AI/rocm-systems
2
0
Vork 0
Code Kwesties Pull-aanvragen Actions Paketten Projecten Publicaties Wiki Activiteit

p2p-latency-tests: fix build by switching to gcnArchName (#1030)

Bladeren bron 
* p2p-latency-tests: fix build by switching to gcnArchName

* rccl-prim-test: switch to gcnArchName
This commit is contained in:
Wenkai Du
2024-01-04 13:36:48 -08:00
gecommit door GitHub
bovenliggende abf265a911
commit cfc04a8aef
3 gewijzigde bestanden met toevoegingen van 13 en 13 verwijderingen
Download Patch-bestand Download Diff-bestand Expand all files Collapse all files
tools/p2p-latency-test/ll_latency_test.cpp
+4 -4
Bestand weergeven
@@ -147,7 +147,7 @@ int main(int argc, char** argv) {
HIPCHECK(hipStreamCreateWithFlags(&stream[0], hipStreamNonBlocking));
HIPCHECK(hipDeviceEnablePeerAccess(device_id[1], 0));
HIPCHECK(hipGetDeviceProperties(&prop[0], device_id[0]));
HIPCHECK(hipExtMallocWithFlags((void**)&flag[0], HIP_IPC_MEM_MIN_SIZE, prop[0].gcnArch / 10 == 94 ? hipDeviceMallocUncached : hipDeviceMallocFinegrained));
HIPCHECK(hipExtMallocWithFlags((void**)&flag[0], HIP_IPC_MEM_MIN_SIZE, strncmp(prop[0].gcnArchName, "gfx94", 5) == 0 ? hipDeviceMallocUncached : hipDeviceMallocFinegrained));
HIPCHECK(hipHostMalloc ((void**)&time_delta[0], sizeof(uint64_t), hipHostMallocDefault));
HIPCHECK(hipMalloc((void**)&abortFlag[0], sizeof(uint32_t)));
HIPCHECK(hipMemsetAsync(flag[0], 0, HIP_IPC_MEM_MIN_SIZE, stream[0]));
@@ -158,7 +158,7 @@ int main(int argc, char** argv) {
HIPCHECK(hipStreamCreateWithFlags(&stream[1], hipStreamNonBlocking));
HIPCHECK(hipDeviceEnablePeerAccess(device_id[0], 0));
HIPCHECK(hipGetDeviceProperties(&prop[1], device_id[1]));
HIPCHECK(hipExtMallocWithFlags((void**)&flag[1], HIP_IPC_MEM_MIN_SIZE, prop[1].gcnArch / 10 == 94 ? hipDeviceMallocUncached : hipDeviceMallocFinegrained));
HIPCHECK(hipExtMallocWithFlags((void**)&flag[1], HIP_IPC_MEM_MIN_SIZE, strncmp(prop[1].gcnArchName, "gfx94", 5) == 0 ? hipDeviceMallocUncached : hipDeviceMallocFinegrained));
HIPCHECK(hipHostMalloc((void**)&time_delta[1], sizeof(uint64_t), hipHostMallocDefault));
HIPCHECK(hipMalloc((void**)&abortFlag[1], sizeof(uint32_t)));
HIPCHECK(hipMemsetAsync(flag[1], 0, HIP_IPC_MEM_MIN_SIZE, stream[1]));
@@ -174,11 +174,11 @@ int main(int argc, char** argv) {
double vega_gpu_rtc_freq;
HIPCHECK(hipStreamSynchronize(stream[0]));
vega_gpu_rtc_freq = (prop[0].gcnArch / 10 == 94) ? 1.0E8 : 2.5E7;
vega_gpu_rtc_freq = strncmp(prop[0].gcnArchName, "gfx94", 5) == 0 ? 1.0E8 : 2.5E7;
fprintf(stdout, "One-way latency in us: %g\n", double(*time_delta[0]) * 1e6 / NUM_LOOPS_RUN / vega_gpu_rtc_freq / 2);
HIPCHECK(hipStreamSynchronize(stream[1]));
vega_gpu_rtc_freq = (prop[1].gcnArch / 10 == 94) ? 1.0E8 : 2.5E7;
vega_gpu_rtc_freq = strncmp(prop[1].gcnArchName, "gfx94", 5) == 0 ? 1.0E8 : 2.5E7;
fprintf(stdout, "One-way latency in us: %g\n", double(*time_delta[1]) * 1e6 / NUM_LOOPS_RUN / vega_gpu_rtc_freq / 2);
HIPCHECK(hipFree(flag[0]));
tools/p2p-latency-test/p2p_latency_test.cpp
+4 -4
Bestand weergeven
@@ -86,7 +86,7 @@ int main(int argc, char** argv) {
HIPCHECK(hipStreamCreateWithFlags(&stream[0], hipStreamNonBlocking));
HIPCHECK(hipDeviceEnablePeerAccess(device_id[1], 0));
HIPCHECK(hipGetDeviceProperties(&prop[0], device_id[0]));
HIPCHECK(hipExtMallocWithFlags((void**)&flag[0], HIP_IPC_MEM_MIN_SIZE, prop[0].gcnArch / 10 == 94 ? hipDeviceMallocUncached : hipDeviceMallocFinegrained));
HIPCHECK(hipExtMallocWithFlags((void**)&flag[0], HIP_IPC_MEM_MIN_SIZE, strncmp(prop[0].gcnArchName, "gfx94", 5) == 0 ? hipDeviceMallocUncached : hipDeviceMallocFinegrained));
HIPCHECK(hipMalloc((void**)&time_delta[0], HIP_IPC_MEM_MIN_SIZE));
HIPCHECK(hipMemsetAsync(flag[0], 0, HIP_IPC_MEM_MIN_SIZE, stream[0]));
HIPCHECK(hipStreamSynchronize(stream[0]));
@@ -95,7 +95,7 @@ int main(int argc, char** argv) {
HIPCHECK(hipStreamCreateWithFlags(&stream[1], hipStreamNonBlocking));
HIPCHECK(hipDeviceEnablePeerAccess(device_id[0], 0));
HIPCHECK(hipGetDeviceProperties(&prop[1], device_id[1]));
HIPCHECK(hipExtMallocWithFlags((void**)&flag[1], HIP_IPC_MEM_MIN_SIZE, prop[1].gcnArch / 10 == 94 ? hipDeviceMallocUncached : hipDeviceMallocFinegrained));
HIPCHECK(hipExtMallocWithFlags((void**)&flag[1], HIP_IPC_MEM_MIN_SIZE, strncmp(prop[1].gcnArchName, "gfx94", 5) == 0 ? hipDeviceMallocUncached : hipDeviceMallocFinegrained));
HIPCHECK(hipMalloc((void**)&time_delta[1], HIP_IPC_MEM_MIN_SIZE));
HIPCHECK(hipMemsetAsync(flag[1], 0, HIP_IPC_MEM_MIN_SIZE, stream[1]));
HIPCHECK(hipStreamSynchronize(stream[1]));
@@ -109,11 +109,11 @@ int main(int argc, char** argv) {
double vega_gpu_rtc_freq;
HIPCHECK(hipStreamSynchronize(stream[0]));
vega_gpu_rtc_freq = (prop[0].gcnArch / 10 == 94) ? 1.0E8 : 2.5E7;
vega_gpu_rtc_freq = strncmp(prop[0].gcnArchName, "gfx94", 5) == 0 ? 1.0E8 : 2.5E7;
fprintf(stdout, "One-way latency in us: %g\n", double(*time_delta[0]) * 1e6 / NUM_LOOPS_RUN / vega_gpu_rtc_freq / 2);
HIPCHECK(hipStreamSynchronize(stream[1]));
vega_gpu_rtc_freq = (prop[1].gcnArch / 10 == 94) ? 1.0E8 : 2.5E7;
vega_gpu_rtc_freq = strncmp(prop[1].gcnArchName, "gfx94", 5) == 0 ? 1.0E8 : 2.5E7;
fprintf(stdout, "One-way latency in us: %g\n", double(*time_delta[1]) * 1e6 / NUM_LOOPS_RUN / vega_gpu_rtc_freq / 2);
HIPCHECK(hipFree(flag[0]));