Skip to content

Commit 8132c4f

Browse files
fhahnfhahn
committed
[VPlan] Also introduce broadcasts for live-ins used in vec preheader.
Slightly generalize materializeLiveInBroadcasts to also introduce broadcasts for live-ins used in the vector preheader. This should cover all live-ins. If the live-in is used in the vector preheader, insert the broadcast at the beginning of the block.
1 parent 74eba97 commit 8132c4f

19 files changed

+101
-97
lines changed

‎llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp‎

Lines changed: 21 additions & 17 deletions
Original file line numberDiff line numberDiff line change
@@ -2184,32 +2184,36 @@ void VPlanTransforms::materializeLiveInBroadcasts(VPlan &Plan) {
21842184
if (Plan.hasScalarVFOnly())
21852185
return;
21862186

2187+
#ifndef NDEBUG
21872188
VPDominatorTree VPDT;
21882189
VPDT.recalculate(Plan);
2190+
#endif
2191+
21892192
auto *VectorPreheader = Plan.getVectorPreheader();
2190-
VPBuilder Builder(VectorPreheader);
21912193
for (VPValue *LiveIn : Plan.getLiveIns()) {
21922194
if (all_of(LiveIn->users(),
2193-
[LiveIn](VPUser *U) {
2194-
return cast<VPRecipeBase>(U)->usesScalars(LiveIn);
2195-
}) ||
2195+
[LiveIn](VPUser *U) { return U->usesScalars(LiveIn); }) ||
21962196
!LiveIn->getLiveInIRValue() ||
21972197
isa<Constant>(LiveIn->getLiveInIRValue()))
21982198
continue;
21992199

2200-
// Add explicit broadcast if the vector preheader dominates all users.
2201-
// TODO: Find valid insert point for all users.
2202-
if (all_of(LiveIn->users(), [&VPDT, VectorPreheader](VPUser *U) {
2203-
return VectorPreheader != cast<VPRecipeBase>(U)->getParent() &&
2204-
VPDT.dominates(VectorPreheader,
2205-
cast<VPRecipeBase>(U)->getParent());
2206-
})) {
2207-
auto *Broadcast =
2208-
Builder.createNaryOp(VPInstruction::Broadcast, {LiveIn});
2209-
LiveIn->replaceUsesWithIf(Broadcast, [LiveIn, Broadcast](VPUser &U,
2210-
unsigned Idx) {
2211-
return Broadcast != &U && !cast<VPRecipeBase>(&U)->usesScalars(LiveIn);
2212-
});
2200+
// Add explicit broadcast at the insert point that dominates all users.
2201+
VPBasicBlock *HoistBlock = VectorPreheader;
2202+
VPBasicBlock::iterator HoistPoint = VectorPreheader->end();
2203+
for (VPUser *User : LiveIn->users()) {
2204+
if (cast<VPRecipeBase>(User)->getParent() == VectorPreheader)
2205+
HoistPoint = HoistBlock->begin();
2206+
else
2207+
assert(VPDT.dominates(VectorPreheader,
2208+
cast<VPRecipeBase>(User)->getParent()) &&
2209+
"All users must be in the vector preheader or dominated by it");
22132210
}
2211+
2212+
VPBuilder Builder(cast<VPBasicBlock>(HoistBlock), HoistPoint);
2213+
auto *Broadcast = Builder.createNaryOp(VPInstruction::Broadcast, {LiveIn});
2214+
LiveIn->replaceUsesWithIf(
2215+
Broadcast, [LiveIn, Broadcast](VPUser &U, unsigned Idx) {
2216+
return Broadcast != &U && !U.usesScalars(LiveIn);
2217+
});
22142218
}
22152219
}

‎llvm/test/Transforms/LoopVectorize/AArch64/force-target-instruction-cost.ll‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -205,10 +205,10 @@ define void @test_exit_branch_cost(ptr %dst, ptr noalias %x.ptr, ptr noalias %y.
205205
; CHECK: [[VECTOR_PH]]:
206206
; CHECK-NEXT: [[BROADCAST_SPLATINSERT40:%.*]] = insertelement <2 x i1> poison, i1 [[C_3]], i64 0
207207
; CHECK-NEXT: [[BROADCAST_SPLAT41:%.*]] = shufflevector <2 x i1> [[BROADCAST_SPLATINSERT40]], <2 x i1> poison, <2 x i32> zeroinitializer
208-
; CHECK-NEXT: [[TMP2:%.*]] = select i1 [[C_4]], <2 x i1> [[BROADCAST_SPLAT41]], <2 x i1> zeroinitializer
209-
; CHECK-NEXT: [[TMP11:%.*]] = xor <2 x i1> [[TMP2]], splat (i1 true)
210208
; CHECK-NEXT: [[BROADCAST_SPLATINSERT56:%.*]] = insertelement <2 x i1> poison, i1 [[C_4]], i64 0
211209
; CHECK-NEXT: [[BROADCAST_SPLAT57:%.*]] = shufflevector <2 x i1> [[BROADCAST_SPLATINSERT56]], <2 x i1> poison, <2 x i32> zeroinitializer
210+
; CHECK-NEXT: [[TMP1:%.*]] = select i1 [[C_4]], <2 x i1> [[BROADCAST_SPLAT41]], <2 x i1> zeroinitializer
211+
; CHECK-NEXT: [[TMP11:%.*]] = xor <2 x i1> [[TMP1]], splat (i1 true)
212212
; CHECK-NEXT: [[TMP33:%.*]] = xor <2 x i1> [[BROADCAST_SPLAT57]], splat (i1 true)
213213
; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
214214
; CHECK: [[VECTOR_BODY]]:

‎llvm/test/Transforms/LoopVectorize/AArch64/induction-costs-sve.ll‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -112,14 +112,14 @@ define void @iv_casts(ptr %dst, ptr %src, i32 %x, i64 %N) #0 {
112112
; PRED-NEXT: [[N_VEC:%.*]] = sub i64 [[N_RND_UP]], [[N_MOD_VF]]
113113
; PRED-NEXT: [[TMP9:%.*]] = call i64 @llvm.vscale.i64()
114114
; PRED-NEXT: [[TMP10:%.*]] = mul i64 [[TMP9]], 8
115+
; PRED-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 8 x i32> poison, i32 [[X]], i64 0
116+
; PRED-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 8 x i32> [[BROADCAST_SPLATINSERT]], <vscale x 8 x i32> poison, <vscale x 8 x i32> zeroinitializer
115117
; PRED-NEXT: [[TMP11:%.*]] = call i64 @llvm.vscale.i64()
116118
; PRED-NEXT: [[TMP12:%.*]] = mul i64 [[TMP11]], 8
117119
; PRED-NEXT: [[TMP13:%.*]] = sub i64 [[TMP0]], [[TMP12]]
118120
; PRED-NEXT: [[TMP14:%.*]] = icmp ugt i64 [[TMP0]], [[TMP12]]
119121
; PRED-NEXT: [[TMP15:%.*]] = select i1 [[TMP14]], i64 [[TMP13]], i64 0
120122
; PRED-NEXT: [[ACTIVE_LANE_MASK_ENTRY:%.*]] = call <vscale x 8 x i1> @llvm.get.active.lane.mask.nxv8i1.i64(i64 0, i64 [[TMP0]])
121-
; PRED-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 8 x i32> poison, i32 [[X]], i64 0
122-
; PRED-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 8 x i32> [[BROADCAST_SPLATINSERT]], <vscale x 8 x i32> poison, <vscale x 8 x i32> zeroinitializer
123123
; PRED-NEXT: [[TMP16:%.*]] = trunc <vscale x 8 x i32> [[BROADCAST_SPLAT]] to <vscale x 8 x i16>
124124
; PRED-NEXT: br label [[VECTOR_BODY:%.*]]
125125
; PRED: vector.body:

‎llvm/test/Transforms/LoopVectorize/AArch64/loop-vectorization-factors.ll‎

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -472,9 +472,9 @@ define void @add_e(ptr noalias nocapture readonly %p, ptr noalias nocapture %q,
472472
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]]
473473
; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <16 x i32> poison, i32 [[CONV13]], i64 0
474474
; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <16 x i32> [[BROADCAST_SPLATINSERT]], <16 x i32> poison, <16 x i32> zeroinitializer
475-
; CHECK-NEXT: [[TMP1:%.*]] = trunc <16 x i32> [[BROADCAST_SPLAT]] to <16 x i8>
476475
; CHECK-NEXT: [[BROADCAST_SPLATINSERT2:%.*]] = insertelement <16 x i32> poison, i32 [[CONV11]], i64 0
477476
; CHECK-NEXT: [[BROADCAST_SPLAT3:%.*]] = shufflevector <16 x i32> [[BROADCAST_SPLATINSERT2]], <16 x i32> poison, <16 x i32> zeroinitializer
477+
; CHECK-NEXT: [[TMP1:%.*]] = trunc <16 x i32> [[BROADCAST_SPLAT]] to <16 x i8>
478478
; CHECK-NEXT: [[TMP2:%.*]] = trunc <16 x i32> [[BROADCAST_SPLAT3]] to <16 x i8>
479479
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
480480
; CHECK: vector.body:
@@ -510,9 +510,9 @@ define void @add_e(ptr noalias nocapture readonly %p, ptr noalias nocapture %q,
510510
; CHECK-NEXT: [[N_VEC5:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF4]]
511511
; CHECK-NEXT: [[BROADCAST_SPLATINSERT6:%.*]] = insertelement <4 x i32> poison, i32 [[CONV13]], i64 0
512512
; CHECK-NEXT: [[BROADCAST_SPLAT7:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT6]], <4 x i32> poison, <4 x i32> zeroinitializer
513-
; CHECK-NEXT: [[TMP17:%.*]] = trunc <4 x i32> [[BROADCAST_SPLAT7]] to <4 x i8>
514513
; CHECK-NEXT: [[BROADCAST_SPLATINSERT8:%.*]] = insertelement <4 x i32> poison, i32 [[CONV11]], i64 0
515514
; CHECK-NEXT: [[BROADCAST_SPLAT9:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT8]], <4 x i32> poison, <4 x i32> zeroinitializer
515+
; CHECK-NEXT: [[TMP17:%.*]] = trunc <4 x i32> [[BROADCAST_SPLAT7]] to <4 x i8>
516516
; CHECK-NEXT: [[TMP18:%.*]] = trunc <4 x i32> [[BROADCAST_SPLAT9]] to <4 x i8>
517517
; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]]
518518
; CHECK: vec.epilog.vector.body:
@@ -620,9 +620,9 @@ define void @add_f(ptr noalias nocapture readonly %p, ptr noalias nocapture %q,
620620
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]]
621621
; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <16 x i32> poison, i32 [[CONV13]], i64 0
622622
; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <16 x i32> [[BROADCAST_SPLATINSERT]], <16 x i32> poison, <16 x i32> zeroinitializer
623-
; CHECK-NEXT: [[TMP1:%.*]] = trunc <16 x i32> [[BROADCAST_SPLAT]] to <16 x i8>
624623
; CHECK-NEXT: [[BROADCAST_SPLATINSERT2:%.*]] = insertelement <16 x i32> poison, i32 [[CONV11]], i64 0
625624
; CHECK-NEXT: [[BROADCAST_SPLAT3:%.*]] = shufflevector <16 x i32> [[BROADCAST_SPLATINSERT2]], <16 x i32> poison, <16 x i32> zeroinitializer
625+
; CHECK-NEXT: [[TMP1:%.*]] = trunc <16 x i32> [[BROADCAST_SPLAT]] to <16 x i8>
626626
; CHECK-NEXT: [[TMP2:%.*]] = trunc <16 x i32> [[BROADCAST_SPLAT3]] to <16 x i8>
627627
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
628628
; CHECK: vector.body:
@@ -660,9 +660,9 @@ define void @add_f(ptr noalias nocapture readonly %p, ptr noalias nocapture %q,
660660
; CHECK-NEXT: [[N_VEC5:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF4]]
661661
; CHECK-NEXT: [[BROADCAST_SPLATINSERT6:%.*]] = insertelement <4 x i32> poison, i32 [[CONV13]], i64 0
662662
; CHECK-NEXT: [[BROADCAST_SPLAT7:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT6]], <4 x i32> poison, <4 x i32> zeroinitializer
663-
; CHECK-NEXT: [[TMP19:%.*]] = trunc <4 x i32> [[BROADCAST_SPLAT7]] to <4 x i8>
664663
; CHECK-NEXT: [[BROADCAST_SPLATINSERT8:%.*]] = insertelement <4 x i32> poison, i32 [[CONV11]], i64 0
665664
; CHECK-NEXT: [[BROADCAST_SPLAT9:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT8]], <4 x i32> poison, <4 x i32> zeroinitializer
665+
; CHECK-NEXT: [[TMP19:%.*]] = trunc <4 x i32> [[BROADCAST_SPLAT7]] to <4 x i8>
666666
; CHECK-NEXT: [[TMP20:%.*]] = trunc <4 x i32> [[BROADCAST_SPLAT9]] to <4 x i8>
667667
; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]]
668668
; CHECK: vec.epilog.vector.body:

‎llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-epilogue.ll‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -160,9 +160,9 @@ define void @dotp_small_epilogue_vf(i64 %idx.neg, i8 %a) #1 {
160160
; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP6]], [[WHILE_BODY]] ], [ 0, [[ENTRY]] ]
161161
; CHECK-NEXT: [[N_MOD_VF4:%.*]] = urem i64 [[TMP0]], 4
162162
; CHECK-NEXT: [[N_VEC5:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF4]]
163-
; CHECK-NEXT: [[IND_END:%.*]] = add i64 [[IDX_NEG]], [[N_VEC5]]
164163
; CHECK-NEXT: [[BROADCAST_SPLATINSERT6:%.*]] = insertelement <4 x i8> poison, i8 [[A]], i64 0
165164
; CHECK-NEXT: [[BROADCAST_SPLAT7:%.*]] = shufflevector <4 x i8> [[BROADCAST_SPLATINSERT6]], <4 x i8> poison, <4 x i32> zeroinitializer
165+
; CHECK-NEXT: [[IND_END:%.*]] = add i64 [[IDX_NEG]], [[N_VEC5]]
166166
; CHECK-NEXT: [[TMP8:%.*]] = sext <4 x i8> [[BROADCAST_SPLAT7]] to <4 x i32>
167167
; CHECK-NEXT: [[TMP10:%.*]] = insertelement <4 x i32> zeroinitializer, i32 [[BC_MERGE_RDX]], i32 0
168168
; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]]

‎llvm/test/Transforms/LoopVectorize/AArch64/reduction-recurrence-costs-sve.ll‎

Lines changed: 8 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -61,15 +61,15 @@ define i32 @chained_recurrences(i32 %x, i64 %y, ptr %src.1, i32 %z, ptr %src.2)
6161
; VSCALEFORTUNING2-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]]
6262
; VSCALEFORTUNING2-NEXT: [[TMP5:%.*]] = call i64 @llvm.vscale.i64()
6363
; VSCALEFORTUNING2-NEXT: [[TMP6:%.*]] = mul i64 [[TMP5]], 8
64-
; VSCALEFORTUNING2-NEXT: [[TMP7:%.*]] = add i64 [[Y]], 1
65-
; VSCALEFORTUNING2-NEXT: [[TMP8:%.*]] = getelementptr i32, ptr [[SRC_1]], i64 [[TMP7]]
64+
; VSCALEFORTUNING2-NEXT: [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <vscale x 4 x i32> poison, i32 [[Z]], i64 0
65+
; VSCALEFORTUNING2-NEXT: [[BROADCAST_SPLAT2:%.*]] = shufflevector <vscale x 4 x i32> [[BROADCAST_SPLATINSERT1]], <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
6666
; VSCALEFORTUNING2-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 4 x i32> poison, i32 [[X]], i64 0
6767
; VSCALEFORTUNING2-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 4 x i32> [[BROADCAST_SPLATINSERT]], <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
68+
; VSCALEFORTUNING2-NEXT: [[TMP7:%.*]] = add i64 [[Y]], 1
69+
; VSCALEFORTUNING2-NEXT: [[TMP8:%.*]] = getelementptr i32, ptr [[SRC_1]], i64 [[TMP7]]
6870
; VSCALEFORTUNING2-NEXT: [[TMP9:%.*]] = lshr <vscale x 4 x i32> [[BROADCAST_SPLAT]], splat (i32 1)
6971
; VSCALEFORTUNING2-NEXT: [[TMP10:%.*]] = shl <vscale x 4 x i32> [[BROADCAST_SPLAT]], splat (i32 1)
7072
; VSCALEFORTUNING2-NEXT: [[TMP11:%.*]] = or <vscale x 4 x i32> [[TMP9]], [[TMP10]]
71-
; VSCALEFORTUNING2-NEXT: [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <vscale x 4 x i32> poison, i32 [[Z]], i64 0
72-
; VSCALEFORTUNING2-NEXT: [[BROADCAST_SPLAT2:%.*]] = shufflevector <vscale x 4 x i32> [[BROADCAST_SPLATINSERT1]], <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
7373
; VSCALEFORTUNING2-NEXT: [[TMP12:%.*]] = or <vscale x 4 x i32> [[BROADCAST_SPLAT2]], [[BROADCAST_SPLAT]]
7474
; VSCALEFORTUNING2-NEXT: [[TMP13:%.*]] = and <vscale x 4 x i32> [[TMP12]], splat (i32 1)
7575
; VSCALEFORTUNING2-NEXT: [[TMP14:%.*]] = xor <vscale x 4 x i32> [[TMP13]], splat (i32 1)
@@ -191,6 +191,10 @@ define i32 @chained_recurrences(i32 %x, i64 %y, ptr %src.1, i32 %z, ptr %src.2)
191191
; PRED-NEXT: [[N_VEC:%.*]] = sub i64 [[N_RND_UP]], [[N_MOD_VF]]
192192
; PRED-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
193193
; PRED-NEXT: [[TMP5:%.*]] = mul i64 [[TMP4]], 4
194+
; PRED-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 4 x i32> poison, i32 [[Z]], i64 0
195+
; PRED-NEXT: [[BROADCAST_SPLAT2:%.*]] = shufflevector <vscale x 4 x i32> [[BROADCAST_SPLATINSERT]], <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
196+
; PRED-NEXT: [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <vscale x 4 x i32> poison, i32 [[X]], i64 0
197+
; PRED-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 4 x i32> [[BROADCAST_SPLATINSERT1]], <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
194198
; PRED-NEXT: [[TMP6:%.*]] = call i64 @llvm.vscale.i64()
195199
; PRED-NEXT: [[TMP7:%.*]] = mul i64 [[TMP6]], 4
196200
; PRED-NEXT: [[TMP8:%.*]] = sub i64 [[TMP0]], [[TMP7]]
@@ -199,13 +203,9 @@ define i32 @chained_recurrences(i32 %x, i64 %y, ptr %src.1, i32 %z, ptr %src.2)
199203
; PRED-NEXT: [[ACTIVE_LANE_MASK_ENTRY:%.*]] = call <vscale x 4 x i1> @llvm.get.active.lane.mask.nxv4i1.i64(i64 0, i64 [[TMP0]])
200204
; PRED-NEXT: [[TMP11:%.*]] = add i64 [[Y]], 1
201205
; PRED-NEXT: [[TMP12:%.*]] = getelementptr i32, ptr [[SRC_1]], i64 [[TMP11]]
202-
; PRED-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 4 x i32> poison, i32 [[X]], i64 0
203-
; PRED-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 4 x i32> [[BROADCAST_SPLATINSERT]], <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
204206
; PRED-NEXT: [[TMP13:%.*]] = lshr <vscale x 4 x i32> [[BROADCAST_SPLAT]], splat (i32 1)
205207
; PRED-NEXT: [[TMP14:%.*]] = shl <vscale x 4 x i32> [[BROADCAST_SPLAT]], splat (i32 1)
206208
; PRED-NEXT: [[TMP15:%.*]] = or <vscale x 4 x i32> [[TMP13]], [[TMP14]]
207-
; PRED-NEXT: [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <vscale x 4 x i32> poison, i32 [[Z]], i64 0
208-
; PRED-NEXT: [[BROADCAST_SPLAT2:%.*]] = shufflevector <vscale x 4 x i32> [[BROADCAST_SPLATINSERT1]], <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
209209
; PRED-NEXT: [[TMP16:%.*]] = or <vscale x 4 x i32> [[BROADCAST_SPLAT2]], [[BROADCAST_SPLAT]]
210210
; PRED-NEXT: [[TMP17:%.*]] = and <vscale x 4 x i32> [[TMP16]], splat (i32 1)
211211
; PRED-NEXT: [[TMP18:%.*]] = xor <vscale x 4 x i32> [[TMP17]], splat (i32 1)

‎llvm/test/Transforms/LoopVectorize/AArch64/store-costs-sve.ll‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -243,9 +243,9 @@ define void @trunc_store(ptr %dst, ptr %src, i16 %x) #1 {
243243
; PRED-NEXT: [[N_VEC:%.*]] = sub i64 [[N_RND_UP]], [[N_MOD_VF]]
244244
; PRED-NEXT: [[TMP3:%.*]] = call i64 @llvm.vscale.i64()
245245
; PRED-NEXT: [[TMP4:%.*]] = mul i64 [[TMP3]], 2
246-
; PRED-NEXT: [[ACTIVE_LANE_MASK_ENTRY:%.*]] = call <vscale x 2 x i1> @llvm.get.active.lane.mask.nxv2i1.i64(i64 0, i64 1000)
247246
; PRED-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 2 x i16> poison, i16 [[X]], i64 0
248247
; PRED-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 2 x i16> [[BROADCAST_SPLATINSERT]], <vscale x 2 x i16> poison, <vscale x 2 x i32> zeroinitializer
248+
; PRED-NEXT: [[ACTIVE_LANE_MASK_ENTRY:%.*]] = call <vscale x 2 x i1> @llvm.get.active.lane.mask.nxv2i1.i64(i64 0, i64 1000)
249249
; PRED-NEXT: [[TMP11:%.*]] = trunc <vscale x 2 x i16> [[BROADCAST_SPLAT]] to <vscale x 2 x i8>
250250
; PRED-NEXT: br label [[VECTOR_BODY:%.*]]
251251
; PRED: vector.body:

‎llvm/test/Transforms/LoopVectorize/RISCV/blend-any-of-reduction-cost.ll‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -73,11 +73,11 @@ define i32 @any_of_reduction_used_in_blend_with_multiple_phis(ptr %src, i64 %N,
7373
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]
7474
; CHECK-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
7575
; CHECK-NEXT: [[TMP5:%.*]] = mul i64 [[TMP4]], 2
76+
; CHECK-NEXT: [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <vscale x 2 x i1> poison, i1 [[C_1]], i64 0
77+
; CHECK-NEXT: [[BROADCAST_SPLAT2:%.*]] = shufflevector <vscale x 2 x i1> [[BROADCAST_SPLATINSERT1]], <vscale x 2 x i1> poison, <vscale x 2 x i32> zeroinitializer
7678
; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 2 x i1> poison, i1 [[C_0]], i64 0
7779
; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 2 x i1> [[BROADCAST_SPLATINSERT]], <vscale x 2 x i1> poison, <vscale x 2 x i32> zeroinitializer
7880
; CHECK-NEXT: [[TMP6:%.*]] = xor <vscale x 2 x i1> [[BROADCAST_SPLAT]], splat (i1 true)
79-
; CHECK-NEXT: [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <vscale x 2 x i1> poison, i1 [[C_1]], i64 0
80-
; CHECK-NEXT: [[BROADCAST_SPLAT2:%.*]] = shufflevector <vscale x 2 x i1> [[BROADCAST_SPLATINSERT1]], <vscale x 2 x i1> poison, <vscale x 2 x i32> zeroinitializer
8181
; CHECK-NEXT: [[TMP7:%.*]] = xor <vscale x 2 x i1> [[BROADCAST_SPLAT2]], splat (i1 true)
8282
; CHECK-NEXT: [[TMP8:%.*]] = select <vscale x 2 x i1> [[TMP6]], <vscale x 2 x i1> [[TMP7]], <vscale x 2 x i1> zeroinitializer
8383
; CHECK-NEXT: [[BROADCAST_SPLATINSERT3:%.*]] = insertelement <vscale x 2 x ptr> poison, ptr [[SRC]], i64 0

‎llvm/test/Transforms/LoopVectorize/RISCV/dead-ops-cost.ll‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -328,9 +328,9 @@ define void @test_phi_in_latch_redundant(ptr %dst, i32 %a) {
328328
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 37, [[N_MOD_VF]]
329329
; CHECK-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
330330
; CHECK-NEXT: [[TMP5:%.*]] = mul i64 [[TMP4]], 2
331-
; CHECK-NEXT: [[IND_END:%.*]] = mul i64 [[N_VEC]], 9
332331
; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 2 x i32> poison, i32 [[A]], i64 0
333332
; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 2 x i32> [[BROADCAST_SPLATINSERT]], <vscale x 2 x i32> poison, <vscale x 2 x i32> zeroinitializer
333+
; CHECK-NEXT: [[TMP7:%.*]] = mul i64 [[N_VEC]], 9
334334
; CHECK-NEXT: [[TMP10:%.*]] = xor <vscale x 2 x i32> [[BROADCAST_SPLAT]], splat (i32 -1)
335335
; CHECK-NEXT: [[TMP6:%.*]] = call <vscale x 2 x i64> @llvm.stepvector.nxv2i64()
336336
; CHECK-NEXT: [[TMP8:%.*]] = mul <vscale x 2 x i64> [[TMP6]], splat (i64 9)
@@ -352,7 +352,7 @@ define void @test_phi_in_latch_redundant(ptr %dst, i32 %a) {
352352
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 37, [[N_VEC]]
353353
; CHECK-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]]
354354
; CHECK: [[SCALAR_PH]]:
355-
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
355+
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[TMP7]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
356356
; CHECK-NEXT: br label %[[LOOP_HEADER:.*]]
357357
; CHECK: [[LOOP_HEADER]]:
358358
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP_LATCH:.*]] ]

‎llvm/test/Transforms/LoopVectorize/RISCV/illegal-type.ll‎

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -107,10 +107,10 @@ define void @uniform_store_i1(ptr noalias %dst, ptr noalias %start, i64 %N) {
107107
; CHECK: vector.ph:
108108
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP0]], 32
109109
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]]
110-
; CHECK-NEXT: [[TMP1:%.*]] = mul i64 [[N_VEC]], 8
111-
; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[START:%.*]], i64 [[TMP1]]
112-
; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <32 x ptr> poison, ptr [[START]], i64 0
110+
; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <32 x ptr> poison, ptr [[START:%.*]], i64 0
113111
; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <32 x ptr> [[BROADCAST_SPLATINSERT]], <32 x ptr> poison, <32 x i32> zeroinitializer
112+
; CHECK-NEXT: [[TMP1:%.*]] = mul i64 [[N_VEC]], 8
113+
; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP1]]
114114
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
115115
; CHECK: vector.body:
116116
; CHECK-NEXT: [[POINTER_PHI:%.*]] = phi ptr [ [[START]], [[VECTOR_PH]] ], [ [[PTR_IND:%.*]], [[VECTOR_BODY]] ]

0 commit comments

Comments
 (0)