[Need help] ninja build failed when compile runtime/common/GPUInfo.h

[Ghjk94522]

I use the dockerfile in the cuda-quantum/docker/build/cudaq.dev.Dockerfile and use the command bash scripts/build_cudaq.sh to build the cuda-quantum from source.
And the compilation failed with:

[342/888] Building CXX object runtime/cudaq/platform/default/rest_server/CMakeFiles/rest-remote-platform-server.dir/helpers/GPUInfo.cpp.o
FAILED: runtime/cudaq/platform/default/rest_server/CMakeFiles/rest-remote-platform-server.dir/helpers/GPUInfo.cpp.o 
/usr/local/llvm/bootstrap/cxx -DCUDAQ_ENABLE_CUDA -DFMT_HEADER_ONLY=1 -D_DEBUG -D__STDC_CONSTANT_MACROS -D__STDC_FORMAT_MACROS -D__STDC_LIMIT_MACROS -Drest_remote_platform_server_EXPORTS -I/workspaces/cuda-quantum/build/include -I/workspaces/cuda-quantum/include -I/workspaces/cuda-quantum/runtime -I/workspaces/cuda-quantum/runtime/cudaq/platform/default/rest_server/. -I/workspaces/cuda-quantum/runtime/cudaq/platform/default/rest_server/../.. -I/workspaces/cuda-quantum/tpls/eigen -I/workspaces/cuda-quantum/tpls/fmt/include -isystem /usr/local/llvm/include -isystem /workspaces/cuda-quantum/tpls/json/include -isystem /usr/local/cuda-12.0/targets/x86_64-linux/include -isystem /usr/local/zlib/include -Wno-ctad-maybe-unsupported -fPIC -fno-semantic-interposition -fvisibility-inlines-hidden -Werror=date-time -Wall -Wextra -Wno-unused-parameter -Wwrite-strings -Wcast-qual -Wno-missing-field-initializers -Wimplicit-fallthrough -Wno-class-memaccess -Wno-redundant-move -Wno-pessimizing-move -Wno-noexcept-type -Wdelete-non-virtual-dtor -Wsuggest-override -Wno-comment -Wno-misleading-indentation -Wctad-maybe-unsupported -fdiagnostics-color -ffunction-sections -fdata-sections -Wno-ctad-maybe-unsupported -Wno-type-limits -O3 -DNDEBUG -std=gnu++20 -fPIC   -D_GNU_SOURCE -D_DEBUG -D__STDC_CONSTANT_MACROS -D__STDC_FORMAT_MACROS -D__STDC_LIMIT_MACROS -UNDEBUG -Werror -MD -MT runtime/cudaq/platform/default/rest_server/CMakeFiles/rest-remote-platform-server.dir/helpers/GPUInfo.cpp.o -MF runtime/cudaq/platform/default/rest_server/CMakeFiles/rest-remote-platform-server.dir/helpers/GPUInfo.cpp.o.d -o runtime/cudaq/platform/default/rest_server/CMakeFiles/rest-remote-platform-server.dir/helpers/GPUInfo.cpp.o -c /workspaces/cuda-quantum/runtime/cudaq/platform/default/rest_server/helpers/GPUInfo.cpp
In file included from /workspaces/cuda-quantum/runtime/cudaq/platform/default/rest_server/helpers/GPUInfo.cpp:9:
/workspaces/cuda-quantum/runtime/common/GPUInfo.h:30:6: error: ‘optional’ in namespace ‘std’ does not name a template type
   30 | std::optional<CudaDeviceProperties> getCudaProperties();
      |      ^~~~~~~~
/workspaces/cuda-quantum/runtime/common/GPUInfo.h:11:1: note: ‘std::optional’ is defined in header ‘<optional>’; did you forget to ‘#include <optional>’?
   10 | #include "nlohmann/json.hpp"
  +++ |+#include <optional>
   11 | namespace cudaq {
/workspaces/cuda-quantum/runtime/cudaq/platform/default/rest_server/helpers/GPUInfo.cpp:24:6: error: ‘optional’ in namespace ‘std’ does not name a template type
   24 | std::optional<CudaDeviceProperties> getCudaProperties() {
      |      ^~~~~~~~
/workspaces/cuda-quantum/runtime/cudaq/platform/default/rest_server/helpers/GPUInfo.cpp:12:1: note: ‘std::optional’ is defined in header ‘<optional>’; did you forget to ‘#include <optional>’?
   11 | #include "cuda_runtime_api.h"
  +++ |+#include <optional>
   12 | #endif

And I have checked the GPUInfo.h, it looks like:

#pragma once
#include "nlohmann/json.hpp"
namespace cudaq {
/// Basic information about the CUDA Device on the remote server.
// Note: this is a subset of `cudaDeviceProp` struct to be populated by the
// server.
struct CudaDeviceProperties {
  std::string deviceName;
  double memoryClockRateMhz;
  double clockRateMhz;
  std::size_t totalGlobalMemMbytes;
  int driverVersion;
  int runtimeVersion;
  NLOHMANN_DEFINE_TYPE_INTRUSIVE(CudaDeviceProperties, deviceName,
                                 memoryClockRateMhz, clockRateMhz,
                                 totalGlobalMemMbytes, driverVersion,
                                 runtimeVersion);
};

/// Helper to query CUDA device info.
// Returns nullopt if no CUDA devices or failed to retrieve the info.
std::optional<CudaDeviceProperties> getCudaProperties();
} // namespace cudaq

Should I add the optional.h to this file? Or I think maybe something wrong with my build steps?

[bmhowe23]

Hi @Ghjk94522 - could you please share your exact steps and host O/S (docker build ..., docker run ..., ...)? Our CI does its build and testing with that docker/build/cudaq.dev.Dockerfile and scripts/build_cudaq.sh, so I think they should work if used in the same way as our CI.

[Ghjk94522]

build docker steps:

docker build -t cuda-q -f docker/build/cudaq.dev.Dockerfile .
docker run -it --network=host --ipc=host --gpus all --privileged --name cuda_q_docker --hostname nv_3090_quant_docker cuda-q-img:latest /bin/bash

host os is: Ubuntu 22.04.4 LTS, and I don't change the env in the docker container.
and in the container I use:

cd /workspaces/cuda-quantum/
bash scripts/build_cudaq.sh -j 32

do I forget some steps?

BTW, I add the #include "llvm/ADT/STLExtras.h" in the GPUInfo.h file, and the compilation success.

[bmhowe23]

Those instructions look correct. Maybe just for reference, here is my full workflow:

git clone --depth=1 https://github.com/NVIDIA/cuda-quantum
cd cuda-quantum
docker pull ghcr.io/nvidia/cuda-quantum-devdeps:ext-cu12.0-gcc11-main
DOCKER_BUILDKIT=0 docker build -t bhowe-cudaq -f docker/build/cudaq.dev.Dockerfile .
docker run -it --rm --name bhowe-temp --gpus all bhowe-cudaq:latest

Then in the container

bash scripts/build_cudaq.sh -v

I do not have any compilation issues with GPUInfo.h. Just for completeness, here is the output of the command when I use the -v option above. It shows configuration options that it might be useful for you to compare against in your environment.

Docker is normally pretty good at removing these "it works for me"-type issues, so I'm not sure what is different between my environment and yours. If the build log (linked above) doesn't reveal any discrepancies with your environment, would you mind pasting the output of lscpu and cat /etc/os-release in this thread just to see if anything jumps out at me?

[Ghjk94522]

It doesn't matter, I'll check this problem, thank u for your help. :)

BTW, actually I found the problem, I delete the llvm dir in the tpls, and when I mkdir llvm in the tpls the compilation success though the llvm dir is empty.