CUDA: Make CCCL configurable + pin it to 3.4.3 for CI jobs (llama/29792)

Pinning to >= 3.4.3 is required to enable DeviceTopK, which was affected by
a race condition https://github.com/NVIDIA/cccl/pull/10627.

We will relax this for future CTK versions which will bundle CCCL >
3.4.X (CTK 13.5 will bundle CCCL 3.5.0 for example)

ggml/CMakeLists.txt

@@ -197,6 +197,7 @@ set(GGML_BLAS_VENDOR ${GGML_BLAS_VENDOR_DEFAULT} CACHE STRING
 option(GGML_LLAMAFILE                       "ggml: use LLAMAFILE"                             ${GGML_LLAMAFILE_DEFAULT})
 
 option(GGML_CUDA                            "ggml: use CUDA"                                  OFF)
+set   (GGML_CUDA_CCCL_VERSION "" CACHE STRING "ggml: CCCL git tag to fetch, empty to use the version bundled with the installed CUDA Toolkit")
 option(GGML_MUSA                            "ggml: use MUSA"                                  OFF)
 option(GGML_CUDA_FORCE_MMQ                  "ggml: use mmq kernels instead of cuBLAS"         OFF)
 option(GGML_CUDA_FORCE_CUBLAS               "ggml: always use cuBLAS instead of mmq kernels"  OFF)

ggml/src/ggml-cuda/CMakeLists.txt

@@ -71,14 +71,13 @@ if (CUDAToolkit_FOUND)
 
     enable_language(CUDA)
 
-    # TODO: Remove once CCCL 3.2 has been released and bundled with CUDA Toolkit
-    if (GGML_CUDA_CUB_3DOT2)
+    if (GGML_CUDA_CCCL_VERSION)
         include(FetchContent)
 
         FetchContent_Declare(
             CCCL
             GIT_REPOSITORY https://github.com/nvidia/cccl.git
-            GIT_TAG        v3.2.0
+            GIT_TAG        "${GGML_CUDA_CCCL_VERSION}"
             GIT_SHALLOW    TRUE
         )
 
@@ -157,14 +156,15 @@ if (CUDAToolkit_FOUND)
         add_compile_definitions(GGML_CUDA_NO_PEER_COPY)
     endif()
 
+    if (GGML_CUDA_CCCL_VERSION)
+        target_link_libraries(ggml-cuda PRIVATE CCCL::CCCL)
+    endif()
+
     if (GGML_STATIC)
         if (WIN32)
             # As of 12.3.1 CUDA Toolkit for Windows does not offer a static cublas library
             target_link_libraries(ggml-cuda PRIVATE CUDA::cudart_static CUDA::cublas)
         else ()
-            if (GGML_CUDA_CUB_3DOT2)
-                target_link_libraries(ggml-cuda PRIVATE  CCCL::CCCL)
-            endif()
             if (CUDAToolkit_VERSION VERSION_GREATER_EQUAL "10.1")
                 target_link_libraries(ggml-cuda PRIVATE  CUDA::cudart_static CUDA::cublas_static CUDA::cublasLt_static)
             else()
@@ -172,9 +172,6 @@ if (CUDAToolkit_FOUND)
             endif()
         endif()
     else()
-        if (GGML_CUDA_CUB_3DOT2)
-            target_link_libraries(ggml-cuda PRIVATE  CCCL::CCCL)
-        endif()
         target_link_libraries(ggml-cuda PRIVATE CUDA::cudart CUDA::cublas)
     endif()
 

ggml/src/ggml-cuda/top-k.cu

@@ -3,11 +3,15 @@
 
 #ifdef GGML_CUDA_USE_CUB
 #    include <cub/cub.cuh>
-#    if (CCCL_MAJOR_VERSION >= 3 && CCCL_MINOR_VERSION >= 2)
+// DeviceTopK has a race condition before CCCL 3.4.3.
+// https://github.com/NVIDIA/cccl/pull/10627
+#    if (CCCL_MAJOR_VERSION > 3 || \
+         (CCCL_MAJOR_VERSION == 3 && CCCL_MINOR_VERSION > 4) || \
+         (CCCL_MAJOR_VERSION == 3 && CCCL_MINOR_VERSION == 4 && CCCL_PATCH_VERSION >= 3))
 #        define CUB_TOP_K_AVAILABLE
 #        include <cuda/iterator>
 using namespace cub;
-#    endif  // CCCL_MAJOR_VERSION >= 3 && CCCL_MINOR_VERSION >= 2
+#    endif  // CCCL >= 3.4.3
 #endif      // GGML_CUDA_USE_CUB
 
 #ifdef CUB_TOP_K_AVAILABLE