[clang] Add _mm512_kunpackd and _mm512_kunpackw to avx512fintrin.h

Add AVX-512 mask unpack intrinsics _mm512_kunpackd and _mm512_kunpackw
to avx512fintrin.h alongside the existing _mm512_kunpackb intrinsic.

These intrinsics extract and concatenate the lower halves of mask
registers, using the existing backend support for __builtin_ia32_kunpckdi
and __builtin_ia32_kunpcksi builtins.

Also adds __mmask32 and __mmask64 type definitions to avx512fintrin.h
for completeness.

Tests already exist in clang/test/CodeGen/X86/avx512bw-builtins.c.

Author: ahmednoursphinx

clang/lib/Headers/avx512fintrin.h

@@ -40,6 +40,8 @@ typedef long long __m512i_u __attribute__((__vector_size__(64), __aligned__(1)))
 
 typedef unsigned char __mmask8;
 typedef unsigned short __mmask16;
+typedef unsigned int __mmask32;
+typedef unsigned long long __mmask64;
 
 /* Rounding mode macros.  */
 #define _MM_FROUND_TO_NEAREST_INT   0x00
@@ -8100,6 +8102,18 @@ _mm512_kunpackb (__mmask16 __A, __mmask16 __B)
   return (__mmask16) __builtin_ia32_kunpckhi ((__mmask16) __A, (__mmask16) __B);
 }
 
+static __inline__ __mmask64 __DEFAULT_FN_ATTRS
+_mm512_kunpackd (__mmask64 __A, __mmask64 __B)
+{
+  return (__mmask64) __builtin_ia32_kunpckdi ((__mmask64) __A, (__mmask64) __B);
+}
+
+static __inline__ __mmask32 __DEFAULT_FN_ATTRS
+_mm512_kunpackw (__mmask32 __A, __mmask32 __B)
+{
+  return (__mmask32) __builtin_ia32_kunpcksi ((__mmask32) __A, (__mmask32) __B);
+}
+
 static __inline__ __mmask16 __DEFAULT_FN_ATTRS_CONSTEXPR
 _mm512_kxnor(__mmask16 __A, __mmask16 __B) {
   return (__mmask16) __builtin_ia32_kxnorhi ((__mmask16) __A, (__mmask16) __B);