This is an archive of the discontinued LLVM Phabricator instance.

Differential D136161

[AArch64][SVE] NFC: extend tests for flag-setting predicate instructions
ClosedPublic

Authored by c-rhodes on Oct 18 2022, 5:15 AM.

Details

Reviewers
efriedma
peterwaller-arm
Commits
rG5621caeb8248: [AArch64][SVE] NFC: extend tests for flag-setting predicate instructions
Summary

A follow on patch will extend existing

PTEST(PG, OP(PG, ...)) -> OP_FLAG_SETTING(PG, ...)

optimization in AArch64InstrInfo::optimizePTestInstr to cover more of
the flag-setting instructions

Diff Detail

Event Timeline

c-rhodes created this revision.Oct 18 2022, 5:15 AM
Herald added a reviewer: efriedma. · View Herald TranscriptOct 18 2022, 5:15 AM
Herald added a project: Restricted Project. · View Herald Transcript
Herald added subscribers: psnobl, kristof.beyls, tschuett. · View Herald Transcript
c-rhodes requested review of this revision.Oct 18 2022, 5:15 AM
Herald added a project: Restricted Project. · View Herald TranscriptOct 18 2022, 5:15 AM
c-rhodes added a child revision: D136083: [AArch64][SVE] Use more flag-setting instructions.Oct 18 2022, 5:15 AM
Harbormaster completed remote builds in B192714: Diff 468492.Oct 18 2022, 6:02 AM
Matt added a subscriber: Matt.Oct 19 2022, 5:19 AM
peterwaller-arm accepted this revision.Oct 24 2022, 6:32 AM
This revision is now accepted and ready to land.Oct 24 2022, 6:32 AM
This revision was landed with ongoing or failed builds.Oct 25 2022, 2:02 AM
Closed by commit rG5621caeb8248: [AArch64][SVE] NFC: extend tests for flag-setting predicate instructions (authored by c-rhodes). · Explain Why
This revision was automatically updated to reflect the committed changes.
c-rhodes added a commit: rG5621caeb8248: [AArch64][SVE] NFC: extend tests for flag-setting predicate instructions.

Revision Contents

PathSize
llvm/
test/
CodeGen/
AArch64/
54 lines
98 lines
17 lines

Diff 470416

llvm/test/CodeGen/AArch64/sve-ptest-removal-brk.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=aarch64--linux-gnu -mattr=+sve %s -o - | FileCheck %s ; RUN: llc -mtriple=aarch64--linux-gnu -mattr=+sve %s -o - | FileCheck %s
; Test that redundant ptest instruction is removed when using a flag setting brk ; Test that redundant ptest instruction is removed when using a flag setting brk
define i32 @brkpa(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a, <vscale x 16 x i1> %b) {
; CHECK-LABEL: brkpa:
; CHECK: // %bb.0:
; CHECK-NEXT: brkpa p1.b, p0/z, p1.b, p2.b
; CHECK-NEXT: ptest p0, p1.b
; CHECK-NEXT: cset w0, ne
; CHECK-NEXT: ret
%1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.brkpa.z.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a, <vscale x 16 x i1> %b)
%2 = tail call i1 @llvm.aarch64.sve.ptest.any.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %1)
%conv = zext i1 %2 to i32
ret i32 %conv
}
define i32 @brkpb(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a, <vscale x 16 x i1> %b) { define i32 @brkpb(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a, <vscale x 16 x i1> %b) {
; CHECK-LABEL: brkpb: ; CHECK-LABEL: brkpb:
; CHECK: // %bb.0: ; CHECK: // %bb.0:
; CHECK-NEXT: brkpbs p0.b, p0/z, p1.b, p2.b ; CHECK-NEXT: brkpbs p0.b, p0/z, p1.b, p2.b
; CHECK-NEXT: cset w0, ne ; CHECK-NEXT: cset w0, ne
; CHECK-NEXT: ret ; CHECK-NEXT: ret
%1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.brkpb.z.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a, <vscale x 16 x i1> %b) %1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.brkpb.z.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a, <vscale x 16 x i1> %b)
%2 = tail call i1 @llvm.aarch64.sve.ptest.any.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %1) %2 = tail call i1 @llvm.aarch64.sve.ptest.any.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %1)
%conv = zext i1 %2 to i32 %conv = zext i1 %2 to i32
ret i32 %conv ret i32 %conv
} }
define i32 @brka(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a) {
; CHECK-LABEL: brka:
; CHECK: // %bb.0:
; CHECK-NEXT: brka p1.b, p0/z, p1.b
; CHECK-NEXT: ptest p0, p1.b
; CHECK-NEXT: cset w0, ne
; CHECK-NEXT: ret
%1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.brka.z.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a)
%2 = tail call i1 @llvm.aarch64.sve.ptest.any.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %1)
%conv = zext i1 %2 to i32
ret i32 %conv
}
define i32 @brkb(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a) { define i32 @brkb(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a) {
; CHECK-LABEL: brkb: ; CHECK-LABEL: brkb:
; CHECK: // %bb.0: ; CHECK: // %bb.0:
; CHECK-NEXT: brkbs p0.b, p0/z, p1.b ; CHECK-NEXT: brkbs p0.b, p0/z, p1.b
; CHECK-NEXT: cset w0, ne ; CHECK-NEXT: cset w0, ne
; CHECK-NEXT: ret ; CHECK-NEXT: ret
%1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.brkb.z.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a) %1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.brkb.z.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a)
%2 = tail call i1 @llvm.aarch64.sve.ptest.any.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %1) %2 = tail call i1 @llvm.aarch64.sve.ptest.any.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %1)
Show All 24 Lines; CHECK-NEXT: ret
%2 = call <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32 31) %2 = call <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32 31)
%3 = tail call i1 @llvm.aarch64.sve.ptest.any.nxv16i1(<vscale x 16 x i1> %2, <vscale x 16 x i1> %1) %3 = tail call i1 @llvm.aarch64.sve.ptest.any.nxv16i1(<vscale x 16 x i1> %2, <vscale x 16 x i1> %1)
%conv = zext i1 %3 to i32 %conv = zext i1 %3 to i32
ret i32 %conv ret i32 %conv
} }
; Test that ptest instruction is not removed when using a non-flag setting brk ; Test that ptest instruction is not removed when using a non-flag setting brk
define i32 @brkpa_neg(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a, <vscale x 16 x i1> %b) {
; CHECK-LABEL: brkpa_neg:
; CHECK: // %bb.0:
; CHECK-NEXT: brkpa p0.b, p0/z, p1.b, p2.b
; CHECK-NEXT: ptest p1, p0.b
; CHECK-NEXT: cset w0, ne
; CHECK-NEXT: ret
%1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.brkpa.z.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a, <vscale x 16 x i1> %b)
%2 = tail call i1 @llvm.aarch64.sve.ptest.any.nxv16i1(<vscale x 16 x i1> %a, <vscale x 16 x i1> %1)
%conv = zext i1 %2 to i32
ret i32 %conv
}
define i32 @brkpb_neg(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a, <vscale x 16 x i1> %b) { define i32 @brkpb_neg(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a, <vscale x 16 x i1> %b) {
; CHECK-LABEL: brkpb_neg: ; CHECK-LABEL: brkpb_neg:
; CHECK: // %bb.0: ; CHECK: // %bb.0:
; CHECK-NEXT: brkpb p0.b, p0/z, p1.b, p2.b ; CHECK-NEXT: brkpb p0.b, p0/z, p1.b, p2.b
; CHECK-NEXT: ptest p1, p0.b ; CHECK-NEXT: ptest p1, p0.b
; CHECK-NEXT: cset w0, ne ; CHECK-NEXT: cset w0, ne
; CHECK-NEXT: ret ; CHECK-NEXT: ret
%1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.brkpb.z.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a, <vscale x 16 x i1> %b) %1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.brkpb.z.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a, <vscale x 16 x i1> %b)
%2 = tail call i1 @llvm.aarch64.sve.ptest.any.nxv16i1(<vscale x 16 x i1> %a, <vscale x 16 x i1> %1) %2 = tail call i1 @llvm.aarch64.sve.ptest.any.nxv16i1(<vscale x 16 x i1> %a, <vscale x 16 x i1> %1)
%conv = zext i1 %2 to i32 %conv = zext i1 %2 to i32
ret i32 %conv ret i32 %conv
} }
define i32 @brka_neg(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a) {
; CHECK-LABEL: brka_neg:
; CHECK: // %bb.0:
; CHECK-NEXT: brka p0.b, p0/z, p1.b
; CHECK-NEXT: ptest p1, p0.b
; CHECK-NEXT: cset w0, ne
; CHECK-NEXT: ret
%1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.brka.z.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a)
%2 = tail call i1 @llvm.aarch64.sve.ptest.any.nxv16i1(<vscale x 16 x i1> %a, <vscale x 16 x i1> %1)
%conv = zext i1 %2 to i32
ret i32 %conv
}
define i32 @brkb_neg(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a) { define i32 @brkb_neg(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a) {
; CHECK-LABEL: brkb_neg: ; CHECK-LABEL: brkb_neg:
; CHECK: // %bb.0: ; CHECK: // %bb.0:
; CHECK-NEXT: brkb p0.b, p0/z, p1.b ; CHECK-NEXT: brkb p0.b, p0/z, p1.b
; CHECK-NEXT: ptest p1, p0.b ; CHECK-NEXT: ptest p1, p0.b
; CHECK-NEXT: cset w0, ne ; CHECK-NEXT: cset w0, ne
; CHECK-NEXT: ret ; CHECK-NEXT: ret
%1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.brkb.z.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a) %1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.brkb.z.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a)
Show All 10 Lines
; CHECK-NEXT: cset w0, ne ; CHECK-NEXT: cset w0, ne
; CHECK-NEXT: ret ; CHECK-NEXT: ret
%1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.brkn.z.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a, <vscale x 16 x i1> %b) %1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.brkn.z.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a, <vscale x 16 x i1> %b)
%2 = tail call i1 @llvm.aarch64.sve.ptest.any.nxv16i1(<vscale x 16 x i1> %a, <vscale x 16 x i1> %1) %2 = tail call i1 @llvm.aarch64.sve.ptest.any.nxv16i1(<vscale x 16 x i1> %a, <vscale x 16 x i1> %1)
%conv = zext i1 %2 to i32 %conv = zext i1 %2 to i32
ret i32 %conv ret i32 %conv
} }
declare <vscale x 16 x i1> @llvm.aarch64.sve.brkpa.z.nxv16i1(<vscale x 16 x i1>, <vscale x 16 x i1>, <vscale x 16 x i1>)
declare <vscale x 16 x i1> @llvm.aarch64.sve.brkpb.z.nxv16i1(<vscale x 16 x i1>, <vscale x 16 x i1>, <vscale x 16 x i1>) declare <vscale x 16 x i1> @llvm.aarch64.sve.brkpb.z.nxv16i1(<vscale x 16 x i1>, <vscale x 16 x i1>, <vscale x 16 x i1>)
declare <vscale x 16 x i1> @llvm.aarch64.sve.brka.z.nxv16i1(<vscale x 16 x i1>, <vscale x 16 x i1>)
declare <vscale x 16 x i1> @llvm.aarch64.sve.brkb.z.nxv16i1(<vscale x 16 x i1>, <vscale x 16 x i1>) declare <vscale x 16 x i1> @llvm.aarch64.sve.brkb.z.nxv16i1(<vscale x 16 x i1>, <vscale x 16 x i1>)
declare <vscale x 16 x i1> @llvm.aarch64.sve.brkn.z.nxv16i1(<vscale x 16 x i1>, <vscale x 16 x i1>, <vscale x 16 x i1>) declare <vscale x 16 x i1> @llvm.aarch64.sve.brkn.z.nxv16i1(<vscale x 16 x i1>, <vscale x 16 x i1>, <vscale x 16 x i1>)
declare i1 @llvm.aarch64.sve.ptest.any.nxv16i1(<vscale x 16 x i1>, <vscale x 16 x i1>) declare i1 @llvm.aarch64.sve.ptest.any.nxv16i1(<vscale x 16 x i1>, <vscale x 16 x i1>)
declare <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32) declare <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32)

llvm/test/CodeGen/AArch64/sve-ptest-removal-log.ll

  • This file was added.
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=aarch64--linux-gnu -mattr=+sve %s -o - | FileCheck %s
; Test flag-setting bitwise logical instructions are used and redundant ptest
; instruction is removed.
define i1 @and(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a, <vscale x 16 x i1> %b) {
; CHECK-LABEL: and:
; CHECK: // %bb.0:
; CHECK-NEXT: and p1.b, p0/z, p1.b, p2.b
; CHECK-NEXT: ptest p0, p1.b
; CHECK-NEXT: cset w0, ne
; CHECK-NEXT: ret
%1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.and.z.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a, <vscale x 16 x i1> %b)
%2 = tail call i1 @llvm.aarch64.sve.ptest.any.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %1)
ret i1 %2
}
define i1 @bic(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a, <vscale x 16 x i1> %b) {
; CHECK-LABEL: bic:
; CHECK: // %bb.0:
; CHECK-NEXT: bic p1.b, p0/z, p1.b, p2.b
; CHECK-NEXT: ptest p0, p1.b
; CHECK-NEXT: cset w0, ne
; CHECK-NEXT: ret
%1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.bic.z.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a, <vscale x 16 x i1> %b)
%2 = tail call i1 @llvm.aarch64.sve.ptest.any.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %1)
ret i1 %2
}
define i1 @eor(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a, <vscale x 16 x i1> %b) {
; CHECK-LABEL: eor:
; CHECK: // %bb.0:
; CHECK-NEXT: eor p1.b, p0/z, p1.b, p2.b
; CHECK-NEXT: ptest p0, p1.b
; CHECK-NEXT: cset w0, ne
; CHECK-NEXT: ret
%1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.eor.z.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a, <vscale x 16 x i1> %b)
%2 = tail call i1 @llvm.aarch64.sve.ptest.any.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %1)
ret i1 %2
}
define i1 @nand(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a, <vscale x 16 x i1> %b) {
; CHECK-LABEL: nand:
; CHECK: // %bb.0:
; CHECK-NEXT: nand p1.b, p0/z, p1.b, p2.b
; CHECK-NEXT: ptest p0, p1.b
; CHECK-NEXT: cset w0, ne
; CHECK-NEXT: ret
%1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.nand.z.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a, <vscale x 16 x i1> %b)
%2 = tail call i1 @llvm.aarch64.sve.ptest.any.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %1)
ret i1 %2
}
define i1 @nor(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a, <vscale x 16 x i1> %b) {
; CHECK-LABEL: nor:
; CHECK: // %bb.0:
; CHECK-NEXT: nor p1.b, p0/z, p1.b, p2.b
; CHECK-NEXT: ptest p0, p1.b
; CHECK-NEXT: cset w0, ne
; CHECK-NEXT: ret
%1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.nor.z.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a, <vscale x 16 x i1> %b)
%2 = tail call i1 @llvm.aarch64.sve.ptest.any.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %1)
ret i1 %2
}
define i1 @orn(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a, <vscale x 16 x i1> %b) {
; CHECK-LABEL: orn:
; CHECK: // %bb.0:
; CHECK-NEXT: orn p1.b, p0/z, p1.b, p2.b
; CHECK-NEXT: ptest p0, p1.b
; CHECK-NEXT: cset w0, ne
; CHECK-NEXT: ret
%1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.orn.z.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a, <vscale x 16 x i1> %b)
%2 = tail call i1 @llvm.aarch64.sve.ptest.any.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %1)
ret i1 %2
}
define i1 @orr(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a, <vscale x 16 x i1> %b) {
; CHECK-LABEL: orr:
; CHECK: // %bb.0:
; CHECK-NEXT: orr p1.b, p0/z, p1.b, p2.b
; CHECK-NEXT: ptest p0, p1.b
; CHECK-NEXT: cset w0, ne
; CHECK-NEXT: ret
%1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.orr.z.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %a, <vscale x 16 x i1> %b)
%2 = tail call i1 @llvm.aarch64.sve.ptest.any.nxv16i1(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %1)
ret i1 %2
}
declare <vscale x 16 x i1> @llvm.aarch64.sve.and.z.nxv16i1(<vscale x 16 x i1>, <vscale x 16 x i1>, <vscale x 16 x i1>)
declare <vscale x 16 x i1> @llvm.aarch64.sve.bic.z.nxv16i1(<vscale x 16 x i1>, <vscale x 16 x i1>, <vscale x 16 x i1>)
declare <vscale x 16 x i1> @llvm.aarch64.sve.eor.z.nxv16i1(<vscale x 16 x i1>, <vscale x 16 x i1>, <vscale x 16 x i1>)
declare <vscale x 16 x i1> @llvm.aarch64.sve.nand.z.nxv16i1(<vscale x 16 x i1>, <vscale x 16 x i1>, <vscale x 16 x i1>)
declare <vscale x 16 x i1> @llvm.aarch64.sve.nor.z.nxv16i1(<vscale x 16 x i1>, <vscale x 16 x i1>, <vscale x 16 x i1>)
declare <vscale x 16 x i1> @llvm.aarch64.sve.orn.z.nxv16i1(<vscale x 16 x i1>, <vscale x 16 x i1>, <vscale x 16 x i1>)
declare <vscale x 16 x i1> @llvm.aarch64.sve.orr.z.nxv16i1(<vscale x 16 x i1>, <vscale x 16 x i1>, <vscale x 16 x i1>)
declare i1 @llvm.aarch64.sve.ptest.any.nxv16i1(<vscale x 16 x i1>, <vscale x 16 x i1>)

llvm/test/CodeGen/AArch64/sve-ptest-removal-ptrue.ll

  • This file was added.
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=aarch64--linux-gnu -mattr=+sve %s -o - | FileCheck %s
define i1 @ptrue() {
; CHECK-LABEL: ptrue:
; CHECK: // %bb.0:
; CHECK-NEXT: ptrue p0.b, pow2
; CHECK-NEXT: ptest p0, p0.b
; CHECK-NEXT: cset w0, ne
; CHECK-NEXT: ret
%1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32 0)
%2 = tail call i1 @llvm.aarch64.sve.ptest.any.nxv16i1(<vscale x 16 x i1> %1, <vscale x 16 x i1> %1)
ret i1 %2
}
declare i1 @llvm.aarch64.sve.ptest.any.nxv16i1(<vscale x 16 x i1>, <vscale x 16 x i1>)
declare <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32 %pattern)