Repository navigation
main doesn't compile with nvcc 13.0 (Linux, GCC 12/13 host): __connect_declfn_v, __transform_completion_signatures, __receiver_adaptor #2403
Description
At current main (1d58877), nvcc can't compile a basic sender chain in a .cu file. Tested in nvidia/cuda:13.0.0-devel-ubuntu24.04 with nvcc V13.0.48 and GCC 13.3 (default) or GCC 12.4 (-ccbin g++-12) as host compiler; both give the same errors. Related: #2402 (nvcc with an MSVC host, which has additional problems).
// nvcc_basic.cu
#include <stdexec/execution.hpp>
int main()
{
auto s = stdexec::just(41) | stdexec::then([](int i) { return i + 1; });
auto [v] = stdexec::sync_wait(std::move(s)).value();
return v == 42 ? 0 : 1;
}nvcc -std=c++20 --extended-lambda --expt-relaxed-constexpr -arch=sm_89 -I stdexec/include nvcc_basic.cu
Errors (paths shortened, repeated lines removed):
include/stdexec/__detail/__connect.hpp(83): error: declaration is incompatible with variable template "stdexec::__connect::__connect_declfn_v" (declared at line 77)
include/stdexec/__detail/__connect.hpp(90): error: declaration is incompatible with variable template "stdexec::__connect::__connect_declfn_v" (declared at line 77)
include/stdexec/__detail/__connect.hpp(103): error: declaration is incompatible with variable template "stdexec::__connect::__connect_declfn_v" (declared at line 97)
include/stdexec/__detail/__connect.hpp(110): error: declaration is incompatible with variable template "stdexec::__connect::__connect_declfn_v" (declared at line 97)
include/stdexec/__detail/__receiver_adaptor.hpp(125): error: expected a ">"
include/stdexec/__detail/__receiver_adaptor.hpp(125): error: too few arguments for concept template "std::derived_from"
include/stdexec/__detail/__receiver_adaptor.hpp(125): error: qualified name is not allowed
include/stdexec/__detail/__typeinfo.hpp(101): error: static assertion failed
include/stdexec/__detail/__transform_completion_signatures.hpp(360): error: type name is not allowed
include/stdexec/__detail/__transform_completion_signatures.hpp(400): error: type name is not allowed
include/stdexec/__detail/__sync_wait.hpp(353): error: no instance of function template "stdexec::connect_t::operator()" matches the argument list
include/stdexec/__detail/__sync_wait.hpp(355): error: static assertion failed with "..."
As far as I can see, the CI doesn't build anything with nvcc: the GPU jobs use clang and nvc++, the CPU jobs don't compile CUDA. That would explain why this goes unnoticed.
HPX compiles stdexec in .cu files and has been carrying patches for nvcc (https://github.com/TheHPXProject/hpx/tree/master/cmake, HPX_StdexecNvcc*.patch). Against 04de75d, the revision HPX uses today, the patched stdexec compiles this program with nvcc 13.0. The largest patch (HPX_StdexecNvccWorkarounds.patch, which changes __connect.hpp, __receiver_adaptor.hpp, __transform_completion_signatures.hpp, __typeinfo.hpp, __env.hpp and __counting_scopes.hpp) no longer applies to main, and with the three HPX patches that still apply (AsyncScopeNvcc, which is #2304, CudaStdAtomic and NvccDiagnostics) main still fails with the errors above.
Is nvcc meant to be supported? If yes, would you take a CI job with nvcc (for example nvcc 13 with GCC 13 as host compiler, compiling the tests as CUDA sources) and PRs that bring HPX's workarounds to main? We'd be happy to help with those.
Activity