diff --git a/cuda-flags.file b/cuda-flags.file index 87cbcf721bc..ebe995cd488 100644 --- a/cuda-flags.file +++ b/cuda-flags.file @@ -2,8 +2,8 @@ ## INCLUDE cpp-standard # define the CUDA compilation flags in a way that can be shared by SCRAM-based and regular tools -# build support for Volta (7.0), Turing (7.5), Ampere (8.x), Lovelace (8.9), Hopper (9.0) and Blackwell (10.0, 12.0) -%define cuda_arch 70 75 80 89 90 100 120 +# build support for Turing (7.5), Ampere (8.x), Lovelace (8.9), Hopper (9.0) and Blackwell (10.0, 12.0) +%define cuda_arch 75 80 89 90 100 120 # LIBOMPTARGET_NVPTX_COMPUTE_CAPABILITIES style for listing the supported CUDA compute architectures %define omptarget_cuda_archs %(echo $(for ARCH in %cuda_arch; do echo "$ARCH"; done) | sed -e"s/ /,/g") diff --git a/cuda.spec b/cuda.spec index 5a55f848f1f..7e42b1185c1 100644 --- a/cuda.spec +++ b/cuda.spec @@ -1,8 +1,8 @@ -### RPM external cuda 12.9.1 +### RPM external cuda 13.3.0 ## INITENV +PATH LD_LIBRARY_PATH %i/lib64 %define runpath_opts -m compute-sanitizer -m drivers -m nvvm -%define driversversion 575.57.08 +%define driversversion 610.43.02 %ifarch x86_64 Source0: https://developer.download.nvidia.com/compute/cuda/%{realversion}/local_installers/%{n}_%{realversion}_%{driversversion}_linux.run @@ -53,13 +53,11 @@ chmod a+x %_builddir/build/extras/CUPTI/lib64/*.so* mv %_builddir/build/extras/CUPTI/lib64/*.so* %{i}/lib64/ mv %_builddir/build/extras/CUPTI/include/*.h %{i}/include/ -# leave out the Nsight and NVVP graphical tools, and package the other binaries -rm -f %_builddir/build/bin/computeprof +# leave out the graphical tools, and package the other binaries rm -f %_builddir/build/bin/cuda-uninstaller rm -f %_builddir/build/bin/ncu* rm -f %_builddir/build/bin/nsight* rm -f %_builddir/build/bin/nsys* -rm -f %_builddir/build/bin/nvvp mv %_builddir/build/bin %{i}/ # package the cuda-gdb support files, and rename the binary to use it via a wrapper diff --git a/cudnn.spec b/cudnn.spec index b7cc5e08e71..a39bb084655 100644 --- a/cudnn.spec +++ b/cudnn.spec @@ -1,7 +1,7 @@ -### RPM external cudnn 9.9.0.52 +### RPM external cudnn 9.23.0.39 ## INITENV +PATH LD_LIBRARY_PATH %i/lib64 -%define cudaver 12 +%define cudaver 13 # NVIDIA uses sbsa for aarch64, and the standard architecture name for ppc64le and x86_64 %ifarch aarch64 diff --git a/onnxruntime/cms-changes.patch b/onnxruntime/cms-changes.patch index 1f1009a6b46..b9896af39f3 100644 --- a/onnxruntime/cms-changes.patch +++ b/onnxruntime/cms-changes.patch @@ -1,16 +1,16 @@ diff --git a/onnxruntime/core/mlas/lib/platform.cpp b/onnxruntime/core/mlas/lib/platform.cpp -index eccde79848e61..4a107f173da07 100644 +index 191ee1ab2f..c958e7148a 100644 --- a/onnxruntime/core/mlas/lib/platform.cpp +++ b/onnxruntime/core/mlas/lib/platform.cpp -@@ -23,6 +23,7 @@ Module Name: - #include "kleidiai/mlasi_kleidiai.h" - #endif - +@@ -30,6 +30,7 @@ Abstract: + #include + #include + #include +#include #include - #include -@@ -355,8 +356,11 @@ Return Value: + #if defined(MLAS_TARGET_POWER) +@@ -436,8 +437,11 @@ Return Value: // uint64_t xcr0 = MlasReadExtendedControlRegister(_XCR_XFEATURE_ENABLED_MASK); @@ -23,7 +23,7 @@ index eccde79848e61..4a107f173da07 100644 this->GemmFloatKernel = MlasGemmFloatKernelAvx; -@@ -390,7 +394,7 @@ Return Value: +@@ -471,7 +475,7 @@ Return Value: __cpuid_count(7, 0, Cpuid7[0], Cpuid7[1], Cpuid7[2], Cpuid7[3]); #endif @@ -33,7 +33,7 @@ index eccde79848e61..4a107f173da07 100644 this->Avx2Supported_ = true; diff --git a/onnxruntime/core/providers/cpu/tensor/scatter.cc b/onnxruntime/core/providers/cpu/tensor/scatter.cc -index c7a2005924836..c938ca9bf073d 100644 +index c7a2005924..c938ca9bf0 100644 --- a/onnxruntime/core/providers/cpu/tensor/scatter.cc +++ b/onnxruntime/core/providers/cpu/tensor/scatter.cc @@ -326,7 +326,7 @@ Status ScatterData( diff --git a/onnxruntime/spec b/onnxruntime/spec index 80ba97847b8..3e91e031f3d 100644 --- a/onnxruntime/spec +++ b/onnxruntime/spec @@ -1,4 +1,4 @@ -### RPM external onnxruntime 1.25.1 +### RPM external onnxruntime 1.26.0 ## INITENV +PATH PYTHON3PATH %{i}/${PYTHON3_LIB_SITE_PACKAGES} ## INCLUDE cuda-flags %define github_user microsoft