Downloafy/suyu-forked
1
0
Fork 1
forked from suyu/suyu
Code Pull requests 1 Activity

Remove references to PICA and rasterizers in video_core

Browse source
This commit is contained in:
James Rowe 2018-01-11 20:07:44 -07:00
parent ebf9a784a9
commit 1d28b2e142
77 changed files with 4 additions and 16444 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

17
src/core/CMakeLists.txt
View file

@ -41,8 +41,6 @@ set(SRCS
hle/service/am/applet_oe.cpp
hle/service/aoc/aoc_u.cpp
hle/service/apm/apm.cpp
hle/service/dsp_dsp.cpp
hle/service/gsp_gpu.cpp
hle/service/hid/hid.cpp
hle/service/lm/lm.cpp
hle/service/nvdrv/devices/nvdisp_disp0.cpp
@ -58,10 +56,6 @@ set(SRCS
hle/service/vi/vi.cpp
hle/service/vi/vi_m.cpp
hle/shared_page.cpp
hw/aes/arithmetic128.cpp
hw/aes/ccm.cpp
hw/aes/key.cpp
hw/gpu.cpp
hw/hw.cpp
hw/lcd.cpp
loader/elf.cpp
@ -130,8 +124,6 @@ set(HEADERS
hle/service/am/applet_oe.h
hle/service/aoc/aoc_u.h
hle/service/apm/apm.h
hle/service/dsp_dsp.h
hle/service/gsp_gpu.h
hle/service/hid/hid.h
hle/service/lm/lm.h
hle/service/nvdrv/devices/nvdevice.h
@ -148,10 +140,6 @@ set(HEADERS
hle/service/vi/vi.h
hle/service/vi/vi_m.h
hle/shared_page.h
hw/aes/arithmetic128.h
hw/aes/ccm.h
hw/aes/key.h
hw/gpu.h
hw/hw.h
hw/lcd.h
loader/elf.h
@ -171,8 +159,5 @@ set(HEADERS
create_directory_groups(${SRCS} ${HEADERS})
add_library(core STATIC ${SRCS} ${HEADERS})
target_link_libraries(core PUBLIC common PRIVATE audio_core dynarmic network video_core)
target_link_libraries(core PUBLIC common PRIVATE dynarmic video_core)
target_link_libraries(core PUBLIC Boost::boost PRIVATE fmt lz4_static unicorn)
if (ENABLE_WEB_SERVICE)
target_link_libraries(core PUBLIC json-headers web_service)
endif()

17
src/core/hle/service/dsp_dsp.cpp
View file

@ -1,17 +0,0 @@
// Copyright 2014 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "audio_core/hle/pipe.h"
#include "core/hle/service/dsp_dsp.h"
using DspPipe = DSP::HLE::DspPipe;
namespace Service {
namespace DSP_DSP {
void SignalPipeInterrupt(DspPipe pipe) {
}
} // namespace DSP_DSP
} // namespace Service

26
src/core/hle/service/dsp_dsp.h
View file

@ -1,26 +0,0 @@
// Copyright 2014 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <string>
#include "core/hle/service/service.h"
namespace DSP {
namespace HLE {
enum class DspPipe;
}
}
namespace Service {
namespace DSP_DSP {
/**
* Signal a specific DSP related interrupt of type == InterruptType::Pipe, pipe == pipe.
* @param pipe The DSP pipe for which to signal an interrupt for.
*/
void SignalPipeInterrupt(DSP::HLE::DspPipe pipe);
} // namespace DSP_DSP
} // namespace Service

11
src/core/hle/service/gsp_gpu.cpp
View file

@ -1,11 +0,0 @@
// Copyright 2014 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "core/hle/service/gsp_gpu.h"
namespace Service {
namespace GSP {
} // namespace GSP
} // namespace Service

195
src/core/hle/service/gsp_gpu.h
View file

@ -1,195 +0,0 @@
// Copyright 2014 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <cstddef>
#include <string>
#include "common/bit_field.h"
#include "common/common_types.h"
#include "core/hle/result.h"
#include "core/hle/service/service.h"
namespace Service {
namespace GSP {
/// GSP interrupt ID
enum class InterruptId : u8 {
PSC0 = 0x00,
PSC1 = 0x01,
PDC0 = 0x02, // Seems called every vertical screen line
PDC1 = 0x03, // Seems called every frame
PPF = 0x04,
P3D = 0x05,
DMA = 0x06,
};
/// GSP command ID
enum class CommandId : u32 {
REQUEST_DMA = 0x00,
/// Submits a commandlist for execution by the GPU.
SUBMIT_GPU_CMDLIST = 0x01,
// Fills a given memory range with a particular value
SET_MEMORY_FILL = 0x02,
// Copies an image and optionally performs color-conversion or scaling.
// This is highly similar to the GameCube's EFB copy feature
SET_DISPLAY_TRANSFER = 0x03,
// Conceptionally similar to SET_DISPLAY_TRANSFER and presumable uses the same hardware path
SET_TEXTURE_COPY = 0x04,
/// Flushes up to 3 cache regions in a single command.
CACHE_FLUSH = 0x05,
};
/// GSP thread interrupt relay queue
struct InterruptRelayQueue {
// Index of last interrupt in the queue
u8 index;
// Number of interrupts remaining to be processed by the userland code
u8 number_interrupts;
// Error code - zero on success, otherwise an error has occurred
u8 error_code;
u8 padding1;
u32 missed_PDC0;
u32 missed_PDC1;
InterruptId slot[0x34]; ///< Interrupt ID slots
};
static_assert(sizeof(InterruptRelayQueue) == 0x40, "InterruptRelayQueue struct has incorrect size");
struct FrameBufferInfo {
BitField<0, 1, u32> active_fb; // 0 = first, 1 = second
u32 address_left;
u32 address_right;
u32 stride; // maps to 0x1EF00X90 ?
u32 format; // maps to 0x1EF00X70 ?
u32 shown_fb; // maps to 0x1EF00X78 ?
u32 unknown;
};
static_assert(sizeof(FrameBufferInfo) == 0x1c, "Struct has incorrect size");
struct FrameBufferUpdate {
BitField<0, 1, u8> index; // Index used for GSP::SetBufferSwap
BitField<0, 1, u8> is_dirty; // true if GSP should update GPU framebuffer registers
u16 pad1;
FrameBufferInfo framebuffer_info[2];
u32 pad2;
};
static_assert(sizeof(FrameBufferUpdate) == 0x40, "Struct has incorrect size");
// TODO: Not sure if this padding is correct.
// Chances are the second block is stored at offset 0x24 rather than 0x20.
#ifndef _MSC_VER
static_assert(offsetof(FrameBufferUpdate, framebuffer_info[1]) == 0x20,
"FrameBufferInfo element has incorrect alignment");
#endif
/// GSP command
struct Command {
BitField<0, 8, CommandId> id;
union {
struct {
u32 source_address;
u32 dest_address;
u32 size;
} dma_request;
struct {
u32 address;
u32 size;
u32 flags;
u32 unused[3];
u32 do_flush;
} submit_gpu_cmdlist;
struct {
u32 start1;
u32 value1;
u32 end1;
u32 start2;
u32 value2;
u32 end2;
u16 control1;
u16 control2;
} memory_fill;
struct {
u32 in_buffer_address;
u32 out_buffer_address;
u32 in_buffer_size;
u32 out_buffer_size;
u32 flags;
} display_transfer;
struct {
u32 in_buffer_address;
u32 out_buffer_address;
u32 size;
u32 in_width_gap;
u32 out_width_gap;
u32 flags;
} texture_copy;
struct {
struct {
u32 address;
u32 size;
} regions[3];
} cache_flush;
u8 raw_data[0x1C];
};
};
static_assert(sizeof(Command) == 0x20, "Command struct has incorrect size");
/// GSP shared memory GX command buffer header
struct CommandBuffer {
union {
u32 hex;
// Current command index. This index is updated by GSP module after loading the command
// data, right before the command is processed. When this index is updated by GSP module,
// the total commands field is decreased by one as well.
BitField<0, 8, u32> index;
// Total commands to process, must not be value 0 when GSP module handles commands. This
// must be <=15 when writing a command to shared memory. This is incremented by the
// application when writing a command to shared memory, after increasing this value
// TriggerCmdReqQueue is only used if this field is value 1.
BitField<8, 8, u32> number_commands;
};
u32 unk[7];
Command commands[0xF];
};
static_assert(sizeof(CommandBuffer) == 0x200, "CommandBuffer struct has incorrect size");
/**
* Signals that the specified interrupt type has occurred to userland code
* @param interrupt_id ID of interrupt that is being signalled
*/
void SignalInterrupt(InterruptId interrupt_id);
ResultCode SetBufferSwap(u32 screen_id, const FrameBufferInfo& info);
/**
* Retrieves the framebuffer info stored in the GSP shared memory for the
* specified screen index and thread id.
* @param thread_id GSP thread id of the process that accesses the structure that we are requesting.
* @param screen_index Index of the screen we are requesting (Top = 0, Bottom = 1).
* @returns FramebufferUpdate Information about the specified framebuffer.
*/
FrameBufferUpdate* GetFrameBufferInfo(u32 thread_id, u32 screen_index);
} // namespace GSP
} // namespace Service

47
src/core/hw/aes/arithmetic128.cpp
View file

@ -1,47 +0,0 @@
// Copyright 2017 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <algorithm>
#include <functional>
#include "core/hw/aes/arithmetic128.h"
namespace HW {
namespace AES {
AESKey Lrot128(const AESKey& in, u32 rot) {
AESKey out;
rot %= 128;
const u32 byte_shift = rot / 8;
const u32 bit_shift = rot % 8;
for (u32 i = 0; i < 16; i++) {
const u32 wrap_index_a = (i + byte_shift) % 16;
const u32 wrap_index_b = (i + byte_shift + 1) % 16;
out[i] = ((in[wrap_index_a] << bit_shift) | (in[wrap_index_b] >> (8 - bit_shift))) & 0xFF;
}
return out;
}
AESKey Add128(const AESKey& a, const AESKey& b) {
AESKey out;
u32 carry = 0;
u32 sum = 0;
for (int i = 15; i >= 0; i--) {
sum = a[i] + b[i] + carry;
carry = sum >> 8;
out[i] = static_cast<u8>(sum & 0xff);
}
return out;
}
AESKey Xor128(const AESKey& a, const AESKey& b) {
AESKey out;
std::transform(a.cbegin(), a.cend(), b.cbegin(), out.begin(), std::bit_xor<>());
return out;
}
} // namespace AES
} // namespace HW

17
src/core/hw/aes/arithmetic128.h
View file

@ -1,17 +0,0 @@
// Copyright 2017 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include "common/common_types.h"
#include "core/hw/aes/key.h"
namespace HW {
namespace AES {
AESKey Lrot128(const AESKey& in, u32 rot);
AESKey Add128(const AESKey& a, const AESKey& b);
AESKey Xor128(const AESKey& a, const AESKey& b);
} // namspace AES
} // namespace HW

40
src/core/hw/aes/ccm.h
View file

@ -1,40 +0,0 @@
// Copyright 2017 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <array>
#include <cstddef>
#include <vector>
#include "common/common_types.h"
namespace HW {
namespace AES {
constexpr size_t CCM_NONCE_SIZE = 12;
constexpr size_t CCM_MAC_SIZE = 16;
using CCMNonce = std::array<u8, CCM_NONCE_SIZE>;
/**
* Encrypts and adds a MAC to the given data using AES-CCM algorithm.
* @param pdata The plain text data to encrypt
* @param nonce The nonce data to use for encryption
* @param slot_id The slot ID of the key to use for encryption
* @returns a vector of u8 containing the encrypted data with MAC at the end
*/
std::vector<u8> EncryptSignCCM(const std::vector<u8>& pdata, const CCMNonce& nonce, size_t slot_id);
/**
* Decrypts and verify the MAC of the given data using AES-CCM algorithm.
* @param cipher The cipher text data to decrypt, with MAC at the end to verify
* @param nonce The nonce data to use for decryption
* @param slot_id The slot ID of the key to use for decryption
* @returns a vector of u8 containing the decrypted data; an empty vector if the verification fails
*/
std::vector<u8> DecryptVerifyCCM(const std::vector<u8>& cipher, const CCMNonce& nonce,
size_t slot_id);
} // namespace AES
} // namespace HW

173
src/core/hw/aes/key.cpp
View file

@ -1,173 +0,0 @@
// Copyright 2017 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <algorithm>
#include <exception>
#include <sstream>
#include <boost/optional.hpp>
#include "common/common_paths.h"
#include "common/file_util.h"
#include "common/logging/log.h"
#include "common/string_util.h"
#include "core/hw/aes/arithmetic128.h"
#include "core/hw/aes/key.h"
namespace HW {
namespace AES {
namespace {
boost::optional<AESKey> generator_constant;
struct KeySlot {
boost::optional<AESKey> x;
boost::optional<AESKey> y;
boost::optional<AESKey> normal;
void SetKeyX(const AESKey& key) {
x = key;
if (y && generator_constant) {
GenerateNormalKey();
}
}
void SetKeyY(const AESKey& key) {
y = key;
if (x && generator_constant) {
GenerateNormalKey();
}
}
void SetNormalKey(const AESKey& key) {
normal = key;
}
void GenerateNormalKey() {
normal = Lrot128(Add128(Xor128(Lrot128(*x, 2), *y), *generator_constant), 87);
}
void Clear() {
x.reset();
y.reset();
normal.reset();
}
};
std::array<KeySlot, KeySlotID::MaxKeySlotID> key_slots;
void ClearAllKeys() {
for (KeySlot& slot : key_slots) {
slot.Clear();
}
generator_constant.reset();
}
AESKey HexToKey(const std::string& hex) {
if (hex.size() < 32) {
throw std::invalid_argument("hex string is too short");
}
AESKey key;
for (size_t i = 0; i < key.size(); ++i) {
key[i] = static_cast<u8>(std::stoi(hex.substr(i * 2, 2), 0, 16));
}
return key;
}
void LoadPresetKeys() {
const std::string filepath = FileUtil::GetUserPath(D_SYSDATA_IDX) + AES_KEYS;
FileUtil::CreateFullPath(filepath); // Create path if not already created
std::ifstream file;
OpenFStream(file, filepath, std::ios_base::in);
if (!file) {
return;
}
while (!file.eof()) {
std::string line;
std::getline(file, line);
std::vector<std::string> parts;
Common::SplitString(line, '=', parts);
if (parts.size() != 2) {
LOG_ERROR(HW_AES, "Failed to parse %s", line.c_str());
continue;
}
const std::string& name = parts[0];
AESKey key;
try {
key = HexToKey(parts[1]);
} catch (const std::logic_error& e) {
LOG_ERROR(HW_AES, "Invalid key %s: %s", parts[1].c_str(), e.what());
continue;
}
if (name == "generator") {
generator_constant = key;
continue;
}
size_t slot_id;
char key_type;
if (std::sscanf(name.c_str(), "slot0x%zXKey%c", &slot_id, &key_type) != 2) {
LOG_ERROR(HW_AES, "Invalid key name %s", name.c_str());
continue;
}
if (slot_id >= MaxKeySlotID) {
LOG_ERROR(HW_AES, "Out of range slot ID 0x%zX", slot_id);
continue;
}
switch (key_type) {
case 'X':
key_slots.at(slot_id).SetKeyX(key);
break;
case 'Y':
key_slots.at(slot_id).SetKeyY(key);
break;
case 'N':
key_slots.at(slot_id).SetNormalKey(key);
break;
default:
LOG_ERROR(HW_AES, "Invalid key type %c", key_type);
break;
}
}
}
} // namespace
void InitKeys() {
ClearAllKeys();
LoadPresetKeys();
}
void SetGeneratorConstant(const AESKey& key) {
generator_constant = key;
}
void SetKeyX(size_t slot_id, const AESKey& key) {
key_slots.at(slot_id).SetKeyX(key);
}
void SetKeyY(size_t slot_id, const AESKey& key) {
key_slots.at(slot_id).SetKeyY(key);
}
void SetNormalKey(size_t slot_id, const AESKey& key) {
key_slots.at(slot_id).SetNormalKey(key);
}
bool IsNormalKeyAvailable(size_t slot_id) {
return key_slots.at(slot_id).normal.is_initialized();
}
AESKey GetNormalKey(size_t slot_id) {
return key_slots.at(slot_id).normal.value_or(AESKey{});
}
} // namespace AES
} // namespace HW

37
src/core/hw/aes/key.h
View file

@ -1,37 +0,0 @@
// Copyright 2017 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <array>
#include <cstddef>
#include "common/common_types.h"
namespace HW {
namespace AES {
enum KeySlotID : size_t {
// AES Keyslot used to generate the UDS data frame CCMP key.
UDSDataKey = 0x2D,
APTWrap = 0x31,
MaxKeySlotID = 0x40,
};
constexpr size_t AES_BLOCK_SIZE = 16;
using AESKey = std::array<u8, AES_BLOCK_SIZE>;
void InitKeys();
void SetGeneratorConstant(const AESKey& key);
void SetKeyX(size_t slot_id, const AESKey& key);
void SetKeyY(size_t slot_id, const AESKey& key);
void SetNormalKey(size_t slot_id, const AESKey& key);
bool IsNormalKeyAvailable(size_t slot_id);
AESKey GetNormalKey(size_t slot_id);
} // namspace AES
} // namespace HW

573
src/core/hw/gpu.cpp
View file

@ -1,573 +0,0 @@
// Copyright 2014 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <cstring>
#include <numeric>
#include <type_traits>
#include "common/alignment.h"
#include "common/color.h"
#include "common/common_types.h"
#include "common/logging/log.h"
#include "common/microprofile.h"
#include "common/vector_math.h"
#include "core/core_timing.h"
#include "core/hle/service/gsp_gpu.h"
#include "core/hw/gpu.h"
#include "core/hw/hw.h"
#include "core/memory.h"
#include "core/tracer/recorder.h"
#include "video_core/command_processor.h"
#include "video_core/debug_utils/debug_utils.h"
#include "video_core/rasterizer_interface.h"
#include "video_core/renderer_base.h"
#include "video_core/utils.h"
#include "video_core/video_core.h"
namespace GPU {
Regs g_regs;
/// 268MHz CPU clocks / 60Hz frames per second
const u64 frame_ticks = static_cast<u64>(BASE_CLOCK_RATE / SCREEN_REFRESH_RATE);
/// Event id for CoreTiming
static CoreTiming::EventType* vblank_event;
template <typename T>
inline void Read(T& var, const u32 raw_addr) {
u32 addr = raw_addr - HW::VADDR_GPU;
u32 index = addr / 4;
// Reads other than u32 are untested, so I'd rather have them abort than silently fail
if (index >= Regs::NumIds() || !std::is_same<T, u32>::value) {
LOG_ERROR(HW_GPU, "unknown Read%lu @ 0x%08X", sizeof(var) * 8, addr);
return;
}
var = g_regs[addr / 4];
}
static Math::Vec4<u8> DecodePixel(Regs::PixelFormat input_format, const u8* src_pixel) {
switch (input_format) {
case Regs::PixelFormat::RGBA8:
return Color::DecodeRGBA8(src_pixel);
case Regs::PixelFormat::RGB8:
return Color::DecodeRGB8(src_pixel);
case Regs::PixelFormat::RGB565:
return Color::DecodeRGB565(src_pixel);
case Regs::PixelFormat::RGB5A1:
return Color::DecodeRGB5A1(src_pixel);
case Regs::PixelFormat::RGBA4:
return Color::DecodeRGBA4(src_pixel);
default:
LOG_ERROR(HW_GPU, "Unknown source framebuffer format %x", input_format);
return {0, 0, 0, 0};
}
}
MICROPROFILE_DEFINE(GPU_DisplayTransfer, "GPU", "DisplayTransfer", MP_RGB(100, 100, 255));
MICROPROFILE_DEFINE(GPU_CmdlistProcessing, "GPU", "Cmdlist Processing", MP_RGB(100, 255, 100));
static void MemoryFill(const Regs::MemoryFillConfig& config) {
const PAddr start_addr = config.GetStartAddress();
const PAddr end_addr = config.GetEndAddress();
// TODO: do hwtest with these cases
if (!Memory::IsValidPhysicalAddress(start_addr)) {
LOG_CRITICAL(HW_GPU, "invalid start address 0x%08X", start_addr);
return;
}
if (!Memory::IsValidPhysicalAddress(end_addr)) {
LOG_CRITICAL(HW_GPU, "invalid end address 0x%08X", end_addr);
return;
}
if (end_addr <= start_addr) {
LOG_CRITICAL(HW_GPU, "invalid memory range from 0x%08X to 0x%08X", start_addr, end_addr);
return;
}
u8* start = Memory::GetPhysicalPointer(start_addr);
u8* end = Memory::GetPhysicalPointer(end_addr);
// TODO: Consider always accelerating and returning vector of
// regions that the accelerated fill did not cover to
// reduce/eliminate the fill that the cpu has to do.
// This would also mean that the flush below is not needed.
// Fill should first flush all surfaces that touch but are
// not completely within the fill range.
// Then fill all completely covered surfaces, and return the
// regions that were between surfaces or within the touching
// ones for cpu to manually fill here.
if (VideoCore::g_renderer->Rasterizer()->AccelerateFill(config))
return;
Memory::RasterizerFlushAndInvalidateRegion(config.GetStartAddress(),
config.GetEndAddress() - config.GetStartAddress());
if (config.fill_24bit) {
// fill with 24-bit values
for (u8* ptr = start; ptr < end; ptr += 3) {
ptr[0] = config.value_24bit_r;
ptr[1] = config.value_24bit_g;
ptr[2] = config.value_24bit_b;
}
} else if (config.fill_32bit) {
// fill with 32-bit values
if (end > start) {
u32 value = config.value_32bit;
size_t len = (end - start) / sizeof(u32);
for (size_t i = 0; i < len; ++i)
memcpy(&start[i * sizeof(u32)], &value, sizeof(u32));
}
} else {
// fill with 16-bit values
u16 value_16bit = config.value_16bit.Value();
for (u8* ptr = start; ptr < end; ptr += sizeof(u16))
memcpy(ptr, &value_16bit, sizeof(u16));
}
}
static void DisplayTransfer(const Regs::DisplayTransferConfig& config) {
const PAddr src_addr = config.GetPhysicalInputAddress();
const PAddr dst_addr = config.GetPhysicalOutputAddress();
// TODO: do hwtest with these cases
if (!Memory::IsValidPhysicalAddress(src_addr)) {
LOG_CRITICAL(HW_GPU, "invalid input address 0x%08X", src_addr);
return;
}
if (!Memory::IsValidPhysicalAddress(dst_addr)) {
LOG_CRITICAL(HW_GPU, "invalid output address 0x%08X", dst_addr);
return;
}
if (config.input_width == 0) {
LOG_CRITICAL(HW_GPU, "zero input width");
return;
}
if (config.input_height == 0) {
LOG_CRITICAL(HW_GPU, "zero input height");
return;
}
if (config.output_width == 0) {
LOG_CRITICAL(HW_GPU, "zero output width");
return;
}
if (config.output_height == 0) {
LOG_CRITICAL(HW_GPU, "zero output height");
return;
}
if (VideoCore::g_renderer->Rasterizer()->AccelerateDisplayTransfer(config))
return;
u8* src_pointer = Memory::GetPhysicalPointer(src_addr);
u8* dst_pointer = Memory::GetPhysicalPointer(dst_addr);
if (config.scaling > config.ScaleXY) {
LOG_CRITICAL(HW_GPU, "Unimplemented display transfer scaling mode %u",
config.scaling.Value());
UNIMPLEMENTED();
return;
}
if (config.input_linear && config.scaling != config.NoScale) {
LOG_CRITICAL(HW_GPU, "Scaling is only implemented on tiled input");
UNIMPLEMENTED();
return;
}
int horizontal_scale = config.scaling != config.NoScale ? 1 : 0;
int vertical_scale = config.scaling == config.ScaleXY ? 1 : 0;
u32 output_width = config.output_width >> horizontal_scale;
u32 output_height = config.output_height >> vertical_scale;
u32 input_size =
config.input_width * config.input_height * GPU::Regs::BytesPerPixel(config.input_format);
u32 output_size = output_width * output_height * GPU::Regs::BytesPerPixel(config.output_format);
Memory::RasterizerFlushRegion(config.GetPhysicalInputAddress(), input_size);
Memory::RasterizerFlushAndInvalidateRegion(config.GetPhysicalOutputAddress(), output_size);
for (u32 y = 0; y < output_height; ++y) {
for (u32 x = 0; x < output_width; ++x) {
Math::Vec4<u8> src_color;
// Calculate the [x,y] position of the input image
// based on the current output position and the scale
u32 input_x = x << horizontal_scale;
u32 input_y = y << vertical_scale;
u32 output_y;
if (config.flip_vertically) {
// Flip the y value of the output data,
// we do this after calculating the [x,y] position of the input image
// to account for the scaling options.
output_y = output_height - y - 1;
} else {
output_y = y;
}
u32 dst_bytes_per_pixel = GPU::Regs::BytesPerPixel(config.output_format);
u32 src_bytes_per_pixel = GPU::Regs::BytesPerPixel(config.input_format);
u32 src_offset;
u32 dst_offset;
if (config.input_linear) {
if (!config.dont_swizzle) {
// Interpret the input as linear and the output as tiled
u32 coarse_y = output_y & ~7;
u32 stride = output_width * dst_bytes_per_pixel;
src_offset = (input_x + input_y * config.input_width) * src_bytes_per_pixel;
dst_offset = VideoCore::GetMortonOffset(x, output_y, dst_bytes_per_pixel) +
coarse_y * stride;
} else {
// Both input and output are linear
src_offset = (input_x + input_y * config.input_width) * src_bytes_per_pixel;
dst_offset = (x + output_y * output_width) * dst_bytes_per_pixel;
}
} else {
if (!config.dont_swizzle) {
// Interpret the input as tiled and the output as linear
u32 coarse_y = input_y & ~7;
u32 stride = config.input_width * src_bytes_per_pixel;
src_offset = VideoCore::GetMortonOffset(input_x, input_y, src_bytes_per_pixel) +
coarse_y * stride;
dst_offset = (x + output_y * output_width) * dst_bytes_per_pixel;
} else {
// Both input and output are tiled
u32 out_coarse_y = output_y & ~7;
u32 out_stride = output_width * dst_bytes_per_pixel;
u32 in_coarse_y = input_y & ~7;
u32 in_stride = config.input_width * src_bytes_per_pixel;
src_offset = VideoCore::GetMortonOffset(input_x, input_y, src_bytes_per_pixel) +
in_coarse_y * in_stride;
dst_offset = VideoCore::GetMortonOffset(x, output_y, dst_bytes_per_pixel) +
out_coarse_y * out_stride;
}
}
const u8* src_pixel = src_pointer + src_offset;
src_color = DecodePixel(config.input_format, src_pixel);
if (config.scaling == config.ScaleX) {
Math::Vec4<u8> pixel =
DecodePixel(config.input_format, src_pixel + src_bytes_per_pixel);
src_color = ((src_color + pixel) / 2).Cast<u8>();
} else if (config.scaling == config.ScaleXY) {
Math::Vec4<u8> pixel1 =
DecodePixel(config.input_format, src_pixel + 1 * src_bytes_per_pixel);
Math::Vec4<u8> pixel2 =
DecodePixel(config.input_format, src_pixel + 2 * src_bytes_per_pixel);
Math::Vec4<u8> pixel3 =
DecodePixel(config.input_format, src_pixel + 3 * src_bytes_per_pixel);
src_color = (((src_color + pixel1) + (pixel2 + pixel3)) / 4).Cast<u8>();
}
u8* dst_pixel = dst_pointer + dst_offset;
switch (config.output_format) {
case Regs::PixelFormat::RGBA8:
Color::EncodeRGBA8(src_color, dst_pixel);
break;
case Regs::PixelFormat::RGB8:
Color::EncodeRGB8(src_color, dst_pixel);
break;
case Regs::PixelFormat::RGB565:
Color::EncodeRGB565(src_color, dst_pixel);
break;
case Regs::PixelFormat::RGB5A1:
Color::EncodeRGB5A1(src_color, dst_pixel);
break;
case Regs::PixelFormat::RGBA4:
Color::EncodeRGBA4(src_color, dst_pixel);
break;
default:
LOG_ERROR(HW_GPU, "Unknown destination framebuffer format %x",
config.output_format.Value());
break;
}
}
}
}
static void TextureCopy(const Regs::DisplayTransferConfig& config) {
const PAddr src_addr = config.GetPhysicalInputAddress();
const PAddr dst_addr = config.GetPhysicalOutputAddress();
// TODO: do hwtest with invalid addresses
if (!Memory::IsValidPhysicalAddress(src_addr)) {
LOG_CRITICAL(HW_GPU, "invalid input address 0x%08X", src_addr);
return;
}
if (!Memory::IsValidPhysicalAddress(dst_addr)) {
LOG_CRITICAL(HW_GPU, "invalid output address 0x%08X", dst_addr);
return;
}
if (VideoCore::g_renderer->Rasterizer()->AccelerateTextureCopy(config))
return;
u8* src_pointer = Memory::GetPhysicalPointer(src_addr);
u8* dst_pointer = Memory::GetPhysicalPointer(dst_addr);
u32 remaining_size = Common::AlignDown(config.texture_copy.size, 16);
if (remaining_size == 0) {
LOG_CRITICAL(HW_GPU, "zero size. Real hardware freezes on this.");
return;
}
u32 input_gap = config.texture_copy.input_gap * 16;
u32 output_gap = config.texture_copy.output_gap * 16;
// Zero gap means contiguous input/output even if width = 0. To avoid infinite loop below, width
// is assigned with the total size if gap = 0.
u32 input_width = input_gap == 0 ? remaining_size : config.texture_copy.input_width * 16;
u32 output_width = output_gap == 0 ? remaining_size : config.texture_copy.output_width * 16;
if (input_width == 0) {
LOG_CRITICAL(HW_GPU, "zero input width. Real hardware freezes on this.");
return;
}
if (output_width == 0) {
LOG_CRITICAL(HW_GPU, "zero output width. Real hardware freezes on this.");
return;
}
size_t contiguous_input_size =
config.texture_copy.size / input_width * (input_width + input_gap);
Memory::RasterizerFlushRegion(config.GetPhysicalInputAddress(),
static_cast<u32>(contiguous_input_size));
size_t contiguous_output_size =
config.texture_copy.size / output_width * (output_width + output_gap);
Memory::RasterizerFlushAndInvalidateRegion(config.GetPhysicalOutputAddress(),
static_cast<u32>(contiguous_output_size));
u32 remaining_input = input_width;
u32 remaining_output = output_width;
while (remaining_size > 0) {
u32 copy_size = std::min({remaining_input, remaining_output, remaining_size});
std::memcpy(dst_pointer, src_pointer, copy_size);
src_pointer += copy_size;
dst_pointer += copy_size;
remaining_input -= copy_size;
remaining_output -= copy_size;
remaining_size -= copy_size;
if (remaining_input == 0) {
remaining_input = input_width;
src_pointer += input_gap;
}
if (remaining_output == 0) {
remaining_output = output_width;
dst_pointer += output_gap;
}
}
}
template <typename T>
inline void Write(u32 addr, const T data) {
addr -= HW::VADDR_GPU;
u32 index = addr / 4;
// Writes other than u32 are untested, so I'd rather have them abort than silently fail
if (index >= Regs::NumIds() || !std::is_same<T, u32>::value) {
LOG_ERROR(HW_GPU, "unknown Write%lu 0x%08X @ 0x%08X", sizeof(data) * 8, (u32)data, addr);
return;
}
g_regs[index] = static_cast<u32>(data);
switch (index) {
// Memory fills are triggered once the fill value is written.
case GPU_REG_INDEX_WORKAROUND(memory_fill_config[0].trigger, 0x00004 + 0x3):
case GPU_REG_INDEX_WORKAROUND(memory_fill_config[1].trigger, 0x00008 + 0x3): {
const bool is_second_filler = (index != GPU_REG_INDEX(memory_fill_config[0].trigger));
auto& config = g_regs.memory_fill_config[is_second_filler];
if (config.trigger) {
MemoryFill(config);
LOG_TRACE(HW_GPU, "MemoryFill from 0x%08x to 0x%08x", config.GetStartAddress(),
config.GetEndAddress());
// It seems that it won't signal interrupt if "address_start" is zero.
// TODO: hwtest this
if (config.GetStartAddress() != 0) {
if (!is_second_filler) {
//Service::GSP::SignalInterrupt(Service::GSP::InterruptId::PSC0);
} else {
//Service::GSP::SignalInterrupt(Service::GSP::InterruptId::PSC1);
}
}
// Reset "trigger" flag and set the "finish" flag
// NOTE: This was confirmed to happen on hardware even if "address_start" is zero.
config.trigger.Assign(0);
config.finished.Assign(1);
}
break;
}
case GPU_REG_INDEX(display_transfer_config.trigger): {
MICROPROFILE_SCOPE(GPU_DisplayTransfer);
const auto& config = g_regs.display_transfer_config;
if (config.trigger & 1) {
if (Pica::g_debug_context)
Pica::g_debug_context->OnEvent(Pica::DebugContext::Event::IncomingDisplayTransfer,
nullptr);
if (config.is_texture_copy) {
TextureCopy(config);
LOG_TRACE(HW_GPU, "TextureCopy: 0x%X bytes from 0x%08X(%u+%u)-> "
"0x%08X(%u+%u), flags 0x%08X",
config.texture_copy.size, config.GetPhysicalInputAddress(),
config.texture_copy.input_width * 16, config.texture_copy.input_gap * 16,
config.GetPhysicalOutputAddress(), config.texture_copy.output_width * 16,
config.texture_copy.output_gap * 16, config.flags);
} else {
DisplayTransfer(config);
LOG_TRACE(HW_GPU, "DisplayTransfer: 0x%08x(%ux%u)-> "
"0x%08x(%ux%u), dst format %x, flags 0x%08X",
config.GetPhysicalInputAddress(), config.input_width.Value(),
config.input_height.Value(), config.GetPhysicalOutputAddress(),
config.output_width.Value(), config.output_height.Value(),
config.output_format.Value(), config.flags);
}
g_regs.display_transfer_config.trigger = 0;
//Service::GSP::SignalInterrupt(Service::GSP::InterruptId::PPF);
}
break;
}
// Seems like writing to this register triggers processing
case GPU_REG_INDEX(command_processor_config.trigger): {
const auto& config = g_regs.command_processor_config;
if (config.trigger & 1) {
MICROPROFILE_SCOPE(GPU_CmdlistProcessing);
u32* buffer = (u32*)Memory::GetPhysicalPointer(config.GetPhysicalAddress());
if (Pica::g_debug_context && Pica::g_debug_context->recorder) {
Pica::g_debug_context->recorder->MemoryAccessed((u8*)buffer, config.size,
config.GetPhysicalAddress());
}
Pica::CommandProcessor::ProcessCommandList(buffer, config.size);
g_regs.command_processor_config.trigger = 0;
}
break;
}
default:
break;
}
// Notify tracer about the register write
// This is happening *after* handling the write to make sure we properly catch all memory reads.
if (Pica::g_debug_context && Pica::g_debug_context->recorder) {
// addr + GPU VBase - IO VBase + IO PBase
Pica::g_debug_context->recorder->RegisterWritten<T>(
addr + 0x1EF00000 - 0x1EC00000 + 0x10100000, data);
}
}
// Explicitly instantiate template functions because we aren't defining this in the header:
template void Read<u64>(u64& var, const u32 addr);
template void Read<u32>(u32& var, const u32 addr);
template void Read<u16>(u16& var, const u32 addr);
template void Read<u8>(u8& var, const u32 addr);
template void Write<u64>(u32 addr, const u64 data);
template void Write<u32>(u32 addr, const u32 data);
template void Write<u16>(u32 addr, const u16 data);
template void Write<u8>(u32 addr, const u8 data);
/// Update hardware
static void VBlankCallback(u64 userdata, int cycles_late) {
//VideoCore::g_renderer->SwapBuffers();
//// Signal to GSP that GPU interrupt has occurred
//// TODO(yuriks): hwtest to determine if PDC0 is for the Top screen and PDC1 for the Sub
//// screen, or if both use the same interrupts and these two instead determine the
//// beginning and end of the VBlank period. If needed, split the interrupt firing into
//// two different intervals.
//Service::GSP::SignalInterrupt(Service::GSP::InterruptId::PDC0);
//Service::GSP::SignalInterrupt(Service::GSP::InterruptId::PDC1);
// Reschedule recurrent event
CoreTiming::ScheduleEvent(frame_ticks - cycles_late, vblank_event);
}
/// Initialize hardware
void Init() {
memset(&g_regs, 0, sizeof(g_regs));
auto& framebuffer_top = g_regs.framebuffer_config[0];
auto& framebuffer_sub = g_regs.framebuffer_config[1];
// Setup default framebuffer addresses (located in VRAM)
// .. or at least these are the ones used by system applets.
// There's probably a smarter way to come up with addresses
// like this which does not require hardcoding.
framebuffer_top.address_left1 = 0x181E6000;
framebuffer_top.address_left2 = 0x1822C800;
framebuffer_top.address_right1 = 0x18273000;
framebuffer_top.address_right2 = 0x182B9800;
framebuffer_sub.address_left1 = 0x1848F000;
framebuffer_sub.address_left2 = 0x184C7800;
framebuffer_top.width.Assign(240);
framebuffer_top.height.Assign(400);
framebuffer_top.stride = 3 * 240;
framebuffer_top.color_format.Assign(Regs::PixelFormat::RGB8);
framebuffer_top.active_fb = 0;
framebuffer_sub.width.Assign(240);
framebuffer_sub.height.Assign(320);
framebuffer_sub.stride = 3 * 240;
framebuffer_sub.color_format.Assign(Regs::PixelFormat::RGB8);
framebuffer_sub.active_fb = 0;
vblank_event = CoreTiming::RegisterEvent("GPU::VBlankCallback", VBlankCallback);
CoreTiming::ScheduleEvent(frame_ticks, vblank_event);
LOG_DEBUG(HW_GPU, "initialized OK");
}
/// Shutdown hardware
void Shutdown() {
LOG_DEBUG(HW_GPU, "shutdown OK");
}
} // namespace

334
src/core/hw/gpu.h
View file

@ -1,334 +0,0 @@
// Copyright 2014 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <cstddef>
#include <type_traits>
#include "common/assert.h"
#include "common/bit_field.h"
#include "common/common_funcs.h"
#include "common/common_types.h"
namespace GPU {
constexpr float SCREEN_REFRESH_RATE = 60;
// Returns index corresponding to the Regs member labeled by field_name
// TODO: Due to Visual studio bug 209229, offsetof does not return constant expressions
// when used with array elements (e.g. GPU_REG_INDEX(memory_fill_config[0])).
// For details cf.
// https://connect.microsoft.com/VisualStudio/feedback/details/209229/offsetof-does-not-produce-a-constant-expression-for-array-members
// Hopefully, this will be fixed sometime in the future.
// For lack of better alternatives, we currently hardcode the offsets when constant
// expressions are needed via GPU_REG_INDEX_WORKAROUND (on sane compilers, static_asserts
// will then make sure the offsets indeed match the automatically calculated ones).
#define GPU_REG_INDEX(field_name) (offsetof(GPU::Regs, field_name) / sizeof(u32))
#if defined(_MSC_VER)
#define GPU_REG_INDEX_WORKAROUND(field_name, backup_workaround_index) (backup_workaround_index)
#else
// NOTE: Yeah, hacking in a static_assert here just to workaround the lacking MSVC compiler
// really is this annoying. This macro just forwards its first argument to GPU_REG_INDEX
// and then performs a (no-op) cast to size_t iff the second argument matches the expected
// field offset. Otherwise, the compiler will fail to compile this code.
#define GPU_REG_INDEX_WORKAROUND(field_name, backup_workaround_index) \
((typename std::enable_if<backup_workaround_index == GPU_REG_INDEX(field_name), size_t>::type) \
GPU_REG_INDEX(field_name))
#endif
// MMIO region 0x1EFxxxxx
struct Regs {
// helper macro to make sure the defined structures are of the expected size.
#if defined(_MSC_VER)
// TODO: MSVC does not support using sizeof() on non-static data members even though this
// is technically allowed since C++11. This macro should be enabled once MSVC adds
// support for that.
#define ASSERT_MEMBER_SIZE(name, size_in_bytes)
#else
#define ASSERT_MEMBER_SIZE(name, size_in_bytes) \
static_assert(sizeof(name) == size_in_bytes, \
"Structure size and register block length don't match")
#endif
// Components are laid out in reverse byte order, most significant bits first.
enum class PixelFormat : u32 {
RGBA8 = 0,
RGB8 = 1,
RGB565 = 2,
RGB5A1 = 3,
RGBA4 = 4,
};
/**
* Returns the number of bytes per pixel.
*/
static int BytesPerPixel(PixelFormat format) {
switch (format) {
case PixelFormat::RGBA8:
return 4;
case PixelFormat::RGB8:
return 3;
case PixelFormat::RGB565:
case PixelFormat::RGB5A1:
case PixelFormat::RGBA4:
return 2;
}
UNREACHABLE();
}
INSERT_PADDING_WORDS(0x4);
struct MemoryFillConfig {
u32 address_start;
u32 address_end;
union {
u32 value_32bit;
BitField<0, 16, u32> value_16bit;
// TODO: Verify component order
BitField<0, 8, u32> value_24bit_r;
BitField<8, 8, u32> value_24bit_g;
BitField<16, 8, u32> value_24bit_b;
};
union {
u32 control;
// Setting this field to 1 triggers the memory fill.
// This field also acts as a status flag, and gets reset to 0 upon completion.
BitField<0, 1, u32> trigger;
// Set to 1 upon completion.
BitField<1, 1, u32> finished;
// If both of these bits are unset, then it will fill the memory with a 16 bit value
// 1: fill with 24-bit wide values
BitField<8, 1, u32> fill_24bit;
// 1: fill with 32-bit wide values
BitField<9, 1, u32> fill_32bit;
};
inline u32 GetStartAddress() const {
return DecodeAddressRegister(address_start);
}
inline u32 GetEndAddress() const {
return DecodeAddressRegister(address_end);
}
} memory_fill_config[2];
ASSERT_MEMBER_SIZE(memory_fill_config[0], 0x10);
INSERT_PADDING_WORDS(0x10b);
struct FramebufferConfig {
union {
u32 size;
BitField<0, 16, u32> width;
BitField<16, 16, u32> height;
};
INSERT_PADDING_WORDS(0x2);
u32 address_left1;
u32 address_left2;
union {
u32 format;
BitField<0, 3, PixelFormat> color_format;
};
INSERT_PADDING_WORDS(0x1);
union {
u32 active_fb;
// 0: Use parameters ending with "1"
// 1: Use parameters ending with "2"
BitField<0, 1, u32> second_fb_active;
};
INSERT_PADDING_WORDS(0x5);
// Distance between two pixel rows, in bytes
u32 stride;
u32 address_right1;
u32 address_right2;
INSERT_PADDING_WORDS(0x30);
} framebuffer_config[2];
ASSERT_MEMBER_SIZE(framebuffer_config[0], 0x100);
INSERT_PADDING_WORDS(0x169);
struct DisplayTransferConfig {
u32 input_address;
u32 output_address;
inline u32 GetPhysicalInputAddress() const {
return DecodeAddressRegister(input_address);
}
inline u32 GetPhysicalOutputAddress() const {
return DecodeAddressRegister(output_address);
}
union {
u32 output_size;
BitField<0, 16, u32> output_width;
BitField<16, 16, u32> output_height;
};
union {
u32 input_size;
BitField<0, 16, u32> input_width;
BitField<16, 16, u32> input_height;
};
enum ScalingMode : u32 {
NoScale = 0, // Doesn't scale the image
ScaleX = 1, // Downscales the image in half in the X axis and applies a box filter
ScaleXY =
2, // Downscales the image in half in both the X and Y axes and applies a box filter
};
union {
u32 flags;
BitField<0, 1, u32> flip_vertically; // flips input data vertically
BitField<1, 1, u32> input_linear; // Converts from linear to tiled format
BitField<2, 1, u32> crop_input_lines;
BitField<3, 1, u32> is_texture_copy; // Copies the data without performing any
// processing and respecting texture copy fields
BitField<5, 1, u32> dont_swizzle;
BitField<8, 3, PixelFormat> input_format;
BitField<12, 3, PixelFormat> output_format;
/// Uses some kind of 32x32 block swizzling mode, instead of the usual 8x8 one.
BitField<16, 1, u32> block_32; // TODO(yuriks): unimplemented
BitField<24, 2, ScalingMode> scaling; // Determines the scaling mode of the transfer
};
INSERT_PADDING_WORDS(0x1);
// it seems that writing to this field triggers the display transfer
u32 trigger;
INSERT_PADDING_WORDS(0x1);
struct {
u32 size; // The lower 4 bits are ignored
union {
u32 input_size;
BitField<0, 16, u32> input_width;
BitField<16, 16, u32> input_gap;
};
union {
u32 output_size;
BitField<0, 16, u32> output_width;
BitField<16, 16, u32> output_gap;
};
} texture_copy;
} display_transfer_config;
ASSERT_MEMBER_SIZE(display_transfer_config, 0x2c);
INSERT_PADDING_WORDS(0x32D);
struct {
// command list size (in bytes)
u32 size;
INSERT_PADDING_WORDS(0x1);
// command list address
u32 address;
INSERT_PADDING_WORDS(0x1);
// it seems that writing to this field triggers command list processing
u32 trigger;
inline u32 GetPhysicalAddress() const {
return DecodeAddressRegister(address);
}
} command_processor_config;
ASSERT_MEMBER_SIZE(command_processor_config, 0x14);
INSERT_PADDING_WORDS(0x9c3);
static constexpr size_t NumIds() {
return sizeof(Regs) / sizeof(u32);
}
const u32& operator[](int index) const {
const u32* content = reinterpret_cast<const u32*>(this);
return content[index];
}
u32& operator[](int index) {
u32* content = reinterpret_cast<u32*>(this);
return content[index];
}
#undef ASSERT_MEMBER_SIZE
private:
/*
* Most physical addresses which GPU registers refer to are 8-byte aligned.
* This function should be used to get the address from a raw register value.
*/
static inline u32 DecodeAddressRegister(u32 register_value) {
return register_value * 8;
}
};
static_assert(std::is_standard_layout<Regs>::value, "Structure does not use standard layout");
// TODO: MSVC does not support using offsetof() on non-static data members even though this
// is technically allowed since C++11. This macro should be enabled once MSVC adds
// support for that.
#ifndef _MSC_VER
#define ASSERT_REG_POSITION(field_name, position) \
static_assert(offsetof(Regs, field_name) == position * 4, \
"Field " #field_name " has invalid position")
ASSERT_REG_POSITION(memory_fill_config[0], 0x00004);
ASSERT_REG_POSITION(memory_fill_config[1], 0x00008);
ASSERT_REG_POSITION(framebuffer_config[0], 0x00117);
ASSERT_REG_POSITION(framebuffer_config[1], 0x00157);
ASSERT_REG_POSITION(display_transfer_config, 0x00300);
ASSERT_REG_POSITION(command_processor_config, 0x00638);
#undef ASSERT_REG_POSITION
#endif // !defined(_MSC_VER)
// The total number of registers is chosen arbitrarily, but let's make sure it's not some odd value
// anyway.
static_assert(sizeof(Regs) == 0x1000 * sizeof(u32), "Invalid total size of register set");
extern Regs g_regs;
template <typename T>
void Read(T& var, const u32 addr);
template <typename T>
void Write(u32 addr, const T data);
/// Initialize hardware
void Init();
/// Shutdown hardware
void Shutdown();
} // namespace

6
src/core/settings.cpp
View file

@ -2,7 +2,6 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "audio_core/audio_core.h"
#include "core/gdbstub/gdbstub.h"
#include "core/hle/service/hid/hid.h"
#include "core/settings.h"
@ -19,8 +18,6 @@ void Apply() {
GDBStub::SetServerPort(values.gdbstub_port);
GDBStub::ToggleServer(values.use_gdbstub);
VideoCore::g_hw_renderer_enabled = values.use_hw_renderer;
VideoCore::g_shader_jit_enabled = values.use_shader_jit;
VideoCore::g_toggle_framelimit_enabled = values.toggle_framelimit;
if (VideoCore::g_emu_window) {
@ -28,9 +25,6 @@ void Apply() {
VideoCore::g_emu_window->UpdateCurrentFramebufferLayout(layout.width, layout.height);
}
AudioCore::SelectSink(values.sink_id);
AudioCore::EnableStretching(values.enable_audio_stretching);
Service::HID::ReloadInputDevices();
}

75
src/video_core/CMakeLists.txt
View file

@ -1,96 +1,23 @@
set(SRCS
command_processor.cpp
debug_utils/debug_utils.cpp
geometry_pipeline.cpp
pica.cpp
primitive_assembly.cpp
regs.cpp
renderer_base.cpp
renderer_opengl/gl_rasterizer.cpp
renderer_opengl/gl_rasterizer_cache.cpp
renderer_opengl/gl_shader_gen.cpp
renderer_opengl/gl_shader_util.cpp
renderer_opengl/gl_state.cpp
renderer_opengl/renderer_opengl.cpp
shader/shader.cpp
shader/shader_interpreter.cpp
swrasterizer/clipper.cpp
swrasterizer/framebuffer.cpp
swrasterizer/lighting.cpp
swrasterizer/proctex.cpp
swrasterizer/rasterizer.cpp
swrasterizer/swrasterizer.cpp
swrasterizer/texturing.cpp
texture/etc1.cpp
texture/texture_decode.cpp
vertex_loader.cpp
video_core.cpp
)
set(HEADERS
command_processor.h
debug_utils/debug_utils.h
geometry_pipeline.h
gpu_debugger.h
pica.h
pica_state.h
pica_types.h
primitive_assembly.h
rasterizer_interface.h
regs.h
regs_framebuffer.h
regs_lighting.h
regs_pipeline.h
regs_rasterizer.h
regs_shader.h
regs_texturing.h
renderer_base.h
renderer_opengl/gl_rasterizer.h
renderer_opengl/gl_rasterizer_cache.h
renderer_opengl/gl_resource_manager.h
renderer_opengl/gl_shader_gen.h
renderer_opengl/gl_shader_util.h
renderer_opengl/gl_state.h
renderer_opengl/pica_to_gl.h
renderer_opengl/renderer_opengl.h
shader/debug_data.h
shader/shader.h
shader/shader_interpreter.h
swrasterizer/clipper.h
swrasterizer/framebuffer.h
swrasterizer/lighting.h
swrasterizer/proctex.h
swrasterizer/rasterizer.h
swrasterizer/swrasterizer.h
swrasterizer/texturing.h
texture/etc1.h
texture/texture_decode.h
utils.h
vertex_loader.h
video_core.h
)
if(ARCHITECTURE_x86_64)
set(SRCS ${SRCS}
shader/shader_jit_x64.cpp
shader/shader_jit_x64_compiler.cpp)
set(HEADERS ${HEADERS}
shader/shader_jit_x64.h
shader/shader_jit_x64_compiler.h)
endif()
create_directory_groups(${SRCS} ${HEADERS})
add_library(video_core STATIC ${SRCS} ${HEADERS})
target_link_libraries(video_core PUBLIC common core)
target_link_libraries(video_core PRIVATE glad nihstro-headers)
if (ARCHITECTURE_x86_64)
target_link_libraries(video_core PRIVATE xbyak)
endif()
if (PNG_FOUND)
target_link_libraries(video_core PRIVATE PNG::PNG)
target_compile_definitions(video_core PRIVATE HAVE_PNG)
endif()
target_link_libraries(video_core PRIVATE glad)

647
src/video_core/command_processor.cpp
View file

@ -1,647 +0,0 @@
// Copyright 2014 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <array>
#include <cstddef>
#include <memory>
#include <utility>
#include "common/assert.h"
#include "common/logging/log.h"
#include "common/microprofile.h"
#include "common/vector_math.h"
#include "core/hle/service/gsp_gpu.h"
#include "core/hw/gpu.h"
#include "core/memory.h"
#include "core/tracer/recorder.h"
#include "video_core/command_processor.h"
#include "video_core/debug_utils/debug_utils.h"
#include "video_core/pica_state.h"
#include "video_core/pica_types.h"
#include "video_core/primitive_assembly.h"
#include "video_core/rasterizer_interface.h"
#include "video_core/regs.h"
#include "video_core/regs_pipeline.h"
#include "video_core/regs_texturing.h"
#include "video_core/renderer_base.h"
#include "video_core/shader/shader.h"
#include "video_core/vertex_loader.h"
#include "video_core/video_core.h"
namespace Pica {
namespace CommandProcessor {
static int vs_float_regs_counter = 0;
static u32 vs_uniform_write_buffer[4];
static int gs_float_regs_counter = 0;
static u32 gs_uniform_write_buffer[4];
static int default_attr_counter = 0;
static u32 default_attr_write_buffer[3];
// Expand a 4-bit mask to 4-byte mask, e.g. 0b0101 -> 0x00FF00FF
static const u32 expand_bits_to_bytes[] = {
0x00000000, 0x000000ff, 0x0000ff00, 0x0000ffff, 0x00ff0000, 0x00ff00ff, 0x00ffff00, 0x00ffffff,
0xff000000, 0xff0000ff, 0xff00ff00, 0xff00ffff, 0xffff0000, 0xffff00ff, 0xffffff00, 0xffffffff,
};
MICROPROFILE_DEFINE(GPU_Drawing, "GPU", "Drawing", MP_RGB(50, 50, 240));
static const char* GetShaderSetupTypeName(Shader::ShaderSetup& setup) {
if (&setup == &g_state.vs) {
return "vertex shader";
}
if (&setup == &g_state.gs) {
return "geometry shader";
}
return "unknown shader";
}
static void WriteUniformBoolReg(Shader::ShaderSetup& setup, u32 value) {
for (unsigned i = 0; i < setup.uniforms.b.size(); ++i)
setup.uniforms.b[i] = (value & (1 << i)) != 0;
}
static void WriteUniformIntReg(Shader::ShaderSetup& setup, unsigned index,
const Math::Vec4<u8>& values) {
ASSERT(index < setup.uniforms.i.size());
setup.uniforms.i[index] = values;
LOG_TRACE(HW_GPU, "Set %s integer uniform %d to %02x %02x %02x %02x",
GetShaderSetupTypeName(setup), index, values.x, values.y, values.z, values.w);
}
static void WriteUniformFloatReg(ShaderRegs& config, Shader::ShaderSetup& setup,
int& float_regs_counter, u32 uniform_write_buffer[4], u32 value) {
auto& uniform_setup = config.uniform_setup;
// TODO: Does actual hardware indeed keep an intermediate buffer or does
// it directly write the values?
uniform_write_buffer[float_regs_counter++] = value;
// Uniforms are written in a packed format such that four float24 values are encoded in
// three 32-bit numbers. We write to internal memory once a full such vector is
// written.
if ((float_regs_counter >= 4 && uniform_setup.IsFloat32()) ||
(float_regs_counter >= 3 && !uniform_setup.IsFloat32())) {
float_regs_counter = 0;
auto& uniform = setup.uniforms.f[uniform_setup.index];
if (uniform_setup.index >= 96) {
LOG_ERROR(HW_GPU, "Invalid %s float uniform index %d", GetShaderSetupTypeName(setup),
(int)uniform_setup.index);
} else {
// NOTE: The destination component order indeed is "backwards"
if (uniform_setup.IsFloat32()) {
for (auto i : {0, 1, 2, 3})
uniform[3 - i] = float24::FromFloat32(*(float*)(&uniform_write_buffer[i]));
} else {
// TODO: Untested
uniform.w = float24::FromRaw(uniform_write_buffer[0] >> 8);
uniform.z = float24::FromRaw(((uniform_write_buffer[0] & 0xFF) << 16) |
((uniform_write_buffer[1] >> 16) & 0xFFFF));
uniform.y = float24::FromRaw(((uniform_write_buffer[1] & 0xFFFF) << 8) |
((uniform_write_buffer[2] >> 24) & 0xFF));
uniform.x = float24::FromRaw(uniform_write_buffer[2] & 0xFFFFFF);
}
LOG_TRACE(HW_GPU, "Set %s float uniform %x to (%f %f %f %f)",
GetShaderSetupTypeName(setup), (int)uniform_setup.index,
uniform.x.ToFloat32(), uniform.y.ToFloat32(), uniform.z.ToFloat32(),
uniform.w.ToFloat32());
// TODO: Verify that this actually modifies the register!
uniform_setup.index.Assign(uniform_setup.index + 1);
}
}
}
static void LoadDefaultVertexAttributes(u32 register_value) {
auto& regs = g_state.regs;
// TODO: Does actual hardware indeed keep an intermediate buffer or does
// it directly write the values?
default_attr_write_buffer[default_attr_counter++] = register_value;
// Default attributes are written in a packed format such that four float24 values are encoded
// in three 32-bit numbers.
// We write to internal memory once a full such vector is written.
if (default_attr_counter >= 3) {
default_attr_counter = 0;
auto& setup = regs.pipeline.vs_default_attributes_setup;
if (setup.index >= 16) {
LOG_ERROR(HW_GPU, "Invalid VS default attribute index %d", (int)setup.index);
return;
}
Math::Vec4<float24> attribute;
// NOTE: The destination component order indeed is "backwards"
attribute.w = float24::FromRaw(default_attr_write_buffer[0] >> 8);
attribute.z = float24::FromRaw(((default_attr_write_buffer[0] & 0xFF) << 16) |
((default_attr_write_buffer[1] >> 16) & 0xFFFF));
attribute.y = float24::FromRaw(((default_attr_write_buffer[1] & 0xFFFF) << 8) |
((default_attr_write_buffer[2] >> 24) & 0xFF));
attribute.x = float24::FromRaw(default_attr_write_buffer[2] & 0xFFFFFF);
LOG_TRACE(HW_GPU, "Set default VS attribute %x to (%f %f %f %f)", (int)setup.index,
attribute.x.ToFloat32(), attribute.y.ToFloat32(), attribute.z.ToFloat32(),
attribute.w.ToFloat32());
// TODO: Verify that this actually modifies the register!
if (setup.index < 15) {
g_state.input_default_attributes.attr[setup.index] = attribute;
setup.index++;
} else {
// Put each attribute into an immediate input buffer. When all specified immediate
// attributes are present, the Vertex Shader is invoked and everything is sent to
// the primitive assembler.
auto& immediate_input = g_state.immediate.input_vertex;
auto& immediate_attribute_id = g_state.immediate.current_attribute;
immediate_input.attr[immediate_attribute_id] = attribute;
if (immediate_attribute_id < regs.pipeline.max_input_attrib_index) {
immediate_attribute_id += 1;
} else {
MICROPROFILE_SCOPE(GPU_Drawing);
immediate_attribute_id = 0;
auto* shader_engine = Shader::GetEngine();
shader_engine->SetupBatch(g_state.vs, regs.vs.main_offset);
// Send to vertex shader
if (g_debug_context)
g_debug_context->OnEvent(DebugContext::Event::VertexShaderInvocation,
static_cast<void*>(&immediate_input));
Shader::UnitState shader_unit;
Shader::AttributeBuffer output{};
shader_unit.LoadInput(regs.vs, immediate_input);
shader_engine->Run(g_state.vs, shader_unit);
shader_unit.WriteOutput(regs.vs, output);
// Send to geometry pipeline
if (g_state.immediate.reset_geometry_pipeline) {
g_state.geometry_pipeline.Reconfigure();
g_state.immediate.reset_geometry_pipeline = false;
}
ASSERT(!g_state.geometry_pipeline.NeedIndexInput());
g_state.geometry_pipeline.Setup(shader_engine);
g_state.geometry_pipeline.SubmitVertex(output);
// TODO: If drawing after every immediate mode triangle kills performance,
// change it to flush triangles whenever a drawing config register changes
// See: https://github.com/citra-emu/citra/pull/2866#issuecomment-327011550
VideoCore::g_renderer->Rasterizer()->DrawTriangles();
if (g_debug_context) {
g_debug_context->OnEvent(DebugContext::Event::FinishedPrimitiveBatch, nullptr);
}
}
}
}
}
static void Draw(u32 command_id) {
MICROPROFILE_SCOPE(GPU_Drawing);
auto& regs = g_state.regs;
#if PICA_LOG_TEV
DebugUtils::DumpTevStageConfig(regs.GetTevStages());
#endif
if (g_debug_context)
g_debug_context->OnEvent(DebugContext::Event::IncomingPrimitiveBatch, nullptr);
// Processes information about internal vertex attributes to figure out how a vertex is
// loaded.
// Later, these can be compiled and cached.
const u32 base_address = regs.pipeline.vertex_attributes.GetPhysicalBaseAddress();
VertexLoader loader(regs.pipeline);
// Load vertices
bool is_indexed = (command_id == PICA_REG_INDEX(pipeline.trigger_draw_indexed));
const auto& index_info = regs.pipeline.index_array;
const u8* index_address_8 = Memory::GetPhysicalPointer(base_address + index_info.offset);
const u16* index_address_16 = reinterpret_cast<const u16*>(index_address_8);
bool index_u16 = index_info.format != 0;
PrimitiveAssembler<Shader::OutputVertex>& primitive_assembler = g_state.primitive_assembler;
if (g_debug_context && g_debug_context->recorder) {
for (int i = 0; i < 3; ++i) {
const auto texture = regs.texturing.GetTextures()[i];
if (!texture.enabled)
continue;
u8* texture_data = Memory::GetPhysicalPointer(texture.config.GetPhysicalAddress());
g_debug_context->recorder->MemoryAccessed(
texture_data, Pica::TexturingRegs::NibblesPerPixel(texture.format) *
texture.config.width / 2 * texture.config.height,
texture.config.GetPhysicalAddress());
}
}
DebugUtils::MemoryAccessTracker memory_accesses;
// Simple circular-replacement vertex cache
// The size has been tuned for optimal balance between hit-rate and the cost of lookup
const size_t VERTEX_CACHE_SIZE = 32;
std::array<u16, VERTEX_CACHE_SIZE> vertex_cache_ids;
std::array<Shader::AttributeBuffer, VERTEX_CACHE_SIZE> vertex_cache;
Shader::AttributeBuffer vs_output;
unsigned int vertex_cache_pos = 0;
vertex_cache_ids.fill(-1);
auto* shader_engine = Shader::GetEngine();
Shader::UnitState shader_unit;
shader_engine->SetupBatch(g_state.vs, regs.vs.main_offset);
g_state.geometry_pipeline.Reconfigure();
g_state.geometry_pipeline.Setup(shader_engine);
if (g_state.geometry_pipeline.NeedIndexInput())
ASSERT(is_indexed);
for (unsigned int index = 0; index < regs.pipeline.num_vertices; ++index) {
// Indexed rendering doesn't use the start offset
unsigned int vertex = is_indexed
? (index_u16 ? index_address_16[index] : index_address_8[index])
: (index + regs.pipeline.vertex_offset);
// -1 is a common special value used for primitive restart. Since it's unknown if
// the PICA supports it, and it would mess up the caching, guard against it here.
ASSERT(vertex != -1);
bool vertex_cache_hit = false;
if (is_indexed) {
if (g_state.geometry_pipeline.NeedIndexInput()) {
g_state.geometry_pipeline.SubmitIndex(vertex);
continue;
}
if (g_debug_context && Pica::g_debug_context->recorder) {
int size = index_u16 ? 2 : 1;
memory_accesses.AddAccess(base_address + index_info.offset + size * index, size);
}
for (unsigned int i = 0; i < VERTEX_CACHE_SIZE; ++i) {
if (vertex == vertex_cache_ids[i]) {
vs_output = vertex_cache[i];
vertex_cache_hit = true;
break;
}
}
}
if (!vertex_cache_hit) {
// Initialize data for the current vertex
Shader::AttributeBuffer input;
loader.LoadVertex(base_address, index, vertex, input, memory_accesses);
// Send to vertex shader
if (g_debug_context)
g_debug_context->OnEvent(DebugContext::Event::VertexShaderInvocation,
(void*)&input);
shader_unit.LoadInput(regs.vs, input);
shader_engine->Run(g_state.vs, shader_unit);
shader_unit.WriteOutput(regs.vs, vs_output);
if (is_indexed) {
vertex_cache[vertex_cache_pos] = vs_output;
vertex_cache_ids[vertex_cache_pos] = vertex;
vertex_cache_pos = (vertex_cache_pos + 1) % VERTEX_CACHE_SIZE;
}
}
// Send to geometry pipeline
g_state.geometry_pipeline.SubmitVertex(vs_output);
}
for (auto& range : memory_accesses.ranges) {
g_debug_context->recorder->MemoryAccessed(Memory::GetPhysicalPointer(range.first),
range.second, range.first);
}
VideoCore::g_renderer->Rasterizer()->DrawTriangles();
if (g_debug_context) {
g_debug_context->OnEvent(DebugContext::Event::FinishedPrimitiveBatch, nullptr);
}
}
static void WritePicaReg(u32 id, u32 value, u32 mask) {
auto& regs = g_state.regs;
if (id >= Regs::NUM_REGS) {
LOG_ERROR(HW_GPU,
"Commandlist tried to write to invalid register 0x%03X (value: %08X, mask: %X)",
id, value, mask);
return;
}
// TODO: Figure out how register masking acts on e.g. vs.uniform_setup.set_value
u32 old_value = regs.reg_array[id];
const u32 write_mask = expand_bits_to_bytes[mask];
regs.reg_array[id] = (old_value & ~write_mask) | (value & write_mask);
// Double check for is_pica_tracing to avoid call overhead
if (DebugUtils::IsPicaTracing()) {
DebugUtils::OnPicaRegWrite({(u16)id, (u16)mask, regs.reg_array[id]});
}
if (g_debug_context)
g_debug_context->OnEvent(DebugContext::Event::PicaCommandLoaded,
reinterpret_cast<void*>(&id));
switch (id) {
// Trigger IRQ
case PICA_REG_INDEX(trigger_irq):
//Service::GSP::SignalInterrupt(Service::GSP::InterruptId::P3D);
break;
case PICA_REG_INDEX(pipeline.triangle_topology):
g_state.primitive_assembler.Reconfigure(regs.pipeline.triangle_topology);
break;
case PICA_REG_INDEX(pipeline.restart_primitive):
g_state.primitive_assembler.Reset();
break;
case PICA_REG_INDEX(pipeline.vs_default_attributes_setup.index):
g_state.immediate.current_attribute = 0;
g_state.immediate.reset_geometry_pipeline = true;
default_attr_counter = 0;
break;
// Load default vertex input attributes
case PICA_REG_INDEX_WORKAROUND(pipeline.vs_default_attributes_setup.set_value[0], 0x233):
case PICA_REG_INDEX_WORKAROUND(pipeline.vs_default_attributes_setup.set_value[1], 0x234):
case PICA_REG_INDEX_WORKAROUND(pipeline.vs_default_attributes_setup.set_value[2], 0x235):
LoadDefaultVertexAttributes(value);
break;
case PICA_REG_INDEX(pipeline.gpu_mode):
// This register likely just enables vertex processing and doesn't need any special handling
break;
case PICA_REG_INDEX_WORKAROUND(pipeline.command_buffer.trigger[0], 0x23c):
case PICA_REG_INDEX_WORKAROUND(pipeline.command_buffer.trigger[1], 0x23d): {
unsigned index =
static_cast<unsigned>(id - PICA_REG_INDEX(pipeline.command_buffer.trigger[0]));
u32* head_ptr = (u32*)Memory::GetPhysicalPointer(
regs.pipeline.command_buffer.GetPhysicalAddress(index));
g_state.cmd_list.head_ptr = g_state.cmd_list.current_ptr = head_ptr;
g_state.cmd_list.length = regs.pipeline.command_buffer.GetSize(index) / sizeof(u32);
break;
}
// It seems like these trigger vertex rendering
case PICA_REG_INDEX(pipeline.trigger_draw):
case PICA_REG_INDEX(pipeline.trigger_draw_indexed):
Draw(id);
break;
case PICA_REG_INDEX(gs.bool_uniforms):
WriteUniformBoolReg(g_state.gs, g_state.regs.gs.bool_uniforms.Value());
break;
case PICA_REG_INDEX_WORKAROUND(gs.int_uniforms[0], 0x281):
case PICA_REG_INDEX_WORKAROUND(gs.int_uniforms[1], 0x282):
case PICA_REG_INDEX_WORKAROUND(gs.int_uniforms[2], 0x283):
case PICA_REG_INDEX_WORKAROUND(gs.int_uniforms[3], 0x284): {
unsigned index = (id - PICA_REG_INDEX_WORKAROUND(gs.int_uniforms[0], 0x281));
auto values = regs.gs.int_uniforms[index];
WriteUniformIntReg(g_state.gs, index,
Math::Vec4<u8>(values.x, values.y, values.z, values.w));
break;
}
case PICA_REG_INDEX_WORKAROUND(gs.uniform_setup.set_value[0], 0x291):
case PICA_REG_INDEX_WORKAROUND(gs.uniform_setup.set_value[1], 0x292):
case PICA_REG_INDEX_WORKAROUND(gs.uniform_setup.set_value[2], 0x293):
case PICA_REG_INDEX_WORKAROUND(gs.uniform_setup.set_value[3], 0x294):
case PICA_REG_INDEX_WORKAROUND(gs.uniform_setup.set_value[4], 0x295):
case PICA_REG_INDEX_WORKAROUND(gs.uniform_setup.set_value[5], 0x296):
case PICA_REG_INDEX_WORKAROUND(gs.uniform_setup.set_value[6], 0x297):
case PICA_REG_INDEX_WORKAROUND(gs.uniform_setup.set_value[7], 0x298): {
WriteUniformFloatReg(g_state.regs.gs, g_state.gs, gs_float_regs_counter,
gs_uniform_write_buffer, value);
break;
}
case PICA_REG_INDEX_WORKAROUND(gs.program.set_word[0], 0x29c):
case PICA_REG_INDEX_WORKAROUND(gs.program.set_word[1], 0x29d):
case PICA_REG_INDEX_WORKAROUND(gs.program.set_word[2], 0x29e):
case PICA_REG_INDEX_WORKAROUND(gs.program.set_word[3], 0x29f):
case PICA_REG_INDEX_WORKAROUND(gs.program.set_word[4], 0x2a0):
case PICA_REG_INDEX_WORKAROUND(gs.program.set_word[5], 0x2a1):
case PICA_REG_INDEX_WORKAROUND(gs.program.set_word[6], 0x2a2):
case PICA_REG_INDEX_WORKAROUND(gs.program.set_word[7], 0x2a3): {
u32& offset = g_state.regs.gs.program.offset;
if (offset >= 4096) {
LOG_ERROR(HW_GPU, "Invalid GS program offset %u", offset);
} else {
g_state.gs.program_code[offset] = value;
offset++;
}
break;
}
case PICA_REG_INDEX_WORKAROUND(gs.swizzle_patterns.set_word[0], 0x2a6):
case PICA_REG_INDEX_WORKAROUND(gs.swizzle_patterns.set_word[1], 0x2a7):
case PICA_REG_INDEX_WORKAROUND(gs.swizzle_patterns.set_word[2], 0x2a8):
case PICA_REG_INDEX_WORKAROUND(gs.swizzle_patterns.set_word[3], 0x2a9):
case PICA_REG_INDEX_WORKAROUND(gs.swizzle_patterns.set_word[4], 0x2aa):
case PICA_REG_INDEX_WORKAROUND(gs.swizzle_patterns.set_word[5], 0x2ab):
case PICA_REG_INDEX_WORKAROUND(gs.swizzle_patterns.set_word[6], 0x2ac):
case PICA_REG_INDEX_WORKAROUND(gs.swizzle_patterns.set_word[7], 0x2ad): {
u32& offset = g_state.regs.gs.swizzle_patterns.offset;
if (offset >= g_state.gs.swizzle_data.size()) {
LOG_ERROR(HW_GPU, "Invalid GS swizzle pattern offset %u", offset);
} else {
g_state.gs.swizzle_data[offset] = value;
offset++;
}
break;
}
case PICA_REG_INDEX(vs.bool_uniforms):
// TODO (wwylele): does regs.pipeline.gs_unit_exclusive_configuration affect this?
WriteUniformBoolReg(g_state.vs, g_state.regs.vs.bool_uniforms.Value());
break;
case PICA_REG_INDEX_WORKAROUND(vs.int_uniforms[0], 0x2b1):
case PICA_REG_INDEX_WORKAROUND(vs.int_uniforms[1], 0x2b2):
case PICA_REG_INDEX_WORKAROUND(vs.int_uniforms[2], 0x2b3):
case PICA_REG_INDEX_WORKAROUND(vs.int_uniforms[3], 0x2b4): {
// TODO (wwylele): does regs.pipeline.gs_unit_exclusive_configuration affect this?
unsigned index = (id - PICA_REG_INDEX_WORKAROUND(vs.int_uniforms[0], 0x2b1));
auto values = regs.vs.int_uniforms[index];
WriteUniformIntReg(g_state.vs, index,
Math::Vec4<u8>(values.x, values.y, values.z, values.w));
break;
}
case PICA_REG_INDEX_WORKAROUND(vs.uniform_setup.set_value[0], 0x2c1):
case PICA_REG_INDEX_WORKAROUND(vs.uniform_setup.set_value[1], 0x2c2):
case PICA_REG_INDEX_WORKAROUND(vs.uniform_setup.set_value[2], 0x2c3):
case PICA_REG_INDEX_WORKAROUND(vs.uniform_setup.set_value[3], 0x2c4):
case PICA_REG_INDEX_WORKAROUND(vs.uniform_setup.set_value[4], 0x2c5):
case PICA_REG_INDEX_WORKAROUND(vs.uniform_setup.set_value[5], 0x2c6):
case PICA_REG_INDEX_WORKAROUND(vs.uniform_setup.set_value[6], 0x2c7):
case PICA_REG_INDEX_WORKAROUND(vs.uniform_setup.set_value[7], 0x2c8): {
// TODO (wwylele): does regs.pipeline.gs_unit_exclusive_configuration affect this?
WriteUniformFloatReg(g_state.regs.vs, g_state.vs, vs_float_regs_counter,
vs_uniform_write_buffer, value);
break;
}
case PICA_REG_INDEX_WORKAROUND(vs.program.set_word[0], 0x2cc):
case PICA_REG_INDEX_WORKAROUND(vs.program.set_word[1], 0x2cd):
case PICA_REG_INDEX_WORKAROUND(vs.program.set_word[2], 0x2ce):
case PICA_REG_INDEX_WORKAROUND(vs.program.set_word[3], 0x2cf):
case PICA_REG_INDEX_WORKAROUND(vs.program.set_word[4], 0x2d0):
case PICA_REG_INDEX_WORKAROUND(vs.program.set_word[5], 0x2d1):
case PICA_REG_INDEX_WORKAROUND(vs.program.set_word[6], 0x2d2):
case PICA_REG_INDEX_WORKAROUND(vs.program.set_word[7], 0x2d3): {
u32& offset = g_state.regs.vs.program.offset;
if (offset >= 512) {
LOG_ERROR(HW_GPU, "Invalid VS program offset %u", offset);
} else {
g_state.vs.program_code[offset] = value;
if (!g_state.regs.pipeline.gs_unit_exclusive_configuration) {
g_state.gs.program_code[offset] = value;
}
offset++;
}
break;
}
case PICA_REG_INDEX_WORKAROUND(vs.swizzle_patterns.set_word[0], 0x2d6):
case PICA_REG_INDEX_WORKAROUND(vs.swizzle_patterns.set_word[1], 0x2d7):
case PICA_REG_INDEX_WORKAROUND(vs.swizzle_patterns.set_word[2], 0x2d8):
case PICA_REG_INDEX_WORKAROUND(vs.swizzle_patterns.set_word[3], 0x2d9):
case PICA_REG_INDEX_WORKAROUND(vs.swizzle_patterns.set_word[4], 0x2da):
case PICA_REG_INDEX_WORKAROUND(vs.swizzle_patterns.set_word[5], 0x2db):
case PICA_REG_INDEX_WORKAROUND(vs.swizzle_patterns.set_word[6], 0x2dc):
case PICA_REG_INDEX_WORKAROUND(vs.swizzle_patterns.set_word[7], 0x2dd): {
u32& offset = g_state.regs.vs.swizzle_patterns.offset;
if (offset >= g_state.vs.swizzle_data.size()) {
LOG_ERROR(HW_GPU, "Invalid VS swizzle pattern offset %u", offset);
} else {
g_state.vs.swizzle_data[offset] = value;
if (!g_state.regs.pipeline.gs_unit_exclusive_configuration) {
g_state.gs.swizzle_data[offset] = value;
}
offset++;
}
break;
}
case PICA_REG_INDEX_WORKAROUND(lighting.lut_data[0], 0x1c8):
case PICA_REG_INDEX_WORKAROUND(lighting.lut_data[1], 0x1c9):
case PICA_REG_INDEX_WORKAROUND(lighting.lut_data[2], 0x1ca):
case PICA_REG_INDEX_WORKAROUND(lighting.lut_data[3], 0x1cb):
case PICA_REG_INDEX_WORKAROUND(lighting.lut_data[4], 0x1cc):
case PICA_REG_INDEX_WORKAROUND(lighting.lut_data[5], 0x1cd):
case PICA_REG_INDEX_WORKAROUND(lighting.lut_data[6], 0x1ce):
case PICA_REG_INDEX_WORKAROUND(lighting.lut_data[7], 0x1cf): {
auto& lut_config = regs.lighting.lut_config;
ASSERT_MSG(lut_config.index < 256, "lut_config.index exceeded maximum value of 255!");
g_state.lighting.luts[lut_config.type][lut_config.index].raw = value;
lut_config.index.Assign(lut_config.index + 1);
break;
}
case PICA_REG_INDEX_WORKAROUND(texturing.fog_lut_data[0], 0xe8):
case PICA_REG_INDEX_WORKAROUND(texturing.fog_lut_data[1], 0xe9):
case PICA_REG_INDEX_WORKAROUND(texturing.fog_lut_data[2], 0xea):
case PICA_REG_INDEX_WORKAROUND(texturing.fog_lut_data[3], 0xeb):
case PICA_REG_INDEX_WORKAROUND(texturing.fog_lut_data[4], 0xec):
case PICA_REG_INDEX_WORKAROUND(texturing.fog_lut_data[5], 0xed):
case PICA_REG_INDEX_WORKAROUND(texturing.fog_lut_data[6], 0xee):
case PICA_REG_INDEX_WORKAROUND(texturing.fog_lut_data[7], 0xef): {
g_state.fog.lut[regs.texturing.fog_lut_offset % 128].raw = value;
regs.texturing.fog_lut_offset.Assign(regs.texturing.fog_lut_offset + 1);
break;
}
case PICA_REG_INDEX_WORKAROUND(texturing.proctex_lut_data[0], 0xb0):
case PICA_REG_INDEX_WORKAROUND(texturing.proctex_lut_data[1], 0xb1):
case PICA_REG_INDEX_WORKAROUND(texturing.proctex_lut_data[2], 0xb2):
case PICA_REG_INDEX_WORKAROUND(texturing.proctex_lut_data[3], 0xb3):
case PICA_REG_INDEX_WORKAROUND(texturing.proctex_lut_data[4], 0xb4):
case PICA_REG_INDEX_WORKAROUND(texturing.proctex_lut_data[5], 0xb5):
case PICA_REG_INDEX_WORKAROUND(texturing.proctex_lut_data[6], 0xb6):
case PICA_REG_INDEX_WORKAROUND(texturing.proctex_lut_data[7], 0xb7): {
auto& index = regs.texturing.proctex_lut_config.index;
auto& pt = g_state.proctex;
switch (regs.texturing.proctex_lut_config.ref_table.Value()) {
case TexturingRegs::ProcTexLutTable::Noise:
pt.noise_table[index % pt.noise_table.size()].raw = value;
break;
case TexturingRegs::ProcTexLutTable::ColorMap:
pt.color_map_table[index % pt.color_map_table.size()].raw = value;
break;
case TexturingRegs::ProcTexLutTable::AlphaMap:
pt.alpha_map_table[index % pt.alpha_map_table.size()].raw = value;
break;
case TexturingRegs::ProcTexLutTable::Color:
pt.color_table[index % pt.color_table.size()].raw = value;
break;
case TexturingRegs::ProcTexLutTable::ColorDiff:
pt.color_diff_table[index % pt.color_diff_table.size()].raw = value;
break;
}
index.Assign(index + 1);
break;
}
default:
break;
}
VideoCore::g_renderer->Rasterizer()->NotifyPicaRegisterChanged(id);
if (g_debug_context)
g_debug_context->OnEvent(DebugContext::Event::PicaCommandProcessed,
reinterpret_cast<void*>(&id));
}
void ProcessCommandList(const u32* list, u32 size) {
g_state.cmd_list.head_ptr = g_state.cmd_list.current_ptr = list;
g_state.cmd_list.length = size / sizeof(u32);
while (g_state.cmd_list.current_ptr < g_state.cmd_list.head_ptr + g_state.cmd_list.length) {
// Align read pointer to 8 bytes
if ((g_state.cmd_list.head_ptr - g_state.cmd_list.current_ptr) % 2 != 0)
++g_state.cmd_list.current_ptr;
u32 value = *g_state.cmd_list.current_ptr++;
const CommandHeader header = {*g_state.cmd_list.current_ptr++};
WritePicaReg(header.cmd_id, value, header.parameter_mask);
for (unsigned i = 0; i < header.extra_data_length; ++i) {
u32 cmd = header.cmd_id + (header.group_commands ? i + 1 : 0);
WritePicaReg(cmd, *g_state.cmd_list.current_ptr++, header.parameter_mask);
}
}
}
} // namespace CommandProcessor
} // namespace Pica

41
src/video_core/command_processor.h
View file

@ -1,41 +0,0 @@
// Copyright 2014 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <type_traits>
#include "common/bit_field.h"
#include "common/common_types.h"
namespace Pica {
namespace CommandProcessor {
union CommandHeader {
u32 hex;
BitField<0, 16, u32> cmd_id;
// parameter_mask:
// Mask applied to the input value to make it possible to update
// parts of a register without overwriting its other fields.
// first bit: 0x000000FF
// second bit: 0x0000FF00
// third bit: 0x00FF0000
// fourth bit: 0xFF000000
BitField<16, 4, u32> parameter_mask;
BitField<20, 11, u32> extra_data_length;
BitField<31, 1, u32> group_commands;
};
static_assert(std::is_standard_layout<CommandHeader>::value == true,
"CommandHeader does not use standard layout");
static_assert(sizeof(CommandHeader) == sizeof(u32), "CommandHeader has incorrect size!");
void ProcessCommandList(const u32* list, u32 size);
} // namespace
} // namespace

577
src/video_core/debug_utils/debug_utils.cpp
View file

@ -1,577 +0,0 @@
// Copyright 2014 Citra Emulator Project
// Licensed under GPLv2
// Refer to the license.txt file included.
#include <algorithm>
#include <condition_variable>
#include <cstdint>
#include <cstring>
#include <fstream>
#include <map>
#include <mutex>
#include <stdexcept>
#include <string>
#ifdef HAVE_PNG
#include <png.h>
#include <setjmp.h>
#endif
#include <nihstro/bit_field.h>
#include <nihstro/float24.h>
#include <nihstro/shader_binary.h>
#include "common/assert.h"
#include "common/bit_field.h"
#include "common/color.h"
#include "common/common_types.h"
#include "common/file_util.h"
#include "common/logging/log.h"
#include "common/math_util.h"
#include "common/vector_math.h"
#include "video_core/debug_utils/debug_utils.h"
#include "video_core/pica_state.h"
#include "video_core/pica_types.h"
#include "video_core/rasterizer_interface.h"
#include "video_core/regs_rasterizer.h"
#include "video_core/regs_shader.h"
#include "video_core/regs_texturing.h"
#include "video_core/renderer_base.h"
#include "video_core/shader/shader.h"
#include "video_core/texture/texture_decode.h"
#include "video_core/utils.h"
#include "video_core/video_core.h"
using nihstro::DVLBHeader;
using nihstro::DVLEHeader;
using nihstro::DVLPHeader;
namespace Pica {
void DebugContext::DoOnEvent(Event event, void* data) {
{
std::unique_lock<std::mutex> lock(breakpoint_mutex);
// Commit the rasterizer's caches so framebuffers, render targets, etc. will show on debug
// widgets
VideoCore::g_renderer->Rasterizer()->FlushAll();
// TODO: Should stop the CPU thread here once we multithread emulation.
active_breakpoint = event;
at_breakpoint = true;
// Tell all observers that we hit a breakpoint
for (auto& breakpoint_observer : breakpoint_observers) {
breakpoint_observer->OnPicaBreakPointHit(event, data);
}
// Wait until another thread tells us to Resume()
resume_from_breakpoint.wait(lock, [&] { return !at_breakpoint; });
}
}
void DebugContext::Resume() {
{
std::lock_guard<std::mutex> lock(breakpoint_mutex);
// Tell all observers that we are about to resume
for (auto& breakpoint_observer : breakpoint_observers) {
breakpoint_observer->OnPicaResume();
}
// Resume the waiting thread (i.e. OnEvent())
at_breakpoint = false;
}
resume_from_breakpoint.notify_one();
}
std::shared_ptr<DebugContext> g_debug_context; // TODO: Get rid of this global
namespace DebugUtils {
void DumpShader(const std::string& filename, const ShaderRegs& config,
const Shader::ShaderSetup& setup,
const RasterizerRegs::VSOutputAttributes* output_attributes) {
struct StuffToWrite {
const u8* pointer;
u32 size;
};
std::vector<StuffToWrite> writing_queue;
u32 write_offset = 0;
auto QueueForWriting = [&writing_queue, &write_offset](const u8* pointer, u32 size) {
writing_queue.push_back({pointer, size});
u32 old_write_offset = write_offset;
write_offset += size;
return old_write_offset;
};
// First off, try to translate Pica state (one enum for output attribute type and component)
// into shbin format (separate type and component mask).
union OutputRegisterInfo {
enum Type : u64 {
POSITION = 0,
QUATERNION = 1,
COLOR = 2,
TEXCOORD0 = 3,
TEXCOORD1 = 5,
TEXCOORD2 = 6,
VIEW = 8,
};
BitField<0, 64, u64> hex;
BitField<0, 16, Type> type;
BitField<16, 16, u64> id;
BitField<32, 4, u64> component_mask;
};
// This is put into a try-catch block to make sure we notice unknown configurations.
std::vector<OutputRegisterInfo> output_info_table;
for (unsigned i = 0; i < 7; ++i) {
using OutputAttributes = Pica::RasterizerRegs::VSOutputAttributes;
// TODO: It's still unclear how the attribute components map to the register!
// Once we know that, this code probably will not make much sense anymore.
std::map<OutputAttributes::Semantic, std::pair<OutputRegisterInfo::Type, u32>> map = {
{OutputAttributes::POSITION_X, {OutputRegisterInfo::POSITION, 1}},
{OutputAttributes::POSITION_Y, {OutputRegisterInfo::POSITION, 2}},
{OutputAttributes::POSITION_Z, {OutputRegisterInfo::POSITION, 4}},
{OutputAttributes::POSITION_W, {OutputRegisterInfo::POSITION, 8}},
{OutputAttributes::QUATERNION_X, {OutputRegisterInfo::QUATERNION, 1}},
{OutputAttributes::QUATERNION_Y, {OutputRegisterInfo::QUATERNION, 2}},
{OutputAttributes::QUATERNION_Z, {OutputRegisterInfo::QUATERNION, 4}},
{OutputAttributes::QUATERNION_W, {OutputRegisterInfo::QUATERNION, 8}},
{OutputAttributes::COLOR_R, {OutputRegisterInfo::COLOR, 1}},
{OutputAttributes::COLOR_G, {OutputRegisterInfo::COLOR, 2}},
{OutputAttributes::COLOR_B, {OutputRegisterInfo::COLOR, 4}},
{OutputAttributes::COLOR_A, {OutputRegisterInfo::COLOR, 8}},
{OutputAttributes::TEXCOORD0_U, {OutputRegisterInfo::TEXCOORD0, 1}},
{OutputAttributes::TEXCOORD0_V, {OutputRegisterInfo::TEXCOORD0, 2}},
{OutputAttributes::TEXCOORD1_U, {OutputRegisterInfo::TEXCOORD1, 1}},
{OutputAttributes::TEXCOORD1_V, {OutputRegisterInfo::TEXCOORD1, 2}},
{OutputAttributes::TEXCOORD2_U, {OutputRegisterInfo::TEXCOORD2, 1}},
{OutputAttributes::TEXCOORD2_V, {OutputRegisterInfo::TEXCOORD2, 2}},
{OutputAttributes::VIEW_X, {OutputRegisterInfo::VIEW, 1}},
{OutputAttributes::VIEW_Y, {OutputRegisterInfo::VIEW, 2}},
{OutputAttributes::VIEW_Z, {OutputRegisterInfo::VIEW, 4}},
};
for (const auto& semantic : std::vector<OutputAttributes::Semantic>{
output_attributes[i].map_x, output_attributes[i].map_y, output_attributes[i].map_z,
output_attributes[i].map_w}) {
if (semantic == OutputAttributes::INVALID)
continue;
try {
OutputRegisterInfo::Type type = map.at(semantic).first;
u32 component_mask = map.at(semantic).second;
auto it = std::find_if(output_info_table.begin(), output_info_table.end(),
[&i, &type](const OutputRegisterInfo& info) {
return info.id == i && info.type == type;
});
if (it == output_info_table.end()) {
output_info_table.emplace_back();
output_info_table.back().type.Assign(type);
output_info_table.back().component_mask.Assign(component_mask);
output_info_table.back().id.Assign(i);
} else {
it->component_mask.Assign(it->component_mask | component_mask);
}
} catch (const std::out_of_range&) {
DEBUG_ASSERT_MSG(false, "Unknown output attribute mapping");
LOG_ERROR(HW_GPU, "Unknown output attribute mapping: %03x, %03x, %03x, %03x",
(int)output_attributes[i].map_x.Value(),
(int)output_attributes[i].map_y.Value(),
(int)output_attributes[i].map_z.Value(),
(int)output_attributes[i].map_w.Value());
}
}
}
struct {
DVLBHeader header;
u32 dvle_offset;
} dvlb{{DVLBHeader::MAGIC_WORD, 1}}; // 1 DVLE
DVLPHeader dvlp{DVLPHeader::MAGIC_WORD};
DVLEHeader dvle{DVLEHeader::MAGIC_WORD};
QueueForWriting(reinterpret_cast<const u8*>(&dvlb), sizeof(dvlb));
u32 dvlp_offset = QueueForWriting(reinterpret_cast<const u8*>(&dvlp), sizeof(dvlp));
dvlb.dvle_offset = QueueForWriting(reinterpret_cast<const u8*>(&dvle), sizeof(dvle));
// TODO: Reduce the amount of binary code written to relevant portions
dvlp.binary_offset = write_offset - dvlp_offset;
dvlp.binary_size_words = static_cast<uint32_t>(setup.program_code.size());
QueueForWriting(reinterpret_cast<const u8*>(setup.program_code.data()),
static_cast<u32>(setup.program_code.size()) * sizeof(u32));
dvlp.swizzle_info_offset = write_offset - dvlp_offset;
dvlp.swizzle_info_num_entries = static_cast<uint32_t>(setup.swizzle_data.size());
u32 dummy = 0;
for (unsigned int i = 0; i < setup.swizzle_data.size(); ++i) {
QueueForWriting(reinterpret_cast<const u8*>(&setup.swizzle_data[i]),
sizeof(setup.swizzle_data[i]));
QueueForWriting(reinterpret_cast<const u8*>(&dummy), sizeof(dummy));
}
dvle.main_offset_words = config.main_offset;
dvle.output_register_table_offset = write_offset - dvlb.dvle_offset;
dvle.output_register_table_size = static_cast<u32>(output_info_table.size());
QueueForWriting(reinterpret_cast<const u8*>(output_info_table.data()),
static_cast<u32>(output_info_table.size() * sizeof(OutputRegisterInfo)));
// TODO: Create a label table for "main"
std::vector<nihstro::ConstantInfo> constant_table;
for (unsigned i = 0; i < setup.uniforms.b.size(); ++i) {
nihstro::ConstantInfo constant;
memset(&constant, 0, sizeof(constant));
constant.type = nihstro::ConstantInfo::Bool;
constant.regid = i;
constant.b = setup.uniforms.b[i];
constant_table.emplace_back(constant);
}
for (unsigned i = 0; i < setup.uniforms.i.size(); ++i) {
nihstro::ConstantInfo constant;
memset(&constant, 0, sizeof(constant));
constant.type = nihstro::ConstantInfo::Int;
constant.regid = i;
constant.i.x = setup.uniforms.i[i].x;
constant.i.y = setup.uniforms.i[i].y;
constant.i.z = setup.uniforms.i[i].z;
constant.i.w = setup.uniforms.i[i].w;
constant_table.emplace_back(constant);
}
for (unsigned i = 0; i < sizeof(setup.uniforms.f) / sizeof(setup.uniforms.f[0]); ++i) {
nihstro::ConstantInfo constant;
memset(&constant, 0, sizeof(constant));
constant.type = nihstro::ConstantInfo::Float;
constant.regid = i;
constant.f.x = nihstro::to_float24(setup.uniforms.f[i].x.ToFloat32());
constant.f.y = nihstro::to_float24(setup.uniforms.f[i].y.ToFloat32());
constant.f.z = nihstro::to_float24(setup.uniforms.f[i].z.ToFloat32());
constant.f.w = nihstro::to_float24(setup.uniforms.f[i].w.ToFloat32());
// Store constant if it's different from zero..
if (setup.uniforms.f[i].x.ToFloat32() != 0.0 || setup.uniforms.f[i].y.ToFloat32() != 0.0 ||
setup.uniforms.f[i].z.ToFloat32() != 0.0 || setup.uniforms.f[i].w.ToFloat32() != 0.0)
constant_table.emplace_back(constant);
}
dvle.constant_table_offset = write_offset - dvlb.dvle_offset;
dvle.constant_table_size = static_cast<uint32_t>(constant_table.size());
for (const auto& constant : constant_table) {
QueueForWriting(reinterpret_cast<const u8*>(&constant), sizeof(constant));
}
// Write data to file
std::ofstream file(filename, std::ios_base::out | std::ios_base::binary);
for (const auto& chunk : writing_queue) {
file.write(reinterpret_cast<const char*>(chunk.pointer), chunk.size);
}
}
static std::unique_ptr<PicaTrace> pica_trace;
static std::mutex pica_trace_mutex;
bool g_is_pica_tracing = false;
void StartPicaTracing() {
if (g_is_pica_tracing) {
LOG_WARNING(HW_GPU, "StartPicaTracing called even though tracing already running!");
return;
}
std::lock_guard<std::mutex> lock(pica_trace_mutex);
pica_trace = std::make_unique<PicaTrace>();
g_is_pica_tracing = true;
}
void OnPicaRegWrite(PicaTrace::Write write) {
std::lock_guard<std::mutex> lock(pica_trace_mutex);
if (!g_is_pica_tracing)
return;
pica_trace->writes.push_back(write);
}
std::unique_ptr<PicaTrace> FinishPicaTracing() {
if (!g_is_pica_tracing) {
LOG_WARNING(HW_GPU, "FinishPicaTracing called even though tracing isn't running!");
return {};
}
// signalize that no further tracing should be performed
g_is_pica_tracing = false;
// Wait until running tracing is finished
std::lock_guard<std::mutex> lock(pica_trace_mutex);
std::unique_ptr<PicaTrace> ret(std::move(pica_trace));
return ret;
}
#ifdef HAVE_PNG
// Adapter functions to libpng to write/flush to File::IOFile instances.
static void WriteIOFile(png_structp png_ptr, png_bytep data, png_size_t length) {
auto* fp = static_cast<FileUtil::IOFile*>(png_get_io_ptr(png_ptr));
if (!fp->WriteBytes(data, length))
png_error(png_ptr, "Failed to write to output PNG file.");
}
static void FlushIOFile(png_structp png_ptr) {
auto* fp = static_cast<FileUtil::IOFile*>(png_get_io_ptr(png_ptr));
if (!fp->Flush())
png_error(png_ptr, "Failed to flush to output PNG file.");
}
#endif
void DumpTexture(const TexturingRegs::TextureConfig& texture_config, u8* data) {
#ifndef HAVE_PNG
return;
#else
if (!data)
return;
// Write data to file
static int dump_index = 0;
std::string filename =
std::string("texture_dump") + std::to_string(++dump_index) + std::string(".png");
u32 row_stride = texture_config.width * 3;
u8* buf;
char title[] = "Citra texture dump";
char title_key[] = "Title";
png_structp png_ptr = nullptr;
png_infop info_ptr = nullptr;
// Open file for writing (binary mode)
FileUtil::IOFile fp(filename, "wb");
// Initialize write structure
png_ptr = png_create_write_struct(PNG_LIBPNG_VER_STRING, nullptr, nullptr, nullptr);
if (png_ptr == nullptr) {
LOG_ERROR(Debug_GPU, "Could not allocate write struct");
goto finalise;
}
// Initialize info structure
info_ptr = png_create_info_struct(png_ptr);
if (info_ptr == nullptr) {
LOG_ERROR(Debug_GPU, "Could not allocate info struct");
goto finalise;
}
// Setup Exception handling
if (setjmp(png_jmpbuf(png_ptr))) {
LOG_ERROR(Debug_GPU, "Error during png creation");
goto finalise;
}
png_set_write_fn(png_ptr, static_cast<void*>(&fp), WriteIOFile, FlushIOFile);
// Write header (8 bit color depth)
png_set_IHDR(png_ptr, info_ptr, texture_config.width, texture_config.height, 8,
PNG_COLOR_TYPE_RGB /*_ALPHA*/, PNG_INTERLACE_NONE, PNG_COMPRESSION_TYPE_BASE,
PNG_FILTER_TYPE_BASE);
png_text title_text;
title_text.compression = PNG_TEXT_COMPRESSION_NONE;
title_text.key = title_key;
title_text.text = title;
png_set_text(png_ptr, info_ptr, &title_text, 1);
png_write_info(png_ptr, info_ptr);
buf = new u8[row_stride * texture_config.height];
for (unsigned y = 0; y < texture_config.height; ++y) {
for (unsigned x = 0; x < texture_config.width; ++x) {
Pica::Texture::TextureInfo info;
info.width = texture_config.width;
info.height = texture_config.height;
info.stride = row_stride;
info.format = g_state.regs.texturing.texture0_format;
Math::Vec4<u8> texture_color = Pica::Texture::LookupTexture(data, x, y, info);
buf[3 * x + y * row_stride] = texture_color.r();
buf[3 * x + y * row_stride + 1] = texture_color.g();
buf[3 * x + y * row_stride + 2] = texture_color.b();
}
}
// Write image data
for (unsigned y = 0; y < texture_config.height; ++y) {
u8* row_ptr = (u8*)buf + y * row_stride;
png_write_row(png_ptr, row_ptr);
}
delete[] buf;
// End write
png_write_end(png_ptr, nullptr);
finalise:
if (info_ptr != nullptr)
png_free_data(png_ptr, info_ptr, PNG_FREE_ALL, -1);
if (png_ptr != nullptr)
png_destroy_write_struct(&png_ptr, (png_infopp) nullptr);
#endif
}
static std::string ReplacePattern(const std::string& input, const std::string& pattern,
const std::string& replacement) {
size_t start = input.find(pattern);
if (start == std::string::npos)
return input;
std::string ret = input;
ret.replace(start, pattern.length(), replacement);
return ret;
}
static std::string GetTevStageConfigSourceString(
const TexturingRegs::TevStageConfig::Source& source) {
using Source = TexturingRegs::TevStageConfig::Source;
static const std::map<Source, std::string> source_map = {
{Source::PrimaryColor, "PrimaryColor"},
{Source::PrimaryFragmentColor, "PrimaryFragmentColor"},
{Source::SecondaryFragmentColor, "SecondaryFragmentColor"},
{Source::Texture0, "Texture0"},
{Source::Texture1, "Texture1"},
{Source::Texture2, "Texture2"},
{Source::Texture3, "Texture3"},
{Source::PreviousBuffer, "PreviousBuffer"},
{Source::Constant, "Constant"},
{Source::Previous, "Previous"},
};
const auto src_it = source_map.find(source);
if (src_it == source_map.end())
return "Unknown";
return src_it->second;
}
static std::string GetTevStageConfigColorSourceString(
const TexturingRegs::TevStageConfig::Source& source,
const TexturingRegs::TevStageConfig::ColorModifier modifier) {
using ColorModifier = TexturingRegs::TevStageConfig::ColorModifier;
static const std::map<ColorModifier, std::string> color_modifier_map = {
{ColorModifier::SourceColor, "%source.rgb"},
{ColorModifier::OneMinusSourceColor, "(1.0 - %source.rgb)"},
{ColorModifier::SourceAlpha, "%source.aaa"},
{ColorModifier::OneMinusSourceAlpha, "(1.0 - %source.aaa)"},
{ColorModifier::SourceRed, "%source.rrr"},
{ColorModifier::OneMinusSourceRed, "(1.0 - %source.rrr)"},
{ColorModifier::SourceGreen, "%source.ggg"},
{ColorModifier::OneMinusSourceGreen, "(1.0 - %source.ggg)"},
{ColorModifier::SourceBlue, "%source.bbb"},
{ColorModifier::OneMinusSourceBlue, "(1.0 - %source.bbb)"},
};
auto src_str = GetTevStageConfigSourceString(source);
auto modifier_it = color_modifier_map.find(modifier);
std::string modifier_str = "%source.????";
if (modifier_it != color_modifier_map.end())
modifier_str = modifier_it->second;
return ReplacePattern(modifier_str, "%source", src_str);
}
static std::string GetTevStageConfigAlphaSourceString(
const TexturingRegs::TevStageConfig::Source& source,
const TexturingRegs::TevStageConfig::AlphaModifier modifier) {
using AlphaModifier = TexturingRegs::TevStageConfig::AlphaModifier;
static const std::map<AlphaModifier, std::string> alpha_modifier_map = {
{AlphaModifier::SourceAlpha, "%source.a"},
{AlphaModifier::OneMinusSourceAlpha, "(1.0 - %source.a)"},
{AlphaModifier::SourceRed, "%source.r"},
{AlphaModifier::OneMinusSourceRed, "(1.0 - %source.r)"},
{AlphaModifier::SourceGreen, "%source.g"},
{AlphaModifier::OneMinusSourceGreen, "(1.0 - %source.g)"},
{AlphaModifier::SourceBlue, "%source.b"},
{AlphaModifier::OneMinusSourceBlue, "(1.0 - %source.b)"},
};
auto src_str = GetTevStageConfigSourceString(source);
auto modifier_it = alpha_modifier_map.find(modifier);
std::string modifier_str = "%source.????";
if (modifier_it != alpha_modifier_map.end())
modifier_str = modifier_it->second;
return ReplacePattern(modifier_str, "%source", src_str);
}
static std::string GetTevStageConfigOperationString(
const TexturingRegs::TevStageConfig::Operation& operation) {
using Operation = TexturingRegs::TevStageConfig::Operation;
static const std::map<Operation, std::string> combiner_map = {
{Operation::Replace, "%source1"},
{Operation::Modulate, "(%source1 * %source2)"},
{Operation::Add, "(%source1 + %source2)"},
{Operation::AddSigned, "(%source1 + %source2) - 0.5"},
{Operation::Lerp, "lerp(%source1, %source2, %source3)"},
{Operation::Subtract, "(%source1 - %source2)"},
{Operation::Dot3_RGB, "dot(%source1, %source2)"},
{Operation::MultiplyThenAdd, "((%source1 * %source2) + %source3)"},
{Operation::AddThenMultiply, "((%source1 + %source2) * %source3)"},
};
const auto op_it = combiner_map.find(operation);
if (op_it == combiner_map.end())
return "Unknown op (%source1, %source2, %source3)";
return op_it->second;
}
std::string GetTevStageConfigColorCombinerString(const TexturingRegs::TevStageConfig& tev_stage) {
auto op_str = GetTevStageConfigOperationString(tev_stage.color_op);
op_str = ReplacePattern(
op_str, "%source1",
GetTevStageConfigColorSourceString(tev_stage.color_source1, tev_stage.color_modifier1));
op_str = ReplacePattern(
op_str, "%source2",
GetTevStageConfigColorSourceString(tev_stage.color_source2, tev_stage.color_modifier2));
return ReplacePattern(
op_str, "%source3",
GetTevStageConfigColorSourceString(tev_stage.color_source3, tev_stage.color_modifier3));
}
std::string GetTevStageConfigAlphaCombinerString(const TexturingRegs::TevStageConfig& tev_stage) {
auto op_str = GetTevStageConfigOperationString(tev_stage.alpha_op);
op_str = ReplacePattern(
op_str, "%source1",
GetTevStageConfigAlphaSourceString(tev_stage.alpha_source1, tev_stage.alpha_modifier1));
op_str = ReplacePattern(
op_str, "%source2",
GetTevStageConfigAlphaSourceString(tev_stage.alpha_source2, tev_stage.alpha_modifier2));
return ReplacePattern(
op_str, "%source3",
GetTevStageConfigAlphaSourceString(tev_stage.alpha_source3, tev_stage.alpha_modifier3));
}
void DumpTevStageConfig(const std::array<TexturingRegs::TevStageConfig, 6>& stages) {
std::string stage_info = "Tev setup:\n";
for (size_t index = 0; index < stages.size(); ++index) {
const auto& tev_stage = stages[index];
stage_info += "Stage " + std::to_string(index) + ": " +
GetTevStageConfigColorCombinerString(tev_stage) + " " +
GetTevStageConfigAlphaCombinerString(tev_stage) + "\n";
}
LOG_TRACE(HW_GPU, "%s", stage_info.c_str());
}
} // namespace
} // namespace

251
src/video_core/debug_utils/debug_utils.h
View file

@ -1,251 +0,0 @@
// Copyright 2014 Citra Emulator Project
// Licensed under GPLv2
// Refer to the license.txt file included.
#pragma once
#include <algorithm>
#include <array>
#include <condition_variable>
#include <iterator>
#include <list>
#include <map>
#include <memory>
#include <mutex>
#include <string>
#include <utility>
#include <vector>
#include "common/common_types.h"
#include "common/vector_math.h"
#include "video_core/regs_rasterizer.h"
#include "video_core/regs_shader.h"
#include "video_core/regs_texturing.h"
namespace CiTrace {
class Recorder;
}
namespace Pica {
namespace Shader {
struct ShaderSetup;
}
class DebugContext {
public:
enum class Event {
FirstEvent = 0,
PicaCommandLoaded = FirstEvent,
PicaCommandProcessed,
IncomingPrimitiveBatch,
FinishedPrimitiveBatch,
VertexShaderInvocation,
IncomingDisplayTransfer,
GSPCommandProcessed,
BufferSwapped,
NumEvents
};
/**
* Inherit from this class to be notified of events registered to some debug context.
* Most importantly this is used for our debugger GUI.
*
* To implement event handling, override the OnPicaBreakPointHit and OnPicaResume methods.
* @warning All BreakPointObservers need to be on the same thread to guarantee thread-safe state
* access
* @todo Evaluate an alternative interface, in which there is only one managing observer and
* multiple child observers running (by design) on the same thread.
*/
class BreakPointObserver {
public:
/// Constructs the object such that it observes events of the given DebugContext.
BreakPointObserver(std::shared_ptr<DebugContext> debug_context)
: context_weak(debug_context) {
std::unique_lock<std::mutex> lock(debug_context->breakpoint_mutex);
debug_context->breakpoint_observers.push_back(this);
}
virtual ~BreakPointObserver() {
auto context = context_weak.lock();
if (context) {
std::unique_lock<std::mutex> lock(context->breakpoint_mutex);
context->breakpoint_observers.remove(this);
// If we are the last observer to be destroyed, tell the debugger context that
// it is free to continue. In particular, this is required for a proper Citra
// shutdown, when the emulation thread is waiting at a breakpoint.
if (context->breakpoint_observers.empty())
context->Resume();
}
}
/**
* Action to perform when a breakpoint was reached.
* @param event Type of event which triggered the breakpoint
* @param data Optional data pointer (if unused, this is a nullptr)
* @note This function will perform nothing unless it is overridden in the child class.
*/
virtual void OnPicaBreakPointHit(Event event, void* data) {}
/**
* Action to perform when emulation is resumed from a breakpoint.
* @note This function will perform nothing unless it is overridden in the child class.
*/
virtual void OnPicaResume() {}
protected:
/**
* Weak context pointer. This need not be valid, so when requesting a shared_ptr via
* context_weak.lock(), always compare the result against nullptr.
*/
std::weak_ptr<DebugContext> context_weak;
};
/**
* Simple structure defining a breakpoint state
*/
struct BreakPoint {
bool enabled = false;
};
/**
* Static constructor used to create a shared_ptr of a DebugContext.
*/
static std::shared_ptr<DebugContext> Construct() {
return std::shared_ptr<DebugContext>(new DebugContext);
}
/**
* Used by the emulation core when a given event has happened. If a breakpoint has been set
* for this event, OnEvent calls the event handlers of the registered breakpoint observers.
* The current thread then is halted until Resume() is called from another thread (or until
* emulation is stopped).
* @param event Event which has happened
* @param data Optional data pointer (pass nullptr if unused). Needs to remain valid until
* Resume() is called.
*/
void OnEvent(Event event, void* data) {
// This check is left in the header to allow the compiler to inline it.
if (!breakpoints[(int)event].enabled)
return;
// For the rest of event handling, call a separate function.
DoOnEvent(event, data);
}
void DoOnEvent(Event event, void* data);
/**
* Resume from the current breakpoint.
* @warning Calling this from the same thread that OnEvent was called in will cause a deadlock.
* Calling from any other thread is safe.
*/
void Resume();
/**
* Delete all set breakpoints and resume emulation.
*/
void ClearBreakpoints() {
for (auto& bp : breakpoints) {
bp.enabled = false;
}
Resume();
}
// TODO: Evaluate if access to these members should be hidden behind a public interface.
std::array<BreakPoint, (int)Event::NumEvents> breakpoints;
Event active_breakpoint;
bool at_breakpoint = false;
std::shared_ptr<CiTrace::Recorder> recorder = nullptr;
private:
/**
* Private default constructor to make sure people always construct this through Construct()
* instead.
*/
DebugContext() = default;
/// Mutex protecting current breakpoint state and the observer list.
std::mutex breakpoint_mutex;
/// Used by OnEvent to wait for resumption.
std::condition_variable resume_from_breakpoint;
/// List of registered observers
std::list<BreakPointObserver*> breakpoint_observers;
};
extern std::shared_ptr<DebugContext> g_debug_context; // TODO: Get rid of this global
namespace DebugUtils {
#define PICA_DUMP_TEXTURES 0
#define PICA_LOG_TEV 0
void DumpShader(const std::string& filename, const ShaderRegs& config,
const Shader::ShaderSetup& setup,
const RasterizerRegs::VSOutputAttributes* output_attributes);
// Utility class to log Pica commands.
struct PicaTrace {
struct Write {
u16 cmd_id;
u16 mask;
u32 value;
};
std::vector<Write> writes;
};
extern bool g_is_pica_tracing;
void StartPicaTracing();
inline bool IsPicaTracing() {
return g_is_pica_tracing;
}
void OnPicaRegWrite(PicaTrace::Write write);
std::unique_ptr<PicaTrace> FinishPicaTracing();
void DumpTexture(const TexturingRegs::TextureConfig& texture_config, u8* data);
std::string GetTevStageConfigColorCombinerString(const TexturingRegs::TevStageConfig& tev_stage);
std::string GetTevStageConfigAlphaCombinerString(const TexturingRegs::TevStageConfig& tev_stage);
/// Dumps the Tev stage config to log at trace level
void DumpTevStageConfig(const std::array<TexturingRegs::TevStageConfig, 6>& stages);
/**
* Used in the vertex loader to merge access records. TODO: Investigate if actually useful.
*/
class MemoryAccessTracker {
/// Combine overlapping and close ranges
void SimplifyRanges() {
for (auto it = ranges.begin(); it != ranges.end(); ++it) {
// NOTE: We add 32 to the range end address to make sure "close" ranges are combined,
// too
auto it2 = std::next(it);
while (it2 != ranges.end() && it->first + it->second + 32 >= it2->first) {
it->second = std::max(it->second, it2->first + it2->second - it->first);
it2 = ranges.erase(it2);
}
}
}
public:
/// Record a particular memory access in the list
void AddAccess(u32 paddr, u32 size) {
// Create new range or extend existing one
ranges[paddr] = std::max(ranges[paddr], size);
// Simplify ranges...
SimplifyRanges();
}
/// Map of accessed ranges (mapping start address to range size)
std::map<u32, u32> ranges;
};
} // namespace
} // namespace

274
src/video_core/geometry_pipeline.cpp
View file

@ -1,274 +0,0 @@
// Copyright 2017 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "video_core/geometry_pipeline.h"
#include "video_core/pica_state.h"
#include "video_core/regs.h"
#include "video_core/renderer_base.h"
#include "video_core/video_core.h"
namespace Pica {
/// An attribute buffering interface for different pipeline modes
class GeometryPipelineBackend {
public:
virtual ~GeometryPipelineBackend() = default;
/// Checks if there is no incomplete data transfer
virtual bool IsEmpty() const = 0;
/// Checks if the pipeline needs a direct input from index buffer
virtual bool NeedIndexInput() const = 0;
/// Submits an index from index buffer
virtual void SubmitIndex(unsigned int val) = 0;
/**
* Submits vertex attributes
* @param input attributes of a vertex output from vertex shader
* @return if the buffer is full and the geometry shader should be invoked
*/
virtual bool SubmitVertex(const Shader::AttributeBuffer& input) = 0;
};
// In the Point mode, vertex attributes are sent to the input registers in the geometry shader unit.
// The size of vertex shader outputs and geometry shader inputs are constants. Geometry shader is
// invoked upon inputs buffer filled up by vertex shader outputs. For example, if we have a geometry
// shader that takes 6 inputs, and the vertex shader outputs 2 attributes, it would take 3 vertices
// for one geometry shader invocation.
// TODO: what happens when the input size is not divisible by the output size?
class GeometryPipeline_Point : public GeometryPipelineBackend {
public:
GeometryPipeline_Point(const Regs& regs, Shader::GSUnitState& unit) : regs(regs), unit(unit) {
ASSERT(regs.pipeline.variable_primitive == 0);
ASSERT(regs.gs.input_to_uniform == 0);
vs_output_num = regs.pipeline.vs_outmap_total_minus_1_a + 1;
size_t gs_input_num = regs.gs.max_input_attribute_index + 1;
ASSERT(gs_input_num % vs_output_num == 0);
buffer_cur = attribute_buffer.attr;
buffer_end = attribute_buffer.attr + gs_input_num;
}
bool IsEmpty() const override {
return buffer_cur == attribute_buffer.attr;
}
bool NeedIndexInput() const override {
return false;
}
void SubmitIndex(unsigned int val) override {
UNREACHABLE();
}
bool SubmitVertex(const Shader::AttributeBuffer& input) override {
buffer_cur = std::copy(input.attr, input.attr + vs_output_num, buffer_cur);
if (buffer_cur == buffer_end) {
buffer_cur = attribute_buffer.attr;
unit.LoadInput(regs.gs, attribute_buffer);
return true;
}
return false;
}
private:
const Regs& regs;
Shader::GSUnitState& unit;
Shader::AttributeBuffer attribute_buffer;
Math::Vec4<float24>* buffer_cur;
Math::Vec4<float24>* buffer_end;
unsigned int vs_output_num;
};
// In VariablePrimitive mode, vertex attributes are buffered into the uniform registers in the
// geometry shader unit. The number of vertex is variable, which is specified by the first index
// value in the batch. This mode is usually used for subdivision.
class GeometryPipeline_VariablePrimitive : public GeometryPipelineBackend {
public:
GeometryPipeline_VariablePrimitive(const Regs& regs, Shader::ShaderSetup& setup)
: regs(regs), setup(setup) {
ASSERT(regs.pipeline.variable_primitive == 1);
ASSERT(regs.gs.input_to_uniform == 1);
vs_output_num = regs.pipeline.vs_outmap_total_minus_1_a + 1;
}
bool IsEmpty() const override {
return need_index;
}
bool NeedIndexInput() const override {
return need_index;
}
void SubmitIndex(unsigned int val) override {
DEBUG_ASSERT(need_index);
// The number of vertex input is put to the uniform register
float24 vertex_num = float24::FromFloat32(static_cast<float>(val));
setup.uniforms.f[0] = Math::MakeVec(vertex_num, vertex_num, vertex_num, vertex_num);
// The second uniform register and so on are used for receiving input vertices
buffer_cur = setup.uniforms.f + 1;
main_vertex_num = regs.pipeline.variable_vertex_main_num_minus_1 + 1;
total_vertex_num = val;
need_index = false;
}
bool SubmitVertex(const Shader::AttributeBuffer& input) override {
DEBUG_ASSERT(!need_index);
if (main_vertex_num != 0) {
// For main vertices, receive all attributes
buffer_cur = std::copy(input.attr, input.attr + vs_output_num, buffer_cur);
--main_vertex_num;
} else {
// For other vertices, only receive the first attribute (usually the position)
*(buffer_cur++) = input.attr[0];
}
--total_vertex_num;
if (total_vertex_num == 0) {
need_index = true;
return true;
}
return false;
}
private:
bool need_index = true;
const Regs& regs;
Shader::ShaderSetup& setup;
unsigned int main_vertex_num;
unsigned int total_vertex_num;
Math::Vec4<float24>* buffer_cur;
unsigned int vs_output_num;
};
// In FixedPrimitive mode, vertex attributes are buffered into the uniform registers in the geometry
// shader unit. The number of vertex per shader invocation is constant. This is usually used for
// particle system.
class GeometryPipeline_FixedPrimitive : public GeometryPipelineBackend {
public:
GeometryPipeline_FixedPrimitive(const Regs& regs, Shader::ShaderSetup& setup)
: regs(regs), setup(setup) {
ASSERT(regs.pipeline.variable_primitive == 0);
ASSERT(regs.gs.input_to_uniform == 1);
vs_output_num = regs.pipeline.vs_outmap_total_minus_1_a + 1;
ASSERT(vs_output_num == regs.pipeline.gs_config.stride_minus_1 + 1);
size_t vertex_num = regs.pipeline.gs_config.fixed_vertex_num_minus_1 + 1;
buffer_cur = buffer_begin = setup.uniforms.f + regs.pipeline.gs_config.start_index;
buffer_end = buffer_begin + vs_output_num * vertex_num;
}
bool IsEmpty() const override {
return buffer_cur == buffer_begin;
}
bool NeedIndexInput() const override {
return false;
}
void SubmitIndex(unsigned int val) override {
UNREACHABLE();
}
bool SubmitVertex(const Shader::AttributeBuffer& input) override {
buffer_cur = std::copy(input.attr, input.attr + vs_output_num, buffer_cur);
if (buffer_cur == buffer_end) {
buffer_cur = buffer_begin;
return true;
}
return false;
}
private:
const Regs& regs;
Shader::ShaderSetup& setup;
Math::Vec4<float24>* buffer_begin;
Math::Vec4<float24>* buffer_cur;
Math::Vec4<float24>* buffer_end;
unsigned int vs_output_num;
};
GeometryPipeline::GeometryPipeline(State& state) : state(state) {}
GeometryPipeline::~GeometryPipeline() = default;
void GeometryPipeline::SetVertexHandler(Shader::VertexHandler vertex_handler) {
this->vertex_handler = vertex_handler;
}
void GeometryPipeline::Setup(Shader::ShaderEngine* shader_engine) {
if (!backend)
return;
this->shader_engine = shader_engine;
shader_engine->SetupBatch(state.gs, state.regs.gs.main_offset);
}
void GeometryPipeline::Reconfigure() {
ASSERT(!backend || backend->IsEmpty());
if (state.regs.pipeline.use_gs == PipelineRegs::UseGS::No) {
backend = nullptr;
return;
}
ASSERT(state.regs.pipeline.use_gs == PipelineRegs::UseGS::Yes);
// The following assumes that when geometry shader is in use, the shader unit 3 is configured as
// a geometry shader unit.
// TODO: what happens if this is not true?
ASSERT(state.regs.pipeline.gs_unit_exclusive_configuration == 1);
ASSERT(state.regs.gs.shader_mode == ShaderRegs::ShaderMode::GS);
state.gs_unit.ConfigOutput(state.regs.gs);
ASSERT(state.regs.pipeline.vs_outmap_total_minus_1_a ==
state.regs.pipeline.vs_outmap_total_minus_1_b);
switch (state.regs.pipeline.gs_config.mode) {
case PipelineRegs::GSMode::Point:
backend = std::make_unique<GeometryPipeline_Point>(state.regs, state.gs_unit);
break;
case PipelineRegs::GSMode::VariablePrimitive:
backend = std::make_unique<GeometryPipeline_VariablePrimitive>(state.regs, state.gs);
break;
case PipelineRegs::GSMode::FixedPrimitive:
backend = std::make_unique<GeometryPipeline_FixedPrimitive>(state.regs, state.gs);
break;
default:
UNREACHABLE();
}
}
bool GeometryPipeline::NeedIndexInput() const {
if (!backend)
return false;
return backend->NeedIndexInput();
}
void GeometryPipeline::SubmitIndex(unsigned int val) {
backend->SubmitIndex(val);
}
void GeometryPipeline::SubmitVertex(const Shader::AttributeBuffer& input) {
if (!backend) {
// No backend means the geometry shader is disabled, so we send the vertex shader output
// directly to the primitive assembler.
vertex_handler(input);
} else {
if (backend->SubmitVertex(input)) {
shader_engine->Run(state.gs, state.gs_unit);
// The uniform b15 is set to true after every geometry shader invocation. This is useful
// for the shader to know if this is the first invocation in a batch, if the program set
// b15 to false first.
state.gs.uniforms.b[15] = true;
}
}
}
} // namespace Pica

49
src/video_core/geometry_pipeline.h
View file

@ -1,49 +0,0 @@
// Copyright 2017 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <memory>
#include "video_core/shader/shader.h"
namespace Pica {
struct State;
class GeometryPipelineBackend;
/// A pipeline receiving from vertex shader and sending to geometry shader and primitive assembler
class GeometryPipeline {
public:
explicit GeometryPipeline(State& state);
~GeometryPipeline();
/// Sets the handler for receiving vertex outputs from vertex shader
void SetVertexHandler(Shader::VertexHandler vertex_handler);
/**
* Setup the geometry shader unit if it is in use
* @param shader_engine the shader engine for the geometry shader to run
*/
void Setup(Shader::ShaderEngine* shader_engine);
/// Reconfigures the pipeline according to current register settings
void Reconfigure();
/// Checks if the pipeline needs a direct input from index buffer
bool NeedIndexInput() const;
/// Submits an index from index buffer. Call this only when NeedIndexInput returns true
void SubmitIndex(unsigned int val);
/// Submits vertex attributes output from vertex shader
void SubmitVertex(const Shader::AttributeBuffer& input);
private:
Shader::VertexHandler vertex_handler;
Shader::ShaderEngine* shader_engine;
std::unique_ptr<GeometryPipelineBackend> backend;
State& state;
};
} // namespace Pica

85
src/video_core/gpu_debugger.h
View file

@ -1,85 +0,0 @@
// Copyright 2014 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <algorithm>
#include <functional>
#include <vector>
#include "core/hle/service/gsp_gpu.h"
class GraphicsDebugger {
public:
// Base class for all objects which need to be notified about GPU events
class DebuggerObserver {
public:
DebuggerObserver() : observed(nullptr) {}
virtual ~DebuggerObserver() {
if (observed)
observed->UnregisterObserver(this);
}
/**
* Called when a GX command has been processed and is ready for being
* read via GraphicsDebugger::ReadGXCommandHistory.
* @param total_command_count Total number of commands in the GX history
* @note All methods in this class are called from the GSP thread
*/
virtual void GXCommandProcessed(int total_command_count) {
const Service::GSP::Command& cmd =
observed->ReadGXCommandHistory(total_command_count - 1);
LOG_TRACE(Debug_GPU, "Received command: id=%x", (int)cmd.id.Value());
}
protected:
const GraphicsDebugger* GetDebugger() const {
return observed;
}
private:
GraphicsDebugger* observed;
friend class GraphicsDebugger;
};
void GXCommandProcessed(u8* command_data) {
if (observers.empty())
return;
gx_command_history.emplace_back();
Service::GSP::Command& cmd = gx_command_history.back();
memcpy(&cmd, command_data, sizeof(Service::GSP::Command));
ForEachObserver([this](DebuggerObserver* observer) {
observer->GXCommandProcessed(static_cast<int>(this->gx_command_history.size()));
});
}
const Service::GSP::Command& ReadGXCommandHistory(int index) const {
// TODO: Is this thread-safe?
return gx_command_history[index];
}
void RegisterObserver(DebuggerObserver* observer) {
// TODO: Check for duplicates
observers.push_back(observer);
observer->observed = this;
}
void UnregisterObserver(DebuggerObserver* observer) {
observers.erase(std::remove(observers.begin(), observers.end(), observer), observers.end());
observer->observed = nullptr;
}
private:
void ForEachObserver(std::function<void(DebuggerObserver*)> func) {
std::for_each(observers.begin(), observers.end(), func);
}
std::vector<DebuggerObserver*> observers;
std::vector<Service::GSP::Command> gx_command_history;
};

54
src/video_core/pica.cpp
View file

@ -1,54 +0,0 @@
// Copyright 2015 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <cstring>
#include "video_core/geometry_pipeline.h"
#include "video_core/pica.h"
#include "video_core/pica_state.h"
#include "video_core/renderer_base.h"
#include "video_core/video_core.h"
namespace Pica {
State g_state;
void Init() {
g_state.Reset();
}
void Shutdown() {
Shader::Shutdown();
}
template <typename T>
void Zero(T& o) {
memset(&o, 0, sizeof(o));
}
State::State() : geometry_pipeline(*this) {
auto SubmitVertex = [this](const Shader::AttributeBuffer& vertex) {
using Pica::Shader::OutputVertex;
auto AddTriangle = [this](const OutputVertex& v0, const OutputVertex& v1,
const OutputVertex& v2) {
VideoCore::g_renderer->Rasterizer()->AddTriangle(v0, v1, v2);
};
primitive_assembler.SubmitVertex(
Shader::OutputVertex::FromAttributeBuffer(regs.rasterizer, vertex), AddTriangle);
};
auto SetWinding = [this]() { primitive_assembler.SetWinding(); };
g_state.gs_unit.SetVertexHandler(SubmitVertex, SetWinding);
g_state.geometry_pipeline.SetVertexHandler(SubmitVertex);
}
void State::Reset() {
Zero(regs);
Zero(vs);
Zero(gs);
Zero(cmd_list);
Zero(immediate);
primitive_assembler.Reconfigure(PipelineRegs::TriangleTopology::List);
}
}

16
src/video_core/pica.h
View file

@ -1,16 +0,0 @@
// Copyright 2014 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include "video_core/regs_texturing.h"
namespace Pica {
/// Initialize Pica state
void Init();
/// Shutdown Pica state
void Shutdown();
} // namespace

159
src/video_core/pica_state.h
View file

@ -1,159 +0,0 @@
// Copyright 2016 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <array>
#include "common/bit_field.h"
#include "common/common_types.h"
#include "common/vector_math.h"
#include "video_core/geometry_pipeline.h"
#include "video_core/primitive_assembly.h"
#include "video_core/regs.h"
#include "video_core/shader/shader.h"
namespace Pica {
/// Struct used to describe current Pica state
struct State {
State();
void Reset();
/// Pica registers
Regs regs;
Shader::ShaderSetup vs;
Shader::ShaderSetup gs;
Shader::AttributeBuffer input_default_attributes;
struct ProcTex {
union ValueEntry {
u32 raw;
// LUT value, encoded as 12-bit fixed point, with 12 fraction bits
BitField<0, 12, u32> value; // 0.0.12 fixed point
// Difference between two entry values. Used for efficient interpolation.
// 0.0.12 fixed point with two's complement. The range is [-0.5, 0.5).
// Note: the type of this is different from the one of lighting LUT
BitField<12, 12, s32> difference;
float ToFloat() const {
return static_cast<float>(value) / 4095.f;
}
float DiffToFloat() const {
return static_cast<float>(difference) / 4095.f;
}
};
union ColorEntry {
u32 raw;
BitField<0, 8, u32> r;
BitField<8, 8, u32> g;
BitField<16, 8, u32> b;
BitField<24, 8, u32> a;
Math::Vec4<u8> ToVector() const {
return {static_cast<u8>(r), static_cast<u8>(g), static_cast<u8>(b),
static_cast<u8>(a)};
}
};
union ColorDifferenceEntry {
u32 raw;
BitField<0, 8, s32> r; // half of the difference between two ColorEntry
BitField<8, 8, s32> g;
BitField<16, 8, s32> b;
BitField<24, 8, s32> a;
Math::Vec4<s32> ToVector() const {
return Math::Vec4<s32>{r, g, b, a} * 2;
}
};
std::array<ValueEntry, 128> noise_table;
std::array<ValueEntry, 128> color_map_table;
std::array<ValueEntry, 128> alpha_map_table;
std::array<ColorEntry, 256> color_table;
std::array<ColorDifferenceEntry, 256> color_diff_table;
} proctex;
struct Lighting {
union LutEntry {
// Used for raw access
u32 raw;
// LUT value, encoded as 12-bit fixed point, with 12 fraction bits
BitField<0, 12, u32> value; // 0.0.12 fixed point
// Used for efficient interpolation.
BitField<12, 11, u32> difference; // 0.0.11 fixed point
BitField<23, 1, u32> neg_difference;
float ToFloat() const {
return static_cast<float>(value) / 4095.f;
}
float DiffToFloat() const {
float diff = static_cast<float>(difference) / 2047.f;
return neg_difference ? -diff : diff;
}
};
std::array<std::array<LutEntry, 256>, 24> luts;
} lighting;
struct {
union LutEntry {
// Used for raw access
u32 raw;
BitField<0, 13, s32> difference; // 1.1.11 fixed point
BitField<13, 11, u32> value; // 0.0.11 fixed point
float ToFloat() const {
return static_cast<float>(value) / 2047.0f;
}
float DiffToFloat() const {
return static_cast<float>(difference) / 2047.0f;
}
};
std::array<LutEntry, 128> lut;
} fog;
/// Current Pica command list
struct {
const u32* head_ptr;
const u32* current_ptr;
u32 length;
} cmd_list;
/// Struct used to describe immediate mode rendering state
struct ImmediateModeState {
// Used to buffer partial vertices for immediate-mode rendering.
Shader::AttributeBuffer input_vertex;
// Index of the next attribute to be loaded into `input_vertex`.
u32 current_attribute = 0;
// Indicates the immediate mode just started and the geometry pipeline needs to reconfigure
bool reset_geometry_pipeline = true;
} immediate;
// the geometry shader needs to be kept in the global state because some shaders relie on
// preserved register value across shader invocation.
// TODO: also bring the three vertex shader units here and implement the shader scheduler.
Shader::GSUnitState gs_unit;
GeometryPipeline geometry_pipeline;
// This is constructed with a dummy triangle topology
PrimitiveAssembler<Shader::OutputVertex> primitive_assembler;
};
extern State g_state; ///< Current Pica state
} // namespace

143
src/video_core/pica_types.h
View file

@ -1,143 +0,0 @@
// Copyright 2015 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <cmath>
#include <cstring>
#include "common/common_types.h"
namespace Pica {
/**
* Template class for converting arbitrary Pica float types to IEEE 754 32-bit single-precision
* floating point.
*
* When decoding, format is as follows:
* - The first `M` bits are the mantissa
* - The next `E` bits are the exponent
* - The last bit is the sign bit
*
* @todo Verify on HW if this conversion is sufficiently accurate.
*/
template <unsigned M, unsigned E>
struct Float {
public:
static Float<M, E> FromFloat32(float val) {
Float<M, E> ret;
ret.value = val;
return ret;
}
static Float<M, E> FromRaw(u32 hex) {
Float<M, E> res;
const int width = M + E + 1;
const int bias = 128 - (1 << (E - 1));
const int exponent = (hex >> M) & ((1 << E) - 1);
const unsigned mantissa = hex & ((1 << M) - 1);
if (hex & ((1 << (width - 1)) - 1))
hex = ((hex >> (E + M)) << 31) | (mantissa << (23 - M)) | ((exponent + bias) << 23);
else
hex = ((hex >> (E + M)) << 31);
std::memcpy(&res.value, &hex, sizeof(float));
return res;
}
static Float<M, E> Zero() {
return FromFloat32(0.f);
}
// Not recommended for anything but logging
float ToFloat32() const {
return value;
}
Float<M, E> operator*(const Float<M, E>& flt) const {
float result = value * flt.ToFloat32();
// PICA gives 0 instead of NaN when multiplying by inf
if (!std::isnan(value) && !std::isnan(flt.ToFloat32()))
if (std::isnan(result))
result = 0.f;
return Float<M, E>::FromFloat32(result);
}
Float<M, E> operator/(const Float<M, E>& flt) const {
return Float<M, E>::FromFloat32(ToFloat32() / flt.ToFloat32());
}
Float<M, E> operator+(const Float<M, E>& flt) const {
return Float<M, E>::FromFloat32(ToFloat32() + flt.ToFloat32());
}
Float<M, E> operator-(const Float<M, E>& flt) const {
return Float<M, E>::FromFloat32(ToFloat32() - flt.ToFloat32());
}
Float<M, E>& operator*=(const Float<M, E>& flt) {
value = operator*(flt).value;
return *this;
}
Float<M, E>& operator/=(const Float<M, E>& flt) {
value /= flt.ToFloat32();
return *this;
}
Float<M, E>& operator+=(const Float<M, E>& flt) {
value += flt.ToFloat32();
return *this;
}
Float<M, E>& operator-=(const Float<M, E>& flt) {
value -= flt.ToFloat32();
return *this;
}
Float<M, E> operator-() const {
return Float<M, E>::FromFloat32(-ToFloat32());
}
bool operator<(const Float<M, E>& flt) const {
return ToFloat32() < flt.ToFloat32();
}
bool operator>(const Float<M, E>& flt) const {
return ToFloat32() > flt.ToFloat32();
}
bool operator>=(const Float<M, E>& flt) const {
return ToFloat32() >= flt.ToFloat32();
}
bool operator<=(const Float<M, E>& flt) const {
return ToFloat32() <= flt.ToFloat32();
}
bool operator==(const Float<M, E>& flt) const {
return ToFloat32() == flt.ToFloat32();
}
bool operator!=(const Float<M, E>& flt) const {
return ToFloat32() != flt.ToFloat32();
}
private:
static const unsigned MASK = (1 << (M + E + 1)) - 1;
static const unsigned MANTISSA_MASK = (1 << M) - 1;
static const unsigned EXPONENT_MASK = (1 << E) - 1;
// Stored as a regular float, merely for convenience
// TODO: Perform proper arithmetic on this!
float value;
};
using float24 = Float<16, 7>;
using float20 = Float<12, 7>;
using float16 = Float<10, 5>;
} // namespace Pica

77
src/video_core/primitive_assembly.cpp
View file

@ -1,77 +0,0 @@
// Copyright 2014 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/logging/log.h"
#include "video_core/primitive_assembly.h"
#include "video_core/regs_pipeline.h"
#include "video_core/shader/shader.h"
namespace Pica {
template <typename VertexType>
PrimitiveAssembler<VertexType>::PrimitiveAssembler(PipelineRegs::TriangleTopology topology)
: topology(topology), buffer_index(0) {}
template <typename VertexType>
void PrimitiveAssembler<VertexType>::SubmitVertex(const VertexType& vtx,
TriangleHandler triangle_handler) {
switch (topology) {
case PipelineRegs::TriangleTopology::List:
case PipelineRegs::TriangleTopology::Shader:
if (buffer_index < 2) {
buffer[buffer_index++] = vtx;
} else {
buffer_index = 0;
if (topology == PipelineRegs::TriangleTopology::Shader && winding) {
triangle_handler(buffer[1], buffer[0], vtx);
winding = false;
} else {
triangle_handler(buffer[0], buffer[1], vtx);
}
}
break;
case PipelineRegs::TriangleTopology::Strip:
case PipelineRegs::TriangleTopology::Fan:
if (strip_ready)
triangle_handler(buffer[0], buffer[1], vtx);
buffer[buffer_index] = vtx;
strip_ready |= (buffer_index == 1);
if (topology == PipelineRegs::TriangleTopology::Strip)
buffer_index = !buffer_index;
else if (topology == PipelineRegs::TriangleTopology::Fan)
buffer_index = 1;
break;
default:
LOG_ERROR(HW_GPU, "Unknown triangle topology %x:", (int)topology);
break;
}
}
template <typename VertexType>
void PrimitiveAssembler<VertexType>::SetWinding() {
winding = true;
}
template <typename VertexType>
void PrimitiveAssembler<VertexType>::Reset() {
buffer_index = 0;
strip_ready = false;
winding = false;
}
template <typename VertexType>
void PrimitiveAssembler<VertexType>::Reconfigure(PipelineRegs::TriangleTopology topology) {
Reset();
this->topology = topology;
}
// explicitly instantiate use cases
template struct PrimitiveAssembler<Shader::OutputVertex>;
} // namespace

57
src/video_core/primitive_assembly.h
View file

@ -1,57 +0,0 @@
// Copyright 2014 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <functional>
#include "video_core/regs_pipeline.h"
namespace Pica {
/*
* Utility class to build triangles from a series of vertices,
* according to a given triangle topology.
*/
template <typename VertexType>
struct PrimitiveAssembler {
using TriangleHandler =
std::function<void(const VertexType& v0, const VertexType& v1, const VertexType& v2)>;
PrimitiveAssembler(
PipelineRegs::TriangleTopology topology = PipelineRegs::TriangleTopology::List);
/*
* Queues a vertex, builds primitives from the vertex queue according to the given
* triangle topology, and calls triangle_handler for each generated primitive.
* NOTE: We could specify the triangle handler in the constructor, but this way we can
* keep event and handler code next to each other.
*/
void SubmitVertex(const VertexType& vtx, TriangleHandler triangle_handler);
/**
* Invert the vertex order of the next triangle. Called by geometry shader emitter.
* This only takes effect for TriangleTopology::Shader.
*/
void SetWinding();
/**
* Resets the internal state of the PrimitiveAssembler.
*/
void Reset();
/**
* Reconfigures the PrimitiveAssembler to use a different triangle topology.
*/
void Reconfigure(PipelineRegs::TriangleTopology topology);
private:
PipelineRegs::TriangleTopology topology;
int buffer_index;
VertexType buffer[2];
bool strip_ready = false;
bool winding = false;
};
} // namespace

67
src/video_core/rasterizer_interface.h
View file

@ -1,67 +0,0 @@
// Copyright 2015 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include "common/common_types.h"
#include "core/hw/gpu.h"
struct ScreenInfo;
namespace Pica {
namespace Shader {
struct OutputVertex;
}
}
namespace VideoCore {
class RasterizerInterface {
public:
virtual ~RasterizerInterface() {}
/// Queues the primitive formed by the given vertices for rendering
virtual void AddTriangle(const Pica::Shader::OutputVertex& v0,
const Pica::Shader::OutputVertex& v1,
const Pica::Shader::OutputVertex& v2) = 0;
/// Draw the current batch of triangles
virtual void DrawTriangles() = 0;
/// Notify rasterizer that the specified PICA register has been changed
virtual void NotifyPicaRegisterChanged(u32 id) = 0;
/// Notify rasterizer that all caches should be flushed to 3DS memory
virtual void FlushAll() = 0;
/// Notify rasterizer that any caches of the specified region should be flushed to 3DS memory
virtual void FlushRegion(PAddr addr, u64 size) = 0;
/// Notify rasterizer that any caches of the specified region should be flushed to 3DS memory
/// and invalidated
virtual void FlushAndInvalidateRegion(PAddr addr, u64 size) = 0;
/// Attempt to use a faster method to perform a display transfer with is_texture_copy = 0
virtual bool AccelerateDisplayTransfer(const GPU::Regs::DisplayTransferConfig& config) {
return false;
}
/// Attempt to use a faster method to perform a display transfer with is_texture_copy = 1
virtual bool AccelerateTextureCopy(const GPU::Regs::DisplayTransferConfig& config) {
return false;
}
/// Attempt to use a faster method to fill a region
virtual bool AccelerateFill(const GPU::Regs::MemoryFillConfig& config) {
return false;
}
/// Attempt to use a faster method to display the framebuffer to screen
virtual bool AccelerateDisplay(const GPU::Regs::FramebufferConfig& config,
PAddr framebuffer_addr, u32 pixel_stride,
ScreenInfo& screen_info) {
return false;
}
};
}

488
src/video_core/regs.cpp
View file

@ -1,488 +0,0 @@
// Copyright 2015 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <algorithm>
#include <iterator>
#include <utility>
#include "common/common_types.h"
#include "video_core/regs.h"
namespace Pica {
static const std::pair<u16, const char*> register_names[] = {
{0x010, "GPUREG_FINALIZE"},
{0x040, "GPUREG_FACECULLING_CONFIG"},
{0x041, "GPUREG_VIEWPORT_WIDTH"},
{0x042, "GPUREG_VIEWPORT_INVW"},
{0x043, "GPUREG_VIEWPORT_HEIGHT"},
{0x044, "GPUREG_VIEWPORT_INVH"},
{0x047, "GPUREG_FRAGOP_CLIP"},
{0x048, "GPUREG_FRAGOP_CLIP_DATA0"},
{0x049, "GPUREG_FRAGOP_CLIP_DATA1"},
{0x04A, "GPUREG_FRAGOP_CLIP_DATA2"},
{0x04B, "GPUREG_FRAGOP_CLIP_DATA3"},
{0x04D, "GPUREG_DEPTHMAP_SCALE"},
{0x04E, "GPUREG_DEPTHMAP_OFFSET"},
{0x04F, "GPUREG_SH_OUTMAP_TOTAL"},
{0x050, "GPUREG_SH_OUTMAP_O0"},
{0x051, "GPUREG_SH_OUTMAP_O1"},
{0x052, "GPUREG_SH_OUTMAP_O2"},
{0x053, "GPUREG_SH_OUTMAP_O3"},
{0x054, "GPUREG_SH_OUTMAP_O4"},
{0x055, "GPUREG_SH_OUTMAP_O5"},
{0x056, "GPUREG_SH_OUTMAP_O6"},
{0x061, "GPUREG_EARLYDEPTH_FUNC"},
{0x062, "GPUREG_EARLYDEPTH_TEST1"},
{0x063, "GPUREG_EARLYDEPTH_CLEAR"},
{0x064, "GPUREG_SH_OUTATTR_MODE"},
{0x065, "GPUREG_SCISSORTEST_MODE"},
{0x066, "GPUREG_SCISSORTEST_POS"},
{0x067, "GPUREG_SCISSORTEST_DIM"},
{0x068, "GPUREG_VIEWPORT_XY"},
{0x06A, "GPUREG_EARLYDEPTH_DATA"},
{0x06D, "GPUREG_DEPTHMAP_ENABLE"},
{0x06E, "GPUREG_RENDERBUF_DIM"},
{0x06F, "GPUREG_SH_OUTATTR_CLOCK"},
{0x080, "GPUREG_TEXUNIT_CONFIG"},
{0x081, "GPUREG_TEXUNIT0_BORDER_COLOR"},
{0x082, "GPUREG_TEXUNIT0_DIM"},
{0x083, "GPUREG_TEXUNIT0_PARAM"},
{0x084, "GPUREG_TEXUNIT0_LOD"},
{0x085, "GPUREG_TEXUNIT0_ADDR1"},
{0x086, "GPUREG_TEXUNIT0_ADDR2"},
{0x087, "GPUREG_TEXUNIT0_ADDR3"},
{0x088, "GPUREG_TEXUNIT0_ADDR4"},
{0x089, "GPUREG_TEXUNIT0_ADDR5"},
{0x08A, "GPUREG_TEXUNIT0_ADDR6"},
{0x08B, "GPUREG_TEXUNIT0_SHADOW"},
{0x08E, "GPUREG_TEXUNIT0_TYPE"},
{0x08F, "GPUREG_LIGHTING_ENABLE0"},
{0x091, "GPUREG_TEXUNIT1_BORDER_COLOR"},
{0x092, "GPUREG_TEXUNIT1_DIM"},
{0x093, "GPUREG_TEXUNIT1_PARAM"},
{0x094, "GPUREG_TEXUNIT1_LOD"},
{0x095, "GPUREG_TEXUNIT1_ADDR"},
{0x096, "GPUREG_TEXUNIT1_TYPE"},
{0x099, "GPUREG_TEXUNIT2_BORDER_COLOR"},
{0x09A, "GPUREG_TEXUNIT2_DIM"},
{0x09B, "GPUREG_TEXUNIT2_PARAM"},
{0x09C, "GPUREG_TEXUNIT2_LOD"},
{0x09D, "GPUREG_TEXUNIT2_ADDR"},
{0x09E, "GPUREG_TEXUNIT2_TYPE"},
{0x0A8, "GPUREG_TEXUNIT3_PROCTEX0"},
{0x0A9, "GPUREG_TEXUNIT3_PROCTEX1"},
{0x0AA, "GPUREG_TEXUNIT3_PROCTEX2"},
{0x0AB, "GPUREG_TEXUNIT3_PROCTEX3"},
{0x0AC, "GPUREG_TEXUNIT3_PROCTEX4"},
{0x0AD, "GPUREG_TEXUNIT3_PROCTEX5"},
{0x0AF, "GPUREG_PROCTEX_LUT"},
{0x0B0, "GPUREG_PROCTEX_LUT_DATA0"},
{0x0B1, "GPUREG_PROCTEX_LUT_DATA1"},
{0x0B2, "GPUREG_PROCTEX_LUT_DATA2"},
{0x0B3, "GPUREG_PROCTEX_LUT_DATA3"},
{0x0B4, "GPUREG_PROCTEX_LUT_DATA4"},
{0x0B5, "GPUREG_PROCTEX_LUT_DATA5"},
{0x0B6, "GPUREG_PROCTEX_LUT_DATA6"},
{0x0B7, "GPUREG_PROCTEX_LUT_DATA7"},
{0x0C0, "GPUREG_TEXENV0_SOURCE"},
{0x0C1, "GPUREG_TEXENV0_OPERAND"},
{0x0C2, "GPUREG_TEXENV0_COMBINER"},
{0x0C3, "GPUREG_TEXENV0_COLOR"},
{0x0C4, "GPUREG_TEXENV0_SCALE"},
{0x0C8, "GPUREG_TEXENV1_SOURCE"},
{0x0C9, "GPUREG_TEXENV1_OPERAND"},
{0x0CA, "GPUREG_TEXENV1_COMBINER"},
{0x0CB, "GPUREG_TEXENV1_COLOR"},
{0x0CC, "GPUREG_TEXENV1_SCALE"},
{0x0D0, "GPUREG_TEXENV2_SOURCE"},
{0x0D1, "GPUREG_TEXENV2_OPERAND"},
{0x0D2, "GPUREG_TEXENV2_COMBINER"},
{0x0D3, "GPUREG_TEXENV2_COLOR"},
{0x0D4, "GPUREG_TEXENV2_SCALE"},
{0x0D8, "GPUREG_TEXENV3_SOURCE"},
{0x0D9, "GPUREG_TEXENV3_OPERAND"},
{0x0DA, "GPUREG_TEXENV3_COMBINER"},
{0x0DB, "GPUREG_TEXENV3_COLOR"},
{0x0DC, "GPUREG_TEXENV3_SCALE"},
{0x0E0, "GPUREG_TEXENV_UPDATE_BUFFER"},
{0x0E1, "GPUREG_FOG_COLOR"},
{0x0E4, "GPUREG_GAS_ATTENUATION"},
{0x0E5, "GPUREG_GAS_ACCMAX"},
{0x0E6, "GPUREG_FOG_LUT_INDEX"},
{0x0E8, "GPUREG_FOG_LUT_DATA0"},
{0x0E9, "GPUREG_FOG_LUT_DATA1"},
{0x0EA, "GPUREG_FOG_LUT_DATA2"},
{0x0EB, "GPUREG_FOG_LUT_DATA3"},
{0x0EC, "GPUREG_FOG_LUT_DATA4"},
{0x0ED, "GPUREG_FOG_LUT_DATA5"},
{0x0EE, "GPUREG_FOG_LUT_DATA6"},
{0x0EF, "GPUREG_FOG_LUT_DATA7"},
{0x0F0, "GPUREG_TEXENV4_SOURCE"},
{0x0F1, "GPUREG_TEXENV4_OPERAND"},
{0x0F2, "GPUREG_TEXENV4_COMBINER"},
{0x0F3, "GPUREG_TEXENV4_COLOR"},
{0x0F4, "GPUREG_TEXENV4_SCALE"},
{0x0F8, "GPUREG_TEXENV5_SOURCE"},
{0x0F9, "GPUREG_TEXENV5_OPERAND"},
{0x0FA, "GPUREG_TEXENV5_COMBINER"},
{0x0FB, "GPUREG_TEXENV5_COLOR"},
{0x0FC, "GPUREG_TEXENV5_SCALE"},
{0x0FD, "GPUREG_TEXENV_BUFFER_COLOR"},
{0x100, "GPUREG_COLOR_OPERATION"},
{0x101, "GPUREG_BLEND_FUNC"},
{0x102, "GPUREG_LOGIC_OP"},
{0x103, "GPUREG_BLEND_COLOR"},
{0x104, "GPUREG_FRAGOP_ALPHA_TEST"},
{0x105, "GPUREG_STENCIL_TEST"},
{0x106, "GPUREG_STENCIL_OP"},
{0x107, "GPUREG_DEPTH_COLOR_MASK"},
{0x110, "GPUREG_FRAMEBUFFER_INVALIDATE"},
{0x111, "GPUREG_FRAMEBUFFER_FLUSH"},
{0x112, "GPUREG_COLORBUFFER_READ"},
{0x113, "GPUREG_COLORBUFFER_WRITE"},
{0x114, "GPUREG_DEPTHBUFFER_READ"},
{0x115, "GPUREG_DEPTHBUFFER_WRITE"},
{0x116, "GPUREG_DEPTHBUFFER_FORMAT"},
{0x117, "GPUREG_COLORBUFFER_FORMAT"},
{0x118, "GPUREG_EARLYDEPTH_TEST2"},
{0x11B, "GPUREG_FRAMEBUFFER_BLOCK32"},
{0x11C, "GPUREG_DEPTHBUFFER_LOC"},
{0x11D, "GPUREG_COLORBUFFER_LOC"},
{0x11E, "GPUREG_FRAMEBUFFER_DIM"},
{0x120, "GPUREG_GAS_LIGHT_XY"},
{0x121, "GPUREG_GAS_LIGHT_Z"},
{0x122, "GPUREG_GAS_LIGHT_Z_COLOR"},
{0x123, "GPUREG_GAS_LUT_INDEX"},
{0x124, "GPUREG_GAS_LUT_DATA"},
{0x126, "GPUREG_GAS_DELTAZ_DEPTH"},
{0x130, "GPUREG_FRAGOP_SHADOW"},
{0x140, "GPUREG_LIGHT0_SPECULAR0"},
{0x141, "GPUREG_LIGHT0_SPECULAR1"},
{0x142, "GPUREG_LIGHT0_DIFFUSE"},
{0x143, "GPUREG_LIGHT0_AMBIENT"},
{0x144, "GPUREG_LIGHT0_XY"},
{0x145, "GPUREG_LIGHT0_Z"},
{0x146, "GPUREG_LIGHT0_SPOTDIR_XY"},
{0x147, "GPUREG_LIGHT0_SPOTDIR_Z"},
{0x149, "GPUREG_LIGHT0_CONFIG"},
{0x14A, "GPUREG_LIGHT0_ATTENUATION_BIAS"},
{0x14B, "GPUREG_LIGHT0_ATTENUATION_SCALE"},
{0x150, "GPUREG_LIGHT1_SPECULAR0"},
{0x151, "GPUREG_LIGHT1_SPECULAR1"},
{0x152, "GPUREG_LIGHT1_DIFFUSE"},
{0x153, "GPUREG_LIGHT1_AMBIENT"},
{0x154, "GPUREG_LIGHT1_XY"},
{0x155, "GPUREG_LIGHT1_Z"},
{0x156, "GPUREG_LIGHT1_SPOTDIR_XY"},
{0x157, "GPUREG_LIGHT1_SPOTDIR_Z"},
{0x159, "GPUREG_LIGHT1_CONFIG"},
{0x15A, "GPUREG_LIGHT1_ATTENUATION_BIAS"},
{0x15B, "GPUREG_LIGHT1_ATTENUATION_SCALE"},
{0x160, "GPUREG_LIGHT2_SPECULAR0"},
{0x161, "GPUREG_LIGHT2_SPECULAR1"},
{0x162, "GPUREG_LIGHT2_DIFFUSE"},
{0x163, "GPUREG_LIGHT2_AMBIENT"},
{0x164, "GPUREG_LIGHT2_XY"},
{0x165, "GPUREG_LIGHT2_Z"},
{0x166, "GPUREG_LIGHT2_SPOTDIR_XY"},
{0x167, "GPUREG_LIGHT2_SPOTDIR_Z"},
{0x169, "GPUREG_LIGHT2_CONFIG"},
{0x16A, "GPUREG_LIGHT2_ATTENUATION_BIAS"},
{0x16B, "GPUREG_LIGHT2_ATTENUATION_SCALE"},
{0x170, "GPUREG_LIGHT3_SPECULAR0"},
{0x171, "GPUREG_LIGHT3_SPECULAR1"},
{0x172, "GPUREG_LIGHT3_DIFFUSE"},
{0x173, "GPUREG_LIGHT3_AMBIENT"},
{0x174, "GPUREG_LIGHT3_XY"},
{0x175, "GPUREG_LIGHT3_Z"},
{0x176, "GPUREG_LIGHT3_SPOTDIR_XY"},
{0x177, "GPUREG_LIGHT3_SPOTDIR_Z"},
{0x179, "GPUREG_LIGHT3_CONFIG"},
{0x17A, "GPUREG_LIGHT3_ATTENUATION_BIAS"},
{0x17B, "GPUREG_LIGHT3_ATTENUATION_SCALE"},
{0x180, "GPUREG_LIGHT4_SPECULAR0"},
{0x181, "GPUREG_LIGHT4_SPECULAR1"},
{0x182, "GPUREG_LIGHT4_DIFFUSE"},
{0x183, "GPUREG_LIGHT4_AMBIENT"},
{0x184, "GPUREG_LIGHT4_XY"},
{0x185, "GPUREG_LIGHT4_Z"},
{0x186, "GPUREG_LIGHT4_SPOTDIR_XY"},
{0x187, "GPUREG_LIGHT4_SPOTDIR_Z"},
{0x189, "GPUREG_LIGHT4_CONFIG"},
{0x18A, "GPUREG_LIGHT4_ATTENUATION_BIAS"},
{0x18B, "GPUREG_LIGHT4_ATTENUATION_SCALE"},
{0x190, "GPUREG_LIGHT5_SPECULAR0"},
{0x191, "GPUREG_LIGHT5_SPECULAR1"},
{0x192, "GPUREG_LIGHT5_DIFFUSE"},
{0x193, "GPUREG_LIGHT5_AMBIENT"},
{0x194, "GPUREG_LIGHT5_XY"},
{0x195, "GPUREG_LIGHT5_Z"},
{0x196, "GPUREG_LIGHT5_SPOTDIR_XY"},
{0x197, "GPUREG_LIGHT5_SPOTDIR_Z"},
{0x199, "GPUREG_LIGHT5_CONFIG"},
{0x19A, "GPUREG_LIGHT5_ATTENUATION_BIAS"},
{0x19B, "GPUREG_LIGHT5_ATTENUATION_SCALE"},
{0x1A0, "GPUREG_LIGHT6_SPECULAR0"},
{0x1A1, "GPUREG_LIGHT6_SPECULAR1"},
{0x1A2, "GPUREG_LIGHT6_DIFFUSE"},
{0x1A3, "GPUREG_LIGHT6_AMBIENT"},
{0x1A4, "GPUREG_LIGHT6_XY"},
{0x1A5, "GPUREG_LIGHT6_Z"},
{0x1A6, "GPUREG_LIGHT6_SPOTDIR_XY"},
{0x1A7, "GPUREG_LIGHT6_SPOTDIR_Z"},
{0x1A9, "GPUREG_LIGHT6_CONFIG"},
{0x1AA, "GPUREG_LIGHT6_ATTENUATION_BIAS"},
{0x1AB, "GPUREG_LIGHT6_ATTENUATION_SCALE"},
{0x1B0, "GPUREG_LIGHT7_SPECULAR0"},
{0x1B1, "GPUREG_LIGHT7_SPECULAR1"},
{0x1B2, "GPUREG_LIGHT7_DIFFUSE"},
{0x1B3, "GPUREG_LIGHT7_AMBIENT"},
{0x1B4, "GPUREG_LIGHT7_XY"},
{0x1B5, "GPUREG_LIGHT7_Z"},
{0x1B6, "GPUREG_LIGHT7_SPOTDIR_XY"},
{0x1B7, "GPUREG_LIGHT7_SPOTDIR_Z"},
{0x1B9, "GPUREG_LIGHT7_CONFIG"},
{0x1BA, "GPUREG_LIGHT7_ATTENUATION_BIAS"},
{0x1BB, "GPUREG_LIGHT7_ATTENUATION_SCALE"},
{0x1C0, "GPUREG_LIGHTING_AMBIENT"},
{0x1C2, "GPUREG_LIGHTING_NUM_LIGHTS"},
{0x1C3, "GPUREG_LIGHTING_CONFIG0"},
{0x1C4, "GPUREG_LIGHTING_CONFIG1"},
{0x1C5, "GPUREG_LIGHTING_LUT_INDEX"},
{0x1C6, "GPUREG_LIGHTING_ENABLE1"},
{0x1C8, "GPUREG_LIGHTING_LUT_DATA0"},
{0x1C9, "GPUREG_LIGHTING_LUT_DATA1"},
{0x1CA, "GPUREG_LIGHTING_LUT_DATA2"},
{0x1CB, "GPUREG_LIGHTING_LUT_DATA3"},
{0x1CC, "GPUREG_LIGHTING_LUT_DATA4"},
{0x1CD, "GPUREG_LIGHTING_LUT_DATA5"},
{0x1CE, "GPUREG_LIGHTING_LUT_DATA6"},
{0x1CF, "GPUREG_LIGHTING_LUT_DATA7"},
{0x1D0, "GPUREG_LIGHTING_LUTINPUT_ABS"},
{0x1D1, "GPUREG_LIGHTING_LUTINPUT_SELECT"},
{0x1D2, "GPUREG_LIGHTING_LUTINPUT_SCALE"},
{0x1D9, "GPUREG_LIGHTING_LIGHT_PERMUTATION"},
{0x200, "GPUREG_ATTRIBBUFFERS_LOC"},
{0x201, "GPUREG_ATTRIBBUFFERS_FORMAT_LOW"},
{0x202, "GPUREG_ATTRIBBUFFERS_FORMAT_HIGH"},
{0x203, "GPUREG_ATTRIBBUFFER0_OFFSET"},
{0x204, "GPUREG_ATTRIBBUFFER0_CONFIG1"},
{0x205, "GPUREG_ATTRIBBUFFER0_CONFIG2"},
{0x206, "GPUREG_ATTRIBBUFFER1_OFFSET"},
{0x207, "GPUREG_ATTRIBBUFFER1_CONFIG1"},
{0x208, "GPUREG_ATTRIBBUFFER1_CONFIG2"},
{0x209, "GPUREG_ATTRIBBUFFER2_OFFSET"},
{0x20A, "GPUREG_ATTRIBBUFFER2_CONFIG1"},
{0x20B, "GPUREG_ATTRIBBUFFER2_CONFIG2"},
{0x20C, "GPUREG_ATTRIBBUFFER3_OFFSET"},
{0x20D, "GPUREG_ATTRIBBUFFER3_CONFIG1"},
{0x20E, "GPUREG_ATTRIBBUFFER3_CONFIG2"},
{0x20F, "GPUREG_ATTRIBBUFFER4_OFFSET"},
{0x210, "GPUREG_ATTRIBBUFFER4_CONFIG1"},
{0x211, "GPUREG_ATTRIBBUFFER4_CONFIG2"},
{0x212, "GPUREG_ATTRIBBUFFER5_OFFSET"},
{0x213, "GPUREG_ATTRIBBUFFER5_CONFIG1"},
{0x214, "GPUREG_ATTRIBBUFFER5_CONFIG2"},
{0x215, "GPUREG_ATTRIBBUFFER6_OFFSET"},
{0x216, "GPUREG_ATTRIBBUFFER6_CONFIG1"},
{0x217, "GPUREG_ATTRIBBUFFER6_CONFIG2"},
{0x218, "GPUREG_ATTRIBBUFFER7_OFFSET"},
{0x219, "GPUREG_ATTRIBBUFFER7_CONFIG1"},
{0x21A, "GPUREG_ATTRIBBUFFER7_CONFIG2"},
{0x21B, "GPUREG_ATTRIBBUFFER8_OFFSET"},
{0x21C, "GPUREG_ATTRIBBUFFER8_CONFIG1"},
{0x21D, "GPUREG_ATTRIBBUFFER8_CONFIG2"},
{0x21E, "GPUREG_ATTRIBBUFFER9_OFFSET"},
{0x21F, "GPUREG_ATTRIBBUFFER9_CONFIG1"},
{0x220, "GPUREG_ATTRIBBUFFER9_CONFIG2"},
{0x221, "GPUREG_ATTRIBBUFFER10_OFFSET"},
{0x222, "GPUREG_ATTRIBBUFFER10_CONFIG1"},
{0x223, "GPUREG_ATTRIBBUFFER10_CONFIG2"},
{0x224, "GPUREG_ATTRIBBUFFER11_OFFSET"},
{0x225, "GPUREG_ATTRIBBUFFER11_CONFIG1"},
{0x226, "GPUREG_ATTRIBBUFFER11_CONFIG2"},
{0x227, "GPUREG_INDEXBUFFER_CONFIG"},
{0x228, "GPUREG_NUMVERTICES"},
{0x229, "GPUREG_GEOSTAGE_CONFIG"},
{0x22A, "GPUREG_VERTEX_OFFSET"},
{0x22D, "GPUREG_POST_VERTEX_CACHE_NUM"},
{0x22E, "GPUREG_DRAWARRAYS"},
{0x22F, "GPUREG_DRAWELEMENTS"},
{0x231, "GPUREG_VTX_FUNC"},
{0x232, "GPUREG_FIXEDATTRIB_INDEX"},
{0x233, "GPUREG_FIXEDATTRIB_DATA0"},
{0x234, "GPUREG_FIXEDATTRIB_DATA1"},
{0x235, "GPUREG_FIXEDATTRIB_DATA2"},
{0x238, "GPUREG_CMDBUF_SIZE0"},
{0x239, "GPUREG_CMDBUF_SIZE1"},
{0x23A, "GPUREG_CMDBUF_ADDR0"},
{0x23B, "GPUREG_CMDBUF_ADDR1"},
{0x23C, "GPUREG_CMDBUF_JUMP0"},
{0x23D, "GPUREG_CMDBUF_JUMP1"},
{0x242, "GPUREG_VSH_NUM_ATTR"},
{0x244, "GPUREG_VSH_COM_MODE"},
{0x245, "GPUREG_START_DRAW_FUNC0"},
{0x24A, "GPUREG_VSH_OUTMAP_TOTAL1"},
{0x251, "GPUREG_VSH_OUTMAP_TOTAL2"},
{0x252, "GPUREG_GSH_MISC0"},
{0x253, "GPUREG_GEOSTAGE_CONFIG2"},
{0x254, "GPUREG_GSH_MISC1"},
{0x25E, "GPUREG_PRIMITIVE_CONFIG"},
{0x25F, "GPUREG_RESTART_PRIMITIVE"},
{0x280, "GPUREG_GSH_BOOLUNIFORM"},
{0x281, "GPUREG_GSH_INTUNIFORM_I0"},
{0x282, "GPUREG_GSH_INTUNIFORM_I1"},
{0x283, "GPUREG_GSH_INTUNIFORM_I2"},
{0x284, "GPUREG_GSH_INTUNIFORM_I3"},
{0x289, "GPUREG_GSH_INPUTBUFFER_CONFIG"},
{0x28A, "GPUREG_GSH_ENTRYPOINT"},
{0x28B, "GPUREG_GSH_ATTRIBUTES_PERMUTATION_LOW"},
{0x28C, "GPUREG_GSH_ATTRIBUTES_PERMUTATION_HIGH"},
{0x28D, "GPUREG_GSH_OUTMAP_MASK"},
{0x28F, "GPUREG_GSH_CODETRANSFER_END"},
{0x290, "GPUREG_GSH_FLOATUNIFORM_INDEX"},
{0x291, "GPUREG_GSH_FLOATUNIFORM_DATA0"},
{0x292, "GPUREG_GSH_FLOATUNIFORM_DATA1"},
{0x293, "GPUREG_GSH_FLOATUNIFORM_DATA2"},
{0x294, "GPUREG_GSH_FLOATUNIFORM_DATA3"},
{0x295, "GPUREG_GSH_FLOATUNIFORM_DATA4"},
{0x296, "GPUREG_GSH_FLOATUNIFORM_DATA5"},
{0x297, "GPUREG_GSH_FLOATUNIFORM_DATA6"},
{0x298, "GPUREG_GSH_FLOATUNIFORM_DATA7"},
{0x29B, "GPUREG_GSH_CODETRANSFER_INDEX"},
{0x29C, "GPUREG_GSH_CODETRANSFER_DATA0"},
{0x29D, "GPUREG_GSH_CODETRANSFER_DATA1"},
{0x29E, "GPUREG_GSH_CODETRANSFER_DATA2"},
{0x29F, "GPUREG_GSH_CODETRANSFER_DATA3"},
{0x2A0, "GPUREG_GSH_CODETRANSFER_DATA4"},
{0x2A1, "GPUREG_GSH_CODETRANSFER_DATA5"},
{0x2A2, "GPUREG_GSH_CODETRANSFER_DATA6"},
{0x2A3, "GPUREG_GSH_CODETRANSFER_DATA7"},
{0x2A5, "GPUREG_GSH_OPDESCS_INDEX"},
{0x2A6, "GPUREG_GSH_OPDESCS_DATA0"},
{0x2A7, "GPUREG_GSH_OPDESCS_DATA1"},
{0x2A8, "GPUREG_GSH_OPDESCS_DATA2"},
{0x2A9, "GPUREG_GSH_OPDESCS_DATA3"},
{0x2AA, "GPUREG_GSH_OPDESCS_DATA4"},
{0x2AB, "GPUREG_GSH_OPDESCS_DATA5"},
{0x2AC, "GPUREG_GSH_OPDESCS_DATA6"},
{0x2AD, "GPUREG_GSH_OPDESCS_DATA7"},
{0x2B0, "GPUREG_VSH_BOOLUNIFORM"},
{0x2B1, "GPUREG_VSH_INTUNIFORM_I0"},
{0x2B2, "GPUREG_VSH_INTUNIFORM_I1"},
{0x2B3, "GPUREG_VSH_INTUNIFORM_I2"},
{0x2B4, "GPUREG_VSH_INTUNIFORM_I3"},
{0x2B9, "GPUREG_VSH_INPUTBUFFER_CONFIG"},
{0x2BA, "GPUREG_VSH_ENTRYPOINT"},
{0x2BB, "GPUREG_VSH_ATTRIBUTES_PERMUTATION_LOW"},
{0x2BC, "GPUREG_VSH_ATTRIBUTES_PERMUTATION_HIGH"},
{0x2BD, "GPUREG_VSH_OUTMAP_MASK"},
{0x2BF, "GPUREG_VSH_CODETRANSFER_END"},
{0x2C0, "GPUREG_VSH_FLOATUNIFORM_INDEX"},
{0x2C1, "GPUREG_VSH_FLOATUNIFORM_DATA0"},
{0x2C2, "GPUREG_VSH_FLOATUNIFORM_DATA1"},
{0x2C3, "GPUREG_VSH_FLOATUNIFORM_DATA2"},
{0x2C4, "GPUREG_VSH_FLOATUNIFORM_DATA3"},
{0x2C5, "GPUREG_VSH_FLOATUNIFORM_DATA4"},
{0x2C6, "GPUREG_VSH_FLOATUNIFORM_DATA5"},
{0x2C7, "GPUREG_VSH_FLOATUNIFORM_DATA6"},
{0x2C8, "GPUREG_VSH_FLOATUNIFORM_DATA7"},
{0x2CB, "GPUREG_VSH_CODETRANSFER_INDEX"},
{0x2CC, "GPUREG_VSH_CODETRANSFER_DATA0"},
{0x2CD, "GPUREG_VSH_CODETRANSFER_DATA1"},
{0x2CE, "GPUREG_VSH_CODETRANSFER_DATA2"},
{0x2CF, "GPUREG_VSH_CODETRANSFER_DATA3"},
{0x2D0, "GPUREG_VSH_CODETRANSFER_DATA4"},
{0x2D1, "GPUREG_VSH_CODETRANSFER_DATA5"},
{0x2D2, "GPUREG_VSH_CODETRANSFER_DATA6"},
{0x2D3, "GPUREG_VSH_CODETRANSFER_DATA7"},
{0x2D5, "GPUREG_VSH_OPDESCS_INDEX"},
{0x2D6, "GPUREG_VSH_OPDESCS_DATA0"},
{0x2D7, "GPUREG_VSH_OPDESCS_DATA1"},
{0x2D8, "GPUREG_VSH_OPDESCS_DATA2"},
{0x2D9, "GPUREG_VSH_OPDESCS_DATA3"},
{0x2DA, "GPUREG_VSH_OPDESCS_DATA4"},
{0x2DB, "GPUREG_VSH_OPDESCS_DATA5"},
{0x2DC, "GPUREG_VSH_OPDESCS_DATA6"},
{0x2DD, "GPUREG_VSH_OPDESCS_DATA7"},
};
const char* Regs::GetRegisterName(u16 index) {
auto found = std::lower_bound(std::begin(register_names), std::end(register_names), index,
[](auto p, auto i) { return p.first < i; });
if (found->first == index) {
return found->second;
} else {
// Return empty string if no match is found
return "";
}
}
} // namespace Pica

149
src/video_core/regs.h
View file

@ -1,149 +0,0 @@
// Copyright 2017 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <array>
#include <cstddef>
#include <string>
#ifndef _MSC_VER
#include <type_traits> // for std::enable_if
#endif
#include "common/common_funcs.h"
#include "common/common_types.h"
#include "video_core/regs_framebuffer.h"
#include "video_core/regs_lighting.h"
#include "video_core/regs_pipeline.h"
#include "video_core/regs_rasterizer.h"
#include "video_core/regs_shader.h"
#include "video_core/regs_texturing.h"
namespace Pica {
// Returns index corresponding to the Regs member labeled by field_name
// TODO: Due to Visual studio bug 209229, offsetof does not return constant expressions
// when used with array elements (e.g. PICA_REG_INDEX(vs_uniform_setup.set_value[1])).
// For details cf.
// https://connect.microsoft.com/VisualStudio/feedback/details/209229/offsetof-does-not-produce-a-constant-expression-for-array-members
// Hopefully, this will be fixed sometime in the future.
// For lack of better alternatives, we currently hardcode the offsets when constant
// expressions are needed via PICA_REG_INDEX_WORKAROUND (on sane compilers, static_asserts
// will then make sure the offsets indeed match the automatically calculated ones).
#define PICA_REG_INDEX(field_name) (offsetof(Pica::Regs, field_name) / sizeof(u32))
#if defined(_MSC_VER)
#define PICA_REG_INDEX_WORKAROUND(field_name, backup_workaround_index) (backup_workaround_index)
#else
// NOTE: Yeah, hacking in a static_assert here just to workaround the lacking MSVC compiler
// really is this annoying. This macro just forwards its first argument to PICA_REG_INDEX
// and then performs a (no-op) cast to size_t iff the second argument matches the expected
// field offset. Otherwise, the compiler will fail to compile this code.
#define PICA_REG_INDEX_WORKAROUND(field_name, backup_workaround_index) \
((typename std::enable_if<backup_workaround_index == PICA_REG_INDEX(field_name), \
size_t>::type)PICA_REG_INDEX(field_name))
#endif // _MSC_VER
struct Regs {
static constexpr size_t NUM_REGS = 0x300;
union {
struct {
INSERT_PADDING_WORDS(0x10);
u32 trigger_irq;
INSERT_PADDING_WORDS(0x2f);
RasterizerRegs rasterizer;
TexturingRegs texturing;
FramebufferRegs framebuffer;
LightingRegs lighting;
PipelineRegs pipeline;
ShaderRegs gs;
ShaderRegs vs;
INSERT_PADDING_WORDS(0x20);
};
std::array<u32, NUM_REGS> reg_array;
};
/// Map register indices to names readable by humans
static const char* GetRegisterName(u16 index);
};
static_assert(sizeof(Regs) == Regs::NUM_REGS * sizeof(u32), "Regs struct has wrong size");
// TODO: MSVC does not support using offsetof() on non-static data members even though this
// is technically allowed since C++11. This macro should be enabled once MSVC adds
// support for that.
#ifndef _MSC_VER
#define ASSERT_REG_POSITION(field_name, position) \
static_assert(offsetof(Regs, field_name) == position * 4, \
"Field " #field_name " has invalid position")
ASSERT_REG_POSITION(trigger_irq, 0x10);
ASSERT_REG_POSITION(rasterizer, 0x40);
ASSERT_REG_POSITION(rasterizer.cull_mode, 0x40);
ASSERT_REG_POSITION(rasterizer.viewport_size_x, 0x41);
ASSERT_REG_POSITION(rasterizer.viewport_size_y, 0x43);
ASSERT_REG_POSITION(rasterizer.viewport_depth_range, 0x4d);
ASSERT_REG_POSITION(rasterizer.viewport_depth_near_plane, 0x4e);
ASSERT_REG_POSITION(rasterizer.vs_output_attributes[0], 0x50);
ASSERT_REG_POSITION(rasterizer.vs_output_attributes[1], 0x51);
ASSERT_REG_POSITION(rasterizer.scissor_test, 0x65);
ASSERT_REG_POSITION(rasterizer.viewport_corner, 0x68);
ASSERT_REG_POSITION(rasterizer.depthmap_enable, 0x6D);
ASSERT_REG_POSITION(texturing, 0x80);
ASSERT_REG_POSITION(texturing.main_config, 0x80);
ASSERT_REG_POSITION(texturing.texture0, 0x81);
ASSERT_REG_POSITION(texturing.texture0_format, 0x8e);
ASSERT_REG_POSITION(texturing.fragment_lighting_enable, 0x8f);
ASSERT_REG_POSITION(texturing.texture1, 0x91);
ASSERT_REG_POSITION(texturing.texture1_format, 0x96);
ASSERT_REG_POSITION(texturing.texture2, 0x99);
ASSERT_REG_POSITION(texturing.texture2_format, 0x9e);
ASSERT_REG_POSITION(texturing.proctex, 0xa8);
ASSERT_REG_POSITION(texturing.proctex_noise_u, 0xa9);
ASSERT_REG_POSITION(texturing.proctex_noise_v, 0xaa);
ASSERT_REG_POSITION(texturing.proctex_noise_frequency, 0xab);
ASSERT_REG_POSITION(texturing.proctex_lut, 0xac);
ASSERT_REG_POSITION(texturing.proctex_lut_offset, 0xad);
ASSERT_REG_POSITION(texturing.proctex_lut_config, 0xaf);
ASSERT_REG_POSITION(texturing.tev_stage0, 0xc0);
ASSERT_REG_POSITION(texturing.tev_stage1, 0xc8);
ASSERT_REG_POSITION(texturing.tev_stage2, 0xd0);
ASSERT_REG_POSITION(texturing.tev_stage3, 0xd8);
ASSERT_REG_POSITION(texturing.tev_combiner_buffer_input, 0xe0);
ASSERT_REG_POSITION(texturing.fog_mode, 0xe0);
ASSERT_REG_POSITION(texturing.fog_color, 0xe1);
ASSERT_REG_POSITION(texturing.fog_lut_offset, 0xe6);
ASSERT_REG_POSITION(texturing.fog_lut_data, 0xe8);
ASSERT_REG_POSITION(texturing.tev_stage4, 0xf0);
ASSERT_REG_POSITION(texturing.tev_stage5, 0xf8);
ASSERT_REG_POSITION(texturing.tev_combiner_buffer_color, 0xfd);
ASSERT_REG_POSITION(framebuffer, 0x100);
ASSERT_REG_POSITION(framebuffer.output_merger, 0x100);
ASSERT_REG_POSITION(framebuffer.framebuffer, 0x110);
ASSERT_REG_POSITION(lighting, 0x140);
ASSERT_REG_POSITION(pipeline, 0x200);
ASSERT_REG_POSITION(pipeline.vertex_attributes, 0x200);
ASSERT_REG_POSITION(pipeline.index_array, 0x227);
ASSERT_REG_POSITION(pipeline.num_vertices, 0x228);
ASSERT_REG_POSITION(pipeline.vertex_offset, 0x22a);
ASSERT_REG_POSITION(pipeline.trigger_draw, 0x22e);
ASSERT_REG_POSITION(pipeline.trigger_draw_indexed, 0x22f);
ASSERT_REG_POSITION(pipeline.vs_default_attributes_setup, 0x232);
ASSERT_REG_POSITION(pipeline.command_buffer, 0x238);
ASSERT_REG_POSITION(pipeline.gpu_mode, 0x245);
ASSERT_REG_POSITION(pipeline.triangle_topology, 0x25e);
ASSERT_REG_POSITION(pipeline.restart_primitive, 0x25f);
ASSERT_REG_POSITION(gs, 0x280);
ASSERT_REG_POSITION(vs, 0x2b0);
#undef ASSERT_REG_POSITION
#endif // !defined(_MSC_VER)
} // namespace Pica

283
src/video_core/regs_framebuffer.h
View file

@ -1,283 +0,0 @@
// Copyright 2017 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <array>
#include "common/assert.h"
#include "common/bit_field.h"
#include "common/common_funcs.h"
#include "common/common_types.h"
#include "common/logging/log.h"
namespace Pica {
struct FramebufferRegs {
enum class LogicOp : u32 {
Clear = 0,
And = 1,
AndReverse = 2,
Copy = 3,
Set = 4,
CopyInverted = 5,
NoOp = 6,
Invert = 7,
Nand = 8,
Or = 9,
Nor = 10,
Xor = 11,
Equiv = 12,
AndInverted = 13,
OrReverse = 14,
OrInverted = 15,
};
enum class BlendEquation : u32 {
Add = 0,
Subtract = 1,
ReverseSubtract = 2,
Min = 3,
Max = 4,
};
enum class BlendFactor : u32 {
Zero = 0,
One = 1,
SourceColor = 2,
OneMinusSourceColor = 3,
DestColor = 4,
OneMinusDestColor = 5,
SourceAlpha = 6,
OneMinusSourceAlpha = 7,
DestAlpha = 8,
OneMinusDestAlpha = 9,
ConstantColor = 10,
OneMinusConstantColor = 11,
ConstantAlpha = 12,
OneMinusConstantAlpha = 13,
SourceAlphaSaturate = 14,
};
enum class CompareFunc : u32 {
Never = 0,
Always = 1,
Equal = 2,
NotEqual = 3,
LessThan = 4,
LessThanOrEqual = 5,
GreaterThan = 6,
GreaterThanOrEqual = 7,
};
enum class StencilAction : u32 {
Keep = 0,
Zero = 1,
Replace = 2,
Increment = 3,
Decrement = 4,
Invert = 5,
IncrementWrap = 6,
DecrementWrap = 7,
};
struct {
union {
// If false, logic blending is used
BitField<8, 1, u32> alphablend_enable;
};
union {
BitField<0, 3, BlendEquation> blend_equation_rgb;
BitField<8, 3, BlendEquation> blend_equation_a;
BitField<16, 4, BlendFactor> factor_source_rgb;
BitField<20, 4, BlendFactor> factor_dest_rgb;
BitField<24, 4, BlendFactor> factor_source_a;
BitField<28, 4, BlendFactor> factor_dest_a;
} alpha_blending;
union {
BitField<0, 4, LogicOp> logic_op;
};
union {
u32 raw;
BitField<0, 8, u32> r;
BitField<8, 8, u32> g;
BitField<16, 8, u32> b;
BitField<24, 8, u32> a;
} blend_const;
union {
BitField<0, 1, u32> enable;
BitField<4, 3, CompareFunc> func;
BitField<8, 8, u32> ref;
} alpha_test;
struct {
union {
// Raw value of this register
u32 raw_func;
// If true, enable stencil testing
BitField<0, 1, u32> enable;
// Comparison operation for stencil testing
BitField<4, 3, CompareFunc> func;
// Mask used to control writing to the stencil buffer
BitField<8, 8, u32> write_mask;
// Value to compare against for stencil testing
BitField<16, 8, u32> reference_value;
// Mask to apply on stencil test inputs
BitField<24, 8, u32> input_mask;
};
union {
// Raw value of this register
u32 raw_op;
// Action to perform when the stencil test fails
BitField<0, 3, StencilAction> action_stencil_fail;
// Action to perform when stencil testing passed but depth testing fails
BitField<4, 3, StencilAction> action_depth_fail;
// Action to perform when both stencil and depth testing pass
BitField<8, 3, StencilAction> action_depth_pass;
};
} stencil_test;
union {
BitField<0, 1, u32> depth_test_enable;
BitField<4, 3, CompareFunc> depth_test_func;
BitField<8, 1, u32> red_enable;
BitField<9, 1, u32> green_enable;
BitField<10, 1, u32> blue_enable;
BitField<11, 1, u32> alpha_enable;
BitField<12, 1, u32> depth_write_enable;
};
INSERT_PADDING_WORDS(0x8);
} output_merger;
// Components are laid out in reverse byte order, most significant bits first.
enum class ColorFormat : u32 {
RGBA8 = 0,
RGB8 = 1,
RGB5A1 = 2,
RGB565 = 3,
RGBA4 = 4,
};
enum class DepthFormat : u32 {
D16 = 0,
D24 = 2,
D24S8 = 3,
};
// Returns the number of bytes in the specified color format
static unsigned BytesPerColorPixel(ColorFormat format) {
switch (format) {
case ColorFormat::RGBA8:
return 4;
case ColorFormat::RGB8:
return 3;
case ColorFormat::RGB5A1:
case ColorFormat::RGB565:
case ColorFormat::RGBA4:
return 2;
default:
LOG_CRITICAL(HW_GPU, "Unknown color format %u", format);
UNIMPLEMENTED();
}
}
struct FramebufferConfig {
INSERT_PADDING_WORDS(0x3);
union {
BitField<0, 4, u32> allow_color_write; // 0 = disable, else enable
};
INSERT_PADDING_WORDS(0x1);
union {
BitField<0, 2, u32> allow_depth_stencil_write; // 0 = disable, else enable
};
BitField<0, 2, DepthFormat> depth_format;
BitField<16, 3, ColorFormat> color_format;
INSERT_PADDING_WORDS(0x4);
BitField<0, 28, u32> depth_buffer_address;
BitField<0, 28, u32> color_buffer_address;
union {
// Apparently, the framebuffer width is stored as expected,
// while the height is stored as the actual height minus one.
// Hence, don't access these fields directly but use the accessors
// GetWidth() and GetHeight() instead.
BitField<0, 11, u32> width;
BitField<12, 10, u32> height;
};
INSERT_PADDING_WORDS(0x1);
inline PAddr GetColorBufferPhysicalAddress() const {
return color_buffer_address * 8;
}
inline PAddr GetDepthBufferPhysicalAddress() const {
return depth_buffer_address * 8;
}
inline u32 GetWidth() const {
return width;
}
inline u32 GetHeight() const {
return height + 1;
}
} framebuffer;
// Returns the number of bytes in the specified depth format
static u32 BytesPerDepthPixel(DepthFormat format) {
switch (format) {
case DepthFormat::D16:
return 2;
case DepthFormat::D24:
return 3;
case DepthFormat::D24S8:
return 4;
}
ASSERT_MSG(false, "Unknown depth format %u", format);
}
// Returns the number of bits per depth component of the specified depth format
static u32 DepthBitsPerPixel(DepthFormat format) {
switch (format) {
case DepthFormat::D16:
return 16;
case DepthFormat::D24:
case DepthFormat::D24S8:
return 24;
}
ASSERT_MSG(false, "Unknown depth format %u", format);
}
INSERT_PADDING_WORDS(0x20);
};
static_assert(sizeof(FramebufferRegs) == 0x40 * sizeof(u32),
"FramebufferRegs struct has incorrect size");
} // namespace Pica

321
src/video_core/regs_lighting.h
View file

@ -1,321 +0,0 @@
// Copyright 2017 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <array>
#include "common/assert.h"
#include "common/bit_field.h"
#include "common/common_funcs.h"
#include "common/common_types.h"
#include "common/vector_math.h"
namespace Pica {
struct LightingRegs {
enum class LightingSampler {
Distribution0 = 0,
Distribution1 = 1,
Fresnel = 3,
ReflectBlue = 4,
ReflectGreen = 5,
ReflectRed = 6,
SpotlightAttenuation = 8,
DistanceAttenuation = 16,
};
static constexpr unsigned NumLightingSampler = 24;
static LightingSampler SpotlightAttenuationSampler(unsigned index) {
return static_cast<LightingSampler>(
static_cast<unsigned>(LightingSampler::SpotlightAttenuation) + index);
}
static LightingSampler DistanceAttenuationSampler(unsigned index) {
return static_cast<LightingSampler>(
static_cast<unsigned>(LightingSampler::DistanceAttenuation) + index);
}
/**
* Pica fragment lighting supports using different LUTs for each lighting component: Reflectance
* R, G, and B channels, distribution function for specular components 0 and 1, fresnel factor,
* and spotlight attenuation. Furthermore, which LUTs are used for each channel (or whether a
* channel is enabled at all) is specified by various pre-defined lighting configurations. With
* configurations that require more LUTs, more cycles are required on HW to perform lighting
* computations.
*/
enum class LightingConfig : u32 {
Config0 = 0, ///< Reflect Red, Distribution 0, Spotlight
Config1 = 1, ///< Reflect Red, Fresnel, Spotlight
Config2 = 2, ///< Reflect Red, Distribution 0/1
Config3 = 3, ///< Distribution 0/1, Fresnel
Config4 = 4, ///< Reflect Red/Green/Blue, Distribution 0/1, Spotlight
Config5 = 5, ///< Reflect Red/Green/Blue, Distribution 0, Fresnel, Spotlight
Config6 = 6, ///< Reflect Red, Distribution 0/1, Fresnel, Spotlight
Config7 = 8, ///< Reflect Red/Green/Blue, Distribution 0/1, Fresnel, Spotlight
///< NOTE: '8' is intentional, '7' does not appear to be a valid configuration
};
/// Selects which lighting components are affected by fresnel
enum class LightingFresnelSelector : u32 {
None = 0, ///< Fresnel is disabled
PrimaryAlpha = 1, ///< Primary (diffuse) lighting alpha is affected by fresnel
SecondaryAlpha = 2, ///< Secondary (specular) lighting alpha is affected by fresnel
Both =
PrimaryAlpha |
SecondaryAlpha, ///< Both primary and secondary lighting alphas are affected by fresnel
};
/// Factor used to scale the output of a lighting LUT
enum class LightingScale : u32 {
Scale1 = 0, ///< Scale is 1x
Scale2 = 1, ///< Scale is 2x
Scale4 = 2, ///< Scale is 4x
Scale8 = 3, ///< Scale is 8x
Scale1_4 = 6, ///< Scale is 0.25x
Scale1_2 = 7, ///< Scale is 0.5x
};
enum class LightingLutInput : u32 {
NH = 0, // Cosine of the angle between the normal and half-angle vectors
VH = 1, // Cosine of the angle between the view and half-angle vectors
NV = 2, // Cosine of the angle between the normal and the view vector
LN = 3, // Cosine of the angle between the light and the normal vectors
SP = 4, // Cosine of the angle between the light and the inverse spotlight vectors
CP = 5, // Cosine of the angle between the tangent and projection of half-angle vectors
};
enum class LightingBumpMode : u32 {
None = 0,
NormalMap = 1,
TangentMap = 2,
};
union LightColor {
BitField<0, 10, u32> b;
BitField<10, 10, u32> g;
BitField<20, 10, u32> r;
Math::Vec3f ToVec3f() const {
// These fields are 10 bits wide, however 255 corresponds to 1.0f for each color
// component
return Math::MakeVec((f32)r / 255.f, (f32)g / 255.f, (f32)b / 255.f);
}
};
/// Returns true if the specified lighting sampler is supported by the current Pica lighting
/// configuration
static bool IsLightingSamplerSupported(LightingConfig config, LightingSampler sampler) {
switch (sampler) {
case LightingSampler::Distribution0:
return (config != LightingConfig::Config1);
case LightingSampler::Distribution1:
return (config != LightingConfig::Config0) && (config != LightingConfig::Config1) &&
(config != LightingConfig::Config5);
case LightingSampler::SpotlightAttenuation:
return (config != LightingConfig::Config2) && (config != LightingConfig::Config3);
case LightingSampler::Fresnel:
return (config != LightingConfig::Config0) && (config != LightingConfig::Config2) &&
(config != LightingConfig::Config4);
case LightingSampler::ReflectRed:
return (config != LightingConfig::Config3);
case LightingSampler::ReflectGreen:
case LightingSampler::ReflectBlue:
return (config == LightingConfig::Config4) || (config == LightingConfig::Config5) ||
(config == LightingConfig::Config7);
default:
UNREACHABLE_MSG("Regs::IsLightingSamplerSupported: Reached unreachable section, "
"sampler should be one of Distribution0, Distribution1, "
"SpotlightAttenuation, Fresnel, ReflectRed, ReflectGreen or "
"ReflectBlue, instead got %i",
static_cast<int>(config));
}
}
struct LightSrc {
LightColor specular_0; // material.specular_0 * light.specular_0
LightColor specular_1; // material.specular_1 * light.specular_1
LightColor diffuse; // material.diffuse * light.diffuse
LightColor ambient; // material.ambient * light.ambient
// Encoded as 16-bit floating point
union {
BitField<0, 16, u32> x;
BitField<16, 16, u32> y;
};
union {
BitField<0, 16, u32> z;
};
// inverse spotlight direction vector, encoded as fixed1.1.11
union {
BitField<0, 13, s32> spot_x;
BitField<16, 13, s32> spot_y;
};
union {
BitField<0, 13, s32> spot_z;
};
INSERT_PADDING_WORDS(0x1);
union {
BitField<0, 1, u32> directional;
BitField<1, 1, u32> two_sided_diffuse; // When disabled, clamp dot-product to 0
BitField<2, 1, u32> geometric_factor_0;
BitField<3, 1, u32> geometric_factor_1;
} config;
BitField<0, 20, u32> dist_atten_bias;
BitField<0, 20, u32> dist_atten_scale;
INSERT_PADDING_WORDS(0x4);
};
static_assert(sizeof(LightSrc) == 0x10 * sizeof(u32), "LightSrc structure must be 0x10 words");
LightSrc light[8];
LightColor global_ambient; // Emission + (material.ambient * lighting.ambient)
INSERT_PADDING_WORDS(0x1);
BitField<0, 3, u32> max_light_index; // Number of enabled lights - 1
union {
BitField<2, 2, LightingFresnelSelector> fresnel_selector;
BitField<4, 4, LightingConfig> config;
BitField<22, 2, u32> bump_selector; // 0: Texture 0, 1: Texture 1, 2: Texture 2
BitField<27, 1, u32> clamp_highlights;
BitField<28, 2, LightingBumpMode> bump_mode;
BitField<30, 1, u32> disable_bump_renorm;
} config0;
union {
u32 raw;
// Each bit specifies whether spot light attenuation should be applied for the corresponding
// light.
BitField<8, 8, u32> disable_spot_atten;
BitField<16, 1, u32> disable_lut_d0;
BitField<17, 1, u32> disable_lut_d1;
// Note: by intuition, BitField<18, 1, u32> should be disable_lut_sp, but it is actually a
// dummy bit which is always set as 1.
BitField<19, 1, u32> disable_lut_fr;
BitField<20, 1, u32> disable_lut_rr;
BitField<21, 1, u32> disable_lut_rg;
BitField<22, 1, u32> disable_lut_rb;
// Each bit specifies whether distance attenuation should be applied for the corresponding
// light.
BitField<24, 8, u32> disable_dist_atten;
} config1;
bool IsDistAttenDisabled(unsigned index) const {
return (config1.disable_dist_atten & (1 << index)) != 0;
}
bool IsSpotAttenDisabled(unsigned index) const {
return (config1.disable_spot_atten & (1 << index)) != 0;
}
union {
BitField<0, 8, u32> index; ///< Index at which to set data in the LUT
BitField<8, 5, u32> type; ///< Type of LUT for which to set data
} lut_config;
BitField<0, 1, u32> disable;
INSERT_PADDING_WORDS(0x1);
// When data is written to any of these registers, it gets written to the lookup table of the
// selected type at the selected index, specified above in the `lut_config` register. With each
// write, `lut_config.index` is incremented. It does not matter which of these registers is
// written to, the behavior will be the same.
u32 lut_data[8];
// These are used to specify if absolute (abs) value should be used for each LUT index. When
// abs mode is disabled, LUT indexes are in the range of (-1.0, 1.0). Otherwise, they are in
// the range of (0.0, 1.0).
union {
BitField<1, 1, u32> disable_d0;
BitField<5, 1, u32> disable_d1;
BitField<9, 1, u32> disable_sp;
BitField<13, 1, u32> disable_fr;
BitField<17, 1, u32> disable_rb;
BitField<21, 1, u32> disable_rg;
BitField<25, 1, u32> disable_rr;
} abs_lut_input;
union {
BitField<0, 3, LightingLutInput> d0;
BitField<4, 3, LightingLutInput> d1;
BitField<8, 3, LightingLutInput> sp;
BitField<12, 3, LightingLutInput> fr;
BitField<16, 3, LightingLutInput> rb;
BitField<20, 3, LightingLutInput> rg;
BitField<24, 3, LightingLutInput> rr;
} lut_input;
union {
BitField<0, 3, LightingScale> d0;
BitField<4, 3, LightingScale> d1;
BitField<8, 3, LightingScale> sp;
BitField<12, 3, LightingScale> fr;
BitField<16, 3, LightingScale> rb;
BitField<20, 3, LightingScale> rg;
BitField<24, 3, LightingScale> rr;
static float GetScale(LightingScale scale) {
switch (scale) {
case LightingScale::Scale1:
return 1.0f;
case LightingScale::Scale2:
return 2.0f;
case LightingScale::Scale4:
return 4.0f;
case LightingScale::Scale8:
return 8.0f;
case LightingScale::Scale1_4:
return 0.25f;
case LightingScale::Scale1_2:
return 0.5f;
}
return 0.0f;
}
} lut_scale;
INSERT_PADDING_WORDS(0x6);
union {
// There are 8 light enable "slots", corresponding to the total number of lights supported
// by Pica. For N enabled lights (specified by register 0x1c2, or 'src_num' above), the
// first N slots below will be set to integers within the range of 0-7, corresponding to the
// actual light that is enabled for each slot.
BitField<0, 3, u32> slot_0;
BitField<4, 3, u32> slot_1;
BitField<8, 3, u32> slot_2;
BitField<12, 3, u32> slot_3;
BitField<16, 3, u32> slot_4;
BitField<20, 3, u32> slot_5;
BitField<24, 3, u32> slot_6;
BitField<28, 3, u32> slot_7;
unsigned GetNum(unsigned index) const {
const unsigned enable_slots[] = {slot_0, slot_1, slot_2, slot_3,
slot_4, slot_5, slot_6, slot_7};
return enable_slots[index];
}
} light_enable;
INSERT_PADDING_WORDS(0x26);
};
static_assert(sizeof(LightingRegs) == 0xC0 * sizeof(u32), "LightingRegs struct has incorrect size");
} // namespace Pica

269
src/video_core/regs_pipeline.h
View file

@ -1,269 +0,0 @@
// Copyright 2017 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <array>
#include "common/assert.h"
#include "common/bit_field.h"
#include "common/common_funcs.h"
#include "common/common_types.h"
namespace Pica {
struct PipelineRegs {
enum class VertexAttributeFormat : u32 {
BYTE = 0,
UBYTE = 1,
SHORT = 2,
FLOAT = 3,
};
struct {
BitField<1, 28, u32> base_address;
PAddr GetPhysicalBaseAddress() const {
return base_address * 16;
}
// Descriptor for internal vertex attributes
union {
BitField<0, 2, VertexAttributeFormat> format0; // size of one element
BitField<2, 2, u32> size0; // number of elements minus 1
BitField<4, 2, VertexAttributeFormat> format1;
BitField<6, 2, u32> size1;
BitField<8, 2, VertexAttributeFormat> format2;
BitField<10, 2, u32> size2;
BitField<12, 2, VertexAttributeFormat> format3;
BitField<14, 2, u32> size3;
BitField<16, 2, VertexAttributeFormat> format4;
BitField<18, 2, u32> size4;
BitField<20, 2, VertexAttributeFormat> format5;
BitField<22, 2, u32> size5;
BitField<24, 2, VertexAttributeFormat> format6;
BitField<26, 2, u32> size6;
BitField<28, 2, VertexAttributeFormat> format7;
BitField<30, 2, u32> size7;
};
union {
BitField<0, 2, VertexAttributeFormat> format8;
BitField<2, 2, u32> size8;
BitField<4, 2, VertexAttributeFormat> format9;
BitField<6, 2, u32> size9;
BitField<8, 2, VertexAttributeFormat> format10;
BitField<10, 2, u32> size10;
BitField<12, 2, VertexAttributeFormat> format11;
BitField<14, 2, u32> size11;
BitField<16, 12, u32> attribute_mask;
// number of total attributes minus 1
BitField<28, 4, u32> max_attribute_index;
};
inline VertexAttributeFormat GetFormat(int n) const {
VertexAttributeFormat formats[] = {format0, format1, format2, format3,
format4, format5, format6, format7,
format8, format9, format10, format11};
return formats[n];
}
inline int GetNumElements(int n) const {
u32 sizes[] = {size0, size1, size2, size3, size4, size5,
size6, size7, size8, size9, size10, size11};
return (int)sizes[n] + 1;
}
inline int GetElementSizeInBytes(int n) const {
return (GetFormat(n) == VertexAttributeFormat::FLOAT)
? 4
: (GetFormat(n) == VertexAttributeFormat::SHORT) ? 2 : 1;
}
inline int GetStride(int n) const {
return GetNumElements(n) * GetElementSizeInBytes(n);
}
inline bool IsDefaultAttribute(int id) const {
return (id >= 12) || (attribute_mask & (1ULL << id)) != 0;
}
inline int GetNumTotalAttributes() const {
return (int)max_attribute_index + 1;
}
// Attribute loaders map the source vertex data to input attributes
// This e.g. allows to load different attributes from different memory locations
struct {
// Source attribute data offset from the base address
BitField<0, 28, u32> data_offset;
union {
BitField<0, 4, u32> comp0;
BitField<4, 4, u32> comp1;
BitField<8, 4, u32> comp2;
BitField<12, 4, u32> comp3;
BitField<16, 4, u32> comp4;
BitField<20, 4, u32> comp5;
BitField<24, 4, u32> comp6;
BitField<28, 4, u32> comp7;
};
union {
BitField<0, 4, u32> comp8;
BitField<4, 4, u32> comp9;
BitField<8, 4, u32> comp10;
BitField<12, 4, u32> comp11;
// bytes for a single vertex in this loader
BitField<16, 8, u32> byte_count;
BitField<28, 4, u32> component_count;
};
inline int GetComponent(int n) const {
u32 components[] = {comp0, comp1, comp2, comp3, comp4, comp5,
comp6, comp7, comp8, comp9, comp10, comp11};
return (int)components[n];
}
} attribute_loaders[12];
} vertex_attributes;
struct {
enum IndexFormat : u32 {
BYTE = 0,
SHORT = 1,
};
union {
BitField<0, 31, u32> offset; // relative to base attribute address
BitField<31, 1, IndexFormat> format;
};
} index_array;
// Number of vertices to render
u32 num_vertices;
enum class UseGS : u32 {
No = 0,
Yes = 2,
};
union {
BitField<0, 2, UseGS> use_gs;
BitField<31, 1, u32> variable_primitive;
};
// The index of the first vertex to render
u32 vertex_offset;
INSERT_PADDING_WORDS(0x3);
// These two trigger rendering of triangles
u32 trigger_draw;
u32 trigger_draw_indexed;
INSERT_PADDING_WORDS(0x2);
// These registers are used to setup the default "fall-back" vertex shader attributes
struct {
// Index of the current default attribute
u32 index;
// Writing to these registers sets the "current" default attribute.
u32 set_value[3];
} vs_default_attributes_setup;
INSERT_PADDING_WORDS(0x2);
struct {
// There are two channels that can be used to configure the next command buffer, which can
// be then executed by writing to the "trigger" registers. There are two reasons why a game
// might use this feature:
// 1) With this, an arbitrary number of additional command buffers may be executed in
// sequence without requiring any intervention of the CPU after the initial one is
// kicked off.
// 2) Games can configure these registers to provide a command list subroutine mechanism.
// TODO: verify the bit length of these two fields
// According to 3dbrew, the bit length of them are 21 and 29, respectively
BitField<0, 20, u32> size[2]; ///< Size (in bytes / 8) of each channel's command buffer
BitField<0, 28, u32> addr[2]; ///< Physical address / 8 of each channel's command buffer
u32 trigger[2]; ///< Triggers execution of the channel's command buffer when written to
unsigned GetSize(unsigned index) const {
ASSERT(index < 2);
return 8 * size[index];
}
PAddr GetPhysicalAddress(unsigned index) const {
ASSERT(index < 2);
return (PAddr)(8 * addr[index]);
}
} command_buffer;
INSERT_PADDING_WORDS(4);
/// Number of input attributes to the vertex shader minus 1
BitField<0, 4, u32> max_input_attrib_index;
INSERT_PADDING_WORDS(1);
// The shader unit 3, which can be used for both vertex and geometry shader, gets its
// configuration depending on this register. If this is not set, unit 3 will share some
// configuration with other units. It is known that program code and swizzle pattern uploaded
// via regs.vs will be also uploaded to unit 3 if this is not set. Although very likely, it is
// still unclear whether uniforms and other configuration can be also shared.
BitField<0, 1, u32> gs_unit_exclusive_configuration;
enum class GPUMode : u32 {
Drawing = 0,
Configuring = 1,
};
GPUMode gpu_mode;
INSERT_PADDING_WORDS(0x4);
BitField<0, 4, u32> vs_outmap_total_minus_1_a;
INSERT_PADDING_WORDS(0x6);
BitField<0, 4, u32> vs_outmap_total_minus_1_b;
enum class GSMode : u32 {
Point = 0,
VariablePrimitive = 1,
FixedPrimitive = 2,
};
union {
BitField<0, 8, GSMode> mode;
BitField<8, 4, u32> fixed_vertex_num_minus_1;
BitField<12, 4, u32> stride_minus_1;
BitField<16, 4, u32> start_index;
} gs_config;
INSERT_PADDING_WORDS(0x1);
u32 variable_vertex_main_num_minus_1;
INSERT_PADDING_WORDS(0x9);
enum class TriangleTopology : u32 {
List = 0,
Strip = 1,
Fan = 2,
Shader = 3, // Programmable setup unit implemented in a geometry shader
};
BitField<8, 2, TriangleTopology> triangle_topology;
u32 restart_primitive;
INSERT_PADDING_WORDS(0x20);
};
static_assert(sizeof(PipelineRegs) == 0x80 * sizeof(u32), "PipelineRegs struct has incorrect size");
} // namespace Pica

139
src/video_core/regs_rasterizer.h
View file

@ -1,139 +0,0 @@
// Copyright 2017 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <array>
#include "common/bit_field.h"
#include "common/common_funcs.h"
#include "common/common_types.h"
#include "video_core/pica_types.h"
namespace Pica {
struct RasterizerRegs {
enum class CullMode : u32 {
// Select which polygons are considered to be "frontfacing".
KeepAll = 0,
KeepClockWise = 1,
KeepCounterClockWise = 2,
// TODO: What does the third value imply?
};
union {
BitField<0, 2, CullMode> cull_mode;
};
BitField<0, 24, u32> viewport_size_x;
INSERT_PADDING_WORDS(0x1);
BitField<0, 24, u32> viewport_size_y;
INSERT_PADDING_WORDS(0x3);
BitField<0, 1, u32> clip_enable;
BitField<0, 24, u32> clip_coef[4]; // float24
Math::Vec4<float24> GetClipCoef() const {
return {float24::FromRaw(clip_coef[0]), float24::FromRaw(clip_coef[1]),
float24::FromRaw(clip_coef[2]), float24::FromRaw(clip_coef[3])};
}
INSERT_PADDING_WORDS(0x1);
BitField<0, 24, u32> viewport_depth_range; // float24
BitField<0, 24, u32> viewport_depth_near_plane; // float24
BitField<0, 3, u32> vs_output_total;
union VSOutputAttributes {
// Maps components of output vertex attributes to semantics
enum Semantic : u32 {
POSITION_X = 0,
POSITION_Y = 1,
POSITION_Z = 2,
POSITION_W = 3,
QUATERNION_X = 4,
QUATERNION_Y = 5,
QUATERNION_Z = 6,
QUATERNION_W = 7,
COLOR_R = 8,
COLOR_G = 9,
COLOR_B = 10,
COLOR_A = 11,
TEXCOORD0_U = 12,
TEXCOORD0_V = 13,
TEXCOORD1_U = 14,
TEXCOORD1_V = 15,
TEXCOORD0_W = 16,
VIEW_X = 18,
VIEW_Y = 19,
VIEW_Z = 20,
TEXCOORD2_U = 22,
TEXCOORD2_V = 23,
INVALID = 31,
};
BitField<0, 5, Semantic> map_x;
BitField<8, 5, Semantic> map_y;
BitField<16, 5, Semantic> map_z;
BitField<24, 5, Semantic> map_w;
} vs_output_attributes[7];
INSERT_PADDING_WORDS(0xe);
enum class ScissorMode : u32 {
Disabled = 0,
Exclude = 1, // Exclude pixels inside the scissor box
Include = 3 // Exclude pixels outside the scissor box
};
struct {
BitField<0, 2, ScissorMode> mode;
union {
BitField<0, 10, u32> x1;
BitField<16, 10, u32> y1;
};
union {
BitField<0, 10, u32> x2;
BitField<16, 10, u32> y2;
};
} scissor_test;
union {
BitField<0, 10, s32> x;
BitField<16, 10, s32> y;
} viewport_corner;
INSERT_PADDING_WORDS(0x1);
// TODO: early depth
INSERT_PADDING_WORDS(0x1);
INSERT_PADDING_WORDS(0x2);
enum DepthBuffering : u32 {
WBuffering = 0,
ZBuffering = 1,
};
BitField<0, 1, DepthBuffering> depthmap_enable;
INSERT_PADDING_WORDS(0x12);
};
static_assert(sizeof(RasterizerRegs) == 0x40 * sizeof(u32),
"RasterizerRegs struct has incorrect size");
} // namespace Pica

111
src/video_core/regs_shader.h
View file

@ -1,111 +0,0 @@
// Copyright 2017 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <array>
#include "common/bit_field.h"
#include "common/common_funcs.h"
#include "common/common_types.h"
namespace Pica {
struct ShaderRegs {
BitField<0, 16, u32> bool_uniforms;
union {
BitField<0, 8, u32> x;
BitField<8, 8, u32> y;
BitField<16, 8, u32> z;
BitField<24, 8, u32> w;
} int_uniforms[4];
INSERT_PADDING_WORDS(0x4);
enum ShaderMode {
GS = 0x08,
VS = 0xA0,
};
union {
// Number of input attributes to shader unit - 1
BitField<0, 4, u32> max_input_attribute_index;
BitField<8, 8, u32> input_to_uniform;
BitField<24, 8, ShaderMode> shader_mode;
};
// Offset to shader program entry point (in words)
BitField<0, 16, u32> main_offset;
/// Maps input attributes to registers. 4-bits per attribute, specifying a register index
u32 input_attribute_to_register_map_low;
u32 input_attribute_to_register_map_high;
unsigned int GetRegisterForAttribute(unsigned int attribute_index) const {
u64 map = ((u64)input_attribute_to_register_map_high << 32) |
(u64)input_attribute_to_register_map_low;
return (map >> (attribute_index * 4)) & 0b1111;
}
BitField<0, 16, u32> output_mask;
// 0x28E, CODETRANSFER_END
INSERT_PADDING_WORDS(0x2);
struct {
enum Format : u32 {
FLOAT24 = 0,
FLOAT32 = 1,
};
bool IsFloat32() const {
return format == FLOAT32;
}
union {
// Index of the next uniform to write to
// TODO: ctrulib uses 8 bits for this, however that seems to yield lots of invalid
// indices
// TODO: Maybe the uppermost index is for the geometry shader? Investigate!
BitField<0, 7, u32> index;
BitField<31, 1, Format> format;
};
// Writing to these registers sets the current uniform.
u32 set_value[8];
} uniform_setup;
INSERT_PADDING_WORDS(0x2);
struct {
// Offset of the next instruction to write code to.
// Incremented with each instruction write.
u32 offset;
// Writing to these registers sets the "current" word in the shader program.
u32 set_word[8];
} program;
INSERT_PADDING_WORDS(0x1);
// This register group is used to load an internal table of swizzling patterns,
// which are indexed by each shader instruction to specify vector component swizzling.
struct {
// Offset of the next swizzle pattern to write code to.
// Incremented with each instruction write.
u32 offset;
// Writing to these registers sets the current swizzle pattern in the table.
u32 set_word[8];
} swizzle_patterns;
INSERT_PADDING_WORDS(0x2);
};
static_assert(sizeof(ShaderRegs) == 0x30 * sizeof(u32), "ShaderRegs struct has incorrect size");
} // namespace Pica

452
src/video_core/regs_texturing.h
View file

@ -1,452 +0,0 @@
// Copyright 2017 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <array>
#include "common/assert.h"
#include "common/bit_field.h"
#include "common/common_funcs.h"
#include "common/common_types.h"
namespace Pica {
struct TexturingRegs {
struct TextureConfig {
enum TextureType : u32 {
Texture2D = 0,
TextureCube = 1,
Shadow2D = 2,
Projection2D = 3,
ShadowCube = 4,
Disabled = 5,
};
enum WrapMode : u32 {
ClampToEdge = 0,
ClampToBorder = 1,
Repeat = 2,
MirroredRepeat = 3,
// Mode 4-7 produces some weird result and may be just invalid:
ClampToEdge2 = 4, // Positive coord: clamp to edge; negative coord: repeat
ClampToBorder2 = 5, // Positive coord: clamp to border; negative coord: repeat
Repeat2 = 6, // Same as Repeat
Repeat3 = 7, // Same as Repeat
};
enum TextureFilter : u32 {
Nearest = 0,
Linear = 1,
};
union {
u32 raw;
BitField<0, 8, u32> r;
BitField<8, 8, u32> g;
BitField<16, 8, u32> b;
BitField<24, 8, u32> a;
} border_color;
union {
BitField<0, 11, u32> height;
BitField<16, 11, u32> width;
};
union {
BitField<1, 1, TextureFilter> mag_filter;
BitField<2, 1, TextureFilter> min_filter;
BitField<8, 3, WrapMode> wrap_t;
BitField<12, 3, WrapMode> wrap_s;
/// @note Only valid for texture 0 according to 3DBrew.
BitField<28, 3, TextureType> type;
};
INSERT_PADDING_WORDS(0x1);
BitField<0, 28, u32> address;
PAddr GetPhysicalAddress() const {
return address * 8;
}
// texture1 and texture2 store the texture format directly after the address
// whereas texture0 inserts some additional flags inbetween.
// Hence, we store the format separately so that all other parameters can be described
// in a single structure.
};
enum class TextureFormat : u32 {
RGBA8 = 0,
RGB8 = 1,
RGB5A1 = 2,
RGB565 = 3,
RGBA4 = 4,
IA8 = 5,
RG8 = 6, ///< @note Also called HILO8 in 3DBrew.
I8 = 7,
A8 = 8,
IA4 = 9,
I4 = 10,
A4 = 11,
ETC1 = 12, // compressed
ETC1A4 = 13, // compressed
};
static unsigned NibblesPerPixel(TextureFormat format) {
switch (format) {
case TextureFormat::RGBA8:
return 8;
case TextureFormat::RGB8:
return 6;
case TextureFormat::RGB5A1:
case TextureFormat::RGB565:
case TextureFormat::RGBA4:
case TextureFormat::IA8:
case TextureFormat::RG8:
return 4;
case TextureFormat::I4:
case TextureFormat::A4:
return 1;
case TextureFormat::I8:
case TextureFormat::A8:
case TextureFormat::IA4:
default: // placeholder for yet unknown formats
UNIMPLEMENTED();
return 0;
}
}
union {
BitField<0, 1, u32> texture0_enable;
BitField<1, 1, u32> texture1_enable;
BitField<2, 1, u32> texture2_enable;
BitField<8, 2, u32> texture3_coordinates;
BitField<10, 1, u32> texture3_enable;
BitField<13, 1, u32> texture2_use_coord1;
BitField<16, 1, u32> clear_texture_cache; // TODO: unimplemented
} main_config;
TextureConfig texture0;
enum class CubeFace {
PositiveX = 0,
NegativeX = 1,
PositiveY = 2,
NegativeY = 3,
PositiveZ = 4,
NegativeZ = 5,
};
BitField<0, 22, u32> cube_address[5];
PAddr GetCubePhysicalAddress(CubeFace face) const {
PAddr address = texture0.address;
if (face != CubeFace::PositiveX) {
// Bits [22:27] from the main texture address is shared with all cubemap additional
// addresses.
auto& face_addr = cube_address[static_cast<size_t>(face) - 1];
address &= ~face_addr.mask;
address |= face_addr;
}
// A multiplier of 8 is also needed in the same way as the main address.
return address * 8;
}
INSERT_PADDING_WORDS(0x3);
BitField<0, 4, TextureFormat> texture0_format;
BitField<0, 1, u32> fragment_lighting_enable;
INSERT_PADDING_WORDS(0x1);
TextureConfig texture1;
BitField<0, 4, TextureFormat> texture1_format;
INSERT_PADDING_WORDS(0x2);
TextureConfig texture2;
BitField<0, 4, TextureFormat> texture2_format;
INSERT_PADDING_WORDS(0x9);
struct FullTextureConfig {
const bool enabled;
const TextureConfig config;
const TextureFormat format;
};
const std::array<FullTextureConfig, 3> GetTextures() const {
return {{
{main_config.texture0_enable.ToBool(), texture0, texture0_format},
{main_config.texture1_enable.ToBool(), texture1, texture1_format},
{main_config.texture2_enable.ToBool(), texture2, texture2_format},
}};
}
// 0xa8-0xad: ProcTex Config
enum class ProcTexClamp : u32 {
ToZero = 0,
ToEdge = 1,
SymmetricalRepeat = 2,
MirroredRepeat = 3,
Pulse = 4,
};
enum class ProcTexCombiner : u32 {
U = 0, // u
U2 = 1, // u * u
V = 2, // v
V2 = 3, // v * v
Add = 4, // (u + v) / 2
Add2 = 5, // (u * u + v * v) / 2
SqrtAdd2 = 6, // sqrt(u * u + v * v)
Min = 7, // min(u, v)
Max = 8, // max(u, v)
RMax = 9, // Average of Max and SqrtAdd2
};
enum class ProcTexShift : u32 {
None = 0,
Odd = 1,
Even = 2,
};
union {
BitField<0, 3, ProcTexClamp> u_clamp;
BitField<3, 3, ProcTexClamp> v_clamp;
BitField<6, 4, ProcTexCombiner> color_combiner;
BitField<10, 4, ProcTexCombiner> alpha_combiner;
BitField<14, 1, u32> separate_alpha;
BitField<15, 1, u32> noise_enable;
BitField<16, 2, ProcTexShift> u_shift;
BitField<18, 2, ProcTexShift> v_shift;
BitField<20, 8, u32> bias_low; // float16 TODO: unimplemented
} proctex;
union ProcTexNoiseConfig {
BitField<0, 16, s32> amplitude; // fixed1.3.12
BitField<16, 16, u32> phase; // float16
};
ProcTexNoiseConfig proctex_noise_u;
ProcTexNoiseConfig proctex_noise_v;
union {
BitField<0, 16, u32> u; // float16
BitField<16, 16, u32> v; // float16
} proctex_noise_frequency;
enum class ProcTexFilter : u32 {
Nearest = 0,
Linear = 1,
NearestMipmapNearest = 2,
LinearMipmapNearest = 3,
NearestMipmapLinear = 4,
LinearMipmapLinear = 5,
};
union {
BitField<0, 3, ProcTexFilter> filter;
BitField<11, 8, u32> width;
BitField<19, 8, u32> bias_high; // TODO: unimplemented
} proctex_lut;
BitField<0, 8, u32> proctex_lut_offset;
INSERT_PADDING_WORDS(0x1);
// 0xaf-0xb7: ProcTex LUT
enum class ProcTexLutTable : u32 {
Noise = 0,
ColorMap = 2,
AlphaMap = 3,
Color = 4,
ColorDiff = 5,
};
union {
BitField<0, 8, u32> index;
BitField<8, 4, ProcTexLutTable> ref_table;
} proctex_lut_config;
u32 proctex_lut_data[8];
INSERT_PADDING_WORDS(0x8);
// 0xc0-0xff: Texture Combiner (akin to glTexEnv)
struct TevStageConfig {
enum class Source : u32 {
PrimaryColor = 0x0,
PrimaryFragmentColor = 0x1,
SecondaryFragmentColor = 0x2,
Texture0 = 0x3,
Texture1 = 0x4,
Texture2 = 0x5,
Texture3 = 0x6,
PreviousBuffer = 0xd,
Constant = 0xe,
Previous = 0xf,
};
enum class ColorModifier : u32 {
SourceColor = 0x0,
OneMinusSourceColor = 0x1,
SourceAlpha = 0x2,
OneMinusSourceAlpha = 0x3,
SourceRed = 0x4,
OneMinusSourceRed = 0x5,
SourceGreen = 0x8,
OneMinusSourceGreen = 0x9,
SourceBlue = 0xc,
OneMinusSourceBlue = 0xd,
};
enum class AlphaModifier : u32 {
SourceAlpha = 0x0,
OneMinusSourceAlpha = 0x1,
SourceRed = 0x2,
OneMinusSourceRed = 0x3,
SourceGreen = 0x4,
OneMinusSourceGreen = 0x5,
SourceBlue = 0x6,
OneMinusSourceBlue = 0x7,
};
enum class Operation : u32 {
Replace = 0,
Modulate = 1,
Add = 2,
AddSigned = 3,
Lerp = 4,
Subtract = 5,
Dot3_RGB = 6,
Dot3_RGBA = 7,
MultiplyThenAdd = 8,
AddThenMultiply = 9,
};
union {
u32 sources_raw;
BitField<0, 4, Source> color_source1;
BitField<4, 4, Source> color_source2;
BitField<8, 4, Source> color_source3;
BitField<16, 4, Source> alpha_source1;
BitField<20, 4, Source> alpha_source2;
BitField<24, 4, Source> alpha_source3;
};
union {
u32 modifiers_raw;
BitField<0, 4, ColorModifier> color_modifier1;
BitField<4, 4, ColorModifier> color_modifier2;
BitField<8, 4, ColorModifier> color_modifier3;
BitField<12, 3, AlphaModifier> alpha_modifier1;
BitField<16, 3, AlphaModifier> alpha_modifier2;
BitField<20, 3, AlphaModifier> alpha_modifier3;
};
union {
u32 ops_raw;
BitField<0, 4, Operation> color_op;
BitField<16, 4, Operation> alpha_op;
};
union {
u32 const_color;
BitField<0, 8, u32> const_r;
BitField<8, 8, u32> const_g;
BitField<16, 8, u32> const_b;
BitField<24, 8, u32> const_a;
};
union {
u32 scales_raw;
BitField<0, 2, u32> color_scale;
BitField<16, 2, u32> alpha_scale;
};
inline unsigned GetColorMultiplier() const {
return (color_scale < 3) ? (1 << color_scale) : 1;
}
inline unsigned GetAlphaMultiplier() const {
return (alpha_scale < 3) ? (1 << alpha_scale) : 1;
}
};
TevStageConfig tev_stage0;
INSERT_PADDING_WORDS(0x3);
TevStageConfig tev_stage1;
INSERT_PADDING_WORDS(0x3);
TevStageConfig tev_stage2;
INSERT_PADDING_WORDS(0x3);
TevStageConfig tev_stage3;
INSERT_PADDING_WORDS(0x3);
enum class FogMode : u32 {
None = 0,
Fog = 5,
Gas = 7,
};
union {
BitField<0, 3, FogMode> fog_mode;
BitField<16, 1, u32> fog_flip;
union {
// Tev stages 0-3 write their output to the combiner buffer if the corresponding bit in
// these masks are set
BitField<8, 4, u32> update_mask_rgb;
BitField<12, 4, u32> update_mask_a;
bool TevStageUpdatesCombinerBufferColor(unsigned stage_index) const {
return (stage_index < 4) && (update_mask_rgb & (1 << stage_index));
}
bool TevStageUpdatesCombinerBufferAlpha(unsigned stage_index) const {
return (stage_index < 4) && (update_mask_a & (1 << stage_index));
}
} tev_combiner_buffer_input;
};
union {
u32 raw;
BitField<0, 8, u32> r;
BitField<8, 8, u32> g;
BitField<16, 8, u32> b;
} fog_color;
INSERT_PADDING_WORDS(0x4);
BitField<0, 16, u32> fog_lut_offset;
INSERT_PADDING_WORDS(0x1);
u32 fog_lut_data[8];
TevStageConfig tev_stage4;
INSERT_PADDING_WORDS(0x3);
TevStageConfig tev_stage5;
union {
u32 raw;
BitField<0, 8, u32> r;
BitField<8, 8, u32> g;
BitField<16, 8, u32> b;
BitField<24, 8, u32> a;
} tev_combiner_buffer_color;
INSERT_PADDING_WORDS(0x2);
const std::array<TevStageConfig, 6> GetTevStages() const {
return {{tev_stage0, tev_stage1, tev_stage2, tev_stage3, tev_stage4, tev_stage5}};
};
};
static_assert(sizeof(TexturingRegs) == 0x80 * sizeof(u32),
"TexturingRegs struct has incorrect size");
} // namespace Pica

15
src/video_core/renderer_base.cpp
View file

@ -5,19 +5,6 @@
#include <atomic>
#include <memory>
#include "video_core/renderer_base.h"
#include "video_core/renderer_opengl/gl_rasterizer.h"
#include "video_core/swrasterizer/swrasterizer.h"
#include "video_core/video_core.h"
void RendererBase::RefreshRasterizerSetting() {
bool hw_renderer_enabled = VideoCore::g_hw_renderer_enabled;
if (rasterizer == nullptr || opengl_rasterizer_active != hw_renderer_enabled) {
opengl_rasterizer_active = hw_renderer_enabled;
if (hw_renderer_enabled) {
rasterizer = std::make_unique<RasterizerOpenGL>();
} else {
rasterizer = std::make_unique<VideoCore::SWRasterizer>();
}
}
}
void RendererBase::RefreshRasterizerSetting() {}

7
src/video_core/renderer_base.h
View file

@ -5,8 +5,8 @@
#pragma once
#include <memory>
#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/rasterizer_interface.h"
class EmuWindow;
@ -72,14 +72,9 @@ public:
return m_current_frame;
}
VideoCore::RasterizerInterface* Rasterizer() const {
return rasterizer.get();
}
void RefreshRasterizerSetting();
protected:
std::unique_ptr<VideoCore::RasterizerInterface> rasterizer;
f32 m_current_fps = 0.0f; ///< Current framerate, should be set by the renderer
int m_current_frame = 0; ///< Current frame, should be set by the renderer

1686
src/video_core/renderer_opengl/gl_rasterizer.cpp
View file

File diff suppressed because it is too large Load diff

316
src/video_core/renderer_opengl/gl_rasterizer.h
View file

@ -1,316 +0,0 @@
// Copyright 2015 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <array>
#include <cstddef>
#include <cstring>
#include <memory>
#include <unordered_map>
#include <vector>
#include <glad/glad.h>
#include "common/bit_field.h"
#include "common/common_types.h"
#include "common/hash.h"
#include "common/vector_math.h"
#include "core/hw/gpu.h"
#include "video_core/pica_state.h"
#include "video_core/pica_types.h"
#include "video_core/rasterizer_interface.h"
#include "video_core/regs_framebuffer.h"
#include "video_core/regs_lighting.h"
#include "video_core/regs_rasterizer.h"
#include "video_core/regs_texturing.h"
#include "video_core/renderer_opengl/gl_rasterizer_cache.h"
#include "video_core/renderer_opengl/gl_resource_manager.h"
#include "video_core/renderer_opengl/gl_shader_gen.h"
#include "video_core/renderer_opengl/gl_state.h"
#include "video_core/renderer_opengl/pica_to_gl.h"
#include "video_core/shader/shader.h"
struct ScreenInfo;
class RasterizerOpenGL : public VideoCore::RasterizerInterface {
public:
RasterizerOpenGL();
~RasterizerOpenGL() override;
void AddTriangle(const Pica::Shader::OutputVertex& v0, const Pica::Shader::OutputVertex& v1,
const Pica::Shader::OutputVertex& v2) override;
void DrawTriangles() override;
void NotifyPicaRegisterChanged(u32 id) override;
void FlushAll() override;
void FlushRegion(PAddr addr, u64 size) override;
void FlushAndInvalidateRegion(PAddr addr, u64 size) override;
bool AccelerateDisplayTransfer(const GPU::Regs::DisplayTransferConfig& config) override;
bool AccelerateTextureCopy(const GPU::Regs::DisplayTransferConfig& config) override;
bool AccelerateFill(const GPU::Regs::MemoryFillConfig& config) override;
bool AccelerateDisplay(const GPU::Regs::FramebufferConfig& config, PAddr framebuffer_addr,
u32 pixel_stride, ScreenInfo& screen_info) override;
/// OpenGL shader generated for a given Pica register state
struct PicaShader {
/// OpenGL shader resource
OGLShader shader;
};
private:
struct SamplerInfo {
using TextureConfig = Pica::TexturingRegs::TextureConfig;
OGLSampler sampler;
/// Creates the sampler object, initializing its state so that it's in sync with the
/// SamplerInfo struct.
void Create();
/// Syncs the sampler object with the config, updating any necessary state.
void SyncWithConfig(const TextureConfig& config);
private:
TextureConfig::TextureFilter mag_filter;
TextureConfig::TextureFilter min_filter;
TextureConfig::WrapMode wrap_s;
TextureConfig::WrapMode wrap_t;
u32 border_color;
};
/// Structure that the hardware rendered vertices are composed of
struct HardwareVertex {
HardwareVertex(const Pica::Shader::OutputVertex& v, bool flip_quaternion) {
position[0] = v.pos.x.ToFloat32();
position[1] = v.pos.y.ToFloat32();
position[2] = v.pos.z.ToFloat32();
position[3] = v.pos.w.ToFloat32();
color[0] = v.color.x.ToFloat32();
color[1] = v.color.y.ToFloat32();
color[2] = v.color.z.ToFloat32();
color[3] = v.color.w.ToFloat32();
tex_coord0[0] = v.tc0.x.ToFloat32();
tex_coord0[1] = v.tc0.y.ToFloat32();
tex_coord1[0] = v.tc1.x.ToFloat32();
tex_coord1[1] = v.tc1.y.ToFloat32();
tex_coord2[0] = v.tc2.x.ToFloat32();
tex_coord2[1] = v.tc2.y.ToFloat32();
tex_coord0_w = v.tc0_w.ToFloat32();
normquat[0] = v.quat.x.ToFloat32();
normquat[1] = v.quat.y.ToFloat32();
normquat[2] = v.quat.z.ToFloat32();
normquat[3] = v.quat.w.ToFloat32();
view[0] = v.view.x.ToFloat32();
view[1] = v.view.y.ToFloat32();
view[2] = v.view.z.ToFloat32();
if (flip_quaternion) {
for (float& x : normquat) {
x = -x;
}
}
}
GLfloat position[4];
GLfloat color[4];
GLfloat tex_coord0[2];
GLfloat tex_coord1[2];
GLfloat tex_coord2[2];
GLfloat tex_coord0_w;
GLfloat normquat[4];
GLfloat view[3];
};
struct LightSrc {
alignas(16) GLvec3 specular_0;
alignas(16) GLvec3 specular_1;
alignas(16) GLvec3 diffuse;
alignas(16) GLvec3 ambient;
alignas(16) GLvec3 position;
alignas(16) GLvec3 spot_direction; // negated
GLfloat dist_atten_bias;
GLfloat dist_atten_scale;
};
/// Uniform structure for the Uniform Buffer Object, all vectors must be 16-byte aligned
// NOTE: Always keep a vec4 at the end. The GL spec is not clear wether the alignment at
// the end of a uniform block is included in UNIFORM_BLOCK_DATA_SIZE or not.
// Not following that rule will cause problems on some AMD drivers.
struct UniformData {
alignas(8) GLvec2 framebuffer_scale;
GLint alphatest_ref;
GLfloat depth_scale;
GLfloat depth_offset;
GLint scissor_x1;
GLint scissor_y1;
GLint scissor_x2;
GLint scissor_y2;
alignas(16) GLvec3 fog_color;
alignas(8) GLvec2 proctex_noise_f;
alignas(8) GLvec2 proctex_noise_a;
alignas(8) GLvec2 proctex_noise_p;
alignas(16) GLvec3 lighting_global_ambient;
LightSrc light_src[8];
alignas(16) GLvec4 const_color[6]; // A vec4 color for each of the six tev stages
alignas(16) GLvec4 tev_combiner_buffer_color;
alignas(16) GLvec4 clip_coef;
};
static_assert(
sizeof(UniformData) == 0x470,
"The size of the UniformData structure has changed, update the structure in the shader");
static_assert(sizeof(UniformData) < 16384,
"UniformData structure must be less than 16kb as per the OpenGL spec");
/// Syncs the clip enabled status to match the PICA register
void SyncClipEnabled();
/// Syncs the clip coefficients to match the PICA register
void SyncClipCoef();
/// Sets the OpenGL shader in accordance with the current PICA register state
void SetShader();
/// Syncs the cull mode to match the PICA register
void SyncCullMode();
/// Syncs the depth scale to match the PICA register
void SyncDepthScale();
/// Syncs the depth offset to match the PICA register
void SyncDepthOffset();
/// Syncs the blend enabled status to match the PICA register
void SyncBlendEnabled();
/// Syncs the blend functions to match the PICA register
void SyncBlendFuncs();
/// Syncs the blend color to match the PICA register
void SyncBlendColor();
/// Syncs the fog states to match the PICA register
void SyncFogColor();
void SyncFogLUT();
/// Sync the procedural texture noise configuration to match the PICA register
void SyncProcTexNoise();
/// Sync the procedural texture lookup tables
void SyncProcTexNoiseLUT();
void SyncProcTexColorMap();
void SyncProcTexAlphaMap();
void SyncProcTexLUT();
void SyncProcTexDiffLUT();
/// Syncs the alpha test states to match the PICA register
void SyncAlphaTest();
/// Syncs the logic op states to match the PICA register
void SyncLogicOp();
/// Syncs the color write mask to match the PICA register state
void SyncColorWriteMask();
/// Syncs the stencil write mask to match the PICA register state
void SyncStencilWriteMask();
/// Syncs the depth write mask to match the PICA register state
void SyncDepthWriteMask();
/// Syncs the stencil test states to match the PICA register
void SyncStencilTest();
/// Syncs the depth test states to match the PICA register
void SyncDepthTest();
/// Syncs the TEV combiner color buffer to match the PICA register
void SyncCombinerColor();
/// Syncs the TEV constant color to match the PICA register
void SyncTevConstColor(int tev_index, const Pica::TexturingRegs::TevStageConfig& tev_stage);
/// Syncs the lighting global ambient color to match the PICA register
void SyncGlobalAmbient();
/// Syncs the lighting lookup tables
void SyncLightingLUT(unsigned index);
/// Syncs the specified light's specular 0 color to match the PICA register
void SyncLightSpecular0(int light_index);
/// Syncs the specified light's specular 1 color to match the PICA register
void SyncLightSpecular1(int light_index);
/// Syncs the specified light's diffuse color to match the PICA register
void SyncLightDiffuse(int light_index);
/// Syncs the specified light's ambient color to match the PICA register
void SyncLightAmbient(int light_index);
/// Syncs the specified light's position to match the PICA register
void SyncLightPosition(int light_index);
/// Syncs the specified spot light direcition to match the PICA register
void SyncLightSpotDirection(int light_index);
/// Syncs the specified light's distance attenuation bias to match the PICA register
void SyncLightDistanceAttenuationBias(int light_index);
/// Syncs the specified light's distance attenuation scale to match the PICA register
void SyncLightDistanceAttenuationScale(int light_index);
OpenGLState state;
RasterizerCacheOpenGL res_cache;
std::vector<HardwareVertex> vertex_batch;
std::unordered_map<GLShader::PicaShaderConfig, std::unique_ptr<PicaShader>> shader_cache;
const PicaShader* current_shader = nullptr;
bool shader_dirty;
struct {
UniformData data;
std::array<bool, Pica::LightingRegs::NumLightingSampler> lut_dirty;
bool fog_lut_dirty;
bool proctex_noise_lut_dirty;
bool proctex_color_map_dirty;
bool proctex_alpha_map_dirty;
bool proctex_lut_dirty;
bool proctex_diff_lut_dirty;
bool dirty;
} uniform_block_data = {};
std::array<SamplerInfo, 3> texture_samplers;
OGLVertexArray vertex_array;
OGLBuffer vertex_buffer;
OGLBuffer uniform_buffer;
OGLFramebuffer framebuffer;
OGLBuffer lighting_lut_buffer;
OGLTexture lighting_lut;
std::array<std::array<GLvec2, 256>, Pica::LightingRegs::NumLightingSampler> lighting_lut_data{};
OGLBuffer fog_lut_buffer;
OGLTexture fog_lut;
std::array<GLvec2, 128> fog_lut_data{};
OGLBuffer proctex_noise_lut_buffer;
OGLTexture proctex_noise_lut;
std::array<GLvec2, 128> proctex_noise_lut_data{};
OGLBuffer proctex_color_map_buffer;
OGLTexture proctex_color_map;
std::array<GLvec2, 128> proctex_color_map_data{};
OGLBuffer proctex_alpha_map_buffer;
OGLTexture proctex_alpha_map;
std::array<GLvec2, 128> proctex_alpha_map_data{};
OGLBuffer proctex_lut_buffer;
OGLTexture proctex_lut;
std::array<GLvec4, 256> proctex_lut_data{};
OGLBuffer proctex_diff_lut_buffer;
OGLTexture proctex_diff_lut;
std::array<GLvec4, 256> proctex_diff_lut_data{};
};

799
src/video_core/renderer_opengl/gl_rasterizer_cache.cpp
View file

@ -1,799 +0,0 @@
// Copyright 2015 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <algorithm>
#include <atomic>
#include <cstring>
#include <iterator>
#include <unordered_set>
#include <utility>
#include <vector>
#include <glad/glad.h>
#include "common/bit_field.h"
#include "common/logging/log.h"
#include "common/math_util.h"
#include "common/microprofile.h"
#include "common/vector_math.h"
#include "core/frontend/emu_window.h"
#include "core/memory.h"
#include "core/settings.h"
#include "video_core/pica_state.h"
#include "video_core/renderer_opengl/gl_rasterizer_cache.h"
#include "video_core/renderer_opengl/gl_state.h"
#include "video_core/texture/texture_decode.h"
#include "video_core/utils.h"
#include "video_core/video_core.h"
struct FormatTuple {
GLint internal_format;
GLenum format;
GLenum type;
};
static const std::array<FormatTuple, 5> fb_format_tuples = {{
{GL_RGBA8, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8}, // RGBA8
{GL_RGB8, GL_BGR, GL_UNSIGNED_BYTE}, // RGB8
{GL_RGB5_A1, GL_RGBA, GL_UNSIGNED_SHORT_5_5_5_1}, // RGB5A1
{GL_RGB565, GL_RGB, GL_UNSIGNED_SHORT_5_6_5}, // RGB565
{GL_RGBA4, GL_RGBA, GL_UNSIGNED_SHORT_4_4_4_4}, // RGBA4
}};
static const std::array<FormatTuple, 4> depth_format_tuples = {{
{GL_DEPTH_COMPONENT16, GL_DEPTH_COMPONENT, GL_UNSIGNED_SHORT}, // D16
{},
{GL_DEPTH_COMPONENT24, GL_DEPTH_COMPONENT, GL_UNSIGNED_INT}, // D24
{GL_DEPTH24_STENCIL8, GL_DEPTH_STENCIL, GL_UNSIGNED_INT_24_8}, // D24S8
}};
RasterizerCacheOpenGL::RasterizerCacheOpenGL() {
transfer_framebuffers[0].Create();
transfer_framebuffers[1].Create();
}
RasterizerCacheOpenGL::~RasterizerCacheOpenGL() {
FlushAll();
}
static void MortonCopyPixels(CachedSurface::PixelFormat pixel_format, u32 width, u32 height,
u32 bytes_per_pixel, u32 gl_bytes_per_pixel, u8* morton_data,
u8* gl_data, bool morton_to_gl) {
using PixelFormat = CachedSurface::PixelFormat;
u8* data_ptrs[2];
u32 depth_stencil_shifts[2] = {24, 8};
if (morton_to_gl) {
std::swap(depth_stencil_shifts[0], depth_stencil_shifts[1]);
}
if (pixel_format == PixelFormat::D24S8) {
for (unsigned y = 0; y < height; ++y) {
for (unsigned x = 0; x < width; ++x) {
const u32 coarse_y = y & ~7;
u32 morton_offset = VideoCore::GetMortonOffset(x, y, bytes_per_pixel) +
coarse_y * width * bytes_per_pixel;
u32 gl_pixel_index = (x + (height - 1 - y) * width) * gl_bytes_per_pixel;
data_ptrs[morton_to_gl] = morton_data + morton_offset;
data_ptrs[!morton_to_gl] = &gl_data[gl_pixel_index];
// Swap depth and stencil value ordering since 3DS does not match OpenGL
u32 depth_stencil;
memcpy(&depth_stencil, data_ptrs[1], sizeof(u32));
depth_stencil = (depth_stencil << depth_stencil_shifts[0]) |
(depth_stencil >> depth_stencil_shifts[1]);
memcpy(data_ptrs[0], &depth_stencil, sizeof(u32));
}
}
} else {
for (unsigned y = 0; y < height; ++y) {
for (unsigned x = 0; x < width; ++x) {
const u32 coarse_y = y & ~7;
u32 morton_offset = VideoCore::GetMortonOffset(x, y, bytes_per_pixel) +
coarse_y * width * bytes_per_pixel;
u32 gl_pixel_index = (x + (height - 1 - y) * width) * gl_bytes_per_pixel;
data_ptrs[morton_to_gl] = morton_data + morton_offset;
data_ptrs[!morton_to_gl] = &gl_data[gl_pixel_index];
memcpy(data_ptrs[0], data_ptrs[1], bytes_per_pixel);
}
}
}
}
void RasterizerCacheOpenGL::BlitTextures(GLuint src_tex, GLuint dst_tex,
CachedSurface::SurfaceType type,
const MathUtil::Rectangle<int>& src_rect,
const MathUtil::Rectangle<int>& dst_rect) {
using SurfaceType = CachedSurface::SurfaceType;
OpenGLState cur_state = OpenGLState::GetCurState();
// Make sure textures aren't bound to texture units, since going to bind them to framebuffer
// components
OpenGLState::ResetTexture(src_tex);
OpenGLState::ResetTexture(dst_tex);
// Keep track of previous framebuffer bindings
GLuint old_fbs[2] = {cur_state.draw.read_framebuffer, cur_state.draw.draw_framebuffer};
cur_state.draw.read_framebuffer = transfer_framebuffers[0].handle;
cur_state.draw.draw_framebuffer = transfer_framebuffers[1].handle;
cur_state.Apply();
u32 buffers = 0;
if (type == SurfaceType::Color || type == SurfaceType::Texture) {
glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, src_tex,
0);
glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0,
0);
glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, dst_tex,
0);
glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0,
0);
buffers = GL_COLOR_BUFFER_BIT;
} else if (type == SurfaceType::Depth) {
glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0);
glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, GL_TEXTURE_2D, src_tex, 0);
glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0);
glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0);
glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, GL_TEXTURE_2D, dst_tex, 0);
glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0);
buffers = GL_DEPTH_BUFFER_BIT;
} else if (type == SurfaceType::DepthStencil) {
glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0);
glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D,
src_tex, 0);
glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0);
glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D,
dst_tex, 0);
buffers = GL_DEPTH_BUFFER_BIT | GL_STENCIL_BUFFER_BIT;
}
glBlitFramebuffer(src_rect.left, src_rect.top, src_rect.right, src_rect.bottom, dst_rect.left,
dst_rect.top, dst_rect.right, dst_rect.bottom, buffers,
buffers == GL_COLOR_BUFFER_BIT ? GL_LINEAR : GL_NEAREST);
// Restore previous framebuffer bindings
cur_state.draw.read_framebuffer = old_fbs[0];
cur_state.draw.draw_framebuffer = old_fbs[1];
cur_state.Apply();
}
bool RasterizerCacheOpenGL::TryBlitSurfaces(CachedSurface* src_surface,
const MathUtil::Rectangle<int>& src_rect,
CachedSurface* dst_surface,
const MathUtil::Rectangle<int>& dst_rect) {
if (!CachedSurface::CheckFormatsBlittable(src_surface->pixel_format,
dst_surface->pixel_format)) {
return false;
}
BlitTextures(src_surface->texture.handle, dst_surface->texture.handle,
CachedSurface::GetFormatType(src_surface->pixel_format), src_rect, dst_rect);
return true;
}
static void AllocateSurfaceTexture(GLuint texture, CachedSurface::PixelFormat pixel_format,
u32 width, u32 height) {
// Allocate an uninitialized texture of appropriate size and format for the surface
using SurfaceType = CachedSurface::SurfaceType;
OpenGLState cur_state = OpenGLState::GetCurState();
// Keep track of previous texture bindings
GLuint old_tex = cur_state.texture_units[0].texture_2d;
cur_state.texture_units[0].texture_2d = texture;
cur_state.Apply();
glActiveTexture(GL_TEXTURE0);
SurfaceType type = CachedSurface::GetFormatType(pixel_format);
FormatTuple tuple;
if (type == SurfaceType::Color) {
ASSERT((size_t)pixel_format < fb_format_tuples.size());
tuple = fb_format_tuples[(unsigned int)pixel_format];
} else if (type == SurfaceType::Depth || type == SurfaceType::DepthStencil) {
size_t tuple_idx = (size_t)pixel_format - 14;
ASSERT(tuple_idx < depth_format_tuples.size());
tuple = depth_format_tuples[tuple_idx];
} else {
tuple = {GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE};
}
glTexImage2D(GL_TEXTURE_2D, 0, tuple.internal_format, width, height, 0, tuple.format,
tuple.type, nullptr);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAX_LEVEL, 0);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
// Restore previous texture bindings
cur_state.texture_units[0].texture_2d = old_tex;
cur_state.Apply();
}
MICROPROFILE_DEFINE(OpenGL_SurfaceUpload, "OpenGL", "Surface Upload", MP_RGB(128, 64, 192));
CachedSurface* RasterizerCacheOpenGL::GetSurface(const CachedSurface& params, bool match_res_scale,
bool load_if_create) {
using PixelFormat = CachedSurface::PixelFormat;
using SurfaceType = CachedSurface::SurfaceType;
if (params.addr == 0) {
return nullptr;
}
u32 params_size =
params.width * params.height * CachedSurface::GetFormatBpp(params.pixel_format) / 8;
// Check for an exact match in existing surfaces
CachedSurface* best_exact_surface = nullptr;
float exact_surface_goodness = -1.f;
auto surface_interval =
boost::icl::interval<PAddr>::right_open(params.addr, params.addr + params_size);
auto range = surface_cache.equal_range(surface_interval);
for (auto it = range.first; it != range.second; ++it) {
for (auto it2 = it->second.begin(); it2 != it->second.end(); ++it2) {
CachedSurface* surface = it2->get();
// Check if the request matches the surface exactly
if (params.addr == surface->addr && params.width == surface->width &&
params.height == surface->height && params.pixel_format == surface->pixel_format) {
// Make sure optional param-matching criteria are fulfilled
bool tiling_match = (params.is_tiled == surface->is_tiled);
bool res_scale_match = (params.res_scale_width == surface->res_scale_width &&
params.res_scale_height == surface->res_scale_height);
if (!match_res_scale || res_scale_match) {
// Prioritize same-tiling and highest resolution surfaces
float match_goodness =
(float)tiling_match + surface->res_scale_width * surface->res_scale_height;
if (match_goodness > exact_surface_goodness || surface->dirty) {
exact_surface_goodness = match_goodness;
best_exact_surface = surface;
}
}
}
}
}
// Return the best exact surface if found
if (best_exact_surface != nullptr) {
return best_exact_surface;
}
// No matching surfaces found, so create a new one
u8* texture_src_data = Memory::GetPhysicalPointer(params.addr);
if (texture_src_data == nullptr) {
return nullptr;
}
MICROPROFILE_SCOPE(OpenGL_SurfaceUpload);
// Stride only applies to linear images.
ASSERT(params.pixel_stride == 0 || !params.is_tiled);
std::shared_ptr<CachedSurface> new_surface = std::make_shared<CachedSurface>();
new_surface->addr = params.addr;
new_surface->size = params_size;
new_surface->texture.Create();
new_surface->width = params.width;
new_surface->height = params.height;
new_surface->pixel_stride = params.pixel_stride;
new_surface->res_scale_width = params.res_scale_width;
new_surface->res_scale_height = params.res_scale_height;
new_surface->is_tiled = params.is_tiled;
new_surface->pixel_format = params.pixel_format;
new_surface->dirty = false;
if (!load_if_create) {
// Don't load any data; just allocate the surface's texture
AllocateSurfaceTexture(new_surface->texture.handle, new_surface->pixel_format,
new_surface->GetScaledWidth(), new_surface->GetScaledHeight());
} else {
// TODO: Consider attempting subrect match in existing surfaces and direct blit here instead
// of memory upload below if that's a common scenario in some game
Memory::RasterizerFlushRegion(params.addr, params_size);
// Load data from memory to the new surface
OpenGLState cur_state = OpenGLState::GetCurState();
GLuint old_tex = cur_state.texture_units[0].texture_2d;
cur_state.texture_units[0].texture_2d = new_surface->texture.handle;
cur_state.Apply();
glActiveTexture(GL_TEXTURE0);
if (!new_surface->is_tiled) {
// TODO: Ensure this will always be a color format, not a depth or other format
ASSERT((size_t)new_surface->pixel_format < fb_format_tuples.size());
const FormatTuple& tuple = fb_format_tuples[(unsigned int)params.pixel_format];
glPixelStorei(GL_UNPACK_ROW_LENGTH, (GLint)new_surface->pixel_stride);
glTexImage2D(GL_TEXTURE_2D, 0, tuple.internal_format, params.width, params.height, 0,
tuple.format, tuple.type, texture_src_data);
glPixelStorei(GL_UNPACK_ROW_LENGTH, 0);
} else {
SurfaceType type = CachedSurface::GetFormatType(new_surface->pixel_format);
if (type != SurfaceType::Depth && type != SurfaceType::DepthStencil) {
FormatTuple tuple;
if ((size_t)params.pixel_format < fb_format_tuples.size()) {
tuple = fb_format_tuples[(unsigned int)params.pixel_format];
} else {
// Texture
tuple = {GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE};
}
std::vector<Math::Vec4<u8>> tex_buffer(params.width * params.height);
Pica::Texture::TextureInfo tex_info;
tex_info.width = params.width;
tex_info.height = params.height;
tex_info.format = (Pica::TexturingRegs::TextureFormat)params.pixel_format;
tex_info.SetDefaultStride();
tex_info.physical_address = params.addr;
for (unsigned y = 0; y < params.height; ++y) {
for (unsigned x = 0; x < params.width; ++x) {
tex_buffer[x + params.width * y] = Pica::Texture::LookupTexture(
texture_src_data, x, params.height - 1 - y, tex_info);
}
}
glTexImage2D(GL_TEXTURE_2D, 0, tuple.internal_format, params.width, params.height,
0, GL_RGBA, GL_UNSIGNED_BYTE, tex_buffer.data());
} else {
// Depth/Stencil formats need special treatment since they aren't sampleable using
// LookupTexture and can't use RGBA format
size_t tuple_idx = (size_t)params.pixel_format - 14;
ASSERT(tuple_idx < depth_format_tuples.size());
const FormatTuple& tuple = depth_format_tuples[tuple_idx];
u32 bytes_per_pixel = CachedSurface::GetFormatBpp(params.pixel_format) / 8;
// OpenGL needs 4 bpp alignment for D24 since using GL_UNSIGNED_INT as type
bool use_4bpp = (params.pixel_format == PixelFormat::D24);
u32 gl_bytes_per_pixel = use_4bpp ? 4 : bytes_per_pixel;
std::vector<u8> temp_fb_depth_buffer(params.width * params.height *
gl_bytes_per_pixel);
u8* temp_fb_depth_buffer_ptr =
use_4bpp ? temp_fb_depth_buffer.data() + 1 : temp_fb_depth_buffer.data();
MortonCopyPixels(params.pixel_format, params.width, params.height, bytes_per_pixel,
gl_bytes_per_pixel, texture_src_data, temp_fb_depth_buffer_ptr,
true);
glTexImage2D(GL_TEXTURE_2D, 0, tuple.internal_format, params.width, params.height,
0, tuple.format, tuple.type, temp_fb_depth_buffer.data());
}
}
// If not 1x scale, blit 1x texture to a new scaled texture and replace texture in surface
if (new_surface->res_scale_width != 1.f || new_surface->res_scale_height != 1.f) {
OGLTexture scaled_texture;
scaled_texture.Create();
AllocateSurfaceTexture(scaled_texture.handle, new_surface->pixel_format,
new_surface->GetScaledWidth(), new_surface->GetScaledHeight());
BlitTextures(new_surface->texture.handle, scaled_texture.handle,
CachedSurface::GetFormatType(new_surface->pixel_format),
MathUtil::Rectangle<int>(0, 0, new_surface->width, new_surface->height),
MathUtil::Rectangle<int>(0, 0, new_surface->GetScaledWidth(),
new_surface->GetScaledHeight()));
new_surface->texture.Release();
new_surface->texture.handle = scaled_texture.handle;
scaled_texture.handle = 0;
cur_state.texture_units[0].texture_2d = new_surface->texture.handle;
cur_state.Apply();
}
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAX_LEVEL, 0);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
cur_state.texture_units[0].texture_2d = old_tex;
cur_state.Apply();
}
Memory::RasterizerMarkRegionCached(new_surface->addr, new_surface->size, 1);
surface_cache.add(std::make_pair(boost::icl::interval<PAddr>::right_open(
new_surface->addr, new_surface->addr + new_surface->size),
std::set<std::shared_ptr<CachedSurface>>({new_surface})));
return new_surface.get();
}
CachedSurface* RasterizerCacheOpenGL::GetSurfaceRect(const CachedSurface& params,
bool match_res_scale, bool load_if_create,
MathUtil::Rectangle<int>& out_rect) {
if (params.addr == 0) {
return nullptr;
}
u32 total_pixels = params.width * params.height;
u32 params_size = total_pixels * CachedSurface::GetFormatBpp(params.pixel_format) / 8;
// Attempt to find encompassing surfaces
CachedSurface* best_subrect_surface = nullptr;
float subrect_surface_goodness = -1.f;
auto surface_interval =
boost::icl::interval<PAddr>::right_open(params.addr, params.addr + params_size);
auto cache_upper_bound = surface_cache.upper_bound(surface_interval);
for (auto it = surface_cache.lower_bound(surface_interval); it != cache_upper_bound; ++it) {
for (auto it2 = it->second.begin(); it2 != it->second.end(); ++it2) {
CachedSurface* surface = it2->get();
// Check if the request is contained in the surface
if (params.addr >= surface->addr &&
params.addr + params_size - 1 <= surface->addr + surface->size - 1 &&
params.pixel_format == surface->pixel_format) {
// Make sure optional param-matching criteria are fulfilled
bool tiling_match = (params.is_tiled == surface->is_tiled);
bool res_scale_match = (params.res_scale_width == surface->res_scale_width &&
params.res_scale_height == surface->res_scale_height);
if (!match_res_scale || res_scale_match) {
// Prioritize same-tiling and highest resolution surfaces
float match_goodness =
(float)tiling_match + surface->res_scale_width * surface->res_scale_height;
if (match_goodness > subrect_surface_goodness || surface->dirty) {
subrect_surface_goodness = match_goodness;
best_subrect_surface = surface;
}
}
}
}
}
// Return the best subrect surface if found
if (best_subrect_surface != nullptr) {
unsigned int bytes_per_pixel =
(CachedSurface::GetFormatBpp(best_subrect_surface->pixel_format) / 8);
int x0, y0;
if (!params.is_tiled) {
u32 begin_pixel_index = (params.addr - best_subrect_surface->addr) / bytes_per_pixel;
x0 = begin_pixel_index % best_subrect_surface->width;
y0 = begin_pixel_index / best_subrect_surface->width;
out_rect = MathUtil::Rectangle<int>(x0, y0, x0 + params.width, y0 + params.height);
} else {
u32 bytes_per_tile = 8 * 8 * bytes_per_pixel;
u32 tiles_per_row = best_subrect_surface->width / 8;
u32 begin_tile_index = (params.addr - best_subrect_surface->addr) / bytes_per_tile;
x0 = begin_tile_index % tiles_per_row * 8;
y0 = begin_tile_index / tiles_per_row * 8;
// Tiled surfaces are flipped vertically in the rasterizer vs. 3DS memory.
out_rect =
MathUtil::Rectangle<int>(x0, best_subrect_surface->height - y0, x0 + params.width,
best_subrect_surface->height - (y0 + params.height));
}
out_rect.left = (int)(out_rect.left * best_subrect_surface->res_scale_width);
out_rect.right = (int)(out_rect.right * best_subrect_surface->res_scale_width);
out_rect.top = (int)(out_rect.top * best_subrect_surface->res_scale_height);
out_rect.bottom = (int)(out_rect.bottom * best_subrect_surface->res_scale_height);
return best_subrect_surface;
}
// No subrect found - create and return a new surface
if (!params.is_tiled) {
out_rect = MathUtil::Rectangle<int>(0, 0, (int)(params.width * params.res_scale_width),
(int)(params.height * params.res_scale_height));
} else {
out_rect = MathUtil::Rectangle<int>(0, (int)(params.height * params.res_scale_height),
(int)(params.width * params.res_scale_width), 0);
}
return GetSurface(params, match_res_scale, load_if_create);
}
CachedSurface* RasterizerCacheOpenGL::GetTextureSurface(
const Pica::TexturingRegs::FullTextureConfig& config) {
Pica::Texture::TextureInfo info =
Pica::Texture::TextureInfo::FromPicaRegister(config.config, config.format);
CachedSurface params;
params.addr = info.physical_address;
params.width = info.width;
params.height = info.height;
params.is_tiled = true;
params.pixel_format = CachedSurface::PixelFormatFromTextureFormat(info.format);
return GetSurface(params, false, true);
}
std::tuple<CachedSurface*, CachedSurface*, MathUtil::Rectangle<int>>
RasterizerCacheOpenGL::GetFramebufferSurfaces(
const Pica::FramebufferRegs::FramebufferConfig& config) {
const auto& regs = Pica::g_state.regs;
// Make sur that framebuffers don't overlap if both color and depth are being used
u32 fb_area = config.GetWidth() * config.GetHeight();
bool framebuffers_overlap =
config.GetColorBufferPhysicalAddress() != 0 &&
config.GetDepthBufferPhysicalAddress() != 0 &&
MathUtil::IntervalsIntersect(
config.GetColorBufferPhysicalAddress(),
fb_area * GPU::Regs::BytesPerPixel(GPU::Regs::PixelFormat(config.color_format.Value())),
config.GetDepthBufferPhysicalAddress(),
fb_area * Pica::FramebufferRegs::BytesPerDepthPixel(config.depth_format));
bool using_color_fb = config.GetColorBufferPhysicalAddress() != 0;
bool depth_write_enable = regs.framebuffer.output_merger.depth_write_enable &&
regs.framebuffer.framebuffer.allow_depth_stencil_write;
bool using_depth_fb = config.GetDepthBufferPhysicalAddress() != 0 &&
(regs.framebuffer.output_merger.depth_test_enable || depth_write_enable ||
!framebuffers_overlap);
if (framebuffers_overlap && using_color_fb && using_depth_fb) {
LOG_CRITICAL(Render_OpenGL, "Color and depth framebuffer memory regions overlap; "
"overlapping framebuffers not supported!");
using_depth_fb = false;
}
// get color and depth surfaces
CachedSurface color_params;
CachedSurface depth_params;
color_params.width = depth_params.width = config.GetWidth();
color_params.height = depth_params.height = config.GetHeight();
color_params.is_tiled = depth_params.is_tiled = true;
// Set the internal resolution, assume the same scaling factor for top and bottom screens
float resolution_scale_factor = Settings::values.resolution_factor;
if (resolution_scale_factor == 0.0f) {
// Auto - scale resolution to the window size
resolution_scale_factor = VideoCore::g_emu_window->GetFramebufferLayout().GetScalingRatio();
}
// Scale the resolution by the specified factor
color_params.res_scale_width = resolution_scale_factor;
depth_params.res_scale_width = resolution_scale_factor;
color_params.res_scale_height = resolution_scale_factor;
depth_params.res_scale_height = resolution_scale_factor;
color_params.addr = config.GetColorBufferPhysicalAddress();
color_params.pixel_format = CachedSurface::PixelFormatFromColorFormat(config.color_format);
depth_params.addr = config.GetDepthBufferPhysicalAddress();
depth_params.pixel_format = CachedSurface::PixelFormatFromDepthFormat(config.depth_format);
MathUtil::Rectangle<int> color_rect;
CachedSurface* color_surface =
using_color_fb ? GetSurfaceRect(color_params, true, true, color_rect) : nullptr;
MathUtil::Rectangle<int> depth_rect;
CachedSurface* depth_surface =
using_depth_fb ? GetSurfaceRect(depth_params, true, true, depth_rect) : nullptr;
// Sanity check to make sure found surfaces aren't the same
if (using_depth_fb && using_color_fb && color_surface == depth_surface) {
LOG_CRITICAL(
Render_OpenGL,
"Color and depth framebuffer surfaces overlap; overlapping surfaces not supported!");
using_depth_fb = false;
depth_surface = nullptr;
}
MathUtil::Rectangle<int> rect;
if (color_surface != nullptr && depth_surface != nullptr &&
(depth_rect.left != color_rect.left || depth_rect.top != color_rect.top)) {
// Can't specify separate color and depth viewport offsets in OpenGL, so re-zero both if
// they don't match
if (color_rect.left != 0 || color_rect.top != 0) {
color_surface = GetSurface(color_params, true, true);
}
if (depth_rect.left != 0 || depth_rect.top != 0) {
depth_surface = GetSurface(depth_params, true, true);
}
if (!color_surface->is_tiled) {
rect = MathUtil::Rectangle<int>(
0, 0, (int)(color_params.width * color_params.res_scale_width),
(int)(color_params.height * color_params.res_scale_height));
} else {
rect = MathUtil::Rectangle<int>(
0, (int)(color_params.height * color_params.res_scale_height),
(int)(color_params.width * color_params.res_scale_width), 0);
}
} else if (color_surface != nullptr) {
rect = color_rect;
} else if (depth_surface != nullptr) {
rect = depth_rect;
} else {
rect = MathUtil::Rectangle<int>(0, 0, 0, 0);
}
return std::make_tuple(color_surface, depth_surface, rect);
}
CachedSurface* RasterizerCacheOpenGL::TryGetFillSurface(const GPU::Regs::MemoryFillConfig& config) {
auto surface_interval =
boost::icl::interval<PAddr>::right_open(config.GetStartAddress(), config.GetEndAddress());
auto range = surface_cache.equal_range(surface_interval);
for (auto it = range.first; it != range.second; ++it) {
for (auto it2 = it->second.begin(); it2 != it->second.end(); ++it2) {
int bits_per_value = 0;
if (config.fill_24bit) {
bits_per_value = 24;
} else if (config.fill_32bit) {
bits_per_value = 32;
} else {
bits_per_value = 16;
}
CachedSurface* surface = it2->get();
if (surface->addr == config.GetStartAddress() &&
CachedSurface::GetFormatBpp(surface->pixel_format) == bits_per_value &&
(surface->width * surface->height *
CachedSurface::GetFormatBpp(surface->pixel_format) / 8) ==
(config.GetEndAddress() - config.GetStartAddress())) {
return surface;
}
}
}
return nullptr;
}
MICROPROFILE_DEFINE(OpenGL_SurfaceDownload, "OpenGL", "Surface Download", MP_RGB(128, 192, 64));
void RasterizerCacheOpenGL::FlushSurface(CachedSurface* surface) {
using PixelFormat = CachedSurface::PixelFormat;
using SurfaceType = CachedSurface::SurfaceType;
if (!surface->dirty) {
return;
}
MICROPROFILE_SCOPE(OpenGL_SurfaceDownload);
u8* dst_buffer = Memory::GetPhysicalPointer(surface->addr);
if (dst_buffer == nullptr) {
return;
}
OpenGLState cur_state = OpenGLState::GetCurState();
GLuint old_tex = cur_state.texture_units[0].texture_2d;
OGLTexture unscaled_tex;
GLuint texture_to_flush = surface->texture.handle;
// If not 1x scale, blit scaled texture to a new 1x texture and use that to flush
if (surface->res_scale_width != 1.f || surface->res_scale_height != 1.f) {
unscaled_tex.Create();
AllocateSurfaceTexture(unscaled_tex.handle, surface->pixel_format, surface->width,
surface->height);
BlitTextures(
surface->texture.handle, unscaled_tex.handle,
CachedSurface::GetFormatType(surface->pixel_format),
MathUtil::Rectangle<int>(0, 0, surface->GetScaledWidth(), surface->GetScaledHeight()),
MathUtil::Rectangle<int>(0, 0, surface->width, surface->height));
texture_to_flush = unscaled_tex.handle;
}
cur_state.texture_units[0].texture_2d = texture_to_flush;
cur_state.Apply();
glActiveTexture(GL_TEXTURE0);
if (!surface->is_tiled) {
// TODO: Ensure this will always be a color format, not a depth or other format
ASSERT((size_t)surface->pixel_format < fb_format_tuples.size());
const FormatTuple& tuple = fb_format_tuples[(unsigned int)surface->pixel_format];
glPixelStorei(GL_PACK_ROW_LENGTH, (GLint)surface->pixel_stride);
glGetTexImage(GL_TEXTURE_2D, 0, tuple.format, tuple.type, dst_buffer);
glPixelStorei(GL_PACK_ROW_LENGTH, 0);
} else {
SurfaceType type = CachedSurface::GetFormatType(surface->pixel_format);
if (type != SurfaceType::Depth && type != SurfaceType::DepthStencil) {
ASSERT((size_t)surface->pixel_format < fb_format_tuples.size());
const FormatTuple& tuple = fb_format_tuples[(unsigned int)surface->pixel_format];
u32 bytes_per_pixel = CachedSurface::GetFormatBpp(surface->pixel_format) / 8;
std::vector<u8> temp_gl_buffer(surface->width * surface->height * bytes_per_pixel);
glGetTexImage(GL_TEXTURE_2D, 0, tuple.format, tuple.type, temp_gl_buffer.data());
// Directly copy pixels. Internal OpenGL color formats are consistent so no conversion
// is necessary.
MortonCopyPixels(surface->pixel_format, surface->width, surface->height,
bytes_per_pixel, bytes_per_pixel, dst_buffer, temp_gl_buffer.data(),
false);
} else {
// Depth/Stencil formats need special treatment since they aren't sampleable using
// LookupTexture and can't use RGBA format
size_t tuple_idx = (size_t)surface->pixel_format - 14;
ASSERT(tuple_idx < depth_format_tuples.size());
const FormatTuple& tuple = depth_format_tuples[tuple_idx];
u32 bytes_per_pixel = CachedSurface::GetFormatBpp(surface->pixel_format) / 8;
// OpenGL needs 4 bpp alignment for D24 since using GL_UNSIGNED_INT as type
bool use_4bpp = (surface->pixel_format == PixelFormat::D24);
u32 gl_bytes_per_pixel = use_4bpp ? 4 : bytes_per_pixel;
std::vector<u8> temp_gl_buffer(surface->width * surface->height * gl_bytes_per_pixel);
glGetTexImage(GL_TEXTURE_2D, 0, tuple.format, tuple.type, temp_gl_buffer.data());
u8* temp_gl_buffer_ptr = use_4bpp ? temp_gl_buffer.data() + 1 : temp_gl_buffer.data();
MortonCopyPixels(surface->pixel_format, surface->width, surface->height,
bytes_per_pixel, gl_bytes_per_pixel, dst_buffer, temp_gl_buffer_ptr,
false);
}
}
surface->dirty = false;
cur_state.texture_units[0].texture_2d = old_tex;
cur_state.Apply();
}
void RasterizerCacheOpenGL::FlushRegion(PAddr addr, u32 size, const CachedSurface* skip_surface,
bool invalidate) {
if (size == 0) {
return;
}
// Gather up unique surfaces that touch the region
std::unordered_set<std::shared_ptr<CachedSurface>> touching_surfaces;
auto surface_interval = boost::icl::interval<PAddr>::right_open(addr, addr + size);
auto cache_upper_bound = surface_cache.upper_bound(surface_interval);
for (auto it = surface_cache.lower_bound(surface_interval); it != cache_upper_bound; ++it) {
std::copy_if(it->second.begin(), it->second.end(),
std::inserter(touching_surfaces, touching_surfaces.end()),
[skip_surface](std::shared_ptr<CachedSurface> surface) {
return (surface.get() != skip_surface);
});
}
// Flush and invalidate surfaces
for (auto surface : touching_surfaces) {
FlushSurface(surface.get());
if (invalidate) {
Memory::RasterizerMarkRegionCached(surface->addr, surface->size, -1);
surface_cache.subtract(
std::make_pair(boost::icl::interval<PAddr>::right_open(
surface->addr, surface->addr + surface->size),
std::set<std::shared_ptr<CachedSurface>>({surface})));
}
}
}
void RasterizerCacheOpenGL::FlushAll() {
for (auto& surfaces : surface_cache) {
for (auto& surface : surfaces.second) {
FlushSurface(surface.get());
}
}
}

239
src/video_core/renderer_opengl/gl_rasterizer_cache.h
View file

@ -1,239 +0,0 @@
// Copyright 2015 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <array>
#include <memory>
#include <set>
#include <tuple>
#ifdef __GNUC__
#pragma GCC diagnostic push
#pragma GCC diagnostic ignored "-Wunused-local-typedef"
#endif
#include <boost/icl/interval_map.hpp>
#ifdef __GNUC__
#pragma GCC diagnostic pop
#endif
#include <glad/glad.h>
#include "common/assert.h"
#include "common/common_funcs.h"
#include "common/common_types.h"
#include "core/hw/gpu.h"
#include "video_core/regs_framebuffer.h"
#include "video_core/regs_texturing.h"
#include "video_core/renderer_opengl/gl_resource_manager.h"
namespace MathUtil {
template <class T>
struct Rectangle;
}
struct CachedSurface;
using SurfaceCache = boost::icl::interval_map<PAddr, std::set<std::shared_ptr<CachedSurface>>>;
struct CachedSurface {
enum class PixelFormat {
// First 5 formats are shared between textures and color buffers
RGBA8 = 0,
RGB8 = 1,
RGB5A1 = 2,
RGB565 = 3,
RGBA4 = 4,
// Texture-only formats
IA8 = 5,
RG8 = 6,
I8 = 7,
A8 = 8,
IA4 = 9,
I4 = 10,
A4 = 11,
ETC1 = 12,
ETC1A4 = 13,
// Depth buffer-only formats
D16 = 14,
// gap
D24 = 16,
D24S8 = 17,
Invalid = 255,
};
enum class SurfaceType {
Color = 0,
Texture = 1,
Depth = 2,
DepthStencil = 3,
Invalid = 4,
};
static unsigned int GetFormatBpp(CachedSurface::PixelFormat format) {
static const std::array<unsigned int, 18> bpp_table = {
32, // RGBA8
24, // RGB8
16, // RGB5A1
16, // RGB565
16, // RGBA4
16, // IA8
16, // RG8
8, // I8
8, // A8
8, // IA4
4, // I4
4, // A4
4, // ETC1
8, // ETC1A4
16, // D16
0,
24, // D24
32, // D24S8
};
ASSERT((unsigned int)format < ARRAY_SIZE(bpp_table));
return bpp_table[(unsigned int)format];
}
static PixelFormat PixelFormatFromTextureFormat(Pica::TexturingRegs::TextureFormat format) {
return ((unsigned int)format < 14) ? (PixelFormat)format : PixelFormat::Invalid;
}
static PixelFormat PixelFormatFromColorFormat(Pica::FramebufferRegs::ColorFormat format) {
return ((unsigned int)format < 5) ? (PixelFormat)format : PixelFormat::Invalid;
}
static PixelFormat PixelFormatFromDepthFormat(Pica::FramebufferRegs::DepthFormat format) {
return ((unsigned int)format < 4) ? (PixelFormat)((unsigned int)format + 14)
: PixelFormat::Invalid;
}
static PixelFormat PixelFormatFromGPUPixelFormat(GPU::Regs::PixelFormat format) {
switch (format) {
// RGB565 and RGB5A1 are switched in PixelFormat compared to ColorFormat
case GPU::Regs::PixelFormat::RGB565:
return PixelFormat::RGB565;
case GPU::Regs::PixelFormat::RGB5A1:
return PixelFormat::RGB5A1;
default:
return ((unsigned int)format < 5) ? (PixelFormat)format : PixelFormat::Invalid;
}
}
static bool CheckFormatsBlittable(PixelFormat pixel_format_a, PixelFormat pixel_format_b) {
SurfaceType a_type = GetFormatType(pixel_format_a);
SurfaceType b_type = GetFormatType(pixel_format_b);
if ((a_type == SurfaceType::Color || a_type == SurfaceType::Texture) &&
(b_type == SurfaceType::Color || b_type == SurfaceType::Texture)) {
return true;
}
if (a_type == SurfaceType::Depth && b_type == SurfaceType::Depth) {
return true;
}
if (a_type == SurfaceType::DepthStencil && b_type == SurfaceType::DepthStencil) {
return true;
}
return false;
}
static SurfaceType GetFormatType(PixelFormat pixel_format) {
if ((unsigned int)pixel_format < 5) {
return SurfaceType::Color;
}
if ((unsigned int)pixel_format < 14) {
return SurfaceType::Texture;
}
if (pixel_format == PixelFormat::D16 || pixel_format == PixelFormat::D24) {
return SurfaceType::Depth;
}
if (pixel_format == PixelFormat::D24S8) {
return SurfaceType::DepthStencil;
}
return SurfaceType::Invalid;
}
u32 GetScaledWidth() const {
return (u32)(width * res_scale_width);
}
u32 GetScaledHeight() const {
return (u32)(height * res_scale_height);
}
PAddr addr;
u32 size;
PAddr min_valid;
PAddr max_valid;
OGLTexture texture;
u32 width;
u32 height;
/// Stride between lines, in pixels. Only valid for images in linear format.
u32 pixel_stride = 0;
float res_scale_width = 1.f;
float res_scale_height = 1.f;
bool is_tiled;
PixelFormat pixel_format;
bool dirty;
};
class RasterizerCacheOpenGL : NonCopyable {
public:
RasterizerCacheOpenGL();
~RasterizerCacheOpenGL();
/// Blits one texture to another
void BlitTextures(GLuint src_tex, GLuint dst_tex, CachedSurface::SurfaceType type,
const MathUtil::Rectangle<int>& src_rect,
const MathUtil::Rectangle<int>& dst_rect);
/// Attempt to blit one surface's texture to another
bool TryBlitSurfaces(CachedSurface* src_surface, const MathUtil::Rectangle<int>& src_rect,
CachedSurface* dst_surface, const MathUtil::Rectangle<int>& dst_rect);
/// Loads a texture from 3DS memory to OpenGL and caches it (if not already cached)
CachedSurface* GetSurface(const CachedSurface& params, bool match_res_scale,
bool load_if_create);
/// Attempt to find a subrect (resolution scaled) of a surface, otherwise loads a texture from
/// 3DS memory to OpenGL and caches it (if not already cached)
CachedSurface* GetSurfaceRect(const CachedSurface& params, bool match_res_scale,
bool load_if_create, MathUtil::Rectangle<int>& out_rect);
/// Gets a surface based on the texture configuration
CachedSurface* GetTextureSurface(const Pica::TexturingRegs::FullTextureConfig& config);
/// Gets the color and depth surfaces and rect (resolution scaled) based on the framebuffer
/// configuration
std::tuple<CachedSurface*, CachedSurface*, MathUtil::Rectangle<int>> GetFramebufferSurfaces(
const Pica::FramebufferRegs::FramebufferConfig& config);
/// Attempt to get a surface that exactly matches the fill region and format
CachedSurface* TryGetFillSurface(const GPU::Regs::MemoryFillConfig& config);
/// Write the surface back to memory
void FlushSurface(CachedSurface* surface);
/// Write any cached resources overlapping the region back to memory (if dirty) and optionally
/// invalidate them in the cache
void FlushRegion(PAddr addr, u32 size, const CachedSurface* skip_surface, bool invalidate);
/// Flush all cached resources tracked by this cache manager
void FlushAll();
private:
SurfaceCache surface_cache;
OGLFramebuffer transfer_framebuffers[2];
};

1231
src/video_core/renderer_opengl/gl_shader_gen.cpp
View file

File diff suppressed because it is too large Load diff

162
src/video_core/renderer_opengl/gl_shader_gen.h
View file

@ -1,162 +0,0 @@
// Copyright 2015 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <array>
#include <cstring>
#include <functional>
#include <string>
#include <type_traits>
#include "video_core/regs.h"
namespace GLShader {
enum Attributes {
ATTRIBUTE_POSITION,
ATTRIBUTE_COLOR,
ATTRIBUTE_TEXCOORD0,
ATTRIBUTE_TEXCOORD1,
ATTRIBUTE_TEXCOORD2,
ATTRIBUTE_TEXCOORD0_W,
ATTRIBUTE_NORMQUAT,
ATTRIBUTE_VIEW,
};
/**
* This struct contains all state used to generate the GLSL shader program that emulates the current
* Pica register configuration. This struct is used as a cache key for generated GLSL shader
* programs. The functions in gl_shader_gen.cpp should retrieve state from this struct only, not by
* directly accessing Pica registers. This should reduce the risk of bugs in shader generation where
* Pica state is not being captured in the shader cache key, thereby resulting in (what should be)
* two separate shaders sharing the same key.
*
* We use a union because "implicitly-defined copy/move constructor for a union X copies the object
* representation of X." and "implicitly-defined copy assignment operator for a union X copies the
* object representation (3.9) of X." = Bytewise copy instead of memberwise copy. This is important
* because the padding bytes are included in the hash and comparison between objects.
*/
union PicaShaderConfig {
/// Construct a PicaShaderConfig with the given Pica register configuration.
static PicaShaderConfig BuildFromRegs(const Pica::Regs& regs);
bool TevStageUpdatesCombinerBufferColor(unsigned stage_index) const {
return (stage_index < 4) && (state.combiner_buffer_input & (1 << stage_index));
}
bool TevStageUpdatesCombinerBufferAlpha(unsigned stage_index) const {
return (stage_index < 4) && ((state.combiner_buffer_input >> 4) & (1 << stage_index));
}
bool operator==(const PicaShaderConfig& o) const {
return std::memcmp(&state, &o.state, sizeof(PicaShaderConfig::State)) == 0;
};
// NOTE: MSVC15 (Update 2) doesn't think `delete`'d constructors and operators are TC.
// This makes BitField not TC when used in a union or struct so we have to resort
// to this ugly hack.
// Once that bug is fixed we can use Pica::Regs::TevStageConfig here.
// Doesn't include const_color because we don't sync it, see comment in BuildFromRegs()
struct TevStageConfigRaw {
u32 sources_raw;
u32 modifiers_raw;
u32 ops_raw;
u32 scales_raw;
explicit operator Pica::TexturingRegs::TevStageConfig() const noexcept {
Pica::TexturingRegs::TevStageConfig stage;
stage.sources_raw = sources_raw;
stage.modifiers_raw = modifiers_raw;
stage.ops_raw = ops_raw;
stage.const_color = 0;
stage.scales_raw = scales_raw;
return stage;
}
};
struct State {
Pica::FramebufferRegs::CompareFunc alpha_test_func;
Pica::RasterizerRegs::ScissorMode scissor_test_mode;
Pica::TexturingRegs::TextureConfig::TextureType texture0_type;
bool texture2_use_coord1;
std::array<TevStageConfigRaw, 6> tev_stages;
u8 combiner_buffer_input;
Pica::RasterizerRegs::DepthBuffering depthmap_enable;
Pica::TexturingRegs::FogMode fog_mode;
bool fog_flip;
struct {
struct {
unsigned num;
bool directional;
bool two_sided_diffuse;
bool dist_atten_enable;
bool spot_atten_enable;
bool geometric_factor_0;
bool geometric_factor_1;
} light[8];
bool enable;
unsigned src_num;
Pica::LightingRegs::LightingBumpMode bump_mode;
unsigned bump_selector;
bool bump_renorm;
bool clamp_highlights;
Pica::LightingRegs::LightingConfig config;
Pica::LightingRegs::LightingFresnelSelector fresnel_selector;
struct {
bool enable;
bool abs_input;
Pica::LightingRegs::LightingLutInput type;
float scale;
} lut_d0, lut_d1, lut_sp, lut_fr, lut_rr, lut_rg, lut_rb;
} lighting;
struct {
bool enable;
u32 coord;
Pica::TexturingRegs::ProcTexClamp u_clamp, v_clamp;
Pica::TexturingRegs::ProcTexCombiner color_combiner, alpha_combiner;
bool separate_alpha;
bool noise_enable;
Pica::TexturingRegs::ProcTexShift u_shift, v_shift;
u32 lut_width;
u32 lut_offset;
Pica::TexturingRegs::ProcTexFilter lut_filter;
} proctex;
} state;
};
#if (__GNUC__ >= 5) || defined(__clang__) || defined(_MSC_VER)
static_assert(std::is_trivially_copyable<PicaShaderConfig::State>::value,
"PicaShaderConfig::State must be trivially copyable");
#endif
/**
* Generates the GLSL vertex shader program source code for the current Pica state
* @returns String of the shader source code
*/
std::string GenerateVertexShader();
/**
* Generates the GLSL fragment shader program source code for the current Pica state
* @param config ShaderCacheKey object generated for the current Pica state, used for the shader
* configuration (NOTE: Use state in this struct only, not the Pica registers!)
* @returns String of the shader source code
*/
std::string GenerateFragmentShader(const PicaShaderConfig& config);
} // namespace GLShader
namespace std {
template <>
struct hash<GLShader::PicaShaderConfig> {
size_t operator()(const GLShader::PicaShaderConfig& k) const {
return Common::ComputeHash64(&k.state, sizeof(GLShader::PicaShaderConfig::State));
}
};
} // namespace std

235
src/video_core/renderer_opengl/pica_to_gl.h
View file

@ -1,235 +0,0 @@
// Copyright 2015 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <array>
#include <cstddef>
#include <glad/glad.h>
#include "common/assert.h"
#include "common/bit_field.h"
#include "common/common_funcs.h"
#include "common/common_types.h"
#include "common/logging/log.h"
#include "core/core.h"
#include "video_core/regs_framebuffer.h"
#include "video_core/regs_lighting.h"
#include "video_core/regs_texturing.h"
using GLvec2 = std::array<GLfloat, 2>;
using GLvec3 = std::array<GLfloat, 3>;
using GLvec4 = std::array<GLfloat, 4>;
namespace PicaToGL {
inline GLenum TextureFilterMode(Pica::TexturingRegs::TextureConfig::TextureFilter mode) {
static const GLenum filter_mode_table[] = {
GL_NEAREST, // TextureFilter::Nearest
GL_LINEAR, // TextureFilter::Linear
};
// Range check table for input
if (static_cast<size_t>(mode) >= ARRAY_SIZE(filter_mode_table)) {
LOG_CRITICAL(Render_OpenGL, "Unknown texture filtering mode %d", mode);
UNREACHABLE();
return GL_LINEAR;
}
GLenum gl_mode = filter_mode_table[mode];
// Check for dummy values indicating an unknown mode
if (gl_mode == 0) {
LOG_CRITICAL(Render_OpenGL, "Unknown texture filtering mode %d", mode);
UNIMPLEMENTED();
return GL_LINEAR;
}
return gl_mode;
}
inline GLenum WrapMode(Pica::TexturingRegs::TextureConfig::WrapMode mode) {
static const GLenum wrap_mode_table[] = {
GL_CLAMP_TO_EDGE, // WrapMode::ClampToEdge
GL_CLAMP_TO_BORDER, // WrapMode::ClampToBorder
GL_REPEAT, // WrapMode::Repeat
GL_MIRRORED_REPEAT, // WrapMode::MirroredRepeat
// TODO(wwylele): ClampToEdge2 and ClampToBorder2 are not properly implemented here. See the
// comments in enum WrapMode.
GL_CLAMP_TO_EDGE, // WrapMode::ClampToEdge2
GL_CLAMP_TO_BORDER, // WrapMode::ClampToBorder2
GL_REPEAT, // WrapMode::Repeat2
GL_REPEAT, // WrapMode::Repeat3
};
// Range check table for input
if (static_cast<size_t>(mode) >= ARRAY_SIZE(wrap_mode_table)) {
LOG_CRITICAL(Render_OpenGL, "Unknown texture wrap mode %d", mode);
UNREACHABLE();
return GL_CLAMP_TO_EDGE;
}
if (static_cast<u32>(mode) > 3) {
Core::Telemetry().AddField(Telemetry::FieldType::Session,
"VideoCore_Pica_UnsupportedTextureWrapMode",
static_cast<u32>(mode));
LOG_WARNING(Render_OpenGL, "Using texture wrap mode %u", static_cast<u32>(mode));
}
GLenum gl_mode = wrap_mode_table[mode];
// Check for dummy values indicating an unknown mode
if (gl_mode == 0) {
LOG_CRITICAL(Render_OpenGL, "Unknown texture wrap mode %d", mode);
UNIMPLEMENTED();
return GL_CLAMP_TO_EDGE;
}
return gl_mode;
}
inline GLenum BlendEquation(Pica::FramebufferRegs::BlendEquation equation) {
static const GLenum blend_equation_table[] = {
GL_FUNC_ADD, // BlendEquation::Add
GL_FUNC_SUBTRACT, // BlendEquation::Subtract
GL_FUNC_REVERSE_SUBTRACT, // BlendEquation::ReverseSubtract
GL_MIN, // BlendEquation::Min
GL_MAX, // BlendEquation::Max
};
// Range check table for input
if (static_cast<size_t>(equation) >= ARRAY_SIZE(blend_equation_table)) {
LOG_CRITICAL(Render_OpenGL, "Unknown blend equation %d", equation);
UNREACHABLE();
return GL_FUNC_ADD;
}
return blend_equation_table[(unsigned)equation];
}
inline GLenum BlendFunc(Pica::FramebufferRegs::BlendFactor factor) {
static const GLenum blend_func_table[] = {
GL_ZERO, // BlendFactor::Zero
GL_ONE, // BlendFactor::One
GL_SRC_COLOR, // BlendFactor::SourceColor
GL_ONE_MINUS_SRC_COLOR, // BlendFactor::OneMinusSourceColor
GL_DST_COLOR, // BlendFactor::DestColor
GL_ONE_MINUS_DST_COLOR, // BlendFactor::OneMinusDestColor
GL_SRC_ALPHA, // BlendFactor::SourceAlpha
GL_ONE_MINUS_SRC_ALPHA, // BlendFactor::OneMinusSourceAlpha
GL_DST_ALPHA, // BlendFactor::DestAlpha
GL_ONE_MINUS_DST_ALPHA, // BlendFactor::OneMinusDestAlpha
GL_CONSTANT_COLOR, // BlendFactor::ConstantColor
GL_ONE_MINUS_CONSTANT_COLOR, // BlendFactor::OneMinusConstantColor
GL_CONSTANT_ALPHA, // BlendFactor::ConstantAlpha
GL_ONE_MINUS_CONSTANT_ALPHA, // BlendFactor::OneMinusConstantAlpha
GL_SRC_ALPHA_SATURATE, // BlendFactor::SourceAlphaSaturate
};
// Range check table for input
if (static_cast<size_t>(factor) >= ARRAY_SIZE(blend_func_table)) {
LOG_CRITICAL(Render_OpenGL, "Unknown blend factor %d", factor);
UNREACHABLE();
return GL_ONE;
}
return blend_func_table[(unsigned)factor];
}
inline GLenum LogicOp(Pica::FramebufferRegs::LogicOp op) {
static const GLenum logic_op_table[] = {
GL_CLEAR, // Clear
GL_AND, // And
GL_AND_REVERSE, // AndReverse
GL_COPY, // Copy
GL_SET, // Set
GL_COPY_INVERTED, // CopyInverted
GL_NOOP, // NoOp
GL_INVERT, // Invert
GL_NAND, // Nand
GL_OR, // Or
GL_NOR, // Nor
GL_XOR, // Xor
GL_EQUIV, // Equiv
GL_AND_INVERTED, // AndInverted
GL_OR_REVERSE, // OrReverse
GL_OR_INVERTED, // OrInverted
};
// Range check table for input
if (static_cast<size_t>(op) >= ARRAY_SIZE(logic_op_table)) {
LOG_CRITICAL(Render_OpenGL, "Unknown logic op %d", op);
UNREACHABLE();
return GL_COPY;
}
return logic_op_table[(unsigned)op];
}
inline GLenum CompareFunc(Pica::FramebufferRegs::CompareFunc func) {
static const GLenum compare_func_table[] = {
GL_NEVER, // CompareFunc::Never
GL_ALWAYS, // CompareFunc::Always
GL_EQUAL, // CompareFunc::Equal
GL_NOTEQUAL, // CompareFunc::NotEqual
GL_LESS, // CompareFunc::LessThan
GL_LEQUAL, // CompareFunc::LessThanOrEqual
GL_GREATER, // CompareFunc::GreaterThan
GL_GEQUAL, // CompareFunc::GreaterThanOrEqual
};
// Range check table for input
if (static_cast<size_t>(func) >= ARRAY_SIZE(compare_func_table)) {
LOG_CRITICAL(Render_OpenGL, "Unknown compare function %d", func);
UNREACHABLE();
return GL_ALWAYS;
}
return compare_func_table[(unsigned)func];
}
inline GLenum StencilOp(Pica::FramebufferRegs::StencilAction action) {
static const GLenum stencil_op_table[] = {
GL_KEEP, // StencilAction::Keep
GL_ZERO, // StencilAction::Zero
GL_REPLACE, // StencilAction::Replace
GL_INCR, // StencilAction::Increment
GL_DECR, // StencilAction::Decrement
GL_INVERT, // StencilAction::Invert
GL_INCR_WRAP, // StencilAction::IncrementWrap
GL_DECR_WRAP, // StencilAction::DecrementWrap
};
// Range check table for input
if (static_cast<size_t>(action) >= ARRAY_SIZE(stencil_op_table)) {
LOG_CRITICAL(Render_OpenGL, "Unknown stencil op %d", action);
UNREACHABLE();
return GL_KEEP;
}
return stencil_op_table[(unsigned)action];
}
inline GLvec4 ColorRGBA8(const u32 color) {
return {{
(color >> 0 & 0xFF) / 255.0f, (color >> 8 & 0xFF) / 255.0f, (color >> 16 & 0xFF) / 255.0f,
(color >> 24 & 0xFF) / 255.0f,
}};
}
inline std::array<GLfloat, 3> LightColor(const Pica::LightingRegs::LightColor& color) {
return {{
color.r / 255.0f, color.g / 255.0f, color.b / 255.0f,
}};
}
} // namespace

7
src/video_core/renderer_opengl/renderer_opengl.cpp
View file

@ -13,14 +13,11 @@
#include "core/core.h"
#include "core/core_timing.h"
#include "core/frontend/emu_window.h"
#include "core/hw/gpu.h"
#include "core/hw/hw.h"
#include "core/hw/lcd.h"
#include "core/memory.h"
#include "core/settings.h"
#include "core/tracer/recorder.h"
#include "video_core/debug_utils/debug_utils.h"
#include "video_core/rasterizer_interface.h"
#include "video_core/renderer_opengl/renderer_opengl.h"
#include "video_core/video_core.h"
@ -128,10 +125,6 @@ void RendererOpenGL::SwapBuffers(const FramebufferInfo& framebuffer_info) {
prev_state.Apply();
RefreshRasterizerSetting();
if (Pica::g_debug_context && Pica::g_debug_context->recorder) {
Pica::g_debug_context->recorder->FrameFinished();
}
}
static inline u32 MortonInterleave128(u32 x, u32 y) {

1
src/video_core/renderer_opengl/renderer_opengl.h
View file

@ -8,7 +8,6 @@
#include <glad/glad.h>
#include "common/common_types.h"
#include "common/math_util.h"
#include "core/hw/gpu.h"
#include "video_core/renderer_base.h"
#include "video_core/renderer_opengl/gl_resource_manager.h"
#include "video_core/renderer_opengl/gl_state.h"

186
src/video_core/shader/debug_data.h
View file

@ -1,186 +0,0 @@
// Copyright 2016 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <vector>
#include "common/common_types.h"
#include "common/vector_math.h"
#include "video_core/pica_types.h"
namespace Pica {
namespace Shader {
/// Helper structure used to keep track of data useful for inspection of shader emulation
template <bool full_debugging>
struct DebugData;
template <>
struct DebugData<false> {
// TODO: Hide these behind and interface and move them to DebugData<true>
u32 max_offset = 0; ///< maximum program counter ever reached
u32 max_opdesc_id = 0; ///< maximum swizzle pattern index ever used
};
template <>
struct DebugData<true> {
/// Records store the input and output operands of a particular instruction.
struct Record {
enum Type {
// Floating point arithmetic operands
SRC1 = 0x1,
SRC2 = 0x2,
SRC3 = 0x4,
// Initial and final output operand value
DEST_IN = 0x8,
DEST_OUT = 0x10,
// Current and next instruction offset (in words)
CUR_INSTR = 0x20,
NEXT_INSTR = 0x40,
// Output address register value
ADDR_REG_OUT = 0x80,
// Result of a comparison instruction
CMP_RESULT = 0x100,
// Input values for conditional flow control instructions
COND_BOOL_IN = 0x200,
COND_CMP_IN = 0x400,
// Input values for a loop
LOOP_INT_IN = 0x800,
};
Math::Vec4<float24> src1;
Math::Vec4<float24> src2;
Math::Vec4<float24> src3;
Math::Vec4<float24> dest_in;
Math::Vec4<float24> dest_out;
s32 address_registers[2];
bool conditional_code[2];
bool cond_bool;
bool cond_cmp[2];
Math::Vec4<u8> loop_int;
u32 instruction_offset;
u32 next_instruction;
/// set of enabled fields (as a combination of Type flags)
unsigned mask = 0;
};
u32 max_offset = 0; ///< maximum program counter ever reached
u32 max_opdesc_id = 0; ///< maximum swizzle pattern index ever used
/// List of records for each executed shader instruction
std::vector<DebugData<true>::Record> records;
};
/// Type alias for better readability
using DebugDataRecord = DebugData<true>::Record;
/// Helper function to set a DebugData<true>::Record field based on the template enum parameter.
template <DebugDataRecord::Type type, typename ValueType>
inline void SetField(DebugDataRecord& record, ValueType value);
template <>
inline void SetField<DebugDataRecord::SRC1>(DebugDataRecord& record, float24* value) {
record.src1.x = value[0];
record.src1.y = value[1];
record.src1.z = value[2];
record.src1.w = value[3];
}
template <>
inline void SetField<DebugDataRecord::SRC2>(DebugDataRecord& record, float24* value) {
record.src2.x = value[0];
record.src2.y = value[1];
record.src2.z = value[2];
record.src2.w = value[3];
}
template <>
inline void SetField<DebugDataRecord::SRC3>(DebugDataRecord& record, float24* value) {
record.src3.x = value[0];
record.src3.y = value[1];
record.src3.z = value[2];
record.src3.w = value[3];
}
template <>
inline void SetField<DebugDataRecord::DEST_IN>(DebugDataRecord& record, float24* value) {
record.dest_in.x = value[0];
record.dest_in.y = value[1];
record.dest_in.z = value[2];
record.dest_in.w = value[3];
}
template <>
inline void SetField<DebugDataRecord::DEST_OUT>(DebugDataRecord& record, float24* value) {
record.dest_out.x = value[0];
record.dest_out.y = value[1];
record.dest_out.z = value[2];
record.dest_out.w = value[3];
}
template <>
inline void SetField<DebugDataRecord::ADDR_REG_OUT>(DebugDataRecord& record, s32* value) {
record.address_registers[0] = value[0];
record.address_registers[1] = value[1];
}
template <>
inline void SetField<DebugDataRecord::CMP_RESULT>(DebugDataRecord& record, bool* value) {
record.conditional_code[0] = value[0];
record.conditional_code[1] = value[1];
}
template <>
inline void SetField<DebugDataRecord::COND_BOOL_IN>(DebugDataRecord& record, bool value) {
record.cond_bool = value;
}
template <>
inline void SetField<DebugDataRecord::COND_CMP_IN>(DebugDataRecord& record, bool* value) {
record.cond_cmp[0] = value[0];
record.cond_cmp[1] = value[1];
}
template <>
inline void SetField<DebugDataRecord::LOOP_INT_IN>(DebugDataRecord& record, Math::Vec4<u8> value) {
record.loop_int = value;
}
template <>
inline void SetField<DebugDataRecord::CUR_INSTR>(DebugDataRecord& record, u32 value) {
record.instruction_offset = value;
}
template <>
inline void SetField<DebugDataRecord::NEXT_INSTR>(DebugDataRecord& record, u32 value) {
record.next_instruction = value;
}
/// Helper function to set debug information on the current shader iteration.
template <DebugDataRecord::Type type, typename ValueType>
inline void Record(DebugData<false>& debug_data, u32 offset, ValueType value) {
// Debugging disabled => nothing to do
}
template <DebugDataRecord::Type type, typename ValueType>
inline void Record(DebugData<true>& debug_data, u32 offset, ValueType value) {
if (offset >= debug_data.records.size())
debug_data.records.resize(offset + 1);
SetField<type, ValueType>(debug_data.records[offset], value);
debug_data.records[offset].mask |= type;
}
} // namespace Shader
} // namespace Pica

154
src/video_core/shader/shader.cpp
View file

@ -1,154 +0,0 @@
// Copyright 2015 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <cmath>
#include <cstring>
#include "common/bit_set.h"
#include "common/logging/log.h"
#include "common/microprofile.h"
#include "video_core/pica_state.h"
#include "video_core/regs_rasterizer.h"
#include "video_core/regs_shader.h"
#include "video_core/shader/shader.h"
#include "video_core/shader/shader_interpreter.h"
#ifdef ARCHITECTURE_x86_64
#include "video_core/shader/shader_jit_x64.h"
#endif // ARCHITECTURE_x86_64
#include "video_core/video_core.h"
namespace Pica {
namespace Shader {
OutputVertex OutputVertex::FromAttributeBuffer(const RasterizerRegs& regs,
const AttributeBuffer& input) {
// Setup output data
union {
OutputVertex ret{};
std::array<float24, 24> vertex_slots;
};
static_assert(sizeof(vertex_slots) == sizeof(ret), "Struct and array have different sizes.");
unsigned int num_attributes = regs.vs_output_total;
ASSERT(num_attributes <= 7);
for (unsigned int i = 0; i < num_attributes; ++i) {
const auto& output_register_map = regs.vs_output_attributes[i];
RasterizerRegs::VSOutputAttributes::Semantic semantics[4] = {
output_register_map.map_x, output_register_map.map_y, output_register_map.map_z,
output_register_map.map_w};
for (unsigned comp = 0; comp < 4; ++comp) {
RasterizerRegs::VSOutputAttributes::Semantic semantic = semantics[comp];
if (semantic < vertex_slots.size()) {
vertex_slots[semantic] = input.attr[i][comp];
} else if (semantic != RasterizerRegs::VSOutputAttributes::INVALID) {
LOG_ERROR(HW_GPU, "Invalid/unknown semantic id: %u", (unsigned int)semantic);
}
}
}
// The hardware takes the absolute and saturates vertex colors like this, *before* doing
// interpolation
for (unsigned i = 0; i < 4; ++i) {
float c = std::fabs(ret.color[i].ToFloat32());
ret.color[i] = float24::FromFloat32(c < 1.0f ? c : 1.0f);
}
LOG_TRACE(HW_GPU, "Output vertex: pos(%.2f, %.2f, %.2f, %.2f), quat(%.2f, %.2f, %.2f, %.2f), "
"col(%.2f, %.2f, %.2f, %.2f), tc0(%.2f, %.2f), view(%.2f, %.2f, %.2f)",
ret.pos.x.ToFloat32(), ret.pos.y.ToFloat32(), ret.pos.z.ToFloat32(),
ret.pos.w.ToFloat32(), ret.quat.x.ToFloat32(), ret.quat.y.ToFloat32(),
ret.quat.z.ToFloat32(), ret.quat.w.ToFloat32(), ret.color.x.ToFloat32(),
ret.color.y.ToFloat32(), ret.color.z.ToFloat32(), ret.color.w.ToFloat32(),
ret.tc0.u().ToFloat32(), ret.tc0.v().ToFloat32(), ret.view.x.ToFloat32(),
ret.view.y.ToFloat32(), ret.view.z.ToFloat32());
return ret;
}
void UnitState::LoadInput(const ShaderRegs& config, const AttributeBuffer& input) {
const unsigned max_attribute = config.max_input_attribute_index;
for (unsigned attr = 0; attr <= max_attribute; ++attr) {
unsigned reg = config.GetRegisterForAttribute(attr);
registers.input[reg] = input.attr[attr];
}
}
void UnitState::WriteOutput(const ShaderRegs& config, AttributeBuffer& output) {
unsigned int output_i = 0;
for (unsigned int reg : Common::BitSet<u32>(config.output_mask)) {
output.attr[output_i++] = registers.output[reg];
}
}
UnitState::UnitState(GSEmitter* emitter) : emitter_ptr(emitter) {}
GSEmitter::GSEmitter() {
handlers = new Handlers;
}
GSEmitter::~GSEmitter() {
delete handlers;
}
void GSEmitter::Emit(Math::Vec4<float24> (&vertex)[16]) {
ASSERT(vertex_id < 3);
std::copy(std::begin(vertex), std::end(vertex), buffer[vertex_id].begin());
if (prim_emit) {
if (winding)
handlers->winding_setter();
for (size_t i = 0; i < buffer.size(); ++i) {
AttributeBuffer output;
unsigned int output_i = 0;
for (unsigned int reg : Common::BitSet<u32>(output_mask)) {
output.attr[output_i++] = buffer[i][reg];
}
handlers->vertex_handler(output);
}
}
}
GSUnitState::GSUnitState() : UnitState(&emitter) {}
void GSUnitState::SetVertexHandler(VertexHandler vertex_handler, WindingSetter winding_setter) {
emitter.handlers->vertex_handler = std::move(vertex_handler);
emitter.handlers->winding_setter = std::move(winding_setter);
}
void GSUnitState::ConfigOutput(const ShaderRegs& config) {
emitter.output_mask = config.output_mask;
}
MICROPROFILE_DEFINE(GPU_Shader, "GPU", "Shader", MP_RGB(50, 50, 240));
#ifdef ARCHITECTURE_x86_64
static std::unique_ptr<JitX64Engine> jit_engine;
#endif // ARCHITECTURE_x86_64
static InterpreterEngine interpreter_engine;
ShaderEngine* GetEngine() {
#ifdef ARCHITECTURE_x86_64
// TODO(yuriks): Re-initialize on each change rather than being persistent
if (VideoCore::g_shader_jit_enabled) {
if (jit_engine == nullptr) {
jit_engine = std::make_unique<JitX64Engine>();
}
return jit_engine.get();
}
#endif // ARCHITECTURE_x86_64
return &interpreter_engine;
}
void Shutdown() {
#ifdef ARCHITECTURE_x86_64
jit_engine = nullptr;
#endif // ARCHITECTURE_x86_64
}
} // namespace Shader
} // namespace Pica

233
src/video_core/shader/shader.h
View file

@ -1,233 +0,0 @@
// Copyright 2015 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <array>
#include <cstddef>
#include <functional>
#include <type_traits>
#include <nihstro/shader_bytecode.h>
#include "common/assert.h"
#include "common/common_funcs.h"
#include "common/common_types.h"
#include "common/vector_math.h"
#include "video_core/pica_types.h"
#include "video_core/regs_rasterizer.h"
#include "video_core/regs_shader.h"
using nihstro::RegisterType;
using nihstro::SourceRegister;
using nihstro::DestRegister;
namespace Pica {
namespace Shader {
constexpr unsigned MAX_PROGRAM_CODE_LENGTH = 4096;
constexpr unsigned MAX_SWIZZLE_DATA_LENGTH = 4096;
struct AttributeBuffer {
alignas(16) Math::Vec4<float24> attr[16];
};
/// Handler type for receiving vertex outputs from vertex shader or geometry shader
using VertexHandler = std::function<void(const AttributeBuffer&)>;
/// Handler type for signaling to invert the vertex order of the next triangle
using WindingSetter = std::function<void()>;
struct OutputVertex {
Math::Vec4<float24> pos;
Math::Vec4<float24> quat;
Math::Vec4<float24> color;
Math::Vec2<float24> tc0;
Math::Vec2<float24> tc1;
float24 tc0_w;
INSERT_PADDING_WORDS(1);
Math::Vec3<float24> view;
INSERT_PADDING_WORDS(1);
Math::Vec2<float24> tc2;
static OutputVertex FromAttributeBuffer(const RasterizerRegs& regs,
const AttributeBuffer& output);
};
#define ASSERT_POS(var, pos) \
static_assert(offsetof(OutputVertex, var) == pos * sizeof(float24), "Semantic at wrong " \
"offset.")
ASSERT_POS(pos, RasterizerRegs::VSOutputAttributes::POSITION_X);
ASSERT_POS(quat, RasterizerRegs::VSOutputAttributes::QUATERNION_X);
ASSERT_POS(color, RasterizerRegs::VSOutputAttributes::COLOR_R);
ASSERT_POS(tc0, RasterizerRegs::VSOutputAttributes::TEXCOORD0_U);
ASSERT_POS(tc1, RasterizerRegs::VSOutputAttributes::TEXCOORD1_U);
ASSERT_POS(tc0_w, RasterizerRegs::VSOutputAttributes::TEXCOORD0_W);
ASSERT_POS(view, RasterizerRegs::VSOutputAttributes::VIEW_X);
ASSERT_POS(tc2, RasterizerRegs::VSOutputAttributes::TEXCOORD2_U);
#undef ASSERT_POS
static_assert(std::is_pod<OutputVertex>::value, "Structure is not POD");
static_assert(sizeof(OutputVertex) == 24 * sizeof(float), "OutputVertex has invalid size");
/**
* This structure contains state information for primitive emitting in geometry shader.
*/
struct GSEmitter {
std::array<std::array<Math::Vec4<float24>, 16>, 3> buffer;
u8 vertex_id;
bool prim_emit;
bool winding;
u32 output_mask;
// Function objects are hidden behind a raw pointer to make the structure standard layout type,
// for JIT to use offsetof to access other members.
struct Handlers {
VertexHandler vertex_handler;
WindingSetter winding_setter;
} * handlers;
GSEmitter();
~GSEmitter();
void Emit(Math::Vec4<float24> (&vertex)[16]);
};
static_assert(std::is_standard_layout<GSEmitter>::value, "GSEmitter is not standard layout type");
/**
* This structure contains the state information that needs to be unique for a shader unit. The 3DS
* has four shader units that process shaders in parallel. At the present, Citra only implements a
* single shader unit that processes all shaders serially. Putting the state information in a struct
* here will make it easier for us to parallelize the shader processing later.
*/
struct UnitState {
explicit UnitState(GSEmitter* emitter = nullptr);
struct Registers {
// The registers are accessed by the shader JIT using SSE instructions, and are therefore
// required to be 16-byte aligned.
alignas(16) Math::Vec4<float24> input[16];
alignas(16) Math::Vec4<float24> temporary[16];
alignas(16) Math::Vec4<float24> output[16];
} registers;
static_assert(std::is_pod<Registers>::value, "Structure is not POD");
bool conditional_code[2];
// Two Address registers and one loop counter
// TODO: How many bits do these actually have?
s32 address_registers[3];
GSEmitter* emitter_ptr;
static size_t InputOffset(const SourceRegister& reg) {
switch (reg.GetRegisterType()) {
case RegisterType::Input:
return offsetof(UnitState, registers.input) +
reg.GetIndex() * sizeof(Math::Vec4<float24>);
case RegisterType::Temporary:
return offsetof(UnitState, registers.temporary) +
reg.GetIndex() * sizeof(Math::Vec4<float24>);
default:
UNREACHABLE();
return 0;
}
}
static size_t OutputOffset(const DestRegister& reg) {
switch (reg.GetRegisterType()) {
case RegisterType::Output:
return offsetof(UnitState, registers.output) +
reg.GetIndex() * sizeof(Math::Vec4<float24>);
case RegisterType::Temporary:
return offsetof(UnitState, registers.temporary) +
reg.GetIndex() * sizeof(Math::Vec4<float24>);
default:
UNREACHABLE();
return 0;
}
}
/**
* Loads the unit state with an input vertex.
*
* @param config Shader configuration registers corresponding to the unit.
* @param input Attribute buffer to load into the input registers.
*/
void LoadInput(const ShaderRegs& config, const AttributeBuffer& input);
void WriteOutput(const ShaderRegs& config, AttributeBuffer& output);
};
/**
* This is an extended shader unit state that represents the special unit that can run both vertex
* shader and geometry shader. It contains an additional primitive emitter and utilities for
* geometry shader.
*/
struct GSUnitState : public UnitState {
GSUnitState();
void SetVertexHandler(VertexHandler vertex_handler, WindingSetter winding_setter);
void ConfigOutput(const ShaderRegs& config);
GSEmitter emitter;
};
struct ShaderSetup {
struct {
// The float uniforms are accessed by the shader JIT using SSE instructions, and are
// therefore required to be 16-byte aligned.
alignas(16) Math::Vec4<float24> f[96];
std::array<bool, 16> b;
std::array<Math::Vec4<u8>, 4> i;
} uniforms;
static size_t GetFloatUniformOffset(unsigned index) {
return offsetof(ShaderSetup, uniforms.f) + index * sizeof(Math::Vec4<float24>);
}
static size_t GetBoolUniformOffset(unsigned index) {
return offsetof(ShaderSetup, uniforms.b) + index * sizeof(bool);
}
static size_t GetIntUniformOffset(unsigned index) {
return offsetof(ShaderSetup, uniforms.i) + index * sizeof(Math::Vec4<u8>);
}
std::array<u32, MAX_PROGRAM_CODE_LENGTH> program_code;
std::array<u32, MAX_SWIZZLE_DATA_LENGTH> swizzle_data;
/// Data private to ShaderEngines
struct EngineData {
unsigned int entry_point;
/// Used by the JIT, points to a compiled shader object.
const void* cached_shader = nullptr;
} engine_data;
};
class ShaderEngine {
public:
virtual ~ShaderEngine() = default;
/**
* Performs any shader unit setup that only needs to happen once per shader (as opposed to once
* per vertex, which would happen within the `Run` function).
*/
virtual void SetupBatch(ShaderSetup& setup, unsigned int entry_point) = 0;
/**
* Runs the currently setup shader.
*
* @param setup Shader engine state, must be setup with SetupBatch on each shader change.
* @param state Shader unit state, must be setup with input data before each shader invocation.
*/
virtual void Run(const ShaderSetup& setup, UnitState& state) const = 0;
};
// TODO(yuriks): Remove and make it non-global state somewhere
ShaderEngine* GetEngine();
void Shutdown();
} // namespace Shader
} // namespace Pica

701
src/video_core/shader/shader_interpreter.cpp
View file

@ -1,701 +0,0 @@
// Copyright 2014 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <algorithm>
#include <array>
#include <cmath>
#include <numeric>
#include <boost/container/static_vector.hpp>
#include <boost/range/algorithm/fill.hpp>
#include <nihstro/shader_bytecode.h>
#include "common/assert.h"
#include "common/common_types.h"
#include "common/logging/log.h"
#include "common/microprofile.h"
#include "common/vector_math.h"
#include "video_core/pica_state.h"
#include "video_core/pica_types.h"
#include "video_core/shader/shader.h"
#include "video_core/shader/shader_interpreter.h"
using nihstro::OpCode;
using nihstro::Instruction;
using nihstro::RegisterType;
using nihstro::SourceRegister;
using nihstro::SwizzlePattern;
namespace Pica {
namespace Shader {
struct CallStackElement {
u32 final_address; // Address upon which we jump to return_address
u32 return_address; // Where to jump when leaving scope
u8 repeat_counter; // How often to repeat until this call stack element is removed
u8 loop_increment; // Which value to add to the loop counter after an iteration
// TODO: Should this be a signed value? Does it even matter?
u32 loop_address; // The address where we'll return to after each loop iteration
};
template <bool Debug>
static void RunInterpreter(const ShaderSetup& setup, UnitState& state, DebugData<Debug>& debug_data,
unsigned offset) {
// TODO: Is there a maximal size for this?
boost::container::static_vector<CallStackElement, 16> call_stack;
u32 program_counter = offset;
state.conditional_code[0] = false;
state.conditional_code[1] = false;
auto call = [&program_counter, &call_stack](u32 offset, u32 num_instructions, u32 return_offset,
u8 repeat_count, u8 loop_increment) {
// -1 to make sure when incrementing the PC we end up at the correct offset
program_counter = offset - 1;
ASSERT(call_stack.size() < call_stack.capacity());
call_stack.push_back(
{offset + num_instructions, return_offset, repeat_count, loop_increment, offset});
};
auto evaluate_condition = [&state](Instruction::FlowControlType flow_control) {
using Op = Instruction::FlowControlType::Op;
bool result_x = flow_control.refx.Value() == state.conditional_code[0];
bool result_y = flow_control.refy.Value() == state.conditional_code[1];
switch (flow_control.op) {
case Op::Or:
return result_x || result_y;
case Op::And:
return result_x && result_y;
case Op::JustX:
return result_x;
case Op::JustY:
return result_y;
default:
UNREACHABLE();
return false;
}
};
const auto& uniforms = setup.uniforms;
const auto& swizzle_data = setup.swizzle_data;
const auto& program_code = setup.program_code;
// Placeholder for invalid inputs
static float24 dummy_vec4_float24[4];
unsigned iteration = 0;
bool exit_loop = false;
while (!exit_loop) {
if (!call_stack.empty()) {
auto& top = call_stack.back();
if (program_counter == top.final_address) {
state.address_registers[2] += top.loop_increment;
if (top.repeat_counter-- == 0) {
program_counter = top.return_address;
call_stack.pop_back();
} else {
program_counter = top.loop_address;
}
// TODO: Is "trying again" accurate to hardware?
continue;
}
}
const Instruction instr = {program_code[program_counter]};
const SwizzlePattern swizzle = {swizzle_data[instr.common.operand_desc_id]};
Record<DebugDataRecord::CUR_INSTR>(debug_data, iteration, program_counter);
if (iteration > 0)
Record<DebugDataRecord::NEXT_INSTR>(debug_data, iteration - 1, program_counter);
debug_data.max_offset = std::max<u32>(debug_data.max_offset, 1 + program_counter);
auto LookupSourceRegister = [&](const SourceRegister& source_reg) -> const float24* {
switch (source_reg.GetRegisterType()) {
case RegisterType::Input:
return &state.registers.input[source_reg.GetIndex()].x;
case RegisterType::Temporary:
return &state.registers.temporary[source_reg.GetIndex()].x;
case RegisterType::FloatUniform:
return &uniforms.f[source_reg.GetIndex()].x;
default:
return dummy_vec4_float24;
}
};
switch (instr.opcode.Value().GetInfo().type) {
case OpCode::Type::Arithmetic: {
const bool is_inverted =
(0 != (instr.opcode.Value().GetInfo().subtype & OpCode::Info::SrcInversed));
const int address_offset =
(instr.common.address_register_index == 0)
? 0
: state.address_registers[instr.common.address_register_index - 1];
const float24* src1_ = LookupSourceRegister(instr.common.GetSrc1(is_inverted) +
(is_inverted ? 0 : address_offset));
const float24* src2_ = LookupSourceRegister(instr.common.GetSrc2(is_inverted) +
(is_inverted ? address_offset : 0));
const bool negate_src1 = ((bool)swizzle.negate_src1 != false);
const bool negate_src2 = ((bool)swizzle.negate_src2 != false);
float24 src1[4] = {
src1_[(int)swizzle.src1_selector_0.Value()],
src1_[(int)swizzle.src1_selector_1.Value()],
src1_[(int)swizzle.src1_selector_2.Value()],
src1_[(int)swizzle.src1_selector_3.Value()],
};
if (negate_src1) {
src1[0] = -src1[0];
src1[1] = -src1[1];
src1[2] = -src1[2];
src1[3] = -src1[3];
}
float24 src2[4] = {
src2_[(int)swizzle.src2_selector_0.Value()],
src2_[(int)swizzle.src2_selector_1.Value()],
src2_[(int)swizzle.src2_selector_2.Value()],
src2_[(int)swizzle.src2_selector_3.Value()],
};
if (negate_src2) {
src2[0] = -src2[0];
src2[1] = -src2[1];
src2[2] = -src2[2];
src2[3] = -src2[3];
}
float24* dest =
(instr.common.dest.Value() < 0x10)
? &state.registers.output[instr.common.dest.Value().GetIndex()][0]
: (instr.common.dest.Value() < 0x20)
? &state.registers.temporary[instr.common.dest.Value().GetIndex()][0]
: dummy_vec4_float24;
debug_data.max_opdesc_id =
std::max<u32>(debug_data.max_opdesc_id, 1 + instr.common.operand_desc_id);
switch (instr.opcode.Value().EffectiveOpCode()) {
case OpCode::Id::ADD: {
Record<DebugDataRecord::SRC1>(debug_data, iteration, src1);
Record<DebugDataRecord::SRC2>(debug_data, iteration, src2);
Record<DebugDataRecord::DEST_IN>(debug_data, iteration, dest);
for (int i = 0; i < 4; ++i) {
if (!swizzle.DestComponentEnabled(i))
continue;
dest[i] = src1[i] + src2[i];
}
Record<DebugDataRecord::DEST_OUT>(debug_data, iteration, dest);
break;
}
case OpCode::Id::MUL: {
Record<DebugDataRecord::SRC1>(debug_data, iteration, src1);
Record<DebugDataRecord::SRC2>(debug_data, iteration, src2);
Record<DebugDataRecord::DEST_IN>(debug_data, iteration, dest);
for (int i = 0; i < 4; ++i) {
if (!swizzle.DestComponentEnabled(i))
continue;
dest[i] = src1[i] * src2[i];
}
Record<DebugDataRecord::DEST_OUT>(debug_data, iteration, dest);
break;
}
case OpCode::Id::FLR:
Record<DebugDataRecord::SRC1>(debug_data, iteration, src1);
Record<DebugDataRecord::DEST_IN>(debug_data, iteration, dest);
for (int i = 0; i < 4; ++i) {
if (!swizzle.DestComponentEnabled(i))
continue;
dest[i] = float24::FromFloat32(std::floor(src1[i].ToFloat32()));
}
Record<DebugDataRecord::DEST_OUT>(debug_data, iteration, dest);
break;
case OpCode::Id::MAX:
Record<DebugDataRecord::SRC1>(debug_data, iteration, src1);
Record<DebugDataRecord::SRC2>(debug_data, iteration, src2);
Record<DebugDataRecord::DEST_IN>(debug_data, iteration, dest);
for (int i = 0; i < 4; ++i) {
if (!swizzle.DestComponentEnabled(i))
continue;
// NOTE: Exact form required to match NaN semantics to hardware:
// max(0, NaN) -> NaN
// max(NaN, 0) -> 0
dest[i] = (src1[i] > src2[i]) ? src1[i] : src2[i];
}
Record<DebugDataRecord::DEST_OUT>(debug_data, iteration, dest);
break;
case OpCode::Id::MIN:
Record<DebugDataRecord::SRC1>(debug_data, iteration, src1);
Record<DebugDataRecord::SRC2>(debug_data, iteration, src2);
Record<DebugDataRecord::DEST_IN>(debug_data, iteration, dest);
for (int i = 0; i < 4; ++i) {
if (!swizzle.DestComponentEnabled(i))
continue;
// NOTE: Exact form required to match NaN semantics to hardware:
// min(0, NaN) -> NaN
// min(NaN, 0) -> 0
dest[i] = (src1[i] < src2[i]) ? src1[i] : src2[i];
}
Record<DebugDataRecord::DEST_OUT>(debug_data, iteration, dest);
break;
case OpCode::Id::DP3:
case OpCode::Id::DP4:
case OpCode::Id::DPH:
case OpCode::Id::DPHI: {
Record<DebugDataRecord::SRC1>(debug_data, iteration, src1);
Record<DebugDataRecord::SRC2>(debug_data, iteration, src2);
Record<DebugDataRecord::DEST_IN>(debug_data, iteration, dest);
OpCode::Id opcode = instr.opcode.Value().EffectiveOpCode();
if (opcode == OpCode::Id::DPH || opcode == OpCode::Id::DPHI)
src1[3] = float24::FromFloat32(1.0f);
int num_components = (opcode == OpCode::Id::DP3) ? 3 : 4;
float24 dot = std::inner_product(src1, src1 + num_components, src2,
float24::FromFloat32(0.f));
for (int i = 0; i < 4; ++i) {
if (!swizzle.DestComponentEnabled(i))
continue;
dest[i] = dot;
}
Record<DebugDataRecord::DEST_OUT>(debug_data, iteration, dest);
break;
}
// Reciprocal
case OpCode::Id::RCP: {
Record<DebugDataRecord::SRC1>(debug_data, iteration, src1);
Record<DebugDataRecord::DEST_IN>(debug_data, iteration, dest);
float24 rcp_res = float24::FromFloat32(1.0f / src1[0].ToFloat32());
for (int i = 0; i < 4; ++i) {
if (!swizzle.DestComponentEnabled(i))
continue;
dest[i] = rcp_res;
}
Record<DebugDataRecord::DEST_OUT>(debug_data, iteration, dest);
break;
}
// Reciprocal Square Root
case OpCode::Id::RSQ: {
Record<DebugDataRecord::SRC1>(debug_data, iteration, src1);
Record<DebugDataRecord::DEST_IN>(debug_data, iteration, dest);
float24 rsq_res = float24::FromFloat32(1.0f / std::sqrt(src1[0].ToFloat32()));
for (int i = 0; i < 4; ++i) {
if (!swizzle.DestComponentEnabled(i))
continue;
dest[i] = rsq_res;
}
Record<DebugDataRecord::DEST_OUT>(debug_data, iteration, dest);
break;
}
case OpCode::Id::MOVA: {
Record<DebugDataRecord::SRC1>(debug_data, iteration, src1);
for (int i = 0; i < 2; ++i) {
if (!swizzle.DestComponentEnabled(i))
continue;
// TODO: Figure out how the rounding is done on hardware
state.address_registers[i] = static_cast<s32>(src1[i].ToFloat32());
}
Record<DebugDataRecord::ADDR_REG_OUT>(debug_data, iteration,
state.address_registers);
break;
}
case OpCode::Id::MOV: {
Record<DebugDataRecord::SRC1>(debug_data, iteration, src1);
Record<DebugDataRecord::DEST_IN>(debug_data, iteration, dest);
for (int i = 0; i < 4; ++i) {
if (!swizzle.DestComponentEnabled(i))
continue;
dest[i] = src1[i];
}
Record<DebugDataRecord::DEST_OUT>(debug_data, iteration, dest);
break;
}
case OpCode::Id::SGE:
case OpCode::Id::SGEI:
Record<DebugDataRecord::SRC1>(debug_data, iteration, src1);
Record<DebugDataRecord::SRC2>(debug_data, iteration, src2);
Record<DebugDataRecord::DEST_IN>(debug_data, iteration, dest);
for (int i = 0; i < 4; ++i) {
if (!swizzle.DestComponentEnabled(i))
continue;
dest[i] = (src1[i] >= src2[i]) ? float24::FromFloat32(1.0f)
: float24::FromFloat32(0.0f);
}
Record<DebugDataRecord::DEST_OUT>(debug_data, iteration, dest);
break;
case OpCode::Id::SLT:
case OpCode::Id::SLTI:
Record<DebugDataRecord::SRC1>(debug_data, iteration, src1);
Record<DebugDataRecord::SRC2>(debug_data, iteration, src2);
Record<DebugDataRecord::DEST_IN>(debug_data, iteration, dest);
for (int i = 0; i < 4; ++i) {
if (!swizzle.DestComponentEnabled(i))
continue;
dest[i] = (src1[i] < src2[i]) ? float24::FromFloat32(1.0f)
: float24::FromFloat32(0.0f);
}
Record<DebugDataRecord::DEST_OUT>(debug_data, iteration, dest);
break;
case OpCode::Id::CMP:
Record<DebugDataRecord::SRC1>(debug_data, iteration, src1);
Record<DebugDataRecord::SRC2>(debug_data, iteration, src2);
for (int i = 0; i < 2; ++i) {
// TODO: Can you restrict to one compare via dest masking?
auto compare_op = instr.common.compare_op;
auto op = (i == 0) ? compare_op.x.Value() : compare_op.y.Value();
switch (op) {
case Instruction::Common::CompareOpType::Equal:
state.conditional_code[i] = (src1[i] == src2[i]);
break;
case Instruction::Common::CompareOpType::NotEqual:
state.conditional_code[i] = (src1[i] != src2[i]);
break;
case Instruction::Common::CompareOpType::LessThan:
state.conditional_code[i] = (src1[i] < src2[i]);
break;
case Instruction::Common::CompareOpType::LessEqual:
state.conditional_code[i] = (src1[i] <= src2[i]);
break;
case Instruction::Common::CompareOpType::GreaterThan:
state.conditional_code[i] = (src1[i] > src2[i]);
break;
case Instruction::Common::CompareOpType::GreaterEqual:
state.conditional_code[i] = (src1[i] >= src2[i]);
break;
default:
LOG_ERROR(HW_GPU, "Unknown compare mode %x", static_cast<int>(op));
break;
}
}
Record<DebugDataRecord::CMP_RESULT>(debug_data, iteration, state.conditional_code);
break;
case OpCode::Id::EX2: {
Record<DebugDataRecord::SRC1>(debug_data, iteration, src1);
Record<DebugDataRecord::DEST_IN>(debug_data, iteration, dest);
// EX2 only takes first component exp2 and writes it to all dest components
float24 ex2_res = float24::FromFloat32(std::exp2(src1[0].ToFloat32()));
for (int i = 0; i < 4; ++i) {
if (!swizzle.DestComponentEnabled(i))
continue;
dest[i] = ex2_res;
}
Record<DebugDataRecord::DEST_OUT>(debug_data, iteration, dest);
break;
}
case OpCode::Id::LG2: {
Record<DebugDataRecord::SRC1>(debug_data, iteration, src1);
Record<DebugDataRecord::DEST_IN>(debug_data, iteration, dest);
// LG2 only takes the first component log2 and writes it to all dest components
float24 lg2_res = float24::FromFloat32(std::log2(src1[0].ToFloat32()));
for (int i = 0; i < 4; ++i) {
if (!swizzle.DestComponentEnabled(i))
continue;
dest[i] = lg2_res;
}
Record<DebugDataRecord::DEST_OUT>(debug_data, iteration, dest);
break;
}
default:
LOG_ERROR(HW_GPU, "Unhandled arithmetic instruction: 0x%02x (%s): 0x%08x",
(int)instr.opcode.Value().EffectiveOpCode(),
instr.opcode.Value().GetInfo().name, instr.hex);
DEBUG_ASSERT(false);
break;
}
break;
}
case OpCode::Type::MultiplyAdd: {
if ((instr.opcode.Value().EffectiveOpCode() == OpCode::Id::MAD) ||
(instr.opcode.Value().EffectiveOpCode() == OpCode::Id::MADI)) {
const SwizzlePattern& swizzle = *reinterpret_cast<const SwizzlePattern*>(
&swizzle_data[instr.mad.operand_desc_id]);
bool is_inverted = (instr.opcode.Value().EffectiveOpCode() == OpCode::Id::MADI);
const int address_offset =
(instr.mad.address_register_index == 0)
? 0
: state.address_registers[instr.mad.address_register_index - 1];
const float24* src1_ = LookupSourceRegister(instr.mad.GetSrc1(is_inverted));
const float24* src2_ = LookupSourceRegister(instr.mad.GetSrc2(is_inverted) +
(!is_inverted * address_offset));
const float24* src3_ = LookupSourceRegister(instr.mad.GetSrc3(is_inverted) +
(is_inverted * address_offset));
const bool negate_src1 = ((bool)swizzle.negate_src1 != false);
const bool negate_src2 = ((bool)swizzle.negate_src2 != false);
const bool negate_src3 = ((bool)swizzle.negate_src3 != false);
float24 src1[4] = {
src1_[(int)swizzle.src1_selector_0.Value()],
src1_[(int)swizzle.src1_selector_1.Value()],
src1_[(int)swizzle.src1_selector_2.Value()],
src1_[(int)swizzle.src1_selector_3.Value()],
};
if (negate_src1) {
src1[0] = -src1[0];
src1[1] = -src1[1];
src1[2] = -src1[2];
src1[3] = -src1[3];
}
float24 src2[4] = {
src2_[(int)swizzle.src2_selector_0.Value()],
src2_[(int)swizzle.src2_selector_1.Value()],
src2_[(int)swizzle.src2_selector_2.Value()],
src2_[(int)swizzle.src2_selector_3.Value()],
};
if (negate_src2) {
src2[0] = -src2[0];
src2[1] = -src2[1];
src2[2] = -src2[2];
src2[3] = -src2[3];
}
float24 src3[4] = {
src3_[(int)swizzle.src3_selector_0.Value()],
src3_[(int)swizzle.src3_selector_1.Value()],
src3_[(int)swizzle.src3_selector_2.Value()],
src3_[(int)swizzle.src3_selector_3.Value()],
};
if (negate_src3) {
src3[0] = -src3[0];
src3[1] = -src3[1];
src3[2] = -src3[2];
src3[3] = -src3[3];
}
float24* dest =
(instr.mad.dest.Value() < 0x10)
? &state.registers.output[instr.mad.dest.Value().GetIndex()][0]
: (instr.mad.dest.Value() < 0x20)
? &state.registers.temporary[instr.mad.dest.Value().GetIndex()][0]
: dummy_vec4_float24;
Record<DebugDataRecord::SRC1>(debug_data, iteration, src1);
Record<DebugDataRecord::SRC2>(debug_data, iteration, src2);
Record<DebugDataRecord::SRC3>(debug_data, iteration, src3);
Record<DebugDataRecord::DEST_IN>(debug_data, iteration, dest);
for (int i = 0; i < 4; ++i) {
if (!swizzle.DestComponentEnabled(i))
continue;
dest[i] = src1[i] * src2[i] + src3[i];
}
Record<DebugDataRecord::DEST_OUT>(debug_data, iteration, dest);
} else {
LOG_ERROR(HW_GPU, "Unhandled multiply-add instruction: 0x%02x (%s): 0x%08x",
(int)instr.opcode.Value().EffectiveOpCode(),
instr.opcode.Value().GetInfo().name, instr.hex);
}
break;
}
default: {
// Handle each instruction on its own
switch (instr.opcode.Value()) {
case OpCode::Id::END:
exit_loop = true;
break;
case OpCode::Id::JMPC:
Record<DebugDataRecord::COND_CMP_IN>(debug_data, iteration, state.conditional_code);
if (evaluate_condition(instr.flow_control)) {
program_counter = instr.flow_control.dest_offset - 1;
}
break;
case OpCode::Id::JMPU:
Record<DebugDataRecord::COND_BOOL_IN>(
debug_data, iteration, uniforms.b[instr.flow_control.bool_uniform_id]);
if (uniforms.b[instr.flow_control.bool_uniform_id] ==
!(instr.flow_control.num_instructions & 1)) {
program_counter = instr.flow_control.dest_offset - 1;
}
break;
case OpCode::Id::CALL:
call(instr.flow_control.dest_offset, instr.flow_control.num_instructions,
program_counter + 1, 0, 0);
break;
case OpCode::Id::CALLU:
Record<DebugDataRecord::COND_BOOL_IN>(
debug_data, iteration, uniforms.b[instr.flow_control.bool_uniform_id]);
if (uniforms.b[instr.flow_control.bool_uniform_id]) {
call(instr.flow_control.dest_offset, instr.flow_control.num_instructions,
program_counter + 1, 0, 0);
}
break;
case OpCode::Id::CALLC:
Record<DebugDataRecord::COND_CMP_IN>(debug_data, iteration, state.conditional_code);
if (evaluate_condition(instr.flow_control)) {
call(instr.flow_control.dest_offset, instr.flow_control.num_instructions,
program_counter + 1, 0, 0);
}
break;
case OpCode::Id::NOP:
break;
case OpCode::Id::IFU:
Record<DebugDataRecord::COND_BOOL_IN>(
debug_data, iteration, uniforms.b[instr.flow_control.bool_uniform_id]);
if (uniforms.b[instr.flow_control.bool_uniform_id]) {
call(program_counter + 1, instr.flow_control.dest_offset - program_counter - 1,
instr.flow_control.dest_offset + instr.flow_control.num_instructions, 0,
0);
} else {
call(instr.flow_control.dest_offset, instr.flow_control.num_instructions,
instr.flow_control.dest_offset + instr.flow_control.num_instructions, 0,
0);
}
break;
case OpCode::Id::IFC: {
// TODO: Do we need to consider swizzlers here?
Record<DebugDataRecord::COND_CMP_IN>(debug_data, iteration, state.conditional_code);
if (evaluate_condition(instr.flow_control)) {
call(program_counter + 1, instr.flow_control.dest_offset - program_counter - 1,
instr.flow_control.dest_offset + instr.flow_control.num_instructions, 0,
0);
} else {
call(instr.flow_control.dest_offset, instr.flow_control.num_instructions,
instr.flow_control.dest_offset + instr.flow_control.num_instructions, 0,
0);
}
break;
}
case OpCode::Id::LOOP: {
Math::Vec4<u8> loop_param(uniforms.i[instr.flow_control.int_uniform_id].x,
uniforms.i[instr.flow_control.int_uniform_id].y,
uniforms.i[instr.flow_control.int_uniform_id].z,
uniforms.i[instr.flow_control.int_uniform_id].w);
state.address_registers[2] = loop_param.y;
Record<DebugDataRecord::LOOP_INT_IN>(debug_data, iteration, loop_param);
call(program_counter + 1, instr.flow_control.dest_offset - program_counter,
instr.flow_control.dest_offset + 1, loop_param.x, loop_param.z);
break;
}
case OpCode::Id::EMIT: {
GSEmitter* emitter = state.emitter_ptr;
ASSERT_MSG(emitter, "Execute EMIT on VS");
emitter->Emit(state.registers.output);
break;
}
case OpCode::Id::SETEMIT: {
GSEmitter* emitter = state.emitter_ptr;
ASSERT_MSG(emitter, "Execute SETEMIT on VS");
emitter->vertex_id = instr.setemit.vertex_id;
emitter->prim_emit = instr.setemit.prim_emit != 0;
emitter->winding = instr.setemit.winding != 0;
break;
}
default:
LOG_ERROR(HW_GPU, "Unhandled instruction: 0x%02x (%s): 0x%08x",
(int)instr.opcode.Value().EffectiveOpCode(),
instr.opcode.Value().GetInfo().name, instr.hex);
break;
}
break;
}
}
++program_counter;
++iteration;
}
}
void InterpreterEngine::SetupBatch(ShaderSetup& setup, unsigned int entry_point) {
ASSERT(entry_point < MAX_PROGRAM_CODE_LENGTH);
setup.engine_data.entry_point = entry_point;
}
MICROPROFILE_DECLARE(GPU_Shader);
void InterpreterEngine::Run(const ShaderSetup& setup, UnitState& state) const {
MICROPROFILE_SCOPE(GPU_Shader);
DebugData<false> dummy_debug_data;
RunInterpreter(setup, state, dummy_debug_data, setup.engine_data.entry_point);
}
DebugData<true> InterpreterEngine::ProduceDebugInfo(const ShaderSetup& setup,
const AttributeBuffer& input,
const ShaderRegs& config) const {
UnitState state;
DebugData<true> debug_data;
// Setup input register table
boost::fill(state.registers.input, Math::Vec4<float24>::AssignToAll(float24::Zero()));
state.LoadInput(config, input);
RunInterpreter(setup, state, debug_data, setup.engine_data.entry_point);
return debug_data;
}
} // namespace
} // namespace

32
src/video_core/shader/shader_interpreter.h
View file

@ -1,32 +0,0 @@
// Copyright 2014 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include "video_core/shader/debug_data.h"
#include "video_core/shader/shader.h"
namespace Pica {
namespace Shader {
class InterpreterEngine final : public ShaderEngine {
public:
void SetupBatch(ShaderSetup& setup, unsigned int entry_point) override;
void Run(const ShaderSetup& setup, UnitState& state) const override;
/**
* Produce debug information based on the given shader and input vertex
* @param setup Shader engine state
* @param input Input vertex into the shader
* @param config Configuration object for the shader pipeline
* @return Debug information for this shader with regards to the given vertex
*/
DebugData<true> ProduceDebugInfo(const ShaderSetup& setup, const AttributeBuffer& input,
const ShaderRegs& config) const;
};
} // namespace
} // namespace

48
src/video_core/shader/shader_jit_x64.cpp
View file

@ -1,48 +0,0 @@
// Copyright 2016 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/hash.h"
#include "common/microprofile.h"
#include "video_core/shader/shader.h"
#include "video_core/shader/shader_jit_x64.h"
#include "video_core/shader/shader_jit_x64_compiler.h"
namespace Pica {
namespace Shader {
JitX64Engine::JitX64Engine() = default;
JitX64Engine::~JitX64Engine() = default;
void JitX64Engine::SetupBatch(ShaderSetup& setup, unsigned int entry_point) {
ASSERT(entry_point < MAX_PROGRAM_CODE_LENGTH);
setup.engine_data.entry_point = entry_point;
u64 code_hash = Common::ComputeHash64(&setup.program_code, sizeof(setup.program_code));
u64 swizzle_hash = Common::ComputeHash64(&setup.swizzle_data, sizeof(setup.swizzle_data));
u64 cache_key = code_hash ^ swizzle_hash;
auto iter = cache.find(cache_key);
if (iter != cache.end()) {
setup.engine_data.cached_shader = iter->second.get();
} else {
auto shader = std::make_unique<JitShader>();
shader->Compile(&setup.program_code, &setup.swizzle_data);
setup.engine_data.cached_shader = shader.get();
cache.emplace_hint(iter, cache_key, std::move(shader));
}
}
MICROPROFILE_DECLARE(GPU_Shader);
void JitX64Engine::Run(const ShaderSetup& setup, UnitState& state) const {
ASSERT(setup.engine_data.cached_shader != nullptr);
MICROPROFILE_SCOPE(GPU_Shader);
const JitShader* shader = static_cast<const JitShader*>(setup.engine_data.cached_shader);
shader->Run(setup, state, setup.engine_data.entry_point);
}
} // namespace Shader
} // namespace Pica

30
src/video_core/shader/shader_jit_x64.h
View file

@ -1,30 +0,0 @@
// Copyright 2016 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <memory>
#include <unordered_map>
#include "common/common_types.h"
#include "video_core/shader/shader.h"
namespace Pica {
namespace Shader {
class JitShader;
class JitX64Engine final : public ShaderEngine {
public:
JitX64Engine();
~JitX64Engine() override;
void SetupBatch(ShaderSetup& setup, unsigned int entry_point) override;
void Run(const ShaderSetup& setup, UnitState& state) const override;
private:
std::unordered_map<u64, std::unique_ptr<JitShader>> cache;
};
} // namespace Shader
} // namespace Pica

942
src/video_core/shader/shader_jit_x64_compiler.cpp
View file

@ -1,942 +0,0 @@
// Copyright 2015 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <algorithm>
#include <cmath>
#include <cstdint>
#include <nihstro/shader_bytecode.h>
#include <smmintrin.h>
#include <xmmintrin.h>
#include "common/assert.h"
#include "common/logging/log.h"
#include "common/vector_math.h"
#include "common/x64/cpu_detect.h"
#include "common/x64/xbyak_abi.h"
#include "common/x64/xbyak_util.h"
#include "video_core/pica_state.h"
#include "video_core/pica_types.h"
#include "video_core/shader/shader.h"
#include "video_core/shader/shader_jit_x64_compiler.h"
using namespace Common::X64;
using namespace Xbyak::util;
using Xbyak::Label;
using Xbyak::Reg32;
using Xbyak::Reg64;
using Xbyak::Xmm;
namespace Pica {
namespace Shader {
typedef void (JitShader::*JitFunction)(Instruction instr);
const JitFunction instr_table[64] = {
&JitShader::Compile_ADD, // add
&JitShader::Compile_DP3, // dp3
&JitShader::Compile_DP4, // dp4
&JitShader::Compile_DPH, // dph
nullptr, // unknown
&JitShader::Compile_EX2, // ex2
&JitShader::Compile_LG2, // lg2
nullptr, // unknown
&JitShader::Compile_MUL, // mul
&JitShader::Compile_SGE, // sge
&JitShader::Compile_SLT, // slt
&JitShader::Compile_FLR, // flr
&JitShader::Compile_MAX, // max
&JitShader::Compile_MIN, // min
&JitShader::Compile_RCP, // rcp
&JitShader::Compile_RSQ, // rsq
nullptr, // unknown
nullptr, // unknown
&JitShader::Compile_MOVA, // mova
&JitShader::Compile_MOV, // mov
nullptr, // unknown
nullptr, // unknown
nullptr, // unknown
nullptr, // unknown
&JitShader::Compile_DPH, // dphi
nullptr, // unknown
&JitShader::Compile_SGE, // sgei
&JitShader::Compile_SLT, // slti
nullptr, // unknown
nullptr, // unknown
nullptr, // unknown
nullptr, // unknown
nullptr, // unknown
&JitShader::Compile_NOP, // nop
&JitShader::Compile_END, // end
nullptr, // break
&JitShader::Compile_CALL, // call
&JitShader::Compile_CALLC, // callc
&JitShader::Compile_CALLU, // callu
&JitShader::Compile_IF, // ifu
&JitShader::Compile_IF, // ifc
&JitShader::Compile_LOOP, // loop
&JitShader::Compile_EMIT, // emit
&JitShader::Compile_SETE, // sete
&JitShader::Compile_JMP, // jmpc
&JitShader::Compile_JMP, // jmpu
&JitShader::Compile_CMP, // cmp
&JitShader::Compile_CMP, // cmp
&JitShader::Compile_MAD, // madi
&JitShader::Compile_MAD, // madi
&JitShader::Compile_MAD, // madi
&JitShader::Compile_MAD, // madi
&JitShader::Compile_MAD, // madi
&JitShader::Compile_MAD, // madi
&JitShader::Compile_MAD, // madi
&JitShader::Compile_MAD, // madi
&JitShader::Compile_MAD, // mad
&JitShader::Compile_MAD, // mad
&JitShader::Compile_MAD, // mad
&JitShader::Compile_MAD, // mad
&JitShader::Compile_MAD, // mad
&JitShader::Compile_MAD, // mad
&JitShader::Compile_MAD, // mad
&JitShader::Compile_MAD, // mad
};
// The following is used to alias some commonly used registers. Generally, RAX-RDX and XMM0-XMM3 can
// be used as scratch registers within a compiler function. The other registers have designated
// purposes, as documented below:
/// Pointer to the uniform memory
static const Reg64 SETUP = r9;
/// The two 32-bit VS address offset registers set by the MOVA instruction
static const Reg64 ADDROFFS_REG_0 = r10;
static const Reg64 ADDROFFS_REG_1 = r11;
/// VS loop count register (Multiplied by 16)
static const Reg32 LOOPCOUNT_REG = r12d;
/// Current VS loop iteration number (we could probably use LOOPCOUNT_REG, but this quicker)
static const Reg32 LOOPCOUNT = esi;
/// Number to increment LOOPCOUNT_REG by on each loop iteration (Multiplied by 16)
static const Reg32 LOOPINC = edi;
/// Result of the previous CMP instruction for the X-component comparison
static const Reg64 COND0 = r13;
/// Result of the previous CMP instruction for the Y-component comparison
static const Reg64 COND1 = r14;
/// Pointer to the UnitState instance for the current VS unit
static const Reg64 STATE = r15;
/// SIMD scratch register
static const Xmm SCRATCH = xmm0;
/// Loaded with the first swizzled source register, otherwise can be used as a scratch register
static const Xmm SRC1 = xmm1;
/// Loaded with the second swizzled source register, otherwise can be used as a scratch register
static const Xmm SRC2 = xmm2;
/// Loaded with the third swizzled source register, otherwise can be used as a scratch register
static const Xmm SRC3 = xmm3;
/// Additional scratch register
static const Xmm SCRATCH2 = xmm4;
/// Constant vector of [1.0f, 1.0f, 1.0f, 1.0f], used to efficiently set a vector to one
static const Xmm ONE = xmm14;
/// Constant vector of [-0.f, -0.f, -0.f, -0.f], used to efficiently negate a vector with XOR
static const Xmm NEGBIT = xmm15;
// State registers that must not be modified by external functions calls
// Scratch registers, e.g., SRC1 and SCRATCH, have to be saved on the side if needed
static const BitSet32 persistent_regs = BuildRegSet({
// Pointers to register blocks
SETUP, STATE,
// Cached registers
ADDROFFS_REG_0, ADDROFFS_REG_1, LOOPCOUNT_REG, COND0, COND1,
// Constants
ONE, NEGBIT,
// Loop variables
LOOPCOUNT, LOOPINC,
});
/// Raw constant for the source register selector that indicates no swizzling is performed
static const u8 NO_SRC_REG_SWIZZLE = 0x1b;
/// Raw constant for the destination register enable mask that indicates all components are enabled
static const u8 NO_DEST_REG_MASK = 0xf;
static void LogCritical(const char* msg) {
LOG_CRITICAL(HW_GPU, "%s", msg);
}
void JitShader::Compile_Assert(bool condition, const char* msg) {
if (!condition) {
mov(ABI_PARAM1, reinterpret_cast<size_t>(msg));
CallFarFunction(*this, LogCritical);
}
}
/**
* Loads and swizzles a source register into the specified XMM register.
* @param instr VS instruction, used for determining how to load the source register
* @param src_num Number indicating which source register to load (1 = src1, 2 = src2, 3 = src3)
* @param src_reg SourceRegister object corresponding to the source register to load
* @param dest Destination XMM register to store the loaded, swizzled source register
*/
void JitShader::Compile_SwizzleSrc(Instruction instr, unsigned src_num, SourceRegister src_reg,
Xmm dest) {
Reg64 src_ptr;
size_t src_offset;
if (src_reg.GetRegisterType() == RegisterType::FloatUniform) {
src_ptr = SETUP;
src_offset = ShaderSetup::GetFloatUniformOffset(src_reg.GetIndex());
} else {
src_ptr = STATE;
src_offset = UnitState::InputOffset(src_reg);
}
int src_offset_disp = (int)src_offset;
ASSERT_MSG(src_offset == src_offset_disp, "Source register offset too large for int type");
unsigned operand_desc_id;
const bool is_inverted =
(0 != (instr.opcode.Value().GetInfo().subtype & OpCode::Info::SrcInversed));
unsigned address_register_index;
unsigned offset_src;
if (instr.opcode.Value().EffectiveOpCode() == OpCode::Id::MAD ||
instr.opcode.Value().EffectiveOpCode() == OpCode::Id::MADI) {
operand_desc_id = instr.mad.operand_desc_id;
offset_src = is_inverted ? 3 : 2;
address_register_index = instr.mad.address_register_index;
} else {
operand_desc_id = instr.common.operand_desc_id;
offset_src = is_inverted ? 2 : 1;
address_register_index = instr.common.address_register_index;
}
if (src_num == offset_src && address_register_index != 0) {
switch (address_register_index) {
case 1: // address offset 1
movaps(dest, xword[src_ptr + ADDROFFS_REG_0 + src_offset_disp]);
break;
case 2: // address offset 2
movaps(dest, xword[src_ptr + ADDROFFS_REG_1 + src_offset_disp]);
break;
case 3: // address offset 3
movaps(dest, xword[src_ptr + LOOPCOUNT_REG.cvt64() + src_offset_disp]);
break;
default:
UNREACHABLE();
break;
}
} else {
// Load the source
movaps(dest, xword[src_ptr + src_offset_disp]);
}
SwizzlePattern swiz = {(*swizzle_data)[operand_desc_id]};
// Generate instructions for source register swizzling as needed
u8 sel = swiz.GetRawSelector(src_num);
if (sel != NO_SRC_REG_SWIZZLE) {
// Selector component order needs to be reversed for the SHUFPS instruction
sel = ((sel & 0xc0) >> 6) | ((sel & 3) << 6) | ((sel & 0xc) << 2) | ((sel & 0x30) >> 2);
// Shuffle inputs for swizzle
shufps(dest, dest, sel);
}
// If the source register should be negated, flip the negative bit using XOR
const bool negate[] = {swiz.negate_src1, swiz.negate_src2, swiz.negate_src3};
if (negate[src_num - 1]) {
xorps(dest, NEGBIT);
}
}
void JitShader::Compile_DestEnable(Instruction instr, Xmm src) {
DestRegister dest;
unsigned operand_desc_id;
if (instr.opcode.Value().EffectiveOpCode() == OpCode::Id::MAD ||
instr.opcode.Value().EffectiveOpCode() == OpCode::Id::MADI) {
operand_desc_id = instr.mad.operand_desc_id;
dest = instr.mad.dest.Value();
} else {
operand_desc_id = instr.common.operand_desc_id;
dest = instr.common.dest.Value();
}
SwizzlePattern swiz = {(*swizzle_data)[operand_desc_id]};
size_t dest_offset_disp = UnitState::OutputOffset(dest);
// If all components are enabled, write the result to the destination register
if (swiz.dest_mask == NO_DEST_REG_MASK) {
// Store dest back to memory
movaps(xword[STATE + dest_offset_disp], src);
} else {
// Not all components are enabled, so mask the result when storing to the destination
// register...
movaps(SCRATCH, xword[STATE + dest_offset_disp]);
if (Common::GetCPUCaps().sse4_1) {
u8 mask = ((swiz.dest_mask & 1) << 3) | ((swiz.dest_mask & 8) >> 3) |
((swiz.dest_mask & 2) << 1) | ((swiz.dest_mask & 4) >> 1);
blendps(SCRATCH, src, mask);
} else {
movaps(SCRATCH2, src);
unpckhps(SCRATCH2, SCRATCH); // Unpack X/Y components of source and destination
unpcklps(SCRATCH, src); // Unpack Z/W components of source and destination
// Compute selector to selectively copy source components to destination for SHUFPS
// instruction
u8 sel = ((swiz.DestComponentEnabled(0) ? 1 : 0) << 0) |
((swiz.DestComponentEnabled(1) ? 3 : 2) << 2) |
((swiz.DestComponentEnabled(2) ? 0 : 1) << 4) |
((swiz.DestComponentEnabled(3) ? 2 : 3) << 6);
shufps(SCRATCH, SCRATCH2, sel);
}
// Store dest back to memory
movaps(xword[STATE + dest_offset_disp], SCRATCH);
}
}
void JitShader::Compile_SanitizedMul(Xmm src1, Xmm src2, Xmm scratch) {
// 0 * inf and inf * 0 in the PICA should return 0 instead of NaN. This can be implemented by
// checking for NaNs before and after the multiplication. If the multiplication result is NaN
// where neither source was, this NaN was generated by a 0 * inf multiplication, and so the
// result should be transformed to 0 to match PICA fp rules.
// Set scratch to mask of (src1 != NaN and src2 != NaN)
movaps(scratch, src1);
cmpordps(scratch, src2);
mulps(src1, src2);
// Set src2 to mask of (result == NaN)
movaps(src2, src1);
cmpunordps(src2, src2);
// Clear components where scratch != src2 (i.e. if result is NaN where neither source was NaN)
xorps(scratch, src2);
andps(src1, scratch);
}
void JitShader::Compile_EvaluateCondition(Instruction instr) {
// Note: NXOR is used below to check for equality
switch (instr.flow_control.op) {
case Instruction::FlowControlType::Or:
mov(eax, COND0);
mov(ebx, COND1);
xor_(eax, (instr.flow_control.refx.Value() ^ 1));
xor_(ebx, (instr.flow_control.refy.Value() ^ 1));
or_(eax, ebx);
break;
case Instruction::FlowControlType::And:
mov(eax, COND0);
mov(ebx, COND1);
xor_(eax, (instr.flow_control.refx.Value() ^ 1));
xor_(ebx, (instr.flow_control.refy.Value() ^ 1));
and_(eax, ebx);
break;
case Instruction::FlowControlType::JustX:
mov(eax, COND0);
xor_(eax, (instr.flow_control.refx.Value() ^ 1));
break;
case Instruction::FlowControlType::JustY:
mov(eax, COND1);
xor_(eax, (instr.flow_control.refy.Value() ^ 1));
break;
}
}
void JitShader::Compile_UniformCondition(Instruction instr) {
size_t offset = ShaderSetup::GetBoolUniformOffset(instr.flow_control.bool_uniform_id);
cmp(byte[SETUP + offset], 0);
}
BitSet32 JitShader::PersistentCallerSavedRegs() {
return persistent_regs & ABI_ALL_CALLER_SAVED;
}
void JitShader::Compile_ADD(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2);
addps(SRC1, SRC2);
Compile_DestEnable(instr, SRC1);
}
void JitShader::Compile_DP3(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2);
Compile_SanitizedMul(SRC1, SRC2, SCRATCH);
movaps(SRC2, SRC1);
shufps(SRC2, SRC2, _MM_SHUFFLE(1, 1, 1, 1));
movaps(SRC3, SRC1);
shufps(SRC3, SRC3, _MM_SHUFFLE(2, 2, 2, 2));
shufps(SRC1, SRC1, _MM_SHUFFLE(0, 0, 0, 0));
addps(SRC1, SRC2);
addps(SRC1, SRC3);
Compile_DestEnable(instr, SRC1);
}
void JitShader::Compile_DP4(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2);
Compile_SanitizedMul(SRC1, SRC2, SCRATCH);
movaps(SRC2, SRC1);
shufps(SRC1, SRC1, _MM_SHUFFLE(2, 3, 0, 1)); // XYZW -> ZWXY
addps(SRC1, SRC2);
movaps(SRC2, SRC1);
shufps(SRC1, SRC1, _MM_SHUFFLE(0, 1, 2, 3)); // XYZW -> WZYX
addps(SRC1, SRC2);
Compile_DestEnable(instr, SRC1);
}
void JitShader::Compile_DPH(Instruction instr) {
if (instr.opcode.Value().EffectiveOpCode() == OpCode::Id::DPHI) {
Compile_SwizzleSrc(instr, 1, instr.common.src1i, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2i, SRC2);
} else {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2);
}
if (Common::GetCPUCaps().sse4_1) {
// Set 4th component to 1.0
blendps(SRC1, ONE, 0b1000);
} else {
// Set 4th component to 1.0
movaps(SCRATCH, SRC1);
unpckhps(SCRATCH, ONE); // XYZW, 1111 -> Z1__
unpcklpd(SRC1, SCRATCH); // XYZW, Z1__ -> XYZ1
}
Compile_SanitizedMul(SRC1, SRC2, SCRATCH);
movaps(SRC2, SRC1);
shufps(SRC1, SRC1, _MM_SHUFFLE(2, 3, 0, 1)); // XYZW -> ZWXY
addps(SRC1, SRC2);
movaps(SRC2, SRC1);
shufps(SRC1, SRC1, _MM_SHUFFLE(0, 1, 2, 3)); // XYZW -> WZYX
addps(SRC1, SRC2);
Compile_DestEnable(instr, SRC1);
}
void JitShader::Compile_EX2(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
movss(xmm0, SRC1); // ABI_PARAM1
ABI_PushRegistersAndAdjustStack(*this, PersistentCallerSavedRegs(), 0);
CallFarFunction(*this, exp2f);
ABI_PopRegistersAndAdjustStack(*this, PersistentCallerSavedRegs(), 0);
shufps(xmm0, xmm0, _MM_SHUFFLE(0, 0, 0, 0)); // ABI_RETURN
movaps(SRC1, xmm0);
Compile_DestEnable(instr, SRC1);
}
void JitShader::Compile_LG2(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
movss(xmm0, SRC1); // ABI_PARAM1
ABI_PushRegistersAndAdjustStack(*this, PersistentCallerSavedRegs(), 0);
CallFarFunction(*this, log2f);
ABI_PopRegistersAndAdjustStack(*this, PersistentCallerSavedRegs(), 0);
shufps(xmm0, xmm0, _MM_SHUFFLE(0, 0, 0, 0)); // ABI_RETURN
movaps(SRC1, xmm0);
Compile_DestEnable(instr, SRC1);
}
void JitShader::Compile_MUL(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2);
Compile_SanitizedMul(SRC1, SRC2, SCRATCH);
Compile_DestEnable(instr, SRC1);
}
void JitShader::Compile_SGE(Instruction instr) {
if (instr.opcode.Value().EffectiveOpCode() == OpCode::Id::SGEI) {
Compile_SwizzleSrc(instr, 1, instr.common.src1i, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2i, SRC2);
} else {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2);
}
cmpleps(SRC2, SRC1);
andps(SRC2, ONE);
Compile_DestEnable(instr, SRC2);
}
void JitShader::Compile_SLT(Instruction instr) {
if (instr.opcode.Value().EffectiveOpCode() == OpCode::Id::SLTI) {
Compile_SwizzleSrc(instr, 1, instr.common.src1i, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2i, SRC2);
} else {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2);
}
cmpltps(SRC1, SRC2);
andps(SRC1, ONE);
Compile_DestEnable(instr, SRC1);
}
void JitShader::Compile_FLR(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
if (Common::GetCPUCaps().sse4_1) {
roundps(SRC1, SRC1, _MM_FROUND_FLOOR);
} else {
cvttps2dq(SRC1, SRC1);
cvtdq2ps(SRC1, SRC1);
}
Compile_DestEnable(instr, SRC1);
}
void JitShader::Compile_MAX(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2);
// SSE semantics match PICA200 ones: In case of NaN, SRC2 is returned.
maxps(SRC1, SRC2);
Compile_DestEnable(instr, SRC1);
}
void JitShader::Compile_MIN(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2);
// SSE semantics match PICA200 ones: In case of NaN, SRC2 is returned.
minps(SRC1, SRC2);
Compile_DestEnable(instr, SRC1);
}
void JitShader::Compile_MOVA(Instruction instr) {
SwizzlePattern swiz = {(*swizzle_data)[instr.common.operand_desc_id]};
if (!swiz.DestComponentEnabled(0) && !swiz.DestComponentEnabled(1)) {
return; // NoOp
}
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
// Convert floats to integers using truncation (only care about X and Y components)
cvttps2dq(SRC1, SRC1);
// Get result
movq(rax, SRC1);
// Handle destination enable
if (swiz.DestComponentEnabled(0) && swiz.DestComponentEnabled(1)) {
// Move and sign-extend low 32 bits
movsxd(ADDROFFS_REG_0, eax);
// Move and sign-extend high 32 bits
shr(rax, 32);
movsxd(ADDROFFS_REG_1, eax);
// Multiply by 16 to be used as an offset later
shl(ADDROFFS_REG_0, 4);
shl(ADDROFFS_REG_1, 4);
} else {
if (swiz.DestComponentEnabled(0)) {
// Move and sign-extend low 32 bits
movsxd(ADDROFFS_REG_0, eax);
// Multiply by 16 to be used as an offset later
shl(ADDROFFS_REG_0, 4);
} else if (swiz.DestComponentEnabled(1)) {
// Move and sign-extend high 32 bits
shr(rax, 32);
movsxd(ADDROFFS_REG_1, eax);
// Multiply by 16 to be used as an offset later
shl(ADDROFFS_REG_1, 4);
}
}
}
void JitShader::Compile_MOV(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
Compile_DestEnable(instr, SRC1);
}
void JitShader::Compile_RCP(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
// TODO(bunnei): RCPSS is a pretty rough approximation, this might cause problems if Pica
// performs this operation more accurately. This should be checked on hardware.
rcpss(SRC1, SRC1);
shufps(SRC1, SRC1, _MM_SHUFFLE(0, 0, 0, 0)); // XYWZ -> XXXX
Compile_DestEnable(instr, SRC1);
}
void JitShader::Compile_RSQ(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
// TODO(bunnei): RSQRTSS is a pretty rough approximation, this might cause problems if Pica
// performs this operation more accurately. This should be checked on hardware.
rsqrtss(SRC1, SRC1);
shufps(SRC1, SRC1, _MM_SHUFFLE(0, 0, 0, 0)); // XYWZ -> XXXX
Compile_DestEnable(instr, SRC1);
}
void JitShader::Compile_NOP(Instruction instr) {}
void JitShader::Compile_END(Instruction instr) {
ABI_PopRegistersAndAdjustStack(*this, ABI_ALL_CALLEE_SAVED, 8, 16);
ret();
}
void JitShader::Compile_CALL(Instruction instr) {
// Push offset of the return
push(qword, (instr.flow_control.dest_offset + instr.flow_control.num_instructions));
// Call the subroutine
call(instruction_labels[instr.flow_control.dest_offset]);
// Skip over the return offset that's on the stack
add(rsp, 8);
}
void JitShader::Compile_CALLC(Instruction instr) {
Compile_EvaluateCondition(instr);
Label b;
jz(b);
Compile_CALL(instr);
L(b);
}
void JitShader::Compile_CALLU(Instruction instr) {
Compile_UniformCondition(instr);
Label b;
jz(b);
Compile_CALL(instr);
L(b);
}
void JitShader::Compile_CMP(Instruction instr) {
using Op = Instruction::Common::CompareOpType::Op;
Op op_x = instr.common.compare_op.x;
Op op_y = instr.common.compare_op.y;
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2);
// SSE doesn't have greater-than (GT) or greater-equal (GE) comparison operators. You need to
// emulate them by swapping the lhs and rhs and using LT and LE. NLT and NLE can't be used here
// because they don't match when used with NaNs.
static const u8 cmp[] = {CMP_EQ, CMP_NEQ, CMP_LT, CMP_LE, CMP_LT, CMP_LE};
bool invert_op_x = (op_x == Op::GreaterThan || op_x == Op::GreaterEqual);
Xmm lhs_x = invert_op_x ? SRC2 : SRC1;
Xmm rhs_x = invert_op_x ? SRC1 : SRC2;
if (op_x == op_y) {
// Compare X-component and Y-component together
cmpps(lhs_x, rhs_x, cmp[op_x]);
movq(COND0, lhs_x);
mov(COND1, COND0);
} else {
bool invert_op_y = (op_y == Op::GreaterThan || op_y == Op::GreaterEqual);
Xmm lhs_y = invert_op_y ? SRC2 : SRC1;
Xmm rhs_y = invert_op_y ? SRC1 : SRC2;
// Compare X-component
movaps(SCRATCH, lhs_x);
cmpss(SCRATCH, rhs_x, cmp[op_x]);
// Compare Y-component
cmpps(lhs_y, rhs_y, cmp[op_y]);
movq(COND0, SCRATCH);
movq(COND1, lhs_y);
}
shr(COND0.cvt32(), 31); // ignores upper 32 bits in source
shr(COND1, 63);
}
void JitShader::Compile_MAD(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.mad.src1, SRC1);
if (instr.opcode.Value().EffectiveOpCode() == OpCode::Id::MADI) {
Compile_SwizzleSrc(instr, 2, instr.mad.src2i, SRC2);
Compile_SwizzleSrc(instr, 3, instr.mad.src3i, SRC3);
} else {
Compile_SwizzleSrc(instr, 2, instr.mad.src2, SRC2);
Compile_SwizzleSrc(instr, 3, instr.mad.src3, SRC3);
}
Compile_SanitizedMul(SRC1, SRC2, SCRATCH);
addps(SRC1, SRC3);
Compile_DestEnable(instr, SRC1);
}
void JitShader::Compile_IF(Instruction instr) {
Compile_Assert(instr.flow_control.dest_offset >= program_counter,
"Backwards if-statements not supported");
Label l_else, l_endif;
// Evaluate the "IF" condition
if (instr.opcode.Value() == OpCode::Id::IFU) {
Compile_UniformCondition(instr);
} else if (instr.opcode.Value() == OpCode::Id::IFC) {
Compile_EvaluateCondition(instr);
}
jz(l_else, T_NEAR);
// Compile the code that corresponds to the condition evaluating as true
Compile_Block(instr.flow_control.dest_offset);
// If there isn't an "ELSE" condition, we are done here
if (instr.flow_control.num_instructions == 0) {
L(l_else);
return;
}
jmp(l_endif, T_NEAR);
L(l_else);
// This code corresponds to the "ELSE" condition
// Comple the code that corresponds to the condition evaluating as false
Compile_Block(instr.flow_control.dest_offset + instr.flow_control.num_instructions);
L(l_endif);
}
void JitShader::Compile_LOOP(Instruction instr) {
Compile_Assert(instr.flow_control.dest_offset >= program_counter,
"Backwards loops not supported");
Compile_Assert(!looping, "Nested loops not supported");
looping = true;
// This decodes the fields from the integer uniform at index instr.flow_control.int_uniform_id.
// The Y (LOOPCOUNT_REG) and Z (LOOPINC) component are kept multiplied by 16 (Left shifted by
// 4 bits) to be used as an offset into the 16-byte vector registers later
size_t offset = ShaderSetup::GetIntUniformOffset(instr.flow_control.int_uniform_id);
mov(LOOPCOUNT, dword[SETUP + offset]);
mov(LOOPCOUNT_REG, LOOPCOUNT);
shr(LOOPCOUNT_REG, 4);
and_(LOOPCOUNT_REG, 0xFF0); // Y-component is the start
mov(LOOPINC, LOOPCOUNT);
shr(LOOPINC, 12);
and_(LOOPINC, 0xFF0); // Z-component is the incrementer
movzx(LOOPCOUNT, LOOPCOUNT.cvt8()); // X-component is iteration count
add(LOOPCOUNT, 1); // Iteration count is X-component + 1
Label l_loop_start;
L(l_loop_start);
Compile_Block(instr.flow_control.dest_offset + 1);
add(LOOPCOUNT_REG, LOOPINC); // Increment LOOPCOUNT_REG by Z-component
sub(LOOPCOUNT, 1); // Increment loop count by 1
jnz(l_loop_start); // Loop if not equal
looping = false;
}
void JitShader::Compile_JMP(Instruction instr) {
if (instr.opcode.Value() == OpCode::Id::JMPC)
Compile_EvaluateCondition(instr);
else if (instr.opcode.Value() == OpCode::Id::JMPU)
Compile_UniformCondition(instr);
else
UNREACHABLE();
bool inverted_condition =
(instr.opcode.Value() == OpCode::Id::JMPU) && (instr.flow_control.num_instructions & 1);
Label& b = instruction_labels[instr.flow_control.dest_offset];
if (inverted_condition) {
jz(b, T_NEAR);
} else {
jnz(b, T_NEAR);
}
}
static void Emit(GSEmitter* emitter, Math::Vec4<float24> (*output)[16]) {
emitter->Emit(*output);
}
void JitShader::Compile_EMIT(Instruction instr) {
Label have_emitter, end;
mov(rax, qword[STATE + offsetof(UnitState, emitter_ptr)]);
test(rax, rax);
jnz(have_emitter);
ABI_PushRegistersAndAdjustStack(*this, PersistentCallerSavedRegs(), 0);
mov(ABI_PARAM1, reinterpret_cast<size_t>("Execute EMIT on VS"));
CallFarFunction(*this, LogCritical);
ABI_PopRegistersAndAdjustStack(*this, PersistentCallerSavedRegs(), 0);
jmp(end);
L(have_emitter);
ABI_PushRegistersAndAdjustStack(*this, PersistentCallerSavedRegs(), 0);
mov(ABI_PARAM1, rax);
mov(ABI_PARAM2, STATE);
add(ABI_PARAM2, static_cast<Xbyak::uint32>(offsetof(UnitState, registers.output)));
CallFarFunction(*this, Emit);
ABI_PopRegistersAndAdjustStack(*this, PersistentCallerSavedRegs(), 0);
L(end);
}
void JitShader::Compile_SETE(Instruction instr) {
Label have_emitter, end;
mov(rax, qword[STATE + offsetof(UnitState, emitter_ptr)]);
test(rax, rax);
jnz(have_emitter);
ABI_PushRegistersAndAdjustStack(*this, PersistentCallerSavedRegs(), 0);
mov(ABI_PARAM1, reinterpret_cast<size_t>("Execute SETEMIT on VS"));
CallFarFunction(*this, LogCritical);
ABI_PopRegistersAndAdjustStack(*this, PersistentCallerSavedRegs(), 0);
jmp(end);
L(have_emitter);
mov(byte[rax + offsetof(GSEmitter, vertex_id)], instr.setemit.vertex_id);
mov(byte[rax + offsetof(GSEmitter, prim_emit)], instr.setemit.prim_emit);
mov(byte[rax + offsetof(GSEmitter, winding)], instr.setemit.winding);
L(end);
}
void JitShader::Compile_Block(unsigned end) {
while (program_counter < end) {
Compile_NextInstr();
}
}
void JitShader::Compile_Return() {
// Peek return offset on the stack and check if we're at that offset
mov(rax, qword[rsp + 8]);
cmp(eax, (program_counter));
// If so, jump back to before CALL
Label b;
jnz(b);
ret();
L(b);
}
void JitShader::Compile_NextInstr() {
if (std::binary_search(return_offsets.begin(), return_offsets.end(), program_counter)) {
Compile_Return();
}
L(instruction_labels[program_counter]);
Instruction instr = {(*program_code)[program_counter++]};
OpCode::Id opcode = instr.opcode.Value();
auto instr_func = instr_table[static_cast<unsigned>(opcode)];
if (instr_func) {
// JIT the instruction!
((*this).*instr_func)(instr);
} else {
// Unhandled instruction
LOG_CRITICAL(HW_GPU, "Unhandled instruction: 0x%02x (0x%08x)",
instr.opcode.Value().EffectiveOpCode(), instr.hex);
}
}
void JitShader::FindReturnOffsets() {
return_offsets.clear();
for (size_t offset = 0; offset < program_code->size(); ++offset) {
Instruction instr = {(*program_code)[offset]};
switch (instr.opcode.Value()) {
case OpCode::Id::CALL:
case OpCode::Id::CALLC:
case OpCode::Id::CALLU:
return_offsets.push_back(instr.flow_control.dest_offset +
instr.flow_control.num_instructions);
break;
default:
break;
}
}
// Sort for efficient binary search later
std::sort(return_offsets.begin(), return_offsets.end());
}
void JitShader::Compile(const std::array<u32, MAX_PROGRAM_CODE_LENGTH>* program_code_,
const std::array<u32, MAX_SWIZZLE_DATA_LENGTH>* swizzle_data_) {
program_code = program_code_;
swizzle_data = swizzle_data_;
// Reset flow control state
program = (CompiledShader*)getCurr();
program_counter = 0;
looping = false;
instruction_labels.fill(Xbyak::Label());
// Find all `CALL` instructions and identify return locations
FindReturnOffsets();
// The stack pointer is 8 modulo 16 at the entry of a procedure
// We reserve 16 bytes and assign a dummy value to the first 8 bytes, to catch any potential
// return checks (see Compile_Return) that happen in shader main routine.
ABI_PushRegistersAndAdjustStack(*this, ABI_ALL_CALLEE_SAVED, 8, 16);
mov(qword[rsp + 8], 0xFFFFFFFFFFFFFFFFULL);
mov(SETUP, ABI_PARAM1);
mov(STATE, ABI_PARAM2);
// Zero address/loop registers
xor_(ADDROFFS_REG_0.cvt32(), ADDROFFS_REG_0.cvt32());
xor_(ADDROFFS_REG_1.cvt32(), ADDROFFS_REG_1.cvt32());
xor_(LOOPCOUNT_REG, LOOPCOUNT_REG);
// Used to set a register to one
static const __m128 one = {1.f, 1.f, 1.f, 1.f};
mov(rax, reinterpret_cast<size_t>(&one));
movaps(ONE, xword[rax]);
// Used to negate registers
static const __m128 neg = {-0.f, -0.f, -0.f, -0.f};
mov(rax, reinterpret_cast<size_t>(&neg));
movaps(NEGBIT, xword[rax]);
// Jump to start of the shader program
jmp(ABI_PARAM3);
// Compile entire program
Compile_Block(static_cast<unsigned>(program_code->size()));
// Free memory that's no longer needed
program_code = nullptr;
swizzle_data = nullptr;
return_offsets.clear();
return_offsets.shrink_to_fit();
ready();
ASSERT_MSG(getSize() <= MAX_SHADER_SIZE, "Compiled a shader that exceeds the allocated size!");
LOG_DEBUG(HW_GPU, "Compiled shader size=%lu", getSize());
}
JitShader::JitShader() : Xbyak::CodeGenerator(MAX_SHADER_SIZE) {}
} // namespace Shader
} // namespace Pica

127
src/video_core/shader/shader_jit_x64_compiler.h
View file

@ -1,127 +0,0 @@
// Copyright 2015 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <array>
#include <cstddef>
#include <utility>
#include <vector>
#include <nihstro/shader_bytecode.h>
#include <xbyak.h>
#include "common/bit_set.h"
#include "common/common_types.h"
#include "video_core/shader/shader.h"
using nihstro::Instruction;
using nihstro::OpCode;
using nihstro::SwizzlePattern;
namespace Pica {
namespace Shader {
/// Memory allocated for each compiled shader
constexpr size_t MAX_SHADER_SIZE = MAX_PROGRAM_CODE_LENGTH * 64;
/**
* This class implements the shader JIT compiler. It recompiles a Pica shader program into x86_64
* code that can be executed on the host machine directly.
*/
class JitShader : public Xbyak::CodeGenerator {
public:
JitShader();
void Run(const ShaderSetup& setup, UnitState& state, unsigned offset) const {
program(&setup, &state, instruction_labels[offset].getAddress());
}
void Compile(const std::array<u32, MAX_PROGRAM_CODE_LENGTH>* program_code,
const std::array<u32, MAX_SWIZZLE_DATA_LENGTH>* swizzle_data);
void Compile_ADD(Instruction instr);
void Compile_DP3(Instruction instr);
void Compile_DP4(Instruction instr);
void Compile_DPH(Instruction instr);
void Compile_EX2(Instruction instr);
void Compile_LG2(Instruction instr);
void Compile_MUL(Instruction instr);
void Compile_SGE(Instruction instr);
void Compile_SLT(Instruction instr);
void Compile_FLR(Instruction instr);
void Compile_MAX(Instruction instr);
void Compile_MIN(Instruction instr);
void Compile_RCP(Instruction instr);
void Compile_RSQ(Instruction instr);
void Compile_MOVA(Instruction instr);
void Compile_MOV(Instruction instr);
void Compile_NOP(Instruction instr);
void Compile_END(Instruction instr);
void Compile_CALL(Instruction instr);
void Compile_CALLC(Instruction instr);
void Compile_CALLU(Instruction instr);
void Compile_IF(Instruction instr);
void Compile_LOOP(Instruction instr);
void Compile_JMP(Instruction instr);
void Compile_CMP(Instruction instr);
void Compile_MAD(Instruction instr);
void Compile_EMIT(Instruction instr);
void Compile_SETE(Instruction instr);
private:
void Compile_Block(unsigned end);
void Compile_NextInstr();
void Compile_SwizzleSrc(Instruction instr, unsigned src_num, SourceRegister src_reg,
Xbyak::Xmm dest);
void Compile_DestEnable(Instruction instr, Xbyak::Xmm dest);
/**
* Compiles a `MUL src1, src2` operation, properly handling the PICA semantics when multiplying
* zero by inf. Clobbers `src2` and `scratch`.
*/
void Compile_SanitizedMul(Xbyak::Xmm src1, Xbyak::Xmm src2, Xbyak::Xmm scratch);
void Compile_EvaluateCondition(Instruction instr);
void Compile_UniformCondition(Instruction instr);
/**
* Emits the code to conditionally return from a subroutine envoked by the `CALL` instruction.
*/
void Compile_Return();
BitSet32 PersistentCallerSavedRegs();
/**
* Assertion evaluated at compile-time, but only triggered if executed at runtime.
* @param condition Condition to be evaluated.
* @param msg Message to be logged if the assertion fails.
*/
void Compile_Assert(bool condition, const char* msg);
/**
* Analyzes the entire shader program for `CALL` instructions before emitting any code,
* identifying the locations where a return needs to be inserted.
*/
void FindReturnOffsets();
const std::array<u32, MAX_PROGRAM_CODE_LENGTH>* program_code = nullptr;
const std::array<u32, MAX_SWIZZLE_DATA_LENGTH>* swizzle_data = nullptr;
/// Mapping of Pica VS instructions to pointers in the emitted code
std::array<Xbyak::Label, MAX_PROGRAM_CODE_LENGTH> instruction_labels;
/// Offsets in code where a return needs to be inserted
std::vector<unsigned> return_offsets;
unsigned program_counter = 0; ///< Offset of the next instruction to decode
bool looping = false; ///< True if compiling a loop, used to check for nested loops
using CompiledShader = void(const void* setup, void* state, const u8* start_addr);
CompiledShader* program = nullptr;
};
} // Shader
} // Pica

197
src/video_core/swrasterizer/clipper.cpp
View file

@ -1,197 +0,0 @@
// Copyright 2014 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <algorithm>
#include <array>
#include <cstddef>
#include <boost/container/static_vector.hpp>
#include <boost/container/vector.hpp>
#include "common/bit_field.h"
#include "common/common_types.h"
#include "common/logging/log.h"
#include "common/vector_math.h"
#include "video_core/pica_state.h"
#include "video_core/pica_types.h"
#include "video_core/shader/shader.h"
#include "video_core/swrasterizer/clipper.h"
#include "video_core/swrasterizer/rasterizer.h"
using Pica::Rasterizer::Vertex;
namespace Pica {
namespace Clipper {
struct ClippingEdge {
public:
ClippingEdge(Math::Vec4<float24> coeffs, Math::Vec4<float24> bias = Math::Vec4<float24>(
float24::FromFloat32(0), float24::FromFloat32(0),
float24::FromFloat32(0), float24::FromFloat32(0)))
: coeffs(coeffs), bias(bias) {}
bool IsInside(const Vertex& vertex) const {
return Math::Dot(vertex.pos + bias, coeffs) >= float24::FromFloat32(0);
}
bool IsOutSide(const Vertex& vertex) const {
return !IsInside(vertex);
}
Vertex GetIntersection(const Vertex& v0, const Vertex& v1) const {
float24 dp = Math::Dot(v0.pos + bias, coeffs);
float24 dp_prev = Math::Dot(v1.pos + bias, coeffs);
float24 factor = dp_prev / (dp_prev - dp);
return Vertex::Lerp(factor, v0, v1);
}
private:
float24 pos;
Math::Vec4<float24> coeffs;
Math::Vec4<float24> bias;
};
static void InitScreenCoordinates(Vertex& vtx) {
struct {
float24 halfsize_x;
float24 offset_x;
float24 halfsize_y;
float24 offset_y;
float24 zscale;
float24 offset_z;
} viewport;
const auto& regs = g_state.regs;
viewport.halfsize_x = float24::FromRaw(regs.rasterizer.viewport_size_x);
viewport.halfsize_y = float24::FromRaw(regs.rasterizer.viewport_size_y);
viewport.offset_x = float24::FromFloat32(static_cast<float>(regs.rasterizer.viewport_corner.x));
viewport.offset_y = float24::FromFloat32(static_cast<float>(regs.rasterizer.viewport_corner.y));
float24 inv_w = float24::FromFloat32(1.f) / vtx.pos.w;
vtx.pos.w = inv_w;
vtx.quat *= inv_w;
vtx.color *= inv_w;
vtx.tc0 *= inv_w;
vtx.tc1 *= inv_w;
vtx.tc0_w *= inv_w;
vtx.view *= inv_w;
vtx.tc2 *= inv_w;
vtx.screenpos[0] =
(vtx.pos.x * inv_w + float24::FromFloat32(1.0)) * viewport.halfsize_x + viewport.offset_x;
vtx.screenpos[1] =
(vtx.pos.y * inv_w + float24::FromFloat32(1.0)) * viewport.halfsize_y + viewport.offset_y;
vtx.screenpos[2] = vtx.pos.z * inv_w;
}
void ProcessTriangle(const OutputVertex& v0, const OutputVertex& v1, const OutputVertex& v2) {
using boost::container::static_vector;
// Clipping a planar n-gon against a plane will remove at least 1 vertex and introduces 2 at
// the new edge (or less in degenerate cases). As such, we can say that each clipping plane
// introduces at most 1 new vertex to the polygon. Since we start with a triangle and have a
// fixed 6 clipping planes, the maximum number of vertices of the clipped polygon is 3 + 6 = 9.
static const size_t MAX_VERTICES = 9;
static_vector<Vertex, MAX_VERTICES> buffer_a = {v0, v1, v2};
static_vector<Vertex, MAX_VERTICES> buffer_b;
auto FlipQuaternionIfOpposite = [](auto& a, const auto& b) {
if (Math::Dot(a, b) < float24::Zero())
a = a * float24::FromFloat32(-1.0f);
};
// Flip the quaternions if they are opposite to prevent interpolating them over the wrong
// direction.
FlipQuaternionIfOpposite(buffer_a[1].quat, buffer_a[0].quat);
FlipQuaternionIfOpposite(buffer_a[2].quat, buffer_a[0].quat);
auto* output_list = &buffer_a;
auto* input_list = &buffer_b;
// NOTE: We clip against a w=epsilon plane to guarantee that the output has a positive w value.
// TODO: Not sure if this is a valid approach. Also should probably instead use the smallest
// epsilon possible within float24 accuracy.
static const float24 EPSILON = float24::FromFloat32(0.00001f);
static const float24 f0 = float24::FromFloat32(0.0);
static const float24 f1 = float24::FromFloat32(1.0);
static const std::array<ClippingEdge, 7> clipping_edges = {{
{Math::MakeVec(-f1, f0, f0, f1)}, // x = +w
{Math::MakeVec(f1, f0, f0, f1)}, // x = -w
{Math::MakeVec(f0, -f1, f0, f1)}, // y = +w
{Math::MakeVec(f0, f1, f0, f1)}, // y = -w
{Math::MakeVec(f0, f0, -f1, f0)}, // z = 0
{Math::MakeVec(f0, f0, f1, f1)}, // z = -w
{Math::MakeVec(f0, f0, f0, f1), Math::Vec4<float24>(f0, f0, f0, EPSILON)}, // w = EPSILON
}};
// Simple implementation of the Sutherland-Hodgman clipping algorithm.
// TODO: Make this less inefficient (currently lots of useless buffering overhead happens here)
auto Clip = [&](const ClippingEdge& edge) {
std::swap(input_list, output_list);
output_list->clear();
const Vertex* reference_vertex = &input_list->back();
for (const auto& vertex : *input_list) {
// NOTE: This algorithm changes vertex order in some cases!
if (edge.IsInside(vertex)) {
if (edge.IsOutSide(*reference_vertex)) {
output_list->push_back(edge.GetIntersection(vertex, *reference_vertex));
}
output_list->push_back(vertex);
} else if (edge.IsInside(*reference_vertex)) {
output_list->push_back(edge.GetIntersection(vertex, *reference_vertex));
}
reference_vertex = &vertex;
}
};
for (auto edge : clipping_edges) {
Clip(edge);
// Need to have at least a full triangle to continue...
if (output_list->size() < 3)
return;
}
if (g_state.regs.rasterizer.clip_enable) {
ClippingEdge custom_edge{g_state.regs.rasterizer.GetClipCoef()};
Clip(custom_edge);
if (output_list->size() < 3)
return;
}
InitScreenCoordinates((*output_list)[0]);
InitScreenCoordinates((*output_list)[1]);
for (size_t i = 0; i < output_list->size() - 2; i++) {
Vertex& vtx0 = (*output_list)[0];
Vertex& vtx1 = (*output_list)[i + 1];
Vertex& vtx2 = (*output_list)[i + 2];
InitScreenCoordinates(vtx2);
LOG_TRACE(Render_Software,
"Triangle %lu/%lu at position (%.3f, %.3f, %.3f, %.3f), "
"(%.3f, %.3f, %.3f, %.3f), (%.3f, %.3f, %.3f, %.3f) and "
"screen position (%.2f, %.2f, %.2f), (%.2f, %.2f, %.2f), (%.2f, %.2f, %.2f)",
i + 1, output_list->size() - 2, vtx0.pos.x.ToFloat32(), vtx0.pos.y.ToFloat32(),
vtx0.pos.z.ToFloat32(), vtx0.pos.w.ToFloat32(), vtx1.pos.x.ToFloat32(),
vtx1.pos.y.ToFloat32(), vtx1.pos.z.ToFloat32(), vtx1.pos.w.ToFloat32(),
vtx2.pos.x.ToFloat32(), vtx2.pos.y.ToFloat32(), vtx2.pos.z.ToFloat32(),
vtx2.pos.w.ToFloat32(), vtx0.screenpos.x.ToFloat32(),
vtx0.screenpos.y.ToFloat32(), vtx0.screenpos.z.ToFloat32(),
vtx1.screenpos.x.ToFloat32(), vtx1.screenpos.y.ToFloat32(),
vtx1.screenpos.z.ToFloat32(), vtx2.screenpos.x.ToFloat32(),
vtx2.screenpos.y.ToFloat32(), vtx2.screenpos.z.ToFloat32());
Rasterizer::ProcessTriangle(vtx0, vtx1, vtx2);
}
}
} // namespace
} // namespace

21
src/video_core/swrasterizer/clipper.h
View file

@ -1,21 +0,0 @@
// Copyright 2014 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
namespace Pica {
namespace Shader {
struct OutputVertex;
}
namespace Clipper {
using Shader::OutputVertex;
void ProcessTriangle(const OutputVertex& v0, const OutputVertex& v1, const OutputVertex& v2);
} // namespace
} // namespace

360
src/video_core/swrasterizer/framebuffer.cpp
View file

@ -1,360 +0,0 @@
// Copyright 2017 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <algorithm>
#include "common/assert.h"
#include "common/color.h"
#include "common/common_types.h"
#include "common/logging/log.h"
#include "common/math_util.h"
#include "common/vector_math.h"
#include "core/hw/gpu.h"
#include "core/memory.h"
#include "video_core/pica_state.h"
#include "video_core/regs_framebuffer.h"
#include "video_core/swrasterizer/framebuffer.h"
#include "video_core/utils.h"
namespace Pica {
namespace Rasterizer {
void DrawPixel(int x, int y, const Math::Vec4<u8>& color) {
const auto& framebuffer = g_state.regs.framebuffer.framebuffer;
const PAddr addr = framebuffer.GetColorBufferPhysicalAddress();
// Similarly to textures, the render framebuffer is laid out from bottom to top, too.
// NOTE: The framebuffer height register contains the actual FB height minus one.
y = framebuffer.height - y;
const u32 coarse_y = y & ~7;
u32 bytes_per_pixel =
GPU::Regs::BytesPerPixel(GPU::Regs::PixelFormat(framebuffer.color_format.Value()));
u32 dst_offset = VideoCore::GetMortonOffset(x, y, bytes_per_pixel) +
coarse_y * framebuffer.width * bytes_per_pixel;
u8* dst_pixel = Memory::GetPhysicalPointer(addr) + dst_offset;
switch (framebuffer.color_format) {
case FramebufferRegs::ColorFormat::RGBA8:
Color::EncodeRGBA8(color, dst_pixel);
break;
case FramebufferRegs::ColorFormat::RGB8:
Color::EncodeRGB8(color, dst_pixel);
break;
case FramebufferRegs::ColorFormat::RGB5A1:
Color::EncodeRGB5A1(color, dst_pixel);
break;
case FramebufferRegs::ColorFormat::RGB565:
Color::EncodeRGB565(color, dst_pixel);
break;
case FramebufferRegs::ColorFormat::RGBA4:
Color::EncodeRGBA4(color, dst_pixel);
break;
default:
LOG_CRITICAL(Render_Software, "Unknown framebuffer color format %x",
framebuffer.color_format.Value());
UNIMPLEMENTED();
}
}
const Math::Vec4<u8> GetPixel(int x, int y) {
const auto& framebuffer = g_state.regs.framebuffer.framebuffer;
const PAddr addr = framebuffer.GetColorBufferPhysicalAddress();
y = framebuffer.height - y;
const u32 coarse_y = y & ~7;
u32 bytes_per_pixel =
GPU::Regs::BytesPerPixel(GPU::Regs::PixelFormat(framebuffer.color_format.Value()));
u32 src_offset = VideoCore::GetMortonOffset(x, y, bytes_per_pixel) +
coarse_y * framebuffer.width * bytes_per_pixel;
u8* src_pixel = Memory::GetPhysicalPointer(addr) + src_offset;
switch (framebuffer.color_format) {
case FramebufferRegs::ColorFormat::RGBA8:
return Color::DecodeRGBA8(src_pixel);
case FramebufferRegs::ColorFormat::RGB8:
return Color::DecodeRGB8(src_pixel);
case FramebufferRegs::ColorFormat::RGB5A1:
return Color::DecodeRGB5A1(src_pixel);
case FramebufferRegs::ColorFormat::RGB565:
return Color::DecodeRGB565(src_pixel);
case FramebufferRegs::ColorFormat::RGBA4:
return Color::DecodeRGBA4(src_pixel);
default:
LOG_CRITICAL(Render_Software, "Unknown framebuffer color format %x",
framebuffer.color_format.Value());
UNIMPLEMENTED();
}
return {0, 0, 0, 0};
}
u32 GetDepth(int x, int y) {
const auto& framebuffer = g_state.regs.framebuffer.framebuffer;
const PAddr addr = framebuffer.GetDepthBufferPhysicalAddress();
u8* depth_buffer = Memory::GetPhysicalPointer(addr);
y = framebuffer.height - y;
const u32 coarse_y = y & ~7;
u32 bytes_per_pixel = FramebufferRegs::BytesPerDepthPixel(framebuffer.depth_format);
u32 stride = framebuffer.width * bytes_per_pixel;
u32 src_offset = VideoCore::GetMortonOffset(x, y, bytes_per_pixel) + coarse_y * stride;
u8* src_pixel = depth_buffer + src_offset;
switch (framebuffer.depth_format) {
case FramebufferRegs::DepthFormat::D16:
return Color::DecodeD16(src_pixel);
case FramebufferRegs::DepthFormat::D24:
return Color::DecodeD24(src_pixel);
case FramebufferRegs::DepthFormat::D24S8:
return Color::DecodeD24S8(src_pixel).x;
default:
LOG_CRITICAL(HW_GPU, "Unimplemented depth format %u", framebuffer.depth_format);
UNIMPLEMENTED();
return 0;
}
}
u8 GetStencil(int x, int y) {
const auto& framebuffer = g_state.regs.framebuffer.framebuffer;
const PAddr addr = framebuffer.GetDepthBufferPhysicalAddress();
u8* depth_buffer = Memory::GetPhysicalPointer(addr);
y = framebuffer.height - y;
const u32 coarse_y = y & ~7;
u32 bytes_per_pixel = Pica::FramebufferRegs::BytesPerDepthPixel(framebuffer.depth_format);
u32 stride = framebuffer.width * bytes_per_pixel;
u32 src_offset = VideoCore::GetMortonOffset(x, y, bytes_per_pixel) + coarse_y * stride;
u8* src_pixel = depth_buffer + src_offset;
switch (framebuffer.depth_format) {
case FramebufferRegs::DepthFormat::D24S8:
return Color::DecodeD24S8(src_pixel).y;
default:
LOG_WARNING(
HW_GPU,
"GetStencil called for function which doesn't have a stencil component (format %u)",
framebuffer.depth_format);
return 0;
}
}
void SetDepth(int x, int y, u32 value) {
const auto& framebuffer = g_state.regs.framebuffer.framebuffer;
const PAddr addr = framebuffer.GetDepthBufferPhysicalAddress();
u8* depth_buffer = Memory::GetPhysicalPointer(addr);
y = framebuffer.height - y;
const u32 coarse_y = y & ~7;
u32 bytes_per_pixel = FramebufferRegs::BytesPerDepthPixel(framebuffer.depth_format);
u32 stride = framebuffer.width * bytes_per_pixel;
u32 dst_offset = VideoCore::GetMortonOffset(x, y, bytes_per_pixel) + coarse_y * stride;
u8* dst_pixel = depth_buffer + dst_offset;
switch (framebuffer.depth_format) {
case FramebufferRegs::DepthFormat::D16:
Color::EncodeD16(value, dst_pixel);
break;
case FramebufferRegs::DepthFormat::D24:
Color::EncodeD24(value, dst_pixel);
break;
case FramebufferRegs::DepthFormat::D24S8:
Color::EncodeD24X8(value, dst_pixel);
break;
default:
LOG_CRITICAL(HW_GPU, "Unimplemented depth format %u", framebuffer.depth_format);
UNIMPLEMENTED();
break;
}
}
void SetStencil(int x, int y, u8 value) {
const auto& framebuffer = g_state.regs.framebuffer.framebuffer;
const PAddr addr = framebuffer.GetDepthBufferPhysicalAddress();
u8* depth_buffer = Memory::GetPhysicalPointer(addr);
y = framebuffer.height - y;
const u32 coarse_y = y & ~7;
u32 bytes_per_pixel = Pica::FramebufferRegs::BytesPerDepthPixel(framebuffer.depth_format);
u32 stride = framebuffer.width * bytes_per_pixel;
u32 dst_offset = VideoCore::GetMortonOffset(x, y, bytes_per_pixel) + coarse_y * stride;
u8* dst_pixel = depth_buffer + dst_offset;
switch (framebuffer.depth_format) {
case Pica::FramebufferRegs::DepthFormat::D16:
case Pica::FramebufferRegs::DepthFormat::D24:
// Nothing to do
break;
case Pica::FramebufferRegs::DepthFormat::D24S8:
Color::EncodeX24S8(value, dst_pixel);
break;
default:
LOG_CRITICAL(HW_GPU, "Unimplemented depth format %u", framebuffer.depth_format);
UNIMPLEMENTED();
break;
}
}
u8 PerformStencilAction(FramebufferRegs::StencilAction action, u8 old_stencil, u8 ref) {
switch (action) {
case FramebufferRegs::StencilAction::Keep:
return old_stencil;
case FramebufferRegs::StencilAction::Zero:
return 0;
case FramebufferRegs::StencilAction::Replace:
return ref;
case FramebufferRegs::StencilAction::Increment:
// Saturated increment
return std::min<u8>(old_stencil, 254) + 1;
case FramebufferRegs::StencilAction::Decrement:
// Saturated decrement
return std::max<u8>(old_stencil, 1) - 1;
case FramebufferRegs::StencilAction::Invert:
return ~old_stencil;
case FramebufferRegs::StencilAction::IncrementWrap:
return old_stencil + 1;
case FramebufferRegs::StencilAction::DecrementWrap:
return old_stencil - 1;
default:
LOG_CRITICAL(HW_GPU, "Unknown stencil action %x", (int)action);
UNIMPLEMENTED();
return 0;
}
}
Math::Vec4<u8> EvaluateBlendEquation(const Math::Vec4<u8>& src, const Math::Vec4<u8>& srcfactor,
const Math::Vec4<u8>& dest, const Math::Vec4<u8>& destfactor,
FramebufferRegs::BlendEquation equation) {
Math::Vec4<int> result;
auto src_result = (src * srcfactor).Cast<int>();
auto dst_result = (dest * destfactor).Cast<int>();
switch (equation) {
case FramebufferRegs::BlendEquation::Add:
result = (src_result + dst_result) / 255;
break;
case FramebufferRegs::BlendEquation::Subtract:
result = (src_result - dst_result) / 255;
break;
case FramebufferRegs::BlendEquation::ReverseSubtract:
result = (dst_result - src_result) / 255;
break;
// TODO: How do these two actually work? OpenGL doesn't include the blend factors in the
// min/max computations, but is this what the 3DS actually does?
case FramebufferRegs::BlendEquation::Min:
result.r() = std::min(src.r(), dest.r());
result.g() = std::min(src.g(), dest.g());
result.b() = std::min(src.b(), dest.b());
result.a() = std::min(src.a(), dest.a());
break;
case FramebufferRegs::BlendEquation::Max:
result.r() = std::max(src.r(), dest.r());
result.g() = std::max(src.g(), dest.g());
result.b() = std::max(src.b(), dest.b());
result.a() = std::max(src.a(), dest.a());
break;
default:
LOG_CRITICAL(HW_GPU, "Unknown RGB blend equation %x", equation);
UNIMPLEMENTED();
}
return Math::Vec4<u8>(MathUtil::Clamp(result.r(), 0, 255), MathUtil::Clamp(result.g(), 0, 255),
MathUtil::Clamp(result.b(), 0, 255), MathUtil::Clamp(result.a(), 0, 255));
};
u8 LogicOp(u8 src, u8 dest, FramebufferRegs::LogicOp op) {
switch (op) {
case FramebufferRegs::LogicOp::Clear:
return 0;
case FramebufferRegs::LogicOp::And:
return src & dest;
case FramebufferRegs::LogicOp::AndReverse:
return src & ~dest;
case FramebufferRegs::LogicOp::Copy:
return src;
case FramebufferRegs::LogicOp::Set:
return 255;
case FramebufferRegs::LogicOp::CopyInverted:
return ~src;
case FramebufferRegs::LogicOp::NoOp:
return dest;
case FramebufferRegs::LogicOp::Invert:
return ~dest;
case FramebufferRegs::LogicOp::Nand:
return ~(src & dest);
case FramebufferRegs::LogicOp::Or:
return src | dest;
case FramebufferRegs::LogicOp::Nor:
return ~(src | dest);
case FramebufferRegs::LogicOp::Xor:
return src ^ dest;
case FramebufferRegs::LogicOp::Equiv:
return ~(src ^ dest);
case FramebufferRegs::LogicOp::AndInverted:
return ~src & dest;
case FramebufferRegs::LogicOp::OrReverse:
return src | ~dest;
case FramebufferRegs::LogicOp::OrInverted:
return ~src | dest;
}
UNREACHABLE();
};
} // namespace Rasterizer
} // namespace Pica

29
src/video_core/swrasterizer/framebuffer.h
View file

@ -1,29 +0,0 @@
// Copyright 2017 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include "common/common_types.h"
#include "common/vector_math.h"
#include "video_core/regs_framebuffer.h"
namespace Pica {
namespace Rasterizer {
void DrawPixel(int x, int y, const Math::Vec4<u8>& color);
const Math::Vec4<u8> GetPixel(int x, int y);
u32 GetDepth(int x, int y);
u8 GetStencil(int x, int y);
void SetDepth(int x, int y, u32 value);
void SetStencil(int x, int y, u8 value);
u8 PerformStencilAction(FramebufferRegs::StencilAction action, u8 old_stencil, u8 ref);
Math::Vec4<u8> EvaluateBlendEquation(const Math::Vec4<u8>& src, const Math::Vec4<u8>& srcfactor,
const Math::Vec4<u8>& dest, const Math::Vec4<u8>& destfactor,
FramebufferRegs::BlendEquation equation);
u8 LogicOp(u8 src, u8 dest, FramebufferRegs::LogicOp op);
} // namespace Rasterizer
} // namespace Pica

308
src/video_core/swrasterizer/lighting.cpp
View file

@ -1,308 +0,0 @@
// Copyright 2017 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/math_util.h"
#include "video_core/swrasterizer/lighting.h"
namespace Pica {
static float LookupLightingLut(const Pica::State::Lighting& lighting, size_t lut_index, u8 index,
float delta) {
ASSERT_MSG(lut_index < lighting.luts.size(), "Out of range lut");
ASSERT_MSG(index < lighting.luts[lut_index].size(), "Out of range index");
const auto& lut = lighting.luts[lut_index][index];
float lut_value = lut.ToFloat();
float lut_diff = lut.DiffToFloat();
return lut_value + lut_diff * delta;
}
std::tuple<Math::Vec4<u8>, Math::Vec4<u8>> ComputeFragmentsColors(
const Pica::LightingRegs& lighting, const Pica::State::Lighting& lighting_state,
const Math::Quaternion<float>& normquat, const Math::Vec3<float>& view,
const Math::Vec4<u8> (&texture_color)[4]) {
Math::Vec3<float> surface_normal;
Math::Vec3<float> surface_tangent;
if (lighting.config0.bump_mode != LightingRegs::LightingBumpMode::None) {
Math::Vec3<float> perturbation =
texture_color[lighting.config0.bump_selector].xyz().Cast<float>() / 127.5f -
Math::MakeVec(1.0f, 1.0f, 1.0f);
if (lighting.config0.bump_mode == LightingRegs::LightingBumpMode::NormalMap) {
if (!lighting.config0.disable_bump_renorm) {
const float z_square = 1 - perturbation.xy().Length2();
perturbation.z = std::sqrt(std::max(z_square, 0.0f));
}
surface_normal = perturbation;
surface_tangent = Math::MakeVec(1.0f, 0.0f, 0.0f);
} else if (lighting.config0.bump_mode == LightingRegs::LightingBumpMode::TangentMap) {
surface_normal = Math::MakeVec(0.0f, 0.0f, 1.0f);
surface_tangent = perturbation;
} else {
LOG_ERROR(HW_GPU, "Unknown bump mode %u", lighting.config0.bump_mode.Value());
}
} else {
surface_normal = Math::MakeVec(0.0f, 0.0f, 1.0f);
surface_tangent = Math::MakeVec(1.0f, 0.0f, 0.0f);
}
// Use the normalized the quaternion when performing the rotation
auto normal = Math::QuaternionRotate(normquat, surface_normal);
auto tangent = Math::QuaternionRotate(normquat, surface_tangent);
Math::Vec4<float> diffuse_sum = {0.0f, 0.0f, 0.0f, 1.0f};
Math::Vec4<float> specular_sum = {0.0f, 0.0f, 0.0f, 1.0f};
for (unsigned light_index = 0; light_index <= lighting.max_light_index; ++light_index) {
unsigned num = lighting.light_enable.GetNum(light_index);
const auto& light_config = lighting.light[num];
Math::Vec3<float> refl_value = {};
Math::Vec3<float> position = {float16::FromRaw(light_config.x).ToFloat32(),
float16::FromRaw(light_config.y).ToFloat32(),
float16::FromRaw(light_config.z).ToFloat32()};
Math::Vec3<float> light_vector;
if (light_config.config.directional)
light_vector = position;
else
light_vector = position + view;
light_vector.Normalize();
Math::Vec3<float> norm_view = view.Normalized();
Math::Vec3<float> half_vector = norm_view + light_vector;
float dist_atten = 1.0f;
if (!lighting.IsDistAttenDisabled(num)) {
auto distance = (-view - position).Length();
float scale = Pica::float20::FromRaw(light_config.dist_atten_scale).ToFloat32();
float bias = Pica::float20::FromRaw(light_config.dist_atten_bias).ToFloat32();
size_t lut =
static_cast<size_t>(LightingRegs::LightingSampler::DistanceAttenuation) + num;
float sample_loc = MathUtil::Clamp(scale * distance + bias, 0.0f, 1.0f);
u8 lutindex =
static_cast<u8>(MathUtil::Clamp(std::floor(sample_loc * 256.0f), 0.0f, 255.0f));
float delta = sample_loc * 256 - lutindex;
dist_atten = LookupLightingLut(lighting_state, lut, lutindex, delta);
}
auto GetLutValue = [&](LightingRegs::LightingLutInput input, bool abs,
LightingRegs::LightingScale scale_enum,
LightingRegs::LightingSampler sampler) {
float result = 0.0f;
switch (input) {
case LightingRegs::LightingLutInput::NH:
result = Math::Dot(normal, half_vector.Normalized());
break;
case LightingRegs::LightingLutInput::VH:
result = Math::Dot(norm_view, half_vector.Normalized());
break;
case LightingRegs::LightingLutInput::NV:
result = Math::Dot(normal, norm_view);
break;
case LightingRegs::LightingLutInput::LN:
result = Math::Dot(light_vector, normal);
break;
case LightingRegs::LightingLutInput::SP: {
Math::Vec3<s32> spot_dir{light_config.spot_x.Value(), light_config.spot_y.Value(),
light_config.spot_z.Value()};
result = Math::Dot(light_vector, spot_dir.Cast<float>() / 2047.0f);
break;
}
case LightingRegs::LightingLutInput::CP:
if (lighting.config0.config == LightingRegs::LightingConfig::Config7) {
const Math::Vec3<float> norm_half_vector = half_vector.Normalized();
const Math::Vec3<float> half_vector_proj =
norm_half_vector - normal * Math::Dot(normal, norm_half_vector);
result = Math::Dot(half_vector_proj, tangent);
} else {
result = 0.0f;
}
break;
default:
LOG_CRITICAL(HW_GPU, "Unknown lighting LUT input %u\n", static_cast<u32>(input));
UNIMPLEMENTED();
result = 0.0f;
}
u8 index;
float delta;
if (abs) {
if (light_config.config.two_sided_diffuse)
result = std::abs(result);
else
result = std::max(result, 0.0f);
float flr = std::floor(result * 256.0f);
index = static_cast<u8>(MathUtil::Clamp(flr, 0.0f, 255.0f));
delta = result * 256 - index;
} else {
float flr = std::floor(result * 128.0f);
s8 signed_index = static_cast<s8>(MathUtil::Clamp(flr, -128.0f, 127.0f));
delta = result * 128.0f - signed_index;
index = static_cast<u8>(signed_index);
}
float scale = lighting.lut_scale.GetScale(scale_enum);
return scale *
LookupLightingLut(lighting_state, static_cast<size_t>(sampler), index, delta);
};
// If enabled, compute spot light attenuation value
float spot_atten = 1.0f;
if (!lighting.IsSpotAttenDisabled(num) &&
LightingRegs::IsLightingSamplerSupported(
lighting.config0.config, LightingRegs::LightingSampler::SpotlightAttenuation)) {
auto lut = LightingRegs::SpotlightAttenuationSampler(num);
spot_atten = GetLutValue(lighting.lut_input.sp, lighting.abs_lut_input.disable_sp == 0,
lighting.lut_scale.sp, lut);
}
// Specular 0 component
float d0_lut_value = 1.0f;
if (lighting.config1.disable_lut_d0 == 0 &&
LightingRegs::IsLightingSamplerSupported(
lighting.config0.config, LightingRegs::LightingSampler::Distribution0)) {
d0_lut_value =
GetLutValue(lighting.lut_input.d0, lighting.abs_lut_input.disable_d0 == 0,
lighting.lut_scale.d0, LightingRegs::LightingSampler::Distribution0);
}
Math::Vec3<float> specular_0 = d0_lut_value * light_config.specular_0.ToVec3f();
// If enabled, lookup ReflectRed value, otherwise, 1.0 is used
if (lighting.config1.disable_lut_rr == 0 &&
LightingRegs::IsLightingSamplerSupported(lighting.config0.config,
LightingRegs::LightingSampler::ReflectRed)) {
refl_value.x =
GetLutValue(lighting.lut_input.rr, lighting.abs_lut_input.disable_rr == 0,
lighting.lut_scale.rr, LightingRegs::LightingSampler::ReflectRed);
} else {
refl_value.x = 1.0f;
}
// If enabled, lookup ReflectGreen value, otherwise, ReflectRed value is used
if (lighting.config1.disable_lut_rg == 0 &&
LightingRegs::IsLightingSamplerSupported(lighting.config0.config,
LightingRegs::LightingSampler::ReflectGreen)) {
refl_value.y =
GetLutValue(lighting.lut_input.rg, lighting.abs_lut_input.disable_rg == 0,
lighting.lut_scale.rg, LightingRegs::LightingSampler::ReflectGreen);
} else {
refl_value.y = refl_value.x;
}
// If enabled, lookup ReflectBlue value, otherwise, ReflectRed value is used
if (lighting.config1.disable_lut_rb == 0 &&
LightingRegs::IsLightingSamplerSupported(lighting.config0.config,
LightingRegs::LightingSampler::ReflectBlue)) {
refl_value.z =
GetLutValue(lighting.lut_input.rb, lighting.abs_lut_input.disable_rb == 0,
lighting.lut_scale.rb, LightingRegs::LightingSampler::ReflectBlue);
} else {
refl_value.z = refl_value.x;
}
// Specular 1 component
float d1_lut_value = 1.0f;
if (lighting.config1.disable_lut_d1 == 0 &&
LightingRegs::IsLightingSamplerSupported(
lighting.config0.config, LightingRegs::LightingSampler::Distribution1)) {
d1_lut_value =
GetLutValue(lighting.lut_input.d1, lighting.abs_lut_input.disable_d1 == 0,
lighting.lut_scale.d1, LightingRegs::LightingSampler::Distribution1);
}
Math::Vec3<float> specular_1 =
d1_lut_value * refl_value * light_config.specular_1.ToVec3f();
// Fresnel
// Note: only the last entry in the light slots applies the Fresnel factor
if (light_index == lighting.max_light_index && lighting.config1.disable_lut_fr == 0 &&
LightingRegs::IsLightingSamplerSupported(lighting.config0.config,
LightingRegs::LightingSampler::Fresnel)) {
float lut_value =
GetLutValue(lighting.lut_input.fr, lighting.abs_lut_input.disable_fr == 0,
lighting.lut_scale.fr, LightingRegs::LightingSampler::Fresnel);
// Enabled for diffuse lighting alpha component
if (lighting.config0.fresnel_selector ==
LightingRegs::LightingFresnelSelector::PrimaryAlpha ||
lighting.config0.fresnel_selector == LightingRegs::LightingFresnelSelector::Both) {
diffuse_sum.a() = lut_value;
}
// Enabled for the specular lighting alpha component
if (lighting.config0.fresnel_selector ==
LightingRegs::LightingFresnelSelector::SecondaryAlpha ||
lighting.config0.fresnel_selector == LightingRegs::LightingFresnelSelector::Both) {
specular_sum.a() = lut_value;
}
}
auto dot_product = Math::Dot(light_vector, normal);
// Calculate clamp highlights before applying the two-sided diffuse configuration to the dot
// product.
float clamp_highlights = 1.0f;
if (lighting.config0.clamp_highlights) {
if (dot_product <= 0.0f)
clamp_highlights = 0.0f;
else
clamp_highlights = 1.0f;
}
if (light_config.config.two_sided_diffuse)
dot_product = std::abs(dot_product);
else
dot_product = std::max(dot_product, 0.0f);
if (light_config.config.geometric_factor_0 || light_config.config.geometric_factor_1) {
float geo_factor = half_vector.Length2();
geo_factor = geo_factor == 0.0f ? 0.0f : std::min(dot_product / geo_factor, 1.0f);
if (light_config.config.geometric_factor_0) {
specular_0 *= geo_factor;
}
if (light_config.config.geometric_factor_1) {
specular_1 *= geo_factor;
}
}
auto diffuse =
light_config.diffuse.ToVec3f() * dot_product + light_config.ambient.ToVec3f();
diffuse_sum += Math::MakeVec(diffuse * dist_atten * spot_atten, 0.0f);
specular_sum += Math::MakeVec(
(specular_0 + specular_1) * clamp_highlights * dist_atten * spot_atten, 0.0f);
}
diffuse_sum += Math::MakeVec(lighting.global_ambient.ToVec3f(), 0.0f);
auto diffuse = Math::MakeVec<float>(MathUtil::Clamp(diffuse_sum.x, 0.0f, 1.0f) * 255,
MathUtil::Clamp(diffuse_sum.y, 0.0f, 1.0f) * 255,
MathUtil::Clamp(diffuse_sum.z, 0.0f, 1.0f) * 255,
MathUtil::Clamp(diffuse_sum.w, 0.0f, 1.0f) * 255)
.Cast<u8>();
auto specular = Math::MakeVec<float>(MathUtil::Clamp(specular_sum.x, 0.0f, 1.0f) * 255,
MathUtil::Clamp(specular_sum.y, 0.0f, 1.0f) * 255,
MathUtil::Clamp(specular_sum.z, 0.0f, 1.0f) * 255,
MathUtil::Clamp(specular_sum.w, 0.0f, 1.0f) * 255)
.Cast<u8>();
return std::make_tuple(diffuse, specular);
}
} // namespace Pica

19
src/video_core/swrasterizer/lighting.h
View file

@ -1,19 +0,0 @@
// Copyright 2017 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <tuple>
#include "common/quaternion.h"
#include "common/vector_math.h"
#include "video_core/pica_state.h"
namespace Pica {
std::tuple<Math::Vec4<u8>, Math::Vec4<u8>> ComputeFragmentsColors(
const Pica::LightingRegs& lighting, const Pica::State::Lighting& lighting_state,
const Math::Quaternion<float>& normquat, const Math::Vec3<float>& view,
const Math::Vec4<u8> (&texture_color)[4]);
} // namespace Pica

223
src/video_core/swrasterizer/proctex.cpp
View file

@ -1,223 +0,0 @@
// Copyright 2017 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <array>
#include <cmath>
#include "common/math_util.h"
#include "video_core/swrasterizer/proctex.h"
namespace Pica {
namespace Rasterizer {
using ProcTexClamp = TexturingRegs::ProcTexClamp;
using ProcTexShift = TexturingRegs::ProcTexShift;
using ProcTexCombiner = TexturingRegs::ProcTexCombiner;
using ProcTexFilter = TexturingRegs::ProcTexFilter;
static float LookupLUT(const std::array<State::ProcTex::ValueEntry, 128>& lut, float coord) {
// For NoiseLUT/ColorMap/AlphaMap, coord=0.0 is lut[0], coord=127.0/128.0 is lut[127] and
// coord=1.0 is lut[127]+lut_diff[127]. For other indices, the result is interpolated using
// value entries and difference entries.
coord *= 128;
const int index_int = std::min(static_cast<int>(coord), 127);
const float frac = coord - index_int;
return lut[index_int].ToFloat() + frac * lut[index_int].DiffToFloat();
}
// These function are used to generate random noise for procedural texture. Their results are
// verified against real hardware, but it's not known if the algorithm is the same as hardware.
static unsigned int NoiseRand1D(unsigned int v) {
static constexpr std::array<unsigned int, 16> table{
{0, 4, 10, 8, 4, 9, 7, 12, 5, 15, 13, 14, 11, 15, 2, 11}};
return ((v % 9 + 2) * 3 & 0xF) ^ table[(v / 9) & 0xF];
}
static float NoiseRand2D(unsigned int x, unsigned int y) {
static constexpr std::array<unsigned int, 16> table{
{10, 2, 15, 8, 0, 7, 4, 5, 5, 13, 2, 6, 13, 9, 3, 14}};
unsigned int u2 = NoiseRand1D(x);
unsigned int v2 = NoiseRand1D(y);
v2 += ((u2 & 3) == 1) ? 4 : 0;
v2 ^= (u2 & 1) * 6;
v2 += 10 + u2;
v2 &= 0xF;
v2 ^= table[u2];
return -1.0f + v2 * 2.0f / 15.0f;
}
static float NoiseCoef(float u, float v, TexturingRegs regs, State::ProcTex state) {
const float freq_u = float16::FromRaw(regs.proctex_noise_frequency.u).ToFloat32();
const float freq_v = float16::FromRaw(regs.proctex_noise_frequency.v).ToFloat32();
const float phase_u = float16::FromRaw(regs.proctex_noise_u.phase).ToFloat32();
const float phase_v = float16::FromRaw(regs.proctex_noise_v.phase).ToFloat32();
const float x = 9 * freq_u * std::abs(u + phase_u);
const float y = 9 * freq_v * std::abs(v + phase_v);
const int x_int = static_cast<int>(x);
const int y_int = static_cast<int>(y);
const float x_frac = x - x_int;
const float y_frac = y - y_int;
const float g0 = NoiseRand2D(x_int, y_int) * (x_frac + y_frac);
const float g1 = NoiseRand2D(x_int + 1, y_int) * (x_frac + y_frac - 1);
const float g2 = NoiseRand2D(x_int, y_int + 1) * (x_frac + y_frac - 1);
const float g3 = NoiseRand2D(x_int + 1, y_int + 1) * (x_frac + y_frac - 2);
const float x_noise = LookupLUT(state.noise_table, x_frac);
const float y_noise = LookupLUT(state.noise_table, y_frac);
return Math::BilinearInterp(g0, g1, g2, g3, x_noise, y_noise);
}
static float GetShiftOffset(float v, ProcTexShift mode, ProcTexClamp clamp_mode) {
const float offset = (clamp_mode == ProcTexClamp::MirroredRepeat) ? 1 : 0.5f;
switch (mode) {
case ProcTexShift::None:
return 0;
case ProcTexShift::Odd:
return offset * (((int)v / 2) % 2);
case ProcTexShift::Even:
return offset * ((((int)v + 1) / 2) % 2);
default:
LOG_CRITICAL(HW_GPU, "Unknown shift mode %u", static_cast<u32>(mode));
return 0;
}
};
static void ClampCoord(float& coord, ProcTexClamp mode) {
switch (mode) {
case ProcTexClamp::ToZero:
if (coord > 1.0f)
coord = 0.0f;
break;
case ProcTexClamp::ToEdge:
coord = std::min(coord, 1.0f);
break;
case ProcTexClamp::SymmetricalRepeat:
coord = coord - std::floor(coord);
break;
case ProcTexClamp::MirroredRepeat: {
int integer = static_cast<int>(coord);
float frac = coord - integer;
coord = (integer % 2) == 0 ? frac : (1.0f - frac);
break;
}
case ProcTexClamp::Pulse:
if (coord <= 0.5f)
coord = 0.0f;
else
coord = 1.0f;
break;
default:
LOG_CRITICAL(HW_GPU, "Unknown clamp mode %u", static_cast<u32>(mode));
coord = std::min(coord, 1.0f);
break;
}
}
float CombineAndMap(float u, float v, ProcTexCombiner combiner,
const std::array<State::ProcTex::ValueEntry, 128>& map_table) {
float f;
switch (combiner) {
case ProcTexCombiner::U:
f = u;
break;
case ProcTexCombiner::U2:
f = u * u;
break;
case TexturingRegs::ProcTexCombiner::V:
f = v;
break;
case TexturingRegs::ProcTexCombiner::V2:
f = v * v;
break;
case TexturingRegs::ProcTexCombiner::Add:
f = (u + v) * 0.5f;
break;
case TexturingRegs::ProcTexCombiner::Add2:
f = (u * u + v * v) * 0.5f;
break;
case TexturingRegs::ProcTexCombiner::SqrtAdd2:
f = std::min(std::sqrt(u * u + v * v), 1.0f);
break;
case TexturingRegs::ProcTexCombiner::Min:
f = std::min(u, v);
break;
case TexturingRegs::ProcTexCombiner::Max:
f = std::max(u, v);
break;
case TexturingRegs::ProcTexCombiner::RMax:
f = std::min(((u + v) * 0.5f + std::sqrt(u * u + v * v)) * 0.5f, 1.0f);
break;
default:
LOG_CRITICAL(HW_GPU, "Unknown combiner %u", static_cast<u32>(combiner));
f = 0.0f;
break;
}
return LookupLUT(map_table, f);
}
Math::Vec4<u8> ProcTex(float u, float v, TexturingRegs regs, State::ProcTex state) {
u = std::abs(u);
v = std::abs(v);
// Get shift offset before noise generation
const float u_shift = GetShiftOffset(v, regs.proctex.u_shift, regs.proctex.u_clamp);
const float v_shift = GetShiftOffset(u, regs.proctex.v_shift, regs.proctex.v_clamp);
// Generate noise
if (regs.proctex.noise_enable) {
float noise = NoiseCoef(u, v, regs, state);
u += noise * regs.proctex_noise_u.amplitude / 4095.0f;
v += noise * regs.proctex_noise_v.amplitude / 4095.0f;
u = std::abs(u);
v = std::abs(v);
}
// Shift
u += u_shift;
v += v_shift;
// Clamp
ClampCoord(u, regs.proctex.u_clamp);
ClampCoord(v, regs.proctex.v_clamp);
// Combine and map
const float lut_coord = CombineAndMap(u, v, regs.proctex.color_combiner, state.color_map_table);
// Look up the color
// For the color lut, coord=0.0 is lut[offset] and coord=1.0 is lut[offset+width-1]
const u32 offset = regs.proctex_lut_offset;
const u32 width = regs.proctex_lut.width;
const float index = offset + (lut_coord * (width - 1));
Math::Vec4<u8> final_color;
// TODO(wwylele): implement mipmap
switch (regs.proctex_lut.filter) {
case ProcTexFilter::Linear:
case ProcTexFilter::LinearMipmapLinear:
case ProcTexFilter::LinearMipmapNearest: {
const int index_int = static_cast<int>(index);
const float frac = index - index_int;
const auto color_value = state.color_table[index_int].ToVector().Cast<float>();
const auto color_diff = state.color_diff_table[index_int].ToVector().Cast<float>();
final_color = (color_value + frac * color_diff).Cast<u8>();
break;
}
case ProcTexFilter::Nearest:
case ProcTexFilter::NearestMipmapLinear:
case ProcTexFilter::NearestMipmapNearest:
final_color = state.color_table[static_cast<int>(std::round(index))].ToVector();
break;
}
if (regs.proctex.separate_alpha) {
// Note: in separate alpha mode, the alpha channel skips the color LUT look up stage. It
// uses the output of CombineAndMap directly instead.
const float final_alpha =
CombineAndMap(u, v, regs.proctex.alpha_combiner, state.alpha_map_table);
return Math::MakeVec<u8>(final_color.rgb(), static_cast<u8>(final_alpha * 255));
} else {
return final_color;
}
}
} // namespace Rasterizer
} // namespace Pica

16
src/video_core/swrasterizer/proctex.h
View file

@ -1,16 +0,0 @@
// Copyright 2017 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/common_types.h"
#include "common/vector_math.h"
#include "video_core/pica_state.h"
namespace Pica {
namespace Rasterizer {
/// Generates procedural texture color for the given coordinates
Math::Vec4<u8> ProcTex(float u, float v, TexturingRegs regs, State::ProcTex state);
} // namespace Rasterizer
} // namespace Pica

853
src/video_core/swrasterizer/rasterizer.cpp
View file

@ -1,853 +0,0 @@
// Copyright 2014 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <algorithm>
#include <array>
#include <cmath>
#include <tuple>
#include "common/assert.h"
#include "common/bit_field.h"
#include "common/color.h"
#include "common/common_types.h"
#include "common/logging/log.h"
#include "common/math_util.h"
#include "common/microprofile.h"
#include "common/quaternion.h"
#include "common/vector_math.h"
#include "core/hw/gpu.h"
#include "core/memory.h"
#include "video_core/debug_utils/debug_utils.h"
#include "video_core/pica_state.h"
#include "video_core/pica_types.h"
#include "video_core/regs_framebuffer.h"
#include "video_core/regs_rasterizer.h"
#include "video_core/regs_texturing.h"
#include "video_core/shader/shader.h"
#include "video_core/swrasterizer/framebuffer.h"
#include "video_core/swrasterizer/lighting.h"
#include "video_core/swrasterizer/proctex.h"
#include "video_core/swrasterizer/rasterizer.h"
#include "video_core/swrasterizer/texturing.h"
#include "video_core/texture/texture_decode.h"
#include "video_core/utils.h"
namespace Pica {
namespace Rasterizer {
// NOTE: Assuming that rasterizer coordinates are 12.4 fixed-point values
struct Fix12P4 {
Fix12P4() {}
Fix12P4(u16 val) : val(val) {}
static u16 FracMask() {
return 0xF;
}
static u16 IntMask() {
return (u16)~0xF;
}
operator u16() const {
return val;
}
bool operator<(const Fix12P4& oth) const {
return (u16) * this < (u16)oth;
}
private:
u16 val;
};
/**
* Calculate signed area of the triangle spanned by the three argument vertices.
* The sign denotes an orientation.
*
* @todo define orientation concretely.
*/
static int SignedArea(const Math::Vec2<Fix12P4>& vtx1, const Math::Vec2<Fix12P4>& vtx2,
const Math::Vec2<Fix12P4>& vtx3) {
const auto vec1 = Math::MakeVec(vtx2 - vtx1, 0);
const auto vec2 = Math::MakeVec(vtx3 - vtx1, 0);
// TODO: There is a very small chance this will overflow for sizeof(int) == 4
return Math::Cross(vec1, vec2).z;
};
/// Convert a 3D vector for cube map coordinates to 2D texture coordinates along with the face name
static std::tuple<float24, float24, PAddr> ConvertCubeCoord(float24 u, float24 v, float24 w,
const TexturingRegs& regs) {
const float abs_u = std::abs(u.ToFloat32());
const float abs_v = std::abs(v.ToFloat32());
const float abs_w = std::abs(w.ToFloat32());
float24 x, y, z;
PAddr addr;
if (abs_u > abs_v && abs_u > abs_w) {
if (u > float24::FromFloat32(0)) {
addr = regs.GetCubePhysicalAddress(TexturingRegs::CubeFace::PositiveX);
y = -v;
} else {
addr = regs.GetCubePhysicalAddress(TexturingRegs::CubeFace::NegativeX);
y = v;
}
x = -w;
z = u;
} else if (abs_v > abs_w) {
if (v > float24::FromFloat32(0)) {
addr = regs.GetCubePhysicalAddress(TexturingRegs::CubeFace::PositiveY);
x = u;
} else {
addr = regs.GetCubePhysicalAddress(TexturingRegs::CubeFace::NegativeY);
x = -u;
}
y = w;
z = v;
} else {
if (w > float24::FromFloat32(0)) {
addr = regs.GetCubePhysicalAddress(TexturingRegs::CubeFace::PositiveZ);
y = -v;
} else {
addr = regs.GetCubePhysicalAddress(TexturingRegs::CubeFace::NegativeZ);
y = v;
}
x = u;
z = w;
}
const float24 half = float24::FromFloat32(0.5f);
return std::make_tuple(x / z * half + half, y / z * half + half, addr);
}
MICROPROFILE_DEFINE(GPU_Rasterization, "GPU", "Rasterization", MP_RGB(50, 50, 240));
/**
* Helper function for ProcessTriangle with the "reversed" flag to allow for implementing
* culling via recursion.
*/
static void ProcessTriangleInternal(const Vertex& v0, const Vertex& v1, const Vertex& v2,
bool reversed = false) {
const auto& regs = g_state.regs;
MICROPROFILE_SCOPE(GPU_Rasterization);
// vertex positions in rasterizer coordinates
static auto FloatToFix = [](float24 flt) {
// TODO: Rounding here is necessary to prevent garbage pixels at
// triangle borders. Is it that the correct solution, though?
return Fix12P4(static_cast<unsigned short>(round(flt.ToFloat32() * 16.0f)));
};
static auto ScreenToRasterizerCoordinates = [](const Math::Vec3<float24>& vec) {
return Math::Vec3<Fix12P4>{FloatToFix(vec.x), FloatToFix(vec.y), FloatToFix(vec.z)};
};
Math::Vec3<Fix12P4> vtxpos[3]{ScreenToRasterizerCoordinates(v0.screenpos),
ScreenToRasterizerCoordinates(v1.screenpos),
ScreenToRasterizerCoordinates(v2.screenpos)};
if (regs.rasterizer.cull_mode == RasterizerRegs::CullMode::KeepAll) {
// Make sure we always end up with a triangle wound counter-clockwise
if (!reversed && SignedArea(vtxpos[0].xy(), vtxpos[1].xy(), vtxpos[2].xy()) <= 0) {
ProcessTriangleInternal(v0, v2, v1, true);
return;
}
} else {
if (!reversed && regs.rasterizer.cull_mode == RasterizerRegs::CullMode::KeepClockWise) {
// Reverse vertex order and use the CCW code path.
ProcessTriangleInternal(v0, v2, v1, true);
return;
}
// Cull away triangles which are wound clockwise.
if (SignedArea(vtxpos[0].xy(), vtxpos[1].xy(), vtxpos[2].xy()) <= 0)
return;
}
u16 min_x = std::min({vtxpos[0].x, vtxpos[1].x, vtxpos[2].x});
u16 min_y = std::min({vtxpos[0].y, vtxpos[1].y, vtxpos[2].y});
u16 max_x = std::max({vtxpos[0].x, vtxpos[1].x, vtxpos[2].x});
u16 max_y = std::max({vtxpos[0].y, vtxpos[1].y, vtxpos[2].y});
// Convert the scissor box coordinates to 12.4 fixed point
u16 scissor_x1 = (u16)(regs.rasterizer.scissor_test.x1 << 4);
u16 scissor_y1 = (u16)(regs.rasterizer.scissor_test.y1 << 4);
// x2,y2 have +1 added to cover the entire sub-pixel area
u16 scissor_x2 = (u16)((regs.rasterizer.scissor_test.x2 + 1) << 4);
u16 scissor_y2 = (u16)((regs.rasterizer.scissor_test.y2 + 1) << 4);
if (regs.rasterizer.scissor_test.mode == RasterizerRegs::ScissorMode::Include) {
// Calculate the new bounds
min_x = std::max(min_x, scissor_x1);
min_y = std::max(min_y, scissor_y1);
max_x = std::min(max_x, scissor_x2);
max_y = std::min(max_y, scissor_y2);
}
min_x &= Fix12P4::IntMask();
min_y &= Fix12P4::IntMask();
max_x = ((max_x + Fix12P4::FracMask()) & Fix12P4::IntMask());
max_y = ((max_y + Fix12P4::FracMask()) & Fix12P4::IntMask());
// Triangle filling rules: Pixels on the right-sided edge or on flat bottom edges are not
// drawn. Pixels on any other triangle border are drawn. This is implemented with three bias
// values which are added to the barycentric coordinates w0, w1 and w2, respectively.
// NOTE: These are the PSP filling rules. Not sure if the 3DS uses the same ones...
auto IsRightSideOrFlatBottomEdge = [](const Math::Vec2<Fix12P4>& vtx,
const Math::Vec2<Fix12P4>& line1,
const Math::Vec2<Fix12P4>& line2) {
if (line1.y == line2.y) {
// just check if vertex is above us => bottom line parallel to x-axis
return vtx.y < line1.y;
} else {
// check if vertex is on our left => right side
// TODO: Not sure how likely this is to overflow
return (int)vtx.x < (int)line1.x +
((int)line2.x - (int)line1.x) * ((int)vtx.y - (int)line1.y) /
((int)line2.y - (int)line1.y);
}
};
int bias0 =
IsRightSideOrFlatBottomEdge(vtxpos[0].xy(), vtxpos[1].xy(), vtxpos[2].xy()) ? -1 : 0;
int bias1 =
IsRightSideOrFlatBottomEdge(vtxpos[1].xy(), vtxpos[2].xy(), vtxpos[0].xy()) ? -1 : 0;
int bias2 =
IsRightSideOrFlatBottomEdge(vtxpos[2].xy(), vtxpos[0].xy(), vtxpos[1].xy()) ? -1 : 0;
auto w_inverse = Math::MakeVec(v0.pos.w, v1.pos.w, v2.pos.w);
auto textures = regs.texturing.GetTextures();
auto tev_stages = regs.texturing.GetTevStages();
bool stencil_action_enable =
g_state.regs.framebuffer.output_merger.stencil_test.enable &&
g_state.regs.framebuffer.framebuffer.depth_format == FramebufferRegs::DepthFormat::D24S8;
const auto stencil_test = g_state.regs.framebuffer.output_merger.stencil_test;
// Enter rasterization loop, starting at the center of the topleft bounding box corner.
// TODO: Not sure if looping through x first might be faster
for (u16 y = min_y + 8; y < max_y; y += 0x10) {
for (u16 x = min_x + 8; x < max_x; x += 0x10) {
// Do not process the pixel if it's inside the scissor box and the scissor mode is set
// to Exclude
if (regs.rasterizer.scissor_test.mode == RasterizerRegs::ScissorMode::Exclude) {
if (x >= scissor_x1 && x < scissor_x2 && y >= scissor_y1 && y < scissor_y2)
continue;
}
// Calculate the barycentric coordinates w0, w1 and w2
int w0 = bias0 + SignedArea(vtxpos[1].xy(), vtxpos[2].xy(), {x, y});
int w1 = bias1 + SignedArea(vtxpos[2].xy(), vtxpos[0].xy(), {x, y});
int w2 = bias2 + SignedArea(vtxpos[0].xy(), vtxpos[1].xy(), {x, y});
int wsum = w0 + w1 + w2;
// If current pixel is not covered by the current primitive
if (w0 < 0 || w1 < 0 || w2 < 0)
continue;
auto baricentric_coordinates =
Math::MakeVec(float24::FromFloat32(static_cast<float>(w0)),
float24::FromFloat32(static_cast<float>(w1)),
float24::FromFloat32(static_cast<float>(w2)));
float24 interpolated_w_inverse =
float24::FromFloat32(1.0f) / Math::Dot(w_inverse, baricentric_coordinates);
// interpolated_z = z / w
float interpolated_z_over_w =
(v0.screenpos[2].ToFloat32() * w0 + v1.screenpos[2].ToFloat32() * w1 +
v2.screenpos[2].ToFloat32() * w2) /
wsum;
// Not fully accurate. About 3 bits in precision are missing.
// Z-Buffer (z / w * scale + offset)
float depth_scale = float24::FromRaw(regs.rasterizer.viewport_depth_range).ToFloat32();
float depth_offset =
float24::FromRaw(regs.rasterizer.viewport_depth_near_plane).ToFloat32();
float depth = interpolated_z_over_w * depth_scale + depth_offset;
// Potentially switch to W-Buffer
if (regs.rasterizer.depthmap_enable ==
Pica::RasterizerRegs::DepthBuffering::WBuffering) {
// W-Buffer (z * scale + w * offset = (z / w * scale + offset) * w)
depth *= interpolated_w_inverse.ToFloat32() * wsum;
}
// Clamp the result
depth = MathUtil::Clamp(depth, 0.0f, 1.0f);
// Perspective correct attribute interpolation:
// Attribute values cannot be calculated by simple linear interpolation since
// they are not linear in screen space. For example, when interpolating a
// texture coordinate across two vertices, something simple like
// u = (u0*w0 + u1*w1)/(w0+w1)
// will not work. However, the attribute value divided by the
// clipspace w-coordinate (u/w) and and the inverse w-coordinate (1/w) are linear
// in screenspace. Hence, we can linearly interpolate these two independently and
// calculate the interpolated attribute by dividing the results.
// I.e.
// u_over_w = ((u0/v0.pos.w)*w0 + (u1/v1.pos.w)*w1)/(w0+w1)
// one_over_w = (( 1/v0.pos.w)*w0 + ( 1/v1.pos.w)*w1)/(w0+w1)
// u = u_over_w / one_over_w
//
// The generalization to three vertices is straightforward in baricentric coordinates.
auto GetInterpolatedAttribute = [&](float24 attr0, float24 attr1, float24 attr2) {
auto attr_over_w = Math::MakeVec(attr0, attr1, attr2);
float24 interpolated_attr_over_w = Math::Dot(attr_over_w, baricentric_coordinates);
return interpolated_attr_over_w * interpolated_w_inverse;
};
Math::Vec4<u8> primary_color{
(u8)(
GetInterpolatedAttribute(v0.color.r(), v1.color.r(), v2.color.r()).ToFloat32() *
255),
(u8)(
GetInterpolatedAttribute(v0.color.g(), v1.color.g(), v2.color.g()).ToFloat32() *
255),
(u8)(
GetInterpolatedAttribute(v0.color.b(), v1.color.b(), v2.color.b()).ToFloat32() *
255),
(u8)(
GetInterpolatedAttribute(v0.color.a(), v1.color.a(), v2.color.a()).ToFloat32() *
255),
};
Math::Vec2<float24> uv[3];
uv[0].u() = GetInterpolatedAttribute(v0.tc0.u(), v1.tc0.u(), v2.tc0.u());
uv[0].v() = GetInterpolatedAttribute(v0.tc0.v(), v1.tc0.v(), v2.tc0.v());
uv[1].u() = GetInterpolatedAttribute(v0.tc1.u(), v1.tc1.u(), v2.tc1.u());
uv[1].v() = GetInterpolatedAttribute(v0.tc1.v(), v1.tc1.v(), v2.tc1.v());
uv[2].u() = GetInterpolatedAttribute(v0.tc2.u(), v1.tc2.u(), v2.tc2.u());
uv[2].v() = GetInterpolatedAttribute(v0.tc2.v(), v1.tc2.v(), v2.tc2.v());
Math::Vec4<u8> texture_color[4]{};
for (int i = 0; i < 3; ++i) {
const auto& texture = textures[i];
if (!texture.enabled)
continue;
DEBUG_ASSERT(0 != texture.config.address);
int coordinate_i =
(i == 2 && regs.texturing.main_config.texture2_use_coord1) ? 1 : i;
float24 u = uv[coordinate_i].u();
float24 v = uv[coordinate_i].v();
// Only unit 0 respects the texturing type (according to 3DBrew)
// TODO: Refactor so cubemaps and shadowmaps can be handled
PAddr texture_address = texture.config.GetPhysicalAddress();
if (i == 0) {
switch (texture.config.type) {
case TexturingRegs::TextureConfig::Texture2D:
break;
case TexturingRegs::TextureConfig::TextureCube: {
auto w = GetInterpolatedAttribute(v0.tc0_w, v1.tc0_w, v2.tc0_w);
std::tie(u, v, texture_address) = ConvertCubeCoord(u, v, w, regs.texturing);
break;
}
case TexturingRegs::TextureConfig::Projection2D: {
auto tc0_w = GetInterpolatedAttribute(v0.tc0_w, v1.tc0_w, v2.tc0_w);
u /= tc0_w;
v /= tc0_w;
break;
}
default:
// TODO: Change to LOG_ERROR when more types are handled.
LOG_DEBUG(HW_GPU, "Unhandled texture type %x", (int)texture.config.type);
UNIMPLEMENTED();
break;
}
}
int s = (int)(u * float24::FromFloat32(static_cast<float>(texture.config.width)))
.ToFloat32();
int t = (int)(v * float24::FromFloat32(static_cast<float>(texture.config.height)))
.ToFloat32();
bool use_border_s = false;
bool use_border_t = false;
if (texture.config.wrap_s == TexturingRegs::TextureConfig::ClampToBorder) {
use_border_s = s < 0 || s >= static_cast<int>(texture.config.width);
} else if (texture.config.wrap_s == TexturingRegs::TextureConfig::ClampToBorder2) {
use_border_s = s >= static_cast<int>(texture.config.width);
}
if (texture.config.wrap_t == TexturingRegs::TextureConfig::ClampToBorder) {
use_border_t = t < 0 || t >= static_cast<int>(texture.config.height);
} else if (texture.config.wrap_t == TexturingRegs::TextureConfig::ClampToBorder2) {
use_border_t = t >= static_cast<int>(texture.config.height);
}
if (use_border_s || use_border_t) {
auto border_color = texture.config.border_color;
texture_color[i] = {border_color.r, border_color.g, border_color.b,
border_color.a};
} else {
// Textures are laid out from bottom to top, hence we invert the t coordinate.
// NOTE: This may not be the right place for the inversion.
// TODO: Check if this applies to ETC textures, too.
s = GetWrappedTexCoord(texture.config.wrap_s, s, texture.config.width);
t = texture.config.height - 1 -
GetWrappedTexCoord(texture.config.wrap_t, t, texture.config.height);
const u8* texture_data = Memory::GetPhysicalPointer(texture_address);
auto info =
Texture::TextureInfo::FromPicaRegister(texture.config, texture.format);
// TODO: Apply the min and mag filters to the texture
texture_color[i] = Texture::LookupTexture(texture_data, s, t, info);
#if PICA_DUMP_TEXTURES
DebugUtils::DumpTexture(texture.config, texture_data);
#endif
}
}
// sample procedural texture
if (regs.texturing.main_config.texture3_enable) {
const auto& proctex_uv = uv[regs.texturing.main_config.texture3_coordinates];
texture_color[3] = ProcTex(proctex_uv.u().ToFloat32(), proctex_uv.v().ToFloat32(),
g_state.regs.texturing, g_state.proctex);
}
// Texture environment - consists of 6 stages of color and alpha combining.
//
// Color combiners take three input color values from some source (e.g. interpolated
// vertex color, texture color, previous stage, etc), perform some very simple
// operations on each of them (e.g. inversion) and then calculate the output color
// with some basic arithmetic. Alpha combiners can be configured separately but work
// analogously.
Math::Vec4<u8> combiner_output;
Math::Vec4<u8> combiner_buffer = {0, 0, 0, 0};
Math::Vec4<u8> next_combiner_buffer = {
regs.texturing.tev_combiner_buffer_color.r,
regs.texturing.tev_combiner_buffer_color.g,
regs.texturing.tev_combiner_buffer_color.b,
regs.texturing.tev_combiner_buffer_color.a,
};
Math::Vec4<u8> primary_fragment_color = {0, 0, 0, 0};
Math::Vec4<u8> secondary_fragment_color = {0, 0, 0, 0};
if (!g_state.regs.lighting.disable) {
Math::Quaternion<float> normquat = Math::Quaternion<float>{
{GetInterpolatedAttribute(v0.quat.x, v1.quat.x, v2.quat.x).ToFloat32(),
GetInterpolatedAttribute(v0.quat.y, v1.quat.y, v2.quat.y).ToFloat32(),
GetInterpolatedAttribute(v0.quat.z, v1.quat.z, v2.quat.z).ToFloat32()},
GetInterpolatedAttribute(v0.quat.w, v1.quat.w, v2.quat.w).ToFloat32(),
}.Normalized();
Math::Vec3<float> view{
GetInterpolatedAttribute(v0.view.x, v1.view.x, v2.view.x).ToFloat32(),
GetInterpolatedAttribute(v0.view.y, v1.view.y, v2.view.y).ToFloat32(),
GetInterpolatedAttribute(v0.view.z, v1.view.z, v2.view.z).ToFloat32(),
};
std::tie(primary_fragment_color, secondary_fragment_color) = ComputeFragmentsColors(
g_state.regs.lighting, g_state.lighting, normquat, view, texture_color);
}
for (unsigned tev_stage_index = 0; tev_stage_index < tev_stages.size();
++tev_stage_index) {
const auto& tev_stage = tev_stages[tev_stage_index];
using Source = TexturingRegs::TevStageConfig::Source;
auto GetSource = [&](Source source) -> Math::Vec4<u8> {
switch (source) {
case Source::PrimaryColor:
return primary_color;
case Source::PrimaryFragmentColor:
return primary_fragment_color;
case Source::SecondaryFragmentColor:
return secondary_fragment_color;
case Source::Texture0:
return texture_color[0];
case Source::Texture1:
return texture_color[1];
case Source::Texture2:
return texture_color[2];
case Source::Texture3:
return texture_color[3];
case Source::PreviousBuffer:
return combiner_buffer;
case Source::Constant:
return {tev_stage.const_r, tev_stage.const_g, tev_stage.const_b,
tev_stage.const_a};
case Source::Previous:
return combiner_output;
default:
LOG_ERROR(HW_GPU, "Unknown color combiner source %d", (int)source);
UNIMPLEMENTED();
return {0, 0, 0, 0};
}
};
// color combiner
// NOTE: Not sure if the alpha combiner might use the color output of the previous
// stage as input. Hence, we currently don't directly write the result to
// combiner_output.rgb(), but instead store it in a temporary variable until
// alpha combining has been done.
Math::Vec3<u8> color_result[3] = {
GetColorModifier(tev_stage.color_modifier1, GetSource(tev_stage.color_source1)),
GetColorModifier(tev_stage.color_modifier2, GetSource(tev_stage.color_source2)),
GetColorModifier(tev_stage.color_modifier3, GetSource(tev_stage.color_source3)),
};
auto color_output = ColorCombine(tev_stage.color_op, color_result);
u8 alpha_output;
if (tev_stage.color_op == TexturingRegs::TevStageConfig::Operation::Dot3_RGBA) {
// result of Dot3_RGBA operation is also placed to the alpha component
alpha_output = color_output.x;
} else {
// alpha combiner
std::array<u8, 3> alpha_result = {{
GetAlphaModifier(tev_stage.alpha_modifier1,
GetSource(tev_stage.alpha_source1)),
GetAlphaModifier(tev_stage.alpha_modifier2,
GetSource(tev_stage.alpha_source2)),
GetAlphaModifier(tev_stage.alpha_modifier3,
GetSource(tev_stage.alpha_source3)),
}};
alpha_output = AlphaCombine(tev_stage.alpha_op, alpha_result);
}
combiner_output[0] =
std::min((unsigned)255, color_output.r() * tev_stage.GetColorMultiplier());
combiner_output[1] =
std::min((unsigned)255, color_output.g() * tev_stage.GetColorMultiplier());
combiner_output[2] =
std::min((unsigned)255, color_output.b() * tev_stage.GetColorMultiplier());
combiner_output[3] =
std::min((unsigned)255, alpha_output * tev_stage.GetAlphaMultiplier());
combiner_buffer = next_combiner_buffer;
if (regs.texturing.tev_combiner_buffer_input.TevStageUpdatesCombinerBufferColor(
tev_stage_index)) {
next_combiner_buffer.r() = combiner_output.r();
next_combiner_buffer.g() = combiner_output.g();
next_combiner_buffer.b() = combiner_output.b();
}
if (regs.texturing.tev_combiner_buffer_input.TevStageUpdatesCombinerBufferAlpha(
tev_stage_index)) {
next_combiner_buffer.a() = combiner_output.a();
}
}
const auto& output_merger = regs.framebuffer.output_merger;
// TODO: Does alpha testing happen before or after stencil?
if (output_merger.alpha_test.enable) {
bool pass = false;
switch (output_merger.alpha_test.func) {
case FramebufferRegs::CompareFunc::Never:
pass = false;
break;
case FramebufferRegs::CompareFunc::Always:
pass = true;
break;
case FramebufferRegs::CompareFunc::Equal:
pass = combiner_output.a() == output_merger.alpha_test.ref;
break;
case FramebufferRegs::CompareFunc::NotEqual:
pass = combiner_output.a() != output_merger.alpha_test.ref;
break;
case FramebufferRegs::CompareFunc::LessThan:
pass = combiner_output.a() < output_merger.alpha_test.ref;
break;
case FramebufferRegs::CompareFunc::LessThanOrEqual:
pass = combiner_output.a() <= output_merger.alpha_test.ref;
break;
case FramebufferRegs::CompareFunc::GreaterThan:
pass = combiner_output.a() > output_merger.alpha_test.ref;
break;
case FramebufferRegs::CompareFunc::GreaterThanOrEqual:
pass = combiner_output.a() >= output_merger.alpha_test.ref;
break;
}
if (!pass)
continue;
}
// Apply fog combiner
// Not fully accurate. We'd have to know what data type is used to
// store the depth etc. Using float for now until we know more
// about Pica datatypes
if (regs.texturing.fog_mode == TexturingRegs::FogMode::Fog) {
const Math::Vec3<u8> fog_color = {
static_cast<u8>(regs.texturing.fog_color.r.Value()),
static_cast<u8>(regs.texturing.fog_color.g.Value()),
static_cast<u8>(regs.texturing.fog_color.b.Value()),
};
// Get index into fog LUT
float fog_index;
if (g_state.regs.texturing.fog_flip) {
fog_index = (1.0f - depth) * 128.0f;
} else {
fog_index = depth * 128.0f;
}
// Generate clamped fog factor from LUT for given fog index
float fog_i = MathUtil::Clamp(floorf(fog_index), 0.0f, 127.0f);
float fog_f = fog_index - fog_i;
const auto& fog_lut_entry = g_state.fog.lut[static_cast<unsigned int>(fog_i)];
float fog_factor = fog_lut_entry.ToFloat() + fog_lut_entry.DiffToFloat() * fog_f;
fog_factor = MathUtil::Clamp(fog_factor, 0.0f, 1.0f);
// Blend the fog
for (unsigned i = 0; i < 3; i++) {
combiner_output[i] = static_cast<u8>(fog_factor * combiner_output[i] +
(1.0f - fog_factor) * fog_color[i]);
}
}
u8 old_stencil = 0;
auto UpdateStencil = [stencil_test, x, y,
&old_stencil](Pica::FramebufferRegs::StencilAction action) {
u8 new_stencil =
PerformStencilAction(action, old_stencil, stencil_test.reference_value);
if (g_state.regs.framebuffer.framebuffer.allow_depth_stencil_write != 0)
SetStencil(x >> 4, y >> 4, (new_stencil & stencil_test.write_mask) |
(old_stencil & ~stencil_test.write_mask));
};
if (stencil_action_enable) {
old_stencil = GetStencil(x >> 4, y >> 4);
u8 dest = old_stencil & stencil_test.input_mask;
u8 ref = stencil_test.reference_value & stencil_test.input_mask;
bool pass = false;
switch (stencil_test.func) {
case FramebufferRegs::CompareFunc::Never:
pass = false;
break;
case FramebufferRegs::CompareFunc::Always:
pass = true;
break;
case FramebufferRegs::CompareFunc::Equal:
pass = (ref == dest);
break;
case FramebufferRegs::CompareFunc::NotEqual:
pass = (ref != dest);
break;
case FramebufferRegs::CompareFunc::LessThan:
pass = (ref < dest);
break;
case FramebufferRegs::CompareFunc::LessThanOrEqual:
pass = (ref <= dest);
break;
case FramebufferRegs::CompareFunc::GreaterThan:
pass = (ref > dest);
break;
case FramebufferRegs::CompareFunc::GreaterThanOrEqual:
pass = (ref >= dest);
break;
}
if (!pass) {
UpdateStencil(stencil_test.action_stencil_fail);
continue;
}
}
// Convert float to integer
unsigned num_bits =
FramebufferRegs::DepthBitsPerPixel(regs.framebuffer.framebuffer.depth_format);
u32 z = (u32)(depth * ((1 << num_bits) - 1));
if (output_merger.depth_test_enable) {
u32 ref_z = GetDepth(x >> 4, y >> 4);
bool pass = false;
switch (output_merger.depth_test_func) {
case FramebufferRegs::CompareFunc::Never:
pass = false;
break;
case FramebufferRegs::CompareFunc::Always:
pass = true;
break;
case FramebufferRegs::CompareFunc::Equal:
pass = z == ref_z;
break;
case FramebufferRegs::CompareFunc::NotEqual:
pass = z != ref_z;
break;
case FramebufferRegs::CompareFunc::LessThan:
pass = z < ref_z;
break;
case FramebufferRegs::CompareFunc::LessThanOrEqual:
pass = z <= ref_z;
break;
case FramebufferRegs::CompareFunc::GreaterThan:
pass = z > ref_z;
break;
case FramebufferRegs::CompareFunc::GreaterThanOrEqual:
pass = z >= ref_z;
break;
}
if (!pass) {
if (stencil_action_enable)
UpdateStencil(stencil_test.action_depth_fail);
continue;
}
}
if (regs.framebuffer.framebuffer.allow_depth_stencil_write != 0 &&
output_merger.depth_write_enable) {
SetDepth(x >> 4, y >> 4, z);
}
// The stencil depth_pass action is executed even if depth testing is disabled
if (stencil_action_enable)
UpdateStencil(stencil_test.action_depth_pass);
auto dest = GetPixel(x >> 4, y >> 4);
Math::Vec4<u8> blend_output = combiner_output;
if (output_merger.alphablend_enable) {
auto params = output_merger.alpha_blending;
auto LookupFactor = [&](unsigned channel,
FramebufferRegs::BlendFactor factor) -> u8 {
DEBUG_ASSERT(channel < 4);
const Math::Vec4<u8> blend_const = {
static_cast<u8>(output_merger.blend_const.r),
static_cast<u8>(output_merger.blend_const.g),
static_cast<u8>(output_merger.blend_const.b),
static_cast<u8>(output_merger.blend_const.a),
};
switch (factor) {
case FramebufferRegs::BlendFactor::Zero:
return 0;
case FramebufferRegs::BlendFactor::One:
return 255;
case FramebufferRegs::BlendFactor::SourceColor:
return combiner_output[channel];
case FramebufferRegs::BlendFactor::OneMinusSourceColor:
return 255 - combiner_output[channel];
case FramebufferRegs::BlendFactor::DestColor:
return dest[channel];
case FramebufferRegs::BlendFactor::OneMinusDestColor:
return 255 - dest[channel];
case FramebufferRegs::BlendFactor::SourceAlpha:
return combiner_output.a();
case FramebufferRegs::BlendFactor::OneMinusSourceAlpha:
return 255 - combiner_output.a();
case FramebufferRegs::BlendFactor::DestAlpha:
return dest.a();
case FramebufferRegs::BlendFactor::OneMinusDestAlpha:
return 255 - dest.a();
case FramebufferRegs::BlendFactor::ConstantColor:
return blend_const[channel];
case FramebufferRegs::BlendFactor::OneMinusConstantColor:
return 255 - blend_const[channel];
case FramebufferRegs::BlendFactor::ConstantAlpha:
return blend_const.a();
case FramebufferRegs::BlendFactor::OneMinusConstantAlpha:
return 255 - blend_const.a();
case FramebufferRegs::BlendFactor::SourceAlphaSaturate:
// Returns 1.0 for the alpha channel
if (channel == 3)
return 255;
return std::min(combiner_output.a(), static_cast<u8>(255 - dest.a()));
default:
LOG_CRITICAL(HW_GPU, "Unknown blend factor %x", factor);
UNIMPLEMENTED();
break;
}
return combiner_output[channel];
};
auto srcfactor = Math::MakeVec(LookupFactor(0, params.factor_source_rgb),
LookupFactor(1, params.factor_source_rgb),
LookupFactor(2, params.factor_source_rgb),
LookupFactor(3, params.factor_source_a));
auto dstfactor = Math::MakeVec(LookupFactor(0, params.factor_dest_rgb),
LookupFactor(1, params.factor_dest_rgb),
LookupFactor(2, params.factor_dest_rgb),
LookupFactor(3, params.factor_dest_a));
blend_output = EvaluateBlendEquation(combiner_output, srcfactor, dest, dstfactor,
params.blend_equation_rgb);
blend_output.a() = EvaluateBlendEquation(combiner_output, srcfactor, dest,
dstfactor, params.blend_equation_a)
.a();
} else {
blend_output =
Math::MakeVec(LogicOp(combiner_output.r(), dest.r(), output_merger.logic_op),
LogicOp(combiner_output.g(), dest.g(), output_merger.logic_op),
LogicOp(combiner_output.b(), dest.b(), output_merger.logic_op),
LogicOp(combiner_output.a(), dest.a(), output_merger.logic_op));
}
const Math::Vec4<u8> result = {
output_merger.red_enable ? blend_output.r() : dest.r(),
output_merger.green_enable ? blend_output.g() : dest.g(),
output_merger.blue_enable ? blend_output.b() : dest.b(),
output_merger.alpha_enable ? blend_output.a() : dest.a(),
};
if (regs.framebuffer.framebuffer.allow_color_write != 0)
DrawPixel(x >> 4, y >> 4, result);
}
}
}
void ProcessTriangle(const Vertex& v0, const Vertex& v1, const Vertex& v2) {
ProcessTriangleInternal(v0, v1, v2);
}
} // namespace Rasterizer
} // namespace Pica

48
src/video_core/swrasterizer/rasterizer.h
View file

@ -1,48 +0,0 @@
// Copyright 2014 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include "video_core/shader/shader.h"
namespace Pica {
namespace Rasterizer {
struct Vertex : Shader::OutputVertex {
Vertex(const OutputVertex& v) : OutputVertex(v) {}
// Attributes used to store intermediate results
// position after perspective divide
Math::Vec3<float24> screenpos;
// Linear interpolation
// factor: 0=this, 1=vtx
// Note: This function cannot be called after perspective divide
void Lerp(float24 factor, const Vertex& vtx) {
pos = pos * factor + vtx.pos * (float24::FromFloat32(1) - factor);
quat = quat * factor + vtx.quat * (float24::FromFloat32(1) - factor);
color = color * factor + vtx.color * (float24::FromFloat32(1) - factor);
tc0 = tc0 * factor + vtx.tc0 * (float24::FromFloat32(1) - factor);
tc1 = tc1 * factor + vtx.tc1 * (float24::FromFloat32(1) - factor);
tc0_w = tc0_w * factor + vtx.tc0_w * (float24::FromFloat32(1) - factor);
view = view * factor + vtx.view * (float24::FromFloat32(1) - factor);
tc2 = tc2 * factor + vtx.tc2 * (float24::FromFloat32(1) - factor);
}
// Linear interpolation
// factor: 0=v0, 1=v1
// Note: This function cannot be called after perspective divide
static Vertex Lerp(float24 factor, const Vertex& v0, const Vertex& v1) {
Vertex ret = v0;
ret.Lerp(factor, v1);
return ret;
}
};
void ProcessTriangle(const Vertex& v0, const Vertex& v1, const Vertex& v2);
} // namespace Rasterizer
} // namespace Pica

15
src/video_core/swrasterizer/swrasterizer.cpp
View file

@ -1,15 +0,0 @@
// Copyright 2015 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "video_core/swrasterizer/clipper.h"
#include "video_core/swrasterizer/swrasterizer.h"
namespace VideoCore {
void SWRasterizer::AddTriangle(const Pica::Shader::OutputVertex& v0,
const Pica::Shader::OutputVertex& v1,
const Pica::Shader::OutputVertex& v2) {
Pica::Clipper::ProcessTriangle(v0, v1, v2);
}
}

27
src/video_core/swrasterizer/swrasterizer.h
View file

@ -1,27 +0,0 @@
// Copyright 2015 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include "common/common_types.h"
#include "video_core/rasterizer_interface.h"
namespace Pica {
namespace Shader {
struct OutputVertex;
}
}
namespace VideoCore {
class SWRasterizer : public RasterizerInterface {
void AddTriangle(const Pica::Shader::OutputVertex& v0, const Pica::Shader::OutputVertex& v1,
const Pica::Shader::OutputVertex& v2) override;
void DrawTriangles() override {}
void NotifyPicaRegisterChanged(u32 id) override {}
void FlushAll() override {}
void FlushRegion(PAddr addr, u64 size) override {}
void FlushAndInvalidateRegion(PAddr addr, u64 size) override {}
};
}

244
src/video_core/swrasterizer/texturing.cpp
View file

@ -1,244 +0,0 @@
// Copyright 2017 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <algorithm>
#include "common/assert.h"
#include "common/common_types.h"
#include "common/math_util.h"
#include "common/vector_math.h"
#include "video_core/regs_texturing.h"
#include "video_core/swrasterizer/texturing.h"
namespace Pica {
namespace Rasterizer {
using TevStageConfig = TexturingRegs::TevStageConfig;
int GetWrappedTexCoord(TexturingRegs::TextureConfig::WrapMode mode, int val, unsigned size) {
switch (mode) {
case TexturingRegs::TextureConfig::ClampToEdge2:
// For negative coordinate, ClampToEdge2 behaves the same as Repeat
if (val < 0) {
return static_cast<int>(static_cast<unsigned>(val) % size);
}
// [[fallthrough]]
case TexturingRegs::TextureConfig::ClampToEdge:
val = std::max(val, 0);
val = std::min(val, static_cast<int>(size) - 1);
return val;
case TexturingRegs::TextureConfig::ClampToBorder:
return val;
case TexturingRegs::TextureConfig::ClampToBorder2:
// For ClampToBorder2, the case of positive coordinate beyond the texture size is already
// handled outside. Here we only handle the negative coordinate in the same way as Repeat.
case TexturingRegs::TextureConfig::Repeat2:
case TexturingRegs::TextureConfig::Repeat3:
case TexturingRegs::TextureConfig::Repeat:
return static_cast<int>(static_cast<unsigned>(val) % size);
case TexturingRegs::TextureConfig::MirroredRepeat: {
unsigned int coord = (static_cast<unsigned>(val) % (2 * size));
if (coord >= size)
coord = 2 * size - 1 - coord;
return static_cast<int>(coord);
}
default:
LOG_ERROR(HW_GPU, "Unknown texture coordinate wrapping mode %x", (int)mode);
UNIMPLEMENTED();
return 0;
}
};
Math::Vec3<u8> GetColorModifier(TevStageConfig::ColorModifier factor,
const Math::Vec4<u8>& values) {
using ColorModifier = TevStageConfig::ColorModifier;
switch (factor) {
case ColorModifier::SourceColor:
return values.rgb();
case ColorModifier::OneMinusSourceColor:
return (Math::Vec3<u8>(255, 255, 255) - values.rgb()).Cast<u8>();
case ColorModifier::SourceAlpha:
return values.aaa();
case ColorModifier::OneMinusSourceAlpha:
return (Math::Vec3<u8>(255, 255, 255) - values.aaa()).Cast<u8>();
case ColorModifier::SourceRed:
return values.rrr();
case ColorModifier::OneMinusSourceRed:
return (Math::Vec3<u8>(255, 255, 255) - values.rrr()).Cast<u8>();
case ColorModifier::SourceGreen:
return values.ggg();
case ColorModifier::OneMinusSourceGreen:
return (Math::Vec3<u8>(255, 255, 255) - values.ggg()).Cast<u8>();
case ColorModifier::SourceBlue:
return values.bbb();
case ColorModifier::OneMinusSourceBlue:
return (Math::Vec3<u8>(255, 255, 255) - values.bbb()).Cast<u8>();
}
UNREACHABLE();
};
u8 GetAlphaModifier(TevStageConfig::AlphaModifier factor, const Math::Vec4<u8>& values) {
using AlphaModifier = TevStageConfig::AlphaModifier;
switch (factor) {
case AlphaModifier::SourceAlpha:
return values.a();
case AlphaModifier::OneMinusSourceAlpha:
return 255 - values.a();
case AlphaModifier::SourceRed:
return values.r();
case AlphaModifier::OneMinusSourceRed:
return 255 - values.r();
case AlphaModifier::SourceGreen:
return values.g();
case AlphaModifier::OneMinusSourceGreen:
return 255 - values.g();
case AlphaModifier::SourceBlue:
return values.b();
case AlphaModifier::OneMinusSourceBlue:
return 255 - values.b();
}
UNREACHABLE();
};
Math::Vec3<u8> ColorCombine(TevStageConfig::Operation op, const Math::Vec3<u8> input[3]) {
using Operation = TevStageConfig::Operation;
switch (op) {
case Operation::Replace:
return input[0];
case Operation::Modulate:
return ((input[0] * input[1]) / 255).Cast<u8>();
case Operation::Add: {
auto result = input[0] + input[1];
result.r() = std::min(255, result.r());
result.g() = std::min(255, result.g());
result.b() = std::min(255, result.b());
return result.Cast<u8>();
}
case Operation::AddSigned: {
// TODO(bunnei): Verify that the color conversion from (float) 0.5f to
// (byte) 128 is correct
auto result =
input[0].Cast<int>() + input[1].Cast<int>() - Math::MakeVec<int>(128, 128, 128);
result.r() = MathUtil::Clamp<int>(result.r(), 0, 255);
result.g() = MathUtil::Clamp<int>(result.g(), 0, 255);
result.b() = MathUtil::Clamp<int>(result.b(), 0, 255);
return result.Cast<u8>();
}
case Operation::Lerp:
return ((input[0] * input[2] +
input[1] * (Math::MakeVec<u8>(255, 255, 255) - input[2]).Cast<u8>()) /
255)
.Cast<u8>();
case Operation::Subtract: {
auto result = input[0].Cast<int>() - input[1].Cast<int>();
result.r() = std::max(0, result.r());
result.g() = std::max(0, result.g());
result.b() = std::max(0, result.b());
return result.Cast<u8>();
}
case Operation::MultiplyThenAdd: {
auto result = (input[0] * input[1] + 255 * input[2].Cast<int>()) / 255;
result.r() = std::min(255, result.r());
result.g() = std::min(255, result.g());
result.b() = std::min(255, result.b());
return result.Cast<u8>();
}
case Operation::AddThenMultiply: {
auto result = input[0] + input[1];
result.r() = std::min(255, result.r());
result.g() = std::min(255, result.g());
result.b() = std::min(255, result.b());
result = (result * input[2].Cast<int>()) / 255;
return result.Cast<u8>();
}
case Operation::Dot3_RGB:
case Operation::Dot3_RGBA: {
// Not fully accurate. Worst case scenario seems to yield a +/-3 error. Some HW results
// indicate that the per-component computation can't have a higher precision than 1/256,
// while dot3_rgb((0x80,g0,b0), (0x7F,g1,b1)) and dot3_rgb((0x80,g0,b0), (0x80,g1,b1)) give
// different results.
int result = ((input[0].r() * 2 - 255) * (input[1].r() * 2 - 255) + 128) / 256 +
((input[0].g() * 2 - 255) * (input[1].g() * 2 - 255) + 128) / 256 +
((input[0].b() * 2 - 255) * (input[1].b() * 2 - 255) + 128) / 256;
result = std::max(0, std::min(255, result));
return {(u8)result, (u8)result, (u8)result};
}
default:
LOG_ERROR(HW_GPU, "Unknown color combiner operation %d", (int)op);
UNIMPLEMENTED();
return {0, 0, 0};
}
};
u8 AlphaCombine(TevStageConfig::Operation op, const std::array<u8, 3>& input) {
switch (op) {
using Operation = TevStageConfig::Operation;
case Operation::Replace:
return input[0];
case Operation::Modulate:
return input[0] * input[1] / 255;
case Operation::Add:
return std::min(255, input[0] + input[1]);
case Operation::AddSigned: {
// TODO(bunnei): Verify that the color conversion from (float) 0.5f to (byte) 128 is correct
auto result = static_cast<int>(input[0]) + static_cast<int>(input[1]) - 128;
return static_cast<u8>(MathUtil::Clamp<int>(result, 0, 255));
}
case Operation::Lerp:
return (input[0] * input[2] + input[1] * (255 - input[2])) / 255;
case Operation::Subtract:
return std::max(0, (int)input[0] - (int)input[1]);
case Operation::MultiplyThenAdd:
return std::min(255, (input[0] * input[1] + 255 * input[2]) / 255);
case Operation::AddThenMultiply:
return (std::min(255, (input[0] + input[1])) * input[2]) / 255;
default:
LOG_ERROR(HW_GPU, "Unknown alpha combiner operation %d", (int)op);
UNIMPLEMENTED();
return 0;
}
};
} // namespace Rasterizer
} // namespace Pica

28
src/video_core/swrasterizer/texturing.h
View file

@ -1,28 +0,0 @@
// Copyright 2017 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include "common/common_types.h"
#include "common/vector_math.h"
#include "video_core/regs_texturing.h"
namespace Pica {
namespace Rasterizer {
int GetWrappedTexCoord(TexturingRegs::TextureConfig::WrapMode mode, int val, unsigned size);
Math::Vec3<u8> GetColorModifier(TexturingRegs::TevStageConfig::ColorModifier factor,
const Math::Vec4<u8>& values);
u8 GetAlphaModifier(TexturingRegs::TevStageConfig::AlphaModifier factor,
const Math::Vec4<u8>& values);
Math::Vec3<u8> ColorCombine(TexturingRegs::TevStageConfig::Operation op,
const Math::Vec3<u8> input[3]);
u8 AlphaCombine(TexturingRegs::TevStageConfig::Operation op, const std::array<u8, 3>& input);
} // namespace Rasterizer
} // namespace Pica

122
src/video_core/texture/etc1.cpp
View file

@ -1,122 +0,0 @@
// Copyright 2017 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <array>
#include "common/bit_field.h"
#include "common/color.h"
#include "common/common_types.h"
#include "common/math_util.h"
#include "common/vector_math.h"
#include "video_core/texture/etc1.h"
namespace Pica {
namespace Texture {
namespace {
constexpr std::array<std::array<u8, 2>, 8> etc1_modifier_table = {{
{2, 8}, {5, 17}, {9, 29}, {13, 42}, {18, 60}, {24, 80}, {33, 106}, {47, 183},
}};
union ETC1Tile {
u64 raw;
// Each of these two is a collection of 16 bits (one per lookup value)
BitField<0, 16, u64> table_subindexes;
BitField<16, 16, u64> negation_flags;
unsigned GetTableSubIndex(unsigned index) const {
return (table_subindexes >> index) & 1;
}
bool GetNegationFlag(unsigned index) const {
return ((negation_flags >> index) & 1) == 1;
}
BitField<32, 1, u64> flip;
BitField<33, 1, u64> differential_mode;
BitField<34, 3, u64> table_index_2;
BitField<37, 3, u64> table_index_1;
union {
// delta value + base value
BitField<40, 3, s64> db;
BitField<43, 5, u64> b;
BitField<48, 3, s64> dg;
BitField<51, 5, u64> g;
BitField<56, 3, s64> dr;
BitField<59, 5, u64> r;
} differential;
union {
BitField<40, 4, u64> b2;
BitField<44, 4, u64> b1;
BitField<48, 4, u64> g2;
BitField<52, 4, u64> g1;
BitField<56, 4, u64> r2;
BitField<60, 4, u64> r1;
} separate;
const Math::Vec3<u8> GetRGB(unsigned int x, unsigned int y) const {
int texel = 4 * x + y;
if (flip)
std::swap(x, y);
// Lookup base value
Math::Vec3<int> ret;
if (differential_mode) {
ret.r() = static_cast<int>(differential.r);
ret.g() = static_cast<int>(differential.g);
ret.b() = static_cast<int>(differential.b);
if (x >= 2) {
ret.r() += static_cast<int>(differential.dr);
ret.g() += static_cast<int>(differential.dg);
ret.b() += static_cast<int>(differential.db);
}
ret.r() = Color::Convert5To8(ret.r());
ret.g() = Color::Convert5To8(ret.g());
ret.b() = Color::Convert5To8(ret.b());
} else {
if (x < 2) {
ret.r() = Color::Convert4To8(static_cast<u8>(separate.r1));
ret.g() = Color::Convert4To8(static_cast<u8>(separate.g1));
ret.b() = Color::Convert4To8(static_cast<u8>(separate.b1));
} else {
ret.r() = Color::Convert4To8(static_cast<u8>(separate.r2));
ret.g() = Color::Convert4To8(static_cast<u8>(separate.g2));
ret.b() = Color::Convert4To8(static_cast<u8>(separate.b2));
}
}
// Add modifier
unsigned table_index =
static_cast<int>((x < 2) ? table_index_1.Value() : table_index_2.Value());
int modifier = etc1_modifier_table[table_index][GetTableSubIndex(texel)];
if (GetNegationFlag(texel))
modifier *= -1;
ret.r() = MathUtil::Clamp(ret.r() + modifier, 0, 255);
ret.g() = MathUtil::Clamp(ret.g() + modifier, 0, 255);
ret.b() = MathUtil::Clamp(ret.b() + modifier, 0, 255);
return ret.Cast<u8>();
}
};
} // anonymous namespace
Math::Vec3<u8> SampleETC1Subtile(u64 value, unsigned int x, unsigned int y) {
ETC1Tile tile{value};
return tile.GetRGB(x, y);
}
} // namespace Texture
} // namespace Pica

16
src/video_core/texture/etc1.h
View file

@ -1,16 +0,0 @@
// Copyright 2017 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include "common/common_types.h"
#include "common/vector_math.h"
namespace Pica {
namespace Texture {
Math::Vec3<u8> SampleETC1Subtile(u64 value, unsigned int x, unsigned int y);
} // namespace Texture
} // namespace Pica

227
src/video_core/texture/texture_decode.cpp
View file

@ -1,227 +0,0 @@
// Copyright 2017 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/assert.h"
#include "common/color.h"
#include "common/logging/log.h"
#include "common/math_util.h"
#include "common/swap.h"
#include "common/vector_math.h"
#include "video_core/regs_texturing.h"
#include "video_core/texture/etc1.h"
#include "video_core/texture/texture_decode.h"
#include "video_core/utils.h"
using TextureFormat = Pica::TexturingRegs::TextureFormat;
namespace Pica {
namespace Texture {
constexpr size_t TILE_SIZE = 8 * 8;
constexpr size_t ETC1_SUBTILES = 2 * 2;
size_t CalculateTileSize(TextureFormat format) {
switch (format) {
case TextureFormat::RGBA8:
return 4 * TILE_SIZE;
case TextureFormat::RGB8:
return 3 * TILE_SIZE;
case TextureFormat::RGB5A1:
case TextureFormat::RGB565:
case TextureFormat::RGBA4:
case TextureFormat::IA8:
case TextureFormat::RG8:
return 2 * TILE_SIZE;
case TextureFormat::I8:
case TextureFormat::A8:
case TextureFormat::IA4:
return 1 * TILE_SIZE;
case TextureFormat::I4:
case TextureFormat::A4:
return TILE_SIZE / 2;
case TextureFormat::ETC1:
return ETC1_SUBTILES * 8;
case TextureFormat::ETC1A4:
return ETC1_SUBTILES * 16;
default: // placeholder for yet unknown formats
UNIMPLEMENTED();
return 0;
}
}
Math::Vec4<u8> LookupTexture(const u8* source, unsigned int x, unsigned int y,
const TextureInfo& info, bool disable_alpha) {
// Coordinate in tiles
const unsigned int coarse_x = x / 8;
const unsigned int coarse_y = y / 8;
// Coordinate inside the tile
const unsigned int fine_x = x % 8;
const unsigned int fine_y = y % 8;
const u8* line = source + coarse_y * info.stride;
const u8* tile = line + coarse_x * CalculateTileSize(info.format);
return LookupTexelInTile(tile, fine_x, fine_y, info, disable_alpha);
}
Math::Vec4<u8> LookupTexelInTile(const u8* source, unsigned int x, unsigned int y,
const TextureInfo& info, bool disable_alpha) {
DEBUG_ASSERT(x < 8);
DEBUG_ASSERT(y < 8);
using VideoCore::MortonInterleave;
switch (info.format) {
case TextureFormat::RGBA8: {
auto res = Color::DecodeRGBA8(source + MortonInterleave(x, y) * 4);
return {res.r(), res.g(), res.b(), static_cast<u8>(disable_alpha ? 255 : res.a())};
}
case TextureFormat::RGB8: {
auto res = Color::DecodeRGB8(source + MortonInterleave(x, y) * 3);
return {res.r(), res.g(), res.b(), 255};
}
case TextureFormat::RGB5A1: {
auto res = Color::DecodeRGB5A1(source + MortonInterleave(x, y) * 2);
return {res.r(), res.g(), res.b(), static_cast<u8>(disable_alpha ? 255 : res.a())};
}
case TextureFormat::RGB565: {
auto res = Color::DecodeRGB565(source + MortonInterleave(x, y) * 2);
return {res.r(), res.g(), res.b(), 255};
}
case TextureFormat::RGBA4: {
auto res = Color::DecodeRGBA4(source + MortonInterleave(x, y) * 2);
return {res.r(), res.g(), res.b(), static_cast<u8>(disable_alpha ? 255 : res.a())};
}
case TextureFormat::IA8: {
const u8* source_ptr = source + MortonInterleave(x, y) * 2;
if (disable_alpha) {
// Show intensity as red, alpha as green
return {source_ptr[1], source_ptr[0], 0, 255};
} else {
return {source_ptr[1], source_ptr[1], source_ptr[1], source_ptr[0]};
}
}
case TextureFormat::RG8: {
auto res = Color::DecodeRG8(source + MortonInterleave(x, y) * 2);
return {res.r(), res.g(), 0, 255};
}
case TextureFormat::I8: {
const u8* source_ptr = source + MortonInterleave(x, y);
return {*source_ptr, *source_ptr, *source_ptr, 255};
}
case TextureFormat::A8: {
const u8* source_ptr = source + MortonInterleave(x, y);
if (disable_alpha) {
return {*source_ptr, *source_ptr, *source_ptr, 255};
} else {
return {0, 0, 0, *source_ptr};
}
}
case TextureFormat::IA4: {
const u8* source_ptr = source + MortonInterleave(x, y);
u8 i = Color::Convert4To8(((*source_ptr) & 0xF0) >> 4);
u8 a = Color::Convert4To8((*source_ptr) & 0xF);
if (disable_alpha) {
// Show intensity as red, alpha as green
return {i, a, 0, 255};
} else {
return {i, i, i, a};
}
}
case TextureFormat::I4: {
u32 morton_offset = MortonInterleave(x, y);
const u8* source_ptr = source + morton_offset / 2;
u8 i = (morton_offset % 2) ? ((*source_ptr & 0xF0) >> 4) : (*source_ptr & 0xF);
i = Color::Convert4To8(i);
return {i, i, i, 255};
}
case TextureFormat::A4: {
u32 morton_offset = MortonInterleave(x, y);
const u8* source_ptr = source + morton_offset / 2;
u8 a = (morton_offset % 2) ? ((*source_ptr & 0xF0) >> 4) : (*source_ptr & 0xF);
a = Color::Convert4To8(a);
if (disable_alpha) {
return {a, a, a, 255};
} else {
return {0, 0, 0, a};
}
}
case TextureFormat::ETC1:
case TextureFormat::ETC1A4: {
bool has_alpha = (info.format == TextureFormat::ETC1A4);
size_t subtile_size = has_alpha ? 16 : 8;
// ETC1 further subdivides each 8x8 tile into four 4x4 subtiles
constexpr unsigned int subtile_width = 4;
constexpr unsigned int subtile_height = 4;
unsigned int subtile_index = (x / subtile_width) + 2 * (y / subtile_height);
x %= subtile_width;
y %= subtile_height;
const u8* subtile_ptr = source + subtile_index * subtile_size;
u8 alpha = 255;
if (has_alpha) {
u64_le packed_alpha;
memcpy(&packed_alpha, subtile_ptr, sizeof(u64));
subtile_ptr += sizeof(u64);
alpha = Color::Convert4To8((packed_alpha >> (4 * (x * subtile_width + y))) & 0xF);
}
u64_le subtile_data;
memcpy(&subtile_data, subtile_ptr, sizeof(u64));
return Math::MakeVec(SampleETC1Subtile(subtile_data, x, y),
disable_alpha ? (u8)255 : alpha);
}
default:
LOG_ERROR(HW_GPU, "Unknown texture format: %x", (u32)info.format);
DEBUG_ASSERT(false);
return {};
}
}
TextureInfo TextureInfo::FromPicaRegister(const TexturingRegs::TextureConfig& config,
const TexturingRegs::TextureFormat& format) {
TextureInfo info;
info.physical_address = config.GetPhysicalAddress();
info.width = config.width;
info.height = config.height;
info.format = format;
info.SetDefaultStride();
return info;
}
} // namespace Texture
} // namespace Pica

60
src/video_core/texture/texture_decode.h
View file

@ -1,60 +0,0 @@
// Copyright 2017 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include "common/common_types.h"
#include "common/vector_math.h"
#include "video_core/regs_texturing.h"
namespace Pica {
namespace Texture {
/// Returns the byte size of a 8*8 tile of the specified texture format.
size_t CalculateTileSize(TexturingRegs::TextureFormat format);
struct TextureInfo {
PAddr physical_address;
unsigned int width;
unsigned int height;
ptrdiff_t stride;
TexturingRegs::TextureFormat format;
static TextureInfo FromPicaRegister(const TexturingRegs::TextureConfig& config,
const TexturingRegs::TextureFormat& format);
/// Calculates stride from format and width, assuming that the entire texture is contiguous.
void SetDefaultStride() {
stride = CalculateTileSize(format) * (width / 8);
}
};
/**
* Lookup texel located at the given coordinates and return an RGBA vector of its color.
* @param source Source pointer to read data from
* @param x,y Texture coordinates to read from
* @param info TextureInfo object describing the texture setup
* @param disable_alpha This is used for debug widgets which use this method to display textures
* without providing a good way to visualize alpha by themselves. If true, this will return 255 for
* the alpha component, and either drop the information entirely or store it in an "unused" color
* channel.
* @todo Eventually we should get rid of the disable_alpha parameter.
*/
Math::Vec4<u8> LookupTexture(const u8* source, unsigned int x, unsigned int y,
const TextureInfo& info, bool disable_alpha = false);
/**
* Looks up a texel from a single 8x8 texture tile.
*
* @param source Pointer to the beginning of the tile.
* @param x, y In-tile coordinates to read from. Must be < 8.
* @param info TextureInfo describing the texture format.
* @param disable_alpha Used for debugging. Sets the result alpha to 255 and either discards the
* real alpha or inserts it in an otherwise unused channel.
*/
Math::Vec4<u8> LookupTexelInTile(const u8* source, unsigned int x, unsigned int y,
const TextureInfo& info, bool disable_alpha);
} // namespace Texture
} // namespace Pica

160
src/video_core/vertex_loader.cpp
View file

@ -1,160 +0,0 @@
#include <memory>
#include <boost/range/algorithm/fill.hpp>
#include "common/alignment.h"
#include "common/assert.h"
#include "common/bit_field.h"
#include "common/common_types.h"
#include "common/logging/log.h"
#include "common/vector_math.h"
#include "core/memory.h"
#include "video_core/debug_utils/debug_utils.h"
#include "video_core/pica_state.h"
#include "video_core/pica_types.h"
#include "video_core/regs_pipeline.h"
#include "video_core/shader/shader.h"
#include "video_core/vertex_loader.h"
namespace Pica {
void VertexLoader::Setup(const PipelineRegs& regs) {
ASSERT_MSG(!is_setup, "VertexLoader is not intended to be setup more than once.");
const auto& attribute_config = regs.vertex_attributes;
num_total_attributes = attribute_config.GetNumTotalAttributes();
boost::fill(vertex_attribute_sources, 0xdeadbeef);
for (int i = 0; i < 16; i++) {
vertex_attribute_is_default[i] = attribute_config.IsDefaultAttribute(i);
}
// Setup attribute data from loaders
for (int loader = 0; loader < 12; ++loader) {
const auto& loader_config = attribute_config.attribute_loaders[loader];
u32 offset = 0;
// TODO: What happens if a loader overwrites a previous one's data?
for (unsigned component = 0; component < loader_config.component_count; ++component) {
if (component >= 12) {
LOG_ERROR(HW_GPU,
"Overflow in the vertex attribute loader %u trying to load component %u",
loader, component);
continue;
}
u32 attribute_index = loader_config.GetComponent(component);
if (attribute_index < 12) {
offset = Common::AlignUp(offset,
attribute_config.GetElementSizeInBytes(attribute_index));
vertex_attribute_sources[attribute_index] = loader_config.data_offset + offset;
vertex_attribute_strides[attribute_index] =
static_cast<u32>(loader_config.byte_count);
vertex_attribute_formats[attribute_index] =
attribute_config.GetFormat(attribute_index);
vertex_attribute_elements[attribute_index] =
attribute_config.GetNumElements(attribute_index);
offset += attribute_config.GetStride(attribute_index);
} else if (attribute_index < 16) {
// Attribute ids 12, 13, 14 and 15 signify 4, 8, 12 and 16-byte paddings,
// respectively
offset = Common::AlignUp(offset, 4);
offset += (attribute_index - 11) * 4;
} else {
UNREACHABLE(); // This is truly unreachable due to the number of bits for each
// component
}
}
}
is_setup = true;
}
void VertexLoader::LoadVertex(u32 base_address, int index, int vertex,
Shader::AttributeBuffer& input,
DebugUtils::MemoryAccessTracker& memory_accesses) {
ASSERT_MSG(is_setup, "A VertexLoader needs to be setup before loading vertices.");
for (int i = 0; i < num_total_attributes; ++i) {
if (vertex_attribute_elements[i] != 0) {
// Load per-vertex data from the loader arrays
u32 source_addr =
base_address + vertex_attribute_sources[i] + vertex_attribute_strides[i] * vertex;
if (g_debug_context && Pica::g_debug_context->recorder) {
memory_accesses.AddAccess(
source_addr,
vertex_attribute_elements[i] *
((vertex_attribute_formats[i] == PipelineRegs::VertexAttributeFormat::FLOAT)
? 4
: (vertex_attribute_formats[i] ==
PipelineRegs::VertexAttributeFormat::SHORT)
? 2
: 1));
}
switch (vertex_attribute_formats[i]) {
case PipelineRegs::VertexAttributeFormat::BYTE: {
const s8* srcdata =
reinterpret_cast<const s8*>(Memory::GetPhysicalPointer(source_addr));
for (unsigned int comp = 0; comp < vertex_attribute_elements[i]; ++comp) {
input.attr[i][comp] = float24::FromFloat32(srcdata[comp]);
}
break;
}
case PipelineRegs::VertexAttributeFormat::UBYTE: {
const u8* srcdata =
reinterpret_cast<const u8*>(Memory::GetPhysicalPointer(source_addr));
for (unsigned int comp = 0; comp < vertex_attribute_elements[i]; ++comp) {
input.attr[i][comp] = float24::FromFloat32(srcdata[comp]);
}
break;
}
case PipelineRegs::VertexAttributeFormat::SHORT: {
const s16* srcdata =
reinterpret_cast<const s16*>(Memory::GetPhysicalPointer(source_addr));
for (unsigned int comp = 0; comp < vertex_attribute_elements[i]; ++comp) {
input.attr[i][comp] = float24::FromFloat32(srcdata[comp]);
}
break;
}
case PipelineRegs::VertexAttributeFormat::FLOAT: {
const float* srcdata =
reinterpret_cast<const float*>(Memory::GetPhysicalPointer(source_addr));
for (unsigned int comp = 0; comp < vertex_attribute_elements[i]; ++comp) {
input.attr[i][comp] = float24::FromFloat32(srcdata[comp]);
}
break;
}
}
// Default attribute values set if array elements have < 4 components. This
// is *not* carried over from the default attribute settings even if they're
// enabled for this attribute.
for (unsigned int comp = vertex_attribute_elements[i]; comp < 4; ++comp) {
input.attr[i][comp] =
comp == 3 ? float24::FromFloat32(1.0f) : float24::FromFloat32(0.0f);
}
LOG_TRACE(HW_GPU, "Loaded %d components of attribute %x for vertex %x (index %x) from "
"0x%08x + 0x%08x + 0x%04x: %f %f %f %f",
vertex_attribute_elements[i], i, vertex, index, base_address,
vertex_attribute_sources[i], vertex_attribute_strides[i] * vertex,
input.attr[i][0].ToFloat32(), input.attr[i][1].ToFloat32(),
input.attr[i][2].ToFloat32(), input.attr[i][3].ToFloat32());
} else if (vertex_attribute_is_default[i]) {
// Load the default attribute if we're configured to do so
input.attr[i] = g_state.input_default_attributes.attr[i];
LOG_TRACE(HW_GPU,
"Loaded default attribute %x for vertex %x (index %x): (%f, %f, %f, %f)", i,
vertex, index, input.attr[i][0].ToFloat32(), input.attr[i][1].ToFloat32(),
input.attr[i][2].ToFloat32(), input.attr[i][3].ToFloat32());
} else {
// TODO(yuriks): In this case, no data gets loaded and the vertex
// remains with the last value it had. This isn't currently maintained
// as global state, however, and so won't work in Citra yet.
}
}
}
} // namespace Pica

42
src/video_core/vertex_loader.h
View file

@ -1,42 +0,0 @@
#pragma once
#include <array>
#include "common/common_types.h"
#include "video_core/regs_pipeline.h"
namespace Pica {
namespace DebugUtils {
class MemoryAccessTracker;
}
namespace Shader {
struct AttributeBuffer;
}
class VertexLoader {
public:
VertexLoader() = default;
explicit VertexLoader(const PipelineRegs& regs) {
Setup(regs);
}
void Setup(const PipelineRegs& regs);
void LoadVertex(u32 base_address, int index, int vertex, Shader::AttributeBuffer& input,
DebugUtils::MemoryAccessTracker& memory_accesses);
int GetNumTotalAttributes() const {
return num_total_attributes;
}
private:
std::array<u32, 16> vertex_attribute_sources;
std::array<u32, 16> vertex_attribute_strides{};
std::array<PipelineRegs::VertexAttributeFormat, 16> vertex_attribute_formats;
std::array<u32, 16> vertex_attribute_elements{};
std::array<bool, 16> vertex_attribute_is_default;
int num_total_attributes = 0;
bool is_setup = false;
};
} // namespace Pica

5
src/video_core/video_core.cpp
View file

@ -4,7 +4,6 @@
#include <memory>
#include "common/logging/log.h"
#include "video_core/pica.h"
#include "video_core/renderer_base.h"
#include "video_core/renderer_opengl/renderer_opengl.h"
#include "video_core/video_core.h"
@ -24,8 +23,6 @@ std::atomic<bool> g_toggle_framelimit_enabled;
/// Initialize the video core
bool Init(EmuWindow* emu_window) {
Pica::Init();
g_emu_window = emu_window;
g_renderer = std::make_unique<RendererOpenGL>();
g_renderer->SetWindow(g_emu_window);
@ -40,8 +37,6 @@ bool Init(EmuWindow* emu_window) {
/// Shutdown the video core
void Shutdown() {
Pica::Shutdown();
g_renderer.reset();
LOG_DEBUG(Render, "shutdown OK");