[BypassSlowDivision] Improve our handling of divisions by constants
Summary:
Don't bail out on constant divisors for divisions that can be narrowed without
introducing control flow . This gives us a 32 bit multiply instead of an
emulated 64 bit multiply in the generated PTX assembly.
Reviewers: jlebar
Subscribers: jholewinski, mcrosier, llvm-commits
Differential Revision: https://reviews.llvm.org/D38265
llvm-svn: 314253
diff --git a/llvm/lib/Transforms/Utils/BypassSlowDivision.cpp b/llvm/lib/Transforms/Utils/BypassSlowDivision.cpp
index d6c31f2..4aed897 100644
--- a/llvm/lib/Transforms/Utils/BypassSlowDivision.cpp
+++ b/llvm/lib/Transforms/Utils/BypassSlowDivision.cpp
@@ -339,11 +339,6 @@
Value *Dividend = SlowDivOrRem->getOperand(0);
Value *Divisor = SlowDivOrRem->getOperand(1);
- if (isa<ConstantInt>(Divisor)) {
- // Keep division by a constant for DAGCombiner.
- return None;
- }
-
VisitedSetTy SetL;
ValueRange DividendRange = getValueRange(Dividend, SetL);
if (DividendRange == VALRNG_LIKELY_LONG)
@@ -359,7 +354,9 @@
if (DividendShort && DivisorShort) {
// If both operands are known to be short then just replace the long
- // division with a short one in-place.
+ // division with a short one in-place. Since we're not introducing control
+ // flow in this case, narrowing the division is always a win, even if the
+ // divisor is a constant (and will later get replaced by a multiplication).
IRBuilder<> Builder(SlowDivOrRem);
Value *TruncDividend = Builder.CreateTrunc(Dividend, BypassType);
@@ -369,7 +366,16 @@
Value *ExtDiv = Builder.CreateZExt(TruncDiv, getSlowType());
Value *ExtRem = Builder.CreateZExt(TruncRem, getSlowType());
return QuotRemPair(ExtDiv, ExtRem);
- } else if (DividendShort && !isSignedOp()) {
+ }
+
+ if (isa<ConstantInt>(Divisor)) {
+ // If the divisor is not a constant, DAGCombiner will convert it to a
+ // multiplication by a magic constant. It isn't clear if it is worth
+ // introducing control flow to get a narrower multiply.
+ return None;
+ }
+
+ if (DividendShort && !isSignedOp()) {
// If the division is unsigned and Dividend is known to be short, then
// either
// 1) Divisor is less or equal to Dividend, and the result can be computed