commit: fad64b065493a4406304cf7cb4d4aa36a7876555 Author: Miezhiko <Miezhiko <AT> gmail <DOT> com> AuthorDate: Mon Dec 27 10:56:29 2021 +0000 Commit: Andrew Ammerlaan <andrewammerlaan <AT> gentoo <DOT> org> CommitDate: Mon Dec 27 11:01:04 2021 +0000 URL: https://gitweb.gentoo.org/proj/sci.git/commit/?id=fad64b06
sci-libs/pytorch: fixes for cuda USE flag, nccl update Signed-off-by: Miezhiko <Miezhiko <AT> gmail.com> Closes: https://github.com/gentoo/sci/pull/1132 Signed-off-by: Andrew Ammerlaan <andrewammerlaan <AT> gentoo.org> .../files/pytorch-1.10.1-nccl-nvccflags.patch | 37 ++++++++++++++++++++++ sci-libs/pytorch/pytorch-1.10.1.ebuild | 7 ++-- 2 files changed, 41 insertions(+), 3 deletions(-) diff --git a/sci-libs/pytorch/files/pytorch-1.10.1-nccl-nvccflags.patch b/sci-libs/pytorch/files/pytorch-1.10.1-nccl-nvccflags.patch new file mode 100644 index 000000000..0ac12a1e7 --- /dev/null +++ b/sci-libs/pytorch/files/pytorch-1.10.1-nccl-nvccflags.patch @@ -0,0 +1,37 @@ +From 7b04cd503fa1a6265893e0320ee246bcc0bd98cf Mon Sep 17 00:00:00 2001 +From: Miezhiko <miezh...@gmail.com> +Date: Mon, 27 Dec 2021 14:48:12 +0400 +Subject: [PATCH] neutral flags + +--- + makefiles/common.mk | 10 +--------- + 1 file changed, 1 insertion(+), 9 deletions(-) + +diff --git a/makefiles/common.mk b/makefiles/common.mk +index 64f8d2dc6..7a979c90f 100644 +--- a/makefiles/common.mk ++++ b/makefiles/common.mk +@@ -55,7 +55,7 @@ CXXFLAGS := -DCUDA_MAJOR=$(CUDA_MAJOR) -DCUDA_MINOR=$(CUDA_MINOR) -fPIC -fvisi + # Maxrregcount needs to be set accordingly to NCCL_MAX_NTHREADS (otherwise it will cause kernel launch errors) + # 512 : 120, 640 : 96, 768 : 80, 1024 : 60 + # We would not have to set this if we used __launch_bounds__, but this only works on kernels, not on functions. +-NVCUFLAGS := -ccbin $(CXX) $(NVCC_GENCODE) -std=c++11 --expt-extended-lambda -Xptxas -maxrregcount=96 -Xfatbin -compress-all ++NVCUFLAGS := $(NVCCFLAGS) $(NVCC_GENCODE) -std=c++11 --expt-extended-lambda -Xptxas -maxrregcount=96 -Xfatbin -compress-all + # Use addprefix so that we can specify more than one path + NVLDFLAGS := -L${CUDA_LIB} -lcudart -lrt + +@@ -69,14 +69,6 @@ NVLDFLAGS += ${GCOV_FLAGS:%=-Xcompiler %} + # $(warning GCOV_FLAGS=${GCOV_FLAGS}) + ########## GCOV ########## + +-ifeq ($(DEBUG), 0) +-NVCUFLAGS += -O3 +-CXXFLAGS += -O3 -g +-else +-NVCUFLAGS += -O0 -G -g +-CXXFLAGS += -O0 -g -ggdb3 +-endif +- + ifneq ($(VERBOSE), 0) + NVCUFLAGS += -Xptxas -v -Xcompiler -Wall,-Wextra,-Wno-unused-parameter + CXXFLAGS += -Wall -Wextra diff --git a/sci-libs/pytorch/pytorch-1.10.1.ebuild b/sci-libs/pytorch/pytorch-1.10.1.ebuild index a7bbb62ac..948d595a7 100644 --- a/sci-libs/pytorch/pytorch-1.10.1.ebuild +++ b/sci-libs/pytorch/pytorch-1.10.1.ebuild @@ -28,7 +28,7 @@ https://github.com/facebookincubator/gloo/archive/c22a5cfba94edf8ea4f53a174d38aa https://github.com/google/googletest/archive/e2239ee6043f73722e7aa812a459f54a28552929.tar.gz -> googletest-e2239ee6043f73722e7aa812a459f54a28552929.tar.gz https://github.com/intel/ideep/archive/9ca27bbfd88fa1469cbf0467bd6f14cd1738fa40.tar.gz -> ideep-9ca27bbfd88fa1469cbf0467bd6f14cd1738fa40.tar.gz https://github.com/intel/mkl-dnn/archive/5ef631a0.tar.gz -> mkl-dnn-5ef631a0.tar.gz -cuda? ( https://github.com/NVIDIA/nccl/archive/033d7995.tar.gz -> nccl-033d7995.tar.gz ) +cuda? ( https://github.com/NVIDIA/nccl/archive/c5790b36722d5b41ee2a9b2bad69e364180ffd22.tar.gz -> nccl-c5790b36722d5b41ee2a9b2bad69e364180ffd22.tar.gz ) https://github.com/Maratyszcza/NNPACK/archive/c07e3a0400713d546e0dea2d5466dd22ea389c73.tar.gz -> NNPACK-c07e3a0400713d546e0dea2d5466dd22ea389c73.tar.gz https://github.com/onnx/onnx/archive/a82c6a70.tar.gz -> onnx-a82c6a70.tar.gz https://github.com/onnx/onnx-tensorrt/archive/c153211418a7c57ce071d9ce2a41f8d1c85a878f.tar.gz -> onnx-tensorrt-c153211418a7c57ce071d9ce2a41f8d1c85a878f.tar.gz @@ -168,7 +168,7 @@ src_prepare() { rmdir third_party/ideep/mkl-dnn || die ln -sv "${WORKDIR}"/mkl-dnn-5ef631a030a6f73131c77892041042805a06064f third_party/ideep/mkl-dnn || die rmdir third_party/nccl/nccl || die - ln -sv "${WORKDIR}"/nccl-033d799524fb97629af5ac2f609de367472b2696 third_party/nccl/nccl || die + ln -sv "${WORKDIR}"/nccl-c5790b36722d5b41ee2a9b2bad69e364180ffd22 third_party/nccl/nccl || die rmdir third_party/NNPACK || die ln -sv "${WORKDIR}"/NNPACK-c07e3a0400713d546e0dea2d5466dd22ea389c73 third_party/NNPACK || die rmdir third_party/onnx || die @@ -208,9 +208,10 @@ src_prepare() { if use cuda; then cd third_party/nccl/nccl || die - eapply "${FILESDIR}"/${PN}-1.6.0-nccl-nvccflags.patch + eapply "${FILESDIR}"/${PN}-1.10.1-nccl-nvccflags.patch cuda_src_prepare export CUDAHOSTCXX=$(cuda_gccdir)/g++ + cd "${S}" fi if use rocm; then
