compilation with `-fno-sycl-rdc` is broken due to `__sycl_cmath_wrapper_impl.hpp` (missing definition of `SYCL_EXTERNAL`)

[romintomasetti]

Describe the bug

4f0afd5 introduced __sycl_cmath_wrapper_impl.hpp.

It seems that this file broke compilation with the -fno-sycl-rdc flag as soon as <cmath> is included and targeting CUDA (haven't tried AMD).
The code is as simple as:

#include <cmath>

int main() {}

And using icpx (2025.3.3.20260319) with compile command:

icpx -x c++ -fsycl -fsycl-targets=nvptx64-nvidia-cuda -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_90 test.cpp

is fine but with:

icpx -x c++ -fsycl -fsycl-targets=nvptx64-nvidia-cuda -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_90 -fno-sycl-rdc test.cpp

it fails with:

# [5/5] RUN icpx -x c++ -fsycl -fsycl-targets=nvptx64-nvidia-cuda -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_86 -fno-sycl-rdc test.cpp
# 0.429 In file included from test.cpp:1:
# 0.429 In file included from /opt/intel/oneapi/compiler/2025.3/bin/compiler/../../include/sycl/stl_wrappers/cmath:41:
# 0.429 /opt/intel/oneapi/compiler/2025.3/bin/compiler/../../include/sycl/stl_wrappers/__sycl_cmath_wrapper_impl.hpp:225:12: error: unknown type name 'SYCL_EXTERNAL'
# 0.429   225 | extern "C" SYCL_EXTERNAL float __nv_nearbyintf(float);
# 0.429       |            ^
# 0.430 /opt/intel/oneapi/compiler/2025.3/bin/compiler/../../include/sycl/stl_wrappers/__sycl_cmath_wrapper_impl.hpp:226:12: error: unknown type name 'SYCL_EXTERNAL'
# 0.430   226 | extern "C" SYCL_EXTERNAL double __nv_nearbyint(double);
# 0.430       |            ^
# 0.487 2 errors generated.
# ERROR: process "/bin/sh -c icpx -x c++ -fsycl -fsycl-targets=nvptx64-nvidia-cuda -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_86 -fno-sycl-rdc test.cpp" did not complete successfully: exit code: 1

Below I've put a Dockerfile to reproduce the issue, along with a typical output.

I think @npmiller @bader introduced that file. It may be related to:

• SYCL_EXTERNAL definition is inconsistent. #20541

To reproduce

The Dockerfile to reproduce.

FROM nvcr.io/nvidia/cuda:12.8.1-devel-ubuntu24.04

ARG COMPILER_VERSION=2025.3

RUN <<EOF
    set -ex

    apt update && apt --no-install-recommends --yes install wget gpg

    wget -O- https://apt.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRODUCTS.PUB \
        | gpg --dearmor | tee /usr/share/keyrings/oneapi-archive-keyring.gpg > /dev/null
    echo "deb [signed-by=/usr/share/keyrings/oneapi-archive-keyring.gpg] https://apt.repos.intel.com/oneapi all main" \
        | tee /etc/apt/sources.list.d/oneAPI.list

    apt update

    apt --no-install-recommends --yes install intel-oneapi-compiler-dpcpp-cpp-${COMPILER_VERSION}
EOF

ENV PATH=/opt/intel/oneapi/compiler/${COMPILER_VERSION}/bin/:$PATH
ENV LD_LIBRARY_PATH=/opt/intel/oneapi/compiler/${COMPILER_VERSION}/lib:$LD_LIBRARY_PATH

COPY <<EOF test.cpp
#include <cmath>

int main() {}
EOF

RUN icpx --version

# Compiling without specifying "-fno-sycl-rdc" is fine.
RUN icpx -x c++ -fsycl -fsycl-targets=nvptx64-nvidia-cuda -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_90 test.cpp

# Compiling with "-fno-sycl-rdc" is broken.
RUN icpx -x c++ -fsycl -fsycl-targets=nvptx64-nvidia-cuda -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_90 -fno-sycl-rdc test.cpp

An example of output for:

docker buildx build --tag test --progress=plain - < repro.dockerfile

#0 building with "default" instance using docker driver

#1 [internal] load build definition from Dockerfile
#1 transferring dockerfile: 2.41kB done
#1 DONE 0.0s

#2 [internal] load metadata for nvcr.io/nvidia/cuda:12.8.1-devel-ubuntu24.04
#2 DONE 1.1s

#3 [internal] load .dockerignore
#3 transferring context: 2B done
#3 DONE 0.0s

#4 [internal] preparing inline document
#4 DONE 0.0s

#5 [1/6] FROM nvcr.io/nvidia/cuda:12.8.1-devel-ubuntu24.04@sha256:520292dbb4f755fd360766059e62956e9379485d9e073bbd2f6e3c20c270ed66
#5 DONE 0.0s

#6 [2/6] RUN <<EOF (set -ex...)
#6 CACHED

#7 [3/6] COPY <<EOF test.cpp
#7 CACHED

#8 [4/6] RUN icpx --version
#8 0.229 Intel(R) oneAPI DPC++/C++ Compiler 2025.3.3 (2025.3.3.20260319)
#8 0.229 Target: x86_64-unknown-linux-gnu
#8 0.229 Thread model: posix
#8 0.229 InstalledDir: /opt/intel/oneapi/compiler/2025.3/bin/compiler
#8 0.229 Configuration file: /opt/intel/oneapi/compiler/2025.3/bin/compiler/../icpx.cfg
#8 DONE 0.3s

#9 [5/6] RUN icpx -x c++ -fsycl -fsycl-targets=nvptx64-nvidia-cuda -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_90 test.cpp
#9 DONE 0.8s

#10 [6/6] RUN icpx -x c++ -fsycl -fsycl-targets=nvptx64-nvidia-cuda -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_90 -fno-sycl-rdc test.cpp
#10 0.393 In file included from test.cpp:1:
#10 0.393 In file included from /opt/intel/oneapi/compiler/2025.3/bin/compiler/../../include/sycl/stl_wrappers/cmath:41:
#10 0.393 /opt/intel/oneapi/compiler/2025.3/bin/compiler/../../include/sycl/stl_wrappers/__sycl_cmath_wrapper_impl.hpp:225:12: error: unknown type name 'SYCL_EXTERNAL'
#10 0.393   225 | extern "C" SYCL_EXTERNAL float __nv_nearbyintf(float);
#10 0.393       |            ^
#10 0.394 /opt/intel/oneapi/compiler/2025.3/bin/compiler/../../include/sycl/stl_wrappers/__sycl_cmath_wrapper_impl.hpp:226:12: error: unknown type name 'SYCL_EXTERNAL'
#10 0.394   226 | extern "C" SYCL_EXTERNAL double __nv_nearbyint(double);
#10 0.394       |            ^
#10 0.436 2 errors generated.
#10 ERROR: process "/bin/sh -c icpx -x c++ -fsycl -fsycl-targets=nvptx64-nvidia-cuda -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_90 -fno-sycl-rdc test.cpp" did not complete successfully: exit code: 1
------
 > [6/6] RUN icpx -x c++ -fsycl -fsycl-targets=nvptx64-nvidia-cuda -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_90 -fno-sycl-rdc test.cpp:
0.393 In file included from test.cpp:1:
0.393 In file included from /opt/intel/oneapi/compiler/2025.3/bin/compiler/../../include/sycl/stl_wrappers/cmath:41:
0.393 /opt/intel/oneapi/compiler/2025.3/bin/compiler/../../include/sycl/stl_wrappers/__sycl_cmath_wrapper_impl.hpp:225:12: error: unknown type name 'SYCL_EXTERNAL'
0.393   225 | extern "C" SYCL_EXTERNAL float __nv_nearbyintf(float);
0.393       |            ^
0.394 /opt/intel/oneapi/compiler/2025.3/bin/compiler/../../include/sycl/stl_wrappers/__sycl_cmath_wrapper_impl.hpp:226:12: error: unknown type name 'SYCL_EXTERNAL'
0.394   226 | extern "C" SYCL_EXTERNAL double __nv_nearbyint(double);
0.394       |            ^
0.436 2 errors generated.
------
Dockerfile:46
--------------------
  44 |     # 0.487 2 errors generated.
  45 |     # ERROR: process "/bin/sh -c icpx -x c++ -fsycl -fsycl-targets=nvptx64-nvidia-cuda -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_86 -fno-sycl-rdc test.cpp" did not complete successfully: exit code: 1
  46 | >>> RUN icpx -x c++ -fsycl -fsycl-targets=nvptx64-nvidia-cuda -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_90 -fno-sycl-rdc test.cpp
  47 |     
--------------------
ERROR: failed to build: failed to solve: process "/bin/sh -c icpx -x c++ -fsycl -fsycl-targets=nvptx64-nvidia-cuda -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_90 -fno-sycl-rdc test.cpp" did not complete successfully: exit code: 1

Environment

No response

Additional context

No response