birkeh/o3de
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Merge pull request #1309 from aws-lumberyard-dev/MetalAsyncBufferFixes

Browse Source 
Metal async buffer fixes related to meshes
This commit is contained in:
moudgils
2021-06-15 09:48:26 -07:00
committed by GitHub
parent 0241538c47 f8f282998b
commit 82c5adaaed
6 changed files with 54 additions and 39 deletions
Download Patch File Download Diff File Expand all files Collapse all files
Gems/Atom/RHI/Metal/Code/Source/RHI/ArgumentBuffer.cpp
+12 -13
View File
@@ -400,15 +400,13 @@ namespace AZ
id<MTLResource> mtlconstantBufferResource = m_constantBuffer.GetGpuAddress<id<MTLResource>>();
if(RHI::CheckBitsAny(srgResourcesVisInfo.m_constantDataStageMask, RHI::ShaderStageMask::Compute))
{
uint16_t arrayIndex = resourcesToMakeResidentCompute[MTLResourceUsageRead].m_resourceArrayLen++;
resourcesToMakeResidentCompute[MTLResourceUsageRead].m_resourceArray[arrayIndex] = mtlconstantBufferResource;
resourcesToMakeResidentCompute[MTLResourceUsageRead].emplace(mtlconstantBufferResource);
}
else
{
MTLRenderStages mtlRenderStages = GetRenderStages(srgResourcesVisInfo.m_constantDataStageMask);
AZStd::pair <MTLResourceUsage,MTLRenderStages> key = AZStd::make_pair(MTLResourceUsageRead, mtlRenderStages);
uint16_t arrayIndex = resourcesToMakeResidentGraphics[key].m_resourceArrayLen++;
resourcesToMakeResidentGraphics[key].m_resourceArray[arrayIndex] = mtlconstantBufferResource;
AZStd::pair <MTLResourceUsage,MTLRenderStages> key = AZStd::make_pair(MTLResourceUsageRead, mtlRenderStages);
resourcesToMakeResidentGraphics[key].emplace(mtlconstantBufferResource);
}
}
}
@@ -440,16 +438,18 @@ namespace AZ
//Call UseResource on all resources for Compute stage
for (const auto& key : resourcesToMakeResidentCompute)
{
[static_cast<id<MTLComputeCommandEncoder>>(commandEncoder) useResources: key.second.m_resourceArray.data()
count: key.second.m_resourceArrayLen
AZStd::vector<id <MTLResource>> resourcesToProcessVec(key.second.begin(), key.second.end());
[static_cast<id<MTLComputeCommandEncoder>>(commandEncoder) useResources: &resourcesToProcessVec[0]
count: resourcesToProcessVec.size()
usage: key.first];
}
//Call UseResource on all resources for Vertex and Fragment stages
for (const auto& key : resourcesToMakeResidentGraphics)
{
[static_cast<id<MTLRenderCommandEncoder>>(commandEncoder) useResources: key.second.m_resourceArray.data()
count: key.second.m_resourceArrayLen
AZStd::vector<id <MTLResource>> resourcesToProcessVec(key.second.begin(), key.second.end());
[static_cast<id<MTLRenderCommandEncoder>>(commandEncoder) useResources: &resourcesToProcessVec[0]
count: resourcesToProcessVec.size()
usage: key.first.first
stages: key.first.second];
}
@@ -480,9 +480,9 @@ namespace AZ
AZ_Assert(false, "Undefined Resource type");
}
}
uint16_t arrayIndex = resourcesToMakeResidentMap[resourceUsage].m_resourceArrayLen++;
id<MTLResource> mtlResourceToBind = resourceBindingData.m_resourcPtr->GetGpuAddress<id<MTLResource>>();
resourcesToMakeResidentMap[resourceUsage].m_resourceArray[arrayIndex] = mtlResourceToBind;
resourcesToMakeResidentMap[resourceUsage].emplace(mtlResourceToBind);
}
}
@@ -516,9 +516,8 @@ namespace AZ
}
AZStd::pair <MTLResourceUsage, MTLRenderStages> key = AZStd::make_pair(resourceUsage, mtlRenderStages);
uint16_t arrayIndex = resourcesToMakeResidentMap[key].m_resourceArrayLen++;
id<MTLResource> mtlResourceToBind = resourceBindingData.m_resourcPtr->GetGpuAddress<id<MTLResource>>();
resourcesToMakeResidentMap[key].m_resourceArray[arrayIndex] = mtlResourceToBind;
resourcesToMakeResidentMap[key].emplace(mtlResourceToBind);
}
}
}
Gems/Atom/RHI/Metal/Code/Source/RHI/ArgumentBuffer.h
+4 -9
View File
@@ -120,15 +120,10 @@ namespace AZ
ResourceBindingsMap m_resourceBindings;
static const int MaxEntriesInArgTable = 31;
struct MetalResourceArray
{
AZStd::array<id <MTLResource>, MaxEntriesInArgTable> m_resourceArray;
uint16_t m_resourceArrayLen = 0;
};
//Map to cache all the resources based on the usage as we can batch all the resources for a given usage
using ComputeResourcesToMakeResidentMap = AZStd::unordered_map<MTLResourceUsage, MetalResourceArray>;
//Map to cache all the resources based on the usage and shader stage as we can batch all the resources for a given usage/shader usage
using GraphicsResourcesToMakeResidentMap = AZStd::unordered_map<AZStd::pair<MTLResourceUsage,MTLRenderStages>, MetalResourceArray>;
//Map to cache all the resources based on the usage as we can batch all the resources for a given usage.
using ComputeResourcesToMakeResidentMap = AZStd::unordered_map<MTLResourceUsage, AZStd::unordered_set<id <MTLResource>>>;
//Map to cache all the resources based on the usage and shader stage as we can batch all the resources for a given usage/shader usage.
using GraphicsResourcesToMakeResidentMap = AZStd::unordered_map<AZStd::pair<MTLResourceUsage,MTLRenderStages>, AZStd::unordered_set<id <MTLResource>>>;
void CollectResourcesForCompute(id<MTLCommandEncoder> encoder,
const ResourceBindingsSet& resourceBindingData,
Gems/Atom/RHI/Metal/Code/Source/RHI/AsyncUploadQueue.cpp
+31 -11
View File
@@ -85,16 +85,36 @@ namespace AZ
uint64_t AsyncUploadQueue::QueueUpload(const RHI::BufferStreamRequest& uploadRequest)
{
uint64_t queueValue = m_uploadFence.Increment();
Buffer& destBuffer = static_cast<Buffer&>(*uploadRequest.m_buffer);
const MemoryView& destMemoryView = destBuffer.GetMemoryView();
MTLStorageMode mtlStorageMode = destBuffer.GetMemoryView().GetStorageMode();
RHI::BufferPool& bufferPool = static_cast<RHI::BufferPool&>(*destBuffer.GetPool());
// No need to use staging buffers since it's host memory.
// We just map, copy and then unmap.
if(mtlStorageMode == MTLStorageModeShared || mtlStorageMode == GetCPUGPUMemoryMode())
{
RHI::BufferMapRequest mapRequest;
mapRequest.m_buffer = uploadRequest.m_buffer;
mapRequest.m_byteCount = uploadRequest.m_byteCount;
mapRequest.m_byteOffset = uploadRequest.m_byteOffset;
RHI::BufferMapResponse mapResponse;
bufferPool.MapBuffer(mapRequest, mapResponse);
::memcpy(mapResponse.m_data, uploadRequest.m_sourceData, uploadRequest.m_byteCount);
bufferPool.UnmapBuffer(*uploadRequest.m_buffer);
if (uploadRequest.m_fenceToSignal)
{
uploadRequest.m_fenceToSignal->SignalOnCpu();
}
return m_uploadFence.GetPendingValue();
}
const MemoryView& memoryView = static_cast<Buffer&>(*uploadRequest.m_buffer).GetMemoryView();
RHI::Ptr<Memory> buffer = memoryView.GetMemory();
Fence* fenceToSignal = nullptr;
uint64_t fenceToSignalValue = 0;
size_t byteCount = uploadRequest.m_byteCount;
size_t byteOffset = memoryView.GetOffset() + uploadRequest.m_byteOffset;
size_t byteOffset = destMemoryView.GetOffset() + uploadRequest.m_byteOffset;
uint64_t queueValue = m_uploadFence.Increment();
const uint8_t* sourceData = reinterpret_cast<const uint8_t*>(uploadRequest.m_sourceData);
if (uploadRequest.m_fenceToSignal)
@@ -125,11 +145,11 @@ namespace AZ
}
id<MTLBlitCommandEncoder> blitEncoder = [framePacket->m_mtlCommandBuffer blitCommandEncoder];
[blitEncoder copyFromBuffer:framePacket->m_stagingResource
sourceOffset:0
toBuffer:buffer->GetGpuAddress<id<MTLBuffer>>()
destinationOffset:byteOffset + pendingByteOffset
size:bytesToCopy];
[blitEncoder copyFromBuffer: framePacket->m_stagingResource
sourceOffset: 0
toBuffer: destMemoryView.GetGpuAddress<id<MTLBuffer>>()
destinationOffset: byteOffset + pendingByteOffset
size: bytesToCopy];
[blitEncoder endEncoding];
blitEncoder = nil;
Gems/Atom/RHI/Metal/Code/Source/RHI/BufferPoolResolver.cpp
+6 -4
View File
@@ -40,9 +40,8 @@ namespace AZ
buffer->m_pendingResolves++;
uploadRequest.m_attachmentBuffer = buffer;
uploadRequest.m_byteOffset = request.m_byteOffset;
uploadRequest.m_byteOffset = buffer->GetMemoryView().GetOffset() + request.m_byteOffset;
uploadRequest.m_stagingBuffer = stagingBuffer;
uploadRequest.m_byteSize = request.m_byteCount;
return stagingBuffer->GetMemoryView().GetCpuAddress();
}
@@ -64,12 +63,15 @@ namespace AZ
AZ_Assert(stagingBuffer, "Staging Buffer is null.");
AZ_Assert(destBuffer, "Attachment Buffer is null.");
//Inform the GPU that the CPU has modified the staging buffer.
Platform::SynchronizeBufferOnCPU(stagingBuffer->GetMemoryView().GetGpuAddress<id<MTLBuffer>>(), stagingBuffer->GetMemoryView().GetOffset(), stagingBuffer->GetMemoryView().GetSize());
RHI::CopyBufferDescriptor copyDescriptor;
copyDescriptor.m_sourceBuffer = stagingBuffer;
copyDescriptor.m_sourceOffset = 0;
copyDescriptor.m_sourceOffset = stagingBuffer->GetMemoryView().GetOffset();
copyDescriptor.m_destinationBuffer = destBuffer;
copyDescriptor.m_destinationOffset = static_cast<uint32_t>(packet.m_byteOffset);
copyDescriptor.m_size = static_cast<uint32_t>(packet.m_byteSize);
copyDescriptor.m_size = stagingBuffer->GetMemoryView().GetSize();
commandList.Submit(RHI::CopyItem(copyDescriptor));
device.QueueForRelease(stagingBuffer->GetMemoryView());
Gems/Atom/RHI/Metal/Code/Source/RHI/BufferPoolResolver.h
-1
View File
@@ -54,7 +54,6 @@ namespace AZ
Buffer* m_attachmentBuffer = nullptr;
RHI::Ptr<Buffer> m_stagingBuffer;
size_t m_byteOffset = 0;
size_t m_byteSize = 0;
};
AZStd::mutex m_uploadPacketsLock;
Gems/Atom/RHI/Metal/Code/Source/RHI/MemoryPageAllocator.cpp
+1 -1
View File
@@ -44,7 +44,7 @@ namespace AZ
if (memoryView.IsValid())
{
heapMemoryUsage.m_residentInBytes += m_descriptor.m_pageSizeInBytes;
memoryView.SetName("BufferPage");
memoryView.SetName(AZStd::string::format("BufferPage_%s", AZ::Uuid::CreateRandom().ToString<AZStd::string>().c_str()));
}
else
{