[llvm] r235079 - TRUNCATE constant folding - minor fix for rL233224
Simon Pilgrim
llvm-dev at redking.me.uk
Sat Apr 25 12:00:33 PDT 2015
Hi Patrik,
Sorry for the delay. I have been using your script, its finding a number of other faults and its taking a while to determine what is responsible. I was a little over enthusiastic with testing other targets as well which slowed things down a great deal as well.
I have a better idea of what is going on now and will create a patch for review shortly.
Simon.
> On 24 Apr 2015, at 16:41, Patrik Hägglund H <patrik.h.hagglund at ericsson.com> wrote:
>
> Hi Simon,
>
> I still see new regression issues pop up, caused by this patch (visible at r235073, and bisected back to this revision). Can you try to use the script I provided, and try for yourself?
>
> (For example: bin/llvm-stress -size 300 -seed 15896 | bin/llc -march=x86-64 -mcpu=corei7 -o /dev/null)
>
> /Patrik Hägglund
>
> -----Original Message-----
> From: llvm-commits-bounces at cs.uiuc.edu [mailto:llvm-commits-bounces at cs.uiuc.edu] On Behalf Of Patrik Hägglund H
> Sent: den 21 april 2015 08:28
> To: Simon Pilgrim; llvm-commits at cs.uiuc.edu
> Subject: RE: [llvm] r235079 - TRUNCATE constant folding - minor fix for rL233224
>
> Thanks. You can find my simple shell script as an attachment (and the usage on line 4). We use it as a quick regression test, before we merge from master into our own development branch.
>
> The size argument to llvm-stress is 300, which seems like a good tradeoff. We currently only test corei7. If you try any other architecture, you will probably find failures quite easily. Also, -verify-machineinstrs and -verify-regalloc are disabled, due to known failures (on corei7).
>
> /Patrik Hägglund
>
> -----Original Message-----
> From: Simon Pilgrim [mailto:llvm-dev at redking.me.uk]
> Sent: den 20 april 2015 23:59
> To: Patrik Hägglund H; llvm-commits at cs.uiuc.edu
> Subject: Re: [llvm] r235079 - TRUNCATE constant folding - minor fix for rL233224
>
> Thanks again Patrik, I've found the next bug (SelectionDag.cpp CONCAT_VECTORS of BUILD_VECTORS without checking that the scalar operands are all of the same type) - I'll get a fix + the test case committed tomorrow morning (UK time).
>
> I'm trying to think of ways to efficiently locate more of these issues - are you using llvm-stress in any particular fashion that is finding these quickly or are you just bulk testing with various seeds and sizes values and cpu targets?
>
> Simon.
>
>> On 20 Apr 2015, at 11:59, Patrik Hägglund H <patrik.h.hagglund at ericsson.com> wrote:
>>
>> Thanks for the fix. However, I now see a second failure (found by llvm-stress) caused by this commit:
>>
>>> bin/llvm-stress -seed 13800 -size 300 | bin/llc -march=x86-64
>>> -mcpu=corei7 -o /dev/null
>>
>> llc: ../lib/CodeGen/SelectionDAG/SelectionDAG.cpp:762: void VerifySDNode(llvm::SDNode*): Assertion `I->getValueType() == N->getOperand(0).getValueType() && "Operands must all have the same type"' failed.
>> 0 llc 0x0000000001f11831 llvm::sys::PrintStackTrace(llvm::raw_ostream&) + 44
>> 1 llc 0x0000000001f11bad
>> 2 llc 0x0000000001f10785
>> 3 libpthread.so.0 0x00007f1bb1b69810
>> 4 libc.so.6 0x00007f1bb1384755 gsignal + 53
>> 5 libc.so.6 0x00007f1bb1385d31 abort + 385
>> 6 libc.so.6 0x00007f1bb137d610 __assert_fail + 240
>> 7 llc 0x0000000001d8b150
>> 8 llc 0x0000000001d8b1ae llvm::SelectionDAG::InsertNode(llvm::SDNode*) + 54
>> 9 llc 0x0000000001da65dd llvm::SelectionDAG::getNode(unsigned int, llvm::SDLoc, llvm::EVT, llvm::ArrayRef<llvm::SDValue>) + 1947
>> 10 llc 0x0000000001d9a930 llvm::SelectionDAG::getNode(unsigned int, llvm::SDLoc, llvm::EVT, llvm::SDValue, llvm::SDValue, bool, bool, bool) + 982
>> 11 llc 0x0000000001e60042
>> 12 llc 0x0000000001e61a1e
>> 13 llc 0x0000000001e5f32f
>> 14 llc 0x0000000001e4710d
>> 15 llc 0x0000000001e4b695 llvm::SelectionDAG::LegalizeTypes() + 59
>> 16 llc 0x0000000001e09802 llvm::SelectionDAGISel::CodeGenAndEmitDAG() + 1360
>> 17 llc 0x0000000001e08f86 llvm::SelectionDAGISel::SelectBasicBlock(llvm::ilist_iterator<llvm::Instruction const>, llvm::ilist_iterator<llvm::Instruction const>, bool&) + 234
>> 18 llc 0x0000000001e0c615 llvm::SelectionDAGISel::SelectAllBasicBlocks(llvm::Function const&) + 2939
>> 19 llc 0x0000000001e08124 llvm::SelectionDAGISel::runOnMachineFunction(llvm::MachineFunction&) + 998
>> 20 llc 0x000000000141b7d3
>> 21 llc 0x000000000180fd4b llvm::MachineFunctionPass::runOnFunction(llvm::Function&) + 95
>> 22 llc 0x0000000001b457d2 llvm::FPPassManager::runOnFunction(llvm::Function&) + 290
>> 23 llc 0x0000000001b45942 llvm::FPPassManager::runOnModule(llvm::Module&) + 84
>> 24 llc 0x0000000001b45c96
>> 25 llc 0x0000000001b46332 llvm::legacy::PassManagerImpl::run(llvm::Module&) + 250
>> 26 llc 0x0000000001b46553 llvm::legacy::PassManager::run(llvm::Module&) + 39
>> 27 llc 0x0000000000c2281d
>> 28 llc 0x0000000000c21903 main + 242
>> 29 libc.so.6 0x00007f1bb1370c16 __libc_start_main + 230
>> 30 llc 0x0000000000c20e29
>> Stack dump:
>> 0. Program arguments: bin/llc -march=x86-64 -mcpu=corei7 -o /dev/null
>> 1. Running pass 'Function Pass Manager' on module '<stdin>'.
>> 2. Running pass 'X86 DAG->DAG Instruction Selection' on function '@autogen_SD13800'
>> Abort (core dumped)
>>
>> /Patrik Hägglund
>>
>> -----Original Message-----
>> From: Simon Pilgrim [mailto:llvm-dev at redking.me.uk]
>> Sent: den 19 april 2015 23:08
>> To: Patrik Hägglund H
>> Cc: llvm-commits at cs.uiuc.edu
>> Subject: Re: [llvm] r235079 - TRUNCATE constant folding - minor fix
>> for rL233224
>>
>> Patrik, thanks for the test case. I've identified the issue - the
>> commit is exposing more code that isn't correctly handling potential
>> build_vector / scalar_to_vector truncation requirements, in this case
>> in getScalarValueForVectorElement in X86ISelLowering.cpp
>>
>> I'll push a patch with the test case shortly.
>>
>> Simon.
>>
>>> On 19 Apr 2015, at 19:55, Patrik Hägglund H <patrik.h.hagglund at ericsson.com> wrote:
>>>
>>> Hi Simon,
>>>
>>> This commit is causing a regression for the following testcase on x86 (generated by llvm-stress and then reduced):
>>>
>>> define void @autogen_SD13708(i32) {
>>> BB:
>>> %Shuff7 = shufflevector <8 x i32> zeroinitializer, <8 x i32>
>>> zeroinitializer, <8 x i32> <i32 8, i32 10, i32 12, i32 14, i32 undef,
>>> i32 2, i32 4, i32 undef> br label %CF
>>>
>>> CF:
>>> %Tr = trunc <8 x i64> zeroinitializer to <8 x i32>
>>> %Shuff20 = shufflevector <8 x i32> %Shuff7, <8 x i32> %Tr, <8 x i32>
>>> <i32 13, i32 15, i32 1, i32 3, i32 5, i32 7, i32 undef, i32 11> br
>>> i1 undef, label %CF, label %CF247
>>>
>>> CF247:
>>> %I171 = insertelement <8 x i32> %Shuff20, i32 %0, i32 0 br i1 undef,
>>> label %CF, label %CF247 }
>>>
>>>
>>> bin/llc -march=x86-64 -mcpu=corei7 foo.ll
>>> llc: ../lib/CodeGen/SelectionDAG/SelectionDAG.cpp:2989: llvm::SDValue llvm::SelectionDAG::getNode(unsigned int, llvm::SDLoc, llvm::EVT, llvm::SDValue): Assertion `VT.getSizeInBits() == Operand.getValueType().getSizeInBits() && "Cannot BITCAST between types of different sizes!"' failed.
>>> 0 llc 0x0000000001f1c7d5 llvm::sys::PrintStackTrace(llvm::raw_ostream&) + 44
>>> 1 llc 0x0000000001f1cb51
>>> 2 llc 0x0000000001f1b729
>>> 3 libpthread.so.0 0x00007ff8431a8810
>>> 4 libc.so.6 0x00007ff8429c3755 gsignal + 53
>>> 5 libc.so.6 0x00007ff8429c4d31 abort + 385
>>> 6 libc.so.6 0x00007ff8429bc610 __assert_fail + 240
>>> 7 llc 0x0000000001da0fb3 llvm::SelectionDAG::getNode(unsigned int, llvm::SDLoc, llvm::EVT, llvm::SDValue) + 8345
>>> 8 llc 0x000000000146668f
>>> 9 llc 0x00000000014668fb
>>> 10 llc 0x000000000146c744
>>> 11 llc 0x0000000001473c93
>>> 12 llc 0x000000000147eb70
>>> 13 llc 0x00000000014b631b llvm::X86TargetLowering::LowerOperation(llvm::SDValue, llvm::SelectionDAG&) const + 389
>>> 14 llc 0x0000000001d5988b
>>> 15 llc 0x0000000001d71358 llvm::SelectionDAG::Legalize() + 298
>>> 16 llc 0x0000000001e11f72 llvm::SelectionDAGISel::CodeGenAndEmitDAG() + 2960
>>> 17 llc 0x0000000001e110b6 llvm::SelectionDAGISel::SelectBasicBlock(llvm::ilist_iterator<llvm::Instruction const>, llvm::ilist_iterator<llvm::Instruction const>, bool&) + 234
>>> 18 llc 0x0000000001e14745 llvm::SelectionDAGISel::SelectAllBasicBlocks(llvm::Function const&) + 2939
>>> 19 llc 0x0000000001e10254 llvm::SelectionDAGISel::runOnMachineFunction(llvm::MachineFunction&) + 998
>>> 20 llc 0x000000000141f675
>>> 21 llc 0x0000000001818007 llvm::MachineFunctionPass::runOnFunction(llvm::Function&) + 95
>>> 22 llc 0x0000000001b4d64c llvm::FPPassManager::runOnFunction(llvm::Function&) + 290
>>> 23 llc 0x0000000001b4d7bc llvm::FPPassManager::runOnModule(llvm::Module&) + 84
>>> 24 llc 0x0000000001b4db10
>>> 25 llc 0x0000000001b4e1ac llvm::legacy::PassManagerImpl::run(llvm::Module&) + 250
>>> 26 llc 0x0000000001b4e3cd llvm::legacy::PassManager::run(llvm::Module&) + 39
>>> 27 llc 0x0000000000c26bf5
>>> 28 llc 0x0000000000c25cdb main + 242
>>> 29 libc.so.6 0x00007ff8429afc16 __libc_start_main + 230
>>> 30 llc 0x0000000000c25269
>>> Stack dump:
>>> 0. Program arguments: bin/llc -march=x86-64 -mcpu=corei7 /home/uabpath/tmp/foo.ll
>>> 1. Running pass 'Function Pass Manager' on module '/home/uabpath/tmp/foo.ll'.
>>> 2. Running pass 'X86 DAG->DAG Instruction Selection' on function '@autogen_SD13708'
>>> Abort (core dumped)
>>>
>>> /Patrik Hägglund
>>>
>>>
>>> -----Original Message-----
>>> From: llvm-commits-bounces at cs.uiuc.edu
>>> [mailto:llvm-commits-bounces at cs.uiuc.edu] On Behalf Of Simon Pilgrim
>>> Sent: den 16 april 2015 10:21
>>> To: llvm-commits at cs.uiuc.edu
>>> Subject: [llvm] r235079 - TRUNCATE constant folding - minor fix for
>>> rL233224
>>>
>>> Author: rksimon
>>> Date: Thu Apr 16 03:21:09 2015
>>> New Revision: 235079
>>>
>>> URL: http://llvm.org/viewvc/llvm-project?rev=235079&view=rev
>>> Log:
>>> TRUNCATE constant folding - minor fix for rL233224
>>>
>>> Fix for test case found by James Molloy - TRUNCATE of constant build vectors can be more simply achieved by simply replacing with a new build vector node with the truncated value type - no need to touch the scalar operands at all.
>>>
>>> Added:
>>> llvm/trunk/test/CodeGen/AArch64/fold-constants.ll
>>> Modified:
>>> llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
>>>
>>> Modified: llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
>>> URL:
>>> http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionD
>>> A G/SelectionDAG.cpp?rev=235079&r1=235078&r2=235079&view=diff
>>> =====================================================================
>>> =
>>> ========
>>> --- llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAG.cpp (original)
>>> +++ llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAG.cpp Thu Apr 16
>>> +++ 03:21:09 2015
>>> @@ -2851,13 +2851,16 @@ SDValue SelectionDAG::getNode(unsigned O
>>> // FIXME: Entirely reasonable to perform folding of other unary
>>> // operations here as the need arises.
>>> break;
>>> + case ISD::TRUNCATE:
>>> + // Constant build vector truncation can be done with the original scalar
>>> + // operands but with a new build vector with the truncated value type.
>>> + return getNode(ISD::BUILD_VECTOR, DL, VT, BV->ops());
>>> case ISD::FNEG:
>>> case ISD::FABS:
>>> case ISD::FCEIL:
>>> case ISD::FTRUNC:
>>> case ISD::FFLOOR:
>>> case ISD::FP_EXTEND:
>>> - case ISD::TRUNCATE:
>>> case ISD::UINT_TO_FP:
>>> case ISD::SINT_TO_FP: {
>>> // Let the above scalar folding handle the folding of each element.
>>>
>>> Added: llvm/trunk/test/CodeGen/AArch64/fold-constants.ll
>>> URL:
>>> http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AArch64/f
>>> o
>>> ld-constants.ll?rev=235079&view=auto
>>> =====================================================================
>>> =
>>> ========
>>> --- llvm/trunk/test/CodeGen/AArch64/fold-constants.ll (added)
>>> +++ llvm/trunk/test/CodeGen/AArch64/fold-constants.ll Thu Apr 16
>>> +++ 03:21:09 2015
>>> @@ -0,0 +1,21 @@
>>> +; RUN: llc -mtriple=aarch64-linux-gnu -o - %s | FileCheck %s
>>> +
>>> +define i64 @dotests_616() {
>>> +; CHECK-LABEL: dotests_616
>>> +; CHECK: movi d0, #0000000000000000
>>> +; CHECK-NEXT: umov w8, v0.b[2]
>>> +; CHECK-NEXT: sbfx w8, w8, #0, #1
>>> +; CHECK-NEXT: fmov s0, w8
>>> +; CHECK-NEXT: fmov x0, d0
>>> +; CHECK-NEXT: ret
>>> +entry:
>>> + %0 = bitcast <2 x i64> zeroinitializer to <8 x i16>
>>> + %1 = and <8 x i16> zeroinitializer, %0
>>> + %2 = icmp ne <8 x i16> %1, zeroinitializer
>>> + %3 = extractelement <8 x i1> %2, i32 2
>>> + %vgetq_lane285 = sext i1 %3 to i16
>>> + %vset_lane = insertelement <4 x i16> undef, i16 %vgetq_lane285,
>>> +i32
>>> +0
>>> + %4 = bitcast <4 x i16> %vset_lane to <1 x i64>
>>> + %vget_lane = extractelement <1 x i64> %4, i32 0
>>> + ret i64 %vget_lane
>>> +}
>>>
>>>
>>> _______________________________________________
>>> 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