suyu/src/video_core/dma_pusher.cpp
ReinUsesLisp 82c2601555 video_core: Reimplement the buffer cache
Reimplement the buffer cache using cached bindings and page level
granularity for modification tracking. This also drops the usage of
shared pointers and virtual functions from the cache.

- Bindings are cached, allowing to skip work when the game changes few
  bits between draws.
- OpenGL Assembly shaders no longer copy when a region has been modified
  from the GPU to emulate constant buffers, instead GL_EXT_memory_object
  is used to alias sub-buffers within the same allocation.
- OpenGL Assembly shaders stream constant buffer data using
  glProgramBufferParametersIuivNV, from NV_parameter_buffer_object. In
  theory this should save one hash table resolve inside the driver
  compared to glBufferSubData.
- A new OpenGL stream buffer is implemented based on fences for drivers
  that are not Nvidia's proprietary, due to their low performance on
  partial glBufferSubData calls synchronized with 3D rendering (that
  some games use a lot).
- Most optimizations are shared between APIs now, allowing Vulkan to
  cache more bindings than before, skipping unnecesarry work.

This commit adds the necessary infrastructure to use Vulkan object from
OpenGL. Overall, it improves performance and fixes some bugs present on
the old cache. There are still some edge cases hit by some games that
harm performance on some vendors, this are planned to be fixed in later
commits.
2021-02-13 02:17:22 -03:00

175 lines
5.9 KiB
C++

// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/cityhash.h"
#include "common/microprofile.h"
#include "core/core.h"
#include "core/memory.h"
#include "video_core/dma_pusher.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/gpu.h"
#include "video_core/memory_manager.h"
namespace Tegra {
DmaPusher::DmaPusher(Core::System& system_, GPU& gpu_) : gpu{gpu_}, system{system_} {}
DmaPusher::~DmaPusher() = default;
MICROPROFILE_DEFINE(DispatchCalls, "GPU", "Execute command buffer", MP_RGB(128, 128, 192));
void DmaPusher::DispatchCalls() {
MICROPROFILE_SCOPE(DispatchCalls);
gpu.SyncGuestHost();
dma_pushbuffer_subindex = 0;
dma_state.is_last_call = true;
while (system.IsPoweredOn()) {
if (!Step()) {
break;
}
}
gpu.FlushCommands();
gpu.SyncGuestHost();
gpu.OnCommandListEnd();
}
bool DmaPusher::Step() {
if (!ib_enable || dma_pushbuffer.empty()) {
// pushbuffer empty and IB empty or nonexistent - nothing to do
return false;
}
CommandList& command_list{dma_pushbuffer.front()};
ASSERT_OR_EXECUTE(
command_list.command_lists.size() || command_list.prefetch_command_list.size(), {
// Somehow the command_list is empty, in order to avoid a crash
// We ignore it and assume its size is 0.
dma_pushbuffer.pop();
dma_pushbuffer_subindex = 0;
return true;
});
if (command_list.prefetch_command_list.size()) {
// Prefetched command list from nvdrv, used for things like synchronization
command_headers = std::move(command_list.prefetch_command_list);
dma_pushbuffer.pop();
} else {
const CommandListHeader command_list_header{
command_list.command_lists[dma_pushbuffer_subindex++]};
const GPUVAddr dma_get = command_list_header.addr;
if (dma_pushbuffer_subindex >= command_list.command_lists.size()) {
// We've gone through the current list, remove it from the queue
dma_pushbuffer.pop();
dma_pushbuffer_subindex = 0;
}
if (command_list_header.size == 0) {
return true;
}
// Push buffer non-empty, read a word
command_headers.resize(command_list_header.size);
gpu.MemoryManager().ReadBlockUnsafe(dma_get, command_headers.data(),
command_list_header.size * sizeof(u32));
}
for (std::size_t index = 0; index < command_headers.size();) {
const CommandHeader& command_header = command_headers[index];
if (dma_state.method_count) {
// Data word of methods command
if (dma_state.non_incrementing) {
const u32 max_write = static_cast<u32>(
std::min<std::size_t>(index + dma_state.method_count, command_headers.size()) -
index);
CallMultiMethod(&command_header.argument, max_write);
dma_state.method_count -= max_write;
dma_state.is_last_call = true;
index += max_write;
continue;
} else {
dma_state.is_last_call = dma_state.method_count <= 1;
CallMethod(command_header.argument);
}
if (!dma_state.non_incrementing) {
dma_state.method++;
}
if (dma_increment_once) {
dma_state.non_incrementing = true;
}
dma_state.method_count--;
} else {
// No command active - this is the first word of a new one
switch (command_header.mode) {
case SubmissionMode::Increasing:
SetState(command_header);
dma_state.non_incrementing = false;
dma_increment_once = false;
break;
case SubmissionMode::NonIncreasing:
SetState(command_header);
dma_state.non_incrementing = true;
dma_increment_once = false;
break;
case SubmissionMode::Inline:
dma_state.method = command_header.method;
dma_state.subchannel = command_header.subchannel;
CallMethod(command_header.arg_count);
dma_state.non_incrementing = true;
dma_increment_once = false;
break;
case SubmissionMode::IncreaseOnce:
SetState(command_header);
dma_state.non_incrementing = false;
dma_increment_once = true;
break;
default:
break;
}
}
index++;
}
return true;
}
void DmaPusher::SetState(const CommandHeader& command_header) {
dma_state.method = command_header.method;
dma_state.subchannel = command_header.subchannel;
dma_state.method_count = command_header.method_count;
}
void DmaPusher::CallMethod(u32 argument) const {
if (dma_state.method < non_puller_methods) {
gpu.CallMethod(GPU::MethodCall{
dma_state.method,
argument,
dma_state.subchannel,
dma_state.method_count,
});
} else {
subchannels[dma_state.subchannel]->CallMethod(dma_state.method, argument,
dma_state.is_last_call);
}
}
void DmaPusher::CallMultiMethod(const u32* base_start, u32 num_methods) const {
if (dma_state.method < non_puller_methods) {
gpu.CallMultiMethod(dma_state.method, dma_state.subchannel, base_start, num_methods,
dma_state.method_count);
} else {
subchannels[dma_state.subchannel]->CallMultiMethod(dma_state.method, base_start,
num_methods, dma_state.method_count);
}
}
} // namespace Tegra