RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-23 19:23:23 +01:00
Code Issues Projects Releases Wiki Activity

This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target.

Browse Source 
The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information.

Please note that the patch extensively affects the X86 MC instr scheduling for SNB.

Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX.

The updated and extended information about each instruction includes the following details:
•static latency of the instruction
•number of uOps from which the instruction consists of
•all ports used by the instruction's' uOPs

For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5:

def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> {
let Latency = 9;
let NumMicroOps = 6;
let ResourceCycles = [1,2,2,1];

}
def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>;
def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>;
def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>;
def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>;

Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script.

Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb

Differential Revision:  https://reviews.llvm.org/D35019#inline-304691

llvm-svn: 307529
This commit is contained in:
Gadi Haber 2017-07-10 09:53:16 +00:00
parent 4b56bf7dea
commit fb7830bb57
13 changed files with 3338 additions and 926 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2472
lib/Target/X86/X86SchedSandyBridge.td
View File

File diff suppressed because it is too large Load Diff

408
test/CodeGen/X86/avx-schedule.ll
View File

@ -10,8 +10,8 @@ define <4 x double> @test_addpd(<4 x double> %a0, <4 x double> %a1, <4 x double>
; SANDY-LABEL: test_addpd:
; SANDY: # BB#0:
; SANDY-NEXT: vaddpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: vaddpd (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vaddpd (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_addpd:
; HASWELL: # BB#0:
@ -40,8 +40,8 @@ define <8 x float> @test_addps(<8 x float> %a0, <8 x float> %a1, <8 x float> *%a
; SANDY-LABEL: test_addps:
; SANDY: # BB#0:
; SANDY-NEXT: vaddps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: vaddps (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vaddps (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_addps:
; HASWELL: # BB#0:
@ -70,8 +70,8 @@ define <4 x double> @test_addsubpd(<4 x double> %a0, <4 x double> %a1, <4 x doub
; SANDY-LABEL: test_addsubpd:
; SANDY: # BB#0:
; SANDY-NEXT: vaddsubpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: vaddsubpd (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vaddsubpd (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_addsubpd:
; HASWELL: # BB#0:
@ -101,8 +101,8 @@ define <8 x float> @test_addsubps(<8 x float> %a0, <8 x float> %a1, <8 x float>
; SANDY-LABEL: test_addsubps:
; SANDY: # BB#0:
; SANDY-NEXT: vaddsubps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: vaddsubps (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vaddsubps (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_addsubps:
; HASWELL: # BB#0:
@ -131,10 +131,10 @@ declare <8 x float> @llvm.x86.avx.addsub.ps.256(<8 x float>, <8 x float>) nounwi
define <4 x double> @test_andnotpd(<4 x double> %a0, <4 x double> %a1, <4 x double> *%a2) {
; SANDY-LABEL: test_andnotpd:
; SANDY: # BB#0:
; SANDY-NEXT: vandnpd %ymm1, %ymm0, %ymm0 # sched: [1:0.33]
; SANDY-NEXT: vandnpd (%rdi), %ymm0, %ymm0 # sched: [5:0.50]
; SANDY-NEXT: vandnpd %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
; SANDY-NEXT: vandnpd (%rdi), %ymm0, %ymm0 # sched: [8:1.00]
; SANDY-NEXT: vaddpd %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_andnotpd:
; HASWELL: # BB#0:
@ -172,10 +172,10 @@ define <4 x double> @test_andnotpd(<4 x double> %a0, <4 x double> %a1, <4 x doub
define <8 x float> @test_andnotps(<8 x float> %a0, <8 x float> %a1, <8 x float> *%a2) {
; SANDY-LABEL: test_andnotps:
; SANDY: # BB#0:
; SANDY-NEXT: vandnps %ymm1, %ymm0, %ymm0 # sched: [1:0.33]
; SANDY-NEXT: vandnps (%rdi), %ymm0, %ymm0 # sched: [5:0.50]
; SANDY-NEXT: vandnps %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
; SANDY-NEXT: vandnps (%rdi), %ymm0, %ymm0 # sched: [8:1.00]
; SANDY-NEXT: vaddps %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_andnotps:
; HASWELL: # BB#0:
@ -213,10 +213,10 @@ define <8 x float> @test_andnotps(<8 x float> %a0, <8 x float> %a1, <8 x float>
define <4 x double> @test_andpd(<4 x double> %a0, <4 x double> %a1, <4 x double> *%a2) {
; SANDY-LABEL: test_andpd:
; SANDY: # BB#0:
; SANDY-NEXT: vandpd %ymm1, %ymm0, %ymm0 # sched: [1:0.33]
; SANDY-NEXT: vandpd (%rdi), %ymm0, %ymm0 # sched: [5:0.50]
; SANDY-NEXT: vandpd %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
; SANDY-NEXT: vandpd (%rdi), %ymm0, %ymm0 # sched: [5:1.00]
; SANDY-NEXT: vaddpd %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_andpd:
; HASWELL: # BB#0:
@ -252,10 +252,10 @@ define <4 x double> @test_andpd(<4 x double> %a0, <4 x double> %a1, <4 x double>
define <8 x float> @test_andps(<8 x float> %a0, <8 x float> %a1, <8 x float> *%a2) {
; SANDY-LABEL: test_andps:
; SANDY: # BB#0:
; SANDY-NEXT: vandps %ymm1, %ymm0, %ymm0 # sched: [1:0.33]
; SANDY-NEXT: vandps (%rdi), %ymm0, %ymm0 # sched: [5:0.50]
; SANDY-NEXT: vandps %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
; SANDY-NEXT: vandps (%rdi), %ymm0, %ymm0 # sched: [5:1.00]
; SANDY-NEXT: vaddps %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_andps:
; HASWELL: # BB#0:
@ -291,10 +291,10 @@ define <8 x float> @test_andps(<8 x float> %a0, <8 x float> %a1, <8 x float> *%a
define <4 x double> @test_blendpd(<4 x double> %a0, <4 x double> %a1, <4 x double> *%a2) {
; SANDY-LABEL: test_blendpd:
; SANDY: # BB#0:
; SANDY-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0],ymm1[1,2],ymm0[3] sched: [1:0.50]
; SANDY-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0],ymm1[1,2],ymm0[3] sched: [1:1.00]
; SANDY-NEXT: vaddpd %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0],mem[1,2],ymm0[3] sched: [5:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0],mem[1,2],ymm0[3] sched: [8:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_blendpd:
; HASWELL: # BB#0:
@ -326,9 +326,9 @@ define <4 x double> @test_blendpd(<4 x double> %a0, <4 x double> %a1, <4 x doubl
define <8 x float> @test_blendps(<8 x float> %a0, <8 x float> %a1, <8 x float> *%a2) {
; SANDY-LABEL: test_blendps:
; SANDY: # BB#0:
; SANDY-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2],ymm0[3,4,5,6,7] sched: [1:0.50]
; SANDY-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0,1],mem[2],ymm0[3],mem[4,5,6],ymm0[7] sched: [5:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2],ymm0[3,4,5,6,7] sched: [1:1.00]
; SANDY-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0,1],mem[2],ymm0[3],mem[4,5,6],ymm0[7] sched: [8:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_blendps:
; HASWELL: # BB#0:
@ -356,9 +356,9 @@ define <8 x float> @test_blendps(<8 x float> %a0, <8 x float> %a1, <8 x float> *
define <4 x double> @test_blendvpd(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, <4 x double> *%a3) {
; SANDY-LABEL: test_blendvpd:
; SANDY: # BB#0:
; SANDY-NEXT: vblendvpd %ymm2, %ymm1, %ymm0, %ymm0 # sched: [2:1.00]
; SANDY-NEXT: vblendvpd %ymm2, (%rdi), %ymm0, %ymm0 # sched: [6:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vblendvpd %ymm2, %ymm1, %ymm0, %ymm0 # sched: [2:2.00]
; SANDY-NEXT: vblendvpd %ymm2, (%rdi), %ymm0, %ymm0 # sched: [9:2.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_blendvpd:
; HASWELL: # BB#0:
@ -387,9 +387,9 @@ declare <4 x double> @llvm.x86.avx.blendv.pd.256(<4 x double>, <4 x double>, <4
define <8 x float> @test_blendvps(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2, <8 x float> *%a3) {
; SANDY-LABEL: test_blendvps:
; SANDY: # BB#0:
; SANDY-NEXT: vblendvps %ymm2, %ymm1, %ymm0, %ymm0 # sched: [2:1.00]
; SANDY-NEXT: vblendvps %ymm2, (%rdi), %ymm0, %ymm0 # sched: [6:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vblendvps %ymm2, %ymm1, %ymm0, %ymm0 # sched: [2:2.00]
; SANDY-NEXT: vblendvps %ymm2, (%rdi), %ymm0, %ymm0 # sched: [9:2.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_blendvps:
; HASWELL: # BB#0:
@ -418,8 +418,8 @@ declare <8 x float> @llvm.x86.avx.blendv.ps.256(<8 x float>, <8 x float>, <8 x f
define <8 x float> @test_broadcastf128(<4 x float> *%a0) {
; SANDY-LABEL: test_broadcastf128:
; SANDY: # BB#0:
; SANDY-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] sched: [5:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] sched: [3:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_broadcastf128:
; HASWELL: # BB#0:
@ -443,8 +443,8 @@ define <8 x float> @test_broadcastf128(<4 x float> *%a0) {
define <4 x double> @test_broadcastsd_ymm(double *%a0) {
; SANDY-LABEL: test_broadcastsd_ymm:
; SANDY: # BB#0:
; SANDY-NEXT: vbroadcastsd (%rdi), %ymm0 # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vbroadcastsd (%rdi), %ymm0 # sched: [7:0.50]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_broadcastsd_ymm:
; HASWELL: # BB#0:
@ -469,8 +469,8 @@ define <4 x double> @test_broadcastsd_ymm(double *%a0) {
define <4 x float> @test_broadcastss(float *%a0) {
; SANDY-LABEL: test_broadcastss:
; SANDY: # BB#0:
; SANDY-NEXT: vbroadcastss (%rdi), %xmm0 # sched: [4:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vbroadcastss (%rdi), %xmm0 # sched: [6:0.50]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_broadcastss:
; HASWELL: # BB#0:
@ -496,7 +496,7 @@ define <8 x float> @test_broadcastss_ymm(float *%a0) {
; SANDY-LABEL: test_broadcastss_ymm:
; SANDY: # BB#0:
; SANDY-NEXT: vbroadcastss (%rdi), %ymm0 # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_broadcastss_ymm:
; HASWELL: # BB#0:
@ -522,9 +522,9 @@ define <4 x double> @test_cmppd(<4 x double> %a0, <4 x double> %a1, <4 x double>
; SANDY-LABEL: test_cmppd:
; SANDY: # BB#0:
; SANDY-NEXT: vcmpeqpd %ymm1, %ymm0, %ymm1 # sched: [3:1.00]
; SANDY-NEXT: vcmpeqpd (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
; SANDY-NEXT: vorpd %ymm0, %ymm1, %ymm0 # sched: [1:0.33]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vcmpeqpd (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
; SANDY-NEXT: vorpd %ymm0, %ymm1, %ymm0 # sched: [1:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_cmppd:
; HASWELL: # BB#0:
@ -560,9 +560,9 @@ define <8 x float> @test_cmpps(<8 x float> %a0, <8 x float> %a1, <8 x float> *%a
; SANDY-LABEL: test_cmpps:
; SANDY: # BB#0:
; SANDY-NEXT: vcmpeqps %ymm1, %ymm0, %ymm1 # sched: [3:1.00]
; SANDY-NEXT: vcmpeqps (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
; SANDY-NEXT: vorps %ymm0, %ymm1, %ymm0 # sched: [1:0.33]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vcmpeqps (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
; SANDY-NEXT: vorps %ymm0, %ymm1, %ymm0 # sched: [1:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_cmpps:
; HASWELL: # BB#0:
@ -598,9 +598,9 @@ define <4 x double> @test_cvtdq2pd(<4 x i32> %a0, <4 x i32> *%a1) {
; SANDY-LABEL: test_cvtdq2pd:
; SANDY: # BB#0:
; SANDY-NEXT: vcvtdq2pd %xmm0, %ymm0 # sched: [4:1.00]
; SANDY-NEXT: vcvtdq2pd (%rdi), %ymm1 # sched: [8:1.00]
; SANDY-NEXT: vcvtdq2pd (%rdi), %ymm1 # sched: [10:1.00]
; SANDY-NEXT: vaddpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_cvtdq2pd:
; HASWELL: # BB#0:
@ -632,12 +632,12 @@ define <4 x double> @test_cvtdq2pd(<4 x i32> %a0, <4 x i32> *%a1) {
define <8 x float> @test_cvtdq2ps(<8 x i32> %a0, <8 x i32> *%a1) {
; SANDY-LABEL: test_cvtdq2ps:
; SANDY: # BB#0:
; SANDY-NEXT: vcvtdq2ps %ymm0, %ymm0 # sched: [4:1.00]
; SANDY-NEXT: vmovaps (%rdi), %xmm1 # sched: [4:0.50]
; SANDY-NEXT: vinsertf128 $1, 16(%rdi), %ymm1, %ymm1 # sched: [5:1.00]
; SANDY-NEXT: vcvtdq2ps %ymm1, %ymm1 # sched: [4:1.00]
; SANDY-NEXT: vcvtdq2ps %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: vmovaps (%rdi), %xmm1 # sched: [6:0.50]
; SANDY-NEXT: vinsertf128 $1, 16(%rdi), %ymm1, %ymm1 # sched: [7:1.00]
; SANDY-NEXT: vcvtdq2ps %ymm1, %ymm1 # sched: [3:1.00]
; SANDY-NEXT: vaddps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_cvtdq2ps:
; HASWELL: # BB#0:
@ -669,10 +669,10 @@ define <8 x float> @test_cvtdq2ps(<8 x i32> %a0, <8 x i32> *%a1) {
define <8 x i32> @test_cvtpd2dq(<4 x double> %a0, <4 x double> *%a1) {
; SANDY-LABEL: test_cvtpd2dq:
; SANDY: # BB#0:
; SANDY-NEXT: vcvttpd2dq %ymm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vcvttpd2dqy (%rdi), %xmm1 # sched: [7:1.00]
; SANDY-NEXT: vcvttpd2dq %ymm0, %xmm0 # sched: [4:1.00]
; SANDY-NEXT: vcvttpd2dqy (%rdi), %xmm1 # sched: [11:1.00]
; SANDY-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 # sched: [1:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_cvtpd2dq:
; HASWELL: # BB#0:
@ -704,10 +704,10 @@ define <8 x i32> @test_cvtpd2dq(<4 x double> %a0, <4 x double> *%a1) {
define <8 x float> @test_cvtpd2ps(<4 x double> %a0, <4 x double> *%a1) {
; SANDY-LABEL: test_cvtpd2ps:
; SANDY: # BB#0:
; SANDY-NEXT: vcvtpd2ps %ymm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vcvtpd2psy (%rdi), %xmm1 # sched: [7:1.00]
; SANDY-NEXT: vcvtpd2ps %ymm0, %xmm0 # sched: [4:1.00]
; SANDY-NEXT: vcvtpd2psy (%rdi), %xmm1 # sched: [11:1.00]
; SANDY-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 # sched: [1:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_cvtpd2ps:
; HASWELL: # BB#0:
@ -741,8 +741,8 @@ define <8 x i32> @test_cvtps2dq(<8 x float> %a0, <8 x float> *%a1) {
; SANDY: # BB#0:
; SANDY-NEXT: vcvttps2dq %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: vcvttps2dq (%rdi), %ymm1 # sched: [7:1.00]
; SANDY-NEXT: vorps %ymm1, %ymm0, %ymm0 # sched: [1:0.33]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vorps %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_cvtps2dq:
; HASWELL: # BB#0:
@ -774,9 +774,9 @@ define <8 x i32> @test_cvtps2dq(<8 x float> %a0, <8 x float> *%a1) {
define <4 x double> @test_divpd(<4 x double> %a0, <4 x double> %a1, <4 x double> *%a2) {
; SANDY-LABEL: test_divpd:
; SANDY: # BB#0:
; SANDY-NEXT: vdivpd %ymm1, %ymm0, %ymm0 # sched: [12:1.00]
; SANDY-NEXT: vdivpd (%rdi), %ymm0, %ymm0 # sched: [16:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vdivpd %ymm1, %ymm0, %ymm0 # sched: [45:3.00]
; SANDY-NEXT: vdivpd (%rdi), %ymm0, %ymm0 # sched: [52:3.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_divpd:
; HASWELL: # BB#0:
@ -804,9 +804,9 @@ define <4 x double> @test_divpd(<4 x double> %a0, <4 x double> %a1, <4 x double>
define <8 x float> @test_divps(<8 x float> %a0, <8 x float> %a1, <8 x float> *%a2) {
; SANDY-LABEL: test_divps:
; SANDY: # BB#0:
; SANDY-NEXT: vdivps %ymm1, %ymm0, %ymm0 # sched: [12:1.00]
; SANDY-NEXT: vdivps (%rdi), %ymm0, %ymm0 # sched: [16:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vdivps %ymm1, %ymm0, %ymm0 # sched: [29:3.00]
; SANDY-NEXT: vdivps (%rdi), %ymm0, %ymm0 # sched: [36:3.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_divps:
; HASWELL: # BB#0:
@ -834,9 +834,9 @@ define <8 x float> @test_divps(<8 x float> %a0, <8 x float> %a1, <8 x float> *%a
define <8 x float> @test_dpps(<8 x float> %a0, <8 x float> %a1, <8 x float> *%a2) {
; SANDY-LABEL: test_dpps:
; SANDY: # BB#0:
; SANDY-NEXT: vdpps $7, %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: vdpps $7, %ymm1, %ymm0, %ymm0 # sched: [12:2.00]
; SANDY-NEXT: vdpps $7, (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_dpps:
; HASWELL: # BB#0:
@ -866,9 +866,9 @@ define <4 x float> @test_extractf128(<8 x float> %a0, <8 x float> %a1, <4 x floa
; SANDY-LABEL: test_extractf128:
; SANDY: # BB#0:
; SANDY-NEXT: vextractf128 $1, %ymm0, %xmm0 # sched: [1:1.00]
; SANDY-NEXT: vextractf128 $1, %ymm1, (%rdi) # sched: [1:1.00]
; SANDY-NEXT: vextractf128 $1, %ymm1, (%rdi) # sched: [5:1.00]
; SANDY-NEXT: vzeroupper # sched: [?:0.000000e+00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_extractf128:
; HASWELL: # BB#0:
@ -900,7 +900,7 @@ define <4 x double> @test_haddpd(<4 x double> %a0, <4 x double> %a1, <4 x double
; SANDY: # BB#0:
; SANDY-NEXT: vhaddpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: vhaddpd (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_haddpd:
; HASWELL: # BB#0:
@ -929,9 +929,9 @@ declare <4 x double> @llvm.x86.avx.hadd.pd.256(<4 x double>, <4 x double>) nounw
define <8 x float> @test_haddps(<8 x float> %a0, <8 x float> %a1, <8 x float> *%a2) {
; SANDY-LABEL: test_haddps:
; SANDY: # BB#0:
; SANDY-NEXT: vhaddps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: vhaddps (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vhaddps %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
; SANDY-NEXT: vhaddps (%rdi), %ymm0, %ymm0 # sched: [12:2.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_haddps:
; HASWELL: # BB#0:
@ -960,9 +960,9 @@ declare <8 x float> @llvm.x86.avx.hadd.ps.256(<8 x float>, <8 x float>) nounwind
define <4 x double> @test_hsubpd(<4 x double> %a0, <4 x double> %a1, <4 x double> *%a2) {
; SANDY-LABEL: test_hsubpd:
; SANDY: # BB#0:
; SANDY-NEXT: vhsubpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: vhsubpd (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vhsubpd %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
; SANDY-NEXT: vhsubpd (%rdi), %ymm0, %ymm0 # sched: [12:2.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_hsubpd:
; HASWELL: # BB#0:
@ -991,9 +991,9 @@ declare <4 x double> @llvm.x86.avx.hsub.pd.256(<4 x double>, <4 x double>) nounw
define <8 x float> @test_hsubps(<8 x float> %a0, <8 x float> %a1, <8 x float> *%a2) {
; SANDY-LABEL: test_hsubps:
; SANDY: # BB#0:
; SANDY-NEXT: vhsubps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: vhsubps (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vhsubps %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
; SANDY-NEXT: vhsubps (%rdi), %ymm0, %ymm0 # sched: [12:2.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_hsubps:
; HASWELL: # BB#0:
@ -1023,9 +1023,9 @@ define <8 x float> @test_insertf128(<8 x float> %a0, <4 x float> %a1, <4 x float
; SANDY-LABEL: test_insertf128:
; SANDY: # BB#0:
; SANDY-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm1 # sched: [1:1.00]
; SANDY-NEXT: vinsertf128 $1, (%rdi), %ymm0, %ymm0 # sched: [5:1.00]
; SANDY-NEXT: vinsertf128 $1, (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
; SANDY-NEXT: vaddps %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_insertf128:
; HASWELL: # BB#0:
@ -1059,8 +1059,8 @@ define <8 x float> @test_insertf128(<8 x float> %a0, <4 x float> %a1, <4 x float
define <32 x i8> @test_lddqu(i8* %a0) {
; SANDY-LABEL: test_lddqu:
; SANDY: # BB#0:
; SANDY-NEXT: vlddqu (%rdi), %ymm0 # sched: [4:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vlddqu (%rdi), %ymm0 # sched: [6:0.50]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_lddqu:
; HASWELL: # BB#0:
@ -1084,10 +1084,10 @@ declare <32 x i8> @llvm.x86.avx.ldu.dq.256(i8*) nounwind readonly
define <2 x double> @test_maskmovpd(i8* %a0, <2 x i64> %a1, <2 x double> %a2) {
; SANDY-LABEL: test_maskmovpd:
; SANDY: # BB#0:
; SANDY-NEXT: vmaskmovpd (%rdi), %xmm0, %xmm2 # sched: [?:0.000000e+00]
; SANDY-NEXT: vmaskmovpd %xmm1, %xmm0, (%rdi) # sched: [?:0.000000e+00]
; SANDY-NEXT: vmaskmovpd (%rdi), %xmm0, %xmm2 # sched: [8:2.00]
; SANDY-NEXT: vmaskmovpd %xmm1, %xmm0, (%rdi) # sched: [5:1.00]
; SANDY-NEXT: vmovapd %xmm2, %xmm0 # sched: [1:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_maskmovpd:
; HASWELL: # BB#0:
@ -1119,10 +1119,10 @@ declare void @llvm.x86.avx.maskstore.pd(i8*, <2 x i64>, <2 x double>) nounwind
define <4 x double> @test_maskmovpd_ymm(i8* %a0, <4 x i64> %a1, <4 x double> %a2) {
; SANDY-LABEL: test_maskmovpd_ymm:
; SANDY: # BB#0:
; SANDY-NEXT: vmaskmovpd (%rdi), %ymm0, %ymm2 # sched: [?:0.000000e+00]
; SANDY-NEXT: vmaskmovpd (%rdi), %ymm0, %ymm2 # sched: [5:1.00]
; SANDY-NEXT: vmaskmovpd %ymm1, %ymm0, (%rdi) # sched: [?:0.000000e+00]
; SANDY-NEXT: vmovapd %ymm2, %ymm0 # sched: [1:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_maskmovpd_ymm:
; HASWELL: # BB#0:
@ -1154,10 +1154,10 @@ declare void @llvm.x86.avx.maskstore.pd.256(i8*, <4 x i64>, <4 x double>) nounwi
define <4 x float> @test_maskmovps(i8* %a0, <4 x i32> %a1, <4 x float> %a2) {
; SANDY-LABEL: test_maskmovps:
; SANDY: # BB#0:
; SANDY-NEXT: vmaskmovps (%rdi), %xmm0, %xmm2 # sched: [?:0.000000e+00]
; SANDY-NEXT: vmaskmovps %xmm1, %xmm0, (%rdi) # sched: [?:0.000000e+00]
; SANDY-NEXT: vmaskmovps (%rdi), %xmm0, %xmm2 # sched: [8:2.00]
; SANDY-NEXT: vmaskmovps %xmm1, %xmm0, (%rdi) # sched: [5:1.00]
; SANDY-NEXT: vmovaps %xmm2, %xmm0 # sched: [1:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_maskmovps:
; HASWELL: # BB#0:
@ -1189,10 +1189,10 @@ declare void @llvm.x86.avx.maskstore.ps(i8*, <4 x i32>, <4 x float>) nounwind
define <8 x float> @test_maskmovps_ymm(i8* %a0, <8 x i32> %a1, <8 x float> %a2) {
; SANDY-LABEL: test_maskmovps_ymm:
; SANDY: # BB#0:
; SANDY-NEXT: vmaskmovps (%rdi), %ymm0, %ymm2 # sched: [?:0.000000e+00]
; SANDY-NEXT: vmaskmovps (%rdi), %ymm0, %ymm2 # sched: [1:0.50]
; SANDY-NEXT: vmaskmovps %ymm1, %ymm0, (%rdi) # sched: [?:0.000000e+00]
; SANDY-NEXT: vmovaps %ymm2, %ymm0 # sched: [1:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_maskmovps_ymm:
; HASWELL: # BB#0:
@ -1225,8 +1225,8 @@ define <4 x double> @test_maxpd(<4 x double> %a0, <4 x double> %a1, <4 x double>
; SANDY-LABEL: test_maxpd:
; SANDY: # BB#0:
; SANDY-NEXT: vmaxpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: vmaxpd (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vmaxpd (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_maxpd:
; HASWELL: # BB#0:
@ -1256,8 +1256,8 @@ define <8 x float> @test_maxps(<8 x float> %a0, <8 x float> %a1, <8 x float> *%a
; SANDY-LABEL: test_maxps:
; SANDY: # BB#0:
; SANDY-NEXT: vmaxps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: vmaxps (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vmaxps (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_maxps:
; HASWELL: # BB#0:
@ -1288,7 +1288,7 @@ define <4 x double> @test_minpd(<4 x double> %a0, <4 x double> %a1, <4 x double>
; SANDY: # BB#0:
; SANDY-NEXT: vminpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: vminpd (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_minpd:
; HASWELL: # BB#0:
@ -1319,7 +1319,7 @@ define <8 x float> @test_minps(<8 x float> %a0, <8 x float> %a1, <8 x float> *%a
; SANDY: # BB#0:
; SANDY-NEXT: vminps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: vminps (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_minps:
; HASWELL: # BB#0:
@ -1348,10 +1348,10 @@ declare <8 x float> @llvm.x86.avx.min.ps.256(<8 x float>, <8 x float>) nounwind
define <4 x double> @test_movapd(<4 x double> *%a0, <4 x double> *%a1) {
; SANDY-LABEL: test_movapd:
; SANDY: # BB#0:
; SANDY-NEXT: vmovapd (%rdi), %ymm0 # sched: [4:0.50]
; SANDY-NEXT: vmovapd (%rdi), %ymm0 # sched: [7:0.50]
; SANDY-NEXT: vaddpd %ymm0, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: vmovapd %ymm0, (%rsi) # sched: [1:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vmovapd %ymm0, (%rsi) # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_movapd:
; HASWELL: # BB#0:
@ -1382,10 +1382,10 @@ define <4 x double> @test_movapd(<4 x double> *%a0, <4 x double> *%a1) {
define <8 x float> @test_movaps(<8 x float> *%a0, <8 x float> *%a1) {
; SANDY-LABEL: test_movaps:
; SANDY: # BB#0:
; SANDY-NEXT: vmovaps (%rdi), %ymm0 # sched: [4:0.50]
; SANDY-NEXT: vmovaps (%rdi), %ymm0 # sched: [7:0.50]
; SANDY-NEXT: vaddps %ymm0, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: vmovaps %ymm0, (%rsi) # sched: [1:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vmovaps %ymm0, (%rsi) # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_movaps:
; HASWELL: # BB#0:
@ -1417,9 +1417,9 @@ define <4 x double> @test_movddup(<4 x double> %a0, <4 x double> *%a1) {
; SANDY-LABEL: test_movddup:
; SANDY: # BB#0:
; SANDY-NEXT: vmovddup {{.*#+}} ymm0 = ymm0[0,0,2,2] sched: [1:1.00]
; SANDY-NEXT: vmovddup {{.*#+}} ymm1 = mem[0,0,2,2] sched: [4:0.50]
; SANDY-NEXT: vmovddup {{.*#+}} ymm1 = mem[0,0,2,2] sched: [7:0.50]
; SANDY-NEXT: vaddpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_movddup:
; HASWELL: # BB#0:
@ -1451,9 +1451,9 @@ define <4 x double> @test_movddup(<4 x double> %a0, <4 x double> *%a1) {
define i32 @test_movmskpd(<4 x double> %a0) {
; SANDY-LABEL: test_movmskpd:
; SANDY: # BB#0:
; SANDY-NEXT: vmovmskpd %ymm0, %eax # sched: [1:0.33]
; SANDY-NEXT: vmovmskpd %ymm0, %eax # sched: [2:1.00]
; SANDY-NEXT: vzeroupper # sched: [?:0.000000e+00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_movmskpd:
; HASWELL: # BB#0:
@ -1479,9 +1479,9 @@ declare i32 @llvm.x86.avx.movmsk.pd.256(<4 x double>) nounwind readnone
define i32 @test_movmskps(<8 x float> %a0) {
; SANDY-LABEL: test_movmskps:
; SANDY: # BB#0:
; SANDY-NEXT: vmovmskps %ymm0, %eax # sched: [1:0.33]
; SANDY-NEXT: vmovmskps %ymm0, %eax # sched: [3:1.00]
; SANDY-NEXT: vzeroupper # sched: [?:0.000000e+00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_movmskps:
; HASWELL: # BB#0:
@ -1508,8 +1508,8 @@ define <4 x double> @test_movntpd(<4 x double> %a0, <4 x double> *%a1) {
; SANDY-LABEL: test_movntpd:
; SANDY: # BB#0:
; SANDY-NEXT: vaddpd %ymm0, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: vmovntpd %ymm0, (%rdi) # sched: [1:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vmovntpd %ymm0, (%rdi) # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_movntpd:
; HASWELL: # BB#0:
@ -1537,8 +1537,8 @@ define <8 x float> @test_movntps(<8 x float> %a0, <8 x float> *%a1) {
; SANDY-LABEL: test_movntps:
; SANDY: # BB#0:
; SANDY-NEXT: vaddps %ymm0, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: vmovntps %ymm0, (%rdi) # sched: [1:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vmovntps %ymm0, (%rdi) # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_movntps:
; HASWELL: # BB#0:
@ -1566,9 +1566,9 @@ define <8 x float> @test_movshdup(<8 x float> %a0, <8 x float> *%a1) {
; SANDY-LABEL: test_movshdup:
; SANDY: # BB#0:
; SANDY-NEXT: vmovshdup {{.*#+}} ymm0 = ymm0[1,1,3,3,5,5,7,7] sched: [1:1.00]
; SANDY-NEXT: vmovshdup {{.*#+}} ymm1 = mem[1,1,3,3,5,5,7,7] sched: [4:0.50]
; SANDY-NEXT: vmovshdup {{.*#+}} ymm1 = mem[1,1,3,3,5,5,7,7] sched: [7:0.50]
; SANDY-NEXT: vaddps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_movshdup:
; HASWELL: # BB#0:
@ -1601,9 +1601,9 @@ define <8 x float> @test_movsldup(<8 x float> %a0, <8 x float> *%a1) {
; SANDY-LABEL: test_movsldup:
; SANDY: # BB#0:
; SANDY-NEXT: vmovsldup {{.*#+}} ymm0 = ymm0[0,0,2,2,4,4,6,6] sched: [1:1.00]
; SANDY-NEXT: vmovsldup {{.*#+}} ymm1 = mem[0,0,2,2,4,4,6,6] sched: [4:0.50]
; SANDY-NEXT: vmovsldup {{.*#+}} ymm1 = mem[0,0,2,2,4,4,6,6] sched: [7:0.50]
; SANDY-NEXT: vaddps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_movsldup:
; HASWELL: # BB#0:
@ -1635,12 +1635,12 @@ define <8 x float> @test_movsldup(<8 x float> %a0, <8 x float> *%a1) {
define <4 x double> @test_movupd(<4 x double> *%a0, <4 x double> *%a1) {
; SANDY-LABEL: test_movupd:
; SANDY: # BB#0:
; SANDY-NEXT: vmovups (%rdi), %xmm0 # sched: [4:0.50]
; SANDY-NEXT: vinsertf128 $1, 16(%rdi), %ymm0, %ymm0 # sched: [5:1.00]
; SANDY-NEXT: vmovups (%rdi), %xmm0 # sched: [6:0.50]
; SANDY-NEXT: vinsertf128 $1, 16(%rdi), %ymm0, %ymm0 # sched: [7:1.00]
; SANDY-NEXT: vaddpd %ymm0, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: vextractf128 $1, %ymm0, 16(%rsi) # sched: [1:1.00]
; SANDY-NEXT: vmovupd %xmm0, (%rsi) # sched: [1:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vextractf128 $1, %ymm0, 16(%rsi) # sched: [5:1.00]
; SANDY-NEXT: vmovupd %xmm0, (%rsi) # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_movupd:
; HASWELL: # BB#0:
@ -1671,12 +1671,12 @@ define <4 x double> @test_movupd(<4 x double> *%a0, <4 x double> *%a1) {
define <8 x float> @test_movups(<8 x float> *%a0, <8 x float> *%a1) {
; SANDY-LABEL: test_movups:
; SANDY: # BB#0:
; SANDY-NEXT: vmovups (%rdi), %xmm0 # sched: [4:0.50]
; SANDY-NEXT: vinsertf128 $1, 16(%rdi), %ymm0, %ymm0 # sched: [5:1.00]
; SANDY-NEXT: vmovups (%rdi), %xmm0 # sched: [6:0.50]
; SANDY-NEXT: vinsertf128 $1, 16(%rdi), %ymm0, %ymm0 # sched: [7:1.00]
; SANDY-NEXT: vaddps %ymm0, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: vextractf128 $1, %ymm0, 16(%rsi) # sched: [1:1.00]
; SANDY-NEXT: vmovups %xmm0, (%rsi) # sched: [1:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vextractf128 $1, %ymm0, 16(%rsi) # sched: [5:1.00]
; SANDY-NEXT: vmovups %xmm0, (%rsi) # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_movups:
; HASWELL: # BB#0:
@ -1708,8 +1708,8 @@ define <4 x double> @test_mulpd(<4 x double> %a0, <4 x double> %a1, <4 x double>
; SANDY-LABEL: test_mulpd:
; SANDY: # BB#0:
; SANDY-NEXT: vmulpd %ymm1, %ymm0, %ymm0 # sched: [5:1.00]
; SANDY-NEXT: vmulpd (%rdi), %ymm0, %ymm0 # sched: [9:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vmulpd (%rdi), %ymm0, %ymm0 # sched: [12:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_mulpd:
; HASWELL: # BB#0:
@ -1738,8 +1738,8 @@ define <8 x float> @test_mulps(<8 x float> %a0, <8 x float> %a1, <8 x float> *%a
; SANDY-LABEL: test_mulps:
; SANDY: # BB#0:
; SANDY-NEXT: vmulps %ymm1, %ymm0, %ymm0 # sched: [5:1.00]
; SANDY-NEXT: vmulps (%rdi), %ymm0, %ymm0 # sched: [9:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vmulps (%rdi), %ymm0, %ymm0 # sched: [12:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_mulps:
; HASWELL: # BB#0:
@ -1767,10 +1767,10 @@ define <8 x float> @test_mulps(<8 x float> %a0, <8 x float> %a1, <8 x float> *%a
define <4 x double> @orpd(<4 x double> %a0, <4 x double> %a1, <4 x double> *%a2) {
; SANDY-LABEL: orpd:
; SANDY: # BB#0:
; SANDY-NEXT: vorpd %ymm1, %ymm0, %ymm0 # sched: [1:0.33]
; SANDY-NEXT: vorpd (%rdi), %ymm0, %ymm0 # sched: [5:0.50]
; SANDY-NEXT: vorpd %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
; SANDY-NEXT: vorpd (%rdi), %ymm0, %ymm0 # sched: [8:1.00]
; SANDY-NEXT: vaddpd %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: orpd:
; HASWELL: # BB#0:
@ -1806,10 +1806,10 @@ define <4 x double> @orpd(<4 x double> %a0, <4 x double> %a1, <4 x double> *%a2)
define <8 x float> @test_orps(<8 x float> %a0, <8 x float> %a1, <8 x float> *%a2) {
; SANDY-LABEL: test_orps:
; SANDY: # BB#0:
; SANDY-NEXT: vorps %ymm1, %ymm0, %ymm0 # sched: [1:0.33]
; SANDY-NEXT: vorps (%rdi), %ymm0, %ymm0 # sched: [5:0.50]
; SANDY-NEXT: vorps %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
; SANDY-NEXT: vorps (%rdi), %ymm0, %ymm0 # sched: [8:1.00]
; SANDY-NEXT: vaddps %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_orps:
; HASWELL: # BB#0:
@ -1846,9 +1846,9 @@ define <2 x double> @test_permilpd(<2 x double> %a0, <2 x double> *%a1) {
; SANDY-LABEL: test_permilpd:
; SANDY: # BB#0:
; SANDY-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0] sched: [1:1.00]
; SANDY-NEXT: vpermilpd {{.*#+}} xmm1 = mem[1,0] sched: [5:1.00]
; SANDY-NEXT: vpermilpd {{.*#+}} xmm1 = mem[1,0] sched: [7:1.00]
; SANDY-NEXT: vaddpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_permilpd:
; HASWELL: # BB#0:
@ -1880,10 +1880,10 @@ define <2 x double> @test_permilpd(<2 x double> %a0, <2 x double> *%a1) {
define <4 x double> @test_permilpd_ymm(<4 x double> %a0, <4 x double> *%a1) {
; SANDY-LABEL: test_permilpd_ymm:
; SANDY: # BB#0:
; SANDY-NEXT: vpermilpd {{.*#+}} ymm0 = ymm0[1,0,2,3] sched: [1:1.00]
; SANDY-NEXT: vpermilpd {{.*#+}} ymm0 = ymm0[1,0,2,3] sched: [8:1.00]
; SANDY-NEXT: vpermilpd {{.*#+}} ymm1 = mem[1,0,2,3] sched: [5:1.00]
; SANDY-NEXT: vaddpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_permilpd_ymm:
; HASWELL: # BB#0:
@ -1916,9 +1916,9 @@ define <4 x float> @test_permilps(<4 x float> %a0, <4 x float> *%a1) {
; SANDY-LABEL: test_permilps:
; SANDY: # BB#0:
; SANDY-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,2,1,0] sched: [1:1.00]
; SANDY-NEXT: vpermilps {{.*#+}} xmm1 = mem[3,2,1,0] sched: [5:1.00]
; SANDY-NEXT: vpermilps {{.*#+}} xmm1 = mem[3,2,1,0] sched: [7:1.00]
; SANDY-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_permilps:
; HASWELL: # BB#0:
@ -1950,10 +1950,10 @@ define <4 x float> @test_permilps(<4 x float> %a0, <4 x float> *%a1) {
define <8 x float> @test_permilps_ymm(<8 x float> %a0, <8 x float> *%a1) {
; SANDY-LABEL: test_permilps_ymm:
; SANDY: # BB#0:
; SANDY-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4] sched: [1:1.00]
; SANDY-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4] sched: [8:1.00]
; SANDY-NEXT: vpermilps {{.*#+}} ymm1 = mem[3,2,1,0,7,6,5,4] sched: [5:1.00]
; SANDY-NEXT: vaddps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_permilps_ymm:
; HASWELL: # BB#0:
@ -1986,8 +1986,8 @@ define <2 x double> @test_permilvarpd(<2 x double> %a0, <2 x i64> %a1, <2 x i64>
; SANDY-LABEL: test_permilvarpd:
; SANDY: # BB#0:
; SANDY-NEXT: vpermilpd %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
; SANDY-NEXT: vpermilpd (%rdi), %xmm0, %xmm0 # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vpermilpd (%rdi), %xmm0, %xmm0 # sched: [1:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_permilvarpd:
; HASWELL: # BB#0:
@ -2018,7 +2018,7 @@ define <4 x double> @test_permilvarpd_ymm(<4 x double> %a0, <4 x i64> %a1, <4 x
; SANDY: # BB#0:
; SANDY-NEXT: vpermilpd %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
; SANDY-NEXT: vpermilpd (%rdi), %ymm0, %ymm0 # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_permilvarpd_ymm:
; HASWELL: # BB#0:
@ -2048,8 +2048,8 @@ define <4 x float> @test_permilvarps(<4 x float> %a0, <4 x i32> %a1, <4 x i32> *
; SANDY-LABEL: test_permilvarps:
; SANDY: # BB#0:
; SANDY-NEXT: vpermilps %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
; SANDY-NEXT: vpermilps (%rdi), %xmm0, %xmm0 # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vpermilps (%rdi), %xmm0, %xmm0 # sched: [1:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_permilvarps:
; HASWELL: # BB#0:
@ -2080,7 +2080,7 @@ define <8 x float> @test_permilvarps_ymm(<8 x float> %a0, <8 x i32> %a1, <8 x i3
; SANDY: # BB#0:
; SANDY-NEXT: vpermilps %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
; SANDY-NEXT: vpermilps (%rdi), %ymm0, %ymm0 # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_permilvarps_ymm:
; HASWELL: # BB#0:
@ -2112,7 +2112,7 @@ define <8 x float> @test_rcpps(<8 x float> %a0, <8 x float> *%a1) {
; SANDY-NEXT: vrcpps %ymm0, %ymm0 # sched: [5:1.00]
; SANDY-NEXT: vrcpps (%rdi), %ymm1 # sched: [9:1.00]
; SANDY-NEXT: vaddps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_rcpps:
; HASWELL: # BB#0:
@ -2148,7 +2148,7 @@ define <4 x double> @test_roundpd(<4 x double> %a0, <4 x double> *%a1) {
; SANDY-NEXT: vroundpd $7, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: vroundpd $7, (%rdi), %ymm1 # sched: [7:1.00]
; SANDY-NEXT: vaddpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_roundpd:
; HASWELL: # BB#0:
@ -2184,7 +2184,7 @@ define <8 x float> @test_roundps(<8 x float> %a0, <8 x float> *%a1) {
; SANDY-NEXT: vroundps $7, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: vroundps $7, (%rdi), %ymm1 # sched: [7:1.00]
; SANDY-NEXT: vaddps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_roundps:
; HASWELL: # BB#0:
@ -2217,10 +2217,10 @@ declare <8 x float> @llvm.x86.avx.round.ps.256(<8 x float>, i32) nounwind readno
define <8 x float> @test_rsqrtps(<8 x float> %a0, <8 x float> *%a1) {
; SANDY-LABEL: test_rsqrtps:
; SANDY: # BB#0:
; SANDY-NEXT: vrsqrtps %ymm0, %ymm0 # sched: [5:1.00]
; SANDY-NEXT: vrsqrtps (%rdi), %ymm1 # sched: [9:1.00]
; SANDY-NEXT: vrsqrtps (%rdi), %ymm1 # sched: [14:3.00]
; SANDY-NEXT: vrsqrtps %ymm0, %ymm0 # sched: [7:3.00]
; SANDY-NEXT: vaddps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_rsqrtps:
; HASWELL: # BB#0:
@ -2254,9 +2254,9 @@ define <4 x double> @test_shufpd(<4 x double> %a0, <4 x double> %a1, <4 x double
; SANDY-LABEL: test_shufpd:
; SANDY: # BB#0:
; SANDY-NEXT: vshufpd {{.*#+}} ymm0 = ymm0[1],ymm1[0],ymm0[2],ymm1[3] sched: [1:1.00]
; SANDY-NEXT: vshufpd {{.*#+}} ymm1 = ymm1[1],mem[0],ymm1[2],mem[3] sched: [5:1.00]
; SANDY-NEXT: vshufpd {{.*#+}} ymm1 = ymm1[1],mem[0],ymm1[2],mem[3] sched: [8:1.00]
; SANDY-NEXT: vaddpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_shufpd:
; HASWELL: # BB#0:
@ -2289,8 +2289,8 @@ define <8 x float> @test_shufps(<8 x float> %a0, <8 x float> %a1, <8 x float> *%
; SANDY-LABEL: test_shufps:
; SANDY: # BB#0:
; SANDY-NEXT: vshufps {{.*#+}} ymm0 = ymm0[0,0],ymm1[0,0],ymm0[4,4],ymm1[4,4] sched: [1:1.00]
; SANDY-NEXT: vshufps {{.*#+}} ymm0 = ymm0[0,3],mem[0,0],ymm0[4,7],mem[4,4] sched: [5:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vshufps {{.*#+}} ymm0 = ymm0[0,3],mem[0,0],ymm0[4,7],mem[4,4] sched: [8:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_shufps:
; HASWELL: # BB#0:
@ -2318,10 +2318,10 @@ define <8 x float> @test_shufps(<8 x float> %a0, <8 x float> %a1, <8 x float> *%
define <4 x double> @test_sqrtpd(<4 x double> %a0, <4 x double> *%a1) {
; SANDY-LABEL: test_sqrtpd:
; SANDY: # BB#0:
; SANDY-NEXT: vsqrtpd %ymm0, %ymm0 # sched: [15:1.00]
; SANDY-NEXT: vsqrtpd (%rdi), %ymm1 # sched: [19:1.00]
; SANDY-NEXT: vsqrtpd (%rdi), %ymm1 # sched: [52:3.00]
; SANDY-NEXT: vsqrtpd %ymm0, %ymm0 # sched: [45:3.00]
; SANDY-NEXT: vaddpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_sqrtpd:
; HASWELL: # BB#0:
@ -2354,10 +2354,10 @@ declare <4 x double> @llvm.x86.avx.sqrt.pd.256(<4 x double>) nounwind readnone
define <8 x float> @test_sqrtps(<8 x float> %a0, <8 x float> *%a1) {
; SANDY-LABEL: test_sqrtps:
; SANDY: # BB#0:
; SANDY-NEXT: vsqrtps %ymm0, %ymm0 # sched: [15:1.00]
; SANDY-NEXT: vsqrtps (%rdi), %ymm1 # sched: [19:1.00]
; SANDY-NEXT: vsqrtps (%rdi), %ymm1 # sched: [36:3.00]
; SANDY-NEXT: vsqrtps %ymm0, %ymm0 # sched: [29:3.00]
; SANDY-NEXT: vaddps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_sqrtps:
; HASWELL: # BB#0:
@ -2391,8 +2391,8 @@ define <4 x double> @test_subpd(<4 x double> %a0, <4 x double> %a1, <4 x double>
; SANDY-LABEL: test_subpd:
; SANDY: # BB#0:
; SANDY-NEXT: vsubpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: vsubpd (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vsubpd (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_subpd:
; HASWELL: # BB#0:
@ -2421,8 +2421,8 @@ define <8 x float> @test_subps(<8 x float> %a0, <8 x float> %a1, <8 x float> *%a
; SANDY-LABEL: test_subps:
; SANDY: # BB#0:
; SANDY-NEXT: vsubps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: vsubps (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vsubps (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_subps:
; HASWELL: # BB#0:
@ -2451,11 +2451,11 @@ define i32 @test_testpd(<2 x double> %a0, <2 x double> %a1, <2 x double> *%a2) {
; SANDY-LABEL: test_testpd:
; SANDY: # BB#0:
; SANDY-NEXT: xorl %eax, %eax # sched: [1:0.33]
; SANDY-NEXT: vtestpd %xmm1, %xmm0 # sched: [1:0.33]
; SANDY-NEXT: setb %al # sched: [1:0.33]
; SANDY-NEXT: vtestpd (%rdi), %xmm0 # sched: [5:0.50]
; SANDY-NEXT: vtestpd %xmm1, %xmm0 # sched: [1:1.00]
; SANDY-NEXT: setb %al # sched: [1:1.00]
; SANDY-NEXT: vtestpd (%rdi), %xmm0 # sched: [7:1.00]
; SANDY-NEXT: adcl $0, %eax # sched: [1:0.33]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_testpd:
; HASWELL: # BB#0:
@ -2495,12 +2495,12 @@ define i32 @test_testpd_ymm(<4 x double> %a0, <4 x double> %a1, <4 x double> *%a
; SANDY-LABEL: test_testpd_ymm:
; SANDY: # BB#0:
; SANDY-NEXT: xorl %eax, %eax # sched: [1:0.33]
; SANDY-NEXT: vtestpd %ymm1, %ymm0 # sched: [1:0.33]
; SANDY-NEXT: setb %al # sched: [1:0.33]
; SANDY-NEXT: vtestpd (%rdi), %ymm0 # sched: [5:0.50]
; SANDY-NEXT: vtestpd %ymm1, %ymm0 # sched: [1:1.00]
; SANDY-NEXT: setb %al # sched: [1:1.00]
; SANDY-NEXT: vtestpd (%rdi), %ymm0 # sched: [8:1.00]
; SANDY-NEXT: adcl $0, %eax # sched: [1:0.33]
; SANDY-NEXT: vzeroupper # sched: [?:0.000000e+00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_testpd_ymm:
; HASWELL: # BB#0:
@ -2542,11 +2542,11 @@ define i32 @test_testps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a2) {
; SANDY-LABEL: test_testps:
; SANDY: # BB#0:
; SANDY-NEXT: xorl %eax, %eax # sched: [1:0.33]
; SANDY-NEXT: vtestps %xmm1, %xmm0 # sched: [1:0.33]
; SANDY-NEXT: setb %al # sched: [1:0.33]
; SANDY-NEXT: vtestps (%rdi), %xmm0 # sched: [5:0.50]
; SANDY-NEXT: vtestps %xmm1, %xmm0 # sched: [1:1.00]
; SANDY-NEXT: setb %al # sched: [1:1.00]
; SANDY-NEXT: vtestps (%rdi), %xmm0 # sched: [7:1.00]
; SANDY-NEXT: adcl $0, %eax # sched: [1:0.33]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_testps:
; HASWELL: # BB#0:
@ -2586,12 +2586,12 @@ define i32 @test_testps_ymm(<8 x float> %a0, <8 x float> %a1, <8 x float> *%a2)
; SANDY-LABEL: test_testps_ymm:
; SANDY: # BB#0:
; SANDY-NEXT: xorl %eax, %eax # sched: [1:0.33]
; SANDY-NEXT: vtestps %ymm1, %ymm0 # sched: [1:0.33]
; SANDY-NEXT: setb %al # sched: [1:0.33]
; SANDY-NEXT: vtestps (%rdi), %ymm0 # sched: [5:0.50]
; SANDY-NEXT: vtestps %ymm1, %ymm0 # sched: [1:1.00]
; SANDY-NEXT: setb %al # sched: [1:1.00]
; SANDY-NEXT: vtestps (%rdi), %ymm0 # sched: [8:1.00]
; SANDY-NEXT: adcl $0, %eax # sched: [1:0.33]
; SANDY-NEXT: vzeroupper # sched: [?:0.000000e+00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_testps_ymm:
; HASWELL: # BB#0:
@ -2635,7 +2635,7 @@ define <4 x double> @test_unpckhpd(<4 x double> %a0, <4 x double> %a1, <4 x doub
; SANDY-NEXT: vunpckhpd {{.*#+}} ymm0 = ymm0[1],ymm1[1],ymm0[3],ymm1[3] sched: [1:1.00]
; SANDY-NEXT: vunpckhpd {{.*#+}} ymm1 = ymm1[1],mem[1],ymm1[3],mem[3] sched: [5:1.00]
; SANDY-NEXT: vaddpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_unpckhpd:
; HASWELL: # BB#0:
@ -2669,7 +2669,7 @@ define <8 x float> @test_unpckhps(<8 x float> %a0, <8 x float> %a1, <8 x float>
; SANDY: # BB#0:
; SANDY-NEXT: vunpckhps {{.*#+}} ymm0 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7] sched: [1:1.00]
; SANDY-NEXT: vunpckhps {{.*#+}} ymm0 = ymm0[2],mem[2],ymm0[3],mem[3],ymm0[6],mem[6],ymm0[7],mem[7] sched: [5:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_unpckhps:
; HASWELL: # BB#0:
@ -2698,9 +2698,9 @@ define <4 x double> @test_unpcklpd(<4 x double> %a0, <4 x double> %a1, <4 x doub
; SANDY-LABEL: test_unpcklpd:
; SANDY: # BB#0:
; SANDY-NEXT: vunpcklpd {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[2],ymm1[2] sched: [1:1.00]
; SANDY-NEXT: vunpcklpd {{.*#+}} ymm1 = ymm1[0],mem[0],ymm1[2],mem[2] sched: [5:1.00]
; SANDY-NEXT: vunpcklpd {{.*#+}} ymm1 = ymm1[0],mem[0],ymm1[2],mem[2] sched: [8:1.00]
; SANDY-NEXT: vaddpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_unpcklpd:
; HASWELL: # BB#0:
@ -2733,8 +2733,8 @@ define <8 x float> @test_unpcklps(<8 x float> %a0, <8 x float> %a1, <8 x float>
; SANDY-LABEL: test_unpcklps:
; SANDY: # BB#0:
; SANDY-NEXT: vunpcklps {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5] sched: [1:1.00]
; SANDY-NEXT: vunpcklps {{.*#+}} ymm0 = ymm0[0],mem[0],ymm0[1],mem[1],ymm0[4],mem[4],ymm0[5],mem[5] sched: [5:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vunpcklps {{.*#+}} ymm0 = ymm0[0],mem[0],ymm0[1],mem[1],ymm0[4],mem[4],ymm0[5],mem[5] sched: [8:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_unpcklps:
; HASWELL: # BB#0:
@ -2762,10 +2762,10 @@ define <8 x float> @test_unpcklps(<8 x float> %a0, <8 x float> %a1, <8 x float>
define <4 x double> @test_xorpd(<4 x double> %a0, <4 x double> %a1, <4 x double> *%a2) {
; SANDY-LABEL: test_xorpd:
; SANDY: # BB#0:
; SANDY-NEXT: vxorpd %ymm1, %ymm0, %ymm0 # sched: [1:0.33]
; SANDY-NEXT: vxorpd (%rdi), %ymm0, %ymm0 # sched: [5:0.50]
; SANDY-NEXT: vxorpd %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
; SANDY-NEXT: vxorpd (%rdi), %ymm0, %ymm0 # sched: [5:1.00]
; SANDY-NEXT: vaddpd %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_xorpd:
; HASWELL: # BB#0:
@ -2801,10 +2801,10 @@ define <4 x double> @test_xorpd(<4 x double> %a0, <4 x double> %a1, <4 x double>
define <8 x float> @test_xorps(<8 x float> %a0, <8 x float> %a1, <8 x float> *%a2) {
; SANDY-LABEL: test_xorps:
; SANDY: # BB#0:
; SANDY-NEXT: vxorps %ymm1, %ymm0, %ymm0 # sched: [1:0.33]
; SANDY-NEXT: vxorps (%rdi), %ymm0, %ymm0 # sched: [5:0.50]
; SANDY-NEXT: vxorps %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
; SANDY-NEXT: vxorps (%rdi), %ymm0, %ymm0 # sched: [5:1.00]
; SANDY-NEXT: vaddps %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_xorps:
; HASWELL: # BB#0:
@ -2841,7 +2841,7 @@ define void @test_zeroall() {
; SANDY-LABEL: test_zeroall:
; SANDY: # BB#0:
; SANDY-NEXT: vzeroall # sched: [?:0.000000e+00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_zeroall:
; HASWELL: # BB#0:
@ -2866,7 +2866,7 @@ define void @test_zeroupper() {
; SANDY-LABEL: test_zeroupper:
; SANDY: # BB#0:
; SANDY-NEXT: vzeroupper # sched: [?:0.000000e+00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_zeroupper:
; HASWELL: # BB#0:

22
test/CodeGen/X86/extractelement-legalization-store-ordering.ll
View File

@ -15,18 +15,18 @@ define void @test_extractelement_legalization_storereuse(<4 x i32> %a, i32* noca
; CHECK-NEXT: pushl %esi
; CHECK-NEXT: movl {{[0-9]+}}(%esp), %eax
; CHECK-NEXT: movl {{[0-9]+}}(%esp), %ecx
; CHECK-NEXT: paddd (%ecx), %xmm0
; CHECK-NEXT: movl {{[0-9]+}}(%esp), %edx
; CHECK-NEXT: paddd (%edx), %xmm0
; CHECK-NEXT: movdqa %xmm0, (%edx)
; CHECK-NEXT: movl (%edx), %esi
; CHECK-NEXT: movl 4(%edx), %edi
; CHECK-NEXT: shll $4, %ecx
; CHECK-NEXT: movl 8(%edx), %ebx
; CHECK-NEXT: movl 12(%edx), %edx
; CHECK-NEXT: movl %esi, 12(%eax,%ecx)
; CHECK-NEXT: movl %edi, (%eax,%ecx)
; CHECK-NEXT: movl %ebx, 8(%eax,%ecx)
; CHECK-NEXT: movl %edx, 4(%eax,%ecx)
; CHECK-NEXT: movdqa %xmm0, (%ecx)
; CHECK-NEXT: movl (%ecx), %esi
; CHECK-NEXT: movl 4(%ecx), %edi
; CHECK-NEXT: shll $4, %edx
; CHECK-NEXT: movl 8(%ecx), %ebx
; CHECK-NEXT: movl 12(%ecx), %ecx
; CHECK-NEXT: movl %esi, 12(%eax,%edx)
; CHECK-NEXT: movl %edi, (%eax,%edx)
; CHECK-NEXT: movl %ebx, 8(%eax,%edx)
; CHECK-NEXT: movl %ecx, 4(%eax,%edx)
; CHECK-NEXT: popl %esi
; CHECK-NEXT: popl %edi
; CHECK-NEXT: popl %ebx

2
test/CodeGen/X86/fp128-i128.ll
View File

@ -50,8 +50,8 @@ define void @TestUnionLD1(fp128 %s, i64 %n) #0 {
; CHECK-NEXT: andq %rdi, %rcx
; CHECK-NEXT: movabsq $-281474976710656, %rdx # imm = 0xFFFF000000000000
; CHECK-NEXT: andq -{{[0-9]+}}(%rsp), %rdx
; CHECK-NEXT: orq %rcx, %rdx
; CHECK-NEXT: movq %rax, -{{[0-9]+}}(%rsp)
; CHECK-NEXT: orq %rcx, %rdx
; CHECK-NEXT: movq %rdx, -{{[0-9]+}}(%rsp)
; CHECK-NEXT: movaps -{{[0-9]+}}(%rsp), %xmm0
; CHECK-NEXT: jmp foo # TAILCALL

16
test/CodeGen/X86/gather-addresses.ll
View File

@ -16,10 +16,10 @@
; LIN: sarq $32, %r[[REG2]]
; LIN: movslq %e[[REG4]], %r[[REG3:.+]]
; LIN: sarq $32, %r[[REG4]]
; LIN: movsd (%rdi,%r[[REG1]],8), %xmm0
; LIN: movhpd (%rdi,%r[[REG2]],8), %xmm0
; LIN: movsd (%rdi,%r[[REG3]],8), %xmm1
; LIN: movhpd (%rdi,%r[[REG4]],8), %xmm1
; LIN: movsd (%rdi,%r[[REG3]],8), %xmm1
; LIN: movhpd (%rdi,%r[[REG4]],8), %xmm1
; LIN: movq %rdi, %xmm1
; LIN: movq %r[[REG3]], %xmm0
; WIN: movdqa (%rdx), %xmm0
; WIN: pand (%r8), %xmm0
@ -29,10 +29,10 @@
; WIN: sarq $32, %r[[REG2]]
; WIN: movslq %e[[REG4]], %r[[REG3:.+]]
; WIN: sarq $32, %r[[REG4]]
; WIN: movsd (%rcx,%r[[REG1]],8), %xmm0
; WIN: movhpd (%rcx,%r[[REG2]],8), %xmm0
; WIN: movsd (%rcx,%r[[REG3]],8), %xmm1
; WIN: movhpd (%rcx,%r[[REG4]],8), %xmm1
; WIN: movsd (%rcx,%r[[REG3]],8), %xmm1
; WIN: movhpd (%rcx,%r[[REG4]],8), %xmm1
; WIN: movdqa (%r[[REG2]]), %xmm0
; WIN: movq %r[[REG2]], %xmm1
define <4 x double> @foo(double* %p, <4 x i32>* %i, <4 x i32>* %h) nounwind {
%a = load <4 x i32>, <4 x i32>* %i

46
test/CodeGen/X86/recip-fastmath.ll
View File

@ -45,9 +45,9 @@ define float @f32_no_estimate(float %x) #0 {
;
; SANDY-LABEL: f32_no_estimate:
; SANDY: # BB#0:
; SANDY-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero sched: [4:0.50]
; SANDY-NEXT: vdivss %xmm0, %xmm1, %xmm0 # sched: [12:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero sched: [6:0.50]
; SANDY-NEXT: vdivss %xmm0, %xmm1, %xmm0 # sched: [14:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: f32_no_estimate:
; HASWELL: # BB#0:
@ -113,11 +113,11 @@ define float @f32_one_step(float %x) #1 {
; SANDY: # BB#0:
; SANDY-NEXT: vrcpss %xmm0, %xmm0, %xmm1 # sched: [5:1.00]
; SANDY-NEXT: vmulss %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
; SANDY-NEXT: vmovss {{.*#+}} xmm2 = mem[0],zero,zero,zero sched: [4:0.50]
; SANDY-NEXT: vmovss {{.*#+}} xmm2 = mem[0],zero,zero,zero sched: [6:0.50]
; SANDY-NEXT: vsubss %xmm0, %xmm2, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vmulss %xmm0, %xmm1, %xmm0 # sched: [5:1.00]
; SANDY-NEXT: vaddss %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: f32_one_step:
; HASWELL: # BB#0:
@ -207,7 +207,7 @@ define float @f32_two_step(float %x) #2 {
; SANDY: # BB#0:
; SANDY-NEXT: vrcpss %xmm0, %xmm0, %xmm1 # sched: [5:1.00]
; SANDY-NEXT: vmulss %xmm1, %xmm0, %xmm2 # sched: [5:1.00]
; SANDY-NEXT: vmovss {{.*#+}} xmm3 = mem[0],zero,zero,zero sched: [4:0.50]
; SANDY-NEXT: vmovss {{.*#+}} xmm3 = mem[0],zero,zero,zero sched: [6:0.50]
; SANDY-NEXT: vsubss %xmm2, %xmm3, %xmm2 # sched: [3:1.00]
; SANDY-NEXT: vmulss %xmm2, %xmm1, %xmm2 # sched: [5:1.00]
; SANDY-NEXT: vaddss %xmm2, %xmm1, %xmm1 # sched: [3:1.00]
@ -215,7 +215,7 @@ define float @f32_two_step(float %x) #2 {
; SANDY-NEXT: vsubss %xmm0, %xmm3, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vmulss %xmm0, %xmm1, %xmm0 # sched: [5:1.00]
; SANDY-NEXT: vaddss %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: f32_two_step:
; HASWELL: # BB#0:
@ -284,9 +284,9 @@ define <4 x float> @v4f32_no_estimate(<4 x float> %x) #0 {
;
; SANDY-LABEL: v4f32_no_estimate:
; SANDY: # BB#0:
; SANDY-NEXT: vmovaps {{.*#+}} xmm1 = [1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00] sched: [4:0.50]
; SANDY-NEXT: vdivps %xmm0, %xmm1, %xmm0 # sched: [12:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vmovaps {{.*#+}} xmm1 = [1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00] sched: [6:0.50]
; SANDY-NEXT: vdivps %xmm0, %xmm1, %xmm0 # sched: [14:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: v4f32_no_estimate:
; HASWELL: # BB#0:
@ -350,13 +350,13 @@ define <4 x float> @v4f32_one_step(<4 x float> %x) #1 {
;
; SANDY-LABEL: v4f32_one_step:
; SANDY: # BB#0:
; SANDY-NEXT: vrcpps %xmm0, %xmm1 # sched: [5:1.00]
; SANDY-NEXT: vrcpps %xmm0, %xmm1 # sched: [7:3.00]
; SANDY-NEXT: vmulps %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
; SANDY-NEXT: vmovaps {{.*#+}} xmm2 = [1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00] sched: [4:0.50]
; SANDY-NEXT: vmovaps {{.*#+}} xmm2 = [1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00] sched: [6:0.50]
; SANDY-NEXT: vsubps %xmm0, %xmm2, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vmulps %xmm0, %xmm1, %xmm0 # sched: [5:1.00]
; SANDY-NEXT: vaddps %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: v4f32_one_step:
; HASWELL: # BB#0:
@ -453,9 +453,9 @@ define <4 x float> @v4f32_two_step(<4 x float> %x) #2 {
;
; SANDY-LABEL: v4f32_two_step:
; SANDY: # BB#0:
; SANDY-NEXT: vrcpps %xmm0, %xmm1 # sched: [5:1.00]
; SANDY-NEXT: vrcpps %xmm0, %xmm1 # sched: [7:3.00]
; SANDY-NEXT: vmulps %xmm1, %xmm0, %xmm2 # sched: [5:1.00]
; SANDY-NEXT: vmovaps {{.*#+}} xmm3 = [1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00] sched: [4:0.50]
; SANDY-NEXT: vmovaps {{.*#+}} xmm3 = [1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00] sched: [6:0.50]
; SANDY-NEXT: vsubps %xmm2, %xmm3, %xmm2 # sched: [3:1.00]
; SANDY-NEXT: vmulps %xmm2, %xmm1, %xmm2 # sched: [5:1.00]
; SANDY-NEXT: vaddps %xmm2, %xmm1, %xmm1 # sched: [3:1.00]
@ -463,7 +463,7 @@ define <4 x float> @v4f32_two_step(<4 x float> %x) #2 {
; SANDY-NEXT: vsubps %xmm0, %xmm3, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vmulps %xmm0, %xmm1, %xmm0 # sched: [5:1.00]
; SANDY-NEXT: vaddps %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: v4f32_two_step:
; HASWELL: # BB#0:
@ -546,9 +546,9 @@ define <8 x float> @v8f32_no_estimate(<8 x float> %x) #0 {
;
; SANDY-LABEL: v8f32_no_estimate:
; SANDY: # BB#0:
; SANDY-NEXT: vmovaps {{.*#+}} ymm1 = [1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00] sched: [4:0.50]
; SANDY-NEXT: vdivps %ymm0, %ymm1, %ymm0 # sched: [12:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vmovaps {{.*#+}} ymm1 = [1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00] sched: [7:0.50]
; SANDY-NEXT: vdivps %ymm0, %ymm1, %ymm0 # sched: [29:3.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: v8f32_no_estimate:
; HASWELL: # BB#0:
@ -621,11 +621,11 @@ define <8 x float> @v8f32_one_step(<8 x float> %x) #1 {
; SANDY: # BB#0:
; SANDY-NEXT: vrcpps %ymm0, %ymm1 # sched: [5:1.00]
; SANDY-NEXT: vmulps %ymm1, %ymm0, %ymm0 # sched: [5:1.00]
; SANDY-NEXT: vmovaps {{.*#+}} ymm2 = [1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00] sched: [4:0.50]
; SANDY-NEXT: vmovaps {{.*#+}} ymm2 = [1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00] sched: [7:0.50]
; SANDY-NEXT: vsubps %ymm0, %ymm2, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: vmulps %ymm0, %ymm1, %ymm0 # sched: [5:1.00]
; SANDY-NEXT: vaddps %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: v8f32_one_step:
; HASWELL: # BB#0:
@ -737,7 +737,7 @@ define <8 x float> @v8f32_two_step(<8 x float> %x) #2 {
; SANDY: # BB#0:
; SANDY-NEXT: vrcpps %ymm0, %ymm1 # sched: [5:1.00]
; SANDY-NEXT: vmulps %ymm1, %ymm0, %ymm2 # sched: [5:1.00]
; SANDY-NEXT: vmovaps {{.*#+}} ymm3 = [1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00] sched: [4:0.50]
; SANDY-NEXT: vmovaps {{.*#+}} ymm3 = [1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00] sched: [7:0.50]
; SANDY-NEXT: vsubps %ymm2, %ymm3, %ymm2 # sched: [3:1.00]
; SANDY-NEXT: vmulps %ymm2, %ymm1, %ymm2 # sched: [5:1.00]
; SANDY-NEXT: vaddps %ymm2, %ymm1, %ymm1 # sched: [3:1.00]
@ -745,7 +745,7 @@ define <8 x float> @v8f32_two_step(<8 x float> %x) #2 {
; SANDY-NEXT: vsubps %ymm0, %ymm3, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: vmulps %ymm0, %ymm1, %ymm0 # sched: [5:1.00]
; SANDY-NEXT: vaddps %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: v8f32_two_step:
; HASWELL: # BB#0:

70
test/CodeGen/X86/recip-fastmath2.ll
View File

@ -39,8 +39,8 @@ define float @f32_no_step_2(float %x) #3 {
; SANDY-LABEL: f32_no_step_2:
; SANDY: # BB#0:
; SANDY-NEXT: vrcpss %xmm0, %xmm0, %xmm0 # sched: [5:1.00]
; SANDY-NEXT: vmulss {{.*}}(%rip), %xmm0, %xmm0 # sched: [9:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vmulss {{.*}}(%rip), %xmm0, %xmm0 # sched: [11:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: f32_no_step_2:
; HASWELL: # BB#0:
@ -110,12 +110,12 @@ define float @f32_one_step_2(float %x) #1 {
; SANDY: # BB#0:
; SANDY-NEXT: vrcpss %xmm0, %xmm0, %xmm1 # sched: [5:1.00]
; SANDY-NEXT: vmulss %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
; SANDY-NEXT: vmovss {{.*#+}} xmm2 = mem[0],zero,zero,zero sched: [4:0.50]
; SANDY-NEXT: vmovss {{.*#+}} xmm2 = mem[0],zero,zero,zero sched: [6:0.50]
; SANDY-NEXT: vsubss %xmm0, %xmm2, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vmulss %xmm0, %xmm1, %xmm0 # sched: [5:1.00]
; SANDY-NEXT: vaddss %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vmulss {{.*}}(%rip), %xmm0, %xmm0 # sched: [9:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vmulss {{.*}}(%rip), %xmm0, %xmm0 # sched: [11:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: f32_one_step_2:
; HASWELL: # BB#0:
@ -198,13 +198,13 @@ define float @f32_one_step_2_divs(float %x) #1 {
; SANDY: # BB#0:
; SANDY-NEXT: vrcpss %xmm0, %xmm0, %xmm1 # sched: [5:1.00]
; SANDY-NEXT: vmulss %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
; SANDY-NEXT: vmovss {{.*#+}} xmm2 = mem[0],zero,zero,zero sched: [4:0.50]
; SANDY-NEXT: vmovss {{.*#+}} xmm2 = mem[0],zero,zero,zero sched: [6:0.50]
; SANDY-NEXT: vsubss %xmm0, %xmm2, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vmulss %xmm0, %xmm1, %xmm0 # sched: [5:1.00]
; SANDY-NEXT: vaddss %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vmulss {{.*}}(%rip), %xmm0, %xmm1 # sched: [9:1.00]
; SANDY-NEXT: vmulss {{.*}}(%rip), %xmm0, %xmm1 # sched: [11:1.00]
; SANDY-NEXT: vmulss %xmm0, %xmm1, %xmm0 # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: f32_one_step_2_divs:
; HASWELL: # BB#0:
@ -305,7 +305,7 @@ define float @f32_two_step_2(float %x) #2 {
; SANDY: # BB#0:
; SANDY-NEXT: vrcpss %xmm0, %xmm0, %xmm1 # sched: [5:1.00]
; SANDY-NEXT: vmulss %xmm1, %xmm0, %xmm2 # sched: [5:1.00]
; SANDY-NEXT: vmovss {{.*#+}} xmm3 = mem[0],zero,zero,zero sched: [4:0.50]
; SANDY-NEXT: vmovss {{.*#+}} xmm3 = mem[0],zero,zero,zero sched: [6:0.50]
; SANDY-NEXT: vsubss %xmm2, %xmm3, %xmm2 # sched: [3:1.00]
; SANDY-NEXT: vmulss %xmm2, %xmm1, %xmm2 # sched: [5:1.00]
; SANDY-NEXT: vaddss %xmm2, %xmm1, %xmm1 # sched: [3:1.00]
@ -313,8 +313,8 @@ define float @f32_two_step_2(float %x) #2 {
; SANDY-NEXT: vsubss %xmm0, %xmm3, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vmulss %xmm0, %xmm1, %xmm0 # sched: [5:1.00]
; SANDY-NEXT: vaddss %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vmulss {{.*}}(%rip), %xmm0, %xmm0 # sched: [9:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vmulss {{.*}}(%rip), %xmm0, %xmm0 # sched: [11:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: f32_two_step_2:
; HASWELL: # BB#0:
@ -403,14 +403,14 @@ define <4 x float> @v4f32_one_step2(<4 x float> %x) #1 {
;
; SANDY-LABEL: v4f32_one_step2:
; SANDY: # BB#0:
; SANDY-NEXT: vrcpps %xmm0, %xmm1 # sched: [5:1.00]
; SANDY-NEXT: vrcpps %xmm0, %xmm1 # sched: [7:3.00]
; SANDY-NEXT: vmulps %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
; SANDY-NEXT: vmovaps {{.*#+}} xmm2 = [1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00] sched: [4:0.50]
; SANDY-NEXT: vmovaps {{.*#+}} xmm2 = [1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00] sched: [6:0.50]
; SANDY-NEXT: vsubps %xmm0, %xmm2, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vmulps %xmm0, %xmm1, %xmm0 # sched: [5:1.00]
; SANDY-NEXT: vaddps %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vmulps {{.*}}(%rip), %xmm0, %xmm0 # sched: [9:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vmulps {{.*}}(%rip), %xmm0, %xmm0 # sched: [11:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: v4f32_one_step2:
; HASWELL: # BB#0:
@ -501,15 +501,15 @@ define <4 x float> @v4f32_one_step_2_divs(<4 x float> %x) #1 {
;
; SANDY-LABEL: v4f32_one_step_2_divs:
; SANDY: # BB#0:
; SANDY-NEXT: vrcpps %xmm0, %xmm1 # sched: [5:1.00]
; SANDY-NEXT: vrcpps %xmm0, %xmm1 # sched: [7:3.00]
; SANDY-NEXT: vmulps %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
; SANDY-NEXT: vmovaps {{.*#+}} xmm2 = [1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00] sched: [4:0.50]
; SANDY-NEXT: vmovaps {{.*#+}} xmm2 = [1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00] sched: [6:0.50]
; SANDY-NEXT: vsubps %xmm0, %xmm2, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vmulps %xmm0, %xmm1, %xmm0 # sched: [5:1.00]
; SANDY-NEXT: vaddps %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vmulps {{.*}}(%rip), %xmm0, %xmm1 # sched: [9:1.00]
; SANDY-NEXT: vmulps {{.*}}(%rip), %xmm0, %xmm1 # sched: [11:1.00]
; SANDY-NEXT: vmulps %xmm0, %xmm1, %xmm0 # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: v4f32_one_step_2_divs:
; HASWELL: # BB#0:
@ -619,9 +619,9 @@ define <4 x float> @v4f32_two_step2(<4 x float> %x) #2 {
;
; SANDY-LABEL: v4f32_two_step2:
; SANDY: # BB#0:
; SANDY-NEXT: vrcpps %xmm0, %xmm1 # sched: [5:1.00]
; SANDY-NEXT: vrcpps %xmm0, %xmm1 # sched: [7:3.00]
; SANDY-NEXT: vmulps %xmm1, %xmm0, %xmm2 # sched: [5:1.00]
; SANDY-NEXT: vmovaps {{.*#+}} xmm3 = [1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00] sched: [4:0.50]
; SANDY-NEXT: vmovaps {{.*#+}} xmm3 = [1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00] sched: [6:0.50]
; SANDY-NEXT: vsubps %xmm2, %xmm3, %xmm2 # sched: [3:1.00]
; SANDY-NEXT: vmulps %xmm2, %xmm1, %xmm2 # sched: [5:1.00]
; SANDY-NEXT: vaddps %xmm2, %xmm1, %xmm1 # sched: [3:1.00]
@ -629,8 +629,8 @@ define <4 x float> @v4f32_two_step2(<4 x float> %x) #2 {
; SANDY-NEXT: vsubps %xmm0, %xmm3, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vmulps %xmm0, %xmm1, %xmm0 # sched: [5:1.00]
; SANDY-NEXT: vaddps %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vmulps {{.*}}(%rip), %xmm0, %xmm0 # sched: [9:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vmulps {{.*}}(%rip), %xmm0, %xmm0 # sched: [11:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: v4f32_two_step2:
; HASWELL: # BB#0:
@ -741,12 +741,12 @@ define <8 x float> @v8f32_one_step2(<8 x float> %x) #1 {
; SANDY: # BB#0:
; SANDY-NEXT: vrcpps %ymm0, %ymm1 # sched: [5:1.00]
; SANDY-NEXT: vmulps %ymm1, %ymm0, %ymm0 # sched: [5:1.00]
; SANDY-NEXT: vmovaps {{.*#+}} ymm2 = [1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00] sched: [4:0.50]
; SANDY-NEXT: vmovaps {{.*#+}} ymm2 = [1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00] sched: [7:0.50]
; SANDY-NEXT: vsubps %ymm0, %ymm2, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: vmulps %ymm0, %ymm1, %ymm0 # sched: [5:1.00]
; SANDY-NEXT: vaddps %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: vmulps {{.*}}(%rip), %ymm0, %ymm0 # sched: [9:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vmulps {{.*}}(%rip), %ymm0, %ymm0 # sched: [12:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: v8f32_one_step2:
; HASWELL: # BB#0:
@ -848,13 +848,13 @@ define <8 x float> @v8f32_one_step_2_divs(<8 x float> %x) #1 {
; SANDY: # BB#0:
; SANDY-NEXT: vrcpps %ymm0, %ymm1 # sched: [5:1.00]
; SANDY-NEXT: vmulps %ymm1, %ymm0, %ymm0 # sched: [5:1.00]
; SANDY-NEXT: vmovaps {{.*#+}} ymm2 = [1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00] sched: [4:0.50]
; SANDY-NEXT: vmovaps {{.*#+}} ymm2 = [1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00] sched: [7:0.50]
; SANDY-NEXT: vsubps %ymm0, %ymm2, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: vmulps %ymm0, %ymm1, %ymm0 # sched: [5:1.00]
; SANDY-NEXT: vaddps %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: vmulps {{.*}}(%rip), %ymm0, %ymm1 # sched: [9:1.00]
; SANDY-NEXT: vmulps {{.*}}(%rip), %ymm0, %ymm1 # sched: [12:1.00]
; SANDY-NEXT: vmulps %ymm0, %ymm1, %ymm0 # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: v8f32_one_step_2_divs:
; HASWELL: # BB#0:
@ -980,7 +980,7 @@ define <8 x float> @v8f32_two_step2(<8 x float> %x) #2 {
; SANDY: # BB#0:
; SANDY-NEXT: vrcpps %ymm0, %ymm1 # sched: [5:1.00]
; SANDY-NEXT: vmulps %ymm1, %ymm0, %ymm2 # sched: [5:1.00]
; SANDY-NEXT: vmovaps {{.*#+}} ymm3 = [1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00] sched: [4:0.50]
; SANDY-NEXT: vmovaps {{.*#+}} ymm3 = [1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00] sched: [7:0.50]
; SANDY-NEXT: vsubps %ymm2, %ymm3, %ymm2 # sched: [3:1.00]
; SANDY-NEXT: vmulps %ymm2, %ymm1, %ymm2 # sched: [5:1.00]
; SANDY-NEXT: vaddps %ymm2, %ymm1, %ymm1 # sched: [3:1.00]
@ -988,8 +988,8 @@ define <8 x float> @v8f32_two_step2(<8 x float> %x) #2 {
; SANDY-NEXT: vsubps %ymm0, %ymm3, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: vmulps %ymm0, %ymm1, %ymm0 # sched: [5:1.00]
; SANDY-NEXT: vaddps %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
; SANDY-NEXT: vmulps {{.*}}(%rip), %ymm0, %ymm0 # sched: [9:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vmulps {{.*}}(%rip), %ymm0, %ymm0 # sched: [12:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: v8f32_two_step2:
; HASWELL: # BB#0:
@ -1070,7 +1070,7 @@ define <8 x float> @v8f32_no_step(<8 x float> %x) #3 {
; SANDY-LABEL: v8f32_no_step:
; SANDY: # BB#0:
; SANDY-NEXT: vrcpps %ymm0, %ymm0 # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: v8f32_no_step:
; HASWELL: # BB#0:
@ -1125,8 +1125,8 @@ define <8 x float> @v8f32_no_step2(<8 x float> %x) #3 {
; SANDY-LABEL: v8f32_no_step2:
; SANDY: # BB#0:
; SANDY-NEXT: vrcpps %ymm0, %ymm0 # sched: [5:1.00]
; SANDY-NEXT: vmulps {{.*}}(%rip), %ymm0, %ymm0 # sched: [9:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vmulps {{.*}}(%rip), %ymm0, %ymm0 # sched: [12:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: v8f32_no_step2:
; HASWELL: # BB#0:

248
test/CodeGen/X86/sse-schedule.ll
View File

@ -31,8 +31,8 @@ define <4 x float> @test_addps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a
; SANDY-LABEL: test_addps:
; SANDY: # BB#0:
; SANDY-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vaddps (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vaddps (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_addps:
; HASWELL: # BB#0:
@ -73,8 +73,8 @@ define float @test_addss(float %a0, float %a1, float *%a2) {
; SANDY-LABEL: test_addss:
; SANDY: # BB#0:
; SANDY-NEXT: vaddss %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vaddss (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vaddss (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_addss:
; HASWELL: # BB#0:
@ -122,9 +122,9 @@ define <4 x float> @test_andps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a
;
; SANDY-LABEL: test_andps:
; SANDY: # BB#0:
; SANDY-NEXT: vandps %xmm1, %xmm0, %xmm0 # sched: [1:0.33]
; SANDY-NEXT: vandps (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vandps %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
; SANDY-NEXT: vandps (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_andps:
; HASWELL: # BB#0:
@ -176,9 +176,9 @@ define <4 x float> @test_andnotps(<4 x float> %a0, <4 x float> %a1, <4 x float>
;
; SANDY-LABEL: test_andnotps:
; SANDY: # BB#0:
; SANDY-NEXT: vandnps %xmm1, %xmm0, %xmm0 # sched: [1:0.33]
; SANDY-NEXT: vandnps (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vandnps %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
; SANDY-NEXT: vandnps (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_andnotps:
; HASWELL: # BB#0:
@ -228,9 +228,9 @@ define <4 x float> @test_cmpps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a
; SANDY-LABEL: test_cmpps:
; SANDY: # BB#0:
; SANDY-NEXT: vcmpeqps %xmm1, %xmm0, %xmm1 # sched: [3:1.00]
; SANDY-NEXT: vcmpeqps (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
; SANDY-NEXT: vorps %xmm0, %xmm1, %xmm0 # sched: [1:0.33]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vcmpeqps (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
; SANDY-NEXT: vorps %xmm0, %xmm1, %xmm0 # sched: [1:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_cmpps:
; HASWELL: # BB#0:
@ -277,7 +277,7 @@ define float @test_cmpss(float %a0, float %a1, float *%a2) {
; SANDY: # BB#0:
; SANDY-NEXT: vcmpeqss %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vcmpeqss (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_cmpss:
; HASWELL: # BB#0:
@ -347,16 +347,16 @@ define i32 @test_comiss(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a2) {
; SANDY-LABEL: test_comiss:
; SANDY: # BB#0:
; SANDY-NEXT: vcomiss %xmm1, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: setnp %al # sched: [1:0.33]
; SANDY-NEXT: sete %cl # sched: [1:0.33]
; SANDY-NEXT: setnp %al # sched: [1:1.00]
; SANDY-NEXT: sete %cl # sched: [1:1.00]
; SANDY-NEXT: andb %al, %cl # sched: [1:0.33]
; SANDY-NEXT: vcomiss (%rdi), %xmm0 # sched: [7:1.00]
; SANDY-NEXT: setnp %al # sched: [1:0.33]
; SANDY-NEXT: sete %dl # sched: [1:0.33]
; SANDY-NEXT: setnp %al # sched: [1:1.00]
; SANDY-NEXT: sete %dl # sched: [1:1.00]
; SANDY-NEXT: andb %al, %dl # sched: [1:0.33]
; SANDY-NEXT: orb %cl, %dl # sched: [1:0.33]
; SANDY-NEXT: movzbl %dl, %eax # sched: [1:0.33]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_comiss:
; HASWELL: # BB#0:
@ -417,10 +417,10 @@ define float @test_cvtsi2ss(i32 %a0, i32 *%a1) {
;
; SANDY-LABEL: test_cvtsi2ss:
; SANDY: # BB#0:
; SANDY-NEXT: vcvtsi2ssl %edi, %xmm0, %xmm0 # sched: [4:1.00]
; SANDY-NEXT: vcvtsi2ssl (%rsi), %xmm1, %xmm1 # sched: [8:1.00]
; SANDY-NEXT: vcvtsi2ssl %edi, %xmm0, %xmm0 # sched: [5:2.00]
; SANDY-NEXT: vcvtsi2ssl (%rsi), %xmm1, %xmm1 # sched: [10:1.00]
; SANDY-NEXT: vaddss %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_cvtsi2ss:
; HASWELL: # BB#0:
@ -466,10 +466,10 @@ define float @test_cvtsi2ssq(i64 %a0, i64 *%a1) {
;
; SANDY-LABEL: test_cvtsi2ssq:
; SANDY: # BB#0:
; SANDY-NEXT: vcvtsi2ssq %rdi, %xmm0, %xmm0 # sched: [4:1.00]
; SANDY-NEXT: vcvtsi2ssq (%rsi), %xmm1, %xmm1 # sched: [8:1.00]
; SANDY-NEXT: vcvtsi2ssq %rdi, %xmm0, %xmm0 # sched: [5:2.00]
; SANDY-NEXT: vcvtsi2ssq (%rsi), %xmm1, %xmm1 # sched: [10:1.00]
; SANDY-NEXT: vaddss %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_cvtsi2ssq:
; HASWELL: # BB#0:
@ -515,10 +515,10 @@ define i32 @test_cvtss2si(float %a0, float *%a1) {
;
; SANDY-LABEL: test_cvtss2si:
; SANDY: # BB#0:
; SANDY-NEXT: vcvtss2si %xmm0, %ecx # sched: [3:1.00]
; SANDY-NEXT: vcvtss2si (%rdi), %eax # sched: [7:1.00]
; SANDY-NEXT: vcvtss2si %xmm0, %ecx # sched: [5:1.00]
; SANDY-NEXT: vcvtss2si (%rdi), %eax # sched: [10:1.00]
; SANDY-NEXT: addl %ecx, %eax # sched: [1:0.33]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_cvtss2si:
; HASWELL: # BB#0:
@ -567,10 +567,10 @@ define i64 @test_cvtss2siq(float %a0, float *%a1) {
;
; SANDY-LABEL: test_cvtss2siq:
; SANDY: # BB#0:
; SANDY-NEXT: vcvtss2si %xmm0, %rcx # sched: [3:1.00]
; SANDY-NEXT: vcvtss2si (%rdi), %rax # sched: [7:1.00]
; SANDY-NEXT: vcvtss2si %xmm0, %rcx # sched: [5:1.00]
; SANDY-NEXT: vcvtss2si (%rdi), %rax # sched: [10:1.00]
; SANDY-NEXT: addq %rcx, %rax # sched: [1:0.33]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_cvtss2siq:
; HASWELL: # BB#0:
@ -619,10 +619,10 @@ define i32 @test_cvttss2si(float %a0, float *%a1) {
;
; SANDY-LABEL: test_cvttss2si:
; SANDY: # BB#0:
; SANDY-NEXT: vcvttss2si %xmm0, %ecx # sched: [3:1.00]
; SANDY-NEXT: vcvttss2si (%rdi), %eax # sched: [7:1.00]
; SANDY-NEXT: vcvttss2si %xmm0, %ecx # sched: [5:1.00]
; SANDY-NEXT: vcvttss2si (%rdi), %eax # sched: [10:1.00]
; SANDY-NEXT: addl %ecx, %eax # sched: [1:0.33]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_cvttss2si:
; HASWELL: # BB#0:
@ -668,10 +668,10 @@ define i64 @test_cvttss2siq(float %a0, float *%a1) {
;
; SANDY-LABEL: test_cvttss2siq:
; SANDY: # BB#0:
; SANDY-NEXT: vcvttss2si %xmm0, %rcx # sched: [3:1.00]
; SANDY-NEXT: vcvttss2si (%rdi), %rax # sched: [7:1.00]
; SANDY-NEXT: vcvttss2si %xmm0, %rcx # sched: [5:1.00]
; SANDY-NEXT: vcvttss2si (%rdi), %rax # sched: [10:1.00]
; SANDY-NEXT: addq %rcx, %rax # sched: [1:0.33]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_cvttss2siq:
; HASWELL: # BB#0:
@ -714,9 +714,9 @@ define <4 x float> @test_divps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a
;
; SANDY-LABEL: test_divps:
; SANDY: # BB#0:
; SANDY-NEXT: vdivps %xmm1, %xmm0, %xmm0 # sched: [12:1.00]
; SANDY-NEXT: vdivps (%rdi), %xmm0, %xmm0 # sched: [16:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vdivps %xmm1, %xmm0, %xmm0 # sched: [14:1.00]
; SANDY-NEXT: vdivps (%rdi), %xmm0, %xmm0 # sched: [20:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_divps:
; HASWELL: # BB#0:
@ -756,9 +756,9 @@ define float @test_divss(float %a0, float %a1, float *%a2) {
;
; SANDY-LABEL: test_divss:
; SANDY: # BB#0:
; SANDY-NEXT: vdivss %xmm1, %xmm0, %xmm0 # sched: [12:1.00]
; SANDY-NEXT: vdivss (%rdi), %xmm0, %xmm0 # sched: [16:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vdivss %xmm1, %xmm0, %xmm0 # sched: [14:1.00]
; SANDY-NEXT: vdivss (%rdi), %xmm0, %xmm0 # sched: [20:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_divss:
; HASWELL: # BB#0:
@ -799,8 +799,8 @@ define void @test_ldmxcsr(i32 %a0) {
; SANDY-LABEL: test_ldmxcsr:
; SANDY: # BB#0:
; SANDY-NEXT: movl %edi, -{{[0-9]+}}(%rsp) # sched: [1:1.00]
; SANDY-NEXT: vldmxcsr -{{[0-9]+}}(%rsp) # sched: [4:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vldmxcsr -{{[0-9]+}}(%rsp) # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_ldmxcsr:
; HASWELL: # BB#0:
@ -843,8 +843,8 @@ define <4 x float> @test_maxps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a
; SANDY-LABEL: test_maxps:
; SANDY: # BB#0:
; SANDY-NEXT: vmaxps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vmaxps (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vmaxps (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_maxps:
; HASWELL: # BB#0:
@ -886,8 +886,8 @@ define <4 x float> @test_maxss(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a
; SANDY-LABEL: test_maxss:
; SANDY: # BB#0:
; SANDY-NEXT: vmaxss %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vmaxss (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vmaxss (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_maxss:
; HASWELL: # BB#0:
@ -929,8 +929,8 @@ define <4 x float> @test_minps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a
; SANDY-LABEL: test_minps:
; SANDY: # BB#0:
; SANDY-NEXT: vminps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vminps (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vminps (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_minps:
; HASWELL: # BB#0:
@ -972,8 +972,8 @@ define <4 x float> @test_minss(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a
; SANDY-LABEL: test_minss:
; SANDY: # BB#0:
; SANDY-NEXT: vminss %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vminss (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vminss (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_minss:
; HASWELL: # BB#0:
@ -1017,10 +1017,10 @@ define void @test_movaps(<4 x float> *%a0, <4 x float> *%a1) {
;
; SANDY-LABEL: test_movaps:
; SANDY: # BB#0:
; SANDY-NEXT: vmovaps (%rdi), %xmm0 # sched: [4:0.50]
; SANDY-NEXT: vmovaps (%rdi), %xmm0 # sched: [6:0.50]
; SANDY-NEXT: vaddps %xmm0, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vmovaps %xmm0, (%rsi) # sched: [1:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vmovaps %xmm0, (%rsi) # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_movaps:
; HASWELL: # BB#0:
@ -1068,7 +1068,7 @@ define <4 x float> @test_movhlps(<4 x float> %a0, <4 x float> %a1) {
; SANDY-LABEL: test_movhlps:
; SANDY: # BB#0:
; SANDY-NEXT: vunpckhpd {{.*#+}} xmm0 = xmm1[1],xmm0[1] sched: [1:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_movhlps:
; HASWELL: # BB#0:
@ -1111,10 +1111,10 @@ define void @test_movhps(<4 x float> %a0, <4 x float> %a1, x86_mmx *%a2) {
;
; SANDY-LABEL: test_movhps:
; SANDY: # BB#0:
; SANDY-NEXT: vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0] sched: [5:1.00]
; SANDY-NEXT: vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0] sched: [7:1.00]
; SANDY-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vpextrq $1, %xmm0, (%rdi) # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_movhps:
; HASWELL: # BB#0:
@ -1164,7 +1164,7 @@ define <4 x float> @test_movlhps(<4 x float> %a0, <4 x float> %a1) {
; SANDY: # BB#0:
; SANDY-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0] sched: [1:1.00]
; SANDY-NEXT: vaddps %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_movlhps:
; HASWELL: # BB#0:
@ -1206,10 +1206,10 @@ define void @test_movlps(<4 x float> %a0, <4 x float> %a1, x86_mmx *%a2) {
;
; SANDY-LABEL: test_movlps:
; SANDY: # BB#0:
; SANDY-NEXT: vmovlpd {{.*#+}} xmm1 = mem[0],xmm1[1] sched: [5:1.00]
; SANDY-NEXT: vmovlpd {{.*#+}} xmm1 = mem[0],xmm1[1] sched: [7:1.00]
; SANDY-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vmovlps %xmm0, (%rdi) # sched: [1:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vmovlps %xmm0, (%rdi) # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_movlps:
; HASWELL: # BB#0:
@ -1254,8 +1254,8 @@ define i32 @test_movmskps(<4 x float> %a0) {
;
; SANDY-LABEL: test_movmskps:
; SANDY: # BB#0:
; SANDY-NEXT: vmovmskps %xmm0, %eax # sched: [1:0.33]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vmovmskps %xmm0, %eax # sched: [2:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_movmskps:
; HASWELL: # BB#0:
@ -1295,8 +1295,8 @@ define void @test_movntps(<4 x float> %a0, <4 x float> *%a1) {
;
; SANDY-LABEL: test_movntps:
; SANDY: # BB#0:
; SANDY-NEXT: vmovntps %xmm0, (%rdi) # sched: [1:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vmovntps %xmm0, (%rdi) # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_movntps:
; HASWELL: # BB#0:
@ -1335,10 +1335,10 @@ define void @test_movss_mem(float* %a0, float* %a1) {
;
; SANDY-LABEL: test_movss_mem:
; SANDY: # BB#0:
; SANDY-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero sched: [4:0.50]
; SANDY-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero sched: [6:0.50]
; SANDY-NEXT: vaddss %xmm0, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vmovss %xmm0, (%rsi) # sched: [1:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vmovss %xmm0, (%rsi) # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_movss_mem:
; HASWELL: # BB#0:
@ -1383,8 +1383,8 @@ define <4 x float> @test_movss_reg(<4 x float> %a0, <4 x float> %a1) {
;
; SANDY-LABEL: test_movss_reg:
; SANDY: # BB#0:
; SANDY-NEXT: vblendps {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3] sched: [1:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vblendps {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3] sched: [1:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_movss_reg:
; HASWELL: # BB#0:
@ -1423,10 +1423,10 @@ define void @test_movups(<4 x float> *%a0, <4 x float> *%a1) {
;
; SANDY-LABEL: test_movups:
; SANDY: # BB#0:
; SANDY-NEXT: vmovups (%rdi), %xmm0 # sched: [4:0.50]
; SANDY-NEXT: vmovups (%rdi), %xmm0 # sched: [6:0.50]
; SANDY-NEXT: vaddps %xmm0, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vmovups %xmm0, (%rsi) # sched: [1:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vmovups %xmm0, (%rsi) # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_movups:
; HASWELL: # BB#0:
@ -1469,8 +1469,8 @@ define <4 x float> @test_mulps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a
; SANDY-LABEL: test_mulps:
; SANDY: # BB#0:
; SANDY-NEXT: vmulps %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
; SANDY-NEXT: vmulps (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vmulps (%rdi), %xmm0, %xmm0 # sched: [11:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_mulps:
; HASWELL: # BB#0:
@ -1511,8 +1511,8 @@ define float @test_mulss(float %a0, float %a1, float *%a2) {
; SANDY-LABEL: test_mulss:
; SANDY: # BB#0:
; SANDY-NEXT: vmulss %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
; SANDY-NEXT: vmulss (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vmulss (%rdi), %xmm0, %xmm0 # sched: [11:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_mulss:
; HASWELL: # BB#0:
@ -1560,9 +1560,9 @@ define <4 x float> @test_orps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a2
;
; SANDY-LABEL: test_orps:
; SANDY: # BB#0:
; SANDY-NEXT: vorps %xmm1, %xmm0, %xmm0 # sched: [1:0.33]
; SANDY-NEXT: vorps (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vorps %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
; SANDY-NEXT: vorps (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_orps:
; HASWELL: # BB#0:
@ -1609,8 +1609,8 @@ define void @test_prefetchnta(i8* %a0) {
;
; SANDY-LABEL: test_prefetchnta:
; SANDY: # BB#0:
; SANDY-NEXT: prefetchnta (%rdi) # sched: [4:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: prefetchnta (%rdi) # sched: [5:0.50]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_prefetchnta:
; HASWELL: # BB#0:
@ -1652,10 +1652,10 @@ define <4 x float> @test_rcpps(<4 x float> %a0, <4 x float> *%a1) {
;
; SANDY-LABEL: test_rcpps:
; SANDY: # BB#0:
; SANDY-NEXT: vrcpps %xmm0, %xmm0 # sched: [5:1.00]
; SANDY-NEXT: vrcpps (%rdi), %xmm1 # sched: [9:1.00]
; SANDY-NEXT: vrcpps %xmm0, %xmm0 # sched: [7:3.00]
; SANDY-NEXT: vrcpps (%rdi), %xmm1 # sched: [11:1.00]
; SANDY-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_rcpps:
; HASWELL: # BB#0:
@ -1708,10 +1708,10 @@ define <4 x float> @test_rcpss(float %a0, float *%a1) {
; SANDY-LABEL: test_rcpss:
; SANDY: # BB#0:
; SANDY-NEXT: vrcpss %xmm0, %xmm0, %xmm0 # sched: [9:1.00]
; SANDY-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero sched: [4:0.50]
; SANDY-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero sched: [6:0.50]
; SANDY-NEXT: vrcpss %xmm1, %xmm1, %xmm1 # sched: [9:1.00]
; SANDY-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_rcpss:
; HASWELL: # BB#0:
@ -1765,9 +1765,9 @@ define <4 x float> @test_rsqrtps(<4 x float> %a0, <4 x float> *%a1) {
; SANDY-LABEL: test_rsqrtps:
; SANDY: # BB#0:
; SANDY-NEXT: vrsqrtps %xmm0, %xmm0 # sched: [5:1.00]
; SANDY-NEXT: vrsqrtps (%rdi), %xmm1 # sched: [9:1.00]
; SANDY-NEXT: vrsqrtps (%rdi), %xmm1 # sched: [11:1.00]
; SANDY-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_rsqrtps:
; HASWELL: # BB#0:
@ -1819,11 +1819,11 @@ define <4 x float> @test_rsqrtss(float %a0, float *%a1) {
;
; SANDY-LABEL: test_rsqrtss:
; SANDY: # BB#0:
; SANDY-NEXT: vrsqrtss %xmm0, %xmm0, %xmm0 # sched: [9:1.00]
; SANDY-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero sched: [4:0.50]
; SANDY-NEXT: vrsqrtss %xmm1, %xmm1, %xmm1 # sched: [9:1.00]
; SANDY-NEXT: vrsqrtss %xmm0, %xmm0, %xmm0 # sched: [5:1.00]
; SANDY-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero sched: [6:0.50]
; SANDY-NEXT: vrsqrtss %xmm1, %xmm1, %xmm1 # sched: [5:1.00]
; SANDY-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_rsqrtss:
; HASWELL: # BB#0:
@ -1875,7 +1875,7 @@ define void @test_sfence() {
; SANDY-LABEL: test_sfence:
; SANDY: # BB#0:
; SANDY-NEXT: sfence # sched: [1:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_sfence:
; HASWELL: # BB#0:
@ -1917,8 +1917,8 @@ define <4 x float> @test_shufps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%
; SANDY-LABEL: test_shufps:
; SANDY: # BB#0:
; SANDY-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[0,0] sched: [1:1.00]
; SANDY-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,3],mem[0,0] sched: [5:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,3],mem[0,0] sched: [7:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_shufps:
; HASWELL: # BB#0:
@ -1962,10 +1962,10 @@ define <4 x float> @test_sqrtps(<4 x float> %a0, <4 x float> *%a1) {
;
; SANDY-LABEL: test_sqrtps:
; SANDY: # BB#0:
; SANDY-NEXT: vsqrtps %xmm0, %xmm0 # sched: [15:1.00]
; SANDY-NEXT: vsqrtps (%rdi), %xmm1 # sched: [19:1.00]
; SANDY-NEXT: vsqrtps %xmm0, %xmm0 # sched: [14:1.00]
; SANDY-NEXT: vsqrtps (%rdi), %xmm1 # sched: [20:1.00]
; SANDY-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_sqrtps:
; HASWELL: # BB#0:
@ -2017,11 +2017,11 @@ define <4 x float> @test_sqrtss(<4 x float> %a0, <4 x float> *%a1) {
;
; SANDY-LABEL: test_sqrtss:
; SANDY: # BB#0:
; SANDY-NEXT: vsqrtss %xmm0, %xmm0, %xmm0 # sched: [19:1.00]
; SANDY-NEXT: vmovaps (%rdi), %xmm1 # sched: [4:0.50]
; SANDY-NEXT: vsqrtss %xmm1, %xmm1, %xmm1 # sched: [19:1.00]
; SANDY-NEXT: vsqrtss %xmm0, %xmm0, %xmm0 # sched: [114:1.00]
; SANDY-NEXT: vmovaps (%rdi), %xmm1 # sched: [6:0.50]
; SANDY-NEXT: vsqrtss %xmm1, %xmm1, %xmm1 # sched: [114:1.00]
; SANDY-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_sqrtss:
; HASWELL: # BB#0:
@ -2067,9 +2067,9 @@ define i32 @test_stmxcsr() {
;
; SANDY-LABEL: test_stmxcsr:
; SANDY: # BB#0:
; SANDY-NEXT: vstmxcsr -{{[0-9]+}}(%rsp) # sched: [1:1.00]
; SANDY-NEXT: movl -{{[0-9]+}}(%rsp), %eax # sched: [4:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vstmxcsr -{{[0-9]+}}(%rsp) # sched: [5:1.00]
; SANDY-NEXT: movl -{{[0-9]+}}(%rsp), %eax # sched: [5:0.50]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_stmxcsr:
; HASWELL: # BB#0:
@ -2112,8 +2112,8 @@ define <4 x float> @test_subps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a
; SANDY-LABEL: test_subps:
; SANDY: # BB#0:
; SANDY-NEXT: vsubps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vsubps (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vsubps (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_subps:
; HASWELL: # BB#0:
@ -2154,8 +2154,8 @@ define float @test_subss(float %a0, float %a1, float *%a2) {
; SANDY-LABEL: test_subss:
; SANDY: # BB#0:
; SANDY-NEXT: vsubss %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vsubss (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vsubss (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_subss:
; HASWELL: # BB#0:
@ -2220,16 +2220,16 @@ define i32 @test_ucomiss(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a2) {
; SANDY-LABEL: test_ucomiss:
; SANDY: # BB#0:
; SANDY-NEXT: vucomiss %xmm1, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: setnp %al # sched: [1:0.33]
; SANDY-NEXT: sete %cl # sched: [1:0.33]
; SANDY-NEXT: setnp %al # sched: [1:1.00]
; SANDY-NEXT: sete %cl # sched: [1:1.00]
; SANDY-NEXT: andb %al, %cl # sched: [1:0.33]
; SANDY-NEXT: vucomiss (%rdi), %xmm0 # sched: [7:1.00]
; SANDY-NEXT: setnp %al # sched: [1:0.33]
; SANDY-NEXT: sete %dl # sched: [1:0.33]
; SANDY-NEXT: setnp %al # sched: [1:1.00]
; SANDY-NEXT: sete %dl # sched: [1:1.00]
; SANDY-NEXT: andb %al, %dl # sched: [1:0.33]
; SANDY-NEXT: orb %cl, %dl # sched: [1:0.33]
; SANDY-NEXT: movzbl %dl, %eax # sched: [1:0.33]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_ucomiss:
; HASWELL: # BB#0:
@ -2292,8 +2292,8 @@ define <4 x float> @test_unpckhps(<4 x float> %a0, <4 x float> %a1, <4 x float>
; SANDY-LABEL: test_unpckhps:
; SANDY: # BB#0:
; SANDY-NEXT: vunpckhps {{.*#+}} xmm0 = xmm0[2],xmm1[2],xmm0[3],xmm1[3] sched: [1:1.00]
; SANDY-NEXT: vunpckhps {{.*#+}} xmm0 = xmm0[2],mem[2],xmm0[3],mem[3] sched: [5:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vunpckhps {{.*#+}} xmm0 = xmm0[2],mem[2],xmm0[3],mem[3] sched: [7:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_unpckhps:
; HASWELL: # BB#0:
@ -2338,8 +2338,8 @@ define <4 x float> @test_unpcklps(<4 x float> %a0, <4 x float> %a1, <4 x float>
; SANDY-LABEL: test_unpcklps:
; SANDY: # BB#0:
; SANDY-NEXT: vunpcklps {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1] sched: [1:1.00]
; SANDY-NEXT: vunpcklps {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[1],mem[1] sched: [5:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vunpcklps {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[1],mem[1] sched: [7:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_unpcklps:
; HASWELL: # BB#0:
@ -2387,9 +2387,9 @@ define <4 x float> @test_xorps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a
;
; SANDY-LABEL: test_xorps:
; SANDY: # BB#0:
; SANDY-NEXT: vxorps %xmm1, %xmm0, %xmm0 # sched: [1:0.33]
; SANDY-NEXT: vxorps (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vxorps %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
; SANDY-NEXT: vxorps (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_xorps:
; HASWELL: # BB#0:

598
test/CodeGen/X86/sse2-schedule.ll
View File

File diff suppressed because it is too large Load Diff

48
test/CodeGen/X86/sse3-schedule.ll
View File

@ -31,8 +31,8 @@ define <2 x double> @test_addsubpd(<2 x double> %a0, <2 x double> %a1, <2 x doub
; SANDY-LABEL: test_addsubpd:
; SANDY: # BB#0:
; SANDY-NEXT: vaddsubpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vaddsubpd (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vaddsubpd (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_addsubpd:
; HASWELL: # BB#0:
@ -74,8 +74,8 @@ define <4 x float> @test_addsubps(<4 x float> %a0, <4 x float> %a1, <4 x float>
; SANDY-LABEL: test_addsubps:
; SANDY: # BB#0:
; SANDY-NEXT: vaddsubps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vaddsubps (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vaddsubps (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_addsubps:
; HASWELL: # BB#0:
@ -116,9 +116,9 @@ define <2 x double> @test_haddpd(<2 x double> %a0, <2 x double> %a1, <2 x double
;
; SANDY-LABEL: test_haddpd:
; SANDY: # BB#0:
; SANDY-NEXT: vhaddpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vhaddpd (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vhaddpd %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
; SANDY-NEXT: vhaddpd (%rdi), %xmm0, %xmm0 # sched: [11:2.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_haddpd:
; HASWELL: # BB#0:
@ -159,9 +159,9 @@ define <4 x float> @test_haddps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%
;
; SANDY-LABEL: test_haddps:
; SANDY: # BB#0:
; SANDY-NEXT: vhaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vhaddps (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vhaddps %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
; SANDY-NEXT: vhaddps (%rdi), %xmm0, %xmm0 # sched: [11:2.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_haddps:
; HASWELL: # BB#0:
@ -202,9 +202,9 @@ define <2 x double> @test_hsubpd(<2 x double> %a0, <2 x double> %a1, <2 x double
;
; SANDY-LABEL: test_hsubpd:
; SANDY: # BB#0:
; SANDY-NEXT: vhsubpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vhsubpd (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vhsubpd %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
; SANDY-NEXT: vhsubpd (%rdi), %xmm0, %xmm0 # sched: [11:2.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_hsubpd:
; HASWELL: # BB#0:
@ -245,9 +245,9 @@ define <4 x float> @test_hsubps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%
;
; SANDY-LABEL: test_hsubps:
; SANDY: # BB#0:
; SANDY-NEXT: vhsubps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vhsubps (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vhsubps %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
; SANDY-NEXT: vhsubps (%rdi), %xmm0, %xmm0 # sched: [11:2.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_hsubps:
; HASWELL: # BB#0:
@ -287,8 +287,8 @@ define <16 x i8> @test_lddqu(i8* %a0) {
;
; SANDY-LABEL: test_lddqu:
; SANDY: # BB#0:
; SANDY-NEXT: vlddqu (%rdi), %xmm0 # sched: [4:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vlddqu (%rdi), %xmm0 # sched: [6:0.50]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_lddqu:
; HASWELL: # BB#0:
@ -330,9 +330,9 @@ define <2 x double> @test_movddup(<2 x double> %a0, <2 x double> *%a1) {
; SANDY-LABEL: test_movddup:
; SANDY: # BB#0:
; SANDY-NEXT: vmovddup {{.*#+}} xmm0 = xmm0[0,0] sched: [1:1.00]
; SANDY-NEXT: vmovddup {{.*#+}} xmm1 = mem[0,0] sched: [4:0.50]
; SANDY-NEXT: vmovddup {{.*#+}} xmm1 = mem[0,0] sched: [6:0.50]
; SANDY-NEXT: vaddpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_movddup:
; HASWELL: # BB#0:
@ -380,9 +380,9 @@ define <4 x float> @test_movshdup(<4 x float> %a0, <4 x float> *%a1) {
; SANDY-LABEL: test_movshdup:
; SANDY: # BB#0:
; SANDY-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3] sched: [1:1.00]
; SANDY-NEXT: vmovshdup {{.*#+}} xmm1 = mem[1,1,3,3] sched: [4:0.50]
; SANDY-NEXT: vmovshdup {{.*#+}} xmm1 = mem[1,1,3,3] sched: [6:0.50]
; SANDY-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_movshdup:
; HASWELL: # BB#0:
@ -430,9 +430,9 @@ define <4 x float> @test_movsldup(<4 x float> %a0, <4 x float> *%a1) {
; SANDY-LABEL: test_movsldup:
; SANDY: # BB#0:
; SANDY-NEXT: vmovsldup {{.*#+}} xmm0 = xmm0[0,0,2,2] sched: [1:1.00]
; SANDY-NEXT: vmovsldup {{.*#+}} xmm1 = mem[0,0,2,2] sched: [4:0.50]
; SANDY-NEXT: vmovsldup {{.*#+}} xmm1 = mem[0,0,2,2] sched: [6:0.50]
; SANDY-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_movsldup:
; HASWELL: # BB#0:

222
test/CodeGen/X86/sse41-schedule.ll
View File

@ -25,10 +25,10 @@ define <2 x double> @test_blendpd(<2 x double> %a0, <2 x double> %a1, <2 x doubl
;
; SANDY-LABEL: test_blendpd:
; SANDY: # BB#0:
; SANDY-NEXT: vblendpd {{.*#+}} xmm0 = xmm0[0],xmm1[1] sched: [1:0.50]
; SANDY-NEXT: vblendpd {{.*#+}} xmm0 = xmm0[0],xmm1[1] sched: [1:1.00]
; 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]
; SANDY-NEXT: vblendpd {{.*#+}} xmm0 = xmm0[0],mem[1] sched: [7:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_blendpd:
; HASWELL: # BB#0:
@ -65,9 +65,9 @@ define <4 x float> @test_blendps(<4 x float> %a0, <4 x float> %a1, <4 x float> *
;
; 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]
; SANDY-NEXT: vblendps {{.*#+}} xmm0 = xmm0[0],xmm1[1,2],xmm0[3] sched: [1:1.00]
; SANDY-NEXT: vblendps {{.*#+}} xmm0 = xmm0[0],mem[1],xmm0[2,3] sched: [7:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_blendps:
; HASWELL: # BB#0:
@ -107,9 +107,9 @@ define <2 x double> @test_blendvpd(<2 x double> %a0, <2 x double> %a1, <2 x doub
;
; 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]
; SANDY-NEXT: vblendvpd %xmm2, %xmm1, %xmm0, %xmm0 # sched: [2:2.00]
; SANDY-NEXT: vblendvpd %xmm2, (%rdi), %xmm0, %xmm0 # sched: [8:2.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_blendvpd:
; HASWELL: # BB#0:
@ -150,9 +150,9 @@ define <4 x float> @test_blendvps(<4 x float> %a0, <4 x float> %a1, <4 x float>
;
; 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]
; SANDY-NEXT: vblendvps %xmm2, %xmm1, %xmm0, %xmm0 # sched: [2:2.00]
; SANDY-NEXT: vblendvps %xmm2, (%rdi), %xmm0, %xmm0 # sched: [8:2.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_blendvps:
; HASWELL: # BB#0:
@ -187,9 +187,9 @@ define <2 x double> @test_dppd(<2 x double> %a0, <2 x double> %a1, <2 x double>
;
; 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]
; SANDY-NEXT: vdppd $7, %xmm1, %xmm0, %xmm0 # sched: [9:1.00]
; SANDY-NEXT: vdppd $7, (%rdi), %xmm0, %xmm0 # sched: [15:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_dppd:
; HASWELL: # BB#0:
@ -224,9 +224,9 @@ define <4 x float> @test_dpps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a2
;
; SANDY-LABEL: test_dpps:
; SANDY: # BB#0:
; SANDY-NEXT: vdpps $7, %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vdpps $7, %xmm1, %xmm0, %xmm0 # sched: [12:2.00]
; SANDY-NEXT: vdpps $7, (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_dpps:
; HASWELL: # BB#0:
@ -262,8 +262,8 @@ define <4 x float> @test_insertps(<4 x float> %a0, <4 x float> %a1, float *%a2)
; 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]
; SANDY-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0,1,2],mem[0] sched: [7:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_insertps:
; HASWELL: # BB#0:
@ -296,8 +296,8 @@ define <2 x i64> @test_movntdqa(i8* %a0) {
;
; SANDY-LABEL: test_movntdqa:
; SANDY: # BB#0:
; SANDY-NEXT: vmovntdqa (%rdi), %xmm0 # sched: [4:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vmovntdqa (%rdi), %xmm0 # sched: [6:0.50]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_movntdqa:
; HASWELL: # BB#0:
@ -328,9 +328,9 @@ define <8 x i16> @test_mpsadbw(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) {
;
; 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]
; SANDY-NEXT: vmpsadbw $7, %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
; SANDY-NEXT: vmpsadbw $7, (%rdi), %xmm0, %xmm0 # sched: [11:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_mpsadbw:
; HASWELL: # BB#0:
@ -367,8 +367,8 @@ define <8 x i16> @test_packusdw(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) {
; 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]
; SANDY-NEXT: vpackusdw (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_packusdw:
; HASWELL: # BB#0:
@ -411,8 +411,8 @@ define <16 x i8> @test_pblendvb(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> %a2, <16
; 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]
; SANDY-NEXT: vpblendvb %xmm2, (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pblendvb:
; HASWELL: # BB#0:
@ -448,8 +448,8 @@ define <8 x i16> @test_pblendw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) {
; 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]
; SANDY-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],mem[2,3],xmm0[4,5,6],mem[7] sched: [7:0.50]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pblendw:
; HASWELL: # BB#0:
@ -483,9 +483,9 @@ define <2 x i64> @test_pcmpeqq(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> *%a2) {
;
; 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]
; SANDY-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vpcmpeqq (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pcmpeqq:
; HASWELL: # BB#0:
@ -521,9 +521,9 @@ define i32 @test_pextrb(<16 x i8> %a0, i8 *%a1) {
;
; SANDY-LABEL: test_pextrb:
; SANDY: # BB#0:
; SANDY-NEXT: vpextrb $3, %xmm0, %eax # sched: [1:0.50]
; SANDY-NEXT: vpextrb $3, %xmm0, %eax # sched: [3:1.00]
; SANDY-NEXT: vpextrb $1, %xmm0, (%rdi) # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pextrb:
; HASWELL: # BB#0:
@ -558,9 +558,9 @@ define i32 @test_pextrd(<4 x i32> %a0, i32 *%a1) {
;
; SANDY-LABEL: test_pextrd:
; SANDY: # BB#0:
; SANDY-NEXT: vpextrd $3, %xmm0, %eax # sched: [1:0.50]
; SANDY-NEXT: vpextrd $3, %xmm0, %eax # sched: [3:1.00]
; SANDY-NEXT: vpextrd $1, %xmm0, (%rdi) # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pextrd:
; HASWELL: # BB#0:
@ -594,9 +594,9 @@ define i64 @test_pextrq(<2 x i64> %a0, <2 x i64> %a1, i64 *%a2) {
;
; SANDY-LABEL: test_pextrq:
; SANDY: # BB#0:
; SANDY-NEXT: vpextrq $1, %xmm0, %rax # sched: [1:0.50]
; SANDY-NEXT: vpextrq $1, %xmm0, %rax # sched: [3:1.00]
; SANDY-NEXT: vpextrq $1, %xmm0, (%rdi) # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pextrq:
; HASWELL: # BB#0:
@ -630,9 +630,9 @@ define i32 @test_pextrw(<8 x i16> %a0, i16 *%a1) {
;
; SANDY-LABEL: test_pextrw:
; SANDY: # BB#0:
; SANDY-NEXT: vpextrw $3, %xmm0, %eax # sched: [1:0.50]
; SANDY-NEXT: vpextrw $3, %xmm0, %eax # sched: [3:1.00]
; SANDY-NEXT: vpextrw $1, %xmm0, (%rdi) # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pextrw:
; HASWELL: # BB#0:
@ -667,9 +667,9 @@ define <8 x i16> @test_phminposuw(<8 x i16> *%a0) {
;
; SANDY-LABEL: test_phminposuw:
; SANDY: # BB#0:
; SANDY-NEXT: vphminposuw (%rdi), %xmm0 # sched: [9:1.00]
; SANDY-NEXT: vphminposuw (%rdi), %xmm0 # sched: [11:1.00]
; SANDY-NEXT: vphminposuw %xmm0, %xmm0 # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_phminposuw:
; HASWELL: # BB#0:
@ -704,9 +704,9 @@ define <16 x i8> @test_pinsrb(<16 x i8> %a0, i8 %a1, i8 *%a2) {
;
; 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]
; SANDY-NEXT: vpinsrb $1, %edi, %xmm0, %xmm0 # sched: [2:1.00]
; SANDY-NEXT: vpinsrb $3, (%rsi), %xmm0, %xmm0 # sched: [7:0.50]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pinsrb:
; HASWELL: # BB#0:
@ -740,9 +740,9 @@ define <4 x i32> @test_pinsrd(<4 x i32> %a0, i32 %a1, i32 *%a2) {
;
; 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]
; SANDY-NEXT: vpinsrd $1, %edi, %xmm0, %xmm0 # sched: [2:1.00]
; SANDY-NEXT: vpinsrd $3, (%rsi), %xmm0, %xmm0 # sched: [7:0.50]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pinsrd:
; HASWELL: # BB#0:
@ -778,10 +778,10 @@ define <2 x i64> @test_pinsrq(<2 x i64> %a0, <2 x i64> %a1, i64 %a2, i64 *%a3) {
;
; 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: vpinsrq $1, %rdi, %xmm0, %xmm0 # sched: [2:1.00]
; SANDY-NEXT: vpinsrq $1, (%rsi), %xmm1, %xmm1 # sched: [7:0.50]
; SANDY-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pinsrq:
; HASWELL: # BB#0:
@ -819,8 +819,8 @@ define <16 x i8> @test_pmaxsb(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) {
; 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]
; SANDY-NEXT: vpmaxsb (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pmaxsb:
; HASWELL: # BB#0:
@ -856,8 +856,8 @@ define <4 x i32> @test_pmaxsd(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) {
; 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]
; SANDY-NEXT: vpmaxsd (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pmaxsd:
; HASWELL: # BB#0:
@ -893,8 +893,8 @@ define <4 x i32> @test_pmaxud(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) {
; 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]
; SANDY-NEXT: vpmaxud (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pmaxud:
; HASWELL: # BB#0:
@ -930,8 +930,8 @@ define <8 x i16> @test_pmaxuw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) {
; 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]
; SANDY-NEXT: vpmaxuw (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pmaxuw:
; HASWELL: # BB#0:
@ -967,8 +967,8 @@ define <16 x i8> @test_pminsb(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) {
; 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]
; SANDY-NEXT: vpminsb (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pminsb:
; HASWELL: # BB#0:
@ -1004,8 +1004,8 @@ define <4 x i32> @test_pminsd(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) {
; 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]
; SANDY-NEXT: vpminsd (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pminsd:
; HASWELL: # BB#0:
@ -1041,8 +1041,8 @@ define <4 x i32> @test_pminud(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) {
; 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]
; SANDY-NEXT: vpminud (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pminud:
; HASWELL: # BB#0:
@ -1078,8 +1078,8 @@ define <8 x i16> @test_pminuw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) {
; 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]
; SANDY-NEXT: vpminuw (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pminuw:
; HASWELL: # BB#0:
@ -1118,9 +1118,9 @@ define <8 x i16> @test_pmovsxbw(<16 x i8> %a0, <8 x i8> *%a1) {
; 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]
; SANDY-NEXT: vpmovsxbw (%rdi), %xmm1 # sched: [7:0.50]
; SANDY-NEXT: vpaddw %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pmovsxbw:
; HASWELL: # BB#0:
@ -1162,9 +1162,9 @@ define <4 x i32> @test_pmovsxbd(<16 x i8> %a0, <4 x i8> *%a1) {
; 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: vpmovsxbd (%rdi), %xmm1 # sched: [7:0.50]
; SANDY-NEXT: vpaddd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pmovsxbd:
; HASWELL: # BB#0:
@ -1206,9 +1206,9 @@ define <2 x i64> @test_pmovsxbq(<16 x i8> %a0, <2 x i8> *%a1) {
; 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: vpmovsxbq (%rdi), %xmm1 # sched: [7:0.50]
; SANDY-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pmovsxbq:
; HASWELL: # BB#0:
@ -1250,9 +1250,9 @@ define <2 x i64> @test_pmovsxdq(<4 x i32> %a0, <2 x i32> *%a1) {
; 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: vpmovsxdq (%rdi), %xmm1 # sched: [7:0.50]
; SANDY-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pmovsxdq:
; HASWELL: # BB#0:
@ -1294,9 +1294,9 @@ define <4 x i32> @test_pmovsxwd(<8 x i16> %a0, <4 x i16> *%a1) {
; 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: vpmovsxwd (%rdi), %xmm1 # sched: [7:0.50]
; SANDY-NEXT: vpaddd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pmovsxwd:
; HASWELL: # BB#0:
@ -1338,9 +1338,9 @@ define <2 x i64> @test_pmovsxwq(<8 x i16> %a0, <2 x i16> *%a1) {
; 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: vpmovsxwq (%rdi), %xmm1 # sched: [7:0.50]
; SANDY-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pmovsxwq:
; HASWELL: # BB#0:
@ -1382,9 +1382,9 @@ define <8 x i16> @test_pmovzxbw(<16 x i8> %a0, <8 x i8> *%a1) {
; 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]
; 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: [7:0.50]
; SANDY-NEXT: vpaddw %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pmovzxbw:
; HASWELL: # BB#0:
@ -1426,9 +1426,9 @@ define <4 x i32> @test_pmovzxbd(<16 x i8> %a0, <4 x i8> *%a1) {
; 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: vpmovzxbd {{.*#+}} xmm1 = mem[0],zero,zero,zero,mem[1],zero,zero,zero,mem[2],zero,zero,zero,mem[3],zero,zero,zero sched: [7:0.50]
; SANDY-NEXT: vpaddd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pmovzxbd:
; HASWELL: # BB#0:
@ -1470,9 +1470,9 @@ define <2 x i64> @test_pmovzxbq(<16 x i8> %a0, <2 x i8> *%a1) {
; 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: vpmovzxbq {{.*#+}} xmm1 = mem[0],zero,zero,zero,zero,zero,zero,zero,mem[1],zero,zero,zero,zero,zero,zero,zero sched: [7:0.50]
; SANDY-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pmovzxbq:
; HASWELL: # BB#0:
@ -1514,9 +1514,9 @@ define <2 x i64> @test_pmovzxdq(<4 x i32> %a0, <2 x i32> *%a1) {
; 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: vpmovzxdq {{.*#+}} xmm1 = mem[0],zero,mem[1],zero sched: [7:0.50]
; SANDY-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pmovzxdq:
; HASWELL: # BB#0:
@ -1558,9 +1558,9 @@ define <4 x i32> @test_pmovzxwd(<8 x i16> %a0, <4 x i16> *%a1) {
; 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: vpmovzxwd {{.*#+}} xmm1 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero sched: [7:0.50]
; SANDY-NEXT: vpaddd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pmovzxwd:
; HASWELL: # BB#0:
@ -1602,9 +1602,9 @@ define <2 x i64> @test_pmovzxwq(<8 x i16> %a0, <2 x i16> *%a1) {
; 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: vpmovzxwq {{.*#+}} xmm1 = mem[0],zero,zero,zero,mem[1],zero,zero,zero sched: [7:0.50]
; SANDY-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pmovzxwq:
; HASWELL: # BB#0:
@ -1642,9 +1642,9 @@ define <2 x i64> @test_pmuldq(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) {
;
; SANDY-LABEL: test_pmuldq:
; SANDY: # BB#0:
; SANDY-NEXT: vpmuldq %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
; SANDY-NEXT: vpmuldq %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vpmuldq (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pmuldq:
; HASWELL: # BB#0:
@ -1680,9 +1680,9 @@ define <4 x i32> @test_pmulld(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) {
;
; SANDY-LABEL: test_pmulld:
; SANDY: # BB#0:
; SANDY-NEXT: vpmulld %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
; SANDY-NEXT: vpmulld %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vpmulld (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pmulld:
; HASWELL: # BB#0:
@ -1724,13 +1724,13 @@ define i32 @test_ptest(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> *%a2) {
;
; 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: vptest %xmm1, %xmm0 # sched: [2:1.00]
; SANDY-NEXT: setb %al # sched: [1:1.00]
; SANDY-NEXT: vptest (%rdi), %xmm0 # sched: [8:1.00]
; SANDY-NEXT: setb %cl # sched: [1:1.00]
; 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]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_ptest:
; HASWELL: # BB#0:
@ -1778,9 +1778,9 @@ define <2 x double> @test_roundpd(<2 x double> %a0, <2 x double> *%a1) {
; 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: vroundpd $7, (%rdi), %xmm1 # sched: [9:1.00]
; SANDY-NEXT: vaddpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_roundpd:
; HASWELL: # BB#0:
@ -1822,9 +1822,9 @@ define <4 x float> @test_roundps(<4 x float> %a0, <4 x float> *%a1) {
; 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: vroundps $7, (%rdi), %xmm1 # sched: [9:1.00]
; SANDY-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_roundps:
; HASWELL: # BB#0:
@ -1867,9 +1867,9 @@ define <2 x double> @test_roundsd(<2 x double> %a0, <2 x double> %a1, <2 x doubl
; 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: vroundsd $7, (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
; SANDY-NEXT: vaddpd %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_roundsd:
; HASWELL: # BB#0:
@ -1912,9 +1912,9 @@ define <4 x float> @test_roundss(<4 x float> %a0, <4 x float> %a1, <4 x float> *
; 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: vroundss $7, (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
; SANDY-NEXT: vaddps %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_roundss:
; HASWELL: # BB#0:

38
test/CodeGen/X86/sse42-schedule.ll
View File

@ -26,9 +26,9 @@ define i32 @crc32_32_8(i32 %a0, i8 %a1, i8 *%a2) {
; SANDY-LABEL: crc32_32_8:
; SANDY: # BB#0:
; SANDY-NEXT: crc32b %sil, %edi # sched: [3:1.00]
; SANDY-NEXT: crc32b (%rdx), %edi # sched: [7:1.00]
; SANDY-NEXT: crc32b (%rdx), %edi # sched: [8:1.00]
; SANDY-NEXT: movl %edi, %eax # sched: [1:0.33]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: crc32_32_8:
; HASWELL: # BB#0:
@ -68,9 +68,9 @@ define i32 @crc32_32_16(i32 %a0, i16 %a1, i16 *%a2) {
; SANDY-LABEL: crc32_32_16:
; SANDY: # BB#0:
; SANDY-NEXT: crc32w %si, %edi # sched: [3:1.00]
; SANDY-NEXT: crc32w (%rdx), %edi # sched: [7:1.00]
; SANDY-NEXT: crc32w (%rdx), %edi # sched: [8:1.00]
; SANDY-NEXT: movl %edi, %eax # sched: [1:0.33]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: crc32_32_16:
; HASWELL: # BB#0:
@ -112,7 +112,7 @@ define i32 @crc32_32_32(i32 %a0, i32 %a1, i32 *%a2) {
; SANDY-NEXT: crc32l %esi, %edi # sched: [3:1.00]
; SANDY-NEXT: crc32l (%rdx), %edi # sched: [7:1.00]
; SANDY-NEXT: movl %edi, %eax # sched: [1:0.33]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: crc32_32_32:
; HASWELL: # BB#0:
@ -152,9 +152,9 @@ define i64 @crc32_64_8(i64 %a0, i8 %a1, i8 *%a2) nounwind {
; SANDY-LABEL: crc32_64_8:
; SANDY: # BB#0:
; SANDY-NEXT: crc32b %sil, %edi # sched: [3:1.00]
; SANDY-NEXT: crc32b (%rdx), %edi # sched: [7:1.00]
; SANDY-NEXT: crc32b (%rdx), %edi # sched: [8:1.00]
; SANDY-NEXT: movq %rdi, %rax # sched: [1:0.33]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: crc32_64_8:
; HASWELL: # BB#0:
@ -196,7 +196,7 @@ define i64 @crc32_64_64(i64 %a0, i64 %a1, i64 *%a2) {
; SANDY-NEXT: crc32q %rsi, %rdi # sched: [3:1.00]
; SANDY-NEXT: crc32q (%rdx), %rdi # sched: [7:1.00]
; SANDY-NEXT: movq %rdi, %rax # sched: [1:0.33]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: crc32_64_64:
; HASWELL: # BB#0:
@ -256,7 +256,7 @@ define i32 @test_pcmpestri(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) {
; SANDY-NEXT: vpcmpestri $7, (%rdi), %xmm0 # sched: [4:2.33]
; SANDY-NEXT: # kill: %ECX<def> %ECX<kill> %RCX<def>
; SANDY-NEXT: leal (%rcx,%rsi), %eax # sched: [1:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pcmpestri:
; HASWELL: # BB#0:
@ -320,7 +320,7 @@ define <16 x i8> @test_pcmpestrm(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) {
; SANDY-NEXT: movl $7, %eax # sched: [1:0.33]
; SANDY-NEXT: movl $7, %edx # sched: [1:0.33]
; SANDY-NEXT: vpcmpestrm $7, (%rdi), %xmm0 # sched: [11:2.33]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pcmpestrm:
; HASWELL: # BB#0:
@ -369,12 +369,12 @@ define i32 @test_pcmpistri(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) {
;
; SANDY-LABEL: test_pcmpistri:
; SANDY: # BB#0:
; SANDY-NEXT: vpcmpistri $7, %xmm1, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vpcmpistri $7, %xmm1, %xmm0 # sched: [11:3.00]
; SANDY-NEXT: movl %ecx, %eax # sched: [1:0.33]
; SANDY-NEXT: vpcmpistri $7, (%rdi), %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vpcmpistri $7, (%rdi), %xmm0 # sched: [17:3.00]
; SANDY-NEXT: # kill: %ECX<def> %ECX<kill> %RCX<def>
; SANDY-NEXT: leal (%rcx,%rax), %eax # sched: [1:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pcmpistri:
; HASWELL: # BB#0:
@ -416,9 +416,9 @@ define <16 x i8> @test_pcmpistrm(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) {
;
; SANDY-LABEL: test_pcmpistrm:
; SANDY: # BB#0:
; SANDY-NEXT: vpcmpistrm $7, %xmm1, %xmm0 # sched: [11:1.00]
; SANDY-NEXT: vpcmpistrm $7, (%rdi), %xmm0 # sched: [11:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vpcmpistrm $7, %xmm1, %xmm0 # sched: [11:3.00]
; SANDY-NEXT: vpcmpistrm $7, (%rdi), %xmm0 # sched: [17:3.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pcmpistrm:
; HASWELL: # BB#0:
@ -453,9 +453,9 @@ define <2 x i64> @test_pcmpgtq(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> *%a2) {
;
; SANDY-LABEL: test_pcmpgtq:
; SANDY: # BB#0:
; SANDY-NEXT: vpcmpgtq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; SANDY-NEXT: vpcmpgtq (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vpcmpgtq %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
; SANDY-NEXT: vpcmpgtq (%rdi), %xmm0, %xmm0 # sched: [11:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pcmpgtq:
; HASWELL: # BB#0:

74
test/CodeGen/X86/ssse3-schedule.ll
View File

@ -35,9 +35,9 @@ define <16 x i8> @test_pabsb(<16 x i8> %a0, <16 x i8> *%a1) {
; SANDY-LABEL: test_pabsb:
; SANDY: # BB#0:
; SANDY-NEXT: vpabsb %xmm0, %xmm0 # sched: [1:0.50]
; SANDY-NEXT: vpabsb (%rdi), %xmm1 # sched: [5:0.50]
; SANDY-NEXT: vpabsb (%rdi), %xmm1 # sched: [7:0.50]
; SANDY-NEXT: vpor %xmm1, %xmm0, %xmm0 # sched: [1:0.33]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pabsb:
; HASWELL: # BB#0:
@ -86,9 +86,9 @@ define <4 x i32> @test_pabsd(<4 x i32> %a0, <4 x i32> *%a1) {
; SANDY-LABEL: test_pabsd:
; SANDY: # BB#0:
; SANDY-NEXT: vpabsd %xmm0, %xmm0 # sched: [1:0.50]
; SANDY-NEXT: vpabsd (%rdi), %xmm1 # sched: [5:0.50]
; SANDY-NEXT: vpabsd (%rdi), %xmm1 # sched: [7:0.50]
; SANDY-NEXT: vpor %xmm1, %xmm0, %xmm0 # sched: [1:0.33]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pabsd:
; HASWELL: # BB#0:
@ -136,7 +136,7 @@ define <8 x i16> @test_pabsw(<8 x i16> %a0, <8 x i16> *%a1) {
; SANDY-LABEL: test_pabsw:
; SANDY: # BB#0:
; SANDY-NEXT: vpabsw %xmm0, %xmm0 # sched: [1:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pabsw:
; HASWELL: # BB#0:
@ -182,8 +182,8 @@ define <8 x i16> @test_palignr(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) {
; SANDY-LABEL: test_palignr:
; SANDY: # BB#0:
; SANDY-NEXT: vpalignr {{.*#+}} xmm0 = xmm0[6,7,8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5] sched: [1:0.50]
; SANDY-NEXT: vpalignr {{.*#+}} xmm0 = mem[14,15],xmm0[0,1,2,3,4,5,6,7,8,9,10,11,12,13] sched: [5:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vpalignr {{.*#+}} xmm0 = mem[14,15],xmm0[0,1,2,3,4,5,6,7,8,9,10,11,12,13] sched: [7:0.50]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_palignr:
; HASWELL: # BB#0:
@ -223,9 +223,9 @@ define <4 x i32> @test_phaddd(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) {
;
; SANDY-LABEL: test_phaddd:
; SANDY: # BB#0:
; SANDY-NEXT: vphaddd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; SANDY-NEXT: vphaddd (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vphaddd %xmm1, %xmm0, %xmm0 # sched: [3:1.50]
; SANDY-NEXT: vphaddd (%rdi), %xmm0, %xmm0 # sched: [9:1.50]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_phaddd:
; HASWELL: # BB#0:
@ -274,9 +274,9 @@ define <8 x i16> @test_phaddsw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) {
;
; SANDY-LABEL: test_phaddsw:
; SANDY: # BB#0:
; SANDY-NEXT: vphaddsw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; SANDY-NEXT: vphaddsw (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vphaddsw %xmm1, %xmm0, %xmm0 # sched: [3:1.50]
; SANDY-NEXT: vphaddsw (%rdi), %xmm0, %xmm0 # sched: [9:1.50]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_phaddsw:
; HASWELL: # BB#0:
@ -317,9 +317,9 @@ define <8 x i16> @test_phaddw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) {
;
; SANDY-LABEL: test_phaddw:
; SANDY: # BB#0:
; SANDY-NEXT: vphaddw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; SANDY-NEXT: vphaddw (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vphaddw %xmm1, %xmm0, %xmm0 # sched: [3:1.50]
; SANDY-NEXT: vphaddw (%rdi), %xmm0, %xmm0 # sched: [9:1.50]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_phaddw:
; HASWELL: # BB#0:
@ -360,9 +360,9 @@ define <4 x i32> @test_phsubd(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) {
;
; SANDY-LABEL: test_phsubd:
; SANDY: # BB#0:
; SANDY-NEXT: vphsubd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; SANDY-NEXT: vphsubd (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vphsubd %xmm1, %xmm0, %xmm0 # sched: [3:1.50]
; SANDY-NEXT: vphsubd (%rdi), %xmm0, %xmm0 # sched: [9:1.50]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_phsubd:
; HASWELL: # BB#0:
@ -411,9 +411,9 @@ define <8 x i16> @test_phsubsw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) {
;
; SANDY-LABEL: test_phsubsw:
; SANDY: # BB#0:
; SANDY-NEXT: vphsubsw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; SANDY-NEXT: vphsubsw (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vphsubsw %xmm1, %xmm0, %xmm0 # sched: [3:1.50]
; SANDY-NEXT: vphsubsw (%rdi), %xmm0, %xmm0 # sched: [9:1.50]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_phsubsw:
; HASWELL: # BB#0:
@ -454,9 +454,9 @@ define <8 x i16> @test_phsubw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) {
;
; SANDY-LABEL: test_phsubw:
; SANDY: # BB#0:
; SANDY-NEXT: vphsubw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; SANDY-NEXT: vphsubw (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vphsubw %xmm1, %xmm0, %xmm0 # sched: [3:1.50]
; SANDY-NEXT: vphsubw (%rdi), %xmm0, %xmm0 # sched: [9:1.50]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_phsubw:
; HASWELL: # BB#0:
@ -497,9 +497,9 @@ define <8 x i16> @test_pmaddubsw(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) {
;
; SANDY-LABEL: test_pmaddubsw:
; SANDY: # BB#0:
; SANDY-NEXT: vpmaddubsw %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
; SANDY-NEXT: vpmaddubsw %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vpmaddubsw (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pmaddubsw:
; HASWELL: # BB#0:
@ -538,8 +538,8 @@ define <8 x i16> @test_pmulhrsw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) {
;
; SANDY-LABEL: test_pmulhrsw:
; SANDY: # BB#0:
; SANDY-NEXT: vpmulhrsw %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vpmulhrsw %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pmulhrsw:
; HASWELL: # BB#0:
@ -579,8 +579,8 @@ define <16 x i8> @test_pshufb(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) {
; SANDY-LABEL: test_pshufb:
; SANDY: # BB#0:
; SANDY-NEXT: vpshufb %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; SANDY-NEXT: vpshufb (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vpshufb (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_pshufb:
; HASWELL: # BB#0:
@ -630,8 +630,8 @@ define <16 x i8> @test_psignb(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) {
; SANDY-LABEL: test_psignb:
; SANDY: # BB#0:
; SANDY-NEXT: vpsignb %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; SANDY-NEXT: vpsignb (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vpsignb (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_psignb:
; HASWELL: # BB#0:
@ -681,8 +681,8 @@ define <4 x i32> @test_psignd(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) {
; SANDY-LABEL: test_psignd:
; SANDY: # BB#0:
; SANDY-NEXT: vpsignd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; SANDY-NEXT: vpsignd (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vpsignd (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_psignd:
; HASWELL: # BB#0:
@ -732,8 +732,8 @@ define <8 x i16> @test_psignw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) {
; SANDY-LABEL: test_psignw:
; SANDY: # BB#0:
; SANDY-NEXT: vpsignw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; SANDY-NEXT: vpsignw (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
; SANDY-NEXT: retq # sched: [5:1.00]
; SANDY-NEXT: vpsignw (%rdi), %xmm0, %xmm0 # sched: [7:0.50]
; SANDY-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_psignw:
; HASWELL: # BB#0: