[X86] Add builtins for vector element insert and extract for different 128 and 256 bit vector types. Use them to implement the extract and insert intrinsics.

Previously we were just using extended vector operations in the header file.

This unfortunately allowed non-constant indices to be used with the intrinsics. This is incompatible with gcc, icc, and MSVC. It also introduces a different performance characteristic because non-constant index gets lowered to a vector store and an element sized load.

By adding the builtins we can check for the index to be a constant and ensure its in range of the vector element count.

User code still has the option to use extended vector operations themselves if they need non-constant indexing.

llvm-svn: 334057

5 files changed, 25 insertions, 35 deletions

diff --git a/clang/test/CodeGen/avx-builtins.c b/clang/test/CodeGen/avx-builtins.c
index 4c904ee71fc..43c502f47f1 100644
--- a/clang/test/CodeGen/avx-builtins.c
+++ b/clang/test/CodeGen/avx-builtins.c
@@ -316,32 +316,28 @@ __m256 test_mm256_dp_ps(__m256 A, __m256 B) {
 
 int test_mm256_extract_epi8(__m256i A) {
   // CHECK-LABEL: test_mm256_extract_epi8
-  // CHECK: and i32 %{{.*}}, 31
-  // CHECK: extractelement <32 x i8> %{{.*}}, i32 %{{.*}}
+  // CHECK: extractelement <32 x i8> %{{.*}}, i32 31
   // CHECK: zext i8 %{{.*}} to i32
-  return _mm256_extract_epi8(A, 32);
+  return _mm256_extract_epi8(A, 31);
 }
 
 int test_mm256_extract_epi16(__m256i A) {
   // CHECK-LABEL: test_mm256_extract_epi16
-  // CHECK: and i32 %{{.*}}, 15
-  // CHECK: extractelement <16 x i16> %{{.*}}, i32 %{{.*}}
+  // CHECK: extractelement <16 x i16> %{{.*}}, i32 15
   // CHECK: zext i16 %{{.*}} to i32
-  return _mm256_extract_epi16(A, 16);
+  return _mm256_extract_epi16(A, 15);
 }
 
 int test_mm256_extract_epi32(__m256i A) {
   // CHECK-LABEL: test_mm256_extract_epi32
-  // CHECK: and i32 %{{.*}}, 7
-  // CHECK: extractelement <8 x i32> %{{.*}}, i32 %{{.*}}
-  return _mm256_extract_epi32(A, 8);
+  // CHECK: extractelement <8 x i32> %{{.*}}, i32 7
+  return _mm256_extract_epi32(A, 7);
 }
 
 long long test_mm256_extract_epi64(__m256i A) {
   // CHECK-LABEL: test_mm256_extract_epi64
-  // CHECK: and i32 %{{.*}}, 3
-  // CHECK: extractelement <4 x i64> %{{.*}}, i32 %{{.*}}
-  return _mm256_extract_epi64(A, 5);
+  // CHECK: extractelement <4 x i64> %{{.*}}, i32 3
+  return _mm256_extract_epi64(A, 3);
 }
 
 __m128d test_mm256_extractf128_pd(__m256d A) {
@@ -400,29 +396,25 @@ __m256 test_mm256_hsub_ps(__m256 A, __m256 B) {
 
 __m256i test_mm256_insert_epi8(__m256i x, char b) {
   // CHECK-LABEL: test_mm256_insert_epi8
-  // CHECK: and i32 %{{.*}}, 31
-  // CHECK: insertelement <32 x i8> %{{.*}}, i8 %{{.*}}, i32 %{{.*}}
-  return _mm256_insert_epi8(x, b, 17);
+  // CHECK: insertelement <32 x i8> %{{.*}}, i8 %{{.*}}, i32 14
+  return _mm256_insert_epi8(x, b, 14);
 }
 
 __m256i test_mm256_insert_epi16(__m256i x, int b) {
   // CHECK-LABEL: test_mm256_insert_epi16
-  // CHECK: and i32 %{{.*}}, 15
-  // CHECK: insertelement <16 x i16> %{{.*}}, i16 %{{.*}}, i32 %{{.*}}
+  // CHECK: insertelement <16 x i16> %{{.*}}, i16 %{{.*}}, i32 4
   return _mm256_insert_epi16(x, b, 4);
 }
 
 __m256i test_mm256_insert_epi32(__m256i x, int b) {
   // CHECK-LABEL: test_mm256_insert_epi32
-  // CHECK: and i32 %{{.*}}, 7
-  // CHECK: insertelement <8 x i32> %{{.*}}, i32 %{{.*}}, i32 %{{.*}}
+  // CHECK: insertelement <8 x i32> %{{.*}}, i32 %{{.*}}, i32 5
   return _mm256_insert_epi32(x, b, 5);
 }
 
 __m256i test_mm256_insert_epi64(__m256i x, long long b) {
   // CHECK-LABEL: test_mm256_insert_epi64
-  // CHECK: and i32 %{{.*}}, 3
-  // CHECK: insertelement <4 x i64> %{{.*}}, i64 %{{.*}}, i32 %{{.*}}
+  // CHECK: insertelement <4 x i64> %{{.*}}, i64 %{{.*}}, i32 2
   return _mm256_insert_epi64(x, b, 2);
 }
 
diff --git a/clang/test/CodeGen/sse2-builtins.c b/clang/test/CodeGen/sse2-builtins.c
index f7094b2fbd0..c0a44795bf4 100644
--- a/clang/test/CodeGen/sse2-builtins.c
+++ b/clang/test/CodeGen/sse2-builtins.c
@@ -613,17 +613,15 @@ __m128d test_mm_div_sd(__m128d A, __m128d B) {
 // Lowering to pextrw requires optimization.
 int test_mm_extract_epi16(__m128i A) {
   // CHECK-LABEL: test_mm_extract_epi16
-  // CHECK: [[x:%.*]] = and i32 %{{.*}}, 7
-  // CHECK: extractelement <8 x i16> %{{.*}}, i32 [[x]]
+  // CHECK: extractelement <8 x i16> %{{.*}}, i32 1
   // CHECK: zext i16 %{{.*}} to i32
-  return _mm_extract_epi16(A, 9);
+  return _mm_extract_epi16(A, 1);
 }
 
 __m128i test_mm_insert_epi16(__m128i A, int B) {
   // CHECK-LABEL: test_mm_insert_epi16
-  // CHECK: [[x:%.*]] = and i32 %{{.*}}, 7
-  // CHECK: insertelement <8 x i16> %{{.*}}, i32 [[x]]
-  return _mm_insert_epi16(A, B, 8);
+  // CHECK: insertelement <8 x i16> %{{.*}}, i32 0
+  return _mm_insert_epi16(A, B, 0);
 }
 
 void test_mm_lfence() {
diff --git a/clang/test/CodeGen/sse41-builtins.c b/clang/test/CodeGen/sse41-builtins.c
index a99ecdabe44..6d6e11e9622 100644
--- a/clang/test/CodeGen/sse41-builtins.c
+++ b/clang/test/CodeGen/sse41-builtins.c
@@ -220,20 +220,20 @@ __m128 test_mm_floor_ss(__m128 x, __m128 y) {
 
 __m128i test_mm_insert_epi8(__m128i x, char b) {
   // CHECK-LABEL: test_mm_insert_epi8
-  // CHECK: insertelement <16 x i8> %{{.*}}, i8 %{{.*}}, i32 0
-  return _mm_insert_epi8(x, b, 16);
+  // CHECK: insertelement <16 x i8> %{{.*}}, i8 %{{.*}}, i32 1
+  return _mm_insert_epi8(x, b, 1);
 }
 
 __m128i test_mm_insert_epi32(__m128i x, int b) {
   // CHECK-LABEL: test_mm_insert_epi32
-  // CHECK: insertelement <4 x i32> %{{.*}}, i32 %{{.*}}, i32 0
-  return _mm_insert_epi32(x, b, 4);
+  // CHECK: insertelement <4 x i32> %{{.*}}, i32 %{{.*}}, i32 1
+  return _mm_insert_epi32(x, b, 1);
 }
 
 __m128i test_mm_insert_epi64(__m128i x, long long b) {
   // CHECK-LABEL: test_mm_insert_epi64
-  // CHECK: insertelement <2 x i64> %{{.*}}, i64 %{{.*}}, i32 0
-  return _mm_insert_epi64(x, b, 2);
+  // CHECK: insertelement <2 x i64> %{{.*}}, i64 %{{.*}}, i32 1
+  return _mm_insert_epi64(x, b, 1);
 }
 
 __m128 test_mm_insert_ps(__m128 x, __m128 y) {
diff --git a/clang/test/CodeGen/target-features-error-2.c b/clang/test/CodeGen/target-features-error-2.c
index 683d9ab99ef..40279fb6dea 100644
--- a/clang/test/CodeGen/target-features-error-2.c
+++ b/clang/test/CodeGen/target-features-error-2.c
@@ -9,7 +9,7 @@
 
 #if NEED_SSE42
 int baz(__m256i a) {
-  return _mm256_extract_epi32(a, 3); // expected-error {{always_inline function '_mm256_extract_epi32' requires target feature 'sse4.2', but would be inlined into function 'baz' that is compiled without support for 'sse4.2'}}
+  return _mm256_extract_epi32(a, 3); // expected-error {{'__builtin_ia32_vec_ext_v8si' needs target feature avx}}
 }
 #endif
 
diff --git a/clang/test/CodeGen/vector.c b/clang/test/CodeGen/vector.c
index ebaea841aa8..98dd82a28a0 100644
--- a/clang/test/CodeGen/vector.c
+++ b/clang/test/CodeGen/vector.c
@@ -1,4 +1,4 @@
-// RUN: %clang_cc1 -ffreestanding -triple i386-apple-darwin9 -O1 -target-cpu core2 -debug-info-kind=limited -emit-llvm %s -o - | FileCheck %s
+// RUN: %clang_cc1 -ffreestanding -triple i386-apple-darwin9 -O1 -target-cpu corei7 -debug-info-kind=limited -emit-llvm %s -o - | FileCheck %s
 typedef short __v4hi __attribute__ ((__vector_size__ (8)));
 
 void test1() {