[llvm] 5e2f094 - [X86] Add common SSE2/SSSE3 check prefix to vector truncation tests
Simon Pilgrim via llvm-commits
llvm-commits at lists.llvm.org
Fri Jun 30 10:49:36 PDT 2023
Author: Simon Pilgrim
Date: 2023-06-30T18:49:26+01:00
New Revision: 5e2f0947c586042083cc1643f2aac90f2492bacb
URL: https://github.com/llvm/llvm-project/commit/5e2f0947c586042083cc1643f2aac90f2492bacb
DIFF: https://github.com/llvm/llvm-project/commit/5e2f0947c586042083cc1643f2aac90f2492bacb.diff
LOG: [X86] Add common SSE2/SSSE3 check prefix to vector truncation tests
Reduced duplicate checks where we can
Added:
Modified:
llvm/test/CodeGen/X86/vector-trunc-packus.ll
llvm/test/CodeGen/X86/vector-trunc-usat.ll
llvm/test/CodeGen/X86/vector-trunc.ll
Removed:
################################################################################
diff --git a/llvm/test/CodeGen/X86/vector-trunc-packus.ll b/llvm/test/CodeGen/X86/vector-trunc-packus.ll
index 804fd89eaf0cb..1babaf9175ea3 100644
--- a/llvm/test/CodeGen/X86/vector-trunc-packus.ll
+++ b/llvm/test/CodeGen/X86/vector-trunc-packus.ll
@@ -1,8 +1,8 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefix=SSE --check-prefix=SSE2
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+ssse3 | FileCheck %s --check-prefix=SSE --check-prefix=SSSE3
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefix=SSE --check-prefix=SSE41
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=AVX --check-prefix=AVX1
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefixes=SSE,SSE2-SSSE3,SSE2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+ssse3 | FileCheck %s --check-prefixes=SSE,SSE2-SSSE3,SSSE3
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefixes=SSE,SSE41
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX1
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-SLOW
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-FAST,AVX2-FAST-ALL
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-FAST,AVX2-FAST-PERLANE
@@ -20,65 +20,35 @@
;
define <2 x i32> @trunc_packus_v2i64_v2i32(<2 x i64> %a0) {
-; SSE2-LABEL: trunc_packus_v2i64_v2i32:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [2147483648,2147483648]
-; SSE2-NEXT: movdqa %xmm0, %xmm2
-; SSE2-NEXT: pxor %xmm1, %xmm2
-; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
-; SSE2-NEXT: pxor %xmm4, %xmm4
-; SSE2-NEXT: pcmpeqd %xmm3, %xmm4
-; SSE2-NEXT: movdqa {{.*#+}} xmm3 = [2147483647,2147483647]
-; SSE2-NEXT: pcmpgtd %xmm2, %xmm3
-; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm3[0,0,2,2]
-; SSE2-NEXT: pand %xmm4, %xmm2
-; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
-; SSE2-NEXT: por %xmm2, %xmm3
-; SSE2-NEXT: pand %xmm3, %xmm0
-; SSE2-NEXT: pandn {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3
-; SSE2-NEXT: por %xmm0, %xmm3
-; SSE2-NEXT: movdqa %xmm3, %xmm0
-; SSE2-NEXT: pxor %xmm1, %xmm0
-; SSE2-NEXT: movdqa %xmm0, %xmm2
-; SSE2-NEXT: pcmpgtd %xmm1, %xmm2
-; SSE2-NEXT: pcmpeqd %xmm1, %xmm0
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; SSE2-NEXT: pand %xmm2, %xmm0
-; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm2[1,1,3,3]
-; SSE2-NEXT: por %xmm0, %xmm1
-; SSE2-NEXT: pand %xmm3, %xmm1
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[0,2,2,3]
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_packus_v2i64_v2i32:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [2147483648,2147483648]
-; SSSE3-NEXT: movdqa %xmm0, %xmm2
-; SSSE3-NEXT: pxor %xmm1, %xmm2
-; SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
-; SSSE3-NEXT: pxor %xmm4, %xmm4
-; SSSE3-NEXT: pcmpeqd %xmm3, %xmm4
-; SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [2147483647,2147483647]
-; SSSE3-NEXT: pcmpgtd %xmm2, %xmm3
-; SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm3[0,0,2,2]
-; SSSE3-NEXT: pand %xmm4, %xmm2
-; SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
-; SSSE3-NEXT: por %xmm2, %xmm3
-; SSSE3-NEXT: pand %xmm3, %xmm0
-; SSSE3-NEXT: pandn {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3
-; SSSE3-NEXT: por %xmm0, %xmm3
-; SSSE3-NEXT: movdqa %xmm3, %xmm0
-; SSSE3-NEXT: pxor %xmm1, %xmm0
-; SSSE3-NEXT: movdqa %xmm0, %xmm2
-; SSSE3-NEXT: pcmpgtd %xmm1, %xmm2
-; SSSE3-NEXT: pcmpeqd %xmm1, %xmm0
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; SSSE3-NEXT: pand %xmm2, %xmm0
-; SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm2[1,1,3,3]
-; SSSE3-NEXT: por %xmm0, %xmm1
-; SSSE3-NEXT: pand %xmm3, %xmm1
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm1[0,2,2,3]
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc_packus_v2i64_v2i32:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [2147483648,2147483648]
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm2
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm2
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
+; SSE2-SSSE3-NEXT: pxor %xmm4, %xmm4
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm3, %xmm4
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [2147483647,2147483647]
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm3[0,0,2,2]
+; SSE2-SSSE3-NEXT: pand %xmm4, %xmm2
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: pand %xmm3, %xmm0
+; SSE2-SSSE3-NEXT: pandn {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3
+; SSE2-SSSE3-NEXT: por %xmm0, %xmm3
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm0
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm2
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm1, %xmm2
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; SSE2-SSSE3-NEXT: pand %xmm2, %xmm0
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm2[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm0, %xmm1
+; SSE2-SSSE3-NEXT: pand %xmm3, %xmm1
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm1[0,2,2,3]
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_packus_v2i64_v2i32:
; SSE41: # %bb.0:
@@ -179,67 +149,36 @@ define <2 x i32> @trunc_packus_v2i64_v2i32(<2 x i64> %a0) {
}
define void @trunc_packus_v2i64_v2i32_store(<2 x i64> %a0, ptr %p1) {
-; SSE2-LABEL: trunc_packus_v2i64_v2i32_store:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [2147483648,2147483648]
-; SSE2-NEXT: movdqa %xmm0, %xmm2
-; SSE2-NEXT: pxor %xmm1, %xmm2
-; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
-; SSE2-NEXT: pxor %xmm4, %xmm4
-; SSE2-NEXT: pcmpeqd %xmm3, %xmm4
-; SSE2-NEXT: movdqa {{.*#+}} xmm3 = [2147483647,2147483647]
-; SSE2-NEXT: pcmpgtd %xmm2, %xmm3
-; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm3[0,0,2,2]
-; SSE2-NEXT: pand %xmm4, %xmm2
-; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
-; SSE2-NEXT: por %xmm2, %xmm3
-; SSE2-NEXT: pand %xmm3, %xmm0
-; SSE2-NEXT: pandn {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3
-; SSE2-NEXT: por %xmm0, %xmm3
-; SSE2-NEXT: movdqa %xmm3, %xmm0
-; SSE2-NEXT: pxor %xmm1, %xmm0
-; SSE2-NEXT: movdqa %xmm0, %xmm2
-; SSE2-NEXT: pcmpgtd %xmm1, %xmm2
-; SSE2-NEXT: pcmpeqd %xmm1, %xmm0
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; SSE2-NEXT: pand %xmm2, %xmm0
-; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm2[1,1,3,3]
-; SSE2-NEXT: por %xmm0, %xmm1
-; SSE2-NEXT: pand %xmm3, %xmm1
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[0,2,2,3]
-; SSE2-NEXT: movq %xmm0, (%rdi)
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_packus_v2i64_v2i32_store:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [2147483648,2147483648]
-; SSSE3-NEXT: movdqa %xmm0, %xmm2
-; SSSE3-NEXT: pxor %xmm1, %xmm2
-; SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
-; SSSE3-NEXT: pxor %xmm4, %xmm4
-; SSSE3-NEXT: pcmpeqd %xmm3, %xmm4
-; SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [2147483647,2147483647]
-; SSSE3-NEXT: pcmpgtd %xmm2, %xmm3
-; SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm3[0,0,2,2]
-; SSSE3-NEXT: pand %xmm4, %xmm2
-; SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
-; SSSE3-NEXT: por %xmm2, %xmm3
-; SSSE3-NEXT: pand %xmm3, %xmm0
-; SSSE3-NEXT: pandn {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3
-; SSSE3-NEXT: por %xmm0, %xmm3
-; SSSE3-NEXT: movdqa %xmm3, %xmm0
-; SSSE3-NEXT: pxor %xmm1, %xmm0
-; SSSE3-NEXT: movdqa %xmm0, %xmm2
-; SSSE3-NEXT: pcmpgtd %xmm1, %xmm2
-; SSSE3-NEXT: pcmpeqd %xmm1, %xmm0
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; SSSE3-NEXT: pand %xmm2, %xmm0
-; SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm2[1,1,3,3]
-; SSSE3-NEXT: por %xmm0, %xmm1
-; SSSE3-NEXT: pand %xmm3, %xmm1
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm1[0,2,2,3]
-; SSSE3-NEXT: movq %xmm0, (%rdi)
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc_packus_v2i64_v2i32_store:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [2147483648,2147483648]
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm2
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm2
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
+; SSE2-SSSE3-NEXT: pxor %xmm4, %xmm4
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm3, %xmm4
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [2147483647,2147483647]
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm3[0,0,2,2]
+; SSE2-SSSE3-NEXT: pand %xmm4, %xmm2
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: pand %xmm3, %xmm0
+; SSE2-SSSE3-NEXT: pandn {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3
+; SSE2-SSSE3-NEXT: por %xmm0, %xmm3
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm0
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm2
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm1, %xmm2
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; SSE2-SSSE3-NEXT: pand %xmm2, %xmm0
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm2[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm0, %xmm1
+; SSE2-SSSE3-NEXT: pand %xmm3, %xmm1
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm1[0,2,2,3]
+; SSE2-SSSE3-NEXT: movq %xmm0, (%rdi)
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_packus_v2i64_v2i32_store:
; SSE41: # %bb.0:
@@ -344,113 +283,59 @@ define void @trunc_packus_v2i64_v2i32_store(<2 x i64> %a0, ptr %p1) {
}
define <4 x i32> @trunc_packus_v4i64_v4i32(<4 x i64> %a0) {
-; SSE2-LABEL: trunc_packus_v4i64_v4i32:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa {{.*#+}} xmm3 = [4294967295,4294967295]
-; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [2147483648,2147483648]
-; SSE2-NEXT: movdqa %xmm0, %xmm4
-; SSE2-NEXT: pxor %xmm2, %xmm4
-; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm4[1,1,3,3]
-; SSE2-NEXT: pxor %xmm6, %xmm6
-; SSE2-NEXT: pcmpeqd %xmm6, %xmm5
-; SSE2-NEXT: movdqa {{.*#+}} xmm7 = [2147483647,2147483647]
-; SSE2-NEXT: movdqa %xmm7, %xmm8
-; SSE2-NEXT: pcmpgtd %xmm4, %xmm8
-; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm8[0,0,2,2]
-; SSE2-NEXT: pand %xmm5, %xmm4
-; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm8[1,1,3,3]
-; SSE2-NEXT: por %xmm4, %xmm5
-; SSE2-NEXT: pand %xmm5, %xmm0
-; SSE2-NEXT: pandn %xmm3, %xmm5
-; SSE2-NEXT: por %xmm5, %xmm0
-; SSE2-NEXT: movdqa %xmm1, %xmm4
-; SSE2-NEXT: pxor %xmm2, %xmm4
-; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm4[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm6, %xmm5
-; SSE2-NEXT: pcmpgtd %xmm4, %xmm7
-; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm7[0,0,2,2]
-; SSE2-NEXT: pand %xmm5, %xmm4
-; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm7[1,1,3,3]
-; SSE2-NEXT: por %xmm4, %xmm5
-; SSE2-NEXT: pand %xmm5, %xmm1
-; SSE2-NEXT: pandn %xmm3, %xmm5
-; SSE2-NEXT: por %xmm1, %xmm5
-; SSE2-NEXT: movdqa %xmm5, %xmm1
-; SSE2-NEXT: pxor %xmm2, %xmm1
-; SSE2-NEXT: movdqa %xmm1, %xmm3
-; SSE2-NEXT: pcmpgtd %xmm2, %xmm3
-; SSE2-NEXT: pcmpeqd %xmm2, %xmm1
-; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; SSE2-NEXT: pand %xmm3, %xmm1
-; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
-; SSE2-NEXT: por %xmm1, %xmm3
-; SSE2-NEXT: pand %xmm5, %xmm3
-; SSE2-NEXT: movdqa %xmm0, %xmm1
-; SSE2-NEXT: pxor %xmm2, %xmm1
-; SSE2-NEXT: movdqa %xmm1, %xmm4
-; SSE2-NEXT: pcmpgtd %xmm2, %xmm4
-; SSE2-NEXT: pcmpeqd %xmm2, %xmm1
-; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; SSE2-NEXT: pand %xmm4, %xmm1
-; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm4[1,1,3,3]
-; SSE2-NEXT: por %xmm1, %xmm2
-; SSE2-NEXT: pand %xmm2, %xmm0
-; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2],xmm3[0,2]
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_packus_v4i64_v4i32:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [4294967295,4294967295]
-; SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [2147483648,2147483648]
-; SSSE3-NEXT: movdqa %xmm0, %xmm4
-; SSSE3-NEXT: pxor %xmm2, %xmm4
-; SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm4[1,1,3,3]
-; SSSE3-NEXT: pxor %xmm6, %xmm6
-; SSSE3-NEXT: pcmpeqd %xmm6, %xmm5
-; SSSE3-NEXT: movdqa {{.*#+}} xmm7 = [2147483647,2147483647]
-; SSSE3-NEXT: movdqa %xmm7, %xmm8
-; SSSE3-NEXT: pcmpgtd %xmm4, %xmm8
-; SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm8[0,0,2,2]
-; SSSE3-NEXT: pand %xmm5, %xmm4
-; SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm8[1,1,3,3]
-; SSSE3-NEXT: por %xmm4, %xmm5
-; SSSE3-NEXT: pand %xmm5, %xmm0
-; SSSE3-NEXT: pandn %xmm3, %xmm5
-; SSSE3-NEXT: por %xmm5, %xmm0
-; SSSE3-NEXT: movdqa %xmm1, %xmm4
-; SSSE3-NEXT: pxor %xmm2, %xmm4
-; SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm4[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm6, %xmm5
-; SSSE3-NEXT: pcmpgtd %xmm4, %xmm7
-; SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm7[0,0,2,2]
-; SSSE3-NEXT: pand %xmm5, %xmm4
-; SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm7[1,1,3,3]
-; SSSE3-NEXT: por %xmm4, %xmm5
-; SSSE3-NEXT: pand %xmm5, %xmm1
-; SSSE3-NEXT: pandn %xmm3, %xmm5
-; SSSE3-NEXT: por %xmm1, %xmm5
-; SSSE3-NEXT: movdqa %xmm5, %xmm1
-; SSSE3-NEXT: pxor %xmm2, %xmm1
-; SSSE3-NEXT: movdqa %xmm1, %xmm3
-; SSSE3-NEXT: pcmpgtd %xmm2, %xmm3
-; SSSE3-NEXT: pcmpeqd %xmm2, %xmm1
-; SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; SSSE3-NEXT: pand %xmm3, %xmm1
-; SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
-; SSSE3-NEXT: por %xmm1, %xmm3
-; SSSE3-NEXT: pand %xmm5, %xmm3
-; SSSE3-NEXT: movdqa %xmm0, %xmm1
-; SSSE3-NEXT: pxor %xmm2, %xmm1
-; SSSE3-NEXT: movdqa %xmm1, %xmm4
-; SSSE3-NEXT: pcmpgtd %xmm2, %xmm4
-; SSSE3-NEXT: pcmpeqd %xmm2, %xmm1
-; SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; SSSE3-NEXT: pand %xmm4, %xmm1
-; SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm4[1,1,3,3]
-; SSSE3-NEXT: por %xmm1, %xmm2
-; SSSE3-NEXT: pand %xmm2, %xmm0
-; SSSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2],xmm3[0,2]
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc_packus_v4i64_v4i32:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [4294967295,4294967295]
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [2147483648,2147483648]
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm4
+; SSE2-SSSE3-NEXT: pxor %xmm2, %xmm4
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm4[1,1,3,3]
+; SSE2-SSSE3-NEXT: pxor %xmm6, %xmm6
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm6, %xmm5
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm7 = [2147483647,2147483647]
+; SSE2-SSSE3-NEXT: movdqa %xmm7, %xmm8
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm4, %xmm8
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm8[0,0,2,2]
+; SSE2-SSSE3-NEXT: pand %xmm5, %xmm4
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm8[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm4, %xmm5
+; SSE2-SSSE3-NEXT: pand %xmm5, %xmm0
+; SSE2-SSSE3-NEXT: pandn %xmm3, %xmm5
+; SSE2-SSSE3-NEXT: por %xmm5, %xmm0
+; SSE2-SSSE3-NEXT: movdqa %xmm1, %xmm4
+; SSE2-SSSE3-NEXT: pxor %xmm2, %xmm4
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm4[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm6, %xmm5
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm4, %xmm7
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm7[0,0,2,2]
+; SSE2-SSSE3-NEXT: pand %xmm5, %xmm4
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm7[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm4, %xmm5
+; SSE2-SSSE3-NEXT: pand %xmm5, %xmm1
+; SSE2-SSSE3-NEXT: pandn %xmm3, %xmm5
+; SSE2-SSSE3-NEXT: por %xmm1, %xmm5
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm1
+; SSE2-SSSE3-NEXT: pxor %xmm2, %xmm1
+; SSE2-SSSE3-NEXT: movdqa %xmm1, %xmm3
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm2, %xmm1
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; SSE2-SSSE3-NEXT: pand %xmm3, %xmm1
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm1, %xmm3
+; SSE2-SSSE3-NEXT: pand %xmm5, %xmm3
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm1
+; SSE2-SSSE3-NEXT: pxor %xmm2, %xmm1
+; SSE2-SSSE3-NEXT: movdqa %xmm1, %xmm4
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm2, %xmm4
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm2, %xmm1
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; SSE2-SSSE3-NEXT: pand %xmm4, %xmm1
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm4[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm1, %xmm2
+; SSE2-SSSE3-NEXT: pand %xmm2, %xmm0
+; SSE2-SSSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2],xmm3[0,2]
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_packus_v4i64_v4i32:
; SSE41: # %bb.0:
@@ -613,215 +498,110 @@ define <4 x i32> @trunc_packus_v4i64_v4i32(<4 x i64> %a0) {
define <8 x i32> @trunc_packus_v8i64_v8i32(ptr %p0) "min-legal-vector-width"="256" {
-; SSE2-LABEL: trunc_packus_v8i64_v8i32:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa (%rdi), %xmm3
-; SSE2-NEXT: movdqa 16(%rdi), %xmm8
-; SSE2-NEXT: movdqa 32(%rdi), %xmm6
-; SSE2-NEXT: movdqa 48(%rdi), %xmm1
-; SSE2-NEXT: movdqa {{.*#+}} xmm4 = [4294967295,4294967295]
-; SSE2-NEXT: movdqa {{.*#+}} xmm0 = [2147483648,2147483648]
-; SSE2-NEXT: movdqa %xmm3, %xmm2
-; SSE2-NEXT: pxor %xmm0, %xmm2
-; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm2[1,1,3,3]
-; SSE2-NEXT: pxor %xmm7, %xmm7
-; SSE2-NEXT: pcmpeqd %xmm7, %xmm9
-; SSE2-NEXT: movdqa {{.*#+}} xmm5 = [2147483647,2147483647]
-; SSE2-NEXT: movdqa %xmm5, %xmm10
-; SSE2-NEXT: pcmpgtd %xmm2, %xmm10
-; SSE2-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
-; SSE2-NEXT: pand %xmm9, %xmm11
-; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm10[1,1,3,3]
-; SSE2-NEXT: por %xmm11, %xmm2
-; SSE2-NEXT: pand %xmm2, %xmm3
-; SSE2-NEXT: pandn %xmm4, %xmm2
-; SSE2-NEXT: por %xmm3, %xmm2
-; SSE2-NEXT: movdqa %xmm8, %xmm3
-; SSE2-NEXT: pxor %xmm0, %xmm3
-; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm3[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm7, %xmm9
-; SSE2-NEXT: movdqa %xmm5, %xmm10
-; SSE2-NEXT: pcmpgtd %xmm3, %xmm10
-; SSE2-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
-; SSE2-NEXT: pand %xmm9, %xmm11
-; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm10[1,1,3,3]
-; SSE2-NEXT: por %xmm11, %xmm3
-; SSE2-NEXT: pand %xmm3, %xmm8
-; SSE2-NEXT: pandn %xmm4, %xmm3
-; SSE2-NEXT: por %xmm8, %xmm3
-; SSE2-NEXT: movdqa %xmm6, %xmm8
-; SSE2-NEXT: pxor %xmm0, %xmm8
-; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm8[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm7, %xmm9
-; SSE2-NEXT: movdqa %xmm5, %xmm10
-; SSE2-NEXT: pcmpgtd %xmm8, %xmm10
-; SSE2-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
-; SSE2-NEXT: pand %xmm9, %xmm11
-; SSE2-NEXT: pshufd {{.*#+}} xmm8 = xmm10[1,1,3,3]
-; SSE2-NEXT: por %xmm11, %xmm8
-; SSE2-NEXT: pand %xmm8, %xmm6
-; SSE2-NEXT: pandn %xmm4, %xmm8
-; SSE2-NEXT: por %xmm6, %xmm8
-; SSE2-NEXT: movdqa %xmm1, %xmm6
-; SSE2-NEXT: pxor %xmm0, %xmm6
-; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm6[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm7, %xmm9
-; SSE2-NEXT: pcmpgtd %xmm6, %xmm5
-; SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm5[0,0,2,2]
-; SSE2-NEXT: pand %xmm9, %xmm6
-; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm5[1,1,3,3]
-; SSE2-NEXT: por %xmm6, %xmm5
-; SSE2-NEXT: pand %xmm5, %xmm1
-; SSE2-NEXT: pandn %xmm4, %xmm5
-; SSE2-NEXT: por %xmm1, %xmm5
-; SSE2-NEXT: movdqa %xmm5, %xmm1
-; SSE2-NEXT: pxor %xmm0, %xmm1
-; SSE2-NEXT: movdqa %xmm1, %xmm4
-; SSE2-NEXT: pcmpgtd %xmm0, %xmm4
-; SSE2-NEXT: pcmpeqd %xmm0, %xmm1
-; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; SSE2-NEXT: pand %xmm4, %xmm1
-; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
-; SSE2-NEXT: por %xmm1, %xmm4
-; SSE2-NEXT: pand %xmm5, %xmm4
-; SSE2-NEXT: movdqa %xmm8, %xmm1
-; SSE2-NEXT: pxor %xmm0, %xmm1
-; SSE2-NEXT: movdqa %xmm1, %xmm5
-; SSE2-NEXT: pcmpgtd %xmm0, %xmm5
-; SSE2-NEXT: pcmpeqd %xmm0, %xmm1
-; SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm1[1,1,3,3]
-; SSE2-NEXT: pand %xmm5, %xmm6
-; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm5[1,1,3,3]
-; SSE2-NEXT: por %xmm6, %xmm1
-; SSE2-NEXT: pand %xmm8, %xmm1
-; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,2],xmm4[0,2]
-; SSE2-NEXT: movdqa %xmm3, %xmm4
-; SSE2-NEXT: pxor %xmm0, %xmm4
-; SSE2-NEXT: movdqa %xmm4, %xmm5
-; SSE2-NEXT: pcmpgtd %xmm0, %xmm5
-; SSE2-NEXT: pcmpeqd %xmm0, %xmm4
-; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
-; SSE2-NEXT: pand %xmm5, %xmm4
-; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm5[1,1,3,3]
-; SSE2-NEXT: por %xmm4, %xmm5
-; SSE2-NEXT: pand %xmm3, %xmm5
-; SSE2-NEXT: movdqa %xmm2, %xmm3
-; SSE2-NEXT: pxor %xmm0, %xmm3
-; SSE2-NEXT: movdqa %xmm3, %xmm4
-; SSE2-NEXT: pcmpgtd %xmm0, %xmm4
-; SSE2-NEXT: pcmpeqd %xmm0, %xmm3
-; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
-; SSE2-NEXT: pand %xmm4, %xmm3
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm4[1,1,3,3]
-; SSE2-NEXT: por %xmm3, %xmm0
-; SSE2-NEXT: pand %xmm2, %xmm0
-; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2],xmm5[0,2]
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_packus_v8i64_v8i32:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa (%rdi), %xmm3
-; SSSE3-NEXT: movdqa 16(%rdi), %xmm8
-; SSSE3-NEXT: movdqa 32(%rdi), %xmm6
-; SSSE3-NEXT: movdqa 48(%rdi), %xmm1
-; SSSE3-NEXT: movdqa {{.*#+}} xmm4 = [4294967295,4294967295]
-; SSSE3-NEXT: movdqa {{.*#+}} xmm0 = [2147483648,2147483648]
-; SSSE3-NEXT: movdqa %xmm3, %xmm2
-; SSSE3-NEXT: pxor %xmm0, %xmm2
-; SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm2[1,1,3,3]
-; SSSE3-NEXT: pxor %xmm7, %xmm7
-; SSSE3-NEXT: pcmpeqd %xmm7, %xmm9
-; SSSE3-NEXT: movdqa {{.*#+}} xmm5 = [2147483647,2147483647]
-; SSSE3-NEXT: movdqa %xmm5, %xmm10
-; SSSE3-NEXT: pcmpgtd %xmm2, %xmm10
-; SSSE3-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
-; SSSE3-NEXT: pand %xmm9, %xmm11
-; SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm10[1,1,3,3]
-; SSSE3-NEXT: por %xmm11, %xmm2
-; SSSE3-NEXT: pand %xmm2, %xmm3
-; SSSE3-NEXT: pandn %xmm4, %xmm2
-; SSSE3-NEXT: por %xmm3, %xmm2
-; SSSE3-NEXT: movdqa %xmm8, %xmm3
-; SSSE3-NEXT: pxor %xmm0, %xmm3
-; SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm3[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm7, %xmm9
-; SSSE3-NEXT: movdqa %xmm5, %xmm10
-; SSSE3-NEXT: pcmpgtd %xmm3, %xmm10
-; SSSE3-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
-; SSSE3-NEXT: pand %xmm9, %xmm11
-; SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm10[1,1,3,3]
-; SSSE3-NEXT: por %xmm11, %xmm3
-; SSSE3-NEXT: pand %xmm3, %xmm8
-; SSSE3-NEXT: pandn %xmm4, %xmm3
-; SSSE3-NEXT: por %xmm8, %xmm3
-; SSSE3-NEXT: movdqa %xmm6, %xmm8
-; SSSE3-NEXT: pxor %xmm0, %xmm8
-; SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm8[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm7, %xmm9
-; SSSE3-NEXT: movdqa %xmm5, %xmm10
-; SSSE3-NEXT: pcmpgtd %xmm8, %xmm10
-; SSSE3-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
-; SSSE3-NEXT: pand %xmm9, %xmm11
-; SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm10[1,1,3,3]
-; SSSE3-NEXT: por %xmm11, %xmm8
-; SSSE3-NEXT: pand %xmm8, %xmm6
-; SSSE3-NEXT: pandn %xmm4, %xmm8
-; SSSE3-NEXT: por %xmm6, %xmm8
-; SSSE3-NEXT: movdqa %xmm1, %xmm6
-; SSSE3-NEXT: pxor %xmm0, %xmm6
-; SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm6[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm7, %xmm9
-; SSSE3-NEXT: pcmpgtd %xmm6, %xmm5
-; SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm5[0,0,2,2]
-; SSSE3-NEXT: pand %xmm9, %xmm6
-; SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm5[1,1,3,3]
-; SSSE3-NEXT: por %xmm6, %xmm5
-; SSSE3-NEXT: pand %xmm5, %xmm1
-; SSSE3-NEXT: pandn %xmm4, %xmm5
-; SSSE3-NEXT: por %xmm1, %xmm5
-; SSSE3-NEXT: movdqa %xmm5, %xmm1
-; SSSE3-NEXT: pxor %xmm0, %xmm1
-; SSSE3-NEXT: movdqa %xmm1, %xmm4
-; SSSE3-NEXT: pcmpgtd %xmm0, %xmm4
-; SSSE3-NEXT: pcmpeqd %xmm0, %xmm1
-; SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; SSSE3-NEXT: pand %xmm4, %xmm1
-; SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
-; SSSE3-NEXT: por %xmm1, %xmm4
-; SSSE3-NEXT: pand %xmm5, %xmm4
-; SSSE3-NEXT: movdqa %xmm8, %xmm1
-; SSSE3-NEXT: pxor %xmm0, %xmm1
-; SSSE3-NEXT: movdqa %xmm1, %xmm5
-; SSSE3-NEXT: pcmpgtd %xmm0, %xmm5
-; SSSE3-NEXT: pcmpeqd %xmm0, %xmm1
-; SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm1[1,1,3,3]
-; SSSE3-NEXT: pand %xmm5, %xmm6
-; SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm5[1,1,3,3]
-; SSSE3-NEXT: por %xmm6, %xmm1
-; SSSE3-NEXT: pand %xmm8, %xmm1
-; SSSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,2],xmm4[0,2]
-; SSSE3-NEXT: movdqa %xmm3, %xmm4
-; SSSE3-NEXT: pxor %xmm0, %xmm4
-; SSSE3-NEXT: movdqa %xmm4, %xmm5
-; SSSE3-NEXT: pcmpgtd %xmm0, %xmm5
-; SSSE3-NEXT: pcmpeqd %xmm0, %xmm4
-; SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
-; SSSE3-NEXT: pand %xmm5, %xmm4
-; SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm5[1,1,3,3]
-; SSSE3-NEXT: por %xmm4, %xmm5
-; SSSE3-NEXT: pand %xmm3, %xmm5
-; SSSE3-NEXT: movdqa %xmm2, %xmm3
-; SSSE3-NEXT: pxor %xmm0, %xmm3
-; SSSE3-NEXT: movdqa %xmm3, %xmm4
-; SSSE3-NEXT: pcmpgtd %xmm0, %xmm4
-; SSSE3-NEXT: pcmpeqd %xmm0, %xmm3
-; SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
-; SSSE3-NEXT: pand %xmm4, %xmm3
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm4[1,1,3,3]
-; SSSE3-NEXT: por %xmm3, %xmm0
-; SSSE3-NEXT: pand %xmm2, %xmm0
-; SSSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2],xmm5[0,2]
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc_packus_v8i64_v8i32:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa (%rdi), %xmm3
+; SSE2-SSSE3-NEXT: movdqa 16(%rdi), %xmm8
+; SSE2-SSSE3-NEXT: movdqa 32(%rdi), %xmm6
+; SSE2-SSSE3-NEXT: movdqa 48(%rdi), %xmm1
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm4 = [4294967295,4294967295]
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm0 = [2147483648,2147483648]
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm2
+; SSE2-SSSE3-NEXT: pxor %xmm0, %xmm2
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm2[1,1,3,3]
+; SSE2-SSSE3-NEXT: pxor %xmm7, %xmm7
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm7, %xmm9
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm5 = [2147483647,2147483647]
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm10
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm2, %xmm10
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
+; SSE2-SSSE3-NEXT: pand %xmm9, %xmm11
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm10[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm11, %xmm2
+; SSE2-SSSE3-NEXT: pand %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: pandn %xmm4, %xmm2
+; SSE2-SSSE3-NEXT: por %xmm3, %xmm2
+; SSE2-SSSE3-NEXT: movdqa %xmm8, %xmm3
+; SSE2-SSSE3-NEXT: pxor %xmm0, %xmm3
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm3[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm7, %xmm9
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm10
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm3, %xmm10
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
+; SSE2-SSSE3-NEXT: pand %xmm9, %xmm11
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm10[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm11, %xmm3
+; SSE2-SSSE3-NEXT: pand %xmm3, %xmm8
+; SSE2-SSSE3-NEXT: pandn %xmm4, %xmm3
+; SSE2-SSSE3-NEXT: por %xmm8, %xmm3
+; SSE2-SSSE3-NEXT: movdqa %xmm6, %xmm8
+; SSE2-SSSE3-NEXT: pxor %xmm0, %xmm8
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm8[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm7, %xmm9
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm10
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm8, %xmm10
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
+; SSE2-SSSE3-NEXT: pand %xmm9, %xmm11
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm10[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm11, %xmm8
+; SSE2-SSSE3-NEXT: pand %xmm8, %xmm6
+; SSE2-SSSE3-NEXT: pandn %xmm4, %xmm8
+; SSE2-SSSE3-NEXT: por %xmm6, %xmm8
+; SSE2-SSSE3-NEXT: movdqa %xmm1, %xmm6
+; SSE2-SSSE3-NEXT: pxor %xmm0, %xmm6
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm6[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm7, %xmm9
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm6, %xmm5
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm5[0,0,2,2]
+; SSE2-SSSE3-NEXT: pand %xmm9, %xmm6
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm5[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm6, %xmm5
+; SSE2-SSSE3-NEXT: pand %xmm5, %xmm1
+; SSE2-SSSE3-NEXT: pandn %xmm4, %xmm5
+; SSE2-SSSE3-NEXT: por %xmm1, %xmm5
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm1
+; SSE2-SSSE3-NEXT: pxor %xmm0, %xmm1
+; SSE2-SSSE3-NEXT: movdqa %xmm1, %xmm4
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm0, %xmm4
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm0, %xmm1
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; SSE2-SSSE3-NEXT: pand %xmm4, %xmm1
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm1, %xmm4
+; SSE2-SSSE3-NEXT: pand %xmm5, %xmm4
+; SSE2-SSSE3-NEXT: movdqa %xmm8, %xmm1
+; SSE2-SSSE3-NEXT: pxor %xmm0, %xmm1
+; SSE2-SSSE3-NEXT: movdqa %xmm1, %xmm5
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm0, %xmm5
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm0, %xmm1
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm1[1,1,3,3]
+; SSE2-SSSE3-NEXT: pand %xmm5, %xmm6
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm5[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm6, %xmm1
+; SSE2-SSSE3-NEXT: pand %xmm8, %xmm1
+; SSE2-SSSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,2],xmm4[0,2]
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm4
+; SSE2-SSSE3-NEXT: pxor %xmm0, %xmm4
+; SSE2-SSSE3-NEXT: movdqa %xmm4, %xmm5
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm0, %xmm5
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm0, %xmm4
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
+; SSE2-SSSE3-NEXT: pand %xmm5, %xmm4
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm5[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm4, %xmm5
+; SSE2-SSSE3-NEXT: pand %xmm3, %xmm5
+; SSE2-SSSE3-NEXT: movdqa %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: pxor %xmm0, %xmm3
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm4
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm0, %xmm4
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm0, %xmm3
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
+; SSE2-SSSE3-NEXT: pand %xmm4, %xmm3
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm4[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm3, %xmm0
+; SSE2-SSSE3-NEXT: pand %xmm2, %xmm0
+; SSE2-SSSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2],xmm5[0,2]
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_packus_v8i64_v8i32:
; SSE41: # %bb.0:
@@ -1037,69 +817,37 @@ define <8 x i32> @trunc_packus_v8i64_v8i32(ptr %p0) "min-legal-vector-width"="25
;
define <2 x i16> @trunc_packus_v2i64_v2i16(<2 x i64> %a0) {
-; SSE2-LABEL: trunc_packus_v2i64_v2i16:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [2147483648,2147483648]
-; SSE2-NEXT: movdqa %xmm0, %xmm2
-; SSE2-NEXT: pxor %xmm1, %xmm2
-; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
-; SSE2-NEXT: pxor %xmm4, %xmm4
-; SSE2-NEXT: pcmpeqd %xmm3, %xmm4
-; SSE2-NEXT: movdqa {{.*#+}} xmm3 = [2147549183,2147549183]
-; SSE2-NEXT: pcmpgtd %xmm2, %xmm3
-; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm3[0,0,2,2]
-; SSE2-NEXT: pand %xmm4, %xmm2
-; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
-; SSE2-NEXT: por %xmm2, %xmm3
-; SSE2-NEXT: pand %xmm3, %xmm0
-; SSE2-NEXT: pandn {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3
-; SSE2-NEXT: por %xmm0, %xmm3
-; SSE2-NEXT: movdqa %xmm3, %xmm0
-; SSE2-NEXT: pxor %xmm1, %xmm0
-; SSE2-NEXT: movdqa %xmm0, %xmm2
-; SSE2-NEXT: pcmpgtd %xmm1, %xmm2
-; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm2[0,0,2,2]
-; SSE2-NEXT: pcmpeqd %xmm1, %xmm0
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; SSE2-NEXT: pand %xmm4, %xmm0
-; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm2[1,1,3,3]
-; SSE2-NEXT: por %xmm0, %xmm1
-; SSE2-NEXT: pand %xmm3, %xmm1
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[0,2,2,3]
-; SSE2-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_packus_v2i64_v2i16:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [2147483648,2147483648]
-; SSSE3-NEXT: movdqa %xmm0, %xmm2
-; SSSE3-NEXT: pxor %xmm1, %xmm2
-; SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
-; SSSE3-NEXT: pxor %xmm4, %xmm4
-; SSSE3-NEXT: pcmpeqd %xmm3, %xmm4
-; SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [2147549183,2147549183]
-; SSSE3-NEXT: pcmpgtd %xmm2, %xmm3
-; SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm3[0,0,2,2]
-; SSSE3-NEXT: pand %xmm4, %xmm2
-; SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
-; SSSE3-NEXT: por %xmm2, %xmm3
-; SSSE3-NEXT: pand %xmm3, %xmm0
-; SSSE3-NEXT: pandn {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3
-; SSSE3-NEXT: por %xmm0, %xmm3
-; SSSE3-NEXT: movdqa %xmm3, %xmm0
-; SSSE3-NEXT: pxor %xmm1, %xmm0
-; SSSE3-NEXT: movdqa %xmm0, %xmm2
-; SSSE3-NEXT: pcmpgtd %xmm1, %xmm2
-; SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm2[0,0,2,2]
-; SSSE3-NEXT: pcmpeqd %xmm1, %xmm0
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; SSSE3-NEXT: pand %xmm4, %xmm0
-; SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm2[1,1,3,3]
-; SSSE3-NEXT: por %xmm0, %xmm1
-; SSSE3-NEXT: pand %xmm3, %xmm1
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm1[0,2,2,3]
-; SSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc_packus_v2i64_v2i16:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [2147483648,2147483648]
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm2
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm2
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
+; SSE2-SSSE3-NEXT: pxor %xmm4, %xmm4
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm3, %xmm4
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [2147549183,2147549183]
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm3[0,0,2,2]
+; SSE2-SSSE3-NEXT: pand %xmm4, %xmm2
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: pand %xmm3, %xmm0
+; SSE2-SSSE3-NEXT: pandn {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3
+; SSE2-SSSE3-NEXT: por %xmm0, %xmm3
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm0
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm2
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm1, %xmm2
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm2[0,0,2,2]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; SSE2-SSSE3-NEXT: pand %xmm4, %xmm0
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm2[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm0, %xmm1
+; SSE2-SSSE3-NEXT: pand %xmm3, %xmm1
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm1[0,2,2,3]
+; SSE2-SSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_packus_v2i64_v2i16:
; SSE41: # %bb.0:
@@ -1212,71 +960,38 @@ define <2 x i16> @trunc_packus_v2i64_v2i16(<2 x i64> %a0) {
}
define void @trunc_packus_v2i64_v2i16_store(<2 x i64> %a0, ptr%p1) {
-; SSE2-LABEL: trunc_packus_v2i64_v2i16_store:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [2147483648,2147483648]
-; SSE2-NEXT: movdqa %xmm0, %xmm2
-; SSE2-NEXT: pxor %xmm1, %xmm2
-; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
-; SSE2-NEXT: pxor %xmm4, %xmm4
-; SSE2-NEXT: pcmpeqd %xmm3, %xmm4
-; SSE2-NEXT: movdqa {{.*#+}} xmm3 = [2147549183,2147549183]
-; SSE2-NEXT: pcmpgtd %xmm2, %xmm3
-; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm3[0,0,2,2]
-; SSE2-NEXT: pand %xmm4, %xmm2
-; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
-; SSE2-NEXT: por %xmm2, %xmm3
-; SSE2-NEXT: pand %xmm3, %xmm0
-; SSE2-NEXT: pandn {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3
-; SSE2-NEXT: por %xmm0, %xmm3
-; SSE2-NEXT: movdqa %xmm3, %xmm0
-; SSE2-NEXT: pxor %xmm1, %xmm0
-; SSE2-NEXT: movdqa %xmm0, %xmm2
-; SSE2-NEXT: pcmpgtd %xmm1, %xmm2
-; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm2[0,0,2,2]
-; SSE2-NEXT: pcmpeqd %xmm1, %xmm0
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; SSE2-NEXT: pand %xmm4, %xmm0
-; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm2[1,1,3,3]
-; SSE2-NEXT: por %xmm0, %xmm1
-; SSE2-NEXT: pand %xmm3, %xmm1
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[0,2,2,3]
-; SSE2-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
-; SSE2-NEXT: movd %xmm0, (%rdi)
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_packus_v2i64_v2i16_store:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [2147483648,2147483648]
-; SSSE3-NEXT: movdqa %xmm0, %xmm2
-; SSSE3-NEXT: pxor %xmm1, %xmm2
-; SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
-; SSSE3-NEXT: pxor %xmm4, %xmm4
-; SSSE3-NEXT: pcmpeqd %xmm3, %xmm4
-; SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [2147549183,2147549183]
-; SSSE3-NEXT: pcmpgtd %xmm2, %xmm3
-; SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm3[0,0,2,2]
-; SSSE3-NEXT: pand %xmm4, %xmm2
-; SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
-; SSSE3-NEXT: por %xmm2, %xmm3
-; SSSE3-NEXT: pand %xmm3, %xmm0
-; SSSE3-NEXT: pandn {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3
-; SSSE3-NEXT: por %xmm0, %xmm3
-; SSSE3-NEXT: movdqa %xmm3, %xmm0
-; SSSE3-NEXT: pxor %xmm1, %xmm0
-; SSSE3-NEXT: movdqa %xmm0, %xmm2
-; SSSE3-NEXT: pcmpgtd %xmm1, %xmm2
-; SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm2[0,0,2,2]
-; SSSE3-NEXT: pcmpeqd %xmm1, %xmm0
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; SSSE3-NEXT: pand %xmm4, %xmm0
-; SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm2[1,1,3,3]
-; SSSE3-NEXT: por %xmm0, %xmm1
-; SSSE3-NEXT: pand %xmm3, %xmm1
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm1[0,2,2,3]
-; SSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
-; SSSE3-NEXT: movd %xmm0, (%rdi)
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc_packus_v2i64_v2i16_store:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [2147483648,2147483648]
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm2
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm2
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
+; SSE2-SSSE3-NEXT: pxor %xmm4, %xmm4
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm3, %xmm4
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [2147549183,2147549183]
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm3[0,0,2,2]
+; SSE2-SSSE3-NEXT: pand %xmm4, %xmm2
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: pand %xmm3, %xmm0
+; SSE2-SSSE3-NEXT: pandn {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3
+; SSE2-SSSE3-NEXT: por %xmm0, %xmm3
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm0
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm2
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm1, %xmm2
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm2[0,0,2,2]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; SSE2-SSSE3-NEXT: pand %xmm4, %xmm0
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm2[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm0, %xmm1
+; SSE2-SSSE3-NEXT: pand %xmm3, %xmm1
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm1[0,2,2,3]
+; SSE2-SSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
+; SSE2-SSSE3-NEXT: movd %xmm0, (%rdi)
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_packus_v2i64_v2i16_store:
; SSE41: # %bb.0:
@@ -1366,155 +1081,95 @@ define void @trunc_packus_v2i64_v2i16_store(<2 x i64> %a0, ptr%p1) {
; AVX512BW-LABEL: trunc_packus_v2i64_v2i16_store:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512BW-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
-; AVX512BW-NEXT: vpmovusqw %zmm0, %xmm0
-; AVX512BW-NEXT: vmovd %xmm0, (%rdi)
-; AVX512BW-NEXT: vzeroupper
-; AVX512BW-NEXT: retq
-;
-; AVX512BWVL-LABEL: trunc_packus_v2i64_v2i16_store:
-; AVX512BWVL: # %bb.0:
-; AVX512BWVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512BWVL-NEXT: vpmaxsq %xmm1, %xmm0, %xmm0
-; AVX512BWVL-NEXT: vpmovusqw %xmm0, (%rdi)
-; AVX512BWVL-NEXT: retq
-;
-; SKX-LABEL: trunc_packus_v2i64_v2i16_store:
-; SKX: # %bb.0:
-; SKX-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; SKX-NEXT: vpmaxsq %xmm1, %xmm0, %xmm0
-; SKX-NEXT: vpmovusqw %xmm0, (%rdi)
-; SKX-NEXT: retq
- %1 = icmp slt <2 x i64> %a0, <i64 65535, i64 65535>
- %2 = select <2 x i1> %1, <2 x i64> %a0, <2 x i64> <i64 65535, i64 65535>
- %3 = icmp sgt <2 x i64> %2, zeroinitializer
- %4 = select <2 x i1> %3, <2 x i64> %2, <2 x i64> zeroinitializer
- %5 = trunc <2 x i64> %4 to <2 x i16>
- store <2 x i16> %5, ptr%p1
- ret void
-}
-
-define <4 x i16> @trunc_packus_v4i64_v4i16(<4 x i64> %a0) {
-; SSE2-LABEL: trunc_packus_v4i64_v4i16:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa {{.*#+}} xmm4 = [65535,65535]
-; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [2147483648,2147483648]
-; SSE2-NEXT: movdqa %xmm1, %xmm3
-; SSE2-NEXT: pxor %xmm2, %xmm3
-; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm3[1,1,3,3]
-; SSE2-NEXT: pxor %xmm6, %xmm6
-; SSE2-NEXT: pcmpeqd %xmm6, %xmm5
-; SSE2-NEXT: movdqa {{.*#+}} xmm7 = [2147549183,2147549183]
-; SSE2-NEXT: movdqa %xmm7, %xmm8
-; SSE2-NEXT: pcmpgtd %xmm3, %xmm8
-; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm8[0,0,2,2]
-; SSE2-NEXT: pand %xmm5, %xmm9
-; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm8[1,1,3,3]
-; SSE2-NEXT: por %xmm9, %xmm3
-; SSE2-NEXT: pand %xmm3, %xmm1
-; SSE2-NEXT: pandn %xmm4, %xmm3
-; SSE2-NEXT: por %xmm1, %xmm3
-; SSE2-NEXT: movdqa %xmm0, %xmm1
-; SSE2-NEXT: pxor %xmm2, %xmm1
-; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm1[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm6, %xmm5
-; SSE2-NEXT: pcmpgtd %xmm1, %xmm7
-; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm7[0,0,2,2]
-; SSE2-NEXT: pand %xmm5, %xmm1
-; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm7[1,1,3,3]
-; SSE2-NEXT: por %xmm1, %xmm5
-; SSE2-NEXT: pand %xmm5, %xmm0
-; SSE2-NEXT: pandn %xmm4, %xmm5
-; SSE2-NEXT: por %xmm0, %xmm5
-; SSE2-NEXT: movdqa %xmm5, %xmm0
-; SSE2-NEXT: pxor %xmm2, %xmm0
-; SSE2-NEXT: movdqa %xmm0, %xmm1
-; SSE2-NEXT: pcmpgtd %xmm2, %xmm1
-; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm1[0,0,2,2]
-; SSE2-NEXT: pcmpeqd %xmm2, %xmm0
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; SSE2-NEXT: pand %xmm4, %xmm0
-; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; SSE2-NEXT: por %xmm0, %xmm1
-; SSE2-NEXT: pand %xmm5, %xmm1
-; SSE2-NEXT: movdqa %xmm3, %xmm0
-; SSE2-NEXT: pxor %xmm2, %xmm0
-; SSE2-NEXT: movdqa %xmm0, %xmm4
-; SSE2-NEXT: pcmpgtd %xmm2, %xmm4
-; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm4[0,0,2,2]
-; SSE2-NEXT: pcmpeqd %xmm2, %xmm0
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; SSE2-NEXT: pand %xmm5, %xmm0
-; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm4[1,1,3,3]
-; SSE2-NEXT: por %xmm0, %xmm2
-; SSE2-NEXT: pand %xmm3, %xmm2
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[0,2,2,3]
-; SSE2-NEXT: pshuflw {{.*#+}} xmm2 = xmm0[0,2,2,3,4,5,6,7]
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[0,2,2,3]
-; SSE2-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
-; SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_packus_v4i64_v4i16:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa {{.*#+}} xmm4 = [65535,65535]
-; SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [2147483648,2147483648]
-; SSSE3-NEXT: movdqa %xmm1, %xmm3
-; SSSE3-NEXT: pxor %xmm2, %xmm3
-; SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm3[1,1,3,3]
-; SSSE3-NEXT: pxor %xmm6, %xmm6
-; SSSE3-NEXT: pcmpeqd %xmm6, %xmm5
-; SSSE3-NEXT: movdqa {{.*#+}} xmm7 = [2147549183,2147549183]
-; SSSE3-NEXT: movdqa %xmm7, %xmm8
-; SSSE3-NEXT: pcmpgtd %xmm3, %xmm8
-; SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm8[0,0,2,2]
-; SSSE3-NEXT: pand %xmm5, %xmm9
-; SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm8[1,1,3,3]
-; SSSE3-NEXT: por %xmm9, %xmm3
-; SSSE3-NEXT: pand %xmm3, %xmm1
-; SSSE3-NEXT: pandn %xmm4, %xmm3
-; SSSE3-NEXT: por %xmm1, %xmm3
-; SSSE3-NEXT: movdqa %xmm0, %xmm1
-; SSSE3-NEXT: pxor %xmm2, %xmm1
-; SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm1[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm6, %xmm5
-; SSSE3-NEXT: pcmpgtd %xmm1, %xmm7
-; SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm7[0,0,2,2]
-; SSSE3-NEXT: pand %xmm5, %xmm1
-; SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm7[1,1,3,3]
-; SSSE3-NEXT: por %xmm1, %xmm5
-; SSSE3-NEXT: pand %xmm5, %xmm0
-; SSSE3-NEXT: pandn %xmm4, %xmm5
-; SSSE3-NEXT: por %xmm0, %xmm5
-; SSSE3-NEXT: movdqa %xmm5, %xmm0
-; SSSE3-NEXT: pxor %xmm2, %xmm0
-; SSSE3-NEXT: movdqa %xmm0, %xmm1
-; SSSE3-NEXT: pcmpgtd %xmm2, %xmm1
-; SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm1[0,0,2,2]
-; SSSE3-NEXT: pcmpeqd %xmm2, %xmm0
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; SSSE3-NEXT: pand %xmm4, %xmm0
-; SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; SSSE3-NEXT: por %xmm0, %xmm1
-; SSSE3-NEXT: pand %xmm5, %xmm1
-; SSSE3-NEXT: movdqa %xmm3, %xmm0
-; SSSE3-NEXT: pxor %xmm2, %xmm0
-; SSSE3-NEXT: movdqa %xmm0, %xmm4
-; SSSE3-NEXT: pcmpgtd %xmm2, %xmm4
-; SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm4[0,0,2,2]
-; SSSE3-NEXT: pcmpeqd %xmm2, %xmm0
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; SSSE3-NEXT: pand %xmm5, %xmm0
-; SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm4[1,1,3,3]
-; SSSE3-NEXT: por %xmm0, %xmm2
-; SSSE3-NEXT: pand %xmm3, %xmm2
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm2[0,2,2,3]
-; SSSE3-NEXT: pshuflw {{.*#+}} xmm2 = xmm0[0,2,2,3,4,5,6,7]
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm1[0,2,2,3]
-; SSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
-; SSSE3-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
-; SSSE3-NEXT: retq
+; AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1
+; AVX512BW-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT: vpmovusqw %zmm0, %xmm0
+; AVX512BW-NEXT: vmovd %xmm0, (%rdi)
+; AVX512BW-NEXT: vzeroupper
+; AVX512BW-NEXT: retq
+;
+; AVX512BWVL-LABEL: trunc_packus_v2i64_v2i16_store:
+; AVX512BWVL: # %bb.0:
+; AVX512BWVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
+; AVX512BWVL-NEXT: vpmaxsq %xmm1, %xmm0, %xmm0
+; AVX512BWVL-NEXT: vpmovusqw %xmm0, (%rdi)
+; AVX512BWVL-NEXT: retq
+;
+; SKX-LABEL: trunc_packus_v2i64_v2i16_store:
+; SKX: # %bb.0:
+; SKX-NEXT: vpxor %xmm1, %xmm1, %xmm1
+; SKX-NEXT: vpmaxsq %xmm1, %xmm0, %xmm0
+; SKX-NEXT: vpmovusqw %xmm0, (%rdi)
+; SKX-NEXT: retq
+ %1 = icmp slt <2 x i64> %a0, <i64 65535, i64 65535>
+ %2 = select <2 x i1> %1, <2 x i64> %a0, <2 x i64> <i64 65535, i64 65535>
+ %3 = icmp sgt <2 x i64> %2, zeroinitializer
+ %4 = select <2 x i1> %3, <2 x i64> %2, <2 x i64> zeroinitializer
+ %5 = trunc <2 x i64> %4 to <2 x i16>
+ store <2 x i16> %5, ptr%p1
+ ret void
+}
+
+define <4 x i16> @trunc_packus_v4i64_v4i16(<4 x i64> %a0) {
+; SSE2-SSSE3-LABEL: trunc_packus_v4i64_v4i16:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm4 = [65535,65535]
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [2147483648,2147483648]
+; SSE2-SSSE3-NEXT: movdqa %xmm1, %xmm3
+; SSE2-SSSE3-NEXT: pxor %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm3[1,1,3,3]
+; SSE2-SSSE3-NEXT: pxor %xmm6, %xmm6
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm6, %xmm5
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm7 = [2147549183,2147549183]
+; SSE2-SSSE3-NEXT: movdqa %xmm7, %xmm8
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm3, %xmm8
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm8[0,0,2,2]
+; SSE2-SSSE3-NEXT: pand %xmm5, %xmm9
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm8[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm9, %xmm3
+; SSE2-SSSE3-NEXT: pand %xmm3, %xmm1
+; SSE2-SSSE3-NEXT: pandn %xmm4, %xmm3
+; SSE2-SSSE3-NEXT: por %xmm1, %xmm3
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm1
+; SSE2-SSSE3-NEXT: pxor %xmm2, %xmm1
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm1[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm6, %xmm5
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm1, %xmm7
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm7[0,0,2,2]
+; SSE2-SSSE3-NEXT: pand %xmm5, %xmm1
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm7[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm1, %xmm5
+; SSE2-SSSE3-NEXT: pand %xmm5, %xmm0
+; SSE2-SSSE3-NEXT: pandn %xmm4, %xmm5
+; SSE2-SSSE3-NEXT: por %xmm0, %xmm5
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm0
+; SSE2-SSSE3-NEXT: pxor %xmm2, %xmm0
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm1
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm2, %xmm1
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm1[0,0,2,2]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm2, %xmm0
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; SSE2-SSSE3-NEXT: pand %xmm4, %xmm0
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm0, %xmm1
+; SSE2-SSSE3-NEXT: pand %xmm5, %xmm1
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm0
+; SSE2-SSSE3-NEXT: pxor %xmm2, %xmm0
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm4
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm2, %xmm4
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm4[0,0,2,2]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm2, %xmm0
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; SSE2-SSSE3-NEXT: pand %xmm5, %xmm0
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm4[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm0, %xmm2
+; SSE2-SSSE3-NEXT: pand %xmm3, %xmm2
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm2[0,2,2,3]
+; SSE2-SSSE3-NEXT: pshuflw {{.*#+}} xmm2 = xmm0[0,2,2,3,4,5,6,7]
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm1[0,2,2,3]
+; SSE2-SSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
+; SSE2-SSSE3-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_packus_v4i64_v4i16:
; SSE41: # %bb.0:
@@ -1649,127 +1304,66 @@ define <4 x i16> @trunc_packus_v4i64_v4i16(<4 x i64> %a0) {
}
define void @trunc_packus_v4i64_v4i16_store(<4 x i64> %a0, ptr%p1) {
-; SSE2-LABEL: trunc_packus_v4i64_v4i16_store:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa {{.*#+}} xmm4 = [65535,65535]
-; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [2147483648,2147483648]
-; SSE2-NEXT: movdqa %xmm1, %xmm3
-; SSE2-NEXT: pxor %xmm2, %xmm3
-; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm3[1,1,3,3]
-; SSE2-NEXT: pxor %xmm6, %xmm6
-; SSE2-NEXT: pcmpeqd %xmm6, %xmm5
-; SSE2-NEXT: movdqa {{.*#+}} xmm7 = [2147549183,2147549183]
-; SSE2-NEXT: movdqa %xmm7, %xmm8
-; SSE2-NEXT: pcmpgtd %xmm3, %xmm8
-; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm8[0,0,2,2]
-; SSE2-NEXT: pand %xmm5, %xmm9
-; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm8[1,1,3,3]
-; SSE2-NEXT: por %xmm9, %xmm3
-; SSE2-NEXT: pand %xmm3, %xmm1
-; SSE2-NEXT: pandn %xmm4, %xmm3
-; SSE2-NEXT: por %xmm1, %xmm3
-; SSE2-NEXT: movdqa %xmm0, %xmm1
-; SSE2-NEXT: pxor %xmm2, %xmm1
-; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm1[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm6, %xmm5
-; SSE2-NEXT: pcmpgtd %xmm1, %xmm7
-; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm7[0,0,2,2]
-; SSE2-NEXT: pand %xmm5, %xmm1
-; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm7[1,1,3,3]
-; SSE2-NEXT: por %xmm1, %xmm5
-; SSE2-NEXT: pand %xmm5, %xmm0
-; SSE2-NEXT: pandn %xmm4, %xmm5
-; SSE2-NEXT: por %xmm0, %xmm5
-; SSE2-NEXT: movdqa %xmm5, %xmm0
-; SSE2-NEXT: pxor %xmm2, %xmm0
-; SSE2-NEXT: movdqa %xmm0, %xmm1
-; SSE2-NEXT: pcmpgtd %xmm2, %xmm1
-; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm1[0,0,2,2]
-; SSE2-NEXT: pcmpeqd %xmm2, %xmm0
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; SSE2-NEXT: pand %xmm4, %xmm0
-; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; SSE2-NEXT: por %xmm0, %xmm1
-; SSE2-NEXT: pand %xmm5, %xmm1
-; SSE2-NEXT: movdqa %xmm3, %xmm0
-; SSE2-NEXT: pxor %xmm2, %xmm0
-; SSE2-NEXT: movdqa %xmm0, %xmm4
-; SSE2-NEXT: pcmpgtd %xmm2, %xmm4
-; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm4[0,0,2,2]
-; SSE2-NEXT: pcmpeqd %xmm2, %xmm0
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; SSE2-NEXT: pand %xmm5, %xmm0
-; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm4[1,1,3,3]
-; SSE2-NEXT: por %xmm0, %xmm2
-; SSE2-NEXT: pand %xmm3, %xmm2
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[0,2,2,3]
-; SSE2-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
-; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; SSE2-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
-; SSE2-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; SSE2-NEXT: movq %xmm1, (%rdi)
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_packus_v4i64_v4i16_store:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa {{.*#+}} xmm4 = [65535,65535]
-; SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [2147483648,2147483648]
-; SSSE3-NEXT: movdqa %xmm1, %xmm3
-; SSSE3-NEXT: pxor %xmm2, %xmm3
-; SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm3[1,1,3,3]
-; SSSE3-NEXT: pxor %xmm6, %xmm6
-; SSSE3-NEXT: pcmpeqd %xmm6, %xmm5
-; SSSE3-NEXT: movdqa {{.*#+}} xmm7 = [2147549183,2147549183]
-; SSSE3-NEXT: movdqa %xmm7, %xmm8
-; SSSE3-NEXT: pcmpgtd %xmm3, %xmm8
-; SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm8[0,0,2,2]
-; SSSE3-NEXT: pand %xmm5, %xmm9
-; SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm8[1,1,3,3]
-; SSSE3-NEXT: por %xmm9, %xmm3
-; SSSE3-NEXT: pand %xmm3, %xmm1
-; SSSE3-NEXT: pandn %xmm4, %xmm3
-; SSSE3-NEXT: por %xmm1, %xmm3
-; SSSE3-NEXT: movdqa %xmm0, %xmm1
-; SSSE3-NEXT: pxor %xmm2, %xmm1
-; SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm1[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm6, %xmm5
-; SSSE3-NEXT: pcmpgtd %xmm1, %xmm7
-; SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm7[0,0,2,2]
-; SSSE3-NEXT: pand %xmm5, %xmm1
-; SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm7[1,1,3,3]
-; SSSE3-NEXT: por %xmm1, %xmm5
-; SSSE3-NEXT: pand %xmm5, %xmm0
-; SSSE3-NEXT: pandn %xmm4, %xmm5
-; SSSE3-NEXT: por %xmm0, %xmm5
-; SSSE3-NEXT: movdqa %xmm5, %xmm0
-; SSSE3-NEXT: pxor %xmm2, %xmm0
-; SSSE3-NEXT: movdqa %xmm0, %xmm1
-; SSSE3-NEXT: pcmpgtd %xmm2, %xmm1
-; SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm1[0,0,2,2]
-; SSSE3-NEXT: pcmpeqd %xmm2, %xmm0
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; SSSE3-NEXT: pand %xmm4, %xmm0
-; SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; SSSE3-NEXT: por %xmm0, %xmm1
-; SSSE3-NEXT: pand %xmm5, %xmm1
-; SSSE3-NEXT: movdqa %xmm3, %xmm0
-; SSSE3-NEXT: pxor %xmm2, %xmm0
-; SSSE3-NEXT: movdqa %xmm0, %xmm4
-; SSSE3-NEXT: pcmpgtd %xmm2, %xmm4
-; SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm4[0,0,2,2]
-; SSSE3-NEXT: pcmpeqd %xmm2, %xmm0
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; SSSE3-NEXT: pand %xmm5, %xmm0
-; SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm4[1,1,3,3]
-; SSSE3-NEXT: por %xmm0, %xmm2
-; SSSE3-NEXT: pand %xmm3, %xmm2
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm2[0,2,2,3]
-; SSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
-; SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; SSSE3-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
-; SSSE3-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; SSSE3-NEXT: movq %xmm1, (%rdi)
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc_packus_v4i64_v4i16_store:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm4 = [65535,65535]
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [2147483648,2147483648]
+; SSE2-SSSE3-NEXT: movdqa %xmm1, %xmm3
+; SSE2-SSSE3-NEXT: pxor %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm3[1,1,3,3]
+; SSE2-SSSE3-NEXT: pxor %xmm6, %xmm6
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm6, %xmm5
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm7 = [2147549183,2147549183]
+; SSE2-SSSE3-NEXT: movdqa %xmm7, %xmm8
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm3, %xmm8
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm8[0,0,2,2]
+; SSE2-SSSE3-NEXT: pand %xmm5, %xmm9
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm8[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm9, %xmm3
+; SSE2-SSSE3-NEXT: pand %xmm3, %xmm1
+; SSE2-SSSE3-NEXT: pandn %xmm4, %xmm3
+; SSE2-SSSE3-NEXT: por %xmm1, %xmm3
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm1
+; SSE2-SSSE3-NEXT: pxor %xmm2, %xmm1
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm1[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm6, %xmm5
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm1, %xmm7
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm7[0,0,2,2]
+; SSE2-SSSE3-NEXT: pand %xmm5, %xmm1
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm7[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm1, %xmm5
+; SSE2-SSSE3-NEXT: pand %xmm5, %xmm0
+; SSE2-SSSE3-NEXT: pandn %xmm4, %xmm5
+; SSE2-SSSE3-NEXT: por %xmm0, %xmm5
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm0
+; SSE2-SSSE3-NEXT: pxor %xmm2, %xmm0
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm1
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm2, %xmm1
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm1[0,0,2,2]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm2, %xmm0
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; SSE2-SSSE3-NEXT: pand %xmm4, %xmm0
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm0, %xmm1
+; SSE2-SSSE3-NEXT: pand %xmm5, %xmm1
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm0
+; SSE2-SSSE3-NEXT: pxor %xmm2, %xmm0
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm4
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm2, %xmm4
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm4[0,0,2,2]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm2, %xmm0
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; SSE2-SSSE3-NEXT: pand %xmm5, %xmm0
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm4[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm0, %xmm2
+; SSE2-SSSE3-NEXT: pand %xmm3, %xmm2
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm2[0,2,2,3]
+; SSE2-SSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
+; SSE2-SSSE3-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
+; SSE2-SSSE3-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
+; SSE2-SSSE3-NEXT: movq %xmm1, (%rdi)
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_packus_v4i64_v4i16_store:
; SSE41: # %bb.0:
@@ -1909,241 +1503,123 @@ define void @trunc_packus_v4i64_v4i16_store(<4 x i64> %a0, ptr%p1) {
}
define <8 x i16> @trunc_packus_v8i64_v8i16(ptr %p0) "min-legal-vector-width"="256" {
-; SSE2-LABEL: trunc_packus_v8i64_v8i16:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa (%rdi), %xmm8
-; SSE2-NEXT: movdqa 16(%rdi), %xmm2
-; SSE2-NEXT: movdqa 32(%rdi), %xmm3
-; SSE2-NEXT: movdqa 48(%rdi), %xmm6
-; SSE2-NEXT: movdqa {{.*#+}} xmm4 = [65535,65535]
-; SSE2-NEXT: movdqa {{.*#+}} xmm0 = [2147483648,2147483648]
-; SSE2-NEXT: movdqa %xmm2, %xmm1
-; SSE2-NEXT: pxor %xmm0, %xmm1
-; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm1[1,1,3,3]
-; SSE2-NEXT: pxor %xmm7, %xmm7
-; SSE2-NEXT: pcmpeqd %xmm7, %xmm9
-; SSE2-NEXT: movdqa {{.*#+}} xmm5 = [2147549183,2147549183]
-; SSE2-NEXT: movdqa %xmm5, %xmm10
-; SSE2-NEXT: pcmpgtd %xmm1, %xmm10
-; SSE2-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
-; SSE2-NEXT: pand %xmm9, %xmm11
-; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm10[1,1,3,3]
-; SSE2-NEXT: por %xmm11, %xmm1
-; SSE2-NEXT: pand %xmm1, %xmm2
-; SSE2-NEXT: pandn %xmm4, %xmm1
-; SSE2-NEXT: por %xmm2, %xmm1
-; SSE2-NEXT: movdqa %xmm8, %xmm2
-; SSE2-NEXT: pxor %xmm0, %xmm2
-; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm2[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm7, %xmm9
-; SSE2-NEXT: movdqa %xmm5, %xmm10
-; SSE2-NEXT: pcmpgtd %xmm2, %xmm10
-; SSE2-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
-; SSE2-NEXT: pand %xmm9, %xmm11
-; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm10[1,1,3,3]
-; SSE2-NEXT: por %xmm11, %xmm2
-; SSE2-NEXT: pand %xmm2, %xmm8
-; SSE2-NEXT: pandn %xmm4, %xmm2
-; SSE2-NEXT: por %xmm8, %xmm2
-; SSE2-NEXT: movdqa %xmm6, %xmm8
-; SSE2-NEXT: pxor %xmm0, %xmm8
-; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm8[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm7, %xmm9
-; SSE2-NEXT: movdqa %xmm5, %xmm10
-; SSE2-NEXT: pcmpgtd %xmm8, %xmm10
-; SSE2-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
-; SSE2-NEXT: pand %xmm9, %xmm11
-; SSE2-NEXT: pshufd {{.*#+}} xmm8 = xmm10[1,1,3,3]
-; SSE2-NEXT: por %xmm11, %xmm8
-; SSE2-NEXT: pand %xmm8, %xmm6
-; SSE2-NEXT: pandn %xmm4, %xmm8
-; SSE2-NEXT: por %xmm6, %xmm8
-; SSE2-NEXT: movdqa %xmm3, %xmm6
-; SSE2-NEXT: pxor %xmm0, %xmm6
-; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm6[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm7, %xmm9
-; SSE2-NEXT: pcmpgtd %xmm6, %xmm5
-; SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm5[0,0,2,2]
-; SSE2-NEXT: pand %xmm9, %xmm6
-; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm5[1,1,3,3]
-; SSE2-NEXT: por %xmm6, %xmm5
-; SSE2-NEXT: pand %xmm5, %xmm3
-; SSE2-NEXT: pandn %xmm4, %xmm5
-; SSE2-NEXT: por %xmm3, %xmm5
-; SSE2-NEXT: movdqa %xmm5, %xmm3
-; SSE2-NEXT: pxor %xmm0, %xmm3
-; SSE2-NEXT: movdqa %xmm3, %xmm4
-; SSE2-NEXT: pcmpgtd %xmm0, %xmm4
-; SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm4[0,0,2,2]
-; SSE2-NEXT: pcmpeqd %xmm0, %xmm3
-; SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm3[1,1,3,3]
-; SSE2-NEXT: pand %xmm6, %xmm7
-; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm4[1,1,3,3]
-; SSE2-NEXT: por %xmm7, %xmm3
-; SSE2-NEXT: pand %xmm5, %xmm3
-; SSE2-NEXT: movdqa %xmm8, %xmm4
-; SSE2-NEXT: pxor %xmm0, %xmm4
-; SSE2-NEXT: movdqa %xmm4, %xmm5
-; SSE2-NEXT: pcmpgtd %xmm0, %xmm5
-; SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm5[0,0,2,2]
-; SSE2-NEXT: pcmpeqd %xmm0, %xmm4
-; SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm4[1,1,3,3]
-; SSE2-NEXT: pand %xmm6, %xmm7
-; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm5[1,1,3,3]
-; SSE2-NEXT: por %xmm7, %xmm4
-; SSE2-NEXT: pand %xmm8, %xmm4
-; SSE2-NEXT: movdqa %xmm2, %xmm5
-; SSE2-NEXT: pxor %xmm0, %xmm5
-; SSE2-NEXT: movdqa %xmm5, %xmm6
-; SSE2-NEXT: pcmpgtd %xmm0, %xmm6
-; SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm6[0,0,2,2]
-; SSE2-NEXT: pcmpeqd %xmm0, %xmm5
-; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm5[1,1,3,3]
-; SSE2-NEXT: pand %xmm7, %xmm5
-; SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
-; SSE2-NEXT: por %xmm5, %xmm6
-; SSE2-NEXT: pand %xmm2, %xmm6
-; SSE2-NEXT: movdqa %xmm1, %xmm2
-; SSE2-NEXT: pxor %xmm0, %xmm2
-; SSE2-NEXT: movdqa %xmm2, %xmm5
-; SSE2-NEXT: pcmpgtd %xmm0, %xmm5
-; SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm5[0,0,2,2]
-; SSE2-NEXT: pcmpeqd %xmm0, %xmm2
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[1,1,3,3]
-; SSE2-NEXT: pand %xmm7, %xmm0
-; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm5[1,1,3,3]
-; SSE2-NEXT: por %xmm0, %xmm2
-; SSE2-NEXT: pand %xmm1, %xmm2
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[0,2,2,3]
-; SSE2-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
-; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm6[0,2,2,3]
-; SSE2-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
-; SSE2-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm4[0,2,2,3]
-; SSE2-NEXT: pshuflw {{.*#+}} xmm2 = xmm0[0,1,0,2,4,5,6,7]
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm3[0,2,2,3]
-; SSE2-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,1,0,2,4,5,6,7]
-; SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
-; SSE2-NEXT: movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_packus_v8i64_v8i16:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa (%rdi), %xmm8
-; SSSE3-NEXT: movdqa 16(%rdi), %xmm2
-; SSSE3-NEXT: movdqa 32(%rdi), %xmm3
-; SSSE3-NEXT: movdqa 48(%rdi), %xmm6
-; SSSE3-NEXT: movdqa {{.*#+}} xmm4 = [65535,65535]
-; SSSE3-NEXT: movdqa {{.*#+}} xmm0 = [2147483648,2147483648]
-; SSSE3-NEXT: movdqa %xmm2, %xmm1
-; SSSE3-NEXT: pxor %xmm0, %xmm1
-; SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm1[1,1,3,3]
-; SSSE3-NEXT: pxor %xmm7, %xmm7
-; SSSE3-NEXT: pcmpeqd %xmm7, %xmm9
-; SSSE3-NEXT: movdqa {{.*#+}} xmm5 = [2147549183,2147549183]
-; SSSE3-NEXT: movdqa %xmm5, %xmm10
-; SSSE3-NEXT: pcmpgtd %xmm1, %xmm10
-; SSSE3-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
-; SSSE3-NEXT: pand %xmm9, %xmm11
-; SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm10[1,1,3,3]
-; SSSE3-NEXT: por %xmm11, %xmm1
-; SSSE3-NEXT: pand %xmm1, %xmm2
-; SSSE3-NEXT: pandn %xmm4, %xmm1
-; SSSE3-NEXT: por %xmm2, %xmm1
-; SSSE3-NEXT: movdqa %xmm8, %xmm2
-; SSSE3-NEXT: pxor %xmm0, %xmm2
-; SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm2[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm7, %xmm9
-; SSSE3-NEXT: movdqa %xmm5, %xmm10
-; SSSE3-NEXT: pcmpgtd %xmm2, %xmm10
-; SSSE3-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
-; SSSE3-NEXT: pand %xmm9, %xmm11
-; SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm10[1,1,3,3]
-; SSSE3-NEXT: por %xmm11, %xmm2
-; SSSE3-NEXT: pand %xmm2, %xmm8
-; SSSE3-NEXT: pandn %xmm4, %xmm2
-; SSSE3-NEXT: por %xmm8, %xmm2
-; SSSE3-NEXT: movdqa %xmm6, %xmm8
-; SSSE3-NEXT: pxor %xmm0, %xmm8
-; SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm8[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm7, %xmm9
-; SSSE3-NEXT: movdqa %xmm5, %xmm10
-; SSSE3-NEXT: pcmpgtd %xmm8, %xmm10
-; SSSE3-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
-; SSSE3-NEXT: pand %xmm9, %xmm11
-; SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm10[1,1,3,3]
-; SSSE3-NEXT: por %xmm11, %xmm8
-; SSSE3-NEXT: pand %xmm8, %xmm6
-; SSSE3-NEXT: pandn %xmm4, %xmm8
-; SSSE3-NEXT: por %xmm6, %xmm8
-; SSSE3-NEXT: movdqa %xmm3, %xmm6
-; SSSE3-NEXT: pxor %xmm0, %xmm6
-; SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm6[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm7, %xmm9
-; SSSE3-NEXT: pcmpgtd %xmm6, %xmm5
-; SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm5[0,0,2,2]
-; SSSE3-NEXT: pand %xmm9, %xmm6
-; SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm5[1,1,3,3]
-; SSSE3-NEXT: por %xmm6, %xmm5
-; SSSE3-NEXT: pand %xmm5, %xmm3
-; SSSE3-NEXT: pandn %xmm4, %xmm5
-; SSSE3-NEXT: por %xmm3, %xmm5
-; SSSE3-NEXT: movdqa %xmm5, %xmm3
-; SSSE3-NEXT: pxor %xmm0, %xmm3
-; SSSE3-NEXT: movdqa %xmm3, %xmm4
-; SSSE3-NEXT: pcmpgtd %xmm0, %xmm4
-; SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm4[0,0,2,2]
-; SSSE3-NEXT: pcmpeqd %xmm0, %xmm3
-; SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm3[1,1,3,3]
-; SSSE3-NEXT: pand %xmm6, %xmm7
-; SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm4[1,1,3,3]
-; SSSE3-NEXT: por %xmm7, %xmm3
-; SSSE3-NEXT: pand %xmm5, %xmm3
-; SSSE3-NEXT: movdqa %xmm8, %xmm4
-; SSSE3-NEXT: pxor %xmm0, %xmm4
-; SSSE3-NEXT: movdqa %xmm4, %xmm5
-; SSSE3-NEXT: pcmpgtd %xmm0, %xmm5
-; SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm5[0,0,2,2]
-; SSSE3-NEXT: pcmpeqd %xmm0, %xmm4
-; SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm4[1,1,3,3]
-; SSSE3-NEXT: pand %xmm6, %xmm7
-; SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm5[1,1,3,3]
-; SSSE3-NEXT: por %xmm7, %xmm4
-; SSSE3-NEXT: pand %xmm8, %xmm4
-; SSSE3-NEXT: movdqa %xmm2, %xmm5
-; SSSE3-NEXT: pxor %xmm0, %xmm5
-; SSSE3-NEXT: movdqa %xmm5, %xmm6
-; SSSE3-NEXT: pcmpgtd %xmm0, %xmm6
-; SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm6[0,0,2,2]
-; SSSE3-NEXT: pcmpeqd %xmm0, %xmm5
-; SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm5[1,1,3,3]
-; SSSE3-NEXT: pand %xmm7, %xmm5
-; SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
-; SSSE3-NEXT: por %xmm5, %xmm6
-; SSSE3-NEXT: pand %xmm2, %xmm6
-; SSSE3-NEXT: movdqa %xmm1, %xmm2
-; SSSE3-NEXT: pxor %xmm0, %xmm2
-; SSSE3-NEXT: movdqa %xmm2, %xmm5
-; SSSE3-NEXT: pcmpgtd %xmm0, %xmm5
-; SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm5[0,0,2,2]
-; SSSE3-NEXT: pcmpeqd %xmm0, %xmm2
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm2[1,1,3,3]
-; SSSE3-NEXT: pand %xmm7, %xmm0
-; SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm5[1,1,3,3]
-; SSSE3-NEXT: por %xmm0, %xmm2
-; SSSE3-NEXT: pand %xmm1, %xmm2
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm2[0,2,2,3]
-; SSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
-; SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm6[0,2,2,3]
-; SSSE3-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
-; SSSE3-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm4[0,2,2,3]
-; SSSE3-NEXT: pshuflw {{.*#+}} xmm2 = xmm0[0,1,0,2,4,5,6,7]
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm3[0,2,2,3]
-; SSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,1,0,2,4,5,6,7]
-; SSSE3-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
-; SSSE3-NEXT: movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc_packus_v8i64_v8i16:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa (%rdi), %xmm8
+; SSE2-SSSE3-NEXT: movdqa 16(%rdi), %xmm2
+; SSE2-SSSE3-NEXT: movdqa 32(%rdi), %xmm3
+; SSE2-SSSE3-NEXT: movdqa 48(%rdi), %xmm6
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm4 = [65535,65535]
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm0 = [2147483648,2147483648]
+; SSE2-SSSE3-NEXT: movdqa %xmm2, %xmm1
+; SSE2-SSSE3-NEXT: pxor %xmm0, %xmm1
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm1[1,1,3,3]
+; SSE2-SSSE3-NEXT: pxor %xmm7, %xmm7
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm7, %xmm9
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm5 = [2147549183,2147549183]
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm10
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm1, %xmm10
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
+; SSE2-SSSE3-NEXT: pand %xmm9, %xmm11
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm10[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm11, %xmm1
+; SSE2-SSSE3-NEXT: pand %xmm1, %xmm2
+; SSE2-SSSE3-NEXT: pandn %xmm4, %xmm1
+; SSE2-SSSE3-NEXT: por %xmm2, %xmm1
+; SSE2-SSSE3-NEXT: movdqa %xmm8, %xmm2
+; SSE2-SSSE3-NEXT: pxor %xmm0, %xmm2
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm2[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm7, %xmm9
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm10
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm2, %xmm10
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
+; SSE2-SSSE3-NEXT: pand %xmm9, %xmm11
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm10[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm11, %xmm2
+; SSE2-SSSE3-NEXT: pand %xmm2, %xmm8
+; SSE2-SSSE3-NEXT: pandn %xmm4, %xmm2
+; SSE2-SSSE3-NEXT: por %xmm8, %xmm2
+; SSE2-SSSE3-NEXT: movdqa %xmm6, %xmm8
+; SSE2-SSSE3-NEXT: pxor %xmm0, %xmm8
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm8[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm7, %xmm9
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm10
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm8, %xmm10
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
+; SSE2-SSSE3-NEXT: pand %xmm9, %xmm11
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm10[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm11, %xmm8
+; SSE2-SSSE3-NEXT: pand %xmm8, %xmm6
+; SSE2-SSSE3-NEXT: pandn %xmm4, %xmm8
+; SSE2-SSSE3-NEXT: por %xmm6, %xmm8
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm6
+; SSE2-SSSE3-NEXT: pxor %xmm0, %xmm6
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm6[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm7, %xmm9
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm6, %xmm5
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm5[0,0,2,2]
+; SSE2-SSSE3-NEXT: pand %xmm9, %xmm6
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm5[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm6, %xmm5
+; SSE2-SSSE3-NEXT: pand %xmm5, %xmm3
+; SSE2-SSSE3-NEXT: pandn %xmm4, %xmm5
+; SSE2-SSSE3-NEXT: por %xmm3, %xmm5
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm3
+; SSE2-SSSE3-NEXT: pxor %xmm0, %xmm3
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm4
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm0, %xmm4
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm4[0,0,2,2]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm0, %xmm3
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm3[1,1,3,3]
+; SSE2-SSSE3-NEXT: pand %xmm6, %xmm7
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm4[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm7, %xmm3
+; SSE2-SSSE3-NEXT: pand %xmm5, %xmm3
+; SSE2-SSSE3-NEXT: movdqa %xmm8, %xmm4
+; SSE2-SSSE3-NEXT: pxor %xmm0, %xmm4
+; SSE2-SSSE3-NEXT: movdqa %xmm4, %xmm5
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm0, %xmm5
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm5[0,0,2,2]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm0, %xmm4
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm4[1,1,3,3]
+; SSE2-SSSE3-NEXT: pand %xmm6, %xmm7
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm5[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm7, %xmm4
+; SSE2-SSSE3-NEXT: pand %xmm8, %xmm4
+; SSE2-SSSE3-NEXT: movdqa %xmm2, %xmm5
+; SSE2-SSSE3-NEXT: pxor %xmm0, %xmm5
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm6
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm0, %xmm6
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm6[0,0,2,2]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm0, %xmm5
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm5[1,1,3,3]
+; SSE2-SSSE3-NEXT: pand %xmm7, %xmm5
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm5, %xmm6
+; SSE2-SSSE3-NEXT: pand %xmm2, %xmm6
+; SSE2-SSSE3-NEXT: movdqa %xmm1, %xmm2
+; SSE2-SSSE3-NEXT: pxor %xmm0, %xmm2
+; SSE2-SSSE3-NEXT: movdqa %xmm2, %xmm5
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm0, %xmm5
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm5[0,0,2,2]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm0, %xmm2
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm2[1,1,3,3]
+; SSE2-SSSE3-NEXT: pand %xmm7, %xmm0
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm5[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm0, %xmm2
+; SSE2-SSSE3-NEXT: pand %xmm1, %xmm2
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm2[0,2,2,3]
+; SSE2-SSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm6[0,2,2,3]
+; SSE2-SSSE3-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
+; SSE2-SSSE3-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm4[0,2,2,3]
+; SSE2-SSSE3-NEXT: pshuflw {{.*#+}} xmm2 = xmm0[0,1,0,2,4,5,6,7]
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm3[0,2,2,3]
+; SSE2-SSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,1,0,2,4,5,6,7]
+; SSE2-SSSE3-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
+; SSE2-SSSE3-NEXT: movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_packus_v8i64_v8i16:
; SSE41: # %bb.0:
@@ -2587,109 +2063,57 @@ define <8 x i16> @trunc_packus_v8i32_v8i16(<8 x i32> %a0) {
}
define <16 x i16> @trunc_packus_v16i32_v16i16(ptr %p0) "min-legal-vector-width"="256" {
-; SSE2-LABEL: trunc_packus_v16i32_v16i16:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa (%rdi), %xmm1
-; SSE2-NEXT: movdqa 16(%rdi), %xmm3
-; SSE2-NEXT: movdqa 32(%rdi), %xmm0
-; SSE2-NEXT: movdqa 48(%rdi), %xmm4
-; SSE2-NEXT: movdqa {{.*#+}} xmm5 = [65535,65535,65535,65535]
-; SSE2-NEXT: movdqa %xmm5, %xmm2
-; SSE2-NEXT: pcmpgtd %xmm3, %xmm2
-; SSE2-NEXT: pand %xmm2, %xmm3
-; SSE2-NEXT: pandn %xmm5, %xmm2
-; SSE2-NEXT: por %xmm3, %xmm2
-; SSE2-NEXT: movdqa %xmm5, %xmm3
-; SSE2-NEXT: pcmpgtd %xmm1, %xmm3
-; SSE2-NEXT: pand %xmm3, %xmm1
-; SSE2-NEXT: pandn %xmm5, %xmm3
-; SSE2-NEXT: por %xmm1, %xmm3
-; SSE2-NEXT: movdqa %xmm5, %xmm6
-; SSE2-NEXT: pcmpgtd %xmm4, %xmm6
-; SSE2-NEXT: pand %xmm6, %xmm4
-; SSE2-NEXT: pandn %xmm5, %xmm6
-; SSE2-NEXT: por %xmm4, %xmm6
-; SSE2-NEXT: movdqa %xmm5, %xmm4
-; SSE2-NEXT: pcmpgtd %xmm0, %xmm4
-; SSE2-NEXT: pand %xmm4, %xmm0
-; SSE2-NEXT: pandn %xmm5, %xmm4
-; SSE2-NEXT: por %xmm0, %xmm4
-; SSE2-NEXT: pxor %xmm5, %xmm5
-; SSE2-NEXT: movdqa %xmm4, %xmm1
-; SSE2-NEXT: pcmpgtd %xmm5, %xmm1
-; SSE2-NEXT: pand %xmm4, %xmm1
-; SSE2-NEXT: movdqa %xmm6, %xmm4
-; SSE2-NEXT: pcmpgtd %xmm5, %xmm4
-; SSE2-NEXT: pand %xmm6, %xmm4
-; SSE2-NEXT: movdqa %xmm3, %xmm0
-; SSE2-NEXT: pcmpgtd %xmm5, %xmm0
-; SSE2-NEXT: pand %xmm3, %xmm0
-; SSE2-NEXT: movdqa %xmm2, %xmm3
-; SSE2-NEXT: pcmpgtd %xmm5, %xmm3
-; SSE2-NEXT: pand %xmm2, %xmm3
-; SSE2-NEXT: pslld $16, %xmm3
-; SSE2-NEXT: psrad $16, %xmm3
-; SSE2-NEXT: pslld $16, %xmm0
-; SSE2-NEXT: psrad $16, %xmm0
-; SSE2-NEXT: packssdw %xmm3, %xmm0
-; SSE2-NEXT: pslld $16, %xmm4
-; SSE2-NEXT: psrad $16, %xmm4
-; SSE2-NEXT: pslld $16, %xmm1
-; SSE2-NEXT: psrad $16, %xmm1
-; SSE2-NEXT: packssdw %xmm4, %xmm1
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_packus_v16i32_v16i16:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa (%rdi), %xmm1
-; SSSE3-NEXT: movdqa 16(%rdi), %xmm3
-; SSSE3-NEXT: movdqa 32(%rdi), %xmm0
-; SSSE3-NEXT: movdqa 48(%rdi), %xmm4
-; SSSE3-NEXT: movdqa {{.*#+}} xmm5 = [65535,65535,65535,65535]
-; SSSE3-NEXT: movdqa %xmm5, %xmm2
-; SSSE3-NEXT: pcmpgtd %xmm3, %xmm2
-; SSSE3-NEXT: pand %xmm2, %xmm3
-; SSSE3-NEXT: pandn %xmm5, %xmm2
-; SSSE3-NEXT: por %xmm3, %xmm2
-; SSSE3-NEXT: movdqa %xmm5, %xmm3
-; SSSE3-NEXT: pcmpgtd %xmm1, %xmm3
-; SSSE3-NEXT: pand %xmm3, %xmm1
-; SSSE3-NEXT: pandn %xmm5, %xmm3
-; SSSE3-NEXT: por %xmm1, %xmm3
-; SSSE3-NEXT: movdqa %xmm5, %xmm6
-; SSSE3-NEXT: pcmpgtd %xmm4, %xmm6
-; SSSE3-NEXT: pand %xmm6, %xmm4
-; SSSE3-NEXT: pandn %xmm5, %xmm6
-; SSSE3-NEXT: por %xmm4, %xmm6
-; SSSE3-NEXT: movdqa %xmm5, %xmm4
-; SSSE3-NEXT: pcmpgtd %xmm0, %xmm4
-; SSSE3-NEXT: pand %xmm4, %xmm0
-; SSSE3-NEXT: pandn %xmm5, %xmm4
-; SSSE3-NEXT: por %xmm0, %xmm4
-; SSSE3-NEXT: pxor %xmm5, %xmm5
-; SSSE3-NEXT: movdqa %xmm4, %xmm1
-; SSSE3-NEXT: pcmpgtd %xmm5, %xmm1
-; SSSE3-NEXT: pand %xmm4, %xmm1
-; SSSE3-NEXT: movdqa %xmm6, %xmm4
-; SSSE3-NEXT: pcmpgtd %xmm5, %xmm4
-; SSSE3-NEXT: pand %xmm6, %xmm4
-; SSSE3-NEXT: movdqa %xmm3, %xmm0
-; SSSE3-NEXT: pcmpgtd %xmm5, %xmm0
-; SSSE3-NEXT: pand %xmm3, %xmm0
-; SSSE3-NEXT: movdqa %xmm2, %xmm3
-; SSSE3-NEXT: pcmpgtd %xmm5, %xmm3
-; SSSE3-NEXT: pand %xmm2, %xmm3
-; SSSE3-NEXT: pslld $16, %xmm3
-; SSSE3-NEXT: psrad $16, %xmm3
-; SSSE3-NEXT: pslld $16, %xmm0
-; SSSE3-NEXT: psrad $16, %xmm0
-; SSSE3-NEXT: packssdw %xmm3, %xmm0
-; SSSE3-NEXT: pslld $16, %xmm4
-; SSSE3-NEXT: psrad $16, %xmm4
-; SSSE3-NEXT: pslld $16, %xmm1
-; SSSE3-NEXT: psrad $16, %xmm1
-; SSSE3-NEXT: packssdw %xmm4, %xmm1
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc_packus_v16i32_v16i16:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa (%rdi), %xmm1
+; SSE2-SSSE3-NEXT: movdqa 16(%rdi), %xmm3
+; SSE2-SSSE3-NEXT: movdqa 32(%rdi), %xmm0
+; SSE2-SSSE3-NEXT: movdqa 48(%rdi), %xmm4
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm5 = [65535,65535,65535,65535]
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm2
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm3, %xmm2
+; SSE2-SSSE3-NEXT: pand %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: pandn %xmm5, %xmm2
+; SSE2-SSSE3-NEXT: por %xmm3, %xmm2
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm3
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm1, %xmm3
+; SSE2-SSSE3-NEXT: pand %xmm3, %xmm1
+; SSE2-SSSE3-NEXT: pandn %xmm5, %xmm3
+; SSE2-SSSE3-NEXT: por %xmm1, %xmm3
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm6
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm4, %xmm6
+; SSE2-SSSE3-NEXT: pand %xmm6, %xmm4
+; SSE2-SSSE3-NEXT: pandn %xmm5, %xmm6
+; SSE2-SSSE3-NEXT: por %xmm4, %xmm6
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm4
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm0, %xmm4
+; SSE2-SSSE3-NEXT: pand %xmm4, %xmm0
+; SSE2-SSSE3-NEXT: pandn %xmm5, %xmm4
+; SSE2-SSSE3-NEXT: por %xmm0, %xmm4
+; SSE2-SSSE3-NEXT: pxor %xmm5, %xmm5
+; SSE2-SSSE3-NEXT: movdqa %xmm4, %xmm1
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm5, %xmm1
+; SSE2-SSSE3-NEXT: pand %xmm4, %xmm1
+; SSE2-SSSE3-NEXT: movdqa %xmm6, %xmm4
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm5, %xmm4
+; SSE2-SSSE3-NEXT: pand %xmm6, %xmm4
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm0
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm5, %xmm0
+; SSE2-SSSE3-NEXT: pand %xmm3, %xmm0
+; SSE2-SSSE3-NEXT: movdqa %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm5, %xmm3
+; SSE2-SSSE3-NEXT: pand %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: pslld $16, %xmm3
+; SSE2-SSSE3-NEXT: psrad $16, %xmm3
+; SSE2-SSSE3-NEXT: pslld $16, %xmm0
+; SSE2-SSSE3-NEXT: psrad $16, %xmm0
+; SSE2-SSSE3-NEXT: packssdw %xmm3, %xmm0
+; SSE2-SSSE3-NEXT: pslld $16, %xmm4
+; SSE2-SSSE3-NEXT: psrad $16, %xmm4
+; SSE2-SSSE3-NEXT: pslld $16, %xmm1
+; SSE2-SSSE3-NEXT: psrad $16, %xmm1
+; SSE2-SSSE3-NEXT: packssdw %xmm4, %xmm1
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_packus_v16i32_v16i16:
; SSE41: # %bb.0:
@@ -3596,227 +3020,116 @@ define void @trunc_packus_v4i64_v4i8_store(<4 x i64> %a0, ptr%p1) {
}
define <8 x i8> @trunc_packus_v8i64_v8i8(ptr %p0) "min-legal-vector-width"="256" {
-; SSE2-LABEL: trunc_packus_v8i64_v8i8:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa (%rdi), %xmm5
-; SSE2-NEXT: movdqa 16(%rdi), %xmm0
-; SSE2-NEXT: movdqa 32(%rdi), %xmm3
-; SSE2-NEXT: movdqa 48(%rdi), %xmm8
-; SSE2-NEXT: movdqa {{.*#+}} xmm4 = [255,255]
-; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [2147483648,2147483648]
-; SSE2-NEXT: movdqa %xmm3, %xmm2
-; SSE2-NEXT: pxor %xmm1, %xmm2
-; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm2[1,1,3,3]
-; SSE2-NEXT: pxor %xmm7, %xmm7
-; SSE2-NEXT: pcmpeqd %xmm7, %xmm9
-; SSE2-NEXT: movdqa {{.*#+}} xmm6 = [2147483903,2147483903]
-; SSE2-NEXT: movdqa %xmm6, %xmm10
-; SSE2-NEXT: pcmpgtd %xmm2, %xmm10
-; SSE2-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
-; SSE2-NEXT: pand %xmm9, %xmm11
-; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm10[1,1,3,3]
-; SSE2-NEXT: por %xmm11, %xmm2
-; SSE2-NEXT: pand %xmm2, %xmm3
-; SSE2-NEXT: pandn %xmm4, %xmm2
-; SSE2-NEXT: por %xmm3, %xmm2
-; SSE2-NEXT: movdqa %xmm8, %xmm3
-; SSE2-NEXT: pxor %xmm1, %xmm3
-; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm3[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm7, %xmm9
-; SSE2-NEXT: movdqa %xmm6, %xmm10
-; SSE2-NEXT: pcmpgtd %xmm3, %xmm10
-; SSE2-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
-; SSE2-NEXT: pand %xmm9, %xmm11
-; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm10[1,1,3,3]
-; SSE2-NEXT: por %xmm11, %xmm3
-; SSE2-NEXT: pand %xmm3, %xmm8
-; SSE2-NEXT: pandn %xmm4, %xmm3
-; SSE2-NEXT: por %xmm8, %xmm3
-; SSE2-NEXT: movdqa %xmm5, %xmm8
-; SSE2-NEXT: pxor %xmm1, %xmm8
-; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm8[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm7, %xmm9
-; SSE2-NEXT: movdqa %xmm6, %xmm10
-; SSE2-NEXT: pcmpgtd %xmm8, %xmm10
-; SSE2-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
-; SSE2-NEXT: pand %xmm9, %xmm11
-; SSE2-NEXT: pshufd {{.*#+}} xmm8 = xmm10[1,1,3,3]
-; SSE2-NEXT: por %xmm11, %xmm8
-; SSE2-NEXT: pand %xmm8, %xmm5
-; SSE2-NEXT: pandn %xmm4, %xmm8
-; SSE2-NEXT: por %xmm5, %xmm8
-; SSE2-NEXT: movdqa %xmm0, %xmm5
-; SSE2-NEXT: pxor %xmm1, %xmm5
-; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm5[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm7, %xmm9
-; SSE2-NEXT: pcmpgtd %xmm5, %xmm6
-; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm6[0,0,2,2]
-; SSE2-NEXT: pand %xmm9, %xmm5
-; SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
-; SSE2-NEXT: por %xmm5, %xmm6
-; SSE2-NEXT: pand %xmm6, %xmm0
-; SSE2-NEXT: pandn %xmm4, %xmm6
-; SSE2-NEXT: por %xmm0, %xmm6
-; SSE2-NEXT: movdqa %xmm6, %xmm0
-; SSE2-NEXT: pxor %xmm1, %xmm0
-; SSE2-NEXT: movdqa %xmm0, %xmm4
-; SSE2-NEXT: pcmpgtd %xmm1, %xmm4
-; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm4[0,0,2,2]
-; SSE2-NEXT: pcmpeqd %xmm1, %xmm0
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; SSE2-NEXT: pand %xmm5, %xmm0
-; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
-; SSE2-NEXT: por %xmm0, %xmm4
-; SSE2-NEXT: pand %xmm6, %xmm4
-; SSE2-NEXT: movdqa %xmm8, %xmm0
-; SSE2-NEXT: pxor %xmm1, %xmm0
-; SSE2-NEXT: movdqa %xmm0, %xmm5
-; SSE2-NEXT: pcmpgtd %xmm1, %xmm5
-; SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm5[0,0,2,2]
-; SSE2-NEXT: pcmpeqd %xmm1, %xmm0
-; SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm0[1,1,3,3]
-; SSE2-NEXT: pand %xmm6, %xmm7
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm5[1,1,3,3]
-; SSE2-NEXT: por %xmm7, %xmm0
-; SSE2-NEXT: pand %xmm8, %xmm0
-; SSE2-NEXT: packuswb %xmm4, %xmm0
-; SSE2-NEXT: movdqa %xmm3, %xmm4
-; SSE2-NEXT: pxor %xmm1, %xmm4
-; SSE2-NEXT: movdqa %xmm4, %xmm5
-; SSE2-NEXT: pcmpgtd %xmm1, %xmm5
-; SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm5[0,0,2,2]
-; SSE2-NEXT: pcmpeqd %xmm1, %xmm4
-; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
-; SSE2-NEXT: pand %xmm6, %xmm4
-; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm5[1,1,3,3]
-; SSE2-NEXT: por %xmm4, %xmm5
-; SSE2-NEXT: pand %xmm3, %xmm5
-; SSE2-NEXT: movdqa %xmm2, %xmm3
-; SSE2-NEXT: pxor %xmm1, %xmm3
-; SSE2-NEXT: movdqa %xmm3, %xmm4
-; SSE2-NEXT: pcmpgtd %xmm1, %xmm4
-; SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm4[0,0,2,2]
-; SSE2-NEXT: pcmpeqd %xmm1, %xmm3
-; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm3[1,1,3,3]
-; SSE2-NEXT: pand %xmm6, %xmm1
-; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm4[1,1,3,3]
-; SSE2-NEXT: por %xmm1, %xmm3
-; SSE2-NEXT: pand %xmm2, %xmm3
-; SSE2-NEXT: packuswb %xmm5, %xmm3
-; SSE2-NEXT: packuswb %xmm3, %xmm0
-; SSE2-NEXT: packuswb %xmm0, %xmm0
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_packus_v8i64_v8i8:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa (%rdi), %xmm5
-; SSSE3-NEXT: movdqa 16(%rdi), %xmm0
-; SSSE3-NEXT: movdqa 32(%rdi), %xmm3
-; SSSE3-NEXT: movdqa 48(%rdi), %xmm8
-; SSSE3-NEXT: movdqa {{.*#+}} xmm4 = [255,255]
-; SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [2147483648,2147483648]
-; SSSE3-NEXT: movdqa %xmm3, %xmm2
-; SSSE3-NEXT: pxor %xmm1, %xmm2
-; SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm2[1,1,3,3]
-; SSSE3-NEXT: pxor %xmm7, %xmm7
-; SSSE3-NEXT: pcmpeqd %xmm7, %xmm9
-; SSSE3-NEXT: movdqa {{.*#+}} xmm6 = [2147483903,2147483903]
-; SSSE3-NEXT: movdqa %xmm6, %xmm10
-; SSSE3-NEXT: pcmpgtd %xmm2, %xmm10
-; SSSE3-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
-; SSSE3-NEXT: pand %xmm9, %xmm11
-; SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm10[1,1,3,3]
-; SSSE3-NEXT: por %xmm11, %xmm2
-; SSSE3-NEXT: pand %xmm2, %xmm3
-; SSSE3-NEXT: pandn %xmm4, %xmm2
-; SSSE3-NEXT: por %xmm3, %xmm2
-; SSSE3-NEXT: movdqa %xmm8, %xmm3
-; SSSE3-NEXT: pxor %xmm1, %xmm3
-; SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm3[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm7, %xmm9
-; SSSE3-NEXT: movdqa %xmm6, %xmm10
-; SSSE3-NEXT: pcmpgtd %xmm3, %xmm10
-; SSSE3-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
-; SSSE3-NEXT: pand %xmm9, %xmm11
-; SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm10[1,1,3,3]
-; SSSE3-NEXT: por %xmm11, %xmm3
-; SSSE3-NEXT: pand %xmm3, %xmm8
-; SSSE3-NEXT: pandn %xmm4, %xmm3
-; SSSE3-NEXT: por %xmm8, %xmm3
-; SSSE3-NEXT: movdqa %xmm5, %xmm8
-; SSSE3-NEXT: pxor %xmm1, %xmm8
-; SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm8[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm7, %xmm9
-; SSSE3-NEXT: movdqa %xmm6, %xmm10
-; SSSE3-NEXT: pcmpgtd %xmm8, %xmm10
-; SSSE3-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
-; SSSE3-NEXT: pand %xmm9, %xmm11
-; SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm10[1,1,3,3]
-; SSSE3-NEXT: por %xmm11, %xmm8
-; SSSE3-NEXT: pand %xmm8, %xmm5
-; SSSE3-NEXT: pandn %xmm4, %xmm8
-; SSSE3-NEXT: por %xmm5, %xmm8
-; SSSE3-NEXT: movdqa %xmm0, %xmm5
-; SSSE3-NEXT: pxor %xmm1, %xmm5
-; SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm5[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm7, %xmm9
-; SSSE3-NEXT: pcmpgtd %xmm5, %xmm6
-; SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm6[0,0,2,2]
-; SSSE3-NEXT: pand %xmm9, %xmm5
-; SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
-; SSSE3-NEXT: por %xmm5, %xmm6
-; SSSE3-NEXT: pand %xmm6, %xmm0
-; SSSE3-NEXT: pandn %xmm4, %xmm6
-; SSSE3-NEXT: por %xmm0, %xmm6
-; SSSE3-NEXT: movdqa %xmm6, %xmm0
-; SSSE3-NEXT: pxor %xmm1, %xmm0
-; SSSE3-NEXT: movdqa %xmm0, %xmm4
-; SSSE3-NEXT: pcmpgtd %xmm1, %xmm4
-; SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm4[0,0,2,2]
-; SSSE3-NEXT: pcmpeqd %xmm1, %xmm0
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; SSSE3-NEXT: pand %xmm5, %xmm0
-; SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
-; SSSE3-NEXT: por %xmm0, %xmm4
-; SSSE3-NEXT: pand %xmm6, %xmm4
-; SSSE3-NEXT: movdqa %xmm8, %xmm0
-; SSSE3-NEXT: pxor %xmm1, %xmm0
-; SSSE3-NEXT: movdqa %xmm0, %xmm5
-; SSSE3-NEXT: pcmpgtd %xmm1, %xmm5
-; SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm5[0,0,2,2]
-; SSSE3-NEXT: pcmpeqd %xmm1, %xmm0
-; SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm0[1,1,3,3]
-; SSSE3-NEXT: pand %xmm6, %xmm7
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm5[1,1,3,3]
-; SSSE3-NEXT: por %xmm7, %xmm0
-; SSSE3-NEXT: pand %xmm8, %xmm0
-; SSSE3-NEXT: packuswb %xmm4, %xmm0
-; SSSE3-NEXT: movdqa %xmm3, %xmm4
-; SSSE3-NEXT: pxor %xmm1, %xmm4
-; SSSE3-NEXT: movdqa %xmm4, %xmm5
-; SSSE3-NEXT: pcmpgtd %xmm1, %xmm5
-; SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm5[0,0,2,2]
-; SSSE3-NEXT: pcmpeqd %xmm1, %xmm4
-; SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
-; SSSE3-NEXT: pand %xmm6, %xmm4
-; SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm5[1,1,3,3]
-; SSSE3-NEXT: por %xmm4, %xmm5
-; SSSE3-NEXT: pand %xmm3, %xmm5
-; SSSE3-NEXT: movdqa %xmm2, %xmm3
-; SSSE3-NEXT: pxor %xmm1, %xmm3
-; SSSE3-NEXT: movdqa %xmm3, %xmm4
-; SSSE3-NEXT: pcmpgtd %xmm1, %xmm4
-; SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm4[0,0,2,2]
-; SSSE3-NEXT: pcmpeqd %xmm1, %xmm3
-; SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm3[1,1,3,3]
-; SSSE3-NEXT: pand %xmm6, %xmm1
-; SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm4[1,1,3,3]
-; SSSE3-NEXT: por %xmm1, %xmm3
-; SSSE3-NEXT: pand %xmm2, %xmm3
-; SSSE3-NEXT: packuswb %xmm5, %xmm3
-; SSSE3-NEXT: packuswb %xmm3, %xmm0
-; SSSE3-NEXT: packuswb %xmm0, %xmm0
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc_packus_v8i64_v8i8:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa (%rdi), %xmm5
+; SSE2-SSSE3-NEXT: movdqa 16(%rdi), %xmm0
+; SSE2-SSSE3-NEXT: movdqa 32(%rdi), %xmm3
+; SSE2-SSSE3-NEXT: movdqa 48(%rdi), %xmm8
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm4 = [255,255]
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [2147483648,2147483648]
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm2
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm2
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm2[1,1,3,3]
+; SSE2-SSSE3-NEXT: pxor %xmm7, %xmm7
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm7, %xmm9
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm6 = [2147483903,2147483903]
+; SSE2-SSSE3-NEXT: movdqa %xmm6, %xmm10
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm2, %xmm10
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
+; SSE2-SSSE3-NEXT: pand %xmm9, %xmm11
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm10[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm11, %xmm2
+; SSE2-SSSE3-NEXT: pand %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: pandn %xmm4, %xmm2
+; SSE2-SSSE3-NEXT: por %xmm3, %xmm2
+; SSE2-SSSE3-NEXT: movdqa %xmm8, %xmm3
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm3
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm3[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm7, %xmm9
+; SSE2-SSSE3-NEXT: movdqa %xmm6, %xmm10
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm3, %xmm10
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
+; SSE2-SSSE3-NEXT: pand %xmm9, %xmm11
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm10[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm11, %xmm3
+; SSE2-SSSE3-NEXT: pand %xmm3, %xmm8
+; SSE2-SSSE3-NEXT: pandn %xmm4, %xmm3
+; SSE2-SSSE3-NEXT: por %xmm8, %xmm3
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm8
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm8
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm8[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm7, %xmm9
+; SSE2-SSSE3-NEXT: movdqa %xmm6, %xmm10
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm8, %xmm10
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
+; SSE2-SSSE3-NEXT: pand %xmm9, %xmm11
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm10[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm11, %xmm8
+; SSE2-SSSE3-NEXT: pand %xmm8, %xmm5
+; SSE2-SSSE3-NEXT: pandn %xmm4, %xmm8
+; SSE2-SSSE3-NEXT: por %xmm5, %xmm8
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm5
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm5
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm5[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm7, %xmm9
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm5, %xmm6
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm6[0,0,2,2]
+; SSE2-SSSE3-NEXT: pand %xmm9, %xmm5
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm5, %xmm6
+; SSE2-SSSE3-NEXT: pand %xmm6, %xmm0
+; SSE2-SSSE3-NEXT: pandn %xmm4, %xmm6
+; SSE2-SSSE3-NEXT: por %xmm0, %xmm6
+; SSE2-SSSE3-NEXT: movdqa %xmm6, %xmm0
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm4
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm1, %xmm4
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm4[0,0,2,2]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; SSE2-SSSE3-NEXT: pand %xmm5, %xmm0
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm0, %xmm4
+; SSE2-SSSE3-NEXT: pand %xmm6, %xmm4
+; SSE2-SSSE3-NEXT: movdqa %xmm8, %xmm0
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm5
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm1, %xmm5
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm5[0,0,2,2]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm0[1,1,3,3]
+; SSE2-SSSE3-NEXT: pand %xmm6, %xmm7
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm5[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm7, %xmm0
+; SSE2-SSSE3-NEXT: pand %xmm8, %xmm0
+; SSE2-SSSE3-NEXT: packuswb %xmm4, %xmm0
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm4
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm4
+; SSE2-SSSE3-NEXT: movdqa %xmm4, %xmm5
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm1, %xmm5
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm5[0,0,2,2]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm1, %xmm4
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
+; SSE2-SSSE3-NEXT: pand %xmm6, %xmm4
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm5[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm4, %xmm5
+; SSE2-SSSE3-NEXT: pand %xmm3, %xmm5
+; SSE2-SSSE3-NEXT: movdqa %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm3
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm4
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm1, %xmm4
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm4[0,0,2,2]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm1, %xmm3
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm3[1,1,3,3]
+; SSE2-SSSE3-NEXT: pand %xmm6, %xmm1
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm4[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm1, %xmm3
+; SSE2-SSSE3-NEXT: pand %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: packuswb %xmm5, %xmm3
+; SSE2-SSSE3-NEXT: packuswb %xmm3, %xmm0
+; SSE2-SSSE3-NEXT: packuswb %xmm0, %xmm0
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_packus_v8i64_v8i8:
; SSE41: # %bb.0:
@@ -3997,229 +3310,117 @@ define <8 x i8> @trunc_packus_v8i64_v8i8(ptr %p0) "min-legal-vector-width"="256"
}
define void @trunc_packus_v8i64_v8i8_store(ptr %p0, ptr%p1) "min-legal-vector-width"="256" {
-; SSE2-LABEL: trunc_packus_v8i64_v8i8_store:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa (%rdi), %xmm5
-; SSE2-NEXT: movdqa 16(%rdi), %xmm3
-; SSE2-NEXT: movdqa 32(%rdi), %xmm2
-; SSE2-NEXT: movdqa 48(%rdi), %xmm8
-; SSE2-NEXT: movdqa {{.*#+}} xmm4 = [255,255]
-; SSE2-NEXT: movdqa {{.*#+}} xmm0 = [2147483648,2147483648]
-; SSE2-NEXT: movdqa %xmm2, %xmm1
-; SSE2-NEXT: pxor %xmm0, %xmm1
-; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm1[1,1,3,3]
-; SSE2-NEXT: pxor %xmm7, %xmm7
-; SSE2-NEXT: pcmpeqd %xmm7, %xmm9
-; SSE2-NEXT: movdqa {{.*#+}} xmm6 = [2147483903,2147483903]
-; SSE2-NEXT: movdqa %xmm6, %xmm10
-; SSE2-NEXT: pcmpgtd %xmm1, %xmm10
-; SSE2-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
-; SSE2-NEXT: pand %xmm9, %xmm11
-; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm10[1,1,3,3]
-; SSE2-NEXT: por %xmm11, %xmm1
-; SSE2-NEXT: pand %xmm1, %xmm2
-; SSE2-NEXT: pandn %xmm4, %xmm1
-; SSE2-NEXT: por %xmm2, %xmm1
-; SSE2-NEXT: movdqa %xmm8, %xmm2
-; SSE2-NEXT: pxor %xmm0, %xmm2
-; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm2[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm7, %xmm9
-; SSE2-NEXT: movdqa %xmm6, %xmm10
-; SSE2-NEXT: pcmpgtd %xmm2, %xmm10
-; SSE2-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
-; SSE2-NEXT: pand %xmm9, %xmm11
-; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm10[1,1,3,3]
-; SSE2-NEXT: por %xmm11, %xmm2
-; SSE2-NEXT: pand %xmm2, %xmm8
-; SSE2-NEXT: pandn %xmm4, %xmm2
-; SSE2-NEXT: por %xmm8, %xmm2
-; SSE2-NEXT: movdqa %xmm5, %xmm8
-; SSE2-NEXT: pxor %xmm0, %xmm8
-; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm8[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm7, %xmm9
-; SSE2-NEXT: movdqa %xmm6, %xmm10
-; SSE2-NEXT: pcmpgtd %xmm8, %xmm10
-; SSE2-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
-; SSE2-NEXT: pand %xmm9, %xmm11
-; SSE2-NEXT: pshufd {{.*#+}} xmm8 = xmm10[1,1,3,3]
-; SSE2-NEXT: por %xmm11, %xmm8
-; SSE2-NEXT: pand %xmm8, %xmm5
-; SSE2-NEXT: pandn %xmm4, %xmm8
-; SSE2-NEXT: por %xmm5, %xmm8
-; SSE2-NEXT: movdqa %xmm3, %xmm5
-; SSE2-NEXT: pxor %xmm0, %xmm5
-; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm5[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm7, %xmm9
-; SSE2-NEXT: pcmpgtd %xmm5, %xmm6
-; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm6[0,0,2,2]
-; SSE2-NEXT: pand %xmm9, %xmm5
-; SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
-; SSE2-NEXT: por %xmm5, %xmm6
-; SSE2-NEXT: pand %xmm6, %xmm3
-; SSE2-NEXT: pandn %xmm4, %xmm6
-; SSE2-NEXT: por %xmm3, %xmm6
-; SSE2-NEXT: movdqa %xmm6, %xmm3
-; SSE2-NEXT: pxor %xmm0, %xmm3
-; SSE2-NEXT: movdqa %xmm3, %xmm4
-; SSE2-NEXT: pcmpgtd %xmm0, %xmm4
-; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm4[0,0,2,2]
-; SSE2-NEXT: pcmpeqd %xmm0, %xmm3
-; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
-; SSE2-NEXT: pand %xmm5, %xmm3
-; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
-; SSE2-NEXT: por %xmm3, %xmm4
-; SSE2-NEXT: pand %xmm6, %xmm4
-; SSE2-NEXT: movdqa %xmm8, %xmm3
-; SSE2-NEXT: pxor %xmm0, %xmm3
-; SSE2-NEXT: movdqa %xmm3, %xmm5
-; SSE2-NEXT: pcmpgtd %xmm0, %xmm5
-; SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm5[0,0,2,2]
-; SSE2-NEXT: pcmpeqd %xmm0, %xmm3
-; SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm3[1,1,3,3]
-; SSE2-NEXT: pand %xmm6, %xmm7
-; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm5[1,1,3,3]
-; SSE2-NEXT: por %xmm7, %xmm3
-; SSE2-NEXT: pand %xmm8, %xmm3
-; SSE2-NEXT: packuswb %xmm4, %xmm3
-; SSE2-NEXT: movdqa %xmm2, %xmm4
-; SSE2-NEXT: pxor %xmm0, %xmm4
-; SSE2-NEXT: movdqa %xmm4, %xmm5
-; SSE2-NEXT: pcmpgtd %xmm0, %xmm5
-; SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm5[0,0,2,2]
-; SSE2-NEXT: pcmpeqd %xmm0, %xmm4
-; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
-; SSE2-NEXT: pand %xmm6, %xmm4
-; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm5[1,1,3,3]
-; SSE2-NEXT: por %xmm4, %xmm5
-; SSE2-NEXT: pand %xmm2, %xmm5
-; SSE2-NEXT: movdqa %xmm1, %xmm2
-; SSE2-NEXT: pxor %xmm0, %xmm2
-; SSE2-NEXT: movdqa %xmm2, %xmm4
-; SSE2-NEXT: pcmpgtd %xmm0, %xmm4
-; SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm4[0,0,2,2]
-; SSE2-NEXT: pcmpeqd %xmm0, %xmm2
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[1,1,3,3]
-; SSE2-NEXT: pand %xmm6, %xmm0
-; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm4[1,1,3,3]
-; SSE2-NEXT: por %xmm0, %xmm2
-; SSE2-NEXT: pand %xmm1, %xmm2
-; SSE2-NEXT: packuswb %xmm5, %xmm2
-; SSE2-NEXT: packuswb %xmm2, %xmm3
-; SSE2-NEXT: packuswb %xmm3, %xmm3
-; SSE2-NEXT: movq %xmm3, (%rsi)
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_packus_v8i64_v8i8_store:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa (%rdi), %xmm5
-; SSSE3-NEXT: movdqa 16(%rdi), %xmm3
-; SSSE3-NEXT: movdqa 32(%rdi), %xmm2
-; SSSE3-NEXT: movdqa 48(%rdi), %xmm8
-; SSSE3-NEXT: movdqa {{.*#+}} xmm4 = [255,255]
-; SSSE3-NEXT: movdqa {{.*#+}} xmm0 = [2147483648,2147483648]
-; SSSE3-NEXT: movdqa %xmm2, %xmm1
-; SSSE3-NEXT: pxor %xmm0, %xmm1
-; SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm1[1,1,3,3]
-; SSSE3-NEXT: pxor %xmm7, %xmm7
-; SSSE3-NEXT: pcmpeqd %xmm7, %xmm9
-; SSSE3-NEXT: movdqa {{.*#+}} xmm6 = [2147483903,2147483903]
-; SSSE3-NEXT: movdqa %xmm6, %xmm10
-; SSSE3-NEXT: pcmpgtd %xmm1, %xmm10
-; SSSE3-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
-; SSSE3-NEXT: pand %xmm9, %xmm11
-; SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm10[1,1,3,3]
-; SSSE3-NEXT: por %xmm11, %xmm1
-; SSSE3-NEXT: pand %xmm1, %xmm2
-; SSSE3-NEXT: pandn %xmm4, %xmm1
-; SSSE3-NEXT: por %xmm2, %xmm1
-; SSSE3-NEXT: movdqa %xmm8, %xmm2
-; SSSE3-NEXT: pxor %xmm0, %xmm2
-; SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm2[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm7, %xmm9
-; SSSE3-NEXT: movdqa %xmm6, %xmm10
-; SSSE3-NEXT: pcmpgtd %xmm2, %xmm10
-; SSSE3-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
-; SSSE3-NEXT: pand %xmm9, %xmm11
-; SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm10[1,1,3,3]
-; SSSE3-NEXT: por %xmm11, %xmm2
-; SSSE3-NEXT: pand %xmm2, %xmm8
-; SSSE3-NEXT: pandn %xmm4, %xmm2
-; SSSE3-NEXT: por %xmm8, %xmm2
-; SSSE3-NEXT: movdqa %xmm5, %xmm8
-; SSSE3-NEXT: pxor %xmm0, %xmm8
-; SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm8[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm7, %xmm9
-; SSSE3-NEXT: movdqa %xmm6, %xmm10
-; SSSE3-NEXT: pcmpgtd %xmm8, %xmm10
-; SSSE3-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
-; SSSE3-NEXT: pand %xmm9, %xmm11
-; SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm10[1,1,3,3]
-; SSSE3-NEXT: por %xmm11, %xmm8
-; SSSE3-NEXT: pand %xmm8, %xmm5
-; SSSE3-NEXT: pandn %xmm4, %xmm8
-; SSSE3-NEXT: por %xmm5, %xmm8
-; SSSE3-NEXT: movdqa %xmm3, %xmm5
-; SSSE3-NEXT: pxor %xmm0, %xmm5
-; SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm5[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm7, %xmm9
-; SSSE3-NEXT: pcmpgtd %xmm5, %xmm6
-; SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm6[0,0,2,2]
-; SSSE3-NEXT: pand %xmm9, %xmm5
-; SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
-; SSSE3-NEXT: por %xmm5, %xmm6
-; SSSE3-NEXT: pand %xmm6, %xmm3
-; SSSE3-NEXT: pandn %xmm4, %xmm6
-; SSSE3-NEXT: por %xmm3, %xmm6
-; SSSE3-NEXT: movdqa %xmm6, %xmm3
-; SSSE3-NEXT: pxor %xmm0, %xmm3
-; SSSE3-NEXT: movdqa %xmm3, %xmm4
-; SSSE3-NEXT: pcmpgtd %xmm0, %xmm4
-; SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm4[0,0,2,2]
-; SSSE3-NEXT: pcmpeqd %xmm0, %xmm3
-; SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
-; SSSE3-NEXT: pand %xmm5, %xmm3
-; SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
-; SSSE3-NEXT: por %xmm3, %xmm4
-; SSSE3-NEXT: pand %xmm6, %xmm4
-; SSSE3-NEXT: movdqa %xmm8, %xmm3
-; SSSE3-NEXT: pxor %xmm0, %xmm3
-; SSSE3-NEXT: movdqa %xmm3, %xmm5
-; SSSE3-NEXT: pcmpgtd %xmm0, %xmm5
-; SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm5[0,0,2,2]
-; SSSE3-NEXT: pcmpeqd %xmm0, %xmm3
-; SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm3[1,1,3,3]
-; SSSE3-NEXT: pand %xmm6, %xmm7
-; SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm5[1,1,3,3]
-; SSSE3-NEXT: por %xmm7, %xmm3
-; SSSE3-NEXT: pand %xmm8, %xmm3
-; SSSE3-NEXT: packuswb %xmm4, %xmm3
-; SSSE3-NEXT: movdqa %xmm2, %xmm4
-; SSSE3-NEXT: pxor %xmm0, %xmm4
-; SSSE3-NEXT: movdqa %xmm4, %xmm5
-; SSSE3-NEXT: pcmpgtd %xmm0, %xmm5
-; SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm5[0,0,2,2]
-; SSSE3-NEXT: pcmpeqd %xmm0, %xmm4
-; SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
-; SSSE3-NEXT: pand %xmm6, %xmm4
-; SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm5[1,1,3,3]
-; SSSE3-NEXT: por %xmm4, %xmm5
-; SSSE3-NEXT: pand %xmm2, %xmm5
-; SSSE3-NEXT: movdqa %xmm1, %xmm2
-; SSSE3-NEXT: pxor %xmm0, %xmm2
-; SSSE3-NEXT: movdqa %xmm2, %xmm4
-; SSSE3-NEXT: pcmpgtd %xmm0, %xmm4
-; SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm4[0,0,2,2]
-; SSSE3-NEXT: pcmpeqd %xmm0, %xmm2
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm2[1,1,3,3]
-; SSSE3-NEXT: pand %xmm6, %xmm0
-; SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm4[1,1,3,3]
-; SSSE3-NEXT: por %xmm0, %xmm2
-; SSSE3-NEXT: pand %xmm1, %xmm2
-; SSSE3-NEXT: packuswb %xmm5, %xmm2
-; SSSE3-NEXT: packuswb %xmm2, %xmm3
-; SSSE3-NEXT: packuswb %xmm3, %xmm3
-; SSSE3-NEXT: movq %xmm3, (%rsi)
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc_packus_v8i64_v8i8_store:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa (%rdi), %xmm5
+; SSE2-SSSE3-NEXT: movdqa 16(%rdi), %xmm3
+; SSE2-SSSE3-NEXT: movdqa 32(%rdi), %xmm2
+; SSE2-SSSE3-NEXT: movdqa 48(%rdi), %xmm8
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm4 = [255,255]
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm0 = [2147483648,2147483648]
+; SSE2-SSSE3-NEXT: movdqa %xmm2, %xmm1
+; SSE2-SSSE3-NEXT: pxor %xmm0, %xmm1
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm1[1,1,3,3]
+; SSE2-SSSE3-NEXT: pxor %xmm7, %xmm7
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm7, %xmm9
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm6 = [2147483903,2147483903]
+; SSE2-SSSE3-NEXT: movdqa %xmm6, %xmm10
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm1, %xmm10
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
+; SSE2-SSSE3-NEXT: pand %xmm9, %xmm11
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm10[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm11, %xmm1
+; SSE2-SSSE3-NEXT: pand %xmm1, %xmm2
+; SSE2-SSSE3-NEXT: pandn %xmm4, %xmm1
+; SSE2-SSSE3-NEXT: por %xmm2, %xmm1
+; SSE2-SSSE3-NEXT: movdqa %xmm8, %xmm2
+; SSE2-SSSE3-NEXT: pxor %xmm0, %xmm2
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm2[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm7, %xmm9
+; SSE2-SSSE3-NEXT: movdqa %xmm6, %xmm10
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm2, %xmm10
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
+; SSE2-SSSE3-NEXT: pand %xmm9, %xmm11
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm10[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm11, %xmm2
+; SSE2-SSSE3-NEXT: pand %xmm2, %xmm8
+; SSE2-SSSE3-NEXT: pandn %xmm4, %xmm2
+; SSE2-SSSE3-NEXT: por %xmm8, %xmm2
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm8
+; SSE2-SSSE3-NEXT: pxor %xmm0, %xmm8
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm8[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm7, %xmm9
+; SSE2-SSSE3-NEXT: movdqa %xmm6, %xmm10
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm8, %xmm10
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
+; SSE2-SSSE3-NEXT: pand %xmm9, %xmm11
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm10[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm11, %xmm8
+; SSE2-SSSE3-NEXT: pand %xmm8, %xmm5
+; SSE2-SSSE3-NEXT: pandn %xmm4, %xmm8
+; SSE2-SSSE3-NEXT: por %xmm5, %xmm8
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm5
+; SSE2-SSSE3-NEXT: pxor %xmm0, %xmm5
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm5[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm7, %xmm9
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm5, %xmm6
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm6[0,0,2,2]
+; SSE2-SSSE3-NEXT: pand %xmm9, %xmm5
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm5, %xmm6
+; SSE2-SSSE3-NEXT: pand %xmm6, %xmm3
+; SSE2-SSSE3-NEXT: pandn %xmm4, %xmm6
+; SSE2-SSSE3-NEXT: por %xmm3, %xmm6
+; SSE2-SSSE3-NEXT: movdqa %xmm6, %xmm3
+; SSE2-SSSE3-NEXT: pxor %xmm0, %xmm3
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm4
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm0, %xmm4
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm4[0,0,2,2]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm0, %xmm3
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
+; SSE2-SSSE3-NEXT: pand %xmm5, %xmm3
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm3, %xmm4
+; SSE2-SSSE3-NEXT: pand %xmm6, %xmm4
+; SSE2-SSSE3-NEXT: movdqa %xmm8, %xmm3
+; SSE2-SSSE3-NEXT: pxor %xmm0, %xmm3
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm5
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm0, %xmm5
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm5[0,0,2,2]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm0, %xmm3
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm3[1,1,3,3]
+; SSE2-SSSE3-NEXT: pand %xmm6, %xmm7
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm5[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm7, %xmm3
+; SSE2-SSSE3-NEXT: pand %xmm8, %xmm3
+; SSE2-SSSE3-NEXT: packuswb %xmm4, %xmm3
+; SSE2-SSSE3-NEXT: movdqa %xmm2, %xmm4
+; SSE2-SSSE3-NEXT: pxor %xmm0, %xmm4
+; SSE2-SSSE3-NEXT: movdqa %xmm4, %xmm5
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm0, %xmm5
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm5[0,0,2,2]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm0, %xmm4
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
+; SSE2-SSSE3-NEXT: pand %xmm6, %xmm4
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm5[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm4, %xmm5
+; SSE2-SSSE3-NEXT: pand %xmm2, %xmm5
+; SSE2-SSSE3-NEXT: movdqa %xmm1, %xmm2
+; SSE2-SSSE3-NEXT: pxor %xmm0, %xmm2
+; SSE2-SSSE3-NEXT: movdqa %xmm2, %xmm4
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm0, %xmm4
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm4[0,0,2,2]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm0, %xmm2
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm2[1,1,3,3]
+; SSE2-SSSE3-NEXT: pand %xmm6, %xmm0
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm4[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm0, %xmm2
+; SSE2-SSSE3-NEXT: pand %xmm1, %xmm2
+; SSE2-SSSE3-NEXT: packuswb %xmm5, %xmm2
+; SSE2-SSSE3-NEXT: packuswb %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: packuswb %xmm3, %xmm3
+; SSE2-SSSE3-NEXT: movq %xmm3, (%rsi)
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_packus_v8i64_v8i8_store:
; SSE41: # %bb.0:
@@ -4404,433 +3605,219 @@ define void @trunc_packus_v8i64_v8i8_store(ptr %p0, ptr%p1) "min-legal-vector-wi
}
define <16 x i8> @trunc_packus_v16i64_v16i8(ptr %p0) "min-legal-vector-width"="256" {
-; SSE2-LABEL: trunc_packus_v16i64_v16i8:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa (%rdi), %xmm7
-; SSE2-NEXT: movdqa 16(%rdi), %xmm0
-; SSE2-NEXT: movdqa 32(%rdi), %xmm12
-; SSE2-NEXT: movdqa 48(%rdi), %xmm11
-; SSE2-NEXT: movdqa 80(%rdi), %xmm10
-; SSE2-NEXT: movdqa 64(%rdi), %xmm5
-; SSE2-NEXT: movdqa 112(%rdi), %xmm4
-; SSE2-NEXT: movdqa 96(%rdi), %xmm3
-; SSE2-NEXT: movdqa {{.*#+}} xmm6 = [255,255]
-; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [2147483648,2147483648]
-; SSE2-NEXT: movdqa %xmm3, %xmm2
-; SSE2-NEXT: pxor %xmm1, %xmm2
-; SSE2-NEXT: pshufd {{.*#+}} xmm13 = xmm2[1,1,3,3]
-; SSE2-NEXT: pxor %xmm9, %xmm9
-; SSE2-NEXT: pcmpeqd %xmm9, %xmm13
-; SSE2-NEXT: movdqa {{.*#+}} xmm8 = [2147483903,2147483903]
-; SSE2-NEXT: movdqa %xmm8, %xmm14
-; SSE2-NEXT: pcmpgtd %xmm2, %xmm14
-; SSE2-NEXT: pshufd {{.*#+}} xmm15 = xmm14[0,0,2,2]
-; SSE2-NEXT: pand %xmm13, %xmm15
-; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm14[1,1,3,3]
-; SSE2-NEXT: por %xmm15, %xmm2
-; SSE2-NEXT: pand %xmm2, %xmm3
-; SSE2-NEXT: pandn %xmm6, %xmm2
-; SSE2-NEXT: por %xmm3, %xmm2
-; SSE2-NEXT: movdqa %xmm4, %xmm3
-; SSE2-NEXT: pxor %xmm1, %xmm3
-; SSE2-NEXT: pshufd {{.*#+}} xmm13 = xmm3[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm9, %xmm13
-; SSE2-NEXT: movdqa %xmm8, %xmm14
-; SSE2-NEXT: pcmpgtd %xmm3, %xmm14
-; SSE2-NEXT: pshufd {{.*#+}} xmm15 = xmm14[0,0,2,2]
-; SSE2-NEXT: pand %xmm13, %xmm15
-; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm14[1,1,3,3]
-; SSE2-NEXT: por %xmm15, %xmm3
-; SSE2-NEXT: pand %xmm3, %xmm4
-; SSE2-NEXT: pandn %xmm6, %xmm3
-; SSE2-NEXT: por %xmm4, %xmm3
-; SSE2-NEXT: movdqa %xmm5, %xmm4
-; SSE2-NEXT: pxor %xmm1, %xmm4
-; SSE2-NEXT: pshufd {{.*#+}} xmm13 = xmm4[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm9, %xmm13
-; SSE2-NEXT: movdqa %xmm8, %xmm14
-; SSE2-NEXT: pcmpgtd %xmm4, %xmm14
-; SSE2-NEXT: pshufd {{.*#+}} xmm15 = xmm14[0,0,2,2]
-; SSE2-NEXT: pand %xmm13, %xmm15
-; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm14[1,1,3,3]
-; SSE2-NEXT: por %xmm15, %xmm4
-; SSE2-NEXT: pand %xmm4, %xmm5
-; SSE2-NEXT: pandn %xmm6, %xmm4
-; SSE2-NEXT: por %xmm5, %xmm4
-; SSE2-NEXT: movdqa %xmm10, %xmm5
-; SSE2-NEXT: pxor %xmm1, %xmm5
-; SSE2-NEXT: pshufd {{.*#+}} xmm13 = xmm5[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm9, %xmm13
-; SSE2-NEXT: movdqa %xmm8, %xmm14
-; SSE2-NEXT: pcmpgtd %xmm5, %xmm14
-; SSE2-NEXT: pshufd {{.*#+}} xmm15 = xmm14[0,0,2,2]
-; SSE2-NEXT: pand %xmm13, %xmm15
-; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm14[1,1,3,3]
-; SSE2-NEXT: por %xmm15, %xmm5
-; SSE2-NEXT: pand %xmm5, %xmm10
-; SSE2-NEXT: pandn %xmm6, %xmm5
-; SSE2-NEXT: por %xmm10, %xmm5
-; SSE2-NEXT: movdqa %xmm12, %xmm10
-; SSE2-NEXT: pxor %xmm1, %xmm10
-; SSE2-NEXT: pshufd {{.*#+}} xmm13 = xmm10[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm9, %xmm13
-; SSE2-NEXT: movdqa %xmm8, %xmm14
-; SSE2-NEXT: pcmpgtd %xmm10, %xmm14
-; SSE2-NEXT: pshufd {{.*#+}} xmm15 = xmm14[0,0,2,2]
-; SSE2-NEXT: pand %xmm13, %xmm15
-; SSE2-NEXT: pshufd {{.*#+}} xmm10 = xmm14[1,1,3,3]
-; SSE2-NEXT: por %xmm15, %xmm10
-; SSE2-NEXT: pand %xmm10, %xmm12
-; SSE2-NEXT: pandn %xmm6, %xmm10
-; SSE2-NEXT: por %xmm12, %xmm10
-; SSE2-NEXT: movdqa %xmm11, %xmm12
-; SSE2-NEXT: pxor %xmm1, %xmm12
-; SSE2-NEXT: pshufd {{.*#+}} xmm13 = xmm12[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm9, %xmm13
-; SSE2-NEXT: movdqa %xmm8, %xmm14
-; SSE2-NEXT: pcmpgtd %xmm12, %xmm14
-; SSE2-NEXT: pshufd {{.*#+}} xmm15 = xmm14[0,0,2,2]
-; SSE2-NEXT: pand %xmm13, %xmm15
-; SSE2-NEXT: pshufd {{.*#+}} xmm12 = xmm14[1,1,3,3]
-; SSE2-NEXT: por %xmm15, %xmm12
-; SSE2-NEXT: pand %xmm12, %xmm11
-; SSE2-NEXT: pandn %xmm6, %xmm12
-; SSE2-NEXT: por %xmm11, %xmm12
-; SSE2-NEXT: movdqa %xmm7, %xmm11
-; SSE2-NEXT: pxor %xmm1, %xmm11
-; SSE2-NEXT: pshufd {{.*#+}} xmm13 = xmm11[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm9, %xmm13
-; SSE2-NEXT: movdqa %xmm8, %xmm14
-; SSE2-NEXT: pcmpgtd %xmm11, %xmm14
-; SSE2-NEXT: pshufd {{.*#+}} xmm15 = xmm14[0,0,2,2]
-; SSE2-NEXT: pand %xmm13, %xmm15
-; SSE2-NEXT: pshufd {{.*#+}} xmm11 = xmm14[1,1,3,3]
-; SSE2-NEXT: por %xmm15, %xmm11
-; SSE2-NEXT: pand %xmm11, %xmm7
-; SSE2-NEXT: pandn %xmm6, %xmm11
-; SSE2-NEXT: por %xmm7, %xmm11
-; SSE2-NEXT: movdqa %xmm0, %xmm7
-; SSE2-NEXT: pxor %xmm1, %xmm7
-; SSE2-NEXT: pshufd {{.*#+}} xmm13 = xmm7[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm9, %xmm13
-; SSE2-NEXT: pcmpgtd %xmm7, %xmm8
-; SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm8[0,0,2,2]
-; SSE2-NEXT: pand %xmm13, %xmm7
-; SSE2-NEXT: pshufd {{.*#+}} xmm8 = xmm8[1,1,3,3]
-; SSE2-NEXT: por %xmm7, %xmm8
-; SSE2-NEXT: pand %xmm8, %xmm0
-; SSE2-NEXT: pandn %xmm6, %xmm8
-; SSE2-NEXT: por %xmm0, %xmm8
-; SSE2-NEXT: movdqa %xmm8, %xmm0
-; SSE2-NEXT: pxor %xmm1, %xmm0
-; SSE2-NEXT: movdqa %xmm0, %xmm6
-; SSE2-NEXT: pcmpgtd %xmm1, %xmm6
-; SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm6[0,0,2,2]
-; SSE2-NEXT: pcmpeqd %xmm1, %xmm0
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; SSE2-NEXT: pand %xmm7, %xmm0
-; SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
-; SSE2-NEXT: por %xmm0, %xmm6
-; SSE2-NEXT: pand %xmm8, %xmm6
-; SSE2-NEXT: movdqa %xmm11, %xmm0
-; SSE2-NEXT: pxor %xmm1, %xmm0
-; SSE2-NEXT: movdqa %xmm0, %xmm7
-; SSE2-NEXT: pcmpgtd %xmm1, %xmm7
-; SSE2-NEXT: pshufd {{.*#+}} xmm8 = xmm7[0,0,2,2]
-; SSE2-NEXT: pcmpeqd %xmm1, %xmm0
-; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm0[1,1,3,3]
-; SSE2-NEXT: pand %xmm8, %xmm9
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm7[1,1,3,3]
-; SSE2-NEXT: por %xmm9, %xmm0
-; SSE2-NEXT: pand %xmm11, %xmm0
-; SSE2-NEXT: packuswb %xmm6, %xmm0
-; SSE2-NEXT: movdqa %xmm12, %xmm6
-; SSE2-NEXT: pxor %xmm1, %xmm6
-; SSE2-NEXT: movdqa %xmm6, %xmm7
-; SSE2-NEXT: pcmpgtd %xmm1, %xmm7
-; SSE2-NEXT: pshufd {{.*#+}} xmm8 = xmm7[0,0,2,2]
-; SSE2-NEXT: pcmpeqd %xmm1, %xmm6
-; SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
-; SSE2-NEXT: pand %xmm8, %xmm6
-; SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm7[1,1,3,3]
-; SSE2-NEXT: por %xmm6, %xmm7
-; SSE2-NEXT: pand %xmm12, %xmm7
-; SSE2-NEXT: movdqa %xmm10, %xmm6
-; SSE2-NEXT: pxor %xmm1, %xmm6
-; SSE2-NEXT: movdqa %xmm6, %xmm8
-; SSE2-NEXT: pcmpgtd %xmm1, %xmm8
-; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm8[0,0,2,2]
-; SSE2-NEXT: pcmpeqd %xmm1, %xmm6
-; SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
-; SSE2-NEXT: pand %xmm9, %xmm6
-; SSE2-NEXT: pshufd {{.*#+}} xmm8 = xmm8[1,1,3,3]
-; SSE2-NEXT: por %xmm6, %xmm8
-; SSE2-NEXT: pand %xmm10, %xmm8
-; SSE2-NEXT: packuswb %xmm7, %xmm8
-; SSE2-NEXT: packuswb %xmm8, %xmm0
-; SSE2-NEXT: movdqa %xmm5, %xmm6
-; SSE2-NEXT: pxor %xmm1, %xmm6
-; SSE2-NEXT: movdqa %xmm6, %xmm7
-; SSE2-NEXT: pcmpgtd %xmm1, %xmm7
-; SSE2-NEXT: pshufd {{.*#+}} xmm8 = xmm7[0,0,2,2]
-; SSE2-NEXT: pcmpeqd %xmm1, %xmm6
-; SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
-; SSE2-NEXT: pand %xmm8, %xmm6
-; SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm7[1,1,3,3]
-; SSE2-NEXT: por %xmm6, %xmm7
-; SSE2-NEXT: pand %xmm5, %xmm7
-; SSE2-NEXT: movdqa %xmm4, %xmm5
-; SSE2-NEXT: pxor %xmm1, %xmm5
-; SSE2-NEXT: movdqa %xmm5, %xmm6
-; SSE2-NEXT: pcmpgtd %xmm1, %xmm6
-; SSE2-NEXT: pshufd {{.*#+}} xmm8 = xmm6[0,0,2,2]
-; SSE2-NEXT: pcmpeqd %xmm1, %xmm5
-; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm5[1,1,3,3]
-; SSE2-NEXT: pand %xmm8, %xmm9
-; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm6[1,1,3,3]
-; SSE2-NEXT: por %xmm9, %xmm5
-; SSE2-NEXT: pand %xmm4, %xmm5
-; SSE2-NEXT: packuswb %xmm7, %xmm5
-; SSE2-NEXT: movdqa %xmm3, %xmm4
-; SSE2-NEXT: pxor %xmm1, %xmm4
-; SSE2-NEXT: movdqa %xmm4, %xmm6
-; SSE2-NEXT: pcmpgtd %xmm1, %xmm6
-; SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm6[0,0,2,2]
-; SSE2-NEXT: pcmpeqd %xmm1, %xmm4
-; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
-; SSE2-NEXT: pand %xmm7, %xmm4
-; SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
-; SSE2-NEXT: por %xmm4, %xmm6
-; SSE2-NEXT: pand %xmm3, %xmm6
-; SSE2-NEXT: movdqa %xmm2, %xmm3
-; SSE2-NEXT: pxor %xmm1, %xmm3
-; SSE2-NEXT: movdqa %xmm3, %xmm4
-; SSE2-NEXT: pcmpgtd %xmm1, %xmm4
-; SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm4[0,0,2,2]
-; SSE2-NEXT: pcmpeqd %xmm1, %xmm3
-; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm3[1,1,3,3]
-; SSE2-NEXT: pand %xmm7, %xmm1
-; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm4[1,1,3,3]
-; SSE2-NEXT: por %xmm1, %xmm3
-; SSE2-NEXT: pand %xmm2, %xmm3
-; SSE2-NEXT: packuswb %xmm6, %xmm3
-; SSE2-NEXT: packuswb %xmm3, %xmm5
-; SSE2-NEXT: packuswb %xmm5, %xmm0
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_packus_v16i64_v16i8:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa (%rdi), %xmm7
-; SSSE3-NEXT: movdqa 16(%rdi), %xmm0
-; SSSE3-NEXT: movdqa 32(%rdi), %xmm12
-; SSSE3-NEXT: movdqa 48(%rdi), %xmm11
-; SSSE3-NEXT: movdqa 80(%rdi), %xmm10
-; SSSE3-NEXT: movdqa 64(%rdi), %xmm5
-; SSSE3-NEXT: movdqa 112(%rdi), %xmm4
-; SSSE3-NEXT: movdqa 96(%rdi), %xmm3
-; SSSE3-NEXT: movdqa {{.*#+}} xmm6 = [255,255]
-; SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [2147483648,2147483648]
-; SSSE3-NEXT: movdqa %xmm3, %xmm2
-; SSSE3-NEXT: pxor %xmm1, %xmm2
-; SSSE3-NEXT: pshufd {{.*#+}} xmm13 = xmm2[1,1,3,3]
-; SSSE3-NEXT: pxor %xmm9, %xmm9
-; SSSE3-NEXT: pcmpeqd %xmm9, %xmm13
-; SSSE3-NEXT: movdqa {{.*#+}} xmm8 = [2147483903,2147483903]
-; SSSE3-NEXT: movdqa %xmm8, %xmm14
-; SSSE3-NEXT: pcmpgtd %xmm2, %xmm14
-; SSSE3-NEXT: pshufd {{.*#+}} xmm15 = xmm14[0,0,2,2]
-; SSSE3-NEXT: pand %xmm13, %xmm15
-; SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm14[1,1,3,3]
-; SSSE3-NEXT: por %xmm15, %xmm2
-; SSSE3-NEXT: pand %xmm2, %xmm3
-; SSSE3-NEXT: pandn %xmm6, %xmm2
-; SSSE3-NEXT: por %xmm3, %xmm2
-; SSSE3-NEXT: movdqa %xmm4, %xmm3
-; SSSE3-NEXT: pxor %xmm1, %xmm3
-; SSSE3-NEXT: pshufd {{.*#+}} xmm13 = xmm3[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm9, %xmm13
-; SSSE3-NEXT: movdqa %xmm8, %xmm14
-; SSSE3-NEXT: pcmpgtd %xmm3, %xmm14
-; SSSE3-NEXT: pshufd {{.*#+}} xmm15 = xmm14[0,0,2,2]
-; SSSE3-NEXT: pand %xmm13, %xmm15
-; SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm14[1,1,3,3]
-; SSSE3-NEXT: por %xmm15, %xmm3
-; SSSE3-NEXT: pand %xmm3, %xmm4
-; SSSE3-NEXT: pandn %xmm6, %xmm3
-; SSSE3-NEXT: por %xmm4, %xmm3
-; SSSE3-NEXT: movdqa %xmm5, %xmm4
-; SSSE3-NEXT: pxor %xmm1, %xmm4
-; SSSE3-NEXT: pshufd {{.*#+}} xmm13 = xmm4[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm9, %xmm13
-; SSSE3-NEXT: movdqa %xmm8, %xmm14
-; SSSE3-NEXT: pcmpgtd %xmm4, %xmm14
-; SSSE3-NEXT: pshufd {{.*#+}} xmm15 = xmm14[0,0,2,2]
-; SSSE3-NEXT: pand %xmm13, %xmm15
-; SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm14[1,1,3,3]
-; SSSE3-NEXT: por %xmm15, %xmm4
-; SSSE3-NEXT: pand %xmm4, %xmm5
-; SSSE3-NEXT: pandn %xmm6, %xmm4
-; SSSE3-NEXT: por %xmm5, %xmm4
-; SSSE3-NEXT: movdqa %xmm10, %xmm5
-; SSSE3-NEXT: pxor %xmm1, %xmm5
-; SSSE3-NEXT: pshufd {{.*#+}} xmm13 = xmm5[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm9, %xmm13
-; SSSE3-NEXT: movdqa %xmm8, %xmm14
-; SSSE3-NEXT: pcmpgtd %xmm5, %xmm14
-; SSSE3-NEXT: pshufd {{.*#+}} xmm15 = xmm14[0,0,2,2]
-; SSSE3-NEXT: pand %xmm13, %xmm15
-; SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm14[1,1,3,3]
-; SSSE3-NEXT: por %xmm15, %xmm5
-; SSSE3-NEXT: pand %xmm5, %xmm10
-; SSSE3-NEXT: pandn %xmm6, %xmm5
-; SSSE3-NEXT: por %xmm10, %xmm5
-; SSSE3-NEXT: movdqa %xmm12, %xmm10
-; SSSE3-NEXT: pxor %xmm1, %xmm10
-; SSSE3-NEXT: pshufd {{.*#+}} xmm13 = xmm10[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm9, %xmm13
-; SSSE3-NEXT: movdqa %xmm8, %xmm14
-; SSSE3-NEXT: pcmpgtd %xmm10, %xmm14
-; SSSE3-NEXT: pshufd {{.*#+}} xmm15 = xmm14[0,0,2,2]
-; SSSE3-NEXT: pand %xmm13, %xmm15
-; SSSE3-NEXT: pshufd {{.*#+}} xmm10 = xmm14[1,1,3,3]
-; SSSE3-NEXT: por %xmm15, %xmm10
-; SSSE3-NEXT: pand %xmm10, %xmm12
-; SSSE3-NEXT: pandn %xmm6, %xmm10
-; SSSE3-NEXT: por %xmm12, %xmm10
-; SSSE3-NEXT: movdqa %xmm11, %xmm12
-; SSSE3-NEXT: pxor %xmm1, %xmm12
-; SSSE3-NEXT: pshufd {{.*#+}} xmm13 = xmm12[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm9, %xmm13
-; SSSE3-NEXT: movdqa %xmm8, %xmm14
-; SSSE3-NEXT: pcmpgtd %xmm12, %xmm14
-; SSSE3-NEXT: pshufd {{.*#+}} xmm15 = xmm14[0,0,2,2]
-; SSSE3-NEXT: pand %xmm13, %xmm15
-; SSSE3-NEXT: pshufd {{.*#+}} xmm12 = xmm14[1,1,3,3]
-; SSSE3-NEXT: por %xmm15, %xmm12
-; SSSE3-NEXT: pand %xmm12, %xmm11
-; SSSE3-NEXT: pandn %xmm6, %xmm12
-; SSSE3-NEXT: por %xmm11, %xmm12
-; SSSE3-NEXT: movdqa %xmm7, %xmm11
-; SSSE3-NEXT: pxor %xmm1, %xmm11
-; SSSE3-NEXT: pshufd {{.*#+}} xmm13 = xmm11[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm9, %xmm13
-; SSSE3-NEXT: movdqa %xmm8, %xmm14
-; SSSE3-NEXT: pcmpgtd %xmm11, %xmm14
-; SSSE3-NEXT: pshufd {{.*#+}} xmm15 = xmm14[0,0,2,2]
-; SSSE3-NEXT: pand %xmm13, %xmm15
-; SSSE3-NEXT: pshufd {{.*#+}} xmm11 = xmm14[1,1,3,3]
-; SSSE3-NEXT: por %xmm15, %xmm11
-; SSSE3-NEXT: pand %xmm11, %xmm7
-; SSSE3-NEXT: pandn %xmm6, %xmm11
-; SSSE3-NEXT: por %xmm7, %xmm11
-; SSSE3-NEXT: movdqa %xmm0, %xmm7
-; SSSE3-NEXT: pxor %xmm1, %xmm7
-; SSSE3-NEXT: pshufd {{.*#+}} xmm13 = xmm7[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm9, %xmm13
-; SSSE3-NEXT: pcmpgtd %xmm7, %xmm8
-; SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm8[0,0,2,2]
-; SSSE3-NEXT: pand %xmm13, %xmm7
-; SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm8[1,1,3,3]
-; SSSE3-NEXT: por %xmm7, %xmm8
-; SSSE3-NEXT: pand %xmm8, %xmm0
-; SSSE3-NEXT: pandn %xmm6, %xmm8
-; SSSE3-NEXT: por %xmm0, %xmm8
-; SSSE3-NEXT: movdqa %xmm8, %xmm0
-; SSSE3-NEXT: pxor %xmm1, %xmm0
-; SSSE3-NEXT: movdqa %xmm0, %xmm6
-; SSSE3-NEXT: pcmpgtd %xmm1, %xmm6
-; SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm6[0,0,2,2]
-; SSSE3-NEXT: pcmpeqd %xmm1, %xmm0
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; SSSE3-NEXT: pand %xmm7, %xmm0
-; SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
-; SSSE3-NEXT: por %xmm0, %xmm6
-; SSSE3-NEXT: pand %xmm8, %xmm6
-; SSSE3-NEXT: movdqa %xmm11, %xmm0
-; SSSE3-NEXT: pxor %xmm1, %xmm0
-; SSSE3-NEXT: movdqa %xmm0, %xmm7
-; SSSE3-NEXT: pcmpgtd %xmm1, %xmm7
-; SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm7[0,0,2,2]
-; SSSE3-NEXT: pcmpeqd %xmm1, %xmm0
-; SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm0[1,1,3,3]
-; SSSE3-NEXT: pand %xmm8, %xmm9
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm7[1,1,3,3]
-; SSSE3-NEXT: por %xmm9, %xmm0
-; SSSE3-NEXT: pand %xmm11, %xmm0
-; SSSE3-NEXT: packuswb %xmm6, %xmm0
-; SSSE3-NEXT: movdqa %xmm12, %xmm6
-; SSSE3-NEXT: pxor %xmm1, %xmm6
-; SSSE3-NEXT: movdqa %xmm6, %xmm7
-; SSSE3-NEXT: pcmpgtd %xmm1, %xmm7
-; SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm7[0,0,2,2]
-; SSSE3-NEXT: pcmpeqd %xmm1, %xmm6
-; SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
-; SSSE3-NEXT: pand %xmm8, %xmm6
-; SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm7[1,1,3,3]
-; SSSE3-NEXT: por %xmm6, %xmm7
-; SSSE3-NEXT: pand %xmm12, %xmm7
-; SSSE3-NEXT: movdqa %xmm10, %xmm6
-; SSSE3-NEXT: pxor %xmm1, %xmm6
-; SSSE3-NEXT: movdqa %xmm6, %xmm8
-; SSSE3-NEXT: pcmpgtd %xmm1, %xmm8
-; SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm8[0,0,2,2]
-; SSSE3-NEXT: pcmpeqd %xmm1, %xmm6
-; SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
-; SSSE3-NEXT: pand %xmm9, %xmm6
-; SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm8[1,1,3,3]
-; SSSE3-NEXT: por %xmm6, %xmm8
-; SSSE3-NEXT: pand %xmm10, %xmm8
-; SSSE3-NEXT: packuswb %xmm7, %xmm8
-; SSSE3-NEXT: packuswb %xmm8, %xmm0
-; SSSE3-NEXT: movdqa %xmm5, %xmm6
-; SSSE3-NEXT: pxor %xmm1, %xmm6
-; SSSE3-NEXT: movdqa %xmm6, %xmm7
-; SSSE3-NEXT: pcmpgtd %xmm1, %xmm7
-; SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm7[0,0,2,2]
-; SSSE3-NEXT: pcmpeqd %xmm1, %xmm6
-; SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
-; SSSE3-NEXT: pand %xmm8, %xmm6
-; SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm7[1,1,3,3]
-; SSSE3-NEXT: por %xmm6, %xmm7
-; SSSE3-NEXT: pand %xmm5, %xmm7
-; SSSE3-NEXT: movdqa %xmm4, %xmm5
-; SSSE3-NEXT: pxor %xmm1, %xmm5
-; SSSE3-NEXT: movdqa %xmm5, %xmm6
-; SSSE3-NEXT: pcmpgtd %xmm1, %xmm6
-; SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm6[0,0,2,2]
-; SSSE3-NEXT: pcmpeqd %xmm1, %xmm5
-; SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm5[1,1,3,3]
-; SSSE3-NEXT: pand %xmm8, %xmm9
-; SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm6[1,1,3,3]
-; SSSE3-NEXT: por %xmm9, %xmm5
-; SSSE3-NEXT: pand %xmm4, %xmm5
-; SSSE3-NEXT: packuswb %xmm7, %xmm5
-; SSSE3-NEXT: movdqa %xmm3, %xmm4
-; SSSE3-NEXT: pxor %xmm1, %xmm4
-; SSSE3-NEXT: movdqa %xmm4, %xmm6
-; SSSE3-NEXT: pcmpgtd %xmm1, %xmm6
-; SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm6[0,0,2,2]
-; SSSE3-NEXT: pcmpeqd %xmm1, %xmm4
-; SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
-; SSSE3-NEXT: pand %xmm7, %xmm4
-; SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
-; SSSE3-NEXT: por %xmm4, %xmm6
-; SSSE3-NEXT: pand %xmm3, %xmm6
-; SSSE3-NEXT: movdqa %xmm2, %xmm3
-; SSSE3-NEXT: pxor %xmm1, %xmm3
-; SSSE3-NEXT: movdqa %xmm3, %xmm4
-; SSSE3-NEXT: pcmpgtd %xmm1, %xmm4
-; SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm4[0,0,2,2]
-; SSSE3-NEXT: pcmpeqd %xmm1, %xmm3
-; SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm3[1,1,3,3]
-; SSSE3-NEXT: pand %xmm7, %xmm1
-; SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm4[1,1,3,3]
-; SSSE3-NEXT: por %xmm1, %xmm3
-; SSSE3-NEXT: pand %xmm2, %xmm3
-; SSSE3-NEXT: packuswb %xmm6, %xmm3
-; SSSE3-NEXT: packuswb %xmm3, %xmm5
-; SSSE3-NEXT: packuswb %xmm5, %xmm0
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc_packus_v16i64_v16i8:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa (%rdi), %xmm7
+; SSE2-SSSE3-NEXT: movdqa 16(%rdi), %xmm0
+; SSE2-SSSE3-NEXT: movdqa 32(%rdi), %xmm12
+; SSE2-SSSE3-NEXT: movdqa 48(%rdi), %xmm11
+; SSE2-SSSE3-NEXT: movdqa 80(%rdi), %xmm10
+; SSE2-SSSE3-NEXT: movdqa 64(%rdi), %xmm5
+; SSE2-SSSE3-NEXT: movdqa 112(%rdi), %xmm4
+; SSE2-SSSE3-NEXT: movdqa 96(%rdi), %xmm3
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm6 = [255,255]
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [2147483648,2147483648]
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm2
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm2
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm13 = xmm2[1,1,3,3]
+; SSE2-SSSE3-NEXT: pxor %xmm9, %xmm9
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm9, %xmm13
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm8 = [2147483903,2147483903]
+; SSE2-SSSE3-NEXT: movdqa %xmm8, %xmm14
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm2, %xmm14
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm15 = xmm14[0,0,2,2]
+; SSE2-SSSE3-NEXT: pand %xmm13, %xmm15
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm14[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm15, %xmm2
+; SSE2-SSSE3-NEXT: pand %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: pandn %xmm6, %xmm2
+; SSE2-SSSE3-NEXT: por %xmm3, %xmm2
+; SSE2-SSSE3-NEXT: movdqa %xmm4, %xmm3
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm3
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm13 = xmm3[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm9, %xmm13
+; SSE2-SSSE3-NEXT: movdqa %xmm8, %xmm14
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm3, %xmm14
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm15 = xmm14[0,0,2,2]
+; SSE2-SSSE3-NEXT: pand %xmm13, %xmm15
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm14[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm15, %xmm3
+; SSE2-SSSE3-NEXT: pand %xmm3, %xmm4
+; SSE2-SSSE3-NEXT: pandn %xmm6, %xmm3
+; SSE2-SSSE3-NEXT: por %xmm4, %xmm3
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm4
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm4
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm13 = xmm4[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm9, %xmm13
+; SSE2-SSSE3-NEXT: movdqa %xmm8, %xmm14
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm4, %xmm14
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm15 = xmm14[0,0,2,2]
+; SSE2-SSSE3-NEXT: pand %xmm13, %xmm15
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm14[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm15, %xmm4
+; SSE2-SSSE3-NEXT: pand %xmm4, %xmm5
+; SSE2-SSSE3-NEXT: pandn %xmm6, %xmm4
+; SSE2-SSSE3-NEXT: por %xmm5, %xmm4
+; SSE2-SSSE3-NEXT: movdqa %xmm10, %xmm5
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm5
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm13 = xmm5[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm9, %xmm13
+; SSE2-SSSE3-NEXT: movdqa %xmm8, %xmm14
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm5, %xmm14
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm15 = xmm14[0,0,2,2]
+; SSE2-SSSE3-NEXT: pand %xmm13, %xmm15
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm14[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm15, %xmm5
+; SSE2-SSSE3-NEXT: pand %xmm5, %xmm10
+; SSE2-SSSE3-NEXT: pandn %xmm6, %xmm5
+; SSE2-SSSE3-NEXT: por %xmm10, %xmm5
+; SSE2-SSSE3-NEXT: movdqa %xmm12, %xmm10
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm10
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm13 = xmm10[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm9, %xmm13
+; SSE2-SSSE3-NEXT: movdqa %xmm8, %xmm14
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm10, %xmm14
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm15 = xmm14[0,0,2,2]
+; SSE2-SSSE3-NEXT: pand %xmm13, %xmm15
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm10 = xmm14[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm15, %xmm10
+; SSE2-SSSE3-NEXT: pand %xmm10, %xmm12
+; SSE2-SSSE3-NEXT: pandn %xmm6, %xmm10
+; SSE2-SSSE3-NEXT: por %xmm12, %xmm10
+; SSE2-SSSE3-NEXT: movdqa %xmm11, %xmm12
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm12
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm13 = xmm12[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm9, %xmm13
+; SSE2-SSSE3-NEXT: movdqa %xmm8, %xmm14
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm12, %xmm14
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm15 = xmm14[0,0,2,2]
+; SSE2-SSSE3-NEXT: pand %xmm13, %xmm15
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm12 = xmm14[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm15, %xmm12
+; SSE2-SSSE3-NEXT: pand %xmm12, %xmm11
+; SSE2-SSSE3-NEXT: pandn %xmm6, %xmm12
+; SSE2-SSSE3-NEXT: por %xmm11, %xmm12
+; SSE2-SSSE3-NEXT: movdqa %xmm7, %xmm11
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm11
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm13 = xmm11[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm9, %xmm13
+; SSE2-SSSE3-NEXT: movdqa %xmm8, %xmm14
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm11, %xmm14
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm15 = xmm14[0,0,2,2]
+; SSE2-SSSE3-NEXT: pand %xmm13, %xmm15
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm11 = xmm14[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm15, %xmm11
+; SSE2-SSSE3-NEXT: pand %xmm11, %xmm7
+; SSE2-SSSE3-NEXT: pandn %xmm6, %xmm11
+; SSE2-SSSE3-NEXT: por %xmm7, %xmm11
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm7
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm7
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm13 = xmm7[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm9, %xmm13
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm7, %xmm8
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm8[0,0,2,2]
+; SSE2-SSSE3-NEXT: pand %xmm13, %xmm7
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm8[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm7, %xmm8
+; SSE2-SSSE3-NEXT: pand %xmm8, %xmm0
+; SSE2-SSSE3-NEXT: pandn %xmm6, %xmm8
+; SSE2-SSSE3-NEXT: por %xmm0, %xmm8
+; SSE2-SSSE3-NEXT: movdqa %xmm8, %xmm0
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm6
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm1, %xmm6
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm6[0,0,2,2]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; SSE2-SSSE3-NEXT: pand %xmm7, %xmm0
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm0, %xmm6
+; SSE2-SSSE3-NEXT: pand %xmm8, %xmm6
+; SSE2-SSSE3-NEXT: movdqa %xmm11, %xmm0
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm7
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm1, %xmm7
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm7[0,0,2,2]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm0[1,1,3,3]
+; SSE2-SSSE3-NEXT: pand %xmm8, %xmm9
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm7[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm9, %xmm0
+; SSE2-SSSE3-NEXT: pand %xmm11, %xmm0
+; SSE2-SSSE3-NEXT: packuswb %xmm6, %xmm0
+; SSE2-SSSE3-NEXT: movdqa %xmm12, %xmm6
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm6
+; SSE2-SSSE3-NEXT: movdqa %xmm6, %xmm7
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm1, %xmm7
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm7[0,0,2,2]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm1, %xmm6
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
+; SSE2-SSSE3-NEXT: pand %xmm8, %xmm6
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm7[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm6, %xmm7
+; SSE2-SSSE3-NEXT: pand %xmm12, %xmm7
+; SSE2-SSSE3-NEXT: movdqa %xmm10, %xmm6
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm6
+; SSE2-SSSE3-NEXT: movdqa %xmm6, %xmm8
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm1, %xmm8
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm8[0,0,2,2]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm1, %xmm6
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
+; SSE2-SSSE3-NEXT: pand %xmm9, %xmm6
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm8[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm6, %xmm8
+; SSE2-SSSE3-NEXT: pand %xmm10, %xmm8
+; SSE2-SSSE3-NEXT: packuswb %xmm7, %xmm8
+; SSE2-SSSE3-NEXT: packuswb %xmm8, %xmm0
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm6
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm6
+; SSE2-SSSE3-NEXT: movdqa %xmm6, %xmm7
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm1, %xmm7
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm7[0,0,2,2]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm1, %xmm6
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
+; SSE2-SSSE3-NEXT: pand %xmm8, %xmm6
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm7[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm6, %xmm7
+; SSE2-SSSE3-NEXT: pand %xmm5, %xmm7
+; SSE2-SSSE3-NEXT: movdqa %xmm4, %xmm5
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm5
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm6
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm1, %xmm6
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm6[0,0,2,2]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm1, %xmm5
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm5[1,1,3,3]
+; SSE2-SSSE3-NEXT: pand %xmm8, %xmm9
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm6[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm9, %xmm5
+; SSE2-SSSE3-NEXT: pand %xmm4, %xmm5
+; SSE2-SSSE3-NEXT: packuswb %xmm7, %xmm5
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm4
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm4
+; SSE2-SSSE3-NEXT: movdqa %xmm4, %xmm6
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm1, %xmm6
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm6[0,0,2,2]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm1, %xmm4
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
+; SSE2-SSSE3-NEXT: pand %xmm7, %xmm4
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm4, %xmm6
+; SSE2-SSSE3-NEXT: pand %xmm3, %xmm6
+; SSE2-SSSE3-NEXT: movdqa %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm3
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm4
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm1, %xmm4
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm4[0,0,2,2]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm1, %xmm3
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm3[1,1,3,3]
+; SSE2-SSSE3-NEXT: pand %xmm7, %xmm1
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm4[1,1,3,3]
+; SSE2-SSSE3-NEXT: por %xmm1, %xmm3
+; SSE2-SSSE3-NEXT: pand %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: packuswb %xmm6, %xmm3
+; SSE2-SSSE3-NEXT: packuswb %xmm3, %xmm5
+; SSE2-SSSE3-NEXT: packuswb %xmm5, %xmm0
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_packus_v16i64_v16i8:
; SSE41: # %bb.0:
diff --git a/llvm/test/CodeGen/X86/vector-trunc-usat.ll b/llvm/test/CodeGen/X86/vector-trunc-usat.ll
index eaa9d24af98fd..a57d61b1c83c5 100644
--- a/llvm/test/CodeGen/X86/vector-trunc-usat.ll
+++ b/llvm/test/CodeGen/X86/vector-trunc-usat.ll
@@ -1,6 +1,6 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefix=SSE2
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+ssse3 | FileCheck %s --check-prefix=SSSE3
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefixes=SSE2-SSSE3,SSE2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+ssse3 | FileCheck %s --check-prefixes=SSE2-SSSE3,SSSE3
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefix=SSE41
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=AVX --check-prefix=AVX1
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-SLOW
@@ -20,39 +20,22 @@
;
define <2 x i32> @trunc_usat_v2i64_v2i32(<2 x i64> %a0) {
-; SSE2-LABEL: trunc_usat_v2i64_v2i32:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [9223372039002259456,9223372039002259456]
-; SSE2-NEXT: pxor %xmm0, %xmm1
-; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm1[0,0,2,2]
-; SSE2-NEXT: movdqa {{.*#+}} xmm3 = [2147483647,2147483647,2147483647,2147483647]
-; SSE2-NEXT: pcmpgtd %xmm2, %xmm3
-; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; SSE2-NEXT: pcmpeqd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
-; SSE2-NEXT: pand %xmm3, %xmm1
-; SSE2-NEXT: pcmpeqd %xmm2, %xmm2
-; SSE2-NEXT: pxor %xmm1, %xmm2
-; SSE2-NEXT: pand %xmm0, %xmm1
-; SSE2-NEXT: por %xmm2, %xmm1
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[0,2,2,3]
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_usat_v2i64_v2i32:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [9223372039002259456,9223372039002259456]
-; SSSE3-NEXT: pxor %xmm0, %xmm1
-; SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm1[0,0,2,2]
-; SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [2147483647,2147483647,2147483647,2147483647]
-; SSSE3-NEXT: pcmpgtd %xmm2, %xmm3
-; SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
-; SSSE3-NEXT: pand %xmm3, %xmm1
-; SSSE3-NEXT: pcmpeqd %xmm2, %xmm2
-; SSSE3-NEXT: pxor %xmm1, %xmm2
-; SSSE3-NEXT: pand %xmm0, %xmm1
-; SSSE3-NEXT: por %xmm2, %xmm1
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm1[0,2,2,3]
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc_usat_v2i64_v2i32:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [9223372039002259456,9223372039002259456]
+; SSE2-SSSE3-NEXT: pxor %xmm0, %xmm1
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm1[0,0,2,2]
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [2147483647,2147483647,2147483647,2147483647]
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
+; SSE2-SSSE3-NEXT: pand %xmm3, %xmm1
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm2, %xmm2
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm2
+; SSE2-SSSE3-NEXT: pand %xmm0, %xmm1
+; SSE2-SSSE3-NEXT: por %xmm2, %xmm1
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm1[0,2,2,3]
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_usat_v2i64_v2i32:
; SSE41: # %bb.0:
@@ -129,41 +112,23 @@ define <2 x i32> @trunc_usat_v2i64_v2i32(<2 x i64> %a0) {
}
define void @trunc_usat_v2i64_v2i32_store(<2 x i64> %a0, ptr %p1) {
-; SSE2-LABEL: trunc_usat_v2i64_v2i32_store:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [9223372039002259456,9223372039002259456]
-; SSE2-NEXT: pxor %xmm0, %xmm1
-; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm1[0,0,2,2]
-; SSE2-NEXT: movdqa {{.*#+}} xmm3 = [2147483647,2147483647,2147483647,2147483647]
-; SSE2-NEXT: pcmpgtd %xmm2, %xmm3
-; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; SSE2-NEXT: pcmpeqd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
-; SSE2-NEXT: pand %xmm3, %xmm1
-; SSE2-NEXT: pcmpeqd %xmm2, %xmm2
-; SSE2-NEXT: pxor %xmm1, %xmm2
-; SSE2-NEXT: pand %xmm0, %xmm1
-; SSE2-NEXT: por %xmm2, %xmm1
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[0,2,2,3]
-; SSE2-NEXT: movq %xmm0, (%rdi)
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_usat_v2i64_v2i32_store:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [9223372039002259456,9223372039002259456]
-; SSSE3-NEXT: pxor %xmm0, %xmm1
-; SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm1[0,0,2,2]
-; SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [2147483647,2147483647,2147483647,2147483647]
-; SSSE3-NEXT: pcmpgtd %xmm2, %xmm3
-; SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
-; SSSE3-NEXT: pand %xmm3, %xmm1
-; SSSE3-NEXT: pcmpeqd %xmm2, %xmm2
-; SSSE3-NEXT: pxor %xmm1, %xmm2
-; SSSE3-NEXT: pand %xmm0, %xmm1
-; SSSE3-NEXT: por %xmm2, %xmm1
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm1[0,2,2,3]
-; SSSE3-NEXT: movq %xmm0, (%rdi)
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc_usat_v2i64_v2i32_store:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [9223372039002259456,9223372039002259456]
+; SSE2-SSSE3-NEXT: pxor %xmm0, %xmm1
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm1[0,0,2,2]
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [2147483647,2147483647,2147483647,2147483647]
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
+; SSE2-SSSE3-NEXT: pand %xmm3, %xmm1
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm2, %xmm2
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm2
+; SSE2-SSSE3-NEXT: pand %xmm0, %xmm1
+; SSE2-SSSE3-NEXT: por %xmm2, %xmm1
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm1[0,2,2,3]
+; SSE2-SSSE3-NEXT: movq %xmm0, (%rdi)
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_usat_v2i64_v2i32_store:
; SSE41: # %bb.0:
@@ -244,63 +209,34 @@ define void @trunc_usat_v2i64_v2i32_store(<2 x i64> %a0, ptr %p1) {
}
define <4 x i32> @trunc_usat_v4i64_v4i32(<4 x i64> %a0) {
-; SSE2-LABEL: trunc_usat_v4i64_v4i32:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [9223372039002259456,9223372039002259456]
-; SSE2-NEXT: movdqa %xmm0, %xmm3
-; SSE2-NEXT: pxor %xmm2, %xmm3
-; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm3[0,0,2,2]
-; SSE2-NEXT: movdqa {{.*#+}} xmm5 = [2147483647,2147483647,2147483647,2147483647]
-; SSE2-NEXT: movdqa %xmm5, %xmm6
-; SSE2-NEXT: pcmpgtd %xmm4, %xmm6
-; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm2, %xmm3
-; SSE2-NEXT: pand %xmm6, %xmm3
-; SSE2-NEXT: movdqa %xmm1, %xmm4
-; SSE2-NEXT: pxor %xmm2, %xmm4
-; SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm4[0,0,2,2]
-; SSE2-NEXT: pcmpgtd %xmm6, %xmm5
-; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm2, %xmm4
-; SSE2-NEXT: pand %xmm5, %xmm4
-; SSE2-NEXT: pand %xmm4, %xmm1
-; SSE2-NEXT: pandn {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm4
-; SSE2-NEXT: por %xmm1, %xmm4
-; SSE2-NEXT: pcmpeqd %xmm1, %xmm1
-; SSE2-NEXT: pxor %xmm3, %xmm1
-; SSE2-NEXT: pand %xmm3, %xmm0
-; SSE2-NEXT: por %xmm1, %xmm0
-; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2],xmm4[0,2]
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_usat_v4i64_v4i32:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [9223372039002259456,9223372039002259456]
-; SSSE3-NEXT: movdqa %xmm0, %xmm3
-; SSSE3-NEXT: pxor %xmm2, %xmm3
-; SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm3[0,0,2,2]
-; SSSE3-NEXT: movdqa {{.*#+}} xmm5 = [2147483647,2147483647,2147483647,2147483647]
-; SSSE3-NEXT: movdqa %xmm5, %xmm6
-; SSSE3-NEXT: pcmpgtd %xmm4, %xmm6
-; SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm2, %xmm3
-; SSSE3-NEXT: pand %xmm6, %xmm3
-; SSSE3-NEXT: movdqa %xmm1, %xmm4
-; SSSE3-NEXT: pxor %xmm2, %xmm4
-; SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm4[0,0,2,2]
-; SSSE3-NEXT: pcmpgtd %xmm6, %xmm5
-; SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm2, %xmm4
-; SSSE3-NEXT: pand %xmm5, %xmm4
-; SSSE3-NEXT: pand %xmm4, %xmm1
-; SSSE3-NEXT: pandn {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm4
-; SSSE3-NEXT: por %xmm1, %xmm4
-; SSSE3-NEXT: pcmpeqd %xmm1, %xmm1
-; SSSE3-NEXT: pxor %xmm3, %xmm1
-; SSSE3-NEXT: pand %xmm3, %xmm0
-; SSSE3-NEXT: por %xmm1, %xmm0
-; SSSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2],xmm4[0,2]
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc_usat_v4i64_v4i32:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [9223372039002259456,9223372039002259456]
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm3
+; SSE2-SSSE3-NEXT: pxor %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm3[0,0,2,2]
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm5 = [2147483647,2147483647,2147483647,2147483647]
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm6
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm4, %xmm6
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: pand %xmm6, %xmm3
+; SSE2-SSSE3-NEXT: movdqa %xmm1, %xmm4
+; SSE2-SSSE3-NEXT: pxor %xmm2, %xmm4
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm4[0,0,2,2]
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm6, %xmm5
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm2, %xmm4
+; SSE2-SSSE3-NEXT: pand %xmm5, %xmm4
+; SSE2-SSSE3-NEXT: pand %xmm4, %xmm1
+; SSE2-SSSE3-NEXT: pandn {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm4
+; SSE2-SSSE3-NEXT: por %xmm1, %xmm4
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm1, %xmm1
+; SSE2-SSSE3-NEXT: pxor %xmm3, %xmm1
+; SSE2-SSSE3-NEXT: pand %xmm3, %xmm0
+; SSE2-SSSE3-NEXT: por %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2],xmm4[0,2]
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_usat_v4i64_v4i32:
; SSE41: # %bb.0:
@@ -445,117 +381,61 @@ define <4 x i32> @trunc_usat_v4i64_v4i32(<4 x i64> %a0) {
}
define <8 x i32> @trunc_usat_v8i64_v8i32(ptr %p0) {
-; SSE2-LABEL: trunc_usat_v8i64_v8i32:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa (%rdi), %xmm2
-; SSE2-NEXT: movdqa 16(%rdi), %xmm0
-; SSE2-NEXT: movdqa 32(%rdi), %xmm6
-; SSE2-NEXT: movdqa 48(%rdi), %xmm1
-; SSE2-NEXT: movdqa {{.*#+}} xmm3 = [9223372039002259456,9223372039002259456]
-; SSE2-NEXT: movdqa %xmm1, %xmm5
-; SSE2-NEXT: pxor %xmm3, %xmm5
-; SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm5[0,0,2,2]
-; SSE2-NEXT: movdqa {{.*#+}} xmm4 = [2147483647,2147483647,2147483647,2147483647]
-; SSE2-NEXT: movdqa %xmm4, %xmm8
-; SSE2-NEXT: pcmpgtd %xmm7, %xmm8
-; SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm5[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm3, %xmm7
-; SSE2-NEXT: pand %xmm8, %xmm7
-; SSE2-NEXT: pcmpeqd %xmm5, %xmm5
-; SSE2-NEXT: pand %xmm7, %xmm1
-; SSE2-NEXT: pxor %xmm5, %xmm7
-; SSE2-NEXT: por %xmm1, %xmm7
-; SSE2-NEXT: movdqa %xmm6, %xmm1
-; SSE2-NEXT: pxor %xmm3, %xmm1
-; SSE2-NEXT: pshufd {{.*#+}} xmm8 = xmm1[0,0,2,2]
-; SSE2-NEXT: movdqa %xmm4, %xmm9
-; SSE2-NEXT: pcmpgtd %xmm8, %xmm9
-; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm3, %xmm1
-; SSE2-NEXT: pand %xmm9, %xmm1
-; SSE2-NEXT: pand %xmm1, %xmm6
-; SSE2-NEXT: pxor %xmm5, %xmm1
-; SSE2-NEXT: por %xmm6, %xmm1
-; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,2],xmm7[0,2]
-; SSE2-NEXT: movdqa %xmm0, %xmm6
-; SSE2-NEXT: pxor %xmm3, %xmm6
-; SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm6[0,0,2,2]
-; SSE2-NEXT: movdqa %xmm4, %xmm8
-; SSE2-NEXT: pcmpgtd %xmm7, %xmm8
-; SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm3, %xmm6
-; SSE2-NEXT: pand %xmm8, %xmm6
-; SSE2-NEXT: pand %xmm6, %xmm0
-; SSE2-NEXT: pxor %xmm5, %xmm6
-; SSE2-NEXT: por %xmm0, %xmm6
-; SSE2-NEXT: movdqa %xmm2, %xmm0
-; SSE2-NEXT: pxor %xmm3, %xmm0
-; SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm0[0,0,2,2]
-; SSE2-NEXT: pcmpgtd %xmm7, %xmm4
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm3, %xmm0
-; SSE2-NEXT: pand %xmm4, %xmm0
-; SSE2-NEXT: pxor %xmm0, %xmm5
-; SSE2-NEXT: pand %xmm2, %xmm0
-; SSE2-NEXT: por %xmm5, %xmm0
-; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2],xmm6[0,2]
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_usat_v8i64_v8i32:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa (%rdi), %xmm2
-; SSSE3-NEXT: movdqa 16(%rdi), %xmm0
-; SSSE3-NEXT: movdqa 32(%rdi), %xmm6
-; SSSE3-NEXT: movdqa 48(%rdi), %xmm1
-; SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [9223372039002259456,9223372039002259456]
-; SSSE3-NEXT: movdqa %xmm1, %xmm5
-; SSSE3-NEXT: pxor %xmm3, %xmm5
-; SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm5[0,0,2,2]
-; SSSE3-NEXT: movdqa {{.*#+}} xmm4 = [2147483647,2147483647,2147483647,2147483647]
-; SSSE3-NEXT: movdqa %xmm4, %xmm8
-; SSSE3-NEXT: pcmpgtd %xmm7, %xmm8
-; SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm5[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm3, %xmm7
-; SSSE3-NEXT: pand %xmm8, %xmm7
-; SSSE3-NEXT: pcmpeqd %xmm5, %xmm5
-; SSSE3-NEXT: pand %xmm7, %xmm1
-; SSSE3-NEXT: pxor %xmm5, %xmm7
-; SSSE3-NEXT: por %xmm1, %xmm7
-; SSSE3-NEXT: movdqa %xmm6, %xmm1
-; SSSE3-NEXT: pxor %xmm3, %xmm1
-; SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm1[0,0,2,2]
-; SSSE3-NEXT: movdqa %xmm4, %xmm9
-; SSSE3-NEXT: pcmpgtd %xmm8, %xmm9
-; SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm3, %xmm1
-; SSSE3-NEXT: pand %xmm9, %xmm1
-; SSSE3-NEXT: pand %xmm1, %xmm6
-; SSSE3-NEXT: pxor %xmm5, %xmm1
-; SSSE3-NEXT: por %xmm6, %xmm1
-; SSSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,2],xmm7[0,2]
-; SSSE3-NEXT: movdqa %xmm0, %xmm6
-; SSSE3-NEXT: pxor %xmm3, %xmm6
-; SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm6[0,0,2,2]
-; SSSE3-NEXT: movdqa %xmm4, %xmm8
-; SSSE3-NEXT: pcmpgtd %xmm7, %xmm8
-; SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm3, %xmm6
-; SSSE3-NEXT: pand %xmm8, %xmm6
-; SSSE3-NEXT: pand %xmm6, %xmm0
-; SSSE3-NEXT: pxor %xmm5, %xmm6
-; SSSE3-NEXT: por %xmm0, %xmm6
-; SSSE3-NEXT: movdqa %xmm2, %xmm0
-; SSSE3-NEXT: pxor %xmm3, %xmm0
-; SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm0[0,0,2,2]
-; SSSE3-NEXT: pcmpgtd %xmm7, %xmm4
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm3, %xmm0
-; SSSE3-NEXT: pand %xmm4, %xmm0
-; SSSE3-NEXT: pxor %xmm0, %xmm5
-; SSSE3-NEXT: pand %xmm2, %xmm0
-; SSSE3-NEXT: por %xmm5, %xmm0
-; SSSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2],xmm6[0,2]
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc_usat_v8i64_v8i32:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa (%rdi), %xmm2
+; SSE2-SSSE3-NEXT: movdqa 16(%rdi), %xmm0
+; SSE2-SSSE3-NEXT: movdqa 32(%rdi), %xmm6
+; SSE2-SSSE3-NEXT: movdqa 48(%rdi), %xmm1
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [9223372039002259456,9223372039002259456]
+; SSE2-SSSE3-NEXT: movdqa %xmm1, %xmm5
+; SSE2-SSSE3-NEXT: pxor %xmm3, %xmm5
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm5[0,0,2,2]
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm4 = [2147483647,2147483647,2147483647,2147483647]
+; SSE2-SSSE3-NEXT: movdqa %xmm4, %xmm8
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm7, %xmm8
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm5[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm3, %xmm7
+; SSE2-SSSE3-NEXT: pand %xmm8, %xmm7
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm5, %xmm5
+; SSE2-SSSE3-NEXT: pand %xmm7, %xmm1
+; SSE2-SSSE3-NEXT: pxor %xmm5, %xmm7
+; SSE2-SSSE3-NEXT: por %xmm1, %xmm7
+; SSE2-SSSE3-NEXT: movdqa %xmm6, %xmm1
+; SSE2-SSSE3-NEXT: pxor %xmm3, %xmm1
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm1[0,0,2,2]
+; SSE2-SSSE3-NEXT: movdqa %xmm4, %xmm9
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm8, %xmm9
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm3, %xmm1
+; SSE2-SSSE3-NEXT: pand %xmm9, %xmm1
+; SSE2-SSSE3-NEXT: pand %xmm1, %xmm6
+; SSE2-SSSE3-NEXT: pxor %xmm5, %xmm1
+; SSE2-SSSE3-NEXT: por %xmm6, %xmm1
+; SSE2-SSSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,2],xmm7[0,2]
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm6
+; SSE2-SSSE3-NEXT: pxor %xmm3, %xmm6
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm6[0,0,2,2]
+; SSE2-SSSE3-NEXT: movdqa %xmm4, %xmm8
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm7, %xmm8
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm3, %xmm6
+; SSE2-SSSE3-NEXT: pand %xmm8, %xmm6
+; SSE2-SSSE3-NEXT: pand %xmm6, %xmm0
+; SSE2-SSSE3-NEXT: pxor %xmm5, %xmm6
+; SSE2-SSSE3-NEXT: por %xmm0, %xmm6
+; SSE2-SSSE3-NEXT: movdqa %xmm2, %xmm0
+; SSE2-SSSE3-NEXT: pxor %xmm3, %xmm0
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm0[0,0,2,2]
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm7, %xmm4
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm3, %xmm0
+; SSE2-SSSE3-NEXT: pand %xmm4, %xmm0
+; SSE2-SSSE3-NEXT: pxor %xmm0, %xmm5
+; SSE2-SSSE3-NEXT: pand %xmm2, %xmm0
+; SSE2-SSSE3-NEXT: por %xmm5, %xmm0
+; SSE2-SSSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2],xmm6[0,2]
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_usat_v8i64_v8i32:
; SSE41: # %bb.0:
@@ -716,41 +596,23 @@ define <8 x i32> @trunc_usat_v8i64_v8i32(ptr %p0) {
;
define <2 x i16> @trunc_usat_v2i64_v2i16(<2 x i64> %a0) {
-; SSE2-LABEL: trunc_usat_v2i64_v2i16:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [9223372039002259456,9223372039002259456]
-; SSE2-NEXT: pxor %xmm0, %xmm1
-; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm1[0,0,2,2]
-; SSE2-NEXT: movdqa {{.*#+}} xmm3 = [2147549183,2147549183,2147549183,2147549183]
-; SSE2-NEXT: pcmpgtd %xmm2, %xmm3
-; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; SSE2-NEXT: pcmpeqd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
-; SSE2-NEXT: pand %xmm3, %xmm1
-; SSE2-NEXT: pcmpeqd %xmm2, %xmm2
-; SSE2-NEXT: pxor %xmm1, %xmm2
-; SSE2-NEXT: pand %xmm0, %xmm1
-; SSE2-NEXT: por %xmm2, %xmm1
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[0,2,2,3]
-; SSE2-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_usat_v2i64_v2i16:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [9223372039002259456,9223372039002259456]
-; SSSE3-NEXT: pxor %xmm0, %xmm1
-; SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm1[0,0,2,2]
-; SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [2147549183,2147549183,2147549183,2147549183]
-; SSSE3-NEXT: pcmpgtd %xmm2, %xmm3
-; SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
-; SSSE3-NEXT: pand %xmm3, %xmm1
-; SSSE3-NEXT: pcmpeqd %xmm2, %xmm2
-; SSSE3-NEXT: pxor %xmm1, %xmm2
-; SSSE3-NEXT: pand %xmm0, %xmm1
-; SSSE3-NEXT: por %xmm2, %xmm1
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm1[0,2,2,3]
-; SSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc_usat_v2i64_v2i16:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [9223372039002259456,9223372039002259456]
+; SSE2-SSSE3-NEXT: pxor %xmm0, %xmm1
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm1[0,0,2,2]
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [2147549183,2147549183,2147549183,2147549183]
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
+; SSE2-SSSE3-NEXT: pand %xmm3, %xmm1
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm2, %xmm2
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm2
+; SSE2-SSSE3-NEXT: pand %xmm0, %xmm1
+; SSE2-SSSE3-NEXT: por %xmm2, %xmm1
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm1[0,2,2,3]
+; SSE2-SSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_usat_v2i64_v2i16:
; SSE41: # %bb.0:
@@ -839,43 +701,24 @@ define <2 x i16> @trunc_usat_v2i64_v2i16(<2 x i64> %a0) {
}
define void @trunc_usat_v2i64_v2i16_store(<2 x i64> %a0, ptr %p1) {
-; SSE2-LABEL: trunc_usat_v2i64_v2i16_store:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [9223372039002259456,9223372039002259456]
-; SSE2-NEXT: pxor %xmm0, %xmm1
-; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm1[0,0,2,2]
-; SSE2-NEXT: movdqa {{.*#+}} xmm3 = [2147549183,2147549183,2147549183,2147549183]
-; SSE2-NEXT: pcmpgtd %xmm2, %xmm3
-; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; SSE2-NEXT: pcmpeqd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
-; SSE2-NEXT: pand %xmm3, %xmm1
-; SSE2-NEXT: pcmpeqd %xmm2, %xmm2
-; SSE2-NEXT: pxor %xmm1, %xmm2
-; SSE2-NEXT: pand %xmm0, %xmm1
-; SSE2-NEXT: por %xmm2, %xmm1
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[0,2,2,3]
-; SSE2-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
-; SSE2-NEXT: movd %xmm0, (%rdi)
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_usat_v2i64_v2i16_store:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [9223372039002259456,9223372039002259456]
-; SSSE3-NEXT: pxor %xmm0, %xmm1
-; SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm1[0,0,2,2]
-; SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [2147549183,2147549183,2147549183,2147549183]
-; SSSE3-NEXT: pcmpgtd %xmm2, %xmm3
-; SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
-; SSSE3-NEXT: pand %xmm3, %xmm1
-; SSSE3-NEXT: pcmpeqd %xmm2, %xmm2
-; SSSE3-NEXT: pxor %xmm1, %xmm2
-; SSSE3-NEXT: pand %xmm0, %xmm1
-; SSSE3-NEXT: por %xmm2, %xmm1
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm1[0,2,2,3]
-; SSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
-; SSSE3-NEXT: movd %xmm0, (%rdi)
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc_usat_v2i64_v2i16_store:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [9223372039002259456,9223372039002259456]
+; SSE2-SSSE3-NEXT: pxor %xmm0, %xmm1
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm1[0,0,2,2]
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [2147549183,2147549183,2147549183,2147549183]
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
+; SSE2-SSSE3-NEXT: pand %xmm3, %xmm1
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm2, %xmm2
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm2
+; SSE2-SSSE3-NEXT: pand %xmm0, %xmm1
+; SSE2-SSSE3-NEXT: por %xmm2, %xmm1
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm1[0,2,2,3]
+; SSE2-SSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
+; SSE2-SSSE3-NEXT: movd %xmm0, (%rdi)
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_usat_v2i64_v2i16_store:
; SSE41: # %bb.0:
@@ -971,71 +814,38 @@ define void @trunc_usat_v2i64_v2i16_store(<2 x i64> %a0, ptr %p1) {
}
define <4 x i16> @trunc_usat_v4i64_v4i16(<4 x i64> %a0) {
-; SSE2-LABEL: trunc_usat_v4i64_v4i16:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [9223372039002259456,9223372039002259456]
-; SSE2-NEXT: movdqa %xmm0, %xmm3
-; SSE2-NEXT: pxor %xmm2, %xmm3
-; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm3[0,0,2,2]
-; SSE2-NEXT: movdqa {{.*#+}} xmm5 = [2147549183,2147549183,2147549183,2147549183]
-; SSE2-NEXT: movdqa %xmm5, %xmm6
-; SSE2-NEXT: pcmpgtd %xmm4, %xmm6
-; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm2, %xmm3
-; SSE2-NEXT: pand %xmm6, %xmm3
-; SSE2-NEXT: pcmpeqd %xmm4, %xmm4
-; SSE2-NEXT: pand %xmm3, %xmm0
-; SSE2-NEXT: pxor %xmm4, %xmm3
-; SSE2-NEXT: por %xmm0, %xmm3
-; SSE2-NEXT: movdqa %xmm1, %xmm0
-; SSE2-NEXT: pxor %xmm2, %xmm0
-; SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm0[0,0,2,2]
-; SSE2-NEXT: pcmpgtd %xmm6, %xmm5
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm2, %xmm0
-; SSE2-NEXT: pand %xmm5, %xmm0
-; SSE2-NEXT: pxor %xmm0, %xmm4
-; SSE2-NEXT: pand %xmm1, %xmm0
-; SSE2-NEXT: por %xmm4, %xmm0
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSE2-NEXT: pshuflw {{.*#+}} xmm1 = xmm0[0,2,2,3,4,5,6,7]
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm3[0,2,2,3]
-; SSE2-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
-; SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_usat_v4i64_v4i16:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [9223372039002259456,9223372039002259456]
-; SSSE3-NEXT: movdqa %xmm0, %xmm3
-; SSSE3-NEXT: pxor %xmm2, %xmm3
-; SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm3[0,0,2,2]
-; SSSE3-NEXT: movdqa {{.*#+}} xmm5 = [2147549183,2147549183,2147549183,2147549183]
-; SSSE3-NEXT: movdqa %xmm5, %xmm6
-; SSSE3-NEXT: pcmpgtd %xmm4, %xmm6
-; SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm2, %xmm3
-; SSSE3-NEXT: pand %xmm6, %xmm3
-; SSSE3-NEXT: pcmpeqd %xmm4, %xmm4
-; SSSE3-NEXT: pand %xmm3, %xmm0
-; SSSE3-NEXT: pxor %xmm4, %xmm3
-; SSSE3-NEXT: por %xmm0, %xmm3
-; SSSE3-NEXT: movdqa %xmm1, %xmm0
-; SSSE3-NEXT: pxor %xmm2, %xmm0
-; SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm0[0,0,2,2]
-; SSSE3-NEXT: pcmpgtd %xmm6, %xmm5
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm2, %xmm0
-; SSSE3-NEXT: pand %xmm5, %xmm0
-; SSSE3-NEXT: pxor %xmm0, %xmm4
-; SSSE3-NEXT: pand %xmm1, %xmm0
-; SSSE3-NEXT: por %xmm4, %xmm0
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSSE3-NEXT: pshuflw {{.*#+}} xmm1 = xmm0[0,2,2,3,4,5,6,7]
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm3[0,2,2,3]
-; SSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
-; SSSE3-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc_usat_v4i64_v4i16:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [9223372039002259456,9223372039002259456]
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm3
+; SSE2-SSSE3-NEXT: pxor %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm3[0,0,2,2]
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm5 = [2147549183,2147549183,2147549183,2147549183]
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm6
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm4, %xmm6
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: pand %xmm6, %xmm3
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm4, %xmm4
+; SSE2-SSSE3-NEXT: pand %xmm3, %xmm0
+; SSE2-SSSE3-NEXT: pxor %xmm4, %xmm3
+; SSE2-SSSE3-NEXT: por %xmm0, %xmm3
+; SSE2-SSSE3-NEXT: movdqa %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: pxor %xmm2, %xmm0
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm0[0,0,2,2]
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm6, %xmm5
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm2, %xmm0
+; SSE2-SSSE3-NEXT: pand %xmm5, %xmm0
+; SSE2-SSSE3-NEXT: pxor %xmm0, %xmm4
+; SSE2-SSSE3-NEXT: pand %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: por %xmm4, %xmm0
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; SSE2-SSSE3-NEXT: pshuflw {{.*#+}} xmm1 = xmm0[0,2,2,3,4,5,6,7]
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm3[0,2,2,3]
+; SSE2-SSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
+; SSE2-SSSE3-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_usat_v4i64_v4i16:
; SSE41: # %bb.0:
@@ -1138,73 +948,39 @@ define <4 x i16> @trunc_usat_v4i64_v4i16(<4 x i64> %a0) {
}
define void @trunc_usat_v4i64_v4i16_store(<4 x i64> %a0, ptr%p1) {
-; SSE2-LABEL: trunc_usat_v4i64_v4i16_store:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [9223372039002259456,9223372039002259456]
-; SSE2-NEXT: movdqa %xmm0, %xmm3
-; SSE2-NEXT: pxor %xmm2, %xmm3
-; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm3[0,0,2,2]
-; SSE2-NEXT: movdqa {{.*#+}} xmm5 = [2147549183,2147549183,2147549183,2147549183]
-; SSE2-NEXT: movdqa %xmm5, %xmm6
-; SSE2-NEXT: pcmpgtd %xmm4, %xmm6
-; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm2, %xmm3
-; SSE2-NEXT: pand %xmm6, %xmm3
-; SSE2-NEXT: pcmpeqd %xmm4, %xmm4
-; SSE2-NEXT: pand %xmm3, %xmm0
-; SSE2-NEXT: pxor %xmm4, %xmm3
-; SSE2-NEXT: por %xmm0, %xmm3
-; SSE2-NEXT: movdqa %xmm1, %xmm0
-; SSE2-NEXT: pxor %xmm2, %xmm0
-; SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm0[0,0,2,2]
-; SSE2-NEXT: pcmpgtd %xmm6, %xmm5
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm2, %xmm0
-; SSE2-NEXT: pand %xmm5, %xmm0
-; SSE2-NEXT: pxor %xmm0, %xmm4
-; SSE2-NEXT: pand %xmm1, %xmm0
-; SSE2-NEXT: por %xmm4, %xmm0
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSE2-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
-; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm3[0,2,2,3]
-; SSE2-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
-; SSE2-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; SSE2-NEXT: movq %xmm1, (%rdi)
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_usat_v4i64_v4i16_store:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [9223372039002259456,9223372039002259456]
-; SSSE3-NEXT: movdqa %xmm0, %xmm3
-; SSSE3-NEXT: pxor %xmm2, %xmm3
-; SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm3[0,0,2,2]
-; SSSE3-NEXT: movdqa {{.*#+}} xmm5 = [2147549183,2147549183,2147549183,2147549183]
-; SSSE3-NEXT: movdqa %xmm5, %xmm6
-; SSSE3-NEXT: pcmpgtd %xmm4, %xmm6
-; SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm2, %xmm3
-; SSSE3-NEXT: pand %xmm6, %xmm3
-; SSSE3-NEXT: pcmpeqd %xmm4, %xmm4
-; SSSE3-NEXT: pand %xmm3, %xmm0
-; SSSE3-NEXT: pxor %xmm4, %xmm3
-; SSSE3-NEXT: por %xmm0, %xmm3
-; SSSE3-NEXT: movdqa %xmm1, %xmm0
-; SSSE3-NEXT: pxor %xmm2, %xmm0
-; SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm0[0,0,2,2]
-; SSSE3-NEXT: pcmpgtd %xmm6, %xmm5
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm2, %xmm0
-; SSSE3-NEXT: pand %xmm5, %xmm0
-; SSSE3-NEXT: pxor %xmm0, %xmm4
-; SSSE3-NEXT: pand %xmm1, %xmm0
-; SSSE3-NEXT: por %xmm4, %xmm0
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
-; SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm3[0,2,2,3]
-; SSSE3-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
-; SSSE3-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; SSSE3-NEXT: movq %xmm1, (%rdi)
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc_usat_v4i64_v4i16_store:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [9223372039002259456,9223372039002259456]
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm3
+; SSE2-SSSE3-NEXT: pxor %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm3[0,0,2,2]
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm5 = [2147549183,2147549183,2147549183,2147549183]
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm6
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm4, %xmm6
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: pand %xmm6, %xmm3
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm4, %xmm4
+; SSE2-SSSE3-NEXT: pand %xmm3, %xmm0
+; SSE2-SSSE3-NEXT: pxor %xmm4, %xmm3
+; SSE2-SSSE3-NEXT: por %xmm0, %xmm3
+; SSE2-SSSE3-NEXT: movdqa %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: pxor %xmm2, %xmm0
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm0[0,0,2,2]
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm6, %xmm5
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm2, %xmm0
+; SSE2-SSSE3-NEXT: pand %xmm5, %xmm0
+; SSE2-SSSE3-NEXT: pxor %xmm0, %xmm4
+; SSE2-SSSE3-NEXT: pand %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: por %xmm4, %xmm0
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; SSE2-SSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm3[0,2,2,3]
+; SSE2-SSSE3-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
+; SSE2-SSSE3-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
+; SSE2-SSSE3-NEXT: movq %xmm1, (%rdi)
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_usat_v4i64_v4i16_store:
; SSE41: # %bb.0:
@@ -1312,135 +1088,70 @@ define void @trunc_usat_v4i64_v4i16_store(<4 x i64> %a0, ptr%p1) {
}
define <8 x i16> @trunc_usat_v8i64_v8i16(ptr %p0) {
-; SSE2-LABEL: trunc_usat_v8i64_v8i16:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa (%rdi), %xmm4
-; SSE2-NEXT: movdqa 16(%rdi), %xmm1
-; SSE2-NEXT: movdqa 32(%rdi), %xmm6
-; SSE2-NEXT: movdqa 48(%rdi), %xmm7
-; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [9223372039002259456,9223372039002259456]
-; SSE2-NEXT: movdqa %xmm6, %xmm0
-; SSE2-NEXT: pxor %xmm2, %xmm0
-; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm0[0,0,2,2]
-; SSE2-NEXT: movdqa {{.*#+}} xmm3 = [2147549183,2147549183,2147549183,2147549183]
-; SSE2-NEXT: movdqa %xmm3, %xmm8
-; SSE2-NEXT: pcmpgtd %xmm5, %xmm8
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm2, %xmm0
-; SSE2-NEXT: pand %xmm8, %xmm0
-; SSE2-NEXT: pcmpeqd %xmm5, %xmm5
-; SSE2-NEXT: pand %xmm0, %xmm6
-; SSE2-NEXT: pxor %xmm5, %xmm0
-; SSE2-NEXT: por %xmm6, %xmm0
-; SSE2-NEXT: movdqa %xmm7, %xmm6
-; SSE2-NEXT: pxor %xmm2, %xmm6
-; SSE2-NEXT: pshufd {{.*#+}} xmm8 = xmm6[0,0,2,2]
-; SSE2-NEXT: movdqa %xmm3, %xmm9
-; SSE2-NEXT: pcmpgtd %xmm8, %xmm9
-; SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm2, %xmm6
-; SSE2-NEXT: pand %xmm9, %xmm6
-; SSE2-NEXT: pand %xmm6, %xmm7
-; SSE2-NEXT: pxor %xmm5, %xmm6
-; SSE2-NEXT: por %xmm7, %xmm6
-; SSE2-NEXT: movdqa %xmm4, %xmm7
-; SSE2-NEXT: pxor %xmm2, %xmm7
-; SSE2-NEXT: pshufd {{.*#+}} xmm8 = xmm7[0,0,2,2]
-; SSE2-NEXT: movdqa %xmm3, %xmm9
-; SSE2-NEXT: pcmpgtd %xmm8, %xmm9
-; SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm7[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm2, %xmm7
-; SSE2-NEXT: pand %xmm9, %xmm7
-; SSE2-NEXT: pand %xmm7, %xmm4
-; SSE2-NEXT: pxor %xmm5, %xmm7
-; SSE2-NEXT: por %xmm4, %xmm7
-; SSE2-NEXT: movdqa %xmm1, %xmm4
-; SSE2-NEXT: pxor %xmm2, %xmm4
-; SSE2-NEXT: pshufd {{.*#+}} xmm8 = xmm4[0,0,2,2]
-; SSE2-NEXT: pcmpgtd %xmm8, %xmm3
-; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm2, %xmm4
-; SSE2-NEXT: pand %xmm3, %xmm4
-; SSE2-NEXT: pxor %xmm4, %xmm5
-; SSE2-NEXT: pand %xmm1, %xmm4
-; SSE2-NEXT: por %xmm5, %xmm4
-; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm4[0,2,2,3]
-; SSE2-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
-; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm7[0,2,2,3]
-; SSE2-NEXT: pshuflw {{.*#+}} xmm2 = xmm2[0,2,2,3,4,5,6,7]
-; SSE2-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
-; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm6[0,2,2,3]
-; SSE2-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,1,0,2,4,5,6,7]
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSE2-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,1,0,2,4,5,6,7]
-; SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; SSE2-NEXT: movsd {{.*#+}} xmm0 = xmm2[0],xmm0[1]
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_usat_v8i64_v8i16:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa (%rdi), %xmm4
-; SSSE3-NEXT: movdqa 16(%rdi), %xmm1
-; SSSE3-NEXT: movdqa 32(%rdi), %xmm6
-; SSSE3-NEXT: movdqa 48(%rdi), %xmm7
-; SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [9223372039002259456,9223372039002259456]
-; SSSE3-NEXT: movdqa %xmm6, %xmm0
-; SSSE3-NEXT: pxor %xmm2, %xmm0
-; SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm0[0,0,2,2]
-; SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [2147549183,2147549183,2147549183,2147549183]
-; SSSE3-NEXT: movdqa %xmm3, %xmm8
-; SSSE3-NEXT: pcmpgtd %xmm5, %xmm8
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm2, %xmm0
-; SSSE3-NEXT: pand %xmm8, %xmm0
-; SSSE3-NEXT: pcmpeqd %xmm5, %xmm5
-; SSSE3-NEXT: pand %xmm0, %xmm6
-; SSSE3-NEXT: pxor %xmm5, %xmm0
-; SSSE3-NEXT: por %xmm6, %xmm0
-; SSSE3-NEXT: movdqa %xmm7, %xmm6
-; SSSE3-NEXT: pxor %xmm2, %xmm6
-; SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm6[0,0,2,2]
-; SSSE3-NEXT: movdqa %xmm3, %xmm9
-; SSSE3-NEXT: pcmpgtd %xmm8, %xmm9
-; SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm2, %xmm6
-; SSSE3-NEXT: pand %xmm9, %xmm6
-; SSSE3-NEXT: pand %xmm6, %xmm7
-; SSSE3-NEXT: pxor %xmm5, %xmm6
-; SSSE3-NEXT: por %xmm7, %xmm6
-; SSSE3-NEXT: movdqa %xmm4, %xmm7
-; SSSE3-NEXT: pxor %xmm2, %xmm7
-; SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm7[0,0,2,2]
-; SSSE3-NEXT: movdqa %xmm3, %xmm9
-; SSSE3-NEXT: pcmpgtd %xmm8, %xmm9
-; SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm7[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm2, %xmm7
-; SSSE3-NEXT: pand %xmm9, %xmm7
-; SSSE3-NEXT: pand %xmm7, %xmm4
-; SSSE3-NEXT: pxor %xmm5, %xmm7
-; SSSE3-NEXT: por %xmm4, %xmm7
-; SSSE3-NEXT: movdqa %xmm1, %xmm4
-; SSSE3-NEXT: pxor %xmm2, %xmm4
-; SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm4[0,0,2,2]
-; SSSE3-NEXT: pcmpgtd %xmm8, %xmm3
-; SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm2, %xmm4
-; SSSE3-NEXT: pand %xmm3, %xmm4
-; SSSE3-NEXT: pxor %xmm4, %xmm5
-; SSSE3-NEXT: pand %xmm1, %xmm4
-; SSSE3-NEXT: por %xmm5, %xmm4
-; SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm4[0,2,2,3]
-; SSSE3-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
-; SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm7[0,2,2,3]
-; SSSE3-NEXT: pshuflw {{.*#+}} xmm2 = xmm2[0,2,2,3,4,5,6,7]
-; SSSE3-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
-; SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm6[0,2,2,3]
-; SSSE3-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,1,0,2,4,5,6,7]
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,1,0,2,4,5,6,7]
-; SSSE3-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; SSSE3-NEXT: movsd {{.*#+}} xmm0 = xmm2[0],xmm0[1]
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc_usat_v8i64_v8i16:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa (%rdi), %xmm4
+; SSE2-SSSE3-NEXT: movdqa 16(%rdi), %xmm1
+; SSE2-SSSE3-NEXT: movdqa 32(%rdi), %xmm6
+; SSE2-SSSE3-NEXT: movdqa 48(%rdi), %xmm7
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [9223372039002259456,9223372039002259456]
+; SSE2-SSSE3-NEXT: movdqa %xmm6, %xmm0
+; SSE2-SSSE3-NEXT: pxor %xmm2, %xmm0
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm0[0,0,2,2]
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [2147549183,2147549183,2147549183,2147549183]
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm8
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm5, %xmm8
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm2, %xmm0
+; SSE2-SSSE3-NEXT: pand %xmm8, %xmm0
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm5, %xmm5
+; SSE2-SSSE3-NEXT: pand %xmm0, %xmm6
+; SSE2-SSSE3-NEXT: pxor %xmm5, %xmm0
+; SSE2-SSSE3-NEXT: por %xmm6, %xmm0
+; SSE2-SSSE3-NEXT: movdqa %xmm7, %xmm6
+; SSE2-SSSE3-NEXT: pxor %xmm2, %xmm6
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm6[0,0,2,2]
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm9
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm8, %xmm9
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm2, %xmm6
+; SSE2-SSSE3-NEXT: pand %xmm9, %xmm6
+; SSE2-SSSE3-NEXT: pand %xmm6, %xmm7
+; SSE2-SSSE3-NEXT: pxor %xmm5, %xmm6
+; SSE2-SSSE3-NEXT: por %xmm7, %xmm6
+; SSE2-SSSE3-NEXT: movdqa %xmm4, %xmm7
+; SSE2-SSSE3-NEXT: pxor %xmm2, %xmm7
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm7[0,0,2,2]
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm9
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm8, %xmm9
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm7[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm2, %xmm7
+; SSE2-SSSE3-NEXT: pand %xmm9, %xmm7
+; SSE2-SSSE3-NEXT: pand %xmm7, %xmm4
+; SSE2-SSSE3-NEXT: pxor %xmm5, %xmm7
+; SSE2-SSSE3-NEXT: por %xmm4, %xmm7
+; SSE2-SSSE3-NEXT: movdqa %xmm1, %xmm4
+; SSE2-SSSE3-NEXT: pxor %xmm2, %xmm4
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm4[0,0,2,2]
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm8, %xmm3
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm2, %xmm4
+; SSE2-SSSE3-NEXT: pand %xmm3, %xmm4
+; SSE2-SSSE3-NEXT: pxor %xmm4, %xmm5
+; SSE2-SSSE3-NEXT: pand %xmm1, %xmm4
+; SSE2-SSSE3-NEXT: por %xmm5, %xmm4
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm4[0,2,2,3]
+; SSE2-SSSE3-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm7[0,2,2,3]
+; SSE2-SSSE3-NEXT: pshuflw {{.*#+}} xmm2 = xmm2[0,2,2,3,4,5,6,7]
+; SSE2-SSSE3-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm6[0,2,2,3]
+; SSE2-SSSE3-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,1,0,2,4,5,6,7]
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; SSE2-SSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,1,0,2,4,5,6,7]
+; SSE2-SSSE3-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; SSE2-SSSE3-NEXT: movsd {{.*#+}} xmm0 = xmm2[0],xmm0[1]
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_usat_v8i64_v8i16:
; SSE41: # %bb.0:
@@ -1853,99 +1564,52 @@ define <8 x i16> @trunc_usat_v8i32_v8i16(<8 x i32> %a0) {
}
define <16 x i16> @trunc_usat_v16i32_v16i16(ptr %p0) {
-; SSE2-LABEL: trunc_usat_v16i32_v16i16:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa (%rdi), %xmm5
-; SSE2-NEXT: movdqa 16(%rdi), %xmm4
-; SSE2-NEXT: movdqa 32(%rdi), %xmm0
-; SSE2-NEXT: movdqa 48(%rdi), %xmm8
-; SSE2-NEXT: movdqa {{.*#+}} xmm6 = [2147483648,2147483648,2147483648,2147483648]
-; SSE2-NEXT: movdqa %xmm0, %xmm3
-; SSE2-NEXT: pxor %xmm6, %xmm3
-; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [2147549183,2147549183,2147549183,2147549183]
-; SSE2-NEXT: movdqa %xmm2, %xmm1
-; SSE2-NEXT: pcmpgtd %xmm3, %xmm1
-; SSE2-NEXT: pcmpeqd %xmm7, %xmm7
-; SSE2-NEXT: pand %xmm1, %xmm0
-; SSE2-NEXT: pxor %xmm7, %xmm1
-; SSE2-NEXT: por %xmm0, %xmm1
-; SSE2-NEXT: movdqa %xmm8, %xmm0
-; SSE2-NEXT: pxor %xmm6, %xmm0
-; SSE2-NEXT: movdqa %xmm2, %xmm3
-; SSE2-NEXT: pcmpgtd %xmm0, %xmm3
-; SSE2-NEXT: pand %xmm3, %xmm8
-; SSE2-NEXT: pxor %xmm7, %xmm3
-; SSE2-NEXT: por %xmm8, %xmm3
-; SSE2-NEXT: movdqa %xmm5, %xmm8
-; SSE2-NEXT: pxor %xmm6, %xmm8
-; SSE2-NEXT: movdqa %xmm2, %xmm0
-; SSE2-NEXT: pcmpgtd %xmm8, %xmm0
-; SSE2-NEXT: pand %xmm0, %xmm5
-; SSE2-NEXT: pxor %xmm7, %xmm0
-; SSE2-NEXT: por %xmm5, %xmm0
-; SSE2-NEXT: pxor %xmm4, %xmm6
-; SSE2-NEXT: pcmpgtd %xmm6, %xmm2
-; SSE2-NEXT: pxor %xmm2, %xmm7
-; SSE2-NEXT: pand %xmm4, %xmm2
-; SSE2-NEXT: por %xmm7, %xmm2
-; SSE2-NEXT: pslld $16, %xmm2
-; SSE2-NEXT: psrad $16, %xmm2
-; SSE2-NEXT: pslld $16, %xmm0
-; SSE2-NEXT: psrad $16, %xmm0
-; SSE2-NEXT: packssdw %xmm2, %xmm0
-; SSE2-NEXT: pslld $16, %xmm3
-; SSE2-NEXT: psrad $16, %xmm3
-; SSE2-NEXT: pslld $16, %xmm1
-; SSE2-NEXT: psrad $16, %xmm1
-; SSE2-NEXT: packssdw %xmm3, %xmm1
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_usat_v16i32_v16i16:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa (%rdi), %xmm5
-; SSSE3-NEXT: movdqa 16(%rdi), %xmm4
-; SSSE3-NEXT: movdqa 32(%rdi), %xmm0
-; SSSE3-NEXT: movdqa 48(%rdi), %xmm8
-; SSSE3-NEXT: movdqa {{.*#+}} xmm6 = [2147483648,2147483648,2147483648,2147483648]
-; SSSE3-NEXT: movdqa %xmm0, %xmm3
-; SSSE3-NEXT: pxor %xmm6, %xmm3
-; SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [2147549183,2147549183,2147549183,2147549183]
-; SSSE3-NEXT: movdqa %xmm2, %xmm1
-; SSSE3-NEXT: pcmpgtd %xmm3, %xmm1
-; SSSE3-NEXT: pcmpeqd %xmm7, %xmm7
-; SSSE3-NEXT: pand %xmm1, %xmm0
-; SSSE3-NEXT: pxor %xmm7, %xmm1
-; SSSE3-NEXT: por %xmm0, %xmm1
-; SSSE3-NEXT: movdqa %xmm8, %xmm0
-; SSSE3-NEXT: pxor %xmm6, %xmm0
-; SSSE3-NEXT: movdqa %xmm2, %xmm3
-; SSSE3-NEXT: pcmpgtd %xmm0, %xmm3
-; SSSE3-NEXT: pand %xmm3, %xmm8
-; SSSE3-NEXT: pxor %xmm7, %xmm3
-; SSSE3-NEXT: por %xmm8, %xmm3
-; SSSE3-NEXT: movdqa %xmm5, %xmm8
-; SSSE3-NEXT: pxor %xmm6, %xmm8
-; SSSE3-NEXT: movdqa %xmm2, %xmm0
-; SSSE3-NEXT: pcmpgtd %xmm8, %xmm0
-; SSSE3-NEXT: pand %xmm0, %xmm5
-; SSSE3-NEXT: pxor %xmm7, %xmm0
-; SSSE3-NEXT: por %xmm5, %xmm0
-; SSSE3-NEXT: pxor %xmm4, %xmm6
-; SSSE3-NEXT: pcmpgtd %xmm6, %xmm2
-; SSSE3-NEXT: pxor %xmm2, %xmm7
-; SSSE3-NEXT: pand %xmm4, %xmm2
-; SSSE3-NEXT: por %xmm7, %xmm2
-; SSSE3-NEXT: pslld $16, %xmm2
-; SSSE3-NEXT: psrad $16, %xmm2
-; SSSE3-NEXT: pslld $16, %xmm0
-; SSSE3-NEXT: psrad $16, %xmm0
-; SSSE3-NEXT: packssdw %xmm2, %xmm0
-; SSSE3-NEXT: pslld $16, %xmm3
-; SSSE3-NEXT: psrad $16, %xmm3
-; SSSE3-NEXT: pslld $16, %xmm1
-; SSSE3-NEXT: psrad $16, %xmm1
-; SSSE3-NEXT: packssdw %xmm3, %xmm1
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc_usat_v16i32_v16i16:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa (%rdi), %xmm5
+; SSE2-SSSE3-NEXT: movdqa 16(%rdi), %xmm4
+; SSE2-SSSE3-NEXT: movdqa 32(%rdi), %xmm0
+; SSE2-SSSE3-NEXT: movdqa 48(%rdi), %xmm8
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm6 = [2147483648,2147483648,2147483648,2147483648]
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm3
+; SSE2-SSSE3-NEXT: pxor %xmm6, %xmm3
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [2147549183,2147549183,2147549183,2147549183]
+; SSE2-SSSE3-NEXT: movdqa %xmm2, %xmm1
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm3, %xmm1
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm7, %xmm7
+; SSE2-SSSE3-NEXT: pand %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: pxor %xmm7, %xmm1
+; SSE2-SSSE3-NEXT: por %xmm0, %xmm1
+; SSE2-SSSE3-NEXT: movdqa %xmm8, %xmm0
+; SSE2-SSSE3-NEXT: pxor %xmm6, %xmm0
+; SSE2-SSSE3-NEXT: movdqa %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm0, %xmm3
+; SSE2-SSSE3-NEXT: pand %xmm3, %xmm8
+; SSE2-SSSE3-NEXT: pxor %xmm7, %xmm3
+; SSE2-SSSE3-NEXT: por %xmm8, %xmm3
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm8
+; SSE2-SSSE3-NEXT: pxor %xmm6, %xmm8
+; SSE2-SSSE3-NEXT: movdqa %xmm2, %xmm0
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm8, %xmm0
+; SSE2-SSSE3-NEXT: pand %xmm0, %xmm5
+; SSE2-SSSE3-NEXT: pxor %xmm7, %xmm0
+; SSE2-SSSE3-NEXT: por %xmm5, %xmm0
+; SSE2-SSSE3-NEXT: pxor %xmm4, %xmm6
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm6, %xmm2
+; SSE2-SSSE3-NEXT: pxor %xmm2, %xmm7
+; SSE2-SSSE3-NEXT: pand %xmm4, %xmm2
+; SSE2-SSSE3-NEXT: por %xmm7, %xmm2
+; SSE2-SSSE3-NEXT: pslld $16, %xmm2
+; SSE2-SSSE3-NEXT: psrad $16, %xmm2
+; SSE2-SSSE3-NEXT: pslld $16, %xmm0
+; SSE2-SSSE3-NEXT: psrad $16, %xmm0
+; SSE2-SSSE3-NEXT: packssdw %xmm2, %xmm0
+; SSE2-SSSE3-NEXT: pslld $16, %xmm3
+; SSE2-SSSE3-NEXT: psrad $16, %xmm3
+; SSE2-SSSE3-NEXT: pslld $16, %xmm1
+; SSE2-SSSE3-NEXT: psrad $16, %xmm1
+; SSE2-SSSE3-NEXT: packssdw %xmm3, %xmm1
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_usat_v16i32_v16i16:
; SSE41: # %bb.0:
@@ -2587,121 +2251,63 @@ define void @trunc_usat_v4i64_v4i8_store(<4 x i64> %a0, ptr%p1) {
}
define <8 x i8> @trunc_usat_v8i64_v8i8(ptr %p0) {
-; SSE2-LABEL: trunc_usat_v8i64_v8i8:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa (%rdi), %xmm6
-; SSE2-NEXT: movdqa 16(%rdi), %xmm0
-; SSE2-NEXT: movdqa 32(%rdi), %xmm1
-; SSE2-NEXT: movdqa 48(%rdi), %xmm5
-; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [255,255]
-; SSE2-NEXT: movdqa {{.*#+}} xmm3 = [9223372039002259456,9223372039002259456]
-; SSE2-NEXT: movdqa %xmm0, %xmm7
-; SSE2-NEXT: pxor %xmm3, %xmm7
-; SSE2-NEXT: pshufd {{.*#+}} xmm8 = xmm7[0,0,2,2]
-; SSE2-NEXT: movdqa {{.*#+}} xmm4 = [2147483903,2147483903,2147483903,2147483903]
-; SSE2-NEXT: movdqa %xmm4, %xmm9
-; SSE2-NEXT: pcmpgtd %xmm8, %xmm9
-; SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm7[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm3, %xmm7
-; SSE2-NEXT: pand %xmm9, %xmm7
-; SSE2-NEXT: pand %xmm7, %xmm0
-; SSE2-NEXT: pandn %xmm2, %xmm7
-; SSE2-NEXT: por %xmm0, %xmm7
-; SSE2-NEXT: movdqa %xmm6, %xmm0
-; SSE2-NEXT: pxor %xmm3, %xmm0
-; SSE2-NEXT: pshufd {{.*#+}} xmm8 = xmm0[0,0,2,2]
-; SSE2-NEXT: movdqa %xmm4, %xmm9
-; SSE2-NEXT: pcmpgtd %xmm8, %xmm9
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm3, %xmm0
-; SSE2-NEXT: pand %xmm9, %xmm0
-; SSE2-NEXT: pand %xmm0, %xmm6
-; SSE2-NEXT: pandn %xmm2, %xmm0
-; SSE2-NEXT: por %xmm6, %xmm0
-; SSE2-NEXT: packuswb %xmm7, %xmm0
-; SSE2-NEXT: movdqa %xmm5, %xmm6
-; SSE2-NEXT: pxor %xmm3, %xmm6
-; SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm6[0,0,2,2]
-; SSE2-NEXT: movdqa %xmm4, %xmm8
-; SSE2-NEXT: pcmpgtd %xmm7, %xmm8
-; SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm3, %xmm6
-; SSE2-NEXT: pand %xmm8, %xmm6
-; SSE2-NEXT: pand %xmm6, %xmm5
-; SSE2-NEXT: pandn %xmm2, %xmm6
-; SSE2-NEXT: por %xmm5, %xmm6
-; SSE2-NEXT: movdqa %xmm1, %xmm5
-; SSE2-NEXT: pxor %xmm3, %xmm5
-; SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm5[0,0,2,2]
-; SSE2-NEXT: pcmpgtd %xmm7, %xmm4
-; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm5[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm3, %xmm5
-; SSE2-NEXT: pand %xmm4, %xmm5
-; SSE2-NEXT: pand %xmm5, %xmm1
-; SSE2-NEXT: pandn %xmm2, %xmm5
-; SSE2-NEXT: por %xmm1, %xmm5
-; SSE2-NEXT: packuswb %xmm6, %xmm5
-; SSE2-NEXT: packuswb %xmm5, %xmm0
-; SSE2-NEXT: packuswb %xmm0, %xmm0
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_usat_v8i64_v8i8:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa (%rdi), %xmm6
-; SSSE3-NEXT: movdqa 16(%rdi), %xmm0
-; SSSE3-NEXT: movdqa 32(%rdi), %xmm1
-; SSSE3-NEXT: movdqa 48(%rdi), %xmm5
-; SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [255,255]
-; SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [9223372039002259456,9223372039002259456]
-; SSSE3-NEXT: movdqa %xmm0, %xmm7
-; SSSE3-NEXT: pxor %xmm3, %xmm7
-; SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm7[0,0,2,2]
-; SSSE3-NEXT: movdqa {{.*#+}} xmm4 = [2147483903,2147483903,2147483903,2147483903]
-; SSSE3-NEXT: movdqa %xmm4, %xmm9
-; SSSE3-NEXT: pcmpgtd %xmm8, %xmm9
-; SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm7[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm3, %xmm7
-; SSSE3-NEXT: pand %xmm9, %xmm7
-; SSSE3-NEXT: pand %xmm7, %xmm0
-; SSSE3-NEXT: pandn %xmm2, %xmm7
-; SSSE3-NEXT: por %xmm0, %xmm7
-; SSSE3-NEXT: movdqa %xmm6, %xmm0
-; SSSE3-NEXT: pxor %xmm3, %xmm0
-; SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm0[0,0,2,2]
-; SSSE3-NEXT: movdqa %xmm4, %xmm9
-; SSSE3-NEXT: pcmpgtd %xmm8, %xmm9
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm3, %xmm0
-; SSSE3-NEXT: pand %xmm9, %xmm0
-; SSSE3-NEXT: pand %xmm0, %xmm6
-; SSSE3-NEXT: pandn %xmm2, %xmm0
-; SSSE3-NEXT: por %xmm6, %xmm0
-; SSSE3-NEXT: packuswb %xmm7, %xmm0
-; SSSE3-NEXT: movdqa %xmm5, %xmm6
-; SSSE3-NEXT: pxor %xmm3, %xmm6
-; SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm6[0,0,2,2]
-; SSSE3-NEXT: movdqa %xmm4, %xmm8
-; SSSE3-NEXT: pcmpgtd %xmm7, %xmm8
-; SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm3, %xmm6
-; SSSE3-NEXT: pand %xmm8, %xmm6
-; SSSE3-NEXT: pand %xmm6, %xmm5
-; SSSE3-NEXT: pandn %xmm2, %xmm6
-; SSSE3-NEXT: por %xmm5, %xmm6
-; SSSE3-NEXT: movdqa %xmm1, %xmm5
-; SSSE3-NEXT: pxor %xmm3, %xmm5
-; SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm5[0,0,2,2]
-; SSSE3-NEXT: pcmpgtd %xmm7, %xmm4
-; SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm5[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm3, %xmm5
-; SSSE3-NEXT: pand %xmm4, %xmm5
-; SSSE3-NEXT: pand %xmm5, %xmm1
-; SSSE3-NEXT: pandn %xmm2, %xmm5
-; SSSE3-NEXT: por %xmm1, %xmm5
-; SSSE3-NEXT: packuswb %xmm6, %xmm5
-; SSSE3-NEXT: packuswb %xmm5, %xmm0
-; SSSE3-NEXT: packuswb %xmm0, %xmm0
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc_usat_v8i64_v8i8:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa (%rdi), %xmm6
+; SSE2-SSSE3-NEXT: movdqa 16(%rdi), %xmm0
+; SSE2-SSSE3-NEXT: movdqa 32(%rdi), %xmm1
+; SSE2-SSSE3-NEXT: movdqa 48(%rdi), %xmm5
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [255,255]
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [9223372039002259456,9223372039002259456]
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm7
+; SSE2-SSSE3-NEXT: pxor %xmm3, %xmm7
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm7[0,0,2,2]
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm4 = [2147483903,2147483903,2147483903,2147483903]
+; SSE2-SSSE3-NEXT: movdqa %xmm4, %xmm9
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm8, %xmm9
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm7[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm3, %xmm7
+; SSE2-SSSE3-NEXT: pand %xmm9, %xmm7
+; SSE2-SSSE3-NEXT: pand %xmm7, %xmm0
+; SSE2-SSSE3-NEXT: pandn %xmm2, %xmm7
+; SSE2-SSSE3-NEXT: por %xmm0, %xmm7
+; SSE2-SSSE3-NEXT: movdqa %xmm6, %xmm0
+; SSE2-SSSE3-NEXT: pxor %xmm3, %xmm0
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm0[0,0,2,2]
+; SSE2-SSSE3-NEXT: movdqa %xmm4, %xmm9
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm8, %xmm9
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm3, %xmm0
+; SSE2-SSSE3-NEXT: pand %xmm9, %xmm0
+; SSE2-SSSE3-NEXT: pand %xmm0, %xmm6
+; SSE2-SSSE3-NEXT: pandn %xmm2, %xmm0
+; SSE2-SSSE3-NEXT: por %xmm6, %xmm0
+; SSE2-SSSE3-NEXT: packuswb %xmm7, %xmm0
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm6
+; SSE2-SSSE3-NEXT: pxor %xmm3, %xmm6
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm6[0,0,2,2]
+; SSE2-SSSE3-NEXT: movdqa %xmm4, %xmm8
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm7, %xmm8
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm3, %xmm6
+; SSE2-SSSE3-NEXT: pand %xmm8, %xmm6
+; SSE2-SSSE3-NEXT: pand %xmm6, %xmm5
+; SSE2-SSSE3-NEXT: pandn %xmm2, %xmm6
+; SSE2-SSSE3-NEXT: por %xmm5, %xmm6
+; SSE2-SSSE3-NEXT: movdqa %xmm1, %xmm5
+; SSE2-SSSE3-NEXT: pxor %xmm3, %xmm5
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm5[0,0,2,2]
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm7, %xmm4
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm5[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm3, %xmm5
+; SSE2-SSSE3-NEXT: pand %xmm4, %xmm5
+; SSE2-SSSE3-NEXT: pand %xmm5, %xmm1
+; SSE2-SSSE3-NEXT: pandn %xmm2, %xmm5
+; SSE2-SSSE3-NEXT: por %xmm1, %xmm5
+; SSE2-SSSE3-NEXT: packuswb %xmm6, %xmm5
+; SSE2-SSSE3-NEXT: packuswb %xmm5, %xmm0
+; SSE2-SSSE3-NEXT: packuswb %xmm0, %xmm0
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_usat_v8i64_v8i8:
; SSE41: # %bb.0:
@@ -2829,123 +2435,64 @@ define <8 x i8> @trunc_usat_v8i64_v8i8(ptr %p0) {
}
define void @trunc_usat_v8i64_v8i8_store(ptr %p0, ptr%p1) {
-; SSE2-LABEL: trunc_usat_v8i64_v8i8_store:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa (%rdi), %xmm6
-; SSE2-NEXT: movdqa 16(%rdi), %xmm5
-; SSE2-NEXT: movdqa 32(%rdi), %xmm0
-; SSE2-NEXT: movdqa 48(%rdi), %xmm4
-; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [255,255]
-; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [9223372039002259456,9223372039002259456]
-; SSE2-NEXT: movdqa %xmm5, %xmm7
-; SSE2-NEXT: pxor %xmm2, %xmm7
-; SSE2-NEXT: pshufd {{.*#+}} xmm8 = xmm7[0,0,2,2]
-; SSE2-NEXT: movdqa {{.*#+}} xmm3 = [2147483903,2147483903,2147483903,2147483903]
-; SSE2-NEXT: movdqa %xmm3, %xmm9
-; SSE2-NEXT: pcmpgtd %xmm8, %xmm9
-; SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm7[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm2, %xmm7
-; SSE2-NEXT: pand %xmm9, %xmm7
-; SSE2-NEXT: pand %xmm7, %xmm5
-; SSE2-NEXT: pandn %xmm1, %xmm7
-; SSE2-NEXT: por %xmm5, %xmm7
-; SSE2-NEXT: movdqa %xmm6, %xmm5
-; SSE2-NEXT: pxor %xmm2, %xmm5
-; SSE2-NEXT: pshufd {{.*#+}} xmm8 = xmm5[0,0,2,2]
-; SSE2-NEXT: movdqa %xmm3, %xmm9
-; SSE2-NEXT: pcmpgtd %xmm8, %xmm9
-; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm5[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm2, %xmm5
-; SSE2-NEXT: pand %xmm9, %xmm5
-; SSE2-NEXT: pand %xmm5, %xmm6
-; SSE2-NEXT: pandn %xmm1, %xmm5
-; SSE2-NEXT: por %xmm6, %xmm5
-; SSE2-NEXT: packuswb %xmm7, %xmm5
-; SSE2-NEXT: movdqa %xmm4, %xmm6
-; SSE2-NEXT: pxor %xmm2, %xmm6
-; SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm6[0,0,2,2]
-; SSE2-NEXT: movdqa %xmm3, %xmm8
-; SSE2-NEXT: pcmpgtd %xmm7, %xmm8
-; SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm2, %xmm6
-; SSE2-NEXT: pand %xmm8, %xmm6
-; SSE2-NEXT: pand %xmm6, %xmm4
-; SSE2-NEXT: pandn %xmm1, %xmm6
-; SSE2-NEXT: por %xmm4, %xmm6
-; SSE2-NEXT: movdqa %xmm0, %xmm4
-; SSE2-NEXT: pxor %xmm2, %xmm4
-; SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm4[0,0,2,2]
-; SSE2-NEXT: pcmpgtd %xmm7, %xmm3
-; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm2, %xmm4
-; SSE2-NEXT: pand %xmm3, %xmm4
-; SSE2-NEXT: pand %xmm4, %xmm0
-; SSE2-NEXT: pandn %xmm1, %xmm4
-; SSE2-NEXT: por %xmm0, %xmm4
-; SSE2-NEXT: packuswb %xmm6, %xmm4
-; SSE2-NEXT: packuswb %xmm4, %xmm5
-; SSE2-NEXT: packuswb %xmm5, %xmm5
-; SSE2-NEXT: movq %xmm5, (%rsi)
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_usat_v8i64_v8i8_store:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa (%rdi), %xmm6
-; SSSE3-NEXT: movdqa 16(%rdi), %xmm5
-; SSSE3-NEXT: movdqa 32(%rdi), %xmm0
-; SSSE3-NEXT: movdqa 48(%rdi), %xmm4
-; SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [255,255]
-; SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [9223372039002259456,9223372039002259456]
-; SSSE3-NEXT: movdqa %xmm5, %xmm7
-; SSSE3-NEXT: pxor %xmm2, %xmm7
-; SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm7[0,0,2,2]
-; SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [2147483903,2147483903,2147483903,2147483903]
-; SSSE3-NEXT: movdqa %xmm3, %xmm9
-; SSSE3-NEXT: pcmpgtd %xmm8, %xmm9
-; SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm7[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm2, %xmm7
-; SSSE3-NEXT: pand %xmm9, %xmm7
-; SSSE3-NEXT: pand %xmm7, %xmm5
-; SSSE3-NEXT: pandn %xmm1, %xmm7
-; SSSE3-NEXT: por %xmm5, %xmm7
-; SSSE3-NEXT: movdqa %xmm6, %xmm5
-; SSSE3-NEXT: pxor %xmm2, %xmm5
-; SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm5[0,0,2,2]
-; SSSE3-NEXT: movdqa %xmm3, %xmm9
-; SSSE3-NEXT: pcmpgtd %xmm8, %xmm9
-; SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm5[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm2, %xmm5
-; SSSE3-NEXT: pand %xmm9, %xmm5
-; SSSE3-NEXT: pand %xmm5, %xmm6
-; SSSE3-NEXT: pandn %xmm1, %xmm5
-; SSSE3-NEXT: por %xmm6, %xmm5
-; SSSE3-NEXT: packuswb %xmm7, %xmm5
-; SSSE3-NEXT: movdqa %xmm4, %xmm6
-; SSSE3-NEXT: pxor %xmm2, %xmm6
-; SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm6[0,0,2,2]
-; SSSE3-NEXT: movdqa %xmm3, %xmm8
-; SSSE3-NEXT: pcmpgtd %xmm7, %xmm8
-; SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm2, %xmm6
-; SSSE3-NEXT: pand %xmm8, %xmm6
-; SSSE3-NEXT: pand %xmm6, %xmm4
-; SSSE3-NEXT: pandn %xmm1, %xmm6
-; SSSE3-NEXT: por %xmm4, %xmm6
-; SSSE3-NEXT: movdqa %xmm0, %xmm4
-; SSSE3-NEXT: pxor %xmm2, %xmm4
-; SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm4[0,0,2,2]
-; SSSE3-NEXT: pcmpgtd %xmm7, %xmm3
-; SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm2, %xmm4
-; SSSE3-NEXT: pand %xmm3, %xmm4
-; SSSE3-NEXT: pand %xmm4, %xmm0
-; SSSE3-NEXT: pandn %xmm1, %xmm4
-; SSSE3-NEXT: por %xmm0, %xmm4
-; SSSE3-NEXT: packuswb %xmm6, %xmm4
-; SSSE3-NEXT: packuswb %xmm4, %xmm5
-; SSSE3-NEXT: packuswb %xmm5, %xmm5
-; SSSE3-NEXT: movq %xmm5, (%rsi)
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc_usat_v8i64_v8i8_store:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa (%rdi), %xmm6
+; SSE2-SSSE3-NEXT: movdqa 16(%rdi), %xmm5
+; SSE2-SSSE3-NEXT: movdqa 32(%rdi), %xmm0
+; SSE2-SSSE3-NEXT: movdqa 48(%rdi), %xmm4
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [255,255]
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [9223372039002259456,9223372039002259456]
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm7
+; SSE2-SSSE3-NEXT: pxor %xmm2, %xmm7
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm7[0,0,2,2]
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [2147483903,2147483903,2147483903,2147483903]
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm9
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm8, %xmm9
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm7[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm2, %xmm7
+; SSE2-SSSE3-NEXT: pand %xmm9, %xmm7
+; SSE2-SSSE3-NEXT: pand %xmm7, %xmm5
+; SSE2-SSSE3-NEXT: pandn %xmm1, %xmm7
+; SSE2-SSSE3-NEXT: por %xmm5, %xmm7
+; SSE2-SSSE3-NEXT: movdqa %xmm6, %xmm5
+; SSE2-SSSE3-NEXT: pxor %xmm2, %xmm5
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm5[0,0,2,2]
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm9
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm8, %xmm9
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm5 = xmm5[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm2, %xmm5
+; SSE2-SSSE3-NEXT: pand %xmm9, %xmm5
+; SSE2-SSSE3-NEXT: pand %xmm5, %xmm6
+; SSE2-SSSE3-NEXT: pandn %xmm1, %xmm5
+; SSE2-SSSE3-NEXT: por %xmm6, %xmm5
+; SSE2-SSSE3-NEXT: packuswb %xmm7, %xmm5
+; SSE2-SSSE3-NEXT: movdqa %xmm4, %xmm6
+; SSE2-SSSE3-NEXT: pxor %xmm2, %xmm6
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm6[0,0,2,2]
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm8
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm7, %xmm8
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm2, %xmm6
+; SSE2-SSSE3-NEXT: pand %xmm8, %xmm6
+; SSE2-SSSE3-NEXT: pand %xmm6, %xmm4
+; SSE2-SSSE3-NEXT: pandn %xmm1, %xmm6
+; SSE2-SSSE3-NEXT: por %xmm4, %xmm6
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm4
+; SSE2-SSSE3-NEXT: pxor %xmm2, %xmm4
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm4[0,0,2,2]
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm7, %xmm3
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm2, %xmm4
+; SSE2-SSSE3-NEXT: pand %xmm3, %xmm4
+; SSE2-SSSE3-NEXT: pand %xmm4, %xmm0
+; SSE2-SSSE3-NEXT: pandn %xmm1, %xmm4
+; SSE2-SSSE3-NEXT: por %xmm0, %xmm4
+; SSE2-SSSE3-NEXT: packuswb %xmm6, %xmm4
+; SSE2-SSSE3-NEXT: packuswb %xmm4, %xmm5
+; SSE2-SSSE3-NEXT: packuswb %xmm5, %xmm5
+; SSE2-SSSE3-NEXT: movq %xmm5, (%rsi)
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_usat_v8i64_v8i8_store:
; SSE41: # %bb.0:
@@ -3076,223 +2623,114 @@ define void @trunc_usat_v8i64_v8i8_store(ptr %p0, ptr%p1) {
}
define <16 x i8> @trunc_usat_v16i64_v16i8(ptr %p0) {
-; SSE2-LABEL: trunc_usat_v16i64_v16i8:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa 96(%rdi), %xmm1
-; SSE2-NEXT: movdqa 112(%rdi), %xmm4
-; SSE2-NEXT: movdqa 64(%rdi), %xmm6
-; SSE2-NEXT: movdqa 80(%rdi), %xmm7
-; SSE2-NEXT: movdqa (%rdi), %xmm10
-; SSE2-NEXT: movdqa 16(%rdi), %xmm0
-; SSE2-NEXT: movdqa 32(%rdi), %xmm8
-; SSE2-NEXT: movdqa 48(%rdi), %xmm9
-; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [255,255]
-; SSE2-NEXT: movdqa {{.*#+}} xmm3 = [9223372039002259456,9223372039002259456]
-; SSE2-NEXT: movdqa %xmm0, %xmm11
-; SSE2-NEXT: pxor %xmm3, %xmm11
-; SSE2-NEXT: pshufd {{.*#+}} xmm12 = xmm11[0,0,2,2]
-; SSE2-NEXT: movdqa {{.*#+}} xmm5 = [2147483903,2147483903,2147483903,2147483903]
-; SSE2-NEXT: movdqa %xmm5, %xmm13
-; SSE2-NEXT: pcmpgtd %xmm12, %xmm13
-; SSE2-NEXT: pshufd {{.*#+}} xmm11 = xmm11[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm3, %xmm11
-; SSE2-NEXT: pand %xmm13, %xmm11
-; SSE2-NEXT: pand %xmm11, %xmm0
-; SSE2-NEXT: pandn %xmm2, %xmm11
-; SSE2-NEXT: por %xmm0, %xmm11
-; SSE2-NEXT: movdqa %xmm10, %xmm0
-; SSE2-NEXT: pxor %xmm3, %xmm0
-; SSE2-NEXT: pshufd {{.*#+}} xmm12 = xmm0[0,0,2,2]
-; SSE2-NEXT: movdqa %xmm5, %xmm13
-; SSE2-NEXT: pcmpgtd %xmm12, %xmm13
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm3, %xmm0
-; SSE2-NEXT: pand %xmm13, %xmm0
-; SSE2-NEXT: pand %xmm0, %xmm10
-; SSE2-NEXT: pandn %xmm2, %xmm0
-; SSE2-NEXT: por %xmm10, %xmm0
-; SSE2-NEXT: packuswb %xmm11, %xmm0
-; SSE2-NEXT: movdqa %xmm9, %xmm10
-; SSE2-NEXT: pxor %xmm3, %xmm10
-; SSE2-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
-; SSE2-NEXT: movdqa %xmm5, %xmm12
-; SSE2-NEXT: pcmpgtd %xmm11, %xmm12
-; SSE2-NEXT: pshufd {{.*#+}} xmm10 = xmm10[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm3, %xmm10
-; SSE2-NEXT: pand %xmm12, %xmm10
-; SSE2-NEXT: pand %xmm10, %xmm9
-; SSE2-NEXT: pandn %xmm2, %xmm10
-; SSE2-NEXT: por %xmm9, %xmm10
-; SSE2-NEXT: movdqa %xmm8, %xmm9
-; SSE2-NEXT: pxor %xmm3, %xmm9
-; SSE2-NEXT: pshufd {{.*#+}} xmm11 = xmm9[0,0,2,2]
-; SSE2-NEXT: movdqa %xmm5, %xmm12
-; SSE2-NEXT: pcmpgtd %xmm11, %xmm12
-; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm9[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm3, %xmm9
-; SSE2-NEXT: pand %xmm12, %xmm9
-; SSE2-NEXT: pand %xmm9, %xmm8
-; SSE2-NEXT: pandn %xmm2, %xmm9
-; SSE2-NEXT: por %xmm8, %xmm9
-; SSE2-NEXT: packuswb %xmm10, %xmm9
-; SSE2-NEXT: packuswb %xmm9, %xmm0
-; SSE2-NEXT: movdqa %xmm7, %xmm8
-; SSE2-NEXT: pxor %xmm3, %xmm8
-; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm8[0,0,2,2]
-; SSE2-NEXT: movdqa %xmm5, %xmm10
-; SSE2-NEXT: pcmpgtd %xmm9, %xmm10
-; SSE2-NEXT: pshufd {{.*#+}} xmm8 = xmm8[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm3, %xmm8
-; SSE2-NEXT: pand %xmm10, %xmm8
-; SSE2-NEXT: pand %xmm8, %xmm7
-; SSE2-NEXT: pandn %xmm2, %xmm8
-; SSE2-NEXT: por %xmm7, %xmm8
-; SSE2-NEXT: movdqa %xmm6, %xmm7
-; SSE2-NEXT: pxor %xmm3, %xmm7
-; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm7[0,0,2,2]
-; SSE2-NEXT: movdqa %xmm5, %xmm10
-; SSE2-NEXT: pcmpgtd %xmm9, %xmm10
-; SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm7[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm3, %xmm7
-; SSE2-NEXT: pand %xmm10, %xmm7
-; SSE2-NEXT: pand %xmm7, %xmm6
-; SSE2-NEXT: pandn %xmm2, %xmm7
-; SSE2-NEXT: por %xmm6, %xmm7
-; SSE2-NEXT: packuswb %xmm8, %xmm7
-; SSE2-NEXT: movdqa %xmm4, %xmm6
-; SSE2-NEXT: pxor %xmm3, %xmm6
-; SSE2-NEXT: pshufd {{.*#+}} xmm8 = xmm6[0,0,2,2]
-; SSE2-NEXT: movdqa %xmm5, %xmm9
-; SSE2-NEXT: pcmpgtd %xmm8, %xmm9
-; SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm3, %xmm6
-; SSE2-NEXT: pand %xmm9, %xmm6
-; SSE2-NEXT: pand %xmm6, %xmm4
-; SSE2-NEXT: pandn %xmm2, %xmm6
-; SSE2-NEXT: por %xmm4, %xmm6
-; SSE2-NEXT: movdqa %xmm1, %xmm4
-; SSE2-NEXT: pxor %xmm3, %xmm4
-; SSE2-NEXT: pshufd {{.*#+}} xmm8 = xmm4[0,0,2,2]
-; SSE2-NEXT: pcmpgtd %xmm8, %xmm5
-; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
-; SSE2-NEXT: pcmpeqd %xmm3, %xmm4
-; SSE2-NEXT: pand %xmm5, %xmm4
-; SSE2-NEXT: pand %xmm4, %xmm1
-; SSE2-NEXT: pandn %xmm2, %xmm4
-; SSE2-NEXT: por %xmm1, %xmm4
-; SSE2-NEXT: packuswb %xmm6, %xmm4
-; SSE2-NEXT: packuswb %xmm4, %xmm7
-; SSE2-NEXT: packuswb %xmm7, %xmm0
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_usat_v16i64_v16i8:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa 96(%rdi), %xmm1
-; SSSE3-NEXT: movdqa 112(%rdi), %xmm4
-; SSSE3-NEXT: movdqa 64(%rdi), %xmm6
-; SSSE3-NEXT: movdqa 80(%rdi), %xmm7
-; SSSE3-NEXT: movdqa (%rdi), %xmm10
-; SSSE3-NEXT: movdqa 16(%rdi), %xmm0
-; SSSE3-NEXT: movdqa 32(%rdi), %xmm8
-; SSSE3-NEXT: movdqa 48(%rdi), %xmm9
-; SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [255,255]
-; SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [9223372039002259456,9223372039002259456]
-; SSSE3-NEXT: movdqa %xmm0, %xmm11
-; SSSE3-NEXT: pxor %xmm3, %xmm11
-; SSSE3-NEXT: pshufd {{.*#+}} xmm12 = xmm11[0,0,2,2]
-; SSSE3-NEXT: movdqa {{.*#+}} xmm5 = [2147483903,2147483903,2147483903,2147483903]
-; SSSE3-NEXT: movdqa %xmm5, %xmm13
-; SSSE3-NEXT: pcmpgtd %xmm12, %xmm13
-; SSSE3-NEXT: pshufd {{.*#+}} xmm11 = xmm11[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm3, %xmm11
-; SSSE3-NEXT: pand %xmm13, %xmm11
-; SSSE3-NEXT: pand %xmm11, %xmm0
-; SSSE3-NEXT: pandn %xmm2, %xmm11
-; SSSE3-NEXT: por %xmm0, %xmm11
-; SSSE3-NEXT: movdqa %xmm10, %xmm0
-; SSSE3-NEXT: pxor %xmm3, %xmm0
-; SSSE3-NEXT: pshufd {{.*#+}} xmm12 = xmm0[0,0,2,2]
-; SSSE3-NEXT: movdqa %xmm5, %xmm13
-; SSSE3-NEXT: pcmpgtd %xmm12, %xmm13
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm3, %xmm0
-; SSSE3-NEXT: pand %xmm13, %xmm0
-; SSSE3-NEXT: pand %xmm0, %xmm10
-; SSSE3-NEXT: pandn %xmm2, %xmm0
-; SSSE3-NEXT: por %xmm10, %xmm0
-; SSSE3-NEXT: packuswb %xmm11, %xmm0
-; SSSE3-NEXT: movdqa %xmm9, %xmm10
-; SSSE3-NEXT: pxor %xmm3, %xmm10
-; SSSE3-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
-; SSSE3-NEXT: movdqa %xmm5, %xmm12
-; SSSE3-NEXT: pcmpgtd %xmm11, %xmm12
-; SSSE3-NEXT: pshufd {{.*#+}} xmm10 = xmm10[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm3, %xmm10
-; SSSE3-NEXT: pand %xmm12, %xmm10
-; SSSE3-NEXT: pand %xmm10, %xmm9
-; SSSE3-NEXT: pandn %xmm2, %xmm10
-; SSSE3-NEXT: por %xmm9, %xmm10
-; SSSE3-NEXT: movdqa %xmm8, %xmm9
-; SSSE3-NEXT: pxor %xmm3, %xmm9
-; SSSE3-NEXT: pshufd {{.*#+}} xmm11 = xmm9[0,0,2,2]
-; SSSE3-NEXT: movdqa %xmm5, %xmm12
-; SSSE3-NEXT: pcmpgtd %xmm11, %xmm12
-; SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm9[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm3, %xmm9
-; SSSE3-NEXT: pand %xmm12, %xmm9
-; SSSE3-NEXT: pand %xmm9, %xmm8
-; SSSE3-NEXT: pandn %xmm2, %xmm9
-; SSSE3-NEXT: por %xmm8, %xmm9
-; SSSE3-NEXT: packuswb %xmm10, %xmm9
-; SSSE3-NEXT: packuswb %xmm9, %xmm0
-; SSSE3-NEXT: movdqa %xmm7, %xmm8
-; SSSE3-NEXT: pxor %xmm3, %xmm8
-; SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm8[0,0,2,2]
-; SSSE3-NEXT: movdqa %xmm5, %xmm10
-; SSSE3-NEXT: pcmpgtd %xmm9, %xmm10
-; SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm8[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm3, %xmm8
-; SSSE3-NEXT: pand %xmm10, %xmm8
-; SSSE3-NEXT: pand %xmm8, %xmm7
-; SSSE3-NEXT: pandn %xmm2, %xmm8
-; SSSE3-NEXT: por %xmm7, %xmm8
-; SSSE3-NEXT: movdqa %xmm6, %xmm7
-; SSSE3-NEXT: pxor %xmm3, %xmm7
-; SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm7[0,0,2,2]
-; SSSE3-NEXT: movdqa %xmm5, %xmm10
-; SSSE3-NEXT: pcmpgtd %xmm9, %xmm10
-; SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm7[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm3, %xmm7
-; SSSE3-NEXT: pand %xmm10, %xmm7
-; SSSE3-NEXT: pand %xmm7, %xmm6
-; SSSE3-NEXT: pandn %xmm2, %xmm7
-; SSSE3-NEXT: por %xmm6, %xmm7
-; SSSE3-NEXT: packuswb %xmm8, %xmm7
-; SSSE3-NEXT: movdqa %xmm4, %xmm6
-; SSSE3-NEXT: pxor %xmm3, %xmm6
-; SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm6[0,0,2,2]
-; SSSE3-NEXT: movdqa %xmm5, %xmm9
-; SSSE3-NEXT: pcmpgtd %xmm8, %xmm9
-; SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm3, %xmm6
-; SSSE3-NEXT: pand %xmm9, %xmm6
-; SSSE3-NEXT: pand %xmm6, %xmm4
-; SSSE3-NEXT: pandn %xmm2, %xmm6
-; SSSE3-NEXT: por %xmm4, %xmm6
-; SSSE3-NEXT: movdqa %xmm1, %xmm4
-; SSSE3-NEXT: pxor %xmm3, %xmm4
-; SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm4[0,0,2,2]
-; SSSE3-NEXT: pcmpgtd %xmm8, %xmm5
-; SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
-; SSSE3-NEXT: pcmpeqd %xmm3, %xmm4
-; SSSE3-NEXT: pand %xmm5, %xmm4
-; SSSE3-NEXT: pand %xmm4, %xmm1
-; SSSE3-NEXT: pandn %xmm2, %xmm4
-; SSSE3-NEXT: por %xmm1, %xmm4
-; SSSE3-NEXT: packuswb %xmm6, %xmm4
-; SSSE3-NEXT: packuswb %xmm4, %xmm7
-; SSSE3-NEXT: packuswb %xmm7, %xmm0
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc_usat_v16i64_v16i8:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa 96(%rdi), %xmm1
+; SSE2-SSSE3-NEXT: movdqa 112(%rdi), %xmm4
+; SSE2-SSSE3-NEXT: movdqa 64(%rdi), %xmm6
+; SSE2-SSSE3-NEXT: movdqa 80(%rdi), %xmm7
+; SSE2-SSSE3-NEXT: movdqa (%rdi), %xmm10
+; SSE2-SSSE3-NEXT: movdqa 16(%rdi), %xmm0
+; SSE2-SSSE3-NEXT: movdqa 32(%rdi), %xmm8
+; SSE2-SSSE3-NEXT: movdqa 48(%rdi), %xmm9
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [255,255]
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [9223372039002259456,9223372039002259456]
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm11
+; SSE2-SSSE3-NEXT: pxor %xmm3, %xmm11
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm12 = xmm11[0,0,2,2]
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm5 = [2147483903,2147483903,2147483903,2147483903]
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm13
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm12, %xmm13
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm11 = xmm11[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm3, %xmm11
+; SSE2-SSSE3-NEXT: pand %xmm13, %xmm11
+; SSE2-SSSE3-NEXT: pand %xmm11, %xmm0
+; SSE2-SSSE3-NEXT: pandn %xmm2, %xmm11
+; SSE2-SSSE3-NEXT: por %xmm0, %xmm11
+; SSE2-SSSE3-NEXT: movdqa %xmm10, %xmm0
+; SSE2-SSSE3-NEXT: pxor %xmm3, %xmm0
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm12 = xmm0[0,0,2,2]
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm13
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm12, %xmm13
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm3, %xmm0
+; SSE2-SSSE3-NEXT: pand %xmm13, %xmm0
+; SSE2-SSSE3-NEXT: pand %xmm0, %xmm10
+; SSE2-SSSE3-NEXT: pandn %xmm2, %xmm0
+; SSE2-SSSE3-NEXT: por %xmm10, %xmm0
+; SSE2-SSSE3-NEXT: packuswb %xmm11, %xmm0
+; SSE2-SSSE3-NEXT: movdqa %xmm9, %xmm10
+; SSE2-SSSE3-NEXT: pxor %xmm3, %xmm10
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm11 = xmm10[0,0,2,2]
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm12
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm11, %xmm12
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm10 = xmm10[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm3, %xmm10
+; SSE2-SSSE3-NEXT: pand %xmm12, %xmm10
+; SSE2-SSSE3-NEXT: pand %xmm10, %xmm9
+; SSE2-SSSE3-NEXT: pandn %xmm2, %xmm10
+; SSE2-SSSE3-NEXT: por %xmm9, %xmm10
+; SSE2-SSSE3-NEXT: movdqa %xmm8, %xmm9
+; SSE2-SSSE3-NEXT: pxor %xmm3, %xmm9
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm11 = xmm9[0,0,2,2]
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm12
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm11, %xmm12
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm9[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm3, %xmm9
+; SSE2-SSSE3-NEXT: pand %xmm12, %xmm9
+; SSE2-SSSE3-NEXT: pand %xmm9, %xmm8
+; SSE2-SSSE3-NEXT: pandn %xmm2, %xmm9
+; SSE2-SSSE3-NEXT: por %xmm8, %xmm9
+; SSE2-SSSE3-NEXT: packuswb %xmm10, %xmm9
+; SSE2-SSSE3-NEXT: packuswb %xmm9, %xmm0
+; SSE2-SSSE3-NEXT: movdqa %xmm7, %xmm8
+; SSE2-SSSE3-NEXT: pxor %xmm3, %xmm8
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm8[0,0,2,2]
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm10
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm9, %xmm10
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm8[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm3, %xmm8
+; SSE2-SSSE3-NEXT: pand %xmm10, %xmm8
+; SSE2-SSSE3-NEXT: pand %xmm8, %xmm7
+; SSE2-SSSE3-NEXT: pandn %xmm2, %xmm8
+; SSE2-SSSE3-NEXT: por %xmm7, %xmm8
+; SSE2-SSSE3-NEXT: movdqa %xmm6, %xmm7
+; SSE2-SSSE3-NEXT: pxor %xmm3, %xmm7
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm9 = xmm7[0,0,2,2]
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm10
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm9, %xmm10
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm7 = xmm7[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm3, %xmm7
+; SSE2-SSSE3-NEXT: pand %xmm10, %xmm7
+; SSE2-SSSE3-NEXT: pand %xmm7, %xmm6
+; SSE2-SSSE3-NEXT: pandn %xmm2, %xmm7
+; SSE2-SSSE3-NEXT: por %xmm6, %xmm7
+; SSE2-SSSE3-NEXT: packuswb %xmm8, %xmm7
+; SSE2-SSSE3-NEXT: movdqa %xmm4, %xmm6
+; SSE2-SSSE3-NEXT: pxor %xmm3, %xmm6
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm6[0,0,2,2]
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm9
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm8, %xmm9
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm3, %xmm6
+; SSE2-SSSE3-NEXT: pand %xmm9, %xmm6
+; SSE2-SSSE3-NEXT: pand %xmm6, %xmm4
+; SSE2-SSSE3-NEXT: pandn %xmm2, %xmm6
+; SSE2-SSSE3-NEXT: por %xmm4, %xmm6
+; SSE2-SSSE3-NEXT: movdqa %xmm1, %xmm4
+; SSE2-SSSE3-NEXT: pxor %xmm3, %xmm4
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm8 = xmm4[0,0,2,2]
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm8, %xmm5
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
+; SSE2-SSSE3-NEXT: pcmpeqd %xmm3, %xmm4
+; SSE2-SSSE3-NEXT: pand %xmm5, %xmm4
+; SSE2-SSSE3-NEXT: pand %xmm4, %xmm1
+; SSE2-SSSE3-NEXT: pandn %xmm2, %xmm4
+; SSE2-SSSE3-NEXT: por %xmm1, %xmm4
+; SSE2-SSSE3-NEXT: packuswb %xmm6, %xmm4
+; SSE2-SSSE3-NEXT: packuswb %xmm4, %xmm7
+; SSE2-SSSE3-NEXT: packuswb %xmm7, %xmm0
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_usat_v16i64_v16i8:
; SSE41: # %bb.0:
@@ -3689,47 +3127,26 @@ define void @trunc_usat_v4i32_v4i8_store(<4 x i32> %a0, ptr%p1) {
}
define <8 x i8> @trunc_usat_v8i32_v8i8(<8 x i32> %a0) {
-; SSE2-LABEL: trunc_usat_v8i32_v8i8:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [255,255,255,255]
-; SSE2-NEXT: movdqa {{.*#+}} xmm3 = [2147483648,2147483648,2147483648,2147483648]
-; SSE2-NEXT: movdqa %xmm1, %xmm4
-; SSE2-NEXT: pxor %xmm3, %xmm4
-; SSE2-NEXT: movdqa {{.*#+}} xmm5 = [2147483903,2147483903,2147483903,2147483903]
-; SSE2-NEXT: movdqa %xmm5, %xmm6
-; SSE2-NEXT: pcmpgtd %xmm4, %xmm6
-; SSE2-NEXT: pand %xmm6, %xmm1
-; SSE2-NEXT: pandn %xmm2, %xmm6
-; SSE2-NEXT: por %xmm1, %xmm6
-; SSE2-NEXT: pxor %xmm0, %xmm3
-; SSE2-NEXT: pcmpgtd %xmm3, %xmm5
-; SSE2-NEXT: pand %xmm5, %xmm0
-; SSE2-NEXT: pandn %xmm2, %xmm5
-; SSE2-NEXT: por %xmm5, %xmm0
-; SSE2-NEXT: packuswb %xmm6, %xmm0
-; SSE2-NEXT: packuswb %xmm0, %xmm0
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_usat_v8i32_v8i8:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [255,255,255,255]
-; SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [2147483648,2147483648,2147483648,2147483648]
-; SSSE3-NEXT: movdqa %xmm1, %xmm4
-; SSSE3-NEXT: pxor %xmm3, %xmm4
-; SSSE3-NEXT: movdqa {{.*#+}} xmm5 = [2147483903,2147483903,2147483903,2147483903]
-; SSSE3-NEXT: movdqa %xmm5, %xmm6
-; SSSE3-NEXT: pcmpgtd %xmm4, %xmm6
-; SSSE3-NEXT: pand %xmm6, %xmm1
-; SSSE3-NEXT: pandn %xmm2, %xmm6
-; SSSE3-NEXT: por %xmm1, %xmm6
-; SSSE3-NEXT: pxor %xmm0, %xmm3
-; SSSE3-NEXT: pcmpgtd %xmm3, %xmm5
-; SSSE3-NEXT: pand %xmm5, %xmm0
-; SSSE3-NEXT: pandn %xmm2, %xmm5
-; SSSE3-NEXT: por %xmm5, %xmm0
-; SSSE3-NEXT: packuswb %xmm6, %xmm0
-; SSSE3-NEXT: packuswb %xmm0, %xmm0
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc_usat_v8i32_v8i8:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [255,255,255,255]
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [2147483648,2147483648,2147483648,2147483648]
+; SSE2-SSSE3-NEXT: movdqa %xmm1, %xmm4
+; SSE2-SSSE3-NEXT: pxor %xmm3, %xmm4
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm5 = [2147483903,2147483903,2147483903,2147483903]
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm6
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm4, %xmm6
+; SSE2-SSSE3-NEXT: pand %xmm6, %xmm1
+; SSE2-SSSE3-NEXT: pandn %xmm2, %xmm6
+; SSE2-SSSE3-NEXT: por %xmm1, %xmm6
+; SSE2-SSSE3-NEXT: pxor %xmm0, %xmm3
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm3, %xmm5
+; SSE2-SSSE3-NEXT: pand %xmm5, %xmm0
+; SSE2-SSSE3-NEXT: pandn %xmm2, %xmm5
+; SSE2-SSSE3-NEXT: por %xmm5, %xmm0
+; SSE2-SSSE3-NEXT: packuswb %xmm6, %xmm0
+; SSE2-SSSE3-NEXT: packuswb %xmm0, %xmm0
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_usat_v8i32_v8i8:
; SSE41: # %bb.0:
@@ -3799,49 +3216,27 @@ define <8 x i8> @trunc_usat_v8i32_v8i8(<8 x i32> %a0) {
}
define void @trunc_usat_v8i32_v8i8_store(<8 x i32> %a0, ptr%p1) {
-; SSE2-LABEL: trunc_usat_v8i32_v8i8_store:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [255,255,255,255]
-; SSE2-NEXT: movdqa {{.*#+}} xmm3 = [2147483648,2147483648,2147483648,2147483648]
-; SSE2-NEXT: movdqa %xmm1, %xmm4
-; SSE2-NEXT: pxor %xmm3, %xmm4
-; SSE2-NEXT: movdqa {{.*#+}} xmm5 = [2147483903,2147483903,2147483903,2147483903]
-; SSE2-NEXT: movdqa %xmm5, %xmm6
-; SSE2-NEXT: pcmpgtd %xmm4, %xmm6
-; SSE2-NEXT: pand %xmm6, %xmm1
-; SSE2-NEXT: pandn %xmm2, %xmm6
-; SSE2-NEXT: por %xmm1, %xmm6
-; SSE2-NEXT: pxor %xmm0, %xmm3
-; SSE2-NEXT: pcmpgtd %xmm3, %xmm5
-; SSE2-NEXT: pand %xmm5, %xmm0
-; SSE2-NEXT: pandn %xmm2, %xmm5
-; SSE2-NEXT: por %xmm0, %xmm5
-; SSE2-NEXT: packuswb %xmm6, %xmm5
-; SSE2-NEXT: packuswb %xmm5, %xmm5
-; SSE2-NEXT: movq %xmm5, (%rdi)
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_usat_v8i32_v8i8_store:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [255,255,255,255]
-; SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [2147483648,2147483648,2147483648,2147483648]
-; SSSE3-NEXT: movdqa %xmm1, %xmm4
-; SSSE3-NEXT: pxor %xmm3, %xmm4
-; SSSE3-NEXT: movdqa {{.*#+}} xmm5 = [2147483903,2147483903,2147483903,2147483903]
-; SSSE3-NEXT: movdqa %xmm5, %xmm6
-; SSSE3-NEXT: pcmpgtd %xmm4, %xmm6
-; SSSE3-NEXT: pand %xmm6, %xmm1
-; SSSE3-NEXT: pandn %xmm2, %xmm6
-; SSSE3-NEXT: por %xmm1, %xmm6
-; SSSE3-NEXT: pxor %xmm0, %xmm3
-; SSSE3-NEXT: pcmpgtd %xmm3, %xmm5
-; SSSE3-NEXT: pand %xmm5, %xmm0
-; SSSE3-NEXT: pandn %xmm2, %xmm5
-; SSSE3-NEXT: por %xmm0, %xmm5
-; SSSE3-NEXT: packuswb %xmm6, %xmm5
-; SSSE3-NEXT: packuswb %xmm5, %xmm5
-; SSSE3-NEXT: movq %xmm5, (%rdi)
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc_usat_v8i32_v8i8_store:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [255,255,255,255]
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [2147483648,2147483648,2147483648,2147483648]
+; SSE2-SSSE3-NEXT: movdqa %xmm1, %xmm4
+; SSE2-SSSE3-NEXT: pxor %xmm3, %xmm4
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm5 = [2147483903,2147483903,2147483903,2147483903]
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm6
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm4, %xmm6
+; SSE2-SSSE3-NEXT: pand %xmm6, %xmm1
+; SSE2-SSSE3-NEXT: pandn %xmm2, %xmm6
+; SSE2-SSSE3-NEXT: por %xmm1, %xmm6
+; SSE2-SSSE3-NEXT: pxor %xmm0, %xmm3
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm3, %xmm5
+; SSE2-SSSE3-NEXT: pand %xmm5, %xmm0
+; SSE2-SSSE3-NEXT: pandn %xmm2, %xmm5
+; SSE2-SSSE3-NEXT: por %xmm0, %xmm5
+; SSE2-SSSE3-NEXT: packuswb %xmm6, %xmm5
+; SSE2-SSSE3-NEXT: packuswb %xmm5, %xmm5
+; SSE2-SSSE3-NEXT: movq %xmm5, (%rdi)
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_usat_v8i32_v8i8_store:
; SSE41: # %bb.0:
@@ -3917,85 +3312,45 @@ define void @trunc_usat_v8i32_v8i8_store(<8 x i32> %a0, ptr%p1) {
}
define <16 x i8> @trunc_usat_v16i32_v16i8(ptr %p0) {
-; SSE2-LABEL: trunc_usat_v16i32_v16i8:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa (%rdi), %xmm6
-; SSE2-NEXT: movdqa 16(%rdi), %xmm0
-; SSE2-NEXT: movdqa 32(%rdi), %xmm1
-; SSE2-NEXT: movdqa 48(%rdi), %xmm5
-; SSE2-NEXT: movdqa {{.*#+}} xmm3 = [255,255,255,255]
-; SSE2-NEXT: movdqa {{.*#+}} xmm4 = [2147483648,2147483648,2147483648,2147483648]
-; SSE2-NEXT: movdqa %xmm0, %xmm7
-; SSE2-NEXT: pxor %xmm4, %xmm7
-; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [2147483903,2147483903,2147483903,2147483903]
-; SSE2-NEXT: movdqa %xmm2, %xmm8
-; SSE2-NEXT: pcmpgtd %xmm7, %xmm8
-; SSE2-NEXT: pand %xmm8, %xmm0
-; SSE2-NEXT: pandn %xmm3, %xmm8
-; SSE2-NEXT: por %xmm0, %xmm8
-; SSE2-NEXT: movdqa %xmm6, %xmm7
-; SSE2-NEXT: pxor %xmm4, %xmm7
-; SSE2-NEXT: movdqa %xmm2, %xmm0
-; SSE2-NEXT: pcmpgtd %xmm7, %xmm0
-; SSE2-NEXT: pand %xmm0, %xmm6
-; SSE2-NEXT: pandn %xmm3, %xmm0
-; SSE2-NEXT: por %xmm6, %xmm0
-; SSE2-NEXT: packuswb %xmm8, %xmm0
-; SSE2-NEXT: movdqa %xmm5, %xmm6
-; SSE2-NEXT: pxor %xmm4, %xmm6
-; SSE2-NEXT: movdqa %xmm2, %xmm7
-; SSE2-NEXT: pcmpgtd %xmm6, %xmm7
-; SSE2-NEXT: pand %xmm7, %xmm5
-; SSE2-NEXT: pandn %xmm3, %xmm7
-; SSE2-NEXT: por %xmm5, %xmm7
-; SSE2-NEXT: pxor %xmm1, %xmm4
-; SSE2-NEXT: pcmpgtd %xmm4, %xmm2
-; SSE2-NEXT: pand %xmm2, %xmm1
-; SSE2-NEXT: pandn %xmm3, %xmm2
-; SSE2-NEXT: por %xmm1, %xmm2
-; SSE2-NEXT: packuswb %xmm7, %xmm2
-; SSE2-NEXT: packuswb %xmm2, %xmm0
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_usat_v16i32_v16i8:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa (%rdi), %xmm6
-; SSSE3-NEXT: movdqa 16(%rdi), %xmm0
-; SSSE3-NEXT: movdqa 32(%rdi), %xmm1
-; SSSE3-NEXT: movdqa 48(%rdi), %xmm5
-; SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [255,255,255,255]
-; SSSE3-NEXT: movdqa {{.*#+}} xmm4 = [2147483648,2147483648,2147483648,2147483648]
-; SSSE3-NEXT: movdqa %xmm0, %xmm7
-; SSSE3-NEXT: pxor %xmm4, %xmm7
-; SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [2147483903,2147483903,2147483903,2147483903]
-; SSSE3-NEXT: movdqa %xmm2, %xmm8
-; SSSE3-NEXT: pcmpgtd %xmm7, %xmm8
-; SSSE3-NEXT: pand %xmm8, %xmm0
-; SSSE3-NEXT: pandn %xmm3, %xmm8
-; SSSE3-NEXT: por %xmm0, %xmm8
-; SSSE3-NEXT: movdqa %xmm6, %xmm7
-; SSSE3-NEXT: pxor %xmm4, %xmm7
-; SSSE3-NEXT: movdqa %xmm2, %xmm0
-; SSSE3-NEXT: pcmpgtd %xmm7, %xmm0
-; SSSE3-NEXT: pand %xmm0, %xmm6
-; SSSE3-NEXT: pandn %xmm3, %xmm0
-; SSSE3-NEXT: por %xmm6, %xmm0
-; SSSE3-NEXT: packuswb %xmm8, %xmm0
-; SSSE3-NEXT: movdqa %xmm5, %xmm6
-; SSSE3-NEXT: pxor %xmm4, %xmm6
-; SSSE3-NEXT: movdqa %xmm2, %xmm7
-; SSSE3-NEXT: pcmpgtd %xmm6, %xmm7
-; SSSE3-NEXT: pand %xmm7, %xmm5
-; SSSE3-NEXT: pandn %xmm3, %xmm7
-; SSSE3-NEXT: por %xmm5, %xmm7
-; SSSE3-NEXT: pxor %xmm1, %xmm4
-; SSSE3-NEXT: pcmpgtd %xmm4, %xmm2
-; SSSE3-NEXT: pand %xmm2, %xmm1
-; SSSE3-NEXT: pandn %xmm3, %xmm2
-; SSSE3-NEXT: por %xmm1, %xmm2
-; SSSE3-NEXT: packuswb %xmm7, %xmm2
-; SSSE3-NEXT: packuswb %xmm2, %xmm0
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc_usat_v16i32_v16i8:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa (%rdi), %xmm6
+; SSE2-SSSE3-NEXT: movdqa 16(%rdi), %xmm0
+; SSE2-SSSE3-NEXT: movdqa 32(%rdi), %xmm1
+; SSE2-SSSE3-NEXT: movdqa 48(%rdi), %xmm5
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [255,255,255,255]
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm4 = [2147483648,2147483648,2147483648,2147483648]
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm7
+; SSE2-SSSE3-NEXT: pxor %xmm4, %xmm7
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [2147483903,2147483903,2147483903,2147483903]
+; SSE2-SSSE3-NEXT: movdqa %xmm2, %xmm8
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm7, %xmm8
+; SSE2-SSSE3-NEXT: pand %xmm8, %xmm0
+; SSE2-SSSE3-NEXT: pandn %xmm3, %xmm8
+; SSE2-SSSE3-NEXT: por %xmm0, %xmm8
+; SSE2-SSSE3-NEXT: movdqa %xmm6, %xmm7
+; SSE2-SSSE3-NEXT: pxor %xmm4, %xmm7
+; SSE2-SSSE3-NEXT: movdqa %xmm2, %xmm0
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm7, %xmm0
+; SSE2-SSSE3-NEXT: pand %xmm0, %xmm6
+; SSE2-SSSE3-NEXT: pandn %xmm3, %xmm0
+; SSE2-SSSE3-NEXT: por %xmm6, %xmm0
+; SSE2-SSSE3-NEXT: packuswb %xmm8, %xmm0
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm6
+; SSE2-SSSE3-NEXT: pxor %xmm4, %xmm6
+; SSE2-SSSE3-NEXT: movdqa %xmm2, %xmm7
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm6, %xmm7
+; SSE2-SSSE3-NEXT: pand %xmm7, %xmm5
+; SSE2-SSSE3-NEXT: pandn %xmm3, %xmm7
+; SSE2-SSSE3-NEXT: por %xmm5, %xmm7
+; SSE2-SSSE3-NEXT: pxor %xmm1, %xmm4
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm4, %xmm2
+; SSE2-SSSE3-NEXT: pand %xmm2, %xmm1
+; SSE2-SSSE3-NEXT: pandn %xmm3, %xmm2
+; SSE2-SSSE3-NEXT: por %xmm1, %xmm2
+; SSE2-SSSE3-NEXT: packuswb %xmm7, %xmm2
+; SSE2-SSSE3-NEXT: packuswb %xmm2, %xmm0
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_usat_v16i32_v16i8:
; SSE41: # %bb.0:
@@ -4057,87 +3412,46 @@ define <16 x i8> @trunc_usat_v16i32_v16i8(ptr %p0) {
}
define void @trunc_usat_v16i32_v16i8_store(ptr %p0, ptr %p1) {
-; SSE2-LABEL: trunc_usat_v16i32_v16i8_store:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa (%rdi), %xmm6
-; SSE2-NEXT: movdqa 16(%rdi), %xmm5
-; SSE2-NEXT: movdqa 32(%rdi), %xmm0
-; SSE2-NEXT: movdqa 48(%rdi), %xmm4
-; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [255,255,255,255]
-; SSE2-NEXT: movdqa {{.*#+}} xmm3 = [2147483648,2147483648,2147483648,2147483648]
-; SSE2-NEXT: movdqa %xmm5, %xmm7
-; SSE2-NEXT: pxor %xmm3, %xmm7
-; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [2147483903,2147483903,2147483903,2147483903]
-; SSE2-NEXT: movdqa %xmm1, %xmm8
-; SSE2-NEXT: pcmpgtd %xmm7, %xmm8
-; SSE2-NEXT: pand %xmm8, %xmm5
-; SSE2-NEXT: pandn %xmm2, %xmm8
-; SSE2-NEXT: por %xmm5, %xmm8
-; SSE2-NEXT: movdqa %xmm6, %xmm7
-; SSE2-NEXT: pxor %xmm3, %xmm7
-; SSE2-NEXT: movdqa %xmm1, %xmm5
-; SSE2-NEXT: pcmpgtd %xmm7, %xmm5
-; SSE2-NEXT: pand %xmm5, %xmm6
-; SSE2-NEXT: pandn %xmm2, %xmm5
-; SSE2-NEXT: por %xmm6, %xmm5
-; SSE2-NEXT: packuswb %xmm8, %xmm5
-; SSE2-NEXT: movdqa %xmm4, %xmm6
-; SSE2-NEXT: pxor %xmm3, %xmm6
-; SSE2-NEXT: movdqa %xmm1, %xmm7
-; SSE2-NEXT: pcmpgtd %xmm6, %xmm7
-; SSE2-NEXT: pand %xmm7, %xmm4
-; SSE2-NEXT: pandn %xmm2, %xmm7
-; SSE2-NEXT: por %xmm4, %xmm7
-; SSE2-NEXT: pxor %xmm0, %xmm3
-; SSE2-NEXT: pcmpgtd %xmm3, %xmm1
-; SSE2-NEXT: pand %xmm1, %xmm0
-; SSE2-NEXT: pandn %xmm2, %xmm1
-; SSE2-NEXT: por %xmm0, %xmm1
-; SSE2-NEXT: packuswb %xmm7, %xmm1
-; SSE2-NEXT: packuswb %xmm1, %xmm5
-; SSE2-NEXT: movdqa %xmm5, (%rsi)
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_usat_v16i32_v16i8_store:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa (%rdi), %xmm6
-; SSSE3-NEXT: movdqa 16(%rdi), %xmm5
-; SSSE3-NEXT: movdqa 32(%rdi), %xmm0
-; SSSE3-NEXT: movdqa 48(%rdi), %xmm4
-; SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [255,255,255,255]
-; SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [2147483648,2147483648,2147483648,2147483648]
-; SSSE3-NEXT: movdqa %xmm5, %xmm7
-; SSSE3-NEXT: pxor %xmm3, %xmm7
-; SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [2147483903,2147483903,2147483903,2147483903]
-; SSSE3-NEXT: movdqa %xmm1, %xmm8
-; SSSE3-NEXT: pcmpgtd %xmm7, %xmm8
-; SSSE3-NEXT: pand %xmm8, %xmm5
-; SSSE3-NEXT: pandn %xmm2, %xmm8
-; SSSE3-NEXT: por %xmm5, %xmm8
-; SSSE3-NEXT: movdqa %xmm6, %xmm7
-; SSSE3-NEXT: pxor %xmm3, %xmm7
-; SSSE3-NEXT: movdqa %xmm1, %xmm5
-; SSSE3-NEXT: pcmpgtd %xmm7, %xmm5
-; SSSE3-NEXT: pand %xmm5, %xmm6
-; SSSE3-NEXT: pandn %xmm2, %xmm5
-; SSSE3-NEXT: por %xmm6, %xmm5
-; SSSE3-NEXT: packuswb %xmm8, %xmm5
-; SSSE3-NEXT: movdqa %xmm4, %xmm6
-; SSSE3-NEXT: pxor %xmm3, %xmm6
-; SSSE3-NEXT: movdqa %xmm1, %xmm7
-; SSSE3-NEXT: pcmpgtd %xmm6, %xmm7
-; SSSE3-NEXT: pand %xmm7, %xmm4
-; SSSE3-NEXT: pandn %xmm2, %xmm7
-; SSSE3-NEXT: por %xmm4, %xmm7
-; SSSE3-NEXT: pxor %xmm0, %xmm3
-; SSSE3-NEXT: pcmpgtd %xmm3, %xmm1
-; SSSE3-NEXT: pand %xmm1, %xmm0
-; SSSE3-NEXT: pandn %xmm2, %xmm1
-; SSSE3-NEXT: por %xmm0, %xmm1
-; SSSE3-NEXT: packuswb %xmm7, %xmm1
-; SSSE3-NEXT: packuswb %xmm1, %xmm5
-; SSSE3-NEXT: movdqa %xmm5, (%rsi)
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc_usat_v16i32_v16i8_store:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa (%rdi), %xmm6
+; SSE2-SSSE3-NEXT: movdqa 16(%rdi), %xmm5
+; SSE2-SSSE3-NEXT: movdqa 32(%rdi), %xmm0
+; SSE2-SSSE3-NEXT: movdqa 48(%rdi), %xmm4
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [255,255,255,255]
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [2147483648,2147483648,2147483648,2147483648]
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm7
+; SSE2-SSSE3-NEXT: pxor %xmm3, %xmm7
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [2147483903,2147483903,2147483903,2147483903]
+; SSE2-SSSE3-NEXT: movdqa %xmm1, %xmm8
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm7, %xmm8
+; SSE2-SSSE3-NEXT: pand %xmm8, %xmm5
+; SSE2-SSSE3-NEXT: pandn %xmm2, %xmm8
+; SSE2-SSSE3-NEXT: por %xmm5, %xmm8
+; SSE2-SSSE3-NEXT: movdqa %xmm6, %xmm7
+; SSE2-SSSE3-NEXT: pxor %xmm3, %xmm7
+; SSE2-SSSE3-NEXT: movdqa %xmm1, %xmm5
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm7, %xmm5
+; SSE2-SSSE3-NEXT: pand %xmm5, %xmm6
+; SSE2-SSSE3-NEXT: pandn %xmm2, %xmm5
+; SSE2-SSSE3-NEXT: por %xmm6, %xmm5
+; SSE2-SSSE3-NEXT: packuswb %xmm8, %xmm5
+; SSE2-SSSE3-NEXT: movdqa %xmm4, %xmm6
+; SSE2-SSSE3-NEXT: pxor %xmm3, %xmm6
+; SSE2-SSSE3-NEXT: movdqa %xmm1, %xmm7
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm6, %xmm7
+; SSE2-SSSE3-NEXT: pand %xmm7, %xmm4
+; SSE2-SSSE3-NEXT: pandn %xmm2, %xmm7
+; SSE2-SSSE3-NEXT: por %xmm4, %xmm7
+; SSE2-SSSE3-NEXT: pxor %xmm0, %xmm3
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm3, %xmm1
+; SSE2-SSSE3-NEXT: pand %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: pandn %xmm2, %xmm1
+; SSE2-SSSE3-NEXT: por %xmm0, %xmm1
+; SSE2-SSSE3-NEXT: packuswb %xmm7, %xmm1
+; SSE2-SSSE3-NEXT: packuswb %xmm1, %xmm5
+; SSE2-SSSE3-NEXT: movdqa %xmm5, (%rsi)
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_usat_v16i32_v16i8_store:
; SSE41: # %bb.0:
@@ -4203,21 +3517,13 @@ define void @trunc_usat_v16i32_v16i8_store(ptr %p0, ptr %p1) {
}
define <8 x i8> @trunc_usat_v8i16_v8i8(<8 x i16> %a0) {
-; SSE2-LABEL: trunc_usat_v8i16_v8i8:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa %xmm0, %xmm1
-; SSE2-NEXT: psubusw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
-; SSE2-NEXT: psubw %xmm1, %xmm0
-; SSE2-NEXT: packuswb %xmm0, %xmm0
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_usat_v8i16_v8i8:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa %xmm0, %xmm1
-; SSSE3-NEXT: psubusw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
-; SSSE3-NEXT: psubw %xmm1, %xmm0
-; SSSE3-NEXT: packuswb %xmm0, %xmm0
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc_usat_v8i16_v8i8:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm1
+; SSE2-SSSE3-NEXT: psubusw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
+; SSE2-SSSE3-NEXT: psubw %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: packuswb %xmm0, %xmm0
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_usat_v8i16_v8i8:
; SSE41: # %bb.0:
@@ -4267,23 +3573,14 @@ define <8 x i8> @trunc_usat_v8i16_v8i8(<8 x i16> %a0) {
}
define void @trunc_usat_v8i16_v8i8_store(<8 x i16> %a0, ptr%p1) {
-; SSE2-LABEL: trunc_usat_v8i16_v8i8_store:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa %xmm0, %xmm1
-; SSE2-NEXT: psubusw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
-; SSE2-NEXT: psubw %xmm1, %xmm0
-; SSE2-NEXT: packuswb %xmm0, %xmm0
-; SSE2-NEXT: movq %xmm0, (%rdi)
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_usat_v8i16_v8i8_store:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa %xmm0, %xmm1
-; SSSE3-NEXT: psubusw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
-; SSSE3-NEXT: psubw %xmm1, %xmm0
-; SSSE3-NEXT: packuswb %xmm0, %xmm0
-; SSSE3-NEXT: movq %xmm0, (%rdi)
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc_usat_v8i16_v8i8_store:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm1
+; SSE2-SSSE3-NEXT: psubusw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
+; SSE2-SSSE3-NEXT: psubw %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: packuswb %xmm0, %xmm0
+; SSE2-SSSE3-NEXT: movq %xmm0, (%rdi)
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_usat_v8i16_v8i8_store:
; SSE41: # %bb.0:
@@ -4338,29 +3635,17 @@ define void @trunc_usat_v8i16_v8i8_store(<8 x i16> %a0, ptr%p1) {
}
define <16 x i8> @trunc_usat_v16i16_v16i8(<16 x i16> %a0) {
-; SSE2-LABEL: trunc_usat_v16i16_v16i8:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255]
-; SSE2-NEXT: movdqa %xmm1, %xmm3
-; SSE2-NEXT: psubusw %xmm2, %xmm3
-; SSE2-NEXT: psubw %xmm3, %xmm1
-; SSE2-NEXT: movdqa %xmm0, %xmm3
-; SSE2-NEXT: psubusw %xmm2, %xmm3
-; SSE2-NEXT: psubw %xmm3, %xmm0
-; SSE2-NEXT: packuswb %xmm1, %xmm0
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_usat_v16i16_v16i8:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255]
-; SSSE3-NEXT: movdqa %xmm1, %xmm3
-; SSSE3-NEXT: psubusw %xmm2, %xmm3
-; SSSE3-NEXT: psubw %xmm3, %xmm1
-; SSSE3-NEXT: movdqa %xmm0, %xmm3
-; SSSE3-NEXT: psubusw %xmm2, %xmm3
-; SSSE3-NEXT: psubw %xmm3, %xmm0
-; SSSE3-NEXT: packuswb %xmm1, %xmm0
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc_usat_v16i16_v16i8:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255]
+; SSE2-SSSE3-NEXT: movdqa %xmm1, %xmm3
+; SSE2-SSSE3-NEXT: psubusw %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: psubw %xmm3, %xmm1
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm3
+; SSE2-SSSE3-NEXT: psubusw %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: psubw %xmm3, %xmm0
+; SSE2-SSSE3-NEXT: packuswb %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_usat_v16i16_v16i8:
; SSE41: # %bb.0:
@@ -4430,51 +3715,28 @@ define <16 x i8> @trunc_usat_v16i16_v16i8(<16 x i16> %a0) {
}
define <32 x i8> @trunc_usat_v32i16_v32i8(ptr %p0) {
-; SSE2-LABEL: trunc_usat_v32i16_v32i8:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa (%rdi), %xmm0
-; SSE2-NEXT: movdqa 16(%rdi), %xmm2
-; SSE2-NEXT: movdqa 32(%rdi), %xmm1
-; SSE2-NEXT: movdqa 48(%rdi), %xmm3
-; SSE2-NEXT: movdqa {{.*#+}} xmm4 = [255,255,255,255,255,255,255,255]
-; SSE2-NEXT: movdqa %xmm3, %xmm5
-; SSE2-NEXT: psubusw %xmm4, %xmm5
-; SSE2-NEXT: psubw %xmm5, %xmm3
-; SSE2-NEXT: movdqa %xmm1, %xmm5
-; SSE2-NEXT: psubusw %xmm4, %xmm5
-; SSE2-NEXT: psubw %xmm5, %xmm1
-; SSE2-NEXT: packuswb %xmm3, %xmm1
-; SSE2-NEXT: movdqa %xmm2, %xmm3
-; SSE2-NEXT: psubusw %xmm4, %xmm3
-; SSE2-NEXT: psubw %xmm3, %xmm2
-; SSE2-NEXT: movdqa %xmm0, %xmm3
-; SSE2-NEXT: psubusw %xmm4, %xmm3
-; SSE2-NEXT: psubw %xmm3, %xmm0
-; SSE2-NEXT: packuswb %xmm2, %xmm0
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_usat_v32i16_v32i8:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa (%rdi), %xmm0
-; SSSE3-NEXT: movdqa 16(%rdi), %xmm2
-; SSSE3-NEXT: movdqa 32(%rdi), %xmm1
-; SSSE3-NEXT: movdqa 48(%rdi), %xmm3
-; SSSE3-NEXT: movdqa {{.*#+}} xmm4 = [255,255,255,255,255,255,255,255]
-; SSSE3-NEXT: movdqa %xmm3, %xmm5
-; SSSE3-NEXT: psubusw %xmm4, %xmm5
-; SSSE3-NEXT: psubw %xmm5, %xmm3
-; SSSE3-NEXT: movdqa %xmm1, %xmm5
-; SSSE3-NEXT: psubusw %xmm4, %xmm5
-; SSSE3-NEXT: psubw %xmm5, %xmm1
-; SSSE3-NEXT: packuswb %xmm3, %xmm1
-; SSSE3-NEXT: movdqa %xmm2, %xmm3
-; SSSE3-NEXT: psubusw %xmm4, %xmm3
-; SSSE3-NEXT: psubw %xmm3, %xmm2
-; SSSE3-NEXT: movdqa %xmm0, %xmm3
-; SSSE3-NEXT: psubusw %xmm4, %xmm3
-; SSSE3-NEXT: psubw %xmm3, %xmm0
-; SSSE3-NEXT: packuswb %xmm2, %xmm0
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc_usat_v32i16_v32i8:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa (%rdi), %xmm0
+; SSE2-SSSE3-NEXT: movdqa 16(%rdi), %xmm2
+; SSE2-SSSE3-NEXT: movdqa 32(%rdi), %xmm1
+; SSE2-SSSE3-NEXT: movdqa 48(%rdi), %xmm3
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm4 = [255,255,255,255,255,255,255,255]
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm5
+; SSE2-SSSE3-NEXT: psubusw %xmm4, %xmm5
+; SSE2-SSSE3-NEXT: psubw %xmm5, %xmm3
+; SSE2-SSSE3-NEXT: movdqa %xmm1, %xmm5
+; SSE2-SSSE3-NEXT: psubusw %xmm4, %xmm5
+; SSE2-SSSE3-NEXT: psubw %xmm5, %xmm1
+; SSE2-SSSE3-NEXT: packuswb %xmm3, %xmm1
+; SSE2-SSSE3-NEXT: movdqa %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: psubusw %xmm4, %xmm3
+; SSE2-SSSE3-NEXT: psubw %xmm3, %xmm2
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm3
+; SSE2-SSSE3-NEXT: psubusw %xmm4, %xmm3
+; SSE2-SSSE3-NEXT: psubw %xmm3, %xmm0
+; SSE2-SSSE3-NEXT: packuswb %xmm2, %xmm0
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_usat_v32i16_v32i8:
; SSE41: # %bb.0:
@@ -4554,155 +3816,80 @@ define <32 x i8> @trunc_usat_v32i16_v32i8(ptr %p0) {
}
define <32 x i8> @trunc_usat_v32i32_v32i8(ptr %p0) {
-; SSE2-LABEL: trunc_usat_v32i32_v32i8:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa (%rdi), %xmm7
-; SSE2-NEXT: movdqa 16(%rdi), %xmm0
-; SSE2-NEXT: movdqa 32(%rdi), %xmm2
-; SSE2-NEXT: movdqa 48(%rdi), %xmm5
-; SSE2-NEXT: movdqa 96(%rdi), %xmm8
-; SSE2-NEXT: movdqa 112(%rdi), %xmm9
-; SSE2-NEXT: movdqa 64(%rdi), %xmm10
-; SSE2-NEXT: movdqa 80(%rdi), %xmm1
-; SSE2-NEXT: movdqa {{.*#+}} xmm4 = [255,255,255,255]
-; SSE2-NEXT: movdqa {{.*#+}} xmm6 = [2147483648,2147483648,2147483648,2147483648]
-; SSE2-NEXT: movdqa %xmm1, %xmm11
-; SSE2-NEXT: pxor %xmm6, %xmm11
-; SSE2-NEXT: movdqa {{.*#+}} xmm3 = [2147483903,2147483903,2147483903,2147483903]
-; SSE2-NEXT: movdqa %xmm3, %xmm12
-; SSE2-NEXT: pcmpgtd %xmm11, %xmm12
-; SSE2-NEXT: pand %xmm12, %xmm1
-; SSE2-NEXT: pandn %xmm4, %xmm12
-; SSE2-NEXT: por %xmm1, %xmm12
-; SSE2-NEXT: movdqa %xmm10, %xmm11
-; SSE2-NEXT: pxor %xmm6, %xmm11
-; SSE2-NEXT: movdqa %xmm3, %xmm1
-; SSE2-NEXT: pcmpgtd %xmm11, %xmm1
-; SSE2-NEXT: pand %xmm1, %xmm10
-; SSE2-NEXT: pandn %xmm4, %xmm1
-; SSE2-NEXT: por %xmm10, %xmm1
-; SSE2-NEXT: packuswb %xmm12, %xmm1
-; SSE2-NEXT: movdqa %xmm9, %xmm10
-; SSE2-NEXT: pxor %xmm6, %xmm10
-; SSE2-NEXT: movdqa %xmm3, %xmm11
-; SSE2-NEXT: pcmpgtd %xmm10, %xmm11
-; SSE2-NEXT: pand %xmm11, %xmm9
-; SSE2-NEXT: pandn %xmm4, %xmm11
-; SSE2-NEXT: por %xmm9, %xmm11
-; SSE2-NEXT: movdqa %xmm8, %xmm9
-; SSE2-NEXT: pxor %xmm6, %xmm9
-; SSE2-NEXT: movdqa %xmm3, %xmm10
-; SSE2-NEXT: pcmpgtd %xmm9, %xmm10
-; SSE2-NEXT: pand %xmm10, %xmm8
-; SSE2-NEXT: pandn %xmm4, %xmm10
-; SSE2-NEXT: por %xmm8, %xmm10
-; SSE2-NEXT: packuswb %xmm11, %xmm10
-; SSE2-NEXT: packuswb %xmm10, %xmm1
-; SSE2-NEXT: movdqa %xmm0, %xmm8
-; SSE2-NEXT: pxor %xmm6, %xmm8
-; SSE2-NEXT: movdqa %xmm3, %xmm9
-; SSE2-NEXT: pcmpgtd %xmm8, %xmm9
-; SSE2-NEXT: pand %xmm9, %xmm0
-; SSE2-NEXT: pandn %xmm4, %xmm9
-; SSE2-NEXT: por %xmm0, %xmm9
-; SSE2-NEXT: movdqa %xmm7, %xmm8
-; SSE2-NEXT: pxor %xmm6, %xmm8
-; SSE2-NEXT: movdqa %xmm3, %xmm0
-; SSE2-NEXT: pcmpgtd %xmm8, %xmm0
-; SSE2-NEXT: pand %xmm0, %xmm7
-; SSE2-NEXT: pandn %xmm4, %xmm0
-; SSE2-NEXT: por %xmm7, %xmm0
-; SSE2-NEXT: packuswb %xmm9, %xmm0
-; SSE2-NEXT: movdqa %xmm5, %xmm7
-; SSE2-NEXT: pxor %xmm6, %xmm7
-; SSE2-NEXT: movdqa %xmm3, %xmm8
-; SSE2-NEXT: pcmpgtd %xmm7, %xmm8
-; SSE2-NEXT: pand %xmm8, %xmm5
-; SSE2-NEXT: pandn %xmm4, %xmm8
-; SSE2-NEXT: por %xmm5, %xmm8
-; SSE2-NEXT: pxor %xmm2, %xmm6
-; SSE2-NEXT: pcmpgtd %xmm6, %xmm3
-; SSE2-NEXT: pand %xmm3, %xmm2
-; SSE2-NEXT: pandn %xmm4, %xmm3
-; SSE2-NEXT: por %xmm2, %xmm3
-; SSE2-NEXT: packuswb %xmm8, %xmm3
-; SSE2-NEXT: packuswb %xmm3, %xmm0
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc_usat_v32i32_v32i8:
-; SSSE3: # %bb.0:
-; SSSE3-NEXT: movdqa (%rdi), %xmm7
-; SSSE3-NEXT: movdqa 16(%rdi), %xmm0
-; SSSE3-NEXT: movdqa 32(%rdi), %xmm2
-; SSSE3-NEXT: movdqa 48(%rdi), %xmm5
-; SSSE3-NEXT: movdqa 96(%rdi), %xmm8
-; SSSE3-NEXT: movdqa 112(%rdi), %xmm9
-; SSSE3-NEXT: movdqa 64(%rdi), %xmm10
-; SSSE3-NEXT: movdqa 80(%rdi), %xmm1
-; SSSE3-NEXT: movdqa {{.*#+}} xmm4 = [255,255,255,255]
-; SSSE3-NEXT: movdqa {{.*#+}} xmm6 = [2147483648,2147483648,2147483648,2147483648]
-; SSSE3-NEXT: movdqa %xmm1, %xmm11
-; SSSE3-NEXT: pxor %xmm6, %xmm11
-; SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [2147483903,2147483903,2147483903,2147483903]
-; SSSE3-NEXT: movdqa %xmm3, %xmm12
-; SSSE3-NEXT: pcmpgtd %xmm11, %xmm12
-; SSSE3-NEXT: pand %xmm12, %xmm1
-; SSSE3-NEXT: pandn %xmm4, %xmm12
-; SSSE3-NEXT: por %xmm1, %xmm12
-; SSSE3-NEXT: movdqa %xmm10, %xmm11
-; SSSE3-NEXT: pxor %xmm6, %xmm11
-; SSSE3-NEXT: movdqa %xmm3, %xmm1
-; SSSE3-NEXT: pcmpgtd %xmm11, %xmm1
-; SSSE3-NEXT: pand %xmm1, %xmm10
-; SSSE3-NEXT: pandn %xmm4, %xmm1
-; SSSE3-NEXT: por %xmm10, %xmm1
-; SSSE3-NEXT: packuswb %xmm12, %xmm1
-; SSSE3-NEXT: movdqa %xmm9, %xmm10
-; SSSE3-NEXT: pxor %xmm6, %xmm10
-; SSSE3-NEXT: movdqa %xmm3, %xmm11
-; SSSE3-NEXT: pcmpgtd %xmm10, %xmm11
-; SSSE3-NEXT: pand %xmm11, %xmm9
-; SSSE3-NEXT: pandn %xmm4, %xmm11
-; SSSE3-NEXT: por %xmm9, %xmm11
-; SSSE3-NEXT: movdqa %xmm8, %xmm9
-; SSSE3-NEXT: pxor %xmm6, %xmm9
-; SSSE3-NEXT: movdqa %xmm3, %xmm10
-; SSSE3-NEXT: pcmpgtd %xmm9, %xmm10
-; SSSE3-NEXT: pand %xmm10, %xmm8
-; SSSE3-NEXT: pandn %xmm4, %xmm10
-; SSSE3-NEXT: por %xmm8, %xmm10
-; SSSE3-NEXT: packuswb %xmm11, %xmm10
-; SSSE3-NEXT: packuswb %xmm10, %xmm1
-; SSSE3-NEXT: movdqa %xmm0, %xmm8
-; SSSE3-NEXT: pxor %xmm6, %xmm8
-; SSSE3-NEXT: movdqa %xmm3, %xmm9
-; SSSE3-NEXT: pcmpgtd %xmm8, %xmm9
-; SSSE3-NEXT: pand %xmm9, %xmm0
-; SSSE3-NEXT: pandn %xmm4, %xmm9
-; SSSE3-NEXT: por %xmm0, %xmm9
-; SSSE3-NEXT: movdqa %xmm7, %xmm8
-; SSSE3-NEXT: pxor %xmm6, %xmm8
-; SSSE3-NEXT: movdqa %xmm3, %xmm0
-; SSSE3-NEXT: pcmpgtd %xmm8, %xmm0
-; SSSE3-NEXT: pand %xmm0, %xmm7
-; SSSE3-NEXT: pandn %xmm4, %xmm0
-; SSSE3-NEXT: por %xmm7, %xmm0
-; SSSE3-NEXT: packuswb %xmm9, %xmm0
-; SSSE3-NEXT: movdqa %xmm5, %xmm7
-; SSSE3-NEXT: pxor %xmm6, %xmm7
-; SSSE3-NEXT: movdqa %xmm3, %xmm8
-; SSSE3-NEXT: pcmpgtd %xmm7, %xmm8
-; SSSE3-NEXT: pand %xmm8, %xmm5
-; SSSE3-NEXT: pandn %xmm4, %xmm8
-; SSSE3-NEXT: por %xmm5, %xmm8
-; SSSE3-NEXT: pxor %xmm2, %xmm6
-; SSSE3-NEXT: pcmpgtd %xmm6, %xmm3
-; SSSE3-NEXT: pand %xmm3, %xmm2
-; SSSE3-NEXT: pandn %xmm4, %xmm3
-; SSSE3-NEXT: por %xmm2, %xmm3
-; SSSE3-NEXT: packuswb %xmm8, %xmm3
-; SSSE3-NEXT: packuswb %xmm3, %xmm0
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc_usat_v32i32_v32i8:
+; SSE2-SSSE3: # %bb.0:
+; SSE2-SSSE3-NEXT: movdqa (%rdi), %xmm7
+; SSE2-SSSE3-NEXT: movdqa 16(%rdi), %xmm0
+; SSE2-SSSE3-NEXT: movdqa 32(%rdi), %xmm2
+; SSE2-SSSE3-NEXT: movdqa 48(%rdi), %xmm5
+; SSE2-SSSE3-NEXT: movdqa 96(%rdi), %xmm8
+; SSE2-SSSE3-NEXT: movdqa 112(%rdi), %xmm9
+; SSE2-SSSE3-NEXT: movdqa 64(%rdi), %xmm10
+; SSE2-SSSE3-NEXT: movdqa 80(%rdi), %xmm1
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm4 = [255,255,255,255]
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm6 = [2147483648,2147483648,2147483648,2147483648]
+; SSE2-SSSE3-NEXT: movdqa %xmm1, %xmm11
+; SSE2-SSSE3-NEXT: pxor %xmm6, %xmm11
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [2147483903,2147483903,2147483903,2147483903]
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm12
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm11, %xmm12
+; SSE2-SSSE3-NEXT: pand %xmm12, %xmm1
+; SSE2-SSSE3-NEXT: pandn %xmm4, %xmm12
+; SSE2-SSSE3-NEXT: por %xmm1, %xmm12
+; SSE2-SSSE3-NEXT: movdqa %xmm10, %xmm11
+; SSE2-SSSE3-NEXT: pxor %xmm6, %xmm11
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm1
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm11, %xmm1
+; SSE2-SSSE3-NEXT: pand %xmm1, %xmm10
+; SSE2-SSSE3-NEXT: pandn %xmm4, %xmm1
+; SSE2-SSSE3-NEXT: por %xmm10, %xmm1
+; SSE2-SSSE3-NEXT: packuswb %xmm12, %xmm1
+; SSE2-SSSE3-NEXT: movdqa %xmm9, %xmm10
+; SSE2-SSSE3-NEXT: pxor %xmm6, %xmm10
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm11
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm10, %xmm11
+; SSE2-SSSE3-NEXT: pand %xmm11, %xmm9
+; SSE2-SSSE3-NEXT: pandn %xmm4, %xmm11
+; SSE2-SSSE3-NEXT: por %xmm9, %xmm11
+; SSE2-SSSE3-NEXT: movdqa %xmm8, %xmm9
+; SSE2-SSSE3-NEXT: pxor %xmm6, %xmm9
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm10
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm9, %xmm10
+; SSE2-SSSE3-NEXT: pand %xmm10, %xmm8
+; SSE2-SSSE3-NEXT: pandn %xmm4, %xmm10
+; SSE2-SSSE3-NEXT: por %xmm8, %xmm10
+; SSE2-SSSE3-NEXT: packuswb %xmm11, %xmm10
+; SSE2-SSSE3-NEXT: packuswb %xmm10, %xmm1
+; SSE2-SSSE3-NEXT: movdqa %xmm0, %xmm8
+; SSE2-SSSE3-NEXT: pxor %xmm6, %xmm8
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm9
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm8, %xmm9
+; SSE2-SSSE3-NEXT: pand %xmm9, %xmm0
+; SSE2-SSSE3-NEXT: pandn %xmm4, %xmm9
+; SSE2-SSSE3-NEXT: por %xmm0, %xmm9
+; SSE2-SSSE3-NEXT: movdqa %xmm7, %xmm8
+; SSE2-SSSE3-NEXT: pxor %xmm6, %xmm8
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm0
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm8, %xmm0
+; SSE2-SSSE3-NEXT: pand %xmm0, %xmm7
+; SSE2-SSSE3-NEXT: pandn %xmm4, %xmm0
+; SSE2-SSSE3-NEXT: por %xmm7, %xmm0
+; SSE2-SSSE3-NEXT: packuswb %xmm9, %xmm0
+; SSE2-SSSE3-NEXT: movdqa %xmm5, %xmm7
+; SSE2-SSSE3-NEXT: pxor %xmm6, %xmm7
+; SSE2-SSSE3-NEXT: movdqa %xmm3, %xmm8
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm7, %xmm8
+; SSE2-SSSE3-NEXT: pand %xmm8, %xmm5
+; SSE2-SSSE3-NEXT: pandn %xmm4, %xmm8
+; SSE2-SSSE3-NEXT: por %xmm5, %xmm8
+; SSE2-SSSE3-NEXT: pxor %xmm2, %xmm6
+; SSE2-SSSE3-NEXT: pcmpgtd %xmm6, %xmm3
+; SSE2-SSSE3-NEXT: pand %xmm3, %xmm2
+; SSE2-SSSE3-NEXT: pandn %xmm4, %xmm3
+; SSE2-SSSE3-NEXT: por %xmm2, %xmm3
+; SSE2-SSSE3-NEXT: packuswb %xmm8, %xmm3
+; SSE2-SSSE3-NEXT: packuswb %xmm3, %xmm0
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc_usat_v32i32_v32i8:
; SSE41: # %bb.0:
diff --git a/llvm/test/CodeGen/X86/vector-trunc.ll b/llvm/test/CodeGen/X86/vector-trunc.ll
index 1688ce6611c8f..129e535b4a841 100644
--- a/llvm/test/CodeGen/X86/vector-trunc.ll
+++ b/llvm/test/CodeGen/X86/vector-trunc.ll
@@ -1,8 +1,8 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefix=SSE --check-prefix=SSE2
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+ssse3 | FileCheck %s --check-prefix=SSE --check-prefix=SSSE3
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefix=SSE --check-prefix=SSE41
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=AVX --check-prefix=AVX1
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefixes=SSE,SSE2-SSSE3,SSE2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+ssse3 | FileCheck %s --check-prefixes=SSE,SSE2-SSSE3,SSSE3
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefixes=SSE,SSE41
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX1
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-SLOW
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-FAST-ALL
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-FAST-PERLANE
@@ -157,35 +157,20 @@ entry:
}
define <8 x i16> @trunc8i64_8i16(<8 x i64> %a) {
-; SSE2-LABEL: trunc8i64_8i16:
-; SSE2: # %bb.0: # %entry
-; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; SSE2-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSE2-NEXT: pshuflw {{.*#+}} xmm4 = xmm0[0,2,2,3,4,5,6,7]
-; SSE2-NEXT: punpckldq {{.*#+}} xmm4 = xmm4[0],xmm1[0],xmm4[1],xmm1[1]
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm3[0,2,2,3]
-; SSE2-NEXT: pshuflw {{.*#+}} xmm1 = xmm0[0,1,0,2,4,5,6,7]
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[0,2,2,3]
-; SSE2-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,1,0,2,4,5,6,7]
-; SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; SSE2-NEXT: movsd {{.*#+}} xmm0 = xmm4[0],xmm0[1]
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc8i64_8i16:
-; SSSE3: # %bb.0: # %entry
-; SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; SSSE3-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSSE3-NEXT: pshuflw {{.*#+}} xmm4 = xmm0[0,2,2,3,4,5,6,7]
-; SSSE3-NEXT: punpckldq {{.*#+}} xmm4 = xmm4[0],xmm1[0],xmm4[1],xmm1[1]
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm3[0,2,2,3]
-; SSSE3-NEXT: pshuflw {{.*#+}} xmm1 = xmm0[0,1,0,2,4,5,6,7]
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm2[0,2,2,3]
-; SSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,1,0,2,4,5,6,7]
-; SSSE3-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; SSSE3-NEXT: movsd {{.*#+}} xmm0 = xmm4[0],xmm0[1]
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc8i64_8i16:
+; SSE2-SSSE3: # %bb.0: # %entry
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
+; SSE2-SSSE3-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; SSE2-SSSE3-NEXT: pshuflw {{.*#+}} xmm4 = xmm0[0,2,2,3,4,5,6,7]
+; SSE2-SSSE3-NEXT: punpckldq {{.*#+}} xmm4 = xmm4[0],xmm1[0],xmm4[1],xmm1[1]
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm3[0,2,2,3]
+; SSE2-SSSE3-NEXT: pshuflw {{.*#+}} xmm1 = xmm0[0,1,0,2,4,5,6,7]
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm2[0,2,2,3]
+; SSE2-SSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,1,0,2,4,5,6,7]
+; SSE2-SSSE3-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; SSE2-SSSE3-NEXT: movsd {{.*#+}} xmm0 = xmm4[0],xmm0[1]
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc8i64_8i16:
; SSE41: # %bb.0: # %entry
@@ -235,33 +220,19 @@ entry:
}
define void @trunc8i64_8i8(<8 x i64> %a) {
-; SSE2-LABEL: trunc8i64_8i8:
-; SSE2: # %bb.0: # %entry
-; SSE2-NEXT: movdqa {{.*#+}} xmm4 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0]
-; SSE2-NEXT: pand %xmm4, %xmm3
-; SSE2-NEXT: pand %xmm4, %xmm2
-; SSE2-NEXT: packuswb %xmm3, %xmm2
-; SSE2-NEXT: pand %xmm4, %xmm1
-; SSE2-NEXT: pand %xmm4, %xmm0
-; SSE2-NEXT: packuswb %xmm1, %xmm0
-; SSE2-NEXT: packuswb %xmm2, %xmm0
-; SSE2-NEXT: packuswb %xmm0, %xmm0
-; SSE2-NEXT: movq %xmm0, (%rax)
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc8i64_8i8:
-; SSSE3: # %bb.0: # %entry
-; SSSE3-NEXT: movdqa {{.*#+}} xmm4 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0]
-; SSSE3-NEXT: pand %xmm4, %xmm3
-; SSSE3-NEXT: pand %xmm4, %xmm2
-; SSSE3-NEXT: packuswb %xmm3, %xmm2
-; SSSE3-NEXT: pand %xmm4, %xmm1
-; SSSE3-NEXT: pand %xmm4, %xmm0
-; SSSE3-NEXT: packuswb %xmm1, %xmm0
-; SSSE3-NEXT: packuswb %xmm2, %xmm0
-; SSSE3-NEXT: packuswb %xmm0, %xmm0
-; SSSE3-NEXT: movq %xmm0, (%rax)
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc8i64_8i8:
+; SSE2-SSSE3: # %bb.0: # %entry
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm4 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0]
+; SSE2-SSSE3-NEXT: pand %xmm4, %xmm3
+; SSE2-SSSE3-NEXT: pand %xmm4, %xmm2
+; SSE2-SSSE3-NEXT: packuswb %xmm3, %xmm2
+; SSE2-SSSE3-NEXT: pand %xmm4, %xmm1
+; SSE2-SSSE3-NEXT: pand %xmm4, %xmm0
+; SSE2-SSSE3-NEXT: packuswb %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: packuswb %xmm2, %xmm0
+; SSE2-SSSE3-NEXT: packuswb %xmm0, %xmm0
+; SSE2-SSSE3-NEXT: movq %xmm0, (%rax)
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc8i64_8i8:
; SSE41: # %bb.0: # %entry
@@ -392,19 +363,12 @@ entry:
}
define <8 x i16> @trunc8i32_8i16_ashr(<8 x i32> %a) {
-; SSE2-LABEL: trunc8i32_8i16_ashr:
-; SSE2: # %bb.0: # %entry
-; SSE2-NEXT: psrad $16, %xmm1
-; SSE2-NEXT: psrad $16, %xmm0
-; SSE2-NEXT: packssdw %xmm1, %xmm0
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc8i32_8i16_ashr:
-; SSSE3: # %bb.0: # %entry
-; SSSE3-NEXT: psrad $16, %xmm1
-; SSSE3-NEXT: psrad $16, %xmm0
-; SSSE3-NEXT: packssdw %xmm1, %xmm0
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc8i32_8i16_ashr:
+; SSE2-SSSE3: # %bb.0: # %entry
+; SSE2-SSSE3-NEXT: psrad $16, %xmm1
+; SSE2-SSSE3-NEXT: psrad $16, %xmm0
+; SSE2-SSSE3-NEXT: packssdw %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc8i32_8i16_ashr:
; SSE41: # %bb.0: # %entry
@@ -466,19 +430,12 @@ entry:
}
define <8 x i16> @trunc8i32_8i16_lshr(<8 x i32> %a) {
-; SSE2-LABEL: trunc8i32_8i16_lshr:
-; SSE2: # %bb.0: # %entry
-; SSE2-NEXT: psrad $16, %xmm1
-; SSE2-NEXT: psrad $16, %xmm0
-; SSE2-NEXT: packssdw %xmm1, %xmm0
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc8i32_8i16_lshr:
-; SSSE3: # %bb.0: # %entry
-; SSSE3-NEXT: psrad $16, %xmm1
-; SSSE3-NEXT: psrad $16, %xmm0
-; SSSE3-NEXT: packssdw %xmm1, %xmm0
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc8i32_8i16_lshr:
+; SSE2-SSSE3: # %bb.0: # %entry
+; SSE2-SSSE3-NEXT: psrad $16, %xmm1
+; SSE2-SSSE3-NEXT: psrad $16, %xmm0
+; SSE2-SSSE3-NEXT: packssdw %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc8i32_8i16_lshr:
; SSE41: # %bb.0: # %entry
@@ -624,37 +581,21 @@ entry:
}
define void @trunc16i32_16i16(<16 x i32> %a) {
-; SSE2-LABEL: trunc16i32_16i16:
-; SSE2: # %bb.0: # %entry
-; SSE2-NEXT: pslld $16, %xmm1
-; SSE2-NEXT: psrad $16, %xmm1
-; SSE2-NEXT: pslld $16, %xmm0
-; SSE2-NEXT: psrad $16, %xmm0
-; SSE2-NEXT: packssdw %xmm1, %xmm0
-; SSE2-NEXT: pslld $16, %xmm3
-; SSE2-NEXT: psrad $16, %xmm3
-; SSE2-NEXT: pslld $16, %xmm2
-; SSE2-NEXT: psrad $16, %xmm2
-; SSE2-NEXT: packssdw %xmm3, %xmm2
-; SSE2-NEXT: movdqu %xmm2, (%rax)
-; SSE2-NEXT: movdqu %xmm0, (%rax)
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc16i32_16i16:
-; SSSE3: # %bb.0: # %entry
-; SSSE3-NEXT: pslld $16, %xmm1
-; SSSE3-NEXT: psrad $16, %xmm1
-; SSSE3-NEXT: pslld $16, %xmm0
-; SSSE3-NEXT: psrad $16, %xmm0
-; SSSE3-NEXT: packssdw %xmm1, %xmm0
-; SSSE3-NEXT: pslld $16, %xmm3
-; SSSE3-NEXT: psrad $16, %xmm3
-; SSSE3-NEXT: pslld $16, %xmm2
-; SSSE3-NEXT: psrad $16, %xmm2
-; SSSE3-NEXT: packssdw %xmm3, %xmm2
-; SSSE3-NEXT: movdqu %xmm2, (%rax)
-; SSSE3-NEXT: movdqu %xmm0, (%rax)
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc16i32_16i16:
+; SSE2-SSSE3: # %bb.0: # %entry
+; SSE2-SSSE3-NEXT: pslld $16, %xmm1
+; SSE2-SSSE3-NEXT: psrad $16, %xmm1
+; SSE2-SSSE3-NEXT: pslld $16, %xmm0
+; SSE2-SSSE3-NEXT: psrad $16, %xmm0
+; SSE2-SSSE3-NEXT: packssdw %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: pslld $16, %xmm3
+; SSE2-SSSE3-NEXT: psrad $16, %xmm3
+; SSE2-SSSE3-NEXT: pslld $16, %xmm2
+; SSE2-SSSE3-NEXT: psrad $16, %xmm2
+; SSE2-SSSE3-NEXT: packssdw %xmm3, %xmm2
+; SSE2-SSSE3-NEXT: movdqu %xmm2, (%rax)
+; SSE2-SSSE3-NEXT: movdqu %xmm0, (%rax)
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc16i32_16i16:
; SSE41: # %bb.0: # %entry
@@ -706,29 +647,17 @@ entry:
}
define void @trunc16i32_16i16_ashr(<16 x i32> %a) {
-; SSE2-LABEL: trunc16i32_16i16_ashr:
-; SSE2: # %bb.0: # %entry
-; SSE2-NEXT: psrad $16, %xmm1
-; SSE2-NEXT: psrad $16, %xmm0
-; SSE2-NEXT: packssdw %xmm1, %xmm0
-; SSE2-NEXT: psrad $16, %xmm3
-; SSE2-NEXT: psrad $16, %xmm2
-; SSE2-NEXT: packssdw %xmm3, %xmm2
-; SSE2-NEXT: movdqu %xmm2, (%rax)
-; SSE2-NEXT: movdqu %xmm0, (%rax)
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc16i32_16i16_ashr:
-; SSSE3: # %bb.0: # %entry
-; SSSE3-NEXT: psrad $16, %xmm1
-; SSSE3-NEXT: psrad $16, %xmm0
-; SSSE3-NEXT: packssdw %xmm1, %xmm0
-; SSSE3-NEXT: psrad $16, %xmm3
-; SSSE3-NEXT: psrad $16, %xmm2
-; SSSE3-NEXT: packssdw %xmm3, %xmm2
-; SSSE3-NEXT: movdqu %xmm2, (%rax)
-; SSSE3-NEXT: movdqu %xmm0, (%rax)
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc16i32_16i16_ashr:
+; SSE2-SSSE3: # %bb.0: # %entry
+; SSE2-SSSE3-NEXT: psrad $16, %xmm1
+; SSE2-SSSE3-NEXT: psrad $16, %xmm0
+; SSE2-SSSE3-NEXT: packssdw %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: psrad $16, %xmm3
+; SSE2-SSSE3-NEXT: psrad $16, %xmm2
+; SSE2-SSSE3-NEXT: packssdw %xmm3, %xmm2
+; SSE2-SSSE3-NEXT: movdqu %xmm2, (%rax)
+; SSE2-SSSE3-NEXT: movdqu %xmm0, (%rax)
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc16i32_16i16_ashr:
; SSE41: # %bb.0: # %entry
@@ -781,29 +710,17 @@ entry:
}
define void @trunc16i32_16i16_lshr(<16 x i32> %a) {
-; SSE2-LABEL: trunc16i32_16i16_lshr:
-; SSE2: # %bb.0: # %entry
-; SSE2-NEXT: psrad $16, %xmm1
-; SSE2-NEXT: psrad $16, %xmm0
-; SSE2-NEXT: packssdw %xmm1, %xmm0
-; SSE2-NEXT: psrad $16, %xmm3
-; SSE2-NEXT: psrad $16, %xmm2
-; SSE2-NEXT: packssdw %xmm3, %xmm2
-; SSE2-NEXT: movdqu %xmm2, (%rax)
-; SSE2-NEXT: movdqu %xmm0, (%rax)
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc16i32_16i16_lshr:
-; SSSE3: # %bb.0: # %entry
-; SSSE3-NEXT: psrad $16, %xmm1
-; SSSE3-NEXT: psrad $16, %xmm0
-; SSSE3-NEXT: packssdw %xmm1, %xmm0
-; SSSE3-NEXT: psrad $16, %xmm3
-; SSSE3-NEXT: psrad $16, %xmm2
-; SSSE3-NEXT: packssdw %xmm3, %xmm2
-; SSSE3-NEXT: movdqu %xmm2, (%rax)
-; SSSE3-NEXT: movdqu %xmm0, (%rax)
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc16i32_16i16_lshr:
+; SSE2-SSSE3: # %bb.0: # %entry
+; SSE2-SSSE3-NEXT: psrad $16, %xmm1
+; SSE2-SSSE3-NEXT: psrad $16, %xmm0
+; SSE2-SSSE3-NEXT: packssdw %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: psrad $16, %xmm3
+; SSE2-SSSE3-NEXT: psrad $16, %xmm2
+; SSE2-SSSE3-NEXT: packssdw %xmm3, %xmm2
+; SSE2-SSSE3-NEXT: movdqu %xmm2, (%rax)
+; SSE2-SSSE3-NEXT: movdqu %xmm0, (%rax)
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc16i32_16i16_lshr:
; SSE41: # %bb.0: # %entry
@@ -856,31 +773,18 @@ entry:
}
define void @trunc16i32_16i8(<16 x i32> %a) {
-; SSE2-LABEL: trunc16i32_16i8:
-; SSE2: # %bb.0: # %entry
-; SSE2-NEXT: movdqa {{.*#+}} xmm4 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
-; SSE2-NEXT: pand %xmm4, %xmm3
-; SSE2-NEXT: pand %xmm4, %xmm2
-; SSE2-NEXT: packuswb %xmm3, %xmm2
-; SSE2-NEXT: pand %xmm4, %xmm1
-; SSE2-NEXT: pand %xmm4, %xmm0
-; SSE2-NEXT: packuswb %xmm1, %xmm0
-; SSE2-NEXT: packuswb %xmm2, %xmm0
-; SSE2-NEXT: movdqu %xmm0, (%rax)
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc16i32_16i8:
-; SSSE3: # %bb.0: # %entry
-; SSSE3-NEXT: movdqa {{.*#+}} xmm4 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
-; SSSE3-NEXT: pand %xmm4, %xmm3
-; SSSE3-NEXT: pand %xmm4, %xmm2
-; SSSE3-NEXT: packuswb %xmm3, %xmm2
-; SSSE3-NEXT: pand %xmm4, %xmm1
-; SSSE3-NEXT: pand %xmm4, %xmm0
-; SSSE3-NEXT: packuswb %xmm1, %xmm0
-; SSSE3-NEXT: packuswb %xmm2, %xmm0
-; SSSE3-NEXT: movdqu %xmm0, (%rax)
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc16i32_16i8:
+; SSE2-SSSE3: # %bb.0: # %entry
+; SSE2-SSSE3-NEXT: movdqa {{.*#+}} xmm4 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
+; SSE2-SSSE3-NEXT: pand %xmm4, %xmm3
+; SSE2-SSSE3-NEXT: pand %xmm4, %xmm2
+; SSE2-SSSE3-NEXT: packuswb %xmm3, %xmm2
+; SSE2-SSSE3-NEXT: pand %xmm4, %xmm1
+; SSE2-SSSE3-NEXT: pand %xmm4, %xmm0
+; SSE2-SSSE3-NEXT: packuswb %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: packuswb %xmm2, %xmm0
+; SSE2-SSSE3-NEXT: movdqu %xmm0, (%rax)
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc16i32_16i8:
; SSE41: # %bb.0: # %entry
@@ -934,29 +838,17 @@ entry:
}
define void @trunc16i32_16i8_ashr(<16 x i32> %a) {
-; SSE2-LABEL: trunc16i32_16i8_ashr:
-; SSE2: # %bb.0: # %entry
-; SSE2-NEXT: psrld $24, %xmm1
-; SSE2-NEXT: psrld $24, %xmm0
-; SSE2-NEXT: packuswb %xmm1, %xmm0
-; SSE2-NEXT: psrld $24, %xmm3
-; SSE2-NEXT: psrld $24, %xmm2
-; SSE2-NEXT: packuswb %xmm3, %xmm2
-; SSE2-NEXT: packuswb %xmm2, %xmm0
-; SSE2-NEXT: movdqu %xmm0, (%rax)
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc16i32_16i8_ashr:
-; SSSE3: # %bb.0: # %entry
-; SSSE3-NEXT: psrld $24, %xmm1
-; SSSE3-NEXT: psrld $24, %xmm0
-; SSSE3-NEXT: packuswb %xmm1, %xmm0
-; SSSE3-NEXT: psrld $24, %xmm3
-; SSSE3-NEXT: psrld $24, %xmm2
-; SSSE3-NEXT: packuswb %xmm3, %xmm2
-; SSSE3-NEXT: packuswb %xmm2, %xmm0
-; SSSE3-NEXT: movdqu %xmm0, (%rax)
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc16i32_16i8_ashr:
+; SSE2-SSSE3: # %bb.0: # %entry
+; SSE2-SSSE3-NEXT: psrld $24, %xmm1
+; SSE2-SSSE3-NEXT: psrld $24, %xmm0
+; SSE2-SSSE3-NEXT: packuswb %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: psrld $24, %xmm3
+; SSE2-SSSE3-NEXT: psrld $24, %xmm2
+; SSE2-SSSE3-NEXT: packuswb %xmm3, %xmm2
+; SSE2-SSSE3-NEXT: packuswb %xmm2, %xmm0
+; SSE2-SSSE3-NEXT: movdqu %xmm0, (%rax)
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc16i32_16i8_ashr:
; SSE41: # %bb.0: # %entry
@@ -1011,29 +903,17 @@ entry:
}
define void @trunc16i32_16i8_lshr(<16 x i32> %a) {
-; SSE2-LABEL: trunc16i32_16i8_lshr:
-; SSE2: # %bb.0: # %entry
-; SSE2-NEXT: psrld $24, %xmm1
-; SSE2-NEXT: psrld $24, %xmm0
-; SSE2-NEXT: packuswb %xmm1, %xmm0
-; SSE2-NEXT: psrld $24, %xmm3
-; SSE2-NEXT: psrld $24, %xmm2
-; SSE2-NEXT: packuswb %xmm3, %xmm2
-; SSE2-NEXT: packuswb %xmm2, %xmm0
-; SSE2-NEXT: movdqu %xmm0, (%rax)
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc16i32_16i8_lshr:
-; SSSE3: # %bb.0: # %entry
-; SSSE3-NEXT: psrld $24, %xmm1
-; SSSE3-NEXT: psrld $24, %xmm0
-; SSSE3-NEXT: packuswb %xmm1, %xmm0
-; SSSE3-NEXT: psrld $24, %xmm3
-; SSSE3-NEXT: psrld $24, %xmm2
-; SSSE3-NEXT: packuswb %xmm3, %xmm2
-; SSSE3-NEXT: packuswb %xmm2, %xmm0
-; SSSE3-NEXT: movdqu %xmm0, (%rax)
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc16i32_16i8_lshr:
+; SSE2-SSSE3: # %bb.0: # %entry
+; SSE2-SSSE3-NEXT: psrld $24, %xmm1
+; SSE2-SSSE3-NEXT: psrld $24, %xmm0
+; SSE2-SSSE3-NEXT: packuswb %xmm1, %xmm0
+; SSE2-SSSE3-NEXT: psrld $24, %xmm3
+; SSE2-SSSE3-NEXT: psrld $24, %xmm2
+; SSE2-SSSE3-NEXT: packuswb %xmm3, %xmm2
+; SSE2-SSSE3-NEXT: packuswb %xmm2, %xmm0
+; SSE2-SSSE3-NEXT: movdqu %xmm0, (%rax)
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc16i32_16i8_lshr:
; SSE41: # %bb.0: # %entry
@@ -1406,35 +1286,20 @@ entry:
}
define <8 x i16> @trunc2x4i64_8i16(<4 x i64> %a, <4 x i64> %b) {
-; SSE2-LABEL: trunc2x4i64_8i16:
-; SSE2: # %bb.0: # %entry
-; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; SSE2-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSE2-NEXT: pshuflw {{.*#+}} xmm4 = xmm0[0,2,2,3,4,5,6,7]
-; SSE2-NEXT: punpckldq {{.*#+}} xmm4 = xmm4[0],xmm1[0],xmm4[1],xmm1[1]
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm3[0,2,2,3]
-; SSE2-NEXT: pshuflw {{.*#+}} xmm1 = xmm0[0,1,0,2,4,5,6,7]
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[0,2,2,3]
-; SSE2-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,1,0,2,4,5,6,7]
-; SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; SSE2-NEXT: movsd {{.*#+}} xmm0 = xmm4[0],xmm0[1]
-; SSE2-NEXT: retq
-;
-; SSSE3-LABEL: trunc2x4i64_8i16:
-; SSSE3: # %bb.0: # %entry
-; SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; SSSE3-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSSE3-NEXT: pshuflw {{.*#+}} xmm4 = xmm0[0,2,2,3,4,5,6,7]
-; SSSE3-NEXT: punpckldq {{.*#+}} xmm4 = xmm4[0],xmm1[0],xmm4[1],xmm1[1]
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm3[0,2,2,3]
-; SSSE3-NEXT: pshuflw {{.*#+}} xmm1 = xmm0[0,1,0,2,4,5,6,7]
-; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm2[0,2,2,3]
-; SSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,1,0,2,4,5,6,7]
-; SSSE3-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; SSSE3-NEXT: movsd {{.*#+}} xmm0 = xmm4[0],xmm0[1]
-; SSSE3-NEXT: retq
+; SSE2-SSSE3-LABEL: trunc2x4i64_8i16:
+; SSE2-SSSE3: # %bb.0: # %entry
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
+; SSE2-SSSE3-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; SSE2-SSSE3-NEXT: pshuflw {{.*#+}} xmm4 = xmm0[0,2,2,3,4,5,6,7]
+; SSE2-SSSE3-NEXT: punpckldq {{.*#+}} xmm4 = xmm4[0],xmm1[0],xmm4[1],xmm1[1]
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm3[0,2,2,3]
+; SSE2-SSSE3-NEXT: pshuflw {{.*#+}} xmm1 = xmm0[0,1,0,2,4,5,6,7]
+; SSE2-SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm2[0,2,2,3]
+; SSE2-SSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,1,0,2,4,5,6,7]
+; SSE2-SSSE3-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; SSE2-SSSE3-NEXT: movsd {{.*#+}} xmm0 = xmm4[0],xmm0[1]
+; SSE2-SSSE3-NEXT: retq
;
; SSE41-LABEL: trunc2x4i64_8i16:
; SSE41: # %bb.0: # %entry
More information about the llvm-commits
mailing list