Skip to content

Commit bce3bfc

Browse files
fhahnfhahn
committed
[LV] Add another epilogue test with an AnyOfReduction of i1.
Additional test case from #78304.
1 parent eea81aa commit bce3bfc

File tree

1 file changed

+149
-0
lines changed

1 file changed

+149
-0
lines changed

‎llvm/test/Transforms/LoopVectorize/epilog-vectorization-any-of-reductions.ll

Lines changed: 149 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -210,6 +210,152 @@ exit:
210210
; uselistorder directives
211211
uselistorder i1 %sel, { 1, 0 }
212212
}
213+
214+
define i1 @any_of_reduction_i1_epilog2(ptr %start, ptr %end, i64 %x) {
215+
; CHECK-LABEL: define i1 @any_of_reduction_i1_epilog2(
216+
; CHECK-SAME: ptr [[START:%.*]], ptr [[END:%.*]], i64 [[X:%.*]]) {
217+
; CHECK-NEXT: iter.check:
218+
; CHECK-NEXT: [[START2:%.*]] = ptrtoint ptr [[START]] to i64
219+
; CHECK-NEXT: [[END1:%.*]] = ptrtoint ptr [[END]] to i64
220+
; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[END1]], -16
221+
; CHECK-NEXT: [[TMP1:%.*]] = sub i64 [[TMP0]], [[START2]]
222+
; CHECK-NEXT: [[TMP2:%.*]] = lshr i64 [[TMP1]], 4
223+
; CHECK-NEXT: [[TMP3:%.*]] = add nuw nsw i64 [[TMP2]], 1
224+
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP3]], 4
225+
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]
226+
; CHECK: vector.main.loop.iter.check:
227+
; CHECK-NEXT: [[MIN_ITERS_CHECK3:%.*]] = icmp ult i64 [[TMP3]], 4
228+
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK3]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]
229+
; CHECK: vector.ph:
230+
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP3]], 4
231+
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP3]], [[N_MOD_VF]]
232+
; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[X]], i64 0
233+
; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer
234+
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
235+
; CHECK: vector.body:
236+
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
237+
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i1> [ <i1 true, i1 true, i1 true, i1 true>, [[VECTOR_PH]] ], [ [[TMP21:%.*]], [[VECTOR_BODY]] ]
238+
; CHECK-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 16
239+
; CHECK-NEXT: [[TMP4:%.*]] = add i64 [[OFFSET_IDX]], 0
240+
; CHECK-NEXT: [[TMP5:%.*]] = add i64 [[OFFSET_IDX]], 16
241+
; CHECK-NEXT: [[TMP6:%.*]] = add i64 [[OFFSET_IDX]], 32
242+
; CHECK-NEXT: [[TMP7:%.*]] = add i64 [[OFFSET_IDX]], 48
243+
; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP4]]
244+
; CHECK-NEXT: [[NEXT_GEP4:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP5]]
245+
; CHECK-NEXT: [[NEXT_GEP5:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP6]]
246+
; CHECK-NEXT: [[NEXT_GEP6:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP7]]
247+
; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds i8, ptr [[NEXT_GEP]], i64 8
248+
; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds i8, ptr [[NEXT_GEP4]], i64 8
249+
; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds i8, ptr [[NEXT_GEP5]], i64 8
250+
; CHECK-NEXT: [[TMP11:%.*]] = getelementptr inbounds i8, ptr [[NEXT_GEP6]], i64 8
251+
; CHECK-NEXT: [[TMP12:%.*]] = load i64, ptr [[TMP8]], align 8
252+
; CHECK-NEXT: [[TMP13:%.*]] = load i64, ptr [[TMP9]], align 8
253+
; CHECK-NEXT: [[TMP14:%.*]] = load i64, ptr [[TMP10]], align 8
254+
; CHECK-NEXT: [[TMP15:%.*]] = load i64, ptr [[TMP11]], align 8
255+
; CHECK-NEXT: [[TMP16:%.*]] = insertelement <4 x i64> poison, i64 [[TMP12]], i32 0
256+
; CHECK-NEXT: [[TMP17:%.*]] = insertelement <4 x i64> [[TMP16]], i64 [[TMP13]], i32 1
257+
; CHECK-NEXT: [[TMP18:%.*]] = insertelement <4 x i64> [[TMP17]], i64 [[TMP14]], i32 2
258+
; CHECK-NEXT: [[TMP19:%.*]] = insertelement <4 x i64> [[TMP18]], i64 [[TMP15]], i32 3
259+
; CHECK-NEXT: [[TMP20:%.*]] = icmp eq <4 x i64> [[TMP19]], [[BROADCAST_SPLAT]]
260+
; CHECK-NEXT: [[TMP21]] = select <4 x i1> [[TMP20]], <4 x i1> [[VEC_PHI]], <4 x i1> zeroinitializer
261+
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
262+
; CHECK-NEXT: [[TMP22:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
263+
; CHECK-NEXT: br i1 [[TMP22]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]
264+
; CHECK: middle.block:
265+
; CHECK-NEXT: [[RDX_SELECT_CMP:%.*]] = icmp ne <4 x i1> [[TMP21]], <i1 true, i1 true, i1 true, i1 true>
266+
; CHECK-NEXT: [[TMP23:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[RDX_SELECT_CMP]])
267+
; CHECK-NEXT: [[RDX_SELECT:%.*]] = select i1 [[TMP23]], i1 false, i1 true
268+
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP3]], [[N_VEC]]
269+
; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]]
270+
; CHECK: vec.epilog.iter.check:
271+
; CHECK-NEXT: [[TMP24:%.*]] = mul i64 [[N_VEC]], 16
272+
; CHECK-NEXT: [[IND_END9:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP24]]
273+
; CHECK-NEXT: [[N_VEC_REMAINING:%.*]] = sub i64 [[TMP3]], [[N_VEC]]
274+
; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_VEC_REMAINING]], 4
275+
; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]]
276+
; CHECK: vec.epilog.ph:
277+
; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i1 [ true, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ], [ [[RDX_SELECT]], [[VEC_EPILOG_ITER_CHECK]] ]
278+
; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
279+
; CHECK-NEXT: [[N_MOD_VF7:%.*]] = urem i64 [[TMP3]], 4
280+
; CHECK-NEXT: [[N_VEC8:%.*]] = sub i64 [[TMP3]], [[N_MOD_VF7]]
281+
; CHECK-NEXT: [[TMP25:%.*]] = mul i64 [[N_VEC8]], 16
282+
; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP25]]
283+
; CHECK-NEXT: [[MINMAX_IDENT_SPLATINSERT:%.*]] = insertelement <4 x i1> poison, i1 [[BC_MERGE_RDX]], i64 0
284+
; CHECK-NEXT: [[MINMAX_IDENT_SPLAT:%.*]] = shufflevector <4 x i1> [[MINMAX_IDENT_SPLATINSERT]], <4 x i1> poison, <4 x i32> zeroinitializer
285+
; CHECK-NEXT: [[BROADCAST_SPLATINSERT18:%.*]] = insertelement <4 x i64> poison, i64 [[X]], i64 0
286+
; CHECK-NEXT: [[BROADCAST_SPLAT19:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT18]], <4 x i64> poison, <4 x i32> zeroinitializer
287+
; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]]
288+
; CHECK: vec.epilog.vector.body:
289+
; CHECK-NEXT: [[INDEX11:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT20:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
290+
; CHECK-NEXT: [[VEC_PHI12:%.*]] = phi <4 x i1> [ [[MINMAX_IDENT_SPLAT]], [[VEC_EPILOG_PH]] ], [ [[TMP43:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
291+
; CHECK-NEXT: [[OFFSET_IDX13:%.*]] = mul i64 [[INDEX11]], 16
292+
; CHECK-NEXT: [[TMP26:%.*]] = add i64 [[OFFSET_IDX13]], 0
293+
; CHECK-NEXT: [[TMP27:%.*]] = add i64 [[OFFSET_IDX13]], 16
294+
; CHECK-NEXT: [[TMP28:%.*]] = add i64 [[OFFSET_IDX13]], 32
295+
; CHECK-NEXT: [[TMP29:%.*]] = add i64 [[OFFSET_IDX13]], 48
296+
; CHECK-NEXT: [[NEXT_GEP14:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP26]]
297+
; CHECK-NEXT: [[NEXT_GEP15:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP27]]
298+
; CHECK-NEXT: [[NEXT_GEP16:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP28]]
299+
; CHECK-NEXT: [[NEXT_GEP17:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP29]]
300+
; CHECK-NEXT: [[TMP30:%.*]] = getelementptr inbounds i8, ptr [[NEXT_GEP14]], i64 8
301+
; CHECK-NEXT: [[TMP31:%.*]] = getelementptr inbounds i8, ptr [[NEXT_GEP15]], i64 8
302+
; CHECK-NEXT: [[TMP32:%.*]] = getelementptr inbounds i8, ptr [[NEXT_GEP16]], i64 8
303+
; CHECK-NEXT: [[TMP33:%.*]] = getelementptr inbounds i8, ptr [[NEXT_GEP17]], i64 8
304+
; CHECK-NEXT: [[TMP34:%.*]] = load i64, ptr [[TMP30]], align 8
305+
; CHECK-NEXT: [[TMP35:%.*]] = load i64, ptr [[TMP31]], align 8
306+
; CHECK-NEXT: [[TMP36:%.*]] = load i64, ptr [[TMP32]], align 8
307+
; CHECK-NEXT: [[TMP37:%.*]] = load i64, ptr [[TMP33]], align 8
308+
; CHECK-NEXT: [[TMP38:%.*]] = insertelement <4 x i64> poison, i64 [[TMP34]], i32 0
309+
; CHECK-NEXT: [[TMP39:%.*]] = insertelement <4 x i64> [[TMP38]], i64 [[TMP35]], i32 1
310+
; CHECK-NEXT: [[TMP40:%.*]] = insertelement <4 x i64> [[TMP39]], i64 [[TMP36]], i32 2
311+
; CHECK-NEXT: [[TMP41:%.*]] = insertelement <4 x i64> [[TMP40]], i64 [[TMP37]], i32 3
312+
; CHECK-NEXT: [[TMP42:%.*]] = icmp eq <4 x i64> [[TMP41]], [[BROADCAST_SPLAT19]]
313+
; CHECK-NEXT: [[TMP43]] = select <4 x i1> [[TMP42]], <4 x i1> [[VEC_PHI12]], <4 x i1> zeroinitializer
314+
; CHECK-NEXT: [[INDEX_NEXT20]] = add nuw i64 [[INDEX11]], 4
315+
; CHECK-NEXT: [[TMP44:%.*]] = icmp eq i64 [[INDEX_NEXT20]], [[N_VEC8]]
316+
; CHECK-NEXT: br i1 [[TMP44]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]]
317+
; CHECK: vec.epilog.middle.block:
318+
; CHECK-NEXT: [[RDX_SELECT_CMP21:%.*]] = icmp ne <4 x i1> [[TMP43]], <i1 true, i1 true, i1 true, i1 true>
319+
; CHECK-NEXT: [[TMP45:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[RDX_SELECT_CMP21]])
320+
; CHECK-NEXT: [[RDX_SELECT22:%.*]] = select i1 [[TMP45]], i1 false, i1 true
321+
; CHECK-NEXT: [[CMP_N10:%.*]] = icmp eq i64 [[TMP3]], [[N_VEC8]]
322+
; CHECK-NEXT: br i1 [[CMP_N10]], label [[EXIT]], label [[VEC_EPILOG_SCALAR_PH]]
323+
; CHECK: vec.epilog.scalar.ph:
324+
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi ptr [ [[IND_END]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[IND_END9]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[START]], [[ITER_CHECK:%.*]] ]
325+
; CHECK-NEXT: [[BC_MERGE_RDX23:%.*]] = phi i1 [ true, [[ITER_CHECK]] ], [ [[RDX_SELECT]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[RDX_SELECT22]], [[VEC_EPILOG_MIDDLE_BLOCK]] ]
326+
; CHECK-NEXT: br label [[LOOP:%.*]]
327+
; CHECK: loop:
328+
; CHECK-NEXT: [[RED:%.*]] = phi i1 [ [[BC_MERGE_RDX23]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[SELECT:%.*]], [[LOOP]] ]
329+
; CHECK-NEXT: [[IV:%.*]] = phi ptr [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
330+
; CHECK-NEXT: [[GETELEMENTPTR5:%.*]] = getelementptr inbounds i8, ptr [[IV]], i64 8
331+
; CHECK-NEXT: [[LOAD6:%.*]] = load i64, ptr [[GETELEMENTPTR5]], align 8
332+
; CHECK-NEXT: [[ICMP7:%.*]] = icmp eq i64 [[LOAD6]], [[X]]
333+
; CHECK-NEXT: [[SELECT]] = select i1 [[ICMP7]], i1 [[RED]], i1 false
334+
; CHECK-NEXT: [[IV_NEXT]] = getelementptr inbounds i8, ptr [[IV]], i64 16
335+
; CHECK-NEXT: [[EC:%.*]] = icmp eq ptr [[IV_NEXT]], [[END]]
336+
; CHECK-NEXT: br i1 [[EC]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP10:![0-9]+]]
337+
; CHECK: exit:
338+
; CHECK-NEXT: [[SELECT_LCSSA:%.*]] = phi i1 [ [[SELECT]], [[LOOP]] ], [ [[RDX_SELECT]], [[MIDDLE_BLOCK]] ], [ [[RDX_SELECT22]], [[VEC_EPILOG_MIDDLE_BLOCK]] ]
339+
; CHECK-NEXT: ret i1 [[SELECT_LCSSA]]
340+
;
341+
entry:
342+
br label %loop
343+
344+
loop: ; preds = %bb3, %bb2
345+
%red = phi i1 [ true, %entry ], [ %select, %loop ]
346+
%iv = phi ptr [ %start, %entry ], [ %iv.next, %loop ]
347+
%getelementptr5 = getelementptr inbounds i8, ptr %iv, i64 8
348+
%load6 = load i64, ptr %getelementptr5, align 8
349+
%icmp7 = icmp eq i64 %load6, %x
350+
%select = select i1 %icmp7, i1 %red, i1 false
351+
%iv.next = getelementptr inbounds i8, ptr %iv, i64 16
352+
%ec = icmp eq ptr %iv.next, %end
353+
br i1 %ec, label %exit, label %loop
354+
355+
exit:
356+
ret i1 %select
357+
}
358+
213359
;.
214360
; CHECK: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]], [[META2:![0-9]+]]}
215361
; CHECK: [[META1]] = !{!"llvm.loop.isvectorized", i32 1}
@@ -219,4 +365,7 @@ exit:
219365
; CHECK: [[LOOP5]] = distinct !{[[LOOP5]], [[META1]], [[META2]]}
220366
; CHECK: [[LOOP6]] = distinct !{[[LOOP6]], [[META1]], [[META2]]}
221367
; CHECK: [[LOOP7]] = distinct !{[[LOOP7]], [[META2]], [[META1]]}
368+
; CHECK: [[LOOP8]] = distinct !{[[LOOP8]], [[META1]], [[META2]]}
369+
; CHECK: [[LOOP9]] = distinct !{[[LOOP9]], [[META1]], [[META2]]}
370+
; CHECK: [[LOOP10]] = distinct !{[[LOOP10]], [[META2]], [[META1]]}
222371
;.

0 commit comments

Comments
 (0)