leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 29 Packages Projects Releases Wiki Activity

support for K=0 for sm100 GG (#2746)

Browse Source
This commit is contained in:
ANIKET SHIVAM
2025-11-04 11:25:39 -05:00
committed by GitHub
parent 8afb19d904
commit 020c700e97
8 changed files with 72 additions and 43 deletions
Download Patch File Download Diff File Expand all files Collapse all files
examples/75_blackwell_grouped_gemm/75_blackwell_grouped_gemm.cu
+6 -6
View File
@@ -317,14 +317,14 @@ struct Options {
int m = cmd_line_m;
int n = cmd_line_n;
int k = cmd_line_k;
if (m < 1) {
m = alignment * ((rand() % 64) + 1);
if (m < 0) {
m = alignment * ((rand() % 64));
}
if (n < 1) {
n = alignment * ((rand() % 64) + 1);
if (n < 0) {
n = alignment * ((rand() % 64));
}
if (k < 1) {
k = alignment * ((rand() % 64) + 1);
if (k < 0) {
k = alignment * ((rand() % 64));
}
problem_sizes_host.push_back({m, n, k});
}
examples/75_blackwell_grouped_gemm/75_blackwell_grouped_gemm_block_scaled.cu
+6 -6
View File
@@ -379,14 +379,14 @@ struct Options {
int m = cmd_line_m;
int n = cmd_line_n;
int k = cmd_line_k;
if (m < 1) {
m = alignment * ((rand() % 64) + 1);
if (m < 0) {
m = alignment * ((rand() % 64));
}
if (n < 1) {
n = alignment * ((rand() % 64) + 1);
if (n < 0) {
n = alignment * ((rand() % 64));
}
if (k < 1) {
k = alignment * ((rand() % 64) + 1);
if (k < 0) {
k = alignment * ((rand() % 64));
}
problem_sizes_host.push_back({m, n, k});
}
examples/75_blackwell_grouped_gemm/CMakeLists.txt
+1 -1
View File
@@ -38,7 +38,7 @@ set(TEST_EPILOGUE --alpha=0.5 --beta=0.5 --iterations=0)
set(TEST_EPILOGUE_LARGE_GROUP --alpha=1.5 --beta=2.0 --groups=50 --iterations=0) # Random problem sizes
set(TEST_EPILOGUE_OP --beta=0.5 --iterations=1) # Random problem sizes
set(TEST_EPILOGUE_OP_LARGE_GROUP --alpha=1.5 --iterations=1) # Random problem sizes
set(TEST_EPILOGUE_OP_LARGE_GROUP --alpha=1.5 --groups=50 --iterations=1) # Random problem sizes
set(TEST_FIXED --m=2048 --n=5120 --k=8192 --iterations=0) # Fixed problem sizes
set(TEST_FIXED_LARGE_GROUP --m=2048 --n=512 --k=512 --groups=51 --iterations=0) # Fixed problem sizes