[X86][Broadwell] Added the instruction scheduling information for the Broadwell CPU.

Adding the scheduling information for the Browadwell (BDW) CPU target.

This patch adds the instruction scheduling information for the Broadwell (BDW) architecture target by adding the file X86SchedBroadwell.td located under the X86 Target.
We used the scheduling information retrieved from the Broadwell architects in order to create the file.
The scheduling information includes latency, number of micro-Ops and used ports by each BDW instruction.

The patch continues the scheduling replacement and insertion effort started with the SandyBridge (SNB) target in r310792, the Haswell (HSW) target in r311879, the SkylakeClient (SKL) target in rL313613 + rL315978 and the SkylakeServer (SKX) in rL315175.

Performance fluctuations may be expected due to code alignment effects.

Reviewers: zvi, RKSimon, craig.topper
Differential Revision: https://reviews.llvm.org/D39054

Change-Id: If6f799e5ff60e1091c8d43b05ea78c53581bae01
llvm-svn: 316492
This commit is contained in:
Gadi Haber 2017-10-24 20:19:47 +00:00
parent f8f4620dc6
commit 323f2e1715
22 changed files with 5450 additions and 1374 deletions

View File

@ -576,7 +576,7 @@ def BDWFeatures : ProcessorFeatures<HSWFeatures.Value, [
FeatureADX,
FeatureRDSEED
]>;
class BroadwellProc<string Name> : ProcModel<Name, HaswellModel,
class BroadwellProc<string Name> : ProcModel<Name, BroadwellModel,
BDWFeatures.Value, [
ProcIntelBDW
]>;

File diff suppressed because it is too large Load Diff

View File

@ -663,10 +663,10 @@ def GenericPostRAModel : GenericX86Model {
include "X86ScheduleAtom.td"
include "X86SchedSandyBridge.td"
include "X86SchedHaswell.td"
include "X86SchedBroadwell.td"
include "X86ScheduleSLM.td"
include "X86ScheduleZnver1.td"
include "X86ScheduleBtVer2.td"
include "X86SchedSkylakeClient.td"
include "X86SchedSkylakeServer.td"

View File

@ -38,8 +38,8 @@ define <2 x i64> @test_aesdec(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> *%a2) {
; BROADWELL-LABEL: test_aesdec:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vaesdec %xmm1, %xmm0, %xmm0 # sched: [7:1.00]
; BROADWELL-NEXT: vaesdec (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vaesdec (%rdi), %xmm0, %xmm0 # sched: [12:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_aesdec:
; SKYLAKE: # BB#0:
@ -93,8 +93,8 @@ define <2 x i64> @test_aesdeclast(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> *%a2)
; BROADWELL-LABEL: test_aesdeclast:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vaesdeclast %xmm1, %xmm0, %xmm0 # sched: [7:1.00]
; BROADWELL-NEXT: vaesdeclast (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vaesdeclast (%rdi), %xmm0, %xmm0 # sched: [12:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_aesdeclast:
; SKYLAKE: # BB#0:
@ -148,8 +148,8 @@ define <2 x i64> @test_aesenc(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> *%a2) {
; BROADWELL-LABEL: test_aesenc:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vaesenc %xmm1, %xmm0, %xmm0 # sched: [7:1.00]
; BROADWELL-NEXT: vaesenc (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vaesenc (%rdi), %xmm0, %xmm0 # sched: [12:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_aesenc:
; SKYLAKE: # BB#0:
@ -203,8 +203,8 @@ define <2 x i64> @test_aesenclast(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> *%a2)
; BROADWELL-LABEL: test_aesenclast:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vaesenclast %xmm1, %xmm0, %xmm0 # sched: [7:1.00]
; BROADWELL-NEXT: vaesenclast (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vaesenclast (%rdi), %xmm0, %xmm0 # sched: [12:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_aesenclast:
; SKYLAKE: # BB#0:
@ -262,9 +262,9 @@ define <2 x i64> @test_aesimc(<2 x i64> %a0, <2 x i64> *%a1) {
; BROADWELL-LABEL: test_aesimc:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vaesimc %xmm0, %xmm0 # sched: [14:2.00]
; BROADWELL-NEXT: vaesimc (%rdi), %xmm1 # sched: [14:2.00]
; BROADWELL-NEXT: vaesimc (%rdi), %xmm1 # sched: [19:2.00]
; BROADWELL-NEXT: vpor %xmm1, %xmm0, %xmm0 # sched: [1:0.33]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_aesimc:
; SKYLAKE: # BB#0:
@ -326,9 +326,9 @@ define <2 x i64> @test_aeskeygenassist(<2 x i64> %a0, <2 x i64> *%a1) {
; BROADWELL-LABEL: test_aeskeygenassist:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vaeskeygenassist $7, %xmm0, %xmm0 # sched: [29:7.00]
; BROADWELL-NEXT: vaeskeygenassist $7, (%rdi), %xmm1 # sched: [28:7.00]
; BROADWELL-NEXT: vaeskeygenassist $7, (%rdi), %xmm1 # sched: [33:7.00]
; BROADWELL-NEXT: vpor %xmm1, %xmm0, %xmm0 # sched: [1:0.33]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_aeskeygenassist:
; SKYLAKE: # BB#0:

View File

@ -31,8 +31,8 @@ define <4 x double> @test_addpd(<4 x double> %a0, <4 x double> %a1, <4 x double>
; BROADWELL-LABEL: test_addpd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vaddpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: vaddpd (%rdi), %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vaddpd (%rdi), %ymm0, %ymm0 # sched: [9:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_addpd:
; SKYLAKE: # BB#0:
@ -85,8 +85,8 @@ define <8 x float> @test_addps(<8 x float> %a0, <8 x float> %a1, <8 x float> *%a
; BROADWELL-LABEL: test_addps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vaddps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: vaddps (%rdi), %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vaddps (%rdi), %ymm0, %ymm0 # sched: [9:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_addps:
; SKYLAKE: # BB#0:
@ -139,8 +139,8 @@ define <4 x double> @test_addsubpd(<4 x double> %a0, <4 x double> %a1, <4 x doub
; BROADWELL-LABEL: test_addsubpd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vaddsubpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: vaddsubpd (%rdi), %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vaddsubpd (%rdi), %ymm0, %ymm0 # sched: [9:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_addsubpd:
; SKYLAKE: # BB#0:
@ -194,8 +194,8 @@ define <8 x float> @test_addsubps(<8 x float> %a0, <8 x float> %a1, <8 x float>
; BROADWELL-LABEL: test_addsubps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vaddsubps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: vaddsubps (%rdi), %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vaddsubps (%rdi), %ymm0, %ymm0 # sched: [9:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_addsubps:
; SKYLAKE: # BB#0:
@ -252,9 +252,9 @@ define <4 x double> @test_andnotpd(<4 x double> %a0, <4 x double> %a1, <4 x doub
; BROADWELL-LABEL: test_andnotpd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vandnpd %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
; BROADWELL-NEXT: vandnpd (%rdi), %ymm0, %ymm0 # sched: [1:1.00]
; BROADWELL-NEXT: vandnpd (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
; BROADWELL-NEXT: vaddpd %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_andnotpd:
; SKYLAKE: # BB#0:
@ -321,9 +321,9 @@ define <8 x float> @test_andnotps(<8 x float> %a0, <8 x float> %a1, <8 x float>
; BROADWELL-LABEL: test_andnotps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vandnps %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
; BROADWELL-NEXT: vandnps (%rdi), %ymm0, %ymm0 # sched: [1:1.00]
; BROADWELL-NEXT: vandnps (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
; BROADWELL-NEXT: vaddps %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_andnotps:
; SKYLAKE: # BB#0:
@ -390,9 +390,9 @@ define <4 x double> @test_andpd(<4 x double> %a0, <4 x double> %a1, <4 x double>
; BROADWELL-LABEL: test_andpd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vandpd %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
; BROADWELL-NEXT: vandpd (%rdi), %ymm0, %ymm0 # sched: [1:1.00]
; BROADWELL-NEXT: vandpd (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
; BROADWELL-NEXT: vaddpd %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_andpd:
; SKYLAKE: # BB#0:
@ -457,9 +457,9 @@ define <8 x float> @test_andps(<8 x float> %a0, <8 x float> %a1, <8 x float> *%a
; BROADWELL-LABEL: test_andps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vandps %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
; BROADWELL-NEXT: vandps (%rdi), %ymm0, %ymm0 # sched: [1:1.00]
; BROADWELL-NEXT: vandps (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
; BROADWELL-NEXT: vaddps %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_andps:
; SKYLAKE: # BB#0:
@ -525,8 +525,8 @@ define <4 x double> @test_blendpd(<4 x double> %a0, <4 x double> %a1, <4 x doubl
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0],ymm1[1,2],ymm0[3] sched: [1:0.33]
; BROADWELL-NEXT: vaddpd %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0],mem[1,2],ymm0[3] sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0],mem[1,2],ymm0[3] sched: [7:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_blendpd:
; SKYLAKE: # BB#0:
@ -584,8 +584,8 @@ define <8 x float> @test_blendps(<8 x float> %a0, <8 x float> %a1, <8 x float> *
; BROADWELL-LABEL: test_blendps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2],ymm0[3,4,5,6,7] sched: [1:0.33]
; BROADWELL-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0,1],mem[2],ymm0[3],mem[4,5,6],ymm0[7] sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0,1],mem[2],ymm0[3],mem[4,5,6],ymm0[7] sched: [7:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_blendps:
; SKYLAKE: # BB#0:
@ -638,8 +638,8 @@ define <4 x double> @test_blendvpd(<4 x double> %a0, <4 x double> %a1, <4 x doub
; BROADWELL-LABEL: test_blendvpd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vblendvpd %ymm2, %ymm1, %ymm0, %ymm0 # sched: [2:2.00]
; BROADWELL-NEXT: vblendvpd %ymm2, (%rdi), %ymm0, %ymm0 # sched: [2:2.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vblendvpd %ymm2, (%rdi), %ymm0, %ymm0 # sched: [8:2.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_blendvpd:
; SKYLAKE: # BB#0:
@ -693,8 +693,8 @@ define <8 x float> @test_blendvps(<8 x float> %a0, <8 x float> %a1, <8 x float>
; BROADWELL-LABEL: test_blendvps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vblendvps %ymm2, %ymm1, %ymm0, %ymm0 # sched: [2:2.00]
; BROADWELL-NEXT: vblendvps %ymm2, (%rdi), %ymm0, %ymm0 # sched: [2:2.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vblendvps %ymm2, (%rdi), %ymm0, %ymm0 # sched: [8:2.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_blendvps:
; SKYLAKE: # BB#0:
@ -744,8 +744,8 @@ define <8 x float> @test_broadcastf128(<4 x float> *%a0) {
;
; BROADWELL-LABEL: test_broadcastf128:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] sched: [6:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_broadcastf128:
; SKYLAKE: # BB#0:
@ -789,8 +789,8 @@ define <4 x double> @test_broadcastsd_ymm(double *%a0) {
;
; BROADWELL-LABEL: test_broadcastsd_ymm:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vbroadcastsd (%rdi), %ymm0 # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vbroadcastsd (%rdi), %ymm0 # sched: [6:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_broadcastsd_ymm:
; SKYLAKE: # BB#0:
@ -835,8 +835,8 @@ define <4 x float> @test_broadcastss(float *%a0) {
;
; BROADWELL-LABEL: test_broadcastss:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vbroadcastss (%rdi), %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vbroadcastss (%rdi), %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_broadcastss:
; SKYLAKE: # BB#0:
@ -881,8 +881,8 @@ define <8 x float> @test_broadcastss_ymm(float *%a0) {
;
; BROADWELL-LABEL: test_broadcastss_ymm:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vbroadcastss (%rdi), %ymm0 # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vbroadcastss (%rdi), %ymm0 # sched: [6:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_broadcastss_ymm:
; SKYLAKE: # BB#0:
@ -934,9 +934,9 @@ define <4 x double> @test_cmppd(<4 x double> %a0, <4 x double> %a1, <4 x double>
; BROADWELL-LABEL: test_cmppd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vcmpeqpd %ymm1, %ymm0, %ymm1 # sched: [3:1.00]
; BROADWELL-NEXT: vcmpeqpd (%rdi), %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: vcmpeqpd (%rdi), %ymm0, %ymm0 # sched: [9:1.00]
; BROADWELL-NEXT: vorpd %ymm0, %ymm1, %ymm0 # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_cmppd:
; SKYLAKE: # BB#0:
@ -1002,9 +1002,9 @@ define <8 x float> @test_cmpps(<8 x float> %a0, <8 x float> %a1, <8 x float> *%a
; BROADWELL-LABEL: test_cmpps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vcmpeqps %ymm1, %ymm0, %ymm1 # sched: [3:1.00]
; BROADWELL-NEXT: vcmpeqps (%rdi), %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: vcmpeqps (%rdi), %ymm0, %ymm0 # sched: [9:1.00]
; BROADWELL-NEXT: vorps %ymm0, %ymm1, %ymm0 # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_cmpps:
; SKYLAKE: # BB#0:
@ -1070,9 +1070,9 @@ define <4 x double> @test_cvtdq2pd(<4 x i32> %a0, <4 x i32> *%a1) {
; BROADWELL-LABEL: test_cvtdq2pd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vcvtdq2pd %xmm0, %ymm0 # sched: [6:1.00]
; BROADWELL-NEXT: vcvtdq2pd (%rdi), %ymm1 # sched: [6:1.00]
; BROADWELL-NEXT: vcvtdq2pd (%rdi), %ymm1 # sched: [11:1.00]
; BROADWELL-NEXT: vaddpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_cvtdq2pd:
; SKYLAKE: # BB#0:
@ -1135,9 +1135,9 @@ define <8 x float> @test_cvtdq2ps(<8 x i32> %a0, <8 x i32> *%a1) {
; BROADWELL-LABEL: test_cvtdq2ps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vcvtdq2ps %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: vcvtdq2ps (%rdi), %ymm1 # sched: [3:1.00]
; BROADWELL-NEXT: vcvtdq2ps (%rdi), %ymm1 # sched: [9:1.00]
; BROADWELL-NEXT: vaddps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_cvtdq2ps:
; SKYLAKE: # BB#0:
@ -1198,9 +1198,9 @@ define <8 x i32> @test_cvtpd2dq(<4 x double> %a0, <4 x double> *%a1) {
; BROADWELL-LABEL: test_cvtpd2dq:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vcvttpd2dq %ymm0, %xmm0 # sched: [6:1.00]
; BROADWELL-NEXT: vcvttpd2dqy (%rdi), %xmm1 # sched: [7:1.00]
; BROADWELL-NEXT: vcvttpd2dqy (%rdi), %xmm1 # sched: [8:1.00]
; BROADWELL-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_cvtpd2dq:
; SKYLAKE: # BB#0:
@ -1261,9 +1261,9 @@ define <8 x float> @test_cvtpd2ps(<4 x double> %a0, <4 x double> *%a1) {
; BROADWELL-LABEL: test_cvtpd2ps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vcvtpd2ps %ymm0, %xmm0 # sched: [6:1.00]
; BROADWELL-NEXT: vcvtpd2psy (%rdi), %xmm1 # sched: [7:1.00]
; BROADWELL-NEXT: vcvtpd2psy (%rdi), %xmm1 # sched: [8:1.00]
; BROADWELL-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_cvtpd2ps:
; SKYLAKE: # BB#0:
@ -1324,9 +1324,9 @@ define <8 x i32> @test_cvtps2dq(<8 x float> %a0, <8 x float> *%a1) {
; BROADWELL-LABEL: test_cvtps2dq:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vcvttps2dq %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: vcvttps2dq (%rdi), %ymm1 # sched: [3:1.00]
; BROADWELL-NEXT: vcvttps2dq (%rdi), %ymm1 # sched: [9:1.00]
; BROADWELL-NEXT: vorps %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_cvtps2dq:
; SKYLAKE: # BB#0:
@ -1383,9 +1383,9 @@ define <4 x double> @test_divpd(<4 x double> %a0, <4 x double> %a1, <4 x double>
;
; BROADWELL-LABEL: test_divpd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vdivpd %ymm1, %ymm0, %ymm0 # sched: [35:2.00]
; BROADWELL-NEXT: vdivpd (%rdi), %ymm0, %ymm0 # sched: [35:2.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vdivpd %ymm1, %ymm0, %ymm0 # sched: [23:2.00]
; BROADWELL-NEXT: vdivpd (%rdi), %ymm0, %ymm0 # sched: [29:2.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_divpd:
; SKYLAKE: # BB#0:
@ -1437,9 +1437,9 @@ define <8 x float> @test_divps(<8 x float> %a0, <8 x float> %a1, <8 x float> *%a
;
; BROADWELL-LABEL: test_divps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vdivps %ymm1, %ymm0, %ymm0 # sched: [21:2.00]
; BROADWELL-NEXT: vdivps (%rdi), %ymm0, %ymm0 # sched: [21:2.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vdivps %ymm1, %ymm0, %ymm0 # sched: [17:2.00]
; BROADWELL-NEXT: vdivps (%rdi), %ymm0, %ymm0 # sched: [23:2.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_divps:
; SKYLAKE: # BB#0:
@ -1492,8 +1492,8 @@ define <8 x float> @test_dpps(<8 x float> %a0, <8 x float> %a1, <8 x float> *%a2
; BROADWELL-LABEL: test_dpps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vdpps $7, %ymm1, %ymm0, %ymm0 # sched: [14:2.00]
; BROADWELL-NEXT: vdpps $7, (%rdi), %ymm0, %ymm0 # sched: [14:2.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vdpps $7, (%rdi), %ymm0, %ymm0 # sched: [20:2.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_dpps:
; SKYLAKE: # BB#0:
@ -1552,7 +1552,7 @@ define <4 x float> @test_extractf128(<8 x float> %a0, <8 x float> %a1, <4 x floa
; BROADWELL-NEXT: vextractf128 $1, %ymm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: vextractf128 $1, %ymm1, (%rdi) # sched: [1:1.00]
; BROADWELL-NEXT: vzeroupper # sched: [4:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_extractf128:
; SKYLAKE: # BB#0:
@ -1608,8 +1608,8 @@ define <4 x double> @test_haddpd(<4 x double> %a0, <4 x double> %a1, <4 x double
; BROADWELL-LABEL: test_haddpd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vhaddpd %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
; BROADWELL-NEXT: vhaddpd (%rdi), %ymm0, %ymm0 # sched: [5:2.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vhaddpd (%rdi), %ymm0, %ymm0 # sched: [11:2.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_haddpd:
; SKYLAKE: # BB#0:
@ -1663,8 +1663,8 @@ define <8 x float> @test_haddps(<8 x float> %a0, <8 x float> %a1, <8 x float> *%
; BROADWELL-LABEL: test_haddps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vhaddps %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
; BROADWELL-NEXT: vhaddps (%rdi), %ymm0, %ymm0 # sched: [5:2.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vhaddps (%rdi), %ymm0, %ymm0 # sched: [11:2.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_haddps:
; SKYLAKE: # BB#0:
@ -1718,8 +1718,8 @@ define <4 x double> @test_hsubpd(<4 x double> %a0, <4 x double> %a1, <4 x double
; BROADWELL-LABEL: test_hsubpd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vhsubpd %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
; BROADWELL-NEXT: vhsubpd (%rdi), %ymm0, %ymm0 # sched: [5:2.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vhsubpd (%rdi), %ymm0, %ymm0 # sched: [11:2.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_hsubpd:
; SKYLAKE: # BB#0:
@ -1773,8 +1773,8 @@ define <8 x float> @test_hsubps(<8 x float> %a0, <8 x float> %a1, <8 x float> *%
; BROADWELL-LABEL: test_hsubps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vhsubps %ymm1, %ymm0, %ymm0 # sched: [5:2.00]
; BROADWELL-NEXT: vhsubps (%rdi), %ymm0, %ymm0 # sched: [5:2.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vhsubps (%rdi), %ymm0, %ymm0 # sched: [11:2.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_hsubps:
; SKYLAKE: # BB#0:
@ -1831,9 +1831,9 @@ define <8 x float> @test_insertf128(<8 x float> %a0, <4 x float> %a1, <4 x float
; BROADWELL-LABEL: test_insertf128:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm1 # sched: [3:1.00]
; BROADWELL-NEXT: vinsertf128 $1, (%rdi), %ymm0, %ymm0 # sched: [1:0.50]
; BROADWELL-NEXT: vinsertf128 $1, (%rdi), %ymm0, %ymm0 # sched: [6:0.50]
; BROADWELL-NEXT: vaddps %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_insertf128:
; SKYLAKE: # BB#0:
@ -1889,8 +1889,8 @@ define <32 x i8> @test_lddqu(i8* %a0) {
;
; BROADWELL-LABEL: test_lddqu:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vlddqu (%rdi), %ymm0 # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vlddqu (%rdi), %ymm0 # sched: [6:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_lddqu:
; SKYLAKE: # BB#0:
@ -1940,10 +1940,10 @@ define <2 x double> @test_maskmovpd(i8* %a0, <2 x i64> %a1, <2 x double> %a2) {
;
; BROADWELL-LABEL: test_maskmovpd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmaskmovpd (%rdi), %xmm0, %xmm2 # sched: [2:2.00]
; BROADWELL-NEXT: vmaskmovpd %xmm1, %xmm0, (%rdi) # sched: [4:1.00]
; BROADWELL-NEXT: vmaskmovpd (%rdi), %xmm0, %xmm2 # sched: [7:2.00]
; BROADWELL-NEXT: vmaskmovpd %xmm1, %xmm0, (%rdi) # sched: [5:1.00]
; BROADWELL-NEXT: vmovapd %xmm2, %xmm0 # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_maskmovpd:
; SKYLAKE: # BB#0:
@ -2003,10 +2003,10 @@ define <4 x double> @test_maskmovpd_ymm(i8* %a0, <4 x i64> %a1, <4 x double> %a2
;
; BROADWELL-LABEL: test_maskmovpd_ymm:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmaskmovpd (%rdi), %ymm0, %ymm2 # sched: [2:2.00]
; BROADWELL-NEXT: vmaskmovpd %ymm1, %ymm0, (%rdi) # sched: [4:1.00]
; BROADWELL-NEXT: vmaskmovpd (%rdi), %ymm0, %ymm2 # sched: [8:2.00]
; BROADWELL-NEXT: vmaskmovpd %ymm1, %ymm0, (%rdi) # sched: [5:1.00]
; BROADWELL-NEXT: vmovapd %ymm2, %ymm0 # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_maskmovpd_ymm:
; SKYLAKE: # BB#0:
@ -2066,10 +2066,10 @@ define <4 x float> @test_maskmovps(i8* %a0, <4 x i32> %a1, <4 x float> %a2) {
;
; BROADWELL-LABEL: test_maskmovps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmaskmovps (%rdi), %xmm0, %xmm2 # sched: [2:2.00]
; BROADWELL-NEXT: vmaskmovps %xmm1, %xmm0, (%rdi) # sched: [4:1.00]
; BROADWELL-NEXT: vmaskmovps (%rdi), %xmm0, %xmm2 # sched: [7:2.00]
; BROADWELL-NEXT: vmaskmovps %xmm1, %xmm0, (%rdi) # sched: [5:1.00]
; BROADWELL-NEXT: vmovaps %xmm2, %xmm0 # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_maskmovps:
; SKYLAKE: # BB#0:
@ -2129,10 +2129,10 @@ define <8 x float> @test_maskmovps_ymm(i8* %a0, <8 x i32> %a1, <8 x float> %a2)
;
; BROADWELL-LABEL: test_maskmovps_ymm:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmaskmovps (%rdi), %ymm0, %ymm2 # sched: [2:2.00]
; BROADWELL-NEXT: vmaskmovps %ymm1, %ymm0, (%rdi) # sched: [4:1.00]
; BROADWELL-NEXT: vmaskmovps (%rdi), %ymm0, %ymm2 # sched: [8:2.00]
; BROADWELL-NEXT: vmaskmovps %ymm1, %ymm0, (%rdi) # sched: [5:1.00]
; BROADWELL-NEXT: vmovaps %ymm2, %ymm0 # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_maskmovps_ymm:
; SKYLAKE: # BB#0:
@ -2190,8 +2190,8 @@ define <4 x double> @test_maxpd(<4 x double> %a0, <4 x double> %a1, <4 x double>
; BROADWELL-LABEL: test_maxpd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmaxpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: vmaxpd (%rdi), %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vmaxpd (%rdi), %ymm0, %ymm0 # sched: [9:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_maxpd:
; SKYLAKE: # BB#0:
@ -2245,8 +2245,8 @@ define <8 x float> @test_maxps(<8 x float> %a0, <8 x float> %a1, <8 x float> *%a
; BROADWELL-LABEL: test_maxps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmaxps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: vmaxps (%rdi), %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vmaxps (%rdi), %ymm0, %ymm0 # sched: [9:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_maxps:
; SKYLAKE: # BB#0:
@ -2300,8 +2300,8 @@ define <4 x double> @test_minpd(<4 x double> %a0, <4 x double> %a1, <4 x double>
; BROADWELL-LABEL: test_minpd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vminpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: vminpd (%rdi), %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vminpd (%rdi), %ymm0, %ymm0 # sched: [9:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_minpd:
; SKYLAKE: # BB#0:
@ -2355,8 +2355,8 @@ define <8 x float> @test_minps(<8 x float> %a0, <8 x float> %a1, <8 x float> *%a
; BROADWELL-LABEL: test_minps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vminps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: vminps (%rdi), %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vminps (%rdi), %ymm0, %ymm0 # sched: [9:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_minps:
; SKYLAKE: # BB#0:
@ -2412,10 +2412,10 @@ define <4 x double> @test_movapd(<4 x double> *%a0, <4 x double> *%a1) {
;
; BROADWELL-LABEL: test_movapd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmovapd (%rdi), %ymm0 # sched: [1:0.50]
; BROADWELL-NEXT: vmovapd (%rdi), %ymm0 # sched: [6:0.50]
; BROADWELL-NEXT: vaddpd %ymm0, %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: vmovapd %ymm0, (%rsi) # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_movapd:
; SKYLAKE: # BB#0:
@ -2474,10 +2474,10 @@ define <8 x float> @test_movaps(<8 x float> *%a0, <8 x float> *%a1) {
;
; BROADWELL-LABEL: test_movaps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmovaps (%rdi), %ymm0 # sched: [1:0.50]
; BROADWELL-NEXT: vmovaps (%rdi), %ymm0 # sched: [6:0.50]
; BROADWELL-NEXT: vaddps %ymm0, %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: vmovaps %ymm0, (%rsi) # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_movaps:
; SKYLAKE: # BB#0:
@ -2537,9 +2537,9 @@ define <4 x double> @test_movddup(<4 x double> %a0, <4 x double> *%a1) {
; BROADWELL-LABEL: test_movddup:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmovddup {{.*#+}} ymm0 = ymm0[0,0,2,2] sched: [1:1.00]
; BROADWELL-NEXT: vmovddup {{.*#+}} ymm1 = mem[0,0,2,2] sched: [1:0.50]
; BROADWELL-NEXT: vmovddup {{.*#+}} ymm1 = mem[0,0,2,2] sched: [6:0.50]
; BROADWELL-NEXT: vaddpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_movddup:
; SKYLAKE: # BB#0:
@ -2598,7 +2598,7 @@ define i32 @test_movmskpd(<4 x double> %a0) {
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmovmskpd %ymm0, %eax # sched: [3:1.00]
; BROADWELL-NEXT: vzeroupper # sched: [4:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_movmskpd:
; SKYLAKE: # BB#0:
@ -2650,7 +2650,7 @@ define i32 @test_movmskps(<8 x float> %a0) {
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmovmskps %ymm0, %eax # sched: [3:1.00]
; BROADWELL-NEXT: vzeroupper # sched: [4:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_movmskps:
; SKYLAKE: # BB#0:
@ -2702,7 +2702,7 @@ define <4 x double> @test_movntpd(<4 x double> %a0, <4 x double> *%a1) {
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vaddpd %ymm0, %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: vmovntpd %ymm0, (%rdi) # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_movntpd:
; SKYLAKE: # BB#0:
@ -2755,7 +2755,7 @@ define <8 x float> @test_movntps(<8 x float> %a0, <8 x float> *%a1) {
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vaddps %ymm0, %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: vmovntps %ymm0, (%rdi) # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_movntps:
; SKYLAKE: # BB#0:
@ -2810,9 +2810,9 @@ define <8 x float> @test_movshdup(<8 x float> %a0, <8 x float> *%a1) {
; BROADWELL-LABEL: test_movshdup:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmovshdup {{.*#+}} ymm0 = ymm0[1,1,3,3,5,5,7,7] sched: [1:1.00]
; BROADWELL-NEXT: vmovshdup {{.*#+}} ymm1 = mem[1,1,3,3,5,5,7,7] sched: [1:0.50]
; BROADWELL-NEXT: vmovshdup {{.*#+}} ymm1 = mem[1,1,3,3,5,5,7,7] sched: [6:0.50]
; BROADWELL-NEXT: vaddps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_movshdup:
; SKYLAKE: # BB#0:
@ -2873,9 +2873,9 @@ define <8 x float> @test_movsldup(<8 x float> %a0, <8 x float> *%a1) {
; BROADWELL-LABEL: test_movsldup:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmovsldup {{.*#+}} ymm0 = ymm0[0,0,2,2,4,4,6,6] sched: [1:1.00]
; BROADWELL-NEXT: vmovsldup {{.*#+}} ymm1 = mem[0,0,2,2,4,4,6,6] sched: [1:0.50]
; BROADWELL-NEXT: vmovsldup {{.*#+}} ymm1 = mem[0,0,2,2,4,4,6,6] sched: [6:0.50]
; BROADWELL-NEXT: vaddps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_movsldup:
; SKYLAKE: # BB#0:
@ -2937,10 +2937,10 @@ define <4 x double> @test_movupd(<4 x double> *%a0, <4 x double> *%a1) {
;
; BROADWELL-LABEL: test_movupd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmovupd (%rdi), %ymm0 # sched: [1:0.50]
; BROADWELL-NEXT: vmovupd (%rdi), %ymm0 # sched: [6:0.50]
; BROADWELL-NEXT: vaddpd %ymm0, %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: vmovupd %ymm0, (%rsi) # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_movupd:
; SKYLAKE: # BB#0:
@ -3001,10 +3001,10 @@ define <8 x float> @test_movups(<8 x float> *%a0, <8 x float> *%a1) {
;
; BROADWELL-LABEL: test_movups:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmovups (%rdi), %ymm0 # sched: [1:0.50]
; BROADWELL-NEXT: vmovups (%rdi), %ymm0 # sched: [6:0.50]
; BROADWELL-NEXT: vaddps %ymm0, %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: vmovups %ymm0, (%rsi) # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_movups:
; SKYLAKE: # BB#0:
@ -3060,9 +3060,9 @@ define <4 x double> @test_mulpd(<4 x double> %a0, <4 x double> %a1, <4 x double>
;
; BROADWELL-LABEL: test_mulpd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmulpd %ymm1, %ymm0, %ymm0 # sched: [5:0.50]
; BROADWELL-NEXT: vmulpd (%rdi), %ymm0, %ymm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vmulpd %ymm1, %ymm0, %ymm0 # sched: [3:0.50]
; BROADWELL-NEXT: vmulpd (%rdi), %ymm0, %ymm0 # sched: [9:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_mulpd:
; SKYLAKE: # BB#0:
@ -3114,9 +3114,9 @@ define <8 x float> @test_mulps(<8 x float> %a0, <8 x float> %a1, <8 x float> *%a
;
; BROADWELL-LABEL: test_mulps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmulps %ymm1, %ymm0, %ymm0 # sched: [5:0.50]
; BROADWELL-NEXT: vmulps (%rdi), %ymm0, %ymm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vmulps %ymm1, %ymm0, %ymm0 # sched: [3:0.50]
; BROADWELL-NEXT: vmulps (%rdi), %ymm0, %ymm0 # sched: [9:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_mulps:
; SKYLAKE: # BB#0:
@ -3172,9 +3172,9 @@ define <4 x double> @orpd(<4 x double> %a0, <4 x double> %a1, <4 x double> *%a2)
; BROADWELL-LABEL: orpd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vorpd %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
; BROADWELL-NEXT: vorpd (%rdi), %ymm0, %ymm0 # sched: [1:1.00]
; BROADWELL-NEXT: vorpd (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
; BROADWELL-NEXT: vaddpd %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: orpd:
; SKYLAKE: # BB#0:
@ -3239,9 +3239,9 @@ define <8 x float> @test_orps(<8 x float> %a0, <8 x float> %a1, <8 x float> *%a2
; BROADWELL-LABEL: test_orps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vorps %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
; BROADWELL-NEXT: vorps (%rdi), %ymm0, %ymm0 # sched: [1:1.00]
; BROADWELL-NEXT: vorps (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
; BROADWELL-NEXT: vaddps %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_orps:
; SKYLAKE: # BB#0:
@ -3306,9 +3306,9 @@ define <4 x double> @test_perm2f128(<4 x double> %a0, <4 x double> %a1, <4 x dou
; BROADWELL-LABEL: test_perm2f128:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vperm2f128 {{.*#+}} ymm1 = ymm0[2,3],ymm1[0,1] sched: [3:1.00]
; BROADWELL-NEXT: vperm2f128 {{.*#+}} ymm0 = ymm0[2,3],mem[0,1] sched: [3:1.00]
; BROADWELL-NEXT: vperm2f128 {{.*#+}} ymm0 = ymm0[2,3],mem[0,1] sched: [9:1.00]
; BROADWELL-NEXT: vaddpd %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_perm2f128:
; SKYLAKE: # BB#0:
@ -3369,9 +3369,9 @@ define <2 x double> @test_permilpd(<2 x double> %a0, <2 x double> *%a1) {
; BROADWELL-LABEL: test_permilpd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0] sched: [1:1.00]
; BROADWELL-NEXT: vpermilpd {{.*#+}} xmm1 = mem[1,0] sched: [1:1.00]
; BROADWELL-NEXT: vpermilpd {{.*#+}} xmm1 = mem[1,0] sched: [6:1.00]
; BROADWELL-NEXT: vaddpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_permilpd:
; SKYLAKE: # BB#0:
@ -3432,9 +3432,9 @@ define <4 x double> @test_permilpd_ymm(<4 x double> %a0, <4 x double> *%a1) {
; BROADWELL-LABEL: test_permilpd_ymm:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpermilpd {{.*#+}} ymm0 = ymm0[1,0,2,3] sched: [1:1.00]
; BROADWELL-NEXT: vpermilpd {{.*#+}} ymm1 = mem[1,0,2,3] sched: [1:1.00]
; BROADWELL-NEXT: vpermilpd {{.*#+}} ymm1 = mem[1,0,2,3] sched: [7:1.00]
; BROADWELL-NEXT: vaddpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_permilpd_ymm:
; SKYLAKE: # BB#0:
@ -3495,9 +3495,9 @@ define <4 x float> @test_permilps(<4 x float> %a0, <4 x float> *%a1) {
; BROADWELL-LABEL: test_permilps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,2,1,0] sched: [1:1.00]
; BROADWELL-NEXT: vpermilps {{.*#+}} xmm1 = mem[3,2,1,0] sched: [1:1.00]
; BROADWELL-NEXT: vpermilps {{.*#+}} xmm1 = mem[3,2,1,0] sched: [6:1.00]
; BROADWELL-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_permilps:
; SKYLAKE: # BB#0:
@ -3558,9 +3558,9 @@ define <8 x float> @test_permilps_ymm(<8 x float> %a0, <8 x float> *%a1) {
; BROADWELL-LABEL: test_permilps_ymm:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4] sched: [1:1.00]
; BROADWELL-NEXT: vpermilps {{.*#+}} ymm1 = mem[3,2,1,0,7,6,5,4] sched: [1:1.00]
; BROADWELL-NEXT: vpermilps {{.*#+}} ymm1 = mem[3,2,1,0,7,6,5,4] sched: [7:1.00]
; BROADWELL-NEXT: vaddps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_permilps_ymm:
; SKYLAKE: # BB#0:
@ -3618,8 +3618,8 @@ define <2 x double> @test_permilvarpd(<2 x double> %a0, <2 x i64> %a1, <2 x i64>
; BROADWELL-LABEL: test_permilvarpd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpermilpd %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
; BROADWELL-NEXT: vpermilpd (%rdi), %xmm0, %xmm0 # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpermilpd (%rdi), %xmm0, %xmm0 # sched: [6:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_permilvarpd:
; SKYLAKE: # BB#0:
@ -3673,8 +3673,8 @@ define <4 x double> @test_permilvarpd_ymm(<4 x double> %a0, <4 x i64> %a1, <4 x
; BROADWELL-LABEL: test_permilvarpd_ymm:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpermilpd %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
; BROADWELL-NEXT: vpermilpd (%rdi), %ymm0, %ymm0 # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpermilpd (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_permilvarpd_ymm:
; SKYLAKE: # BB#0:
@ -3728,8 +3728,8 @@ define <4 x float> @test_permilvarps(<4 x float> %a0, <4 x i32> %a1, <4 x i32> *
; BROADWELL-LABEL: test_permilvarps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpermilps %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
; BROADWELL-NEXT: vpermilps (%rdi), %xmm0, %xmm0 # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpermilps (%rdi), %xmm0, %xmm0 # sched: [6:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_permilvarps:
; SKYLAKE: # BB#0:
@ -3783,8 +3783,8 @@ define <8 x float> @test_permilvarps_ymm(<8 x float> %a0, <8 x i32> %a1, <8 x i3
; BROADWELL-LABEL: test_permilvarps_ymm:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpermilps %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
; BROADWELL-NEXT: vpermilps (%rdi), %ymm0, %ymm0 # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpermilps (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_permilvarps_ymm:
; SKYLAKE: # BB#0:
@ -3840,10 +3840,10 @@ define <8 x float> @test_rcpps(<8 x float> %a0, <8 x float> *%a1) {
;
; BROADWELL-LABEL: test_rcpps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vrcpps (%rdi), %ymm1 # sched: [11:2.00]
; BROADWELL-NEXT: vrcpps (%rdi), %ymm1 # sched: [17:2.00]
; BROADWELL-NEXT: vrcpps %ymm0, %ymm0 # sched: [11:2.00]
; BROADWELL-NEXT: vaddps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_rcpps:
; SKYLAKE: # BB#0:
@ -3904,10 +3904,10 @@ define <4 x double> @test_roundpd(<4 x double> %a0, <4 x double> *%a1) {
;
; BROADWELL-LABEL: test_roundpd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vroundpd $7, %ymm0, %ymm0 # sched: [5:1.25]
; BROADWELL-NEXT: vroundpd $7, (%rdi), %ymm1 # sched: [6:2.00]
; BROADWELL-NEXT: vroundpd $7, %ymm0, %ymm0 # sched: [6:0.50]
; BROADWELL-NEXT: vroundpd $7, (%rdi), %ymm1 # sched: [12:2.00]
; BROADWELL-NEXT: vaddpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_roundpd:
; SKYLAKE: # BB#0:
@ -3968,10 +3968,10 @@ define <8 x float> @test_roundps(<8 x float> %a0, <8 x float> *%a1) {
;
; BROADWELL-LABEL: test_roundps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vroundps $7, %ymm0, %ymm0 # sched: [5:1.25]
; BROADWELL-NEXT: vroundps $7, (%rdi), %ymm1 # sched: [6:2.00]
; BROADWELL-NEXT: vroundps $7, %ymm0, %ymm0 # sched: [6:0.50]
; BROADWELL-NEXT: vroundps $7, (%rdi), %ymm1 # sched: [12:2.00]
; BROADWELL-NEXT: vaddps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_roundps:
; SKYLAKE: # BB#0:
@ -4032,10 +4032,10 @@ define <8 x float> @test_rsqrtps(<8 x float> %a0, <8 x float> *%a1) {
;
; BROADWELL-LABEL: test_rsqrtps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vrsqrtps (%rdi), %ymm1 # sched: [11:2.00]
; BROADWELL-NEXT: vrsqrtps (%rdi), %ymm1 # sched: [17:2.00]
; BROADWELL-NEXT: vrsqrtps %ymm0, %ymm0 # sched: [11:2.00]
; BROADWELL-NEXT: vaddps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_rsqrtps:
; SKYLAKE: # BB#0:
@ -4097,9 +4097,9 @@ define <4 x double> @test_shufpd(<4 x double> %a0, <4 x double> %a1, <4 x double
; BROADWELL-LABEL: test_shufpd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vshufpd {{.*#+}} ymm0 = ymm0[1],ymm1[0],ymm0[2],ymm1[3] sched: [1:1.00]
; BROADWELL-NEXT: vshufpd {{.*#+}} ymm1 = ymm1[1],mem[0],ymm1[2],mem[3] sched: [1:1.00]
; BROADWELL-NEXT: vshufpd {{.*#+}} ymm1 = ymm1[1],mem[0],ymm1[2],mem[3] sched: [7:1.00]
; BROADWELL-NEXT: vaddpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_shufpd:
; SKYLAKE: # BB#0:
@ -4157,8 +4157,8 @@ define <8 x float> @test_shufps(<8 x float> %a0, <8 x float> %a1, <8 x float> *%
; BROADWELL-LABEL: test_shufps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vshufps {{.*#+}} ymm0 = ymm0[0,0],ymm1[0,0],ymm0[4,4],ymm1[4,4] sched: [1:1.00]
; BROADWELL-NEXT: vshufps {{.*#+}} ymm0 = ymm0[0,3],mem[0,0],ymm0[4,7],mem[4,4] sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vshufps {{.*#+}} ymm0 = ymm0[0,3],mem[0,0],ymm0[4,7],mem[4,4] sched: [7:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_shufps:
; SKYLAKE: # BB#0:
@ -4213,10 +4213,10 @@ define <4 x double> @test_sqrtpd(<4 x double> %a0, <4 x double> *%a1) {
;
; BROADWELL-LABEL: test_sqrtpd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vsqrtpd (%rdi), %ymm1 # sched: [35:2.00]
; BROADWELL-NEXT: vsqrtpd %ymm0, %ymm0 # sched: [35:2.00]
; BROADWELL-NEXT: vsqrtpd (%rdi), %ymm1 # sched: [40:2.00]
; BROADWELL-NEXT: vsqrtpd %ymm0, %ymm0 # sched: [34:2.00]
; BROADWELL-NEXT: vaddpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_sqrtpd:
; SKYLAKE: # BB#0:
@ -4277,10 +4277,10 @@ define <8 x float> @test_sqrtps(<8 x float> %a0, <8 x float> *%a1) {
;
; BROADWELL-LABEL: test_sqrtps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vsqrtps (%rdi), %ymm1 # sched: [21:2.00]
; BROADWELL-NEXT: vsqrtps (%rdi), %ymm1 # sched: [27:2.00]
; BROADWELL-NEXT: vsqrtps %ymm0, %ymm0 # sched: [21:2.00]
; BROADWELL-NEXT: vaddps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_sqrtps:
; SKYLAKE: # BB#0:
@ -4339,8 +4339,8 @@ define <4 x double> @test_subpd(<4 x double> %a0, <4 x double> %a1, <4 x double>
; BROADWELL-LABEL: test_subpd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vsubpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: vsubpd (%rdi), %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vsubpd (%rdi), %ymm0, %ymm0 # sched: [9:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_subpd:
; SKYLAKE: # BB#0:
@ -4393,8 +4393,8 @@ define <8 x float> @test_subps(<8 x float> %a0, <8 x float> %a1, <8 x float> *%a
; BROADWELL-LABEL: test_subps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vsubps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: vsubps (%rdi), %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vsubps (%rdi), %ymm0, %ymm0 # sched: [9:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_subps:
; SKYLAKE: # BB#0:
@ -4458,9 +4458,9 @@ define i32 @test_testpd(<2 x double> %a0, <2 x double> %a1, <2 x double> *%a2) {
; BROADWELL-NEXT: xorl %eax, %eax # sched: [1:0.25]
; BROADWELL-NEXT: vtestpd %xmm1, %xmm0 # sched: [1:1.00]
; BROADWELL-NEXT: setb %al # sched: [1:0.50]
; BROADWELL-NEXT: vtestpd (%rdi), %xmm0 # sched: [1:1.00]
; BROADWELL-NEXT: adcl $0, %eax # sched: [2:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vtestpd (%rdi), %xmm0 # sched: [6:1.00]
; BROADWELL-NEXT: adcl $0, %eax # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_testpd:
; SKYLAKE: # BB#0:
@ -4541,10 +4541,10 @@ define i32 @test_testpd_ymm(<4 x double> %a0, <4 x double> %a1, <4 x double> *%a
; BROADWELL-NEXT: xorl %eax, %eax # sched: [1:0.25]
; BROADWELL-NEXT: vtestpd %ymm1, %ymm0 # sched: [1:1.00]
; BROADWELL-NEXT: setb %al # sched: [1:0.50]
; BROADWELL-NEXT: vtestpd (%rdi), %ymm0 # sched: [1:1.00]
; BROADWELL-NEXT: adcl $0, %eax # sched: [2:0.50]
; BROADWELL-NEXT: vtestpd (%rdi), %ymm0 # sched: [7:1.00]
; BROADWELL-NEXT: adcl $0, %eax # sched: [1:0.50]
; BROADWELL-NEXT: vzeroupper # sched: [4:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_testpd_ymm:
; SKYLAKE: # BB#0:
@ -4625,9 +4625,9 @@ define i32 @test_testps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a2) {
; BROADWELL-NEXT: xorl %eax, %eax # sched: [1:0.25]
; BROADWELL-NEXT: vtestps %xmm1, %xmm0 # sched: [1:1.00]
; BROADWELL-NEXT: setb %al # sched: [1:0.50]
; BROADWELL-NEXT: vtestps (%rdi), %xmm0 # sched: [1:1.00]
; BROADWELL-NEXT: adcl $0, %eax # sched: [2:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vtestps (%rdi), %xmm0 # sched: [6:1.00]
; BROADWELL-NEXT: adcl $0, %eax # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_testps:
; SKYLAKE: # BB#0:
@ -4708,10 +4708,10 @@ define i32 @test_testps_ymm(<8 x float> %a0, <8 x float> %a1, <8 x float> *%a2)
; BROADWELL-NEXT: xorl %eax, %eax # sched: [1:0.25]
; BROADWELL-NEXT: vtestps %ymm1, %ymm0 # sched: [1:1.00]
; BROADWELL-NEXT: setb %al # sched: [1:0.50]
; BROADWELL-NEXT: vtestps (%rdi), %ymm0 # sched: [1:1.00]
; BROADWELL-NEXT: adcl $0, %eax # sched: [2:0.50]
; BROADWELL-NEXT: vtestps (%rdi), %ymm0 # sched: [7:1.00]
; BROADWELL-NEXT: adcl $0, %eax # sched: [1:0.50]
; BROADWELL-NEXT: vzeroupper # sched: [4:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_testps_ymm:
; SKYLAKE: # BB#0:
@ -4784,9 +4784,9 @@ define <4 x double> @test_unpckhpd(<4 x double> %a0, <4 x double> %a1, <4 x doub
; BROADWELL-LABEL: test_unpckhpd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vunpckhpd {{.*#+}} ymm0 = ymm0[1],ymm1[1],ymm0[3],ymm1[3] sched: [1:1.00]
; BROADWELL-NEXT: vunpckhpd {{.*#+}} ymm1 = ymm1[1],mem[1],ymm1[3],mem[3] sched: [1:1.00]
; BROADWELL-NEXT: vunpckhpd {{.*#+}} ymm1 = ymm1[1],mem[1],ymm1[3],mem[3] sched: [7:1.00]
; BROADWELL-NEXT: vaddpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_unpckhpd:
; SKYLAKE: # BB#0:
@ -4844,8 +4844,8 @@ define <8 x float> @test_unpckhps(<8 x float> %a0, <8 x float> %a1, <8 x float>
; BROADWELL-LABEL: test_unpckhps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vunpckhps {{.*#+}} ymm0 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7] sched: [1:1.00]
; BROADWELL-NEXT: vunpckhps {{.*#+}} ymm0 = ymm0[2],mem[2],ymm0[3],mem[3],ymm0[6],mem[6],ymm0[7],mem[7] sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vunpckhps {{.*#+}} ymm0 = ymm0[2],mem[2],ymm0[3],mem[3],ymm0[6],mem[6],ymm0[7],mem[7] sched: [7:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_unpckhps:
; SKYLAKE: # BB#0:
@ -4901,9 +4901,9 @@ define <4 x double> @test_unpcklpd(<4 x double> %a0, <4 x double> %a1, <4 x doub
; BROADWELL-LABEL: test_unpcklpd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vunpcklpd {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[2],ymm1[2] sched: [1:1.00]
; BROADWELL-NEXT: vunpcklpd {{.*#+}} ymm1 = ymm1[0],mem[0],ymm1[2],mem[2] sched: [1:1.00]
; BROADWELL-NEXT: vunpcklpd {{.*#+}} ymm1 = ymm1[0],mem[0],ymm1[2],mem[2] sched: [7:1.00]
; BROADWELL-NEXT: vaddpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_unpcklpd:
; SKYLAKE: # BB#0:
@ -4961,8 +4961,8 @@ define <8 x float> @test_unpcklps(<8 x float> %a0, <8 x float> %a1, <8 x float>
; BROADWELL-LABEL: test_unpcklps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vunpcklps {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5] sched: [1:1.00]
; BROADWELL-NEXT: vunpcklps {{.*#+}} ymm0 = ymm0[0],mem[0],ymm0[1],mem[1],ymm0[4],mem[4],ymm0[5],mem[5] sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vunpcklps {{.*#+}} ymm0 = ymm0[0],mem[0],ymm0[1],mem[1],ymm0[4],mem[4],ymm0[5],mem[5] sched: [7:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_unpcklps:
; SKYLAKE: # BB#0:
@ -5018,9 +5018,9 @@ define <4 x double> @test_xorpd(<4 x double> %a0, <4 x double> %a1, <4 x double>
; BROADWELL-LABEL: test_xorpd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vxorpd %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
; BROADWELL-NEXT: vxorpd (%rdi), %ymm0, %ymm0 # sched: [1:1.00]
; BROADWELL-NEXT: vxorpd (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
; BROADWELL-NEXT: vaddpd %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_xorpd:
; SKYLAKE: # BB#0:
@ -5085,9 +5085,9 @@ define <8 x float> @test_xorps(<8 x float> %a0, <8 x float> %a1, <8 x float> *%a
; BROADWELL-LABEL: test_xorps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vxorps %ymm1, %ymm0, %ymm0 # sched: [1:1.00]
; BROADWELL-NEXT: vxorps (%rdi), %ymm0, %ymm0 # sched: [1:1.00]
; BROADWELL-NEXT: vxorps (%rdi), %ymm0, %ymm0 # sched: [7:1.00]
; BROADWELL-NEXT: vaddps %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_xorps:
; SKYLAKE: # BB#0:
@ -5146,7 +5146,7 @@ define void @test_zeroall() {
; BROADWELL-LABEL: test_zeroall:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vzeroall # sched: [16:16.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_zeroall:
; SKYLAKE: # BB#0:
@ -5191,7 +5191,7 @@ define void @test_zeroupper() {
; BROADWELL-LABEL: test_zeroupper:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vzeroupper # sched: [4:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_zeroupper:
; SKYLAKE: # BB#0:

File diff suppressed because it is too large Load Diff

View File

@ -30,10 +30,10 @@ define i16 @test_andn_i16(i16 zeroext %a0, i16 zeroext %a1, i16 *%a2) {
; BROADWELL: # BB#0:
; BROADWELL-NEXT: andnl %esi, %edi, %eax # sched: [1:0.50]
; BROADWELL-NEXT: notl %edi # sched: [1:0.25]
; BROADWELL-NEXT: andw (%rdx), %di # sched: [1:0.50]
; BROADWELL-NEXT: andw (%rdx), %di # sched: [6:0.50]
; BROADWELL-NEXT: addl %edi, %eax # sched: [1:0.25]
; BROADWELL-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill>
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_andn_i16:
; SKYLAKE: # BB#0:
@ -87,9 +87,9 @@ define i32 @test_andn_i32(i32 %a0, i32 %a1, i32 *%a2) {
; BROADWELL-LABEL: test_andn_i32:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: andnl %esi, %edi, %ecx # sched: [1:0.50]
; BROADWELL-NEXT: andnl (%rdx), %edi, %eax # sched: [1:0.50]
; BROADWELL-NEXT: andnl (%rdx), %edi, %eax # sched: [6:0.50]
; BROADWELL-NEXT: addl %ecx, %eax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_andn_i32:
; SKYLAKE: # BB#0:
@ -137,9 +137,9 @@ define i64 @test_andn_i64(i64 %a0, i64 %a1, i64 *%a2) {
; BROADWELL-LABEL: test_andn_i64:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: andnq %rsi, %rdi, %rcx # sched: [1:0.50]
; BROADWELL-NEXT: andnq (%rdx), %rdi, %rax # sched: [1:0.50]
; BROADWELL-NEXT: andnq (%rdx), %rdi, %rax # sched: [6:0.50]
; BROADWELL-NEXT: addq %rcx, %rax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_andn_i64:
; SKYLAKE: # BB#0:
@ -186,10 +186,10 @@ define i32 @test_bextr_i32(i32 %a0, i32 %a1, i32 *%a2) {
;
; BROADWELL-LABEL: test_bextr_i32:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: bextrl %edi, (%rdx), %ecx # sched: [2:0.50]
; BROADWELL-NEXT: bextrl %edi, (%rdx), %ecx # sched: [7:0.50]
; BROADWELL-NEXT: bextrl %edi, %esi, %eax # sched: [2:0.50]
; BROADWELL-NEXT: addl %ecx, %eax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_bextr_i32:
; SKYLAKE: # BB#0:
@ -236,10 +236,10 @@ define i64 @test_bextr_i64(i64 %a0, i64 %a1, i64 *%a2) {
;
; BROADWELL-LABEL: test_bextr_i64:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: bextrq %rdi, (%rdx), %rcx # sched: [2:0.50]
; BROADWELL-NEXT: bextrq %rdi, (%rdx), %rcx # sched: [7:0.50]
; BROADWELL-NEXT: bextrq %rdi, %rsi, %rax # sched: [2:0.50]
; BROADWELL-NEXT: addq %rcx, %rax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_bextr_i64:
; SKYLAKE: # BB#0:
@ -286,10 +286,10 @@ define i32 @test_blsi_i32(i32 %a0, i32 *%a1) {
;
; BROADWELL-LABEL: test_blsi_i32:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: blsil (%rsi), %ecx # sched: [1:0.50]
; BROADWELL-NEXT: blsil (%rsi), %ecx # sched: [6:0.50]
; BROADWELL-NEXT: blsil %edi, %eax # sched: [1:0.50]
; BROADWELL-NEXT: addl %ecx, %eax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_blsi_i32:
; SKYLAKE: # BB#0:
@ -337,10 +337,10 @@ define i64 @test_blsi_i64(i64 %a0, i64 *%a1) {
;
; BROADWELL-LABEL: test_blsi_i64:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: blsiq (%rsi), %rcx # sched: [1:0.50]
; BROADWELL-NEXT: blsiq (%rsi), %rcx # sched: [6:0.50]
; BROADWELL-NEXT: blsiq %rdi, %rax # sched: [1:0.50]
; BROADWELL-NEXT: addq %rcx, %rax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_blsi_i64:
; SKYLAKE: # BB#0:
@ -388,10 +388,10 @@ define i32 @test_blsmsk_i32(i32 %a0, i32 *%a1) {
;
; BROADWELL-LABEL: test_blsmsk_i32:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: blsmskl (%rsi), %ecx # sched: [1:0.50]
; BROADWELL-NEXT: blsmskl (%rsi), %ecx # sched: [6:0.50]
; BROADWELL-NEXT: blsmskl %edi, %eax # sched: [1:0.50]
; BROADWELL-NEXT: addl %ecx, %eax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_blsmsk_i32:
; SKYLAKE: # BB#0:
@ -439,10 +439,10 @@ define i64 @test_blsmsk_i64(i64 %a0, i64 *%a1) {
;
; BROADWELL-LABEL: test_blsmsk_i64:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: blsmskq (%rsi), %rcx # sched: [1:0.50]
; BROADWELL-NEXT: blsmskq (%rsi), %rcx # sched: [6:0.50]
; BROADWELL-NEXT: blsmskq %rdi, %rax # sched: [1:0.50]
; BROADWELL-NEXT: addq %rcx, %rax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_blsmsk_i64:
; SKYLAKE: # BB#0:
@ -490,10 +490,10 @@ define i32 @test_blsr_i32(i32 %a0, i32 *%a1) {
;
; BROADWELL-LABEL: test_blsr_i32:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: blsrl (%rsi), %ecx # sched: [1:0.50]
; BROADWELL-NEXT: blsrl (%rsi), %ecx # sched: [6:0.50]
; BROADWELL-NEXT: blsrl %edi, %eax # sched: [1:0.50]
; BROADWELL-NEXT: addl %ecx, %eax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_blsr_i32:
; SKYLAKE: # BB#0:
@ -541,10 +541,10 @@ define i64 @test_blsr_i64(i64 %a0, i64 *%a1) {
;
; BROADWELL-LABEL: test_blsr_i64:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: blsrq (%rsi), %rcx # sched: [1:0.50]
; BROADWELL-NEXT: blsrq (%rsi), %rcx # sched: [6:0.50]
; BROADWELL-NEXT: blsrq %rdi, %rax # sched: [1:0.50]
; BROADWELL-NEXT: addq %rcx, %rax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_blsr_i64:
; SKYLAKE: # BB#0:
@ -594,11 +594,11 @@ define i16 @test_cttz_i16(i16 zeroext %a0, i16 *%a1) {
;
; BROADWELL-LABEL: test_cttz_i16:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: tzcntw (%rsi), %cx # sched: [3:1.00]
; BROADWELL-NEXT: tzcntw (%rsi), %cx # sched: [8:1.00]
; BROADWELL-NEXT: tzcntw %di, %ax # sched: [3:1.00]
; BROADWELL-NEXT: orl %ecx, %eax # sched: [1:0.25]
; BROADWELL-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill>
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_cttz_i16:
; SKYLAKE: # BB#0:
@ -648,10 +648,10 @@ define i32 @test_cttz_i32(i32 %a0, i32 *%a1) {
;
; BROADWELL-LABEL: test_cttz_i32:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: tzcntl (%rsi), %ecx # sched: [3:1.00]
; BROADWELL-NEXT: tzcntl (%rsi), %ecx # sched: [8:1.00]
; BROADWELL-NEXT: tzcntl %edi, %eax # sched: [3:1.00]
; BROADWELL-NEXT: orl %ecx, %eax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_cttz_i32:
; SKYLAKE: # BB#0:
@ -698,10 +698,10 @@ define i64 @test_cttz_i64(i64 %a0, i64 *%a1) {
;
; BROADWELL-LABEL: test_cttz_i64:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: tzcntq (%rsi), %rcx # sched: [3:1.00]
; BROADWELL-NEXT: tzcntq (%rsi), %rcx # sched: [8:1.00]
; BROADWELL-NEXT: tzcntq %rdi, %rax # sched: [3:1.00]
; BROADWELL-NEXT: orq %rcx, %rax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_cttz_i64:
; SKYLAKE: # BB#0:

View File

@ -23,10 +23,10 @@ define i32 @test_bzhi_i32(i32 %a0, i32 %a1, i32 *%a2) {
;
; BROADWELL-LABEL: test_bzhi_i32:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: bzhil %edi, (%rdx), %ecx # sched: [1:0.50]
; BROADWELL-NEXT: bzhil %edi, (%rdx), %ecx # sched: [6:0.50]
; BROADWELL-NEXT: bzhil %edi, %esi, %eax # sched: [1:0.50]
; BROADWELL-NEXT: addl %ecx, %eax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_bzhi_i32:
; SKYLAKE: # BB#0:
@ -73,10 +73,10 @@ define i64 @test_bzhi_i64(i64 %a0, i64 %a1, i64 *%a2) {
;
; BROADWELL-LABEL: test_bzhi_i64:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: bzhiq %rdi, (%rdx), %rcx # sched: [1:0.50]
; BROADWELL-NEXT: bzhiq %rdi, (%rdx), %rcx # sched: [6:0.50]
; BROADWELL-NEXT: bzhiq %rdi, %rsi, %rax # sched: [1:0.50]
; BROADWELL-NEXT: addq %rcx, %rax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_bzhi_i64:
; SKYLAKE: # BB#0:
@ -132,9 +132,9 @@ define i64 @test_mulx_i64(i64 %a0, i64 %a1, i64 *%a2) {
; BROADWELL-NEXT: movq %rdx, %rax # sched: [1:0.25]
; BROADWELL-NEXT: movq %rdi, %rdx # sched: [1:0.25]
; BROADWELL-NEXT: mulxq %rsi, %rsi, %rcx # sched: [4:1.00]
; BROADWELL-NEXT: mulxq (%rax), %rdx, %rax # sched: [4:1.00]
; BROADWELL-NEXT: mulxq (%rax), %rdx, %rax # sched: [9:1.00]
; BROADWELL-NEXT: orq %rcx, %rax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_mulx_i64:
; SKYLAKE: # BB#0:
@ -193,10 +193,10 @@ define i32 @test_pdep_i32(i32 %a0, i32 %a1, i32 *%a2) {
;
; BROADWELL-LABEL: test_pdep_i32:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pdepl (%rdx), %edi, %ecx # sched: [3:1.00]
; BROADWELL-NEXT: pdepl (%rdx), %edi, %ecx # sched: [8:1.00]
; BROADWELL-NEXT: pdepl %esi, %edi, %eax # sched: [3:1.00]
; BROADWELL-NEXT: addl %ecx, %eax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pdep_i32:
; SKYLAKE: # BB#0:
@ -243,10 +243,10 @@ define i64 @test_pdep_i64(i64 %a0, i64 %a1, i64 *%a2) {
;
; BROADWELL-LABEL: test_pdep_i64:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pdepq (%rdx), %rdi, %rcx # sched: [3:1.00]
; BROADWELL-NEXT: pdepq (%rdx), %rdi, %rcx # sched: [8:1.00]
; BROADWELL-NEXT: pdepq %rsi, %rdi, %rax # sched: [3:1.00]
; BROADWELL-NEXT: addq %rcx, %rax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pdep_i64:
; SKYLAKE: # BB#0:
@ -293,10 +293,10 @@ define i32 @test_pext_i32(i32 %a0, i32 %a1, i32 *%a2) {
;
; BROADWELL-LABEL: test_pext_i32:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pextl (%rdx), %edi, %ecx # sched: [3:1.00]
; BROADWELL-NEXT: pextl (%rdx), %edi, %ecx # sched: [8:1.00]
; BROADWELL-NEXT: pextl %esi, %edi, %eax # sched: [3:1.00]
; BROADWELL-NEXT: addl %ecx, %eax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pext_i32:
; SKYLAKE: # BB#0:
@ -343,10 +343,10 @@ define i64 @test_pext_i64(i64 %a0, i64 %a1, i64 *%a2) {
;
; BROADWELL-LABEL: test_pext_i64:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pextq (%rdx), %rdi, %rcx # sched: [3:1.00]
; BROADWELL-NEXT: pextq (%rdx), %rdi, %rcx # sched: [8:1.00]
; BROADWELL-NEXT: pextq %rsi, %rdi, %rax # sched: [3:1.00]
; BROADWELL-NEXT: addq %rcx, %rax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pext_i64:
; SKYLAKE: # BB#0:
@ -394,9 +394,9 @@ define i32 @test_rorx_i32(i32 %a0, i32 %a1, i32 *%a2) {
; BROADWELL-LABEL: test_rorx_i32:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: rorxl $5, %edi, %ecx # sched: [1:0.50]
; BROADWELL-NEXT: rorxl $5, (%rdx), %eax # sched: [1:0.50]
; BROADWELL-NEXT: rorxl $5, (%rdx), %eax # sched: [6:0.50]
; BROADWELL-NEXT: addl %ecx, %eax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_rorx_i32:
; SKYLAKE: # BB#0:
@ -447,9 +447,9 @@ define i64 @test_rorx_i64(i64 %a0, i64 %a1, i64 *%a2) {
; BROADWELL-LABEL: test_rorx_i64:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: rorxq $5, %rdi, %rcx # sched: [1:0.50]
; BROADWELL-NEXT: rorxq $5, (%rdx), %rax # sched: [1:0.50]
; BROADWELL-NEXT: rorxq $5, (%rdx), %rax # sched: [6:0.50]
; BROADWELL-NEXT: addq %rcx, %rax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_rorx_i64:
; SKYLAKE: # BB#0:
@ -500,9 +500,9 @@ define i32 @test_sarx_i32(i32 %a0, i32 %a1, i32 *%a2) {
; BROADWELL-LABEL: test_sarx_i32:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: sarxl %esi, %edi, %ecx # sched: [1:0.50]
; BROADWELL-NEXT: sarxl %esi, (%rdx), %eax # sched: [1:0.50]
; BROADWELL-NEXT: sarxl %esi, (%rdx), %eax # sched: [6:0.50]
; BROADWELL-NEXT: addl %ecx, %eax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_sarx_i32:
; SKYLAKE: # BB#0:
@ -549,9 +549,9 @@ define i64 @test_sarx_i64(i64 %a0, i64 %a1, i64 *%a2) {
; BROADWELL-LABEL: test_sarx_i64:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: sarxq %rsi, %rdi, %rcx # sched: [1:0.50]
; BROADWELL-NEXT: sarxq %rsi, (%rdx), %rax # sched: [1:0.50]
; BROADWELL-NEXT: sarxq %rsi, (%rdx), %rax # sched: [6:0.50]
; BROADWELL-NEXT: addq %rcx, %rax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_sarx_i64:
; SKYLAKE: # BB#0:
@ -598,9 +598,9 @@ define i32 @test_shlx_i32(i32 %a0, i32 %a1, i32 *%a2) {
; BROADWELL-LABEL: test_shlx_i32:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: shlxl %esi, %edi, %ecx # sched: [1:0.50]
; BROADWELL-NEXT: shlxl %esi, (%rdx), %eax # sched: [1:0.50]
; BROADWELL-NEXT: shlxl %esi, (%rdx), %eax # sched: [6:0.50]
; BROADWELL-NEXT: addl %ecx, %eax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_shlx_i32:
; SKYLAKE: # BB#0:
@ -647,9 +647,9 @@ define i64 @test_shlx_i64(i64 %a0, i64 %a1, i64 *%a2) {
; BROADWELL-LABEL: test_shlx_i64:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: shlxq %rsi, %rdi, %rcx # sched: [1:0.50]
; BROADWELL-NEXT: shlxq %rsi, (%rdx), %rax # sched: [1:0.50]
; BROADWELL-NEXT: shlxq %rsi, (%rdx), %rax # sched: [6:0.50]
; BROADWELL-NEXT: addq %rcx, %rax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_shlx_i64:
; SKYLAKE: # BB#0:
@ -696,9 +696,9 @@ define i32 @test_shrx_i32(i32 %a0, i32 %a1, i32 *%a2) {
; BROADWELL-LABEL: test_shrx_i32:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: shrxl %esi, %edi, %ecx # sched: [1:0.50]
; BROADWELL-NEXT: shrxl %esi, (%rdx), %eax # sched: [1:0.50]
; BROADWELL-NEXT: shrxl %esi, (%rdx), %eax # sched: [6:0.50]
; BROADWELL-NEXT: addl %ecx, %eax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_shrx_i32:
; SKYLAKE: # BB#0:
@ -745,9 +745,9 @@ define i64 @test_shrx_i64(i64 %a0, i64 %a1, i64 *%a2) {
; BROADWELL-LABEL: test_shrx_i64:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: shrxq %rsi, %rdi, %rcx # sched: [1:0.50]
; BROADWELL-NEXT: shrxq %rsi, (%rdx), %rax # sched: [1:0.50]
; BROADWELL-NEXT: shrxq %rsi, (%rdx), %rax # sched: [6:0.50]
; BROADWELL-NEXT: addq %rcx, %rax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_shrx_i64:
; SKYLAKE: # BB#0:

View File

@ -31,10 +31,10 @@ define <4 x float> @test_vcvtph2ps_128(<8 x i16> %a0, <8 x i16> *%a1) {
;
; BROADWELL-LABEL: test_vcvtph2ps_128:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vcvtph2ps (%rdi), %xmm1 # sched: [1:1.00]
; BROADWELL-NEXT: vcvtph2ps (%rdi), %xmm1 # sched: [6:1.00]
; BROADWELL-NEXT: vcvtph2ps %xmm0, %xmm0 # sched: [2:1.00]
; BROADWELL-NEXT: vaddps %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vcvtph2ps_128:
; SKYLAKE: # BB#0:
@ -88,10 +88,10 @@ define <8 x float> @test_vcvtph2ps_256(<8 x i16> %a0, <8 x i16> *%a1) {
;
; BROADWELL-LABEL: test_vcvtph2ps_256:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vcvtph2ps (%rdi), %ymm1 # sched: [1:1.00]
; BROADWELL-NEXT: vcvtph2ps (%rdi), %ymm1 # sched: [6:1.00]
; BROADWELL-NEXT: vcvtph2ps %xmm0, %ymm0 # sched: [2:1.00]
; BROADWELL-NEXT: vaddps %ymm0, %ymm1, %ymm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vcvtph2ps_256:
; SKYLAKE: # BB#0:
@ -144,7 +144,7 @@ define <8 x i16> @test_vcvtps2ph_128(<4 x float> %a0, <4 x float> %a1, <4 x i16>
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vcvtps2ph $0, %xmm0, %xmm0 # sched: [4:1.00]
; BROADWELL-NEXT: vcvtps2ph $0, %xmm1, (%rdi) # sched: [4:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vcvtps2ph_128:
; SKYLAKE: # BB#0:
@ -196,9 +196,9 @@ define <8 x i16> @test_vcvtps2ph_256(<8 x float> %a0, <8 x float> %a1, <8 x i16>
; BROADWELL-LABEL: test_vcvtps2ph_256:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vcvtps2ph $0, %ymm0, %xmm0 # sched: [6:1.00]
; BROADWELL-NEXT: vcvtps2ph $0, %ymm1, (%rdi) # sched: [6:1.00]
; BROADWELL-NEXT: vcvtps2ph $0, %ymm1, (%rdi) # sched: [4:1.00]
; BROADWELL-NEXT: vzeroupper # sched: [4:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vcvtps2ph_256:
; SKYLAKE: # BB#0:

View File

@ -31,8 +31,8 @@ define <2 x double> @test_vfmadd213pd(<2 x double> %a0, <2 x double> %a1, <2 x d
; BROADWELL-LABEL: test_vfmadd213pd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vfmadd213pd %xmm2, %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: vfmadd213pd (%rdi), %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vfmadd213pd (%rdi), %xmm1, %xmm0 # sched: [10:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vfmadd213pd:
; SKYLAKE: # BB#0:
@ -79,8 +79,8 @@ define <4 x double> @test_vfmadd213pd_ymm(<4 x double> %a0, <4 x double> %a1, <4
; BROADWELL-LABEL: test_vfmadd213pd_ymm:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vfmadd213pd %ymm2, %ymm1, %ymm0 # sched: [5:0.50]
; BROADWELL-NEXT: vfmadd213pd (%rdi), %ymm1, %ymm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vfmadd213pd (%rdi), %ymm1, %ymm0 # sched: [11:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vfmadd213pd_ymm:
; SKYLAKE: # BB#0:
@ -127,8 +127,8 @@ define <4 x float> @test_vfmadd213ps(<4 x float> %a0, <4 x float> %a1, <4 x floa
; BROADWELL-LABEL: test_vfmadd213ps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vfmadd213ps %xmm2, %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: vfmadd213ps (%rdi), %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vfmadd213ps (%rdi), %xmm1, %xmm0 # sched: [10:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vfmadd213ps:
; SKYLAKE: # BB#0:
@ -175,8 +175,8 @@ define <8 x float> @test_vfmadd213ps_ymm(<8 x float> %a0, <8 x float> %a1, <8 x
; BROADWELL-LABEL: test_vfmadd213ps_ymm:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vfmadd213ps %ymm2, %ymm1, %ymm0 # sched: [5:0.50]
; BROADWELL-NEXT: vfmadd213ps (%rdi), %ymm1, %ymm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vfmadd213ps (%rdi), %ymm1, %ymm0 # sched: [11:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vfmadd213ps_ymm:
; SKYLAKE: # BB#0:
@ -223,8 +223,8 @@ define <2 x double> @test_vfmadd213sd(<2 x double> %a0, <2 x double> %a1, <2 x d
; BROADWELL-LABEL: test_vfmadd213sd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vfmadd213sd %xmm2, %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: vfmadd213sd (%rdi), %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vfmadd213sd (%rdi), %xmm1, %xmm0 # sched: [10:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vfmadd213sd:
; SKYLAKE: # BB#0:
@ -271,8 +271,8 @@ define <4 x float> @test_vfmadd213ss(<4 x float> %a0, <4 x float> %a1, <4 x floa
; BROADWELL-LABEL: test_vfmadd213ss:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vfmadd213ss %xmm2, %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: vfmadd213ss (%rdi), %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vfmadd213ss (%rdi), %xmm1, %xmm0 # sched: [10:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vfmadd213ss:
; SKYLAKE: # BB#0:
@ -331,8 +331,8 @@ define <2 x double> @test_vfmaddsubpd(<2 x double> %a0, <2 x double> %a1, <2 x d
; BROADWELL-LABEL: test_vfmaddsubpd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vfmaddsub213pd %xmm2, %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: vfmaddsub213pd (%rdi), %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vfmaddsub213pd (%rdi), %xmm1, %xmm0 # sched: [10:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vfmaddsubpd:
; SKYLAKE: # BB#0:
@ -379,8 +379,8 @@ define <4 x double> @test_vfmaddsubpd_ymm(<4 x double> %a0, <4 x double> %a1, <4
; BROADWELL-LABEL: test_vfmaddsubpd_ymm:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vfmaddsub213pd %ymm2, %ymm1, %ymm0 # sched: [5:0.50]
; BROADWELL-NEXT: vfmaddsub213pd (%rdi), %ymm1, %ymm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vfmaddsub213pd (%rdi), %ymm1, %ymm0 # sched: [11:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vfmaddsubpd_ymm:
; SKYLAKE: # BB#0:
@ -427,8 +427,8 @@ define <4 x float> @test_vfmaddsubps(<4 x float> %a0, <4 x float> %a1, <4 x floa
; BROADWELL-LABEL: test_vfmaddsubps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vfmaddsub213ps %xmm2, %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: vfmaddsub213ps (%rdi), %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vfmaddsub213ps (%rdi), %xmm1, %xmm0 # sched: [10:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vfmaddsubps:
; SKYLAKE: # BB#0:
@ -475,8 +475,8 @@ define <8 x float> @test_vfmaddsubps_ymm(<8 x float> %a0, <8 x float> %a1, <8 x
; BROADWELL-LABEL: test_vfmaddsubps_ymm:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vfmaddsub213ps %ymm2, %ymm1, %ymm0 # sched: [5:0.50]
; BROADWELL-NEXT: vfmaddsub213ps (%rdi), %ymm1, %ymm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vfmaddsub213ps (%rdi), %ymm1, %ymm0 # sched: [11:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vfmaddsubps_ymm:
; SKYLAKE: # BB#0:
@ -535,8 +535,8 @@ define <2 x double> @test_vfmsubaddpd(<2 x double> %a0, <2 x double> %a1, <2 x d
; BROADWELL-LABEL: test_vfmsubaddpd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vfmsubadd213pd %xmm2, %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: vfmsubadd213pd (%rdi), %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vfmsubadd213pd (%rdi), %xmm1, %xmm0 # sched: [10:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vfmsubaddpd:
; SKYLAKE: # BB#0:
@ -583,8 +583,8 @@ define <4 x double> @test_vfmsubaddpd_ymm(<4 x double> %a0, <4 x double> %a1, <4
; BROADWELL-LABEL: test_vfmsubaddpd_ymm:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vfmsubadd213pd %ymm2, %ymm1, %ymm0 # sched: [5:0.50]
; BROADWELL-NEXT: vfmsubadd213pd (%rdi), %ymm1, %ymm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vfmsubadd213pd (%rdi), %ymm1, %ymm0 # sched: [11:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vfmsubaddpd_ymm:
; SKYLAKE: # BB#0:
@ -631,8 +631,8 @@ define <4 x float> @test_vfmsubaddps(<4 x float> %a0, <4 x float> %a1, <4 x floa
; BROADWELL-LABEL: test_vfmsubaddps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vfmsubadd213ps %xmm2, %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: vfmsubadd213ps (%rdi), %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vfmsubadd213ps (%rdi), %xmm1, %xmm0 # sched: [10:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vfmsubaddps:
; SKYLAKE: # BB#0:
@ -679,8 +679,8 @@ define <8 x float> @test_vfmsubaddps_ymm(<8 x float> %a0, <8 x float> %a1, <8 x
; BROADWELL-LABEL: test_vfmsubaddps_ymm:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vfmsubadd213ps %ymm2, %ymm1, %ymm0 # sched: [5:0.50]
; BROADWELL-NEXT: vfmsubadd213ps (%rdi), %ymm1, %ymm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vfmsubadd213ps (%rdi), %ymm1, %ymm0 # sched: [11:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vfmsubaddps_ymm:
; SKYLAKE: # BB#0:
@ -739,8 +739,8 @@ define <2 x double> @test_vfmsub213pd(<2 x double> %a0, <2 x double> %a1, <2 x d
; BROADWELL-LABEL: test_vfmsub213pd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vfmsub213pd %xmm2, %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: vfmsub213pd (%rdi), %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vfmsub213pd (%rdi), %xmm1, %xmm0 # sched: [10:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vfmsub213pd:
; SKYLAKE: # BB#0:
@ -787,8 +787,8 @@ define <4 x double> @test_vfmsub213pd_ymm(<4 x double> %a0, <4 x double> %a1, <4
; BROADWELL-LABEL: test_vfmsub213pd_ymm:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vfmsub213pd %ymm2, %ymm1, %ymm0 # sched: [5:0.50]
; BROADWELL-NEXT: vfmsub213pd (%rdi), %ymm1, %ymm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vfmsub213pd (%rdi), %ymm1, %ymm0 # sched: [11:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vfmsub213pd_ymm:
; SKYLAKE: # BB#0:
@ -835,8 +835,8 @@ define <4 x float> @test_vfmsub213ps(<4 x float> %a0, <4 x float> %a1, <4 x floa
; BROADWELL-LABEL: test_vfmsub213ps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vfmsub213ps %xmm2, %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: vfmsub213ps (%rdi), %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vfmsub213ps (%rdi), %xmm1, %xmm0 # sched: [10:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vfmsub213ps:
; SKYLAKE: # BB#0:
@ -883,8 +883,8 @@ define <8 x float> @test_vfmsub213ps_ymm(<8 x float> %a0, <8 x float> %a1, <8 x
; BROADWELL-LABEL: test_vfmsub213ps_ymm:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vfmsub213ps %ymm2, %ymm1, %ymm0 # sched: [5:0.50]
; BROADWELL-NEXT: vfmsub213ps (%rdi), %ymm1, %ymm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vfmsub213ps (%rdi), %ymm1, %ymm0 # sched: [11:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vfmsub213ps_ymm:
; SKYLAKE: # BB#0:
@ -931,8 +931,8 @@ define <2 x double> @test_vfmsub213sd(<2 x double> %a0, <2 x double> %a1, <2 x d
; BROADWELL-LABEL: test_vfmsub213sd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vfmsub213sd %xmm2, %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: vfmsub213sd (%rdi), %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vfmsub213sd (%rdi), %xmm1, %xmm0 # sched: [10:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vfmsub213sd:
; SKYLAKE: # BB#0:
@ -979,8 +979,8 @@ define <4 x float> @test_vfmsub213ss(<4 x float> %a0, <4 x float> %a1, <4 x floa
; BROADWELL-LABEL: test_vfmsub213ss:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vfmsub213ss %xmm2, %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: vfmsub213ss (%rdi), %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vfmsub213ss (%rdi), %xmm1, %xmm0 # sched: [10:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vfmsub213ss:
; SKYLAKE: # BB#0:
@ -1039,8 +1039,8 @@ define <2 x double> @test_vfnmadd213pd(<2 x double> %a0, <2 x double> %a1, <2 x
; BROADWELL-LABEL: test_vfnmadd213pd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vfnmadd213pd %xmm2, %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: vfnmadd213pd (%rdi), %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vfnmadd213pd (%rdi), %xmm1, %xmm0 # sched: [10:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vfnmadd213pd:
; SKYLAKE: # BB#0:
@ -1087,8 +1087,8 @@ define <4 x double> @test_vfnmadd213pd_ymm(<4 x double> %a0, <4 x double> %a1, <
; BROADWELL-LABEL: test_vfnmadd213pd_ymm:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vfnmadd213pd %ymm2, %ymm1, %ymm0 # sched: [5:0.50]
; BROADWELL-NEXT: vfnmadd213pd (%rdi), %ymm1, %ymm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vfnmadd213pd (%rdi), %ymm1, %ymm0 # sched: [11:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vfnmadd213pd_ymm:
; SKYLAKE: # BB#0:
@ -1135,8 +1135,8 @@ define <4 x float> @test_vfnmadd213ps(<4 x float> %a0, <4 x float> %a1, <4 x flo
; BROADWELL-LABEL: test_vfnmadd213ps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vfnmadd213ps %xmm2, %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: vfnmadd213ps (%rdi), %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vfnmadd213ps (%rdi), %xmm1, %xmm0 # sched: [10:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vfnmadd213ps:
; SKYLAKE: # BB#0:
@ -1183,8 +1183,8 @@ define <8 x float> @test_vfnmadd213ps_ymm(<8 x float> %a0, <8 x float> %a1, <8 x
; BROADWELL-LABEL: test_vfnmadd213ps_ymm:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vfnmadd213ps %ymm2, %ymm1, %ymm0 # sched: [5:0.50]
; BROADWELL-NEXT: vfnmadd213ps (%rdi), %ymm1, %ymm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vfnmadd213ps (%rdi), %ymm1, %ymm0 # sched: [11:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vfnmadd213ps_ymm:
; SKYLAKE: # BB#0:
@ -1231,8 +1231,8 @@ define <2 x double> @test_vfnmadd213sd(<2 x double> %a0, <2 x double> %a1, <2 x
; BROADWELL-LABEL: test_vfnmadd213sd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vfnmadd213sd %xmm2, %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: vfnmadd213sd (%rdi), %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vfnmadd213sd (%rdi), %xmm1, %xmm0 # sched: [10:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vfnmadd213sd:
; SKYLAKE: # BB#0:
@ -1279,8 +1279,8 @@ define <4 x float> @test_vfnmadd213ss(<4 x float> %a0, <4 x float> %a1, <4 x flo
; BROADWELL-LABEL: test_vfnmadd213ss:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vfnmadd213ss %xmm2, %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: vfnmadd213ss (%rdi), %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vfnmadd213ss (%rdi), %xmm1, %xmm0 # sched: [10:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vfnmadd213ss:
; SKYLAKE: # BB#0:
@ -1339,8 +1339,8 @@ define <2 x double> @test_vfnmsub213pd(<2 x double> %a0, <2 x double> %a1, <2 x
; BROADWELL-LABEL: test_vfnmsub213pd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vfnmsub213pd %xmm2, %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: vfnmsub213pd (%rdi), %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vfnmsub213pd (%rdi), %xmm1, %xmm0 # sched: [10:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vfnmsub213pd:
; SKYLAKE: # BB#0:
@ -1387,8 +1387,8 @@ define <4 x double> @test_vfnmsub213pd_ymm(<4 x double> %a0, <4 x double> %a1, <
; BROADWELL-LABEL: test_vfnmsub213pd_ymm:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vfnmsub213pd %ymm2, %ymm1, %ymm0 # sched: [5:0.50]
; BROADWELL-NEXT: vfnmsub213pd (%rdi), %ymm1, %ymm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vfnmsub213pd (%rdi), %ymm1, %ymm0 # sched: [11:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vfnmsub213pd_ymm:
; SKYLAKE: # BB#0:
@ -1435,8 +1435,8 @@ define <4 x float> @test_vfnmsub213ps(<4 x float> %a0, <4 x float> %a1, <4 x flo
; BROADWELL-LABEL: test_vfnmsub213ps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vfnmsub213ps %xmm2, %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: vfnmsub213ps (%rdi), %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vfnmsub213ps (%rdi), %xmm1, %xmm0 # sched: [10:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vfnmsub213ps:
; SKYLAKE: # BB#0:
@ -1483,8 +1483,8 @@ define <8 x float> @test_vfnmsub213ps_ymm(<8 x float> %a0, <8 x float> %a1, <8 x
; BROADWELL-LABEL: test_vfnmsub213ps_ymm:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vfnmsub213ps %ymm2, %ymm1, %ymm0 # sched: [5:0.50]
; BROADWELL-NEXT: vfnmsub213ps (%rdi), %ymm1, %ymm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vfnmsub213ps (%rdi), %ymm1, %ymm0 # sched: [11:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vfnmsub213ps_ymm:
; SKYLAKE: # BB#0:
@ -1531,8 +1531,8 @@ define <2 x double> @test_vfnmsub213sd(<2 x double> %a0, <2 x double> %a1, <2 x
; BROADWELL-LABEL: test_vfnmsub213sd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vfnmsub213sd %xmm2, %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: vfnmsub213sd (%rdi), %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vfnmsub213sd (%rdi), %xmm1, %xmm0 # sched: [10:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vfnmsub213sd:
; SKYLAKE: # BB#0:
@ -1579,8 +1579,8 @@ define <4 x float> @test_vfnmsub213ss(<4 x float> %a0, <4 x float> %a1, <4 x flo
; BROADWELL-LABEL: test_vfnmsub213ss:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vfnmsub213ss %xmm2, %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: vfnmsub213ss (%rdi), %xmm1, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vfnmsub213ss (%rdi), %xmm1, %xmm0 # sched: [10:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_vfnmsub213ss:
; SKYLAKE: # BB#0:

View File

@ -52,7 +52,7 @@ define i32 @test_lea_offset(i32) {
; BROADWELL: # BB#0:
; BROADWELL-NEXT: # kill: %EDI<def> %EDI<kill> %RDI<def>
; BROADWELL-NEXT: leal -24(%rdi), %eax # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_lea_offset:
; SKYLAKE: # BB#0:
@ -116,7 +116,7 @@ define i32 @test_lea_offset_big(i32) {
; BROADWELL: # BB#0:
; BROADWELL-NEXT: # kill: %EDI<def> %EDI<kill> %RDI<def>
; BROADWELL-NEXT: leal 1024(%rdi), %eax # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_lea_offset_big:
; SKYLAKE: # BB#0:
@ -187,7 +187,7 @@ define i32 @test_lea_add(i32, i32) {
; BROADWELL-NEXT: # kill: %ESI<def> %ESI<kill> %RSI<def>
; BROADWELL-NEXT: # kill: %EDI<def> %EDI<kill> %RDI<def>
; BROADWELL-NEXT: leal (%rdi,%rsi), %eax # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_lea_add:
; SKYLAKE: # BB#0:
@ -264,7 +264,7 @@ define i32 @test_lea_add_offset(i32, i32) {
; BROADWELL-NEXT: # kill: %EDI<def> %EDI<kill> %RDI<def>
; BROADWELL-NEXT: leal (%rdi,%rsi), %eax # sched: [1:0.50]
; BROADWELL-NEXT: addl $16, %eax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_lea_add_offset:
; SKYLAKE: # BB#0:
@ -347,7 +347,7 @@ define i32 @test_lea_add_offset_big(i32, i32) {
; BROADWELL-NEXT: leal (%rdi,%rsi), %eax # sched: [1:0.50]
; BROADWELL-NEXT: addl $-4096, %eax # imm = 0xF000
; BROADWELL-NEXT: # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_lea_add_offset_big:
; SKYLAKE: # BB#0:
@ -417,7 +417,7 @@ define i32 @test_lea_mul(i32) {
; BROADWELL: # BB#0:
; BROADWELL-NEXT: # kill: %EDI<def> %EDI<kill> %RDI<def>
; BROADWELL-NEXT: leal (%rdi,%rdi,2), %eax # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_lea_mul:
; SKYLAKE: # BB#0:
@ -485,7 +485,7 @@ define i32 @test_lea_mul_offset(i32) {
; BROADWELL-NEXT: # kill: %EDI<def> %EDI<kill> %RDI<def>
; BROADWELL-NEXT: leal (%rdi,%rdi,2), %eax # sched: [1:0.50]
; BROADWELL-NEXT: addl $-32, %eax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_lea_mul_offset:
; SKYLAKE: # BB#0:
@ -559,7 +559,7 @@ define i32 @test_lea_mul_offset_big(i32) {
; BROADWELL-NEXT: leal (%rdi,%rdi,8), %eax # sched: [1:0.50]
; BROADWELL-NEXT: addl $10000, %eax # imm = 0x2710
; BROADWELL-NEXT: # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_lea_mul_offset_big:
; SKYLAKE: # BB#0:
@ -632,7 +632,7 @@ define i32 @test_lea_add_scale(i32, i32) {
; BROADWELL-NEXT: # kill: %ESI<def> %ESI<kill> %RSI<def>
; BROADWELL-NEXT: # kill: %EDI<def> %EDI<kill> %RDI<def>
; BROADWELL-NEXT: leal (%rdi,%rsi,2), %eax # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_lea_add_scale:
; SKYLAKE: # BB#0:
@ -710,7 +710,7 @@ define i32 @test_lea_add_scale_offset(i32, i32) {
; BROADWELL-NEXT: # kill: %EDI<def> %EDI<kill> %RDI<def>
; BROADWELL-NEXT: leal (%rdi,%rsi,4), %eax # sched: [1:0.50]
; BROADWELL-NEXT: addl $96, %eax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_lea_add_scale_offset:
; SKYLAKE: # BB#0:
@ -794,7 +794,7 @@ define i32 @test_lea_add_scale_offset_big(i32, i32) {
; BROADWELL-NEXT: leal (%rdi,%rsi,8), %eax # sched: [1:0.50]
; BROADWELL-NEXT: addl $-1200, %eax # imm = 0xFB50
; BROADWELL-NEXT: # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_lea_add_scale_offset_big:
; SKYLAKE: # BB#0:

View File

@ -46,7 +46,7 @@ define i64 @test_lea_offset(i64) {
; BROADWELL-LABEL: test_lea_offset:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: leaq -24(%rdi), %rax # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_lea_offset:
; SKYLAKE: # BB#0:
@ -101,7 +101,7 @@ define i64 @test_lea_offset_big(i64) {
; BROADWELL-LABEL: test_lea_offset_big:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: leaq 1024(%rdi), %rax # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_lea_offset_big:
; SKYLAKE: # BB#0:
@ -157,7 +157,7 @@ define i64 @test_lea_add(i64, i64) {
; BROADWELL-LABEL: test_lea_add:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: leaq (%rdi,%rsi), %rax # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_lea_add:
; SKYLAKE: # BB#0:
@ -216,7 +216,7 @@ define i64 @test_lea_add_offset(i64, i64) {
; BROADWELL: # BB#0:
; BROADWELL-NEXT: leaq (%rdi,%rsi), %rax # sched: [1:0.50]
; BROADWELL-NEXT: addq $16, %rax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_lea_add_offset:
; SKYLAKE: # BB#0:
@ -281,7 +281,7 @@ define i64 @test_lea_add_offset_big(i64, i64) {
; BROADWELL-NEXT: leaq (%rdi,%rsi), %rax # sched: [1:0.50]
; BROADWELL-NEXT: addq $-4096, %rax # imm = 0xF000
; BROADWELL-NEXT: # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_lea_add_offset_big:
; SKYLAKE: # BB#0:
@ -339,7 +339,7 @@ define i64 @test_lea_mul(i64) {
; BROADWELL-LABEL: test_lea_mul:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: leaq (%rdi,%rdi,2), %rax # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_lea_mul:
; SKYLAKE: # BB#0:
@ -398,7 +398,7 @@ define i64 @test_lea_mul_offset(i64) {
; BROADWELL: # BB#0:
; BROADWELL-NEXT: leaq (%rdi,%rdi,2), %rax # sched: [1:0.50]
; BROADWELL-NEXT: addq $-32, %rax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_lea_mul_offset:
; SKYLAKE: # BB#0:
@ -463,7 +463,7 @@ define i64 @test_lea_mul_offset_big(i64) {
; BROADWELL-NEXT: leaq (%rdi,%rdi,8), %rax # sched: [1:0.50]
; BROADWELL-NEXT: addq $10000, %rax # imm = 0x2710
; BROADWELL-NEXT: # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_lea_mul_offset_big:
; SKYLAKE: # BB#0:
@ -521,7 +521,7 @@ define i64 @test_lea_add_scale(i64, i64) {
; BROADWELL-LABEL: test_lea_add_scale:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: leaq (%rdi,%rsi,2), %rax # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_lea_add_scale:
; SKYLAKE: # BB#0:
@ -581,7 +581,7 @@ define i64 @test_lea_add_scale_offset(i64, i64) {
; BROADWELL: # BB#0:
; BROADWELL-NEXT: leaq (%rdi,%rsi,4), %rax # sched: [1:0.50]
; BROADWELL-NEXT: addq $96, %rax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_lea_add_scale_offset:
; SKYLAKE: # BB#0:
@ -647,7 +647,7 @@ define i64 @test_lea_add_scale_offset_big(i64, i64) {
; BROADWELL-NEXT: leaq (%rdi,%rsi,8), %rax # sched: [1:0.50]
; BROADWELL-NEXT: addq $-1200, %rax # imm = 0xFB50
; BROADWELL-NEXT: # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_lea_add_scale_offset_big:
; SKYLAKE: # BB#0:

View File

@ -26,11 +26,11 @@ define i16 @test_ctlz_i16(i16 zeroext %a0, i16 *%a1) {
;
; BROADWELL-LABEL: test_ctlz_i16:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: lzcntw (%rsi), %cx # sched: [3:1.00]
; BROADWELL-NEXT: lzcntw (%rsi), %cx # sched: [8:1.00]
; BROADWELL-NEXT: lzcntw %di, %ax # sched: [3:1.00]
; BROADWELL-NEXT: orl %ecx, %eax # sched: [1:0.25]
; BROADWELL-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill>
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_ctlz_i16:
; SKYLAKE: # BB#0:
@ -80,10 +80,10 @@ define i32 @test_ctlz_i32(i32 %a0, i32 *%a1) {
;
; BROADWELL-LABEL: test_ctlz_i32:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: lzcntl (%rsi), %ecx # sched: [3:1.00]
; BROADWELL-NEXT: lzcntl (%rsi), %ecx # sched: [8:1.00]
; BROADWELL-NEXT: lzcntl %edi, %eax # sched: [3:1.00]
; BROADWELL-NEXT: orl %ecx, %eax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_ctlz_i32:
; SKYLAKE: # BB#0:
@ -130,10 +130,10 @@ define i64 @test_ctlz_i64(i64 %a0, i64 *%a1) {
;
; BROADWELL-LABEL: test_ctlz_i64:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: lzcntq (%rsi), %rcx # sched: [3:1.00]
; BROADWELL-NEXT: lzcntq (%rsi), %rcx # sched: [8:1.00]
; BROADWELL-NEXT: lzcntq %rdi, %rax # sched: [3:1.00]
; BROADWELL-NEXT: orq %rcx, %rax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_ctlz_i64:
; SKYLAKE: # BB#0:

View File

@ -54,11 +54,11 @@ define i64 @test_cvtpd2pi(<2 x double> %a0, <2 x double>* %a1) optsize {
;
; BROADWELL-LABEL: test_cvtpd2pi:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: cvtpd2pi (%rdi), %mm0 # sched: [4:1.00]
; BROADWELL-NEXT: cvtpd2pi (%rdi), %mm0 # sched: [9:1.00]
; BROADWELL-NEXT: cvtpd2pi %xmm0, %mm1 # sched: [4:1.00]
; BROADWELL-NEXT: por %mm1, %mm0 # sched: [1:0.33]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_cvtpd2pi:
; SKYLAKE: # BB#0:
@ -139,9 +139,9 @@ define <2 x double> @test_cvtpi2pd(x86_mmx %a0, x86_mmx* %a1) optsize {
; BROADWELL-LABEL: test_cvtpi2pd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: cvtpi2pd %mm0, %xmm0 # sched: [4:1.00]
; BROADWELL-NEXT: cvtpi2pd (%rdi), %xmm1 # sched: [4:1.00]
; BROADWELL-NEXT: cvtpi2pd (%rdi), %xmm1 # sched: [9:1.00]
; BROADWELL-NEXT: vaddpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_cvtpi2pd:
; SKYLAKE: # BB#0:
@ -217,9 +217,9 @@ define <4 x float> @test_cvtpi2ps(x86_mmx %a0, x86_mmx* %a1, <4 x float> %a2, <4
; BROADWELL-LABEL: test_cvtpi2ps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: cvtpi2ps %mm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: cvtpi2ps (%rdi), %xmm1 # sched: [3:1.00]
; BROADWELL-NEXT: cvtpi2ps (%rdi), %xmm1 # sched: [8:1.00]
; BROADWELL-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_cvtpi2ps:
; SKYLAKE: # BB#0:
@ -300,10 +300,10 @@ define i64 @test_cvtps2pi(<4 x float> %a0, <4 x float>* %a1) optsize {
; BROADWELL-LABEL: test_cvtps2pi:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: cvtps2pi %xmm0, %mm0 # sched: [4:1.00]
; BROADWELL-NEXT: cvtps2pi (%rdi), %mm1 # sched: [3:1.00]
; BROADWELL-NEXT: cvtps2pi (%rdi), %mm1 # sched: [8:1.00]
; BROADWELL-NEXT: por %mm0, %mm1 # sched: [1:0.33]
; BROADWELL-NEXT: movd %mm1, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_cvtps2pi:
; SKYLAKE: # BB#0:
@ -388,11 +388,11 @@ define i64 @test_cvttpd2pi(<2 x double> %a0, <2 x double>* %a1) optsize {
;
; BROADWELL-LABEL: test_cvttpd2pi:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: cvttpd2pi (%rdi), %mm0 # sched: [4:1.00]
; BROADWELL-NEXT: cvttpd2pi (%rdi), %mm0 # sched: [9:1.00]
; BROADWELL-NEXT: cvttpd2pi %xmm0, %mm1 # sched: [4:1.00]
; BROADWELL-NEXT: por %mm1, %mm0 # sched: [1:0.33]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_cvttpd2pi:
; SKYLAKE: # BB#0:
@ -478,10 +478,10 @@ define i64 @test_cvttps2pi(<4 x float> %a0, <4 x float>* %a1) optsize {
; BROADWELL-LABEL: test_cvttps2pi:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: cvttps2pi %xmm0, %mm0 # sched: [4:1.00]
; BROADWELL-NEXT: cvttps2pi (%rdi), %mm1 # sched: [3:1.00]
; BROADWELL-NEXT: cvttps2pi (%rdi), %mm1 # sched: [8:1.00]
; BROADWELL-NEXT: por %mm0, %mm1 # sched: [1:0.33]
; BROADWELL-NEXT: movd %mm1, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_cvttps2pi:
; SKYLAKE: # BB#0:
@ -552,7 +552,7 @@ define void @test_emms() optsize {
; BROADWELL-LABEL: test_emms:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: emms # sched: [31:10.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_emms:
; SKYLAKE: # BB#0:
@ -607,7 +607,7 @@ define void @test_maskmovq(x86_mmx %a0, x86_mmx %a1, i8* %a2) optsize {
; BROADWELL-LABEL: test_maskmovq:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: maskmovq %mm1, %mm0 # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_maskmovq:
; SKYLAKE: # BB#0:
@ -708,15 +708,15 @@ define i32 @test_movd(x86_mmx %a0, i32 %a1, i32 *%a2) {
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmovd %edi, %xmm0 # sched: [1:1.00]
; BROADWELL-NEXT: vmovq %xmm0, -{{[0-9]+}}(%rsp) # sched: [1:1.00]
; BROADWELL-NEXT: movq -{{[0-9]+}}(%rsp), %mm1 # sched: [1:0.50]
; BROADWELL-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero sched: [1:0.50]
; BROADWELL-NEXT: movq -{{[0-9]+}}(%rsp), %mm1 # sched: [5:0.50]
; BROADWELL-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero sched: [5:0.50]
; BROADWELL-NEXT: vmovlps %xmm0, -{{[0-9]+}}(%rsp) # sched: [1:1.00]
; BROADWELL-NEXT: paddd -{{[0-9]+}}(%rsp), %mm1 # sched: [1:0.50]
; BROADWELL-NEXT: paddd -{{[0-9]+}}(%rsp), %mm1 # sched: [6:0.50]
; BROADWELL-NEXT: paddd %mm1, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: movd %mm1, %ecx # sched: [1:1.00]
; BROADWELL-NEXT: movd %mm0, %eax # sched: [1:1.00]
; BROADWELL-NEXT: movl %ecx, (%rsi) # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_movd:
; SKYLAKE: # BB#0:
@ -829,7 +829,7 @@ define i64 @test_movdq2q(<2 x i64> %a0) optsize {
; BROADWELL-NEXT: movdq2q %xmm0, %mm0 # sched: [2:0.67]
; BROADWELL-NEXT: paddd %mm0, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_movdq2q:
; SKYLAKE: # BB#0:
@ -894,7 +894,7 @@ define void @test_movntq(x86_mmx* %a0, x86_mmx %a1) optsize {
; BROADWELL-LABEL: test_movntq:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: movntq %mm0, (%rdi) # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_movntq:
; SKYLAKE: # BB#0:
@ -960,10 +960,10 @@ define void @test_movq(i64 *%a0) {
;
; BROADWELL-LABEL: test_movq:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: movq (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: movq (%rdi), %mm0 # sched: [5:0.50]
; BROADWELL-NEXT: paddd %mm0, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: movq %mm0, (%rdi) # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_movq:
; SKYLAKE: # BB#0:
@ -1029,7 +1029,7 @@ define <2 x i64> @test_movq2dq(x86_mmx %a0) optsize {
; BROADWELL-LABEL: test_movq2dq:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: movq2dq %mm0, %xmm0 # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_movq2dq:
; SKYLAKE: # BB#0:
@ -1093,10 +1093,10 @@ define i64 @test_pabsb(x86_mmx *%a0) optsize {
;
; BROADWELL-LABEL: test_pabsb:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pabsb (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: pabsb (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: pabsb %mm0, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pabsb:
; SKYLAKE: # BB#0:
@ -1171,10 +1171,10 @@ define i64 @test_pabsd(x86_mmx *%a0) optsize {
;
; BROADWELL-LABEL: test_pabsd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pabsd (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: pabsd (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: pabsd %mm0, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pabsd:
; SKYLAKE: # BB#0:
@ -1249,10 +1249,10 @@ define i64 @test_pabsw(x86_mmx *%a0) optsize {
;
; BROADWELL-LABEL: test_pabsw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pabsw (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: pabsw (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: pabsw %mm0, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pabsw:
; SKYLAKE: # BB#0:
@ -1328,9 +1328,9 @@ define i64 @test_packssdw(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_packssdw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: packssdw %mm1, %mm0 # sched: [3:2.00]
; BROADWELL-NEXT: packssdw (%rdi), %mm0 # sched: [2:2.00]
; BROADWELL-NEXT: packssdw (%rdi), %mm0 # sched: [7:2.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_packssdw:
; SKYLAKE: # BB#0:
@ -1406,9 +1406,9 @@ define i64 @test_packsswb(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_packsswb:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: packsswb %mm1, %mm0 # sched: [3:2.00]
; BROADWELL-NEXT: packsswb (%rdi), %mm0 # sched: [2:2.00]
; BROADWELL-NEXT: packsswb (%rdi), %mm0 # sched: [7:2.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_packsswb:
; SKYLAKE: # BB#0:
@ -1484,9 +1484,9 @@ define i64 @test_packuswb(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_packuswb:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: packuswb %mm1, %mm0 # sched: [3:2.00]
; BROADWELL-NEXT: packuswb (%rdi), %mm0 # sched: [2:2.00]
; BROADWELL-NEXT: packuswb (%rdi), %mm0 # sched: [7:2.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_packuswb:
; SKYLAKE: # BB#0:
@ -1562,9 +1562,9 @@ define i64 @test_paddb(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_paddb:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: paddb %mm1, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: paddb (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: paddb (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_paddb:
; SKYLAKE: # BB#0:
@ -1640,9 +1640,9 @@ define i64 @test_paddd(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_paddd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: paddd %mm1, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: paddd (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: paddd (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_paddd:
; SKYLAKE: # BB#0:
@ -1718,9 +1718,9 @@ define i64 @test_paddq(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_paddq:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: paddq %mm1, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: paddq (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: paddq (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_paddq:
; SKYLAKE: # BB#0:
@ -1796,9 +1796,9 @@ define i64 @test_paddsb(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_paddsb:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: paddsb %mm1, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: paddsb (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: paddsb (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_paddsb:
; SKYLAKE: # BB#0:
@ -1874,9 +1874,9 @@ define i64 @test_paddsw(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_paddsw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: paddsw %mm1, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: paddsw (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: paddsw (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_paddsw:
; SKYLAKE: # BB#0:
@ -1952,9 +1952,9 @@ define i64 @test_paddusb(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_paddusb:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: paddusb %mm1, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: paddusb (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: paddusb (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_paddusb:
; SKYLAKE: # BB#0:
@ -2030,9 +2030,9 @@ define i64 @test_paddusw(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_paddusw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: paddusw %mm1, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: paddusw (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: paddusw (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_paddusw:
; SKYLAKE: # BB#0:
@ -2108,9 +2108,9 @@ define i64 @test_paddw(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_paddw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: paddw %mm1, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: paddw (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: paddw (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_paddw:
; SKYLAKE: # BB#0:
@ -2186,9 +2186,9 @@ define i64 @test_palignr(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_palignr:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: palignr $1, %mm1, %mm0 # sched: [1:1.00]
; BROADWELL-NEXT: palignr $1, (%rdi), %mm0 # sched: [1:1.00]
; BROADWELL-NEXT: palignr $1, (%rdi), %mm0 # sched: [6:1.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_palignr:
; SKYLAKE: # BB#0:
@ -2264,9 +2264,9 @@ define i64 @test_pand(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_pand:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pand %mm1, %mm0 # sched: [1:0.33]
; BROADWELL-NEXT: pand (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: pand (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pand:
; SKYLAKE: # BB#0:
@ -2342,9 +2342,9 @@ define i64 @test_pandn(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_pandn:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pandn %mm1, %mm0 # sched: [1:0.33]
; BROADWELL-NEXT: pandn (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: pandn (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pandn:
; SKYLAKE: # BB#0:
@ -2420,9 +2420,9 @@ define i64 @test_pavgb(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_pavgb:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pavgb %mm1, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: pavgb (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: pavgb (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pavgb:
; SKYLAKE: # BB#0:
@ -2498,9 +2498,9 @@ define i64 @test_pavgw(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_pavgw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pavgw %mm1, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: pavgw (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: pavgw (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pavgw:
; SKYLAKE: # BB#0:
@ -2576,9 +2576,9 @@ define i64 @test_pcmpeqb(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_pcmpeqb:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pcmpeqb %mm1, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: pcmpeqb (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: pcmpeqb (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pcmpeqb:
; SKYLAKE: # BB#0:
@ -2654,9 +2654,9 @@ define i64 @test_pcmpeqd(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_pcmpeqd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pcmpeqd %mm1, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: pcmpeqd (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: pcmpeqd (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pcmpeqd:
; SKYLAKE: # BB#0:
@ -2732,9 +2732,9 @@ define i64 @test_pcmpeqw(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_pcmpeqw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pcmpeqw %mm1, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: pcmpeqw (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: pcmpeqw (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pcmpeqw:
; SKYLAKE: # BB#0:
@ -2810,9 +2810,9 @@ define i64 @test_pcmpgtb(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_pcmpgtb:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pcmpgtb %mm1, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: pcmpgtb (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: pcmpgtb (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pcmpgtb:
; SKYLAKE: # BB#0:
@ -2888,9 +2888,9 @@ define i64 @test_pcmpgtd(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_pcmpgtd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pcmpgtd %mm1, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: pcmpgtd (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: pcmpgtd (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pcmpgtd:
; SKYLAKE: # BB#0:
@ -2966,9 +2966,9 @@ define i64 @test_pcmpgtw(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_pcmpgtw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pcmpgtw %mm1, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: pcmpgtw (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: pcmpgtw (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pcmpgtw:
; SKYLAKE: # BB#0:
@ -3034,7 +3034,7 @@ define i32 @test_pextrw(x86_mmx %a0) optsize {
; BROADWELL-LABEL: test_pextrw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pextrw $0, %mm0, %eax # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pextrw:
; SKYLAKE: # BB#0:
@ -3099,9 +3099,9 @@ define i64 @test_phaddd(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_phaddd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: phaddd %mm1, %mm0 # sched: [3:2.00]
; BROADWELL-NEXT: phaddd (%rdi), %mm0 # sched: [3:2.00]
; BROADWELL-NEXT: phaddd (%rdi), %mm0 # sched: [8:2.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_phaddd:
; SKYLAKE: # BB#0:
@ -3177,9 +3177,9 @@ define i64 @test_phaddsw(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_phaddsw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: phaddsw %mm1, %mm0 # sched: [3:2.00]
; BROADWELL-NEXT: phaddsw (%rdi), %mm0 # sched: [3:2.00]
; BROADWELL-NEXT: phaddsw (%rdi), %mm0 # sched: [8:2.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_phaddsw:
; SKYLAKE: # BB#0:
@ -3255,9 +3255,9 @@ define i64 @test_phaddw(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_phaddw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: phaddw %mm1, %mm0 # sched: [3:2.00]
; BROADWELL-NEXT: phaddw (%rdi), %mm0 # sched: [3:2.00]
; BROADWELL-NEXT: phaddw (%rdi), %mm0 # sched: [8:2.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_phaddw:
; SKYLAKE: # BB#0:
@ -3333,9 +3333,9 @@ define i64 @test_phsubd(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_phsubd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: phsubd %mm1, %mm0 # sched: [3:2.00]
; BROADWELL-NEXT: phsubd (%rdi), %mm0 # sched: [3:2.00]
; BROADWELL-NEXT: phsubd (%rdi), %mm0 # sched: [8:2.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_phsubd:
; SKYLAKE: # BB#0:
@ -3411,9 +3411,9 @@ define i64 @test_phsubsw(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_phsubsw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: phsubsw %mm1, %mm0 # sched: [3:2.00]
; BROADWELL-NEXT: phsubsw (%rdi), %mm0 # sched: [3:2.00]
; BROADWELL-NEXT: phsubsw (%rdi), %mm0 # sched: [8:2.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_phsubsw:
; SKYLAKE: # BB#0:
@ -3489,9 +3489,9 @@ define i64 @test_phsubw(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_phsubw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: phsubw %mm1, %mm0 # sched: [3:2.00]
; BROADWELL-NEXT: phsubw (%rdi), %mm0 # sched: [3:2.00]
; BROADWELL-NEXT: phsubw (%rdi), %mm0 # sched: [8:2.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_phsubw:
; SKYLAKE: # BB#0:
@ -3572,10 +3572,10 @@ define i64 @test_pinsrw(x86_mmx %a0, i32 %a1, i16* %a2) optsize {
; BROADWELL-LABEL: test_pinsrw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pinsrw $0, %edi, %mm0 # sched: [2:2.00]
; BROADWELL-NEXT: movswl (%rsi), %eax # sched: [4:0.50]
; BROADWELL-NEXT: movswl (%rsi), %eax # sched: [5:0.50]
; BROADWELL-NEXT: pinsrw $1, %eax, %mm0 # sched: [2:2.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pinsrw:
; SKYLAKE: # BB#0:
@ -3656,9 +3656,9 @@ define i64 @test_pmaddwd(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_pmaddwd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pmaddwd %mm1, %mm0 # sched: [5:1.00]
; BROADWELL-NEXT: pmaddwd (%rdi), %mm0 # sched: [5:1.00]
; BROADWELL-NEXT: pmaddwd (%rdi), %mm0 # sched: [10:1.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pmaddwd:
; SKYLAKE: # BB#0:
@ -3734,9 +3734,9 @@ define i64 @test_pmaddubsw(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_pmaddubsw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pmaddubsw %mm1, %mm0 # sched: [5:1.00]
; BROADWELL-NEXT: pmaddubsw (%rdi), %mm0 # sched: [5:1.00]
; BROADWELL-NEXT: pmaddubsw (%rdi), %mm0 # sched: [10:1.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pmaddubsw:
; SKYLAKE: # BB#0:
@ -3812,9 +3812,9 @@ define i64 @test_pmaxsw(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_pmaxsw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pmaxsw %mm1, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: pmaxsw (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: pmaxsw (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pmaxsw:
; SKYLAKE: # BB#0:
@ -3890,9 +3890,9 @@ define i64 @test_pmaxub(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_pmaxub:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pmaxub %mm1, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: pmaxub (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: pmaxub (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pmaxub:
; SKYLAKE: # BB#0:
@ -3968,9 +3968,9 @@ define i64 @test_pminsw(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_pminsw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pminsw %mm1, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: pminsw (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: pminsw (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pminsw:
; SKYLAKE: # BB#0:
@ -4046,9 +4046,9 @@ define i64 @test_pminub(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_pminub:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pminub %mm1, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: pminub (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: pminub (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pminub:
; SKYLAKE: # BB#0:
@ -4113,8 +4113,8 @@ define i32 @test_pmovmskb(x86_mmx %a0) optsize {
;
; BROADWELL-LABEL: test_pmovmskb:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pmovmskb %mm0, %eax # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: pmovmskb %mm0, %eax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pmovmskb:
; SKYLAKE: # BB#0:
@ -4179,9 +4179,9 @@ define i64 @test_pmulhrsw(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_pmulhrsw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pmulhrsw %mm1, %mm0 # sched: [5:1.00]
; BROADWELL-NEXT: pmulhrsw (%rdi), %mm0 # sched: [5:1.00]
; BROADWELL-NEXT: pmulhrsw (%rdi), %mm0 # sched: [10:1.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pmulhrsw:
; SKYLAKE: # BB#0:
@ -4257,9 +4257,9 @@ define i64 @test_pmulhw(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_pmulhw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pmulhw %mm1, %mm0 # sched: [5:1.00]
; BROADWELL-NEXT: pmulhw (%rdi), %mm0 # sched: [5:1.00]
; BROADWELL-NEXT: pmulhw (%rdi), %mm0 # sched: [10:1.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pmulhw:
; SKYLAKE: # BB#0:
@ -4335,9 +4335,9 @@ define i64 @test_pmulhuw(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_pmulhuw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pmulhuw %mm1, %mm0 # sched: [5:1.00]
; BROADWELL-NEXT: pmulhuw (%rdi), %mm0 # sched: [5:1.00]
; BROADWELL-NEXT: pmulhuw (%rdi), %mm0 # sched: [10:1.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pmulhuw:
; SKYLAKE: # BB#0:
@ -4413,9 +4413,9 @@ define i64 @test_pmullw(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_pmullw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pmullw %mm1, %mm0 # sched: [5:1.00]
; BROADWELL-NEXT: pmullw (%rdi), %mm0 # sched: [5:1.00]
; BROADWELL-NEXT: pmullw (%rdi), %mm0 # sched: [10:1.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pmullw:
; SKYLAKE: # BB#0:
@ -4491,9 +4491,9 @@ define i64 @test_pmuludq(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_pmuludq:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pmuludq %mm1, %mm0 # sched: [5:1.00]
; BROADWELL-NEXT: pmuludq (%rdi), %mm0 # sched: [5:1.00]
; BROADWELL-NEXT: pmuludq (%rdi), %mm0 # sched: [10:1.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pmuludq:
; SKYLAKE: # BB#0:
@ -4569,9 +4569,9 @@ define i64 @test_por(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_por:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: por %mm1, %mm0 # sched: [1:0.33]
; BROADWELL-NEXT: por (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: por (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_por:
; SKYLAKE: # BB#0:
@ -4647,9 +4647,9 @@ define i64 @test_psadbw(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_psadbw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: psadbw %mm1, %mm0 # sched: [5:1.00]
; BROADWELL-NEXT: psadbw (%rdi), %mm0 # sched: [5:1.00]
; BROADWELL-NEXT: psadbw (%rdi), %mm0 # sched: [10:1.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_psadbw:
; SKYLAKE: # BB#0:
@ -4725,9 +4725,9 @@ define i64 @test_pshufb(x86_mmx %a0, x86_mmx %a1, x86_mmx *%a2) optsize {
; BROADWELL-LABEL: test_pshufb:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pshufb %mm1, %mm0 # sched: [1:1.00]
; BROADWELL-NEXT: pshufb (%rdi), %mm0 # sched: [1:1.00]
; BROADWELL-NEXT: pshufb (%rdi), %mm0 # sched: [6:1.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pshufb:
; SKYLAKE: # BB#0:
@ -4802,10 +4802,10 @@ define i64 @test_pshufw(x86_mmx *%a0) optsize {
;
; BROADWELL-LABEL: test_pshufw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pshufw $0, (%rdi), %mm0 # mm0 = mem[0,0,0,0] sched: [1:1.00]
; BROADWELL-NEXT: pshufw $0, (%rdi), %mm0 # mm0 = mem[0,0,0,0] sched: [6:1.00]
; BROADWELL-NEXT: pshufw $0, %mm0, %mm0 # mm0 = mm0[0,0,0,0] sched: [1:1.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pshufw:
; SKYLAKE: # BB#0:
@ -4881,9 +4881,9 @@ define i64 @test_psignb(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_psignb:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: psignb %mm1, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: psignb (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: psignb (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_psignb:
; SKYLAKE: # BB#0:
@ -4959,9 +4959,9 @@ define i64 @test_psignd(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_psignd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: psignd %mm1, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: psignd (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: psignd (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_psignd:
; SKYLAKE: # BB#0:
@ -5037,9 +5037,9 @@ define i64 @test_psignw(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_psignw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: psignw %mm1, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: psignw (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: psignw (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_psignw:
; SKYLAKE: # BB#0:
@ -5120,10 +5120,10 @@ define i64 @test_pslld(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_pslld:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pslld %mm1, %mm0 # sched: [1:1.00]
; BROADWELL-NEXT: pslld (%rdi), %mm0 # sched: [1:1.00]
; BROADWELL-NEXT: pslld (%rdi), %mm0 # sched: [6:1.00]
; BROADWELL-NEXT: pslld $7, %mm0 # sched: [1:1.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pslld:
; SKYLAKE: # BB#0:
@ -5210,10 +5210,10 @@ define i64 @test_psllq(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_psllq:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: psllq %mm1, %mm0 # sched: [1:1.00]
; BROADWELL-NEXT: psllq (%rdi), %mm0 # sched: [1:1.00]
; BROADWELL-NEXT: psllq (%rdi), %mm0 # sched: [6:1.00]
; BROADWELL-NEXT: psllq $7, %mm0 # sched: [1:1.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_psllq:
; SKYLAKE: # BB#0:
@ -5300,10 +5300,10 @@ define i64 @test_psllw(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_psllw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: psllw %mm1, %mm0 # sched: [1:1.00]
; BROADWELL-NEXT: psllw (%rdi), %mm0 # sched: [1:1.00]
; BROADWELL-NEXT: psllw (%rdi), %mm0 # sched: [6:1.00]
; BROADWELL-NEXT: psllw $7, %mm0 # sched: [1:1.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_psllw:
; SKYLAKE: # BB#0:
@ -5390,10 +5390,10 @@ define i64 @test_psrad(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_psrad:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: psrad %mm1, %mm0 # sched: [1:1.00]
; BROADWELL-NEXT: psrad (%rdi), %mm0 # sched: [1:1.00]
; BROADWELL-NEXT: psrad (%rdi), %mm0 # sched: [6:1.00]
; BROADWELL-NEXT: psrad $7, %mm0 # sched: [1:1.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_psrad:
; SKYLAKE: # BB#0:
@ -5480,10 +5480,10 @@ define i64 @test_psraw(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_psraw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: psraw %mm1, %mm0 # sched: [1:1.00]
; BROADWELL-NEXT: psraw (%rdi), %mm0 # sched: [1:1.00]
; BROADWELL-NEXT: psraw (%rdi), %mm0 # sched: [6:1.00]
; BROADWELL-NEXT: psraw $7, %mm0 # sched: [1:1.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_psraw:
; SKYLAKE: # BB#0:
@ -5570,10 +5570,10 @@ define i64 @test_psrld(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_psrld:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: psrld %mm1, %mm0 # sched: [1:1.00]
; BROADWELL-NEXT: psrld (%rdi), %mm0 # sched: [1:1.00]
; BROADWELL-NEXT: psrld (%rdi), %mm0 # sched: [6:1.00]
; BROADWELL-NEXT: psrld $7, %mm0 # sched: [1:1.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_psrld:
; SKYLAKE: # BB#0:
@ -5660,10 +5660,10 @@ define i64 @test_psrlq(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_psrlq:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: psrlq %mm1, %mm0 # sched: [1:1.00]
; BROADWELL-NEXT: psrlq (%rdi), %mm0 # sched: [1:1.00]
; BROADWELL-NEXT: psrlq (%rdi), %mm0 # sched: [6:1.00]
; BROADWELL-NEXT: psrlq $7, %mm0 # sched: [1:1.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_psrlq:
; SKYLAKE: # BB#0:
@ -5750,10 +5750,10 @@ define i64 @test_psrlw(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_psrlw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: psrlw %mm1, %mm0 # sched: [1:1.00]
; BROADWELL-NEXT: psrlw (%rdi), %mm0 # sched: [1:1.00]
; BROADWELL-NEXT: psrlw (%rdi), %mm0 # sched: [6:1.00]
; BROADWELL-NEXT: psrlw $7, %mm0 # sched: [1:1.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_psrlw:
; SKYLAKE: # BB#0:
@ -5835,9 +5835,9 @@ define i64 @test_psubb(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_psubb:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: psubb %mm1, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: psubb (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: psubb (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_psubb:
; SKYLAKE: # BB#0:
@ -5913,9 +5913,9 @@ define i64 @test_psubd(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_psubd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: psubd %mm1, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: psubd (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: psubd (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_psubd:
; SKYLAKE: # BB#0:
@ -5991,9 +5991,9 @@ define i64 @test_psubq(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_psubq:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: psubq %mm1, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: psubq (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: psubq (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_psubq:
; SKYLAKE: # BB#0:
@ -6069,9 +6069,9 @@ define i64 @test_psubsb(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_psubsb:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: psubsb %mm1, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: psubsb (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: psubsb (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_psubsb:
; SKYLAKE: # BB#0:
@ -6147,9 +6147,9 @@ define i64 @test_psubsw(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_psubsw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: psubsw %mm1, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: psubsw (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: psubsw (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_psubsw:
; SKYLAKE: # BB#0:
@ -6225,9 +6225,9 @@ define i64 @test_psubusb(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_psubusb:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: psubusb %mm1, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: psubusb (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: psubusb (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_psubusb:
; SKYLAKE: # BB#0:
@ -6303,9 +6303,9 @@ define i64 @test_psubusw(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_psubusw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: psubusw %mm1, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: psubusw (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: psubusw (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_psubusw:
; SKYLAKE: # BB#0:
@ -6381,9 +6381,9 @@ define i64 @test_psubw(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_psubw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: psubw %mm1, %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: psubw (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: psubw (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_psubw:
; SKYLAKE: # BB#0:
@ -6459,9 +6459,9 @@ define i64 @test_punpckhbw(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_punpckhbw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: punpckhbw %mm1, %mm0 # mm0 = mm0[4],mm1[4],mm0[5],mm1[5],mm0[6],mm1[6],mm0[7],mm1[7] sched: [1:1.00]
; BROADWELL-NEXT: punpckhbw (%rdi), %mm0 # mm0 = mm0[4],mem[4],mm0[5],mem[5],mm0[6],mem[6],mm0[7],mem[7] sched: [1:1.00]
; BROADWELL-NEXT: punpckhbw (%rdi), %mm0 # mm0 = mm0[4],mem[4],mm0[5],mem[5],mm0[6],mem[6],mm0[7],mem[7] sched: [6:1.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_punpckhbw:
; SKYLAKE: # BB#0:
@ -6537,9 +6537,9 @@ define i64 @test_punpckhdq(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_punpckhdq:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: punpckhdq %mm1, %mm0 # mm0 = mm0[1],mm1[1] sched: [1:1.00]
; BROADWELL-NEXT: punpckhdq (%rdi), %mm0 # mm0 = mm0[1],mem[1] sched: [1:1.00]
; BROADWELL-NEXT: punpckhdq (%rdi), %mm0 # mm0 = mm0[1],mem[1] sched: [6:1.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_punpckhdq:
; SKYLAKE: # BB#0:
@ -6615,9 +6615,9 @@ define i64 @test_punpckhwd(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_punpckhwd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: punpckhwd %mm1, %mm0 # mm0 = mm0[2],mm1[2],mm0[3],mm1[3] sched: [1:1.00]
; BROADWELL-NEXT: punpckhwd (%rdi), %mm0 # mm0 = mm0[2],mem[2],mm0[3],mem[3] sched: [1:1.00]
; BROADWELL-NEXT: punpckhwd (%rdi), %mm0 # mm0 = mm0[2],mem[2],mm0[3],mem[3] sched: [6:1.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_punpckhwd:
; SKYLAKE: # BB#0:
@ -6693,9 +6693,9 @@ define i64 @test_punpcklbw(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_punpcklbw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: punpcklbw %mm1, %mm0 # mm0 = mm0[0],mm1[0],mm0[1],mm1[1],mm0[2],mm1[2],mm0[3],mm1[3] sched: [1:1.00]
; BROADWELL-NEXT: punpcklbw (%rdi), %mm0 # mm0 = mm0[0],mem[0],mm0[1],mem[1],mm0[2],mem[2],mm0[3],mem[3] sched: [1:1.00]
; BROADWELL-NEXT: punpcklbw (%rdi), %mm0 # mm0 = mm0[0],mem[0],mm0[1],mem[1],mm0[2],mem[2],mm0[3],mem[3] sched: [6:1.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_punpcklbw:
; SKYLAKE: # BB#0:
@ -6771,9 +6771,9 @@ define i64 @test_punpckldq(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_punpckldq:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: punpckldq %mm1, %mm0 # mm0 = mm0[0],mm1[0] sched: [1:1.00]
; BROADWELL-NEXT: punpckldq (%rdi), %mm0 # mm0 = mm0[0],mem[0] sched: [1:1.00]
; BROADWELL-NEXT: punpckldq (%rdi), %mm0 # mm0 = mm0[0],mem[0] sched: [6:1.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_punpckldq:
; SKYLAKE: # BB#0:
@ -6849,9 +6849,9 @@ define i64 @test_punpcklwd(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_punpcklwd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: punpcklwd %mm1, %mm0 # mm0 = mm0[0],mm1[0],mm0[1],mm1[1] sched: [1:1.00]
; BROADWELL-NEXT: punpcklwd (%rdi), %mm0 # mm0 = mm0[0],mem[0],mm0[1],mem[1] sched: [1:1.00]
; BROADWELL-NEXT: punpcklwd (%rdi), %mm0 # mm0 = mm0[0],mem[0],mm0[1],mem[1] sched: [6:1.00]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_punpcklwd:
; SKYLAKE: # BB#0:
@ -6927,9 +6927,9 @@ define i64 @test_pxor(x86_mmx %a0, x86_mmx %a1, x86_mmx* %a2) optsize {
; BROADWELL-LABEL: test_pxor:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: pxor %mm1, %mm0 # sched: [1:0.33]
; BROADWELL-NEXT: pxor (%rdi), %mm0 # sched: [1:0.50]
; BROADWELL-NEXT: pxor (%rdi), %mm0 # sched: [6:0.50]
; BROADWELL-NEXT: movd %mm0, %rax # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pxor:
; SKYLAKE: # BB#0:

View File

@ -40,9 +40,9 @@ define i16 @test_movbe_i16(i16 *%a0, i16 %a1, i16 *%a2) {
;
; BROADWELL-LABEL: test_movbe_i16:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: movbew (%rdi), %ax # sched: [1:0.50]
; BROADWELL-NEXT: movbew %si, (%rdx) # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: movbew (%rdi), %ax # sched: [6:0.50]
; BROADWELL-NEXT: movbew %si, (%rdx) # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_movbe_i16:
; SKYLAKE: # BB#0:
@ -100,9 +100,9 @@ define i32 @test_movbe_i32(i32 *%a0, i32 %a1, i32 *%a2) {
;
; BROADWELL-LABEL: test_movbe_i32:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: movbel (%rdi), %eax # sched: [1:0.50]
; BROADWELL-NEXT: movbel %esi, (%rdx) # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: movbel (%rdi), %eax # sched: [6:0.50]
; BROADWELL-NEXT: movbel %esi, (%rdx) # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_movbe_i32:
; SKYLAKE: # BB#0:
@ -160,9 +160,9 @@ define i64 @test_movbe_i64(i64 *%a0, i64 %a1, i64 *%a2) {
;
; BROADWELL-LABEL: test_movbe_i64:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: movbeq (%rdi), %rax # sched: [1:0.50]
; BROADWELL-NEXT: movbeq %rsi, (%rdx) # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: movbeq (%rdi), %rax # sched: [6:0.50]
; BROADWELL-NEXT: movbeq %rsi, (%rdx) # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_movbe_i64:
; SKYLAKE: # BB#0:

View File

@ -46,11 +46,11 @@ define i16 @test_ctpop_i16(i16 zeroext %a0, i16 *%a1) {
;
; BROADWELL-LABEL: test_ctpop_i16:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: popcntw (%rsi), %cx # sched: [3:1.00]
; BROADWELL-NEXT: popcntw (%rsi), %cx # sched: [8:1.00]
; BROADWELL-NEXT: popcntw %di, %ax # sched: [3:1.00]
; BROADWELL-NEXT: orl %ecx, %eax # sched: [1:0.25]
; BROADWELL-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill>
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_ctpop_i16:
; SKYLAKE: # BB#0:
@ -114,10 +114,10 @@ define i32 @test_ctpop_i32(i32 %a0, i32 *%a1) {
;
; BROADWELL-LABEL: test_ctpop_i32:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: popcntl (%rsi), %ecx # sched: [3:1.00]
; BROADWELL-NEXT: popcntl (%rsi), %ecx # sched: [8:1.00]
; BROADWELL-NEXT: popcntl %edi, %eax # sched: [3:1.00]
; BROADWELL-NEXT: orl %ecx, %eax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_ctpop_i32:
; SKYLAKE: # BB#0:
@ -178,10 +178,10 @@ define i64 @test_ctpop_i64(i64 %a0, i64 *%a1) {
;
; BROADWELL-LABEL: test_ctpop_i64:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: popcntq (%rsi), %rcx # sched: [3:1.00]
; BROADWELL-NEXT: popcntq (%rsi), %rcx # sched: [8:1.00]
; BROADWELL-NEXT: popcntq %rdi, %rax # sched: [3:1.00]
; BROADWELL-NEXT: orq %rcx, %rax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_ctpop_i64:
; SKYLAKE: # BB#0:

View File

@ -45,8 +45,8 @@ define <4 x float> @test_addps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a
; BROADWELL-LABEL: test_addps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: vaddps (%rdi), %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vaddps (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_addps:
; SKYLAKE: # BB#0:
@ -111,8 +111,8 @@ define float @test_addss(float %a0, float %a1, float *%a2) {
; BROADWELL-LABEL: test_addss:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vaddss %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: vaddss (%rdi), %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vaddss (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_addss:
; SKYLAKE: # BB#0:
@ -181,8 +181,8 @@ define <4 x float> @test_andps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a
; BROADWELL-LABEL: test_andps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vandps %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
; BROADWELL-NEXT: vandps (%rdi), %xmm0, %xmm0 # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vandps (%rdi), %xmm0, %xmm0 # sched: [6:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_andps:
; SKYLAKE: # BB#0:
@ -255,8 +255,8 @@ define <4 x float> @test_andnotps(<4 x float> %a0, <4 x float> %a1, <4 x float>
; BROADWELL-LABEL: test_andnotps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vandnps %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
; BROADWELL-NEXT: vandnps (%rdi), %xmm0, %xmm0 # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vandnps (%rdi), %xmm0, %xmm0 # sched: [6:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_andnotps:
; SKYLAKE: # BB#0:
@ -332,9 +332,9 @@ define <4 x float> @test_cmpps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a
; BROADWELL-LABEL: test_cmpps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vcmpeqps %xmm1, %xmm0, %xmm1 # sched: [3:1.00]
; BROADWELL-NEXT: vcmpeqps (%rdi), %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: vcmpeqps (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
; BROADWELL-NEXT: vorps %xmm0, %xmm1, %xmm0 # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_cmpps:
; SKYLAKE: # BB#0:
@ -407,8 +407,8 @@ define float @test_cmpss(float %a0, float %a1, float *%a2) {
; BROADWELL-LABEL: test_cmpss:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vcmpeqss %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: vcmpeqss (%rdi), %xmm0, %xmm0 # sched: [7:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vcmpeqss (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_cmpss:
; SKYLAKE: # BB#0:
@ -521,13 +521,13 @@ define i32 @test_comiss(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a2) {
; BROADWELL-NEXT: setnp %al # sched: [1:0.50]
; BROADWELL-NEXT: sete %cl # sched: [1:0.50]
; BROADWELL-NEXT: andb %al, %cl # sched: [1:0.25]
; BROADWELL-NEXT: vcomiss (%rdi), %xmm0 # sched: [7:1.00]
; BROADWELL-NEXT: vcomiss (%rdi), %xmm0 # sched: [8:1.00]
; BROADWELL-NEXT: setnp %al # sched: [1:0.50]
; BROADWELL-NEXT: sete %dl # sched: [1:0.50]
; BROADWELL-NEXT: andb %al, %dl # sched: [1:0.25]
; BROADWELL-NEXT: orb %cl, %dl # sched: [1:0.25]
; BROADWELL-NEXT: movzbl %dl, %eax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_comiss:
; SKYLAKE: # BB#0:
@ -631,9 +631,9 @@ define float @test_cvtsi2ss(i32 %a0, i32 *%a1) {
; BROADWELL-LABEL: test_cvtsi2ss:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vcvtsi2ssl %edi, %xmm0, %xmm0 # sched: [4:1.00]
; BROADWELL-NEXT: vcvtsi2ssl (%rsi), %xmm1, %xmm1 # sched: [8:1.00]
; BROADWELL-NEXT: vcvtsi2ssl (%rsi), %xmm1, %xmm1 # sched: [9:1.00]
; BROADWELL-NEXT: vaddss %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_cvtsi2ss:
; SKYLAKE: # BB#0:
@ -708,9 +708,9 @@ define float @test_cvtsi2ssq(i64 %a0, i64 *%a1) {
; BROADWELL-LABEL: test_cvtsi2ssq:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vcvtsi2ssq %rdi, %xmm0, %xmm0 # sched: [5:2.00]
; BROADWELL-NEXT: vcvtsi2ssq (%rsi), %xmm1, %xmm1 # sched: [8:1.00]
; BROADWELL-NEXT: vcvtsi2ssq (%rsi), %xmm1, %xmm1 # sched: [9:1.00]
; BROADWELL-NEXT: vaddss %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_cvtsi2ssq:
; SKYLAKE: # BB#0:
@ -785,9 +785,9 @@ define i32 @test_cvtss2si(float %a0, float *%a1) {
; BROADWELL-LABEL: test_cvtss2si:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vcvtss2si %xmm0, %ecx # sched: [4:1.00]
; BROADWELL-NEXT: vcvtss2si (%rdi), %eax # sched: [4:1.00]
; BROADWELL-NEXT: vcvtss2si (%rdi), %eax # sched: [9:1.00]
; BROADWELL-NEXT: addl %ecx, %eax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_cvtss2si:
; SKYLAKE: # BB#0:
@ -865,9 +865,9 @@ define i64 @test_cvtss2siq(float %a0, float *%a1) {
; BROADWELL-LABEL: test_cvtss2siq:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vcvtss2si %xmm0, %rcx # sched: [4:1.00]
; BROADWELL-NEXT: vcvtss2si (%rdi), %rax # sched: [4:1.00]
; BROADWELL-NEXT: vcvtss2si (%rdi), %rax # sched: [9:1.00]
; BROADWELL-NEXT: addq %rcx, %rax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_cvtss2siq:
; SKYLAKE: # BB#0:
@ -945,9 +945,9 @@ define i32 @test_cvttss2si(float %a0, float *%a1) {
; BROADWELL-LABEL: test_cvttss2si:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vcvttss2si %xmm0, %ecx # sched: [4:1.00]
; BROADWELL-NEXT: vcvttss2si (%rdi), %eax # sched: [4:1.00]
; BROADWELL-NEXT: vcvttss2si (%rdi), %eax # sched: [9:1.00]
; BROADWELL-NEXT: addl %ecx, %eax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_cvttss2si:
; SKYLAKE: # BB#0:
@ -1022,9 +1022,9 @@ define i64 @test_cvttss2siq(float %a0, float *%a1) {
; BROADWELL-LABEL: test_cvttss2siq:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vcvttss2si %xmm0, %rcx # sched: [4:1.00]
; BROADWELL-NEXT: vcvttss2si (%rdi), %rax # sched: [4:1.00]
; BROADWELL-NEXT: vcvttss2si (%rdi), %rax # sched: [9:1.00]
; BROADWELL-NEXT: addq %rcx, %rax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_cvttss2siq:
; SKYLAKE: # BB#0:
@ -1093,9 +1093,9 @@ define <4 x float> @test_divps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a
;
; BROADWELL-LABEL: test_divps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vdivps %xmm1, %xmm0, %xmm0 # sched: [13:1.00]
; BROADWELL-NEXT: vdivps (%rdi), %xmm0, %xmm0 # sched: [13:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vdivps %xmm1, %xmm0, %xmm0 # sched: [11:1.00]
; BROADWELL-NEXT: vdivps (%rdi), %xmm0, %xmm0 # sched: [16:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_divps:
; SKYLAKE: # BB#0:
@ -1159,9 +1159,9 @@ define float @test_divss(float %a0, float %a1, float *%a2) {
;
; BROADWELL-LABEL: test_divss:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vdivss %xmm1, %xmm0, %xmm0 # sched: [13:1.00]
; BROADWELL-NEXT: vdivss (%rdi), %xmm0, %xmm0 # sched: [13:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vdivss %xmm1, %xmm0, %xmm0 # sched: [11:1.00]
; BROADWELL-NEXT: vdivss (%rdi), %xmm0, %xmm0 # sched: [16:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_divss:
; SKYLAKE: # BB#0:
@ -1226,8 +1226,8 @@ define void @test_ldmxcsr(i32 %a0) {
; BROADWELL-LABEL: test_ldmxcsr:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: movl %edi, -{{[0-9]+}}(%rsp) # sched: [1:1.00]
; BROADWELL-NEXT: vldmxcsr -{{[0-9]+}}(%rsp) # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vldmxcsr -{{[0-9]+}}(%rsp) # sched: [7:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_ldmxcsr:
; SKYLAKE: # BB#0:
@ -1294,8 +1294,8 @@ define <4 x float> @test_maxps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a
; BROADWELL-LABEL: test_maxps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmaxps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: vmaxps (%rdi), %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vmaxps (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_maxps:
; SKYLAKE: # BB#0:
@ -1361,8 +1361,8 @@ define <4 x float> @test_maxss(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a
; BROADWELL-LABEL: test_maxss:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmaxss %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: vmaxss (%rdi), %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vmaxss (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_maxss:
; SKYLAKE: # BB#0:
@ -1428,8 +1428,8 @@ define <4 x float> @test_minps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a
; BROADWELL-LABEL: test_minps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vminps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: vminps (%rdi), %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vminps (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_minps:
; SKYLAKE: # BB#0:
@ -1495,8 +1495,8 @@ define <4 x float> @test_minss(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a
; BROADWELL-LABEL: test_minss:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vminss %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: vminss (%rdi), %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vminss (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_minss:
; SKYLAKE: # BB#0:
@ -1566,10 +1566,10 @@ define void @test_movaps(<4 x float> *%a0, <4 x float> *%a1) {
;
; BROADWELL-LABEL: test_movaps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmovaps (%rdi), %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: vmovaps (%rdi), %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: vaddps %xmm0, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: vmovaps %xmm0, (%rsi) # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_movaps:
; SKYLAKE: # BB#0:
@ -1641,7 +1641,7 @@ define <4 x float> @test_movhlps(<4 x float> %a0, <4 x float> %a1) {
; BROADWELL-LABEL: test_movhlps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vunpckhpd {{.*#+}} xmm0 = xmm1[1],xmm0[1] sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_movhlps:
; SKYLAKE: # BB#0:
@ -1708,10 +1708,10 @@ define void @test_movhps(<4 x float> %a0, <4 x float> %a1, x86_mmx *%a2) {
;
; BROADWELL-LABEL: test_movhps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0] sched: [1:1.00]
; BROADWELL-NEXT: vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0] sched: [6:1.00]
; BROADWELL-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: vpextrq $1, %xmm0, (%rdi) # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpextrq $1, %xmm0, (%rdi) # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_movhps:
; SKYLAKE: # BB#0:
@ -1787,7 +1787,7 @@ define <4 x float> @test_movlhps(<4 x float> %a0, <4 x float> %a1) {
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmovlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0] sched: [1:1.00]
; BROADWELL-NEXT: vaddps %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_movlhps:
; SKYLAKE: # BB#0:
@ -1855,10 +1855,10 @@ define void @test_movlps(<4 x float> %a0, <4 x float> %a1, x86_mmx *%a2) {
;
; BROADWELL-LABEL: test_movlps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmovlpd {{.*#+}} xmm1 = mem[0],xmm1[1] sched: [1:1.00]
; BROADWELL-NEXT: vmovlpd {{.*#+}} xmm1 = mem[0],xmm1[1] sched: [6:1.00]
; BROADWELL-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: vmovlps %xmm0, (%rdi) # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_movlps:
; SKYLAKE: # BB#0:
@ -1928,7 +1928,7 @@ define i32 @test_movmskps(<4 x float> %a0) {
; BROADWELL-LABEL: test_movmskps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmovmskps %xmm0, %eax # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_movmskps:
; SKYLAKE: # BB#0:
@ -1989,7 +1989,7 @@ define void @test_movntps(<4 x float> %a0, <4 x float> *%a1) {
; BROADWELL-LABEL: test_movntps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmovntps %xmm0, (%rdi) # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_movntps:
; SKYLAKE: # BB#0:
@ -2052,10 +2052,10 @@ define void @test_movss_mem(float* %a0, float* %a1) {
;
; BROADWELL-LABEL: test_movss_mem:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero sched: [1:0.50]
; BROADWELL-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero sched: [5:0.50]
; BROADWELL-NEXT: vaddss %xmm0, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: vmovss %xmm0, (%rsi) # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_movss_mem:
; SKYLAKE: # BB#0:
@ -2125,7 +2125,7 @@ define <4 x float> @test_movss_reg(<4 x float> %a0, <4 x float> %a1) {
; BROADWELL-LABEL: test_movss_reg:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vblendps {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3] sched: [1:0.33]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_movss_reg:
; SKYLAKE: # BB#0:
@ -2188,10 +2188,10 @@ define void @test_movups(<4 x float> *%a0, <4 x float> *%a1) {
;
; BROADWELL-LABEL: test_movups:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmovups (%rdi), %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: vmovups (%rdi), %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: vaddps %xmm0, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: vmovups %xmm0, (%rsi) # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_movups:
; SKYLAKE: # BB#0:
@ -2259,9 +2259,9 @@ define <4 x float> @test_mulps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a
;
; BROADWELL-LABEL: test_mulps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmulps %xmm1, %xmm0, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: vmulps (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vmulps %xmm1, %xmm0, %xmm0 # sched: [3:0.50]
; BROADWELL-NEXT: vmulps (%rdi), %xmm0, %xmm0 # sched: [8:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_mulps:
; SKYLAKE: # BB#0:
@ -2325,9 +2325,9 @@ define float @test_mulss(float %a0, float %a1, float *%a2) {
;
; BROADWELL-LABEL: test_mulss:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmulss %xmm1, %xmm0, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: vmulss (%rdi), %xmm0, %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vmulss %xmm1, %xmm0, %xmm0 # sched: [3:0.50]
; BROADWELL-NEXT: vmulss (%rdi), %xmm0, %xmm0 # sched: [8:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_mulss:
; SKYLAKE: # BB#0:
@ -2396,8 +2396,8 @@ define <4 x float> @test_orps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a2
; BROADWELL-LABEL: test_orps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vorps %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
; BROADWELL-NEXT: vorps (%rdi), %xmm0, %xmm0 # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vorps (%rdi), %xmm0, %xmm0 # sched: [6:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_orps:
; SKYLAKE: # BB#0:
@ -2466,8 +2466,8 @@ define void @test_prefetchnta(i8* %a0) {
;
; BROADWELL-LABEL: test_prefetchnta:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: prefetchnta (%rdi) # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: prefetchnta (%rdi) # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_prefetchnta:
; SKYLAKE: # BB#0:
@ -2534,9 +2534,9 @@ define <4 x float> @test_rcpps(<4 x float> %a0, <4 x float> *%a1) {
; BROADWELL-LABEL: test_rcpps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vrcpps %xmm0, %xmm0 # sched: [5:1.00]
; BROADWELL-NEXT: vrcpps (%rdi), %xmm1 # sched: [5:1.00]
; BROADWELL-NEXT: vrcpps (%rdi), %xmm1 # sched: [10:1.00]
; BROADWELL-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_rcpps:
; SKYLAKE: # BB#0:
@ -2619,10 +2619,10 @@ define <4 x float> @test_rcpss(float %a0, float *%a1) {
; BROADWELL-LABEL: test_rcpss:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vrcpss %xmm0, %xmm0, %xmm0 # sched: [5:1.00]
; BROADWELL-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero sched: [1:0.50]
; BROADWELL-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero sched: [5:0.50]
; BROADWELL-NEXT: vrcpss %xmm1, %xmm1, %xmm1 # sched: [5:1.00]
; BROADWELL-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_rcpss:
; SKYLAKE: # BB#0:
@ -2706,9 +2706,9 @@ define <4 x float> @test_rsqrtps(<4 x float> %a0, <4 x float> *%a1) {
; BROADWELL-LABEL: test_rsqrtps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vrsqrtps %xmm0, %xmm0 # sched: [5:1.00]
; BROADWELL-NEXT: vrsqrtps (%rdi), %xmm1 # sched: [5:1.00]
; BROADWELL-NEXT: vrsqrtps (%rdi), %xmm1 # sched: [10:1.00]
; BROADWELL-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_rsqrtps:
; SKYLAKE: # BB#0:
@ -2791,10 +2791,10 @@ define <4 x float> @test_rsqrtss(float %a0, float *%a1) {
; BROADWELL-LABEL: test_rsqrtss:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vrsqrtss %xmm0, %xmm0, %xmm0 # sched: [5:1.00]
; BROADWELL-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero sched: [1:0.50]
; BROADWELL-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero sched: [5:0.50]
; BROADWELL-NEXT: vrsqrtss %xmm1, %xmm1, %xmm1 # sched: [5:1.00]
; BROADWELL-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_rsqrtss:
; SKYLAKE: # BB#0:
@ -2871,8 +2871,8 @@ define void @test_sfence() {
;
; BROADWELL-LABEL: test_sfence:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: sfence # sched: [1:0.33]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: sfence # sched: [2:0.33]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_sfence:
; SKYLAKE: # BB#0:
@ -2936,8 +2936,8 @@ define <4 x float> @test_shufps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%
; BROADWELL-LABEL: test_shufps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[0,0] sched: [1:1.00]
; BROADWELL-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,3],mem[0,0] sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,3],mem[0,0] sched: [6:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_shufps:
; SKYLAKE: # BB#0:
@ -3008,9 +3008,9 @@ define <4 x float> @test_sqrtps(<4 x float> %a0, <4 x float> *%a1) {
; BROADWELL-LABEL: test_sqrtps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vsqrtps %xmm0, %xmm0 # sched: [14:1.00]
; BROADWELL-NEXT: vsqrtps (%rdi), %xmm1 # sched: [14:1.00]
; BROADWELL-NEXT: vsqrtps (%rdi), %xmm1 # sched: [19:1.00]
; BROADWELL-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_sqrtps:
; SKYLAKE: # BB#0:
@ -3093,10 +3093,10 @@ define <4 x float> @test_sqrtss(<4 x float> %a0, <4 x float> *%a1) {
; BROADWELL-LABEL: test_sqrtss:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vsqrtss %xmm0, %xmm0, %xmm0 # sched: [14:1.00]
; BROADWELL-NEXT: vmovaps (%rdi), %xmm1 # sched: [1:0.50]
; BROADWELL-NEXT: vmovaps (%rdi), %xmm1 # sched: [5:0.50]
; BROADWELL-NEXT: vsqrtss %xmm1, %xmm1, %xmm1 # sched: [14:1.00]
; BROADWELL-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_sqrtss:
; SKYLAKE: # BB#0:
@ -3170,9 +3170,9 @@ define i32 @test_stmxcsr() {
;
; BROADWELL-LABEL: test_stmxcsr:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vstmxcsr -{{[0-9]+}}(%rsp) # sched: [1:1.00]
; BROADWELL-NEXT: movl -{{[0-9]+}}(%rsp), %eax # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vstmxcsr -{{[0-9]+}}(%rsp) # sched: [2:1.00]
; BROADWELL-NEXT: movl -{{[0-9]+}}(%rsp), %eax # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_stmxcsr:
; SKYLAKE: # BB#0:
@ -3239,8 +3239,8 @@ define <4 x float> @test_subps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a
; BROADWELL-LABEL: test_subps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vsubps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: vsubps (%rdi), %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vsubps (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_subps:
; SKYLAKE: # BB#0:
@ -3305,8 +3305,8 @@ define float @test_subss(float %a0, float %a1, float *%a2) {
; BROADWELL-LABEL: test_subss:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vsubss %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: vsubss (%rdi), %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vsubss (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_subss:
; SKYLAKE: # BB#0:
@ -3414,13 +3414,13 @@ define i32 @test_ucomiss(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a2) {
; BROADWELL-NEXT: setnp %al # sched: [1:0.50]
; BROADWELL-NEXT: sete %cl # sched: [1:0.50]
; BROADWELL-NEXT: andb %al, %cl # sched: [1:0.25]
; BROADWELL-NEXT: vucomiss (%rdi), %xmm0 # sched: [7:1.00]
; BROADWELL-NEXT: vucomiss (%rdi), %xmm0 # sched: [8:1.00]
; BROADWELL-NEXT: setnp %al # sched: [1:0.50]
; BROADWELL-NEXT: sete %dl # sched: [1:0.50]
; BROADWELL-NEXT: andb %al, %dl # sched: [1:0.25]
; BROADWELL-NEXT: orb %cl, %dl # sched: [1:0.25]
; BROADWELL-NEXT: movzbl %dl, %eax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_ucomiss:
; SKYLAKE: # BB#0:
@ -3523,8 +3523,8 @@ define <4 x float> @test_unpckhps(<4 x float> %a0, <4 x float> %a1, <4 x float>
; BROADWELL-LABEL: test_unpckhps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vunpckhps {{.*#+}} xmm0 = xmm0[2],xmm1[2],xmm0[3],xmm1[3] sched: [1:1.00]
; BROADWELL-NEXT: vunpckhps {{.*#+}} xmm0 = xmm0[2],mem[2],xmm0[3],mem[3] sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vunpckhps {{.*#+}} xmm0 = xmm0[2],mem[2],xmm0[3],mem[3] sched: [6:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_unpckhps:
; SKYLAKE: # BB#0:
@ -3593,8 +3593,8 @@ define <4 x float> @test_unpcklps(<4 x float> %a0, <4 x float> %a1, <4 x float>
; BROADWELL-LABEL: test_unpcklps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vunpcklps {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1] sched: [1:1.00]
; BROADWELL-NEXT: vunpcklps {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[1],mem[1] sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vunpcklps {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[1],mem[1] sched: [6:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_unpcklps:
; SKYLAKE: # BB#0:
@ -3663,8 +3663,8 @@ define <4 x float> @test_xorps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a
; BROADWELL-LABEL: test_xorps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vxorps %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
; BROADWELL-NEXT: vxorps (%rdi), %xmm0, %xmm0 # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vxorps (%rdi), %xmm0, %xmm0 # sched: [6:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_xorps:
; SKYLAKE: # BB#0:

File diff suppressed because it is too large Load Diff

View File

@ -45,8 +45,8 @@ define <2 x double> @test_addsubpd(<2 x double> %a0, <2 x double> %a1, <2 x doub
; BROADWELL-LABEL: test_addsubpd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vaddsubpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: vaddsubpd (%rdi), %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vaddsubpd (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_addsubpd:
; SKYLAKE: # BB#0:
@ -112,8 +112,8 @@ define <4 x float> @test_addsubps(<4 x float> %a0, <4 x float> %a1, <4 x float>
; BROADWELL-LABEL: test_addsubps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vaddsubps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: vaddsubps (%rdi), %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vaddsubps (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_addsubps:
; SKYLAKE: # BB#0:
@ -179,8 +179,8 @@ define <2 x double> @test_haddpd(<2 x double> %a0, <2 x double> %a1, <2 x double
; BROADWELL-LABEL: test_haddpd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vhaddpd %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
; BROADWELL-NEXT: vhaddpd (%rdi), %xmm0, %xmm0 # sched: [5:2.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vhaddpd (%rdi), %xmm0, %xmm0 # sched: [10:2.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_haddpd:
; SKYLAKE: # BB#0:
@ -246,8 +246,8 @@ define <4 x float> @test_haddps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%
; BROADWELL-LABEL: test_haddps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vhaddps %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
; BROADWELL-NEXT: vhaddps (%rdi), %xmm0, %xmm0 # sched: [5:2.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vhaddps (%rdi), %xmm0, %xmm0 # sched: [10:2.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_haddps:
; SKYLAKE: # BB#0:
@ -313,8 +313,8 @@ define <2 x double> @test_hsubpd(<2 x double> %a0, <2 x double> %a1, <2 x double
; BROADWELL-LABEL: test_hsubpd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vhsubpd %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
; BROADWELL-NEXT: vhsubpd (%rdi), %xmm0, %xmm0 # sched: [5:2.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vhsubpd (%rdi), %xmm0, %xmm0 # sched: [10:2.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_hsubpd:
; SKYLAKE: # BB#0:
@ -380,8 +380,8 @@ define <4 x float> @test_hsubps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%
; BROADWELL-LABEL: test_hsubps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vhsubps %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
; BROADWELL-NEXT: vhsubps (%rdi), %xmm0, %xmm0 # sched: [5:2.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vhsubps (%rdi), %xmm0, %xmm0 # sched: [10:2.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_hsubps:
; SKYLAKE: # BB#0:
@ -443,8 +443,8 @@ define <16 x i8> @test_lddqu(i8* %a0) {
;
; BROADWELL-LABEL: test_lddqu:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vlddqu (%rdi), %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vlddqu (%rdi), %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_lddqu:
; SKYLAKE: # BB#0:
@ -511,7 +511,7 @@ define void @test_monitor(i8* %a0, i32 %a1, i32 %a2) {
; BROADWELL-NEXT: leaq (%rdi), %rax # sched: [1:0.50]
; BROADWELL-NEXT: movl %esi, %ecx # sched: [1:0.25]
; BROADWELL-NEXT: monitor # sched: [100:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_monitor:
; SKYLAKE: # BB#0:
@ -585,9 +585,9 @@ define <2 x double> @test_movddup(<2 x double> %a0, <2 x double> *%a1) {
; BROADWELL-LABEL: test_movddup:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmovddup {{.*#+}} xmm0 = xmm0[0,0] sched: [1:1.00]
; BROADWELL-NEXT: vmovddup {{.*#+}} xmm1 = mem[0,0] sched: [1:0.50]
; BROADWELL-NEXT: vmovddup {{.*#+}} xmm1 = mem[0,0] sched: [5:0.50]
; BROADWELL-NEXT: vsubpd %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_movddup:
; SKYLAKE: # BB#0:
@ -663,9 +663,9 @@ define <4 x float> @test_movshdup(<4 x float> %a0, <4 x float> *%a1) {
; BROADWELL-LABEL: test_movshdup:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3] sched: [1:1.00]
; BROADWELL-NEXT: vmovshdup {{.*#+}} xmm1 = mem[1,1,3,3] sched: [1:0.50]
; BROADWELL-NEXT: vmovshdup {{.*#+}} xmm1 = mem[1,1,3,3] sched: [5:0.50]
; BROADWELL-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_movshdup:
; SKYLAKE: # BB#0:
@ -741,9 +741,9 @@ define <4 x float> @test_movsldup(<4 x float> %a0, <4 x float> *%a1) {
; BROADWELL-LABEL: test_movsldup:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmovsldup {{.*#+}} xmm0 = xmm0[0,0,2,2] sched: [1:1.00]
; BROADWELL-NEXT: vmovsldup {{.*#+}} xmm1 = mem[0,0,2,2] sched: [1:0.50]
; BROADWELL-NEXT: vmovsldup {{.*#+}} xmm1 = mem[0,0,2,2] sched: [5:0.50]
; BROADWELL-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_movsldup:
; SKYLAKE: # BB#0:
@ -819,8 +819,8 @@ define void @test_mwait(i32 %a0, i32 %a1) {
; BROADWELL: # BB#0:
; BROADWELL-NEXT: movl %edi, %ecx # sched: [1:0.25]
; BROADWELL-NEXT: movl %esi, %eax # sched: [1:0.25]
; BROADWELL-NEXT: mwait # sched: [20:2.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: mwait # sched: [100:0.25]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_mwait:
; SKYLAKE: # BB#0:

View File

@ -43,8 +43,8 @@ define <2 x double> @test_blendpd(<2 x double> %a0, <2 x double> %a1, <2 x doubl
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vblendpd {{.*#+}} xmm0 = xmm0[0],xmm1[1] sched: [1:0.33]
; BROADWELL-NEXT: vaddpd %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: vblendpd {{.*#+}} xmm0 = xmm0[0],mem[1] sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vblendpd {{.*#+}} xmm0 = xmm0[0],mem[1] sched: [6:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_blendpd:
; SKYLAKE: # BB#0:
@ -109,8 +109,8 @@ define <4 x float> @test_blendps(<4 x float> %a0, <4 x float> %a1, <4 x float> *
; BROADWELL-LABEL: test_blendps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vblendps {{.*#+}} xmm0 = xmm0[0],xmm1[1,2],xmm0[3] sched: [1:0.33]
; BROADWELL-NEXT: vblendps {{.*#+}} xmm0 = xmm0[0],mem[1],xmm0[2,3] sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vblendps {{.*#+}} xmm0 = xmm0[0],mem[1],xmm0[2,3] sched: [6:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_blendps:
; SKYLAKE: # BB#0:
@ -175,8 +175,8 @@ define <2 x double> @test_blendvpd(<2 x double> %a0, <2 x double> %a1, <2 x doub
; BROADWELL-LABEL: test_blendvpd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vblendvpd %xmm2, %xmm1, %xmm0, %xmm0 # sched: [2:2.00]
; BROADWELL-NEXT: vblendvpd %xmm2, (%rdi), %xmm0, %xmm0 # sched: [2:2.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vblendvpd %xmm2, (%rdi), %xmm0, %xmm0 # sched: [7:2.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_blendvpd:
; SKYLAKE: # BB#0:
@ -242,8 +242,8 @@ define <4 x float> @test_blendvps(<4 x float> %a0, <4 x float> %a1, <4 x float>
; BROADWELL-LABEL: test_blendvps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vblendvps %xmm2, %xmm1, %xmm0, %xmm0 # sched: [2:2.00]
; BROADWELL-NEXT: vblendvps %xmm2, (%rdi), %xmm0, %xmm0 # sched: [2:2.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vblendvps %xmm2, (%rdi), %xmm0, %xmm0 # sched: [7:2.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_blendvps:
; SKYLAKE: # BB#0:
@ -303,8 +303,8 @@ define <2 x double> @test_dppd(<2 x double> %a0, <2 x double> %a1, <2 x double>
; BROADWELL-LABEL: test_dppd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vdppd $7, %xmm1, %xmm0, %xmm0 # sched: [9:1.00]
; BROADWELL-NEXT: vdppd $7, (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vdppd $7, (%rdi), %xmm0, %xmm0 # sched: [14:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_dppd:
; SKYLAKE: # BB#0:
@ -364,8 +364,8 @@ define <4 x float> @test_dpps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a2
; BROADWELL-LABEL: test_dpps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vdpps $7, %xmm1, %xmm0, %xmm0 # sched: [14:2.00]
; BROADWELL-NEXT: vdpps $7, (%rdi), %xmm0, %xmm0 # sched: [14:2.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vdpps $7, (%rdi), %xmm0, %xmm0 # sched: [19:2.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_dpps:
; SKYLAKE: # BB#0:
@ -425,8 +425,8 @@ define i32 @test_extractps(<4 x float> %a0, i32 *%a1) {
; BROADWELL-LABEL: test_extractps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vextractps $3, %xmm0, %eax # sched: [2:1.00]
; BROADWELL-NEXT: vextractps $1, %xmm0, (%rdi) # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vextractps $1, %xmm0, (%rdi) # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_extractps:
; SKYLAKE: # BB#0:
@ -487,8 +487,8 @@ define <4 x float> @test_insertps(<4 x float> %a0, <4 x float> %a1, float *%a2)
; BROADWELL-LABEL: test_insertps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vinsertps {{.*#+}} xmm0 = zero,xmm1[0],xmm0[2,3] sched: [1:1.00]
; BROADWELL-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0,1,2],mem[0] sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0,1,2],mem[0] sched: [6:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_insertps:
; SKYLAKE: # BB#0:
@ -543,8 +543,8 @@ define <2 x i64> @test_movntdqa(i8* %a0) {
;
; BROADWELL-LABEL: test_movntdqa:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmovntdqa (%rdi), %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vmovntdqa (%rdi), %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_movntdqa:
; SKYLAKE: # BB#0:
@ -598,8 +598,8 @@ define <8 x i16> @test_mpsadbw(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) {
; BROADWELL-LABEL: test_mpsadbw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vmpsadbw $7, %xmm1, %xmm0, %xmm0 # sched: [7:2.00]
; BROADWELL-NEXT: vmpsadbw $7, (%rdi), %xmm0, %xmm0 # sched: [7:2.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vmpsadbw $7, (%rdi), %xmm0, %xmm0 # sched: [12:2.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_mpsadbw:
; SKYLAKE: # BB#0:
@ -660,8 +660,8 @@ define <8 x i16> @test_packusdw(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) {
; BROADWELL-LABEL: test_packusdw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpackusdw %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
; BROADWELL-NEXT: vpackusdw (%rdi), %xmm0, %xmm0 # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpackusdw (%rdi), %xmm0, %xmm0 # sched: [6:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_packusdw:
; SKYLAKE: # BB#0:
@ -728,8 +728,8 @@ define <16 x i8> @test_pblendvb(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> %a2, <16
; BROADWELL-LABEL: test_pblendvb:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpblendvb %xmm2, %xmm1, %xmm0, %xmm0 # sched: [2:2.00]
; BROADWELL-NEXT: vpblendvb %xmm2, (%rdi), %xmm0, %xmm0 # sched: [2:2.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpblendvb %xmm2, (%rdi), %xmm0, %xmm0 # sched: [7:2.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pblendvb:
; SKYLAKE: # BB#0:
@ -789,8 +789,8 @@ define <8 x i16> @test_pblendw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) {
; BROADWELL-LABEL: test_pblendw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7] sched: [1:1.00]
; BROADWELL-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],mem[2,3],xmm0[4,5,6],mem[7] sched: [4:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],mem[2,3],xmm0[4,5,6],mem[7] sched: [6:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pblendw:
; SKYLAKE: # BB#0:
@ -849,8 +849,8 @@ define <2 x i64> @test_pcmpeqq(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> *%a2) {
; BROADWELL-LABEL: test_pcmpeqq:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: vpcmpeqq (%rdi), %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpcmpeqq (%rdi), %xmm0, %xmm0 # sched: [6:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pcmpeqq:
; SKYLAKE: # BB#0:
@ -913,8 +913,8 @@ define i32 @test_pextrb(<16 x i8> %a0, i8 *%a1) {
; BROADWELL-LABEL: test_pextrb:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpextrb $3, %xmm0, %eax # sched: [2:1.00]
; BROADWELL-NEXT: vpextrb $1, %xmm0, (%rdi) # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpextrb $1, %xmm0, (%rdi) # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pextrb:
; SKYLAKE: # BB#0:
@ -979,8 +979,8 @@ define i32 @test_pextrd(<4 x i32> %a0, i32 *%a1) {
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpaddd %xmm0, %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: vpextrd $3, %xmm0, %eax # sched: [2:1.00]
; BROADWELL-NEXT: vpextrd $1, %xmm0, (%rdi) # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpextrd $1, %xmm0, (%rdi) # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pextrd:
; SKYLAKE: # BB#0:
@ -1044,8 +1044,8 @@ define i64 @test_pextrq(<2 x i64> %a0, <2 x i64> %a1, i64 *%a2) {
; BROADWELL-LABEL: test_pextrq:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpextrq $1, %xmm0, %rax # sched: [2:1.00]
; BROADWELL-NEXT: vpextrq $1, %xmm0, (%rdi) # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpextrq $1, %xmm0, (%rdi) # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pextrq:
; SKYLAKE: # BB#0:
@ -1104,8 +1104,8 @@ define i32 @test_pextrw(<8 x i16> %a0, i16 *%a1) {
; BROADWELL-LABEL: test_pextrw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpextrw $3, %xmm0, %eax # sched: [2:1.00]
; BROADWELL-NEXT: vpextrw $1, %xmm0, (%rdi) # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpextrw $1, %xmm0, (%rdi) # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pextrw:
; SKYLAKE: # BB#0:
@ -1164,9 +1164,9 @@ define <8 x i16> @test_phminposuw(<8 x i16> *%a0) {
;
; BROADWELL-LABEL: test_phminposuw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vphminposuw (%rdi), %xmm0 # sched: [5:1.00]
; BROADWELL-NEXT: vphminposuw (%rdi), %xmm0 # sched: [10:1.00]
; BROADWELL-NEXT: vphminposuw %xmm0, %xmm0 # sched: [5:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_phminposuw:
; SKYLAKE: # BB#0:
@ -1226,8 +1226,8 @@ define <16 x i8> @test_pinsrb(<16 x i8> %a0, i8 %a1, i8 *%a2) {
; BROADWELL-LABEL: test_pinsrb:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpinsrb $1, %edi, %xmm0, %xmm0 # sched: [2:2.00]
; BROADWELL-NEXT: vpinsrb $3, (%rsi), %xmm0, %xmm0 # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpinsrb $3, (%rsi), %xmm0, %xmm0 # sched: [6:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pinsrb:
; SKYLAKE: # BB#0:
@ -1286,8 +1286,8 @@ define <4 x i32> @test_pinsrd(<4 x i32> %a0, i32 %a1, i32 *%a2) {
; BROADWELL-LABEL: test_pinsrd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpinsrd $1, %edi, %xmm0, %xmm0 # sched: [2:2.00]
; BROADWELL-NEXT: vpinsrd $3, (%rsi), %xmm0, %xmm0 # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpinsrd $3, (%rsi), %xmm0, %xmm0 # sched: [6:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pinsrd:
; SKYLAKE: # BB#0:
@ -1350,9 +1350,9 @@ define <2 x i64> @test_pinsrq(<2 x i64> %a0, <2 x i64> %a1, i64 %a2, i64 *%a3) {
; BROADWELL-LABEL: test_pinsrq:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpinsrq $1, %rdi, %xmm0, %xmm0 # sched: [2:2.00]
; BROADWELL-NEXT: vpinsrq $1, (%rsi), %xmm1, %xmm1 # sched: [1:1.00]
; BROADWELL-NEXT: vpinsrq $1, (%rsi), %xmm1, %xmm1 # sched: [6:1.00]
; BROADWELL-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pinsrq:
; SKYLAKE: # BB#0:
@ -1416,8 +1416,8 @@ define <16 x i8> @test_pmaxsb(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) {
; BROADWELL-LABEL: test_pmaxsb:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpmaxsb %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: vpmaxsb (%rdi), %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpmaxsb (%rdi), %xmm0, %xmm0 # sched: [6:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pmaxsb:
; SKYLAKE: # BB#0:
@ -1477,8 +1477,8 @@ define <4 x i32> @test_pmaxsd(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) {
; BROADWELL-LABEL: test_pmaxsd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: vpmaxsd (%rdi), %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpmaxsd (%rdi), %xmm0, %xmm0 # sched: [6:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pmaxsd:
; SKYLAKE: # BB#0:
@ -1538,8 +1538,8 @@ define <4 x i32> @test_pmaxud(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) {
; BROADWELL-LABEL: test_pmaxud:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpmaxud %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: vpmaxud (%rdi), %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpmaxud (%rdi), %xmm0, %xmm0 # sched: [6:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pmaxud:
; SKYLAKE: # BB#0:
@ -1599,8 +1599,8 @@ define <8 x i16> @test_pmaxuw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) {
; BROADWELL-LABEL: test_pmaxuw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpmaxuw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: vpmaxuw (%rdi), %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpmaxuw (%rdi), %xmm0, %xmm0 # sched: [6:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pmaxuw:
; SKYLAKE: # BB#0:
@ -1660,8 +1660,8 @@ define <16 x i8> @test_pminsb(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) {
; BROADWELL-LABEL: test_pminsb:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpminsb %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: vpminsb (%rdi), %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpminsb (%rdi), %xmm0, %xmm0 # sched: [6:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pminsb:
; SKYLAKE: # BB#0:
@ -1721,8 +1721,8 @@ define <4 x i32> @test_pminsd(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) {
; BROADWELL-LABEL: test_pminsd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpminsd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: vpminsd (%rdi), %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpminsd (%rdi), %xmm0, %xmm0 # sched: [6:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pminsd:
; SKYLAKE: # BB#0:
@ -1782,8 +1782,8 @@ define <4 x i32> @test_pminud(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) {
; BROADWELL-LABEL: test_pminud:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpminud %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: vpminud (%rdi), %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpminud (%rdi), %xmm0, %xmm0 # sched: [6:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pminud:
; SKYLAKE: # BB#0:
@ -1843,8 +1843,8 @@ define <8 x i16> @test_pminuw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) {
; BROADWELL-LABEL: test_pminuw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpminuw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: vpminuw (%rdi), %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpminuw (%rdi), %xmm0, %xmm0 # sched: [6:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pminuw:
; SKYLAKE: # BB#0:
@ -1909,9 +1909,9 @@ define <8 x i16> @test_pmovsxbw(<16 x i8> %a0, <8 x i8> *%a1) {
; BROADWELL-LABEL: test_pmovsxbw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpmovsxbw %xmm0, %xmm0 # sched: [1:1.00]
; BROADWELL-NEXT: vpmovsxbw (%rdi), %xmm1 # sched: [1:1.00]
; BROADWELL-NEXT: vpmovsxbw (%rdi), %xmm1 # sched: [6:1.00]
; BROADWELL-NEXT: vpaddw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pmovsxbw:
; SKYLAKE: # BB#0:
@ -1981,9 +1981,9 @@ define <4 x i32> @test_pmovsxbd(<16 x i8> %a0, <4 x i8> *%a1) {
; BROADWELL-LABEL: test_pmovsxbd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpmovsxbd %xmm0, %xmm0 # sched: [1:1.00]
; BROADWELL-NEXT: vpmovsxbd (%rdi), %xmm1 # sched: [1:1.00]
; BROADWELL-NEXT: vpmovsxbd (%rdi), %xmm1 # sched: [6:1.00]
; BROADWELL-NEXT: vpaddd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pmovsxbd:
; SKYLAKE: # BB#0:
@ -2053,9 +2053,9 @@ define <2 x i64> @test_pmovsxbq(<16 x i8> %a0, <2 x i8> *%a1) {
; BROADWELL-LABEL: test_pmovsxbq:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpmovsxbq %xmm0, %xmm0 # sched: [1:1.00]
; BROADWELL-NEXT: vpmovsxbq (%rdi), %xmm1 # sched: [1:1.00]
; BROADWELL-NEXT: vpmovsxbq (%rdi), %xmm1 # sched: [6:1.00]
; BROADWELL-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pmovsxbq:
; SKYLAKE: # BB#0:
@ -2125,9 +2125,9 @@ define <2 x i64> @test_pmovsxdq(<4 x i32> %a0, <2 x i32> *%a1) {
; BROADWELL-LABEL: test_pmovsxdq:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpmovsxdq %xmm0, %xmm0 # sched: [1:1.00]
; BROADWELL-NEXT: vpmovsxdq (%rdi), %xmm1 # sched: [1:1.00]
; BROADWELL-NEXT: vpmovsxdq (%rdi), %xmm1 # sched: [6:1.00]
; BROADWELL-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pmovsxdq:
; SKYLAKE: # BB#0:
@ -2197,9 +2197,9 @@ define <4 x i32> @test_pmovsxwd(<8 x i16> %a0, <4 x i16> *%a1) {
; BROADWELL-LABEL: test_pmovsxwd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpmovsxwd %xmm0, %xmm0 # sched: [1:1.00]
; BROADWELL-NEXT: vpmovsxwd (%rdi), %xmm1 # sched: [1:1.00]
; BROADWELL-NEXT: vpmovsxwd (%rdi), %xmm1 # sched: [6:1.00]
; BROADWELL-NEXT: vpaddd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pmovsxwd:
; SKYLAKE: # BB#0:
@ -2269,9 +2269,9 @@ define <2 x i64> @test_pmovsxwq(<8 x i16> %a0, <2 x i16> *%a1) {
; BROADWELL-LABEL: test_pmovsxwq:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpmovsxwq %xmm0, %xmm0 # sched: [1:1.00]
; BROADWELL-NEXT: vpmovsxwq (%rdi), %xmm1 # sched: [1:1.00]
; BROADWELL-NEXT: vpmovsxwq (%rdi), %xmm1 # sched: [6:1.00]
; BROADWELL-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pmovsxwq:
; SKYLAKE: # BB#0:
@ -2341,9 +2341,9 @@ define <8 x i16> @test_pmovzxbw(<16 x i8> %a0, <8 x i8> *%a1) {
; BROADWELL-LABEL: test_pmovzxbw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero sched: [1:1.00]
; BROADWELL-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: [1:1.00]
; BROADWELL-NEXT: vpmovzxbw {{.*#+}} xmm1 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero sched: [6:1.00]
; BROADWELL-NEXT: vpaddw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pmovzxbw:
; SKYLAKE: # BB#0:
@ -2413,9 +2413,9 @@ define <4 x i32> @test_pmovzxbd(<16 x i8> %a0, <4 x i8> *%a1) {
; BROADWELL-LABEL: test_pmovzxbd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpmovzxbd {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero sched: [1:1.00]
; BROADWELL-NEXT: vpmovzxbd {{.*#+}} xmm1 = mem[0],zero,zero,zero,mem[1],zero,zero,zero,mem[2],zero,zero,zero,mem[3],zero,zero,zero sched: [1:1.00]
; BROADWELL-NEXT: vpmovzxbd {{.*#+}} xmm1 = mem[0],zero,zero,zero,mem[1],zero,zero,zero,mem[2],zero,zero,zero,mem[3],zero,zero,zero sched: [6:1.00]
; BROADWELL-NEXT: vpaddd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pmovzxbd:
; SKYLAKE: # BB#0:
@ -2485,9 +2485,9 @@ define <2 x i64> @test_pmovzxbq(<16 x i8> %a0, <2 x i8> *%a1) {
; BROADWELL-LABEL: test_pmovzxbq:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero sched: [1:1.00]
; BROADWELL-NEXT: vpmovzxbq {{.*#+}} xmm1 = mem[0],zero,zero,zero,zero,zero,zero,zero,mem[1],zero,zero,zero,zero,zero,zero,zero sched: [1:1.00]
; BROADWELL-NEXT: vpmovzxbq {{.*#+}} xmm1 = mem[0],zero,zero,zero,zero,zero,zero,zero,mem[1],zero,zero,zero,zero,zero,zero,zero sched: [6:1.00]
; BROADWELL-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pmovzxbq:
; SKYLAKE: # BB#0:
@ -2557,9 +2557,9 @@ define <2 x i64> @test_pmovzxdq(<4 x i32> %a0, <2 x i32> *%a1) {
; BROADWELL-LABEL: test_pmovzxdq:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero sched: [1:1.00]
; BROADWELL-NEXT: vpmovzxdq {{.*#+}} xmm1 = mem[0],zero,mem[1],zero sched: [1:1.00]
; BROADWELL-NEXT: vpmovzxdq {{.*#+}} xmm1 = mem[0],zero,mem[1],zero sched: [6:1.00]
; BROADWELL-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pmovzxdq:
; SKYLAKE: # BB#0:
@ -2629,9 +2629,9 @@ define <4 x i32> @test_pmovzxwd(<8 x i16> %a0, <4 x i16> *%a1) {
; BROADWELL-LABEL: test_pmovzxwd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero sched: [1:1.00]
; BROADWELL-NEXT: vpmovzxwd {{.*#+}} xmm1 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero sched: [1:1.00]
; BROADWELL-NEXT: vpmovzxwd {{.*#+}} xmm1 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero sched: [6:1.00]
; BROADWELL-NEXT: vpaddd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pmovzxwd:
; SKYLAKE: # BB#0:
@ -2701,9 +2701,9 @@ define <2 x i64> @test_pmovzxwq(<8 x i16> %a0, <2 x i16> *%a1) {
; BROADWELL-LABEL: test_pmovzxwq:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero sched: [1:1.00]
; BROADWELL-NEXT: vpmovzxwq {{.*#+}} xmm1 = mem[0],zero,zero,zero,mem[1],zero,zero,zero sched: [1:1.00]
; BROADWELL-NEXT: vpmovzxwq {{.*#+}} xmm1 = mem[0],zero,zero,zero,mem[1],zero,zero,zero sched: [6:1.00]
; BROADWELL-NEXT: vpaddq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pmovzxwq:
; SKYLAKE: # BB#0:
@ -2768,8 +2768,8 @@ define <2 x i64> @test_pmuldq(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) {
; BROADWELL-LABEL: test_pmuldq:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpmuldq %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
; BROADWELL-NEXT: vpmuldq (%rdi), %xmm0, %xmm0 # sched: [5:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpmuldq (%rdi), %xmm0, %xmm0 # sched: [10:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pmuldq:
; SKYLAKE: # BB#0:
@ -2830,8 +2830,8 @@ define <4 x i32> @test_pmulld(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) {
; BROADWELL-LABEL: test_pmulld:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpmulld %xmm1, %xmm0, %xmm0 # sched: [10:2.00]
; BROADWELL-NEXT: vpmulld (%rdi), %xmm0, %xmm0 # sched: [10:2.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpmulld (%rdi), %xmm0, %xmm0 # sched: [15:2.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pmulld:
; SKYLAKE: # BB#0:
@ -2907,11 +2907,11 @@ define i32 @test_ptest(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> *%a2) {
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vptest %xmm1, %xmm0 # sched: [2:1.00]
; BROADWELL-NEXT: setb %al # sched: [1:0.50]
; BROADWELL-NEXT: vptest (%rdi), %xmm0 # sched: [2:1.00]
; BROADWELL-NEXT: vptest (%rdi), %xmm0 # sched: [7:1.00]
; BROADWELL-NEXT: setb %cl # sched: [1:0.50]
; BROADWELL-NEXT: andb %al, %cl # sched: [1:0.25]
; BROADWELL-NEXT: movzbl %cl, %eax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_ptest:
; SKYLAKE: # BB#0:
@ -2992,10 +2992,10 @@ define <2 x double> @test_roundpd(<2 x double> %a0, <2 x double> *%a1) {
;
; BROADWELL-LABEL: test_roundpd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vroundpd $7, %xmm0, %xmm0 # sched: [5:1.25]
; BROADWELL-NEXT: vroundpd $7, (%rdi), %xmm1 # sched: [6:2.00]
; BROADWELL-NEXT: vroundpd $7, %xmm0, %xmm0 # sched: [6:0.50]
; BROADWELL-NEXT: vroundpd $7, (%rdi), %xmm1 # sched: [11:2.00]
; BROADWELL-NEXT: vaddpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_roundpd:
; SKYLAKE: # BB#0:
@ -3064,10 +3064,10 @@ define <4 x float> @test_roundps(<4 x float> %a0, <4 x float> *%a1) {
;
; BROADWELL-LABEL: test_roundps:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vroundps $7, %xmm0, %xmm0 # sched: [5:1.25]
; BROADWELL-NEXT: vroundps $7, (%rdi), %xmm1 # sched: [6:2.00]
; BROADWELL-NEXT: vroundps $7, %xmm0, %xmm0 # sched: [6:0.50]
; BROADWELL-NEXT: vroundps $7, (%rdi), %xmm1 # sched: [11:2.00]
; BROADWELL-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_roundps:
; SKYLAKE: # BB#0:
@ -3137,10 +3137,10 @@ define <2 x double> @test_roundsd(<2 x double> %a0, <2 x double> %a1, <2 x doubl
;
; BROADWELL-LABEL: test_roundsd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vroundsd $7, %xmm1, %xmm0, %xmm1 # sched: [5:1.25]
; BROADWELL-NEXT: vroundsd $7, (%rdi), %xmm0, %xmm0 # sched: [6:2.00]
; BROADWELL-NEXT: vroundsd $7, %xmm1, %xmm0, %xmm1 # sched: [6:0.50]
; BROADWELL-NEXT: vroundsd $7, (%rdi), %xmm0, %xmm0 # sched: [11:2.00]
; BROADWELL-NEXT: vaddpd %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_roundsd:
; SKYLAKE: # BB#0:
@ -3210,10 +3210,10 @@ define <4 x float> @test_roundss(<4 x float> %a0, <4 x float> %a1, <4 x float> *
;
; BROADWELL-LABEL: test_roundss:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vroundss $7, %xmm1, %xmm0, %xmm1 # sched: [5:1.25]
; BROADWELL-NEXT: vroundss $7, (%rdi), %xmm0, %xmm0 # sched: [6:2.00]
; BROADWELL-NEXT: vroundss $7, %xmm1, %xmm0, %xmm1 # sched: [6:0.50]
; BROADWELL-NEXT: vroundss $7, (%rdi), %xmm0, %xmm0 # sched: [11:2.00]
; BROADWELL-NEXT: vaddps %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_roundss:
; SKYLAKE: # BB#0:

View File

@ -42,9 +42,9 @@ define i32 @crc32_32_8(i32 %a0, i8 %a1, i8 *%a2) {
; BROADWELL-LABEL: crc32_32_8:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: crc32b %sil, %edi # sched: [3:1.00]
; BROADWELL-NEXT: crc32b (%rdx), %edi # sched: [7:1.00]
; BROADWELL-NEXT: crc32b (%rdx), %edi # sched: [8:1.00]
; BROADWELL-NEXT: movl %edi, %eax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: crc32_32_8:
; SKYLAKE: # BB#0:
@ -112,9 +112,9 @@ define i32 @crc32_32_16(i32 %a0, i16 %a1, i16 *%a2) {
; BROADWELL-LABEL: crc32_32_16:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: crc32w %si, %edi # sched: [3:1.00]
; BROADWELL-NEXT: crc32w (%rdx), %edi # sched: [7:1.00]
; BROADWELL-NEXT: crc32w (%rdx), %edi # sched: [8:1.00]
; BROADWELL-NEXT: movl %edi, %eax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: crc32_32_16:
; SKYLAKE: # BB#0:
@ -182,9 +182,9 @@ define i32 @crc32_32_32(i32 %a0, i32 %a1, i32 *%a2) {
; BROADWELL-LABEL: crc32_32_32:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: crc32l %esi, %edi # sched: [3:1.00]
; BROADWELL-NEXT: crc32l (%rdx), %edi # sched: [7:1.00]
; BROADWELL-NEXT: crc32l (%rdx), %edi # sched: [8:1.00]
; BROADWELL-NEXT: movl %edi, %eax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: crc32_32_32:
; SKYLAKE: # BB#0:
@ -252,9 +252,9 @@ define i64 @crc32_64_8(i64 %a0, i8 %a1, i8 *%a2) nounwind {
; BROADWELL-LABEL: crc32_64_8:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: crc32b %sil, %edi # sched: [3:1.00]
; BROADWELL-NEXT: crc32b (%rdx), %edi # sched: [7:1.00]
; BROADWELL-NEXT: crc32b (%rdx), %edi # sched: [8:1.00]
; BROADWELL-NEXT: movq %rdi, %rax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: crc32_64_8:
; SKYLAKE: # BB#0:
@ -322,9 +322,9 @@ define i64 @crc32_64_64(i64 %a0, i64 %a1, i64 *%a2) {
; BROADWELL-LABEL: crc32_64_64:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: crc32q %rsi, %rdi # sched: [3:1.00]
; BROADWELL-NEXT: crc32q (%rdx), %rdi # sched: [7:1.00]
; BROADWELL-NEXT: crc32q (%rdx), %rdi # sched: [8:1.00]
; BROADWELL-NEXT: movq %rdi, %rax # sched: [1:0.25]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: crc32_64_64:
; SKYLAKE: # BB#0:
@ -421,10 +421,10 @@ define i32 @test_pcmpestri(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) {
; BROADWELL-NEXT: movl %ecx, %esi # sched: [1:0.25]
; BROADWELL-NEXT: movl $7, %eax # sched: [1:0.25]
; BROADWELL-NEXT: movl $7, %edx # sched: [1:0.25]
; BROADWELL-NEXT: vpcmpestri $7, (%rdi), %xmm0 # sched: [18:4.00]
; BROADWELL-NEXT: vpcmpestri $7, (%rdi), %xmm0 # sched: [23:4.00]
; BROADWELL-NEXT: # kill: %ECX<def> %ECX<kill> %RCX<def>
; BROADWELL-NEXT: leal (%rcx,%rsi), %eax # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pcmpestri:
; SKYLAKE: # BB#0:
@ -533,8 +533,8 @@ define <16 x i8> @test_pcmpestrm(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) {
; BROADWELL-NEXT: vpcmpestrm $7, %xmm1, %xmm0 # sched: [19:4.00]
; BROADWELL-NEXT: movl $7, %eax # sched: [1:0.25]
; BROADWELL-NEXT: movl $7, %edx # sched: [1:0.25]
; BROADWELL-NEXT: vpcmpestrm $7, (%rdi), %xmm0 # sched: [19:4.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpcmpestrm $7, (%rdi), %xmm0 # sched: [24:4.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pcmpestrm:
; SKYLAKE: # BB#0:
@ -623,10 +623,10 @@ define i32 @test_pcmpistri(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) {
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpcmpistri $7, %xmm1, %xmm0 # sched: [11:3.00]
; BROADWELL-NEXT: movl %ecx, %eax # sched: [1:0.25]
; BROADWELL-NEXT: vpcmpistri $7, (%rdi), %xmm0 # sched: [11:3.00]
; BROADWELL-NEXT: vpcmpistri $7, (%rdi), %xmm0 # sched: [16:3.00]
; BROADWELL-NEXT: # kill: %ECX<def> %ECX<kill> %RCX<def>
; BROADWELL-NEXT: leal (%rcx,%rax), %eax # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pcmpistri:
; SKYLAKE: # BB#0:
@ -699,8 +699,8 @@ define <16 x i8> @test_pcmpistrm(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) {
; BROADWELL-LABEL: test_pcmpistrm:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpcmpistrm $7, %xmm1, %xmm0 # sched: [11:3.00]
; BROADWELL-NEXT: vpcmpistrm $7, (%rdi), %xmm0 # sched: [11:3.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpcmpistrm $7, (%rdi), %xmm0 # sched: [16:3.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pcmpistrm:
; SKYLAKE: # BB#0:
@ -760,8 +760,8 @@ define <2 x i64> @test_pcmpgtq(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> *%a2) {
; BROADWELL-LABEL: test_pcmpgtq:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpcmpgtq %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
; BROADWELL-NEXT: vpcmpgtq (%rdi), %xmm0, %xmm0 # sched: [5:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpcmpgtq (%rdi), %xmm0, %xmm0 # sched: [10:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pcmpgtq:
; SKYLAKE: # BB#0:
@ -823,9 +823,9 @@ define <2 x i64> @test_pclmulqdq(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> *%a2) {
;
; BROADWELL-LABEL: test_pclmulqdq:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpclmulqdq $0, %xmm1, %xmm0, %xmm0 # sched: [11:2.00]
; BROADWELL-NEXT: vpclmulqdq $0, (%rdi), %xmm0, %xmm0 # sched: [11:2.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpclmulqdq $0, %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
; BROADWELL-NEXT: vpclmulqdq $0, (%rdi), %xmm0, %xmm0 # sched: [10:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pclmulqdq:
; SKYLAKE: # BB#0:

View File

@ -51,9 +51,9 @@ define <16 x i8> @test_pabsb(<16 x i8> %a0, <16 x i8> *%a1) {
; BROADWELL-LABEL: test_pabsb:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpabsb %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: vpabsb (%rdi), %xmm1 # sched: [1:0.50]
; BROADWELL-NEXT: vpabsb (%rdi), %xmm1 # sched: [6:0.50]
; BROADWELL-NEXT: vpor %xmm1, %xmm0, %xmm0 # sched: [1:0.33]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pabsb:
; SKYLAKE: # BB#0:
@ -130,9 +130,9 @@ define <4 x i32> @test_pabsd(<4 x i32> %a0, <4 x i32> *%a1) {
; BROADWELL-LABEL: test_pabsd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpabsd %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: vpabsd (%rdi), %xmm1 # sched: [1:0.50]
; BROADWELL-NEXT: vpabsd (%rdi), %xmm1 # sched: [6:0.50]
; BROADWELL-NEXT: vpor %xmm1, %xmm0, %xmm0 # sched: [1:0.33]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pabsd:
; SKYLAKE: # BB#0:
@ -209,9 +209,9 @@ define <8 x i16> @test_pabsw(<8 x i16> %a0, <8 x i16> *%a1) {
; BROADWELL-LABEL: test_pabsw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpabsw %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: vpabsw (%rdi), %xmm1 # sched: [1:0.50]
; BROADWELL-NEXT: vpabsw (%rdi), %xmm1 # sched: [6:0.50]
; BROADWELL-NEXT: vpor %xmm1, %xmm0, %xmm0 # sched: [1:0.33]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pabsw:
; SKYLAKE: # BB#0:
@ -287,8 +287,8 @@ define <8 x i16> @test_palignr(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) {
; BROADWELL-LABEL: test_palignr:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpalignr {{.*#+}} xmm0 = xmm0[6,7,8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5] sched: [1:1.00]
; BROADWELL-NEXT: vpalignr {{.*#+}} xmm0 = mem[14,15],xmm0[0,1,2,3,4,5,6,7,8,9,10,11,12,13] sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpalignr {{.*#+}} xmm0 = mem[14,15],xmm0[0,1,2,3,4,5,6,7,8,9,10,11,12,13] sched: [6:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_palignr:
; SKYLAKE: # BB#0:
@ -353,8 +353,8 @@ define <4 x i32> @test_phaddd(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) {
; BROADWELL-LABEL: test_phaddd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vphaddd %xmm1, %xmm0, %xmm0 # sched: [3:2.00]
; BROADWELL-NEXT: vphaddd (%rdi), %xmm0, %xmm0 # sched: [3:2.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vphaddd (%rdi), %xmm0, %xmm0 # sched: [8:2.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_phaddd:
; SKYLAKE: # BB#0:
@ -420,8 +420,8 @@ define <8 x i16> @test_phaddsw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) {
; BROADWELL-LABEL: test_phaddsw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vphaddsw %xmm1, %xmm0, %xmm0 # sched: [3:2.00]
; BROADWELL-NEXT: vphaddsw (%rdi), %xmm0, %xmm0 # sched: [3:2.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vphaddsw (%rdi), %xmm0, %xmm0 # sched: [8:2.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_phaddsw:
; SKYLAKE: # BB#0:
@ -487,8 +487,8 @@ define <8 x i16> @test_phaddw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) {
; BROADWELL-LABEL: test_phaddw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vphaddw %xmm1, %xmm0, %xmm0 # sched: [3:2.00]
; BROADWELL-NEXT: vphaddw (%rdi), %xmm0, %xmm0 # sched: [3:2.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vphaddw (%rdi), %xmm0, %xmm0 # sched: [8:2.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_phaddw:
; SKYLAKE: # BB#0:
@ -554,8 +554,8 @@ define <4 x i32> @test_phsubd(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) {
; BROADWELL-LABEL: test_phsubd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vphsubd %xmm1, %xmm0, %xmm0 # sched: [3:2.00]
; BROADWELL-NEXT: vphsubd (%rdi), %xmm0, %xmm0 # sched: [3:2.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vphsubd (%rdi), %xmm0, %xmm0 # sched: [8:2.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_phsubd:
; SKYLAKE: # BB#0:
@ -621,8 +621,8 @@ define <8 x i16> @test_phsubsw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) {
; BROADWELL-LABEL: test_phsubsw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vphsubsw %xmm1, %xmm0, %xmm0 # sched: [3:2.00]
; BROADWELL-NEXT: vphsubsw (%rdi), %xmm0, %xmm0 # sched: [3:2.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vphsubsw (%rdi), %xmm0, %xmm0 # sched: [8:2.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_phsubsw:
; SKYLAKE: # BB#0:
@ -688,8 +688,8 @@ define <8 x i16> @test_phsubw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) {
; BROADWELL-LABEL: test_phsubw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vphsubw %xmm1, %xmm0, %xmm0 # sched: [3:2.00]
; BROADWELL-NEXT: vphsubw (%rdi), %xmm0, %xmm0 # sched: [3:2.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vphsubw (%rdi), %xmm0, %xmm0 # sched: [8:2.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_phsubw:
; SKYLAKE: # BB#0:
@ -755,8 +755,8 @@ define <8 x i16> @test_pmaddubsw(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) {
; BROADWELL-LABEL: test_pmaddubsw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpmaddubsw %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
; BROADWELL-NEXT: vpmaddubsw (%rdi), %xmm0, %xmm0 # sched: [5:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpmaddubsw (%rdi), %xmm0, %xmm0 # sched: [10:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pmaddubsw:
; SKYLAKE: # BB#0:
@ -823,8 +823,8 @@ define <8 x i16> @test_pmulhrsw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) {
; BROADWELL-LABEL: test_pmulhrsw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpmulhrsw %xmm1, %xmm0, %xmm0 # sched: [5:1.00]
; BROADWELL-NEXT: vpmulhrsw (%rdi), %xmm0, %xmm0 # sched: [5:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpmulhrsw (%rdi), %xmm0, %xmm0 # sched: [10:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pmulhrsw:
; SKYLAKE: # BB#0:
@ -890,8 +890,8 @@ define <16 x i8> @test_pshufb(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) {
; BROADWELL-LABEL: test_pshufb:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpshufb %xmm1, %xmm0, %xmm0 # sched: [1:1.00]
; BROADWELL-NEXT: vpshufb (%rdi), %xmm0, %xmm0 # sched: [1:1.00]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpshufb (%rdi), %xmm0, %xmm0 # sched: [6:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_pshufb:
; SKYLAKE: # BB#0:
@ -961,8 +961,8 @@ define <16 x i8> @test_psignb(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) {
; BROADWELL-LABEL: test_psignb:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpsignb %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: vpsignb (%rdi), %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpsignb (%rdi), %xmm0, %xmm0 # sched: [6:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_psignb:
; SKYLAKE: # BB#0:
@ -1032,8 +1032,8 @@ define <4 x i32> @test_psignd(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) {
; BROADWELL-LABEL: test_psignd:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpsignd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: vpsignd (%rdi), %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpsignd (%rdi), %xmm0, %xmm0 # sched: [6:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_psignd:
; SKYLAKE: # BB#0:
@ -1103,8 +1103,8 @@ define <8 x i16> @test_psignw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) {
; BROADWELL-LABEL: test_psignw:
; BROADWELL: # BB#0:
; BROADWELL-NEXT: vpsignw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: vpsignw (%rdi), %xmm0, %xmm0 # sched: [1:0.50]
; BROADWELL-NEXT: retq # sched: [2:1.00]
; BROADWELL-NEXT: vpsignw (%rdi), %xmm0, %xmm0 # sched: [6:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
; SKYLAKE-LABEL: test_psignw:
; SKYLAKE: # BB#0: