SYCL `long double` expression requires 128 bit size 'long double' type support

[abagusetty]

Describe the bug

Compilation flag (-mlong-double-64) doesnt seem to apply for the device-side, results in long double` error

error: expression requires 128 bit size 'long double' type support, but target 'nvptx64-nvidia-cuda' does not support it

Compiled via:
clang++ -fsycl -mlong-double-64 -fsycl-targets=nvptx64-nvidia-cuda -Xsycl-target-backend --cuda-gpu-arch=sm_80 sycl_long_double.cpp -o sycl_long_double.out

To reproduce

1. Include code snippet as short as possible
2. Specify the command which should be used to compile the program
3. Specify the command which should be used to launch the program
4. Indicate what is wrong and what was expected

#include <sycl/sycl.hpp>

int main(int argc, char **argv) {
  sycl::queue myQueue(sycl::gpu_selector_v);

  long double* ptr = sycl::malloc_device<long double>(512, myQueue);
  myQueue.parallel_for( sycl::range<1>(512), [=](sycl::id<1> idx) {
    ptr[idx] = 1.1;
  });

  return 0;
}

Environment

• OS: [e.g Windows/Linux]
• Target device and vendor: [e.g. Nvidia GPU]
• DPC++ version: [e.g. commit hash or output of clang++ --version]
• Dependencies version: [e.g. the output of nvidia-smi and sycl-ls --verbose]

clang version 19.0.0git (https://github.com/intel/llvm.git 99d3ce4093d0a9b697f7ce02fc4543dd059b6dc8)
Target: x86_64-unknown-linux-gnu
Thread model: posix
InstalledDir: /soft/compilers/oneapi/upstream/install-12.2.91_06112024/install/bin
Build config: +assertions

Additional context

Works fine on PVC with Compiler (2025.x.0.20240629)
Haven't got a chance to test it with HIP backend

[JackAKirk]

long double should be using standard 64 bit double in nvptx backend but long double itself currently gives the same error. This is a valid request.

[jchlanda]

Seems like this option is lost when we construct a compilation invocation for the device, the long double 64 option is not being included. So the host is compiled with long double equal to 64, but the device is still on 128. I'll have a go at fixing it.

[jchlanda]

I believe this issue should be addressed by: #16441. Feel free to reopen this ticket if more work is needed.