Skip to content

Commit 53b49fa

Browse files
fhahnfhahn
committed
[LV] Use SCEV::getElementCount in selectEpilogueVectorizationFactor.
Follow-up to llvm#149789 to use getElementCount to compute the remaining iterations in selectEpilogueVectrizationFactor.
1 parent bd2b7eb commit 53b49fa

File tree

2 files changed

+128
-52
lines changed

2 files changed

+128
-52
lines changed

‎llvm/lib/Transforms/Vectorize/LoopVectorize.cpp‎

Lines changed: 10 additions & 12 deletions
Original file line numberDiff line numberDiff line change
@@ -4500,19 +4500,17 @@ VectorizationFactor LoopVectorizationPlanner::selectEpilogueVectorizationFactor(
45004500
Type *TCType = Legal->getWidestInductionType();
45014501
const SCEV *RemainingIterations = nullptr;
45024502
unsigned MaxTripCount = 0;
4503-
if (MainLoopVF.isFixed()) {
4504-
// TODO: extend to support scalable VFs.
4505-
const SCEV *TC = vputils::getSCEVExprForVPValue(
4506-
getPlanFor(MainLoopVF).getTripCount(), SE);
4507-
assert(!isa<SCEVCouldNotCompute>(TC) &&
4508-
"Trip count SCEV must be computable");
4509-
RemainingIterations = SE.getURemExpr(
4510-
TC, SE.getConstant(TCType, MainLoopVF.getFixedValue() * IC));
4511-
4512-
// No iterations left to process in the epilogue.
4513-
if (RemainingIterations->isZero())
4514-
return Result;
4503+
const SCEV *TC =
4504+
vputils::getSCEVExprForVPValue(getPlanFor(MainLoopVF).getTripCount(), SE);
4505+
assert(!isa<SCEVCouldNotCompute>(TC) && "Trip count SCEV must be computable");
4506+
RemainingIterations =
4507+
SE.getURemExpr(TC, SE.getElementCount(TCType, MainLoopVF * IC));
4508+
4509+
// No iterations left to process in the epilogue.
4510+
if (RemainingIterations->isZero())
4511+
return Result;
45154512

4513+
if (MainLoopVF.isFixed()) {
45164514
MaxTripCount = MainLoopVF.getFixedValue() * IC - 1;
45174515
if (SE.isKnownPredicate(CmpInst::ICMP_ULT, RemainingIterations,
45184516
SE.getConstant(TCType, MaxTripCount))) {

‎llvm/test/Transforms/LoopVectorize/AArch64/sve-epilog-vect.ll‎

Lines changed: 118 additions & 40 deletions
Original file line numberDiff line numberDiff line change
@@ -148,6 +148,72 @@ exit:
148148
ret void
149149
}
150150

151+
define void @main_vf_vscale_x_2_no_epi_iteration(ptr %A) #0 vscale_range(8, 8) {
152+
; CHECK-LABEL: @main_vf_vscale_x_2_no_epi_iteration(
153+
; CHECK-NEXT: entry:
154+
; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
155+
; CHECK-NEXT: [[TMP1:%.*]] = mul nuw i64 [[TMP0]], 4
156+
; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
157+
; CHECK: vector.ph:
158+
; CHECK-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
159+
; CHECK-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 4
160+
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]]
161+
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]]
162+
; CHECK-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
163+
; CHECK-NEXT: [[TMP5:%.*]] = mul nuw i64 [[TMP4]], 4
164+
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
165+
; CHECK: vector.body:
166+
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
167+
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP5]]
168+
; CHECK-NEXT: [[TMP11:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
169+
; CHECK-NEXT: br i1 [[TMP11]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
170+
; CHECK: middle.block:
171+
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]]
172+
; CHECK-NEXT: br label [[FOR_BODY:%.*]]
173+
; CHECK: for.body:
174+
;
175+
; CHECK-VF8-LABEL: @main_vf_vscale_x_2_no_epi_iteration(
176+
; CHECK-VF8-NEXT: iter.check:
177+
; CHECK-VF8-NEXT: br i1 false, label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]
178+
; CHECK-VF8: vector.main.loop.iter.check:
179+
; CHECK-VF8-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
180+
; CHECK-VF8-NEXT: [[TMP1:%.*]] = mul nuw i64 [[TMP0]], 4
181+
; CHECK-VF8-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 1024, [[TMP1]]
182+
; CHECK-VF8-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]
183+
; CHECK-VF8: vector.ph:
184+
; CHECK-VF8-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
185+
; CHECK-VF8-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 4
186+
; CHECK-VF8-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]]
187+
; CHECK-VF8-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]]
188+
; CHECK-VF8-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
189+
; CHECK-VF8-NEXT: [[TMP5:%.*]] = mul nuw i64 [[TMP4]], 4
190+
; CHECK-VF8-NEXT: br label [[VECTOR_BODY:%.*]]
191+
; CHECK-VF8: vector.body:
192+
; CHECK-VF8-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
193+
; CHECK-VF8-NEXT: [[INDEX_NEXT2]] = add nuw i64 [[INDEX1]], 8
194+
; CHECK-VF8-NEXT: [[TMP14:%.*]] = icmp eq i64 [[INDEX_NEXT2]], 1024
195+
; CHECK-VF8-NEXT: br i1 [[TMP14]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]
196+
; CHECK-VF8: vec.epilog.middle.block:
197+
; CHECK-VF8-NEXT: br i1 true, label [[EXIT]], label [[VEC_EPILOG_SCALAR_PH]]
198+
; CHECK-VF8: vec.epilog.scalar.ph:
199+
; CHECK-VF8-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 1024, [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[ITER_CHECK:%.*]] ]
200+
; CHECK-VF8-NEXT: br label [[FOR_BODY:%.*]]
201+
; CHECK-VF8: for.body:
202+
;
203+
entry:
204+
br label %for.body
205+
206+
for.body:
207+
%iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
208+
%arrayidx = getelementptr inbounds i64, ptr %A, i64 %iv
209+
store i64 1, ptr %arrayidx, align 1
210+
%iv.next = add nuw nsw i64 %iv, 1
211+
%exitcond = icmp ne i64 %iv.next, 1024
212+
br i1 %exitcond, label %for.body, label %exit
213+
214+
exit:
215+
ret void
216+
}
151217

152218
; DEBUG: LV: Checking a loop in 'main_vf_vscale_x_2'
153219
; DEBUG: Create Skeleton for epilogue vectorized loop (first pass)
@@ -163,106 +229,118 @@ exit:
163229
; fixed-width VF=8 for the epilogue if the vectors are known to be
164230
; sufficiently wide. This information can be deduced from vscale_range or
165231
; VScaleForTuning (set by mcpu/mtune).
166-
define void @main_vf_vscale_x_2(ptr %A) #0 vscale_range(8, 8) {
232+
define void @main_vf_vscale_x_2(ptr %A, i64 %n) #0 vscale_range(8, 8) {
167233
; CHECK-LABEL: @main_vf_vscale_x_2(
168234
; CHECK-NEXT: iter.check:
169-
; CHECK-NEXT: br i1 false, label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]
235+
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 8
236+
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]
170237
; CHECK: vector.main.loop.iter.check:
171238
; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
172239
; CHECK-NEXT: [[TMP1:%.*]] = mul nuw i64 [[TMP0]], 4
173-
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 1024, [[TMP1]]
174-
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]
240+
; CHECK-NEXT: [[MIN_ITERS_CHECK1:%.*]] = icmp ult i64 [[N]], [[TMP1]]
241+
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK1]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]
175242
; CHECK: vector.ph:
176243
; CHECK-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
177244
; CHECK-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 4
178-
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]]
179-
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]]
245+
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], [[TMP3]]
246+
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]
180247
; CHECK-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
181248
; CHECK-NEXT: [[TMP5:%.*]] = mul nuw i64 [[TMP4]], 4
182249
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
183250
; CHECK: vector.body:
184251
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
185252
; CHECK-NEXT: [[TMP12:%.*]] = getelementptr inbounds i64, ptr [[A:%.*]], i64 [[INDEX]]
253+
; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds i64, ptr [[TMP12]], i32 0
186254
; CHECK-NEXT: [[TMP15:%.*]] = call i64 @llvm.vscale.i64()
187255
; CHECK-NEXT: [[TMP16:%.*]] = mul nuw i64 [[TMP15]], 2
188256
; CHECK-NEXT: [[TMP17:%.*]] = getelementptr inbounds i64, ptr [[TMP12]], i64 [[TMP16]]
189-
; CHECK-NEXT: store <vscale x 2 x i64> splat (i64 1), ptr [[TMP12]], align 1
257+
; CHECK-NEXT: store <vscale x 2 x i64> splat (i64 1), ptr [[TMP14]], align 1
190258
; CHECK-NEXT: store <vscale x 2 x i64> splat (i64 1), ptr [[TMP17]], align 1
191259
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP5]]
192260
; CHECK-NEXT: [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
193-
; CHECK-NEXT: br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
261+
; CHECK-NEXT: br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP7:![0-9]+]]
194262
; CHECK: middle.block:
195-
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]]
196-
; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]]
263+
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
264+
; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH:%.*]]
197265
; CHECK: vec.epilog.iter.check:
198-
; CHECK-NEXT: [[N_VEC_REMAINING:%.*]] = sub i64 1024, [[N_VEC]]
266+
; CHECK-NEXT: [[N_VEC_REMAINING:%.*]] = sub i64 [[N]], [[N_VEC]]
199267
; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_VEC_REMAINING]], 8
200268
; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]]
201269
; CHECK: vec.epilog.ph:
202-
; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
203-
; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]]
270+
; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[SCALAR_PH]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
271+
; CHECK-NEXT: [[N_MOD_VF2:%.*]] = urem i64 [[N]], 8
272+
; CHECK-NEXT: [[N_VEC3:%.*]] = sub i64 [[N]], [[N_MOD_VF2]]
273+
; CHECK-NEXT: br label [[FOR_BODY:%.*]]
204274
; CHECK: vec.epilog.vector.body:
205-
; CHECK-NEXT: [[INDEX1:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT2:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
206-
; CHECK-NEXT: [[TMP20:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[INDEX1]]
207-
; CHECK-NEXT: store <8 x i64> splat (i64 1), ptr [[TMP20]], align 1
208-
; CHECK-NEXT: [[INDEX_NEXT2]] = add nuw i64 [[INDEX1]], 8
209-
; CHECK-NEXT: [[TMP22:%.*]] = icmp eq i64 [[INDEX_NEXT2]], 1024
210-
; CHECK-NEXT: br i1 [[TMP22]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]
275+
; CHECK-NEXT: [[INDEX4:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT5:%.*]], [[FOR_BODY]] ]
276+
; CHECK-NEXT: [[TMP19:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[INDEX4]]
277+
; CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds i64, ptr [[TMP19]], i32 0
278+
; CHECK-NEXT: store <8 x i64> splat (i64 1), ptr [[TMP13]], align 1
279+
; CHECK-NEXT: [[INDEX_NEXT5]] = add nuw i64 [[INDEX4]], 8
280+
; CHECK-NEXT: [[TMP20:%.*]] = icmp eq i64 [[INDEX_NEXT5]], [[N_VEC3]]
281+
; CHECK-NEXT: br i1 [[TMP20]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[FOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]
211282
; CHECK: vec.epilog.middle.block:
212-
; CHECK-NEXT: br i1 true, label [[EXIT]], label [[VEC_EPILOG_SCALAR_PH]]
283+
; CHECK-NEXT: [[CMP_N6:%.*]] = icmp eq i64 [[N]], [[N_VEC3]]
284+
; CHECK-NEXT: br i1 [[CMP_N6]], label [[EXIT]], label [[VEC_EPILOG_SCALAR_PH]]
213285
; CHECK: vec.epilog.scalar.ph:
214-
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 1024, [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[ITER_CHECK:%.*]] ]
215-
; CHECK-NEXT: br label [[FOR_BODY:%.*]]
286+
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC3]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], [[SCALAR_PH]] ], [ 0, [[ITER_CHECK:%.*]] ]
287+
; CHECK-NEXT: br label [[FOR_BODY1:%.*]]
216288
; CHECK: for.body:
217289
;
218290
; CHECK-VF8-LABEL: @main_vf_vscale_x_2(
219291
; CHECK-VF8-NEXT: iter.check:
220-
; CHECK-VF8-NEXT: br i1 false, label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]
292+
; CHECK-VF8-NEXT: [[MIN_ITERS_CHECK1:%.*]] = icmp ult i64 [[N:%.*]], 8
293+
; CHECK-VF8-NEXT: br i1 [[MIN_ITERS_CHECK1]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]
221294
; CHECK-VF8: vector.main.loop.iter.check:
222295
; CHECK-VF8-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
223296
; CHECK-VF8-NEXT: [[TMP1:%.*]] = mul nuw i64 [[TMP0]], 4
224-
; CHECK-VF8-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 1024, [[TMP1]]
297+
; CHECK-VF8-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], [[TMP1]]
225298
; CHECK-VF8-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]
226299
; CHECK-VF8: vector.ph:
227300
; CHECK-VF8-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
228301
; CHECK-VF8-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 4
229-
; CHECK-VF8-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]]
230-
; CHECK-VF8-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]]
302+
; CHECK-VF8-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], [[TMP3]]
303+
; CHECK-VF8-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]
231304
; CHECK-VF8-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
232305
; CHECK-VF8-NEXT: [[TMP5:%.*]] = mul nuw i64 [[TMP4]], 4
233306
; CHECK-VF8-NEXT: br label [[VECTOR_BODY:%.*]]
234307
; CHECK-VF8: vector.body:
235308
; CHECK-VF8-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
236309
; CHECK-VF8-NEXT: [[TMP12:%.*]] = getelementptr inbounds i64, ptr [[A:%.*]], i64 [[INDEX]]
310+
; CHECK-VF8-NEXT: [[TMP14:%.*]] = getelementptr inbounds i64, ptr [[TMP12]], i32 0
237311
; CHECK-VF8-NEXT: [[TMP15:%.*]] = call i64 @llvm.vscale.i64()
238312
; CHECK-VF8-NEXT: [[TMP16:%.*]] = mul nuw i64 [[TMP15]], 2
239313
; CHECK-VF8-NEXT: [[TMP17:%.*]] = getelementptr inbounds i64, ptr [[TMP12]], i64 [[TMP16]]
240-
; CHECK-VF8-NEXT: store <vscale x 2 x i64> splat (i64 1), ptr [[TMP12]], align 1
314+
; CHECK-VF8-NEXT: store <vscale x 2 x i64> splat (i64 1), ptr [[TMP14]], align 1
241315
; CHECK-VF8-NEXT: store <vscale x 2 x i64> splat (i64 1), ptr [[TMP17]], align 1
242316
; CHECK-VF8-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP5]]
243317
; CHECK-VF8-NEXT: [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
244-
; CHECK-VF8-NEXT: br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
318+
; CHECK-VF8-NEXT: br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]
245319
; CHECK-VF8: middle.block:
246-
; CHECK-VF8-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]]
320+
; CHECK-VF8-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
247321
; CHECK-VF8-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]]
248322
; CHECK-VF8: vec.epilog.iter.check:
249-
; CHECK-VF8-NEXT: [[N_VEC_REMAINING:%.*]] = sub i64 1024, [[N_VEC]]
323+
; CHECK-VF8-NEXT: [[N_VEC_REMAINING:%.*]] = sub i64 [[N]], [[N_VEC]]
250324
; CHECK-VF8-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_VEC_REMAINING]], 8
251325
; CHECK-VF8-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]]
252326
; CHECK-VF8: vec.epilog.ph:
253327
; CHECK-VF8-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
328+
; CHECK-VF8-NEXT: [[N_MOD_VF2:%.*]] = urem i64 [[N]], 8
329+
; CHECK-VF8-NEXT: [[N_VEC3:%.*]] = sub i64 [[N]], [[N_MOD_VF2]]
254330
; CHECK-VF8-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]]
255331
; CHECK-VF8: vec.epilog.vector.body:
256332
; CHECK-VF8-NEXT: [[INDEX1:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT2:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
257333
; CHECK-VF8-NEXT: [[TMP20:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[INDEX1]]
258-
; CHECK-VF8-NEXT: store <8 x i64> splat (i64 1), ptr [[TMP20]], align 1
334+
; CHECK-VF8-NEXT: [[TMP21:%.*]] = getelementptr inbounds i64, ptr [[TMP20]], i32 0
335+
; CHECK-VF8-NEXT: store <8 x i64> splat (i64 1), ptr [[TMP21]], align 1
259336
; CHECK-VF8-NEXT: [[INDEX_NEXT2]] = add nuw i64 [[INDEX1]], 8
260-
; CHECK-VF8-NEXT: [[TMP22:%.*]] = icmp eq i64 [[INDEX_NEXT2]], 1024
261-
; CHECK-VF8-NEXT: br i1 [[TMP22]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]
337+
; CHECK-VF8-NEXT: [[TMP19:%.*]] = icmp eq i64 [[INDEX_NEXT2]], [[N_VEC3]]
338+
; CHECK-VF8-NEXT: br i1 [[TMP19]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]]
262339
; CHECK-VF8: vec.epilog.middle.block:
263-
; CHECK-VF8-NEXT: br i1 true, label [[EXIT]], label [[VEC_EPILOG_SCALAR_PH]]
340+
; CHECK-VF8-NEXT: [[CMP_N6:%.*]] = icmp eq i64 [[N]], [[N_VEC3]]
341+
; CHECK-VF8-NEXT: br i1 [[CMP_N6]], label [[EXIT]], label [[VEC_EPILOG_SCALAR_PH]]
264342
; CHECK-VF8: vec.epilog.scalar.ph:
265-
; CHECK-VF8-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 1024, [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[ITER_CHECK:%.*]] ]
343+
; CHECK-VF8-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC3]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[ITER_CHECK:%.*]] ]
266344
; CHECK-VF8-NEXT: br label [[FOR_BODY:%.*]]
267345
; CHECK-VF8: for.body:
268346
;
@@ -274,7 +352,7 @@ for.body:
274352
%arrayidx = getelementptr inbounds i64, ptr %A, i64 %iv
275353
store i64 1, ptr %arrayidx, align 1
276354
%iv.next = add nuw nsw i64 %iv, 1
277-
%exitcond = icmp ne i64 %iv.next, 1024
355+
%exitcond = icmp ne i64 %iv.next, %n
278356
br i1 %exitcond, label %for.body, label %exit
279357

280358
exit:
@@ -313,7 +391,7 @@ define void @test_pr57912_pointer_induction(ptr %start) #0 {
313391
; CHECK-NEXT: store <vscale x 16 x i8> zeroinitializer, ptr [[TMP19]], align 1
314392
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP7]]
315393
; CHECK-NEXT: [[TMP20:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
316-
; CHECK-NEXT: br i1 [[TMP20]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]
394+
; CHECK-NEXT: br i1 [[TMP20]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]
317395
; CHECK: middle.block:
318396
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 10000, [[N_VEC]]
319397
; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]]
@@ -340,7 +418,7 @@ define void @test_pr57912_pointer_induction(ptr %start) #0 {
340418
; CHECK-NEXT: store <vscale x 8 x i8> zeroinitializer, ptr [[TMP28]], align 1
341419
; CHECK-NEXT: [[INDEX_NEXT8]] = add nuw i64 [[INDEX7]], [[TMP26]]
342420
; CHECK-NEXT: [[TMP30:%.*]] = icmp eq i64 [[INDEX_NEXT8]], [[N_VEC3]]
343-
; CHECK-NEXT: br i1 [[TMP30]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]]
421+
; CHECK-NEXT: br i1 [[TMP30]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP11:![0-9]+]]
344422
; CHECK: vec.epilog.middle.block:
345423
; CHECK-NEXT: [[CMP_N6:%.*]] = icmp eq i64 10000, [[N_VEC3]]
346424
; CHECK-NEXT: br i1 [[CMP_N6]], label [[EXIT]], label [[VEC_EPILOG_SCALAR_PH]]
@@ -376,7 +454,7 @@ define void @test_pr57912_pointer_induction(ptr %start) #0 {
376454
; CHECK-VF8-NEXT: store <vscale x 16 x i8> zeroinitializer, ptr [[TMP17]], align 1
377455
; CHECK-VF8-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP5]]
378456
; CHECK-VF8-NEXT: [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
379-
; CHECK-VF8-NEXT: br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]
457+
; CHECK-VF8-NEXT: br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP11:![0-9]+]]
380458
; CHECK-VF8: middle.block:
381459
; CHECK-VF8-NEXT: [[CMP_N:%.*]] = icmp eq i64 10000, [[N_VEC]]
382460
; CHECK-VF8-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]]
@@ -395,7 +473,7 @@ define void @test_pr57912_pointer_induction(ptr %start) #0 {
395473
; CHECK-VF8-NEXT: store <8 x i8> zeroinitializer, ptr [[TMP20]], align 1
396474
; CHECK-VF8-NEXT: [[INDEX_NEXT4]] = add nuw i64 [[INDEX3]], 8
397475
; CHECK-VF8-NEXT: [[TMP22:%.*]] = icmp eq i64 [[INDEX_NEXT4]], 10000
398-
; CHECK-VF8-NEXT: br i1 [[TMP22]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]]
476+
; CHECK-VF8-NEXT: br i1 [[TMP22]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP12:![0-9]+]]
399477
; CHECK-VF8: vec.epilog.middle.block:
400478
; CHECK-VF8-NEXT: br i1 true, label [[EXIT]], label [[VEC_EPILOG_SCALAR_PH]]
401479
; CHECK-VF8: vec.epilog.scalar.ph:

0 commit comments

Comments
 (0)