[PowerPC] hoist xxspltib out of loop body #127121
Merged
Conversation
This file contains hidden or 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
Member
|
@llvm/pr-subscribers-backend-powerpc Author: zhijian lin (diggerlin) Changesthe patch fix the issue #127119 Full diff: https://github.com/llvm/llvm-project/pull/127121.diff 3 Files Affected:
diff --git a/llvm/lib/Target/PowerPC/PPCInstrVSX.td b/llvm/lib/Target/PowerPC/PPCInstrVSX.td
index 8e400bc63b785..d9e88b283a749 100644
--- a/llvm/lib/Target/PowerPC/PPCInstrVSX.td
+++ b/llvm/lib/Target/PowerPC/PPCInstrVSX.td
@@ -1659,8 +1659,6 @@ let Predicates = [HasVSX, HasP9Vector] in {
RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">;
// Vector Splat Immediate Byte
- // FIXME: Setting the hasSideEffects flag here to match current behaviour.
- let hasSideEffects = 1 in
def XXSPLTIB : X_RD6_IMM8<60, 360, (outs vsrc:$XT), (ins u8imm:$IMM8),
"xxspltib $XT, $IMM8", IIC_VecPerm, []>;
diff --git a/llvm/test/CodeGen/PowerPC/memset-tail.ll b/llvm/test/CodeGen/PowerPC/memset-tail.ll
index 31c136d009ba5..ec9929e0d90dd 100644
--- a/llvm/test/CodeGen/PowerPC/memset-tail.ll
+++ b/llvm/test/CodeGen/PowerPC/memset-tail.ll
@@ -380,9 +380,9 @@ define dso_local void @memsetTailV1B2(ptr nocapture noundef writeonly %p) local_
;
; P9-BE-LABEL: memsetTailV1B2:
; P9-BE: # %bb.0: # %entry
+; P9-BE-NEXT: xxspltib 0, 15
; P9-BE-NEXT: li 4, 3855
; P9-BE-NEXT: sth 4, 16(3)
-; P9-BE-NEXT: xxspltib 0, 15
; P9-BE-NEXT: stxv 0, 0(3)
; P9-BE-NEXT: blr
;
@@ -433,9 +433,9 @@ define dso_local void @memsetTailV1B1(ptr nocapture noundef writeonly %p) local_
;
; P9-BE-LABEL: memsetTailV1B1:
; P9-BE: # %bb.0: # %entry
+; P9-BE-NEXT: xxspltib 0, 15
; P9-BE-NEXT: li 4, 15
; P9-BE-NEXT: stb 4, 16(3)
-; P9-BE-NEXT: xxspltib 0, 15
; P9-BE-NEXT: stxv 0, 0(3)
; P9-BE-NEXT: blr
;
@@ -861,9 +861,9 @@ define dso_local void @memset2TailV1B2(ptr nocapture noundef writeonly %p) local
;
; P9-BE-LABEL: memset2TailV1B2:
; P9-BE: # %bb.0: # %entry
+; P9-BE-NEXT: xxspltib 0, 165
; P9-BE-NEXT: li 4, -23131
; P9-BE-NEXT: sth 4, 16(3)
-; P9-BE-NEXT: xxspltib 0, 165
; P9-BE-NEXT: stxv 0, 0(3)
; P9-BE-NEXT: blr
;
@@ -917,9 +917,9 @@ define dso_local void @memset2TailV1B1(ptr nocapture noundef writeonly %p) local
;
; P9-BE-LABEL: memset2TailV1B1:
; P9-BE: # %bb.0: # %entry
+; P9-BE-NEXT: xxspltib 0, 165
; P9-BE-NEXT: li 4, -91
; P9-BE-NEXT: stb 4, 16(3)
-; P9-BE-NEXT: xxspltib 0, 165
; P9-BE-NEXT: stxv 0, 0(3)
; P9-BE-NEXT: blr
;
diff --git a/llvm/test/CodeGen/PowerPC/pr127119.ll b/llvm/test/CodeGen/PowerPC/pr127119.ll
new file mode 100644
index 0000000000000..2856433777712
--- /dev/null
+++ b/llvm/test/CodeGen/PowerPC/pr127119.ll
@@ -0,0 +1,95 @@
+;; Test `xxspltib` hoist out loop.
+
+; RUN: llc -verify-machineinstrs -mtriple powerpc64-ibm-aix-xcoff \
+; RUN: %s -o - 2>&1 | FileCheck %s
+
+define void @_Z3fooPfS_Pi(ptr noalias nocapture noundef writeonly %_a, ptr noalias nocapture noundef readonly %In_a, ptr noalias nocapture noundef readonly %n) local_unnamed_addr #0 {
+entry:
+ %0 = load i32, ptr %n, align 4
+ %cmp9 = icmp sgt i32 %0, 0
+ br i1 %cmp9, label %for.body.preheader, label %for.cond.cleanup
+
+for.body.preheader:
+ %wide.trip.count = zext nneg i32 %0 to i64
+ %xtraiter = and i64 %wide.trip.count, 1
+ %1 = icmp eq i32 %0, 1
+ br i1 %1, label %for.cond.cleanup.loopexit.unr-lcssa, label %for.body.preheader.new
+
+for.body.preheader.new:
+ %unroll_iter = and i64 %wide.trip.count, 2147483646
+ br label %for.body
+
+for.cond.cleanup.loopexit.unr-lcssa:
+ %indvars.iv.unr = phi i64 [ 0, %for.body.preheader ], [ %indvars.iv.next.1, %for.body ]
+ %lcmp.mod.not = icmp eq i64 %xtraiter, 0
+ br i1 %lcmp.mod.not, label %for.cond.cleanup, label %for.body.epil
+
+for.body.epil:
+ %arrayidx.epil = getelementptr inbounds nuw float, ptr %In_a, i64 %indvars.iv.unr
+ %2 = load float, ptr %arrayidx.epil, align 4
+ %vecins.i.epil = insertelement <4 x float> poison, float %2, i64 0
+ %3 = bitcast <4 x float> %vecins.i.epil to <16 x i8>
+ %and1.i.epil = and <16 x i8> %3, <i8 6, i8 6, i8 6, i8 6, i8 poison, i8 poison, i8 poison, i8 poison, i8 poison, i8 poison, i8 poison, i8 poison, i8 poison, i8 poison, i8 poison, i8 poison>
+ %4 = bitcast <16 x i8> %and1.i.epil to <4 x float>
+ %vecext.i.epil = extractelement <4 x float> %4, i64 0
+ %arrayidx5.epil = getelementptr inbounds nuw float, ptr %_a, i64 %indvars.iv.unr
+ store float %vecext.i.epil, ptr %arrayidx5.epil, align 4
+ br label %for.cond.cleanup
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %indvars.iv = phi i64 [ 0, %for.body.preheader.new ], [ %indvars.iv.next.1, %for.body ]
+ %niter = phi i64 [ 0, %for.body.preheader.new ], [ %niter.next.1, %for.body ]
+ %arrayidx = getelementptr inbounds nuw float, ptr %In_a, i64 %indvars.iv
+ %5 = load float, ptr %arrayidx, align 4
+ %vecins.i = insertelement <4 x float> poison, float %5, i64 0
+ %6 = bitcast <4 x float> %vecins.i to <16 x i8>
+ %and1.i = and <16 x i8> %6, <i8 6, i8 6, i8 6, i8 6, i8 poison, i8 poison, i8 poison, i8 poison, i8 poison, i8 poison, i8 poison, i8 poison, i8 poison, i8 poison, i8 poison, i8 poison>
+ %7 = bitcast <16 x i8> %and1.i to <4 x float>
+ %vecext.i = extractelement <4 x float> %7, i64 0
+ %arrayidx5 = getelementptr inbounds nuw float, ptr %_a, i64 %indvars.iv
+ store float %vecext.i, ptr %arrayidx5, align 4
+ %indvars.iv.next = or disjoint i64 %indvars.iv, 1
+ %arrayidx.1 = getelementptr inbounds nuw float, ptr %In_a, i64 %indvars.iv.next
+ %8 = load float, ptr %arrayidx.1, align 4
+ %vecins.i.1 = insertelement <4 x float> poison, float %8, i64 0
+ %9 = bitcast <4 x float> %vecins.i.1 to <16 x i8>
+ %and1.i.1 = and <16 x i8> %9, <i8 6, i8 6, i8 6, i8 6, i8 poison, i8 poison, i8 poison, i8 poison, i8 poison, i8 poison, i8 poison, i8 poison, i8 poison, i8 poison, i8 poison, i8 poison>
+ %10 = bitcast <16 x i8> %and1.i.1 to <4 x float>
+ %vecext.i.1 = extractelement <4 x float> %10, i64 0
+ %arrayidx5.1 = getelementptr inbounds nuw float, ptr %_a, i64 %indvars.iv.next
+ store float %vecext.i.1, ptr %arrayidx5.1, align 4
+ %indvars.iv.next.1 = add nuw nsw i64 %indvars.iv, 2
+ %niter.next.1 = add i64 %niter, 2
+ %niter.ncmp.1 = icmp eq i64 %niter.next.1, %unroll_iter
+ br i1 %niter.ncmp.1, label %for.cond.cleanup.loopexit.unr-lcssa, label %for.body
+}
+
+attributes #0 = { mustprogress nofree norecurse nosync nounwind memory(argmem: readwrite) "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="pwr10" "target-features"="+altivec,+bpermd,+crbits,+crypto,+direct-move,+extdiv,+isa-v206-instructions,+isa-v207-instructions,+isa-v30-instructions,+isa-v31-instructions,+mma,+paired-vector-memops,+pcrelative-memops,+power10-vector,+power8-vector,+power9-vector,+prefix-instrs,+quadword-atomics,+vsx,-aix-shared-lib-tls-model-opt,-aix-small-local-dynamic-tls,-aix-small-local-exec-tls,-htm,-privileged,-rop-protect,-spe" }
+
+; CHECK: ._Z3fooPfS_Pi:
+; CHECK-NEXT: # %bb.0: # %entry
+; CHECK-NEXT: lwz 5, 0(5)
+; CHECK-NEXT: cmpwi 5, 1
+; CHECK-NEXT: bltlr 0
+; CHECK-NEXT: # %bb.1: # %for.body.preheader
+; CHECK-NEXT: li 6, 0
+; CHECK-NEXT: cmplwi 5, 1
+; CHECK-NEXT: beq 0, L..BB0_4
+; CHECK-NEXT: # %bb.2: # %for.body.preheader.new
+; CHECK-NEXT: rlwinm 6, 5, 0, 1, 30
+; CHECK-NEXT: xxspltib 0, 6
+; CHECK-NEXT: addi 9, 4, -8
+; CHECK-NEXT: addi 7, 3, -8
+; CHECK-NEXT: li 8, 8
+; CHECK-NEXT: li 10, 12
+; CHECK-NEXT: li 11, 4
+; CHECK-NEXT: addi 6, 6, -2
+; CHECK-NEXT: rldicl 6, 6, 63, 1
+; CHECK-NEXT: addi 6, 6, 1
+; CHECK-NEXT: mtctr 6
+; CHECK-NEXT: li 6, 0
+; CHECK-NEXT: .align 4
+; CHECK-NEXT: L..BB0_3:
|
lei137
reviewed
Feb 14, 2025
Contributor
lei137
left a comment
lei137
left a comment
There was a problem hiding this comment.
In general this looks good to me. However, would be good if you can do a NFC pre-commit of the new test added, so this patch can show the changes.
diggerlin
changed the title
Contributor
Author
I created a separated NFC patch for the new test case added. #127701 |
llvm-sync bot
pushed a commit
to arm/arm-toolchain
that referenced
this pull request
Feb 21, 2025
…ists xxsplitib out of loop (#127701) This is a pre-commit test case for patch llvm/llvm-project#127121 that hoists xxsplitib out of loop
diggerlin
force-pushed
the
digger/127119
branch
from
ad69826 to
aedecc5
Compare
diggerlin
changed the title
Contributor
Author
|
rebased the code based on the pre-commit 481e1eb |
lei137
changed the title
lei137
approved these changes
Feb 28, 2025
Contributor
lei137
left a comment
lei137
left a comment
There was a problem hiding this comment.
LGTM.
I made a small update to your subject title and descripton body. Hope that is okay.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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.
Fixes #127119
Remove
hasSideEffectsfromxxspltibsince there is no special restriction specified in the ISA that prevent it from being reordered, move, CSE, or LICM. Removing this restriction will allowxxspltibto be hoisted out of loop bodies.