Fix new CUDA FA on Touring (#413)

ikawrakow · Iwan Kawrakow · web-flow · commit 627f40643759 · 2025-05-12T15:09:33.000+03:00
Co-authored-by: Iwan Kawrakow &lt;iwan.kawrakow@gmail.com&gt;
diff --git a/ggml/src/ggml-cuda/fattn-new-mma.cu b/ggml/src/ggml-cuda/fattn-new-mma.cu
@@ -273,11 +273,11 @@ struct fattn_mma_f16_config<576, 512> {
     }
 
     static constexpr __device__ int get_nbatch_V2_device(int ncols) {
-#if __CUDA_ARCH__ == GML_CUDA_CC_TURING
+#if __CUDA_ARCH__ == CC_TURING
         return ncols <= 16 ? 64 : 128;
 #else
         return ncols <= 16 ? 256 : 128;
-#endif // __CUDA_ARCH__ == GML_CUDA_CC_TURING
+#endif // __CUDA_ARCH__ == CC_TURING
     }
 
     static int get_nbatch_combine_host(const int /*cc*/, const int /*ncols*/) {

Original file line number	Diff line number	Diff line change
`@@ -273,11 +273,11 @@ struct fattn_mma_f16_config<576, 512> {`
`273`	`273`	`}`
`274`	`274`
`275`	`275`	`static constexpr __device__ int get_nbatch_V2_device(int ncols) {`
`276`		`-#if __CUDA_ARCH__ == GML_CUDA_CC_TURING`
	`276`	`+#if __CUDA_ARCH__ == CC_TURING`
`277`	`277`	`return ncols <= 16 ? 64 : 128;`
`278`	`278`	`#else`
`279`	`279`	`return ncols <= 16 ? 256 : 128;`
`280`		`-#endif // __CUDA_ARCH__ == GML_CUDA_CC_TURING`
	`280`	`+#endif // __CUDA_ARCH__ == CC_TURING`
`281`	`281`	`}`
`282`	`282`
`283`	`283`	`static int get_nbatch_combine_host(const int /cc/, const int /ncols/) {`