mpegvideo_armv5te.c 7.9 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213
  1. /*
  2. * Optimization of some functions from mpegvideo.c for armv5te
  3. * Copyright (c) 2007 Siarhei Siamashka <ssvb@users.sourceforge.net>
  4. *
  5. * This file is part of FFmpeg.
  6. *
  7. * FFmpeg is free software; you can redistribute it and/or
  8. * modify it under the terms of the GNU Lesser General Public
  9. * License as published by the Free Software Foundation; either
  10. * version 2.1 of the License, or (at your option) any later version.
  11. *
  12. * FFmpeg is distributed in the hope that it will be useful,
  13. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  15. * Lesser General Public License for more details.
  16. *
  17. * You should have received a copy of the GNU Lesser General Public
  18. * License along with FFmpeg; if not, write to the Free Software
  19. * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  20. */
  21. /*
  22. * Some useful links for those who may be interested in optimizing code for ARM.
  23. * ARM Architecture Reference Manual: http://www.arm.com/community/academy/resources.html
  24. * Instructions timings and optimization guide for ARM9E: http://www.arm.com/pdfs/DDI0222B_9EJS_r1p2.pdf
  25. */
  26. #include "../dsputil.h"
  27. #include "../mpegvideo.h"
  28. #include "../avcodec.h"
  29. #ifdef ENABLE_ARM_TESTS
  30. /**
  31. * h263 dequantizer supplementary function, it is performance critical and needs to
  32. * have optimized implementations for each architecture. Is also used as a reference
  33. * implementation in regression tests
  34. */
  35. static inline void dct_unquantize_h263_helper_c(DCTELEM *block, int qmul, int qadd, int count)
  36. {
  37. int i, level;
  38. for (i = 0; i < count; i++) {
  39. level = block[i];
  40. if (level) {
  41. if (level < 0) {
  42. level = level * qmul - qadd;
  43. } else {
  44. level = level * qmul + qadd;
  45. }
  46. block[i] = level;
  47. }
  48. }
  49. }
  50. #endif
  51. /* GCC 3.1 or higher is required to support symbolic names in assembly code */
  52. #if (__GNUC__ > 3) || ((__GNUC__ == 3) && (__GNUC_MINOR__ >= 1))
  53. /**
  54. * Special optimized version of dct_unquantize_h263_helper_c, it requires the block
  55. * to be at least 8 bytes aligned, and may process more elements than requested.
  56. * But it is guaranteed to never process more than 64 elements provided that
  57. * xxcount argument is <= 64, so it is safe. This macro is optimized for a common
  58. * distribution of values for nCoeffs (they are mostly multiple of 8 plus one or
  59. * two extra elements). So this macro processes data as 8 elements per loop iteration
  60. * and contains optional 2 elements processing in the end.
  61. *
  62. * Inner loop should take 6 cycles per element on arm926ej-s (Nokia 770)
  63. */
  64. #define dct_unquantize_h263_special_helper_armv5te(xxblock, xxqmul, xxqadd, xxcount) \
  65. ({ DCTELEM *xblock = xxblock; \
  66. int xqmul = xxqmul, xqadd = xxqadd, xcount = xxcount, xtmp; \
  67. int xdata1, xdata2; \
  68. __asm__ __volatile__( \
  69. "subs %[count], %[count], #2 \n\t" \
  70. "ble 2f \n\t" \
  71. "ldrd r4, [%[block], #0] \n\t" \
  72. "1: \n\t" \
  73. "ldrd r6, [%[block], #8] \n\t" \
  74. \
  75. "rsbs %[data1], %[zero], r4, asr #16 \n\t" \
  76. "addgt %[data1], %[qadd], #0 \n\t" \
  77. "rsblt %[data1], %[qadd], #0 \n\t" \
  78. "smlatbne %[data1], r4, %[qmul], %[data1] \n\t" \
  79. \
  80. "rsbs %[data2], %[zero], r5, asr #16 \n\t" \
  81. "addgt %[data2], %[qadd], #0 \n\t" \
  82. "rsblt %[data2], %[qadd], #0 \n\t" \
  83. "smlatbne %[data2], r5, %[qmul], %[data2] \n\t" \
  84. \
  85. "rsbs %[tmp], %[zero], r4, asl #16 \n\t" \
  86. "addgt %[tmp], %[qadd], #0 \n\t" \
  87. "rsblt %[tmp], %[qadd], #0 \n\t" \
  88. "smlabbne r4, r4, %[qmul], %[tmp] \n\t" \
  89. \
  90. "rsbs %[tmp], %[zero], r5, asl #16 \n\t" \
  91. "addgt %[tmp], %[qadd], #0 \n\t" \
  92. "rsblt %[tmp], %[qadd], #0 \n\t" \
  93. "smlabbne r5, r5, %[qmul], %[tmp] \n\t" \
  94. \
  95. "strh r4, [%[block]], #2 \n\t" \
  96. "strh %[data1], [%[block]], #2 \n\t" \
  97. "strh r5, [%[block]], #2 \n\t" \
  98. "strh %[data2], [%[block]], #2 \n\t" \
  99. \
  100. "rsbs %[data1], %[zero], r6, asr #16 \n\t" \
  101. "addgt %[data1], %[qadd], #0 \n\t" \
  102. "rsblt %[data1], %[qadd], #0 \n\t" \
  103. "smlatbne %[data1], r6, %[qmul], %[data1] \n\t" \
  104. \
  105. "rsbs %[data2], %[zero], r7, asr #16 \n\t" \
  106. "addgt %[data2], %[qadd], #0 \n\t" \
  107. "rsblt %[data2], %[qadd], #0 \n\t" \
  108. "smlatbne %[data2], r7, %[qmul], %[data2] \n\t" \
  109. \
  110. "rsbs %[tmp], %[zero], r6, asl #16 \n\t" \
  111. "addgt %[tmp], %[qadd], #0 \n\t" \
  112. "rsblt %[tmp], %[qadd], #0 \n\t" \
  113. "smlabbne r6, r6, %[qmul], %[tmp] \n\t" \
  114. \
  115. "rsbs %[tmp], %[zero], r7, asl #16 \n\t" \
  116. "addgt %[tmp], %[qadd], #0 \n\t" \
  117. "rsblt %[tmp], %[qadd], #0 \n\t" \
  118. "smlabbne r7, r7, %[qmul], %[tmp] \n\t" \
  119. \
  120. "strh r6, [%[block]], #2 \n\t" \
  121. "strh %[data1], [%[block]], #2 \n\t" \
  122. "strh r7, [%[block]], #2 \n\t" \
  123. "strh %[data2], [%[block]], #2 \n\t" \
  124. \
  125. "subs %[count], %[count], #8 \n\t" \
  126. "ldrgtd r4, [%[block], #0] \n\t" /* load data early to avoid load/use pipeline stall */ \
  127. "bgt 1b \n\t" \
  128. \
  129. "adds %[count], %[count], #2 \n\t" \
  130. "ble 3f \n\t" \
  131. "2: \n\t" \
  132. "ldrsh %[data1], [%[block], #0] \n\t" \
  133. "ldrsh %[data2], [%[block], #2] \n\t" \
  134. "mov %[tmp], %[qadd] \n\t" \
  135. "cmp %[data1], #0 \n\t" \
  136. "rsblt %[tmp], %[qadd], #0 \n\t" \
  137. "smlabbne %[data1], %[data1], %[qmul], %[tmp] \n\t" \
  138. "mov %[tmp], %[qadd] \n\t" \
  139. "cmp %[data2], #0 \n\t" \
  140. "rsblt %[tmp], %[qadd], #0 \n\t" \
  141. "smlabbne %[data2], %[data2], %[qmul], %[tmp] \n\t" \
  142. "strh %[data1], [%[block]], #2 \n\t" \
  143. "strh %[data2], [%[block]], #2 \n\t" \
  144. "3: \n\t" \
  145. : [block] "+&r" (xblock), [count] "+&r" (xcount), [tmp] "=&r" (xtmp), \
  146. [data1] "=&r" (xdata1), [data2] "=&r" (xdata2) \
  147. : [qmul] "r" (xqmul), [qadd] "r" (xqadd), [zero] "r" (0) \
  148. : "r4", "r5", "r6", "r7", "cc", "memory" \
  149. ); \
  150. })
  151. static void dct_unquantize_h263_intra_armv5te(MpegEncContext *s,
  152. DCTELEM *block, int n, int qscale)
  153. {
  154. int i, level, qmul, qadd;
  155. int nCoeffs;
  156. assert(s->block_last_index[n]>=0);
  157. qmul = qscale << 1;
  158. if (!s->h263_aic) {
  159. if (n < 4)
  160. level = block[0] * s->y_dc_scale;
  161. else
  162. level = block[0] * s->c_dc_scale;
  163. qadd = (qscale - 1) | 1;
  164. }else{
  165. qadd = 0;
  166. level = block[0];
  167. }
  168. if(s->ac_pred)
  169. nCoeffs=63;
  170. else
  171. nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
  172. dct_unquantize_h263_special_helper_armv5te(block, qmul, qadd, nCoeffs + 1);
  173. block[0] = level;
  174. }
  175. static void dct_unquantize_h263_inter_armv5te(MpegEncContext *s,
  176. DCTELEM *block, int n, int qscale)
  177. {
  178. int i, level, qmul, qadd;
  179. int nCoeffs;
  180. assert(s->block_last_index[n]>=0);
  181. qadd = (qscale - 1) | 1;
  182. qmul = qscale << 1;
  183. nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
  184. dct_unquantize_h263_special_helper_armv5te(block, qmul, qadd, nCoeffs + 1);
  185. }
  186. #define HAVE_DCT_UNQUANTIZE_H263_ARMV5TE_OPTIMIZED
  187. #endif
  188. void MPV_common_init_armv5te(MpegEncContext *s)
  189. {
  190. #ifdef HAVE_DCT_UNQUANTIZE_H263_ARMV5TE_OPTIMIZED
  191. s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_armv5te;
  192. s->dct_unquantize_h263_inter = dct_unquantize_h263_inter_armv5te;
  193. #endif
  194. }