jfdctint-avx2.asm 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320
  1. ;
  2. ; jfdctint.asm - accurate integer FDCT (64-bit AVX2)
  3. ;
  4. ; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
  5. ; Copyright (C) 2009, 2016, 2018, D. R. Commander.
  6. ;
  7. ; Based on the x86 SIMD extension for IJG JPEG library
  8. ; Copyright (C) 1999-2006, MIYASAKA Masaru.
  9. ; For conditions of distribution and use, see copyright notice in jsimdext.inc
  10. ;
  11. ; This file should be assembled with NASM (Netwide Assembler),
  12. ; can *not* be assembled with Microsoft's MASM or any compatible
  13. ; assembler (including Borland's Turbo Assembler).
  14. ; NASM is available from http://nasm.sourceforge.net/ or
  15. ; http://sourceforge.net/project/showfiles.php?group_id=6208
  16. ;
  17. ; This file contains a slow-but-accurate integer implementation of the
  18. ; forward DCT (Discrete Cosine Transform). The following code is based
  19. ; directly on the IJG's original jfdctint.c; see the jfdctint.c for
  20. ; more details.
  21. %include "jsimdext.inc"
  22. %include "jdct.inc"
  23. ; --------------------------------------------------------------------------
  24. %define CONST_BITS 13
  25. %define PASS1_BITS 2
  26. %define DESCALE_P1 (CONST_BITS - PASS1_BITS)
  27. %define DESCALE_P2 (CONST_BITS + PASS1_BITS)
  28. %if CONST_BITS == 13
  29. F_0_298 equ 2446 ; FIX(0.298631336)
  30. F_0_390 equ 3196 ; FIX(0.390180644)
  31. F_0_541 equ 4433 ; FIX(0.541196100)
  32. F_0_765 equ 6270 ; FIX(0.765366865)
  33. F_0_899 equ 7373 ; FIX(0.899976223)
  34. F_1_175 equ 9633 ; FIX(1.175875602)
  35. F_1_501 equ 12299 ; FIX(1.501321110)
  36. F_1_847 equ 15137 ; FIX(1.847759065)
  37. F_1_961 equ 16069 ; FIX(1.961570560)
  38. F_2_053 equ 16819 ; FIX(2.053119869)
  39. F_2_562 equ 20995 ; FIX(2.562915447)
  40. F_3_072 equ 25172 ; FIX(3.072711026)
  41. %else
  42. ; NASM cannot do compile-time arithmetic on floating-point constants.
  43. %define DESCALE(x, n) (((x) + (1 << ((n) - 1))) >> (n))
  44. F_0_298 equ DESCALE( 320652955, 30 - CONST_BITS) ; FIX(0.298631336)
  45. F_0_390 equ DESCALE( 418953276, 30 - CONST_BITS) ; FIX(0.390180644)
  46. F_0_541 equ DESCALE( 581104887, 30 - CONST_BITS) ; FIX(0.541196100)
  47. F_0_765 equ DESCALE( 821806413, 30 - CONST_BITS) ; FIX(0.765366865)
  48. F_0_899 equ DESCALE( 966342111, 30 - CONST_BITS) ; FIX(0.899976223)
  49. F_1_175 equ DESCALE(1262586813, 30 - CONST_BITS) ; FIX(1.175875602)
  50. F_1_501 equ DESCALE(1612031267, 30 - CONST_BITS) ; FIX(1.501321110)
  51. F_1_847 equ DESCALE(1984016188, 30 - CONST_BITS) ; FIX(1.847759065)
  52. F_1_961 equ DESCALE(2106220350, 30 - CONST_BITS) ; FIX(1.961570560)
  53. F_2_053 equ DESCALE(2204520673, 30 - CONST_BITS) ; FIX(2.053119869)
  54. F_2_562 equ DESCALE(2751909506, 30 - CONST_BITS) ; FIX(2.562915447)
  55. F_3_072 equ DESCALE(3299298341, 30 - CONST_BITS) ; FIX(3.072711026)
  56. %endif
  57. ; --------------------------------------------------------------------------
  58. ; In-place 8x8x16-bit matrix transpose using AVX2 instructions
  59. ; %1-%4: Input/output registers
  60. ; %5-%8: Temp registers
  61. %macro dotranspose 8
  62. ; %1=(00 01 02 03 04 05 06 07 40 41 42 43 44 45 46 47)
  63. ; %2=(10 11 12 13 14 15 16 17 50 51 52 53 54 55 56 57)
  64. ; %3=(20 21 22 23 24 25 26 27 60 61 62 63 64 65 66 67)
  65. ; %4=(30 31 32 33 34 35 36 37 70 71 72 73 74 75 76 77)
  66. vpunpcklwd %5, %1, %2
  67. vpunpckhwd %6, %1, %2
  68. vpunpcklwd %7, %3, %4
  69. vpunpckhwd %8, %3, %4
  70. ; transpose coefficients(phase 1)
  71. ; %5=(00 10 01 11 02 12 03 13 40 50 41 51 42 52 43 53)
  72. ; %6=(04 14 05 15 06 16 07 17 44 54 45 55 46 56 47 57)
  73. ; %7=(20 30 21 31 22 32 23 33 60 70 61 71 62 72 63 73)
  74. ; %8=(24 34 25 35 26 36 27 37 64 74 65 75 66 76 67 77)
  75. vpunpckldq %1, %5, %7
  76. vpunpckhdq %2, %5, %7
  77. vpunpckldq %3, %6, %8
  78. vpunpckhdq %4, %6, %8
  79. ; transpose coefficients(phase 2)
  80. ; %1=(00 10 20 30 01 11 21 31 40 50 60 70 41 51 61 71)
  81. ; %2=(02 12 22 32 03 13 23 33 42 52 62 72 43 53 63 73)
  82. ; %3=(04 14 24 34 05 15 25 35 44 54 64 74 45 55 65 75)
  83. ; %4=(06 16 26 36 07 17 27 37 46 56 66 76 47 57 67 77)
  84. vpermq %1, %1, 0x8D
  85. vpermq %2, %2, 0x8D
  86. vpermq %3, %3, 0xD8
  87. vpermq %4, %4, 0xD8
  88. ; transpose coefficients(phase 3)
  89. ; %1=(01 11 21 31 41 51 61 71 00 10 20 30 40 50 60 70)
  90. ; %2=(03 13 23 33 43 53 63 73 02 12 22 32 42 52 62 72)
  91. ; %3=(04 14 24 34 44 54 64 74 05 15 25 35 45 55 65 75)
  92. ; %4=(06 16 26 36 46 56 66 76 07 17 27 37 47 57 67 77)
  93. %endmacro
  94. ; --------------------------------------------------------------------------
  95. ; In-place 8x8x16-bit slow integer forward DCT using AVX2 instructions
  96. ; %1-%4: Input/output registers
  97. ; %5-%8: Temp registers
  98. ; %9: Pass (1 or 2)
  99. %macro dodct 9
  100. vpsubw %5, %1, %4 ; %5=data1_0-data6_7=tmp6_7
  101. vpaddw %6, %1, %4 ; %6=data1_0+data6_7=tmp1_0
  102. vpaddw %7, %2, %3 ; %7=data3_2+data4_5=tmp3_2
  103. vpsubw %8, %2, %3 ; %8=data3_2-data4_5=tmp4_5
  104. ; -- Even part
  105. vperm2i128 %6, %6, %6, 0x01 ; %6=tmp0_1
  106. vpaddw %1, %6, %7 ; %1=tmp0_1+tmp3_2=tmp10_11
  107. vpsubw %6, %6, %7 ; %6=tmp0_1-tmp3_2=tmp13_12
  108. vperm2i128 %7, %1, %1, 0x01 ; %7=tmp11_10
  109. vpsignw %1, %1, [rel PW_1_NEG1] ; %1=tmp10_neg11
  110. vpaddw %7, %7, %1 ; %7=(tmp10+tmp11)_(tmp10-tmp11)
  111. %if %9 == 1
  112. vpsllw %1, %7, PASS1_BITS ; %1=data0_4
  113. %else
  114. vpaddw %7, %7, [rel PW_DESCALE_P2X]
  115. vpsraw %1, %7, PASS1_BITS ; %1=data0_4
  116. %endif
  117. ; (Original)
  118. ; z1 = (tmp12 + tmp13) * 0.541196100;
  119. ; data2 = z1 + tmp13 * 0.765366865;
  120. ; data6 = z1 + tmp12 * -1.847759065;
  121. ;
  122. ; (This implementation)
  123. ; data2 = tmp13 * (0.541196100 + 0.765366865) + tmp12 * 0.541196100;
  124. ; data6 = tmp13 * 0.541196100 + tmp12 * (0.541196100 - 1.847759065);
  125. vperm2i128 %7, %6, %6, 0x01 ; %7=tmp12_13
  126. vpunpcklwd %2, %6, %7
  127. vpunpckhwd %6, %6, %7
  128. vpmaddwd %2, %2, [rel PW_F130_F054_MF130_F054] ; %2=data2_6L
  129. vpmaddwd %6, %6, [rel PW_F130_F054_MF130_F054] ; %6=data2_6H
  130. vpaddd %2, %2, [rel PD_DESCALE_P %+ %9]
  131. vpaddd %6, %6, [rel PD_DESCALE_P %+ %9]
  132. vpsrad %2, %2, DESCALE_P %+ %9
  133. vpsrad %6, %6, DESCALE_P %+ %9
  134. vpackssdw %3, %2, %6 ; %6=data2_6
  135. ; -- Odd part
  136. vpaddw %7, %8, %5 ; %7=tmp4_5+tmp6_7=z3_4
  137. ; (Original)
  138. ; z5 = (z3 + z4) * 1.175875602;
  139. ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644;
  140. ; z3 += z5; z4 += z5;
  141. ;
  142. ; (This implementation)
  143. ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602;
  144. ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644);
  145. vperm2i128 %2, %7, %7, 0x01 ; %2=z4_3
  146. vpunpcklwd %6, %7, %2
  147. vpunpckhwd %7, %7, %2
  148. vpmaddwd %6, %6, [rel PW_MF078_F117_F078_F117] ; %6=z3_4L
  149. vpmaddwd %7, %7, [rel PW_MF078_F117_F078_F117] ; %7=z3_4H
  150. ; (Original)
  151. ; z1 = tmp4 + tmp7; z2 = tmp5 + tmp6;
  152. ; tmp4 = tmp4 * 0.298631336; tmp5 = tmp5 * 2.053119869;
  153. ; tmp6 = tmp6 * 3.072711026; tmp7 = tmp7 * 1.501321110;
  154. ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447;
  155. ; data7 = tmp4 + z1 + z3; data5 = tmp5 + z2 + z4;
  156. ; data3 = tmp6 + z2 + z3; data1 = tmp7 + z1 + z4;
  157. ;
  158. ; (This implementation)
  159. ; tmp4 = tmp4 * (0.298631336 - 0.899976223) + tmp7 * -0.899976223;
  160. ; tmp5 = tmp5 * (2.053119869 - 2.562915447) + tmp6 * -2.562915447;
  161. ; tmp6 = tmp5 * -2.562915447 + tmp6 * (3.072711026 - 2.562915447);
  162. ; tmp7 = tmp4 * -0.899976223 + tmp7 * (1.501321110 - 0.899976223);
  163. ; data7 = tmp4 + z3; data5 = tmp5 + z4;
  164. ; data3 = tmp6 + z3; data1 = tmp7 + z4;
  165. vperm2i128 %4, %5, %5, 0x01 ; %4=tmp7_6
  166. vpunpcklwd %2, %8, %4
  167. vpunpckhwd %4, %8, %4
  168. vpmaddwd %2, %2, [rel PW_MF060_MF089_MF050_MF256] ; %2=tmp4_5L
  169. vpmaddwd %4, %4, [rel PW_MF060_MF089_MF050_MF256] ; %4=tmp4_5H
  170. vpaddd %2, %2, %6 ; %2=data7_5L
  171. vpaddd %4, %4, %7 ; %4=data7_5H
  172. vpaddd %2, %2, [rel PD_DESCALE_P %+ %9]
  173. vpaddd %4, %4, [rel PD_DESCALE_P %+ %9]
  174. vpsrad %2, %2, DESCALE_P %+ %9
  175. vpsrad %4, %4, DESCALE_P %+ %9
  176. vpackssdw %4, %2, %4 ; %4=data7_5
  177. vperm2i128 %2, %8, %8, 0x01 ; %2=tmp5_4
  178. vpunpcklwd %8, %5, %2
  179. vpunpckhwd %5, %5, %2
  180. vpmaddwd %8, %8, [rel PW_F050_MF256_F060_MF089] ; %8=tmp6_7L
  181. vpmaddwd %5, %5, [rel PW_F050_MF256_F060_MF089] ; %5=tmp6_7H
  182. vpaddd %8, %8, %6 ; %8=data3_1L
  183. vpaddd %5, %5, %7 ; %5=data3_1H
  184. vpaddd %8, %8, [rel PD_DESCALE_P %+ %9]
  185. vpaddd %5, %5, [rel PD_DESCALE_P %+ %9]
  186. vpsrad %8, %8, DESCALE_P %+ %9
  187. vpsrad %5, %5, DESCALE_P %+ %9
  188. vpackssdw %2, %8, %5 ; %2=data3_1
  189. %endmacro
  190. ; --------------------------------------------------------------------------
  191. SECTION SEG_CONST
  192. alignz 32
  193. GLOBAL_DATA(jconst_fdct_islow_avx2)
  194. EXTN(jconst_fdct_islow_avx2):
  195. PW_F130_F054_MF130_F054 times 4 dw (F_0_541 + F_0_765), F_0_541
  196. times 4 dw (F_0_541 - F_1_847), F_0_541
  197. PW_MF078_F117_F078_F117 times 4 dw (F_1_175 - F_1_961), F_1_175
  198. times 4 dw (F_1_175 - F_0_390), F_1_175
  199. PW_MF060_MF089_MF050_MF256 times 4 dw (F_0_298 - F_0_899), -F_0_899
  200. times 4 dw (F_2_053 - F_2_562), -F_2_562
  201. PW_F050_MF256_F060_MF089 times 4 dw (F_3_072 - F_2_562), -F_2_562
  202. times 4 dw (F_1_501 - F_0_899), -F_0_899
  203. PD_DESCALE_P1 times 8 dd 1 << (DESCALE_P1 - 1)
  204. PD_DESCALE_P2 times 8 dd 1 << (DESCALE_P2 - 1)
  205. PW_DESCALE_P2X times 16 dw 1 << (PASS1_BITS - 1)
  206. PW_1_NEG1 times 8 dw 1
  207. times 8 dw -1
  208. alignz 32
  209. ; --------------------------------------------------------------------------
  210. SECTION SEG_TEXT
  211. BITS 64
  212. ;
  213. ; Perform the forward DCT on one block of samples.
  214. ;
  215. ; GLOBAL(void)
  216. ; jsimd_fdct_islow_avx2(DCTELEM *data)
  217. ;
  218. ; r10 = DCTELEM *data
  219. align 32
  220. GLOBAL_FUNCTION(jsimd_fdct_islow_avx2)
  221. EXTN(jsimd_fdct_islow_avx2):
  222. push rbp
  223. mov rax, rsp
  224. mov rbp, rsp
  225. collect_args 1
  226. ; ---- Pass 1: process rows.
  227. vmovdqu ymm4, YMMWORD [YMMBLOCK(0,0,r10,SIZEOF_DCTELEM)]
  228. vmovdqu ymm5, YMMWORD [YMMBLOCK(2,0,r10,SIZEOF_DCTELEM)]
  229. vmovdqu ymm6, YMMWORD [YMMBLOCK(4,0,r10,SIZEOF_DCTELEM)]
  230. vmovdqu ymm7, YMMWORD [YMMBLOCK(6,0,r10,SIZEOF_DCTELEM)]
  231. ; ymm4=(00 01 02 03 04 05 06 07 10 11 12 13 14 15 16 17)
  232. ; ymm5=(20 21 22 23 24 25 26 27 30 31 32 33 34 35 36 37)
  233. ; ymm6=(40 41 42 43 44 45 46 47 50 51 52 53 54 55 56 57)
  234. ; ymm7=(60 61 62 63 64 65 66 67 70 71 72 73 74 75 76 77)
  235. vperm2i128 ymm0, ymm4, ymm6, 0x20
  236. vperm2i128 ymm1, ymm4, ymm6, 0x31
  237. vperm2i128 ymm2, ymm5, ymm7, 0x20
  238. vperm2i128 ymm3, ymm5, ymm7, 0x31
  239. ; ymm0=(00 01 02 03 04 05 06 07 40 41 42 43 44 45 46 47)
  240. ; ymm1=(10 11 12 13 14 15 16 17 50 51 52 53 54 55 56 57)
  241. ; ymm2=(20 21 22 23 24 25 26 27 60 61 62 63 64 65 66 67)
  242. ; ymm3=(30 31 32 33 34 35 36 37 70 71 72 73 74 75 76 77)
  243. dotranspose ymm0, ymm1, ymm2, ymm3, ymm4, ymm5, ymm6, ymm7
  244. dodct ymm0, ymm1, ymm2, ymm3, ymm4, ymm5, ymm6, ymm7, 1
  245. ; ymm0=data0_4, ymm1=data3_1, ymm2=data2_6, ymm3=data7_5
  246. ; ---- Pass 2: process columns.
  247. vperm2i128 ymm4, ymm1, ymm3, 0x20 ; ymm4=data3_7
  248. vperm2i128 ymm1, ymm1, ymm3, 0x31 ; ymm1=data1_5
  249. dotranspose ymm0, ymm1, ymm2, ymm4, ymm3, ymm5, ymm6, ymm7
  250. dodct ymm0, ymm1, ymm2, ymm4, ymm3, ymm5, ymm6, ymm7, 2
  251. ; ymm0=data0_4, ymm1=data3_1, ymm2=data2_6, ymm4=data7_5
  252. vperm2i128 ymm3, ymm0, ymm1, 0x30 ; ymm3=data0_1
  253. vperm2i128 ymm5, ymm2, ymm1, 0x20 ; ymm5=data2_3
  254. vperm2i128 ymm6, ymm0, ymm4, 0x31 ; ymm6=data4_5
  255. vperm2i128 ymm7, ymm2, ymm4, 0x21 ; ymm7=data6_7
  256. vmovdqu YMMWORD [YMMBLOCK(0,0,r10,SIZEOF_DCTELEM)], ymm3
  257. vmovdqu YMMWORD [YMMBLOCK(2,0,r10,SIZEOF_DCTELEM)], ymm5
  258. vmovdqu YMMWORD [YMMBLOCK(4,0,r10,SIZEOF_DCTELEM)], ymm6
  259. vmovdqu YMMWORD [YMMBLOCK(6,0,r10,SIZEOF_DCTELEM)], ymm7
  260. vzeroupper
  261. uncollect_args 1
  262. pop rbp
  263. ret
  264. ; For some reason, the OS X linker does not honor the request to align the
  265. ; segment unless we do this.
  266. align 32