[llvm] [MachineOutliner] Efficient Implementation of MachineOutliner::findCandidates() (PR #90260)
Xuan Zhang via llvm-commits
llvm-commits at lists.llvm.org
Wed May 29 07:35:01 PDT 2024
https://github.com/xuanzh-meta updated https://github.com/llvm/llvm-project/pull/90260
>From 046ddd4db1589e830339a7fb5b42c507f3d6015c Mon Sep 17 00:00:00 2001
From: Xuan Zhang <xuanzh at meta.com>
Date: Fri, 26 Apr 2024 12:44:02 -0700
Subject: [PATCH 1/4] efficient implementation of
MachineOutliner::findCandidates()
---
llvm/lib/CodeGen/MachineOutliner.cpp | 22 ++++++++++---------
.../AArch64/machine-outliner-overlap.mir | 12 +++++-----
2 files changed, 18 insertions(+), 16 deletions(-)
diff --git a/llvm/lib/CodeGen/MachineOutliner.cpp b/llvm/lib/CodeGen/MachineOutliner.cpp
index dc2f5ef15206e..d553c0e6d2477 100644
--- a/llvm/lib/CodeGen/MachineOutliner.cpp
+++ b/llvm/lib/CodeGen/MachineOutliner.cpp
@@ -593,7 +593,11 @@ void MachineOutliner::findCandidates(
unsigned NumDiscarded = 0;
unsigned NumKept = 0;
#endif
- for (const unsigned &StartIdx : RS.StartIndices) {
+ // Sort the start indices so that we can efficiently check if candidates
+ // overlap with each other in MachineOutliner::findCandidates().
+ SmallVector<unsigned> SortedStartIndices(RS.StartIndices);
+ llvm::sort(SortedStartIndices);
+ for (const unsigned &StartIdx : SortedStartIndices) {
// Trick: Discard some candidates that would be incompatible with the
// ones we've already found for this sequence. This will save us some
// work in candidate selection.
@@ -616,17 +620,15 @@ void MachineOutliner::findCandidates(
// * End before the other starts
// * Start after the other ends
unsigned EndIdx = StartIdx + StringLen - 1;
- auto FirstOverlap = find_if(
- CandidatesForRepeatedSeq, [StartIdx, EndIdx](const Candidate &C) {
- return EndIdx >= C.getStartIdx() && StartIdx <= C.getEndIdx();
- });
- if (FirstOverlap != CandidatesForRepeatedSeq.end()) {
+ if (CandidatesForRepeatedSeq.size() > 0 &&
+ StartIdx <= CandidatesForRepeatedSeq.back().getEndIdx()) {
#ifndef NDEBUG
++NumDiscarded;
- LLVM_DEBUG(dbgs() << " .. DISCARD candidate @ [" << StartIdx
- << ", " << EndIdx << "]; overlaps with candidate @ ["
- << FirstOverlap->getStartIdx() << ", "
- << FirstOverlap->getEndIdx() << "]\n");
+ LLVM_DEBUG(dbgs() << " .. DISCARD candidate @ [" << StartIdx << ", "
+ << EndIdx << "]; overlaps with candidate @ ["
+ << CandidatesForRepeatedSeq.back().getStartIdx()
+ << ", " << CandidatesForRepeatedSeq.back().getEndIdx()
+ << "]\n");
#endif
continue;
}
diff --git a/llvm/test/CodeGen/AArch64/machine-outliner-overlap.mir b/llvm/test/CodeGen/AArch64/machine-outliner-overlap.mir
index 649bb33828c32..c6bd4c1d04d87 100644
--- a/llvm/test/CodeGen/AArch64/machine-outliner-overlap.mir
+++ b/llvm/test/CodeGen/AArch64/machine-outliner-overlap.mir
@@ -8,27 +8,27 @@
# CHECK-NEXT: Candidates discarded: 0
# CHECK-NEXT: Candidates kept: 2
# CHECK-DAG: Sequence length: 8
-# CHECK-NEXT: .. DISCARD candidate @ [5, 12]; overlaps with candidate @ [12, 19]
+# CHECK-NEXT: .. DISCARD candidate @ [12, 19]; overlaps with candidate @ [5, 12]
# CHECK-NEXT: Candidates discarded: 1
# CHECK-NEXT: Candidates kept: 1
# CHECK-DAG: Sequence length: 9
-# CHECK-NEXT: .. DISCARD candidate @ [4, 12]; overlaps with candidate @ [11, 19]
+# CHECK-NEXT: .. DISCARD candidate @ [11, 19]; overlaps with candidate @ [4, 12]
# CHECK-NEXT: Candidates discarded: 1
# CHECK-NEXT: Candidates kept: 1
# CHECK-DAG: Sequence length: 10
-# CHECK-NEXT: .. DISCARD candidate @ [3, 12]; overlaps with candidate @ [10, 19]
+# CHECK-NEXT: .. DISCARD candidate @ [10, 19]; overlaps with candidate @ [3, 12]
# CHECK-NEXT: Candidates discarded: 1
# CHECK-NEXT: Candidates kept: 1
# CHECK-DAG: Sequence length: 11
-# CHECK-NEXT: .. DISCARD candidate @ [2, 12]; overlaps with candidate @ [9, 19]
+# CHECK-NEXT: .. DISCARD candidate @ [9, 19]; overlaps with candidate @ [2, 12]
# CHECK-NEXT: Candidates discarded: 1
# CHECK-NEXT: Candidates kept: 1
# CHECK-DAG: Sequence length: 12
-# CHECK-NEXT: .. DISCARD candidate @ [1, 12]; overlaps with candidate @ [8, 19]
+# CHECK-NEXT: .. DISCARD candidate @ [8, 19]; overlaps with candidate @ [1, 12]
# CHECK-NEXT: Candidates discarded: 1
# CHECK-NEXT: Candidates kept: 1
# CHECK-DAG: Sequence length: 13
-# CHECK-NEXT: .. DISCARD candidate @ [0, 12]; overlaps with candidate @ [7, 19]
+# CHECK-NEXT: .. DISCARD candidate @ [7, 19]; overlaps with candidate @ [0, 12]
# CHECK-NEXT: Candidates discarded: 1
# CHECK-NEXT: Candidates kept: 1
>From de64e6fa3a4e8410e1f16763b15777bc22953291 Mon Sep 17 00:00:00 2001
From: Xuan Zhang <xuanzh at meta.com>
Date: Fri, 3 May 2024 08:32:20 -0700
Subject: [PATCH 2/4] update check on empty vector
---
llvm/lib/CodeGen/MachineOutliner.cpp | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/llvm/lib/CodeGen/MachineOutliner.cpp b/llvm/lib/CodeGen/MachineOutliner.cpp
index d553c0e6d2477..4b75f1cd4ae43 100644
--- a/llvm/lib/CodeGen/MachineOutliner.cpp
+++ b/llvm/lib/CodeGen/MachineOutliner.cpp
@@ -620,7 +620,7 @@ void MachineOutliner::findCandidates(
// * End before the other starts
// * Start after the other ends
unsigned EndIdx = StartIdx + StringLen - 1;
- if (CandidatesForRepeatedSeq.size() > 0 &&
+ if (!CandidatesForRepeatedSeq.empty() &&
StartIdx <= CandidatesForRepeatedSeq.back().getEndIdx()) {
#ifndef NDEBUG
++NumDiscarded;
>From 358c9541645156646b0de844982a4ccb193ecfd9 Mon Sep 17 00:00:00 2001
From: Xuan Zhang <xuanzh at meta.com>
Date: Mon, 6 May 2024 07:51:51 -0700
Subject: [PATCH 3/4] retrigger checks
>From 632f4820ee172cba6acb7b8cb85a116ffd729fb1 Mon Sep 17 00:00:00 2001
From: Xuan Zhang <xuanzh at meta.com>
Date: Tue, 28 May 2024 13:22:02 -0700
Subject: [PATCH 4/4] remove const and sort in place
---
llvm/lib/CodeGen/MachineOutliner.cpp | 7 +++----
1 file changed, 3 insertions(+), 4 deletions(-)
diff --git a/llvm/lib/CodeGen/MachineOutliner.cpp b/llvm/lib/CodeGen/MachineOutliner.cpp
index 4b75f1cd4ae43..9d192f67e6579 100644
--- a/llvm/lib/CodeGen/MachineOutliner.cpp
+++ b/llvm/lib/CodeGen/MachineOutliner.cpp
@@ -584,7 +584,7 @@ void MachineOutliner::findCandidates(
LLVM_DEBUG(dbgs() << "*** Discarding overlapping candidates *** \n");
LLVM_DEBUG(
dbgs() << "Searching for overlaps in all repeated sequences...\n");
- for (const SuffixTree::RepeatedSubstring &RS : ST) {
+ for (SuffixTree::RepeatedSubstring &RS : ST) {
CandidatesForRepeatedSeq.clear();
unsigned StringLen = RS.Length;
LLVM_DEBUG(dbgs() << " Sequence length: " << StringLen << "\n");
@@ -595,9 +595,8 @@ void MachineOutliner::findCandidates(
#endif
// Sort the start indices so that we can efficiently check if candidates
// overlap with each other in MachineOutliner::findCandidates().
- SmallVector<unsigned> SortedStartIndices(RS.StartIndices);
- llvm::sort(SortedStartIndices);
- for (const unsigned &StartIdx : SortedStartIndices) {
+ llvm::sort(RS.StartIndices);
+ for (const unsigned &StartIdx : RS.StartIndices) {
// Trick: Discard some candidates that would be incompatible with the
// ones we've already found for this sequence. This will save us some
// work in candidate selection.
More information about the llvm-commits
mailing list