Skip to content

Commit 5fa6039

Browse files
RKSimonRKSimon
committed
[SLP][X86] Add llvm.isnan intrinsic test coverage
We still need to tag the llvm.isnan.? intrinsic as vectorizable
1 parent 26ed14f commit 5fa6039

File tree

1 file changed

+54
-12
lines changed

1 file changed

+54
-12
lines changed

‎llvm/test/Transforms/SLPVectorizer/X86/intrinsic.ll

+54-12
Original file line numberDiff line numberDiff line change
@@ -195,25 +195,25 @@ define void @vec_ctlz_i32_neg(i32* %a, i32* %b, i32* %c, i1) {
195195
; CHECK-NEXT: [[I0:%.*]] = load i32, i32* [[A:%.*]], align 4
196196
; CHECK-NEXT: [[I1:%.*]] = load i32, i32* [[B:%.*]], align 4
197197
; CHECK-NEXT: [[ADD1:%.*]] = add i32 [[I0]], [[I1]]
198-
; CHECK-NEXT: [[CALL1:%.*]] = tail call i32 @llvm.ctlz.i32(i32 [[ADD1]], i1 true) #[[ATTR3:[0-9]+]]
198+
; CHECK-NEXT: [[CALL1:%.*]] = tail call i32 @llvm.ctlz.i32(i32 [[ADD1]], i1 true) #[[ATTR4:[0-9]+]]
199199
; CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 1
200200
; CHECK-NEXT: [[I2:%.*]] = load i32, i32* [[ARRAYIDX2]], align 4
201201
; CHECK-NEXT: [[ARRAYIDX3:%.*]] = getelementptr inbounds i32, i32* [[B]], i32 1
202202
; CHECK-NEXT: [[I3:%.*]] = load i32, i32* [[ARRAYIDX3]], align 4
203203
; CHECK-NEXT: [[ADD2:%.*]] = add i32 [[I2]], [[I3]]
204-
; CHECK-NEXT: [[CALL2:%.*]] = tail call i32 @llvm.ctlz.i32(i32 [[ADD2]], i1 false) #[[ATTR3]]
204+
; CHECK-NEXT: [[CALL2:%.*]] = tail call i32 @llvm.ctlz.i32(i32 [[ADD2]], i1 false) #[[ATTR4]]
205205
; CHECK-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 2
206206
; CHECK-NEXT: [[I4:%.*]] = load i32, i32* [[ARRAYIDX4]], align 4
207207
; CHECK-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds i32, i32* [[B]], i32 2
208208
; CHECK-NEXT: [[I5:%.*]] = load i32, i32* [[ARRAYIDX5]], align 4
209209
; CHECK-NEXT: [[ADD3:%.*]] = add i32 [[I4]], [[I5]]
210-
; CHECK-NEXT: [[CALL3:%.*]] = tail call i32 @llvm.ctlz.i32(i32 [[ADD3]], i1 true) #[[ATTR3]]
210+
; CHECK-NEXT: [[CALL3:%.*]] = tail call i32 @llvm.ctlz.i32(i32 [[ADD3]], i1 true) #[[ATTR4]]
211211
; CHECK-NEXT: [[ARRAYIDX6:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 3
212212
; CHECK-NEXT: [[I6:%.*]] = load i32, i32* [[ARRAYIDX6]], align 4
213213
; CHECK-NEXT: [[ARRAYIDX7:%.*]] = getelementptr inbounds i32, i32* [[B]], i32 3
214214
; CHECK-NEXT: [[I7:%.*]] = load i32, i32* [[ARRAYIDX7]], align 4
215215
; CHECK-NEXT: [[ADD4:%.*]] = add i32 [[I6]], [[I7]]
216-
; CHECK-NEXT: [[CALL4:%.*]] = tail call i32 @llvm.ctlz.i32(i32 [[ADD4]], i1 false) #[[ATTR3]]
216+
; CHECK-NEXT: [[CALL4:%.*]] = tail call i32 @llvm.ctlz.i32(i32 [[ADD4]], i1 false) #[[ATTR4]]
217217
; CHECK-NEXT: store i32 [[CALL1]], i32* [[C:%.*]], align 4
218218
; CHECK-NEXT: [[ARRAYIDX8:%.*]] = getelementptr inbounds i32, i32* [[C]], i32 1
219219
; CHECK-NEXT: store i32 [[CALL2]], i32* [[ARRAYIDX8]], align 4
@@ -322,25 +322,25 @@ define void @vec_cttz_i32_neg(i32* %a, i32* %b, i32* %c, i1) {
322322
; CHECK-NEXT: [[I0:%.*]] = load i32, i32* [[A:%.*]], align 4
323323
; CHECK-NEXT: [[I1:%.*]] = load i32, i32* [[B:%.*]], align 4
324324
; CHECK-NEXT: [[ADD1:%.*]] = add i32 [[I0]], [[I1]]
325-
; CHECK-NEXT: [[CALL1:%.*]] = tail call i32 @llvm.cttz.i32(i32 [[ADD1]], i1 true) #[[ATTR3]]
325+
; CHECK-NEXT: [[CALL1:%.*]] = tail call i32 @llvm.cttz.i32(i32 [[ADD1]], i1 true) #[[ATTR4]]
326326
; CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 1
327327
; CHECK-NEXT: [[I2:%.*]] = load i32, i32* [[ARRAYIDX2]], align 4
328328
; CHECK-NEXT: [[ARRAYIDX3:%.*]] = getelementptr inbounds i32, i32* [[B]], i32 1
329329
; CHECK-NEXT: [[I3:%.*]] = load i32, i32* [[ARRAYIDX3]], align 4
330330
; CHECK-NEXT: [[ADD2:%.*]] = add i32 [[I2]], [[I3]]
331-
; CHECK-NEXT: [[CALL2:%.*]] = tail call i32 @llvm.cttz.i32(i32 [[ADD2]], i1 false) #[[ATTR3]]
331+
; CHECK-NEXT: [[CALL2:%.*]] = tail call i32 @llvm.cttz.i32(i32 [[ADD2]], i1 false) #[[ATTR4]]
332332
; CHECK-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 2
333333
; CHECK-NEXT: [[I4:%.*]] = load i32, i32* [[ARRAYIDX4]], align 4
334334
; CHECK-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds i32, i32* [[B]], i32 2
335335
; CHECK-NEXT: [[I5:%.*]] = load i32, i32* [[ARRAYIDX5]], align 4
336336
; CHECK-NEXT: [[ADD3:%.*]] = add i32 [[I4]], [[I5]]
337-
; CHECK-NEXT: [[CALL3:%.*]] = tail call i32 @llvm.cttz.i32(i32 [[ADD3]], i1 true) #[[ATTR3]]
337+
; CHECK-NEXT: [[CALL3:%.*]] = tail call i32 @llvm.cttz.i32(i32 [[ADD3]], i1 true) #[[ATTR4]]
338338
; CHECK-NEXT: [[ARRAYIDX6:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 3
339339
; CHECK-NEXT: [[I6:%.*]] = load i32, i32* [[ARRAYIDX6]], align 4
340340
; CHECK-NEXT: [[ARRAYIDX7:%.*]] = getelementptr inbounds i32, i32* [[B]], i32 3
341341
; CHECK-NEXT: [[I7:%.*]] = load i32, i32* [[ARRAYIDX7]], align 4
342342
; CHECK-NEXT: [[ADD4:%.*]] = add i32 [[I6]], [[I7]]
343-
; CHECK-NEXT: [[CALL4:%.*]] = tail call i32 @llvm.cttz.i32(i32 [[ADD4]], i1 false) #[[ATTR3]]
343+
; CHECK-NEXT: [[CALL4:%.*]] = tail call i32 @llvm.cttz.i32(i32 [[ADD4]], i1 false) #[[ATTR4]]
344344
; CHECK-NEXT: store i32 [[CALL1]], i32* [[C:%.*]], align 4
345345
; CHECK-NEXT: [[ARRAYIDX8:%.*]] = getelementptr inbounds i32, i32* [[C]], i32 1
346346
; CHECK-NEXT: store i32 [[CALL2]], i32* [[ARRAYIDX8]], align 4
@@ -448,25 +448,25 @@ define void @vec_powi_f32_neg(float* %a, float* %b, float* %c, i32 %P, i32 %Q) {
448448
; CHECK-NEXT: [[I0:%.*]] = load float, float* [[A:%.*]], align 4
449449
; CHECK-NEXT: [[I1:%.*]] = load float, float* [[B:%.*]], align 4
450450
; CHECK-NEXT: [[ADD1:%.*]] = fadd float [[I0]], [[I1]]
451-
; CHECK-NEXT: [[CALL1:%.*]] = tail call float @llvm.powi.f32.i32(float [[ADD1]], i32 [[P:%.*]]) #[[ATTR3]]
451+
; CHECK-NEXT: [[CALL1:%.*]] = tail call float @llvm.powi.f32.i32(float [[ADD1]], i32 [[P:%.*]]) #[[ATTR4]]
452452
; CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds float, float* [[A]], i32 1
453453
; CHECK-NEXT: [[I2:%.*]] = load float, float* [[ARRAYIDX2]], align 4
454454
; CHECK-NEXT: [[ARRAYIDX3:%.*]] = getelementptr inbounds float, float* [[B]], i32 1
455455
; CHECK-NEXT: [[I3:%.*]] = load float, float* [[ARRAYIDX3]], align 4
456456
; CHECK-NEXT: [[ADD2:%.*]] = fadd float [[I2]], [[I3]]
457-
; CHECK-NEXT: [[CALL2:%.*]] = tail call float @llvm.powi.f32.i32(float [[ADD2]], i32 [[Q:%.*]]) #[[ATTR3]]
457+
; CHECK-NEXT: [[CALL2:%.*]] = tail call float @llvm.powi.f32.i32(float [[ADD2]], i32 [[Q:%.*]]) #[[ATTR4]]
458458
; CHECK-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds float, float* [[A]], i32 2
459459
; CHECK-NEXT: [[I4:%.*]] = load float, float* [[ARRAYIDX4]], align 4
460460
; CHECK-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds float, float* [[B]], i32 2
461461
; CHECK-NEXT: [[I5:%.*]] = load float, float* [[ARRAYIDX5]], align 4
462462
; CHECK-NEXT: [[ADD3:%.*]] = fadd float [[I4]], [[I5]]
463-
; CHECK-NEXT: [[CALL3:%.*]] = tail call float @llvm.powi.f32.i32(float [[ADD3]], i32 [[P]]) #[[ATTR3]]
463+
; CHECK-NEXT: [[CALL3:%.*]] = tail call float @llvm.powi.f32.i32(float [[ADD3]], i32 [[P]]) #[[ATTR4]]
464464
; CHECK-NEXT: [[ARRAYIDX6:%.*]] = getelementptr inbounds float, float* [[A]], i32 3
465465
; CHECK-NEXT: [[I6:%.*]] = load float, float* [[ARRAYIDX6]], align 4
466466
; CHECK-NEXT: [[ARRAYIDX7:%.*]] = getelementptr inbounds float, float* [[B]], i32 3
467467
; CHECK-NEXT: [[I7:%.*]] = load float, float* [[ARRAYIDX7]], align 4
468468
; CHECK-NEXT: [[ADD4:%.*]] = fadd float [[I6]], [[I7]]
469-
; CHECK-NEXT: [[CALL4:%.*]] = tail call float @llvm.powi.f32.i32(float [[ADD4]], i32 [[Q]]) #[[ATTR3]]
469+
; CHECK-NEXT: [[CALL4:%.*]] = tail call float @llvm.powi.f32.i32(float [[ADD4]], i32 [[Q]]) #[[ATTR4]]
470470
; CHECK-NEXT: store float [[CALL1]], float* [[C:%.*]], align 4
471471
; CHECK-NEXT: [[ARRAYIDX8:%.*]] = getelementptr inbounds float, float* [[C]], i32 1
472472
; CHECK-NEXT: store float [[CALL2]], float* [[ARRAYIDX8]], align 4
@@ -513,3 +513,45 @@ entry:
513513
ret void
514514

515515
}
516+
517+
518+
declare i1 @llvm.isnan.f64(double)
519+
520+
define void @vec_isnan_f64(double* %a, double* %b, double* %c, double* %d) {
521+
; CHECK-LABEL: @vec_isnan_f64(
522+
; CHECK-NEXT: entry:
523+
; CHECK-NEXT: [[AIDX1:%.*]] = getelementptr inbounds double, double* [[A:%.*]], i64 1
524+
; CHECK-NEXT: [[A0:%.*]] = load double, double* [[A]], align 8
525+
; CHECK-NEXT: [[A1:%.*]] = load double, double* [[AIDX1]], align 8
526+
; CHECK-NEXT: [[TMP0:%.*]] = bitcast double* [[B:%.*]] to <2 x double>*
527+
; CHECK-NEXT: [[TMP1:%.*]] = load <2 x double>, <2 x double>* [[TMP0]], align 8
528+
; CHECK-NEXT: [[TMP2:%.*]] = bitcast double* [[C:%.*]] to <2 x double>*
529+
; CHECK-NEXT: [[TMP3:%.*]] = load <2 x double>, <2 x double>* [[TMP2]], align 8
530+
; CHECK-NEXT: [[ISNAN0:%.*]] = tail call i1 @llvm.isnan.f64(double [[A0]])
531+
; CHECK-NEXT: [[ISNAN1:%.*]] = tail call i1 @llvm.isnan.f64(double [[A1]])
532+
; CHECK-NEXT: [[TMP4:%.*]] = insertelement <2 x i1> poison, i1 [[ISNAN0]], i32 0
533+
; CHECK-NEXT: [[TMP5:%.*]] = insertelement <2 x i1> [[TMP4]], i1 [[ISNAN1]], i32 1
534+
; CHECK-NEXT: [[TMP6:%.*]] = select <2 x i1> [[TMP5]], <2 x double> [[TMP1]], <2 x double> [[TMP3]]
535+
; CHECK-NEXT: [[TMP7:%.*]] = bitcast double* [[D:%.*]] to <2 x double>*
536+
; CHECK-NEXT: store <2 x double> [[TMP6]], <2 x double>* [[TMP7]], align 8
537+
; CHECK-NEXT: ret void
538+
;
539+
entry:
540+
%aidx1 = getelementptr inbounds double, double* %a, i64 1
541+
%bidx1 = getelementptr inbounds double, double* %b, i64 1
542+
%cidx1 = getelementptr inbounds double, double* %c, i64 1
543+
%didx1 = getelementptr inbounds double, double* %d, i64 1
544+
%a0 = load double, double* %a, align 8
545+
%b0 = load double, double* %b, align 8
546+
%c0 = load double, double* %c, align 8
547+
%a1 = load double, double* %aidx1, align 8
548+
%b1 = load double, double* %bidx1, align 8
549+
%c1 = load double, double* %cidx1, align 8
550+
%isnan0 = tail call i1 @llvm.isnan.f64(double %a0)
551+
%isnan1 = tail call i1 @llvm.isnan.f64(double %a1)
552+
%r0 = select i1 %isnan0, double %b0, double %c0
553+
%r1 = select i1 %isnan1, double %b1, double %c1
554+
store double %r0, double* %d, align 8
555+
store double %r1, double* %didx1, align 8
556+
ret void
557+
}

0 commit comments

Comments
 (0)