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

SWDEV-283515 - Fix crashing in kernel launch on MGPUS

Browse Source 
Fix wrong mixing of current device and stream device in
ihipModuleLaunchKernel() and hipLaunchCooperativeKernel().

Fix missing hipSetDevice() in hipMemcpyWithStream* tests.

Change-Id: I09333bb40d239bb42c832df5ea16d17eeaeff5e7
This commit is contained in:
Tao Sang
2021-04-27 19:03:22 -04:00
committed by Tao Sang
parent 6b14050424
commit a95ff95bf7
7 changed files with 25 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/src/runtimeApi/memory/hipMemcpyWithStreamMultiThread.cpp
+2 -4
View File
@@ -463,6 +463,7 @@ void HipMemcpyWithStreamMultiThreadtests::TestkindDefaultForDtoD(void) {
int *A_h[numDevices], *B_h[numDevices], *C_h[numDevices];
// Initialize and create the host and device elements for first device
HIPCHECK(hipSetDevice(0));
HipTest::initArrays(&A_d[0], &B_d[0], &C_d[0],
&A_h[0], &B_h[0], &C_h[0], N, false);
@@ -481,7 +482,6 @@ void HipMemcpyWithStreamMultiThreadtests::TestkindDefaultForDtoD(void) {
HIPCHECK(hipStreamCreate(&stream[i]));
}
HIPCHECK(hipSetDevice(0));
HIPCHECK(hipMemcpyWithStream(A_d[0], A_h[0], Nbytes,
hipMemcpyHostToDevice, stream[0]));
HIPCHECK(hipMemcpyWithStream(B_d[0], B_h[0], Nbytes,
@@ -491,7 +491,6 @@ void HipMemcpyWithStreamMultiThreadtests::TestkindDefaultForDtoD(void) {
// using hipMemcpyDefault kind that is numDevices in the setup.
// 1st GPU start numbering from 0,1,2..n etc.
for (int i=1; i < numDevices; ++i) {
HIPCHECK(hipSetDevice(i));
HIPCHECK(hipMemcpyWithStream(A_d[i], A_d[0], Nbytes,
hipMemcpyDefault, stream[i]));
HIPCHECK(hipMemcpyWithStream(B_d[i], B_d[0], Nbytes,
@@ -499,14 +498,13 @@ void HipMemcpyWithStreamMultiThreadtests::TestkindDefaultForDtoD(void) {
}
for (int i=0; i < numDevices; ++i) {
HIPCHECK(hipSetDevice(i));
hipLaunchKernelGGL(HipTest::vectorADD, dim3(blocks), dim3(threadsPerBlock),
0, stream[i], static_cast<const int*>(A_d[i]),
static_cast<const int*>(B_d[i]), C_d[i], N);
}
for (int i=0; i < numDevices; ++i) {
HIPCHECK(hipSetDevice(i));
HIPCHECK(hipSetDevice(i)); // hipMemcpy will be on this device
HIPCHECK(hipStreamSynchronize(stream[i]));
HIPCHECK(hipMemcpy(C_h[i], C_d[i], Nbytes, hipMemcpyDeviceToHost));
// Output of each GPU is getting validated with input of 1st GPU.