correct behavior of some intrinsics

Author: sayantn

crates/core_arch/src/x86/avx512bw.rs

@@ -10705,9 +10705,13 @@ pub fn _mm_cvtepi16_epi8(a: __m128i) -> __m128i {
 #[cfg_attr(test, assert_instr(vpmovwb))]
 pub fn _mm_mask_cvtepi16_epi8(src: __m128i, k: __mmask8, a: __m128i) -> __m128i {
     unsafe {
-        let convert = _mm_cvtepi16_epi8(a).as_i8x16();
-        let k: __mmask16 = 0b11111111_11111111 & k as __mmask16;
-        transmute(simd_select_bitmask(k, convert, src.as_i8x16()))
+        let a = _mm_cvtepi16_epi8(a).as_i8x16();
+        let src = simd_shuffle!(
+            src.as_i8x16(),
+            i8x16::ZERO,
+            [0, 1, 2, 3, 4, 5, 6, 7, 16, 16, 16, 16, 16, 16, 16, 16]
+        );
+        simd_select_bitmask(k as u16, a, src).as_m128i()
     }
 }
 
@@ -10719,11 +10723,7 @@ pub fn _mm_mask_cvtepi16_epi8(src: __m128i, k: __mmask8, a: __m128i) -> __m128i
 #[stable(feature = "stdarch_x86_avx512", since = "1.89")]
 #[cfg_attr(test, assert_instr(vpmovwb))]
 pub fn _mm_maskz_cvtepi16_epi8(k: __mmask8, a: __m128i) -> __m128i {
-    unsafe {
-        let convert = _mm_cvtepi16_epi8(a).as_i8x16();
-        let k: __mmask16 = 0b11111111_11111111 & k as __mmask16;
-        transmute(simd_select_bitmask(k, convert, i8x16::ZERO))
-    }
+    _mm_mask_cvtepi16_epi8(_mm_setzero_si128(), k, a)
 }
 
 /// Convert packed signed 16-bit integers in a to packed 8-bit integers with signed saturation, and store the results in dst.

crates/core_arch/src/x86/avx512f.rs

@@ -11149,10 +11149,7 @@ pub fn _mm256_maskz_cvtpd_ps(k: __mmask8, a: __m256d) -> __m128 {
 #[stable(feature = "stdarch_x86_avx512", since = "1.89")]
 #[cfg_attr(test, assert_instr(vcvtpd2ps))]
 pub fn _mm_mask_cvtpd_ps(src: __m128, k: __mmask8, a: __m128d) -> __m128 {
-    unsafe {
-        let convert = _mm_cvtpd_ps(a);
-        transmute(simd_select_bitmask(k, convert.as_f32x4(), src.as_f32x4()))
-    }
+    unsafe { vcvtpd2ps128(a.as_f64x2(), src.as_f32x4(), k).as_m128() }
 }
 
 /// Convert packed double-precision (64-bit) floating-point elements in a to packed single-precision (32-bit) floating-point elements, and store the results in dst using zeromask k (elements are zeroed out when the corresponding mask bit is not set).
@@ -11259,10 +11256,7 @@ pub fn _mm256_maskz_cvtpd_epi32(k: __mmask8, a: __m256d) -> __m128i {
 #[stable(feature = "stdarch_x86_avx512", since = "1.89")]
 #[cfg_attr(test, assert_instr(vcvtpd2dq))]
 pub fn _mm_mask_cvtpd_epi32(src: __m128i, k: __mmask8, a: __m128d) -> __m128i {
-    unsafe {
-        let convert = _mm_cvtpd_epi32(a);
-        transmute(simd_select_bitmask(k, convert.as_i32x4(), src.as_i32x4()))
-    }
+    unsafe { vcvtpd2dq128(a.as_f64x2(), src.as_i32x4(), k).as_m128i() }
 }
 
 /// Convert packed double-precision (64-bit) floating-point elements in a to packed 32-bit integers, and store the results in dst using zeromask k (elements are zeroed out when the corresponding mask bit is not set).
@@ -42483,9 +42477,13 @@ unsafe extern "C" {
 
     #[link_name = "llvm.x86.avx512.mask.cvtps2pd.512"]
     fn vcvtps2pd(a: f32x8, src: f64x8, mask: u8, sae: i32) -> f64x8;
+    #[link_name = "llvm.x86.avx512.mask.cvtpd2ps"]
+    fn vcvtpd2ps128(a: f64x2, src: f32x4, mask: u8) -> f32x4;
     #[link_name = "llvm.x86.avx512.mask.cvtpd2ps.512"]
     fn vcvtpd2ps(a: f64x8, src: f32x8, mask: u8, rounding: i32) -> f32x8;
 
+    #[link_name = "llvm.x86.avx512.mask.cvtpd2dq.128"]
+    fn vcvtpd2dq128(a: f64x2, src: i32x4, k: u8) -> i32x4;
     #[link_name = "llvm.x86.avx512.mask.cvtpd2dq.512"]
     fn vcvtpd2dq(a: f64x8, src: i32x8, mask: u8, rounding: i32) -> i32x8;
 

crates/core_arch/src/x86/avx512fp16.rs

@@ -1007,7 +1007,7 @@ pub fn _mm_comilt_sh(a: __m128h, b: __m128h) -> i32 {
 #[target_feature(enable = "avx512fp16")]
 #[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")]
 pub fn _mm_comineq_sh(a: __m128h, b: __m128h) -> i32 {
-    _mm_comi_sh::<_CMP_NEQ_OS>(a, b)
+    _mm_comi_sh::<_CMP_NEQ_US>(a, b)
 }
 
 /// Compare the lower half-precision (16-bit) floating-point elements in a and b for equality, and
@@ -1073,7 +1073,7 @@ pub fn _mm_ucomilt_sh(a: __m128h, b: __m128h) -> i32 {
 #[target_feature(enable = "avx512fp16")]
 #[unstable(feature = "stdarch_x86_avx512_f16", issue = "127213")]
 pub fn _mm_ucomineq_sh(a: __m128h, b: __m128h) -> i32 {
-    _mm_comi_sh::<_CMP_NEQ_OQ>(a, b)
+    _mm_comi_sh::<_CMP_NEQ_UQ>(a, b)
 }
 
 /// Load 128-bits (composed of 8 packed half-precision (16-bit) floating-point elements) from memory into