summaryrefslogtreecommitdiffstats
path: root/llvm/lib/Transforms/Vectorize/BBVectorize.cpp
diff options
context:
space:
mode:
authorHal Finkel <hfinkel@anl.gov>2013-02-15 04:28:42 +0000
committerHal Finkel <hfinkel@anl.gov>2013-02-15 04:28:42 +0000
commit283f4f0e66f547b96dcc86e649c2ef393029ea44 (patch)
treeaf14dcb59d819bba6918505de4950e2a6c2c1e2c /llvm/lib/Transforms/Vectorize/BBVectorize.cpp
parent08b088dbbe7716039088930e4f0a25c497f64546 (diff)
downloadbcm5719-llvm-283f4f0e66f547b96dcc86e649c2ef393029ea44.tar.gz
bcm5719-llvm-283f4f0e66f547b96dcc86e649c2ef393029ea44.zip
BBVectorize: Cap the number of candidate pairs in each instruction group
For some basic blocks, it is possible to generate many candidate pairs for relatively few pairable instructions. When many (tens of thousands) of these pairs are generated for a single instruction group, the time taken to generate and rank the different vectorization plans can become quite large. As a result, we now cap the number of candidate pairs within each instruction group. This is done by closing out the group once the threshold is reached (set now at 3000 pairs). Although this will limit the overall compile-time impact, this may not be the best way to achieve this result. It might be better, for example, to prune excessive candidate pairs after the fact the prevent the generation of short, but highly-connected groups. We can experiment with this in the future. This change reduces the overall compile-time slowdown of the csa.ll test case in PR15222 to ~5x. If 5x is still considered too large, a lower limit can be used as the default. This represents a functionality change, but only for very large inputs (thus, there is no regression test). llvm-svn: 175251
Diffstat (limited to 'llvm/lib/Transforms/Vectorize/BBVectorize.cpp')
-rw-r--r--llvm/lib/Transforms/Vectorize/BBVectorize.cpp10
1 files changed, 9 insertions, 1 deletions
diff --git a/llvm/lib/Transforms/Vectorize/BBVectorize.cpp b/llvm/lib/Transforms/Vectorize/BBVectorize.cpp
index 37638ca3c80..4849a968b0d 100644
--- a/llvm/lib/Transforms/Vectorize/BBVectorize.cpp
+++ b/llvm/lib/Transforms/Vectorize/BBVectorize.cpp
@@ -88,6 +88,10 @@ MaxInsts("bb-vectorize-max-instr-per-group", cl::init(500), cl::Hidden,
cl::desc("The maximum number of pairable instructions per group"));
static cl::opt<unsigned>
+MaxPairs("bb-vectorize-max-pairs-per-group", cl::init(3000), cl::Hidden,
+ cl::desc("The maximum number of candidate instruction pairs per group"));
+
+static cl::opt<unsigned>
MaxCandPairsForCycleCheck("bb-vectorize-max-cycle-check-pairs", cl::init(200),
cl::Hidden, cl::desc("The maximum number of candidate pairs with which to use"
" a full cycle check"));
@@ -1164,6 +1168,7 @@ namespace {
DenseSet<ValuePair> &FixedOrderPairs,
DenseMap<ValuePair, int> &CandidatePairCostSavings,
std::vector<Value *> &PairableInsts, bool NonPow2Len) {
+ size_t TotalPairs = 0;
BasicBlock::iterator E = BB.end();
if (Start == E) return false;
@@ -1210,6 +1215,7 @@ namespace {
}
CandidatePairs[I].push_back(J);
+ ++TotalPairs;
if (TTI)
CandidatePairCostSavings.insert(ValuePairWithCost(ValuePair(I, J),
CostSavings));
@@ -1233,7 +1239,8 @@ namespace {
// If we have already found too many pairs, break here and this function
// will be called again starting after the last instruction selected
// during this invocation.
- if (PairableInsts.size() >= Config.MaxInsts) {
+ if (PairableInsts.size() >= Config.MaxInsts ||
+ TotalPairs >= Config.MaxPairs) {
ShouldContinue = true;
break;
}
@@ -3165,6 +3172,7 @@ VectorizeConfig::VectorizeConfig() {
MaxCandPairsForCycleCheck = ::MaxCandPairsForCycleCheck;
SplatBreaksChain = ::SplatBreaksChain;
MaxInsts = ::MaxInsts;
+ MaxPairs = ::MaxPairs;
MaxIter = ::MaxIter;
Pow2LenOnly = ::Pow2LenOnly;
NoMemOpBoost = ::NoMemOpBoost;
OpenPOWER on IntegriCloud