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

SWDEV-282354 - Some AMD GPUs cause clock64() to roll over faster, this changeset prevents infinite loops due to that.

Browse Source 
Change-Id: I19d026073afa16104ffbee4705b74d37003a04d9
This commit is contained in:
Sourabh Betigeri
2021-04-19 12:28:34 -07:00
committed by Sourabh Betigeri
parent 09a41f2cf4
commit 6bcbb61e5f
5 changed files with 114 additions and 40 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/src/runtimeApi/cooperativeGrps/multi_gpu_streams.cpp
+22 -4
View File
@@ -161,8 +161,17 @@ __global__ void test_coop_kernel(unsigned int loops, long long *array,
}
for (int i = 0; i < loops; i++) {
long long start_clock = clock64();
while (clock64() < (start_clock+1000000)) {}
long long time_diff = 0;
long long last_clock = clock64();
do {
long long cur_clock = clock64();
if (cur_clock > last_clock) {
time_diff += (cur_clock - last_clock);
}
// If it rolls over, we don't know how much to add to catch up.
// So just ignore those slipped cycles.
last_clock = cur_clock;
} while(time_diff < 1000000);
array[rank] += clock64();
}
}
@@ -171,8 +180,17 @@ __global__ void test_kernel(uint32_t loops, unsigned long long *array) {
unsigned int rank = blockIdx.x * blockDim.x + threadIdx.x;
for (int i = 0; i < loops; i++) {
long long start_clock = clock64();
while (clock64() < (start_clock+1000000)) {}
long long time_diff = 0;
long long last_clock = clock64();
do {
long long cur_clock = clock64();
if (cur_clock > last_clock) {
time_diff += (cur_clock - last_clock);
}
// If it rolls over, we don't know how much to add to catch up.
// So just ignore those slipped cycles.
last_clock = cur_clock;
} while(time_diff < 1000000);
array[rank] += clock64();
}
}