[Bug] won't build with HipBLAS. sd.cpp tries to use CUDA regardless.

[Username1020382920382]

Git commit

newest as of Jan 30th

Operating System & Version

CachyOS (kernel 6.18.7-2)

GGML backends

HIP

Command-line arguments used

mkdir build && cd build if command -v rocminfo; then export GFX_NAME=$(rocminfo | awk '/ *Name: +gfx[1-9]/ {print $2; exit}'); else echo "rocminfo missing!"; fi if [ -z "${GFX_NAME}" ]; then echo "Error: Couldn't detect GPU!"; else echo "Building for GPU: ${GFX_NAME}"; fi cmake .. -G "Ninja" -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ -DSD_HIPBLAS=ON -DCMAKE_BUILD_TYPE=Release -DGPU_TARGETS=$GFX_NAME -DAMDGPU_TARGETS=$GFX_NAME -DCMAKE_BUILD_WITH_INSTALL_RPATH=ON -DCMAKE_POSITION_INDEPENDENT_CODE=ON cmake --build . --config Release

Steps to reproduce

followed the building guide from: https://github.com/leejet/stable-diffusion.cpp/blob/master/docs/build.md for HipBLAS on RX 6600. SD.CPP tries to compile with CUDA regardless of HIP being enabled.

here is the output of 'cmake -LAH | grep -i CUDA'

// ggml: use CUDA
GGML_CUDA:BOOL=OFF
// ggml: cuda link binary compression mode; requires cuda 12.8+
GGML_CUDA_COMPRESSION_MODE:STRING=size
// ggml: compile ggml FlashAttention CUDA kernels
GGML_CUDA_FA:BOOL=OFF
GGML_CUDA_FA_ALL_QUANTS:BOOL=OFF
GGML_CUDA_FORCE_CUBLAS:BOOL=OFF
GGML_CUDA_FORCE_MMQ:BOOL=OFF
// ggml: use CUDA graphs (llama.cpp only)
GGML_CUDA_GRAPHS:BOOL=OFF
GGML_CUDA_NO_PEER_COPY:BOOL=OFF
// ggml: do not try to use CUDA VMM
GGML_CUDA_NO_VMM:BOOL=OFF
GGML_CUDA_PEER_MAX_BATCH_SIZE:STRING=128
// sd: cuda backend
SD_CUDA:BOOL=OFF
// sd: x1.5 faster softmax, indeterministic (sometimes, same seed don't generate same image), cuda only

What you expected to happen

finished compile and usable rocm.

What actually happened

clang++: error: unsupported CUDA gpu architecture: gfx1032

Logs / error messages / stack trace

[ 1%] Built target zip
[ 7%] Built target ggml-base
[ 8%] Building HIP object ggml/src/ggml-hip/CMakeFiles/ggml-hip.dir//ggml-cuda/acc.cu.o
clang++: error: unsupported CUDA gpu architecture: gfx1032
clang++: error: cannot find CUDA installation; provide its path via '--cuda-path', or pass '-nocudainc' to build without CUDA includes
make[2]: *** [ggml/src/ggml-hip/CMakeFiles/ggml-hip.dir/build.make:78: ggml/src/ggml-hip/CMakeFiles/ggml-hip.dir//ggml-cuda/acc.cu.o] Error 1
make[1]: *** [CMakeFiles/Makefile2:401: ggml/src/ggml-hip/CMakeFiles/ggml-hip.dir/all] Error 2
make: *** [Makefile:136: all] Error 2

Additional context / environment details

No response