[llvm] [SCEV] Generalize (C * A /u C) -> A fold to (C1 * A /u C2) -> C1/C2 * A. (PR #157159)
Florian Hahn via llvm-commits
llvm-commits at lists.llvm.org
Fri Sep 5 11:56:21 PDT 2025
https://github.com/fhahn created https://github.com/llvm/llvm-project/pull/157159
Generalize fold added in https://github.com/llvm/llvm-project/commit/74ec38fad0a1289f936e5388fa8bbe74653c55d9 to support multiplying and dividing
by different constants, given they are both powers-of-2 and C1 is a
multiple of C2, checked via their trailing zeros.
https://alive2.llvm.org/ce/z/eqJ2xj
>From 1b5045525e051c9f65f381b4feff39822c5e5f2c Mon Sep 17 00:00:00 2001
From: Florian Hahn <flo at fhahn.com>
Date: Fri, 5 Sep 2025 16:33:40 +0100
Subject: [PATCH 1/2] Add test multiplying by 6
---
.../ScalarEvolution/mul-udiv-folds.ll | 46 +++++++++++++++++++
1 file changed, 46 insertions(+)
diff --git a/llvm/test/Analysis/ScalarEvolution/mul-udiv-folds.ll b/llvm/test/Analysis/ScalarEvolution/mul-udiv-folds.ll
index 9f4360d2ae383..684c30d9f2cde 100644
--- a/llvm/test/Analysis/ScalarEvolution/mul-udiv-folds.ll
+++ b/llvm/test/Analysis/ScalarEvolution/mul-udiv-folds.ll
@@ -48,6 +48,52 @@ exit:
ret void
}
+define void @udiv4_and_udiv2_mul_6(i1 %c, ptr %A) {
+; CHECK-LABEL: 'udiv4_and_udiv2_mul_6'
+; CHECK-NEXT: Classifying expressions for: @udiv4_and_udiv2_mul_6
+; CHECK-NEXT: %start = select i1 %c, i32 512, i32 0
+; CHECK-NEXT: --> %start U: [0,513) S: [0,513)
+; CHECK-NEXT: %div.2 = lshr i32 %start, 1
+; CHECK-NEXT: --> (%start /u 2) U: [0,257) S: [0,257)
+; CHECK-NEXT: %div.4 = lshr i32 %start, 2
+; CHECK-NEXT: --> (%start /u 4) U: [0,129) S: [0,129)
+; CHECK-NEXT: %iv.start = zext i32 %div.4 to i64
+; CHECK-NEXT: --> ((zext i32 %start to i64) /u 4) U: [0,129) S: [0,129)
+; CHECK-NEXT: %wide.trip.count = zext i32 %div.2 to i64
+; CHECK-NEXT: --> ((zext i32 %start to i64) /u 2) U: [0,257) S: [0,257)
+; CHECK-NEXT: %iv = phi i64 [ %iv.start, %entry ], [ %iv.next, %loop ]
+; CHECK-NEXT: --> {((zext i32 %start to i64) /u 4),+,1}<%loop> U: full-set S: full-set Exits: ((zext i32 %start to i64) /u 2) LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: %gep = getelementptr <{ i32, i16 }>, ptr %A, i64 %iv
+; CHECK-NEXT: --> {((6 * ((zext i32 %start to i64) /u 4))<nuw><nsw> + %A),+,6}<%loop> U: full-set S: full-set Exits: ((6 * ((zext i32 %start to i64) /u 2))<nuw><nsw> + %A) LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: %iv.next = add i64 %iv, 1
+; CHECK-NEXT: --> {(1 + ((zext i32 %start to i64) /u 4))<nuw><nsw>,+,1}<%loop> U: full-set S: full-set Exits: (1 + ((zext i32 %start to i64) /u 2))<nuw><nsw> LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: Determining loop execution counts for: @udiv4_and_udiv2_mul_6
+; CHECK-NEXT: Loop %loop: backedge-taken count is ((-1 * ((zext i32 %start to i64) /u 4))<nsw> + ((zext i32 %start to i64) /u 2))
+; CHECK-NEXT: Loop %loop: constant max backedge-taken count is i64 -1
+; CHECK-NEXT: Loop %loop: symbolic max backedge-taken count is ((-1 * ((zext i32 %start to i64) /u 4))<nsw> + ((zext i32 %start to i64) /u 2))
+; CHECK-NEXT: Loop %loop: Trip multiple is 1
+;
+entry:
+ %start = select i1 %c, i32 512, i32 0
+ %div.2 = lshr i32 %start, 1
+ %div.4 = lshr i32 %start, 2
+ %iv.start = zext i32 %div.4 to i64
+ %wide.trip.count = zext i32 %div.2 to i64
+ br label %loop
+
+loop:
+ %iv = phi i64 [ %iv.start, %entry ], [ %iv.next, %loop ]
+ %gep = getelementptr <{i32, i16}>, ptr %A, i64 %iv
+ call void @use(ptr %gep)
+ %iv.next = add i64 %iv, 1
+ %ec = icmp eq i64 %iv, %wide.trip.count
+ br i1 %ec, label %exit, label %loop
+
+exit:
+ ret void
+}
+
+
define void @udiv4_and_udiv2_mul_1(i1 %c, ptr %A) {
; CHECK-LABEL: 'udiv4_and_udiv2_mul_1'
; CHECK-NEXT: Classifying expressions for: @udiv4_and_udiv2_mul_1
>From a81c0a35a149010ab55ae9a5f4104bb3a46d2c2b Mon Sep 17 00:00:00 2001
From: Florian Hahn <flo at fhahn.com>
Date: Fri, 5 Sep 2025 19:54:59 +0100
Subject: [PATCH 2/2] [SCEV] Generalize (C * A /u C) -> A fold to (C1 * A /u
C2) -> C1/C2 * A.
Generalize fold added in 74ec38fad0a to support multiplying and dividing
by different constants, given they are both powers-of-2 and C1 is a
multiple of C2, checked via their trailing zeros.
https://alive2.llvm.org/ce/z/eqJ2xj
---
llvm/lib/Analysis/ScalarEvolution.cpp | 15 +++++++++------
.../Analysis/ScalarEvolution/mul-udiv-folds.ll | 2 +-
2 files changed, 10 insertions(+), 7 deletions(-)
diff --git a/llvm/lib/Analysis/ScalarEvolution.cpp b/llvm/lib/Analysis/ScalarEvolution.cpp
index bd57d1192eb94..f9fbcb05798da 100644
--- a/llvm/lib/Analysis/ScalarEvolution.cpp
+++ b/llvm/lib/Analysis/ScalarEvolution.cpp
@@ -3216,13 +3216,16 @@ const SCEV *ScalarEvolution::getMulExpr(SmallVectorImpl<const SCEV *> &Ops,
};
}
- // Try to fold (C * D /u C) -> D, if C is a power-of-2 and D is a multiple
- // of C.
+ // Try to fold (C1 * D /u C2) -> C1/C2 * D, if C1 and C2 are powers-of-2,
+ // D is a multiple of C2, and C1 is a multiple of C1.
const SCEV *D;
- if (match(Ops[1], m_scev_UDiv(m_SCEV(D), m_scev_Specific(LHSC))) &&
- LHSC->getAPInt().isPowerOf2() &&
- LHSC->getAPInt().logBase2() <= getMinTrailingZeros(D)) {
- return D;
+ const SCEVConstant *C2;
+ if (LHSC->getAPInt().isPowerOf2() &&
+ match(Ops[1], m_scev_UDiv(m_SCEV(D), m_SCEVConstant(C2))) &&
+ C2->getAPInt().isPowerOf2() &&
+ getMinTrailingZeros(LHSC) >= getMinTrailingZeros(C2) &&
+ getMinTrailingZeros(LHSC) <= getMinTrailingZeros(D)) {
+ return getMulExpr(getUDivExpr(LHSC, C2), D);
}
}
}
diff --git a/llvm/test/Analysis/ScalarEvolution/mul-udiv-folds.ll b/llvm/test/Analysis/ScalarEvolution/mul-udiv-folds.ll
index 684c30d9f2cde..afe69ceb148aa 100644
--- a/llvm/test/Analysis/ScalarEvolution/mul-udiv-folds.ll
+++ b/llvm/test/Analysis/ScalarEvolution/mul-udiv-folds.ll
@@ -19,7 +19,7 @@ define void @udiv4_and_udiv2_mul_4(i1 %c, ptr %A) {
; CHECK-NEXT: %iv = phi i64 [ %iv.start, %entry ], [ %iv.next, %loop ]
; CHECK-NEXT: --> {((zext i32 %start to i64) /u 4),+,1}<%loop> U: full-set S: full-set Exits: ((zext i32 %start to i64) /u 2) LoopDispositions: { %loop: Computable }
; CHECK-NEXT: %gep = getelementptr i32, ptr %A, i64 %iv
-; CHECK-NEXT: --> {((zext i32 %start to i64) + %A),+,4}<%loop> U: full-set S: full-set Exits: ((zext i32 %start to i64) + (4 * ((zext i32 %start to i64) /u 2))<nuw><nsw> + (-4 * ((zext i32 %start to i64) /u 4))<nsw> + %A) LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: --> {((zext i32 %start to i64) + %A),+,4}<%loop> U: full-set S: full-set Exits: ((3 * (zext i32 %start to i64))<nuw><nsw> + (-4 * ((zext i32 %start to i64) /u 4))<nsw> + %A) LoopDispositions: { %loop: Computable }
; CHECK-NEXT: %iv.next = add i64 %iv, 1
; CHECK-NEXT: --> {(1 + ((zext i32 %start to i64) /u 4))<nuw><nsw>,+,1}<%loop> U: full-set S: full-set Exits: (1 + ((zext i32 %start to i64) /u 2))<nuw><nsw> LoopDispositions: { %loop: Computable }
; CHECK-NEXT: Determining loop execution counts for: @udiv4_and_udiv2_mul_4
More information about the llvm-commits
mailing list