NASP/cuda-samples
85
0
Fork 0
mirror of https://github.com/NVIDIA/cuda-samples.git synced 2026-10-25 22:28:17 +08:00
Code Issues Packages Projects Releases Wiki Activity

Bug 5196362, 5184356, 5212196, 5214258 and 5214259: Update sameples for CUDA13.0 API changes

Browse Source
This commit is contained in:
shawnz
2025-04-10 18:27:18 +08:00
parent 278f4adbd2
commit 02fdb070ad
27 changed files with 181 additions and 116 deletions
Download Patch File Download Diff File Expand all files Collapse all files

25
Samples/6_Performance/UnifiedMemoryPerf/matrixMultiplyPerf.cu
View File

@@ -335,9 +335,11 @@ void runMatrixMultiplyKernel(unsigned int matrixDim,
checkCudaErrors(cudaMallocManaged(&dptrA, size));
checkCudaErrors(cudaMallocManaged(&dptrB, size));
checkCudaErrors(cudaMallocManaged(&dptrC, size));
checkCudaErrors(cudaMemPrefetchAsync(dptrA, size, cudaCpuDeviceId));
checkCudaErrors(cudaMemPrefetchAsync(dptrB, size, cudaCpuDeviceId));
checkCudaErrors(cudaMemPrefetchAsync(dptrC, size, cudaCpuDeviceId));
cudaMemLocation hostLoc;
hostLoc.type = cudaMemLocationTypeHost;
checkCudaErrors(cudaMemPrefetchAsync(dptrA, size, hostLoc, 0));
checkCudaErrors(cudaMemPrefetchAsync(dptrB, size, hostLoc, 0));
checkCudaErrors(cudaMemPrefetchAsync(dptrC, size, hostLoc, 0));
}
else {
checkCudaErrors(cudaMallocManaged(&dptrA, size, cudaMemAttachHost));
@@ -402,9 +404,12 @@ void runMatrixMultiplyKernel(unsigned int matrixDim,
}
if (hintsRequired) {
if (deviceProp.concurrentManagedAccess) {
checkCudaErrors(cudaMemPrefetchAsync(dptrA, size, device_id, streamToRunOn));
checkCudaErrors(cudaMemPrefetchAsync(dptrB, size, device_id, streamToRunOn));
checkCudaErrors(cudaMemPrefetchAsync(dptrC, size, device_id, streamToRunOn));
cudaMemLocation deviceLoc;
deviceLoc.type = cudaMemLocationTypeDevice;
deviceLoc.id = device_id;
checkCudaErrors(cudaMemPrefetchAsync(dptrA, size, deviceLoc, 0, streamToRunOn));
checkCudaErrors(cudaMemPrefetchAsync(dptrB, size, deviceLoc, 0, streamToRunOn));
checkCudaErrors(cudaMemPrefetchAsync(dptrC, size, deviceLoc, 0, streamToRunOn));
}
else {
checkCudaErrors(cudaStreamAttachMemAsync(streamToRunOn, dptrA, 0, cudaMemAttachGlobal));
@@ -437,9 +442,11 @@ void runMatrixMultiplyKernel(unsigned int matrixDim,
sdkStartTimer(&gpuTransferCallsTimer);
if (hintsRequired) {
if (deviceProp.concurrentManagedAccess) {
checkCudaErrors(cudaMemPrefetchAsync(dptrA, size, cudaCpuDeviceId));
checkCudaErrors(cudaMemPrefetchAsync(dptrB, size, cudaCpuDeviceId));
checkCudaErrors(cudaMemPrefetchAsync(dptrC, size, cudaCpuDeviceId));
cudaMemLocation hostLoc;
hostLoc.type = cudaMemLocationTypeHost;
checkCudaErrors(cudaMemPrefetchAsync(dptrA, size, hostLoc, 0));
checkCudaErrors(cudaMemPrefetchAsync(dptrB, size, hostLoc, 0));
checkCudaErrors(cudaMemPrefetchAsync(dptrC, size, hostLoc, 0));
}
else {
checkCudaErrors(cudaStreamAttachMemAsync(streamToRunOn, dptrA, 0, cudaMemAttachHost));