diff options
Diffstat (limited to 'common/arm64/ihevc_intra_pred_filters_chroma_mode_19_to_25.s')
-rw-r--r-- | common/arm64/ihevc_intra_pred_filters_chroma_mode_19_to_25.s | 61 |
1 files changed, 33 insertions, 28 deletions
diff --git a/common/arm64/ihevc_intra_pred_filters_chroma_mode_19_to_25.s b/common/arm64/ihevc_intra_pred_filters_chroma_mode_19_to_25.s index 3af2da7..261c591 100644 --- a/common/arm64/ihevc_intra_pred_filters_chroma_mode_19_to_25.s +++ b/common/arm64/ihevc_intra_pred_filters_chroma_mode_19_to_25.s @@ -102,8 +102,11 @@ ihevc_intra_pred_chroma_mode_19_to_25_av8: - // stmfd sp!, {x4-x12, x14} //stack stores the values of the arguments - push_v_regs + // stmfd sp!, {x4-x12, x14} //stack stores the values of the arguments + + stp d12,d13,[sp,#-16]! + stp d8,d14,[sp,#-16]! // Storing d14 using { sub sp,sp,#8; str d14,[sp] } is giving bus error. + // d8 is used as dummy register and stored along with d14 using stp. d8 is not used in the function. stp x19, x20,[sp,#-16]! adrp x7, :got:gai4_ihevc_ang_table @@ -264,10 +267,10 @@ prologue: add x10,x8,x9 //(i row)*pu1_ref[ref_main_idx] - ld1 {v8.8b},[x10],x11 //(i row)ref_main_idx + ld1 {v7.8b},[x10],x11 //(i row)ref_main_idx sbfx x9,x14,#8,#8 - ld1 {v9.8b},[x10] //(i row)ref_main_idx_1 + ld1 {v19.8b},[x10] //(i row)ref_main_idx_1 add x12,x8,x9 //(ii)*pu1_ref[ref_main_idx] sbfx x9,x14,#16,#8 @@ -275,10 +278,10 @@ prologue: add x10,x8,x9 //(iii)*pu1_ref[ref_main_idx] ld1 {v12.8b},[x12],x11 //(ii)ref_main_idx - umull v10.8h, v8.8b, v30.8b //(i row)vmull_u8(ref_main_idx, dup_const_32_fract) + umull v23.8h, v7.8b, v30.8b //(i row)vmull_u8(ref_main_idx, dup_const_32_fract) ld1 {v13.8b},[x12] //(ii)ref_main_idx_1 - umlal v10.8h, v9.8b, v31.8b //(i row)vmull_u8(ref_main_idx_1, dup_const_fract) + umlal v23.8h, v19.8b, v31.8b //(i row)vmull_u8(ref_main_idx_1, dup_const_fract) dup v27.8b, v4.8b[2] //(iii) sub v28.8b, v1.8b , v29.8b //(ii)32-fract(dup_const_32_fract) @@ -292,7 +295,7 @@ prologue: umlal v14.8h, v13.8b, v29.8b //(ii)vmull_u8(ref_main_idx_1, dup_const_fract) ld1 {v17.8b},[x10] //(iii)ref_main_idx_1 - rshrn v10.8b, v10.8h,#5 //(i row)shift_res = vrshrn_n_u16(add_res, 5) + rshrn v23.8b, v23.8h,#5 //(i row)shift_res = vrshrn_n_u16(add_res, 5) ld1 {v20.8b},[x12],x11 //(iv)ref_main_idx sub v26.8b, v1.8b , v27.8b //(iii)32-fract(dup_const_32_fract) @@ -306,20 +309,20 @@ prologue: umlal v18.8h, v17.8b, v27.8b //(iii)vmull_u8(ref_main_idx_1, dup_const_fract) // lsl x14,x14,#1 - st1 {v10.8b},[x2],#8 //(i row) + st1 {v23.8b},[x2],#8 //(i row) rshrn v14.8b, v14.8h,#5 //(ii)shift_res = vrshrn_n_u16(add_res, 5) sbfx x9,x14,#0,#8 dup v29.8b, v4.8b[5] //(vi) add x10,x8,x9 //(v)*pu1_ref[ref_main_idx] - ld1 {v8.8b},[x10],x11 //(v)ref_main_idx + ld1 {v7.8b},[x10],x11 //(v)ref_main_idx sub v24.8b, v1.8b , v25.8b //(iv)32-fract(dup_const_32_fract) umull v22.8h, v20.8b, v24.8b //(iv)vmull_u8(ref_main_idx, dup_const_32_fract) sbfx x9,x14,#8,#8 - ld1 {v9.8b},[x10] //(v)ref_main_idx_1 + ld1 {v19.8b},[x10] //(v)ref_main_idx_1 umlal v22.8h, v21.8b, v25.8b //(iv)vmull_u8(ref_main_idx_1, dup_const_fract) st1 {v14.8b},[x0],x3 //(ii) @@ -333,10 +336,10 @@ prologue: add x10,x8,x9 //(vii)*pu1_ref[ref_main_idx] ld1 {v12.8b},[x12],x11 //(vi)ref_main_idx - umull v10.8h, v8.8b, v30.8b //(v)vmull_u8(ref_main_idx, dup_const_32_fract) + umull v23.8h, v7.8b, v30.8b //(v)vmull_u8(ref_main_idx, dup_const_32_fract) ld1 {v13.8b},[x12] //(vi)ref_main_idx_1 - umlal v10.8h, v9.8b, v31.8b //(v)vmull_u8(ref_main_idx_1, dup_const_fract) + umlal v23.8h, v19.8b, v31.8b //(v)vmull_u8(ref_main_idx_1, dup_const_fract) st1 {v18.8b},[x0],x3 //(iii) rshrn v22.8b, v22.8h,#5 //(iv)shift_res = vrshrn_n_u16(add_res, 5) @@ -358,7 +361,7 @@ prologue: cmp x4,#8 // go to end if 4x4 beq end_loops - rshrn v10.8b, v10.8h,#5 //(v)shift_res = vrshrn_n_u16(add_res, 5) + rshrn v23.8b, v23.8h,#5 //(v)shift_res = vrshrn_n_u16(add_res, 5) ld1 {v20.8b},[x12],x11 //(viii)ref_main_idx sub v26.8b, v1.8b , v27.8b //(vii)32-fract(dup_const_32_fract) @@ -372,7 +375,7 @@ prologue: sub x20,x4,#8 csel x4, x20, x4,gt - st1 {v10.8b},[x0],x3 //(v) + st1 {v23.8b},[x0],x3 //(v) rshrn v14.8b, v14.8h,#5 //(vi)shift_res = vrshrn_n_u16(add_res, 5) beq epilogue @@ -393,14 +396,14 @@ kernel_8_rows: subs x4,x4,#8 sbfx x9,x14,#8,#8 - ld1 {v8.8b},[x10],x11 //(i)ref_main_idx + ld1 {v7.8b},[x10],x11 //(i)ref_main_idx sub v24.8b, v1.8b , v25.8b //(viii)32-fract(dup_const_32_fract) add x20,x6,#8 //increment the row value csel x6, x20, x6,le add x12,x8,x9 //(ii)*pu1_ref[ref_main_idx] - ld1 {v9.8b},[x10] //(i)ref_main_idx_1 + ld1 {v19.8b},[x10] //(i)ref_main_idx_1 umull v22.8h, v20.8b, v24.8b //(viii)vmull_u8(ref_main_idx, dup_const_32_fract) ld1 {v5.8b},[x6] //loads the row value @@ -417,10 +420,10 @@ kernel_8_rows: add x10,x8,x9 //(iii)*pu1_ref[ref_main_idx] ld1 {v12.8b},[x12],x11 //(ii)ref_main_idx - umull v10.8h, v8.8b, v30.8b //(i)vmull_u8(ref_main_idx, dup_const_32_fract) + umull v23.8h, v7.8b, v30.8b //(i)vmull_u8(ref_main_idx, dup_const_32_fract) ld1 {v13.8b},[x12] //(ii)ref_main_idx_1 - umlal v10.8h, v9.8b, v31.8b //(i)vmull_u8(ref_main_idx_1, dup_const_fract) + umlal v23.8h, v19.8b, v31.8b //(i)vmull_u8(ref_main_idx_1, dup_const_fract) sbfx x9,x14,#24,#8 csel x4, x5, x4,le //reload nt @@ -439,7 +442,7 @@ kernel_8_rows: umlal v14.8h, v13.8b, v29.8b //(ii)vmull_u8(ref_main_idx_1, dup_const_fract) ld1 {v17.8b},[x10] //(iii)ref_main_idx_1 - rshrn v10.8b, v10.8h,#5 //(i)shift_res = vrshrn_n_u16(add_res, 5) + rshrn v23.8b, v23.8h,#5 //(i)shift_res = vrshrn_n_u16(add_res, 5) dup v25.8b, v4.8b[3] //(iv) smull v2.8h, v5.8b, v0.8b //pos = ((row + 1) * intra_pred_ang) @@ -463,7 +466,7 @@ kernel_8_rows: add x10,x8,x9 //(v)*pu1_ref[ref_main_idx] sbfx x9,x14,#8,#8 - st1 {v10.8b},[x2],#8 //(i) + st1 {v23.8b},[x2],#8 //(i) sub v24.8b, v1.8b , v25.8b //(iv)32-fract(dup_const_32_fract) dup v29.8b, v4.8b[5] //(vi) @@ -478,10 +481,10 @@ kernel_8_rows: dup v25.8b, v4.8b[7] //(viii) rshrn v18.8b, v18.8h,#5 //(iii)shift_res = vrshrn_n_u16(add_res, 5) - ld1 {v8.8b},[x10],x11 //(v)ref_main_idx + ld1 {v7.8b},[x10],x11 //(v)ref_main_idx and v4.16b, v2.16b , v6.16b //dup_const_fract(fract = pos & (31)) - ld1 {v9.8b},[x10] //(v)ref_main_idx_1 + ld1 {v19.8b},[x10] //(v)ref_main_idx_1 shrn v3.8b, v2.8h,#5 //idx = pos >> 5 st1 {v14.8b},[x0],x3 //(ii) @@ -496,10 +499,10 @@ kernel_8_rows: shl v3.8b, v3.8b,#1 ld1 {v13.8b},[x12] //(vi)ref_main_idx_1 - umull v10.8h, v8.8b, v30.8b //(v)vmull_u8(ref_main_idx, dup_const_32_fract) + umull v23.8h, v7.8b, v30.8b //(v)vmull_u8(ref_main_idx, dup_const_32_fract) smov x14, v3.2s[0] //(i)extract idx to the r register - umlal v10.8h, v9.8b, v31.8b //(v)vmull_u8(ref_main_idx_1, dup_const_fract) + umlal v23.8h, v19.8b, v31.8b //(v)vmull_u8(ref_main_idx_1, dup_const_fract) add x12,x8,x9 //(viii)*pu1_ref[ref_main_idx] csel x8, x1, x8,le //reload the source to pu1_src+2nt @@ -514,7 +517,7 @@ kernel_8_rows: umlal v14.8h, v13.8b, v29.8b //(vi)vmull_u8(ref_main_idx_1, dup_const_fract) ld1 {v20.8b},[x12],x11 //(viii)ref_main_idx - rshrn v10.8b, v10.8h,#5 //(v)shift_res = vrshrn_n_u16(add_res, 5) + rshrn v23.8b, v23.8h,#5 //(v)shift_res = vrshrn_n_u16(add_res, 5) ld1 {v21.8b},[x12] //(viii)ref_main_idx_1 sub v26.8b, v1.8b , v27.8b //(vii)32-fract(dup_const_32_fract) @@ -529,7 +532,7 @@ kernel_8_rows: st1 {v22.8b},[x0],x3 //(iv) umull v18.8h, v16.8b, v26.8b //(vii)vmull_u8(ref_main_idx, dup_const_32_fract) - st1 {v10.8b},[x0],x3 //(v) + st1 {v23.8b},[x0],x3 //(v) umlal v18.8h, v17.8b, v27.8b //(vii)vmull_u8(ref_main_idx_1, dup_const_fract) add x20,x2,x12 //increment the dst pointer to 8*dst_strd - nt @@ -563,9 +566,11 @@ core_loop_4: end_loops: add sp, sp, #132 - // ldmfd sp!,{x4-x12,x15} //reload the registers from sp + // ldmfd sp!,{x4-x12,x15} //reload the registers from sp ldp x19, x20,[sp],#16 - pop_v_regs + ldp d8,d14,[sp],#16 // Loading d14 using { ldr d14,[sp]; add sp,sp,#8 } is giving bus error. + // d8 is used as dummy register and loaded along with d14 using ldp. d8 is not used in the function. + ldp d12,d13,[sp],#16 ret |