Disable bfloat16 operations on HIP

Author: stijnh

.github/workflows/cmake-run-cuda.yml

@@ -6,6 +6,10 @@ on:
       cuda-version:
         required: true
         type: string
+      cuda-architectures:
+        required: false
+        type: string
+        default: "80-real"
 
 env:
   # Customize the CMake build type here (Release, Debug, RelWithDebInfo, etc.)
@@ -33,7 +37,7 @@ jobs:
     - name: Configure CMake
       # Configure CMake in a 'build' subdirectory. `CMAKE_BUILD_TYPE` is only required if you are using a single-configuration generator such as make.
       # See https://cmake.org/cmake/help/latest/variable/CMAKE_BUILD_TYPE.html?highlight=cmake_build_type
-      run: CUDAARCHS=all cmake -B ${{github.workspace}}/build -DCMAKE_BUILD_TYPE=${{env.BUILD_TYPE}} -DKERNEL_FLOAT_BUILD_TEST=1 -DKERNEL_FLOAT_BUILD_EXAMPLE=1
+      run: CUDAARCHS=all cmake -B ${{github.workspace}}/build -DCMAKE_BUILD_TYPE=${{env.BUILD_TYPE}} -DCMAKE_CUDA_ARCHITECTURES=\"${{ inputs.cuda-architectures }}\" -DKERNEL_FLOAT_BUILD_TEST=1 -DKERNEL_FLOAT_BUILD_EXAMPLE=1
 
     - name: Build
       # Build your program with the given configuration

.github/workflows/cmake.yml

@@ -12,20 +12,21 @@ jobs:
   build-cuda:
     uses: ./.github/workflows/cmake-run-cuda.yml
     with:
-      cuda-version: "12.8.0"
+      cuda-version: "12.9.0"
+      cuda-architectures: 60-real;70-real;75-real;80-real;86-real
 
   build-hip:
     needs: build-cuda #  Only attempt HIP after CUDA was succesfull
     uses: ./.github/workflows/cmake-run-hip.yml
     with:
-      rocm-version: "6.3.0"
+      rocm-version: "6.4.0"
 
-  build-cuda-12-9:
+  build-cuda-12-8:
     if: github.ref_name == 'main'
     needs: build-cuda
     uses: ./.github/workflows/cmake-run-cuda.yml
     with:
-      cuda-version: "12.9.0"
+      cuda-version: "12.8.0"
 
   build-cuda-12-6:
     if: github.ref_name == 'main'
@@ -48,9 +49,9 @@ jobs:
     with:
       rocm-version: "6.2.0"
 
-  build-hip-6-4:
+  build-hip-6-3:
     if: github.ref_name == 'main'
     needs: build-hip
     uses: ./.github/workflows/cmake-run-hip.yml
     with:
-      rocm-version: "6.4.0"
+      rocm-version: "6.3.0"

include/kernel_float/macros.h

@@ -46,7 +46,7 @@
 #endif  // KERNEL_FLOAT_BF16_AVAILABLE
 
 #ifndef KERNEL_FLOAT_BF16_OPS_AVAILABLE
-#define KERNEL_FLOAT_BF16_OPS_AVAILABLE ((KERNEL_FLOAT_IS_CUDA && __CUDA_ARCH__ >= 800) || KERNEL_FLOAT_IS_HIP)
+#define KERNEL_FLOAT_BF16_OPS_AVAILABLE (KERNEL_FLOAT_IS_CUDA && __CUDA_ARCH__ >= 800)
 #endif
 
 #ifndef KERNEL_FLOAT_FP8_AVAILABLE

single_include/kernel_float.h

@@ -16,8 +16,8 @@
 
 //================================================================================
 // this file has been auto-generated, do not modify its contents!
-// date: 2025-09-02 18:31:16.281730
-// git hash: 023bc75e8ec67145cdcb447c5fd9aa7d7f180cc6
+// date: 2025-09-09 10:07:16.211257
+// git hash: 812619a8a211c35416f42945e01be86eb22efbb3
 //================================================================================
 
 #ifndef KERNEL_FLOAT_MACROS_H
@@ -68,7 +68,7 @@
 #endif  // KERNEL_FLOAT_BF16_AVAILABLE
 
 #ifndef KERNEL_FLOAT_BF16_OPS_AVAILABLE
-#define KERNEL_FLOAT_BF16_OPS_AVAILABLE ((KERNEL_FLOAT_IS_CUDA && __CUDA_ARCH__ >= 800) || KERNEL_FLOAT_IS_HIP)
+#define KERNEL_FLOAT_BF16_OPS_AVAILABLE (KERNEL_FLOAT_IS_CUDA && __CUDA_ARCH__ >= 800)
 #endif
 
 #ifndef KERNEL_FLOAT_FP8_AVAILABLE