[LLVMdev] how to use "new instruction()"
Nick Lewycky
nicholas at mxc.ca
Fri Apr 17 01:56:46 PDT 2015
zhi chen wrote:
> It seems that the problem was because I used builder.CreateFAdd to
> create a <2 x double> vectortype FADD instruction. It works if I use it
> to create the scalar version FADD. I want to have an instruction like:
> *%2 = fadd <2 x double> undef, <2 x double> undef. *The following is the
> way I used to create the vectorized FADD instruction:
>
> //pInst is a double type instruction
>
> Type *vecTy = VectorType::get(pInst->getType(), 2);
> Value *emptyVec = UndefValue::get(vecTy);
> IRBuilder<> builder(&*pInst);
> Value *dupVal = builder.CreateFAdd(emptyVec, emptyVec, instName);
> std::cout << " dupVal " << *dupVal << "\n";
>
> It outputs: dupVal <2 x double> <double fadd (double undef, double
> undef), double fadd (double undef, double undef)>
>
> If I dyn_cast the dupVal to instruction type (dupInst) and print
> dupInst, it outputs: "dupInst printing a <null> value"
> But if I use Instruction *dupInst = (Instruction *) dupVal and print it,
> I'll get:
> dupInst <2 x double> <double fadd (double undef, double undef), double
> fadd (double undef, double undef)>
>
> It seems that if simply fails to generate the vectorized FADD
> instruction. Anything wrong with my code?
IRBuilder gave you back a constant instead of an Instruction. This is
why it returns a Value*. For a simple example, if you ask it to create
"add i32 1, 2" it will not return an add instruction, it will instead
return "i32 3" which is a ConstantInt.
In your case, it returned to you a ConstantExpr whose getOpcode() shows
Instruction::FAdd, and getOperand(0) and getOperand(1) show 'undef', but
it is not an instruction. "Undef" is treated as a constant, so an fadd
between two constants gets you a constant instead of an instruction.
Usually it won't matter, just build the function top down and don't look
at whether you're getting an Instruction* or Constant* and everything
will be fine.
Nick
>
> Best,
> Zhi
>
>
>
>
>
> On Thu, Apr 16, 2015 at 11:55 PM, zhi chen <zchenhn at gmail.com
> <mailto:zchenhn at gmail.com>> wrote:
>
> Yes. I was using this. It seems the produced instruction is not
> correct. There are probably some other problems. I need to recheck
> it. Thanks for your help, Daniel.
>
> Best,
> Zhi
>
> On Thu, Apr 16, 2015 at 11:40 PM, Daniel Berlin <dberlin at dberlin.org
> <mailto:dberlin at dberlin.org>> wrote:
>
> Value * is the instruction.
>
> use dyn_cast<Instruction> to get to it.
>
>
> On Thu, Apr 16, 2015 at 11:39 PM zhi chen <zchenhn at gmail.com
> <mailto:zchenhn at gmail.com>> wrote:
>
> But IRBuilder.CreateXYZ only returns a "VALUE" type. Can I
> get the instruction created by it? For example,
>
> IRBuilder<> builder(&*pinst);
> Value *val = builder.CreateFAdd(LV, RV, "");
>
> How can I get the fadd instruction created by builder?
>
> On Thu, Apr 16, 2015 at 8:52 PM, zhi chen <zchenhn at gmail.com
> <mailto:zchenhn at gmail.com>> wrote:
>
> Yes. That's what I was the solution in my mind. But I
> just wanted to know if there was a generic way to save
> some code...
>
> On Thu, Apr 16, 2015 at 8:32 PM, Tim Northover
> <t.p.northover at gmail.com
> <mailto:t.p.northover at gmail.com>> wrote:
>
> > I understand that I can detect the operation
> first, and use "create" to
> > create for each of them. But I don't if there is a
> generic way to do this
> > because if might be add/sub/mul... operations.
>
> I don't think there is. Realistically, just blindly
> replacing
> instructions with vector equivalents is only going
> to work in a few
> cases anyway. You're probably best to intentionally
> detect those cases
> and call the correct CreateXYZ function.
>
> Cheers.
>
> Tim.
>
>
>
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu <mailto:LLVMdev at cs.uiuc.edu>
> http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>
>
>
>
>
> _______________________________________________
> 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