ThreadManager: Sync async reads on accurate gpu.
This commit is contained in:
parent
57fdbd9b89
commit
165ae823f5
9 changed files with 48 additions and 8 deletions
|
@ -145,6 +145,18 @@ public:
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
|
bool MustFlushRegion(VAddr addr, std::size_t size) {
|
||||||
|
std::lock_guard lock{mutex};
|
||||||
|
|
||||||
|
std::vector<MapInterval> objects = GetMapsInRange(addr, size);
|
||||||
|
for (auto& object : objects) {
|
||||||
|
if (object->IsModified() && object->IsRegistered()) {
|
||||||
|
return true;
|
||||||
|
}
|
||||||
|
}
|
||||||
|
return false;
|
||||||
|
}
|
||||||
|
|
||||||
/// Mark the specified region as being invalidated
|
/// Mark the specified region as being invalidated
|
||||||
void InvalidateRegion(VAddr addr, u64 size) {
|
void InvalidateRegion(VAddr addr, u64 size) {
|
||||||
std::lock_guard lock{mutex};
|
std::lock_guard lock{mutex};
|
||||||
|
|
|
@ -147,7 +147,7 @@ void GPU::SyncGuestHost() {
|
||||||
}
|
}
|
||||||
|
|
||||||
void GPU::OnCommandListEnd() {
|
void GPU::OnCommandListEnd() {
|
||||||
renderer.Rasterizer().ReleaseFences();
|
renderer->Rasterizer().ReleaseFences();
|
||||||
}
|
}
|
||||||
// Note that, traditionally, methods are treated as 4-byte addressable locations, and hence
|
// Note that, traditionally, methods are treated as 4-byte addressable locations, and hence
|
||||||
// their numbers are written down multiplied by 4 in Docs. Here we are not multiply by 4.
|
// their numbers are written down multiplied by 4 in Docs. Here we are not multiply by 4.
|
||||||
|
|
|
@ -6,6 +6,7 @@
|
||||||
#include "common/microprofile.h"
|
#include "common/microprofile.h"
|
||||||
#include "core/core.h"
|
#include "core/core.h"
|
||||||
#include "core/frontend/emu_window.h"
|
#include "core/frontend/emu_window.h"
|
||||||
|
#include "core/settings.h"
|
||||||
#include "video_core/dma_pusher.h"
|
#include "video_core/dma_pusher.h"
|
||||||
#include "video_core/gpu.h"
|
#include "video_core/gpu.h"
|
||||||
#include "video_core/gpu_thread.h"
|
#include "video_core/gpu_thread.h"
|
||||||
|
@ -80,7 +81,11 @@ void ThreadManager::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
|
||||||
}
|
}
|
||||||
|
|
||||||
void ThreadManager::FlushRegion(VAddr addr, u64 size) {
|
void ThreadManager::FlushRegion(VAddr addr, u64 size) {
|
||||||
PushCommand(FlushRegionCommand(addr, size));
|
if (system.Renderer().Rasterizer().MustFlushRegion(addr, size)) {
|
||||||
|
u64 fence = PushCommand(FlushRegionCommand(addr, size));
|
||||||
|
while (fence < state.signaled_fence.load(std::memory_order_relaxed)) {
|
||||||
|
}
|
||||||
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
void ThreadManager::InvalidateRegion(VAddr addr, u64 size) {
|
void ThreadManager::InvalidateRegion(VAddr addr, u64 size) {
|
||||||
|
|
|
@ -49,13 +49,9 @@ public:
|
||||||
/// Records a GPU query and caches it
|
/// Records a GPU query and caches it
|
||||||
virtual void Query(GPUVAddr gpu_addr, QueryType type, std::optional<u64> timestamp) = 0;
|
virtual void Query(GPUVAddr gpu_addr, QueryType type, std::optional<u64> timestamp) = 0;
|
||||||
|
|
||||||
virtual void SignalFence(GPUVAddr addr, u32 value) {
|
virtual void SignalFence(GPUVAddr addr, u32 value) {}
|
||||||
|
|
||||||
}
|
virtual void ReleaseFences() {}
|
||||||
|
|
||||||
virtual void ReleaseFences() {
|
|
||||||
|
|
||||||
}
|
|
||||||
|
|
||||||
/// Notify rasterizer that all caches should be flushed to Switch memory
|
/// Notify rasterizer that all caches should be flushed to Switch memory
|
||||||
virtual void FlushAll() = 0;
|
virtual void FlushAll() = 0;
|
||||||
|
@ -63,6 +59,8 @@ public:
|
||||||
/// Notify rasterizer that any caches of the specified region should be flushed to Switch memory
|
/// Notify rasterizer that any caches of the specified region should be flushed to Switch memory
|
||||||
virtual void FlushRegion(VAddr addr, u64 size) = 0;
|
virtual void FlushRegion(VAddr addr, u64 size) = 0;
|
||||||
|
|
||||||
|
virtual bool MustFlushRegion(VAddr addr, u64 size) = 0;
|
||||||
|
|
||||||
/// Notify rasterizer that any caches of the specified region should be invalidated
|
/// Notify rasterizer that any caches of the specified region should be invalidated
|
||||||
virtual void InvalidateRegion(VAddr addr, u64 size) = 0;
|
virtual void InvalidateRegion(VAddr addr, u64 size) = 0;
|
||||||
|
|
||||||
|
|
|
@ -650,6 +650,10 @@ void RasterizerOpenGL::FlushRegion(VAddr addr, u64 size) {
|
||||||
query_cache.FlushRegion(addr, size);
|
query_cache.FlushRegion(addr, size);
|
||||||
}
|
}
|
||||||
|
|
||||||
|
bool RasterizerOpenGL::MustFlushRegion(VAddr addr, u64 size) {
|
||||||
|
return texture_cache.MustFlushRegion(addr, size) || buffer_cache.MustFlushRegion(addr, size);
|
||||||
|
}
|
||||||
|
|
||||||
void RasterizerOpenGL::InvalidateRegion(VAddr addr, u64 size) {
|
void RasterizerOpenGL::InvalidateRegion(VAddr addr, u64 size) {
|
||||||
MICROPROFILE_SCOPE(OpenGL_CacheManagement);
|
MICROPROFILE_SCOPE(OpenGL_CacheManagement);
|
||||||
if (addr == 0 || size == 0) {
|
if (addr == 0 || size == 0) {
|
||||||
|
|
|
@ -67,6 +67,7 @@ public:
|
||||||
void Query(GPUVAddr gpu_addr, VideoCore::QueryType type, std::optional<u64> timestamp) override;
|
void Query(GPUVAddr gpu_addr, VideoCore::QueryType type, std::optional<u64> timestamp) override;
|
||||||
void FlushAll() override;
|
void FlushAll() override;
|
||||||
void FlushRegion(VAddr addr, u64 size) override;
|
void FlushRegion(VAddr addr, u64 size) override;
|
||||||
|
bool MustFlushRegion(VAddr addr, u64 size) override;
|
||||||
void InvalidateRegion(VAddr addr, u64 size) override;
|
void InvalidateRegion(VAddr addr, u64 size) override;
|
||||||
void OnCPUWrite(VAddr addr, u64 size) override;
|
void OnCPUWrite(VAddr addr, u64 size) override;
|
||||||
void SyncGuestHost() override;
|
void SyncGuestHost() override;
|
||||||
|
|
|
@ -514,6 +514,10 @@ void RasterizerVulkan::FlushRegion(VAddr addr, u64 size) {
|
||||||
query_cache.FlushRegion(addr, size);
|
query_cache.FlushRegion(addr, size);
|
||||||
}
|
}
|
||||||
|
|
||||||
|
bool RasterizerVulkan::MustFlushRegion(VAddr addr, u64 size) {
|
||||||
|
return texture_cache.MustFlushRegion(addr, size) || buffer_cache.MustFlushRegion(addr, size);
|
||||||
|
}
|
||||||
|
|
||||||
void RasterizerVulkan::InvalidateRegion(VAddr addr, u64 size) {
|
void RasterizerVulkan::InvalidateRegion(VAddr addr, u64 size) {
|
||||||
if (addr == 0 || size == 0) {
|
if (addr == 0 || size == 0) {
|
||||||
return;
|
return;
|
||||||
|
|
|
@ -118,6 +118,7 @@ public:
|
||||||
void Query(GPUVAddr gpu_addr, VideoCore::QueryType type, std::optional<u64> timestamp) override;
|
void Query(GPUVAddr gpu_addr, VideoCore::QueryType type, std::optional<u64> timestamp) override;
|
||||||
void FlushAll() override;
|
void FlushAll() override;
|
||||||
void FlushRegion(VAddr addr, u64 size) override;
|
void FlushRegion(VAddr addr, u64 size) override;
|
||||||
|
bool MustFlushRegion(VAddr addr, u64 size) override;
|
||||||
void InvalidateRegion(VAddr addr, u64 size) override;
|
void InvalidateRegion(VAddr addr, u64 size) override;
|
||||||
void OnCPUWrite(VAddr addr, u64 size) override;
|
void OnCPUWrite(VAddr addr, u64 size) override;
|
||||||
void SyncGuestHost() override;
|
void SyncGuestHost() override;
|
||||||
|
|
|
@ -116,6 +116,21 @@ public:
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
|
bool MustFlushRegion(VAddr addr, std::size_t size) {
|
||||||
|
std::lock_guard lock{mutex};
|
||||||
|
|
||||||
|
auto surfaces = GetSurfacesInRegion(addr, size);
|
||||||
|
if (surfaces.empty()) {
|
||||||
|
return false;
|
||||||
|
}
|
||||||
|
for (const auto& surface : surfaces) {
|
||||||
|
if (surface->IsModified()) {
|
||||||
|
return true;
|
||||||
|
}
|
||||||
|
}
|
||||||
|
return false;
|
||||||
|
}
|
||||||
|
|
||||||
TView GetTextureSurface(const Tegra::Texture::TICEntry& tic,
|
TView GetTextureSurface(const Tegra::Texture::TICEntry& tic,
|
||||||
const VideoCommon::Shader::Sampler& entry) {
|
const VideoCommon::Shader::Sampler& entry) {
|
||||||
std::lock_guard lock{mutex};
|
std::lock_guard lock{mutex};
|
||||||
|
|
Loading…
Reference in a new issue