[llvm] [InstCombine] Extend (icmp eq/ne (and Z, X), (and Z, Y)) folds (PR #94867)
via llvm-commits
llvm-commits at lists.llvm.org
Sun Jun 9 16:08:04 PDT 2024
https://github.com/goldsteinn updated https://github.com/llvm/llvm-project/pull/94867
>From 4c9f9fa953f864b67ef82fb57eabeb4f88984dec Mon Sep 17 00:00:00 2001
From: Noah Goldstein <goldstein.w.n at gmail.com>
Date: Sat, 8 Jun 2024 14:22:35 -0500
Subject: [PATCH 1/2] [InstCombine] Add tests for combining `(icmp eq/ne (and
X, P2), (and X, -P2))`; NFC
---
.../Transforms/InstCombine/and-compare.ll | 99 +++++++++++++++++++
1 file changed, 99 insertions(+)
diff --git a/llvm/test/Transforms/InstCombine/and-compare.ll b/llvm/test/Transforms/InstCombine/and-compare.ll
index 14379ebf3a905..3f714b3640492 100644
--- a/llvm/test/Transforms/InstCombine/and-compare.ll
+++ b/llvm/test/Transforms/InstCombine/and-compare.ll
@@ -4,6 +4,8 @@
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"
+declare void @use.i8(i8)
+
; Should be optimized to one and.
define i1 @test1(i32 %a, i32 %b) {
; CHECK-LABEL: @test1(
@@ -75,3 +77,100 @@ define <2 x i1> @test3vec(<2 x i64> %A) {
ret <2 x i1> %cmp
}
+define i1 @test_ne_cp2(i8 %x, i8 %yy) {
+; CHECK-LABEL: @test_ne_cp2(
+; CHECK-NEXT: [[AND_X_NEG_Y:%.*]] = and i8 [[X:%.*]], -16
+; CHECK-NEXT: [[AND_X_Y:%.*]] = and i8 [[X]], 16
+; CHECK-NEXT: call void @use.i8(i8 [[AND_X_NEG_Y]])
+; CHECK-NEXT: call void @use.i8(i8 [[AND_X_Y]])
+; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[AND_X_NEG_Y]], [[AND_X_Y]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %and_x_neg_y = and i8 %x, -16
+ %and_x_y = and i8 %x, 16
+ call void @use.i8(i8 %and_x_neg_y)
+ call void @use.i8(i8 %and_x_y)
+ %r = icmp ne i8 %and_x_neg_y, %and_x_y
+ ret i1 %r
+}
+
+define i1 @test_ne_cp2_2(i8 %x, i8 %yy) {
+; CHECK-LABEL: @test_ne_cp2_2(
+; CHECK-NEXT: [[AND_X_NEG_Y:%.*]] = and i8 [[X:%.*]], -4
+; CHECK-NEXT: [[AND_X_Y:%.*]] = and i8 [[X]], 4
+; CHECK-NEXT: call void @use.i8(i8 [[AND_X_NEG_Y]])
+; CHECK-NEXT: call void @use.i8(i8 [[AND_X_Y]])
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[AND_X_Y]], [[AND_X_NEG_Y]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %and_x_neg_y = and i8 %x, -4
+ %and_x_y = and i8 %x, 4
+ call void @use.i8(i8 %and_x_neg_y)
+ call void @use.i8(i8 %and_x_y)
+ %r = icmp eq i8 %and_x_y, %and_x_neg_y
+ ret i1 %r
+}
+
+define i1 @test_ne_cp2_other_okay_all_ones(i8 %x, i8 %yy) {
+; CHECK-LABEL: @test_ne_cp2_other_okay_all_ones(
+; CHECK-NEXT: [[AND_X_NEG_Y:%.*]] = and i8 [[X:%.*]], -17
+; CHECK-NEXT: [[AND_X_Y:%.*]] = and i8 [[X]], 16
+; CHECK-NEXT: call void @use.i8(i8 [[AND_X_NEG_Y]])
+; CHECK-NEXT: call void @use.i8(i8 [[AND_X_Y]])
+; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[AND_X_NEG_Y]], [[AND_X_Y]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %and_x_neg_y = and i8 %x, -17
+ %and_x_y = and i8 %x, 16
+ call void @use.i8(i8 %and_x_neg_y)
+ call void @use.i8(i8 %and_x_y)
+ %r = icmp ne i8 %and_x_neg_y, %and_x_y
+ ret i1 %r
+}
+
+define i1 @test_ne_cp2_other_fail2(i8 %x, i8 %yy) {
+; CHECK-LABEL: @test_ne_cp2_other_fail2(
+; CHECK-NEXT: [[AND_X_NEG_Y:%.*]] = and i8 [[X:%.*]], -16
+; CHECK-NEXT: [[AND_X_Y:%.*]] = and i8 [[X]], 17
+; CHECK-NEXT: call void @use.i8(i8 [[AND_X_NEG_Y]])
+; CHECK-NEXT: call void @use.i8(i8 [[AND_X_Y]])
+; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[AND_X_NEG_Y]], [[AND_X_Y]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %and_x_neg_y = and i8 %x, -16
+ %and_x_y = and i8 %x, 17
+ call void @use.i8(i8 %and_x_neg_y)
+ call void @use.i8(i8 %and_x_y)
+ %r = icmp ne i8 %and_x_neg_y, %and_x_y
+ ret i1 %r
+}
+
+define i1 @test_ne_cp2_other_okay(i8 %x, i8 %yy) {
+; CHECK-LABEL: @test_ne_cp2_other_okay(
+; CHECK-NEXT: [[AND_X_NEG_Y:%.*]] = and i8 [[X:%.*]], -17
+; CHECK-NEXT: [[AND_X_Y:%.*]] = and i8 [[X]], 16
+; CHECK-NEXT: call void @use.i8(i8 [[AND_X_Y]])
+; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[AND_X_NEG_Y]], [[AND_X_Y]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %and_x_neg_y = and i8 %x, -17
+ %and_x_y = and i8 %x, 16
+ call void @use.i8(i8 %and_x_y)
+ %r = icmp ne i8 %and_x_neg_y, %and_x_y
+ ret i1 %r
+}
+
+define i1 @test_ne_cp2_other_okay2(i8 %x, i8 %yy) {
+; CHECK-LABEL: @test_ne_cp2_other_okay2(
+; CHECK-NEXT: [[AND_X_NEG_Y:%.*]] = and i8 [[X:%.*]], -17
+; CHECK-NEXT: [[AND_X_Y:%.*]] = and i8 [[X]], 16
+; CHECK-NEXT: call void @use.i8(i8 [[AND_X_Y]])
+; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[AND_X_Y]], [[AND_X_NEG_Y]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %and_x_neg_y = and i8 %x, -17
+ %and_x_y = and i8 %x, 16
+ call void @use.i8(i8 %and_x_y)
+ %r = icmp ne i8 %and_x_y, %and_x_neg_y
+ ret i1 %r
+}
>From 4d2e4f58721e1de1cdbdfed6ae56496dc8d7c553 Mon Sep 17 00:00:00 2001
From: Noah Goldstein <goldstein.w.n at gmail.com>
Date: Sat, 8 Jun 2024 14:22:37 -0500
Subject: [PATCH 2/2] [InstCombine] Extend `(icmp eq/ne (and Z, X), (and Z,
Y))` folds
Two ways to relaxed:
1) Only require one of the `and` ops to be single-use if both `X`
and `Y` are constant.
2) Special case, if `X ^ Y` is a negative power of 2, then
`Z ^ NegP2 ==/!= 0` will fold to `Z u</u>= -NegP2` which
creates no additional instructions so fold irrelivant of
use counts.
---
.../InstCombine/InstCombineCompares.cpp | 27 ++++++++++++++-----
.../Transforms/InstCombine/and-compare.ll | 16 +++++------
2 files changed, 28 insertions(+), 15 deletions(-)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 4203147bc6a54..da98d1a166b1a 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -5548,8 +5548,8 @@ Instruction *InstCombinerImpl::foldICmpEquality(ICmpInst &I) {
}
// (X&Z) == (Y&Z) -> (X^Y) & Z == 0
- if (match(Op0, m_OneUse(m_And(m_Value(A), m_Value(B)))) &&
- match(Op1, m_OneUse(m_And(m_Value(C), m_Value(D))))) {
+ if (match(Op0, m_And(m_Value(A), m_Value(B))) &&
+ match(Op1, m_And(m_Value(C), m_Value(D)))) {
Value *X = nullptr, *Y = nullptr, *Z = nullptr;
if (A == C) {
@@ -5570,10 +5570,25 @@ Instruction *InstCombinerImpl::foldICmpEquality(ICmpInst &I) {
Z = B;
}
- if (X) { // Build (X^Y) & Z
- Op1 = Builder.CreateXor(X, Y);
- Op1 = Builder.CreateAnd(Op1, Z);
- return new ICmpInst(Pred, Op1, Constant::getNullValue(Op1->getType()));
+ if (X) {
+ // If X^Y is a negative power of two, then `icmp eq/ne (Z & NegP2), 0`
+ // will fold to `icmp ult/uge Z, -NegP2` incuring no additional uses.
+ const APInt *C0, *C1;
+ bool XorIsNegP2 = match(X, m_APInt(C0)) && match(Y, m_APInt(C1)) &&
+ (*C0 ^ *C1).isNegatedPowerOf2();
+
+ // If either Op0/Op1 are both one use or X^Y will constant fold and one of
+ // Op0/Op1 are one use proceeed. In those cases we are instruction neutral
+ // but `icmp eq/ne A, 0` is easier to analyze than `icmp eq/ne A, B`
+ int UseCnt =
+ int(Op0->hasOneUse()) + int(Op1->hasOneUse()) +
+ (int(match(X, m_ImmConstant()) && match(Y, m_ImmConstant())));
+ if (XorIsNegP2 || UseCnt >= 2) {
+ // Build (X^Y) & Z
+ Op1 = Builder.CreateXor(X, Y);
+ Op1 = Builder.CreateAnd(Op1, Z);
+ return new ICmpInst(Pred, Op1, Constant::getNullValue(Op1->getType()));
+ }
}
}
diff --git a/llvm/test/Transforms/InstCombine/and-compare.ll b/llvm/test/Transforms/InstCombine/and-compare.ll
index 3f714b3640492..5a9767a64a2ce 100644
--- a/llvm/test/Transforms/InstCombine/and-compare.ll
+++ b/llvm/test/Transforms/InstCombine/and-compare.ll
@@ -83,7 +83,7 @@ define i1 @test_ne_cp2(i8 %x, i8 %yy) {
; CHECK-NEXT: [[AND_X_Y:%.*]] = and i8 [[X]], 16
; CHECK-NEXT: call void @use.i8(i8 [[AND_X_NEG_Y]])
; CHECK-NEXT: call void @use.i8(i8 [[AND_X_Y]])
-; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[AND_X_NEG_Y]], [[AND_X_Y]]
+; CHECK-NEXT: [[R:%.*]] = icmp ugt i8 [[X]], 31
; CHECK-NEXT: ret i1 [[R]]
;
%and_x_neg_y = and i8 %x, -16
@@ -100,7 +100,7 @@ define i1 @test_ne_cp2_2(i8 %x, i8 %yy) {
; CHECK-NEXT: [[AND_X_Y:%.*]] = and i8 [[X]], 4
; CHECK-NEXT: call void @use.i8(i8 [[AND_X_NEG_Y]])
; CHECK-NEXT: call void @use.i8(i8 [[AND_X_Y]])
-; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[AND_X_Y]], [[AND_X_NEG_Y]]
+; CHECK-NEXT: [[R:%.*]] = icmp ult i8 [[X]], 8
; CHECK-NEXT: ret i1 [[R]]
;
%and_x_neg_y = and i8 %x, -4
@@ -117,7 +117,7 @@ define i1 @test_ne_cp2_other_okay_all_ones(i8 %x, i8 %yy) {
; CHECK-NEXT: [[AND_X_Y:%.*]] = and i8 [[X]], 16
; CHECK-NEXT: call void @use.i8(i8 [[AND_X_NEG_Y]])
; CHECK-NEXT: call void @use.i8(i8 [[AND_X_Y]])
-; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[AND_X_NEG_Y]], [[AND_X_Y]]
+; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[X]], 0
; CHECK-NEXT: ret i1 [[R]]
;
%and_x_neg_y = and i8 %x, -17
@@ -147,10 +147,9 @@ define i1 @test_ne_cp2_other_fail2(i8 %x, i8 %yy) {
define i1 @test_ne_cp2_other_okay(i8 %x, i8 %yy) {
; CHECK-LABEL: @test_ne_cp2_other_okay(
-; CHECK-NEXT: [[AND_X_NEG_Y:%.*]] = and i8 [[X:%.*]], -17
-; CHECK-NEXT: [[AND_X_Y:%.*]] = and i8 [[X]], 16
+; CHECK-NEXT: [[AND_X_Y:%.*]] = and i8 [[X:%.*]], 16
; CHECK-NEXT: call void @use.i8(i8 [[AND_X_Y]])
-; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[AND_X_NEG_Y]], [[AND_X_Y]]
+; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[X]], 0
; CHECK-NEXT: ret i1 [[R]]
;
%and_x_neg_y = and i8 %x, -17
@@ -162,10 +161,9 @@ define i1 @test_ne_cp2_other_okay(i8 %x, i8 %yy) {
define i1 @test_ne_cp2_other_okay2(i8 %x, i8 %yy) {
; CHECK-LABEL: @test_ne_cp2_other_okay2(
-; CHECK-NEXT: [[AND_X_NEG_Y:%.*]] = and i8 [[X:%.*]], -17
-; CHECK-NEXT: [[AND_X_Y:%.*]] = and i8 [[X]], 16
+; CHECK-NEXT: [[AND_X_Y:%.*]] = and i8 [[X:%.*]], 16
; CHECK-NEXT: call void @use.i8(i8 [[AND_X_Y]])
-; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[AND_X_Y]], [[AND_X_NEG_Y]]
+; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[X]], 0
; CHECK-NEXT: ret i1 [[R]]
;
%and_x_neg_y = and i8 %x, -17
More information about the llvm-commits
mailing list