search for: gicombinerul

...allows us to omit the defs for MI's involved in debug info. Any names used solely in the apply section and defined by a value-constructing operator (e.g. create*) can be omitted With these rules, the defs section isn't completely gone but is often empty and can be omitted with a variant of GICombineRule. The examples would be as below. I've indicated which rule eliminates which def with color coding and strike-through: // Omitted due to rule 1 // Omitted due to rule 2 // We haven't been listing any defs related to rule 3 since these are the formerly anonymous // operands t...

...I's involved in debug info. >> 6. Any names used solely in the apply section and defined by a >> value-constructing operator (e.g. create*) can be omitted >> With these rules, the defs section isn't completely gone but is often empty and can be omitted with a variant of GICombineRule. The examples would be as below. I've indicated which rule eliminates which def with color coding and strike-through: >> // Omitted due to rule 1 >> // Omitted due to rule 2 >> // We haven't been listing any defs related to rule 3 since these are the formerly...

...the examples to the DAG style we were talking about. Hopefully I haven't forgotten anything, there was a lot to keep track of :-). Overall, I think there's a couple places where things get a a little awkward (mainly debug info) but things generally look good to me. A Simple Example def : GICombineRule<(defs reg:$D, reg:$S), (match (G_ZEXT s32:$t1, s8:$S), (G_TRUNC s16:$D, s32:$t1)), (apply (G_ZEXT s16:$D, s8:$S))>; This has been converted to the style that fits within tblgen's DAG type but isn't (directly) a DAG....

...I haven't forgotten anything, there was a lot to keep track of :-). Overall, I think there's a couple places where things get a a little awkward (mainly debug info) but things generally look good to me. > > That's awesome :) > > >> _A Simple Example_ >> def : GICombineRule<(defs reg:$D, reg:$S), >> (match (G_ZEXT s32:$t1, s8:$S), >> (G_TRUNC s16:$D, s32:$t1)), >> (apply (G_ZEXT s16:$D, s8:$S))>; >> This has been converted to the style that fits within tblgen's DAG ty...

...with_shift : GIMatchPredicate< >> (ins reg:$A), (outs uint64_t:$shift), MachineOperandPtr, [{ >> // MachineOperand *matchARMShiftedOp2(const MachineOperand &, uint64_t); >> return matchARMShiftedOp2(${A}, ${imm}); >> }]>; >> def : GICombineRule< >> (defs root:$D, reg:$S1, reg_with_shift:$S2), >> (match [{MIR %D = G_ADD %S1, %S2 }]), >> (apply [{MIR %D = ADD %S1, %S2, %S2.shift }])>; >> (I've just realized there's a gap in the ability to name metadata when used in this way. I've...

...re Combine Rules in tablegen and with the help of the people credited above I think we've ended up somewhere that should be nice to work with and opens up a lot of possibilities. Individual Combine Rules A Simple Example Here's a simple example that eliminates a redundant G_TRUNC. def : GICombineRule<(defs root:$D, operand:$S), (match [{MIR %1(s32) = G_ZEXT %S(s8) %D(s16) = G_TRUNC %1(s32) }]), (apply [{MIR %D(s16) = G_ZEXT %S(s8) }])>; There's three parts to this rule: defs declares the interface for the rule....

...an example of a non-bool result: def reg_with_shift : GIMatchPredicate< (ins reg:$A), (outs uint64_t:$shift), MachineOperandPtr, [{ // MachineOperand *matchARMShiftedOp2(const MachineOperand &, uint64_t); return matchARMShiftedOp2(${A}, ${imm}); }]>; def : GICombineRule< (defs root:$D, reg:$S1, reg_with_shift:$S2), (match [{MIR %D = G_ADD %S1, %S2 }]), (apply [{MIR %D = ADD %S1, %S2, %S2.shift }])>; (I've just realized there's a gap in the ability to name metadata when used in this way. I've gone with 'S2.shift' for now...

...asses? It would be nice to > re-use many of the rules, for example, except they'd match LLVM IR > rather than MIR. As you go about implementation, maybe keep this idea > in mind? > > > Here's a simple example that eliminates a redundant G_TRUNC. > > > > def : GICombineRule<(defs root:$D, operand:$S), > > (match [{MIR %1(s32) = G_ZEXT %S(s8) > > %D(s16) = G_TRUNC %1(s32) }]), > > (apply [{MIR %D(s16) = G_ZEXT %S(s8) }])>; > > The use of '%' vs. '$' here really threw me for a loop. I would have > expected '$D'...

...tCombine and MIR for combines that make sense to both isn't something I intend to work on but it does fit within the syntax proposed. > It would be something along the lines of: > class IRIndependentBinaryOp; > def IRIndependentAdd; > def IRIndependentMul; > def : GICombineRule<(defs ...), (match (IRIndependentMul $C, $A, 2)), (apply (IRIndependentAdd $C, $A, $A))>; > The backend(s) would need to know how to match and create LLVM-IR and MIR versions of IRIndependentMul and IRIndependentAdd much like the GlobalISel Combiner backend needs to know how to match and...

Hi Daniel, Thanks for the reply! Le lun. 26 nov. 2018 à 10:00, Daniel Sanders <daniel_l_sanders at apple.com> a écrit : > > Hi Quentin, > > Sorry for the slow reply. > > > On Nov 16, 2018, at 09:25, Quentin Colombet <quentin.colombet at gmail.com> wrote: > > > > Hi Daniel, > > > > I finally read the proposal. > > > > I have a

...ne and MIR for combines that > make sense to both isn't something I intend to work on but it does fit > within the syntax proposed. > It would be something along the lines of: > class IRIndependentBinaryOp; > def IRIndependentAdd; > def IRIndependentMul; > def : GICombineRule<(defs ...), (match (IRIndependentMul $C, $A, 2)), > (apply (IRIndependentAdd $C, $A, $A))>; > The backend(s) would need to know how to match and create LLVM-IR and MIR > versions of IRIndependentMul and IRIndependentAdd much like the GlobalISel > Combiner backend needs to know ho...

...e G_* opcodes directly correspond to LLVM-IR instructions. The tricky bit will be the escape hatches into C++ would need to either have Instruction/MachineInstr versions or would need to accept both. >> Here's a simple example that eliminates a redundant G_TRUNC. >> >> def : GICombineRule<(defs root:$D, operand:$S), >> (match [{MIR %1(s32) = G_ZEXT %S(s8) >> %D(s16) = G_TRUNC %1(s32) }]), >> (apply [{MIR %D(s16) = G_ZEXT %S(s8) }])>; > > The use of '%' vs. '$' here really threw me for a loop. I would have > expected '$D' an...