[llvm-commits] [llvm] r167324 - in /llvm/trunk: include/llvm/Analysis/Passes.h include/llvm/InitializePasses.h include/llvm/LinkAllPasses.h lib/Analysis/Analysis.cpp lib/Analysis/CMakeLists.txt lib/Analysis/CostModel.cpp test/Analysis/CostModel/ te...
David Tweed
david.tweed at arm.com
Mon Nov 5 05:10:03 PST 2012
ReHi,
About the only comment I've got is that, while this is clearly an initial
stub, supposing one were to obtain independent information about the
vectorisation cost from somewhere for various CPUs, would there be enough
freedom to move the internals to accommodate a huge set of tables detailing
costs per CPU? It looks like there's freedom to plug in any other
implementation behind the interface, but it's probably as well to check. (I
also note that this is based up on the cost for a single instruction, which
is probably the way to go for simplicity, at least for the initial
implementation.)
Regards,
Dave
-----Original Message-----
From: llvm-commits-bounces at cs.uiuc.edu
[mailto:llvm-commits-bounces at cs.uiuc.edu] On Behalf Of Hal Finkel
Sent: 02 November 2012 22:00
To: Nadav Rotem
Cc: llvm-commits at cs.uiuc.edu
Subject: Re: [llvm-commits] [llvm] r167324 - in /llvm/trunk:
include/llvm/Analysis/Passes.h include/llvm/InitializePasses.h
include/llvm/LinkAllPasses.h lib/Analysis/Analysis.cpp
lib/Analysis/CMakeLists.txt lib/Analysis/CostModel.cpp
test/Analysis/CostModel/ te...
----- Original Message -----
> From: "Nadav Rotem" <nrotem at apple.com>
> To: llvm-commits at cs.uiuc.edu
> Sent: Friday, November 2, 2012 4:48:18 PM
> Subject: [llvm-commits] [llvm] r167324 - in /llvm/trunk:
include/llvm/Analysis/Passes.h
> include/llvm/InitializePasses.h include/llvm/LinkAllPasses.h
lib/Analysis/Analysis.cpp lib/Analysis/CMakeLists.txt
> lib/Analysis/CostModel.cpp test/Analysis/CostModel/
test/Analysis/CostModel/X86/
> test/Analysis/CostModel/X86/lit.local.cfg
test/Analysis/CostModel/X86/tiny.ll
> test/Analysis/CostModel/X86/vectorized-loop.ll
test/Analysis/CostModel/lit.local.cfg
> test/Analysis/CostModel/no_info.ll
>
> Author: nadav
> Date: Fri Nov 2 16:48:17 2012
> New Revision: 167324
>
> URL: http://llvm.org/viewvc/llvm-project?rev=167324&view=rev
> Log:
> Add a cost model analysis that allows us to estimate the cost of
> IR-level instructions.
>
> Added:
> llvm/trunk/lib/Analysis/CostModel.cpp
> llvm/trunk/test/Analysis/CostModel/
> llvm/trunk/test/Analysis/CostModel/X86/
> llvm/trunk/test/Analysis/CostModel/X86/lit.local.cfg
> llvm/trunk/test/Analysis/CostModel/X86/tiny.ll
> llvm/trunk/test/Analysis/CostModel/X86/vectorized-loop.ll
> llvm/trunk/test/Analysis/CostModel/lit.local.cfg
> llvm/trunk/test/Analysis/CostModel/no_info.ll
> Modified:
> llvm/trunk/include/llvm/Analysis/Passes.h
> llvm/trunk/include/llvm/InitializePasses.h
> llvm/trunk/include/llvm/LinkAllPasses.h
> llvm/trunk/lib/Analysis/Analysis.cpp
> llvm/trunk/lib/Analysis/CMakeLists.txt
>
> Modified: llvm/trunk/include/llvm/Analysis/Passes.h
> URL:
>
http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Analysis/Passes.
h?rev=167324&r1=167323&r2=167324&view=diff
>
============================================================================
==
> --- llvm/trunk/include/llvm/Analysis/Passes.h (original)
> +++ llvm/trunk/include/llvm/Analysis/Passes.h Fri Nov 2 16:48:17
> 2012
> @@ -187,6 +187,13 @@
>
>
//===--------------------------------------------------------------------===
//
> //
> + // createCostModelAnalysisPass - This creates an instance of the
> + // CostModelAnalysis pass.
> + //
> + FunctionPass *createCostModelAnalysisPass();
> +
> +
>
//===--------------------------------------------------------------------===
//
> + //
> // Minor pass prototypes, allowing us to expose them through
> bugpoint and
> // analyze.
> FunctionPass *createInstCountPass();
>
> Modified: llvm/trunk/include/llvm/InitializePasses.h
> URL:
>
http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/InitializePasses
.h?rev=167324&r1=167323&r2=167324&view=diff
>
============================================================================
==
> --- llvm/trunk/include/llvm/InitializePasses.h (original)
> +++ llvm/trunk/include/llvm/InitializePasses.h Fri Nov 2 16:48:17
> 2012
> @@ -88,6 +88,7 @@
> void initializeConstantMergePass(PassRegistry&);
> void initializeConstantPropagationPass(PassRegistry&);
> void initializeMachineCopyPropagationPass(PassRegistry&);
> +void initializeCostModelAnalysisPass(PassRegistry&);
> void initializeCorrelatedValuePropagationPass(PassRegistry&);
> void initializeDAEPass(PassRegistry&);
> void initializeDAHPass(PassRegistry&);
>
> Modified: llvm/trunk/include/llvm/LinkAllPasses.h
> URL:
>
http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/LinkAllPasses.h?
rev=167324&r1=167323&r2=167324&view=diff
>
============================================================================
==
> --- llvm/trunk/include/llvm/LinkAllPasses.h (original)
> +++ llvm/trunk/include/llvm/LinkAllPasses.h Fri Nov 2 16:48:17 2012
> @@ -60,6 +60,7 @@
> (void) llvm::createCFGSimplificationPass();
> (void) llvm::createConstantMergePass();
> (void) llvm::createConstantPropagationPass();
> + (void) llvm::createCostModelAnalysisPass();
> (void) llvm::createDeadArgEliminationPass();
> (void) llvm::createDeadCodeEliminationPass();
> (void) llvm::createDeadInstEliminationPass();
>
> Modified: llvm/trunk/lib/Analysis/Analysis.cpp
> URL:
>
http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Analysis/Analysis.cpp?rev
=167324&r1=167323&r2=167324&view=diff
>
============================================================================
==
> --- llvm/trunk/lib/Analysis/Analysis.cpp (original)
> +++ llvm/trunk/lib/Analysis/Analysis.cpp Fri Nov 2 16:48:17 2012
> @@ -26,6 +26,7 @@
> initializeBasicAliasAnalysisPass(Registry);
> initializeBlockFrequencyInfoPass(Registry);
> initializeBranchProbabilityInfoPass(Registry);
> + initializeCostModelAnalysisPass(Registry);
> initializeCFGViewerPass(Registry);
> initializeCFGPrinterPass(Registry);
> initializeCFGOnlyViewerPass(Registry);
>
> Modified: llvm/trunk/lib/Analysis/CMakeLists.txt
> URL:
>
http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Analysis/CMakeLists.txt?r
ev=167324&r1=167323&r2=167324&view=diff
>
============================================================================
==
> --- llvm/trunk/lib/Analysis/CMakeLists.txt (original)
> +++ llvm/trunk/lib/Analysis/CMakeLists.txt Fri Nov 2 16:48:17 2012
> @@ -10,6 +10,7 @@
> BranchProbabilityInfo.cpp
> CFGPrinter.cpp
> CaptureTracking.cpp
> + CostModel.cpp
> CodeMetrics.cpp
> ConstantFolding.cpp
> DbgInfoPrinter.cpp
>
> Added: llvm/trunk/lib/Analysis/CostModel.cpp
> URL:
>
http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Analysis/CostModel.cpp?re
v=167324&view=auto
>
============================================================================
==
> --- llvm/trunk/lib/Analysis/CostModel.cpp (added)
> +++ llvm/trunk/lib/Analysis/CostModel.cpp Fri Nov 2 16:48:17 2012
> @@ -0,0 +1,175 @@
> +//===- CostModel.cpp ------ Cost Model Analysis
> ---------------------------===//
> +//
> +// The LLVM Compiler Infrastructure
> +//
> +// This file is distributed under the University of Illinois Open
> Source
> +// License. See LICENSE.TXT for details.
> +//
>
+//===----------------------------------------------------------------------
===//
> +//
> +// This file defines the cost model analysis. It provides a very
> basic cost
> +// estimation for LLVM-IR. The cost result can be thought of as
> cycles, but it
> +// is really unit-less. The estimated cost is ment to be used for
> comparing
> +// alternatives.
> +//
>
+//===----------------------------------------------------------------------
===//
> +
> +#define CM_NAME "cost-model"
> +#define DEBUG_TYPE CM_NAME
> +#include "llvm/Analysis/Passes.h"
> +#include "llvm/Function.h"
> +#include "llvm/Instructions.h"
> +#include "llvm/Pass.h"
> +#include "llvm/TargetTransformInfo.h"
> +#include "llvm/Value.h"
> +#include "llvm/Support/Debug.h"
> +#include "llvm/Support/raw_ostream.h"
> +using namespace llvm;
> +
> +namespace {
> + class CostModelAnalysis : public FunctionPass {
> +
> + public:
> + static char ID; // Class identification, replacement for
> typeinfo
> + CostModelAnalysis() : FunctionPass(ID), F(0), VTTI(0) {
> + initializeCostModelAnalysisPass(
> + *PassRegistry::getPassRegistry());
> + }
> +
> + /// Returns the expected cost of the instruction.
> + /// Returns -1 if the cost is unknown.
> + /// Note, this method does not cache the cost calculation and it
> + /// can be expensive in some cases.
> + unsigned getInstructionCost(Instruction *I) const;
> +
> + private:
> + virtual void getAnalysisUsage(AnalysisUsage &AU) const;
> + virtual bool runOnFunction(Function &F);
> + virtual void print(raw_ostream &OS, const Module*) const;
> +
> + /// The function that we analyze.
> + Function *F;
> + /// Vector target information.
> + const VectorTargetTransformInfo *VTTI;
> + };
> +} // End of anonymous namespace
> +
> +// Register this pass.
> +char CostModelAnalysis::ID = 0;
> +static const char cm_name[] = "Cost Model Analysis";
> +INITIALIZE_PASS_BEGIN(CostModelAnalysis, CM_NAME, cm_name, false,
> true)
> +INITIALIZE_PASS_END (CostModelAnalysis, CM_NAME, cm_name, false,
> true)
> +
> +FunctionPass *llvm::createCostModelAnalysisPass() {
> + return new CostModelAnalysis();
> +}
> +
> +void
> +CostModelAnalysis::getAnalysisUsage(AnalysisUsage &AU) const {
> + AU.setPreservesAll();
> +}
> +
> +bool
> +CostModelAnalysis::runOnFunction(Function &F) {
> + this->F = &F;
> +
> + // Target information.
> + TargetTransformInfo *TTI;
> + TTI = getAnalysisIfAvailable<TargetTransformInfo>();
> + if (TTI)
> + VTTI = TTI->getVectorTargetTransformInfo();
> +
> + return false;
> +}
> +
> +unsigned CostModelAnalysis::getInstructionCost(Instruction *I) const
> {
> + if (!VTTI)
> + return -1;
> +
> + switch (I->getOpcode()) {
> + case Instruction::Ret:
> + case Instruction::PHI:
> + case Instruction::Br: {
> + return VTTI->getCFInstrCost(I->getOpcode());
> + }
> + case Instruction::Add:
> + case Instruction::FAdd:
> + case Instruction::Sub:
> + case Instruction::FSub:
> + case Instruction::Mul:
> + case Instruction::FMul:
> + case Instruction::UDiv:
> + case Instruction::SDiv:
> + case Instruction::FDiv:
> + case Instruction::URem:
> + case Instruction::SRem:
> + case Instruction::FRem:
> + case Instruction::Shl:
> + case Instruction::LShr:
> + case Instruction::AShr:
> + case Instruction::And:
> + case Instruction::Or:
> + case Instruction::Xor: {
> + return VTTI->getArithmeticInstrCost(I->getOpcode(),
> I->getType());
> + }
> + case Instruction::Select: {
> + SelectInst *SI = cast<SelectInst>(I);
> + Type *CondTy = SI->getCondition()->getType();
> + return VTTI->getCmpSelInstrCost(I->getOpcode(), I->getType(),
> CondTy);
> + }
> + case Instruction::ICmp:
> + case Instruction::FCmp: {
> + Type *ValTy = I->getOperand(0)->getType();
> + return VTTI->getCmpSelInstrCost(I->getOpcode(), ValTy);
> + }
> + case Instruction::Store: {
> + StoreInst *SI = cast<StoreInst>(I);
> + Type *ValTy = SI->getValueOperand()->getType();
> + return VTTI->getMemoryOpCost(I->getOpcode(), ValTy,
> + SI->getAlignment(),
> + SI->getPointerAddressSpace());
> + }
> + case Instruction::Load: {
> + LoadInst *LI = cast<LoadInst>(I);
> + return VTTI->getMemoryOpCost(I->getOpcode(), I->getType(),
> + LI->getAlignment(),
> + LI->getPointerAddressSpace());
> + }
> + case Instruction::ZExt:
> + case Instruction::SExt:
> + case Instruction::FPToUI:
> + case Instruction::FPToSI:
> + case Instruction::FPExt:
> + case Instruction::PtrToInt:
> + case Instruction::IntToPtr:
> + case Instruction::SIToFP:
> + case Instruction::UIToFP:
> + case Instruction::Trunc:
> + case Instruction::FPTrunc:
> + case Instruction::BitCast: {
> + Type *SrcTy = I->getOperand(0)->getType();
> + return VTTI->getCastInstrCost(I->getOpcode(), I->getType(),
> SrcTy);
> + }
> + default:
> + // We don't have any information on this instruction.
This is great! We need the vector instructions too ;)
Thanks again,
Hal
> + return -1;
> + }
> +}
> +
> +void CostModelAnalysis::print(raw_ostream &OS, const Module*) const
> {
> + if (!F)
> + return;
> +
> + for (Function::iterator B = F->begin(), BE = F->end(); B != BE;
> ++B) {
> + for (BasicBlock::iterator it = B->begin(), e = B->end(); it !=
> e; ++it) {
> + Instruction *Inst = it;
> + unsigned Cost = getInstructionCost(Inst);
> + if (Cost != (unsigned)-1)
> + OS << "Cost Model: Found an estimated cost of " << Cost;
> + else
> + OS << "Cost Model: Unknown cost";
> +
> + OS << " for instruction: "<< *Inst << "\n";
> + }
> + }
> +}
>
> Added: llvm/trunk/test/Analysis/CostModel/X86/lit.local.cfg
> URL:
>
http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Analysis/CostModel/X86/l
it.local.cfg?rev=167324&view=auto
>
============================================================================
==
> --- llvm/trunk/test/Analysis/CostModel/X86/lit.local.cfg (added)
> +++ llvm/trunk/test/Analysis/CostModel/X86/lit.local.cfg Fri Nov 2
> 16:48:17 2012
> @@ -0,0 +1,6 @@
> +config.suffixes = ['.ll', '.c', '.cpp']
> +
> +targets = set(config.root.targets_to_build.split())
> +if not 'X86' in targets:
> + config.unsupported = True
> +
>
> Added: llvm/trunk/test/Analysis/CostModel/X86/tiny.ll
> URL:
>
http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Analysis/CostModel/X86/t
iny.ll?rev=167324&view=auto
>
============================================================================
==
> --- llvm/trunk/test/Analysis/CostModel/X86/tiny.ll (added)
> +++ llvm/trunk/test/Analysis/CostModel/X86/tiny.ll Fri Nov 2
> 16:48:17 2012
> @@ -0,0 +1,11 @@
> +; RUN: opt < %s -cost-model -analyze
> -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx | FileCheck %s
> +
> +target datalayout =
>
"e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:6
4-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
> +target triple = "x86_64-apple-macosx10.8.0"
> +
> +;CHECK: cost of 1 {{.*}} add
> +;CHECK: cost of 1 {{.*}} ret
> +define i32 @no_info(i32 %arg) {
> + %e = add i32 %arg, %arg
> + ret i32 %e
> +}
>
> Added: llvm/trunk/test/Analysis/CostModel/X86/vectorized-loop.ll
> URL:
>
http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Analysis/CostModel/X86/v
ectorized-loop.ll?rev=167324&view=auto
>
============================================================================
==
> --- llvm/trunk/test/Analysis/CostModel/X86/vectorized-loop.ll (added)
> +++ llvm/trunk/test/Analysis/CostModel/X86/vectorized-loop.ll Fri Nov
> 2 16:48:17 2012
> @@ -0,0 +1,76 @@
> +; RUN: opt < %s -cost-model -analyze
> -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx | FileCheck %s
> +
> +target datalayout =
>
"e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:6
4-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
> +target triple = "x86_64-apple-macosx10.8.0"
> +
> +define i32 @foo(i32* noalias nocapture %A, i32* noalias nocapture
> %B, i32 %start, i32 %end) nounwind uwtable ssp {
> +entry:
> + ;CHECK: cost of 1 {{.*}} icmp
> + %cmp7 = icmp slt i32 %start, %end
> + br i1 %cmp7, label %for.body.lr.ph, label %for.end
> +
> +for.body.lr.ph: ; preds = %entry
> + ;CHECK: cost of 1 {{.*}} sext
> + %0 = sext i32 %start to i64
> + %1 = sub i32 %end, %start
> + %2 = zext i32 %1 to i64
> + %end.idx = add i64 %2, %0
> + ;CHECK: cost of 1 {{.*}} add
> + %n.vec = and i64 %2, 4294967288
> + %end.idx.rnd.down = add i64 %n.vec, %0
> + ;CHECK: cost of 1 {{.*}} icmp
> + %cmp.zero = icmp eq i64 %n.vec, 0
> + br i1 %cmp.zero, label %middle.block, label %vector.body
> +
> +vector.body: ; preds =
> %for.body.lr.ph, %vector.body
> + %index = phi i64 [ %index.next, %vector.body ], [ %0,
> %for.body.lr.ph ]
> + %3 = add i64 %index, 2
> + %4 = getelementptr inbounds i32* %B, i64 %3
> + ;CHECK: cost of 0 {{.*}} bitcast
> + %5 = bitcast i32* %4 to <8 x i32>*
> + ;CHECK: cost of 1 {{.*}} load
> + %6 = load <8 x i32>* %5, align 4
> + %7 = mul nsw <8 x i32> %6, <i32 5, i32 5, i32 5, i32 5, i32 5, i32
> 5, i32 5, i32 5>
> + %8 = getelementptr inbounds i32* %A, i64 %index
> + %9 = bitcast i32* %8 to <8 x i32>*
> + %10 = load <8 x i32>* %9, align 4
> + %11 = add nsw <8 x i32> %10, %7
> + ;CHECK: cost of 1 {{.*}} store
> + store <8 x i32> %11, <8 x i32>* %9, align 4
> + %index.next = add i64 %index, 8
> + %12 = icmp eq i64 %index.next, %end.idx.rnd.down
> + ;CHECK: cost of 1 {{.*}} br
> + br i1 %12, label %middle.block, label %vector.body
> +
> +middle.block: ; preds =
> %vector.body, %for.body.lr.ph
> + %cmp.n = icmp eq i64 %end.idx, %end.idx.rnd.down
> + br i1 %cmp.n, label %for.end, label %for.body
> +
> +for.body: ; preds =
> %middle.block, %for.body
> + %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [
> %end.idx.rnd.down, %middle.block ]
> + %13 = add nsw i64 %indvars.iv, 2
> + %arrayidx = getelementptr inbounds i32* %B, i64 %13
> + ;CHECK: cost of 1 {{.*}} load
> + %14 = load i32* %arrayidx, align 4, !tbaa !0
> + ;CHECK: cost of 1 {{.*}} mul
> + %mul = mul nsw i32 %14, 5
> + %arrayidx2 = getelementptr inbounds i32* %A, i64 %indvars.iv
> + ;CHECK: cost of 1 {{.*}} load
> + %15 = load i32* %arrayidx2, align 4, !tbaa !0
> + %add3 = add nsw i32 %15, %mul
> + store i32 %add3, i32* %arrayidx2, align 4, !tbaa !0
> + %indvars.iv.next = add i64 %indvars.iv, 1
> + ;CHECK: cost of 0 {{.*}} trunc
> + %16 = trunc i64 %indvars.iv.next to i32
> + %cmp = icmp slt i32 %16, %end
> + ;CHECK: cost of 1 {{.*}} br
> + br i1 %cmp, label %for.body, label %for.end
> +
> +for.end: ; preds =
> %middle.block, %for.body, %entry
> + ;CHECK: cost of 1 {{.*}} ret
> + ret i32 undef
> +}
> +
> +!0 = metadata !{metadata !"int", metadata !1}
> +!1 = metadata !{metadata !"omnipotent char", metadata !2}
> +!2 = metadata !{metadata !"Simple C/C++ TBAA"}
>
> Added: llvm/trunk/test/Analysis/CostModel/lit.local.cfg
> URL:
>
http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Analysis/CostModel/lit.l
ocal.cfg?rev=167324&view=auto
>
============================================================================
==
> --- llvm/trunk/test/Analysis/CostModel/lit.local.cfg (added)
> +++ llvm/trunk/test/Analysis/CostModel/lit.local.cfg Fri Nov 2
> 16:48:17 2012
> @@ -0,0 +1 @@
> +config.suffixes = ['.ll', '.c', '.cpp']
>
> Added: llvm/trunk/test/Analysis/CostModel/no_info.ll
> URL:
>
http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Analysis/CostModel/no_in
fo.ll?rev=167324&view=auto
>
============================================================================
==
> --- llvm/trunk/test/Analysis/CostModel/no_info.ll (added)
> +++ llvm/trunk/test/Analysis/CostModel/no_info.ll Fri Nov 2 16:48:17
> 2012
> @@ -0,0 +1,15 @@
> +; RUN: opt < %s -cost-model -analyze | FileCheck %s
> +
> +; The cost model does not have any target information so it can't
> make a decision.
> +; Notice that OPT does not read the triple information from the
> module itself, only through the command line.
> +
> +; This info ignored:
> +target datalayout =
>
"e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:6
4-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
> +target triple = "x86_64-apple-macosx10.8.0"
> +
> +;CHECK: Unknown cost {{.*}} add
> +;CHECK: Unknown cost {{.*}} ret
> +define i32 @no_info(i32 %arg) {
> + %e = add i32 %arg, %arg
> + ret i32 %e
> +}
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>
--
Hal Finkel
Postdoctoral Appointee
Leadership Computing Facility
Argonne National Laboratory
_______________________________________________
llvm-commits mailing list
llvm-commits at cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
More information about the llvm-commits
mailing list