This is an archive of the discontinued LLVM Phabricator instance.

Differential D115752

[X86] Adjust some IceLake fp shuffle schedule classes (PR48110)
ClosedPublic

Authored by RKSimon on Dec 14 2021, 12:40 PM.

Details

Reviewers
pengfei
LuoYuanke
craig.topper
andreadb
HaohaiWen
Commits
rG67cce1ceee2d: [X86] Adjust some IceLake fp shuffle schedule classes (PR48110)
Summary

The IceLake scheduler model is still mainly a copy of the SkylakeServer model.

This patch adjusts the fp shuffle classes to account for most instructions now working on Port 1 as well as Port 5.

This is based off Agner + uops.info as well as the PR48110 report.

Diff Detail

Event Timeline

RKSimon created this revision.Dec 14 2021, 12:40 PM
Herald added subscribers: gbedwell, hiraditya. · View Herald TranscriptDec 14 2021, 12:40 PM
RKSimon requested review of this revision.Dec 14 2021, 12:40 PM
Herald added a project: Restricted Project. · View Herald TranscriptDec 14 2021, 12:40 PM
Harbormaster completed remote builds in B139285: Diff 394354.Dec 14 2021, 1:45 PM
HaohaiWen added a comment.Dec 16 2021, 4:20 AM

LGTM

HaohaiWen accepted this revision.Dec 16 2021, 4:20 AM
This revision is now accepted and ready to land.Dec 16 2021, 4:20 AM
This revision was landed with ongoing or failed builds.Dec 19 2021, 5:00 AM
Closed by commit rG67cce1ceee2d: [X86] Adjust some IceLake fp shuffle schedule classes (PR48110) (authored by RKSimon). · Explain Why
This revision was automatically updated to reflect the committed changes.
RKSimon added a commit: rG67cce1ceee2d: [X86] Adjust some IceLake fp shuffle schedule classes (PR48110).

Revision Contents

PathSize
llvm/
lib/
Target/
X86/
35 lines
test/
tools/
llvm-mca/
X86/
IceLakeServer/
58 lines
50 lines
14 lines
14 lines
10 lines

Diff 395318

llvm/lib/Target/X86/X86SchedIceLake.td

Show First 20 Lines • Show All 325 Lines • ▼ Show 20 Lines
defm : ICXWriteResPair<WriteFRndY, [ICXPort01], 8, [2], 2, 7>; defm : ICXWriteResPair<WriteFRndY, [ICXPort01], 8, [2], 2, 7>;
defm : ICXWriteResPair<WriteFRndZ, [ICXPort05], 8, [2], 2, 7>; defm : ICXWriteResPair<WriteFRndZ, [ICXPort05], 8, [2], 2, 7>;
defm : ICXWriteResPair<WriteFLogic, [ICXPort015], 1, [1], 1, 6>; // Floating point and/or/xor logicals. defm : ICXWriteResPair<WriteFLogic, [ICXPort015], 1, [1], 1, 6>; // Floating point and/or/xor logicals.
defm : ICXWriteResPair<WriteFLogicY, [ICXPort015], 1, [1], 1, 7>; defm : ICXWriteResPair<WriteFLogicY, [ICXPort015], 1, [1], 1, 7>;
defm : ICXWriteResPair<WriteFLogicZ, [ICXPort05], 1, [1], 1, 7>; defm : ICXWriteResPair<WriteFLogicZ, [ICXPort05], 1, [1], 1, 7>;
defm : ICXWriteResPair<WriteFTest, [ICXPort0], 2, [1], 1, 6>; // Floating point TEST instructions. defm : ICXWriteResPair<WriteFTest, [ICXPort0], 2, [1], 1, 6>; // Floating point TEST instructions.
defm : ICXWriteResPair<WriteFTestY, [ICXPort0], 2, [1], 1, 7>; defm : ICXWriteResPair<WriteFTestY, [ICXPort0], 2, [1], 1, 7>;
defm : ICXWriteResPair<WriteFTestZ, [ICXPort0], 2, [1], 1, 7>; defm : ICXWriteResPair<WriteFTestZ, [ICXPort0], 2, [1], 1, 7>;
defm : ICXWriteResPair<WriteFShuffle, [ICXPort5], 1, [1], 1, 6>; // Floating point vector shuffles. defm : ICXWriteResPair<WriteFShuffle, [ICXPort15], 1, [1], 1, 6>; // Floating point vector shuffles.
defm : ICXWriteResPair<WriteFShuffleY, [ICXPort5], 1, [1], 1, 7>; defm : ICXWriteResPair<WriteFShuffleY, [ICXPort15], 1, [1], 1, 7>;
defm : ICXWriteResPair<WriteFShuffleZ, [ICXPort5], 1, [1], 1, 7>; defm : ICXWriteResPair<WriteFShuffleZ, [ICXPort5], 1, [1], 1, 7>;
defm : ICXWriteResPair<WriteFVarShuffle, [ICXPort5], 1, [1], 1, 6>; // Floating point vector variable shuffles. defm : ICXWriteResPair<WriteFVarShuffle, [ICXPort15], 1, [1], 1, 6>; // Floating point vector variable shuffles.
defm : ICXWriteResPair<WriteFVarShuffleY, [ICXPort5], 1, [1], 1, 7>; defm : ICXWriteResPair<WriteFVarShuffleY, [ICXPort15], 1, [1], 1, 7>;
defm : ICXWriteResPair<WriteFVarShuffleZ, [ICXPort5], 1, [1], 1, 7>; defm : ICXWriteResPair<WriteFVarShuffleZ, [ICXPort5], 1, [1], 1, 7>;
defm : ICXWriteResPair<WriteFBlend, [ICXPort015], 1, [1], 1, 6>; // Floating point vector blends. defm : ICXWriteResPair<WriteFBlend, [ICXPort015], 1, [1], 1, 6>; // Floating point vector blends.
defm : ICXWriteResPair<WriteFBlendY,[ICXPort015], 1, [1], 1, 7>; defm : ICXWriteResPair<WriteFBlendY,[ICXPort015], 1, [1], 1, 7>;
defm : ICXWriteResPair<WriteFBlendZ,[ICXPort015], 1, [1], 1, 7>; defm : ICXWriteResPair<WriteFBlendZ,[ICXPort015], 1, [1], 1, 7>;
defm : ICXWriteResPair<WriteFVarBlend, [ICXPort015], 2, [2], 2, 6>; // Fp vector variable blends. defm : ICXWriteResPair<WriteFVarBlend, [ICXPort015], 2, [2], 2, 6>; // Fp vector variable blends.
defm : ICXWriteResPair<WriteFVarBlendY,[ICXPort015], 2, [2], 2, 7>; defm : ICXWriteResPair<WriteFVarBlendY,[ICXPort015], 2, [2], 2, 7>;
defm : ICXWriteResPair<WriteFVarBlendZ,[ICXPort015], 2, [2], 2, 7>; defm : ICXWriteResPair<WriteFVarBlendZ,[ICXPort015], 2, [2], 2, 7>;
// FMA Scheduling helper class. // FMA Scheduling helper class.
▲ Show 20 Lines • Show All 312 Lines • ▼ Show 20 Linesdef ICXWriteResGroup3 : SchedWriteRes<[ICXPort5]> {
let Latency = 1; let Latency = 1;
let NumMicroOps = 1; let NumMicroOps = 1;
let ResourceCycles = [1]; let ResourceCycles = [1];
} }
def: InstRW<[ICXWriteResGroup3], (instregex "COM(P?)_FST0r", def: InstRW<[ICXWriteResGroup3], (instregex "COM(P?)_FST0r",
"KMOV(B|D|Q|W)kr", "KMOV(B|D|Q|W)kr",
"UCOM_F(P?)r", "UCOM_F(P?)r",
"VPBROADCAST(D|Q)rr", "VPBROADCAST(D|Q)rr",
"(V?)INSERTPS(Z?)rr",
"(V?)MOV(HL|LH)PS(Z?)rr",
"(V?)MOVDDUP(Y|Z|Z128|Z256)?rr",
"(V?)PALIGNR(Y|Z|Z128|Z256)?rri", "(V?)PALIGNR(Y|Z|Z128|Z256)?rri",
"(V?)PACK(U|S)S(DW|WB)(Y|Z|Z128|Z256)?rr")>; "(V?)PERMIL(PD|PS)(Y|Z|Z128|Z256)?ri",
"(V?)PERMIL(PD|PS)(Y|Z|Z128|Z256)?rr",
"(V?)PACK(U|S)S(DW|WB)(Y|Z|Z128|Z256)?rr",
"(V?)UNPCK(L|H)(PD|PS)(Y|Z|Z128|Z256)?rr")>;
def ICXWriteResGroup4 : SchedWriteRes<[ICXPort6]> { def ICXWriteResGroup4 : SchedWriteRes<[ICXPort6]> {
let Latency = 1; let Latency = 1;
let NumMicroOps = 1; let NumMicroOps = 1;
let ResourceCycles = [1]; let ResourceCycles = [1];
} }
def: InstRW<[ICXWriteResGroup4], (instregex "JMP(16|32|64)r")>; def: InstRW<[ICXWriteResGroup4], (instregex "JMP(16|32|64)r")>;
Show All 22 Linesdef ICXWriteResGroup9 : SchedWriteRes<[ICXPort015]> {
let Latency = 1; let Latency = 1;
let NumMicroOps = 1; let NumMicroOps = 1;
let ResourceCycles = [1]; let ResourceCycles = [1];
} }
def: InstRW<[ICXWriteResGroup9], (instregex "VBLENDMPD(Z128|Z256)rr", def: InstRW<[ICXWriteResGroup9], (instregex "VBLENDMPD(Z128|Z256)rr",
"VBLENDMPS(Z128|Z256)rr", "VBLENDMPS(Z128|Z256)rr",
"VPADD(B|D|Q|W)(Y|Z|Z128|Z256)rr", "VPADD(B|D|Q|W)(Y|Z|Z128|Z256)rr",
"(V?)PADD(B|D|Q|W)rr", "(V?)PADD(B|D|Q|W)rr",
"(V?)MOV(SD|SS)(Z?)rr",
"VPBLENDD(Y?)rri", "VPBLENDD(Y?)rri",
"VPBLENDMB(Z128|Z256)rr", "VPBLENDMB(Z128|Z256)rr",
"VPBLENDMD(Z128|Z256)rr", "VPBLENDMD(Z128|Z256)rr",
"VPBLENDMQ(Z128|Z256)rr", "VPBLENDMQ(Z128|Z256)rr",
"VPBLENDMW(Z128|Z256)rr", "VPBLENDMW(Z128|Z256)rr",
"VPSUB(B|D|Q|W)(Y|Z|Z128|Z256)rrk", "VPSUB(B|D|Q|W)(Y|Z|Z128|Z256)rrk",
"VPTERNLOGD(Z|Z128|Z256)rri", "VPTERNLOGD(Z|Z128|Z256)rri",
"VPTERNLOGQ(Z|Z128|Z256)rri")>; "VPTERNLOGQ(Z|Z128|Z256)rri")>;
▲ Show 20 Lines • Show All 577 Lines • ▼ Show 20 Lines
} }
def: InstRW<[ICXWriteResGroup90], (instrs VCVTDQ2PDYrr)>; def: InstRW<[ICXWriteResGroup90], (instrs VCVTDQ2PDYrr)>;
def ICXWriteResGroup92 : SchedWriteRes<[ICXPort5,ICXPort23]> { def ICXWriteResGroup92 : SchedWriteRes<[ICXPort5,ICXPort23]> {
let Latency = 7; let Latency = 7;
let NumMicroOps = 2; let NumMicroOps = 2;
let ResourceCycles = [1,1]; let ResourceCycles = [1,1];
} }
def: InstRW<[ICXWriteResGroup92], (instregex "VMOVSDZrm(b?)", def: InstRW<[ICXWriteResGroup92], (instregex "VMOV(SD|SS)Zrm(b?)",
"VMOVSSZrm(b?)",
"VPBROADCAST(B|W)(Z128)?rm", "VPBROADCAST(B|W)(Z128)?rm",
"(V?)INSERTPS(Z?)rm",
"(V?)PALIGNR(Z128)?rmi", "(V?)PALIGNR(Z128)?rmi",
"(V?)PACK(U|S)S(DW|WB)(Z128)?rm")>; "(V?)PERMIL(PD|PS)(Z128)?m(b?)i",
"(V?)PERMIL(PD|PS)(Z128)?rm",
"(V?)PACK(U|S)S(DW|WB)(Z128)?rm",
"(V?)UNPCK(L|H)(PD|PS)(Z128)?rm")>;
def ICXWriteResGroup93 : SchedWriteRes<[ICXPort5,ICXPort015]> { def ICXWriteResGroup93 : SchedWriteRes<[ICXPort5,ICXPort015]> {
let Latency = 7; let Latency = 7;
let NumMicroOps = 2; let NumMicroOps = 2;
let ResourceCycles = [1,1]; let ResourceCycles = [1,1];
} }
def: InstRW<[ICXWriteResGroup93], (instregex "VCVTDQ2PDZ256rr", def: InstRW<[ICXWriteResGroup93], (instregex "VCVTDQ2PDZ256rr",
"VCVTPD2DQ(Y|Z256)rr", "VCVTPD2DQ(Y|Z256)rr",
▲ Show 20 Lines • Show All 223 Lines • ▼ Show 20 Linesdef ICXWriteResGroup119 : SchedWriteRes<[ICXPort5,ICXPort23]> {
let Latency = 8; let Latency = 8;
let NumMicroOps = 2; let NumMicroOps = 2;
let ResourceCycles = [1,1]; let ResourceCycles = [1,1];
} }
def: InstRW<[ICXWriteResGroup119], (instregex "FCOM(P?)(32|64)m", def: InstRW<[ICXWriteResGroup119], (instregex "FCOM(P?)(32|64)m",
"VPBROADCASTB(Z|Z256)rm(b?)", "VPBROADCASTB(Z|Z256)rm(b?)",
"VPBROADCASTW(Z|Z256)rm(b?)", "VPBROADCASTW(Z|Z256)rm(b?)",
"(V?)PALIGNR(Y|Z|Z256)rmi", "(V?)PALIGNR(Y|Z|Z256)rmi",
"(V?)PACK(U|S)S(DW|WB)(Y|Z|Z256)rm")>; "(V?)PERMIL(PD|PS)(Y|Z|Z256)m(b?)i",
"(V?)PERMIL(PD|PS)(Y|Z|Z256)rm",
"(V?)PACK(U|S)S(DW|WB)(Y|Z|Z256)rm",
"(V?)UNPCK(L|H)(PD|PS)(Y|Z|Z256)rm")>;
def: InstRW<[ICXWriteResGroup119], (instrs VPBROADCASTBYrm, def: InstRW<[ICXWriteResGroup119], (instrs VPBROADCASTBYrm,
VPBROADCASTWYrm, VPBROADCASTWYrm,
VPMOVSXBDYrm, VPMOVSXBDYrm,
VPMOVSXBQYrm, VPMOVSXBQYrm,
VPMOVSXWQYrm)>; VPMOVSXWQYrm)>;
def ICXWriteResGroup121 : SchedWriteRes<[ICXPort23,ICXPort015]> { def ICXWriteResGroup121 : SchedWriteRes<[ICXPort23,ICXPort015]> {
let Latency = 8; let Latency = 8;
▲ Show 20 Lines • Show All 1,067 Lines • Show Last 20 Lines

llvm/test/tools/llvm-mca/X86/IceLakeServer/resources-avx1.s

Show First 20 Lines • Show All 1,305 Lines • ▼ Show 20 Lines
# CHECK-NEXT: 2 1 1.00 * vmovntpd %ymm0, (%rax) # CHECK-NEXT: 2 1 1.00 * vmovntpd %ymm0, (%rax)
# CHECK-NEXT: 2 1 1.00 * vmovntps %xmm0, (%rax) # CHECK-NEXT: 2 1 1.00 * vmovntps %xmm0, (%rax)
# CHECK-NEXT: 2 1 1.00 * vmovntps %ymm0, (%rax) # CHECK-NEXT: 2 1 1.00 * vmovntps %ymm0, (%rax)
# CHECK-NEXT: 1 1 0.33 vmovq %xmm0, %xmm2 # CHECK-NEXT: 1 1 0.33 vmovq %xmm0, %xmm2
# CHECK-NEXT: 1 1 1.00 vmovq %rax, %xmm2 # CHECK-NEXT: 1 1 1.00 vmovq %rax, %xmm2
# CHECK-NEXT: 1 5 0.50 * vmovq (%rax), %xmm2 # CHECK-NEXT: 1 5 0.50 * vmovq (%rax), %xmm2
# CHECK-NEXT: 1 2 1.00 vmovq %xmm0, %rcx # CHECK-NEXT: 1 2 1.00 vmovq %xmm0, %rcx
# CHECK-NEXT: 2 1 1.00 * vmovq %xmm0, (%rax) # CHECK-NEXT: 2 1 1.00 * vmovq %xmm0, (%rax)
# CHECK-NEXT: 1 1 1.00 vmovsd %xmm0, %xmm1, %xmm2 # CHECK-NEXT: 1 1 0.33 vmovsd %xmm0, %xmm1, %xmm2
# CHECK-NEXT: 2 1 1.00 * vmovsd %xmm0, (%rax) # CHECK-NEXT: 2 1 1.00 * vmovsd %xmm0, (%rax)
# CHECK-NEXT: 1 5 0.50 * vmovsd (%rax), %xmm2 # CHECK-NEXT: 1 5 0.50 * vmovsd (%rax), %xmm2
# CHECK-NEXT: 1 1 1.00 vmovshdup %xmm0, %xmm2 # CHECK-NEXT: 1 1 0.50 vmovshdup %xmm0, %xmm2
# CHECK-NEXT: 1 6 0.50 * vmovshdup (%rax), %xmm2 # CHECK-NEXT: 1 6 0.50 * vmovshdup (%rax), %xmm2
# CHECK-NEXT: 1 1 1.00 vmovshdup %ymm0, %ymm2 # CHECK-NEXT: 1 1 0.50 vmovshdup %ymm0, %ymm2
# CHECK-NEXT: 1 7 0.50 * vmovshdup (%rax), %ymm2 # CHECK-NEXT: 1 7 0.50 * vmovshdup (%rax), %ymm2
# CHECK-NEXT: 1 1 1.00 vmovsldup %xmm0, %xmm2 # CHECK-NEXT: 1 1 0.50 vmovsldup %xmm0, %xmm2
# CHECK-NEXT: 1 6 0.50 * vmovsldup (%rax), %xmm2 # CHECK-NEXT: 1 6 0.50 * vmovsldup (%rax), %xmm2
# CHECK-NEXT: 1 1 1.00 vmovsldup %ymm0, %ymm2 # CHECK-NEXT: 1 1 0.50 vmovsldup %ymm0, %ymm2
# CHECK-NEXT: 1 7 0.50 * vmovsldup (%rax), %ymm2 # CHECK-NEXT: 1 7 0.50 * vmovsldup (%rax), %ymm2
# CHECK-NEXT: 1 1 1.00 vmovss %xmm0, %xmm1, %xmm2 # CHECK-NEXT: 1 1 0.33 vmovss %xmm0, %xmm1, %xmm2
# CHECK-NEXT: 2 1 1.00 * vmovss %xmm0, (%rax) # CHECK-NEXT: 2 1 1.00 * vmovss %xmm0, (%rax)
# CHECK-NEXT: 1 5 0.50 * vmovss (%rax), %xmm2 # CHECK-NEXT: 1 5 0.50 * vmovss (%rax), %xmm2
# CHECK-NEXT: 1 1 0.33 vmovupd %xmm0, %xmm2 # CHECK-NEXT: 1 1 0.33 vmovupd %xmm0, %xmm2
# CHECK-NEXT: 2 1 1.00 * vmovupd %xmm0, (%rax) # CHECK-NEXT: 2 1 1.00 * vmovupd %xmm0, (%rax)
# CHECK-NEXT: 1 6 0.50 * vmovupd (%rax), %xmm2 # CHECK-NEXT: 1 6 0.50 * vmovupd (%rax), %xmm2
# CHECK-NEXT: 1 1 0.33 vmovupd %ymm0, %ymm2 # CHECK-NEXT: 1 1 0.33 vmovupd %ymm0, %ymm2
# CHECK-NEXT: 2 1 1.00 * vmovupd %ymm0, (%rax) # CHECK-NEXT: 2 1 1.00 * vmovupd %ymm0, (%rax)
# CHECK-NEXT: 1 7 0.50 * vmovupd (%rax), %ymm2 # CHECK-NEXT: 1 7 0.50 * vmovupd (%rax), %ymm2
▲ Show 20 Lines • Show All 311 Lines • ▼ Show 20 Lines
# CHECK-NEXT: 2 8 1.00 vroundss $1, %xmm0, %xmm1, %xmm2 # CHECK-NEXT: 2 8 1.00 vroundss $1, %xmm0, %xmm1, %xmm2
# CHECK-NEXT: 3 14 1.00 * vroundss $1, (%rax), %xmm1, %xmm2 # CHECK-NEXT: 3 14 1.00 * vroundss $1, (%rax), %xmm1, %xmm2
# CHECK-NEXT: 1 4 1.00 vrsqrtps %xmm0, %xmm2 # CHECK-NEXT: 1 4 1.00 vrsqrtps %xmm0, %xmm2
# CHECK-NEXT: 2 10 1.00 * vrsqrtps (%rax), %xmm2 # CHECK-NEXT: 2 10 1.00 * vrsqrtps (%rax), %xmm2
# CHECK-NEXT: 1 4 1.00 vrsqrtps %ymm0, %ymm2 # CHECK-NEXT: 1 4 1.00 vrsqrtps %ymm0, %ymm2
# CHECK-NEXT: 2 11 1.00 * vrsqrtps (%rax), %ymm2 # CHECK-NEXT: 2 11 1.00 * vrsqrtps (%rax), %ymm2
# CHECK-NEXT: 1 4 1.00 vrsqrtss %xmm0, %xmm1, %xmm2 # CHECK-NEXT: 1 4 1.00 vrsqrtss %xmm0, %xmm1, %xmm2
# CHECK-NEXT: 2 9 1.00 * vrsqrtss (%rax), %xmm1, %xmm2 # CHECK-NEXT: 2 9 1.00 * vrsqrtss (%rax), %xmm1, %xmm2
# CHECK-NEXT: 1 1 1.00 vshufpd $1, %xmm0, %xmm1, %xmm2 # CHECK-NEXT: 1 1 0.50 vshufpd $1, %xmm0, %xmm1, %xmm2
# CHECK-NEXT: 2 7 1.00 * vshufpd $1, (%rax), %xmm1, %xmm2 # CHECK-NEXT: 2 7 0.50 * vshufpd $1, (%rax), %xmm1, %xmm2
# CHECK-NEXT: 1 1 1.00 vshufpd $1, %ymm0, %ymm1, %ymm2 # CHECK-NEXT: 1 1 0.50 vshufpd $1, %ymm0, %ymm1, %ymm2
# CHECK-NEXT: 2 8 1.00 * vshufpd $1, (%rax), %ymm1, %ymm2 # CHECK-NEXT: 2 8 0.50 * vshufpd $1, (%rax), %ymm1, %ymm2
# CHECK-NEXT: 1 1 1.00 vshufps $1, %xmm0, %xmm1, %xmm2 # CHECK-NEXT: 1 1 0.50 vshufps $1, %xmm0, %xmm1, %xmm2
# CHECK-NEXT: 2 7 1.00 * vshufps $1, (%rax), %xmm1, %xmm2 # CHECK-NEXT: 2 7 0.50 * vshufps $1, (%rax), %xmm1, %xmm2
# CHECK-NEXT: 1 1 1.00 vshufps $1, %ymm0, %ymm1, %ymm2 # CHECK-NEXT: 1 1 0.50 vshufps $1, %ymm0, %ymm1, %ymm2
# CHECK-NEXT: 2 8 1.00 * vshufps $1, (%rax), %ymm1, %ymm2 # CHECK-NEXT: 2 8 0.50 * vshufps $1, (%rax), %ymm1, %ymm2
# CHECK-NEXT: 1 18 6.00 vsqrtpd %xmm0, %xmm2 # CHECK-NEXT: 1 18 6.00 vsqrtpd %xmm0, %xmm2
# CHECK-NEXT: 2 24 6.00 * vsqrtpd (%rax), %xmm2 # CHECK-NEXT: 2 24 6.00 * vsqrtpd (%rax), %xmm2
# CHECK-NEXT: 1 18 12.00 vsqrtpd %ymm0, %ymm2 # CHECK-NEXT: 1 18 12.00 vsqrtpd %ymm0, %ymm2
# CHECK-NEXT: 2 25 12.00 * vsqrtpd (%rax), %ymm2 # CHECK-NEXT: 2 25 12.00 * vsqrtpd (%rax), %ymm2
# CHECK-NEXT: 1 12 3.00 vsqrtps %xmm0, %xmm2 # CHECK-NEXT: 1 12 3.00 vsqrtps %xmm0, %xmm2
# CHECK-NEXT: 2 18 3.00 * vsqrtps (%rax), %xmm2 # CHECK-NEXT: 2 18 3.00 * vsqrtps (%rax), %xmm2
# CHECK-NEXT: 1 12 6.00 vsqrtps %ymm0, %ymm2 # CHECK-NEXT: 1 12 6.00 vsqrtps %ymm0, %ymm2
# CHECK-NEXT: 2 19 6.00 * vsqrtps (%rax), %ymm2 # CHECK-NEXT: 2 19 6.00 * vsqrtps (%rax), %ymm2
▲ Show 20 Lines • Show All 64 Lines • ▼ Show 20 Lines
# CHECK-NEXT: [7] - ICXPort5 # CHECK-NEXT: [7] - ICXPort5
# CHECK-NEXT: [8] - ICXPort6 # CHECK-NEXT: [8] - ICXPort6
# CHECK-NEXT: [9] - ICXPort7 # CHECK-NEXT: [9] - ICXPort7
# CHECK-NEXT: [10] - ICXPort8 # CHECK-NEXT: [10] - ICXPort8
# CHECK-NEXT: [11] - ICXPort9 # CHECK-NEXT: [11] - ICXPort9
# CHECK: Resource pressure per iteration: # CHECK: Resource pressure per iteration:
# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] # CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11]
# CHECK-NEXT: - 123.00 317.42 221.42 173.17 173.17 34.00 312.92 6.25 12.67 - - # CHECK-NEXT: - 123.00 318.08 228.08 173.17 173.17 34.00 305.58 6.25 12.67 - -
# CHECK: Resource pressure by instruction: # CHECK: Resource pressure by instruction:
# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] Instructions: # CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] Instructions:
# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - vaddpd %xmm0, %xmm1, %xmm2 # CHECK-NEXT: - - 0.50 0.50 - - - - - - - - vaddpd %xmm0, %xmm1, %xmm2
# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - vaddpd (%rax), %xmm1, %xmm2 # CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - vaddpd (%rax), %xmm1, %xmm2
# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - vaddpd %ymm0, %ymm1, %ymm2 # CHECK-NEXT: - - 0.50 0.50 - - - - - - - - vaddpd %ymm0, %ymm1, %ymm2
# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - vaddpd (%rax), %ymm1, %ymm2 # CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - vaddpd (%rax), %ymm1, %ymm2
# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - vaddps %xmm0, %xmm1, %xmm2 # CHECK-NEXT: - - 0.50 0.50 - - - - - - - - vaddps %xmm0, %xmm1, %xmm2
▲ Show 20 Lines • Show All 268 Lines • ▼ Show 20 Lines
# CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - vmovntpd %ymm0, (%rax) # CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - vmovntpd %ymm0, (%rax)
# CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - vmovntps %xmm0, (%rax) # CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - vmovntps %xmm0, (%rax)
# CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - vmovntps %ymm0, (%rax) # CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - vmovntps %ymm0, (%rax)
# CHECK-NEXT: - - 0.33 0.33 - - - 0.33 - - - - vmovq %xmm0, %xmm2 # CHECK-NEXT: - - 0.33 0.33 - - - 0.33 - - - - vmovq %xmm0, %xmm2
# CHECK-NEXT: - - - - - - - 1.00 - - - - vmovq %rax, %xmm2 # CHECK-NEXT: - - - - - - - 1.00 - - - - vmovq %rax, %xmm2
# CHECK-NEXT: - - - - 0.50 0.50 - - - - - - vmovq (%rax), %xmm2 # CHECK-NEXT: - - - - 0.50 0.50 - - - - - - vmovq (%rax), %xmm2
# CHECK-NEXT: - - 1.00 - - - - - - - - - vmovq %xmm0, %rcx # CHECK-NEXT: - - 1.00 - - - - - - - - - vmovq %xmm0, %rcx
# CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - vmovq %xmm0, (%rax) # CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - vmovq %xmm0, (%rax)
# CHECK-NEXT: - - - - - - - 1.00 - - - - vmovsd %xmm0, %xmm1, %xmm2 # CHECK-NEXT: - - 0.33 0.33 - - - 0.33 - - - - vmovsd %xmm0, %xmm1, %xmm2
# CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - vmovsd %xmm0, (%rax) # CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - vmovsd %xmm0, (%rax)
# CHECK-NEXT: - - - - 0.50 0.50 - - - - - - vmovsd (%rax), %xmm2 # CHECK-NEXT: - - - - 0.50 0.50 - - - - - - vmovsd (%rax), %xmm2
# CHECK-NEXT: - - - - - - - 1.00 - - - - vmovshdup %xmm0, %xmm2 # CHECK-NEXT: - - - 0.50 - - - 0.50 - - - - vmovshdup %xmm0, %xmm2
# CHECK-NEXT: - - - - 0.50 0.50 - - - - - - vmovshdup (%rax), %xmm2 # CHECK-NEXT: - - - - 0.50 0.50 - - - - - - vmovshdup (%rax), %xmm2
# CHECK-NEXT: - - - - - - - 1.00 - - - - vmovshdup %ymm0, %ymm2 # CHECK-NEXT: - - - 0.50 - - - 0.50 - - - - vmovshdup %ymm0, %ymm2
# CHECK-NEXT: - - - - 0.50 0.50 - - - - - - vmovshdup (%rax), %ymm2 # CHECK-NEXT: - - - - 0.50 0.50 - - - - - - vmovshdup (%rax), %ymm2
# CHECK-NEXT: - - - - - - - 1.00 - - - - vmovsldup %xmm0, %xmm2 # CHECK-NEXT: - - - 0.50 - - - 0.50 - - - - vmovsldup %xmm0, %xmm2
# CHECK-NEXT: - - - - 0.50 0.50 - - - - - - vmovsldup (%rax), %xmm2 # CHECK-NEXT: - - - - 0.50 0.50 - - - - - - vmovsldup (%rax), %xmm2
# CHECK-NEXT: - - - - - - - 1.00 - - - - vmovsldup %ymm0, %ymm2 # CHECK-NEXT: - - - 0.50 - - - 0.50 - - - - vmovsldup %ymm0, %ymm2
# CHECK-NEXT: - - - - 0.50 0.50 - - - - - - vmovsldup (%rax), %ymm2 # CHECK-NEXT: - - - - 0.50 0.50 - - - - - - vmovsldup (%rax), %ymm2
# CHECK-NEXT: - - - - - - - 1.00 - - - - vmovss %xmm0, %xmm1, %xmm2 # CHECK-NEXT: - - 0.33 0.33 - - - 0.33 - - - - vmovss %xmm0, %xmm1, %xmm2
# CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - vmovss %xmm0, (%rax) # CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - vmovss %xmm0, (%rax)
# CHECK-NEXT: - - - - 0.50 0.50 - - - - - - vmovss (%rax), %xmm2 # CHECK-NEXT: - - - - 0.50 0.50 - - - - - - vmovss (%rax), %xmm2
# CHECK-NEXT: - - 0.33 0.33 - - - 0.33 - - - - vmovupd %xmm0, %xmm2 # CHECK-NEXT: - - 0.33 0.33 - - - 0.33 - - - - vmovupd %xmm0, %xmm2
# CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - vmovupd %xmm0, (%rax) # CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - vmovupd %xmm0, (%rax)
# CHECK-NEXT: - - - - 0.50 0.50 - - - - - - vmovupd (%rax), %xmm2 # CHECK-NEXT: - - - - 0.50 0.50 - - - - - - vmovupd (%rax), %xmm2
# CHECK-NEXT: - - 0.33 0.33 - - - 0.33 - - - - vmovupd %ymm0, %ymm2 # CHECK-NEXT: - - 0.33 0.33 - - - 0.33 - - - - vmovupd %ymm0, %ymm2
# CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - vmovupd %ymm0, (%rax) # CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - vmovupd %ymm0, (%rax)
# CHECK-NEXT: - - - - 0.50 0.50 - - - - - - vmovupd (%rax), %ymm2 # CHECK-NEXT: - - - - 0.50 0.50 - - - - - - vmovupd (%rax), %ymm2
▲ Show 20 Lines • Show All 311 Lines • ▼ Show 20 Lines
# CHECK-NEXT: - - 1.00 1.00 - - - - - - - - vroundss $1, %xmm0, %xmm1, %xmm2 # CHECK-NEXT: - - 1.00 1.00 - - - - - - - - vroundss $1, %xmm0, %xmm1, %xmm2
# CHECK-NEXT: - - 1.00 1.00 0.50 0.50 - - - - - - vroundss $1, (%rax), %xmm1, %xmm2 # CHECK-NEXT: - - 1.00 1.00 0.50 0.50 - - - - - - vroundss $1, (%rax), %xmm1, %xmm2
# CHECK-NEXT: - - 1.00 - - - - - - - - - vrsqrtps %xmm0, %xmm2 # CHECK-NEXT: - - 1.00 - - - - - - - - - vrsqrtps %xmm0, %xmm2
# CHECK-NEXT: - - 1.00 - 0.50 0.50 - - - - - - vrsqrtps (%rax), %xmm2 # CHECK-NEXT: - - 1.00 - 0.50 0.50 - - - - - - vrsqrtps (%rax), %xmm2
# CHECK-NEXT: - - 1.00 - - - - - - - - - vrsqrtps %ymm0, %ymm2 # CHECK-NEXT: - - 1.00 - - - - - - - - - vrsqrtps %ymm0, %ymm2
# CHECK-NEXT: - - 1.00 - 0.50 0.50 - - - - - - vrsqrtps (%rax), %ymm2 # CHECK-NEXT: - - 1.00 - 0.50 0.50 - - - - - - vrsqrtps (%rax), %ymm2
# CHECK-NEXT: - - 1.00 - - - - - - - - - vrsqrtss %xmm0, %xmm1, %xmm2 # CHECK-NEXT: - - 1.00 - - - - - - - - - vrsqrtss %xmm0, %xmm1, %xmm2
# CHECK-NEXT: - - 1.00 - 0.50 0.50 - - - - - - vrsqrtss (%rax), %xmm1, %xmm2 # CHECK-NEXT: - - 1.00 - 0.50 0.50 - - - - - - vrsqrtss (%rax), %xmm1, %xmm2
# CHECK-NEXT: - - - - - - - 1.00 - - - - vshufpd $1, %xmm0, %xmm1, %xmm2 # CHECK-NEXT: - - - 0.50 - - - 0.50 - - - - vshufpd $1, %xmm0, %xmm1, %xmm2
# CHECK-NEXT: - - - - 0.50 0.50 - 1.00 - - - - vshufpd $1, (%rax), %xmm1, %xmm2 # CHECK-NEXT: - - - 0.50 0.50 0.50 - 0.50 - - - - vshufpd $1, (%rax), %xmm1, %xmm2
# CHECK-NEXT: - - - - - - - 1.00 - - - - vshufpd $1, %ymm0, %ymm1, %ymm2 # CHECK-NEXT: - - - 0.50 - - - 0.50 - - - - vshufpd $1, %ymm0, %ymm1, %ymm2
# CHECK-NEXT: - - - - 0.50 0.50 - 1.00 - - - - vshufpd $1, (%rax), %ymm1, %ymm2 # CHECK-NEXT: - - - 0.50 0.50 0.50 - 0.50 - - - - vshufpd $1, (%rax), %ymm1, %ymm2
# CHECK-NEXT: - - - - - - - 1.00 - - - - vshufps $1, %xmm0, %xmm1, %xmm2 # CHECK-NEXT: - - - 0.50 - - - 0.50 - - - - vshufps $1, %xmm0, %xmm1, %xmm2
# CHECK-NEXT: - - - - 0.50 0.50 - 1.00 - - - - vshufps $1, (%rax), %xmm1, %xmm2 # CHECK-NEXT: - - - 0.50 0.50 0.50 - 0.50 - - - - vshufps $1, (%rax), %xmm1, %xmm2
# CHECK-NEXT: - - - - - - - 1.00 - - - - vshufps $1, %ymm0, %ymm1, %ymm2 # CHECK-NEXT: - - - 0.50 - - - 0.50 - - - - vshufps $1, %ymm0, %ymm1, %ymm2
# CHECK-NEXT: - - - - 0.50 0.50 - 1.00 - - - - vshufps $1, (%rax), %ymm1, %ymm2 # CHECK-NEXT: - - - 0.50 0.50 0.50 - 0.50 - - - - vshufps $1, (%rax), %ymm1, %ymm2
# CHECK-NEXT: - 6.00 1.00 - - - - - - - - - vsqrtpd %xmm0, %xmm2 # CHECK-NEXT: - 6.00 1.00 - - - - - - - - - vsqrtpd %xmm0, %xmm2
# CHECK-NEXT: - 6.00 1.00 - 0.50 0.50 - - - - - - vsqrtpd (%rax), %xmm2 # CHECK-NEXT: - 6.00 1.00 - 0.50 0.50 - - - - - - vsqrtpd (%rax), %xmm2
# CHECK-NEXT: - 12.00 1.00 - - - - - - - - - vsqrtpd %ymm0, %ymm2 # CHECK-NEXT: - 12.00 1.00 - - - - - - - - - vsqrtpd %ymm0, %ymm2
# CHECK-NEXT: - 12.00 1.00 - 0.50 0.50 - - - - - - vsqrtpd (%rax), %ymm2 # CHECK-NEXT: - 12.00 1.00 - 0.50 0.50 - - - - - - vsqrtpd (%rax), %ymm2
# CHECK-NEXT: - 3.00 1.00 - - - - - - - - - vsqrtps %xmm0, %xmm2 # CHECK-NEXT: - 3.00 1.00 - - - - - - - - - vsqrtps %xmm0, %xmm2
# CHECK-NEXT: - 3.00 1.00 - 0.50 0.50 - - - - - - vsqrtps (%rax), %xmm2 # CHECK-NEXT: - 3.00 1.00 - 0.50 0.50 - - - - - - vsqrtps (%rax), %xmm2
# CHECK-NEXT: - 6.00 1.00 - - - - - - - - - vsqrtps %ymm0, %ymm2 # CHECK-NEXT: - 6.00 1.00 - - - - - - - - - vsqrtps %ymm0, %ymm2
# CHECK-NEXT: - 6.00 1.00 - 0.50 0.50 - - - - - - vsqrtps (%rax), %ymm2 # CHECK-NEXT: - 6.00 1.00 - 0.50 0.50 - - - - - - vsqrtps (%rax), %ymm2
▲ Show 20 Lines • Show All 55 Lines • Show Last 20 Lines

llvm/test/tools/llvm-mca/X86/IceLakeServer/resources-avx512vl.s

  • This file is larger than 256 KB, so syntax highlighting is disabled by default.
Show First 20 Lines • Show All 1,523 Lines • ▼ Show 20 Lines
# CHECK-NEXT: 1 1 1.00 vmovddup %xmm16, %xmm19 {%k1} {z} # CHECK-NEXT: 1 1 1.00 vmovddup %xmm16, %xmm19 {%k1} {z}
# CHECK-NEXT: 2 7 0.50 * vmovddup (%rax), %xmm19 {%k1} {z} # CHECK-NEXT: 2 7 0.50 * vmovddup (%rax), %xmm19 {%k1} {z}
# CHECK-NEXT: 1 1 1.00 vmovddup %ymm16, %ymm19 # CHECK-NEXT: 1 1 1.00 vmovddup %ymm16, %ymm19
# CHECK-NEXT: 2 8 0.50 * vmovddup (%rax), %ymm19 # CHECK-NEXT: 2 8 0.50 * vmovddup (%rax), %ymm19
# CHECK-NEXT: 1 1 1.00 vmovddup %ymm16, %ymm19 {%k1} # CHECK-NEXT: 1 1 1.00 vmovddup %ymm16, %ymm19 {%k1}
# CHECK-NEXT: 2 8 0.50 * vmovddup (%rax), %ymm19 {%k1} # CHECK-NEXT: 2 8 0.50 * vmovddup (%rax), %ymm19 {%k1}
# CHECK-NEXT: 1 1 1.00 vmovddup %ymm16, %ymm19 {%k1} {z} # CHECK-NEXT: 1 1 1.00 vmovddup %ymm16, %ymm19 {%k1} {z}
# CHECK-NEXT: 2 8 0.50 * vmovddup (%rax), %ymm19 {%k1} {z} # CHECK-NEXT: 2 8 0.50 * vmovddup (%rax), %ymm19 {%k1} {z}
# CHECK-NEXT: 1 1 1.00 vmovshdup %xmm16, %xmm19 # CHECK-NEXT: 1 1 0.50 vmovshdup %xmm16, %xmm19
# CHECK-NEXT: 2 7 0.50 * vmovshdup (%rax), %xmm19 # CHECK-NEXT: 2 7 0.50 * vmovshdup (%rax), %xmm19
# CHECK-NEXT: 1 1 1.00 vmovshdup %xmm16, %xmm19 {%k1} # CHECK-NEXT: 1 1 0.50 vmovshdup %xmm16, %xmm19 {%k1}
# CHECK-NEXT: 2 7 0.50 * vmovshdup (%rax), %xmm19 {%k1} # CHECK-NEXT: 2 7 0.50 * vmovshdup (%rax), %xmm19 {%k1}
# CHECK-NEXT: 1 1 1.00 vmovshdup %xmm16, %xmm19 {%k1} {z} # CHECK-NEXT: 1 1 0.50 vmovshdup %xmm16, %xmm19 {%k1} {z}
# CHECK-NEXT: 2 7 0.50 * vmovshdup (%rax), %xmm19 {%k1} {z} # CHECK-NEXT: 2 7 0.50 * vmovshdup (%rax), %xmm19 {%k1} {z}
# CHECK-NEXT: 1 1 1.00 vmovshdup %ymm16, %ymm19 # CHECK-NEXT: 1 1 0.50 vmovshdup %ymm16, %ymm19
# CHECK-NEXT: 2 8 0.50 * vmovshdup (%rax), %ymm19 # CHECK-NEXT: 2 8 0.50 * vmovshdup (%rax), %ymm19
# CHECK-NEXT: 1 1 1.00 vmovshdup %ymm16, %ymm19 {%k1} # CHECK-NEXT: 1 1 0.50 vmovshdup %ymm16, %ymm19 {%k1}
# CHECK-NEXT: 2 8 0.50 * vmovshdup (%rax), %ymm19 {%k1} # CHECK-NEXT: 2 8 0.50 * vmovshdup (%rax), %ymm19 {%k1}
# CHECK-NEXT: 1 1 1.00 vmovshdup %ymm16, %ymm19 {%k1} {z} # CHECK-NEXT: 1 1 0.50 vmovshdup %ymm16, %ymm19 {%k1} {z}
# CHECK-NEXT: 2 8 0.50 * vmovshdup (%rax), %ymm19 {%k1} {z} # CHECK-NEXT: 2 8 0.50 * vmovshdup (%rax), %ymm19 {%k1} {z}
# CHECK-NEXT: 1 1 1.00 vmovsldup %xmm16, %xmm19 # CHECK-NEXT: 1 1 0.50 vmovsldup %xmm16, %xmm19
# CHECK-NEXT: 2 7 0.50 * vmovsldup (%rax), %xmm19 # CHECK-NEXT: 2 7 0.50 * vmovsldup (%rax), %xmm19
# CHECK-NEXT: 1 1 1.00 vmovsldup %xmm16, %xmm19 {%k1} # CHECK-NEXT: 1 1 0.50 vmovsldup %xmm16, %xmm19 {%k1}
# CHECK-NEXT: 2 7 0.50 * vmovsldup (%rax), %xmm19 {%k1} # CHECK-NEXT: 2 7 0.50 * vmovsldup (%rax), %xmm19 {%k1}
# CHECK-NEXT: 1 1 1.00 vmovsldup %xmm16, %xmm19 {%k1} {z} # CHECK-NEXT: 1 1 0.50 vmovsldup %xmm16, %xmm19 {%k1} {z}
# CHECK-NEXT: 2 7 0.50 * vmovsldup (%rax), %xmm19 {%k1} {z} # CHECK-NEXT: 2 7 0.50 * vmovsldup (%rax), %xmm19 {%k1} {z}
# CHECK-NEXT: 1 1 1.00 vmovsldup %ymm16, %ymm19 # CHECK-NEXT: 1 1 0.50 vmovsldup %ymm16, %ymm19
# CHECK-NEXT: 2 8 0.50 * vmovsldup (%rax), %ymm19 # CHECK-NEXT: 2 8 0.50 * vmovsldup (%rax), %ymm19
# CHECK-NEXT: 1 1 1.00 vmovsldup %ymm16, %ymm19 {%k1} # CHECK-NEXT: 1 1 0.50 vmovsldup %ymm16, %ymm19 {%k1}
# CHECK-NEXT: 2 8 0.50 * vmovsldup (%rax), %ymm19 {%k1} # CHECK-NEXT: 2 8 0.50 * vmovsldup (%rax), %ymm19 {%k1}
# CHECK-NEXT: 1 1 1.00 vmovsldup %ymm16, %ymm19 {%k1} {z} # CHECK-NEXT: 1 1 0.50 vmovsldup %ymm16, %ymm19 {%k1} {z}
# CHECK-NEXT: 2 8 0.50 * vmovsldup (%rax), %ymm19 {%k1} {z} # CHECK-NEXT: 2 8 0.50 * vmovsldup (%rax), %ymm19 {%k1} {z}
# CHECK-NEXT: 1 4 0.50 vmulpd %xmm16, %xmm17, %xmm19 # CHECK-NEXT: 1 4 0.50 vmulpd %xmm16, %xmm17, %xmm19
# CHECK-NEXT: 2 10 0.50 * vmulpd (%rax), %xmm17, %xmm19 # CHECK-NEXT: 2 10 0.50 * vmulpd (%rax), %xmm17, %xmm19
# CHECK-NEXT: 2 10 0.50 * vmulpd (%rax){1to2}, %xmm17, %xmm19 # CHECK-NEXT: 2 10 0.50 * vmulpd (%rax){1to2}, %xmm17, %xmm19
# CHECK-NEXT: 1 4 0.50 vmulpd %xmm16, %xmm17, %xmm19 {%k1} # CHECK-NEXT: 1 4 0.50 vmulpd %xmm16, %xmm17, %xmm19 {%k1}
# CHECK-NEXT: 2 10 0.50 * vmulpd (%rax), %xmm17, %xmm19 {%k1} # CHECK-NEXT: 2 10 0.50 * vmulpd (%rax), %xmm17, %xmm19 {%k1}
# CHECK-NEXT: 2 10 0.50 * vmulpd (%rax){1to2}, %xmm17, %xmm19 {%k1} # CHECK-NEXT: 2 10 0.50 * vmulpd (%rax){1to2}, %xmm17, %xmm19 {%k1}
# CHECK-NEXT: 1 4 0.50 vmulpd %xmm16, %xmm17, %xmm19 {%k1} {z} # CHECK-NEXT: 1 4 0.50 vmulpd %xmm16, %xmm17, %xmm19 {%k1} {z}
▲ Show 20 Lines • Show All 784 Lines • ▼ Show 20 Lines
# CHECK-NEXT: [7] - ICXPort5 # CHECK-NEXT: [7] - ICXPort5
# CHECK-NEXT: [8] - ICXPort6 # CHECK-NEXT: [8] - ICXPort6
# CHECK-NEXT: [9] - ICXPort7 # CHECK-NEXT: [9] - ICXPort7
# CHECK-NEXT: [10] - ICXPort8 # CHECK-NEXT: [10] - ICXPort8
# CHECK-NEXT: [11] - ICXPort9 # CHECK-NEXT: [11] - ICXPort9
# CHECK: Resource pressure per iteration: # CHECK: Resource pressure per iteration:
# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] # CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11]
# CHECK-NEXT: - 423.00 257.33 226.33 372.50 372.50 - 658.33 4.00 - - - # CHECK-NEXT: - 423.00 257.33 232.33 372.50 372.50 - 652.33 4.00 - - -
# CHECK: Resource pressure by instruction: # CHECK: Resource pressure by instruction:
# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] Instructions: # CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] Instructions:
# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - vaddpd %xmm16, %xmm17, %xmm19 # CHECK-NEXT: - - 0.50 0.50 - - - - - - - - vaddpd %xmm16, %xmm17, %xmm19
# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - vaddpd (%rax), %xmm17, %xmm19 # CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - vaddpd (%rax), %xmm17, %xmm19
# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - vaddpd (%rax){1to2}, %xmm17, %xmm19 # CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - vaddpd (%rax){1to2}, %xmm17, %xmm19
# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - vaddpd %xmm16, %xmm17, %xmm19 {%k1} # CHECK-NEXT: - - 0.50 0.50 - - - - - - - - vaddpd %xmm16, %xmm17, %xmm19 {%k1}
# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - vaddpd (%rax), %xmm17, %xmm19 {%k1} # CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - vaddpd (%rax), %xmm17, %xmm19 {%k1}
▲ Show 20 Lines • Show All 271 Lines • ▼ Show 20 Lines
# CHECK-NEXT: - - - - - - - 1.00 - - - - vmovddup %xmm16, %xmm19 {%k1} {z} # CHECK-NEXT: - - - - - - - 1.00 - - - - vmovddup %xmm16, %xmm19 {%k1} {z}
# CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovddup (%rax), %xmm19 {%k1} {z} # CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovddup (%rax), %xmm19 {%k1} {z}
# CHECK-NEXT: - - - - - - - 1.00 - - - - vmovddup %ymm16, %ymm19 # CHECK-NEXT: - - - - - - - 1.00 - - - - vmovddup %ymm16, %ymm19
# CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovddup (%rax), %ymm19 # CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovddup (%rax), %ymm19
# CHECK-NEXT: - - - - - - - 1.00 - - - - vmovddup %ymm16, %ymm19 {%k1} # CHECK-NEXT: - - - - - - - 1.00 - - - - vmovddup %ymm16, %ymm19 {%k1}
# CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovddup (%rax), %ymm19 {%k1} # CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovddup (%rax), %ymm19 {%k1}
# CHECK-NEXT: - - - - - - - 1.00 - - - - vmovddup %ymm16, %ymm19 {%k1} {z} # CHECK-NEXT: - - - - - - - 1.00 - - - - vmovddup %ymm16, %ymm19 {%k1} {z}
# CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovddup (%rax), %ymm19 {%k1} {z} # CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovddup (%rax), %ymm19 {%k1} {z}
# CHECK-NEXT: - - - - - - - 1.00 - - - - vmovshdup %xmm16, %xmm19 # CHECK-NEXT: - - - 0.50 - - - 0.50 - - - - vmovshdup %xmm16, %xmm19
# CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovshdup (%rax), %xmm19 # CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovshdup (%rax), %xmm19
# CHECK-NEXT: - - - - - - - 1.00 - - - - vmovshdup %xmm16, %xmm19 {%k1} # CHECK-NEXT: - - - 0.50 - - - 0.50 - - - - vmovshdup %xmm16, %xmm19 {%k1}
# CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovshdup (%rax), %xmm19 {%k1} # CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovshdup (%rax), %xmm19 {%k1}
# CHECK-NEXT: - - - - - - - 1.00 - - - - vmovshdup %xmm16, %xmm19 {%k1} {z} # CHECK-NEXT: - - - 0.50 - - - 0.50 - - - - vmovshdup %xmm16, %xmm19 {%k1} {z}
# CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovshdup (%rax), %xmm19 {%k1} {z} # CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovshdup (%rax), %xmm19 {%k1} {z}
# CHECK-NEXT: - - - - - - - 1.00 - - - - vmovshdup %ymm16, %ymm19 # CHECK-NEXT: - - - 0.50 - - - 0.50 - - - - vmovshdup %ymm16, %ymm19
# CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovshdup (%rax), %ymm19 # CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovshdup (%rax), %ymm19
# CHECK-NEXT: - - - - - - - 1.00 - - - - vmovshdup %ymm16, %ymm19 {%k1} # CHECK-NEXT: - - - 0.50 - - - 0.50 - - - - vmovshdup %ymm16, %ymm19 {%k1}
# CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovshdup (%rax), %ymm19 {%k1} # CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovshdup (%rax), %ymm19 {%k1}
# CHECK-NEXT: - - - - - - - 1.00 - - - - vmovshdup %ymm16, %ymm19 {%k1} {z} # CHECK-NEXT: - - - 0.50 - - - 0.50 - - - - vmovshdup %ymm16, %ymm19 {%k1} {z}
# CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovshdup (%rax), %ymm19 {%k1} {z} # CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovshdup (%rax), %ymm19 {%k1} {z}
# CHECK-NEXT: - - - - - - - 1.00 - - - - vmovsldup %xmm16, %xmm19 # CHECK-NEXT: - - - 0.50 - - - 0.50 - - - - vmovsldup %xmm16, %xmm19
# CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovsldup (%rax), %xmm19 # CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovsldup (%rax), %xmm19
# CHECK-NEXT: - - - - - - - 1.00 - - - - vmovsldup %xmm16, %xmm19 {%k1} # CHECK-NEXT: - - - 0.50 - - - 0.50 - - - - vmovsldup %xmm16, %xmm19 {%k1}
# CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovsldup (%rax), %xmm19 {%k1} # CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovsldup (%rax), %xmm19 {%k1}
# CHECK-NEXT: - - - - - - - 1.00 - - - - vmovsldup %xmm16, %xmm19 {%k1} {z} # CHECK-NEXT: - - - 0.50 - - - 0.50 - - - - vmovsldup %xmm16, %xmm19 {%k1} {z}
# CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovsldup (%rax), %xmm19 {%k1} {z} # CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovsldup (%rax), %xmm19 {%k1} {z}
# CHECK-NEXT: - - - - - - - 1.00 - - - - vmovsldup %ymm16, %ymm19 # CHECK-NEXT: - - - 0.50 - - - 0.50 - - - - vmovsldup %ymm16, %ymm19
# CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovsldup (%rax), %ymm19 # CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovsldup (%rax), %ymm19
# CHECK-NEXT: - - - - - - - 1.00 - - - - vmovsldup %ymm16, %ymm19 {%k1} # CHECK-NEXT: - - - 0.50 - - - 0.50 - - - - vmovsldup %ymm16, %ymm19 {%k1}
# CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovsldup (%rax), %ymm19 {%k1} # CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovsldup (%rax), %ymm19 {%k1}
# CHECK-NEXT: - - - - - - - 1.00 - - - - vmovsldup %ymm16, %ymm19 {%k1} {z} # CHECK-NEXT: - - - 0.50 - - - 0.50 - - - - vmovsldup %ymm16, %ymm19 {%k1} {z}
# CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovsldup (%rax), %ymm19 {%k1} {z} # CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovsldup (%rax), %ymm19 {%k1} {z}
# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - vmulpd %xmm16, %xmm17, %xmm19 # CHECK-NEXT: - - 0.50 0.50 - - - - - - - - vmulpd %xmm16, %xmm17, %xmm19
# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - vmulpd (%rax), %xmm17, %xmm19 # CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - vmulpd (%rax), %xmm17, %xmm19
# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - vmulpd (%rax){1to2}, %xmm17, %xmm19 # CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - vmulpd (%rax){1to2}, %xmm17, %xmm19
# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - vmulpd %xmm16, %xmm17, %xmm19 {%k1} # CHECK-NEXT: - - 0.50 0.50 - - - - - - - - vmulpd %xmm16, %xmm17, %xmm19 {%k1}
# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - vmulpd (%rax), %xmm17, %xmm19 {%k1} # CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - vmulpd (%rax), %xmm17, %xmm19 {%k1}
# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - vmulpd (%rax){1to2}, %xmm17, %xmm19 {%k1} # CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - vmulpd (%rax){1to2}, %xmm17, %xmm19 {%k1}
# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - vmulpd %xmm16, %xmm17, %xmm19 {%k1} {z} # CHECK-NEXT: - - 0.50 0.50 - - - - - - - - vmulpd %xmm16, %xmm17, %xmm19 {%k1} {z}
▲ Show 20 Lines • Show All 775 Lines • Show Last 20 Lines

llvm/test/tools/llvm-mca/X86/IceLakeServer/resources-sse1.s

Show First 20 Lines • Show All 246 Lines • ▼ Show 20 Lines
# CHECK-NEXT: 1 1 1.00 movlhps %xmm0, %xmm2 # CHECK-NEXT: 1 1 1.00 movlhps %xmm0, %xmm2
# CHECK-NEXT: 2 1 1.00 * movhps %xmm0, (%rax) # CHECK-NEXT: 2 1 1.00 * movhps %xmm0, (%rax)
# CHECK-NEXT: 2 6 1.00 * movhps (%rax), %xmm2 # CHECK-NEXT: 2 6 1.00 * movhps (%rax), %xmm2
# CHECK-NEXT: 2 1 1.00 * movlps %xmm0, (%rax) # CHECK-NEXT: 2 1 1.00 * movlps %xmm0, (%rax)
# CHECK-NEXT: 2 6 1.00 * movlps (%rax), %xmm2 # CHECK-NEXT: 2 6 1.00 * movlps (%rax), %xmm2
# CHECK-NEXT: 1 2 1.00 movmskps %xmm0, %ecx # CHECK-NEXT: 1 2 1.00 movmskps %xmm0, %ecx
# CHECK-NEXT: 2 1 1.00 * movntps %xmm0, (%rax) # CHECK-NEXT: 2 1 1.00 * movntps %xmm0, (%rax)
# CHECK-NEXT: 2 1 1.00 * * U movntq %mm0, (%rax) # CHECK-NEXT: 2 1 1.00 * * U movntq %mm0, (%rax)
# CHECK-NEXT: 1 1 1.00 movss %xmm0, %xmm2 # CHECK-NEXT: 1 1 0.33 movss %xmm0, %xmm2
# CHECK-NEXT: 2 1 1.00 * movss %xmm0, (%rax) # CHECK-NEXT: 2 1 1.00 * movss %xmm0, (%rax)
# CHECK-NEXT: 1 5 0.50 * movss (%rax), %xmm2 # CHECK-NEXT: 1 5 0.50 * movss (%rax), %xmm2
# CHECK-NEXT: 1 1 0.33 movups %xmm0, %xmm2 # CHECK-NEXT: 1 1 0.33 movups %xmm0, %xmm2
# CHECK-NEXT: 2 1 1.00 * movups %xmm0, (%rax) # CHECK-NEXT: 2 1 1.00 * movups %xmm0, (%rax)
# CHECK-NEXT: 1 6 0.50 * movups (%rax), %xmm2 # CHECK-NEXT: 1 6 0.50 * movups (%rax), %xmm2
# CHECK-NEXT: 1 4 0.50 mulps %xmm0, %xmm2 # CHECK-NEXT: 1 4 0.50 mulps %xmm0, %xmm2
# CHECK-NEXT: 2 10 0.50 * mulps (%rax), %xmm2 # CHECK-NEXT: 2 10 0.50 * mulps (%rax), %xmm2
# CHECK-NEXT: 1 4 0.50 mulss %xmm0, %xmm2 # CHECK-NEXT: 1 4 0.50 mulss %xmm0, %xmm2
Show All 30 Lines
# CHECK-NEXT: 2 10 1.00 * rcpps (%rax), %xmm2 # CHECK-NEXT: 2 10 1.00 * rcpps (%rax), %xmm2
# CHECK-NEXT: 1 4 1.00 rcpss %xmm0, %xmm2 # CHECK-NEXT: 1 4 1.00 rcpss %xmm0, %xmm2
# CHECK-NEXT: 2 9 1.00 * rcpss (%rax), %xmm2 # CHECK-NEXT: 2 9 1.00 * rcpss (%rax), %xmm2
# CHECK-NEXT: 1 4 1.00 rsqrtps %xmm0, %xmm2 # CHECK-NEXT: 1 4 1.00 rsqrtps %xmm0, %xmm2
# CHECK-NEXT: 2 10 1.00 * rsqrtps (%rax), %xmm2 # CHECK-NEXT: 2 10 1.00 * rsqrtps (%rax), %xmm2
# CHECK-NEXT: 1 4 1.00 rsqrtss %xmm0, %xmm2 # CHECK-NEXT: 1 4 1.00 rsqrtss %xmm0, %xmm2
# CHECK-NEXT: 2 9 1.00 * rsqrtss (%rax), %xmm2 # CHECK-NEXT: 2 9 1.00 * rsqrtss (%rax), %xmm2
# CHECK-NEXT: 2 2 0.33 * * U sfence # CHECK-NEXT: 2 2 0.33 * * U sfence
# CHECK-NEXT: 1 1 1.00 shufps $1, %xmm0, %xmm2 # CHECK-NEXT: 1 1 0.50 shufps $1, %xmm0, %xmm2
# CHECK-NEXT: 2 7 1.00 * shufps $1, (%rax), %xmm2 # CHECK-NEXT: 2 7 0.50 * shufps $1, (%rax), %xmm2
# CHECK-NEXT: 1 12 3.00 sqrtps %xmm0, %xmm2 # CHECK-NEXT: 1 12 3.00 sqrtps %xmm0, %xmm2
# CHECK-NEXT: 2 18 3.00 * sqrtps (%rax), %xmm2 # CHECK-NEXT: 2 18 3.00 * sqrtps (%rax), %xmm2
# CHECK-NEXT: 1 12 3.00 sqrtss %xmm0, %xmm2 # CHECK-NEXT: 1 12 3.00 sqrtss %xmm0, %xmm2
# CHECK-NEXT: 2 17 3.00 * sqrtss (%rax), %xmm2 # CHECK-NEXT: 2 17 3.00 * sqrtss (%rax), %xmm2
# CHECK-NEXT: 3 2 1.00 * U stmxcsr (%rax) # CHECK-NEXT: 3 2 1.00 * U stmxcsr (%rax)
# CHECK-NEXT: 1 4 0.50 subps %xmm0, %xmm2 # CHECK-NEXT: 1 4 0.50 subps %xmm0, %xmm2
# CHECK-NEXT: 2 10 0.50 * subps (%rax), %xmm2 # CHECK-NEXT: 2 10 0.50 * subps (%rax), %xmm2
# CHECK-NEXT: 1 4 0.50 subss %xmm0, %xmm2 # CHECK-NEXT: 1 4 0.50 subss %xmm0, %xmm2
Show All 18 Lines
# CHECK-NEXT: [7] - ICXPort5 # CHECK-NEXT: [7] - ICXPort5
# CHECK-NEXT: [8] - ICXPort6 # CHECK-NEXT: [8] - ICXPort6
# CHECK-NEXT: [9] - ICXPort7 # CHECK-NEXT: [9] - ICXPort7
# CHECK-NEXT: [10] - ICXPort8 # CHECK-NEXT: [10] - ICXPort8
# CHECK-NEXT: [11] - ICXPort9 # CHECK-NEXT: [11] - ICXPort9
# CHECK: Resource pressure per iteration: # CHECK: Resource pressure per iteration:
# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] # CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11]
# CHECK-NEXT: - 26.00 65.83 25.83 32.00 32.00 8.00 37.83 0.50 3.00 - - # CHECK-NEXT: - 26.00 66.17 27.17 32.00 32.00 8.00 36.17 0.50 3.00 - -
# CHECK: Resource pressure by instruction: # CHECK: Resource pressure by instruction:
# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] Instructions: # CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] Instructions:
# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - addps %xmm0, %xmm2 # CHECK-NEXT: - - 0.50 0.50 - - - - - - - - addps %xmm0, %xmm2
# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - addps (%rax), %xmm2 # CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - addps (%rax), %xmm2
# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - addss %xmm0, %xmm2 # CHECK-NEXT: - - 0.50 0.50 - - - - - - - - addss %xmm0, %xmm2
# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - addss (%rax), %xmm2 # CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - addss (%rax), %xmm2
# CHECK-NEXT: - - 0.33 0.33 - - - 0.33 - - - - andnps %xmm0, %xmm2 # CHECK-NEXT: - - 0.33 0.33 - - - 0.33 - - - - andnps %xmm0, %xmm2
▲ Show 20 Lines • Show All 45 Lines • ▼ Show 20 Lines
# CHECK-NEXT: - - - - - - - 1.00 - - - - movlhps %xmm0, %xmm2 # CHECK-NEXT: - - - - - - - 1.00 - - - - movlhps %xmm0, %xmm2
# CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - movhps %xmm0, (%rax) # CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - movhps %xmm0, (%rax)
# CHECK-NEXT: - - - - 0.50 0.50 - 1.00 - - - - movhps (%rax), %xmm2 # CHECK-NEXT: - - - - 0.50 0.50 - 1.00 - - - - movhps (%rax), %xmm2
# CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - movlps %xmm0, (%rax) # CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - movlps %xmm0, (%rax)
# CHECK-NEXT: - - - - 0.50 0.50 - 1.00 - - - - movlps (%rax), %xmm2 # CHECK-NEXT: - - - - 0.50 0.50 - 1.00 - - - - movlps (%rax), %xmm2
# CHECK-NEXT: - - 1.00 - - - - - - - - - movmskps %xmm0, %ecx # CHECK-NEXT: - - 1.00 - - - - - - - - - movmskps %xmm0, %ecx
# CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - movntps %xmm0, (%rax) # CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - movntps %xmm0, (%rax)
# CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - movntq %mm0, (%rax) # CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - movntq %mm0, (%rax)
# CHECK-NEXT: - - - - - - - 1.00 - - - - movss %xmm0, %xmm2 # CHECK-NEXT: - - 0.33 0.33 - - - 0.33 - - - - movss %xmm0, %xmm2
# CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - movss %xmm0, (%rax) # CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - movss %xmm0, (%rax)
# CHECK-NEXT: - - - - 0.50 0.50 - - - - - - movss (%rax), %xmm2 # CHECK-NEXT: - - - - 0.50 0.50 - - - - - - movss (%rax), %xmm2
# CHECK-NEXT: - - 0.33 0.33 - - - 0.33 - - - - movups %xmm0, %xmm2 # CHECK-NEXT: - - 0.33 0.33 - - - 0.33 - - - - movups %xmm0, %xmm2
# CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - movups %xmm0, (%rax) # CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - movups %xmm0, (%rax)
# CHECK-NEXT: - - - - 0.50 0.50 - - - - - - movups (%rax), %xmm2 # CHECK-NEXT: - - - - 0.50 0.50 - - - - - - movups (%rax), %xmm2
# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - mulps %xmm0, %xmm2 # CHECK-NEXT: - - 0.50 0.50 - - - - - - - - mulps %xmm0, %xmm2
# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - mulps (%rax), %xmm2 # CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - mulps (%rax), %xmm2
# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - mulss %xmm0, %xmm2 # CHECK-NEXT: - - 0.50 0.50 - - - - - - - - mulss %xmm0, %xmm2
Show All 30 Lines
# CHECK-NEXT: - - 1.00 - 0.50 0.50 - - - - - - rcpps (%rax), %xmm2 # CHECK-NEXT: - - 1.00 - 0.50 0.50 - - - - - - rcpps (%rax), %xmm2
# CHECK-NEXT: - - 1.00 - - - - - - - - - rcpss %xmm0, %xmm2 # CHECK-NEXT: - - 1.00 - - - - - - - - - rcpss %xmm0, %xmm2
# CHECK-NEXT: - - 1.00 - 0.50 0.50 - - - - - - rcpss (%rax), %xmm2 # CHECK-NEXT: - - 1.00 - 0.50 0.50 - - - - - - rcpss (%rax), %xmm2
# CHECK-NEXT: - - 1.00 - - - - - - - - - rsqrtps %xmm0, %xmm2 # CHECK-NEXT: - - 1.00 - - - - - - - - - rsqrtps %xmm0, %xmm2
# CHECK-NEXT: - - 1.00 - 0.50 0.50 - - - - - - rsqrtps (%rax), %xmm2 # CHECK-NEXT: - - 1.00 - 0.50 0.50 - - - - - - rsqrtps (%rax), %xmm2
# CHECK-NEXT: - - 1.00 - - - - - - - - - rsqrtss %xmm0, %xmm2 # CHECK-NEXT: - - 1.00 - - - - - - - - - rsqrtss %xmm0, %xmm2
# CHECK-NEXT: - - 1.00 - 0.50 0.50 - - - - - - rsqrtss (%rax), %xmm2 # CHECK-NEXT: - - 1.00 - 0.50 0.50 - - - - - - rsqrtss (%rax), %xmm2
# CHECK-NEXT: - - 0.25 0.25 0.33 0.33 - 0.25 0.25 0.33 - - sfence # CHECK-NEXT: - - 0.25 0.25 0.33 0.33 - 0.25 0.25 0.33 - - sfence
# CHECK-NEXT: - - - - - - - 1.00 - - - - shufps $1, %xmm0, %xmm2 # CHECK-NEXT: - - - 0.50 - - - 0.50 - - - - shufps $1, %xmm0, %xmm2
# CHECK-NEXT: - - - - 0.50 0.50 - 1.00 - - - - shufps $1, (%rax), %xmm2 # CHECK-NEXT: - - - 0.50 0.50 0.50 - 0.50 - - - - shufps $1, (%rax), %xmm2
# CHECK-NEXT: - 3.00 1.00 - - - - - - - - - sqrtps %xmm0, %xmm2 # CHECK-NEXT: - 3.00 1.00 - - - - - - - - - sqrtps %xmm0, %xmm2
# CHECK-NEXT: - 3.00 1.00 - 0.50 0.50 - - - - - - sqrtps (%rax), %xmm2 # CHECK-NEXT: - 3.00 1.00 - 0.50 0.50 - - - - - - sqrtps (%rax), %xmm2
# CHECK-NEXT: - 3.00 1.00 - - - - - - - - - sqrtss %xmm0, %xmm2 # CHECK-NEXT: - 3.00 1.00 - - - - - - - - - sqrtss %xmm0, %xmm2
# CHECK-NEXT: - 3.00 1.00 - 0.50 0.50 - - - - - - sqrtss (%rax), %xmm2 # CHECK-NEXT: - 3.00 1.00 - 0.50 0.50 - - - - - - sqrtss (%rax), %xmm2
# CHECK-NEXT: - - - - 0.33 0.33 1.00 1.00 - 0.33 - - stmxcsr (%rax) # CHECK-NEXT: - - - - 0.33 0.33 1.00 1.00 - 0.33 - - stmxcsr (%rax)
# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - subps %xmm0, %xmm2 # CHECK-NEXT: - - 0.50 0.50 - - - - - - - - subps %xmm0, %xmm2
# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - subps (%rax), %xmm2 # CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - subps (%rax), %xmm2
# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - subss %xmm0, %xmm2 # CHECK-NEXT: - - 0.50 0.50 - - - - - - - - subss %xmm0, %xmm2
Show All 9 Lines

llvm/test/tools/llvm-mca/X86/IceLakeServer/resources-sse2.s

Show First 20 Lines • Show All 498 Lines • ▼ Show 20 Lines
# CHECK-NEXT: 2 1 1.00 * movntdq %xmm0, (%rax) # CHECK-NEXT: 2 1 1.00 * movntdq %xmm0, (%rax)
# CHECK-NEXT: 2 1 1.00 * movntpd %xmm0, (%rax) # CHECK-NEXT: 2 1 1.00 * movntpd %xmm0, (%rax)
# CHECK-NEXT: 1 1 0.33 movq %xmm0, %xmm2 # CHECK-NEXT: 1 1 0.33 movq %xmm0, %xmm2
# CHECK-NEXT: 1 1 1.00 movq %rax, %xmm2 # CHECK-NEXT: 1 1 1.00 movq %rax, %xmm2
# CHECK-NEXT: 1 5 0.50 * movq (%rax), %xmm2 # CHECK-NEXT: 1 5 0.50 * movq (%rax), %xmm2
# CHECK-NEXT: 1 2 1.00 movq %xmm0, %rcx # CHECK-NEXT: 1 2 1.00 movq %xmm0, %rcx
# CHECK-NEXT: 2 1 1.00 * movq %xmm0, (%rax) # CHECK-NEXT: 2 1 1.00 * movq %xmm0, (%rax)
# CHECK-NEXT: 2 2 2.00 movq2dq %mm0, %xmm2 # CHECK-NEXT: 2 2 2.00 movq2dq %mm0, %xmm2
# CHECK-NEXT: 1 1 1.00 movsd %xmm0, %xmm2 # CHECK-NEXT: 1 1 0.33 movsd %xmm0, %xmm2
# CHECK-NEXT: 2 1 1.00 * movsd %xmm0, (%rax) # CHECK-NEXT: 2 1 1.00 * movsd %xmm0, (%rax)
# CHECK-NEXT: 1 5 0.50 * movsd (%rax), %xmm2 # CHECK-NEXT: 1 5 0.50 * movsd (%rax), %xmm2
# CHECK-NEXT: 1 1 0.33 movupd %xmm0, %xmm2 # CHECK-NEXT: 1 1 0.33 movupd %xmm0, %xmm2
# CHECK-NEXT: 2 1 1.00 * movupd %xmm0, (%rax) # CHECK-NEXT: 2 1 1.00 * movupd %xmm0, (%rax)
# CHECK-NEXT: 1 6 0.50 * movupd (%rax), %xmm2 # CHECK-NEXT: 1 6 0.50 * movupd (%rax), %xmm2
# CHECK-NEXT: 1 4 0.50 mulpd %xmm0, %xmm2 # CHECK-NEXT: 1 4 0.50 mulpd %xmm0, %xmm2
# CHECK-NEXT: 2 10 0.50 * mulpd (%rax), %xmm2 # CHECK-NEXT: 2 10 0.50 * mulpd (%rax), %xmm2
# CHECK-NEXT: 1 4 0.50 mulsd %xmm0, %xmm2 # CHECK-NEXT: 1 4 0.50 mulsd %xmm0, %xmm2
▲ Show 20 Lines • Show All 135 Lines • ▼ Show 20 Lines
# CHECK-NEXT: 1 1 0.50 punpckldq %xmm0, %xmm2 # CHECK-NEXT: 1 1 0.50 punpckldq %xmm0, %xmm2
# CHECK-NEXT: 2 7 0.50 * punpckldq (%rax), %xmm2 # CHECK-NEXT: 2 7 0.50 * punpckldq (%rax), %xmm2
# CHECK-NEXT: 1 1 0.50 punpcklqdq %xmm0, %xmm2 # CHECK-NEXT: 1 1 0.50 punpcklqdq %xmm0, %xmm2
# CHECK-NEXT: 2 7 0.50 * punpcklqdq (%rax), %xmm2 # CHECK-NEXT: 2 7 0.50 * punpcklqdq (%rax), %xmm2
# CHECK-NEXT: 1 1 0.50 punpcklwd %xmm0, %xmm2 # CHECK-NEXT: 1 1 0.50 punpcklwd %xmm0, %xmm2
# CHECK-NEXT: 2 7 0.50 * punpcklwd (%rax), %xmm2 # CHECK-NEXT: 2 7 0.50 * punpcklwd (%rax), %xmm2
# CHECK-NEXT: 1 1 0.33 pxor %xmm0, %xmm2 # CHECK-NEXT: 1 1 0.33 pxor %xmm0, %xmm2
# CHECK-NEXT: 2 7 0.50 * pxor (%rax), %xmm2 # CHECK-NEXT: 2 7 0.50 * pxor (%rax), %xmm2
# CHECK-NEXT: 1 1 1.00 shufpd $1, %xmm0, %xmm2 # CHECK-NEXT: 1 1 0.50 shufpd $1, %xmm0, %xmm2
# CHECK-NEXT: 2 7 1.00 * shufpd $1, (%rax), %xmm2 # CHECK-NEXT: 2 7 0.50 * shufpd $1, (%rax), %xmm2
# CHECK-NEXT: 1 18 6.00 sqrtpd %xmm0, %xmm2 # CHECK-NEXT: 1 18 6.00 sqrtpd %xmm0, %xmm2
# CHECK-NEXT: 2 24 6.00 * sqrtpd (%rax), %xmm2 # CHECK-NEXT: 2 24 6.00 * sqrtpd (%rax), %xmm2
# CHECK-NEXT: 1 18 6.00 sqrtsd %xmm0, %xmm2 # CHECK-NEXT: 1 18 6.00 sqrtsd %xmm0, %xmm2
# CHECK-NEXT: 2 23 6.00 * sqrtsd (%rax), %xmm2 # CHECK-NEXT: 2 23 6.00 * sqrtsd (%rax), %xmm2
# CHECK-NEXT: 1 4 0.50 subpd %xmm0, %xmm2 # CHECK-NEXT: 1 4 0.50 subpd %xmm0, %xmm2
# CHECK-NEXT: 2 10 0.50 * subpd (%rax), %xmm2 # CHECK-NEXT: 2 10 0.50 * subpd (%rax), %xmm2
# CHECK-NEXT: 1 4 0.50 subsd %xmm0, %xmm2 # CHECK-NEXT: 1 4 0.50 subsd %xmm0, %xmm2
# CHECK-NEXT: 2 9 0.50 * subsd (%rax), %xmm2 # CHECK-NEXT: 2 9 0.50 * subsd (%rax), %xmm2
Show All 17 Lines
# CHECK-NEXT: [7] - ICXPort5 # CHECK-NEXT: [7] - ICXPort5
# CHECK-NEXT: [8] - ICXPort6 # CHECK-NEXT: [8] - ICXPort6
# CHECK-NEXT: [9] - ICXPort7 # CHECK-NEXT: [9] - ICXPort7
# CHECK-NEXT: [10] - ICXPort8 # CHECK-NEXT: [10] - ICXPort8
# CHECK-NEXT: [11] - ICXPort9 # CHECK-NEXT: [11] - ICXPort9
# CHECK: Resource pressure per iteration: # CHECK: Resource pressure per iteration:
# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] # CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11]
# CHECK-NEXT: - 38.00 103.58 94.58 63.50 63.50 14.00 85.58 2.25 5.00 - - # CHECK-NEXT: - 38.00 103.92 95.92 63.50 63.50 14.00 83.92 2.25 5.00 - -
# CHECK: Resource pressure by instruction: # CHECK: Resource pressure by instruction:
# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] Instructions: # CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] Instructions:
# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - addpd %xmm0, %xmm2 # CHECK-NEXT: - - 0.50 0.50 - - - - - - - - addpd %xmm0, %xmm2
# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - addpd (%rax), %xmm2 # CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - addpd (%rax), %xmm2
# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - addsd %xmm0, %xmm2 # CHECK-NEXT: - - 0.50 0.50 - - - - - - - - addsd %xmm0, %xmm2
# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - addsd (%rax), %xmm2 # CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - addsd (%rax), %xmm2
# CHECK-NEXT: - - 0.33 0.33 - - - 0.33 - - - - andnpd %xmm0, %xmm2 # CHECK-NEXT: - - 0.33 0.33 - - - 0.33 - - - - andnpd %xmm0, %xmm2
▲ Show 20 Lines • Show All 84 Lines • ▼ Show 20 Lines
# CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - movntdq %xmm0, (%rax) # CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - movntdq %xmm0, (%rax)
# CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - movntpd %xmm0, (%rax) # CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - movntpd %xmm0, (%rax)
# CHECK-NEXT: - - 0.33 0.33 - - - 0.33 - - - - movq %xmm0, %xmm2 # CHECK-NEXT: - - 0.33 0.33 - - - 0.33 - - - - movq %xmm0, %xmm2
# CHECK-NEXT: - - - - - - - 1.00 - - - - movq %rax, %xmm2 # CHECK-NEXT: - - - - - - - 1.00 - - - - movq %rax, %xmm2
# CHECK-NEXT: - - - - 0.50 0.50 - - - - - - movq (%rax), %xmm2 # CHECK-NEXT: - - - - 0.50 0.50 - - - - - - movq (%rax), %xmm2
# CHECK-NEXT: - - 1.00 - - - - - - - - - movq %xmm0, %rcx # CHECK-NEXT: - - 1.00 - - - - - - - - - movq %xmm0, %rcx
# CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - movq %xmm0, (%rax) # CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - movq %xmm0, (%rax)
# CHECK-NEXT: - - - - - - - 2.00 - - - - movq2dq %mm0, %xmm2 # CHECK-NEXT: - - - - - - - 2.00 - - - - movq2dq %mm0, %xmm2
# CHECK-NEXT: - - - - - - - 1.00 - - - - movsd %xmm0, %xmm2 # CHECK-NEXT: - - 0.33 0.33 - - - 0.33 - - - - movsd %xmm0, %xmm2
# CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - movsd %xmm0, (%rax) # CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - movsd %xmm0, (%rax)
# CHECK-NEXT: - - - - 0.50 0.50 - - - - - - movsd (%rax), %xmm2 # CHECK-NEXT: - - - - 0.50 0.50 - - - - - - movsd (%rax), %xmm2
# CHECK-NEXT: - - 0.33 0.33 - - - 0.33 - - - - movupd %xmm0, %xmm2 # CHECK-NEXT: - - 0.33 0.33 - - - 0.33 - - - - movupd %xmm0, %xmm2
# CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - movupd %xmm0, (%rax) # CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 - - movupd %xmm0, (%rax)
# CHECK-NEXT: - - - - 0.50 0.50 - - - - - - movupd (%rax), %xmm2 # CHECK-NEXT: - - - - 0.50 0.50 - - - - - - movupd (%rax), %xmm2
# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - mulpd %xmm0, %xmm2 # CHECK-NEXT: - - 0.50 0.50 - - - - - - - - mulpd %xmm0, %xmm2
# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - mulpd (%rax), %xmm2 # CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - mulpd (%rax), %xmm2
# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - mulsd %xmm0, %xmm2 # CHECK-NEXT: - - 0.50 0.50 - - - - - - - - mulsd %xmm0, %xmm2
▲ Show 20 Lines • Show All 135 Lines • ▼ Show 20 Lines
# CHECK-NEXT: - - - 0.50 - - - 0.50 - - - - punpckldq %xmm0, %xmm2 # CHECK-NEXT: - - - 0.50 - - - 0.50 - - - - punpckldq %xmm0, %xmm2
# CHECK-NEXT: - - - 0.50 0.50 0.50 - 0.50 - - - - punpckldq (%rax), %xmm2 # CHECK-NEXT: - - - 0.50 0.50 0.50 - 0.50 - - - - punpckldq (%rax), %xmm2
# CHECK-NEXT: - - - 0.50 - - - 0.50 - - - - punpcklqdq %xmm0, %xmm2 # CHECK-NEXT: - - - 0.50 - - - 0.50 - - - - punpcklqdq %xmm0, %xmm2
# CHECK-NEXT: - - - 0.50 0.50 0.50 - 0.50 - - - - punpcklqdq (%rax), %xmm2 # CHECK-NEXT: - - - 0.50 0.50 0.50 - 0.50 - - - - punpcklqdq (%rax), %xmm2
# CHECK-NEXT: - - - 0.50 - - - 0.50 - - - - punpcklwd %xmm0, %xmm2 # CHECK-NEXT: - - - 0.50 - - - 0.50 - - - - punpcklwd %xmm0, %xmm2
# CHECK-NEXT: - - - 0.50 0.50 0.50 - 0.50 - - - - punpcklwd (%rax), %xmm2 # CHECK-NEXT: - - - 0.50 0.50 0.50 - 0.50 - - - - punpcklwd (%rax), %xmm2
# CHECK-NEXT: - - 0.33 0.33 - - - 0.33 - - - - pxor %xmm0, %xmm2 # CHECK-NEXT: - - 0.33 0.33 - - - 0.33 - - - - pxor %xmm0, %xmm2
# CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - pxor (%rax), %xmm2 # CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - pxor (%rax), %xmm2
# CHECK-NEXT: - - - - - - - 1.00 - - - - shufpd $1, %xmm0, %xmm2 # CHECK-NEXT: - - - 0.50 - - - 0.50 - - - - shufpd $1, %xmm0, %xmm2
# CHECK-NEXT: - - - - 0.50 0.50 - 1.00 - - - - shufpd $1, (%rax), %xmm2 # CHECK-NEXT: - - - 0.50 0.50 0.50 - 0.50 - - - - shufpd $1, (%rax), %xmm2
# CHECK-NEXT: - 6.00 1.00 - - - - - - - - - sqrtpd %xmm0, %xmm2 # CHECK-NEXT: - 6.00 1.00 - - - - - - - - - sqrtpd %xmm0, %xmm2
# CHECK-NEXT: - 6.00 1.00 - 0.50 0.50 - - - - - - sqrtpd (%rax), %xmm2 # CHECK-NEXT: - 6.00 1.00 - 0.50 0.50 - - - - - - sqrtpd (%rax), %xmm2
# CHECK-NEXT: - 6.00 1.00 - - - - - - - - - sqrtsd %xmm0, %xmm2 # CHECK-NEXT: - 6.00 1.00 - - - - - - - - - sqrtsd %xmm0, %xmm2
# CHECK-NEXT: - 6.00 1.00 - 0.50 0.50 - - - - - - sqrtsd (%rax), %xmm2 # CHECK-NEXT: - 6.00 1.00 - 0.50 0.50 - - - - - - sqrtsd (%rax), %xmm2
# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - subpd %xmm0, %xmm2 # CHECK-NEXT: - - 0.50 0.50 - - - - - - - - subpd %xmm0, %xmm2
# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - subpd (%rax), %xmm2 # CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - subpd (%rax), %xmm2
# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - subsd %xmm0, %xmm2 # CHECK-NEXT: - - 0.50 0.50 - - - - - - - - subsd %xmm0, %xmm2
# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - subsd (%rax), %xmm2 # CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - subsd (%rax), %xmm2
# CHECK-NEXT: - - 1.00 - - - - - - - - - ucomisd %xmm0, %xmm1 # CHECK-NEXT: - - 1.00 - - - - - - - - - ucomisd %xmm0, %xmm1
# CHECK-NEXT: - - 1.00 - 0.50 0.50 - - - - - - ucomisd (%rax), %xmm1 # CHECK-NEXT: - - 1.00 - 0.50 0.50 - - - - - - ucomisd (%rax), %xmm1
# CHECK-NEXT: - - - - - - - 1.00 - - - - unpckhpd %xmm0, %xmm2 # CHECK-NEXT: - - - - - - - 1.00 - - - - unpckhpd %xmm0, %xmm2
# CHECK-NEXT: - - - - 0.50 0.50 - 1.00 - - - - unpckhpd (%rax), %xmm2 # CHECK-NEXT: - - - - 0.50 0.50 - 1.00 - - - - unpckhpd (%rax), %xmm2
# CHECK-NEXT: - - - - - - - 1.00 - - - - unpcklpd %xmm0, %xmm2 # CHECK-NEXT: - - - - - - - 1.00 - - - - unpcklpd %xmm0, %xmm2
# CHECK-NEXT: - - - - 0.50 0.50 - 1.00 - - - - unpcklpd (%rax), %xmm2 # CHECK-NEXT: - - - - 0.50 0.50 - 1.00 - - - - unpcklpd (%rax), %xmm2
# CHECK-NEXT: - - 0.33 0.33 - - - 0.33 - - - - xorpd %xmm0, %xmm2 # CHECK-NEXT: - - 0.33 0.33 - - - 0.33 - - - - xorpd %xmm0, %xmm2
# CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - xorpd (%rax), %xmm2 # CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - xorpd (%rax), %xmm2

llvm/test/tools/llvm-mca/X86/IceLakeServer/resources-sse3.s

Show First 20 Lines • Show All 53 Lines • ▼ Show 20 Lines
# CHECK-NEXT: 3 6 2.00 hsubpd %xmm0, %xmm2 # CHECK-NEXT: 3 6 2.00 hsubpd %xmm0, %xmm2
# CHECK-NEXT: 4 12 2.00 * hsubpd (%rax), %xmm2 # CHECK-NEXT: 4 12 2.00 * hsubpd (%rax), %xmm2
# CHECK-NEXT: 3 6 2.00 hsubps %xmm0, %xmm2 # CHECK-NEXT: 3 6 2.00 hsubps %xmm0, %xmm2
# CHECK-NEXT: 4 12 2.00 * hsubps (%rax), %xmm2 # CHECK-NEXT: 4 12 2.00 * hsubps (%rax), %xmm2
# CHECK-NEXT: 1 6 0.50 * lddqu (%rax), %xmm2 # CHECK-NEXT: 1 6 0.50 * lddqu (%rax), %xmm2
# CHECK-NEXT: 1 100 0.25 U monitor # CHECK-NEXT: 1 100 0.25 U monitor
# CHECK-NEXT: 1 1 1.00 movddup %xmm0, %xmm2 # CHECK-NEXT: 1 1 1.00 movddup %xmm0, %xmm2
# CHECK-NEXT: 1 5 0.50 * movddup (%rax), %xmm2 # CHECK-NEXT: 1 5 0.50 * movddup (%rax), %xmm2
# CHECK-NEXT: 1 1 1.00 movshdup %xmm0, %xmm2 # CHECK-NEXT: 1 1 0.50 movshdup %xmm0, %xmm2
# CHECK-NEXT: 1 6 0.50 * movshdup (%rax), %xmm2 # CHECK-NEXT: 1 6 0.50 * movshdup (%rax), %xmm2
# CHECK-NEXT: 1 1 1.00 movsldup %xmm0, %xmm2 # CHECK-NEXT: 1 1 0.50 movsldup %xmm0, %xmm2
# CHECK-NEXT: 1 6 0.50 * movsldup (%rax), %xmm2 # CHECK-NEXT: 1 6 0.50 * movsldup (%rax), %xmm2
# CHECK-NEXT: 10 20 2.50 * * U mwait # CHECK-NEXT: 10 20 2.50 * * U mwait
# CHECK: Resources: # CHECK: Resources:
# CHECK-NEXT: [0] - ICXDivider # CHECK-NEXT: [0] - ICXDivider
# CHECK-NEXT: [1] - ICXFPDivider # CHECK-NEXT: [1] - ICXFPDivider
# CHECK-NEXT: [2] - ICXPort0 # CHECK-NEXT: [2] - ICXPort0
# CHECK-NEXT: [3] - ICXPort1 # CHECK-NEXT: [3] - ICXPort1
# CHECK-NEXT: [4] - ICXPort2 # CHECK-NEXT: [4] - ICXPort2
# CHECK-NEXT: [5] - ICXPort3 # CHECK-NEXT: [5] - ICXPort3
# CHECK-NEXT: [6] - ICXPort4 # CHECK-NEXT: [6] - ICXPort4
# CHECK-NEXT: [7] - ICXPort5 # CHECK-NEXT: [7] - ICXPort5
# CHECK-NEXT: [8] - ICXPort6 # CHECK-NEXT: [8] - ICXPort6
# CHECK-NEXT: [9] - ICXPort7 # CHECK-NEXT: [9] - ICXPort7
# CHECK-NEXT: [10] - ICXPort8 # CHECK-NEXT: [10] - ICXPort8
# CHECK-NEXT: [11] - ICXPort9 # CHECK-NEXT: [11] - ICXPort9
# CHECK: Resource pressure per iteration: # CHECK: Resource pressure per iteration:
# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] # CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11]
# CHECK-NEXT: - - 6.67 6.67 5.00 5.00 - 24.67 4.00 - - - # CHECK-NEXT: - - 6.67 7.67 5.00 5.00 - 23.67 4.00 - - -
# CHECK: Resource pressure by instruction: # CHECK: Resource pressure by instruction:
# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] Instructions: # CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] Instructions:
# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - addsubpd %xmm0, %xmm2 # CHECK-NEXT: - - 0.50 0.50 - - - - - - - - addsubpd %xmm0, %xmm2
# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - addsubpd (%rax), %xmm2 # CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - addsubpd (%rax), %xmm2
# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - addsubps %xmm0, %xmm2 # CHECK-NEXT: - - 0.50 0.50 - - - - - - - - addsubps %xmm0, %xmm2
# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - addsubps (%rax), %xmm2 # CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - addsubps (%rax), %xmm2
# CHECK-NEXT: - - 0.33 0.33 - - - 2.33 - - - - haddpd %xmm0, %xmm2 # CHECK-NEXT: - - 0.33 0.33 - - - 2.33 - - - - haddpd %xmm0, %xmm2
# CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 2.33 - - - - haddpd (%rax), %xmm2 # CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 2.33 - - - - haddpd (%rax), %xmm2
# CHECK-NEXT: - - 0.33 0.33 - - - 2.33 - - - - haddps %xmm0, %xmm2 # CHECK-NEXT: - - 0.33 0.33 - - - 2.33 - - - - haddps %xmm0, %xmm2
# CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 2.33 - - - - haddps (%rax), %xmm2 # CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 2.33 - - - - haddps (%rax), %xmm2
# CHECK-NEXT: - - 0.33 0.33 - - - 2.33 - - - - hsubpd %xmm0, %xmm2 # CHECK-NEXT: - - 0.33 0.33 - - - 2.33 - - - - hsubpd %xmm0, %xmm2
# CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 2.33 - - - - hsubpd (%rax), %xmm2 # CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 2.33 - - - - hsubpd (%rax), %xmm2
# CHECK-NEXT: - - 0.33 0.33 - - - 2.33 - - - - hsubps %xmm0, %xmm2 # CHECK-NEXT: - - 0.33 0.33 - - - 2.33 - - - - hsubps %xmm0, %xmm2
# CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 2.33 - - - - hsubps (%rax), %xmm2 # CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 2.33 - - - - hsubps (%rax), %xmm2
# CHECK-NEXT: - - - - 0.50 0.50 - - - - - - lddqu (%rax), %xmm2 # CHECK-NEXT: - - - - 0.50 0.50 - - - - - - lddqu (%rax), %xmm2
# CHECK-NEXT: - - 0.25 0.25 - - - 0.25 0.25 - - - monitor # CHECK-NEXT: - - 0.25 0.25 - - - 0.25 0.25 - - - monitor
# CHECK-NEXT: - - - - - - - 1.00 - - - - movddup %xmm0, %xmm2 # CHECK-NEXT: - - - - - - - 1.00 - - - - movddup %xmm0, %xmm2
# CHECK-NEXT: - - - - 0.50 0.50 - - - - - - movddup (%rax), %xmm2 # CHECK-NEXT: - - - - 0.50 0.50 - - - - - - movddup (%rax), %xmm2
# CHECK-NEXT: - - - - - - - 1.00 - - - - movshdup %xmm0, %xmm2 # CHECK-NEXT: - - - 0.50 - - - 0.50 - - - - movshdup %xmm0, %xmm2
# CHECK-NEXT: - - - - 0.50 0.50 - - - - - - movshdup (%rax), %xmm2 # CHECK-NEXT: - - - - 0.50 0.50 - - - - - - movshdup (%rax), %xmm2
# CHECK-NEXT: - - - - - - - 1.00 - - - - movsldup %xmm0, %xmm2 # CHECK-NEXT: - - - 0.50 - - - 0.50 - - - - movsldup %xmm0, %xmm2
# CHECK-NEXT: - - - - 0.50 0.50 - - - - - - movsldup (%rax), %xmm2 # CHECK-NEXT: - - - - 0.50 0.50 - - - - - - movsldup (%rax), %xmm2
# CHECK-NEXT: - - 1.75 1.75 - - - 2.75 3.75 - - - mwait # CHECK-NEXT: - - 1.75 1.75 - - - 2.75 3.75 - - - mwait