diff options
| author | Jens Arnold <amiconn@rockbox.org> | 2008-10-07 19:40:17 +0000 |
|---|---|---|
| committer | Jens Arnold <amiconn@rockbox.org> | 2008-10-07 19:40:17 +0000 |
| commit | 6b84f600466ab02f5a671d5004cc5b13f18460af (patch) | |
| tree | 2dddfb3838ec570cd415e10f0251c3d3f0d0cf10 | |
| parent | 46573019a53dca411f754d40d0f21c1e4eafaedf (diff) | |
| download | rockbox-6b84f600466ab02f5a671d5004cc5b13f18460af.tar.gz rockbox-6b84f600466ab02f5a671d5004cc5b13f18460af.zip | |
APE: Further ARMv6 filter optimisations: Save 4 'ror's per round by utilising the shift feature of the 'pack halfword' instructions in the unaligned vector addition/ subtraction, better pipelining in the aligned scalarproduct(), and a new method to calculate the unaligned scalarproduct().
git-svn-id: svn://svn.rockbox.org/rockbox/trunk@18736 a1c6a512-1295-4272-9138-f99709370657
| -rw-r--r-- | apps/codecs/demac/libdemac/vector_math16_armv6.h | 147 |
1 files changed, 66 insertions, 81 deletions
diff --git a/apps/codecs/demac/libdemac/vector_math16_armv6.h b/apps/codecs/demac/libdemac/vector_math16_armv6.h index bf50d9cabd..e180429193 100644 --- a/apps/codecs/demac/libdemac/vector_math16_armv6.h +++ b/apps/codecs/demac/libdemac/vector_math16_armv6.h | |||
| @@ -39,36 +39,33 @@ static inline void vector_add(int16_t* v1, int16_t* v2) | |||
| 39 | 39 | ||
| 40 | "10: \n" | 40 | "10: \n" |
| 41 | "ldrh r4, [%[v2]], #2 \n" | 41 | "ldrh r4, [%[v2]], #2 \n" |
| 42 | "mov r4, r4, lsl #16 \n" | ||
| 42 | "1: \n" | 43 | "1: \n" |
| 43 | "ldmia %[v2]!, {r5-r8} \n" | 44 | "ldmia %[v2]!, {r5-r8} \n" |
| 44 | "ldmia %[v1], {r0-r3} \n" | 45 | "ldmia %[v1], {r0-r3} \n" |
| 45 | "mov r5, r5, ror #16 \n" | 46 | "mov r5, r5, ror #16 \n" |
| 46 | "pkhbt r4, r4, r5 \n" | 47 | "pkhtb r4, r5, r4, asr #16 \n" |
| 47 | "sadd16 r0, r0, r4 \n" | 48 | "sadd16 r0, r0, r4 \n" |
| 48 | "mov r6, r6, ror #16 \n" | 49 | "pkhbt r5, r5, r6, lsl #16 \n" |
| 49 | "pkhbt r5, r5, r6 \n" | ||
| 50 | "sadd16 r1, r1, r5 \n" | 50 | "sadd16 r1, r1, r5 \n" |
| 51 | "mov r7, r7, ror #16 \n" | 51 | "mov r7, r7, ror #16 \n" |
| 52 | "pkhbt r6, r6, r7 \n" | 52 | "pkhtb r6, r7, r6, asr #16 \n" |
| 53 | "sadd16 r2, r2, r6 \n" | 53 | "sadd16 r2, r2, r6 \n" |
| 54 | "mov r8, r8, ror #16 \n" | 54 | "pkhbt r7, r7, r8, lsl #16 \n" |
| 55 | "pkhbt r7, r7, r8 \n" | ||
| 56 | "sadd16 r3, r3, r7 \n" | 55 | "sadd16 r3, r3, r7 \n" |
| 57 | "stmia %[v1]!, {r0-r3} \n" | 56 | "stmia %[v1]!, {r0-r3} \n" |
| 58 | "mov r4, r8 \n" | 57 | "mov r4, r8 \n" |
| 59 | "ldmia %[v2]!, {r5-r8} \n" | 58 | "ldmia %[v2]!, {r5-r8} \n" |
| 60 | "ldmia %[v1], {r0-r3} \n" | 59 | "ldmia %[v1], {r0-r3} \n" |
| 61 | "mov r5, r5, ror #16 \n" | 60 | "mov r5, r5, ror #16 \n" |
| 62 | "pkhbt r4, r4, r5 \n" | 61 | "pkhtb r4, r5, r4, asr #16 \n" |
| 63 | "sadd16 r0, r0, r4 \n" | 62 | "sadd16 r0, r0, r4 \n" |
| 64 | "mov r6, r6, ror #16 \n" | 63 | "pkhbt r5, r5, r6, lsl #16 \n" |
| 65 | "pkhbt r5, r5, r6 \n" | ||
| 66 | "sadd16 r1, r1, r5 \n" | 64 | "sadd16 r1, r1, r5 \n" |
| 67 | "mov r7, r7, ror #16 \n" | 65 | "mov r7, r7, ror #16 \n" |
| 68 | "pkhbt r6, r6, r7 \n" | 66 | "pkhtb r6, r7, r6, asr #16 \n" |
| 69 | "sadd16 r2, r2, r6 \n" | 67 | "sadd16 r2, r2, r6 \n" |
| 70 | "mov r8, r8, ror #16 \n" | 68 | "pkhbt r7, r7, r8, lsl #16 \n" |
| 71 | "pkhbt r7, r7, r8 \n" | ||
| 72 | "sadd16 r3, r3, r7 \n" | 69 | "sadd16 r3, r3, r7 \n" |
| 73 | "stmia %[v1]!, {r0-r3} \n" | 70 | "stmia %[v1]!, {r0-r3} \n" |
| 74 | #if ORDER > 16 | 71 | #if ORDER > 16 |
| @@ -128,36 +125,33 @@ static inline void vector_sub(int16_t* v1, int16_t* v2) | |||
| 128 | 125 | ||
| 129 | "10: \n" | 126 | "10: \n" |
| 130 | "ldrh r4, [%[v2]], #2 \n" | 127 | "ldrh r4, [%[v2]], #2 \n" |
| 128 | "mov r4, r4, lsl #16 \n" | ||
| 131 | "1: \n" | 129 | "1: \n" |
| 132 | "ldmia %[v2]!, {r5-r8} \n" | 130 | "ldmia %[v2]!, {r5-r8} \n" |
| 133 | "ldmia %[v1], {r0-r3} \n" | 131 | "ldmia %[v1], {r0-r3} \n" |
| 134 | "mov r5, r5, ror #16 \n" | 132 | "mov r5, r5, ror #16 \n" |
| 135 | "pkhbt r4, r4, r5 \n" | 133 | "pkhtb r4, r5, r4, asr #16 \n" |
| 136 | "ssub16 r0, r0, r4 \n" | 134 | "ssub16 r0, r0, r4 \n" |
| 137 | "mov r6, r6, ror #16 \n" | 135 | "pkhbt r5, r5, r6, lsl #16 \n" |
| 138 | "pkhbt r5, r5, r6 \n" | ||
| 139 | "ssub16 r1, r1, r5 \n" | 136 | "ssub16 r1, r1, r5 \n" |
| 140 | "mov r7, r7, ror #16 \n" | 137 | "mov r7, r7, ror #16 \n" |
| 141 | "pkhbt r6, r6, r7 \n" | 138 | "pkhtb r6, r7, r6, asr #16 \n" |
| 142 | "ssub16 r2, r2, r6 \n" | 139 | "ssub16 r2, r2, r6 \n" |
| 143 | "mov r8, r8, ror #16 \n" | 140 | "pkhbt r7, r7, r8, lsl #16 \n" |
| 144 | "pkhbt r7, r7, r8 \n" | ||
| 145 | "ssub16 r3, r3, r7 \n" | 141 | "ssub16 r3, r3, r7 \n" |
| 146 | "stmia %[v1]!, {r0-r3} \n" | 142 | "stmia %[v1]!, {r0-r3} \n" |
| 147 | "mov r4, r8 \n" | 143 | "mov r4, r8 \n" |
| 148 | "ldmia %[v2]!, {r5-r8} \n" | 144 | "ldmia %[v2]!, {r5-r8} \n" |
| 149 | "ldmia %[v1], {r0-r3} \n" | 145 | "ldmia %[v1], {r0-r3} \n" |
| 150 | "mov r5, r5, ror #16 \n" | 146 | "mov r5, r5, ror #16 \n" |
| 151 | "pkhbt r4, r4, r5 \n" | 147 | "pkhtb r4, r5, r4, asr #16 \n" |
| 152 | "ssub16 r0, r0, r4 \n" | 148 | "ssub16 r0, r0, r4 \n" |
| 153 | "mov r6, r6, ror #16 \n" | 149 | "pkhbt r5, r5, r6, lsl #16 \n" |
| 154 | "pkhbt r5, r5, r6 \n" | ||
| 155 | "ssub16 r1, r1, r5 \n" | 150 | "ssub16 r1, r1, r5 \n" |
| 156 | "mov r7, r7, ror #16 \n" | 151 | "mov r7, r7, ror #16 \n" |
| 157 | "pkhbt r6, r6, r7 \n" | 152 | "pkhtb r6, r7, r6, asr #16 \n" |
| 158 | "ssub16 r2, r2, r6 \n" | 153 | "ssub16 r2, r2, r6 \n" |
| 159 | "mov r8, r8, ror #16 \n" | 154 | "pkhbt r7, r7, r8, lsl #16 \n" |
| 160 | "pkhbt r7, r7, r8 \n" | ||
| 161 | "ssub16 r3, r3, r7 \n" | 155 | "ssub16 r3, r3, r7 \n" |
| 162 | "stmia %[v1]!, {r0-r3} \n" | 156 | "stmia %[v1]!, {r0-r3} \n" |
| 163 | #if ORDER > 16 | 157 | #if ORDER > 16 |
| @@ -217,80 +211,71 @@ static inline int32_t scalarproduct(int16_t* v1, int16_t* v2) | |||
| 217 | "beq 20f \n" | 211 | "beq 20f \n" |
| 218 | 212 | ||
| 219 | "10: \n" | 213 | "10: \n" |
| 220 | "ldrh r2, [%[v2]], #2 \n" | 214 | "ldrh r7, [%[v2]], #2 \n" |
| 221 | "ldr r0, [%[v1]], #4 \n" | 215 | "ldmia %[v2]!, {r4-r5} \n" |
| 222 | "ldr r3, [%[v2]], #4 \n" | 216 | "ldmia %[v1]!, {r0-r1} \n" |
| 223 | "mov r2, r2, lsl #16 \n" | 217 | "mov r7, r7, lsl #16 \n" |
| 224 | "1: \n" | 218 | "1: \n" |
| 225 | "ldr r1, [%[v1]], #4 \n" | 219 | "pkhbt r8, r4, r7 \n" |
| 226 | "smlabt %[res], r0, r2, %[res] \n" | 220 | "ldmia %[v2]!, {r6-r7} \n" |
| 227 | "ldr r4, [%[v2]], #4 \n" | 221 | "smladx %[res], r0, r8, %[res] \n" |
| 228 | "smlatb %[res], r0, r3, %[res] \n" | 222 | "pkhbt r8, r5, r4 \n" |
| 229 | "ldr r0, [%[v1]], #4 \n" | 223 | "ldmia %[v1]!, {r2-r3} \n" |
| 230 | "smlabt %[res], r1, r3, %[res] \n" | 224 | "smladx %[res], r1, r8, %[res] \n" |
| 231 | "ldr r5, [%[v2]], #4 \n" | 225 | "pkhbt r8, r6, r5 \n" |
| 232 | "smlatb %[res], r1, r4, %[res] \n" | 226 | "ldmia %[v2]!, {r4-r5} \n" |
| 233 | "ldr r1, [%[v1]], #4 \n" | 227 | "smladx %[res], r2, r8, %[res] \n" |
| 234 | "smlabt %[res], r0, r4, %[res] \n" | 228 | "pkhbt r8, r7, r6 \n" |
| 235 | "ldr r6, [%[v2]], #4 \n" | 229 | "ldmia %[v1]!, {r0-r1} \n" |
| 236 | "smlatb %[res], r0, r5, %[res] \n" | 230 | "smladx %[res], r3, r8, %[res] \n" |
| 237 | "ldr r0, [%[v1]], #4 \n" | 231 | "pkhbt r8, r4, r7 \n" |
| 238 | "smlabt %[res], r1, r5, %[res] \n" | 232 | "ldmia %[v2]!, {r6-r7} \n" |
| 239 | "ldr r3, [%[v2]], #4 \n" | 233 | "smladx %[res], r0, r8, %[res] \n" |
| 240 | "smlatb %[res], r1, r6, %[res] \n" | 234 | "pkhbt r8, r5, r4 \n" |
| 241 | "mov r2, r6 \n" | 235 | "ldmia %[v1]!, {r2-r3} \n" |
| 242 | "ldr r1, [%[v1]], #4 \n" | 236 | "smladx %[res], r1, r8, %[res] \n" |
| 243 | "smlabt %[res], r0, r2, %[res] \n" | 237 | "pkhbt r8, r6, r5 \n" |
| 244 | "ldr r4, [%[v2]], #4 \n" | ||
| 245 | "smlatb %[res], r0, r3, %[res] \n" | ||
| 246 | "ldr r0, [%[v1]], #4 \n" | ||
| 247 | "smlabt %[res], r1, r3, %[res] \n" | ||
| 248 | "ldr r5, [%[v2]], #4 \n" | ||
| 249 | "smlatb %[res], r1, r4, %[res] \n" | ||
| 250 | "ldr r1, [%[v1]], #4 \n" | ||
| 251 | "smlabt %[res], r0, r4, %[res] \n" | ||
| 252 | "ldr r6, [%[v2]], #4 \n" | ||
| 253 | "smlatb %[res], r0, r5, %[res] \n" | ||
| 254 | #if ORDER > 16 | 238 | #if ORDER > 16 |
| 255 | "subs %[cnt], %[cnt], #1 \n" | 239 | "subs %[cnt], %[cnt], #1 \n" |
| 256 | "ldrne r0, [%[v1]], #4 \n" | 240 | "ldmneia %[v2]!, {r4-r5} \n" |
| 257 | "smlabt %[res], r1, r5, %[res] \n" | 241 | "smladx %[res], r2, r8, %[res] \n" |
| 258 | "ldrne r3, [%[v2]], #4 \n" | 242 | "pkhbt r8, r7, r6 \n" |
| 259 | "smlatb %[res], r1, r6, %[res] \n" | 243 | "ldmneia %[v1]!, {r0-r1} \n" |
| 260 | "mov r2, r6 \n" | 244 | "smladx %[res], r3, r8, %[res] \n" |
| 261 | "bne 1b \n" | 245 | "bne 1b \n" |
| 262 | #else | 246 | #else |
| 263 | "smlabt %[res], r1, r5, %[res] \n" | 247 | "pkhbt r7, r7, r6 \n" |
| 264 | "smlatb %[res], r1, r6, %[res] \n" | 248 | "smladx %[res], r2, r8, %[res] \n" |
| 249 | "smladx %[res], r3, r7, %[res] \n" | ||
| 265 | #endif | 250 | #endif |
| 266 | "b 99f \n" | 251 | "b 99f \n" |
| 267 | 252 | ||
| 268 | "20: \n" | 253 | "20: \n" |
| 269 | "ldmia %[v1]!, {r0-r1} \n" | 254 | "ldmia %[v1]!, {r0-r1} \n" |
| 270 | "ldmia %[v2]!, {r4-r5} \n" | 255 | "ldmia %[v2]!, {r5-r7} \n" |
| 271 | "1: \n" | 256 | "1: \n" |
| 272 | "ldmia %[v1]!, {r2-r3} \n" | 257 | "ldmia %[v1]!, {r2-r3} \n" |
| 273 | "smlad %[res], r0, r4, %[res] \n" | 258 | "smlad %[res], r0, r5, %[res] \n" |
| 274 | "ldmia %[v2]!, {r6-r7} \n" | ||
| 275 | "smlad %[res], r1, r5, %[res] \n" | ||
| 276 | "ldmia %[v1]!, {r0-r1} \n" | ||
| 277 | "smlad %[res], r2, r6, %[res] \n" | ||
| 278 | "ldmia %[v2]!, {r4-r5} \n" | 259 | "ldmia %[v2]!, {r4-r5} \n" |
| 279 | "smlad %[res], r3, r7, %[res] \n" | 260 | "smlad %[res], r1, r6, %[res] \n" |
| 280 | "ldmia %[v1]!, {r2-r3} \n" | 261 | "ldmia %[v1]!, {r0-r1} \n" |
| 281 | "smlad %[res], r0, r4, %[res] \n" | 262 | "smlad %[res], r2, r7, %[res] \n" |
| 282 | "ldmia %[v2]!, {r6-r7} \n" | 263 | "ldmia %[v2]!, {r6-r7} \n" |
| 283 | "smlad %[res], r1, r5, %[res] \n" | 264 | "smlad %[res], r3, r4, %[res] \n" |
| 265 | "ldmia %[v1]!, {r2-r3} \n" | ||
| 266 | "smlad %[res], r0, r5, %[res] \n" | ||
| 267 | "ldmia %[v2]!, {r4-r5} \n" | ||
| 268 | "smlad %[res], r1, r6, %[res] \n" | ||
| 284 | #if ORDER > 16 | 269 | #if ORDER > 16 |
| 285 | "subs %[cnt], %[cnt], #1 \n" | 270 | "subs %[cnt], %[cnt], #1 \n" |
| 286 | "ldmneia %[v1]!, {r0-r1} \n" | 271 | "ldmneia %[v1]!, {r0-r1} \n" |
| 287 | "smlad %[res], r2, r6, %[res] \n" | 272 | "smlad %[res], r2, r7, %[res] \n" |
| 288 | "ldmneia %[v2]!, {r4-r5} \n" | 273 | "ldmneia %[v2]!, {r6-r7} \n" |
| 289 | "smlad %[res], r3, r7, %[res] \n" | 274 | "smlad %[res], r3, r4, %[res] \n" |
| 290 | "bne 1b \n" | 275 | "bne 1b \n" |
| 291 | #else | 276 | #else |
| 292 | "smlad %[res], r2, r6, %[res] \n" | 277 | "smlad %[res], r2, r7, %[res] \n" |
| 293 | "smlad %[res], r3, r7, %[res] \n" | 278 | "smlad %[res], r3, r4, %[res] \n" |
| 294 | #endif | 279 | #endif |
| 295 | 280 | ||
| 296 | "99: \n" | 281 | "99: \n" |
| @@ -303,8 +288,8 @@ static inline int32_t scalarproduct(int16_t* v1, int16_t* v2) | |||
| 303 | [res]"+r"(res) | 288 | [res]"+r"(res) |
| 304 | : /* inputs */ | 289 | : /* inputs */ |
| 305 | : /* clobbers */ | 290 | : /* clobbers */ |
| 306 | "r0", "r1", "r2", "r3", | 291 | "r0", "r1", "r2", "r3", "r4", |
| 307 | "r4", "r5", "r6", "r7" | 292 | "r5", "r6", "r7", "r8" |
| 308 | ); | 293 | ); |
| 309 | return res; | 294 | return res; |
| 310 | } | 295 | } |