Arm64 SHA512 intrinsics fail to compile with GCC ("target specific option mismatch")

Consider the following source file test-sha512.c, using SHA512 intrinsics on Arm64 (aarch64):

#include <arm_neon.h>
const uint64_t data[256] = {0,};

void test()
{
    uint64x2_t a = vld1q_u64(data);
    a = vsha512h2q_u64(a, a, a);
}

On Ubuntu 22.10 (virtual machine on an MacBook M1), with gcc 12.2.0, I have the error "inlining failed in call to ‘always_inline’" and "target specific option mismatch":

$ gcc -c test-sha512.c -march=armv8-a+sha3
In file included from test-sha512.c:1:
/usr/lib/gcc/aarch64-linux-gnu/12/include/arm_neon.h: In function ‘test’:
/usr/lib/gcc/aarch64-linux-gnu/12/include/arm_neon.h:29671:1: error: inlining failed in call to ‘always_inline’ ‘vsha512h2q_u64’: target specific option mismatch
29671 | vsha512h2q_u64 (uint64x2_t __a, uint64x2_t __b, uint64x2_t __c)
      | ^~~~~~~~~~~~~~
test-sha512.c:7:9: note: called from here
    7 |     a = vsha512h2q_u64(a, a, a);
      |         ^~~~~~~~~~~~~~~~~~~~~~~
/usr/lib/gcc/aarch64-linux-gnu/12/include/arm_neon.h:29671:1: error: inlining failed in call to ‘always_inline’ ‘vsha512h2q_u64’: target specific option mismatch
29671 | vsha512h2q_u64 (uint64x2_t __a, uint64x2_t __b, uint64x2_t __c)
      | ^~~~~~~~~~~~~~
test-sha512.c:7:9: note: called from here
    7 |     a = vsha512h2q_u64(a, a, a);
      |         ^~~~~~~~~~~~~~~~~~~~~~~

With clang 15.0.6, it compiles correctly and a full SHA512 implementation using Arm64 intrinsics and compiled with clang works correctly.

$ clang -c test-sha512.c -march=armv8-a+sha3

Note: the Arm architecture defines distinct features for SHA1, SHA256, SHA512 and SHA3. However, gcc and clang know crypto, sha2 and sha3 only. The SHA512 instructions (cryptographically part of SHA2) are activated with sha3. Weird. Anyway...

The similar Arm64 intrinsics for AES, SHA1 and SHA256 compile correctly with gcc. The problem is specific to SHA512.

Other tests, without success, same error:

• The error "target specific option mismatch" may suggest a misinterpretation of the -march option. I tried with all Armv8 options (-march=armv8-a+fp+simd+crypto+crc+lse+fp16+rcpc+rdma+dotprod+aes+sha2+sha3+sm4+fp16fml+sve+profile+rng+memtag+sb+ssbs+predres+sve2+sve2-sm4+sve2-aes+sve2-sha3+sve2-bitperm+tme+i8mm+f32mm+f64mm+bf16+flagm+pauth+ls64+mops) and with -march=armv9-a.
• Using -march=native, considering that the M1 supports SHA512.
• Using -mcpu=neoverse-v1 or -mcpu=neoverse-n2 or other known Arm cores which support SHA512.
• Various types of optimization options.
• Tried various (but not all) suggestions from gcc --target-help.

Is this a known error? I did not find any reference online about this error on Arm64 SHA512 intrinsics.

EDIT: SHA256 intrinsics also fail with the same error. Only SHA1 intrinsics work. My previous tests with SHA256 were made using clang, sorry.

EDIT 2: SHA256 intrinsics work with -march=armv8-a+sha2+crypto but not with -march=armv8-a+sha2. SHA512 still don't work, even with all -march options.

The SHA3/SHA512 extension is documented by ARM only for ARMv8.2-A onward. As such, gcc requires you to use -march=armv8.2-a+sha3 (or v8.3-a, etc.)