1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
|
From 1ae3479e42a2f52ef07410b16d963ccdae8b62f7 Mon Sep 17 00:00:00 2001
From: drisspg <drisspguessous@gmail.com>
Date: Mon, 16 Dec 2024 09:15:07 -0800
Subject: [PATCH 1/5] Update
[ghstack-poisoned]
---
test/inductor/test_flex_attention.py | 5 +++++
torch/_inductor/kernel/flex_attention.py | 3 ---
2 files changed, 5 insertions(+), 3 deletions(-)
Index: pytorch/test/inductor/test_flex_attention.py
===================================================================
--- pytorch.orig/test/inductor/test_flex_attention.py
+++ pytorch/test/inductor/test_flex_attention.py
@@ -3231,6 +3231,11 @@ def forward(self, arg0_1, arg1_1, arg2_1
self.run_test_with_call(attention, Q_S=Q_S, KV_S=KV_S)
+ @supported_platform
+ def test_num_warps_8_error(self):
+ attention = functools.partial(flex_attention, score_mod=_identity)
+ self.run_test_with_call(attention, Q_S=128, KV_S=128, Q_D=128, V_D=128)
+
@unittest.skipIf(not TEST_MULTIGPU, "detected only one GPU")
def test_qkv_and_block_mask_on_the_same_device(self):
make_tensor = functools.partial(
Index: pytorch/torch/_inductor/kernel/flex_attention.py
===================================================================
--- pytorch.orig/torch/_inductor/kernel/flex_attention.py
+++ pytorch/torch/_inductor/kernel/flex_attention.py
@@ -746,11 +746,13 @@ def _get_nv_config(query, mode: Mode) ->
return (64, 128, 8, 3)
else:
return (64, 64, 4, 2)
- elif capability >= (8, 0): # A100
- if head_dim == 64:
+ elif capability >= (8, 0):
+ if head_dim >= 64:
return (32, 128, 4, 3)
elif head_dim == 128:
- return (64, 128, 8, 3)
+ # SM86/89 have smaller shared memory sizes
+ num_stages = 3 if capability[-1] == 0 else 2
+ return (64, 64, 4, num_stages)
else:
return (64, 64, 4, 2)
else: # modest hardware or extremely large head_dim
@@ -2273,9 +2275,6 @@ def flex_attention_backward(*args, **kwa
or SPARSE_Q_BLOCK_SIZE % BLOCK2 != 0
):
continue
- if num_warps == 8:
- # Working around https://github.com/pytorch/pytorch/issues/141603
- continue
# Performance tuning
cur_kernel_options = original_kernel_options.copy()
|
