Skip to content

Commit c7ded05

Browse files
q10facebook-github-bot
q10
authored and
facebook-github-bot
committed
Remove sm_100 and sm_120 (#4024)
Summary: X-link: facebookresearch/FBGEMM#1133 - Remove sm_100 and sm_120 from architectures list and keep just sm_100a and sm_120a instead, to enable compilation for FP4 CUTLASS quantization kernels (#4004), since we are running into the following error: ``` Instruction 'cvt with .e2m1x2' not supported on .target 'sm_100' ``` Pull Request resolved: #4024 Reviewed By: spcyppt Differential Revision: D73901832 Pulled By: q10 fbshipit-source-id: 690c58b214aee80374e43a93bf39fe70e430da9a
1 parent 7d7af72 commit c7ded05

File tree

2 files changed

+28
-29
lines changed

2 files changed

+28
-29
lines changed

‎.github/scripts/nova_dir.bash

+1-1
Original file line numberDiff line numberDiff line change
@@ -23,7 +23,7 @@ fi
2323
if [[ "$CONDA_ENV" != "" ]]; then export CONDA_RUN="conda run --no-capture-output -p ${CONDA_ENV}" && echo "$CONDA_RUN"; fi
2424

2525
if [[ "$CU_VERSION" == "cu128" ]]; then
26-
export TORCH_CUDA_ARCH_LIST="7.0;8.0;9.0;9.0a;10.0;10.0a;12.0;12.0a"
26+
export TORCH_CUDA_ARCH_LIST="7.0;8.0;9.0;9.0a;10.0a;12.0a"
2727
echo "Set TORCH_CUDA_ARCH_LIST to: ${TORCH_CUDA_ARCH_LIST}"
2828

2929
elif [[ "$CU_VERSION" == "cu126" ]] ||

‎fbgemm_gpu/docs/src/general/Releases.rst

+27-28
Original file line numberDiff line numberDiff line change
@@ -10,34 +10,33 @@ FBGEMM is released in accordance to the PyTorch release schedule, and is each
1010
release has no guarantee to work in conjunction with PyTorch releases that are
1111
older than the one that the FBGEMM release corresponds to.
1212

13-
+-----------------+------------------+------------------+----------------+----------------------+---------------------------+---------------------------+
14-
| FBGEMM Release | Corresponding | Supported | Supported | Supported CUDA | (Experimental) Supported | (Experimental) Supported |
15-
| | PyTorch Release | Python Versions | CUDA Versions | Architectures | ROCm Versions | ROCm Architectures |
16-
+=================+==================+==================+================+======================+===========================+===========================+
17-
| 1.2.0 | 2.7.x | 3.9, 3.10, 3.11, | 11.8, 12.6, | 7.0, 8.0, 9.0, 9.0a, | 6.1, 6.2.4, 6.3 | gfx908, gfx90a, gfx942 |
18-
| | | 3.12, 3.13 | 12.8 | 10.0, 10.0a, 12.0, | | |
19-
| | | | | 12.0a | | |
20-
+-----------------+------------------+------------------+----------------+----------------------+---------------------------+---------------------------+
21-
| 1.1.0 | 2.6.x | 3.9, 3.10, 3.11, | 11.8, 12.4, | 7.0, 8.0, 9.0, | 6.1, 6.2.4, 6.3 | gfx908, gfx90a, gfx942 |
22-
| | | 3.12, 3.13 | 12.6 | 9.0a | | |
23-
+-----------------+------------------+------------------+----------------+----------------------+---------------------------+---------------------------+
24-
| 1.0.0 | 2.5.x | 3.9, 3.10, 3.11, | 11.8, 12.1, | 7.0, 8.0, 9.0, | 6.0, 6.1 | gfx908, gfx90a |
25-
| | | 3.12 | 12.4 | 9.0a | | |
26-
+-----------------+------------------+------------------+----------------+----------------------+---------------------------+---------------------------+
27-
| 0.8.0 | 2.4.x | 3.8, 3.9, 3.10, | 11.8, 12.1, | 7.0, 8.0, 9.0, | 6.0, 6.1 | gfx908, gfx90a |
28-
| | | 3.11, 3.12 | 12.4 | 9.0a | | |
29-
+-----------------+------------------+------------------+----------------+----------------------+---------------------------+---------------------------+
30-
| 0.7.0 | 2.3.x | 3.8, 3.9, 3.10, | 11.8, 12.1 | 7.0, 8.0, 9.0 | 6.0 | gfx908, gfx90a |
31-
| | | 3.11, 3.12 | | | | |
32-
+-----------------+------------------+------------------+----------------+----------------------+---------------------------+---------------------------+
33-
| 0.6.0 | 2.2.x | 3.8, 3.9, 3.10, | 11.8, 12.1 | 7.0, 8.0, 9.0 | 5.7 | gfx90a |
34-
| | | 3.11, 3.12 | | | | |
35-
+-----------------+------------------+------------------+----------------+----------------------+---------------------------+---------------------------+
36-
| 0.5.0 | 2.1.x | 3.8, 3.9, 3.10, | 11.8, 12.1 | 7.0, 8.0, 9.0 | 5.5, 5.6 | gfx90a |
37-
| | | 3.11 | | | | |
38-
+-----------------+------------------+------------------+----------------+----------------------+---------------------------+---------------------------+
39-
| 0.4.0 | 2.0.x | 3.8, 3.9, 3.10 | 11.7, 11.8 | 7.0, 8.0 | 5.3, 5.4 | gfx90a |
40-
+-----------------+------------------+------------------+----------------+----------------------+---------------------------+---------------------------+
13+
+-----------------+------------------+------------------+----------------+--------------------+---------------------------+---------------------------+
14+
| FBGEMM Release | Corresponding | Supported | Supported | Supported CUDA | (Experimental) Supported | (Experimental) Supported |
15+
| | PyTorch Release | Python Versions | CUDA Versions | Architectures | ROCm Versions | ROCm Architectures |
16+
+=================+==================+==================+================+====================+===========================+===========================+
17+
| 1.2.0 | 2.7.x | 3.9, 3.10, 3.11, | 11.8, 12.6, | 7.0, 8.0, 9.0, | 6.1, 6.2.4, 6.3 | gfx908, gfx90a, gfx942 |
18+
| | | 3.12, 3.13 | 12.8 | 9.0a, 10.0a, 12.0a | | |
19+
+-----------------+------------------+------------------+----------------+----------------------+---------------------------+---------------------------+
20+
| 1.1.0 | 2.6.x | 3.9, 3.10, 3.11, | 11.8, 12.4, | 7.0, 8.0, 9.0, | 6.1, 6.2.4, 6.3 | gfx908, gfx90a, gfx942 |
21+
| | | 3.12, 3.13 | 12.6 | 9.0a | | |
22+
+-----------------+------------------+------------------+----------------+--------------------+---------------------------+---------------------------+
23+
| 1.0.0 | 2.5.x | 3.9, 3.10, 3.11, | 11.8, 12.1, | 7.0, 8.0, 9.0, | 6.0, 6.1 | gfx908, gfx90a |
24+
| | | 3.12 | 12.4 | 9.0a | | |
25+
+-----------------+------------------+------------------+----------------+--------------------+---------------------------+---------------------------+
26+
| 0.8.0 | 2.4.x | 3.8, 3.9, 3.10, | 11.8, 12.1, | 7.0, 8.0, 9.0, | 6.0, 6.1 | gfx908, gfx90a |
27+
| | | 3.11, 3.12 | 12.4 | 9.0a | | |
28+
+-----------------+------------------+------------------+----------------+--------------------+---------------------------+---------------------------+
29+
| 0.7.0 | 2.3.x | 3.8, 3.9, 3.10, | 11.8, 12.1 | 7.0, 8.0, 9.0 | 6.0 | gfx908, gfx90a |
30+
| | | 3.11, 3.12 | | | | |
31+
+-----------------+------------------+------------------+----------------+--------------------+---------------------------+---------------------------+
32+
| 0.6.0 | 2.2.x | 3.8, 3.9, 3.10, | 11.8, 12.1 | 7.0, 8.0, 9.0 | 5.7 | gfx90a |
33+
| | | 3.11, 3.12 | | | | |
34+
+-----------------+------------------+------------------+----------------+--------------------+---------------------------+---------------------------+
35+
| 0.5.0 | 2.1.x | 3.8, 3.9, 3.10, | 11.8, 12.1 | 7.0, 8.0, 9.0 | 5.5, 5.6 | gfx90a |
36+
| | | 3.11 | | | | |
37+
+-----------------+------------------+------------------+----------------+--------------------+---------------------------+---------------------------+
38+
| 0.4.0 | 2.0.x | 3.8, 3.9, 3.10 | 11.7, 11.8 | 7.0, 8.0 | 5.3, 5.4 | gfx90a |
39+
+-----------------+------------------+------------------+----------------+--------------------+---------------------------+---------------------------+
4140

4241
Note that the list of supported CUDA and ROCm architectures refer to the targets
4342
support available in the default installation packages, and that building for

0 commit comments

Comments
 (0)