This is an archive of the discontinued LLVM Phabricator instance.

Differential D109118

[AArch64][SVE][InstCombine] Canonicalize aarch64_sve_dup_x intrinsic to IR splat operation
ClosedPublic

Authored by mnadeem on Sep 1 2021, 6:17 PM.

Details

Reviewers
efriedma
david-arm
paulwalker-arm
sdesmalen
Commits
rGab111e982fd1: Revert "Revert "[AArch64][SVE][InstCombine] Canonicalize aarch64_sve_dup_x…
rGeee7d225ded9: Revert "[AArch64][SVE][InstCombine] Canonicalize aarch64_sve_dup_x intrinsic to…
rG98c37247d81d: [AArch64][SVE][InstCombine] Canonicalize aarch64_sve_dup_x intrinsic to IR…

Diff Detail

Event Timeline

mnadeem created this revision.Sep 1 2021, 6:17 PM
Herald added a reviewer: efriedma. · View Herald TranscriptSep 1 2021, 6:17 PM
Herald added subscribers: psnobl, hiraditya, kristof.beyls, tschuett. · View Herald Transcript
mnadeem requested review of this revision.Sep 1 2021, 6:17 PM
Herald added a project: Restricted Project. · View Herald TranscriptSep 1 2021, 6:17 PM
Herald added a subscriber: llvm-commits. · View Herald Transcript
mnadeem added reviewers: david-arm, paulwalker-arm, sdesmalen.Sep 1 2021, 6:18 PM
Harbormaster completed remote builds in B122216: Diff 370130.Sep 1 2021, 6:53 PM
paulwalker-arm added a comment.Sep 2 2021, 2:54 AM

The patch looks good but please add an explicit test for the dup_x->splat transform to sve-intrinsic-opts-dup.ll.

The above failures are because many of the SVE ACLE tests use dup_x so unfortunately those will need updating as well. I guess this is the trade off, the SVE ACLE tests use -O1 so that they can ignore local variable stack access but then it means updates are required when we canonicalising the intrinsics to LLVM constants/instructions. Another option is to perhaps disable instcombine for those tests but I've not looked to see if that is possible, but my gut feeling is probably not.

Matt added a subscriber: Matt.Sep 2 2021, 7:05 AM
mnadeem updated this revision to Diff 370463.Sep 2 2021, 6:36 PM

Updated acle tests and added an explicit test.

Harbormaster completed remote builds in B122435: Diff 370463.Sep 2 2021, 8:59 PM
paulwalker-arm accepted this revision.Sep 3 2021, 2:55 AM
This revision is now accepted and ready to land.Sep 3 2021, 2:55 AM
This revision was landed with ongoing or failed builds.Sep 10 2021, 12:53 PM
Closed by commit rG98c37247d81d: [AArch64][SVE][InstCombine] Canonicalize aarch64_sve_dup_x intrinsic to IR… (authored by mnadeem). · Explain Why
This revision was automatically updated to reflect the committed changes.
mnadeem added a commit: rG98c37247d81d: [AArch64][SVE][InstCombine] Canonicalize aarch64_sve_dup_x intrinsic to IR….
Herald added a project: Restricted Project. · View Herald TranscriptSep 10 2021, 12:53 PM
Herald added a subscriber: cfe-commits. · View Herald Transcript
mnadeem added a reverting change: rGeee7d225ded9: Revert "[AArch64][SVE][InstCombine] Canonicalize aarch64_sve_dup_x intrinsic to….Sep 10 2021, 1:02 PM
mnadeem added commits: rGeee7d225ded9: Revert "[AArch64][SVE][InstCombine] Canonicalize aarch64_sve_dup_x intrinsic to…, rGab111e982fd1: Revert "Revert "[AArch64][SVE][InstCombine] Canonicalize aarch64_sve_dup_x….Sep 10 2021, 6:26 PM

Large Diff

This large diff affects 129 files. Files without inline comments have been collapsed. Expand All Files

Revision Contents

PathSize
clang/
test/
CodeGen/
aarch64-sve-intrinsics/
199 lines
12 lines
12 lines
12 lines
12 lines
198 lines
144 lines
126 lines
6 lines
6 lines
6 lines
144 lines
84 lines
102 lines
102 lines
102 lines
102 lines
84 lines
18 lines
126 lines
126 lines
24 lines
12 lines
156 lines
52 lines
144 lines
54 lines
54 lines
198 lines
198 lines
54 lines
198 lines
54 lines
198 lines
198 lines
198 lines
198 lines
144 lines
54 lines
54 lines
54 lines
54 lines
54 lines
144 lines
48 lines
48 lines
54 lines
198 lines
198 lines
6 lines
6 lines
aarch64-sve2-intrinsics/
48 lines
36 lines
36 lines
36 lines
36 lines
12 lines
12 lines
36 lines
36 lines
36 lines
18 lines
36 lines
36 lines
36 lines
48 lines
24 lines
24 lines
24 lines
48 lines
48 lines
48 lines
48 lines
48 lines
48 lines
144 lines
144 lines
144 lines
42 lines
42 lines
42 lines
42 lines
36 lines
36 lines
48 lines
6 lines
24 lines
6 lines
24 lines
6 lines
144 lines
18 lines
18 lines
18 lines
18 lines
18 lines
18 lines
24 lines
18 lines
18 lines
24 lines
24 lines
24 lines
144 lines
144 lines
144 lines
144 lines
36 lines
36 lines
144 lines
144 lines
36 lines
36 lines
12 lines
12 lines
72 lines
36 lines
36 lines
36 lines
18 lines
36 lines
18 lines
36 lines
36 lines
72 lines
llvm/
lib/
Target/
AArch64/
48 lines
test/
Transforms/
InstCombine/
AArch64/
18 lines
18 lines
11 lines

Diff 371992

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_abd.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_acge.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_acgt.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_acle.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_aclt.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_add.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_and.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_asr.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_bfdot.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_bfmlalb.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_bfmlalt.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_bic.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_cmpeq.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_cmpge.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_cmpgt.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_cmple.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_cmplt.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_cmpne.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_cmpuo.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_div.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_divr.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_dot.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_dup-bfloat.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_dup.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_dupq.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_eor.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_lsl.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_lsr.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_mad.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_max.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_maxnm.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_min.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_minnm.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_mla.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_mls.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_msb.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_mul.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_mulh.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_mulx.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_nmad.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_nmla.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_nmls.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_nmsb.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_orr.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_qadd.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_qsub.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_scale.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_sub.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_subr.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_sudot.c

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_usdot.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_aba.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_abalb.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_abalt.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_abdlb.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_abdlt.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_adclb.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_adclt.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_addhnb.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_addhnt.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_addlb.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_addlbt.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_addlt.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_addwb.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_addwt.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_bcax.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_bdep.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_bext.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_bgrp.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_bsl.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_bsl1n.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_bsl2n.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_eor3.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_eorbt.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_eortb.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_hadd.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_hsub.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_hsubr.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_mlalb.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_mlalt.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_mlslb.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_mlslt.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_mullb.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_mullt.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_nbsl.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_pmul.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_pmullb.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_pmullb_128.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_pmullt.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_pmullt_128.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_qadd.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_qdmlalb.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_qdmlalbt.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_qdmlalt.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_qdmlslb.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_qdmlslbt.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_qdmlslt.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_qdmulh.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_qdmullb.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_qdmullt.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_qrdmlah.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_qrdmlsh.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_qrdmulh.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_qrshl.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_qshl.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_qsub.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_qsubr.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_raddhnb.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_raddhnt.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_rhadd.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_rshl.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_rsubhnb.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_rsubhnt.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_sbclb.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_sbclt.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_sqadd.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_subhnb.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_subhnt.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_sublb.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_sublbt.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_sublt.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_subltb.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_subwb.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_subwt.c

clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_uqadd.c

llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp

llvm/test/Transforms/InstCombine/AArch64/sve-intrinsic-fmul-idempotency.ll

llvm/test/Transforms/InstCombine/AArch64/sve-intrinsic-mul-idempotency.ll

llvm/test/Transforms/InstCombine/AArch64/sve-intrinsic-opts-dup.ll