[STYLE] run autopep8 and isort (#421)

Run:
```
isort ./python
autopep8 -i --ignore E501,E701,E731 $(find ./python/ -name '*.py')
```
with an `.isort.cfg` and then clean up a few warts. This PR should be a no-op; the idea is that this is all boring whitespace changes, and any config file changes will be in a different change to make it easier to review.
This commit is contained in:
Madeleine Thompson
2022-01-06 14:34:17 -08:00
committed by GitHub
parent 120cda015e
commit 8bf551ae7a
30 changed files with 742 additions and 623 deletions

View File

@@ -1,6 +1,7 @@
import torch
import triton
import pytest
import torch
import triton
@pytest.mark.parametrize("MODE", ["sdd", "dds", "dsd"])
@@ -71,7 +72,8 @@ def test_softmax(BLOCK, WIDTH, DTYPE):
# torch result
rx = triton.testing.mask_tensor(x, layout, BLOCK, value=float("-inf"))
# broadcast at_mask to the same shape as rx
if is_causal: at_mask = torch.tril(at_mask)
if is_causal:
at_mask = torch.tril(at_mask)
M = at_mask[None, None, :, :] + torch.zeros_like(rx)
rx[M == 0] = float("-inf")
# rx += kp_mask[:, None, None, :]

View File

@@ -1,14 +1,16 @@
import torch
import triton
import pytest
import torch
import triton
@pytest.mark.parametrize("M, N, dtype, mode",
[
(M, N, dtype, mode) for M in [1024, 821]
for N in [512, 857, 1871, 2089, 8573, 31000]
for dtype in ['float16', 'float32']\
for mode in ['forward', 'backward']
]
[
(M, N, dtype, mode) for M in [1024, 821]
for N in [512, 857, 1871, 2089, 8573, 31000]
for dtype in ['float16', 'float32']
for mode in ['forward', 'backward']
]
)
def test_op(M, N, dtype, mode):
dtype = {'float16': torch.float16, 'float32': torch.float32}[dtype]
@@ -30,4 +32,4 @@ def test_op(M, N, dtype, mode):
x.grad.zero_()
th_y.backward(dy)
th_dx = x.grad.clone()
triton.testing.assert_almost_equal(th_dx, tt_dx)
triton.testing.assert_almost_equal(th_dx, tt_dx)

View File

@@ -1,8 +1,10 @@
import pytest
import itertools
import triton
import pytest
import torch
import triton
@pytest.mark.parametrize(
"BLOCK_M, BLOCK_N, BLOCK_K, SPLIT_K, NWARP, NSTAGE, M, N, K, AT, BT, DTYPE",
@@ -80,11 +82,11 @@ def test_op(BLOCK_M, BLOCK_N, BLOCK_K, SPLIT_K, NWARP, NSTAGE, M, N, K, AT, BT,
K = BLOCK_K * SPLIT_K if K is None else K
# allocate/transpose inputs
DTYPE = {"float16": torch.float16, "float32": torch.float32}[DTYPE]
a = .1*torch.randn((K, M) if AT else (M, K), device="cuda", dtype=DTYPE)
b = .1*torch.randn((N, K) if BT else (K, N), device="cuda", dtype=DTYPE)
a = .1 * torch.randn((K, M) if AT else (M, K), device="cuda", dtype=DTYPE)
b = .1 * torch.randn((N, K) if BT else (K, N), device="cuda", dtype=DTYPE)
a = a.t() if AT else a
b = b.t() if BT else b
# run test
th_c = torch.matmul(a, b)
tt_c = triton.testing.catch_oor(lambda : triton.ops.matmul(a, b), pytest)
tt_c = triton.testing.catch_oor(lambda: triton.ops.matmul(a, b), pytest)
triton.testing.assert_almost_equal(th_c, tt_c)