1 /* 2 * 3 * Copyright 2012 Samsung Electronics S.LSI Co. LTD 4 * 5 * Licensed under the Apache License, Version 2.0 (the "License") 6 * you may not use this file except in compliance with the License. 7 * You may obtain a copy of the License at 8 * 9 * http://www.apache.org/licenses/LICENSE-2.0 10 * 11 * Unless required by applicable law or agreed to in writing, software 12 * distributed under the License is distributed on an "AS IS" BASIS, 13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 14 * See the License for the specific language governing permissions and 15 * limitations under the License. 16 */ 17 18 /* 19 * @file csc_linear_to_tiled_crop_neon.s 20 * @brief SEC_OMX specific define 21 * @author ShinWon Lee (shinwon.lee (at) samsung.com) 22 * @version 1.0 23 * @history 24 * 2012.02.01 : Create 25 */ 26 27 /* 28 * Interleave src1, src2 to dst 29 * 30 * @param dest 31 * dst address[out] 32 * 33 * @param src1 34 * src1 address[in] 35 * 36 * @param src2 37 * src2 address[in] 38 * 39 * @param src_size 40 * src_size or src1 41 */ 42 43 .arch armv7-a 44 .text 45 .global csc_interleave_memcpy_neon 46 .type csc_interleave_memcpy_neon, %function 47 csc_interleave_memcpy_neon: 48 .fnstart 49 50 @r0 dest 51 @r1 src1 52 @r2 src2 53 @r3 src_size 54 @r4 55 @r5 56 @r6 57 @r7 58 @r8 temp1 59 @r9 temp2 60 @r10 dest_addr 61 @r11 src1_addr 62 @r12 src2_addr 63 @r14 i 64 65 stmfd sp!, {r8-r12,r14} @ backup registers 66 67 mov r10, r0 68 mov r11, r1 69 mov r12, r2 70 mov r14, r3 71 72 cmp r14, #128 73 blt LESS_THAN_128 74 75 LOOP_128: 76 vld1.8 {q0}, [r11]! 77 vld1.8 {q2}, [r11]! 78 vld1.8 {q4}, [r11]! 79 vld1.8 {q6}, [r11]! 80 vld1.8 {q8}, [r11]! 81 vld1.8 {q10}, [r11]! 82 vld1.8 {q12}, [r11]! 83 vld1.8 {q14}, [r11]! 84 vld1.8 {q1}, [r12]! 85 vld1.8 {q3}, [r12]! 86 vld1.8 {q5}, [r12]! 87 vld1.8 {q7}, [r12]! 88 vld1.8 {q9}, [r12]! 89 vld1.8 {q11}, [r12]! 90 vld1.8 {q13}, [r12]! 91 vld1.8 {q15}, [r12]! 92 93 vst2.8 {q0, q1}, [r10]! 94 vst2.8 {q2, q3}, [r10]! 95 vst2.8 {q4, q5}, [r10]! 96 vst2.8 {q6, q7}, [r10]! 97 vst2.8 {q8, q9}, [r10]! 98 vst2.8 {q10, q11}, [r10]! 99 vst2.8 {q12, q13}, [r10]! 100 vst2.8 {q14, q15}, [r10]! 101 102 sub r14, #128 103 cmp r14, #128 104 bgt LOOP_128 105 106 LESS_THAN_128: 107 cmp r14, #0 108 beq RESTORE_REG 109 110 LOOP_1: 111 ldrb r8, [r11], #1 112 ldrb r9, [r12], #1 113 strb r8, [r10], #1 114 strb r9, [r10], #1 115 subs r14, #1 116 bne LOOP_1 117 118 RESTORE_REG: 119 ldmfd sp!, {r8-r12,r15} @ restore registers 120 .fnend 121
