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

Catch2 Test Fixes (#2961)

Browse Source 
* test fixes

* address PR comment

* PR comment fixing Nvidia pass

* add additional kernel launch checks

* pr comments
This commit is contained in:
Paulius Velesko
2022-10-01 01:07:25 +03:00
committed by GitHub
parent 609edf1c9e
commit da453211ed
94 changed files with 177 additions and 34 deletions
Download Patch File Download Diff File Expand all files Collapse all files
catch/unit/multiThread/hipMultiThreadStreams2.cc
+3
View File
@@ -56,6 +56,7 @@ void run1(size_t size, hipStream_t stream) {
HIPCHECK(hipMemcpyAsync(Bh, Ah, size, hipMemcpyHostToHost, stream));
HIPCHECK(hipMemcpyAsync(Cd, Bh, size, hipMemcpyHostToDevice, stream));
hipLaunchKernelGGL(HIP_KERNEL_NAME(Inc), dim3(N / 500), dim3(500), 0, stream, Cd);
HIP_CHECK(hipGetLastError());
HIPCHECK(hipMemcpyAsync(Dd, Cd, size, hipMemcpyDeviceToDevice, stream));
HIPCHECK(hipMemcpyAsync(Eh, Dd, size, hipMemcpyDeviceToHost, stream));
HIPCHECK(hipDeviceSynchronize());
@@ -97,7 +98,9 @@ void run(size_t size, hipStream_t stream1, hipStream_t stream2) {
HIPCHECK(hipMemcpyAsync(Cd, Bh, size, hipMemcpyHostToDevice, stream1));
HIPCHECK(hipMemcpyAsync(Cdd, Bhh, size, hipMemcpyHostToDevice, stream2));
hipLaunchKernelGGL(HIP_KERNEL_NAME(Inc), dim3(N / 500), dim3(500), 0, stream1, Cd);
HIP_CHECK(hipGetLastError());
hipLaunchKernelGGL(HIP_KERNEL_NAME(Inc), dim3(N / 500), dim3(500), 0, stream2, Cdd);
HIP_CHECK(hipGetLastError());
HIPCHECK(hipMemcpyAsync(Dd, Cd, size, hipMemcpyDeviceToDevice, stream1));
HIPCHECK(hipMemcpyAsync(Ddd, Cdd, size, hipMemcpyDeviceToDevice, stream2));
HIPCHECK(hipMemcpyAsync(Eh, Dd, size, hipMemcpyDeviceToHost, stream1));