purgatory/yuzu
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Merge pull request #7658 from ameerj/sparse-fixes

Browse Source 
video_core/memory_manager: Fixes for sparse memory management
This commit is contained in:
Fernando S 2022-01-06 13:50:14 +01:00 committed by GitHub
commit a95c49e7d0
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 46 additions and 63 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

90
src/video_core/memory_manager.cpp
View File

@ -73,12 +73,12 @@ void MemoryManager::Unmap(GPUVAddr gpu_addr, std::size_t size) {
} }
const auto submapped_ranges = GetSubmappedRange(gpu_addr, size); const auto submapped_ranges = GetSubmappedRange(gpu_addr, size);
for (const auto& map : submapped_ranges) { for (const auto& [map_addr, map_size] : submapped_ranges) {
// Flush and invalidate through the GPU interface, to be asynchronous if possible. // Flush and invalidate through the GPU interface, to be asynchronous if possible.
const std::optional<VAddr> cpu_addr = GpuToCpuAddress(map.first); const std::optional<VAddr> cpu_addr = GpuToCpuAddress(map_addr);
ASSERT(cpu_addr); ASSERT(cpu_addr);
rasterizer->UnmapMemory(*cpu_addr, map.second); rasterizer->UnmapMemory(*cpu_addr, map_size);
} }
UpdateRange(gpu_addr, PageEntry::State::Unmapped, size); UpdateRange(gpu_addr, PageEntry::State::Unmapped, size);
@ -265,7 +265,8 @@ size_t MemoryManager::BytesToMapEnd(GPUVAddr gpu_addr) const noexcept {
return it->second - (gpu_addr - it->first); return it->second - (gpu_addr - it->first);
} }
void MemoryManager::ReadBlock(GPUVAddr gpu_src_addr, void* dest_buffer, std::size_t size) const { void MemoryManager::ReadBlockImpl(GPUVAddr gpu_src_addr, void* dest_buffer, std::size_t size,
bool is_safe) const {
std::size_t remaining_size{size}; std::size_t remaining_size{size};
std::size_t page_index{gpu_src_addr >> page_bits}; std::size_t page_index{gpu_src_addr >> page_bits};
std::size_t page_offset{gpu_src_addr & page_mask}; std::size_t page_offset{gpu_src_addr & page_mask};
@ -273,35 +274,15 @@ void MemoryManager::ReadBlock(GPUVAddr gpu_src_addr, void* dest_buffer, std::siz
while (remaining_size > 0) { while (remaining_size > 0) {
const std::size_t copy_amount{ const std::size_t copy_amount{
std::min(static_cast<std::size_t>(page_size) - page_offset, remaining_size)}; std::min(static_cast<std::size_t>(page_size) - page_offset, remaining_size)};
const auto page_addr{GpuToCpuAddress(page_index << page_bits)};
if (const auto page_addr{GpuToCpuAddress(page_index << page_bits)}; page_addr) { if (page_addr && *page_addr != 0) {
const auto src_addr{*page_addr + page_offset}; const auto src_addr{*page_addr + page_offset};
if (is_safe) {
// Flush must happen on the rasterizer interface, such that memory is always synchronous // Flush must happen on the rasterizer interface, such that memory is always
// when it is read (even when in asynchronous GPU mode). Fixes Dead Cells title menu. // synchronous when it is read (even when in asynchronous GPU mode).
// Fixes Dead Cells title menu.
rasterizer->FlushRegion(src_addr, copy_amount); rasterizer->FlushRegion(src_addr, copy_amount);
system.Memory().ReadBlockUnsafe(src_addr, dest_buffer, copy_amount);
} }
page_index++;
page_offset = 0;
dest_buffer = static_cast<u8*>(dest_buffer) + copy_amount;
remaining_size -= copy_amount;
}
}
void MemoryManager::ReadBlockUnsafe(GPUVAddr gpu_src_addr, void* dest_buffer,
const std::size_t size) const {
std::size_t remaining_size{size};
std::size_t page_index{gpu_src_addr >> page_bits};
std::size_t page_offset{gpu_src_addr & page_mask};
while (remaining_size > 0) {
const std::size_t copy_amount{
std::min(static_cast<std::size_t>(page_size) - page_offset, remaining_size)};
if (const auto page_addr{GpuToCpuAddress(page_index << page_bits)}; page_addr) {
const auto src_addr{*page_addr + page_offset};
system.Memory().ReadBlockUnsafe(src_addr, dest_buffer, copy_amount); system.Memory().ReadBlockUnsafe(src_addr, dest_buffer, copy_amount);
} else { } else {
std::memset(dest_buffer, 0, copy_amount); std::memset(dest_buffer, 0, copy_amount);
@ -314,7 +295,17 @@ void MemoryManager::ReadBlockUnsafe(GPUVAddr gpu_src_addr, void* dest_buffer,
} }
} }
void MemoryManager::WriteBlock(GPUVAddr gpu_dest_addr, const void* src_buffer, std::size_t size) { void MemoryManager::ReadBlock(GPUVAddr gpu_src_addr, void* dest_buffer, std::size_t size) const {
ReadBlockImpl(gpu_src_addr, dest_buffer, size, true);
}
void MemoryManager::ReadBlockUnsafe(GPUVAddr gpu_src_addr, void* dest_buffer,
const std::size_t size) const {
ReadBlockImpl(gpu_src_addr, dest_buffer, size, false);
}
void MemoryManager::WriteBlockImpl(GPUVAddr gpu_dest_addr, const void* src_buffer, std::size_t size,
bool is_safe) {
std::size_t remaining_size{size}; std::size_t remaining_size{size};
std::size_t page_index{gpu_dest_addr >> page_bits}; std::size_t page_index{gpu_dest_addr >> page_bits};
std::size_t page_offset{gpu_dest_addr & page_mask}; std::size_t page_offset{gpu_dest_addr & page_mask};
@ -322,13 +313,15 @@ void MemoryManager::WriteBlock(GPUVAddr gpu_dest_addr, const void* src_buffer, s
while (remaining_size > 0) { while (remaining_size > 0) {
const std::size_t copy_amount{ const std::size_t copy_amount{
std::min(static_cast<std::size_t>(page_size) - page_offset, remaining_size)}; std::min(static_cast<std::size_t>(page_size) - page_offset, remaining_size)};
const auto page_addr{GpuToCpuAddress(page_index << page_bits)};
if (const auto page_addr{GpuToCpuAddress(page_index << page_bits)}; page_addr) { if (page_addr && *page_addr != 0) {
const auto dest_addr{*page_addr + page_offset}; const auto dest_addr{*page_addr + page_offset};
if (is_safe) {
// Invalidate must happen on the rasterizer interface, such that memory is always // Invalidate must happen on the rasterizer interface, such that memory is always
// synchronous when it is written (even when in asynchronous GPU mode). // synchronous when it is written (even when in asynchronous GPU mode).
rasterizer->InvalidateRegion(dest_addr, copy_amount); rasterizer->InvalidateRegion(dest_addr, copy_amount);
}
system.Memory().WriteBlockUnsafe(dest_addr, src_buffer, copy_amount); system.Memory().WriteBlockUnsafe(dest_addr, src_buffer, copy_amount);
} }
@ -339,26 +332,13 @@ void MemoryManager::WriteBlock(GPUVAddr gpu_dest_addr, const void* src_buffer, s
} }
} }
void MemoryManager::WriteBlock(GPUVAddr gpu_dest_addr, const void* src_buffer, std::size_t size) {
WriteBlockImpl(gpu_dest_addr, src_buffer, size, true);
}
void MemoryManager::WriteBlockUnsafe(GPUVAddr gpu_dest_addr, const void* src_buffer, void MemoryManager::WriteBlockUnsafe(GPUVAddr gpu_dest_addr, const void* src_buffer,
std::size_t size) { std::size_t size) {
std::size_t remaining_size{size}; WriteBlockImpl(gpu_dest_addr, src_buffer, size, false);
std::size_t page_index{gpu_dest_addr >> page_bits};
std::size_t page_offset{gpu_dest_addr & page_mask};
while (remaining_size > 0) {
const std::size_t copy_amount{
std::min(static_cast<std::size_t>(page_size) - page_offset, remaining_size)};
if (const auto page_addr{GpuToCpuAddress(page_index << page_bits)}; page_addr) {
const auto dest_addr{*page_addr + page_offset};
system.Memory().WriteBlockUnsafe(dest_addr, src_buffer, copy_amount);
}
page_index++;
page_offset = 0;
src_buffer = static_cast<const u8*>(src_buffer) + copy_amount;
remaining_size -= copy_amount;
}
} }
void MemoryManager::FlushRegion(GPUVAddr gpu_addr, size_t size) const { void MemoryManager::FlushRegion(GPUVAddr gpu_addr, size_t size) const {
@ -435,15 +415,15 @@ std::vector<std::pair<GPUVAddr, std::size_t>> MemoryManager::GetSubmappedRange(
size_t page_offset{gpu_addr & page_mask}; size_t page_offset{gpu_addr & page_mask};
std::optional<std::pair<GPUVAddr, std::size_t>> last_segment{}; std::optional<std::pair<GPUVAddr, std::size_t>> last_segment{};
std::optional<VAddr> old_page_addr{}; std::optional<VAddr> old_page_addr{};
const auto extend_size = [this, &last_segment, &page_index](std::size_t bytes) { const auto extend_size = [&last_segment, &page_index, &page_offset](std::size_t bytes) {
if (!last_segment) { if (!last_segment) {
GPUVAddr new_base_addr = page_index << page_bits; const GPUVAddr new_base_addr = (page_index << page_bits) + page_offset;
last_segment = {new_base_addr, bytes}; last_segment = {new_base_addr, bytes};
} else { } else {
last_segment->second += bytes; last_segment->second += bytes;
} }
}; };
const auto split = [this, &last_segment, &result] { const auto split = [&last_segment, &result] {
if (last_segment) { if (last_segment) {
result.push_back(*last_segment); result.push_back(*last_segment);
last_segment = std::nullopt; last_segment = std::nullopt;
@ -452,7 +432,7 @@ std::vector<std::pair<GPUVAddr, std::size_t>> MemoryManager::GetSubmappedRange(
while (remaining_size > 0) { while (remaining_size > 0) {
const size_t num_bytes{std::min(page_size - page_offset, remaining_size)}; const size_t num_bytes{std::min(page_size - page_offset, remaining_size)};
const auto page_addr{GpuToCpuAddress(page_index << page_bits)}; const auto page_addr{GpuToCpuAddress(page_index << page_bits)};
if (!page_addr) { if (!page_addr || *page_addr == 0) {
split(); split();
} else if (old_page_addr) { } else if (old_page_addr) {
if (*old_page_addr + page_size != *page_addr) { if (*old_page_addr + page_size != *page_addr) {

5
src/video_core/memory_manager.h
View File

@ -155,6 +155,11 @@ private:
void FlushRegion(GPUVAddr gpu_addr, size_t size) const; void FlushRegion(GPUVAddr gpu_addr, size_t size) const;
void ReadBlockImpl(GPUVAddr gpu_src_addr, void* dest_buffer, std::size_t size,
bool is_safe) const;
void WriteBlockImpl(GPUVAddr gpu_dest_addr, const void* src_buffer, std::size_t size,
bool is_safe);
[[nodiscard]] static constexpr std::size_t PageEntryIndex(GPUVAddr gpu_addr) { [[nodiscard]] static constexpr std::size_t PageEntryIndex(GPUVAddr gpu_addr) {
return (gpu_addr >> page_bits) & page_table_mask; return (gpu_addr >> page_bits) & page_table_mask;
} }

4
src/video_core/texture_cache/texture_cache.h
View File

@ -1376,9 +1376,7 @@ void TextureCache<P>::ForEachSparseSegment(ImageBase& image, Func&& func) {
using FuncReturn = typename std::invoke_result<Func, GPUVAddr, VAddr, size_t>::type; using FuncReturn = typename std::invoke_result<Func, GPUVAddr, VAddr, size_t>::type;
static constexpr bool RETURNS_BOOL = std::is_same_v<FuncReturn, bool>; static constexpr bool RETURNS_BOOL = std::is_same_v<FuncReturn, bool>;
const auto segments = gpu_memory.GetSubmappedRange(image.gpu_addr, image.guest_size_bytes); const auto segments = gpu_memory.GetSubmappedRange(image.gpu_addr, image.guest_size_bytes);
for (auto& segment : segments) { for (const auto& [gpu_addr, size] : segments) {
const auto gpu_addr = segment.first;
const auto size = segment.second;
std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr); std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr);
ASSERT(cpu_addr); ASSERT(cpu_addr);
if constexpr (RETURNS_BOOL) { if constexpr (RETURNS_BOOL) {