CUDA build failed

[emkman]

using CUDA 11.2 on Ubuntu 18.04

error: linking with `cc` failed: exit status: 1
  |
  = note: "cc" "-m64" "/tmp/rustc34h9Yw/symbols.o" "/root/aleo-prover/target/release/deps/aleo_prover-18d199d49d2a8ded.aleo_prover.1d97f3f0-cgu.0.rcgu.o" "-Wl,--as-needed" "-L" "/root/aleo-prover/target/release/deps" "-L" "/root/aleo-prover/target/release/build/snarkvm-cuda-ecfceb3bcf9c7dd3/out" "-L" "/root/aleo-prover/target/release/build/blst-a1efdb917a5ef45b/out" "-L" "/root/aleo-prover/target/release/build/curl-sys-57f21fcae3bbf520/out/build" "-L" "/root/aleo-prover/target/release/build/libz-sys-299ca258f4d02532/out/lib" "-L" "/root/aleo-prover/target/release/build/libz-sys-299ca258f4d02532/out/lib" "-L" "/root/aleo-prover/target/release/build/ring-e88af84875cf3f54/out" "-L" "/root/.rustup/toolchains/stable-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/lib" "-Wl,-Bstatic" "/tmp/rustc34h9Yw/libring-c300d20d950cad5b.rlib" "/tmp/rustc34h9Yw/libsnarkvm_cuda-abdc56e5a0f6a673.rlib" "/tmp/rustc34h9Yw/libblst-da5eb185f18af887.rlib" "/tmp/rustc34h9Yw/libcurl_sys-3bf80d05cdb1333e.rlib" "/tmp/rustc34h9Yw/liblibz_sys-97c515cd179fdaae.rlib" "/root/.rustup/toolchains/stable-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/lib/libcompiler_builtins-272ca28f0b8538d5.rlib" "-Wl,-Bdynamic" "-lssl" "-lcrypto" "-lgcc_s" "-lutil" "-lrt" "-lpthread" "-lm" "-ldl" "-lc" "-Wl,--eh-frame-hdr" "-Wl,-znoexecstack" "-L" "/root/.rustup/toolchains/stable-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/lib" "-o" "/root/aleo-prover/target/release/deps/aleo_prover-18d199d49d2a8ded" "-Wl,--gc-sections" "-pie" "-Wl,-zrelro,-znow" "-Wl,-O1" "-nodefaultlibs"
  = note: /root/aleo-prover/target/release/deps/aleo_prover-18d199d49d2a8ded.aleo_prover.1d97f3f0-cgu.0.rcgu.o: In function `rayon_core::thread_pool::ThreadPool::install::{{closure}}':
          aleo_prover.1d97f3f0-cgu.0:(.text._ZN10rayon_core11thread_pool10ThreadPool7install28_$u7b$$u7b$closure$u7d$$u7d$17h2a6910cc806e82e5E+0x192): undefined reference to `snarkvm_ntt'
          /root/aleo-prover/target/release/deps/aleo_prover-18d199d49d2a8ded.aleo_prover.1d97f3f0-cgu.0.rcgu.o: In function `snarkvm_algorithms::msm::variable_base::VariableBase::msm':
          aleo_prover.1d97f3f0-cgu.0:(.text._ZN18snarkvm_algorithms3msm13variable_base12VariableBase3msm17hd2f277f0ea0033ebE+0x53): undefined reference to `snarkvm_msm'
          collect2: error: ld returned 1 exit status

  = help: some `extern` functions couldn't be found; some native libraries may need to be installed or have their path specified
  = note: use the `-l` flag to specify native libraries to link
  = note: use the `cargo:rustc-link-lib` directive to specify the native libraries to link with Cargo (see https://doc.rust-lang.org/cargo/reference/build-scripts.html#cargorustc-link-libkindname)

[HarukaMa]

nvcc needs to be in PATH, or use NVCC=/path/to/nvcc

[emkman]

Got it working, thanks! It seems that there is max 1 thread pool per gpu enforced, correct? I am actually getting worse performance with the GPU than without due to this limitation. Is that normal?

[HarukaMa]

While you can actually use multiple thread pools per GPU, CUDA calls are not parallelized so the GPU usage will be low. I know nothing about GPU programming so don't know how to fix that.

[emkman]

Makes sense but in my testing the thread flag gets ignored when activating the GPU flag. Closing this issue, appreciate the help.