Skip to content

Commit 32ed303

Browse files
RKSimonRKSimon
committed
[X86] Add test coverage for Issue #63946 
1 parent 70893b6 commit 32ed303

File tree

1 file changed

+157
-0
lines changed

1 file changed

+157
-0
lines changed

‎llvm/test/CodeGen/X86/sext-vsetcc.ll

Lines changed: 157 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -566,3 +566,160 @@ define <4 x i64> @PR50055_signed(ptr %src, ptr %dst) {
566566
store <8 x i32> %sext, ptr %dst, align 32
567567
ret <4 x i64> %t2
568568
}
569+
570+
define <8 x i32> @PR63946(<8 x i32> %a0, <8 x i32> %b0) nounwind {
571+
; SSE-LABEL: PR63946:
572+
; SSE: # %bb.0: # %entry
573+
; SSE-NEXT: movdqa %xmm1, %xmm4
574+
; SSE-NEXT: pshufd {{.*#+}} xmm6 = xmm2[1,2,3,0]
575+
; SSE-NEXT: pshufd {{.*#+}} xmm8 = xmm3[1,2,3,0]
576+
; SSE-NEXT: pshufd {{.*#+}} xmm5 = xmm3[2,3,0,1]
577+
; SSE-NEXT: pshufd {{.*#+}} xmm9 = xmm2[2,3,0,1]
578+
; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm3[3,0,1,2]
579+
; SSE-NEXT: pshufd {{.*#+}} xmm10 = xmm2[3,0,1,2]
580+
; SSE-NEXT: movdqa %xmm0, %xmm7
581+
; SSE-NEXT: pcmpeqd %xmm2, %xmm7
582+
; SSE-NEXT: movdqa %xmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
583+
; SSE-NEXT: movdqa %xmm8, %xmm11
584+
; SSE-NEXT: pcmpeqd %xmm4, %xmm11
585+
; SSE-NEXT: movdqa %xmm6, %xmm12
586+
; SSE-NEXT: movdqa %xmm9, %xmm13
587+
; SSE-NEXT: movdqa %xmm5, %xmm14
588+
; SSE-NEXT: pcmpeqd %xmm4, %xmm14
589+
; SSE-NEXT: movdqa %xmm1, %xmm15
590+
; SSE-NEXT: pcmpeqd %xmm4, %xmm15
591+
; SSE-NEXT: pcmpeqd %xmm4, %xmm2
592+
; SSE-NEXT: pcmpeqd %xmm4, %xmm6
593+
; SSE-NEXT: pcmpeqd %xmm4, %xmm9
594+
; SSE-NEXT: movdqa %xmm10, %xmm7
595+
; SSE-NEXT: pcmpeqd %xmm4, %xmm10
596+
; SSE-NEXT: pcmpeqd %xmm3, %xmm4
597+
; SSE-NEXT: por %xmm4, %xmm11
598+
; SSE-NEXT: pcmpeqd %xmm0, %xmm12
599+
; SSE-NEXT: por {{[-0-9]+}}(%r{{[sb]}}p), %xmm12 # 16-byte Folded Reload
600+
; SSE-NEXT: pcmpeqd %xmm0, %xmm13
601+
; SSE-NEXT: pcmpeqd %xmm0, %xmm7
602+
; SSE-NEXT: por %xmm14, %xmm2
603+
; SSE-NEXT: por %xmm11, %xmm2
604+
; SSE-NEXT: pcmpeqd %xmm0, %xmm3
605+
; SSE-NEXT: por %xmm3, %xmm13
606+
; SSE-NEXT: por %xmm12, %xmm13
607+
; SSE-NEXT: por %xmm15, %xmm6
608+
; SSE-NEXT: pcmpeqd %xmm0, %xmm8
609+
; SSE-NEXT: por %xmm7, %xmm8
610+
; SSE-NEXT: pcmpeqd %xmm0, %xmm5
611+
; SSE-NEXT: por %xmm8, %xmm5
612+
; SSE-NEXT: por %xmm13, %xmm5
613+
; SSE-NEXT: por %xmm6, %xmm9
614+
; SSE-NEXT: por %xmm2, %xmm9
615+
; SSE-NEXT: packssdw %xmm9, %xmm5
616+
; SSE-NEXT: pcmpeqd %xmm0, %xmm1
617+
; SSE-NEXT: packssdw %xmm10, %xmm1
618+
; SSE-NEXT: por %xmm5, %xmm1
619+
; SSE-NEXT: movdqa %xmm1, %xmm0
620+
; SSE-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3]
621+
; SSE-NEXT: pslld $31, %xmm0
622+
; SSE-NEXT: psrad $31, %xmm0
623+
; SSE-NEXT: punpckhwd {{.*#+}} xmm1 = xmm1[4,4,5,5,6,6,7,7]
624+
; SSE-NEXT: pslld $31, %xmm1
625+
; SSE-NEXT: psrad $31, %xmm1
626+
; SSE-NEXT: retq
627+
;
628+
; AVX2-LABEL: PR63946:
629+
; AVX2: # %bb.0: # %entry
630+
; AVX2-NEXT: vpshufd {{.*#+}} ymm2 = ymm1[1,2,3,0,5,6,7,4]
631+
; AVX2-NEXT: vpshufd {{.*#+}} ymm3 = ymm1[2,3,0,1,6,7,4,5]
632+
; AVX2-NEXT: vpshufd {{.*#+}} ymm4 = ymm1[3,0,1,2,7,4,5,6]
633+
; AVX2-NEXT: vpermq {{.*#+}} ymm5 = ymm1[2,3,0,1]
634+
; AVX2-NEXT: vpermq {{.*#+}} ymm6 = ymm2[2,3,0,1]
635+
; AVX2-NEXT: vpermq {{.*#+}} ymm7 = ymm3[2,3,0,1]
636+
; AVX2-NEXT: vpermq {{.*#+}} ymm8 = ymm4[2,3,0,1]
637+
; AVX2-NEXT: vpcmpeqd %ymm1, %ymm0, %ymm1
638+
; AVX2-NEXT: vpcmpeqd %ymm0, %ymm2, %ymm2
639+
; AVX2-NEXT: vpor %ymm1, %ymm2, %ymm1
640+
; AVX2-NEXT: vpcmpeqd %ymm0, %ymm3, %ymm2
641+
; AVX2-NEXT: vextracti128 $1, %ymm2, %xmm3
642+
; AVX2-NEXT: vpcmpeqd %ymm0, %ymm4, %ymm4
643+
; AVX2-NEXT: vextracti128 $1, %ymm4, %xmm9
644+
; AVX2-NEXT: vpcmpeqd %ymm0, %ymm5, %ymm5
645+
; AVX2-NEXT: vextracti128 $1, %ymm5, %xmm10
646+
; AVX2-NEXT: vpor %xmm3, %xmm10, %xmm3
647+
; AVX2-NEXT: vpcmpeqd %ymm0, %ymm6, %ymm6
648+
; AVX2-NEXT: vextracti128 $1, %ymm6, %xmm10
649+
; AVX2-NEXT: vpor %xmm10, %xmm9, %xmm9
650+
; AVX2-NEXT: vpcmpeqd %ymm0, %ymm7, %ymm7
651+
; AVX2-NEXT: vextracti128 $1, %ymm7, %xmm10
652+
; AVX2-NEXT: vpackssdw %xmm10, %xmm7, %xmm7
653+
; AVX2-NEXT: vpcmpeqd %ymm0, %ymm8, %ymm0
654+
; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm8
655+
; AVX2-NEXT: vpackssdw %xmm8, %xmm0, %xmm0
656+
; AVX2-NEXT: vpor %xmm0, %xmm7, %xmm0
657+
; AVX2-NEXT: vextracti128 $1, %ymm1, %xmm7
658+
; AVX2-NEXT: vpor %xmm3, %xmm7, %xmm3
659+
; AVX2-NEXT: vpor %xmm3, %xmm9, %xmm3
660+
; AVX2-NEXT: vpor %xmm5, %xmm2, %xmm2
661+
; AVX2-NEXT: vpor %xmm2, %xmm1, %xmm1
662+
; AVX2-NEXT: vpor %xmm6, %xmm4, %xmm2
663+
; AVX2-NEXT: vpor %xmm2, %xmm1, %xmm1
664+
; AVX2-NEXT: vpackssdw %xmm3, %xmm1, %xmm1
665+
; AVX2-NEXT: vpor %xmm0, %xmm1, %xmm0
666+
; AVX2-NEXT: vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
667+
; AVX2-NEXT: vpslld $31, %ymm0, %ymm0
668+
; AVX2-NEXT: vpsrad $31, %ymm0, %ymm0
669+
; AVX2-NEXT: retq
670+
;
671+
; AVX512-LABEL: PR63946:
672+
; AVX512: # %bb.0: # %entry
673+
; AVX512-NEXT: # kill: def $ymm1 killed $ymm1 def $zmm1
674+
; AVX512-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
675+
; AVX512-NEXT: vpshufd {{.*#+}} ymm2 = ymm1[1,2,3,0,5,6,7,4]
676+
; AVX512-NEXT: vpshufd {{.*#+}} ymm3 = ymm1[2,3,0,1,6,7,4,5]
677+
; AVX512-NEXT: vpshufd {{.*#+}} ymm4 = ymm1[3,0,1,2,7,4,5,6]
678+
; AVX512-NEXT: vpermq {{.*#+}} ymm5 = ymm1[2,3,0,1]
679+
; AVX512-NEXT: vpermq {{.*#+}} ymm6 = ymm2[2,3,0,1]
680+
; AVX512-NEXT: vpermq {{.*#+}} ymm7 = ymm3[2,3,0,1]
681+
; AVX512-NEXT: vpermq {{.*#+}} ymm8 = ymm4[2,3,0,1]
682+
; AVX512-NEXT: vpcmpeqd %zmm1, %zmm0, %k0
683+
; AVX512-NEXT: vpcmpeqd %zmm0, %zmm2, %k1
684+
; AVX512-NEXT: vpcmpeqd %zmm0, %zmm3, %k3
685+
; AVX512-NEXT: vpcmpeqd %zmm0, %zmm4, %k2
686+
; AVX512-NEXT: vpcmpeqd %zmm0, %zmm5, %k4
687+
; AVX512-NEXT: vpcmpeqd %zmm0, %zmm6, %k5
688+
; AVX512-NEXT: vpcmpeqd %zmm0, %zmm7, %k6
689+
; AVX512-NEXT: vpcmpeqd %zmm0, %zmm8, %k7
690+
; AVX512-NEXT: korw %k0, %k1, %k0
691+
; AVX512-NEXT: korw %k3, %k0, %k0
692+
; AVX512-NEXT: korw %k4, %k0, %k0
693+
; AVX512-NEXT: korw %k2, %k0, %k0
694+
; AVX512-NEXT: korw %k5, %k0, %k0
695+
; AVX512-NEXT: korw %k6, %k0, %k0
696+
; AVX512-NEXT: korw %k7, %k0, %k1
697+
; AVX512-NEXT: vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
698+
; AVX512-NEXT: # kill: def $ymm0 killed $ymm0 killed $zmm0
699+
; AVX512-NEXT: retq
700+
entry:
701+
%shuffle = shufflevector <8 x i32> %b0, <8 x i32> poison, <8 x i32> <i32 1, i32 2, i32 3, i32 0, i32 5, i32 6, i32 7, i32 4>
702+
%shuffle1 = shufflevector <8 x i32> %b0, <8 x i32> poison, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 6, i32 7, i32 4, i32 5>
703+
%shuffle2 = shufflevector <8 x i32> %shuffle, <8 x i32> poison, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 6, i32 7, i32 4, i32 5>
704+
%shuffle3 = shufflevector <8 x i32> %b0, <8 x i32> poison, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
705+
%shuffle4 = shufflevector <8 x i32> %shuffle, <8 x i32> poison, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
706+
%shuffle5 = shufflevector <8 x i32> %shuffle1, <8 x i32> poison, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
707+
%shuffle6 = shufflevector <8 x i32> %shuffle2, <8 x i32> poison, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
708+
%cmp = icmp eq <8 x i32> %a0, %b0
709+
%cmp7 = icmp eq <8 x i32> %shuffle, %a0
710+
%cmp9 = icmp eq <8 x i32> %shuffle1, %a0
711+
%cmp11 = icmp eq <8 x i32> %shuffle2, %a0
712+
%cmp13 = icmp eq <8 x i32> %shuffle3, %a0
713+
%cmp15 = icmp eq <8 x i32> %shuffle4, %a0
714+
%cmp17 = icmp eq <8 x i32> %shuffle5, %a0
715+
%cmp19 = icmp eq <8 x i32> %shuffle6, %a0
716+
%or2365 = or <8 x i1> %cmp7, %cmp
717+
%or2264 = or <8 x i1> %or2365, %cmp9
718+
%or2567 = or <8 x i1> %or2264, %cmp13
719+
%or2163 = or <8 x i1> %or2567, %cmp11
720+
%or62 = or <8 x i1> %or2163, %cmp15
721+
%or2466 = or <8 x i1> %or62, %cmp17
722+
%or2668 = or <8 x i1> %or2466, %cmp19
723+
%or26 = sext <8 x i1> %or2668 to <8 x i32>
724+
ret <8 x i32> %or26
725+
}

0 commit comments

Comments
 (0)