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

[TOOLS] Update p2p-latency-test for gfx950 (#1730)

Browse Source
This commit is contained in:
Nilesh M Negi
2025-07-10 12:13:29 -05:00
committed by GitHub
parent 2c099fe29a
commit f839e4edef
5 changed files with 42 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tools/p2p-latency-test/p2p_latency_test.cpp
+5 -4
View File
@@ -1,5 +1,6 @@
/*************************************************************************
* Copyright (c) Microsoft Corporation.
* Modifications Copyright (c) 2023 Advanced Micro Devices, Inc. All rights reserved.
* Licensed under the MIT License.
************************************************************************/
@@ -86,7 +87,7 @@ 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, strncmp(prop[0].gcnArchName, "gfx942", 6) == 0 ? hipDeviceMallocUncached : hipDeviceMallocFinegrained));
HIPCHECK(hipExtMallocWithFlags((void**)&flag[0], HIP_IPC_MEM_MIN_SIZE, (strncmp(prop[0].gcnArchName, "gfx942", 6) == 0 || strncmp(prop[0].gcnArchName, "gfx950", 6) == 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 +96,7 @@ 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, strncmp(prop[1].gcnArchName, "gfx942", 6) == 0 ? hipDeviceMallocUncached : hipDeviceMallocFinegrained));
HIPCHECK(hipExtMallocWithFlags((void**)&flag[1], HIP_IPC_MEM_MIN_SIZE, (strncmp(prop[1].gcnArchName, "gfx942", 6) == 0 || strncmp(prop[1].gcnArchName, "gfx950", 6) == 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 +110,11 @@ int main(int argc, char** argv) {
double vega_gpu_rtc_freq;
HIPCHECK(hipStreamSynchronize(stream[0]));
vega_gpu_rtc_freq = strncmp(prop[0].gcnArchName, "gfx942", 6) == 0 ? 1.0E8 : 2.5E7;
vega_gpu_rtc_freq = (strncmp(prop[0].gcnArchName, "gfx942", 6) == 0 || strncmp(prop[0].gcnArchName, "gfx950", 6) == 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 = strncmp(prop[1].gcnArchName, "gfx942", 6) == 0 ? 1.0E8 : 2.5E7;
vega_gpu_rtc_freq = (strncmp(prop[1].gcnArchName, "gfx942", 6) == 0 || strncmp(prop[1].gcnArchName, "gfx950", 6) == 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]));