[x86][AVX512CD] Add conflict and lzcnt intrinsics in their 512bit versions

include tests

review
http://reviews.llvm.org/D10795

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@240941 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/include/clang/Basic/BuiltinsX86.def b/include/clang/Basic/BuiltinsX86.def
index 0873cb3d709c18d1e83cbe6a1f75c3a641a7413a..a0b5218d78c403b8f26886895824ee1e6f81e151 100644 (file)
--- a/include/clang/Basic/BuiltinsX86.def
+++ b/include/clang/Basic/BuiltinsX86.def
@@ -1142,5 +1142,9 @@ BUILTIN(__builtin_ia32_psubusw512_mask, "V32sV32sV32sV32sUi", "")
 BUILTIN(__builtin_ia32_vpermi2varhi512_mask, "V32sV32sV32sV32sUi", "")
 BUILTIN(__builtin_ia32_vpermt2varhi512_mask, "V32sV32sV32sV32sUi", "")
 BUILTIN(__builtin_ia32_vpermt2varhi512_maskz, "V32sV32sV32sV32sUi", "")
+BUILTIN(__builtin_ia32_vpconflictdi_512_mask, "V8LLiV8LLiV8LLiUc", "")
+BUILTIN(__builtin_ia32_vpconflictsi_512_mask, "V16iV16iV16iUs", "")
+BUILTIN(__builtin_ia32_vplzcntd_512_mask, "V16iV16iV16iUs", "")
+BUILTIN(__builtin_ia32_vplzcntq_512_mask, "V8LLiV8LLiV8LLiUc", "")
 
 #undef BUILTIN

diff --git a/lib/Headers/CMakeLists.txt b/lib/Headers/CMakeLists.txt
index 6f3ce52613c827f0ee21f9a6a60e55aed8cca78e..7e95113564c8078e2e0d26769bc8764cf28f131c 100644 (file)
--- a/lib/Headers/CMakeLists.txt
+++ b/lib/Headers/CMakeLists.txt
@@ -5,6 +5,7 @@ set(files
   arm_acle.h
   avx2intrin.h
   avx512bwintrin.h
+  avx512cdintrin.h
   avx512erintrin.h
   avx512fintrin.h
   avx512vlbwintrin.h

diff --git a/lib/Headers/avx512cdintrin.h b/lib/Headers/avx512cdintrin.h
new file mode 100644 (file)
index 0000000..f7a7970
--- /dev/null
+++ b/lib/Headers/avx512cdintrin.h
@@ -0,0 +1,131 @@
+/*===------------- avx512cdintrin.h - AVX512CD intrinsics ------------------===\r
+ *\r
+ *\r
+ * Permission is hereby granted, free of charge, to any person obtaining a copy\r
+ * of this software and associated documentation files (the "Software"), to deal\r
+ * in the Software without restriction, including without limitation the rights\r
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell\r
+ * copies of the Software, and to permit persons to whom the Software is\r
+ * furnished to do so, subject to the following conditions:\r
+ *\r
+ * The above copyright notice and this permission notice shall be included in\r
+ * all copies or substantial portions of the Software.\r
+ *\r
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR\r
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,\r
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE\r
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER\r
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,\r
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN\r
+ * THE SOFTWARE.\r
+ *\r
+ *===-----------------------------------------------------------------------===\r
+ */\r
+#ifndef __IMMINTRIN_H\r
+#error "Never use <avx512cdintrin.h> directly; include <immintrin.h> instead."\r
+#endif\r
+\r
+#ifndef __AVX512CDINTRIN_H\r
+#define __AVX512CDINTRIN_H\r
+\r
+/* Define the default attributes for the functions in this file. */\r
+#define DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("avx512cd")))\r
+\r
+static __inline__ __m512i DEFAULT_FN_ATTRS\r
+_mm512_conflict_epi64 (__m512i __A)\r
+{\r
+  return (__m512i) __builtin_ia32_vpconflictdi_512_mask ((__v8di) __A,\r
+                 (__v8di) _mm512_setzero_si512 (),\r
+                 (__mmask8) -1);\r
+}\r
+\r
+static __inline__ __m512i DEFAULT_FN_ATTRS\r
+_mm512_mask_conflict_epi64 (__m512i __W, __mmask8 __U, __m512i __A)\r
+{\r
+  return (__m512i) __builtin_ia32_vpconflictdi_512_mask ((__v8di) __A,\r
+               (__v8di) __W,\r
+               (__mmask8) __U);\r
+}\r
+\r
+static __inline__ __m512i DEFAULT_FN_ATTRS\r
+_mm512_maskz_conflict_epi64 (__mmask8 __U, __m512i __A)\r
+{\r
+  return (__m512i) __builtin_ia32_vpconflictdi_512_mask ((__v8di) __A,\r
+                 (__v8di) _mm512_setzero_si512 (),\r
+                 (__mmask8) __U);\r
+}\r
+\r
+static __inline__ __m512i DEFAULT_FN_ATTRS\r
+_mm512_conflict_epi32 (__m512i __A)\r
+{\r
+  return (__m512i) __builtin_ia32_vpconflictsi_512_mask ((__v16si) __A,\r
+                 (__v16si) _mm512_setzero_si512 (),\r
+                 (__mmask16) -1);\r
+}\r
+\r
+static __inline__ __m512i DEFAULT_FN_ATTRS\r
+_mm512_mask_conflict_epi32 (__m512i __W, __mmask16 __U, __m512i __A)\r
+{\r
+  return (__m512i) __builtin_ia32_vpconflictsi_512_mask ((__v16si) __A,\r
+               (__v16si) __W,\r
+               (__mmask16) __U);\r
+}\r
+\r
+static __inline__ __m512i DEFAULT_FN_ATTRS\r
+_mm512_maskz_conflict_epi32 (__mmask16 __U, __m512i __A)\r
+{\r
+  return (__m512i) __builtin_ia32_vpconflictsi_512_mask ((__v16si) __A,\r
+                 (__v16si) _mm512_setzero_si512 (),\r
+                 (__mmask16) __U);\r
+}\r
+\r
+static __inline__ __m512i DEFAULT_FN_ATTRS\r
+_mm512_lzcnt_epi32 (__m512i __A)\r
+{\r
+  return (__m512i) __builtin_ia32_vplzcntd_512_mask ((__v16si) __A,\r
+             (__v16si) _mm512_setzero_si512 (),\r
+             (__mmask16) -1);\r
+}\r
+\r
+static __inline__ __m512i DEFAULT_FN_ATTRS\r
+_mm512_mask_lzcnt_epi32 (__m512i __W, __mmask16 __U, __m512i __A)\r
+{\r
+  return (__m512i) __builtin_ia32_vplzcntd_512_mask ((__v16si) __A,\r
+                 (__v16si) __W,\r
+                 (__mmask16) __U);\r
+}\r
+\r
+static __inline__ __m512i DEFAULT_FN_ATTRS\r
+_mm512_maskz_lzcnt_epi32 (__mmask16 __U, __m512i __A)\r
+{\r
+  return (__m512i) __builtin_ia32_vplzcntd_512_mask ((__v16si) __A,\r
+             (__v16si) _mm512_setzero_si512 (),\r
+             (__mmask16) __U);\r
+}\r
+\r
+static __inline__ __m512i DEFAULT_FN_ATTRS\r
+_mm512_lzcnt_epi64 (__m512i __A)\r
+{\r
+  return (__m512i) __builtin_ia32_vplzcntq_512_mask ((__v8di) __A,\r
+             (__v8di) _mm512_setzero_si512 (),\r
+             (__mmask8) -1);\r
+}\r
+\r
+static __inline__ __m512i DEFAULT_FN_ATTRS\r
+_mm512_mask_lzcnt_epi64 (__m512i __W, __mmask8 __U, __m512i __A)\r
+{\r
+  return (__m512i) __builtin_ia32_vplzcntq_512_mask ((__v8di) __A,\r
+                 (__v8di) __W,\r
+                 (__mmask8) __U);\r
+}\r
+\r
+static __inline__ __m512i DEFAULT_FN_ATTRS\r
+_mm512_maskz_lzcnt_epi64 (__mmask8 __U, __m512i __A)\r
+{\r
+  return (__m512i) __builtin_ia32_vplzcntq_512_mask ((__v8di) __A,\r
+             (__v8di) _mm512_setzero_si512 (),\r
+             (__mmask8) __U);\r
+}\r
+#undef DEFAULT_FN_ATTRS\r
+\r
+#endif\r

diff --git a/lib/Headers/immintrin.h b/lib/Headers/immintrin.h
index 4af407dfc7e87d9a3d57fe9268944ff8707a0712..4d2638a11e250340363a35a16350d2904e461b84 100644 (file)
--- a/lib/Headers/immintrin.h
+++ b/lib/Headers/immintrin.h
@@ -56,6 +56,8 @@
 
 #include <avx512bwintrin.h>
 
+#include <avx512cdintrin.h>
+
 #include <avx512dqintrin.h>
 
 #include <avx512vlbwintrin.h>

diff --git a/test/CodeGen/avx512cdintrin.c b/test/CodeGen/avx512cdintrin.c
new file mode 100644 (file)
index 0000000..2b31cef
--- /dev/null
+++ b/test/CodeGen/avx512cdintrin.c
@@ -0,0 +1,62 @@
+// RUN: %clang_cc1 %s -O0 -triple=x86_64-apple-darwin -ffreestanding -target-feature +avx512cd -emit-llvm -o - -Werror | FileCheck %s\r
+#include <immintrin.h>\r
+__m512i test_mm512_conflict_epi64(__m512i __A) {\r
+  // CHECK-LABEL: @test_mm512_conflict_epi64\r
+  // CHECK: @llvm.x86.avx512.mask.conflict.q.512\r
+  return _mm512_conflict_epi64(__A); \r
+}\r
+__m512i test_mm512_mask_conflict_epi64(__m512i __W, __mmask8 __U, __m512i __A) {\r
+  // CHECK-LABEL: @test_mm512_mask_conflict_epi64\r
+  // CHECK: @llvm.x86.avx512.mask.conflict.q.512\r
+  return _mm512_mask_conflict_epi64(__W,__U,__A); \r
+}\r
+__m512i test_mm512_maskz_conflict_epi64(__mmask8 __U, __m512i __A) {\r
+  // CHECK-LABEL: @test_mm512_maskz_conflict_epi64\r
+  // CHECK: @llvm.x86.avx512.mask.conflict.q.512\r
+  return _mm512_maskz_conflict_epi64(__U,__A); \r
+}\r
+__m512i test_mm512_conflict_epi32(__m512i __A) {\r
+  // CHECK-LABEL: @test_mm512_conflict_epi32\r
+  // CHECK: @llvm.x86.avx512.mask.conflict.d.512\r
+  return _mm512_conflict_epi32(__A); \r
+}\r
+__m512i test_mm512_mask_conflict_epi32(__m512i __W, __mmask16 __U, __m512i __A) {\r
+  // CHECK-LABEL: @test_mm512_mask_conflict_epi32\r
+  // CHECK: @llvm.x86.avx512.mask.conflict.d.512\r
+  return _mm512_mask_conflict_epi32(__W,__U,__A); \r
+}\r
+__m512i test_mm512_maskz_conflict_epi32(__mmask16 __U, __m512i __A) {\r
+  // CHECK-LABEL: @test_mm512_maskz_conflict_epi32\r
+  // CHECK: @llvm.x86.avx512.mask.conflict.d.512\r
+  return _mm512_maskz_conflict_epi32(__U,__A); \r
+}\r
+__m512i test_mm512_lzcnt_epi32(__m512i __A) {\r
+  // CHECK-LABEL: @test_mm512_lzcnt_epi32\r
+  // CHECK: @llvm.x86.avx512.mask.lzcnt.d.512\r
+  return _mm512_lzcnt_epi32(__A); \r
+}\r
+__m512i test_mm512_mask_lzcnt_epi32(__m512i __W, __mmask16 __U, __m512i __A) {\r
+  // CHECK-LABEL: @test_mm512_mask_lzcnt_epi32\r
+  // CHECK: @llvm.x86.avx512.mask.lzcnt.d.512\r
+  return _mm512_mask_lzcnt_epi32(__W,__U,__A); \r
+}\r
+__m512i test_mm512_maskz_lzcnt_epi32(__mmask16 __U, __m512i __A) {\r
+  // CHECK-LABEL: @test_mm512_maskz_lzcnt_epi32\r
+  // CHECK: @llvm.x86.avx512.mask.lzcnt.d.512\r
+  return _mm512_maskz_lzcnt_epi32(__U,__A); \r
+}\r
+__m512i test_mm512_lzcnt_epi64(__m512i __A) {\r
+  // CHECK-LABEL: @test_mm512_lzcnt_epi64\r
+  // CHECK: @llvm.x86.avx512.mask.lzcnt.q.512\r
+  return _mm512_lzcnt_epi64(__A); \r
+}\r
+__m512i test_mm512_mask_lzcnt_epi64(__m512i __W, __mmask8 __U, __m512i __A) {\r
+  // CHECK-LABEL: @test_mm512_mask_lzcnt_epi64\r
+  // CHECK: @llvm.x86.avx512.mask.lzcnt.q.512\r
+  return _mm512_mask_lzcnt_epi64(__W,__U,__A); \r
+}\r
+__m512i test_mm512_maskz_lzcnt_epi64(__mmask8 __U, __m512i __A) {\r
+  // CHECK-LABEL: @test_mm512_maskz_lzcnt_epi64\r
+  // CHECK: @llvm.x86.avx512.mask.lzcnt.q.512\r
+  return _mm512_maskz_lzcnt_epi64(__U,__A); \r
+}\r