AI/rocm-systems
2
0
Garpu 0
Kode Masalah Tarik Permintaan Actions Packages Projects Rilis Wiki Kegiatan

SWDEV-428408 - Add waitingSignal for hsa_amd_memory_async_copy calls in hsaCopyStaged

Telusuri Sumber 
Change-Id: I3c42ef1ef3ed2f0b00f0a50d402a32106e5978ba
This commit is contained in:
Ioannis Assiouras
2023-10-23 00:19:03 +01:00
orang tua d4799b2a3f
melakukan b0c9fb84fd
1 mengubah file dengan 8 tambahan dan 4 penghapusan
Download Patch File Download Diff File Expand all files Collapse all files
rocclr/device/rocm/rocblit.cpp
+8 -4
Melihat File
@@ -802,11 +802,13 @@ bool DmaBlitManager::hsaCopyStaged(const_address hostSrc, address hostDst, size_
engine = HwQueueEngine::SdmaWrite;
}
gpu().Barriers().SetActiveEngine(engine);
auto wait_events = gpu().Barriers().WaitingSignal(engine);
hsa_signal_t active = gpu().Barriers().ActiveSignal(kInitSignalValueOne, gpu().timestamp());
memcpy(hsaBuffer, hostSrc + offset, size);
status = hsa_amd_memory_async_copy(hostDst + offset, dev().getBackendDevice(), hsaBuffer,
srcAgent, size, 0, nullptr, active);
status = hsa_amd_memory_async_copy(
hostDst + offset, dev().getBackendDevice(), hsaBuffer, srcAgent, size,
wait_events.size(), wait_events.data(), active);
ClPrint(amd::LOG_DEBUG, amd::LOG_COPY,
"HSA Async Copy staged H2D dst=0x%zx, src=0x%zx, size=%ld, completion_signal=0x%zx",
hostDst + offset, hsaBuffer, size, active.handle);
@@ -834,11 +836,13 @@ bool DmaBlitManager::hsaCopyStaged(const_address hostSrc, address hostDst, size_
engine = HwQueueEngine::SdmaRead;
}
gpu().Barriers().SetActiveEngine(engine);
auto wait_events = gpu().Barriers().WaitingSignal(engine);
hsa_signal_t active = gpu().Barriers().ActiveSignal(kInitSignalValueOne, gpu().timestamp());
// Copy data from Device to Host
status = hsa_amd_memory_async_copy(hsaBuffer, dstAgent, hostSrc + offset,
dev().getBackendDevice(), size, 0, nullptr, active);
status = hsa_amd_memory_async_copy(
hsaBuffer, dstAgent, hostSrc + offset, dev().getBackendDevice(), size,
wait_events.size(), wait_events.data(), active);
ClPrint(amd::LOG_DEBUG, amd::LOG_COPY,
"HSA Async Copy staged D2H dst=0x%zx, src=0x%zx, size=%ld, completion_signal=0x%zx",
hsaBuffer, hostSrc + offset, size, active.handle);