cor_h_vec_opt.s revision 41050cdb033641ddf26831d9272c0930f7b40a2d
1@/* 2@ ** Copyright 2003-2010, VisualOn, Inc. 3@ ** 4@ ** Licensed under the Apache License, Version 2.0 (the "License"); 5@ ** you may not use this file except in compliance with the License. 6@ ** You may obtain a copy of the License at 7@ ** 8@ ** http://www.apache.org/licenses/LICENSE-2.0 9@ ** 10@ ** Unless required by applicable law or agreed to in writing, software 11@ ** distributed under the License is distributed on an "AS IS" BASIS, 12@ ** WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 13@ ** See the License for the specific language governing permissions and 14@ ** limitations under the License. 15@ */ 16@static void cor_h_vec_012( 17@ Word16 h[], /* (i) scaled impulse response */ 18@ Word16 vec[], /* (i) scaled vector (/8) to correlate with h[] */ 19@ Word16 track, /* (i) track to use */ 20@ Word16 sign[], /* (i) sign vector */ 21@ Word16 rrixix[][NB_POS], /* (i) correlation of h[x] with h[x] */ 22@ Word16 cor_1[], /* (o) result of correlation (NB_POS elements) */ 23@ Word16 cor_2[] /* (o) result of correlation (NB_POS elements) */ 24@) 25@r0 ---- h[] 26@r1 ---- vec[] 27@r2 ---- track 28@r3 ---- sign[] 29@r4 ---- rrixix[][NB_POS] 30@r5 ---- cor_1[] 31@r6 ---- cor_2[] 32 33 34 .section .text 35 .global cor_h_vec_012_asm 36 37cor_h_vec_012_asm: 38 39 STMFD r13!, {r4 - r12, r14} 40 LDR r4, [r13, #40] @load rrixix[][NB_POS] 41 ADD r7, r4, r2, LSL #5 @r7 --- p0 = rrixix[track] 42 MOV r4, #0 @i=0 43 44 @r0 --- h[], r1 --- vec[], r2 --- pos 45 @r3 --- sign[], r4 --- i, r7 --- p0 46LOOPi: 47 MOV r5, #0 @L_sum1 = 0 48 MOV r6, #0 @L_sum2 = 0 49 ADD r9, r1, r2, LSL #1 @p2 = &vec[pos] 50 MOV r10, r0 @p1 = h 51 RSB r11, r2, #62 @j=62-pos 52 53LOOPj1: 54 LDRSH r12, [r10], #2 55 LDRSH r8, [r9], #2 56 LDRSH r14, [r9] 57 SUBS r11, r11, #1 58 MLA r5, r12, r8, r5 59 MLA r6, r12, r14, r6 60 BGE LOOPj1 61 62 LDRSH r12, [r10], #2 @*p1++ 63 MOV r6, r6, LSL #2 @L_sum2 = (L_sum2 << 2) 64 MLA r5, r12, r14, r5 65 MOV r14, #0x8000 66 MOV r5, r5, LSL #2 @L_sum1 = (L_sum1 << 2) 67 ADD r10, r6, r14 68 ADD r9, r5, r14 69 MOV r5, r9, ASR #16 70 MOV r6, r10, ASR #16 71 ADD r9, r3, r2, LSL #1 @address of sign[pos] 72 ADD r8, r7, #32 73 LDRSH r10, [r9], #2 @sign[pos] 74 LDRSH r11, [r9] @sign[pos + 1] 75 MUL r12, r5, r10 76 MUL r14, r6, r11 77 MOV r5, r12, ASR #15 78 MOV r6, r14, ASR #15 79 LDR r9, [r13, #44] 80 LDR r12, [r13, #48] 81 LDRSH r10, [r7], #2 @*p0++ 82 LDRSH r11, [r8] @*p3++ 83 ADD r9, r9, r4, LSL #1 84 ADD r12, r12, r4, LSL #1 85 ADD r5, r5, r10 86 ADD r6, r6, r11 87 STRH r5, [r9] 88 STRH r6, [r12] 89 90 ADD r2, r2, #4 91 92 MOV r5, #0 @L_sum1 = 0 93 MOV r6, #0 @L_sum2 = 0 94 ADD r9, r1, r2, LSL #1 @p2 = &vec[pos] 95 MOV r10, r0 @p1 = h 96 RSB r11, r2, #62 @j=62-pos 97 ADD r4, r4, #1 @i++ 98 99LOOPj2: 100 LDRSH r12, [r10], #2 101 LDRSH r8, [r9], #2 102 LDRSH r14, [r9] 103 SUBS r11, r11, #1 104 MLA r5, r12, r8, r5 105 MLA r6, r12, r14, r6 106 BGE LOOPj2 107 108 LDRSH r12, [r10], #2 @*p1++ 109 MOV r6, r6, LSL #2 @L_sum2 = (L_sum2 << 2) 110 MLA r5, r12, r14, r5 111 MOV r14, #0x8000 112 MOV r5, r5, LSL #2 @L_sum1 = (L_sum1 << 2) 113 ADD r10, r6, r14 114 ADD r9, r5, r14 115 116 MOV r5, r9, ASR #16 117 MOV r6, r10, ASR #16 118 ADD r9, r3, r2, LSL #1 @address of sign[pos] 119 ADD r8, r7, #32 120 LDRSH r10, [r9], #2 @sign[pos] 121 LDRSH r11, [r9] @sign[pos + 1] 122 MUL r12, r5, r10 123 MUL r14, r6, r11 124 MOV r5, r12, ASR #15 125 MOV r6, r14, ASR #15 126 LDR r9, [r13, #44] 127 LDR r12, [r13, #48] 128 LDRSH r10, [r7], #2 @*p0++ 129 LDRSH r11, [r8] @*p3++ 130 ADD r9, r9, r4, LSL #1 131 ADD r12, r12, r4, LSL #1 132 ADD r5, r5, r10 133 ADD r6, r6, r11 134 STRH r5, [r9] 135 STRH r6, [r12] 136 ADD r4, r4, #1 @i+1 137 ADD r2, r2, #4 @pos += STEP 138 CMP r4, #16 139 140 BLT LOOPi 141 142the_end: 143 LDMFD r13!, {r4 - r12, r15} 144 145 @ENDFUNC 146 .END 147 148 149 150 151 152