AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

P4 to Git Change 1092261 by gandryey@gera-dev-w7 on 2014/10/30 13:01:38

Browse Source 
ECR #304775 - Align the queue size to match the multidispatch scheduler requirements

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuvirtual.cpp#337 edit


[ROCm/clr commit: 7cac96b9a5]
This commit is contained in:
foreman
2014-10-30 13:11:11 -04:00
parent 9f3d6e0a69
commit 59b9cd758e
1 changed files with 8 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/clr/rocclr/runtime/device/gpu/gpuvirtual.cpp
+8 -3
View File
@@ -264,6 +264,12 @@ VirtualGPU::releasePinnedMem()
bool
VirtualGPU::createVirtualQueue(uint deviceQueueSize)
{
uint MinDeviceQueueSize = 16 * 1024;
deviceQueueSize = std::max(deviceQueueSize, MinDeviceQueueSize);
// Align the queue size for the multiple dispatch scheduler.
// Each thread works with 32 entries
deviceQueueSize = amd::alignUp(deviceQueueSize, sizeof(AmdAqlWrap) * 32);
if (deviceQueueSize_ == deviceQueueSize) {
return true;
}
@@ -281,19 +287,18 @@ VirtualGPU::createVirtualQueue(uint deviceQueueSize)
schedParamIdx_ = 0;
deviceQueueSize_ = 0;
}
uint numSlots = deviceQueueSize / sizeof(AmdAqlWrap);
uint allocSize = deviceQueueSize;
// Add the virtual queue header
allocSize += sizeof(AmdVQueueHeader);
allocSize = amd::alignUp(allocSize, 128);
allocSize = amd::alignUp(allocSize, sizeof(AmdAqlWrap));
uint argOffs = allocSize;
// Add the kernel arguments and wait events
uint singleArgSize = amd::alignUp(dev().info().maxParameterSize_ + 64 +
dev().settings().numWaitEvents_ * sizeof(uint64_t), 128);
dev().settings().numWaitEvents_ * sizeof(uint64_t), sizeof(AmdAqlWrap));
allocSize += singleArgSize * numSlots;
uint eventsOffs = allocSize;