[clang] add masked sqrt 512

Signed-off-by: Shreeyash Pandey <[email protected]>

Author: bojle

clang/lib/Headers/avx512fintrin.h

@@ -1464,17 +1464,15 @@ _mm512_sqrt_pd(__m512d __A) {
                                            _MM_FROUND_CUR_DIRECTION);
 }
 
-static __inline__ __m512d __DEFAULT_FN_ATTRS512
-_mm512_mask_sqrt_pd (__m512d __W, __mmask8 __U, __m512d __A)
-{
+static __inline__ __m512d __DEFAULT_FN_ATTRS512_CONSTEXPR
+_mm512_mask_sqrt_pd(__m512d __W, __mmask8 __U, __m512d __A) {
   return (__m512d)__builtin_ia32_selectpd_512(__U,
                                               (__v8df)_mm512_sqrt_pd(__A),
                                               (__v8df)__W);
 }
 
-static __inline__ __m512d __DEFAULT_FN_ATTRS512
-_mm512_maskz_sqrt_pd (__mmask8 __U, __m512d __A)
-{
+static __inline__ __m512d __DEFAULT_FN_ATTRS512_CONSTEXPR
+_mm512_maskz_sqrt_pd(__mmask8 __U, __m512d __A) {
   return (__m512d)__builtin_ia32_selectpd_512(__U,
                                               (__v8df)_mm512_sqrt_pd(__A),
                                               (__v8df)_mm512_setzero_pd());
@@ -1499,17 +1497,15 @@ _mm512_sqrt_ps(__m512 __A) {
                                           _MM_FROUND_CUR_DIRECTION);
 }
 
-static  __inline__ __m512 __DEFAULT_FN_ATTRS512
-_mm512_mask_sqrt_ps(__m512 __W, __mmask16 __U, __m512 __A)
-{
+static __inline__ __m512 __DEFAULT_FN_ATTRS512_CONSTEXPR
+_mm512_mask_sqrt_ps(__m512 __W, __mmask16 __U, __m512 __A) {
   return (__m512)__builtin_ia32_selectps_512(__U,
                                              (__v16sf)_mm512_sqrt_ps(__A),
                                              (__v16sf)__W);
 }
 
-static  __inline__ __m512 __DEFAULT_FN_ATTRS512
-_mm512_maskz_sqrt_ps( __mmask16 __U, __m512 __A)
-{
+static __inline__ __m512 __DEFAULT_FN_ATTRS512_CONSTEXPR
+_mm512_maskz_sqrt_ps(__mmask16 __U, __m512 __A) {
   return (__m512)__builtin_ia32_selectps_512(__U,
                                              (__v16sf)_mm512_sqrt_ps(__A),
                                              (__v16sf)_mm512_setzero_ps());

clang/test/CodeGen/X86/avx512f-builtins.c

@@ -29,6 +29,12 @@ __m512d test_mm512_mask_sqrt_pd (__m512d __W, __mmask8 __U, __m512d __A)
   return _mm512_mask_sqrt_pd (__W,__U,__A);
 }
 
+TEST_CONSTEXPR(match_m512d(_mm512_mask_sqrt_pd(
+    _mm512_set_pd(800.0, 700.0, 600.0, 500.0, 400.0, 300.0, 200.0, 100.0),
+    0b11000011,
+    _mm512_set_pd(64.0, 49.0, 36.0, 25.0, 16.0, 9.0, 4.0, 1.0)),
+    1.0, 2.0, 300.0, 400.0, 500.0, 600.0, 7.0, 8.0));
+
 __m512d test_mm512_maskz_sqrt_pd (__mmask8 __U, __m512d __A)
 {
   // CHECK-LABEL: test_mm512_maskz_sqrt_pd 
@@ -38,6 +44,10 @@ __m512d test_mm512_maskz_sqrt_pd (__mmask8 __U, __m512d __A)
   return _mm512_maskz_sqrt_pd (__U,__A);
 }
 
+TEST_CONSTEXPR(match_m512d(_mm512_maskz_sqrt_pd(0b00001111,
+    _mm512_set_pd(64.0, 49.0, 36.0, 25.0, 16.0, 9.0, 4.0, 1.0)),
+    1.0, 2.0, 3.0, 4.0, 0.0, 0.0, 0.0, 0.0));
+
 __m512d test_mm512_mask_sqrt_round_pd(__m512d __W,__mmask8 __U,__m512d __A)
 {
   // CHECK-LABEL: test_mm512_mask_sqrt_round_pd