[gentoo-commits] repo/gentoo:master commit in: sci-ml/caffe2/

[Alfredo Tupone]

commit:     883310f11ac13b90635ce2c768437f09bc255c2f
Author:     Alfredo Tupone <tupone <AT> gentoo <DOT> org>
AuthorDate: Fri May 23 06:49:52 2025 +0000
Commit:     Alfredo Tupone <tupone <AT> gentoo <DOT> org>
CommitDate: Fri May 23 06:50:28 2025 +0000
URL:        https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=883310f1

sci-ml/caffe2: add usage of cusparselt

Signed-off-by: Alfredo Tupone <tupone <AT> gentoo.org>

 sci-ml/caffe2/{caffe2-2.7.0-r1.ebuild => caffe2-2.7.0-r2.ebuild} | 4 +++-
 sci-ml/caffe2/metadata.xml                                       | 1 +
 2 files changed, 4 insertions(+), 1 deletion(-)

diff --git a/sci-ml/caffe2/caffe2-2.7.0-r1.ebuild 
b/sci-ml/caffe2/caffe2-2.7.0-r2.ebuild
similarity index 98%
rename from sci-ml/caffe2/caffe2-2.7.0-r1.ebuild
rename to sci-ml/caffe2/caffe2-2.7.0-r2.ebuild
index e4485984f80d..f9d324654e49 100644
--- a/sci-ml/caffe2/caffe2-2.7.0-r1.ebuild
+++ b/sci-ml/caffe2/caffe2-2.7.0-r2.ebuild
@@ -27,7 +27,7 @@ S="${WORKDIR}"/${MYP}
 LICENSE="BSD"
 SLOT="0"
 KEYWORDS="~amd64 ~arm64"
-IUSE="cuda distributed fbgemm flash gloo memefficient mkl mpi nnpack +numpy
+IUSE="cuda cusparselt distributed fbgemm flash gloo memefficient mkl mpi 
nnpack +numpy
        onednn openblas opencl openmp qnnpack rocm xnnpack"
 RESTRICT="test"
 REQUIRED_USE="
@@ -61,6 +61,7 @@ RDEPEND="
                dev-libs/cudnn
                >=sci-ml/cudnn-frontend-1.0.3:0/8
                dev-util/nvidia-cuda-toolkit:=[profiler]
+               cusparselt? ( dev-libs/cusparselt )
        )
        fbgemm? ( sci-ml/FBGEMM )
        gloo? ( sci-ml/gloo[cuda?] )
@@ -284,6 +285,7 @@ src_configure() {
                        -DTORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST:-3.5 
7.0}"
                        -DUSE_NCCL=OFF # TODO: NVIDIA Collective Communication 
Library
                        -DCMAKE_CUDA_FLAGS="$(cuda_gccdir -f | tr -d \")"
+                       -DUSE_CUSPARSELT=$(usex cusparselt)
                )
        elif use rocm; then
                export PYTORCH_ROCM_ARCH="$(get_amdgpu_flags)"

diff --git a/sci-ml/caffe2/metadata.xml b/sci-ml/caffe2/metadata.xml
index 2483e58ed39e..961dd624f404 100644
--- a/sci-ml/caffe2/metadata.xml
+++ b/sci-ml/caffe2/metadata.xml
@@ -6,6 +6,7 @@
                <name>Tupone Alfredo</name>
        </maintainer>
        <use>
+               <flag name="cusparselt">Use the CUDA Sparse Matrix 
Multiplication</flag>
                <flag name="distributed">Support distributed applications</flag>
                <flag name="fbgemm">Use <pkg>sci-ml/FBGEMM</pkg></flag>
                <flag name="flash">Enable flash attention</flag>