Skip to content

Commit e613a54

Browse files
RKSimonRKSimon
committed
[Analysis] getIntrinsicForCallSite - add vectorization support for acos/asin/atan and cosh/sinh/tanh libcalls
Followup to #llvm#106584 - ensure acos/asin/atan and cosh/sinh/tanh libcalls correctly map to the llvm intrinsic equivalents
1 parent 4d412be commit e613a54

File tree

4 files changed

+113
-113
lines changed

4 files changed

+113
-113
lines changed

‎llvm/lib/Analysis/ValueTracking.cpp

Lines changed: 24 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -4132,6 +4132,30 @@ Intrinsic::ID llvm::getIntrinsicForCallSite(const CallBase &CB,
41324132
case LibFunc_tanf:
41334133
case LibFunc_tanl:
41344134
return Intrinsic::tan;
4135+
case LibFunc_asin:
4136+
case LibFunc_asinf:
4137+
case LibFunc_asinl:
4138+
return Intrinsic::asin;
4139+
case LibFunc_acos:
4140+
case LibFunc_acosf:
4141+
case LibFunc_acosl:
4142+
return Intrinsic::acos;
4143+
case LibFunc_atan:
4144+
case LibFunc_atanf:
4145+
case LibFunc_atanl:
4146+
return Intrinsic::atan;
4147+
case LibFunc_sinh:
4148+
case LibFunc_sinhf:
4149+
case LibFunc_sinhl:
4150+
return Intrinsic::sinh;
4151+
case LibFunc_cosh:
4152+
case LibFunc_coshf:
4153+
case LibFunc_coshl:
4154+
return Intrinsic::cosh;
4155+
case LibFunc_tanh:
4156+
case LibFunc_tanhf:
4157+
case LibFunc_tanhl:
4158+
return Intrinsic::tanh;
41354159
case LibFunc_exp:
41364160
case LibFunc_expf:
41374161
case LibFunc_expl:

‎llvm/test/Transforms/LoopVectorize/X86/amdlibm-calls.ll

Lines changed: 29 additions & 29 deletions
Original file line numberDiff line numberDiff line change
@@ -398,10 +398,10 @@ for.end:
398398

399399
define void @acos_f64(ptr nocapture %varray) {
400400
; CHECK-LABEL: @acos_f64(
401-
; CHECK-VF2-NOT:[[TMP5:%.*]] = call <2 x double> @llvm.acos.v2f64(<2 x double> [[TMP4:%.*]])
402-
; CHECK-VF4-NOT:[[TMP5:%.*]] = call <4 x double> @llvm.acos.v4f64(<4 x double> [[TMP4:%.*]])
403-
; CHECK-VF8-NOT:[[TMP5:%.*]] = call <8 x double> @llvm.acos.v8f64(<8 x double> [[TMP4:%.*]])
404-
; CHECK-VF16-NOT:[[TMP5:%.*]] = call <16 x double> @llvm.acos.v16f64(<16 x double> [[TMP4:%.*]])
401+
; CHECK-VF2: [[TMP5:%.*]] = call <2 x double> @llvm.acos.v2f64(<2 x double> [[TMP4:%.*]])
402+
; CHECK-VF4: [[TMP5:%.*]] = call <4 x double> @llvm.acos.v4f64(<4 x double> [[TMP4:%.*]])
403+
; CHECK-VF8: [[TMP5:%.*]] = call <8 x double> @llvm.acos.v8f64(<8 x double> [[TMP4:%.*]])
404+
; CHECK-VF16: [[TMP5:%.*]] = call <16 x double> @llvm.acos.v16f64(<16 x double> [[TMP4:%.*]])
405405
; CHECK: ret void
406406
;
407407
entry:
@@ -424,7 +424,7 @@ for.end:
424424

425425
define void @acos_f32(ptr nocapture %varray) {
426426
; CHECK-LABEL: @acos_f32(
427-
; CHECK-VF2-NOT:[[TMP5:%.*]] = call <2 x float> @llvm.acos.v2f32(<2 x float> [[TMP4:%.*]])
427+
; CHECK-VF2: [[TMP5:%.*]] = call <2 x float> @llvm.acos.v2f32(<2 x float> [[TMP4:%.*]])
428428
; CHECK-VF4: [[TMP5:%.*]] = call <4 x float> @amd_vrs4_acosf(<4 x float> [[TMP4:%.*]])
429429
; CHECK-VF8: [[TMP5:%.*]] = call <8 x float> @amd_vrs8_acosf(<8 x float> [[TMP4:%.*]])
430430
; CHECK-VF16: [[TMP5:%.*]] = call <16 x float> @amd_vrs16_acosf(<16 x float> [[TMP4:%.*]])
@@ -502,10 +502,10 @@ for.end:
502502

503503
define void @asin_f64(ptr nocapture %varray) {
504504
; CHECK-LABEL: @asin_f64(
505-
; CHECK-VF2-NOT:[[TMP5:%.*]] = call <2 x double> @llvm.asin.v2f64(<2 x double> [[TMP4:%.*]])
506-
; CHECK-VF4-NOT:[[TMP5:%.*]] = call <4 x double> @llvm.asin.v4f64(<4 x double> [[TMP4:%.*]])
505+
; CHECK-VF2: [[TMP5:%.*]] = call <2 x double> @llvm.asin.v2f64(<2 x double> [[TMP4:%.*]])
506+
; CHECK-VF4: [[TMP5:%.*]] = call <4 x double> @llvm.asin.v4f64(<4 x double> [[TMP4:%.*]])
507507
; CHECK-VF8: [[TMP5:%.*]] = call <8 x double> @amd_vrd8_asin(<8 x double> [[TMP4:%.*]])
508-
; CHECK-VF16-NOT:[[TMP5:%.*]] = call <16 x double> @llvm.asin.v16f64(<16 x double> [[TMP4:%.*]])
508+
; CHECK-VF16: [[TMP5:%.*]] = call <16 x double> @llvm.asin.v16f64(<16 x double> [[TMP4:%.*]])
509509
; CHECK: ret void
510510
;
511511
entry:
@@ -528,7 +528,7 @@ for.end:
528528

529529
define void @asin_f32(ptr nocapture %varray) {
530530
; CHECK-LABEL: @asin_f32(
531-
; CHECK-VF2-NOT:[[TMP5:%.*]] = call <2 x float> @llvm.asin.v2f32(<2 x float> [[TMP4:%.*]])
531+
; CHECK-VF2: [[TMP5:%.*]] = call <2 x float> @llvm.asin.v2f32(<2 x float> [[TMP4:%.*]])
532532
; CHECK-VF4: [[TMP5:%.*]] = call <4 x float> @amd_vrs4_asinf(<4 x float> [[TMP4:%.*]])
533533
; CHECK-VF8: [[TMP5:%.*]] = call <8 x float> @amd_vrs8_asinf(<8 x float> [[TMP4:%.*]])
534534
; CHECK-VF16: [[TMP5:%.*]] = call <16 x float> @amd_vrs16_asinf(<16 x float> [[TMP4:%.*]])
@@ -609,7 +609,7 @@ define void @atan_f64(ptr nocapture %varray) {
609609
; CHECK-VF2: [[TMP5:%.*]] = call <2 x double> @amd_vrd2_atan(<2 x double> [[TMP4:%.*]])
610610
; CHECK-VF4: [[TMP5:%.*]] = call <4 x double> @amd_vrd4_atan(<4 x double> [[TMP4:%.*]])
611611
; CHECK-VF8: [[TMP5:%.*]] = call <8 x double> @amd_vrd8_atan(<8 x double> [[TMP4:%.*]])
612-
; CHECK-VF16-NOT:[[TMP5:%.*]] = call <16 x double> @llvm.atan.v16f64(<16 x double> [[TMP4:%.*]])
612+
; CHECK-VF16: [[TMP5:%.*]] = call <16 x double> @llvm.atan.v16f64(<16 x double> [[TMP4:%.*]])
613613
; CHECK: ret void
614614
;
615615
entry:
@@ -632,7 +632,7 @@ for.end:
632632

633633
define void @atan_f32(ptr nocapture %varray) {
634634
; CHECK-LABEL: @atan_f32(
635-
; CHECK-VF2-NOT:[[TMP5:%.*]] = call <2 x float> @llvm.atan.v2f32(<2 x float> [[TMP4:%.*]])
635+
; CHECK-VF2: [[TMP5:%.*]] = call <2 x float> @llvm.atan.v2f32(<2 x float> [[TMP4:%.*]])
636636
; CHECK-VF4: [[TMP5:%.*]] = call <4 x float> @amd_vrs4_atanf(<4 x float> [[TMP4:%.*]])
637637
; CHECK-VF8: [[TMP5:%.*]] = call <8 x float> @amd_vrs8_atanf(<8 x float> [[TMP4:%.*]])
638638
; CHECK-VF16: [[TMP5:%.*]] = call <16 x float> @amd_vrs16_atanf(<16 x float> [[TMP4:%.*]])
@@ -710,10 +710,10 @@ for.end:
710710

711711
define void @sinh_f64(ptr nocapture %varray) {
712712
; CHECK-LABEL: @sinh_f64(
713-
; CHECK-VF2-NOT:[[TMP5:%.*]] = call <2 x double> @llvm.sinh.v2f64(<2 x double> [[TMP4:%.*]])
714-
; CHECK-VF4-NOT:[[TMP5:%.*]] = call <4 x double> @llvm.sinh.v4f64(<4 x double> [[TMP4:%.*]])
715-
; CHECK-VF8-NOT:[[TMP5:%.*]] = call <8 x double> @llvm.sinh.v8f64(<8 x double> [[TMP4:%.*]])
716-
; CHECK-VF16-NOT:[[TMP5:%.*]] = call <16 x double> @llvm.sinh.v16f64(<16 x double> [[TMP4:%.*]])
713+
; CHECK-VF2: [[TMP5:%.*]] = call <2 x double> @llvm.sinh.v2f64(<2 x double> [[TMP4:%.*]])
714+
; CHECK-VF4: [[TMP5:%.*]] = call <4 x double> @llvm.sinh.v4f64(<4 x double> [[TMP4:%.*]])
715+
; CHECK-VF8: [[TMP5:%.*]] = call <8 x double> @llvm.sinh.v8f64(<8 x double> [[TMP4:%.*]])
716+
; CHECK-VF16: [[TMP5:%.*]] = call <16 x double> @llvm.sinh.v16f64(<16 x double> [[TMP4:%.*]])
717717
; CHECK: ret void
718718
;
719719
entry:
@@ -736,10 +736,10 @@ for.end:
736736

737737
define void @sinh_f32(ptr nocapture %varray) {
738738
; CHECK-LABEL: @sinh_f32(
739-
; CHECK-VF2-NOT:[[TMP5:%.*]] = call <2 x float> @llvm.sinh.v2f32(<2 x float> [[TMP4:%.*]])
740-
; CHECK-VF4-NOT:[[TMP5:%.*]] = call <4 x float> @llvm.sinh.v4f32(<4 x float> [[TMP4:%.*]])
741-
; CHECK-VF8-NOT:[[TMP5:%.*]] = call <8 x float> @llvm.sinh.v8f32(<8 x float> [[TMP4:%.*]])
742-
; CHECK-VF16-NOT:[[TMP5:%.*]] = call <16 x float> @llvm.sinh.v16f32(<16 x float> [[TMP4:%.*]])
739+
; CHECK-VF2: [[TMP5:%.*]] = call <2 x float> @llvm.sinh.v2f32(<2 x float> [[TMP4:%.*]])
740+
; CHECK-VF4: [[TMP5:%.*]] = call <4 x float> @llvm.sinh.v4f32(<4 x float> [[TMP4:%.*]])
741+
; CHECK-VF8: [[TMP5:%.*]] = call <8 x float> @llvm.sinh.v8f32(<8 x float> [[TMP4:%.*]])
742+
; CHECK-VF16: [[TMP5:%.*]] = call <16 x float> @llvm.sinh.v16f32(<16 x float> [[TMP4:%.*]])
743743
; CHECK: ret void
744744
;
745745
entry:
@@ -815,9 +815,9 @@ for.end:
815815
define void @cosh_f64(ptr nocapture %varray) {
816816
; CHECK-LABEL: @cosh_f64(
817817
; CHECK-VF2: [[TMP5:%.*]] = call <2 x double> @amd_vrd2_cosh(<2 x double> [[TMP4:%.*]])
818-
; CHECK-VF4-NOT:[[TMP5:%.*]] = call <4 x double> @llvm.cosh.v4f64(<4 x double> [[TMP4:%.*]])
819-
; CHECK-VF8-NOT:[[TMP5:%.*]] = call <8 x double> @llvm.cosh.v8f64(<8 x double> [[TMP4:%.*]])
820-
; CHECK-VF16-NOT:[[TMP5:%.*]] = call <16 x double> @llvm.cosh.v16f64(<16 x double> [[TMP4:%.*]])
818+
; CHECK-VF4: [[TMP5:%.*]] = call <4 x double> @llvm.cosh.v4f64(<4 x double> [[TMP4:%.*]])
819+
; CHECK-VF8: [[TMP5:%.*]] = call <8 x double> @llvm.cosh.v8f64(<8 x double> [[TMP4:%.*]])
820+
; CHECK-VF16: [[TMP5:%.*]] = call <16 x double> @llvm.cosh.v16f64(<16 x double> [[TMP4:%.*]])
821821
; CHECK: ret void
822822
;
823823
entry:
@@ -840,10 +840,10 @@ for.end:
840840

841841
define void @cosh_f32(ptr nocapture %varray) {
842842
; CHECK-LABEL: @cosh_f32(
843-
; CHECK-VF2-NOT:[[TMP5:%.*]] = call <2 x float> @llvm.cosh.v2f32(<2 x float> [[TMP4:%.*]])
843+
; CHECK-VF2: [[TMP5:%.*]] = call <2 x float> @llvm.cosh.v2f32(<2 x float> [[TMP4:%.*]])
844844
; CHECK-VF4: [[TMP5:%.*]] = call <4 x float> @amd_vrs4_coshf(<4 x float> [[TMP4:%.*]])
845845
; CHECK-VF8: [[TMP5:%.*]] = call <8 x float> @amd_vrs8_coshf(<8 x float> [[TMP4:%.*]])
846-
; CHECK-VF16-NOT:[[TMP5:%.*]] = call <16 x float> @llvm.cosh.v16f32(<16 x float> [[TMP4:%.*]])
846+
; CHECK-VF16: [[TMP5:%.*]] = call <16 x float> @llvm.cosh.v16f32(<16 x float> [[TMP4:%.*]])
847847
; CHECK: ret void
848848
;
849849
entry:
@@ -918,10 +918,10 @@ for.end:
918918

919919
define void @tanh_f64(ptr nocapture %varray) {
920920
; CHECK-LABEL: @tanh_f64(
921-
; CHECK-VF2-NOT:[[TMP5:%.*]] = call <2 x double> @llvm.tanh.v2f64(<2 x double> [[TMP4:%.*]])
922-
; CHECK-VF4-NOT:[[TMP5:%.*]] = call <4 x double> @llvm.tanh.v4f64(<4 x double> [[TMP4:%.*]])
923-
; CHECK-VF8-NOT:[[TMP5:%.*]] = call <8 x double> @llvm.tanh.v8f64(<8 x double> [[TMP4:%.*]])
924-
; CHECK-VF16-NOT:[[TMP5:%.*]] = call <16 x double> @llvm.tanh.v16f64(<16 x double> [[TMP4:%.*]])
921+
; CHECK-VF2: [[TMP5:%.*]] = call <2 x double> @llvm.tanh.v2f64(<2 x double> [[TMP4:%.*]])
922+
; CHECK-VF4: [[TMP5:%.*]] = call <4 x double> @llvm.tanh.v4f64(<4 x double> [[TMP4:%.*]])
923+
; CHECK-VF8: [[TMP5:%.*]] = call <8 x double> @llvm.tanh.v8f64(<8 x double> [[TMP4:%.*]])
924+
; CHECK-VF16: [[TMP5:%.*]] = call <16 x double> @llvm.tanh.v16f64(<16 x double> [[TMP4:%.*]])
925925
; CHECK: ret void
926926
;
927927
entry:
@@ -944,7 +944,7 @@ for.end:
944944

945945
define void @tanh_f32(ptr nocapture %varray) {
946946
; CHECK-LABEL: @tanh_f32(
947-
; CHECK-VF2-NOT:[[TMP5:%.*]] = call <2 x float> @llvm.tanh.v2f32(<2 x float> [[TMP4:%.*]])
947+
; CHECK-VF2: [[TMP5:%.*]] = call <2 x float> @llvm.tanh.v2f32(<2 x float> [[TMP4:%.*]])
948948
; CHECK-VF4: [[TMP5:%.*]] = call <4 x float> @amd_vrs4_tanhf(<4 x float> [[TMP4:%.*]])
949949
; CHECK-VF8: [[TMP5:%.*]] = call <8 x float> @amd_vrs8_tanhf(<8 x float> [[TMP4:%.*]])
950950
; CHECK-VF16: [[TMP5:%.*]] = call <16 x float> @amd_vrs16_tanhf(<16 x float> [[TMP4:%.*]])

‎llvm/test/Transforms/SLPVectorizer/AArch64/accelerate-vector-functions-inseltpoison.ll

Lines changed: 30 additions & 42 deletions
Original file line numberDiff line numberDiff line change
@@ -587,13 +587,11 @@ define <4 x float> @asin_4x(ptr %a) {
587587
; NOACCELERATE-NEXT: [[VECEXT_1:%.*]] = extractelement <4 x float> [[TMP0]], i32 1
588588
; NOACCELERATE-NEXT: [[TMP2:%.*]] = tail call fast float @asinf(float [[VECEXT_1]])
589589
; NOACCELERATE-NEXT: [[VECINS_1:%.*]] = insertelement <4 x float> [[VECINS]], float [[TMP2]], i32 1
590-
; NOACCELERATE-NEXT: [[VECEXT_2:%.*]] = extractelement <4 x float> [[TMP0]], i32 2
591-
; NOACCELERATE-NEXT: [[TMP3:%.*]] = tail call fast float @asinf(float [[VECEXT_2]])
592-
; NOACCELERATE-NEXT: [[VECINS_2:%.*]] = insertelement <4 x float> [[VECINS_1]], float [[TMP3]], i32 2
593-
; NOACCELERATE-NEXT: [[VECEXT_3:%.*]] = extractelement <4 x float> [[TMP0]], i32 3
594-
; NOACCELERATE-NEXT: [[TMP4:%.*]] = tail call fast float @asinf(float [[VECEXT_3]])
595-
; NOACCELERATE-NEXT: [[VECINS_3:%.*]] = insertelement <4 x float> [[VECINS_2]], float [[TMP4]], i32 3
596-
; NOACCELERATE-NEXT: ret <4 x float> [[VECINS_3]]
590+
; NOACCELERATE-NEXT: [[TMP3:%.*]] = shufflevector <4 x float> [[TMP0]], <4 x float> poison, <2 x i32> <i32 2, i32 3>
591+
; NOACCELERATE-NEXT: [[TMP4:%.*]] = call fast <2 x float> @llvm.asin.v2f32(<2 x float> [[TMP3]])
592+
; NOACCELERATE-NEXT: [[TMP5:%.*]] = shufflevector <2 x float> [[TMP4]], <2 x float> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison>
593+
; NOACCELERATE-NEXT: [[VECINS_31:%.*]] = shufflevector <4 x float> [[VECINS_1]], <4 x float> [[TMP5]], <4 x i32> <i32 0, i32 1, i32 4, i32 5>
594+
; NOACCELERATE-NEXT: ret <4 x float> [[VECINS_31]]
597595
;
598596
entry:
599597
%0 = load <4 x float>, ptr %a, align 16
@@ -666,13 +664,11 @@ define <4 x float> @acos_4x(ptr %a) {
666664
; NOACCELERATE-NEXT: [[VECEXT_1:%.*]] = extractelement <4 x float> [[TMP0]], i32 1
667665
; NOACCELERATE-NEXT: [[TMP2:%.*]] = tail call fast float @acosf(float [[VECEXT_1]])
668666
; NOACCELERATE-NEXT: [[VECINS_1:%.*]] = insertelement <4 x float> [[VECINS]], float [[TMP2]], i32 1
669-
; NOACCELERATE-NEXT: [[VECEXT_2:%.*]] = extractelement <4 x float> [[TMP0]], i32 2
670-
; NOACCELERATE-NEXT: [[TMP3:%.*]] = tail call fast float @acosf(float [[VECEXT_2]])
671-
; NOACCELERATE-NEXT: [[VECINS_2:%.*]] = insertelement <4 x float> [[VECINS_1]], float [[TMP3]], i32 2
672-
; NOACCELERATE-NEXT: [[VECEXT_3:%.*]] = extractelement <4 x float> [[TMP0]], i32 3
673-
; NOACCELERATE-NEXT: [[TMP4:%.*]] = tail call fast float @acosf(float [[VECEXT_3]])
674-
; NOACCELERATE-NEXT: [[VECINS_3:%.*]] = insertelement <4 x float> [[VECINS_2]], float [[TMP4]], i32 3
675-
; NOACCELERATE-NEXT: ret <4 x float> [[VECINS_3]]
667+
; NOACCELERATE-NEXT: [[TMP3:%.*]] = shufflevector <4 x float> [[TMP0]], <4 x float> poison, <2 x i32> <i32 2, i32 3>
668+
; NOACCELERATE-NEXT: [[TMP4:%.*]] = call fast <2 x float> @llvm.acos.v2f32(<2 x float> [[TMP3]])
669+
; NOACCELERATE-NEXT: [[TMP5:%.*]] = shufflevector <2 x float> [[TMP4]], <2 x float> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison>
670+
; NOACCELERATE-NEXT: [[VECINS_31:%.*]] = shufflevector <4 x float> [[VECINS_1]], <4 x float> [[TMP5]], <4 x i32> <i32 0, i32 1, i32 4, i32 5>
671+
; NOACCELERATE-NEXT: ret <4 x float> [[VECINS_31]]
676672
;
677673
entry:
678674
%0 = load <4 x float>, ptr %a, align 16
@@ -745,13 +741,11 @@ define <4 x float> @atan_4x(ptr %a) {
745741
; NOACCELERATE-NEXT: [[VECEXT_1:%.*]] = extractelement <4 x float> [[TMP0]], i32 1
746742
; NOACCELERATE-NEXT: [[TMP2:%.*]] = tail call fast float @atanf(float [[VECEXT_1]])
747743
; NOACCELERATE-NEXT: [[VECINS_1:%.*]] = insertelement <4 x float> [[VECINS]], float [[TMP2]], i32 1
748-
; NOACCELERATE-NEXT: [[VECEXT_2:%.*]] = extractelement <4 x float> [[TMP0]], i32 2
749-
; NOACCELERATE-NEXT: [[TMP3:%.*]] = tail call fast float @atanf(float [[VECEXT_2]])
750-
; NOACCELERATE-NEXT: [[VECINS_2:%.*]] = insertelement <4 x float> [[VECINS_1]], float [[TMP3]], i32 2
751-
; NOACCELERATE-NEXT: [[VECEXT_3:%.*]] = extractelement <4 x float> [[TMP0]], i32 3
752-
; NOACCELERATE-NEXT: [[TMP4:%.*]] = tail call fast float @atanf(float [[VECEXT_3]])
753-
; NOACCELERATE-NEXT: [[VECINS_3:%.*]] = insertelement <4 x float> [[VECINS_2]], float [[TMP4]], i32 3
754-
; NOACCELERATE-NEXT: ret <4 x float> [[VECINS_3]]
744+
; NOACCELERATE-NEXT: [[TMP3:%.*]] = shufflevector <4 x float> [[TMP0]], <4 x float> poison, <2 x i32> <i32 2, i32 3>
745+
; NOACCELERATE-NEXT: [[TMP4:%.*]] = call fast <2 x float> @llvm.atan.v2f32(<2 x float> [[TMP3]])
746+
; NOACCELERATE-NEXT: [[TMP5:%.*]] = shufflevector <2 x float> [[TMP4]], <2 x float> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison>
747+
; NOACCELERATE-NEXT: [[VECINS_31:%.*]] = shufflevector <4 x float> [[VECINS_1]], <4 x float> [[TMP5]], <4 x i32> <i32 0, i32 1, i32 4, i32 5>
748+
; NOACCELERATE-NEXT: ret <4 x float> [[VECINS_31]]
755749
;
756750
entry:
757751
%0 = load <4 x float>, ptr %a, align 16
@@ -824,13 +818,11 @@ define <4 x float> @sinh_4x(ptr %a) {
824818
; NOACCELERATE-NEXT: [[VECEXT_1:%.*]] = extractelement <4 x float> [[TMP0]], i32 1
825819
; NOACCELERATE-NEXT: [[TMP2:%.*]] = tail call fast float @sinhf(float [[VECEXT_1]])
826820
; NOACCELERATE-NEXT: [[VECINS_1:%.*]] = insertelement <4 x float> [[VECINS]], float [[TMP2]], i32 1
827-
; NOACCELERATE-NEXT: [[VECEXT_2:%.*]] = extractelement <4 x float> [[TMP0]], i32 2
828-
; NOACCELERATE-NEXT: [[TMP3:%.*]] = tail call fast float @sinhf(float [[VECEXT_2]])
829-
; NOACCELERATE-NEXT: [[VECINS_2:%.*]] = insertelement <4 x float> [[VECINS_1]], float [[TMP3]], i32 2
830-
; NOACCELERATE-NEXT: [[VECEXT_3:%.*]] = extractelement <4 x float> [[TMP0]], i32 3
831-
; NOACCELERATE-NEXT: [[TMP4:%.*]] = tail call fast float @sinhf(float [[VECEXT_3]])
832-
; NOACCELERATE-NEXT: [[VECINS_3:%.*]] = insertelement <4 x float> [[VECINS_2]], float [[TMP4]], i32 3
833-
; NOACCELERATE-NEXT: ret <4 x float> [[VECINS_3]]
821+
; NOACCELERATE-NEXT: [[TMP3:%.*]] = shufflevector <4 x float> [[TMP0]], <4 x float> poison, <2 x i32> <i32 2, i32 3>
822+
; NOACCELERATE-NEXT: [[TMP4:%.*]] = call fast <2 x float> @llvm.sinh.v2f32(<2 x float> [[TMP3]])
823+
; NOACCELERATE-NEXT: [[TMP5:%.*]] = shufflevector <2 x float> [[TMP4]], <2 x float> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison>
824+
; NOACCELERATE-NEXT: [[VECINS_31:%.*]] = shufflevector <4 x float> [[VECINS_1]], <4 x float> [[TMP5]], <4 x i32> <i32 0, i32 1, i32 4, i32 5>
825+
; NOACCELERATE-NEXT: ret <4 x float> [[VECINS_31]]
834826
;
835827
entry:
836828
%0 = load <4 x float>, ptr %a, align 16
@@ -903,13 +895,11 @@ define <4 x float> @cosh_4x(ptr %a) {
903895
; NOACCELERATE-NEXT: [[VECEXT_1:%.*]] = extractelement <4 x float> [[TMP0]], i32 1
904896
; NOACCELERATE-NEXT: [[TMP2:%.*]] = tail call fast float @coshf(float [[VECEXT_1]])
905897
; NOACCELERATE-NEXT: [[VECINS_1:%.*]] = insertelement <4 x float> [[VECINS]], float [[TMP2]], i32 1
906-
; NOACCELERATE-NEXT: [[VECEXT_2:%.*]] = extractelement <4 x float> [[TMP0]], i32 2
907-
; NOACCELERATE-NEXT: [[TMP3:%.*]] = tail call fast float @coshf(float [[VECEXT_2]])
908-
; NOACCELERATE-NEXT: [[VECINS_2:%.*]] = insertelement <4 x float> [[VECINS_1]], float [[TMP3]], i32 2
909-
; NOACCELERATE-NEXT: [[VECEXT_3:%.*]] = extractelement <4 x float> [[TMP0]], i32 3
910-
; NOACCELERATE-NEXT: [[TMP4:%.*]] = tail call fast float @coshf(float [[VECEXT_3]])
911-
; NOACCELERATE-NEXT: [[VECINS_3:%.*]] = insertelement <4 x float> [[VECINS_2]], float [[TMP4]], i32 3
912-
; NOACCELERATE-NEXT: ret <4 x float> [[VECINS_3]]
898+
; NOACCELERATE-NEXT: [[TMP3:%.*]] = shufflevector <4 x float> [[TMP0]], <4 x float> poison, <2 x i32> <i32 2, i32 3>
899+
; NOACCELERATE-NEXT: [[TMP4:%.*]] = call fast <2 x float> @llvm.cosh.v2f32(<2 x float> [[TMP3]])
900+
; NOACCELERATE-NEXT: [[TMP5:%.*]] = shufflevector <2 x float> [[TMP4]], <2 x float> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison>
901+
; NOACCELERATE-NEXT: [[VECINS_31:%.*]] = shufflevector <4 x float> [[VECINS_1]], <4 x float> [[TMP5]], <4 x i32> <i32 0, i32 1, i32 4, i32 5>
902+
; NOACCELERATE-NEXT: ret <4 x float> [[VECINS_31]]
913903
;
914904
entry:
915905
%0 = load <4 x float>, ptr %a, align 16
@@ -982,13 +972,11 @@ define <4 x float> @tanh_4x(ptr %a) {
982972
; NOACCELERATE-NEXT: [[VECEXT_1:%.*]] = extractelement <4 x float> [[TMP0]], i32 1
983973
; NOACCELERATE-NEXT: [[TMP2:%.*]] = tail call fast float @tanhf(float [[VECEXT_1]])
984974
; NOACCELERATE-NEXT: [[VECINS_1:%.*]] = insertelement <4 x float> [[VECINS]], float [[TMP2]], i32 1
985-
; NOACCELERATE-NEXT: [[VECEXT_2:%.*]] = extractelement <4 x float> [[TMP0]], i32 2
986-
; NOACCELERATE-NEXT: [[TMP3:%.*]] = tail call fast float @tanhf(float [[VECEXT_2]])
987-
; NOACCELERATE-NEXT: [[VECINS_2:%.*]] = insertelement <4 x float> [[VECINS_1]], float [[TMP3]], i32 2
988-
; NOACCELERATE-NEXT: [[VECEXT_3:%.*]] = extractelement <4 x float> [[TMP0]], i32 3
989-
; NOACCELERATE-NEXT: [[TMP4:%.*]] = tail call fast float @tanhf(float [[VECEXT_3]])
990-
; NOACCELERATE-NEXT: [[VECINS_3:%.*]] = insertelement <4 x float> [[VECINS_2]], float [[TMP4]], i32 3
991-
; NOACCELERATE-NEXT: ret <4 x float> [[VECINS_3]]
975+
; NOACCELERATE-NEXT: [[TMP3:%.*]] = shufflevector <4 x float> [[TMP0]], <4 x float> poison, <2 x i32> <i32 2, i32 3>
976+
; NOACCELERATE-NEXT: [[TMP4:%.*]] = call fast <2 x float> @llvm.tanh.v2f32(<2 x float> [[TMP3]])
977+
; NOACCELERATE-NEXT: [[TMP5:%.*]] = shufflevector <2 x float> [[TMP4]], <2 x float> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison>
978+
; NOACCELERATE-NEXT: [[VECINS_31:%.*]] = shufflevector <4 x float> [[VECINS_1]], <4 x float> [[TMP5]], <4 x i32> <i32 0, i32 1, i32 4, i32 5>
979+
; NOACCELERATE-NEXT: ret <4 x float> [[VECINS_31]]
992980
;
993981
entry:
994982
%0 = load <4 x float>, ptr %a, align 16

0 commit comments

Comments
 (0)