similar to: [LLVMdev] Adding instructions to MachineBlock

On Wed, Jun 3, 2009 at 12:46 PM, Manjunath Kudlur<keveman at gmail.com> wrote: > Hello, > > I am writing a MachineFunction pass that converts vector instructions > to a sequence of scalar instructions. Why? That really isn't the level you want to be doing that sort of thing normally. Usually, legalization turns illegal vector operations into legal scalar operations. -Eli

I am seeing the following assertion in leak detector. /llvm/lib/VMCore/LeakDetector.cpp:43: void<unnamed>::LeakDetectorImpl<T>::addGarbage(const T*) [with T = void]: Assertion `Ts.count(Cache) == 0 && "Object already in set!"' failed. I am creating a list of instructions using BuildMI() and adding them to a basic block using BB->insert(). I am seeing this

On Wed, Jun 3, 2009 at 5:10 PM, Manjunath Kudlur <keveman at gmail.com> wrote: > I am seeing the following assertion in leak detector. > > /llvm/lib/VMCore/LeakDetector.cpp:43: > void<unnamed>::LeakDetectorImpl<T>::addGarbage(const T*) [with T = > void]: Assertion `Ts.count(Cache) == 0 && "Object already in set!"' > failed. > > I am

Hi Bill, I am using the following version of BuildMI : MachineInstrBuilder BuildMI(MachineFunction &MF, const TargetInstrDesc &TID, unsigned DestReg) I do the following : void createInstrs(std::vector<MachineInstr *>& ilist) { Machine Instr *mi; mi = BuildMI(MF, someTID, somereg);

I am trying to JIT some code containing an indirect branch (and the corresponding store i8* blockaddress(@label)). I am using LLVM 2.7 code base. I build the ExecutionEngine using EngineBuilder, and call engine->getPointerToFunction(func). When I use setOptLevel(llvm::CodeGenOpt::None), the JITing fails with the following message : JIT.h:131: virtual void*

Dan, Thanks a lot. Using a modifier in the assembly string works for this case. I am trying to solve a related problem. I am trying to print out a set of "mov" ops for the vector_shuffle node. Since the source of the "mov" is from one of the sources to vector_shuffle, depending on the mask, I am not sure what assembly string to emit. For example, if I have d <-

Hello, I am trying to create a machine instruction for "extractelement". I want to translate r <- extractelement v, 0 to mov r, v.x I was looking at the dag I can use and I found vector_extract. The inputs for this SDnode are a register and a iPtr constant. With that, I need to create 4 separate def's to extract element 0, 1, 2, and 3 and translate to v.x, v.y, v.z, and v.w. I

Manjunath, I had a very similar problem and I solved it using a custom vector shuffle and addition instead of mov. For example, Vector_shuffle s1, s2, <0,3> is mapped to a custom instruction where I transform the swizzle to a 32bit integer mask and an inverted mask. So I have dst, src0, src1, imm1, imm2 And I have my asm look similar to: Add dst, src0.imm1, src1.imm2 and then in the asm

I want to add a set of "properties" to each instruction in my instruction set, and want to be able to query that in my machine-specific optimizations. My Insts.td file looks as follows : class InstProperty; def propX : InstProperty; def propY : InstProperty; def propZ : InstProperty; class myInst<..., list<InstProperty> props> : Instruction { ... ...

One way to do this is to handle this in the AsmPrinter, with operand modifiers. For example, on x86 there are instructions with ${dst:call} in their asm string. The "call" part is interpreted as an operand modifier. The assembly printer looks for the "call" modifier on MachineOperand::MO_Immediate operands (in X86ATTAsmPrinter::printOperand), which lets it perform custom

Manjunath, I asked this question recently, but instead of telling you to search the archive I'm going to take it as a chance to recall how to do it (because I'll do so anyway and even telling tablegen to use an enum for instructions is not as trivial to do as you might think : ) I wrote it up in the wiki at http://wiki.llvm.org/HowTo:_Add_arbitrary_properties_to_instructions Regards,

Christian, Thanks for your reply and the wiki entry. I did search the archives, but evidently I didn't search for the right thing. My bad. Anyways, I am still wondering about the other part of my question. Why aren't there Tablegen backends specific to some architecture backends? Let me describe a different scenario. Suppose my architecture has vector and scalar units, and suppose I want

Hello, I am trying to convert the subtree (vector_shuffle v2f32, v2f32 (build_vector imm1, imm2)) to a machine instruction that takes 2 v2f32's and 2 immediates. I tried the following table gen pattern : (set v2f32Reg:$dst, (vector_shuffle v2f32Reg:$src1, v2f32Reg:$src2, (build_vector imm:$c1, imm:$c2))) Table gen barfs about type

Dan, > In general, it's a design goal to structure CodeGen features as > target-independent code parameterized with target-specific data. > The degree to which this goal is met in LLVM CodeGen features today > varies, but features that have been implemented using TableGen have > been relatively successful. Can you give an example of a relatively successful instance where

I was going through the function DAGTypeLegalizer::SplitVecRes_LOAD in LegalizeVectorTypes.cpp. I noticed that it is using getSizeInBits()/8 to compute IncrementSize, which is the offset for the load of second half of the vector. I have a situation where the frontend is producing load for a <2 x i1> type, and the architecture has i1 registers (but not v2i1 registers). The store size of i1 is

Hello, I want to have an operation foo with variable number of operands, and I am trying to achieve it using multiple operations. Let's say I want to have [FOO r1, r2], I am constructing the DAG as follows : consumeArg(r1) -> consumeArg(r2) -> FOO -> arg(r1) -> arg(r2) Note that the arrows are all "Chain"s. I need to have consumeArg(r1) because, I don't want the

Hello, Given a TargetRegisterClass *RC, I was wondering if there is a way to find out what register class it is directly, instead of comparing it against all the &mytarget::Class1, &mytarget::Class2 etc. This goes back to my original intention of having special query functions for a subset of register classes. Suppose I wanted vector register classes (more than one, say 2-elements and

> All of the tablegen backends work this way. As you mentioned, > there are no target-specific tablegen backends at present. > > The underlying observation here is that features are never > fundamentally "specific for a target". For example, a mapping > between vector opcodes and associated scalar opcodes could > reasonably be made on many architectures. Even >

I am looking for the slot index of a register around the given slot index Min. Is there a better way than the linear search: ... findDefUsesAroundIndex( LiveInterval* li, SlotIndex Min ) ... for( MachineOperand * mo = MRI->getRegUseDefListHead(li->reg); mo; mo = mo->getNextOperandForReg() ) { SlotIndex si = SI->getInstructionIndex( use.getOperand().getParent() ); if(

On Jun 8, 2009, at 9:58 AM, Manjunath Kudlur wrote: > > Now, I can write a tablegen backend that can output a opcode->opcode > map, which can be used by the pass. Is this and intended use case for > Tablegen? Or, are tablegen backends supposed to be generic to all > architecture backends? In general, it's a design goal to structure CodeGen features as target-independent code