nongnu: Add cuda-python and nvidia-cutlass.

* nongnu/packages/nvidia.scm (cuda-python, nvidia-cutlass): Add variables.
* nongnu/packages/patches: Add necessary patches.

nongnu/packages/nvidia.scm

@@ -1893,6 +1893,52 @@ libraries for NVIDIA GPUs, all of which are proprietary.")
     (home-page "https://developer.nvidia.com/cuda-toolkit")
     (license (package-license cuda-cudart))))
 
+(define-public cuda-python
+  (package
+    (name "cuda-python")
+    (version "12.1.0")
+    (source
+     (origin
+       (method git-fetch)
+       (uri (git-reference
+             (url "https://github.com/NVIDIA/cuda-python")
+             (commit (string-append "v" version))))
+       (file-name (git-file-name name version))
+       (sha256
+        (base32 "0i0wvx5kxckphsf1n02rr86hrnc2r6p8wlrvq1n1w9c3l6m24d13"))))
+    (build-system pyproject-build-system)
+    (arguments
+     (list
+      #:tests? #f  ; FIXME: most tests fail.
+      #:phases
+      #~(modify-phases %standard-phases
+          (add-after 'unpack 'fix-setup.py
+            (lambda _
+              (substitute* "setup.py"
+                (("import versioneer" all)
+                 (format #f "~a~%import pyparsing" all)))))
+          (add-before 'build 'set_cuda_paths
+            (lambda _
+              (setenv "CUDA_HOME"
+                      #$(this-package-input "cuda-dev"))
+              (setenv "PARALLEL_LEVEL"
+                      (number->string (parallel-job-count))))))))
+    (native-inputs (list python-cython
+                         python-numpy
+                         python-pytest
+                         python-pytest-benchmark
+                         python-setuptools
+                         python-wheel))
+    (inputs (list cuda-dev))
+    (propagated-inputs (list python-pyclibrary))
+    (home-page "https://github.com/NVIDIA/cuda-python")
+    (synopsis "CUDA Python low-level bindings")
+    (description "This package provides Python low-level bindings for NVIDIA
+CUDA toolkit.")
+    (license
+     (license:nonfree
+      "https://github.com/NVIDIA/cuda-python/blob/main/LICENSE"))))
+
 (define (nvidia-cudnn-samples system version)
   (origin
     (method url-fetch)
@@ -2051,6 +2097,83 @@ functionality on top of the backend API, such as errata filters and
 autotuning.")
     (license license-gnu:expat)))
 
+(define-public nvidia-cutlass
+  (package
+    (name "nvidia-cutlass")
+    (version "3.2.2")
+    (source
+     (origin
+       (method git-fetch)
+       (uri (git-reference
+             (url "https://github.com/NVIDIA/cutlass")
+             (commit (string-append "v" version))))
+       (file-name (git-file-name name version))
+       (sha256
+        (base32 "0qyxkp3pmndlzm3aw9xwrx57znj9p4xlvqahavgzq8c1nd7bj3wp"))
+       (patches
+        (parameterize
+            ((%patch-path
+              (map
+               (lambda (directory)
+                 (string-append directory "/nongnu/packages/patches"))
+               %load-path)))
+          (search-patches "nvidia-cutlass-3.2.2_disable_static_lib.patch")))))
+    (build-system cmake-build-system)
+    (arguments
+     (list
+      ;; XXX: Cutlass is incredibly heavy to build when not specifying target
+      ;; GPU architecture (4G), avoid tests, examples and static library.
+      ;; Setting the contrary often runs out of RAM even on beefy laptops.
+      #:configure-flags ''("-DCUTLASS_ENABLE_TESTS=OFF"
+                           "-DCUTLASS_INSTALL_TESTS=OFF"
+                           "-DCUTLASS_BUILD_STATIC_LIBRARY=OFF"
+                           "-DCUTLASS_ENABLE_EXAMPLES=OFF"
+                           "-DCUTLASS_UNITY_BUILD_ENABLED=ON")
+      #:phases
+      #~(modify-phases %standard-phases
+          ;; XXX: This phase is not necessary on earlier versions.
+          ;; Remove it when updating.
+          (add-after 'unpack 'fix-cuda-build
+            (lambda _
+              (substitute* "CMakeLists.txt"
+                (("--user")
+                 (string-append "--prefix=" #$output)))
+              (setenv "PYTHONPATH"
+                      (string-append (getcwd) "/python"))))
+          (add-before 'build 'set_cuda_paths
+            (lambda _
+              (setenv "CUDACXX"
+                      #$(file-append (this-package-input "cuda-toolkit")
+                                     "/bin/nvcc"))))
+          (add-after 'install 'cleanup
+            (lambda _
+              (delete-file-recursively
+               (string-append #$output "/test")))))))
+    (native-inputs (list python python-setuptools))
+    (inputs (list cuda-toolkit))
+    (propagated-inputs (list cuda-python
+                             python-networkx
+                             python-numpy
+                             python-pydot
+                             python-scipy
+                             python-treelib))
+    (home-page "https://developer.nvidia.com/blog/cutlass-linear-algebra-cuda")
+    (synopsis "CUDA Templates for Linear Algebra Subroutines")
+    (description
+     "This package provides a collection of CUDA C++ template abstractions for
+implementing high-performance matrix-matrix multiplication (GEMM) and related
+computations at all levels and scales within CUDA.  It incorporates strategies
+for hierarchical decomposition and data movement similar to those used to
+implement cuBLAS and cuDNN.  CUTLASS decomposes these moving parts into
+reusable, modular software components abstracted by C++ template
+classes.  Primitives for different levels of a conceptual parallelization
+hierarchy can be specialized and tuned via custom tiling sizes, data types,
+and other algorithmic policy.  The resulting flexibility simplifies their use
+as building blocks within custom kernels and applications.")
+    (license
+     (license:nonfree
+      "https://github.com/NVIDIA/cutlass/blob/main/LICENSE.txt"))))
+
 
 
 ;;;

nongnu/packages/patches/nvidia-cutlass-3.2.2_disable_static_lib.patch

@@ -0,0 +1,73 @@
+From 7ee9ec4c2636cca833761d3466df27edc4e3f952 Mon Sep 17 00:00:00 2001
+From: Nicolas Graves <ngraves@ngraves.fr>
+Date: Tue, 30 Jul 2024 14:13:09 +0200
+Subject: [PATCH] Add CUTLASS_BUILD_STATIC_LIBRARY option
+
+---
+ tools/library/CMakeLists.txt | 22 ++++++++++++++++++----
+ 1 file changed, 18 insertions(+), 4 deletions(-)
+
+diff --git a/tools/library/CMakeLists.txt b/tools/library/CMakeLists.txt
+index a11ebcf6..79f7ccd1 100644
+--- a/tools/library/CMakeLists.txt
++++ b/tools/library/CMakeLists.txt
+@@ -34,6 +34,7 @@ include(GNUInstallDirs)
+ 
+ set(CUTLASS_BUILD_MONO_LIBRARY OFF CACHE BOOL 
+   "Determines whether the cutlass library is generated as a single file or multiple files.")
++option(CUTLASS_BUILD_STATIC_LIBRARY "Build static libary for CUTLASS" ON)
+ 
+ ################################################################################
+ 
+@@ -126,7 +127,9 @@ function(cutlass_add_cutlass_library)
+     # simply link the generated object files to the default library. 
+ 
+     target_link_libraries(${DEFAULT_NAME} PRIVATE $<BUILD_INTERFACE:${__NAME}_objs>)
+-    target_link_libraries(${DEFAULT_NAME}_static PRIVATE $<BUILD_INTERFACE:${__NAME}_objs>)
++    if (CUTLASS_BUILD_STATIC_LIBRARY)
++        target_link_libraries(${DEFAULT_NAME}_static PRIVATE $<BUILD_INTERFACE:${__NAME}_objs>)
++    endif()
+ 
+   else()
+ 
+@@ -152,7 +155,7 @@ function(cutlass_add_cutlass_library)
+       )
+     
+     set_target_properties(${__NAME} PROPERTIES DEBUG_POSTFIX "${CUTLASS_LIBRARY_DEBUG_POSTFIX}")
+-    
++    if (CUTLASS_BUILD_STATIC_LIBRARY)
+     cutlass_add_library(
+       ${__NAME}_static
+       STATIC
+@@ -189,6 +192,15 @@ function(cutlass_add_cutlass_library)
+       LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR}
+       ARCHIVE DESTINATION ${CMAKE_INSTALL_LIBDIR}
+       )
++    else()
++    install(
++      TARGETS ${__NAME}
++      EXPORT NvidiaCutlass
++      RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR}
++      LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR}
++      ARCHIVE DESTINATION ${CMAKE_INSTALL_LIBDIR}
++      )
++    endif()
+     
+     if (__SUFFIX)
+     
+@@ -197,8 +209,10 @@ function(cutlass_add_cutlass_library)
+       # commands to pull in all kernels by default.
+     
+       target_link_libraries(${DEFAULT_NAME} INTERFACE ${__NAME})
+-      target_link_libraries(${DEFAULT_NAME}_static INTERFACE ${__NAME}_static)
+-    
++      if (CUTLASS_BUILD_STATIC_LIBRARY)
++         target_link_libraries(${DEFAULT_NAME}_static INTERFACE ${__NAME}_static)
++      endif()
++
+     endif()
+ 
+   endif()
+-- 
+2.45.2
+

nongnu/packages/patches/nvidia-cutlass-3.4.0_disable_static_lib.patch

@@ -0,0 +1,82 @@
+From ce4a14ae4041d6cfb69987fef5a65c50754c89b6 Mon Sep 17 00:00:00 2001
+From: Nicolas Graves <ngraves@ngraves.fr>
+Date: Sun, 28 Jul 2024 16:57:16 +0200
+Subject: [PATCH] Add option CUTLASS_BUILD_STATIC_LIBRARY
+
+---
+ tools/library/CMakeLists.txt | 26 +++++++++++++++++++++-----
+ 1 file changed, 20 insertions(+), 4 deletions(-)
+
+diff --git a/tools/library/CMakeLists.txt b/tools/library/CMakeLists.txt
+index 60a6cca5..f096c84d 100644
+--- a/tools/library/CMakeLists.txt
++++ b/tools/library/CMakeLists.txt
+@@ -34,6 +34,7 @@ include(GNUInstallDirs)
+ 
+ set(CUTLASS_BUILD_MONO_LIBRARY OFF CACHE BOOL 
+   "Determines whether the cutlass library is generated as a single file or multiple files.")
++option(CUTLASS_BUILD_STATIC_LIBRARY "Build static libary for CUTLASS" ON)
+ 
+ ################################################################################
+ 
+@@ -126,7 +127,9 @@ function(cutlass_add_cutlass_library)
+     # simply link the generated object files to the default library. 
+ 
+     target_link_libraries(${DEFAULT_NAME} PRIVATE $<BUILD_INTERFACE:${__NAME}_objs>)
+-    target_link_libraries(${DEFAULT_NAME}_static PRIVATE $<BUILD_INTERFACE:${__NAME}_objs>)
++    if (CUTLASS_BUILD_STATIC_LIBRARY)
++        target_link_libraries(${DEFAULT_NAME}_static PRIVATE $<BUILD_INTERFACE:${__NAME}_objs>)
++    endif()
+ 
+   else()
+ 
+@@ -154,7 +157,7 @@ function(cutlass_add_cutlass_library)
+       )
+     
+     set_target_properties(${__NAME} PROPERTIES DEBUG_POSTFIX "${CUTLASS_LIBRARY_DEBUG_POSTFIX}")
+-    
++    if (CUTLASS_BUILD_STATIC_LIBRARY)
+     cutlass_add_library(
+       ${__NAME}_static
+       STATIC
+@@ -193,6 +196,15 @@ function(cutlass_add_cutlass_library)
+       LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR}
+       ARCHIVE DESTINATION ${CMAKE_INSTALL_LIBDIR}
+       )
++    else()
++    install(
++      TARGETS ${__NAME}
++      EXPORT NvidiaCutlass
++      RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR}
++      LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR}
++      ARCHIVE DESTINATION ${CMAKE_INSTALL_LIBDIR}
++      )
++    endif()
+     
+     if (__SUFFIX)
+     
+@@ -201,7 +213,9 @@ function(cutlass_add_cutlass_library)
+       # commands to pull in all kernels by default.
+     
+       target_link_libraries(${DEFAULT_NAME} PUBLIC ${__NAME})
+-      target_link_libraries(${DEFAULT_NAME}_static PUBLIC ${__NAME}_static)
++      if (CUTLASS_BUILD_STATIC_LIBRARY)
++         target_link_libraries(${DEFAULT_NAME}_static PUBLIC ${__NAME}_static)
++      endif()
+     
+     endif()
+ 
+@@ -250,7 +264,9 @@ cutlass_add_cutlass_library(
+ 
+ # For backward compatibility with the old name
+ add_library(cutlass_lib ALIAS cutlass_library)
+-add_library(cutlass_lib_static ALIAS cutlass_library_static)
++if (CUTLASS_BUILD_STATIC_LIBRARY)
++   add_library(cutlass_lib_static ALIAS cutlass_library_static)
++endif()
+ 
+ ################################################################################
+ 
+-- 
+2.45.2
+