archive/Cemu
1
0
Fork 0
mirror of https://github.com/cemu-project/Cemu.git synced 2025-07-05 06:21:19 +12:00
Code Issues Projects Releases Packages Wiki Activity Actions

do vertex restride on the GPU & don't over-sync

Browse source
This commit is contained in:
Samuliak 2024-09-23 16:47:22 +02:00
parent 8b68df0c59
commit 3cf831d46a
3 changed files with 14 additions and 12 deletions
Download patch file Download diff file Expand all files Collapse all files

12
src/Cafe/HW/Latte/Renderer/Metal/MetalMemoryManager.cpp
View file

@ -9,7 +9,7 @@ MetalVertexBufferCache::~MetalVertexBufferCache()
{
}
MetalRestridedBufferRange MetalVertexBufferCache::RestrideBufferIfNeeded(MTL::Buffer* bufferCache, uint32 bufferIndex, size_t stride)
MetalRestridedBufferRange MetalVertexBufferCache::RestrideBufferIfNeeded(MTL::Buffer* bufferCache, uint32 bufferIndex, size_t stride, std::vector<MTL::Resource*>& barrierBuffers)
{
auto vertexBufferRange = m_bufferRanges[bufferIndex];
auto& restrideInfo = *vertexBufferRange.restrideInfo;
@ -28,14 +28,14 @@ MetalRestridedBufferRange MetalVertexBufferCache::RestrideBufferIfNeeded(MTL::Bu
restrideInfo.allocation = m_bufferAllocator.GetBufferAllocation(newSize);
buffer = m_bufferAllocator.GetBuffer(restrideInfo.allocation.bufferIndex);
// HACK: the restriding is done on the CPU, since doing it on the GPU was causing over-synchronization
/*
uint8* oldPtr = (uint8*)bufferCache->contents() + vertexBufferRange.offset;
uint8* newPtr = (uint8*)buffer->contents() + restrideInfo.allocation.offset;
for (size_t elem = 0; elem < vertexBufferRange.size / stride; elem++)
memcpy(newPtr + elem * newStride, oldPtr + elem * stride, stride);
*/
/*
if (m_mtlr->GetEncoderType() == MetalEncoderType::Render)
{
auto renderCommandEncoder = static_cast<MTL::RenderCommandEncoder*>(m_mtlr->GetCommandEncoder());
@ -56,16 +56,12 @@ MetalRestridedBufferRange MetalVertexBufferCache::RestrideBufferIfNeeded(MTL::Bu
renderCommandEncoder->drawPrimitives(MTL::PrimitiveTypeTriangleStrip, NS::UInteger(0), vertexBufferRange.size / stride);
// TODO: do the barriers in one call?
MTL::Resource* barrierBuffers[] = {buffer};
renderCommandEncoder->memoryBarrier(barrierBuffers, 1, MTL::RenderStageVertex, MTL::RenderStageVertex);
vectorAppendUnique(barrierBuffers, static_cast<MTL::Resource*>(buffer));
}
else
{
debug_printf("vertex buffer restride needs an active render command encoder\n");
cemu_assert_suspicious();
}
*/
restrideInfo.memoryInvalidated = false;
restrideInfo.lastStride = newStride;