[llvm] r367322 - [InstCombine] Fold "x ?% y ==/!= 0" to "x & (y-1) ==/!= 0" iff y is power-of-two
Roman Lebedev via llvm-commits
llvm-commits at lists.llvm.org
Tue Jul 30 08:28:22 PDT 2019
Author: lebedevri
Date: Tue Jul 30 08:28:22 2019
New Revision: 367322
URL: http://llvm.org/viewvc/llvm-project?rev=367322&view=rev
Log:
[InstCombine] Fold "x ?% y ==/!= 0" to "x & (y-1) ==/!= 0" iff y is power-of-two
Summary:
I have stumbled into this by accident while preparing to extend backend `x s% C ==/!= 0` handling.
While we did happen to handle this fold in most of the cases,
the folding is indirect - we fold `x u% y` to `x & (y-1)` (iff `y` is power-of-two),
or first turn `x s% -y` to `x u% y`; that does handle most of the cases.
But we can't turn `x s% INT_MIN` to `x u% -INT_MIN`,
and thus we end up being stuck with `(x s% INT_MIN) == 0`.
There is no such restriction for the more general fold:
https://rise4fun.com/Alive/IIeS
To be noted, the fold does not enforce that `y` is a constant,
so it may indeed increase instruction count.
This is consistent with what `x u% y`->`x & (y-1)` already does.
I think it makes sense, it's at most one (simple) extra instruction,
while `rem`ainder is really much more un-simple (and likely **very** costly).
Reviewers: spatel, RKSimon, nikic, xbolva00, craig.topper
Reviewed By: RKSimon
Subscribers: hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D65046
Modified:
llvm/trunk/include/llvm/IR/PatternMatch.h
llvm/trunk/lib/Transforms/InstCombine/InstCombineCompares.cpp
llvm/trunk/lib/Transforms/InstCombine/InstCombineInternal.h
llvm/trunk/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp
llvm/trunk/test/Transforms/InstCombine/rem.ll
Modified: llvm/trunk/include/llvm/IR/PatternMatch.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/IR/PatternMatch.h?rev=367322&r1=367321&r2=367322&view=diff
==============================================================================
--- llvm/trunk/include/llvm/IR/PatternMatch.h (original)
+++ llvm/trunk/include/llvm/IR/PatternMatch.h Tue Jul 30 08:28:22 2019
@@ -1003,6 +1003,12 @@ struct is_idiv_op {
}
};
+struct is_irem_op {
+ bool isOpType(unsigned Opcode) {
+ return Opcode == Instruction::SRem || Opcode == Instruction::URem;
+ }
+};
+
/// Matches shift operations.
template <typename LHS, typename RHS>
inline BinOpPred_match<LHS, RHS, is_shift_op> m_Shift(const LHS &L,
@@ -1038,6 +1044,13 @@ inline BinOpPred_match<LHS, RHS, is_idiv
return BinOpPred_match<LHS, RHS, is_idiv_op>(L, R);
}
+/// Matches integer remainder operations.
+template <typename LHS, typename RHS>
+inline BinOpPred_match<LHS, RHS, is_irem_op> m_IRem(const LHS &L,
+ const RHS &R) {
+ return BinOpPred_match<LHS, RHS, is_irem_op>(L, R);
+}
+
//===----------------------------------------------------------------------===//
// Class that matches exact binary ops.
//
Modified: llvm/trunk/lib/Transforms/InstCombine/InstCombineCompares.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/InstCombine/InstCombineCompares.cpp?rev=367322&r1=367321&r2=367322&view=diff
==============================================================================
--- llvm/trunk/lib/Transforms/InstCombine/InstCombineCompares.cpp (original)
+++ llvm/trunk/lib/Transforms/InstCombine/InstCombineCompares.cpp Tue Jul 30 08:28:22 2019
@@ -1317,6 +1317,27 @@ static Instruction *processUGT_ADDCST_AD
return ExtractValueInst::Create(Call, 1, "sadd.overflow");
}
+/// If we have:
+/// icmp eq/ne (urem/srem %x, %y), 0
+/// iff %y is a power-of-two, we can replace this with a bit test:
+/// icmp eq/ne (and %x, (add %y, -1)), 0
+Instruction *InstCombiner::foldIRemByPowerOfTwoToBitTest(ICmpInst &I) {
+ // This fold is only valid for equality predicates.
+ if (!I.isEquality())
+ return nullptr;
+ ICmpInst::Predicate Pred;
+ Value *X, *Y, *Zero;
+ if (!match(&I, m_ICmp(Pred, m_OneUse(m_IRem(m_Value(X), m_Value(Y))),
+ m_CombineAnd(m_Zero(), m_Value(Zero)))))
+ return nullptr;
+ if (!isKnownToBeAPowerOfTwo(Y, /*OrZero*/ true, 0, &I))
+ return nullptr;
+ // This may increase instruction count, we don't enforce that Y is a constant.
+ Value *Mask = Builder.CreateAdd(Y, Constant::getAllOnesValue(Y->getType()));
+ Value *Masked = Builder.CreateAnd(X, Mask);
+ return ICmpInst::Create(Instruction::ICmp, Pred, Masked, Zero);
+}
+
// Handle icmp pred X, 0
Instruction *InstCombiner::foldICmpWithZero(ICmpInst &Cmp) {
CmpInst::Predicate Pred = Cmp.getPredicate();
@@ -1335,6 +1356,9 @@ Instruction *InstCombiner::foldICmpWithZ
}
}
+ if (Instruction *New = foldIRemByPowerOfTwoToBitTest(Cmp))
+ return New;
+
// Given:
// icmp eq/ne (urem %x, %y), 0
// Iff %x has 0 or 1 bits set, and %y has at least 2 bits set, omit 'urem':
Modified: llvm/trunk/lib/Transforms/InstCombine/InstCombineInternal.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/InstCombine/InstCombineInternal.h?rev=367322&r1=367321&r2=367322&view=diff
==============================================================================
--- llvm/trunk/lib/Transforms/InstCombine/InstCombineInternal.h (original)
+++ llvm/trunk/lib/Transforms/InstCombine/InstCombineInternal.h Tue Jul 30 08:28:22 2019
@@ -839,6 +839,7 @@ private:
Instruction *foldICmpInstWithConstantNotInt(ICmpInst &Cmp);
Instruction *foldICmpBinOp(ICmpInst &Cmp);
Instruction *foldICmpEquality(ICmpInst &Cmp);
+ Instruction *foldIRemByPowerOfTwoToBitTest(ICmpInst &I);
Instruction *foldICmpWithZero(ICmpInst &Cmp);
Instruction *foldICmpSelectConstant(ICmpInst &Cmp, SelectInst *Select,
Modified: llvm/trunk/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp?rev=367322&r1=367321&r2=367322&view=diff
==============================================================================
--- llvm/trunk/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp (original)
+++ llvm/trunk/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp Tue Jul 30 08:28:22 2019
@@ -1319,6 +1319,8 @@ Instruction *InstCombiner::visitURem(Bin
Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);
Type *Ty = I.getType();
if (isKnownToBeAPowerOfTwo(Op1, /*OrZero*/ true, 0, &I)) {
+ // This may increase instruction count, we don't enforce that Y is a
+ // constant.
Constant *N1 = Constant::getAllOnesValue(Ty);
Value *Add = Builder.CreateAdd(Op1, N1);
return BinaryOperator::CreateAnd(Op0, Add);
Modified: llvm/trunk/test/Transforms/InstCombine/rem.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/InstCombine/rem.ll?rev=367322&r1=367321&r2=367322&view=diff
==============================================================================
--- llvm/trunk/test/Transforms/InstCombine/rem.ll (original)
+++ llvm/trunk/test/Transforms/InstCombine/rem.ll Tue Jul 30 08:28:22 2019
@@ -201,8 +201,8 @@ define <2 x i19> @weird_vec_power_of_2_c
define i1 @test3a(i32 %A) {
; CHECK-LABEL: @test3a(
-; CHECK-NEXT: [[B1:%.*]] = and i32 [[A:%.*]], 7
-; CHECK-NEXT: [[C:%.*]] = icmp ne i32 [[B1]], 0
+; CHECK-NEXT: [[TMP1:%.*]] = and i32 [[A:%.*]], 7
+; CHECK-NEXT: [[C:%.*]] = icmp ne i32 [[TMP1]], 0
; CHECK-NEXT: ret i1 [[C]]
;
%B = srem i32 %A, -8
@@ -212,8 +212,8 @@ define i1 @test3a(i32 %A) {
define <2 x i1> @test3a_vec(<2 x i32> %A) {
; CHECK-LABEL: @test3a_vec(
-; CHECK-NEXT: [[B1:%.*]] = and <2 x i32> [[A:%.*]], <i32 7, i32 7>
-; CHECK-NEXT: [[C:%.*]] = icmp ne <2 x i32> [[B1]], zeroinitializer
+; CHECK-NEXT: [[TMP1:%.*]] = and <2 x i32> [[A:%.*]], <i32 7, i32 7>
+; CHECK-NEXT: [[C:%.*]] = icmp ne <2 x i32> [[TMP1]], zeroinitializer
; CHECK-NEXT: ret <2 x i1> [[C]]
;
%B = srem <2 x i32> %A, <i32 -8, i32 -8>
@@ -681,8 +681,8 @@ define <2 x i1> @test24_vec(<2 x i32> %A
define i1 @test25(i32 %A) {
; CHECK-LABEL: @test25(
-; CHECK-NEXT: [[B:%.*]] = srem i32 [[A:%.*]], -2147483648
-; CHECK-NEXT: [[C:%.*]] = icmp ne i32 [[B]], 0
+; CHECK-NEXT: [[TMP1:%.*]] = and i32 [[A:%.*]], 2147483647
+; CHECK-NEXT: [[C:%.*]] = icmp ne i32 [[TMP1]], 0
; CHECK-NEXT: ret i1 [[C]]
;
%B = srem i32 %A, 2147483648 ; signbit
@@ -692,8 +692,8 @@ define i1 @test25(i32 %A) {
define <2 x i1> @test25_vec(<2 x i32> %A) {
; CHECK-LABEL: @test25_vec(
-; CHECK-NEXT: [[B:%.*]] = srem <2 x i32> [[A:%.*]], <i32 -2147483648, i32 -2147483648>
-; CHECK-NEXT: [[C:%.*]] = icmp ne <2 x i32> [[B]], zeroinitializer
+; CHECK-NEXT: [[TMP1:%.*]] = and <2 x i32> [[A:%.*]], <i32 2147483647, i32 2147483647>
+; CHECK-NEXT: [[C:%.*]] = icmp ne <2 x i32> [[TMP1]], zeroinitializer
; CHECK-NEXT: ret <2 x i1> [[C]]
;
%B = srem <2 x i32> %A, <i32 2147483648, i32 2147483648>
@@ -703,9 +703,10 @@ define <2 x i1> @test25_vec(<2 x i32> %A
define i1 @test26(i32 %A, i32 %B) {
; CHECK-LABEL: @test26(
-; CHECK-NEXT: [[C:%.*]] = shl nuw i32 1, [[B:%.*]]
-; CHECK-NEXT: [[D:%.*]] = srem i32 [[A:%.*]], [[C]]
-; CHECK-NEXT: [[E:%.*]] = icmp ne i32 [[D]], 0
+; CHECK-NEXT: [[NOTMASK:%.*]] = shl nsw i32 -1, [[B:%.*]]
+; CHECK-NEXT: [[TMP1:%.*]] = xor i32 [[NOTMASK]], -1
+; CHECK-NEXT: [[TMP2:%.*]] = and i32 [[TMP1]], [[A:%.*]]
+; CHECK-NEXT: [[E:%.*]] = icmp ne i32 [[TMP2]], 0
; CHECK-NEXT: ret i1 [[E]]
;
%C = shl i32 1, %B ; not a constant
@@ -729,8 +730,8 @@ define i1 @test27(i32 %A, i32* %remdst)
define i1 @test28(i32 %A) {
; CHECK-LABEL: @test28(
-; CHECK-NEXT: [[B:%.*]] = srem i32 [[A:%.*]], -2147483648
-; CHECK-NEXT: [[C:%.*]] = icmp eq i32 [[B]], 0
+; CHECK-NEXT: [[TMP1:%.*]] = and i32 [[A:%.*]], 2147483647
+; CHECK-NEXT: [[C:%.*]] = icmp eq i32 [[TMP1]], 0
; CHECK-NEXT: ret i1 [[C]]
;
%B = srem i32 %A, 2147483648 ; signbit
More information about the llvm-commits
mailing list