1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc < %s -mtriple=i686-linux-pc -mcpu=corei7 | FileCheck %s --check-prefix=X32 3; RUN: llc < %s -mtriple=x86_64-linux-pc -mcpu=corei7 | FileCheck %s --check-prefix=X64 4 5; uitofp <2 x i32> codegen from buildvector or legalization is different but gives the same results 6; across the full 0 - 0xFFFFFFFF u32 range. 7 8define <2 x float> @uitofp_2i32_cvt_buildvector(i32 %x, i32 %y, <2 x float> %v) { 9; X32-LABEL: uitofp_2i32_cvt_buildvector: 10; X32: # %bb.0: 11; X32-NEXT: movq {{.*#+}} xmm1 = mem[0],zero 12; X32-NEXT: movdqa {{.*#+}} xmm2 = [1258291200,1258291200,1258291200,1258291200] 13; X32-NEXT: pblendw {{.*#+}} xmm2 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7] 14; X32-NEXT: psrld $16, %xmm1 15; X32-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0],mem[1],xmm1[2],mem[3],xmm1[4],mem[5],xmm1[6],mem[7] 16; X32-NEXT: addps {{\.LCPI.*}}, %xmm1 17; X32-NEXT: addps %xmm2, %xmm1 18; X32-NEXT: mulps %xmm1, %xmm0 19; X32-NEXT: retl 20; 21; X64-LABEL: uitofp_2i32_cvt_buildvector: 22; X64: # %bb.0: 23; X64-NEXT: movd %edi, %xmm1 24; X64-NEXT: pinsrd $1, %esi, %xmm1 25; X64-NEXT: movdqa {{.*#+}} xmm2 = [1258291200,1258291200,1258291200,1258291200] 26; X64-NEXT: pblendw {{.*#+}} xmm2 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7] 27; X64-NEXT: psrld $16, %xmm1 28; X64-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0],mem[1],xmm1[2],mem[3],xmm1[4],mem[5],xmm1[6],mem[7] 29; X64-NEXT: addps {{.*}}(%rip), %xmm1 30; X64-NEXT: addps %xmm2, %xmm1 31; X64-NEXT: mulps %xmm1, %xmm0 32; X64-NEXT: retq 33 %t1 = uitofp i32 %x to float 34 %t2 = insertelement <2 x float> undef, float %t1, i32 0 35 %t3 = uitofp i32 %y to float 36 %t4 = insertelement <2 x float> %t2, float %t3, i32 1 37 %t5 = fmul <2 x float> %v, %t4 38 ret <2 x float> %t5 39} 40 41define <2 x float> @uitofp_2i32_buildvector_cvt(i32 %x, i32 %y, <2 x float> %v) { 42; X32-LABEL: uitofp_2i32_buildvector_cvt: 43; X32: # %bb.0: 44; X32-NEXT: movss {{.*#+}} xmm1 = mem[0],zero,zero,zero 45; X32-NEXT: movss {{.*#+}} xmm2 = mem[0],zero,zero,zero 46; X32-NEXT: unpcklpd {{.*#+}} xmm2 = xmm2[0],xmm1[0] 47; X32-NEXT: movapd {{.*#+}} xmm1 = [4.503600e+15,4.503600e+15] 48; X32-NEXT: orpd %xmm1, %xmm2 49; X32-NEXT: subpd %xmm1, %xmm2 50; X32-NEXT: cvtpd2ps %xmm2, %xmm1 51; X32-NEXT: mulps %xmm1, %xmm0 52; X32-NEXT: retl 53; 54; X64-LABEL: uitofp_2i32_buildvector_cvt: 55; X64: # %bb.0: 56; X64-NEXT: movd %esi, %xmm1 57; X64-NEXT: movd %edi, %xmm2 58; X64-NEXT: punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm1[0] 59; X64-NEXT: movdqa {{.*#+}} xmm1 = [4.503600e+15,4.503600e+15] 60; X64-NEXT: por %xmm1, %xmm2 61; X64-NEXT: subpd %xmm1, %xmm2 62; X64-NEXT: cvtpd2ps %xmm2, %xmm1 63; X64-NEXT: mulps %xmm1, %xmm0 64; X64-NEXT: retq 65 %t1 = insertelement <2 x i32> undef, i32 %x, i32 0 66 %t2 = insertelement <2 x i32> %t1, i32 %y, i32 1 67 %t3 = uitofp <2 x i32> %t2 to <2 x float> 68 %t4 = fmul <2 x float> %v, %t3 69 ret <2 x float> %t4 70} 71 72define <2 x float> @uitofp_2i32_legalized(<2 x i32> %in, <2 x float> %v) { 73; X32-LABEL: uitofp_2i32_legalized: 74; X32: # %bb.0: 75; X32-NEXT: xorps %xmm2, %xmm2 76; X32-NEXT: blendps {{.*#+}} xmm2 = xmm0[0],xmm2[1],xmm0[2],xmm2[3] 77; X32-NEXT: movaps {{.*#+}} xmm0 = [4.503600e+15,4.503600e+15] 78; X32-NEXT: orps %xmm0, %xmm2 79; X32-NEXT: subpd %xmm0, %xmm2 80; X32-NEXT: cvtpd2ps %xmm2, %xmm0 81; X32-NEXT: mulps %xmm1, %xmm0 82; X32-NEXT: retl 83; 84; X64-LABEL: uitofp_2i32_legalized: 85; X64: # %bb.0: 86; X64-NEXT: xorps %xmm2, %xmm2 87; X64-NEXT: blendps {{.*#+}} xmm2 = xmm0[0],xmm2[1],xmm0[2],xmm2[3] 88; X64-NEXT: movaps {{.*#+}} xmm0 = [4.503600e+15,4.503600e+15] 89; X64-NEXT: orps %xmm0, %xmm2 90; X64-NEXT: subpd %xmm0, %xmm2 91; X64-NEXT: cvtpd2ps %xmm2, %xmm0 92; X64-NEXT: mulps %xmm1, %xmm0 93; X64-NEXT: retq 94 %t1 = uitofp <2 x i32> %in to <2 x float> 95 %t2 = fmul <2 x float> %v, %t1 96 ret <2 x float> %t2 97} 98