gaspersic/triton
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Fix for binop % (mod) unit test failures. (#13)

Browse Source 
If the either data type if fp, then fmod should be used for the
reference computation.
This commit is contained in:
rsanthanam-amd
2022-10-28 14:06:17 -05:00
committed by GitHub
parent 5f0d90db7e
commit 531ef18cb6
1 changed files with 1 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

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

@@ -234,7 +234,6 @@ def _mod_operation_ill_conditioned(dtype_x, dtype_y) -> bool:
('int64', 'float32'),
('int64', 'float64'),
('uint16', 'bfloat16'),
('uint16', 'float16'),
('uint16', 'float32'),
('uint32', 'bfloat16'),
('uint32', 'float16'),
@@ -258,7 +257,7 @@ def _mod_operation_ill_conditioned(dtype_x, dtype_y) -> bool:
])
def test_bin_op(dtype_x, dtype_y, op, device='cuda'):
expr = f' x {op} y'
if op == '%' and dtype_x in int_dtypes + uint_dtypes and dtype_y in int_dtypes + uint_dtypes:
if op == '%' and (dtype_x in dtypes and dtype_y in dtypes):
# LLVM has 'numpy.fmod', not 'numpy.remainder', semantics on integer remainders.
numpy_expr = 'np.fmod(x, y)'
elif op in ('/', '%') and dtype_x in ('int16', 'float16', 'bfloat16') and dtype_y in ('int16', 'float16', 'bfloat16'):