-
Notifications
You must be signed in to change notification settings - Fork 14.3k
[RISCV] [MachineOutliner] Analyze all candidates #127659
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
@llvm/pr-subscribers-backend-risc-v Author: Sudharsan Veeravalli (svs-quic) Changes#117700 made a change from analyzing all the candidates to analyzing just the first candidate before deciding to either delete or keep all of them. Even though the candidates all have the same instructions, the basic blocks in which they are present are different and we will need to check each of them before deciding whether to keep or erase them. Particularly, Similarly, deleting all the candidates if the first one is not viable will result in missed outlining opportunities. Full diff: https://github.com/llvm/llvm-project/pull/127659.diff 2 Files Affected:
diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfo.cpp b/llvm/lib/Target/RISCV/RISCVInstrInfo.cpp
index 456fb66917216..bc859887c4aac 100644
--- a/llvm/lib/Target/RISCV/RISCVInstrInfo.cpp
+++ b/llvm/lib/Target/RISCV/RISCVInstrInfo.cpp
@@ -3015,30 +3015,25 @@ static bool cannotInsertTailCall(const MachineBasicBlock &MBB) {
return false;
}
-static std::optional<MachineOutlinerConstructionID>
-analyzeCandidate(outliner::Candidate &C) {
+static bool analyzeCandidate(outliner::Candidate &C) {
// If last instruction is return then we can rely on
// the verification already performed in the getOutliningTypeImpl.
if (C.back().isReturn()) {
assert(!cannotInsertTailCall(*C.getMBB()) &&
"The candidate who uses return instruction must be outlined "
"using tail call");
- return MachineOutlinerTailCall;
+ return false;
}
- auto CandidateUsesX5 = [](outliner::Candidate &C) {
- const TargetRegisterInfo *TRI = C.getMF()->getSubtarget().getRegisterInfo();
- if (std::any_of(C.begin(), C.end(), [TRI](const MachineInstr &MI) {
- return isMIModifiesReg(MI, TRI, RISCV::X5);
- }))
- return true;
- return !C.isAvailableAcrossAndOutOfSeq(RISCV::X5, *TRI);
- };
-
- if (!CandidateUsesX5(C))
- return MachineOutlinerDefault;
+ // Filter out candidates where the X5 register (t0) can't be used to setup
+ // the function call.
+ const TargetRegisterInfo *TRI = C.getMF()->getSubtarget().getRegisterInfo();
+ if (std::any_of(C.begin(), C.end(), [TRI](const MachineInstr &MI) {
+ return isMIModifiesReg(MI, TRI, RISCV::X5);
+ }))
+ return true;
- return std::nullopt;
+ return !C.isAvailableAcrossAndOutOfSeq(RISCV::X5, *TRI);
}
std::optional<std::unique_ptr<outliner::OutlinedFunction>>
@@ -3047,35 +3042,32 @@ RISCVInstrInfo::getOutliningCandidateInfo(
std::vector<outliner::Candidate> &RepeatedSequenceLocs,
unsigned MinRepeats) const {
- // Each RepeatedSequenceLoc is identical.
- outliner::Candidate &Candidate = RepeatedSequenceLocs[0];
- auto CandidateInfo = analyzeCandidate(Candidate);
- if (!CandidateInfo)
- RepeatedSequenceLocs.clear();
+ // Analyze each candidate and erase the ones that are not viable.
+ llvm::erase_if(RepeatedSequenceLocs, analyzeCandidate);
// If the sequence doesn't have enough candidates left, then we're done.
if (RepeatedSequenceLocs.size() < MinRepeats)
return std::nullopt;
+ // Each RepeatedSequenceLoc is identical.
+ outliner::Candidate &Candidate = RepeatedSequenceLocs[0];
unsigned InstrSizeCExt =
Candidate.getMF()->getSubtarget<RISCVSubtarget>().hasStdExtCOrZca() ? 2
: 4;
unsigned CallOverhead = 0, FrameOverhead = 0;
- MachineOutlinerConstructionID MOCI = CandidateInfo.value();
- switch (MOCI) {
- case MachineOutlinerDefault:
- // call t0, function = 8 bytes.
- CallOverhead = 8;
- // jr t0 = 4 bytes, 2 bytes if compressed instructions are enabled.
- FrameOverhead = InstrSizeCExt;
- break;
- case MachineOutlinerTailCall:
+ MachineOutlinerConstructionID MOCI = MachineOutlinerDefault;
+ if (Candidate.back().isReturn()) {
+ MOCI = MachineOutlinerTailCall;
// tail call = auipc + jalr in the worst case without linker relaxation.
CallOverhead = 4 + InstrSizeCExt;
// Using tail call we move ret instruction from caller to callee.
FrameOverhead = 0;
- break;
+ } else {
+ // call t0, function = 8 bytes.
+ CallOverhead = 8;
+ // jr t0 = 4 bytes, 2 bytes if compressed instructions are enabled.
+ FrameOverhead = InstrSizeCExt;
}
for (auto &C : RepeatedSequenceLocs)
diff --git a/llvm/test/CodeGen/RISCV/machine-outliner-call-x5-liveout.mir b/llvm/test/CodeGen/RISCV/machine-outliner-call-x5-liveout.mir
new file mode 100644
index 0000000000000..5da06850e5a67
--- /dev/null
+++ b/llvm/test/CodeGen/RISCV/machine-outliner-call-x5-liveout.mir
@@ -0,0 +1,146 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 5
+# RUN: llc -mtriple=riscv32 -x mir -run-pass=machine-outliner -simplify-mir -verify-machineinstrs < %s \
+# RUN: | FileCheck -check-prefixes=RV32I-MO %s
+# RUN: llc -mtriple=riscv64 -x mir -run-pass=machine-outliner -simplify-mir -verify-machineinstrs < %s \
+# RUN: | FileCheck -check-prefixes=RV64I-MO %s
+
+# MIR has been edited by hand to have x5 as live out in @dont_outline
+
+--- |
+ define i32 @outline_0(i32 %a, i32 %b) { ret i32 0 }
+
+ ; Can't outline if x5 is live out of the MBB
+ define i32 @dont_outline(i32 %a, i32 %b) { ret i32 0 }
+
+ define i32 @outline_1(i32 %a, i32 %b) { ret i32 0 }
+
+ define i32 @outline_2(i32 %a, i32 %b) { ret i32 0 }
+...
+---
+
+name: outline_0
+tracksRegLiveness: true
+isOutlined: false
+body: |
+ bb.0:
+ liveins: $x10, $x11
+
+ ; RV32I-MO-LABEL: name: outline_0
+ ; RV32I-MO: liveins: $x10, $x11
+ ; RV32I-MO-NEXT: {{ $}}
+ ; RV32I-MO-NEXT: $x5 = PseudoCALLReg target-flags(riscv-call) @OUTLINED_FUNCTION_0, implicit-def $x5, implicit-def $x10, implicit-def $x11, implicit-def $x12, implicit $x10, implicit $x11
+ ; RV32I-MO-NEXT: PseudoRET implicit $x11
+ ;
+ ; RV64I-MO-LABEL: name: outline_0
+ ; RV64I-MO: liveins: $x10, $x11
+ ; RV64I-MO-NEXT: {{ $}}
+ ; RV64I-MO-NEXT: $x5 = PseudoCALLReg target-flags(riscv-call) @OUTLINED_FUNCTION_0, implicit-def $x5, implicit-def $x10, implicit-def $x11, implicit-def $x12, implicit $x10, implicit $x11
+ ; RV64I-MO-NEXT: PseudoRET implicit $x11
+ $x11 = ORI $x11, 1023
+ $x12 = ADDI $x10, 17
+ $x11 = AND $x12, $x11
+ $x10 = SUB $x10, $x11
+ PseudoRET implicit $x11
+
+...
+---
+
+name: dont_outline
+tracksRegLiveness: true
+isOutlined: false
+body: |
+ ; RV32I-MO-LABEL: name: dont_outline
+ ; RV32I-MO: bb.0:
+ ; RV32I-MO-NEXT: liveins: $x10, $x11
+ ; RV32I-MO-NEXT: {{ $}}
+ ; RV32I-MO-NEXT: $x11 = ORI $x11, 1023
+ ; RV32I-MO-NEXT: $x12 = ADDI $x10, 17
+ ; RV32I-MO-NEXT: $x11 = AND $x12, $x11
+ ; RV32I-MO-NEXT: $x10 = SUB $x10, $x11
+ ; RV32I-MO-NEXT: BEQ $x10, $x11, %bb.1
+ ; RV32I-MO-NEXT: {{ $}}
+ ; RV32I-MO-NEXT: bb.1:
+ ; RV32I-MO-NEXT: liveins: $x10, $x5
+ ; RV32I-MO-NEXT: {{ $}}
+ ; RV32I-MO-NEXT: PseudoRET implicit $x10, implicit $x5
+ ;
+ ; RV64I-MO-LABEL: name: dont_outline
+ ; RV64I-MO: bb.0:
+ ; RV64I-MO-NEXT: liveins: $x10, $x11
+ ; RV64I-MO-NEXT: {{ $}}
+ ; RV64I-MO-NEXT: $x11 = ORI $x11, 1023
+ ; RV64I-MO-NEXT: $x12 = ADDI $x10, 17
+ ; RV64I-MO-NEXT: $x11 = AND $x12, $x11
+ ; RV64I-MO-NEXT: $x10 = SUB $x10, $x11
+ ; RV64I-MO-NEXT: BEQ $x10, $x11, %bb.1
+ ; RV64I-MO-NEXT: {{ $}}
+ ; RV64I-MO-NEXT: bb.1:
+ ; RV64I-MO-NEXT: liveins: $x10, $x5
+ ; RV64I-MO-NEXT: {{ $}}
+ ; RV64I-MO-NEXT: PseudoRET implicit $x10, implicit $x5
+ bb.0:
+ liveins: $x10, $x11
+
+ $x11 = ORI $x11, 1023
+ $x12 = ADDI $x10, 17
+ $x11 = AND $x12, $x11
+ $x10 = SUB $x10, $x11
+ BEQ $x10, $x11, %bb.1
+
+ bb.1:
+ liveins: $x10, $x5
+ PseudoRET implicit $x10, implicit $x5
+
+...
+---
+
+name: outline_1
+tracksRegLiveness: true
+isOutlined: false
+body: |
+ bb.0:
+ liveins: $x10, $x11
+
+ ; RV32I-MO-LABEL: name: outline_1
+ ; RV32I-MO: liveins: $x10, $x11
+ ; RV32I-MO-NEXT: {{ $}}
+ ; RV32I-MO-NEXT: $x5 = PseudoCALLReg target-flags(riscv-call) @OUTLINED_FUNCTION_0, implicit-def $x5, implicit-def $x10, implicit-def $x11, implicit-def $x12, implicit $x10, implicit $x11
+ ; RV32I-MO-NEXT: PseudoRET implicit $x10
+ ;
+ ; RV64I-MO-LABEL: name: outline_1
+ ; RV64I-MO: liveins: $x10, $x11
+ ; RV64I-MO-NEXT: {{ $}}
+ ; RV64I-MO-NEXT: $x5 = PseudoCALLReg target-flags(riscv-call) @OUTLINED_FUNCTION_0, implicit-def $x5, implicit-def $x10, implicit-def $x11, implicit-def $x12, implicit $x10, implicit $x11
+ ; RV64I-MO-NEXT: PseudoRET implicit $x10
+ $x11 = ORI $x11, 1023
+ $x12 = ADDI $x10, 17
+ $x11 = AND $x12, $x11
+ $x10 = SUB $x10, $x11
+ PseudoRET implicit $x10
+
+...
+---
+name: outline_2
+tracksRegLiveness: true
+isOutlined: false
+body: |
+ bb.0:
+ liveins: $x10, $x11
+
+ ; RV32I-MO-LABEL: name: outline_2
+ ; RV32I-MO: liveins: $x10, $x11
+ ; RV32I-MO-NEXT: {{ $}}
+ ; RV32I-MO-NEXT: $x5 = PseudoCALLReg target-flags(riscv-call) @OUTLINED_FUNCTION_0, implicit-def $x5, implicit-def $x10, implicit-def $x11, implicit-def $x12, implicit $x10, implicit $x11
+ ; RV32I-MO-NEXT: PseudoRET implicit $x12
+ ;
+ ; RV64I-MO-LABEL: name: outline_2
+ ; RV64I-MO: liveins: $x10, $x11
+ ; RV64I-MO-NEXT: {{ $}}
+ ; RV64I-MO-NEXT: $x5 = PseudoCALLReg target-flags(riscv-call) @OUTLINED_FUNCTION_0, implicit-def $x5, implicit-def $x10, implicit-def $x11, implicit-def $x12, implicit $x10, implicit $x11
+ ; RV64I-MO-NEXT: PseudoRET implicit $x12
+ $x11 = ORI $x11, 1023
+ $x12 = ADDI $x10, 17
+ $x11 = AND $x12, $x11
+ $x10 = SUB $x10, $x11
+ PseudoRET implicit $x12
+...
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Nice Find.
I think we can afford to be a little clearer about the difference between checks on the candidate's context (i.e. where it would be called from) vs its content (the instructions inside it), but maybe that's not helpful as when looking at the instructions most of our analysis is about e.g. liveness.
I'm not sure we need to fix the costing issue today, but I'm not sure.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I'm happy with how this works and fixes the issue.
I think this should also be backported to the release branch, as I think it will affect a lot of tests that use -Oz
.
I've added this to the 20.x release as I do think this is a big enough bug that we should fix it on the branch. I hope that's not too controversial. |
@mga-sc @wangpc-pp @resistor Please can one of you also take a look at this review? I feel relatively new to the MachineOutliner, so I would appreciate a second +1 before this is merged. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM.
/cherry-pick 6757cf4 |
Failed to cherry-pick: 6757cf4 https://github.com/llvm/llvm-project/actions/runs/13452069337 Please manually backport the fix and push it to your github fork. Once this is done, please create a pull request |
/pull-request #128146 |
llvm#117700 made a change from analyzing all the candidates to analyzing just the first candidate before deciding to either delete or keep all of them. Even though the candidates all have the same instructions, the basic blocks in which they are present are different and we will need to check each of them before deciding whether to keep or erase them. Particularly, `isAvailableAcrossAndOutOfSeq` checks to see if the register (x5 in this case) is available from the end of the MBB to the beginning of the candidate and not checking this for each candidate led to incorrect candidates being outlined resulting in correctness issues in a few downstream benchmarks. Similarly, deleting all the candidates if the first one is not viable will result in missed outlining opportunities. (cherry picked from commit 6757cf4)
#117700 made a change from analyzing all the candidates to analyzing just the first candidate before deciding to either delete or keep all of them.
Even though the candidates all have the same instructions, the basic blocks in which they are present are different and we will need to check each of them before deciding whether to keep or erase them. Particularly,
isAvailableAcrossAndOutOfSeq
checks to see if the register (x5 in this case) is available from the end of the MBB to the beginning of the candidate and not checking this for each candidate led to incorrect candidates being outlined resulting in correctness issues in a few downstream benchmarks.Similarly, deleting all the candidates if the first one is not viable will result in missed outlining opportunities.