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

removed gfx940 and gfx941 (#1606)

Browse Source 
* removed gfx940 and gfx941

* removed gfx940 and gfx941

* Update "gfx94" to "gfx942" in init.cc

* Updated remaining "gfx94" updates to "gfx942"

* Update filenames and variables from gfx940 to gfx942

---------

Co-authored-by: akolliasAMD <akollias@amd.com>
This commit is contained in:
corey-derochie-amd
2025-03-20 09:34:53 -06:00
committed by GitHub
parent 6403afff4a
commit 6505639cf4
30 changed files with 74 additions and 79 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tools/rccl-prim-test/rccl_prim_test.cpp
+7 -7
View File
@@ -426,7 +426,7 @@ int main(int argc,char* argv[])
static const char *ring_4p3l = "0 1 2 3|0 1 3 2|0 2 1 3|0 2 3 1|0 3 1 2|0 3 2 1";
static const char *ring_8p1h = "0 1 3 2 4 5 7 6|6 7 5 4 2 3 1 0|0 1 5 4 6 7 3 2|2 3 7 6 4 5 1 0";
static const char *ring_16p1h = "0 1 3 2 6 7 15 14 10 11 9 8 12 13 5 4|0 1 2 3 7 6 13 12 8 9 10 11 15 14 5 4|0 2 3 7 6 14 15 11 10 8 9 13 12 4 5 1|4 5 13 12 8 9 11 10 14 15 7 6 2 3 1 0|4 5 14 15 11 10 9 8 12 13 6 7 3 2 1 0|1 5 4 12 13 9 8 10 11 15 14 6 7 3 2 0";
static const char *ring_gfx940_8p = "0 1 2 3 4 5 6 7|0 1 2 3 4 5 7 6|0 2 4 1 3 6 5 7|0 2 4 6 1 7 3 5|0 3 1 5 2 7 4 6|0 3 5 1 6 2 7 4|0 4 1 7 3 6 2 5|7 6 5 4 3 2 1 0|6 7 5 4 3 2 1 0|7 5 6 3 1 4 2 0|5 3 7 1 6 4 2 0|6 4 7 2 5 1 3 0|4 7 2 6 1 5 3 0|5 2 6 3 7 1 4 0";
static const char *ring_gfx942_8p = "0 1 2 3 4 5 6 7|0 1 2 3 4 5 7 6|0 2 4 1 3 6 5 7|0 2 4 6 1 7 3 5|0 3 1 5 2 7 4 6|0 3 5 1 6 2 7 4|0 4 1 7 3 6 2 5|7 6 5 4 3 2 1 0|6 7 5 4 3 2 1 0|7 5 6 3 1 4 2 0|5 3 7 1 6 4 2 0|6 4 7 2 5 1 3 0|4 7 2 6 1 5 3 0|5 2 6 3 7 1 4 0";
setupPeers(connection_info, &is_xgmi);
if (!r) {
@@ -436,8 +436,8 @@ int main(int argc,char* argv[])
if (nGpu == 8 && !cr8g) {
hipDeviceProp_t prop;
HIPCHECK(hipGetDeviceProperties(&prop, 0));
if (strncmp(prop.gcnArchName, "gfx94", 5) == 0) {
r = (char *)ring_gfx940_8p;
if (strncmp(prop.gcnArchName, "gfx942", 6) == 0) {
r = (char *)ring_gfx942_8p;
if(!workgroups) workgroups = 28;
} else {
r = (char *)ring_8p1h;
@@ -521,11 +521,11 @@ int main(int argc,char* argv[])
profiling_data[i] = (struct profiling_data_t *)malloc(sizeof(struct profiling_data_t)*iters);
HIPCHECK(hipMalloc((void**) &d_profiling_data[i], sizeof(struct profiling_data_t)*iters));
HIPCHECK(hipExtMallocWithFlags((void**) &transfer_data[i], sizeof(struct transfer_data_t), strncmp(prop.gcnArchName, "gfx94", 5) == 0 ? hipDeviceMallocUncached : hipDeviceMallocFinegrained));
HIPCHECK(hipExtMallocWithFlags((void**) &transfer_data[i], sizeof(struct transfer_data_t), strncmp(prop.gcnArchName, "gfx942", 6) == 0 ? hipDeviceMallocUncached : hipDeviceMallocFinegrained));
for (int j = 0; j < workgroups; j++) {
HIPCHECK(hipExtMallocWithFlags((void**) &buff[i*MAX_WORKGROUPS+j], 2*N*sizeof(float), strncmp(prop.gcnArchName, "gfx94", 5) == 0 ? hipDeviceMallocUncached : hipDeviceMallocFinegrained));
HIPCHECK(hipExtMallocWithFlags((void**) &buff[i*MAX_WORKGROUPS+j], 2*N*sizeof(float), strncmp(prop.gcnArchName, "gfx942", 6) == 0 ? hipDeviceMallocUncached : hipDeviceMallocFinegrained));
// additional fine grained buffer for local doublecopy, only need 1 buffer (not used by remote)
HIPCHECK(hipExtMallocWithFlags((void**) &buff_fine[i*MAX_WORKGROUPS+j], N*sizeof(float), strncmp(prop.gcnArchName, "gfx94", 5) == 0 ? hipDeviceMallocUncached : hipDeviceMallocFinegrained));
HIPCHECK(hipExtMallocWithFlags((void**) &buff_fine[i*MAX_WORKGROUPS+j], N*sizeof(float), strncmp(prop.gcnArchName, "gfx942", 6) == 0 ? hipDeviceMallocUncached : hipDeviceMallocFinegrained));
HIPCHECK(hipMalloc((void**) &buff_coarse[i*MAX_WORKGROUPS+j], 2*N*sizeof(float)));
//randomize test data
hipLaunchKernelGGL(initTestDataKernel,
@@ -670,7 +670,7 @@ int main(int argc,char* argv[])
hipDeviceProp_t prop;
HIPCHECK(hipGetDeviceProperties(&prop, i));
double vega_gpu_rtc_freq, bw_std_dev = 0, mean_write_cycle = 0;
if (strncmp(prop.gcnArchName, "gfx94", 5) == 0)
if (strncmp(prop.gcnArchName, "gfx942", 6) == 0)
vega_gpu_rtc_freq = 1.0E8;
else
vega_gpu_rtc_freq = 2.5E7;