From eefec589e8bc4db55d7da9070ace47befa021999 Mon Sep 17 00:00:00 2001 From: Adam Nemet Date: Wed, 18 Feb 2015 03:42:43 +0000 Subject: [PATCH] [LoopAccesses] Make VectorizerParams global As LAA is becoming a pass, we can no longer pass the params to its constructor. This changes the command line flags to have external storage. These can now be accessed both from LV and LAA. VectorizerParams is moved out of LoopAccessInfo in order to shorten the code to access it. This is part of the patchset that converts LoopAccessAnalysis into an actual analysis pass. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@229622 91177308-0d34-0410-b5e6-96231b3b80d8 --- include/llvm/Analysis/LoopAccessAnalysis.h | 49 ++++++++-------------- lib/Analysis/LoopAccessAnalysis.cpp | 32 +++++++------- lib/Transforms/Vectorize/LoopVectorize.cpp | 32 +++++++------- 3 files changed, 51 insertions(+), 62 deletions(-) diff --git a/include/llvm/Analysis/LoopAccessAnalysis.h b/include/llvm/Analysis/LoopAccessAnalysis.h index c46276b8698..9af3bf07065 100644 --- a/include/llvm/Analysis/LoopAccessAnalysis.h +++ b/include/llvm/Analysis/LoopAccessAnalysis.h @@ -60,6 +60,22 @@ public: const Loop *TheLoop); }; +/// \brief Collection of parameters shared beetween the Loop Vectorizer and the +/// Loop Access Analysis. +struct VectorizerParams { + /// \brief Maximum SIMD width. + static const unsigned MaxVectorWidth; + + /// \brief VF as overridden by the user. + static unsigned VectorizationFactor; + /// \brief Interleave factor as overridden by the user. + static unsigned VectorizationInterleave; + + /// \\brief When performing memory disambiguation checks at runtime do not + /// make more than this number of comparisons. + static const unsigned RuntimeMemoryCheckThreshold; +}; + /// \brief Drive the analysis of memory accesses in the loop /// /// This class is responsible for analyzing the memory accesses of a loop. It @@ -76,30 +92,6 @@ public: /// RuntimePointerCheck class. class LoopAccessInfo { public: - /// \brief Collection of parameters used from the vectorizer. - struct VectorizerParams { - /// \brief Maximum simd width. - unsigned MaxVectorWidth; - - /// \brief VF as overridden by the user. - unsigned VectorizationFactor; - /// \brief Interleave factor as overridden by the user. - unsigned VectorizationInterleave; - - /// \\brief When performing memory disambiguation checks at runtime do not - /// make more than this number of comparisons. - unsigned RuntimeMemoryCheckThreshold; - - VectorizerParams(unsigned MaxVectorWidth, - unsigned VectorizationFactor, - unsigned VectorizationInterleave, - unsigned RuntimeMemoryCheckThreshold) : - MaxVectorWidth(MaxVectorWidth), - VectorizationFactor(VectorizationFactor), - VectorizationInterleave(VectorizationInterleave), - RuntimeMemoryCheckThreshold(RuntimeMemoryCheckThreshold) {} - }; - /// This struct holds information about the memory runtime legality check that /// a group of pointers do not overlap. struct RuntimePointerCheck { @@ -139,11 +131,9 @@ public: LoopAccessInfo(Function *F, Loop *L, ScalarEvolution *SE, const DataLayout *DL, const TargetLibraryInfo *TLI, - AliasAnalysis *AA, DominatorTree *DT, - const VectorizerParams &VectParams) : + AliasAnalysis *AA, DominatorTree *DT) : TheFunction(F), TheLoop(L), SE(SE), DL(DL), TLI(TLI), AA(AA), DT(DT), - NumLoads(0), NumStores(0), MaxSafeDepDistBytes(-1U), - VectParams(VectParams) {} + NumLoads(0), NumStores(0), MaxSafeDepDistBytes(-1U) {} /// Return true we can analyze the memory accesses in the loop and there are /// no memory dependence cycles. Replaces symbolic strides using Strides. @@ -187,9 +177,6 @@ private: unsigned NumStores; unsigned MaxSafeDepDistBytes; - - /// \brief Vectorizer parameters used by the analysis. - VectorizerParams VectParams; }; Value *stripIntegerCast(Value *V); diff --git a/lib/Analysis/LoopAccessAnalysis.cpp b/lib/Analysis/LoopAccessAnalysis.cpp index 0c30cb63059..9ac27ddb46f 100644 --- a/lib/Analysis/LoopAccessAnalysis.cpp +++ b/lib/Analysis/LoopAccessAnalysis.cpp @@ -437,10 +437,9 @@ public: typedef PointerIntPair MemAccessInfo; typedef SmallPtrSet MemAccessInfoSet; - MemoryDepChecker(ScalarEvolution *Se, const DataLayout *Dl, const Loop *L, - const LoopAccessInfo::VectorizerParams &VectParams) + MemoryDepChecker(ScalarEvolution *Se, const DataLayout *Dl, const Loop *L) : SE(Se), DL(Dl), InnermostLoop(L), AccessIdx(0), - ShouldRetryWithRuntimeCheck(false), VectParams(VectParams) {} + ShouldRetryWithRuntimeCheck(false) {} /// \brief Register the location (instructions are given increasing numbers) /// of a write access. @@ -495,9 +494,6 @@ private: /// vectorize this loop with runtime checks. bool ShouldRetryWithRuntimeCheck; - /// \brief Vectorizer parameters used by the analysis. - LoopAccessInfo::VectorizerParams VectParams; - /// \brief Check whether there is a plausible dependence between the two /// accesses. /// @@ -621,7 +617,8 @@ bool MemoryDepChecker::couldPreventStoreLoadForward(unsigned Distance, // Store-load forwarding distance. const unsigned NumCyclesForStoreLoadThroughMemory = 8*TypeByteSize; // Maximum vector factor. - unsigned MaxVFWithoutSLForwardIssues = VectParams.MaxVectorWidth*TypeByteSize; + unsigned MaxVFWithoutSLForwardIssues = + VectorizerParams::MaxVectorWidth * TypeByteSize; if(MaxSafeDepDistBytes < MaxVFWithoutSLForwardIssues) MaxVFWithoutSLForwardIssues = MaxSafeDepDistBytes; @@ -640,7 +637,8 @@ bool MemoryDepChecker::couldPreventStoreLoadForward(unsigned Distance, } if (MaxVFWithoutSLForwardIssues < MaxSafeDepDistBytes && - MaxVFWithoutSLForwardIssues != VectParams.MaxVectorWidth*TypeByteSize) + MaxVFWithoutSLForwardIssues != + VectorizerParams::MaxVectorWidth * TypeByteSize) MaxSafeDepDistBytes = MaxVFWithoutSLForwardIssues; return false; } @@ -745,10 +743,10 @@ bool MemoryDepChecker::isDependent(const MemAccessInfo &A, unsigned AIdx, unsigned Distance = (unsigned) Val.getZExtValue(); // Bail out early if passed-in parameters make vectorization not feasible. - unsigned ForcedFactor = (VectParams.VectorizationFactor ? - VectParams.VectorizationFactor : 1); - unsigned ForcedUnroll = (VectParams.VectorizationInterleave ? - VectParams.VectorizationInterleave : 1); + unsigned ForcedFactor = (VectorizerParams::VectorizationFactor ? + VectorizerParams::VectorizationFactor : 1); + unsigned ForcedUnroll = (VectorizerParams::VectorizationInterleave ? + VectorizerParams::VectorizationInterleave : 1); // The distance must be bigger than the size needed for a vectorized version // of the operation and the size of the vectorized operation must not be @@ -831,7 +829,7 @@ bool LoopAccessInfo::canVectorizeMemory(ValueToValueMap &Strides) { PtrRtCheck.Need = false; const bool IsAnnotatedParallel = TheLoop->isAnnotatedParallel(); - MemoryDepChecker DepChecker(SE, DL, TheLoop, VectParams); + MemoryDepChecker DepChecker(SE, DL, TheLoop); // For each block. for (Loop::block_iterator bb = TheLoop->block_begin(), @@ -1000,7 +998,8 @@ bool LoopAccessInfo::canVectorizeMemory(ValueToValueMap &Strides) { // Check that we did not collect too many pointers or found an unsizeable // pointer. - if (!CanDoRT || NumComparisons > VectParams.RuntimeMemoryCheckThreshold) { + if (!CanDoRT || + NumComparisons > VectorizerParams::RuntimeMemoryCheckThreshold) { PtrRtCheck.reset(); CanDoRT = false; } @@ -1040,14 +1039,15 @@ bool LoopAccessInfo::canVectorizeMemory(ValueToValueMap &Strides) { TheLoop, Strides, true); // Check that we did not collect too many pointers or found an unsizeable // pointer. - if (!CanDoRT || NumComparisons > VectParams.RuntimeMemoryCheckThreshold) { + if (!CanDoRT || + NumComparisons > VectorizerParams::RuntimeMemoryCheckThreshold) { if (!CanDoRT && NumComparisons > 0) emitAnalysis(VectorizationReport() << "cannot check memory dependencies at runtime"); else emitAnalysis(VectorizationReport() << NumComparisons << " exceeds limit of " - << VectParams.RuntimeMemoryCheckThreshold + << VectorizerParams::RuntimeMemoryCheckThreshold << " dependent memory operations checked at runtime"); DEBUG(dbgs() << "LV: Can't vectorize with memory checks\n"); PtrRtCheck.reset(); diff --git a/lib/Transforms/Vectorize/LoopVectorize.cpp b/lib/Transforms/Vectorize/LoopVectorize.cpp index bd39b8048b1..3f6660203c0 100644 --- a/lib/Transforms/Vectorize/LoopVectorize.cpp +++ b/lib/Transforms/Vectorize/LoopVectorize.cpp @@ -106,14 +106,19 @@ using namespace llvm::PatternMatch; STATISTIC(LoopsVectorized, "Number of loops vectorized"); STATISTIC(LoopsAnalyzed, "Number of loops analyzed for vectorization"); -static cl::opt -VectorizationFactor("force-vector-width", cl::init(0), cl::Hidden, - cl::desc("Sets the SIMD width. Zero is autoselect.")); - -static cl::opt -VectorizationInterleave("force-vector-interleave", cl::init(0), cl::Hidden, - cl::desc("Sets the vectorization interleave count. " - "Zero is autoselect.")); +static cl::opt +VectorizationFactor("force-vector-width", cl::Hidden, + cl::desc("Sets the SIMD width. Zero is autoselect."), + cl::location(VectorizerParams::VectorizationFactor)); +unsigned VectorizerParams::VectorizationFactor = 0; + +static cl::opt +VectorizationInterleave("force-vector-interleave", cl::Hidden, + cl::desc("Sets the vectorization interleave count. " + "Zero is autoselect."), + cl::location( + VectorizerParams::VectorizationInterleave)); +unsigned VectorizerParams::VectorizationInterleave = 0; static cl::opt EnableIfConversion("enable-if-conversion", cl::init(true), cl::Hidden, @@ -147,10 +152,10 @@ static const unsigned TinyTripCountUnrollThreshold = 128; /// When performing memory disambiguation checks at runtime do not make more /// than this number of comparisons. -static const unsigned RuntimeMemoryCheckThreshold = 8; +const unsigned VectorizerParams::RuntimeMemoryCheckThreshold = 8; /// Maximum simd width. -static const unsigned MaxVectorWidth = 64; +const unsigned VectorizerParams::MaxVectorWidth = 64; static cl::opt ForceTargetNumScalarRegs( "force-target-num-scalar-regs", cl::init(0), cl::Hidden, @@ -551,10 +556,7 @@ public: : NumPredStores(0), TheLoop(L), SE(SE), DL(DL), TLI(TLI), TheFunction(F), TTI(TTI), Induction(nullptr), WidestIndTy(nullptr), - LAI(F, L, SE, DL, TLI, AA, DT, - LoopAccessInfo::VectorizerParams( - MaxVectorWidth, VectorizationFactor, VectorizationInterleave, - RuntimeMemoryCheckThreshold)), + LAI(F, L, SE, DL, TLI, AA, DT), HasFunNoNaNAttr(false) {} /// This enum represents the kinds of reductions that we support. @@ -1019,7 +1021,7 @@ class LoopVectorizeHints { bool validate(unsigned Val) { switch (Kind) { case HK_WIDTH: - return isPowerOf2_32(Val) && Val <= MaxVectorWidth; + return isPowerOf2_32(Val) && Val <= VectorizerParams::MaxVectorWidth; case HK_UNROLL: return isPowerOf2_32(Val) && Val <= MaxInterleaveFactor; case HK_FORCE: -- 2.34.1