maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

[X86] Fix WriteMPSAD/WritePSADBW values to allow us to remove unnecessary instrw overrides. 

llvm-svn: 330542
This commit is contained in:
Simon Pilgrim 2018-04-22 10:39:16 +00:00
parent 523fd335b1
commit a41ae2f005
6 changed files with 10 additions and 73 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
llvm/lib/Target/X86/X86SchedBroadwell.td
View File

@ -187,7 +187,7 @@ defm : BWWriteResPair<WriteShuffle, [BWPort5], 1>; // Vector shuffles.
defm : BWWriteResPair<WriteVarShuffle, [BWPort5], 1>; // Vector variable shuffles.
defm : BWWriteResPair<WriteBlend, [BWPort15], 1>; // Vector blends.
defm : BWWriteResPair<WriteVarBlend, [BWPort5], 2, [2]>; // Vector variable blends.
defm : BWWriteResPair<WriteMPSAD, [BWPort0, BWPort5], 6, [1, 2]>; // Vector MPSAD.
defm : BWWriteResPair<WriteMPSAD, [BWPort0, BWPort5], 7, [1, 2], 3, 5>; // Vector MPSAD.
defm : BWWriteResPair<WritePSADBW, [BWPort0], 5>; // Vector PSADBW.
// Conversion between integer and float.
@ -1335,13 +1335,6 @@ def: InstRW<[BWWriteResGroup77], (instregex "VBLENDPDYrmi",
"VPORYrm",
"VPXORYrm")>;
def BWWriteResGroup78 : SchedWriteRes<[BWPort0,BWPort5]> {
let Latency = 7;
let NumMicroOps = 3;
let ResourceCycles = [1,2];
}
def: InstRW<[BWWriteResGroup78], (instregex "(V?)MPSADBW(Y?)rri")>;
def BWWriteResGroup79 : SchedWriteRes<[BWPort5,BWPort23]> {
let Latency = 7;
let NumMicroOps = 3;
@ -1923,13 +1916,6 @@ def: InstRW<[BWWriteResGroup135], (instregex "ADD_FI16m",
"VROUNDPDYm",
"VROUNDPSYm")>;
def BWWriteResGroup136 : SchedWriteRes<[BWPort0,BWPort5,BWPort23]> {
let Latency = 12;
let NumMicroOps = 4;
let ResourceCycles = [1,2,1];
}
def: InstRW<[BWWriteResGroup136], (instregex "(V?)MPSADBWrmi")>;
def BWWriteResGroup137 : SchedWriteRes<[BWPort0,BWFPDivider]> {
let Latency = 11;
let NumMicroOps = 1;

16
llvm/lib/Target/X86/X86SchedHaswell.td
View File

@ -186,7 +186,7 @@ defm : HWWriteResPair<WriteShuffle256, [HWPort5], 3>;
defm : HWWriteResPair<WriteVarShuffle256, [HWPort5], 3>;
defm : HWWriteResPair<WriteVarBlend, [HWPort5], 2, [2]>;
defm : HWWriteResPair<WriteVarVecShift, [HWPort0, HWPort5], 2, [2, 1]>;
defm : HWWriteResPair<WriteMPSAD, [HWPort0, HWPort5], 6, [1, 2]>;
defm : HWWriteResPair<WriteMPSAD, [HWPort0, HWPort5], 7, [1, 2], 3, 6>;
defm : HWWriteResPair<WritePSADBW, [HWPort0], 5>;
// String instructions.
@ -2227,20 +2227,6 @@ def HWWriteResGroup109 : SchedWriteRes<[HWPort1,HWPort23,HWPort237,HWPort06,HWPo
def: InstRW<[HWWriteResGroup109], (instregex "SHLD(16|32|64)mrCL",
"SHRD(16|32|64)mrCL")>;
def HWWriteResGroup112 : SchedWriteRes<[HWPort0,HWPort5]> {
let Latency = 7;
let NumMicroOps = 3;
let ResourceCycles = [1,2];
}
def: InstRW<[HWWriteResGroup112], (instregex "(V?)MPSADBW(Y?)rri")>;
def HWWriteResGroup113 : SchedWriteRes<[HWPort0,HWPort5,HWPort23]> {
let Latency = 13;
let NumMicroOps = 4;
let ResourceCycles = [1,2,1];
}
def: InstRW<[HWWriteResGroup113], (instregex "(V?)MPSADBWrmi")>;
def HWWriteResGroup113_1 : SchedWriteRes<[HWPort0,HWPort5,HWPort23]> {
let Latency = 14;
let NumMicroOps = 4;

16
llvm/lib/Target/X86/X86SchedSandyBridge.td
View File

@ -170,7 +170,7 @@ defm : SBWriteResPair<WriteShuffle, [SBPort5], 1>;
defm : SBWriteResPair<WriteVarShuffle, [SBPort15], 1>;
defm : SBWriteResPair<WriteBlend, [SBPort15], 1>;
defm : SBWriteResPair<WriteVarBlend, [SBPort1, SBPort5], 2>;
defm : SBWriteResPair<WriteMPSAD, [SBPort0, SBPort15], 5, [1,2], 3>;
defm : SBWriteResPair<WriteMPSAD, [SBPort0, SBPort15], 7, [1,2], 3, 6>;
defm : SBWriteResPair<WritePSADBW, [SBPort0], 5>;
////////////////////////////////////////////////////////////////////////////////
@ -789,13 +789,6 @@ def SBWriteResGroup33 : SchedWriteRes<[SBPort4,SBPort23]> {
def: InstRW<[SBWriteResGroup33], (instregex "PUSH64i8",
"PUSH(16|32|64)r")>;
def SBWriteResGroup34 : SchedWriteRes<[SBPort0,SBPort15]> {
let Latency = 7;
let NumMicroOps = 3;
let ResourceCycles = [1,2];
}
def: InstRW<[SBWriteResGroup34], (instregex "(V?)MPSADBWrri")>;
def SBWriteResGroup35 : SchedWriteRes<[SBPort1,SBPort5]> {
let Latency = 5;
let NumMicroOps = 3;
@ -1649,13 +1642,6 @@ def: InstRW<[SBWriteResGroup107], (instregex "VCVTPD2DQYrm",
"VCVTPD2PSYrm",
"VCVTTPD2DQYrm")>;
def SBWriteResGroup108 : SchedWriteRes<[SBPort0,SBPort23,SBPort15]> {
let Latency = 13;
let NumMicroOps = 4;
let ResourceCycles = [1,1,2];
}
def: InstRW<[SBWriteResGroup108], (instregex "(V?)MPSADBWrmi")>;
def SBWriteResGroup109 : SchedWriteRes<[SBPort1,SBPort5,SBPort23]> {
let Latency = 11;
let NumMicroOps = 4;

16
llvm/lib/Target/X86/X86SchedSkylakeClient.td
View File

@ -184,7 +184,7 @@ defm : SKLWriteResPair<WriteShuffle, [SKLPort5], 1>; // Vector shuffles.
defm : SKLWriteResPair<WriteVarShuffle, [SKLPort5], 1>; // Vector shuffles.
defm : SKLWriteResPair<WriteBlend, [SKLPort15], 1>; // Vector blends.
defm : SKLWriteResPair<WriteVarBlend, [SKLPort5], 2, [2]>; // Vector variable blends.
defm : SKLWriteResPair<WriteMPSAD, [SKLPort0, SKLPort5], 6, [1, 2]>; // Vector MPSAD.
defm : SKLWriteResPair<WriteMPSAD, [SKLPort5], 4, [2], 2, 6>; // Vector MPSAD.
defm : SKLWriteResPair<WritePSADBW, [SKLPort5], 3>; // Vector PSADBW.
// Conversion between integer and float.
@ -958,13 +958,6 @@ def: InstRW<[SKLWriteResGroup48], (instregex "(V?)ADDPD(Y?)rr",
"(V?)SUBSDrr",
"(V?)SUBSSrr")>;
def SKLWriteResGroup50 : SchedWriteRes<[SKLPort5]> {
let Latency = 4;
let NumMicroOps = 2;
let ResourceCycles = [2];
}
def: InstRW<[SKLWriteResGroup50], (instregex "(V?)MPSADBW(Y?)rri")>;
def SKLWriteResGroup51 : SchedWriteRes<[SKLPort1,SKLPort5]> {
let Latency = 4;
let NumMicroOps = 2;
@ -1993,13 +1986,6 @@ def: InstRW<[SKLWriteResGroup134], (instregex "(V?)ADDPDrm",
def: InstRW<[SKLWriteResGroup134],
(instregex "VF(N)?M(ADD|SUB|ADDSUB|SUBADD)(132|213|231)P(D|S)m")>;
def SKLWriteResGroup137 : SchedWriteRes<[SKLPort5,SKLPort23]> {
let Latency = 10;
let NumMicroOps = 3;
let ResourceCycles = [2,1];
}
def: InstRW<[SKLWriteResGroup137], (instregex "(V?)MPSADBWrmi")>;
def SKLWriteResGroup138 : SchedWriteRes<[SKLPort0,SKLPort5,SKLPort23]> {
let Latency = 10;
let NumMicroOps = 3;

15
llvm/lib/Target/X86/X86SchedSkylakeServer.td
View File

@ -184,8 +184,8 @@ defm : SKXWriteResPair<WriteShuffle, [SKXPort5], 1>; // Vector shuffles.
defm : SKXWriteResPair<WriteVarShuffle, [SKXPort5], 1>; // Vector variable shuffles.
defm : SKXWriteResPair<WriteBlend, [SKXPort15], 1>; // Vector blends.
defm : SKXWriteResPair<WriteVarBlend, [SKXPort5], 2, [2]>; // Vector variable blends.
defm : SKXWriteResPair<WriteMPSAD, [SKXPort0, SKXPort5], 6, [1, 2]>; // Vector MPSAD.
defm : SKXWriteResPair<WritePSADBW, [SKXPort5], 3>; // Vector PSADBW.
defm : SKXWriteResPair<WriteMPSAD, [SKXPort5], 4, [2], 2, 6>; // Vector MPSAD.
defm : SKXWriteResPair<WritePSADBW, [SKXPort5], 3, [1,1], 1, 6>; // Vector PSADBW.
// Conversion between integer and float.
defm : SKXWriteResPair<WriteCvtF2I, [SKXPort1], 3>; // Float -> Integer.
@ -1869,15 +1869,12 @@ def SKXWriteResGroup51 : SchedWriteRes<[SKXPort5]> {
let NumMicroOps = 2;
let ResourceCycles = [2];
}
def: InstRW<[SKXWriteResGroup51], (instregex "MPSADBWrri",
"VEXPANDPDZ128rr",
def: InstRW<[SKXWriteResGroup51], (instregex "VEXPANDPDZ128rr",
"VEXPANDPDZ256rr",
"VEXPANDPDZrr",
"VEXPANDPSZ128rr",
"VEXPANDPSZ256rr",
"VEXPANDPSZrr",
"VMPSADBWYrri",
"VMPSADBWrri",
"VPEXPANDDZ128rr",
"VPEXPANDDZ256rr",
"VPEXPANDDZrr",
@ -3702,7 +3699,6 @@ def SKXWriteResGroup136 : SchedWriteRes<[SKXPort5,SKXPort23]> {
let ResourceCycles = [1,1];
}
def: InstRW<[SKXWriteResGroup136], (instregex "PCMPGTQrm",
"PSADBWrm",
"VALIGNDZ128rm(b?)i",
"VALIGNQZ128rm(b?)i",
"VCMPPDZ128rm(b?)i",
@ -3757,7 +3753,6 @@ def: InstRW<[SKXWriteResGroup136], (instregex "PCMPGTQrm",
"VPMOVZXWDZ128rm(b?)",
"VPMOVZXWQZ128rm(b?)",
"VPSADBWZ128rm(b?)",
"VPSADBWrm",
"VPTESTMBZ128rm(b?)",
"VPTESTMDZ128rm(b?)",
"VPTESTMQZ128rm(b?)",
@ -4179,10 +4174,8 @@ def SKXWriteResGroup151 : SchedWriteRes<[SKXPort5,SKXPort23]> {
let NumMicroOps = 3;
let ResourceCycles = [2,1];
}
def: InstRW<[SKXWriteResGroup151], (instregex "MPSADBWrmi",
"VEXPANDPDZ128rm(b?)",
def: InstRW<[SKXWriteResGroup151], (instregex "VEXPANDPDZ128rm(b?)",
"VEXPANDPSZ128rm(b?)",
"VMPSADBWrmi",
"VPEXPANDDZ128rm(b?)",
"VPEXPANDQZ128rm(b?)")>;

4
llvm/test/CodeGen/X86/avx2-schedule.ll
View File

@ -608,8 +608,8 @@ declare <4 x i64> @llvm.x86.avx2.movntdqa(i8*) nounwind readonly
define <16 x i16> @test_mpsadbw(<32 x i8> %a0, <32 x i8> %a1, <32 x i8> *%a2) {
; GENERIC-LABEL: test_mpsadbw:
; GENERIC: # %bb.0:
; GENERIC-NEXT: vmpsadbw $7, %ymm1, %ymm0, %ymm0 # sched: [5:1.00]
; GENERIC-NEXT: vmpsadbw $7, (%rdi), %ymm0, %ymm0 # sched: [10:1.00]
; GENERIC-NEXT: vmpsadbw $7, %ymm1, %ymm0, %ymm0 # sched: [7:1.00]
; GENERIC-NEXT: vmpsadbw $7, (%rdi), %ymm0, %ymm0 # sched: [13:1.00]
; GENERIC-NEXT: retq # sched: [1:1.00]
;
; HASWELL-LABEL: test_mpsadbw: