hdkv/yuzu
1
Fork 0
Code Issues Pull Requests Releases 1 Activity

early-access version 1335

Browse Source
This commit is contained in:
pineappleEA
2021-01-18 00:31:15 +01:00
parent 2a0ecc5567
commit 2b31de3238
36 changed files with 531 additions and 341 deletions
Download Patch File Download Diff File Expand all files Collapse all files

45
src/video_core/buffer_cache/buffer_cache.h
View File

@@ -239,8 +239,7 @@ private:
void ImmediateUploadMemory(Buffer& buffer, u64 largest_copy,
std::span<const BufferCopy> copies);
void MappedUploadMemory(Buffer& buffer, u64 total_size_bytes,
std::span<const BufferCopy> copies);
void MappedUploadMemory(Buffer& buffer, u64 total_size_bytes, std::span<BufferCopy> copies);
void DeleteBuffer(BufferId buffer_id);
@@ -362,11 +361,17 @@ void BufferCache<P>::DownloadMemory(VAddr cpu_addr, u64 size) {
auto download_staging = runtime.DownloadStagingBuffer(total_size_bytes);
const u8* const mapped_memory = download_staging.mapped_span.data();
const std::span<BufferCopy> copies_span(copies.data(), copies.data() + copies.size());
for (BufferCopy& copy : copies) {
// Modify copies to have the staging offset in mind
copy.dst_offset += download_staging.offset;
}
runtime.CopyBuffer(download_staging.buffer, buffer, copies_span);
runtime.Finish();
for (const BufferCopy& copy : copies) {
const VAddr copy_cpu_addr = buffer.CpuAddr() + copy.src_offset;
const u8* copy_mapped_memory = mapped_memory + copy.dst_offset;
// Undo the modified offset
const u64 dst_offset = copy.dst_offset - download_staging.offset;
const u8* copy_mapped_memory = mapped_memory + dst_offset;
cpu_memory.WriteBlockUnsafe(copy_cpu_addr, copy_mapped_memory, copy.size);
}
} else {
@@ -554,7 +559,9 @@ void BufferCache<P>::PopAsyncFlushes() {
}
if constexpr (USE_MEMORY_MAPS) {
auto download_staging = runtime.DownloadStagingBuffer(total_size_bytes);
for (const auto [copy, buffer_id] : downloads) {
for (auto& [copy, buffer_id] : downloads) {
// Have in mind the staging buffer offset for the copy
copy.dst_offset += download_staging.offset;
const std::array copies{copy};
runtime.CopyBuffer(download_staging.buffer, slot_buffers[buffer_id], copies);
}
@@ -562,7 +569,9 @@ void BufferCache<P>::PopAsyncFlushes() {
for (const auto [copy, buffer_id] : downloads) {
const Buffer& buffer = slot_buffers[buffer_id];
const VAddr cpu_addr = buffer.CpuAddr() + copy.src_offset;
const u8* read_mapped_memory = download_staging.mapped_span.data() + copy.dst_offset;
// Undo the modified offset
const u64 dst_offset = copy.dst_offset - download_staging.offset;
const u8* read_mapped_memory = download_staging.mapped_span.data() + dst_offset;
cpu_memory.WriteBlockUnsafe(cpu_addr, read_mapped_memory, copy.size);
}
} else {
@@ -651,24 +660,25 @@ void BufferCache<P>::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32
const VAddr cpu_addr = binding.cpu_addr;
const u32 size = binding.size;
Buffer& buffer = slot_buffers[binding.buffer_id];
if constexpr (IS_OPENGL) {
if (size <= SKIP_CACHE_SIZE && !buffer.IsRegionGpuModified(cpu_addr, size)) {
if (size <= SKIP_CACHE_SIZE && !buffer.IsRegionGpuModified(cpu_addr, size)) {
if constexpr (IS_OPENGL) {
if (runtime.HasFastBufferSubData()) {
// Fast path for Nvidia
if (!HasFastUniformBufferBound(stage, binding_index)) {
// We only have to bind when the currently bound buffer is not the fast version
fast_bound_uniform_buffers[stage] |= 1U << binding_index;
runtime.BindFastUniformBuffer(stage, binding_index, size);
}
const auto span = ImmediateBufferWithData(cpu_addr, size);
runtime.PushFastUniformBuffer(stage, binding_index, span);
} else {
// Stream buffer path to avoid stalling on non-Nvidia drivers
const auto span = runtime.BindMappedUniformBuffer(stage, binding_index, size);
cpu_memory.ReadBlockUnsafe(cpu_addr, span.data(), size);
return;
}
return;
}
fast_bound_uniform_buffers[stage] |= 1U << binding_index;
// Stream buffer path to avoid stalling on non-Nvidia drivers or Vulkan
const std::span<u8> span = runtime.BindMappedUniformBuffer(stage, binding_index, size);
cpu_memory.ReadBlockUnsafe(cpu_addr, span.data(), size);
return;
}
// Classic cached path
SynchronizeBuffer(buffer, cpu_addr, size);
@@ -1117,13 +1127,16 @@ void BufferCache<P>::ImmediateUploadMemory(Buffer& buffer, u64 largest_copy,
template <class P>
void BufferCache<P>::MappedUploadMemory(Buffer& buffer, u64 total_size_bytes,
std::span<const BufferCopy> copies) {
std::span<BufferCopy> copies) {
auto upload_staging = runtime.UploadStagingBuffer(total_size_bytes);
const std::span<u8> staging_pointer = upload_staging.mapped_span;
for (const BufferCopy& copy : copies) {
const VAddr cpu_addr = buffer.CpuAddr() + copy.dst_offset;
for (BufferCopy& copy : copies) {
u8* const src_pointer = staging_pointer.data() + copy.src_offset;
const VAddr cpu_addr = buffer.CpuAddr() + copy.dst_offset;
cpu_memory.ReadBlockUnsafe(cpu_addr, src_pointer, copy.size);
// Apply the staging offset
copy.src_offset += upload_staging.offset;
}
runtime.CopyBuffer(buffer, upload_staging.buffer, copies);
}