[AIE2] Use OR to mimic MOV when copying GPR to GPR #221
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
krishnamtibrewala
force-pushed
the
aie2-mov-add
branch
from
5617017
to
3e3143a
Compare
krishnamtibrewala
changed the title
krishnamtibrewala
force-pushed
the
aie2-mov-add
branch
2 times, most recently
from
3fee812
to
10dafa6
Compare
|
| @@ -94,6 +94,8 @@ let Itinerary = II_NE in { | |||
| } | |||
| let Itinerary = II_OR in { | |||
| def OR : AIE2_alu_r_rr_inst_alu<0b0101, (outs eR:$mRx), (ins eR:$mRx0, eR:$mRy), "or", "$mRx, $mRx0, $mRy">; | |||
| let isCodeGenOnly = 1 in | |||
| def MOV_OR : AIE2_alu_r_rr_inst_alu_mov_gpr<0b0101, (outs eR:$mRx), (ins eR:$mRx0), "or", "$mRx, $mRx0, $mRx0">; | |||
There was a problem hiding this comment.
I wouldn't mind defining this directly in terms of AIE2_inst_alu_instr32, computing the alu image here.
There was a problem hiding this comment.
Hi @martien-de-jong kindly clarify on what you mean by "alu image". Thank you.
There was a problem hiding this comment.
let alu = {mRx0,mRx,mRx0,op,0b1}; as used in AIE2_alu_r_rr_inst_alu_mov_gpr
There was a problem hiding this comment.
I would like to be consistent with how all the other instructions are defined in the file.
If you think exposing alu image here is better, I will update it :).
gbossu
reviewed
Oct 24, 2024
krishnamtibrewala
force-pushed
the
aie2-mov-add
branch
from
10dafa6
to
b22a5b9
Compare
krishnamtibrewala
requested review from
abhinay-anubola,
abnikant,
khallouh,
konstantinschwarz,
SagarMaheshwari99 and
stephenneuendorffer
as code owners
krishnamtibrewala
force-pushed
the
aie2-mov-add
branch
from
b22a5b9
to
275b8c4
Compare
| @@ -829,7 +838,8 @@ bool AIE2InstrInfo::expandPostRAPseudo(MachineInstr &MI) const { | |||
| case AIE2::PseudoMove: { | |||
| Register Dst = MI.getOperand(0).getReg(); | |||
| Register Src = MI.getOperand(1).getReg(); | |||
| BuildMI(MBB, MI, DL, get(AIE2::MOV_mv_scl), Dst) | |||
| const unsigned MOVSclOpcode = getScalarMovOpcode(Dst, Src); | |||
There was a problem hiding this comment.
Summarizing, this is 'late code selection' since the move is inserted by the register allocator. We check the registers here, and expand it to MOV_mv_scl immediately if there's no other way, or we expand it to a pseudo for which the scheduler decides whether to put in in MOV or in ALU.
I think it's fine, I'm just wondering whether the decision could already be made by RegAlloc to avoid this multi-stage expansion of pseudos.
martien-de-jong
approved these changes
Nov 11, 2024
Standalone this PR does not create regression, its the combination of #217 that leads to couple of cycles of regression. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR allows to mimic MOV behavior using OR instr when MOV slot is not available.
Note : This works only when MOV was for GPR to GPR due to limitation of OR to only work with GPRs