LAPACK ArmPL backend doesn't compile with ArmPL 25.04

[rafbiels]

Summary

ArmPL 25.04 has the following change in the release notes:

By default Arm Performance Libraries now uses C++ std::complex types when compiled with C++ programs.

https://developer.arm.com/documentation/109681/2504/Release-overview/Differences-from-previous-release/Additions-and-changes-?lang=en

Presumably this is the cause of a compilation failure of oneMath LAPACK ArmPL backend:

In file included from /work/oneMath/src/lapack/backends/armpl/armpl_wrappers.cpp:23:
/work/oneMath/src/lapack/backends/armpl/armpl_common.hpp:145:16: error: cannot convert 'std::complex<float>' to 'std::int64_t' (aka 'long') without a conversion operator
  145 |         return static_cast<std::int64_t>((*((T*)&alpha)));
      |                ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Version

3f8f4f3

Environment

• Arm Neoverse-V1 CPU
• arm-performance-libraries_25.04.1_deb_gcc
• Ubuntu 24.04
• Open-source DPC++ tag nightly-2025-04-01

Steps to reproduce

cmake \
  -DTARGET_DOMAINS=lapack \
  -DENABLE_MKLCPU_BACKEND=off -DENABLE_MKLGPU_BACKEND=off \
  -DENABLE_ARMPL_BACKEND=on -DARMPL_ROOT=/path/to/armpl \
  -B build-lapack -G Ninja -DREF_LAPACK_ROOT=/path/to/netlib/
cmake --build build-lapack

Observed behavior

Compilation failure

Expected behavior

Compilation succeeds

[rafbiels]

@adegomme would you be able to look into this as the author of the code?

[adegomme]

Hi, indeed, and this change should make it easier in the future. I added a small patch that should hopefully help in this case.
Tests should pass now, but I'll still have to check if the return value for the scratchpad size is actually given correctly in real() by the lib, but this will have to wait a bit.
Thanks for the report

[sknepper]

PR #667 was merged; @rafbiels , please verify that it fixes this ominous issue and close if so. Thanks!

[rafbiels]

I can confirm the backend works well with the head of the develop branch now. Many thanks @adegomme for the prompt fix!