Mercurial > libavcodec.hg
comparison armv4l/simple_idct_armv6.S @ 4452:c66326f1f635 libavcodec
optimize IDCT of rows with mostly zero coefficients
| author | mru |
|---|---|
| date | Wed, 31 Jan 2007 23:04:56 +0000 |
| parents | cab2986ffc0b |
| children | 99e960878498 |
comparison
equal
deleted
inserted
replaced
| 4451:9fa2c8a7e4d8 | 4452:c66326f1f635 |
|---|---|
| 88 smlad v1, a3, ip, v1 /* A0 += W4*row[4] + W6*row[6] */ | 88 smlad v1, a3, ip, v1 /* A0 += W4*row[4] + W6*row[6] */ |
| 89 smlsd v4, a3, ip, v4 /* A3 += W4*row[4] - W6*row[6] */ | 89 smlsd v4, a3, ip, v4 /* A3 += W4*row[4] - W6*row[6] */ |
| 90 .endm | 90 .endm |
| 91 | 91 |
| 92 /* | 92 /* |
| 93 Compute partial IDCT of half row. | |
| 94 shift = left-shift amount | |
| 95 a3 = row[2,0] | |
| 96 a4 = row[3,1] | |
| 97 | |
| 98 Output in registers v1--v8 | |
| 99 */ | |
| 100 .macro idct_row4 shift | |
| 101 ldr ip, [pc, #(w42-.-8)] /* ip = W4 | (W2 << 16) */ | |
| 102 ldr lr, [pc, #(w46-.-8)] /* lr = W4 | (W6 << 16) */ | |
| 103 ldr v7, [pc, #(w57-.-8)] /* v7 = W5 | (W7 << 16) */ | |
| 104 mov a2, #(1<<(\shift-1)) | |
| 105 smlad v1, a3, ip, a2 | |
| 106 smlsd v4, a3, ip, a2 | |
| 107 ldr ip, [pc, #(w13-.-8)] /* ip = W1 | (W3 << 16) */ | |
| 108 smlad v2, a3, lr, a2 | |
| 109 smlsd v3, a3, lr, a2 | |
| 110 smusdx fp, a4, v7 /* fp = B3 = W7*row[1] - W5*row[3] */ | |
| 111 smuad v5, a4, ip /* v5 = B0 = W1*row[1] + W3*row[3] */ | |
| 112 pkhtb a3, ip, v7, asr #16 /* a4 = W7 | (W3 << 16) */ | |
| 113 pkhbt a2, ip, v7, lsl #16 /* a2 = W1 | (W5 << 16) */ | |
| 114 smusdx v6, a3, a4 /* v6 = -B1 = W7*row[3] - W3*row[1] */ | |
| 115 smusdx v7, a4, a2 /* v7 = B2 = W5*row[1] - W1*row[3] */ | |
| 116 .endm | |
| 117 | |
| 118 /* | |
| 93 Compute final part of IDCT single row without shift. | 119 Compute final part of IDCT single row without shift. |
| 94 Input in registers v1--v8 | 120 Input in registers v1--v8 |
| 95 Output in registers ip, v1--v3, lr, v5--v7 | 121 Output in registers ip, v1--v3, lr, v5--v7 |
| 96 */ | 122 */ |
| 97 .macro idct_finish | 123 .macro idct_finish |
| 165 a2 = dest | 191 a2 = dest |
| 166 */ | 192 */ |
| 167 .align | 193 .align |
| 168 .func idct_row_armv6 | 194 .func idct_row_armv6 |
| 169 idct_row_armv6: | 195 idct_row_armv6: |
| 170 stmfd sp!, {a2, lr} | 196 str lr, [sp, #-4]! |
| 197 | |
| 198 ldr lr, [a1, #12] /* lr = row[7,5] */ | |
| 199 ldr ip, [a1, #4] /* ip = row[6,4] */ | |
| 200 ldr a4, [a1, #8] /* a4 = row[3,1] */ | |
| 201 ldr a3, [a1] /* a3 = row[2,0] */ | |
| 202 orrs lr, lr, ip | |
| 203 cmpeq lr, a4 | |
| 204 cmpeq lr, a3, lsr #16 | |
| 205 beq 1f | |
| 206 str a2, [sp, #-4]! | |
| 207 cmp lr, #0 | |
| 208 beq 2f | |
| 171 | 209 |
| 172 idct_row ROW_SHIFT | 210 idct_row ROW_SHIFT |
| 173 ldr a2, [sp], #4 | 211 b 3f |
| 212 | |
| 213 2: idct_row4 ROW_SHIFT | |
| 214 | |
| 215 3: ldr a2, [sp], #4 | |
| 174 idct_finish_shift ROW_SHIFT | 216 idct_finish_shift ROW_SHIFT |
| 175 | 217 |
| 176 strh v1, [a2] | 218 strh v1, [a2] |
| 177 strh v2, [a2, #(16*2)] | 219 strh v2, [a2, #(16*2)] |
| 178 strh v3, [a2, #(16*4)] | 220 strh v3, [a2, #(16*4)] |
| 180 strh fp, [a2, #(16*1)] | 222 strh fp, [a2, #(16*1)] |
| 181 strh v7, [a2, #(16*3)] | 223 strh v7, [a2, #(16*3)] |
| 182 strh v6, [a2, #(16*5)] | 224 strh v6, [a2, #(16*5)] |
| 183 strh v5, [a2, #(16*7)] | 225 strh v5, [a2, #(16*7)] |
| 184 | 226 |
| 227 ldr pc, [sp], #4 | |
| 228 | |
| 229 1: mov a3, a3, lsl #3 | |
| 230 strh a3, [a2] | |
| 231 strh a3, [a2, #(16*2)] | |
| 232 strh a3, [a2, #(16*4)] | |
| 233 strh a3, [a2, #(16*6)] | |
| 234 strh a3, [a2, #(16*1)] | |
| 235 strh a3, [a2, #(16*3)] | |
| 236 strh a3, [a2, #(16*5)] | |
| 237 strh a3, [a2, #(16*7)] | |
| 185 ldr pc, [sp], #4 | 238 ldr pc, [sp], #4 |
| 186 .endfunc | 239 .endfunc |
| 187 | 240 |
| 188 /* | 241 /* |
| 189 Compute IDCT of single column, read as row. | 242 Compute IDCT of single column, read as row. |