lower __mm512_andnot_ps/__mm512_andnot_pd to IR

[lsaba]

Summary

lower mm512_andnot_ps/mm512_andnot_pd to IR instead of using builtin_ia32_andnps512_mask/builtin_ia32_andnpd512_mask builtins

[igorb]

LGTM

cfe/trunk/lib/Headers/avx512dqintrin.h

return (__m512) __builtin_ia32_andps512_mask ((__v16sf) __A,
             (__v16sf) __B,
             (__v16sf)
             _mm512_setzero_ps (),
             (__mmask16) __U);
 }
 
 static __inline__ __m512d __DEFAULT_FN_ATTRS
 _mm512_andnot_pd (__m512d __A, __m512d __B) {
-  return (__m512d) __builtin_ia32_andnpd512_mask ((__v8df) __A,
-              (__v8df) __B,
-              (__v8df)
-              _mm512_setzero_pd (),
-              (__mmask8) -1);
+  return (__m512d)(~(__v8du)__A & (__v8du)__B);
 }
 
 static __inline__ __m512d __DEFAULT_FN_ATTRS
 _mm512_mask_andnot_pd (__m512d __W, __mmask8 __U, __m512d __A, __m512d __B) {
   return (__m512d) __builtin_ia32_andnpd512_mask ((__v8df) __A,
               (__v8df) __B,
               (__v8df) __W,
               (__mmask8) __U);
 }
 
 static __inline__ __m512d __DEFAULT_FN_ATTRS
 _mm512_maskz_andnot_pd (__mmask8 __U, __m512d __A, __m512d __B) {
   return (__m512d) __builtin_ia32_andnpd512_mask ((__v8df) __A,
               (__v8df) __B,
               (__v8df)
               _mm512_setzero_pd (),
               (__mmask8) __U);
 }
 
 static __inline__ __m512 __DEFAULT_FN_ATTRS
 _mm512_andnot_ps (__m512 __A, __m512 __B) {
-  return (__m512) __builtin_ia32_andnps512_mask ((__v16sf) __A,
-             (__v16sf) __B,
-             (__v16sf)
-             _mm512_setzero_ps (),
-             (__mmask16) -1);
+  return (__m512)(~(__v16su)__A & (__v16su)__B);
 }
 
 static __inline__ __m512 __DEFAULT_FN_ATTRS
 _mm512_mask_andnot_ps (__m512 __W, __mmask16 __U, __m512 __A, __m512 __B) {
   return (__m512) __builtin_ia32_andnps512_mask ((__v16sf) __A,
              (__v16sf) __B,
              (__v16sf) __W,
              (__mmask16) __U);

cfe/trunk/test/CodeGen/avx512dq-builtins.c

 __m512 test_mm512_maskz_and_ps (__mmask16 __U, __m512 __A, __m512 __B) {
   // CHECK-LABEL: @test_mm512_maskz_and_ps
   // CHECK: @llvm.x86.avx512.mask.and.ps.512
   return (__m512) _mm512_maskz_and_ps(__U, __A, __B);
 }
 
 __m512d test_mm512_andnot_pd (__m512d __A, __m512d __B) {
   // CHECK-LABEL: @test_mm512_andnot_pd
-  // CHECK: @llvm.x86.avx512.mask.andn.pd.512
+  // CHECK: xor <8 x i64> %{{.*}}, <i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1>
+  // CHECK: and <8 x i64>
   return (__m512d) _mm512_andnot_pd(__A, __B);
 }
 
 __m512d test_mm512_mask_andnot_pd (__m512d __W, __mmask8 __U, __m512d __A, __m512d __B) {
   // CHECK-LABEL: @test_mm512_mask_andnot_pd
   // CHECK: @llvm.x86.avx512.mask.andn.pd.512
   return (__m512d) _mm512_mask_andnot_pd(__W, __U, __A, __B);
 }
 
 __m512d test_mm512_maskz_andnot_pd (__mmask8 __U, __m512d __A, __m512d __B) {
   // CHECK-LABEL: @test_mm512_maskz_andnot_pd
   // CHECK: @llvm.x86.avx512.mask.andn.pd.512
   return (__m512d) _mm512_maskz_andnot_pd(__U, __A, __B);
 }
 
 __m512 test_mm512_andnot_ps (__m512 __A, __m512 __B) {
   // CHECK-LABEL: @test_mm512_andnot_ps
-  // CHECK: @llvm.x86.avx512.mask.andn.ps.512
+  // CHECK: xor <16 x i32> %{{.*}}, <i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1>
+  // CHECK: and <16 x i32>
   return (__m512) _mm512_andnot_ps(__A, __B);
 }
 
 __m512 test_mm512_mask_andnot_ps (__m512 __W, __mmask16 __U, __m512 __A, __m512 __B) {
   // CHECK-LABEL: @test_mm512_mask_andnot_ps
   // CHECK: @llvm.x86.avx512.mask.andn.ps.512
   return (__m512) _mm512_mask_andnot_ps(__W, __U, __A, __B);
 }