[llvm] [AArch64][LoopVectorize] Use upper bound trip count instead of the constant TC when choosing max VF (PR #67697)
via llvm-commits
llvm-commits at lists.llvm.org
Fri Oct 6 02:41:24 PDT 2023
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-llvm-transforms
<details>
<summary>Changes</summary>
This patch is based off of https://github.com/llvm/llvm-project/pull/67543. It should not be merged before the previous PR.
We are currently using the exact trip count to make decisions regarding the maximum VF. We can instead use the upper bound TC, which will be the same as the constant trip count when that is known.
---
Full diff: https://github.com/llvm/llvm-project/pull/67697.diff
1 Files Affected:
- (modified) llvm/lib/Transforms/Vectorize/LoopVectorize.cpp (+34-31)
``````````diff
diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
index 53ad37bf3599b5c..26bf92d7d7c02be 100644
--- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -1663,17 +1663,17 @@ class LoopVectorizationCostModel {
/// disabled or unsupported, then the scalable part will be equal to
/// ElementCount::getScalable(0).
FixedScalableVFPair computeFeasibleMaxVF(unsigned ConstTripCount,
+ unsigned MaxTripCount,
ElementCount UserVF,
bool FoldTailByMasking);
/// \return the maximized element count based on the targets vector
/// registers and the loop trip-count, but limited to a maximum safe VF.
/// This is a helper function of computeFeasibleMaxVF.
- ElementCount getMaximizedVFForTarget(unsigned ConstTripCount,
- unsigned SmallestType,
- unsigned WidestType,
- ElementCount MaxSafeVF,
- bool FoldTailByMasking);
+ ElementCount
+ getMaximizedVFForTarget(unsigned ConstTripCount, unsigned MaxTripCount,
+ unsigned SmallestType, unsigned WidestType,
+ ElementCount MaxSafeVF, bool FoldTailByMasking);
/// \return the maximum legal scalable VF, based on the safe max number
/// of elements.
@@ -4811,7 +4811,8 @@ LoopVectorizationCostModel::getMaxLegalScalableVF(unsigned MaxSafeElements) {
}
FixedScalableVFPair LoopVectorizationCostModel::computeFeasibleMaxVF(
- unsigned ConstTripCount, ElementCount UserVF, bool FoldTailByMasking) {
+ unsigned ConstTripCount, unsigned MaxTripCount, ElementCount UserVF,
+ bool FoldTailByMasking) {
MinBWs = computeMinimumValueSizes(TheLoop->getBlocks(), *DB, &TTI);
unsigned SmallestType, WidestType;
std::tie(SmallestType, WidestType) = getSmallestAndWidestTypes();
@@ -4898,14 +4899,14 @@ FixedScalableVFPair LoopVectorizationCostModel::computeFeasibleMaxVF(
FixedScalableVFPair Result(ElementCount::getFixed(1),
ElementCount::getScalable(0));
- if (auto MaxVF =
- getMaximizedVFForTarget(ConstTripCount, SmallestType, WidestType,
- MaxSafeFixedVF, FoldTailByMasking))
+ if (auto MaxVF = getMaximizedVFForTarget(ConstTripCount, MaxTripCount,
+ SmallestType, WidestType,
+ MaxSafeFixedVF, FoldTailByMasking))
Result.FixedVF = MaxVF;
- if (auto MaxVF =
- getMaximizedVFForTarget(ConstTripCount, SmallestType, WidestType,
- MaxSafeScalableVF, FoldTailByMasking))
+ if (auto MaxVF = getMaximizedVFForTarget(
+ ConstTripCount, MaxTripCount, SmallestType, WidestType,
+ MaxSafeScalableVF, FoldTailByMasking))
if (MaxVF.isScalable()) {
Result.ScalableVF = MaxVF;
LLVM_DEBUG(dbgs() << "LV: Found feasible scalable VF = " << MaxVF
@@ -4928,6 +4929,7 @@ LoopVectorizationCostModel::computeMaxVF(ElementCount UserVF, unsigned UserIC) {
}
unsigned TC = PSE.getSE()->getSmallConstantTripCount(TheLoop);
+ unsigned MaxTC = PSE.getSE()->getSmallConstantMaxTripCount(TheLoop);
LLVM_DEBUG(dbgs() << "LV: Found trip count: " << TC << '\n');
if (TC == 1) {
reportVectorizationFailure("Single iteration (non) loop",
@@ -4938,7 +4940,7 @@ LoopVectorizationCostModel::computeMaxVF(ElementCount UserVF, unsigned UserIC) {
switch (ScalarEpilogueStatus) {
case CM_ScalarEpilogueAllowed:
- return computeFeasibleMaxVF(TC, UserVF, false);
+ return computeFeasibleMaxVF(TC, MaxTC, UserVF, false);
case CM_ScalarEpilogueNotAllowedUsePredicate:
[[fallthrough]];
case CM_ScalarEpilogueNotNeededUsePredicate:
@@ -4976,7 +4978,7 @@ LoopVectorizationCostModel::computeMaxVF(ElementCount UserVF, unsigned UserIC) {
LLVM_DEBUG(dbgs() << "LV: Cannot fold tail by masking: vectorize with a "
"scalar epilogue instead.\n");
ScalarEpilogueStatus = CM_ScalarEpilogueAllowed;
- return computeFeasibleMaxVF(TC, UserVF, false);
+ return computeFeasibleMaxVF(TC, MaxTC, UserVF, false);
}
return FixedScalableVFPair::getNone();
}
@@ -4993,7 +4995,8 @@ LoopVectorizationCostModel::computeMaxVF(ElementCount UserVF, unsigned UserIC) {
InterleaveInfo.invalidateGroupsRequiringScalarEpilogue();
}
- FixedScalableVFPair MaxFactors = computeFeasibleMaxVF(TC, UserVF, true);
+ FixedScalableVFPair MaxFactors =
+ computeFeasibleMaxVF(TC, MaxTC, UserVF, true);
// Avoid tail folding if the trip count is known to be a multiple of any VF
// we choose.
@@ -5069,8 +5072,8 @@ LoopVectorizationCostModel::computeMaxVF(ElementCount UserVF, unsigned UserIC) {
}
ElementCount LoopVectorizationCostModel::getMaximizedVFForTarget(
- unsigned ConstTripCount, unsigned SmallestType, unsigned WidestType,
- ElementCount MaxSafeVF, bool FoldTailByMasking) {
+ unsigned ConstTripCount, unsigned MaxTripCount, unsigned SmallestType,
+ unsigned WidestType, ElementCount MaxSafeVF, bool FoldTailByMasking) {
bool ComputeScalableMaxVF = MaxSafeVF.isScalable();
const TypeSize WidestRegister = TTI.getRegisterBitWidth(
ComputeScalableMaxVF ? TargetTransformInfo::RGK_ScalableVector
@@ -5108,24 +5111,24 @@ ElementCount LoopVectorizationCostModel::getMaximizedVFForTarget(
}
// When a scalar epilogue is required, at least one iteration of the scalar
- // loop has to execute. Adjust ConstTripCount accordingly to avoid picking a
+ // loop has to execute. Adjust MaxTripCount accordingly to avoid picking a
// max VF that results in a dead vector loop.
- if (ConstTripCount > 0 && requiresScalarEpilogue(true))
- ConstTripCount -= 1;
-
- if (ConstTripCount && ConstTripCount <= WidestRegisterMinEC &&
- (!FoldTailByMasking || isPowerOf2_32(ConstTripCount))) {
- // If loop trip count (TC) is known at compile time there is no point in
- // choosing VF greater than TC (as done in the loop below). Select maximum
- // power of two which doesn't exceed TC.
- // If MaxVectorElementCount is scalable, we only fall back on a fixed VF
- // when the TC is less than or equal to the known number of lanes.
- auto ClampedConstTripCount = llvm::bit_floor(ConstTripCount);
+ if (MaxTripCount > 0 && requiresScalarEpilogue(true))
+ MaxTripCount -= 1;
+
+ if (MaxTripCount && MaxTripCount <= WidestRegisterMinEC &&
+ (!FoldTailByMasking || isPowerOf2_32(MaxTripCount))) {
+ // If upper bound loop trip count (TC) is known at compile time there is no
+ // point in choosing VF greater than TC (as done in the loop below). Select
+ // maximum power of two which doesn't exceed TC. If MaxVectorElementCount is
+ // scalable, we only fall back on a fixed VF when the TC is less than or
+ // equal to the known number of lanes.
+ auto ClampedUpperTripCount = llvm::bit_floor(MaxTripCount);
LLVM_DEBUG(dbgs() << "LV: Clamping the MaxVF to maximum power of two not "
"exceeding the constant trip count: "
- << ClampedConstTripCount << "\n");
+ << ClampedUpperTripCount << "\n");
return ElementCount::get(
- ClampedConstTripCount,
+ ClampedUpperTripCount,
FoldTailByMasking ? MaxVectorElementCount.isScalable() : false);
}
``````````
</details>
https://github.com/llvm/llvm-project/pull/67697
More information about the llvm-commits
mailing list