[X86] Add other flavors of AVX512 cmpps/cmppd intrinsics.

llvm-svn: 227773

2 files changed, 73 insertions, 19 deletions

diff --git a/clang/lib/Headers/avx512fintrin.h b/clang/lib/Headers/avx512fintrin.h
index 09b29fa0496..e8c449d71a9 100644
--- a/clang/lib/Headers/avx512fintrin.h
+++ b/clang/lib/Headers/avx512fintrin.h
@@ -672,19 +672,37 @@ _mm512_mask_blend_epi32(__mmask16 __U, __m512i __A, __m512i __W)
 
 /* Compare */
 
-#define _mm512_cmp_ps_mask(a, b, p) __extension__ ({ \
-  __m512 __a = (a); \
-  __m512 __b = (b); \
-  (__mmask16)__builtin_ia32_cmpps512_mask((__v16sf)__a, (__v16sf)__b, (p), \
-                                          (__mmask16)-1,                   \
-                                          _MM_FROUND_CUR_DIRECTION); })
-
-#define _mm512_cmp_pd_mask(a, b, p) __extension__ ({ \
-  __m512 __a = (a); \
-  __m512 __b = (b); \
-  (__mmask8)__builtin_ia32_cmppd512_mask((__v8df)__a, (__v8df)__b, (p), \
-                                          (__mmask8)-1,                 \
-                                          _MM_FROUND_CUR_DIRECTION); })
+#define _mm512_cmp_round_ps_mask(A, B, P, R) __extension__ ({ \
+  (__mmask16)__builtin_ia32_cmpps512_mask((__v16sf)(__m512)(A), \
+                                          (__v16sf)(__m512)(B), \
+                                          (P), (__mmask16)-1, (R)); })
+
+#define _mm512_mask_cmp_round_ps_mask(U, A, B, P, R) __extension__ ({ \
+  (__mmask16)__builtin_ia32_cmpps512_mask((__v16sf)(__m512)(A), \
+                                          (__v16sf)(__m512)(B), \
+                                          (P), (__mmask16)(U), (R)); })
+
+#define _mm512_cmp_ps_mask(A, B, P) \
+  _mm512_cmp_round_ps_mask((A), (B), (P), _MM_FROUND_CUR_DIRECTION)
+
+#define _mm512_mask_cmp_ps_mask(U, A, B, P) \
+  _mm512_mask_cmp_round_ps_mask((U), (A), (B), (P), _MM_FROUND_CUR_DIRECTION)
+
+#define _mm512_cmp_round_pd_mask(A, B, P, R) __extension__ ({ \
+  (__mmask8)__builtin_ia32_cmppd512_mask((__v8df)(__m512d)(A), \
+                                         (__v8df)(__m512d)(B), \
+                                         (P), (__mmask8)-1, (R)); })
+
+#define _mm512_mask_cmp_round_pd_mask(U, A, B, P, R) __extension__ ({ \
+  (__mmask8)__builtin_ia32_cmppd512_mask((__v8df)(__m512d)(A), \
+                                         (__v8df)(__m512d)(B), \
+                                         (P), (__mmask8)(U), (R)); })
+
+#define _mm512_cmp_pd_mask(A, B, P) \
+  _mm512_cmp_round_pd_mask((A), (B), (P), _MM_FROUND_CUR_DIRECTION)
+
+#define _mm512_mask_cmp_pd_mask(U, A, B, P) \
+  _mm512_mask_cmp_round_pd_mask((U), (A), (B), (P), _MM_FROUND_CUR_DIRECTION)
 
 /* Conversion */
 
diff --git a/clang/test/CodeGen/avx512f-builtins.c b/clang/test/CodeGen/avx512f-builtins.c
index 1e22da8ba5f..f7f7df54174 100644
--- a/clang/test/CodeGen/avx512f-builtins.c
+++ b/clang/test/CodeGen/avx512f-builtins.c
@@ -284,16 +284,52 @@ __m512 test_mm512_unpacklo_ps(__m512 a, __m512 b)
   return _mm512_unpacklo_ps(a, b);
 }
 
-__mmask16 test_mm512_cmp_ps_mask(__m512 __a, __m512 __b) {
-  // CHECK-LABEL: @test_mm512_cmp_ps_mask
+__mmask16 test_mm512_cmp_round_ps_mask(__m512 a, __m512 b) {
+  // CHECK-LABEL: @test_mm512_cmp_round_ps_mask
   // CHECK: @llvm.x86.avx512.mask.cmp.ps.512
-  return _mm512_cmp_ps_mask(__a, __b, 0);
+  return _mm512_cmp_round_ps_mask(a, b, 0, _MM_FROUND_TO_NEAREST_INT);
 }
 
-__mmask8 test_mm512_cmp_pd_mask(__m512 __a, __m512 __b) {
-  // CHECK-LABEL: @test_mm512_cmp_pd_mask
+__mmask16 test_mm512_mask_cmp_round_ps_mask(__mmask16 m, __m512 a, __m512 b) {
+  // CHECK-LABEL: @test_mm512_mask_cmp_round_ps_mask
+  // CHECK: @llvm.x86.avx512.mask.cmp.ps.512
+  return _mm512_mask_cmp_round_ps_mask(m, a, b, 0, _MM_FROUND_TO_NEAREST_INT);
+}
+
+__mmask16 test_mm512_cmp_ps_mask(__m512 a, __m512 b) {
+  // check-label: @test_mm512_cmp_ps_mask
+  // check: @llvm.x86.avx512.mask.cmp.ps.512
+  return _mm512_cmp_ps_mask(a, b, 0);
+}
+
+__mmask16 test_mm512_mask_cmp_ps_mask(__mmask16 m, __m512 a, __m512 b) {
+  // CHECK-LABEL: @test_mm512_mask_cmp_ps_mask
+  // CHECK: @llvm.x86.avx512.mask.cmp.ps.512
+  return _mm512_mask_cmp_ps_mask(m, a, b, 0);
+}
+
+__mmask8 test_mm512_cmp_round_pd_mask(__m512d a, __m512d b) {
+  // CHECK-LABEL: @test_mm512_cmp_round_pd_mask
+  // CHECK: @llvm.x86.avx512.mask.cmp.pd.512
+  return _mm512_cmp_round_pd_mask(a, b, 0, _MM_FROUND_TO_NEAREST_INT);
+}
+
+__mmask8 test_mm512_mask_cmp_round_pd_mask(__mmask8 m, __m512d a, __m512d b) {
+  // CHECK-LABEL: @test_mm512_mask_cmp_round_pd_mask
+  // CHECK: @llvm.x86.avx512.mask.cmp.pd.512
+  return _mm512_mask_cmp_round_pd_mask(m, a, b, 0, _MM_FROUND_TO_NEAREST_INT);
+}
+
+__mmask8 test_mm512_cmp_pd_mask(__m512d a, __m512d b) {
+  // check-label: @test_mm512_cmp_pd_mask
+  // check: @llvm.x86.avx512.mask.cmp.pd.512
+  return _mm512_cmp_pd_mask(a, b, 0);
+}
+
+__mmask8 test_mm512_mask_cmp_pd_mask(__mmask8 m, __m512d a, __m512d b) {
+  // CHECK-LABEL: @test_mm512_mask_cmp_pd_mask
   // CHECK: @llvm.x86.avx512.mask.cmp.pd.512
-  return _mm512_cmp_pd_mask(__a, __b, 0);
+  return _mm512_mask_cmp_pd_mask(m, a, b, 0);
 }
 
 __m256d test_mm512_extractf64x4_pd(__m512d a)