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

SWDEV-301667 - Improve logging

Browse Source 
Change-Id: Ifa6da876b85cb503967cf09aac6d477b10db8e63


[ROCm/clr commit: feca11d5e3]
This commit is contained in:
Saleel Kudchadker
2022-11-02 19:51:33 -07:00
parent 994c9384e7
commit b157c8eeea
2 changed files with 6 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/clr/rocclr/device/rocm/rocblit.cpp
+5 -2
View File
@@ -1812,12 +1812,15 @@ bool KernelBlitManager::writeBuffer(const void* srcHost, device::Memory& dstMemo
bool result = false;
if (dev().info().largeBar_ && size[0] <= kMaxH2dMemcpySize) {
if ((dstMemory.owner()->getHostMem() == nullptr) && (dstMemory.owner()->getSvmPtr() != nullptr)) {
if ((dstMemory.owner()->getHostMem() == nullptr) &&
(dstMemory.owner()->getSvmPtr() != nullptr)) {
// CPU read ahead, hence release GPU memory
ClPrint(amd::LOG_DEBUG, amd::LOG_COPY, "Host memcpy for map wait_event");
gpu().releaseGpuMemoryFence();
char* dst = reinterpret_cast<char*>(dstMemory.owner()->getSvmPtr());
std::memcpy(dst + origin[0], srcHost, size[0]);
// Set hasPendingDispatch_ flag. Then releaseGpuMemoryFence() will use barrier to invalidate cache
// Set hasPendingDispatch_ flag. Then releaseGpuMemoryFence() will use barrier
// to invalidate cache
gpu().hasPendingDispatch();
gpu().releaseGpuMemoryFence();
return true;