From: Manman Ren <manman.ren@gmail.com>
Date: Wed, 23 Oct 2013 20:33:14 +0000 (+0000)
Subject: Intrinsics: fix extract & insert when index is out of bound.
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=8484375b0fc442c704b3ec06e2e795d88591606f;p=clang

Intrinsics: fix extract & insert when index is out of bound.

Now, all extract & insert intrinsics should have the correct and operation
to ignore higher bits.

rdar://15250497


git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@193267 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Headers/avxintrin.h b/lib/Headers/avxintrin.h
index 50454f2653..141c4d994b 100644
--- a/lib/Headers/avxintrin.h
+++ b/lib/Headers/avxintrin.h
@@ -435,21 +435,21 @@ static __inline int __attribute__((__always_inline__, __nodebug__))
 _mm256_extract_epi32(__m256i __a, int const __imm)
 {
   __v8si __b = (__v8si)__a;
-  return __b[__imm];
+  return __b[__imm & 7];
 }
 
 static __inline int __attribute__((__always_inline__, __nodebug__))
 _mm256_extract_epi16(__m256i __a, int const __imm)
 {
   __v16hi __b = (__v16hi)__a;
-  return __b[__imm];
+  return __b[__imm & 15];
 }
 
 static __inline int __attribute__((__always_inline__, __nodebug__))
 _mm256_extract_epi8(__m256i __a, int const __imm)
 {
   __v32qi __b = (__v32qi)__a;
-  return __b[__imm];
+  return __b[__imm & 31];
 }
 
 #ifdef __x86_64__
@@ -457,7 +457,7 @@ static __inline long long  __attribute__((__always_inline__, __nodebug__))
 _mm256_extract_epi64(__m256i __a, const int __imm)
 {
   __v4di __b = (__v4di)__a;
-  return __b[__imm];
+  return __b[__imm & 3];
 }
 #endif
 
diff --git a/lib/Headers/smmintrin.h b/lib/Headers/smmintrin.h
index 5b6db6c445..53b3ccb431 100644
--- a/lib/Headers/smmintrin.h
+++ b/lib/Headers/smmintrin.h
@@ -197,7 +197,7 @@ _mm_max_epu32 (__m128i __V1, __m128i __V2)
 #define _mm_extract_ps(X, N) (__extension__                      \
                               ({ union { int __i; float __f; } __t;  \
                                  __v4sf __a = (__v4sf)(X);       \
-                                 __t.__f = __a[N];                 \
+                                 __t.__f = __a[(N) & 3];                 \
                                  __t.__i;}))
 
 /* Miscellaneous insert and extract macros.  */
@@ -215,14 +215,14 @@ _mm_max_epu32 (__m128i __V1, __m128i __V2)
                                              
 /* Insert int into packed integer array at index.  */
 #define _mm_insert_epi8(X, I, N) (__extension__ ({ __v16qi __a = (__v16qi)(X); \
-                                                   __a[(N)] = (I);             \
+                                                   __a[(N) & 15] = (I);             \
                                                    __a;}))
 #define _mm_insert_epi32(X, I, N) (__extension__ ({ __v4si __a = (__v4si)(X); \
-                                                    __a[(N)] = (I);           \
+                                                    __a[(N) & 3] = (I);           \
                                                     __a;}))
 #ifdef __x86_64__
 #define _mm_insert_epi64(X, I, N) (__extension__ ({ __v2di __a = (__v2di)(X); \
-                                                    __a[(N)] = (I);           \
+                                                    __a[(N) & 1] = (I);           \
                                                     __a;}))
 #endif /* __x86_64__ */
 
@@ -231,12 +231,12 @@ _mm_max_epu32 (__m128i __V1, __m128i __V2)
  */
 #define _mm_extract_epi8(X, N) (__extension__ ({ __v16qi __a = (__v16qi)(X); \
                                                  (int)(unsigned char) \
-                                                     __a[(N)];}))
+                                                     __a[(N) & 15];}))
 #define _mm_extract_epi32(X, N) (__extension__ ({ __v4si __a = (__v4si)(X); \
-                                                  __a[(N)];}))
+                                                  __a[(N) & 3];}))
 #ifdef __x86_64__
 #define _mm_extract_epi64(X, N) (__extension__ ({ __v2di __a = (__v2di)(X); \
-                                                  __a[(N)];}))
+                                                  __a[(N) & 1];}))
 #endif /* __x86_64 */
 
 /* SSE4 128-bit Packed Integer Comparisons.  */
diff --git a/test/CodeGen/avx-builtins.c b/test/CodeGen/avx-builtins.c
index 0e5a741bcf..c88946fe85 100644
--- a/test/CodeGen/avx-builtins.c
+++ b/test/CodeGen/avx-builtins.c
@@ -93,3 +93,21 @@ int test_mm_cmpistrz(__m128i A, __m128i B) {
   // CHECK: @llvm.x86.sse42.pcmpistriz128
   return _mm_cmpistrz(A, B, 7);
 }
+
+int test_extract_epi32(__m256i __a) {
+  // CHECK-LABEL: @test_extract_epi32
+  // CHECK: extractelement <8 x i32> %{{.*}}, i32 0
+  return _mm256_extract_epi32(__a, 8);
+}
+
+int test_extract_epi16(__m256i __a) {
+  // CHECK-LABEL: @test_extract_epi16
+  // CHECK: extractelement <16 x i16> %{{.*}}, i32 0
+  return _mm256_extract_epi16(__a, 16);
+}
+
+int test_extract_epi8(__m256i __a) {
+  // CHECK-LABEL: @test_extract_epi8
+  // CHECK: extractelement <32 x i8> %{{.*}}, i32 0
+  return _mm256_extract_epi8(__a, 32);
+}
diff --git a/test/CodeGen/sse-builtins.c b/test/CodeGen/sse-builtins.c
index cee9b3c2a5..1f5cb8e9e1 100644
--- a/test/CodeGen/sse-builtins.c
+++ b/test/CodeGen/sse-builtins.c
@@ -213,3 +213,27 @@ void test_extract_epi16(__m128i __a) {
   // CHECK: extractelement <8 x i16> %{{.*}}, i32 [[x]]
   _mm_extract_epi16(__a, 8);
 }
+
+int test_extract_ps(__m128i __a) {
+  // CHECK-LABEL: @test_extract_ps
+  // CHECK: extractelement <4 x float> %{{.*}}, i32 0
+  return _mm_extract_ps(__a, 4);
+}
+
+int test_extract_epi8(__m128i __a) {
+  // CHECK-LABEL: @test_extract_epi8
+  // CHECK: extractelement <16 x i8> %{{.*}}, i32 0
+  return _mm_extract_epi8(__a, 16);
+}
+
+int test_extract_epi32(__m128i __a) {
+  // CHECK-LABEL: @test_extract_epi32
+  // CHECK: extractelement <4 x i32> %{{.*}}, i32 0
+  return _mm_extract_epi32(__a, 4);
+}
+
+void test_insert_epi32(__m128i __a, int b) {
+  // CHECK-LABEL: @test_insert_epi32
+  // CHECK: insertelement <4 x i32> %{{.*}}, i32 %{{.*}}, i32 0
+   _mm_insert_epi32(__a, b, 4);
+}