github/ROCm
1
1
Fork 0
mirror of https://github.com/ROCm/ROCm.git synced 2026-04-05 03:01:17 -04:00
Code Issues Packages Projects Releases Wiki Activity

Disable pipelining (#276)

Browse Source 
This PR sets default value of pipeline stages to 1 for amd.
Add explicit num stages in test_gemm test
This commit is contained in:
Alexander Efimov
2023-08-01 21:54:17 +02:00
committed by GitHub
parent 82db24cde7
commit a318bc5777
2 changed files with 3 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
python/test/unit/language/test_core_amd.py
View File

@@ -1511,7 +1511,8 @@ def test_gemm(SIZE_M, SIZE_N, SIZE_K, NUM_WARPS, BLOCK_SIZE_M, BLOCK_SIZE_N, BLO
stride_cm=c.stride(0), stride_cn=c.stride(1),
M=a.shape[0], N=b.shape[1], K=a.shape[1],
BLOCK_SIZE_M=BLOCK_SIZE_M, BLOCK_SIZE_N=BLOCK_SIZE_N, BLOCK_SIZE_K=BLOCK_SIZE_K,
num_warps=NUM_WARPS)
num_warps=NUM_WARPS,
num_stages=2)
golden = torch.matmul(a, b)
# It's not easy to get a proper error threshold in different size

2
python/triton/compiler/compiler.py
View File

@@ -399,7 +399,7 @@ def compile(fn, **kwargs):
context = ir.context()
constants = kwargs.get("constants", dict())
num_warps = kwargs.get("num_warps", 4)
num_stages = kwargs.get("num_stages", 3 if is_cuda and arch >= 75 else 2)
num_stages = kwargs.get("num_stages", 3 if is_cuda and arch >= 75 else (1 if is_hip else 2))
extern_libs = kwargs.get("extern_libs", dict())
if extern_libs is None:
extern_libs = dict()