[llvm] [InstCombine] Improve `(icmp pred (and X, Y), ...)` fold. (PR #66787)
via llvm-commits
llvm-commits at lists.llvm.org
Sat Oct 21 22:30:48 PDT 2023
https://github.com/goldsteinn updated https://github.com/llvm/llvm-project/pull/66787
>From 62e7e3aa46eac5913d0de7e1fea65d035ca63a1e Mon Sep 17 00:00:00 2001
From: Noah Goldstein <goldstein.w.n at gmail.com>
Date: Wed, 13 Sep 2023 13:45:47 -0500
Subject: [PATCH 01/13] [InstCombine] Make `isFreeToInvert` check recursively.
Some Instructions (select/min/max) are inverted by just inverting the
operands. So the answer of whether they are free to invert is really
just whether the operands are free to invert.
Differential Revision: https://reviews.llvm.org/D159056
---
.../Transforms/InstCombine/InstCombiner.h | 55 +++++++++++--------
.../InstCombine/minmax-intrinsics.ll | 13 ++---
llvm/test/Transforms/InstCombine/pr63791.ll | 2 +-
.../LoopVectorize/reduction-inloop.ll | 25 +++++----
4 files changed, 51 insertions(+), 44 deletions(-)
diff --git a/llvm/include/llvm/Transforms/InstCombine/InstCombiner.h b/llvm/include/llvm/Transforms/InstCombine/InstCombiner.h
index f8b3874267ded3b..3a1ac917da6a6cb 100644
--- a/llvm/include/llvm/Transforms/InstCombine/InstCombiner.h
+++ b/llvm/include/llvm/Transforms/InstCombine/InstCombiner.h
@@ -239,7 +239,8 @@ class LLVM_LIBRARY_VISIBILITY InstCombiner {
/// uses of V and only keep uses of ~V.
///
/// See also: canFreelyInvertAllUsersOf()
- static bool isFreeToInvert(Value *V, bool WillInvertAllUses) {
+ static bool isFreeToInvert(Value *V, bool WillInvertAllUses,
+ unsigned Depth = 0) {
// ~(~(X)) -> X.
if (match(V, m_Not(PatternMatch::m_Value())))
return true;
@@ -248,32 +249,38 @@ class LLVM_LIBRARY_VISIBILITY InstCombiner {
if (match(V, PatternMatch::m_AnyIntegralConstant()))
return true;
+ if (Depth++ >= MaxAnalysisRecursionDepth)
+ return false;
+
+ // The rest of the cases require that we invert all uses so don't bother
+ // doing the analysis if we know we can't use the result.
+ if (!WillInvertAllUses)
+ return false;
+
// Compares can be inverted if all of their uses are being modified to use
// the ~V.
if (isa<CmpInst>(V))
- return WillInvertAllUses;
-
- // If `V` is of the form `A + Constant` then `-1 - V` can be folded into
- // `(-1 - Constant) - A` if we are willing to invert all of the uses.
- if (match(V, m_Add(PatternMatch::m_Value(), PatternMatch::m_ImmConstant())))
- return WillInvertAllUses;
-
- // If `V` is of the form `Constant - A` then `-1 - V` can be folded into
- // `A + (-1 - Constant)` if we are willing to invert all of the uses.
- if (match(V, m_Sub(PatternMatch::m_ImmConstant(), PatternMatch::m_Value())))
- return WillInvertAllUses;
-
- // Selects with invertible operands are freely invertible
- if (match(V,
- m_Select(PatternMatch::m_Value(), m_Not(PatternMatch::m_Value()),
- m_Not(PatternMatch::m_Value()))))
- return WillInvertAllUses;
-
- // Min/max may be in the form of intrinsics, so handle those identically
- // to select patterns.
- if (match(V, m_MaxOrMin(m_Not(PatternMatch::m_Value()),
- m_Not(PatternMatch::m_Value()))))
- return WillInvertAllUses;
+ return true;
+
+ Value *A, *B;
+ // If `V` is of the form `A + B` then `-1 - V` can be folded into
+ // `~B - A` or `~A - B` if we are willing to invert all of the uses.
+ if (match(V, m_Add(PatternMatch::m_Value(A), PatternMatch::m_Value(B))))
+ return isFreeToInvert(A, A->hasOneUse(), Depth) ||
+ isFreeToInvert(B, B->hasOneUse(), Depth);
+
+ // If `V` is of the form `A - B` then `-1 - V` can be folded into
+ // `~A + B` if we are willing to invert all of the uses.
+ if (match(V, m_Sub(PatternMatch::m_Value(A), PatternMatch::m_Value())))
+ return isFreeToInvert(A, A->hasOneUse(), Depth);
+
+ // Selects/min/max with invertible operands are freely invertible
+ if (match(V, m_Select(PatternMatch::m_Value(), PatternMatch::m_Value(A),
+ PatternMatch::m_Value(B))) ||
+ match(V,
+ m_MaxOrMin(PatternMatch::m_Value(A), PatternMatch::m_Value(B))))
+ return isFreeToInvert(A, A->hasOneUse(), Depth) &&
+ isFreeToInvert(B, B->hasOneUse(), Depth);
return false;
}
diff --git a/llvm/test/Transforms/InstCombine/minmax-intrinsics.ll b/llvm/test/Transforms/InstCombine/minmax-intrinsics.ll
index 7a4da66ae2151c8..9ed25b358c887f1 100644
--- a/llvm/test/Transforms/InstCombine/minmax-intrinsics.ll
+++ b/llvm/test/Transforms/InstCombine/minmax-intrinsics.ll
@@ -1494,14 +1494,13 @@ define i8 @freeToInvert_two_minmax_ops_use3(i8 %x, i8 %y, i8 %z, i8 %w) {
define i8 @sub_not_min_max(i8 %r, i8 %g, i8 %b) {
; CHECK-LABEL: @sub_not_min_max(
-; CHECK-NEXT: [[NOTR:%.*]] = xor i8 [[R:%.*]], -1
; CHECK-NEXT: [[NOTG:%.*]] = xor i8 [[G:%.*]], -1
; CHECK-NEXT: call void @use(i8 [[NOTG]])
; CHECK-NEXT: [[NOTB:%.*]] = xor i8 [[B:%.*]], -1
; CHECK-NEXT: call void @use(i8 [[NOTB]])
-; CHECK-NEXT: [[M:%.*]] = call i8 @llvm.smin.i8(i8 [[NOTR]], i8 [[NOTG]])
-; CHECK-NEXT: [[K:%.*]] = call i8 @llvm.smin.i8(i8 [[M]], i8 [[NOTB]])
-; CHECK-NEXT: [[CK:%.*]] = sub i8 [[NOTR]], [[K]]
+; CHECK-NEXT: [[TMP1:%.*]] = call i8 @llvm.smax.i8(i8 [[R:%.*]], i8 [[G]])
+; CHECK-NEXT: [[TMP2:%.*]] = call i8 @llvm.smax.i8(i8 [[TMP1]], i8 [[B]])
+; CHECK-NEXT: [[CK:%.*]] = sub i8 [[TMP2]], [[R]]
; CHECK-NEXT: ret i8 [[CK]]
;
%notr = xor i8 %r, -1
@@ -1523,9 +1522,9 @@ define i8 @sub_not_min_max_uses1(i8 %r, i8 %g, i8 %b) {
; CHECK-NEXT: call void @use(i8 [[NOTG]])
; CHECK-NEXT: [[NOTB:%.*]] = xor i8 [[B:%.*]], -1
; CHECK-NEXT: call void @use(i8 [[NOTB]])
-; CHECK-NEXT: [[M:%.*]] = call i8 @llvm.smin.i8(i8 [[NOTR]], i8 [[NOTG]])
-; CHECK-NEXT: [[K:%.*]] = call i8 @llvm.smin.i8(i8 [[M]], i8 [[NOTB]])
-; CHECK-NEXT: [[CK:%.*]] = sub i8 [[NOTR]], [[K]]
+; CHECK-NEXT: [[TMP1:%.*]] = call i8 @llvm.smax.i8(i8 [[R]], i8 [[G]])
+; CHECK-NEXT: [[TMP2:%.*]] = call i8 @llvm.smax.i8(i8 [[TMP1]], i8 [[B]])
+; CHECK-NEXT: [[CK:%.*]] = sub i8 [[TMP2]], [[R]]
; CHECK-NEXT: ret i8 [[CK]]
;
%notr = xor i8 %r, -1
diff --git a/llvm/test/Transforms/InstCombine/pr63791.ll b/llvm/test/Transforms/InstCombine/pr63791.ll
index a489b2e3e6221a0..0fac19d583fe96c 100644
--- a/llvm/test/Transforms/InstCombine/pr63791.ll
+++ b/llvm/test/Transforms/InstCombine/pr63791.ll
@@ -17,7 +17,7 @@ define void @y() {
; CHECK: for.cond5.preheader.i:
; CHECK-NEXT: br i1 false, label [[FOR_INC19_I:%.*]], label [[FOR_COND1_LOOPEXIT_I:%.*]]
; CHECK: for.inc19.i:
-; CHECK-NEXT: br i1 false, label [[FOR_INC19_I]], label [[FOR_COND1_LOOPEXIT_I]]
+; CHECK-NEXT: br i1 true, label [[FOR_COND1_LOOPEXIT_I]], label [[FOR_INC19_I]]
;
entry:
br label %for.cond.i
diff --git a/llvm/test/Transforms/LoopVectorize/reduction-inloop.ll b/llvm/test/Transforms/LoopVectorize/reduction-inloop.ll
index 18b05c05d9b9d21..305da23709fa5e7 100644
--- a/llvm/test/Transforms/LoopVectorize/reduction-inloop.ll
+++ b/llvm/test/Transforms/LoopVectorize/reduction-inloop.ll
@@ -1292,7 +1292,7 @@ define i32 @predicated_or_dominates_reduction(ptr %b) {
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_LOAD_CONTINUE6:%.*]] ]
-; CHECK-NEXT: [[VEC_PHI:%.*]] = phi i32 [ undef, [[VECTOR_PH]] ], [ [[TMP51:%.*]], [[PRED_LOAD_CONTINUE6]] ]
+; CHECK-NEXT: [[VEC_PHI:%.*]] = phi i32 [ undef, [[VECTOR_PH]] ], [ [[TMP49:%.*]], [[PRED_LOAD_CONTINUE6]] ]
; CHECK-NEXT: [[TMP0:%.*]] = or i32 [[INDEX]], 1
; CHECK-NEXT: [[TMP1:%.*]] = or i32 [[INDEX]], 2
; CHECK-NEXT: [[TMP2:%.*]] = or i32 [[INDEX]], 3
@@ -1353,23 +1353,24 @@ define i32 @predicated_or_dominates_reduction(ptr %b) {
; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE6]]
; CHECK: pred.load.continue6:
; CHECK-NEXT: [[TMP43:%.*]] = phi <4 x i32> [ [[TMP37]], [[PRED_LOAD_CONTINUE4]] ], [ [[TMP42]], [[PRED_LOAD_IF5]] ]
-; CHECK-NEXT: [[TMP44:%.*]] = icmp ne <4 x i32> [[TMP43]], zeroinitializer
-; CHECK-NEXT: [[TMP45:%.*]] = select <4 x i1> [[TMP19]], <4 x i1> [[TMP44]], <4 x i1> zeroinitializer
-; CHECK-NEXT: [[TMP46:%.*]] = xor <4 x i1> [[TMP19]], <i1 true, i1 true, i1 true, i1 true>
-; CHECK-NEXT: [[TMP47:%.*]] = or <4 x i1> [[TMP45]], [[TMP46]]
-; CHECK-NEXT: [[TMP48:%.*]] = bitcast <4 x i1> [[TMP47]] to i4
-; CHECK-NEXT: [[TMP49:%.*]] = call i4 @llvm.ctpop.i4(i4 [[TMP48]]), !range [[RNG42:![0-9]+]]
-; CHECK-NEXT: [[TMP50:%.*]] = zext i4 [[TMP49]] to i32
-; CHECK-NEXT: [[TMP51]] = add i32 [[VEC_PHI]], [[TMP50]]
+; CHECK-NEXT: [[TMP44:%.*]] = icmp eq <4 x i32> [[TMP43]], zeroinitializer
+; CHECK-NEXT: [[NOT_:%.*]] = xor <4 x i1> [[TMP19]], <i1 true, i1 true, i1 true, i1 true>
+; CHECK-NEXT: [[TMP45:%.*]] = select <4 x i1> [[NOT_]], <4 x i1> <i1 true, i1 true, i1 true, i1 true>, <4 x i1> [[TMP44]]
+; CHECK-NEXT: [[DOTNOT7:%.*]] = and <4 x i1> [[TMP45]], [[TMP19]]
+; CHECK-NEXT: [[NOT__NOT7:%.*]] = xor <4 x i1> [[DOTNOT7]], <i1 true, i1 true, i1 true, i1 true>
+; CHECK-NEXT: [[TMP46:%.*]] = bitcast <4 x i1> [[NOT__NOT7]] to i4
+; CHECK-NEXT: [[TMP47:%.*]] = call i4 @llvm.ctpop.i4(i4 [[TMP46]]), !range [[RNG42:![0-9]+]]
+; CHECK-NEXT: [[TMP48:%.*]] = zext i4 [[TMP47]] to i32
+; CHECK-NEXT: [[TMP49]] = add i32 [[VEC_PHI]], [[TMP48]]
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
-; CHECK-NEXT: [[TMP52:%.*]] = icmp eq i32 [[INDEX_NEXT]], 1000
-; CHECK-NEXT: br i1 [[TMP52]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP43:![0-9]+]]
+; CHECK-NEXT: [[TMP50:%.*]] = icmp eq i32 [[INDEX_NEXT]], 1000
+; CHECK-NEXT: br i1 [[TMP50]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP43:![0-9]+]]
; CHECK: middle.block:
; CHECK-NEXT: br i1 true, label [[FOR_COND_CLEANUP:%.*]], label [[SCALAR_PH]]
; CHECK: scalar.ph:
; CHECK-NEXT: br label [[FOR_BODY:%.*]]
; CHECK: for.cond.cleanup:
-; CHECK-NEXT: [[A_1_LCSSA:%.*]] = phi i32 [ poison, [[FOR_INC:%.*]] ], [ [[TMP51]], [[MIDDLE_BLOCK]] ]
+; CHECK-NEXT: [[A_1_LCSSA:%.*]] = phi i32 [ poison, [[FOR_INC:%.*]] ], [ [[TMP49]], [[MIDDLE_BLOCK]] ]
; CHECK-NEXT: ret i32 [[A_1_LCSSA]]
; CHECK: for.body:
; CHECK-NEXT: br i1 poison, label [[LOR_LHS_FALSE:%.*]], label [[IF_THEN:%.*]]
>From e7bd0e174eb53df002c9f910390d310c99c0612e Mon Sep 17 00:00:00 2001
From: Noah Goldstein <goldstein.w.n at gmail.com>
Date: Mon, 18 Sep 2023 14:59:58 -0500
Subject: [PATCH 02/13] [InstCombine] Add additional tests for free inversion;
NFC
---
.../Transforms/InstCombine/free-inversion.ll | 403 ++++++++++++++++++
1 file changed, 403 insertions(+)
create mode 100644 llvm/test/Transforms/InstCombine/free-inversion.ll
diff --git a/llvm/test/Transforms/InstCombine/free-inversion.ll b/llvm/test/Transforms/InstCombine/free-inversion.ll
new file mode 100644
index 000000000000000..a2931fb9a081109
--- /dev/null
+++ b/llvm/test/Transforms/InstCombine/free-inversion.ll
@@ -0,0 +1,403 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -passes=instcombine -S | FileCheck %s
+
+declare i8 @llvm.smin.i8(i8, i8)
+declare i8 @llvm.umin.i8(i8, i8)
+declare i8 @llvm.smax.i8(i8, i8)
+declare i8 @llvm.umax.i8(i8, i8)
+
+declare void @use.i8(i8)
+
+define i8 @xor_1(i8 %a, i1 %c, i8 %x, i8 %y) {
+; CHECK-LABEL: @xor_1(
+; CHECK-NEXT: [[NX:%.*]] = xor i8 [[X:%.*]], -1
+; CHECK-NEXT: [[YY:%.*]] = xor i8 [[Y:%.*]], 123
+; CHECK-NEXT: [[B:%.*]] = select i1 [[C:%.*]], i8 [[NX]], i8 [[YY]]
+; CHECK-NEXT: [[BA:%.*]] = xor i8 [[B]], [[A:%.*]]
+; CHECK-NEXT: [[NOT_BA:%.*]] = xor i8 [[BA]], -1
+; CHECK-NEXT: ret i8 [[NOT_BA]]
+;
+ %nx = xor i8 %x, -1
+ %yy = xor i8 %y, 123
+ %b = select i1 %c, i8 %nx, i8 %yy
+ %ba = xor i8 %b, %a
+ %not_ba = xor i8 %ba, -1
+ ret i8 %not_ba
+}
+
+define i8 @xor_2(i8 %a, i1 %c, i8 %x, i8 %y) {
+; CHECK-LABEL: @xor_2(
+; CHECK-NEXT: [[NX:%.*]] = xor i8 [[X:%.*]], -1
+; CHECK-NEXT: [[YY:%.*]] = xor i8 [[Y:%.*]], 123
+; CHECK-NEXT: [[B:%.*]] = select i1 [[C:%.*]], i8 [[NX]], i8 [[YY]]
+; CHECK-NEXT: [[AB:%.*]] = xor i8 [[B]], [[A:%.*]]
+; CHECK-NEXT: [[NOT_AB:%.*]] = xor i8 [[AB]], -1
+; CHECK-NEXT: ret i8 [[NOT_AB]]
+;
+ %nx = xor i8 %x, -1
+ %yy = xor i8 %y, 123
+ %b = select i1 %c, i8 %nx, i8 %yy
+ %ab = xor i8 %a, %b
+ %not_ab = xor i8 %ab, -1
+ ret i8 %not_ab
+}
+
+define i8 @xor_fail(i8 %a, i1 %c, i8 %x, i8 %y) {
+; CHECK-LABEL: @xor_fail(
+; CHECK-NEXT: [[NX:%.*]] = xor i8 [[X:%.*]], -1
+; CHECK-NEXT: [[B:%.*]] = select i1 [[C:%.*]], i8 [[NX]], i8 [[Y:%.*]]
+; CHECK-NEXT: [[AB:%.*]] = xor i8 [[B]], [[A:%.*]]
+; CHECK-NEXT: [[NOT_AB:%.*]] = xor i8 [[AB]], -1
+; CHECK-NEXT: ret i8 [[NOT_AB]]
+;
+ %nx = xor i8 %x, -1
+ %b = select i1 %c, i8 %nx, i8 %y
+ %ab = xor i8 %a, %b
+ %not_ab = xor i8 %ab, -1
+ ret i8 %not_ab
+}
+
+define i8 @add_1(i8 %a, i1 %c, i8 %x, i8 %y) {
+; CHECK-LABEL: @add_1(
+; CHECK-NEXT: [[NX:%.*]] = xor i8 [[X:%.*]], -1
+; CHECK-NEXT: [[YY:%.*]] = xor i8 [[Y:%.*]], 123
+; CHECK-NEXT: [[B:%.*]] = select i1 [[C:%.*]], i8 [[NX]], i8 [[YY]]
+; CHECK-NEXT: [[BA:%.*]] = add i8 [[B]], [[A:%.*]]
+; CHECK-NEXT: [[NOT_BA:%.*]] = xor i8 [[BA]], -1
+; CHECK-NEXT: ret i8 [[NOT_BA]]
+;
+ %nx = xor i8 %x, -1
+ %yy = xor i8 %y, 123
+ %b = select i1 %c, i8 %nx, i8 %yy
+ %ba = add i8 %b, %a
+ %not_ba = xor i8 %ba, -1
+ ret i8 %not_ba
+}
+
+define i8 @add_2(i8 %a, i1 %c, i8 %x, i8 %y) {
+; CHECK-LABEL: @add_2(
+; CHECK-NEXT: [[NX:%.*]] = xor i8 [[X:%.*]], -1
+; CHECK-NEXT: [[YY:%.*]] = xor i8 [[Y:%.*]], 123
+; CHECK-NEXT: [[B:%.*]] = select i1 [[C:%.*]], i8 [[NX]], i8 [[YY]]
+; CHECK-NEXT: [[AB:%.*]] = add i8 [[B]], [[A:%.*]]
+; CHECK-NEXT: [[NOT_AB:%.*]] = xor i8 [[AB]], -1
+; CHECK-NEXT: ret i8 [[NOT_AB]]
+;
+ %nx = xor i8 %x, -1
+ %yy = xor i8 %y, 123
+ %b = select i1 %c, i8 %nx, i8 %yy
+ %ab = add i8 %a, %b
+ %not_ab = xor i8 %ab, -1
+ ret i8 %not_ab
+}
+
+define i8 @add_fail(i8 %a, i1 %c, i8 %x, i8 %y) {
+; CHECK-LABEL: @add_fail(
+; CHECK-NEXT: [[NX:%.*]] = xor i8 [[X:%.*]], [[A:%.*]]
+; CHECK-NEXT: [[YY:%.*]] = xor i8 [[Y:%.*]], 123
+; CHECK-NEXT: [[B:%.*]] = select i1 [[C:%.*]], i8 [[NX]], i8 [[YY]]
+; CHECK-NEXT: [[AB:%.*]] = add i8 [[B]], [[A]]
+; CHECK-NEXT: [[NOT_AB:%.*]] = xor i8 [[AB]], -1
+; CHECK-NEXT: ret i8 [[NOT_AB]]
+;
+ %nx = xor i8 %x, %a
+ %yy = xor i8 %y, 123
+ %b = select i1 %c, i8 %nx, i8 %yy
+ %ab = add i8 %a, %b
+ %not_ab = xor i8 %ab, -1
+ ret i8 %not_ab
+}
+
+define i8 @sub_1(i8 %a, i1 %c, i8 %x, i8 %y) {
+; CHECK-LABEL: @sub_1(
+; CHECK-NEXT: [[NX:%.*]] = xor i8 [[X:%.*]], -1
+; CHECK-NEXT: [[YY:%.*]] = xor i8 [[Y:%.*]], 123
+; CHECK-NEXT: [[B:%.*]] = select i1 [[C:%.*]], i8 [[NX]], i8 [[YY]]
+; CHECK-NEXT: [[TMP1:%.*]] = xor i8 [[B]], -1
+; CHECK-NEXT: [[NOT_BA:%.*]] = add i8 [[TMP1]], [[A:%.*]]
+; CHECK-NEXT: ret i8 [[NOT_BA]]
+;
+ %nx = xor i8 %x, -1
+ %yy = xor i8 %y, 123
+ %b = select i1 %c, i8 %nx, i8 %yy
+ %ba = sub i8 %b, %a
+ %not_ba = xor i8 %ba, -1
+ ret i8 %not_ba
+}
+
+define i8 @sub_2(i8 %a, i1 %c, i8 %x, i8 %y) {
+; CHECK-LABEL: @sub_2(
+; CHECK-NEXT: [[TMP1:%.*]] = xor i8 [[Y:%.*]], -124
+; CHECK-NEXT: [[B_NEG_V:%.*]] = select i1 [[C:%.*]], i8 [[X:%.*]], i8 [[TMP1]]
+; CHECK-NEXT: [[B_NEG:%.*]] = add i8 [[B_NEG_V]], 1
+; CHECK-NEXT: [[AB:%.*]] = add i8 [[B_NEG]], [[A:%.*]]
+; CHECK-NEXT: [[NOT_AB:%.*]] = xor i8 [[AB]], -1
+; CHECK-NEXT: ret i8 [[NOT_AB]]
+;
+ %nx = xor i8 %x, -1
+ %yy = xor i8 %y, 123
+ %b = select i1 %c, i8 %nx, i8 %yy
+ %ab = sub i8 %a, %b
+ %not_ab = xor i8 %ab, -1
+ ret i8 %not_ab
+}
+
+define i8 @sub_fail(i8 %a, i1 %c, i8 %x, i8 %y) {
+; CHECK-LABEL: @sub_fail(
+; CHECK-NEXT: [[NX:%.*]] = xor i8 [[X:%.*]], -1
+; CHECK-NEXT: [[YY:%.*]] = xor i8 [[Y:%.*]], 123
+; CHECK-NEXT: call void @use.i8(i8 [[NX]])
+; CHECK-NEXT: [[B:%.*]] = select i1 [[C:%.*]], i8 [[NX]], i8 [[YY]]
+; CHECK-NEXT: [[TMP1:%.*]] = xor i8 [[B]], -1
+; CHECK-NEXT: [[NOT_BA:%.*]] = add i8 [[TMP1]], [[A:%.*]]
+; CHECK-NEXT: ret i8 [[NOT_BA]]
+;
+ %nx = xor i8 %x, -1
+ %yy = xor i8 %y, 123
+ call void @use.i8(i8 %nx)
+ %b = select i1 %c, i8 %nx, i8 %yy
+ %ba = sub i8 %b, %a
+ %not_ba = xor i8 %ba, -1
+ ret i8 %not_ba
+}
+
+define i8 @ashr_1(i8 %a, i1 %c, i8 %x, i8 %y) {
+; CHECK-LABEL: @ashr_1(
+; CHECK-NEXT: [[NX:%.*]] = xor i8 [[X:%.*]], -1
+; CHECK-NEXT: [[YY:%.*]] = xor i8 [[Y:%.*]], 123
+; CHECK-NEXT: [[B:%.*]] = select i1 [[C:%.*]], i8 [[NX]], i8 [[YY]]
+; CHECK-NEXT: [[BA:%.*]] = ashr i8 [[B]], [[A:%.*]]
+; CHECK-NEXT: [[NOT_BA:%.*]] = xor i8 [[BA]], -1
+; CHECK-NEXT: ret i8 [[NOT_BA]]
+;
+ %nx = xor i8 %x, -1
+ %yy = xor i8 %y, 123
+ %b = select i1 %c, i8 %nx, i8 %yy
+ %ba = ashr i8 %b, %a
+ %not_ba = xor i8 %ba, -1
+ ret i8 %not_ba
+}
+
+define i8 @ashr_2_fail(i8 %a, i1 %c, i8 %x, i8 %y) {
+; CHECK-LABEL: @ashr_2_fail(
+; CHECK-NEXT: [[NX:%.*]] = xor i8 [[X:%.*]], -1
+; CHECK-NEXT: [[YY:%.*]] = xor i8 [[Y:%.*]], 123
+; CHECK-NEXT: [[B:%.*]] = select i1 [[C:%.*]], i8 [[NX]], i8 [[YY]]
+; CHECK-NEXT: [[AB:%.*]] = ashr i8 [[A:%.*]], [[B]]
+; CHECK-NEXT: [[NOT_AB:%.*]] = xor i8 [[AB]], -1
+; CHECK-NEXT: ret i8 [[NOT_AB]]
+;
+ %nx = xor i8 %x, -1
+ %yy = xor i8 %y, 123
+ %b = select i1 %c, i8 %nx, i8 %yy
+ %ab = ashr i8 %a, %b
+ %not_ab = xor i8 %ab, -1
+ ret i8 %not_ab
+}
+
+define i8 @select_1(i1 %cc, i8 %na, i8 %aa, i1 %c, i8 %x, i8 %y) {
+; CHECK-LABEL: @select_1(
+; CHECK-NEXT: [[NX:%.*]] = xor i8 [[X:%.*]], -1
+; CHECK-NEXT: [[YY:%.*]] = xor i8 [[Y:%.*]], 123
+; CHECK-NEXT: [[B:%.*]] = select i1 [[C:%.*]], i8 [[NX]], i8 [[YY]]
+; CHECK-NEXT: [[TMP1:%.*]] = xor i8 [[NA:%.*]], [[AA:%.*]]
+; CHECK-NEXT: [[A:%.*]] = xor i8 [[TMP1]], 45
+; CHECK-NEXT: [[AB:%.*]] = select i1 [[CC:%.*]], i8 [[A]], i8 [[B]]
+; CHECK-NEXT: [[NOT_AB:%.*]] = xor i8 [[AB]], -1
+; CHECK-NEXT: ret i8 [[NOT_AB]]
+;
+ %nx = xor i8 %x, -1
+ %yy = xor i8 %y, 123
+ %b = select i1 %c, i8 %nx, i8 %yy
+ %nna = xor i8 %na, 45
+ %a = xor i8 %aa, %nna
+ %ab = select i1 %cc, i8 %a, i8 %b
+ %not_ab = xor i8 %ab, -1
+ ret i8 %not_ab
+}
+
+define i8 @select_2(i1 %cc, i8 %na, i1 %c, i8 %x, i8 %y) {
+; CHECK-LABEL: @select_2(
+; CHECK-NEXT: [[NX:%.*]] = xor i8 [[X:%.*]], -1
+; CHECK-NEXT: [[YY:%.*]] = xor i8 [[Y:%.*]], 123
+; CHECK-NEXT: [[B:%.*]] = select i1 [[C:%.*]], i8 [[NX]], i8 [[YY]]
+; CHECK-NEXT: [[A:%.*]] = xor i8 [[NA:%.*]], 45
+; CHECK-NEXT: [[BA:%.*]] = select i1 [[CC:%.*]], i8 [[B]], i8 [[A]]
+; CHECK-NEXT: [[NOT_BA:%.*]] = xor i8 [[BA]], -1
+; CHECK-NEXT: ret i8 [[NOT_BA]]
+;
+ %nx = xor i8 %x, -1
+ %yy = xor i8 %y, 123
+ %b = select i1 %c, i8 %nx, i8 %yy
+ %a = xor i8 %na, 45
+ %ba = select i1 %cc, i8 %b, i8 %a
+ %not_ba = xor i8 %ba, -1
+ ret i8 %not_ba
+}
+
+define i1 @select_logic_or_fail(i1 %cc, i1 %c, i1 %x, i8 %y) {
+; CHECK-LABEL: @select_logic_or_fail(
+; CHECK-NEXT: [[NX:%.*]] = xor i1 [[X:%.*]], true
+; CHECK-NEXT: [[YY:%.*]] = icmp eq i8 [[Y:%.*]], 123
+; CHECK-NEXT: [[B:%.*]] = select i1 [[C:%.*]], i1 [[NX]], i1 [[YY]]
+; CHECK-NEXT: [[AB:%.*]] = select i1 [[CC:%.*]], i1 [[B]], i1 false
+; CHECK-NEXT: [[NOT_AB:%.*]] = xor i1 [[AB]], true
+; CHECK-NEXT: ret i1 [[NOT_AB]]
+;
+ %nx = xor i1 %x, -1
+ %yy = icmp eq i8 %y, 123
+ %b = select i1 %c, i1 %nx, i1 %yy
+ %ab = select i1 %cc, i1 %b, i1 false
+ %not_ab = xor i1 %ab, -1
+ ret i1 %not_ab
+}
+
+define i1 @select_logic_and_fail(i1 %cc, i1 %c, i1 %x, i8 %y) {
+; CHECK-LABEL: @select_logic_and_fail(
+; CHECK-NEXT: [[NX:%.*]] = xor i1 [[X:%.*]], true
+; CHECK-NEXT: [[YY:%.*]] = icmp eq i8 [[Y:%.*]], 123
+; CHECK-NEXT: [[B:%.*]] = select i1 [[C:%.*]], i1 [[NX]], i1 [[YY]]
+; CHECK-NEXT: [[AB:%.*]] = select i1 [[CC:%.*]], i1 true, i1 [[B]]
+; CHECK-NEXT: [[NOT_AB:%.*]] = xor i1 [[AB]], true
+; CHECK-NEXT: ret i1 [[NOT_AB]]
+;
+ %nx = xor i1 %x, -1
+ %yy = icmp eq i8 %y, 123
+ %b = select i1 %c, i1 %nx, i1 %yy
+ %ab = select i1 %cc, i1 true, i1 %b
+ %not_ab = xor i1 %ab, -1
+ ret i1 %not_ab
+}
+
+define i8 @smin_1(i8 %aa, i8 %na, i1 %c, i8 %x, i8 %y) {
+; CHECK-LABEL: @smin_1(
+; CHECK-NEXT: [[NX:%.*]] = xor i8 [[X:%.*]], -1
+; CHECK-NEXT: [[YY:%.*]] = xor i8 [[Y:%.*]], 123
+; CHECK-NEXT: [[B:%.*]] = select i1 [[C:%.*]], i8 [[NX]], i8 [[YY]]
+; CHECK-NEXT: [[NNA:%.*]] = xor i8 [[NA:%.*]], -1
+; CHECK-NEXT: [[A:%.*]] = add i8 [[NNA]], [[AA:%.*]]
+; CHECK-NEXT: [[AB:%.*]] = call i8 @llvm.smin.i8(i8 [[A]], i8 [[B]])
+; CHECK-NEXT: [[NOT_AB:%.*]] = xor i8 [[AB]], -1
+; CHECK-NEXT: ret i8 [[NOT_AB]]
+;
+ %nx = xor i8 %x, -1
+ %yy = xor i8 %y, 123
+ %b = select i1 %c, i8 %nx, i8 %yy
+ %nna = xor i8 %na, -1
+ %a = add i8 %aa, %nna
+ %ab = call i8 @llvm.smin.i8(i8 %a, i8 %b)
+ %not_ab = xor i8 %ab, -1
+ ret i8 %not_ab
+}
+
+define i8 @smin_1_fail(i8 %a, i1 %c, i8 %x, i8 %y) {
+; CHECK-LABEL: @smin_1_fail(
+; CHECK-NEXT: [[NX:%.*]] = xor i8 [[X:%.*]], -1
+; CHECK-NEXT: [[YY:%.*]] = xor i8 [[Y:%.*]], 123
+; CHECK-NEXT: [[B:%.*]] = select i1 [[C:%.*]], i8 [[NX]], i8 [[YY]]
+; CHECK-NEXT: [[AB:%.*]] = call i8 @llvm.smin.i8(i8 [[A:%.*]], i8 [[B]])
+; CHECK-NEXT: [[NOT_AB:%.*]] = xor i8 [[AB]], -1
+; CHECK-NEXT: ret i8 [[NOT_AB]]
+;
+ %nx = xor i8 %x, -1
+ %yy = xor i8 %y, 123
+ %b = select i1 %c, i8 %nx, i8 %yy
+ %ab = call i8 @llvm.smin.i8(i8 %a, i8 %b)
+ %not_ab = xor i8 %ab, -1
+ ret i8 %not_ab
+}
+
+define i8 @umin_1_fail(i1 %c, i8 %x, i8 %y) {
+; CHECK-LABEL: @umin_1_fail(
+; CHECK-NEXT: [[NX:%.*]] = xor i8 [[X:%.*]], -1
+; CHECK-NEXT: [[B:%.*]] = select i1 [[C:%.*]], i8 [[NX]], i8 [[Y:%.*]]
+; CHECK-NEXT: [[BA:%.*]] = call i8 @llvm.umin.i8(i8 [[B]], i8 85)
+; CHECK-NEXT: [[NOT_BA:%.*]] = xor i8 [[BA]], -1
+; CHECK-NEXT: ret i8 [[NOT_BA]]
+;
+ %nx = xor i8 %x, -1
+ %b = select i1 %c, i8 %nx, i8 %y
+ %ba = call i8 @llvm.umin.i8(i8 %b, i8 85)
+ %not_ba = xor i8 %ba, -1
+ ret i8 %not_ba
+}
+
+define i8 @smax_1(i8 %aa, i8 %na, i1 %c, i8 %x, i8 %y) {
+; CHECK-LABEL: @smax_1(
+; CHECK-NEXT: [[NX:%.*]] = xor i8 [[X:%.*]], -1
+; CHECK-NEXT: [[YY:%.*]] = xor i8 [[Y:%.*]], 123
+; CHECK-NEXT: [[B:%.*]] = select i1 [[C:%.*]], i8 [[NX]], i8 [[YY]]
+; CHECK-NEXT: [[NNA:%.*]] = xor i8 [[NA:%.*]], -1
+; CHECK-NEXT: [[A:%.*]] = sub i8 [[NNA]], [[AA:%.*]]
+; CHECK-NEXT: [[AB:%.*]] = call i8 @llvm.smax.i8(i8 [[A]], i8 [[B]])
+; CHECK-NEXT: [[NOT_AB:%.*]] = xor i8 [[AB]], -1
+; CHECK-NEXT: ret i8 [[NOT_AB]]
+;
+ %nx = xor i8 %x, -1
+ %yy = xor i8 %y, 123
+ %b = select i1 %c, i8 %nx, i8 %yy
+ %nna = xor i8 %na, -1
+ %a = sub i8 %nna, %aa
+ %ab = call i8 @llvm.smax.i8(i8 %a, i8 %b)
+ %not_ab = xor i8 %ab, -1
+ ret i8 %not_ab
+}
+
+define i8 @smax_1_fail(i8 %aa, i8 %na, i1 %c, i8 %x, i8 %y) {
+; CHECK-LABEL: @smax_1_fail(
+; CHECK-NEXT: [[NX:%.*]] = xor i8 [[X:%.*]], -1
+; CHECK-NEXT: [[YY:%.*]] = xor i8 [[Y:%.*]], 123
+; CHECK-NEXT: call void @use.i8(i8 [[YY]])
+; CHECK-NEXT: [[B:%.*]] = select i1 [[C:%.*]], i8 [[NX]], i8 [[YY]]
+; CHECK-NEXT: [[NNA:%.*]] = xor i8 [[NA:%.*]], -1
+; CHECK-NEXT: [[A:%.*]] = sub i8 [[NNA]], [[AA:%.*]]
+; CHECK-NEXT: [[AB:%.*]] = call i8 @llvm.smax.i8(i8 [[A]], i8 [[B]])
+; CHECK-NEXT: [[NOT_AB:%.*]] = xor i8 [[AB]], -1
+; CHECK-NEXT: ret i8 [[NOT_AB]]
+;
+ %nx = xor i8 %x, -1
+ %yy = xor i8 %y, 123
+ call void @use.i8(i8 %yy)
+ %b = select i1 %c, i8 %nx, i8 %yy
+ %nna = xor i8 %na, -1
+ %a = sub i8 %nna, %aa
+ %ab = call i8 @llvm.smax.i8(i8 %a, i8 %b)
+ %not_ab = xor i8 %ab, -1
+ ret i8 %not_ab
+}
+
+define i8 @umax_1(i8 %na, i1 %c, i8 %x, i8 %y) {
+; CHECK-LABEL: @umax_1(
+; CHECK-NEXT: [[NX:%.*]] = xor i8 [[X:%.*]], -1
+; CHECK-NEXT: [[YY:%.*]] = xor i8 [[Y:%.*]], 123
+; CHECK-NEXT: [[B:%.*]] = select i1 [[C:%.*]], i8 [[NX]], i8 [[YY]]
+; CHECK-NEXT: [[BA:%.*]] = call i8 @llvm.umax.i8(i8 [[B]], i8 85)
+; CHECK-NEXT: [[NOT_BA:%.*]] = xor i8 [[BA]], -1
+; CHECK-NEXT: ret i8 [[NOT_BA]]
+;
+ %nx = xor i8 %x, -1
+ %yy = xor i8 %y, 123
+ %b = select i1 %c, i8 %nx, i8 %yy
+ %ba = call i8 @llvm.umax.i8(i8 %b, i8 85)
+ %not_ba = xor i8 %ba, -1
+ ret i8 %not_ba
+}
+
+define i8 @umax_1_fail(i8 %na, i1 %c, i8 %x, i8 %y) {
+; CHECK-LABEL: @umax_1_fail(
+; CHECK-NEXT: [[NX:%.*]] = xor i8 [[X:%.*]], -1
+; CHECK-NEXT: [[YY:%.*]] = xor i8 [[Y:%.*]], 123
+; CHECK-NEXT: [[B:%.*]] = select i1 [[C:%.*]], i8 [[NX]], i8 [[YY]]
+; CHECK-NEXT: call void @use.i8(i8 [[B]])
+; CHECK-NEXT: [[BA:%.*]] = call i8 @llvm.umax.i8(i8 [[B]], i8 85)
+; CHECK-NEXT: [[NOT_BA:%.*]] = xor i8 [[BA]], -1
+; CHECK-NEXT: ret i8 [[NOT_BA]]
+;
+ %nx = xor i8 %x, -1
+ %yy = xor i8 %y, 123
+ %b = select i1 %c, i8 %nx, i8 %yy
+ call void @use.i8(i8 %b)
+ %ba = call i8 @llvm.umax.i8(i8 %b, i8 85)
+ %not_ba = xor i8 %ba, -1
+ ret i8 %not_ba
+}
>From 4ee977f6b7f964d92844b41f400e9b3431637e81 Mon Sep 17 00:00:00 2001
From: Noah Goldstein <goldstein.w.n at gmail.com>
Date: Mon, 18 Sep 2023 15:12:51 -0500
Subject: [PATCH 03/13] [InstCombine] add `getFreeInverted` to perform folds
for free inversion of op
With the current logic of `if(isFreeToInvert(Op)) return Not(Op)` its
fairly easy to either 1) cause regressions or 2) infinite loops
if the folds we have for `Not(Op)` ever de-sync with the cases we
know are freely invertible.
This patch adds `getFreeInverted` which is able to build the free
inverted op along with check for free inversion to alleviate this
problem.
---
.../Transforms/InstCombine/InstCombiner.h | 128 +++++++++++++-----
.../InstCombine/InstCombineAndOrXor.cpp | 5 +
.../Transforms/InstCombine/free-inversion.ll | 110 ++++++---------
.../Transforms/InstCombine/icmp-of-or-x.ll | 6 +-
llvm/test/Transforms/InstCombine/pr63791.ll | 2 +-
.../LoopVectorize/ARM/mve-selectandorcost.ll | 2 +-
.../LoopVectorize/reduction-inloop.ll | 25 ++--
7 files changed, 158 insertions(+), 120 deletions(-)
diff --git a/llvm/include/llvm/Transforms/InstCombine/InstCombiner.h b/llvm/include/llvm/Transforms/InstCombine/InstCombiner.h
index 3a1ac917da6a6cb..a2e3c176cfab5df 100644
--- a/llvm/include/llvm/Transforms/InstCombine/InstCombiner.h
+++ b/llvm/include/llvm/Transforms/InstCombine/InstCombiner.h
@@ -233,56 +233,116 @@ class LLVM_LIBRARY_VISIBILITY InstCombiner {
PatternMatch::m_Value()));
}
- /// Return true if the specified value is free to invert (apply ~ to).
- /// This happens in cases where the ~ can be eliminated. If WillInvertAllUses
- /// is true, work under the assumption that the caller intends to remove all
- /// uses of V and only keep uses of ~V.
- ///
- /// See also: canFreelyInvertAllUsersOf()
- static bool isFreeToInvert(Value *V, bool WillInvertAllUses,
- unsigned Depth = 0) {
+ /// Return nonnull value if V is free to invert (with condition) regarding
+ /// WillInvertAllUses.
+ /// If Builder is nonnull, it will return a simplified ~V
+ /// If builder is null, it will return an arbitrary nonnull value (not
+ /// dereferenceable).
+ static Value *getFreeInverted(Value *V, bool WillInvertAllUses,
+ BuilderTy *Builder, unsigned Depth = 0) {
+ static Value *const kNonNull = reinterpret_cast<Value *>(uintptr_t(1));
// ~(~(X)) -> X.
- if (match(V, m_Not(PatternMatch::m_Value())))
- return true;
+ Value *A, *B;
+ if (match(V, m_Not(PatternMatch::m_Value(A))))
+ return A;
+ Constant *C;
// Constants can be considered to be not'ed values.
- if (match(V, PatternMatch::m_AnyIntegralConstant()))
- return true;
+ if (match(V, PatternMatch::m_ImmConstant(C)))
+ return ConstantExpr::getNot(C);
if (Depth++ >= MaxAnalysisRecursionDepth)
- return false;
+ return nullptr;
// The rest of the cases require that we invert all uses so don't bother
// doing the analysis if we know we can't use the result.
if (!WillInvertAllUses)
- return false;
+ return nullptr;
// Compares can be inverted if all of their uses are being modified to use
// the ~V.
- if (isa<CmpInst>(V))
- return true;
+ if (auto *I = dyn_cast<CmpInst>(V)) {
+ if (Builder != nullptr)
+ return Builder->CreateCmp(I->getInversePredicate(), I->getOperand(0),
+ I->getOperand(1));
+ return kNonNull;
+ }
- Value *A, *B;
- // If `V` is of the form `A + B` then `-1 - V` can be folded into
- // `~B - A` or `~A - B` if we are willing to invert all of the uses.
- if (match(V, m_Add(PatternMatch::m_Value(A), PatternMatch::m_Value(B))))
- return isFreeToInvert(A, A->hasOneUse(), Depth) ||
- isFreeToInvert(B, B->hasOneUse(), Depth);
+ // If `V` is of the form `A + Constant` then `-1 - V` can be folded into
+ // `(-1 - Constant) - A` if we are willing to invert all of the uses.
+ if (match(V, m_Add(PatternMatch::m_Value(A), PatternMatch::m_Value(B)))) {
+ if (auto *BV = getFreeInverted(B, B->hasOneUse(), Builder, Depth))
+ return Builder ? Builder->CreateSub(BV, A) : kNonNull;
+ if (auto *AV = getFreeInverted(A, A->hasOneUse(), Builder, Depth))
+ return Builder ? Builder->CreateSub(AV, B) : kNonNull;
+ return nullptr;
+ }
- // If `V` is of the form `A - B` then `-1 - V` can be folded into
- // `~A + B` if we are willing to invert all of the uses.
- if (match(V, m_Sub(PatternMatch::m_Value(A), PatternMatch::m_Value())))
- return isFreeToInvert(A, A->hasOneUse(), Depth);
+ // If `V` is of the form `A ^ ~B` then `~(A ^ ~B)` can be folded
+ // into `A ^ B` if we are willing to invert all of the uses.
+ if (match(V, m_Xor(PatternMatch::m_Value(A), PatternMatch::m_Value(B)))) {
+ if (auto *BV = getFreeInverted(B, B->hasOneUse(), Builder, Depth))
+ return Builder ? Builder->CreateXor(A, BV) : kNonNull;
+ if (auto *AV = getFreeInverted(A, A->hasOneUse(), Builder, Depth))
+ return Builder ? Builder->CreateXor(AV, B) : kNonNull;
+ return nullptr;
+ }
+ // If `V` is of the form `Constant - A` then `-1 - V` can be folded into
+ // `A + (-1 - Constant)` if we are willing to invert all of the uses.
+ if (match(V, m_Sub(PatternMatch::m_Value(A), PatternMatch::m_Value(B)))) {
+ if (auto *AV = getFreeInverted(A, A->hasOneUse(), Builder, Depth))
+ return Builder ? Builder->CreateAdd(AV, B) : kNonNull;
+ return nullptr;
+ }
+
+ // If `V` is of the form `Constant - A` then `-1 - V` can be folded into
+ // `A + (-1 - Constant)` if we are willing to invert all of the uses.
+ if (match(V, m_AShr(PatternMatch::m_Value(A), PatternMatch::m_Value(B)))) {
+ if (auto *AV = getFreeInverted(A, A->hasOneUse(), Builder, Depth))
+ return Builder ? Builder->CreateAShr(AV, B) : kNonNull;
+ return nullptr;
+ }
+
+ Value *Cond;
+ // LogicOps are special in that we canonicalize them at the cost of an
+ // instruction.
+ bool IsSelect =
+ match(V, m_Select(PatternMatch::m_Value(Cond), PatternMatch::m_Value(A),
+ PatternMatch::m_Value(B))) &&
+ !match(V, PatternMatch::m_LogicalOp(PatternMatch::m_Value(),
+ PatternMatch::m_Value()));
// Selects/min/max with invertible operands are freely invertible
- if (match(V, m_Select(PatternMatch::m_Value(), PatternMatch::m_Value(A),
- PatternMatch::m_Value(B))) ||
- match(V,
- m_MaxOrMin(PatternMatch::m_Value(A), PatternMatch::m_Value(B))))
- return isFreeToInvert(A, A->hasOneUse(), Depth) &&
- isFreeToInvert(B, B->hasOneUse(), Depth);
-
- return false;
+ if (IsSelect || match(V, m_MaxOrMin(PatternMatch::m_Value(A),
+ PatternMatch::m_Value(B)))) {
+ if (!getFreeInverted(A, A->hasOneUse(), /*Builder*/ nullptr, Depth) ||
+ !getFreeInverted(B, B->hasOneUse(), /*Builder*/ nullptr, Depth))
+ return nullptr;
+ if (Builder != nullptr) {
+ A = getFreeInverted(A, A->hasOneUse(), Builder, Depth);
+ B = getFreeInverted(B, B->hasOneUse(), Builder, Depth);
+ assert(
+ A != nullptr && B != nullptr &&
+ "Unable to build inverted values for known freely invertable ops");
+ if (auto *II = dyn_cast<IntrinsicInst>(V))
+ return Builder->CreateBinaryIntrinsic(II->getIntrinsicID(), A, B);
+ return Builder->CreateSelect(Cond, A, B);
+ }
+ return kNonNull;
+ }
+
+ return nullptr;
+ }
+
+ /// Return true if the specified value is free to invert (apply ~ to).
+ /// This happens in cases where the ~ can be eliminated. If WillInvertAllUses
+ /// is true, work under the assumption that the caller intends to remove all
+ /// uses of V and only keep uses of ~V.
+ ///
+ /// See also: canFreelyInvertAllUsersOf()
+ static bool isFreeToInvert(Value *V, bool WillInvertAllUses) {
+ return getFreeInverted(V, WillInvertAllUses, /*Builder*/ nullptr,
+ /*Depth*/ 0) != nullptr;
}
/// Given i1 V, can every user of V be freely adapted if V is changed to !V ?
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp b/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
index 3e0218d9b76d1f7..5a3d6dd20d60a08 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
@@ -4410,6 +4410,11 @@ Instruction *InstCombinerImpl::foldNot(BinaryOperator &I) {
if (Instruction *NewXor = foldNotXor(I, Builder))
return NewXor;
+ // TODO: Could handle multi-use better by checking if all uses of NotOp (other
+ // than I) can be inverted.
+ if (Value *R = getFreeInverted(NotOp, NotOp->hasOneUse(), &Builder))
+ return replaceInstUsesWith(I, R);
+
return nullptr;
}
diff --git a/llvm/test/Transforms/InstCombine/free-inversion.ll b/llvm/test/Transforms/InstCombine/free-inversion.ll
index a2931fb9a081109..0d694d3f72e3afa 100644
--- a/llvm/test/Transforms/InstCombine/free-inversion.ll
+++ b/llvm/test/Transforms/InstCombine/free-inversion.ll
@@ -10,11 +10,9 @@ declare void @use.i8(i8)
define i8 @xor_1(i8 %a, i1 %c, i8 %x, i8 %y) {
; CHECK-LABEL: @xor_1(
-; CHECK-NEXT: [[NX:%.*]] = xor i8 [[X:%.*]], -1
-; CHECK-NEXT: [[YY:%.*]] = xor i8 [[Y:%.*]], 123
-; CHECK-NEXT: [[B:%.*]] = select i1 [[C:%.*]], i8 [[NX]], i8 [[YY]]
-; CHECK-NEXT: [[BA:%.*]] = xor i8 [[B]], [[A:%.*]]
-; CHECK-NEXT: [[NOT_BA:%.*]] = xor i8 [[BA]], -1
+; CHECK-NEXT: [[TMP1:%.*]] = xor i8 [[Y:%.*]], -124
+; CHECK-NEXT: [[B_NOT:%.*]] = select i1 [[C:%.*]], i8 [[X:%.*]], i8 [[TMP1]]
+; CHECK-NEXT: [[NOT_BA:%.*]] = xor i8 [[B_NOT]], [[A:%.*]]
; CHECK-NEXT: ret i8 [[NOT_BA]]
;
%nx = xor i8 %x, -1
@@ -27,11 +25,9 @@ define i8 @xor_1(i8 %a, i1 %c, i8 %x, i8 %y) {
define i8 @xor_2(i8 %a, i1 %c, i8 %x, i8 %y) {
; CHECK-LABEL: @xor_2(
-; CHECK-NEXT: [[NX:%.*]] = xor i8 [[X:%.*]], -1
-; CHECK-NEXT: [[YY:%.*]] = xor i8 [[Y:%.*]], 123
-; CHECK-NEXT: [[B:%.*]] = select i1 [[C:%.*]], i8 [[NX]], i8 [[YY]]
-; CHECK-NEXT: [[AB:%.*]] = xor i8 [[B]], [[A:%.*]]
-; CHECK-NEXT: [[NOT_AB:%.*]] = xor i8 [[AB]], -1
+; CHECK-NEXT: [[TMP1:%.*]] = xor i8 [[Y:%.*]], -124
+; CHECK-NEXT: [[B_NOT:%.*]] = select i1 [[C:%.*]], i8 [[X:%.*]], i8 [[TMP1]]
+; CHECK-NEXT: [[NOT_AB:%.*]] = xor i8 [[B_NOT]], [[A:%.*]]
; CHECK-NEXT: ret i8 [[NOT_AB]]
;
%nx = xor i8 %x, -1
@@ -59,11 +55,9 @@ define i8 @xor_fail(i8 %a, i1 %c, i8 %x, i8 %y) {
define i8 @add_1(i8 %a, i1 %c, i8 %x, i8 %y) {
; CHECK-LABEL: @add_1(
-; CHECK-NEXT: [[NX:%.*]] = xor i8 [[X:%.*]], -1
-; CHECK-NEXT: [[YY:%.*]] = xor i8 [[Y:%.*]], 123
-; CHECK-NEXT: [[B:%.*]] = select i1 [[C:%.*]], i8 [[NX]], i8 [[YY]]
-; CHECK-NEXT: [[BA:%.*]] = add i8 [[B]], [[A:%.*]]
-; CHECK-NEXT: [[NOT_BA:%.*]] = xor i8 [[BA]], -1
+; CHECK-NEXT: [[TMP1:%.*]] = xor i8 [[Y:%.*]], -124
+; CHECK-NEXT: [[TMP2:%.*]] = select i1 [[C:%.*]], i8 [[X:%.*]], i8 [[TMP1]]
+; CHECK-NEXT: [[NOT_BA:%.*]] = sub i8 [[TMP2]], [[A:%.*]]
; CHECK-NEXT: ret i8 [[NOT_BA]]
;
%nx = xor i8 %x, -1
@@ -76,11 +70,9 @@ define i8 @add_1(i8 %a, i1 %c, i8 %x, i8 %y) {
define i8 @add_2(i8 %a, i1 %c, i8 %x, i8 %y) {
; CHECK-LABEL: @add_2(
-; CHECK-NEXT: [[NX:%.*]] = xor i8 [[X:%.*]], -1
-; CHECK-NEXT: [[YY:%.*]] = xor i8 [[Y:%.*]], 123
-; CHECK-NEXT: [[B:%.*]] = select i1 [[C:%.*]], i8 [[NX]], i8 [[YY]]
-; CHECK-NEXT: [[AB:%.*]] = add i8 [[B]], [[A:%.*]]
-; CHECK-NEXT: [[NOT_AB:%.*]] = xor i8 [[AB]], -1
+; CHECK-NEXT: [[TMP1:%.*]] = xor i8 [[Y:%.*]], -124
+; CHECK-NEXT: [[TMP2:%.*]] = select i1 [[C:%.*]], i8 [[X:%.*]], i8 [[TMP1]]
+; CHECK-NEXT: [[NOT_AB:%.*]] = sub i8 [[TMP2]], [[A:%.*]]
; CHECK-NEXT: ret i8 [[NOT_AB]]
;
%nx = xor i8 %x, -1
@@ -110,11 +102,9 @@ define i8 @add_fail(i8 %a, i1 %c, i8 %x, i8 %y) {
define i8 @sub_1(i8 %a, i1 %c, i8 %x, i8 %y) {
; CHECK-LABEL: @sub_1(
-; CHECK-NEXT: [[NX:%.*]] = xor i8 [[X:%.*]], -1
-; CHECK-NEXT: [[YY:%.*]] = xor i8 [[Y:%.*]], 123
-; CHECK-NEXT: [[B:%.*]] = select i1 [[C:%.*]], i8 [[NX]], i8 [[YY]]
-; CHECK-NEXT: [[TMP1:%.*]] = xor i8 [[B]], -1
-; CHECK-NEXT: [[NOT_BA:%.*]] = add i8 [[TMP1]], [[A:%.*]]
+; CHECK-NEXT: [[TMP1:%.*]] = xor i8 [[Y:%.*]], -124
+; CHECK-NEXT: [[TMP2:%.*]] = select i1 [[C:%.*]], i8 [[X:%.*]], i8 [[TMP1]]
+; CHECK-NEXT: [[NOT_BA:%.*]] = add i8 [[TMP2]], [[A:%.*]]
; CHECK-NEXT: ret i8 [[NOT_BA]]
;
%nx = xor i8 %x, -1
@@ -129,9 +119,8 @@ define i8 @sub_2(i8 %a, i1 %c, i8 %x, i8 %y) {
; CHECK-LABEL: @sub_2(
; CHECK-NEXT: [[TMP1:%.*]] = xor i8 [[Y:%.*]], -124
; CHECK-NEXT: [[B_NEG_V:%.*]] = select i1 [[C:%.*]], i8 [[X:%.*]], i8 [[TMP1]]
-; CHECK-NEXT: [[B_NEG:%.*]] = add i8 [[B_NEG_V]], 1
-; CHECK-NEXT: [[AB:%.*]] = add i8 [[B_NEG]], [[A:%.*]]
-; CHECK-NEXT: [[NOT_AB:%.*]] = xor i8 [[AB]], -1
+; CHECK-NEXT: [[TMP2:%.*]] = add i8 [[B_NEG_V]], [[A:%.*]]
+; CHECK-NEXT: [[NOT_AB:%.*]] = sub i8 -2, [[TMP2]]
; CHECK-NEXT: ret i8 [[NOT_AB]]
;
%nx = xor i8 %x, -1
@@ -145,11 +134,10 @@ define i8 @sub_2(i8 %a, i1 %c, i8 %x, i8 %y) {
define i8 @sub_fail(i8 %a, i1 %c, i8 %x, i8 %y) {
; CHECK-LABEL: @sub_fail(
; CHECK-NEXT: [[NX:%.*]] = xor i8 [[X:%.*]], -1
-; CHECK-NEXT: [[YY:%.*]] = xor i8 [[Y:%.*]], 123
; CHECK-NEXT: call void @use.i8(i8 [[NX]])
-; CHECK-NEXT: [[B:%.*]] = select i1 [[C:%.*]], i8 [[NX]], i8 [[YY]]
-; CHECK-NEXT: [[TMP1:%.*]] = xor i8 [[B]], -1
-; CHECK-NEXT: [[NOT_BA:%.*]] = add i8 [[TMP1]], [[A:%.*]]
+; CHECK-NEXT: [[TMP1:%.*]] = xor i8 [[Y:%.*]], -124
+; CHECK-NEXT: [[TMP2:%.*]] = select i1 [[C:%.*]], i8 [[X]], i8 [[TMP1]]
+; CHECK-NEXT: [[NOT_BA:%.*]] = add i8 [[TMP2]], [[A:%.*]]
; CHECK-NEXT: ret i8 [[NOT_BA]]
;
%nx = xor i8 %x, -1
@@ -163,11 +151,9 @@ define i8 @sub_fail(i8 %a, i1 %c, i8 %x, i8 %y) {
define i8 @ashr_1(i8 %a, i1 %c, i8 %x, i8 %y) {
; CHECK-LABEL: @ashr_1(
-; CHECK-NEXT: [[NX:%.*]] = xor i8 [[X:%.*]], -1
-; CHECK-NEXT: [[YY:%.*]] = xor i8 [[Y:%.*]], 123
-; CHECK-NEXT: [[B:%.*]] = select i1 [[C:%.*]], i8 [[NX]], i8 [[YY]]
-; CHECK-NEXT: [[BA:%.*]] = ashr i8 [[B]], [[A:%.*]]
-; CHECK-NEXT: [[NOT_BA:%.*]] = xor i8 [[BA]], -1
+; CHECK-NEXT: [[TMP1:%.*]] = xor i8 [[Y:%.*]], -124
+; CHECK-NEXT: [[TMP2:%.*]] = select i1 [[C:%.*]], i8 [[X:%.*]], i8 [[TMP1]]
+; CHECK-NEXT: [[NOT_BA:%.*]] = ashr i8 [[TMP2]], [[A:%.*]]
; CHECK-NEXT: ret i8 [[NOT_BA]]
;
%nx = xor i8 %x, -1
@@ -197,13 +183,11 @@ define i8 @ashr_2_fail(i8 %a, i1 %c, i8 %x, i8 %y) {
define i8 @select_1(i1 %cc, i8 %na, i8 %aa, i1 %c, i8 %x, i8 %y) {
; CHECK-LABEL: @select_1(
-; CHECK-NEXT: [[NX:%.*]] = xor i8 [[X:%.*]], -1
-; CHECK-NEXT: [[YY:%.*]] = xor i8 [[Y:%.*]], 123
-; CHECK-NEXT: [[B:%.*]] = select i1 [[C:%.*]], i8 [[NX]], i8 [[YY]]
; CHECK-NEXT: [[TMP1:%.*]] = xor i8 [[NA:%.*]], [[AA:%.*]]
-; CHECK-NEXT: [[A:%.*]] = xor i8 [[TMP1]], 45
-; CHECK-NEXT: [[AB:%.*]] = select i1 [[CC:%.*]], i8 [[A]], i8 [[B]]
-; CHECK-NEXT: [[NOT_AB:%.*]] = xor i8 [[AB]], -1
+; CHECK-NEXT: [[TMP2:%.*]] = xor i8 [[TMP1]], -46
+; CHECK-NEXT: [[TMP3:%.*]] = xor i8 [[Y:%.*]], -124
+; CHECK-NEXT: [[TMP4:%.*]] = select i1 [[C:%.*]], i8 [[X:%.*]], i8 [[TMP3]]
+; CHECK-NEXT: [[NOT_AB:%.*]] = select i1 [[CC:%.*]], i8 [[TMP2]], i8 [[TMP4]]
; CHECK-NEXT: ret i8 [[NOT_AB]]
;
%nx = xor i8 %x, -1
@@ -218,12 +202,10 @@ define i8 @select_1(i1 %cc, i8 %na, i8 %aa, i1 %c, i8 %x, i8 %y) {
define i8 @select_2(i1 %cc, i8 %na, i1 %c, i8 %x, i8 %y) {
; CHECK-LABEL: @select_2(
-; CHECK-NEXT: [[NX:%.*]] = xor i8 [[X:%.*]], -1
-; CHECK-NEXT: [[YY:%.*]] = xor i8 [[Y:%.*]], 123
-; CHECK-NEXT: [[B:%.*]] = select i1 [[C:%.*]], i8 [[NX]], i8 [[YY]]
-; CHECK-NEXT: [[A:%.*]] = xor i8 [[NA:%.*]], 45
-; CHECK-NEXT: [[BA:%.*]] = select i1 [[CC:%.*]], i8 [[B]], i8 [[A]]
-; CHECK-NEXT: [[NOT_BA:%.*]] = xor i8 [[BA]], -1
+; CHECK-NEXT: [[TMP1:%.*]] = xor i8 [[Y:%.*]], -124
+; CHECK-NEXT: [[TMP2:%.*]] = select i1 [[C:%.*]], i8 [[X:%.*]], i8 [[TMP1]]
+; CHECK-NEXT: [[TMP3:%.*]] = xor i8 [[NA:%.*]], -46
+; CHECK-NEXT: [[NOT_BA:%.*]] = select i1 [[CC:%.*]], i8 [[TMP2]], i8 [[TMP3]]
; CHECK-NEXT: ret i8 [[NOT_BA]]
;
%nx = xor i8 %x, -1
@@ -271,13 +253,10 @@ define i1 @select_logic_and_fail(i1 %cc, i1 %c, i1 %x, i8 %y) {
define i8 @smin_1(i8 %aa, i8 %na, i1 %c, i8 %x, i8 %y) {
; CHECK-LABEL: @smin_1(
-; CHECK-NEXT: [[NX:%.*]] = xor i8 [[X:%.*]], -1
-; CHECK-NEXT: [[YY:%.*]] = xor i8 [[Y:%.*]], 123
-; CHECK-NEXT: [[B:%.*]] = select i1 [[C:%.*]], i8 [[NX]], i8 [[YY]]
-; CHECK-NEXT: [[NNA:%.*]] = xor i8 [[NA:%.*]], -1
-; CHECK-NEXT: [[A:%.*]] = add i8 [[NNA]], [[AA:%.*]]
-; CHECK-NEXT: [[AB:%.*]] = call i8 @llvm.smin.i8(i8 [[A]], i8 [[B]])
-; CHECK-NEXT: [[NOT_AB:%.*]] = xor i8 [[AB]], -1
+; CHECK-NEXT: [[TMP1:%.*]] = sub i8 [[NA:%.*]], [[AA:%.*]]
+; CHECK-NEXT: [[TMP2:%.*]] = xor i8 [[Y:%.*]], -124
+; CHECK-NEXT: [[TMP3:%.*]] = select i1 [[C:%.*]], i8 [[X:%.*]], i8 [[TMP2]]
+; CHECK-NEXT: [[NOT_AB:%.*]] = call i8 @llvm.smax.i8(i8 [[TMP1]], i8 [[TMP3]])
; CHECK-NEXT: ret i8 [[NOT_AB]]
;
%nx = xor i8 %x, -1
@@ -324,13 +303,10 @@ define i8 @umin_1_fail(i1 %c, i8 %x, i8 %y) {
define i8 @smax_1(i8 %aa, i8 %na, i1 %c, i8 %x, i8 %y) {
; CHECK-LABEL: @smax_1(
-; CHECK-NEXT: [[NX:%.*]] = xor i8 [[X:%.*]], -1
-; CHECK-NEXT: [[YY:%.*]] = xor i8 [[Y:%.*]], 123
-; CHECK-NEXT: [[B:%.*]] = select i1 [[C:%.*]], i8 [[NX]], i8 [[YY]]
-; CHECK-NEXT: [[NNA:%.*]] = xor i8 [[NA:%.*]], -1
-; CHECK-NEXT: [[A:%.*]] = sub i8 [[NNA]], [[AA:%.*]]
-; CHECK-NEXT: [[AB:%.*]] = call i8 @llvm.smax.i8(i8 [[A]], i8 [[B]])
-; CHECK-NEXT: [[NOT_AB:%.*]] = xor i8 [[AB]], -1
+; CHECK-NEXT: [[TMP1:%.*]] = add i8 [[NA:%.*]], [[AA:%.*]]
+; CHECK-NEXT: [[TMP2:%.*]] = xor i8 [[Y:%.*]], -124
+; CHECK-NEXT: [[TMP3:%.*]] = select i1 [[C:%.*]], i8 [[X:%.*]], i8 [[TMP2]]
+; CHECK-NEXT: [[NOT_AB:%.*]] = call i8 @llvm.smin.i8(i8 [[TMP1]], i8 [[TMP3]])
; CHECK-NEXT: ret i8 [[NOT_AB]]
;
%nx = xor i8 %x, -1
@@ -368,11 +344,9 @@ define i8 @smax_1_fail(i8 %aa, i8 %na, i1 %c, i8 %x, i8 %y) {
define i8 @umax_1(i8 %na, i1 %c, i8 %x, i8 %y) {
; CHECK-LABEL: @umax_1(
-; CHECK-NEXT: [[NX:%.*]] = xor i8 [[X:%.*]], -1
-; CHECK-NEXT: [[YY:%.*]] = xor i8 [[Y:%.*]], 123
-; CHECK-NEXT: [[B:%.*]] = select i1 [[C:%.*]], i8 [[NX]], i8 [[YY]]
-; CHECK-NEXT: [[BA:%.*]] = call i8 @llvm.umax.i8(i8 [[B]], i8 85)
-; CHECK-NEXT: [[NOT_BA:%.*]] = xor i8 [[BA]], -1
+; CHECK-NEXT: [[TMP1:%.*]] = xor i8 [[Y:%.*]], -124
+; CHECK-NEXT: [[TMP2:%.*]] = select i1 [[C:%.*]], i8 [[X:%.*]], i8 [[TMP1]]
+; CHECK-NEXT: [[NOT_BA:%.*]] = call i8 @llvm.umin.i8(i8 [[TMP2]], i8 -86)
; CHECK-NEXT: ret i8 [[NOT_BA]]
;
%nx = xor i8 %x, -1
diff --git a/llvm/test/Transforms/InstCombine/icmp-of-or-x.ll b/llvm/test/Transforms/InstCombine/icmp-of-or-x.ll
index 839aa98a8b24e2f..4b8df439b846fb6 100644
--- a/llvm/test/Transforms/InstCombine/icmp-of-or-x.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-of-or-x.ll
@@ -121,9 +121,9 @@ define i1 @or_ne_notY_eq_1s(i8 %x, i8 %y) {
define i1 @or_ne_notY_eq_1s_fail_bad_not(i8 %x, i8 %y) {
; CHECK-LABEL: @or_ne_notY_eq_1s_fail_bad_not(
-; CHECK-NEXT: [[NY:%.*]] = xor i8 [[Y:%.*]], -2
-; CHECK-NEXT: [[OR:%.*]] = or i8 [[NY]], [[X:%.*]]
-; CHECK-NEXT: [[CMP:%.*]] = icmp ne i8 [[OR]], [[X]]
+; CHECK-NEXT: [[TMP1:%.*]] = xor i8 [[Y:%.*]], 1
+; CHECK-NEXT: [[TMP2:%.*]] = or i8 [[TMP1]], [[X:%.*]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp ne i8 [[TMP2]], -1
; CHECK-NEXT: ret i1 [[CMP]]
;
%ny = xor i8 %y, -2
diff --git a/llvm/test/Transforms/InstCombine/pr63791.ll b/llvm/test/Transforms/InstCombine/pr63791.ll
index 0fac19d583fe96c..a489b2e3e6221a0 100644
--- a/llvm/test/Transforms/InstCombine/pr63791.ll
+++ b/llvm/test/Transforms/InstCombine/pr63791.ll
@@ -17,7 +17,7 @@ define void @y() {
; CHECK: for.cond5.preheader.i:
; CHECK-NEXT: br i1 false, label [[FOR_INC19_I:%.*]], label [[FOR_COND1_LOOPEXIT_I:%.*]]
; CHECK: for.inc19.i:
-; CHECK-NEXT: br i1 true, label [[FOR_COND1_LOOPEXIT_I]], label [[FOR_INC19_I]]
+; CHECK-NEXT: br i1 false, label [[FOR_INC19_I]], label [[FOR_COND1_LOOPEXIT_I]]
;
entry:
br label %for.cond.i
diff --git a/llvm/test/Transforms/LoopVectorize/ARM/mve-selectandorcost.ll b/llvm/test/Transforms/LoopVectorize/ARM/mve-selectandorcost.ll
index 5d85a4cd73fddfd..c4d1607ff640af5 100644
--- a/llvm/test/Transforms/LoopVectorize/ARM/mve-selectandorcost.ll
+++ b/llvm/test/Transforms/LoopVectorize/ARM/mve-selectandorcost.ll
@@ -86,7 +86,7 @@ define float @test(ptr nocapture readonly %pA, ptr nocapture readonly %pB, i32 %
; CHECK-NEXT: [[ACCUM_1]] = phi float [ [[ADD4]], [[IF_THEN]] ], [ [[ACCUM_017]], [[WHILE_BODY]] ]
; CHECK-NEXT: [[DEC]] = add i32 [[BLOCKSIZE_ADDR_018]], -1
; CHECK-NEXT: [[CMP_NOT:%.*]] = icmp eq i32 [[DEC]], 0
-; CHECK-NEXT: br i1 [[CMP_NOT]], label [[WHILE_END]], label [[WHILE_BODY]], !llvm.loop [[LOOP2:![0-9]+]]
+; CHECK-NEXT: br i1 [[CMP_NOT]], label [[WHILE_END]], label [[WHILE_BODY]], !llvm.loop [[LOOP3:![0-9]+]]
; CHECK: while.end:
; CHECK-NEXT: [[ACCUM_0_LCSSA:%.*]] = phi float [ 0.000000e+00, [[ENTRY:%.*]] ], [ [[ACCUM_1]], [[IF_END]] ], [ [[TMP14]], [[MIDDLE_BLOCK]] ]
; CHECK-NEXT: ret float [[ACCUM_0_LCSSA]]
diff --git a/llvm/test/Transforms/LoopVectorize/reduction-inloop.ll b/llvm/test/Transforms/LoopVectorize/reduction-inloop.ll
index 305da23709fa5e7..18b05c05d9b9d21 100644
--- a/llvm/test/Transforms/LoopVectorize/reduction-inloop.ll
+++ b/llvm/test/Transforms/LoopVectorize/reduction-inloop.ll
@@ -1292,7 +1292,7 @@ define i32 @predicated_or_dominates_reduction(ptr %b) {
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_LOAD_CONTINUE6:%.*]] ]
-; CHECK-NEXT: [[VEC_PHI:%.*]] = phi i32 [ undef, [[VECTOR_PH]] ], [ [[TMP49:%.*]], [[PRED_LOAD_CONTINUE6]] ]
+; CHECK-NEXT: [[VEC_PHI:%.*]] = phi i32 [ undef, [[VECTOR_PH]] ], [ [[TMP51:%.*]], [[PRED_LOAD_CONTINUE6]] ]
; CHECK-NEXT: [[TMP0:%.*]] = or i32 [[INDEX]], 1
; CHECK-NEXT: [[TMP1:%.*]] = or i32 [[INDEX]], 2
; CHECK-NEXT: [[TMP2:%.*]] = or i32 [[INDEX]], 3
@@ -1353,24 +1353,23 @@ define i32 @predicated_or_dominates_reduction(ptr %b) {
; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE6]]
; CHECK: pred.load.continue6:
; CHECK-NEXT: [[TMP43:%.*]] = phi <4 x i32> [ [[TMP37]], [[PRED_LOAD_CONTINUE4]] ], [ [[TMP42]], [[PRED_LOAD_IF5]] ]
-; CHECK-NEXT: [[TMP44:%.*]] = icmp eq <4 x i32> [[TMP43]], zeroinitializer
-; CHECK-NEXT: [[NOT_:%.*]] = xor <4 x i1> [[TMP19]], <i1 true, i1 true, i1 true, i1 true>
-; CHECK-NEXT: [[TMP45:%.*]] = select <4 x i1> [[NOT_]], <4 x i1> <i1 true, i1 true, i1 true, i1 true>, <4 x i1> [[TMP44]]
-; CHECK-NEXT: [[DOTNOT7:%.*]] = and <4 x i1> [[TMP45]], [[TMP19]]
-; CHECK-NEXT: [[NOT__NOT7:%.*]] = xor <4 x i1> [[DOTNOT7]], <i1 true, i1 true, i1 true, i1 true>
-; CHECK-NEXT: [[TMP46:%.*]] = bitcast <4 x i1> [[NOT__NOT7]] to i4
-; CHECK-NEXT: [[TMP47:%.*]] = call i4 @llvm.ctpop.i4(i4 [[TMP46]]), !range [[RNG42:![0-9]+]]
-; CHECK-NEXT: [[TMP48:%.*]] = zext i4 [[TMP47]] to i32
-; CHECK-NEXT: [[TMP49]] = add i32 [[VEC_PHI]], [[TMP48]]
+; CHECK-NEXT: [[TMP44:%.*]] = icmp ne <4 x i32> [[TMP43]], zeroinitializer
+; CHECK-NEXT: [[TMP45:%.*]] = select <4 x i1> [[TMP19]], <4 x i1> [[TMP44]], <4 x i1> zeroinitializer
+; CHECK-NEXT: [[TMP46:%.*]] = xor <4 x i1> [[TMP19]], <i1 true, i1 true, i1 true, i1 true>
+; CHECK-NEXT: [[TMP47:%.*]] = or <4 x i1> [[TMP45]], [[TMP46]]
+; CHECK-NEXT: [[TMP48:%.*]] = bitcast <4 x i1> [[TMP47]] to i4
+; CHECK-NEXT: [[TMP49:%.*]] = call i4 @llvm.ctpop.i4(i4 [[TMP48]]), !range [[RNG42:![0-9]+]]
+; CHECK-NEXT: [[TMP50:%.*]] = zext i4 [[TMP49]] to i32
+; CHECK-NEXT: [[TMP51]] = add i32 [[VEC_PHI]], [[TMP50]]
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
-; CHECK-NEXT: [[TMP50:%.*]] = icmp eq i32 [[INDEX_NEXT]], 1000
-; CHECK-NEXT: br i1 [[TMP50]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP43:![0-9]+]]
+; CHECK-NEXT: [[TMP52:%.*]] = icmp eq i32 [[INDEX_NEXT]], 1000
+; CHECK-NEXT: br i1 [[TMP52]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP43:![0-9]+]]
; CHECK: middle.block:
; CHECK-NEXT: br i1 true, label [[FOR_COND_CLEANUP:%.*]], label [[SCALAR_PH]]
; CHECK: scalar.ph:
; CHECK-NEXT: br label [[FOR_BODY:%.*]]
; CHECK: for.cond.cleanup:
-; CHECK-NEXT: [[A_1_LCSSA:%.*]] = phi i32 [ poison, [[FOR_INC:%.*]] ], [ [[TMP49]], [[MIDDLE_BLOCK]] ]
+; CHECK-NEXT: [[A_1_LCSSA:%.*]] = phi i32 [ poison, [[FOR_INC:%.*]] ], [ [[TMP51]], [[MIDDLE_BLOCK]] ]
; CHECK-NEXT: ret i32 [[A_1_LCSSA]]
; CHECK: for.body:
; CHECK-NEXT: br i1 poison, label [[LOR_LHS_FALSE:%.*]], label [[IF_THEN:%.*]]
>From 56ed3e9ba9ed99fb2229b111449e42fd9b46ef1d Mon Sep 17 00:00:00 2001
From: Noah Goldstein <goldstein.w.n at gmail.com>
Date: Wed, 13 Sep 2023 13:45:49 -0500
Subject: [PATCH 04/13] [InstCombine] Add tests for expanding
`foldICmpWithLowBitMaskedVal`; NFC
Differential Revision: https://reviews.llvm.org/D159057
---
.../InstCombine/icmp-and-lowbit-mask.ll | 651 ++++++++++++++++++
1 file changed, 651 insertions(+)
create mode 100644 llvm/test/Transforms/InstCombine/icmp-and-lowbit-mask.ll
diff --git a/llvm/test/Transforms/InstCombine/icmp-and-lowbit-mask.ll b/llvm/test/Transforms/InstCombine/icmp-and-lowbit-mask.ll
new file mode 100644
index 000000000000000..d815183577bbcbf
--- /dev/null
+++ b/llvm/test/Transforms/InstCombine/icmp-and-lowbit-mask.ll
@@ -0,0 +1,651 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -passes=instcombine -S | FileCheck %s
+
+declare void @llvm.assume(i1)
+declare i8 @llvm.ctpop.i8(i8)
+declare i8 @llvm.umin.i8(i8, i8)
+declare i8 @llvm.umax.i8(i8, i8)
+declare i8 @llvm.smin.i8(i8, i8)
+declare i8 @llvm.smax.i8(i8, i8)
+declare i8 @llvm.bitreverse.i8(i8)
+declare void @use.i8(i8)
+declare void @use.i16(i16)
+define i1 @src_is_mask_zext(i16 %x_in, i8 %y) {
+; CHECK-LABEL: @src_is_mask_zext(
+; CHECK-NEXT: [[X:%.*]] = xor i16 [[X_IN:%.*]], 123
+; CHECK-NEXT: [[M_IN:%.*]] = lshr i8 -1, [[Y:%.*]]
+; CHECK-NEXT: [[MASK:%.*]] = zext i8 [[M_IN]] to i16
+; CHECK-NEXT: [[AND:%.*]] = and i16 [[X]], [[MASK]]
+; CHECK-NEXT: [[R:%.*]] = icmp eq i16 [[AND]], [[X]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = xor i16 %x_in, 123
+ %m_in = lshr i8 -1, %y
+ %mask = zext i8 %m_in to i16
+
+ %and = and i16 %x, %mask
+ %r = icmp eq i16 %and, %x
+ ret i1 %r
+}
+
+define i1 @src_is_mask_zext_fail_not_mask(i16 %x_in, i8 %y) {
+; CHECK-LABEL: @src_is_mask_zext_fail_not_mask(
+; CHECK-NEXT: [[X:%.*]] = xor i16 [[X_IN:%.*]], 123
+; CHECK-NEXT: [[M_IN:%.*]] = lshr i8 -2, [[Y:%.*]]
+; CHECK-NEXT: [[MASK:%.*]] = zext i8 [[M_IN]] to i16
+; CHECK-NEXT: [[AND:%.*]] = and i16 [[X]], [[MASK]]
+; CHECK-NEXT: [[R:%.*]] = icmp eq i16 [[AND]], [[X]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = xor i16 %x_in, 123
+ %m_in = lshr i8 -2, %y
+ %mask = zext i8 %m_in to i16
+
+ %and = and i16 %x, %mask
+ %r = icmp eq i16 %and, %x
+ ret i1 %r
+}
+
+define i1 @src_is_mask_sext(i16 %x_in, i8 %y) {
+; CHECK-LABEL: @src_is_mask_sext(
+; CHECK-NEXT: [[X:%.*]] = xor i16 [[X_IN:%.*]], 123
+; CHECK-NEXT: [[M_IN:%.*]] = lshr i8 31, [[Y:%.*]]
+; CHECK-NEXT: [[MASK:%.*]] = zext i8 [[M_IN]] to i16
+; CHECK-NEXT: [[NOTMASK:%.*]] = xor i16 [[MASK]], -1
+; CHECK-NEXT: [[AND:%.*]] = and i16 [[X]], [[NOTMASK]]
+; CHECK-NEXT: [[R:%.*]] = icmp eq i16 [[AND]], 0
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = xor i16 %x_in, 123
+ %m_in = lshr i8 31, %y
+ %mask = sext i8 %m_in to i16
+ %notmask = xor i16 %mask, -1
+
+ %and = and i16 %notmask, %x
+ %r = icmp eq i16 %and, 0
+ ret i1 %r
+}
+
+define i1 @src_is_mask_sext_fail_multiuse(i16 %x_in, i8 %y) {
+; CHECK-LABEL: @src_is_mask_sext_fail_multiuse(
+; CHECK-NEXT: [[X:%.*]] = xor i16 [[X_IN:%.*]], 122
+; CHECK-NEXT: [[M_IN:%.*]] = lshr i8 -1, [[Y:%.*]]
+; CHECK-NEXT: [[TMP1:%.*]] = xor i8 [[M_IN]], -1
+; CHECK-NEXT: [[NOTMASK:%.*]] = sext i8 [[TMP1]] to i16
+; CHECK-NEXT: [[AND:%.*]] = and i16 [[X]], [[NOTMASK]]
+; CHECK-NEXT: call void @use.i16(i16 [[AND]])
+; CHECK-NEXT: [[R:%.*]] = icmp eq i16 [[AND]], 0
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = xor i16 %x_in, 123
+ %m_in = lshr i8 -1, %y
+ %mask = sext i8 %m_in to i16
+ %notmask = xor i16 %mask, -1
+
+ %and = and i16 %notmask, %x
+ call void @use.i16(i16 %and)
+ %r = icmp eq i16 %and, 0
+ ret i1 %r
+}
+
+define i1 @src_is_mask_and(i8 %x_in, i8 %y, i8 %z) {
+; CHECK-LABEL: @src_is_mask_and(
+; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
+; CHECK-NEXT: [[MY:%.*]] = lshr i8 7, [[Y:%.*]]
+; CHECK-NEXT: [[MZ:%.*]] = lshr i8 -1, [[Z:%.*]]
+; CHECK-NEXT: [[MASK:%.*]] = and i8 [[MY]], [[MZ]]
+; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], [[MASK]]
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[X]], [[AND]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = xor i8 %x_in, 123
+ %my = ashr i8 7, %y
+ %mz = lshr i8 -1, %z
+ %mask = and i8 %my, %mz
+
+ %and = and i8 %x, %mask
+ %r = icmp eq i8 %x, %and
+ ret i1 %r
+}
+
+define i1 @src_is_mask_and_fail_mixed(i8 %x_in, i8 %y, i8 %z) {
+; CHECK-LABEL: @src_is_mask_and_fail_mixed(
+; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
+; CHECK-NEXT: [[MY:%.*]] = ashr i8 -8, [[Y:%.*]]
+; CHECK-NEXT: [[MZ:%.*]] = lshr i8 -1, [[Z:%.*]]
+; CHECK-NEXT: [[MASK:%.*]] = and i8 [[MY]], [[MZ]]
+; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], [[MASK]]
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[X]], [[AND]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = xor i8 %x_in, 123
+ %my = ashr i8 -8, %y
+ %mz = lshr i8 -1, %z
+ %mask = and i8 %my, %mz
+
+ %and = and i8 %x, %mask
+ %r = icmp eq i8 %x, %and
+ ret i1 %r
+}
+
+define i1 @src_is_mask_or(i8 %x_in, i8 %y) {
+; CHECK-LABEL: @src_is_mask_or(
+; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
+; CHECK-NEXT: [[MY:%.*]] = lshr i8 -1, [[Y:%.*]]
+; CHECK-NEXT: [[MASK:%.*]] = and i8 [[MY]], 7
+; CHECK-NEXT: [[AND:%.*]] = and i8 [[MASK]], [[X]]
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[X]], [[AND]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = xor i8 %x_in, 123
+ %my = lshr i8 -1, %y
+ %mask = and i8 %my, 7
+
+ %and = and i8 %mask, %x
+ %r = icmp eq i8 %x, %and
+ ret i1 %r
+}
+
+define i1 @src_is_mask_xor(i8 %x_in, i8 %y) {
+; CHECK-LABEL: @src_is_mask_xor(
+; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
+; CHECK-NEXT: [[Y_M1:%.*]] = add i8 [[Y:%.*]], -1
+; CHECK-NEXT: [[MASK:%.*]] = xor i8 [[Y_M1]], [[Y]]
+; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], [[MASK]]
+; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[AND]], [[X]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = xor i8 %x_in, 123
+ %y_m1 = add i8 %y, -1
+ %mask = xor i8 %y, %y_m1
+ %and = and i8 %x, %mask
+ %r = icmp ne i8 %and, %x
+ ret i1 %r
+}
+
+define i1 @src_is_mask_xor_fail_notmask(i8 %x_in, i8 %y) {
+; CHECK-LABEL: @src_is_mask_xor_fail_notmask(
+; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
+; CHECK-NEXT: [[Y_M1_NOT:%.*]] = sub i8 0, [[Y:%.*]]
+; CHECK-NEXT: [[NOTMASK:%.*]] = xor i8 [[Y_M1_NOT]], [[Y]]
+; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], [[NOTMASK]]
+; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[AND]], [[X]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = xor i8 %x_in, 123
+ %y_m1 = add i8 %y, -1
+ %mask = xor i8 %y, %y_m1
+ %notmask = xor i8 %mask, -1
+ %and = and i8 %x, %notmask
+ %r = icmp ne i8 %and, %x
+ ret i1 %r
+}
+
+define i1 @src_is_mask_select(i8 %x_in, i8 %y, i1 %cond) {
+; CHECK-LABEL: @src_is_mask_select(
+; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
+; CHECK-NEXT: [[Y_M1:%.*]] = add i8 [[Y:%.*]], -1
+; CHECK-NEXT: [[YMASK:%.*]] = xor i8 [[Y_M1]], [[Y]]
+; CHECK-NEXT: [[MASK:%.*]] = select i1 [[COND:%.*]], i8 [[YMASK]], i8 15
+; CHECK-NEXT: [[AND:%.*]] = and i8 [[MASK]], [[X]]
+; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[AND]], [[X]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = xor i8 %x_in, 123
+ %y_m1 = add i8 %y, -1
+ %ymask = xor i8 %y, %y_m1
+ %mask = select i1 %cond, i8 %ymask, i8 15
+
+ %and = and i8 %mask, %x
+ %r = icmp ne i8 %and, %x
+ ret i1 %r
+}
+
+define i1 @src_is_mask_select_fail_wrong_pattern(i8 %x_in, i8 %y, i1 %cond, i8 %z) {
+; CHECK-LABEL: @src_is_mask_select_fail_wrong_pattern(
+; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
+; CHECK-NEXT: [[Y_M1:%.*]] = add i8 [[Y:%.*]], -1
+; CHECK-NEXT: [[YMASK:%.*]] = xor i8 [[Y_M1]], [[Y]]
+; CHECK-NEXT: [[MASK:%.*]] = select i1 [[COND:%.*]], i8 [[YMASK]], i8 15
+; CHECK-NEXT: [[AND:%.*]] = and i8 [[MASK]], [[X]]
+; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[AND]], [[Z:%.*]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = xor i8 %x_in, 123
+ %y_m1 = add i8 %y, -1
+ %ymask = xor i8 %y, %y_m1
+ %mask = select i1 %cond, i8 %ymask, i8 15
+
+ %and = and i8 %mask, %x
+ %r = icmp ne i8 %and, %z
+ ret i1 %r
+}
+
+define i1 @src_is_mask_shl_lshr(i8 %x_in, i8 %y, i1 %cond) {
+; CHECK-LABEL: @src_is_mask_shl_lshr(
+; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 122
+; CHECK-NEXT: [[TMP1:%.*]] = lshr i8 -1, [[Y:%.*]]
+; CHECK-NEXT: [[NOTMASK:%.*]] = xor i8 [[TMP1]], -1
+; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], [[NOTMASK]]
+; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[AND]], 0
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = xor i8 %x_in, 123
+ %m_shl = shl i8 -1, %y
+ %mask = lshr i8 %m_shl, %y
+ %notmask = xor i8 %mask, -1
+
+ %and = and i8 %x, %notmask
+ %r = icmp ne i8 0, %and
+ ret i1 %r
+}
+
+define i1 @src_is_mask_shl_lshr_fail_not_allones(i8 %x_in, i8 %y, i1 %cond) {
+; CHECK-LABEL: @src_is_mask_shl_lshr_fail_not_allones(
+; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
+; CHECK-NEXT: [[TMP1:%.*]] = lshr i8 -1, [[Y:%.*]]
+; CHECK-NEXT: [[MASK:%.*]] = and i8 [[TMP1]], -2
+; CHECK-NEXT: [[NOTMASK:%.*]] = xor i8 [[MASK]], -1
+; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], [[NOTMASK]]
+; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[AND]], 0
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = xor i8 %x_in, 123
+ %m_shl = shl i8 -2, %y
+ %mask = lshr i8 %m_shl, %y
+ %notmask = xor i8 %mask, -1
+
+ %and = and i8 %x, %notmask
+ %r = icmp ne i8 0, %and
+ ret i1 %r
+}
+
+define i1 @src_is_mask_lshr(i8 %x_in, i8 %y, i8 %z, i1 %cond) {
+; CHECK-LABEL: @src_is_mask_lshr(
+; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
+; CHECK-NEXT: [[Y_M1:%.*]] = add i8 [[Y:%.*]], -1
+; CHECK-NEXT: [[YMASK:%.*]] = xor i8 [[Y_M1]], [[Y]]
+; CHECK-NEXT: [[SMASK:%.*]] = select i1 [[COND:%.*]], i8 [[YMASK]], i8 15
+; CHECK-NEXT: [[MASK:%.*]] = lshr i8 [[SMASK]], [[Z:%.*]]
+; CHECK-NEXT: [[AND:%.*]] = and i8 [[MASK]], [[X]]
+; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[X]], [[AND]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = xor i8 %x_in, 123
+ %y_m1 = add i8 %y, -1
+ %ymask = xor i8 %y, %y_m1
+ %smask = select i1 %cond, i8 %ymask, i8 15
+ %mask = lshr i8 %smask, %z
+ %and = and i8 %mask, %x
+ %r = icmp ne i8 %x, %and
+ ret i1 %r
+}
+
+define i1 @src_is_mask_ashr(i8 %x_in, i8 %y, i8 %z, i1 %cond) {
+; CHECK-LABEL: @src_is_mask_ashr(
+; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
+; CHECK-NEXT: [[Y_M1:%.*]] = add i8 [[Y:%.*]], -1
+; CHECK-NEXT: [[YMASK:%.*]] = xor i8 [[Y_M1]], [[Y]]
+; CHECK-NEXT: [[SMASK:%.*]] = select i1 [[COND:%.*]], i8 [[YMASK]], i8 15
+; CHECK-NEXT: [[MASK:%.*]] = ashr i8 [[SMASK]], [[Z:%.*]]
+; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], [[MASK]]
+; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[AND]], [[X]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = xor i8 %x_in, 123
+ %y_m1 = add i8 %y, -1
+ %ymask = xor i8 %y, %y_m1
+ %smask = select i1 %cond, i8 %ymask, i8 15
+ %mask = ashr i8 %smask, %z
+ %and = and i8 %x, %mask
+ %r = icmp ult i8 %and, %x
+ ret i1 %r
+}
+
+define i1 @src_is_mask_p2_m1(i8 %x_in, i8 %y) {
+; CHECK-LABEL: @src_is_mask_p2_m1(
+; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
+; CHECK-NEXT: [[P2ORZ:%.*]] = shl i8 2, [[Y:%.*]]
+; CHECK-NEXT: [[MASK:%.*]] = add i8 [[P2ORZ]], -1
+; CHECK-NEXT: [[AND:%.*]] = and i8 [[MASK]], [[X]]
+; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[AND]], [[X]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = xor i8 %x_in, 123
+ %p2orz = shl i8 2, %y
+ %mask = add i8 %p2orz, -1
+ %and = and i8 %mask, %x
+ %r = icmp ult i8 %and, %x
+ ret i1 %r
+}
+
+define i1 @src_is_mask_umax(i8 %x_in, i8 %y) {
+; CHECK-LABEL: @src_is_mask_umax(
+; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
+; CHECK-NEXT: [[Y_M1:%.*]] = add i8 [[Y:%.*]], -1
+; CHECK-NEXT: [[YMASK:%.*]] = xor i8 [[Y_M1]], [[Y]]
+; CHECK-NEXT: [[MASK:%.*]] = call i8 @llvm.umax.i8(i8 [[YMASK]], i8 3)
+; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], [[MASK]]
+; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[AND]], [[X]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = xor i8 %x_in, 123
+ %y_m1 = add i8 %y, -1
+ %ymask = xor i8 %y, %y_m1
+ %mask = call i8 @llvm.umax.i8(i8 %ymask, i8 3)
+
+ %and = and i8 %x, %mask
+ %r = icmp ugt i8 %x, %and
+ ret i1 %r
+}
+
+define i1 @src_is_mask_umin(i8 %x_in, i8 %y, i8 %z) {
+; CHECK-LABEL: @src_is_mask_umin(
+; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
+; CHECK-NEXT: [[Y_M1:%.*]] = add i8 [[Y:%.*]], -1
+; CHECK-NEXT: [[YMASK:%.*]] = xor i8 [[Y_M1]], [[Y]]
+; CHECK-NEXT: [[ZMASK:%.*]] = lshr i8 15, [[Z:%.*]]
+; CHECK-NEXT: [[MASK:%.*]] = call i8 @llvm.umin.i8(i8 [[YMASK]], i8 [[ZMASK]])
+; CHECK-NEXT: [[AND:%.*]] = and i8 [[MASK]], [[X]]
+; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[AND]], [[X]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = xor i8 %x_in, 123
+ %y_m1 = add i8 %y, -1
+ %ymask = xor i8 %y, %y_m1
+ %zmask = lshr i8 15, %z
+ %mask = call i8 @llvm.umin.i8(i8 %ymask, i8 %zmask)
+
+ %and = and i8 %mask, %x
+ %r = icmp ugt i8 %x, %and
+ ret i1 %r
+}
+
+define i1 @src_is_mask_umin_fail_mismatch(i8 %x_in, i8 %y) {
+; CHECK-LABEL: @src_is_mask_umin_fail_mismatch(
+; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
+; CHECK-NEXT: [[Y_M1:%.*]] = add i8 [[Y:%.*]], -1
+; CHECK-NEXT: [[YMASK:%.*]] = xor i8 [[Y_M1]], [[Y]]
+; CHECK-NEXT: [[MASK:%.*]] = call i8 @llvm.umin.i8(i8 [[YMASK]], i8 -32)
+; CHECK-NEXT: [[AND:%.*]] = and i8 [[MASK]], [[X]]
+; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[AND]], [[X]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = xor i8 %x_in, 123
+ %y_m1 = add i8 %y, -1
+ %ymask = xor i8 %y, %y_m1
+ %mask = call i8 @llvm.umin.i8(i8 %ymask, i8 -32)
+
+ %and = and i8 %mask, %x
+ %r = icmp ugt i8 %x, %and
+ ret i1 %r
+}
+
+define i1 @src_is_mask_smax(i8 %x_in, i8 %y) {
+; CHECK-LABEL: @src_is_mask_smax(
+; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
+; CHECK-NEXT: [[Y_M1:%.*]] = add i8 [[Y:%.*]], -1
+; CHECK-NEXT: [[YMASK:%.*]] = xor i8 [[Y_M1]], [[Y]]
+; CHECK-NEXT: [[MASK:%.*]] = call i8 @llvm.smax.i8(i8 [[YMASK]], i8 -1)
+; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], [[MASK]]
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[AND]], [[X]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = xor i8 %x_in, 123
+ %y_m1 = add i8 %y, -1
+ %ymask = xor i8 %y, %y_m1
+ %mask = call i8 @llvm.smax.i8(i8 %ymask, i8 -1)
+
+ %and = and i8 %x, %mask
+ %r = icmp uge i8 %and, %x
+ ret i1 %r
+}
+
+define i1 @src_is_mask_smin(i8 %x_in, i8 %y) {
+; CHECK-LABEL: @src_is_mask_smin(
+; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
+; CHECK-NEXT: [[Y_M1:%.*]] = add i8 [[Y:%.*]], -1
+; CHECK-NEXT: [[YMASK:%.*]] = xor i8 [[Y_M1]], [[Y]]
+; CHECK-NEXT: [[MASK:%.*]] = call i8 @llvm.smin.i8(i8 [[YMASK]], i8 0)
+; CHECK-NEXT: [[AND:%.*]] = and i8 [[MASK]], [[X]]
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[AND]], [[X]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = xor i8 %x_in, 123
+ %y_m1 = add i8 %y, -1
+ %ymask = xor i8 %y, %y_m1
+ %mask = call i8 @llvm.smin.i8(i8 %ymask, i8 0)
+
+ %and = and i8 %mask, %x
+ %r = icmp uge i8 %and, %x
+ ret i1 %r
+}
+
+define i1 @src_is_mask_bitreverse_not_mask(i8 %x_in, i8 %y) {
+; CHECK-LABEL: @src_is_mask_bitreverse_not_mask(
+; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
+; CHECK-NEXT: [[NMASK:%.*]] = shl i8 -1, [[Y:%.*]]
+; CHECK-NEXT: [[MASK:%.*]] = call i8 @llvm.bitreverse.i8(i8 [[NMASK]])
+; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], [[MASK]]
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[AND]], [[X]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = xor i8 %x_in, 123
+ %nmask = shl i8 -1, %y
+ %mask = call i8 @llvm.bitreverse.i8(i8 %nmask)
+
+ %and = and i8 %x, %mask
+ %r = icmp ule i8 %x, %and
+ ret i1 %r
+}
+
+define i1 @src_is_notmask_sext(i16 %x_in, i8 %y) {
+; CHECK-LABEL: @src_is_notmask_sext(
+; CHECK-NEXT: [[X:%.*]] = xor i16 [[X_IN:%.*]], 123
+; CHECK-NEXT: [[M_IN:%.*]] = shl i8 -8, [[Y:%.*]]
+; CHECK-NEXT: [[TMP1:%.*]] = xor i8 [[M_IN]], -1
+; CHECK-NEXT: [[MASK:%.*]] = sext i8 [[TMP1]] to i16
+; CHECK-NEXT: [[AND:%.*]] = and i16 [[X]], [[MASK]]
+; CHECK-NEXT: [[R:%.*]] = icmp eq i16 [[AND]], [[X]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = xor i16 %x_in, 123
+ %m_in = shl i8 -8, %y
+ %nmask = sext i8 %m_in to i16
+ %mask = xor i16 %nmask, -1
+ %and = and i16 %mask, %x
+ %r = icmp ule i16 %x, %and
+ ret i1 %r
+}
+
+define i1 @src_is_notmask_shl(i8 %x_in, i8 %y, i1 %cond) {
+; CHECK-LABEL: @src_is_notmask_shl(
+; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 122
+; CHECK-NEXT: [[NMASK:%.*]] = shl i8 -1, [[Y:%.*]]
+; CHECK-NEXT: [[MASK:%.*]] = call i8 @llvm.bitreverse.i8(i8 [[NMASK]])
+; CHECK-NEXT: [[NOTMASK0:%.*]] = xor i8 [[MASK]], -1
+; CHECK-NEXT: [[NOTMASK:%.*]] = select i1 [[COND:%.*]], i8 [[NOTMASK0]], i8 -8
+; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], [[NOTMASK]]
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[AND]], 0
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = xor i8 %x_in, 123
+ %nmask = shl i8 -1, %y
+ %mask = call i8 @llvm.bitreverse.i8(i8 %nmask)
+ %notmask0 = xor i8 %mask, -1
+ %notmask = select i1 %cond, i8 %notmask0, i8 -8
+ %and = and i8 %x, %notmask
+ %r = icmp eq i8 %and, 0
+ ret i1 %r
+}
+
+define i1 @src_is_notmask_shl_fail_multiuse_invert(i8 %x_in, i8 %y, i1 %cond) {
+; CHECK-LABEL: @src_is_notmask_shl_fail_multiuse_invert(
+; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 122
+; CHECK-NEXT: [[NMASK:%.*]] = shl i8 -1, [[Y:%.*]]
+; CHECK-NEXT: [[MASK:%.*]] = call i8 @llvm.bitreverse.i8(i8 [[NMASK]])
+; CHECK-NEXT: [[NOTMASK0:%.*]] = xor i8 [[MASK]], -1
+; CHECK-NEXT: [[NOTMASK:%.*]] = select i1 [[COND:%.*]], i8 [[NOTMASK0]], i8 -8
+; CHECK-NEXT: call void @use.i8(i8 [[NOTMASK]])
+; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], [[NOTMASK]]
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[AND]], 0
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = xor i8 %x_in, 123
+ %nmask = shl i8 -1, %y
+ %mask = call i8 @llvm.bitreverse.i8(i8 %nmask)
+ %notmask0 = xor i8 %mask, -1
+ %notmask = select i1 %cond, i8 %notmask0, i8 -8
+ call void @use.i8(i8 %notmask)
+ %and = and i8 %x, %notmask
+ %r = icmp eq i8 %and, 0
+ ret i1 %r
+}
+
+define i1 @src_is_notmask_lshr_shl(i8 %x_in, i8 %y) {
+; CHECK-LABEL: @src_is_notmask_lshr_shl(
+; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
+; CHECK-NEXT: [[X_HIGHBITS:%.*]] = lshr i8 [[X]], [[Y:%.*]]
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[X_HIGHBITS]], 0
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = xor i8 %x_in, 123
+ %mask_shr = lshr i8 -1, %y
+ %nmask = shl i8 %mask_shr, %y
+ %mask = xor i8 %nmask, -1
+ %and = and i8 %mask, %x
+ %r = icmp eq i8 %and, %x
+ ret i1 %r
+}
+
+define i1 @src_is_notmask_lshr_shl_fail_mismatch_shifts(i8 %x_in, i8 %y, i8 %z) {
+; CHECK-LABEL: @src_is_notmask_lshr_shl_fail_mismatch_shifts(
+; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
+; CHECK-NEXT: [[MASK_SHR:%.*]] = lshr i8 -1, [[Y:%.*]]
+; CHECK-NEXT: [[NMASK:%.*]] = shl i8 [[MASK_SHR]], [[Z:%.*]]
+; CHECK-NEXT: [[MASK:%.*]] = xor i8 [[NMASK]], -1
+; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], [[MASK]]
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[AND]], [[X]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = xor i8 %x_in, 123
+ %mask_shr = lshr i8 -1, %y
+ %nmask = shl i8 %mask_shr, %z
+ %mask = xor i8 %nmask, -1
+ %and = and i8 %mask, %x
+ %r = icmp eq i8 %and, %x
+ ret i1 %r
+}
+
+define i1 @src_is_notmask_ashr(i16 %x_in, i8 %y, i16 %z) {
+; CHECK-LABEL: @src_is_notmask_ashr(
+; CHECK-NEXT: [[X:%.*]] = xor i16 [[X_IN:%.*]], 123
+; CHECK-NEXT: [[M_IN:%.*]] = shl i8 -32, [[Y:%.*]]
+; CHECK-NEXT: [[NMASK:%.*]] = sext i8 [[M_IN]] to i16
+; CHECK-NEXT: [[NMASK_SHR:%.*]] = ashr i16 [[NMASK]], [[Z:%.*]]
+; CHECK-NEXT: [[MASK:%.*]] = xor i16 [[NMASK_SHR]], -1
+; CHECK-NEXT: [[AND:%.*]] = and i16 [[X]], [[MASK]]
+; CHECK-NEXT: [[R:%.*]] = icmp eq i16 [[X]], [[AND]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = xor i16 %x_in, 123
+ %m_in = shl i8 -32, %y
+ %nmask = sext i8 %m_in to i16
+ %nmask_shr = ashr i16 %nmask, %z
+ %mask = xor i16 %nmask_shr, -1
+ %and = and i16 %x, %mask
+ %r = icmp eq i16 %x, %and
+ ret i1 %r
+}
+
+define i1 @src_is_notmask_neg_p2(i8 %x_in, i8 %y) {
+; CHECK-LABEL: @src_is_notmask_neg_p2(
+; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
+; CHECK-NEXT: [[NY:%.*]] = sub i8 0, [[Y:%.*]]
+; CHECK-NEXT: [[P2:%.*]] = and i8 [[NY]], [[Y]]
+; CHECK-NEXT: [[NMASK:%.*]] = sub i8 0, [[P2]]
+; CHECK-NEXT: [[MASK:%.*]] = call i8 @llvm.bitreverse.i8(i8 [[NMASK]])
+; CHECK-NEXT: [[NOTMASK:%.*]] = xor i8 [[MASK]], -1
+; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], [[NOTMASK]]
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[AND]], 0
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = xor i8 %x_in, 123
+ %ny = sub i8 0, %y
+ %p2 = and i8 %ny, %y
+ %nmask = sub i8 0, %p2
+ %mask = call i8 @llvm.bitreverse.i8(i8 %nmask)
+ %notmask = xor i8 %mask, -1
+ %and = and i8 %notmask, %x
+ %r = icmp eq i8 0, %and
+ ret i1 %r
+}
+
+define i1 @src_is_notmask_neg_p2_fail_not_invertable(i8 %x_in, i8 %y) {
+; CHECK-LABEL: @src_is_notmask_neg_p2_fail_not_invertable(
+; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
+; CHECK-NEXT: [[NY:%.*]] = sub i8 0, [[Y:%.*]]
+; CHECK-NEXT: [[P2:%.*]] = and i8 [[NY]], [[Y]]
+; CHECK-NEXT: [[NOTMASK:%.*]] = sub i8 0, [[P2]]
+; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], [[NOTMASK]]
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[AND]], 0
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = xor i8 %x_in, 123
+ %ny = sub i8 0, %y
+ %p2 = and i8 %ny, %y
+ %notmask = sub i8 0, %p2
+ %and = and i8 %notmask, %x
+ %r = icmp eq i8 0, %and
+ ret i1 %r
+}
+
+define i1 @src_is_mask_const_slt(i8 %x_in) {
+; CHECK-LABEL: @src_is_mask_const_slt(
+; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
+; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], 7
+; CHECK-NEXT: [[R:%.*]] = icmp slt i8 [[X]], [[AND]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = xor i8 %x_in, 123
+ %and = and i8 %x, 7
+ %r = icmp slt i8 %x, %and
+ ret i1 %r
+}
+
+define i1 @src_is_mask_const_sgt(i8 %x_in) {
+; CHECK-LABEL: @src_is_mask_const_sgt(
+; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
+; CHECK-NEXT: [[R:%.*]] = icmp sgt i8 [[X]], 7
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = xor i8 %x_in, 123
+ %and = and i8 %x, 7
+ %r = icmp sgt i8 %x, %and
+ ret i1 %r
+}
+
+define i1 @src_is_mask_const_sle(i8 %x_in) {
+; CHECK-LABEL: @src_is_mask_const_sle(
+; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
+; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], 31
+; CHECK-NEXT: [[R:%.*]] = icmp sle i8 [[AND]], [[X]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = xor i8 %x_in, 123
+ %and = and i8 %x, 31
+ %r = icmp sle i8 %and, %x
+ ret i1 %r
+}
+
+define i1 @src_is_mask_const_sge(i8 %x_in) {
+; CHECK-LABEL: @src_is_mask_const_sge(
+; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
+; CHECK-NEXT: [[R:%.*]] = icmp slt i8 [[X]], 32
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = xor i8 %x_in, 123
+ %and = and i8 %x, 31
+ %r = icmp sge i8 %and, %x
+ ret i1 %r
+}
>From d6c6cbc96268fd2c4e97393bf4a22b6e2d37e7d5 Mon Sep 17 00:00:00 2001
From: Noah Goldstein <goldstein.w.n at gmail.com>
Date: Wed, 13 Sep 2023 13:45:52 -0500
Subject: [PATCH 05/13] [InstCombine] Improve mask detection in
`foldICmpWithLowBitMaskedVal`
Make recursive matcher that is able to detect a lot more patterns.
Proofs for all supported patterns: https://alive2.llvm.org/ce/z/fSQ3nZ
Differential Revision: https://reviews.llvm.org/D159058
---
llvm/include/llvm/IR/PatternMatch.h | 25 +++
.../InstCombine/InstCombineCompares.cpp | 149 ++++++++++++++++--
...nt-low-bit-mask-and-icmp-eq-to-icmp-ule.ll | 3 +-
...nt-low-bit-mask-and-icmp-ne-to-icmp-ugt.ll | 3 +-
...t-low-bit-mask-and-icmp-sge-to-icmp-sle.ll | 3 +-
...t-low-bit-mask-and-icmp-sgt-to-icmp-sgt.ll | 3 +-
...t-low-bit-mask-and-icmp-sle-to-icmp-sle.ll | 3 +-
...t-low-bit-mask-and-icmp-slt-to-icmp-sgt.ll | 3 +-
...t-low-bit-mask-and-icmp-uge-to-icmp-ule.ll | 3 +-
...t-low-bit-mask-and-icmp-ugt-to-icmp-ugt.ll | 3 +-
...t-low-bit-mask-and-icmp-ule-to-icmp-ule.ll | 3 +-
...t-low-bit-mask-and-icmp-ult-to-icmp-ugt.ll | 3 +-
.../InstCombine/icmp-and-lowbit-mask.ll | 45 ++----
13 files changed, 184 insertions(+), 65 deletions(-)
diff --git a/llvm/include/llvm/IR/PatternMatch.h b/llvm/include/llvm/IR/PatternMatch.h
index 13877538f79de6d..0d50df19a326725 100644
--- a/llvm/include/llvm/IR/PatternMatch.h
+++ b/llvm/include/llvm/IR/PatternMatch.h
@@ -564,6 +564,19 @@ inline api_pred_ty<is_negated_power2> m_NegatedPower2(const APInt *&V) {
return V;
}
+struct is_negated_power2_or_zero {
+ bool isValue(const APInt &C) { return !C || C.isNegatedPowerOf2(); }
+};
+/// Match a integer or vector negated power-of-2.
+/// For vectors, this includes constants with undefined elements.
+inline cst_pred_ty<is_negated_power2_or_zero> m_NegatedPower2OrZero() {
+ return cst_pred_ty<is_negated_power2_or_zero>();
+}
+inline api_pred_ty<is_negated_power2_or_zero>
+m_NegatedPower2OrZero(const APInt *&V) {
+ return V;
+}
+
struct is_power2_or_zero {
bool isValue(const APInt &C) { return !C || C.isPowerOf2(); }
};
@@ -595,6 +608,18 @@ inline cst_pred_ty<is_lowbit_mask> m_LowBitMask() {
}
inline api_pred_ty<is_lowbit_mask> m_LowBitMask(const APInt *&V) { return V; }
+struct is_lowbit_mask_or_zero {
+ bool isValue(const APInt &C) { return !C || C.isMask(); }
+};
+/// Match an integer or vector with only the low bit(s) set.
+/// For vectors, this includes constants with undefined elements.
+inline cst_pred_ty<is_lowbit_mask_or_zero> m_LowBitMaskOrZero() {
+ return cst_pred_ty<is_lowbit_mask_or_zero>();
+}
+inline api_pred_ty<is_lowbit_mask_or_zero> m_LowBitMaskOrZero(const APInt *&V) {
+ return V;
+}
+
struct icmp_pred_with_threshold {
ICmpInst::Predicate Pred;
const APInt *Thr;
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 66e2b6c72cce46c..30e0e3e33e7ec6b 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -4054,6 +4054,109 @@ Instruction *InstCombinerImpl::foldSelectICmp(ICmpInst::Predicate Pred,
return nullptr;
}
+// Returns of V is a Mask ((X + 1) & X == 0) or ~Mask (-Pow2OrZero)
+static bool isMaskOrZero(const Value *V, bool Not, const SimplifyQuery &Q,
+ unsigned Depth = 0) {
+ if (Not ? match(V, m_NegatedPower2OrZero()) : match(V, m_LowBitMaskOrZero()))
+ return true;
+ if (V->getType()->getScalarSizeInBits() == 1)
+ return true;
+ if (Depth++ >= MaxAnalysisRecursionDepth)
+ return false;
+ Value *X;
+ if (match(V, m_Not(m_Value(X))))
+ return isMaskOrZero(X, !Not, Q, Depth);
+ const Operator *I = dyn_cast<Operator>(V);
+ if (I == nullptr)
+ return false;
+ switch (I->getOpcode()) {
+ case Instruction::ZExt:
+ // ZExt(Mask) is a Mask.
+ return !Not && isMaskOrZero(I->getOperand(0), Not, Q, Depth);
+ case Instruction::SExt:
+ // SExt(Mask) is a Mask.
+ // SExt(~Mask) is a ~Mask.
+ return isMaskOrZero(I->getOperand(0), Not, Q, Depth);
+ case Instruction::And:
+ case Instruction::Or:
+ // Mask0 | Mask1 is a Mask.
+ // Mask0 & Mask1 is a Mask.
+ // ~Mask0 | ~Mask1 is a ~Mask.
+ // ~Mask0 & ~Mask1 is a ~Mask.
+ return isMaskOrZero(I->getOperand(1), Not, Q, Depth) &&
+ isMaskOrZero(I->getOperand(0), Not, Q, Depth);
+ case Instruction::Xor:
+ // (X ^ (X - 1)) is a Mask
+ return match(V, m_c_Xor(m_Value(X), m_Add(m_Deferred(X), m_AllOnes())));
+ case Instruction::Select:
+ // c ? Mask0 : Mask1 is a Mask.
+ return isMaskOrZero(I->getOperand(1), Not, Q, Depth) &&
+ isMaskOrZero(I->getOperand(2), Not, Q, Depth);
+ case Instruction::Shl:
+ if (Not) {
+ // (-1 >> X) << X is ~Mask
+ if (match(I->getOperand(0),
+ m_Shr(m_AllOnes(), m_Specific(I->getOperand(1)))))
+ return true;
+
+ // (~Mask) << X is a ~Mask.
+ return isMaskOrZero(I->getOperand(0), Not, Q, Depth);
+ }
+ break;
+ case Instruction::LShr:
+ if (!Not) {
+ // (-1 << X) >> X is a Mask
+ if (match(I->getOperand(0),
+ m_Shl(m_AllOnes(), m_Specific(I->getOperand(1)))))
+ return true;
+ // Mask >> X is a Mask.
+ return isMaskOrZero(I->getOperand(0), Not, Q, Depth);
+ }
+ return false;
+ case Instruction::AShr:
+ // Mask s>> X is a Mask.
+ // ~Mask s>> X is a ~Mask.
+ return isMaskOrZero(I->getOperand(0), Not, Q, Depth);
+ case Instruction::Add:
+ // Pow2 - 1 is a Mask.
+ if (!Not && match(I->getOperand(1), m_AllOnes()))
+ return isKnownToBeAPowerOfTwo(I->getOperand(0), Q.DL, /*OrZero*/ true,
+ Depth, Q.AC, Q.CxtI, Q.DT);
+ break;
+ case Instruction::Sub:
+ // -Pow2 is a ~Mask.
+ if (Not && match(I->getOperand(0), m_Zero()))
+ return isKnownToBeAPowerOfTwo(I->getOperand(1), Q.DL, /*OrZero*/ true,
+ Depth, Q.AC, Q.CxtI, Q.DT);
+ break;
+ case Instruction::Invoke:
+ case Instruction::Call: {
+ if (auto *II = dyn_cast<IntrinsicInst>(I)) {
+ switch (II->getIntrinsicID()) {
+ // min/max(Mask0, Mask1) is a Mask.
+ // min/max(~Mask0, ~Mask1) is a ~Mask.
+ case Intrinsic::umax:
+ case Intrinsic::smax:
+ case Intrinsic::umin:
+ case Intrinsic::smin:
+ return isMaskOrZero(II->getArgOperand(1), Not, Q, Depth) &&
+ isMaskOrZero(II->getArgOperand(0), Not, Q, Depth);
+
+ // In the context of masks, bitreverse(Mask) == ~Mask
+ case Intrinsic::bitreverse:
+ return isMaskOrZero(II->getArgOperand(0), !Not, Q, Depth);
+ default:
+ break;
+ }
+ }
+ break;
+ }
+ default:
+ break;
+ }
+ return false;
+}
+
/// Some comparisons can be simplified.
/// In this case, we are looking for comparisons that look like
/// a check for a lossy truncation.
@@ -4067,21 +4170,35 @@ Instruction *InstCombinerImpl::foldSelectICmp(ICmpInst::Predicate Pred,
/// The Mask can be a constant, too.
/// For some predicates, the operands are commutative.
/// For others, x can only be on a specific side.
-static Value *foldICmpWithLowBitMaskedVal(ICmpInst &I,
- InstCombiner::BuilderTy &Builder) {
+static Value *foldICmpWithLowBitMaskedVal(ICmpInst &I, const SimplifyQuery &Q,
+ InstCombiner &IC) {
+
+ Value *X, *M;
+ ICmpInst::Predicate Pred = I.getPredicate();
ICmpInst::Predicate SrcPred;
- Value *X, *M, *Y;
- auto m_VariableMask = m_CombineOr(
- m_CombineOr(m_Not(m_Shl(m_AllOnes(), m_Value())),
- m_Add(m_Shl(m_One(), m_Value()), m_AllOnes())),
- m_CombineOr(m_LShr(m_AllOnes(), m_Value()),
- m_LShr(m_Shl(m_AllOnes(), m_Value(Y)), m_Deferred(Y))));
- auto m_Mask = m_CombineOr(m_VariableMask, m_LowBitMask());
- if (!match(&I, m_c_ICmp(SrcPred,
- m_c_And(m_CombineAnd(m_Mask, m_Value(M)), m_Value(X)),
- m_Deferred(X))))
- return nullptr;
+ bool NeedsNot = false;
+
+ auto CheckMask = [&](Value *V, bool Not) {
+ if (!ICmpInst::isSigned(Pred))
+ return isMaskOrZero(V, Not, Q);
+ return Not ? match(V, m_NegatedPower2OrZero())
+ : match(V, m_LowBitMaskOrZero());
+ };
+ auto TryMatch = [&](unsigned OpNo) {
+ SrcPred = Pred;
+ if (match(I.getOperand(OpNo),
+ m_c_And(m_Specific(I.getOperand(1 - OpNo)), m_Value(M)))) {
+ X = I.getOperand(1 - OpNo);
+ if (OpNo)
+ SrcPred = ICmpInst::getSwappedPredicate(Pred);
+ return CheckMask(M, /*Not*/ false);
+ }
+ return false;
+ };
+
+ if (!TryMatch(0) && !TryMatch(1))
+ return nullptr;
ICmpInst::Predicate DstPred;
switch (SrcPred) {
case ICmpInst::Predicate::ICMP_EQ:
@@ -4149,7 +4266,9 @@ static Value *foldICmpWithLowBitMaskedVal(ICmpInst &I,
M = Constant::replaceUndefsWith(VecC, SafeReplacementConstant);
}
- return Builder.CreateICmp(DstPred, X, M);
+ if (NeedsNot)
+ M = IC.Builder.CreateNot(M);
+ return IC.Builder.CreateICmp(DstPred, X, M);
}
/// Some comparisons can be simplified.
@@ -5061,7 +5180,7 @@ Instruction *InstCombinerImpl::foldICmpBinOp(ICmpInst &I,
if (Value *V = foldMultiplicationOverflowCheck(I))
return replaceInstUsesWith(I, V);
- if (Value *V = foldICmpWithLowBitMaskedVal(I, Builder))
+ if (Value *V = foldICmpWithLowBitMaskedVal(I, Q, *this))
return replaceInstUsesWith(I, V);
if (Instruction *R = foldICmpAndXX(I, Q, *this))
diff --git a/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-eq-to-icmp-ule.ll b/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-eq-to-icmp-ule.ll
index a957fb2d088ef46..5b7a99d53c308ca 100644
--- a/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-eq-to-icmp-ule.ll
+++ b/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-eq-to-icmp-ule.ll
@@ -62,8 +62,7 @@ define <2 x i1> @p2_vec_nonsplat(<2 x i8> %x) {
define <2 x i1> @p2_vec_nonsplat_edgecase0(<2 x i8> %x) {
; CHECK-LABEL: @p2_vec_nonsplat_edgecase0(
-; CHECK-NEXT: [[TMP1:%.*]] = and <2 x i8> [[X:%.*]], <i8 -4, i8 -1>
-; CHECK-NEXT: [[RET:%.*]] = icmp eq <2 x i8> [[TMP1]], zeroinitializer
+; CHECK-NEXT: [[RET:%.*]] = icmp ult <2 x i8> [[X:%.*]], <i8 4, i8 1>
; CHECK-NEXT: ret <2 x i1> [[RET]]
;
%tmp0 = and <2 x i8> %x, <i8 3, i8 0>
diff --git a/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-ne-to-icmp-ugt.ll b/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-ne-to-icmp-ugt.ll
index 57361cdf38977c1..160d968b9ac4c77 100644
--- a/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-ne-to-icmp-ugt.ll
+++ b/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-ne-to-icmp-ugt.ll
@@ -62,8 +62,7 @@ define <2 x i1> @p2_vec_nonsplat(<2 x i8> %x) {
define <2 x i1> @p2_vec_nonsplat_edgecase0(<2 x i8> %x) {
; CHECK-LABEL: @p2_vec_nonsplat_edgecase0(
-; CHECK-NEXT: [[TMP1:%.*]] = and <2 x i8> [[X:%.*]], <i8 -4, i8 -1>
-; CHECK-NEXT: [[RET:%.*]] = icmp ne <2 x i8> [[TMP1]], zeroinitializer
+; CHECK-NEXT: [[RET:%.*]] = icmp ugt <2 x i8> [[X:%.*]], <i8 3, i8 0>
; CHECK-NEXT: ret <2 x i1> [[RET]]
;
%tmp0 = and <2 x i8> %x, <i8 3, i8 0>
diff --git a/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-sge-to-icmp-sle.ll b/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-sge-to-icmp-sle.ll
index 0dfc9f51baf9c26..60921042d524356 100644
--- a/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-sge-to-icmp-sle.ll
+++ b/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-sge-to-icmp-sle.ll
@@ -50,8 +50,7 @@ define <2 x i1> @p2_vec_nonsplat(<2 x i8> %x) {
define <2 x i1> @p2_vec_nonsplat_edgecase(<2 x i8> %x) {
; CHECK-LABEL: @p2_vec_nonsplat_edgecase(
-; CHECK-NEXT: [[TMP0:%.*]] = and <2 x i8> [[X:%.*]], <i8 3, i8 0>
-; CHECK-NEXT: [[RET:%.*]] = icmp sge <2 x i8> [[TMP0]], [[X]]
+; CHECK-NEXT: [[RET:%.*]] = icmp slt <2 x i8> [[X:%.*]], <i8 4, i8 1>
; CHECK-NEXT: ret <2 x i1> [[RET]]
;
%tmp0 = and <2 x i8> %x, <i8 3, i8 0>
diff --git a/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-sgt-to-icmp-sgt.ll b/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-sgt-to-icmp-sgt.ll
index e0893ce4cf2ecb1..6345e70d7220e29 100644
--- a/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-sgt-to-icmp-sgt.ll
+++ b/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-sgt-to-icmp-sgt.ll
@@ -63,8 +63,7 @@ define <2 x i1> @p2_vec_nonsplat() {
define <2 x i1> @p2_vec_nonsplat_edgecase() {
; CHECK-LABEL: @p2_vec_nonsplat_edgecase(
; CHECK-NEXT: [[X:%.*]] = call <2 x i8> @gen2x8()
-; CHECK-NEXT: [[TMP0:%.*]] = and <2 x i8> [[X]], <i8 3, i8 0>
-; CHECK-NEXT: [[RET:%.*]] = icmp sgt <2 x i8> [[X]], [[TMP0]]
+; CHECK-NEXT: [[RET:%.*]] = icmp sgt <2 x i8> [[X]], <i8 3, i8 0>
; CHECK-NEXT: ret <2 x i1> [[RET]]
;
%x = call <2 x i8> @gen2x8()
diff --git a/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-sle-to-icmp-sle.ll b/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-sle-to-icmp-sle.ll
index 81887a390915730..b7aec53fed6760f 100644
--- a/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-sle-to-icmp-sle.ll
+++ b/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-sle-to-icmp-sle.ll
@@ -63,8 +63,7 @@ define <2 x i1> @p2_vec_nonsplat() {
define <2 x i1> @p2_vec_nonsplat_edgecase() {
; CHECK-LABEL: @p2_vec_nonsplat_edgecase(
; CHECK-NEXT: [[X:%.*]] = call <2 x i8> @gen2x8()
-; CHECK-NEXT: [[TMP0:%.*]] = and <2 x i8> [[X]], <i8 3, i8 0>
-; CHECK-NEXT: [[RET:%.*]] = icmp sle <2 x i8> [[X]], [[TMP0]]
+; CHECK-NEXT: [[RET:%.*]] = icmp slt <2 x i8> [[X]], <i8 4, i8 1>
; CHECK-NEXT: ret <2 x i1> [[RET]]
;
%x = call <2 x i8> @gen2x8()
diff --git a/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-slt-to-icmp-sgt.ll b/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-slt-to-icmp-sgt.ll
index 8ce8687f1984462..56661d335c4f602 100644
--- a/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-slt-to-icmp-sgt.ll
+++ b/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-slt-to-icmp-sgt.ll
@@ -50,8 +50,7 @@ define <2 x i1> @p2_vec_nonsplat(<2 x i8> %x) {
define <2 x i1> @p2_vec_nonsplat_edgecase(<2 x i8> %x) {
; CHECK-LABEL: @p2_vec_nonsplat_edgecase(
-; CHECK-NEXT: [[TMP0:%.*]] = and <2 x i8> [[X:%.*]], <i8 3, i8 0>
-; CHECK-NEXT: [[RET:%.*]] = icmp slt <2 x i8> [[TMP0]], [[X]]
+; CHECK-NEXT: [[RET:%.*]] = icmp sgt <2 x i8> [[X:%.*]], <i8 3, i8 0>
; CHECK-NEXT: ret <2 x i1> [[RET]]
;
%tmp0 = and <2 x i8> %x, <i8 3, i8 0>
diff --git a/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-uge-to-icmp-ule.ll b/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-uge-to-icmp-ule.ll
index ff09e255185b5ad..a93e8f779435fc8 100644
--- a/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-uge-to-icmp-ule.ll
+++ b/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-uge-to-icmp-ule.ll
@@ -62,8 +62,7 @@ define <2 x i1> @p2_vec_nonsplat(<2 x i8> %x) {
define <2 x i1> @p2_vec_nonsplat_edgecase0(<2 x i8> %x) {
; CHECK-LABEL: @p2_vec_nonsplat_edgecase0(
-; CHECK-NEXT: [[TMP1:%.*]] = and <2 x i8> [[X:%.*]], <i8 -4, i8 -1>
-; CHECK-NEXT: [[RET:%.*]] = icmp eq <2 x i8> [[TMP1]], zeroinitializer
+; CHECK-NEXT: [[RET:%.*]] = icmp ult <2 x i8> [[X:%.*]], <i8 4, i8 1>
; CHECK-NEXT: ret <2 x i1> [[RET]]
;
%tmp0 = and <2 x i8> %x, <i8 3, i8 0>
diff --git a/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-ugt-to-icmp-ugt.ll b/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-ugt-to-icmp-ugt.ll
index 4ad04710fd7bb96..73ea4d456d2462e 100644
--- a/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-ugt-to-icmp-ugt.ll
+++ b/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-ugt-to-icmp-ugt.ll
@@ -75,8 +75,7 @@ define <2 x i1> @p2_vec_nonsplat() {
define <2 x i1> @p2_vec_nonsplat_edgecase0() {
; CHECK-LABEL: @p2_vec_nonsplat_edgecase0(
; CHECK-NEXT: [[X:%.*]] = call <2 x i8> @gen2x8()
-; CHECK-NEXT: [[TMP1:%.*]] = and <2 x i8> [[X]], <i8 -4, i8 -1>
-; CHECK-NEXT: [[RET:%.*]] = icmp ne <2 x i8> [[TMP1]], zeroinitializer
+; CHECK-NEXT: [[RET:%.*]] = icmp ugt <2 x i8> [[X]], <i8 3, i8 0>
; CHECK-NEXT: ret <2 x i1> [[RET]]
;
%x = call <2 x i8> @gen2x8()
diff --git a/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-ule-to-icmp-ule.ll b/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-ule-to-icmp-ule.ll
index 8e513dcbf4ef3a3..53886b5f2dc9c33 100644
--- a/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-ule-to-icmp-ule.ll
+++ b/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-ule-to-icmp-ule.ll
@@ -75,8 +75,7 @@ define <2 x i1> @p2_vec_nonsplat() {
define <2 x i1> @p2_vec_nonsplat_edgecase0() {
; CHECK-LABEL: @p2_vec_nonsplat_edgecase0(
; CHECK-NEXT: [[X:%.*]] = call <2 x i8> @gen2x8()
-; CHECK-NEXT: [[TMP1:%.*]] = and <2 x i8> [[X]], <i8 -4, i8 -1>
-; CHECK-NEXT: [[RET:%.*]] = icmp eq <2 x i8> [[TMP1]], zeroinitializer
+; CHECK-NEXT: [[RET:%.*]] = icmp ult <2 x i8> [[X]], <i8 4, i8 1>
; CHECK-NEXT: ret <2 x i1> [[RET]]
;
%x = call <2 x i8> @gen2x8()
diff --git a/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-ult-to-icmp-ugt.ll b/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-ult-to-icmp-ugt.ll
index d02ecf6965e878f..d66be571008c2f7 100644
--- a/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-ult-to-icmp-ugt.ll
+++ b/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-ult-to-icmp-ugt.ll
@@ -62,8 +62,7 @@ define <2 x i1> @p2_vec_nonsplat(<2 x i8> %x) {
define <2 x i1> @p2_vec_nonsplat_edgecase0(<2 x i8> %x) {
; CHECK-LABEL: @p2_vec_nonsplat_edgecase0(
-; CHECK-NEXT: [[TMP1:%.*]] = and <2 x i8> [[X:%.*]], <i8 -4, i8 -1>
-; CHECK-NEXT: [[RET:%.*]] = icmp ne <2 x i8> [[TMP1]], zeroinitializer
+; CHECK-NEXT: [[RET:%.*]] = icmp ugt <2 x i8> [[X:%.*]], <i8 3, i8 0>
; CHECK-NEXT: ret <2 x i1> [[RET]]
;
%tmp0 = and <2 x i8> %x, <i8 3, i8 0>
diff --git a/llvm/test/Transforms/InstCombine/icmp-and-lowbit-mask.ll b/llvm/test/Transforms/InstCombine/icmp-and-lowbit-mask.ll
index d815183577bbcbf..c1990c3e061154c 100644
--- a/llvm/test/Transforms/InstCombine/icmp-and-lowbit-mask.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-and-lowbit-mask.ll
@@ -15,8 +15,7 @@ define i1 @src_is_mask_zext(i16 %x_in, i8 %y) {
; CHECK-NEXT: [[X:%.*]] = xor i16 [[X_IN:%.*]], 123
; CHECK-NEXT: [[M_IN:%.*]] = lshr i8 -1, [[Y:%.*]]
; CHECK-NEXT: [[MASK:%.*]] = zext i8 [[M_IN]] to i16
-; CHECK-NEXT: [[AND:%.*]] = and i16 [[X]], [[MASK]]
-; CHECK-NEXT: [[R:%.*]] = icmp eq i16 [[AND]], [[X]]
+; CHECK-NEXT: [[R:%.*]] = icmp ule i16 [[X]], [[MASK]]
; CHECK-NEXT: ret i1 [[R]]
;
%x = xor i16 %x_in, 123
@@ -94,8 +93,7 @@ define i1 @src_is_mask_and(i8 %x_in, i8 %y, i8 %z) {
; CHECK-NEXT: [[MY:%.*]] = lshr i8 7, [[Y:%.*]]
; CHECK-NEXT: [[MZ:%.*]] = lshr i8 -1, [[Z:%.*]]
; CHECK-NEXT: [[MASK:%.*]] = and i8 [[MY]], [[MZ]]
-; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], [[MASK]]
-; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[X]], [[AND]]
+; CHECK-NEXT: [[R:%.*]] = icmp ule i8 [[X]], [[MASK]]
; CHECK-NEXT: ret i1 [[R]]
;
%x = xor i8 %x_in, 123
@@ -133,8 +131,7 @@ define i1 @src_is_mask_or(i8 %x_in, i8 %y) {
; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
; CHECK-NEXT: [[MY:%.*]] = lshr i8 -1, [[Y:%.*]]
; CHECK-NEXT: [[MASK:%.*]] = and i8 [[MY]], 7
-; CHECK-NEXT: [[AND:%.*]] = and i8 [[MASK]], [[X]]
-; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[X]], [[AND]]
+; CHECK-NEXT: [[R:%.*]] = icmp ule i8 [[X]], [[MASK]]
; CHECK-NEXT: ret i1 [[R]]
;
%x = xor i8 %x_in, 123
@@ -151,8 +148,7 @@ define i1 @src_is_mask_xor(i8 %x_in, i8 %y) {
; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
; CHECK-NEXT: [[Y_M1:%.*]] = add i8 [[Y:%.*]], -1
; CHECK-NEXT: [[MASK:%.*]] = xor i8 [[Y_M1]], [[Y]]
-; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], [[MASK]]
-; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[AND]], [[X]]
+; CHECK-NEXT: [[R:%.*]] = icmp ugt i8 [[X]], [[MASK]]
; CHECK-NEXT: ret i1 [[R]]
;
%x = xor i8 %x_in, 123
@@ -187,8 +183,7 @@ define i1 @src_is_mask_select(i8 %x_in, i8 %y, i1 %cond) {
; CHECK-NEXT: [[Y_M1:%.*]] = add i8 [[Y:%.*]], -1
; CHECK-NEXT: [[YMASK:%.*]] = xor i8 [[Y_M1]], [[Y]]
; CHECK-NEXT: [[MASK:%.*]] = select i1 [[COND:%.*]], i8 [[YMASK]], i8 15
-; CHECK-NEXT: [[AND:%.*]] = and i8 [[MASK]], [[X]]
-; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[AND]], [[X]]
+; CHECK-NEXT: [[R:%.*]] = icmp ugt i8 [[X]], [[MASK]]
; CHECK-NEXT: ret i1 [[R]]
;
%x = xor i8 %x_in, 123
@@ -267,8 +262,7 @@ define i1 @src_is_mask_lshr(i8 %x_in, i8 %y, i8 %z, i1 %cond) {
; CHECK-NEXT: [[YMASK:%.*]] = xor i8 [[Y_M1]], [[Y]]
; CHECK-NEXT: [[SMASK:%.*]] = select i1 [[COND:%.*]], i8 [[YMASK]], i8 15
; CHECK-NEXT: [[MASK:%.*]] = lshr i8 [[SMASK]], [[Z:%.*]]
-; CHECK-NEXT: [[AND:%.*]] = and i8 [[MASK]], [[X]]
-; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[X]], [[AND]]
+; CHECK-NEXT: [[R:%.*]] = icmp ugt i8 [[X]], [[MASK]]
; CHECK-NEXT: ret i1 [[R]]
;
%x = xor i8 %x_in, 123
@@ -288,8 +282,7 @@ define i1 @src_is_mask_ashr(i8 %x_in, i8 %y, i8 %z, i1 %cond) {
; CHECK-NEXT: [[YMASK:%.*]] = xor i8 [[Y_M1]], [[Y]]
; CHECK-NEXT: [[SMASK:%.*]] = select i1 [[COND:%.*]], i8 [[YMASK]], i8 15
; CHECK-NEXT: [[MASK:%.*]] = ashr i8 [[SMASK]], [[Z:%.*]]
-; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], [[MASK]]
-; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[AND]], [[X]]
+; CHECK-NEXT: [[R:%.*]] = icmp ugt i8 [[X]], [[MASK]]
; CHECK-NEXT: ret i1 [[R]]
;
%x = xor i8 %x_in, 123
@@ -307,8 +300,7 @@ define i1 @src_is_mask_p2_m1(i8 %x_in, i8 %y) {
; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
; CHECK-NEXT: [[P2ORZ:%.*]] = shl i8 2, [[Y:%.*]]
; CHECK-NEXT: [[MASK:%.*]] = add i8 [[P2ORZ]], -1
-; CHECK-NEXT: [[AND:%.*]] = and i8 [[MASK]], [[X]]
-; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[AND]], [[X]]
+; CHECK-NEXT: [[R:%.*]] = icmp ugt i8 [[X]], [[MASK]]
; CHECK-NEXT: ret i1 [[R]]
;
%x = xor i8 %x_in, 123
@@ -325,8 +317,7 @@ define i1 @src_is_mask_umax(i8 %x_in, i8 %y) {
; CHECK-NEXT: [[Y_M1:%.*]] = add i8 [[Y:%.*]], -1
; CHECK-NEXT: [[YMASK:%.*]] = xor i8 [[Y_M1]], [[Y]]
; CHECK-NEXT: [[MASK:%.*]] = call i8 @llvm.umax.i8(i8 [[YMASK]], i8 3)
-; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], [[MASK]]
-; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[AND]], [[X]]
+; CHECK-NEXT: [[R:%.*]] = icmp ugt i8 [[X]], [[MASK]]
; CHECK-NEXT: ret i1 [[R]]
;
%x = xor i8 %x_in, 123
@@ -346,8 +337,7 @@ define i1 @src_is_mask_umin(i8 %x_in, i8 %y, i8 %z) {
; CHECK-NEXT: [[YMASK:%.*]] = xor i8 [[Y_M1]], [[Y]]
; CHECK-NEXT: [[ZMASK:%.*]] = lshr i8 15, [[Z:%.*]]
; CHECK-NEXT: [[MASK:%.*]] = call i8 @llvm.umin.i8(i8 [[YMASK]], i8 [[ZMASK]])
-; CHECK-NEXT: [[AND:%.*]] = and i8 [[MASK]], [[X]]
-; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[AND]], [[X]]
+; CHECK-NEXT: [[R:%.*]] = icmp ugt i8 [[X]], [[MASK]]
; CHECK-NEXT: ret i1 [[R]]
;
%x = xor i8 %x_in, 123
@@ -387,8 +377,7 @@ define i1 @src_is_mask_smax(i8 %x_in, i8 %y) {
; CHECK-NEXT: [[Y_M1:%.*]] = add i8 [[Y:%.*]], -1
; CHECK-NEXT: [[YMASK:%.*]] = xor i8 [[Y_M1]], [[Y]]
; CHECK-NEXT: [[MASK:%.*]] = call i8 @llvm.smax.i8(i8 [[YMASK]], i8 -1)
-; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], [[MASK]]
-; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[AND]], [[X]]
+; CHECK-NEXT: [[R:%.*]] = icmp ule i8 [[X]], [[MASK]]
; CHECK-NEXT: ret i1 [[R]]
;
%x = xor i8 %x_in, 123
@@ -407,8 +396,7 @@ define i1 @src_is_mask_smin(i8 %x_in, i8 %y) {
; CHECK-NEXT: [[Y_M1:%.*]] = add i8 [[Y:%.*]], -1
; CHECK-NEXT: [[YMASK:%.*]] = xor i8 [[Y_M1]], [[Y]]
; CHECK-NEXT: [[MASK:%.*]] = call i8 @llvm.smin.i8(i8 [[YMASK]], i8 0)
-; CHECK-NEXT: [[AND:%.*]] = and i8 [[MASK]], [[X]]
-; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[AND]], [[X]]
+; CHECK-NEXT: [[R:%.*]] = icmp ule i8 [[X]], [[MASK]]
; CHECK-NEXT: ret i1 [[R]]
;
%x = xor i8 %x_in, 123
@@ -426,8 +414,7 @@ define i1 @src_is_mask_bitreverse_not_mask(i8 %x_in, i8 %y) {
; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
; CHECK-NEXT: [[NMASK:%.*]] = shl i8 -1, [[Y:%.*]]
; CHECK-NEXT: [[MASK:%.*]] = call i8 @llvm.bitreverse.i8(i8 [[NMASK]])
-; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], [[MASK]]
-; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[AND]], [[X]]
+; CHECK-NEXT: [[R:%.*]] = icmp ule i8 [[X]], [[MASK]]
; CHECK-NEXT: ret i1 [[R]]
;
%x = xor i8 %x_in, 123
@@ -445,8 +432,7 @@ define i1 @src_is_notmask_sext(i16 %x_in, i8 %y) {
; CHECK-NEXT: [[M_IN:%.*]] = shl i8 -8, [[Y:%.*]]
; CHECK-NEXT: [[TMP1:%.*]] = xor i8 [[M_IN]], -1
; CHECK-NEXT: [[MASK:%.*]] = sext i8 [[TMP1]] to i16
-; CHECK-NEXT: [[AND:%.*]] = and i16 [[X]], [[MASK]]
-; CHECK-NEXT: [[R:%.*]] = icmp eq i16 [[AND]], [[X]]
+; CHECK-NEXT: [[R:%.*]] = icmp ule i16 [[X]], [[MASK]]
; CHECK-NEXT: ret i1 [[R]]
;
%x = xor i16 %x_in, 123
@@ -544,8 +530,7 @@ define i1 @src_is_notmask_ashr(i16 %x_in, i8 %y, i16 %z) {
; CHECK-NEXT: [[NMASK:%.*]] = sext i8 [[M_IN]] to i16
; CHECK-NEXT: [[NMASK_SHR:%.*]] = ashr i16 [[NMASK]], [[Z:%.*]]
; CHECK-NEXT: [[MASK:%.*]] = xor i16 [[NMASK_SHR]], -1
-; CHECK-NEXT: [[AND:%.*]] = and i16 [[X]], [[MASK]]
-; CHECK-NEXT: [[R:%.*]] = icmp eq i16 [[X]], [[AND]]
+; CHECK-NEXT: [[R:%.*]] = icmp ule i16 [[X]], [[MASK]]
; CHECK-NEXT: ret i1 [[R]]
;
%x = xor i16 %x_in, 123
>From ecf16c7303178be4f33a9eeb519743902fff708c Mon Sep 17 00:00:00 2001
From: Noah Goldstein <goldstein.w.n at gmail.com>
Date: Wed, 13 Sep 2023 13:45:55 -0500
Subject: [PATCH 06/13] [InstCombine] Recognize `(icmp eq/ne (and X, ~Mask),
0)` pattern in `foldICmpWithLowBitMaskedVal`
`(icmp eq/ne (and X, ~Mask), 0)` is equivilent to `(icmp eq/ne (and X,
Mask), X` and we sometimes generate the former pattern intentionally
to reduce number of uses of `X`.
Proof: https://alive2.llvm.org/ce/z/3u-usC
Differential Revision: https://reviews.llvm.org/D159329
---
.../InstCombine/InstCombineCompares.cpp | 11 ++++++++
.../InstCombine/icmp-and-lowbit-mask.ll | 27 +++++++------------
.../InstCombine/lshr-and-negC-icmpeq-zero.ll | 9 +++----
.../lshr-and-signbit-icmpeq-zero.ll | 9 +++----
.../InstCombine/shl-and-negC-icmpeq-zero.ll | 9 +++----
.../shl-and-signbit-icmpeq-zero.ll | 9 +++----
6 files changed, 32 insertions(+), 42 deletions(-)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 30e0e3e33e7ec6b..74279d47ac913fd 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -4162,6 +4162,7 @@ static bool isMaskOrZero(const Value *V, bool Not, const SimplifyQuery &Q,
/// a check for a lossy truncation.
/// Folds:
/// icmp SrcPred (x & Mask), x to icmp DstPred x, Mask
+/// icmp eq/ne (x & ~Mask), 0 to icmp DstPred x, Mask
/// Where Mask is some pattern that produces all-ones in low bits:
/// (-1 >> y)
/// ((-1 << y) >> y) <- non-canonical, has extra uses
@@ -4194,6 +4195,16 @@ static Value *foldICmpWithLowBitMaskedVal(ICmpInst &I, const SimplifyQuery &Q,
SrcPred = ICmpInst::getSwappedPredicate(Pred);
return CheckMask(M, /*Not*/ false);
}
+ if (OpNo == 1 && match(I.getOperand(1), m_Zero()) &&
+ ICmpInst::isEquality(Pred) &&
+ match(I.getOperand(0), m_OneUse(m_And(m_Value(X), m_Value(M))))) {
+ NeedsNot = true;
+ if (IC.isFreeToInvert(X, X->hasOneUse()) && CheckMask(X, /*Not*/ true)) {
+ std::swap(X, M);
+ return true;
+ }
+ return IC.isFreeToInvert(M, M->hasOneUse()) && CheckMask(M, /*Not*/ true);
+ }
return false;
};
diff --git a/llvm/test/Transforms/InstCombine/icmp-and-lowbit-mask.ll b/llvm/test/Transforms/InstCombine/icmp-and-lowbit-mask.ll
index c1990c3e061154c..5578b9a66e1f219 100644
--- a/llvm/test/Transforms/InstCombine/icmp-and-lowbit-mask.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-and-lowbit-mask.ll
@@ -50,9 +50,7 @@ define i1 @src_is_mask_sext(i16 %x_in, i8 %y) {
; CHECK-NEXT: [[X:%.*]] = xor i16 [[X_IN:%.*]], 123
; CHECK-NEXT: [[M_IN:%.*]] = lshr i8 31, [[Y:%.*]]
; CHECK-NEXT: [[MASK:%.*]] = zext i8 [[M_IN]] to i16
-; CHECK-NEXT: [[NOTMASK:%.*]] = xor i16 [[MASK]], -1
-; CHECK-NEXT: [[AND:%.*]] = and i16 [[X]], [[NOTMASK]]
-; CHECK-NEXT: [[R:%.*]] = icmp eq i16 [[AND]], 0
+; CHECK-NEXT: [[R:%.*]] = icmp ule i16 [[X]], [[MASK]]
; CHECK-NEXT: ret i1 [[R]]
;
%x = xor i16 %x_in, 123
@@ -220,9 +218,7 @@ define i1 @src_is_mask_shl_lshr(i8 %x_in, i8 %y, i1 %cond) {
; CHECK-LABEL: @src_is_mask_shl_lshr(
; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 122
; CHECK-NEXT: [[TMP1:%.*]] = lshr i8 -1, [[Y:%.*]]
-; CHECK-NEXT: [[NOTMASK:%.*]] = xor i8 [[TMP1]], -1
-; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], [[NOTMASK]]
-; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[AND]], 0
+; CHECK-NEXT: [[R:%.*]] = icmp ugt i8 [[X]], [[TMP1]]
; CHECK-NEXT: ret i1 [[R]]
;
%x = xor i8 %x_in, 123
@@ -449,10 +445,8 @@ define i1 @src_is_notmask_shl(i8 %x_in, i8 %y, i1 %cond) {
; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 122
; CHECK-NEXT: [[NMASK:%.*]] = shl i8 -1, [[Y:%.*]]
; CHECK-NEXT: [[MASK:%.*]] = call i8 @llvm.bitreverse.i8(i8 [[NMASK]])
-; CHECK-NEXT: [[NOTMASK0:%.*]] = xor i8 [[MASK]], -1
-; CHECK-NEXT: [[NOTMASK:%.*]] = select i1 [[COND:%.*]], i8 [[NOTMASK0]], i8 -8
-; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], [[NOTMASK]]
-; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[AND]], 0
+; CHECK-NEXT: [[TMP1:%.*]] = select i1 [[COND:%.*]], i8 [[MASK]], i8 7
+; CHECK-NEXT: [[R:%.*]] = icmp ule i8 [[X]], [[TMP1]]
; CHECK-NEXT: ret i1 [[R]]
;
%x = xor i8 %x_in, 123
@@ -550,9 +544,7 @@ define i1 @src_is_notmask_neg_p2(i8 %x_in, i8 %y) {
; CHECK-NEXT: [[P2:%.*]] = and i8 [[NY]], [[Y]]
; CHECK-NEXT: [[NMASK:%.*]] = sub i8 0, [[P2]]
; CHECK-NEXT: [[MASK:%.*]] = call i8 @llvm.bitreverse.i8(i8 [[NMASK]])
-; CHECK-NEXT: [[NOTMASK:%.*]] = xor i8 [[MASK]], -1
-; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], [[NOTMASK]]
-; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[AND]], 0
+; CHECK-NEXT: [[R:%.*]] = icmp ule i8 [[X]], [[MASK]]
; CHECK-NEXT: ret i1 [[R]]
;
%x = xor i8 %x_in, 123
@@ -569,11 +561,10 @@ define i1 @src_is_notmask_neg_p2(i8 %x_in, i8 %y) {
define i1 @src_is_notmask_neg_p2_fail_not_invertable(i8 %x_in, i8 %y) {
; CHECK-LABEL: @src_is_notmask_neg_p2_fail_not_invertable(
; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
-; CHECK-NEXT: [[NY:%.*]] = sub i8 0, [[Y:%.*]]
-; CHECK-NEXT: [[P2:%.*]] = and i8 [[NY]], [[Y]]
-; CHECK-NEXT: [[NOTMASK:%.*]] = sub i8 0, [[P2]]
-; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], [[NOTMASK]]
-; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[AND]], 0
+; CHECK-NEXT: [[TMP1:%.*]] = add i8 [[Y:%.*]], -1
+; CHECK-NEXT: [[TMP2:%.*]] = xor i8 [[Y]], -1
+; CHECK-NEXT: [[TMP3:%.*]] = and i8 [[TMP1]], [[TMP2]]
+; CHECK-NEXT: [[R:%.*]] = icmp ule i8 [[X]], [[TMP3]]
; CHECK-NEXT: ret i1 [[R]]
;
%x = xor i8 %x_in, 123
diff --git a/llvm/test/Transforms/InstCombine/lshr-and-negC-icmpeq-zero.ll b/llvm/test/Transforms/InstCombine/lshr-and-negC-icmpeq-zero.ll
index 79aef3a5406cfb0..847a7940bad8c74 100644
--- a/llvm/test/Transforms/InstCombine/lshr-and-negC-icmpeq-zero.ll
+++ b/llvm/test/Transforms/InstCombine/lshr-and-negC-icmpeq-zero.ll
@@ -84,8 +84,7 @@ define <4 x i1> @vec_4xi32_lshr_and_negC_eq(<4 x i32> %x, <4 x i32> %y) {
define <4 x i1> @vec_lshr_and_negC_eq_undef1(<4 x i32> %x, <4 x i32> %y) {
; CHECK-LABEL: @vec_lshr_and_negC_eq_undef1(
; CHECK-NEXT: [[LSHR:%.*]] = lshr <4 x i32> [[X:%.*]], [[Y:%.*]]
-; CHECK-NEXT: [[AND:%.*]] = and <4 x i32> [[LSHR]], <i32 -8, i32 undef, i32 -8, i32 -8>
-; CHECK-NEXT: [[R:%.*]] = icmp eq <4 x i32> [[AND]], zeroinitializer
+; CHECK-NEXT: [[R:%.*]] = icmp ult <4 x i32> [[LSHR]], <i32 8, i32 8, i32 8, i32 8>
; CHECK-NEXT: ret <4 x i1> [[R]]
;
%lshr = lshr <4 x i32> %x, %y
@@ -97,8 +96,7 @@ define <4 x i1> @vec_lshr_and_negC_eq_undef1(<4 x i32> %x, <4 x i32> %y) {
define <4 x i1> @vec_lshr_and_negC_eq_undef2(<4 x i32> %x, <4 x i32> %y) {
; CHECK-LABEL: @vec_lshr_and_negC_eq_undef2(
; CHECK-NEXT: [[LSHR:%.*]] = lshr <4 x i32> [[X:%.*]], [[Y:%.*]]
-; CHECK-NEXT: [[AND:%.*]] = and <4 x i32> [[LSHR]], <i32 -8, i32 -8, i32 -8, i32 -8>
-; CHECK-NEXT: [[R:%.*]] = icmp eq <4 x i32> [[AND]], <i32 0, i32 0, i32 0, i32 undef>
+; CHECK-NEXT: [[R:%.*]] = icmp ult <4 x i32> [[LSHR]], <i32 8, i32 8, i32 8, i32 8>
; CHECK-NEXT: ret <4 x i1> [[R]]
;
%lshr = lshr <4 x i32> %x, %y
@@ -110,8 +108,7 @@ define <4 x i1> @vec_lshr_and_negC_eq_undef2(<4 x i32> %x, <4 x i32> %y) {
define <4 x i1> @vec_lshr_and_negC_eq_undef3(<4 x i32> %x, <4 x i32> %y) {
; CHECK-LABEL: @vec_lshr_and_negC_eq_undef3(
; CHECK-NEXT: [[LSHR:%.*]] = lshr <4 x i32> [[X:%.*]], [[Y:%.*]]
-; CHECK-NEXT: [[AND:%.*]] = and <4 x i32> [[LSHR]], <i32 -8, i32 -8, i32 undef, i32 -8>
-; CHECK-NEXT: [[R:%.*]] = icmp eq <4 x i32> [[AND]], <i32 0, i32 0, i32 0, i32 undef>
+; CHECK-NEXT: [[R:%.*]] = icmp ult <4 x i32> [[LSHR]], <i32 8, i32 8, i32 8, i32 8>
; CHECK-NEXT: ret <4 x i1> [[R]]
;
%lshr = lshr <4 x i32> %x, %y
diff --git a/llvm/test/Transforms/InstCombine/lshr-and-signbit-icmpeq-zero.ll b/llvm/test/Transforms/InstCombine/lshr-and-signbit-icmpeq-zero.ll
index 5335a4736896d7d..39f4e58b25dc849 100644
--- a/llvm/test/Transforms/InstCombine/lshr-and-signbit-icmpeq-zero.ll
+++ b/llvm/test/Transforms/InstCombine/lshr-and-signbit-icmpeq-zero.ll
@@ -84,8 +84,7 @@ define <4 x i1> @vec_4xi32_lshr_and_signbit_eq(<4 x i32> %x, <4 x i32> %y) {
define <4 x i1> @vec_4xi32_lshr_and_signbit_eq_undef1(<4 x i32> %x, <4 x i32> %y) {
; CHECK-LABEL: @vec_4xi32_lshr_and_signbit_eq_undef1(
; CHECK-NEXT: [[LSHR:%.*]] = lshr <4 x i32> [[X:%.*]], [[Y:%.*]]
-; CHECK-NEXT: [[AND:%.*]] = and <4 x i32> [[LSHR]], <i32 -2147483648, i32 undef, i32 -2147483648, i32 -2147483648>
-; CHECK-NEXT: [[R:%.*]] = icmp eq <4 x i32> [[AND]], zeroinitializer
+; CHECK-NEXT: [[R:%.*]] = icmp sgt <4 x i32> [[LSHR]], <i32 -1, i32 -1, i32 -1, i32 -1>
; CHECK-NEXT: ret <4 x i1> [[R]]
;
%lshr = lshr <4 x i32> %x, %y
@@ -97,8 +96,7 @@ define <4 x i1> @vec_4xi32_lshr_and_signbit_eq_undef1(<4 x i32> %x, <4 x i32> %y
define <4 x i1> @vec_4xi32_lshr_and_signbit_eq_undef2(<4 x i32> %x, <4 x i32> %y) {
; CHECK-LABEL: @vec_4xi32_lshr_and_signbit_eq_undef2(
; CHECK-NEXT: [[LSHR:%.*]] = lshr <4 x i32> [[X:%.*]], [[Y:%.*]]
-; CHECK-NEXT: [[AND:%.*]] = and <4 x i32> [[LSHR]], <i32 -2147483648, i32 -2147483648, i32 -2147483648, i32 -2147483648>
-; CHECK-NEXT: [[R:%.*]] = icmp eq <4 x i32> [[AND]], <i32 undef, i32 0, i32 0, i32 0>
+; CHECK-NEXT: [[R:%.*]] = icmp sgt <4 x i32> [[LSHR]], <i32 -1, i32 -1, i32 -1, i32 -1>
; CHECK-NEXT: ret <4 x i1> [[R]]
;
%lshr = lshr <4 x i32> %x, %y
@@ -110,8 +108,7 @@ define <4 x i1> @vec_4xi32_lshr_and_signbit_eq_undef2(<4 x i32> %x, <4 x i32> %y
define <4 x i1> @vec_4xi32_lshr_and_signbit_eq_undef3(<4 x i32> %x, <4 x i32> %y) {
; CHECK-LABEL: @vec_4xi32_lshr_and_signbit_eq_undef3(
; CHECK-NEXT: [[LSHR:%.*]] = lshr <4 x i32> [[X:%.*]], [[Y:%.*]]
-; CHECK-NEXT: [[AND:%.*]] = and <4 x i32> [[LSHR]], <i32 -2147483648, i32 undef, i32 -2147483648, i32 -2147483648>
-; CHECK-NEXT: [[R:%.*]] = icmp eq <4 x i32> [[AND]], <i32 0, i32 0, i32 0, i32 undef>
+; CHECK-NEXT: [[R:%.*]] = icmp sgt <4 x i32> [[LSHR]], <i32 -1, i32 -1, i32 -1, i32 -1>
; CHECK-NEXT: ret <4 x i1> [[R]]
;
%lshr = lshr <4 x i32> %x, %y
diff --git a/llvm/test/Transforms/InstCombine/shl-and-negC-icmpeq-zero.ll b/llvm/test/Transforms/InstCombine/shl-and-negC-icmpeq-zero.ll
index d8e7fe2e2a2c163..406dc72f2646e5b 100644
--- a/llvm/test/Transforms/InstCombine/shl-and-negC-icmpeq-zero.ll
+++ b/llvm/test/Transforms/InstCombine/shl-and-negC-icmpeq-zero.ll
@@ -84,8 +84,7 @@ define <4 x i1> @vec_4xi32_shl_and_negC_eq(<4 x i32> %x, <4 x i32> %y) {
define <4 x i1> @vec_shl_and_negC_eq_undef1(<4 x i32> %x, <4 x i32> %y) {
; CHECK-LABEL: @vec_shl_and_negC_eq_undef1(
; CHECK-NEXT: [[SHL:%.*]] = shl <4 x i32> [[X:%.*]], [[Y:%.*]]
-; CHECK-NEXT: [[AND:%.*]] = and <4 x i32> [[SHL]], <i32 -8, i32 undef, i32 -8, i32 -8>
-; CHECK-NEXT: [[R:%.*]] = icmp eq <4 x i32> [[AND]], zeroinitializer
+; CHECK-NEXT: [[R:%.*]] = icmp ult <4 x i32> [[SHL]], <i32 8, i32 8, i32 8, i32 8>
; CHECK-NEXT: ret <4 x i1> [[R]]
;
%shl = shl <4 x i32> %x, %y
@@ -97,8 +96,7 @@ define <4 x i1> @vec_shl_and_negC_eq_undef1(<4 x i32> %x, <4 x i32> %y) {
define <4 x i1> @vec_shl_and_negC_eq_undef2(<4 x i32> %x, <4 x i32> %y) {
; CHECK-LABEL: @vec_shl_and_negC_eq_undef2(
; CHECK-NEXT: [[SHL:%.*]] = shl <4 x i32> [[X:%.*]], [[Y:%.*]]
-; CHECK-NEXT: [[AND:%.*]] = and <4 x i32> [[SHL]], <i32 -8, i32 -8, i32 -8, i32 -8>
-; CHECK-NEXT: [[R:%.*]] = icmp eq <4 x i32> [[AND]], <i32 0, i32 0, i32 0, i32 undef>
+; CHECK-NEXT: [[R:%.*]] = icmp ult <4 x i32> [[SHL]], <i32 8, i32 8, i32 8, i32 8>
; CHECK-NEXT: ret <4 x i1> [[R]]
;
%shl = shl <4 x i32> %x, %y
@@ -110,8 +108,7 @@ define <4 x i1> @vec_shl_and_negC_eq_undef2(<4 x i32> %x, <4 x i32> %y) {
define <4 x i1> @vec_shl_and_negC_eq_undef3(<4 x i32> %x, <4 x i32> %y) {
; CHECK-LABEL: @vec_shl_and_negC_eq_undef3(
; CHECK-NEXT: [[SHL:%.*]] = shl <4 x i32> [[X:%.*]], [[Y:%.*]]
-; CHECK-NEXT: [[AND:%.*]] = and <4 x i32> [[SHL]], <i32 -8, i32 -8, i32 undef, i32 -8>
-; CHECK-NEXT: [[R:%.*]] = icmp eq <4 x i32> [[AND]], <i32 0, i32 0, i32 0, i32 undef>
+; CHECK-NEXT: [[R:%.*]] = icmp ult <4 x i32> [[SHL]], <i32 8, i32 8, i32 8, i32 8>
; CHECK-NEXT: ret <4 x i1> [[R]]
;
%shl = shl <4 x i32> %x, %y
diff --git a/llvm/test/Transforms/InstCombine/shl-and-signbit-icmpeq-zero.ll b/llvm/test/Transforms/InstCombine/shl-and-signbit-icmpeq-zero.ll
index 42b755f51a97137..4c2c876e3925bf9 100644
--- a/llvm/test/Transforms/InstCombine/shl-and-signbit-icmpeq-zero.ll
+++ b/llvm/test/Transforms/InstCombine/shl-and-signbit-icmpeq-zero.ll
@@ -84,8 +84,7 @@ define <4 x i1> @vec_4xi32_shl_and_signbit_eq(<4 x i32> %x, <4 x i32> %y) {
define <4 x i1> @vec_4xi32_shl_and_signbit_eq_undef1(<4 x i32> %x, <4 x i32> %y) {
; CHECK-LABEL: @vec_4xi32_shl_and_signbit_eq_undef1(
; CHECK-NEXT: [[SHL:%.*]] = shl <4 x i32> [[X:%.*]], [[Y:%.*]]
-; CHECK-NEXT: [[AND:%.*]] = and <4 x i32> [[SHL]], <i32 -2147483648, i32 undef, i32 -2147483648, i32 -2147483648>
-; CHECK-NEXT: [[R:%.*]] = icmp eq <4 x i32> [[AND]], zeroinitializer
+; CHECK-NEXT: [[R:%.*]] = icmp sgt <4 x i32> [[SHL]], <i32 -1, i32 -1, i32 -1, i32 -1>
; CHECK-NEXT: ret <4 x i1> [[R]]
;
%shl = shl <4 x i32> %x, %y
@@ -97,8 +96,7 @@ define <4 x i1> @vec_4xi32_shl_and_signbit_eq_undef1(<4 x i32> %x, <4 x i32> %y)
define <4 x i1> @vec_4xi32_shl_and_signbit_eq_undef2(<4 x i32> %x, <4 x i32> %y) {
; CHECK-LABEL: @vec_4xi32_shl_and_signbit_eq_undef2(
; CHECK-NEXT: [[SHL:%.*]] = shl <4 x i32> [[X:%.*]], [[Y:%.*]]
-; CHECK-NEXT: [[AND:%.*]] = and <4 x i32> [[SHL]], <i32 -2147483648, i32 -2147483648, i32 -2147483648, i32 -2147483648>
-; CHECK-NEXT: [[R:%.*]] = icmp eq <4 x i32> [[AND]], <i32 undef, i32 0, i32 0, i32 0>
+; CHECK-NEXT: [[R:%.*]] = icmp sgt <4 x i32> [[SHL]], <i32 -1, i32 -1, i32 -1, i32 -1>
; CHECK-NEXT: ret <4 x i1> [[R]]
;
%shl = shl <4 x i32> %x, %y
@@ -110,8 +108,7 @@ define <4 x i1> @vec_4xi32_shl_and_signbit_eq_undef2(<4 x i32> %x, <4 x i32> %y)
define <4 x i1> @vec_4xi32_shl_and_signbit_eq_undef3(<4 x i32> %x, <4 x i32> %y) {
; CHECK-LABEL: @vec_4xi32_shl_and_signbit_eq_undef3(
; CHECK-NEXT: [[SHL:%.*]] = shl <4 x i32> [[X:%.*]], [[Y:%.*]]
-; CHECK-NEXT: [[AND:%.*]] = and <4 x i32> [[SHL]], <i32 -2147483648, i32 undef, i32 -2147483648, i32 -2147483648>
-; CHECK-NEXT: [[R:%.*]] = icmp eq <4 x i32> [[AND]], <i32 0, i32 0, i32 0, i32 undef>
+; CHECK-NEXT: [[R:%.*]] = icmp sgt <4 x i32> [[SHL]], <i32 -1, i32 -1, i32 -1, i32 -1>
; CHECK-NEXT: ret <4 x i1> [[R]]
;
%shl = shl <4 x i32> %x, %y
>From 582d6de7e0a9f9b6872f594eb99207bc3aea2787 Mon Sep 17 00:00:00 2001
From: Noah Goldstein <goldstein.w.n at gmail.com>
Date: Wed, 13 Sep 2023 13:45:58 -0500
Subject: [PATCH 07/13] [InstCombine] Make the `(icmp eq/ne (and X, Y), X)`
canonicalization work for non-const operands
We currently do:
`(icmp eq/ne (and X, Y), Y)` -> `(icmp eq/ne (and ~X, Y), 0)`
if `X` is constant. We can make this more general and do it if `X` is
freely invertable (i.e say `X = ~Z`).
As well, we can also do:
`(icmp eq/ne (and X, Y), Y)` -> `(icmp eq/ne (or X, ~Y), -1)`
If `Y` is freely invertible.
Proofs: https://alive2.llvm.org/ce/z/yeWH3E
Differential Revision: https://reviews.llvm.org/D159059
---
.../InstCombine/InstCombineCompares.cpp | 31 ++++++++++---------
...low-bit-mask-v2-and-icmp-eq-to-icmp-ule.ll | 11 +++----
...low-bit-mask-v2-and-icmp-ne-to-icmp-ugt.ll | 11 +++----
...low-bit-mask-v3-and-icmp-eq-to-icmp-ule.ll | 12 +++----
...low-bit-mask-v3-and-icmp-ne-to-icmp-ugt.ll | 12 +++----
.../InstCombine/icmp-and-lowbit-mask.ll | 30 +++++++++---------
.../Transforms/InstCombine/icmp-of-and-x.ll | 22 ++++++-------
7 files changed, 63 insertions(+), 66 deletions(-)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 74279d47ac913fd..27f236a4c4b39d6 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -4662,6 +4662,22 @@ static Instruction *foldICmpAndXX(ICmpInst &I, const SimplifyQuery &Q,
if (Pred == ICmpInst::ICMP_UGE)
return new ICmpInst(ICmpInst::ICMP_EQ, Op0, Op1);
+ if (ICmpInst::isEquality(Pred) && Op0->hasOneUse()) {
+ // icmp (X & Y) eq/ne Y --> (X | ~Y) eq/ne -1 if Y is freely invertible and
+ // Y is non-constant. If Y is constant this form is preferable (and
+ // canonicalize too it elsewhere).
+ if (!match(Op1, m_ImmConstant()) &&
+ IC.isFreeToInvert(Op1, Op1->hasOneUse() || Op1->hasNUses(2)))
+ return new ICmpInst(Pred,
+ IC.Builder.CreateOr(A, IC.Builder.CreateNot(Op1)),
+ Constant::getAllOnesValue(Op1->getType()));
+ // icmp (X & Y) eq/ne Y --> (~X & Y) eq/ne 0 if X is freely invertible.
+ if (IC.isFreeToInvert(A, A->hasOneUse()))
+ return new ICmpInst(Pred,
+ IC.Builder.CreateAnd(Op1, IC.Builder.CreateNot(A)),
+ Constant::getNullValue(Op1->getType()));
+ }
+
return nullptr;
}
@@ -5446,21 +5462,6 @@ Instruction *InstCombinerImpl::foldICmpEquality(ICmpInst &I) {
}
}
- // canoncalize:
- // (icmp eq/ne (and X, C), X)
- // -> (icmp eq/ne (and X, ~C), 0)
- {
- Constant *CMask;
- A = nullptr;
- if (match(Op0, m_OneUse(m_And(m_Specific(Op1), m_ImmConstant(CMask)))))
- A = Op1;
- else if (match(Op1, m_OneUse(m_And(m_Specific(Op0), m_ImmConstant(CMask)))))
- A = Op0;
- if (A)
- return new ICmpInst(Pred, Builder.CreateAnd(A, Builder.CreateNot(CMask)),
- Constant::getNullValue(A->getType()));
- }
-
if (match(Op1, m_Xor(m_Value(A), m_Value(B))) && (A == Op0 || B == Op0)) {
// A == (A^B) -> B == 0
Value *OtherVal = A == Op0 ? B : A;
diff --git a/llvm/test/Transforms/InstCombine/canonicalize-low-bit-mask-v2-and-icmp-eq-to-icmp-ule.ll b/llvm/test/Transforms/InstCombine/canonicalize-low-bit-mask-v2-and-icmp-eq-to-icmp-ule.ll
index 624b9baa372810a..af1c377d63b84c8 100644
--- a/llvm/test/Transforms/InstCombine/canonicalize-low-bit-mask-v2-and-icmp-eq-to-icmp-ule.ll
+++ b/llvm/test/Transforms/InstCombine/canonicalize-low-bit-mask-v2-and-icmp-eq-to-icmp-ule.ll
@@ -269,9 +269,8 @@ define i1 @n0(i8 %x, i8 %y, i8 %notx) {
define i1 @n1(i8 %x, i8 %y) {
; CHECK-LABEL: @n1(
; CHECK-NEXT: [[T0:%.*]] = shl nuw i8 1, [[Y:%.*]]
-; CHECK-NEXT: [[T1:%.*]] = xor i8 [[T0]], -1
-; CHECK-NEXT: [[T2:%.*]] = and i8 [[T1]], [[X:%.*]]
-; CHECK-NEXT: [[RET:%.*]] = icmp eq i8 [[T2]], [[X]]
+; CHECK-NEXT: [[TMP1:%.*]] = and i8 [[T0]], [[X:%.*]]
+; CHECK-NEXT: [[RET:%.*]] = icmp eq i8 [[TMP1]], 0
; CHECK-NEXT: ret i1 [[RET]]
;
%t0 = shl i8 1, %y ; not -1
@@ -284,9 +283,9 @@ define i1 @n1(i8 %x, i8 %y) {
define i1 @n2(i8 %x, i8 %y) {
; CHECK-LABEL: @n2(
; CHECK-NEXT: [[T0:%.*]] = shl i8 -1, [[Y:%.*]]
-; CHECK-NEXT: [[T1:%.*]] = xor i8 [[T0]], 1
-; CHECK-NEXT: [[T2:%.*]] = and i8 [[T1]], [[X:%.*]]
-; CHECK-NEXT: [[RET:%.*]] = icmp eq i8 [[T2]], [[X]]
+; CHECK-NEXT: [[TMP1:%.*]] = xor i8 [[T0]], -2
+; CHECK-NEXT: [[TMP2:%.*]] = and i8 [[TMP1]], [[X:%.*]]
+; CHECK-NEXT: [[RET:%.*]] = icmp eq i8 [[TMP2]], 0
; CHECK-NEXT: ret i1 [[RET]]
;
%t0 = shl i8 -1, %y
diff --git a/llvm/test/Transforms/InstCombine/canonicalize-low-bit-mask-v2-and-icmp-ne-to-icmp-ugt.ll b/llvm/test/Transforms/InstCombine/canonicalize-low-bit-mask-v2-and-icmp-ne-to-icmp-ugt.ll
index 4e1b90a476a2601..68188b1360bc27d 100644
--- a/llvm/test/Transforms/InstCombine/canonicalize-low-bit-mask-v2-and-icmp-ne-to-icmp-ugt.ll
+++ b/llvm/test/Transforms/InstCombine/canonicalize-low-bit-mask-v2-and-icmp-ne-to-icmp-ugt.ll
@@ -269,9 +269,8 @@ define i1 @n0(i8 %x, i8 %y, i8 %notx) {
define i1 @n1(i8 %x, i8 %y) {
; CHECK-LABEL: @n1(
; CHECK-NEXT: [[T0:%.*]] = shl nuw i8 1, [[Y:%.*]]
-; CHECK-NEXT: [[T1:%.*]] = xor i8 [[T0]], -1
-; CHECK-NEXT: [[T2:%.*]] = and i8 [[T1]], [[X:%.*]]
-; CHECK-NEXT: [[RET:%.*]] = icmp ne i8 [[T2]], [[X]]
+; CHECK-NEXT: [[TMP1:%.*]] = and i8 [[T0]], [[X:%.*]]
+; CHECK-NEXT: [[RET:%.*]] = icmp ne i8 [[TMP1]], 0
; CHECK-NEXT: ret i1 [[RET]]
;
%t0 = shl i8 1, %y ; not -1
@@ -284,9 +283,9 @@ define i1 @n1(i8 %x, i8 %y) {
define i1 @n2(i8 %x, i8 %y) {
; CHECK-LABEL: @n2(
; CHECK-NEXT: [[T0:%.*]] = shl i8 -1, [[Y:%.*]]
-; CHECK-NEXT: [[T1:%.*]] = xor i8 [[T0]], 1
-; CHECK-NEXT: [[T2:%.*]] = and i8 [[T1]], [[X:%.*]]
-; CHECK-NEXT: [[RET:%.*]] = icmp ne i8 [[T2]], [[X]]
+; CHECK-NEXT: [[TMP1:%.*]] = xor i8 [[T0]], -2
+; CHECK-NEXT: [[TMP2:%.*]] = and i8 [[TMP1]], [[X:%.*]]
+; CHECK-NEXT: [[RET:%.*]] = icmp ne i8 [[TMP2]], 0
; CHECK-NEXT: ret i1 [[RET]]
;
%t0 = shl i8 -1, %y
diff --git a/llvm/test/Transforms/InstCombine/canonicalize-low-bit-mask-v3-and-icmp-eq-to-icmp-ule.ll b/llvm/test/Transforms/InstCombine/canonicalize-low-bit-mask-v3-and-icmp-eq-to-icmp-ule.ll
index 75de1dc6d07a222..8ae4d755f977f56 100644
--- a/llvm/test/Transforms/InstCombine/canonicalize-low-bit-mask-v3-and-icmp-eq-to-icmp-ule.ll
+++ b/llvm/test/Transforms/InstCombine/canonicalize-low-bit-mask-v3-and-icmp-eq-to-icmp-ule.ll
@@ -251,9 +251,9 @@ define i1 @n1(i8 %x, i8 %y) {
; CHECK-LABEL: @n1(
; CHECK-NEXT: [[T0:%.*]] = shl i8 -1, [[Y:%.*]]
; CHECK-NEXT: call void @use8(i8 [[T0]])
-; CHECK-NEXT: [[T1:%.*]] = add i8 [[T0]], -1
-; CHECK-NEXT: [[T2:%.*]] = and i8 [[T1]], [[X:%.*]]
-; CHECK-NEXT: [[RET:%.*]] = icmp eq i8 [[T2]], [[X]]
+; CHECK-NEXT: [[TMP1:%.*]] = sub i8 0, [[T0]]
+; CHECK-NEXT: [[TMP2:%.*]] = and i8 [[TMP1]], [[X:%.*]]
+; CHECK-NEXT: [[RET:%.*]] = icmp eq i8 [[TMP2]], 0
; CHECK-NEXT: ret i1 [[RET]]
;
%t0 = shl i8 -1, %y ; not 1
@@ -268,9 +268,9 @@ define i1 @n2(i8 %x, i8 %y) {
; CHECK-LABEL: @n2(
; CHECK-NEXT: [[T0:%.*]] = shl nuw i8 1, [[Y:%.*]]
; CHECK-NEXT: call void @use8(i8 [[T0]])
-; CHECK-NEXT: [[T1:%.*]] = add nuw i8 [[T0]], 1
-; CHECK-NEXT: [[T2:%.*]] = and i8 [[T1]], [[X:%.*]]
-; CHECK-NEXT: [[RET:%.*]] = icmp eq i8 [[T2]], [[X]]
+; CHECK-NEXT: [[TMP1:%.*]] = sub nuw i8 -2, [[T0]]
+; CHECK-NEXT: [[TMP2:%.*]] = and i8 [[TMP1]], [[X:%.*]]
+; CHECK-NEXT: [[RET:%.*]] = icmp eq i8 [[TMP2]], 0
; CHECK-NEXT: ret i1 [[RET]]
;
%t0 = shl i8 1, %y
diff --git a/llvm/test/Transforms/InstCombine/canonicalize-low-bit-mask-v3-and-icmp-ne-to-icmp-ugt.ll b/llvm/test/Transforms/InstCombine/canonicalize-low-bit-mask-v3-and-icmp-ne-to-icmp-ugt.ll
index a92f4a9d9543112..622a78419a2ffb0 100644
--- a/llvm/test/Transforms/InstCombine/canonicalize-low-bit-mask-v3-and-icmp-ne-to-icmp-ugt.ll
+++ b/llvm/test/Transforms/InstCombine/canonicalize-low-bit-mask-v3-and-icmp-ne-to-icmp-ugt.ll
@@ -251,9 +251,9 @@ define i1 @n1(i8 %x, i8 %y) {
; CHECK-LABEL: @n1(
; CHECK-NEXT: [[T0:%.*]] = shl i8 -1, [[Y:%.*]]
; CHECK-NEXT: call void @use8(i8 [[T0]])
-; CHECK-NEXT: [[T1:%.*]] = add i8 [[T0]], -1
-; CHECK-NEXT: [[T2:%.*]] = and i8 [[T1]], [[X:%.*]]
-; CHECK-NEXT: [[RET:%.*]] = icmp ne i8 [[T2]], [[X]]
+; CHECK-NEXT: [[TMP1:%.*]] = sub i8 0, [[T0]]
+; CHECK-NEXT: [[TMP2:%.*]] = and i8 [[TMP1]], [[X:%.*]]
+; CHECK-NEXT: [[RET:%.*]] = icmp ne i8 [[TMP2]], 0
; CHECK-NEXT: ret i1 [[RET]]
;
%t0 = shl i8 -1, %y ; not 1
@@ -268,9 +268,9 @@ define i1 @n2(i8 %x, i8 %y) {
; CHECK-LABEL: @n2(
; CHECK-NEXT: [[T0:%.*]] = shl nuw i8 1, [[Y:%.*]]
; CHECK-NEXT: call void @use8(i8 [[T0]])
-; CHECK-NEXT: [[T1:%.*]] = add nuw i8 [[T0]], 1
-; CHECK-NEXT: [[T2:%.*]] = and i8 [[T1]], [[X:%.*]]
-; CHECK-NEXT: [[RET:%.*]] = icmp ne i8 [[T2]], [[X]]
+; CHECK-NEXT: [[TMP1:%.*]] = sub nuw i8 -2, [[T0]]
+; CHECK-NEXT: [[TMP2:%.*]] = and i8 [[TMP1]], [[X:%.*]]
+; CHECK-NEXT: [[RET:%.*]] = icmp ne i8 [[TMP2]], 0
; CHECK-NEXT: ret i1 [[RET]]
;
%t0 = shl i8 1, %y
diff --git a/llvm/test/Transforms/InstCombine/icmp-and-lowbit-mask.ll b/llvm/test/Transforms/InstCombine/icmp-and-lowbit-mask.ll
index 5578b9a66e1f219..43f6f36134ce338 100644
--- a/llvm/test/Transforms/InstCombine/icmp-and-lowbit-mask.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-and-lowbit-mask.ll
@@ -29,11 +29,11 @@ define i1 @src_is_mask_zext(i16 %x_in, i8 %y) {
define i1 @src_is_mask_zext_fail_not_mask(i16 %x_in, i8 %y) {
; CHECK-LABEL: @src_is_mask_zext_fail_not_mask(
-; CHECK-NEXT: [[X:%.*]] = xor i16 [[X_IN:%.*]], 123
; CHECK-NEXT: [[M_IN:%.*]] = lshr i8 -2, [[Y:%.*]]
; CHECK-NEXT: [[MASK:%.*]] = zext i8 [[M_IN]] to i16
-; CHECK-NEXT: [[AND:%.*]] = and i16 [[X]], [[MASK]]
-; CHECK-NEXT: [[R:%.*]] = icmp eq i16 [[AND]], [[X]]
+; CHECK-NEXT: [[TMP1:%.*]] = xor i16 [[X_IN:%.*]], -124
+; CHECK-NEXT: [[TMP2:%.*]] = or i16 [[TMP1]], [[MASK]]
+; CHECK-NEXT: [[R:%.*]] = icmp eq i16 [[TMP2]], -1
; CHECK-NEXT: ret i1 [[R]]
;
%x = xor i16 %x_in, 123
@@ -106,12 +106,12 @@ define i1 @src_is_mask_and(i8 %x_in, i8 %y, i8 %z) {
define i1 @src_is_mask_and_fail_mixed(i8 %x_in, i8 %y, i8 %z) {
; CHECK-LABEL: @src_is_mask_and_fail_mixed(
-; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
; CHECK-NEXT: [[MY:%.*]] = ashr i8 -8, [[Y:%.*]]
; CHECK-NEXT: [[MZ:%.*]] = lshr i8 -1, [[Z:%.*]]
; CHECK-NEXT: [[MASK:%.*]] = and i8 [[MY]], [[MZ]]
-; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], [[MASK]]
-; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[X]], [[AND]]
+; CHECK-NEXT: [[TMP1:%.*]] = xor i8 [[X_IN:%.*]], -124
+; CHECK-NEXT: [[TMP2:%.*]] = or i8 [[MASK]], [[TMP1]]
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[TMP2]], -1
; CHECK-NEXT: ret i1 [[R]]
;
%x = xor i8 %x_in, 123
@@ -159,11 +159,11 @@ define i1 @src_is_mask_xor(i8 %x_in, i8 %y) {
define i1 @src_is_mask_xor_fail_notmask(i8 %x_in, i8 %y) {
; CHECK-LABEL: @src_is_mask_xor_fail_notmask(
-; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
; CHECK-NEXT: [[Y_M1_NOT:%.*]] = sub i8 0, [[Y:%.*]]
; CHECK-NEXT: [[NOTMASK:%.*]] = xor i8 [[Y_M1_NOT]], [[Y]]
-; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], [[NOTMASK]]
-; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[AND]], [[X]]
+; CHECK-NEXT: [[TMP1:%.*]] = xor i8 [[X_IN:%.*]], -124
+; CHECK-NEXT: [[TMP2:%.*]] = or i8 [[NOTMASK]], [[TMP1]]
+; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[TMP2]], -1
; CHECK-NEXT: ret i1 [[R]]
;
%x = xor i8 %x_in, 123
@@ -349,12 +349,12 @@ define i1 @src_is_mask_umin(i8 %x_in, i8 %y, i8 %z) {
define i1 @src_is_mask_umin_fail_mismatch(i8 %x_in, i8 %y) {
; CHECK-LABEL: @src_is_mask_umin_fail_mismatch(
-; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
; CHECK-NEXT: [[Y_M1:%.*]] = add i8 [[Y:%.*]], -1
; CHECK-NEXT: [[YMASK:%.*]] = xor i8 [[Y_M1]], [[Y]]
; CHECK-NEXT: [[MASK:%.*]] = call i8 @llvm.umin.i8(i8 [[YMASK]], i8 -32)
-; CHECK-NEXT: [[AND:%.*]] = and i8 [[MASK]], [[X]]
-; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[AND]], [[X]]
+; CHECK-NEXT: [[TMP1:%.*]] = xor i8 [[X_IN:%.*]], -124
+; CHECK-NEXT: [[TMP2:%.*]] = or i8 [[MASK]], [[TMP1]]
+; CHECK-NEXT: [[R:%.*]] = icmp ne i8 [[TMP2]], -1
; CHECK-NEXT: ret i1 [[R]]
;
%x = xor i8 %x_in, 123
@@ -500,12 +500,12 @@ define i1 @src_is_notmask_lshr_shl(i8 %x_in, i8 %y) {
define i1 @src_is_notmask_lshr_shl_fail_mismatch_shifts(i8 %x_in, i8 %y, i8 %z) {
; CHECK-LABEL: @src_is_notmask_lshr_shl_fail_mismatch_shifts(
-; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
; CHECK-NEXT: [[MASK_SHR:%.*]] = lshr i8 -1, [[Y:%.*]]
; CHECK-NEXT: [[NMASK:%.*]] = shl i8 [[MASK_SHR]], [[Z:%.*]]
; CHECK-NEXT: [[MASK:%.*]] = xor i8 [[NMASK]], -1
-; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], [[MASK]]
-; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[AND]], [[X]]
+; CHECK-NEXT: [[TMP1:%.*]] = xor i8 [[X_IN:%.*]], -124
+; CHECK-NEXT: [[TMP2:%.*]] = or i8 [[TMP1]], [[MASK]]
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[TMP2]], -1
; CHECK-NEXT: ret i1 [[R]]
;
%x = xor i8 %x_in, 123
diff --git a/llvm/test/Transforms/InstCombine/icmp-of-and-x.ll b/llvm/test/Transforms/InstCombine/icmp-of-and-x.ll
index e95c72b75f97dfd..a83572bff3251ae 100644
--- a/llvm/test/Transforms/InstCombine/icmp-of-and-x.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-of-and-x.ll
@@ -238,9 +238,9 @@ define i1 @icmp_sle_negx_y_fail_maybe_zero(i8 %x, i8 %y) {
define i1 @icmp_eq_x_invertable_y_todo(i8 %x, i1 %y) {
; CHECK-LABEL: @icmp_eq_x_invertable_y_todo(
-; CHECK-NEXT: [[YY:%.*]] = select i1 [[Y:%.*]], i8 7, i8 24
-; CHECK-NEXT: [[AND:%.*]] = and i8 [[YY]], [[X:%.*]]
-; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[AND]], [[X]]
+; CHECK-NEXT: [[TMP1:%.*]] = select i1 [[Y:%.*]], i8 -8, i8 -25
+; CHECK-NEXT: [[TMP2:%.*]] = and i8 [[TMP1]], [[X:%.*]]
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[TMP2]], 0
; CHECK-NEXT: ret i1 [[R]]
;
%yy = select i1 %y, i8 7, i8 24
@@ -251,9 +251,8 @@ define i1 @icmp_eq_x_invertable_y_todo(i8 %x, i1 %y) {
define i1 @icmp_eq_x_invertable_y(i8 %x, i8 %y) {
; CHECK-LABEL: @icmp_eq_x_invertable_y(
-; CHECK-NEXT: [[YY:%.*]] = xor i8 [[Y:%.*]], -1
-; CHECK-NEXT: [[AND:%.*]] = and i8 [[YY]], [[X:%.*]]
-; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[AND]], [[X]]
+; CHECK-NEXT: [[TMP1:%.*]] = and i8 [[X:%.*]], [[Y:%.*]]
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[TMP1]], 0
; CHECK-NEXT: ret i1 [[R]]
;
%yy = xor i8 %y, -1
@@ -264,9 +263,9 @@ define i1 @icmp_eq_x_invertable_y(i8 %x, i8 %y) {
define i1 @icmp_eq_x_invertable_y2_todo(i8 %x, i1 %y) {
; CHECK-LABEL: @icmp_eq_x_invertable_y2_todo(
-; CHECK-NEXT: [[YY:%.*]] = select i1 [[Y:%.*]], i8 7, i8 24
-; CHECK-NEXT: [[AND:%.*]] = and i8 [[YY]], [[X:%.*]]
-; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[YY]], [[AND]]
+; CHECK-NEXT: [[TMP1:%.*]] = select i1 [[Y:%.*]], i8 -8, i8 -25
+; CHECK-NEXT: [[TMP2:%.*]] = or i8 [[TMP1]], [[X:%.*]]
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[TMP2]], -1
; CHECK-NEXT: ret i1 [[R]]
;
%yy = select i1 %y, i8 7, i8 24
@@ -277,9 +276,8 @@ define i1 @icmp_eq_x_invertable_y2_todo(i8 %x, i1 %y) {
define i1 @icmp_eq_x_invertable_y2(i8 %x, i8 %y) {
; CHECK-LABEL: @icmp_eq_x_invertable_y2(
-; CHECK-NEXT: [[YY:%.*]] = xor i8 [[Y:%.*]], -1
-; CHECK-NEXT: [[AND:%.*]] = and i8 [[YY]], [[X:%.*]]
-; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[AND]], [[YY]]
+; CHECK-NEXT: [[TMP1:%.*]] = or i8 [[X:%.*]], [[Y:%.*]]
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[TMP1]], -1
; CHECK-NEXT: ret i1 [[R]]
;
%yy = xor i8 %y, -1
>From c1e31c16863951ed2977c3ab1519f6d16f02ca39 Mon Sep 17 00:00:00 2001
From: Noah Goldstein <goldstein.w.n at gmail.com>
Date: Wed, 13 Sep 2023 13:46:00 -0500
Subject: [PATCH 08/13] [InstCombine] Make `getKnownSign` a member function of
InstCombiner; NFC
This is prep for using `getKnownSign` outside of just InstCombineCalls.
Differential Revision: https://reviews.llvm.org/D159060
---
.../InstCombine/InstCombineCalls.cpp | 29 ++++++++-----------
.../InstCombine/InstCombineInternal.h | 7 +++++
2 files changed, 19 insertions(+), 17 deletions(-)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
index e29fb869686ca0b..9930421ec53bca4 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
@@ -1016,10 +1016,9 @@ Instruction *InstCombinerImpl::foldIntrinsicIsFPClass(IntrinsicInst &II) {
return nullptr;
}
-static std::optional<bool> getKnownSign(Value *Op, Instruction *CxtI,
- const DataLayout &DL, AssumptionCache *AC,
- DominatorTree *DT) {
- KnownBits Known = computeKnownBits(Op, DL, 0, AC, CxtI, DT);
+std::optional<bool> InstCombinerImpl::getKnownSign(Value *Op,
+ Instruction *CxtI) const {
+ KnownBits Known = llvm::computeKnownBits(Op, DL, /*Depth*/ 0, &AC, CxtI, &DT);
if (Known.isNonNegative())
return false;
if (Known.isNegative())
@@ -1033,11 +1032,9 @@ static std::optional<bool> getKnownSign(Value *Op, Instruction *CxtI,
ICmpInst::ICMP_SLT, Op, Constant::getNullValue(Op->getType()), CxtI, DL);
}
-static std::optional<bool> getKnownSignOrZero(Value *Op, Instruction *CxtI,
- const DataLayout &DL,
- AssumptionCache *AC,
- DominatorTree *DT) {
- if (std::optional<bool> Sign = getKnownSign(Op, CxtI, DL, AC, DT))
+std::optional<bool>
+InstCombinerImpl::getKnownSignOrZero(Value *Op, Instruction *CxtI) const {
+ if (std::optional<bool> Sign = getKnownSign(Op, CxtI))
return Sign;
Value *X, *Y;
@@ -1049,12 +1046,11 @@ static std::optional<bool> getKnownSignOrZero(Value *Op, Instruction *CxtI,
/// Return true if two values \p Op0 and \p Op1 are known to have the same sign.
static bool signBitMustBeTheSame(Value *Op0, Value *Op1, Instruction *CxtI,
- const DataLayout &DL, AssumptionCache *AC,
- DominatorTree *DT) {
- std::optional<bool> Known1 = getKnownSign(Op1, CxtI, DL, AC, DT);
+ InstCombinerImpl &IC) {
+ std::optional<bool> Known1 = IC.getKnownSign(Op1, CxtI);
if (!Known1)
return false;
- std::optional<bool> Known0 = getKnownSign(Op0, CxtI, DL, AC, DT);
+ std::optional<bool> Known0 = IC.getKnownSign(Op0, CxtI);
if (!Known0)
return false;
return *Known0 == *Known1;
@@ -1547,8 +1543,7 @@ Instruction *InstCombinerImpl::visitCallInst(CallInst &CI) {
if (match(IIOperand, m_Select(m_Value(), m_Neg(m_Value(X)), m_Deferred(X))))
return replaceOperand(*II, 0, X);
- if (std::optional<bool> Known =
- getKnownSignOrZero(IIOperand, II, DL, &AC, &DT)) {
+ if (std::optional<bool> Known = getKnownSignOrZero(IIOperand, II)) {
// abs(x) -> x if x >= 0 (include abs(x-y) --> x - y where x >= y)
// abs(x) -> x if x > 0 (include abs(x-y) --> x - y where x > y)
if (!*Known)
@@ -1673,7 +1668,7 @@ Instruction *InstCombinerImpl::visitCallInst(CallInst &CI) {
bool UseAndN = IID == Intrinsic::smin || IID == Intrinsic::umin;
if (IID == Intrinsic::smax || IID == Intrinsic::smin) {
- auto KnownSign = getKnownSign(X, II, DL, &AC, &DT);
+ auto KnownSign = getKnownSign(X, II);
if (KnownSign == std::nullopt) {
UseOr = false;
UseAndN = false;
@@ -2446,7 +2441,7 @@ Instruction *InstCombinerImpl::visitCallInst(CallInst &CI) {
FastMathFlags InnerFlags = cast<FPMathOperator>(Src)->getFastMathFlags();
if ((FMF.allowReassoc() && InnerFlags.allowReassoc()) ||
- signBitMustBeTheSame(Exp, InnerExp, II, DL, &AC, &DT)) {
+ signBitMustBeTheSame(Exp, InnerExp, II, *this)) {
// TODO: Add nsw/nuw probably safe if integer type exceeds exponent
// width.
Value *NewExp = Builder.CreateAdd(InnerExp, Exp);
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineInternal.h b/llvm/lib/Transforms/InstCombine/InstCombineInternal.h
index a53d67b2899b700..c60d0754ec3b219 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineInternal.h
+++ b/llvm/lib/Transforms/InstCombine/InstCombineInternal.h
@@ -470,6 +470,13 @@ class LLVM_LIBRARY_VISIBILITY InstCombinerImpl final
Instruction::BinaryOps BinaryOp, bool IsSigned,
Value *LHS, Value *RHS, Instruction *CxtI) const;
+ // Return true if known negative, false if known positive, and nullopt if
+ // unknown.
+ std::optional<bool> getKnownSign(Value *Op, Instruction *CxtI) const;
+ // Return true if known negative or zero, false if known non-zero positive,
+ // and nullopt if unknown.
+ std::optional<bool> getKnownSignOrZero(Value *Op, Instruction *CxtI) const;
+
/// Performs a few simplifications for operators which are associative
/// or commutative.
bool SimplifyAssociativeOrCommutative(BinaryOperator &I);
>From 9f8857a426ad6d4c13dd945163540379477e9bc2 Mon Sep 17 00:00:00 2001
From: Noah Goldstein <goldstein.w.n at gmail.com>
Date: Wed, 13 Sep 2023 13:46:03 -0500
Subject: [PATCH 09/13] [InstCombine] Add transforms `(icmp spred (and X, Y),
X)` if `X` or `Y` are known signed/unsigned
Several transforms:
1) If known `Y < 0`:
- slt -> ult: https://alive2.llvm.org/ce/z/unpsxu
- sle -> ule: https://alive2.llvm.org/ce/z/a5cpmr
- sgt -> ugt: https://alive2.llvm.org/ce/z/qgsgya
- sge -> uge: https://alive2.llvm.org/ce/z/ek_3ls
2) If known `Y > 0`:
- `(X & PosY) s> X --> X s< 0`
- https://alive2.llvm.org/ce/z/drewij
- `(X & PosY) s> X --> X s< 0`
- https://alive2.llvm.org/ce/z/9arL3D
3) If known `X < 0`:
- `(NegX & Y) s> NegX --> Y s>= 0`
- https://alive2.llvm.org/ce/z/ApkaEh
- `(NegX & Y) s<= NegX --> Y s< 0`
- https://alive2.llvm.org/ce/z/oRnfHp
Differential Revision: https://reviews.llvm.org/D145426
---
.../InstCombine/InstCombineCompares.cpp | 45 ++++++++++++++++++-
...t-low-bit-mask-and-icmp-sge-to-icmp-sle.ll | 3 +-
...t-low-bit-mask-and-icmp-sgt-to-icmp-sgt.ll | 3 +-
...t-low-bit-mask-and-icmp-sle-to-icmp-sle.ll | 3 +-
...t-low-bit-mask-and-icmp-slt-to-icmp-sgt.ll | 3 +-
.../InstCombine/icmp-and-lowbit-mask.ll | 8 +---
.../Transforms/InstCombine/icmp-of-and-x.ll | 28 ++++--------
7 files changed, 58 insertions(+), 35 deletions(-)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 27f236a4c4b39d6..e2cdf0cd5c4349b 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -4641,7 +4641,7 @@ static Instruction *foldICmpXNegX(ICmpInst &I,
return nullptr;
}
-static Instruction *foldICmpAndXX(ICmpInst &I, const SimplifyQuery &Q,
+static Instruction *foldICmpAndXX(ICmpInst &I, const SimplifyQuery &,
InstCombinerImpl &IC) {
Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1), *A;
// Normalize and operand as operand 0.
@@ -4678,6 +4678,49 @@ static Instruction *foldICmpAndXX(ICmpInst &I, const SimplifyQuery &Q,
Constant::getNullValue(Op1->getType()));
}
+ if (ICmpInst::isEquality(Pred) || ICmpInst::getSignedPredicate(Pred) != Pred)
+ return nullptr;
+
+ auto IsTrue = [](std::optional<bool> OptBool) {
+ return OptBool.has_value() && OptBool.value();
+ };
+
+ auto IsFalse = [](std::optional<bool> OptBool) {
+ return OptBool.has_value() && !OptBool.value();
+ };
+
+ auto KnownSignY = IC.getKnownSign(A, &I);
+
+ // (X & NegY) spred X --> (icmp (X & NegY) upred X
+ if (IsTrue(KnownSignY))
+ return new ICmpInst(ICmpInst::getUnsignedPredicate(Pred), Op0, Op1);
+
+ if (Pred != ICmpInst::ICMP_SLE && Pred != ICmpInst::ICMP_SGT)
+ return nullptr;
+
+ if (IsFalse(KnownSignY)) {
+ // (X & PosY) s<= X --> X s>= 0
+ if (Pred == ICmpInst::ICMP_SLE)
+ return new ICmpInst(ICmpInst::ICMP_SGE, Op1,
+ Constant::getNullValue(Op1->getType()));
+ // (X & PosY) s> X --> X s< 0
+ if (Pred == ICmpInst::ICMP_SGT)
+ return new ICmpInst(ICmpInst::ICMP_SLT, Op1,
+ Constant::getNullValue(Op1->getType()));
+ }
+
+ if (IsTrue(IC.getKnownSign(Op1, &I))) {
+ // (NegX & Y) s> NegX --> Y s>= 0
+ if (Pred == ICmpInst::ICMP_SGT)
+ return new ICmpInst(ICmpInst::ICMP_SGE, A,
+ Constant::getNullValue(A->getType()));
+
+ // (NegX & Y) s<= NegX --> Y s< 0
+ if (Pred == ICmpInst::ICMP_SLE)
+ return new ICmpInst(ICmpInst::ICMP_SLT, A,
+ Constant::getNullValue(A->getType()));
+ }
+
return nullptr;
}
diff --git a/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-sge-to-icmp-sle.ll b/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-sge-to-icmp-sle.ll
index 60921042d524356..0380c052d5d025d 100644
--- a/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-sge-to-icmp-sle.ll
+++ b/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-sge-to-icmp-sle.ll
@@ -98,8 +98,7 @@ declare i8 @gen8()
define i1 @c0() {
; CHECK-LABEL: @c0(
; CHECK-NEXT: [[X:%.*]] = call i8 @gen8()
-; CHECK-NEXT: [[TMP0:%.*]] = and i8 [[X]], 3
-; CHECK-NEXT: [[RET:%.*]] = icmp sge i8 [[X]], [[TMP0]]
+; CHECK-NEXT: [[RET:%.*]] = icmp sgt i8 [[X]], -1
; CHECK-NEXT: ret i1 [[RET]]
;
%x = call i8 @gen8()
diff --git a/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-sgt-to-icmp-sgt.ll b/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-sgt-to-icmp-sgt.ll
index 6345e70d7220e29..4a8b4d4ba595cd7 100644
--- a/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-sgt-to-icmp-sgt.ll
+++ b/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-sgt-to-icmp-sgt.ll
@@ -125,8 +125,7 @@ define i1 @oneuse0() {
define i1 @c0(i8 %x) {
; CHECK-LABEL: @c0(
-; CHECK-NEXT: [[TMP0:%.*]] = and i8 [[X:%.*]], 3
-; CHECK-NEXT: [[RET:%.*]] = icmp sgt i8 [[TMP0]], [[X]]
+; CHECK-NEXT: [[RET:%.*]] = icmp slt i8 [[X:%.*]], 0
; CHECK-NEXT: ret i1 [[RET]]
;
%tmp0 = and i8 %x, 3
diff --git a/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-sle-to-icmp-sle.ll b/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-sle-to-icmp-sle.ll
index b7aec53fed6760f..8d00f445243a402 100644
--- a/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-sle-to-icmp-sle.ll
+++ b/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-sle-to-icmp-sle.ll
@@ -113,8 +113,7 @@ define i1 @oneuse0() {
define i1 @c0(i8 %x) {
; CHECK-LABEL: @c0(
-; CHECK-NEXT: [[TMP0:%.*]] = and i8 [[X:%.*]], 3
-; CHECK-NEXT: [[RET:%.*]] = icmp sle i8 [[TMP0]], [[X]]
+; CHECK-NEXT: [[RET:%.*]] = icmp sgt i8 [[X:%.*]], -1
; CHECK-NEXT: ret i1 [[RET]]
;
%tmp0 = and i8 %x, 3
diff --git a/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-slt-to-icmp-sgt.ll b/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-slt-to-icmp-sgt.ll
index 56661d335c4f602..8130029a3dd58e1 100644
--- a/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-slt-to-icmp-sgt.ll
+++ b/llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-slt-to-icmp-sgt.ll
@@ -108,8 +108,7 @@ declare i8 @gen8()
define i1 @c0() {
; CHECK-LABEL: @c0(
; CHECK-NEXT: [[X:%.*]] = call i8 @gen8()
-; CHECK-NEXT: [[TMP0:%.*]] = and i8 [[X]], 3
-; CHECK-NEXT: [[RET:%.*]] = icmp slt i8 [[X]], [[TMP0]]
+; CHECK-NEXT: [[RET:%.*]] = icmp slt i8 [[X]], 0
; CHECK-NEXT: ret i1 [[RET]]
;
%x = call i8 @gen8()
diff --git a/llvm/test/Transforms/InstCombine/icmp-and-lowbit-mask.ll b/llvm/test/Transforms/InstCombine/icmp-and-lowbit-mask.ll
index 43f6f36134ce338..a134a3a5ddc8b9c 100644
--- a/llvm/test/Transforms/InstCombine/icmp-and-lowbit-mask.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-and-lowbit-mask.ll
@@ -578,9 +578,7 @@ define i1 @src_is_notmask_neg_p2_fail_not_invertable(i8 %x_in, i8 %y) {
define i1 @src_is_mask_const_slt(i8 %x_in) {
; CHECK-LABEL: @src_is_mask_const_slt(
-; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
-; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], 7
-; CHECK-NEXT: [[R:%.*]] = icmp slt i8 [[X]], [[AND]]
+; CHECK-NEXT: [[R:%.*]] = icmp slt i8 [[X_IN:%.*]], 0
; CHECK-NEXT: ret i1 [[R]]
;
%x = xor i8 %x_in, 123
@@ -603,9 +601,7 @@ define i1 @src_is_mask_const_sgt(i8 %x_in) {
define i1 @src_is_mask_const_sle(i8 %x_in) {
; CHECK-LABEL: @src_is_mask_const_sle(
-; CHECK-NEXT: [[X:%.*]] = xor i8 [[X_IN:%.*]], 123
-; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], 31
-; CHECK-NEXT: [[R:%.*]] = icmp sle i8 [[AND]], [[X]]
+; CHECK-NEXT: [[R:%.*]] = icmp sgt i8 [[X_IN:%.*]], -1
; CHECK-NEXT: ret i1 [[R]]
;
%x = xor i8 %x_in, 123
diff --git a/llvm/test/Transforms/InstCombine/icmp-of-and-x.ll b/llvm/test/Transforms/InstCombine/icmp-of-and-x.ll
index a83572bff3251ae..bfb4abc80ae379b 100644
--- a/llvm/test/Transforms/InstCombine/icmp-of-and-x.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-of-and-x.ll
@@ -57,7 +57,7 @@ define i1 @icmp_sge_x_negy(i8 %x, i8 %y) {
; CHECK-NEXT: [[CY:%.*]] = icmp slt i8 [[Y:%.*]], 0
; CHECK-NEXT: call void @llvm.assume(i1 [[CY]])
; CHECK-NEXT: [[AND:%.*]] = and i8 [[X:%.*]], [[Y]]
-; CHECK-NEXT: [[Z:%.*]] = icmp sge i8 [[AND]], [[X]]
+; CHECK-NEXT: [[Z:%.*]] = icmp eq i8 [[AND]], [[X]]
; CHECK-NEXT: ret i1 [[Z]]
;
%cy = icmp slt i8 %y, 0
@@ -73,7 +73,7 @@ define i1 @icmp_slt_x_negy(i8 %x, i8 %y) {
; CHECK-NEXT: br i1 [[CY]], label [[NEGY:%.*]], label [[POSY:%.*]]
; CHECK: negy:
; CHECK-NEXT: [[AND:%.*]] = and i8 [[X:%.*]], [[Y]]
-; CHECK-NEXT: [[Z:%.*]] = icmp slt i8 [[AND]], [[X]]
+; CHECK-NEXT: [[Z:%.*]] = icmp ne i8 [[AND]], [[X]]
; CHECK-NEXT: ret i1 [[Z]]
; CHECK: posy:
; CHECK-NEXT: [[R:%.*]] = call i1 @barrier()
@@ -115,10 +115,7 @@ posy:
define i1 @icmp_sle_x_negy(i8 %x, i8 %yy) {
; CHECK-LABEL: @icmp_sle_x_negy(
-; CHECK-NEXT: [[Y:%.*]] = or i8 [[YY:%.*]], -128
-; CHECK-NEXT: [[AND:%.*]] = and i8 [[Y]], [[X:%.*]]
-; CHECK-NEXT: [[Z:%.*]] = icmp sle i8 [[AND]], [[X]]
-; CHECK-NEXT: ret i1 [[Z]]
+; CHECK-NEXT: ret i1 true
;
%y = or i8 %yy, 128
%and = and i8 %y, %x
@@ -128,10 +125,7 @@ define i1 @icmp_sle_x_negy(i8 %x, i8 %yy) {
define <2 x i1> @icmp_sgt_x_negy(<2 x i8> %x, <2 x i8> %yy) {
; CHECK-LABEL: @icmp_sgt_x_negy(
-; CHECK-NEXT: [[Y:%.*]] = or <2 x i8> [[YY:%.*]], <i8 -128, i8 -128>
-; CHECK-NEXT: [[AND:%.*]] = and <2 x i8> [[Y]], [[X:%.*]]
-; CHECK-NEXT: [[Z:%.*]] = icmp sgt <2 x i8> [[AND]], [[X]]
-; CHECK-NEXT: ret <2 x i1> [[Z]]
+; CHECK-NEXT: ret <2 x i1> zeroinitializer
;
%y = or <2 x i8> %yy, <i8 128, i8 128>
%and = and <2 x i8> %y, %x
@@ -154,9 +148,7 @@ define <2 x i1> @icmp_sgt_x_negy_fail_partial(<2 x i8> %x, <2 x i8> %yy) {
define <2 x i1> @icmp_sle_x_posy(<2 x i8> %x, <2 x i8> %yy) {
; CHECK-LABEL: @icmp_sle_x_posy(
-; CHECK-NEXT: [[Y:%.*]] = and <2 x i8> [[YY:%.*]], <i8 127, i8 127>
-; CHECK-NEXT: [[AND:%.*]] = and <2 x i8> [[Y]], [[X:%.*]]
-; CHECK-NEXT: [[Z:%.*]] = icmp sle <2 x i8> [[AND]], [[X]]
+; CHECK-NEXT: [[Z:%.*]] = icmp sgt <2 x i8> [[X:%.*]], <i8 -1, i8 -1>
; CHECK-NEXT: ret <2 x i1> [[Z]]
;
%y = and <2 x i8> %yy, <i8 127, i8 127>
@@ -182,8 +174,7 @@ define i1 @icmp_sgt_x_posy(i8 %x, i8 %y) {
; CHECK-LABEL: @icmp_sgt_x_posy(
; CHECK-NEXT: [[CY:%.*]] = icmp sgt i8 [[Y:%.*]], -1
; CHECK-NEXT: call void @llvm.assume(i1 [[CY]])
-; CHECK-NEXT: [[AND:%.*]] = and i8 [[X:%.*]], [[Y]]
-; CHECK-NEXT: [[Z:%.*]] = icmp sgt i8 [[AND]], [[X]]
+; CHECK-NEXT: [[Z:%.*]] = icmp slt i8 [[X:%.*]], 0
; CHECK-NEXT: ret i1 [[Z]]
;
%cy = icmp sge i8 %y, 0
@@ -195,9 +186,7 @@ define i1 @icmp_sgt_x_posy(i8 %x, i8 %y) {
define <2 x i1> @icmp_sgt_negx_y(<2 x i8> %xx, <2 x i8> %y) {
; CHECK-LABEL: @icmp_sgt_negx_y(
-; CHECK-NEXT: [[X:%.*]] = or <2 x i8> [[XX:%.*]], <i8 -128, i8 -128>
-; CHECK-NEXT: [[AND:%.*]] = and <2 x i8> [[X]], [[Y:%.*]]
-; CHECK-NEXT: [[Z:%.*]] = icmp sgt <2 x i8> [[AND]], [[X]]
+; CHECK-NEXT: [[Z:%.*]] = icmp sgt <2 x i8> [[Y:%.*]], <i8 -1, i8 -1>
; CHECK-NEXT: ret <2 x i1> [[Z]]
;
%x = or <2 x i8> %xx, <i8 128, i8 128>
@@ -210,8 +199,7 @@ define i1 @icmp_sle_negx_y(i8 %x, i8 %y) {
; CHECK-LABEL: @icmp_sle_negx_y(
; CHECK-NEXT: [[CX:%.*]] = icmp slt i8 [[X:%.*]], 0
; CHECK-NEXT: call void @llvm.assume(i1 [[CX]])
-; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], [[Y:%.*]]
-; CHECK-NEXT: [[Z:%.*]] = icmp sle i8 [[AND]], [[X]]
+; CHECK-NEXT: [[Z:%.*]] = icmp slt i8 [[Y:%.*]], 0
; CHECK-NEXT: ret i1 [[Z]]
;
%cx = icmp slt i8 %x, 0
>From 302e32b3a906e8ae0a94a72c1fd3bdb9e90c2b05 Mon Sep 17 00:00:00 2001
From: Noah Goldstein <goldstein.w.n at gmail.com>
Date: Wed, 13 Sep 2023 13:46:06 -0500
Subject: [PATCH 10/13] [InstCombine] Add tests for folding multiuse `(icmp
eq/ne (or X, Y), Y)`; NFC
Differential Revision: https://reviews.llvm.org/D159061
---
.../Transforms/InstCombine/icmp-of-or-x.ll | 27 +++++++++++++++++++
1 file changed, 27 insertions(+)
diff --git a/llvm/test/Transforms/InstCombine/icmp-of-or-x.ll b/llvm/test/Transforms/InstCombine/icmp-of-or-x.ll
index 4b8df439b846fb6..b8773057d18fc17 100644
--- a/llvm/test/Transforms/InstCombine/icmp-of-or-x.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-of-or-x.ll
@@ -377,3 +377,30 @@ define i1 @pr64610(ptr %b) {
%r = icmp ugt i32 %or, %s
ret i1 %r
}
+
+define i1 @icmp_eq_x_invertable_y2_todo(i8 %x, i1 %y, i8 %z) {
+; CHECK-LABEL: @icmp_eq_x_invertable_y2_todo(
+; CHECK-NEXT: [[ZZ:%.*]] = xor i8 [[Z:%.*]], -1
+; CHECK-NEXT: [[YY:%.*]] = select i1 [[Y:%.*]], i8 7, i8 [[ZZ]]
+; CHECK-NEXT: [[OR:%.*]] = or i8 [[YY]], [[X:%.*]]
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[YY]], [[OR]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %zz = xor i8 %z, -1
+ %yy = select i1 %y, i8 7, i8 %zz
+ %or = or i8 %x, %yy
+ %r = icmp eq i8 %yy, %or
+ ret i1 %r
+}
+
+define i1 @icmp_eq_x_invertable_y2(i8 %x, i8 %y) {
+; CHECK-LABEL: @icmp_eq_x_invertable_y2(
+; CHECK-NEXT: [[TMP1:%.*]] = and i8 [[X:%.*]], [[Y:%.*]]
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[TMP1]], 0
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %yy = xor i8 %y, -1
+ %or = or i8 %x, %yy
+ %r = icmp eq i8 %yy, %or
+ ret i1 %r
+}
>From 9211164eb9c80cb6813aa3ea4a5b609bc375e0a3 Mon Sep 17 00:00:00 2001
From: Noah Goldstein <goldstein.w.n at gmail.com>
Date: Wed, 13 Sep 2023 13:46:08 -0500
Subject: [PATCH 11/13] [InstCombine] Folding multiuse `(icmp eq/ne (or X, Y),
Y)` for 2 uses of `Y`
The fold will replace 2 uses of `Y` we should also do fold if `Y` has
2 uses (not only oneuse).
Reviewed By: nikic
Differential Revision: https://reviews.llvm.org/D159062
---
llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp | 2 +-
llvm/test/Transforms/InstCombine/icmp-of-or-x.ll | 7 +++----
2 files changed, 4 insertions(+), 5 deletions(-)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index e2cdf0cd5c4349b..47dcb653649e211 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -4747,7 +4747,7 @@ static Instruction *foldICmpOrXX(ICmpInst &I, const SimplifyQuery &Q,
if (ICmpInst::isEquality(Pred) && Op0->hasOneUse()) {
// icmp (X | Y) eq/ne Y --> (X & ~Y) eq/ne 0 if Y is freely invertible
- if (IC.isFreeToInvert(Op1, Op1->hasOneUse()))
+ if (IC.isFreeToInvert(Op1, !Op1->hasNUsesOrMore(3)))
return new ICmpInst(Pred,
IC.Builder.CreateAnd(A, IC.Builder.CreateNot(Op1)),
Constant::getNullValue(Op1->getType()));
diff --git a/llvm/test/Transforms/InstCombine/icmp-of-or-x.ll b/llvm/test/Transforms/InstCombine/icmp-of-or-x.ll
index b8773057d18fc17..050636e3332e9cc 100644
--- a/llvm/test/Transforms/InstCombine/icmp-of-or-x.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-of-or-x.ll
@@ -380,10 +380,9 @@ define i1 @pr64610(ptr %b) {
define i1 @icmp_eq_x_invertable_y2_todo(i8 %x, i1 %y, i8 %z) {
; CHECK-LABEL: @icmp_eq_x_invertable_y2_todo(
-; CHECK-NEXT: [[ZZ:%.*]] = xor i8 [[Z:%.*]], -1
-; CHECK-NEXT: [[YY:%.*]] = select i1 [[Y:%.*]], i8 7, i8 [[ZZ]]
-; CHECK-NEXT: [[OR:%.*]] = or i8 [[YY]], [[X:%.*]]
-; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[YY]], [[OR]]
+; CHECK-NEXT: [[TMP1:%.*]] = select i1 [[Y:%.*]], i8 -8, i8 [[Z:%.*]]
+; CHECK-NEXT: [[TMP2:%.*]] = and i8 [[TMP1]], [[X:%.*]]
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[TMP2]], 0
; CHECK-NEXT: ret i1 [[R]]
;
%zz = xor i8 %z, -1
>From 17d51d0323202d10eae38bf866bd90650141c87b Mon Sep 17 00:00:00 2001
From: Noah Goldstein <goldstein.w.n at gmail.com>
Date: Wed, 13 Sep 2023 13:46:11 -0500
Subject: [PATCH 12/13] [InstCombine] Add tests for transforming `(or/and (icmp
eq/ne X,0),(icmp eq/ne X,Pow2OrZero))`; NFC
Differential Revision: https://reviews.llvm.org/D157311
---
.../Transforms/InstCombine/and-or-icmps.ll | 145 ++++++++++++++++++
1 file changed, 145 insertions(+)
diff --git a/llvm/test/Transforms/InstCombine/and-or-icmps.ll b/llvm/test/Transforms/InstCombine/and-or-icmps.ll
index 881a9b7ff129dbb..3b43b008367e788 100644
--- a/llvm/test/Transforms/InstCombine/and-or-icmps.ll
+++ b/llvm/test/Transforms/InstCombine/and-or-icmps.ll
@@ -3044,3 +3044,148 @@ define i32 @icmp_slt_0_or_icmp_add_1_sge_100_i32_fail(i32 %x) {
%D = or i32 %C, %B
ret i32 %D
}
+
+define i1 @icmp_eq_or_z_or_pow2orz(i8 %x, i8 %y) {
+; CHECK-LABEL: @icmp_eq_or_z_or_pow2orz(
+; CHECK-NEXT: [[NY:%.*]] = sub i8 0, [[Y:%.*]]
+; CHECK-NEXT: [[POW2ORZ:%.*]] = and i8 [[NY]], [[Y]]
+; CHECK-NEXT: [[C0:%.*]] = icmp eq i8 [[X:%.*]], 0
+; CHECK-NEXT: [[CP2:%.*]] = icmp eq i8 [[POW2ORZ]], [[X]]
+; CHECK-NEXT: [[R:%.*]] = or i1 [[C0]], [[CP2]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %ny = sub i8 0, %y
+ %pow2orz = and i8 %ny, %y
+
+ %c0 = icmp eq i8 %x, 0
+ %cp2 = icmp eq i8 %x, %pow2orz
+ %r = or i1 %c0, %cp2
+ ret i1 %r
+}
+
+
+define i1 @icmp_eq_or_z_or_pow2orz_fail_logic_or(i8 %x, i8 %y) {
+; CHECK-LABEL: @icmp_eq_or_z_or_pow2orz_fail_logic_or(
+; CHECK-NEXT: [[NY:%.*]] = sub i8 0, [[Y:%.*]]
+; CHECK-NEXT: [[POW2ORZ:%.*]] = and i8 [[NY]], [[Y]]
+; CHECK-NEXT: [[C0:%.*]] = icmp eq i8 [[X:%.*]], 0
+; CHECK-NEXT: [[CP2:%.*]] = icmp eq i8 [[POW2ORZ]], [[X]]
+; CHECK-NEXT: [[R:%.*]] = select i1 [[C0]], i1 true, i1 [[CP2]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %ny = sub i8 0, %y
+ %pow2orz = and i8 %ny, %y
+
+ %c0 = icmp eq i8 %x, 0
+ %cp2 = icmp eq i8 %x, %pow2orz
+ %r = select i1 %c0, i1 true, i1 %cp2
+ ret i1 %r
+}
+
+
+define <2 x i1> @icmp_ne_and_z_and_pow2orz(<2 x i8> %x, <2 x i8> %y) {
+; CHECK-LABEL: @icmp_ne_and_z_and_pow2orz(
+; CHECK-NEXT: [[NY:%.*]] = sub <2 x i8> zeroinitializer, [[Y:%.*]]
+; CHECK-NEXT: [[POW2ORZ:%.*]] = and <2 x i8> [[NY]], [[Y]]
+; CHECK-NEXT: [[C0:%.*]] = icmp ne <2 x i8> [[X:%.*]], zeroinitializer
+; CHECK-NEXT: [[CP2:%.*]] = icmp ne <2 x i8> [[POW2ORZ]], [[X]]
+; CHECK-NEXT: [[R:%.*]] = and <2 x i1> [[C0]], [[CP2]]
+; CHECK-NEXT: ret <2 x i1> [[R]]
+;
+ %ny = sub <2 x i8> zeroinitializer, %y
+ %pow2orz = and <2 x i8> %ny, %y
+
+ %c0 = icmp ne <2 x i8> %x, zeroinitializer
+ %cp2 = icmp ne <2 x i8> %x, %pow2orz
+ %r = and <2 x i1> %c0, %cp2
+ ret <2 x i1> %r
+}
+
+define <2 x i1> @icmp_ne_and_z_and_pow2orz_fail_logic_and(<2 x i8> %x, <2 x i8> %y) {
+; CHECK-LABEL: @icmp_ne_and_z_and_pow2orz_fail_logic_and(
+; CHECK-NEXT: [[NY:%.*]] = sub <2 x i8> zeroinitializer, [[Y:%.*]]
+; CHECK-NEXT: [[POW2ORZ:%.*]] = and <2 x i8> [[NY]], [[Y]]
+; CHECK-NEXT: [[C0:%.*]] = icmp ne <2 x i8> [[X:%.*]], zeroinitializer
+; CHECK-NEXT: [[CP2:%.*]] = icmp ne <2 x i8> [[POW2ORZ]], [[X]]
+; CHECK-NEXT: [[R:%.*]] = select <2 x i1> [[C0]], <2 x i1> [[CP2]], <2 x i1> zeroinitializer
+; CHECK-NEXT: ret <2 x i1> [[R]]
+;
+ %ny = sub <2 x i8> zeroinitializer, %y
+ %pow2orz = and <2 x i8> %ny, %y
+
+ %c0 = icmp ne <2 x i8> %x, zeroinitializer
+ %cp2 = icmp ne <2 x i8> %x, %pow2orz
+ %r = select <2 x i1> %c0, <2 x i1> %cp2, <2 x i1> zeroinitializer
+ ret <2 x i1> %r
+}
+
+define i1 @icmp_eq_or_z_or_pow2orz_fail_not_pow2(i8 %x, i8 %y) {
+; CHECK-LABEL: @icmp_eq_or_z_or_pow2orz_fail_not_pow2(
+; CHECK-NEXT: [[NY:%.*]] = sub i8 1, [[Y:%.*]]
+; CHECK-NEXT: [[POW2ORZ:%.*]] = and i8 [[NY]], [[Y]]
+; CHECK-NEXT: [[C0:%.*]] = icmp eq i8 [[X:%.*]], 0
+; CHECK-NEXT: [[CP2:%.*]] = icmp eq i8 [[POW2ORZ]], [[X]]
+; CHECK-NEXT: [[R:%.*]] = or i1 [[C0]], [[CP2]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %ny = sub i8 1, %y
+ %pow2orz = and i8 %ny, %y
+
+ %c0 = icmp eq i8 %x, 0
+ %cp2 = icmp eq i8 %x, %pow2orz
+ %r = or i1 %c0, %cp2
+ ret i1 %r
+}
+
+define i1 @icmp_eq_or_z_or_pow2orz_fail_nonzero_const(i8 %x, i8 %y) {
+; CHECK-LABEL: @icmp_eq_or_z_or_pow2orz_fail_nonzero_const(
+; CHECK-NEXT: [[NY:%.*]] = sub i8 0, [[Y:%.*]]
+; CHECK-NEXT: [[POW2ORZ:%.*]] = and i8 [[NY]], [[Y]]
+; CHECK-NEXT: [[C0:%.*]] = icmp eq i8 [[X:%.*]], 1
+; CHECK-NEXT: [[CP2:%.*]] = icmp eq i8 [[POW2ORZ]], [[X]]
+; CHECK-NEXT: [[R:%.*]] = or i1 [[C0]], [[CP2]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %ny = sub i8 0, %y
+ %pow2orz = and i8 %ny, %y
+
+ %c0 = icmp eq i8 %x, 1
+ %cp2 = icmp eq i8 %x, %pow2orz
+ %r = or i1 %c0, %cp2
+ ret i1 %r
+}
+
+define <2 x i1> @icmp_ne_and_z_and_pow2orz_fail_bad_pred(<2 x i8> %x, <2 x i8> %y) {
+; CHECK-LABEL: @icmp_ne_and_z_and_pow2orz_fail_bad_pred(
+; CHECK-NEXT: [[NY:%.*]] = sub <2 x i8> zeroinitializer, [[Y:%.*]]
+; CHECK-NEXT: [[POW2ORZ:%.*]] = and <2 x i8> [[NY]], [[Y]]
+; CHECK-NEXT: [[TMP1:%.*]] = or <2 x i8> [[POW2ORZ]], [[X:%.*]]
+; CHECK-NEXT: [[R:%.*]] = icmp eq <2 x i8> [[TMP1]], zeroinitializer
+; CHECK-NEXT: ret <2 x i1> [[R]]
+;
+ %ny = sub <2 x i8> zeroinitializer, %y
+ %pow2orz = and <2 x i8> %ny, %y
+
+ %c0 = icmp eq <2 x i8> %x, zeroinitializer
+ %cp2 = icmp eq <2 x i8> %x, %pow2orz
+ %r = and <2 x i1> %c0, %cp2
+ ret <2 x i1> %r
+}
+
+define i1 @icmp_eq_or_z_or_pow2orz_fail_bad_pred2(i8 %x, i8 %y) {
+; CHECK-LABEL: @icmp_eq_or_z_or_pow2orz_fail_bad_pred2(
+; CHECK-NEXT: [[NY:%.*]] = sub i8 0, [[Y:%.*]]
+; CHECK-NEXT: [[POW2ORZ:%.*]] = and i8 [[NY]], [[Y]]
+; CHECK-NEXT: [[C0:%.*]] = icmp slt i8 [[X:%.*]], 1
+; CHECK-NEXT: [[CP2:%.*]] = icmp sge i8 [[POW2ORZ]], [[X]]
+; CHECK-NEXT: [[R:%.*]] = or i1 [[C0]], [[CP2]]
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %ny = sub i8 0, %y
+ %pow2orz = and i8 %ny, %y
+
+ %c0 = icmp sle i8 %x, 0
+ %cp2 = icmp sle i8 %x, %pow2orz
+ %r = or i1 %c0, %cp2
+ ret i1 %r
+}
>From 98aa59b063469556d8d98195f123dcb9d7e31245 Mon Sep 17 00:00:00 2001
From: Noah Goldstein <goldstein.w.n at gmail.com>
Date: Fri, 15 Sep 2023 12:47:54 -0500
Subject: [PATCH 13/13] [InstCombine] Add transforms for `(or/and (icmp eq/ne
X,0),(icmp eq/ne X,Pow2OrZero))`
`(or (icmp eq X, 0), (icmp eq X, Pow2OrZero))`
--> `(icmp eq (and X, Pow2OrZero), X)`
`(and (icmp ne X, 0), (icmp ne X, Pow2OrZero))`
--> `(icmp ne (and X, Pow2OrZero), X)`
Proofs: https://alive2.llvm.org/ce/z/nPo2BN
Reviewed By: nikic
Differential Revision: https://reviews.llvm.org/D157312
---
.../InstCombine/InstCombineAndOrXor.cpp | 37 +++++++++++++++++++
.../Transforms/InstCombine/and-or-icmps.ll | 10 ++---
2 files changed, 41 insertions(+), 6 deletions(-)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp b/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
index 5a3d6dd20d60a08..b080cbfed17401d 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
@@ -739,6 +739,38 @@ Value *InstCombinerImpl::simplifyRangeCheck(ICmpInst *Cmp0, ICmpInst *Cmp1,
return Builder.CreateICmp(NewPred, Input, RangeEnd);
}
+// (or (icmp eq X, 0), (icmp eq X, Pow2OrZero))
+// -> (icmp eq (and X, Pow2OrZero), X)
+// (and (icmp ne X, 0), (icmp ne X, Pow2OrZero))
+// -> (icmp ne (and X, Pow2OrZero), X)
+static Value *
+foldAndOrOfICmpsWithPow2AndWithZero(InstCombiner::BuilderTy &Builder,
+ ICmpInst *LHS, ICmpInst *RHS, bool IsAnd,
+ const SimplifyQuery &Q) {
+ CmpInst::Predicate Pred = IsAnd ? CmpInst::ICMP_NE : CmpInst::ICMP_EQ;
+ // Make sure we have right compares for our op.
+ if (LHS->getPredicate() != Pred || RHS->getPredicate() != Pred)
+ return nullptr;
+
+ // Make it so we can match LHS against the (icmp eq/ne X, 0) just for
+ // simplicity.
+ if (match(RHS->getOperand(1), m_Zero()))
+ std::swap(LHS, RHS);
+
+ Value *Pow2, *Op;
+ // Match the desired pattern:
+ // LHS: (icmp eq/ne X, 0)
+ // RHS: (icmp eq/ne X, Pow2OrZero)
+ if (!match(LHS, m_ICmp(Pred, m_Value(Op), m_Zero())) ||
+ !match(RHS, m_c_ICmp(Pred, m_Specific(Op), m_Value(Pow2))) ||
+ !isKnownToBeAPowerOfTwo(Pow2, Q.DL, /*OrZero*/ true, /*Depth*/ 0, Q.AC,
+ Q.CxtI, Q.DT))
+ return nullptr;
+
+ Value *And = Builder.CreateAnd(Op, Pow2);
+ return Builder.CreateICmp(Pred, And, Op);
+}
+
// Fold (iszero(A & K1) | iszero(A & K2)) -> (A & (K1 | K2)) != (K1 | K2)
// Fold (!iszero(A & K1) & !iszero(A & K2)) -> (A & (K1 | K2)) == (K1 | K2)
Value *InstCombinerImpl::foldAndOrOfICmpsOfAndWithPow2(ICmpInst *LHS,
@@ -3102,6 +3134,11 @@ Value *InstCombinerImpl::foldAndOrOfICmps(ICmpInst *LHS, ICmpInst *RHS,
ICmpInst::Predicate PredL = LHS->getPredicate(), PredR = RHS->getPredicate();
Value *LHS0 = LHS->getOperand(0), *RHS0 = RHS->getOperand(0);
Value *LHS1 = LHS->getOperand(1), *RHS1 = RHS->getOperand(1);
+ if (!IsLogical)
+ if (Value *V =
+ foldAndOrOfICmpsWithPow2AndWithZero(Builder, LHS, RHS, IsAnd, Q))
+ return V;
+
const APInt *LHSC = nullptr, *RHSC = nullptr;
match(LHS1, m_APInt(LHSC));
match(RHS1, m_APInt(RHSC));
diff --git a/llvm/test/Transforms/InstCombine/and-or-icmps.ll b/llvm/test/Transforms/InstCombine/and-or-icmps.ll
index 3b43b008367e788..eb91bd7cbf13d7b 100644
--- a/llvm/test/Transforms/InstCombine/and-or-icmps.ll
+++ b/llvm/test/Transforms/InstCombine/and-or-icmps.ll
@@ -3049,9 +3049,8 @@ define i1 @icmp_eq_or_z_or_pow2orz(i8 %x, i8 %y) {
; CHECK-LABEL: @icmp_eq_or_z_or_pow2orz(
; CHECK-NEXT: [[NY:%.*]] = sub i8 0, [[Y:%.*]]
; CHECK-NEXT: [[POW2ORZ:%.*]] = and i8 [[NY]], [[Y]]
-; CHECK-NEXT: [[C0:%.*]] = icmp eq i8 [[X:%.*]], 0
-; CHECK-NEXT: [[CP2:%.*]] = icmp eq i8 [[POW2ORZ]], [[X]]
-; CHECK-NEXT: [[R:%.*]] = or i1 [[C0]], [[CP2]]
+; CHECK-NEXT: [[TMP1:%.*]] = and i8 [[POW2ORZ]], [[X:%.*]]
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[TMP1]], [[X]]
; CHECK-NEXT: ret i1 [[R]]
;
%ny = sub i8 0, %y
@@ -3087,9 +3086,8 @@ define <2 x i1> @icmp_ne_and_z_and_pow2orz(<2 x i8> %x, <2 x i8> %y) {
; CHECK-LABEL: @icmp_ne_and_z_and_pow2orz(
; CHECK-NEXT: [[NY:%.*]] = sub <2 x i8> zeroinitializer, [[Y:%.*]]
; CHECK-NEXT: [[POW2ORZ:%.*]] = and <2 x i8> [[NY]], [[Y]]
-; CHECK-NEXT: [[C0:%.*]] = icmp ne <2 x i8> [[X:%.*]], zeroinitializer
-; CHECK-NEXT: [[CP2:%.*]] = icmp ne <2 x i8> [[POW2ORZ]], [[X]]
-; CHECK-NEXT: [[R:%.*]] = and <2 x i1> [[C0]], [[CP2]]
+; CHECK-NEXT: [[TMP1:%.*]] = and <2 x i8> [[POW2ORZ]], [[X:%.*]]
+; CHECK-NEXT: [[R:%.*]] = icmp ne <2 x i8> [[TMP1]], [[X]]
; CHECK-NEXT: ret <2 x i1> [[R]]
;
%ny = sub <2 x i8> zeroinitializer, %y
More information about the llvm-commits
mailing list