ahmed2007/suyu
1
0
Fork 0
forked from suyu/suyu
Code Pull requests Activity

Introduce skeleton of the GPU Compute Engine.

Browse source
This commit is contained in:
Fernando Sahmkow 2019-04-22 19:05:43 -04:00
parent a91d3fc639
commit e4ff140b99
3 changed files with 202 additions and 8 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

37
src/video_core/engines/kepler_compute.cpp
View file

@ -4,12 +4,21 @@
#include "common/assert.h"
#include "common/logging/log.h"
#include "core/core.h"
#include "video_core/engines/kepler_compute.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/memory_manager.h"
#include "video_core/rasterizer_interface.h"
#include "video_core/renderer_base.h"
#include "video_core/textures/decoders.h"
namespace Tegra::Engines {
KeplerCompute::KeplerCompute(MemoryManager& memory_manager) : memory_manager{memory_manager} {}
KeplerCompute::KeplerCompute(Core::System& system, VideoCore::RasterizerInterface& rasterizer,
MemoryManager& memory_manager)
: system{system}, rasterizer{rasterizer}, memory_manager{memory_manager}, upload_state{
memory_manager,
regs.upload} {}
KeplerCompute::~KeplerCompute() = default;
@ -20,14 +29,34 @@ void KeplerCompute::CallMethod(const GPU::MethodCall& method_call) {
regs.reg_array[method_call.method] = method_call.argument;
switch (method_call.method) {
case KEPLER_COMPUTE_REG_INDEX(exec_upload): {
upload_state.ProcessExec(regs.exec_upload.linear != 0);
break;
}
case KEPLER_COMPUTE_REG_INDEX(data_upload): {
bool is_last_call = method_call.IsLastCall();
upload_state.ProcessData(method_call.argument, is_last_call);
if (is_last_call) {
system.GPU().Maxwell3D().dirty_flags.OnMemoryWrite();
}
break;
}
case KEPLER_COMPUTE_REG_INDEX(launch):
// Abort execution since compute shaders can be used to alter game memory (e.g. CUDA
// kernels)
UNREACHABLE_MSG("Compute shaders are not implemented");
ProcessLaunch();
break;
default:
break;
}
}
void KeplerCompute::ProcessLaunch() {
const GPUVAddr launch_desc_loc = regs.launch_desc_loc.Address();
memory_manager.ReadBlockUnsafe(launch_desc_loc, &launch_description,
LaunchParams::NUM_LAUNCH_PARAMETERS * sizeof(u32));
const GPUVAddr code_loc = regs.code_loc.Address() + launch_description.program_start;
LOG_WARNING(HW_GPU, "Compute Kernel Execute at Address 0x{:016x}, STUBBED", code_loc);
}
} // namespace Tegra::Engines

171
src/video_core/engines/kepler_compute.h
View file

@ -6,14 +6,25 @@
#include <array>
#include <cstddef>
#include <vector>
#include "common/bit_field.h"
#include "common/common_funcs.h"
#include "common/common_types.h"
#include "video_core/engines/engine_upload.h"
#include "video_core/gpu.h"
namespace Core {
class System;
}
namespace Tegra {
class MemoryManager;
}
namespace VideoCore {
class RasterizerInterface;
}
namespace Tegra::Engines {
#define KEPLER_COMPUTE_REG_INDEX(field_name) \
@ -21,7 +32,8 @@ namespace Tegra::Engines {
class KeplerCompute final {
public:
explicit KeplerCompute(MemoryManager& memory_manager);
explicit KeplerCompute(Core::System& system, VideoCore::RasterizerInterface& rasterizer,
MemoryManager& memory_manager);
~KeplerCompute();
static constexpr std::size_t NumConstBuffers = 8;
@ -31,30 +43,183 @@ public:
union {
struct {
INSERT_PADDING_WORDS(0xAF);
INSERT_PADDING_WORDS(0x60);
Upload::Data upload;
struct {
union {
BitField<0, 1, u32> linear;
};
} exec_upload;
u32 data_upload;
INSERT_PADDING_WORDS(0x3F);
struct {
u32 address;
GPUVAddr Address() const {
return static_cast<GPUVAddr>((static_cast<GPUVAddr>(address) << 8));
}
} launch_desc_loc;
INSERT_PADDING_WORDS(0x1);
u32 launch;
INSERT_PADDING_WORDS(0xC48);
INSERT_PADDING_WORDS(0x4A7);
struct {
u32 address_high;
u32 address_low;
u32 limit;
GPUVAddr Address() const {
return static_cast<GPUVAddr>((static_cast<GPUVAddr>(address_high) << 32) |
address_low);
}
} tsc;
INSERT_PADDING_WORDS(0x3);
struct {
u32 address_high;
u32 address_low;
u32 limit;
GPUVAddr Address() const {
return static_cast<GPUVAddr>((static_cast<GPUVAddr>(address_high) << 32) |
address_low);
}
} tic;
INSERT_PADDING_WORDS(0x22);
struct {
u32 address_high;
u32 address_low;
GPUVAddr Address() const {
return static_cast<GPUVAddr>((static_cast<GPUVAddr>(address_high) << 32) |
address_low);
}
} code_loc;
INSERT_PADDING_WORDS(0x3FE);
u32 texture_const_buffer_index;
INSERT_PADDING_WORDS(0x374);
};
std::array<u32, NUM_REGS> reg_array;
};
} regs{};
struct LaunchParams {
static constexpr std::size_t NUM_LAUNCH_PARAMETERS = 0x40;
INSERT_PADDING_WORDS(0x8);
u32 program_start;
INSERT_PADDING_WORDS(0x2);
BitField<30, 1, u32> linked_tsc;
BitField<0, 31, u32> grid_dim_x;
union {
BitField<0, 16, u32> grid_dim_y;
BitField<16, 16, u32> grid_dim_z;
};
INSERT_PADDING_WORDS(0x3);
BitField<0, 16, u32> shared_alloc;
BitField<0, 31, u32> block_dim_x;
union {
BitField<0, 16, u32> block_dim_y;
BitField<16, 16, u32> block_dim_z;
};
union {
BitField<0, 8, u32> const_buffer_enable_mask;
BitField<29, 2, u32> cache_layout;
} memory_config;
INSERT_PADDING_WORDS(0x8);
struct {
u32 address_low;
union {
BitField<0, 8, u32> address_high;
BitField<15, 17, u32> size;
};
GPUVAddr Address() const {
return static_cast<GPUVAddr>((static_cast<GPUVAddr>(address_high.Value()) << 32) |
address_low);
}
} const_buffer_config[8];
union {
BitField<0, 20, u32> local_pos_alloc;
BitField<27, 5, u32> barrier_alloc;
};
union {
BitField<0, 20, u32> local_neg_alloc;
BitField<24, 5, u32> gpr_alloc;
};
INSERT_PADDING_WORDS(0x11);
} launch_description;
struct {
u32 write_offset = 0;
u32 copy_size = 0;
std::vector<u8> inner_buffer;
} state{};
static_assert(sizeof(Regs) == Regs::NUM_REGS * sizeof(u32),
"KeplerCompute Regs has wrong size");
static_assert(sizeof(LaunchParams) == LaunchParams::NUM_LAUNCH_PARAMETERS * sizeof(u32),
"KeplerCompute LaunchParams has wrong size");
/// Write the value to the register identified by method.
void CallMethod(const GPU::MethodCall& method_call);
private:
Core::System& system;
VideoCore::RasterizerInterface& rasterizer;
MemoryManager& memory_manager;
Upload::State upload_state;
void ProcessLaunch();
};
#define ASSERT_REG_POSITION(field_name, position) \
static_assert(offsetof(KeplerCompute::Regs, field_name) == position * 4, \
"Field " #field_name " has invalid position")
#define ASSERT_LAUNCH_PARAM_POSITION(field_name, position) \
static_assert(offsetof(KeplerCompute::LaunchParams, field_name) == position * 4, \
"Field " #field_name " has invalid position")
ASSERT_REG_POSITION(upload, 0x60);
ASSERT_REG_POSITION(exec_upload, 0x6C);
ASSERT_REG_POSITION(data_upload, 0x6D);
ASSERT_REG_POSITION(launch, 0xAF);
ASSERT_REG_POSITION(tsc, 0x557);
ASSERT_REG_POSITION(tic, 0x55D);
ASSERT_REG_POSITION(code_loc, 0x582);
ASSERT_REG_POSITION(texture_const_buffer_index, 0x982);
ASSERT_LAUNCH_PARAM_POSITION(program_start, 0x8);
ASSERT_LAUNCH_PARAM_POSITION(grid_dim_x, 0xC);
ASSERT_LAUNCH_PARAM_POSITION(shared_alloc, 0x11);
ASSERT_LAUNCH_PARAM_POSITION(block_dim_x, 0x12);
ASSERT_LAUNCH_PARAM_POSITION(memory_config, 0x14);
ASSERT_LAUNCH_PARAM_POSITION(const_buffer_config, 0x1D);
#undef ASSERT_REG_POSITION

2
src/video_core/gpu.cpp
View file

@ -35,7 +35,7 @@ GPU::GPU(Core::System& system, VideoCore::RendererBase& renderer) : renderer{ren
dma_pusher = std::make_unique<Tegra::DmaPusher>(*this);
maxwell_3d = std::make_unique<Engines::Maxwell3D>(system, rasterizer, *memory_manager);
fermi_2d = std::make_unique<Engines::Fermi2D>(rasterizer, *memory_manager);
kepler_compute = std::make_unique<Engines::KeplerCompute>(*memory_manager);
kepler_compute = std::make_unique<Engines::KeplerCompute>(system, rasterizer, *memory_manager);
maxwell_dma = std::make_unique<Engines::MaxwellDMA>(system, rasterizer, *memory_manager);
kepler_memory = std::make_unique<Engines::KeplerMemory>(system, *memory_manager);
}