[Clang][AVX512][INTRINSICS] adding round cvt and fix regular cvtps_ph

Differential Revision: http://reviews.llvm.org/D20870 llvm-svn: 271498
llvm · Jun 2, 2016 · 9e7d0a9 · 9e7d0a9
1 parent 1c2cb1d
commit 9e7d0a9
Show file tree

Hide file tree

Showing 2 changed files with 63 additions and 8 deletions.
diff --git a/clang/lib/Headers/avx512vlintrin.h b/clang/lib/Headers/avx512vlintrin.h
@@ -9383,26 +9383,58 @@ _mm256_maskz_cvtph_ps (__mmask8 __U, __m128i __A)
                 (__mmask8) __U);
 }
 
-#define _mm_mask_cvtps_ph(W, U, A, I) __extension__ ({ \
+static __inline __m128i __DEFAULT_FN_ATTRS
+_mm_mask_cvtps_ph (__m128i __W, __mmask8 __U, __m128 __A)
+{
+  return (__m128i) __builtin_ia32_vcvtps2ph_mask ((__v4sf) __A, _MM_FROUND_CUR_DIRECTION,
+                                                  (__v8hi) __W,
+                                                  (__mmask8) __U);
+}
+
+static __inline __m128i __DEFAULT_FN_ATTRS
+_mm_maskz_cvtps_ph (__mmask8 __U, __m128 __A)
+{
+  return (__m128i) __builtin_ia32_vcvtps2ph_mask ((__v4sf) __A, _MM_FROUND_CUR_DIRECTION,
+                                                  (__v8hi) _mm_setzero_si128 (),
+                                                  (__mmask8) __U);
+}
+
+#define _mm_mask_cvt_roundps_ph(W, U, A, I) __extension__ ({ \
   (__m128i)__builtin_ia32_vcvtps2ph_mask((__v4sf)(__m128)(A), (int)(I), \
                                          (__v8hi)(__m128i)(W), \
                                          (__mmask8)(U)); })
 
-#define _mm_maskz_cvtps_ph(U, A, I) __extension__ ({ \
+#define _mm_maskz_cvt_roundps_ph(U, A, I) __extension__ ({ \
   (__m128i)__builtin_ia32_vcvtps2ph_mask((__v4sf)(__m128)(A), (int)(I), \
                                          (__v8hi)_mm_setzero_si128(), \
                                          (__mmask8)(U)); })
 
-#define _mm256_mask_cvtps_ph(W, U, A, I) __extension__ ({ \
+static __inline __m128i __DEFAULT_FN_ATTRS
+_mm256_mask_cvtps_ph (__m128i __W, __mmask8 __U, __m256 __A)
+{
+  return (__m128i) __builtin_ia32_vcvtps2ph256_mask ((__v8sf) __A, _MM_FROUND_CUR_DIRECTION,
+                                                      (__v8hi) __W,
+                                                      (__mmask8) __U);
+}
+
+static __inline __m128i __DEFAULT_FN_ATTRS
+_mm256_maskz_cvtps_ph ( __mmask8 __U, __m256 __A)
+{
+  return (__m128i) __builtin_ia32_vcvtps2ph256_mask ((__v8sf) __A, _MM_FROUND_CUR_DIRECTION,
+                                                      (__v8hi) _mm_setzero_si128(),
+                                                      (__mmask8) __U);
+}
+#define _mm256_mask_cvt_roundps_ph(W, U, A, I) __extension__ ({ \
   (__m128i)__builtin_ia32_vcvtps2ph256_mask((__v8sf)(__m256)(A), (int)(I), \
                                             (__v8hi)(__m128i)(W), \
                                             (__mmask8)(U)); })
 
-#define _mm256_maskz_cvtps_ph(U, A, I) __extension__ ({ \
+#define _mm256_maskz_cvt_roundps_ph(U, A, I) __extension__ ({ \
   (__m128i)__builtin_ia32_vcvtps2ph256_mask((__v8sf)(__m256)(A), (int)(I), \
                                             (__v8hi)_mm_setzero_si128(), \
                                             (__mmask8)(U)); })
 
+
 #undef __DEFAULT_FN_ATTRS
 #undef __DEFAULT_FN_ATTRS_BOTH
 

diff --git a/clang/test/CodeGen/avx512vl-builtins.c b/clang/test/CodeGen/avx512vl-builtins.c
@@ -6726,24 +6726,47 @@ __m256 test_mm256_maskz_cvtph_ps(__mmask8 __U, __m128i __A) {
 __m128i test_mm_mask_cvtps_ph(__m128i __W, __mmask8 __U, __m128 __A) {
   // CHECK-LABEL: @test_mm_mask_cvtps_ph
   // CHECK: @llvm.x86.avx512.mask.vcvtps2ph.128
-  return _mm_mask_cvtps_ph(__W, __U, __A, _MM_FROUND_CUR_DIRECTION); 
+  return _mm_mask_cvtps_ph(__W, __U, __A); 
 }
 
 __m128i test_mm_maskz_cvtps_ph(__mmask8 __U, __m128 __A) {
   // CHECK-LABEL: @test_mm_maskz_cvtps_ph
   // CHECK: @llvm.x86.avx512.mask.vcvtps2ph.128
-  return _mm_maskz_cvtps_ph(__U, __A, _MM_FROUND_CUR_DIRECTION); 
+  return _mm_maskz_cvtps_ph(__U, __A); 
 }
 
 __m128i test_mm256_mask_cvtps_ph(__m128i __W, __mmask8 __U, __m256 __A) {
   // CHECK-LABEL: @test_mm256_mask_cvtps_ph
   // CHECK: @llvm.x86.avx512.mask.vcvtps2ph.256
-  return _mm256_mask_cvtps_ph(__W, __U, __A, _MM_FROUND_CUR_DIRECTION); 
+  return _mm256_mask_cvtps_ph(__W, __U, __A); 
 }
 
 __m128i test_mm256_maskz_cvtps_ph(__mmask8 __U, __m256 __A) {
   // CHECK-LABEL: @test_mm256_maskz_cvtps_ph
   // CHECK: @llvm.x86.avx512.mask.vcvtps2ph.256
-  return _mm256_maskz_cvtps_ph(__U, __A, _MM_FROUND_CUR_DIRECTION); 
+  return _mm256_maskz_cvtps_ph(__U, __A);
 }
 
+__m128i test_mm_mask_cvt_roundps_ph(__m128i __W, __mmask8 __U, __m128 __A) {
+  // CHECK-LABEL: @test_mm_mask_cvt_roundps_ph
+  // CHECK: @llvm.x86.avx512.mask.vcvtps2ph.128
+  return _mm_mask_cvt_roundps_ph(__W, __U, __A, _MM_FROUND_CUR_DIRECTION); 
+}
+
+__m128i test_mm_maskz_cvt_roundps_ph(__mmask8 __U, __m128 __A) {
+  // CHECK-LABEL: @test_mm_maskz_cvt_roundps_ph
+  // CHECK: @llvm.x86.avx512.mask.vcvtps2ph.128
+  return _mm_maskz_cvt_roundps_ph(__U, __A, _MM_FROUND_CUR_DIRECTION); 
+}
+
+__m128i test_mm256_mask_cvt_roundps_ph(__m128i __W, __mmask8 __U, __m256 __A) {
+  // CHECK-LABEL: @test_mm256_mask_cvt_roundps_ph
+  // CHECK: @llvm.x86.avx512.mask.vcvtps2ph.256
+  return _mm256_mask_cvt_roundps_ph(__W, __U, __A, _MM_FROUND_CUR_DIRECTION); 
+}
+
+__m128i test_mm256_maskz_cvt_roundps_ph(__mmask8 __U, __m256 __A) {
+  // CHECK-LABEL: @test_mm256_maskz_cvt_roundps_ph
+  // CHECK: @llvm.x86.avx512.mask.vcvtps2ph.256
+  return _mm256_maskz_cvt_roundps_ph(__U, __A, _MM_FROUND_CUR_DIRECTION); 
+}