Update ggml-cuda/mmvq.cu

Co-authored-by: Johannes Gäßler <johannesg@5d6.de>

.github/workflows/build.yml

@@ -459,7 +459,7 @@ jobs:
           path: build/bin/${{ matrix.build }}
 
   windows-cublas:
-    runs-on: windows-2019
+    runs-on: windows-latest
 
     strategy:
       matrix:

Makefile

@@ -297,10 +297,10 @@ ggml-cuda/%.o: ggml-cuda/%.cu ggml-cuda/%.cuh ggml.h ggml-common.h ggml-cuda/com
 
 ggml-cuda.o: ggml-cuda.cu ggml-cuda.h ggml.h ggml-backend.h ggml-backend-impl.h ggml-common.h $(wildcard ggml-cuda/*.cuh)
 	$(NVCC) $(NVCCFLAGS) $(CXXFLAGS) -Wno-pedantic -c $< -o $@
+endif
 
 whisper-mel-cuda.o: whisper-mel-cuda.cu whisper.h ggml.h ggml-backend.h whisper-mel.hpp whisper-mel-cuda.hpp
 	$(NVCC) $(NVCCFLAGS) $(CXXFLAGS) -Wno-pedantic -c $< -o $@
-endif
 
 ifdef WHISPER_HIPBLAS
 	ROCM_PATH   ?= /opt/rocm

whisper.cpp

@@ -3167,7 +3167,7 @@ struct mel_calc_cpu : public whisper_mel_calc {
 }
 
 whisper_mel_calc * whisper_mel_calc_create(ggml_backend_t backend, const whisper_filters & filters) {
-#if defined(GGML_USE_CUDA) && !defined(GGML_USE_HIPBLAS)
+#if GGML_USE_CUDA
     if (ggml_backend_is_cuda(backend)) {
         auto ret = whisper_mel_calc_create_cuda(backend, filters);
         // run a warmup to avoid the first kernel launch overhead (thus we get the best perf even on the first run)