Mirrors/yuzu
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

DMAEngine: Accelerate BufferClear

Browse source
This commit is contained in:
Fernando Sahmkow 2021-07-13 03:33:08 +02:00
parent 776f391ff6
commit b780d5b5c5
11 changed files with 115 additions and 6 deletions
Download patch file Download diff file Expand all files Collapse all files

65
src/video_core/buffer_cache/buffer_cache.h
View file

@ -164,11 +164,16 @@ public:
/// Pop asynchronous downloads
void PopAsyncFlushes();
[[nodiscard]] bool DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 amount);
bool DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 amount);
bool DMAClear(GPUVAddr src_address, u64 amount, u32 value);
/// Return true when a CPU region is modified from the GPU
[[nodiscard]] bool IsRegionGpuModified(VAddr addr, size_t size);
/// Return true when a region is registered on the cache
[[nodiscard]] bool IsRegionRegistered(VAddr addr, size_t size);
/// Return true when a CPU region is modified from the CPU
[[nodiscard]] bool IsRegionCpuModified(VAddr addr, size_t size);
@ -469,8 +474,8 @@ bool BufferCache<P>::DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 am
if (!cpu_src_address || !cpu_dest_address) {
return false;
}
const bool source_dirty = IsRegionGpuModified(*cpu_src_address, amount);
const bool dest_dirty = IsRegionGpuModified(*cpu_dest_address, amount);
const bool source_dirty = IsRegionRegistered(*cpu_src_address, amount);
const bool dest_dirty = IsRegionRegistered(*cpu_dest_address, amount);
if (!source_dirty && !dest_dirty) {
return false;
}
@ -515,7 +520,7 @@ bool BufferCache<P>::DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 am
}
runtime.CopyBuffer(dest_buffer, src_buffer, copies);
if (source_dirty) {
if (IsRegionGpuModified(*cpu_src_address, amount)) {
dest_buffer.MarkRegionAsGpuModified(*cpu_dest_address, amount);
}
std::vector<u8> tmp_buffer(amount);
@ -524,6 +529,37 @@ bool BufferCache<P>::DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 am
return true;
}
template <class P>
bool BufferCache<P>::DMAClear(GPUVAddr dst_address, u64 amount, u32 value) {
const std::optional<VAddr> cpu_dst_address = gpu_memory.GpuToCpuAddress(dst_address);
if (!cpu_dst_address) {
return false;
}
const bool dest_dirty = IsRegionRegistered(*cpu_dst_address, amount);
if (!dest_dirty) {
return false;
}
const IntervalType subtract_interval{*cpu_dst_address, *cpu_dst_address + amount * sizeof(u32)};
uncommitted_ranges.subtract(subtract_interval);
for (auto& interval_set : committed_ranges) {
interval_set.subtract(subtract_interval);
}
common_ranges.subtract(subtract_interval);
const size_t size = amount * sizeof(u32);
BufferId buffer;
do {
has_deleted_buffers = false;
buffer = FindBuffer(*cpu_dst_address, static_cast<u32>(size));
} while (has_deleted_buffers);
auto& dest_buffer = slot_buffers[buffer];
const u32 offset = static_cast<u32>(*cpu_dst_address - dest_buffer.CpuAddr());
runtime.ClearBuffer(dest_buffer, offset, size, value);
return true;
}
template <class P>
void BufferCache<P>::BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr,
u32 size) {
@ -781,6 +817,27 @@ bool BufferCache<P>::IsRegionGpuModified(VAddr addr, size_t size) {
return false;
}
template <class P>
bool BufferCache<P>::IsRegionRegistered(VAddr addr, size_t size) {
const VAddr end_addr = addr + size;
const u64 page_end = Common::DivCeil(end_addr, PAGE_SIZE);
for (u64 page = addr >> PAGE_BITS; page < page_end;) {
const BufferId buffer_id = page_table[page];
if (!buffer_id) {
++page;
continue;
}
Buffer& buffer = slot_buffers[buffer_id];
const VAddr buf_start_addr = buffer.CpuAddr();
const VAddr buf_end_addr = buf_start_addr + buffer.SizeBytes();
if (buf_start_addr < end_addr && addr < buf_end_addr) {
return true;
}
page = Common::DivCeil(end_addr, PAGE_SIZE);
}
return false;
}
template <class P>
bool BufferCache<P>::IsRegionCpuModified(VAddr addr, size_t size) {
const u64 page_end = Common::DivCeil(addr + size, PAGE_SIZE);