[Bug target/99665] New: GCC can generate FPU instructions not supported by the .fpu directive it emits

[clyon at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99665

            Bug ID: 99665
           Summary: GCC can generate FPU instructions not supported by the
                    .fpu directive it emits
           Product: gcc
           Version: 11.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: clyon at gcc dot gnu.org
  Target Milestone: ---

While looking at the gcc.target/arm/simd/vmmla_1.c testcase, I noticed that
when GCC is configured --target=arm-none-linux-gnueabihf --with-float=hard
--with-mode=arm --with-cpu=cortex-a9 --with-fpu=neon-fp16

and invoked with -march=armv8.2-a+i8mm only (ie. no -mfpu, no -mfloat-abi
option) we have #define __ARM_FEATURE_MATMUL_INT8 1,
(so check_effective_target_arm_v8_2a_i8mm_ok_nocache decides no float-abi/fpu
option is necessary)

and the generated code has:
        .arch armv8.2-a
        .fpu neon-fp16
        vsmmla.s8       q0, q1, q2

this instruction is not supported by neon-fp16, but the compiler still decided
it could generate it.

Using -mfpu=auto fixes this by emitting .fpu neon-fp-armv8, but it seems
surprising that the compiler generates illegal code.