This is the final patch, based on Art's original submission that I've committed. It fixes the concerns I had in the second patch of the set by reworking the framework tests to properly set -mfpu=auto and to select the architecture on Arm using the appropriate feature extensions.
I'm still not entirely sure that it is right for the framework to pick one set of flags and then for an individual test to then try and pick a different set (either the framework should choose, or the test, but never both); but at least these tests now handle this consistently. Artemiy Volkov (1): testsuite: require arm_v8_3a_fp16_complex_neon for complex fp16 tests Richard Earnshaw (1): testsuite: arm: rework some target-supports checks to use -mfpu=auto .../fast-math-bb-slp-complex-add-half-float.c | 2 +- ...th-bb-slp-complex-add-pattern-half-float.c | 2 +- .../fast-math-bb-slp-complex-mla-half-float.c | 2 +- .../fast-math-bb-slp-complex-mls-half-float.c | 2 +- .../fast-math-bb-slp-complex-mul-half-float.c | 2 +- .../fast-math-complex-add-half-float.c | 2 +- ...fast-math-complex-add-pattern-half-float.c | 2 +- .../fast-math-complex-mla-half-float.c | 4 +- .../fast-math-complex-mls-half-float.c | 2 +- .../fast-math-complex-mul-half-float.c | 2 +- .../advsimd-intrinsics/vector-complex_f16.c | 8 +- gcc/testsuite/lib/target-supports.exp | 119 ++++++++++++------ 12 files changed, 95 insertions(+), 54 deletions(-) -- 2.43.0
