PageRenderTime 122ms CodeModel.GetById 29ms RepoModel.GetById 9ms app.codeStats 0ms

/webkit-efl/Source/WebCore/platform/graphics/arm/ShadowBlurNEON.cpp

https://review.tizen.org/git/
C++ | 460 lines | 425 code | 8 blank | 27 comment | 0 complexity | 5013c73cd0c376aae6943f6e084143da MD5 | raw file
Possible License(s): GPL-3.0, AGPL-3.0, GPL-2.0, MPL-2.0, JSON, WTFPL, CC-BY-SA-4.0, CC-BY-3.0, BSD-3-Clause, LGPL-2.0, MPL-2.0-no-copyleft-exception, AGPL-1.0, 0BSD, Zlib, Unlicense, BSD-2-Clause, Apache-2.0, LGPL-3.0, ISC, MIT, CC-BY-SA-3.0, CC0-1.0, LGPL-2.1 
    

  1. /*
    2. 

  2.   This file includes NEON optimization codes for 3 BoxBlurs filter.
    3. 

  3.   It is 2.2-2.3 times faster than C implementation in ContextShadow.cpp.
    4. 

    5. 

  5.   Copyright (C) 2011 Hyunki Baik <hyunki.baik@samsung.com>
    6. 

    7. 

  7. */
    8. 

  8. 

  9. #include "config.h"
    10. 

  10. #include "ShadowBlurNEON.h"
    11. 

  11. 

  12. #if ENABLE(TIZEN_CONTEXTSHADOW_BLUR_NEON)
    13. 

  13. #define ASSTRING(str) #str
    14. 

  14. #define TOSTRING(value) ASSTRING(value)
    15. 

  15. #define DMAX_OFFSET TOSTRING(0)
    16. 

  16. #define DMIN_OFFSET TOSTRING(4)
    17. 

  17. #define STRIDE_OFFSET TOSTRING(8)
    18. 

  18. #define STRIDE_WIDTH_OFFSET TOSTRING(12)
    19. 

  19. #define DELTA_OFFSET TOSTRING(16)
    20. 

  20. #define SOURCE_LINE_END_OFFSET TOSTRING(20)
    21. 

  21. #define DIM_OFFSET TOSTRING(24)
    22. 

  22. #define REMAINING_STRIDES_OFFSET TOSTRING(28)
    23. 

  23. #define STEP_COUNT_OFFSET TOSTRING(32)
    24. 

  24. #define STEP_READALPHA_CONSTANTS_OFFSET TOSTRING(36)
    25. 

  25. 

  26. #define NL "\n"
    27. 

  27. // Register allocation.
    28. 

  28. #define SOURCE_R    "r0"
    29. 

  29. #define LIMIT_R    "r1"
    30. 

  30. #define SIDE1_R    "r2"
    31. 

  31. #define SIDE2_R    "r3"
    32. 

  32. #define SOURCE_END_R    "r4"
    33. 

  33. #define DMAX_R    "r5"
    34. 

  34. #define DMIN_R    "r6"
    35. 

  35. #define STRIDE_R    "r7"
    36. 

  36. #define DELTA_R    "r8"
    37. 

  37. #define STEP_COUNT_R    "r9"
    38. 

  38. #define SOURCE_LINE_END_R    "r10"
    39. 

  39. #define DIM_R    "r11"
    40. 

  40. #define STEP_R    "r12"
    41. 

  41. #define PIXELCOUNT_R    "lr"
    42. 

  42. 

  43. // Alternate names.
    44. 

  44. #define INVCOUNT_R LIMIT_R
    45. 

  45. #define REMAINING_STRIDES_R SOURCE_LINE_END_R
    46. 

  46. #define INIT_STEP_READALPHA_R LIMIT_R
    47. 

  47. #define INIT_STEP_STOREALPHA_R PIXELCOUNT_R
    48. 

  48. #define SIDE2_PLUS_ONE_R DMAX_R
    49. 

  49. #define ALPHA_INDEX_R DMIN_R
    50. 

  50. #define INIT_SUM_R PIXELCOUNT_R
    51. 

  51. #define LOOP_INDEX_R PIXELCOUNT_R
    52. 

  52. #define LOOP_TEMP_R SIDE1_R
    53. 

  53. 

  54. // NEON register allocation
    55. 

  55. #define INVCOUNT_Q    "q0"
    56. 

  56. #define SUM_Q    "q1"
    57. 

  57. #define PIXEL_Q    "q2"
    58. 

  58. #define PIXEL_D0    "d4"
    59. 

  59. #define PIXEL_D1    "d5"
    60. 

  60. #define PIXEL_D00    "d4[0]"
    61. 

  61. #define PIXEL_D01    "d4[1]"
    62. 

  62. #define PIXEL_S1    "s9"
    63. 

  63. #define PIXEL_D10    "d5[0]"
    64. 

  64. #define PIXEL_S2    "s10"
    65. 

  65. #define PIXEL_D11    "d5[1]"
    66. 

  66. #define LOCAL_PIXEL_Q    "q3"
    67. 

  67. #define LOCAL_PIXEL_D0    "d6"
    68. 

  68. #define LOCAL_PIXEL_D1    "d7"
    69. 

  69. #define LOCAL_PIXEL_D00    "d6[0]"
    70. 

  70. #define LOCAL_PIXEL_D01    "d6[1]"
    71. 

  71. #define LOCAL_PIXEL_D10     "d7[0]"
    72. 

  72. #define LOCAL_PIXEL_D11    "d7[1]"
    73. 

  73. #define REMAINING_STRIDES_S    "s16"
    74. 

  74. #define STRIDE_WIDTH_S    "s17"
    75. 

  75. #define DMAX_S    "s18"
    76. 

  76. #define DMIN_S    "s19"
    77. 

  77. #define TEMP_Q    "q5"
    78. 

  78. #define ALPHA1_Q    "q6"
    79. 

  79. #define ALPHA1_D0    "d12"
    80. 

  80. #define ALPHA1_D1    "d13"
    81. 

  81. #define ALPHA2_Q    "q7"
    82. 

  82. #define ALPHA2_D0    "d14"
    83. 

  83. #define ALPHA2_D1    "d15"
    84. 

  84. #define REMAP_STEP_1_LOADALPHA_Q    "d16"
    85. 

  85. #define REMAP_STEP_2_LOADALPHA_Q    "d17"
    86. 

  86. #define REMAP_STEP_1_STOREALPHA_Q    "d18"
    87. 

  87. #define REMAP_STEP_2_STOREALPHA_Q    "d19"
    88. 

  88. #define READALPHA_RANGE    "d20-d25"
    89. 

  89. #define REMAP_STEP00_LOADALPHA_Q    "d20"
    90. 

  90. #define REMAP_STEP01_LOADALPHA_Q    "d21"
    91. 

  91. #define REMAP_STEP10_LOADALPHA_Q    "d22"
    92. 

  92. #define REMAP_STEP11_LOADALPHA_Q    "d23"
    93. 

  93. #define REMAP_STEP20_LOADALPHA_Q    "d24"
    94. 

  94. #define REMAP_STEP21_LOADALPHA_Q    "d25"
    95. 

  95. #define STOREALPHA_RANGE    "d26-31"
    96. 

  96. #define REMAP_STEP00_STOREALPHA_Q    "d26"
    97. 

  97. #define REMAP_STEP01_STOREALPHA_Q    "d27"
    98. 

  98. #define REMAP_STEP10_STOREALPHA_Q    "d28"
    99. 

  99. #define REMAP_STEP11_STOREALPHA_Q    "d29"
    100. 

  100. #define REMAP_STEP20_STOREALPHA_Q    "d30"
    101. 

  101. #define REMAP_STEP21_STOREALPHA_Q    "d31"
    102. 

  102. 

  103. #define DATA_TRANSFER4(command, base) \
    104. 

  104.     command " " PIXEL_D00 ", [" base "], " DELTA_R NL \
    105. 

  105.     command " " PIXEL_D01 ", [" base "], " DELTA_R NL \
    106. 

  106.     command " " PIXEL_D10 ", [" base "], " DELTA_R NL \
    107. 

  107.     command " " PIXEL_D11 ", [" base "], " DELTA_R NL \
    108. 

  108.     "sub " base ", " base ", " DELTA_R ", lsl #2" NL
    109. 

  109. 

  110. // The number of reads depend on REMAINING_STRIDES_R, but it is always >= 1 and <= 3
    111. 

  111. #define CONDITIONAL_DATA_TRANSFER4(command1, command2, base) \
    112. 

  112.     command1 " " PIXEL_D00 ", [" base "], " DELTA_R NL \
    113. 

  113.     "cmp " REMAINING_STRIDES_R ", #2" NL \
    114. 

  114.     command2 "cs " PIXEL_S1 ", [" base "]" NL \
    115. 

  115.     "add " base ", " base ", " DELTA_R NL \
    116. 

  116.     "cmp " REMAINING_STRIDES_R ", #3" NL \
    117. 

  117.     command2 "cs " PIXEL_S2 ", [" base "]" NL \
    118. 

  118.     "sub " base ", " base ", " DELTA_R ", lsl #1" NL
    119. 

  119. 

  120. asm ( // NOLINT
    121. 

  121. ".globl " TOSTRING(boxBlurNeon) NL
    122. 

  122. TOSTRING(boxBlurNeon) ":" NL
    123. 

  123.     ".fpu neon" NL
    124. 

  124.     "stmdb sp!, {r4-r12, lr}" NL
    125. 

  125.     "vpush {d7-d15}" NL
    126. 

  126.     "vldr.u32 " DMAX_S ", [r1, #" DMAX_OFFSET "]" NL
    127. 

  127.     "vldr.u32 " DMIN_S ", [r1, #" DMIN_OFFSET "]" NL
    128. 

  128.     "ldr " STRIDE_R ", [r1, #" STRIDE_OFFSET "]" NL
    129. 

  129.     "vldr.u32 " STRIDE_WIDTH_S ", [r1, #" STRIDE_WIDTH_OFFSET "]" NL
    130. 

  130.     "ldr " DELTA_R ", [r1, #" DELTA_OFFSET "]" NL
    131. 

  131.     "ldr " SOURCE_LINE_END_R ", [r1, #" SOURCE_LINE_END_OFFSET "]" NL
    132. 

  132.     "ldr " DIM_R ", [r1, #" DIM_OFFSET "]" NL
    133. 

  133.     "vldr.u32 " REMAINING_STRIDES_S ", [r1, #" REMAINING_STRIDES_OFFSET "]" NL
    134. 

  134.     "ldr " STEP_COUNT_R ", [r1, #" STEP_COUNT_OFFSET "]" NL
    135. 

  135.     "ldr " INIT_STEP_READALPHA_R ", [r1, #" STEP_READALPHA_CONSTANTS_OFFSET "]" NL
    136. 

  136.     "mla " SOURCE_LINE_END_R ", " SOURCE_LINE_END_R ", " DELTA_R ", " SOURCE_R NL
    137. 

  137.     "cmp " SOURCE_LINE_END_R ", " SOURCE_R NL
    138. 

  138.     "beq .EarlyLeave" NL
    139. 

  139.     "vld1.u32 { d20-d22 }, [" INIT_STEP_READALPHA_R "]!" NL
    140. 

  140.     "vld1.u32 { d23-d25 }, [" INIT_STEP_READALPHA_R "]!" NL
    141. 

  141.     "vld1.u32 { d26-d28 }, [" INIT_STEP_READALPHA_R "]!" NL
    142. 

  142.     "vld1.u32 { d29-d31 }, [" INIT_STEP_READALPHA_R "]!" NL
    143. 

  143. ".MainLoop:" NL
    144. 

  144.     // Processing 4 strides parallelly.
    145. 

  145.     "mov " STEP_R ", #0"  NL
    146. 

  146. ".StepLoop:" NL
    147. 

  147.     "vmov.u32 " DMAX_R ", " DMAX_S NL
    148. 

  148.     "vmov.u32 " DMIN_R ", " DMIN_S NL
    149. 

  149.     "cmp " STEP_R ", #2" NL
    150. 

  150.     "beq .InitStep2" NL
    151. 

  151.     "cmp " STEP_R ", #1" NL
    152. 

  152.     "beq .InitStep1" NL
    153. 

  153. ".InitStep0:" NL
    154. 

  154.     "mov " SIDE1_R ", " DMIN_R NL
    155. 

  155.     "mov " SIDE2_R ", " DMAX_R NL
    156. 

  156.     "vmov.u8 " REMAP_STEP_1_LOADALPHA_Q "," REMAP_STEP00_LOADALPHA_Q NL
    157. 

  157.     "vmov.u8 " REMAP_STEP_2_LOADALPHA_Q "," REMAP_STEP01_LOADALPHA_Q NL
    158. 

  158.     "vmov.u8 " REMAP_STEP_1_STOREALPHA_Q "," REMAP_STEP00_STOREALPHA_Q NL
    159. 

  159.     "vmov.u8 " REMAP_STEP_2_STOREALPHA_Q "," REMAP_STEP01_STOREALPHA_Q NL
    160. 

  160.     "bal .EndInitStep" NL
    161. 

  161. ".InitStep1:" NL
    162. 

  162.     "mov " SIDE1_R ", " DMAX_R NL
    163. 

  163.     "mov " SIDE2_R ", " DMIN_R NL
    164. 

  164.     "vmov.u8 " REMAP_STEP_1_LOADALPHA_Q "," REMAP_STEP10_LOADALPHA_Q NL
    165. 

  165.     "vmov.u8 " REMAP_STEP_2_LOADALPHA_Q "," REMAP_STEP11_LOADALPHA_Q NL
    166. 

  166.     "vmov.u8 " REMAP_STEP_1_STOREALPHA_Q "," REMAP_STEP10_STOREALPHA_Q NL
    167. 

  167.     "vmov.u8 " REMAP_STEP_2_STOREALPHA_Q "," REMAP_STEP11_STOREALPHA_Q NL
    168. 

  168.     "bal .EndInitStep" NL
    169. 

  169. ".InitStep2:" NL
    170. 

  170.     "mov " SIDE1_R ", " DMAX_R NL
    171. 

  171.     "mov " SIDE2_R ", " DMAX_R NL
    172. 

  172.     "vmov.u8 " REMAP_STEP_1_LOADALPHA_Q "," REMAP_STEP20_LOADALPHA_Q NL
    173. 

  173.     "vmov.u8 " REMAP_STEP_2_LOADALPHA_Q "," REMAP_STEP21_LOADALPHA_Q NL
    174. 

  174.     "vmov.u8 " REMAP_STEP_1_STOREALPHA_Q "," REMAP_STEP20_STOREALPHA_Q NL
    175. 

  175.     "vmov.u8 " REMAP_STEP_2_STOREALPHA_Q "," REMAP_STEP21_STOREALPHA_Q NL
    176. 

  176. ".EndInitStep:" NL
    177. 

  177.     "add " PIXELCOUNT_R ", " SIDE1_R ", " SIDE2_R NL
    178. 

  178.     "add " PIXELCOUNT_R ", " PIXELCOUNT_R ", #1" NL
    179. 

  179.     "mov " INVCOUNT_R ", #1" NL
    180. 

  180.     "add " INVCOUNT_R ", " PIXELCOUNT_R ", " INVCOUNT_R ", lsl #15" NL
    181. 

  181.     "subs " INVCOUNT_R ", " INVCOUNT_R ", #1" NL
    182. 

  182. //////////////////////////////////
    183. 

  183. // integer div code from http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.dui0204h/CEGECDGD.html
    184. 

  184. // DIV: DMAX_R(r5), BOT: PIXELCOUNT_R(lr), TOP: INVCOUNT_R(r1), TEMP: DMIN_R(r6)
    185. 

  185.     "mov r6, lr" NL
    186. 

  186.     "cmp r6, r1, LSR #1" NL
    187. 

  187. ".DivLable1:" NL
    188. 

  188.     "movls  r6, r6, LSL #1" NL
    189. 

  189.     "cmp r6, r1, LSR #1" NL
    190. 

  190.     "bls .DivLable1" NL
    191. 

  191.     "mov r5, #0" NL
    192. 

  192. ".DivLable2:" NL
    193. 

  193.     "cmp r1, r6" NL
    194. 

  194.     "subcs r1, r1, r6" NL
    195. 

  195.     "adc r5, r5, r5" NL
    196. 

  196.     "mov r6, r6, LSR #1" NL
    197. 

  197.     "cmp r6, lr" NL
    198. 

  198.     "bhs .DivLable2" NL
    199. 

  199.     "vdup.u32 " INVCOUNT_Q ", " DMAX_R NL
    200. 

  200. ".EndDiv:" NL
    201. 

  201.     DATA_TRANSFER4("vld1.u32", SOURCE_R)
    202. 

  202.     "vtbl.8 " ALPHA1_D0 ", {" PIXEL_D0 "-" PIXEL_D1 "}, " REMAP_STEP_1_LOADALPHA_Q NL
    203. 

  203.     "vtbl.8 " ALPHA1_D1 ", {" PIXEL_D0 "-" PIXEL_D1 "}, " REMAP_STEP_2_LOADALPHA_Q NL
    204. 

  204.     "sub " ALPHA_INDEX_R ", " DIM_R ", #1" NL
    205. 

  205.     "mla " ALPHA_INDEX_R ", " ALPHA_INDEX_R ", " STRIDE_R ", " SOURCE_R NL
    206. 

  206.     DATA_TRANSFER4("vld1.u32", ALPHA_INDEX_R)
    207. 

  207.     "vtbl.8 " ALPHA2_D0 ", {" PIXEL_D0 "-" PIXEL_D1 "}, " REMAP_STEP_1_LOADALPHA_Q NL
    208. 

  208.     "vtbl.8 " ALPHA2_D1 ", {" PIXEL_D0 "-" PIXEL_D1 "}, " REMAP_STEP_2_LOADALPHA_Q NL
    209. 

  209.     "vdup.u32 " TEMP_Q ", " SIDE1_R NL
    210. 

  210.     "vmul.u32 " TEMP_Q ", " TEMP_Q ", " ALPHA1_Q NL
    211. 

  211.     "vadd.u32 " SUM_Q ", " TEMP_Q ", " ALPHA1_Q NL
    212. 

  212.     "add " SIDE2_PLUS_ONE_R ", " SIDE2_R ", #1" NL
    213. 

  213.     "cmp " DIM_R ", " SIDE2_PLUS_ONE_R NL
    214. 

  214.     "bcc .DimSet" NL
    215. 

  215. ".Side2PlusOneSet:" NL
    216. 

  216.     "mov " LIMIT_R ", " SIDE2_PLUS_ONE_R NL
    217. 

  217.     "bal .EndSetLimit1" NL
    218. 

  218. ".DimSet:" NL
    219. 

  219.     "mov " LIMIT_R ", " DIM_R NL
    220. 

  220. ".EndSetLimit1:" NL
    221. 

  221.     "mov " INIT_SUM_R ", " SOURCE_R NL
    222. 

  222.     "mla " SOURCE_END_R ", " LIMIT_R ", " STRIDE_R ", " SOURCE_R NL
    223. 

  223.     "add " INIT_SUM_R ", " INIT_SUM_R ", " STRIDE_R NL
    224. 

  224.     "cmp " INIT_SUM_R ", " SOURCE_END_R NL
    225. 

  225.     "bcs .InitSumDone" NL
    226. 

  226. ".InitSum:" NL
    227. 

  227.     DATA_TRANSFER4("vld1.u32", INIT_SUM_R)
    228. 

  228.     "vtbl.8 " LOCAL_PIXEL_D0 ", {" PIXEL_D0 "-" PIXEL_D1 "}, " REMAP_STEP_1_LOADALPHA_Q NL
    229. 

  229.     "vtbl.8 " LOCAL_PIXEL_D1 ", {" PIXEL_D0 "-" PIXEL_D1 "}, " REMAP_STEP_2_LOADALPHA_Q NL
    230. 

  230.     "vadd.u32 " SUM_Q ", " SUM_Q ", " LOCAL_PIXEL_Q NL
    231. 

  231.     "add " INIT_SUM_R ", " INIT_SUM_R ", " STRIDE_R NL
    232. 

  232.     "cmp " INIT_SUM_R ", " SOURCE_END_R NL
    233. 

  233.     "bcc .InitSum" NL
    234. 

  234. ".InitSumDone:" NL
    235. 

  235.     "cmp " SIDE2_R ", " LIMIT_R NL
    236. 

  236.     "bcc .EndSetSum" NL
    237. 

  237. ".SetSum:" NL
    238. 

  238.     "sub " INIT_SUM_R ", " SIDE2_PLUS_ONE_R ", " LIMIT_R NL
    239. 

  239.     "vdup.u32 " TEMP_Q ", " INIT_SUM_R  NL
    240. 

  240.     "vmul.u32 " TEMP_Q ", " ALPHA2_Q ", " TEMP_Q NL
    241. 

  241.     "vadd.u32 " SUM_Q ", " SUM_Q ", " TEMP_Q NL
    242. 

  242. ".EndSetSum:" NL
    243. 

  243.     "cmp " SIDE1_R ", " DIM_R NL
    244. 

  244.     "bcc .SetLimit2" NL
    245. 

  245.     "mov " LIMIT_R ", " DIM_R NL
    246. 

  246.     "bal .EndSetLimit2" NL
    247. 

  247. ".SetLimit2:" NL
    248. 

  248.     "mov " LIMIT_R ", " SIDE1_R NL
    249. 

  249. ".EndSetLimit2:" NL
    250. 

  250.     // Blurring.
    251. 

  251.     "mov " LOOP_INDEX_R ", #0" NL
    252. 

  252. ".Blur:" NL
    253. 

  253.     "vmul.u32 " LOCAL_PIXEL_Q ", " SUM_Q ", " INVCOUNT_Q NL
    254. 

  254.     "vshr.u32 " LOCAL_PIXEL_Q ", " LOCAL_PIXEL_Q ", #15" NL
    255. 

  255.     "mla " LOOP_TEMP_R ", " LOOP_INDEX_R ", " STRIDE_R ", " SOURCE_R NL
    256. 

  256.     DATA_TRANSFER4("vld1.u32", LOOP_TEMP_R)
    257. 

  257.     "vtbl.8 " PIXEL_D0 ", {" PIXEL_D0 "-" LOCAL_PIXEL_D1 "}, " REMAP_STEP_1_STOREALPHA_Q NL
    258. 

  258.     "vtbl.8 " PIXEL_D1 ", {" PIXEL_D0 "-" LOCAL_PIXEL_D1 "}, " REMAP_STEP_2_STOREALPHA_Q NL
    259. 

  259.     DATA_TRANSFER4("vst1.u32", LOOP_TEMP_R)
    260. 

  260.     "cmp " LOOP_INDEX_R ", " LIMIT_R NL
    261. 

  261.     "bcc .SubtractAlpha1" NL
    262. 

  262.     "sub " LOOP_TEMP_R ", " LOOP_INDEX_R ", " LIMIT_R NL
    263. 

  263.     "mla " LOOP_TEMP_R ", " LOOP_TEMP_R ", " STRIDE_R ", " SOURCE_R NL
    264. 

  264.     DATA_TRANSFER4("vld1.u32", LOOP_TEMP_R)
    265. 

  265.     "vtbl.8 " LOCAL_PIXEL_D0 ", {" PIXEL_D0 "-" PIXEL_D1 "}, " REMAP_STEP_1_LOADALPHA_Q NL
    266. 

  266.     "vtbl.8 " LOCAL_PIXEL_D1 ", {" PIXEL_D0 "-" PIXEL_D1 "}, " REMAP_STEP_2_LOADALPHA_Q NL
    267. 

  267.     "vsub.u32 " SUM_Q ", " SUM_Q ", " LOCAL_PIXEL_Q NL
    268. 

  268.     "bal .EndLeft" NL
    269. 

  269. ".SubtractAlpha1: " NL
    270. 

  270.     "vsub.u32 " SUM_Q ", " SUM_Q ", " ALPHA1_Q NL
    271. 

  271. ".EndLeft: "
    272. 

  272.     "add " LOOP_TEMP_R ", " LOOP_INDEX_R ", " SIDE2_R NL
    273. 

  273.     "add " LOOP_TEMP_R ", " LOOP_TEMP_R ", #1" NL
    274. 

  274.     "cmp " LOOP_TEMP_R ", " DIM_R NL
    275. 

  275.     "bcc .SetRight" NL
    276. 

  276.     "vadd.u32 " SUM_Q ", " SUM_Q ", " ALPHA2_Q NL
    277. 

  277.     "bal .EndRight" NL
    278. 

  278. ".SetRight: " NL
    279. 

  279.     "mla " LOOP_TEMP_R ", " LOOP_TEMP_R ", " STRIDE_R ", " SOURCE_R NL
    280. 

  280.     DATA_TRANSFER4("vld1.u32", LOOP_TEMP_R)
    281. 

  281.     "vtbl.8 " LOCAL_PIXEL_D0 ", {" PIXEL_D0 "-" PIXEL_D1 "}, " REMAP_STEP_1_LOADALPHA_Q NL
    282. 

  282.     "vtbl.8 " LOCAL_PIXEL_D1 ", {" PIXEL_D0 "-" PIXEL_D1 "}, " REMAP_STEP_2_LOADALPHA_Q NL
    283. 

  283.     "vadd.u32 " SUM_Q ", " SUM_Q ", " LOCAL_PIXEL_Q NL
    284. 

  284. ".EndRight: " NL
    285. 

  285.     "add " LOOP_INDEX_R ", " LOOP_INDEX_R ", #1" NL
    286. 

  286.     "cmp " LOOP_INDEX_R ", " DIM_R NL
    287. 

  287.     "bcc .Blur" NL
    288. 

  288. ".EndBlurLine: " NL
    289. 

  289.     // 3 step check
    290. 

  290.     "add " STEP_R ", " STEP_R ", #1" NL
    291. 

  291.     "cmp " STEP_R ", " STEP_COUNT_R NL
    292. 

  292.     "bcc .StepLoop" NL
    293. 

  293. ".EndStepLoop: " NL
    294. 

  294.     // Line check
    295. 

  295.     "add " SOURCE_R ", " SOURCE_R ", " DELTA_R ", lsl #2" NL // next 4 lines
    296. 

  296.     "cmp " SOURCE_R ", " SOURCE_LINE_END_R NL // check all lines are processed
    297. 

  297.     "bcc .MainLoop" NL
    298. 

  298. /////////////////////////////////////////////////////////////////////////////////////////////////
    299. 

  299.     // Processing the remaining strides (0 - 3).
    300. 

  300. ".EarlyLeave:" NL
    301. 

  301.     "vmov.u32 " REMAINING_STRIDES_R ", " REMAINING_STRIDES_S NL
    302. 

  302.     // Early return for 0 strides.
    303. 

  303.     "cmp " REMAINING_STRIDES_R ", #1" NL
    304. 

  304.     "bcs .SecondStepLoopStart" NL
    305. 

  305.     "vpop {d7-d15}" NL
    306. 

  306.     "ldmia sp!, {r4-r12, pc}" NL
    307. 

  307. ".SecondStepLoopStart:" NL
    308. 

  308.     // initialize step variable
    309. 

  309.     "mov " STEP_R ", #0"  NL
    310. 

  310. ".SecondStepLoop:" NL
    311. 

  311.     "vmov.u32 " DMAX_R ", " DMAX_S NL
    312. 

  312.     "vmov.u32 " DMIN_R ", " DMIN_S NL
    313. 

  313.     "cmp " STEP_R ", #2" NL
    314. 

  314.     "beq .SecondInitStep2" NL
    315. 

  315.     "cmp " STEP_R ", #1" NL
    316. 

  316.     "beq .SecondInitStep1" NL
    317. 

  317. ".SecondInitStep0:" NL
    318. 

  318.     "mov " SIDE1_R ", " DMIN_R NL
    319. 

  319.     "mov " SIDE2_R ", " DMAX_R NL
    320. 

  320.     "vmov.u8 " REMAP_STEP_1_LOADALPHA_Q "," REMAP_STEP00_LOADALPHA_Q NL
    321. 

  321.     "vmov.u8 " REMAP_STEP_2_LOADALPHA_Q "," REMAP_STEP01_LOADALPHA_Q NL
    322. 

  322.     "vmov.u8 " REMAP_STEP_1_STOREALPHA_Q "," REMAP_STEP00_STOREALPHA_Q NL
    323. 

  323.     "vmov.u8 " REMAP_STEP_2_STOREALPHA_Q "," REMAP_STEP01_STOREALPHA_Q NL
    324. 

  324.     "bal .SecondEndInitStep" NL
    325. 

  325. ".SecondInitStep1:" NL
    326. 

  326.     "mov " SIDE1_R ", " DMAX_R NL
    327. 

  327.     "mov " SIDE2_R ", " DMIN_R NL
    328. 

  328.     "vmov.u8 " REMAP_STEP_1_LOADALPHA_Q "," REMAP_STEP10_LOADALPHA_Q NL
    329. 

  329.     "vmov.u8 " REMAP_STEP_2_LOADALPHA_Q "," REMAP_STEP11_LOADALPHA_Q NL
    330. 

  330.     "vmov.u8 " REMAP_STEP_1_STOREALPHA_Q "," REMAP_STEP10_STOREALPHA_Q NL
    331. 

  331.     "vmov.u8 " REMAP_STEP_2_STOREALPHA_Q "," REMAP_STEP11_STOREALPHA_Q NL
    332. 

  332.     "bal .SecondEndInitStep" NL
    333. 

  333. ".SecondInitStep2:" NL
    334. 

  334.     "mov " SIDE1_R ", " DMAX_R NL
    335. 

  335.     "mov " SIDE2_R ", " DMAX_R NL
    336. 

  336.     "vmov.u8 " REMAP_STEP_1_LOADALPHA_Q "," REMAP_STEP20_LOADALPHA_Q NL
    337. 

  337.     "vmov.u8 " REMAP_STEP_2_LOADALPHA_Q "," REMAP_STEP21_LOADALPHA_Q NL
    338. 

  338.     "vmov.u8 " REMAP_STEP_1_STOREALPHA_Q "," REMAP_STEP20_STOREALPHA_Q NL
    339. 

  339.     "vmov.u8 " REMAP_STEP_2_STOREALPHA_Q "," REMAP_STEP21_STOREALPHA_Q NL
    340. 

  340. ".SecondEndInitStep:" NL
    341. 

  341.     "add " PIXELCOUNT_R ", " SIDE1_R ", " SIDE2_R NL
    342. 

  342.     "add " PIXELCOUNT_R ", " PIXELCOUNT_R ", #1" NL
    343. 

  343.     "mov " INVCOUNT_R ", #1" NL
    344. 

  344.     "add " INVCOUNT_R ", " PIXELCOUNT_R ", " INVCOUNT_R ", lsl #15" NL
    345. 

  345.     "subs " INVCOUNT_R ", " INVCOUNT_R ", #1" NL
    346. 

  346. //////////////////////////////////
    347. 

  347. // integer div code from http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.dui0204h/CEGECDGD.html
    348. 

  348. // DIV: DMAX_R(r5), BOT: PIXELCOUNT_R(lr), TOP: INVCOUNT_R(r1), TEMP: DMIN_R(r6)
    349. 

  349.     "mov r6, lr" NL
    350. 

  350.     "cmp r6, r1, LSR #1" NL
    351. 

  351. ".SecondDivLable1:" NL
    352. 

  352.     "movls  r6, r6, LSL #1" NL
    353. 

  353.     "cmp r6, r1, LSR #1" NL
    354. 

  354.     "bls .SecondDivLable1" NL
    355. 

  355.     "mov r5, #0" NL
    356. 

  356. ".SecondDivLable2:" NL
    357. 

  357.     "cmp r1, r6" NL
    358. 

  358.     "subcs r1, r1, r6" NL
    359. 

  359.     "adc r5, r5, r5" NL
    360. 

  360.     "mov r6, r6, LSR #1" NL
    361. 

  361.     "cmp r6, lr" NL
    362. 

  362.     "bhs .SecondDivLable2" NL
    363. 

  363.     "vdup.u32 " INVCOUNT_Q ", " DMAX_R NL
    364. 

  364. ".SecondEndDiv:" NL
    365. 

  365.     CONDITIONAL_DATA_TRANSFER4("vld1.u32", "vldr", SOURCE_R)
    366. 

  366.     "vtbl.8 " ALPHA1_D0 ", {" PIXEL_D0 "-" PIXEL_D1 "}, " REMAP_STEP_1_LOADALPHA_Q NL
    367. 

  367.     "vtbl.8 " ALPHA1_D1 ", {" PIXEL_D0 "-" PIXEL_D1 "}, " REMAP_STEP_2_LOADALPHA_Q NL
    368. 

  368.     "sub " ALPHA_INDEX_R ", " DIM_R ", #1" NL
    369. 

  369.     "mla " ALPHA_INDEX_R ", " ALPHA_INDEX_R ", " STRIDE_R ", " SOURCE_R NL
    370. 

  370.     CONDITIONAL_DATA_TRANSFER4("vld1.u32", "vldr", ALPHA_INDEX_R)
    371. 

  371.     "vtbl.8 " ALPHA2_D0 ", {" PIXEL_D0 "-" PIXEL_D1 "}, " REMAP_STEP_1_LOADALPHA_Q NL
    372. 

  372.     "vtbl.8 " ALPHA2_D1 ", {" PIXEL_D0 "-" PIXEL_D1 "}, " REMAP_STEP_2_LOADALPHA_Q NL
    373. 

  373.     "vdup.u32 " TEMP_Q ", " SIDE1_R NL
    374. 

  374.     "vmul.u32 " TEMP_Q ", " TEMP_Q ", " ALPHA1_Q NL
    375. 

  375.     "vadd.u32 " SUM_Q ", " TEMP_Q ", " ALPHA1_Q NL
    376. 

  376.     "add " SIDE2_PLUS_ONE_R ", " SIDE2_R ", #1" NL
    377. 

  377.     "cmp " DIM_R ", " SIDE2_PLUS_ONE_R NL
    378. 

  378.     "bcc .SecondDimSet" NL
    379. 

  379. ".SecondSide2PlusOneSet:" NL
    380. 

  380.     "mov " LIMIT_R ", " SIDE2_PLUS_ONE_R NL
    381. 

  381.     "bal .SecondEndSetLimit1" NL
    382. 

  382. ".SecondDimSet:" NL
    383. 

  383.     "mov " LIMIT_R ", " DIM_R NL
    384. 

  384. ".SecondEndSetLimit1:" NL
    385. 

  385.     "mov " INIT_SUM_R ", " SOURCE_R NL
    386. 

  386.     "mla " SOURCE_END_R ", " LIMIT_R ", " STRIDE_R ", " SOURCE_R NL
    387. 

  387.     "add " INIT_SUM_R ", " INIT_SUM_R ", " STRIDE_R NL
    388. 

  388.     "cmp " INIT_SUM_R ", " SOURCE_END_R NL
    389. 

  389.     "bcs .SecondInitSumDone" NL
    390. 

  390. ".SecondInitSum:" NL
    391. 

  391.     CONDITIONAL_DATA_TRANSFER4("vld1.u32", "vldr", INIT_SUM_R)
    392. 

  392.     "vtbl.8 " LOCAL_PIXEL_D0 ", {" PIXEL_D0 "-" PIXEL_D1 "}, " REMAP_STEP_1_LOADALPHA_Q NL
    393. 

  393.     "vtbl.8 " LOCAL_PIXEL_D1 ", {" PIXEL_D0 "-" PIXEL_D1 "}, " REMAP_STEP_2_LOADALPHA_Q NL
    394. 

  394.     "vadd.u32 " SUM_Q ", " SUM_Q ", " LOCAL_PIXEL_Q NL
    395. 

  395.     "add " INIT_SUM_R ", " INIT_SUM_R ", " STRIDE_R NL
    396. 

  396.     "cmp " INIT_SUM_R ", " SOURCE_END_R NL
    397. 

  397.     "bcc .SecondInitSum" NL
    398. 

  398. ".SecondInitSumDone:" NL
    399. 

  399.     "cmp " SIDE2_R ", " LIMIT_R NL
    400. 

  400.     "bcc .SecondEndSetSum" NL
    401. 

  401. ".SecondSetSum:" NL
    402. 

  402.     "sub " INIT_SUM_R ", " SIDE2_PLUS_ONE_R ", " LIMIT_R NL
    403. 

  403.     "vdup.u32 " TEMP_Q ", " INIT_SUM_R  NL
    404. 

  404.     "vmul.u32 " TEMP_Q ", " ALPHA2_Q ", " TEMP_Q NL
    405. 

  405.     "vadd.u32 " SUM_Q ", " SUM_Q ", " TEMP_Q NL
    406. 

  406. ".SecondEndSetSum:" NL
    407. 

  407.     "cmp " SIDE1_R ", " DIM_R NL
    408. 

  408.     "bcc .SecondSetLimit2" NL
    409. 

  409.     "mov " LIMIT_R ", " DIM_R NL
    410. 

  410.     "bal .SecondEndSetLimit2" NL
    411. 

  411. ".SecondSetLimit2:" NL
    412. 

  412.     "mov " LIMIT_R ", " SIDE1_R NL
    413. 

  413. ".SecondEndSetLimit2:" NL
    414. 

  414.     // Blurring.
    415. 

  415.     "mov " LOOP_INDEX_R ", #0" NL
    416. 

  416. ".SecondBlur:" NL
    417. 

  417.     "vmul.u32 " LOCAL_PIXEL_Q ", " SUM_Q ", " INVCOUNT_Q NL
    418. 

  418.     "vshr.u32 " LOCAL_PIXEL_Q ", " LOCAL_PIXEL_Q ", #15" NL
    419. 

  419.     "mla " LOOP_TEMP_R ", " LOOP_INDEX_R ", " STRIDE_R ", " SOURCE_R NL
    420. 

  420.     CONDITIONAL_DATA_TRANSFER4("vld1.u32", "vldr", LOOP_TEMP_R)
    421. 

  421.     "vtbl.8 " PIXEL_D0 ", {" PIXEL_D0 "-" LOCAL_PIXEL_D1 "}, " REMAP_STEP_1_STOREALPHA_Q NL
    422. 

  422.     "vtbl.8 " PIXEL_D1 ", {" PIXEL_D0 "-" LOCAL_PIXEL_D1 "}, " REMAP_STEP_2_STOREALPHA_Q NL
    423. 

  423.     CONDITIONAL_DATA_TRANSFER4("vst1.u32", "vstr", LOOP_TEMP_R)
    424. 

  424.     "cmp " LOOP_INDEX_R ", " LIMIT_R NL
    425. 

  425.     "bcc .SecondSubtractAlpha1" NL
    426. 

  426.     "sub " LOOP_TEMP_R ", " LOOP_INDEX_R ", " LIMIT_R NL
    427. 

  427.     "mla " LOOP_TEMP_R ", " LOOP_TEMP_R ", " STRIDE_R ", " SOURCE_R NL
    428. 

  428.     CONDITIONAL_DATA_TRANSFER4("vld1.u32", "vldr", LOOP_TEMP_R)
    429. 

  429.     "vtbl.8 " LOCAL_PIXEL_D0 ", {" PIXEL_D0 "-" PIXEL_D1 "}, " REMAP_STEP_1_LOADALPHA_Q NL
    430. 

  430.     "vtbl.8 " LOCAL_PIXEL_D1 ", {" PIXEL_D0 "-" PIXEL_D1 "}, " REMAP_STEP_2_LOADALPHA_Q NL
    431. 

  431.     "vsub.u32 " SUM_Q ", " SUM_Q ", " LOCAL_PIXEL_Q NL
    432. 

  432.     "bal .SecondEndLeft" NL
    433. 

  433. ".SecondSubtractAlpha1: " NL
    434. 

  434.     "vsub.u32 " SUM_Q ", " SUM_Q ", " ALPHA1_Q NL
    435. 

  435. ".SecondEndLeft: "
    436. 

  436.     "add " LOOP_TEMP_R ", " LOOP_INDEX_R ", " SIDE2_R NL
    437. 

  437.     "add " LOOP_TEMP_R ", " LOOP_TEMP_R ", #1" NL
    438. 

  438.     "cmp " LOOP_TEMP_R ", " DIM_R NL
    439. 

  439.     "bcc .SecondSetRight" NL
    440. 

  440.     "vadd.u32 " SUM_Q ", " SUM_Q ", " ALPHA2_Q NL
    441. 

  441.     "bal .SecondEndRight" NL
    442. 

  442. ".SecondSetRight: " NL
    443. 

  443.     "mla " LOOP_TEMP_R ", " LOOP_TEMP_R ", " STRIDE_R ", " SOURCE_R NL
    444. 

  444.     CONDITIONAL_DATA_TRANSFER4("vld1.u32", "vldr", LOOP_TEMP_R)
    445. 

  445.     "vtbl.8 " LOCAL_PIXEL_D0 ", {" PIXEL_D0 "-" PIXEL_D1 "}, " REMAP_STEP_1_LOADALPHA_Q NL
    446. 

  446.     "vtbl.8 " LOCAL_PIXEL_D1 ", {" PIXEL_D0 "-" PIXEL_D1 "}, " REMAP_STEP_2_LOADALPHA_Q NL
    447. 

  447.     "vadd.u32 " SUM_Q ", " SUM_Q ", " LOCAL_PIXEL_Q NL
    448. 

  448. ".SecondEndRight: " NL
    449. 

  449.     "add " LOOP_INDEX_R ", " LOOP_INDEX_R ", #1" NL
    450. 

  450.     "cmp " LOOP_INDEX_R ", " DIM_R NL
    451. 

  451.     "bcc .SecondBlur" NL
    452. 

  452. ".SecondEndBlurLine: " NL
    453. 

  453.     // step check
    454. 

  454.     "add " STEP_R ", " STEP_R ", #1" NL
    455. 

  455.     "cmp " STEP_R ", " STEP_COUNT_R NL
    456. 

  456.     "bcc .SecondStepLoop" NL
    457. 

  457.     "vpop {d7-d15}" NL
    458. 

  458.     "ldmia sp!, {r4-r12, pc}" NL
    459. 

  459. );
    460. 

  460. #endif
    461. 

  461.