[llvm-commits] [llvm] r146578 - in /llvm/trunk: lib/Transforms/Scalar/LoopUnswitch.cpp test/Transforms/LoopUnswitch/2011-11-18-SimpleSwitch.ll test/Transforms/LoopUnswitch/2011-11-18-TwoSwitches-Threshold.ll test/Transforms/LoopUnswitch/2011-11-18-TwoSwitches.ll
Chad Rosier
mcrosier at apple.com
Thu Dec 22 13:15:57 PST 2011
Stepan,
BTW, I reverted this in r147131 to see if it made a difference in out nightly tester reports. It has now been reinstated in r147175+r147176.
Chad
On Dec 22, 2011, at 12:57 PM, Chad Rosier wrote:
>
> On Dec 22, 2011, at 12:31 PM, Stepan Dyatkovskiy wrote:
>
>> Hi, Chad. I found the problem. The execution time depends on -loop-unswitch-threshold opt param. It is magic param that controls loop-unswitch optimization. Now it is 50 by default. 350 optimal for this test.
>
> I was referring to a compile-time regression, so I'm not sure how increasing the threshold from 50 to 350 would improve compile time? I could certainly see how it would improve execution-time.
>
>> A few words about loop-unswitch optimization. Optimizer looking for loop invariant variables and moves it out of loop. E.g.:
>>
>> for (...) {
>> if (A) // A is not changed in loop.
>> do_A();
>> else
>> do_else();
>> }
>>
>> Optimizer replaces this loop with two loops: loop when A is true and with loop when A is false:
>>
>> if (A) {
>> for (...)
>> do_A();
>> } else {
>> for (...)
>> do_else();
>> }
>>
>> If loop-unswitch-threshold is small, we ignore most of these cases and compile time is small. If this param in "middle" we unswitch something: some branches moved out of loop, but some ones stays in loop and cloned in new loops, and we scan it two or more times. I hope I submit optimized patch soon.
>>
>> The next information helps me a lot. Is it possible to pass opt params through the clang?
>
> I believe you're looking for the -mllvm option. You should be able to do something like 'clang -mllvm -loop-unswitch-threshold 200' or maybe its 'clang -mllvm -loop-unswitch-threshold -mllvm 200'.
>
>> What about all another tests? Was they regressed too?
>
> I see two compile-time regressions:
>
> Performance Regressions - Compile
> MultiSource/Benchmarks/Prolangs-C++/shapes/shapes 7.96%
> MultiSource/Applications/sqlite3/sqlite3 2.21%
>
> My suggestion here would be to leave the threshold at the default 50 and look for opportunities to improve the implementation. Determining the optimal threshold is an entirely separate problem.
>
> I do see a number of compile-time improvements as well (so thanks!):
>
> Performance Improvements - Compile
> SingleSource/Benchmarks/Misc/himenobmtxpa -17.99%
> SingleSource/Benchmarks/Adobe-C++/simple_types_loop_invariant -15.76%
> SingleSource/Benchmarks/Adobe-C++/simple_types_constant_folding -14.44%
> SingleSource/Benchmarks/Adobe-C++/stepanov_vector -8.30%
> SingleSource/Benchmarks/Misc-C++/stepanov_v1p2 -7.77%
> SingleSource/Benchmarks/Adobe-C++/stepanov_abstraction -6.10%
> MultiSource/Benchmarks/ASCI_Purple/SMG2000/smg2000 -2.39%
> External/SPEC/CINT2000/254_gap/254_gap -2.31%
> External/SPEC/CINT95/132_ijpeg/132_ijpeg -1.79%
> MultiSource/Benchmarks/MiBench/consumer-jpeg/consumer-jpeg -1.63%
> External/SPEC/CINT2006/400_perlbench/400_perlbench -1.49%
> MultiSource/Benchmarks/mediabench/jpeg/jpeg-6a/cjpeg -1.48%
>
> My investigation of the execution-time regressions is leading me to believe this patch is *not* the culprit.
>
> Chad
>
>>
>> Thanks.
>> -Stepan.
>>
>> Stepan Dyatkovskiy wrote:
>>> OK. Thanks, Chad. Now I have a time and I look at it.
>>>
>>> -Stepan.
>>>
>>> Chad Rosier wrote:
>>>> Hi Stepan,
>>>> I noticed this patch caused an ~8% compile-time regression for MultiSource/Benchmarks/Prolangs-C++/shapes/shapes for ARMv7 -O3 -mthumb. If you have a moment would you mind taking a look? The test-suite can be downloaded from the llvm repository (See: http://llvm.org/docs/TestingGuide.html#testsuite). Here are the command line arguments I used to reproduce the regression:
>>>>
>>>> /Users/mcrosier/llvm-clean/Release+Asserts/Release+Asserts/bin/clang++ -I/Users/mcrosier/llvm-clean/Release+Asserts/projects/test-suite/MultiSource/Benchmarks/Prolangs-C++/shapes -I/Users/mcrosier/llvm-clean/llvm/projects/test-suite/MultiSource/Benchmarks/Prolangs-C++/shapes -I/Users/mcrosier/llvm-clean/Release+Asserts/projects/test-suite/../../../llvm/projects/test-suite/include -I../../../../include -I/Users/mcrosier/llvm-clean/Release+Asserts/include -I/Users/mcrosier/llvm-clean/llvm/include -D_GNU_SOURCE -D__STDC_LIMIT_MACROS -DNDEBUG -DSMALL_PROBLEM_SIZE -O3 -mthumb -miphoneos-version-min=4.0 -Wl,--no-demangle -ccc-install-dir /Developer/Platforms/iPhoneOS.platform/Developer/usr/bin -arch armv7 -isysroot /Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS6.0.Internal.sdk -c /Users/mcrosier/llvm-clean/llvm/projects/test-suite/MultiSource/Benchmarks/Prolangs-C++/shapes/bjarne.cpp -o Output/bjarne.llvm.o
>>>>
>>>> Chad
>>>>
>>>> For Apple's record this is being tracked by<rdar://problem/10601498>.
>>>>
>>>> On Dec 14, 2011, at 11:19 AM, Stepan Dyatkovskiy wrote:
>>>>
>>>>> Author: dyatkovskiy
>>>>> Date: Wed Dec 14 13:19:17 2011
>>>>> New Revision: 146578
>>>>>
>>>>> URL: http://llvm.org/viewvc/llvm-project?rev=146578&view=rev
>>>>> Log:
>>>>> Fix for bug #11429: Wrong behaviour for switches. Small improvement for code size heuristics.
>>>>>
>>>>> Added:
>>>>> llvm/trunk/test/Transforms/LoopUnswitch/2011-11-18-SimpleSwitch.ll
>>>>> llvm/trunk/test/Transforms/LoopUnswitch/2011-11-18-TwoSwitches-Threshold.ll
>>>>> llvm/trunk/test/Transforms/LoopUnswitch/2011-11-18-TwoSwitches.ll
>>>>> Modified:
>>>>> llvm/trunk/lib/Transforms/Scalar/LoopUnswitch.cpp
>>>>>
>>>>> Modified: llvm/trunk/lib/Transforms/Scalar/LoopUnswitch.cpp
>>>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/LoopUnswitch.cpp?rev=146578&r1=146577&r2=146578&view=diff
>>>>> ==============================================================================
>>>>> --- llvm/trunk/lib/Transforms/Scalar/LoopUnswitch.cpp (original)
>>>>> +++ llvm/trunk/lib/Transforms/Scalar/LoopUnswitch.cpp Wed Dec 14 13:19:17 2011
>>>>> @@ -71,7 +71,9 @@
>>>>> // LoopProcessWorklist - Used to check if second loop needs processing
>>>>> // after RewriteLoopBodyWithConditionConstant rewrites first loop.
>>>>> std::vector<Loop*> LoopProcessWorklist;
>>>>> - SmallPtrSet<Value *,8> UnswitchedVals;
>>>>> +
>>>>> + // FIXME: Consider custom class for this.
>>>>> + std::map<const SwitchInst*, SmallPtrSet<const Value *,8> > UnswitchedVals;
>>>>>
>>>>> bool OptimizeForSize;
>>>>> bool redoLoop;
>>>>> @@ -117,7 +119,15 @@
>>>>> private:
>>>>>
>>>>> virtual void releaseMemory() {
>>>>> - UnswitchedVals.clear();
>>>>> + // We need to forget about all switches in the current loop.
>>>>> + // FIXME: Do it better than enumerating all blocks of code
>>>>> + // and see if it is a switch instruction.
>>>>> + for (Loop::block_iterator I = currentLoop->block_begin(),
>>>>> + E = currentLoop->block_end(); I != E; ++I) {
>>>>> + SwitchInst* SI = dyn_cast<SwitchInst>((*I)->getTerminator());
>>>>> + if (SI)
>>>>> + UnswitchedVals.erase(SI);
>>>>> + }
>>>>> }
>>>>>
>>>>> /// RemoveLoopFromWorklist - If the specified loop is on the loop worklist,
>>>>> @@ -128,6 +138,12 @@
>>>>> if (I != LoopProcessWorklist.end())
>>>>> LoopProcessWorklist.erase(I);
>>>>> }
>>>>> +
>>>>> + /// For new loop switches we clone info about values that was
>>>>> + /// already unswitched and has redundant successors.
>>>>> + /// Note, that new loop data is stored inside the VMap.
>>>>> + void CloneUnswitchedVals(const ValueToValueMapTy& VMap,
>>>>> + const BasicBlock* SrcBB);
>>>>>
>>>>> void initLoopData() {
>>>>> loopHeader = currentLoop->getHeader();
>>>>> @@ -255,13 +271,25 @@
>>>>> } else if (SwitchInst *SI = dyn_cast<SwitchInst>(TI)) {
>>>>> Value *LoopCond = FindLIVLoopCondition(SI->getCondition(),
>>>>> currentLoop, Changed);
>>>>> - if (LoopCond&& SI->getNumCases()> 1) {
>>>>> + unsigned NumCases = SI->getNumCases();
>>>>> + if (LoopCond&& NumCases> 1) {
>>>>> // Find a value to unswitch on:
>>>>> // FIXME: this should chose the most expensive case!
>>>>> // FIXME: scan for a case with a non-critical edge?
>>>>> - Constant *UnswitchVal = SI->getCaseValue(1);
>>>>> + Constant *UnswitchVal = NULL;
>>>>> +
>>>>> // Do not process same value again and again.
>>>>> - if (!UnswitchedVals.insert(UnswitchVal))
>>>>> + // At this point we have some cases already unswitched and
>>>>> + // some not yet unswitched. Let's find the first not yet unswitched one.
>>>>> + for (unsigned i = 1; i< NumCases; ++i) {
>>>>> + Constant* UnswitchValCandidate = SI->getCaseValue(i);
>>>>> + if (!UnswitchedVals[SI].count(UnswitchValCandidate)) {
>>>>> + UnswitchVal = UnswitchValCandidate;
>>>>> + break;
>>>>> + }
>>>>> + }
>>>>> +
>>>>> + if (!UnswitchVal)
>>>>> continue;
>>>>>
>>>>> if (UnswitchIfProfitable(LoopCond, UnswitchVal)) {
>>>>> @@ -287,6 +315,23 @@
>>>>> return Changed;
>>>>> }
>>>>>
>>>>> +/// For new loop switches we clone info about values that was
>>>>> +/// already unswitched and has redundant successors.
>>>>> +/// Not that new loop data is stored inside the VMap.
>>>>> +void LoopUnswitch::CloneUnswitchedVals(const ValueToValueMapTy& VMap,
>>>>> + const BasicBlock* SrcBB) {
>>>>> +
>>>>> + const SwitchInst* SI = dyn_cast<SwitchInst>(SrcBB->getTerminator());
>>>>> + if (SI&& UnswitchedVals.count(SI)) {
>>>>> + // Don't clone a totally simplified switch.
>>>>> + if (isa<Constant>(SI->getCondition()))
>>>>> + return;
>>>>> + Value* I = VMap.lookup(SI);
>>>>> + assert(I&& "All instructions that are in SrcBB must be in VMap.");
>>>>> + UnswitchedVals[cast<SwitchInst>(I)] = UnswitchedVals[SI];
>>>>> + }
>>>>> +}
>>>>> +
>>>>> /// isTrivialLoopExitBlock - Check to see if all paths from BB exit the
>>>>> /// loop with no side effects (including infinite loops).
>>>>> ///
>>>>> @@ -378,14 +423,25 @@
>>>>> // Check to see if a successor of the switch is guaranteed to go to the
>>>>> // latch block or exit through a one exit block without having any
>>>>> // side-effects. If so, determine the value of Cond that causes it to do
>>>>> - // this. Note that we can't trivially unswitch on the default case.
>>>>> - for (unsigned i = 1, e = SI->getNumSuccessors(); i != e; ++i)
>>>>> - if ((LoopExitBB = isTrivialLoopExitBlock(currentLoop,
>>>>> + // this.
>>>>> + // Note that we can't trivially unswitch on the default case or
>>>>> + // on already unswitched cases.
>>>>> + for (unsigned i = 1, e = SI->getNumSuccessors(); i != e; ++i) {
>>>>> + BasicBlock* LoopExitCandidate;
>>>>> + if ((LoopExitCandidate = isTrivialLoopExitBlock(currentLoop,
>>>>> SI->getSuccessor(i)))) {
>>>>> // Okay, we found a trivial case, remember the value that is trivial.
>>>>> - if (Val) *Val = SI->getCaseValue(i);
>>>>> + ConstantInt* CaseVal = SI->getCaseValue(i);
>>>>> +
>>>>> + // Check that it was not unswitched before, since already unswitched
>>>>> + // trivial vals are looks trivial too.
>>>>> + if (UnswitchedVals[SI].count(CaseVal))
>>>>> + continue;
>>>>> + LoopExitBB = LoopExitCandidate;
>>>>> + if (Val) *Val = CaseVal;
>>>>> break;
>>>>> }
>>>>> + }
>>>>> }
>>>>>
>>>>> // If we didn't find a single unique LoopExit block, or if the loop exit block
>>>>> @@ -447,8 +503,14 @@
>>>>> // expansion, and the number of basic blocks, to avoid loops with
>>>>> // large numbers of branches which cause loop unswitching to go crazy.
>>>>> // This is a very ad-hoc heuristic.
>>>>> - if (Metrics.NumInsts> Threshold ||
>>>>> - Metrics.NumBlocks * 5> Threshold ||
>>>>> +
>>>>> + unsigned NumUnswitched =
>>>>> + (NumSwitches + NumBranches) + 1 /*take in account current iteration*/;
>>>>> +
>>>>> + unsigned NumInsts = Metrics.NumInsts * NumUnswitched;
>>>>> + unsigned NumBlocks = Metrics.NumBlocks * NumUnswitched;
>>>>> +
>>>>> + if (NumInsts> Threshold || NumBlocks * 5> Threshold ||
>>>>> Metrics.containsIndirectBr || Metrics.isRecursive) {
>>>>> DEBUG(dbgs()<< "NOT unswitching loop %"
>>>>> << currentLoop->getHeader()->getName()<< ", cost too high:"
>>>>> @@ -620,6 +682,12 @@
>>>>> ValueToValueMapTy VMap;
>>>>> for (unsigned i = 0, e = LoopBlocks.size(); i != e; ++i) {
>>>>> BasicBlock *NewBB = CloneBasicBlock(LoopBlocks[i], VMap, ".us", F);
>>>>> +
>>>>> + // Inherit simplified switches info for NewBB
>>>>> + // We needn't pass NewBB since its instructions are already contained
>>>>> + // inside the VMap.
>>>>> + CloneUnswitchedVals(VMap, LoopBlocks[i]);
>>>>> +
>>>>> NewBlocks.push_back(NewBB);
>>>>> VMap[LoopBlocks[i]] = NewBB; // Keep the BB mapping.
>>>>> LPM->cloneBasicBlockSimpleAnalysis(LoopBlocks[i], NewBB, L);
>>>>> @@ -945,6 +1013,9 @@
>>>>> BasicBlock *Switch = SI->getParent();
>>>>> BasicBlock *SISucc = SI->getSuccessor(DeadCase);
>>>>> BasicBlock *Latch = L->getLoopLatch();
>>>>> +
>>>>> + UnswitchedVals[SI].insert(Val);
>>>>> +
>>>>> if (!SI->findCaseDest(SISucc)) continue; // Edge is critical.
>>>>> // If the DeadCase successor dominates the loop latch, then the
>>>>> // transformation isn't safe since it will delete the sole predecessor edge
>>>>>
>>>>> Added: llvm/trunk/test/Transforms/LoopUnswitch/2011-11-18-SimpleSwitch.ll
>>>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopUnswitch/2011-11-18-SimpleSwitch.ll?rev=146578&view=auto
>>>>> ==============================================================================
>>>>> --- llvm/trunk/test/Transforms/LoopUnswitch/2011-11-18-SimpleSwitch.ll (added)
>>>>> +++ llvm/trunk/test/Transforms/LoopUnswitch/2011-11-18-SimpleSwitch.ll Wed Dec 14 13:19:17 2011
>>>>> @@ -0,0 +1,91 @@
>>>>> +; RUN: opt -loop-unswitch -disable-output -stats -info-output-file -< %s | FileCheck --check-prefix=STATS %s
>>>>> +; RUN: opt -S -loop-unswitch -verify-loop-info -verify-dom-info %s | FileCheck %s
>>>>> +
>>>>> +; STATS: 1 loop-simplify - Number of pre-header or exit blocks inserted
>>>>> +; STATS: 2 loop-unswitch - Number of switches unswitched
>>>>> +
>>>>> +; CHECK: %1 = icmp eq i32 %c, 1
>>>>> +; CHECK-NEXT: br i1 %1, label %.split.us, label %..split_crit_edge
>>>>> +
>>>>> +; CHECK: ..split_crit_edge: ; preds = %0
>>>>> +; CHECK-NEXT: br label %.split
>>>>> +
>>>>> +; CHECK: .split.us: ; preds = %0
>>>>> +; CHECK-NEXT: br label %loop_begin.us
>>>>> +
>>>>> +; CHECK: loop_begin.us: ; preds = %loop_begin.backedge.us, %.split.us
>>>>> +; CHECK-NEXT: %var_val.us = load i32* %var
>>>>> +; CHECK-NEXT: switch i32 1, label %default.us-lcssa.us [
>>>>> +; CHECK-NEXT: i32 1, label %inc.us
>>>>> +
>>>>> +; CHECK: inc.us: ; preds = %loop_begin.us
>>>>> +; CHECK-NEXT: call void @incf() noreturn nounwind
>>>>> +; CHECK-NEXT: br label %loop_begin.backedge.us
>>>>> +
>>>>> +; CHECK: .split: ; preds = %..split_crit_edge
>>>>> +; CHECK-NEXT: %2 = icmp eq i32 %c, 2
>>>>> +; CHECK-NEXT: br i1 %2, label %.split.split.us, label %.split..split.split_crit_edge
>>>>> +
>>>>> +; CHECK: .split..split.split_crit_edge: ; preds = %.split
>>>>> +; CHECK-NEXT: br label %.split.split
>>>>> +
>>>>> +; CHECK: .split.split.us: ; preds = %.split
>>>>> +; CHECK-NEXT: br label %loop_begin.us1
>>>>> +
>>>>> +; CHECK: loop_begin.us1: ; preds = %loop_begin.backedge.us5, %.split.split.us
>>>>> +; CHECK-NEXT: %var_val.us2 = load i32* %var
>>>>> +; CHECK-NEXT: switch i32 2, label %default.us-lcssa.us-lcssa.us [
>>>>> +; CHECK-NEXT: i32 1, label %inc.us3
>>>>> +; CHECK-NEXT: i32 2, label %dec.us4
>>>>> +; CHECK-NEXT: ]
>>>>> +
>>>>> +; CHECK: dec.us4: ; preds = %loop_begin.us1
>>>>> +; CHECK-NEXT: call void @decf() noreturn nounwind
>>>>> +; CHECK-NEXT: br label %loop_begin.backedge.us5
>>>>> +
>>>>> +; CHECK: .split.split: ; preds = %.split..split.split_crit_edge
>>>>> +; CHECK-NEXT: br label %loop_begin
>>>>> +
>>>>> +; CHECK: loop_begin: ; preds = %loop_begin.backedge, %.split.split
>>>>> +; CHECK-NEXT: %var_val = load i32* %var
>>>>> +; CHECK-NEXT: switch i32 %c, label %default.us-lcssa.us-lcssa [
>>>>> +; CHECK-NEXT: i32 1, label %inc
>>>>> +; CHECK-NEXT: i32 2, label %dec
>>>>> +; CHECK-NEXT: ]
>>>>> +
>>>>> +; CHECK: inc: ; preds = %loop_begin
>>>>> +; CHECK-NEXT: br i1 true, label %us-unreachable.us-lcssa, label %inc.split
>>>>> +
>>>>> +; CHECK: dec: ; preds = %loop_begin
>>>>> +; CHECK-NEXT: br i1 true, label %us-unreachable6, label %dec.split
>>>>> +
>>>>> +define i32 @test(i32* %var) {
>>>>> + %mem = alloca i32
>>>>> + store i32 2, i32* %mem
>>>>> + %c = load i32* %mem
>>>>> +
>>>>> + br label %loop_begin
>>>>> +
>>>>> +loop_begin:
>>>>> +
>>>>> + %var_val = load i32* %var
>>>>> +
>>>>> + switch i32 %c, label %default [
>>>>> + i32 1, label %inc
>>>>> + i32 2, label %dec
>>>>> + ]
>>>>> +
>>>>> +inc:
>>>>> + call void @incf() noreturn nounwind
>>>>> + br label %loop_begin
>>>>> +dec:
>>>>> + call void @decf() noreturn nounwind
>>>>> + br label %loop_begin
>>>>> +default:
>>>>> + br label %loop_exit
>>>>> +loop_exit:
>>>>> + ret i32 0
>>>>> +}
>>>>> +
>>>>> +declare void @incf() noreturn
>>>>> +declare void @decf() noreturn
>>>>>
>>>>> Added: llvm/trunk/test/Transforms/LoopUnswitch/2011-11-18-TwoSwitches-Threshold.ll
>>>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopUnswitch/2011-11-18-TwoSwitches-Threshold.ll?rev=146578&view=auto
>>>>> ==============================================================================
>>>>> --- llvm/trunk/test/Transforms/LoopUnswitch/2011-11-18-TwoSwitches-Threshold.ll (added)
>>>>> +++ llvm/trunk/test/Transforms/LoopUnswitch/2011-11-18-TwoSwitches-Threshold.ll Wed Dec 14 13:19:17 2011
>>>>> @@ -0,0 +1,84 @@
>>>>> +; RUN: opt -loop-unswitch -loop-unswitch-threshold 30 -disable-output -stats -info-output-file -< %s | FileCheck --check-prefix=STATS %s
>>>>> +; RUN: opt -S -loop-unswitch -loop-unswitch-threshold 30 -verify-loop-info -verify-dom-info %s | FileCheck %s
>>>>> +
>>>>> +; STATS: 1 loop-simplify - Number of pre-header or exit blocks inserted
>>>>> +; STATS: 1 loop-unswitch - Number of switches unswitched
>>>>> +
>>>>> +; ModuleID = '../llvm/test/Transforms/LoopUnswitch/2011-11-18-TwoSwitches.ll'
>>>>> +
>>>>> +; CHECK: %1 = icmp eq i32 %c, 1
>>>>> +; CHECK-NEXT: br i1 %1, label %.split.us, label %..split_crit_edge
>>>>> +
>>>>> +; CHECK: ..split_crit_edge: ; preds = %0
>>>>> +; CHECK-NEXT: br label %.split
>>>>> +
>>>>> +; CHECK: .split.us: ; preds = %0
>>>>> +; CHECK-NEXT: br label %loop_begin.us
>>>>> +
>>>>> +; CHECK: loop_begin.us: ; preds = %loop_begin.backedge.us, %.split.us
>>>>> +; CHECK: switch i32 1, label %second_switch.us [
>>>>> +; CHECK-NEXT: i32 1, label %inc.us
>>>>> +
>>>>> +; CHECK: inc.us: ; preds = %second_switch.us, %loop_begin.us
>>>>> +; CHECK-NEXT: call void @incf() noreturn nounwind
>>>>> +; CHECK-NEXT: br label %loop_begin.backedge.us
>>>>> +
>>>>> +; CHECK: second_switch.us: ; preds = %loop_begin.us
>>>>> +; CHECK-NEXT: switch i32 %d, label %default.us [
>>>>> +; CHECK-NEXT: i32 1, label %inc.us
>>>>> +; CHECK-NEXT: ]
>>>>> +
>>>>> +; CHECK: .split: ; preds = %..split_crit_edge
>>>>> +; CHECK-NEXT: br label %loop_begin
>>>>> +
>>>>> +; CHECK: loop_begin: ; preds = %loop_begin.backedge, %.split
>>>>> +; CHECK: switch i32 %c, label %second_switch [
>>>>> +; CHECK-NEXT: i32 1, label %loop_begin.inc_crit_edge
>>>>> +; CHECK-NEXT: ]
>>>>> +
>>>>> +; CHECK: loop_begin.inc_crit_edge: ; preds = %loop_begin
>>>>> +; CHECK-NEXT: br i1 true, label %us-unreachable, label %inc
>>>>> +
>>>>> +; CHECK: second_switch: ; preds = %loop_begin
>>>>> +; CHECK-NEXT: switch i32 %d, label %default [
>>>>> +; CHECK-NEXT: i32 1, label %inc
>>>>> +; CHECK-NEXT: ]
>>>>> +
>>>>> +; CHECK: inc: ; preds = %loop_begin.inc_crit_edge, %second_switch
>>>>> +; CHECK-NEXT: call void @incf() noreturn nounwind
>>>>> +; CHECK-NEXT: br label %loop_begin.backedge
>>>>> +
>>>>> +define i32 @test(i32* %var) {
>>>>> + %mem = alloca i32
>>>>> + store i32 2, i32* %mem
>>>>> + %c = load i32* %mem
>>>>> + %d = load i32* %mem
>>>>> +
>>>>> + br label %loop_begin
>>>>> +
>>>>> +loop_begin:
>>>>> +
>>>>> + %var_val = load i32* %var
>>>>> +
>>>>> + switch i32 %c, label %second_switch [
>>>>> + i32 1, label %inc
>>>>> + ]
>>>>> +
>>>>> +second_switch:
>>>>> + switch i32 %d, label %default [
>>>>> + i32 1, label %inc
>>>>> + ]
>>>>> +
>>>>> +inc:
>>>>> + call void @incf() noreturn nounwind
>>>>> + br label %loop_begin
>>>>> +
>>>>> +default:
>>>>> + br label %loop_begin
>>>>> +
>>>>> +loop_exit:
>>>>> + ret i32 0
>>>>> +}
>>>>> +
>>>>> +declare void @incf() noreturn
>>>>> +declare void @decf() noreturn
>>>>>
>>>>> Added: llvm/trunk/test/Transforms/LoopUnswitch/2011-11-18-TwoSwitches.ll
>>>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopUnswitch/2011-11-18-TwoSwitches.ll?rev=146578&view=auto
>>>>> ==============================================================================
>>>>> --- llvm/trunk/test/Transforms/LoopUnswitch/2011-11-18-TwoSwitches.ll (added)
>>>>> +++ llvm/trunk/test/Transforms/LoopUnswitch/2011-11-18-TwoSwitches.ll Wed Dec 14 13:19:17 2011
>>>>> @@ -0,0 +1,138 @@
>>>>> +; RUN: opt -loop-unswitch -loop-unswitch-threshold 1000 -disable-output -stats -info-output-file -< %s | FileCheck --check-prefix=STATS %s
>>>>> +; RUN: opt -S -loop-unswitch -loop-unswitch-threshold 1000 -verify-loop-info -verify-dom-info %s | FileCheck %s
>>>>> +
>>>>> +; STATS: 1 loop-simplify - Number of pre-header or exit blocks inserted
>>>>> +; STATS: 3 loop-unswitch - Number of switches unswitched
>>>>> +
>>>>> +; CHECK: %1 = icmp eq i32 %c, 1
>>>>> +; CHECK-NEXT: br i1 %1, label %.split.us, label %..split_crit_edge
>>>>> +
>>>>> +; CHECK: ..split_crit_edge: ; preds = %0
>>>>> +; CHECK-NEXT: br label %.split
>>>>> +
>>>>> +; CHECK: .split.us: ; preds = %0
>>>>> +; CHECK-NEXT: %2 = icmp eq i32 %d, 1
>>>>> +; CHECK-NEXT: br i1 %2, label %.split.us.split.us, label %.split.us..split.us.split_crit_edge
>>>>> +
>>>>> +; CHECK: .split.us..split.us.split_crit_edge: ; preds = %.split.us
>>>>> +; CHECK-NEXT: br label %.split.us.split
>>>>> +
>>>>> +; CHECK: .split.us.split.us: ; preds = %.split.us
>>>>> +; CHECK-NEXT: br label %loop_begin.us.us
>>>>> +
>>>>> +; CHECK: loop_begin.us.us: ; preds = %loop_begin.backedge.us.us, %.split.us.split.us
>>>>> +; CHECK-NEXT: %var_val.us.us = load i32* %var
>>>>> +; CHECK-NEXT: switch i32 1, label %second_switch.us.us [
>>>>> +; CHECK-NEXT: i32 1, label %inc.us.us
>>>>> +
>>>>> +; CHECK: inc.us.us: ; preds = %second_switch.us.us, %loop_begin.us.us
>>>>> +; CHECK-NEXT: call void @incf() noreturn nounwind
>>>>> +; CHECK-NEXT: br label %loop_begin.backedge.us.us
>>>>> +
>>>>> +; CHECK: second_switch.us.us: ; preds = %loop_begin.us.us
>>>>> +; CHECK-NEXT: switch i32 1, label %default.us.us [
>>>>> +; CHECK-NEXT: i32 1, label %inc.us.us
>>>>> +
>>>>> +; CHECK: .split.us.split: ; preds = %.split.us..split.us.split_crit_edge
>>>>> +; CHECK-NEXT: br label %loop_begin.us
>>>>> +
>>>>> +; CHECK: loop_begin.us: ; preds = %loop_begin.backedge.us, %.split.us.split
>>>>> +; CHECK-NEXT: %var_val.us = load i32* %var
>>>>> +; CHECK-NEXT: switch i32 1, label %second_switch.us [
>>>>> +; CHECK-NEXT: i32 1, label %inc.us
>>>>> +
>>>>> +; CHECK: inc.us: ; preds = %second_switch.us.inc.us_crit_edge, %loop_begin.us
>>>>> +; CHECK-NEXT: call void @incf() noreturn nounwind
>>>>> +; CHECK-NEXT: br label %loop_begin.backedge.us
>>>>> +
>>>>> +; CHECK: second_switch.us: ; preds = %loop_begin.us
>>>>> +; CHECK-NEXT: switch i32 %d, label %default.us [
>>>>> +; CHECK-NEXT: i32 1, label %second_switch.us.inc.us_crit_edge
>>>>> +; CHECK-NEXT: ]
>>>>> +
>>>>> +; CHECK: second_switch.us.inc.us_crit_edge: ; preds = %second_switch.us
>>>>> +; CHECK-NEXT: br i1 true, label %us-unreachable8, label %inc.us
>>>>> +
>>>>> +; CHECK: .split: ; preds = %..split_crit_edge
>>>>> +; CHECK-NEXT: %3 = icmp eq i32 %d, 1
>>>>> +; CHECK-NEXT: br i1 %3, label %.split.split.us, label %.split..split.split_crit_edge
>>>>> +
>>>>> +; CHECK: .split..split.split_crit_edge: ; preds = %.split
>>>>> +; CHECK-NEXT: br label %.split.split
>>>>> +
>>>>> +; CHECK: .split.split.us: ; preds = %.split
>>>>> +; CHECK-NEXT: br label %loop_begin.us1
>>>>> +
>>>>> +; CHECK: loop_begin.us1: ; preds = %loop_begin.backedge.us6, %.split.split.us
>>>>> +; CHECK-NEXT: %var_val.us2 = load i32* %var
>>>>> +; CHECK-NEXT: switch i32 %c, label %second_switch.us4 [
>>>>> +; CHECK-NEXT: i32 1, label %loop_begin.inc_crit_edge.us
>>>>> +; CHECK-NEXT: ]
>>>>> +
>>>>> +; CHECK: inc.us3: ; preds = %loop_begin.inc_crit_edge.us, %second_switch.us4
>>>>> +; CHECK-NEXT: call void @incf() noreturn nounwind
>>>>> +; CHECK-NEXT: br label %loop_begin.backedge.us6
>>>>> +
>>>>> +; CHECK: second_switch.us4: ; preds = %loop_begin.us1
>>>>> +; CHECK-NEXT: switch i32 1, label %default.us5 [
>>>>> +; CHECK-NEXT: i32 1, label %inc.us3
>>>>> +; CHECK-NEXT: ]
>>>>> +
>>>>> +; CHECK: loop_begin.inc_crit_edge.us: ; preds = %loop_begin.us1
>>>>> +; CHECK-NEXT: br i1 true, label %us-unreachable.us-lcssa.us, label %inc.us3
>>>>> +
>>>>> +; CHECK: .split.split: ; preds = %.split..split.split_crit_edge
>>>>> +; CHECK-NEXT: br label %loop_begin
>>>>> +
>>>>> +; CHECK: loop_begin: ; preds = %loop_begin.backedge, %.split.split
>>>>> +; CHECK-NEXT: %var_val = load i32* %var
>>>>> +; CHECK-NEXT: switch i32 %c, label %second_switch [
>>>>> +; CHECK-NEXT: i32 1, label %loop_begin.inc_crit_edge
>>>>> +; CHECK-NEXT: ]
>>>>> +
>>>>> +; CHECK: loop_begin.inc_crit_edge: ; preds = %loop_begin
>>>>> +; CHECK-NEXT: br i1 true, label %us-unreachable.us-lcssa, label %inc
>>>>> +
>>>>> +; CHECK: second_switch: ; preds = %loop_begin
>>>>> +; CHECK-NEXT: switch i32 %d, label %default [
>>>>> +; CHECK-NEXT: i32 1, label %second_switch.inc_crit_edge
>>>>> +; CHECK-NEXT: ]
>>>>> +
>>>>> +; CHECK: second_switch.inc_crit_edge: ; preds = %second_switch
>>>>> +; CHECK-NEXT: br i1 true, label %us-unreachable7, label %inc
>>>>> +
>>>>> +
>>>>> +define i32 @test(i32* %var) {
>>>>> + %mem = alloca i32
>>>>> + store i32 2, i32* %mem
>>>>> + %c = load i32* %mem
>>>>> + %d = load i32* %mem
>>>>> +
>>>>> + br label %loop_begin
>>>>> +
>>>>> +loop_begin:
>>>>> +
>>>>> + %var_val = load i32* %var
>>>>> +
>>>>> + switch i32 %c, label %second_switch [
>>>>> + i32 1, label %inc
>>>>> + ]
>>>>> +
>>>>> +second_switch:
>>>>> + switch i32 %d, label %default [
>>>>> + i32 1, label %inc
>>>>> + ]
>>>>> +
>>>>> +inc:
>>>>> + call void @incf() noreturn nounwind
>>>>> + br label %loop_begin
>>>>> +
>>>>> +default:
>>>>> + br label %loop_begin
>>>>> +
>>>>> +loop_exit:
>>>>> + ret i32 0
>>>>> +}
>>>>> +
>>>>> +declare void @incf() noreturn
>>>>> +declare void @decf() noreturn
>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> llvm-commits mailing list
>>>>> llvm-commits at cs.uiuc.edu
>>>>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>>>>
>>>
>>> _______________________________________________
>>> llvm-commits mailing list
>>> llvm-commits at cs.uiuc.edu
>>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>>
>
> _______________________________________________
> 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