Don't claim the udiv created in BypassSlowDivision is exact.

[jlebar]

Summary

In BypassSlowDivision's short-dividend path, we would create e.g.

udiv exact i32 %a, %b

"exact" here means that we are asserting that %a is a multiple of %b.
But we have no reason to believe this must be true -- this is just a
bug, as far as I can tell.

[hfinkel]

But we have no reason to believe this must be true -- this is just a bug, as far as I can tell.

I agree. All we know is that the truncation of the operands to the smaller type did not drop any set bits. That says nothing about the division having no remainder. LGTM.

[jlebar]

Thank you for the review!

llvm/trunk/lib/Transforms/Utils/BypassSlowDivision.cpp

static bool insertFastDiv(Instruction *I, IntegerType *BypassType,
   FastBB->moveBefore(SlowBB);
   IRBuilder<> FastBuilder(FastBB, FastBB->begin());
   Value *ShortDivisorV = FastBuilder.CreateCast(Instruction::Trunc, Divisor,
                                                 BypassType);
   Value *ShortDividendV = FastBuilder.CreateCast(Instruction::Trunc, Dividend,
                                                  BypassType);
 
   // udiv/urem because optimization only handles positive numbers
-  Value *ShortQuotientV = FastBuilder.CreateExactUDiv(ShortDividendV,
-                                                      ShortDivisorV);
+  Value *ShortQuotientV = FastBuilder.CreateUDiv(ShortDividendV, ShortDivisorV);
   Value *ShortRemainderV = FastBuilder.CreateURem(ShortDividendV,
                                                   ShortDivisorV);
   Value *FastQuotientV = FastBuilder.CreateCast(Instruction::ZExt,
                                                 ShortQuotientV,
                                                 Dividend->getType());
   Value *FastRemainderV = FastBuilder.CreateCast(Instruction::ZExt,
                                                  ShortRemainderV,
                                                  Dividend->getType());

llvm/trunk/test/Transforms/CodeGenPrepare/NVPTX/bypass-slow-div-not-exact.ll

+; RUN: opt -S -codegenprepare < %s | FileCheck %s
+
+target datalayout = "e-i64:64-v16:16-v32:32-n16:32:64"
+target triple = "nvptx64-nvidia-cuda"
+
+; Check that the smaller-width division that the BypassSlowDivision pass
+; creates is not marked as "exact" (that is, it doesn't claim that the
+; numerator is a multiple of the denominator).
+;
+; CHECK-LABEL: @test
+define void @test(i64 %a, i64 %b, i64* %retptr) {
+  ; CHECK: udiv i32
+  %d = sdiv i64 %a, %b
+  store i64 %d, i64* %retptr
+  ret void
+}