intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-03 14:55:24 +08:00
Code Issues Packages Projects Releases Wiki Activity

Detect GPU hang in command list immediate

Browse Source 
This change introduces checking of values returned
by blocking calls used in cmdlist_hw_immediate.inl.

Signed-off-by: Patryk Wrobel <patryk.wrobel@intel.com>
Related-To: NEO-6681
This commit is contained in:
Patryk Wrobel
2022-03-23 16:20:37 +00:00
committed by Compute-Runtime-Automation
parent f3bf5498a4
commit 14954acd12
3 changed files with 209 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
level_zero/core/source/cmdlist/cmdlist_hw_immediate.inl
View File

@@ -7,6 +7,7 @@
#pragma once
#include "shared/source/command_stream/wait_status.h"
#include "shared/source/helpers/hw_helper.h"
#include "shared/source/helpers/hw_info.h"
#include "shared/source/memory_manager/internal_allocation_storage.h"
@@ -113,7 +114,10 @@ ze_result_t CommandListCoreFamilyImmediate<gfxCoreFamily>::executeCommandListImm
if (this->isSyncModeQueue) {
auto timeoutMicroseconds = NEO::TimeoutControls::maxTimeout;
this->csr->waitForCompletionWithTimeout(false, timeoutMicroseconds, completionStamp.taskCount);
const auto waitStatus = this->csr->waitForCompletionWithTimeout(false, timeoutMicroseconds, completionStamp.taskCount);
if (waitStatus == NEO::WaitStatus::GpuHang) {
return ZE_RESULT_ERROR_DEVICE_LOST;
}
this->csr->getInternalAllocationStorage()->cleanAllocationList(completionStamp.taskCount, NEO::AllocationUsage::TEMPORARY_ALLOCATION);
}
@@ -180,7 +184,10 @@ ze_result_t CommandListCoreFamilyImmediate<gfxCoreFamily>::appendBarrier(
this->csr->flushNonKernelTask(nullptr, 0, 0, args, false, false, false);
if (this->isSyncModeQueue) {
auto timeoutMicroseconds = NEO::TimeoutControls::maxTimeout;
this->csr->waitForCompletionWithTimeout(false, timeoutMicroseconds, this->csr->peekTaskCount());
const auto waitStatus = this->csr->waitForCompletionWithTimeout(false, timeoutMicroseconds, this->csr->peekTaskCount());
if (waitStatus == NEO::WaitStatus::GpuHang) {
return ZE_RESULT_ERROR_DEVICE_LOST;
}
}
} else {
ret = CommandListCoreFamilyImmediate<gfxCoreFamily>::appendSignalEvent(hSignalEvent);
@@ -263,7 +270,10 @@ ze_result_t CommandListCoreFamilyImmediate<gfxCoreFamily>::appendSignalEvent(ze_
this->csr->flushNonKernelTask(&event->getAllocation(this->device), event->getGpuAddress(this->device), Event::STATE_SIGNALED, args, false, false, false);
if (this->isSyncModeQueue) {
auto timeoutMicroseconds = NEO::TimeoutControls::maxTimeout;
this->csr->waitForCompletionWithTimeout(false, timeoutMicroseconds, this->csr->peekTaskCount());
const auto waitStatus = this->csr->waitForCompletionWithTimeout(false, timeoutMicroseconds, this->csr->peekTaskCount());
if (waitStatus == NEO::WaitStatus::GpuHang) {
return ZE_RESULT_ERROR_DEVICE_LOST;
}
}
}
return ret;
@@ -289,7 +299,10 @@ ze_result_t CommandListCoreFamilyImmediate<gfxCoreFamily>::appendEventReset(ze_e
this->csr->flushNonKernelTask(&event->getAllocation(this->device), event->getGpuAddress(this->device), Event::STATE_CLEARED, args, false, false, false);
if (this->isSyncModeQueue) {
auto timeoutMicroseconds = NEO::TimeoutControls::maxTimeout;
this->csr->waitForCompletionWithTimeout(false, timeoutMicroseconds, this->csr->peekTaskCount());
const auto waitStatus = this->csr->waitForCompletionWithTimeout(false, timeoutMicroseconds, this->csr->peekTaskCount());
if (waitStatus == NEO::WaitStatus::GpuHang) {
return ZE_RESULT_ERROR_DEVICE_LOST;
}
}
}
return ret;