Skip to content

Commit 20d798b

Browse files
fhahnfhahn
committed
Recommit "[SCEV] Look through single value PHIs." (take 3)
This reverts commit 1fbdbb5. All known issues surfaced by this patch should have been fixed now. The fixes included fixing issues with SCEV expansion in LV and DA's reliance on LCSSA phis.
1 parent dd38caf commit 20d798b

File tree

9 files changed

+55
-67
lines changed

9 files changed

+55
-67
lines changed

‎llvm/lib/Analysis/ScalarEvolution.cpp

Lines changed: 1 addition & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -5927,13 +5927,8 @@ const SCEV *ScalarEvolution::createNodeForPHI(PHINode *PN) {
59275927
if (const SCEV *S = createNodeFromSelectLikePHI(PN))
59285928
return S;
59295929

5930-
// If the PHI has a single incoming value, follow that value, unless the
5931-
// PHI's incoming blocks are in a different loop, in which case doing so
5932-
// risks breaking LCSSA form. Instcombine would normally zap these, but
5933-
// it doesn't have DominatorTree information, so it may miss cases.
59345930
if (Value *V = simplifyInstruction(PN, {getDataLayout(), &TLI, &DT, &AC}))
5935-
if (LI.replacementPreservesLCSSAForm(PN, V))
5936-
return getSCEV(V);
5931+
return getSCEV(V);
59375932

59385933
// If it's not a loop phi, we can't handle it yet.
59395934
return getUnknown(PN);

‎llvm/test/Analysis/DependenceAnalysis/lcssa.ll

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -2,7 +2,7 @@
22
; RUN: "-aa-pipeline=basic-aa,tbaa" 2>&1 | FileCheck %s
33

44
; CHECK: Src: %v = load i32, i32* %arrayidx1, align 4 --> Dst: store i32 %add, i32* %a.lcssa, align 4
5-
; CHECK-NEXT: da analyze - confused!
5+
; CHECK-NEXT: da analyze - anti [*|<]!
66

77
define void @f(i32 *%a, i32 %n, i64 %n2) {
88
entry:

‎llvm/test/Analysis/ScalarEvolution/cycled_phis.ll

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -46,7 +46,7 @@ define void @test_02(i32* %p, i32* %q) {
4646
; CHECK-NEXT: %inner_cond = call i1 @cond()
4747
; CHECK-NEXT: --> %inner_cond U: full-set S: full-set Exits: <<Unknown>> LoopDispositions: { %inner_loop: Variant, %outer_loop: Variant }
4848
; CHECK-NEXT: %inner_lcssa = phi i32 [ %inner_phi, %inner_loop ]
49-
; CHECK-NEXT: --> %inner_lcssa U: full-set S: full-set Exits: <<Unknown>> LoopDispositions: { %outer_loop: Variant, %inner_loop: Invariant }
49+
; CHECK-NEXT: --> %inner_phi U: full-set S: full-set Exits: <<Unknown>> LoopDispositions: { %outer_loop: Variant, %inner_loop: Variant }
5050
; CHECK-NEXT: %outer_cond = call i1 @cond()
5151
; CHECK-NEXT: --> %outer_cond U: full-set S: full-set Exits: <<Unknown>> LoopDispositions: { %outer_loop: Variant, %inner_loop: Invariant }
5252
; CHECK-NEXT: Determining loop execution counts for: @test_02
@@ -97,7 +97,7 @@ define void @test_03(i32* %p, i32* %q) {
9797
; CHECK-NEXT: %inner_cond = call i1 @cond()
9898
; CHECK-NEXT: --> %inner_cond U: full-set S: full-set Exits: <<Unknown>> LoopDispositions: { %inner_loop: Variant, %outer_loop: Variant }
9999
; CHECK-NEXT: %inner_lcssa = phi i32 [ %inner_phi_1, %inner_loop ]
100-
; CHECK-NEXT: --> %inner_lcssa U: full-set S: full-set Exits: <<Unknown>> LoopDispositions: { %outer_loop: Variant, %inner_loop: Invariant }
100+
; CHECK-NEXT: --> %inner_phi_1 U: full-set S: full-set Exits: <<Unknown>> LoopDispositions: { %outer_loop: Variant, %inner_loop: Variant }
101101
; CHECK-NEXT: %outer_cond = call i1 @cond()
102102
; CHECK-NEXT: --> %outer_cond U: full-set S: full-set Exits: <<Unknown>> LoopDispositions: { %outer_loop: Variant, %inner_loop: Invariant }
103103
; CHECK-NEXT: Determining loop execution counts for: @test_03

‎llvm/test/Analysis/ScalarEvolution/incorrect-exit-count.ll

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -53,7 +53,7 @@ define dso_local i32 @f() {
5353
; CHECK-NEXT: %dec.3 = add nsw i32 %storemerge1921.3, -1
5454
; CHECK-NEXT: --> {2,+,-1}<nsw><%inner.loop> U: [2,3) S: [2,3) Exits: <<Unknown>> LoopDispositions: { %inner.loop: Computable, %outer.loop: Variant }
5555
; CHECK-NEXT: %storemerge1921.lcssa25.3 = phi i32 [ %storemerge1921.3, %for.end.3 ]
56-
; CHECK-NEXT: --> %storemerge1921.lcssa25.3 U: [3,4) S: [3,4) Exits: <<Unknown>> LoopDispositions: { %outer.loop: Variant, %for.cond6: Invariant, %inner.loop: Invariant }
56+
; CHECK-NEXT: --> {3,+,-1}<nuw><nsw><%inner.loop> U: [3,4) S: [3,4) Exits: <<Unknown>> LoopDispositions: { %outer.loop: Variant, %for.cond6: Variant, %inner.loop: Computable }
5757
; CHECK-NEXT: %dec16 = add nsw i32 %storemerge23, -1
5858
; CHECK-NEXT: --> {2,+,-1}<nsw><%outer.loop> U: [0,3) S: [0,3) Exits: <<Unknown>> LoopDispositions: { %outer.loop: Computable, %for.cond6: Invariant, %inner.loop: Invariant }
5959
; CHECK-NEXT: Determining loop execution counts for: @f

‎llvm/test/Analysis/ScalarEvolution/solve-quadratic-i1.ll

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -59,9 +59,9 @@ define void @f1() #0 {
5959
; CHECK-NEXT: %v6 = add nuw nsw i32 %v1, 1
6060
; CHECK-NEXT: --> {4,+,1}<nuw><nsw><%b1> U: [4,7) S: [4,7) Exits: 6 LoopDispositions: { %b1: Computable }
6161
; CHECK-NEXT: %v7 = phi i32 [ %v1, %b1 ]
62-
; CHECK-NEXT: --> %v7 U: [3,6) S: [3,6) --> 5 U: [5,6) S: [5,6)
62+
; CHECK-NEXT: --> {3,+,1}<nuw><nsw><%b1> U: [3,6) S: [3,6) --> 5 U: [5,6) S: [5,6)
6363
; CHECK-NEXT: %v8 = phi i16 [ %v3, %b1 ]
64-
; CHECK-NEXT: --> %v8 U: full-set S: full-set --> 12 U: [12,13) S: [12,13)
64+
; CHECK-NEXT: --> {3,+,4,+,1}<%b1> U: full-set S: full-set --> 12 U: [12,13) S: [12,13)
6565
; CHECK-NEXT: Determining loop execution counts for: @f1
6666
; CHECK-NEXT: Loop %b3: <multiple exits> Unpredictable backedge-taken count.
6767
; CHECK-NEXT: Loop %b3: Unpredictable max backedge-taken count.

‎llvm/test/Analysis/ScalarEvolution/solve-quadratic-overflow.ll

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -12,11 +12,11 @@
1212
; CHECK-NEXT: %v3 = mul i16 %v2, %v2
1313
; CHECK-NEXT: --> {1,+,3,+,2}<%b1> U: full-set S: full-set Exits: 0 LoopDispositions: { %b1: Computable }
1414
; CHECK-NEXT: %v5 = phi i16 [ %v2, %b1 ]
15-
; CHECK-NEXT: --> %v5 U: [-256,0) S: [-256,0)
15+
; CHECK-NEXT: --> {-1,+,-1}<%b1> U: [-256,0) S: [-256,0) --> -256 U: [-256,-255) S: [-256,-255)
1616
; CHECK-NEXT: %v6 = phi i16 [ %v3, %b1 ]
17-
; CHECK-NEXT: --> %v6 U: full-set S: full-set
17+
; CHECK-NEXT: --> {1,+,3,+,2}<%b1> U: full-set S: full-set --> 0 U: [0,1) S: [0,1)
1818
; CHECK-NEXT: %v7 = sext i16 %v5 to i32
19-
; CHECK-NEXT: --> (sext i16 %v5 to i32) U: [-256,0) S: [-256,0)
19+
; CHECK-NEXT: --> {-1,+,-1}<nsw><%b1> U: [-256,0) S: [-256,0) --> -256 U: [-256,-255) S: [-256,-255)
2020
; CHECK-NEXT: Determining loop execution counts for: @f0
2121
; CHECK-NEXT: Loop %b1: backedge-taken count is 255
2222
; CHECK-NEXT: Loop %b1: max backedge-taken count is 255

‎llvm/test/Analysis/ScalarEvolution/trivial-phis.ll

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -2,7 +2,7 @@
22

33
; CHECK-LABEL: @test1
44
; CHECK: %add.lcssa.wide = phi i64 [ %indvars.iv.next, %do.body ]
5-
; CHECK-NEXT: --> %add.lcssa.wide U: [1,2147483648) S: [1,2147483648)
5+
; CHECK-NEXT: --> {1,+,1}<nuw><nsw><%do.body> U: [1,2147483648) S: [1,2147483648)
66

77
define i64 @test1(i32 signext %n, float* %A) {
88
entry:

‎llvm/test/Transforms/LoopStrengthReduce/funclet.ll

Lines changed: 16 additions & 24 deletions
Original file line numberDiff line numberDiff line change
@@ -15,23 +15,21 @@ define void @f() personality i32 (...)* @_except_handler3 {
1515
; CHECK-NEXT: entry:
1616
; CHECK-NEXT: br label [[THROW:%.*]]
1717
; CHECK: throw:
18-
; CHECK-NEXT: [[TMP96:%.*]] = getelementptr inbounds i8, i8* undef, i32 1
1918
; CHECK-NEXT: invoke void @reserve()
2019
; CHECK-NEXT: to label [[THROW]] unwind label [[PAD:%.*]]
2120
; CHECK: pad:
22-
; CHECK-NEXT: [[PHI2:%.*]] = phi i8* [ [[TMP96]], [[THROW]] ]
2321
; CHECK-NEXT: [[CS:%.*]] = catchswitch within none [label %unreachable] unwind label [[BLAH2:%.*]]
2422
; CHECK: unreachable:
2523
; CHECK-NEXT: [[TMP0:%.*]] = catchpad within [[CS]] []
2624
; CHECK-NEXT: unreachable
2725
; CHECK: blah2:
2826
; CHECK-NEXT: [[CLEANUPPADI4_I_I_I:%.*]] = cleanuppad within none []
29-
; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, i8* [[PHI2]], i32 -1
3027
; CHECK-NEXT: br label [[LOOP_BODY:%.*]]
3128
; CHECK: loop_body:
32-
; CHECK-NEXT: [[LSR_IV:%.*]] = phi i8* [ [[SCEVGEP1:%.*]], [[ITER:%.*]] ], [ [[SCEVGEP]], [[BLAH2]] ]
33-
; CHECK-NEXT: [[SCEVGEP1]] = getelementptr i8, i8* [[LSR_IV]], i32 1
34-
; CHECK-NEXT: [[TMP100:%.*]] = icmp eq i8* [[SCEVGEP1]], undef
29+
; CHECK-NEXT: [[LSR_IV:%.*]] = phi i32 [ [[LSR_IV_NEXT:%.*]], [[ITER:%.*]] ], [ 0, [[BLAH2]] ]
30+
; CHECK-NEXT: [[LSR_IV_NEXT]] = add nuw nsw i32 [[LSR_IV]], -1
31+
; CHECK-NEXT: [[LSR_IV_NEXT1:%.*]] = inttoptr i32 [[LSR_IV_NEXT]] to i8*
32+
; CHECK-NEXT: [[TMP100:%.*]] = icmp eq i8* [[LSR_IV_NEXT1]], null
3533
; CHECK-NEXT: br i1 [[TMP100]], label [[UNWIND_OUT:%.*]], label [[ITER]]
3634
; CHECK: iter:
3735
; CHECK-NEXT: br i1 true, label [[UNWIND_OUT]], label [[LOOP_BODY]]
@@ -76,27 +74,25 @@ define void @g() personality i32 (...)* @_except_handler3 {
7674
; CHECK-NEXT: entry:
7775
; CHECK-NEXT: br label [[THROW:%.*]]
7876
; CHECK: throw:
79-
; CHECK-NEXT: [[TMP96:%.*]] = getelementptr inbounds i8, i8* undef, i32 1
8077
; CHECK-NEXT: invoke void @reserve()
8178
; CHECK-NEXT: to label [[THROW]] unwind label [[PAD:%.*]]
8279
; CHECK: pad:
83-
; CHECK-NEXT: [[PHI2:%.*]] = phi i8* [ [[TMP96]], [[THROW]] ]
8480
; CHECK-NEXT: [[CS:%.*]] = catchswitch within none [label [[UNREACHABLE:%.*]], label %blah] unwind to caller
8581
; CHECK: unreachable:
8682
; CHECK-NEXT: [[TMP0:%.*]] = catchpad within [[CS]] []
8783
; CHECK-NEXT: unreachable
8884
; CHECK: blah:
8985
; CHECK-NEXT: [[CATCHPAD:%.*]] = catchpad within [[CS]] []
90-
; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, i8* [[PHI2]], i32 -1
9186
; CHECK-NEXT: br label [[LOOP_BODY:%.*]]
9287
; CHECK: unwind_out:
9388
; CHECK-NEXT: catchret from [[CATCHPAD]] to label [[LEAVE:%.*]]
9489
; CHECK: leave:
9590
; CHECK-NEXT: ret void
9691
; CHECK: loop_body:
97-
; CHECK-NEXT: [[LSR_IV:%.*]] = phi i8* [ [[SCEVGEP1:%.*]], [[ITER:%.*]] ], [ [[SCEVGEP]], [[BLAH:%.*]] ]
98-
; CHECK-NEXT: [[SCEVGEP1]] = getelementptr i8, i8* [[LSR_IV]], i32 1
99-
; CHECK-NEXT: [[TMP100:%.*]] = icmp eq i8* [[SCEVGEP1]], undef
92+
; CHECK-NEXT: [[LSR_IV:%.*]] = phi i32 [ [[LSR_IV_NEXT:%.*]], [[ITER:%.*]] ], [ 0, [[BLAH:%.*]] ]
93+
; CHECK-NEXT: [[LSR_IV_NEXT]] = add nuw nsw i32 [[LSR_IV]], -1
94+
; CHECK-NEXT: [[LSR_IV_NEXT1:%.*]] = inttoptr i32 [[LSR_IV_NEXT]] to i8*
95+
; CHECK-NEXT: [[TMP100:%.*]] = icmp eq i8* [[LSR_IV_NEXT1]], null
10096
; CHECK-NEXT: br i1 [[TMP100]], label [[UNWIND_OUT:%.*]], label [[ITER]]
10197
; CHECK: iter:
10298
; CHECK-NEXT: br i1 true, label [[UNWIND_OUT]], label [[LOOP_BODY]]
@@ -142,7 +138,6 @@ define void @h() personality i32 (...)* @_except_handler3 {
142138
; CHECK-NEXT: entry:
143139
; CHECK-NEXT: br label [[THROW:%.*]]
144140
; CHECK: throw:
145-
; CHECK-NEXT: [[TMP96:%.*]] = getelementptr inbounds i8, i8* undef, i32 1
146141
; CHECK-NEXT: invoke void @reserve()
147142
; CHECK-NEXT: to label [[THROW]] unwind label [[PAD:%.*]]
148143
; CHECK: pad:
@@ -151,18 +146,17 @@ define void @h() personality i32 (...)* @_except_handler3 {
151146
; CHECK-NEXT: [[TMP0:%.*]] = catchpad within [[CS]] []
152147
; CHECK-NEXT: unreachable
153148
; CHECK: blug:
154-
; CHECK-NEXT: [[PHI2:%.*]] = phi i8* [ [[TMP96]], [[PAD]] ]
155149
; CHECK-NEXT: [[CATCHPAD:%.*]] = catchpad within [[CS]] []
156-
; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, i8* [[PHI2]], i32 -1
157150
; CHECK-NEXT: br label [[LOOP_BODY:%.*]]
158151
; CHECK: unwind_out:
159152
; CHECK-NEXT: catchret from [[CATCHPAD]] to label [[LEAVE:%.*]]
160153
; CHECK: leave:
161154
; CHECK-NEXT: ret void
162155
; CHECK: loop_body:
163-
; CHECK-NEXT: [[LSR_IV:%.*]] = phi i8* [ [[SCEVGEP1:%.*]], [[ITER:%.*]] ], [ [[SCEVGEP]], [[BLUG:%.*]] ]
164-
; CHECK-NEXT: [[SCEVGEP1]] = getelementptr i8, i8* [[LSR_IV]], i32 1
165-
; CHECK-NEXT: [[TMP100:%.*]] = icmp eq i8* [[SCEVGEP1]], undef
156+
; CHECK-NEXT: [[LSR_IV:%.*]] = phi i32 [ [[LSR_IV_NEXT:%.*]], [[ITER:%.*]] ], [ 0, [[BLUG:%.*]] ]
157+
; CHECK-NEXT: [[LSR_IV_NEXT]] = add nuw nsw i32 [[LSR_IV]], -1
158+
; CHECK-NEXT: [[LSR_IV_NEXT1:%.*]] = inttoptr i32 [[LSR_IV_NEXT]] to i8*
159+
; CHECK-NEXT: [[TMP100:%.*]] = icmp eq i8* [[LSR_IV_NEXT1]], null
166160
; CHECK-NEXT: br i1 [[TMP100]], label [[UNWIND_OUT:%.*]], label [[ITER]]
167161
; CHECK: iter:
168162
; CHECK-NEXT: br i1 true, label [[UNWIND_OUT]], label [[LOOP_BODY]]
@@ -208,11 +202,9 @@ define void @i() personality i32 (...)* @_except_handler3 {
208202
; CHECK-NEXT: entry:
209203
; CHECK-NEXT: br label [[THROW:%.*]]
210204
; CHECK: throw:
211-
; CHECK-NEXT: [[TMP96:%.*]] = getelementptr inbounds i8, i8* undef, i32 1
212205
; CHECK-NEXT: invoke void @reserve()
213206
; CHECK-NEXT: to label [[THROW]] unwind label [[CATCHPAD:%.*]]
214207
; CHECK: catchpad:
215-
; CHECK-NEXT: [[PHI2:%.*]] = phi i8* [ [[TMP96]], [[THROW]] ]
216208
; CHECK-NEXT: [[CS:%.*]] = catchswitch within none [label %cp_body] unwind label [[CLEANUPPAD:%.*]]
217209
; CHECK: cp_body:
218210
; CHECK-NEXT: [[TMP0:%.*]] = catchpad within [[CS]] []
@@ -221,12 +213,12 @@ define void @i() personality i32 (...)* @_except_handler3 {
221213
; CHECK-NEXT: [[TMP1:%.*]] = cleanuppad within none []
222214
; CHECK-NEXT: br label [[LOOP_HEAD]]
223215
; CHECK: loop_head:
224-
; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, i8* [[PHI2]], i32 -1
225216
; CHECK-NEXT: br label [[LOOP_BODY:%.*]]
226217
; CHECK: loop_body:
227-
; CHECK-NEXT: [[LSR_IV:%.*]] = phi i8* [ [[SCEVGEP1:%.*]], [[ITER:%.*]] ], [ [[SCEVGEP]], [[LOOP_HEAD]] ]
228-
; CHECK-NEXT: [[SCEVGEP1]] = getelementptr i8, i8* [[LSR_IV]], i32 1
229-
; CHECK-NEXT: [[TMP100:%.*]] = icmp eq i8* [[SCEVGEP1]], undef
218+
; CHECK-NEXT: [[LSR_IV:%.*]] = phi i32 [ [[LSR_IV_NEXT:%.*]], [[ITER:%.*]] ], [ 0, [[LOOP_HEAD]] ]
219+
; CHECK-NEXT: [[LSR_IV_NEXT]] = add nuw nsw i32 [[LSR_IV]], -1
220+
; CHECK-NEXT: [[LSR_IV_NEXT1:%.*]] = inttoptr i32 [[LSR_IV_NEXT]] to i8*
221+
; CHECK-NEXT: [[TMP100:%.*]] = icmp eq i8* [[LSR_IV_NEXT1]], null
230222
; CHECK-NEXT: br i1 [[TMP100]], label [[UNWIND_OUT:%.*]], label [[ITER]]
231223
; CHECK: iter:
232224
; CHECK-NEXT: br i1 true, label [[UNWIND_OUT]], label [[LOOP_BODY]]

‎llvm/test/Transforms/LoopVectorize/pr45259.ll

Lines changed: 28 additions & 27 deletions
Original file line numberDiff line numberDiff line change
@@ -6,53 +6,54 @@
66
define i8 @widget(i8* %arr, i8 %t9) {
77
; CHECK-LABEL: @widget(
88
; CHECK-NEXT: bb:
9-
; CHECK-NEXT: [[ARR2:%.*]] = ptrtoint i8* [[ARR:%.*]] to i64
9+
; CHECK-NEXT: [[ARR1:%.*]] = ptrtoint i8* [[ARR:%.*]] to i64
1010
; CHECK-NEXT: br label [[BB6:%.*]]
1111
; CHECK: bb6:
1212
; CHECK-NEXT: [[T1_0:%.*]] = phi i8* [ [[ARR]], [[BB:%.*]] ], [ null, [[BB6]] ]
1313
; CHECK-NEXT: [[C:%.*]] = call i1 @cond()
1414
; CHECK-NEXT: br i1 [[C]], label [[FOR_PREHEADER:%.*]], label [[BB6]]
1515
; CHECK: for.preheader:
1616
; CHECK-NEXT: [[T1_0_LCSSA:%.*]] = phi i8* [ [[T1_0]], [[BB6]] ]
17-
; CHECK-NEXT: [[T1_0_LCSSA1:%.*]] = ptrtoint i8* [[T1_0_LCSSA]] to i64
18-
; CHECK-NEXT: [[TMP0:%.*]] = trunc i64 [[T1_0_LCSSA1]] to i32
19-
; CHECK-NEXT: [[TMP1:%.*]] = trunc i64 [[ARR2]] to i32
20-
; CHECK-NEXT: [[TMP2:%.*]] = sub i32 [[TMP0]], [[TMP1]]
21-
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP2]], 4
17+
; CHECK-NEXT: [[T1_0_LCSSA2:%.*]] = ptrtoint i8* [[T1_0_LCSSA]] to i64
18+
; CHECK-NEXT: [[TMP0:%.*]] = trunc i64 [[ARR1]] to i32
19+
; CHECK-NEXT: [[TMP1:%.*]] = sub i32 0, [[TMP0]]
20+
; CHECK-NEXT: [[TMP2:%.*]] = trunc i64 [[T1_0_LCSSA2]] to i32
21+
; CHECK-NEXT: [[TMP3:%.*]] = add i32 [[TMP1]], [[TMP2]]
22+
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP3]], 4
2223
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]]
2324
; CHECK: vector.scevcheck:
24-
; CHECK-NEXT: [[TMP3:%.*]] = add i64 [[T1_0_LCSSA1]], -1
25-
; CHECK-NEXT: [[TMP4:%.*]] = sub i64 [[TMP3]], [[ARR2]]
26-
; CHECK-NEXT: [[TMP5:%.*]] = trunc i64 [[TMP4]] to i8
27-
; CHECK-NEXT: [[TMP6:%.*]] = add i8 1, [[TMP5]]
28-
; CHECK-NEXT: [[TMP9:%.*]] = icmp slt i8 [[TMP6]], 1
29-
; CHECK-NEXT: [[TMP11:%.*]] = icmp ugt i64 [[TMP4]], 255
30-
; CHECK-NEXT: [[TMP12:%.*]] = or i1 [[TMP9]], [[TMP11]]
31-
; CHECK-NEXT: br i1 [[TMP12]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
25+
; CHECK-NEXT: [[TMP4:%.*]] = sub i64 -1, [[ARR1]]
26+
; CHECK-NEXT: [[TMP5:%.*]] = add i64 [[TMP4]], [[T1_0_LCSSA2]]
27+
; CHECK-NEXT: [[TMP6:%.*]] = trunc i64 [[TMP5]] to i8
28+
; CHECK-NEXT: [[TMP7:%.*]] = add i8 1, [[TMP6]]
29+
; CHECK-NEXT: [[TMP8:%.*]] = icmp slt i8 [[TMP7]], 1
30+
; CHECK-NEXT: [[TMP9:%.*]] = icmp ugt i64 [[TMP5]], 255
31+
; CHECK-NEXT: [[TMP10:%.*]] = or i1 [[TMP8]], [[TMP9]]
32+
; CHECK-NEXT: br i1 [[TMP10]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
3233
; CHECK: vector.ph:
33-
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[TMP2]], 4
34-
; CHECK-NEXT: [[N_VEC:%.*]] = sub i32 [[TMP2]], [[N_MOD_VF]]
34+
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[TMP3]], 4
35+
; CHECK-NEXT: [[N_VEC:%.*]] = sub i32 [[TMP3]], [[N_MOD_VF]]
3536
; CHECK-NEXT: [[IND_END:%.*]] = trunc i32 [[N_VEC]] to i8
3637
; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i8> poison, i8 [[T9:%.*]], i32 0
3738
; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i8> [[BROADCAST_SPLATINSERT]], <4 x i8> poison, <4 x i32> zeroinitializer
3839
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
3940
; CHECK: vector.body:
4041
; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
4142
; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i8> [ <i8 0, i8 1, i8 2, i8 3>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
42-
; CHECK-NEXT: [[TMP14:%.*]] = add <4 x i8> [[VEC_IND]], <i8 1, i8 1, i8 1, i8 1>
43-
; CHECK-NEXT: [[TMP15:%.*]] = extractelement <4 x i8> [[TMP14]], i32 0
44-
; CHECK-NEXT: [[TMP16:%.*]] = getelementptr inbounds i8, i8* [[ARR]], i8 [[TMP15]]
45-
; CHECK-NEXT: [[TMP17:%.*]] = icmp slt <4 x i8> [[TMP14]], [[BROADCAST_SPLAT]]
46-
; CHECK-NEXT: [[TMP18:%.*]] = zext <4 x i1> [[TMP17]] to <4 x i8>
47-
; CHECK-NEXT: [[TMP19:%.*]] = getelementptr inbounds i8, i8* [[TMP16]], i32 0
48-
; CHECK-NEXT: [[TMP20:%.*]] = bitcast i8* [[TMP19]] to <4 x i8>*
49-
; CHECK-NEXT: store <4 x i8> [[TMP18]], <4 x i8>* [[TMP20]], align 1
43+
; CHECK-NEXT: [[TMP11:%.*]] = add <4 x i8> [[VEC_IND]], <i8 1, i8 1, i8 1, i8 1>
44+
; CHECK-NEXT: [[TMP12:%.*]] = extractelement <4 x i8> [[TMP11]], i32 0
45+
; CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds i8, i8* [[ARR]], i8 [[TMP12]]
46+
; CHECK-NEXT: [[TMP14:%.*]] = icmp slt <4 x i8> [[TMP11]], [[BROADCAST_SPLAT]]
47+
; CHECK-NEXT: [[TMP15:%.*]] = zext <4 x i1> [[TMP14]] to <4 x i8>
48+
; CHECK-NEXT: [[TMP16:%.*]] = getelementptr inbounds i8, i8* [[TMP13]], i32 0
49+
; CHECK-NEXT: [[TMP17:%.*]] = bitcast i8* [[TMP16]] to <4 x i8>*
50+
; CHECK-NEXT: store <4 x i8> [[TMP15]], <4 x i8>* [[TMP17]], align 1
5051
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
5152
; CHECK-NEXT: [[VEC_IND_NEXT]] = add <4 x i8> [[VEC_IND]], <i8 4, i8 4, i8 4, i8 4>
52-
; CHECK-NEXT: [[TMP21:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
53-
; CHECK-NEXT: br i1 [[TMP21]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
53+
; CHECK-NEXT: [[TMP18:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
54+
; CHECK-NEXT: br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
5455
; CHECK: middle.block:
55-
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[TMP2]], [[N_VEC]]
56+
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[TMP3]], [[N_VEC]]
5657
; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH]]
5758
; CHECK: scalar.ph:
5859
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i8 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_PREHEADER]] ], [ 0, [[VECTOR_SCEVCHECK]] ]

0 commit comments

Comments
 (0)