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

SWDEV-383056 - Don't sync with dst device for hipMemcpyAsync.

Browse Source 
Change-Id: I28530e6bd870d617507592576295fc9e7eed1475
This commit is contained in:
Jaydeep Patel
2023-02-27 17:23:01 +00:00
committed by Jaydeepkumar Patel
parent 88dc5cd386
commit 57fa5938fe
6 changed files with 16 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files
hipamd/src/hip_memory.cpp
+7 -12
View File
@@ -451,7 +451,7 @@ void ihipHtoHMemcpy(void* dst, const void* src, size_t sizeBytes, hip::Stream& s
}
// ================================================================================================
hipError_t ihipMemcpy(void* dst, const void* src, size_t sizeBytes, hipMemcpyKind kind,
hip::Stream& stream, bool isAsync = false) {
hip::Stream& stream, bool isHostAsync, bool isGPUAsync) {
hipError_t status;
if (sizeBytes == 0) {
// Skip if nothing needs writing.
@@ -464,7 +464,6 @@ hipError_t ihipMemcpy(void* dst, const void* src, size_t sizeBytes, hipMemcpyKin
if (src == dst && kind == hipMemcpyDefault) {
return hipSuccess;
}
bool isP2P = false;
size_t sOffset = 0;
amd::Memory* srcMemory = getMemoryObject(src, sOffset);
size_t dOffset = 0;
@@ -473,24 +472,20 @@ hipError_t ihipMemcpy(void* dst, const void* src, size_t sizeBytes, hipMemcpyKin
ihipHtoHMemcpy(dst, src, sizeBytes, stream);
return hipSuccess;
} else if ((srcMemory == nullptr) && (dstMemory != nullptr)) {
isAsync = false;
isHostAsync = false;
} else if ((srcMemory != nullptr) && (dstMemory == nullptr)) {
isAsync = false;
} else if ((srcMemory->getContext().devices()[0] != dstMemory->getContext().devices()[0]) &&
(srcMemory->getContext().devices().size() == 1) &&
(dstMemory->getContext().devices().size() == 1)) {
isAsync = true;
isP2P = true;
isHostAsync = false;
}
amd::Command* command = nullptr;
status = ihipMemcpyCommand(command, dst, src, sizeBytes, kind, stream, isAsync);
status = ihipMemcpyCommand(command, dst, src, sizeBytes, kind, stream, isHostAsync);
if (status != hipSuccess) {
return status;
}
command->enqueue();
if (!isAsync) {
if (!isHostAsync) {
command->awaitCompletion();
} else if (isP2P) {
} else if (!isGPUAsync) {
hip::Stream* pStream = hip::getNullStream(dstMemory->getContext());
amd::Command::EventWaitList waitList;
waitList.push_back(command);