Skip to content

Commit 0664b60

Browse files
nikicnikic
committed
[InstCombine] Strip trailing zero GEP indices
Zero indices at the end do not change the GEP offset and can be removed. (Doing the same at the start requires adjusting the source element type.)
1 parent 3696120 commit 0664b60

File tree

10 files changed

+74
-73
lines changed

10 files changed

+74
-73
lines changed

‎llvm/lib/Transforms/InstCombine/InstructionCombining.cpp‎

Lines changed: 8 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -3205,6 +3205,14 @@ Instruction *InstCombinerImpl::visitGetElementPtrInst(GetElementPtrInst &GEP) {
32053205
return replaceInstUsesWith(GEP, NewGEP);
32063206
}
32073207

3208+
// Strip trailing zero indices.
3209+
auto *LastIdx = dyn_cast<Constant>(Indices.back());
3210+
if (LastIdx && LastIdx->isNullValue() && !LastIdx->getType()->isVectorTy()) {
3211+
return replaceInstUsesWith(
3212+
GEP, Builder.CreateGEP(GEP.getSourceElementType(), PtrOp,
3213+
drop_end(Indices), "", GEP.getNoWrapFlags()));
3214+
}
3215+
32083216
// Scalarize vector operands; prefer splat-of-gep.as canonical form.
32093217
// Note that this looses information about undef lanes; we run it after
32103218
// demanded bits to partially mitigate that loss.

‎llvm/test/Transforms/InstCombine/load-cmp.ll‎

Lines changed: 2 additions & 9 deletions
Original file line numberDiff line numberDiff line change
@@ -215,10 +215,7 @@ define i1 @test10_struct(i32 %x) {
215215

216216
define i1 @test10_struct_noinbounds(i32 %x) {
217217
; CHECK-LABEL: @test10_struct_noinbounds(
218-
; CHECK-NEXT: [[P:%.*]] = getelementptr [[FOO:%.*]], ptr @GS, i32 [[X:%.*]], i32 0
219-
; CHECK-NEXT: [[Q:%.*]] = load i32, ptr [[P]], align 4
220-
; CHECK-NEXT: [[R:%.*]] = icmp eq i32 [[Q]], 9
221-
; CHECK-NEXT: ret i1 [[R]]
218+
; CHECK-NEXT: ret i1 false
222219
;
223220
%p = getelementptr %Foo, ptr @GS, i32 %x, i32 0
224221
%q = load i32, ptr %p
@@ -252,11 +249,7 @@ define i1 @test10_struct_i64(i64 %x){
252249

253250
define i1 @test10_struct_noinbounds_i16(i16 %x) {
254251
; CHECK-LABEL: @test10_struct_noinbounds_i16(
255-
; CHECK-NEXT: [[TMP1:%.*]] = sext i16 [[X:%.*]] to i32
256-
; CHECK-NEXT: [[P:%.*]] = getelementptr [[FOO:%.*]], ptr @GS, i32 [[TMP1]], i32 0
257-
; CHECK-NEXT: [[Q:%.*]] = load i32, ptr [[P]], align 4
258-
; CHECK-NEXT: [[R:%.*]] = icmp eq i32 [[Q]], 0
259-
; CHECK-NEXT: ret i1 [[R]]
252+
; CHECK-NEXT: ret i1 false
260253
;
261254
%p = getelementptr %Foo, ptr @GS, i16 %x, i32 0
262255
%q = load i32, ptr %p

‎llvm/test/Transforms/InstCombine/strcmp-3.ll‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -25,7 +25,7 @@ define i32 @fold_strcmp_a5i0_a5i1_to_0() {
2525

2626
define i32 @call_strcmp_a5i0_a5iI(i64 %I) {
2727
; CHECK-LABEL: @call_strcmp_a5i0_a5iI(
28-
; CHECK-NEXT: [[Q:%.*]] = getelementptr [5 x [4 x i8]], ptr @a5, i64 0, i64 [[I:%.*]], i64 0
28+
; CHECK-NEXT: [[Q:%.*]] = getelementptr [5 x [4 x i8]], ptr @a5, i64 0, i64 [[I:%.*]]
2929
; CHECK-NEXT: [[CMP:%.*]] = call i32 @strcmp(ptr noundef nonnull dereferenceable(4) @a5, ptr noundef nonnull dereferenceable(1) [[Q]])
3030
; CHECK-NEXT: ret i32 [[CMP]]
3131
;
@@ -40,7 +40,7 @@ define i32 @call_strcmp_a5i0_a5iI(i64 %I) {
4040

4141
define i32 @call_strcmp_a5iI_a5i0(i64 %I) {
4242
; CHECK-LABEL: @call_strcmp_a5iI_a5i0(
43-
; CHECK-NEXT: [[P:%.*]] = getelementptr [5 x [4 x i8]], ptr @a5, i64 0, i64 [[I:%.*]], i64 0
43+
; CHECK-NEXT: [[P:%.*]] = getelementptr [5 x [4 x i8]], ptr @a5, i64 0, i64 [[I:%.*]]
4444
; CHECK-NEXT: [[CMP:%.*]] = call i32 @strcmp(ptr noundef nonnull dereferenceable(1) [[P]], ptr noundef nonnull dereferenceable(4) @a5)
4545
; CHECK-NEXT: ret i32 [[CMP]]
4646
;

‎llvm/test/Transforms/InstCombine/vector_gep1-inseltpoison.ll‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -55,7 +55,7 @@ define <2 x i1> @test5(<2 x ptr> %a) {
5555

5656
define <2 x ptr> @test7(<2 x ptr> %a) {
5757
; CHECK-LABEL: @test7(
58-
; CHECK-NEXT: [[W:%.*]] = getelementptr { i32, i32 }, <2 x ptr> [[A:%.*]], <2 x i64> <i64 5, i64 9>, i32 0
58+
; CHECK-NEXT: [[W:%.*]] = getelementptr { i32, i32 }, <2 x ptr> [[A:%.*]], <2 x i64> <i64 5, i64 9>
5959
; CHECK-NEXT: ret <2 x ptr> [[W]]
6060
;
6161
%w = getelementptr {i32, i32}, <2 x ptr> %a, <2 x i32> <i32 5, i32 9>, <2 x i32> zeroinitializer

‎llvm/test/Transforms/InstCombine/vector_gep1.ll‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -55,7 +55,7 @@ define <2 x i1> @test5(<2 x ptr> %a) {
5555

5656
define <2 x ptr> @test7(<2 x ptr> %a) {
5757
; CHECK-LABEL: @test7(
58-
; CHECK-NEXT: [[W:%.*]] = getelementptr { i32, i32 }, <2 x ptr> [[A:%.*]], <2 x i64> <i64 5, i64 9>, i32 0
58+
; CHECK-NEXT: [[W:%.*]] = getelementptr { i32, i32 }, <2 x ptr> [[A:%.*]], <2 x i64> <i64 5, i64 9>
5959
; CHECK-NEXT: ret <2 x ptr> [[W]]
6060
;
6161
%w = getelementptr {i32, i32}, <2 x ptr> %a, <2 x i32> <i32 5, i32 9>, <2 x i32> zeroinitializer

‎llvm/test/Transforms/InstCombine/vectorgep-crash.ll‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -14,7 +14,7 @@ define <8 x ptr> @test_vector_gep(ptr %arg1, <8 x i64> %arg2) {
1414
; CHECK-LABEL: define <8 x ptr> @test_vector_gep(
1515
; CHECK-SAME: ptr [[ARG1:%.*]], <8 x i64> [[ARG2:%.*]]) {
1616
; CHECK-NEXT: [[TOP:.*:]]
17-
; CHECK-NEXT: [[VECTORGEP14:%.*]] = getelementptr inbounds [[DUAL:%.*]], ptr [[ARG1]], <8 x i64> [[ARG2]], i32 1, i32 0, i64 0, i32 1, i32 0, i64 0
17+
; CHECK-NEXT: [[VECTORGEP14:%.*]] = getelementptr inbounds [[DUAL:%.*]], ptr [[ARG1]], <8 x i64> [[ARG2]], i32 1, i32 0, i64 0, i32 1
1818
; CHECK-NEXT: ret <8 x ptr> [[VECTORGEP14]]
1919
;
2020
top:

‎llvm/test/Transforms/LoopVectorize/AArch64/sve-interleaved-accesses.ll‎

Lines changed: 11 additions & 11 deletions
Original file line numberDiff line numberDiff line change
@@ -267,7 +267,7 @@ define i32 @test_struct_load6(ptr %S) #1 {
267267
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
268268
; CHECK-NEXT: [[VEC_IND:%.*]] = phi <vscale x 4 x i64> [ [[TMP2]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
269269
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP16:%.*]], [[VECTOR_BODY]] ]
270-
; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT_ST6:%.*]], ptr [[S:%.*]], <vscale x 4 x i64> [[VEC_IND]], i32 0
270+
; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT_ST6:%.*]], ptr [[S:%.*]], <vscale x 4 x i64> [[VEC_IND]]
271271
; CHECK-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 4 x i32> @llvm.masked.gather.nxv4i32.nxv4p0(<vscale x 4 x ptr> [[TMP5]], i32 4, <vscale x 4 x i1> splat (i1 true), <vscale x 4 x i32> poison)
272272
; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT_ST6]], ptr [[S]], <vscale x 4 x i64> [[VEC_IND]], i32 1
273273
; CHECK-NEXT: [[WIDE_MASKED_GATHER1:%.*]] = call <vscale x 4 x i32> @llvm.masked.gather.nxv4i32.nxv4p0(<vscale x 4 x ptr> [[TMP6]], i32 4, <vscale x 4 x i1> splat (i1 true), <vscale x 4 x i32> poison)
@@ -366,7 +366,7 @@ define void @test_reversed_load2_store2(ptr noalias nocapture readonly %A, ptr n
366366
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
367367
; CHECK-NEXT: [[VEC_IND:%.*]] = phi <vscale x 4 x i32> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
368368
; CHECK-NEXT: [[OFFSET_IDX:%.*]] = sub i64 1023, [[INDEX]]
369-
; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds [[STRUCT_ST2:%.*]], ptr [[A:%.*]], i64 [[OFFSET_IDX]], i32 0
369+
; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds [[STRUCT_ST2:%.*]], ptr [[A:%.*]], i64 [[OFFSET_IDX]]
370370
; CHECK-NEXT: [[TMP6:%.*]] = shl nuw nsw i64 [[TMP0]], 3
371371
; CHECK-NEXT: [[TMP8:%.*]] = sub nsw i64 2, [[TMP6]]
372372
; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds i32, ptr [[TMP4]], i64 [[TMP8]]
@@ -378,7 +378,7 @@ define void @test_reversed_load2_store2(ptr noalias nocapture readonly %A, ptr n
378378
; CHECK-NEXT: [[REVERSE1:%.*]] = call <vscale x 4 x i32> @llvm.vector.reverse.nxv4i32(<vscale x 4 x i32> [[TMP11]])
379379
; CHECK-NEXT: [[TMP12:%.*]] = add nsw <vscale x 4 x i32> [[REVERSE]], [[VEC_IND]]
380380
; CHECK-NEXT: [[TMP13:%.*]] = sub nsw <vscale x 4 x i32> [[REVERSE1]], [[VEC_IND]]
381-
; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT_ST2]], ptr [[B:%.*]], i64 [[OFFSET_IDX]], i32 0
381+
; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT_ST2]], ptr [[B:%.*]], i64 [[OFFSET_IDX]]
382382
; CHECK-NEXT: [[TMP15:%.*]] = shl nuw nsw i64 [[TMP0]], 3
383383
; CHECK-NEXT: [[TMP18:%.*]] = sub nsw i64 2, [[TMP15]]
384384
; CHECK-NEXT: [[TMP19:%.*]] = getelementptr inbounds i32, ptr [[TMP14]], i64 [[TMP18]]
@@ -587,7 +587,7 @@ define void @load_gap_reverse(ptr noalias nocapture readonly %P1, ptr noalias no
587587
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
588588
; CHECK-NEXT: [[VEC_IND:%.*]] = phi <vscale x 4 x i64> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
589589
; CHECK-NEXT: [[TMP4:%.*]] = add nsw <vscale x 4 x i64> [[BROADCAST_SPLAT1]], [[VEC_IND]]
590-
; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[PAIR:%.*]], ptr [[P1:%.*]], <vscale x 4 x i64> [[VEC_IND]], i32 0
590+
; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[PAIR:%.*]], ptr [[P1:%.*]], <vscale x 4 x i64> [[VEC_IND]]
591591
; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[PAIR]], ptr [[P2:%.*]], <vscale x 4 x i64> [[VEC_IND]], i32 1
592592
; CHECK-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 4 x i64> @llvm.masked.gather.nxv4i64.nxv4p0(<vscale x 4 x ptr> [[TMP6]], i32 8, <vscale x 4 x i1> splat (i1 true), <vscale x 4 x i64> poison)
593593
; CHECK-NEXT: [[TMP7:%.*]] = sub nsw <vscale x 4 x i64> [[WIDE_MASKED_GATHER]], [[VEC_IND]]
@@ -729,7 +729,7 @@ define void @int_float_struct(ptr nocapture readonly %p) #0 {
729729
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
730730
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 4 x float> [ insertelement (<vscale x 4 x float> zeroinitializer, float undef, i32 0), [[VECTOR_PH]] ], [ [[TMP7:%.*]], [[VECTOR_BODY]] ]
731731
; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <vscale x 4 x i32> [ insertelement (<vscale x 4 x i32> zeroinitializer, i32 undef, i32 0), [[VECTOR_PH]] ], [ [[TMP6:%.*]], [[VECTOR_BODY]] ]
732-
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT_INTFLOAT:%.*]], ptr [[P:%.*]], i64 [[INDEX]], i32 0
732+
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT_INTFLOAT:%.*]], ptr [[P:%.*]], i64 [[INDEX]]
733733
; CHECK-NEXT: [[WIDE_VEC:%.*]] = load <vscale x 8 x i32>, ptr [[TMP2]], align 4
734734
; CHECK-NEXT: [[STRIDED_VEC:%.*]] = call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.vector.deinterleave2.nxv8i32(<vscale x 8 x i32> [[WIDE_VEC]])
735735
; CHECK-NEXT: [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[STRIDED_VEC]], 0
@@ -814,7 +814,7 @@ define void @PR27626_0(ptr %p, i32 %z, i64 %n) #1 {
814814
; CHECK: vector.body:
815815
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
816816
; CHECK-NEXT: [[VEC_IND:%.*]] = phi <vscale x 4 x i64> [ [[TMP9]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
817-
; CHECK-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[PAIR_I32:%.*]], ptr [[P:%.*]], <vscale x 4 x i64> [[VEC_IND]], i32 0
817+
; CHECK-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[PAIR_I32:%.*]], ptr [[P:%.*]], <vscale x 4 x i64> [[VEC_IND]]
818818
; CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], <vscale x 4 x i64> [[VEC_IND]], i32 1
819819
; CHECK-NEXT: call void @llvm.masked.scatter.nxv4i32.nxv4p0(<vscale x 4 x i32> [[BROADCAST_SPLAT]], <vscale x 4 x ptr> [[TMP12]], i32 4, <vscale x 4 x i1> splat (i1 true))
820820
; CHECK-NEXT: [[TMP14:%.*]] = extractelement <vscale x 4 x ptr> [[TMP12]], i64 0
@@ -885,7 +885,7 @@ define i32 @PR27626_1(ptr %p, i64 %n) #1 {
885885
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
886886
; CHECK-NEXT: [[VEC_IND:%.*]] = phi <vscale x 4 x i64> [ [[TMP9]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
887887
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP17:%.*]], [[VECTOR_BODY]] ]
888-
; CHECK-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[PAIR_I32:%.*]], ptr [[P:%.*]], i64 [[INDEX]], i32 0
888+
; CHECK-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[PAIR_I32:%.*]], ptr [[P:%.*]], i64 [[INDEX]]
889889
; CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], <vscale x 4 x i64> [[VEC_IND]], i32 1
890890
; CHECK-NEXT: [[WIDE_VEC:%.*]] = load <vscale x 8 x i32>, ptr [[TMP12]], align 4
891891
; CHECK-NEXT: [[STRIDED_VEC:%.*]] = call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.vector.deinterleave2.nxv8i32(<vscale x 8 x i32> [[WIDE_VEC]])
@@ -963,7 +963,7 @@ define void @PR27626_2(ptr %p, i64 %n, i32 %z) #1 {
963963
; CHECK: vector.body:
964964
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
965965
; CHECK-NEXT: [[VEC_IND:%.*]] = phi <vscale x 4 x i64> [ [[TMP9]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
966-
; CHECK-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[PAIR_I32:%.*]], ptr [[P:%.*]], <vscale x 4 x i64> [[VEC_IND]], i32 0
966+
; CHECK-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[PAIR_I32:%.*]], ptr [[P:%.*]], <vscale x 4 x i64> [[VEC_IND]]
967967
; CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds i8, ptr [[P]], i64 -8
968968
; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], <vscale x 4 x i64> [[VEC_IND]], i32 1
969969
; CHECK-NEXT: call void @llvm.masked.scatter.nxv4i32.nxv4p0(<vscale x 4 x i32> [[BROADCAST_SPLAT]], <vscale x 4 x ptr> [[TMP12]], i32 4, <vscale x 4 x i1> splat (i1 true))
@@ -1036,7 +1036,7 @@ define i32 @PR27626_3(ptr %p, i64 %n, i32 %z) #1 {
10361036
; CHECK-NEXT: [[VEC_IND:%.*]] = phi <vscale x 4 x i64> [ [[TMP9]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
10371037
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP18:%.*]], [[VECTOR_BODY]] ]
10381038
; CHECK-NEXT: [[TMP12:%.*]] = add nuw nsw <vscale x 4 x i64> [[VEC_IND]], splat (i64 1)
1039-
; CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[PAIR_I32:%.*]], ptr [[P:%.*]], i64 [[INDEX]], i32 0
1039+
; CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[PAIR_I32:%.*]], ptr [[P:%.*]], i64 [[INDEX]]
10401040
; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[INDEX]], i32 1
10411041
; CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], <vscale x 4 x i64> [[TMP12]], i32 1
10421042
; CHECK-NEXT: [[WIDE_VEC:%.*]] = load <vscale x 8 x i32>, ptr [[TMP13]], align 4
@@ -1572,7 +1572,7 @@ define void @interleave_deinterleave_reverse(ptr noalias nocapture readonly %A,
15721572
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
15731573
; CHECK-NEXT: [[VEC_IND:%.*]] = phi <vscale x 4 x i32> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
15741574
; CHECK-NEXT: [[OFFSET_IDX:%.*]] = sub i64 1023, [[INDEX]]
1575-
; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT_XYZT:%.*]], ptr [[A:%.*]], i64 [[OFFSET_IDX]], i32 0
1575+
; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT_XYZT:%.*]], ptr [[A:%.*]], i64 [[OFFSET_IDX]]
15761576
; CHECK-NEXT: [[TMP6:%.*]] = shl nuw nsw i64 [[TMP0]], 4
15771577
; CHECK-NEXT: [[TMP9:%.*]] = sub nsw i64 4, [[TMP6]]
15781578
; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds i32, ptr [[TMP5]], i64 [[TMP9]]
@@ -1590,7 +1590,7 @@ define void @interleave_deinterleave_reverse(ptr noalias nocapture readonly %A,
15901590
; CHECK-NEXT: [[TMP18:%.*]] = sub nsw <vscale x 4 x i32> [[REVERSE3]], [[VEC_IND]]
15911591
; CHECK-NEXT: [[TMP19:%.*]] = mul nsw <vscale x 4 x i32> [[REVERSE4]], [[VEC_IND]]
15921592
; CHECK-NEXT: [[TMP20:%.*]] = shl nuw nsw <vscale x 4 x i32> [[REVERSE5]], [[VEC_IND]]
1593-
; CHECK-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT_XYZT]], ptr [[B:%.*]], i64 [[OFFSET_IDX]], i32 0
1593+
; CHECK-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT_XYZT]], ptr [[B:%.*]], i64 [[OFFSET_IDX]]
15941594
; CHECK-NEXT: [[TMP22:%.*]] = shl nuw nsw i64 [[TMP0]], 4
15951595
; CHECK-NEXT: [[TMP25:%.*]] = sub nsw i64 4, [[TMP22]]
15961596
; CHECK-NEXT: [[TMP26:%.*]] = getelementptr inbounds i32, ptr [[TMP21]], i64 [[TMP25]]

‎llvm/test/Transforms/LoopVectorize/consecutive-ptr-uniforms.ll‎

Lines changed: 17 additions & 17 deletions
Original file line numberDiff line numberDiff line change
@@ -87,10 +87,10 @@ for.end:
8787
; CHECK: %[[I1:.+]] = or disjoint i64 %index, 1
8888
; CHECK: %[[I2:.+]] = or disjoint i64 %index, 2
8989
; CHECK: %[[I3:.+]] = or disjoint i64 %index, 3
90-
; CHECK: getelementptr inbounds %pair, ptr %p, i64 %index, i32 0
91-
; CHECK: getelementptr inbounds %pair, ptr %p, i64 %[[I1]], i32 0
92-
; CHECK: getelementptr inbounds %pair, ptr %p, i64 %[[I2]], i32 0
93-
; CHECK: getelementptr inbounds %pair, ptr %p, i64 %[[I3]], i32 0
90+
; CHECK: getelementptr inbounds %pair, ptr %p, i64 %index
91+
; CHECK: getelementptr inbounds %pair, ptr %p, i64 %[[I1]]
92+
; CHECK: getelementptr inbounds %pair, ptr %p, i64 %[[I2]]
93+
; CHECK: getelementptr inbounds %pair, ptr %p, i64 %[[I3]]
9494
; CHECK: getelementptr inbounds %pair, ptr %p, i64 %index, i32 1
9595
; CHECK: getelementptr inbounds %pair, ptr %p, i64 %[[I1]], i32 1
9696
; CHECK: getelementptr inbounds %pair, ptr %p, i64 %[[I2]], i32 1
@@ -102,7 +102,7 @@ for.end:
102102
; INTER: vector.body
103103
; INTER: %index = phi i64 [ 0, %vector.ph ], [ %index.next, %vector.body ]
104104
; INTER-NOT: getelementptr
105-
; INTER: getelementptr inbounds %pair, ptr %p, i64 %index, i32 0
105+
; INTER: getelementptr inbounds %pair, ptr %p, i64 %index
106106
; INTER-NOT: getelementptr
107107
; INTER: br i1 {{.*}}, label %middle.block, label %vector.body
108108
;
@@ -145,10 +145,10 @@ for.end:
145145
; CHECK: %[[I1:.+]] = add i64 [[OFFSET_IDX]], -1
146146
; CHECK: %[[I2:.+]] = add i64 [[OFFSET_IDX]], -2
147147
; CHECK: %[[I3:.+]] = add i64 [[OFFSET_IDX]], -3
148-
; CHECK: getelementptr inbounds %pair, ptr %p, i64 [[OFFSET_IDX]], i32 0
149-
; CHECK: getelementptr inbounds %pair, ptr %p, i64 %[[I1]], i32 0
150-
; CHECK: getelementptr inbounds %pair, ptr %p, i64 %[[I2]], i32 0
151-
; CHECK: getelementptr inbounds %pair, ptr %p, i64 %[[I3]], i32 0
148+
; CHECK: getelementptr inbounds %pair, ptr %p, i64 [[OFFSET_IDX]]
149+
; CHECK: getelementptr inbounds %pair, ptr %p, i64 %[[I1]]
150+
; CHECK: getelementptr inbounds %pair, ptr %p, i64 %[[I2]]
151+
; CHECK: getelementptr inbounds %pair, ptr %p, i64 %[[I3]]
152152
; CHECK: getelementptr inbounds %pair, ptr %p, i64 [[OFFSET_IDX]], i32 1
153153
; CHECK: getelementptr inbounds %pair, ptr %p, i64 %[[I1]], i32 1
154154
; CHECK: getelementptr inbounds %pair, ptr %p, i64 %[[I2]], i32 1
@@ -161,7 +161,7 @@ for.end:
161161
; INTER: %index = phi i64 [ 0, %vector.ph ], [ %index.next, %vector.body ]
162162
; INTER: [[OFFSET_IDX:%.+]] = sub i64 %n, %index
163163
; INTER-NOT: getelementptr
164-
; INTER: %[[G0:.+]] = getelementptr inbounds %pair, ptr %p, i64 [[OFFSET_IDX]], i32 0
164+
; INTER: %[[G0:.+]] = getelementptr inbounds %pair, ptr %p, i64 [[OFFSET_IDX]]
165165
; INTER: getelementptr inbounds i8, ptr %[[G0]], i64 -24
166166
; INTER-NOT: getelementptr
167167
; INTER: br i1 {{.*}}, label %middle.block, label %vector.body
@@ -200,14 +200,14 @@ for.end:
200200
; INTER-NOT: LV: Found uniform instruction: %tmp0 = getelementptr inbounds %pair, ptr %p, i64 %i, i32 0
201201
; INTER: vector.body
202202
; INTER: %index = phi i64 [ 0, %vector.ph ], [ %index.next, {{.*}} ]
203-
; INTER: %[[G0:.+]] = getelementptr inbounds %pair, ptr %p, i64 %index, i32 0
203+
; INTER: %[[G0:.+]] = getelementptr inbounds %pair, ptr %p, i64 %index
204204
; INTER: %wide.vec = load <8 x i32>, ptr %[[G0]], align 8
205-
; INTER: %[[I1:.+]] = or disjoint i64 %index, 1
206-
; INTER: getelementptr inbounds %pair, ptr %p, i64 %[[I1]], i32 0
207-
; INTER: %[[I2:.+]] = or disjoint i64 %index, 2
208-
; INTER: getelementptr inbounds %pair, ptr %p, i64 %[[I2]], i32 0
209-
; INTER: %[[I3:.+]] = or disjoint i64 %index, 3
210-
; INTER: getelementptr inbounds %pair, ptr %p, i64 %[[I3]], i32 0
205+
; INTER: %[[G1:.+]] = getelementptr %pair, ptr %p, i64 %index
206+
; INTER: getelementptr i8, ptr %[[G1]], i64 8
207+
; INTER: %[[G2:.+]] = getelementptr %pair, ptr %p, i64 %index
208+
; INTER: getelementptr i8, ptr %[[G2]], i64 16
209+
; INTER: %[[G3:.+]] = getelementptr %pair, ptr %p, i64 %index
210+
; INTER: getelementptr i8, ptr %[[G3]], i64 24
211211
; INTER: br i1 {{.*}}, label %middle.block, label %vector.body
212212
;
213213
define void @predicated_store(ptr %p, i32 %x, i64 %n) {

0 commit comments

Comments
 (0)