[PATCH] D28587: [SCEV] Make getUDivExactExpr handle non-nuw multiplies correctly
Eli Friedman via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Wed Jan 11 16:56:18 PST 2017
efriedma removed rL LLVM as the repository for this revision.
efriedma updated this revision to Diff 84055.
efriedma added a comment.
Uploaded wrong version of the patch.
https://reviews.llvm.org/D28587
Files:
lib/Analysis/ScalarEvolution.cpp
test/Analysis/ScalarEvolution/pr24757.ll
Index: test/Analysis/ScalarEvolution/pr24757.ll
===================================================================
--- test/Analysis/ScalarEvolution/pr24757.ll
+++ test/Analysis/ScalarEvolution/pr24757.ll
@@ -1,6 +1,6 @@
; RUN: opt -analyze -scalar-evolution < %s | FileCheck %s
-; CHECK: Loop %bb1: backedge-taken count is (zext i7 (trunc i8 %a.promoted to i7) to i8)
+; CHECK: Loop %bb1: backedge-taken count is ((2 * %a.promoted) /u 2)
target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-apple-macosx10.10.0"
Index: lib/Analysis/ScalarEvolution.cpp
===================================================================
--- lib/Analysis/ScalarEvolution.cpp
+++ lib/Analysis/ScalarEvolution.cpp
@@ -2884,7 +2884,7 @@
// end of this file for inspiration.
const SCEVMulExpr *Mul = dyn_cast<SCEVMulExpr>(LHS);
- if (!Mul)
+ if (!Mul || !Mul->hasNoUnsignedWrap())
return getUDivExpr(LHS, RHS);
if (const SCEVConstant *RHSCst = dyn_cast<SCEVConstant>(RHS)) {
@@ -5143,12 +5143,27 @@
APInt EffectiveMask =
APInt::getLowBitsSet(BitWidth, BitWidth - LZ - TZ).shl(TZ);
if ((LZ != 0 || TZ != 0) && !((~A & ~KnownZero) & EffectiveMask)) {
- const SCEV *MulCount = getConstant(ConstantInt::get(
- getContext(), APInt::getOneBitSet(BitWidth, TZ)));
+ const SCEV *MulCount = getConstant(APInt::getOneBitSet(BitWidth, TZ));
+ const SCEV *LHS = getSCEV(BO->LHS);
+ const SCEV *ShiftedLHS = nullptr;
+ if (auto *LHSMul = dyn_cast<SCEVMulExpr>(LHS)) {
+ if (auto *OpC = dyn_cast<SCEVConstant>(LHSMul->getOperand(0))) {
+ // For an expression like (x * 8) & 8, simplify the multiply.
+ unsigned MulZeros = OpC->getAPInt().countTrailingZeros();
+ unsigned GCD = std::min(MulZeros, TZ);
+ APInt DivAmt = APInt::getOneBitSet(BitWidth, TZ - GCD);
+ SmallVector<const SCEV*, 4> MulOps;
+ MulOps.push_back(getConstant(OpC->getAPInt().lshr(GCD)));
+ MulOps.append(LHSMul->op_begin() + 1, LHSMul->op_end());
+ auto *NewMul = getMulExpr(MulOps, LHSMul->getNoWrapFlags());
+ ShiftedLHS = getUDivExpr(NewMul, getConstant(DivAmt));
+ }
+ }
+ if (!ShiftedLHS)
+ ShiftedLHS = getUDivExpr(LHS, MulCount);
return getMulExpr(
getZeroExtendExpr(
- getTruncateExpr(
- getUDivExactExpr(getSCEV(BO->LHS), MulCount),
+ getTruncateExpr(ShiftedLHS,
IntegerType::get(getContext(), BitWidth - LZ - TZ)),
BO->LHS->getType()),
MulCount);
@@ -7260,17 +7275,7 @@
// E.g. if Val is i8 -127 then the smallest value of X that satisfies (3)
// is i8 1, not i8 -127
- const auto *ModuloResult = getUDivExactExpr(Distance, Step);
-
- // Since SCEV does not have a URem node, we construct one using a truncate
- // and a zero extend.
-
- unsigned NarrowWidth = StepV.getBitWidth() - StepV.countTrailingZeros();
- auto *NarrowTy = IntegerType::get(getContext(), NarrowWidth);
- auto *WideTy = Distance->getType();
-
- const SCEV *Limit =
- getZeroExtendExpr(getTruncateExpr(ModuloResult, NarrowTy), WideTy);
+ const auto *Limit = getUDivExactExpr(Distance, Step);
return ExitLimit(Limit, Limit, false, Predicates);
}
}
-------------- next part --------------
A non-text attachment was scrubbed...
Name: D28587.84055.patch
Type: text/x-patch
Size: 3496 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20170112/4458a56c/attachment.bin>
More information about the llvm-commits
mailing list