API tools faq
paste
Guest User

Untitled

a guest
Dec 8th, 2016
65
0
Never
Add comment
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
text 83.87 KB | None | 0 0
raw download clone embed print report
  1. /*============================================================================
  2.  
  3.  
  4. NVIDIA FXAA 3.11 by TIMOTHY LOTTES
  5.  
  6.  
  7. ------------------------------------------------------------------------------
  8. COPYRIGHT (C) 2010, 2011 NVIDIA CORPORATION. ALL RIGHTS RESERVED.
  9. ------------------------------------------------------------------------------
  10. TO THE MAXIMUM EXTENT PERMITTED BY APPLICABLE LAW, THIS SOFTWARE IS PROVIDED
  11. *AS IS* AND NVIDIA AND ITS SUPPLIERS DISCLAIM ALL WARRANTIES, EITHER EXPRESS
  12. OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, IMPLIED WARRANTIES OF
  13. MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. IN NO EVENT SHALL NVIDIA
  14. OR ITS SUPPLIERS BE LIABLE FOR ANY SPECIAL, INCIDENTAL, INDIRECT, OR
  15. CONSEQUENTIAL DAMAGES WHATSOEVER (INCLUDING, WITHOUT LIMITATION, DAMAGES FOR
  16. LOSS OF BUSINESS PROFITS, BUSINESS INTERRUPTION, LOSS OF BUSINESS INFORMATION,
  17. OR ANY OTHER PECUNIARY LOSS) ARISING OUT OF THE USE OF OR INABILITY TO USE
  18. THIS SOFTWARE, EVEN IF NVIDIA HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH
  19. DAMAGES.
  20.  
  21. ------------------------------------------------------------------------------
  22. INTEGRATION CHECKLIST
  23. ------------------------------------------------------------------------------
  24. (1.)
  25. In the shader source, setup defines for the desired configuration.
  26. When providing multiple shaders (for different presets),
  27. simply setup the defines differently in multiple files.
  28. Example,
  29.  
  30. #define FXAA_PC 1
  31. #define FXAA_HLSL_5 1
  32. #define FXAA_QUALITY__PRESET 12
  33.  
  34. Or,
  35.  
  36. #define FXAA_360 1
  37.  
  38. Or,
  39.  
  40. #define FXAA_PS3 1
  41.  
  42. Etc.
  43.  
  44. (2.)
  45. Then include this file,
  46.  
  47. #include "Fxaa3_11.h"
  48.  
  49. (3.)
  50. Then call the FXAA pixel shader from within your desired shader.
  51. Look at the FXAA Quality FxaaPixelShader() for docs on inputs.
  52. As for FXAA 3.11 all inputs for all shaders are the same
  53. to enable easy porting between platforms.
  54.  
  55. return FxaaPixelShader(...);
  56.  
  57. (4.)
  58. Insure pass prior to FXAA outputs RGBL (see next section).
  59. Or use,
  60.  
  61. #define FXAA_GREEN_AS_LUMA 1
  62.  
  63. (5.)
  64. Setup engine to provide the following constants
  65. which are used in the FxaaPixelShader() inputs,
  66.  
  67. FxaaFloat2 fxaaQualityRcpFrame,
  68. FxaaFloat4 fxaaConsoleRcpFrameOpt,
  69. FxaaFloat4 fxaaConsoleRcpFrameOpt2,
  70. FxaaFloat4 fxaaConsole360RcpFrameOpt2,
  71. FxaaFloat fxaaQualitySubpix,
  72. FxaaFloat fxaaQualityEdgeThreshold,
  73. FxaaFloat fxaaQualityEdgeThresholdMin,
  74. FxaaFloat fxaaConsoleEdgeSharpness,
  75. FxaaFloat fxaaConsoleEdgeThreshold,
  76. FxaaFloat fxaaConsoleEdgeThresholdMin,
  77. FxaaFloat4 fxaaConsole360ConstDir
  78.  
  79. Look at the FXAA Quality FxaaPixelShader() for docs on inputs.
  80.  
  81. (6.)
  82. Have FXAA vertex shader run as a full screen triangle,
  83. and output "pos" and "fxaaConsolePosPos"
  84. such that inputs in the pixel shader provide,
  85.  
  86. // {xy} = center of pixel
  87. FxaaFloat2 pos,
  88.  
  89. // {xy__} = upper left of pixel
  90. // {__zw} = lower right of pixel
  91. FxaaFloat4 fxaaConsolePosPos,
  92.  
  93. (7.)
  94. Insure the texture sampler(s) used by FXAA are set to bilinear filtering.
  95.  
  96.  
  97. ------------------------------------------------------------------------------
  98. INTEGRATION - RGBL AND COLORSPACE
  99. ------------------------------------------------------------------------------
  100. FXAA3 requires RGBL as input unless the following is set,
  101.  
  102. #define FXAA_GREEN_AS_LUMA 1
  103.  
  104. In which case the engine uses green in place of luma,
  105. and requires RGB input is in a non-linear colorspace.
  106.  
  107. RGB should be LDR (low dynamic range).
  108. Specifically do FXAA after tonemapping.
  109.  
  110. RGB data as returned by a texture fetch can be non-linear,
  111. or linear when FXAA_GREEN_AS_LUMA is not set.
  112. Note an "sRGB format" texture counts as linear,
  113. because the result of a texture fetch is linear data.
  114. Regular "RGBA8" textures in the sRGB colorspace are non-linear.
  115.  
  116. If FXAA_GREEN_AS_LUMA is not set,
  117. luma must be stored in the alpha channel prior to running FXAA.
  118. This luma should be in a perceptual space (could be gamma 2.0).
  119. Example pass before FXAA where output is gamma 2.0 encoded,
  120.  
  121. color.rgb = ToneMap(color.rgb); // linear color output
  122. color.rgb = sqrt(color.rgb); // gamma 2.0 color output
  123. return color;
  124.  
  125. To use FXAA,
  126.  
  127. color.rgb = ToneMap(color.rgb); // linear color output
  128. color.rgb = sqrt(color.rgb); // gamma 2.0 color output
  129. color.a = dot(color.rgb, FxaaFloat3(0.299, 0.587, 0.114)); // compute luma
  130. return color;
  131.  
  132. Another example where output is linear encoded,
  133. say for instance writing to an sRGB formated render target,
  134. where the render target does the conversion back to sRGB after blending,
  135.  
  136. color.rgb = ToneMap(color.rgb); // linear color output
  137. return color;
  138.  
  139. To use FXAA,
  140.  
  141. color.rgb = ToneMap(color.rgb); // linear color output
  142. color.a = sqrt(dot(color.rgb, FxaaFloat3(0.299, 0.587, 0.114))); // compute luma
  143. return color;
  144.  
  145. Getting luma correct is required for the algorithm to work correctly.
  146.  
  147.  
  148. ------------------------------------------------------------------------------
  149. BEING LINEARLY CORRECT?
  150. ------------------------------------------------------------------------------
  151. Applying FXAA to a framebuffer with linear RGB color will look worse.
  152. This is very counter intuitive, but happends to be true in this case.
  153. The reason is because dithering artifacts will be more visiable
  154. in a linear colorspace.
  155.  
  156.  
  157. ------------------------------------------------------------------------------
  158. COMPLEX INTEGRATION
  159. ------------------------------------------------------------------------------
  160. Q. What if the engine is blending into RGB before wanting to run FXAA?
  161.  
  162. A. In the last opaque pass prior to FXAA,
  163. have the pass write out luma into alpha.
  164. Then blend into RGB only.
  165. FXAA should be able to run ok
  166. assuming the blending pass did not any add aliasing.
  167. This should be the common case for particles and common blending passes.
  168.  
  169. A. Or use FXAA_GREEN_AS_LUMA.
  170.  
  171. ============================================================================*/
  172.  
  173. /*============================================================================
  174.  
  175. INTEGRATION KNOBS
  176.  
  177. ============================================================================*/
  178. //
  179. // FXAA_PS3 and FXAA_360 choose the console algorithm (FXAA3 CONSOLE).
  180. // FXAA_360_OPT is a prototype for the new optimized 360 version.
  181. //
  182. // 1 = Use API.
  183. // 0 = Don't use API.
  184. //
  185. /*--------------------------------------------------------------------------*/
  186. #ifndef FXAA_PS3
  187. #define FXAA_PS3 0
  188. #endif
  189. /*--------------------------------------------------------------------------*/
  190. #ifndef FXAA_360
  191. #define FXAA_360 0
  192. #endif
  193. /*--------------------------------------------------------------------------*/
  194. #ifndef FXAA_360_OPT
  195. #define FXAA_360_OPT 0
  196. #endif
  197. /*==========================================================================*/
  198. #ifndef FXAA_PC
  199. //
  200. // FXAA Quality
  201. // The high quality PC algorithm.
  202. //
  203. #define FXAA_PC 0
  204. #endif
  205. /*--------------------------------------------------------------------------*/
  206. #ifndef FXAA_PC_CONSOLE
  207. //
  208. // The console algorithm for PC is included
  209. // for developers targeting really low spec machines.
  210. // Likely better to just run FXAA_PC, and use a really low preset.
  211. //
  212. #define FXAA_PC_CONSOLE 0
  213. #endif
  214. /*--------------------------------------------------------------------------*/
  215. #ifndef FXAA_GLSL_120
  216. #define FXAA_GLSL_120 0
  217. #endif
  218. /*--------------------------------------------------------------------------*/
  219. #ifndef FXAA_GLSL_130
  220. #define FXAA_GLSL_130 0
  221. #endif
  222. /*--------------------------------------------------------------------------*/
  223. #ifndef FXAA_HLSL_3
  224. #define FXAA_HLSL_3 0
  225. #endif
  226. /*--------------------------------------------------------------------------*/
  227. #ifndef FXAA_HLSL_4
  228. #define FXAA_HLSL_4 0
  229. #endif
  230. /*--------------------------------------------------------------------------*/
  231. #ifndef FXAA_HLSL_5
  232. #define FXAA_HLSL_5 0
  233. #endif
  234. /*==========================================================================*/
  235. #ifndef FXAA_GREEN_AS_LUMA
  236. //
  237. // For those using non-linear color,
  238. // and either not able to get luma in alpha, or not wanting to,
  239. // this enables FXAA to run using green as a proxy for luma.
  240. // So with this enabled, no need to pack luma in alpha.
  241. //
  242. // This will turn off AA on anything which lacks some amount of green.
  243. // Pure red and blue or combination of only R and B, will get no AA.
  244. //
  245. // Might want to lower the settings for both,
  246. // fxaaConsoleEdgeThresholdMin
  247. // fxaaQualityEdgeThresholdMin
  248. // In order to insure AA does not get turned off on colors
  249. // which contain a minor amount of green.
  250. //
  251. // 1 = On.
  252. // 0 = Off.
  253. //
  254. #define FXAA_GREEN_AS_LUMA 0
  255. #endif
  256. /*--------------------------------------------------------------------------*/
  257. #ifndef FXAA_EARLY_EXIT
  258. //
  259. // Controls algorithm's early exit path.
  260. // On PS3 turning this ON adds 2 cycles to the shader.
  261. // On 360 turning this OFF adds 10ths of a millisecond to the shader.
  262. // Turning this off on console will result in a more blurry image.
  263. // So this defaults to on.
  264. //
  265. // 1 = On.
  266. // 0 = Off.
  267. //
  268. #define FXAA_EARLY_EXIT 1
  269. #endif
  270. /*--------------------------------------------------------------------------*/
  271. #ifndef FXAA_DISCARD
  272. //
  273. // Only valid for PC OpenGL currently.
  274. // Probably will not work when FXAA_GREEN_AS_LUMA = 1.
  275. //
  276. // 1 = Use discard on pixels which don't need AA.
  277. // For APIs which enable concurrent TEX+ROP from same surface.
  278. // 0 = Return unchanged color on pixels which don't need AA.
  279. //
  280. #define FXAA_DISCARD 0
  281. #endif
  282. /*--------------------------------------------------------------------------*/
  283. #ifndef FXAA_FAST_PIXEL_OFFSET
  284. //
  285. // Used for GLSL 120 only.
  286. //
  287. // 1 = GL API supports fast pixel offsets
  288. // 0 = do not use fast pixel offsets
  289. //
  290. #ifdef GL_EXT_gpu_shader4
  291. #define FXAA_FAST_PIXEL_OFFSET 1
  292. #endif
  293. #ifdef GL_NV_gpu_shader5
  294. #define FXAA_FAST_PIXEL_OFFSET 1
  295. #endif
  296. #ifdef GL_ARB_gpu_shader5
  297. #define FXAA_FAST_PIXEL_OFFSET 1
  298. #endif
  299. #ifndef FXAA_FAST_PIXEL_OFFSET
  300. #define FXAA_FAST_PIXEL_OFFSET 0
  301. #endif
  302. #endif
  303. /*--------------------------------------------------------------------------*/
  304. #ifndef FXAA_GATHER4_ALPHA
  305. //
  306. // 1 = API supports gather4 on alpha channel.
  307. // 0 = API does not support gather4 on alpha channel.
  308. //
  309. #if (FXAA_HLSL_5 == 1)
  310. #define FXAA_GATHER4_ALPHA 1
  311. #endif
  312. #ifdef GL_ARB_gpu_shader5
  313. #define FXAA_GATHER4_ALPHA 1
  314. #endif
  315. #ifdef GL_NV_gpu_shader5
  316. #define FXAA_GATHER4_ALPHA 1
  317. #endif
  318. #ifndef FXAA_GATHER4_ALPHA
  319. #define FXAA_GATHER4_ALPHA 0
  320. #endif
  321. #endif
  322.  
  323. /*============================================================================
  324. FXAA CONSOLE PS3 - TUNING KNOBS
  325. ============================================================================*/
  326. #ifndef FXAA_CONSOLE__PS3_EDGE_SHARPNESS
  327. //
  328. // Consoles the sharpness of edges on PS3 only.
  329. // Non-PS3 tuning is done with shader input.
  330. //
  331. // Due to the PS3 being ALU bound,
  332. // there are only two safe values here: 4 and 8.
  333. // These options use the shaders ability to a free *|/ by 2|4|8.
  334. //
  335. // 8.0 is sharper
  336. // 4.0 is softer
  337. // 2.0 is really soft (good for vector graphics inputs)
  338. //
  339. #if 1
  340. #define FXAA_CONSOLE__PS3_EDGE_SHARPNESS 8.0
  341. #endif
  342. #if 0
  343. #define FXAA_CONSOLE__PS3_EDGE_SHARPNESS 4.0
  344. #endif
  345. #if 0
  346. #define FXAA_CONSOLE__PS3_EDGE_SHARPNESS 2.0
  347. #endif
  348. #endif
  349. /*--------------------------------------------------------------------------*/
  350. #ifndef FXAA_CONSOLE__PS3_EDGE_THRESHOLD
  351. //
  352. // Only effects PS3.
  353. // Non-PS3 tuning is done with shader input.
  354. //
  355. // The minimum amount of local contrast required to apply algorithm.
  356. // The console setting has a different mapping than the quality setting.
  357. //
  358. // This only applies when FXAA_EARLY_EXIT is 1.
  359. //
  360. // Due to the PS3 being ALU bound,
  361. // there are only two safe values here: 0.25 and 0.125.
  362. // These options use the shaders ability to a free *|/ by 2|4|8.
  363. //
  364. // 0.125 leaves less aliasing, but is softer
  365. // 0.25 leaves more aliasing, and is sharper
  366. //
  367. #if 1
  368. #define FXAA_CONSOLE__PS3_EDGE_THRESHOLD 0.125
  369. #else
  370. #define FXAA_CONSOLE__PS3_EDGE_THRESHOLD 0.25
  371. #endif
  372. #endif
  373.  
  374. /*============================================================================
  375. FXAA QUALITY - TUNING KNOBS
  376. ------------------------------------------------------------------------------
  377. NOTE the other tuning knobs are now in the shader function inputs!
  378. ============================================================================*/
  379. #ifndef FXAA_QUALITY__PRESET
  380. //
  381. // Choose the quality preset.
  382. // This needs to be compiled into the shader as it effects code.
  383. // Best option to include multiple presets is to
  384. // in each shader define the preset, then include this file.
  385. //
  386. // OPTIONS
  387. // -----------------------------------------------------------------------
  388. // 10 to 15 - default medium dither (10=fastest, 15=highest quality)
  389. // 20 to 29 - less dither, more expensive (20=fastest, 29=highest quality)
  390. // 39 - no dither, very expensive
  391. //
  392. // NOTES
  393. // -----------------------------------------------------------------------
  394. // 12 = slightly faster then FXAA 3.9 and higher edge quality (default)
  395. // 13 = about same speed as FXAA 3.9 and better than 12
  396. // 23 = closest to FXAA 3.9 visually and performance wise
  397. // _ = the lowest digit is directly related to performance
  398. // _ = the highest digit is directly related to style
  399. //
  400. #define FXAA_QUALITY__PRESET 12
  401. #endif
  402.  
  403.  
  404. /*============================================================================
  405.  
  406. FXAA QUALITY - PRESETS
  407.  
  408. ============================================================================*/
  409.  
  410. /*============================================================================
  411. FXAA QUALITY - MEDIUM DITHER PRESETS
  412. ============================================================================*/
  413. #if (FXAA_QUALITY__PRESET == 10)
  414. #define FXAA_QUALITY__PS 3
  415. #define FXAA_QUALITY__P0 1.5
  416. #define FXAA_QUALITY__P1 3.0
  417. #define FXAA_QUALITY__P2 12.0
  418. #endif
  419. /*--------------------------------------------------------------------------*/
  420. #if (FXAA_QUALITY__PRESET == 11)
  421. #define FXAA_QUALITY__PS 4
  422. #define FXAA_QUALITY__P0 1.0
  423. #define FXAA_QUALITY__P1 1.5
  424. #define FXAA_QUALITY__P2 3.0
  425. #define FXAA_QUALITY__P3 12.0
  426. #endif
  427. /*--------------------------------------------------------------------------*/
  428. #if (FXAA_QUALITY__PRESET == 12)
  429. #define FXAA_QUALITY__PS 5
  430. #define FXAA_QUALITY__P0 1.0
  431. #define FXAA_QUALITY__P1 1.5
  432. #define FXAA_QUALITY__P2 2.0
  433. #define FXAA_QUALITY__P3 4.0
  434. #define FXAA_QUALITY__P4 12.0
  435. #endif
  436. /*--------------------------------------------------------------------------*/
  437. #if (FXAA_QUALITY__PRESET == 13)
  438. #define FXAA_QUALITY__PS 6
  439. #define FXAA_QUALITY__P0 1.0
  440. #define FXAA_QUALITY__P1 1.5
  441. #define FXAA_QUALITY__P2 2.0
  442. #define FXAA_QUALITY__P3 2.0
  443. #define FXAA_QUALITY__P4 4.0
  444. #define FXAA_QUALITY__P5 12.0
  445. #endif
  446. /*--------------------------------------------------------------------------*/
  447. #if (FXAA_QUALITY__PRESET == 14)
  448. #define FXAA_QUALITY__PS 7
  449. #define FXAA_QUALITY__P0 1.0
  450. #define FXAA_QUALITY__P1 1.5
  451. #define FXAA_QUALITY__P2 2.0
  452. #define FXAA_QUALITY__P3 2.0
  453. #define FXAA_QUALITY__P4 2.0
  454. #define FXAA_QUALITY__P5 4.0
  455. #define FXAA_QUALITY__P6 12.0
  456. #endif
  457. /*--------------------------------------------------------------------------*/
  458. #if (FXAA_QUALITY__PRESET == 15)
  459. #define FXAA_QUALITY__PS 8
  460. #define FXAA_QUALITY__P0 1.0
  461. #define FXAA_QUALITY__P1 1.5
  462. #define FXAA_QUALITY__P2 2.0
  463. #define FXAA_QUALITY__P3 2.0
  464. #define FXAA_QUALITY__P4 2.0
  465. #define FXAA_QUALITY__P5 2.0
  466. #define FXAA_QUALITY__P6 4.0
  467. #define FXAA_QUALITY__P7 12.0
  468. #endif
  469.  
  470. /*============================================================================
  471. FXAA QUALITY - LOW DITHER PRESETS
  472. ============================================================================*/
  473. #if (FXAA_QUALITY__PRESET == 20)
  474. #define FXAA_QUALITY__PS 3
  475. #define FXAA_QUALITY__P0 1.5
  476. #define FXAA_QUALITY__P1 2.0
  477. #define FXAA_QUALITY__P2 8.0
  478. #endif
  479. /*--------------------------------------------------------------------------*/
  480. #if (FXAA_QUALITY__PRESET == 21)
  481. #define FXAA_QUALITY__PS 4
  482. #define FXAA_QUALITY__P0 1.0
  483. #define FXAA_QUALITY__P1 1.5
  484. #define FXAA_QUALITY__P2 2.0
  485. #define FXAA_QUALITY__P3 8.0
  486. #endif
  487. /*--------------------------------------------------------------------------*/
  488. #if (FXAA_QUALITY__PRESET == 22)
  489. #define FXAA_QUALITY__PS 5
  490. #define FXAA_QUALITY__P0 1.0
  491. #define FXAA_QUALITY__P1 1.5
  492. #define FXAA_QUALITY__P2 2.0
  493. #define FXAA_QUALITY__P3 2.0
  494. #define FXAA_QUALITY__P4 8.0
  495. #endif
  496. /*--------------------------------------------------------------------------*/
  497. #if (FXAA_QUALITY__PRESET == 23)
  498. #define FXAA_QUALITY__PS 6
  499. #define FXAA_QUALITY__P0 1.0
  500. #define FXAA_QUALITY__P1 1.5
  501. #define FXAA_QUALITY__P2 2.0
  502. #define FXAA_QUALITY__P3 2.0
  503. #define FXAA_QUALITY__P4 2.0
  504. #define FXAA_QUALITY__P5 8.0
  505. #endif
  506. /*--------------------------------------------------------------------------*/
  507. #if (FXAA_QUALITY__PRESET == 24)
  508. #define FXAA_QUALITY__PS 7
  509. #define FXAA_QUALITY__P0 1.0
  510. #define FXAA_QUALITY__P1 1.5
  511. #define FXAA_QUALITY__P2 2.0
  512. #define FXAA_QUALITY__P3 2.0
  513. #define FXAA_QUALITY__P4 2.0
  514. #define FXAA_QUALITY__P5 3.0
  515. #define FXAA_QUALITY__P6 8.0
  516. #endif
  517. /*--------------------------------------------------------------------------*/
  518. #if (FXAA_QUALITY__PRESET == 25)
  519. #define FXAA_QUALITY__PS 8
  520. #define FXAA_QUALITY__P0 1.0
  521. #define FXAA_QUALITY__P1 1.5
  522. #define FXAA_QUALITY__P2 2.0
  523. #define FXAA_QUALITY__P3 2.0
  524. #define FXAA_QUALITY__P4 2.0
  525. #define FXAA_QUALITY__P5 2.0
  526. #define FXAA_QUALITY__P6 4.0
  527. #define FXAA_QUALITY__P7 8.0
  528. #endif
  529. /*--------------------------------------------------------------------------*/
  530. #if (FXAA_QUALITY__PRESET == 26)
  531. #define FXAA_QUALITY__PS 9
  532. #define FXAA_QUALITY__P0 1.0
  533. #define FXAA_QUALITY__P1 1.5
  534. #define FXAA_QUALITY__P2 2.0
  535. #define FXAA_QUALITY__P3 2.0
  536. #define FXAA_QUALITY__P4 2.0
  537. #define FXAA_QUALITY__P5 2.0
  538. #define FXAA_QUALITY__P6 2.0
  539. #define FXAA_QUALITY__P7 4.0
  540. #define FXAA_QUALITY__P8 8.0
  541. #endif
  542. /*--------------------------------------------------------------------------*/
  543. #if (FXAA_QUALITY__PRESET == 27)
  544. #define FXAA_QUALITY__PS 10
  545. #define FXAA_QUALITY__P0 1.0
  546. #define FXAA_QUALITY__P1 1.5
  547. #define FXAA_QUALITY__P2 2.0
  548. #define FXAA_QUALITY__P3 2.0
  549. #define FXAA_QUALITY__P4 2.0
  550. #define FXAA_QUALITY__P5 2.0
  551. #define FXAA_QUALITY__P6 2.0
  552. #define FXAA_QUALITY__P7 2.0
  553. #define FXAA_QUALITY__P8 4.0
  554. #define FXAA_QUALITY__P9 8.0
  555. #endif
  556. /*--------------------------------------------------------------------------*/
  557. #if (FXAA_QUALITY__PRESET == 28)
  558. #define FXAA_QUALITY__PS 11
  559. #define FXAA_QUALITY__P0 1.0
  560. #define FXAA_QUALITY__P1 1.5
  561. #define FXAA_QUALITY__P2 2.0
  562. #define FXAA_QUALITY__P3 2.0
  563. #define FXAA_QUALITY__P4 2.0
  564. #define FXAA_QUALITY__P5 2.0
  565. #define FXAA_QUALITY__P6 2.0
  566. #define FXAA_QUALITY__P7 2.0
  567. #define FXAA_QUALITY__P8 2.0
  568. #define FXAA_QUALITY__P9 4.0
  569. #define FXAA_QUALITY__P10 8.0
  570. #endif
  571. /*--------------------------------------------------------------------------*/
  572. #if (FXAA_QUALITY__PRESET == 29)
  573. #define FXAA_QUALITY__PS 12
  574. #define FXAA_QUALITY__P0 1.0
  575. #define FXAA_QUALITY__P1 1.5
  576. #define FXAA_QUALITY__P2 2.0
  577. #define FXAA_QUALITY__P3 2.0
  578. #define FXAA_QUALITY__P4 2.0
  579. #define FXAA_QUALITY__P5 2.0
  580. #define FXAA_QUALITY__P6 2.0
  581. #define FXAA_QUALITY__P7 2.0
  582. #define FXAA_QUALITY__P8 2.0
  583. #define FXAA_QUALITY__P9 2.0
  584. #define FXAA_QUALITY__P10 4.0
  585. #define FXAA_QUALITY__P11 8.0
  586. #endif
  587.  
  588. /*============================================================================
  589. FXAA QUALITY - EXTREME QUALITY
  590. ============================================================================*/
  591. #if (FXAA_QUALITY__PRESET == 39)
  592. #define FXAA_QUALITY__PS 12
  593. #define FXAA_QUALITY__P0 1.0
  594. #define FXAA_QUALITY__P1 1.0
  595. #define FXAA_QUALITY__P2 1.0
  596. #define FXAA_QUALITY__P3 1.0
  597. #define FXAA_QUALITY__P4 1.0
  598. #define FXAA_QUALITY__P5 1.5
  599. #define FXAA_QUALITY__P6 2.0
  600. #define FXAA_QUALITY__P7 2.0
  601. #define FXAA_QUALITY__P8 2.0
  602. #define FXAA_QUALITY__P9 2.0
  603. #define FXAA_QUALITY__P10 4.0
  604. #define FXAA_QUALITY__P11 8.0
  605. #endif
  606.  
  607.  
  608.  
  609. /*============================================================================
  610.  
  611. API PORTING
  612.  
  613. ============================================================================*/
  614. #if (FXAA_GLSL_120 == 1) || (FXAA_GLSL_130 == 1)
  615. #define FxaaBool bool
  616. #define FxaaDiscard discard
  617. #define FxaaFloat float
  618. #define FxaaFloat2 vec2
  619. #define FxaaFloat3 vec3
  620. #define FxaaFloat4 vec4
  621. #define FxaaHalf float
  622. #define FxaaHalf2 vec2
  623. #define FxaaHalf3 vec3
  624. #define FxaaHalf4 vec4
  625. #define FxaaInt2 ivec2
  626. #define FxaaSat(x) clamp(x, 0.0, 1.0)
  627. #define FxaaTex sampler2D
  628. #else
  629. #define FxaaBool bool
  630. #define FxaaDiscard clip(-1)
  631. #define FxaaFloat float
  632. #define FxaaFloat2 float2
  633. #define FxaaFloat3 float3
  634. #define FxaaFloat4 float4
  635. #define FxaaHalf half
  636. #define FxaaHalf2 half2
  637. #define FxaaHalf3 half3
  638. #define FxaaHalf4 half4
  639. #define FxaaSat(x) saturate(x)
  640. #endif
  641. /*--------------------------------------------------------------------------*/
  642. #if (FXAA_GLSL_120 == 1)
  643. // Requires,
  644. // #version 120
  645. // And at least,
  646. // #extension GL_EXT_gpu_shader4 : enable
  647. // (or set FXAA_FAST_PIXEL_OFFSET 1 to work like DX9)
  648. #define FxaaTexTop(t, p) texture2DLod(t, p, 0.0)
  649. #if (FXAA_FAST_PIXEL_OFFSET == 1)
  650. #define FxaaTexOff(t, p, o, r) texture2DLodOffset(t, p, 0.0, o)
  651. #else
  652. #define FxaaTexOff(t, p, o, r) texture2DLod(t, p + (o * r), 0.0)
  653. #endif
  654. #if (FXAA_GATHER4_ALPHA == 1)
  655. // use #extension GL_ARB_gpu_shader5 : enable
  656. #define FxaaTexAlpha4(t, p) textureGather(t, p, 3)
  657. #define FxaaTexOffAlpha4(t, p, o) textureGatherOffset(t, p, o, 3)
  658. #define FxaaTexGreen4(t, p) textureGather(t, p, 1)
  659. #define FxaaTexOffGreen4(t, p, o) textureGatherOffset(t, p, o, 1)
  660. #endif
  661. #endif
  662. /*--------------------------------------------------------------------------*/
  663. #if (FXAA_GLSL_130 == 1)
  664. // Requires "#version 130" or better
  665. #define FxaaTexTop(t, p) textureLod(t, p, 0.0)
  666. #define FxaaTexOff(t, p, o, r) textureLodOffset(t, p, 0.0, o)
  667. #if (FXAA_GATHER4_ALPHA == 1)
  668. // use #extension GL_ARB_gpu_shader5 : enable
  669. #define FxaaTexAlpha4(t, p) textureGather(t, p, 3)
  670. #define FxaaTexOffAlpha4(t, p, o) textureGatherOffset(t, p, o, 3)
  671. #define FxaaTexGreen4(t, p) textureGather(t, p, 1)
  672. #define FxaaTexOffGreen4(t, p, o) textureGatherOffset(t, p, o, 1)
  673. #endif
  674. #endif
  675. /*--------------------------------------------------------------------------*/
  676. #if (FXAA_HLSL_3 == 1) || (FXAA_360 == 1) || (FXAA_PS3 == 1)
  677. #define FxaaInt2 float2
  678. #define FxaaTex sampler2D
  679. #define FxaaTexTop(t, p) tex2Dlod(t, float4(p, 0.0, 0.0))
  680. #define FxaaTexOff(t, p, o, r) tex2Dlod(t, float4(p + (o * r), 0, 0))
  681. #endif
  682. /*--------------------------------------------------------------------------*/
  683. #if (FXAA_HLSL_4 == 1)
  684. #define FxaaInt2 int2
  685. struct FxaaTex { SamplerState smpl; Texture2D tex; };
  686. #define FxaaTexTop(t, p) t.tex.SampleLevel(t.smpl, p, 0.0)
  687. #define FxaaTexOff(t, p, o, r) t.tex.SampleLevel(t.smpl, p, 0.0, o)
  688. #endif
  689. /*--------------------------------------------------------------------------*/
  690. #if (FXAA_HLSL_5 == 1)
  691. #define FxaaInt2 int2
  692. struct FxaaTex { SamplerState smpl; Texture2D tex; };
  693. #define FxaaTexTop(t, p) t.tex.SampleLevel(t.smpl, p, 0.0)
  694. #define FxaaTexOff(t, p, o, r) t.tex.SampleLevel(t.smpl, p, 0.0, o)
  695. #define FxaaTexAlpha4(t, p) t.tex.GatherAlpha(t.smpl, p)
  696. #define FxaaTexOffAlpha4(t, p, o) t.tex.GatherAlpha(t.smpl, p, o)
  697. #define FxaaTexGreen4(t, p) t.tex.GatherGreen(t.smpl, p)
  698. #define FxaaTexOffGreen4(t, p, o) t.tex.GatherGreen(t.smpl, p, o)
  699. #endif
  700.  
  701.  
  702. /*============================================================================
  703. GREEN AS LUMA OPTION SUPPORT FUNCTION
  704. ============================================================================*/
  705. #if (FXAA_GREEN_AS_LUMA == 0)
  706. FxaaFloat FxaaLuma(FxaaFloat4 rgba) { return rgba.w; }
  707. #else
  708. FxaaFloat FxaaLuma(FxaaFloat4 rgba) { return rgba.y; }
  709. #endif
  710.  
  711.  
  712.  
  713.  
  714. /*============================================================================
  715.  
  716. FXAA3 QUALITY - PC
  717.  
  718. ============================================================================*/
  719. #if (FXAA_PC == 1)
  720. /*--------------------------------------------------------------------------*/
  721. FxaaFloat4 FxaaPixelShader(
  722. //
  723. // Use noperspective interpolation here (turn off perspective interpolation).
  724. // {xy} = center of pixel
  725. FxaaFloat2 pos,
  726. //
  727. // Used only for FXAA Console, and not used on the 360 version.
  728. // Use noperspective interpolation here (turn off perspective interpolation).
  729. // {xy__} = upper left of pixel
  730. // {__zw} = lower right of pixel
  731. FxaaFloat4 fxaaConsolePosPos,
  732. //
  733. // Input color texture.
  734. // {rgb_} = color in linear or perceptual color space
  735. // if (FXAA_GREEN_AS_LUMA == 0)
  736. // {___a} = luma in perceptual color space (not linear)
  737. FxaaTex tex,
  738. //
  739. // Only used on the optimized 360 version of FXAA Console.
  740. // For everything but 360, just use the same input here as for "tex".
  741. // For 360, same texture, just alias with a 2nd sampler.
  742. // This sampler needs to have an exponent bias of -1.
  743. FxaaTex fxaaConsole360TexExpBiasNegOne,
  744. //
  745. // Only used on the optimized 360 version of FXAA Console.
  746. // For everything but 360, just use the same input here as for "tex".
  747. // For 360, same texture, just alias with a 3nd sampler.
  748. // This sampler needs to have an exponent bias of -2.
  749. FxaaTex fxaaConsole360TexExpBiasNegTwo,
  750. //
  751. // Only used on FXAA Quality.
  752. // This must be from a constant/uniform.
  753. // {x_} = 1.0/screenWidthInPixels
  754. // {_y} = 1.0/screenHeightInPixels
  755. FxaaFloat2 fxaaQualityRcpFrame,
  756. //
  757. // Only used on FXAA Console.
  758. // This must be from a constant/uniform.
  759. // This effects sub-pixel AA quality and inversely sharpness.
  760. // Where N ranges between,
  761. // N = 0.50 (default)
  762. // N = 0.33 (sharper)
  763. // {x___} = -N/screenWidthInPixels
  764. // {_y__} = -N/screenHeightInPixels
  765. // {__z_} = N/screenWidthInPixels
  766. // {___w} = N/screenHeightInPixels
  767. FxaaFloat4 fxaaConsoleRcpFrameOpt,
  768. //
  769. // Only used on FXAA Console.
  770. // Not used on 360, but used on PS3 and PC.
  771. // This must be from a constant/uniform.
  772. // {x___} = -2.0/screenWidthInPixels
  773. // {_y__} = -2.0/screenHeightInPixels
  774. // {__z_} = 2.0/screenWidthInPixels
  775. // {___w} = 2.0/screenHeightInPixels
  776. FxaaFloat4 fxaaConsoleRcpFrameOpt2,
  777. //
  778. // Only used on FXAA Console.
  779. // Only used on 360 in place of fxaaConsoleRcpFrameOpt2.
  780. // This must be from a constant/uniform.
  781. // {x___} = 8.0/screenWidthInPixels
  782. // {_y__} = 8.0/screenHeightInPixels
  783. // {__z_} = -4.0/screenWidthInPixels
  784. // {___w} = -4.0/screenHeightInPixels
  785. FxaaFloat4 fxaaConsole360RcpFrameOpt2,
  786. //
  787. // Only used on FXAA Quality.
  788. // This used to be the FXAA_QUALITY__SUBPIX define.
  789. // It is here now to allow easier tuning.
  790. // Choose the amount of sub-pixel aliasing removal.
  791. // This can effect sharpness.
  792. // 1.00 - upper limit (softer)
  793. // 0.75 - default amount of filtering
  794. // 0.50 - lower limit (sharper, less sub-pixel aliasing removal)
  795. // 0.25 - almost off
  796. // 0.00 - completely off
  797. FxaaFloat fxaaQualitySubpix,
  798. //
  799. // Only used on FXAA Quality.
  800. // This used to be the FXAA_QUALITY__EDGE_THRESHOLD define.
  801. // It is here now to allow easier tuning.
  802. // The minimum amount of local contrast required to apply algorithm.
  803. // 0.333 - too little (faster)
  804. // 0.250 - low quality
  805. // 0.166 - default
  806. // 0.125 - high quality
  807. // 0.063 - overkill (slower)
  808. FxaaFloat fxaaQualityEdgeThreshold,
  809. //
  810. // Only used on FXAA Quality.
  811. // This used to be the FXAA_QUALITY__EDGE_THRESHOLD_MIN define.
  812. // It is here now to allow easier tuning.
  813. // Trims the algorithm from processing darks.
  814. // 0.0833 - upper limit (default, the start of visible unfiltered edges)
  815. // 0.0625 - high quality (faster)
  816. // 0.0312 - visible limit (slower)
  817. // Special notes when using FXAA_GREEN_AS_LUMA,
  818. // Likely want to set this to zero.
  819. // As colors that are mostly not-green
  820. // will appear very dark in the green channel!
  821. // Tune by looking at mostly non-green content,
  822. // then start at zero and increase until aliasing is a problem.
  823. FxaaFloat fxaaQualityEdgeThresholdMin,
  824. //
  825. // Only used on FXAA Console.
  826. // This used to be the FXAA_CONSOLE__EDGE_SHARPNESS define.
  827. // It is here now to allow easier tuning.
  828. // This does not effect PS3, as this needs to be compiled in.
  829. // Use FXAA_CONSOLE__PS3_EDGE_SHARPNESS for PS3.
  830. // Due to the PS3 being ALU bound,
  831. // there are only three safe values here: 2 and 4 and 8.
  832. // These options use the shaders ability to a free *|/ by 2|4|8.
  833. // For all other platforms can be a non-power of two.
  834. // 8.0 is sharper (default!!!)
  835. // 4.0 is softer
  836. // 2.0 is really soft (good only for vector graphics inputs)
  837. FxaaFloat fxaaConsoleEdgeSharpness,
  838. //
  839. // Only used on FXAA Console.
  840. // This used to be the FXAA_CONSOLE__EDGE_THRESHOLD define.
  841. // It is here now to allow easier tuning.
  842. // This does not effect PS3, as this needs to be compiled in.
  843. // Use FXAA_CONSOLE__PS3_EDGE_THRESHOLD for PS3.
  844. // Due to the PS3 being ALU bound,
  845. // there are only two safe values here: 1/4 and 1/8.
  846. // These options use the shaders ability to a free *|/ by 2|4|8.
  847. // The console setting has a different mapping than the quality setting.
  848. // Other platforms can use other values.
  849. // 0.125 leaves less aliasing, but is softer (default!!!)
  850. // 0.25 leaves more aliasing, and is sharper
  851. FxaaFloat fxaaConsoleEdgeThreshold,
  852. //
  853. // Only used on FXAA Console.
  854. // This used to be the FXAA_CONSOLE__EDGE_THRESHOLD_MIN define.
  855. // It is here now to allow easier tuning.
  856. // Trims the algorithm from processing darks.
  857. // The console setting has a different mapping than the quality setting.
  858. // This only applies when FXAA_EARLY_EXIT is 1.
  859. // This does not apply to PS3,
  860. // PS3 was simplified to avoid more shader instructions.
  861. // 0.06 - faster but more aliasing in darks
  862. // 0.05 - default
  863. // 0.04 - slower and less aliasing in darks
  864. // Special notes when using FXAA_GREEN_AS_LUMA,
  865. // Likely want to set this to zero.
  866. // As colors that are mostly not-green
  867. // will appear very dark in the green channel!
  868. // Tune by looking at mostly non-green content,
  869. // then start at zero and increase until aliasing is a problem.
  870. FxaaFloat fxaaConsoleEdgeThresholdMin,
  871. //
  872. // Extra constants for 360 FXAA Console only.
  873. // Use zeros or anything else for other platforms.
  874. // These must be in physical constant registers and NOT immedates.
  875. // Immedates will result in compiler un-optimizing.
  876. // {xyzw} = float4(1.0, -1.0, 0.25, -0.25)
  877. FxaaFloat4 fxaaConsole360ConstDir
  878. ) {
  879. /*--------------------------------------------------------------------------*/
  880. FxaaFloat2 posM;
  881. posM.x = pos.x;
  882. posM.y = pos.y;
  883. #if (FXAA_GATHER4_ALPHA == 1)
  884. #if (FXAA_DISCARD == 0)
  885. FxaaFloat4 rgbyM = FxaaTexTop(tex, posM);
  886. #if (FXAA_GREEN_AS_LUMA == 0)
  887. #define lumaM rgbyM.w
  888. #else
  889. #define lumaM rgbyM.y
  890. #endif
  891. #endif
  892. #if (FXAA_GREEN_AS_LUMA == 0)
  893. FxaaFloat4 luma4A = FxaaTexAlpha4(tex, posM);
  894. FxaaFloat4 luma4B = FxaaTexOffAlpha4(tex, posM, FxaaInt2(-1, -1));
  895. #else
  896. FxaaFloat4 luma4A = FxaaTexGreen4(tex, posM);
  897. FxaaFloat4 luma4B = FxaaTexOffGreen4(tex, posM, FxaaInt2(-1, -1));
  898. #endif
  899. #if (FXAA_DISCARD == 1)
  900. #define lumaM luma4A.w
  901. #endif
  902. #define lumaE luma4A.z
  903. #define lumaS luma4A.x
  904. #define lumaSE luma4A.y
  905. #define lumaNW luma4B.w
  906. #define lumaN luma4B.z
  907. #define lumaW luma4B.x
  908. #else
  909. FxaaFloat4 rgbyM = FxaaTexTop(tex, posM);
  910. #if (FXAA_GREEN_AS_LUMA == 0)
  911. #define lumaM rgbyM.w
  912. #else
  913. #define lumaM rgbyM.y
  914. #endif
  915. FxaaFloat lumaS = FxaaLuma(FxaaTexOff(tex, posM, FxaaInt2( 0, 1), fxaaQualityRcpFrame.xy));
  916. FxaaFloat lumaE = FxaaLuma(FxaaTexOff(tex, posM, FxaaInt2( 1, 0), fxaaQualityRcpFrame.xy));
  917. FxaaFloat lumaN = FxaaLuma(FxaaTexOff(tex, posM, FxaaInt2( 0,-1), fxaaQualityRcpFrame.xy));
  918. FxaaFloat lumaW = FxaaLuma(FxaaTexOff(tex, posM, FxaaInt2(-1, 0), fxaaQualityRcpFrame.xy));
  919. #endif
  920. /*--------------------------------------------------------------------------*/
  921. FxaaFloat maxSM = max(lumaS, lumaM);
  922. FxaaFloat minSM = min(lumaS, lumaM);
  923. FxaaFloat maxESM = max(lumaE, maxSM);
  924. FxaaFloat minESM = min(lumaE, minSM);
  925. FxaaFloat maxWN = max(lumaN, lumaW);
  926. FxaaFloat minWN = min(lumaN, lumaW);
  927. FxaaFloat rangeMax = max(maxWN, maxESM);
  928. FxaaFloat rangeMin = min(minWN, minESM);
  929. FxaaFloat rangeMaxScaled = rangeMax * fxaaQualityEdgeThreshold;
  930. FxaaFloat range = rangeMax - rangeMin;
  931. FxaaFloat rangeMaxClamped = max(fxaaQualityEdgeThresholdMin, rangeMaxScaled);
  932. FxaaBool earlyExit = range < rangeMaxClamped;
  933. /*--------------------------------------------------------------------------*/
  934. if(earlyExit)
  935. #if (FXAA_DISCARD == 1)
  936. FxaaDiscard;
  937. #else
  938. return rgbyM;
  939. #endif
  940. /*--------------------------------------------------------------------------*/
  941. #if (FXAA_GATHER4_ALPHA == 0)
  942. FxaaFloat lumaNW = FxaaLuma(FxaaTexOff(tex, posM, FxaaInt2(-1,-1), fxaaQualityRcpFrame.xy));
  943. FxaaFloat lumaSE = FxaaLuma(FxaaTexOff(tex, posM, FxaaInt2( 1, 1), fxaaQualityRcpFrame.xy));
  944. FxaaFloat lumaNE = FxaaLuma(FxaaTexOff(tex, posM, FxaaInt2( 1,-1), fxaaQualityRcpFrame.xy));
  945. FxaaFloat lumaSW = FxaaLuma(FxaaTexOff(tex, posM, FxaaInt2(-1, 1), fxaaQualityRcpFrame.xy));
  946. #else
  947. FxaaFloat lumaNE = FxaaLuma(FxaaTexOff(tex, posM, FxaaInt2(1, -1), fxaaQualityRcpFrame.xy));
  948. FxaaFloat lumaSW = FxaaLuma(FxaaTexOff(tex, posM, FxaaInt2(-1, 1), fxaaQualityRcpFrame.xy));
  949. #endif
  950. /*--------------------------------------------------------------------------*/
  951. FxaaFloat lumaNS = lumaN + lumaS;
  952. FxaaFloat lumaWE = lumaW + lumaE;
  953. FxaaFloat subpixRcpRange = 1.0/range;
  954. FxaaFloat subpixNSWE = lumaNS + lumaWE;
  955. FxaaFloat edgeHorz1 = (-2.0 * lumaM) + lumaNS;
  956. FxaaFloat edgeVert1 = (-2.0 * lumaM) + lumaWE;
  957. /*--------------------------------------------------------------------------*/
  958. FxaaFloat lumaNESE = lumaNE + lumaSE;
  959. FxaaFloat lumaNWNE = lumaNW + lumaNE;
  960. FxaaFloat edgeHorz2 = (-2.0 * lumaE) + lumaNESE;
  961. FxaaFloat edgeVert2 = (-2.0 * lumaN) + lumaNWNE;
  962. /*--------------------------------------------------------------------------*/
  963. FxaaFloat lumaNWSW = lumaNW + lumaSW;
  964. FxaaFloat lumaSWSE = lumaSW + lumaSE;
  965. FxaaFloat edgeHorz4 = (abs(edgeHorz1) * 2.0) + abs(edgeHorz2);
  966. FxaaFloat edgeVert4 = (abs(edgeVert1) * 2.0) + abs(edgeVert2);
  967. FxaaFloat edgeHorz3 = (-2.0 * lumaW) + lumaNWSW;
  968. FxaaFloat edgeVert3 = (-2.0 * lumaS) + lumaSWSE;
  969. FxaaFloat edgeHorz = abs(edgeHorz3) + edgeHorz4;
  970. FxaaFloat edgeVert = abs(edgeVert3) + edgeVert4;
  971. /*--------------------------------------------------------------------------*/
  972. FxaaFloat subpixNWSWNESE = lumaNWSW + lumaNESE;
  973. FxaaFloat lengthSign = fxaaQualityRcpFrame.x;
  974. FxaaBool horzSpan = edgeHorz >= edgeVert;
  975. FxaaFloat subpixA = subpixNSWE * 2.0 + subpixNWSWNESE;
  976. /*--------------------------------------------------------------------------*/
  977. if(!horzSpan) lumaN = lumaW;
  978. if(!horzSpan) lumaS = lumaE;
  979. if(horzSpan) lengthSign = fxaaQualityRcpFrame.y;
  980. FxaaFloat subpixB = (subpixA * (1.0/12.0)) - lumaM;
  981. /*--------------------------------------------------------------------------*/
  982. FxaaFloat gradientN = lumaN - lumaM;
  983. FxaaFloat gradientS = lumaS - lumaM;
  984. FxaaFloat lumaNN = lumaN + lumaM;
  985. FxaaFloat lumaSS = lumaS + lumaM;
  986. FxaaBool pairN = abs(gradientN) >= abs(gradientS);
  987. FxaaFloat gradient = max(abs(gradientN), abs(gradientS));
  988. if(pairN) lengthSign = -lengthSign;
  989. FxaaFloat subpixC = FxaaSat(abs(subpixB) * subpixRcpRange);
  990. /*--------------------------------------------------------------------------*/
  991. FxaaFloat2 posB;
  992. posB.x = posM.x;
  993. posB.y = posM.y;
  994. FxaaFloat2 offNP;
  995. offNP.x = (!horzSpan) ? 0.0 : fxaaQualityRcpFrame.x;
  996. offNP.y = ( horzSpan) ? 0.0 : fxaaQualityRcpFrame.y;
  997. if(!horzSpan) posB.x += lengthSign * 0.5;
  998. if( horzSpan) posB.y += lengthSign * 0.5;
  999. /*--------------------------------------------------------------------------*/
  1000. FxaaFloat2 posN;
  1001. posN.x = posB.x - offNP.x * FXAA_QUALITY__P0;
  1002. posN.y = posB.y - offNP.y * FXAA_QUALITY__P0;
  1003. FxaaFloat2 posP;
  1004. posP.x = posB.x + offNP.x * FXAA_QUALITY__P0;
  1005. posP.y = posB.y + offNP.y * FXAA_QUALITY__P0;
  1006. FxaaFloat subpixD = ((-2.0)*subpixC) + 3.0;
  1007. FxaaFloat lumaEndN = FxaaLuma(FxaaTexTop(tex, posN));
  1008. FxaaFloat subpixE = subpixC * subpixC;
  1009. FxaaFloat lumaEndP = FxaaLuma(FxaaTexTop(tex, posP));
  1010. /*--------------------------------------------------------------------------*/
  1011. if(!pairN) lumaNN = lumaSS;
  1012. FxaaFloat gradientScaled = gradient * 1.0/4.0;
  1013. FxaaFloat lumaMM = lumaM - lumaNN * 0.5;
  1014. FxaaFloat subpixF = subpixD * subpixE;
  1015. FxaaBool lumaMLTZero = lumaMM < 0.0;
  1016. /*--------------------------------------------------------------------------*/
  1017. lumaEndN -= lumaNN * 0.5;
  1018. lumaEndP -= lumaNN * 0.5;
  1019. FxaaBool doneN = abs(lumaEndN) >= gradientScaled;
  1020. FxaaBool doneP = abs(lumaEndP) >= gradientScaled;
  1021. if(!doneN) posN.x -= offNP.x * FXAA_QUALITY__P1;
  1022. if(!doneN) posN.y -= offNP.y * FXAA_QUALITY__P1;
  1023. FxaaBool doneNP = (!doneN) || (!doneP);
  1024. if(!doneP) posP.x += offNP.x * FXAA_QUALITY__P1;
  1025. if(!doneP) posP.y += offNP.y * FXAA_QUALITY__P1;
  1026. /*--------------------------------------------------------------------------*/
  1027. if(doneNP) {
  1028. if(!doneN) lumaEndN = FxaaLuma(FxaaTexTop(tex, posN.xy));
  1029. if(!doneP) lumaEndP = FxaaLuma(FxaaTexTop(tex, posP.xy));
  1030. if(!doneN) lumaEndN = lumaEndN - lumaNN * 0.5;
  1031. if(!doneP) lumaEndP = lumaEndP - lumaNN * 0.5;
  1032. doneN = abs(lumaEndN) >= gradientScaled;
  1033. doneP = abs(lumaEndP) >= gradientScaled;
  1034. if(!doneN) posN.x -= offNP.x * FXAA_QUALITY__P2;
  1035. if(!doneN) posN.y -= offNP.y * FXAA_QUALITY__P2;
  1036. doneNP = (!doneN) || (!doneP);
  1037. if(!doneP) posP.x += offNP.x * FXAA_QUALITY__P2;
  1038. if(!doneP) posP.y += offNP.y * FXAA_QUALITY__P2;
  1039. /*--------------------------------------------------------------------------*/
  1040. #if (FXAA_QUALITY__PS > 3)
  1041. if(doneNP) {
  1042. if(!doneN) lumaEndN = FxaaLuma(FxaaTexTop(tex, posN.xy));
  1043. if(!doneP) lumaEndP = FxaaLuma(FxaaTexTop(tex, posP.xy));
  1044. if(!doneN) lumaEndN = lumaEndN - lumaNN * 0.5;
  1045. if(!doneP) lumaEndP = lumaEndP - lumaNN * 0.5;
  1046. doneN = abs(lumaEndN) >= gradientScaled;
  1047. doneP = abs(lumaEndP) >= gradientScaled;
  1048. if(!doneN) posN.x -= offNP.x * FXAA_QUALITY__P3;
  1049. if(!doneN) posN.y -= offNP.y * FXAA_QUALITY__P3;
  1050. doneNP = (!doneN) || (!doneP);
  1051. if(!doneP) posP.x += offNP.x * FXAA_QUALITY__P3;
  1052. if(!doneP) posP.y += offNP.y * FXAA_QUALITY__P3;
  1053. /*--------------------------------------------------------------------------*/
  1054. #if (FXAA_QUALITY__PS > 4)
  1055. if(doneNP) {
  1056. if(!doneN) lumaEndN = FxaaLuma(FxaaTexTop(tex, posN.xy));
  1057. if(!doneP) lumaEndP = FxaaLuma(FxaaTexTop(tex, posP.xy));
  1058. if(!doneN) lumaEndN = lumaEndN - lumaNN * 0.5;
  1059. if(!doneP) lumaEndP = lumaEndP - lumaNN * 0.5;
  1060. doneN = abs(lumaEndN) >= gradientScaled;
  1061. doneP = abs(lumaEndP) >= gradientScaled;
  1062. if(!doneN) posN.x -= offNP.x * FXAA_QUALITY__P4;
  1063. if(!doneN) posN.y -= offNP.y * FXAA_QUALITY__P4;
  1064. doneNP = (!doneN) || (!doneP);
  1065. if(!doneP) posP.x += offNP.x * FXAA_QUALITY__P4;
  1066. if(!doneP) posP.y += offNP.y * FXAA_QUALITY__P4;
  1067. /*--------------------------------------------------------------------------*/
  1068. #if (FXAA_QUALITY__PS > 5)
  1069. if(doneNP) {
  1070. if(!doneN) lumaEndN = FxaaLuma(FxaaTexTop(tex, posN.xy));
  1071. if(!doneP) lumaEndP = FxaaLuma(FxaaTexTop(tex, posP.xy));
  1072. if(!doneN) lumaEndN = lumaEndN - lumaNN * 0.5;
  1073. if(!doneP) lumaEndP = lumaEndP - lumaNN * 0.5;
  1074. doneN = abs(lumaEndN) >= gradientScaled;
  1075. doneP = abs(lumaEndP) >= gradientScaled;
  1076. if(!doneN) posN.x -= offNP.x * FXAA_QUALITY__P5;
  1077. if(!doneN) posN.y -= offNP.y * FXAA_QUALITY__P5;
  1078. doneNP = (!doneN) || (!doneP);
  1079. if(!doneP) posP.x += offNP.x * FXAA_QUALITY__P5;
  1080. if(!doneP) posP.y += offNP.y * FXAA_QUALITY__P5;
  1081. /*--------------------------------------------------------------------------*/
  1082. #if (FXAA_QUALITY__PS > 6)
  1083. if(doneNP) {
  1084. if(!doneN) lumaEndN = FxaaLuma(FxaaTexTop(tex, posN.xy));
  1085. if(!doneP) lumaEndP = FxaaLuma(FxaaTexTop(tex, posP.xy));
  1086. if(!doneN) lumaEndN = lumaEndN - lumaNN * 0.5;
  1087. if(!doneP) lumaEndP = lumaEndP - lumaNN * 0.5;
  1088. doneN = abs(lumaEndN) >= gradientScaled;
  1089. doneP = abs(lumaEndP) >= gradientScaled;
  1090. if(!doneN) posN.x -= offNP.x * FXAA_QUALITY__P6;
  1091. if(!doneN) posN.y -= offNP.y * FXAA_QUALITY__P6;
  1092. doneNP = (!doneN) || (!doneP);
  1093. if(!doneP) posP.x += offNP.x * FXAA_QUALITY__P6;
  1094. if(!doneP) posP.y += offNP.y * FXAA_QUALITY__P6;
  1095. /*--------------------------------------------------------------------------*/
  1096. #if (FXAA_QUALITY__PS > 7)
  1097. if(doneNP) {
  1098. if(!doneN) lumaEndN = FxaaLuma(FxaaTexTop(tex, posN.xy));
  1099. if(!doneP) lumaEndP = FxaaLuma(FxaaTexTop(tex, posP.xy));
  1100. if(!doneN) lumaEndN = lumaEndN - lumaNN * 0.5;
  1101. if(!doneP) lumaEndP = lumaEndP - lumaNN * 0.5;
  1102. doneN = abs(lumaEndN) >= gradientScaled;
  1103. doneP = abs(lumaEndP) >= gradientScaled;
  1104. if(!doneN) posN.x -= offNP.x * FXAA_QUALITY__P7;
  1105. if(!doneN) posN.y -= offNP.y * FXAA_QUALITY__P7;
  1106. doneNP = (!doneN) || (!doneP);
  1107. if(!doneP) posP.x += offNP.x * FXAA_QUALITY__P7;
  1108. if(!doneP) posP.y += offNP.y * FXAA_QUALITY__P7;
  1109. /*--------------------------------------------------------------------------*/
  1110. #if (FXAA_QUALITY__PS > 8)
  1111. if(doneNP) {
  1112. if(!doneN) lumaEndN = FxaaLuma(FxaaTexTop(tex, posN.xy));
  1113. if(!doneP) lumaEndP = FxaaLuma(FxaaTexTop(tex, posP.xy));
  1114. if(!doneN) lumaEndN = lumaEndN - lumaNN * 0.5;
  1115. if(!doneP) lumaEndP = lumaEndP - lumaNN * 0.5;
  1116. doneN = abs(lumaEndN) >= gradientScaled;
  1117. doneP = abs(lumaEndP) >= gradientScaled;
  1118. if(!doneN) posN.x -= offNP.x * FXAA_QUALITY__P8;
  1119. if(!doneN) posN.y -= offNP.y * FXAA_QUALITY__P8;
  1120. doneNP = (!doneN) || (!doneP);
  1121. if(!doneP) posP.x += offNP.x * FXAA_QUALITY__P8;
  1122. if(!doneP) posP.y += offNP.y * FXAA_QUALITY__P8;
  1123. /*--------------------------------------------------------------------------*/
  1124. #if (FXAA_QUALITY__PS > 9)
  1125. if(doneNP) {
  1126. if(!doneN) lumaEndN = FxaaLuma(FxaaTexTop(tex, posN.xy));
  1127. if(!doneP) lumaEndP = FxaaLuma(FxaaTexTop(tex, posP.xy));
  1128. if(!doneN) lumaEndN = lumaEndN - lumaNN * 0.5;
  1129. if(!doneP) lumaEndP = lumaEndP - lumaNN * 0.5;
  1130. doneN = abs(lumaEndN) >= gradientScaled;
  1131. doneP = abs(lumaEndP) >= gradientScaled;
  1132. if(!doneN) posN.x -= offNP.x * FXAA_QUALITY__P9;
  1133. if(!doneN) posN.y -= offNP.y * FXAA_QUALITY__P9;
  1134. doneNP = (!doneN) || (!doneP);
  1135. if(!doneP) posP.x += offNP.x * FXAA_QUALITY__P9;
  1136. if(!doneP) posP.y += offNP.y * FXAA_QUALITY__P9;
  1137. /*--------------------------------------------------------------------------*/
  1138. #if (FXAA_QUALITY__PS > 10)
  1139. if(doneNP) {
  1140. if(!doneN) lumaEndN = FxaaLuma(FxaaTexTop(tex, posN.xy));
  1141. if(!doneP) lumaEndP = FxaaLuma(FxaaTexTop(tex, posP.xy));
  1142. if(!doneN) lumaEndN = lumaEndN - lumaNN * 0.5;
  1143. if(!doneP) lumaEndP = lumaEndP - lumaNN * 0.5;
  1144. doneN = abs(lumaEndN) >= gradientScaled;
  1145. doneP = abs(lumaEndP) >= gradientScaled;
  1146. if(!doneN) posN.x -= offNP.x * FXAA_QUALITY__P10;
  1147. if(!doneN) posN.y -= offNP.y * FXAA_QUALITY__P10;
  1148. doneNP = (!doneN) || (!doneP);
  1149. if(!doneP) posP.x += offNP.x * FXAA_QUALITY__P10;
  1150. if(!doneP) posP.y += offNP.y * FXAA_QUALITY__P10;
  1151. /*--------------------------------------------------------------------------*/
  1152. #if (FXAA_QUALITY__PS > 11)
  1153. if(doneNP) {
  1154. if(!doneN) lumaEndN = FxaaLuma(FxaaTexTop(tex, posN.xy));
  1155. if(!doneP) lumaEndP = FxaaLuma(FxaaTexTop(tex, posP.xy));
  1156. if(!doneN) lumaEndN = lumaEndN - lumaNN * 0.5;
  1157. if(!doneP) lumaEndP = lumaEndP - lumaNN * 0.5;
  1158. doneN = abs(lumaEndN) >= gradientScaled;
  1159. doneP = abs(lumaEndP) >= gradientScaled;
  1160. if(!doneN) posN.x -= offNP.x * FXAA_QUALITY__P11;
  1161. if(!doneN) posN.y -= offNP.y * FXAA_QUALITY__P11;
  1162. doneNP = (!doneN) || (!doneP);
  1163. if(!doneP) posP.x += offNP.x * FXAA_QUALITY__P11;
  1164. if(!doneP) posP.y += offNP.y * FXAA_QUALITY__P11;
  1165. /*--------------------------------------------------------------------------*/
  1166. #if (FXAA_QUALITY__PS > 12)
  1167. if(doneNP) {
  1168. if(!doneN) lumaEndN = FxaaLuma(FxaaTexTop(tex, posN.xy));
  1169. if(!doneP) lumaEndP = FxaaLuma(FxaaTexTop(tex, posP.xy));
  1170. if(!doneN) lumaEndN = lumaEndN - lumaNN * 0.5;
  1171. if(!doneP) lumaEndP = lumaEndP - lumaNN * 0.5;
  1172. doneN = abs(lumaEndN) >= gradientScaled;
  1173. doneP = abs(lumaEndP) >= gradientScaled;
  1174. if(!doneN) posN.x -= offNP.x * FXAA_QUALITY__P12;
  1175. if(!doneN) posN.y -= offNP.y * FXAA_QUALITY__P12;
  1176. doneNP = (!doneN) || (!doneP);
  1177. if(!doneP) posP.x += offNP.x * FXAA_QUALITY__P12;
  1178. if(!doneP) posP.y += offNP.y * FXAA_QUALITY__P12;
  1179. /*--------------------------------------------------------------------------*/
  1180. }
  1181. #endif
  1182. /*--------------------------------------------------------------------------*/
  1183. }
  1184. #endif
  1185. /*--------------------------------------------------------------------------*/
  1186. }
  1187. #endif
  1188. /*--------------------------------------------------------------------------*/
  1189. }
  1190. #endif
  1191. /*--------------------------------------------------------------------------*/
  1192. }
  1193. #endif
  1194. /*--------------------------------------------------------------------------*/
  1195. }
  1196. #endif
  1197. /*--------------------------------------------------------------------------*/
  1198. }
  1199. #endif
  1200. /*--------------------------------------------------------------------------*/
  1201. }
  1202. #endif
  1203. /*--------------------------------------------------------------------------*/
  1204. }
  1205. #endif
  1206. /*--------------------------------------------------------------------------*/
  1207. }
  1208. #endif
  1209. /*--------------------------------------------------------------------------*/
  1210. }
  1211. /*--------------------------------------------------------------------------*/
  1212. FxaaFloat dstN = posM.x - posN.x;
  1213. FxaaFloat dstP = posP.x - posM.x;
  1214. if(!horzSpan) dstN = posM.y - posN.y;
  1215. if(!horzSpan) dstP = posP.y - posM.y;
  1216. /*--------------------------------------------------------------------------*/
  1217. FxaaBool goodSpanN = (lumaEndN < 0.0) != lumaMLTZero;
  1218. FxaaFloat spanLength = (dstP + dstN);
  1219. FxaaBool goodSpanP = (lumaEndP < 0.0) != lumaMLTZero;
  1220. FxaaFloat spanLengthRcp = 1.0/spanLength;
  1221. /*--------------------------------------------------------------------------*/
  1222. FxaaBool directionN = dstN < dstP;
  1223. FxaaFloat dst = min(dstN, dstP);
  1224. FxaaBool goodSpan = directionN ? goodSpanN : goodSpanP;
  1225. FxaaFloat subpixG = subpixF * subpixF;
  1226. FxaaFloat pixelOffset = (dst * (-spanLengthRcp)) + 0.5;
  1227. FxaaFloat subpixH = subpixG * fxaaQualitySubpix;
  1228. /*--------------------------------------------------------------------------*/
  1229. FxaaFloat pixelOffsetGood = goodSpan ? pixelOffset : 0.0;
  1230. FxaaFloat pixelOffsetSubpix = max(pixelOffsetGood, subpixH);
  1231. if(!horzSpan) posM.x += pixelOffsetSubpix * lengthSign;
  1232. if( horzSpan) posM.y += pixelOffsetSubpix * lengthSign;
  1233. #if (FXAA_DISCARD == 1)
  1234. return FxaaTexTop(tex, posM);
  1235. #else
  1236. return FxaaFloat4(FxaaTexTop(tex, posM).xyz, lumaM);
  1237. #endif
  1238. }
  1239. /*==========================================================================*/
  1240. #endif
  1241.  
  1242.  
  1243.  
  1244.  
  1245. /*============================================================================
  1246.  
  1247. FXAA3 CONSOLE - PC VERSION
  1248.  
  1249. ------------------------------------------------------------------------------
  1250. Instead of using this on PC, I'd suggest just using FXAA Quality with
  1251. #define FXAA_QUALITY__PRESET 10
  1252. Or
  1253. #define FXAA_QUALITY__PRESET 20
  1254. Either are higher qualilty and almost as fast as this on modern PC GPUs.
  1255. ============================================================================*/
  1256. #if (FXAA_PC_CONSOLE == 1)
  1257. /*--------------------------------------------------------------------------*/
  1258. FxaaFloat4 FxaaPixelShader(
  1259. // See FXAA Quality FxaaPixelShader() source for docs on Inputs!
  1260. FxaaFloat2 pos,
  1261. FxaaFloat4 fxaaConsolePosPos,
  1262. FxaaTex tex,
  1263. FxaaTex fxaaConsole360TexExpBiasNegOne,
  1264. FxaaTex fxaaConsole360TexExpBiasNegTwo,
  1265. FxaaFloat2 fxaaQualityRcpFrame,
  1266. FxaaFloat4 fxaaConsoleRcpFrameOpt,
  1267. FxaaFloat4 fxaaConsoleRcpFrameOpt2,
  1268. FxaaFloat4 fxaaConsole360RcpFrameOpt2,
  1269. FxaaFloat fxaaQualitySubpix,
  1270. FxaaFloat fxaaQualityEdgeThreshold,
  1271. FxaaFloat fxaaQualityEdgeThresholdMin,
  1272. FxaaFloat fxaaConsoleEdgeSharpness,
  1273. FxaaFloat fxaaConsoleEdgeThreshold,
  1274. FxaaFloat fxaaConsoleEdgeThresholdMin,
  1275. FxaaFloat4 fxaaConsole360ConstDir
  1276. ) {
  1277. /*--------------------------------------------------------------------------*/
  1278. FxaaFloat lumaNw = FxaaLuma(FxaaTexTop(tex, fxaaConsolePosPos.xy));
  1279. FxaaFloat lumaSw = FxaaLuma(FxaaTexTop(tex, fxaaConsolePosPos.xw));
  1280. FxaaFloat lumaNe = FxaaLuma(FxaaTexTop(tex, fxaaConsolePosPos.zy));
  1281. FxaaFloat lumaSe = FxaaLuma(FxaaTexTop(tex, fxaaConsolePosPos.zw));
  1282. /*--------------------------------------------------------------------------*/
  1283. FxaaFloat4 rgbyM = FxaaTexTop(tex, pos.xy);
  1284. #if (FXAA_GREEN_AS_LUMA == 0)
  1285. FxaaFloat lumaM = rgbyM.w;
  1286. #else
  1287. FxaaFloat lumaM = rgbyM.y;
  1288. #endif
  1289. /*--------------------------------------------------------------------------*/
  1290. FxaaFloat lumaMaxNwSw = max(lumaNw, lumaSw);
  1291. lumaNe += 1.0/384.0;
  1292. FxaaFloat lumaMinNwSw = min(lumaNw, lumaSw);
  1293. /*--------------------------------------------------------------------------*/
  1294. FxaaFloat lumaMaxNeSe = max(lumaNe, lumaSe);
  1295. FxaaFloat lumaMinNeSe = min(lumaNe, lumaSe);
  1296. /*--------------------------------------------------------------------------*/
  1297. FxaaFloat lumaMax = max(lumaMaxNeSe, lumaMaxNwSw);
  1298. FxaaFloat lumaMin = min(lumaMinNeSe, lumaMinNwSw);
  1299. /*--------------------------------------------------------------------------*/
  1300. FxaaFloat lumaMaxScaled = lumaMax * fxaaConsoleEdgeThreshold;
  1301. /*--------------------------------------------------------------------------*/
  1302. FxaaFloat lumaMinM = min(lumaMin, lumaM);
  1303. FxaaFloat lumaMaxScaledClamped = max(fxaaConsoleEdgeThresholdMin, lumaMaxScaled);
  1304. FxaaFloat lumaMaxM = max(lumaMax, lumaM);
  1305. FxaaFloat dirSwMinusNe = lumaSw - lumaNe;
  1306. FxaaFloat lumaMaxSubMinM = lumaMaxM - lumaMinM;
  1307. FxaaFloat dirSeMinusNw = lumaSe - lumaNw;
  1308. if(lumaMaxSubMinM < lumaMaxScaledClamped) return rgbyM;
  1309. /*--------------------------------------------------------------------------*/
  1310. FxaaFloat2 dir;
  1311. dir.x = dirSwMinusNe + dirSeMinusNw;
  1312. dir.y = dirSwMinusNe - dirSeMinusNw;
  1313. /*--------------------------------------------------------------------------*/
  1314. FxaaFloat2 dir1 = normalize(dir.xy);
  1315. FxaaFloat4 rgbyN1 = FxaaTexTop(tex, pos.xy - dir1 * fxaaConsoleRcpFrameOpt.zw);
  1316. FxaaFloat4 rgbyP1 = FxaaTexTop(tex, pos.xy + dir1 * fxaaConsoleRcpFrameOpt.zw);
  1317. /*--------------------------------------------------------------------------*/
  1318. FxaaFloat dirAbsMinTimesC = min(abs(dir1.x), abs(dir1.y)) * fxaaConsoleEdgeSharpness;
  1319. FxaaFloat2 dir2 = clamp(dir1.xy / dirAbsMinTimesC, -2.0, 2.0);
  1320. /*--------------------------------------------------------------------------*/
  1321. FxaaFloat4 rgbyN2 = FxaaTexTop(tex, pos.xy - dir2 * fxaaConsoleRcpFrameOpt2.zw);
  1322. FxaaFloat4 rgbyP2 = FxaaTexTop(tex, pos.xy + dir2 * fxaaConsoleRcpFrameOpt2.zw);
  1323. /*--------------------------------------------------------------------------*/
  1324. FxaaFloat4 rgbyA = rgbyN1 + rgbyP1;
  1325. FxaaFloat4 rgbyB = ((rgbyN2 + rgbyP2) * 0.25) + (rgbyA * 0.25);
  1326. /*--------------------------------------------------------------------------*/
  1327. #if (FXAA_GREEN_AS_LUMA == 0)
  1328. FxaaBool twoTap = (rgbyB.w < lumaMin) || (rgbyB.w > lumaMax);
  1329. #else
  1330. FxaaBool twoTap = (rgbyB.y < lumaMin) || (rgbyB.y > lumaMax);
  1331. #endif
  1332. if(twoTap) rgbyB.xyz = rgbyA.xyz * 0.5;
  1333. return rgbyB; }
  1334. /*==========================================================================*/
  1335. #endif
  1336.  
  1337.  
  1338.  
  1339. /*============================================================================
  1340.  
  1341. FXAA3 CONSOLE - 360 PIXEL SHADER
  1342.  
  1343. ------------------------------------------------------------------------------
  1344. This optimized version thanks to suggestions from Andy Luedke.
  1345. Should be fully tex bound in all cases.
  1346. As of the FXAA 3.11 release, I have still not tested this code,
  1347. however I fixed a bug which was in both FXAA 3.9 and FXAA 3.10.
  1348. And note this is replacing the old unoptimized version.
  1349. If it does not work, please let me know so I can fix it.
  1350. ============================================================================*/
  1351. #if (FXAA_360 == 1)
  1352. /*--------------------------------------------------------------------------*/
  1353. [reduceTempRegUsage(4)]
  1354. float4 FxaaPixelShader(
  1355. // See FXAA Quality FxaaPixelShader() source for docs on Inputs!
  1356. FxaaFloat2 pos,
  1357. FxaaFloat4 fxaaConsolePosPos,
  1358. FxaaTex tex,
  1359. FxaaTex fxaaConsole360TexExpBiasNegOne,
  1360. FxaaTex fxaaConsole360TexExpBiasNegTwo,
  1361. FxaaFloat2 fxaaQualityRcpFrame,
  1362. FxaaFloat4 fxaaConsoleRcpFrameOpt,
  1363. FxaaFloat4 fxaaConsoleRcpFrameOpt2,
  1364. FxaaFloat4 fxaaConsole360RcpFrameOpt2,
  1365. FxaaFloat fxaaQualitySubpix,
  1366. FxaaFloat fxaaQualityEdgeThreshold,
  1367. FxaaFloat fxaaQualityEdgeThresholdMin,
  1368. FxaaFloat fxaaConsoleEdgeSharpness,
  1369. FxaaFloat fxaaConsoleEdgeThreshold,
  1370. FxaaFloat fxaaConsoleEdgeThresholdMin,
  1371. FxaaFloat4 fxaaConsole360ConstDir
  1372. ) {
  1373. /*--------------------------------------------------------------------------*/
  1374. float4 lumaNwNeSwSe;
  1375. #if (FXAA_GREEN_AS_LUMA == 0)
  1376. asm {
  1377. tfetch2D lumaNwNeSwSe.w___, tex, pos.xy, OffsetX = -0.5, OffsetY = -0.5, UseComputedLOD=false
  1378. tfetch2D lumaNwNeSwSe._w__, tex, pos.xy, OffsetX = 0.5, OffsetY = -0.5, UseComputedLOD=false
  1379. tfetch2D lumaNwNeSwSe.__w_, tex, pos.xy, OffsetX = -0.5, OffsetY = 0.5, UseComputedLOD=false
  1380. tfetch2D lumaNwNeSwSe.___w, tex, pos.xy, OffsetX = 0.5, OffsetY = 0.5, UseComputedLOD=false
  1381. };
  1382. #else
  1383. asm {
  1384. tfetch2D lumaNwNeSwSe.y___, tex, pos.xy, OffsetX = -0.5, OffsetY = -0.5, UseComputedLOD=false
  1385. tfetch2D lumaNwNeSwSe._y__, tex, pos.xy, OffsetX = 0.5, OffsetY = -0.5, UseComputedLOD=false
  1386. tfetch2D lumaNwNeSwSe.__y_, tex, pos.xy, OffsetX = -0.5, OffsetY = 0.5, UseComputedLOD=false
  1387. tfetch2D lumaNwNeSwSe.___y, tex, pos.xy, OffsetX = 0.5, OffsetY = 0.5, UseComputedLOD=false
  1388. };
  1389. #endif
  1390. /*--------------------------------------------------------------------------*/
  1391. lumaNwNeSwSe.y += 1.0/384.0;
  1392. float2 lumaMinTemp = min(lumaNwNeSwSe.xy, lumaNwNeSwSe.zw);
  1393. float2 lumaMaxTemp = max(lumaNwNeSwSe.xy, lumaNwNeSwSe.zw);
  1394. float lumaMin = min(lumaMinTemp.x, lumaMinTemp.y);
  1395. float lumaMax = max(lumaMaxTemp.x, lumaMaxTemp.y);
  1396. /*--------------------------------------------------------------------------*/
  1397. float4 rgbyM = tex2Dlod(tex, float4(pos.xy, 0.0, 0.0));
  1398. #if (FXAA_GREEN_AS_LUMA == 0)
  1399. float lumaMinM = min(lumaMin, rgbyM.w);
  1400. float lumaMaxM = max(lumaMax, rgbyM.w);
  1401. #else
  1402. float lumaMinM = min(lumaMin, rgbyM.y);
  1403. float lumaMaxM = max(lumaMax, rgbyM.y);
  1404. #endif
  1405. if((lumaMaxM - lumaMinM) < max(fxaaConsoleEdgeThresholdMin, lumaMax * fxaaConsoleEdgeThreshold)) return rgbyM;
  1406. /*--------------------------------------------------------------------------*/
  1407. float2 dir;
  1408. dir.x = dot(lumaNwNeSwSe, fxaaConsole360ConstDir.yyxx);
  1409. dir.y = dot(lumaNwNeSwSe, fxaaConsole360ConstDir.xyxy);
  1410. dir = normalize(dir);
  1411. /*--------------------------------------------------------------------------*/
  1412. float4 dir1 = dir.xyxy * fxaaConsoleRcpFrameOpt.xyzw;
  1413. /*--------------------------------------------------------------------------*/
  1414. float4 dir2;
  1415. float dirAbsMinTimesC = min(abs(dir.x), abs(dir.y)) * fxaaConsoleEdgeSharpness;
  1416. dir2 = saturate(fxaaConsole360ConstDir.zzww * dir.xyxy / dirAbsMinTimesC + 0.5);
  1417. dir2 = dir2 * fxaaConsole360RcpFrameOpt2.xyxy + fxaaConsole360RcpFrameOpt2.zwzw;
  1418. /*--------------------------------------------------------------------------*/
  1419. float4 rgbyN1 = tex2Dlod(fxaaConsole360TexExpBiasNegOne, float4(pos.xy + dir1.xy, 0.0, 0.0));
  1420. float4 rgbyP1 = tex2Dlod(fxaaConsole360TexExpBiasNegOne, float4(pos.xy + dir1.zw, 0.0, 0.0));
  1421. float4 rgbyN2 = tex2Dlod(fxaaConsole360TexExpBiasNegTwo, float4(pos.xy + dir2.xy, 0.0, 0.0));
  1422. float4 rgbyP2 = tex2Dlod(fxaaConsole360TexExpBiasNegTwo, float4(pos.xy + dir2.zw, 0.0, 0.0));
  1423. /*--------------------------------------------------------------------------*/
  1424. float4 rgbyA = rgbyN1 + rgbyP1;
  1425. float4 rgbyB = rgbyN2 + rgbyP2 + rgbyA * 0.5;
  1426. /*--------------------------------------------------------------------------*/
  1427. float4 rgbyR = ((FxaaLuma(rgbyB) - lumaMax) > 0.0) ? rgbyA : rgbyB;
  1428. rgbyR = ((FxaaLuma(rgbyB) - lumaMin) > 0.0) ? rgbyR : rgbyA;
  1429. return rgbyR; }
  1430. /*==========================================================================*/
  1431. #endif
  1432.  
  1433.  
  1434.  
  1435. /*============================================================================
  1436.  
  1437. FXAA3 CONSOLE - OPTIMIZED PS3 PIXEL SHADER (NO EARLY EXIT)
  1438.  
  1439. ==============================================================================
  1440. The code below does not exactly match the assembly.
  1441. I have a feeling that 12 cycles is possible, but was not able to get there.
  1442. Might have to increase register count to get full performance.
  1443. Note this shader does not use perspective interpolation.
  1444.  
  1445. Use the following cgc options,
  1446.  
  1447. --fenable-bx2 --fastmath --fastprecision --nofloatbindings
  1448.  
  1449. ------------------------------------------------------------------------------
  1450. NVSHADERPERF OUTPUT
  1451. ------------------------------------------------------------------------------
  1452. For reference and to aid in debug, output of NVShaderPerf should match this,
  1453.  
  1454. Shader to schedule:
  1455. 0: texpkb h0.w(TRUE), v5.zyxx, #0
  1456. 2: addh h2.z(TRUE), h0.w, constant(0.001953, 0.000000, 0.000000, 0.000000).x
  1457. 4: texpkb h0.w(TRUE), v5.xwxx, #0
  1458. 6: addh h0.z(TRUE), -h2, h0.w
  1459. 7: texpkb h1.w(TRUE), v5, #0
  1460. 9: addh h0.x(TRUE), h0.z, -h1.w
  1461. 10: addh h3.w(TRUE), h0.z, h1
  1462. 11: texpkb h2.w(TRUE), v5.zwzz, #0
  1463. 13: addh h0.z(TRUE), h3.w, -h2.w
  1464. 14: addh h0.x(TRUE), h2.w, h0
  1465. 15: nrmh h1.xz(TRUE), h0_n
  1466. 16: minh_m8 h0.x(TRUE), |h1|, |h1.z|
  1467. 17: maxh h4.w(TRUE), h0, h1
  1468. 18: divx h2.xy(TRUE), h1_n.xzzw, h0_n
  1469. 19: movr r1.zw(TRUE), v4.xxxy
  1470. 20: madr r2.xz(TRUE), -h1, constant(cConst5.x, cConst5.y, cConst5.z, cConst5.w).zzww, r1.zzww
  1471. 22: minh h5.w(TRUE), h0, h1
  1472. 23: texpkb h0(TRUE), r2.xzxx, #0
  1473. 25: madr r0.zw(TRUE), h1.xzxz, constant(cConst5.x, cConst5.y, cConst5.z, cConst5.w), r1
  1474. 27: maxh h4.x(TRUE), h2.z, h2.w
  1475. 28: texpkb h1(TRUE), r0.zwzz, #0
  1476. 30: addh_d2 h1(TRUE), h0, h1
  1477. 31: madr r0.xy(TRUE), -h2, constant(cConst5.x, cConst5.y, cConst5.z, cConst5.w).xyxx, r1.zwzz
  1478. 33: texpkb h0(TRUE), r0, #0
  1479. 35: minh h4.z(TRUE), h2, h2.w
  1480. 36: fenct TRUE
  1481. 37: madr r1.xy(TRUE), h2, constant(cConst5.x, cConst5.y, cConst5.z, cConst5.w).xyxx, r1.zwzz
  1482. 39: texpkb h2(TRUE), r1, #0
  1483. 41: addh_d2 h0(TRUE), h0, h2
  1484. 42: maxh h2.w(TRUE), h4, h4.x
  1485. 43: minh h2.x(TRUE), h5.w, h4.z
  1486. 44: addh_d2 h0(TRUE), h0, h1
  1487. 45: slth h2.x(TRUE), h0.w, h2
  1488. 46: sgth h2.w(TRUE), h0, h2
  1489. 47: movh h0(TRUE), h0
  1490. 48: addx.c0 rc(TRUE), h2, h2.w
  1491. 49: movh h0(c0.NE.x), h1
  1492.  
  1493. IPU0 ------ Simplified schedule: --------
  1494. Pass | Unit | uOp | PC: Op
  1495. -----+--------+------+-------------------------
  1496. 1 | SCT0/1 | mov | 0: TXLr h0.w, g[TEX1].zyxx, const.xxxx, TEX0;
  1497. | TEX | txl | 0: TXLr h0.w, g[TEX1].zyxx, const.xxxx, TEX0;
  1498. | SCB1 | add | 2: ADDh h2.z, h0.--w-, const.--x-;
  1499. | | |
  1500. 2 | SCT0/1 | mov | 4: TXLr h0.w, g[TEX1].xwxx, const.xxxx, TEX0;
  1501. | TEX | txl | 4: TXLr h0.w, g[TEX1].xwxx, const.xxxx, TEX0;
  1502. | SCB1 | add | 6: ADDh h0.z,-h2, h0.--w-;
  1503. | | |
  1504. 3 | SCT0/1 | mov | 7: TXLr h1.w, g[TEX1], const.xxxx, TEX0;
  1505. | TEX | txl | 7: TXLr h1.w, g[TEX1], const.xxxx, TEX0;
  1506. | SCB0 | add | 9: ADDh h0.x, h0.z---,-h1.w---;
  1507. | SCB1 | add | 10: ADDh h3.w, h0.---z, h1;
  1508. | | |
  1509. 4 | SCT0/1 | mov | 11: TXLr h2.w, g[TEX1].zwzz, const.xxxx, TEX0;
  1510. | TEX | txl | 11: TXLr h2.w, g[TEX1].zwzz, const.xxxx, TEX0;
  1511. | SCB0 | add | 14: ADDh h0.x, h2.w---, h0;
  1512. | SCB1 | add | 13: ADDh h0.z, h3.--w-,-h2.--w-;
  1513. | | |
  1514. 5 | SCT1 | mov | 15: NRMh h1.xz, h0;
  1515. | SRB | nrm | 15: NRMh h1.xz, h0;
  1516. | SCB0 | min | 16: MINh*8 h0.x, |h1|, |h1.z---|;
  1517. | SCB1 | max | 17: MAXh h4.w, h0, h1;
  1518. | | |
  1519. 6 | SCT0 | div | 18: DIVx h2.xy, h1.xz--, h0;
  1520. | SCT1 | mov | 19: MOVr r1.zw, g[TEX0].--xy;
  1521. | SCB0 | mad | 20: MADr r2.xz,-h1, const.z-w-, r1.z-w-;
  1522. | SCB1 | min | 22: MINh h5.w, h0, h1;
  1523. | | |
  1524. 7 | SCT0/1 | mov | 23: TXLr h0, r2.xzxx, const.xxxx, TEX0;
  1525. | TEX | txl | 23: TXLr h0, r2.xzxx, const.xxxx, TEX0;
  1526. | SCB0 | max | 27: MAXh h4.x, h2.z---, h2.w---;
  1527. | SCB1 | mad | 25: MADr r0.zw, h1.--xz, const, r1;
  1528. | | |
  1529. 8 | SCT0/1 | mov | 28: TXLr h1, r0.zwzz, const.xxxx, TEX0;
  1530. | TEX | txl | 28: TXLr h1, r0.zwzz, const.xxxx, TEX0;
  1531. | SCB0/1 | add | 30: ADDh/2 h1, h0, h1;
  1532. | | |
  1533. 9 | SCT0 | mad | 31: MADr r0.xy,-h2, const.xy--, r1.zw--;
  1534. | SCT1 | mov | 33: TXLr h0, r0, const.zzzz, TEX0;
  1535. | TEX | txl | 33: TXLr h0, r0, const.zzzz, TEX0;
  1536. | SCB1 | min | 35: MINh h4.z, h2, h2.--w-;
  1537. | | |
  1538. 10 | SCT0 | mad | 37: MADr r1.xy, h2, const.xy--, r1.zw--;
  1539. | SCT1 | mov | 39: TXLr h2, r1, const.zzzz, TEX0;
  1540. | TEX | txl | 39: TXLr h2, r1, const.zzzz, TEX0;
  1541. | SCB0/1 | add | 41: ADDh/2 h0, h0, h2;
  1542. | | |
  1543. 11 | SCT0 | min | 43: MINh h2.x, h5.w---, h4.z---;
  1544. | SCT1 | max | 42: MAXh h2.w, h4, h4.---x;
  1545. | SCB0/1 | add | 44: ADDh/2 h0, h0, h1;
  1546. | | |
  1547. 12 | SCT0 | set | 45: SLTh h2.x, h0.w---, h2;
  1548. | SCT1 | set | 46: SGTh h2.w, h0, h2;
  1549. | SCB0/1 | mul | 47: MOVh h0, h0;
  1550. | | |
  1551. 13 | SCT0 | mad | 48: ADDxc0_s rc, h2, h2.w---;
  1552. | SCB0/1 | mul | 49: MOVh h0(NE0.xxxx), h1;
  1553.  
  1554. Pass SCT TEX SCB
  1555. 1: 0% 100% 25%
  1556. 2: 0% 100% 25%
  1557. 3: 0% 100% 50%
  1558. 4: 0% 100% 50%
  1559. 5: 0% 0% 50%
  1560. 6: 100% 0% 75%
  1561. 7: 0% 100% 75%
  1562. 8: 0% 100% 100%
  1563. 9: 0% 100% 25%
  1564. 10: 0% 100% 100%
  1565. 11: 50% 0% 100%
  1566. 12: 50% 0% 100%
  1567. 13: 25% 0% 100%
  1568.  
  1569. MEAN: 17% 61% 67%
  1570.  
  1571. Pass SCT0 SCT1 TEX SCB0 SCB1
  1572. 1: 0% 0% 100% 0% 100%
  1573. 2: 0% 0% 100% 0% 100%
  1574. 3: 0% 0% 100% 100% 100%
  1575. 4: 0% 0% 100% 100% 100%
  1576. 5: 0% 0% 0% 100% 100%
  1577. 6: 100% 100% 0% 100% 100%
  1578. 7: 0% 0% 100% 100% 100%
  1579. 8: 0% 0% 100% 100% 100%
  1580. 9: 0% 0% 100% 0% 100%
  1581. 10: 0% 0% 100% 100% 100%
  1582. 11: 100% 100% 0% 100% 100%
  1583. 12: 100% 100% 0% 100% 100%
  1584. 13: 100% 0% 0% 100% 100%
  1585.  
  1586. MEAN: 30% 23% 61% 76% 100%
  1587. Fragment Performance Setup: Driver RSX Compiler, GPU RSX, Flags 0x5
  1588. Results 13 cycles, 3 r regs, 923,076,923 pixels/s
  1589. ============================================================================*/
  1590. #if (FXAA_PS3 == 1) && (FXAA_EARLY_EXIT == 0)
  1591. /*--------------------------------------------------------------------------*/
  1592. #pragma regcount 7
  1593. #pragma disablepc all
  1594. #pragma option O3
  1595. #pragma option OutColorPrec=fp16
  1596. #pragma texformat default RGBA8
  1597. /*==========================================================================*/
  1598. half4 FxaaPixelShader(
  1599. // See FXAA Quality FxaaPixelShader() source for docs on Inputs!
  1600. FxaaFloat2 pos,
  1601. FxaaFloat4 fxaaConsolePosPos,
  1602. FxaaTex tex,
  1603. FxaaTex fxaaConsole360TexExpBiasNegOne,
  1604. FxaaTex fxaaConsole360TexExpBiasNegTwo,
  1605. FxaaFloat2 fxaaQualityRcpFrame,
  1606. FxaaFloat4 fxaaConsoleRcpFrameOpt,
  1607. FxaaFloat4 fxaaConsoleRcpFrameOpt2,
  1608. FxaaFloat4 fxaaConsole360RcpFrameOpt2,
  1609. FxaaFloat fxaaQualitySubpix,
  1610. FxaaFloat fxaaQualityEdgeThreshold,
  1611. FxaaFloat fxaaQualityEdgeThresholdMin,
  1612. FxaaFloat fxaaConsoleEdgeSharpness,
  1613. FxaaFloat fxaaConsoleEdgeThreshold,
  1614. FxaaFloat fxaaConsoleEdgeThresholdMin,
  1615. FxaaFloat4 fxaaConsole360ConstDir
  1616. ) {
  1617. /*--------------------------------------------------------------------------*/
  1618. // (1)
  1619. half4 dir;
  1620. half4 lumaNe = h4tex2Dlod(tex, half4(fxaaConsolePosPos.zy, 0, 0));
  1621. #if (FXAA_GREEN_AS_LUMA == 0)
  1622. lumaNe.w += half(1.0/512.0);
  1623. dir.x = -lumaNe.w;
  1624. dir.z = -lumaNe.w;
  1625. #else
  1626. lumaNe.y += half(1.0/512.0);
  1627. dir.x = -lumaNe.y;
  1628. dir.z = -lumaNe.y;
  1629. #endif
  1630. /*--------------------------------------------------------------------------*/
  1631. // (2)
  1632. half4 lumaSw = h4tex2Dlod(tex, half4(fxaaConsolePosPos.xw, 0, 0));
  1633. #if (FXAA_GREEN_AS_LUMA == 0)
  1634. dir.x += lumaSw.w;
  1635. dir.z += lumaSw.w;
  1636. #else
  1637. dir.x += lumaSw.y;
  1638. dir.z += lumaSw.y;
  1639. #endif
  1640. /*--------------------------------------------------------------------------*/
  1641. // (3)
  1642. half4 lumaNw = h4tex2Dlod(tex, half4(fxaaConsolePosPos.xy, 0, 0));
  1643. #if (FXAA_GREEN_AS_LUMA == 0)
  1644. dir.x -= lumaNw.w;
  1645. dir.z += lumaNw.w;
  1646. #else
  1647. dir.x -= lumaNw.y;
  1648. dir.z += lumaNw.y;
  1649. #endif
  1650. /*--------------------------------------------------------------------------*/
  1651. // (4)
  1652. half4 lumaSe = h4tex2Dlod(tex, half4(fxaaConsolePosPos.zw, 0, 0));
  1653. #if (FXAA_GREEN_AS_LUMA == 0)
  1654. dir.x += lumaSe.w;
  1655. dir.z -= lumaSe.w;
  1656. #else
  1657. dir.x += lumaSe.y;
  1658. dir.z -= lumaSe.y;
  1659. #endif
  1660. /*--------------------------------------------------------------------------*/
  1661. // (5)
  1662. half4 dir1_pos;
  1663. dir1_pos.xy = normalize(dir.xyz).xz;
  1664. half dirAbsMinTimesC = min(abs(dir1_pos.x), abs(dir1_pos.y)) * half(FXAA_CONSOLE__PS3_EDGE_SHARPNESS);
  1665. /*--------------------------------------------------------------------------*/
  1666. // (6)
  1667. half4 dir2_pos;
  1668. dir2_pos.xy = clamp(dir1_pos.xy / dirAbsMinTimesC, half(-2.0), half(2.0));
  1669. dir1_pos.zw = pos.xy;
  1670. dir2_pos.zw = pos.xy;
  1671. half4 temp1N;
  1672. temp1N.xy = dir1_pos.zw - dir1_pos.xy * fxaaConsoleRcpFrameOpt.zw;
  1673. /*--------------------------------------------------------------------------*/
  1674. // (7)
  1675. temp1N = h4tex2Dlod(tex, half4(temp1N.xy, 0.0, 0.0));
  1676. half4 rgby1;
  1677. rgby1.xy = dir1_pos.zw + dir1_pos.xy * fxaaConsoleRcpFrameOpt.zw;
  1678. /*--------------------------------------------------------------------------*/
  1679. // (8)
  1680. rgby1 = h4tex2Dlod(tex, half4(rgby1.xy, 0.0, 0.0));
  1681. rgby1 = (temp1N + rgby1) * 0.5;
  1682. /*--------------------------------------------------------------------------*/
  1683. // (9)
  1684. half4 temp2N;
  1685. temp2N.xy = dir2_pos.zw - dir2_pos.xy * fxaaConsoleRcpFrameOpt2.zw;
  1686. temp2N = h4tex2Dlod(tex, half4(temp2N.xy, 0.0, 0.0));
  1687. /*--------------------------------------------------------------------------*/
  1688. // (10)
  1689. half4 rgby2;
  1690. rgby2.xy = dir2_pos.zw + dir2_pos.xy * fxaaConsoleRcpFrameOpt2.zw;
  1691. rgby2 = h4tex2Dlod(tex, half4(rgby2.xy, 0.0, 0.0));
  1692. rgby2 = (temp2N + rgby2) * 0.5;
  1693. /*--------------------------------------------------------------------------*/
  1694. // (11)
  1695. // compilier moves these scalar ops up to other cycles
  1696. #if (FXAA_GREEN_AS_LUMA == 0)
  1697. half lumaMin = min(min(lumaNw.w, lumaSw.w), min(lumaNe.w, lumaSe.w));
  1698. half lumaMax = max(max(lumaNw.w, lumaSw.w), max(lumaNe.w, lumaSe.w));
  1699. #else
  1700. half lumaMin = min(min(lumaNw.y, lumaSw.y), min(lumaNe.y, lumaSe.y));
  1701. half lumaMax = max(max(lumaNw.y, lumaSw.y), max(lumaNe.y, lumaSe.y));
  1702. #endif
  1703. rgby2 = (rgby2 + rgby1) * 0.5;
  1704. /*--------------------------------------------------------------------------*/
  1705. // (12)
  1706. #if (FXAA_GREEN_AS_LUMA == 0)
  1707. bool twoTapLt = rgby2.w < lumaMin;
  1708. bool twoTapGt = rgby2.w > lumaMax;
  1709. #else
  1710. bool twoTapLt = rgby2.y < lumaMin;
  1711. bool twoTapGt = rgby2.y > lumaMax;
  1712. #endif
  1713. /*--------------------------------------------------------------------------*/
  1714. // (13)
  1715. if(twoTapLt || twoTapGt) rgby2 = rgby1;
  1716. /*--------------------------------------------------------------------------*/
  1717. return rgby2; }
  1718. /*==========================================================================*/
  1719. #endif
  1720.  
  1721.  
  1722.  
  1723. /*============================================================================
  1724.  
  1725. FXAA3 CONSOLE - OPTIMIZED PS3 PIXEL SHADER (WITH EARLY EXIT)
  1726.  
  1727. ==============================================================================
  1728. The code mostly matches the assembly.
  1729. I have a feeling that 14 cycles is possible, but was not able to get there.
  1730. Might have to increase register count to get full performance.
  1731. Note this shader does not use perspective interpolation.
  1732.  
  1733. Use the following cgc options,
  1734.  
  1735. --fenable-bx2 --fastmath --fastprecision --nofloatbindings
  1736.  
  1737. Use of FXAA_GREEN_AS_LUMA currently adds a cycle (16 clks).
  1738. Will look at fixing this for FXAA 3.12.
  1739. ------------------------------------------------------------------------------
  1740. NVSHADERPERF OUTPUT
  1741. ------------------------------------------------------------------------------
  1742. For reference and to aid in debug, output of NVShaderPerf should match this,
  1743.  
  1744. Shader to schedule:
  1745. 0: texpkb h0.w(TRUE), v5.zyxx, #0
  1746. 2: addh h2.y(TRUE), h0.w, constant(0.001953, 0.000000, 0.000000, 0.000000).x
  1747. 4: texpkb h1.w(TRUE), v5.xwxx, #0
  1748. 6: addh h0.x(TRUE), h1.w, -h2.y
  1749. 7: texpkb h2.w(TRUE), v5.zwzz, #0
  1750. 9: minh h4.w(TRUE), h2.y, h2
  1751. 10: maxh h5.x(TRUE), h2.y, h2.w
  1752. 11: texpkb h0.w(TRUE), v5, #0
  1753. 13: addh h3.w(TRUE), -h0, h0.x
  1754. 14: addh h0.x(TRUE), h0.w, h0
  1755. 15: addh h0.z(TRUE), -h2.w, h0.x
  1756. 16: addh h0.x(TRUE), h2.w, h3.w
  1757. 17: minh h5.y(TRUE), h0.w, h1.w
  1758. 18: nrmh h2.xz(TRUE), h0_n
  1759. 19: minh_m8 h2.w(TRUE), |h2.x|, |h2.z|
  1760. 20: divx h4.xy(TRUE), h2_n.xzzw, h2_n.w
  1761. 21: movr r1.zw(TRUE), v4.xxxy
  1762. 22: maxh h2.w(TRUE), h0, h1
  1763. 23: fenct TRUE
  1764. 24: madr r0.xy(TRUE), -h2.xzzw, constant(cConst5.x, cConst5.y, cConst5.z, cConst5.w).zwzz, r1.zwzz
  1765. 26: texpkb h0(TRUE), r0, #0
  1766. 28: maxh h5.x(TRUE), h2.w, h5
  1767. 29: minh h5.w(TRUE), h5.y, h4
  1768. 30: madr r1.xy(TRUE), h2.xzzw, constant(cConst5.x, cConst5.y, cConst5.z, cConst5.w).zwzz, r1.zwzz
  1769. 32: texpkb h2(TRUE), r1, #0
  1770. 34: addh_d2 h2(TRUE), h0, h2
  1771. 35: texpkb h1(TRUE), v4, #0
  1772. 37: maxh h5.y(TRUE), h5.x, h1.w
  1773. 38: minh h4.w(TRUE), h1, h5
  1774. 39: madr r0.xy(TRUE), -h4, constant(cConst5.x, cConst5.y, cConst5.z, cConst5.w).xyxx, r1.zwzz
  1775. 41: texpkb h0(TRUE), r0, #0
  1776. 43: addh_m8 h5.z(TRUE), h5.y, -h4.w
  1777. 44: madr r2.xy(TRUE), h4, constant(cConst5.x, cConst5.y, cConst5.z, cConst5.w).xyxx, r1.zwzz
  1778. 46: texpkb h3(TRUE), r2, #0
  1779. 48: addh_d2 h0(TRUE), h0, h3
  1780. 49: addh_d2 h3(TRUE), h0, h2
  1781. 50: movh h0(TRUE), h3
  1782. 51: slth h3.x(TRUE), h3.w, h5.w
  1783. 52: sgth h3.w(TRUE), h3, h5.x
  1784. 53: addx.c0 rc(TRUE), h3.x, h3
  1785. 54: slth.c0 rc(TRUE), h5.z, h5
  1786. 55: movh h0(c0.NE.w), h2
  1787. 56: movh h0(c0.NE.x), h1
  1788.  
  1789. IPU0 ------ Simplified schedule: --------
  1790. Pass | Unit | uOp | PC: Op
  1791. -----+--------+------+-------------------------
  1792. 1 | SCT0/1 | mov | 0: TXLr h0.w, g[TEX1].zyxx, const.xxxx, TEX0;
  1793. | TEX | txl | 0: TXLr h0.w, g[TEX1].zyxx, const.xxxx, TEX0;
  1794. | SCB0 | add | 2: ADDh h2.y, h0.-w--, const.-x--;
  1795. | | |
  1796. 2 | SCT0/1 | mov | 4: TXLr h1.w, g[TEX1].xwxx, const.xxxx, TEX0;
  1797. | TEX | txl | 4: TXLr h1.w, g[TEX1].xwxx, const.xxxx, TEX0;
  1798. | SCB0 | add | 6: ADDh h0.x, h1.w---,-h2.y---;
  1799. | | |
  1800. 3 | SCT0/1 | mov | 7: TXLr h2.w, g[TEX1].zwzz, const.xxxx, TEX0;
  1801. | TEX | txl | 7: TXLr h2.w, g[TEX1].zwzz, const.xxxx, TEX0;
  1802. | SCB0 | max | 10: MAXh h5.x, h2.y---, h2.w---;
  1803. | SCB1 | min | 9: MINh h4.w, h2.---y, h2;
  1804. | | |
  1805. 4 | SCT0/1 | mov | 11: TXLr h0.w, g[TEX1], const.xxxx, TEX0;
  1806. | TEX | txl | 11: TXLr h0.w, g[TEX1], const.xxxx, TEX0;
  1807. | SCB0 | add | 14: ADDh h0.x, h0.w---, h0;
  1808. | SCB1 | add | 13: ADDh h3.w,-h0, h0.---x;
  1809. | | |
  1810. 5 | SCT0 | mad | 16: ADDh h0.x, h2.w---, h3.w---;
  1811. | SCT1 | mad | 15: ADDh h0.z,-h2.--w-, h0.--x-;
  1812. | SCB0 | min | 17: MINh h5.y, h0.-w--, h1.-w--;
  1813. | | |
  1814. 6 | SCT1 | mov | 18: NRMh h2.xz, h0;
  1815. | SRB | nrm | 18: NRMh h2.xz, h0;
  1816. | SCB1 | min | 19: MINh*8 h2.w, |h2.---x|, |h2.---z|;
  1817. | | |
  1818. 7 | SCT0 | div | 20: DIVx h4.xy, h2.xz--, h2.ww--;
  1819. | SCT1 | mov | 21: MOVr r1.zw, g[TEX0].--xy;
  1820. | SCB1 | max | 22: MAXh h2.w, h0, h1;
  1821. | | |
  1822. 8 | SCT0 | mad | 24: MADr r0.xy,-h2.xz--, const.zw--, r1.zw--;
  1823. | SCT1 | mov | 26: TXLr h0, r0, const.xxxx, TEX0;
  1824. | TEX | txl | 26: TXLr h0, r0, const.xxxx, TEX0;
  1825. | SCB0 | max | 28: MAXh h5.x, h2.w---, h5;
  1826. | SCB1 | min | 29: MINh h5.w, h5.---y, h4;
  1827. | | |
  1828. 9 | SCT0 | mad | 30: MADr r1.xy, h2.xz--, const.zw--, r1.zw--;
  1829. | SCT1 | mov | 32: TXLr h2, r1, const.xxxx, TEX0;
  1830. | TEX | txl | 32: TXLr h2, r1, const.xxxx, TEX0;
  1831. | SCB0/1 | add | 34: ADDh/2 h2, h0, h2;
  1832. | | |
  1833. 10 | SCT0/1 | mov | 35: TXLr h1, g[TEX0], const.xxxx, TEX0;
  1834. | TEX | txl | 35: TXLr h1, g[TEX0], const.xxxx, TEX0;
  1835. | SCB0 | max | 37: MAXh h5.y, h5.-x--, h1.-w--;
  1836. | SCB1 | min | 38: MINh h4.w, h1, h5;
  1837. | | |
  1838. 11 | SCT0 | mad | 39: MADr r0.xy,-h4, const.xy--, r1.zw--;
  1839. | SCT1 | mov | 41: TXLr h0, r0, const.zzzz, TEX0;
  1840. | TEX | txl | 41: TXLr h0, r0, const.zzzz, TEX0;
  1841. | SCB0 | mad | 44: MADr r2.xy, h4, const.xy--, r1.zw--;
  1842. | SCB1 | add | 43: ADDh*8 h5.z, h5.--y-,-h4.--w-;
  1843. | | |
  1844. 12 | SCT0/1 | mov | 46: TXLr h3, r2, const.xxxx, TEX0;
  1845. | TEX | txl | 46: TXLr h3, r2, const.xxxx, TEX0;
  1846. | SCB0/1 | add | 48: ADDh/2 h0, h0, h3;
  1847. | | |
  1848. 13 | SCT0/1 | mad | 49: ADDh/2 h3, h0, h2;
  1849. | SCB0/1 | mul | 50: MOVh h0, h3;
  1850. | | |
  1851. 14 | SCT0 | set | 51: SLTh h3.x, h3.w---, h5.w---;
  1852. | SCT1 | set | 52: SGTh h3.w, h3, h5.---x;
  1853. | SCB0 | set | 54: SLThc0 rc, h5.z---, h5;
  1854. | SCB1 | add | 53: ADDxc0_s rc, h3.---x, h3;
  1855. | | |
  1856. 15 | SCT0/1 | mul | 55: MOVh h0(NE0.wwww), h2;
  1857. | SCB0/1 | mul | 56: MOVh h0(NE0.xxxx), h1;
  1858.  
  1859. Pass SCT TEX SCB
  1860. 1: 0% 100% 25%
  1861. 2: 0% 100% 25%
  1862. 3: 0% 100% 50%
  1863. 4: 0% 100% 50%
  1864. 5: 50% 0% 25%
  1865. 6: 0% 0% 25%
  1866. 7: 100% 0% 25%
  1867. 8: 0% 100% 50%
  1868. 9: 0% 100% 100%
  1869. 10: 0% 100% 50%
  1870. 11: 0% 100% 75%
  1871. 12: 0% 100% 100%
  1872. 13: 100% 0% 100%
  1873. 14: 50% 0% 50%
  1874. 15: 100% 0% 100%
  1875.  
  1876. MEAN: 26% 60% 56%
  1877.  
  1878. Pass SCT0 SCT1 TEX SCB0 SCB1
  1879. 1: 0% 0% 100% 100% 0%
  1880. 2: 0% 0% 100% 100% 0%
  1881. 3: 0% 0% 100% 100% 100%
  1882. 4: 0% 0% 100% 100% 100%
  1883. 5: 100% 100% 0% 100% 0%
  1884. 6: 0% 0% 0% 0% 100%
  1885. 7: 100% 100% 0% 0% 100%
  1886. 8: 0% 0% 100% 100% 100%
  1887. 9: 0% 0% 100% 100% 100%
  1888. 10: 0% 0% 100% 100% 100%
  1889. 11: 0% 0% 100% 100% 100%
  1890. 12: 0% 0% 100% 100% 100%
  1891. 13: 100% 100% 0% 100% 100%
  1892. 14: 100% 100% 0% 100% 100%
  1893. 15: 100% 100% 0% 100% 100%
  1894.  
  1895. MEAN: 33% 33% 60% 86% 80%
  1896. Fragment Performance Setup: Driver RSX Compiler, GPU RSX, Flags 0x5
  1897. Results 15 cycles, 3 r regs, 800,000,000 pixels/s
  1898. ============================================================================*/
  1899. #if (FXAA_PS3 == 1) && (FXAA_EARLY_EXIT == 1)
  1900. /*--------------------------------------------------------------------------*/
  1901. #pragma regcount 7
  1902. #pragma disablepc all
  1903. #pragma option O2
  1904. #pragma option OutColorPrec=fp16
  1905. #pragma texformat default RGBA8
  1906. /*==========================================================================*/
  1907. half4 FxaaPixelShader(
  1908. // See FXAA Quality FxaaPixelShader() source for docs on Inputs!
  1909. FxaaFloat2 pos,
  1910. FxaaFloat4 fxaaConsolePosPos,
  1911. FxaaTex tex,
  1912. FxaaTex fxaaConsole360TexExpBiasNegOne,
  1913. FxaaTex fxaaConsole360TexExpBiasNegTwo,
  1914. FxaaFloat2 fxaaQualityRcpFrame,
  1915. FxaaFloat4 fxaaConsoleRcpFrameOpt,
  1916. FxaaFloat4 fxaaConsoleRcpFrameOpt2,
  1917. FxaaFloat4 fxaaConsole360RcpFrameOpt2,
  1918. FxaaFloat fxaaQualitySubpix,
  1919. FxaaFloat fxaaQualityEdgeThreshold,
  1920. FxaaFloat fxaaQualityEdgeThresholdMin,
  1921. FxaaFloat fxaaConsoleEdgeSharpness,
  1922. FxaaFloat fxaaConsoleEdgeThreshold,
  1923. FxaaFloat fxaaConsoleEdgeThresholdMin,
  1924. FxaaFloat4 fxaaConsole360ConstDir
  1925. ) {
  1926. /*--------------------------------------------------------------------------*/
  1927. // (1)
  1928. half4 rgbyNe = h4tex2Dlod(tex, half4(fxaaConsolePosPos.zy, 0, 0));
  1929. #if (FXAA_GREEN_AS_LUMA == 0)
  1930. half lumaNe = rgbyNe.w + half(1.0/512.0);
  1931. #else
  1932. half lumaNe = rgbyNe.y + half(1.0/512.0);
  1933. #endif
  1934. /*--------------------------------------------------------------------------*/
  1935. // (2)
  1936. half4 lumaSw = h4tex2Dlod(tex, half4(fxaaConsolePosPos.xw, 0, 0));
  1937. #if (FXAA_GREEN_AS_LUMA == 0)
  1938. half lumaSwNegNe = lumaSw.w - lumaNe;
  1939. #else
  1940. half lumaSwNegNe = lumaSw.y - lumaNe;
  1941. #endif
  1942. /*--------------------------------------------------------------------------*/
  1943. // (3)
  1944. half4 lumaNw = h4tex2Dlod(tex, half4(fxaaConsolePosPos.xy, 0, 0));
  1945. #if (FXAA_GREEN_AS_LUMA == 0)
  1946. half lumaMaxNwSw = max(lumaNw.w, lumaSw.w);
  1947. half lumaMinNwSw = min(lumaNw.w, lumaSw.w);
  1948. #else
  1949. half lumaMaxNwSw = max(lumaNw.y, lumaSw.y);
  1950. half lumaMinNwSw = min(lumaNw.y, lumaSw.y);
  1951. #endif
  1952. /*--------------------------------------------------------------------------*/
  1953. // (4)
  1954. half4 lumaSe = h4tex2Dlod(tex, half4(fxaaConsolePosPos.zw, 0, 0));
  1955. #if (FXAA_GREEN_AS_LUMA == 0)
  1956. half dirZ = lumaNw.w + lumaSwNegNe;
  1957. half dirX = -lumaNw.w + lumaSwNegNe;
  1958. #else
  1959. half dirZ = lumaNw.y + lumaSwNegNe;
  1960. half dirX = -lumaNw.y + lumaSwNegNe;
  1961. #endif
  1962. /*--------------------------------------------------------------------------*/
  1963. // (5)
  1964. half3 dir;
  1965. dir.y = 0.0;
  1966. #if (FXAA_GREEN_AS_LUMA == 0)
  1967. dir.x = lumaSe.w + dirX;
  1968. dir.z = -lumaSe.w + dirZ;
  1969. half lumaMinNeSe = min(lumaNe, lumaSe.w);
  1970. #else
  1971. dir.x = lumaSe.y + dirX;
  1972. dir.z = -lumaSe.y + dirZ;
  1973. half lumaMinNeSe = min(lumaNe, lumaSe.y);
  1974. #endif
  1975. /*--------------------------------------------------------------------------*/
  1976. // (6)
  1977. half4 dir1_pos;
  1978. dir1_pos.xy = normalize(dir).xz;
  1979. half dirAbsMinTimes8 = min(abs(dir1_pos.x), abs(dir1_pos.y)) * half(FXAA_CONSOLE__PS3_EDGE_SHARPNESS);
  1980. /*--------------------------------------------------------------------------*/
  1981. // (7)
  1982. half4 dir2_pos;
  1983. dir2_pos.xy = clamp(dir1_pos.xy / dirAbsMinTimes8, half(-2.0), half(2.0));
  1984. dir1_pos.zw = pos.xy;
  1985. dir2_pos.zw = pos.xy;
  1986. #if (FXAA_GREEN_AS_LUMA == 0)
  1987. half lumaMaxNeSe = max(lumaNe, lumaSe.w);
  1988. #else
  1989. half lumaMaxNeSe = max(lumaNe, lumaSe.y);
  1990. #endif
  1991. /*--------------------------------------------------------------------------*/
  1992. // (8)
  1993. half4 temp1N;
  1994. temp1N.xy = dir1_pos.zw - dir1_pos.xy * fxaaConsoleRcpFrameOpt.zw;
  1995. temp1N = h4tex2Dlod(tex, half4(temp1N.xy, 0.0, 0.0));
  1996. half lumaMax = max(lumaMaxNwSw, lumaMaxNeSe);
  1997. half lumaMin = min(lumaMinNwSw, lumaMinNeSe);
  1998. /*--------------------------------------------------------------------------*/
  1999. // (9)
  2000. half4 rgby1;
  2001. rgby1.xy = dir1_pos.zw + dir1_pos.xy * fxaaConsoleRcpFrameOpt.zw;
  2002. rgby1 = h4tex2Dlod(tex, half4(rgby1.xy, 0.0, 0.0));
  2003. rgby1 = (temp1N + rgby1) * 0.5;
  2004. /*--------------------------------------------------------------------------*/
  2005. // (10)
  2006. half4 rgbyM = h4tex2Dlod(tex, half4(pos.xy, 0.0, 0.0));
  2007. #if (FXAA_GREEN_AS_LUMA == 0)
  2008. half lumaMaxM = max(lumaMax, rgbyM.w);
  2009. half lumaMinM = min(lumaMin, rgbyM.w);
  2010. #else
  2011. half lumaMaxM = max(lumaMax, rgbyM.y);
  2012. half lumaMinM = min(lumaMin, rgbyM.y);
  2013. #endif
  2014. /*--------------------------------------------------------------------------*/
  2015. // (11)
  2016. half4 temp2N;
  2017. temp2N.xy = dir2_pos.zw - dir2_pos.xy * fxaaConsoleRcpFrameOpt2.zw;
  2018. temp2N = h4tex2Dlod(tex, half4(temp2N.xy, 0.0, 0.0));
  2019. half4 rgby2;
  2020. rgby2.xy = dir2_pos.zw + dir2_pos.xy * fxaaConsoleRcpFrameOpt2.zw;
  2021. half lumaRangeM = (lumaMaxM - lumaMinM) / FXAA_CONSOLE__PS3_EDGE_THRESHOLD;
  2022. /*--------------------------------------------------------------------------*/
  2023. // (12)
  2024. rgby2 = h4tex2Dlod(tex, half4(rgby2.xy, 0.0, 0.0));
  2025. rgby2 = (temp2N + rgby2) * 0.5;
  2026. /*--------------------------------------------------------------------------*/
  2027. // (13)
  2028. rgby2 = (rgby2 + rgby1) * 0.5;
  2029. /*--------------------------------------------------------------------------*/
  2030. // (14)
  2031. #if (FXAA_GREEN_AS_LUMA == 0)
  2032. bool twoTapLt = rgby2.w < lumaMin;
  2033. bool twoTapGt = rgby2.w > lumaMax;
  2034. #else
  2035. bool twoTapLt = rgby2.y < lumaMin;
  2036. bool twoTapGt = rgby2.y > lumaMax;
  2037. #endif
  2038. bool earlyExit = lumaRangeM < lumaMax;
  2039. bool twoTap = twoTapLt || twoTapGt;
  2040. /*--------------------------------------------------------------------------*/
  2041. // (15)
  2042. if(twoTap) rgby2 = rgby1;
  2043. if(earlyExit) rgby2 = rgbyM;
  2044. /*--------------------------------------------------------------------------*/
  2045. return rgby2; }
  2046. /*==========================================================================*/
  2047. #endif
Add Comment
Please, Sign In to add comment
Public Pastes
We use cookies for various purposes including analytics. By continuing to use Pastebin, you agree to our use of cookies as described in the Cookies Policy.  OK, I Understand
Not a member of Pastebin yet?
Sign Up, it unlocks many cool features!