From: Craig Topper <craig.topper@intel.com>
Date: Sat, 24 Feb 2018 06:46:42 +0000 (+0000)
Subject: [X86] Remove __builtin_ia32_permvarsf256_mask and __builtin_ia32_permvarsi256_mask... 
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=8d120f1d0c2667d1aebb32d128ad0cbe4d6484ad;p=clang

[X86] Remove __builtin_ia32_permvarsf256_mask and __builtin_ia32_permvarsi256_mask and use the avx2 unmasked versions and a select instead.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@326022 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/include/clang/Basic/BuiltinsX86.def b/include/clang/Basic/BuiltinsX86.def
index 74c0c039c9..bb53628e91 100644
--- a/include/clang/Basic/BuiltinsX86.def
+++ b/include/clang/Basic/BuiltinsX86.def
@@ -1799,8 +1799,6 @@ TARGET_BUILTIN(__builtin_ia32_permvarhi128_mask, "V8sV8sV8sV8sUc","","avx512bw,a
 TARGET_BUILTIN(__builtin_ia32_permvarhi256_mask, "V16sV16sV16sV16sUs","","avx512bw,avx512vl")
 TARGET_BUILTIN(__builtin_ia32_permvardf256_mask, "V4dV4dV4LLiV4dUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_permvardi256_mask, "V4LLiV4LLiV4LLiV4LLiUc","","avx512vl")
-TARGET_BUILTIN(__builtin_ia32_permvarsf256_mask, "V8fV8fV8iV8fUc","","avx512vl")
-TARGET_BUILTIN(__builtin_ia32_permvarsi256_mask, "V8iV8iV8iV8iUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_fpclasspd128_mask, "UcV2dIiUc","","avx512dq,avx512vl")
 TARGET_BUILTIN(__builtin_ia32_fpclasspd256_mask, "UcV4dIiUc","","avx512dq,avx512vl")
 TARGET_BUILTIN(__builtin_ia32_fpclassps128_mask, "UcV4fIiUc","","avx512dq,avx512vl")
diff --git a/lib/Headers/avx512vlintrin.h b/lib/Headers/avx512vlintrin.h
index fb8056e3f8..e6d4cea56d 100644
--- a/lib/Headers/avx512vlintrin.h
+++ b/lib/Headers/avx512vlintrin.h
@@ -8178,60 +8178,41 @@ _mm256_mask_permutexvar_epi64 (__m256i __W, __mmask8 __M, __m256i __X,
                  __M);
 }
 
-static __inline__ __m256 __DEFAULT_FN_ATTRS
-_mm256_mask_permutexvar_ps (__m256 __W, __mmask8 __U, __m256i __X,
-          __m256 __Y)
-{
-  return (__m256) __builtin_ia32_permvarsf256_mask ((__v8sf) __Y,
-                (__v8si) __X,
-                (__v8sf) __W,
-                (__mmask8) __U);
-}
+#define _mm256_permutexvar_ps(A, B) _mm256_permutevar8x32_ps((B), (A))
 
 static __inline__ __m256 __DEFAULT_FN_ATTRS
-_mm256_maskz_permutexvar_ps (__mmask8 __U, __m256i __X, __m256 __Y)
+_mm256_mask_permutexvar_ps(__m256 __W, __mmask8 __U, __m256i __X, __m256 __Y)
 {
-  return (__m256) __builtin_ia32_permvarsf256_mask ((__v8sf) __Y,
-                (__v8si) __X,
-                (__v8sf) _mm256_setzero_ps (),
-                (__mmask8) __U);
+  return (__m256)__builtin_ia32_selectps_256((__mmask8)__U,
+                                        (__v8sf)_mm256_permutexvar_ps(__X, __Y),
+                                        (__v8sf)__W);
 }
 
 static __inline__ __m256 __DEFAULT_FN_ATTRS
-_mm256_permutexvar_ps (__m256i __X, __m256 __Y)
+_mm256_maskz_permutexvar_ps(__mmask8 __U, __m256i __X, __m256 __Y)
 {
-  return (__m256) __builtin_ia32_permvarsf256_mask ((__v8sf) __Y,
-                (__v8si) __X,
-                (__v8sf) _mm256_undefined_si256 (),
-                (__mmask8) -1);
+  return (__m256)__builtin_ia32_selectps_256((__mmask8)__U,
+                                        (__v8sf)_mm256_permutexvar_ps(__X, __Y),
+                                        (__v8sf)_mm256_setzero_ps());
 }
 
-static __inline__ __m256i __DEFAULT_FN_ATTRS
-_mm256_maskz_permutexvar_epi32 (__mmask8 __M, __m256i __X, __m256i __Y)
-{
-  return (__m256i) __builtin_ia32_permvarsi256_mask ((__v8si) __Y,
-                 (__v8si) __X,
-                 (__v8si) _mm256_setzero_si256 (),
-                 __M);
-}
+#define _mm256_permutexvar_epi32(A, B) _mm256_permutevar8x32_epi32((B), (A))
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS
-_mm256_mask_permutexvar_epi32 (__m256i __W, __mmask8 __M, __m256i __X,
-             __m256i __Y)
+_mm256_mask_permutexvar_epi32(__m256i __W, __mmask8 __M, __m256i __X,
+                              __m256i __Y)
 {
-  return (__m256i) __builtin_ia32_permvarsi256_mask ((__v8si) __Y,
-                 (__v8si) __X,
-                 (__v8si) __W,
-                 (__mmask8) __M);
+  return (__m256i)__builtin_ia32_selectd_256((__mmask8)__M,
+                                     (__v8si)_mm256_permutexvar_epi32(__X, __Y),
+                                     (__v8si)__W);
 }
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS
-_mm256_permutexvar_epi32 (__m256i __X, __m256i __Y)
+_mm256_maskz_permutexvar_epi32(__mmask8 __M, __m256i __X, __m256i __Y)
 {
-  return (__m256i) __builtin_ia32_permvarsi256_mask ((__v8si) __Y,
-                 (__v8si) __X,
-                 (__v8si) _mm256_undefined_si256(),
-                 (__mmask8) -1);
+  return (__m256i)__builtin_ia32_selectd_256((__mmask8)__M,
+                                     (__v8si)_mm256_permutexvar_epi32(__X, __Y),
+                                     (__v8si)_mm256_setzero_si256());
 }
 
 #define _mm_alignr_epi32(A, B, imm) __extension__ ({ \
diff --git a/test/CodeGen/avx512vl-builtins.c b/test/CodeGen/avx512vl-builtins.c
index 008eb09b8b..00a0813801 100644
--- a/test/CodeGen/avx512vl-builtins.c
+++ b/test/CodeGen/avx512vl-builtins.c
@@ -7019,37 +7019,40 @@ __m256i test_mm256_mask_permutexvar_epi64(__m256i __W, __mmask8 __M, __m256i __X
 
 __m256 test_mm256_mask_permutexvar_ps(__m256 __W, __mmask8 __U, __m256i __X, __m256 __Y) {
   // CHECK-LABEL: @test_mm256_mask_permutexvar_ps
-  // CHECK: @llvm.x86.avx512.mask.permvar.sf.256
+  // CHECK: @llvm.x86.avx2.permps
+  // CHECK: select <8 x i1> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}}
   return _mm256_mask_permutexvar_ps(__W, __U, __X, __Y);
 }
 
 __m256 test_mm256_maskz_permutexvar_ps(__mmask8 __U, __m256i __X, __m256 __Y) {
   // CHECK-LABEL: @test_mm256_maskz_permutexvar_ps
-  // CHECK: @llvm.x86.avx512.mask.permvar.sf.256
+  // CHECK: select <8 x i1> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}}
   return _mm256_maskz_permutexvar_ps(__U, __X, __Y);
 }
 
 __m256 test_mm256_permutexvar_ps(__m256i __X, __m256 __Y) {
   // CHECK-LABEL: @test_mm256_permutexvar_ps
-  // CHECK: @llvm.x86.avx512.mask.permvar.sf.256
+  // CHECK: @llvm.x86.avx2.permps
   return _mm256_permutexvar_ps( __X, __Y);
 }
 
 __m256i test_mm256_maskz_permutexvar_epi32(__mmask8 __M, __m256i __X, __m256i __Y) {
   // CHECK-LABEL: @test_mm256_maskz_permutexvar_epi32
-  // CHECK: @llvm.x86.avx512.mask.permvar.si.256
+  // CHECK: @llvm.x86.avx2.permd
+  // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}}
   return _mm256_maskz_permutexvar_epi32(__M, __X, __Y);
 }
 
 __m256i test_mm256_permutexvar_epi32(__m256i __X, __m256i __Y) {
   // CHECK-LABEL: @test_mm256_permutexvar_epi32
-  // CHECK: @llvm.x86.avx512.mask.permvar.si.256
+  // CHECK: @llvm.x86.avx2.permd
   return _mm256_permutexvar_epi32(__X, __Y);
 }
 
 __m256i test_mm256_mask_permutexvar_epi32(__m256i __W, __mmask8 __M, __m256i __X, __m256i __Y) {
   // CHECK-LABEL: @test_mm256_mask_permutexvar_epi32
-  // CHECK: @llvm.x86.avx512.mask.permvar.si.256
+  // CHECK: @llvm.x86.avx2.permd
+  // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}}
   return _mm256_mask_permutexvar_epi32(__W, __M, __X, __Y);
 }