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

Enable gfx94x (#808) (#816)

Browse Source 
(cherry picked from commit 94da229a7788d74685d1591a4e75a8341de64f41)

[ROCm/rccl commit: a7fcd58a97]
This commit is contained in:
Wenkai Du
2023-07-21 07:31:27 -07:00
committed by GitHub
parent f4b106ec94
commit dfda1d6fab
15 changed files with 159 additions and 56 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rccl/src/misc/msccl/msccl_lifecycle.cc
+2 -1
View File
@@ -23,6 +23,7 @@
#include "msccl/msccl_status.h"
RCCL_PARAM(MscclEnabled, "MSCCL_ENABLE", 1);
RCCL_PARAM(MscclForceEnabled, "MSCCL_FORCE_ENABLE", 0);
static const char* mscclAlgoFilePathEnv = "MSCCL_ALGO_FILE_PATH";
static std::atomic<bool> mscclInitialized;
static bool mscclSchedulerTriedLoadAlgo = false;
@@ -257,7 +258,7 @@ static ncclResult_t mscclSchedulerSelectAlgo(struct mscclSavedSchedulerParam* pa
if (status.mscclSchedulerPtr) {
NCCLCHECK(status.mscclSchedulerPtr->selectAlgo(&(param->p)));
} else {
if (param->comm->topo->mscclEnabled) {
if (param->comm->topo->mscclEnabled || rcclParamMscclForceEnabled()) {
NCCLCHECK(mscclInternalSchedulerSelectAlgo(&(param->p)));
} else {
param->p.scheduled = false;
projects/rccl/src/misc/npkit.cc
+7 -1
View File
@@ -120,7 +120,13 @@ ncclResult_t NpKit::Dump(const std::string& dump_dir) {
dump_file_path = dump_dir;
dump_file_path += "/gpu_clock_rate_rank_";
dump_file_path += std::to_string(rank_);
constexpr int vega_gpu_rtc_freq_in_khz = 25000;
hipDeviceProp_t devProp;
int vega_gpu_rtc_freq_in_khz;
CUDACHECK(hipGetDeviceProperties(&devProp, 0));
if (devProp.gcnArch/10 == 94)
vega_gpu_rtc_freq_in_khz = 100000;
else
vega_gpu_rtc_freq_in_khz = 25000;
std::string clock_rate_str = std::to_string(vega_gpu_rtc_freq_in_khz);
auto gpu_clock_rate_file = std::fstream(dump_file_path, std::ios::out);
gpu_clock_rate_file.write(clock_rate_str.c_str(), clock_rate_str.length());