notsoyuzu
/
yuzu-mainline
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

video_core/memory_manager: Fixes for sparse memory management

This commit is contained in:
ameerj 2021-12-31 02:07:59 -05:00
parent 2428214c4b
commit 285b6dbc39
2 changed files with 12 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

22
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);
@ -274,8 +274,8 @@ void MemoryManager::ReadBlockImpl(GPUVAddr gpu_src_addr, void* dest_buffer, std:
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) { if (is_safe) {
// Flush must happen on the rasterizer interface, such that memory is always // Flush must happen on the rasterizer interface, such that memory is always
@ -313,8 +313,8 @@ void MemoryManager::WriteBlockImpl(GPUVAddr gpu_dest_addr, const void* src_buffe
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) { if (is_safe) {
@ -415,15 +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;
@ -432,7 +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) {

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) {