[InstCombine] Generalize trunc-shift-icmp fold from (1 << Y) to (Pow2 << Y) #169163
Conversation
This file contains hidden or 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
… << Y) Extends the `icmp(trunc(shl))` fold to handle any power of 2 constant as the shift base, not just 1. This generalizes the following patterns by adjusting the comparison offsets by `log2(Pow2)`. ```llvm (trunc (1 << Y) to iN) == 0 --> Y u>= N (trunc (1 << Y) to iN) != 0 --> Y u< N (trunc (1 << Y) to iN) == 2**C --> Y == C (trunc (1 << Y) to iN) != 2**C --> Y != C ; to (trunc (Pow2 << Y) to iN) == 0 --> Y u>= N - log2(Pow2) (trunc (Pow2 << Y) to iN) != 0 --> Y u< N - log2(Pow2) (trunc (Pow2 << Y) to iN) == 2**C --> Y == C - log2(Pow2) (trunc (Pow2 << Y) to iN) != 2**C --> Y != C - log2(Pow2) ``` Proof: https://alive2.llvm.org/ce/z/2zwTkp
llvmbot
added
llvm:instcombine
Member
|
@llvm/pr-subscribers-llvm-transforms Author: Pedro Lobo (pedroclobo) ChangesExtends the (trunc (1 << Y) to iN) == 0 --> Y u>= N
(trunc (1 << Y) to iN) != 0 --> Y u< N
(trunc (1 << Y) to iN) == 2**C --> Y == C
(trunc (1 << Y) to iN) != 2**C --> Y != C
; to
(trunc (Pow2 << Y) to iN) == 0 --> Y u>= N - log2(Pow2)
(trunc (Pow2 << Y) to iN) != 0 --> Y u< N - log2(Pow2)
(trunc (Pow2 << Y) to iN) == 2**C --> Y == C - log2(Pow2)
(trunc (Pow2 << Y) to iN) != 2**C --> Y != C - log2(Pow2)Proof: https://alive2.llvm.org/ce/z/2zwTkp Full diff: https://github.com/llvm/llvm-project/pull/169163.diff 2 Files Affected:
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index f153db177cac1..cf6e7315114dc 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -1465,20 +1465,24 @@ Instruction *InstCombinerImpl::foldICmpTruncConstant(ICmpInst &Cmp,
ConstantInt::get(V->getType(), 1));
}
- // TODO: Handle any shifted constant by subtracting trailing zeros.
// TODO: Handle non-equality predicates.
Value *Y;
- if (Cmp.isEquality() && match(X, m_Shl(m_One(), m_Value(Y)))) {
- // (trunc (1 << Y) to iN) == 0 --> Y u>= N
- // (trunc (1 << Y) to iN) != 0 --> Y u< N
+ const APInt *Pow2;
+ if (Cmp.isEquality() && match(X, m_Shl(m_Power2(Pow2), m_Value(Y))) &&
+ DstBits > Pow2->logBase2()) {
+ // (trunc (Pow2 << Y) to iN) == 0 --> Y u>= N - log2(Pow2)
+ // (trunc (Pow2 << Y) to iN) != 0 --> Y u< N - log2(Pow2)
+ // iff N > log2(Pow2)
if (C.isZero()) {
auto NewPred = (Pred == Cmp.ICMP_EQ) ? Cmp.ICMP_UGE : Cmp.ICMP_ULT;
- return new ICmpInst(NewPred, Y, ConstantInt::get(SrcTy, DstBits));
+ return new ICmpInst(NewPred, Y,
+ ConstantInt::get(SrcTy, DstBits - Pow2->logBase2()));
}
- // (trunc (1 << Y) to iN) == 2**C --> Y == C
- // (trunc (1 << Y) to iN) != 2**C --> Y != C
+ // (trunc (Pow2 << Y) to iN) == 2**C --> Y == C - log2(Pow2)
+ // (trunc (Pow2 << Y) to iN) != 2**C --> Y != C - log2(Pow2)
if (C.isPowerOf2())
- return new ICmpInst(Pred, Y, ConstantInt::get(SrcTy, C.logBase2()));
+ return new ICmpInst(
+ Pred, Y, ConstantInt::get(SrcTy, C.logBase2() - Pow2->logBase2()));
}
if (Cmp.isEquality() && (Trunc->hasOneUse() || Trunc->hasNoUnsignedWrap())) {
diff --git a/llvm/test/Transforms/InstCombine/icmp-trunc.ll b/llvm/test/Transforms/InstCombine/icmp-trunc.ll
index ad76ef7329b0a..6e7b63cfd9589 100644
--- a/llvm/test/Transforms/InstCombine/icmp-trunc.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-trunc.ll
@@ -407,13 +407,9 @@ define i1 @shl1_trunc_ne0_use2(i37 %a) {
ret i1 %r
}
-; TODO: A > 4
-
define i1 @shl2_trunc_eq0(i9 %a) {
; CHECK-LABEL: @shl2_trunc_eq0(
-; CHECK-NEXT: [[SHL:%.*]] = shl i9 2, [[A:%.*]]
-; CHECK-NEXT: [[T:%.*]] = trunc i9 [[SHL]] to i6
-; CHECK-NEXT: [[R:%.*]] = icmp eq i6 [[T]], 0
+; CHECK-NEXT: [[R:%.*]] = icmp ugt i9 [[A:%.*]], 4
; CHECK-NEXT: ret i1 [[R]]
;
%shl = shl i9 2, %a
@@ -424,9 +420,7 @@ define i1 @shl2_trunc_eq0(i9 %a) {
define i1 @shl2_trunc_ne0(i9 %a) {
; CHECK-LABEL: @shl2_trunc_ne0(
-; CHECK-NEXT: [[SHL:%.*]] = shl i9 2, [[A:%.*]]
-; CHECK-NEXT: [[T:%.*]] = trunc i9 [[SHL]] to i6
-; CHECK-NEXT: [[R:%.*]] = icmp ne i6 [[T]], 0
+; CHECK-NEXT: [[R:%.*]] = icmp ult i9 [[A:%.*]], 5
; CHECK-NEXT: ret i1 [[R]]
;
%shl = shl i9 2, %a
@@ -450,9 +444,7 @@ define i1 @shl3_trunc_eq0(i9 %a) {
define <2 x i1> @shl4_trunc_ne0(<2 x i8> %a) {
; CHECK-LABEL: @shl4_trunc_ne0(
-; CHECK-NEXT: [[SHL:%.*]] = shl <2 x i8> <i8 4, i8 poison>, [[A:%.*]]
-; CHECK-NEXT: [[T:%.*]] = trunc <2 x i8> [[SHL]] to <2 x i5>
-; CHECK-NEXT: [[R:%.*]] = icmp ne <2 x i5> [[T]], zeroinitializer
+; CHECK-NEXT: [[R:%.*]] = icmp ult <2 x i8> [[A:%.*]], splat (i8 3)
; CHECK-NEXT: ret <2 x i1> [[R]]
;
%shl = shl <2 x i8> <i8 4, i8 poison>, %a
@@ -461,6 +453,16 @@ define <2 x i1> @shl4_trunc_ne0(<2 x i8> %a) {
ret <2 x i1> %r
}
+define i1 @shl5_trunc_ne0(i9 %a) {
+; CHECK-LABEL: @shl5_trunc_ne0(
+; CHECK-NEXT: [[R:%.*]] = icmp ult i9 [[A:%.*]], 4
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %shl = shl i9 4, %a
+ %t = trunc i9 %shl to i6
+ %r = icmp ne i6 %t, 0
+ ret i1 %r
+}
; TODO: A < 5
@@ -507,17 +509,9 @@ define i1 @shl1_trunc_ne32(i8 %a) {
}
define i1 @shl2_trunc_eq8_i32(i32 %a) {
-; DL64-LABEL: @shl2_trunc_eq8_i32(
-; DL64-NEXT: [[SHL:%.*]] = shl i32 2, [[A:%.*]]
-; DL64-NEXT: [[TMP1:%.*]] = and i32 [[SHL]], 65534
-; DL64-NEXT: [[R:%.*]] = icmp eq i32 [[TMP1]], 8
-; DL64-NEXT: ret i1 [[R]]
-;
-; DL8-LABEL: @shl2_trunc_eq8_i32(
-; DL8-NEXT: [[SHL:%.*]] = shl i32 2, [[A:%.*]]
-; DL8-NEXT: [[T:%.*]] = trunc i32 [[SHL]] to i16
-; DL8-NEXT: [[R:%.*]] = icmp eq i16 [[T]], 8
-; DL8-NEXT: ret i1 [[R]]
+; CHECK-LABEL: @shl2_trunc_eq8_i32(
+; CHECK-NEXT: [[R:%.*]] = icmp eq i32 [[A:%.*]], 2
+; CHECK-NEXT: ret i1 [[R]]
;
%shl = shl i32 2, %a
%t = trunc i32 %shl to i16
@@ -526,17 +520,9 @@ define i1 @shl2_trunc_eq8_i32(i32 %a) {
}
define i1 @shl2_trunc_ne8_i32(i32 %a) {
-; DL64-LABEL: @shl2_trunc_ne8_i32(
-; DL64-NEXT: [[SHL:%.*]] = shl i32 2, [[A:%.*]]
-; DL64-NEXT: [[TMP1:%.*]] = and i32 [[SHL]], 65534
-; DL64-NEXT: [[R:%.*]] = icmp ne i32 [[TMP1]], 8
-; DL64-NEXT: ret i1 [[R]]
-;
-; DL8-LABEL: @shl2_trunc_ne8_i32(
-; DL8-NEXT: [[SHL:%.*]] = shl i32 2, [[A:%.*]]
-; DL8-NEXT: [[T:%.*]] = trunc i32 [[SHL]] to i16
-; DL8-NEXT: [[R:%.*]] = icmp ne i16 [[T]], 8
-; DL8-NEXT: ret i1 [[R]]
+; CHECK-LABEL: @shl2_trunc_ne8_i32(
+; CHECK-NEXT: [[R:%.*]] = icmp ne i32 [[A:%.*]], 2
+; CHECK-NEXT: ret i1 [[R]]
;
%shl = shl i32 2, %a
%t = trunc i32 %shl to i16
@@ -544,6 +530,26 @@ define i1 @shl2_trunc_ne8_i32(i32 %a) {
ret i1 %r
}
+define i1 @neg_shl2_trunc_eq0_i8(i8 %a) {
+; CHECK-LABEL: @neg_shl2_trunc_eq0_i8(
+; CHECK-NEXT: ret i1 true
+;
+ %shl = shl i8 256, %a
+ %t = trunc i8 %shl to i6
+ %r = icmp eq i6 %t, 0
+ ret i1 %r
+}
+
+define i1 @neg_shl2_trunc_ne0_i8(i8 %a) {
+; CHECK-LABEL: @neg_shl2_trunc_ne0_i8(
+; CHECK-NEXT: ret i1 false
+;
+ %shl = shl i8 128, %a
+ %t = trunc i8 %shl to i6
+ %r = icmp ne i6 %t, 0
+ ret i1 %r
+}
+
define i1 @shl1_trunc_sgt4(i32 %a) {
; CHECK-LABEL: @shl1_trunc_sgt4(
; CHECK-NEXT: [[SHL:%.*]] = shl nuw i32 1, [[A:%.*]]
|
Collaborator
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/52/builds/13052 Here is the relevant piece of the build log for the reference |
Collaborator
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/14/builds/4724 Here is the relevant piece of the build log for the reference |
aadeshps-mcw
pushed a commit
to aadeshps-mcw/llvm-project
that referenced
this pull request
Nov 26, 2025
… << Y) (llvm#169163) Extends the `icmp(trunc(shl))` fold to handle any power of 2 constant as the shift base, not just 1. This generalizes the following patterns by adjusting the comparison offsets by `log2(Pow2)`. ```llvm (trunc (1 << Y) to iN) == 0 --> Y u>= N (trunc (1 << Y) to iN) != 0 --> Y u< N (trunc (1 << Y) to iN) == 2**C --> Y == C (trunc (1 << Y) to iN) != 2**C --> Y != C ; to (trunc (Pow2 << Y) to iN) == 0 --> Y u>= N - log2(Pow2) (trunc (Pow2 << Y) to iN) != 0 --> Y u< N - log2(Pow2) (trunc (Pow2 << Y) to iN) == 2**C --> Y == C - log2(Pow2) (trunc (Pow2 << Y) to iN) != 2**C --> Y != C - log2(Pow2) ``` Proof: https://alive2.llvm.org/ce/z/2zwTkp
Priyanshu3820
pushed a commit
to Priyanshu3820/llvm-project
that referenced
this pull request
Nov 26, 2025
… << Y) (llvm#169163) Extends the `icmp(trunc(shl))` fold to handle any power of 2 constant as the shift base, not just 1. This generalizes the following patterns by adjusting the comparison offsets by `log2(Pow2)`. ```llvm (trunc (1 << Y) to iN) == 0 --> Y u>= N (trunc (1 << Y) to iN) != 0 --> Y u< N (trunc (1 << Y) to iN) == 2**C --> Y == C (trunc (1 << Y) to iN) != 2**C --> Y != C ; to (trunc (Pow2 << Y) to iN) == 0 --> Y u>= N - log2(Pow2) (trunc (Pow2 << Y) to iN) != 0 --> Y u< N - log2(Pow2) (trunc (Pow2 << Y) to iN) == 2**C --> Y == C - log2(Pow2) (trunc (Pow2 << Y) to iN) != 2**C --> Y != C - log2(Pow2) ``` Proof: https://alive2.llvm.org/ce/z/2zwTkp
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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.
Extends the
icmp(trunc(shl))fold to handle any power of 2 constant as theshift base, not just 1. This generalizes the following patterns by
adjusting the comparison offsets by
log2(Pow2).Proof: https://alive2.llvm.org/ce/z/2zwTkp