SWDEV-451594 - Fix HDP reg readback

Change-Id: I478a968330f85c3b60ff39fb40bf3cd91acd610e


[ROCm/clr commit: 1ba74c3ce3]
このコミットが含まれているのは:
Saleel Kudchadker
2024-05-14 17:52:14 +00:00
committed by Maneesh Gupta
コミット efc2de44f0
2個のファイルの変更5行の追加5行の削除
+3 -3
ファイルの表示
@@ -403,15 +403,15 @@ hipError_t GraphExec::CaptureAQLPackets() {
if (kernArgImpl == KernelArgImpl::DeviceKernelArgsHDP) {
*device->info().hdpMemFlushCntl = 1u;
volatile auto kSentinel = *device->info().hdpMemFlushCntl;
auto kSentinel = *reinterpret_cast<volatile int*>(device->info().hdpMemFlushCntl);
} else if (kernArgImpl == KernelArgImpl::DeviceKernelArgsReadback &&
kernarg_pool_size_graph_ != 0) {
address dev_ptr = kernarg_pool_graph_ + kernarg_pool_size_graph_;
volatile auto kSentinel = *(dev_ptr - 1);
auto kSentinel = *reinterpret_cast<volatile int*>(dev_ptr);
_mm_sfence();
*(dev_ptr - 1) = kSentinel;
_mm_mfence();
kSentinel = *(dev_ptr - 1);
kSentinel = *reinterpret_cast<volatile int*>(dev_ptr - 1);
}
}
+2 -2
ファイルの表示
@@ -3239,13 +3239,13 @@ bool VirtualGPU::submitKernelInternal(const amd::NDRangeContainer& sizes,
if (kernArgImpl == KernelArgImpl::DeviceKernelArgsHDP) {
*dev().info().hdpMemFlushCntl = 1u;
volatile auto kSentinel = *dev().info().hdpMemFlushCntl;
auto kSentinel = *reinterpret_cast<volatile int*>(dev().info().hdpMemFlushCntl);
} else if (kernArgImpl == KernelArgImpl::DeviceKernelArgsReadback &&
argSize != 0) {
_mm_sfence();
*(argBuffer + argSize - 1) = *(parameters + argSize - 1);
_mm_mfence();
volatile auto kSentinel = *(argBuffer + argSize - 1);
auto kSentinel = *reinterpret_cast<volatile address*>(argBuffer + argSize - 1);
}
}
}