[Headers][X86] Add constexpr support for some AVX512 masked extension/truncation intrinsics.

[moorabbit]

The following AVX[512] intrinsics are now constexpr:

• _mm512_mask_cvtepi8_epi32
• _mm512_maskz_cvtepi8_epi32
• _mm512_mask_cvtepi8_epi64
• _mm512_maskz_cvtepi8_epi64
• _mm512_mask_cvtepi16_epi32
• _mm512_maskz_cvtepi16_epi32
• _mm512_mask_cvtepi16_epi64
• _mm512_maskz_cvtepi16_epi64
• _mm512_mask_cvtepi32_epi64
• _mm512_maskz_cvtepi32_epi64
• _mm512_mask_cvtepu8_epi32
• _mm512_maskz_cvtepu8_epi32
• _mm512_mask_cvtepu8_epi64
• _mm512_maskz_cvtepu8_epi64
• _mm512_mask_cvtepu16_epi32
• _mm512_maskz_cvtepu16_epi32
• _mm512_mask_cvtepu16_epi64
• _mm512_maskz_cvtepu16_epi64
• _mm512_mask_cvtepu32_epi64
• _mm512_maskz_cvtepu32_epi64
• _mm512_mask_cvtepi8_epi16
• _mm512_maskz_cvtepi8_epi16
• _mm512_mask_cvtepu8_epi16
• _mm512_maskz_cvtepu8_epi16
• _mm_cvtepi16_epi8
• _mm256_cvtepi16_epi8
• _mm256_mask_cvtepi16_epi8
• _mm256_maskz_cvtepi16_epi8

This PR is part 1 of a series of PRs fixing #154539

[moorabbit]

@RKSimon these functions:

llvm-project/clang/lib/Headers/avx512vlbwintrin.h

Lines 1447 to 1459 in f5d3cf4

	static __inline__ __m128i __DEFAULT_FN_ATTRS128
	_mm_mask_cvtepi16_epi8 (__m128i __O, __mmask8 __M, __m128i __A) {
	return (__m128i) __builtin_ia32_pmovwb128_mask ((__v8hi) __A,
	(__v16qi) __O,
	__M);
	}
	static __inline__ __m128i __DEFAULT_FN_ATTRS128
	_mm_maskz_cvtepi16_epi8 (__mmask8 __M, __m128i __A) {
	return (__m128i) __builtin_ia32_pmovwb128_mask ((__v8hi) __A,
	(__v16qi) _mm_setzero_si128(),
	__M);
	}

Can't be made constexpr because __builtin_ia32_pmovwb128_mask is not constexpr.

Is there a plan to remove __builtin_ia32_pmovwb128_mask and replace it with a __builtin_ia32_selectb_128? Following the approach here:

llvm-project/clang/lib/Headers/avx512vlbwintrin.h

Lines 1485 to 1490 in f5d3cf4

	static __inline__ __m128i __DEFAULT_FN_ATTRS256
	_mm256_mask_cvtepi16_epi8 (__m128i __O, __mmask16 __M, __m256i __A) {
	return (__m128i)__builtin_ia32_selectb_128((__mmask16)__M,
	(__v16qi)_mm256_cvtepi16_epi8(__A),
	(__v16qi)__O);
	}

[RKSimon]

At the moment we need to keep _mm_mask_cvtepi16_epi8 etc as the selection mask only acts on part of the vector (we're truncating v8i16 to v8i8, using the mask8 on that and passing through all the upper half of the passthrough to create a v16i8) - the best we might be able to do is a custom __builtin_ia32_selectb_64 but it'll be messy - we'll have to come back to it later, maybe when the _mm512_kunpack intrinsics are constexpr?

[RKSimon]

LGTM - cheers