[LLVMdev] Crash in SLP for vector data type as function argument.
Suyog Kamal Sarda
suyog.sarda at samsung.com
Thu Jan 8 03:34:02 PST 2015
Hi Arnold, Shahid,
Thanks for reply and suggestions.
I would go with Arnold's suggestion not to cast VectorizedValue to Instruction and use it as it is.
I can commit patch on this (trivial change in assignment to TempVec).
However, this issue hits only when we have http://reviews.llvm.org/D6818 properly committed.
It will take some time for me to improve that patch.
This crash issue doesn't have anything to do with above patch code, so I am not getting how to put a test case for this.
Can I simply commit it without test case? If yes, should I go ahead with the commit after check-all without putting it for approval?
Please suggest a proper way to do this.
Regards,
Suyog
------- Original Message -------
Sender : Arnold Schwaighofer<aschwaighofer at apple.com>
Date : Jan 08, 2015 02:05 (GMT+09:00)
Title : Re: [LLVMdev] Crash in SLP for vector data type as function argument.
The code in emitReduction has to be fixed. As your example shows it is not safe to assume we will always have an instruction as a result of vectorizeTree(). It seems to me that we can just remove the line that performs the cast. All subsequent uses of the value ‘ValToReduce’ actually are uses of “Value *TmpVec”. The IRBuilder in the variable “Builder” carries the insertion point for all operations in this function (inserting after the instruction “ValToReduce” would be a reason why we need an “Instruction”).
/// \brief Emit a horizontal reduction of the vectorized value.
Value *emitReduction(Value *VectorizedValue, IRBuilder<> &Builder) {
assert(VectorizedValue && "Need to have a vectorized tree node");
Instruction *ValToReduce = dyn_cast(VectorizedValue);
assert(isPowerOf2_32(ReduxWidth) &&
"We only handle power-of-two reductions for now");
Value *TmpVec = ValToReduce;
for (unsigned i = ReduxWidth / 2; i != 0; i >>= 1) {
if (IsPairwiseReduction) {
Value *LeftMask =
createRdxShuffleMask(ReduxWidth, i, true, true, Builder);
Value *RightMask =
createRdxShuffleMask(ReduxWidth, i, true, false, Builder);
Value *LeftShuf = Builder.CreateShuffleVector(
TmpVec, UndefValue::get(TmpVec->getType()), LeftMask, "rdx.shuf.l");
Value *RightShuf = Builder.CreateShuffleVector(
TmpVec, UndefValue::get(TmpVec->getType()), (RightMask),
"rdx.shuf.r");
TmpVec = createBinOp(Builder, ReductionOpcode, LeftShuf, RightShuf,
"bin.rdx");
} else {
Value *UpperHalf =
createRdxShuffleMask(ReduxWidth, i, false, false, Builder);
Value *Shuf = Builder.CreateShuffleVector(
TmpVec, UndefValue::get(TmpVec->getType()), UpperHalf, "rdx.shuf");
TmpVec = createBinOp(Builder, ReductionOpcode, TmpVec, Shuf, "bin.rdx");
}
}
Thanks,
Arnold
> On Jan 7, 2015, at 3:34 AM, Suyog Kamal Sarda wrote:
>
> Hi Shahid,
>
> Thanks for the reply.
>
> Actually, yes, the emitreduction() takes vectorizedvalue which is leaf of the tree. '
> I got confused by the name of the argument passed while calling emitReduction().
>
> Value *ReducedSubTree = emitReduction(VectorizedRoot, Builder)
>
> Anyways, that should hardly matter.
>
> I had mentioned the test case :
>
> int foo(uint32x4_t a) {
> return a[0] + a[1] + a[2] + a[3];
> }
>
> LLVM IR :
>
> define i32 @hadd(<4 x i32> %a) {
> entry:
> %vecext = extractelement <4 x i32> %a, i32 0
> %vecext1 = extractelement <4 x i32> %a, i32 1
> %add = add i32 %vecext, %vecext1
> %vecext2 = extractelement <4 x i32> %a, i32 2
> %add3 = add i32 %add, %vecext2
> %vecext4 = extractelement <4 x i32> %a, i32 3
> %add5 = add i32 %add3, %vecext4
> ret i32 %add5
> }
>
> Now, when leaf %vecext is reached, the vectorizeTree() function call sets the VectorizedValue to 0th operand of extractelement instruction.
>
> case Instruction::ExtractElelement: {
> if(CanReuseExtract(E->Scalars)) {
> Value *V = VL0->getOperand(0);
> E->VectorizedValue = V;
> return V;
> }
> return Gather(E->Scalars, VecTy);
> }
>
> Now in emitReduction(), the VectorizedValue is dyn_cast to Instruction.
> In above IR, %a is not an instruction (function argument), hence while referring the casted value which is null,
> crash occurs.
>
> Instruction *ValToReduce = dyn_cast(VectorizedValue);
>
> Note : The above test case won't crash with current svn version, since code for parsing the tree for above IR is yet
> to be included in svn. Initial patch was submitted in http://reviews.llvm.org/D6818.
> I am working on refining it, however, the above code flow is not disturbed at all in my patch of parsing.
> You can try to reproduce the problem by importing above patch in local code.
>
> When the vector data type 'a' is in global scope, a 'load' instruction is generated in basic block of the function:
>
> test case 2:
>
> unint32x4_t a;
> int foo() {
> return a[0] + a[1] + a[2] + a[3];
> }
>
> IR for above test case :
>
> @a = common global <4 x i32> zeroinitializer, align 16
>
> define i32 @hadd() #0 {
> entry:
> %0 = load <4 x i32>* @a, align 16, !tbaa !1
> %vecext = extractelement <4 x i32> %0, i32 0
> %vecext1 = extractelement <4 x i32> %0, i32 1
> %add = add i32 %vecext, %vecext1
> %vecext2 = extractelement <4 x i32> %0, i32 2
> %add3 = add i32 %add, %vecext2
> %vecext4 = extractelement <4 x i32> %0, i32 3
> %add5 = add i32 %add3, %vecext4
> ret i32 %add5
> }
>
> Now, since here, 0th operand of leaf %vecext is a load instruction,
> the dyn_casting into an instruction will succeed here and reduction will be emitted properly.
>
> How can we solve this problem? What type of casting should a function argument belong to?
>
> Regards,
> Suyog
>
>
>
> ------- Original Message -------
> Sender : Shahid, Asghar-ahmad
> Date : Jan 07, 2015 20:05 (GMT+09:00)
> Title : RE: [LLVMdev] Crash in SLP for vector data type as function argument.
>
> Hi Suyog,
>
> IMO emitReduction() takes a vectorized value which is the leafs of the matched pattern/tree.
> So what you are thinking as root is actually the leaf of the tree.
> Root should actually be the value which is being feed to the "return" statement.
>
> It would be of great help if you could, share the sample test?
>
> Regards,
> Shahid
>
>> -----Original Message-----
>> From: llvmdev-bounces at cs.uiuc.edu [mailto:llvmdev-bounces at cs.uiuc.edu]
>> On Behalf Of Suyog Kamal Sarda
>> Sent: Monday, January 05, 2015 5:40 PM
>> To: nrotem at apple.com; aschwaighofer at apple.com;
>> mzolotukhin at apple.com; james.molloy at arm.com
>> Cc: llvmdev at cs.uiuc.edu
>> Subject: [LLVMdev] Crash in SLP for vector data type as function argument.
>>
>> Hi all,
>>
>> Came across a crash in SLP vectorization while testing following code for
>> AArch64 :
>>
>> int foo(uint32x4_t a) {
>> return a[0] + a[1] + a[2] + a[3];
>> }
>>
>> The LLVM IR for above code will be:
>>
>> define i32 @hadd(<4 x i32> %a) {
>> entry:
>> %vecext = extractelement <4 x i32> %a, i32 0
>> %vecext1 = extractelement <4 x i32> %a, i32 1
>> %add = add i32 %vecext, %vecext1
>> %vecext2 = extractelement <4 x i32> %a, i32 2
>> %add3 = add i32 %add, %vecext2
>> %vecext4 = extractelement <4 x i32> %a, i32 3
>> %add5 = add i32 %add3, %vecext4
>> ret i32 %add5
>> }
>>
>> I somehow try to recognize this pattern and try to vectorize it using existing
>> code for horizontal reductions (I just recognize the pattern and fill up the
>> data, rest is done by already existing code.
>> I do pattern matching very badly though, but that's a different story).
>>
>>
>> Please note that whatever follows is with existing code, I haven't modified
>> any bit of it.
>>
>> Now, once the pattern is recognized, we call "trytoReduce()" where we try
>> to vectorize tree by function call "vectorizeTree()" which returns root of the
>> vectorized tree. Then we emit the reduction using call "emitRedcution()"
>> which takes the root of the vector tree as argument. Inside
>> "emitReduction()", we cast root of the tree into an instruction.
>>
>> Now, for above case, while setting the root of the vectorized tree,
>> extractelement instruction is encountered, and its 0th operand is set as the
>> root of the tree, which in above case is "%a". However, this is not an
>> instruction and hence, when we cast it into an instruction in
>> "emitReduction()" code, it returns nullptr which causes a crash ahead when
>> referencing it.
>>
>> Take a second case where the vector data type is in global scope.
>>
>> unint32x4_t a;
>> int foo() {
>> return a[0] + a[1] + a[2] + a[3];
>> }
>>
>> The IR for above code is:
>>
>> @a = common global <4 x i32> zeroinitializer, align 16
>>
>> define i32 @hadd() #0 {
>> entry:
>> %0 = load <4 x i32>* @a, align 16, !tbaa !1
>> %vecext = extractelement <4 x i32> %0, i32 0
>> %vecext1 = extractelement <4 x i32> %0, i32 1
>> %add = add i32 %vecext, %vecext1
>> %vecext2 = extractelement <4 x i32> %0, i32 2
>> %add3 = add i32 %add, %vecext2
>> %vecext4 = extractelement <4 x i32> %0, i32 3
>> %add5 = add i32 %add3, %vecext4
>> ret i32 %add5
>> }
>>
>> Now in above case, 0th operand of extractelement %0 is a load instruction,
>> and hence it doesn't crash while casting into an instruction and runs smoothly
>> further.
>>
>> Can someone please suggest how to resolve this? Is there something I am
>> missing or is it a basic problem with IR itself ?
>>
>> Regards,
>> Suyog
>>
>>
>> _______________________________________________
>> LLVM Developers mailing list
>> LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
More information about the llvm-dev
mailing list