; SSE-LABEL: combine_vec_sdiv_by_pow2b_v16i8:
; SSE: # %bb.0:
; SSE-NEXT: movdqa %xmm0, %xmm1
-; SSE-NEXT: pxor %xmm2, %xmm2
-; SSE-NEXT: pcmpgtb %xmm0, %xmm2
-; SSE-NEXT: movdqa %xmm2, %xmm3
-; SSE-NEXT: psrlw $4, %xmm3
-; SSE-NEXT: pand {{.*}}(%rip), %xmm3
-; SSE-NEXT: movdqa {{.*#+}} xmm0 = [49408,32992,24736,57408,49408,32992,24736,57408]
-; SSE-NEXT: pblendvb %xmm0, %xmm3, %xmm2
-; SSE-NEXT: movdqa %xmm2, %xmm3
-; SSE-NEXT: psrlw $2, %xmm3
-; SSE-NEXT: pand {{.*}}(%rip), %xmm3
-; SSE-NEXT: paddb %xmm0, %xmm0
-; SSE-NEXT: pblendvb %xmm0, %xmm3, %xmm2
-; SSE-NEXT: movdqa %xmm2, %xmm3
-; SSE-NEXT: psrlw $1, %xmm3
-; SSE-NEXT: pand {{.*}}(%rip), %xmm3
-; SSE-NEXT: paddb %xmm0, %xmm0
-; SSE-NEXT: pblendvb %xmm0, %xmm3, %xmm2
-; SSE-NEXT: paddb %xmm1, %xmm2
-; SSE-NEXT: punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm2[8],xmm3[9],xmm2[9],xmm3[10],xmm2[10],xmm3[11],xmm2[11],xmm3[12],xmm2[12],xmm3[13],xmm2[13],xmm3[14],xmm2[14],xmm3[15],xmm2[15]
-; SSE-NEXT: movdqa %xmm3, %xmm4
-; SSE-NEXT: psraw $4, %xmm4
-; SSE-NEXT: movdqa {{.*#+}} xmm5 = [16384,32800,41056,8384,16384,32800,41056,8384]
-; SSE-NEXT: punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm5[8],xmm0[9],xmm5[9],xmm0[10],xmm5[10],xmm0[11],xmm5[11],xmm0[12],xmm5[12],xmm0[13],xmm5[13],xmm0[14],xmm5[14],xmm0[15],xmm5[15]
-; SSE-NEXT: pblendvb %xmm0, %xmm4, %xmm3
-; SSE-NEXT: movdqa %xmm3, %xmm4
-; SSE-NEXT: psraw $2, %xmm4
-; SSE-NEXT: paddw %xmm0, %xmm0
-; SSE-NEXT: pblendvb %xmm0, %xmm4, %xmm3
-; SSE-NEXT: movdqa %xmm3, %xmm4
-; SSE-NEXT: psraw $1, %xmm4
-; SSE-NEXT: paddw %xmm0, %xmm0
-; SSE-NEXT: pblendvb %xmm0, %xmm4, %xmm3
+; SSE-NEXT: pxor %xmm0, %xmm0
+; SSE-NEXT: pcmpgtb %xmm1, %xmm0
+; SSE-NEXT: pmovzxbw {{.*#+}} xmm3 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; SSE-NEXT: movdqa {{.*#+}} xmm2 = [1,4,2,16,8,32,64,2]
+; SSE-NEXT: pmullw %xmm2, %xmm3
; SSE-NEXT: psrlw $8, %xmm3
-; SSE-NEXT: punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
+; SSE-NEXT: pmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; SSE-NEXT: pmullw %xmm2, %xmm0
+; SSE-NEXT: psrlw $8, %xmm0
+; SSE-NEXT: packuswb %xmm0, %xmm3
+; SSE-NEXT: paddb %xmm1, %xmm3
+; SSE-NEXT: punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm3[8],xmm2[9],xmm3[9],xmm2[10],xmm3[10],xmm2[11],xmm3[11],xmm2[12],xmm3[12],xmm2[13],xmm3[13],xmm2[14],xmm3[14],xmm2[15],xmm3[15]
; SSE-NEXT: movdqa %xmm2, %xmm4
; SSE-NEXT: psraw $4, %xmm4
-; SSE-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm5[0],xmm0[1],xmm5[1],xmm0[2],xmm5[2],xmm0[3],xmm5[3],xmm0[4],xmm5[4],xmm0[5],xmm5[5],xmm0[6],xmm5[6],xmm0[7],xmm5[7]
+; SSE-NEXT: movdqa {{.*#+}} xmm5 = [16384,32800,41056,8384,16384,32800,41056,8384]
+; SSE-NEXT: punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm5[8],xmm0[9],xmm5[9],xmm0[10],xmm5[10],xmm0[11],xmm5[11],xmm0[12],xmm5[12],xmm0[13],xmm5[13],xmm0[14],xmm5[14],xmm0[15],xmm5[15]
; SSE-NEXT: pblendvb %xmm0, %xmm4, %xmm2
; SSE-NEXT: movdqa %xmm2, %xmm4
; SSE-NEXT: psraw $2, %xmm4
; SSE-NEXT: paddw %xmm0, %xmm0
; SSE-NEXT: pblendvb %xmm0, %xmm4, %xmm2
; SSE-NEXT: psrlw $8, %xmm2
-; SSE-NEXT: packuswb %xmm3, %xmm2
+; SSE-NEXT: punpcklbw {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE-NEXT: movdqa %xmm3, %xmm4
+; SSE-NEXT: psraw $4, %xmm4
+; SSE-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm5[0],xmm0[1],xmm5[1],xmm0[2],xmm5[2],xmm0[3],xmm5[3],xmm0[4],xmm5[4],xmm0[5],xmm5[5],xmm0[6],xmm5[6],xmm0[7],xmm5[7]
+; SSE-NEXT: pblendvb %xmm0, %xmm4, %xmm3
+; SSE-NEXT: movdqa %xmm3, %xmm4
+; SSE-NEXT: psraw $2, %xmm4
+; SSE-NEXT: paddw %xmm0, %xmm0
+; SSE-NEXT: pblendvb %xmm0, %xmm4, %xmm3
+; SSE-NEXT: movdqa %xmm3, %xmm4
+; SSE-NEXT: psraw $1, %xmm4
+; SSE-NEXT: paddw %xmm0, %xmm0
+; SSE-NEXT: pblendvb %xmm0, %xmm4, %xmm3
+; SSE-NEXT: psrlw $8, %xmm3
+; SSE-NEXT: packuswb %xmm2, %xmm3
; SSE-NEXT: movaps {{.*#+}} xmm0 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
-; SSE-NEXT: pblendvb %xmm0, %xmm2, %xmm1
+; SSE-NEXT: pblendvb %xmm0, %xmm3, %xmm1
; SSE-NEXT: movdqa %xmm1, %xmm0
; SSE-NEXT: retq
;
; AVX1: # %bb.0:
; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX1-NEXT: vpcmpgtb %xmm0, %xmm1, %xmm1
-; AVX1-NEXT: vpsrlw $4, %xmm1, %xmm2
-; AVX1-NEXT: vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [49408,32992,24736,57408,49408,32992,24736,57408]
-; AVX1-NEXT: vpblendvb %xmm3, %xmm2, %xmm1, %xmm1
-; AVX1-NEXT: vpsrlw $2, %xmm1, %xmm2
-; AVX1-NEXT: vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX1-NEXT: vpaddb %xmm3, %xmm3, %xmm3
-; AVX1-NEXT: vpblendvb %xmm3, %xmm2, %xmm1, %xmm1
-; AVX1-NEXT: vpsrlw $1, %xmm1, %xmm2
-; AVX1-NEXT: vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX1-NEXT: vpaddb %xmm3, %xmm3, %xmm3
-; AVX1-NEXT: vpblendvb %xmm3, %xmm2, %xmm1, %xmm1
+; AVX1-NEXT: vpmovzxbw {{.*#+}} xmm2 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [1,4,2,16,8,32,64,2]
+; AVX1-NEXT: vpmullw %xmm3, %xmm2, %xmm2
+; AVX1-NEXT: vpsrlw $8, %xmm2, %xmm2
+; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
+; AVX1-NEXT: vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; AVX1-NEXT: vpmullw %xmm3, %xmm1, %xmm1
+; AVX1-NEXT: vpsrlw $8, %xmm1, %xmm1
+; AVX1-NEXT: vpackuswb %xmm1, %xmm2, %xmm1
; AVX1-NEXT: vpaddb %xmm1, %xmm0, %xmm1
; AVX1-NEXT: vpunpckhbw {{.*#+}} xmm2 = xmm0[8],xmm1[8],xmm0[9],xmm1[9],xmm0[10],xmm1[10],xmm0[11],xmm1[11],xmm0[12],xmm1[12],xmm0[13],xmm1[13],xmm0[14],xmm1[14],xmm0[15],xmm1[15]
; AVX1-NEXT: vpsraw $4, %xmm2, %xmm3
; AVX2: # %bb.0:
; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX2-NEXT: vpcmpgtb %xmm0, %xmm1, %xmm1
-; AVX2-NEXT: vpsrlw $4, %xmm1, %xmm2
-; AVX2-NEXT: vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX2-NEXT: vmovdqa {{.*#+}} xmm3 = [49408,32992,24736,57408,49408,32992,24736,57408]
-; AVX2-NEXT: vpblendvb %xmm3, %xmm2, %xmm1, %xmm1
-; AVX2-NEXT: vpsrlw $2, %xmm1, %xmm2
-; AVX2-NEXT: vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX2-NEXT: vpaddb %xmm3, %xmm3, %xmm3
-; AVX2-NEXT: vpblendvb %xmm3, %xmm2, %xmm1, %xmm1
-; AVX2-NEXT: vpsrlw $1, %xmm1, %xmm2
-; AVX2-NEXT: vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX2-NEXT: vpaddb %xmm3, %xmm3, %xmm3
-; AVX2-NEXT: vpblendvb %xmm3, %xmm2, %xmm1, %xmm1
+; AVX2-NEXT: vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
+; AVX2-NEXT: vpmullw {{.*}}(%rip), %ymm1, %ymm1
+; AVX2-NEXT: vpsrlw $8, %ymm1, %ymm1
+; AVX2-NEXT: vextracti128 $1, %ymm1, %xmm2
+; AVX2-NEXT: vpackuswb %xmm2, %xmm1, %xmm1
; AVX2-NEXT: vpaddb %xmm1, %xmm0, %xmm1
; AVX2-NEXT: vpunpckhbw {{.*#+}} xmm2 = xmm0[8],xmm1[8],xmm0[9],xmm1[9],xmm0[10],xmm1[10],xmm0[11],xmm1[11],xmm0[12],xmm1[12],xmm0[13],xmm1[13],xmm0[14],xmm1[14],xmm0[15],xmm1[15]
; AVX2-NEXT: vpsraw $4, %xmm2, %xmm3
; AVX2-NEXT: vpackuswb %xmm2, %xmm1, %xmm1
; AVX2-NEXT: vmovdqa {{.*#+}} xmm2 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
; AVX2-NEXT: vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
+; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
;
; AVX512F-LABEL: combine_vec_sdiv_by_pow2b_v16i8:
; SSE: # %bb.0:
; SSE-NEXT: movdqa %xmm0, %xmm1
; SSE-NEXT: psraw $15, %xmm1
-; SSE-NEXT: movdqa %xmm1, %xmm2
-; SSE-NEXT: psrlw $8, %xmm2
-; SSE-NEXT: pblendw {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3,4,5,6,7]
-; SSE-NEXT: movdqa %xmm2, %xmm1
-; SSE-NEXT: psrlw $4, %xmm1
-; SSE-NEXT: pblendw {{.*#+}} xmm1 = xmm2[0],xmm1[1,2,3,4],xmm2[5,6],xmm1[7]
-; SSE-NEXT: movdqa %xmm1, %xmm2
-; SSE-NEXT: psrlw $2, %xmm2
-; SSE-NEXT: pblendw {{.*#+}} xmm2 = xmm1[0],xmm2[1,2],xmm1[3,4],xmm2[5,6,7]
-; SSE-NEXT: movdqa %xmm2, %xmm1
-; SSE-NEXT: psrlw $1, %xmm1
-; SSE-NEXT: pblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2],xmm2[3],xmm1[4,5],xmm2[6],xmm1[7]
+; SSE-NEXT: pmulhuw {{.*}}(%rip), %xmm1
; SSE-NEXT: paddw %xmm0, %xmm1
; SSE-NEXT: movdqa %xmm1, %xmm2
; SSE-NEXT: psraw $4, %xmm2
; AVX1-LABEL: combine_vec_sdiv_by_pow2b_v8i16:
; AVX1: # %bb.0:
; AVX1-NEXT: vpsraw $15, %xmm0, %xmm1
-; AVX1-NEXT: vpsrlw $8, %xmm1, %xmm2
-; AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3,4,5,6,7]
-; AVX1-NEXT: vpsrlw $4, %xmm1, %xmm2
-; AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3,4],xmm1[5,6],xmm2[7]
-; AVX1-NEXT: vpsrlw $2, %xmm1, %xmm2
-; AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2],xmm1[3,4],xmm2[5,6,7]
-; AVX1-NEXT: vpsrlw $1, %xmm1, %xmm2
-; AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2],xmm1[3],xmm2[4,5],xmm1[6],xmm2[7]
+; AVX1-NEXT: vpmulhuw {{.*}}(%rip), %xmm1, %xmm1
; AVX1-NEXT: vpaddw %xmm1, %xmm0, %xmm1
; AVX1-NEXT: vpsraw $4, %xmm1, %xmm2
; AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1,2],xmm2[3],xmm1[4],xmm2[5,6],xmm1[7]
; AVX2-LABEL: combine_vec_sdiv_by_pow2b_v8i16:
; AVX2: # %bb.0:
; AVX2-NEXT: vpsraw $15, %xmm0, %xmm1
-; AVX2-NEXT: vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
-; AVX2-NEXT: vpsrlvd {{.*}}(%rip), %ymm1, %ymm1
-; AVX2-NEXT: vpshufb {{.*#+}} ymm1 = ymm1[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
-; AVX2-NEXT: vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
+; AVX2-NEXT: vpmulhuw {{.*}}(%rip), %xmm1, %xmm1
; AVX2-NEXT: vpaddw %xmm1, %xmm0, %xmm1
; AVX2-NEXT: vpmovsxwd %xmm1, %ymm1
; AVX2-NEXT: vpsravd {{.*}}(%rip), %ymm1, %ymm1
; AVX512F-LABEL: combine_vec_sdiv_by_pow2b_v8i16:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vpsraw $15, %xmm0, %xmm1
-; AVX512F-NEXT: vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
-; AVX512F-NEXT: vpsrlvd {{.*}}(%rip), %ymm1, %ymm1
-; AVX512F-NEXT: vpmovdw %zmm1, %ymm1
+; AVX512F-NEXT: vpmulhuw {{.*}}(%rip), %xmm1, %xmm1
; AVX512F-NEXT: vpaddw %xmm1, %xmm0, %xmm1
; AVX512F-NEXT: vpmovsxwd %xmm1, %ymm1
; AVX512F-NEXT: vpsravd {{.*}}(%rip), %ymm1, %ymm1
; SSE-LABEL: combine_vec_sdiv_by_pow2b_v16i16:
; SSE: # %bb.0:
; SSE-NEXT: movdqa %xmm0, %xmm2
+; SSE-NEXT: psraw $15, %xmm2
+; SSE-NEXT: movdqa {{.*#+}} xmm3 = [1,4,2,16,8,32,64,2]
+; SSE-NEXT: pmulhuw %xmm3, %xmm2
+; SSE-NEXT: paddw %xmm0, %xmm2
+; SSE-NEXT: movdqa %xmm2, %xmm4
+; SSE-NEXT: psraw $4, %xmm4
+; SSE-NEXT: pblendw {{.*#+}} xmm4 = xmm2[0,1,2],xmm4[3],xmm2[4],xmm4[5,6],xmm2[7]
+; SSE-NEXT: movdqa %xmm4, %xmm5
+; SSE-NEXT: psraw $2, %xmm5
+; SSE-NEXT: pblendw {{.*#+}} xmm5 = xmm4[0],xmm5[1],xmm4[2,3],xmm5[4],xmm4[5],xmm5[6],xmm4[7]
+; SSE-NEXT: movdqa %xmm5, %xmm2
+; SSE-NEXT: psraw $1, %xmm2
+; SSE-NEXT: pblendw {{.*#+}} xmm2 = xmm5[0,1],xmm2[2],xmm5[3],xmm2[4,5],xmm5[6],xmm2[7]
+; SSE-NEXT: pblendw {{.*#+}} xmm2 = xmm0[0],xmm2[1,2,3,4,5,6,7]
+; SSE-NEXT: movdqa %xmm1, %xmm0
; SSE-NEXT: psraw $15, %xmm0
-; SSE-NEXT: movdqa %xmm0, %xmm3
-; SSE-NEXT: psrlw $8, %xmm3
-; SSE-NEXT: pblendw {{.*#+}} xmm3 = xmm0[0],xmm3[1,2,3,4,5,6,7]
-; SSE-NEXT: movdqa %xmm3, %xmm0
-; SSE-NEXT: psrlw $4, %xmm0
-; SSE-NEXT: pblendw {{.*#+}} xmm0 = xmm3[0],xmm0[1,2,3,4],xmm3[5,6],xmm0[7]
-; SSE-NEXT: movdqa %xmm0, %xmm3
-; SSE-NEXT: psrlw $2, %xmm3
-; SSE-NEXT: pblendw {{.*#+}} xmm3 = xmm0[0],xmm3[1,2],xmm0[3,4],xmm3[5,6,7]
-; SSE-NEXT: movdqa %xmm3, %xmm0
-; SSE-NEXT: psrlw $1, %xmm0
-; SSE-NEXT: pblendw {{.*#+}} xmm0 = xmm3[0,1],xmm0[2],xmm3[3],xmm0[4,5],xmm3[6],xmm0[7]
-; SSE-NEXT: paddw %xmm2, %xmm0
+; SSE-NEXT: pmulhuw %xmm3, %xmm0
+; SSE-NEXT: paddw %xmm1, %xmm0
; SSE-NEXT: movdqa %xmm0, %xmm3
; SSE-NEXT: psraw $4, %xmm3
; SSE-NEXT: pblendw {{.*#+}} xmm3 = xmm0[0,1,2],xmm3[3],xmm0[4],xmm3[5,6],xmm0[7]
-; SSE-NEXT: movdqa %xmm3, %xmm4
-; SSE-NEXT: psraw $2, %xmm4
-; SSE-NEXT: pblendw {{.*#+}} xmm4 = xmm3[0],xmm4[1],xmm3[2,3],xmm4[4],xmm3[5],xmm4[6],xmm3[7]
-; SSE-NEXT: movdqa %xmm4, %xmm0
-; SSE-NEXT: psraw $1, %xmm0
-; SSE-NEXT: pblendw {{.*#+}} xmm0 = xmm4[0,1],xmm0[2],xmm4[3],xmm0[4,5],xmm4[6],xmm0[7]
-; SSE-NEXT: pblendw {{.*#+}} xmm0 = xmm2[0],xmm0[1,2,3,4,5,6,7]
-; SSE-NEXT: movdqa %xmm1, %xmm2
-; SSE-NEXT: psraw $15, %xmm2
-; SSE-NEXT: movdqa %xmm2, %xmm3
-; SSE-NEXT: psrlw $8, %xmm3
-; SSE-NEXT: pblendw {{.*#+}} xmm3 = xmm2[0],xmm3[1,2,3,4,5,6,7]
-; SSE-NEXT: movdqa %xmm3, %xmm2
-; SSE-NEXT: psrlw $4, %xmm2
-; SSE-NEXT: pblendw {{.*#+}} xmm2 = xmm3[0],xmm2[1,2,3,4],xmm3[5,6],xmm2[7]
-; SSE-NEXT: movdqa %xmm2, %xmm3
-; SSE-NEXT: psrlw $2, %xmm3
-; SSE-NEXT: pblendw {{.*#+}} xmm3 = xmm2[0],xmm3[1,2],xmm2[3,4],xmm3[5,6,7]
-; SSE-NEXT: movdqa %xmm3, %xmm2
-; SSE-NEXT: psrlw $1, %xmm2
-; SSE-NEXT: pblendw {{.*#+}} xmm2 = xmm3[0,1],xmm2[2],xmm3[3],xmm2[4,5],xmm3[6],xmm2[7]
-; SSE-NEXT: paddw %xmm1, %xmm2
-; SSE-NEXT: movdqa %xmm2, %xmm3
-; SSE-NEXT: psraw $4, %xmm3
-; SSE-NEXT: pblendw {{.*#+}} xmm3 = xmm2[0,1,2],xmm3[3],xmm2[4],xmm3[5,6],xmm2[7]
-; SSE-NEXT: movdqa %xmm3, %xmm4
-; SSE-NEXT: psraw $2, %xmm4
-; SSE-NEXT: pblendw {{.*#+}} xmm4 = xmm3[0],xmm4[1],xmm3[2,3],xmm4[4],xmm3[5],xmm4[6],xmm3[7]
-; SSE-NEXT: movdqa %xmm4, %xmm2
-; SSE-NEXT: psraw $1, %xmm2
-; SSE-NEXT: pblendw {{.*#+}} xmm2 = xmm4[0,1],xmm2[2],xmm4[3],xmm2[4,5],xmm4[6],xmm2[7]
-; SSE-NEXT: pblendw {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3,4,5,6,7]
-; SSE-NEXT: movdqa %xmm2, %xmm1
+; SSE-NEXT: movdqa %xmm3, %xmm0
+; SSE-NEXT: psraw $2, %xmm0
+; SSE-NEXT: pblendw {{.*#+}} xmm0 = xmm3[0],xmm0[1],xmm3[2,3],xmm0[4],xmm3[5],xmm0[6],xmm3[7]
+; SSE-NEXT: movdqa %xmm0, %xmm3
+; SSE-NEXT: psraw $1, %xmm3
+; SSE-NEXT: pblendw {{.*#+}} xmm3 = xmm0[0,1],xmm3[2],xmm0[3],xmm3[4,5],xmm0[6],xmm3[7]
+; SSE-NEXT: pblendw {{.*#+}} xmm3 = xmm1[0],xmm3[1,2,3,4,5,6,7]
+; SSE-NEXT: movdqa %xmm2, %xmm0
+; SSE-NEXT: movdqa %xmm3, %xmm1
; SSE-NEXT: retq
;
; AVX1-LABEL: combine_vec_sdiv_by_pow2b_v16i16:
; AVX1: # %bb.0:
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
; AVX1-NEXT: vpsraw $15, %xmm1, %xmm2
-; AVX1-NEXT: vpsrlw $8, %xmm2, %xmm3
-; AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0],xmm3[1,2,3,4,5,6,7]
-; AVX1-NEXT: vpsrlw $4, %xmm2, %xmm3
-; AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0],xmm3[1,2,3,4],xmm2[5,6],xmm3[7]
-; AVX1-NEXT: vpsrlw $2, %xmm2, %xmm3
-; AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0],xmm3[1,2],xmm2[3,4],xmm3[5,6,7]
-; AVX1-NEXT: vpsrlw $1, %xmm2, %xmm3
-; AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0,1],xmm3[2],xmm2[3],xmm3[4,5],xmm2[6],xmm3[7]
+; AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [1,4,2,16,8,32,64,2]
+; AVX1-NEXT: vpmulhuw %xmm3, %xmm2, %xmm2
; AVX1-NEXT: vpaddw %xmm2, %xmm1, %xmm1
; AVX1-NEXT: vpsraw $4, %xmm1, %xmm2
; AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1,2],xmm2[3],xmm1[4],xmm2[5,6],xmm1[7]
; AVX1-NEXT: vpsraw $1, %xmm1, %xmm2
; AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2],xmm1[3],xmm2[4,5],xmm1[6],xmm2[7]
; AVX1-NEXT: vpsraw $15, %xmm0, %xmm2
-; AVX1-NEXT: vpsrlw $8, %xmm2, %xmm3
-; AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0],xmm3[1,2,3,4,5,6,7]
-; AVX1-NEXT: vpsrlw $4, %xmm2, %xmm3
-; AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0],xmm3[1,2,3,4],xmm2[5,6],xmm3[7]
-; AVX1-NEXT: vpsrlw $2, %xmm2, %xmm3
-; AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0],xmm3[1,2],xmm2[3,4],xmm3[5,6,7]
-; AVX1-NEXT: vpsrlw $1, %xmm2, %xmm3
-; AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0,1],xmm3[2],xmm2[3],xmm3[4,5],xmm2[6],xmm3[7]
+; AVX1-NEXT: vpmulhuw %xmm3, %xmm2, %xmm2
; AVX1-NEXT: vpaddw %xmm2, %xmm0, %xmm2
; AVX1-NEXT: vpsraw $4, %xmm2, %xmm3
; AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0,1,2],xmm3[3],xmm2[4],xmm3[5,6],xmm2[7]
; AVX2-LABEL: combine_vec_sdiv_by_pow2b_v16i16:
; AVX2: # %bb.0:
; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX2-NEXT: vmovdqa {{.*#+}} ymm2 = [16,14,15,12,13,11,10,15,16,14,15,12,13,11,10,15]
+; AVX2-NEXT: vmovdqa {{.*#+}} ymm2 = [0,2,1,4,3,5,6,1,0,2,1,4,3,5,6,1]
; AVX2-NEXT: vpunpckhwd {{.*#+}} ymm3 = ymm2[4],ymm1[4],ymm2[5],ymm1[5],ymm2[6],ymm1[6],ymm2[7],ymm1[7],ymm2[12],ymm1[12],ymm2[13],ymm1[13],ymm2[14],ymm1[14],ymm2[15],ymm1[15]
; AVX2-NEXT: vpsraw $15, %ymm0, %ymm4
+; AVX2-NEXT: vpmulhuw {{.*}}(%rip), %ymm4, %ymm4
+; AVX2-NEXT: vpaddw %ymm4, %ymm0, %ymm4
; AVX2-NEXT: vpunpckhwd {{.*#+}} ymm5 = ymm1[4],ymm4[4],ymm1[5],ymm4[5],ymm1[6],ymm4[6],ymm1[7],ymm4[7],ymm1[12],ymm4[12],ymm1[13],ymm4[13],ymm1[14],ymm4[14],ymm1[15],ymm4[15]
-; AVX2-NEXT: vpsrlvd %ymm3, %ymm5, %ymm3
+; AVX2-NEXT: vpsravd %ymm3, %ymm5, %ymm3
; AVX2-NEXT: vpsrld $16, %ymm3, %ymm3
; AVX2-NEXT: vpunpcklwd {{.*#+}} ymm2 = ymm2[0],ymm1[0],ymm2[1],ymm1[1],ymm2[2],ymm1[2],ymm2[3],ymm1[3],ymm2[8],ymm1[8],ymm2[9],ymm1[9],ymm2[10],ymm1[10],ymm2[11],ymm1[11]
-; AVX2-NEXT: vpunpcklwd {{.*#+}} ymm4 = ymm1[0],ymm4[0],ymm1[1],ymm4[1],ymm1[2],ymm4[2],ymm1[3],ymm4[3],ymm1[8],ymm4[8],ymm1[9],ymm4[9],ymm1[10],ymm4[10],ymm1[11],ymm4[11]
-; AVX2-NEXT: vpsrlvd %ymm2, %ymm4, %ymm2
-; AVX2-NEXT: vpsrld $16, %ymm2, %ymm2
-; AVX2-NEXT: vpackusdw %ymm3, %ymm2, %ymm2
-; AVX2-NEXT: vpaddw %ymm2, %ymm0, %ymm2
-; AVX2-NEXT: vpunpckhwd {{.*#+}} ymm3 = ymm1[4],ymm2[4],ymm1[5],ymm2[5],ymm1[6],ymm2[6],ymm1[7],ymm2[7],ymm1[12],ymm2[12],ymm1[13],ymm2[13],ymm1[14],ymm2[14],ymm1[15],ymm2[15]
-; AVX2-NEXT: vmovdqa {{.*#+}} ymm4 = [0,2,1,4,3,5,6,1,0,2,1,4,3,5,6,1]
-; AVX2-NEXT: vpunpckhwd {{.*#+}} ymm5 = ymm4[4],ymm1[4],ymm4[5],ymm1[5],ymm4[6],ymm1[6],ymm4[7],ymm1[7],ymm4[12],ymm1[12],ymm4[13],ymm1[13],ymm4[14],ymm1[14],ymm4[15],ymm1[15]
-; AVX2-NEXT: vpsravd %ymm5, %ymm3, %ymm3
-; AVX2-NEXT: vpsrld $16, %ymm3, %ymm3
-; AVX2-NEXT: vpunpcklwd {{.*#+}} ymm2 = ymm1[0],ymm2[0],ymm1[1],ymm2[1],ymm1[2],ymm2[2],ymm1[3],ymm2[3],ymm1[8],ymm2[8],ymm1[9],ymm2[9],ymm1[10],ymm2[10],ymm1[11],ymm2[11]
-; AVX2-NEXT: vpunpcklwd {{.*#+}} ymm1 = ymm4[0],ymm1[0],ymm4[1],ymm1[1],ymm4[2],ymm1[2],ymm4[3],ymm1[3],ymm4[8],ymm1[8],ymm4[9],ymm1[9],ymm4[10],ymm1[10],ymm4[11],ymm1[11]
-; AVX2-NEXT: vpsravd %ymm1, %ymm2, %ymm1
+; AVX2-NEXT: vpunpcklwd {{.*#+}} ymm1 = ymm1[0],ymm4[0],ymm1[1],ymm4[1],ymm1[2],ymm4[2],ymm1[3],ymm4[3],ymm1[8],ymm4[8],ymm1[9],ymm4[9],ymm1[10],ymm4[10],ymm1[11],ymm4[11]
+; AVX2-NEXT: vpsravd %ymm2, %ymm1, %ymm1
; AVX2-NEXT: vpsrld $16, %ymm1, %ymm1
; AVX2-NEXT: vpackusdw %ymm3, %ymm1, %ymm1
; AVX2-NEXT: vpblendw {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3,4,5,6,7],ymm0[8],ymm1[9,10,11,12,13,14,15]
; AVX512F-LABEL: combine_vec_sdiv_by_pow2b_v16i16:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vpsraw $15, %ymm0, %ymm1
-; AVX512F-NEXT: vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
-; AVX512F-NEXT: vpsrlvd {{.*}}(%rip), %zmm1, %zmm1
-; AVX512F-NEXT: vpmovdw %zmm1, %ymm1
+; AVX512F-NEXT: vpmulhuw {{.*}}(%rip), %ymm1, %ymm1
; AVX512F-NEXT: vpaddw %ymm1, %ymm0, %ymm1
; AVX512F-NEXT: vpmovsxwd %ymm1, %zmm1
; AVX512F-NEXT: vpsravd {{.*}}(%rip), %zmm1, %zmm1
; SSE-NEXT: movdqa %xmm1, %xmm4
; SSE-NEXT: movdqa %xmm0, %xmm1
; SSE-NEXT: psraw $15, %xmm0
-; SSE-NEXT: movdqa %xmm0, %xmm5
-; SSE-NEXT: psrlw $8, %xmm5
-; SSE-NEXT: pblendw {{.*#+}} xmm5 = xmm0[0],xmm5[1,2,3,4,5,6,7]
-; SSE-NEXT: movdqa %xmm5, %xmm0
-; SSE-NEXT: psrlw $4, %xmm0
-; SSE-NEXT: pblendw {{.*#+}} xmm0 = xmm5[0],xmm0[1,2,3,4],xmm5[5,6],xmm0[7]
-; SSE-NEXT: movdqa %xmm0, %xmm5
-; SSE-NEXT: psrlw $2, %xmm5
-; SSE-NEXT: pblendw {{.*#+}} xmm5 = xmm0[0],xmm5[1,2],xmm0[3,4],xmm5[5,6,7]
-; SSE-NEXT: movdqa %xmm5, %xmm0
-; SSE-NEXT: psrlw $1, %xmm0
-; SSE-NEXT: pblendw {{.*#+}} xmm0 = xmm5[0,1],xmm0[2],xmm5[3],xmm0[4,5],xmm5[6],xmm0[7]
+; SSE-NEXT: movdqa {{.*#+}} xmm5 = [1,4,2,16,8,32,64,2]
+; SSE-NEXT: pmulhuw %xmm5, %xmm0
; SSE-NEXT: paddw %xmm1, %xmm0
-; SSE-NEXT: movdqa %xmm0, %xmm5
-; SSE-NEXT: psraw $4, %xmm5
-; SSE-NEXT: pblendw {{.*#+}} xmm5 = xmm0[0,1,2],xmm5[3],xmm0[4],xmm5[5,6],xmm0[7]
-; SSE-NEXT: movdqa %xmm5, %xmm6
-; SSE-NEXT: psraw $2, %xmm6
-; SSE-NEXT: pblendw {{.*#+}} xmm6 = xmm5[0],xmm6[1],xmm5[2,3],xmm6[4],xmm5[5],xmm6[6],xmm5[7]
-; SSE-NEXT: movdqa %xmm6, %xmm0
+; SSE-NEXT: movdqa %xmm0, %xmm6
+; SSE-NEXT: psraw $4, %xmm6
+; SSE-NEXT: pblendw {{.*#+}} xmm6 = xmm0[0,1,2],xmm6[3],xmm0[4],xmm6[5,6],xmm0[7]
+; SSE-NEXT: movdqa %xmm6, %xmm7
+; SSE-NEXT: psraw $2, %xmm7
+; SSE-NEXT: pblendw {{.*#+}} xmm7 = xmm6[0],xmm7[1],xmm6[2,3],xmm7[4],xmm6[5],xmm7[6],xmm6[7]
+; SSE-NEXT: movdqa %xmm7, %xmm0
; SSE-NEXT: psraw $1, %xmm0
-; SSE-NEXT: pblendw {{.*#+}} xmm0 = xmm6[0,1],xmm0[2],xmm6[3],xmm0[4,5],xmm6[6],xmm0[7]
+; SSE-NEXT: pblendw {{.*#+}} xmm0 = xmm7[0,1],xmm0[2],xmm7[3],xmm0[4,5],xmm7[6],xmm0[7]
; SSE-NEXT: pblendw {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3,4,5,6,7]
; SSE-NEXT: movdqa %xmm4, %xmm1
; SSE-NEXT: psraw $15, %xmm1
-; SSE-NEXT: movdqa %xmm1, %xmm5
-; SSE-NEXT: psrlw $8, %xmm5
-; SSE-NEXT: pblendw {{.*#+}} xmm5 = xmm1[0],xmm5[1,2,3,4,5,6,7]
-; SSE-NEXT: movdqa %xmm5, %xmm1
-; SSE-NEXT: psrlw $4, %xmm1
-; SSE-NEXT: pblendw {{.*#+}} xmm1 = xmm5[0],xmm1[1,2,3,4],xmm5[5,6],xmm1[7]
-; SSE-NEXT: movdqa %xmm1, %xmm5
-; SSE-NEXT: psrlw $2, %xmm5
-; SSE-NEXT: pblendw {{.*#+}} xmm5 = xmm1[0],xmm5[1,2],xmm1[3,4],xmm5[5,6,7]
-; SSE-NEXT: movdqa %xmm5, %xmm1
-; SSE-NEXT: psrlw $1, %xmm1
-; SSE-NEXT: pblendw {{.*#+}} xmm1 = xmm5[0,1],xmm1[2],xmm5[3],xmm1[4,5],xmm5[6],xmm1[7]
+; SSE-NEXT: pmulhuw %xmm5, %xmm1
; SSE-NEXT: paddw %xmm4, %xmm1
-; SSE-NEXT: movdqa %xmm1, %xmm5
-; SSE-NEXT: psraw $4, %xmm5
-; SSE-NEXT: pblendw {{.*#+}} xmm5 = xmm1[0,1,2],xmm5[3],xmm1[4],xmm5[5,6],xmm1[7]
-; SSE-NEXT: movdqa %xmm5, %xmm6
-; SSE-NEXT: psraw $2, %xmm6
-; SSE-NEXT: pblendw {{.*#+}} xmm6 = xmm5[0],xmm6[1],xmm5[2,3],xmm6[4],xmm5[5],xmm6[6],xmm5[7]
-; SSE-NEXT: movdqa %xmm6, %xmm1
+; SSE-NEXT: movdqa %xmm1, %xmm6
+; SSE-NEXT: psraw $4, %xmm6
+; SSE-NEXT: pblendw {{.*#+}} xmm6 = xmm1[0,1,2],xmm6[3],xmm1[4],xmm6[5,6],xmm1[7]
+; SSE-NEXT: movdqa %xmm6, %xmm7
+; SSE-NEXT: psraw $2, %xmm7
+; SSE-NEXT: pblendw {{.*#+}} xmm7 = xmm6[0],xmm7[1],xmm6[2,3],xmm7[4],xmm6[5],xmm7[6],xmm6[7]
+; SSE-NEXT: movdqa %xmm7, %xmm1
; SSE-NEXT: psraw $1, %xmm1
-; SSE-NEXT: pblendw {{.*#+}} xmm1 = xmm6[0,1],xmm1[2],xmm6[3],xmm1[4,5],xmm6[6],xmm1[7]
+; SSE-NEXT: pblendw {{.*#+}} xmm1 = xmm7[0,1],xmm1[2],xmm7[3],xmm1[4,5],xmm7[6],xmm1[7]
; SSE-NEXT: pblendw {{.*#+}} xmm1 = xmm4[0],xmm1[1,2,3,4,5,6,7]
; SSE-NEXT: movdqa %xmm2, %xmm4
; SSE-NEXT: psraw $15, %xmm4
-; SSE-NEXT: movdqa %xmm4, %xmm5
-; SSE-NEXT: psrlw $8, %xmm5
-; SSE-NEXT: pblendw {{.*#+}} xmm5 = xmm4[0],xmm5[1,2,3,4,5,6,7]
-; SSE-NEXT: movdqa %xmm5, %xmm4
-; SSE-NEXT: psrlw $4, %xmm4
-; SSE-NEXT: pblendw {{.*#+}} xmm4 = xmm5[0],xmm4[1,2,3,4],xmm5[5,6],xmm4[7]
-; SSE-NEXT: movdqa %xmm4, %xmm5
-; SSE-NEXT: psrlw $2, %xmm5
-; SSE-NEXT: pblendw {{.*#+}} xmm5 = xmm4[0],xmm5[1,2],xmm4[3,4],xmm5[5,6,7]
-; SSE-NEXT: movdqa %xmm5, %xmm4
-; SSE-NEXT: psrlw $1, %xmm4
-; SSE-NEXT: pblendw {{.*#+}} xmm4 = xmm5[0,1],xmm4[2],xmm5[3],xmm4[4,5],xmm5[6],xmm4[7]
+; SSE-NEXT: pmulhuw %xmm5, %xmm4
; SSE-NEXT: paddw %xmm2, %xmm4
-; SSE-NEXT: movdqa %xmm4, %xmm5
-; SSE-NEXT: psraw $4, %xmm5
-; SSE-NEXT: pblendw {{.*#+}} xmm5 = xmm4[0,1,2],xmm5[3],xmm4[4],xmm5[5,6],xmm4[7]
-; SSE-NEXT: movdqa %xmm5, %xmm6
-; SSE-NEXT: psraw $2, %xmm6
-; SSE-NEXT: pblendw {{.*#+}} xmm6 = xmm5[0],xmm6[1],xmm5[2,3],xmm6[4],xmm5[5],xmm6[6],xmm5[7]
-; SSE-NEXT: movdqa %xmm6, %xmm4
+; SSE-NEXT: movdqa %xmm4, %xmm6
+; SSE-NEXT: psraw $4, %xmm6
+; SSE-NEXT: pblendw {{.*#+}} xmm6 = xmm4[0,1,2],xmm6[3],xmm4[4],xmm6[5,6],xmm4[7]
+; SSE-NEXT: movdqa %xmm6, %xmm7
+; SSE-NEXT: psraw $2, %xmm7
+; SSE-NEXT: pblendw {{.*#+}} xmm7 = xmm6[0],xmm7[1],xmm6[2,3],xmm7[4],xmm6[5],xmm7[6],xmm6[7]
+; SSE-NEXT: movdqa %xmm7, %xmm4
; SSE-NEXT: psraw $1, %xmm4
-; SSE-NEXT: pblendw {{.*#+}} xmm4 = xmm6[0,1],xmm4[2],xmm6[3],xmm4[4,5],xmm6[6],xmm4[7]
+; SSE-NEXT: pblendw {{.*#+}} xmm4 = xmm7[0,1],xmm4[2],xmm7[3],xmm4[4,5],xmm7[6],xmm4[7]
; SSE-NEXT: pblendw {{.*#+}} xmm4 = xmm2[0],xmm4[1,2,3,4,5,6,7]
; SSE-NEXT: movdqa %xmm3, %xmm2
; SSE-NEXT: psraw $15, %xmm2
-; SSE-NEXT: movdqa %xmm2, %xmm5
-; SSE-NEXT: psrlw $8, %xmm5
-; SSE-NEXT: pblendw {{.*#+}} xmm5 = xmm2[0],xmm5[1,2,3,4,5,6,7]
-; SSE-NEXT: movdqa %xmm5, %xmm2
-; SSE-NEXT: psrlw $4, %xmm2
-; SSE-NEXT: pblendw {{.*#+}} xmm2 = xmm5[0],xmm2[1,2,3,4],xmm5[5,6],xmm2[7]
-; SSE-NEXT: movdqa %xmm2, %xmm5
-; SSE-NEXT: psrlw $2, %xmm5
-; SSE-NEXT: pblendw {{.*#+}} xmm5 = xmm2[0],xmm5[1,2],xmm2[3,4],xmm5[5,6,7]
-; SSE-NEXT: movdqa %xmm5, %xmm2
-; SSE-NEXT: psrlw $1, %xmm2
-; SSE-NEXT: pblendw {{.*#+}} xmm2 = xmm5[0,1],xmm2[2],xmm5[3],xmm2[4,5],xmm5[6],xmm2[7]
+; SSE-NEXT: pmulhuw %xmm5, %xmm2
; SSE-NEXT: paddw %xmm3, %xmm2
; SSE-NEXT: movdqa %xmm2, %xmm5
; SSE-NEXT: psraw $4, %xmm5
;
; AVX1-LABEL: combine_vec_sdiv_by_pow2b_v32i16:
; AVX1: # %bb.0:
-; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT: vpsraw $15, %xmm2, %xmm3
-; AVX1-NEXT: vpsrlw $8, %xmm3, %xmm4
-; AVX1-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0],xmm4[1,2,3,4,5,6,7]
-; AVX1-NEXT: vpsrlw $4, %xmm3, %xmm4
-; AVX1-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0],xmm4[1,2,3,4],xmm3[5,6],xmm4[7]
-; AVX1-NEXT: vpsrlw $2, %xmm3, %xmm4
-; AVX1-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0],xmm4[1,2],xmm3[3,4],xmm4[5,6,7]
-; AVX1-NEXT: vpsrlw $1, %xmm3, %xmm4
-; AVX1-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0,1],xmm4[2],xmm3[3],xmm4[4,5],xmm3[6],xmm4[7]
-; AVX1-NEXT: vpaddw %xmm3, %xmm2, %xmm2
-; AVX1-NEXT: vpsraw $4, %xmm2, %xmm3
-; AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0,1,2],xmm3[3],xmm2[4],xmm3[5,6],xmm2[7]
-; AVX1-NEXT: vpsraw $2, %xmm2, %xmm3
-; AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0],xmm3[1],xmm2[2,3],xmm3[4],xmm2[5],xmm3[6],xmm2[7]
-; AVX1-NEXT: vpsraw $1, %xmm2, %xmm3
-; AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0,1],xmm3[2],xmm2[3],xmm3[4,5],xmm2[6],xmm3[7]
-; AVX1-NEXT: vpsraw $15, %xmm0, %xmm3
-; AVX1-NEXT: vpsrlw $8, %xmm3, %xmm4
-; AVX1-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0],xmm4[1,2,3,4,5,6,7]
-; AVX1-NEXT: vpsrlw $4, %xmm3, %xmm4
-; AVX1-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0],xmm4[1,2,3,4],xmm3[5,6],xmm4[7]
-; AVX1-NEXT: vpsrlw $2, %xmm3, %xmm4
-; AVX1-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0],xmm4[1,2],xmm3[3,4],xmm4[5,6,7]
-; AVX1-NEXT: vpsrlw $1, %xmm3, %xmm4
-; AVX1-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0,1],xmm4[2],xmm3[3],xmm4[4,5],xmm3[6],xmm4[7]
-; AVX1-NEXT: vpaddw %xmm3, %xmm0, %xmm3
-; AVX1-NEXT: vpsraw $4, %xmm3, %xmm4
-; AVX1-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0,1,2],xmm4[3],xmm3[4],xmm4[5,6],xmm3[7]
-; AVX1-NEXT: vpsraw $2, %xmm3, %xmm4
-; AVX1-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0],xmm4[1],xmm3[2,3],xmm4[4],xmm3[5],xmm4[6],xmm3[7]
-; AVX1-NEXT: vpsraw $1, %xmm3, %xmm4
-; AVX1-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0,1],xmm4[2],xmm3[3],xmm4[4,5],xmm3[6],xmm4[7]
-; AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm3, %ymm3
-; AVX1-NEXT: vmovaps {{.*#+}} ymm2 = [0,65535,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,65535]
-; AVX1-NEXT: vandps %ymm2, %ymm3, %ymm3
-; AVX1-NEXT: vandnps %ymm0, %ymm2, %ymm0
-; AVX1-NEXT: vorps %ymm0, %ymm3, %ymm0
-; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm3
+; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3
; AVX1-NEXT: vpsraw $15, %xmm3, %xmm4
-; AVX1-NEXT: vpsrlw $8, %xmm4, %xmm5
-; AVX1-NEXT: vpblendw {{.*#+}} xmm4 = xmm4[0],xmm5[1,2,3,4,5,6,7]
-; AVX1-NEXT: vpsrlw $4, %xmm4, %xmm5
-; AVX1-NEXT: vpblendw {{.*#+}} xmm4 = xmm4[0],xmm5[1,2,3,4],xmm4[5,6],xmm5[7]
-; AVX1-NEXT: vpsrlw $2, %xmm4, %xmm5
-; AVX1-NEXT: vpblendw {{.*#+}} xmm4 = xmm4[0],xmm5[1,2],xmm4[3,4],xmm5[5,6,7]
-; AVX1-NEXT: vpsrlw $1, %xmm4, %xmm5
-; AVX1-NEXT: vpblendw {{.*#+}} xmm4 = xmm4[0,1],xmm5[2],xmm4[3],xmm5[4,5],xmm4[6],xmm5[7]
+; AVX1-NEXT: vmovdqa {{.*#+}} xmm2 = [1,4,2,16,8,32,64,2]
+; AVX1-NEXT: vpmulhuw %xmm2, %xmm4, %xmm4
; AVX1-NEXT: vpaddw %xmm4, %xmm3, %xmm3
; AVX1-NEXT: vpsraw $4, %xmm3, %xmm4
; AVX1-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0,1,2],xmm4[3],xmm3[4],xmm4[5,6],xmm3[7]
; AVX1-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0],xmm4[1],xmm3[2,3],xmm4[4],xmm3[5],xmm4[6],xmm3[7]
; AVX1-NEXT: vpsraw $1, %xmm3, %xmm4
; AVX1-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0,1],xmm4[2],xmm3[3],xmm4[4,5],xmm3[6],xmm4[7]
-; AVX1-NEXT: vpsraw $15, %xmm1, %xmm4
-; AVX1-NEXT: vpsrlw $8, %xmm4, %xmm5
-; AVX1-NEXT: vpblendw {{.*#+}} xmm4 = xmm4[0],xmm5[1,2,3,4,5,6,7]
-; AVX1-NEXT: vpsrlw $4, %xmm4, %xmm5
-; AVX1-NEXT: vpblendw {{.*#+}} xmm4 = xmm4[0],xmm5[1,2,3,4],xmm4[5,6],xmm5[7]
-; AVX1-NEXT: vpsrlw $2, %xmm4, %xmm5
-; AVX1-NEXT: vpblendw {{.*#+}} xmm4 = xmm4[0],xmm5[1,2],xmm4[3,4],xmm5[5,6,7]
-; AVX1-NEXT: vpsrlw $1, %xmm4, %xmm5
-; AVX1-NEXT: vpblendw {{.*#+}} xmm4 = xmm4[0,1],xmm5[2],xmm4[3],xmm5[4,5],xmm4[6],xmm5[7]
-; AVX1-NEXT: vpaddw %xmm4, %xmm1, %xmm4
+; AVX1-NEXT: vpsraw $15, %xmm0, %xmm4
+; AVX1-NEXT: vpmulhuw %xmm2, %xmm4, %xmm4
+; AVX1-NEXT: vpaddw %xmm4, %xmm0, %xmm4
; AVX1-NEXT: vpsraw $4, %xmm4, %xmm5
; AVX1-NEXT: vpblendw {{.*#+}} xmm4 = xmm4[0,1,2],xmm5[3],xmm4[4],xmm5[5,6],xmm4[7]
; AVX1-NEXT: vpsraw $2, %xmm4, %xmm5
; AVX1-NEXT: vpsraw $1, %xmm4, %xmm5
; AVX1-NEXT: vpblendw {{.*#+}} xmm4 = xmm4[0,1],xmm5[2],xmm4[3],xmm5[4,5],xmm4[6],xmm5[7]
; AVX1-NEXT: vinsertf128 $1, %xmm3, %ymm4, %ymm3
-; AVX1-NEXT: vandps %ymm2, %ymm3, %ymm3
-; AVX1-NEXT: vandnps %ymm1, %ymm2, %ymm1
-; AVX1-NEXT: vorps %ymm1, %ymm3, %ymm1
+; AVX1-NEXT: vmovaps {{.*#+}} ymm4 = [0,65535,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,65535]
+; AVX1-NEXT: vandps %ymm4, %ymm3, %ymm3
+; AVX1-NEXT: vandnps %ymm0, %ymm4, %ymm0
+; AVX1-NEXT: vorps %ymm0, %ymm3, %ymm0
+; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm3
+; AVX1-NEXT: vpsraw $15, %xmm3, %xmm5
+; AVX1-NEXT: vpmulhuw %xmm2, %xmm5, %xmm5
+; AVX1-NEXT: vpaddw %xmm5, %xmm3, %xmm3
+; AVX1-NEXT: vpsraw $4, %xmm3, %xmm5
+; AVX1-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0,1,2],xmm5[3],xmm3[4],xmm5[5,6],xmm3[7]
+; AVX1-NEXT: vpsraw $2, %xmm3, %xmm5
+; AVX1-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0],xmm5[1],xmm3[2,3],xmm5[4],xmm3[5],xmm5[6],xmm3[7]
+; AVX1-NEXT: vpsraw $1, %xmm3, %xmm5
+; AVX1-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0,1],xmm5[2],xmm3[3],xmm5[4,5],xmm3[6],xmm5[7]
+; AVX1-NEXT: vpsraw $15, %xmm1, %xmm5
+; AVX1-NEXT: vpmulhuw %xmm2, %xmm5, %xmm2
+; AVX1-NEXT: vpaddw %xmm2, %xmm1, %xmm2
+; AVX1-NEXT: vpsraw $4, %xmm2, %xmm5
+; AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0,1,2],xmm5[3],xmm2[4],xmm5[5,6],xmm2[7]
+; AVX1-NEXT: vpsraw $2, %xmm2, %xmm5
+; AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0],xmm5[1],xmm2[2,3],xmm5[4],xmm2[5],xmm5[6],xmm2[7]
+; AVX1-NEXT: vpsraw $1, %xmm2, %xmm5
+; AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0,1],xmm5[2],xmm2[3],xmm5[4,5],xmm2[6],xmm5[7]
+; AVX1-NEXT: vinsertf128 $1, %xmm3, %ymm2, %ymm2
+; AVX1-NEXT: vandps %ymm4, %ymm2, %ymm2
+; AVX1-NEXT: vandnps %ymm1, %ymm4, %ymm1
+; AVX1-NEXT: vorps %ymm1, %ymm2, %ymm1
; AVX1-NEXT: retq
;
; AVX2-LABEL: combine_vec_sdiv_by_pow2b_v32i16:
; AVX2: # %bb.0:
; AVX2-NEXT: vpxor %xmm2, %xmm2, %xmm2
-; AVX2-NEXT: vmovdqa {{.*#+}} ymm3 = [16,14,15,12,13,11,10,15,16,14,15,12,13,11,10,15]
+; AVX2-NEXT: vmovdqa {{.*#+}} ymm3 = [0,2,1,4,3,5,6,1,0,2,1,4,3,5,6,1]
; AVX2-NEXT: vpunpckhwd {{.*#+}} ymm4 = ymm3[4],ymm2[4],ymm3[5],ymm2[5],ymm3[6],ymm2[6],ymm3[7],ymm2[7],ymm3[12],ymm2[12],ymm3[13],ymm2[13],ymm3[14],ymm2[14],ymm3[15],ymm2[15]
; AVX2-NEXT: vpsraw $15, %ymm0, %ymm5
-; AVX2-NEXT: vpunpckhwd {{.*#+}} ymm6 = ymm2[4],ymm5[4],ymm2[5],ymm5[5],ymm2[6],ymm5[6],ymm2[7],ymm5[7],ymm2[12],ymm5[12],ymm2[13],ymm5[13],ymm2[14],ymm5[14],ymm2[15],ymm5[15]
-; AVX2-NEXT: vpsrlvd %ymm4, %ymm6, %ymm6
-; AVX2-NEXT: vpsrld $16, %ymm6, %ymm6
-; AVX2-NEXT: vpunpcklwd {{.*#+}} ymm3 = ymm3[0],ymm2[0],ymm3[1],ymm2[1],ymm3[2],ymm2[2],ymm3[3],ymm2[3],ymm3[8],ymm2[8],ymm3[9],ymm2[9],ymm3[10],ymm2[10],ymm3[11],ymm2[11]
-; AVX2-NEXT: vpunpcklwd {{.*#+}} ymm5 = ymm2[0],ymm5[0],ymm2[1],ymm5[1],ymm2[2],ymm5[2],ymm2[3],ymm5[3],ymm2[8],ymm5[8],ymm2[9],ymm5[9],ymm2[10],ymm5[10],ymm2[11],ymm5[11]
-; AVX2-NEXT: vpsrlvd %ymm3, %ymm5, %ymm5
-; AVX2-NEXT: vpsrld $16, %ymm5, %ymm5
-; AVX2-NEXT: vpackusdw %ymm6, %ymm5, %ymm5
+; AVX2-NEXT: vbroadcasti128 {{.*#+}} ymm6 = [1,4,2,16,8,32,64,2,1,4,2,16,8,32,64,2]
+; AVX2-NEXT: # ymm6 = mem[0,1,0,1]
+; AVX2-NEXT: vpmulhuw %ymm6, %ymm5, %ymm5
; AVX2-NEXT: vpaddw %ymm5, %ymm0, %ymm5
-; AVX2-NEXT: vpunpckhwd {{.*#+}} ymm6 = ymm2[4],ymm5[4],ymm2[5],ymm5[5],ymm2[6],ymm5[6],ymm2[7],ymm5[7],ymm2[12],ymm5[12],ymm2[13],ymm5[13],ymm2[14],ymm5[14],ymm2[15],ymm5[15]
-; AVX2-NEXT: vmovdqa {{.*#+}} ymm7 = [0,2,1,4,3,5,6,1,0,2,1,4,3,5,6,1]
-; AVX2-NEXT: vpunpckhwd {{.*#+}} ymm8 = ymm7[4],ymm2[4],ymm7[5],ymm2[5],ymm7[6],ymm2[6],ymm7[7],ymm2[7],ymm7[12],ymm2[12],ymm7[13],ymm2[13],ymm7[14],ymm2[14],ymm7[15],ymm2[15]
-; AVX2-NEXT: vpsravd %ymm8, %ymm6, %ymm6
-; AVX2-NEXT: vpsrld $16, %ymm6, %ymm6
+; AVX2-NEXT: vpunpckhwd {{.*#+}} ymm7 = ymm2[4],ymm5[4],ymm2[5],ymm5[5],ymm2[6],ymm5[6],ymm2[7],ymm5[7],ymm2[12],ymm5[12],ymm2[13],ymm5[13],ymm2[14],ymm5[14],ymm2[15],ymm5[15]
+; AVX2-NEXT: vpsravd %ymm4, %ymm7, %ymm7
+; AVX2-NEXT: vpsrld $16, %ymm7, %ymm7
+; AVX2-NEXT: vpunpcklwd {{.*#+}} ymm3 = ymm3[0],ymm2[0],ymm3[1],ymm2[1],ymm3[2],ymm2[2],ymm3[3],ymm2[3],ymm3[8],ymm2[8],ymm3[9],ymm2[9],ymm3[10],ymm2[10],ymm3[11],ymm2[11]
; AVX2-NEXT: vpunpcklwd {{.*#+}} ymm5 = ymm2[0],ymm5[0],ymm2[1],ymm5[1],ymm2[2],ymm5[2],ymm2[3],ymm5[3],ymm2[8],ymm5[8],ymm2[9],ymm5[9],ymm2[10],ymm5[10],ymm2[11],ymm5[11]
-; AVX2-NEXT: vpunpcklwd {{.*#+}} ymm7 = ymm7[0],ymm2[0],ymm7[1],ymm2[1],ymm7[2],ymm2[2],ymm7[3],ymm2[3],ymm7[8],ymm2[8],ymm7[9],ymm2[9],ymm7[10],ymm2[10],ymm7[11],ymm2[11]
-; AVX2-NEXT: vpsravd %ymm7, %ymm5, %ymm5
+; AVX2-NEXT: vpsravd %ymm3, %ymm5, %ymm5
; AVX2-NEXT: vpsrld $16, %ymm5, %ymm5
-; AVX2-NEXT: vpackusdw %ymm6, %ymm5, %ymm5
+; AVX2-NEXT: vpackusdw %ymm7, %ymm5, %ymm5
; AVX2-NEXT: vpblendw {{.*#+}} ymm0 = ymm0[0],ymm5[1,2,3,4,5,6,7],ymm0[8],ymm5[9,10,11,12,13,14,15]
; AVX2-NEXT: vpsraw $15, %ymm1, %ymm5
+; AVX2-NEXT: vpmulhuw %ymm6, %ymm5, %ymm5
+; AVX2-NEXT: vpaddw %ymm5, %ymm1, %ymm5
; AVX2-NEXT: vpunpckhwd {{.*#+}} ymm6 = ymm2[4],ymm5[4],ymm2[5],ymm5[5],ymm2[6],ymm5[6],ymm2[7],ymm5[7],ymm2[12],ymm5[12],ymm2[13],ymm5[13],ymm2[14],ymm5[14],ymm2[15],ymm5[15]
-; AVX2-NEXT: vpsrlvd %ymm4, %ymm6, %ymm4
-; AVX2-NEXT: vpsrld $16, %ymm4, %ymm4
-; AVX2-NEXT: vpunpcklwd {{.*#+}} ymm5 = ymm2[0],ymm5[0],ymm2[1],ymm5[1],ymm2[2],ymm5[2],ymm2[3],ymm5[3],ymm2[8],ymm5[8],ymm2[9],ymm5[9],ymm2[10],ymm5[10],ymm2[11],ymm5[11]
-; AVX2-NEXT: vpsrlvd %ymm3, %ymm5, %ymm3
-; AVX2-NEXT: vpsrld $16, %ymm3, %ymm3
-; AVX2-NEXT: vpackusdw %ymm4, %ymm3, %ymm3
-; AVX2-NEXT: vpaddw %ymm3, %ymm1, %ymm3
-; AVX2-NEXT: vpunpckhwd {{.*#+}} ymm4 = ymm2[4],ymm3[4],ymm2[5],ymm3[5],ymm2[6],ymm3[6],ymm2[7],ymm3[7],ymm2[12],ymm3[12],ymm2[13],ymm3[13],ymm2[14],ymm3[14],ymm2[15],ymm3[15]
-; AVX2-NEXT: vpsravd %ymm8, %ymm4, %ymm4
+; AVX2-NEXT: vpsravd %ymm4, %ymm6, %ymm4
; AVX2-NEXT: vpsrld $16, %ymm4, %ymm4
-; AVX2-NEXT: vpunpcklwd {{.*#+}} ymm2 = ymm2[0],ymm3[0],ymm2[1],ymm3[1],ymm2[2],ymm3[2],ymm2[3],ymm3[3],ymm2[8],ymm3[8],ymm2[9],ymm3[9],ymm2[10],ymm3[10],ymm2[11],ymm3[11]
-; AVX2-NEXT: vpsravd %ymm7, %ymm2, %ymm2
+; AVX2-NEXT: vpunpcklwd {{.*#+}} ymm2 = ymm2[0],ymm5[0],ymm2[1],ymm5[1],ymm2[2],ymm5[2],ymm2[3],ymm5[3],ymm2[8],ymm5[8],ymm2[9],ymm5[9],ymm2[10],ymm5[10],ymm2[11],ymm5[11]
+; AVX2-NEXT: vpsravd %ymm3, %ymm2, %ymm2
; AVX2-NEXT: vpsrld $16, %ymm2, %ymm2
; AVX2-NEXT: vpackusdw %ymm4, %ymm2, %ymm2
; AVX2-NEXT: vpblendw {{.*#+}} ymm1 = ymm1[0],ymm2[1,2,3,4,5,6,7],ymm1[8],ymm2[9,10,11,12,13,14,15]
; AVX512F-LABEL: combine_vec_sdiv_by_pow2b_v32i16:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vpsraw $15, %ymm0, %ymm2
-; AVX512F-NEXT: vpmovzxwd {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
-; AVX512F-NEXT: vbroadcasti64x4 {{.*#+}} zmm3 = [16,14,15,12,13,11,10,15,16,14,15,12,13,11,10,15]
-; AVX512F-NEXT: # zmm3 = mem[0,1,2,3,0,1,2,3]
-; AVX512F-NEXT: vpsrlvd %zmm3, %zmm2, %zmm2
-; AVX512F-NEXT: vpmovdw %zmm2, %ymm2
+; AVX512F-NEXT: vbroadcasti128 {{.*#+}} ymm3 = [1,4,2,16,8,32,64,2,1,4,2,16,8,32,64,2]
+; AVX512F-NEXT: # ymm3 = mem[0,1,0,1]
+; AVX512F-NEXT: vpmulhuw %ymm3, %ymm2, %ymm2
; AVX512F-NEXT: vpaddw %ymm2, %ymm0, %ymm2
; AVX512F-NEXT: vpmovsxwd %ymm2, %zmm2
; AVX512F-NEXT: vbroadcasti64x4 {{.*#+}} zmm4 = [0,2,1,4,3,5,6,1,0,2,1,4,3,5,6,1]
; AVX512F-NEXT: vpmovdw %zmm2, %ymm2
; AVX512F-NEXT: vpblendw {{.*#+}} ymm0 = ymm0[0],ymm2[1,2,3,4,5,6,7],ymm0[8],ymm2[9,10,11,12,13,14,15]
; AVX512F-NEXT: vpsraw $15, %ymm1, %ymm2
-; AVX512F-NEXT: vpmovzxwd {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
-; AVX512F-NEXT: vpsrlvd %zmm3, %zmm2, %zmm2
-; AVX512F-NEXT: vpmovdw %zmm2, %ymm2
+; AVX512F-NEXT: vpmulhuw %ymm3, %ymm2, %ymm2
; AVX512F-NEXT: vpaddw %ymm2, %ymm1, %ymm2
; AVX512F-NEXT: vpmovsxwd %ymm2, %zmm2
; AVX512F-NEXT: vpsravd %zmm4, %zmm2, %zmm2
; SSE: # %bb.0:
; SSE-NEXT: movdqa %xmm0, %xmm1
; SSE-NEXT: pxor %xmm2, %xmm2
-; SSE-NEXT: pxor %xmm3, %xmm3
-; SSE-NEXT: pcmpgtb %xmm0, %xmm3
-; SSE-NEXT: movdqa %xmm3, %xmm4
-; SSE-NEXT: psrlw $4, %xmm4
-; SSE-NEXT: pand {{.*}}(%rip), %xmm4
-; SSE-NEXT: movdqa {{.*#+}} xmm0 = [256,224,256,224,57600,57568,8416,8416]
-; SSE-NEXT: pblendvb %xmm0, %xmm4, %xmm3
-; SSE-NEXT: movdqa %xmm3, %xmm4
-; SSE-NEXT: psrlw $2, %xmm4
-; SSE-NEXT: pand {{.*}}(%rip), %xmm4
-; SSE-NEXT: paddb %xmm0, %xmm0
-; SSE-NEXT: pblendvb %xmm0, %xmm4, %xmm3
-; SSE-NEXT: movdqa %xmm3, %xmm4
-; SSE-NEXT: psrlw $1, %xmm4
-; SSE-NEXT: pand {{.*}}(%rip), %xmm4
-; SSE-NEXT: paddb %xmm0, %xmm0
-; SSE-NEXT: pblendvb %xmm0, %xmm4, %xmm3
-; SSE-NEXT: paddb %xmm1, %xmm3
-; SSE-NEXT: punpckhbw {{.*#+}} xmm4 = xmm4[8],xmm3[8],xmm4[9],xmm3[9],xmm4[10],xmm3[10],xmm4[11],xmm3[11],xmm4[12],xmm3[12],xmm4[13],xmm3[13],xmm4[14],xmm3[14],xmm4[15],xmm3[15]
-; SSE-NEXT: movdqa %xmm4, %xmm5
-; SSE-NEXT: psraw $4, %xmm5
-; SSE-NEXT: movdqa {{.*#+}} xmm6 = [0,32,0,32,8192,8224,57376,57376]
-; SSE-NEXT: punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm6[8],xmm0[9],xmm6[9],xmm0[10],xmm6[10],xmm0[11],xmm6[11],xmm0[12],xmm6[12],xmm0[13],xmm6[13],xmm0[14],xmm6[14],xmm0[15],xmm6[15]
-; SSE-NEXT: pblendvb %xmm0, %xmm5, %xmm4
-; SSE-NEXT: movdqa %xmm4, %xmm5
-; SSE-NEXT: psraw $2, %xmm5
-; SSE-NEXT: paddw %xmm0, %xmm0
-; SSE-NEXT: pblendvb %xmm0, %xmm5, %xmm4
-; SSE-NEXT: movdqa %xmm4, %xmm5
-; SSE-NEXT: psraw $1, %xmm5
-; SSE-NEXT: paddw %xmm0, %xmm0
-; SSE-NEXT: pblendvb %xmm0, %xmm5, %xmm4
+; SSE-NEXT: pxor %xmm0, %xmm0
+; SSE-NEXT: pcmpgtb %xmm1, %xmm0
+; SSE-NEXT: pmovzxbw {{.*#+}} xmm4 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; SSE-NEXT: pmullw {{.*}}(%rip), %xmm4
; SSE-NEXT: psrlw $8, %xmm4
-; SSE-NEXT: punpcklbw {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
+; SSE-NEXT: pmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; SSE-NEXT: pmullw {{.*}}(%rip), %xmm0
+; SSE-NEXT: psrlw $8, %xmm0
+; SSE-NEXT: packuswb %xmm0, %xmm4
+; SSE-NEXT: paddb %xmm1, %xmm4
+; SSE-NEXT: punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm4[8],xmm3[9],xmm4[9],xmm3[10],xmm4[10],xmm3[11],xmm4[11],xmm3[12],xmm4[12],xmm3[13],xmm4[13],xmm3[14],xmm4[14],xmm3[15],xmm4[15]
; SSE-NEXT: movdqa %xmm3, %xmm5
; SSE-NEXT: psraw $4, %xmm5
-; SSE-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm6[0],xmm0[1],xmm6[1],xmm0[2],xmm6[2],xmm0[3],xmm6[3],xmm0[4],xmm6[4],xmm0[5],xmm6[5],xmm0[6],xmm6[6],xmm0[7],xmm6[7]
+; SSE-NEXT: movdqa {{.*#+}} xmm6 = [0,32,0,32,8192,8224,57376,57376]
+; SSE-NEXT: punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm6[8],xmm0[9],xmm6[9],xmm0[10],xmm6[10],xmm0[11],xmm6[11],xmm0[12],xmm6[12],xmm0[13],xmm6[13],xmm0[14],xmm6[14],xmm0[15],xmm6[15]
; SSE-NEXT: pblendvb %xmm0, %xmm5, %xmm3
; SSE-NEXT: movdqa %xmm3, %xmm5
; SSE-NEXT: psraw $2, %xmm5
; SSE-NEXT: paddw %xmm0, %xmm0
; SSE-NEXT: pblendvb %xmm0, %xmm5, %xmm3
; SSE-NEXT: psrlw $8, %xmm3
-; SSE-NEXT: packuswb %xmm4, %xmm3
+; SSE-NEXT: punpcklbw {{.*#+}} xmm4 = xmm4[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE-NEXT: movdqa %xmm4, %xmm5
+; SSE-NEXT: psraw $4, %xmm5
+; SSE-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm6[0],xmm0[1],xmm6[1],xmm0[2],xmm6[2],xmm0[3],xmm6[3],xmm0[4],xmm6[4],xmm0[5],xmm6[5],xmm0[6],xmm6[6],xmm0[7],xmm6[7]
+; SSE-NEXT: pblendvb %xmm0, %xmm5, %xmm4
+; SSE-NEXT: movdqa %xmm4, %xmm5
+; SSE-NEXT: psraw $2, %xmm5
+; SSE-NEXT: paddw %xmm0, %xmm0
+; SSE-NEXT: pblendvb %xmm0, %xmm5, %xmm4
+; SSE-NEXT: movdqa %xmm4, %xmm5
+; SSE-NEXT: psraw $1, %xmm5
+; SSE-NEXT: paddw %xmm0, %xmm0
+; SSE-NEXT: pblendvb %xmm0, %xmm5, %xmm4
+; SSE-NEXT: psrlw $8, %xmm4
+; SSE-NEXT: packuswb %xmm3, %xmm4
; SSE-NEXT: movaps {{.*#+}} xmm0 = [0,0,255,0,0,0,255,0,0,255,255,255,255,255,255,255]
-; SSE-NEXT: pblendvb %xmm0, %xmm3, %xmm1
+; SSE-NEXT: pblendvb %xmm0, %xmm4, %xmm1
; SSE-NEXT: psubb %xmm1, %xmm2
; SSE-NEXT: movaps {{.*#+}} xmm0 = [255,255,0,255,255,255,0,255,255,0,0,0,0,255,0,255]
; SSE-NEXT: pblendvb %xmm0, %xmm2, %xmm1
; AVX1: # %bb.0:
; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX1-NEXT: vpcmpgtb %xmm0, %xmm1, %xmm2
-; AVX1-NEXT: vpsrlw $4, %xmm2, %xmm3
-; AVX1-NEXT: vpand {{.*}}(%rip), %xmm3, %xmm3
-; AVX1-NEXT: vmovdqa {{.*#+}} xmm4 = [256,224,256,224,57600,57568,8416,8416]
-; AVX1-NEXT: vpblendvb %xmm4, %xmm3, %xmm2, %xmm2
-; AVX1-NEXT: vpsrlw $2, %xmm2, %xmm3
-; AVX1-NEXT: vpand {{.*}}(%rip), %xmm3, %xmm3
-; AVX1-NEXT: vpaddb %xmm4, %xmm4, %xmm4
-; AVX1-NEXT: vpblendvb %xmm4, %xmm3, %xmm2, %xmm2
-; AVX1-NEXT: vpsrlw $1, %xmm2, %xmm3
-; AVX1-NEXT: vpand {{.*}}(%rip), %xmm3, %xmm3
-; AVX1-NEXT: vpaddb %xmm4, %xmm4, %xmm4
-; AVX1-NEXT: vpblendvb %xmm4, %xmm3, %xmm2, %xmm2
+; AVX1-NEXT: vpmovzxbw {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
+; AVX1-NEXT: vpmullw {{.*}}(%rip), %xmm3, %xmm3
+; AVX1-NEXT: vpsrlw $8, %xmm3, %xmm3
+; AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
+; AVX1-NEXT: vpmovzxbw {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
+; AVX1-NEXT: vpmullw {{.*}}(%rip), %xmm2, %xmm2
+; AVX1-NEXT: vpsrlw $8, %xmm2, %xmm2
+; AVX1-NEXT: vpackuswb %xmm2, %xmm3, %xmm2
; AVX1-NEXT: vpaddb %xmm2, %xmm0, %xmm2
; AVX1-NEXT: vpunpckhbw {{.*#+}} xmm3 = xmm0[8],xmm2[8],xmm0[9],xmm2[9],xmm0[10],xmm2[10],xmm0[11],xmm2[11],xmm0[12],xmm2[12],xmm0[13],xmm2[13],xmm0[14],xmm2[14],xmm0[15],xmm2[15]
; AVX1-NEXT: vpsraw $4, %xmm3, %xmm4
; AVX2: # %bb.0:
; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX2-NEXT: vpcmpgtb %xmm0, %xmm1, %xmm2
-; AVX2-NEXT: vpsrlw $4, %xmm2, %xmm3
-; AVX2-NEXT: vpand {{.*}}(%rip), %xmm3, %xmm3
-; AVX2-NEXT: vmovdqa {{.*#+}} xmm4 = [256,224,256,224,57600,57568,8416,8416]
-; AVX2-NEXT: vpblendvb %xmm4, %xmm3, %xmm2, %xmm2
-; AVX2-NEXT: vpsrlw $2, %xmm2, %xmm3
-; AVX2-NEXT: vpand {{.*}}(%rip), %xmm3, %xmm3
-; AVX2-NEXT: vpaddb %xmm4, %xmm4, %xmm4
-; AVX2-NEXT: vpblendvb %xmm4, %xmm3, %xmm2, %xmm2
-; AVX2-NEXT: vpsrlw $1, %xmm2, %xmm3
-; AVX2-NEXT: vpand {{.*}}(%rip), %xmm3, %xmm3
-; AVX2-NEXT: vpaddb %xmm4, %xmm4, %xmm4
-; AVX2-NEXT: vpblendvb %xmm4, %xmm3, %xmm2, %xmm2
+; AVX2-NEXT: vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
+; AVX2-NEXT: vpmullw {{.*}}(%rip), %ymm2, %ymm2
+; AVX2-NEXT: vpsrlw $8, %ymm2, %ymm2
+; AVX2-NEXT: vextracti128 $1, %ymm2, %xmm3
+; AVX2-NEXT: vpackuswb %xmm3, %xmm2, %xmm2
; AVX2-NEXT: vpaddb %xmm2, %xmm0, %xmm2
; AVX2-NEXT: vpunpckhbw {{.*#+}} xmm3 = xmm0[8],xmm2[8],xmm0[9],xmm2[9],xmm0[10],xmm2[10],xmm0[11],xmm2[11],xmm0[12],xmm2[12],xmm0[13],xmm2[13],xmm0[14],xmm2[14],xmm0[15],xmm2[15]
; AVX2-NEXT: vpsraw $4, %xmm3, %xmm4
; AVX2-NEXT: vpsubb %xmm0, %xmm1, %xmm1
; AVX2-NEXT: vmovdqa {{.*#+}} xmm2 = [255,255,0,255,255,255,0,255,255,0,0,0,0,255,0,255]
; AVX2-NEXT: vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
+; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
;
; AVX512F-LABEL: non_splat_minus_one_divisor_1: