[x86] Revert r330322 (& r330323): Lowering x86 adds/addus/subs/subus intrinsics

The LLVM commit introduces a crash in LLVM's instruction selection.

I filed http://llvm.org/PR37260 with the test case.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@330997 91177308-0d34-0410-b5e6-96231b3b80d8

4 files changed, 89 insertions, 558 deletions

diff --git a/test/CodeGen/avx2-builtins.c b/test/CodeGen/avx2-builtins.c
index e142eea586..15a17628f4 100644
--- a/test/CodeGen/avx2-builtins.c
+++ b/test/CodeGen/avx2-builtins.c
@@ -56,53 +56,25 @@ __m256i test_mm256_add_epi64(__m256i a, __m256i b) {
 
 __m256i test_mm256_adds_epi8(__m256i a, __m256i b) {
   // CHECK-LABEL: test_mm256_adds_epi8
-  // CHECK-NOT: call <32 x i8> @llvm.x86.avx2.padds.b(<32 x i8> %{{.*}}, <32 x i8> %{{.*}})
-  // CHECK: sext <32 x i8> %{{.*}} to <32 x i16>
-  // CHECK: sext <32 x i8> %{{.*}} to <32 x i16>
-  // CHECK: add <32 x i16> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <32 x i16> %{{.*}}, <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: icmp slt <32 x i16> %{{.*}}, <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i16> <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>, <32 x i16> %{{.*}}
-  // CHECK: trunc <32 x i16> %{{.*}} to <32 x i8>
+  // CHECK: call <32 x i8> @llvm.x86.avx2.padds.b(<32 x i8> %{{.*}}, <32 x i8> %{{.*}})
   return _mm256_adds_epi8(a, b);
 }
 
 __m256i test_mm256_adds_epi16(__m256i a, __m256i b) {
   // CHECK-LABEL: test_mm256_adds_epi16
-  // CHECK-NOT: call <16 x i16> @llvm.x86.avx2.padds.w(<16 x i16> %{{.*}}, <16 x i16> %{{.*}})
-  // CHECK: sext <16 x i16> %{{.*}} to <16 x i32>
-  // CHECK: sext <16 x i16> %{{.*}} to <16 x i32>
-  // CHECK: add <16 x i32> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <16 x i32> %{{.*}}, <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: icmp slt <16 x i32> %{{.*}}, <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>
-  // CHECK: select <16 x i1> %{{.*}}, <16 x i32> <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>, <16 x i32> %{{.*}}
-  // CHECK: trunc <16 x i32> %{{.*}} to <16 x i16>
+  // CHECK: call <16 x i16> @llvm.x86.avx2.padds.w(<16 x i16> %{{.*}}, <16 x i16> %{{.*}})
   return _mm256_adds_epi16(a, b);
 }
 
 __m256i test_mm256_adds_epu8(__m256i a, __m256i b) {
   // CHECK-LABEL: test_mm256_adds_epu8
-  // CHECK-NOT: call <32 x i8> @llvm.x86.avx2.paddus.b(<32 x i8> %{{.*}}, <32 x i8> %{{.*}})
-  // CHECK: zext <32 x i8> %{{.*}} to <32 x i16>
-  // CHECK: zext <32 x i8> %{{.*}} to <32 x i16>
-  // CHECK: add <32 x i16> %{{.*}}, %{{.*}}
-  // CHECK: icmp ule <32 x i16> %{{.*}}, <i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> <i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255>
-  // CHECK: trunc <32 x i16> %{{.*}} to <32 x i8>
+  // CHECK: call <32 x i8> @llvm.x86.avx2.paddus.b(<32 x i8> %{{.*}}, <32 x i8> %{{.*}})
   return _mm256_adds_epu8(a, b);
 }
 
 __m256i test_mm256_adds_epu16(__m256i a, __m256i b) {
   // CHECK-LABEL: test_mm256_adds_epu16
-  // CHECK-NOT: call <16 x i16> @llvm.x86.avx2.paddus.w(<16 x i16> %{{.*}}, <16 x i16> %{{.*}})
-  // CHECK: zext <16 x i16> %{{.*}} to <16 x i32>
-  // CHECK: zext <16 x i16> %{{.*}} to <16 x i32>
-  // CHECK: add <16 x i32> %{{.*}}, %{{.*}}
-  // CHECK: icmp ule <16 x i32> %{{.*}}, <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
-  // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
-  // CHECK: trunc <16 x i32> %{{.*}} to <16 x i16>
+  // CHECK: call <16 x i16> @llvm.x86.avx2.paddus.w(<16 x i16> %{{.*}}, <16 x i16> %{{.*}})
   return _mm256_adds_epu16(a, b);
 }
 
@@ -1199,47 +1171,25 @@ __m256i test_mm256_sub_epi64(__m256i a, __m256i b) {
 
 __m256i test_mm256_subs_epi8(__m256i a, __m256i b) {
   // CHECK-LABEL: test_mm256_subs_epi8
-  // CHECK-NOT: call <32 x i8> @llvm.x86.avx2.psubs.b(<32 x i8> %{{.*}}, <32 x i8> %{{.*}})
-  // CHECK: sext <32 x i8> %{{.*}} to <32 x i16>
-  // CHECK: sext <32 x i8> %{{.*}} to <32 x i16>
-  // CHECK: sub <32 x i16> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <32 x i16> %{{.*}}, <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: icmp slt <32 x i16> %{{.*}}, <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i16> <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>, <32 x i16> %{{.*}}
-  // CHECK: trunc <32 x i16> %{{.*}} to <32 x i8>
+  // CHECK: call <32 x i8> @llvm.x86.avx2.psubs.b(<32 x i8> %{{.*}}, <32 x i8> %{{.*}})
   return _mm256_subs_epi8(a, b);
 }
 
 __m256i test_mm256_subs_epi16(__m256i a, __m256i b) {
   // CHECK-LABEL: test_mm256_subs_epi16
-  // CHECK-NOT: call <16 x i16> @llvm.x86.avx2.psubs.w(<16 x i16> %{{.*}}, <16 x i16> %{{.*}})
-  // CHECK: sext <16 x i16> %{{.*}} to <16 x i32>
-  // CHECK: sext <16 x i16> %{{.*}} to <16 x i32>
-  // CHECK: sub <16 x i32> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <16 x i32> %{{.*}}, <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: icmp slt <16 x i32> %{{.*}}, <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>
-  // CHECK: select <16 x i1> %{{.*}}, <16 x i32> <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>, <16 x i32> %{{.*}}
-  // CHECK: trunc <16 x i32> %{{.*}} to <16 x i16>
+  // CHECK: call <16 x i16> @llvm.x86.avx2.psubs.w(<16 x i16> %{{.*}}, <16 x i16> %{{.*}})
   return _mm256_subs_epi16(a, b);
 }
 
 __m256i test_mm256_subs_epu8(__m256i a, __m256i b) {
   // CHECK-LABEL: test_mm256_subs_epu8
-  // CHECK-NOT: call <32 x i8> @llvm.x86.avx2.psubus.b(<32 x i8> %{{.*}}, <32 x i8> %{{.*}})
-  // CHECK: icmp ugt <32 x i8> {{.*}}, {{.*}}
-  // CHECK: select <32 x i1> {{.*}}, <32 x i8> {{.*}}, <32 x i8> {{.*}}
-  // CHECK: sub <32 x i8> {{.*}}, {{.*}}
+  // CHECK: call <32 x i8> @llvm.x86.avx2.psubus.b(<32 x i8> %{{.*}}, <32 x i8> %{{.*}})
   return _mm256_subs_epu8(a, b);
 }
 
 __m256i test_mm256_subs_epu16(__m256i a, __m256i b) {
   // CHECK-LABEL: test_mm256_subs_epu16
-  // CHECK-NOT: call <16 x i16> @llvm.x86.avx2.psubus.w(<16 x i16> %{{.*}}, <16 x i16> %{{.*}})
-  // CHECK: icmp ugt <16 x i16> {{.*}}, {{.*}}
-  // CHECK: select <16 x i1> {{.*}}, <16 x i16> {{.*}}, <16 x i16> {{.*}}
-  // CHECK: sub <16 x i16> {{.*}}, {{.*}}
+  // CHECK: call <16 x i16> @llvm.x86.avx2.psubus.w(<16 x i16> %{{.*}}, <16 x i16> %{{.*}})
   return _mm256_subs_epu16(a, b);
 }
 
diff --git a/test/CodeGen/avx512bw-builtins.c b/test/CodeGen/avx512bw-builtins.c
index 2c5c82b105..bb644c4423 100644
--- a/test/CodeGen/avx512bw-builtins.c
+++ b/test/CodeGen/avx512bw-builtins.c
@@ -594,154 +594,62 @@ __m512i test_mm512_maskz_packus_epi16(__mmask64 __M, __m512i __A, __m512i __B) {
 }
 __m512i test_mm512_adds_epi8(__m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_adds_epi8
-  // CHECK-NOT: @llvm.x86.avx512.mask.padds.b.512
-  // CHECK: sext <64 x i8> %{{.*}} to <64 x i16>
-  // CHECK: sext <64 x i8> %{{.*}} to <64 x i16>
-  // CHECK: add <64 x i16> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <64 x i16> %{{.*}}, <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: select <64 x i1> %{{.*}}, <64 x i16> %{{.*}}, <64 x i16> <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: icmp slt <64 x i16> %{{.*}}, <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>
-  // CHECK: select <64 x i1> %{{.*}}, <64 x i16> <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>, <64 x i16> %{{.*}}
-  // CHECK: trunc <64 x i16> %{{.*}} to <64 x i8>
+  // CHECK: @llvm.x86.avx512.mask.padds.b.512
   return _mm512_adds_epi8(__A,__B); 
 }
 __m512i test_mm512_mask_adds_epi8(__m512i __W, __mmask64 __U, __m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_mask_adds_epi8
-  // CHECK-NOT: @llvm.x86.avx512.mask.padds.b.512
-  // CHECK: sext <64 x i8> %{{.*}} to <64 x i16>
-  // CHECK: sext <64 x i8> %{{.*}} to <64 x i16>
-  // CHECK: add <64 x i16> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <64 x i16> %{{.*}}, <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: select <64 x i1> %{{.*}}, <64 x i16> %{{.*}}, <64 x i16> <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: icmp slt <64 x i16> %{{.*}}, <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>
-  // CHECK: select <64 x i1> %{{.*}}, <64 x i16> <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>, <64 x i16> %{{.*}}
-  // CHECK: trunc <64 x i16> %{{.*}} to <64 x i8>
-  // CHECK: select <64 x i1> %{{.*}}, <64 x i8> %{{.*}}, <64 x i8> %{{.*}}
- return _mm512_mask_adds_epi8(__W,__U,__A,__B); 
+  // CHECK: @llvm.x86.avx512.mask.padds.b.512
+  return _mm512_mask_adds_epi8(__W,__U,__A,__B); 
 }
 __m512i test_mm512_maskz_adds_epi8(__mmask64 __U, __m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_maskz_adds_epi8
-  // CHECK-NOT: @llvm.x86.avx512.mask.padds.b.512
-  // CHECK: sext <64 x i8> %{{.*}} to <64 x i16>
-  // CHECK: sext <64 x i8> %{{.*}} to <64 x i16>
-  // CHECK: add <64 x i16> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <64 x i16> %{{.*}}, <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: select <64 x i1> %{{.*}}, <64 x i16> %{{.*}}, <64 x i16> <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: icmp slt <64 x i16> %{{.*}}, <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>
-  // CHECK: select <64 x i1> %{{.*}}, <64 x i16> <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>, <64 x i16> %{{.*}}
-  // CHECK: trunc <64 x i16> %{{.*}} to <64 x i8>
-  // CHECK: select <64 x i1> %{{.*}}, <64 x i8> %{{.*}}, <64 x i8> %{{.*}}
+  // CHECK: @llvm.x86.avx512.mask.padds.b.512
   return _mm512_maskz_adds_epi8(__U,__A,__B); 
 }
 __m512i test_mm512_adds_epi16(__m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_adds_epi16
-  // CHECK-NOT: @llvm.x86.avx512.mask.padds.w.512
-  // CHECK: sext <32 x i16> %{{.*}} to <32 x i32>
-  // CHECK: sext <32 x i16> %{{.*}} to <32 x i32>
-  // CHECK: add <32 x i32> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <32 x i32> %{{.*}}, <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i32> %{{.*}}, <32 x i32> <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: icmp slt <32 x i32> %{{.*}}, <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i32> <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>, <32 x i32> %{{.*}}
-  // CHECK: trunc <32 x i32> %{{.*}} to <32 x i16>
- return _mm512_adds_epi16(__A,__B); 
+  // CHECK: @llvm.x86.avx512.mask.padds.w.512
+  return _mm512_adds_epi16(__A,__B); 
 }
 __m512i test_mm512_mask_adds_epi16(__m512i __W, __mmask32 __U, __m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_mask_adds_epi16
-  // CHECK-NOT: @llvm.x86.avx512.mask.padds.w.512
-  // CHECK: sext <32 x i16> %{{.*}} to <32 x i32>
-  // CHECK: sext <32 x i16> %{{.*}} to <32 x i32>
-  // CHECK: add <32 x i32> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <32 x i32> %{{.*}}, <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i32> %{{.*}}, <32 x i32> <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: icmp slt <32 x i32> %{{.*}}, <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i32> <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>, <32 x i32> %{{.*}}
-  // CHECK: trunc <32 x i32> %{{.*}} to <32 x i16>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}}
+  // CHECK: @llvm.x86.avx512.mask.padds.w.512
   return _mm512_mask_adds_epi16(__W,__U,__A,__B); 
 }
 __m512i test_mm512_maskz_adds_epi16(__mmask32 __U, __m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_maskz_adds_epi16
-  // CHECK-NOT: @llvm.x86.avx512.mask.padds.w.512
-  // CHECK: sext <32 x i16> %{{.*}} to <32 x i32>
-  // CHECK: sext <32 x i16> %{{.*}} to <32 x i32>
-  // CHECK: add <32 x i32> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <32 x i32> %{{.*}}, <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i32> %{{.*}}, <32 x i32> <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: icmp slt <32 x i32> %{{.*}}, <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i32> <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>, <32 x i32> %{{.*}}
-  // CHECK: trunc <32 x i32> %{{.*}} to <32 x i16>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}}
-return _mm512_maskz_adds_epi16(__U,__A,__B); 
+  // CHECK: @llvm.x86.avx512.mask.padds.w.512
+  return _mm512_maskz_adds_epi16(__U,__A,__B); 
 }
 __m512i test_mm512_adds_epu8(__m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_adds_epu8
-  // CHECK-NOT: @llvm.x86.avx512.mask.paddus.b.512
-  // CHECK: zext <64 x i8> %{{.*}} to <64 x i16>
-  // CHECK: zext <64 x i8> %{{.*}} to <64 x i16>
-  // CHECK: add <64 x i16> %{{.*}}, %{{.*}}
-  // CHECK: icmp ule <64 x i16> %{{.*}}, <i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255>
-  // CHECK: select <64 x i1> %{{.*}}, <64 x i16> %{{.*}}, <64 x i16> <i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255>
-  // CHECK: trunc <64 x i16> %{{.*}} to <64 x i8>
+  // CHECK: @llvm.x86.avx512.mask.paddus.b.512
   return _mm512_adds_epu8(__A,__B); 
 }
 __m512i test_mm512_mask_adds_epu8(__m512i __W, __mmask64 __U, __m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_mask_adds_epu8
-  // CHECK-NOT: @llvm.x86.avx512.mask.paddus.b.512
-  // CHECK: zext <64 x i8> %{{.*}} to <64 x i16>
-  // CHECK: zext <64 x i8> %{{.*}} to <64 x i16>
-  // CHECK: add <64 x i16> %{{.*}}, %{{.*}}
-  // CHECK: icmp ule <64 x i16> %{{.*}}, <i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255>
-  // CHECK: select <64 x i1> %{{.*}}, <64 x i16> %{{.*}}, <64 x i16> <i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255>
-  // CHECK: trunc <64 x i16> %{{.*}} to <64 x i8>
-  // CHECK: select <64 x i1> %{{.*}}, <64 x i8> %{{.*}}, <64 x i8> %{{.*}}
+  // CHECK: @llvm.x86.avx512.mask.paddus.b.512
   return _mm512_mask_adds_epu8(__W,__U,__A,__B); 
 }
 __m512i test_mm512_maskz_adds_epu8(__mmask64 __U, __m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_maskz_adds_epu8
-  // CHECK-NOT: @llvm.x86.avx512.mask.paddus.b.512
-  // CHECK: zext <64 x i8> %{{.*}} to <64 x i16>
-  // CHECK: zext <64 x i8> %{{.*}} to <64 x i16>
-  // CHECK: add <64 x i16> %{{.*}}, %{{.*}}
-  // CHECK: icmp ule <64 x i16> %{{.*}}, <i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255>
-  // CHECK: select <64 x i1> %{{.*}}, <64 x i16> %{{.*}}, <64 x i16> <i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255>
-  // CHECK: trunc <64 x i16> %{{.*}} to <64 x i8>
-  // CHECK: select <64 x i1> %{{.*}}, <64 x i8> %{{.*}}, <64 x i8> %{{.*}}
+  // CHECK: @llvm.x86.avx512.mask.paddus.b.512
   return _mm512_maskz_adds_epu8(__U,__A,__B); 
 }
 __m512i test_mm512_adds_epu16(__m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_adds_epu16
-  // CHECK-NOT: @llvm.x86.avx512.mask.paddus.w.512
-  // CHECK: zext <32 x i16> %{{.*}} to <32 x i32>
-  // CHECK: zext <32 x i16> %{{.*}} to <32 x i32>
-  // CHECK: add <32 x i32> %{{.*}}, %{{.*}}
-  // CHECK: icmp ule <32 x i32> %{{.*}}, <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i32> %{{.*}}, <32 x i32> <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
-  // CHECK: trunc <32 x i32> %{{.*}} to <32 x i16>
+  // CHECK: @llvm.x86.avx512.mask.paddus.w.512
   return _mm512_adds_epu16(__A,__B); 
 }
 __m512i test_mm512_mask_adds_epu16(__m512i __W, __mmask32 __U, __m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_mask_adds_epu16
-  // CHECK-NOT: @llvm.x86.avx512.mask.paddus.w.512
-  // CHECK: zext <32 x i16> %{{.*}} to <32 x i32>
-  // CHECK: zext <32 x i16> %{{.*}} to <32 x i32>
-  // CHECK: add <32 x i32> %{{.*}}, %{{.*}}
-  // CHECK: icmp ule <32 x i32> %{{.*}}, <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i32> %{{.*}}, <32 x i32> <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
-  // CHECK: trunc <32 x i32> %{{.*}} to <32 x i16>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}}
+  // CHECK: @llvm.x86.avx512.mask.paddus.w.512
   return _mm512_mask_adds_epu16(__W,__U,__A,__B); 
 }
 __m512i test_mm512_maskz_adds_epu16(__mmask32 __U, __m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_maskz_adds_epu16
-  // CHECK-NOT: @llvm.x86.avx512.mask.paddus.w.512
-  // CHECK: zext <32 x i16> %{{.*}} to <32 x i32>
-  // CHECK: zext <32 x i16> %{{.*}} to <32 x i32>
-  // CHECK: add <32 x i32> %{{.*}}, %{{.*}}
-  // CHECK: icmp ule <32 x i32> %{{.*}}, <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i32> %{{.*}}, <32 x i32> <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
-  // CHECK: trunc <32 x i32> %{{.*}} to <32 x i16>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}}
+  // CHECK: @llvm.x86.avx512.mask.paddus.w.512
   return _mm512_maskz_adds_epu16(__U,__A,__B); 
 }
 __m512i test_mm512_avg_epu8(__m512i __A, __m512i __B) {
@@ -995,137 +903,63 @@ __m512i test_mm512_maskz_shuffle_epi8(__mmask64 __U, __m512i __A, __m512i __B) {
 }
 __m512i test_mm512_subs_epi8(__m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_subs_epi8
-  // CHECK-NOT: @llvm.x86.avx512.mask.psubs.b.512
-  // CHECK: sext <64 x i8> %{{.*}} to <64 x i16>
-  // CHECK: sext <64 x i8> %{{.*}} to <64 x i16>
-  // CHECK: sub <64 x i16> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <64 x i16> %{{.*}}, <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: select <64 x i1> %{{.*}}, <64 x i16> %{{.*}}, <64 x i16> <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: icmp slt <64 x i16> %{{.*}}, <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>
-  // CHECK: select <64 x i1> %{{.*}}, <64 x i16> <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>, <64 x i16> %{{.*}}
-  // CHECK: trunc <64 x i16> %{{.*}} to <64 x i8>
-return _mm512_subs_epi8(__A,__B); 
+  // CHECK: @llvm.x86.avx512.mask.psubs.b.512
+  return _mm512_subs_epi8(__A,__B); 
 }
 __m512i test_mm512_mask_subs_epi8(__m512i __W, __mmask64 __U, __m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_mask_subs_epi8
-  // CHECK-NOT: @llvm.x86.avx512.mask.psubs.b.512
-  // CHECK: sext <64 x i8> %{{.*}} to <64 x i16>
-  // CHECK: sext <64 x i8> %{{.*}} to <64 x i16>
-  // CHECK: sub <64 x i16> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <64 x i16> %{{.*}}, <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: select <64 x i1> %{{.*}}, <64 x i16> %{{.*}}, <64 x i16> <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: icmp slt <64 x i16> %{{.*}}, <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>
-  // CHECK: select <64 x i1> %{{.*}}, <64 x i16> <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>, <64 x i16> %{{.*}}
-  // CHECK: trunc <64 x i16> %{{.*}} to <64 x i8>
-  // CHECK: select <64 x i1> %{{.*}}, <64 x i8> %{{.*}}, <64 x i8> %{{.*}}
-return _mm512_mask_subs_epi8(__W,__U,__A,__B); 
+  // CHECK: @llvm.x86.avx512.mask.psubs.b.512
+  return _mm512_mask_subs_epi8(__W,__U,__A,__B); 
 }
 __m512i test_mm512_maskz_subs_epi8(__mmask64 __U, __m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_maskz_subs_epi8
-  // CHECK-NOT: @llvm.x86.avx512.mask.psubs.b.512
-  // CHECK: sext <64 x i8> %{{.*}} to <64 x i16>
-  // CHECK: sext <64 x i8> %{{.*}} to <64 x i16>
-  // CHECK: sub <64 x i16> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <64 x i16> %{{.*}}, <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: select <64 x i1> %{{.*}}, <64 x i16> %{{.*}}, <64 x i16> <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: icmp slt <64 x i16> %{{.*}}, <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>
-  // CHECK: select <64 x i1> %{{.*}}, <64 x i16> <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>, <64 x i16> %{{.*}}
-  // CHECK: trunc <64 x i16> %{{.*}} to <64 x i8>
-  // CHECK: select <64 x i1> %{{.*}}, <64 x i8> %{{.*}}, <64 x i8> %{{.*}}
-return _mm512_maskz_subs_epi8(__U,__A,__B); 
+  // CHECK: @llvm.x86.avx512.mask.psubs.b.512
+  return _mm512_maskz_subs_epi8(__U,__A,__B); 
 }
 __m512i test_mm512_subs_epi16(__m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_subs_epi16
-  // CHECK-NOT: @llvm.x86.avx512.mask.psubs.w.512
-  // CHECK: sext <32 x i16> %{{.*}} to <32 x i32>
-  // CHECK: sext <32 x i16> %{{.*}} to <32 x i32>
-  // CHECK: sub <32 x i32> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <32 x i32> %{{.*}}, <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i32> %{{.*}}, <32 x i32> <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: icmp slt <32 x i32> %{{.*}}, <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i32> <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>, <32 x i32> %{{.*}}
-  // CHECK: trunc <32 x i32> %{{.*}} to <32 x i16>
-return _mm512_subs_epi16(__A,__B); 
+  // CHECK: @llvm.x86.avx512.mask.psubs.w.512
+  return _mm512_subs_epi16(__A,__B); 
 }
 __m512i test_mm512_mask_subs_epi16(__m512i __W, __mmask32 __U, __m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_mask_subs_epi16
-  // CHECK-NOT: @llvm.x86.avx512.mask.psubs.w.512
-  // CHECK: sext <32 x i16> %{{.*}} to <32 x i32>
-  // CHECK: sext <32 x i16> %{{.*}} to <32 x i32>
-  // CHECK: sub <32 x i32> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <32 x i32> %{{.*}}, <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i32> %{{.*}}, <32 x i32> <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: icmp slt <32 x i32> %{{.*}}, <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i32> <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>, <32 x i32> %{{.*}}
-  // CHECK: trunc <32 x i32> %{{.*}} to <32 x i16>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}}
-return _mm512_mask_subs_epi16(__W,__U,__A,__B); 
+  // CHECK: @llvm.x86.avx512.mask.psubs.w.512
+  return _mm512_mask_subs_epi16(__W,__U,__A,__B); 
 }
 __m512i test_mm512_maskz_subs_epi16(__mmask32 __U, __m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_maskz_subs_epi16
-  // CHECK-NOT: @llvm.x86.avx512.mask.psubs.w.512
-  // CHECK: sext <32 x i16> %{{.*}} to <32 x i32>
-  // CHECK: sext <32 x i16> %{{.*}} to <32 x i32>
-  // CHECK: sub <32 x i32> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <32 x i32> %{{.*}}, <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i32> %{{.*}}, <32 x i32> <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: icmp slt <32 x i32> %{{.*}}, <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i32> <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>, <32 x i32> %{{.*}}
-  // CHECK: trunc <32 x i32> %{{.*}} to <32 x i16>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}}
-return _mm512_maskz_subs_epi16(__U,__A,__B); 
+  // CHECK: @llvm.x86.avx512.mask.psubs.w.512
+  return _mm512_maskz_subs_epi16(__U,__A,__B); 
 }
 __m512i test_mm512_subs_epu8(__m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_subs_epu8
-  // CHECK-NOT: @llvm.x86.avx512.mask.psubus.b.512
-  // CHECK: icmp ugt <64 x i8> {{.*}}, {{.*}}
-  // CHECK: select <64 x i1> {{.*}}, <64 x i8> {{.*}}, <64 x i8> {{.*}}
-  // CHECK: sub <64 x i8> {{.*}}, {{.*}}
-return _mm512_subs_epu8(__A,__B); 
+  // CHECK: @llvm.x86.avx512.mask.psubus.b.512
+  return _mm512_subs_epu8(__A,__B); 
 }
 __m512i test_mm512_mask_subs_epu8(__m512i __W, __mmask64 __U, __m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_mask_subs_epu8
-  // CHECK-NOT: @llvm.x86.avx512.mask.psubus.b.512
-  // CHECK: icmp ugt <64 x i8> {{.*}}, {{.*}}
-  // CHECK: select <64 x i1> {{.*}}, <64 x i8> {{.*}}, <64 x i8> {{.*}}
-  // CHECK: sub <64 x i8> {{.*}}, {{.*}}
-  // CHECK: select <64 x i1> %{{.*}}, <64 x i8> %{{.*}}, <64 x i8> %{{.*}}
-return _mm512_mask_subs_epu8(__W,__U,__A,__B); 
+  // CHECK: @llvm.x86.avx512.mask.psubus.b.512
+  return _mm512_mask_subs_epu8(__W,__U,__A,__B); 
 }
 __m512i test_mm512_maskz_subs_epu8(__mmask64 __U, __m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_maskz_subs_epu8
-  // CHECK-NOT: @llvm.x86.avx512.mask.psubus.b.512
-  // CHECK: icmp ugt <64 x i8> {{.*}}, {{.*}}
-  // CHECK: select <64 x i1> {{.*}}, <64 x i8> {{.*}}, <64 x i8> {{.*}}
-  // CHECK: sub <64 x i8> {{.*}}, {{.*}}
-  // CHECK: select <64 x i1> %{{.*}}, <64 x i8> %{{.*}}, <64 x i8> %{{.*}}
-return _mm512_maskz_subs_epu8(__U,__A,__B); 
+  // CHECK: @llvm.x86.avx512.mask.psubus.b.512
+  return _mm512_maskz_subs_epu8(__U,__A,__B); 
 }
 __m512i test_mm512_subs_epu16(__m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_subs_epu16
-  // CHECK-NOT: @llvm.x86.avx512.mask.psubus.w.512
-  // CHECK: icmp ugt <32 x i16> {{.*}}, {{.*}}
-  // CHECK: select <32 x i1> {{.*}}, <32 x i16> {{.*}}, <32 x i16> {{.*}}
-  // CHECK: sub <32 x i16> {{.*}}, {{.*}}
-return _mm512_subs_epu16(__A,__B); 
+  // CHECK: @llvm.x86.avx512.mask.psubus.w.512
+  return _mm512_subs_epu16(__A,__B); 
 }
 __m512i test_mm512_mask_subs_epu16(__m512i __W, __mmask32 __U, __m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_mask_subs_epu16
-  // CHECK-NOT: @llvm.x86.avx512.mask.psubus.w.512
-  // CHECK: icmp ugt <32 x i16> {{.*}}, {{.*}}
-  // CHECK: select <32 x i1> {{.*}}, <32 x i16> {{.*}}, <32 x i16> {{.*}}
-  // CHECK: sub <32 x i16> {{.*}}, {{.*}}
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}}
-return _mm512_mask_subs_epu16(__W,__U,__A,__B); 
+  // CHECK: @llvm.x86.avx512.mask.psubus.w.512
+  return _mm512_mask_subs_epu16(__W,__U,__A,__B); 
 }
 __m512i test_mm512_maskz_subs_epu16(__mmask32 __U, __m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_maskz_subs_epu16
-  // CHECK-NOT: @llvm.x86.avx512.mask.psubus.w.512
-  // CHECK: icmp ugt <32 x i16> {{.*}}, {{.*}}
-  // CHECK: select <32 x i1> {{.*}}, <32 x i16> {{.*}}, <32 x i16> {{.*}}
-  // CHECK: sub <32 x i16> {{.*}}, {{.*}}
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}}
-return _mm512_maskz_subs_epu16(__U,__A,__B); 
+  // CHECK: @llvm.x86.avx512.mask.psubus.w.512
+  return _mm512_maskz_subs_epu16(__U,__A,__B); 
 }
 __m512i test_mm512_mask2_permutex2var_epi16(__m512i __A, __m512i __I, __mmask32 __U, __m512i __B) {
   // CHECK-LABEL: @test_mm512_mask2_permutex2var_epi16
diff --git a/test/CodeGen/avx512vlbw-builtins.c b/test/CodeGen/avx512vlbw-builtins.c
index a41a2efd25..7adc50c231 100644
--- a/test/CodeGen/avx512vlbw-builtins.c
+++ b/test/CodeGen/avx512vlbw-builtins.c
@@ -1075,211 +1075,97 @@ __m256i test_mm256_mask_packus_epi16(__m256i __W, __mmask32 __M, __m256i __A,
 
 __m128i test_mm_mask_adds_epi8(__m128i __W, __mmask16 __U, __m128i __A,        __m128i __B) {
   // CHECK-LABEL: @test_mm_mask_adds_epi8
-  // CHECK-NOT: @llvm.x86.sse2.padds.b
-  // CHECK: sext <16 x i8> %{{.*}} to <16 x i16>
-  // CHECK: sext <16 x i8> %{{.*}} to <16 x i16>
-  // CHECK: add <16 x i16> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <16 x i16> %{{.*}}, <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: select <16 x i1> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: icmp slt <16 x i16> %{{.*}}, <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>
-
-  // CHECK: select <16 x i1> %{{.*}}, <16 x i16> <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>, <16 x i16> %{{.*}}
-  // CHECK: trunc <16 x i16> %{{.*}} to <16 x i8>
+  // CHECK: @llvm.x86.sse2.padds.b
   // CHECK: select <16 x i1> %{{.*}}, <16 x i8> %{{.*}}, <16 x i8> %{{.*}}
   return _mm_mask_adds_epi8(__W,__U,__A,__B); 
 }
 __m128i test_mm_maskz_adds_epi8(__mmask16 __U, __m128i __A, __m128i __B) {
   // CHECK-LABEL: @test_mm_maskz_adds_epi8
-  // CHECK-NOT: @llvm.x86.sse2.padds.b
-  // CHECK: sext <16 x i8> %{{.*}} to <16 x i16>
-  // CHECK: sext <16 x i8> %{{.*}} to <16 x i16>
-  // CHECK: add <16 x i16> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <16 x i16> %{{.*}}, <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: select <16 x i1> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: icmp slt <16 x i16> %{{.*}}, <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>
-
-  // CHECK: select <16 x i1> %{{.*}}, <16 x i16> <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>, <16 x i16> %{{.*}}
-  // CHECK: trunc <16 x i16> %{{.*}} to <16 x i8>
+  // CHECK: @llvm.x86.sse2.padds.b
   // CHECK: select <16 x i1> %{{.*}}, <16 x i8> %{{.*}}, <16 x i8> %{{.*}}
   return _mm_maskz_adds_epi8(__U,__A,__B); 
 }
 __m256i test_mm256_mask_adds_epi8(__m256i __W, __mmask32 __U, __m256i __A,           __m256i __B) {
   // CHECK-LABEL: @test_mm256_mask_adds_epi8
-  // CHECK-NOT: @llvm.x86.avx2.padds.b
-  // CHECK: sext <32 x i8> %{{.*}} to <32 x i16>
-  // CHECK: sext <32 x i8> %{{.*}} to <32 x i16>
-  // CHECK: add <32 x i16> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <32 x i16> %{{.*}}, <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: icmp slt <32 x i16> %{{.*}}, <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i16> <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>, <32 x i16> %{{.*}}
-  // CHECK: trunc <32 x i16> %{{.*}} to <32 x i8>
+  // CHECK: @llvm.x86.avx2.padds.b
   // CHECK: select <32 x i1> %{{.*}}, <32 x i8> %{{.*}}, <32 x i8> %{{.*}}
   return _mm256_mask_adds_epi8(__W,__U,__A,__B); 
 }
 __m256i test_mm256_maskz_adds_epi8(__mmask32 __U, __m256i __A, __m256i __B) {
   // CHECK-LABEL: @test_mm256_maskz_adds_epi8
-  // CHECK-NOT: @llvm.x86.avx2.padds.b
-  // CHECK: sext <32 x i8> %{{.*}} to <32 x i16>
-  // CHECK: sext <32 x i8> %{{.*}} to <32 x i16>
-  // CHECK: add <32 x i16> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <32 x i16> %{{.*}}, <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: icmp slt <32 x i16> %{{.*}}, <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i16> <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>, <32 x i16> %{{.*}}
-  // CHECK: trunc <32 x i16> %{{.*}} to <32 x i8>
+  // CHECK: @llvm.x86.avx2.padds.b
   // CHECK: select <32 x i1> %{{.*}}, <32 x i8> %{{.*}}, <32 x i8> %{{.*}}
   return _mm256_maskz_adds_epi8(__U,__A,__B); 
 }
 __m128i test_mm_mask_adds_epi16(__m128i __W, __mmask8 __U, __m128i __A,         __m128i __B) {
   // CHECK-LABEL: @test_mm_mask_adds_epi16
-  // CHECK-NOT: @llvm.x86.sse2.padds.w
-  // CHECK: sext <8 x i16> %{{.*}} to <8 x i32>
-  // CHECK: sext <8 x i16> %{{.*}} to <8 x i32>
-  // CHECK: add <8 x i32> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <8 x i32> %{{.*}}, <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: icmp slt <8 x i32> %{{.*}}, <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>
-  // CHECK: select <8 x i1> %{{.*}}, <8 x i32> <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>, <8 x i32> %{{.*}}
-  // CHECK: trunc <8 x i32> %{{.*}} to <8 x i16>
+  // CHECK: @llvm.x86.sse2.padds.w
   // CHECK: select <8 x i1> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}}
   return _mm_mask_adds_epi16(__W,__U,__A,__B); 
 }
 __m128i test_mm_maskz_adds_epi16(__mmask8 __U, __m128i __A, __m128i __B) {
   // CHECK-LABEL: @test_mm_maskz_adds_epi16
-  // CHECK-NOT: @llvm.x86.sse2.padds.w
-  // CHECK: sext <8 x i16> %{{.*}} to <8 x i32>
-  // CHECK: sext <8 x i16> %{{.*}} to <8 x i32>
-  // CHECK: add <8 x i32> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <8 x i32> %{{.*}}, <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: icmp slt <8 x i32> %{{.*}}, <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>
-  // CHECK: select <8 x i1> %{{.*}}, <8 x i32> <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>, <8 x i32> %{{.*}}
-  // CHECK: trunc <8 x i32> %{{.*}} to <8 x i16>
+  // CHECK: @llvm.x86.sse2.padds.w
   // CHECK: select <8 x i1> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}}
   return _mm_maskz_adds_epi16(__U,__A,__B); 
 }
 __m256i test_mm256_mask_adds_epi16(__m256i __W, __mmask16 __U, __m256i __A,      __m256i __B) {
   // CHECK-LABEL: @test_mm256_mask_adds_epi16
-  // CHECK-NOT: @llvm.x86.avx2.padds.w
-  // CHECK: sext <16 x i16> %{{.*}} to <16 x i32>
-  // CHECK: sext <16 x i16> %{{.*}} to <16 x i32>
-  // CHECK: add <16 x i32> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <16 x i32> %{{.*}}, <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: icmp slt <16 x i32> %{{.*}}, <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>
-  // CHECK: select <16 x i1> %{{.*}}, <16 x i32> <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>, <16 x i32> %{{.*}}
-  // CHECK: trunc <16 x i32> %{{.*}} to <16 x i16>
+  // CHECK: @llvm.x86.avx2.padds.w
   // CHECK: select <16 x i1> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}}
   return _mm256_mask_adds_epi16(__W,__U,__A,__B); 
 }
 __m256i test_mm256_maskz_adds_epi16(__mmask16 __U, __m256i __A, __m256i __B) {
   // CHECK-LABEL: @test_mm256_maskz_adds_epi16
-  // CHECK-NOT: @llvm.x86.avx2.padds.w
-  // CHECK: sext <16 x i16> %{{.*}} to <16 x i32>
-  // CHECK: sext <16 x i16> %{{.*}} to <16 x i32>
-  // CHECK: add <16 x i32> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <16 x i32> %{{.*}}, <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: icmp slt <16 x i32> %{{.*}}, <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>
-  // CHECK: select <16 x i1> %{{.*}}, <16 x i32> <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>, <16 x i32> %{{.*}}
-  // CHECK: trunc <16 x i32> %{{.*}} to <16 x i16>
+  // CHECK: @llvm.x86.avx2.padds.w
   // CHECK: select <16 x i1> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}}
   return _mm256_maskz_adds_epi16(__U,__A,__B); 
 }
-__m128i test_mm_mask_adds_epu8(__m128i __W, __mmask16 __U, __m128i __A, __m128i __B) {
+__m128i test_mm_mask_adds_epu8(__m128i __W, __mmask16 __U, __m128i __A,        __m128i __B) {
   // CHECK-LABEL: @test_mm_mask_adds_epu8
-  // CHECK-NOT: @llvm.x86.sse2.paddus.b
-  // CHECK: zext <16 x i8> %{{.*}} to <16 x i16>
-  // CHECK: zext <16 x i8> %{{.*}} to <16 x i16>
-  // CHECK: add <16 x i16> %{{.*}}, %{{.*}}
-  // CHECK: icmp ule <16 x i16> %{{.*}}, <i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255>
-  // CHECK: select <16 x i1> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> <i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255>
-  // CHECK: trunc <16 x i16> %{{.*}} to <16 x i8>
+  // CHECK: @llvm.x86.sse2.paddus.b
   // CHECK: select <16 x i1> %{{.*}}, <16 x i8> %{{.*}}, <16 x i8> %{{.*}}
   return _mm_mask_adds_epu8(__W,__U,__A,__B); 
 }
 __m128i test_mm_maskz_adds_epu8(__mmask16 __U, __m128i __A, __m128i __B) {
   // CHECK-LABEL: @test_mm_maskz_adds_epu8
-  // CHECK-NOT: @llvm.x86.sse2.paddus.b
-  // CHECK: zext <16 x i8> %{{.*}} to <16 x i16>
-  // CHECK: zext <16 x i8> %{{.*}} to <16 x i16>
-  // CHECK: add <16 x i16> %{{.*}}, %{{.*}}
-  // CHECK: icmp ule <16 x i16> %{{.*}}, <i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255>
-  // CHECK: select <16 x i1> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> <i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255>
-  // CHECK: trunc <16 x i16> %{{.*}} to <16 x i8>
+  // CHECK: @llvm.x86.sse2.paddus.b
   // CHECK: select <16 x i1> %{{.*}}, <16 x i8> %{{.*}}, <16 x i8> %{{.*}}
   return _mm_maskz_adds_epu8(__U,__A,__B); 
 }
 __m256i test_mm256_mask_adds_epu8(__m256i __W, __mmask32 __U, __m256i __A,           __m256i __B) {
   // CHECK-LABEL: @test_mm256_mask_adds_epu8
-  // CHECK-NOT: @llvm.x86.avx2.paddus.b
-  // CHECK: zext <32 x i8> %{{.*}} to <32 x i16>
-  // CHECK: zext <32 x i8> %{{.*}} to <32 x i16>
-  // CHECK: add <32 x i16> %{{.*}}, %{{.*}}
-  // CHECK: icmp ule <32 x i16> %{{.*}}, <i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> <i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255>
-  // CHECK: trunc <32 x i16> %{{.*}} to <32 x i8>
+  // CHECK: @llvm.x86.avx2.paddus.b
   // CHECK: select <32 x i1> %{{.*}}, <32 x i8> %{{.*}}, <32 x i8> %{{.*}}
   return _mm256_mask_adds_epu8(__W,__U,__A,__B); 
 }
 __m256i test_mm256_maskz_adds_epu8(__mmask32 __U, __m256i __A, __m256i __B) {
   // CHECK-LABEL: @test_mm256_maskz_adds_epu8
-  // CHECK-NOT: @llvm.x86.avx2.paddus.b
-  // CHECK: zext <32 x i8> %{{.*}} to <32 x i16>
-  // CHECK: zext <32 x i8> %{{.*}} to <32 x i16>
-  // CHECK: add <32 x i16> %{{.*}}, %{{.*}}
-  // CHECK: icmp ule <32 x i16> %{{.*}}, <i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> <i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255>
-  // CHECK: trunc <32 x i16> %{{.*}} to <32 x i8>
+  // CHECK: @llvm.x86.avx2.paddus.b
   // CHECK: select <32 x i1> %{{.*}}, <32 x i8> %{{.*}}, <32 x i8> %{{.*}}
   return _mm256_maskz_adds_epu8(__U,__A,__B); 
 }
 __m128i test_mm_mask_adds_epu16(__m128i __W, __mmask8 __U, __m128i __A,         __m128i __B) {
   // CHECK-LABEL: @test_mm_mask_adds_epu16
-  // CHECK-NOT: @llvm.x86.sse2.paddus.w
-  // CHECK: zext <8 x i16> %{{.*}} to <8 x i32>
-  // CHECK: zext <8 x i16> %{{.*}} to <8 x i32>
-  // CHECK: add <8 x i32> %{{.*}}, %{{.*}}
-  // CHECK: icmp ule <8 x i32> %{{.*}}, <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
-  // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
-  // CHECK: trunc <8 x i32> %{{.*}} to <8 x i16>
+  // CHECK: @llvm.x86.sse2.paddus.w
   // CHECK: select <8 x i1> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}}
   return _mm_mask_adds_epu16(__W,__U,__A,__B); 
 }
 __m128i test_mm_maskz_adds_epu16(__mmask8 __U, __m128i __A, __m128i __B) {
   // CHECK-LABEL: @test_mm_maskz_adds_epu16
-  // CHECK-NOT: @llvm.x86.sse2.paddus.w
-  // CHECK: zext <8 x i16> %{{.*}} to <8 x i32>
-  // CHECK: zext <8 x i16> %{{.*}} to <8 x i32>
-  // CHECK: add <8 x i32> %{{.*}}, %{{.*}}
-  // CHECK: icmp ule <8 x i32> %{{.*}}, <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
-  // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
-  // CHECK: trunc <8 x i32> %{{.*}} to <8 x i16>
+  // CHECK: @llvm.x86.sse2.paddus.w
   // CHECK: select <8 x i1> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}}
   return _mm_maskz_adds_epu16(__U,__A,__B); 
 }
 __m256i test_mm256_mask_adds_epu16(__m256i __W, __mmask16 __U, __m256i __A,      __m256i __B) {
   // CHECK-LABEL: @test_mm256_mask_adds_epu16
-  // CHECK-NOT: @llvm.x86.avx2.paddus.w
-  // CHECK: zext <16 x i16> %{{.*}} to <16 x i32>
-  // CHECK: zext <16 x i16> %{{.*}} to <16 x i32>
-  // CHECK: add <16 x i32> %{{.*}}, %{{.*}}
-  // CHECK: icmp ule <16 x i32> %{{.*}}, <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
-  // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
-  // CHECK: trunc <16 x i32> %{{.*}} to <16 x i16>
+  // CHECK: @llvm.x86.avx2.paddus.w
   // CHECK: select <16 x i1> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}}
   return _mm256_mask_adds_epu16(__W,__U,__A,__B); 
 }
 __m256i test_mm256_maskz_adds_epu16(__mmask16 __U, __m256i __A, __m256i __B) {
   // CHECK-LABEL: @test_mm256_maskz_adds_epu16
-  // CHECK-NOT: @llvm.x86.avx2.paddus.w
-  // CHECK: zext <16 x i16> %{{.*}} to <16 x i32>
-  // CHECK: zext <16 x i16> %{{.*}} to <16 x i32>
-  // CHECK: add <16 x i32> %{{.*}}, %{{.*}}
-  // CHECK: icmp ule <16 x i32> %{{.*}}, <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
-  // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
-  // CHECK: trunc <16 x i32> %{{.*}} to <16 x i16>
+  // CHECK: @llvm.x86.avx2.paddus.w
   // CHECK: select <16 x i1> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}}
   return _mm256_maskz_adds_epu16(__U,__A,__B); 
 }
@@ -1633,191 +1519,102 @@ __m256i test_mm256_maskz_shuffle_epi8(__mmask32 __U, __m256i __A, __m256i __B) {
 }
 __m128i test_mm_mask_subs_epi8(__m128i __W, __mmask16 __U, __m128i __A,        __m128i __B) {
   // CHECK-LABEL: @test_mm_mask_subs_epi8
-  // CHECK-NOT: @llvm.x86.sse2.psubs.b
-  // CHECK: sext <16 x i8> %{{.*}} to <16 x i16>
-  // CHECK: sext <16 x i8> %{{.*}} to <16 x i16>
-  // CHECK: sub <16 x i16> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <16 x i16> %{{.*}}, <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: select <16 x i1> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: icmp slt <16 x i16> %{{.*}}, <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>
-
-  // CHECK: select <16 x i1> %{{.*}}, <16 x i16> <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>, <16 x i16> %{{.*}}
-  // CHECK: trunc <16 x i16> %{{.*}} to <16 x i8>
+  // CHECK: @llvm.x86.sse2.psubs.b
   // CHECK: select <16 x i1> %{{.*}}, <16 x i8> %{{.*}}, <16 x i8> %{{.*}}
   return _mm_mask_subs_epi8(__W,__U,__A,__B); 
 }
 __m128i test_mm_maskz_subs_epi8(__mmask16 __U, __m128i __A, __m128i __B) {
   // CHECK-LABEL: @test_mm_maskz_subs_epi8
-  // CHECK-NOT: @llvm.x86.sse2.psubs.b
-  // CHECK: sext <16 x i8> %{{.*}} to <16 x i16>
-  // CHECK: sext <16 x i8> %{{.*}} to <16 x i16>
-  // CHECK: sub <16 x i16> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <16 x i16> %{{.*}}, <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: select <16 x i1> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: icmp slt <16 x i16> %{{.*}}, <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>
-
-  // CHECK: select <16 x i1> %{{.*}}, <16 x i16> <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>, <16 x i16> %{{.*}}
-  // CHECK: trunc <16 x i16> %{{.*}} to <16 x i8>
+  // CHECK: @llvm.x86.sse2.psubs.b
   // CHECK: select <16 x i1> %{{.*}}, <16 x i8> %{{.*}}, <16 x i8> %{{.*}}
   return _mm_maskz_subs_epi8(__U,__A,__B); 
 }
 __m256i test_mm256_mask_subs_epi8(__m256i __W, __mmask32 __U, __m256i __A,           __m256i __B) {
   // CHECK-LABEL: @test_mm256_mask_subs_epi8
-  // CHECK-NOT: @llvm.x86.avx2.psubs.b
-  // CHECK: sext <32 x i8> %{{.*}} to <32 x i16>
-  // CHECK: sext <32 x i8> %{{.*}} to <32 x i16>
-  // CHECK: sub <32 x i16> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <32 x i16> %{{.*}}, <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: icmp slt <32 x i16> %{{.*}}, <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i16> <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>, <32 x i16> %{{.*}}
-  // CHECK: trunc <32 x i16> %{{.*}} to <32 x i8>
+  // CHECK: @llvm.x86.avx2.psubs.b
   // CHECK: select <32 x i1> %{{.*}}, <32 x i8> %{{.*}}, <32 x i8> %{{.*}}
   return _mm256_mask_subs_epi8(__W,__U,__A,__B); 
 }
 __m256i test_mm256_maskz_subs_epi8(__mmask32 __U, __m256i __A, __m256i __B) {
   // CHECK-LABEL: @test_mm256_maskz_subs_epi8
-  // CHECK-NOT: @llvm.x86.avx2.psubs.b
-  // CHECK: sext <32 x i8> %{{.*}} to <32 x i16>
-  // CHECK: sext <32 x i8> %{{.*}} to <32 x i16>
-  // CHECK: sub <32 x i16> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <32 x i16> %{{.*}}, <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: icmp slt <32 x i16> %{{.*}}, <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>
-  // CHECK: select <32 x i1> %{{.*}}, <32 x i16> <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>, <32 x i16> %{{.*}}
-  // CHECK: trunc <32 x i16> %{{.*}} to <32 x i8>
+  // CHECK: @llvm.x86.avx2.psubs.b
   // CHECK: select <32 x i1> %{{.*}}, <32 x i8> %{{.*}}, <32 x i8> %{{.*}}
   return _mm256_maskz_subs_epi8(__U,__A,__B); 
 }
 __m128i test_mm_mask_subs_epi16(__m128i __W, __mmask8 __U, __m128i __A,         __m128i __B) {
   // CHECK-LABEL: @test_mm_mask_subs_epi16
-  // CHECK-NOT: @llvm.x86.sse2.psubs.w
-  // CHECK: sext <8 x i16> %{{.*}} to <8 x i32>
-  // CHECK: sext <8 x i16> %{{.*}} to <8 x i32>
-  // CHECK: sub <8 x i32> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <8 x i32> %{{.*}}, <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: icmp slt <8 x i32> %{{.*}}, <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>
-  // CHECK: select <8 x i1> %{{.*}}, <8 x i32> <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>, <8 x i32> %{{.*}}
-  // CHECK: trunc <8 x i32> %{{.*}} to <8 x i16>
+  // CHECK: @llvm.x86.sse2.psubs.w
   // CHECK: select <8 x i1> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}}
   return _mm_mask_subs_epi16(__W,__U,__A,__B); 
 }
 __m128i test_mm_maskz_subs_epi16(__mmask8 __U, __m128i __A, __m128i __B) {
   // CHECK-LABEL: @test_mm_maskz_subs_epi16
-  // CHECK-NOT: @llvm.x86.sse2.psubs.w
-  // CHECK: sext <8 x i16> %{{.*}} to <8 x i32>
-  // CHECK: sext <8 x i16> %{{.*}} to <8 x i32>
-  // CHECK: sub <8 x i32> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <8 x i32> %{{.*}}, <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: icmp slt <8 x i32> %{{.*}}, <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>
-  // CHECK: select <8 x i1> %{{.*}}, <8 x i32> <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>, <8 x i32> %{{.*}}
-  // CHECK: trunc <8 x i32> %{{.*}} to <8 x i16>
+  // CHECK: @llvm.x86.sse2.psubs.w
   // CHECK: select <8 x i1> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}}
   return _mm_maskz_subs_epi16(__U,__A,__B); 
 }
 __m256i test_mm256_mask_subs_epi16(__m256i __W, __mmask16 __U, __m256i __A,      __m256i __B) {
   // CHECK-LABEL: @test_mm256_mask_subs_epi16
-  // CHECK-NOT: @llvm.x86.avx2.psubs.w
-  // CHECK: sext <16 x i16> %{{.*}} to <16 x i32>
-  // CHECK: sext <16 x i16> %{{.*}} to <16 x i32>
-  // CHECK: sub <16 x i32> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <16 x i32> %{{.*}}, <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: icmp slt <16 x i32> %{{.*}}, <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>
-  // CHECK: select <16 x i1> %{{.*}}, <16 x i32> <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>, <16 x i32> %{{.*}}
-  // CHECK: trunc <16 x i32> %{{.*}} to <16 x i16>
+  // CHECK: @llvm.x86.avx2.psubs.w
   // CHECK: select <16 x i1> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}}
   return _mm256_mask_subs_epi16(__W,__U,__A,__B); 
 }
 __m256i test_mm256_maskz_subs_epi16(__mmask16 __U, __m256i __A, __m256i __B) {
   // CHECK-LABEL: @test_mm256_maskz_subs_epi16
-  // CHECK-NOT: @llvm.x86.avx2.psubs.w
-  // CHECK: sext <16 x i16> %{{.*}} to <16 x i32>
-  // CHECK: sext <16 x i16> %{{.*}} to <16 x i32>
-  // CHECK: sub <16 x i32> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <16 x i32> %{{.*}}, <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: icmp slt <16 x i32> %{{.*}}, <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>
-  // CHECK: select <16 x i1> %{{.*}}, <16 x i32> <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>, <16 x i32> %{{.*}}
-  // CHECK: trunc <16 x i32> %{{.*}} to <16 x i16>
+  // CHECK: @llvm.x86.avx2.psubs.w
   // CHECK: select <16 x i1> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}}
   return _mm256_maskz_subs_epi16(__U,__A,__B); 
 }
 __m128i test_mm_mask_subs_epu8(__m128i __W, __mmask16 __U, __m128i __A,        __m128i __B) {
   // CHECK-LABEL: @test_mm_mask_subs_epu8
-  // CHECK-NOT: @llvm.x86.sse2.psubus.b
-  // CHECK: icmp ugt <16 x i8> {{.*}}, {{.*}}
-  // CHECK: select <16 x i1> {{.*}}, <16 x i8> {{.*}}, <16 x i8> {{.*}}
-  // CHECK: sub <16 x i8> {{.*}}, {{.*}}
+  // CHECK: @llvm.x86.sse2.psubus.b
   // CHECK: select <16 x i1> %{{.*}}, <16 x i8> %{{.*}}, <16 x i8> %{{.*}}
   return _mm_mask_subs_epu8(__W,__U,__A,__B); 
 }
 __m128i test_mm_maskz_subs_epu8(__mmask16 __U, __m128i __A, __m128i __B) {
   // CHECK-LABEL: @test_mm_maskz_subs_epu8
-  // CHECK-NOT: @llvm.x86.sse2.psubus.b
-  // CHECK: icmp ugt <16 x i8> {{.*}}, {{.*}}
-  // CHECK: select <16 x i1> {{.*}}, <16 x i8> {{.*}}, <16 x i8> {{.*}}
-  // CHECK: sub <16 x i8> {{.*}}, {{.*}}
+  // CHECK: @llvm.x86.sse2.psubus.b
   // CHECK: select <16 x i1> %{{.*}}, <16 x i8> %{{.*}}, <16 x i8> %{{.*}}
   return _mm_maskz_subs_epu8(__U,__A,__B); 
 }
 __m256i test_mm256_mask_subs_epu8(__m256i __W, __mmask32 __U, __m256i __A,           __m256i __B) {
   // CHECK-LABEL: @test_mm256_mask_subs_epu8
-  // CHECK-NOT: @llvm.x86.avx2.psubus.b
-  // CHECK: icmp ugt <32 x i8> {{.*}}, {{.*}}
-  // CHECK: select <32 x i1> {{.*}}, <32 x i8> {{.*}}, <32 x i8> {{.*}}
-  // CHECK: sub <32 x i8> {{.*}}, {{.*}}
+  // CHECK: @llvm.x86.avx2.psubus.b
   // CHECK: select <32 x i1> %{{.*}}, <32 x i8> %{{.*}}, <32 x i8> %{{.*}}
   return _mm256_mask_subs_epu8(__W,__U,__A,__B); 
 }
 __m256i test_mm256_maskz_subs_epu8(__mmask32 __U, __m256i __A, __m256i __B) {
   // CHECK-LABEL: @test_mm256_maskz_subs_epu8
-  // CHECK-NOT: @llvm.x86.avx2.psubus.b
-  // CHECK: icmp ugt <32 x i8> {{.*}}, {{.*}}
-  // CHECK: select <32 x i1> {{.*}}, <32 x i8> {{.*}}, <32 x i8> {{.*}}
-  // CHECK: sub <32 x i8> {{.*}}, {{.*}}
+  // CHECK: @llvm.x86.avx2.psubus.b
   // CHECK: select <32 x i1> %{{.*}}, <32 x i8> %{{.*}}, <32 x i8> %{{.*}}
   return _mm256_maskz_subs_epu8(__U,__A,__B); 
 }
 __m128i test_mm_mask_subs_epu16(__m128i __W, __mmask8 __U, __m128i __A,         __m128i __B) {
   // CHECK-LABEL: @test_mm_mask_subs_epu16
-  // CHECK-NOT: @llvm.x86.sse2.psubus.w
-  // CHECK: icmp ugt <8 x i16> {{.*}}, {{.*}}
-  // CHECK: select <8 x i1> {{.*}}, <8 x i16> {{.*}}, <8 x i16> {{.*}}
-  // CHECK: sub <8 x i16> {{.*}}, {{.*}}
+  // CHECK: @llvm.x86.sse2.psubus.w
   // CHECK: select <8 x i1> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}}
   return _mm_mask_subs_epu16(__W,__U,__A,__B); 
 }
 __m128i test_mm_maskz_subs_epu16(__mmask8 __U, __m128i __A, __m128i __B) {
   // CHECK-LABEL: @test_mm_maskz_subs_epu16
-  // CHECK-NOT: @llvm.x86.sse2.psubus.w
-  // CHECK: icmp ugt <8 x i16> {{.*}}, {{.*}}
-  // CHECK: select <8 x i1> {{.*}}, <8 x i16> {{.*}}, <8 x i16> {{.*}}
-  // CHECK: sub <8 x i16> {{.*}}, {{.*}}
+  // CHECK: @llvm.x86.sse2.psubus.w
   // CHECK: select <8 x i1> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}}
   return _mm_maskz_subs_epu16(__U,__A,__B); 
 }
 __m256i test_mm256_mask_subs_epu16(__m256i __W, __mmask16 __U, __m256i __A,      __m256i __B) {
   // CHECK-LABEL: @test_mm256_mask_subs_epu16
-  // CHECK-NOT: @llvm.x86.avx2.psubus.w
-  // CHECK: icmp ugt <16 x i16> {{.*}}, {{.*}}
-  // CHECK: select <16 x i1> {{.*}}, <16 x i16> {{.*}}, <16 x i16> {{.*}}
-  // CHECK: sub <16 x i16> {{.*}}, {{.*}}
+  // CHECK: @llvm.x86.avx2.psubus.w
   // CHECK: select <16 x i1> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}}
   return _mm256_mask_subs_epu16(__W,__U,__A,__B); 
 }
 __m256i test_mm256_maskz_subs_epu16(__mmask16 __U, __m256i __A, __m256i __B) {
   // CHECK-LABEL: @test_mm256_maskz_subs_epu16
-  // CHECK-NOT: @llvm.x86.avx2.psubus.w
-  // CHECK: icmp ugt <16 x i16> {{.*}}, {{.*}}
-  // CHECK: select <16 x i1> {{.*}}, <16 x i16> {{.*}}, <16 x i16> {{.*}}
-  // CHECK: sub <16 x i16> {{.*}}, {{.*}}
+  // CHECK: @llvm.x86.avx2.psubus.w
   // CHECK: select <16 x i1> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}}
   return _mm256_maskz_subs_epu16(__U,__A,__B); 
 }
 
+
 __m128i test_mm_mask2_permutex2var_epi16(__m128i __A, __m128i __I, __mmask8 __U,            __m128i __B) {
   // CHECK-LABEL: @test_mm_mask2_permutex2var_epi16
   // CHECK: @llvm.x86.avx512.mask.vpermi2var.hi.128
diff --git a/test/CodeGen/sse2-builtins.c b/test/CodeGen/sse2-builtins.c
index 26fc939b91..4ddb121ad1 100644
--- a/test/CodeGen/sse2-builtins.c
+++ b/test/CodeGen/sse2-builtins.c
@@ -47,53 +47,25 @@ __m128d test_mm_add_sd(__m128d A, __m128d B) {
 
 __m128i test_mm_adds_epi8(__m128i A, __m128i B) {
   // CHECK-LABEL: test_mm_adds_epi8
-  // CHECK-NOT: call <16 x i8> @llvm.x86.sse2.padds.b(<16 x i8> %{{.*}}, <16 x i8> %{{.*}})
-  // CHECK: sext <16 x i8> %{{.*}} to <16 x i16>
-  // CHECK: sext <16 x i8> %{{.*}} to <16 x i16>
-  // CHECK: add <16 x i16> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <16 x i16> %{{.*}}, <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: select <16 x i1> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: icmp slt <16 x i16> %{{.*}}, <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>
-  // CHECK: select <16 x i1> %{{.*}}, <16 x i16> <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>, <16 x i16> %{{.*}}
-  // CHECK: trunc <16 x i16> %{{.*}} to <16 x i8>
+  // CHECK: call <16 x i8> @llvm.x86.sse2.padds.b(<16 x i8> %{{.*}}, <16 x i8> %{{.*}})
   return _mm_adds_epi8(A, B);
 }
 
 __m128i test_mm_adds_epi16(__m128i A, __m128i B) {
   // CHECK-LABEL: test_mm_adds_epi16
-  // CHECK-NOT: call <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16> %{{.*}}, <8 x i16> %{{.*}})
-  // CHECK: sext <8 x i16> %{{.*}} to <8 x i32>
-  // CHECK: sext <8 x i16> %{{.*}} to <8 x i32>
-  // CHECK: add <8 x i32> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <8 x i32> %{{.*}}, <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: icmp slt <8 x i32> %{{.*}}, <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>
-  // CHECK: select <8 x i1> %{{.*}}, <8 x i32> <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>, <8 x i32> %{{.*}}
-  // CHECK: trunc <8 x i32> %{{.*}} to <8 x i16>
+  // CHECK: call <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16> %{{.*}}, <8 x i16> %{{.*}})
   return _mm_adds_epi16(A, B);
 }
 
 __m128i test_mm_adds_epu8(__m128i A, __m128i B) {
   // CHECK-LABEL: test_mm_adds_epu8
-  // CHECK-NOT: call <16 x i8> @llvm.x86.sse2.paddus.b(<16 x i8> %{{.*}}, <16 x i8> %{{.*}})
-  // CHECK: zext <16 x i8> %{{.*}} to <16 x i16>
-  // CHECK: zext <16 x i8> %{{.*}} to <16 x i16>
-  // CHECK: add <16 x i16> %{{.*}}, %{{.*}}
-  // CHECK: icmp ule <16 x i16> %{{.*}}, <i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255>
-  // CHECK: select <16 x i1> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> <i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255>
-  // CHECK: trunc <16 x i16> %{{.*}} to <16 x i8>
+  // CHECK: call <16 x i8> @llvm.x86.sse2.paddus.b(<16 x i8> %{{.*}}, <16 x i8> %{{.*}})
   return _mm_adds_epu8(A, B);
 }
 
 __m128i test_mm_adds_epu16(__m128i A, __m128i B) {
   // CHECK-LABEL: test_mm_adds_epu16
-  // CHECK-NOT: call <8 x i16> @llvm.x86.sse2.paddus.w(<8 x i16> %{{.*}}, <8 x i16> %{{.*}})
-  // CHECK: zext <8 x i16> %{{.*}} to <8 x i32>
-  // CHECK: zext <8 x i16> %{{.*}} to <8 x i32>
-  // CHECK: add <8 x i32> %{{.*}}, %{{.*}}
-  // CHECK: icmp ule <8 x i32> %{{.*}}, <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
-  // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
-  // CHECK: trunc <8 x i32> %{{.*}} to <8 x i16>
+  // CHECK: call <8 x i16> @llvm.x86.sse2.paddus.w(<8 x i16> %{{.*}}, <8 x i16> %{{.*}})
   return _mm_adds_epu16(A, B);
 }
 
@@ -1444,47 +1416,25 @@ __m128d test_mm_sub_sd(__m128d A, __m128d B) {
 
 __m128i test_mm_subs_epi8(__m128i A, __m128i B) {
   // CHECK-LABEL: test_mm_subs_epi8
-  // CHECK-NOT: call <16 x i8> @llvm.x86.sse2.psubs.b(<16 x i8> %{{.*}}, <16 x i8> %{{.*}})
-  // CHECK: sext <16 x i8> %{{.*}} to <16 x i16>
-  // CHECK: sext <16 x i8> %{{.*}} to <16 x i16>
-  // CHECK: sub <16 x i16> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <16 x i16> %{{.*}}, <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: select <16 x i1> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> <i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127, i16 127>
-  // CHECK: icmp slt <16 x i16> %{{.*}}, <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>
-  // CHECK: select <16 x i1> %{{.*}}, <16 x i16> <i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128, i16 -128>, <16 x i16> %{{.*}}
-  // CHECK: trunc <16 x i16> %{{.*}} to <16 x i8>
+  // CHECK: call <16 x i8> @llvm.x86.sse2.psubs.b(<16 x i8> %{{.*}}, <16 x i8> %{{.*}})
   return _mm_subs_epi8(A, B);
 }
 
 __m128i test_mm_subs_epi16(__m128i A, __m128i B) {
   // CHECK-LABEL: test_mm_subs_epi16
-  // CHECK-NOT: call <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16> %{{.*}}, <8 x i16> %{{.*}})
-  // CHECK: sext <8 x i16> %{{.*}} to <8 x i32>
-  // CHECK: sext <8 x i16> %{{.*}} to <8 x i32>
-  // CHECK: sub <8 x i32> %{{.*}}, %{{.*}}
-  // CHECK: icmp sle <8 x i32> %{{.*}}, <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
-  // CHECK: icmp slt <8 x i32> %{{.*}},  <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>
-  // CHECK: select <8 x i1> %{{.*}}, <8 x i32>  <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>, <8 x i32> %{{.*}}
-  // CHECK: trunc <8 x i32> %{{.*}} to <8 x i16>
+  // CHECK: call <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16> %{{.*}}, <8 x i16> %{{.*}})
   return _mm_subs_epi16(A, B);
 }
 
 __m128i test_mm_subs_epu8(__m128i A, __m128i B) {
   // CHECK-LABEL: test_mm_subs_epu8
-  // CHECK-NOT: call <16 x i8> @llvm.x86.sse2.psubus.b(<16 x i8> %{{.*}}, <16 x i8> %{{.*}})
-  // CHECK: icmp ugt <16 x i8> {{.*}}, {{.*}}
-  // CHECK: select <16 x i1> {{.*}}, <16 x i8> {{.*}}, <16 x i8> {{.*}}
-  // CHECK: sub <16 x i8> {{.*}}, {{.*}}
+  // CHECK: call <16 x i8> @llvm.x86.sse2.psubus.b(<16 x i8> %{{.*}}, <16 x i8> %{{.*}})
   return _mm_subs_epu8(A, B);
 }
 
 __m128i test_mm_subs_epu16(__m128i A, __m128i B) {
   // CHECK-LABEL: test_mm_subs_epu16
-  // CHECK-NOT: call <8 x i16> @llvm.x86.sse2.psubus.w(<8 x i16> %{{.*}}, <8 x i16> %{{.*}})
-  // CHECK: icmp ugt <8 x i16> {{.*}}, {{.*}}
-  // CHECK: select <8 x i1> {{.*}}, <8 x i16> {{.*}}, <8 x i16> {{.*}}
-  // CHECK: sub <8 x i16> {{.*}}, {{.*}}
+  // CHECK: call <8 x i16> @llvm.x86.sse2.psubus.w(<8 x i16> %{{.*}}, <8 x i16> %{{.*}})
   return _mm_subs_epu16(A, B);
 }