Buffer Cache: Clear sync code.

This commit is contained in:
Fernando Sahmkow 2023-05-12 14:46:38 +02:00
parent 9c739f1506
commit c4bfbc6d25
2 changed files with 4 additions and 28 deletions

View File

@ -131,33 +131,15 @@ std::optional<VideoCore::RasterizerDownloadArea> BufferCache<P>::GetFlushArea(VA
template <class P> template <class P>
void BufferCache<P>::DownloadMemory(VAddr cpu_addr, u64 size) { void BufferCache<P>::DownloadMemory(VAddr cpu_addr, u64 size) {
WaitOnAsyncFlushes(cpu_addr, size);
ForEachBufferInRange(cpu_addr, size, [&](BufferId, Buffer& buffer) { ForEachBufferInRange(cpu_addr, size, [&](BufferId, Buffer& buffer) {
DownloadBufferMemory(buffer, cpu_addr, size); DownloadBufferMemory(buffer, cpu_addr, size);
}); });
} }
template <class P>
void BufferCache<P>::WaitOnAsyncFlushes(VAddr cpu_addr, u64 size) {
bool must_wait = false;
ForEachInOverlapCounter(async_downloads, cpu_addr, size,
[&](VAddr, VAddr, int) { must_wait = true; });
bool must_release = false;
ForEachInRangeSet(pending_ranges, cpu_addr, size, [&](VAddr, VAddr) { must_release = true; });
if (must_release) {
std::function<void()> tmp([]() {});
rasterizer.SignalFence(std::move(tmp));
}
if (must_wait || must_release) {
rasterizer.ReleaseFences();
}
}
template <class P> template <class P>
void BufferCache<P>::ClearDownload(IntervalType subtract_interval) { void BufferCache<P>::ClearDownload(IntervalType subtract_interval) {
RemoveEachInOverlapCounter(async_downloads, subtract_interval, -1024); RemoveEachInOverlapCounter(async_downloads, subtract_interval, -1024);
uncommitted_ranges.subtract(subtract_interval); uncommitted_ranges.subtract(subtract_interval);
pending_ranges.subtract(subtract_interval);
for (auto& interval_set : committed_ranges) { for (auto& interval_set : committed_ranges) {
interval_set.subtract(subtract_interval); interval_set.subtract(subtract_interval);
} }
@ -177,7 +159,6 @@ bool BufferCache<P>::DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 am
} }
const IntervalType subtract_interval{*cpu_dest_address, *cpu_dest_address + amount}; const IntervalType subtract_interval{*cpu_dest_address, *cpu_dest_address + amount};
WaitOnAsyncFlushes(*cpu_src_address, static_cast<u32>(amount));
ClearDownload(subtract_interval); ClearDownload(subtract_interval);
BufferId buffer_a; BufferId buffer_a;
@ -205,7 +186,6 @@ bool BufferCache<P>::DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 am
const IntervalType add_interval{new_base_address, new_base_address + size}; const IntervalType add_interval{new_base_address, new_base_address + size};
tmp_intervals.push_back(add_interval); tmp_intervals.push_back(add_interval);
uncommitted_ranges.add(add_interval); uncommitted_ranges.add(add_interval);
pending_ranges.add(add_interval);
}; };
ForEachInRangeSet(common_ranges, *cpu_src_address, amount, mirror); ForEachInRangeSet(common_ranges, *cpu_src_address, amount, mirror);
// This subtraction in this order is important for overlapping copies. // This subtraction in this order is important for overlapping copies.
@ -492,7 +472,6 @@ void BufferCache<P>::CommitAsyncFlushesHigh() {
} }
MICROPROFILE_SCOPE(GPU_DownloadMemory); MICROPROFILE_SCOPE(GPU_DownloadMemory);
pending_ranges.clear();
auto it = committed_ranges.begin(); auto it = committed_ranges.begin();
while (it != committed_ranges.end()) { while (it != committed_ranges.end()) {
auto& current_intervals = *it; auto& current_intervals = *it;
@ -1232,7 +1211,6 @@ void BufferCache<P>::MarkWrittenBuffer(BufferId buffer_id, VAddr cpu_addr, u32 s
const IntervalType base_interval{cpu_addr, cpu_addr + size}; const IntervalType base_interval{cpu_addr, cpu_addr + size};
common_ranges.add(base_interval); common_ranges.add(base_interval);
uncommitted_ranges.add(base_interval); uncommitted_ranges.add(base_interval);
pending_ranges.add(base_interval);
} }
template <class P> template <class P>
@ -1677,14 +1655,15 @@ typename BufferCache<P>::Binding BufferCache<P>::StorageBufferBinding(GPUVAddr s
const bool is_nvn_cbuf = cbuf_index == 0; const bool is_nvn_cbuf = cbuf_index == 0;
// The NVN driver buffer (index 0) is known to pack the SSBO address followed by its size. // The NVN driver buffer (index 0) is known to pack the SSBO address followed by its size.
if (is_nvn_cbuf) { if (is_nvn_cbuf) {
return gpu_memory->Read<u32>(ssbo_addr + 8); const u32 ssbo_size = gpu_memory->Read<u32>(ssbo_addr + 8);
if (ssbo_size != 0) {
return ssbo_size;
}
} }
// Other titles (notably Doom Eternal) may use STG/LDG on buffer addresses in custom defined // Other titles (notably Doom Eternal) may use STG/LDG on buffer addresses in custom defined
// cbufs, which do not store the sizes adjacent to the addresses, so use the fully // cbufs, which do not store the sizes adjacent to the addresses, so use the fully
// mapped buffer size for now. // mapped buffer size for now.
const u32 memory_layout_size = static_cast<u32>(gpu_memory->GetMemoryLayoutSize(gpu_addr)); const u32 memory_layout_size = static_cast<u32>(gpu_memory->GetMemoryLayoutSize(gpu_addr));
LOG_INFO(HW_GPU, "Binding storage buffer for cbuf index {}, MemoryLayoutSize 0x{:X}",
cbuf_index, memory_layout_size);
return memory_layout_size; return memory_layout_size;
}(); }();
const std::optional<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr); const std::optional<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr);

View File

@ -381,8 +381,6 @@ private:
void RunGarbageCollector(); void RunGarbageCollector();
void WaitOnAsyncFlushes(VAddr cpu_addr, u64 size);
void BindHostIndexBuffer(); void BindHostIndexBuffer();
void BindHostVertexBuffers(); void BindHostVertexBuffers();
@ -547,7 +545,6 @@ private:
IntervalSet uncommitted_ranges; IntervalSet uncommitted_ranges;
IntervalSet common_ranges; IntervalSet common_ranges;
IntervalSet cached_ranges; IntervalSet cached_ranges;
IntervalSet pending_ranges;
std::deque<IntervalSet> committed_ranges; std::deque<IntervalSet> committed_ranges;
// Async Buffers // Async Buffers