| OLD | NEW |
| (Empty) |
| 1 ; | |
| 2 ; Copyright (c) 2010 The WebM project authors. All Rights Reserved. | |
| 3 ; | |
| 4 ; Use of this source code is governed by a BSD-style license | |
| 5 ; that can be found in the LICENSE file in the root of the source | |
| 6 ; tree. An additional intellectual property rights grant can be found | |
| 7 ; in the file PATENTS. All contributing project authors may | |
| 8 ; be found in the AUTHORS file in the root of the source tree. | |
| 9 ; | |
| 10 | |
| 11 | |
| 12 EXPORT |vpx_yv12_copy_y_neon| | |
| 13 | |
| 14 ARM | |
| 15 REQUIRE8 | |
| 16 PRESERVE8 | |
| 17 | |
| 18 INCLUDE vpx_scale_asm_offsets.asm | |
| 19 | |
| 20 AREA ||.text||, CODE, READONLY, ALIGN=2 | |
| 21 | |
| 22 ;void vpx_yv12_copy_y_neon(const YV12_BUFFER_CONFIG *src_ybc, | |
| 23 ; YV12_BUFFER_CONFIG *dst_ybc) | |
| 24 |vpx_yv12_copy_y_neon| PROC | |
| 25 push {r4 - r11, lr} | |
| 26 vpush {d8-d15} | |
| 27 | |
| 28 ldr r4, [r0, #yv12_buffer_config_y_height] | |
| 29 ldr r5, [r0, #yv12_buffer_config_y_width] | |
| 30 ldr r6, [r0, #yv12_buffer_config_y_stride] | |
| 31 ldr r7, [r1, #yv12_buffer_config_y_stride] | |
| 32 ldr r2, [r0, #yv12_buffer_config_y_buffer] ;srcptr1 | |
| 33 ldr r3, [r1, #yv12_buffer_config_y_buffer] ;dstptr1 | |
| 34 | |
| 35 ; copy two rows at one time | |
| 36 mov lr, r4, lsr #1 | |
| 37 | |
| 38 cp_src_to_dst_height_loop1 | |
| 39 mov r8, r2 | |
| 40 mov r9, r3 | |
| 41 add r10, r2, r6 | |
| 42 add r11, r3, r7 | |
| 43 movs r12, r5, lsr #7 | |
| 44 ble extra_copy_needed ; y_width < 128 | |
| 45 | |
| 46 cp_src_to_dst_width_loop1 | |
| 47 vld1.8 {q0, q1}, [r8]! | |
| 48 vld1.8 {q8, q9}, [r10]! | |
| 49 vld1.8 {q2, q3}, [r8]! | |
| 50 vld1.8 {q10, q11}, [r10]! | |
| 51 vld1.8 {q4, q5}, [r8]! | |
| 52 vld1.8 {q12, q13}, [r10]! | |
| 53 vld1.8 {q6, q7}, [r8]! | |
| 54 vld1.8 {q14, q15}, [r10]! | |
| 55 | |
| 56 subs r12, r12, #1 | |
| 57 | |
| 58 vst1.8 {q0, q1}, [r9]! | |
| 59 vst1.8 {q8, q9}, [r11]! | |
| 60 vst1.8 {q2, q3}, [r9]! | |
| 61 vst1.8 {q10, q11}, [r11]! | |
| 62 vst1.8 {q4, q5}, [r9]! | |
| 63 vst1.8 {q12, q13}, [r11]! | |
| 64 vst1.8 {q6, q7}, [r9]! | |
| 65 vst1.8 {q14, q15}, [r11]! | |
| 66 | |
| 67 bne cp_src_to_dst_width_loop1 | |
| 68 | |
| 69 subs lr, lr, #1 | |
| 70 add r2, r2, r6, lsl #1 | |
| 71 add r3, r3, r7, lsl #1 | |
| 72 | |
| 73 bne cp_src_to_dst_height_loop1 | |
| 74 | |
| 75 extra_copy_needed | |
| 76 ands r10, r5, #0x7f ;check to see if extra copy
is needed | |
| 77 sub r11, r5, r10 | |
| 78 ldr r2, [r0, #yv12_buffer_config_y_buffer] ;srcptr1 | |
| 79 ldr r3, [r1, #yv12_buffer_config_y_buffer] ;dstptr1 | |
| 80 bne extra_cp_src_to_dst_width1 | |
| 81 end_of_cp_src_to_dst1 | |
| 82 | |
| 83 vpop {d8 - d15} | |
| 84 pop {r4-r11, pc} | |
| 85 | |
| 86 ;============================= | |
| 87 extra_cp_src_to_dst_width1 | |
| 88 add r2, r2, r11 | |
| 89 add r3, r3, r11 | |
| 90 add r0, r8, r6 | |
| 91 add r11, r9, r7 | |
| 92 | |
| 93 mov lr, r4, lsr #1 | |
| 94 extra_cp_src_to_dst_height_loop1 | |
| 95 mov r8, r2 | |
| 96 mov r9, r3 | |
| 97 add r0, r8, r6 | |
| 98 add r11, r9, r7 | |
| 99 | |
| 100 mov r12, r10 | |
| 101 | |
| 102 extra_cp_src_to_dst_width_loop1 | |
| 103 vld1.8 {q0}, [r8]! | |
| 104 vld1.8 {q1}, [r0]! | |
| 105 | |
| 106 subs r12, r12, #16 | |
| 107 | |
| 108 vst1.8 {q0}, [r9]! | |
| 109 vst1.8 {q1}, [r11]! | |
| 110 bne extra_cp_src_to_dst_width_loop1 | |
| 111 | |
| 112 subs lr, lr, #1 | |
| 113 | |
| 114 add r2, r2, r6, lsl #1 | |
| 115 add r3, r3, r7, lsl #1 | |
| 116 | |
| 117 bne extra_cp_src_to_dst_height_loop1 | |
| 118 | |
| 119 b end_of_cp_src_to_dst1 | |
| 120 | |
| 121 ENDP | |
| 122 | |
| 123 END | |
| OLD | NEW |