diff options
Diffstat (limited to 'common/arm64/ihevc_inter_pred_filters_luma_vert.s')
| -rw-r--r-- | common/arm64/ihevc_inter_pred_filters_luma_vert.s | 272 |
1 files changed, 136 insertions, 136 deletions
diff --git a/common/arm64/ihevc_inter_pred_filters_luma_vert.s b/common/arm64/ihevc_inter_pred_filters_luma_vert.s index 48dc30f..bd8b3c4 100644 --- a/common/arm64/ihevc_inter_pred_filters_luma_vert.s +++ b/common/arm64/ihevc_inter_pred_filters_luma_vert.s @@ -115,7 +115,7 @@ ihevc_inter_pred_luma_vert_av8: // stmfd sp!, {x4-x12, x14} //stack stores the values of the arguments - push_v_regs + stp x19, x20,[sp,#-16]! mov x15,x4 // pi1_coeff @@ -161,87 +161,87 @@ prolog: ld1 {v0.8b},[x0],#8 //src_tmp1 = vld1_u8(pu1_src_tmp)// subs x4,x4,#8 ld1 {v2.8b},[x3],x2 //src_tmp3 = vld1_u8(pu1_src_tmp)// - umull v8.8h, v1.8b, v23.8b //mul_res1 = vmull_u8(src_tmp2, coeffabs_1)// + umull v19.8h, v1.8b, v23.8b //mul_res1 = vmull_u8(src_tmp2, coeffabs_1)// ld1 {v3.8b},[x3],x2 //src_tmp4 = vld1_u8(pu1_src_tmp)// - umlsl v8.8h, v0.8b, v22.8b //mul_res1 = vmlsl_u8(mul_res1, src_tmp1, coeffabs_0)// + umlsl v19.8h, v0.8b, v22.8b //mul_res1 = vmlsl_u8(mul_res1, src_tmp1, coeffabs_0)// ld1 {v4.8b},[x3],x2 //src_tmp1 = vld1_u8(pu1_src_tmp)// - umlsl v8.8h, v2.8b, v24.8b //mul_res1 = vmlsl_u8(mul_res1, src_tmp3, coeffabs_2)// + umlsl v19.8h, v2.8b, v24.8b //mul_res1 = vmlsl_u8(mul_res1, src_tmp3, coeffabs_2)// ld1 {v5.8b},[x3],x2 //src_tmp2 = vld1_u8(pu1_src_tmp)// - umlal v8.8h, v3.8b, v25.8b //mul_res1 = vmlal_u8(mul_res1, src_tmp4, coeffabs_3)// + umlal v19.8h, v3.8b, v25.8b //mul_res1 = vmlal_u8(mul_res1, src_tmp4, coeffabs_3)// ld1 {v6.8b},[x3],x2 //src_tmp3 = vld1_u8(pu1_src_tmp)// - umlal v8.8h, v4.8b, v26.8b //mul_res1 = vmlal_u8(mul_res1, src_tmp1, coeffabs_4)// + umlal v19.8h, v4.8b, v26.8b //mul_res1 = vmlal_u8(mul_res1, src_tmp1, coeffabs_4)// ld1 {v7.8b},[x3],x2 //src_tmp4 = vld1_u8(pu1_src_tmp)// - umlsl v8.8h, v5.8b, v27.8b //mul_res1 = vmlsl_u8(mul_res1, src_tmp2, coeffabs_5)// + umlsl v19.8h, v5.8b, v27.8b //mul_res1 = vmlsl_u8(mul_res1, src_tmp2, coeffabs_5)// ld1 {v16.8b},[x3],x2 //src_tmp1 = vld1_u8(pu1_src_tmp)// - umlal v8.8h, v6.8b, v28.8b //mul_res1 = vmlal_u8(mul_res1, src_tmp3, coeffabs_6)// + umlal v19.8h, v6.8b, v28.8b //mul_res1 = vmlal_u8(mul_res1, src_tmp3, coeffabs_6)// ld1 {v17.8b},[x3],x2 //src_tmp2 = vld1_u8(pu1_src_tmp)// - umlsl v8.8h, v7.8b, v29.8b //mul_res1 = vmlsl_u8(mul_res1, src_tmp4, coeffabs_7)// + umlsl v19.8h, v7.8b, v29.8b //mul_res1 = vmlsl_u8(mul_res1, src_tmp4, coeffabs_7)// ld1 {v18.8b},[x3],x2 //src_tmp3 = vld1_u8(pu1_src_tmp)// - umull v10.8h, v2.8b, v23.8b //mul_res2 = vmull_u8(src_tmp3, coeffabs_1)// + umull v20.8h, v2.8b, v23.8b //mul_res2 = vmull_u8(src_tmp3, coeffabs_1)// add x20,x0,x8 csel x0, x20, x0,le - umlsl v10.8h, v1.8b, v22.8b //mul_res2 = vmlsl_u8(mul_res2, src_tmp2, coeffabs_0)// + umlsl v20.8h, v1.8b, v22.8b //mul_res2 = vmlsl_u8(mul_res2, src_tmp2, coeffabs_0)// bic x20,x5,#7 //x5 ->wd csel x4, x20, x4,le - umlsl v10.8h, v3.8b, v24.8b //mul_res2 = vmlsl_u8(mul_res2, src_tmp4, coeffabs_2)// + umlsl v20.8h, v3.8b, v24.8b //mul_res2 = vmlsl_u8(mul_res2, src_tmp4, coeffabs_2)// prfm PLDL1KEEP,[x3] - umlal v10.8h, v4.8b, v25.8b //mul_res2 = vmlal_u8(mul_res2, src_tmp1, coeffabs_3)// + umlal v20.8h, v4.8b, v25.8b //mul_res2 = vmlal_u8(mul_res2, src_tmp1, coeffabs_3)// add x20,x3, x2 prfm PLDL1KEEP,[x20] - umlal v10.8h, v5.8b, v26.8b //mul_res2 = vmlal_u8(mul_res2, src_tmp2, coeffabs_4)// + umlal v20.8h, v5.8b, v26.8b //mul_res2 = vmlal_u8(mul_res2, src_tmp2, coeffabs_4)// add x20,x3, x2, lsl #1 prfm PLDL1KEEP,[x20] - umlsl v10.8h, v6.8b, v27.8b //mul_res2 = vmlsl_u8(mul_res2, src_tmp3, coeffabs_5)// + umlsl v20.8h, v6.8b, v27.8b //mul_res2 = vmlsl_u8(mul_res2, src_tmp3, coeffabs_5)// add x3, x3, x2 - umlal v10.8h, v7.8b, v28.8b //mul_res2 = vmlal_u8(mul_res2, src_tmp4, coeffabs_6)// + umlal v20.8h, v7.8b, v28.8b //mul_res2 = vmlal_u8(mul_res2, src_tmp4, coeffabs_6)// add x20,x3, x2, lsl #1 prfm PLDL1KEEP,[x20] - umlsl v10.8h, v16.8b, v29.8b //mul_res2 = vmlsl_u8(mul_res2, src_tmp1, coeffabs_7)// + umlsl v20.8h, v16.8b, v29.8b //mul_res2 = vmlsl_u8(mul_res2, src_tmp1, coeffabs_7)// add x3,x0,x2 //pu1_src_tmp += src_strd// - sqrshrun v8.8b, v8.8h,#6 //sto_res = vqmovun_s16(sto_res_tmp)// + sqrshrun v19.8b, v19.8h,#6 //sto_res = vqmovun_s16(sto_res_tmp)// ld1 {v1.8b},[x3],x2 //src_tmp3 = vld1_u8(pu1_src_tmp)// - umull v12.8h, v3.8b, v23.8b + umull v21.8h, v3.8b, v23.8b ld1 {v0.8b},[x0],#8 //src_tmp1 = vld1_u8(pu1_src_tmp)// - umlsl v12.8h, v2.8b, v22.8b + umlsl v21.8h, v2.8b, v22.8b ld1 {v2.8b},[x3],x2 //src_tmp3 = vld1_u8(pu1_src_tmp)// - umlsl v12.8h, v4.8b, v24.8b - umlal v12.8h, v5.8b, v25.8b - umlal v12.8h, v6.8b, v26.8b - umlsl v12.8h, v7.8b, v27.8b - umlal v12.8h, v16.8b, v28.8b - umlsl v12.8h, v17.8b, v29.8b + umlsl v21.8h, v4.8b, v24.8b + umlal v21.8h, v5.8b, v25.8b + umlal v21.8h, v6.8b, v26.8b + umlsl v21.8h, v7.8b, v27.8b + umlal v21.8h, v16.8b, v28.8b + umlsl v21.8h, v17.8b, v29.8b add x14,x1,x6 - st1 {v8.8b},[x1],#8 //vst1_u8(pu1_dst,sto_res)// - sqrshrun v10.8b, v10.8h,#6 //sto_res = vqmovun_s16(sto_res_tmp)// + st1 {v19.8b},[x1],#8 //vst1_u8(pu1_dst,sto_res)// + sqrshrun v20.8b, v20.8h,#6 //sto_res = vqmovun_s16(sto_res_tmp)// add x20,x1,x9 csel x1, x20, x1,le - umull v14.8h, v4.8b, v23.8b + umull v30.8h, v4.8b, v23.8b subs x7,x7,#4 - umlsl v14.8h, v3.8b, v22.8b - umlsl v14.8h, v5.8b, v24.8b - umlal v14.8h, v6.8b, v25.8b + umlsl v30.8h, v3.8b, v22.8b + umlsl v30.8h, v5.8b, v24.8b + umlal v30.8h, v6.8b, v25.8b ld1 {v3.8b},[x3],x2 //src_tmp4 = vld1_u8(pu1_src_tmp)// - umlal v14.8h, v7.8b, v26.8b + umlal v30.8h, v7.8b, v26.8b ld1 {v4.8b},[x3],x2 //src_tmp1 = vld1_u8(pu1_src_tmp)// - umlsl v14.8h, v16.8b, v27.8b + umlsl v30.8h, v16.8b, v27.8b ld1 {v5.8b},[x3],x2 //src_tmp2 = vld1_u8(pu1_src_tmp)// - umlal v14.8h, v17.8b, v28.8b + umlal v30.8h, v17.8b, v28.8b ld1 {v6.8b},[x3],x2 //src_tmp3 = vld1_u8(pu1_src_tmp)// - umlsl v14.8h, v18.8b, v29.8b + umlsl v30.8h, v18.8b, v29.8b ld1 {v7.8b},[x3],x2 //src_tmp4 = vld1_u8(pu1_src_tmp)// - st1 {v10.8b},[x14],x6 //vst1_u8(pu1_dst_tmp,sto_res)// - sqrshrun v12.8b, v12.8h,#6 + st1 {v20.8b},[x14],x6 //vst1_u8(pu1_dst_tmp,sto_res)// + sqrshrun v21.8b, v21.8h,#6 blt epilog_end //jumps to epilog_end @@ -250,111 +250,111 @@ prolog: kernel_8: subs x4,x4,#8 - umull v8.8h, v1.8b, v23.8b //mul_res1 = vmull_u8(src_tmp2, coeffabs_1)// + umull v19.8h, v1.8b, v23.8b //mul_res1 = vmull_u8(src_tmp2, coeffabs_1)// add x20,x0,x8 csel x0, x20, x0,le - umlsl v8.8h, v0.8b, v22.8b //mul_res1 = vmlsl_u8(mul_res1, src_tmp1, coeffabs_0)// + umlsl v19.8h, v0.8b, v22.8b //mul_res1 = vmlsl_u8(mul_res1, src_tmp1, coeffabs_0)// bic x20,x5,#7 //x5 ->wd csel x4, x20, x4,le - umlsl v8.8h, v2.8b, v24.8b //mul_res1 = vmlsl_u8(mul_res1, src_tmp3, coeffabs_2)// + umlsl v19.8h, v2.8b, v24.8b //mul_res1 = vmlsl_u8(mul_res1, src_tmp3, coeffabs_2)// ld1 {v16.8b},[x3],x2 //src_tmp1 = vld1_u8(pu1_src_tmp)// - umlal v8.8h, v3.8b, v25.8b //mul_res1 = vmlal_u8(mul_res1, src_tmp4, coeffabs_3)// + umlal v19.8h, v3.8b, v25.8b //mul_res1 = vmlal_u8(mul_res1, src_tmp4, coeffabs_3)// ld1 {v17.8b},[x3],x2 //src_tmp2 = vld1_u8(pu1_src_tmp)// - umlal v8.8h, v4.8b, v26.8b //mul_res1 = vmlal_u8(mul_res1, src_tmp1, coeffabs_4)// + umlal v19.8h, v4.8b, v26.8b //mul_res1 = vmlal_u8(mul_res1, src_tmp1, coeffabs_4)// ld1 {v18.8b},[x3],x2 //src_tmp3 = vld1_u8(pu1_src_tmp)// - umlsl v8.8h, v5.8b, v27.8b //mul_res1 = vmlsl_u8(mul_res1, src_tmp2, coeffabs_5)// + umlsl v19.8h, v5.8b, v27.8b //mul_res1 = vmlsl_u8(mul_res1, src_tmp2, coeffabs_5)// - umlal v8.8h, v6.8b, v28.8b //mul_res1 = vmlal_u8(mul_res1, src_tmp3, coeffabs_6)// + umlal v19.8h, v6.8b, v28.8b //mul_res1 = vmlal_u8(mul_res1, src_tmp3, coeffabs_6)// - umlsl v8.8h, v7.8b, v29.8b //mul_res1 = vmlsl_u8(mul_res1, src_tmp4, coeffabs_7)// - st1 {v12.8b},[x14],x6 + umlsl v19.8h, v7.8b, v29.8b //mul_res1 = vmlsl_u8(mul_res1, src_tmp4, coeffabs_7)// + st1 {v21.8b},[x14],x6 // and x11, x0, #31 - sqrshrun v14.8b, v14.8h,#6 + sqrshrun v30.8b, v30.8h,#6 add x3,x0,x2 //pu1_src_tmp += src_strd// - umull v10.8h, v2.8b, v23.8b //mul_res2 = vmull_u8(src_tmp3, coeffabs_1)// + umull v20.8h, v2.8b, v23.8b //mul_res2 = vmull_u8(src_tmp3, coeffabs_1)// ld1 {v0.8b},[x0],#8 //src_tmp1 = vld1_u8(pu1_src_tmp)// - umlsl v10.8h, v1.8b, v22.8b //mul_res2 = vmlsl_u8(mul_res2, src_tmp2, coeffabs_0)// + umlsl v20.8h, v1.8b, v22.8b //mul_res2 = vmlsl_u8(mul_res2, src_tmp2, coeffabs_0)// - umlsl v10.8h, v3.8b, v24.8b //mul_res2 = vmlsl_u8(mul_res2, src_tmp4, coeffabs_2)// + umlsl v20.8h, v3.8b, v24.8b //mul_res2 = vmlsl_u8(mul_res2, src_tmp4, coeffabs_2)// ld1 {v1.8b},[x3],x2 //src_tmp2 = vld1_u8(pu1_src_tmp)// - umlal v10.8h, v4.8b, v25.8b //mul_res2 = vmlal_u8(mul_res2, src_tmp1, coeffabs_3)// + umlal v20.8h, v4.8b, v25.8b //mul_res2 = vmlal_u8(mul_res2, src_tmp1, coeffabs_3)// - st1 {v14.8b},[x14],x6 - umlal v10.8h, v5.8b, v26.8b //mul_res2 = vmlal_u8(mul_res2, src_tmp2, coeffabs_4)// + st1 {v30.8b},[x14],x6 + umlal v20.8h, v5.8b, v26.8b //mul_res2 = vmlal_u8(mul_res2, src_tmp2, coeffabs_4)// add x14,x1,#0 - umlsl v10.8h, v6.8b, v27.8b //mul_res2 = vmlsl_u8(mul_res2, src_tmp3, coeffabs_5)// + umlsl v20.8h, v6.8b, v27.8b //mul_res2 = vmlsl_u8(mul_res2, src_tmp3, coeffabs_5)// add x1, x1, #8 - umlal v10.8h, v7.8b, v28.8b //mul_res2 = vmlal_u8(mul_res2, src_tmp4, coeffabs_6)// + umlal v20.8h, v7.8b, v28.8b //mul_res2 = vmlal_u8(mul_res2, src_tmp4, coeffabs_6)// - umlsl v10.8h, v16.8b, v29.8b //mul_res2 = vmlsl_u8(mul_res2, src_tmp1, coeffabs_7)// + umlsl v20.8h, v16.8b, v29.8b //mul_res2 = vmlsl_u8(mul_res2, src_tmp1, coeffabs_7)// add x20,x1,x9 csel x1, x20, x1,le - sqrshrun v8.8b, v8.8h,#6 //sto_res = vqmovun_s16(sto_res_tmp)// + sqrshrun v19.8b, v19.8h,#6 //sto_res = vqmovun_s16(sto_res_tmp)// // cmp x11, x10 - umull v12.8h, v3.8b, v23.8b + umull v21.8h, v3.8b, v23.8b add x10, x3, x2, lsl #3 // 10*strd - 8+2 - umlsl v12.8h, v2.8b, v22.8b + umlsl v21.8h, v2.8b, v22.8b add x10, x10, x2 // 11*strd - umlsl v12.8h, v4.8b, v24.8b + umlsl v21.8h, v4.8b, v24.8b ld1 {v2.8b},[x3],x2 //src_tmp3 = vld1_u8(pu1_src_tmp)// - umlal v12.8h, v5.8b, v25.8b + umlal v21.8h, v5.8b, v25.8b - umlal v12.8h, v6.8b, v26.8b - st1 {v8.8b},[x14],x6 //vst1_u8(pu1_dst,sto_res)// + umlal v21.8h, v6.8b, v26.8b + st1 {v19.8b},[x14],x6 //vst1_u8(pu1_dst,sto_res)// prfm PLDL1KEEP,[x10] //11+ 0 - umlsl v12.8h, v7.8b, v27.8b + umlsl v21.8h, v7.8b, v27.8b add x20,x10, x2 prfm PLDL1KEEP,[x20] //11+ 1*strd - umlal v12.8h, v16.8b, v28.8b + umlal v21.8h, v16.8b, v28.8b add x20,x10, x2, lsl #1 prfm PLDL1KEEP,[x20] //11+ 2*strd - umlsl v12.8h, v17.8b, v29.8b + umlsl v21.8h, v17.8b, v29.8b add x10, x10, x2 //12*strd - sqrshrun v10.8b, v10.8h,#6 //sto_res = vqmovun_s16(sto_res_tmp)// + sqrshrun v20.8b, v20.8h,#6 //sto_res = vqmovun_s16(sto_res_tmp)// add x20,x10, x2, lsl #1 prfm PLDL1KEEP,[x20] //11+ 3*strd - umull v14.8h, v4.8b, v23.8b + umull v30.8h, v4.8b, v23.8b // mov x10, x11 - umlsl v14.8h, v3.8b, v22.8b + umlsl v30.8h, v3.8b, v22.8b subs x7,x7,#4 - umlsl v14.8h, v5.8b, v24.8b + umlsl v30.8h, v5.8b, v24.8b - umlal v14.8h, v6.8b, v25.8b + umlal v30.8h, v6.8b, v25.8b ld1 {v3.8b},[x3],x2 //src_tmp4 = vld1_u8(pu1_src_tmp)// - umlal v14.8h, v7.8b, v26.8b + umlal v30.8h, v7.8b, v26.8b ld1 {v4.8b},[x3],x2 //src_tmp1 = vld1_u8(pu1_src_tmp)// - umlsl v14.8h, v16.8b, v27.8b + umlsl v30.8h, v16.8b, v27.8b ld1 {v5.8b},[x3],x2 //src_tmp2 = vld1_u8(pu1_src_tmp)// - umlal v14.8h, v17.8b, v28.8b + umlal v30.8h, v17.8b, v28.8b ld1 {v6.8b},[x3],x2 //src_tmp3 = vld1_u8(pu1_src_tmp)// - umlsl v14.8h, v18.8b, v29.8b + umlsl v30.8h, v18.8b, v29.8b ld1 {v7.8b},[x3],x2 //src_tmp4 = vld1_u8(pu1_src_tmp)// - sqrshrun v12.8b, v12.8h,#6 - st1 {v10.8b},[x14],x6 //vst1_u8(pu1_dst_tmp,sto_res)// + sqrshrun v21.8b, v21.8h,#6 + st1 {v20.8b},[x14],x6 //vst1_u8(pu1_dst_tmp,sto_res)// @@ -362,62 +362,62 @@ kernel_8: epilog: - umull v8.8h, v1.8b, v23.8b //mul_res1 = vmull_u8(src_tmp2, coeffabs_1)// - umlsl v8.8h, v0.8b, v22.8b //mul_res1 = vmlsl_u8(mul_res1, src_tmp1, coeffabs_0)// - umlsl v8.8h, v2.8b, v24.8b //mul_res1 = vmlsl_u8(mul_res1, src_tmp3, coeffabs_2)// - umlal v8.8h, v3.8b, v25.8b //mul_res1 = vmlal_u8(mul_res1, src_tmp4, coeffabs_3)// - umlal v8.8h, v4.8b, v26.8b //mul_res1 = vmlal_u8(mul_res1, src_tmp1, coeffabs_4)// - umlsl v8.8h, v5.8b, v27.8b //mul_res1 = vmlsl_u8(mul_res1, src_tmp2, coeffabs_5)// - umlal v8.8h, v6.8b, v28.8b //mul_res1 = vmlal_u8(mul_res1, src_tmp3, coeffabs_6)// - umlsl v8.8h, v7.8b, v29.8b //mul_res1 = vmlsl_u8(mul_res1, src_tmp4, coeffabs_7)// - st1 {v12.8b},[x14],x6 + umull v19.8h, v1.8b, v23.8b //mul_res1 = vmull_u8(src_tmp2, coeffabs_1)// + umlsl v19.8h, v0.8b, v22.8b //mul_res1 = vmlsl_u8(mul_res1, src_tmp1, coeffabs_0)// + umlsl v19.8h, v2.8b, v24.8b //mul_res1 = vmlsl_u8(mul_res1, src_tmp3, coeffabs_2)// + umlal v19.8h, v3.8b, v25.8b //mul_res1 = vmlal_u8(mul_res1, src_tmp4, coeffabs_3)// + umlal v19.8h, v4.8b, v26.8b //mul_res1 = vmlal_u8(mul_res1, src_tmp1, coeffabs_4)// + umlsl v19.8h, v5.8b, v27.8b //mul_res1 = vmlsl_u8(mul_res1, src_tmp2, coeffabs_5)// + umlal v19.8h, v6.8b, v28.8b //mul_res1 = vmlal_u8(mul_res1, src_tmp3, coeffabs_6)// + umlsl v19.8h, v7.8b, v29.8b //mul_res1 = vmlsl_u8(mul_res1, src_tmp4, coeffabs_7)// + st1 {v21.8b},[x14],x6 - sqrshrun v14.8b, v14.8h,#6 + sqrshrun v30.8b, v30.8h,#6 ld1 {v16.8b},[x3],x2 //src_tmp1 = vld1_u8(pu1_src_tmp)// - umull v10.8h, v2.8b, v23.8b //mul_res2 = vmull_u8(src_tmp3, coeffabs_1)// - umlsl v10.8h, v1.8b, v22.8b //mul_res2 = vmlsl_u8(mul_res2, src_tmp2, coeffabs_0)// - umlsl v10.8h, v3.8b, v24.8b //mul_res2 = vmlsl_u8(mul_res2, src_tmp4, coeffabs_2)// - umlal v10.8h, v4.8b, v25.8b //mul_res2 = vmlal_u8(mul_res2, src_tmp1, coeffabs_3)// - umlal v10.8h, v5.8b, v26.8b //mul_res2 = vmlal_u8(mul_res2, src_tmp2, coeffabs_4)// - umlsl v10.8h, v6.8b, v27.8b //mul_res2 = vmlsl_u8(mul_res2, src_tmp3, coeffabs_5)// - umlal v10.8h, v7.8b, v28.8b //mul_res2 = vmlal_u8(mul_res2, src_tmp4, coeffabs_6)// - umlsl v10.8h, v16.8b, v29.8b //mul_res2 = vmlsl_u8(mul_res2, src_tmp1, coeffabs_7)// - st1 {v14.8b},[x14],x6 - - sqrshrun v8.8b, v8.8h,#6 //sto_res = vqmovun_s16(sto_res_tmp)// + umull v20.8h, v2.8b, v23.8b //mul_res2 = vmull_u8(src_tmp3, coeffabs_1)// + umlsl v20.8h, v1.8b, v22.8b //mul_res2 = vmlsl_u8(mul_res2, src_tmp2, coeffabs_0)// + umlsl v20.8h, v3.8b, v24.8b //mul_res2 = vmlsl_u8(mul_res2, src_tmp4, coeffabs_2)// + umlal v20.8h, v4.8b, v25.8b //mul_res2 = vmlal_u8(mul_res2, src_tmp1, coeffabs_3)// + umlal v20.8h, v5.8b, v26.8b //mul_res2 = vmlal_u8(mul_res2, src_tmp2, coeffabs_4)// + umlsl v20.8h, v6.8b, v27.8b //mul_res2 = vmlsl_u8(mul_res2, src_tmp3, coeffabs_5)// + umlal v20.8h, v7.8b, v28.8b //mul_res2 = vmlal_u8(mul_res2, src_tmp4, coeffabs_6)// + umlsl v20.8h, v16.8b, v29.8b //mul_res2 = vmlsl_u8(mul_res2, src_tmp1, coeffabs_7)// + st1 {v30.8b},[x14],x6 + + sqrshrun v19.8b, v19.8h,#6 //sto_res = vqmovun_s16(sto_res_tmp)// ld1 {v17.8b},[x3],x2 //src_tmp2 = vld1_u8(pu1_src_tmp)// - umull v12.8h, v3.8b, v23.8b - umlsl v12.8h, v2.8b, v22.8b - umlsl v12.8h, v4.8b, v24.8b - umlal v12.8h, v5.8b, v25.8b - umlal v12.8h, v6.8b, v26.8b - umlsl v12.8h, v7.8b, v27.8b - umlal v12.8h, v16.8b, v28.8b - umlsl v12.8h, v17.8b, v29.8b + umull v21.8h, v3.8b, v23.8b + umlsl v21.8h, v2.8b, v22.8b + umlsl v21.8h, v4.8b, v24.8b + umlal v21.8h, v5.8b, v25.8b + umlal v21.8h, v6.8b, v26.8b + umlsl v21.8h, v7.8b, v27.8b + umlal v21.8h, v16.8b, v28.8b + umlsl v21.8h, v17.8b, v29.8b add x14,x1,x6 - st1 {v8.8b},[x1],#8 //vst1_u8(pu1_dst,sto_res)// - sqrshrun v10.8b, v10.8h,#6 //sto_res = vqmovun_s16(sto_res_tmp)// + st1 {v19.8b},[x1],#8 //vst1_u8(pu1_dst,sto_res)// + sqrshrun v20.8b, v20.8h,#6 //sto_res = vqmovun_s16(sto_res_tmp)// ld1 {v18.8b},[x3],x2 //src_tmp3 = vld1_u8(pu1_src_tmp)// - umull v14.8h, v4.8b, v23.8b - umlsl v14.8h, v3.8b, v22.8b - umlsl v14.8h, v5.8b, v24.8b - umlal v14.8h, v6.8b, v25.8b - umlal v14.8h, v7.8b, v26.8b - umlsl v14.8h, v16.8b, v27.8b - umlal v14.8h, v17.8b, v28.8b - umlsl v14.8h, v18.8b, v29.8b - - st1 {v10.8b},[x14],x6 //vst1_u8(pu1_dst_tmp,sto_res)// - sqrshrun v12.8b, v12.8h,#6 + umull v30.8h, v4.8b, v23.8b + umlsl v30.8h, v3.8b, v22.8b + umlsl v30.8h, v5.8b, v24.8b + umlal v30.8h, v6.8b, v25.8b + umlal v30.8h, v7.8b, v26.8b + umlsl v30.8h, v16.8b, v27.8b + umlal v30.8h, v17.8b, v28.8b + umlsl v30.8h, v18.8b, v29.8b + + st1 {v20.8b},[x14],x6 //vst1_u8(pu1_dst_tmp,sto_res)// + sqrshrun v21.8b, v21.8h,#6 epilog_end: - st1 {v12.8b},[x14],x6 - sqrshrun v14.8b, v14.8h,#6 + st1 {v21.8b},[x14],x6 + sqrshrun v30.8b, v30.8h,#6 - st1 {v14.8b},[x14],x6 + st1 {v30.8b},[x14],x6 end_loops: @@ -427,7 +427,7 @@ end_loops: // ldmeqfd sp!,{x4-x12,x15} //reload the registers from sp bne lbl409 ldp x19, x20,[sp], #16 - pop_v_regs + ret lbl409: mov x5, #4 @@ -465,34 +465,34 @@ inner_loop_wd_4: ld1 {v7.s}[1],[x3],x2 //src_tmp4 = vld1_lane_u32((uint32_t *)pu1_src_tmp, src_tmp4, 1)// umlsl v0.8h, v6.8b, v24.8b //mul_res1 = vmlsl_u8(mul_res1, vreinterpret_u8_u32(src_tmp3), coeffabs_2)// - umull v8.8h, v7.8b, v23.8b + umull v19.8h, v7.8b, v23.8b dup v4.2s, v7.2s[1] //src_tmp1 = vdup_lane_u32(src_tmp4, 1)// umull v2.8h, v7.8b, v25.8b //mul_res2 = vmull_u8(vreinterpret_u8_u32(src_tmp4), coeffabs_3)// ld1 {v4.s}[1],[x3],x2 //src_tmp1 = vld1_lane_u32((uint32_t *)pu1_src_tmp, src_tmp1, 1)// - umlsl v8.8h, v6.8b, v22.8b + umlsl v19.8h, v6.8b, v22.8b umlal v0.8h, v4.8b, v26.8b //mul_res1 = vmlal_u8(mul_res1, vreinterpret_u8_u32(src_tmp1), coeffabs_4)// dup v5.2s, v4.2s[1] //src_tmp2 = vdup_lane_u32(src_tmp1, 1)// - umlsl v8.8h, v4.8b, v24.8b + umlsl v19.8h, v4.8b, v24.8b ld1 {v5.s}[1],[x3],x2 //src_tmp2 = vld1_lane_u32((uint32_t *)pu1_src_tmp, src_tmp2, 1)// umlsl v2.8h, v5.8b, v27.8b //mul_res2 = vmlsl_u8(mul_res2, vreinterpret_u8_u32(src_tmp2), coeffabs_5)// dup v6.2s, v5.2s[1] //src_tmp3 = vdup_lane_u32(src_tmp2, 1)// - umlal v8.8h, v5.8b, v25.8b + umlal v19.8h, v5.8b, v25.8b ld1 {v6.s}[1],[x3],x2 //src_tmp3 = vld1_lane_u32((uint32_t *)pu1_src_tmp, src_tmp3, 1)// umlal v0.8h, v6.8b, v28.8b //mul_res1 = vmlal_u8(mul_res1, vreinterpret_u8_u32(src_tmp3), coeffabs_6)// dup v7.2s, v6.2s[1] //src_tmp4 = vdup_lane_u32(src_tmp3, 1)// - umlal v8.8h, v6.8b, v26.8b + umlal v19.8h, v6.8b, v26.8b ld1 {v7.s}[1],[x3],x2 //src_tmp4 = vld1_lane_u32((uint32_t *)pu1_src_tmp, src_tmp4, 1)// umlsl v2.8h, v7.8b, v29.8b //mul_res2 = vmlsl_u8(mul_res2, vreinterpret_u8_u32(src_tmp4), coeffabs_7)// dup v4.2s, v7.2s[1] add v0.8h, v0.8h , v2.8h //mul_res1 = vaddq_u16(mul_res1, mul_res2)// - umlsl v8.8h, v7.8b, v27.8b + umlsl v19.8h, v7.8b, v27.8b ld1 {v4.s}[1],[x3],x2 - umlal v8.8h, v4.8b, v28.8b + umlal v19.8h, v4.8b, v28.8b dup v5.2s, v4.2s[1] sqrshrun v0.8b, v0.8h,#6 //sto_res = vqmovun_s16(sto_res_tmp)// @@ -500,13 +500,13 @@ inner_loop_wd_4: add x3,x1,x6 st1 {v0.s}[0],[x1] //vst1_lane_u32((uint32_t *)pu1_dst, vreinterpret_u32_u8(sto_res), 0)// - umlsl v8.8h, v5.8b, v29.8b + umlsl v19.8h, v5.8b, v29.8b st1 {v0.s}[1],[x3],x6 //vst1_lane_u32((uint32_t *)pu1_dst_tmp, vreinterpret_u32_u8(sto_res), 1)// - sqrshrun v8.8b, v8.8h,#6 + sqrshrun v19.8b, v19.8h,#6 - st1 {v8.s}[0],[x3],x6 + st1 {v19.s}[0],[x3],x6 add x1,x1,#4 - st1 {v8.s}[1],[x3] + st1 {v19.s}[1],[x3] bgt inner_loop_wd_4 end_inner_loop_wd_4: @@ -517,6 +517,6 @@ end_inner_loop_wd_4: // ldmfd sp!, {x4-x12, x15} //reload the registers from sp ldp x19, x20,[sp], #16 - pop_v_regs + ret |