Cannot compile with CUBLAS

[y10ab1]

I'm not sure which PR caused this issue, but after syncing up to the latest version, I can't compile with make LLAMA_CUBLAS=1

Here is the log:

I UNAME_S:   Linux
I UNAME_P:   x86_64
I UNAME_M:   x86_64
I CFLAGS:    -I. -Icommon -D_XOPEN_SOURCE=600 -D_GNU_SOURCE -DNDEBUG  -std=c11   -fPIC -O3 -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function -Wshadow -Wstrict-prototypes -Wpointer-arith -Wmissing-prototypes -Werror=implicit-int -Werror=implicit-function-declaration -pthread -march=native -mtune=native -Wdouble-promotion 
I CXXFLAGS:  -I. -Icommon -D_XOPEN_SOURCE=600 -D_GNU_SOURCE -DNDEBUG  -std=c++11 -fPIC -O3 -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function -Wmissing-declarations -Wmissing-noreturn -pthread  -march=native -mtune=native -Wno-array-bounds -Wno-format-truncation -Wextra-semi
I NVCCFLAGS:  
I LDFLAGS:    
I CC:        cc (Ubuntu 9.4.0-1ubuntu1~18.04) 9.4.0
I CXX:       g++ (Ubuntu 9.4.0-1ubuntu1~18.04) 9.4.0

rm -vrf *.o tests/*.o *.so *.dll benchmark-matmult common/build-info.cpp *.dot *.gcno tests/*.gcno *.gcda tests/*.gcda *.gcov tests/*.gcov lcov-report gcovr-report main quantize quantize-stats perplexity embedding vdot q8dot train-text-from-scratch convert-llama2c-to-ggml simple batched batched-bench save-load-state server gguf llama-bench libllava.a llava-cli baby-llama beam-search speculative infill tokenize benchmark-matmult parallel finetune export-lora lookahead lookup tests/test-c.o metal tests/test-llama-grammar tests/test-grammar-parser tests/test-double-float tests/test-grad0 tests/test-opt tests/test-quantize-fns tests/test-quantize-perf tests/test-sampling tests/test-tokenizer-0-llama tests/test-tokenizer-0-falcon tests/test-tokenizer-1-llama tests/test-tokenizer-1-bpe tests/test-rope tests/test-backend-ops
removed 'build-info.o'
removed 'common.o'
removed 'console.o'
removed 'ggml-alloc.o'
removed 'ggml-backend.o'
removed 'ggml-cuda.o'
removed 'ggml-quants.o'
removed 'ggml.o'
removed 'grammar-parser.o'
removed 'llama.o'
removed 'sampling.o'
removed 'common/build-info.cpp'
I llama.cpp build info: 
I UNAME_S:   Linux
I UNAME_P:   x86_64
I UNAME_M:   x86_64
I CFLAGS:    -I. -Icommon -D_XOPEN_SOURCE=600 -D_GNU_SOURCE -DNDEBUG -DGGML_USE_CUBLAS -I/usr/local/cuda/include -I/opt/cuda/include -I/home/ubuntu/anaconda3/targets/x86_64-linux/include -I/usr/local/cuda/targets/aarch64-linux/include  -std=c11   -fPIC -O3 -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function -Wshadow -Wstrict-prototypes -Wpointer-arith -Wmissing-prototypes -Werror=implicit-int -Werror=implicit-function-declaration -pthread -march=native -mtune=native -Wdouble-promotion 
I CXXFLAGS:  -I. -Icommon -D_XOPEN_SOURCE=600 -D_GNU_SOURCE -DNDEBUG -DGGML_USE_CUBLAS -I/usr/local/cuda/include -I/opt/cuda/include -I/home/ubuntu/anaconda3/targets/x86_64-linux/include -I/usr/local/cuda/targets/aarch64-linux/include  -std=c++11 -fPIC -O3 -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function -Wmissing-declarations -Wmissing-noreturn -pthread  -march=native -mtune=native -Wno-array-bounds -Wno-format-truncation -Wextra-semi
I NVCCFLAGS: -use_fast_math --forward-unknown-to-host-compiler -arch=native -DGGML_CUDA_DMMV_X=32 -DGGML_CUDA_MMV_Y=1 -DK_QUANTS_PER_ITERATION=2 -DGGML_CUDA_PEER_MAX_BATCH_SIZE=128 
I LDFLAGS:   -lcuda -lcublas -lculibos -lcudart -lcublasLt -lpthread -ldl -lrt -L/usr/local/cuda/lib64 -L/opt/cuda/lib64 -L/home/ubuntu/anaconda3/targets/x86_64-linux/lib -L/usr/local/cuda/targets/aarch64-linux/lib -L/usr/lib/wsl/lib 
I CC:        cc (Ubuntu 9.4.0-1ubuntu1~18.04) 9.4.0
I CXX:       g++ (Ubuntu 9.4.0-1ubuntu1~18.04) 9.4.0

cc  -I. -Icommon -D_XOPEN_SOURCE=600 -D_GNU_SOURCE -DNDEBUG -DGGML_USE_CUBLAS -I/usr/local/cuda/include -I/opt/cuda/include -I/home/ubuntu/anaconda3/targets/x86_64-linux/include -I/usr/local/cuda/targets/aarch64-linux/include  -std=c11   -fPIC -O3 -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function -Wshadow -Wstrict-prototypes -Wpointer-arith -Wmissing-prototypes -Werror=implicit-int -Werror=implicit-function-declaration -pthread -march=native -mtune=native -Wdouble-promotion    -c ggml.c -o ggml.o
g++ -I. -Icommon -D_XOPEN_SOURCE=600 -D_GNU_SOURCE -DNDEBUG -DGGML_USE_CUBLAS -I/usr/local/cuda/include -I/opt/cuda/include -I/home/ubuntu/anaconda3/targets/x86_64-linux/include -I/usr/local/cuda/targets/aarch64-linux/include  -std=c++11 -fPIC -O3 -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function -Wmissing-declarations -Wmissing-noreturn -pthread  -march=native -mtune=native -Wno-array-bounds -Wno-format-truncation -Wextra-semi -c llama.cpp -o llama.o
g++ -I. -Icommon -D_XOPEN_SOURCE=600 -D_GNU_SOURCE -DNDEBUG -DGGML_USE_CUBLAS -I/usr/local/cuda/include -I/opt/cuda/include -I/home/ubuntu/anaconda3/targets/x86_64-linux/include -I/usr/local/cuda/targets/aarch64-linux/include  -std=c++11 -fPIC -O3 -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function -Wmissing-declarations -Wmissing-noreturn -pthread  -march=native -mtune=native -Wno-array-bounds -Wno-format-truncation -Wextra-semi -c common/common.cpp -o common.o
g++ -I. -Icommon -D_XOPEN_SOURCE=600 -D_GNU_SOURCE -DNDEBUG -DGGML_USE_CUBLAS -I/usr/local/cuda/include -I/opt/cuda/include -I/home/ubuntu/anaconda3/targets/x86_64-linux/include -I/usr/local/cuda/targets/aarch64-linux/include  -std=c++11 -fPIC -O3 -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function -Wmissing-declarations -Wmissing-noreturn -pthread  -march=native -mtune=native -Wno-array-bounds -Wno-format-truncation -Wextra-semi -c common/sampling.cpp -o sampling.o
g++ -I. -Icommon -D_XOPEN_SOURCE=600 -D_GNU_SOURCE -DNDEBUG -DGGML_USE_CUBLAS -I/usr/local/cuda/include -I/opt/cuda/include -I/home/ubuntu/anaconda3/targets/x86_64-linux/include -I/usr/local/cuda/targets/aarch64-linux/include  -std=c++11 -fPIC -O3 -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function -Wmissing-declarations -Wmissing-noreturn -pthread  -march=native -mtune=native -Wno-array-bounds -Wno-format-truncation -Wextra-semi -c common/grammar-parser.cpp -o grammar-parser.o
g++ -I. -Icommon -D_XOPEN_SOURCE=600 -D_GNU_SOURCE -DNDEBUG -DGGML_USE_CUBLAS -I/usr/local/cuda/include -I/opt/cuda/include -I/home/ubuntu/anaconda3/targets/x86_64-linux/include -I/usr/local/cuda/targets/aarch64-linux/include  -std=c++11 -fPIC -O3 -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function -Wmissing-declarations -Wmissing-noreturn -pthread  -march=native -mtune=native -Wno-array-bounds -Wno-format-truncation -Wextra-semi -c common/build-info.cpp -o build-info.o
g++ -I. -Icommon -D_XOPEN_SOURCE=600 -D_GNU_SOURCE -DNDEBUG -DGGML_USE_CUBLAS -I/usr/local/cuda/include -I/opt/cuda/include -I/home/ubuntu/anaconda3/targets/x86_64-linux/include -I/usr/local/cuda/targets/aarch64-linux/include  -std=c++11 -fPIC -O3 -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function -Wmissing-declarations -Wmissing-noreturn -pthread  -march=native -mtune=native -Wno-array-bounds -Wno-format-truncation -Wextra-semi -c common/console.cpp -o console.o
nvcc -I. -Icommon -D_XOPEN_SOURCE=600 -D_GNU_SOURCE -DNDEBUG -DGGML_USE_CUBLAS -I/usr/local/cuda/include -I/opt/cuda/include -I/home/ubuntu/anaconda3/targets/x86_64-linux/include -I/usr/local/cuda/targets/aarch64-linux/include  -std=c++11 -fPIC -O3 -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function -Wmissing-declarations -Wmissing-noreturn -pthread  -use_fast_math --forward-unknown-to-host-compiler -arch=native -DGGML_CUDA_DMMV_X=32 -DGGML_CUDA_MMV_Y=1 -DK_QUANTS_PER_ITERATION=2 -DGGML_CUDA_PEER_MAX_BATCH_SIZE=128  -Wno-pedantic -Xcompiler "-Wno-array-bounds -Wno-format-truncation -Wextra-semi" -c ggml-cuda.cu -o ggml-cuda.o
cc  -I. -Icommon -D_XOPEN_SOURCE=600 -D_GNU_SOURCE -DNDEBUG -DGGML_USE_CUBLAS -I/usr/local/cuda/include -I/opt/cuda/include -I/home/ubuntu/anaconda3/targets/x86_64-linux/include -I/usr/local/cuda/targets/aarch64-linux/include  -std=c11   -fPIC -O3 -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function -Wshadow -Wstrict-prototypes -Wpointer-arith -Wmissing-prototypes -Werror=implicit-int -Werror=implicit-function-declaration -pthread -march=native -mtune=native -Wdouble-promotion    -c ggml-alloc.c -o ggml-alloc.o
cc  -I. -Icommon -D_XOPEN_SOURCE=600 -D_GNU_SOURCE -DNDEBUG -DGGML_USE_CUBLAS -I/usr/local/cuda/include -I/opt/cuda/include -I/home/ubuntu/anaconda3/targets/x86_64-linux/include -I/usr/local/cuda/targets/aarch64-linux/include  -std=c11   -fPIC -O3 -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function -Wshadow -Wstrict-prototypes -Wpointer-arith -Wmissing-prototypes -Werror=implicit-int -Werror=implicit-function-declaration -pthread -march=native -mtune=native -Wdouble-promotion    -c ggml-backend.c -o ggml-backend.o
cc -I. -Icommon -D_XOPEN_SOURCE=600 -D_GNU_SOURCE -DNDEBUG -DGGML_USE_CUBLAS -I/usr/local/cuda/include -I/opt/cuda/include -I/home/ubuntu/anaconda3/targets/x86_64-linux/include -I/usr/local/cuda/targets/aarch64-linux/include  -std=c11   -fPIC -O3 -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function -Wshadow -Wstrict-prototypes -Wpointer-arith -Wmissing-prototypes -Werror=implicit-int -Werror=implicit-function-declaration -pthread -march=native -mtune=native -Wdouble-promotion     -c ggml-quants.c -o ggml-quants.o
g++ -I. -Icommon -D_XOPEN_SOURCE=600 -D_GNU_SOURCE -DNDEBUG -DGGML_USE_CUBLAS -I/usr/local/cuda/include -I/opt/cuda/include -I/home/ubuntu/anaconda3/targets/x86_64-linux/include -I/usr/local/cuda/targets/aarch64-linux/include  -std=c++11 -fPIC -O3 -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function -Wmissing-declarations -Wmissing-noreturn -pthread  -march=native -mtune=native -Wno-array-bounds -Wno-format-truncation -Wextra-semi examples/main/main.cpp ggml.o llama.o common.o sampling.o grammar-parser.o build-info.o console.o ggml-cuda.o ggml-alloc.o ggml-backend.o ggml-quants.o -o main -lcuda -lcublas -lculibos -lcudart -lcublasLt -lpthread -ldl -lrt -L/usr/local/cuda/lib64 -L/opt/cuda/lib64 -L/home/ubuntu/anaconda3/targets/x86_64-linux/lib -L/usr/local/cuda/targets/aarch64-linux/lib -L/usr/lib/wsl/lib 
/usr/bin/ld: cannot find -lcuda
collect2: error: ld returned 1 exit status
Makefile:588: recipe for target 'main' failed
make: *** [main] Error 1```

[y10ab1]

I solved the issue by manually adding my cuda path to the makefile.

[tastypear]

I meet the same problem. I can compile it correctly in Kaggle last time.

[slaren]

After #4606 we need to link with the CUDA driver library (-lcuda), and in some systems this may require adding a new library path to MK_LDFLAGS in the Makefile. PRs to add the path of this library in your system are welcome. The cmake build should find the path automatically.

[tastypear]

@slaren Thank you. I have compiled it successfully~

For those who have the same trouble:
In Kaggle, libcuda.so is in /usr/local/nvidia/lib64. So just add -L/usr/local/nvidia/lib64 before -lcuda then it should work. Order seems important.
I have little experience with writing MakeFile, this might not be a good solution.

[henk717]

@slaren How can we do this in a universal way including Github CI? There is no driver installed there and our binary needs to support all the drivers not just the one the system would have.

[slaren]

I don't think this is different than linking to any other CUDA library. Our CI is able to build the binaries despite having no driver installed either.