Skip to content

Commit 092b6e7

Browse files
bjopebjope
authored
[InstCombine] Handle "add like" in ADD+GEP->GEP+GEP rewrites (#135156)
Considering that "or disjoint" is the canonical for certain add operations, then I think we want to support such "add like" operations when doing ADD+GEP->GEP+GEP rewrites to make things more consistent. Problem was found when improving ValueTracking, which turned an ADD into OR, and then suddenly optimizations got worse due to these rewrites no longer triggering.
1 parent a485abb commit 092b6e7

22 files changed

+406
-409
lines changed

‎llvm/lib/Transforms/InstCombine/InstructionCombining.cpp

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -3132,7 +3132,7 @@ Instruction *InstCombinerImpl::visitGetElementPtrInst(GetElementPtrInst &GEP) {
31323132
// Try to replace ADD + GEP with GEP + GEP.
31333133
Value *Idx1, *Idx2;
31343134
if (match(GEP.getOperand(1),
3135-
m_OneUse(m_Add(m_Value(Idx1), m_Value(Idx2))))) {
3135+
m_OneUse(m_AddLike(m_Value(Idx1), m_Value(Idx2))))) {
31363136
// %idx = add i64 %idx1, %idx2
31373137
// %gep = getelementptr i32, ptr %ptr, i64 %idx
31383138
// as:
@@ -3148,7 +3148,7 @@ Instruction *InstCombinerImpl::visitGetElementPtrInst(GetElementPtrInst &GEP) {
31483148
NewPtr, Idx2, "", NWFlags));
31493149
}
31503150
ConstantInt *C;
3151-
if (match(GEP.getOperand(1), m_OneUse(m_SExtLike(m_OneUse(m_NSWAdd(
3151+
if (match(GEP.getOperand(1), m_OneUse(m_SExtLike(m_OneUse(m_NSWAddLike(
31523152
m_Value(Idx1), m_ConstantInt(C))))))) {
31533153
// %add = add nsw i32 %idx1, idx2
31543154
// %sidx = sext i32 %add to i64

‎llvm/test/Transforms/InstCombine/array.ll

Lines changed: 5 additions & 9 deletions
Original file line numberDiff line numberDiff line change
@@ -109,12 +109,11 @@ entry:
109109
ret void
110110
}
111111

112-
; FIXME: Should be transformed as OR+GEP -> GEP+GEP (similar to gep_inbounds_add_nuw below).
113112
define ptr @gep_inbounds_nuwaddlike(ptr %ptr, i64 %a, i64 %b) {
114113
; CHECK-LABEL: define ptr @gep_inbounds_nuwaddlike(
115114
; CHECK-SAME: ptr [[PTR:%.*]], i64 [[A:%.*]], i64 [[B:%.*]]) {
116-
; CHECK-NEXT: [[ADD:%.*]] = or disjoint i64 [[A]], [[B]]
117-
; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds nuw i32, ptr [[PTR]], i64 [[ADD]]
115+
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds nuw i32, ptr [[PTR]], i64 [[A]]
116+
; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds nuw i32, ptr [[TMP1]], i64 [[B]]
118117
; CHECK-NEXT: ret ptr [[GEP]]
119118
;
120119
%add = or disjoint i64 %a, %b
@@ -268,18 +267,15 @@ define ptr @gep_inbounds_sext_add_nonneg(ptr %ptr, i32 %a) {
268267
ret ptr %gep
269268
}
270269

271-
; FIXME: Could be optimized similar to gep_inbounds_sext_add_nonneg above
272-
; (difference is that we are using disjoint OR which is canonical form
273-
; of ADD with disjoint operands).
274270
define ptr @gep_inbounds_sext_addlike_nonneg(ptr %ptr, i32 %a) {
275271
; CHECK-LABEL: define ptr @gep_inbounds_sext_addlike_nonneg(
276272
; CHECK-SAME: ptr [[PTR:%.*]], i32 [[A:%.*]]) {
277273
; CHECK-NEXT: [[A_NNEG:%.*]] = icmp sgt i32 [[A]], -1
278274
; CHECK-NEXT: call void @llvm.assume(i1 [[A_NNEG]])
279-
; CHECK-NEXT: [[ADD:%.*]] = or disjoint i32 [[A]], 10
280-
; CHECK-NEXT: [[IDX:%.*]] = zext nneg i32 [[ADD]] to i64
275+
; CHECK-NEXT: [[IDX:%.*]] = zext nneg i32 [[A]] to i64
281276
; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds nuw i32, ptr [[PTR]], i64 [[IDX]]
282-
; CHECK-NEXT: ret ptr [[GEP]]
277+
; CHECK-NEXT: [[GEP1:%.*]] = getelementptr inbounds nuw i8, ptr [[GEP]], i64 40
278+
; CHECK-NEXT: ret ptr [[GEP1]]
283279
;
284280
%a.nneg = icmp sgt i32 %a, -1
285281
call void @llvm.assume(i1 %a.nneg)

‎llvm/test/Transforms/InstCombine/gep-merge-constant-indices.ll

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -157,8 +157,8 @@ define ptr @partialConstant3(ptr %p) {
157157
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds nuw i8, ptr [[P:%.*]], i64 4
158158
; CHECK-NEXT: [[TMP2:%.*]] = ptrtoint ptr [[TMP1]] to i64
159159
; CHECK-NEXT: [[DOTIDX:%.*]] = shl nsw i64 [[TMP2]], 5
160-
; CHECK-NEXT: [[DOTOFFS:%.*]] = or disjoint i64 [[DOTIDX]], 16
161-
; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i8, ptr [[TMP1]], i64 [[DOTOFFS]]
160+
; CHECK-NEXT: [[TMP4:%.*]] = getelementptr i8, ptr [[TMP1]], i64 [[DOTIDX]]
161+
; CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[TMP4]], i64 16
162162
; CHECK-NEXT: ret ptr [[TMP3]]
163163
;
164164
%1 = getelementptr inbounds i32, ptr %p, i64 1

‎llvm/test/Transforms/InstCombine/gep-vector.ll

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -129,8 +129,8 @@ define ptr @test_accumulate_constant_offset_vscale_nonzero(<vscale x 16 x i1> %p
129129
; CHECK-LABEL: @test_accumulate_constant_offset_vscale_nonzero(
130130
; CHECK-NEXT: [[TMP1:%.*]] = call i64 @llvm.vscale.i64()
131131
; CHECK-NEXT: [[TMP2:%.*]] = shl i64 [[TMP1]], 4
132-
; CHECK-NEXT: [[GEP_OFFS:%.*]] = or disjoint i64 [[TMP2]], 4
133-
; CHECK-NEXT: [[GEP:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[GEP_OFFS]]
132+
; CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[TMP2]]
133+
; CHECK-NEXT: [[GEP:%.*]] = getelementptr i8, ptr [[TMP3]], i64 4
134134
; CHECK-NEXT: ret ptr [[GEP]]
135135
;
136136
%gep = getelementptr <vscale x 16 x i8>, ptr %base, i64 1, i64 4

‎llvm/test/Transforms/InstCombine/vscale_gep.ll

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -60,8 +60,8 @@ define i32 @gep_alloca_inbounds_vscale_nonzero() {
6060
; CHECK-NEXT: [[A:%.*]] = alloca <vscale x 4 x i32>, align 16
6161
; CHECK-NEXT: [[TMP1:%.*]] = call i64 @llvm.vscale.i64()
6262
; CHECK-NEXT: [[TMP2:%.*]] = shl i64 [[TMP1]], 4
63-
; CHECK-NEXT: [[TMP_OFFS:%.*]] = or disjoint i64 [[TMP2]], 8
64-
; CHECK-NEXT: [[TMP:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP_OFFS]]
63+
; CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP2]]
64+
; CHECK-NEXT: [[TMP:%.*]] = getelementptr i8, ptr [[TMP3]], i64 8
6565
; CHECK-NEXT: [[LOAD:%.*]] = load i32, ptr [[TMP]], align 4
6666
; CHECK-NEXT: ret i32 [[LOAD]]
6767
;

‎llvm/test/Transforms/LoopUnroll/runtime-multiexit-heuristic.ll

Lines changed: 28 additions & 28 deletions
Original file line numberDiff line numberDiff line change
@@ -35,60 +35,60 @@ define i32 @test1(ptr nocapture %a, i64 %n) {
3535
; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDVARS_IV]]
3636
; CHECK-NEXT: [[TMP3:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
3737
; CHECK-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP3]], [[SUM_02]]
38-
; CHECK-NEXT: [[INDVARS_IV_NEXT:%.*]] = or disjoint i64 [[INDVARS_IV]], 1
3938
; CHECK-NEXT: br label [[FOR_EXITING_BLOCK_1:%.*]]
4039
; CHECK: for.exiting_block.1:
4140
; CHECK-NEXT: br i1 false, label [[OTHEREXIT_LOOPEXIT]], label [[LATCH_1:%.*]]
4241
; CHECK: latch.1:
43-
; CHECK-NEXT: [[ARRAYIDX_1:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT]]
42+
; CHECK-NEXT: [[TMP13:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDVARS_IV]]
43+
; CHECK-NEXT: [[ARRAYIDX_1:%.*]] = getelementptr i8, ptr [[TMP13]], i64 4
4444
; CHECK-NEXT: [[TMP4:%.*]] = load i32, ptr [[ARRAYIDX_1]], align 4
4545
; CHECK-NEXT: [[ADD_1:%.*]] = add nsw i32 [[TMP4]], [[ADD]]
46-
; CHECK-NEXT: [[INDVARS_IV_NEXT_1:%.*]] = or disjoint i64 [[INDVARS_IV]], 2
4746
; CHECK-NEXT: br label [[FOR_EXITING_BLOCK_2:%.*]]
4847
; CHECK: for.exiting_block.2:
4948
; CHECK-NEXT: br i1 false, label [[OTHEREXIT_LOOPEXIT]], label [[LATCH_2:%.*]]
5049
; CHECK: latch.2:
51-
; CHECK-NEXT: [[ARRAYIDX_2:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_1]]
50+
; CHECK-NEXT: [[TMP15:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDVARS_IV]]
51+
; CHECK-NEXT: [[ARRAYIDX_2:%.*]] = getelementptr i8, ptr [[TMP15]], i64 8
5252
; CHECK-NEXT: [[TMP5:%.*]] = load i32, ptr [[ARRAYIDX_2]], align 4
5353
; CHECK-NEXT: [[ADD_2:%.*]] = add nsw i32 [[TMP5]], [[ADD_1]]
54-
; CHECK-NEXT: [[INDVARS_IV_NEXT_2:%.*]] = or disjoint i64 [[INDVARS_IV]], 3
5554
; CHECK-NEXT: br label [[FOR_EXITING_BLOCK_3:%.*]]
5655
; CHECK: for.exiting_block.3:
5756
; CHECK-NEXT: br i1 false, label [[OTHEREXIT_LOOPEXIT]], label [[LATCH_3:%.*]]
5857
; CHECK: latch.3:
59-
; CHECK-NEXT: [[ARRAYIDX_3:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_2]]
58+
; CHECK-NEXT: [[TMP17:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDVARS_IV]]
59+
; CHECK-NEXT: [[ARRAYIDX_3:%.*]] = getelementptr i8, ptr [[TMP17]], i64 12
6060
; CHECK-NEXT: [[TMP6:%.*]] = load i32, ptr [[ARRAYIDX_3]], align 4
6161
; CHECK-NEXT: [[ADD_3:%.*]] = add nsw i32 [[TMP6]], [[ADD_2]]
62-
; CHECK-NEXT: [[INDVARS_IV_NEXT_3:%.*]] = or disjoint i64 [[INDVARS_IV]], 4
6362
; CHECK-NEXT: br label [[FOR_EXITING_BLOCK_4:%.*]]
6463
; CHECK: for.exiting_block.4:
6564
; CHECK-NEXT: br i1 false, label [[OTHEREXIT_LOOPEXIT]], label [[LATCH_4:%.*]]
6665
; CHECK: latch.4:
67-
; CHECK-NEXT: [[ARRAYIDX_4:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_3]]
66+
; CHECK-NEXT: [[TMP18:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDVARS_IV]]
67+
; CHECK-NEXT: [[ARRAYIDX_4:%.*]] = getelementptr i8, ptr [[TMP18]], i64 16
6868
; CHECK-NEXT: [[TMP7:%.*]] = load i32, ptr [[ARRAYIDX_4]], align 4
6969
; CHECK-NEXT: [[ADD_4:%.*]] = add nsw i32 [[TMP7]], [[ADD_3]]
70-
; CHECK-NEXT: [[INDVARS_IV_NEXT_4:%.*]] = or disjoint i64 [[INDVARS_IV]], 5
7170
; CHECK-NEXT: br label [[FOR_EXITING_BLOCK_5:%.*]]
7271
; CHECK: for.exiting_block.5:
7372
; CHECK-NEXT: br i1 false, label [[OTHEREXIT_LOOPEXIT]], label [[LATCH_5:%.*]]
7473
; CHECK: latch.5:
75-
; CHECK-NEXT: [[ARRAYIDX_5:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_4]]
74+
; CHECK-NEXT: [[TMP12:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDVARS_IV]]
75+
; CHECK-NEXT: [[ARRAYIDX_5:%.*]] = getelementptr i8, ptr [[TMP12]], i64 20
7676
; CHECK-NEXT: [[TMP8:%.*]] = load i32, ptr [[ARRAYIDX_5]], align 4
7777
; CHECK-NEXT: [[ADD_5:%.*]] = add nsw i32 [[TMP8]], [[ADD_4]]
78-
; CHECK-NEXT: [[INDVARS_IV_NEXT_5:%.*]] = or disjoint i64 [[INDVARS_IV]], 6
7978
; CHECK-NEXT: br label [[FOR_EXITING_BLOCK_6:%.*]]
8079
; CHECK: for.exiting_block.6:
8180
; CHECK-NEXT: br i1 false, label [[OTHEREXIT_LOOPEXIT]], label [[LATCH_6:%.*]]
8281
; CHECK: latch.6:
83-
; CHECK-NEXT: [[ARRAYIDX_6:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_5]]
82+
; CHECK-NEXT: [[TMP14:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDVARS_IV]]
83+
; CHECK-NEXT: [[ARRAYIDX_6:%.*]] = getelementptr i8, ptr [[TMP14]], i64 24
8484
; CHECK-NEXT: [[TMP9:%.*]] = load i32, ptr [[ARRAYIDX_6]], align 4
8585
; CHECK-NEXT: [[ADD_6:%.*]] = add nsw i32 [[TMP9]], [[ADD_5]]
86-
; CHECK-NEXT: [[INDVARS_IV_NEXT_6:%.*]] = or disjoint i64 [[INDVARS_IV]], 7
8786
; CHECK-NEXT: br label [[FOR_EXITING_BLOCK_7:%.*]]
8887
; CHECK: for.exiting_block.7:
8988
; CHECK-NEXT: br i1 false, label [[OTHEREXIT_LOOPEXIT]], label [[LATCH_7]]
9089
; CHECK: latch.7:
91-
; CHECK-NEXT: [[ARRAYIDX_7:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_6]]
90+
; CHECK-NEXT: [[TMP16:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDVARS_IV]]
91+
; CHECK-NEXT: [[ARRAYIDX_7:%.*]] = getelementptr i8, ptr [[TMP16]], i64 28
9292
; CHECK-NEXT: [[TMP10:%.*]] = load i32, ptr [[ARRAYIDX_7]], align 4
9393
; CHECK-NEXT: [[ADD_7]] = add nsw i32 [[TMP10]], [[ADD_6]]
9494
; CHECK-NEXT: [[INDVARS_IV_NEXT_7]] = add i64 [[INDVARS_IV]], 8
@@ -763,60 +763,60 @@ define i32 @test5(ptr nocapture %a, i64 %n) {
763763
; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDVARS_IV]]
764764
; CHECK-NEXT: [[TMP3:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
765765
; CHECK-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP3]], [[SUM_02]]
766-
; CHECK-NEXT: [[INDVARS_IV_NEXT:%.*]] = or disjoint i64 [[INDVARS_IV]], 1
767766
; CHECK-NEXT: br label [[FOR_EXITING_BLOCK_1:%.*]]
768767
; CHECK: for.exiting_block.1:
769768
; CHECK-NEXT: br i1 false, label [[OTHEREXIT_LOOPEXIT]], label [[LATCH_1:%.*]]
770769
; CHECK: latch.1:
771-
; CHECK-NEXT: [[ARRAYIDX_1:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT]]
770+
; CHECK-NEXT: [[TMP13:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDVARS_IV]]
771+
; CHECK-NEXT: [[ARRAYIDX_1:%.*]] = getelementptr i8, ptr [[TMP13]], i64 4
772772
; CHECK-NEXT: [[TMP4:%.*]] = load i32, ptr [[ARRAYIDX_1]], align 4
773773
; CHECK-NEXT: [[ADD_1:%.*]] = add nsw i32 [[TMP4]], [[ADD]]
774-
; CHECK-NEXT: [[INDVARS_IV_NEXT_1:%.*]] = or disjoint i64 [[INDVARS_IV]], 2
775774
; CHECK-NEXT: br label [[FOR_EXITING_BLOCK_2:%.*]]
776775
; CHECK: for.exiting_block.2:
777776
; CHECK-NEXT: br i1 false, label [[OTHEREXIT_LOOPEXIT]], label [[LATCH_2:%.*]]
778777
; CHECK: latch.2:
779-
; CHECK-NEXT: [[ARRAYIDX_2:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_1]]
778+
; CHECK-NEXT: [[TMP15:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDVARS_IV]]
779+
; CHECK-NEXT: [[ARRAYIDX_2:%.*]] = getelementptr i8, ptr [[TMP15]], i64 8
780780
; CHECK-NEXT: [[TMP5:%.*]] = load i32, ptr [[ARRAYIDX_2]], align 4
781781
; CHECK-NEXT: [[ADD_2:%.*]] = add nsw i32 [[TMP5]], [[ADD_1]]
782-
; CHECK-NEXT: [[INDVARS_IV_NEXT_2:%.*]] = or disjoint i64 [[INDVARS_IV]], 3
783782
; CHECK-NEXT: br label [[FOR_EXITING_BLOCK_3:%.*]]
784783
; CHECK: for.exiting_block.3:
785784
; CHECK-NEXT: br i1 false, label [[OTHEREXIT_LOOPEXIT]], label [[LATCH_3:%.*]]
786785
; CHECK: latch.3:
787-
; CHECK-NEXT: [[ARRAYIDX_3:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_2]]
786+
; CHECK-NEXT: [[TMP17:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDVARS_IV]]
787+
; CHECK-NEXT: [[ARRAYIDX_3:%.*]] = getelementptr i8, ptr [[TMP17]], i64 12
788788
; CHECK-NEXT: [[TMP6:%.*]] = load i32, ptr [[ARRAYIDX_3]], align 4
789789
; CHECK-NEXT: [[ADD_3:%.*]] = add nsw i32 [[TMP6]], [[ADD_2]]
790-
; CHECK-NEXT: [[INDVARS_IV_NEXT_3:%.*]] = or disjoint i64 [[INDVARS_IV]], 4
791790
; CHECK-NEXT: br label [[FOR_EXITING_BLOCK_4:%.*]]
792791
; CHECK: for.exiting_block.4:
793792
; CHECK-NEXT: br i1 false, label [[OTHEREXIT_LOOPEXIT]], label [[LATCH_4:%.*]]
794793
; CHECK: latch.4:
795-
; CHECK-NEXT: [[ARRAYIDX_4:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_3]]
794+
; CHECK-NEXT: [[TMP18:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDVARS_IV]]
795+
; CHECK-NEXT: [[ARRAYIDX_4:%.*]] = getelementptr i8, ptr [[TMP18]], i64 16
796796
; CHECK-NEXT: [[TMP7:%.*]] = load i32, ptr [[ARRAYIDX_4]], align 4
797797
; CHECK-NEXT: [[ADD_4:%.*]] = add nsw i32 [[TMP7]], [[ADD_3]]
798-
; CHECK-NEXT: [[INDVARS_IV_NEXT_4:%.*]] = or disjoint i64 [[INDVARS_IV]], 5
799798
; CHECK-NEXT: br label [[FOR_EXITING_BLOCK_5:%.*]]
800799
; CHECK: for.exiting_block.5:
801800
; CHECK-NEXT: br i1 false, label [[OTHEREXIT_LOOPEXIT]], label [[LATCH_5:%.*]]
802801
; CHECK: latch.5:
803-
; CHECK-NEXT: [[ARRAYIDX_5:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_4]]
802+
; CHECK-NEXT: [[TMP12:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDVARS_IV]]
803+
; CHECK-NEXT: [[ARRAYIDX_5:%.*]] = getelementptr i8, ptr [[TMP12]], i64 20
804804
; CHECK-NEXT: [[TMP8:%.*]] = load i32, ptr [[ARRAYIDX_5]], align 4
805805
; CHECK-NEXT: [[ADD_5:%.*]] = add nsw i32 [[TMP8]], [[ADD_4]]
806-
; CHECK-NEXT: [[INDVARS_IV_NEXT_5:%.*]] = or disjoint i64 [[INDVARS_IV]], 6
807806
; CHECK-NEXT: br label [[FOR_EXITING_BLOCK_6:%.*]]
808807
; CHECK: for.exiting_block.6:
809808
; CHECK-NEXT: br i1 false, label [[OTHEREXIT_LOOPEXIT]], label [[LATCH_6:%.*]]
810809
; CHECK: latch.6:
811-
; CHECK-NEXT: [[ARRAYIDX_6:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_5]]
810+
; CHECK-NEXT: [[TMP14:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDVARS_IV]]
811+
; CHECK-NEXT: [[ARRAYIDX_6:%.*]] = getelementptr i8, ptr [[TMP14]], i64 24
812812
; CHECK-NEXT: [[TMP9:%.*]] = load i32, ptr [[ARRAYIDX_6]], align 4
813813
; CHECK-NEXT: [[ADD_6:%.*]] = add nsw i32 [[TMP9]], [[ADD_5]]
814-
; CHECK-NEXT: [[INDVARS_IV_NEXT_6:%.*]] = or disjoint i64 [[INDVARS_IV]], 7
815814
; CHECK-NEXT: br label [[FOR_EXITING_BLOCK_7:%.*]]
816815
; CHECK: for.exiting_block.7:
817816
; CHECK-NEXT: br i1 false, label [[OTHEREXIT_LOOPEXIT]], label [[LATCH_7]]
818817
; CHECK: latch.7:
819-
; CHECK-NEXT: [[ARRAYIDX_7:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_6]]
818+
; CHECK-NEXT: [[TMP16:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDVARS_IV]]
819+
; CHECK-NEXT: [[ARRAYIDX_7:%.*]] = getelementptr i8, ptr [[TMP16]], i64 28
820820
; CHECK-NEXT: [[TMP10:%.*]] = load i32, ptr [[ARRAYIDX_7]], align 4
821821
; CHECK-NEXT: [[ADD_7]] = add nsw i32 [[TMP10]], [[ADD_6]]
822822
; CHECK-NEXT: [[INDVARS_IV_NEXT_7]] = add i64 [[INDVARS_IV]], 8

‎llvm/test/Transforms/LoopVectorize/ARM/mve-reductions.ll

Lines changed: 4 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -1335,10 +1335,8 @@ define i32 @reduction_interleave_group(i32 %n, ptr %arr) #0 {
13351335
; CHECK: vector.body:
13361336
; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
13371337
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[TMP9:%.*]], [[VECTOR_BODY]] ]
1338-
; CHECK-NEXT: [[OFFSET_IDX:%.*]] = shl i32 [[INDEX]], 1
1339-
; CHECK-NEXT: [[TMP3:%.*]] = or disjoint i32 [[OFFSET_IDX]], 1
1340-
; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i32, ptr [[ARR:%.*]], i32 [[TMP3]]
1341-
; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i8, ptr [[TMP4]], i32 -4
1338+
; CHECK-NEXT: [[DOTIDX:%.*]] = shl i32 [[INDEX]], 3
1339+
; CHECK-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[ARR:%.*]], i32 [[DOTIDX]]
13421340
; CHECK-NEXT: [[WIDE_VEC:%.*]] = load <8 x i32>, ptr [[TMP5]], align 4
13431341
; CHECK-NEXT: [[STRIDED_VEC:%.*]] = shufflevector <8 x i32> [[WIDE_VEC]], <8 x i32> poison, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
13441342
; CHECK-NEXT: [[STRIDED_VEC1:%.*]] = shufflevector <8 x i32> [[WIDE_VEC]], <8 x i32> poison, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
@@ -1359,8 +1357,8 @@ define i32 @reduction_interleave_group(i32 %n, ptr %arr) #0 {
13591357
; CHECK: for.body:
13601358
; CHECK-NEXT: [[IV:%.*]] = phi i32 [ [[IV_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
13611359
; CHECK-NEXT: [[RED_PHI:%.*]] = phi i32 [ [[RED_2:%.*]], [[FOR_BODY]] ], [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ]
1362-
; CHECK-NEXT: [[ADD:%.*]] = or disjoint i32 [[IV]], 1
1363-
; CHECK-NEXT: [[GEP_0:%.*]] = getelementptr inbounds i32, ptr [[ARR]], i32 [[ADD]]
1360+
; CHECK-NEXT: [[TMP11:%.*]] = getelementptr i32, ptr [[ARR]], i32 [[IV]]
1361+
; CHECK-NEXT: [[GEP_0:%.*]] = getelementptr i8, ptr [[TMP11]], i32 4
13641362
; CHECK-NEXT: [[L_0:%.*]] = load i32, ptr [[GEP_0]], align 4
13651363
; CHECK-NEXT: [[GEP_1:%.*]] = getelementptr inbounds i32, ptr [[ARR]], i32 [[IV]]
13661364
; CHECK-NEXT: [[L_1:%.*]] = load i32, ptr [[GEP_1]], align 4

‎llvm/test/Transforms/LoopVectorize/SystemZ/addressing.ll

Lines changed: 5 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -14,11 +14,11 @@ define i32 @foo(ptr nocapture %A) {
1414
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
1515
; CHECK: vector.body:
1616
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
17-
; CHECK-NEXT: [[TMP0:%.*]] = shl i64 [[INDEX]], 2
18-
; CHECK-NEXT: [[TMP1:%.*]] = or disjoint i64 [[TMP0]], 4
1917
; CHECK-NEXT: [[DOTIDX:%.*]] = shl nsw i64 [[INDEX]], 4
2018
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i8, ptr [[A:%.*]], i64 [[DOTIDX]]
21-
; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP1]]
19+
; CHECK-NEXT: [[DOTIDX1:%.*]] = shl i64 [[INDEX]], 4
20+
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[A]], i64 [[DOTIDX1]]
21+
; CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[TMP1]], i64 16
2222
; CHECK-NEXT: store i32 4, ptr [[TMP2]], align 4
2323
; CHECK-NEXT: store i32 4, ptr [[TMP3]], align 4
2424
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
@@ -61,9 +61,9 @@ define i32 @foo1(ptr nocapture noalias %A, ptr nocapture %PtrPtr) {
6161
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
6262
; CHECK: vector.body:
6363
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
64-
; CHECK-NEXT: [[TMP0:%.*]] = or disjoint i64 [[INDEX]], 1
6564
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds ptr, ptr [[PTRPTR:%.*]], i64 [[INDEX]]
66-
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds ptr, ptr [[PTRPTR]], i64 [[TMP0]]
65+
; CHECK-NEXT: [[TMP11:%.*]] = getelementptr ptr, ptr [[PTRPTR]], i64 [[INDEX]]
66+
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[TMP11]], i64 8
6767
; CHECK-NEXT: [[TMP3:%.*]] = load ptr, ptr [[TMP1]], align 8
6868
; CHECK-NEXT: [[TMP4:%.*]] = load ptr, ptr [[TMP2]], align 8
6969
; CHECK-NEXT: [[TMP5:%.*]] = load i32, ptr [[TMP3]], align 4

0 commit comments

Comments
 (0)