[RISCV] Assign separate PseudoVSHA2MS_VV opcodes for each SEW #114317
+33
−24
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
The vsha2ms.vv from Zvknh[ab] currently supports both SEW=32 and SEW=64. It might have different performance characteristics depending on the SEW on some processors. This patch splits these two different SEWs into their own VPsuedo opcodes and scheduling classes. This is effectively a NFC change.
mshockwave
requested review from
michaelmaitland,
dtcxzyw,
topperc and
wangpc-pp
|
@llvm/pr-subscribers-backend-risc-v Author: Min-Yih Hsu (mshockwave) ChangesThe vsha2ms.vv from Zvknh[ab] currently supports both SEW=32 and SEW=64. It might have different performance characteristics depending on the SEW on some processors. This patch splits these two different SEWs into their own VPsuedo opcodes and scheduling classes. This is effectively a NFC change. Full diff: https://github.com/llvm/llvm-project/pull/114317.diff 5 Files Affected:
diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfoVPseudos.td b/llvm/lib/Target/RISCV/RISCVInstrInfoVPseudos.td
index d5b0fa340684b4..be8bc94a76ff0a 100644
--- a/llvm/lib/Target/RISCV/RISCVInstrInfoVPseudos.td
+++ b/llvm/lib/Target/RISCV/RISCVInstrInfoVPseudos.td
@@ -4454,21 +4454,22 @@ class VPatTernaryNoMaskWithPolicy<string intrinsic,
ValueType result_type,
ValueType op1_type,
ValueType op2_type,
- int sew,
+ int log2sew,
LMULInfo vlmul,
VReg result_reg_class,
RegisterClass op1_reg_class,
- DAGOperand op2_kind> :
+ DAGOperand op2_kind,
+ bit sew_aware = false> :
Pat<(result_type (!cast<Intrinsic>(intrinsic)
(result_type result_reg_class:$rs3),
(op1_type op1_reg_class:$rs1),
(op2_type op2_kind:$rs2),
VLOpFrag, (XLenVT timm:$policy))),
- (!cast<Instruction>(inst#"_"#kind#"_"#vlmul.MX)
+ (!cast<Instruction>(inst#"_"#kind#"_"#vlmul.MX#!if(sew_aware, "_E"#!shl(1, log2sew), ""))
result_reg_class:$rs3,
(op1_type op1_reg_class:$rs1),
op2_kind:$rs2,
- GPR:$vl, sew, (XLenVT timm:$policy))>;
+ GPR:$vl, log2sew, (XLenVT timm:$policy))>;
class VPatTernaryNoMaskWithPolicyRoundingMode<string intrinsic,
string inst,
diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfoZvk.td b/llvm/lib/Target/RISCV/RISCVInstrInfoZvk.td
index 782651fd6d0197..93020d3fc5aecc 100644
--- a/llvm/lib/Target/RISCV/RISCVInstrInfoZvk.td
+++ b/llvm/lib/Target/RISCV/RISCVInstrInfoZvk.td
@@ -274,9 +274,11 @@ multiclass VPseudoBinaryNoMaskPolicy_Zvk<VReg RetClass,
multiclass VPseudoTernaryNoMask_Zvk<VReg RetClass,
VReg Op1Class,
DAGOperand Op2Class,
- LMULInfo MInfo> {
- let VLMul = MInfo.value in
- def "_" # MInfo.MX : VPseudoTernaryNoMask_Zvk<RetClass, Op1Class, Op2Class>;
+ LMULInfo MInfo, int sew = 0> {
+ let VLMul = MInfo.value, SEW = sew in {
+ defvar suffix = !if(sew, "_" # MInfo.MX # "_E" # sew, "_" # MInfo.MX);
+ def suffix : VPseudoTernaryNoMask_Zvk<RetClass, Op1Class, Op2Class>;
+ }
}
multiclass VPseudoBinaryV_V_NoMask_Zvk<LMULInfo m> {
@@ -348,12 +350,12 @@ multiclass VPseudoVSHA2CL {
}
}
-multiclass VPseudoVSHA2MS {
- foreach m = MxListVF4 in {
+multiclass VPseudoVSHA2MS<int sew = 0> {
+ foreach m = !if(!eq(sew, 64), MxListVF8, MxListVF4) in {
defvar mx = m.MX;
- defm _VV : VPseudoTernaryNoMask_Zvk<m.vrclass, m.vrclass, m.vrclass, m>,
+ defm _VV : VPseudoTernaryNoMask_Zvk<m.vrclass, m.vrclass, m.vrclass, m, sew = sew>,
SchedTernary<"WriteVSHA2MSV", "ReadVSHA2MSV", "ReadVSHA2MSV",
- "ReadVSHA2MSV", mx>;
+ "ReadVSHA2MSV", mx, sew>;
}
}
@@ -564,7 +566,9 @@ let Predicates = [HasStdExtZvkned] in {
let Predicates = [HasStdExtZvknhaOrZvknhb] in {
defm PseudoVSHA2CH : VPseudoVSHA2CH;
defm PseudoVSHA2CL : VPseudoVSHA2CL;
- defm PseudoVSHA2MS : VPseudoVSHA2MS;
+ defm PseudoVSHA2MS : VPseudoVSHA2MS<sew=32>;
+ let Predicates = [HasStdExtZvknhb] in
+ defm PseudoVSHA2MS : VPseudoVSHA2MS<sew=64>;
} // Predicates = [HasStdExtZvknhaOrZvknhb]
let Predicates = [HasStdExtZvksed] in {
@@ -944,12 +948,14 @@ multiclass VPatUnaryV_V_S_NoMask_Zvk<string intrinsic, string instruction,
}
multiclass VPatBinaryV_VV_NoMask<string intrinsic, string instruction,
- list<VTypeInfo> vtilist> {
+ list<VTypeInfo> vtilist,
+ bit isSEWAware = false> {
foreach vti = vtilist in
def : VPatTernaryNoMaskWithPolicy<intrinsic, instruction, "VV",
vti.Vector, vti.Vector, vti.Vector,
vti.Log2SEW, vti.LMul, vti.RegClass,
- vti.RegClass, vti.RegClass>;
+ vti.RegClass, vti.RegClass,
+ sew_aware = isSEWAware>;
}
multiclass VPatBinaryV_VI_NoMask<string intrinsic, string instruction,
@@ -1101,13 +1107,13 @@ let Predicates = [HasStdExtZvkned] in {
let Predicates = [HasStdExtZvknha] in {
defm : VPatBinaryV_VV_NoMask<"int_riscv_vsha2ch", "PseudoVSHA2CH", I32IntegerVectors>;
defm : VPatBinaryV_VV_NoMask<"int_riscv_vsha2cl", "PseudoVSHA2CH", I32IntegerVectors>;
- defm : VPatBinaryV_VV_NoMask<"int_riscv_vsha2ms", "PseudoVSHA2MS", I32IntegerVectors>;
+ defm : VPatBinaryV_VV_NoMask<"int_riscv_vsha2ms", "PseudoVSHA2MS", I32IntegerVectors, isSEWAware=true>;
} // Predicates = [HasStdExtZvknha]
let Predicates = [HasStdExtZvknhb] in {
defm : VPatBinaryV_VV_NoMask<"int_riscv_vsha2ch", "PseudoVSHA2CH", I32I64IntegerVectors>;
defm : VPatBinaryV_VV_NoMask<"int_riscv_vsha2cl", "PseudoVSHA2CH", I32I64IntegerVectors>;
- defm : VPatBinaryV_VV_NoMask<"int_riscv_vsha2ms", "PseudoVSHA2MS", I32I64IntegerVectors>;
+ defm : VPatBinaryV_VV_NoMask<"int_riscv_vsha2ms", "PseudoVSHA2MS", I32I64IntegerVectors, isSEWAware=true>;
} // Predicates = [HasStdExtZvknhb]
let Predicates = [HasStdExtZvksed] in {
diff --git a/llvm/lib/Target/RISCV/RISCVSchedSiFiveP400.td b/llvm/lib/Target/RISCV/RISCVSchedSiFiveP400.td
index 7a54d2fe108080..1af89903e0068c 100644
--- a/llvm/lib/Target/RISCV/RISCVSchedSiFiveP400.td
+++ b/llvm/lib/Target/RISCV/RISCVSchedSiFiveP400.td
@@ -883,7 +883,8 @@ foreach mx = SchedMxList in {
let Latency = 3, ReleaseAtCycles = [LMulLat] in {
defm "" : LMULWriteResMX<"WriteVSHA2CHV", [SiFiveP400VEXQ0], mx, IsWorstCase>;
defm "" : LMULWriteResMX<"WriteVSHA2CLV", [SiFiveP400VEXQ0], mx, IsWorstCase>;
- defm "" : LMULWriteResMX<"WriteVSHA2MSV", [SiFiveP400VEXQ0], mx, IsWorstCase>;
+ foreach sew = !listremove(SchedSEWSet<mx>.val, [8, 16]) in
+ defm "" : LMULSEWWriteResMXSEW<"WriteVSHA2MSV", [SiFiveP400VEXQ0], mx, sew, IsWorstCase>;
}
// Zvkned
let Latency = 2, ReleaseAtCycles = [LMulLat] in {
@@ -1213,7 +1214,7 @@ defm "" : LMULReadAdvance<"ReadVGMULV", 0>;
// Zvknha or Zvknhb
defm "" : LMULReadAdvance<"ReadVSHA2CHV", 0>;
defm "" : LMULReadAdvance<"ReadVSHA2CLV", 0>;
-defm "" : LMULReadAdvance<"ReadVSHA2MSV", 0>;
+defm "" : LMULSEWReadAdvance<"ReadVSHA2MSV", 0>;
// Zvkned
defm "" : LMULReadAdvance<"ReadVAESMVV", 0>;
defm "" : LMULReadAdvance<"ReadVAESKF1V", 0>;
diff --git a/llvm/lib/Target/RISCV/RISCVSchedSiFiveP600.td b/llvm/lib/Target/RISCV/RISCVSchedSiFiveP600.td
index c685a6d2b094be..51aa003139fbad 100644
--- a/llvm/lib/Target/RISCV/RISCVSchedSiFiveP600.td
+++ b/llvm/lib/Target/RISCV/RISCVSchedSiFiveP600.td
@@ -791,7 +791,8 @@ foreach mx = SchedMxList in {
let Latency = 3, ReleaseAtCycles = [LMulLat] in {
defm "" : LMULWriteResMX<"WriteVSHA2CHV", [SiFiveP600VectorArith], mx, IsWorstCase>;
defm "" : LMULWriteResMX<"WriteVSHA2CLV", [SiFiveP600VectorArith], mx, IsWorstCase>;
- defm "" : LMULWriteResMX<"WriteVSHA2MSV", [SiFiveP600VectorArith], mx, IsWorstCase>;
+ foreach sew = !listremove(SchedSEWSet<mx>.val, [8, 16]) in
+ defm "" : LMULSEWWriteResMXSEW<"WriteVSHA2MSV", [SiFiveP600VectorArith], mx, sew, IsWorstCase>;
}
// Zvkned
let Latency = 2, ReleaseAtCycles = [LMulLat] in {
@@ -1119,7 +1120,7 @@ defm "" : LMULReadAdvance<"ReadVGMULV", 0>;
// Zvknha or Zvknhb
defm "" : LMULReadAdvance<"ReadVSHA2CHV", 0>;
defm "" : LMULReadAdvance<"ReadVSHA2CLV", 0>;
-defm "" : LMULReadAdvance<"ReadVSHA2MSV", 0>;
+defm "" : LMULSEWReadAdvance<"ReadVSHA2MSV", 0>;
// Zvkned
defm "" : LMULReadAdvance<"ReadVAESMVV", 0>;
defm "" : LMULReadAdvance<"ReadVAESKF1V", 0>;
diff --git a/llvm/lib/Target/RISCV/RISCVScheduleZvk.td b/llvm/lib/Target/RISCV/RISCVScheduleZvk.td
index 640c456322f022..62d9bab0fac85f 100644
--- a/llvm/lib/Target/RISCV/RISCVScheduleZvk.td
+++ b/llvm/lib/Target/RISCV/RISCVScheduleZvk.td
@@ -36,7 +36,7 @@ defm "" : LMULSchedWrites<"WriteVGMULV">;
/// Zvknha or Zvknhb extensions
defm "" : LMULSchedWrites<"WriteVSHA2CHV">;
defm "" : LMULSchedWrites<"WriteVSHA2CLV">;
-defm "" : LMULSchedWrites<"WriteVSHA2MSV">;
+defm "" : LMULSEWSchedWrites<"WriteVSHA2MSV">;
/// Zvkned extension
defm "" : LMULSchedWrites<"WriteVAESMVV">;
@@ -79,7 +79,7 @@ defm "" : LMULSchedReads<"ReadVGMULV">;
/// Zvknha or Zvknhb extensions
defm "" : LMULSchedReads<"ReadVSHA2CHV">;
defm "" : LMULSchedReads<"ReadVSHA2CLV">;
-defm "" : LMULSchedReads<"ReadVSHA2MSV">;
+defm "" : LMULSEWSchedReads<"ReadVSHA2MSV">;
/// Zvkned extension
defm "" : LMULSchedReads<"ReadVAESMVV">;
@@ -153,11 +153,11 @@ multiclass UnsupportedSchedZvknhaOrZvknhb {
let Unsupported = true in {
defm "" : LMULWriteRes<"WriteVSHA2CHV", []>;
defm "" : LMULWriteRes<"WriteVSHA2CLV", []>;
-defm "" : LMULWriteRes<"WriteVSHA2MSV", []>;
+defm "" : LMULSEWWriteRes<"WriteVSHA2MSV", []>;
defm "" : LMULReadAdvance<"ReadVSHA2CHV", 0>;
defm "" : LMULReadAdvance<"ReadVSHA2CLV", 0>;
-defm "" : LMULReadAdvance<"ReadVSHA2MSV", 0>;
+defm "" : LMULSEWReadAdvance<"ReadVSHA2MSV", 0>;
}
}
|
topperc
reviewed
Oct 30, 2024
topperc
reviewed
Oct 30, 2024
| @@ -564,7 +566,9 @@ let Predicates = [HasStdExtZvkned] in { | |||
| let Predicates = [HasStdExtZvknhaOrZvknhb] in { | |||
| defm PseudoVSHA2CH : VPseudoVSHA2CH; | |||
| defm PseudoVSHA2CL : VPseudoVSHA2CL; | |||
| defm PseudoVSHA2MS : VPseudoVSHA2MS; | |||
| defm PseudoVSHA2MS : VPseudoVSHA2MS<sew=32>; | |||
| let Predicates = [HasStdExtZvknhb] in | |||
There was a problem hiding this comment.
Which let wins here. This one or the one on line 566?
There was a problem hiding this comment.
this one (line 570) wins.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The vsha2ms.vv from Zvknh[ab] currently supports both SEW=32 and SEW=64. It might have different performance characteristics depending on the SEW on some processors. This patch splits these two different SEWs into their own VPsuedo opcodes and scheduling classes.
This is effectively a NFC change.