[llvm] r301137 - [X86][SSE] Add scheduling latency/throughput tests for (most) SSE41 instructions
Simon Pilgrim via llvm-commits
llvm-commits at lists.llvm.org
Sun Apr 23 13:05:21 PDT 2017
Author: rksimon
Date: Sun Apr 23 15:05:21 2017
New Revision: 301137
URL: http://llvm.org/viewvc/llvm-project?rev=301137&view=rev
Log:
[X86][SSE] Add scheduling latency/throughput tests for (most) SSE41 instructions
Added:
llvm/trunk/test/CodeGen/X86/sse41-schedule.ll
Added: llvm/trunk/test/CodeGen/X86/sse41-schedule.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/sse41-schedule.ll?rev=301137&view=auto
==============================================================================
--- llvm/trunk/test/CodeGen/X86/sse41-schedule.ll (added)
+++ llvm/trunk/test/CodeGen/X86/sse41-schedule.ll Sun Apr 23 15:05:21 2017
@@ -0,0 +1,1938 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mattr=+sse4.1 | FileCheck %s --check-prefix=CHECK --check-prefix=GENERIC
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=slm | FileCheck %s --check-prefix=CHECK --check-prefix=SLM
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=sandybridge | FileCheck %s --check-prefix=CHECK --check-prefix=SANDY
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=ivybridge | FileCheck %s --check-prefix=CHECK --check-prefix=SANDY
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=haswell | FileCheck %s --check-prefix=CHECK --check-prefix=HASWELL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skylake | FileCheck %s --check-prefix=CHECK --check-prefix=HASWELL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=btver2 | FileCheck %s --check-prefix=CHECK --check-prefix=BTVER2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=znver1 | FileCheck %s --check-prefix=CHECK --check-prefix=BTVER2
+
+define <2 x double> @test_blendpd(<2 x double> %a0, <2 x double> %a1, <2 x double> *%a2) {
+; GENERIC-LABEL: test_blendpd:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: blendpd {{.*#+}} xmm0 = xmm0[0],xmm1[1]
+; GENERIC-NEXT: addpd %xmm1, %xmm0
+; GENERIC-NEXT: blendpd {{.*#+}} xmm0 = xmm0[0],mem[1]
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_blendpd:
+; SLM: # BB#0:
+; SLM-NEXT: blendpd {{.*#+}} xmm0 = xmm0[0],xmm1[1] sched: [1:1.00]
+; SLM-NEXT: addpd %xmm1, %xmm0 # sched: [3:1.00]
+; SLM-NEXT: blendpd {{.*#+}} xmm0 = xmm0[0],mem[1] sched: [4:1.00]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_blendpd:
+; SANDY: # BB#0:
+; SANDY-NEXT: vblendpd {{.*#+}} xmm0 = xmm0[0],xmm1[1] sched: [1:0.50]
+; SANDY-NEXT: vaddpd %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
+; SANDY-NEXT: vblendpd {{.*#+}} xmm0 = xmm0[0],mem[1] sched: [5:0.50]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_blendpd:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vblendpd {{.*#+}} xmm0 = xmm0[0],xmm1[1] sched: [1:0.33]
+; HASWELL-NEXT: vaddpd %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
+; HASWELL-NEXT: vblendpd {{.*#+}} xmm0 = xmm0[0],mem[1] sched: [5:0.50]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_blendpd:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vblendpd {{.*#+}} xmm0 = xmm0[0],xmm1[1] sched: [1:0.50]
+; BTVER2-NEXT: vaddpd %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
+; BTVER2-NEXT: vblendpd {{.*#+}} xmm0 = xmm0[0],mem[1] sched: [6:1.00]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = shufflevector <2 x double> %a0, <2 x double> %a1, <2 x i32> <i32 0, i32 3>
+ %2 = load <2 x double>, <2 x double> *%a2, align 16
+ %3 = fadd <2 x double> %a1, %1
+ %4 = shufflevector <2 x double> %3, <2 x double> %2, <2 x i32> <i32 0, i32 3>
+ ret <2 x double> %4
+}
+
+define <4 x float> @test_blendps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a2) {
+; GENERIC-LABEL: test_blendps:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: blendps {{.*#+}} xmm0 = xmm0[0],xmm1[1,2],xmm0[3]
+; GENERIC-NEXT: blendps {{.*#+}} xmm0 = xmm0[0],mem[1],xmm0[2,3]
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_blendps:
+; SLM: # BB#0:
+; SLM-NEXT: blendps {{.*#+}} xmm0 = xmm0[0],xmm1[1,2],xmm0[3] sched: [1:1.00]
+; SLM-NEXT: blendps {{.*#+}} xmm0 = xmm0[0],mem[1],xmm0[2,3] sched: [4:1.00]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_blendps:
+; SANDY: # BB#0:
+; SANDY-NEXT: vblendps {{.*#+}} xmm0 = xmm0[0],xmm1[1,2],xmm0[3] sched: [1:0.50]
+; SANDY-NEXT: vblendps {{.*#+}} xmm0 = xmm0[0],mem[1],xmm0[2,3] sched: [5:0.50]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_blendps:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vblendps {{.*#+}} xmm0 = xmm0[0],xmm1[1,2],xmm0[3] sched: [1:0.33]
+; HASWELL-NEXT: vblendps {{.*#+}} xmm0 = xmm0[0],mem[1],xmm0[2,3] sched: [5:0.50]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_blendps:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vblendps {{.*#+}} xmm0 = xmm0[0],xmm1[1,2],xmm0[3] sched: [1:0.50]
+; BTVER2-NEXT: vblendps {{.*#+}} xmm0 = xmm0[0],mem[1],xmm0[2,3] sched: [6:1.00]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = shufflevector <4 x float> %a0, <4 x float> %a1, <4 x i32> <i32 0, i32 5, i32 6, i32 3>
+ %2 = load <4 x float>, <4 x float> *%a2, align 16
+ %3 = shufflevector <4 x float> %1, <4 x float> %2, <4 x i32> <i32 0, i32 5, i32 2, i32 3>
+ ret <4 x float> %3
+}
+
+define <2 x double> @test_blendvpd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, <2 x double> *%a3) {
+; GENERIC-LABEL: test_blendvpd:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: movapd %xmm0, %xmm3
+; GENERIC-NEXT: movaps %xmm2, %xmm0
+; GENERIC-NEXT: blendvpd %xmm0, %xmm1, %xmm3
+; GENERIC-NEXT: blendvpd %xmm0, (%rdi), %xmm3
+; GENERIC-NEXT: movapd %xmm3, %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_blendvpd:
+; SLM: # BB#0:
+; SLM-NEXT: movapd %xmm0, %xmm3 # sched: [1:1.00]
+; SLM-NEXT: movaps %xmm2, %xmm0 # sched: [1:1.00]
+; SLM-NEXT: blendvpd %xmm0, %xmm1, %xmm3 # sched: [1:1.00]
+; SLM-NEXT: blendvpd %xmm0, (%rdi), %xmm3 # sched: [4:1.00]
+; SLM-NEXT: movapd %xmm3, %xmm0 # sched: [1:1.00]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_blendvpd:
+; SANDY: # BB#0:
+; SANDY-NEXT: vblendvpd %xmm2, %xmm1, %xmm0, %xmm0 # sched: [2:1.00]
+; SANDY-NEXT: vblendvpd %xmm2, (%rdi), %xmm0, %xmm0 # sched: [6:1.00]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_blendvpd:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vblendvpd %xmm2, %xmm1, %xmm0, %xmm0 # sched: [2:2.00]
+; HASWELL-NEXT: vblendvpd %xmm2, (%rdi), %xmm0, %xmm0 # sched: [6:2.00]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_blendvpd:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vblendvpd %xmm2, %xmm1, %xmm0, %xmm0 # sched: [2:1.00]
+; BTVER2-NEXT: vblendvpd %xmm2, (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = call <2 x double> @llvm.x86.sse41.blendvpd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2)
+ %2 = load <2 x double>, <2 x double> *%a3, align 16
+ %3 = call <2 x double> @llvm.x86.sse41.blendvpd(<2 x double> %1, <2 x double> %2, <2 x double> %a2)
+ ret <2 x double> %3
+}
+declare <2 x double> @llvm.x86.sse41.blendvpd(<2 x double>, <2 x double>, <2 x double>) nounwind readnone
+
+define <4 x float> @test_blendvps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, <4 x float> *%a3) {
+; GENERIC-LABEL: test_blendvps:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: movaps %xmm0, %xmm3
+; GENERIC-NEXT: movaps %xmm2, %xmm0
+; GENERIC-NEXT: blendvps %xmm0, %xmm1, %xmm3
+; GENERIC-NEXT: blendvps %xmm0, (%rdi), %xmm3
+; GENERIC-NEXT: movaps %xmm3, %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_blendvps:
+; SLM: # BB#0:
+; SLM-NEXT: movaps %xmm0, %xmm3 # sched: [1:1.00]
+; SLM-NEXT: movaps %xmm2, %xmm0 # sched: [1:1.00]
+; SLM-NEXT: blendvps %xmm0, %xmm1, %xmm3 # sched: [1:1.00]
+; SLM-NEXT: blendvps %xmm0, (%rdi), %xmm3 # sched: [4:1.00]
+; SLM-NEXT: movaps %xmm3, %xmm0 # sched: [1:1.00]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_blendvps:
+; SANDY: # BB#0:
+; SANDY-NEXT: vblendvps %xmm2, %xmm1, %xmm0, %xmm0 # sched: [2:1.00]
+; SANDY-NEXT: vblendvps %xmm2, (%rdi), %xmm0, %xmm0 # sched: [6:1.00]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_blendvps:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vblendvps %xmm2, %xmm1, %xmm0, %xmm0 # sched: [2:2.00]
+; HASWELL-NEXT: vblendvps %xmm2, (%rdi), %xmm0, %xmm0 # sched: [6:2.00]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_blendvps:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vblendvps %xmm2, %xmm1, %xmm0, %xmm0 # sched: [2:1.00]
+; BTVER2-NEXT: vblendvps %xmm2, (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = call <4 x float> @llvm.x86.sse41.blendvps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2)
+ %2 = load <4 x float>, <4 x float> *%a3
+ %3 = call <4 x float> @llvm.x86.sse41.blendvps(<4 x float> %1, <4 x float> %2, <4 x float> %a2)
+ ret <4 x float> %3
+}
+declare <4 x float> @llvm.x86.sse41.blendvps(<4 x float>, <4 x float>, <4 x float>) nounwind readnone
+
+define <2 x double> @test_dppd(<2 x double> %a0, <2 x double> %a1, <2 x double> *%a2) {
+; GENERIC-LABEL: test_dppd:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: dppd $7, %xmm1, %xmm0
+; GENERIC-NEXT: dppd $7, (%rdi), %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_dppd:
+; SLM: # BB#0:
+; SLM-NEXT: dppd $7, %xmm1, %xmm0 # sched: [3:1.00]
+; SLM-NEXT: dppd $7, (%rdi), %xmm0 # sched: [6:1.00]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_dppd:
+; SANDY: # BB#0:
+; SANDY-NEXT: vdppd $7, %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
+; SANDY-NEXT: vdppd $7, (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_dppd:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vdppd $7, %xmm1, %xmm0, %xmm0 # sched: [9:1.00]
+; HASWELL-NEXT: vdppd $7, (%rdi), %xmm0, %xmm0 # sched: [13:1.00]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_dppd:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vdppd $7, %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
+; BTVER2-NEXT: vdppd $7, (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = call <2 x double> @llvm.x86.sse41.dppd(<2 x double> %a0, <2 x double> %a1, i8 7)
+ %2 = load <2 x double>, <2 x double> *%a2, align 16
+ %3 = call <2 x double> @llvm.x86.sse41.dppd(<2 x double> %1, <2 x double> %2, i8 7)
+ ret <2 x double> %3
+}
+declare <2 x double> @llvm.x86.sse41.dppd(<2 x double>, <2 x double>, i8) nounwind readnone
+
+define <4 x float> @test_dpps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a2) {
+; GENERIC-LABEL: test_dpps:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: dpps $7, %xmm1, %xmm0
+; GENERIC-NEXT: dpps $7, (%rdi), %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_dpps:
+; SLM: # BB#0:
+; SLM-NEXT: dpps $7, %xmm1, %xmm0 # sched: [3:1.00]
+; SLM-NEXT: dpps $7, (%rdi), %xmm0 # sched: [6:1.00]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_dpps:
+; SANDY: # BB#0:
+; SANDY-NEXT: vdpps $7, %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
+; SANDY-NEXT: vdpps $7, (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_dpps:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vdpps $7, %xmm1, %xmm0, %xmm0 # sched: [14:2.00]
+; HASWELL-NEXT: vdpps $7, (%rdi), %xmm0, %xmm0 # sched: [18:2.00]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_dpps:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vdpps $7, %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
+; BTVER2-NEXT: vdpps $7, (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = call <4 x float> @llvm.x86.sse41.dpps(<4 x float> %a0, <4 x float> %a1, i8 7)
+ %2 = load <4 x float>, <4 x float> *%a2, align 16
+ %3 = call <4 x float> @llvm.x86.sse41.dpps(<4 x float> %1, <4 x float> %2, i8 7)
+ ret <4 x float> %3
+}
+declare <4 x float> @llvm.x86.sse41.dpps(<4 x float>, <4 x float>, i8) nounwind readnone
+
+define <4 x float> @test_insertps(<4 x float> %a0, <4 x float> %a1, float *%a2) {
+; GENERIC-LABEL: test_insertps:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: insertps {{.*#+}} xmm0 = zero,xmm1[0],xmm0[2,3]
+; GENERIC-NEXT: insertps {{.*#+}} xmm0 = xmm0[0,1,2],mem[0]
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_insertps:
+; SLM: # BB#0:
+; SLM-NEXT: insertps {{.*#+}} xmm0 = zero,xmm1[0],xmm0[2,3] sched: [1:1.00]
+; SLM-NEXT: insertps {{.*#+}} xmm0 = xmm0[0,1,2],mem[0] sched: [4:1.00]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_insertps:
+; SANDY: # BB#0:
+; SANDY-NEXT: vinsertps {{.*#+}} xmm0 = zero,xmm1[0],xmm0[2,3] sched: [1:1.00]
+; SANDY-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0,1,2],mem[0] sched: [5:1.00]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_insertps:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vinsertps {{.*#+}} xmm0 = zero,xmm1[0],xmm0[2,3] sched: [1:1.00]
+; HASWELL-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0,1,2],mem[0] sched: [5:1.00]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_insertps:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vinsertps {{.*#+}} xmm0 = zero,xmm1[0],xmm0[2,3] sched: [1:0.50]
+; BTVER2-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0,1,2],mem[0] sched: [6:1.00]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = call <4 x float> @llvm.x86.sse41.insertps(<4 x float> %a0, <4 x float> %a1, i8 17)
+ %2 = load float, float *%a2
+ %3 = insertelement <4 x float> %1, float %2, i32 3
+ ret <4 x float> %3
+}
+declare <4 x float> @llvm.x86.sse41.insertps(<4 x float>, <4 x float>, i8) nounwind readnone
+
+define <2 x i64> @test_movntdqa(i8* %a0) {
+; GENERIC-LABEL: test_movntdqa:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: movntdqa (%rdi), %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_movntdqa:
+; SLM: # BB#0:
+; SLM-NEXT: movntdqa (%rdi), %xmm0 # sched: [3:1.00]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_movntdqa:
+; SANDY: # BB#0:
+; SANDY-NEXT: vmovntdqa (%rdi), %xmm0 # sched: [4:0.50]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_movntdqa:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vmovntdqa (%rdi), %xmm0 # sched: [4:0.50]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_movntdqa:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vmovntdqa (%rdi), %xmm0 # sched: [5:1.00]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = call <2 x i64> @llvm.x86.sse41.movntdqa(i8* %a0)
+ ret <2 x i64> %1
+}
+declare <2 x i64> @llvm.x86.sse41.movntdqa(i8*) nounwind readnone
+
+define <8 x i16> @test_mpsadbw(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) {
+; GENERIC-LABEL: test_mpsadbw:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: mpsadbw $7, %xmm1, %xmm0
+; GENERIC-NEXT: mpsadbw $7, (%rdi), %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_mpsadbw:
+; SLM: # BB#0:
+; SLM-NEXT: mpsadbw $7, %xmm1, %xmm0 # sched: [7:1.00]
+; SLM-NEXT: mpsadbw $7, (%rdi), %xmm0 # sched: [10:1.00]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_mpsadbw:
+; SANDY: # BB#0:
+; SANDY-NEXT: vmpsadbw $7, %xmm1, %xmm0, %xmm0 # sched: [6:1.00]
+; SANDY-NEXT: vmpsadbw $7, (%rdi), %xmm0, %xmm0 # sched: [6:1.00]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_mpsadbw:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vmpsadbw $7, %xmm1, %xmm0, %xmm0 # sched: [6:2.00]
+; HASWELL-NEXT: vmpsadbw $7, (%rdi), %xmm0, %xmm0 # sched: [6:2.00]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_mpsadbw:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vmpsadbw $7, %xmm1, %xmm0, %xmm0 # sched: [3:2.00]
+; BTVER2-NEXT: vmpsadbw $7, (%rdi), %xmm0, %xmm0 # sched: [8:2.00]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = call <8 x i16> @llvm.x86.sse41.mpsadbw(<16 x i8> %a0, <16 x i8> %a1, i8 7)
+ %2 = bitcast <8 x i16> %1 to <16 x i8>
+ %3 = load <16 x i8>, <16 x i8> *%a2, align 16
+ %4 = call <8 x i16> @llvm.x86.sse41.mpsadbw(<16 x i8> %2, <16 x i8> %3, i8 7)
+ ret <8 x i16> %4
+}
+declare <8 x i16> @llvm.x86.sse41.mpsadbw(<16 x i8>, <16 x i8>, i8) nounwind readnone
+
+define <8 x i16> @test_packusdw(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) {
+; GENERIC-LABEL: test_packusdw:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: packusdw %xmm1, %xmm0
+; GENERIC-NEXT: packusdw (%rdi), %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_packusdw:
+; SLM: # BB#0:
+; SLM-NEXT: packusdw %xmm1, %xmm0 # sched: [1:1.00]
+; SLM-NEXT: packusdw (%rdi), %xmm0 # sched: [4:1.00]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_packusdw:
+; SANDY: # BB#0:
+; SANDY-NEXT: vpackusdw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; SANDY-NEXT: vpackusdw (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_packusdw:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vpackusdw %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
+; HASWELL-NEXT: vpackusdw (%rdi), %xmm0, %xmm0 # sched: [5:1.00]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_packusdw:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vpackusdw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BTVER2-NEXT: vpackusdw (%rdi), %xmm0, %xmm0 # sched: [6:1.00]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = call <8 x i16> @llvm.x86.sse41.packusdw(<4 x i32> %a0, <4 x i32> %a1)
+ %2 = bitcast <8 x i16> %1 to <4 x i32>
+ %3 = load <4 x i32>, <4 x i32> *%a2, align 16
+ %4 = call <8 x i16> @llvm.x86.sse41.packusdw(<4 x i32> %2, <4 x i32> %3)
+ ret <8 x i16> %4
+}
+declare <8 x i16> @llvm.x86.sse41.packusdw(<4 x i32>, <4 x i32>) nounwind readnone
+
+define <16 x i8> @test_pblendvb(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> %a2, <16 x i8> *%a3) {
+; GENERIC-LABEL: test_pblendvb:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: movdqa %xmm0, %xmm3
+; GENERIC-NEXT: movaps %xmm2, %xmm0
+; GENERIC-NEXT: pblendvb %xmm0, %xmm1, %xmm3
+; GENERIC-NEXT: pblendvb %xmm0, (%rdi), %xmm3
+; GENERIC-NEXT: movdqa %xmm3, %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_pblendvb:
+; SLM: # BB#0:
+; SLM-NEXT: movdqa %xmm0, %xmm3 # sched: [1:0.50]
+; SLM-NEXT: movaps %xmm2, %xmm0 # sched: [1:1.00]
+; SLM-NEXT: pblendvb %xmm0, %xmm1, %xmm3 # sched: [1:1.00]
+; SLM-NEXT: pblendvb %xmm0, (%rdi), %xmm3 # sched: [4:1.00]
+; SLM-NEXT: movdqa %xmm3, %xmm0 # sched: [1:0.50]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_pblendvb:
+; SANDY: # BB#0:
+; SANDY-NEXT: vpblendvb %xmm2, %xmm1, %xmm0, %xmm0 # sched: [2:1.00]
+; SANDY-NEXT: vpblendvb %xmm2, (%rdi), %xmm0, %xmm0 # sched: [6:1.00]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_pblendvb:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vpblendvb %xmm2, %xmm1, %xmm0, %xmm0 # sched: [2:2.00]
+; HASWELL-NEXT: vpblendvb %xmm2, (%rdi), %xmm0, %xmm0 # sched: [6:2.00]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_pblendvb:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vpblendvb %xmm2, %xmm1, %xmm0, %xmm0 # sched: [2:1.00]
+; BTVER2-NEXT: vpblendvb %xmm2, (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> %a2)
+ %2 = load <16 x i8>, <16 x i8> *%a3, align 16
+ %3 = call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %1, <16 x i8> %2, <16 x i8> %a2)
+ ret <16 x i8> %3
+}
+declare <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8>, <16 x i8>, <16 x i8>) nounwind readnone
+
+define <8 x i16> @test_pblendw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) {
+; GENERIC-LABEL: test_pblendw:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7]
+; GENERIC-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1],mem[2,3],xmm0[4,5,6],mem[7]
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_pblendw:
+; SLM: # BB#0:
+; SLM-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7] sched: [1:1.00]
+; SLM-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1],mem[2,3],xmm0[4,5,6],mem[7] sched: [4:1.00]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_pblendw:
+; SANDY: # BB#0:
+; SANDY-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7] sched: [1:0.50]
+; SANDY-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],mem[2,3],xmm0[4,5,6],mem[7] sched: [5:0.50]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_pblendw:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7] sched: [1:1.00]
+; HASWELL-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],mem[2,3],xmm0[4,5,6],mem[7] sched: [4:1.00]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_pblendw:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7] sched: [1:0.50]
+; BTVER2-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],mem[2,3],xmm0[4,5,6],mem[7] sched: [6:1.00]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = shufflevector <8 x i16> %a0, <8 x i16> %a1, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+ %2 = load <8 x i16>, <8 x i16> *%a2, align 16
+ %3 = shufflevector <8 x i16> %1, <8 x i16> %2, <8 x i32> <i32 0, i32 1, i32 10, i32 11, i32 4, i32 5, i32 6, i32 15>
+ ret <8 x i16> %3
+}
+
+define <2 x i64> @test_pcmpeqq(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> *%a2) {
+; GENERIC-LABEL: test_pcmpeqq:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: pcmpeqq %xmm1, %xmm0
+; GENERIC-NEXT: pcmpeqq (%rdi), %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_pcmpeqq:
+; SLM: # BB#0:
+; SLM-NEXT: pcmpeqq %xmm1, %xmm0 # sched: [1:0.50]
+; SLM-NEXT: pcmpeqq (%rdi), %xmm0 # sched: [4:1.00]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_pcmpeqq:
+; SANDY: # BB#0:
+; SANDY-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; SANDY-NEXT: vpcmpeqq (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_pcmpeqq:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; HASWELL-NEXT: vpcmpeqq (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_pcmpeqq:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BTVER2-NEXT: vpcmpeqq (%rdi), %xmm0, %xmm0 # sched: [6:1.00]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = icmp eq <2 x i64> %a0, %a1
+ %2 = sext <2 x i1> %1 to <2 x i64>
+ %3 = load <2 x i64>, <2 x i64>*%a2, align 16
+ %4 = icmp eq <2 x i64> %2, %3
+ %5 = sext <2 x i1> %4 to <2 x i64>
+ ret <2 x i64> %5
+}
+
+define i32 @test_pextrb(<16 x i8> %a0, i8 *%a1) {
+; GENERIC-LABEL: test_pextrb:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: pextrb $3, %xmm0, %eax
+; GENERIC-NEXT: pextrb $1, %xmm0, (%rdi)
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_pextrb:
+; SLM: # BB#0:
+; SLM-NEXT: pextrb $3, %xmm0, %eax # sched: [1:1.00]
+; SLM-NEXT: pextrb $1, %xmm0, (%rdi) # sched: [4:2.00]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_pextrb:
+; SANDY: # BB#0:
+; SANDY-NEXT: vpextrb $3, %xmm0, %eax # sched: [1:0.50]
+; SANDY-NEXT: vpextrb $1, %xmm0, (%rdi) # sched: [5:1.00]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_pextrb:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vpextrb $3, %xmm0, %eax # sched: [1:1.00]
+; HASWELL-NEXT: vpextrb $1, %xmm0, (%rdi) # sched: [5:1.00]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_pextrb:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vpextrb $3, %xmm0, %eax # sched: [1:0.50]
+; BTVER2-NEXT: vpextrb $1, %xmm0, (%rdi) # sched: [6:1.00]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = extractelement <16 x i8> %a0, i32 3
+ %2 = extractelement <16 x i8> %a0, i32 1
+ store i8 %2, i8 *%a1
+ %3 = zext i8 %1 to i32
+ ret i32 %3
+}
+
+define i32 @test_pextrd(<4 x i32> %a0, i32 *%a1) {
+; GENERIC-LABEL: test_pextrd:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: pextrd $3, %xmm0, %eax
+; GENERIC-NEXT: pextrd $1, %xmm0, (%rdi)
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_pextrd:
+; SLM: # BB#0:
+; SLM-NEXT: pextrd $3, %xmm0, %eax # sched: [1:1.00]
+; SLM-NEXT: pextrd $1, %xmm0, (%rdi) # sched: [4:2.00]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_pextrd:
+; SANDY: # BB#0:
+; SANDY-NEXT: vpextrd $3, %xmm0, %eax # sched: [1:0.50]
+; SANDY-NEXT: vpextrd $1, %xmm0, (%rdi) # sched: [5:1.00]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_pextrd:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vpextrd $3, %xmm0, %eax # sched: [1:1.00]
+; HASWELL-NEXT: vpextrd $1, %xmm0, (%rdi) # sched: [5:1.00]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_pextrd:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vpextrd $3, %xmm0, %eax # sched: [1:0.50]
+; BTVER2-NEXT: vpextrd $1, %xmm0, (%rdi) # sched: [6:1.00]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = extractelement <4 x i32> %a0, i32 3
+ %2 = extractelement <4 x i32> %a0, i32 1
+ store i32 %2, i32 *%a1
+ ret i32 %1
+}
+
+define i64 @test_pextrq(<2 x i64> %a0, <2 x i64> %a1, i64 *%a2) {
+; GENERIC-LABEL: test_pextrq:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: pextrq $1, %xmm0, %rax
+; GENERIC-NEXT: pextrq $1, %xmm0, (%rdi)
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_pextrq:
+; SLM: # BB#0:
+; SLM-NEXT: pextrq $1, %xmm0, %rax # sched: [1:1.00]
+; SLM-NEXT: pextrq $1, %xmm0, (%rdi) # sched: [4:2.00]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_pextrq:
+; SANDY: # BB#0:
+; SANDY-NEXT: vpextrq $1, %xmm0, %rax # sched: [1:0.50]
+; SANDY-NEXT: vpextrq $1, %xmm0, (%rdi) # sched: [5:1.00]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_pextrq:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vpextrq $1, %xmm0, %rax # sched: [1:1.00]
+; HASWELL-NEXT: vpextrq $1, %xmm0, (%rdi) # sched: [5:1.00]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_pextrq:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vpextrq $1, %xmm0, %rax # sched: [1:0.50]
+; BTVER2-NEXT: vpextrq $1, %xmm0, (%rdi) # sched: [6:1.00]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = extractelement <2 x i64> %a0, i32 1
+ %2 = extractelement <2 x i64> %a0, i32 1
+ store i64 %2, i64 *%a2
+ ret i64 %1
+}
+
+define i32 @test_pextrw(<8 x i16> %a0, i16 *%a1) {
+; GENERIC-LABEL: test_pextrw:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: pextrw $3, %xmm0, %eax
+; GENERIC-NEXT: pextrw $1, %xmm0, (%rdi)
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_pextrw:
+; SLM: # BB#0:
+; SLM-NEXT: pextrw $3, %xmm0, %eax # sched: [4:1.00]
+; SLM-NEXT: pextrw $1, %xmm0, (%rdi) # sched: [4:2.00]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_pextrw:
+; SANDY: # BB#0:
+; SANDY-NEXT: vpextrw $3, %xmm0, %eax # sched: [1:0.50]
+; SANDY-NEXT: vpextrw $1, %xmm0, (%rdi) # sched: [5:1.00]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_pextrw:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vpextrw $3, %xmm0, %eax # sched: [1:1.00]
+; HASWELL-NEXT: vpextrw $1, %xmm0, (%rdi) # sched: [5:1.00]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_pextrw:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vpextrw $3, %xmm0, %eax # sched: [1:0.50]
+; BTVER2-NEXT: vpextrw $1, %xmm0, (%rdi) # sched: [6:1.00]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = extractelement <8 x i16> %a0, i32 3
+ %2 = extractelement <8 x i16> %a0, i32 1
+ store i16 %2, i16 *%a1
+ %3 = zext i16 %1 to i32
+ ret i32 %3
+}
+
+define <8 x i16> @test_phminposuw(<8 x i16> *%a0) {
+; GENERIC-LABEL: test_phminposuw:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: phminposuw (%rdi), %xmm0
+; GENERIC-NEXT: phminposuw %xmm0, %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_phminposuw:
+; SLM: # BB#0:
+; SLM-NEXT: phminposuw (%rdi), %xmm0 # sched: [7:1.00]
+; SLM-NEXT: phminposuw %xmm0, %xmm0 # sched: [4:1.00]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_phminposuw:
+; SANDY: # BB#0:
+; SANDY-NEXT: vphminposuw (%rdi), %xmm0 # sched: [9:1.00]
+; SANDY-NEXT: vphminposuw %xmm0, %xmm0 # sched: [5:1.00]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_phminposuw:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vphminposuw (%rdi), %xmm0 # sched: [9:1.00]
+; HASWELL-NEXT: vphminposuw %xmm0, %xmm0 # sched: [5:1.00]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_phminposuw:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vphminposuw (%rdi), %xmm0 # sched: [7:1.00]
+; BTVER2-NEXT: vphminposuw %xmm0, %xmm0 # sched: [2:1.00]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = load <8 x i16>, <8 x i16> *%a0, align 16
+ %2 = call <8 x i16> @llvm.x86.sse41.phminposuw(<8 x i16> %1)
+ %3 = call <8 x i16> @llvm.x86.sse41.phminposuw(<8 x i16> %2)
+ ret <8 x i16> %3
+}
+declare <8 x i16> @llvm.x86.sse41.phminposuw(<8 x i16>) nounwind readnone
+
+define <16 x i8> @test_pinsrb(<16 x i8> %a0, i8 %a1, i8 *%a2) {
+; GENERIC-LABEL: test_pinsrb:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: pinsrb $1, %edi, %xmm0
+; GENERIC-NEXT: pinsrb $3, (%rsi), %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_pinsrb:
+; SLM: # BB#0:
+; SLM-NEXT: pinsrb $1, %edi, %xmm0 # sched: [1:1.00]
+; SLM-NEXT: pinsrb $3, (%rsi), %xmm0 # sched: [4:1.00]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_pinsrb:
+; SANDY: # BB#0:
+; SANDY-NEXT: vpinsrb $1, %edi, %xmm0, %xmm0 # sched: [1:0.50]
+; SANDY-NEXT: vpinsrb $3, (%rsi), %xmm0, %xmm0 # sched: [5:0.50]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_pinsrb:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vpinsrb $1, %edi, %xmm0, %xmm0 # sched: [1:1.00]
+; HASWELL-NEXT: vpinsrb $3, (%rsi), %xmm0, %xmm0 # sched: [5:1.00]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_pinsrb:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vpinsrb $1, %edi, %xmm0, %xmm0 # sched: [1:0.50]
+; BTVER2-NEXT: vpinsrb $3, (%rsi), %xmm0, %xmm0 # sched: [6:1.00]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = insertelement <16 x i8> %a0, i8 %a1, i32 1
+ %2 = load i8, i8 *%a2
+ %3 = insertelement <16 x i8> %1, i8 %2, i32 3
+ ret <16 x i8> %3
+}
+
+define <4 x i32> @test_pinsrd(<4 x i32> %a0, i32 %a1, i32 *%a2) {
+; GENERIC-LABEL: test_pinsrd:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: pinsrd $1, %edi, %xmm0
+; GENERIC-NEXT: pinsrd $3, (%rsi), %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_pinsrd:
+; SLM: # BB#0:
+; SLM-NEXT: pinsrd $1, %edi, %xmm0 # sched: [1:1.00]
+; SLM-NEXT: pinsrd $3, (%rsi), %xmm0 # sched: [4:1.00]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_pinsrd:
+; SANDY: # BB#0:
+; SANDY-NEXT: vpinsrd $1, %edi, %xmm0, %xmm0 # sched: [1:0.50]
+; SANDY-NEXT: vpinsrd $3, (%rsi), %xmm0, %xmm0 # sched: [5:0.50]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_pinsrd:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vpinsrd $1, %edi, %xmm0, %xmm0 # sched: [1:1.00]
+; HASWELL-NEXT: vpinsrd $3, (%rsi), %xmm0, %xmm0 # sched: [5:1.00]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_pinsrd:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vpinsrd $1, %edi, %xmm0, %xmm0 # sched: [1:0.50]
+; BTVER2-NEXT: vpinsrd $3, (%rsi), %xmm0, %xmm0 # sched: [6:1.00]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = insertelement <4 x i32> %a0, i32 %a1, i32 1
+ %2 = load i32, i32 *%a2
+ %3 = insertelement <4 x i32> %1, i32 %2, i32 3
+ ret <4 x i32> %3
+}
+
+define <2 x i64> @test_pinsrq(<2 x i64> %a0, <2 x i64> %a1, i64 %a2, i64 *%a3) {
+; GENERIC-LABEL: test_pinsrq:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: pinsrq $1, %rdi, %xmm0
+; GENERIC-NEXT: pinsrq $1, (%rsi), %xmm1
+; GENERIC-NEXT: paddq %xmm1, %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_pinsrq:
+; SLM: # BB#0:
+; SLM-NEXT: pinsrq $1, (%rsi), %xmm1 # sched: [4:1.00]
+; SLM-NEXT: pinsrq $1, %rdi, %xmm0 # sched: [1:1.00]
+; SLM-NEXT: paddq %xmm1, %xmm0 # sched: [1:0.50]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_pinsrq:
+; SANDY: # BB#0:
+; SANDY-NEXT: vpinsrq $1, %rdi, %xmm0, %xmm0 # sched: [1:0.50]
+; SANDY-NEXT: vpinsrq $1, (%rsi), %xmm1, %xmm1 # sched: [5:0.50]
+; SANDY-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_pinsrq:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vpinsrq $1, %rdi, %xmm0, %xmm0 # sched: [1:1.00]
+; HASWELL-NEXT: vpinsrq $1, (%rsi), %xmm1, %xmm1 # sched: [5:1.00]
+; HASWELL-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_pinsrq:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vpinsrq $1, (%rsi), %xmm1, %xmm1 # sched: [6:1.00]
+; BTVER2-NEXT: vpinsrq $1, %rdi, %xmm0, %xmm0 # sched: [1:0.50]
+; BTVER2-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = insertelement <2 x i64> %a0, i64 %a2, i32 1
+ %2 = load i64, i64 *%a3
+ %3 = insertelement <2 x i64> %a1, i64 %2, i32 1
+ %4 = add <2 x i64> %1, %3
+ ret <2 x i64> %4
+}
+
+define <16 x i8> @test_pmaxsb(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) {
+; GENERIC-LABEL: test_pmaxsb:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: pmaxsb %xmm1, %xmm0
+; GENERIC-NEXT: pmaxsb (%rdi), %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_pmaxsb:
+; SLM: # BB#0:
+; SLM-NEXT: pmaxsb %xmm1, %xmm0 # sched: [1:0.50]
+; SLM-NEXT: pmaxsb (%rdi), %xmm0 # sched: [4:1.00]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_pmaxsb:
+; SANDY: # BB#0:
+; SANDY-NEXT: vpmaxsb %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; SANDY-NEXT: vpmaxsb (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_pmaxsb:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vpmaxsb %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; HASWELL-NEXT: vpmaxsb (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_pmaxsb:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vpmaxsb %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BTVER2-NEXT: vpmaxsb (%rdi), %xmm0, %xmm0 # sched: [6:1.00]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = call <16 x i8> @llvm.x86.sse41.pmaxsb(<16 x i8> %a0, <16 x i8> %a1)
+ %2 = load <16 x i8>, <16 x i8> *%a2, align 16
+ %3 = call <16 x i8> @llvm.x86.sse41.pmaxsb(<16 x i8> %1, <16 x i8> %2)
+ ret <16 x i8> %3
+}
+declare <16 x i8> @llvm.x86.sse41.pmaxsb(<16 x i8>, <16 x i8>) nounwind readnone
+
+define <4 x i32> @test_pmaxsd(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) {
+; GENERIC-LABEL: test_pmaxsd:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: pmaxsd %xmm1, %xmm0
+; GENERIC-NEXT: pmaxsd (%rdi), %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_pmaxsd:
+; SLM: # BB#0:
+; SLM-NEXT: pmaxsd %xmm1, %xmm0 # sched: [1:0.50]
+; SLM-NEXT: pmaxsd (%rdi), %xmm0 # sched: [4:1.00]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_pmaxsd:
+; SANDY: # BB#0:
+; SANDY-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; SANDY-NEXT: vpmaxsd (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_pmaxsd:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; HASWELL-NEXT: vpmaxsd (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_pmaxsd:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BTVER2-NEXT: vpmaxsd (%rdi), %xmm0, %xmm0 # sched: [6:1.00]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = call <4 x i32> @llvm.x86.sse41.pmaxsd(<4 x i32> %a0, <4 x i32> %a1)
+ %2 = load <4 x i32>, <4 x i32> *%a2, align 16
+ %3 = call <4 x i32> @llvm.x86.sse41.pmaxsd(<4 x i32> %1, <4 x i32> %2)
+ ret <4 x i32> %3
+}
+declare <4 x i32> @llvm.x86.sse41.pmaxsd(<4 x i32>, <4 x i32>) nounwind readnone
+
+define <4 x i32> @test_pmaxud(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) {
+; GENERIC-LABEL: test_pmaxud:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: pmaxud %xmm1, %xmm0
+; GENERIC-NEXT: pmaxud (%rdi), %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_pmaxud:
+; SLM: # BB#0:
+; SLM-NEXT: pmaxud %xmm1, %xmm0 # sched: [1:0.50]
+; SLM-NEXT: pmaxud (%rdi), %xmm0 # sched: [4:1.00]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_pmaxud:
+; SANDY: # BB#0:
+; SANDY-NEXT: vpmaxud %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; SANDY-NEXT: vpmaxud (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_pmaxud:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vpmaxud %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; HASWELL-NEXT: vpmaxud (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_pmaxud:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vpmaxud %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BTVER2-NEXT: vpmaxud (%rdi), %xmm0, %xmm0 # sched: [6:1.00]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = call <4 x i32> @llvm.x86.sse41.pmaxud(<4 x i32> %a0, <4 x i32> %a1)
+ %2 = load <4 x i32>, <4 x i32> *%a2, align 16
+ %3 = call <4 x i32> @llvm.x86.sse41.pmaxud(<4 x i32> %1, <4 x i32> %2)
+ ret <4 x i32> %3
+}
+declare <4 x i32> @llvm.x86.sse41.pmaxud(<4 x i32>, <4 x i32>) nounwind readnone
+
+define <8 x i16> @test_pmaxuw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) {
+; GENERIC-LABEL: test_pmaxuw:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: pmaxuw %xmm1, %xmm0
+; GENERIC-NEXT: pmaxuw (%rdi), %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_pmaxuw:
+; SLM: # BB#0:
+; SLM-NEXT: pmaxuw %xmm1, %xmm0 # sched: [1:0.50]
+; SLM-NEXT: pmaxuw (%rdi), %xmm0 # sched: [4:1.00]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_pmaxuw:
+; SANDY: # BB#0:
+; SANDY-NEXT: vpmaxuw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; SANDY-NEXT: vpmaxuw (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_pmaxuw:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vpmaxuw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; HASWELL-NEXT: vpmaxuw (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_pmaxuw:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vpmaxuw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BTVER2-NEXT: vpmaxuw (%rdi), %xmm0, %xmm0 # sched: [6:1.00]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = call <8 x i16> @llvm.x86.sse41.pmaxuw(<8 x i16> %a0, <8 x i16> %a1)
+ %2 = load <8 x i16>, <8 x i16> *%a2, align 16
+ %3 = call <8 x i16> @llvm.x86.sse41.pmaxuw(<8 x i16> %1, <8 x i16> %2)
+ ret <8 x i16> %3
+}
+declare <8 x i16> @llvm.x86.sse41.pmaxuw(<8 x i16>, <8 x i16>) nounwind readnone
+
+define <16 x i8> @test_pminsb(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) {
+; GENERIC-LABEL: test_pminsb:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: pminsb %xmm1, %xmm0
+; GENERIC-NEXT: pminsb (%rdi), %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_pminsb:
+; SLM: # BB#0:
+; SLM-NEXT: pminsb %xmm1, %xmm0 # sched: [1:0.50]
+; SLM-NEXT: pminsb (%rdi), %xmm0 # sched: [4:1.00]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_pminsb:
+; SANDY: # BB#0:
+; SANDY-NEXT: vpminsb %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; SANDY-NEXT: vpminsb (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_pminsb:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vpminsb %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; HASWELL-NEXT: vpminsb (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_pminsb:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vpminsb %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BTVER2-NEXT: vpminsb (%rdi), %xmm0, %xmm0 # sched: [6:1.00]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = call <16 x i8> @llvm.x86.sse41.pminsb(<16 x i8> %a0, <16 x i8> %a1)
+ %2 = load <16 x i8>, <16 x i8> *%a2, align 16
+ %3 = call <16 x i8> @llvm.x86.sse41.pminsb(<16 x i8> %1, <16 x i8> %2)
+ ret <16 x i8> %3
+}
+declare <16 x i8> @llvm.x86.sse41.pminsb(<16 x i8>, <16 x i8>) nounwind readnone
+
+define <4 x i32> @test_pminsd(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) {
+; GENERIC-LABEL: test_pminsd:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: pminsd %xmm1, %xmm0
+; GENERIC-NEXT: pminsd (%rdi), %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_pminsd:
+; SLM: # BB#0:
+; SLM-NEXT: pminsd %xmm1, %xmm0 # sched: [1:0.50]
+; SLM-NEXT: pminsd (%rdi), %xmm0 # sched: [4:1.00]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_pminsd:
+; SANDY: # BB#0:
+; SANDY-NEXT: vpminsd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; SANDY-NEXT: vpminsd (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_pminsd:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vpminsd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; HASWELL-NEXT: vpminsd (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_pminsd:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vpminsd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BTVER2-NEXT: vpminsd (%rdi), %xmm0, %xmm0 # sched: [6:1.00]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = call <4 x i32> @llvm.x86.sse41.pminsd(<4 x i32> %a0, <4 x i32> %a1)
+ %2 = load <4 x i32>, <4 x i32> *%a2, align 16
+ %3 = call <4 x i32> @llvm.x86.sse41.pminsd(<4 x i32> %1, <4 x i32> %2)
+ ret <4 x i32> %3
+}
+declare <4 x i32> @llvm.x86.sse41.pminsd(<4 x i32>, <4 x i32>) nounwind readnone
+
+define <4 x i32> @test_pminud(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) {
+; GENERIC-LABEL: test_pminud:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: pminud %xmm1, %xmm0
+; GENERIC-NEXT: pminud (%rdi), %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_pminud:
+; SLM: # BB#0:
+; SLM-NEXT: pminud %xmm1, %xmm0 # sched: [1:0.50]
+; SLM-NEXT: pminud (%rdi), %xmm0 # sched: [4:1.00]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_pminud:
+; SANDY: # BB#0:
+; SANDY-NEXT: vpminud %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; SANDY-NEXT: vpminud (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_pminud:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vpminud %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; HASWELL-NEXT: vpminud (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_pminud:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vpminud %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BTVER2-NEXT: vpminud (%rdi), %xmm0, %xmm0 # sched: [6:1.00]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = call <4 x i32> @llvm.x86.sse41.pminud(<4 x i32> %a0, <4 x i32> %a1)
+ %2 = load <4 x i32>, <4 x i32> *%a2, align 16
+ %3 = call <4 x i32> @llvm.x86.sse41.pminud(<4 x i32> %1, <4 x i32> %2)
+ ret <4 x i32> %3
+}
+declare <4 x i32> @llvm.x86.sse41.pminud(<4 x i32>, <4 x i32>) nounwind readnone
+
+define <8 x i16> @test_pminuw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) {
+; GENERIC-LABEL: test_pminuw:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: pminuw %xmm1, %xmm0
+; GENERIC-NEXT: pminuw (%rdi), %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_pminuw:
+; SLM: # BB#0:
+; SLM-NEXT: pminuw %xmm1, %xmm0 # sched: [1:0.50]
+; SLM-NEXT: pminuw (%rdi), %xmm0 # sched: [4:1.00]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_pminuw:
+; SANDY: # BB#0:
+; SANDY-NEXT: vpminuw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; SANDY-NEXT: vpminuw (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_pminuw:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vpminuw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; HASWELL-NEXT: vpminuw (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_pminuw:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vpminuw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BTVER2-NEXT: vpminuw (%rdi), %xmm0, %xmm0 # sched: [6:1.00]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = call <8 x i16> @llvm.x86.sse41.pminuw(<8 x i16> %a0, <8 x i16> %a1)
+ %2 = load <8 x i16>, <8 x i16> *%a2, align 16
+ %3 = call <8 x i16> @llvm.x86.sse41.pminuw(<8 x i16> %1, <8 x i16> %2)
+ ret <8 x i16> %3
+}
+declare <8 x i16> @llvm.x86.sse41.pminuw(<8 x i16>, <8 x i16>) nounwind readnone
+
+define <8 x i16> @test_pmovsxbw(<16 x i8> %a0, <8 x i8> *%a1) {
+; GENERIC-LABEL: test_pmovsxbw:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: pmovsxbw %xmm0, %xmm1
+; GENERIC-NEXT: pmovsxbw (%rdi), %xmm0
+; GENERIC-NEXT: paddw %xmm1, %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_pmovsxbw:
+; SLM: # BB#0:
+; SLM-NEXT: pmovsxbw (%rdi), %xmm1 # sched: [4:1.00]
+; SLM-NEXT: pmovsxbw %xmm0, %xmm0 # sched: [1:1.00]
+; SLM-NEXT: paddw %xmm0, %xmm1 # sched: [1:0.50]
+; SLM-NEXT: movdqa %xmm1, %xmm0 # sched: [1:0.50]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_pmovsxbw:
+; SANDY: # BB#0:
+; SANDY-NEXT: vpmovsxbw %xmm0, %xmm0 # sched: [1:0.50]
+; SANDY-NEXT: vpmovsxbw (%rdi), %xmm1 # sched: [5:0.50]
+; SANDY-NEXT: vpaddw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_pmovsxbw:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vpmovsxbw %xmm0, %xmm0 # sched: [1:1.00]
+; HASWELL-NEXT: vpmovsxbw (%rdi), %xmm1 # sched: [5:1.00]
+; HASWELL-NEXT: vpaddw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_pmovsxbw:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vpmovsxbw (%rdi), %xmm1 # sched: [6:1.00]
+; BTVER2-NEXT: vpmovsxbw %xmm0, %xmm0 # sched: [1:0.50]
+; BTVER2-NEXT: vpaddw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = shufflevector <16 x i8> %a0, <16 x i8> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+ %2 = sext <8 x i8> %1 to <8 x i16>
+ %3 = load <8 x i8>, <8 x i8>* %a1, align 1
+ %4 = sext <8 x i8> %3 to <8 x i16>
+ %5 = add <8 x i16> %2, %4
+ ret <8 x i16> %5
+}
+
+define <4 x i32> @test_pmovsxbd(<16 x i8> %a0, <4 x i8> *%a1) {
+; GENERIC-LABEL: test_pmovsxbd:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: pmovsxbd %xmm0, %xmm1
+; GENERIC-NEXT: pmovsxbd (%rdi), %xmm0
+; GENERIC-NEXT: paddd %xmm1, %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_pmovsxbd:
+; SLM: # BB#0:
+; SLM-NEXT: pmovsxbd (%rdi), %xmm1 # sched: [4:1.00]
+; SLM-NEXT: pmovsxbd %xmm0, %xmm0 # sched: [1:1.00]
+; SLM-NEXT: paddd %xmm0, %xmm1 # sched: [1:0.50]
+; SLM-NEXT: movdqa %xmm1, %xmm0 # sched: [1:0.50]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_pmovsxbd:
+; SANDY: # BB#0:
+; SANDY-NEXT: vpmovsxbd %xmm0, %xmm0 # sched: [1:0.50]
+; SANDY-NEXT: vpmovsxbd (%rdi), %xmm1 # sched: [5:0.50]
+; SANDY-NEXT: vpaddd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_pmovsxbd:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vpmovsxbd %xmm0, %xmm0 # sched: [1:1.00]
+; HASWELL-NEXT: vpmovsxbd (%rdi), %xmm1 # sched: [5:1.00]
+; HASWELL-NEXT: vpaddd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_pmovsxbd:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vpmovsxbd (%rdi), %xmm1 # sched: [6:1.00]
+; BTVER2-NEXT: vpmovsxbd %xmm0, %xmm0 # sched: [1:0.50]
+; BTVER2-NEXT: vpaddd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = shufflevector <16 x i8> %a0, <16 x i8> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %2 = sext <4 x i8> %1 to <4 x i32>
+ %3 = load <4 x i8>, <4 x i8>* %a1, align 1
+ %4 = sext <4 x i8> %3 to <4 x i32>
+ %5 = add <4 x i32> %2, %4
+ ret <4 x i32> %5
+}
+
+define <2 x i64> @test_pmovsxbq(<16 x i8> %a0, <2 x i8> *%a1) {
+; GENERIC-LABEL: test_pmovsxbq:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: pmovsxbq %xmm0, %xmm1
+; GENERIC-NEXT: pmovsxbq (%rdi), %xmm0
+; GENERIC-NEXT: paddq %xmm1, %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_pmovsxbq:
+; SLM: # BB#0:
+; SLM-NEXT: pmovsxbq (%rdi), %xmm1 # sched: [4:1.00]
+; SLM-NEXT: pmovsxbq %xmm0, %xmm0 # sched: [1:1.00]
+; SLM-NEXT: paddq %xmm0, %xmm1 # sched: [1:0.50]
+; SLM-NEXT: movdqa %xmm1, %xmm0 # sched: [1:0.50]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_pmovsxbq:
+; SANDY: # BB#0:
+; SANDY-NEXT: vpmovsxbq %xmm0, %xmm0 # sched: [1:0.50]
+; SANDY-NEXT: vpmovsxbq (%rdi), %xmm1 # sched: [5:0.50]
+; SANDY-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_pmovsxbq:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vpmovsxbq %xmm0, %xmm0 # sched: [1:1.00]
+; HASWELL-NEXT: vpmovsxbq (%rdi), %xmm1 # sched: [5:1.00]
+; HASWELL-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_pmovsxbq:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vpmovsxbq (%rdi), %xmm1 # sched: [6:1.00]
+; BTVER2-NEXT: vpmovsxbq %xmm0, %xmm0 # sched: [1:0.50]
+; BTVER2-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = shufflevector <16 x i8> %a0, <16 x i8> undef, <2 x i32> <i32 0, i32 1>
+ %2 = sext <2 x i8> %1 to <2 x i64>
+ %3 = load <2 x i8>, <2 x i8>* %a1, align 1
+ %4 = sext <2 x i8> %3 to <2 x i64>
+ %5 = add <2 x i64> %2, %4
+ ret <2 x i64> %5
+}
+
+define <2 x i64> @test_pmovsxdq(<4 x i32> %a0, <2 x i32> *%a1) {
+; GENERIC-LABEL: test_pmovsxdq:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: pmovsxdq %xmm0, %xmm1
+; GENERIC-NEXT: pmovsxdq (%rdi), %xmm0
+; GENERIC-NEXT: paddq %xmm1, %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_pmovsxdq:
+; SLM: # BB#0:
+; SLM-NEXT: pmovsxdq (%rdi), %xmm1 # sched: [4:1.00]
+; SLM-NEXT: pmovsxdq %xmm0, %xmm0 # sched: [1:1.00]
+; SLM-NEXT: paddq %xmm0, %xmm1 # sched: [1:0.50]
+; SLM-NEXT: movdqa %xmm1, %xmm0 # sched: [1:0.50]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_pmovsxdq:
+; SANDY: # BB#0:
+; SANDY-NEXT: vpmovsxdq %xmm0, %xmm0 # sched: [1:0.50]
+; SANDY-NEXT: vpmovsxdq (%rdi), %xmm1 # sched: [5:0.50]
+; SANDY-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_pmovsxdq:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vpmovsxdq %xmm0, %xmm0 # sched: [1:1.00]
+; HASWELL-NEXT: vpmovsxdq (%rdi), %xmm1 # sched: [5:1.00]
+; HASWELL-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_pmovsxdq:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vpmovsxdq (%rdi), %xmm1 # sched: [6:1.00]
+; BTVER2-NEXT: vpmovsxdq %xmm0, %xmm0 # sched: [1:0.50]
+; BTVER2-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = shufflevector <4 x i32> %a0, <4 x i32> undef, <2 x i32> <i32 0, i32 1>
+ %2 = sext <2 x i32> %1 to <2 x i64>
+ %3 = load <2 x i32>, <2 x i32>* %a1, align 1
+ %4 = sext <2 x i32> %3 to <2 x i64>
+ %5 = add <2 x i64> %2, %4
+ ret <2 x i64> %5
+}
+
+define <4 x i32> @test_pmovsxwd(<8 x i16> %a0, <4 x i16> *%a1) {
+; GENERIC-LABEL: test_pmovsxwd:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: pmovsxwd %xmm0, %xmm1
+; GENERIC-NEXT: pmovsxwd (%rdi), %xmm0
+; GENERIC-NEXT: paddd %xmm1, %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_pmovsxwd:
+; SLM: # BB#0:
+; SLM-NEXT: pmovsxwd (%rdi), %xmm1 # sched: [4:1.00]
+; SLM-NEXT: pmovsxwd %xmm0, %xmm0 # sched: [1:1.00]
+; SLM-NEXT: paddd %xmm0, %xmm1 # sched: [1:0.50]
+; SLM-NEXT: movdqa %xmm1, %xmm0 # sched: [1:0.50]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_pmovsxwd:
+; SANDY: # BB#0:
+; SANDY-NEXT: vpmovsxwd %xmm0, %xmm0 # sched: [1:0.50]
+; SANDY-NEXT: vpmovsxwd (%rdi), %xmm1 # sched: [5:0.50]
+; SANDY-NEXT: vpaddd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_pmovsxwd:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vpmovsxwd %xmm0, %xmm0 # sched: [1:1.00]
+; HASWELL-NEXT: vpmovsxwd (%rdi), %xmm1 # sched: [5:1.00]
+; HASWELL-NEXT: vpaddd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_pmovsxwd:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vpmovsxwd (%rdi), %xmm1 # sched: [6:1.00]
+; BTVER2-NEXT: vpmovsxwd %xmm0, %xmm0 # sched: [1:0.50]
+; BTVER2-NEXT: vpaddd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = shufflevector <8 x i16> %a0, <8 x i16> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %2 = sext <4 x i16> %1 to <4 x i32>
+ %3 = load <4 x i16>, <4 x i16>* %a1, align 1
+ %4 = sext <4 x i16> %3 to <4 x i32>
+ %5 = add <4 x i32> %2, %4
+ ret <4 x i32> %5
+}
+
+define <2 x i64> @test_pmovsxwq(<8 x i16> %a0, <2 x i16> *%a1) {
+; GENERIC-LABEL: test_pmovsxwq:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: pmovsxwq %xmm0, %xmm1
+; GENERIC-NEXT: pmovsxwq (%rdi), %xmm0
+; GENERIC-NEXT: paddq %xmm1, %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_pmovsxwq:
+; SLM: # BB#0:
+; SLM-NEXT: pmovsxwq (%rdi), %xmm1 # sched: [4:1.00]
+; SLM-NEXT: pmovsxwq %xmm0, %xmm0 # sched: [1:1.00]
+; SLM-NEXT: paddq %xmm0, %xmm1 # sched: [1:0.50]
+; SLM-NEXT: movdqa %xmm1, %xmm0 # sched: [1:0.50]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_pmovsxwq:
+; SANDY: # BB#0:
+; SANDY-NEXT: vpmovsxwq %xmm0, %xmm0 # sched: [1:0.50]
+; SANDY-NEXT: vpmovsxwq (%rdi), %xmm1 # sched: [5:0.50]
+; SANDY-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_pmovsxwq:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vpmovsxwq %xmm0, %xmm0 # sched: [1:1.00]
+; HASWELL-NEXT: vpmovsxwq (%rdi), %xmm1 # sched: [5:1.00]
+; HASWELL-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_pmovsxwq:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vpmovsxwq (%rdi), %xmm1 # sched: [6:1.00]
+; BTVER2-NEXT: vpmovsxwq %xmm0, %xmm0 # sched: [1:0.50]
+; BTVER2-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = shufflevector <8 x i16> %a0, <8 x i16> undef, <2 x i32> <i32 0, i32 1>
+ %2 = sext <2 x i16> %1 to <2 x i64>
+ %3 = load <2 x i16>, <2 x i16>* %a1, align 1
+ %4 = sext <2 x i16> %3 to <2 x i64>
+ %5 = add <2 x i64> %2, %4
+ ret <2 x i64> %5
+}
+
+define <8 x i16> @test_pmovzxbw(<16 x i8> %a0, <8 x i8> *%a1) {
+; GENERIC-LABEL: test_pmovzxbw:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: pmovzxbw {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; GENERIC-NEXT: pmovzxbw {{.*#+}} xmm0 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero
+; GENERIC-NEXT: paddw %xmm1, %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_pmovzxbw:
+; SLM: # BB#0:
+; SLM-NEXT: pmovzxbw {{.*#+}} xmm1 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero sched: [4:1.00]
+; SLM-NEXT: pmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero sched: [1:1.00]
+; SLM-NEXT: paddw %xmm0, %xmm1 # sched: [1:0.50]
+; SLM-NEXT: movdqa %xmm1, %xmm0 # sched: [1:0.50]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_pmovzxbw:
+; SANDY: # BB#0:
+; SANDY-NEXT: vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero sched: [1:0.50]
+; SANDY-NEXT: vpmovzxbw {{.*#+}} xmm1 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero sched: [5:0.50]
+; SANDY-NEXT: vpaddw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_pmovzxbw:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero sched: [1:1.00]
+; HASWELL-NEXT: vpmovzxbw {{.*#+}} xmm1 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero sched: [5:1.00]
+; HASWELL-NEXT: vpaddw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_pmovzxbw:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vpmovzxbw {{.*#+}} xmm1 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero sched: [6:1.00]
+; BTVER2-NEXT: vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero sched: [1:0.50]
+; BTVER2-NEXT: vpaddw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = shufflevector <16 x i8> %a0, <16 x i8> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+ %2 = zext <8 x i8> %1 to <8 x i16>
+ %3 = load <8 x i8>, <8 x i8>* %a1, align 1
+ %4 = zext <8 x i8> %3 to <8 x i16>
+ %5 = add <8 x i16> %2, %4
+ ret <8 x i16> %5
+}
+
+define <4 x i32> @test_pmovzxbd(<16 x i8> %a0, <4 x i8> *%a1) {
+; GENERIC-LABEL: test_pmovzxbd:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: pmovzxbd {{.*#+}} xmm1 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero
+; GENERIC-NEXT: pmovzxbd {{.*#+}} xmm0 = mem[0],zero,zero,zero,mem[1],zero,zero,zero,mem[2],zero,zero,zero,mem[3],zero,zero,zero
+; GENERIC-NEXT: paddd %xmm1, %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_pmovzxbd:
+; SLM: # BB#0:
+; SLM-NEXT: pmovzxbd {{.*#+}} xmm1 = mem[0],zero,zero,zero,mem[1],zero,zero,zero,mem[2],zero,zero,zero,mem[3],zero,zero,zero sched: [4:1.00]
+; SLM-NEXT: pmovzxbd {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero sched: [1:1.00]
+; SLM-NEXT: paddd %xmm0, %xmm1 # sched: [1:0.50]
+; SLM-NEXT: movdqa %xmm1, %xmm0 # sched: [1:0.50]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_pmovzxbd:
+; SANDY: # BB#0:
+; SANDY-NEXT: vpmovzxbd {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero sched: [1:0.50]
+; SANDY-NEXT: vpmovzxbd {{.*#+}} xmm1 = mem[0],zero,zero,zero,mem[1],zero,zero,zero,mem[2],zero,zero,zero,mem[3],zero,zero,zero sched: [5:0.50]
+; SANDY-NEXT: vpaddd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_pmovzxbd:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vpmovzxbd {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero sched: [1:1.00]
+; HASWELL-NEXT: vpmovzxbd {{.*#+}} xmm1 = mem[0],zero,zero,zero,mem[1],zero,zero,zero,mem[2],zero,zero,zero,mem[3],zero,zero,zero sched: [5:1.00]
+; HASWELL-NEXT: vpaddd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_pmovzxbd:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vpmovzxbd {{.*#+}} xmm1 = mem[0],zero,zero,zero,mem[1],zero,zero,zero,mem[2],zero,zero,zero,mem[3],zero,zero,zero sched: [6:1.00]
+; BTVER2-NEXT: vpmovzxbd {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero sched: [1:0.50]
+; BTVER2-NEXT: vpaddd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = shufflevector <16 x i8> %a0, <16 x i8> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %2 = zext <4 x i8> %1 to <4 x i32>
+ %3 = load <4 x i8>, <4 x i8>* %a1, align 1
+ %4 = zext <4 x i8> %3 to <4 x i32>
+ %5 = add <4 x i32> %2, %4
+ ret <4 x i32> %5
+}
+
+define <2 x i64> @test_pmovzxbq(<16 x i8> %a0, <2 x i8> *%a1) {
+; GENERIC-LABEL: test_pmovzxbq:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: pmovzxbq {{.*#+}} xmm1 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
+; GENERIC-NEXT: pmovzxbq {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero,mem[1],zero,zero,zero,zero,zero,zero,zero
+; GENERIC-NEXT: paddq %xmm1, %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_pmovzxbq:
+; SLM: # BB#0:
+; SLM-NEXT: pmovzxbq {{.*#+}} xmm1 = mem[0],zero,zero,zero,zero,zero,zero,zero,mem[1],zero,zero,zero,zero,zero,zero,zero sched: [4:1.00]
+; SLM-NEXT: pmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero sched: [1:1.00]
+; SLM-NEXT: paddq %xmm0, %xmm1 # sched: [1:0.50]
+; SLM-NEXT: movdqa %xmm1, %xmm0 # sched: [1:0.50]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_pmovzxbq:
+; SANDY: # BB#0:
+; SANDY-NEXT: vpmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero sched: [1:0.50]
+; SANDY-NEXT: vpmovzxbq {{.*#+}} xmm1 = mem[0],zero,zero,zero,zero,zero,zero,zero,mem[1],zero,zero,zero,zero,zero,zero,zero sched: [5:0.50]
+; SANDY-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_pmovzxbq:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vpmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero sched: [1:1.00]
+; HASWELL-NEXT: vpmovzxbq {{.*#+}} xmm1 = mem[0],zero,zero,zero,zero,zero,zero,zero,mem[1],zero,zero,zero,zero,zero,zero,zero sched: [5:1.00]
+; HASWELL-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_pmovzxbq:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vpmovzxbq {{.*#+}} xmm1 = mem[0],zero,zero,zero,zero,zero,zero,zero,mem[1],zero,zero,zero,zero,zero,zero,zero sched: [6:1.00]
+; BTVER2-NEXT: vpmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero sched: [1:0.50]
+; BTVER2-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = shufflevector <16 x i8> %a0, <16 x i8> undef, <2 x i32> <i32 0, i32 1>
+ %2 = zext <2 x i8> %1 to <2 x i64>
+ %3 = load <2 x i8>, <2 x i8>* %a1, align 1
+ %4 = zext <2 x i8> %3 to <2 x i64>
+ %5 = add <2 x i64> %2, %4
+ ret <2 x i64> %5
+}
+
+define <2 x i64> @test_pmovzxdq(<4 x i32> %a0, <2 x i32> *%a1) {
+; GENERIC-LABEL: test_pmovzxdq:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: pmovzxdq {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero
+; GENERIC-NEXT: pmovzxdq {{.*#+}} xmm0 = mem[0],zero,mem[1],zero
+; GENERIC-NEXT: paddq %xmm1, %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_pmovzxdq:
+; SLM: # BB#0:
+; SLM-NEXT: pmovzxdq {{.*#+}} xmm1 = mem[0],zero,mem[1],zero sched: [4:1.00]
+; SLM-NEXT: pmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero sched: [1:1.00]
+; SLM-NEXT: paddq %xmm0, %xmm1 # sched: [1:0.50]
+; SLM-NEXT: movdqa %xmm1, %xmm0 # sched: [1:0.50]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_pmovzxdq:
+; SANDY: # BB#0:
+; SANDY-NEXT: vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero sched: [1:0.50]
+; SANDY-NEXT: vpmovzxdq {{.*#+}} xmm1 = mem[0],zero,mem[1],zero sched: [5:0.50]
+; SANDY-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_pmovzxdq:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero sched: [1:1.00]
+; HASWELL-NEXT: vpmovzxdq {{.*#+}} xmm1 = mem[0],zero,mem[1],zero sched: [5:1.00]
+; HASWELL-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_pmovzxdq:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vpmovzxdq {{.*#+}} xmm1 = mem[0],zero,mem[1],zero sched: [6:1.00]
+; BTVER2-NEXT: vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero sched: [1:0.50]
+; BTVER2-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = shufflevector <4 x i32> %a0, <4 x i32> undef, <2 x i32> <i32 0, i32 1>
+ %2 = zext <2 x i32> %1 to <2 x i64>
+ %3 = load <2 x i32>, <2 x i32>* %a1, align 1
+ %4 = zext <2 x i32> %3 to <2 x i64>
+ %5 = add <2 x i64> %2, %4
+ ret <2 x i64> %5
+}
+
+define <4 x i32> @test_pmovzxwd(<8 x i16> %a0, <4 x i16> *%a1) {
+; GENERIC-LABEL: test_pmovzxwd:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: pmovzxwd {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
+; GENERIC-NEXT: pmovzxwd {{.*#+}} xmm0 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero
+; GENERIC-NEXT: paddd %xmm1, %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_pmovzxwd:
+; SLM: # BB#0:
+; SLM-NEXT: pmovzxwd {{.*#+}} xmm1 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero sched: [4:1.00]
+; SLM-NEXT: pmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero sched: [1:1.00]
+; SLM-NEXT: paddd %xmm0, %xmm1 # sched: [1:0.50]
+; SLM-NEXT: movdqa %xmm1, %xmm0 # sched: [1:0.50]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_pmovzxwd:
+; SANDY: # BB#0:
+; SANDY-NEXT: vpmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero sched: [1:0.50]
+; SANDY-NEXT: vpmovzxwd {{.*#+}} xmm1 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero sched: [5:0.50]
+; SANDY-NEXT: vpaddd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_pmovzxwd:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vpmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero sched: [1:1.00]
+; HASWELL-NEXT: vpmovzxwd {{.*#+}} xmm1 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero sched: [5:1.00]
+; HASWELL-NEXT: vpaddd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_pmovzxwd:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vpmovzxwd {{.*#+}} xmm1 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero sched: [6:1.00]
+; BTVER2-NEXT: vpmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero sched: [1:0.50]
+; BTVER2-NEXT: vpaddd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = shufflevector <8 x i16> %a0, <8 x i16> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %2 = zext <4 x i16> %1 to <4 x i32>
+ %3 = load <4 x i16>, <4 x i16>* %a1, align 1
+ %4 = zext <4 x i16> %3 to <4 x i32>
+ %5 = add <4 x i32> %2, %4
+ ret <4 x i32> %5
+}
+
+define <2 x i64> @test_pmovzxwq(<8 x i16> %a0, <2 x i16> *%a1) {
+; GENERIC-LABEL: test_pmovzxwq:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: pmovzxwq {{.*#+}} xmm1 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero
+; GENERIC-NEXT: pmovzxwq {{.*#+}} xmm0 = mem[0],zero,zero,zero,mem[1],zero,zero,zero
+; GENERIC-NEXT: paddq %xmm1, %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_pmovzxwq:
+; SLM: # BB#0:
+; SLM-NEXT: pmovzxwq {{.*#+}} xmm1 = mem[0],zero,zero,zero,mem[1],zero,zero,zero sched: [4:1.00]
+; SLM-NEXT: pmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero sched: [1:1.00]
+; SLM-NEXT: paddq %xmm0, %xmm1 # sched: [1:0.50]
+; SLM-NEXT: movdqa %xmm1, %xmm0 # sched: [1:0.50]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_pmovzxwq:
+; SANDY: # BB#0:
+; SANDY-NEXT: vpmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero sched: [1:0.50]
+; SANDY-NEXT: vpmovzxwq {{.*#+}} xmm1 = mem[0],zero,zero,zero,mem[1],zero,zero,zero sched: [5:0.50]
+; SANDY-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_pmovzxwq:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vpmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero sched: [1:1.00]
+; HASWELL-NEXT: vpmovzxwq {{.*#+}} xmm1 = mem[0],zero,zero,zero,mem[1],zero,zero,zero sched: [5:1.00]
+; HASWELL-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_pmovzxwq:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vpmovzxwq {{.*#+}} xmm1 = mem[0],zero,zero,zero,mem[1],zero,zero,zero sched: [6:1.00]
+; BTVER2-NEXT: vpmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero sched: [1:0.50]
+; BTVER2-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = shufflevector <8 x i16> %a0, <8 x i16> undef, <2 x i32> <i32 0, i32 1>
+ %2 = zext <2 x i16> %1 to <2 x i64>
+ %3 = load <2 x i16>, <2 x i16>* %a1, align 1
+ %4 = zext <2 x i16> %3 to <2 x i64>
+ %5 = add <2 x i64> %2, %4
+ ret <2 x i64> %5
+}
+
+define <2 x i64> @test_pmuldq(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) {
+; GENERIC-LABEL: test_pmuldq:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: pmuldq %xmm1, %xmm0
+; GENERIC-NEXT: pmuldq (%rdi), %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_pmuldq:
+; SLM: # BB#0:
+; SLM-NEXT: pmuldq %xmm1, %xmm0 # sched: [4:1.00]
+; SLM-NEXT: pmuldq (%rdi), %xmm0 # sched: [7:1.00]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_pmuldq:
+; SANDY: # BB#0:
+; SANDY-NEXT: vpmuldq %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; SANDY-NEXT: vpmuldq (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_pmuldq:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vpmuldq %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; HASWELL-NEXT: vpmuldq (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_pmuldq:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vpmuldq %xmm1, %xmm0, %xmm0 # sched: [2:1.00]
+; BTVER2-NEXT: vpmuldq (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = call <2 x i64> @llvm.x86.sse41.pmuldq(<4 x i32> %a0, <4 x i32> %a1)
+ %2 = bitcast <2 x i64> %1 to <4 x i32>
+ %3 = load <4 x i32>, <4 x i32> *%a2, align 16
+ %4 = call <2 x i64> @llvm.x86.sse41.pmuldq(<4 x i32> %2, <4 x i32> %3)
+ ret <2 x i64> %4
+}
+declare <2 x i64> @llvm.x86.sse41.pmuldq(<4 x i32>, <4 x i32>) nounwind readnone
+
+define <4 x i32> @test_pmulld(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) {
+; GENERIC-LABEL: test_pmulld:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: pmulld %xmm1, %xmm0
+; GENERIC-NEXT: pmulld (%rdi), %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_pmulld:
+; SLM: # BB#0:
+; SLM-NEXT: pmulld %xmm1, %xmm0 # sched: [4:1.00]
+; SLM-NEXT: pmulld (%rdi), %xmm0 # sched: [7:1.00]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_pmulld:
+; SANDY: # BB#0:
+; SANDY-NEXT: vpmulld %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
+; SANDY-NEXT: vpmulld (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_pmulld:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vpmulld %xmm1, %xmm0, %xmm0 # sched: [10:2.00]
+; HASWELL-NEXT: vpmulld (%rdi), %xmm0, %xmm0 # sched: [10:2.00]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_pmulld:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vpmulld %xmm1, %xmm0, %xmm0 # sched: [2:1.00]
+; BTVER2-NEXT: vpmulld (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = mul <4 x i32> %a0, %a1
+ %2 = load <4 x i32>, <4 x i32> *%a2, align 16
+ %3 = mul <4 x i32> %1, %2
+ ret <4 x i32> %3
+}
+
+define i32 @test_ptest(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> *%a2) {
+; GENERIC-LABEL: test_ptest:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: ptest %xmm1, %xmm0
+; GENERIC-NEXT: setb %al
+; GENERIC-NEXT: ptest (%rdi), %xmm0
+; GENERIC-NEXT: setb %cl
+; GENERIC-NEXT: andb %al, %cl
+; GENERIC-NEXT: movzbl %cl, %eax
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_ptest:
+; SLM: # BB#0:
+; SLM-NEXT: ptest %xmm1, %xmm0 # sched: [1:0.50]
+; SLM-NEXT: setb %al # sched: [1:0.50]
+; SLM-NEXT: ptest (%rdi), %xmm0 # sched: [4:1.00]
+; SLM-NEXT: setb %cl # sched: [1:0.50]
+; SLM-NEXT: andb %al, %cl # sched: [1:0.50]
+; SLM-NEXT: movzbl %cl, %eax # sched: [1:0.50]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_ptest:
+; SANDY: # BB#0:
+; SANDY-NEXT: vptest %xmm1, %xmm0 # sched: [1:0.33]
+; SANDY-NEXT: setb %al # sched: [1:0.33]
+; SANDY-NEXT: vptest (%rdi), %xmm0 # sched: [5:0.50]
+; SANDY-NEXT: setb %cl # sched: [1:0.33]
+; SANDY-NEXT: andb %al, %cl # sched: [1:0.33]
+; SANDY-NEXT: movzbl %cl, %eax # sched: [1:0.33]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_ptest:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vptest %xmm1, %xmm0 # sched: [2:1.00]
+; HASWELL-NEXT: setb %al # sched: [1:0.50]
+; HASWELL-NEXT: vptest (%rdi), %xmm0 # sched: [2:1.00]
+; HASWELL-NEXT: setb %cl # sched: [1:0.50]
+; HASWELL-NEXT: andb %al, %cl # sched: [1:0.25]
+; HASWELL-NEXT: movzbl %cl, %eax # sched: [1:0.25]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_ptest:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vptest %xmm1, %xmm0 # sched: [1:0.50]
+; BTVER2-NEXT: setb %al # sched: [1:0.50]
+; BTVER2-NEXT: vptest (%rdi), %xmm0 # sched: [6:1.00]
+; BTVER2-NEXT: setb %cl # sched: [1:0.50]
+; BTVER2-NEXT: andb %al, %cl # sched: [1:0.50]
+; BTVER2-NEXT: movzbl %cl, %eax # sched: [1:0.50]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = call i32 @llvm.x86.sse41.ptestc(<2 x i64> %a0, <2 x i64> %a1)
+ %2 = load <2 x i64>, <2 x i64> *%a2, align 16
+ %3 = call i32 @llvm.x86.sse41.ptestc(<2 x i64> %a0, <2 x i64> %2)
+ %4 = and i32 %1, %3
+ ret i32 %4
+}
+declare i32 @llvm.x86.sse41.ptestc(<2 x i64>, <2 x i64>) nounwind readnone
+
+define <2 x double> @test_roundpd(<2 x double> %a0, <2 x double> *%a1) {
+; GENERIC-LABEL: test_roundpd:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: roundpd $7, %xmm0, %xmm1
+; GENERIC-NEXT: roundpd $7, (%rdi), %xmm0
+; GENERIC-NEXT: addpd %xmm1, %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_roundpd:
+; SLM: # BB#0:
+; SLM-NEXT: roundpd $7, (%rdi), %xmm1 # sched: [6:1.00]
+; SLM-NEXT: roundpd $7, %xmm0, %xmm0 # sched: [3:1.00]
+; SLM-NEXT: addpd %xmm0, %xmm1 # sched: [3:1.00]
+; SLM-NEXT: movapd %xmm1, %xmm0 # sched: [1:1.00]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_roundpd:
+; SANDY: # BB#0:
+; SANDY-NEXT: vroundpd $7, %xmm0, %xmm0 # sched: [3:1.00]
+; SANDY-NEXT: vroundpd $7, (%rdi), %xmm1 # sched: [7:1.00]
+; SANDY-NEXT: vaddpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_roundpd:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vroundpd $7, %xmm0, %xmm0 # sched: [6:2.00]
+; HASWELL-NEXT: vroundpd $7, (%rdi), %xmm1 # sched: [10:2.00]
+; HASWELL-NEXT: vaddpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_roundpd:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vroundpd $7, (%rdi), %xmm1 # sched: [8:1.00]
+; BTVER2-NEXT: vroundpd $7, %xmm0, %xmm0 # sched: [3:1.00]
+; BTVER2-NEXT: vaddpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = call <2 x double> @llvm.x86.sse41.round.pd(<2 x double> %a0, i32 7)
+ %2 = load <2 x double>, <2 x double> *%a1, align 16
+ %3 = call <2 x double> @llvm.x86.sse41.round.pd(<2 x double> %2, i32 7)
+ %4 = fadd <2 x double> %1, %3
+ ret <2 x double> %4
+}
+declare <2 x double> @llvm.x86.sse41.round.pd(<2 x double>, i32) nounwind readnone
+
+define <4 x float> @test_roundps(<4 x float> %a0, <4 x float> *%a1) {
+; GENERIC-LABEL: test_roundps:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: roundps $7, %xmm0, %xmm1
+; GENERIC-NEXT: roundps $7, (%rdi), %xmm0
+; GENERIC-NEXT: addps %xmm1, %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_roundps:
+; SLM: # BB#0:
+; SLM-NEXT: roundps $7, (%rdi), %xmm1 # sched: [6:1.00]
+; SLM-NEXT: roundps $7, %xmm0, %xmm0 # sched: [3:1.00]
+; SLM-NEXT: addps %xmm0, %xmm1 # sched: [3:1.00]
+; SLM-NEXT: movaps %xmm1, %xmm0 # sched: [1:1.00]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_roundps:
+; SANDY: # BB#0:
+; SANDY-NEXT: vroundps $7, %xmm0, %xmm0 # sched: [3:1.00]
+; SANDY-NEXT: vroundps $7, (%rdi), %xmm1 # sched: [7:1.00]
+; SANDY-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_roundps:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vroundps $7, %xmm0, %xmm0 # sched: [6:2.00]
+; HASWELL-NEXT: vroundps $7, (%rdi), %xmm1 # sched: [10:2.00]
+; HASWELL-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_roundps:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vroundps $7, (%rdi), %xmm1 # sched: [8:1.00]
+; BTVER2-NEXT: vroundps $7, %xmm0, %xmm0 # sched: [3:1.00]
+; BTVER2-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = call <4 x float> @llvm.x86.sse41.round.ps(<4 x float> %a0, i32 7)
+ %2 = load <4 x float>, <4 x float> *%a1, align 16
+ %3 = call <4 x float> @llvm.x86.sse41.round.ps(<4 x float> %2, i32 7)
+ %4 = fadd <4 x float> %1, %3
+ ret <4 x float> %4
+}
+declare <4 x float> @llvm.x86.sse41.round.ps(<4 x float>, i32) nounwind readnone
+
+define <2 x double> @test_roundsd(<2 x double> %a0, <2 x double> %a1, <2 x double> *%a2) {
+; GENERIC-LABEL: test_roundsd:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: movaps %xmm0, %xmm2
+; GENERIC-NEXT: roundsd $7, %xmm1, %xmm2
+; GENERIC-NEXT: roundsd $7, (%rdi), %xmm0
+; GENERIC-NEXT: addpd %xmm2, %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_roundsd:
+; SLM: # BB#0:
+; SLM-NEXT: movaps %xmm0, %xmm2 # sched: [1:1.00]
+; SLM-NEXT: roundsd $7, (%rdi), %xmm0 # sched: [6:1.00]
+; SLM-NEXT: roundsd $7, %xmm1, %xmm2 # sched: [3:1.00]
+; SLM-NEXT: addpd %xmm2, %xmm0 # sched: [3:1.00]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_roundsd:
+; SANDY: # BB#0:
+; SANDY-NEXT: vroundsd $7, %xmm1, %xmm0, %xmm1 # sched: [3:1.00]
+; SANDY-NEXT: vroundsd $7, (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
+; SANDY-NEXT: vaddpd %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_roundsd:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vroundsd $7, %xmm1, %xmm0, %xmm1 # sched: [6:2.00]
+; HASWELL-NEXT: vroundsd $7, (%rdi), %xmm0, %xmm0 # sched: [10:2.00]
+; HASWELL-NEXT: vaddpd %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_roundsd:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vroundsd $7, %xmm1, %xmm0, %xmm1 # sched: [3:1.00]
+; BTVER2-NEXT: vroundsd $7, (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
+; BTVER2-NEXT: vaddpd %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = call <2 x double> @llvm.x86.sse41.round.sd(<2 x double> %a0, <2 x double> %a1, i32 7)
+ %2 = load <2 x double>, <2 x double>* %a2, align 16
+ %3 = call <2 x double> @llvm.x86.sse41.round.sd(<2 x double> %a0, <2 x double> %2, i32 7)
+ %4 = fadd <2 x double> %1, %3
+ ret <2 x double> %4
+}
+declare <2 x double> @llvm.x86.sse41.round.sd(<2 x double>, <2 x double>, i32) nounwind readnone
+
+define <4 x float> @test_roundss(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a2) {
+; GENERIC-LABEL: test_roundss:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: movaps %xmm0, %xmm2
+; GENERIC-NEXT: roundss $7, %xmm1, %xmm2
+; GENERIC-NEXT: roundss $7, (%rdi), %xmm0
+; GENERIC-NEXT: addps %xmm2, %xmm0
+; GENERIC-NEXT: retq
+;
+; SLM-LABEL: test_roundss:
+; SLM: # BB#0:
+; SLM-NEXT: movaps %xmm0, %xmm2 # sched: [1:1.00]
+; SLM-NEXT: roundss $7, (%rdi), %xmm0 # sched: [6:1.00]
+; SLM-NEXT: roundss $7, %xmm1, %xmm2 # sched: [3:1.00]
+; SLM-NEXT: addps %xmm2, %xmm0 # sched: [3:1.00]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_roundss:
+; SANDY: # BB#0:
+; SANDY-NEXT: vroundss $7, %xmm1, %xmm0, %xmm1 # sched: [3:1.00]
+; SANDY-NEXT: vroundss $7, (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
+; SANDY-NEXT: vaddps %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
+; SANDY-NEXT: retq # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_roundss:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: vroundss $7, %xmm1, %xmm0, %xmm1 # sched: [6:2.00]
+; HASWELL-NEXT: vroundss $7, (%rdi), %xmm0, %xmm0 # sched: [10:2.00]
+; HASWELL-NEXT: vaddps %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
+; HASWELL-NEXT: retq # sched: [1:1.00]
+;
+; BTVER2-LABEL: test_roundss:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: vroundss $7, %xmm1, %xmm0, %xmm1 # sched: [3:1.00]
+; BTVER2-NEXT: vroundss $7, (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
+; BTVER2-NEXT: vaddps %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+ %1 = call <4 x float> @llvm.x86.sse41.round.ss(<4 x float> %a0, <4 x float> %a1, i32 7)
+ %2 = load <4 x float>, <4 x float> *%a2, align 16
+ %3 = call <4 x float> @llvm.x86.sse41.round.ss(<4 x float> %a0, <4 x float> %2, i32 7)
+ %4 = fadd <4 x float> %1, %3
+ ret <4 x float> %4
+}
+declare <4 x float> @llvm.x86.sse41.round.ss(<4 x float>, <4 x float>, i32) nounwind readnone
More information about the llvm-commits
mailing list