diff options
Diffstat (limited to 'llvm/test/CodeGen/X86/vector-shift-lshr-sub128.ll')
| -rw-r--r-- | llvm/test/CodeGen/X86/vector-shift-lshr-sub128.ll | 2729 |
1 files changed, 1451 insertions, 1278 deletions
diff --git a/llvm/test/CodeGen/X86/vector-shift-lshr-sub128.ll b/llvm/test/CodeGen/X86/vector-shift-lshr-sub128.ll index f351bfb2196..546f723e68b 100644 --- a/llvm/test/CodeGen/X86/vector-shift-lshr-sub128.ll +++ b/llvm/test/CodeGen/X86/vector-shift-lshr-sub128.ll @@ -20,6 +20,103 @@ define <2 x i32> @var_shift_v2i32(<2 x i32> %a, <2 x i32> %b) nounwind { ; SSE2-LABEL: var_shift_v2i32: ; SSE2: # %bb.0: +; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [4294967295,0,4294967295,0] +; SSE2-NEXT: pand %xmm2, %xmm1 +; SSE2-NEXT: pand %xmm2, %xmm0 +; SSE2-NEXT: movdqa %xmm0, %xmm2 +; SSE2-NEXT: psrlq %xmm1, %xmm2 +; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1] +; SSE2-NEXT: psrlq %xmm1, %xmm0 +; SSE2-NEXT: movsd {{.*#+}} xmm0 = xmm2[0],xmm0[1] +; SSE2-NEXT: retq +; +; SSE41-LABEL: var_shift_v2i32: +; SSE41: # %bb.0: +; SSE41-NEXT: pxor %xmm2, %xmm2 +; SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7] +; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3],xmm0[4,5],xmm2[6,7] +; SSE41-NEXT: movdqa %xmm0, %xmm2 +; SSE41-NEXT: psrlq %xmm1, %xmm2 +; SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1] +; SSE41-NEXT: psrlq %xmm1, %xmm0 +; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm2[0,1,2,3],xmm0[4,5,6,7] +; SSE41-NEXT: retq +; +; AVX1-LABEL: var_shift_v2i32: +; AVX1: # %bb.0: +; AVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7] +; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3],xmm0[4,5],xmm2[6,7] +; AVX1-NEXT: vpsrlq %xmm1, %xmm0, %xmm2 +; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1] +; AVX1-NEXT: vpsrlq %xmm1, %xmm0, %xmm0 +; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm2[0,1,2,3],xmm0[4,5,6,7] +; AVX1-NEXT: retq +; +; AVX2-LABEL: var_shift_v2i32: +; AVX2: # %bb.0: +; AVX2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX2-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] +; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3] +; AVX2-NEXT: vpsrlvq %xmm1, %xmm0, %xmm0 +; AVX2-NEXT: retq +; +; XOPAVX1-LABEL: var_shift_v2i32: +; XOPAVX1: # %bb.0: +; XOPAVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; XOPAVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3],xmm0[4,5],xmm2[6,7] +; XOPAVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7] +; XOPAVX1-NEXT: vpsubq %xmm1, %xmm2, %xmm1 +; XOPAVX1-NEXT: vpshlq %xmm1, %xmm0, %xmm0 +; XOPAVX1-NEXT: retq +; +; XOPAVX2-LABEL: var_shift_v2i32: +; XOPAVX2: # %bb.0: +; XOPAVX2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; XOPAVX2-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] +; XOPAVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3] +; XOPAVX2-NEXT: vpsrlvq %xmm1, %xmm0, %xmm0 +; XOPAVX2-NEXT: retq +; +; AVX512-LABEL: var_shift_v2i32: +; AVX512: # %bb.0: +; AVX512-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] +; AVX512-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3] +; AVX512-NEXT: vpsrlvq %xmm1, %xmm0, %xmm0 +; AVX512-NEXT: retq +; +; AVX512VL-LABEL: var_shift_v2i32: +; AVX512VL: # %bb.0: +; AVX512VL-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512VL-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] +; AVX512VL-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3] +; AVX512VL-NEXT: vpsrlvq %xmm1, %xmm0, %xmm0 +; AVX512VL-NEXT: retq +; +; X32-SSE-LABEL: var_shift_v2i32: +; X32-SSE: # %bb.0: +; X32-SSE-NEXT: movdqa {{.*#+}} xmm2 = [4294967295,0,4294967295,0] +; X32-SSE-NEXT: pand %xmm2, %xmm1 +; X32-SSE-NEXT: pand %xmm2, %xmm0 +; X32-SSE-NEXT: movdqa %xmm0, %xmm2 +; X32-SSE-NEXT: psrlq %xmm1, %xmm2 +; X32-SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1] +; X32-SSE-NEXT: xorps %xmm3, %xmm3 +; X32-SSE-NEXT: movss {{.*#+}} xmm3 = xmm1[0],xmm3[1,2,3] +; X32-SSE-NEXT: psrlq %xmm3, %xmm0 +; X32-SSE-NEXT: movsd {{.*#+}} xmm0 = xmm2[0],xmm0[1] +; X32-SSE-NEXT: retl + %shift = lshr <2 x i32> %a, %b + ret <2 x i32> %shift +} + +define <4 x i16> @var_shift_v4i16(<4 x i16> %a, <4 x i16> %b) nounwind { +; SSE2-LABEL: var_shift_v4i16: +; SSE2: # %bb.0: +; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [65535,0,65535,0,65535,0,65535,0] +; SSE2-NEXT: pand %xmm2, %xmm0 +; SSE2-NEXT: pand %xmm2, %xmm1 ; SSE2-NEXT: pshuflw {{.*#+}} xmm2 = xmm1[2,3,3,3,4,5,6,7] ; SSE2-NEXT: movdqa %xmm0, %xmm3 ; SSE2-NEXT: psrld %xmm2, %xmm3 @@ -38,70 +135,93 @@ define <2 x i32> @var_shift_v2i32(<2 x i32> %a, <2 x i32> %b) nounwind { ; SSE2-NEXT: movaps %xmm2, %xmm0 ; SSE2-NEXT: retq ; -; SSE41-LABEL: var_shift_v2i32: +; SSE41-LABEL: var_shift_v4i16: ; SSE41: # %bb.0: -; SSE41-NEXT: pshuflw {{.*#+}} xmm2 = xmm1[2,3,3,3,4,5,6,7] -; SSE41-NEXT: movdqa %xmm0, %xmm3 -; SSE41-NEXT: psrld %xmm2, %xmm3 -; SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm1[2,3,0,1] -; SSE41-NEXT: pshuflw {{.*#+}} xmm4 = xmm2[2,3,3,3,4,5,6,7] +; SSE41-NEXT: pxor %xmm2, %xmm2 +; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3],xmm0[4],xmm2[5],xmm0[6],xmm2[7] +; SSE41-NEXT: movdqa %xmm1, %xmm3 +; SSE41-NEXT: pblendw {{.*#+}} xmm3 = xmm3[0],xmm2[1],xmm3[2],xmm2[3],xmm3[4],xmm2[5],xmm3[6],xmm2[7] +; SSE41-NEXT: pshuflw {{.*#+}} xmm4 = xmm3[2,3,3,3,4,5,6,7] ; SSE41-NEXT: movdqa %xmm0, %xmm5 ; SSE41-NEXT: psrld %xmm4, %xmm5 -; SSE41-NEXT: pblendw {{.*#+}} xmm5 = xmm3[0,1,2,3],xmm5[4,5,6,7] -; SSE41-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,1,1,1,4,5,6,7] -; SSE41-NEXT: movdqa %xmm0, %xmm3 -; SSE41-NEXT: psrld %xmm1, %xmm3 -; SSE41-NEXT: pshuflw {{.*#+}} xmm1 = xmm2[0,1,1,1,4,5,6,7] +; SSE41-NEXT: pshufd {{.*#+}} xmm3 = xmm3[2,3,0,1] +; SSE41-NEXT: pshuflw {{.*#+}} xmm4 = xmm3[2,3,3,3,4,5,6,7] +; SSE41-NEXT: movdqa %xmm0, %xmm6 +; SSE41-NEXT: psrld %xmm4, %xmm6 +; SSE41-NEXT: pblendw {{.*#+}} xmm6 = xmm5[0,1,2,3],xmm6[4,5,6,7] +; SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5],xmm1[6],xmm2[7] +; SSE41-NEXT: movdqa %xmm0, %xmm2 +; SSE41-NEXT: psrld %xmm1, %xmm2 +; SSE41-NEXT: pshuflw {{.*#+}} xmm1 = xmm3[0,1,1,1,4,5,6,7] ; SSE41-NEXT: psrld %xmm1, %xmm0 -; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm3[0,1,2,3],xmm0[4,5,6,7] -; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm5[2,3],xmm0[4,5],xmm5[6,7] +; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm2[0,1,2,3],xmm0[4,5,6,7] +; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm6[2,3],xmm0[4,5],xmm6[6,7] ; SSE41-NEXT: retq ; -; AVX1-LABEL: var_shift_v2i32: +; AVX1-LABEL: var_shift_v4i16: ; AVX1: # %bb.0: -; AVX1-NEXT: vpsrldq {{.*#+}} xmm2 = xmm1[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero -; AVX1-NEXT: vpsrld %xmm2, %xmm0, %xmm2 -; AVX1-NEXT: vpsrlq $32, %xmm1, %xmm3 -; AVX1-NEXT: vpsrld %xmm3, %xmm0, %xmm3 -; AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm3[0,1,2,3],xmm2[4,5,6,7] -; AVX1-NEXT: vpxor %xmm3, %xmm3, %xmm3 -; AVX1-NEXT: vpunpckhdq {{.*#+}} xmm3 = xmm1[2],xmm3[2],xmm1[3],xmm3[3] +; AVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3],xmm0[4],xmm2[5],xmm0[6],xmm2[7] +; AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7] +; AVX1-NEXT: vpsrldq {{.*#+}} xmm3 = xmm1[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero ; AVX1-NEXT: vpsrld %xmm3, %xmm0, %xmm3 +; AVX1-NEXT: vpsrlq $32, %xmm1, %xmm4 +; AVX1-NEXT: vpsrld %xmm4, %xmm0, %xmm4 +; AVX1-NEXT: vpblendw {{.*#+}} xmm3 = xmm4[0,1,2,3],xmm3[4,5,6,7] +; AVX1-NEXT: vpunpckhdq {{.*#+}} xmm2 = xmm1[2],xmm2[2],xmm1[3],xmm2[3] +; AVX1-NEXT: vpsrld %xmm2, %xmm0, %xmm2 ; AVX1-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero ; AVX1-NEXT: vpsrld %xmm1, %xmm0, %xmm0 -; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm3[4,5,6,7] -; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3],xmm0[4,5],xmm2[6,7] +; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm2[4,5,6,7] +; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm3[2,3],xmm0[4,5],xmm3[6,7] ; AVX1-NEXT: retq ; -; AVX2-LABEL: var_shift_v2i32: +; AVX2-LABEL: var_shift_v4i16: ; AVX2: # %bb.0: +; AVX2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX2-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7] +; AVX2-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3],xmm0[4],xmm2[5],xmm0[6],xmm2[7] ; AVX2-NEXT: vpsrlvd %xmm1, %xmm0, %xmm0 ; AVX2-NEXT: retq ; -; XOPAVX1-LABEL: var_shift_v2i32: +; XOPAVX1-LABEL: var_shift_v4i16: ; XOPAVX1: # %bb.0: ; XOPAVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; XOPAVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3],xmm0[4],xmm2[5],xmm0[6],xmm2[7] +; XOPAVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7] ; XOPAVX1-NEXT: vpsubd %xmm1, %xmm2, %xmm1 ; XOPAVX1-NEXT: vpshld %xmm1, %xmm0, %xmm0 ; XOPAVX1-NEXT: retq ; -; XOPAVX2-LABEL: var_shift_v2i32: +; XOPAVX2-LABEL: var_shift_v4i16: ; XOPAVX2: # %bb.0: +; XOPAVX2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; XOPAVX2-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7] +; XOPAVX2-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3],xmm0[4],xmm2[5],xmm0[6],xmm2[7] ; XOPAVX2-NEXT: vpsrlvd %xmm1, %xmm0, %xmm0 ; XOPAVX2-NEXT: retq ; -; AVX512-LABEL: var_shift_v2i32: +; AVX512-LABEL: var_shift_v4i16: ; AVX512: # %bb.0: +; AVX512-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7] +; AVX512-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3],xmm0[4],xmm2[5],xmm0[6],xmm2[7] ; AVX512-NEXT: vpsrlvd %xmm1, %xmm0, %xmm0 ; AVX512-NEXT: retq ; -; AVX512VL-LABEL: var_shift_v2i32: +; AVX512VL-LABEL: var_shift_v4i16: ; AVX512VL: # %bb.0: +; AVX512VL-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512VL-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7] +; AVX512VL-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3],xmm0[4],xmm2[5],xmm0[6],xmm2[7] ; AVX512VL-NEXT: vpsrlvd %xmm1, %xmm0, %xmm0 ; AVX512VL-NEXT: retq ; -; X32-SSE-LABEL: var_shift_v2i32: +; X32-SSE-LABEL: var_shift_v4i16: ; X32-SSE: # %bb.0: +; X32-SSE-NEXT: movdqa {{.*#+}} xmm2 = [65535,0,65535,0,65535,0,65535,0] +; X32-SSE-NEXT: pand %xmm2, %xmm0 +; X32-SSE-NEXT: pand %xmm2, %xmm1 ; X32-SSE-NEXT: pshuflw {{.*#+}} xmm2 = xmm1[2,3,3,3,4,5,6,7] ; X32-SSE-NEXT: movdqa %xmm0, %xmm3 ; X32-SSE-NEXT: psrld %xmm2, %xmm3 @@ -119,195 +239,111 @@ define <2 x i32> @var_shift_v2i32(<2 x i32> %a, <2 x i32> %b) nounwind { ; X32-SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,3],xmm0[0,3] ; X32-SSE-NEXT: movaps %xmm2, %xmm0 ; X32-SSE-NEXT: retl - %shift = lshr <2 x i32> %a, %b - ret <2 x i32> %shift + %shift = lshr <4 x i16> %a, %b + ret <4 x i16> %shift } -define <4 x i16> @var_shift_v4i16(<4 x i16> %a, <4 x i16> %b) nounwind { -; SSE2-LABEL: var_shift_v4i16: +define <2 x i16> @var_shift_v2i16(<2 x i16> %a, <2 x i16> %b) nounwind { +; SSE2-LABEL: var_shift_v2i16: ; SSE2: # %bb.0: -; SSE2-NEXT: psllw $12, %xmm1 -; SSE2-NEXT: movdqa %xmm1, %xmm2 -; SSE2-NEXT: psraw $15, %xmm2 -; SSE2-NEXT: movdqa %xmm2, %xmm3 -; SSE2-NEXT: pandn %xmm0, %xmm3 -; SSE2-NEXT: psrlw $8, %xmm0 -; SSE2-NEXT: pand %xmm2, %xmm0 -; SSE2-NEXT: por %xmm3, %xmm0 -; SSE2-NEXT: paddw %xmm1, %xmm1 -; SSE2-NEXT: movdqa %xmm1, %xmm2 -; SSE2-NEXT: psraw $15, %xmm2 -; SSE2-NEXT: movdqa %xmm2, %xmm3 -; SSE2-NEXT: pandn %xmm0, %xmm3 -; SSE2-NEXT: psrlw $4, %xmm0 -; SSE2-NEXT: pand %xmm2, %xmm0 -; SSE2-NEXT: por %xmm3, %xmm0 -; SSE2-NEXT: paddw %xmm1, %xmm1 -; SSE2-NEXT: movdqa %xmm1, %xmm2 -; SSE2-NEXT: psraw $15, %xmm2 -; SSE2-NEXT: movdqa %xmm2, %xmm3 -; SSE2-NEXT: pandn %xmm0, %xmm3 -; SSE2-NEXT: psrlw $2, %xmm0 +; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [65535,0,0,0,65535,0,0,0] +; SSE2-NEXT: pand %xmm2, %xmm1 ; SSE2-NEXT: pand %xmm2, %xmm0 -; SSE2-NEXT: por %xmm3, %xmm0 -; SSE2-NEXT: paddw %xmm1, %xmm1 -; SSE2-NEXT: psraw $15, %xmm1 -; SSE2-NEXT: movdqa %xmm1, %xmm2 -; SSE2-NEXT: pandn %xmm0, %xmm2 -; SSE2-NEXT: psrlw $1, %xmm0 -; SSE2-NEXT: pand %xmm1, %xmm0 -; SSE2-NEXT: por %xmm2, %xmm0 +; SSE2-NEXT: movdqa %xmm0, %xmm2 +; SSE2-NEXT: psrlq %xmm1, %xmm2 +; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1] +; SSE2-NEXT: psrlq %xmm1, %xmm0 +; SSE2-NEXT: movsd {{.*#+}} xmm0 = xmm2[0],xmm0[1] ; SSE2-NEXT: retq ; -; SSE41-LABEL: var_shift_v4i16: +; SSE41-LABEL: var_shift_v2i16: ; SSE41: # %bb.0: -; SSE41-NEXT: movdqa %xmm1, %xmm2 -; SSE41-NEXT: movdqa %xmm0, %xmm1 -; SSE41-NEXT: movdqa %xmm2, %xmm0 -; SSE41-NEXT: psllw $12, %xmm0 -; SSE41-NEXT: psllw $4, %xmm2 -; SSE41-NEXT: por %xmm0, %xmm2 -; SSE41-NEXT: movdqa %xmm2, %xmm3 -; SSE41-NEXT: paddw %xmm2, %xmm3 -; SSE41-NEXT: movdqa %xmm1, %xmm4 -; SSE41-NEXT: psrlw $8, %xmm4 -; SSE41-NEXT: movdqa %xmm2, %xmm0 -; SSE41-NEXT: pblendvb %xmm0, %xmm4, %xmm1 -; SSE41-NEXT: movdqa %xmm1, %xmm2 -; SSE41-NEXT: psrlw $4, %xmm2 -; SSE41-NEXT: movdqa %xmm3, %xmm0 -; SSE41-NEXT: pblendvb %xmm0, %xmm2, %xmm1 -; SSE41-NEXT: movdqa %xmm1, %xmm2 -; SSE41-NEXT: psrlw $2, %xmm2 -; SSE41-NEXT: paddw %xmm3, %xmm3 -; SSE41-NEXT: movdqa %xmm3, %xmm0 -; SSE41-NEXT: pblendvb %xmm0, %xmm2, %xmm1 -; SSE41-NEXT: movdqa %xmm1, %xmm2 -; SSE41-NEXT: psrlw $1, %xmm2 -; SSE41-NEXT: paddw %xmm3, %xmm3 -; SSE41-NEXT: movdqa %xmm3, %xmm0 -; SSE41-NEXT: pblendvb %xmm0, %xmm2, %xmm1 -; SSE41-NEXT: movdqa %xmm1, %xmm0 +; SSE41-NEXT: pxor %xmm2, %xmm2 +; SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7] +; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1,2,3],xmm0[4],xmm2[5,6,7] +; SSE41-NEXT: movdqa %xmm0, %xmm2 +; SSE41-NEXT: psrlq %xmm1, %xmm2 +; SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1] +; SSE41-NEXT: psrlq %xmm1, %xmm0 +; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm2[0,1,2,3],xmm0[4,5,6,7] ; SSE41-NEXT: retq ; -; AVX1-LABEL: var_shift_v4i16: +; AVX1-LABEL: var_shift_v2i16: ; AVX1: # %bb.0: -; AVX1-NEXT: vpsllw $12, %xmm1, %xmm2 -; AVX1-NEXT: vpsllw $4, %xmm1, %xmm1 -; AVX1-NEXT: vpor %xmm2, %xmm1, %xmm1 -; AVX1-NEXT: vpaddw %xmm1, %xmm1, %xmm2 -; AVX1-NEXT: vpsrlw $8, %xmm0, %xmm3 -; AVX1-NEXT: vpblendvb %xmm1, %xmm3, %xmm0, %xmm0 -; AVX1-NEXT: vpsrlw $4, %xmm0, %xmm1 -; AVX1-NEXT: vpblendvb %xmm2, %xmm1, %xmm0, %xmm0 -; AVX1-NEXT: vpsrlw $2, %xmm0, %xmm1 -; AVX1-NEXT: vpaddw %xmm2, %xmm2, %xmm2 -; AVX1-NEXT: vpblendvb %xmm2, %xmm1, %xmm0, %xmm0 -; AVX1-NEXT: vpsrlw $1, %xmm0, %xmm1 -; AVX1-NEXT: vpaddw %xmm2, %xmm2, %xmm2 -; AVX1-NEXT: vpblendvb %xmm2, %xmm1, %xmm0, %xmm0 +; AVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7] +; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1,2,3],xmm0[4],xmm2[5,6,7] +; AVX1-NEXT: vpsrlq %xmm1, %xmm0, %xmm2 +; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1] +; AVX1-NEXT: vpsrlq %xmm1, %xmm0, %xmm0 +; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm2[0,1,2,3],xmm0[4,5,6,7] ; AVX1-NEXT: retq ; -; AVX2-LABEL: var_shift_v4i16: +; AVX2-LABEL: var_shift_v2i16: ; AVX2: # %bb.0: -; 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: 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 -; AVX2-NEXT: vpsrlvd %ymm1, %ymm0, %ymm0 -; AVX2-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[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 {{.*#+}} ymm0 = ymm0[0,2,2,3] -; AVX2-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 -; AVX2-NEXT: vzeroupper +; AVX2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX2-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7] +; AVX2-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1,2,3],xmm0[4],xmm2[5,6,7] +; AVX2-NEXT: vpsrlvq %xmm1, %xmm0, %xmm0 ; AVX2-NEXT: retq ; -; XOP-LABEL: var_shift_v4i16: -; XOP: # %bb.0: -; XOP-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; XOP-NEXT: vpsubw %xmm1, %xmm2, %xmm1 -; XOP-NEXT: vpshlw %xmm1, %xmm0, %xmm0 -; XOP-NEXT: retq -; -; AVX512DQ-LABEL: var_shift_v4i16: -; AVX512DQ: # %bb.0: -; AVX512DQ-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 -; AVX512DQ-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 -; AVX512DQ-NEXT: vpsrlvd %ymm1, %ymm0, %ymm0 -; AVX512DQ-NEXT: vpmovdw %zmm0, %ymm0 -; AVX512DQ-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 -; AVX512DQ-NEXT: vzeroupper -; AVX512DQ-NEXT: retq +; XOPAVX1-LABEL: var_shift_v2i16: +; XOPAVX1: # %bb.0: +; XOPAVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; XOPAVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1,2,3],xmm0[4],xmm2[5,6,7] +; XOPAVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7] +; XOPAVX1-NEXT: vpsubq %xmm1, %xmm2, %xmm1 +; XOPAVX1-NEXT: vpshlq %xmm1, %xmm0, %xmm0 +; XOPAVX1-NEXT: retq ; -; AVX512BW-LABEL: var_shift_v4i16: -; AVX512BW: # %bb.0: -; AVX512BW-NEXT: # kill: def $xmm1 killed $xmm1 def $zmm1 -; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0 -; AVX512BW-NEXT: vpsrlvw %zmm1, %zmm0, %zmm0 -; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0 -; AVX512BW-NEXT: vzeroupper -; AVX512BW-NEXT: retq +; XOPAVX2-LABEL: var_shift_v2i16: +; XOPAVX2: # %bb.0: +; XOPAVX2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; XOPAVX2-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7] +; XOPAVX2-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1,2,3],xmm0[4],xmm2[5,6,7] +; XOPAVX2-NEXT: vpsrlvq %xmm1, %xmm0, %xmm0 +; XOPAVX2-NEXT: retq ; -; AVX512DQVL-LABEL: var_shift_v4i16: -; AVX512DQVL: # %bb.0: -; AVX512DQVL-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 -; AVX512DQVL-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 -; AVX512DQVL-NEXT: vpsrlvd %ymm1, %ymm0, %ymm0 -; AVX512DQVL-NEXT: vpmovdw %ymm0, %xmm0 -; AVX512DQVL-NEXT: vzeroupper -; AVX512DQVL-NEXT: retq +; AVX512-LABEL: var_shift_v2i16: +; AVX512: # %bb.0: +; AVX512-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7] +; AVX512-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1,2,3],xmm0[4],xmm2[5,6,7] +; AVX512-NEXT: vpsrlvq %xmm1, %xmm0, %xmm0 +; AVX512-NEXT: retq ; -; AVX512BWVL-LABEL: var_shift_v4i16: -; AVX512BWVL: # %bb.0: -; AVX512BWVL-NEXT: vpsrlvw %xmm1, %xmm0, %xmm0 -; AVX512BWVL-NEXT: retq +; AVX512VL-LABEL: var_shift_v2i16: +; AVX512VL: # %bb.0: +; AVX512VL-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512VL-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7] +; AVX512VL-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1,2,3],xmm0[4],xmm2[5,6,7] +; AVX512VL-NEXT: vpsrlvq %xmm1, %xmm0, %xmm0 +; AVX512VL-NEXT: retq ; -; X32-SSE-LABEL: var_shift_v4i16: +; X32-SSE-LABEL: var_shift_v2i16: ; X32-SSE: # %bb.0: -; X32-SSE-NEXT: psllw $12, %xmm1 -; X32-SSE-NEXT: movdqa %xmm1, %xmm2 -; X32-SSE-NEXT: psraw $15, %xmm2 -; X32-SSE-NEXT: movdqa %xmm2, %xmm3 -; X32-SSE-NEXT: pandn %xmm0, %xmm3 -; X32-SSE-NEXT: psrlw $8, %xmm0 -; X32-SSE-NEXT: pand %xmm2, %xmm0 -; X32-SSE-NEXT: por %xmm3, %xmm0 -; X32-SSE-NEXT: paddw %xmm1, %xmm1 -; X32-SSE-NEXT: movdqa %xmm1, %xmm2 -; X32-SSE-NEXT: psraw $15, %xmm2 -; X32-SSE-NEXT: movdqa %xmm2, %xmm3 -; X32-SSE-NEXT: pandn %xmm0, %xmm3 -; X32-SSE-NEXT: psrlw $4, %xmm0 -; X32-SSE-NEXT: pand %xmm2, %xmm0 -; X32-SSE-NEXT: por %xmm3, %xmm0 -; X32-SSE-NEXT: paddw %xmm1, %xmm1 -; X32-SSE-NEXT: movdqa %xmm1, %xmm2 -; X32-SSE-NEXT: psraw $15, %xmm2 -; X32-SSE-NEXT: movdqa %xmm2, %xmm3 -; X32-SSE-NEXT: pandn %xmm0, %xmm3 -; X32-SSE-NEXT: psrlw $2, %xmm0 +; X32-SSE-NEXT: movdqa {{.*#+}} xmm2 = [65535,0,0,0,65535,0,0,0] +; X32-SSE-NEXT: pand %xmm2, %xmm1 ; X32-SSE-NEXT: pand %xmm2, %xmm0 -; X32-SSE-NEXT: por %xmm3, %xmm0 -; X32-SSE-NEXT: paddw %xmm1, %xmm1 -; X32-SSE-NEXT: psraw $15, %xmm1 -; X32-SSE-NEXT: movdqa %xmm1, %xmm2 -; X32-SSE-NEXT: pandn %xmm0, %xmm2 -; X32-SSE-NEXT: psrlw $1, %xmm0 -; X32-SSE-NEXT: pand %xmm1, %xmm0 -; X32-SSE-NEXT: por %xmm2, %xmm0 +; X32-SSE-NEXT: movdqa %xmm0, %xmm2 +; X32-SSE-NEXT: psrlq %xmm1, %xmm2 +; X32-SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1] +; X32-SSE-NEXT: psrlq %xmm1, %xmm0 +; X32-SSE-NEXT: movsd {{.*#+}} xmm0 = xmm2[0],xmm0[1] ; X32-SSE-NEXT: retl - %shift = lshr <4 x i16> %a, %b - ret <4 x i16> %shift + %shift = lshr <2 x i16> %a, %b + ret <2 x i16> %shift } -define <2 x i16> @var_shift_v2i16(<2 x i16> %a, <2 x i16> %b) nounwind { -; SSE2-LABEL: var_shift_v2i16: +define <8 x i8> @var_shift_v8i8(<8 x i8> %a, <8 x i8> %b) nounwind { +; SSE2-LABEL: var_shift_v8i8: ; SSE2: # %bb.0: +; SSE2-NEXT: movdqa %xmm0, %xmm2 +; SSE2-NEXT: pand {{.*}}(%rip), %xmm2 ; SSE2-NEXT: psllw $12, %xmm1 -; SSE2-NEXT: movdqa %xmm1, %xmm2 -; SSE2-NEXT: psraw $15, %xmm2 -; SSE2-NEXT: movdqa %xmm2, %xmm3 -; SSE2-NEXT: pandn %xmm0, %xmm3 -; SSE2-NEXT: psrlw $8, %xmm0 -; SSE2-NEXT: pand %xmm2, %xmm0 -; SSE2-NEXT: por %xmm3, %xmm0 +; SSE2-NEXT: movdqa %xmm1, %xmm0 +; SSE2-NEXT: psraw $15, %xmm0 +; SSE2-NEXT: pandn %xmm2, %xmm0 ; SSE2-NEXT: paddw %xmm1, %xmm1 ; SSE2-NEXT: movdqa %xmm1, %xmm2 ; SSE2-NEXT: psraw $15, %xmm2 @@ -333,18 +369,20 @@ define <2 x i16> @var_shift_v2i16(<2 x i16> %a, <2 x i16> %b) nounwind { ; SSE2-NEXT: por %xmm2, %xmm0 ; SSE2-NEXT: retq ; -; SSE41-LABEL: var_shift_v2i16: +; SSE41-LABEL: var_shift_v8i8: ; SSE41: # %bb.0: ; SSE41-NEXT: movdqa %xmm1, %xmm2 ; SSE41-NEXT: movdqa %xmm0, %xmm1 +; SSE41-NEXT: movdqa {{.*#+}} xmm0 = [255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0] +; SSE41-NEXT: pand %xmm0, %xmm1 +; SSE41-NEXT: pand %xmm0, %xmm2 ; SSE41-NEXT: movdqa %xmm2, %xmm0 ; SSE41-NEXT: psllw $12, %xmm0 ; SSE41-NEXT: psllw $4, %xmm2 ; SSE41-NEXT: por %xmm0, %xmm2 ; SSE41-NEXT: movdqa %xmm2, %xmm3 ; SSE41-NEXT: paddw %xmm2, %xmm3 -; SSE41-NEXT: movdqa %xmm1, %xmm4 -; SSE41-NEXT: psrlw $8, %xmm4 +; SSE41-NEXT: pxor %xmm4, %xmm4 ; SSE41-NEXT: movdqa %xmm2, %xmm0 ; SSE41-NEXT: pblendvb %xmm0, %xmm4, %xmm1 ; SSE41-NEXT: movdqa %xmm1, %xmm2 @@ -364,13 +402,16 @@ define <2 x i16> @var_shift_v2i16(<2 x i16> %a, <2 x i16> %b) nounwind { ; SSE41-NEXT: movdqa %xmm1, %xmm0 ; SSE41-NEXT: retq ; -; AVX1-LABEL: var_shift_v2i16: +; AVX1-LABEL: var_shift_v8i8: ; AVX1: # %bb.0: +; AVX1-NEXT: vmovdqa {{.*#+}} xmm2 = [255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0] +; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm0 +; AVX1-NEXT: vpand %xmm2, %xmm1, %xmm1 ; AVX1-NEXT: vpsllw $12, %xmm1, %xmm2 ; AVX1-NEXT: vpsllw $4, %xmm1, %xmm1 ; AVX1-NEXT: vpor %xmm2, %xmm1, %xmm1 ; AVX1-NEXT: vpaddw %xmm1, %xmm1, %xmm2 -; AVX1-NEXT: vpsrlw $8, %xmm0, %xmm3 +; AVX1-NEXT: vpxor %xmm3, %xmm3, %xmm3 ; AVX1-NEXT: vpblendvb %xmm1, %xmm3, %xmm0, %xmm0 ; AVX1-NEXT: vpsrlw $4, %xmm0, %xmm1 ; AVX1-NEXT: vpblendvb %xmm2, %xmm1, %xmm0, %xmm0 @@ -382,8 +423,11 @@ define <2 x i16> @var_shift_v2i16(<2 x i16> %a, <2 x i16> %b) nounwind { ; AVX1-NEXT: vpblendvb %xmm2, %xmm1, %xmm0, %xmm0 ; AVX1-NEXT: retq ; -; AVX2-LABEL: var_shift_v2i16: +; AVX2-LABEL: var_shift_v8i8: ; AVX2: # %bb.0: +; AVX2-NEXT: vpbroadcastw {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255] +; AVX2-NEXT: vpand %xmm2, %xmm0, %xmm0 +; AVX2-NEXT: vpand %xmm2, %xmm1, %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: 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 ; AVX2-NEXT: vpsrlvd %ymm1, %ymm0, %ymm0 @@ -393,15 +437,31 @@ define <2 x i16> @var_shift_v2i16(<2 x i16> %a, <2 x i16> %b) nounwind { ; AVX2-NEXT: vzeroupper ; AVX2-NEXT: retq ; -; XOP-LABEL: var_shift_v2i16: -; XOP: # %bb.0: -; XOP-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; XOP-NEXT: vpsubw %xmm1, %xmm2, %xmm1 -; XOP-NEXT: vpshlw %xmm1, %xmm0, %xmm0 -; XOP-NEXT: retq +; XOPAVX1-LABEL: var_shift_v8i8: +; XOPAVX1: # %bb.0: +; XOPAVX1-NEXT: vmovdqa {{.*#+}} xmm2 = [255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0] +; XOPAVX1-NEXT: vpand %xmm2, %xmm0, %xmm0 +; XOPAVX1-NEXT: vpand %xmm2, %xmm1, %xmm1 +; XOPAVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; XOPAVX1-NEXT: vpsubw %xmm1, %xmm2, %xmm1 +; XOPAVX1-NEXT: vpshlw %xmm1, %xmm0, %xmm0 +; XOPAVX1-NEXT: retq ; -; AVX512DQ-LABEL: var_shift_v2i16: +; XOPAVX2-LABEL: var_shift_v8i8: +; XOPAVX2: # %bb.0: +; XOPAVX2-NEXT: vpbroadcastw {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255] +; XOPAVX2-NEXT: vpand %xmm2, %xmm0, %xmm0 +; XOPAVX2-NEXT: vpand %xmm2, %xmm1, %xmm1 +; XOPAVX2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; XOPAVX2-NEXT: vpsubw %xmm1, %xmm2, %xmm1 +; XOPAVX2-NEXT: vpshlw %xmm1, %xmm0, %xmm0 +; XOPAVX2-NEXT: retq +; +; AVX512DQ-LABEL: var_shift_v8i8: ; AVX512DQ: # %bb.0: +; AVX512DQ-NEXT: vpbroadcastw {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255] +; AVX512DQ-NEXT: vpand %xmm2, %xmm0, %xmm0 +; AVX512DQ-NEXT: vpand %xmm2, %xmm1, %xmm1 ; AVX512DQ-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 ; AVX512DQ-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 ; AVX512DQ-NEXT: vpsrlvd %ymm1, %ymm0, %ymm0 @@ -410,17 +470,21 @@ define <2 x i16> @var_shift_v2i16(<2 x i16> %a, <2 x i16> %b) nounwind { ; AVX512DQ-NEXT: vzeroupper ; AVX512DQ-NEXT: retq ; -; AVX512BW-LABEL: var_shift_v2i16: +; AVX512BW-LABEL: var_shift_v8i8: ; AVX512BW: # %bb.0: -; AVX512BW-NEXT: # kill: def $xmm1 killed $xmm1 def $zmm1 -; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0 +; AVX512BW-NEXT: vpbroadcastw {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255] +; AVX512BW-NEXT: vpand %xmm2, %xmm0, %xmm0 +; AVX512BW-NEXT: vpand %xmm2, %xmm1, %xmm1 ; AVX512BW-NEXT: vpsrlvw %zmm1, %zmm0, %zmm0 ; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0 ; AVX512BW-NEXT: vzeroupper ; AVX512BW-NEXT: retq ; -; AVX512DQVL-LABEL: var_shift_v2i16: +; AVX512DQVL-LABEL: var_shift_v8i8: ; AVX512DQVL: # %bb.0: +; AVX512DQVL-NEXT: vpbroadcastw {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255] +; AVX512DQVL-NEXT: vpand %xmm2, %xmm0, %xmm0 +; AVX512DQVL-NEXT: vpand %xmm2, %xmm1, %xmm1 ; AVX512DQVL-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 ; AVX512DQVL-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 ; AVX512DQVL-NEXT: vpsrlvd %ymm1, %ymm0, %ymm0 @@ -428,21 +492,22 @@ define <2 x i16> @var_shift_v2i16(<2 x i16> %a, <2 x i16> %b) nounwind { ; AVX512DQVL-NEXT: vzeroupper ; AVX512DQVL-NEXT: retq ; -; AVX512BWVL-LABEL: var_shift_v2i16: +; AVX512BWVL-LABEL: var_shift_v8i8: ; AVX512BWVL: # %bb.0: +; AVX512BWVL-NEXT: vpbroadcastw {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255] +; AVX512BWVL-NEXT: vpand %xmm2, %xmm1, %xmm1 +; AVX512BWVL-NEXT: vpand %xmm2, %xmm0, %xmm0 ; AVX512BWVL-NEXT: vpsrlvw %xmm1, %xmm0, %xmm0 ; AVX512BWVL-NEXT: retq ; -; X32-SSE-LABEL: var_shift_v2i16: +; X32-SSE-LABEL: var_shift_v8i8: ; X32-SSE: # %bb.0: +; X32-SSE-NEXT: movdqa %xmm0, %xmm2 +; X32-SSE-NEXT: pand {{\.LCPI.*}}, %xmm2 ; X32-SSE-NEXT: psllw $12, %xmm1 -; X32-SSE-NEXT: movdqa %xmm1, %xmm2 -; X32-SSE-NEXT: psraw $15, %xmm2 -; X32-SSE-NEXT: movdqa %xmm2, %xmm3 -; X32-SSE-NEXT: pandn %xmm0, %xmm3 -; X32-SSE-NEXT: psrlw $8, %xmm0 -; X32-SSE-NEXT: pand %xmm2, %xmm0 -; X32-SSE-NEXT: por %xmm3, %xmm0 +; X32-SSE-NEXT: movdqa %xmm1, %xmm0 +; X32-SSE-NEXT: psraw $15, %xmm0 +; X32-SSE-NEXT: pandn %xmm2, %xmm0 ; X32-SSE-NEXT: paddw %xmm1, %xmm1 ; X32-SSE-NEXT: movdqa %xmm1, %xmm2 ; X32-SSE-NEXT: psraw $15, %xmm2 @@ -467,156 +532,6 @@ define <2 x i16> @var_shift_v2i16(<2 x i16> %a, <2 x i16> %b) nounwind { ; X32-SSE-NEXT: pand %xmm1, %xmm0 ; X32-SSE-NEXT: por %xmm2, %xmm0 ; X32-SSE-NEXT: retl - %shift = lshr <2 x i16> %a, %b - ret <2 x i16> %shift -} - -define <8 x i8> @var_shift_v8i8(<8 x i8> %a, <8 x i8> %b) nounwind { -; SSE2-LABEL: var_shift_v8i8: -; SSE2: # %bb.0: -; SSE2-NEXT: psllw $5, %xmm1 -; SSE2-NEXT: pxor %xmm2, %xmm2 -; SSE2-NEXT: pxor %xmm3, %xmm3 -; SSE2-NEXT: pcmpgtb %xmm1, %xmm3 -; SSE2-NEXT: movdqa %xmm3, %xmm4 -; SSE2-NEXT: pandn %xmm0, %xmm4 -; SSE2-NEXT: psrlw $4, %xmm0 -; SSE2-NEXT: pand %xmm3, %xmm0 -; SSE2-NEXT: pand {{.*}}(%rip), %xmm0 -; SSE2-NEXT: por %xmm4, %xmm0 -; SSE2-NEXT: paddb %xmm1, %xmm1 -; SSE2-NEXT: pxor %xmm3, %xmm3 -; SSE2-NEXT: pcmpgtb %xmm1, %xmm3 -; SSE2-NEXT: movdqa %xmm3, %xmm4 -; SSE2-NEXT: pandn %xmm0, %xmm4 -; SSE2-NEXT: psrlw $2, %xmm0 -; SSE2-NEXT: pand %xmm3, %xmm0 -; SSE2-NEXT: pand {{.*}}(%rip), %xmm0 -; SSE2-NEXT: por %xmm4, %xmm0 -; SSE2-NEXT: paddb %xmm1, %xmm1 -; SSE2-NEXT: pcmpgtb %xmm1, %xmm2 -; SSE2-NEXT: movdqa %xmm2, %xmm1 -; SSE2-NEXT: pandn %xmm0, %xmm1 -; SSE2-NEXT: psrlw $1, %xmm0 -; SSE2-NEXT: pand %xmm2, %xmm0 -; SSE2-NEXT: pand {{.*}}(%rip), %xmm0 -; SSE2-NEXT: por %xmm1, %xmm0 -; SSE2-NEXT: retq -; -; SSE41-LABEL: var_shift_v8i8: -; SSE41: # %bb.0: -; SSE41-NEXT: movdqa %xmm0, %xmm2 -; SSE41-NEXT: psllw $5, %xmm1 -; SSE41-NEXT: movdqa %xmm0, %xmm3 -; SSE41-NEXT: psrlw $4, %xmm3 -; SSE41-NEXT: pand {{.*}}(%rip), %xmm3 -; SSE41-NEXT: movdqa %xmm1, %xmm0 -; SSE41-NEXT: pblendvb %xmm0, %xmm3, %xmm2 -; SSE41-NEXT: movdqa %xmm2, %xmm3 -; SSE41-NEXT: psrlw $2, %xmm3 -; SSE41-NEXT: pand {{.*}}(%rip), %xmm3 -; SSE41-NEXT: paddb %xmm1, %xmm1 -; SSE41-NEXT: movdqa %xmm1, %xmm0 -; SSE41-NEXT: pblendvb %xmm0, %xmm3, %xmm2 -; SSE41-NEXT: movdqa %xmm2, %xmm3 -; SSE41-NEXT: psrlw $1, %xmm3 -; SSE41-NEXT: pand {{.*}}(%rip), %xmm3 -; SSE41-NEXT: paddb %xmm1, %xmm1 -; SSE41-NEXT: movdqa %xmm1, %xmm0 -; SSE41-NEXT: pblendvb %xmm0, %xmm3, %xmm2 -; SSE41-NEXT: movdqa %xmm2, %xmm0 -; SSE41-NEXT: retq -; -; AVX-LABEL: var_shift_v8i8: -; AVX: # %bb.0: -; AVX-NEXT: vpsllw $5, %xmm1, %xmm1 -; AVX-NEXT: vpsrlw $4, %xmm0, %xmm2 -; AVX-NEXT: vpand {{.*}}(%rip), %xmm2, %xmm2 -; AVX-NEXT: vpblendvb %xmm1, %xmm2, %xmm0, %xmm0 -; AVX-NEXT: vpsrlw $2, %xmm0, %xmm2 -; AVX-NEXT: vpand {{.*}}(%rip), %xmm2, %xmm2 -; AVX-NEXT: vpaddb %xmm1, %xmm1, %xmm1 -; AVX-NEXT: vpblendvb %xmm1, %xmm2, %xmm0, %xmm0 -; AVX-NEXT: vpsrlw $1, %xmm0, %xmm2 -; AVX-NEXT: vpand {{.*}}(%rip), %xmm2, %xmm2 -; AVX-NEXT: vpaddb %xmm1, %xmm1, %xmm1 -; AVX-NEXT: vpblendvb %xmm1, %xmm2, %xmm0, %xmm0 -; AVX-NEXT: retq -; -; XOP-LABEL: var_shift_v8i8: -; XOP: # %bb.0: -; XOP-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; XOP-NEXT: vpsubb %xmm1, %xmm2, %xmm1 -; XOP-NEXT: vpshlb %xmm1, %xmm0, %xmm0 -; XOP-NEXT: retq -; -; AVX512DQ-LABEL: var_shift_v8i8: -; AVX512DQ: # %bb.0: -; AVX512DQ-NEXT: vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero -; AVX512DQ-NEXT: vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero -; AVX512DQ-NEXT: vpsrlvd %zmm1, %zmm0, %zmm0 -; AVX512DQ-NEXT: vpmovdb %zmm0, %xmm0 -; AVX512DQ-NEXT: vzeroupper -; AVX512DQ-NEXT: retq -; -; AVX512BW-LABEL: var_shift_v8i8: -; AVX512BW: # %bb.0: -; AVX512BW-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 -; AVX512BW-NEXT: vpmovzxbw {{.*#+}} 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,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero -; AVX512BW-NEXT: vpsrlvw %zmm1, %zmm0, %zmm0 -; AVX512BW-NEXT: vpmovwb %zmm0, %ymm0 -; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 -; AVX512BW-NEXT: vzeroupper -; AVX512BW-NEXT: retq -; -; AVX512DQVL-LABEL: var_shift_v8i8: -; AVX512DQVL: # %bb.0: -; AVX512DQVL-NEXT: vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero -; AVX512DQVL-NEXT: vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero -; AVX512DQVL-NEXT: vpsrlvd %zmm1, %zmm0, %zmm0 -; AVX512DQVL-NEXT: vpmovdb %zmm0, %xmm0 -; AVX512DQVL-NEXT: vzeroupper -; AVX512DQVL-NEXT: retq -; -; AVX512BWVL-LABEL: var_shift_v8i8: -; AVX512BWVL: # %bb.0: -; AVX512BWVL-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 -; AVX512BWVL-NEXT: vpmovzxbw {{.*#+}} 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,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero -; AVX512BWVL-NEXT: vpsrlvw %ymm1, %ymm0, %ymm0 -; AVX512BWVL-NEXT: vpmovwb %ymm0, %xmm0 -; AVX512BWVL-NEXT: vzeroupper -; AVX512BWVL-NEXT: retq -; -; X32-SSE-LABEL: var_shift_v8i8: -; X32-SSE: # %bb.0: -; X32-SSE-NEXT: psllw $5, %xmm1 -; X32-SSE-NEXT: pxor %xmm2, %xmm2 -; X32-SSE-NEXT: pxor %xmm3, %xmm3 -; X32-SSE-NEXT: pcmpgtb %xmm1, %xmm3 -; X32-SSE-NEXT: movdqa %xmm3, %xmm4 -; X32-SSE-NEXT: pandn %xmm0, %xmm4 -; X32-SSE-NEXT: psrlw $4, %xmm0 -; X32-SSE-NEXT: pand %xmm3, %xmm0 -; X32-SSE-NEXT: pand {{\.LCPI.*}}, %xmm0 -; X32-SSE-NEXT: por %xmm4, %xmm0 -; X32-SSE-NEXT: paddb %xmm1, %xmm1 -; X32-SSE-NEXT: pxor %xmm3, %xmm3 -; X32-SSE-NEXT: pcmpgtb %xmm1, %xmm3 -; X32-SSE-NEXT: movdqa %xmm3, %xmm4 -; X32-SSE-NEXT: pandn %xmm0, %xmm4 -; X32-SSE-NEXT: psrlw $2, %xmm0 -; X32-SSE-NEXT: pand %xmm3, %xmm0 -; X32-SSE-NEXT: pand {{\.LCPI.*}}, %xmm0 -; X32-SSE-NEXT: por %xmm4, %xmm0 -; X32-SSE-NEXT: paddb %xmm1, %xmm1 -; X32-SSE-NEXT: pcmpgtb %xmm1, %xmm2 -; X32-SSE-NEXT: movdqa %xmm2, %xmm1 -; X32-SSE-NEXT: pandn %xmm0, %xmm1 -; X32-SSE-NEXT: psrlw $1, %xmm0 -; X32-SSE-NEXT: pand %xmm2, %xmm0 -; X32-SSE-NEXT: pand {{\.LCPI.*}}, %xmm0 -; X32-SSE-NEXT: por %xmm1, %xmm0 -; X32-SSE-NEXT: retl %shift = lshr <8 x i8> %a, %b ret <8 x i8> %shift } @@ -624,148 +539,131 @@ define <8 x i8> @var_shift_v8i8(<8 x i8> %a, <8 x i8> %b) nounwind { define <4 x i8> @var_shift_v4i8(<4 x i8> %a, <4 x i8> %b) nounwind { ; SSE2-LABEL: var_shift_v4i8: ; SSE2: # %bb.0: -; SSE2-NEXT: psllw $5, %xmm1 -; SSE2-NEXT: pxor %xmm2, %xmm2 -; SSE2-NEXT: pxor %xmm3, %xmm3 -; SSE2-NEXT: pcmpgtb %xmm1, %xmm3 -; SSE2-NEXT: movdqa %xmm3, %xmm4 -; SSE2-NEXT: pandn %xmm0, %xmm4 -; SSE2-NEXT: psrlw $4, %xmm0 -; SSE2-NEXT: pand %xmm3, %xmm0 -; SSE2-NEXT: pand {{.*}}(%rip), %xmm0 -; SSE2-NEXT: por %xmm4, %xmm0 -; SSE2-NEXT: paddb %xmm1, %xmm1 -; SSE2-NEXT: pxor %xmm3, %xmm3 -; SSE2-NEXT: pcmpgtb %xmm1, %xmm3 -; SSE2-NEXT: movdqa %xmm3, %xmm4 -; SSE2-NEXT: pandn %xmm0, %xmm4 -; SSE2-NEXT: psrlw $2, %xmm0 -; SSE2-NEXT: pand %xmm3, %xmm0 -; SSE2-NEXT: pand {{.*}}(%rip), %xmm0 -; SSE2-NEXT: por %xmm4, %xmm0 -; SSE2-NEXT: paddb %xmm1, %xmm1 -; SSE2-NEXT: pcmpgtb %xmm1, %xmm2 -; SSE2-NEXT: movdqa %xmm2, %xmm1 -; SSE2-NEXT: pandn %xmm0, %xmm1 -; SSE2-NEXT: psrlw $1, %xmm0 +; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0] ; SSE2-NEXT: pand %xmm2, %xmm0 -; SSE2-NEXT: pand {{.*}}(%rip), %xmm0 -; SSE2-NEXT: por %xmm1, %xmm0 +; SSE2-NEXT: pand %xmm2, %xmm1 +; SSE2-NEXT: pshuflw {{.*#+}} xmm2 = xmm1[2,3,3,3,4,5,6,7] +; SSE2-NEXT: movdqa %xmm0, %xmm3 +; SSE2-NEXT: psrld %xmm2, %xmm3 +; SSE2-NEXT: pshuflw {{.*#+}} xmm4 = xmm1[0,1,1,1,4,5,6,7] +; SSE2-NEXT: movdqa %xmm0, %xmm2 +; SSE2-NEXT: psrld %xmm4, %xmm2 +; SSE2-NEXT: punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm3[0] +; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1] +; SSE2-NEXT: pshuflw {{.*#+}} xmm3 = xmm1[2,3,3,3,4,5,6,7] +; SSE2-NEXT: movdqa %xmm0, %xmm4 +; SSE2-NEXT: psrld %xmm3, %xmm4 +; SSE2-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,1,1,1,4,5,6,7] +; SSE2-NEXT: psrld %xmm1, %xmm0 +; SSE2-NEXT: punpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm4[1] +; SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,3],xmm0[0,3] +; SSE2-NEXT: movaps %xmm2, %xmm0 ; SSE2-NEXT: retq ; ; SSE41-LABEL: var_shift_v4i8: ; SSE41: # %bb.0: -; SSE41-NEXT: movdqa %xmm0, %xmm2 -; SSE41-NEXT: psllw $5, %xmm1 +; SSE41-NEXT: movdqa {{.*#+}} xmm2 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0] +; SSE41-NEXT: pand %xmm2, %xmm0 +; SSE41-NEXT: pand %xmm2, %xmm1 +; SSE41-NEXT: pshuflw {{.*#+}} xmm2 = xmm1[2,3,3,3,4,5,6,7] ; SSE41-NEXT: movdqa %xmm0, %xmm3 -; SSE41-NEXT: psrlw $4, %xmm3 -; SSE41-NEXT: pand {{.*}}(%rip), %xmm3 -; SSE41-NEXT: movdqa %xmm1, %xmm0 -; SSE41-NEXT: pblendvb %xmm0, %xmm3, %xmm2 -; SSE41-NEXT: movdqa %xmm2, %xmm3 -; SSE41-NEXT: psrlw $2, %xmm3 -; SSE41-NEXT: pand {{.*}}(%rip), %xmm3 -; SSE41-NEXT: paddb %xmm1, %xmm1 -; SSE41-NEXT: movdqa %xmm1, %xmm0 -; SSE41-NEXT: pblendvb %xmm0, %xmm3, %xmm2 -; SSE41-NEXT: movdqa %xmm2, %xmm3 -; SSE41-NEXT: psrlw $1, %xmm3 -; SSE41-NEXT: pand {{.*}}(%rip), %xmm3 -; SSE41-NEXT: paddb %xmm1, %xmm1 -; SSE41-NEXT: movdqa %xmm1, %xmm0 -; SSE41-NEXT: pblendvb %xmm0, %xmm3, %xmm2 -; SSE41-NEXT: movdqa %xmm2, %xmm0 +; SSE41-NEXT: psrld %xmm2, %xmm3 +; SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm1[2,3,0,1] +; SSE41-NEXT: pshuflw {{.*#+}} xmm4 = xmm2[2,3,3,3,4,5,6,7] +; SSE41-NEXT: movdqa %xmm0, %xmm5 +; SSE41-NEXT: psrld %xmm4, %xmm5 +; SSE41-NEXT: pblendw {{.*#+}} xmm5 = xmm3[0,1,2,3],xmm5[4,5,6,7] +; SSE41-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,1,1,1,4,5,6,7] +; SSE41-NEXT: movdqa %xmm0, %xmm3 +; SSE41-NEXT: psrld %xmm1, %xmm3 +; SSE41-NEXT: pshuflw {{.*#+}} xmm1 = xmm2[0,1,1,1,4,5,6,7] +; SSE41-NEXT: psrld %xmm1, %xmm0 +; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm3[0,1,2,3],xmm0[4,5,6,7] +; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm5[2,3],xmm0[4,5],xmm5[6,7] ; SSE41-NEXT: retq ; -; AVX-LABEL: var_shift_v4i8: -; AVX: # %bb.0: -; AVX-NEXT: vpsllw $5, %xmm1, %xmm1 -; AVX-NEXT: vpsrlw $4, %xmm0, %xmm2 -; AVX-NEXT: vpand {{.*}}(%rip), %xmm2, %xmm2 -; AVX-NEXT: vpblendvb %xmm1, %xmm2, %xmm0, %xmm0 -; AVX-NEXT: vpsrlw $2, %xmm0, %xmm2 -; AVX-NEXT: vpand {{.*}}(%rip), %xmm2, %xmm2 -; AVX-NEXT: vpaddb %xmm1, %xmm1, %xmm1 -; AVX-NEXT: vpblendvb %xmm1, %xmm2, %xmm0, %xmm0 -; AVX-NEXT: vpsrlw $1, %xmm0, %xmm2 -; AVX-NEXT: vpand {{.*}}(%rip), %xmm2, %xmm2 -; AVX-NEXT: vpaddb %xmm1, %xmm1, %xmm1 -; AVX-NEXT: vpblendvb %xmm1, %xmm2, %xmm0, %xmm0 -; AVX-NEXT: retq +; AVX1-LABEL: var_shift_v4i8: +; AVX1: # %bb.0: +; AVX1-NEXT: vbroadcastss {{.*#+}} xmm2 = [3.57331108E-43,3.57331108E-43,3.57331108E-43,3.57331108E-43] +; AVX1-NEXT: vandps %xmm2, %xmm0, %xmm0 +; AVX1-NEXT: vandps %xmm2, %xmm1, %xmm1 +; AVX1-NEXT: vpsrldq {{.*#+}} xmm2 = xmm1[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero +; AVX1-NEXT: vpsrld %xmm2, %xmm0, %xmm2 +; AVX1-NEXT: vpsrlq $32, %xmm1, %xmm3 +; AVX1-NEXT: vpsrld %xmm3, %xmm0, %xmm3 +; AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm3[0,1,2,3],xmm2[4,5,6,7] +; AVX1-NEXT: vpxor %xmm3, %xmm3, %xmm3 +; AVX1-NEXT: vpunpckhdq {{.*#+}} xmm3 = xmm1[2],xmm3[2],xmm1[3],xmm3[3] +; AVX1-NEXT: vpsrld %xmm3, %xmm0, %xmm3 +; AVX1-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero +; AVX1-NEXT: vpsrld %xmm1, %xmm0, %xmm0 +; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm3[4,5,6,7] +; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3],xmm0[4,5],xmm2[6,7] +; AVX1-NEXT: retq ; -; XOP-LABEL: var_shift_v4i8: -; XOP: # %bb.0: -; XOP-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; XOP-NEXT: vpsubb %xmm1, %xmm2, %xmm1 -; XOP-NEXT: vpshlb %xmm1, %xmm0, %xmm0 -; XOP-NEXT: retq +; AVX2-LABEL: var_shift_v4i8: +; AVX2: # %bb.0: +; AVX2-NEXT: vpbroadcastd {{.*#+}} xmm2 = [255,255,255,255] +; AVX2-NEXT: vpand %xmm2, %xmm1, %xmm1 +; AVX2-NEXT: vpand %xmm2, %xmm0, %xmm0 +; AVX2-NEXT: vpsrlvd %xmm1, %xmm0, %xmm0 +; AVX2-NEXT: retq ; -; AVX512DQ-LABEL: var_shift_v4i8: -; AVX512DQ: # %bb.0: -; AVX512DQ-NEXT: vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero -; AVX512DQ-NEXT: vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero -; AVX512DQ-NEXT: vpsrlvd %zmm1, %zmm0, %zmm0 -; AVX512DQ-NEXT: vpmovdb %zmm0, %xmm0 -; AVX512DQ-NEXT: vzeroupper -; AVX512DQ-NEXT: retq +; XOPAVX1-LABEL: var_shift_v4i8: +; XOPAVX1: # %bb.0: +; XOPAVX1-NEXT: vbroadcastss {{.*#+}} xmm2 = [3.57331108E-43,3.57331108E-43,3.57331108E-43,3.57331108E-43] +; XOPAVX1-NEXT: vandps %xmm2, %xmm0, %xmm0 +; XOPAVX1-NEXT: vandps %xmm2, %xmm1, %xmm1 +; XOPAVX1-NEXT: vxorps %xmm2, %xmm2, %xmm2 +; XOPAVX1-NEXT: vpsubd %xmm1, %xmm2, %xmm1 +; XOPAVX1-NEXT: vpshld %xmm1, %xmm0, %xmm0 +; XOPAVX1-NEXT: retq ; -; AVX512BW-LABEL: var_shift_v4i8: -; AVX512BW: # %bb.0: -; AVX512BW-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 -; AVX512BW-NEXT: vpmovzxbw {{.*#+}} 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,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero -; AVX512BW-NEXT: vpsrlvw %zmm1, %zmm0, %zmm0 -; AVX512BW-NEXT: vpmovwb %zmm0, %ymm0 -; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 -; AVX512BW-NEXT: vzeroupper -; AVX512BW-NEXT: retq +; XOPAVX2-LABEL: var_shift_v4i8: +; XOPAVX2: # %bb.0: +; XOPAVX2-NEXT: vpbroadcastd {{.*#+}} xmm2 = [255,255,255,255] +; XOPAVX2-NEXT: vpand %xmm2, %xmm1, %xmm1 +; XOPAVX2-NEXT: vpand %xmm2, %xmm0, %xmm0 +; XOPAVX2-NEXT: vpsrlvd %xmm1, %xmm0, %xmm0 +; XOPAVX2-NEXT: retq ; -; AVX512DQVL-LABEL: var_shift_v4i8: -; AVX512DQVL: # %bb.0: -; AVX512DQVL-NEXT: vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero -; AVX512DQVL-NEXT: vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero -; AVX512DQVL-NEXT: vpsrlvd %zmm1, %zmm0, %zmm0 -; AVX512DQVL-NEXT: vpmovdb %zmm0, %xmm0 -; AVX512DQVL-NEXT: vzeroupper -; AVX512DQVL-NEXT: retq +; AVX512-LABEL: var_shift_v4i8: +; AVX512: # %bb.0: +; AVX512-NEXT: vpbroadcastd {{.*#+}} xmm2 = [255,255,255,255] +; AVX512-NEXT: vpand %xmm2, %xmm1, %xmm1 +; AVX512-NEXT: vpand %xmm2, %xmm0, %xmm0 +; AVX512-NEXT: vpsrlvd %xmm1, %xmm0, %xmm0 +; AVX512-NEXT: retq ; -; AVX512BWVL-LABEL: var_shift_v4i8: -; AVX512BWVL: # %bb.0: -; AVX512BWVL-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 -; AVX512BWVL-NEXT: vpmovzxbw {{.*#+}} 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,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero -; AVX512BWVL-NEXT: vpsrlvw %ymm1, %ymm0, %ymm0 -; AVX512BWVL-NEXT: vpmovwb %ymm0, %xmm0 -; AVX512BWVL-NEXT: vzeroupper -; AVX512BWVL-NEXT: retq +; AVX512VL-LABEL: var_shift_v4i8: +; AVX512VL: # %bb.0: +; AVX512VL-NEXT: vpbroadcastd {{.*#+}} xmm2 = [255,255,255,255] +; AVX512VL-NEXT: vpand %xmm2, %xmm1, %xmm1 +; AVX512VL-NEXT: vpand %xmm2, %xmm0, %xmm0 +; AVX512VL-NEXT: vpsrlvd %xmm1, %xmm0, %xmm0 +; AVX512VL-NEXT: retq ; ; X32-SSE-LABEL: var_shift_v4i8: ; X32-SSE: # %bb.0: -; X32-SSE-NEXT: psllw $5, %xmm1 -; X32-SSE-NEXT: pxor %xmm2, %xmm2 -; X32-SSE-NEXT: pxor %xmm3, %xmm3 -; X32-SSE-NEXT: pcmpgtb %xmm1, %xmm3 -; X32-SSE-NEXT: movdqa %xmm3, %xmm4 -; X32-SSE-NEXT: pandn %xmm0, %xmm4 -; X32-SSE-NEXT: psrlw $4, %xmm0 -; X32-SSE-NEXT: pand %xmm3, %xmm0 -; X32-SSE-NEXT: pand {{\.LCPI.*}}, %xmm0 -; X32-SSE-NEXT: por %xmm4, %xmm0 -; X32-SSE-NEXT: paddb %xmm1, %xmm1 -; X32-SSE-NEXT: pxor %xmm3, %xmm3 -; X32-SSE-NEXT: pcmpgtb %xmm1, %xmm3 -; X32-SSE-NEXT: movdqa %xmm3, %xmm4 -; X32-SSE-NEXT: pandn %xmm0, %xmm4 -; X32-SSE-NEXT: psrlw $2, %xmm0 -; X32-SSE-NEXT: pand %xmm3, %xmm0 -; X32-SSE-NEXT: pand {{\.LCPI.*}}, %xmm0 -; X32-SSE-NEXT: por %xmm4, %xmm0 -; X32-SSE-NEXT: paddb %xmm1, %xmm1 -; X32-SSE-NEXT: pcmpgtb %xmm1, %xmm2 -; X32-SSE-NEXT: movdqa %xmm2, %xmm1 -; X32-SSE-NEXT: pandn %xmm0, %xmm1 -; X32-SSE-NEXT: psrlw $1, %xmm0 +; X32-SSE-NEXT: movdqa {{.*#+}} xmm2 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0] ; X32-SSE-NEXT: pand %xmm2, %xmm0 -; X32-SSE-NEXT: pand {{\.LCPI.*}}, %xmm0 -; X32-SSE-NEXT: por %xmm1, %xmm0 +; X32-SSE-NEXT: pand %xmm2, %xmm1 +; X32-SSE-NEXT: pshuflw {{.*#+}} xmm2 = xmm1[2,3,3,3,4,5,6,7] +; X32-SSE-NEXT: movdqa %xmm0, %xmm3 +; X32-SSE-NEXT: psrld %xmm2, %xmm3 +; X32-SSE-NEXT: pshuflw {{.*#+}} xmm4 = xmm1[0,1,1,1,4,5,6,7] +; X32-SSE-NEXT: movdqa %xmm0, %xmm2 +; X32-SSE-NEXT: psrld %xmm4, %xmm2 +; X32-SSE-NEXT: punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm3[0] +; X32-SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1] +; X32-SSE-NEXT: pshuflw {{.*#+}} xmm3 = xmm1[2,3,3,3,4,5,6,7] +; X32-SSE-NEXT: movdqa %xmm0, %xmm4 +; X32-SSE-NEXT: psrld %xmm3, %xmm4 +; X32-SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,1,1,1,4,5,6,7] +; X32-SSE-NEXT: psrld %xmm1, %xmm0 +; X32-SSE-NEXT: punpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm4[1] +; X32-SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,3],xmm0[0,3] +; X32-SSE-NEXT: movaps %xmm2, %xmm0 ; X32-SSE-NEXT: retl %shift = lshr <4 x i8> %a, %b ret <4 x i8> %shift @@ -774,148 +672,93 @@ define <4 x i8> @var_shift_v4i8(<4 x i8> %a, <4 x i8> %b) nounwind { define <2 x i8> @var_shift_v2i8(<2 x i8> %a, <2 x i8> %b) nounwind { ; SSE2-LABEL: var_shift_v2i8: ; SSE2: # %bb.0: -; SSE2-NEXT: psllw $5, %xmm1 -; SSE2-NEXT: pxor %xmm2, %xmm2 -; SSE2-NEXT: pxor %xmm3, %xmm3 -; SSE2-NEXT: pcmpgtb %xmm1, %xmm3 -; SSE2-NEXT: movdqa %xmm3, %xmm4 -; SSE2-NEXT: pandn %xmm0, %xmm4 -; SSE2-NEXT: psrlw $4, %xmm0 -; SSE2-NEXT: pand %xmm3, %xmm0 -; SSE2-NEXT: pand {{.*}}(%rip), %xmm0 -; SSE2-NEXT: por %xmm4, %xmm0 -; SSE2-NEXT: paddb %xmm1, %xmm1 -; SSE2-NEXT: pxor %xmm3, %xmm3 -; SSE2-NEXT: pcmpgtb %xmm1, %xmm3 -; SSE2-NEXT: movdqa %xmm3, %xmm4 -; SSE2-NEXT: pandn %xmm0, %xmm4 -; SSE2-NEXT: psrlw $2, %xmm0 -; SSE2-NEXT: pand %xmm3, %xmm0 -; SSE2-NEXT: pand {{.*}}(%rip), %xmm0 -; SSE2-NEXT: por %xmm4, %xmm0 -; SSE2-NEXT: paddb %xmm1, %xmm1 -; SSE2-NEXT: pcmpgtb %xmm1, %xmm2 -; SSE2-NEXT: movdqa %xmm2, %xmm1 -; SSE2-NEXT: pandn %xmm0, %xmm1 -; SSE2-NEXT: psrlw $1, %xmm0 +; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0] +; SSE2-NEXT: pand %xmm2, %xmm1 ; SSE2-NEXT: pand %xmm2, %xmm0 -; SSE2-NEXT: pand {{.*}}(%rip), %xmm0 -; SSE2-NEXT: por %xmm1, %xmm0 +; SSE2-NEXT: movdqa %xmm0, %xmm2 +; SSE2-NEXT: psrlq %xmm1, %xmm2 +; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1] +; SSE2-NEXT: psrlq %xmm1, %xmm0 +; SSE2-NEXT: movsd {{.*#+}} xmm0 = xmm2[0],xmm0[1] ; SSE2-NEXT: retq ; ; SSE41-LABEL: var_shift_v2i8: ; SSE41: # %bb.0: +; SSE41-NEXT: movdqa {{.*#+}} xmm2 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0] +; SSE41-NEXT: pand %xmm2, %xmm1 +; SSE41-NEXT: pand %xmm2, %xmm0 ; SSE41-NEXT: movdqa %xmm0, %xmm2 -; SSE41-NEXT: psllw $5, %xmm1 -; SSE41-NEXT: movdqa %xmm0, %xmm3 -; SSE41-NEXT: psrlw $4, %xmm3 -; SSE41-NEXT: pand {{.*}}(%rip), %xmm3 -; SSE41-NEXT: movdqa %xmm1, %xmm0 -; SSE41-NEXT: pblendvb %xmm0, %xmm3, %xmm2 -; SSE41-NEXT: movdqa %xmm2, %xmm3 -; SSE41-NEXT: psrlw $2, %xmm3 -; SSE41-NEXT: pand {{.*}}(%rip), %xmm3 -; SSE41-NEXT: paddb %xmm1, %xmm1 -; SSE41-NEXT: movdqa %xmm1, %xmm0 -; SSE41-NEXT: pblendvb %xmm0, %xmm3, %xmm2 -; SSE41-NEXT: movdqa %xmm2, %xmm3 -; SSE41-NEXT: psrlw $1, %xmm3 -; SSE41-NEXT: pand {{.*}}(%rip), %xmm3 -; SSE41-NEXT: paddb %xmm1, %xmm1 -; SSE41-NEXT: movdqa %xmm1, %xmm0 -; SSE41-NEXT: pblendvb %xmm0, %xmm3, %xmm2 -; SSE41-NEXT: movdqa %xmm2, %xmm0 +; SSE41-NEXT: psrlq %xmm1, %xmm2 +; SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1] +; SSE41-NEXT: psrlq %xmm1, %xmm0 +; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm2[0,1,2,3],xmm0[4,5,6,7] ; SSE41-NEXT: retq ; -; AVX-LABEL: var_shift_v2i8: -; AVX: # %bb.0: -; AVX-NEXT: vpsllw $5, %xmm1, %xmm1 -; AVX-NEXT: vpsrlw $4, %xmm0, %xmm2 -; AVX-NEXT: vpand {{.*}}(%rip), %xmm2, %xmm2 -; AVX-NEXT: vpblendvb %xmm1, %xmm2, %xmm0, %xmm0 -; AVX-NEXT: vpsrlw $2, %xmm0, %xmm2 -; AVX-NEXT: vpand {{.*}}(%rip), %xmm2, %xmm2 -; AVX-NEXT: vpaddb %xmm1, %xmm1, %xmm1 -; AVX-NEXT: vpblendvb %xmm1, %xmm2, %xmm0, %xmm0 -; AVX-NEXT: vpsrlw $1, %xmm0, %xmm2 -; AVX-NEXT: vpand {{.*}}(%rip), %xmm2, %xmm2 -; AVX-NEXT: vpaddb %xmm1, %xmm1, %xmm1 -; AVX-NEXT: vpblendvb %xmm1, %xmm2, %xmm0, %xmm0 -; AVX-NEXT: retq +; AVX1-LABEL: var_shift_v2i8: +; AVX1: # %bb.0: +; AVX1-NEXT: vmovddup {{.*#+}} xmm2 = [1.2598673968951787E-321,1.2598673968951787E-321] +; AVX1-NEXT: # xmm2 = mem[0,0] +; AVX1-NEXT: vandps %xmm2, %xmm1, %xmm1 +; AVX1-NEXT: vandps %xmm2, %xmm0, %xmm0 +; AVX1-NEXT: vpsrlq %xmm1, %xmm0, %xmm2 +; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1] +; AVX1-NEXT: vpsrlq %xmm1, %xmm0, %xmm0 +; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm2[0,1,2,3],xmm0[4,5,6,7] +; AVX1-NEXT: retq ; -; XOP-LABEL: var_shift_v2i8: -; XOP: # %bb.0: -; XOP-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; XOP-NEXT: vpsubb %xmm1, %xmm2, %xmm1 -; XOP-NEXT: vpshlb %xmm1, %xmm0, %xmm0 -; XOP-NEXT: retq +; AVX2-LABEL: var_shift_v2i8: +; AVX2: # %bb.0: +; AVX2-NEXT: vpbroadcastq {{.*#+}} xmm2 = [255,255] +; AVX2-NEXT: vpand %xmm2, %xmm1, %xmm1 +; AVX2-NEXT: vpand %xmm2, %xmm0, %xmm0 +; AVX2-NEXT: vpsrlvq %xmm1, %xmm0, %xmm0 +; AVX2-NEXT: retq ; -; AVX512DQ-LABEL: var_shift_v2i8: -; AVX512DQ: # %bb.0: -; AVX512DQ-NEXT: vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero -; AVX512DQ-NEXT: vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero -; AVX512DQ-NEXT: vpsrlvd %zmm1, %zmm0, %zmm0 -; AVX512DQ-NEXT: vpmovdb %zmm0, %xmm0 -; AVX512DQ-NEXT: vzeroupper -; AVX512DQ-NEXT: retq +; XOPAVX1-LABEL: var_shift_v2i8: +; XOPAVX1: # %bb.0: +; XOPAVX1-NEXT: vmovddup {{.*#+}} xmm2 = [1.2598673968951787E-321,1.2598673968951787E-321] +; XOPAVX1-NEXT: # xmm2 = mem[0,0] +; XOPAVX1-NEXT: vandps %xmm2, %xmm0, %xmm0 +; XOPAVX1-NEXT: vandps %xmm2, %xmm1, %xmm1 +; XOPAVX1-NEXT: vxorps %xmm2, %xmm2, %xmm2 +; XOPAVX1-NEXT: vpsubq %xmm1, %xmm2, %xmm1 +; XOPAVX1-NEXT: vpshlq %xmm1, %xmm0, %xmm0 +; XOPAVX1-NEXT: retq ; -; AVX512BW-LABEL: var_shift_v2i8: -; AVX512BW: # %bb.0: -; AVX512BW-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 -; AVX512BW-NEXT: vpmovzxbw {{.*#+}} 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,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero -; AVX512BW-NEXT: vpsrlvw %zmm1, %zmm0, %zmm0 -; AVX512BW-NEXT: vpmovwb %zmm0, %ymm0 -; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 -; AVX512BW-NEXT: vzeroupper -; AVX512BW-NEXT: retq +; XOPAVX2-LABEL: var_shift_v2i8: +; XOPAVX2: # %bb.0: +; XOPAVX2-NEXT: vpbroadcastq {{.*#+}} xmm2 = [255,255] +; XOPAVX2-NEXT: vpand %xmm2, %xmm1, %xmm1 +; XOPAVX2-NEXT: vpand %xmm2, %xmm0, %xmm0 +; XOPAVX2-NEXT: vpsrlvq %xmm1, %xmm0, %xmm0 +; XOPAVX2-NEXT: retq ; -; AVX512DQVL-LABEL: var_shift_v2i8: -; AVX512DQVL: # %bb.0: -; AVX512DQVL-NEXT: vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero -; AVX512DQVL-NEXT: vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero -; AVX512DQVL-NEXT: vpsrlvd %zmm1, %zmm0, %zmm0 -; AVX512DQVL-NEXT: vpmovdb %zmm0, %xmm0 -; AVX512DQVL-NEXT: vzeroupper -; AVX512DQVL-NEXT: retq +; AVX512-LABEL: var_shift_v2i8: +; AVX512: # %bb.0: +; AVX512-NEXT: vpbroadcastq {{.*#+}} xmm2 = [255,255] +; AVX512-NEXT: vpand %xmm2, %xmm1, %xmm1 +; AVX512-NEXT: vpand %xmm2, %xmm0, %xmm0 +; AVX512-NEXT: vpsrlvq %xmm1, %xmm0, %xmm0 +; AVX512-NEXT: retq ; -; AVX512BWVL-LABEL: var_shift_v2i8: -; AVX512BWVL: # %bb.0: -; AVX512BWVL-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 -; AVX512BWVL-NEXT: vpmovzxbw {{.*#+}} 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,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero -; AVX512BWVL-NEXT: vpsrlvw %ymm1, %ymm0, %ymm0 -; AVX512BWVL-NEXT: vpmovwb %ymm0, %xmm0 -; AVX512BWVL-NEXT: vzeroupper -; AVX512BWVL-NEXT: retq +; AVX512VL-LABEL: var_shift_v2i8: +; AVX512VL: # %bb.0: +; AVX512VL-NEXT: vpbroadcastq {{.*#+}} xmm2 = [255,255] +; AVX512VL-NEXT: vpand %xmm2, %xmm1, %xmm1 +; AVX512VL-NEXT: vpand %xmm2, %xmm0, %xmm0 +; AVX512VL-NEXT: vpsrlvq %xmm1, %xmm0, %xmm0 +; AVX512VL-NEXT: retq ; ; X32-SSE-LABEL: var_shift_v2i8: ; X32-SSE: # %bb.0: -; X32-SSE-NEXT: psllw $5, %xmm1 -; X32-SSE-NEXT: pxor %xmm2, %xmm2 -; X32-SSE-NEXT: pxor %xmm3, %xmm3 -; X32-SSE-NEXT: pcmpgtb %xmm1, %xmm3 -; X32-SSE-NEXT: movdqa %xmm3, %xmm4 -; X32-SSE-NEXT: pandn %xmm0, %xmm4 -; X32-SSE-NEXT: psrlw $4, %xmm0 -; X32-SSE-NEXT: pand %xmm3, %xmm0 -; X32-SSE-NEXT: pand {{\.LCPI.*}}, %xmm0 -; X32-SSE-NEXT: por %xmm4, %xmm0 -; X32-SSE-NEXT: paddb %xmm1, %xmm1 -; X32-SSE-NEXT: pxor %xmm3, %xmm3 -; X32-SSE-NEXT: pcmpgtb %xmm1, %xmm3 -; X32-SSE-NEXT: movdqa %xmm3, %xmm4 -; X32-SSE-NEXT: pandn %xmm0, %xmm4 -; X32-SSE-NEXT: psrlw $2, %xmm0 -; X32-SSE-NEXT: pand %xmm3, %xmm0 -; X32-SSE-NEXT: pand {{\.LCPI.*}}, %xmm0 -; X32-SSE-NEXT: por %xmm4, %xmm0 -; X32-SSE-NEXT: paddb %xmm1, %xmm1 -; X32-SSE-NEXT: pcmpgtb %xmm1, %xmm2 -; X32-SSE-NEXT: movdqa %xmm2, %xmm1 -; X32-SSE-NEXT: pandn %xmm0, %xmm1 -; X32-SSE-NEXT: psrlw $1, %xmm0 +; X32-SSE-NEXT: movdqa {{.*#+}} xmm2 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0] +; X32-SSE-NEXT: pand %xmm2, %xmm1 ; X32-SSE-NEXT: pand %xmm2, %xmm0 -; X32-SSE-NEXT: pand {{\.LCPI.*}}, %xmm0 -; X32-SSE-NEXT: por %xmm1, %xmm0 +; X32-SSE-NEXT: movdqa %xmm0, %xmm2 +; X32-SSE-NEXT: psrlq %xmm1, %xmm2 +; X32-SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1] +; X32-SSE-NEXT: psrlq %xmm1, %xmm0 +; X32-SSE-NEXT: movsd {{.*#+}} xmm0 = xmm2[0],xmm0[1] ; X32-SSE-NEXT: retl %shift = lshr <2 x i8> %a, %b ret <2 x i8> %shift @@ -928,46 +771,99 @@ define <2 x i8> @var_shift_v2i8(<2 x i8> %a, <2 x i8> %b) nounwind { define <2 x i32> @splatvar_shift_v2i32(<2 x i32> %a, <2 x i32> %b) nounwind { ; SSE2-LABEL: splatvar_shift_v2i32: ; SSE2: # %bb.0: -; SSE2-NEXT: xorps %xmm2, %xmm2 -; SSE2-NEXT: movss {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3] -; SSE2-NEXT: psrld %xmm2, %xmm0 +; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [4294967295,0,4294967295,0] +; SSE2-NEXT: pand %xmm2, %xmm0 +; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,1,0,1] +; SSE2-NEXT: pand %xmm2, %xmm1 +; SSE2-NEXT: movdqa %xmm0, %xmm2 +; SSE2-NEXT: psrlq %xmm1, %xmm2 +; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1] +; SSE2-NEXT: psrlq %xmm1, %xmm0 +; SSE2-NEXT: movsd {{.*#+}} xmm0 = xmm2[0],xmm0[1] ; SSE2-NEXT: retq ; ; SSE41-LABEL: splatvar_shift_v2i32: ; SSE41: # %bb.0: -; SSE41-NEXT: pmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero -; SSE41-NEXT: psrld %xmm1, %xmm0 +; SSE41-NEXT: pxor %xmm2, %xmm2 +; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3],xmm0[4,5],xmm2[6,7] +; SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,1,0,1] +; SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7] +; SSE41-NEXT: movdqa %xmm0, %xmm2 +; SSE41-NEXT: psrlq %xmm1, %xmm2 +; SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1] +; SSE41-NEXT: psrlq %xmm1, %xmm0 +; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm2[0,1,2,3],xmm0[4,5,6,7] ; SSE41-NEXT: retq ; -; AVX-LABEL: splatvar_shift_v2i32: -; AVX: # %bb.0: -; AVX-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero -; AVX-NEXT: vpsrld %xmm1, %xmm0, %xmm0 -; AVX-NEXT: retq +; AVX1-LABEL: splatvar_shift_v2i32: +; AVX1: # %bb.0: +; AVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3],xmm0[4,5],xmm2[6,7] +; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[0,1,0,1] +; AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7] +; AVX1-NEXT: vpsrlq %xmm1, %xmm0, %xmm2 +; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1] +; AVX1-NEXT: vpsrlq %xmm1, %xmm0, %xmm0 +; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm2[0,1,2,3],xmm0[4,5,6,7] +; AVX1-NEXT: retq ; -; XOP-LABEL: splatvar_shift_v2i32: -; XOP: # %bb.0: -; XOP-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero -; XOP-NEXT: vpsrld %xmm1, %xmm0, %xmm0 -; XOP-NEXT: retq +; AVX2-LABEL: splatvar_shift_v2i32: +; AVX2: # %bb.0: +; AVX2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3] +; AVX2-NEXT: vpbroadcastq %xmm1, %xmm1 +; AVX2-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] +; AVX2-NEXT: vpsrlvq %xmm1, %xmm0, %xmm0 +; AVX2-NEXT: retq +; +; XOPAVX1-LABEL: splatvar_shift_v2i32: +; XOPAVX1: # %bb.0: +; XOPAVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; XOPAVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3],xmm0[4,5],xmm2[6,7] +; XOPAVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[0,1,0,1] +; XOPAVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7] +; XOPAVX1-NEXT: vpsubq %xmm1, %xmm2, %xmm1 +; XOPAVX1-NEXT: vpshlq %xmm1, %xmm0, %xmm0 +; XOPAVX1-NEXT: retq +; +; XOPAVX2-LABEL: splatvar_shift_v2i32: +; XOPAVX2: # %bb.0: +; XOPAVX2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; XOPAVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3] +; XOPAVX2-NEXT: vpbroadcastq %xmm1, %xmm1 +; XOPAVX2-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] +; XOPAVX2-NEXT: vpsrlvq %xmm1, %xmm0, %xmm0 +; XOPAVX2-NEXT: retq ; ; AVX512-LABEL: splatvar_shift_v2i32: ; AVX512: # %bb.0: -; AVX512-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero -; AVX512-NEXT: vpsrld %xmm1, %xmm0, %xmm0 +; AVX512-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3] +; AVX512-NEXT: vpbroadcastq %xmm1, %xmm1 +; AVX512-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] +; AVX512-NEXT: vpsrlvq %xmm1, %xmm0, %xmm0 ; AVX512-NEXT: retq ; ; AVX512VL-LABEL: splatvar_shift_v2i32: ; AVX512VL: # %bb.0: -; AVX512VL-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero -; AVX512VL-NEXT: vpsrld %xmm1, %xmm0, %xmm0 +; AVX512VL-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512VL-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3] +; AVX512VL-NEXT: vpbroadcastq %xmm1, %xmm1 +; AVX512VL-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] +; AVX512VL-NEXT: vpsrlvq %xmm1, %xmm0, %xmm0 ; AVX512VL-NEXT: retq ; ; X32-SSE-LABEL: splatvar_shift_v2i32: ; X32-SSE: # %bb.0: -; X32-SSE-NEXT: xorps %xmm2, %xmm2 +; X32-SSE-NEXT: movdqa {{.*#+}} xmm2 = [4294967295,0,4294967295,0] +; X32-SSE-NEXT: pand %xmm2, %xmm0 +; X32-SSE-NEXT: pand %xmm1, %xmm2 +; X32-SSE-NEXT: movdqa %xmm0, %xmm3 +; X32-SSE-NEXT: psrlq %xmm2, %xmm3 +; X32-SSE-NEXT: pxor %xmm2, %xmm2 ; X32-SSE-NEXT: movss {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3] -; X32-SSE-NEXT: psrld %xmm2, %xmm0 +; X32-SSE-NEXT: psrlq %xmm2, %xmm0 +; X32-SSE-NEXT: movsd {{.*#+}} xmm0 = xmm3[0],xmm0[1] ; X32-SSE-NEXT: retl %splat = shufflevector <2 x i32> %b, <2 x i32> undef, <2 x i32> zeroinitializer %shift = lshr <2 x i32> %a, %splat @@ -977,46 +873,139 @@ define <2 x i32> @splatvar_shift_v2i32(<2 x i32> %a, <2 x i32> %b) nounwind { define <4 x i16> @splatvar_shift_v4i16(<4 x i16> %a, <4 x i16> %b) nounwind { ; SSE2-LABEL: splatvar_shift_v4i16: ; SSE2: # %bb.0: -; SSE2-NEXT: pslldq {{.*#+}} xmm1 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0,1] -; SSE2-NEXT: psrldq {{.*#+}} xmm1 = xmm1[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero -; SSE2-NEXT: psrlw %xmm1, %xmm0 +; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [65535,0,65535,0,65535,0,65535,0] +; SSE2-NEXT: pand %xmm2, %xmm0 +; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm1[0,0,0,0] +; SSE2-NEXT: pand %xmm2, %xmm3 +; SSE2-NEXT: pshuflw {{.*#+}} xmm1 = xmm3[2,3,3,3,4,5,6,7] +; SSE2-NEXT: movdqa %xmm0, %xmm2 +; SSE2-NEXT: psrld %xmm1, %xmm2 +; SSE2-NEXT: pshuflw {{.*#+}} xmm4 = xmm3[0,1,1,1,4,5,6,7] +; SSE2-NEXT: movdqa %xmm0, %xmm1 +; SSE2-NEXT: psrld %xmm4, %xmm1 +; SSE2-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0] +; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm3[2,3,0,1] +; SSE2-NEXT: pshuflw {{.*#+}} xmm3 = xmm2[2,3,3,3,4,5,6,7] +; SSE2-NEXT: movdqa %xmm0, %xmm4 +; SSE2-NEXT: psrld %xmm3, %xmm4 +; SSE2-NEXT: pshuflw {{.*#+}} xmm2 = xmm2[0,1,1,1,4,5,6,7] +; SSE2-NEXT: psrld %xmm2, %xmm0 +; SSE2-NEXT: punpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm4[1] +; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,3],xmm0[0,3] +; SSE2-NEXT: movaps %xmm1, %xmm0 ; SSE2-NEXT: retq ; ; SSE41-LABEL: splatvar_shift_v4i16: ; SSE41: # %bb.0: -; SSE41-NEXT: pmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero -; SSE41-NEXT: psrlw %xmm1, %xmm0 +; SSE41-NEXT: pxor %xmm2, %xmm2 +; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3],xmm0[4],xmm2[5],xmm0[6],xmm2[7] +; SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0] +; SSE41-NEXT: movdqa %xmm1, %xmm3 +; SSE41-NEXT: pblendw {{.*#+}} xmm3 = xmm3[0],xmm2[1],xmm3[2],xmm2[3],xmm3[4],xmm2[5],xmm3[6],xmm2[7] +; SSE41-NEXT: pshuflw {{.*#+}} xmm4 = xmm3[2,3,3,3,4,5,6,7] +; SSE41-NEXT: movdqa %xmm0, %xmm5 +; SSE41-NEXT: psrld %xmm4, %xmm5 +; SSE41-NEXT: pshufd {{.*#+}} xmm3 = xmm3[2,3,0,1] +; SSE41-NEXT: pshuflw {{.*#+}} xmm4 = xmm3[2,3,3,3,4,5,6,7] +; SSE41-NEXT: movdqa %xmm0, %xmm6 +; SSE41-NEXT: psrld %xmm4, %xmm6 +; SSE41-NEXT: pblendw {{.*#+}} xmm6 = xmm5[0,1,2,3],xmm6[4,5,6,7] +; SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5],xmm1[6],xmm2[7] +; SSE41-NEXT: movdqa %xmm0, %xmm2 +; SSE41-NEXT: psrld %xmm1, %xmm2 +; SSE41-NEXT: pshuflw {{.*#+}} xmm1 = xmm3[0,1,1,1,4,5,6,7] +; SSE41-NEXT: psrld %xmm1, %xmm0 +; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm2[0,1,2,3],xmm0[4,5,6,7] +; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm6[2,3],xmm0[4,5],xmm6[6,7] ; SSE41-NEXT: retq ; -; AVX-LABEL: splatvar_shift_v4i16: -; AVX: # %bb.0: -; AVX-NEXT: vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero -; AVX-NEXT: vpsrlw %xmm1, %xmm0, %xmm0 -; AVX-NEXT: retq +; AVX1-LABEL: splatvar_shift_v4i16: +; AVX1: # %bb.0: +; AVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3],xmm0[4],xmm2[5],xmm0[6],xmm2[7] +; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0] +; AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7] +; AVX1-NEXT: vpsrldq {{.*#+}} xmm3 = xmm1[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero +; AVX1-NEXT: vpsrld %xmm3, %xmm0, %xmm3 +; AVX1-NEXT: vpsrlq $32, %xmm1, %xmm4 +; AVX1-NEXT: vpsrld %xmm4, %xmm0, %xmm4 +; AVX1-NEXT: vpblendw {{.*#+}} xmm3 = xmm4[0,1,2,3],xmm3[4,5,6,7] +; AVX1-NEXT: vpunpckhdq {{.*#+}} xmm2 = xmm1[2],xmm2[2],xmm1[3],xmm2[3] +; AVX1-NEXT: vpsrld %xmm2, %xmm0, %xmm2 +; AVX1-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero +; AVX1-NEXT: vpsrld %xmm1, %xmm0, %xmm0 +; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm2[4,5,6,7] +; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm3[2,3],xmm0[4,5],xmm3[6,7] +; AVX1-NEXT: retq ; -; XOP-LABEL: splatvar_shift_v4i16: -; XOP: # %bb.0: -; XOP-NEXT: vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero -; XOP-NEXT: vpsrlw %xmm1, %xmm0, %xmm0 -; XOP-NEXT: retq +; AVX2-LABEL: splatvar_shift_v4i16: +; AVX2: # %bb.0: +; AVX2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX2-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3],xmm0[4],xmm2[5],xmm0[6],xmm2[7] +; AVX2-NEXT: vpbroadcastd %xmm1, %xmm1 +; AVX2-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7] +; AVX2-NEXT: vpsrlvd %xmm1, %xmm0, %xmm0 +; AVX2-NEXT: retq +; +; XOPAVX1-LABEL: splatvar_shift_v4i16: +; XOPAVX1: # %bb.0: +; XOPAVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; XOPAVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3],xmm0[4],xmm2[5],xmm0[6],xmm2[7] +; XOPAVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0] +; XOPAVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7] +; XOPAVX1-NEXT: vpsubd %xmm1, %xmm2, %xmm1 +; XOPAVX1-NEXT: vpshld %xmm1, %xmm0, %xmm0 +; XOPAVX1-NEXT: retq +; +; XOPAVX2-LABEL: splatvar_shift_v4i16: +; XOPAVX2: # %bb.0: +; XOPAVX2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; XOPAVX2-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3],xmm0[4],xmm2[5],xmm0[6],xmm2[7] +; XOPAVX2-NEXT: vpbroadcastd %xmm1, %xmm1 +; XOPAVX2-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7] +; XOPAVX2-NEXT: vpsrlvd %xmm1, %xmm0, %xmm0 +; XOPAVX2-NEXT: retq ; ; AVX512-LABEL: splatvar_shift_v4i16: ; AVX512: # %bb.0: -; AVX512-NEXT: vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero -; AVX512-NEXT: vpsrlw %xmm1, %xmm0, %xmm0 +; AVX512-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3],xmm0[4],xmm2[5],xmm0[6],xmm2[7] +; AVX512-NEXT: vpbroadcastd %xmm1, %xmm1 +; AVX512-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7] +; AVX512-NEXT: vpsrlvd %xmm1, %xmm0, %xmm0 ; AVX512-NEXT: retq ; ; AVX512VL-LABEL: splatvar_shift_v4i16: ; AVX512VL: # %bb.0: -; AVX512VL-NEXT: vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero -; AVX512VL-NEXT: vpsrlw %xmm1, %xmm0, %xmm0 +; AVX512VL-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512VL-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3],xmm0[4],xmm2[5],xmm0[6],xmm2[7] +; AVX512VL-NEXT: vpbroadcastd %xmm1, %xmm1 +; AVX512VL-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7] +; AVX512VL-NEXT: vpsrlvd %xmm1, %xmm0, %xmm0 ; AVX512VL-NEXT: retq ; ; X32-SSE-LABEL: splatvar_shift_v4i16: ; X32-SSE: # %bb.0: -; X32-SSE-NEXT: pslldq {{.*#+}} xmm1 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0,1] -; X32-SSE-NEXT: psrldq {{.*#+}} xmm1 = xmm1[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero -; X32-SSE-NEXT: psrlw %xmm1, %xmm0 +; X32-SSE-NEXT: movdqa {{.*#+}} xmm2 = [65535,0,65535,0,65535,0,65535,0] +; X32-SSE-NEXT: pand %xmm2, %xmm0 +; X32-SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm1[0,0,0,0] +; X32-SSE-NEXT: pand %xmm2, %xmm3 +; X32-SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm3[2,3,3,3,4,5,6,7] +; X32-SSE-NEXT: movdqa %xmm0, %xmm2 +; X32-SSE-NEXT: psrld %xmm1, %xmm2 +; X32-SSE-NEXT: pshuflw {{.*#+}} xmm4 = xmm3[0,1,1,1,4,5,6,7] +; X32-SSE-NEXT: movdqa %xmm0, %xmm1 +; X32-SSE-NEXT: psrld %xmm4, %xmm1 +; X32-SSE-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0] +; X32-SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm3[2,3,0,1] +; X32-SSE-NEXT: pshuflw {{.*#+}} xmm3 = xmm2[2,3,3,3,4,5,6,7] +; X32-SSE-NEXT: movdqa %xmm0, %xmm4 +; X32-SSE-NEXT: psrld %xmm3, %xmm4 +; X32-SSE-NEXT: pshuflw {{.*#+}} xmm2 = xmm2[0,1,1,1,4,5,6,7] +; X32-SSE-NEXT: psrld %xmm2, %xmm0 +; X32-SSE-NEXT: punpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm4[1] +; X32-SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,3],xmm0[0,3] +; X32-SSE-NEXT: movaps %xmm1, %xmm0 ; X32-SSE-NEXT: retl %splat = shufflevector <4 x i16> %b, <4 x i16> undef, <4 x i32> zeroinitializer %shift = lshr <4 x i16> %a, %splat @@ -1026,46 +1015,99 @@ define <4 x i16> @splatvar_shift_v4i16(<4 x i16> %a, <4 x i16> %b) nounwind { define <2 x i16> @splatvar_shift_v2i16(<2 x i16> %a, <2 x i16> %b) nounwind { ; SSE2-LABEL: splatvar_shift_v2i16: ; SSE2: # %bb.0: -; SSE2-NEXT: pslldq {{.*#+}} xmm1 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0,1] -; SSE2-NEXT: psrldq {{.*#+}} xmm1 = xmm1[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero -; SSE2-NEXT: psrlw %xmm1, %xmm0 +; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [65535,0,0,0,65535,0,0,0] +; SSE2-NEXT: pand %xmm2, %xmm0 +; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,1,0,1] +; SSE2-NEXT: pand %xmm2, %xmm1 +; SSE2-NEXT: movdqa %xmm0, %xmm2 +; SSE2-NEXT: psrlq %xmm1, %xmm2 +; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1] +; SSE2-NEXT: psrlq %xmm1, %xmm0 +; SSE2-NEXT: movsd {{.*#+}} xmm0 = xmm2[0],xmm0[1] ; SSE2-NEXT: retq ; ; SSE41-LABEL: splatvar_shift_v2i16: ; SSE41: # %bb.0: -; SSE41-NEXT: pmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero -; SSE41-NEXT: psrlw %xmm1, %xmm0 +; SSE41-NEXT: pxor %xmm2, %xmm2 +; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1,2,3],xmm0[4],xmm2[5,6,7] +; SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,1,0,1] +; SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7] +; SSE41-NEXT: movdqa %xmm0, %xmm2 +; SSE41-NEXT: psrlq %xmm1, %xmm2 +; SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1] +; SSE41-NEXT: psrlq %xmm1, %xmm0 +; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm2[0,1,2,3],xmm0[4,5,6,7] ; SSE41-NEXT: retq ; -; AVX-LABEL: splatvar_shift_v2i16: -; AVX: # %bb.0: -; AVX-NEXT: vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero -; AVX-NEXT: vpsrlw %xmm1, %xmm0, %xmm0 -; AVX-NEXT: retq +; AVX1-LABEL: splatvar_shift_v2i16: +; AVX1: # %bb.0: +; AVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1,2,3],xmm0[4],xmm2[5,6,7] +; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[0,1,0,1] +; AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7] +; AVX1-NEXT: vpsrlq %xmm1, %xmm0, %xmm2 +; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1] +; AVX1-NEXT: vpsrlq %xmm1, %xmm0, %xmm0 +; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm2[0,1,2,3],xmm0[4,5,6,7] +; AVX1-NEXT: retq ; -; XOP-LABEL: splatvar_shift_v2i16: -; XOP: # %bb.0: -; XOP-NEXT: vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero -; XOP-NEXT: vpsrlw %xmm1, %xmm0, %xmm0 -; XOP-NEXT: retq +; AVX2-LABEL: splatvar_shift_v2i16: +; AVX2: # %bb.0: +; AVX2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX2-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1,2,3],xmm0[4],xmm2[5,6,7] +; AVX2-NEXT: vpbroadcastq %xmm1, %xmm1 +; AVX2-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7] +; AVX2-NEXT: vpsrlvq %xmm1, %xmm0, %xmm0 +; AVX2-NEXT: retq +; +; XOPAVX1-LABEL: splatvar_shift_v2i16: +; XOPAVX1: # %bb.0: +; XOPAVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; XOPAVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1,2,3],xmm0[4],xmm2[5,6,7] +; XOPAVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[0,1,0,1] +; XOPAVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7] +; XOPAVX1-NEXT: vpsubq %xmm1, %xmm2, %xmm1 +; XOPAVX1-NEXT: vpshlq %xmm1, %xmm0, %xmm0 +; XOPAVX1-NEXT: retq +; +; XOPAVX2-LABEL: splatvar_shift_v2i16: +; XOPAVX2: # %bb.0: +; XOPAVX2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; XOPAVX2-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1,2,3],xmm0[4],xmm2[5,6,7] +; XOPAVX2-NEXT: vpbroadcastq %xmm1, %xmm1 +; XOPAVX2-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7] +; XOPAVX2-NEXT: vpsrlvq %xmm1, %xmm0, %xmm0 +; XOPAVX2-NEXT: retq ; ; AVX512-LABEL: splatvar_shift_v2i16: ; AVX512: # %bb.0: -; AVX512-NEXT: vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero -; AVX512-NEXT: vpsrlw %xmm1, %xmm0, %xmm0 +; AVX512-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1,2,3],xmm0[4],xmm2[5,6,7] +; AVX512-NEXT: vpbroadcastq %xmm1, %xmm1 +; AVX512-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7] +; AVX512-NEXT: vpsrlvq %xmm1, %xmm0, %xmm0 ; AVX512-NEXT: retq ; ; AVX512VL-LABEL: splatvar_shift_v2i16: ; AVX512VL: # %bb.0: -; AVX512VL-NEXT: vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero -; AVX512VL-NEXT: vpsrlw %xmm1, %xmm0, %xmm0 +; AVX512VL-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512VL-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1,2,3],xmm0[4],xmm2[5,6,7] +; AVX512VL-NEXT: vpbroadcastq %xmm1, %xmm1 +; AVX512VL-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7] +; AVX512VL-NEXT: vpsrlvq %xmm1, %xmm0, %xmm0 ; AVX512VL-NEXT: retq ; ; X32-SSE-LABEL: splatvar_shift_v2i16: ; X32-SSE: # %bb.0: -; X32-SSE-NEXT: pslldq {{.*#+}} xmm1 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0,1] -; X32-SSE-NEXT: psrldq {{.*#+}} xmm1 = xmm1[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero -; X32-SSE-NEXT: psrlw %xmm1, %xmm0 +; X32-SSE-NEXT: movdqa {{.*#+}} xmm2 = [65535,0,0,0,65535,0,0,0] +; X32-SSE-NEXT: pand %xmm2, %xmm0 +; X32-SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,1,0,1] +; X32-SSE-NEXT: pand %xmm2, %xmm1 +; X32-SSE-NEXT: movdqa %xmm0, %xmm2 +; X32-SSE-NEXT: psrlq %xmm1, %xmm2 +; X32-SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1] +; X32-SSE-NEXT: psrlq %xmm1, %xmm0 +; X32-SSE-NEXT: movsd {{.*#+}} xmm0 = xmm2[0],xmm0[1] ; X32-SSE-NEXT: retl %splat = shufflevector <2 x i16> %b, <2 x i16> undef, <2 x i32> zeroinitializer %shift = lshr <2 x i16> %a, %splat @@ -1075,119 +1117,200 @@ define <2 x i16> @splatvar_shift_v2i16(<2 x i16> %a, <2 x i16> %b) nounwind { define <8 x i8> @splatvar_shift_v8i8(<8 x i8> %a, <8 x i8> %b) nounwind { ; SSE2-LABEL: splatvar_shift_v8i8: ; SSE2: # %bb.0: -; SSE2-NEXT: pslldq {{.*#+}} xmm1 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0] -; SSE2-NEXT: psrldq {{.*#+}} xmm1 = xmm1[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero -; SSE2-NEXT: psrlw %xmm1, %xmm0 -; SSE2-NEXT: pcmpeqd %xmm2, %xmm2 -; SSE2-NEXT: psrlw %xmm1, %xmm2 -; SSE2-NEXT: psrlw $8, %xmm2 -; SSE2-NEXT: punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7] -; SSE2-NEXT: pshuflw {{.*#+}} xmm1 = xmm2[0,0,2,3,4,5,6,7] -; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0] +; SSE2-NEXT: movdqa %xmm0, %xmm2 +; SSE2-NEXT: pand {{.*}}(%rip), %xmm2 +; SSE2-NEXT: pshuflw {{.*#+}} xmm0 = xmm1[0,0,2,3,4,5,6,7] +; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[0,0,0,0] +; SSE2-NEXT: psllw $12, %xmm1 +; SSE2-NEXT: movdqa %xmm1, %xmm0 +; SSE2-NEXT: psraw $15, %xmm0 +; SSE2-NEXT: pandn %xmm2, %xmm0 +; SSE2-NEXT: paddw %xmm1, %xmm1 +; SSE2-NEXT: movdqa %xmm1, %xmm2 +; SSE2-NEXT: psraw $15, %xmm2 +; SSE2-NEXT: movdqa %xmm2, %xmm3 +; SSE2-NEXT: pandn %xmm0, %xmm3 +; SSE2-NEXT: psrlw $4, %xmm0 +; SSE2-NEXT: pand %xmm2, %xmm0 +; SSE2-NEXT: por %xmm3, %xmm0 +; SSE2-NEXT: paddw %xmm1, %xmm1 +; SSE2-NEXT: movdqa %xmm1, %xmm2 +; SSE2-NEXT: psraw $15, %xmm2 +; SSE2-NEXT: movdqa %xmm2, %xmm3 +; SSE2-NEXT: pandn %xmm0, %xmm3 +; SSE2-NEXT: psrlw $2, %xmm0 +; SSE2-NEXT: pand %xmm2, %xmm0 +; SSE2-NEXT: por %xmm3, %xmm0 +; SSE2-NEXT: paddw %xmm1, %xmm1 +; SSE2-NEXT: psraw $15, %xmm1 +; SSE2-NEXT: movdqa %xmm1, %xmm2 +; SSE2-NEXT: pandn %xmm0, %xmm2 +; SSE2-NEXT: psrlw $1, %xmm0 ; SSE2-NEXT: pand %xmm1, %xmm0 +; SSE2-NEXT: por %xmm2, %xmm0 ; SSE2-NEXT: retq ; ; SSE41-LABEL: splatvar_shift_v8i8: ; SSE41: # %bb.0: -; SSE41-NEXT: pmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero -; SSE41-NEXT: psrlw %xmm1, %xmm0 -; SSE41-NEXT: pcmpeqd %xmm2, %xmm2 -; SSE41-NEXT: psrlw %xmm1, %xmm2 -; SSE41-NEXT: pshufb {{.*#+}} xmm2 = xmm2[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1] -; SSE41-NEXT: pand %xmm2, %xmm0 +; SSE41-NEXT: movdqa %xmm1, %xmm2 +; SSE41-NEXT: movdqa %xmm0, %xmm1 +; SSE41-NEXT: pand {{.*}}(%rip), %xmm1 +; SSE41-NEXT: pshufb {{.*#+}} xmm2 = xmm2[0],zero,xmm2[0],zero,xmm2[0],zero,xmm2[0],zero,xmm2[0],zero,xmm2[0],zero,xmm2[0],zero,xmm2[0],zero +; SSE41-NEXT: movdqa %xmm2, %xmm0 +; SSE41-NEXT: psllw $12, %xmm0 +; SSE41-NEXT: psllw $4, %xmm2 +; SSE41-NEXT: por %xmm0, %xmm2 +; SSE41-NEXT: movdqa %xmm2, %xmm3 +; SSE41-NEXT: paddw %xmm2, %xmm3 +; SSE41-NEXT: pxor %xmm4, %xmm4 +; SSE41-NEXT: movdqa %xmm2, %xmm0 +; SSE41-NEXT: pblendvb %xmm0, %xmm4, %xmm1 +; SSE41-NEXT: movdqa %xmm1, %xmm2 +; SSE41-NEXT: psrlw $4, %xmm2 +; SSE41-NEXT: movdqa %xmm3, %xmm0 +; SSE41-NEXT: pblendvb %xmm0, %xmm2, %xmm1 +; SSE41-NEXT: movdqa %xmm1, %xmm2 +; SSE41-NEXT: psrlw $2, %xmm2 +; SSE41-NEXT: paddw %xmm3, %xmm3 +; SSE41-NEXT: movdqa %xmm3, %xmm0 +; SSE41-NEXT: pblendvb %xmm0, %xmm2, %xmm1 +; SSE41-NEXT: movdqa %xmm1, %xmm2 +; SSE41-NEXT: psrlw $1, %xmm2 +; SSE41-NEXT: paddw %xmm3, %xmm3 +; SSE41-NEXT: movdqa %xmm3, %xmm0 +; SSE41-NEXT: pblendvb %xmm0, %xmm2, %xmm1 +; SSE41-NEXT: movdqa %xmm1, %xmm0 ; SSE41-NEXT: retq ; ; AVX1-LABEL: splatvar_shift_v8i8: ; AVX1: # %bb.0: -; AVX1-NEXT: vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero -; AVX1-NEXT: vpsrlw %xmm1, %xmm0, %xmm0 -; AVX1-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2 -; AVX1-NEXT: vpsrlw %xmm1, %xmm2, %xmm1 -; AVX1-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1] -; AVX1-NEXT: vpand %xmm1, %xmm0, %xmm0 +; AVX1-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; AVX1-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero +; AVX1-NEXT: vpsllw $12, %xmm1, %xmm2 +; AVX1-NEXT: vpsllw $4, %xmm1, %xmm1 +; AVX1-NEXT: vpor %xmm2, %xmm1, %xmm1 +; AVX1-NEXT: vpaddw %xmm1, %xmm1, %xmm2 +; AVX1-NEXT: vpxor %xmm3, %xmm3, %xmm3 +; AVX1-NEXT: vpblendvb %xmm1, %xmm3, %xmm0, %xmm0 +; AVX1-NEXT: vpsrlw $4, %xmm0, %xmm1 +; AVX1-NEXT: vpblendvb %xmm2, %xmm1, %xmm0, %xmm0 +; AVX1-NEXT: vpsrlw $2, %xmm0, %xmm1 +; AVX1-NEXT: vpaddw %xmm2, %xmm2, %xmm2 +; AVX1-NEXT: vpblendvb %xmm2, %xmm1, %xmm0, %xmm0 +; AVX1-NEXT: vpsrlw $1, %xmm0, %xmm1 +; AVX1-NEXT: vpaddw %xmm2, %xmm2, %xmm2 +; AVX1-NEXT: vpblendvb %xmm2, %xmm1, %xmm0, %xmm0 ; AVX1-NEXT: retq ; ; AVX2-LABEL: splatvar_shift_v8i8: ; AVX2: # %bb.0: -; AVX2-NEXT: vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero -; AVX2-NEXT: vpsrlw %xmm1, %xmm0, %xmm0 -; AVX2-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2 -; AVX2-NEXT: vpsrlw %xmm1, %xmm2, %xmm1 -; AVX2-NEXT: vpsrlw $8, %xmm1, %xmm1 -; AVX2-NEXT: vpbroadcastb %xmm1, %xmm1 -; AVX2-NEXT: vpand %xmm1, %xmm0, %xmm0 +; AVX2-NEXT: vpbroadcastw {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255] +; AVX2-NEXT: vpand %xmm2, %xmm0, %xmm0 +; AVX2-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero +; 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 +; 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 %ymm1, %ymm0, %ymm0 +; AVX2-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[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 {{.*#+}} ymm0 = ymm0[0,2,2,3] +; AVX2-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 +; AVX2-NEXT: vzeroupper ; AVX2-NEXT: retq ; ; XOPAVX1-LABEL: splatvar_shift_v8i8: ; XOPAVX1: # %bb.0: -; XOPAVX1-NEXT: vpunpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7] -; XOPAVX1-NEXT: vpshuflw {{.*#+}} xmm1 = xmm1[0,0,0,0,4,5,6,7] +; XOPAVX1-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; XOPAVX1-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero ; XOPAVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; XOPAVX1-NEXT: vpsubb %xmm1, %xmm2, %xmm1 -; XOPAVX1-NEXT: vpshlb %xmm1, %xmm0, %xmm0 +; XOPAVX1-NEXT: vpsubw %xmm1, %xmm2, %xmm1 +; XOPAVX1-NEXT: vpshlw %xmm1, %xmm0, %xmm0 ; XOPAVX1-NEXT: retq ; ; XOPAVX2-LABEL: splatvar_shift_v8i8: ; XOPAVX2: # %bb.0: -; XOPAVX2-NEXT: vpbroadcastb %xmm1, %xmm1 +; XOPAVX2-NEXT: vpbroadcastw {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255] +; XOPAVX2-NEXT: vpand %xmm2, %xmm0, %xmm0 +; XOPAVX2-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero ; XOPAVX2-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; XOPAVX2-NEXT: vpsubb %xmm1, %xmm2, %xmm1 -; XOPAVX2-NEXT: vpshlb %xmm1, %xmm0, %xmm0 +; XOPAVX2-NEXT: vpsubw %xmm1, %xmm2, %xmm1 +; XOPAVX2-NEXT: vpshlw %xmm1, %xmm0, %xmm0 ; XOPAVX2-NEXT: retq ; ; AVX512DQ-LABEL: splatvar_shift_v8i8: ; AVX512DQ: # %bb.0: -; AVX512DQ-NEXT: vpbroadcastb %xmm1, %xmm1 -; AVX512DQ-NEXT: vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero -; AVX512DQ-NEXT: vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero -; AVX512DQ-NEXT: vpsrlvd %zmm1, %zmm0, %zmm0 -; AVX512DQ-NEXT: vpmovdb %zmm0, %xmm0 +; AVX512DQ-NEXT: vpbroadcastw {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255] +; AVX512DQ-NEXT: vpand %xmm2, %xmm0, %xmm0 +; AVX512DQ-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero +; AVX512DQ-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 +; AVX512DQ-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 +; AVX512DQ-NEXT: vpsrlvd %ymm1, %ymm0, %ymm0 +; AVX512DQ-NEXT: vpmovdw %zmm0, %ymm0 +; AVX512DQ-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 ; AVX512DQ-NEXT: vzeroupper ; AVX512DQ-NEXT: retq ; ; AVX512BW-LABEL: splatvar_shift_v8i8: ; AVX512BW: # %bb.0: -; AVX512BW-NEXT: vpbroadcastb %xmm1, %xmm1 -; AVX512BW-NEXT: vpmovzxbw {{.*#+}} 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,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero -; AVX512BW-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 +; AVX512BW-NEXT: vpbroadcastw {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255] +; AVX512BW-NEXT: vpand %xmm2, %xmm0, %xmm0 +; AVX512BW-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero ; AVX512BW-NEXT: vpsrlvw %zmm1, %zmm0, %zmm0 -; AVX512BW-NEXT: vpmovwb %zmm0, %ymm0 -; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 +; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0 ; AVX512BW-NEXT: vzeroupper ; AVX512BW-NEXT: retq ; ; AVX512DQVL-LABEL: splatvar_shift_v8i8: ; AVX512DQVL: # %bb.0: -; AVX512DQVL-NEXT: vpbroadcastb %xmm1, %xmm1 -; AVX512DQVL-NEXT: vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero -; AVX512DQVL-NEXT: vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero -; AVX512DQVL-NEXT: vpsrlvd %zmm1, %zmm0, %zmm0 -; AVX512DQVL-NEXT: vpmovdb %zmm0, %xmm0 +; AVX512DQVL-NEXT: vpbroadcastw {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255] +; AVX512DQVL-NEXT: vpand %xmm2, %xmm0, %xmm0 +; AVX512DQVL-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero +; AVX512DQVL-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 +; AVX512DQVL-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 +; AVX512DQVL-NEXT: vpsrlvd %ymm1, %ymm0, %ymm0 +; AVX512DQVL-NEXT: vpmovdw %ymm0, %xmm0 ; AVX512DQVL-NEXT: vzeroupper ; AVX512DQVL-NEXT: retq ; ; AVX512BWVL-LABEL: splatvar_shift_v8i8: ; AVX512BWVL: # %bb.0: -; AVX512BWVL-NEXT: vpbroadcastb %xmm1, %xmm1 -; AVX512BWVL-NEXT: vpmovzxbw {{.*#+}} 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,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero -; AVX512BWVL-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 -; AVX512BWVL-NEXT: vpsrlvw %ymm1, %ymm0, %ymm0 -; AVX512BWVL-NEXT: vpmovwb %ymm0, %xmm0 -; AVX512BWVL-NEXT: vzeroupper +; AVX512BWVL-NEXT: vpbroadcastw {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255] +; AVX512BWVL-NEXT: vpand %xmm2, %xmm0, %xmm0 +; AVX512BWVL-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero +; AVX512BWVL-NEXT: vpsrlvw %xmm1, %xmm0, %xmm0 ; AVX512BWVL-NEXT: retq ; ; X32-SSE-LABEL: splatvar_shift_v8i8: ; X32-SSE: # %bb.0: -; X32-SSE-NEXT: pslldq {{.*#+}} xmm1 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0] -; X32-SSE-NEXT: psrldq {{.*#+}} xmm1 = xmm1[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero -; X32-SSE-NEXT: psrlw %xmm1, %xmm0 -; X32-SSE-NEXT: pcmpeqd %xmm2, %xmm2 -; X32-SSE-NEXT: psrlw %xmm1, %xmm2 -; X32-SSE-NEXT: psrlw $8, %xmm2 -; X32-SSE-NEXT: punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7] -; X32-SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm2[0,0,2,3,4,5,6,7] -; X32-SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0] +; X32-SSE-NEXT: movdqa %xmm0, %xmm2 +; X32-SSE-NEXT: pand {{\.LCPI.*}}, %xmm2 +; X32-SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm1[0,0,2,3,4,5,6,7] +; X32-SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm0[0,0,0,0] +; X32-SSE-NEXT: psllw $12, %xmm1 +; X32-SSE-NEXT: movdqa %xmm1, %xmm0 +; X32-SSE-NEXT: psraw $15, %xmm0 +; X32-SSE-NEXT: pandn %xmm2, %xmm0 +; X32-SSE-NEXT: paddw %xmm1, %xmm1 +; X32-SSE-NEXT: movdqa %xmm1, %xmm2 +; X32-SSE-NEXT: psraw $15, %xmm2 +; X32-SSE-NEXT: movdqa %xmm2, %xmm3 +; X32-SSE-NEXT: pandn %xmm0, %xmm3 +; X32-SSE-NEXT: psrlw $4, %xmm0 +; X32-SSE-NEXT: pand %xmm2, %xmm0 +; X32-SSE-NEXT: por %xmm3, %xmm0 +; X32-SSE-NEXT: paddw %xmm1, %xmm1 +; X32-SSE-NEXT: movdqa %xmm1, %xmm2 +; X32-SSE-NEXT: psraw $15, %xmm2 +; X32-SSE-NEXT: movdqa %xmm2, %xmm3 +; X32-SSE-NEXT: pandn %xmm0, %xmm3 +; X32-SSE-NEXT: psrlw $2, %xmm0 +; X32-SSE-NEXT: pand %xmm2, %xmm0 +; X32-SSE-NEXT: por %xmm3, %xmm0 +; X32-SSE-NEXT: paddw %xmm1, %xmm1 +; X32-SSE-NEXT: psraw $15, %xmm1 +; X32-SSE-NEXT: movdqa %xmm1, %xmm2 +; X32-SSE-NEXT: pandn %xmm0, %xmm2 +; X32-SSE-NEXT: psrlw $1, %xmm0 ; X32-SSE-NEXT: pand %xmm1, %xmm0 +; X32-SSE-NEXT: por %xmm2, %xmm0 ; X32-SSE-NEXT: retl %splat = shufflevector <8 x i8> %b, <8 x i8> undef, <8 x i32> zeroinitializer %shift = lshr <8 x i8> %a, %splat @@ -1197,119 +1320,132 @@ define <8 x i8> @splatvar_shift_v8i8(<8 x i8> %a, <8 x i8> %b) nounwind { define <4 x i8> @splatvar_shift_v4i8(<4 x i8> %a, <4 x i8> %b) nounwind { ; SSE2-LABEL: splatvar_shift_v4i8: ; SSE2: # %bb.0: -; SSE2-NEXT: pslldq {{.*#+}} xmm1 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0] -; SSE2-NEXT: psrldq {{.*#+}} xmm1 = xmm1[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero -; SSE2-NEXT: psrlw %xmm1, %xmm0 -; SSE2-NEXT: pcmpeqd %xmm2, %xmm2 -; SSE2-NEXT: psrlw %xmm1, %xmm2 -; SSE2-NEXT: psrlw $8, %xmm2 -; SSE2-NEXT: punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7] -; SSE2-NEXT: pshuflw {{.*#+}} xmm1 = xmm2[0,0,2,3,4,5,6,7] -; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0] -; SSE2-NEXT: pand %xmm1, %xmm0 +; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0] +; SSE2-NEXT: pand %xmm2, %xmm0 +; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm1[0,0,0,0] +; SSE2-NEXT: pand %xmm2, %xmm3 +; SSE2-NEXT: pshuflw {{.*#+}} xmm1 = xmm3[2,3,3,3,4,5,6,7] +; SSE2-NEXT: movdqa %xmm0, %xmm2 +; SSE2-NEXT: psrld %xmm1, %xmm2 +; SSE2-NEXT: pshuflw {{.*#+}} xmm4 = xmm3[0,1,1,1,4,5,6,7] +; SSE2-NEXT: movdqa %xmm0, %xmm1 +; SSE2-NEXT: psrld %xmm4, %xmm1 +; SSE2-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0] +; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm3[2,3,0,1] +; SSE2-NEXT: pshuflw {{.*#+}} xmm3 = xmm2[2,3,3,3,4,5,6,7] +; SSE2-NEXT: movdqa %xmm0, %xmm4 +; SSE2-NEXT: psrld %xmm3, %xmm4 +; SSE2-NEXT: pshuflw {{.*#+}} xmm2 = xmm2[0,1,1,1,4,5,6,7] +; SSE2-NEXT: psrld %xmm2, %xmm0 +; SSE2-NEXT: punpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm4[1] +; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,3],xmm0[0,3] +; SSE2-NEXT: movaps %xmm1, %xmm0 ; SSE2-NEXT: retq ; ; SSE41-LABEL: splatvar_shift_v4i8: ; SSE41: # %bb.0: -; SSE41-NEXT: pmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero -; SSE41-NEXT: psrlw %xmm1, %xmm0 -; SSE41-NEXT: pcmpeqd %xmm2, %xmm2 -; SSE41-NEXT: psrlw %xmm1, %xmm2 -; SSE41-NEXT: pshufb {{.*#+}} xmm2 = xmm2[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1] -; SSE41-NEXT: pand %xmm2, %xmm0 +; SSE41-NEXT: pand {{.*}}(%rip), %xmm0 +; SSE41-NEXT: pshufb {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero +; SSE41-NEXT: pshuflw {{.*#+}} xmm2 = xmm1[2,3,3,3,4,5,6,7] +; SSE41-NEXT: movdqa %xmm0, %xmm3 +; SSE41-NEXT: psrld %xmm2, %xmm3 +; SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm1[2,3,0,1] +; SSE41-NEXT: pshuflw {{.*#+}} xmm4 = xmm2[2,3,3,3,4,5,6,7] +; SSE41-NEXT: movdqa %xmm0, %xmm5 +; SSE41-NEXT: psrld %xmm4, %xmm5 +; SSE41-NEXT: pblendw {{.*#+}} xmm5 = xmm3[0,1,2,3],xmm5[4,5,6,7] +; SSE41-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,1,1,1,4,5,6,7] +; SSE41-NEXT: movdqa %xmm0, %xmm3 +; SSE41-NEXT: psrld %xmm1, %xmm3 +; SSE41-NEXT: pshuflw {{.*#+}} xmm1 = xmm2[0,1,1,1,4,5,6,7] +; SSE41-NEXT: psrld %xmm1, %xmm0 +; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm3[0,1,2,3],xmm0[4,5,6,7] +; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm5[2,3],xmm0[4,5],xmm5[6,7] ; SSE41-NEXT: retq ; ; AVX1-LABEL: splatvar_shift_v4i8: ; AVX1: # %bb.0: -; AVX1-NEXT: vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero -; AVX1-NEXT: vpsrlw %xmm1, %xmm0, %xmm0 -; AVX1-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2 -; AVX1-NEXT: vpsrlw %xmm1, %xmm2, %xmm1 -; AVX1-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1] -; AVX1-NEXT: vpand %xmm1, %xmm0, %xmm0 +; AVX1-NEXT: vbroadcastss {{.*#+}} xmm2 = [3.57331108E-43,3.57331108E-43,3.57331108E-43,3.57331108E-43] +; AVX1-NEXT: vandps %xmm2, %xmm0, %xmm0 +; AVX1-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero +; AVX1-NEXT: vpsrldq {{.*#+}} xmm2 = xmm1[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero +; AVX1-NEXT: vpsrld %xmm2, %xmm0, %xmm2 +; AVX1-NEXT: vpsrlq $32, %xmm1, %xmm3 +; AVX1-NEXT: vpsrld %xmm3, %xmm0, %xmm3 +; AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm3[0,1,2,3],xmm2[4,5,6,7] +; AVX1-NEXT: vpxor %xmm3, %xmm3, %xmm3 +; AVX1-NEXT: vpunpckhdq {{.*#+}} xmm3 = xmm1[2],xmm3[2],xmm1[3],xmm3[3] +; AVX1-NEXT: vpsrld %xmm3, %xmm0, %xmm3 +; AVX1-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero +; AVX1-NEXT: vpsrld %xmm1, %xmm0, %xmm0 +; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm3[4,5,6,7] +; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3],xmm0[4,5],xmm2[6,7] ; AVX1-NEXT: retq ; ; AVX2-LABEL: splatvar_shift_v4i8: ; AVX2: # %bb.0: -; AVX2-NEXT: vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero -; AVX2-NEXT: vpsrlw %xmm1, %xmm0, %xmm0 -; AVX2-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2 -; AVX2-NEXT: vpsrlw %xmm1, %xmm2, %xmm1 -; AVX2-NEXT: vpsrlw $8, %xmm1, %xmm1 -; AVX2-NEXT: vpbroadcastb %xmm1, %xmm1 -; AVX2-NEXT: vpand %xmm1, %xmm0, %xmm0 +; AVX2-NEXT: vpbroadcastd {{.*#+}} xmm2 = [255,255,255,255] +; AVX2-NEXT: vpand %xmm2, %xmm0, %xmm0 +; AVX2-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero +; AVX2-NEXT: vpsrlvd %xmm1, %xmm0, %xmm0 ; AVX2-NEXT: retq ; ; XOPAVX1-LABEL: splatvar_shift_v4i8: ; XOPAVX1: # %bb.0: -; XOPAVX1-NEXT: vpunpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7] -; XOPAVX1-NEXT: vpshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7] -; XOPAVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; XOPAVX1-NEXT: vpsubb %xmm1, %xmm2, %xmm1 -; XOPAVX1-NEXT: vpshlb %xmm1, %xmm0, %xmm0 +; XOPAVX1-NEXT: vbroadcastss {{.*#+}} xmm2 = [3.57331108E-43,3.57331108E-43,3.57331108E-43,3.57331108E-43] +; XOPAVX1-NEXT: vandps %xmm2, %xmm0, %xmm0 +; XOPAVX1-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero +; XOPAVX1-NEXT: vxorps %xmm2, %xmm2, %xmm2 +; XOPAVX1-NEXT: vpsubd %xmm1, %xmm2, %xmm1 +; XOPAVX1-NEXT: vpshld %xmm1, %xmm0, %xmm0 ; XOPAVX1-NEXT: retq ; ; XOPAVX2-LABEL: splatvar_shift_v4i8: ; XOPAVX2: # %bb.0: -; XOPAVX2-NEXT: vpbroadcastb %xmm1, %xmm1 -; XOPAVX2-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; XOPAVX2-NEXT: vpsubb %xmm1, %xmm2, %xmm1 -; XOPAVX2-NEXT: vpshlb %xmm1, %xmm0, %xmm0 +; XOPAVX2-NEXT: vpbroadcastd {{.*#+}} xmm2 = [255,255,255,255] +; XOPAVX2-NEXT: vpand %xmm2, %xmm0, %xmm0 +; XOPAVX2-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero +; XOPAVX2-NEXT: vpsrlvd %xmm1, %xmm0, %xmm0 ; XOPAVX2-NEXT: retq ; -; AVX512DQ-LABEL: splatvar_shift_v4i8: -; AVX512DQ: # %bb.0: -; AVX512DQ-NEXT: vpbroadcastb %xmm1, %xmm1 -; AVX512DQ-NEXT: vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero -; AVX512DQ-NEXT: vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero -; AVX512DQ-NEXT: vpsrlvd %zmm1, %zmm0, %zmm0 -; AVX512DQ-NEXT: vpmovdb %zmm0, %xmm0 -; AVX512DQ-NEXT: vzeroupper -; AVX512DQ-NEXT: retq -; -; AVX512BW-LABEL: splatvar_shift_v4i8: -; AVX512BW: # %bb.0: -; AVX512BW-NEXT: vpbroadcastb %xmm1, %xmm1 -; AVX512BW-NEXT: vpmovzxbw {{.*#+}} 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,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero -; AVX512BW-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 -; AVX512BW-NEXT: vpsrlvw %zmm1, %zmm0, %zmm0 -; AVX512BW-NEXT: vpmovwb %zmm0, %ymm0 -; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 -; AVX512BW-NEXT: vzeroupper -; AVX512BW-NEXT: retq -; -; AVX512DQVL-LABEL: splatvar_shift_v4i8: -; AVX512DQVL: # %bb.0: -; AVX512DQVL-NEXT: vpbroadcastb %xmm1, %xmm1 -; AVX512DQVL-NEXT: vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero -; AVX512DQVL-NEXT: vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero -; AVX512DQVL-NEXT: vpsrlvd %zmm1, %zmm0, %zmm0 -; AVX512DQVL-NEXT: vpmovdb %zmm0, %xmm0 -; AVX512DQVL-NEXT: vzeroupper -; AVX512DQVL-NEXT: retq +; AVX512-LABEL: splatvar_shift_v4i8: +; AVX512: # %bb.0: +; AVX512-NEXT: vpbroadcastd {{.*#+}} xmm2 = [255,255,255,255] +; AVX512-NEXT: vpand %xmm2, %xmm0, %xmm0 +; AVX512-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero +; AVX512-NEXT: vpsrlvd %xmm1, %xmm0, %xmm0 +; AVX512-NEXT: retq ; -; AVX512BWVL-LABEL: splatvar_shift_v4i8: -; AVX512BWVL: # %bb.0: -; AVX512BWVL-NEXT: vpbroadcastb %xmm1, %xmm1 -; AVX512BWVL-NEXT: vpmovzxbw {{.*#+}} 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,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero -; AVX512BWVL-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 -; AVX512BWVL-NEXT: vpsrlvw %ymm1, %ymm0, %ymm0 -; AVX512BWVL-NEXT: vpmovwb %ymm0, %xmm0 -; AVX512BWVL-NEXT: vzeroupper -; AVX512BWVL-NEXT: retq +; AVX512VL-LABEL: splatvar_shift_v4i8: +; AVX512VL: # %bb.0: +; AVX512VL-NEXT: vpbroadcastd {{.*#+}} xmm2 = [255,255,255,255] +; AVX512VL-NEXT: vpand %xmm2, %xmm0, %xmm0 +; AVX512VL-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero +; AVX512VL-NEXT: vpsrlvd %xmm1, %xmm0, %xmm0 +; AVX512VL-NEXT: retq ; ; X32-SSE-LABEL: splatvar_shift_v4i8: ; X32-SSE: # %bb.0: -; X32-SSE-NEXT: pslldq {{.*#+}} xmm1 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0] -; X32-SSE-NEXT: psrldq {{.*#+}} xmm1 = xmm1[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero -; X32-SSE-NEXT: psrlw %xmm1, %xmm0 -; X32-SSE-NEXT: pcmpeqd %xmm2, %xmm2 -; X32-SSE-NEXT: psrlw %xmm1, %xmm2 -; X32-SSE-NEXT: psrlw $8, %xmm2 -; X32-SSE-NEXT: punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7] -; X32-SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm2[0,0,2,3,4,5,6,7] -; X32-SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0] -; X32-SSE-NEXT: pand %xmm1, %xmm0 +; X32-SSE-NEXT: movdqa {{.*#+}} xmm2 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0] +; X32-SSE-NEXT: pand %xmm2, %xmm0 +; X32-SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm1[0,0,0,0] +; X32-SSE-NEXT: pand %xmm2, %xmm3 +; X32-SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm3[2,3,3,3,4,5,6,7] +; X32-SSE-NEXT: movdqa %xmm0, %xmm2 +; X32-SSE-NEXT: psrld %xmm1, %xmm2 +; X32-SSE-NEXT: pshuflw {{.*#+}} xmm4 = xmm3[0,1,1,1,4,5,6,7] +; X32-SSE-NEXT: movdqa %xmm0, %xmm1 +; X32-SSE-NEXT: psrld %xmm4, %xmm1 +; X32-SSE-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0] +; X32-SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm3[2,3,0,1] +; X32-SSE-NEXT: pshuflw {{.*#+}} xmm3 = xmm2[2,3,3,3,4,5,6,7] +; X32-SSE-NEXT: movdqa %xmm0, %xmm4 +; X32-SSE-NEXT: psrld %xmm3, %xmm4 +; X32-SSE-NEXT: pshuflw {{.*#+}} xmm2 = xmm2[0,1,1,1,4,5,6,7] +; X32-SSE-NEXT: psrld %xmm2, %xmm0 +; X32-SSE-NEXT: punpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm4[1] +; X32-SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,3],xmm0[0,3] +; X32-SSE-NEXT: movaps %xmm1, %xmm0 ; X32-SSE-NEXT: retl %splat = shufflevector <4 x i8> %b, <4 x i8> undef, <4 x i32> zeroinitializer %shift = lshr <4 x i8> %a, %splat @@ -1319,110 +1455,94 @@ define <4 x i8> @splatvar_shift_v4i8(<4 x i8> %a, <4 x i8> %b) nounwind { define <2 x i8> @splatvar_shift_v2i8(<2 x i8> %a, <2 x i8> %b) nounwind { ; SSE2-LABEL: splatvar_shift_v2i8: ; SSE2: # %bb.0: -; SSE2-NEXT: pslldq {{.*#+}} xmm1 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0] -; SSE2-NEXT: psrldq {{.*#+}} xmm1 = xmm1[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero -; SSE2-NEXT: psrlw %xmm1, %xmm0 -; SSE2-NEXT: pcmpeqd %xmm2, %xmm2 -; SSE2-NEXT: psrlw %xmm1, %xmm2 -; SSE2-NEXT: psrlw $8, %xmm2 -; SSE2-NEXT: punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7] -; SSE2-NEXT: pshuflw {{.*#+}} xmm1 = xmm2[0,0,2,3,4,5,6,7] -; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0] -; SSE2-NEXT: pand %xmm1, %xmm0 +; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0] +; SSE2-NEXT: pand %xmm2, %xmm0 +; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,1,0,1] +; SSE2-NEXT: pand %xmm2, %xmm1 +; SSE2-NEXT: movdqa %xmm0, %xmm2 +; SSE2-NEXT: psrlq %xmm1, %xmm2 +; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1] +; SSE2-NEXT: psrlq %xmm1, %xmm0 +; SSE2-NEXT: movsd {{.*#+}} xmm0 = xmm2[0],xmm0[1] ; SSE2-NEXT: retq ; ; SSE41-LABEL: splatvar_shift_v2i8: ; SSE41: # %bb.0: -; SSE41-NEXT: pmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero -; SSE41-NEXT: psrlw %xmm1, %xmm0 -; SSE41-NEXT: pcmpeqd %xmm2, %xmm2 -; SSE41-NEXT: psrlw %xmm1, %xmm2 -; SSE41-NEXT: pshufb {{.*#+}} xmm2 = xmm2[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1] -; SSE41-NEXT: pand %xmm2, %xmm0 +; SSE41-NEXT: pand {{.*}}(%rip), %xmm0 +; SSE41-NEXT: pshufb {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[0],zero,zero,zero,zero,zero,zero,zero +; SSE41-NEXT: movdqa %xmm0, %xmm2 +; SSE41-NEXT: psrlq %xmm1, %xmm2 +; SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1] +; SSE41-NEXT: psrlq %xmm1, %xmm0 +; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm2[0,1,2,3],xmm0[4,5,6,7] ; SSE41-NEXT: retq ; ; AVX1-LABEL: splatvar_shift_v2i8: ; AVX1: # %bb.0: -; AVX1-NEXT: vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero -; AVX1-NEXT: vpsrlw %xmm1, %xmm0, %xmm0 -; AVX1-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2 -; AVX1-NEXT: vpsrlw %xmm1, %xmm2, %xmm1 -; AVX1-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1] -; AVX1-NEXT: vpand %xmm1, %xmm0, %xmm0 +; AVX1-NEXT: vmovddup {{.*#+}} xmm2 = [1.2598673968951787E-321,1.2598673968951787E-321] +; AVX1-NEXT: # xmm2 = mem[0,0] +; AVX1-NEXT: vandps %xmm2, %xmm0, %xmm0 +; AVX1-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[0],zero,zero,zero,zero,zero,zero,zero +; AVX1-NEXT: vpsrlq %xmm1, %xmm0, %xmm2 +; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1] +; AVX1-NEXT: vpsrlq %xmm1, %xmm0, %xmm0 +; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm2[0,1,2,3],xmm0[4,5,6,7] ; AVX1-NEXT: retq ; ; AVX2-LABEL: splatvar_shift_v2i8: ; AVX2: # %bb.0: -; AVX2-NEXT: vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero -; AVX2-NEXT: vpsrlw %xmm1, %xmm0, %xmm0 -; AVX2-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2 -; AVX2-NEXT: vpsrlw %xmm1, %xmm2, %xmm1 -; AVX2-NEXT: vpsrlw $8, %xmm1, %xmm1 -; AVX2-NEXT: vpbroadcastb %xmm1, %xmm1 -; AVX2-NEXT: vpand %xmm1, %xmm0, %xmm0 +; AVX2-NEXT: vpbroadcastq {{.*#+}} xmm2 = [255,255] +; AVX2-NEXT: vpand %xmm2, %xmm0, %xmm0 +; AVX2-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[0],zero,zero,zero,zero,zero,zero,zero +; AVX2-NEXT: vpsrlvq %xmm1, %xmm0, %xmm0 ; AVX2-NEXT: retq ; -; XOP-LABEL: splatvar_shift_v2i8: -; XOP: # %bb.0: -; XOP-NEXT: insertq {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7,u,u,u,u,u,u,u,u] -; XOP-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; XOP-NEXT: vpsubb %xmm1, %xmm2, %xmm1 -; XOP-NEXT: vpshlb %xmm1, %xmm0, %xmm0 -; XOP-NEXT: retq -; -; AVX512DQ-LABEL: splatvar_shift_v2i8: -; AVX512DQ: # %bb.0: -; AVX512DQ-NEXT: vpbroadcastb %xmm1, %xmm1 -; AVX512DQ-NEXT: vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero -; AVX512DQ-NEXT: vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero -; AVX512DQ-NEXT: vpsrlvd %zmm1, %zmm0, %zmm0 -; AVX512DQ-NEXT: vpmovdb %zmm0, %xmm0 -; AVX512DQ-NEXT: vzeroupper -; AVX512DQ-NEXT: retq +; XOPAVX1-LABEL: splatvar_shift_v2i8: +; XOPAVX1: # %bb.0: +; XOPAVX1-NEXT: vmovddup {{.*#+}} xmm2 = [1.2598673968951787E-321,1.2598673968951787E-321] +; XOPAVX1-NEXT: # xmm2 = mem[0,0] +; XOPAVX1-NEXT: vandps %xmm2, %xmm0, %xmm0 +; XOPAVX1-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[0],zero,zero,zero,zero,zero,zero,zero +; XOPAVX1-NEXT: vxorps %xmm2, %xmm2, %xmm2 +; XOPAVX1-NEXT: vpsubq %xmm1, %xmm2, %xmm1 +; XOPAVX1-NEXT: vpshlq %xmm1, %xmm0, %xmm0 +; XOPAVX1-NEXT: retq ; -; AVX512BW-LABEL: splatvar_shift_v2i8: -; AVX512BW: # %bb.0: -; AVX512BW-NEXT: vpbroadcastb %xmm1, %xmm1 -; AVX512BW-NEXT: vpmovzxbw {{.*#+}} 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,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero -; AVX512BW-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 -; AVX512BW-NEXT: vpsrlvw %zmm1, %zmm0, %zmm0 -; AVX512BW-NEXT: vpmovwb %zmm0, %ymm0 -; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 -; AVX512BW-NEXT: vzeroupper -; AVX512BW-NEXT: retq +; XOPAVX2-LABEL: splatvar_shift_v2i8: +; XOPAVX2: # %bb.0: +; XOPAVX2-NEXT: vpbroadcastq {{.*#+}} xmm2 = [255,255] +; XOPAVX2-NEXT: vpand %xmm2, %xmm0, %xmm0 +; XOPAVX2-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[0],zero,zero,zero,zero,zero,zero,zero +; XOPAVX2-NEXT: vpsrlvq %xmm1, %xmm0, %xmm0 +; XOPAVX2-NEXT: retq ; -; AVX512DQVL-LABEL: splatvar_shift_v2i8: -; AVX512DQVL: # %bb.0: -; AVX512DQVL-NEXT: vpbroadcastb %xmm1, %xmm1 -; AVX512DQVL-NEXT: vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero -; AVX512DQVL-NEXT: vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero -; AVX512DQVL-NEXT: vpsrlvd %zmm1, %zmm0, %zmm0 -; AVX512DQVL-NEXT: vpmovdb %zmm0, %xmm0 -; AVX512DQVL-NEXT: vzeroupper -; AVX512DQVL-NEXT: retq +; AVX512-LABEL: splatvar_shift_v2i8: +; AVX512: # %bb.0: +; AVX512-NEXT: vpbroadcastq {{.*#+}} xmm2 = [255,255] +; AVX512-NEXT: vpand %xmm2, %xmm0, %xmm0 +; AVX512-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[0],zero,zero,zero,zero,zero,zero,zero +; AVX512-NEXT: vpsrlvq %xmm1, %xmm0, %xmm0 +; AVX512-NEXT: retq ; -; AVX512BWVL-LABEL: splatvar_shift_v2i8: -; AVX512BWVL: # %bb.0: -; AVX512BWVL-NEXT: vpbroadcastb %xmm1, %xmm1 -; AVX512BWVL-NEXT: vpmovzxbw {{.*#+}} 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,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero -; AVX512BWVL-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 -; AVX512BWVL-NEXT: vpsrlvw %ymm1, %ymm0, %ymm0 -; AVX512BWVL-NEXT: vpmovwb %ymm0, %xmm0 -; AVX512BWVL-NEXT: vzeroupper -; AVX512BWVL-NEXT: retq +; AVX512VL-LABEL: splatvar_shift_v2i8: +; AVX512VL: # %bb.0: +; AVX512VL-NEXT: vpbroadcastq {{.*#+}} xmm2 = [255,255] +; AVX512VL-NEXT: vpand %xmm2, %xmm0, %xmm0 +; AVX512VL-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[0],zero,zero,zero,zero,zero,zero,zero +; AVX512VL-NEXT: vpsrlvq %xmm1, %xmm0, %xmm0 +; AVX512VL-NEXT: retq ; ; X32-SSE-LABEL: splatvar_shift_v2i8: ; X32-SSE: # %bb.0: -; X32-SSE-NEXT: pslldq {{.*#+}} xmm1 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0] -; X32-SSE-NEXT: psrldq {{.*#+}} xmm1 = xmm1[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero -; X32-SSE-NEXT: psrlw %xmm1, %xmm0 -; X32-SSE-NEXT: pcmpeqd %xmm2, %xmm2 -; X32-SSE-NEXT: psrlw %xmm1, %xmm2 -; X32-SSE-NEXT: psrlw $8, %xmm2 -; X32-SSE-NEXT: punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7] -; X32-SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm2[0,0,2,3,4,5,6,7] -; X32-SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0] -; X32-SSE-NEXT: pand %xmm1, %xmm0 +; X32-SSE-NEXT: movdqa {{.*#+}} xmm2 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0] +; X32-SSE-NEXT: pand %xmm2, %xmm0 +; X32-SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,1,0,1] +; X32-SSE-NEXT: pand %xmm2, %xmm1 +; X32-SSE-NEXT: movdqa %xmm0, %xmm2 +; X32-SSE-NEXT: psrlq %xmm1, %xmm2 +; X32-SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1] +; X32-SSE-NEXT: psrlq %xmm1, %xmm0 +; X32-SSE-NEXT: movsd {{.*#+}} xmm0 = xmm2[0],xmm0[1] ; X32-SSE-NEXT: retl %splat = shufflevector <2 x i8> %b, <2 x i8> undef, <2 x i32> zeroinitializer %shift = lshr <2 x i8> %a, %splat @@ -1436,62 +1556,75 @@ define <2 x i8> @splatvar_shift_v2i8(<2 x i8> %a, <2 x i8> %b) nounwind { define <2 x i32> @constant_shift_v2i32(<2 x i32> %a) nounwind { ; SSE2-LABEL: constant_shift_v2i32: ; SSE2: # %bb.0: +; SSE2-NEXT: pand {{.*}}(%rip), %xmm0 ; SSE2-NEXT: movdqa %xmm0, %xmm1 -; SSE2-NEXT: psrld $4, %xmm1 -; SSE2-NEXT: psrld $5, %xmm0 -; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3] -; SSE2-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1] -; SSE2-NEXT: movdqa %xmm1, %xmm0 +; SSE2-NEXT: psrlq $4, %xmm1 +; SSE2-NEXT: psrlq $5, %xmm0 +; SSE2-NEXT: movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1] ; SSE2-NEXT: retq ; ; SSE41-LABEL: constant_shift_v2i32: ; SSE41: # %bb.0: -; SSE41-NEXT: movdqa %xmm0, %xmm1 -; SSE41-NEXT: psrld $5, %xmm1 -; SSE41-NEXT: psrld $4, %xmm0 -; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5,6,7] +; SSE41-NEXT: pxor %xmm1, %xmm1 +; SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7] +; SSE41-NEXT: movdqa %xmm1, %xmm0 +; SSE41-NEXT: psrlq $5, %xmm0 +; SSE41-NEXT: psrlq $4, %xmm1 +; SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm0[4,5,6,7] +; SSE41-NEXT: movdqa %xmm1, %xmm0 ; SSE41-NEXT: retq ; ; AVX1-LABEL: constant_shift_v2i32: ; AVX1: # %bb.0: -; AVX1-NEXT: vpsrld $5, %xmm0, %xmm1 -; AVX1-NEXT: vpsrld $4, %xmm0, %xmm0 -; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5,6,7] +; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7] +; AVX1-NEXT: vpsrlq $5, %xmm0, %xmm1 +; AVX1-NEXT: vpsrlq $4, %xmm0, %xmm0 +; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm1[4,5,6,7] ; AVX1-NEXT: retq ; ; AVX2-LABEL: constant_shift_v2i32: ; AVX2: # %bb.0: -; AVX2-NEXT: vpsrlvd {{.*}}(%rip), %xmm0, %xmm0 +; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3] +; AVX2-NEXT: vpsrlvq {{.*}}(%rip), %xmm0, %xmm0 ; AVX2-NEXT: retq ; ; XOPAVX1-LABEL: constant_shift_v2i32: ; XOPAVX1: # %bb.0: -; XOPAVX1-NEXT: vpshld {{.*}}(%rip), %xmm0, %xmm0 +; XOPAVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; XOPAVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7] +; XOPAVX1-NEXT: vpshlq {{.*}}(%rip), %xmm0, %xmm0 ; XOPAVX1-NEXT: retq ; ; XOPAVX2-LABEL: constant_shift_v2i32: ; XOPAVX2: # %bb.0: -; XOPAVX2-NEXT: vpsrlvd {{.*}}(%rip), %xmm0, %xmm0 +; XOPAVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; XOPAVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3] +; XOPAVX2-NEXT: vpsrlvq {{.*}}(%rip), %xmm0, %xmm0 ; XOPAVX2-NEXT: retq ; ; AVX512-LABEL: constant_shift_v2i32: ; AVX512: # %bb.0: -; AVX512-NEXT: vpsrlvd {{.*}}(%rip), %xmm0, %xmm0 +; AVX512-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; AVX512-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3] +; AVX512-NEXT: vpsrlvq {{.*}}(%rip), %xmm0, %xmm0 ; AVX512-NEXT: retq ; ; AVX512VL-LABEL: constant_shift_v2i32: ; AVX512VL: # %bb.0: -; AVX512VL-NEXT: vpsrlvd {{.*}}(%rip), %xmm0, %xmm0 +; AVX512VL-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; AVX512VL-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3] +; AVX512VL-NEXT: vpsrlvq {{.*}}(%rip), %xmm0, %xmm0 ; AVX512VL-NEXT: retq ; ; X32-SSE-LABEL: constant_shift_v2i32: ; X32-SSE: # %bb.0: +; X32-SSE-NEXT: pand {{\.LCPI.*}}, %xmm0 ; X32-SSE-NEXT: movdqa %xmm0, %xmm1 -; X32-SSE-NEXT: psrld $4, %xmm1 -; X32-SSE-NEXT: psrld $5, %xmm0 -; X32-SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3] -; X32-SSE-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1] -; X32-SSE-NEXT: movdqa %xmm1, %xmm0 +; X32-SSE-NEXT: psrlq $4, %xmm1 +; X32-SSE-NEXT: psrlq $5, %xmm0 +; X32-SSE-NEXT: movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1] ; X32-SSE-NEXT: retl %shift = lshr <2 x i32> %a, <i32 4, i32 5> ret <2 x i32> %shift @@ -1500,66 +1633,91 @@ define <2 x i32> @constant_shift_v2i32(<2 x i32> %a) nounwind { define <4 x i16> @constant_shift_v4i16(<4 x i16> %a) nounwind { ; SSE2-LABEL: constant_shift_v4i16: ; SSE2: # %bb.0: -; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [0,65535,65535,65535,65535,65535,65535,65535] -; SSE2-NEXT: movdqa %xmm1, %xmm2 -; SSE2-NEXT: pandn %xmm0, %xmm2 -; SSE2-NEXT: pmulhuw {{.*}}(%rip), %xmm0 -; SSE2-NEXT: pand %xmm1, %xmm0 -; SSE2-NEXT: por %xmm2, %xmm0 +; SSE2-NEXT: pand {{.*}}(%rip), %xmm0 +; SSE2-NEXT: movdqa %xmm0, %xmm1 +; SSE2-NEXT: psrld $3, %xmm1 +; SSE2-NEXT: movdqa %xmm0, %xmm2 +; SSE2-NEXT: psrld $2, %xmm2 +; SSE2-NEXT: punpckhqdq {{.*#+}} xmm2 = xmm2[1],xmm1[1] +; SSE2-NEXT: movdqa %xmm0, %xmm1 +; SSE2-NEXT: psrld $1, %xmm1 +; SSE2-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0] +; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,3],xmm2[0,3] ; SSE2-NEXT: retq ; ; SSE41-LABEL: constant_shift_v4i16: ; SSE41: # %bb.0: -; SSE41-NEXT: movdqa {{.*#+}} xmm1 = <u,32768,16384,8192,u,u,u,u> -; SSE41-NEXT: pmulhuw %xmm0, %xmm1 -; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3,4,5,6,7] +; SSE41-NEXT: pxor %xmm1, %xmm1 +; SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7] +; SSE41-NEXT: movdqa %xmm1, %xmm2 +; SSE41-NEXT: movdqa %xmm1, %xmm0 +; SSE41-NEXT: psrld $2, %xmm0 +; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm0[4,5,6,7] +; SSE41-NEXT: psrld $3, %xmm1 +; SSE41-NEXT: psrld $1, %xmm2 +; SSE41-NEXT: pblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm1[4,5,6,7] +; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3],xmm0[4,5],xmm2[6,7] ; SSE41-NEXT: retq ; -; AVX-LABEL: constant_shift_v4i16: -; AVX: # %bb.0: -; AVX-NEXT: vpmulhuw {{.*}}(%rip), %xmm0, %xmm1 -; AVX-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3,4,5,6,7] -; AVX-NEXT: retq +; AVX1-LABEL: constant_shift_v4i16: +; AVX1: # %bb.0: +; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7] +; AVX1-NEXT: vpsrld $3, %xmm0, %xmm1 +; AVX1-NEXT: vpsrld $1, %xmm0, %xmm2 +; AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0,1,2,3],xmm1[4,5,6,7] +; AVX1-NEXT: vpsrld $2, %xmm0, %xmm2 +; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm2[4,5,6,7] +; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7] +; AVX1-NEXT: retq ; -; XOP-LABEL: constant_shift_v4i16: -; XOP: # %bb.0: -; XOP-NEXT: vpshlw {{.*}}(%rip), %xmm0, %xmm0 -; XOP-NEXT: retq +; AVX2-LABEL: constant_shift_v4i16: +; AVX2: # %bb.0: +; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; AVX2-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7] +; AVX2-NEXT: vpsrlvd {{.*}}(%rip), %xmm0, %xmm0 +; AVX2-NEXT: retq ; -; AVX512DQ-LABEL: constant_shift_v4i16: -; AVX512DQ: # %bb.0: -; AVX512DQ-NEXT: vpmulhuw {{.*}}(%rip), %xmm0, %xmm1 -; AVX512DQ-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3,4,5,6,7] -; AVX512DQ-NEXT: retq +; XOPAVX1-LABEL: constant_shift_v4i16: +; XOPAVX1: # %bb.0: +; XOPAVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; XOPAVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7] +; XOPAVX1-NEXT: vpshld {{.*}}(%rip), %xmm0, %xmm0 +; XOPAVX1-NEXT: retq ; -; AVX512BW-LABEL: constant_shift_v4i16: -; AVX512BW: # %bb.0: -; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0 -; AVX512BW-NEXT: vmovdqa {{.*#+}} xmm1 = <0,1,2,3,u,u,u,u> -; AVX512BW-NEXT: vpsrlvw %zmm1, %zmm0, %zmm0 -; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0 -; AVX512BW-NEXT: vzeroupper -; AVX512BW-NEXT: retq +; XOPAVX2-LABEL: constant_shift_v4i16: +; XOPAVX2: # %bb.0: +; XOPAVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; XOPAVX2-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7] +; XOPAVX2-NEXT: vpsrlvd {{.*}}(%rip), %xmm0, %xmm0 +; XOPAVX2-NEXT: retq ; -; AVX512DQVL-LABEL: constant_shift_v4i16: -; AVX512DQVL: # %bb.0: -; AVX512DQVL-NEXT: vpmulhuw {{.*}}(%rip), %xmm0, %xmm1 -; AVX512DQVL-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3,4,5,6,7] -; AVX512DQVL-NEXT: retq +; AVX512-LABEL: constant_shift_v4i16: +; AVX512: # %bb.0: +; AVX512-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; AVX512-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7] +; AVX512-NEXT: vpsrlvd {{.*}}(%rip), %xmm0, %xmm0 +; AVX512-NEXT: retq ; -; AVX512BWVL-LABEL: constant_shift_v4i16: -; AVX512BWVL: # %bb.0: -; AVX512BWVL-NEXT: vpsrlvw {{.*}}(%rip), %xmm0, %xmm0 -; AVX512BWVL-NEXT: retq +; AVX512VL-LABEL: constant_shift_v4i16: +; AVX512VL: # %bb.0: +; AVX512VL-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; AVX512VL-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7] +; AVX512VL-NEXT: vpsrlvd {{.*}}(%rip), %xmm0, %xmm0 +; AVX512VL-NEXT: retq ; ; X32-SSE-LABEL: constant_shift_v4i16: ; X32-SSE: # %bb.0: -; X32-SSE-NEXT: movdqa {{.*#+}} xmm1 = [0,65535,65535,65535,65535,65535,65535,65535] -; X32-SSE-NEXT: movdqa %xmm1, %xmm2 -; X32-SSE-NEXT: pandn %xmm0, %xmm2 -; X32-SSE-NEXT: pmulhuw {{\.LCPI.*}}, %xmm0 -; X32-SSE-NEXT: pand %xmm1, %xmm0 -; X32-SSE-NEXT: por %xmm2, %xmm0 +; X32-SSE-NEXT: pand {{\.LCPI.*}}, %xmm0 +; X32-SSE-NEXT: movdqa %xmm0, %xmm1 +; X32-SSE-NEXT: psrld $3, %xmm1 +; X32-SSE-NEXT: movdqa %xmm0, %xmm2 +; X32-SSE-NEXT: psrld $2, %xmm2 +; X32-SSE-NEXT: punpckhqdq {{.*#+}} xmm2 = xmm2[1],xmm1[1] +; X32-SSE-NEXT: movdqa %xmm0, %xmm1 +; X32-SSE-NEXT: psrld $1, %xmm1 +; X32-SSE-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0] +; X32-SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,3],xmm2[0,3] ; X32-SSE-NEXT: retl %shift = lshr <4 x i16> %a, <i16 0, i16 1, i16 2, i16 3> ret <4 x i16> %shift @@ -1568,72 +1726,75 @@ define <4 x i16> @constant_shift_v4i16(<4 x i16> %a) nounwind { define <2 x i16> @constant_shift_v2i16(<2 x i16> %a) nounwind { ; SSE2-LABEL: constant_shift_v2i16: ; SSE2: # %bb.0: +; SSE2-NEXT: pand {{.*}}(%rip), %xmm0 ; SSE2-NEXT: movdqa %xmm0, %xmm1 -; SSE2-NEXT: psrlw $3, %xmm1 -; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [65535,0,65535,65535,65535,65535,65535,65535] -; SSE2-NEXT: psrlw $2, %xmm0 -; SSE2-NEXT: pand %xmm2, %xmm0 -; SSE2-NEXT: pandn %xmm1, %xmm2 -; SSE2-NEXT: por %xmm2, %xmm0 +; SSE2-NEXT: psrlq $2, %xmm1 +; SSE2-NEXT: psrlq $3, %xmm0 +; SSE2-NEXT: movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1] ; SSE2-NEXT: retq ; ; SSE41-LABEL: constant_shift_v2i16: ; SSE41: # %bb.0: -; SSE41-NEXT: movdqa %xmm0, %xmm1 -; SSE41-NEXT: psrlw $3, %xmm1 -; SSE41-NEXT: psrlw $2, %xmm0 -; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3,4,5,6,7] +; SSE41-NEXT: pxor %xmm1, %xmm1 +; SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3],xmm0[4],xmm1[5,6,7] +; SSE41-NEXT: movdqa %xmm1, %xmm0 +; SSE41-NEXT: psrlq $3, %xmm0 +; SSE41-NEXT: psrlq $2, %xmm1 +; SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm0[4,5,6,7] +; SSE41-NEXT: movdqa %xmm1, %xmm0 ; SSE41-NEXT: retq ; -; AVX-LABEL: constant_shift_v2i16: -; AVX: # %bb.0: -; AVX-NEXT: vpsrlw $3, %xmm0, %xmm1 -; AVX-NEXT: vpsrlw $2, %xmm0, %xmm0 -; AVX-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3,4,5,6,7] -; AVX-NEXT: retq +; AVX1-LABEL: constant_shift_v2i16: +; AVX1: # %bb.0: +; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3],xmm0[4],xmm1[5,6,7] +; AVX1-NEXT: vpsrlq $3, %xmm0, %xmm1 +; AVX1-NEXT: vpsrlq $2, %xmm0, %xmm0 +; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm1[4,5,6,7] +; AVX1-NEXT: retq ; -; XOP-LABEL: constant_shift_v2i16: -; XOP: # %bb.0: -; XOP-NEXT: vpshlw {{.*}}(%rip), %xmm0, %xmm0 -; XOP-NEXT: retq +; AVX2-LABEL: constant_shift_v2i16: +; AVX2: # %bb.0: +; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; AVX2-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3],xmm0[4],xmm1[5,6,7] +; AVX2-NEXT: vpsrlvq {{.*}}(%rip), %xmm0, %xmm0 +; AVX2-NEXT: retq ; -; AVX512DQ-LABEL: constant_shift_v2i16: -; AVX512DQ: # %bb.0: -; AVX512DQ-NEXT: vpsrlw $3, %xmm0, %xmm1 -; AVX512DQ-NEXT: vpsrlw $2, %xmm0, %xmm0 -; AVX512DQ-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3,4,5,6,7] -; AVX512DQ-NEXT: retq +; XOPAVX1-LABEL: constant_shift_v2i16: +; XOPAVX1: # %bb.0: +; XOPAVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; XOPAVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3],xmm0[4],xmm1[5,6,7] +; XOPAVX1-NEXT: vpshlq {{.*}}(%rip), %xmm0, %xmm0 +; XOPAVX1-NEXT: retq ; -; AVX512BW-LABEL: constant_shift_v2i16: -; AVX512BW: # %bb.0: -; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0 -; AVX512BW-NEXT: vmovdqa {{.*#+}} xmm1 = <2,3,u,u,u,u,u,u> -; AVX512BW-NEXT: vpsrlvw %zmm1, %zmm0, %zmm0 -; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0 -; AVX512BW-NEXT: vzeroupper -; AVX512BW-NEXT: retq +; XOPAVX2-LABEL: constant_shift_v2i16: +; XOPAVX2: # %bb.0: +; XOPAVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; XOPAVX2-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3],xmm0[4],xmm1[5,6,7] +; XOPAVX2-NEXT: vpsrlvq {{.*}}(%rip), %xmm0, %xmm0 +; XOPAVX2-NEXT: retq ; -; AVX512DQVL-LABEL: constant_shift_v2i16: -; AVX512DQVL: # %bb.0: -; AVX512DQVL-NEXT: vpsrlw $3, %xmm0, %xmm1 -; AVX512DQVL-NEXT: vpsrlw $2, %xmm0, %xmm0 -; AVX512DQVL-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3,4,5,6,7] -; AVX512DQVL-NEXT: retq +; AVX512-LABEL: constant_shift_v2i16: +; AVX512: # %bb.0: +; AVX512-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; AVX512-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3],xmm0[4],xmm1[5,6,7] +; AVX512-NEXT: vpsrlvq {{.*}}(%rip), %xmm0, %xmm0 +; AVX512-NEXT: retq ; -; AVX512BWVL-LABEL: constant_shift_v2i16: -; AVX512BWVL: # %bb.0: -; AVX512BWVL-NEXT: vpsrlvw {{.*}}(%rip), %xmm0, %xmm0 -; AVX512BWVL-NEXT: retq +; AVX512VL-LABEL: constant_shift_v2i16: +; AVX512VL: # %bb.0: +; AVX512VL-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; AVX512VL-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3],xmm0[4],xmm1[5,6,7] +; AVX512VL-NEXT: vpsrlvq {{.*}}(%rip), %xmm0, %xmm0 +; AVX512VL-NEXT: retq ; ; X32-SSE-LABEL: constant_shift_v2i16: ; X32-SSE: # %bb.0: +; X32-SSE-NEXT: pand {{\.LCPI.*}}, %xmm0 ; X32-SSE-NEXT: movdqa %xmm0, %xmm1 -; X32-SSE-NEXT: psrlw $3, %xmm1 -; X32-SSE-NEXT: movdqa {{.*#+}} xmm2 = [65535,0,65535,65535,65535,65535,65535,65535] -; X32-SSE-NEXT: psrlw $2, %xmm0 -; X32-SSE-NEXT: pand %xmm2, %xmm0 -; X32-SSE-NEXT: pandn %xmm1, %xmm2 -; X32-SSE-NEXT: por %xmm2, %xmm0 +; X32-SSE-NEXT: psrlq $2, %xmm1 +; X32-SSE-NEXT: psrlq $3, %xmm0 +; X32-SSE-NEXT: movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1] ; X32-SSE-NEXT: retl %shift = lshr <2 x i16> %a, <i16 2, i16 3> ret <2 x i16> %shift @@ -1642,94 +1803,74 @@ define <2 x i16> @constant_shift_v2i16(<2 x i16> %a) nounwind { define <8 x i8> @constant_shift_v8i8(<8 x i8> %a) nounwind { ; SSE2-LABEL: constant_shift_v8i8: ; SSE2: # %bb.0: -; SSE2-NEXT: pxor %xmm1, %xmm1 -; SSE2-NEXT: movdqa %xmm0, %xmm2 -; SSE2-NEXT: punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm1[8],xmm2[9],xmm1[9],xmm2[10],xmm1[10],xmm2[11],xmm1[11],xmm2[12],xmm1[12],xmm2[13],xmm1[13],xmm2[14],xmm1[14],xmm2[15],xmm1[15] -; SSE2-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7] -; SSE2-NEXT: pmullw {{.*}}(%rip), %xmm0 -; SSE2-NEXT: psrlw $8, %xmm0 -; SSE2-NEXT: packuswb %xmm2, %xmm0 +; SSE2-NEXT: pand {{.*}}(%rip), %xmm0 +; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [0,65535,65535,65535,65535,65535,65535,65535] +; SSE2-NEXT: movdqa %xmm1, %xmm2 +; SSE2-NEXT: pandn %xmm0, %xmm2 +; SSE2-NEXT: pmulhuw {{.*}}(%rip), %xmm0 +; SSE2-NEXT: pand %xmm1, %xmm0 +; SSE2-NEXT: por %xmm2, %xmm0 ; SSE2-NEXT: retq ; ; SSE41-LABEL: constant_shift_v8i8: ; SSE41: # %bb.0: -; SSE41-NEXT: pxor %xmm2, %xmm2 -; SSE41-NEXT: pmovzxbw {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero -; SSE41-NEXT: punpckhbw {{.*#+}} xmm0 = 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] -; SSE41-NEXT: pmullw {{.*}}(%rip), %xmm1 -; SSE41-NEXT: psrlw $8, %xmm1 -; SSE41-NEXT: packuswb %xmm0, %xmm1 -; SSE41-NEXT: movdqa %xmm1, %xmm0 +; SSE41-NEXT: pand {{.*}}(%rip), %xmm0 +; SSE41-NEXT: movdqa {{.*#+}} xmm1 = <u,32768,16384,8192,4096,2048,1024,512> +; SSE41-NEXT: pmulhuw %xmm0, %xmm1 +; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3,4,5,6,7] ; SSE41-NEXT: retq ; -; AVX1-LABEL: constant_shift_v8i8: -; AVX1: # %bb.0: -; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1 -; AVX1-NEXT: vpunpckhbw {{.*#+}} xmm1 = 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: vpmovzxbw {{.*#+}} 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 -; AVX1-NEXT: vpmullw {{.*}}(%rip), %xmm0, %xmm0 -; AVX1-NEXT: vpsrlw $8, %xmm0, %xmm0 -; AVX1-NEXT: vpackuswb %xmm1, %xmm0, %xmm0 -; AVX1-NEXT: retq -; -; AVX2-LABEL: constant_shift_v8i8: -; AVX2: # %bb.0: -; AVX2-NEXT: vpmovzxbw {{.*#+}} 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,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero -; AVX2-NEXT: vpmullw {{.*}}(%rip), %ymm0, %ymm0 -; AVX2-NEXT: vpsrlw $8, %ymm0, %ymm0 -; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1 -; AVX2-NEXT: vpackuswb %xmm1, %xmm0, %xmm0 -; AVX2-NEXT: vzeroupper -; AVX2-NEXT: retq +; AVX-LABEL: constant_shift_v8i8: +; AVX: # %bb.0: +; AVX-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; AVX-NEXT: vpmulhuw {{.*}}(%rip), %xmm0, %xmm1 +; AVX-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3,4,5,6,7] +; AVX-NEXT: retq ; ; XOP-LABEL: constant_shift_v8i8: ; XOP: # %bb.0: -; XOP-NEXT: vpshlb {{.*}}(%rip), %xmm0, %xmm0 +; XOP-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; XOP-NEXT: vpshlw {{.*}}(%rip), %xmm0, %xmm0 ; XOP-NEXT: retq ; ; AVX512DQ-LABEL: constant_shift_v8i8: ; AVX512DQ: # %bb.0: -; AVX512DQ-NEXT: vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero -; AVX512DQ-NEXT: vpsrlvd {{.*}}(%rip), %zmm0, %zmm0 -; AVX512DQ-NEXT: vpmovdb %zmm0, %xmm0 -; AVX512DQ-NEXT: vzeroupper +; AVX512DQ-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; AVX512DQ-NEXT: vpmulhuw {{.*}}(%rip), %xmm0, %xmm1 +; AVX512DQ-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3,4,5,6,7] ; AVX512DQ-NEXT: retq ; ; AVX512BW-LABEL: constant_shift_v8i8: ; AVX512BW: # %bb.0: -; AVX512BW-NEXT: vmovdqa {{.*#+}} ymm1 = [0,1,2,3,4,5,6,7,0,0,0,0,0,0,0,0] -; AVX512BW-NEXT: vpmovzxbw {{.*#+}} 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,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero +; AVX512BW-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; AVX512BW-NEXT: vmovdqa {{.*#+}} xmm1 = [0,1,2,3,4,5,6,7] ; AVX512BW-NEXT: vpsrlvw %zmm1, %zmm0, %zmm0 -; AVX512BW-NEXT: vpmovwb %zmm0, %ymm0 -; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 +; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0 ; AVX512BW-NEXT: vzeroupper ; AVX512BW-NEXT: retq ; ; AVX512DQVL-LABEL: constant_shift_v8i8: ; AVX512DQVL: # %bb.0: -; AVX512DQVL-NEXT: vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero -; AVX512DQVL-NEXT: vpsrlvd {{.*}}(%rip), %zmm0, %zmm0 -; AVX512DQVL-NEXT: vpmovdb %zmm0, %xmm0 -; AVX512DQVL-NEXT: vzeroupper +; AVX512DQVL-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; AVX512DQVL-NEXT: vpmulhuw {{.*}}(%rip), %xmm0, %xmm1 +; AVX512DQVL-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3,4,5,6,7] ; AVX512DQVL-NEXT: retq ; ; AVX512BWVL-LABEL: constant_shift_v8i8: ; AVX512BWVL: # %bb.0: -; AVX512BWVL-NEXT: vpmovzxbw {{.*#+}} 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,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero -; AVX512BWVL-NEXT: vpsrlvw {{.*}}(%rip), %ymm0, %ymm0 -; AVX512BWVL-NEXT: vpmovwb %ymm0, %xmm0 -; AVX512BWVL-NEXT: vzeroupper +; AVX512BWVL-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; AVX512BWVL-NEXT: vpsrlvw {{.*}}(%rip), %xmm0, %xmm0 ; AVX512BWVL-NEXT: retq ; ; X32-SSE-LABEL: constant_shift_v8i8: ; X32-SSE: # %bb.0: -; X32-SSE-NEXT: pxor %xmm1, %xmm1 -; X32-SSE-NEXT: movdqa %xmm0, %xmm2 -; X32-SSE-NEXT: punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm1[8],xmm2[9],xmm1[9],xmm2[10],xmm1[10],xmm2[11],xmm1[11],xmm2[12],xmm1[12],xmm2[13],xmm1[13],xmm2[14],xmm1[14],xmm2[15],xmm1[15] -; X32-SSE-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7] -; X32-SSE-NEXT: pmullw {{\.LCPI.*}}, %xmm0 -; X32-SSE-NEXT: psrlw $8, %xmm0 -; X32-SSE-NEXT: packuswb %xmm2, %xmm0 +; X32-SSE-NEXT: pand {{\.LCPI.*}}, %xmm0 +; X32-SSE-NEXT: movdqa {{.*#+}} xmm1 = [0,65535,65535,65535,65535,65535,65535,65535] +; X32-SSE-NEXT: movdqa %xmm1, %xmm2 +; X32-SSE-NEXT: pandn %xmm0, %xmm2 +; X32-SSE-NEXT: pmulhuw {{\.LCPI.*}}, %xmm0 +; X32-SSE-NEXT: pand %xmm1, %xmm0 +; X32-SSE-NEXT: por %xmm2, %xmm0 ; X32-SSE-NEXT: retl %shift = lshr <8 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7> ret <8 x i8> %shift @@ -1738,94 +1879,85 @@ define <8 x i8> @constant_shift_v8i8(<8 x i8> %a) nounwind { define <4 x i8> @constant_shift_v4i8(<4 x i8> %a) nounwind { ; SSE2-LABEL: constant_shift_v4i8: ; SSE2: # %bb.0: -; SSE2-NEXT: pxor %xmm1, %xmm1 +; SSE2-NEXT: pand {{.*}}(%rip), %xmm0 +; SSE2-NEXT: movdqa %xmm0, %xmm1 +; SSE2-NEXT: psrld $3, %xmm1 ; SSE2-NEXT: movdqa %xmm0, %xmm2 -; SSE2-NEXT: punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm1[8],xmm2[9],xmm1[9],xmm2[10],xmm1[10],xmm2[11],xmm1[11],xmm2[12],xmm1[12],xmm2[13],xmm1[13],xmm2[14],xmm1[14],xmm2[15],xmm1[15] -; SSE2-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7] -; SSE2-NEXT: pmullw {{.*}}(%rip), %xmm0 -; SSE2-NEXT: psrlw $8, %xmm0 -; SSE2-NEXT: packuswb %xmm2, %xmm0 +; SSE2-NEXT: psrld $2, %xmm2 +; SSE2-NEXT: punpckhqdq {{.*#+}} xmm2 = xmm2[1],xmm1[1] +; SSE2-NEXT: movdqa %xmm0, %xmm1 +; SSE2-NEXT: psrld $1, %xmm1 +; SSE2-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0] +; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,3],xmm2[0,3] ; SSE2-NEXT: retq ; ; SSE41-LABEL: constant_shift_v4i8: ; SSE41: # %bb.0: -; SSE41-NEXT: pxor %xmm2, %xmm2 -; SSE41-NEXT: pmovzxbw {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero -; SSE41-NEXT: punpckhbw {{.*#+}} xmm0 = 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] -; SSE41-NEXT: pmullw {{.*}}(%rip), %xmm1 -; SSE41-NEXT: psrlw $8, %xmm1 -; SSE41-NEXT: packuswb %xmm0, %xmm1 +; SSE41-NEXT: pand {{.*}}(%rip), %xmm0 +; SSE41-NEXT: movdqa %xmm0, %xmm2 +; SSE41-NEXT: movdqa %xmm0, %xmm1 +; SSE41-NEXT: psrld $2, %xmm1 +; SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm0[0,1,2,3],xmm1[4,5,6,7] +; SSE41-NEXT: psrld $3, %xmm0 +; SSE41-NEXT: psrld $1, %xmm2 +; SSE41-NEXT: pblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm0[4,5,6,7] +; SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7] ; SSE41-NEXT: movdqa %xmm1, %xmm0 ; SSE41-NEXT: retq ; ; AVX1-LABEL: constant_shift_v4i8: ; AVX1: # %bb.0: -; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1 -; AVX1-NEXT: vpunpckhbw {{.*#+}} xmm1 = 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: vpmovzxbw {{.*#+}} 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 -; AVX1-NEXT: vpmullw {{.*}}(%rip), %xmm0, %xmm0 -; AVX1-NEXT: vpsrlw $8, %xmm0, %xmm0 -; AVX1-NEXT: vpackuswb %xmm1, %xmm0, %xmm0 +; AVX1-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; AVX1-NEXT: vpsrld $3, %xmm0, %xmm1 +; AVX1-NEXT: vpsrld $1, %xmm0, %xmm2 +; AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0,1,2,3],xmm1[4,5,6,7] +; AVX1-NEXT: vpsrld $2, %xmm0, %xmm2 +; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm2[4,5,6,7] +; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7] ; AVX1-NEXT: retq ; ; AVX2-LABEL: constant_shift_v4i8: ; AVX2: # %bb.0: -; AVX2-NEXT: vpmovzxbw {{.*#+}} 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,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero -; AVX2-NEXT: vpmullw {{.*}}(%rip), %ymm0, %ymm0 -; AVX2-NEXT: vpsrlw $8, %ymm0, %ymm0 -; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1 -; AVX2-NEXT: vpackuswb %xmm1, %xmm0, %xmm0 -; AVX2-NEXT: vzeroupper +; AVX2-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; AVX2-NEXT: vpsrlvd {{.*}}(%rip), %xmm0, %xmm0 ; AVX2-NEXT: retq ; -; XOP-LABEL: constant_shift_v4i8: -; XOP: # %bb.0: -; XOP-NEXT: vpshlb {{.*}}(%rip), %xmm0, %xmm0 -; XOP-NEXT: retq -; -; AVX512DQ-LABEL: constant_shift_v4i8: -; AVX512DQ: # %bb.0: -; AVX512DQ-NEXT: vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero -; AVX512DQ-NEXT: vpsrlvd {{.*}}(%rip), %zmm0, %zmm0 -; AVX512DQ-NEXT: vpmovdb %zmm0, %xmm0 -; AVX512DQ-NEXT: vzeroupper -; AVX512DQ-NEXT: retq +; XOPAVX1-LABEL: constant_shift_v4i8: +; XOPAVX1: # %bb.0: +; XOPAVX1-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; XOPAVX1-NEXT: vpshld {{.*}}(%rip), %xmm0, %xmm0 +; XOPAVX1-NEXT: retq ; -; AVX512BW-LABEL: constant_shift_v4i8: -; AVX512BW: # %bb.0: -; AVX512BW-NEXT: vmovdqa {{.*#+}} ymm1 = [0,1,2,3,0,0,0,0,0,0,0,0,0,0,0,0] -; AVX512BW-NEXT: vpmovzxbw {{.*#+}} 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,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero -; AVX512BW-NEXT: vpsrlvw %zmm1, %zmm0, %zmm0 -; AVX512BW-NEXT: vpmovwb %zmm0, %ymm0 -; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 -; AVX512BW-NEXT: vzeroupper -; AVX512BW-NEXT: retq +; XOPAVX2-LABEL: constant_shift_v4i8: +; XOPAVX2: # %bb.0: +; XOPAVX2-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; XOPAVX2-NEXT: vpsrlvd {{.*}}(%rip), %xmm0, %xmm0 +; XOPAVX2-NEXT: retq ; -; AVX512DQVL-LABEL: constant_shift_v4i8: -; AVX512DQVL: # %bb.0: -; AVX512DQVL-NEXT: vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero -; AVX512DQVL-NEXT: vpsrlvd {{.*}}(%rip), %zmm0, %zmm0 -; AVX512DQVL-NEXT: vpmovdb %zmm0, %xmm0 -; AVX512DQVL-NEXT: vzeroupper -; AVX512DQVL-NEXT: retq +; AVX512-LABEL: constant_shift_v4i8: +; AVX512: # %bb.0: +; AVX512-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; AVX512-NEXT: vpsrlvd {{.*}}(%rip), %xmm0, %xmm0 +; AVX512-NEXT: retq ; -; AVX512BWVL-LABEL: constant_shift_v4i8: -; AVX512BWVL: # %bb.0: -; AVX512BWVL-NEXT: vpmovzxbw {{.*#+}} 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,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero -; AVX512BWVL-NEXT: vpsrlvw {{.*}}(%rip), %ymm0, %ymm0 -; AVX512BWVL-NEXT: vpmovwb %ymm0, %xmm0 -; AVX512BWVL-NEXT: vzeroupper -; AVX512BWVL-NEXT: retq +; AVX512VL-LABEL: constant_shift_v4i8: +; AVX512VL: # %bb.0: +; AVX512VL-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; AVX512VL-NEXT: vpsrlvd {{.*}}(%rip), %xmm0, %xmm0 +; AVX512VL-NEXT: retq ; ; X32-SSE-LABEL: constant_shift_v4i8: ; X32-SSE: # %bb.0: -; X32-SSE-NEXT: pxor %xmm1, %xmm1 +; X32-SSE-NEXT: pand {{\.LCPI.*}}, %xmm0 +; X32-SSE-NEXT: movdqa %xmm0, %xmm1 +; X32-SSE-NEXT: psrld $3, %xmm1 ; X32-SSE-NEXT: movdqa %xmm0, %xmm2 -; X32-SSE-NEXT: punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm1[8],xmm2[9],xmm1[9],xmm2[10],xmm1[10],xmm2[11],xmm1[11],xmm2[12],xmm1[12],xmm2[13],xmm1[13],xmm2[14],xmm1[14],xmm2[15],xmm1[15] -; X32-SSE-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7] -; X32-SSE-NEXT: pmullw {{\.LCPI.*}}, %xmm0 -; X32-SSE-NEXT: psrlw $8, %xmm0 -; X32-SSE-NEXT: packuswb %xmm2, %xmm0 +; X32-SSE-NEXT: psrld $2, %xmm2 +; X32-SSE-NEXT: punpckhqdq {{.*#+}} xmm2 = xmm2[1],xmm1[1] +; X32-SSE-NEXT: movdqa %xmm0, %xmm1 +; X32-SSE-NEXT: psrld $1, %xmm1 +; X32-SSE-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0] +; X32-SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,3],xmm2[0,3] ; X32-SSE-NEXT: retl %shift = lshr <4 x i8> %a, <i8 0, i8 1, i8 2, i8 3> ret <4 x i8> %shift @@ -1834,94 +1966,67 @@ define <4 x i8> @constant_shift_v4i8(<4 x i8> %a) nounwind { define <2 x i8> @constant_shift_v2i8(<2 x i8> %a) nounwind { ; SSE2-LABEL: constant_shift_v2i8: ; SSE2: # %bb.0: -; SSE2-NEXT: pxor %xmm1, %xmm1 -; SSE2-NEXT: movdqa %xmm0, %xmm2 -; SSE2-NEXT: punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm1[8],xmm2[9],xmm1[9],xmm2[10],xmm1[10],xmm2[11],xmm1[11],xmm2[12],xmm1[12],xmm2[13],xmm1[13],xmm2[14],xmm1[14],xmm2[15],xmm1[15] -; SSE2-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7] -; SSE2-NEXT: pmullw {{.*}}(%rip), %xmm0 -; SSE2-NEXT: psrlw $8, %xmm0 -; SSE2-NEXT: packuswb %xmm2, %xmm0 +; SSE2-NEXT: pand {{.*}}(%rip), %xmm0 +; SSE2-NEXT: movdqa %xmm0, %xmm1 +; SSE2-NEXT: psrlq $2, %xmm1 +; SSE2-NEXT: psrlq $3, %xmm0 +; SSE2-NEXT: movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1] ; SSE2-NEXT: retq ; ; SSE41-LABEL: constant_shift_v2i8: ; SSE41: # %bb.0: -; SSE41-NEXT: pxor %xmm2, %xmm2 -; SSE41-NEXT: pmovzxbw {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero -; SSE41-NEXT: punpckhbw {{.*#+}} xmm0 = 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] -; SSE41-NEXT: pmullw {{.*}}(%rip), %xmm1 -; SSE41-NEXT: psrlw $8, %xmm1 -; SSE41-NEXT: packuswb %xmm0, %xmm1 -; SSE41-NEXT: movdqa %xmm1, %xmm0 +; SSE41-NEXT: pand {{.*}}(%rip), %xmm0 +; SSE41-NEXT: movdqa %xmm0, %xmm1 +; SSE41-NEXT: psrlq $3, %xmm1 +; SSE41-NEXT: psrlq $2, %xmm0 +; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm1[4,5,6,7] ; SSE41-NEXT: retq ; ; AVX1-LABEL: constant_shift_v2i8: ; AVX1: # %bb.0: -; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1 -; AVX1-NEXT: vpunpckhbw {{.*#+}} xmm1 = 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: vpmovzxbw {{.*#+}} 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 -; AVX1-NEXT: vpmullw {{.*}}(%rip), %xmm0, %xmm0 -; AVX1-NEXT: vpsrlw $8, %xmm0, %xmm0 -; AVX1-NEXT: vpackuswb %xmm1, %xmm0, %xmm0 +; AVX1-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; AVX1-NEXT: vpsrlq $3, %xmm0, %xmm1 +; AVX1-NEXT: vpsrlq $2, %xmm0, %xmm0 +; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm1[4,5,6,7] ; AVX1-NEXT: retq ; ; AVX2-LABEL: constant_shift_v2i8: ; AVX2: # %bb.0: -; AVX2-NEXT: vpmovzxbw {{.*#+}} 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,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero -; AVX2-NEXT: vpmullw {{.*}}(%rip), %ymm0, %ymm0 -; AVX2-NEXT: vpsrlw $8, %ymm0, %ymm0 -; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1 -; AVX2-NEXT: vpackuswb %xmm1, %xmm0, %xmm0 -; AVX2-NEXT: vzeroupper +; AVX2-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; AVX2-NEXT: vpsrlvq {{.*}}(%rip), %xmm0, %xmm0 ; AVX2-NEXT: retq ; -; XOP-LABEL: constant_shift_v2i8: -; XOP: # %bb.0: -; XOP-NEXT: vpshlb {{.*}}(%rip), %xmm0, %xmm0 -; XOP-NEXT: retq -; -; AVX512DQ-LABEL: constant_shift_v2i8: -; AVX512DQ: # %bb.0: -; AVX512DQ-NEXT: vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero -; AVX512DQ-NEXT: vpsrlvd {{.*}}(%rip), %zmm0, %zmm0 -; AVX512DQ-NEXT: vpmovdb %zmm0, %xmm0 -; AVX512DQ-NEXT: vzeroupper -; AVX512DQ-NEXT: retq +; XOPAVX1-LABEL: constant_shift_v2i8: +; XOPAVX1: # %bb.0: +; XOPAVX1-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; XOPAVX1-NEXT: vpshlq {{.*}}(%rip), %xmm0, %xmm0 +; XOPAVX1-NEXT: retq ; -; AVX512BW-LABEL: constant_shift_v2i8: -; AVX512BW: # %bb.0: -; AVX512BW-NEXT: vmovdqa {{.*#+}} ymm1 = [2,3,0,0,0,0,0,0,0,0,0,0,0,0,0,0] -; AVX512BW-NEXT: vpmovzxbw {{.*#+}} 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,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero -; AVX512BW-NEXT: vpsrlvw %zmm1, %zmm0, %zmm0 -; AVX512BW-NEXT: vpmovwb %zmm0, %ymm0 -; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 -; AVX512BW-NEXT: vzeroupper -; AVX512BW-NEXT: retq +; XOPAVX2-LABEL: constant_shift_v2i8: +; XOPAVX2: # %bb.0: +; XOPAVX2-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; XOPAVX2-NEXT: vpsrlvq {{.*}}(%rip), %xmm0, %xmm0 +; XOPAVX2-NEXT: retq ; -; AVX512DQVL-LABEL: constant_shift_v2i8: -; AVX512DQVL: # %bb.0: -; AVX512DQVL-NEXT: vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero -; AVX512DQVL-NEXT: vpsrlvd {{.*}}(%rip), %zmm0, %zmm0 -; AVX512DQVL-NEXT: vpmovdb %zmm0, %xmm0 -; AVX512DQVL-NEXT: vzeroupper -; AVX512DQVL-NEXT: retq +; AVX512-LABEL: constant_shift_v2i8: +; AVX512: # %bb.0: +; AVX512-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; AVX512-NEXT: vpsrlvq {{.*}}(%rip), %xmm0, %xmm0 +; AVX512-NEXT: retq ; -; AVX512BWVL-LABEL: constant_shift_v2i8: -; AVX512BWVL: # %bb.0: -; AVX512BWVL-NEXT: vpmovzxbw {{.*#+}} 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,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero -; AVX512BWVL-NEXT: vpsrlvw {{.*}}(%rip), %ymm0, %ymm0 -; AVX512BWVL-NEXT: vpmovwb %ymm0, %xmm0 -; AVX512BWVL-NEXT: vzeroupper -; AVX512BWVL-NEXT: retq +; AVX512VL-LABEL: constant_shift_v2i8: +; AVX512VL: # %bb.0: +; AVX512VL-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; AVX512VL-NEXT: vpsrlvq {{.*}}(%rip), %xmm0, %xmm0 +; AVX512VL-NEXT: retq ; ; X32-SSE-LABEL: constant_shift_v2i8: ; X32-SSE: # %bb.0: -; X32-SSE-NEXT: pxor %xmm1, %xmm1 -; X32-SSE-NEXT: movdqa %xmm0, %xmm2 -; X32-SSE-NEXT: punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm1[8],xmm2[9],xmm1[9],xmm2[10],xmm1[10],xmm2[11],xmm1[11],xmm2[12],xmm1[12],xmm2[13],xmm1[13],xmm2[14],xmm1[14],xmm2[15],xmm1[15] -; X32-SSE-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7] -; X32-SSE-NEXT: pmullw {{\.LCPI.*}}, %xmm0 -; X32-SSE-NEXT: psrlw $8, %xmm0 -; X32-SSE-NEXT: packuswb %xmm2, %xmm0 +; X32-SSE-NEXT: pand {{\.LCPI.*}}, %xmm0 +; X32-SSE-NEXT: movdqa %xmm0, %xmm1 +; X32-SSE-NEXT: psrlq $2, %xmm1 +; X32-SSE-NEXT: psrlq $3, %xmm0 +; X32-SSE-NEXT: movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1] ; X32-SSE-NEXT: retl %shift = lshr <2 x i8> %a, <i8 2, i8 3> ret <2 x i8> %shift @@ -1932,102 +2037,167 @@ define <2 x i8> @constant_shift_v2i8(<2 x i8> %a) nounwind { ; define <2 x i32> @splatconstant_shift_v2i32(<2 x i32> %a) nounwind { -; SSE-LABEL: splatconstant_shift_v2i32: -; SSE: # %bb.0: -; SSE-NEXT: psrld $5, %xmm0 -; SSE-NEXT: retq +; SSE2-LABEL: splatconstant_shift_v2i32: +; SSE2: # %bb.0: +; SSE2-NEXT: pand {{.*}}(%rip), %xmm0 +; SSE2-NEXT: psrlq $5, %xmm0 +; SSE2-NEXT: retq ; -; AVX-LABEL: splatconstant_shift_v2i32: -; AVX: # %bb.0: -; AVX-NEXT: vpsrld $5, %xmm0, %xmm0 -; AVX-NEXT: retq +; SSE41-LABEL: splatconstant_shift_v2i32: +; SSE41: # %bb.0: +; SSE41-NEXT: pxor %xmm1, %xmm1 +; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7] +; SSE41-NEXT: psrlq $5, %xmm0 +; SSE41-NEXT: retq ; -; XOP-LABEL: splatconstant_shift_v2i32: -; XOP: # %bb.0: -; XOP-NEXT: vpsrld $5, %xmm0, %xmm0 -; XOP-NEXT: retq +; AVX1-LABEL: splatconstant_shift_v2i32: +; AVX1: # %bb.0: +; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7] +; AVX1-NEXT: vpsrlq $5, %xmm0, %xmm0 +; AVX1-NEXT: retq +; +; AVX2-LABEL: splatconstant_shift_v2i32: +; AVX2: # %bb.0: +; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3] +; AVX2-NEXT: vpsrlq $5, %xmm0, %xmm0 +; AVX2-NEXT: retq +; +; XOPAVX1-LABEL: splatconstant_shift_v2i32: +; XOPAVX1: # %bb.0: +; XOPAVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; XOPAVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7] +; XOPAVX1-NEXT: vpsrlq $5, %xmm0, %xmm0 +; XOPAVX1-NEXT: retq +; +; XOPAVX2-LABEL: splatconstant_shift_v2i32: +; XOPAVX2: # %bb.0: +; XOPAVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; XOPAVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3] +; XOPAVX2-NEXT: vpsrlq $5, %xmm0, %xmm0 +; XOPAVX2-NEXT: retq ; ; AVX512-LABEL: splatconstant_shift_v2i32: ; AVX512: # %bb.0: -; AVX512-NEXT: vpsrld $5, %xmm0, %xmm0 +; AVX512-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; AVX512-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3] +; AVX512-NEXT: vpsrlq $5, %xmm0, %xmm0 ; AVX512-NEXT: retq ; ; AVX512VL-LABEL: splatconstant_shift_v2i32: ; AVX512VL: # %bb.0: -; AVX512VL-NEXT: vpsrld $5, %xmm0, %xmm0 +; AVX512VL-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; AVX512VL-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3] +; AVX512VL-NEXT: vpsrlq $5, %xmm0, %xmm0 ; AVX512VL-NEXT: retq ; ; X32-SSE-LABEL: splatconstant_shift_v2i32: ; X32-SSE: # %bb.0: -; X32-SSE-NEXT: psrld $5, %xmm0 +; X32-SSE-NEXT: pand {{\.LCPI.*}}, %xmm0 +; X32-SSE-NEXT: psrlq $5, %xmm0 ; X32-SSE-NEXT: retl %shift = lshr <2 x i32> %a, <i32 5, i32 5> ret <2 x i32> %shift } define <4 x i16> @splatconstant_shift_v4i16(<4 x i16> %a) nounwind { -; SSE-LABEL: splatconstant_shift_v4i16: -; SSE: # %bb.0: -; SSE-NEXT: psrlw $3, %xmm0 -; SSE-NEXT: retq +; SSE2-LABEL: splatconstant_shift_v4i16: +; SSE2: # %bb.0: +; SSE2-NEXT: pand {{.*}}(%rip), %xmm0 +; SSE2-NEXT: psrld $3, %xmm0 +; SSE2-NEXT: retq +; +; SSE41-LABEL: splatconstant_shift_v4i16: +; SSE41: # %bb.0: +; SSE41-NEXT: pxor %xmm1, %xmm1 +; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7] +; SSE41-NEXT: psrld $3, %xmm0 +; SSE41-NEXT: retq ; ; AVX-LABEL: splatconstant_shift_v4i16: ; AVX: # %bb.0: -; AVX-NEXT: vpsrlw $3, %xmm0, %xmm0 +; AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; AVX-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7] +; AVX-NEXT: vpsrld $3, %xmm0, %xmm0 ; AVX-NEXT: retq ; ; XOP-LABEL: splatconstant_shift_v4i16: ; XOP: # %bb.0: -; XOP-NEXT: vpsrlw $3, %xmm0, %xmm0 +; XOP-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; XOP-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7] +; XOP-NEXT: vpsrld $3, %xmm0, %xmm0 ; XOP-NEXT: retq ; ; AVX512-LABEL: splatconstant_shift_v4i16: ; AVX512: # %bb.0: -; AVX512-NEXT: vpsrlw $3, %xmm0, %xmm0 +; AVX512-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; AVX512-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7] +; AVX512-NEXT: vpsrld $3, %xmm0, %xmm0 ; AVX512-NEXT: retq ; ; AVX512VL-LABEL: splatconstant_shift_v4i16: ; AVX512VL: # %bb.0: -; AVX512VL-NEXT: vpsrlw $3, %xmm0, %xmm0 +; AVX512VL-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; AVX512VL-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7] +; AVX512VL-NEXT: vpsrld $3, %xmm0, %xmm0 ; AVX512VL-NEXT: retq ; ; X32-SSE-LABEL: splatconstant_shift_v4i16: ; X32-SSE: # %bb.0: -; X32-SSE-NEXT: psrlw $3, %xmm0 +; X32-SSE-NEXT: pand {{\.LCPI.*}}, %xmm0 +; X32-SSE-NEXT: psrld $3, %xmm0 ; X32-SSE-NEXT: retl %shift = lshr <4 x i16> %a, <i16 3, i16 3, i16 3, i16 3> ret <4 x i16> %shift } define <2 x i16> @splatconstant_shift_v2i16(<2 x i16> %a) nounwind { -; SSE-LABEL: splatconstant_shift_v2i16: -; SSE: # %bb.0: -; SSE-NEXT: psrlw $3, %xmm0 -; SSE-NEXT: retq +; SSE2-LABEL: splatconstant_shift_v2i16: +; SSE2: # %bb.0: +; SSE2-NEXT: pand {{.*}}(%rip), %xmm0 +; SSE2-NEXT: psrlq $3, %xmm0 +; SSE2-NEXT: retq +; +; SSE41-LABEL: splatconstant_shift_v2i16: +; SSE41: # %bb.0: +; SSE41-NEXT: pxor %xmm1, %xmm1 +; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3],xmm0[4],xmm1[5,6,7] +; SSE41-NEXT: psrlq $3, %xmm0 +; SSE41-NEXT: retq ; ; AVX-LABEL: splatconstant_shift_v2i16: ; AVX: # %bb.0: -; AVX-NEXT: vpsrlw $3, %xmm0, %xmm0 +; AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; AVX-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3],xmm0[4],xmm1[5,6,7] +; AVX-NEXT: vpsrlq $3, %xmm0, %xmm0 ; AVX-NEXT: retq ; ; XOP-LABEL: splatconstant_shift_v2i16: ; XOP: # %bb.0: -; XOP-NEXT: vpsrlw $3, %xmm0, %xmm0 +; XOP-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; XOP-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3],xmm0[4],xmm1[5,6,7] +; XOP-NEXT: vpsrlq $3, %xmm0, %xmm0 ; XOP-NEXT: retq ; ; AVX512-LABEL: splatconstant_shift_v2i16: ; AVX512: # %bb.0: -; AVX512-NEXT: vpsrlw $3, %xmm0, %xmm0 +; AVX512-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; AVX512-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3],xmm0[4],xmm1[5,6,7] +; AVX512-NEXT: vpsrlq $3, %xmm0, %xmm0 ; AVX512-NEXT: retq ; ; AVX512VL-LABEL: splatconstant_shift_v2i16: ; AVX512VL: # %bb.0: -; AVX512VL-NEXT: vpsrlw $3, %xmm0, %xmm0 +; AVX512VL-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; AVX512VL-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3],xmm0[4],xmm1[5,6,7] +; AVX512VL-NEXT: vpsrlq $3, %xmm0, %xmm0 ; AVX512VL-NEXT: retq ; ; X32-SSE-LABEL: splatconstant_shift_v2i16: ; X32-SSE: # %bb.0: -; X32-SSE-NEXT: psrlw $3, %xmm0 +; X32-SSE-NEXT: pand {{\.LCPI.*}}, %xmm0 +; X32-SSE-NEXT: psrlq $3, %xmm0 ; X32-SSE-NEXT: retl %shift = lshr <2 x i16> %a, <i16 3, i16 3> ret <2 x i16> %shift @@ -2036,37 +2206,38 @@ define <2 x i16> @splatconstant_shift_v2i16(<2 x i16> %a) nounwind { define <8 x i8> @splatconstant_shift_v8i8(<8 x i8> %a) nounwind { ; SSE-LABEL: splatconstant_shift_v8i8: ; SSE: # %bb.0: -; SSE-NEXT: psrlw $3, %xmm0 ; SSE-NEXT: pand {{.*}}(%rip), %xmm0 +; SSE-NEXT: psrlw $3, %xmm0 ; SSE-NEXT: retq ; ; AVX-LABEL: splatconstant_shift_v8i8: ; AVX: # %bb.0: -; AVX-NEXT: vpsrlw $3, %xmm0, %xmm0 ; AVX-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; AVX-NEXT: vpsrlw $3, %xmm0, %xmm0 ; AVX-NEXT: retq ; ; XOP-LABEL: splatconstant_shift_v8i8: ; XOP: # %bb.0: -; XOP-NEXT: vpshlb {{.*}}(%rip), %xmm0, %xmm0 +; XOP-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; XOP-NEXT: vpsrlw $3, %xmm0, %xmm0 ; XOP-NEXT: retq ; ; AVX512-LABEL: splatconstant_shift_v8i8: ; AVX512: # %bb.0: -; AVX512-NEXT: vpsrlw $3, %xmm0, %xmm0 ; AVX512-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; AVX512-NEXT: vpsrlw $3, %xmm0, %xmm0 ; AVX512-NEXT: retq ; ; AVX512VL-LABEL: splatconstant_shift_v8i8: ; AVX512VL: # %bb.0: -; AVX512VL-NEXT: vpsrlw $3, %xmm0, %xmm0 ; AVX512VL-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; AVX512VL-NEXT: vpsrlw $3, %xmm0, %xmm0 ; AVX512VL-NEXT: retq ; ; X32-SSE-LABEL: splatconstant_shift_v8i8: ; X32-SSE: # %bb.0: -; X32-SSE-NEXT: psrlw $3, %xmm0 ; X32-SSE-NEXT: pand {{\.LCPI.*}}, %xmm0 +; X32-SSE-NEXT: psrlw $3, %xmm0 ; X32-SSE-NEXT: retl %shift = lshr <8 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3> ret <8 x i8> %shift @@ -2075,37 +2246,38 @@ define <8 x i8> @splatconstant_shift_v8i8(<8 x i8> %a) nounwind { define <4 x i8> @splatconstant_shift_v4i8(<4 x i8> %a) nounwind { ; SSE-LABEL: splatconstant_shift_v4i8: ; SSE: # %bb.0: -; SSE-NEXT: psrlw $3, %xmm0 ; SSE-NEXT: pand {{.*}}(%rip), %xmm0 +; SSE-NEXT: psrld $3, %xmm0 ; SSE-NEXT: retq ; ; AVX-LABEL: splatconstant_shift_v4i8: ; AVX: # %bb.0: -; AVX-NEXT: vpsrlw $3, %xmm0, %xmm0 ; AVX-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; AVX-NEXT: vpsrld $3, %xmm0, %xmm0 ; AVX-NEXT: retq ; ; XOP-LABEL: splatconstant_shift_v4i8: ; XOP: # %bb.0: -; XOP-NEXT: vpshlb {{.*}}(%rip), %xmm0, %xmm0 +; XOP-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; XOP-NEXT: vpsrld $3, %xmm0, %xmm0 ; XOP-NEXT: retq ; ; AVX512-LABEL: splatconstant_shift_v4i8: ; AVX512: # %bb.0: -; AVX512-NEXT: vpsrlw $3, %xmm0, %xmm0 ; AVX512-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; AVX512-NEXT: vpsrld $3, %xmm0, %xmm0 ; AVX512-NEXT: retq ; ; AVX512VL-LABEL: splatconstant_shift_v4i8: ; AVX512VL: # %bb.0: -; AVX512VL-NEXT: vpsrlw $3, %xmm0, %xmm0 ; AVX512VL-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; AVX512VL-NEXT: vpsrld $3, %xmm0, %xmm0 ; AVX512VL-NEXT: retq ; ; X32-SSE-LABEL: splatconstant_shift_v4i8: ; X32-SSE: # %bb.0: -; X32-SSE-NEXT: psrlw $3, %xmm0 ; X32-SSE-NEXT: pand {{\.LCPI.*}}, %xmm0 +; X32-SSE-NEXT: psrld $3, %xmm0 ; X32-SSE-NEXT: retl %shift = lshr <4 x i8> %a, <i8 3, i8 3, i8 3, i8 3> ret <4 x i8> %shift @@ -2114,37 +2286,38 @@ define <4 x i8> @splatconstant_shift_v4i8(<4 x i8> %a) nounwind { define <2 x i8> @splatconstant_shift_v2i8(<2 x i8> %a) nounwind { ; SSE-LABEL: splatconstant_shift_v2i8: ; SSE: # %bb.0: -; SSE-NEXT: psrlw $3, %xmm0 ; SSE-NEXT: pand {{.*}}(%rip), %xmm0 +; SSE-NEXT: psrlq $3, %xmm0 ; SSE-NEXT: retq ; ; AVX-LABEL: splatconstant_shift_v2i8: ; AVX: # %bb.0: -; AVX-NEXT: vpsrlw $3, %xmm0, %xmm0 ; AVX-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; AVX-NEXT: vpsrlq $3, %xmm0, %xmm0 ; AVX-NEXT: retq ; ; XOP-LABEL: splatconstant_shift_v2i8: ; XOP: # %bb.0: -; XOP-NEXT: vpshlb {{.*}}(%rip), %xmm0, %xmm0 +; XOP-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; XOP-NEXT: vpsrlq $3, %xmm0, %xmm0 ; XOP-NEXT: retq ; ; AVX512-LABEL: splatconstant_shift_v2i8: ; AVX512: # %bb.0: -; AVX512-NEXT: vpsrlw $3, %xmm0, %xmm0 ; AVX512-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; AVX512-NEXT: vpsrlq $3, %xmm0, %xmm0 ; AVX512-NEXT: retq ; ; AVX512VL-LABEL: splatconstant_shift_v2i8: ; AVX512VL: # %bb.0: -; AVX512VL-NEXT: vpsrlw $3, %xmm0, %xmm0 ; AVX512VL-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 +; AVX512VL-NEXT: vpsrlq $3, %xmm0, %xmm0 ; AVX512VL-NEXT: retq ; ; X32-SSE-LABEL: splatconstant_shift_v2i8: ; X32-SSE: # %bb.0: -; X32-SSE-NEXT: psrlw $3, %xmm0 ; X32-SSE-NEXT: pand {{\.LCPI.*}}, %xmm0 +; X32-SSE-NEXT: psrlq $3, %xmm0 ; X32-SSE-NEXT: retl %shift = lshr <2 x i8> %a, <i8 3, i8 3> ret <2 x i8> %shift |
