Index: llvm/lib/Analysis/ScalarEvolution.cpp =================================================================== --- llvm/lib/Analysis/ScalarEvolution.cpp +++ llvm/lib/Analysis/ScalarEvolution.cpp @@ -2397,6 +2397,14 @@ Ops[0]->isZero() && IsKnownNonNegative(Ops[1])) Flags = ScalarEvolution::setFlags(Flags, SCEV::FlagNUW); + // (udiv X, Y) * Y is always NUW + if (Type == scMulExpr && !ScalarEvolution::hasFlags(Flags, SCEV::FlagNUW) && + Ops.size() == 2) { + auto *UDiv = dyn_cast(Ops[0]); + if (UDiv && UDiv->getNumOperands() == 2 && UDiv->getOperand(1) == Ops[1]) + Flags = ScalarEvolution::setFlags(Flags, SCEV::FlagNUW); + } + return Flags; } Index: llvm/test/Analysis/ScalarEvolution/extract-highbits-variablemask.ll =================================================================== --- llvm/test/Analysis/ScalarEvolution/extract-highbits-variablemask.ll +++ llvm/test/Analysis/ScalarEvolution/extract-highbits-variablemask.ll @@ -10,7 +10,7 @@ ; CHECK-NEXT: %tmp1 = udiv i32 %val, %num ; CHECK-NEXT: --> (%val /u %num) U: full-set S: full-set ; CHECK-NEXT: %tmp2 = mul i32 %tmp1, %num -; CHECK-NEXT: --> ((%val /u %num) * %num) U: full-set S: full-set +; CHECK-NEXT: --> ((%val /u %num) * %num) U: full-set S: full-set ; CHECK-NEXT: Determining loop execution counts for: @div ; %tmp1 = udiv i32 %val, %num Index: llvm/test/Analysis/ScalarEvolution/mul.ll =================================================================== --- /dev/null +++ llvm/test/Analysis/ScalarEvolution/mul.ll @@ -0,0 +1,16 @@ +; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py +; RUN: opt < %s "-passes=print" -disable-output 2>&1 | FileCheck %s + +define i8 @test(i8 %x, i8 %y) { +; CHECK-LABEL: 'test' +; CHECK-NEXT: Classifying expressions for: @test +; CHECK-NEXT: %udiv = udiv i8 %x, %y +; CHECK-NEXT: --> (%x /u %y) U: full-set S: full-set +; CHECK-NEXT: %res = mul i8 %udiv, %y +; CHECK-NEXT: --> ((%x /u %y) * %y) U: full-set S: full-set +; CHECK-NEXT: Determining loop execution counts for: @test +; + %udiv = udiv i8 %x, %y + %res = mul i8 %udiv, %y + ret i8 %res +}