Cannot compile for HIPBLAS ( undefined reference due to --no-allow-shlib-undefined: void ggml_cuda_flash_attn_ext_wmma_f16_case )

Open mega-ice opened this issue 1 year ago • 0 comments

ROCm: 6.1.2 Ubuntu 6.5.0-41-generic / Ubuntu 22.04.4 LTS RDNA3 GPU commit bf4cb4abad4e35c74b387df034cc4ac7b22e5fe6

i have tried to compile it with cmake or make, and i get the same error. CC=/opt/rocm/llvm/bin/clang CXX=/opt/rocm/llvm/bin/clang++ cmake -DWHISPER_HIPBLAS=ON -DCMAKE_CXX_FLAGS="--offload-arch=gfx1100" OR CXX=hipcc CC=hipcc cmake -DWHISPER_HIPBLAS=ON -DAMDGPU_TARGETS="gfx1100" -DBUILD_SHARED_LIBS=ON ../whisper.cpp OR WHISPER_HIPBLAS=1 make -j8

. . [ 74%] Built target whisper [ 76%] Building CXX object examples/CMakeFiles/common.dir/common.cpp.o [ 78%] Building CXX object examples/CMakeFiles/common.dir/common-ggml.cpp.o [ 80%] Building CXX object examples/CMakeFiles/common.dir/grammar-parser.cpp.o [ 82%] Linking CXX static library libcommon.a [ 82%] Built target common [ 85%] Building CXX object examples/main/CMakeFiles/main.dir/main.cpp.o [ 87%] Linking CXX executable ../../bin/main ld.lld: error: undefined reference due to --no-allow-shlib-undefined: void ggml_cuda_flash_attn_ext_wmma_f16_case<64, 32, float>(ggml_backend_cuda_context&, ggml_tensor*)

ld.lld: error: undefined reference due to --no-allow-shlib-undefined: void ggml_cuda_flash_attn_ext_wmma_f16_case<64, 8, __half>(ggml_backend_cuda_context&, ggml_tensor*)

ld.lld: error: undefined reference due to --no-allow-shlib-undefined: void ggml_cuda_flash_attn_ext_wmma_f16_case<64, 16, float>(ggml_backend_cuda_context&, ggml_tensor*)

ld.lld: error: undefined reference due to --no-allow-shlib-undefined: void ggml_cuda_flash_attn_ext_wmma_f16_case<64, 16, __half>(ggml_backend_cuda_context&, ggml_tensor*)

ld.lld: error: undefined reference due to --no-allow-shlib-undefined: void ggml_cuda_flash_attn_ext_wmma_f16_case<64, 32, __half>(ggml_backend_cuda_context&, ggml_tensor*)

ld.lld: error: undefined reference due to --no-allow-shlib-undefined: void ggml_cuda_flash_attn_ext_wmma_f16_case<80, 32, float>(ggml_backend_cuda_context&, ggml_tensor*)

ld.lld: error: undefined reference due to --no-allow-shlib-undefined: void ggml_cuda_flash_attn_ext_wmma_f16_case<96, 32, float>(ggml_backend_cuda_context&, ggml_tensor*)

ld.lld: error: undefined reference due to --no-allow-shlib-undefined: void ggml_cuda_flash_attn_ext_wmma_f16_case<112, 32, float>(ggml_backend_cuda_context&, ggml_tensor*)

ld.lld: error: undefined reference due to --no-allow-shlib-undefined: void ggml_cuda_flash_attn_ext_wmma_f16_case<128, 32, float>(ggml_backend_cuda_context&, ggml_tensor*)

ld.lld: error: undefined reference due to --no-allow-shlib-undefined: void ggml_cuda_flash_attn_ext_wmma_f16_case<80, 16, float>(ggml_backend_cuda_context&, ggml_tensor*)

ld.lld: error: undefined reference due to --no-allow-shlib-undefined: void ggml_cuda_flash_attn_ext_wmma_f16_case<96, 16, float>(ggml_backend_cuda_context&, ggml_tensor*)

ld.lld: error: undefined reference due to --no-allow-shlib-undefined: void ggml_cuda_flash_attn_ext_wmma_f16_case<112, 16, float>(ggml_backend_cuda_context&, ggml_tensor*)

ld.lld: error: undefined reference due to --no-allow-shlib-undefined: void ggml_cuda_flash_attn_ext_wmma_f16_case<128, 16, float>(ggml_backend_cuda_context&, ggml_tensor*)

ld.lld: error: undefined reference due to --no-allow-shlib-undefined: void ggml_cuda_flash_attn_ext_wmma_f16_case<256, 16, float>(ggml_backend_cuda_context&, ggml_tensor*)

ld.lld: error: undefined reference due to --no-allow-shlib-undefined: void ggml_cuda_flash_attn_ext_wmma_f16_case<80, 16, __half>(ggml_backend_cuda_context&, ggml_tensor*)

ld.lld: error: undefined reference due to --no-allow-shlib-undefined: void ggml_cuda_flash_attn_ext_wmma_f16_case<96, 16, __half>(ggml_backend_cuda_context&, ggml_tensor*)

ld.lld: error: undefined reference due to --no-allow-shlib-undefined: void ggml_cuda_flash_attn_ext_wmma_f16_case<112, 16, __half>(ggml_backend_cuda_context&, ggml_tensor*)

ld.lld: error: undefined reference due to --no-allow-shlib-undefined: void ggml_cuda_flash_attn_ext_wmma_f16_case<128, 16, __half>(ggml_backend_cuda_context&, ggml_tensor*)

ld.lld: error: undefined reference due to --no-allow-shlib-undefined: void ggml_cuda_flash_attn_ext_wmma_f16_case<256, 16, __half>(ggml_backend_cuda_context&, ggml_tensor*)

ld.lld: error: undefined reference due to --no-allow-shlib-undefined: void ggml_cuda_flash_attn_ext_wmma_f16_case<96, 8, __half>(ggml_backend_cuda_context&, ggml_tensor*)

ld.lld: error: too many errors emitted, stopping now (use --error-limit=0 to see all errors) clang: error: linker command failed with exit code 1 (use -v to see invocation)

Jun 19 '24 09:06 mega-ice