Date: Friday, November 4, 2022 @ 02:38:15
Author: kgizdov
Revision: 1341767
upgpkg: python-pytorch 1.13.0-3: add 9.0 (Hopper) CUDA compute architecture as
well
Modified:
python-pytorch/trunk/PKGBUILD
python-pytorch/trunk/cuda_arch_update.patch
------------------------+
PKGBUILD | 4 ++--
cuda_arch_update.patch | 10 +++++++---
2 files changed, 9 insertions(+), 5 deletions(-)
Modified: PKGBUILD
===================================================================
--- PKGBUILD 2022-11-04 01:31:05 UTC (rev 1341766)
+++ PKGBUILD 2022-11-04 02:38:15 UTC (rev 1341767)
@@ -6,7 +6,7 @@
pkgname=("${pkgbase}" "${pkgbase}-opt" "${pkgbase}-cuda" "${pkgbase}-opt-cuda")
pkgver=1.13.0
_pkgver=1.13.0
-pkgrel=2
+pkgrel=3
_pkgdesc='Tensors and Dynamic neural networks in Python with strong GPU
acceleration'
pkgdesc="${_pkgdesc}"
arch=('x86_64')
@@ -112,7 +112,7 @@
'1f7ce593fa9fc62535ca1c3d85c996a73006cc614c7b7258160c3fc53cd52a1cfddcb18baf897f2e1223ecdfee52ca1471b91c9f845368ed6ac51b66f6e0e676'
'fdea0b815d7750a4233c1d4668593020da017aea43cf4cb63b4c00d0852c7d34f0333e618fcf98b8df2185313a2089b8c2e9fe8ec3cfb0bf693598f9c61461a8'
'0a8fc110a306e81beeb9ddfb3a1ddfd26aeda5e3f7adfb0f7c9bc3fd999c2dde62e0b407d3eca573097a53fd97329214e30e8767fb38d770197c7ec2b53daf18'
-
'47ace3b87c4d33c7a1aec13b5bc88f434a0c7d38de43aedbae15f635111633ae1119394e9ead5cf16b5230baa525c518c0b500a3c38cac253170921caf90d3f8'
+
'2a540c5beb978bcda1e3375d82526fb088409cd9ba0be3aa8f411477dd935b75bab2b4a4a79cecffeee91e8c6a3a716884508d17b9a558979dbb5059458bd0d3'
'6286b05d5b5143f117363e3ce3c7d693910f53845aeb6f501b3eea64aa71778cb2d7dcd4ac945d5321ef23b4da02446e86dedc6a9b6a998df4a7f3b1ce50550a')
options=('!lto')
Modified: cuda_arch_update.patch
===================================================================
--- cuda_arch_update.patch 2022-11-04 01:31:05 UTC (rev 1341766)
+++ cuda_arch_update.patch 2022-11-04 02:38:15 UTC (rev 1341767)
@@ -15,7 +15,7 @@
index 7f22d476d2fbe..822c041ee5268 100644
--- a/cmake/Modules_CUDA_fix/upstream/FindCUDA/select_compute_arch.cmake
+++ b/cmake/Modules_CUDA_fix/upstream/FindCUDA/select_compute_arch.cmake
-@@ -98,8 +98,19 @@ if(NOT CUDA_VERSION VERSION_LESS "11.1")
+@@ -98,8 +98,22 @@ if(NOT CUDA_VERSION VERSION_LESS "11.1")
list(APPEND CUDA_ALL_GPU_ARCHITECTURES "8.6")
set(CUDA_LIMIT_GPU_ARCHITECUTRE "8.6")
@@ -28,6 +28,9 @@
+ list(APPEND CUDA_KNOWN_GPU_ARCHITECTURES "Ada")
+ list(APPEND CUDA_COMMON_GPU_ARCHITECTURES "8.9")
+ list(APPEND CUDA_ALL_GPU_ARCHITECTURES "8.9")
++ list(APPEND CUDA_KNOWN_GPU_ARCHITECTURES "Hopper")
++ list(APPEND CUDA_COMMON_GPU_ARCHITECTURES "9.0")
++ list(APPEND CUDA_ALL_GPU_ARCHITECTURES "9.0")
+
if(CUDA_VERSION VERSION_LESS "12.0")
set(CUDA_LIMIT_GPU_ARCHITECTURE "9.0")
@@ -39,16 +42,17 @@
index 36811bf22dedc..612ae9fdf0785 100644
--- a/torch/utils/cpp_extension.py
+++ b/torch/utils/cpp_extension.py
-@@ -1729,10 +1729,11 @@ def _get_cuda_arch_flags(cflags: Optional[List[str]] =
None) -> List[str]:
+@@ -1729,10 +1729,12 @@ def _get_cuda_arch_flags(cflags: Optional[List[str]] =
None) -> List[str]:
('Volta', '7.0+PTX'),
('Turing', '7.5+PTX'),
('Ampere', '8.0;8.6+PTX'),
+ ('Ada', '8.9+PTX'),
++ ('Hopper', '9.0+PTX'),
])
supported_arches = ['3.5', '3.7', '5.0', '5.2', '5.3', '6.0', '6.1',
'6.2',
- '7.0', '7.2', '7.5', '8.0', '8.6']
-+ '7.0', '7.2', '7.5', '8.0', '8.6', '8.9']
++ '7.0', '7.2', '7.5', '8.0', '8.6', '8.9', '9.0']
valid_arch_strings = supported_arches + [s + "+PTX" for s in
supported_arches]
# The default is sm_30 for CUDA 9.x and 10.x
