Skip to content

Commit eb5ffc7

Browse files
rotaterightrotateright
committed
[x86] add tests with constants for FP scalarization; NFC
llvm-svn: 357587
1 parent 3b8c90b commit eb5ffc7

File tree

1 file changed

+231
-0
lines changed

1 file changed

+231
-0
lines changed

‎llvm/test/CodeGen/X86/scalarize-fp.ll

Lines changed: 231 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -466,6 +466,122 @@ define <8 x float> @fdiv_splat_splat_v8f32(<8 x float> %vx, <8 x float> %vy) {
466466
ret <8 x float> %r
467467
}
468468

469+
define <2 x double> @fadd_splat_const_op1_v2f64(<2 x double> %vx) {
470+
; SSE-LABEL: fadd_splat_const_op1_v2f64:
471+
; SSE: # %bb.0:
472+
; SSE-NEXT: unpcklpd {{.*#+}} xmm0 = xmm0[0,0]
473+
; SSE-NEXT: addpd {{.*}}(%rip), %xmm0
474+
; SSE-NEXT: retq
475+
;
476+
; AVX-LABEL: fadd_splat_const_op1_v2f64:
477+
; AVX: # %bb.0:
478+
; AVX-NEXT: vmovddup {{.*#+}} xmm0 = xmm0[0,0]
479+
; AVX-NEXT: vaddpd {{.*}}(%rip), %xmm0, %xmm0
480+
; AVX-NEXT: retq
481+
%splatx = shufflevector <2 x double> %vx, <2 x double> undef, <2 x i32> zeroinitializer
482+
%r = fadd <2 x double> %splatx, <double 42.0, double 42.0>
483+
ret <2 x double> %r
484+
}
485+
486+
define <4 x double> @fsub_const_op0_splat_v4f64(double %x) {
487+
; SSE-LABEL: fsub_const_op0_splat_v4f64:
488+
; SSE: # %bb.0:
489+
; SSE-NEXT: movsd {{.*#+}} xmm1 = mem[0],zero
490+
; SSE-NEXT: subpd %xmm0, %xmm1
491+
; SSE-NEXT: unpcklpd {{.*#+}} xmm1 = xmm1[0,0]
492+
; SSE-NEXT: movapd %xmm1, %xmm0
493+
; SSE-NEXT: retq
494+
;
495+
; AVX-LABEL: fsub_const_op0_splat_v4f64:
496+
; AVX: # %bb.0:
497+
; AVX-NEXT: vmovsd {{.*#+}} xmm1 = mem[0],zero
498+
; AVX-NEXT: vsubpd %xmm0, %xmm1, %xmm0
499+
; AVX-NEXT: vmovddup {{.*#+}} xmm0 = xmm0[0,0]
500+
; AVX-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0
501+
; AVX-NEXT: retq
502+
%vx = insertelement <4 x double> undef, double 8.0, i32 0
503+
%vy = insertelement <4 x double> undef, double %x, i32 0
504+
%splatx = shufflevector <4 x double> %vx, <4 x double> undef, <4 x i32> zeroinitializer
505+
%splaty = shufflevector <4 x double> %vy, <4 x double> undef, <4 x i32> zeroinitializer
506+
%r = fsub <4 x double> %splatx, %splaty
507+
ret <4 x double> %r
508+
}
509+
510+
define <4 x float> @fmul_splat_const_op1_v4f32(<4 x float> %vx, <4 x float> %vy) {
511+
; SSE-LABEL: fmul_splat_const_op1_v4f32:
512+
; SSE: # %bb.0:
513+
; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0,0,0]
514+
; SSE-NEXT: mulps {{.*}}(%rip), %xmm0
515+
; SSE-NEXT: retq
516+
;
517+
; AVX-LABEL: fmul_splat_const_op1_v4f32:
518+
; AVX: # %bb.0:
519+
; AVX-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[0,0,0,0]
520+
; AVX-NEXT: vmulps {{.*}}(%rip), %xmm0, %xmm0
521+
; AVX-NEXT: retq
522+
%splatx = shufflevector <4 x float> %vx, <4 x float> undef, <4 x i32> zeroinitializer
523+
%r = fmul fast <4 x float> %splatx, <float 17.0, float 17.0, float 17.0, float 17.0>
524+
ret <4 x float> %r
525+
}
526+
527+
define <8 x float> @fdiv_splat_const_op0_v8f32(<8 x float> %vy) {
528+
; SSE-LABEL: fdiv_splat_const_op0_v8f32:
529+
; SSE: # %bb.0:
530+
; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0,0,0]
531+
; SSE-NEXT: rcpps %xmm0, %xmm2
532+
; SSE-NEXT: mulps %xmm2, %xmm0
533+
; SSE-NEXT: movaps {{.*#+}} xmm1 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0]
534+
; SSE-NEXT: subps %xmm0, %xmm1
535+
; SSE-NEXT: mulps %xmm2, %xmm1
536+
; SSE-NEXT: addps %xmm2, %xmm1
537+
; SSE-NEXT: mulps {{.*}}(%rip), %xmm1
538+
; SSE-NEXT: movaps %xmm1, %xmm0
539+
; SSE-NEXT: retq
540+
;
541+
; AVX-LABEL: fdiv_splat_const_op0_v8f32:
542+
; AVX: # %bb.0:
543+
; AVX-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[0,0,0,0]
544+
; AVX-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0
545+
; AVX-NEXT: vrcpps %ymm0, %ymm1
546+
; AVX-NEXT: vmulps %ymm1, %ymm0, %ymm0
547+
; AVX-NEXT: vmovaps {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
548+
; AVX-NEXT: vsubps %ymm0, %ymm2, %ymm0
549+
; AVX-NEXT: vmulps %ymm0, %ymm1, %ymm0
550+
; AVX-NEXT: vaddps %ymm0, %ymm1, %ymm0
551+
; AVX-NEXT: vmulps {{.*}}(%rip), %ymm0, %ymm0
552+
; AVX-NEXT: retq
553+
%splatx = shufflevector <8 x float> <float 4.5, float 1.0, float 2.0, float 3.0, float 4.0, float 5.0, float 6.0, float 7.0>, <8 x float> undef, <8 x i32> zeroinitializer
554+
%splaty = shufflevector <8 x float> %vy, <8 x float> undef, <8 x i32> zeroinitializer
555+
%r = fdiv fast <8 x float> %splatx, %splaty
556+
ret <8 x float> %r
557+
}
558+
559+
define <8 x float> @fdiv_const_op1_splat_v8f32(<8 x float> %vx) {
560+
; SSE-LABEL: fdiv_const_op1_splat_v8f32:
561+
; SSE: # %bb.0:
562+
; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0,0,0]
563+
; SSE-NEXT: xorps %xmm1, %xmm1
564+
; SSE-NEXT: rcpps %xmm1, %xmm1
565+
; SSE-NEXT: addps %xmm1, %xmm1
566+
; SSE-NEXT: mulps %xmm0, %xmm1
567+
; SSE-NEXT: movaps %xmm1, %xmm0
568+
; SSE-NEXT: retq
569+
;
570+
; AVX-LABEL: fdiv_const_op1_splat_v8f32:
571+
; AVX: # %bb.0:
572+
; AVX-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[0,0,0,0]
573+
; AVX-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0
574+
; AVX-NEXT: vxorps %xmm1, %xmm1, %xmm1
575+
; AVX-NEXT: vrcpps %ymm1, %ymm1
576+
; AVX-NEXT: vaddps %ymm1, %ymm1, %ymm1
577+
; AVX-NEXT: vmulps %ymm1, %ymm0, %ymm0
578+
; AVX-NEXT: retq
579+
%splatx = shufflevector <8 x float> %vx, <8 x float> undef, <8 x i32> zeroinitializer
580+
%splaty = shufflevector <8 x float> <float 0.0, float 1.0, float 2.0, float 3.0, float 4.0, float 5.0, float 6.0, float 7.0>, <8 x float> undef, <8 x i32> zeroinitializer
581+
%r = fdiv fast <8 x float> %splatx, %splaty
582+
ret <8 x float> %r
583+
}
584+
469585
define <2 x double> @splat0_fadd_v2f64(<2 x double> %vx, <2 x double> %vy) {
470586
; SSE-LABEL: splat0_fadd_v2f64:
471587
; SSE: # %bb.0:
@@ -551,3 +667,118 @@ define <8 x float> @splat0_fdiv_v8f32(<8 x float> %vx, <8 x float> %vy) {
551667
%r = shufflevector <8 x float> %b, <8 x float> undef, <8 x i32> zeroinitializer
552668
ret <8 x float> %r
553669
}
670+
671+
define <2 x double> @splat0_fadd_const_op1_v2f64(<2 x double> %vx) {
672+
; SSE-LABEL: splat0_fadd_const_op1_v2f64:
673+
; SSE: # %bb.0:
674+
; SSE-NEXT: movsd {{.*#+}} xmm1 = mem[0],zero
675+
; SSE-NEXT: addpd %xmm0, %xmm1
676+
; SSE-NEXT: unpcklpd {{.*#+}} xmm1 = xmm1[0,0]
677+
; SSE-NEXT: movapd %xmm1, %xmm0
678+
; SSE-NEXT: retq
679+
;
680+
; AVX-LABEL: splat0_fadd_const_op1_v2f64:
681+
; AVX: # %bb.0:
682+
; AVX-NEXT: vmovsd {{.*#+}} xmm1 = mem[0],zero
683+
; AVX-NEXT: vaddpd %xmm1, %xmm0, %xmm0
684+
; AVX-NEXT: vmovddup {{.*#+}} xmm0 = xmm0[0,0]
685+
; AVX-NEXT: retq
686+
%b = fadd <2 x double> %vx, <double 42.0, double 12.0>
687+
%r = shufflevector <2 x double> %b, <2 x double> undef, <2 x i32> zeroinitializer
688+
ret <2 x double> %r
689+
}
690+
691+
define <4 x double> @splat0_fsub_const_op0_v4f64(double %x) {
692+
; SSE-LABEL: splat0_fsub_const_op0_v4f64:
693+
; SSE: # %bb.0:
694+
; SSE-NEXT: movsd {{.*#+}} xmm1 = mem[0],zero
695+
; SSE-NEXT: subpd %xmm0, %xmm1
696+
; SSE-NEXT: unpcklpd {{.*#+}} xmm1 = xmm1[0,0]
697+
; SSE-NEXT: movapd %xmm1, %xmm0
698+
; SSE-NEXT: retq
699+
;
700+
; AVX-LABEL: splat0_fsub_const_op0_v4f64:
701+
; AVX: # %bb.0:
702+
; AVX-NEXT: vmovsd {{.*#+}} xmm1 = mem[0],zero
703+
; AVX-NEXT: vsubpd %xmm0, %xmm1, %xmm0
704+
; AVX-NEXT: vmovddup {{.*#+}} xmm0 = xmm0[0,0]
705+
; AVX-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0
706+
; AVX-NEXT: retq
707+
%vx = insertelement <4 x double> undef, double %x, i32 0
708+
%b = fsub <4 x double> <double -42.0, double 42.0, double 0.0, double 1.0>, %vx
709+
%r = shufflevector <4 x double> %b, <4 x double> undef, <4 x i32> zeroinitializer
710+
ret <4 x double> %r
711+
}
712+
713+
define <4 x float> @splat0_fmul_const_op1_v4f32(<4 x float> %vx) {
714+
; SSE-LABEL: splat0_fmul_const_op1_v4f32:
715+
; SSE: # %bb.0:
716+
; SSE-NEXT: movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
717+
; SSE-NEXT: mulps %xmm0, %xmm1
718+
; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,0,0,0]
719+
; SSE-NEXT: movaps %xmm1, %xmm0
720+
; SSE-NEXT: retq
721+
;
722+
; AVX-LABEL: splat0_fmul_const_op1_v4f32:
723+
; AVX: # %bb.0:
724+
; AVX-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
725+
; AVX-NEXT: vmulps %xmm1, %xmm0, %xmm0
726+
; AVX-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[0,0,0,0]
727+
; AVX-NEXT: retq
728+
%b = fmul fast <4 x float> %vx, <float 6.0, float -1.0, float 1.0, float 7.0>
729+
%r = shufflevector <4 x float> %b, <4 x float> undef, <4 x i32> zeroinitializer
730+
ret <4 x float> %r
731+
}
732+
733+
define <8 x float> @splat0_fdiv_const_op1_v8f32(<8 x float> %vx) {
734+
; SSE-LABEL: splat0_fdiv_const_op1_v8f32:
735+
; SSE: # %bb.0:
736+
; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0,0,0]
737+
; SSE-NEXT: movaps %xmm0, %xmm1
738+
; SSE-NEXT: retq
739+
;
740+
; AVX-LABEL: splat0_fdiv_const_op1_v8f32:
741+
; AVX: # %bb.0:
742+
; AVX-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
743+
; AVX-NEXT: vrcpps %ymm1, %ymm1
744+
; AVX-NEXT: vmovaps {{.*#+}} xmm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
745+
; AVX-NEXT: vsubps %xmm1, %xmm2, %xmm2
746+
; AVX-NEXT: vmulps %xmm2, %xmm1, %xmm2
747+
; AVX-NEXT: vaddps %xmm2, %xmm1, %xmm1
748+
; AVX-NEXT: vmulps %xmm1, %xmm0, %xmm0
749+
; AVX-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[0,0,0,0]
750+
; AVX-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0
751+
; AVX-NEXT: retq
752+
%b = fdiv fast <8 x float> %vx, <float 1.0, float 2.0, float 3.0, float 4.0, float 5.0, float 6.0, float 7.0, float 8.0>
753+
%r = shufflevector <8 x float> %b, <8 x float> undef, <8 x i32> zeroinitializer
754+
ret <8 x float> %r
755+
}
756+
757+
define <8 x float> @splat0_fdiv_const_op0_v8f32(<8 x float> %vx) {
758+
; SSE-LABEL: splat0_fdiv_const_op0_v8f32:
759+
; SSE: # %bb.0:
760+
; SSE-NEXT: rcpps %xmm0, %xmm2
761+
; SSE-NEXT: mulps %xmm2, %xmm0
762+
; SSE-NEXT: movaps {{.*#+}} xmm1 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0]
763+
; SSE-NEXT: subps %xmm0, %xmm1
764+
; SSE-NEXT: mulps %xmm2, %xmm1
765+
; SSE-NEXT: addps %xmm2, %xmm1
766+
; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,0,0,0]
767+
; SSE-NEXT: movaps %xmm1, %xmm0
768+
; SSE-NEXT: retq
769+
;
770+
; AVX-LABEL: splat0_fdiv_const_op0_v8f32:
771+
; AVX: # %bb.0:
772+
; AVX-NEXT: vrcpps %ymm0, %ymm1
773+
; AVX-NEXT: vmulps %xmm1, %xmm0, %xmm0
774+
; AVX-NEXT: vmovaps {{.*#+}} xmm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
775+
; AVX-NEXT: vsubps %xmm0, %xmm2, %xmm0
776+
; AVX-NEXT: vmulps %xmm0, %xmm1, %xmm0
777+
; AVX-NEXT: vaddps %xmm0, %xmm1, %xmm0
778+
; AVX-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[0,0,0,0]
779+
; AVX-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0
780+
; AVX-NEXT: retq
781+
%b = fdiv fast <8 x float> <float 1.0, float 2.0, float 3.0, float 4.0, float 5.0, float 6.0, float 7.0, float 8.0>, %vx
782+
%r = shufflevector <8 x float> %b, <8 x float> undef, <8 x i32> zeroinitializer
783+
ret <8 x float> %r
784+
}

0 commit comments

Comments
 (0)