Panda3D
Loading...
Searching...
No Matches
stb_image.h
1/* stb_image - v2.15 - public domain image loader - http://nothings.org/stb_image.h
2 no warranty implied; use at your own risk
3
4 Do this:
5 #define STB_IMAGE_IMPLEMENTATION
6 before you include this file in *one* C or C++ file to create the implementation.
7
8 // i.e. it should look like this:
9 #include ...
10 #include ...
11 #include ...
12 #define STB_IMAGE_IMPLEMENTATION
13 #include "stb_image.h"
14
15 You can #define STBI_ASSERT(x) before the #include to avoid using assert.h.
16 And #define STBI_MALLOC, STBI_REALLOC, and STBI_FREE to avoid using malloc,realloc,free
17
18
19 QUICK NOTES:
20 Primarily of interest to game developers and other people who can
21 avoid problematic images and only need the trivial interface
22
23 JPEG baseline & progressive (12 bpc/arithmetic not supported, same as stock IJG lib)
24 PNG 1/2/4/8/16-bit-per-channel
25
26 TGA (not sure what subset, if a subset)
27 BMP non-1bpp, non-RLE
28 PSD (composited view only, no extra channels, 8/16 bit-per-channel)
29
30 GIF (*comp always reports as 4-channel)
31 HDR (radiance rgbE format)
32 PIC (Softimage PIC)
33 PNM (PPM and PGM binary only)
34
35 Animated GIF still needs a proper API, but here's one way to do it:
36 http://gist.github.com/urraka/685d9a6340b26b830d49
37
38 - decode from memory or through FILE (define STBI_NO_STDIO to remove code)
39 - decode from arbitrary I/O callbacks
40 - SIMD acceleration on x86/x64 (SSE2) and ARM (NEON)
41
42 Full documentation under "DOCUMENTATION" below.
43
44
45LICENSE
46
47 See end of file for license information.
48
49RECENT REVISION HISTORY:
50
51 2.15 (2017-03-18) fix png-1,2,4; all Imagenet JPGs; no runtime SSE detection on GCC
52 2.14 (2017-03-03) remove deprecated STBI_JPEG_OLD; fixes for Imagenet JPGs
53 2.13 (2016-12-04) experimental 16-bit API, only for PNG so far; fixes
54 2.12 (2016-04-02) fix typo in 2.11 PSD fix that caused crashes
55 2.11 (2016-04-02) 16-bit PNGS; enable SSE2 in non-gcc x64
56 RGB-format JPEG; remove white matting in PSD;
57 allocate large structures on the stack;
58 correct channel count for PNG & BMP
59 2.10 (2016-01-22) avoid warning introduced in 2.09
60 2.09 (2016-01-16) 16-bit TGA; comments in PNM files; STBI_REALLOC_SIZED
61 2.08 (2015-09-13) fix to 2.07 cleanup, reading RGB PSD as RGBA
62 2.07 (2015-09-13) partial animated GIF support
63 limited 16-bit PSD support
64 minor bugs, code cleanup, and compiler warnings
65
66 See end of file for full revision history.
67
68
69 ============================ Contributors =========================
70
71 Image formats Extensions, features
72 Sean Barrett (jpeg, png, bmp) Jetro Lauha (stbi_info)
73 Nicolas Schulz (hdr, psd) Martin "SpartanJ" Golini (stbi_info)
74 Jonathan Dummer (tga) James "moose2000" Brown (iPhone PNG)
75 Jean-Marc Lienher (gif) Ben "Disch" Wenger (io callbacks)
76 Tom Seddon (pic) Omar Cornut (1/2/4-bit PNG)
77 Thatcher Ulrich (psd) Nicolas Guillemot (vertical flip)
78 Ken Miller (pgm, ppm) Richard Mitton (16-bit PSD)
79 github:urraka (animated gif) Junggon Kim (PNM comments)
80 Daniel Gibson (16-bit TGA)
81 socks-the-fox (16-bit PNG)
82 Jeremy Sawicki (handle all ImageNet JPGs)
83 Optimizations & bugfixes
84 Fabian "ryg" Giesen
85 Arseny Kapoulkine
86
87 Bug & warning fixes
88 Marc LeBlanc David Woo Guillaume George Martins Mozeiko
89 Christpher Lloyd Jerry Jansson Joseph Thomson Phil Jordan
90 Dave Moore Roy Eltham Hayaki Saito Nathan Reed
91 Won Chun Luke Graham Johan Duparc Nick Verigakis
92 the Horde3D community Thomas Ruf Ronny Chevalier Baldur Karlsson
93 Janez Zemva John Bartholomew Michal Cichon github:rlyeh
94 Jonathan Blow Ken Hamada Tero Hanninen github:romigrou
95 Laurent Gomila Cort Stratton Sergio Gonzalez github:svdijk
96 Aruelien Pocheville Thibault Reuille Cass Everitt github:snagar
97 Ryamond Barbiero Paul Du Bois Engin Manap github:Zelex
98 Michaelangel007@github Philipp Wiesemann Dale Weiler github:grim210
99 Oriol Ferrer Mesia Josh Tobin Matthew Gregan github:sammyhw
100 Blazej Dariusz Roszkowski Gregory Mullen github:phprus
101
102*/
103
104#ifndef STBI_INCLUDE_STB_IMAGE_H
105#define STBI_INCLUDE_STB_IMAGE_H
106
107// DOCUMENTATION
108//
109// Limitations:
110// - no 16-bit-per-channel PNG
111// - no 12-bit-per-channel JPEG
112// - no JPEGs with arithmetic coding
113// - no 1-bit BMP
114// - GIF always returns *comp=4
115//
116// Basic usage (see HDR discussion below for HDR usage):
117// int x,y,n;
118// unsigned char *data = stbi_load(filename, &x, &y, &n, 0);
119// // ... process data if not NULL ...
120// // ... x = width, y = height, n = # 8-bit components per pixel ...
121// // ... replace '0' with '1'..'4' to force that many components per pixel
122// // ... but 'n' will always be the number that it would have been if you said 0
123// stbi_image_free(data)
124//
125// Standard parameters:
126// int *x -- outputs image width in pixels
127// int *y -- outputs image height in pixels
128// int *channels_in_file -- outputs # of image components in image file
129// int desired_channels -- if non-zero, # of image components requested in result
130//
131// The return value from an image loader is an 'unsigned char *' which points
132// to the pixel data, or NULL on an allocation failure or if the image is
133// corrupt or invalid. The pixel data consists of *y scanlines of *x pixels,
134// with each pixel consisting of N interleaved 8-bit components; the first
135// pixel pointed to is top-left-most in the image. There is no padding between
136// image scanlines or between pixels, regardless of format. The number of
137// components N is 'req_comp' if req_comp is non-zero, or *comp otherwise.
138// If req_comp is non-zero, *comp has the number of components that _would_
139// have been output otherwise. E.g. if you set req_comp to 4, you will always
140// get RGBA output, but you can check *comp to see if it's trivially opaque
141// because e.g. there were only 3 channels in the source image.
142//
143// An output image with N components has the following components interleaved
144// in this order in each pixel:
145//
146// N=#comp components
147// 1 grey
148// 2 grey, alpha
149// 3 red, green, blue
150// 4 red, green, blue, alpha
151//
152// If image loading fails for any reason, the return value will be NULL,
153// and *x, *y, *comp will be unchanged. The function stbi_failure_reason()
154// can be queried for an extremely brief, end-user unfriendly explanation
155// of why the load failed. Define STBI_NO_FAILURE_STRINGS to avoid
156// compiling these strings at all, and STBI_FAILURE_USERMSG to get slightly
157// more user-friendly ones.
158//
159// Paletted PNG, BMP, GIF, and PIC images are automatically depalettized.
160//
161// ===========================================================================
162//
163// Philosophy
164//
165// stb libraries are designed with the following priorities:
166//
167// 1. easy to use
168// 2. easy to maintain
169// 3. good performance
170//
171// Sometimes I let "good performance" creep up in priority over "easy to maintain",
172// and for best performance I may provide less-easy-to-use APIs that give higher
173// performance, in addition to the easy to use ones. Nevertheless, it's important
174// to keep in mind that from the standpoint of you, a client of this library,
175// all you care about is #1 and #3, and stb libraries DO NOT emphasize #3 above all.
176//
177// Some secondary priorities arise directly from the first two, some of which
178// make more explicit reasons why performance can't be emphasized.
179//
180// - Portable ("ease of use")
181// - Small source code footprint ("easy to maintain")
182// - No dependencies ("ease of use")
183//
184// ===========================================================================
185//
186// I/O callbacks
187//
188// I/O callbacks allow you to read from arbitrary sources, like packaged
189// files or some other source. Data read from callbacks are processed
190// through a small internal buffer (currently 128 bytes) to try to reduce
191// overhead.
192//
193// The three functions you must define are "read" (reads some bytes of data),
194// "skip" (skips some bytes of data), "eof" (reports if the stream is at the end).
195//
196// ===========================================================================
197//
198// SIMD support
199//
200// The JPEG decoder will try to automatically use SIMD kernels on x86 when
201// supported by the compiler. For ARM Neon support, you must explicitly
202// request it.
203//
204// (The old do-it-yourself SIMD API is no longer supported in the current
205// code.)
206//
207// On x86, SSE2 will automatically be used when available based on a run-time
208// test; if not, the generic C versions are used as a fall-back. On ARM targets,
209// the typical path is to have separate builds for NEON and non-NEON devices
210// (at least this is true for iOS and Android). Therefore, the NEON support is
211// toggled by a build flag: define STBI_NEON to get NEON loops.
212//
213// If for some reason you do not want to use any of SIMD code, or if
214// you have issues compiling it, you can disable it entirely by
215// defining STBI_NO_SIMD.
216//
217// ===========================================================================
218//
219// HDR image support (disable by defining STBI_NO_HDR)
220//
221// stb_image now supports loading HDR images in general, and currently
222// the Radiance .HDR file format, although the support is provided
223// generically. You can still load any file through the existing interface;
224// if you attempt to load an HDR file, it will be automatically remapped to
225// LDR, assuming gamma 2.2 and an arbitrary scale factor defaulting to 1;
226// both of these constants can be reconfigured through this interface:
227//
228// stbi_hdr_to_ldr_gamma(2.2f);
229// stbi_hdr_to_ldr_scale(1.0f);
230//
231// (note, do not use _inverse_ constants; stbi_image will invert them
232// appropriately).
233//
234// Additionally, there is a new, parallel interface for loading files as
235// (linear) floats to preserve the full dynamic range:
236//
237// float *data = stbi_loadf(filename, &x, &y, &n, 0);
238//
239// If you load LDR images through this interface, those images will
240// be promoted to floating point values, run through the inverse of
241// constants corresponding to the above:
242//
243// stbi_ldr_to_hdr_scale(1.0f);
244// stbi_ldr_to_hdr_gamma(2.2f);
245//
246// Finally, given a filename (or an open file or memory block--see header
247// file for details) containing image data, you can query for the "most
248// appropriate" interface to use (that is, whether the image is HDR or
249// not), using:
250//
251// stbi_is_hdr(char *filename);
252//
253// ===========================================================================
254//
255// iPhone PNG support:
256//
257// By default we convert iphone-formatted PNGs back to RGB, even though
258// they are internally encoded differently. You can disable this conversion
259// by by calling stbi_convert_iphone_png_to_rgb(0), in which case
260// you will always just get the native iphone "format" through (which
261// is BGR stored in RGB).
262//
263// Call stbi_set_unpremultiply_on_load(1) as well to force a divide per
264// pixel to remove any premultiplied alpha *only* if the image file explicitly
265// says there's premultiplied data (currently only happens in iPhone images,
266// and only if iPhone convert-to-rgb processing is on).
267//
268// ===========================================================================
269//
270// ADDITIONAL CONFIGURATION
271//
272// - You can suppress implementation of any of the decoders to reduce
273// your code footprint by #defining one or more of the following
274// symbols before creating the implementation.
275//
276// STBI_NO_JPEG
277// STBI_NO_PNG
278// STBI_NO_BMP
279// STBI_NO_PSD
280// STBI_NO_TGA
281// STBI_NO_GIF
282// STBI_NO_HDR
283// STBI_NO_PIC
284// STBI_NO_PNM (.ppm and .pgm)
285//
286// - You can request *only* certain decoders and suppress all other ones
287// (this will be more forward-compatible, as addition of new decoders
288// doesn't require you to disable them explicitly):
289//
290// STBI_ONLY_JPEG
291// STBI_ONLY_PNG
292// STBI_ONLY_BMP
293// STBI_ONLY_PSD
294// STBI_ONLY_TGA
295// STBI_ONLY_GIF
296// STBI_ONLY_HDR
297// STBI_ONLY_PIC
298// STBI_ONLY_PNM (.ppm and .pgm)
299//
300// - If you use STBI_NO_PNG (or _ONLY_ without PNG), and you still
301// want the zlib decoder to be available, #define STBI_SUPPORT_ZLIB
302//
303
304
305#ifndef STBI_NO_STDIO
306#include <stdio.h>
307#endif // STBI_NO_STDIO
308
309#define STBI_VERSION 1
310
311enum
312{
313 STBI_default = 0, // only used for req_comp
314
315 STBI_grey = 1,
316 STBI_grey_alpha = 2,
317 STBI_rgb = 3,
318 STBI_rgb_alpha = 4
319};
320
321typedef unsigned char stbi_uc;
322typedef unsigned short stbi_us;
323
324#ifdef __cplusplus
325extern "C" {
326#endif
327
328#ifdef STB_IMAGE_STATIC
329#define STBIDEF static
330#else
331#define STBIDEF extern
332#endif
333
334//////////////////////////////////////////////////////////////////////////////
335//
336// PRIMARY API - works on images of any type
337//
338
339//
340// load image by filename, open file, or memory buffer
341//
342
343typedef struct
344{
345 int (*read) (void *user,char *data,int size); // fill 'data' with 'size' bytes. return number of bytes actually read
346 void (*skip) (void *user,int n); // skip the next 'n' bytes, or 'unget' the last -n bytes if negative
347 int (*eof) (void *user); // returns nonzero if we are at end of file/data
349
350////////////////////////////////////
351//
352// 8-bits-per-channel interface
353//
354
355STBIDEF stbi_uc *stbi_load (char const *filename, int *x, int *y, int *channels_in_file, int desired_channels);
356STBIDEF stbi_uc *stbi_load_from_memory (stbi_uc const *buffer, int len , int *x, int *y, int *channels_in_file, int desired_channels);
357STBIDEF stbi_uc *stbi_load_from_callbacks(stbi_io_callbacks const *clbk , void *user, int *x, int *y, int *channels_in_file, int desired_channels);
358
359#ifndef STBI_NO_STDIO
360STBIDEF stbi_uc *stbi_load_from_file (FILE *f, int *x, int *y, int *channels_in_file, int desired_channels);
361// for stbi_load_from_file, file pointer is left pointing immediately after image
362#endif
363
364////////////////////////////////////
365//
366// 16-bits-per-channel interface
367//
368
369STBIDEF stbi_us *stbi_load_16(char const *filename, int *x, int *y, int *channels_in_file, int desired_channels);
370#ifndef STBI_NO_STDIO
371STBIDEF stbi_us *stbi_load_from_file_16(FILE *f, int *x, int *y, int *channels_in_file, int desired_channels);
372#endif
373// @TODO the other variants
374
375////////////////////////////////////
376//
377// float-per-channel interface
378//
379#ifndef STBI_NO_LINEAR
380 STBIDEF float *stbi_loadf (char const *filename, int *x, int *y, int *channels_in_file, int desired_channels);
381 STBIDEF float *stbi_loadf_from_memory (stbi_uc const *buffer, int len, int *x, int *y, int *channels_in_file, int desired_channels);
382 STBIDEF float *stbi_loadf_from_callbacks (stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *channels_in_file, int desired_channels);
383
384 #ifndef STBI_NO_STDIO
385 STBIDEF float *stbi_loadf_from_file (FILE *f, int *x, int *y, int *channels_in_file, int desired_channels);
386 #endif
387#endif
388
389#ifndef STBI_NO_HDR
390 STBIDEF void stbi_hdr_to_ldr_gamma(float gamma);
391 STBIDEF void stbi_hdr_to_ldr_scale(float scale);
392#endif // STBI_NO_HDR
393
394#ifndef STBI_NO_LINEAR
395 STBIDEF void stbi_ldr_to_hdr_gamma(float gamma);
396 STBIDEF void stbi_ldr_to_hdr_scale(float scale);
397#endif // STBI_NO_LINEAR
398
399// stbi_is_hdr is always defined, but always returns false if STBI_NO_HDR
400STBIDEF int stbi_is_hdr_from_callbacks(stbi_io_callbacks const *clbk, void *user);
401STBIDEF int stbi_is_hdr_from_memory(stbi_uc const *buffer, int len);
402#ifndef STBI_NO_STDIO
403STBIDEF int stbi_is_hdr (char const *filename);
404STBIDEF int stbi_is_hdr_from_file(FILE *f);
405#endif // STBI_NO_STDIO
406
407
408// get a VERY brief reason for failure
409// NOT THREADSAFE
410STBIDEF const char *stbi_failure_reason (void);
411
412// free the loaded image -- this is just free()
413STBIDEF void stbi_image_free (void *retval_from_stbi_load);
414
415// get image dimensions & components without fully decoding
416STBIDEF int stbi_info_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp);
417STBIDEF int stbi_info_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp);
418
419#ifndef STBI_NO_STDIO
420STBIDEF int stbi_info (char const *filename, int *x, int *y, int *comp);
421STBIDEF int stbi_info_from_file (FILE *f, int *x, int *y, int *comp);
422
423#endif
424
425
426
427// for image formats that explicitly notate that they have premultiplied alpha,
428// we just return the colors as stored in the file. set this flag to force
429// unpremultiplication. results are undefined if the unpremultiply overflow.
430STBIDEF void stbi_set_unpremultiply_on_load(int flag_true_if_should_unpremultiply);
431
432// indicate whether we should process iphone images back to canonical format,
433// or just pass them through "as-is"
434STBIDEF void stbi_convert_iphone_png_to_rgb(int flag_true_if_should_convert);
435
436// flip the image vertically, so the first pixel in the output array is the bottom left
437STBIDEF void stbi_set_flip_vertically_on_load(int flag_true_if_should_flip);
438
439// ZLIB client - used by PNG, available for other purposes
440
441STBIDEF char *stbi_zlib_decode_malloc_guesssize(const char *buffer, int len, int initial_size, int *outlen);
442STBIDEF char *stbi_zlib_decode_malloc_guesssize_headerflag(const char *buffer, int len, int initial_size, int *outlen, int parse_header);
443STBIDEF char *stbi_zlib_decode_malloc(const char *buffer, int len, int *outlen);
444STBIDEF int stbi_zlib_decode_buffer(char *obuffer, int olen, const char *ibuffer, int ilen);
445
446STBIDEF char *stbi_zlib_decode_noheader_malloc(const char *buffer, int len, int *outlen);
447STBIDEF int stbi_zlib_decode_noheader_buffer(char *obuffer, int olen, const char *ibuffer, int ilen);
448
449
450#ifdef __cplusplus
451}
452#endif
453
454//
455//
456//// end header file /////////////////////////////////////////////////////
457#endif // STBI_INCLUDE_STB_IMAGE_H
458
459#ifdef STB_IMAGE_IMPLEMENTATION
460
461#if defined(STBI_ONLY_JPEG) || defined(STBI_ONLY_PNG) || defined(STBI_ONLY_BMP) \
462 || defined(STBI_ONLY_TGA) || defined(STBI_ONLY_GIF) || defined(STBI_ONLY_PSD) \
463 || defined(STBI_ONLY_HDR) || defined(STBI_ONLY_PIC) || defined(STBI_ONLY_PNM) \
464 || defined(STBI_ONLY_ZLIB)
465 #ifndef STBI_ONLY_JPEG
466 #define STBI_NO_JPEG
467 #endif
468 #ifndef STBI_ONLY_PNG
469 #define STBI_NO_PNG
470 #endif
471 #ifndef STBI_ONLY_BMP
472 #define STBI_NO_BMP
473 #endif
474 #ifndef STBI_ONLY_PSD
475 #define STBI_NO_PSD
476 #endif
477 #ifndef STBI_ONLY_TGA
478 #define STBI_NO_TGA
479 #endif
480 #ifndef STBI_ONLY_GIF
481 #define STBI_NO_GIF
482 #endif
483 #ifndef STBI_ONLY_HDR
484 #define STBI_NO_HDR
485 #endif
486 #ifndef STBI_ONLY_PIC
487 #define STBI_NO_PIC
488 #endif
489 #ifndef STBI_ONLY_PNM
490 #define STBI_NO_PNM
491 #endif
492#endif
493
494#if defined(STBI_NO_PNG) && !defined(STBI_SUPPORT_ZLIB) && !defined(STBI_NO_ZLIB)
495#define STBI_NO_ZLIB
496#endif
497
498
499#include <stdarg.h>
500#include <stddef.h> // ptrdiff_t on osx
501#include <stdlib.h>
502#include <string.h>
503#include <limits.h>
504
505#if !defined(STBI_NO_LINEAR) || !defined(STBI_NO_HDR)
506#include <math.h> // ldexp
507#endif
508
509#ifndef STBI_NO_STDIO
510#include <stdio.h>
511#endif
512
513#ifndef STBI_ASSERT
514#include <assert.h>
515#define STBI_ASSERT(x) assert(x)
516#endif
517
518
519#ifndef _MSC_VER
520 #ifdef __cplusplus
521 #define stbi_inline inline
522 #else
523 #define stbi_inline
524 #endif
525#else
526 #define stbi_inline __forceinline
527#endif
528
529
530#ifdef _MSC_VER
531typedef unsigned short stbi__uint16;
532typedef signed short stbi__int16;
533typedef unsigned int stbi__uint32;
534typedef signed int stbi__int32;
535#else
536#include <stdint.h>
537typedef uint16_t stbi__uint16;
538typedef int16_t stbi__int16;
539typedef uint32_t stbi__uint32;
540typedef int32_t stbi__int32;
541#endif
542
543// should produce compiler error if size is wrong
544typedef unsigned char validate_uint32[sizeof(stbi__uint32)==4 ? 1 : -1];
545
546#ifdef _MSC_VER
547#define STBI_NOTUSED(v) (void)(v)
548#else
549#define STBI_NOTUSED(v) (void)sizeof(v)
550#endif
551
552#ifdef _MSC_VER
553#define STBI_HAS_LROTL
554#endif
555
556#ifdef STBI_HAS_LROTL
557 #define stbi_lrot(x,y) _lrotl(x,y)
558#else
559 #define stbi_lrot(x,y) (((x) << (y)) | ((x) >> (32 - (y))))
560#endif
561
562#if defined(STBI_MALLOC) && defined(STBI_FREE) && (defined(STBI_REALLOC) || defined(STBI_REALLOC_SIZED))
563// ok
564#elif !defined(STBI_MALLOC) && !defined(STBI_FREE) && !defined(STBI_REALLOC) && !defined(STBI_REALLOC_SIZED)
565// ok
566#else
567#error "Must define all or none of STBI_MALLOC, STBI_FREE, and STBI_REALLOC (or STBI_REALLOC_SIZED)."
568#endif
569
570#ifndef STBI_MALLOC
571#define STBI_MALLOC(sz) malloc(sz)
572#define STBI_REALLOC(p,newsz) realloc(p,newsz)
573#define STBI_FREE(p) free(p)
574#endif
575
576#ifndef STBI_REALLOC_SIZED
577#define STBI_REALLOC_SIZED(p,oldsz,newsz) STBI_REALLOC(p,newsz)
578#endif
579
580// x86/x64 detection
581#if defined(__x86_64__) || defined(_M_X64)
582#define STBI__X64_TARGET
583#elif defined(__i386) || defined(_M_IX86)
584#define STBI__X86_TARGET
585#endif
586
587#if defined(__GNUC__) && defined(STBI__X86_TARGET) && !defined(__SSE2__) && !defined(STBI_NO_SIMD)
588// gcc doesn't support sse2 intrinsics unless you compile with -msse2,
589// which in turn means it gets to use SSE2 everywhere. This is unfortunate,
590// but previous attempts to provide the SSE2 functions with runtime
591// detection caused numerous issues. The way architecture extensions are
592// exposed in GCC/Clang is, sadly, not really suited for one-file libs.
593// New behavior: if compiled with -msse2, we use SSE2 without any
594// detection; if not, we don't use it at all.
595#define STBI_NO_SIMD
596#endif
597
598#if defined(__MINGW32__) && defined(STBI__X86_TARGET) && !defined(STBI_MINGW_ENABLE_SSE2) && !defined(STBI_NO_SIMD)
599// Note that __MINGW32__ doesn't actually mean 32-bit, so we have to avoid STBI__X64_TARGET
600//
601// 32-bit MinGW wants ESP to be 16-byte aligned, but this is not in the
602// Windows ABI and VC++ as well as Windows DLLs don't maintain that invariant.
603// As a result, enabling SSE2 on 32-bit MinGW is dangerous when not
604// simultaneously enabling "-mstackrealign".
605//
606// See https://github.com/nothings/stb/issues/81 for more information.
607//
608// So default to no SSE2 on 32-bit MinGW. If you've read this far and added
609// -mstackrealign to your build settings, feel free to #define STBI_MINGW_ENABLE_SSE2.
610#define STBI_NO_SIMD
611#endif
612
613#if !defined(STBI_NO_SIMD) && (defined(STBI__X86_TARGET) || defined(STBI__X64_TARGET))
614#define STBI_SSE2
615#include <emmintrin.h>
616
617#ifdef _MSC_VER
618
619#if _MSC_VER >= 1400 // not VC6
620#include <intrin.h> // __cpuid
621static int stbi__cpuid3(void)
622{
623 int info[4];
624 __cpuid(info,1);
625 return info[3];
626}
627#else
628static int stbi__cpuid3(void)
629{
630 int res;
631 __asm {
632 mov eax,1
633 cpuid
634 mov res,edx
635 }
636 return res;
637}
638#endif
639
640#define STBI_SIMD_ALIGN(type, name) __declspec(align(16)) type name
641
642static int stbi__sse2_available()
643{
644 int info3 = stbi__cpuid3();
645 return ((info3 >> 26) & 1) != 0;
646}
647#else // assume GCC-style if not VC++
648#define STBI_SIMD_ALIGN(type, name) type name __attribute__((aligned(16)))
649
650static int stbi__sse2_available()
651{
652 // If we're even attempting to compile this on GCC/Clang, that means
653 // -msse2 is on, which means the compiler is allowed to use SSE2
654 // instructions at will, and so are we.
655 return 1;
656}
657#endif
658#endif
659
660// ARM NEON
661#if defined(STBI_NO_SIMD) && defined(STBI_NEON)
662#undef STBI_NEON
663#endif
664
665#ifdef STBI_NEON
666#include <arm_neon.h>
667// assume GCC or Clang on ARM targets
668#define STBI_SIMD_ALIGN(type, name) type name __attribute__((aligned(16)))
669#endif
670
671#ifndef STBI_SIMD_ALIGN
672#define STBI_SIMD_ALIGN(type, name) type name
673#endif
674
675///////////////////////////////////////////////
676//
677// stbi__context struct and start_xxx functions
678
679// stbi__context structure is our basic context used by all images, so it
680// contains all the IO context, plus some basic image information
681typedef struct
682{
683 stbi__uint32 img_x, img_y;
684 int img_n, img_out_n;
685
687 void *io_user_data;
688
689 int read_from_callbacks;
690 int buflen;
691 stbi_uc buffer_start[128];
692
693 stbi_uc *img_buffer, *img_buffer_end;
694 stbi_uc *img_buffer_original, *img_buffer_original_end;
695} stbi__context;
696
697
698static void stbi__refill_buffer(stbi__context *s);
699
700// initialize a memory-decode context
701static void stbi__start_mem(stbi__context *s, stbi_uc const *buffer, int len)
702{
703 s->io.read = NULL;
704 s->read_from_callbacks = 0;
705 s->img_buffer = s->img_buffer_original = (stbi_uc *) buffer;
706 s->img_buffer_end = s->img_buffer_original_end = (stbi_uc *) buffer+len;
707}
708
709// initialize a callback-based context
710static void stbi__start_callbacks(stbi__context *s, stbi_io_callbacks *c, void *user)
711{
712 s->io = *c;
713 s->io_user_data = user;
714 s->buflen = sizeof(s->buffer_start);
715 s->read_from_callbacks = 1;
716 s->img_buffer_original = s->buffer_start;
717 stbi__refill_buffer(s);
718 s->img_buffer_original_end = s->img_buffer_end;
719}
720
721#ifndef STBI_NO_STDIO
722
723static int stbi__stdio_read(void *user, char *data, int size)
724{
725 return (int) fread(data,1,size,(FILE*) user);
726}
727
728static void stbi__stdio_skip(void *user, int n)
729{
730 fseek((FILE*) user, n, SEEK_CUR);
731}
732
733static int stbi__stdio_eof(void *user)
734{
735 return feof((FILE*) user);
736}
737
738static stbi_io_callbacks stbi__stdio_callbacks =
739{
740 stbi__stdio_read,
741 stbi__stdio_skip,
742 stbi__stdio_eof,
743};
744
745static void stbi__start_file(stbi__context *s, FILE *f)
746{
747 stbi__start_callbacks(s, &stbi__stdio_callbacks, (void *) f);
748}
749
750//static void stop_file(stbi__context *s) { }
751
752#endif // !STBI_NO_STDIO
753
754static void stbi__rewind(stbi__context *s)
755{
756 // conceptually rewind SHOULD rewind to the beginning of the stream,
757 // but we just rewind to the beginning of the initial buffer, because
758 // we only use it after doing 'test', which only ever looks at at most 92 bytes
759 s->img_buffer = s->img_buffer_original;
760 s->img_buffer_end = s->img_buffer_original_end;
761}
762
763enum
764{
765 STBI_ORDER_RGB,
766 STBI_ORDER_BGR
767};
768
769typedef struct
770{
771 int bits_per_channel;
772 int num_channels;
773 int channel_order;
774} stbi__result_info;
775
776#ifndef STBI_NO_JPEG
777static int stbi__jpeg_test(stbi__context *s);
778static void *stbi__jpeg_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
779static int stbi__jpeg_info(stbi__context *s, int *x, int *y, int *comp);
780#endif
781
782#ifndef STBI_NO_PNG
783static int stbi__png_test(stbi__context *s);
784static void *stbi__png_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
785static int stbi__png_info(stbi__context *s, int *x, int *y, int *comp);
786#endif
787
788#ifndef STBI_NO_BMP
789static int stbi__bmp_test(stbi__context *s);
790static void *stbi__bmp_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
791static int stbi__bmp_info(stbi__context *s, int *x, int *y, int *comp);
792#endif
793
794#ifndef STBI_NO_TGA
795static int stbi__tga_test(stbi__context *s);
796static void *stbi__tga_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
797static int stbi__tga_info(stbi__context *s, int *x, int *y, int *comp);
798#endif
799
800#ifndef STBI_NO_PSD
801static int stbi__psd_test(stbi__context *s);
802static void *stbi__psd_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri, int bpc);
803static int stbi__psd_info(stbi__context *s, int *x, int *y, int *comp);
804#endif
805
806#ifndef STBI_NO_HDR
807static int stbi__hdr_test(stbi__context *s);
808static float *stbi__hdr_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
809static int stbi__hdr_info(stbi__context *s, int *x, int *y, int *comp);
810#endif
811
812#ifndef STBI_NO_PIC
813static int stbi__pic_test(stbi__context *s);
814static void *stbi__pic_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
815static int stbi__pic_info(stbi__context *s, int *x, int *y, int *comp);
816#endif
817
818#ifndef STBI_NO_GIF
819static int stbi__gif_test(stbi__context *s);
820static void *stbi__gif_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
821static int stbi__gif_info(stbi__context *s, int *x, int *y, int *comp);
822#endif
823
824#ifndef STBI_NO_PNM
825static int stbi__pnm_test(stbi__context *s);
826static void *stbi__pnm_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
827static int stbi__pnm_info(stbi__context *s, int *x, int *y, int *comp);
828#endif
829
830// this is not threadsafe
831static const char *stbi__g_failure_reason;
832
833STBIDEF const char *stbi_failure_reason(void)
834{
835 return stbi__g_failure_reason;
836}
837
838static int stbi__err(const char *str)
839{
840 stbi__g_failure_reason = str;
841 return 0;
842}
843
844static void *stbi__malloc(size_t size)
845{
846 return STBI_MALLOC(size);
847}
848
849// stb_image uses ints pervasively, including for offset calculations.
850// therefore the largest decoded image size we can support with the
851// current code, even on 64-bit targets, is INT_MAX. this is not a
852// significant limitation for the intended use case.
853//
854// we do, however, need to make sure our size calculations don't
855// overflow. hence a few helper functions for size calculations that
856// multiply integers together, making sure that they're non-negative
857// and no overflow occurs.
858
859// return 1 if the sum is valid, 0 on overflow.
860// negative terms are considered invalid.
861static int stbi__addsizes_valid(int a, int b)
862{
863 if (b < 0) return 0;
864 // now 0 <= b <= INT_MAX, hence also
865 // 0 <= INT_MAX - b <= INTMAX.
866 // And "a + b <= INT_MAX" (which might overflow) is the
867 // same as a <= INT_MAX - b (no overflow)
868 return a <= INT_MAX - b;
869}
870
871// returns 1 if the product is valid, 0 on overflow.
872// negative factors are considered invalid.
873static int stbi__mul2sizes_valid(int a, int b)
874{
875 if (a < 0 || b < 0) return 0;
876 if (b == 0) return 1; // mul-by-0 is always safe
877 // portable way to check for no overflows in a*b
878 return a <= INT_MAX/b;
879}
880
881// returns 1 if "a*b + add" has no negative terms/factors and doesn't overflow
882static int stbi__mad2sizes_valid(int a, int b, int add)
883{
884 return stbi__mul2sizes_valid(a, b) && stbi__addsizes_valid(a*b, add);
885}
886
887// returns 1 if "a*b*c + add" has no negative terms/factors and doesn't overflow
888static int stbi__mad3sizes_valid(int a, int b, int c, int add)
889{
890 return stbi__mul2sizes_valid(a, b) && stbi__mul2sizes_valid(a*b, c) &&
891 stbi__addsizes_valid(a*b*c, add);
892}
893
894// returns 1 if "a*b*c*d + add" has no negative terms/factors and doesn't overflow
895static int stbi__mad4sizes_valid(int a, int b, int c, int d, int add)
896{
897 return stbi__mul2sizes_valid(a, b) && stbi__mul2sizes_valid(a*b, c) &&
898 stbi__mul2sizes_valid(a*b*c, d) && stbi__addsizes_valid(a*b*c*d, add);
899}
900
901// mallocs with size overflow checking
902static void *stbi__malloc_mad2(int a, int b, int add)
903{
904 if (!stbi__mad2sizes_valid(a, b, add)) return NULL;
905 return stbi__malloc(a*b + add);
906}
907
908static void *stbi__malloc_mad3(int a, int b, int c, int add)
909{
910 if (!stbi__mad3sizes_valid(a, b, c, add)) return NULL;
911 return stbi__malloc(a*b*c + add);
912}
913
914static void *stbi__malloc_mad4(int a, int b, int c, int d, int add)
915{
916 if (!stbi__mad4sizes_valid(a, b, c, d, add)) return NULL;
917 return stbi__malloc(a*b*c*d + add);
918}
919
920// stbi__err - error
921// stbi__errpf - error returning pointer to float
922// stbi__errpuc - error returning pointer to unsigned char
923
924#ifdef STBI_NO_FAILURE_STRINGS
925 #define stbi__err(x,y) 0
926#elif defined(STBI_FAILURE_USERMSG)
927 #define stbi__err(x,y) stbi__err(y)
928#else
929 #define stbi__err(x,y) stbi__err(x)
930#endif
931
932#define stbi__errpf(x,y) ((float *)(size_t) (stbi__err(x,y)?NULL:NULL))
933#define stbi__errpuc(x,y) ((unsigned char *)(size_t) (stbi__err(x,y)?NULL:NULL))
934
935STBIDEF void stbi_image_free(void *retval_from_stbi_load)
936{
937 STBI_FREE(retval_from_stbi_load);
938}
939
940#ifndef STBI_NO_LINEAR
941static float *stbi__ldr_to_hdr(stbi_uc *data, int x, int y, int comp);
942#endif
943
944#ifndef STBI_NO_HDR
945static stbi_uc *stbi__hdr_to_ldr(float *data, int x, int y, int comp);
946#endif
947
948static int stbi__vertically_flip_on_load = 0;
949
950STBIDEF void stbi_set_flip_vertically_on_load(int flag_true_if_should_flip)
951{
952 stbi__vertically_flip_on_load = flag_true_if_should_flip;
953}
954
955static void *stbi__load_main(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri, int bpc)
956{
957 memset(ri, 0, sizeof(*ri)); // make sure it's initialized if we add new fields
958 ri->bits_per_channel = 8; // default is 8 so most paths don't have to be changed
959 ri->channel_order = STBI_ORDER_RGB; // all current input & output are this, but this is here so we can add BGR order
960 ri->num_channels = 0;
961
962 #ifndef STBI_NO_JPEG
963 if (stbi__jpeg_test(s)) return stbi__jpeg_load(s,x,y,comp,req_comp, ri);
964 #endif
965 #ifndef STBI_NO_PNG
966 if (stbi__png_test(s)) return stbi__png_load(s,x,y,comp,req_comp, ri);
967 #endif
968 #ifndef STBI_NO_BMP
969 if (stbi__bmp_test(s)) return stbi__bmp_load(s,x,y,comp,req_comp, ri);
970 #endif
971 #ifndef STBI_NO_GIF
972 if (stbi__gif_test(s)) return stbi__gif_load(s,x,y,comp,req_comp, ri);
973 #endif
974 #ifndef STBI_NO_PSD
975 if (stbi__psd_test(s)) return stbi__psd_load(s,x,y,comp,req_comp, ri, bpc);
976 #endif
977 #ifndef STBI_NO_PIC
978 if (stbi__pic_test(s)) return stbi__pic_load(s,x,y,comp,req_comp, ri);
979 #endif
980 #ifndef STBI_NO_PNM
981 if (stbi__pnm_test(s)) return stbi__pnm_load(s,x,y,comp,req_comp, ri);
982 #endif
983
984 #ifndef STBI_NO_HDR
985 if (stbi__hdr_test(s)) {
986 float *hdr = stbi__hdr_load(s, x,y,comp,req_comp, ri);
987 return stbi__hdr_to_ldr(hdr, *x, *y, req_comp ? req_comp : *comp);
988 }
989 #endif
990
991 #ifndef STBI_NO_TGA
992 // test tga last because it's a crappy test!
993 if (stbi__tga_test(s))
994 return stbi__tga_load(s,x,y,comp,req_comp, ri);
995 #endif
996
997 return stbi__errpuc("unknown image type", "Image not of any known type, or corrupt");
998}
999
1000static stbi_uc *stbi__convert_16_to_8(stbi__uint16 *orig, int w, int h, int channels)
1001{
1002 int i;
1003 int img_len = w * h * channels;
1004 stbi_uc *reduced;
1005
1006 reduced = (stbi_uc *) stbi__malloc(img_len);
1007 if (reduced == NULL) return stbi__errpuc("outofmem", "Out of memory");
1008
1009 for (i = 0; i < img_len; ++i)
1010 reduced[i] = (stbi_uc)((orig[i] >> 8) & 0xFF); // top half of each byte is sufficient approx of 16->8 bit scaling
1011
1012 STBI_FREE(orig);
1013 return reduced;
1014}
1015
1016static stbi__uint16 *stbi__convert_8_to_16(stbi_uc *orig, int w, int h, int channels)
1017{
1018 int i;
1019 int img_len = w * h * channels;
1020 stbi__uint16 *enlarged;
1021
1022 enlarged = (stbi__uint16 *) stbi__malloc(img_len*2);
1023 if (enlarged == NULL) return (stbi__uint16 *) stbi__errpuc("outofmem", "Out of memory");
1024
1025 for (i = 0; i < img_len; ++i)
1026 enlarged[i] = (stbi__uint16)((orig[i] << 8) + orig[i]); // replicate to high and low byte, maps 0->0, 255->0xffff
1027
1028 STBI_FREE(orig);
1029 return enlarged;
1030}
1031
1032static unsigned char *stbi__load_and_postprocess_8bit(stbi__context *s, int *x, int *y, int *comp, int req_comp)
1033{
1034 stbi__result_info ri;
1035 void *result = stbi__load_main(s, x, y, comp, req_comp, &ri, 8);
1036
1037 if (result == NULL)
1038 return NULL;
1039
1040 if (ri.bits_per_channel != 8) {
1041 STBI_ASSERT(ri.bits_per_channel == 16);
1042 result = stbi__convert_16_to_8((stbi__uint16 *) result, *x, *y, req_comp == 0 ? *comp : req_comp);
1043 ri.bits_per_channel = 8;
1044 }
1045
1046 // @TODO: move stbi__convert_format to here
1047
1048 if (stbi__vertically_flip_on_load) {
1049 int w = *x, h = *y;
1050 int channels = req_comp ? req_comp : *comp;
1051 int row,col,z;
1052 stbi_uc *image = (stbi_uc *) result;
1053
1054 // @OPTIMIZE: use a bigger temp buffer and memcpy multiple pixels at once
1055 for (row = 0; row < (h>>1); row++) {
1056 for (col = 0; col < w; col++) {
1057 for (z = 0; z < channels; z++) {
1058 stbi_uc temp = image[(row * w + col) * channels + z];
1059 image[(row * w + col) * channels + z] = image[((h - row - 1) * w + col) * channels + z];
1060 image[((h - row - 1) * w + col) * channels + z] = temp;
1061 }
1062 }
1063 }
1064 }
1065
1066 return (unsigned char *) result;
1067}
1068
1069static stbi__uint16 *stbi__load_and_postprocess_16bit(stbi__context *s, int *x, int *y, int *comp, int req_comp)
1070{
1071 stbi__result_info ri;
1072 void *result = stbi__load_main(s, x, y, comp, req_comp, &ri, 16);
1073
1074 if (result == NULL)
1075 return NULL;
1076
1077 if (ri.bits_per_channel != 16) {
1078 STBI_ASSERT(ri.bits_per_channel == 8);
1079 result = stbi__convert_8_to_16((stbi_uc *) result, *x, *y, req_comp == 0 ? *comp : req_comp);
1080 ri.bits_per_channel = 16;
1081 }
1082
1083 // @TODO: move stbi__convert_format16 to here
1084 // @TODO: special case RGB-to-Y (and RGBA-to-YA) for 8-bit-to-16-bit case to keep more precision
1085
1086 if (stbi__vertically_flip_on_load) {
1087 int w = *x, h = *y;
1088 int channels = req_comp ? req_comp : *comp;
1089 int row,col,z;
1090 stbi__uint16 *image = (stbi__uint16 *) result;
1091
1092 // @OPTIMIZE: use a bigger temp buffer and memcpy multiple pixels at once
1093 for (row = 0; row < (h>>1); row++) {
1094 for (col = 0; col < w; col++) {
1095 for (z = 0; z < channels; z++) {
1096 stbi__uint16 temp = image[(row * w + col) * channels + z];
1097 image[(row * w + col) * channels + z] = image[((h - row - 1) * w + col) * channels + z];
1098 image[((h - row - 1) * w + col) * channels + z] = temp;
1099 }
1100 }
1101 }
1102 }
1103
1104 return (stbi__uint16 *) result;
1105}
1106
1107#ifndef STBI_NO_HDR
1108static void stbi__float_postprocess(float *result, int *x, int *y, int *comp, int req_comp)
1109{
1110 if (stbi__vertically_flip_on_load && result != NULL) {
1111 int w = *x, h = *y;
1112 int depth = req_comp ? req_comp : *comp;
1113 int row,col,z;
1114 float temp;
1115
1116 // @OPTIMIZE: use a bigger temp buffer and memcpy multiple pixels at once
1117 for (row = 0; row < (h>>1); row++) {
1118 for (col = 0; col < w; col++) {
1119 for (z = 0; z < depth; z++) {
1120 temp = result[(row * w + col) * depth + z];
1121 result[(row * w + col) * depth + z] = result[((h - row - 1) * w + col) * depth + z];
1122 result[((h - row - 1) * w + col) * depth + z] = temp;
1123 }
1124 }
1125 }
1126 }
1127}
1128#endif
1129
1130#ifndef STBI_NO_STDIO
1131
1132static FILE *stbi__fopen(char const *filename, char const *mode)
1133{
1134 FILE *f;
1135#if defined(_MSC_VER) && _MSC_VER >= 1400
1136 if (0 != fopen_s(&f, filename, mode))
1137 f=0;
1138#else
1139 f = fopen(filename, mode);
1140#endif
1141 return f;
1142}
1143
1144
1145STBIDEF stbi_uc *stbi_load(char const *filename, int *x, int *y, int *comp, int req_comp)
1146{
1147 FILE *f = stbi__fopen(filename, "rb");
1148 unsigned char *result;
1149 if (!f) return stbi__errpuc("can't fopen", "Unable to open file");
1150 result = stbi_load_from_file(f,x,y,comp,req_comp);
1151 fclose(f);
1152 return result;
1153}
1154
1155STBIDEF stbi_uc *stbi_load_from_file(FILE *f, int *x, int *y, int *comp, int req_comp)
1156{
1157 unsigned char *result;
1158 stbi__context s;
1159 stbi__start_file(&s,f);
1160 result = stbi__load_and_postprocess_8bit(&s,x,y,comp,req_comp);
1161 if (result) {
1162 // need to 'unget' all the characters in the IO buffer
1163 fseek(f, - (int) (s.img_buffer_end - s.img_buffer), SEEK_CUR);
1164 }
1165 return result;
1166}
1167
1168STBIDEF stbi__uint16 *stbi_load_from_file_16(FILE *f, int *x, int *y, int *comp, int req_comp)
1169{
1170 stbi__uint16 *result;
1171 stbi__context s;
1172 stbi__start_file(&s,f);
1173 result = stbi__load_and_postprocess_16bit(&s,x,y,comp,req_comp);
1174 if (result) {
1175 // need to 'unget' all the characters in the IO buffer
1176 fseek(f, - (int) (s.img_buffer_end - s.img_buffer), SEEK_CUR);
1177 }
1178 return result;
1179}
1180
1181STBIDEF stbi_us *stbi_load_16(char const *filename, int *x, int *y, int *comp, int req_comp)
1182{
1183 FILE *f = stbi__fopen(filename, "rb");
1184 stbi__uint16 *result;
1185 if (!f) return (stbi_us *) stbi__errpuc("can't fopen", "Unable to open file");
1186 result = stbi_load_from_file_16(f,x,y,comp,req_comp);
1187 fclose(f);
1188 return result;
1189}
1190
1191
1192#endif //!STBI_NO_STDIO
1193
1194STBIDEF stbi_uc *stbi_load_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp)
1195{
1196 stbi__context s;
1197 stbi__start_mem(&s,buffer,len);
1198 return stbi__load_and_postprocess_8bit(&s,x,y,comp,req_comp);
1199}
1200
1201STBIDEF stbi_uc *stbi_load_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp, int req_comp)
1202{
1203 stbi__context s;
1204 stbi__start_callbacks(&s, (stbi_io_callbacks *) clbk, user);
1205 return stbi__load_and_postprocess_8bit(&s,x,y,comp,req_comp);
1206}
1207
1208#ifndef STBI_NO_LINEAR
1209static float *stbi__loadf_main(stbi__context *s, int *x, int *y, int *comp, int req_comp)
1210{
1211 unsigned char *data;
1212 #ifndef STBI_NO_HDR
1213 if (stbi__hdr_test(s)) {
1214 stbi__result_info ri;
1215 float *hdr_data = stbi__hdr_load(s,x,y,comp,req_comp, &ri);
1216 if (hdr_data)
1217 stbi__float_postprocess(hdr_data,x,y,comp,req_comp);
1218 return hdr_data;
1219 }
1220 #endif
1221 data = stbi__load_and_postprocess_8bit(s, x, y, comp, req_comp);
1222 if (data)
1223 return stbi__ldr_to_hdr(data, *x, *y, req_comp ? req_comp : *comp);
1224 return stbi__errpf("unknown image type", "Image not of any known type, or corrupt");
1225}
1226
1227STBIDEF float *stbi_loadf_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp)
1228{
1229 stbi__context s;
1230 stbi__start_mem(&s,buffer,len);
1231 return stbi__loadf_main(&s,x,y,comp,req_comp);
1232}
1233
1234STBIDEF float *stbi_loadf_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp, int req_comp)
1235{
1236 stbi__context s;
1237 stbi__start_callbacks(&s, (stbi_io_callbacks *) clbk, user);
1238 return stbi__loadf_main(&s,x,y,comp,req_comp);
1239}
1240
1241#ifndef STBI_NO_STDIO
1242STBIDEF float *stbi_loadf(char const *filename, int *x, int *y, int *comp, int req_comp)
1243{
1244 float *result;
1245 FILE *f = stbi__fopen(filename, "rb");
1246 if (!f) return stbi__errpf("can't fopen", "Unable to open file");
1247 result = stbi_loadf_from_file(f,x,y,comp,req_comp);
1248 fclose(f);
1249 return result;
1250}
1251
1252STBIDEF float *stbi_loadf_from_file(FILE *f, int *x, int *y, int *comp, int req_comp)
1253{
1254 stbi__context s;
1255 stbi__start_file(&s,f);
1256 return stbi__loadf_main(&s,x,y,comp,req_comp);
1257}
1258#endif // !STBI_NO_STDIO
1259
1260#endif // !STBI_NO_LINEAR
1261
1262// these is-hdr-or-not is defined independent of whether STBI_NO_LINEAR is
1263// defined, for API simplicity; if STBI_NO_LINEAR is defined, it always
1264// reports false!
1265
1266STBIDEF int stbi_is_hdr_from_memory(stbi_uc const *buffer, int len)
1267{
1268 #ifndef STBI_NO_HDR
1269 stbi__context s;
1270 stbi__start_mem(&s,buffer,len);
1271 return stbi__hdr_test(&s);
1272 #else
1273 STBI_NOTUSED(buffer);
1274 STBI_NOTUSED(len);
1275 return 0;
1276 #endif
1277}
1278
1279#ifndef STBI_NO_STDIO
1280STBIDEF int stbi_is_hdr (char const *filename)
1281{
1282 FILE *f = stbi__fopen(filename, "rb");
1283 int result=0;
1284 if (f) {
1285 result = stbi_is_hdr_from_file(f);
1286 fclose(f);
1287 }
1288 return result;
1289}
1290
1291STBIDEF int stbi_is_hdr_from_file(FILE *f)
1292{
1293 #ifndef STBI_NO_HDR
1294 stbi__context s;
1295 stbi__start_file(&s,f);
1296 return stbi__hdr_test(&s);
1297 #else
1298 STBI_NOTUSED(f);
1299 return 0;
1300 #endif
1301}
1302#endif // !STBI_NO_STDIO
1303
1304STBIDEF int stbi_is_hdr_from_callbacks(stbi_io_callbacks const *clbk, void *user)
1305{
1306 #ifndef STBI_NO_HDR
1307 stbi__context s;
1308 stbi__start_callbacks(&s, (stbi_io_callbacks *) clbk, user);
1309 return stbi__hdr_test(&s);
1310 #else
1311 STBI_NOTUSED(clbk);
1312 STBI_NOTUSED(user);
1313 return 0;
1314 #endif
1315}
1316
1317#ifndef STBI_NO_LINEAR
1318static float stbi__l2h_gamma=2.2f, stbi__l2h_scale=1.0f;
1319
1320STBIDEF void stbi_ldr_to_hdr_gamma(float gamma) { stbi__l2h_gamma = gamma; }
1321STBIDEF void stbi_ldr_to_hdr_scale(float scale) { stbi__l2h_scale = scale; }
1322#endif
1323
1324static float stbi__h2l_gamma_i=1.0f/2.2f, stbi__h2l_scale_i=1.0f;
1325
1326STBIDEF void stbi_hdr_to_ldr_gamma(float gamma) { stbi__h2l_gamma_i = 1/gamma; }
1327STBIDEF void stbi_hdr_to_ldr_scale(float scale) { stbi__h2l_scale_i = 1/scale; }
1328
1329
1330//////////////////////////////////////////////////////////////////////////////
1331//
1332// Common code used by all image loaders
1333//
1334
1335enum
1336{
1337 STBI__SCAN_load=0,
1338 STBI__SCAN_type,
1339 STBI__SCAN_header
1340};
1341
1342static void stbi__refill_buffer(stbi__context *s)
1343{
1344 int n = (s->io.read)(s->io_user_data,(char*)s->buffer_start,s->buflen);
1345 if (n == 0) {
1346 // at end of file, treat same as if from memory, but need to handle case
1347 // where s->img_buffer isn't pointing to safe memory, e.g. 0-byte file
1348 s->read_from_callbacks = 0;
1349 s->img_buffer = s->buffer_start;
1350 s->img_buffer_end = s->buffer_start+1;
1351 *s->img_buffer = 0;
1352 } else {
1353 s->img_buffer = s->buffer_start;
1354 s->img_buffer_end = s->buffer_start + n;
1355 }
1356}
1357
1358stbi_inline static stbi_uc stbi__get8(stbi__context *s)
1359{
1360 if (s->img_buffer < s->img_buffer_end)
1361 return *s->img_buffer++;
1362 if (s->read_from_callbacks) {
1363 stbi__refill_buffer(s);
1364 return *s->img_buffer++;
1365 }
1366 return 0;
1367}
1368
1369stbi_inline static int stbi__at_eof(stbi__context *s)
1370{
1371 if (s->io.read) {
1372 if (!(s->io.eof)(s->io_user_data)) return 0;
1373 // if feof() is true, check if buffer = end
1374 // special case: we've only got the special 0 character at the end
1375 if (s->read_from_callbacks == 0) return 1;
1376 }
1377
1378 return s->img_buffer >= s->img_buffer_end;
1379}
1380
1381static void stbi__skip(stbi__context *s, int n)
1382{
1383 if (n < 0) {
1384 s->img_buffer = s->img_buffer_end;
1385 return;
1386 }
1387 if (s->io.read) {
1388 int blen = (int) (s->img_buffer_end - s->img_buffer);
1389 if (blen < n) {
1390 s->img_buffer = s->img_buffer_end;
1391 (s->io.skip)(s->io_user_data, n - blen);
1392 return;
1393 }
1394 }
1395 s->img_buffer += n;
1396}
1397
1398static int stbi__getn(stbi__context *s, stbi_uc *buffer, int n)
1399{
1400 if (s->io.read) {
1401 int blen = (int) (s->img_buffer_end - s->img_buffer);
1402 if (blen < n) {
1403 int res, count;
1404
1405 memcpy(buffer, s->img_buffer, blen);
1406
1407 count = (s->io.read)(s->io_user_data, (char*) buffer + blen, n - blen);
1408 res = (count == (n-blen));
1409 s->img_buffer = s->img_buffer_end;
1410 return res;
1411 }
1412 }
1413
1414 if (s->img_buffer+n <= s->img_buffer_end) {
1415 memcpy(buffer, s->img_buffer, n);
1416 s->img_buffer += n;
1417 return 1;
1418 } else
1419 return 0;
1420}
1421
1422static int stbi__get16be(stbi__context *s)
1423{
1424 int z = stbi__get8(s);
1425 return (z << 8) + stbi__get8(s);
1426}
1427
1428static stbi__uint32 stbi__get32be(stbi__context *s)
1429{
1430 stbi__uint32 z = stbi__get16be(s);
1431 return (z << 16) + stbi__get16be(s);
1432}
1433
1434#if defined(STBI_NO_BMP) && defined(STBI_NO_TGA) && defined(STBI_NO_GIF)
1435// nothing
1436#else
1437static int stbi__get16le(stbi__context *s)
1438{
1439 int z = stbi__get8(s);
1440 return z + (stbi__get8(s) << 8);
1441}
1442#endif
1443
1444#ifndef STBI_NO_BMP
1445static stbi__uint32 stbi__get32le(stbi__context *s)
1446{
1447 stbi__uint32 z = stbi__get16le(s);
1448 return z + (stbi__get16le(s) << 16);
1449}
1450#endif
1451
1452#define STBI__BYTECAST(x) ((stbi_uc) ((x) & 255)) // truncate int to byte without warnings
1453
1454
1455//////////////////////////////////////////////////////////////////////////////
1456//
1457// generic converter from built-in img_n to req_comp
1458// individual types do this automatically as much as possible (e.g. jpeg
1459// does all cases internally since it needs to colorspace convert anyway,
1460// and it never has alpha, so very few cases ). png can automatically
1461// interleave an alpha=255 channel, but falls back to this for other cases
1462//
1463// assume data buffer is malloced, so malloc a new one and free that one
1464// only failure mode is malloc failing
1465
1466static stbi_uc stbi__compute_y(int r, int g, int b)
1467{
1468 return (stbi_uc) (((r*77) + (g*150) + (29*b)) >> 8);
1469}
1470
1471static unsigned char *stbi__convert_format(unsigned char *data, int img_n, int req_comp, unsigned int x, unsigned int y)
1472{
1473 int i,j;
1474 unsigned char *good;
1475
1476 if (req_comp == img_n) return data;
1477 STBI_ASSERT(req_comp >= 1 && req_comp <= 4);
1478
1479 good = (unsigned char *) stbi__malloc_mad3(req_comp, x, y, 0);
1480 if (good == NULL) {
1481 STBI_FREE(data);
1482 return stbi__errpuc("outofmem", "Out of memory");
1483 }
1484
1485 for (j=0; j < (int) y; ++j) {
1486 unsigned char *src = data + j * x * img_n ;
1487 unsigned char *dest = good + j * x * req_comp;
1488
1489 #define STBI__COMBO(a,b) ((a)*8+(b))
1490 #define STBI__CASE(a,b) case STBI__COMBO(a,b): for(i=x-1; i >= 0; --i, src += a, dest += b)
1491 // convert source image with img_n components to one with req_comp components;
1492 // avoid switch per pixel, so use switch per scanline and massive macros
1493 switch (STBI__COMBO(img_n, req_comp)) {
1494 STBI__CASE(1,2) { dest[0]=src[0], dest[1]=255; } break;
1495 STBI__CASE(1,3) { dest[0]=dest[1]=dest[2]=src[0]; } break;
1496 STBI__CASE(1,4) { dest[0]=dest[1]=dest[2]=src[0], dest[3]=255; } break;
1497 STBI__CASE(2,1) { dest[0]=src[0]; } break;
1498 STBI__CASE(2,3) { dest[0]=dest[1]=dest[2]=src[0]; } break;
1499 STBI__CASE(2,4) { dest[0]=dest[1]=dest[2]=src[0], dest[3]=src[1]; } break;
1500 STBI__CASE(3,4) { dest[0]=src[0],dest[1]=src[1],dest[2]=src[2],dest[3]=255; } break;
1501 STBI__CASE(3,1) { dest[0]=stbi__compute_y(src[0],src[1],src[2]); } break;
1502 STBI__CASE(3,2) { dest[0]=stbi__compute_y(src[0],src[1],src[2]), dest[1] = 255; } break;
1503 STBI__CASE(4,1) { dest[0]=stbi__compute_y(src[0],src[1],src[2]); } break;
1504 STBI__CASE(4,2) { dest[0]=stbi__compute_y(src[0],src[1],src[2]), dest[1] = src[3]; } break;
1505 STBI__CASE(4,3) { dest[0]=src[0],dest[1]=src[1],dest[2]=src[2]; } break;
1506 default: STBI_ASSERT(0);
1507 }
1508 #undef STBI__CASE
1509 }
1510
1511 STBI_FREE(data);
1512 return good;
1513}
1514
1515static stbi__uint16 stbi__compute_y_16(int r, int g, int b)
1516{
1517 return (stbi__uint16) (((r*77) + (g*150) + (29*b)) >> 8);
1518}
1519
1520static stbi__uint16 *stbi__convert_format16(stbi__uint16 *data, int img_n, int req_comp, unsigned int x, unsigned int y)
1521{
1522 int i,j;
1523 stbi__uint16 *good;
1524
1525 if (req_comp == img_n) return data;
1526 STBI_ASSERT(req_comp >= 1 && req_comp <= 4);
1527
1528 good = (stbi__uint16 *) stbi__malloc(req_comp * x * y * 2);
1529 if (good == NULL) {
1530 STBI_FREE(data);
1531 return (stbi__uint16 *) stbi__errpuc("outofmem", "Out of memory");
1532 }
1533
1534 for (j=0; j < (int) y; ++j) {
1535 stbi__uint16 *src = data + j * x * img_n ;
1536 stbi__uint16 *dest = good + j * x * req_comp;
1537
1538 #define STBI__COMBO(a,b) ((a)*8+(b))
1539 #define STBI__CASE(a,b) case STBI__COMBO(a,b): for(i=x-1; i >= 0; --i, src += a, dest += b)
1540 // convert source image with img_n components to one with req_comp components;
1541 // avoid switch per pixel, so use switch per scanline and massive macros
1542 switch (STBI__COMBO(img_n, req_comp)) {
1543 STBI__CASE(1,2) { dest[0]=src[0], dest[1]=0xffff; } break;
1544 STBI__CASE(1,3) { dest[0]=dest[1]=dest[2]=src[0]; } break;
1545 STBI__CASE(1,4) { dest[0]=dest[1]=dest[2]=src[0], dest[3]=0xffff; } break;
1546 STBI__CASE(2,1) { dest[0]=src[0]; } break;
1547 STBI__CASE(2,3) { dest[0]=dest[1]=dest[2]=src[0]; } break;
1548 STBI__CASE(2,4) { dest[0]=dest[1]=dest[2]=src[0], dest[3]=src[1]; } break;
1549 STBI__CASE(3,4) { dest[0]=src[0],dest[1]=src[1],dest[2]=src[2],dest[3]=0xffff; } break;
1550 STBI__CASE(3,1) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]); } break;
1551 STBI__CASE(3,2) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]), dest[1] = 0xffff; } break;
1552 STBI__CASE(4,1) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]); } break;
1553 STBI__CASE(4,2) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]), dest[1] = src[3]; } break;
1554 STBI__CASE(4,3) { dest[0]=src[0],dest[1]=src[1],dest[2]=src[2]; } break;
1555 default: STBI_ASSERT(0);
1556 }
1557 #undef STBI__CASE
1558 }
1559
1560 STBI_FREE(data);
1561 return good;
1562}
1563
1564#ifndef STBI_NO_LINEAR
1565static float *stbi__ldr_to_hdr(stbi_uc *data, int x, int y, int comp)
1566{
1567 int i,k,n;
1568 float *output;
1569 if (!data) return NULL;
1570 output = (float *) stbi__malloc_mad4(x, y, comp, sizeof(float), 0);
1571 if (output == NULL) { STBI_FREE(data); return stbi__errpf("outofmem", "Out of memory"); }
1572 // compute number of non-alpha components
1573 if (comp & 1) n = comp; else n = comp-1;
1574 for (i=0; i < x*y; ++i) {
1575 for (k=0; k < n; ++k) {
1576 output[i*comp + k] = (float) (pow(data[i*comp+k]/255.0f, stbi__l2h_gamma) * stbi__l2h_scale);
1577 }
1578 if (k < comp) output[i*comp + k] = data[i*comp+k]/255.0f;
1579 }
1580 STBI_FREE(data);
1581 return output;
1582}
1583#endif
1584
1585#ifndef STBI_NO_HDR
1586#define stbi__float2int(x) ((int) (x))
1587static stbi_uc *stbi__hdr_to_ldr(float *data, int x, int y, int comp)
1588{
1589 int i,k,n;
1590 stbi_uc *output;
1591 if (!data) return NULL;
1592 output = (stbi_uc *) stbi__malloc_mad3(x, y, comp, 0);
1593 if (output == NULL) { STBI_FREE(data); return stbi__errpuc("outofmem", "Out of memory"); }
1594 // compute number of non-alpha components
1595 if (comp & 1) n = comp; else n = comp-1;
1596 for (i=0; i < x*y; ++i) {
1597 for (k=0; k < n; ++k) {
1598 float z = (float) pow(data[i*comp+k]*stbi__h2l_scale_i, stbi__h2l_gamma_i) * 255 + 0.5f;
1599 if (z < 0) z = 0;
1600 if (z > 255) z = 255;
1601 output[i*comp + k] = (stbi_uc) stbi__float2int(z);
1602 }
1603 if (k < comp) {
1604 float z = data[i*comp+k] * 255 + 0.5f;
1605 if (z < 0) z = 0;
1606 if (z > 255) z = 255;
1607 output[i*comp + k] = (stbi_uc) stbi__float2int(z);
1608 }
1609 }
1610 STBI_FREE(data);
1611 return output;
1612}
1613#endif
1614
1615//////////////////////////////////////////////////////////////////////////////
1616//
1617// "baseline" JPEG/JFIF decoder
1618//
1619// simple implementation
1620// - doesn't support delayed output of y-dimension
1621// - simple interface (only one output format: 8-bit interleaved RGB)
1622// - doesn't try to recover corrupt jpegs
1623// - doesn't allow partial loading, loading multiple at once
1624// - still fast on x86 (copying globals into locals doesn't help x86)
1625// - allocates lots of intermediate memory (full size of all components)
1626// - non-interleaved case requires this anyway
1627// - allows good upsampling (see next)
1628// high-quality
1629// - upsampled channels are bilinearly interpolated, even across blocks
1630// - quality integer IDCT derived from IJG's 'slow'
1631// performance
1632// - fast huffman; reasonable integer IDCT
1633// - some SIMD kernels for common paths on targets with SSE2/NEON
1634// - uses a lot of intermediate memory, could cache poorly
1635
1636#ifndef STBI_NO_JPEG
1637
1638// huffman decoding acceleration
1639#define FAST_BITS 9 // larger handles more cases; smaller stomps less cache
1640
1641typedef struct
1642{
1643 stbi_uc fast[1 << FAST_BITS];
1644 // weirdly, repacking this into AoS is a 10% speed loss, instead of a win
1645 stbi__uint16 code[256];
1646 stbi_uc values[256];
1647 stbi_uc size[257];
1648 unsigned int maxcode[18];
1649 int delta[17]; // old 'firstsymbol' - old 'firstcode'
1650} stbi__huffman;
1651
1652typedef struct
1653{
1654 stbi__context *s;
1655 stbi__huffman huff_dc[4];
1656 stbi__huffman huff_ac[4];
1657 stbi__uint16 dequant[4][64];
1658 stbi__int16 fast_ac[4][1 << FAST_BITS];
1659
1660// sizes for components, interleaved MCUs
1661 int img_h_max, img_v_max;
1662 int img_mcu_x, img_mcu_y;
1663 int img_mcu_w, img_mcu_h;
1664
1665// definition of jpeg image component
1666 struct
1667 {
1668 int id;
1669 int h,v;
1670 int tq;
1671 int hd,ha;
1672 int dc_pred;
1673
1674 int x,y,w2,h2;
1675 stbi_uc *data;
1676 void *raw_data, *raw_coeff;
1677 stbi_uc *linebuf;
1678 short *coeff; // progressive only
1679 int coeff_w, coeff_h; // number of 8x8 coefficient blocks
1680 } img_comp[4];
1681
1682 stbi__uint32 code_buffer; // jpeg entropy-coded buffer
1683 int code_bits; // number of valid bits
1684 unsigned char marker; // marker seen while filling entropy buffer
1685 int nomore; // flag if we saw a marker so must stop
1686
1687 int progressive;
1688 int spec_start;
1689 int spec_end;
1690 int succ_high;
1691 int succ_low;
1692 int eob_run;
1693 int jfif;
1694 int app14_color_transform; // Adobe APP14 tag
1695 int rgb;
1696
1697 int scan_n, order[4];
1698 int restart_interval, todo;
1699
1700// kernels
1701 void (*idct_block_kernel)(stbi_uc *out, int out_stride, short data[64]);
1702 void (*YCbCr_to_RGB_kernel)(stbi_uc *out, const stbi_uc *y, const stbi_uc *pcb, const stbi_uc *pcr, int count, int step);
1703 stbi_uc *(*resample_row_hv_2_kernel)(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs);
1704} stbi__jpeg;
1705
1706static int stbi__build_huffman(stbi__huffman *h, int *count)
1707{
1708 int i,j,k=0,code;
1709 // build size list for each symbol (from JPEG spec)
1710 for (i=0; i < 16; ++i)
1711 for (j=0; j < count[i]; ++j)
1712 h->size[k++] = (stbi_uc) (i+1);
1713 h->size[k] = 0;
1714
1715 // compute actual symbols (from jpeg spec)
1716 code = 0;
1717 k = 0;
1718 for(j=1; j <= 16; ++j) {
1719 // compute delta to add to code to compute symbol id
1720 h->delta[j] = k - code;
1721 if (h->size[k] == j) {
1722 while (h->size[k] == j)
1723 h->code[k++] = (stbi__uint16) (code++);
1724 if (code-1 >= (1 << j)) return stbi__err("bad code lengths","Corrupt JPEG");
1725 }
1726 // compute largest code + 1 for this size, preshifted as needed later
1727 h->maxcode[j] = code << (16-j);
1728 code <<= 1;
1729 }
1730 h->maxcode[j] = 0xffffffff;
1731
1732 // build non-spec acceleration table; 255 is flag for not-accelerated
1733 memset(h->fast, 255, 1 << FAST_BITS);
1734 for (i=0; i < k; ++i) {
1735 int s = h->size[i];
1736 if (s <= FAST_BITS) {
1737 int c = h->code[i] << (FAST_BITS-s);
1738 int m = 1 << (FAST_BITS-s);
1739 for (j=0; j < m; ++j) {
1740 h->fast[c+j] = (stbi_uc) i;
1741 }
1742 }
1743 }
1744 return 1;
1745}
1746
1747// build a table that decodes both magnitude and value of small ACs in
1748// one go.
1749static void stbi__build_fast_ac(stbi__int16 *fast_ac, stbi__huffman *h)
1750{
1751 int i;
1752 for (i=0; i < (1 << FAST_BITS); ++i) {
1753 stbi_uc fast = h->fast[i];
1754 fast_ac[i] = 0;
1755 if (fast < 255) {
1756 int rs = h->values[fast];
1757 int run = (rs >> 4) & 15;
1758 int magbits = rs & 15;
1759 int len = h->size[fast];
1760
1761 if (magbits && len + magbits <= FAST_BITS) {
1762 // magnitude code followed by receive_extend code
1763 int k = ((i << len) & ((1 << FAST_BITS) - 1)) >> (FAST_BITS - magbits);
1764 int m = 1 << (magbits - 1);
1765 if (k < m) k += (~0U << magbits) + 1;
1766 // if the result is small enough, we can fit it in fast_ac table
1767 if (k >= -128 && k <= 127)
1768 fast_ac[i] = (stbi__int16) ((k << 8) + (run << 4) + (len + magbits));
1769 }
1770 }
1771 }
1772}
1773
1774static void stbi__grow_buffer_unsafe(stbi__jpeg *j)
1775{
1776 do {
1777 int b = j->nomore ? 0 : stbi__get8(j->s);
1778 if (b == 0xff) {
1779 int c = stbi__get8(j->s);
1780 while (c == 0xff) c = stbi__get8(j->s); // consume fill bytes
1781 if (c != 0) {
1782 j->marker = (unsigned char) c;
1783 j->nomore = 1;
1784 return;
1785 }
1786 }
1787 j->code_buffer |= b << (24 - j->code_bits);
1788 j->code_bits += 8;
1789 } while (j->code_bits <= 24);
1790}
1791
1792// (1 << n) - 1
1793static stbi__uint32 stbi__bmask[17]={0,1,3,7,15,31,63,127,255,511,1023,2047,4095,8191,16383,32767,65535};
1794
1795// decode a jpeg huffman value from the bitstream
1796stbi_inline static int stbi__jpeg_huff_decode(stbi__jpeg *j, stbi__huffman *h)
1797{
1798 unsigned int temp;
1799 int c,k;
1800
1801 if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
1802
1803 // look at the top FAST_BITS and determine what symbol ID it is,
1804 // if the code is <= FAST_BITS
1805 c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS)-1);
1806 k = h->fast[c];
1807 if (k < 255) {
1808 int s = h->size[k];
1809 if (s > j->code_bits)
1810 return -1;
1811 j->code_buffer <<= s;
1812 j->code_bits -= s;
1813 return h->values[k];
1814 }
1815
1816 // naive test is to shift the code_buffer down so k bits are
1817 // valid, then test against maxcode. To speed this up, we've
1818 // preshifted maxcode left so that it has (16-k) 0s at the
1819 // end; in other words, regardless of the number of bits, it
1820 // wants to be compared against something shifted to have 16;
1821 // that way we don't need to shift inside the loop.
1822 temp = j->code_buffer >> 16;
1823 for (k=FAST_BITS+1 ; ; ++k)
1824 if (temp < h->maxcode[k])
1825 break;
1826 if (k == 17) {
1827 // error! code not found
1828 j->code_bits -= 16;
1829 return -1;
1830 }
1831
1832 if (k > j->code_bits)
1833 return -1;
1834
1835 // convert the huffman code to the symbol id
1836 c = ((j->code_buffer >> (32 - k)) & stbi__bmask[k]) + h->delta[k];
1837 STBI_ASSERT((((j->code_buffer) >> (32 - h->size[c])) & stbi__bmask[h->size[c]]) == h->code[c]);
1838
1839 // convert the id to a symbol
1840 j->code_bits -= k;
1841 j->code_buffer <<= k;
1842 return h->values[c];
1843}
1844
1845// bias[n] = (-1<<n) + 1
1846static int const stbi__jbias[16] = {0,-1,-3,-7,-15,-31,-63,-127,-255,-511,-1023,-2047,-4095,-8191,-16383,-32767};
1847
1848// combined JPEG 'receive' and JPEG 'extend', since baseline
1849// always extends everything it receives.
1850stbi_inline static int stbi__extend_receive(stbi__jpeg *j, int n)
1851{
1852 unsigned int k;
1853 int sgn;
1854 if (j->code_bits < n) stbi__grow_buffer_unsafe(j);
1855
1856 sgn = (stbi__int32)j->code_buffer >> 31; // sign bit is always in MSB
1857 k = stbi_lrot(j->code_buffer, n);
1858 STBI_ASSERT(n >= 0 && n < (int) (sizeof(stbi__bmask)/sizeof(*stbi__bmask)));
1859 j->code_buffer = k & ~stbi__bmask[n];
1860 k &= stbi__bmask[n];
1861 j->code_bits -= n;
1862 return k + (stbi__jbias[n] & ~sgn);
1863}
1864
1865// get some unsigned bits
1866stbi_inline static int stbi__jpeg_get_bits(stbi__jpeg *j, int n)
1867{
1868 unsigned int k;
1869 if (j->code_bits < n) stbi__grow_buffer_unsafe(j);
1870 k = stbi_lrot(j->code_buffer, n);
1871 j->code_buffer = k & ~stbi__bmask[n];
1872 k &= stbi__bmask[n];
1873 j->code_bits -= n;
1874 return k;
1875}
1876
1877stbi_inline static int stbi__jpeg_get_bit(stbi__jpeg *j)
1878{
1879 unsigned int k;
1880 if (j->code_bits < 1) stbi__grow_buffer_unsafe(j);
1881 k = j->code_buffer;
1882 j->code_buffer <<= 1;
1883 --j->code_bits;
1884 return k & 0x80000000;
1885}
1886
1887// given a value that's at position X in the zigzag stream,
1888// where does it appear in the 8x8 matrix coded as row-major?
1889static stbi_uc stbi__jpeg_dezigzag[64+15] =
1890{
1891 0, 1, 8, 16, 9, 2, 3, 10,
1892 17, 24, 32, 25, 18, 11, 4, 5,
1893 12, 19, 26, 33, 40, 48, 41, 34,
1894 27, 20, 13, 6, 7, 14, 21, 28,
1895 35, 42, 49, 56, 57, 50, 43, 36,
1896 29, 22, 15, 23, 30, 37, 44, 51,
1897 58, 59, 52, 45, 38, 31, 39, 46,
1898 53, 60, 61, 54, 47, 55, 62, 63,
1899 // let corrupt input sample past end
1900 63, 63, 63, 63, 63, 63, 63, 63,
1901 63, 63, 63, 63, 63, 63, 63
1902};
1903
1904// decode one 64-entry block--
1905static int stbi__jpeg_decode_block(stbi__jpeg *j, short data[64], stbi__huffman *hdc, stbi__huffman *hac, stbi__int16 *fac, int b, stbi__uint16 *dequant)
1906{
1907 int diff,dc,k;
1908 int t;
1909
1910 if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
1911 t = stbi__jpeg_huff_decode(j, hdc);
1912 if (t < 0) return stbi__err("bad huffman code","Corrupt JPEG");
1913
1914 // 0 all the ac values now so we can do it 32-bits at a time
1915 memset(data,0,64*sizeof(data[0]));
1916
1917 diff = t ? stbi__extend_receive(j, t) : 0;
1918 dc = j->img_comp[b].dc_pred + diff;
1919 j->img_comp[b].dc_pred = dc;
1920 data[0] = (short) (dc * dequant[0]);
1921
1922 // decode AC components, see JPEG spec
1923 k = 1;
1924 do {
1925 unsigned int zig;
1926 int c,r,s;
1927 if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
1928 c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS)-1);
1929 r = fac[c];
1930 if (r) { // fast-AC path
1931 k += (r >> 4) & 15; // run
1932 s = r & 15; // combined length
1933 j->code_buffer <<= s;
1934 j->code_bits -= s;
1935 // decode into unzigzag'd location
1936 zig = stbi__jpeg_dezigzag[k++];
1937 data[zig] = (short) ((r >> 8) * dequant[zig]);
1938 } else {
1939 int rs = stbi__jpeg_huff_decode(j, hac);
1940 if (rs < 0) return stbi__err("bad huffman code","Corrupt JPEG");
1941 s = rs & 15;
1942 r = rs >> 4;
1943 if (s == 0) {
1944 if (rs != 0xf0) break; // end block
1945 k += 16;
1946 } else {
1947 k += r;
1948 // decode into unzigzag'd location
1949 zig = stbi__jpeg_dezigzag[k++];
1950 data[zig] = (short) (stbi__extend_receive(j,s) * dequant[zig]);
1951 }
1952 }
1953 } while (k < 64);
1954 return 1;
1955}
1956
1957static int stbi__jpeg_decode_block_prog_dc(stbi__jpeg *j, short data[64], stbi__huffman *hdc, int b)
1958{
1959 int diff,dc;
1960 int t;
1961 if (j->spec_end != 0) return stbi__err("can't merge dc and ac", "Corrupt JPEG");
1962
1963 if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
1964
1965 if (j->succ_high == 0) {
1966 // first scan for DC coefficient, must be first
1967 memset(data,0,64*sizeof(data[0])); // 0 all the ac values now
1968 t = stbi__jpeg_huff_decode(j, hdc);
1969 diff = t ? stbi__extend_receive(j, t) : 0;
1970
1971 dc = j->img_comp[b].dc_pred + diff;
1972 j->img_comp[b].dc_pred = dc;
1973 data[0] = (short) (dc << j->succ_low);
1974 } else {
1975 // refinement scan for DC coefficient
1976 if (stbi__jpeg_get_bit(j))
1977 data[0] += (short) (1 << j->succ_low);
1978 }
1979 return 1;
1980}
1981
1982// @OPTIMIZE: store non-zigzagged during the decode passes,
1983// and only de-zigzag when dequantizing
1984static int stbi__jpeg_decode_block_prog_ac(stbi__jpeg *j, short data[64], stbi__huffman *hac, stbi__int16 *fac)
1985{
1986 int k;
1987 if (j->spec_start == 0) return stbi__err("can't merge dc and ac", "Corrupt JPEG");
1988
1989 if (j->succ_high == 0) {
1990 int shift = j->succ_low;
1991
1992 if (j->eob_run) {
1993 --j->eob_run;
1994 return 1;
1995 }
1996
1997 k = j->spec_start;
1998 do {
1999 unsigned int zig;
2000 int c,r,s;
2001 if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
2002 c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS)-1);
2003 r = fac[c];
2004 if (r) { // fast-AC path
2005 k += (r >> 4) & 15; // run
2006 s = r & 15; // combined length
2007 j->code_buffer <<= s;
2008 j->code_bits -= s;
2009 zig = stbi__jpeg_dezigzag[k++];
2010 data[zig] = (short) ((r >> 8) << shift);
2011 } else {
2012 int rs = stbi__jpeg_huff_decode(j, hac);
2013 if (rs < 0) return stbi__err("bad huffman code","Corrupt JPEG");
2014 s = rs & 15;
2015 r = rs >> 4;
2016 if (s == 0) {
2017 if (r < 15) {
2018 j->eob_run = (1 << r);
2019 if (r)
2020 j->eob_run += stbi__jpeg_get_bits(j, r);
2021 --j->eob_run;
2022 break;
2023 }
2024 k += 16;
2025 } else {
2026 k += r;
2027 zig = stbi__jpeg_dezigzag[k++];
2028 data[zig] = (short) (stbi__extend_receive(j,s) << shift);
2029 }
2030 }
2031 } while (k <= j->spec_end);
2032 } else {
2033 // refinement scan for these AC coefficients
2034
2035 short bit = (short) (1 << j->succ_low);
2036
2037 if (j->eob_run) {
2038 --j->eob_run;
2039 for (k = j->spec_start; k <= j->spec_end; ++k) {
2040 short *p = &data[stbi__jpeg_dezigzag[k]];
2041 if (*p != 0)
2042 if (stbi__jpeg_get_bit(j))
2043 if ((*p & bit)==0) {
2044 if (*p > 0)
2045 *p += bit;
2046 else
2047 *p -= bit;
2048 }
2049 }
2050 } else {
2051 k = j->spec_start;
2052 do {
2053 int r,s;
2054 int rs = stbi__jpeg_huff_decode(j, hac); // @OPTIMIZE see if we can use the fast path here, advance-by-r is so slow, eh
2055 if (rs < 0) return stbi__err("bad huffman code","Corrupt JPEG");
2056 s = rs & 15;
2057 r = rs >> 4;
2058 if (s == 0) {
2059 if (r < 15) {
2060 j->eob_run = (1 << r) - 1;
2061 if (r)
2062 j->eob_run += stbi__jpeg_get_bits(j, r);
2063 r = 64; // force end of block
2064 } else {
2065 // r=15 s=0 should write 16 0s, so we just do
2066 // a run of 15 0s and then write s (which is 0),
2067 // so we don't have to do anything special here
2068 }
2069 } else {
2070 if (s != 1) return stbi__err("bad huffman code", "Corrupt JPEG");
2071 // sign bit
2072 if (stbi__jpeg_get_bit(j))
2073 s = bit;
2074 else
2075 s = -bit;
2076 }
2077
2078 // advance by r
2079 while (k <= j->spec_end) {
2080 short *p = &data[stbi__jpeg_dezigzag[k++]];
2081 if (*p != 0) {
2082 if (stbi__jpeg_get_bit(j))
2083 if ((*p & bit)==0) {
2084 if (*p > 0)
2085 *p += bit;
2086 else
2087 *p -= bit;
2088 }
2089 } else {
2090 if (r == 0) {
2091 *p = (short) s;
2092 break;
2093 }
2094 --r;
2095 }
2096 }
2097 } while (k <= j->spec_end);
2098 }
2099 }
2100 return 1;
2101}
2102
2103// take a -128..127 value and stbi__clamp it and convert to 0..255
2104stbi_inline static stbi_uc stbi__clamp(int x)
2105{
2106 // trick to use a single test to catch both cases
2107 if ((unsigned int) x > 255) {
2108 if (x < 0) return 0;
2109 if (x > 255) return 255;
2110 }
2111 return (stbi_uc) x;
2112}
2113
2114#define stbi__f2f(x) ((int) (((x) * 4096 + 0.5)))
2115#define stbi__fsh(x) ((x) << 12)
2116
2117// derived from jidctint -- DCT_ISLOW
2118#define STBI__IDCT_1D(s0,s1,s2,s3,s4,s5,s6,s7) \
2119 int t0,t1,t2,t3,p1,p2,p3,p4,p5,x0,x1,x2,x3; \
2120 p2 = s2; \
2121 p3 = s6; \
2122 p1 = (p2+p3) * stbi__f2f(0.5411961f); \
2123 t2 = p1 + p3*stbi__f2f(-1.847759065f); \
2124 t3 = p1 + p2*stbi__f2f( 0.765366865f); \
2125 p2 = s0; \
2126 p3 = s4; \
2127 t0 = stbi__fsh(p2+p3); \
2128 t1 = stbi__fsh(p2-p3); \
2129 x0 = t0+t3; \
2130 x3 = t0-t3; \
2131 x1 = t1+t2; \
2132 x2 = t1-t2; \
2133 t0 = s7; \
2134 t1 = s5; \
2135 t2 = s3; \
2136 t3 = s1; \
2137 p3 = t0+t2; \
2138 p4 = t1+t3; \
2139 p1 = t0+t3; \
2140 p2 = t1+t2; \
2141 p5 = (p3+p4)*stbi__f2f( 1.175875602f); \
2142 t0 = t0*stbi__f2f( 0.298631336f); \
2143 t1 = t1*stbi__f2f( 2.053119869f); \
2144 t2 = t2*stbi__f2f( 3.072711026f); \
2145 t3 = t3*stbi__f2f( 1.501321110f); \
2146 p1 = p5 + p1*stbi__f2f(-0.899976223f); \
2147 p2 = p5 + p2*stbi__f2f(-2.562915447f); \
2148 p3 = p3*stbi__f2f(-1.961570560f); \
2149 p4 = p4*stbi__f2f(-0.390180644f); \
2150 t3 += p1+p4; \
2151 t2 += p2+p3; \
2152 t1 += p2+p4; \
2153 t0 += p1+p3;
2154
2155static void stbi__idct_block(stbi_uc *out, int out_stride, short data[64])
2156{
2157 int i,val[64],*v=val;
2158 stbi_uc *o;
2159 short *d = data;
2160
2161 // columns
2162 for (i=0; i < 8; ++i,++d, ++v) {
2163 // if all zeroes, shortcut -- this avoids dequantizing 0s and IDCTing
2164 if (d[ 8]==0 && d[16]==0 && d[24]==0 && d[32]==0
2165 && d[40]==0 && d[48]==0 && d[56]==0) {
2166 // no shortcut 0 seconds
2167 // (1|2|3|4|5|6|7)==0 0 seconds
2168 // all separate -0.047 seconds
2169 // 1 && 2|3 && 4|5 && 6|7: -0.047 seconds
2170 int dcterm = d[0] << 2;
2171 v[0] = v[8] = v[16] = v[24] = v[32] = v[40] = v[48] = v[56] = dcterm;
2172 } else {
2173 STBI__IDCT_1D(d[ 0],d[ 8],d[16],d[24],d[32],d[40],d[48],d[56])
2174 // constants scaled things up by 1<<12; let's bring them back
2175 // down, but keep 2 extra bits of precision
2176 x0 += 512; x1 += 512; x2 += 512; x3 += 512;
2177 v[ 0] = (x0+t3) >> 10;
2178 v[56] = (x0-t3) >> 10;
2179 v[ 8] = (x1+t2) >> 10;
2180 v[48] = (x1-t2) >> 10;
2181 v[16] = (x2+t1) >> 10;
2182 v[40] = (x2-t1) >> 10;
2183 v[24] = (x3+t0) >> 10;
2184 v[32] = (x3-t0) >> 10;
2185 }
2186 }
2187
2188 for (i=0, v=val, o=out; i < 8; ++i,v+=8,o+=out_stride) {
2189 // no fast case since the first 1D IDCT spread components out
2190 STBI__IDCT_1D(v[0],v[1],v[2],v[3],v[4],v[5],v[6],v[7])
2191 // constants scaled things up by 1<<12, plus we had 1<<2 from first
2192 // loop, plus horizontal and vertical each scale by sqrt(8) so together
2193 // we've got an extra 1<<3, so 1<<17 total we need to remove.
2194 // so we want to round that, which means adding 0.5 * 1<<17,
2195 // aka 65536. Also, we'll end up with -128 to 127 that we want
2196 // to encode as 0..255 by adding 128, so we'll add that before the shift
2197 x0 += 65536 + (128<<17);
2198 x1 += 65536 + (128<<17);
2199 x2 += 65536 + (128<<17);
2200 x3 += 65536 + (128<<17);
2201 // tried computing the shifts into temps, or'ing the temps to see
2202 // if any were out of range, but that was slower
2203 o[0] = stbi__clamp((x0+t3) >> 17);
2204 o[7] = stbi__clamp((x0-t3) >> 17);
2205 o[1] = stbi__clamp((x1+t2) >> 17);
2206 o[6] = stbi__clamp((x1-t2) >> 17);
2207 o[2] = stbi__clamp((x2+t1) >> 17);
2208 o[5] = stbi__clamp((x2-t1) >> 17);
2209 o[3] = stbi__clamp((x3+t0) >> 17);
2210 o[4] = stbi__clamp((x3-t0) >> 17);
2211 }
2212}
2213
2214#ifdef STBI_SSE2
2215// sse2 integer IDCT. not the fastest possible implementation but it
2216// produces bit-identical results to the generic C version so it's
2217// fully "transparent".
2218static void stbi__idct_simd(stbi_uc *out, int out_stride, short data[64])
2219{
2220 // This is constructed to match our regular (generic) integer IDCT exactly.
2221 __m128i row0, row1, row2, row3, row4, row5, row6, row7;
2222 __m128i tmp;
2223
2224 // dot product constant: even elems=x, odd elems=y
2225 #define dct_const(x,y) _mm_setr_epi16((x),(y),(x),(y),(x),(y),(x),(y))
2226
2227 // out(0) = c0[even]*x + c0[odd]*y (c0, x, y 16-bit, out 32-bit)
2228 // out(1) = c1[even]*x + c1[odd]*y
2229 #define dct_rot(out0,out1, x,y,c0,c1) \
2230 __m128i c0##lo = _mm_unpacklo_epi16((x),(y)); \
2231 __m128i c0##hi = _mm_unpackhi_epi16((x),(y)); \
2232 __m128i out0##_l = _mm_madd_epi16(c0##lo, c0); \
2233 __m128i out0##_h = _mm_madd_epi16(c0##hi, c0); \
2234 __m128i out1##_l = _mm_madd_epi16(c0##lo, c1); \
2235 __m128i out1##_h = _mm_madd_epi16(c0##hi, c1)
2236
2237 // out = in << 12 (in 16-bit, out 32-bit)
2238 #define dct_widen(out, in) \
2239 __m128i out##_l = _mm_srai_epi32(_mm_unpacklo_epi16(_mm_setzero_si128(), (in)), 4); \
2240 __m128i out##_h = _mm_srai_epi32(_mm_unpackhi_epi16(_mm_setzero_si128(), (in)), 4)
2241
2242 // wide add
2243 #define dct_wadd(out, a, b) \
2244 __m128i out##_l = _mm_add_epi32(a##_l, b##_l); \
2245 __m128i out##_h = _mm_add_epi32(a##_h, b##_h)
2246
2247 // wide sub
2248 #define dct_wsub(out, a, b) \
2249 __m128i out##_l = _mm_sub_epi32(a##_l, b##_l); \
2250 __m128i out##_h = _mm_sub_epi32(a##_h, b##_h)
2251
2252 // butterfly a/b, add bias, then shift by "s" and pack
2253 #define dct_bfly32o(out0, out1, a,b,bias,s) \
2254 { \
2255 __m128i abiased_l = _mm_add_epi32(a##_l, bias); \
2256 __m128i abiased_h = _mm_add_epi32(a##_h, bias); \
2257 dct_wadd(sum, abiased, b); \
2258 dct_wsub(dif, abiased, b); \
2259 out0 = _mm_packs_epi32(_mm_srai_epi32(sum_l, s), _mm_srai_epi32(sum_h, s)); \
2260 out1 = _mm_packs_epi32(_mm_srai_epi32(dif_l, s), _mm_srai_epi32(dif_h, s)); \
2261 }
2262
2263 // 8-bit interleave step (for transposes)
2264 #define dct_interleave8(a, b) \
2265 tmp = a; \
2266 a = _mm_unpacklo_epi8(a, b); \
2267 b = _mm_unpackhi_epi8(tmp, b)
2268
2269 // 16-bit interleave step (for transposes)
2270 #define dct_interleave16(a, b) \
2271 tmp = a; \
2272 a = _mm_unpacklo_epi16(a, b); \
2273 b = _mm_unpackhi_epi16(tmp, b)
2274
2275 #define dct_pass(bias,shift) \
2276 { \
2277 /* even part */ \
2278 dct_rot(t2e,t3e, row2,row6, rot0_0,rot0_1); \
2279 __m128i sum04 = _mm_add_epi16(row0, row4); \
2280 __m128i dif04 = _mm_sub_epi16(row0, row4); \
2281 dct_widen(t0e, sum04); \
2282 dct_widen(t1e, dif04); \
2283 dct_wadd(x0, t0e, t3e); \
2284 dct_wsub(x3, t0e, t3e); \
2285 dct_wadd(x1, t1e, t2e); \
2286 dct_wsub(x2, t1e, t2e); \
2287 /* odd part */ \
2288 dct_rot(y0o,y2o, row7,row3, rot2_0,rot2_1); \
2289 dct_rot(y1o,y3o, row5,row1, rot3_0,rot3_1); \
2290 __m128i sum17 = _mm_add_epi16(row1, row7); \
2291 __m128i sum35 = _mm_add_epi16(row3, row5); \
2292 dct_rot(y4o,y5o, sum17,sum35, rot1_0,rot1_1); \
2293 dct_wadd(x4, y0o, y4o); \
2294 dct_wadd(x5, y1o, y5o); \
2295 dct_wadd(x6, y2o, y5o); \
2296 dct_wadd(x7, y3o, y4o); \
2297 dct_bfly32o(row0,row7, x0,x7,bias,shift); \
2298 dct_bfly32o(row1,row6, x1,x6,bias,shift); \
2299 dct_bfly32o(row2,row5, x2,x5,bias,shift); \
2300 dct_bfly32o(row3,row4, x3,x4,bias,shift); \
2301 }
2302
2303 __m128i rot0_0 = dct_const(stbi__f2f(0.5411961f), stbi__f2f(0.5411961f) + stbi__f2f(-1.847759065f));
2304 __m128i rot0_1 = dct_const(stbi__f2f(0.5411961f) + stbi__f2f( 0.765366865f), stbi__f2f(0.5411961f));
2305 __m128i rot1_0 = dct_const(stbi__f2f(1.175875602f) + stbi__f2f(-0.899976223f), stbi__f2f(1.175875602f));
2306 __m128i rot1_1 = dct_const(stbi__f2f(1.175875602f), stbi__f2f(1.175875602f) + stbi__f2f(-2.562915447f));
2307 __m128i rot2_0 = dct_const(stbi__f2f(-1.961570560f) + stbi__f2f( 0.298631336f), stbi__f2f(-1.961570560f));
2308 __m128i rot2_1 = dct_const(stbi__f2f(-1.961570560f), stbi__f2f(-1.961570560f) + stbi__f2f( 3.072711026f));
2309 __m128i rot3_0 = dct_const(stbi__f2f(-0.390180644f) + stbi__f2f( 2.053119869f), stbi__f2f(-0.390180644f));
2310 __m128i rot3_1 = dct_const(stbi__f2f(-0.390180644f), stbi__f2f(-0.390180644f) + stbi__f2f( 1.501321110f));
2311
2312 // rounding biases in column/row passes, see stbi__idct_block for explanation.
2313 __m128i bias_0 = _mm_set1_epi32(512);
2314 __m128i bias_1 = _mm_set1_epi32(65536 + (128<<17));
2315
2316 // load
2317 row0 = _mm_load_si128((const __m128i *) (data + 0*8));
2318 row1 = _mm_load_si128((const __m128i *) (data + 1*8));
2319 row2 = _mm_load_si128((const __m128i *) (data + 2*8));
2320 row3 = _mm_load_si128((const __m128i *) (data + 3*8));
2321 row4 = _mm_load_si128((const __m128i *) (data + 4*8));
2322 row5 = _mm_load_si128((const __m128i *) (data + 5*8));
2323 row6 = _mm_load_si128((const __m128i *) (data + 6*8));
2324 row7 = _mm_load_si128((const __m128i *) (data + 7*8));
2325
2326 // column pass
2327 dct_pass(bias_0, 10);
2328
2329 {
2330 // 16bit 8x8 transpose pass 1
2331 dct_interleave16(row0, row4);
2332 dct_interleave16(row1, row5);
2333 dct_interleave16(row2, row6);
2334 dct_interleave16(row3, row7);
2335
2336 // transpose pass 2
2337 dct_interleave16(row0, row2);
2338 dct_interleave16(row1, row3);
2339 dct_interleave16(row4, row6);
2340 dct_interleave16(row5, row7);
2341
2342 // transpose pass 3
2343 dct_interleave16(row0, row1);
2344 dct_interleave16(row2, row3);
2345 dct_interleave16(row4, row5);
2346 dct_interleave16(row6, row7);
2347 }
2348
2349 // row pass
2350 dct_pass(bias_1, 17);
2351
2352 {
2353 // pack
2354 __m128i p0 = _mm_packus_epi16(row0, row1); // a0a1a2a3...a7b0b1b2b3...b7
2355 __m128i p1 = _mm_packus_epi16(row2, row3);
2356 __m128i p2 = _mm_packus_epi16(row4, row5);
2357 __m128i p3 = _mm_packus_epi16(row6, row7);
2358
2359 // 8bit 8x8 transpose pass 1
2360 dct_interleave8(p0, p2); // a0e0a1e1...
2361 dct_interleave8(p1, p3); // c0g0c1g1...
2362
2363 // transpose pass 2
2364 dct_interleave8(p0, p1); // a0c0e0g0...
2365 dct_interleave8(p2, p3); // b0d0f0h0...
2366
2367 // transpose pass 3
2368 dct_interleave8(p0, p2); // a0b0c0d0...
2369 dct_interleave8(p1, p3); // a4b4c4d4...
2370
2371 // store
2372 _mm_storel_epi64((__m128i *) out, p0); out += out_stride;
2373 _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p0, 0x4e)); out += out_stride;
2374 _mm_storel_epi64((__m128i *) out, p2); out += out_stride;
2375 _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p2, 0x4e)); out += out_stride;
2376 _mm_storel_epi64((__m128i *) out, p1); out += out_stride;
2377 _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p1, 0x4e)); out += out_stride;
2378 _mm_storel_epi64((__m128i *) out, p3); out += out_stride;
2379 _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p3, 0x4e));
2380 }
2381
2382#undef dct_const
2383#undef dct_rot
2384#undef dct_widen
2385#undef dct_wadd
2386#undef dct_wsub
2387#undef dct_bfly32o
2388#undef dct_interleave8
2389#undef dct_interleave16
2390#undef dct_pass
2391}
2392
2393#endif // STBI_SSE2
2394
2395#ifdef STBI_NEON
2396
2397// NEON integer IDCT. should produce bit-identical
2398// results to the generic C version.
2399static void stbi__idct_simd(stbi_uc *out, int out_stride, short data[64])
2400{
2401 int16x8_t row0, row1, row2, row3, row4, row5, row6, row7;
2402
2403 int16x4_t rot0_0 = vdup_n_s16(stbi__f2f(0.5411961f));
2404 int16x4_t rot0_1 = vdup_n_s16(stbi__f2f(-1.847759065f));
2405 int16x4_t rot0_2 = vdup_n_s16(stbi__f2f( 0.765366865f));
2406 int16x4_t rot1_0 = vdup_n_s16(stbi__f2f( 1.175875602f));
2407 int16x4_t rot1_1 = vdup_n_s16(stbi__f2f(-0.899976223f));
2408 int16x4_t rot1_2 = vdup_n_s16(stbi__f2f(-2.562915447f));
2409 int16x4_t rot2_0 = vdup_n_s16(stbi__f2f(-1.961570560f));
2410 int16x4_t rot2_1 = vdup_n_s16(stbi__f2f(-0.390180644f));
2411 int16x4_t rot3_0 = vdup_n_s16(stbi__f2f( 0.298631336f));
2412 int16x4_t rot3_1 = vdup_n_s16(stbi__f2f( 2.053119869f));
2413 int16x4_t rot3_2 = vdup_n_s16(stbi__f2f( 3.072711026f));
2414 int16x4_t rot3_3 = vdup_n_s16(stbi__f2f( 1.501321110f));
2415
2416#define dct_long_mul(out, inq, coeff) \
2417 int32x4_t out##_l = vmull_s16(vget_low_s16(inq), coeff); \
2418 int32x4_t out##_h = vmull_s16(vget_high_s16(inq), coeff)
2419
2420#define dct_long_mac(out, acc, inq, coeff) \
2421 int32x4_t out##_l = vmlal_s16(acc##_l, vget_low_s16(inq), coeff); \
2422 int32x4_t out##_h = vmlal_s16(acc##_h, vget_high_s16(inq), coeff)
2423
2424#define dct_widen(out, inq) \
2425 int32x4_t out##_l = vshll_n_s16(vget_low_s16(inq), 12); \
2426 int32x4_t out##_h = vshll_n_s16(vget_high_s16(inq), 12)
2427
2428// wide add
2429#define dct_wadd(out, a, b) \
2430 int32x4_t out##_l = vaddq_s32(a##_l, b##_l); \
2431 int32x4_t out##_h = vaddq_s32(a##_h, b##_h)
2432
2433// wide sub
2434#define dct_wsub(out, a, b) \
2435 int32x4_t out##_l = vsubq_s32(a##_l, b##_l); \
2436 int32x4_t out##_h = vsubq_s32(a##_h, b##_h)
2437
2438// butterfly a/b, then shift using "shiftop" by "s" and pack
2439#define dct_bfly32o(out0,out1, a,b,shiftop,s) \
2440 { \
2441 dct_wadd(sum, a, b); \
2442 dct_wsub(dif, a, b); \
2443 out0 = vcombine_s16(shiftop(sum_l, s), shiftop(sum_h, s)); \
2444 out1 = vcombine_s16(shiftop(dif_l, s), shiftop(dif_h, s)); \
2445 }
2446
2447#define dct_pass(shiftop, shift) \
2448 { \
2449 /* even part */ \
2450 int16x8_t sum26 = vaddq_s16(row2, row6); \
2451 dct_long_mul(p1e, sum26, rot0_0); \
2452 dct_long_mac(t2e, p1e, row6, rot0_1); \
2453 dct_long_mac(t3e, p1e, row2, rot0_2); \
2454 int16x8_t sum04 = vaddq_s16(row0, row4); \
2455 int16x8_t dif04 = vsubq_s16(row0, row4); \
2456 dct_widen(t0e, sum04); \
2457 dct_widen(t1e, dif04); \
2458 dct_wadd(x0, t0e, t3e); \
2459 dct_wsub(x3, t0e, t3e); \
2460 dct_wadd(x1, t1e, t2e); \
2461 dct_wsub(x2, t1e, t2e); \
2462 /* odd part */ \
2463 int16x8_t sum15 = vaddq_s16(row1, row5); \
2464 int16x8_t sum17 = vaddq_s16(row1, row7); \
2465 int16x8_t sum35 = vaddq_s16(row3, row5); \
2466 int16x8_t sum37 = vaddq_s16(row3, row7); \
2467 int16x8_t sumodd = vaddq_s16(sum17, sum35); \
2468 dct_long_mul(p5o, sumodd, rot1_0); \
2469 dct_long_mac(p1o, p5o, sum17, rot1_1); \
2470 dct_long_mac(p2o, p5o, sum35, rot1_2); \
2471 dct_long_mul(p3o, sum37, rot2_0); \
2472 dct_long_mul(p4o, sum15, rot2_1); \
2473 dct_wadd(sump13o, p1o, p3o); \
2474 dct_wadd(sump24o, p2o, p4o); \
2475 dct_wadd(sump23o, p2o, p3o); \
2476 dct_wadd(sump14o, p1o, p4o); \
2477 dct_long_mac(x4, sump13o, row7, rot3_0); \
2478 dct_long_mac(x5, sump24o, row5, rot3_1); \
2479 dct_long_mac(x6, sump23o, row3, rot3_2); \
2480 dct_long_mac(x7, sump14o, row1, rot3_3); \
2481 dct_bfly32o(row0,row7, x0,x7,shiftop,shift); \
2482 dct_bfly32o(row1,row6, x1,x6,shiftop,shift); \
2483 dct_bfly32o(row2,row5, x2,x5,shiftop,shift); \
2484 dct_bfly32o(row3,row4, x3,x4,shiftop,shift); \
2485 }
2486
2487 // load
2488 row0 = vld1q_s16(data + 0*8);
2489 row1 = vld1q_s16(data + 1*8);
2490 row2 = vld1q_s16(data + 2*8);
2491 row3 = vld1q_s16(data + 3*8);
2492 row4 = vld1q_s16(data + 4*8);
2493 row5 = vld1q_s16(data + 5*8);
2494 row6 = vld1q_s16(data + 6*8);
2495 row7 = vld1q_s16(data + 7*8);
2496
2497 // add DC bias
2498 row0 = vaddq_s16(row0, vsetq_lane_s16(1024, vdupq_n_s16(0), 0));
2499
2500 // column pass
2501 dct_pass(vrshrn_n_s32, 10);
2502
2503 // 16bit 8x8 transpose
2504 {
2505// these three map to a single VTRN.16, VTRN.32, and VSWP, respectively.
2506// whether compilers actually get this is another story, sadly.
2507#define dct_trn16(x, y) { int16x8x2_t t = vtrnq_s16(x, y); x = t.val[0]; y = t.val[1]; }
2508#define dct_trn32(x, y) { int32x4x2_t t = vtrnq_s32(vreinterpretq_s32_s16(x), vreinterpretq_s32_s16(y)); x = vreinterpretq_s16_s32(t.val[0]); y = vreinterpretq_s16_s32(t.val[1]); }
2509#define dct_trn64(x, y) { int16x8_t x0 = x; int16x8_t y0 = y; x = vcombine_s16(vget_low_s16(x0), vget_low_s16(y0)); y = vcombine_s16(vget_high_s16(x0), vget_high_s16(y0)); }
2510
2511 // pass 1
2512 dct_trn16(row0, row1); // a0b0a2b2a4b4a6b6
2513 dct_trn16(row2, row3);
2514 dct_trn16(row4, row5);
2515 dct_trn16(row6, row7);
2516
2517 // pass 2
2518 dct_trn32(row0, row2); // a0b0c0d0a4b4c4d4
2519 dct_trn32(row1, row3);
2520 dct_trn32(row4, row6);
2521 dct_trn32(row5, row7);
2522
2523 // pass 3
2524 dct_trn64(row0, row4); // a0b0c0d0e0f0g0h0
2525 dct_trn64(row1, row5);
2526 dct_trn64(row2, row6);
2527 dct_trn64(row3, row7);
2528
2529#undef dct_trn16
2530#undef dct_trn32
2531#undef dct_trn64
2532 }
2533
2534 // row pass
2535 // vrshrn_n_s32 only supports shifts up to 16, we need
2536 // 17. so do a non-rounding shift of 16 first then follow
2537 // up with a rounding shift by 1.
2538 dct_pass(vshrn_n_s32, 16);
2539
2540 {
2541 // pack and round
2542 uint8x8_t p0 = vqrshrun_n_s16(row0, 1);
2543 uint8x8_t p1 = vqrshrun_n_s16(row1, 1);
2544 uint8x8_t p2 = vqrshrun_n_s16(row2, 1);
2545 uint8x8_t p3 = vqrshrun_n_s16(row3, 1);
2546 uint8x8_t p4 = vqrshrun_n_s16(row4, 1);
2547 uint8x8_t p5 = vqrshrun_n_s16(row5, 1);
2548 uint8x8_t p6 = vqrshrun_n_s16(row6, 1);
2549 uint8x8_t p7 = vqrshrun_n_s16(row7, 1);
2550
2551 // again, these can translate into one instruction, but often don't.
2552#define dct_trn8_8(x, y) { uint8x8x2_t t = vtrn_u8(x, y); x = t.val[0]; y = t.val[1]; }
2553#define dct_trn8_16(x, y) { uint16x4x2_t t = vtrn_u16(vreinterpret_u16_u8(x), vreinterpret_u16_u8(y)); x = vreinterpret_u8_u16(t.val[0]); y = vreinterpret_u8_u16(t.val[1]); }
2554#define dct_trn8_32(x, y) { uint32x2x2_t t = vtrn_u32(vreinterpret_u32_u8(x), vreinterpret_u32_u8(y)); x = vreinterpret_u8_u32(t.val[0]); y = vreinterpret_u8_u32(t.val[1]); }
2555
2556 // sadly can't use interleaved stores here since we only write
2557 // 8 bytes to each scan line!
2558
2559 // 8x8 8-bit transpose pass 1
2560 dct_trn8_8(p0, p1);
2561 dct_trn8_8(p2, p3);
2562 dct_trn8_8(p4, p5);
2563 dct_trn8_8(p6, p7);
2564
2565 // pass 2
2566 dct_trn8_16(p0, p2);
2567 dct_trn8_16(p1, p3);
2568 dct_trn8_16(p4, p6);
2569 dct_trn8_16(p5, p7);
2570
2571 // pass 3
2572 dct_trn8_32(p0, p4);
2573 dct_trn8_32(p1, p5);
2574 dct_trn8_32(p2, p6);
2575 dct_trn8_32(p3, p7);
2576
2577 // store
2578 vst1_u8(out, p0); out += out_stride;
2579 vst1_u8(out, p1); out += out_stride;
2580 vst1_u8(out, p2); out += out_stride;
2581 vst1_u8(out, p3); out += out_stride;
2582 vst1_u8(out, p4); out += out_stride;
2583 vst1_u8(out, p5); out += out_stride;
2584 vst1_u8(out, p6); out += out_stride;
2585 vst1_u8(out, p7);
2586
2587#undef dct_trn8_8
2588#undef dct_trn8_16
2589#undef dct_trn8_32
2590 }
2591
2592#undef dct_long_mul
2593#undef dct_long_mac
2594#undef dct_widen
2595#undef dct_wadd
2596#undef dct_wsub
2597#undef dct_bfly32o
2598#undef dct_pass
2599}
2600
2601#endif // STBI_NEON
2602
2603#define STBI__MARKER_none 0xff
2604// if there's a pending marker from the entropy stream, return that
2605// otherwise, fetch from the stream and get a marker. if there's no
2606// marker, return 0xff, which is never a valid marker value
2607static stbi_uc stbi__get_marker(stbi__jpeg *j)
2608{
2609 stbi_uc x;
2610 if (j->marker != STBI__MARKER_none) { x = j->marker; j->marker = STBI__MARKER_none; return x; }
2611 x = stbi__get8(j->s);
2612 if (x != 0xff) return STBI__MARKER_none;
2613 while (x == 0xff)
2614 x = stbi__get8(j->s); // consume repeated 0xff fill bytes
2615 return x;
2616}
2617
2618// in each scan, we'll have scan_n components, and the order
2619// of the components is specified by order[]
2620#define STBI__RESTART(x) ((x) >= 0xd0 && (x) <= 0xd7)
2621
2622// after a restart interval, stbi__jpeg_reset the entropy decoder and
2623// the dc prediction
2624static void stbi__jpeg_reset(stbi__jpeg *j)
2625{
2626 j->code_bits = 0;
2627 j->code_buffer = 0;
2628 j->nomore = 0;
2629 j->img_comp[0].dc_pred = j->img_comp[1].dc_pred = j->img_comp[2].dc_pred = j->img_comp[3].dc_pred = 0;
2630 j->marker = STBI__MARKER_none;
2631 j->todo = j->restart_interval ? j->restart_interval : 0x7fffffff;
2632 j->eob_run = 0;
2633 // no more than 1<<31 MCUs if no restart_interal? that's plenty safe,
2634 // since we don't even allow 1<<30 pixels
2635}
2636
2637static int stbi__parse_entropy_coded_data(stbi__jpeg *z)
2638{
2639 stbi__jpeg_reset(z);
2640 if (!z->progressive) {
2641 if (z->scan_n == 1) {
2642 int i,j;
2643 STBI_SIMD_ALIGN(short, data[64]);
2644 int n = z->order[0];
2645 // non-interleaved data, we just need to process one block at a time,
2646 // in trivial scanline order
2647 // number of blocks to do just depends on how many actual "pixels" this
2648 // component has, independent of interleaved MCU blocking and such
2649 int w = (z->img_comp[n].x+7) >> 3;
2650 int h = (z->img_comp[n].y+7) >> 3;
2651 for (j=0; j < h; ++j) {
2652 for (i=0; i < w; ++i) {
2653 int ha = z->img_comp[n].ha;
2654 if (!stbi__jpeg_decode_block(z, data, z->huff_dc+z->img_comp[n].hd, z->huff_ac+ha, z->fast_ac[ha], n, z->dequant[z->img_comp[n].tq])) return 0;
2655 z->idct_block_kernel(z->img_comp[n].data+z->img_comp[n].w2*j*8+i*8, z->img_comp[n].w2, data);
2656 // every data block is an MCU, so countdown the restart interval
2657 if (--z->todo <= 0) {
2658 if (z->code_bits < 24) stbi__grow_buffer_unsafe(z);
2659 // if it's NOT a restart, then just bail, so we get corrupt data
2660 // rather than no data
2661 if (!STBI__RESTART(z->marker)) return 1;
2662 stbi__jpeg_reset(z);
2663 }
2664 }
2665 }
2666 return 1;
2667 } else { // interleaved
2668 int i,j,k,x,y;
2669 STBI_SIMD_ALIGN(short, data[64]);
2670 for (j=0; j < z->img_mcu_y; ++j) {
2671 for (i=0; i < z->img_mcu_x; ++i) {
2672 // scan an interleaved mcu... process scan_n components in order
2673 for (k=0; k < z->scan_n; ++k) {
2674 int n = z->order[k];
2675 // scan out an mcu's worth of this component; that's just determined
2676 // by the basic H and V specified for the component
2677 for (y=0; y < z->img_comp[n].v; ++y) {
2678 for (x=0; x < z->img_comp[n].h; ++x) {
2679 int x2 = (i*z->img_comp[n].h + x)*8;
2680 int y2 = (j*z->img_comp[n].v + y)*8;
2681 int ha = z->img_comp[n].ha;
2682 if (!stbi__jpeg_decode_block(z, data, z->huff_dc+z->img_comp[n].hd, z->huff_ac+ha, z->fast_ac[ha], n, z->dequant[z->img_comp[n].tq])) return 0;
2683 z->idct_block_kernel(z->img_comp[n].data+z->img_comp[n].w2*y2+x2, z->img_comp[n].w2, data);
2684 }
2685 }
2686 }
2687 // after all interleaved components, that's an interleaved MCU,
2688 // so now count down the restart interval
2689 if (--z->todo <= 0) {
2690 if (z->code_bits < 24) stbi__grow_buffer_unsafe(z);
2691 if (!STBI__RESTART(z->marker)) return 1;
2692 stbi__jpeg_reset(z);
2693 }
2694 }
2695 }
2696 return 1;
2697 }
2698 } else {
2699 if (z->scan_n == 1) {
2700 int i,j;
2701 int n = z->order[0];
2702 // non-interleaved data, we just need to process one block at a time,
2703 // in trivial scanline order
2704 // number of blocks to do just depends on how many actual "pixels" this
2705 // component has, independent of interleaved MCU blocking and such
2706 int w = (z->img_comp[n].x+7) >> 3;
2707 int h = (z->img_comp[n].y+7) >> 3;
2708 for (j=0; j < h; ++j) {
2709 for (i=0; i < w; ++i) {
2710 short *data = z->img_comp[n].coeff + 64 * (i + j * z->img_comp[n].coeff_w);
2711 if (z->spec_start == 0) {
2712 if (!stbi__jpeg_decode_block_prog_dc(z, data, &z->huff_dc[z->img_comp[n].hd], n))
2713 return 0;
2714 } else {
2715 int ha = z->img_comp[n].ha;
2716 if (!stbi__jpeg_decode_block_prog_ac(z, data, &z->huff_ac[ha], z->fast_ac[ha]))
2717 return 0;
2718 }
2719 // every data block is an MCU, so countdown the restart interval
2720 if (--z->todo <= 0) {
2721 if (z->code_bits < 24) stbi__grow_buffer_unsafe(z);
2722 if (!STBI__RESTART(z->marker)) return 1;
2723 stbi__jpeg_reset(z);
2724 }
2725 }
2726 }
2727 return 1;
2728 } else { // interleaved
2729 int i,j,k,x,y;
2730 for (j=0; j < z->img_mcu_y; ++j) {
2731 for (i=0; i < z->img_mcu_x; ++i) {
2732 // scan an interleaved mcu... process scan_n components in order
2733 for (k=0; k < z->scan_n; ++k) {
2734 int n = z->order[k];
2735 // scan out an mcu's worth of this component; that's just determined
2736 // by the basic H and V specified for the component
2737 for (y=0; y < z->img_comp[n].v; ++y) {
2738 for (x=0; x < z->img_comp[n].h; ++x) {
2739 int x2 = (i*z->img_comp[n].h + x);
2740 int y2 = (j*z->img_comp[n].v + y);
2741 short *data = z->img_comp[n].coeff + 64 * (x2 + y2 * z->img_comp[n].coeff_w);
2742 if (!stbi__jpeg_decode_block_prog_dc(z, data, &z->huff_dc[z->img_comp[n].hd], n))
2743 return 0;
2744 }
2745 }
2746 }
2747 // after all interleaved components, that's an interleaved MCU,
2748 // so now count down the restart interval
2749 if (--z->todo <= 0) {
2750 if (z->code_bits < 24) stbi__grow_buffer_unsafe(z);
2751 if (!STBI__RESTART(z->marker)) return 1;
2752 stbi__jpeg_reset(z);
2753 }
2754 }
2755 }
2756 return 1;
2757 }
2758 }
2759}
2760
2761static void stbi__jpeg_dequantize(short *data, stbi__uint16 *dequant)
2762{
2763 int i;
2764 for (i=0; i < 64; ++i)
2765 data[i] *= dequant[i];
2766}
2767
2768static void stbi__jpeg_finish(stbi__jpeg *z)
2769{
2770 if (z->progressive) {
2771 // dequantize and idct the data
2772 int i,j,n;
2773 for (n=0; n < z->s->img_n; ++n) {
2774 int w = (z->img_comp[n].x+7) >> 3;
2775 int h = (z->img_comp[n].y+7) >> 3;
2776 for (j=0; j < h; ++j) {
2777 for (i=0; i < w; ++i) {
2778 short *data = z->img_comp[n].coeff + 64 * (i + j * z->img_comp[n].coeff_w);
2779 stbi__jpeg_dequantize(data, z->dequant[z->img_comp[n].tq]);
2780 z->idct_block_kernel(z->img_comp[n].data+z->img_comp[n].w2*j*8+i*8, z->img_comp[n].w2, data);
2781 }
2782 }
2783 }
2784 }
2785}
2786
2787static int stbi__process_marker(stbi__jpeg *z, int m)
2788{
2789 int L;
2790 switch (m) {
2791 case STBI__MARKER_none: // no marker found
2792 return stbi__err("expected marker","Corrupt JPEG");
2793
2794 case 0xDD: // DRI - specify restart interval
2795 if (stbi__get16be(z->s) != 4) return stbi__err("bad DRI len","Corrupt JPEG");
2796 z->restart_interval = stbi__get16be(z->s);
2797 return 1;
2798
2799 case 0xDB: // DQT - define quantization table
2800 L = stbi__get16be(z->s)-2;
2801 while (L > 0) {
2802 int q = stbi__get8(z->s);
2803 int p = q >> 4, sixteen = (p != 0);
2804 int t = q & 15,i;
2805 if (p != 0 && p != 1) return stbi__err("bad DQT type","Corrupt JPEG");
2806 if (t > 3) return stbi__err("bad DQT table","Corrupt JPEG");
2807
2808 for (i=0; i < 64; ++i)
2809 z->dequant[t][stbi__jpeg_dezigzag[i]] = sixteen ? stbi__get16be(z->s) : stbi__get8(z->s);
2810 L -= (sixteen ? 129 : 65);
2811 }
2812 return L==0;
2813
2814 case 0xC4: // DHT - define huffman table
2815 L = stbi__get16be(z->s)-2;
2816 while (L > 0) {
2817 stbi_uc *v;
2818 int sizes[16],i,n=0;
2819 int q = stbi__get8(z->s);
2820 int tc = q >> 4;
2821 int th = q & 15;
2822 if (tc > 1 || th > 3) return stbi__err("bad DHT header","Corrupt JPEG");
2823 for (i=0; i < 16; ++i) {
2824 sizes[i] = stbi__get8(z->s);
2825 n += sizes[i];
2826 }
2827 L -= 17;
2828 if (tc == 0) {
2829 if (!stbi__build_huffman(z->huff_dc+th, sizes)) return 0;
2830 v = z->huff_dc[th].values;
2831 } else {
2832 if (!stbi__build_huffman(z->huff_ac+th, sizes)) return 0;
2833 v = z->huff_ac[th].values;
2834 }
2835 for (i=0; i < n; ++i)
2836 v[i] = stbi__get8(z->s);
2837 if (tc != 0)
2838 stbi__build_fast_ac(z->fast_ac[th], z->huff_ac + th);
2839 L -= n;
2840 }
2841 return L==0;
2842 }
2843
2844 // check for comment block or APP blocks
2845 if ((m >= 0xE0 && m <= 0xEF) || m == 0xFE) {
2846 L = stbi__get16be(z->s);
2847 if (L < 2) {
2848 if (m == 0xFE)
2849 return stbi__err("bad COM len","Corrupt JPEG");
2850 else
2851 return stbi__err("bad APP len","Corrupt JPEG");
2852 }
2853 L -= 2;
2854
2855 if (m == 0xE0 && L >= 5) { // JFIF APP0 segment
2856 static const unsigned char tag[5] = {'J','F','I','F','\0'};
2857 int ok = 1;
2858 int i;
2859 for (i=0; i < 5; ++i)
2860 if (stbi__get8(z->s) != tag[i])
2861 ok = 0;
2862 L -= 5;
2863 if (ok)
2864 z->jfif = 1;
2865 } else if (m == 0xEE && L >= 12) { // Adobe APP14 segment
2866 static const unsigned char tag[6] = {'A','d','o','b','e','\0'};
2867 int ok = 1;
2868 int i;
2869 for (i=0; i < 6; ++i)
2870 if (stbi__get8(z->s) != tag[i])
2871 ok = 0;
2872 L -= 6;
2873 if (ok) {
2874 stbi__get8(z->s); // version
2875 stbi__get16be(z->s); // flags0
2876 stbi__get16be(z->s); // flags1
2877 z->app14_color_transform = stbi__get8(z->s); // color transform
2878 L -= 6;
2879 }
2880 }
2881
2882 stbi__skip(z->s, L);
2883 return 1;
2884 }
2885
2886 return stbi__err("unknown marker","Corrupt JPEG");
2887}
2888
2889// after we see SOS
2890static int stbi__process_scan_header(stbi__jpeg *z)
2891{
2892 int i;
2893 int Ls = stbi__get16be(z->s);
2894 z->scan_n = stbi__get8(z->s);
2895 if (z->scan_n < 1 || z->scan_n > 4 || z->scan_n > (int) z->s->img_n) return stbi__err("bad SOS component count","Corrupt JPEG");
2896 if (Ls != 6+2*z->scan_n) return stbi__err("bad SOS len","Corrupt JPEG");
2897 for (i=0; i < z->scan_n; ++i) {
2898 int id = stbi__get8(z->s), which;
2899 int q = stbi__get8(z->s);
2900 for (which = 0; which < z->s->img_n; ++which)
2901 if (z->img_comp[which].id == id)
2902 break;
2903 if (which == z->s->img_n) return 0; // no match
2904 z->img_comp[which].hd = q >> 4; if (z->img_comp[which].hd > 3) return stbi__err("bad DC huff","Corrupt JPEG");
2905 z->img_comp[which].ha = q & 15; if (z->img_comp[which].ha > 3) return stbi__err("bad AC huff","Corrupt JPEG");
2906 z->order[i] = which;
2907 }
2908
2909 {
2910 int aa;
2911 z->spec_start = stbi__get8(z->s);
2912 z->spec_end = stbi__get8(z->s); // should be 63, but might be 0
2913 aa = stbi__get8(z->s);
2914 z->succ_high = (aa >> 4);
2915 z->succ_low = (aa & 15);
2916 if (z->progressive) {
2917 if (z->spec_start > 63 || z->spec_end > 63 || z->spec_start > z->spec_end || z->succ_high > 13 || z->succ_low > 13)
2918 return stbi__err("bad SOS", "Corrupt JPEG");
2919 } else {
2920 if (z->spec_start != 0) return stbi__err("bad SOS","Corrupt JPEG");
2921 if (z->succ_high != 0 || z->succ_low != 0) return stbi__err("bad SOS","Corrupt JPEG");
2922 z->spec_end = 63;
2923 }
2924 }
2925
2926 return 1;
2927}
2928
2929static int stbi__free_jpeg_components(stbi__jpeg *z, int ncomp, int why)
2930{
2931 int i;
2932 for (i=0; i < ncomp; ++i) {
2933 if (z->img_comp[i].raw_data) {
2934 STBI_FREE(z->img_comp[i].raw_data);
2935 z->img_comp[i].raw_data = NULL;
2936 z->img_comp[i].data = NULL;
2937 }
2938 if (z->img_comp[i].raw_coeff) {
2939 STBI_FREE(z->img_comp[i].raw_coeff);
2940 z->img_comp[i].raw_coeff = 0;
2941 z->img_comp[i].coeff = 0;
2942 }
2943 if (z->img_comp[i].linebuf) {
2944 STBI_FREE(z->img_comp[i].linebuf);
2945 z->img_comp[i].linebuf = NULL;
2946 }
2947 }
2948 return why;
2949}
2950
2951static int stbi__process_frame_header(stbi__jpeg *z, int scan)
2952{
2953 stbi__context *s = z->s;
2954 int Lf,p,i,q, h_max=1,v_max=1,c;
2955 Lf = stbi__get16be(s); if (Lf < 11) return stbi__err("bad SOF len","Corrupt JPEG"); // JPEG
2956 p = stbi__get8(s); if (p != 8) return stbi__err("only 8-bit","JPEG format not supported: 8-bit only"); // JPEG baseline
2957 s->img_y = stbi__get16be(s); if (s->img_y == 0) return stbi__err("no header height", "JPEG format not supported: delayed height"); // Legal, but we don't handle it--but neither does IJG
2958 s->img_x = stbi__get16be(s); if (s->img_x == 0) return stbi__err("0 width","Corrupt JPEG"); // JPEG requires
2959 c = stbi__get8(s);
2960 if (c != 3 && c != 1 && c != 4) return stbi__err("bad component count","Corrupt JPEG");
2961 s->img_n = c;
2962 for (i=0; i < c; ++i) {
2963 z->img_comp[i].data = NULL;
2964 z->img_comp[i].linebuf = NULL;
2965 }
2966
2967 if (Lf != 8+3*s->img_n) return stbi__err("bad SOF len","Corrupt JPEG");
2968
2969 z->rgb = 0;
2970 for (i=0; i < s->img_n; ++i) {
2971 static unsigned char rgb[3] = { 'R', 'G', 'B' };
2972 z->img_comp[i].id = stbi__get8(s);
2973 if (s->img_n == 3 && z->img_comp[i].id == rgb[i])
2974 ++z->rgb;
2975 q = stbi__get8(s);
2976 z->img_comp[i].h = (q >> 4); if (!z->img_comp[i].h || z->img_comp[i].h > 4) return stbi__err("bad H","Corrupt JPEG");
2977 z->img_comp[i].v = q & 15; if (!z->img_comp[i].v || z->img_comp[i].v > 4) return stbi__err("bad V","Corrupt JPEG");
2978 z->img_comp[i].tq = stbi__get8(s); if (z->img_comp[i].tq > 3) return stbi__err("bad TQ","Corrupt JPEG");
2979 }
2980
2981 if (scan != STBI__SCAN_load) return 1;
2982
2983 if (!stbi__mad3sizes_valid(s->img_x, s->img_y, s->img_n, 0)) return stbi__err("too large", "Image too large to decode");
2984
2985 for (i=0; i < s->img_n; ++i) {
2986 if (z->img_comp[i].h > h_max) h_max = z->img_comp[i].h;
2987 if (z->img_comp[i].v > v_max) v_max = z->img_comp[i].v;
2988 }
2989
2990 // compute interleaved mcu info
2991 z->img_h_max = h_max;
2992 z->img_v_max = v_max;
2993 z->img_mcu_w = h_max * 8;
2994 z->img_mcu_h = v_max * 8;
2995 // these sizes can't be more than 17 bits
2996 z->img_mcu_x = (s->img_x + z->img_mcu_w-1) / z->img_mcu_w;
2997 z->img_mcu_y = (s->img_y + z->img_mcu_h-1) / z->img_mcu_h;
2998
2999 for (i=0; i < s->img_n; ++i) {
3000 // number of effective pixels (e.g. for non-interleaved MCU)
3001 z->img_comp[i].x = (s->img_x * z->img_comp[i].h + h_max-1) / h_max;
3002 z->img_comp[i].y = (s->img_y * z->img_comp[i].v + v_max-1) / v_max;
3003 // to simplify generation, we'll allocate enough memory to decode
3004 // the bogus oversized data from using interleaved MCUs and their
3005 // big blocks (e.g. a 16x16 iMCU on an image of width 33); we won't
3006 // discard the extra data until colorspace conversion
3007 //
3008 // img_mcu_x, img_mcu_y: <=17 bits; comp[i].h and .v are <=4 (checked earlier)
3009 // so these muls can't overflow with 32-bit ints (which we require)
3010 z->img_comp[i].w2 = z->img_mcu_x * z->img_comp[i].h * 8;
3011 z->img_comp[i].h2 = z->img_mcu_y * z->img_comp[i].v * 8;
3012 z->img_comp[i].coeff = 0;
3013 z->img_comp[i].raw_coeff = 0;
3014 z->img_comp[i].linebuf = NULL;
3015 z->img_comp[i].raw_data = stbi__malloc_mad2(z->img_comp[i].w2, z->img_comp[i].h2, 15);
3016 if (z->img_comp[i].raw_data == NULL)
3017 return stbi__free_jpeg_components(z, i+1, stbi__err("outofmem", "Out of memory"));
3018 // align blocks for idct using mmx/sse
3019 z->img_comp[i].data = (stbi_uc*) (((size_t) z->img_comp[i].raw_data + 15) & ~15);
3020 if (z->progressive) {
3021 // w2, h2 are multiples of 8 (see above)
3022 z->img_comp[i].coeff_w = z->img_comp[i].w2 / 8;
3023 z->img_comp[i].coeff_h = z->img_comp[i].h2 / 8;
3024 z->img_comp[i].raw_coeff = stbi__malloc_mad3(z->img_comp[i].w2, z->img_comp[i].h2, sizeof(short), 15);
3025 if (z->img_comp[i].raw_coeff == NULL)
3026 return stbi__free_jpeg_components(z, i+1, stbi__err("outofmem", "Out of memory"));
3027 z->img_comp[i].coeff = (short*) (((size_t) z->img_comp[i].raw_coeff + 15) & ~15);
3028 }
3029 }
3030
3031 return 1;
3032}
3033
3034// use comparisons since in some cases we handle more than one case (e.g. SOF)
3035#define stbi__DNL(x) ((x) == 0xdc)
3036#define stbi__SOI(x) ((x) == 0xd8)
3037#define stbi__EOI(x) ((x) == 0xd9)
3038#define stbi__SOF(x) ((x) == 0xc0 || (x) == 0xc1 || (x) == 0xc2)
3039#define stbi__SOS(x) ((x) == 0xda)
3040
3041#define stbi__SOF_progressive(x) ((x) == 0xc2)
3042
3043static int stbi__decode_jpeg_header(stbi__jpeg *z, int scan)
3044{
3045 int m;
3046 z->jfif = 0;
3047 z->app14_color_transform = -1; // valid values are 0,1,2
3048 z->marker = STBI__MARKER_none; // initialize cached marker to empty
3049 m = stbi__get_marker(z);
3050 if (!stbi__SOI(m)) return stbi__err("no SOI","Corrupt JPEG");
3051 if (scan == STBI__SCAN_type) return 1;
3052 m = stbi__get_marker(z);
3053 while (!stbi__SOF(m)) {
3054 if (!stbi__process_marker(z,m)) return 0;
3055 m = stbi__get_marker(z);
3056 while (m == STBI__MARKER_none) {
3057 // some files have extra padding after their blocks, so ok, we'll scan
3058 if (stbi__at_eof(z->s)) return stbi__err("no SOF", "Corrupt JPEG");
3059 m = stbi__get_marker(z);
3060 }
3061 }
3062 z->progressive = stbi__SOF_progressive(m);
3063 if (!stbi__process_frame_header(z, scan)) return 0;
3064 return 1;
3065}
3066
3067// decode image to YCbCr format
3068static int stbi__decode_jpeg_image(stbi__jpeg *j)
3069{
3070 int m;
3071 for (m = 0; m < 4; m++) {
3072 j->img_comp[m].raw_data = NULL;
3073 j->img_comp[m].raw_coeff = NULL;
3074 }
3075 j->restart_interval = 0;
3076 if (!stbi__decode_jpeg_header(j, STBI__SCAN_load)) return 0;
3077 m = stbi__get_marker(j);
3078 while (!stbi__EOI(m)) {
3079 if (stbi__SOS(m)) {
3080 if (!stbi__process_scan_header(j)) return 0;
3081 if (!stbi__parse_entropy_coded_data(j)) return 0;
3082 if (j->marker == STBI__MARKER_none ) {
3083 // handle 0s at the end of image data from IP Kamera 9060
3084 while (!stbi__at_eof(j->s)) {
3085 int x = stbi__get8(j->s);
3086 if (x == 255) {
3087 j->marker = stbi__get8(j->s);
3088 break;
3089 }
3090 }
3091 // if we reach eof without hitting a marker, stbi__get_marker() below will fail and we'll eventually return 0
3092 }
3093 } else if (stbi__DNL(m)) {
3094 int Ld = stbi__get16be(j->s);
3095 stbi__uint32 NL = stbi__get16be(j->s);
3096 if (Ld != 4) stbi__err("bad DNL len", "Corrupt JPEG");
3097 if (NL != j->s->img_y) stbi__err("bad DNL height", "Corrupt JPEG");
3098 } else {
3099 if (!stbi__process_marker(j, m)) return 0;
3100 }
3101 m = stbi__get_marker(j);
3102 }
3103 if (j->progressive)
3104 stbi__jpeg_finish(j);
3105 return 1;
3106}
3107
3108// static jfif-centered resampling (across block boundaries)
3109
3110typedef stbi_uc *(*resample_row_func)(stbi_uc *out, stbi_uc *in0, stbi_uc *in1,
3111 int w, int hs);
3112
3113#define stbi__div4(x) ((stbi_uc) ((x) >> 2))
3114
3115static stbi_uc *resample_row_1(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
3116{
3117 STBI_NOTUSED(out);
3118 STBI_NOTUSED(in_far);
3119 STBI_NOTUSED(w);
3120 STBI_NOTUSED(hs);
3121 return in_near;
3122}
3123
3124static stbi_uc* stbi__resample_row_v_2(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
3125{
3126 // need to generate two samples vertically for every one in input
3127 int i;
3128 STBI_NOTUSED(hs);
3129 for (i=0; i < w; ++i)
3130 out[i] = stbi__div4(3*in_near[i] + in_far[i] + 2);
3131 return out;
3132}
3133
3134static stbi_uc* stbi__resample_row_h_2(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
3135{
3136 // need to generate two samples horizontally for every one in input
3137 int i;
3138 stbi_uc *input = in_near;
3139
3140 if (w == 1) {
3141 // if only one sample, can't do any interpolation
3142 out[0] = out[1] = input[0];
3143 return out;
3144 }
3145
3146 out[0] = input[0];
3147 out[1] = stbi__div4(input[0]*3 + input[1] + 2);
3148 for (i=1; i < w-1; ++i) {
3149 int n = 3*input[i]+2;
3150 out[i*2+0] = stbi__div4(n+input[i-1]);
3151 out[i*2+1] = stbi__div4(n+input[i+1]);
3152 }
3153 out[i*2+0] = stbi__div4(input[w-2]*3 + input[w-1] + 2);
3154 out[i*2+1] = input[w-1];
3155
3156 STBI_NOTUSED(in_far);
3157 STBI_NOTUSED(hs);
3158
3159 return out;
3160}
3161
3162#define stbi__div16(x) ((stbi_uc) ((x) >> 4))
3163
3164static stbi_uc *stbi__resample_row_hv_2(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
3165{
3166 // need to generate 2x2 samples for every one in input
3167 int i,t0,t1;
3168 if (w == 1) {
3169 out[0] = out[1] = stbi__div4(3*in_near[0] + in_far[0] + 2);
3170 return out;
3171 }
3172
3173 t1 = 3*in_near[0] + in_far[0];
3174 out[0] = stbi__div4(t1+2);
3175 for (i=1; i < w; ++i) {
3176 t0 = t1;
3177 t1 = 3*in_near[i]+in_far[i];
3178 out[i*2-1] = stbi__div16(3*t0 + t1 + 8);
3179 out[i*2 ] = stbi__div16(3*t1 + t0 + 8);
3180 }
3181 out[w*2-1] = stbi__div4(t1+2);
3182
3183 STBI_NOTUSED(hs);
3184
3185 return out;
3186}
3187
3188#if defined(STBI_SSE2) || defined(STBI_NEON)
3189static stbi_uc *stbi__resample_row_hv_2_simd(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
3190{
3191 // need to generate 2x2 samples for every one in input
3192 int i=0,t0,t1;
3193
3194 if (w == 1) {
3195 out[0] = out[1] = stbi__div4(3*in_near[0] + in_far[0] + 2);
3196 return out;
3197 }
3198
3199 t1 = 3*in_near[0] + in_far[0];
3200 // process groups of 8 pixels for as long as we can.
3201 // note we can't handle the last pixel in a row in this loop
3202 // because we need to handle the filter boundary conditions.
3203 for (; i < ((w-1) & ~7); i += 8) {
3204#if defined(STBI_SSE2)
3205 // load and perform the vertical filtering pass
3206 // this uses 3*x + y = 4*x + (y - x)
3207 __m128i zero = _mm_setzero_si128();
3208 __m128i farb = _mm_loadl_epi64((__m128i *) (in_far + i));
3209 __m128i nearb = _mm_loadl_epi64((__m128i *) (in_near + i));
3210 __m128i farw = _mm_unpacklo_epi8(farb, zero);
3211 __m128i nearw = _mm_unpacklo_epi8(nearb, zero);
3212 __m128i diff = _mm_sub_epi16(farw, nearw);
3213 __m128i nears = _mm_slli_epi16(nearw, 2);
3214 __m128i curr = _mm_add_epi16(nears, diff); // current row
3215
3216 // horizontal filter works the same based on shifted vers of current
3217 // row. "prev" is current row shifted right by 1 pixel; we need to
3218 // insert the previous pixel value (from t1).
3219 // "next" is current row shifted left by 1 pixel, with first pixel
3220 // of next block of 8 pixels added in.
3221 __m128i prv0 = _mm_slli_si128(curr, 2);
3222 __m128i nxt0 = _mm_srli_si128(curr, 2);
3223 __m128i prev = _mm_insert_epi16(prv0, t1, 0);
3224 __m128i next = _mm_insert_epi16(nxt0, 3*in_near[i+8] + in_far[i+8], 7);
3225
3226 // horizontal filter, polyphase implementation since it's convenient:
3227 // even pixels = 3*cur + prev = cur*4 + (prev - cur)
3228 // odd pixels = 3*cur + next = cur*4 + (next - cur)
3229 // note the shared term.
3230 __m128i bias = _mm_set1_epi16(8);
3231 __m128i curs = _mm_slli_epi16(curr, 2);
3232 __m128i prvd = _mm_sub_epi16(prev, curr);
3233 __m128i nxtd = _mm_sub_epi16(next, curr);
3234 __m128i curb = _mm_add_epi16(curs, bias);
3235 __m128i even = _mm_add_epi16(prvd, curb);
3236 __m128i odd = _mm_add_epi16(nxtd, curb);
3237
3238 // interleave even and odd pixels, then undo scaling.
3239 __m128i int0 = _mm_unpacklo_epi16(even, odd);
3240 __m128i int1 = _mm_unpackhi_epi16(even, odd);
3241 __m128i de0 = _mm_srli_epi16(int0, 4);
3242 __m128i de1 = _mm_srli_epi16(int1, 4);
3243
3244 // pack and write output
3245 __m128i outv = _mm_packus_epi16(de0, de1);
3246 _mm_storeu_si128((__m128i *) (out + i*2), outv);
3247#elif defined(STBI_NEON)
3248 // load and perform the vertical filtering pass
3249 // this uses 3*x + y = 4*x + (y - x)
3250 uint8x8_t farb = vld1_u8(in_far + i);
3251 uint8x8_t nearb = vld1_u8(in_near + i);
3252 int16x8_t diff = vreinterpretq_s16_u16(vsubl_u8(farb, nearb));
3253 int16x8_t nears = vreinterpretq_s16_u16(vshll_n_u8(nearb, 2));
3254 int16x8_t curr = vaddq_s16(nears, diff); // current row
3255
3256 // horizontal filter works the same based on shifted vers of current
3257 // row. "prev" is current row shifted right by 1 pixel; we need to
3258 // insert the previous pixel value (from t1).
3259 // "next" is current row shifted left by 1 pixel, with first pixel
3260 // of next block of 8 pixels added in.
3261 int16x8_t prv0 = vextq_s16(curr, curr, 7);
3262 int16x8_t nxt0 = vextq_s16(curr, curr, 1);
3263 int16x8_t prev = vsetq_lane_s16(t1, prv0, 0);
3264 int16x8_t next = vsetq_lane_s16(3*in_near[i+8] + in_far[i+8], nxt0, 7);
3265
3266 // horizontal filter, polyphase implementation since it's convenient:
3267 // even pixels = 3*cur + prev = cur*4 + (prev - cur)
3268 // odd pixels = 3*cur + next = cur*4 + (next - cur)
3269 // note the shared term.
3270 int16x8_t curs = vshlq_n_s16(curr, 2);
3271 int16x8_t prvd = vsubq_s16(prev, curr);
3272 int16x8_t nxtd = vsubq_s16(next, curr);
3273 int16x8_t even = vaddq_s16(curs, prvd);
3274 int16x8_t odd = vaddq_s16(curs, nxtd);
3275
3276 // undo scaling and round, then store with even/odd phases interleaved
3277 uint8x8x2_t o;
3278 o.val[0] = vqrshrun_n_s16(even, 4);
3279 o.val[1] = vqrshrun_n_s16(odd, 4);
3280 vst2_u8(out + i*2, o);
3281#endif
3282
3283 // "previous" value for next iter
3284 t1 = 3*in_near[i+7] + in_far[i+7];
3285 }
3286
3287 t0 = t1;
3288 t1 = 3*in_near[i] + in_far[i];
3289 out[i*2] = stbi__div16(3*t1 + t0 + 8);
3290
3291 for (++i; i < w; ++i) {
3292 t0 = t1;
3293 t1 = 3*in_near[i]+in_far[i];
3294 out[i*2-1] = stbi__div16(3*t0 + t1 + 8);
3295 out[i*2 ] = stbi__div16(3*t1 + t0 + 8);
3296 }
3297 out[w*2-1] = stbi__div4(t1+2);
3298
3299 STBI_NOTUSED(hs);
3300
3301 return out;
3302}
3303#endif
3304
3305static stbi_uc *stbi__resample_row_generic(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
3306{
3307 // resample with nearest-neighbor
3308 int i,j;
3309 STBI_NOTUSED(in_far);
3310 for (i=0; i < w; ++i)
3311 for (j=0; j < hs; ++j)
3312 out[i*hs+j] = in_near[i];
3313 return out;
3314}
3315
3316// this is a reduced-precision calculation of YCbCr-to-RGB introduced
3317// to make sure the code produces the same results in both SIMD and scalar
3318#define stbi__float2fixed(x) (((int) ((x) * 4096.0f + 0.5f)) << 8)
3319static void stbi__YCbCr_to_RGB_row(stbi_uc *out, const stbi_uc *y, const stbi_uc *pcb, const stbi_uc *pcr, int count, int step)
3320{
3321 int i;
3322 for (i=0; i < count; ++i) {
3323 int y_fixed = (y[i] << 20) + (1<<19); // rounding
3324 int r,g,b;
3325 int cr = pcr[i] - 128;
3326 int cb = pcb[i] - 128;
3327 r = y_fixed + cr* stbi__float2fixed(1.40200f);
3328 g = y_fixed + (cr*-stbi__float2fixed(0.71414f)) + ((cb*-stbi__float2fixed(0.34414f)) & 0xffff0000);
3329 b = y_fixed + cb* stbi__float2fixed(1.77200f);
3330 r >>= 20;
3331 g >>= 20;
3332 b >>= 20;
3333 if ((unsigned) r > 255) { if (r < 0) r = 0; else r = 255; }
3334 if ((unsigned) g > 255) { if (g < 0) g = 0; else g = 255; }
3335 if ((unsigned) b > 255) { if (b < 0) b = 0; else b = 255; }
3336 out[0] = (stbi_uc)r;
3337 out[1] = (stbi_uc)g;
3338 out[2] = (stbi_uc)b;
3339 out[3] = 255;
3340 out += step;
3341 }
3342}
3343
3344#if defined(STBI_SSE2) || defined(STBI_NEON)
3345static void stbi__YCbCr_to_RGB_simd(stbi_uc *out, stbi_uc const *y, stbi_uc const *pcb, stbi_uc const *pcr, int count, int step)
3346{
3347 int i = 0;
3348
3349#ifdef STBI_SSE2
3350 // step == 3 is pretty ugly on the final interleave, and i'm not convinced
3351 // it's useful in practice (you wouldn't use it for textures, for example).
3352 // so just accelerate step == 4 case.
3353 if (step == 4) {
3354 // this is a fairly straightforward implementation and not super-optimized.
3355 __m128i signflip = _mm_set1_epi8(-0x80);
3356 __m128i cr_const0 = _mm_set1_epi16( (short) ( 1.40200f*4096.0f+0.5f));
3357 __m128i cr_const1 = _mm_set1_epi16( - (short) ( 0.71414f*4096.0f+0.5f));
3358 __m128i cb_const0 = _mm_set1_epi16( - (short) ( 0.34414f*4096.0f+0.5f));
3359 __m128i cb_const1 = _mm_set1_epi16( (short) ( 1.77200f*4096.0f+0.5f));
3360 __m128i y_bias = _mm_set1_epi8((char) (unsigned char) 128);
3361 __m128i xw = _mm_set1_epi16(255); // alpha channel
3362
3363 for (; i+7 < count; i += 8) {
3364 // load
3365 __m128i y_bytes = _mm_loadl_epi64((__m128i *) (y+i));
3366 __m128i cr_bytes = _mm_loadl_epi64((__m128i *) (pcr+i));
3367 __m128i cb_bytes = _mm_loadl_epi64((__m128i *) (pcb+i));
3368 __m128i cr_biased = _mm_xor_si128(cr_bytes, signflip); // -128
3369 __m128i cb_biased = _mm_xor_si128(cb_bytes, signflip); // -128
3370
3371 // unpack to short (and left-shift cr, cb by 8)
3372 __m128i yw = _mm_unpacklo_epi8(y_bias, y_bytes);
3373 __m128i crw = _mm_unpacklo_epi8(_mm_setzero_si128(), cr_biased);
3374 __m128i cbw = _mm_unpacklo_epi8(_mm_setzero_si128(), cb_biased);
3375
3376 // color transform
3377 __m128i yws = _mm_srli_epi16(yw, 4);
3378 __m128i cr0 = _mm_mulhi_epi16(cr_const0, crw);
3379 __m128i cb0 = _mm_mulhi_epi16(cb_const0, cbw);
3380 __m128i cb1 = _mm_mulhi_epi16(cbw, cb_const1);
3381 __m128i cr1 = _mm_mulhi_epi16(crw, cr_const1);
3382 __m128i rws = _mm_add_epi16(cr0, yws);
3383 __m128i gwt = _mm_add_epi16(cb0, yws);
3384 __m128i bws = _mm_add_epi16(yws, cb1);
3385 __m128i gws = _mm_add_epi16(gwt, cr1);
3386
3387 // descale
3388 __m128i rw = _mm_srai_epi16(rws, 4);
3389 __m128i bw = _mm_srai_epi16(bws, 4);
3390 __m128i gw = _mm_srai_epi16(gws, 4);
3391
3392 // back to byte, set up for transpose
3393 __m128i brb = _mm_packus_epi16(rw, bw);
3394 __m128i gxb = _mm_packus_epi16(gw, xw);
3395
3396 // transpose to interleave channels
3397 __m128i t0 = _mm_unpacklo_epi8(brb, gxb);
3398 __m128i t1 = _mm_unpackhi_epi8(brb, gxb);
3399 __m128i o0 = _mm_unpacklo_epi16(t0, t1);
3400 __m128i o1 = _mm_unpackhi_epi16(t0, t1);
3401
3402 // store
3403 _mm_storeu_si128((__m128i *) (out + 0), o0);
3404 _mm_storeu_si128((__m128i *) (out + 16), o1);
3405 out += 32;
3406 }
3407 }
3408#endif
3409
3410#ifdef STBI_NEON
3411 // in this version, step=3 support would be easy to add. but is there demand?
3412 if (step == 4) {
3413 // this is a fairly straightforward implementation and not super-optimized.
3414 uint8x8_t signflip = vdup_n_u8(0x80);
3415 int16x8_t cr_const0 = vdupq_n_s16( (short) ( 1.40200f*4096.0f+0.5f));
3416 int16x8_t cr_const1 = vdupq_n_s16( - (short) ( 0.71414f*4096.0f+0.5f));
3417 int16x8_t cb_const0 = vdupq_n_s16( - (short) ( 0.34414f*4096.0f+0.5f));
3418 int16x8_t cb_const1 = vdupq_n_s16( (short) ( 1.77200f*4096.0f+0.5f));
3419
3420 for (; i+7 < count; i += 8) {
3421 // load
3422 uint8x8_t y_bytes = vld1_u8(y + i);
3423 uint8x8_t cr_bytes = vld1_u8(pcr + i);
3424 uint8x8_t cb_bytes = vld1_u8(pcb + i);
3425 int8x8_t cr_biased = vreinterpret_s8_u8(vsub_u8(cr_bytes, signflip));
3426 int8x8_t cb_biased = vreinterpret_s8_u8(vsub_u8(cb_bytes, signflip));
3427
3428 // expand to s16
3429 int16x8_t yws = vreinterpretq_s16_u16(vshll_n_u8(y_bytes, 4));
3430 int16x8_t crw = vshll_n_s8(cr_biased, 7);
3431 int16x8_t cbw = vshll_n_s8(cb_biased, 7);
3432
3433 // color transform
3434 int16x8_t cr0 = vqdmulhq_s16(crw, cr_const0);
3435 int16x8_t cb0 = vqdmulhq_s16(cbw, cb_const0);
3436 int16x8_t cr1 = vqdmulhq_s16(crw, cr_const1);
3437 int16x8_t cb1 = vqdmulhq_s16(cbw, cb_const1);
3438 int16x8_t rws = vaddq_s16(yws, cr0);
3439 int16x8_t gws = vaddq_s16(vaddq_s16(yws, cb0), cr1);
3440 int16x8_t bws = vaddq_s16(yws, cb1);
3441
3442 // undo scaling, round, convert to byte
3443 uint8x8x4_t o;
3444 o.val[0] = vqrshrun_n_s16(rws, 4);
3445 o.val[1] = vqrshrun_n_s16(gws, 4);
3446 o.val[2] = vqrshrun_n_s16(bws, 4);
3447 o.val[3] = vdup_n_u8(255);
3448
3449 // store, interleaving r/g/b/a
3450 vst4_u8(out, o);
3451 out += 8*4;
3452 }
3453 }
3454#endif
3455
3456 for (; i < count; ++i) {
3457 int y_fixed = (y[i] << 20) + (1<<19); // rounding
3458 int r,g,b;
3459 int cr = pcr[i] - 128;
3460 int cb = pcb[i] - 128;
3461 r = y_fixed + cr* stbi__float2fixed(1.40200f);
3462 g = y_fixed + cr*-stbi__float2fixed(0.71414f) + ((cb*-stbi__float2fixed(0.34414f)) & 0xffff0000);
3463 b = y_fixed + cb* stbi__float2fixed(1.77200f);
3464 r >>= 20;
3465 g >>= 20;
3466 b >>= 20;
3467 if ((unsigned) r > 255) { if (r < 0) r = 0; else r = 255; }
3468 if ((unsigned) g > 255) { if (g < 0) g = 0; else g = 255; }
3469 if ((unsigned) b > 255) { if (b < 0) b = 0; else b = 255; }
3470 out[0] = (stbi_uc)r;
3471 out[1] = (stbi_uc)g;
3472 out[2] = (stbi_uc)b;
3473 out[3] = 255;
3474 out += step;
3475 }
3476}
3477#endif
3478
3479// set up the kernels
3480static void stbi__setup_jpeg(stbi__jpeg *j)
3481{
3482 j->idct_block_kernel = stbi__idct_block;
3483 j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_row;
3484 j->resample_row_hv_2_kernel = stbi__resample_row_hv_2;
3485
3486#ifdef STBI_SSE2
3487 if (stbi__sse2_available()) {
3488 j->idct_block_kernel = stbi__idct_simd;
3489 j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_simd;
3490 j->resample_row_hv_2_kernel = stbi__resample_row_hv_2_simd;
3491 }
3492#endif
3493
3494#ifdef STBI_NEON
3495 j->idct_block_kernel = stbi__idct_simd;
3496 j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_simd;
3497 j->resample_row_hv_2_kernel = stbi__resample_row_hv_2_simd;
3498#endif
3499}
3500
3501// clean up the temporary component buffers
3502static void stbi__cleanup_jpeg(stbi__jpeg *j)
3503{
3504 stbi__free_jpeg_components(j, j->s->img_n, 0);
3505}
3506
3507typedef struct
3508{
3509 resample_row_func resample;
3510 stbi_uc *line0,*line1;
3511 int hs,vs; // expansion factor in each axis
3512 int w_lores; // horizontal pixels pre-expansion
3513 int ystep; // how far through vertical expansion we are
3514 int ypos; // which pre-expansion row we're on
3515} stbi__resample;
3516
3517// fast 0..255 * 0..255 => 0..255 rounded multiplication
3518static stbi_uc stbi__blinn_8x8(stbi_uc x, stbi_uc y)
3519{
3520 unsigned int t = x*y + 128;
3521 return (stbi_uc) ((t + (t >>8)) >> 8);
3522}
3523
3524static stbi_uc *load_jpeg_image(stbi__jpeg *z, int *out_x, int *out_y, int *comp, int req_comp)
3525{
3526 int n, decode_n, is_rgb;
3527 z->s->img_n = 0; // make stbi__cleanup_jpeg safe
3528
3529 // validate req_comp
3530 if (req_comp < 0 || req_comp > 4) return stbi__errpuc("bad req_comp", "Internal error");
3531
3532 // load a jpeg image from whichever source, but leave in YCbCr format
3533 if (!stbi__decode_jpeg_image(z)) { stbi__cleanup_jpeg(z); return NULL; }
3534
3535 // determine actual number of components to generate
3536 n = req_comp ? req_comp : z->s->img_n >= 3 ? 3 : 1;
3537
3538 is_rgb = z->s->img_n == 3 && (z->rgb == 3 || (z->app14_color_transform == 0 && !z->jfif));
3539
3540 if (z->s->img_n == 3 && n < 3 && !is_rgb)
3541 decode_n = 1;
3542 else
3543 decode_n = z->s->img_n;
3544
3545 // resample and color-convert
3546 {
3547 int k;
3548 unsigned int i,j;
3549 stbi_uc *output;
3550 stbi_uc *coutput[4];
3551
3552 stbi__resample res_comp[4];
3553
3554 for (k=0; k < decode_n; ++k) {
3555 stbi__resample *r = &res_comp[k];
3556
3557 // allocate line buffer big enough for upsampling off the edges
3558 // with upsample factor of 4
3559 z->img_comp[k].linebuf = (stbi_uc *) stbi__malloc(z->s->img_x + 3);
3560 if (!z->img_comp[k].linebuf) { stbi__cleanup_jpeg(z); return stbi__errpuc("outofmem", "Out of memory"); }
3561
3562 r->hs = z->img_h_max / z->img_comp[k].h;
3563 r->vs = z->img_v_max / z->img_comp[k].v;
3564 r->ystep = r->vs >> 1;
3565 r->w_lores = (z->s->img_x + r->hs-1) / r->hs;
3566 r->ypos = 0;
3567 r->line0 = r->line1 = z->img_comp[k].data;
3568
3569 if (r->hs == 1 && r->vs == 1) r->resample = resample_row_1;
3570 else if (r->hs == 1 && r->vs == 2) r->resample = stbi__resample_row_v_2;
3571 else if (r->hs == 2 && r->vs == 1) r->resample = stbi__resample_row_h_2;
3572 else if (r->hs == 2 && r->vs == 2) r->resample = z->resample_row_hv_2_kernel;
3573 else r->resample = stbi__resample_row_generic;
3574 }
3575
3576 // can't error after this so, this is safe
3577 output = (stbi_uc *) stbi__malloc_mad3(n, z->s->img_x, z->s->img_y, 1);
3578 if (!output) { stbi__cleanup_jpeg(z); return stbi__errpuc("outofmem", "Out of memory"); }
3579
3580 // now go ahead and resample
3581 for (j=0; j < z->s->img_y; ++j) {
3582 stbi_uc *out = output + n * z->s->img_x * j;
3583 for (k=0; k < decode_n; ++k) {
3584 stbi__resample *r = &res_comp[k];
3585 int y_bot = r->ystep >= (r->vs >> 1);
3586 coutput[k] = r->resample(z->img_comp[k].linebuf,
3587 y_bot ? r->line1 : r->line0,
3588 y_bot ? r->line0 : r->line1,
3589 r->w_lores, r->hs);
3590 if (++r->ystep >= r->vs) {
3591 r->ystep = 0;
3592 r->line0 = r->line1;
3593 if (++r->ypos < z->img_comp[k].y)
3594 r->line1 += z->img_comp[k].w2;
3595 }
3596 }
3597 if (n >= 3) {
3598 stbi_uc *y = coutput[0];
3599 if (z->s->img_n == 3) {
3600 if (is_rgb) {
3601 for (i=0; i < z->s->img_x; ++i) {
3602 out[0] = y[i];
3603 out[1] = coutput[1][i];
3604 out[2] = coutput[2][i];
3605 out[3] = 255;
3606 out += n;
3607 }
3608 } else {
3609 z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n);
3610 }
3611 } else if (z->s->img_n == 4) {
3612 if (z->app14_color_transform == 0) { // CMYK
3613 for (i=0; i < z->s->img_x; ++i) {
3614 stbi_uc k = coutput[3][i];
3615 out[0] = stbi__blinn_8x8(coutput[0][i], k);
3616 out[1] = stbi__blinn_8x8(coutput[1][i], k);
3617 out[2] = stbi__blinn_8x8(coutput[2][i], k);
3618 out[3] = 255;
3619 out += n;
3620 }
3621 } else if (z->app14_color_transform == 2) { // YCCK
3622 z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n);
3623 for (i=0; i < z->s->img_x; ++i) {
3624 stbi_uc k = coutput[3][i];
3625 out[0] = stbi__blinn_8x8(255 - out[0], k);
3626 out[1] = stbi__blinn_8x8(255 - out[1], k);
3627 out[2] = stbi__blinn_8x8(255 - out[2], k);
3628 out += n;
3629 }
3630 } else { // YCbCr + alpha? Ignore the fourth channel for now
3631 z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n);
3632 }
3633 } else
3634 for (i=0; i < z->s->img_x; ++i) {
3635 out[0] = out[1] = out[2] = y[i];
3636 out[3] = 255; // not used if n==3
3637 out += n;
3638 }
3639 } else {
3640 if (is_rgb) {
3641 if (n == 1)
3642 for (i=0; i < z->s->img_x; ++i)
3643 *out++ = stbi__compute_y(coutput[0][i], coutput[1][i], coutput[2][i]);
3644 else {
3645 for (i=0; i < z->s->img_x; ++i, out += 2) {
3646 out[0] = stbi__compute_y(coutput[0][i], coutput[1][i], coutput[2][i]);
3647 out[1] = 255;
3648 }
3649 }
3650 } else if (z->s->img_n == 4 && z->app14_color_transform == 0) {
3651 for (i=0; i < z->s->img_x; ++i) {
3652 stbi_uc k = coutput[3][i];
3653 stbi_uc r = stbi__blinn_8x8(coutput[0][i], k);
3654 stbi_uc g = stbi__blinn_8x8(coutput[1][i], k);
3655 stbi_uc b = stbi__blinn_8x8(coutput[2][i], k);
3656 out[0] = stbi__compute_y(r, g, b);
3657 out[1] = 255;
3658 out += n;
3659 }
3660 } else if (z->s->img_n == 4 && z->app14_color_transform == 2) {
3661 for (i=0; i < z->s->img_x; ++i) {
3662 out[0] = stbi__blinn_8x8(255 - coutput[0][i], coutput[3][i]);
3663 out[1] = 255;
3664 out += n;
3665 }
3666 } else {
3667 stbi_uc *y = coutput[0];
3668 if (n == 1)
3669 for (i=0; i < z->s->img_x; ++i) out[i] = y[i];
3670 else
3671 for (i=0; i < z->s->img_x; ++i) *out++ = y[i], *out++ = 255;
3672 }
3673 }
3674 }
3675 stbi__cleanup_jpeg(z);
3676 *out_x = z->s->img_x;
3677 *out_y = z->s->img_y;
3678 if (comp) *comp = z->s->img_n >= 3 ? 3 : 1; // report original components, not output
3679 return output;
3680 }
3681}
3682
3683static void *stbi__jpeg_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
3684{
3685 unsigned char* result;
3686 stbi__jpeg* j = (stbi__jpeg*) stbi__malloc(sizeof(stbi__jpeg));
3687 STBI_NOTUSED(ri);
3688 j->s = s;
3689 stbi__setup_jpeg(j);
3690 result = load_jpeg_image(j, x,y,comp,req_comp);
3691 STBI_FREE(j);
3692 return result;
3693}
3694
3695static int stbi__jpeg_test(stbi__context *s)
3696{
3697 int r;
3698 stbi__jpeg* j = (stbi__jpeg*)stbi__malloc(sizeof(stbi__jpeg));
3699 j->s = s;
3700 stbi__setup_jpeg(j);
3701 r = stbi__decode_jpeg_header(j, STBI__SCAN_type);
3702 stbi__rewind(s);
3703 STBI_FREE(j);
3704 return r;
3705}
3706
3707static int stbi__jpeg_info_raw(stbi__jpeg *j, int *x, int *y, int *comp)
3708{
3709 if (!stbi__decode_jpeg_header(j, STBI__SCAN_header)) {
3710 stbi__rewind( j->s );
3711 return 0;
3712 }
3713 if (x) *x = j->s->img_x;
3714 if (y) *y = j->s->img_y;
3715 if (comp) *comp = j->s->img_n >= 3 ? 3 : 1;
3716 return 1;
3717}
3718
3719static int stbi__jpeg_info(stbi__context *s, int *x, int *y, int *comp)
3720{
3721 int result;
3722 stbi__jpeg* j = (stbi__jpeg*) (stbi__malloc(sizeof(stbi__jpeg)));
3723 j->s = s;
3724 result = stbi__jpeg_info_raw(j, x, y, comp);
3725 STBI_FREE(j);
3726 return result;
3727}
3728#endif
3729
3730// public domain zlib decode v0.2 Sean Barrett 2006-11-18
3731// simple implementation
3732// - all input must be provided in an upfront buffer
3733// - all output is written to a single output buffer (can malloc/realloc)
3734// performance
3735// - fast huffman
3736
3737#ifndef STBI_NO_ZLIB
3738
3739// fast-way is faster to check than jpeg huffman, but slow way is slower
3740#define STBI__ZFAST_BITS 9 // accelerate all cases in default tables
3741#define STBI__ZFAST_MASK ((1 << STBI__ZFAST_BITS) - 1)
3742
3743// zlib-style huffman encoding
3744// (jpegs packs from left, zlib from right, so can't share code)
3745typedef struct
3746{
3747 stbi__uint16 fast[1 << STBI__ZFAST_BITS];
3748 stbi__uint16 firstcode[16];
3749 int maxcode[17];
3750 stbi__uint16 firstsymbol[16];
3751 stbi_uc size[288];
3752 stbi__uint16 value[288];
3753} stbi__zhuffman;
3754
3755stbi_inline static int stbi__bitreverse16(int n)
3756{
3757 n = ((n & 0xAAAA) >> 1) | ((n & 0x5555) << 1);
3758 n = ((n & 0xCCCC) >> 2) | ((n & 0x3333) << 2);
3759 n = ((n & 0xF0F0) >> 4) | ((n & 0x0F0F) << 4);
3760 n = ((n & 0xFF00) >> 8) | ((n & 0x00FF) << 8);
3761 return n;
3762}
3763
3764stbi_inline static int stbi__bit_reverse(int v, int bits)
3765{
3766 STBI_ASSERT(bits <= 16);
3767 // to bit reverse n bits, reverse 16 and shift
3768 // e.g. 11 bits, bit reverse and shift away 5
3769 return stbi__bitreverse16(v) >> (16-bits);
3770}
3771
3772static int stbi__zbuild_huffman(stbi__zhuffman *z, const stbi_uc *sizelist, int num)
3773{
3774 int i,k=0;
3775 int code, next_code[16], sizes[17];
3776
3777 // DEFLATE spec for generating codes
3778 memset(sizes, 0, sizeof(sizes));
3779 memset(z->fast, 0, sizeof(z->fast));
3780 for (i=0; i < num; ++i)
3781 ++sizes[sizelist[i]];
3782 sizes[0] = 0;
3783 for (i=1; i < 16; ++i)
3784 if (sizes[i] > (1 << i))
3785 return stbi__err("bad sizes", "Corrupt PNG");
3786 code = 0;
3787 for (i=1; i < 16; ++i) {
3788 next_code[i] = code;
3789 z->firstcode[i] = (stbi__uint16) code;
3790 z->firstsymbol[i] = (stbi__uint16) k;
3791 code = (code + sizes[i]);
3792 if (sizes[i])
3793 if (code-1 >= (1 << i)) return stbi__err("bad codelengths","Corrupt PNG");
3794 z->maxcode[i] = code << (16-i); // preshift for inner loop
3795 code <<= 1;
3796 k += sizes[i];
3797 }
3798 z->maxcode[16] = 0x10000; // sentinel
3799 for (i=0; i < num; ++i) {
3800 int s = sizelist[i];
3801 if (s) {
3802 int c = next_code[s] - z->firstcode[s] + z->firstsymbol[s];
3803 stbi__uint16 fastv = (stbi__uint16) ((s << 9) | i);
3804 z->size [c] = (stbi_uc ) s;
3805 z->value[c] = (stbi__uint16) i;
3806 if (s <= STBI__ZFAST_BITS) {
3807 int j = stbi__bit_reverse(next_code[s],s);
3808 while (j < (1 << STBI__ZFAST_BITS)) {
3809 z->fast[j] = fastv;
3810 j += (1 << s);
3811 }
3812 }
3813 ++next_code[s];
3814 }
3815 }
3816 return 1;
3817}
3818
3819// zlib-from-memory implementation for PNG reading
3820// because PNG allows splitting the zlib stream arbitrarily,
3821// and it's annoying structurally to have PNG call ZLIB call PNG,
3822// we require PNG read all the IDATs and combine them into a single
3823// memory buffer
3824
3825typedef struct
3826{
3827 stbi_uc *zbuffer, *zbuffer_end;
3828 int num_bits;
3829 stbi__uint32 code_buffer;
3830
3831 char *zout;
3832 char *zout_start;
3833 char *zout_end;
3834 int z_expandable;
3835
3836 stbi__zhuffman z_length, z_distance;
3837} stbi__zbuf;
3838
3839stbi_inline static stbi_uc stbi__zget8(stbi__zbuf *z)
3840{
3841 if (z->zbuffer >= z->zbuffer_end) return 0;
3842 return *z->zbuffer++;
3843}
3844
3845static void stbi__fill_bits(stbi__zbuf *z)
3846{
3847 do {
3848 STBI_ASSERT(z->code_buffer < (1U << z->num_bits));
3849 z->code_buffer |= (unsigned int) stbi__zget8(z) << z->num_bits;
3850 z->num_bits += 8;
3851 } while (z->num_bits <= 24);
3852}
3853
3854stbi_inline static unsigned int stbi__zreceive(stbi__zbuf *z, int n)
3855{
3856 unsigned int k;
3857 if (z->num_bits < n) stbi__fill_bits(z);
3858 k = z->code_buffer & ((1 << n) - 1);
3859 z->code_buffer >>= n;
3860 z->num_bits -= n;
3861 return k;
3862}
3863
3864static int stbi__zhuffman_decode_slowpath(stbi__zbuf *a, stbi__zhuffman *z)
3865{
3866 int b,s,k;
3867 // not resolved by fast table, so compute it the slow way
3868 // use jpeg approach, which requires MSbits at top
3869 k = stbi__bit_reverse(a->code_buffer, 16);
3870 for (s=STBI__ZFAST_BITS+1; ; ++s)
3871 if (k < z->maxcode[s])
3872 break;
3873 if (s == 16) return -1; // invalid code!
3874 // code size is s, so:
3875 b = (k >> (16-s)) - z->firstcode[s] + z->firstsymbol[s];
3876 STBI_ASSERT(z->size[b] == s);
3877 a->code_buffer >>= s;
3878 a->num_bits -= s;
3879 return z->value[b];
3880}
3881
3882stbi_inline static int stbi__zhuffman_decode(stbi__zbuf *a, stbi__zhuffman *z)
3883{
3884 int b,s;
3885 if (a->num_bits < 16) stbi__fill_bits(a);
3886 b = z->fast[a->code_buffer & STBI__ZFAST_MASK];
3887 if (b) {
3888 s = b >> 9;
3889 a->code_buffer >>= s;
3890 a->num_bits -= s;
3891 return b & 511;
3892 }
3893 return stbi__zhuffman_decode_slowpath(a, z);
3894}
3895
3896static int stbi__zexpand(stbi__zbuf *z, char *zout, int n) // need to make room for n bytes
3897{
3898 char *q;
3899 int cur, limit, old_limit;
3900 z->zout = zout;
3901 if (!z->z_expandable) return stbi__err("output buffer limit","Corrupt PNG");
3902 cur = (int) (z->zout - z->zout_start);
3903 limit = old_limit = (int) (z->zout_end - z->zout_start);
3904 while (cur + n > limit)
3905 limit *= 2;
3906 q = (char *) STBI_REALLOC_SIZED(z->zout_start, old_limit, limit);
3907 STBI_NOTUSED(old_limit);
3908 if (q == NULL) return stbi__err("outofmem", "Out of memory");
3909 z->zout_start = q;
3910 z->zout = q + cur;
3911 z->zout_end = q + limit;
3912 return 1;
3913}
3914
3915static int stbi__zlength_base[31] = {
3916 3,4,5,6,7,8,9,10,11,13,
3917 15,17,19,23,27,31,35,43,51,59,
3918 67,83,99,115,131,163,195,227,258,0,0 };
3919
3920static int stbi__zlength_extra[31]=
3921{ 0,0,0,0,0,0,0,0,1,1,1,1,2,2,2,2,3,3,3,3,4,4,4,4,5,5,5,5,0,0,0 };
3922
3923static int stbi__zdist_base[32] = { 1,2,3,4,5,7,9,13,17,25,33,49,65,97,129,193,
3924257,385,513,769,1025,1537,2049,3073,4097,6145,8193,12289,16385,24577,0,0};
3925
3926static int stbi__zdist_extra[32] =
3927{ 0,0,0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,8,8,9,9,10,10,11,11,12,12,13,13};
3928
3929static int stbi__parse_huffman_block(stbi__zbuf *a)
3930{
3931 char *zout = a->zout;
3932 for(;;) {
3933 int z = stbi__zhuffman_decode(a, &a->z_length);
3934 if (z < 256) {
3935 if (z < 0) return stbi__err("bad huffman code","Corrupt PNG"); // error in huffman codes
3936 if (zout >= a->zout_end) {
3937 if (!stbi__zexpand(a, zout, 1)) return 0;
3938 zout = a->zout;
3939 }
3940 *zout++ = (char) z;
3941 } else {
3942 stbi_uc *p;
3943 int len,dist;
3944 if (z == 256) {
3945 a->zout = zout;
3946 return 1;
3947 }
3948 z -= 257;
3949 len = stbi__zlength_base[z];
3950 if (stbi__zlength_extra[z]) len += stbi__zreceive(a, stbi__zlength_extra[z]);
3951 z = stbi__zhuffman_decode(a, &a->z_distance);
3952 if (z < 0) return stbi__err("bad huffman code","Corrupt PNG");
3953 dist = stbi__zdist_base[z];
3954 if (stbi__zdist_extra[z]) dist += stbi__zreceive(a, stbi__zdist_extra[z]);
3955 if (zout - a->zout_start < dist) return stbi__err("bad dist","Corrupt PNG");
3956 if (zout + len > a->zout_end) {
3957 if (!stbi__zexpand(a, zout, len)) return 0;
3958 zout = a->zout;
3959 }
3960 p = (stbi_uc *) (zout - dist);
3961 if (dist == 1) { // run of one byte; common in images.
3962 stbi_uc v = *p;
3963 if (len) { do *zout++ = v; while (--len); }
3964 } else {
3965 if (len) { do *zout++ = *p++; while (--len); }
3966 }
3967 }
3968 }
3969}
3970
3971static int stbi__compute_huffman_codes(stbi__zbuf *a)
3972{
3973 static stbi_uc length_dezigzag[19] = { 16,17,18,0,8,7,9,6,10,5,11,4,12,3,13,2,14,1,15 };
3974 stbi__zhuffman z_codelength;
3975 stbi_uc lencodes[286+32+137];//padding for maximum single op
3976 stbi_uc codelength_sizes[19];
3977 int i,n;
3978
3979 int hlit = stbi__zreceive(a,5) + 257;
3980 int hdist = stbi__zreceive(a,5) + 1;
3981 int hclen = stbi__zreceive(a,4) + 4;
3982 int ntot = hlit + hdist;
3983
3984 memset(codelength_sizes, 0, sizeof(codelength_sizes));
3985 for (i=0; i < hclen; ++i) {
3986 int s = stbi__zreceive(a,3);
3987 codelength_sizes[length_dezigzag[i]] = (stbi_uc) s;
3988 }
3989 if (!stbi__zbuild_huffman(&z_codelength, codelength_sizes, 19)) return 0;
3990
3991 n = 0;
3992 while (n < ntot) {
3993 int c = stbi__zhuffman_decode(a, &z_codelength);
3994 if (c < 0 || c >= 19) return stbi__err("bad codelengths", "Corrupt PNG");
3995 if (c < 16)
3996 lencodes[n++] = (stbi_uc) c;
3997 else {
3998 stbi_uc fill = 0;
3999 if (c == 16) {
4000 c = stbi__zreceive(a,2)+3;
4001 if (n == 0) return stbi__err("bad codelengths", "Corrupt PNG");
4002 fill = lencodes[n-1];
4003 } else if (c == 17)
4004 c = stbi__zreceive(a,3)+3;
4005 else {
4006 STBI_ASSERT(c == 18);
4007 c = stbi__zreceive(a,7)+11;
4008 }
4009 if (ntot - n < c) return stbi__err("bad codelengths", "Corrupt PNG");
4010 memset(lencodes+n, fill, c);
4011 n += c;
4012 }
4013 }
4014 if (n != ntot) return stbi__err("bad codelengths","Corrupt PNG");
4015 if (!stbi__zbuild_huffman(&a->z_length, lencodes, hlit)) return 0;
4016 if (!stbi__zbuild_huffman(&a->z_distance, lencodes+hlit, hdist)) return 0;
4017 return 1;
4018}
4019
4020static int stbi__parse_uncompressed_block(stbi__zbuf *a)
4021{
4022 stbi_uc header[4];
4023 int len,nlen,k;
4024 if (a->num_bits & 7)
4025 stbi__zreceive(a, a->num_bits & 7); // discard
4026 // drain the bit-packed data into header
4027 k = 0;
4028 while (a->num_bits > 0) {
4029 header[k++] = (stbi_uc) (a->code_buffer & 255); // suppress MSVC run-time check
4030 a->code_buffer >>= 8;
4031 a->num_bits -= 8;
4032 }
4033 STBI_ASSERT(a->num_bits == 0);
4034 // now fill header the normal way
4035 while (k < 4)
4036 header[k++] = stbi__zget8(a);
4037 len = header[1] * 256 + header[0];
4038 nlen = header[3] * 256 + header[2];
4039 if (nlen != (len ^ 0xffff)) return stbi__err("zlib corrupt","Corrupt PNG");
4040 if (a->zbuffer + len > a->zbuffer_end) return stbi__err("read past buffer","Corrupt PNG");
4041 if (a->zout + len > a->zout_end)
4042 if (!stbi__zexpand(a, a->zout, len)) return 0;
4043 memcpy(a->zout, a->zbuffer, len);
4044 a->zbuffer += len;
4045 a->zout += len;
4046 return 1;
4047}
4048
4049static int stbi__parse_zlib_header(stbi__zbuf *a)
4050{
4051 int cmf = stbi__zget8(a);
4052 int cm = cmf & 15;
4053 /* int cinfo = cmf >> 4; */
4054 int flg = stbi__zget8(a);
4055 if ((cmf*256+flg) % 31 != 0) return stbi__err("bad zlib header","Corrupt PNG"); // zlib spec
4056 if (flg & 32) return stbi__err("no preset dict","Corrupt PNG"); // preset dictionary not allowed in png
4057 if (cm != 8) return stbi__err("bad compression","Corrupt PNG"); // DEFLATE required for png
4058 // window = 1 << (8 + cinfo)... but who cares, we fully buffer output
4059 return 1;
4060}
4061
4062static const stbi_uc stbi__zdefault_length[288] =
4063{
4064 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,
4065 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,
4066 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,
4067 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,
4068 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,
4069 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,
4070 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,
4071 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,
4072 7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7, 7,7,7,7,7,7,7,7,8,8,8,8,8,8,8,8
4073};
4074static const stbi_uc stbi__zdefault_distance[32] =
4075{
4076 5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5
4077};
4078/*
4079Init algorithm:
4080{
4081 int i; // use <= to match clearly with spec
4082 for (i=0; i <= 143; ++i) stbi__zdefault_length[i] = 8;
4083 for ( ; i <= 255; ++i) stbi__zdefault_length[i] = 9;
4084 for ( ; i <= 279; ++i) stbi__zdefault_length[i] = 7;
4085 for ( ; i <= 287; ++i) stbi__zdefault_length[i] = 8;
4086
4087 for (i=0; i <= 31; ++i) stbi__zdefault_distance[i] = 5;
4088}
4089*/
4090
4091static int stbi__parse_zlib(stbi__zbuf *a, int parse_header)
4092{
4093 int final, type;
4094 if (parse_header)
4095 if (!stbi__parse_zlib_header(a)) return 0;
4096 a->num_bits = 0;
4097 a->code_buffer = 0;
4098 do {
4099 final = stbi__zreceive(a,1);
4100 type = stbi__zreceive(a,2);
4101 if (type == 0) {
4102 if (!stbi__parse_uncompressed_block(a)) return 0;
4103 } else if (type == 3) {
4104 return 0;
4105 } else {
4106 if (type == 1) {
4107 // use fixed code lengths
4108 if (!stbi__zbuild_huffman(&a->z_length , stbi__zdefault_length , 288)) return 0;
4109 if (!stbi__zbuild_huffman(&a->z_distance, stbi__zdefault_distance, 32)) return 0;
4110 } else {
4111 if (!stbi__compute_huffman_codes(a)) return 0;
4112 }
4113 if (!stbi__parse_huffman_block(a)) return 0;
4114 }
4115 } while (!final);
4116 return 1;
4117}
4118
4119static int stbi__do_zlib(stbi__zbuf *a, char *obuf, int olen, int exp, int parse_header)
4120{
4121 a->zout_start = obuf;
4122 a->zout = obuf;
4123 a->zout_end = obuf + olen;
4124 a->z_expandable = exp;
4125
4126 return stbi__parse_zlib(a, parse_header);
4127}
4128
4129STBIDEF char *stbi_zlib_decode_malloc_guesssize(const char *buffer, int len, int initial_size, int *outlen)
4130{
4131 stbi__zbuf a;
4132 char *p = (char *) stbi__malloc(initial_size);
4133 if (p == NULL) return NULL;
4134 a.zbuffer = (stbi_uc *) buffer;
4135 a.zbuffer_end = (stbi_uc *) buffer + len;
4136 if (stbi__do_zlib(&a, p, initial_size, 1, 1)) {
4137 if (outlen) *outlen = (int) (a.zout - a.zout_start);
4138 return a.zout_start;
4139 } else {
4140 STBI_FREE(a.zout_start);
4141 return NULL;
4142 }
4143}
4144
4145STBIDEF char *stbi_zlib_decode_malloc(char const *buffer, int len, int *outlen)
4146{
4147 return stbi_zlib_decode_malloc_guesssize(buffer, len, 16384, outlen);
4148}
4149
4150STBIDEF char *stbi_zlib_decode_malloc_guesssize_headerflag(const char *buffer, int len, int initial_size, int *outlen, int parse_header)
4151{
4152 stbi__zbuf a;
4153 char *p = (char *) stbi__malloc(initial_size);
4154 if (p == NULL) return NULL;
4155 a.zbuffer = (stbi_uc *) buffer;
4156 a.zbuffer_end = (stbi_uc *) buffer + len;
4157 if (stbi__do_zlib(&a, p, initial_size, 1, parse_header)) {
4158 if (outlen) *outlen = (int) (a.zout - a.zout_start);
4159 return a.zout_start;
4160 } else {
4161 STBI_FREE(a.zout_start);
4162 return NULL;
4163 }
4164}
4165
4166STBIDEF int stbi_zlib_decode_buffer(char *obuffer, int olen, char const *ibuffer, int ilen)
4167{
4168 stbi__zbuf a;
4169 a.zbuffer = (stbi_uc *) ibuffer;
4170 a.zbuffer_end = (stbi_uc *) ibuffer + ilen;
4171 if (stbi__do_zlib(&a, obuffer, olen, 0, 1))
4172 return (int) (a.zout - a.zout_start);
4173 else
4174 return -1;
4175}
4176
4177STBIDEF char *stbi_zlib_decode_noheader_malloc(char const *buffer, int len, int *outlen)
4178{
4179 stbi__zbuf a;
4180 char *p = (char *) stbi__malloc(16384);
4181 if (p == NULL) return NULL;
4182 a.zbuffer = (stbi_uc *) buffer;
4183 a.zbuffer_end = (stbi_uc *) buffer+len;
4184 if (stbi__do_zlib(&a, p, 16384, 1, 0)) {
4185 if (outlen) *outlen = (int) (a.zout - a.zout_start);
4186 return a.zout_start;
4187 } else {
4188 STBI_FREE(a.zout_start);
4189 return NULL;
4190 }
4191}
4192
4193STBIDEF int stbi_zlib_decode_noheader_buffer(char *obuffer, int olen, const char *ibuffer, int ilen)
4194{
4195 stbi__zbuf a;
4196 a.zbuffer = (stbi_uc *) ibuffer;
4197 a.zbuffer_end = (stbi_uc *) ibuffer + ilen;
4198 if (stbi__do_zlib(&a, obuffer, olen, 0, 0))
4199 return (int) (a.zout - a.zout_start);
4200 else
4201 return -1;
4202}
4203#endif
4204
4205// public domain "baseline" PNG decoder v0.10 Sean Barrett 2006-11-18
4206// simple implementation
4207// - only 8-bit samples
4208// - no CRC checking
4209// - allocates lots of intermediate memory
4210// - avoids problem of streaming data between subsystems
4211// - avoids explicit window management
4212// performance
4213// - uses stb_zlib, a PD zlib implementation with fast huffman decoding
4214
4215#ifndef STBI_NO_PNG
4216typedef struct
4217{
4218 stbi__uint32 length;
4219 stbi__uint32 type;
4220} stbi__pngchunk;
4221
4222static stbi__pngchunk stbi__get_chunk_header(stbi__context *s)
4223{
4224 stbi__pngchunk c;
4225 c.length = stbi__get32be(s);
4226 c.type = stbi__get32be(s);
4227 return c;
4228}
4229
4230static int stbi__check_png_header(stbi__context *s)
4231{
4232 static stbi_uc png_sig[8] = { 137,80,78,71,13,10,26,10 };
4233 int i;
4234 for (i=0; i < 8; ++i)
4235 if (stbi__get8(s) != png_sig[i]) return stbi__err("bad png sig","Not a PNG");
4236 return 1;
4237}
4238
4239typedef struct
4240{
4241 stbi__context *s;
4242 stbi_uc *idata, *expanded, *out;
4243 int depth;
4244} stbi__png;
4245
4246
4247enum {
4248 STBI__F_none=0,
4249 STBI__F_sub=1,
4250 STBI__F_up=2,
4251 STBI__F_avg=3,
4252 STBI__F_paeth=4,
4253 // synthetic filters used for first scanline to avoid needing a dummy row of 0s
4254 STBI__F_avg_first,
4255 STBI__F_paeth_first
4256};
4257
4258static stbi_uc first_row_filter[5] =
4259{
4260 STBI__F_none,
4261 STBI__F_sub,
4262 STBI__F_none,
4263 STBI__F_avg_first,
4264 STBI__F_paeth_first
4265};
4266
4267static int stbi__paeth(int a, int b, int c)
4268{
4269 int p = a + b - c;
4270 int pa = abs(p-a);
4271 int pb = abs(p-b);
4272 int pc = abs(p-c);
4273 if (pa <= pb && pa <= pc) return a;
4274 if (pb <= pc) return b;
4275 return c;
4276}
4277
4278static stbi_uc stbi__depth_scale_table[9] = { 0, 0xff, 0x55, 0, 0x11, 0,0,0, 0x01 };
4279
4280// create the png data from post-deflated data
4281static int stbi__create_png_image_raw(stbi__png *a, stbi_uc *raw, stbi__uint32 raw_len, int out_n, stbi__uint32 x, stbi__uint32 y, int depth, int color)
4282{
4283 int bytes = (depth == 16? 2 : 1);
4284 stbi__context *s = a->s;
4285 stbi__uint32 i,j,stride = x*out_n*bytes;
4286 stbi__uint32 img_len, img_width_bytes;
4287 int k;
4288 int img_n = s->img_n; // copy it into a local for later
4289
4290 int output_bytes = out_n*bytes;
4291 int filter_bytes = img_n*bytes;
4292 int width = x;
4293
4294 STBI_ASSERT(out_n == s->img_n || out_n == s->img_n+1);
4295 a->out = (stbi_uc *) stbi__malloc_mad3(x, y, output_bytes, 0); // extra bytes to write off the end into
4296 if (!a->out) return stbi__err("outofmem", "Out of memory");
4297
4298 img_width_bytes = (((img_n * x * depth) + 7) >> 3);
4299 img_len = (img_width_bytes + 1) * y;
4300 if (s->img_x == x && s->img_y == y) {
4301 if (raw_len != img_len) return stbi__err("not enough pixels","Corrupt PNG");
4302 } else { // interlaced:
4303 if (raw_len < img_len) return stbi__err("not enough pixels","Corrupt PNG");
4304 }
4305
4306 for (j=0; j < y; ++j) {
4307 stbi_uc *cur = a->out + stride*j;
4308 stbi_uc *prior;
4309 int filter = *raw++;
4310
4311 if (filter > 4)
4312 return stbi__err("invalid filter","Corrupt PNG");
4313
4314 if (depth < 8) {
4315 STBI_ASSERT(img_width_bytes <= x);
4316 cur += x*out_n - img_width_bytes; // store output to the rightmost img_len bytes, so we can decode in place
4317 filter_bytes = 1;
4318 width = img_width_bytes;
4319 }
4320 prior = cur - stride; // bugfix: need to compute this after 'cur +=' computation above
4321
4322 // if first row, use special filter that doesn't sample previous row
4323 if (j == 0) filter = first_row_filter[filter];
4324
4325 // handle first byte explicitly
4326 for (k=0; k < filter_bytes; ++k) {
4327 switch (filter) {
4328 case STBI__F_none : cur[k] = raw[k]; break;
4329 case STBI__F_sub : cur[k] = raw[k]; break;
4330 case STBI__F_up : cur[k] = STBI__BYTECAST(raw[k] + prior[k]); break;
4331 case STBI__F_avg : cur[k] = STBI__BYTECAST(raw[k] + (prior[k]>>1)); break;
4332 case STBI__F_paeth : cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(0,prior[k],0)); break;
4333 case STBI__F_avg_first : cur[k] = raw[k]; break;
4334 case STBI__F_paeth_first: cur[k] = raw[k]; break;
4335 }
4336 }
4337
4338 if (depth == 8) {
4339 if (img_n != out_n)
4340 cur[img_n] = 255; // first pixel
4341 raw += img_n;
4342 cur += out_n;
4343 prior += out_n;
4344 } else if (depth == 16) {
4345 if (img_n != out_n) {
4346 cur[filter_bytes] = 255; // first pixel top byte
4347 cur[filter_bytes+1] = 255; // first pixel bottom byte
4348 }
4349 raw += filter_bytes;
4350 cur += output_bytes;
4351 prior += output_bytes;
4352 } else {
4353 raw += 1;
4354 cur += 1;
4355 prior += 1;
4356 }
4357
4358 // this is a little gross, so that we don't switch per-pixel or per-component
4359 if (depth < 8 || img_n == out_n) {
4360 int nk = (width - 1)*filter_bytes;
4361 #define STBI__CASE(f) \
4362 case f: \
4363 for (k=0; k < nk; ++k)
4364 switch (filter) {
4365 // "none" filter turns into a memcpy here; make that explicit.
4366 case STBI__F_none: memcpy(cur, raw, nk); break;
4367 STBI__CASE(STBI__F_sub) { cur[k] = STBI__BYTECAST(raw[k] + cur[k-filter_bytes]); } break;
4368 STBI__CASE(STBI__F_up) { cur[k] = STBI__BYTECAST(raw[k] + prior[k]); } break;
4369 STBI__CASE(STBI__F_avg) { cur[k] = STBI__BYTECAST(raw[k] + ((prior[k] + cur[k-filter_bytes])>>1)); } break;
4370 STBI__CASE(STBI__F_paeth) { cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k-filter_bytes],prior[k],prior[k-filter_bytes])); } break;
4371 STBI__CASE(STBI__F_avg_first) { cur[k] = STBI__BYTECAST(raw[k] + (cur[k-filter_bytes] >> 1)); } break;
4372 STBI__CASE(STBI__F_paeth_first) { cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k-filter_bytes],0,0)); } break;
4373 }
4374 #undef STBI__CASE
4375 raw += nk;
4376 } else {
4377 STBI_ASSERT(img_n+1 == out_n);
4378 #define STBI__CASE(f) \
4379 case f: \
4380 for (i=x-1; i >= 1; --i, cur[filter_bytes]=255,raw+=filter_bytes,cur+=output_bytes,prior+=output_bytes) \
4381 for (k=0; k < filter_bytes; ++k)
4382 switch (filter) {
4383 STBI__CASE(STBI__F_none) { cur[k] = raw[k]; } break;
4384 STBI__CASE(STBI__F_sub) { cur[k] = STBI__BYTECAST(raw[k] + cur[k- output_bytes]); } break;
4385 STBI__CASE(STBI__F_up) { cur[k] = STBI__BYTECAST(raw[k] + prior[k]); } break;
4386 STBI__CASE(STBI__F_avg) { cur[k] = STBI__BYTECAST(raw[k] + ((prior[k] + cur[k- output_bytes])>>1)); } break;
4387 STBI__CASE(STBI__F_paeth) { cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k- output_bytes],prior[k],prior[k- output_bytes])); } break;
4388 STBI__CASE(STBI__F_avg_first) { cur[k] = STBI__BYTECAST(raw[k] + (cur[k- output_bytes] >> 1)); } break;
4389 STBI__CASE(STBI__F_paeth_first) { cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k- output_bytes],0,0)); } break;
4390 }
4391 #undef STBI__CASE
4392
4393 // the loop above sets the high byte of the pixels' alpha, but for
4394 // 16 bit png files we also need the low byte set. we'll do that here.
4395 if (depth == 16) {
4396 cur = a->out + stride*j; // start at the beginning of the row again
4397 for (i=0; i < x; ++i,cur+=output_bytes) {
4398 cur[filter_bytes+1] = 255;
4399 }
4400 }
4401 }
4402 }
4403
4404 // we make a separate pass to expand bits to pixels; for performance,
4405 // this could run two scanlines behind the above code, so it won't
4406 // intefere with filtering but will still be in the cache.
4407 if (depth < 8) {
4408 for (j=0; j < y; ++j) {
4409 stbi_uc *cur = a->out + stride*j;
4410 stbi_uc *in = a->out + stride*j + x*out_n - img_width_bytes;
4411 // unpack 1/2/4-bit into a 8-bit buffer. allows us to keep the common 8-bit path optimal at minimal cost for 1/2/4-bit
4412 // png guarante byte alignment, if width is not multiple of 8/4/2 we'll decode dummy trailing data that will be skipped in the later loop
4413 stbi_uc scale = (color == 0) ? stbi__depth_scale_table[depth] : 1; // scale grayscale values to 0..255 range
4414
4415 // note that the final byte might overshoot and write more data than desired.
4416 // we can allocate enough data that this never writes out of memory, but it
4417 // could also overwrite the next scanline. can it overwrite non-empty data
4418 // on the next scanline? yes, consider 1-pixel-wide scanlines with 1-bit-per-pixel.
4419 // so we need to explicitly clamp the final ones
4420
4421 if (depth == 4) {
4422 for (k=x*img_n; k >= 2; k-=2, ++in) {
4423 *cur++ = scale * ((*in >> 4) );
4424 *cur++ = scale * ((*in ) & 0x0f);
4425 }
4426 if (k > 0) *cur++ = scale * ((*in >> 4) );
4427 } else if (depth == 2) {
4428 for (k=x*img_n; k >= 4; k-=4, ++in) {
4429 *cur++ = scale * ((*in >> 6) );
4430 *cur++ = scale * ((*in >> 4) & 0x03);
4431 *cur++ = scale * ((*in >> 2) & 0x03);
4432 *cur++ = scale * ((*in ) & 0x03);
4433 }
4434 if (k > 0) *cur++ = scale * ((*in >> 6) );
4435 if (k > 1) *cur++ = scale * ((*in >> 4) & 0x03);
4436 if (k > 2) *cur++ = scale * ((*in >> 2) & 0x03);
4437 } else if (depth == 1) {
4438 for (k=x*img_n; k >= 8; k-=8, ++in) {
4439 *cur++ = scale * ((*in >> 7) );
4440 *cur++ = scale * ((*in >> 6) & 0x01);
4441 *cur++ = scale * ((*in >> 5) & 0x01);
4442 *cur++ = scale * ((*in >> 4) & 0x01);
4443 *cur++ = scale * ((*in >> 3) & 0x01);
4444 *cur++ = scale * ((*in >> 2) & 0x01);
4445 *cur++ = scale * ((*in >> 1) & 0x01);
4446 *cur++ = scale * ((*in ) & 0x01);
4447 }
4448 if (k > 0) *cur++ = scale * ((*in >> 7) );
4449 if (k > 1) *cur++ = scale * ((*in >> 6) & 0x01);
4450 if (k > 2) *cur++ = scale * ((*in >> 5) & 0x01);
4451 if (k > 3) *cur++ = scale * ((*in >> 4) & 0x01);
4452 if (k > 4) *cur++ = scale * ((*in >> 3) & 0x01);
4453 if (k > 5) *cur++ = scale * ((*in >> 2) & 0x01);
4454 if (k > 6) *cur++ = scale * ((*in >> 1) & 0x01);
4455 }
4456 if (img_n != out_n) {
4457 int q;
4458 // insert alpha = 255
4459 cur = a->out + stride*j;
4460 if (img_n == 1) {
4461 for (q=x-1; q >= 0; --q) {
4462 cur[q*2+1] = 255;
4463 cur[q*2+0] = cur[q];
4464 }
4465 } else {
4466 STBI_ASSERT(img_n == 3);
4467 for (q=x-1; q >= 0; --q) {
4468 cur[q*4+3] = 255;
4469 cur[q*4+2] = cur[q*3+2];
4470 cur[q*4+1] = cur[q*3+1];
4471 cur[q*4+0] = cur[q*3+0];
4472 }
4473 }
4474 }
4475 }
4476 } else if (depth == 16) {
4477 // force the image data from big-endian to platform-native.
4478 // this is done in a separate pass due to the decoding relying
4479 // on the data being untouched, but could probably be done
4480 // per-line during decode if care is taken.
4481 stbi_uc *cur = a->out;
4482 stbi__uint16 *cur16 = (stbi__uint16*)cur;
4483
4484 for(i=0; i < x*y*out_n; ++i,cur16++,cur+=2) {
4485 *cur16 = (cur[0] << 8) | cur[1];
4486 }
4487 }
4488
4489 return 1;
4490}
4491
4492static int stbi__create_png_image(stbi__png *a, stbi_uc *image_data, stbi__uint32 image_data_len, int out_n, int depth, int color, int interlaced)
4493{
4494 int bytes = (depth == 16 ? 2 : 1);
4495 int out_bytes = out_n * bytes;
4496 stbi_uc *final;
4497 int p;
4498 if (!interlaced)
4499 return stbi__create_png_image_raw(a, image_data, image_data_len, out_n, a->s->img_x, a->s->img_y, depth, color);
4500
4501 // de-interlacing
4502 final = (stbi_uc *) stbi__malloc_mad3(a->s->img_x, a->s->img_y, out_bytes, 0);
4503 for (p=0; p < 7; ++p) {
4504 int xorig[] = { 0,4,0,2,0,1,0 };
4505 int yorig[] = { 0,0,4,0,2,0,1 };
4506 int xspc[] = { 8,8,4,4,2,2,1 };
4507 int yspc[] = { 8,8,8,4,4,2,2 };
4508 int i,j,x,y;
4509 // pass1_x[4] = 0, pass1_x[5] = 1, pass1_x[12] = 1
4510 x = (a->s->img_x - xorig[p] + xspc[p]-1) / xspc[p];
4511 y = (a->s->img_y - yorig[p] + yspc[p]-1) / yspc[p];
4512 if (x && y) {
4513 stbi__uint32 img_len = ((((a->s->img_n * x * depth) + 7) >> 3) + 1) * y;
4514 if (!stbi__create_png_image_raw(a, image_data, image_data_len, out_n, x, y, depth, color)) {
4515 STBI_FREE(final);
4516 return 0;
4517 }
4518 for (j=0; j < y; ++j) {
4519 for (i=0; i < x; ++i) {
4520 int out_y = j*yspc[p]+yorig[p];
4521 int out_x = i*xspc[p]+xorig[p];
4522 memcpy(final + out_y*a->s->img_x*out_bytes + out_x*out_bytes,
4523 a->out + (j*x+i)*out_bytes, out_bytes);
4524 }
4525 }
4526 STBI_FREE(a->out);
4527 image_data += img_len;
4528 image_data_len -= img_len;
4529 }
4530 }
4531 a->out = final;
4532
4533 return 1;
4534}
4535
4536static int stbi__compute_transparency(stbi__png *z, stbi_uc tc[3], int out_n)
4537{
4538 stbi__context *s = z->s;
4539 stbi__uint32 i, pixel_count = s->img_x * s->img_y;
4540 stbi_uc *p = z->out;
4541
4542 // compute color-based transparency, assuming we've
4543 // already got 255 as the alpha value in the output
4544 STBI_ASSERT(out_n == 2 || out_n == 4);
4545
4546 if (out_n == 2) {
4547 for (i=0; i < pixel_count; ++i) {
4548 p[1] = (p[0] == tc[0] ? 0 : 255);
4549 p += 2;
4550 }
4551 } else {
4552 for (i=0; i < pixel_count; ++i) {
4553 if (p[0] == tc[0] && p[1] == tc[1] && p[2] == tc[2])
4554 p[3] = 0;
4555 p += 4;
4556 }
4557 }
4558 return 1;
4559}
4560
4561static int stbi__compute_transparency16(stbi__png *z, stbi__uint16 tc[3], int out_n)
4562{
4563 stbi__context *s = z->s;
4564 stbi__uint32 i, pixel_count = s->img_x * s->img_y;
4565 stbi__uint16 *p = (stbi__uint16*) z->out;
4566
4567 // compute color-based transparency, assuming we've
4568 // already got 65535 as the alpha value in the output
4569 STBI_ASSERT(out_n == 2 || out_n == 4);
4570
4571 if (out_n == 2) {
4572 for (i = 0; i < pixel_count; ++i) {
4573 p[1] = (p[0] == tc[0] ? 0 : 65535);
4574 p += 2;
4575 }
4576 } else {
4577 for (i = 0; i < pixel_count; ++i) {
4578 if (p[0] == tc[0] && p[1] == tc[1] && p[2] == tc[2])
4579 p[3] = 0;
4580 p += 4;
4581 }
4582 }
4583 return 1;
4584}
4585
4586static int stbi__expand_png_palette(stbi__png *a, stbi_uc *palette, int len, int pal_img_n)
4587{
4588 stbi__uint32 i, pixel_count = a->s->img_x * a->s->img_y;
4589 stbi_uc *p, *temp_out, *orig = a->out;
4590
4591 p = (stbi_uc *) stbi__malloc_mad2(pixel_count, pal_img_n, 0);
4592 if (p == NULL) return stbi__err("outofmem", "Out of memory");
4593
4594 // between here and free(out) below, exitting would leak
4595 temp_out = p;
4596
4597 if (pal_img_n == 3) {
4598 for (i=0; i < pixel_count; ++i) {
4599 int n = orig[i]*4;
4600 p[0] = palette[n ];
4601 p[1] = palette[n+1];
4602 p[2] = palette[n+2];
4603 p += 3;
4604 }
4605 } else {
4606 for (i=0; i < pixel_count; ++i) {
4607 int n = orig[i]*4;
4608 p[0] = palette[n ];
4609 p[1] = palette[n+1];
4610 p[2] = palette[n+2];
4611 p[3] = palette[n+3];
4612 p += 4;
4613 }
4614 }
4615 STBI_FREE(a->out);
4616 a->out = temp_out;
4617
4618 STBI_NOTUSED(len);
4619
4620 return 1;
4621}
4622
4623static int stbi__unpremultiply_on_load = 0;
4624static int stbi__de_iphone_flag = 0;
4625
4626STBIDEF void stbi_set_unpremultiply_on_load(int flag_true_if_should_unpremultiply)
4627{
4628 stbi__unpremultiply_on_load = flag_true_if_should_unpremultiply;
4629}
4630
4631STBIDEF void stbi_convert_iphone_png_to_rgb(int flag_true_if_should_convert)
4632{
4633 stbi__de_iphone_flag = flag_true_if_should_convert;
4634}
4635
4636static void stbi__de_iphone(stbi__png *z)
4637{
4638 stbi__context *s = z->s;
4639 stbi__uint32 i, pixel_count = s->img_x * s->img_y;
4640 stbi_uc *p = z->out;
4641
4642 if (s->img_out_n == 3) { // convert bgr to rgb
4643 for (i=0; i < pixel_count; ++i) {
4644 stbi_uc t = p[0];
4645 p[0] = p[2];
4646 p[2] = t;
4647 p += 3;
4648 }
4649 } else {
4650 STBI_ASSERT(s->img_out_n == 4);
4651 if (stbi__unpremultiply_on_load) {
4652 // convert bgr to rgb and unpremultiply
4653 for (i=0; i < pixel_count; ++i) {
4654 stbi_uc a = p[3];
4655 stbi_uc t = p[0];
4656 if (a) {
4657 p[0] = p[2] * 255 / a;
4658 p[1] = p[1] * 255 / a;
4659 p[2] = t * 255 / a;
4660 } else {
4661 p[0] = p[2];
4662 p[2] = t;
4663 }
4664 p += 4;
4665 }
4666 } else {
4667 // convert bgr to rgb
4668 for (i=0; i < pixel_count; ++i) {
4669 stbi_uc t = p[0];
4670 p[0] = p[2];
4671 p[2] = t;
4672 p += 4;
4673 }
4674 }
4675 }
4676}
4677
4678#define STBI__PNG_TYPE(a,b,c,d) (((a) << 24) + ((b) << 16) + ((c) << 8) + (d))
4679
4680static int stbi__parse_png_file(stbi__png *z, int scan, int req_comp)
4681{
4682 stbi_uc palette[1024], pal_img_n=0;
4683 stbi_uc has_trans=0, tc[3];
4684 stbi__uint16 tc16[3];
4685 stbi__uint32 ioff=0, idata_limit=0, i, pal_len=0;
4686 int first=1,k,interlace=0, color=0, is_iphone=0;
4687 stbi__context *s = z->s;
4688
4689 z->expanded = NULL;
4690 z->idata = NULL;
4691 z->out = NULL;
4692
4693 if (!stbi__check_png_header(s)) return 0;
4694
4695 if (scan == STBI__SCAN_type) return 1;
4696
4697 for (;;) {
4698 stbi__pngchunk c = stbi__get_chunk_header(s);
4699 switch (c.type) {
4700 case STBI__PNG_TYPE('C','g','B','I'):
4701 is_iphone = 1;
4702 stbi__skip(s, c.length);
4703 break;
4704 case STBI__PNG_TYPE('I','H','D','R'): {
4705 int comp,filter;
4706 if (!first) return stbi__err("multiple IHDR","Corrupt PNG");
4707 first = 0;
4708 if (c.length != 13) return stbi__err("bad IHDR len","Corrupt PNG");
4709 s->img_x = stbi__get32be(s); if (s->img_x > (1 << 24)) return stbi__err("too large","Very large image (corrupt?)");
4710 s->img_y = stbi__get32be(s); if (s->img_y > (1 << 24)) return stbi__err("too large","Very large image (corrupt?)");
4711 z->depth = stbi__get8(s); if (z->depth != 1 && z->depth != 2 && z->depth != 4 && z->depth != 8 && z->depth != 16) return stbi__err("1/2/4/8/16-bit only","PNG not supported: 1/2/4/8/16-bit only");
4712 color = stbi__get8(s); if (color > 6) return stbi__err("bad ctype","Corrupt PNG");
4713 if (color == 3 && z->depth == 16) return stbi__err("bad ctype","Corrupt PNG");
4714 if (color == 3) pal_img_n = 3; else if (color & 1) return stbi__err("bad ctype","Corrupt PNG");
4715 comp = stbi__get8(s); if (comp) return stbi__err("bad comp method","Corrupt PNG");
4716 filter= stbi__get8(s); if (filter) return stbi__err("bad filter method","Corrupt PNG");
4717 interlace = stbi__get8(s); if (interlace>1) return stbi__err("bad interlace method","Corrupt PNG");
4718 if (!s->img_x || !s->img_y) return stbi__err("0-pixel image","Corrupt PNG");
4719 if (!pal_img_n) {
4720 s->img_n = (color & 2 ? 3 : 1) + (color & 4 ? 1 : 0);
4721 if ((1 << 30) / s->img_x / s->img_n < s->img_y) return stbi__err("too large", "Image too large to decode");
4722 if (scan == STBI__SCAN_header) return 1;
4723 } else {
4724 // if paletted, then pal_n is our final components, and
4725 // img_n is # components to decompress/filter.
4726 s->img_n = 1;
4727 if ((1 << 30) / s->img_x / 4 < s->img_y) return stbi__err("too large","Corrupt PNG");
4728 // if SCAN_header, have to scan to see if we have a tRNS
4729 }
4730 break;
4731 }
4732
4733 case STBI__PNG_TYPE('P','L','T','E'): {
4734 if (first) return stbi__err("first not IHDR", "Corrupt PNG");
4735 if (c.length > 256*3) return stbi__err("invalid PLTE","Corrupt PNG");
4736 pal_len = c.length / 3;
4737 if (pal_len * 3 != c.length) return stbi__err("invalid PLTE","Corrupt PNG");
4738 for (i=0; i < pal_len; ++i) {
4739 palette[i*4+0] = stbi__get8(s);
4740 palette[i*4+1] = stbi__get8(s);
4741 palette[i*4+2] = stbi__get8(s);
4742 palette[i*4+3] = 255;
4743 }
4744 break;
4745 }
4746
4747 case STBI__PNG_TYPE('t','R','N','S'): {
4748 if (first) return stbi__err("first not IHDR", "Corrupt PNG");
4749 if (z->idata) return stbi__err("tRNS after IDAT","Corrupt PNG");
4750 if (pal_img_n) {
4751 if (scan == STBI__SCAN_header) { s->img_n = 4; return 1; }
4752 if (pal_len == 0) return stbi__err("tRNS before PLTE","Corrupt PNG");
4753 if (c.length > pal_len) return stbi__err("bad tRNS len","Corrupt PNG");
4754 pal_img_n = 4;
4755 for (i=0; i < c.length; ++i)
4756 palette[i*4+3] = stbi__get8(s);
4757 } else {
4758 if (!(s->img_n & 1)) return stbi__err("tRNS with alpha","Corrupt PNG");
4759 if (c.length != (stbi__uint32) s->img_n*2) return stbi__err("bad tRNS len","Corrupt PNG");
4760 has_trans = 1;
4761 if (z->depth == 16) {
4762 for (k = 0; k < s->img_n; ++k) tc16[k] = (stbi__uint16)stbi__get16be(s); // copy the values as-is
4763 } else {
4764 for (k = 0; k < s->img_n; ++k) tc[k] = (stbi_uc)(stbi__get16be(s) & 255) * stbi__depth_scale_table[z->depth]; // non 8-bit images will be larger
4765 }
4766 }
4767 break;
4768 }
4769
4770 case STBI__PNG_TYPE('I','D','A','T'): {
4771 if (first) return stbi__err("first not IHDR", "Corrupt PNG");
4772 if (pal_img_n && !pal_len) return stbi__err("no PLTE","Corrupt PNG");
4773 if (scan == STBI__SCAN_header) { s->img_n = pal_img_n; return 1; }
4774 if ((int)(ioff + c.length) < (int)ioff) return 0;
4775 if (ioff + c.length > idata_limit) {
4776 stbi__uint32 idata_limit_old = idata_limit;
4777 stbi_uc *p;
4778 if (idata_limit == 0) idata_limit = c.length > 4096 ? c.length : 4096;
4779 while (ioff + c.length > idata_limit)
4780 idata_limit *= 2;
4781 STBI_NOTUSED(idata_limit_old);
4782 p = (stbi_uc *) STBI_REALLOC_SIZED(z->idata, idata_limit_old, idata_limit); if (p == NULL) return stbi__err("outofmem", "Out of memory");
4783 z->idata = p;
4784 }
4785 if (!stbi__getn(s, z->idata+ioff,c.length)) return stbi__err("outofdata","Corrupt PNG");
4786 ioff += c.length;
4787 break;
4788 }
4789
4790 case STBI__PNG_TYPE('I','E','N','D'): {
4791 stbi__uint32 raw_len, bpl;
4792 if (first) return stbi__err("first not IHDR", "Corrupt PNG");
4793 if (scan != STBI__SCAN_load) return 1;
4794 if (z->idata == NULL) return stbi__err("no IDAT","Corrupt PNG");
4795 // initial guess for decoded data size to avoid unnecessary reallocs
4796 bpl = (s->img_x * z->depth + 7) / 8; // bytes per line, per component
4797 raw_len = bpl * s->img_y * s->img_n /* pixels */ + s->img_y /* filter mode per row */;
4798 z->expanded = (stbi_uc *) stbi_zlib_decode_malloc_guesssize_headerflag((char *) z->idata, ioff, raw_len, (int *) &raw_len, !is_iphone);
4799 if (z->expanded == NULL) return 0; // zlib should set error
4800 STBI_FREE(z->idata); z->idata = NULL;
4801 if ((req_comp == s->img_n+1 && req_comp != 3 && !pal_img_n) || has_trans)
4802 s->img_out_n = s->img_n+1;
4803 else
4804 s->img_out_n = s->img_n;
4805 if (!stbi__create_png_image(z, z->expanded, raw_len, s->img_out_n, z->depth, color, interlace)) return 0;
4806 if (has_trans) {
4807 if (z->depth == 16) {
4808 if (!stbi__compute_transparency16(z, tc16, s->img_out_n)) return 0;
4809 } else {
4810 if (!stbi__compute_transparency(z, tc, s->img_out_n)) return 0;
4811 }
4812 }
4813 if (is_iphone && stbi__de_iphone_flag && s->img_out_n > 2)
4814 stbi__de_iphone(z);
4815 if (pal_img_n) {
4816 // pal_img_n == 3 or 4
4817 s->img_n = pal_img_n; // record the actual colors we had
4818 s->img_out_n = pal_img_n;
4819 if (req_comp >= 3) s->img_out_n = req_comp;
4820 if (!stbi__expand_png_palette(z, palette, pal_len, s->img_out_n))
4821 return 0;
4822 }
4823 STBI_FREE(z->expanded); z->expanded = NULL;
4824 return 1;
4825 }
4826
4827 default:
4828 // if critical, fail
4829 if (first) return stbi__err("first not IHDR", "Corrupt PNG");
4830 if ((c.type & (1 << 29)) == 0) {
4831 #ifndef STBI_NO_FAILURE_STRINGS
4832 // not threadsafe
4833 static char invalid_chunk[] = "XXXX PNG chunk not known";
4834 invalid_chunk[0] = STBI__BYTECAST(c.type >> 24);
4835 invalid_chunk[1] = STBI__BYTECAST(c.type >> 16);
4836 invalid_chunk[2] = STBI__BYTECAST(c.type >> 8);
4837 invalid_chunk[3] = STBI__BYTECAST(c.type >> 0);
4838 #endif
4839 return stbi__err(invalid_chunk, "PNG not supported: unknown PNG chunk type");
4840 }
4841 stbi__skip(s, c.length);
4842 break;
4843 }
4844 // end of PNG chunk, read and skip CRC
4845 stbi__get32be(s);
4846 }
4847}
4848
4849static void *stbi__do_png(stbi__png *p, int *x, int *y, int *n, int req_comp, stbi__result_info *ri)
4850{
4851 void *result=NULL;
4852 if (req_comp < 0 || req_comp > 4) return stbi__errpuc("bad req_comp", "Internal error");
4853 if (stbi__parse_png_file(p, STBI__SCAN_load, req_comp)) {
4854 if (p->depth < 8)
4855 ri->bits_per_channel = 8;
4856 else
4857 ri->bits_per_channel = p->depth;
4858 result = p->out;
4859 p->out = NULL;
4860 if (req_comp && req_comp != p->s->img_out_n) {
4861 if (ri->bits_per_channel == 8)
4862 result = stbi__convert_format((unsigned char *) result, p->s->img_out_n, req_comp, p->s->img_x, p->s->img_y);
4863 else
4864 result = stbi__convert_format16((stbi__uint16 *) result, p->s->img_out_n, req_comp, p->s->img_x, p->s->img_y);
4865 p->s->img_out_n = req_comp;
4866 if (result == NULL) return result;
4867 }
4868 *x = p->s->img_x;
4869 *y = p->s->img_y;
4870 if (n) *n = p->s->img_n;
4871 }
4872 STBI_FREE(p->out); p->out = NULL;
4873 STBI_FREE(p->expanded); p->expanded = NULL;
4874 STBI_FREE(p->idata); p->idata = NULL;
4875
4876 return result;
4877}
4878
4879static void *stbi__png_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
4880{
4881 stbi__png p;
4882 p.s = s;
4883 return stbi__do_png(&p, x,y,comp,req_comp, ri);
4884}
4885
4886static int stbi__png_test(stbi__context *s)
4887{
4888 int r;
4889 r = stbi__check_png_header(s);
4890 stbi__rewind(s);
4891 return r;
4892}
4893
4894static int stbi__png_info_raw(stbi__png *p, int *x, int *y, int *comp)
4895{
4896 if (!stbi__parse_png_file(p, STBI__SCAN_header, 0)) {
4897 stbi__rewind( p->s );
4898 return 0;
4899 }
4900 if (x) *x = p->s->img_x;
4901 if (y) *y = p->s->img_y;
4902 if (comp) *comp = p->s->img_n;
4903 return 1;
4904}
4905
4906static int stbi__png_info(stbi__context *s, int *x, int *y, int *comp)
4907{
4908 stbi__png p;
4909 p.s = s;
4910 return stbi__png_info_raw(&p, x, y, comp);
4911}
4912#endif
4913
4914// Microsoft/Windows BMP image
4915
4916#ifndef STBI_NO_BMP
4917static int stbi__bmp_test_raw(stbi__context *s)
4918{
4919 int r;
4920 int sz;
4921 if (stbi__get8(s) != 'B') return 0;
4922 if (stbi__get8(s) != 'M') return 0;
4923 stbi__get32le(s); // discard filesize
4924 stbi__get16le(s); // discard reserved
4925 stbi__get16le(s); // discard reserved
4926 stbi__get32le(s); // discard data offset
4927 sz = stbi__get32le(s);
4928 r = (sz == 12 || sz == 40 || sz == 56 || sz == 108 || sz == 124);
4929 return r;
4930}
4931
4932static int stbi__bmp_test(stbi__context *s)
4933{
4934 int r = stbi__bmp_test_raw(s);
4935 stbi__rewind(s);
4936 return r;
4937}
4938
4939
4940// returns 0..31 for the highest set bit
4941static int stbi__high_bit(unsigned int z)
4942{
4943 int n=0;
4944 if (z == 0) return -1;
4945 if (z >= 0x10000) n += 16, z >>= 16;
4946 if (z >= 0x00100) n += 8, z >>= 8;
4947 if (z >= 0x00010) n += 4, z >>= 4;
4948 if (z >= 0x00004) n += 2, z >>= 2;
4949 if (z >= 0x00002) n += 1, z >>= 1;
4950 return n;
4951}
4952
4953static int stbi__bitcount(unsigned int a)
4954{
4955 a = (a & 0x55555555) + ((a >> 1) & 0x55555555); // max 2
4956 a = (a & 0x33333333) + ((a >> 2) & 0x33333333); // max 4
4957 a = (a + (a >> 4)) & 0x0f0f0f0f; // max 8 per 4, now 8 bits
4958 a = (a + (a >> 8)); // max 16 per 8 bits
4959 a = (a + (a >> 16)); // max 32 per 8 bits
4960 return a & 0xff;
4961}
4962
4963static int stbi__shiftsigned(int v, int shift, int bits)
4964{
4965 int result;
4966 int z=0;
4967
4968 if (shift < 0) v <<= -shift;
4969 else v >>= shift;
4970 result = v;
4971
4972 z = bits;
4973 while (z < 8) {
4974 result += v >> z;
4975 z += bits;
4976 }
4977 return result;
4978}
4979
4980typedef struct
4981{
4982 int bpp, offset, hsz;
4983 unsigned int mr,mg,mb,ma, all_a;
4984} stbi__bmp_data;
4985
4986static void *stbi__bmp_parse_header(stbi__context *s, stbi__bmp_data *info)
4987{
4988 int hsz;
4989 if (stbi__get8(s) != 'B' || stbi__get8(s) != 'M') return stbi__errpuc("not BMP", "Corrupt BMP");
4990 stbi__get32le(s); // discard filesize
4991 stbi__get16le(s); // discard reserved
4992 stbi__get16le(s); // discard reserved
4993 info->offset = stbi__get32le(s);
4994 info->hsz = hsz = stbi__get32le(s);
4995 info->mr = info->mg = info->mb = info->ma = 0;
4996
4997 if (hsz != 12 && hsz != 40 && hsz != 56 && hsz != 108 && hsz != 124) return stbi__errpuc("unknown BMP", "BMP type not supported: unknown");
4998 if (hsz == 12) {
4999 s->img_x = stbi__get16le(s);
5000 s->img_y = stbi__get16le(s);
5001 } else {
5002 s->img_x = stbi__get32le(s);
5003 s->img_y = stbi__get32le(s);
5004 }
5005 if (stbi__get16le(s) != 1) return stbi__errpuc("bad BMP", "bad BMP");
5006 info->bpp = stbi__get16le(s);
5007 if (info->bpp == 1) return stbi__errpuc("monochrome", "BMP type not supported: 1-bit");
5008 if (hsz != 12) {
5009 int compress = stbi__get32le(s);
5010 if (compress == 1 || compress == 2) return stbi__errpuc("BMP RLE", "BMP type not supported: RLE");
5011 stbi__get32le(s); // discard sizeof
5012 stbi__get32le(s); // discard hres
5013 stbi__get32le(s); // discard vres
5014 stbi__get32le(s); // discard colorsused
5015 stbi__get32le(s); // discard max important
5016 if (hsz == 40 || hsz == 56) {
5017 if (hsz == 56) {
5018 stbi__get32le(s);
5019 stbi__get32le(s);
5020 stbi__get32le(s);
5021 stbi__get32le(s);
5022 }
5023 if (info->bpp == 16 || info->bpp == 32) {
5024 if (compress == 0) {
5025 if (info->bpp == 32) {
5026 info->mr = 0xffu << 16;
5027 info->mg = 0xffu << 8;
5028 info->mb = 0xffu << 0;
5029 info->ma = 0xffu << 24;
5030 info->all_a = 0; // if all_a is 0 at end, then we loaded alpha channel but it was all 0
5031 } else {
5032 info->mr = 31u << 10;
5033 info->mg = 31u << 5;
5034 info->mb = 31u << 0;
5035 }
5036 } else if (compress == 3) {
5037 info->mr = stbi__get32le(s);
5038 info->mg = stbi__get32le(s);
5039 info->mb = stbi__get32le(s);
5040 // not documented, but generated by photoshop and handled by mspaint
5041 if (info->mr == info->mg && info->mg == info->mb) {
5042 // ?!?!?
5043 return stbi__errpuc("bad BMP", "bad BMP");
5044 }
5045 } else
5046 return stbi__errpuc("bad BMP", "bad BMP");
5047 }
5048 } else {
5049 int i;
5050 if (hsz != 108 && hsz != 124)
5051 return stbi__errpuc("bad BMP", "bad BMP");
5052 info->mr = stbi__get32le(s);
5053 info->mg = stbi__get32le(s);
5054 info->mb = stbi__get32le(s);
5055 info->ma = stbi__get32le(s);
5056 stbi__get32le(s); // discard color space
5057 for (i=0; i < 12; ++i)
5058 stbi__get32le(s); // discard color space parameters
5059 if (hsz == 124) {
5060 stbi__get32le(s); // discard rendering intent
5061 stbi__get32le(s); // discard offset of profile data
5062 stbi__get32le(s); // discard size of profile data
5063 stbi__get32le(s); // discard reserved
5064 }
5065 }
5066 }
5067 return (void *) 1;
5068}
5069
5070
5071static void *stbi__bmp_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
5072{
5073 stbi_uc *out;
5074 unsigned int mr=0,mg=0,mb=0,ma=0, all_a;
5075 stbi_uc pal[256][4];
5076 int psize=0,i,j,width;
5077 int flip_vertically, pad, target;
5078 stbi__bmp_data info;
5079 STBI_NOTUSED(ri);
5080
5081 info.all_a = 255;
5082 if (stbi__bmp_parse_header(s, &info) == NULL)
5083 return NULL; // error code already set
5084
5085 flip_vertically = ((int) s->img_y) > 0;
5086 s->img_y = abs((int) s->img_y);
5087
5088 mr = info.mr;
5089 mg = info.mg;
5090 mb = info.mb;
5091 ma = info.ma;
5092 all_a = info.all_a;
5093
5094 if (info.hsz == 12) {
5095 if (info.bpp < 24)
5096 psize = (info.offset - 14 - 24) / 3;
5097 } else {
5098 if (info.bpp < 16)
5099 psize = (info.offset - 14 - info.hsz) >> 2;
5100 }
5101
5102 s->img_n = ma ? 4 : 3;
5103 if (req_comp && req_comp >= 3) // we can directly decode 3 or 4
5104 target = req_comp;
5105 else
5106 target = s->img_n; // if they want monochrome, we'll post-convert
5107
5108 // sanity-check size
5109 if (!stbi__mad3sizes_valid(target, s->img_x, s->img_y, 0))
5110 return stbi__errpuc("too large", "Corrupt BMP");
5111
5112 out = (stbi_uc *) stbi__malloc_mad3(target, s->img_x, s->img_y, 0);
5113 if (!out) return stbi__errpuc("outofmem", "Out of memory");
5114 if (info.bpp < 16) {
5115 int z=0;
5116 if (psize == 0 || psize > 256) { STBI_FREE(out); return stbi__errpuc("invalid", "Corrupt BMP"); }
5117 for (i=0; i < psize; ++i) {
5118 pal[i][2] = stbi__get8(s);
5119 pal[i][1] = stbi__get8(s);
5120 pal[i][0] = stbi__get8(s);
5121 if (info.hsz != 12) stbi__get8(s);
5122 pal[i][3] = 255;
5123 }
5124 stbi__skip(s, info.offset - 14 - info.hsz - psize * (info.hsz == 12 ? 3 : 4));
5125 if (info.bpp == 4) width = (s->img_x + 1) >> 1;
5126 else if (info.bpp == 8) width = s->img_x;
5127 else { STBI_FREE(out); return stbi__errpuc("bad bpp", "Corrupt BMP"); }
5128 pad = (-width)&3;
5129 for (j=0; j < (int) s->img_y; ++j) {
5130 for (i=0; i < (int) s->img_x; i += 2) {
5131 int v=stbi__get8(s),v2=0;
5132 if (info.bpp == 4) {
5133 v2 = v & 15;
5134 v >>= 4;
5135 }
5136 out[z++] = pal[v][0];
5137 out[z++] = pal[v][1];
5138 out[z++] = pal[v][2];
5139 if (target == 4) out[z++] = 255;
5140 if (i+1 == (int) s->img_x) break;
5141 v = (info.bpp == 8) ? stbi__get8(s) : v2;
5142 out[z++] = pal[v][0];
5143 out[z++] = pal[v][1];
5144 out[z++] = pal[v][2];
5145 if (target == 4) out[z++] = 255;
5146 }
5147 stbi__skip(s, pad);
5148 }
5149 } else {
5150 int rshift=0,gshift=0,bshift=0,ashift=0,rcount=0,gcount=0,bcount=0,acount=0;
5151 int z = 0;
5152 int easy=0;
5153 stbi__skip(s, info.offset - 14 - info.hsz);
5154 if (info.bpp == 24) width = 3 * s->img_x;
5155 else if (info.bpp == 16) width = 2*s->img_x;
5156 else /* bpp = 32 and pad = 0 */ width=0;
5157 pad = (-width) & 3;
5158 if (info.bpp == 24) {
5159 easy = 1;
5160 } else if (info.bpp == 32) {
5161 if (mb == 0xff && mg == 0xff00 && mr == 0x00ff0000 && ma == 0xff000000)
5162 easy = 2;
5163 }
5164 if (!easy) {
5165 if (!mr || !mg || !mb) { STBI_FREE(out); return stbi__errpuc("bad masks", "Corrupt BMP"); }
5166 // right shift amt to put high bit in position #7
5167 rshift = stbi__high_bit(mr)-7; rcount = stbi__bitcount(mr);
5168 gshift = stbi__high_bit(mg)-7; gcount = stbi__bitcount(mg);
5169 bshift = stbi__high_bit(mb)-7; bcount = stbi__bitcount(mb);
5170 ashift = stbi__high_bit(ma)-7; acount = stbi__bitcount(ma);
5171 }
5172 for (j=0; j < (int) s->img_y; ++j) {
5173 if (easy) {
5174 for (i=0; i < (int) s->img_x; ++i) {
5175 unsigned char a;
5176 out[z+2] = stbi__get8(s);
5177 out[z+1] = stbi__get8(s);
5178 out[z+0] = stbi__get8(s);
5179 z += 3;
5180 a = (easy == 2 ? stbi__get8(s) : 255);
5181 all_a |= a;
5182 if (target == 4) out[z++] = a;
5183 }
5184 } else {
5185 int bpp = info.bpp;
5186 for (i=0; i < (int) s->img_x; ++i) {
5187 stbi__uint32 v = (bpp == 16 ? (stbi__uint32) stbi__get16le(s) : stbi__get32le(s));
5188 int a;
5189 out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mr, rshift, rcount));
5190 out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mg, gshift, gcount));
5191 out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mb, bshift, bcount));
5192 a = (ma ? stbi__shiftsigned(v & ma, ashift, acount) : 255);
5193 all_a |= a;
5194 if (target == 4) out[z++] = STBI__BYTECAST(a);
5195 }
5196 }
5197 stbi__skip(s, pad);
5198 }
5199 }
5200
5201 // if alpha channel is all 0s, replace with all 255s
5202 if (target == 4 && all_a == 0)
5203 for (i=4*s->img_x*s->img_y-1; i >= 0; i -= 4)
5204 out[i] = 255;
5205
5206 if (flip_vertically) {
5207 stbi_uc t;
5208 for (j=0; j < (int) s->img_y>>1; ++j) {
5209 stbi_uc *p1 = out + j *s->img_x*target;
5210 stbi_uc *p2 = out + (s->img_y-1-j)*s->img_x*target;
5211 for (i=0; i < (int) s->img_x*target; ++i) {
5212 t = p1[i], p1[i] = p2[i], p2[i] = t;
5213 }
5214 }
5215 }
5216
5217 if (req_comp && req_comp != target) {
5218 out = stbi__convert_format(out, target, req_comp, s->img_x, s->img_y);
5219 if (out == NULL) return out; // stbi__convert_format frees input on failure
5220 }
5221
5222 *x = s->img_x;
5223 *y = s->img_y;
5224 if (comp) *comp = s->img_n;
5225 return out;
5226}
5227#endif
5228
5229// Targa Truevision - TGA
5230// by Jonathan Dummer
5231#ifndef STBI_NO_TGA
5232// returns STBI_rgb or whatever, 0 on error
5233static int stbi__tga_get_comp(int bits_per_pixel, int is_grey, int* is_rgb16)
5234{
5235 // only RGB or RGBA (incl. 16bit) or grey allowed
5236 if(is_rgb16) *is_rgb16 = 0;
5237 switch(bits_per_pixel) {
5238 case 8: return STBI_grey;
5239 case 16: if(is_grey) return STBI_grey_alpha;
5240 // else: fall-through
5241 case 15: if(is_rgb16) *is_rgb16 = 1;
5242 return STBI_rgb;
5243 case 24: // fall-through
5244 case 32: return bits_per_pixel/8;
5245 default: return 0;
5246 }
5247}
5248
5249static int stbi__tga_info(stbi__context *s, int *x, int *y, int *comp)
5250{
5251 int tga_w, tga_h, tga_comp, tga_image_type, tga_bits_per_pixel, tga_colormap_bpp;
5252 int sz, tga_colormap_type;
5253 stbi__get8(s); // discard Offset
5254 tga_colormap_type = stbi__get8(s); // colormap type
5255 if( tga_colormap_type > 1 ) {
5256 stbi__rewind(s);
5257 return 0; // only RGB or indexed allowed
5258 }
5259 tga_image_type = stbi__get8(s); // image type
5260 if ( tga_colormap_type == 1 ) { // colormapped (paletted) image
5261 if (tga_image_type != 1 && tga_image_type != 9) {
5262 stbi__rewind(s);
5263 return 0;
5264 }
5265 stbi__skip(s,4); // skip index of first colormap entry and number of entries
5266 sz = stbi__get8(s); // check bits per palette color entry
5267 if ( (sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32) ) {
5268 stbi__rewind(s);
5269 return 0;
5270 }
5271 stbi__skip(s,4); // skip image x and y origin
5272 tga_colormap_bpp = sz;
5273 } else { // "normal" image w/o colormap - only RGB or grey allowed, +/- RLE
5274 if ( (tga_image_type != 2) && (tga_image_type != 3) && (tga_image_type != 10) && (tga_image_type != 11) ) {
5275 stbi__rewind(s);
5276 return 0; // only RGB or grey allowed, +/- RLE
5277 }
5278 stbi__skip(s,9); // skip colormap specification and image x/y origin
5279 tga_colormap_bpp = 0;
5280 }
5281 tga_w = stbi__get16le(s);
5282 if( tga_w < 1 ) {
5283 stbi__rewind(s);
5284 return 0; // test width
5285 }
5286 tga_h = stbi__get16le(s);
5287 if( tga_h < 1 ) {
5288 stbi__rewind(s);
5289 return 0; // test height
5290 }
5291 tga_bits_per_pixel = stbi__get8(s); // bits per pixel
5292 stbi__get8(s); // ignore alpha bits
5293 if (tga_colormap_bpp != 0) {
5294 if((tga_bits_per_pixel != 8) && (tga_bits_per_pixel != 16)) {
5295 // when using a colormap, tga_bits_per_pixel is the size of the indexes
5296 // I don't think anything but 8 or 16bit indexes makes sense
5297 stbi__rewind(s);
5298 return 0;
5299 }
5300 tga_comp = stbi__tga_get_comp(tga_colormap_bpp, 0, NULL);
5301 } else {
5302 tga_comp = stbi__tga_get_comp(tga_bits_per_pixel, (tga_image_type == 3) || (tga_image_type == 11), NULL);
5303 }
5304 if(!tga_comp) {
5305 stbi__rewind(s);
5306 return 0;
5307 }
5308 if (x) *x = tga_w;
5309 if (y) *y = tga_h;
5310 if (comp) *comp = tga_comp;
5311 return 1; // seems to have passed everything
5312}
5313
5314static int stbi__tga_test(stbi__context *s)
5315{
5316 int res = 0;
5317 int sz, tga_color_type;
5318 stbi__get8(s); // discard Offset
5319 tga_color_type = stbi__get8(s); // color type
5320 if ( tga_color_type > 1 ) goto errorEnd; // only RGB or indexed allowed
5321 sz = stbi__get8(s); // image type
5322 if ( tga_color_type == 1 ) { // colormapped (paletted) image
5323 if (sz != 1 && sz != 9) goto errorEnd; // colortype 1 demands image type 1 or 9
5324 stbi__skip(s,4); // skip index of first colormap entry and number of entries
5325 sz = stbi__get8(s); // check bits per palette color entry
5326 if ( (sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32) ) goto errorEnd;
5327 stbi__skip(s,4); // skip image x and y origin
5328 } else { // "normal" image w/o colormap
5329 if ( (sz != 2) && (sz != 3) && (sz != 10) && (sz != 11) ) goto errorEnd; // only RGB or grey allowed, +/- RLE
5330 stbi__skip(s,9); // skip colormap specification and image x/y origin
5331 }
5332 if ( stbi__get16le(s) < 1 ) goto errorEnd; // test width
5333 if ( stbi__get16le(s) < 1 ) goto errorEnd; // test height
5334 sz = stbi__get8(s); // bits per pixel
5335 if ( (tga_color_type == 1) && (sz != 8) && (sz != 16) ) goto errorEnd; // for colormapped images, bpp is size of an index
5336 if ( (sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32) ) goto errorEnd;
5337
5338 res = 1; // if we got this far, everything's good and we can return 1 instead of 0
5339
5340errorEnd:
5341 stbi__rewind(s);
5342 return res;
5343}
5344
5345// read 16bit value and convert to 24bit RGB
5346static void stbi__tga_read_rgb16(stbi__context *s, stbi_uc* out)
5347{
5348 stbi__uint16 px = (stbi__uint16)stbi__get16le(s);
5349 stbi__uint16 fiveBitMask = 31;
5350 // we have 3 channels with 5bits each
5351 int r = (px >> 10) & fiveBitMask;
5352 int g = (px >> 5) & fiveBitMask;
5353 int b = px & fiveBitMask;
5354 // Note that this saves the data in RGB(A) order, so it doesn't need to be swapped later
5355 out[0] = (stbi_uc)((r * 255)/31);
5356 out[1] = (stbi_uc)((g * 255)/31);
5357 out[2] = (stbi_uc)((b * 255)/31);
5358
5359 // some people claim that the most significant bit might be used for alpha
5360 // (possibly if an alpha-bit is set in the "image descriptor byte")
5361 // but that only made 16bit test images completely translucent..
5362 // so let's treat all 15 and 16bit TGAs as RGB with no alpha.
5363}
5364
5365static void *stbi__tga_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
5366{
5367 // read in the TGA header stuff
5368 int tga_offset = stbi__get8(s);
5369 int tga_indexed = stbi__get8(s);
5370 int tga_image_type = stbi__get8(s);
5371 int tga_is_RLE = 0;
5372 int tga_palette_start = stbi__get16le(s);
5373 int tga_palette_len = stbi__get16le(s);
5374 int tga_palette_bits = stbi__get8(s);
5375 int tga_x_origin = stbi__get16le(s);
5376 int tga_y_origin = stbi__get16le(s);
5377 int tga_width = stbi__get16le(s);
5378 int tga_height = stbi__get16le(s);
5379 int tga_bits_per_pixel = stbi__get8(s);
5380 int tga_comp, tga_rgb16=0;
5381 int tga_inverted = stbi__get8(s);
5382 // int tga_alpha_bits = tga_inverted & 15; // the 4 lowest bits - unused (useless?)
5383 // image data
5384 unsigned char *tga_data;
5385 unsigned char *tga_palette = NULL;
5386 int i, j;
5387 unsigned char raw_data[4] = {0};
5388 int RLE_count = 0;
5389 int RLE_repeating = 0;
5390 int read_next_pixel = 1;
5391 STBI_NOTUSED(ri);
5392
5393 // do a tiny bit of precessing
5394 if ( tga_image_type >= 8 )
5395 {
5396 tga_image_type -= 8;
5397 tga_is_RLE = 1;
5398 }
5399 tga_inverted = 1 - ((tga_inverted >> 5) & 1);
5400
5401 // If I'm paletted, then I'll use the number of bits from the palette
5402 if ( tga_indexed ) tga_comp = stbi__tga_get_comp(tga_palette_bits, 0, &tga_rgb16);
5403 else tga_comp = stbi__tga_get_comp(tga_bits_per_pixel, (tga_image_type == 3), &tga_rgb16);
5404
5405 if(!tga_comp) // shouldn't really happen, stbi__tga_test() should have ensured basic consistency
5406 return stbi__errpuc("bad format", "Can't find out TGA pixelformat");
5407
5408 // tga info
5409 *x = tga_width;
5410 *y = tga_height;
5411 if (comp) *comp = tga_comp;
5412
5413 if (!stbi__mad3sizes_valid(tga_width, tga_height, tga_comp, 0))
5414 return stbi__errpuc("too large", "Corrupt TGA");
5415
5416 tga_data = (unsigned char*)stbi__malloc_mad3(tga_width, tga_height, tga_comp, 0);
5417 if (!tga_data) return stbi__errpuc("outofmem", "Out of memory");
5418
5419 // skip to the data's starting position (offset usually = 0)
5420 stbi__skip(s, tga_offset );
5421
5422 if ( !tga_indexed && !tga_is_RLE && !tga_rgb16 ) {
5423 for (i=0; i < tga_height; ++i) {
5424 int row = tga_inverted ? tga_height -i - 1 : i;
5425 stbi_uc *tga_row = tga_data + row*tga_width*tga_comp;
5426 stbi__getn(s, tga_row, tga_width * tga_comp);
5427 }
5428 } else {
5429 // do I need to load a palette?
5430 if ( tga_indexed)
5431 {
5432 // any data to skip? (offset usually = 0)
5433 stbi__skip(s, tga_palette_start );
5434 // load the palette
5435 tga_palette = (unsigned char*)stbi__malloc_mad2(tga_palette_len, tga_comp, 0);
5436 if (!tga_palette) {
5437 STBI_FREE(tga_data);
5438 return stbi__errpuc("outofmem", "Out of memory");
5439 }
5440 if (tga_rgb16) {
5441 stbi_uc *pal_entry = tga_palette;
5442 STBI_ASSERT(tga_comp == STBI_rgb);
5443 for (i=0; i < tga_palette_len; ++i) {
5444 stbi__tga_read_rgb16(s, pal_entry);
5445 pal_entry += tga_comp;
5446 }
5447 } else if (!stbi__getn(s, tga_palette, tga_palette_len * tga_comp)) {
5448 STBI_FREE(tga_data);
5449 STBI_FREE(tga_palette);
5450 return stbi__errpuc("bad palette", "Corrupt TGA");
5451 }
5452 }
5453 // load the data
5454 for (i=0; i < tga_width * tga_height; ++i)
5455 {
5456 // if I'm in RLE mode, do I need to get a RLE stbi__pngchunk?
5457 if ( tga_is_RLE )
5458 {
5459 if ( RLE_count == 0 )
5460 {
5461 // yep, get the next byte as a RLE command
5462 int RLE_cmd = stbi__get8(s);
5463 RLE_count = 1 + (RLE_cmd & 127);
5464 RLE_repeating = RLE_cmd >> 7;
5465 read_next_pixel = 1;
5466 } else if ( !RLE_repeating )
5467 {
5468 read_next_pixel = 1;
5469 }
5470 } else
5471 {
5472 read_next_pixel = 1;
5473 }
5474 // OK, if I need to read a pixel, do it now
5475 if ( read_next_pixel )
5476 {
5477 // load however much data we did have
5478 if ( tga_indexed )
5479 {
5480 // read in index, then perform the lookup
5481 int pal_idx = (tga_bits_per_pixel == 8) ? stbi__get8(s) : stbi__get16le(s);
5482 if ( pal_idx >= tga_palette_len ) {
5483 // invalid index
5484 pal_idx = 0;
5485 }
5486 pal_idx *= tga_comp;
5487 for (j = 0; j < tga_comp; ++j) {
5488 raw_data[j] = tga_palette[pal_idx+j];
5489 }
5490 } else if(tga_rgb16) {
5491 STBI_ASSERT(tga_comp == STBI_rgb);
5492 stbi__tga_read_rgb16(s, raw_data);
5493 } else {
5494 // read in the data raw
5495 for (j = 0; j < tga_comp; ++j) {
5496 raw_data[j] = stbi__get8(s);
5497 }
5498 }
5499 // clear the reading flag for the next pixel
5500 read_next_pixel = 0;
5501 } // end of reading a pixel
5502
5503 // copy data
5504 for (j = 0; j < tga_comp; ++j)
5505 tga_data[i*tga_comp+j] = raw_data[j];
5506
5507 // in case we're in RLE mode, keep counting down
5508 --RLE_count;
5509 }
5510 // do I need to invert the image?
5511 if ( tga_inverted )
5512 {
5513 for (j = 0; j*2 < tga_height; ++j)
5514 {
5515 int index1 = j * tga_width * tga_comp;
5516 int index2 = (tga_height - 1 - j) * tga_width * tga_comp;
5517 for (i = tga_width * tga_comp; i > 0; --i)
5518 {
5519 unsigned char temp = tga_data[index1];
5520 tga_data[index1] = tga_data[index2];
5521 tga_data[index2] = temp;
5522 ++index1;
5523 ++index2;
5524 }
5525 }
5526 }
5527 // clear my palette, if I had one
5528 if ( tga_palette != NULL )
5529 {
5530 STBI_FREE( tga_palette );
5531 }
5532 }
5533
5534 // swap RGB - if the source data was RGB16, it already is in the right order
5535 if (tga_comp >= 3 && !tga_rgb16)
5536 {
5537 unsigned char* tga_pixel = tga_data;
5538 for (i=0; i < tga_width * tga_height; ++i)
5539 {
5540 unsigned char temp = tga_pixel[0];
5541 tga_pixel[0] = tga_pixel[2];
5542 tga_pixel[2] = temp;
5543 tga_pixel += tga_comp;
5544 }
5545 }
5546
5547 // convert to target component count
5548 if (req_comp && req_comp != tga_comp)
5549 tga_data = stbi__convert_format(tga_data, tga_comp, req_comp, tga_width, tga_height);
5550
5551 // the things I do to get rid of an error message, and yet keep
5552 // Microsoft's C compilers happy... [8^(
5553 tga_palette_start = tga_palette_len = tga_palette_bits =
5554 tga_x_origin = tga_y_origin = 0;
5555 // OK, done
5556 return tga_data;
5557}
5558#endif
5559
5560// *************************************************************************************************
5561// Photoshop PSD loader -- PD by Thatcher Ulrich, integration by Nicolas Schulz, tweaked by STB
5562
5563#ifndef STBI_NO_PSD
5564static int stbi__psd_test(stbi__context *s)
5565{
5566 int r = (stbi__get32be(s) == 0x38425053);
5567 stbi__rewind(s);
5568 return r;
5569}
5570
5571static int stbi__psd_decode_rle(stbi__context *s, stbi_uc *p, int pixelCount)
5572{
5573 int count, nleft, len;
5574
5575 count = 0;
5576 while ((nleft = pixelCount - count) > 0) {
5577 len = stbi__get8(s);
5578 if (len == 128) {
5579 // No-op.
5580 } else if (len < 128) {
5581 // Copy next len+1 bytes literally.
5582 len++;
5583 if (len > nleft) return 0; // corrupt data
5584 count += len;
5585 while (len) {
5586 *p = stbi__get8(s);
5587 p += 4;
5588 len--;
5589 }
5590 } else if (len > 128) {
5591 stbi_uc val;
5592 // Next -len+1 bytes in the dest are replicated from next source byte.
5593 // (Interpret len as a negative 8-bit int.)
5594 len = 257 - len;
5595 if (len > nleft) return 0; // corrupt data
5596 val = stbi__get8(s);
5597 count += len;
5598 while (len) {
5599 *p = val;
5600 p += 4;
5601 len--;
5602 }
5603 }
5604 }
5605
5606 return 1;
5607}
5608
5609static void *stbi__psd_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri, int bpc)
5610{
5611 int pixelCount;
5612 int channelCount, compression;
5613 int channel, i;
5614 int bitdepth;
5615 int w,h;
5616 stbi_uc *out;
5617 STBI_NOTUSED(ri);
5618
5619 // Check identifier
5620 if (stbi__get32be(s) != 0x38425053) // "8BPS"
5621 return stbi__errpuc("not PSD", "Corrupt PSD image");
5622
5623 // Check file type version.
5624 if (stbi__get16be(s) != 1)
5625 return stbi__errpuc("wrong version", "Unsupported version of PSD image");
5626
5627 // Skip 6 reserved bytes.
5628 stbi__skip(s, 6 );
5629
5630 // Read the number of channels (R, G, B, A, etc).
5631 channelCount = stbi__get16be(s);
5632 if (channelCount < 0 || channelCount > 16)
5633 return stbi__errpuc("wrong channel count", "Unsupported number of channels in PSD image");
5634
5635 // Read the rows and columns of the image.
5636 h = stbi__get32be(s);
5637 w = stbi__get32be(s);
5638
5639 // Make sure the depth is 8 bits.
5640 bitdepth = stbi__get16be(s);
5641 if (bitdepth != 8 && bitdepth != 16)
5642 return stbi__errpuc("unsupported bit depth", "PSD bit depth is not 8 or 16 bit");
5643
5644 // Make sure the color mode is RGB.
5645 // Valid options are:
5646 // 0: Bitmap
5647 // 1: Grayscale
5648 // 2: Indexed color
5649 // 3: RGB color
5650 // 4: CMYK color
5651 // 7: Multichannel
5652 // 8: Duotone
5653 // 9: Lab color
5654 if (stbi__get16be(s) != 3)
5655 return stbi__errpuc("wrong color format", "PSD is not in RGB color format");
5656
5657 // Skip the Mode Data. (It's the palette for indexed color; other info for other modes.)
5658 stbi__skip(s,stbi__get32be(s) );
5659
5660 // Skip the image resources. (resolution, pen tool paths, etc)
5661 stbi__skip(s, stbi__get32be(s) );
5662
5663 // Skip the reserved data.
5664 stbi__skip(s, stbi__get32be(s) );
5665
5666 // Find out if the data is compressed.
5667 // Known values:
5668 // 0: no compression
5669 // 1: RLE compressed
5670 compression = stbi__get16be(s);
5671 if (compression > 1)
5672 return stbi__errpuc("bad compression", "PSD has an unknown compression format");
5673
5674 // Check size
5675 if (!stbi__mad3sizes_valid(4, w, h, 0))
5676 return stbi__errpuc("too large", "Corrupt PSD");
5677
5678 // Create the destination image.
5679
5680 if (!compression && bitdepth == 16 && bpc == 16) {
5681 out = (stbi_uc *) stbi__malloc_mad3(8, w, h, 0);
5682 ri->bits_per_channel = 16;
5683 } else
5684 out = (stbi_uc *) stbi__malloc(4 * w*h);
5685
5686 if (!out) return stbi__errpuc("outofmem", "Out of memory");
5687 pixelCount = w*h;
5688
5689 // Initialize the data to zero.
5690 //memset( out, 0, pixelCount * 4 );
5691
5692 // Finally, the image data.
5693 if (compression) {
5694 // RLE as used by .PSD and .TIFF
5695 // Loop until you get the number of unpacked bytes you are expecting:
5696 // Read the next source byte into n.
5697 // If n is between 0 and 127 inclusive, copy the next n+1 bytes literally.
5698 // Else if n is between -127 and -1 inclusive, copy the next byte -n+1 times.
5699 // Else if n is 128, noop.
5700 // Endloop
5701
5702 // The RLE-compressed data is preceeded by a 2-byte data count for each row in the data,
5703 // which we're going to just skip.
5704 stbi__skip(s, h * channelCount * 2 );
5705
5706 // Read the RLE data by channel.
5707 for (channel = 0; channel < 4; channel++) {
5708 stbi_uc *p;
5709
5710 p = out+channel;
5711 if (channel >= channelCount) {
5712 // Fill this channel with default data.
5713 for (i = 0; i < pixelCount; i++, p += 4)
5714 *p = (channel == 3 ? 255 : 0);
5715 } else {
5716 // Read the RLE data.
5717 if (!stbi__psd_decode_rle(s, p, pixelCount)) {
5718 STBI_FREE(out);
5719 return stbi__errpuc("corrupt", "bad RLE data");
5720 }
5721 }
5722 }
5723
5724 } else {
5725 // We're at the raw image data. It's each channel in order (Red, Green, Blue, Alpha, ...)
5726 // where each channel consists of an 8-bit (or 16-bit) value for each pixel in the image.
5727
5728 // Read the data by channel.
5729 for (channel = 0; channel < 4; channel++) {
5730 if (channel >= channelCount) {
5731 // Fill this channel with default data.
5732 if (bitdepth == 16 && bpc == 16) {
5733 stbi__uint16 *q = ((stbi__uint16 *) out) + channel;
5734 stbi__uint16 val = channel == 3 ? 65535 : 0;
5735 for (i = 0; i < pixelCount; i++, q += 4)
5736 *q = val;
5737 } else {
5738 stbi_uc *p = out+channel;
5739 stbi_uc val = channel == 3 ? 255 : 0;
5740 for (i = 0; i < pixelCount; i++, p += 4)
5741 *p = val;
5742 }
5743 } else {
5744 if (ri->bits_per_channel == 16) { // output bpc
5745 stbi__uint16 *q = ((stbi__uint16 *) out) + channel;
5746 for (i = 0; i < pixelCount; i++, q += 4)
5747 *q = (stbi__uint16) stbi__get16be(s);
5748 } else {
5749 stbi_uc *p = out+channel;
5750 if (bitdepth == 16) { // input bpc
5751 for (i = 0; i < pixelCount; i++, p += 4)
5752 *p = (stbi_uc) (stbi__get16be(s) >> 8);
5753 } else {
5754 for (i = 0; i < pixelCount; i++, p += 4)
5755 *p = stbi__get8(s);
5756 }
5757 }
5758 }
5759 }
5760 }
5761
5762 // remove weird white matte from PSD
5763 if (channelCount >= 4) {
5764 if (ri->bits_per_channel == 16) {
5765 for (i=0; i < w*h; ++i) {
5766 stbi__uint16 *pixel = (stbi__uint16 *) out + 4*i;
5767 if (pixel[3] != 0 && pixel[3] != 65535) {
5768 float a = pixel[3] / 65535.0f;
5769 float ra = 1.0f / a;
5770 float inv_a = 65535.0f * (1 - ra);
5771 pixel[0] = (stbi__uint16) (pixel[0]*ra + inv_a);
5772 pixel[1] = (stbi__uint16) (pixel[1]*ra + inv_a);
5773 pixel[2] = (stbi__uint16) (pixel[2]*ra + inv_a);
5774 }
5775 }
5776 } else {
5777 for (i=0; i < w*h; ++i) {
5778 unsigned char *pixel = out + 4*i;
5779 if (pixel[3] != 0 && pixel[3] != 255) {
5780 float a = pixel[3] / 255.0f;
5781 float ra = 1.0f / a;
5782 float inv_a = 255.0f * (1 - ra);
5783 pixel[0] = (unsigned char) (pixel[0]*ra + inv_a);
5784 pixel[1] = (unsigned char) (pixel[1]*ra + inv_a);
5785 pixel[2] = (unsigned char) (pixel[2]*ra + inv_a);
5786 }
5787 }
5788 }
5789 }
5790
5791 // convert to desired output format
5792 if (req_comp && req_comp != 4) {
5793 if (ri->bits_per_channel == 16)
5794 out = (stbi_uc *) stbi__convert_format16((stbi__uint16 *) out, 4, req_comp, w, h);
5795 else
5796 out = stbi__convert_format(out, 4, req_comp, w, h);
5797 if (out == NULL) return out; // stbi__convert_format frees input on failure
5798 }
5799
5800 if (comp) *comp = 4;
5801 *y = h;
5802 *x = w;
5803
5804 return out;
5805}
5806#endif
5807
5808// *************************************************************************************************
5809// Softimage PIC loader
5810// by Tom Seddon
5811//
5812// See http://softimage.wiki.softimage.com/index.php/INFO:_PIC_file_format
5813// See http://ozviz.wasp.uwa.edu.au/~pbourke/dataformats/softimagepic/
5814
5815#ifndef STBI_NO_PIC
5816static int stbi__pic_is4(stbi__context *s,const char *str)
5817{
5818 int i;
5819 for (i=0; i<4; ++i)
5820 if (stbi__get8(s) != (stbi_uc)str[i])
5821 return 0;
5822
5823 return 1;
5824}
5825
5826static int stbi__pic_test_core(stbi__context *s)
5827{
5828 int i;
5829
5830 if (!stbi__pic_is4(s,"\x53\x80\xF6\x34"))
5831 return 0;
5832
5833 for(i=0;i<84;++i)
5834 stbi__get8(s);
5835
5836 if (!stbi__pic_is4(s,"PICT"))
5837 return 0;
5838
5839 return 1;
5840}
5841
5842typedef struct
5843{
5844 stbi_uc size,type,channel;
5845} stbi__pic_packet;
5846
5847static stbi_uc *stbi__readval(stbi__context *s, int channel, stbi_uc *dest)
5848{
5849 int mask=0x80, i;
5850
5851 for (i=0; i<4; ++i, mask>>=1) {
5852 if (channel & mask) {
5853 if (stbi__at_eof(s)) return stbi__errpuc("bad file","PIC file too short");
5854 dest[i]=stbi__get8(s);
5855 }
5856 }
5857
5858 return dest;
5859}
5860
5861static void stbi__copyval(int channel,stbi_uc *dest,const stbi_uc *src)
5862{
5863 int mask=0x80,i;
5864
5865 for (i=0;i<4; ++i, mask>>=1)
5866 if (channel&mask)
5867 dest[i]=src[i];
5868}
5869
5870static stbi_uc *stbi__pic_load_core(stbi__context *s,int width,int height,int *comp, stbi_uc *result)
5871{
5872 int act_comp=0,num_packets=0,y,chained;
5873 stbi__pic_packet packets[10];
5874
5875 // this will (should...) cater for even some bizarre stuff like having data
5876 // for the same channel in multiple packets.
5877 do {
5878 stbi__pic_packet *packet;
5879
5880 if (num_packets==sizeof(packets)/sizeof(packets[0]))
5881 return stbi__errpuc("bad format","too many packets");
5882
5883 packet = &packets[num_packets++];
5884
5885 chained = stbi__get8(s);
5886 packet->size = stbi__get8(s);
5887 packet->type = stbi__get8(s);
5888 packet->channel = stbi__get8(s);
5889
5890 act_comp |= packet->channel;
5891
5892 if (stbi__at_eof(s)) return stbi__errpuc("bad file","file too short (reading packets)");
5893 if (packet->size != 8) return stbi__errpuc("bad format","packet isn't 8bpp");
5894 } while (chained);
5895
5896 *comp = (act_comp & 0x10 ? 4 : 3); // has alpha channel?
5897
5898 for(y=0; y<height; ++y) {
5899 int packet_idx;
5900
5901 for(packet_idx=0; packet_idx < num_packets; ++packet_idx) {
5902 stbi__pic_packet *packet = &packets[packet_idx];
5903 stbi_uc *dest = result+y*width*4;
5904
5905 switch (packet->type) {
5906 default:
5907 return stbi__errpuc("bad format","packet has bad compression type");
5908
5909 case 0: {//uncompressed
5910 int x;
5911
5912 for(x=0;x<width;++x, dest+=4)
5913 if (!stbi__readval(s,packet->channel,dest))
5914 return 0;
5915 break;
5916 }
5917
5918 case 1://Pure RLE
5919 {
5920 int left=width, i;
5921
5922 while (left>0) {
5923 stbi_uc count,value[4];
5924
5925 count=stbi__get8(s);
5926 if (stbi__at_eof(s)) return stbi__errpuc("bad file","file too short (pure read count)");
5927
5928 if (count > left)
5929 count = (stbi_uc) left;
5930
5931 if (!stbi__readval(s,packet->channel,value)) return 0;
5932
5933 for(i=0; i<count; ++i,dest+=4)
5934 stbi__copyval(packet->channel,dest,value);
5935 left -= count;
5936 }
5937 }
5938 break;
5939
5940 case 2: {//Mixed RLE
5941 int left=width;
5942 while (left>0) {
5943 int count = stbi__get8(s), i;
5944 if (stbi__at_eof(s)) return stbi__errpuc("bad file","file too short (mixed read count)");
5945
5946 if (count >= 128) { // Repeated
5947 stbi_uc value[4];
5948
5949 if (count==128)
5950 count = stbi__get16be(s);
5951 else
5952 count -= 127;
5953 if (count > left)
5954 return stbi__errpuc("bad file","scanline overrun");
5955
5956 if (!stbi__readval(s,packet->channel,value))
5957 return 0;
5958
5959 for(i=0;i<count;++i, dest += 4)
5960 stbi__copyval(packet->channel,dest,value);
5961 } else { // Raw
5962 ++count;
5963 if (count>left) return stbi__errpuc("bad file","scanline overrun");
5964
5965 for(i=0;i<count;++i, dest+=4)
5966 if (!stbi__readval(s,packet->channel,dest))
5967 return 0;
5968 }
5969 left-=count;
5970 }
5971 break;
5972 }
5973 }
5974 }
5975 }
5976
5977 return result;
5978}
5979
5980static void *stbi__pic_load(stbi__context *s,int *px,int *py,int *comp,int req_comp, stbi__result_info *ri)
5981{
5982 stbi_uc *result;
5983 int i, x,y, internal_comp;
5984 STBI_NOTUSED(ri);
5985
5986 if (!comp) comp = &internal_comp;
5987
5988 for (i=0; i<92; ++i)
5989 stbi__get8(s);
5990
5991 x = stbi__get16be(s);
5992 y = stbi__get16be(s);
5993 if (stbi__at_eof(s)) return stbi__errpuc("bad file","file too short (pic header)");
5994 if (!stbi__mad3sizes_valid(x, y, 4, 0)) return stbi__errpuc("too large", "PIC image too large to decode");
5995
5996 stbi__get32be(s); //skip `ratio'
5997 stbi__get16be(s); //skip `fields'
5998 stbi__get16be(s); //skip `pad'
5999
6000 // intermediate buffer is RGBA
6001 result = (stbi_uc *) stbi__malloc_mad3(x, y, 4, 0);
6002 memset(result, 0xff, x*y*4);
6003
6004 if (!stbi__pic_load_core(s,x,y,comp, result)) {
6005 STBI_FREE(result);
6006 result=0;
6007 }
6008 *px = x;
6009 *py = y;
6010 if (req_comp == 0) req_comp = *comp;
6011 result=stbi__convert_format(result,4,req_comp,x,y);
6012
6013 return result;
6014}
6015
6016static int stbi__pic_test(stbi__context *s)
6017{
6018 int r = stbi__pic_test_core(s);
6019 stbi__rewind(s);
6020 return r;
6021}
6022#endif
6023
6024// *************************************************************************************************
6025// GIF loader -- public domain by Jean-Marc Lienher -- simplified/shrunk by stb
6026
6027#ifndef STBI_NO_GIF
6028typedef struct
6029{
6030 stbi__int16 prefix;
6031 stbi_uc first;
6032 stbi_uc suffix;
6033} stbi__gif_lzw;
6034
6035typedef struct
6036{
6037 int w,h;
6038 stbi_uc *out, *old_out; // output buffer (always 4 components)
6039 int flags, bgindex, ratio, transparent, eflags, delay;
6040 stbi_uc pal[256][4];
6041 stbi_uc lpal[256][4];
6042 stbi__gif_lzw codes[4096];
6043 stbi_uc *color_table;
6044 int parse, step;
6045 int lflags;
6046 int start_x, start_y;
6047 int max_x, max_y;
6048 int cur_x, cur_y;
6049 int line_size;
6050} stbi__gif;
6051
6052static int stbi__gif_test_raw(stbi__context *s)
6053{
6054 int sz;
6055 if (stbi__get8(s) != 'G' || stbi__get8(s) != 'I' || stbi__get8(s) != 'F' || stbi__get8(s) != '8') return 0;
6056 sz = stbi__get8(s);
6057 if (sz != '9' && sz != '7') return 0;
6058 if (stbi__get8(s) != 'a') return 0;
6059 return 1;
6060}
6061
6062static int stbi__gif_test(stbi__context *s)
6063{
6064 int r = stbi__gif_test_raw(s);
6065 stbi__rewind(s);
6066 return r;
6067}
6068
6069static void stbi__gif_parse_colortable(stbi__context *s, stbi_uc pal[256][4], int num_entries, int transp)
6070{
6071 int i;
6072 for (i=0; i < num_entries; ++i) {
6073 pal[i][2] = stbi__get8(s);
6074 pal[i][1] = stbi__get8(s);
6075 pal[i][0] = stbi__get8(s);
6076 pal[i][3] = transp == i ? 0 : 255;
6077 }
6078}
6079
6080static int stbi__gif_header(stbi__context *s, stbi__gif *g, int *comp, int is_info)
6081{
6082 stbi_uc version;
6083 if (stbi__get8(s) != 'G' || stbi__get8(s) != 'I' || stbi__get8(s) != 'F' || stbi__get8(s) != '8')
6084 return stbi__err("not GIF", "Corrupt GIF");
6085
6086 version = stbi__get8(s);
6087 if (version != '7' && version != '9') return stbi__err("not GIF", "Corrupt GIF");
6088 if (stbi__get8(s) != 'a') return stbi__err("not GIF", "Corrupt GIF");
6089
6090 stbi__g_failure_reason = "";
6091 g->w = stbi__get16le(s);
6092 g->h = stbi__get16le(s);
6093 g->flags = stbi__get8(s);
6094 g->bgindex = stbi__get8(s);
6095 g->ratio = stbi__get8(s);
6096 g->transparent = -1;
6097
6098 if (comp != 0) *comp = 4; // can't actually tell whether it's 3 or 4 until we parse the comments
6099
6100 if (is_info) return 1;
6101
6102 if (g->flags & 0x80)
6103 stbi__gif_parse_colortable(s,g->pal, 2 << (g->flags & 7), -1);
6104
6105 return 1;
6106}
6107
6108static int stbi__gif_info_raw(stbi__context *s, int *x, int *y, int *comp)
6109{
6110 stbi__gif* g = (stbi__gif*) stbi__malloc(sizeof(stbi__gif));
6111 if (!stbi__gif_header(s, g, comp, 1)) {
6112 STBI_FREE(g);
6113 stbi__rewind( s );
6114 return 0;
6115 }
6116 if (x) *x = g->w;
6117 if (y) *y = g->h;
6118 STBI_FREE(g);
6119 return 1;
6120}
6121
6122static void stbi__out_gif_code(stbi__gif *g, stbi__uint16 code)
6123{
6124 stbi_uc *p, *c;
6125
6126 // recurse to decode the prefixes, since the linked-list is backwards,
6127 // and working backwards through an interleaved image would be nasty
6128 if (g->codes[code].prefix >= 0)
6129 stbi__out_gif_code(g, g->codes[code].prefix);
6130
6131 if (g->cur_y >= g->max_y) return;
6132
6133 p = &g->out[g->cur_x + g->cur_y];
6134 c = &g->color_table[g->codes[code].suffix * 4];
6135
6136 if (c[3] >= 128) {
6137 p[0] = c[2];
6138 p[1] = c[1];
6139 p[2] = c[0];
6140 p[3] = c[3];
6141 }
6142 g->cur_x += 4;
6143
6144 if (g->cur_x >= g->max_x) {
6145 g->cur_x = g->start_x;
6146 g->cur_y += g->step;
6147
6148 while (g->cur_y >= g->max_y && g->parse > 0) {
6149 g->step = (1 << g->parse) * g->line_size;
6150 g->cur_y = g->start_y + (g->step >> 1);
6151 --g->parse;
6152 }
6153 }
6154}
6155
6156static stbi_uc *stbi__process_gif_raster(stbi__context *s, stbi__gif *g)
6157{
6158 stbi_uc lzw_cs;
6159 stbi__int32 len, init_code;
6160 stbi__uint32 first;
6161 stbi__int32 codesize, codemask, avail, oldcode, bits, valid_bits, clear;
6162 stbi__gif_lzw *p;
6163
6164 lzw_cs = stbi__get8(s);
6165 if (lzw_cs > 12) return NULL;
6166 clear = 1 << lzw_cs;
6167 first = 1;
6168 codesize = lzw_cs + 1;
6169 codemask = (1 << codesize) - 1;
6170 bits = 0;
6171 valid_bits = 0;
6172 for (init_code = 0; init_code < clear; init_code++) {
6173 g->codes[init_code].prefix = -1;
6174 g->codes[init_code].first = (stbi_uc) init_code;
6175 g->codes[init_code].suffix = (stbi_uc) init_code;
6176 }
6177
6178 // support no starting clear code
6179 avail = clear+2;
6180 oldcode = -1;
6181
6182 len = 0;
6183 for(;;) {
6184 if (valid_bits < codesize) {
6185 if (len == 0) {
6186 len = stbi__get8(s); // start new block
6187 if (len == 0)
6188 return g->out;
6189 }
6190 --len;
6191 bits |= (stbi__int32) stbi__get8(s) << valid_bits;
6192 valid_bits += 8;
6193 } else {
6194 stbi__int32 code = bits & codemask;
6195 bits >>= codesize;
6196 valid_bits -= codesize;
6197 // @OPTIMIZE: is there some way we can accelerate the non-clear path?
6198 if (code == clear) { // clear code
6199 codesize = lzw_cs + 1;
6200 codemask = (1 << codesize) - 1;
6201 avail = clear + 2;
6202 oldcode = -1;
6203 first = 0;
6204 } else if (code == clear + 1) { // end of stream code
6205 stbi__skip(s, len);
6206 while ((len = stbi__get8(s)) > 0)
6207 stbi__skip(s,len);
6208 return g->out;
6209 } else if (code <= avail) {
6210 if (first) return stbi__errpuc("no clear code", "Corrupt GIF");
6211
6212 if (oldcode >= 0) {
6213 p = &g->codes[avail++];
6214 if (avail > 4096) return stbi__errpuc("too many codes", "Corrupt GIF");
6215 p->prefix = (stbi__int16) oldcode;
6216 p->first = g->codes[oldcode].first;
6217 p->suffix = (code == avail) ? p->first : g->codes[code].first;
6218 } else if (code == avail)
6219 return stbi__errpuc("illegal code in raster", "Corrupt GIF");
6220
6221 stbi__out_gif_code(g, (stbi__uint16) code);
6222
6223 if ((avail & codemask) == 0 && avail <= 0x0FFF) {
6224 codesize++;
6225 codemask = (1 << codesize) - 1;
6226 }
6227
6228 oldcode = code;
6229 } else {
6230 return stbi__errpuc("illegal code in raster", "Corrupt GIF");
6231 }
6232 }
6233 }
6234}
6235
6236static void stbi__fill_gif_background(stbi__gif *g, int x0, int y0, int x1, int y1)
6237{
6238 int x, y;
6239 stbi_uc *c = g->pal[g->bgindex];
6240 for (y = y0; y < y1; y += 4 * g->w) {
6241 for (x = x0; x < x1; x += 4) {
6242 stbi_uc *p = &g->out[y + x];
6243 p[0] = c[2];
6244 p[1] = c[1];
6245 p[2] = c[0];
6246 p[3] = 0;
6247 }
6248 }
6249}
6250
6251// this function is designed to support animated gifs, although stb_image doesn't support it
6252static stbi_uc *stbi__gif_load_next(stbi__context *s, stbi__gif *g, int *comp, int req_comp)
6253{
6254 int i;
6255 stbi_uc *prev_out = 0;
6256
6257 if (g->out == 0 && !stbi__gif_header(s, g, comp,0))
6258 return 0; // stbi__g_failure_reason set by stbi__gif_header
6259
6260 if (!stbi__mad3sizes_valid(g->w, g->h, 4, 0))
6261 return stbi__errpuc("too large", "GIF too large");
6262
6263 prev_out = g->out;
6264 g->out = (stbi_uc *) stbi__malloc_mad3(4, g->w, g->h, 0);
6265 if (g->out == 0) return stbi__errpuc("outofmem", "Out of memory");
6266
6267 switch ((g->eflags & 0x1C) >> 2) {
6268 case 0: // unspecified (also always used on 1st frame)
6269 stbi__fill_gif_background(g, 0, 0, 4 * g->w, 4 * g->w * g->h);
6270 break;
6271 case 1: // do not dispose
6272 if (prev_out) memcpy(g->out, prev_out, 4 * g->w * g->h);
6273 g->old_out = prev_out;
6274 break;
6275 case 2: // dispose to background
6276 if (prev_out) memcpy(g->out, prev_out, 4 * g->w * g->h);
6277 stbi__fill_gif_background(g, g->start_x, g->start_y, g->max_x, g->max_y);
6278 break;
6279 case 3: // dispose to previous
6280 if (g->old_out) {
6281 for (i = g->start_y; i < g->max_y; i += 4 * g->w)
6282 memcpy(&g->out[i + g->start_x], &g->old_out[i + g->start_x], g->max_x - g->start_x);
6283 }
6284 break;
6285 }
6286
6287 for (;;) {
6288 switch (stbi__get8(s)) {
6289 case 0x2C: /* Image Descriptor */
6290 {
6291 int prev_trans = -1;
6292 stbi__int32 x, y, w, h;
6293 stbi_uc *o;
6294
6295 x = stbi__get16le(s);
6296 y = stbi__get16le(s);
6297 w = stbi__get16le(s);
6298 h = stbi__get16le(s);
6299 if (((x + w) > (g->w)) || ((y + h) > (g->h)))
6300 return stbi__errpuc("bad Image Descriptor", "Corrupt GIF");
6301
6302 g->line_size = g->w * 4;
6303 g->start_x = x * 4;
6304 g->start_y = y * g->line_size;
6305 g->max_x = g->start_x + w * 4;
6306 g->max_y = g->start_y + h * g->line_size;
6307 g->cur_x = g->start_x;
6308 g->cur_y = g->start_y;
6309
6310 g->lflags = stbi__get8(s);
6311
6312 if (g->lflags & 0x40) {
6313 g->step = 8 * g->line_size; // first interlaced spacing
6314 g->parse = 3;
6315 } else {
6316 g->step = g->line_size;
6317 g->parse = 0;
6318 }
6319
6320 if (g->lflags & 0x80) {
6321 stbi__gif_parse_colortable(s,g->lpal, 2 << (g->lflags & 7), g->eflags & 0x01 ? g->transparent : -1);
6322 g->color_table = (stbi_uc *) g->lpal;
6323 } else if (g->flags & 0x80) {
6324 if (g->transparent >= 0 && (g->eflags & 0x01)) {
6325 prev_trans = g->pal[g->transparent][3];
6326 g->pal[g->transparent][3] = 0;
6327 }
6328 g->color_table = (stbi_uc *) g->pal;
6329 } else
6330 return stbi__errpuc("missing color table", "Corrupt GIF");
6331
6332 o = stbi__process_gif_raster(s, g);
6333 if (o == NULL) return NULL;
6334
6335 if (prev_trans != -1)
6336 g->pal[g->transparent][3] = (stbi_uc) prev_trans;
6337
6338 return o;
6339 }
6340
6341 case 0x21: // Comment Extension.
6342 {
6343 int len;
6344 if (stbi__get8(s) == 0xF9) { // Graphic Control Extension.
6345 len = stbi__get8(s);
6346 if (len == 4) {
6347 g->eflags = stbi__get8(s);
6348 g->delay = stbi__get16le(s);
6349 g->transparent = stbi__get8(s);
6350 } else {
6351 stbi__skip(s, len);
6352 break;
6353 }
6354 }
6355 while ((len = stbi__get8(s)) != 0)
6356 stbi__skip(s, len);
6357 break;
6358 }
6359
6360 case 0x3B: // gif stream termination code
6361 return (stbi_uc *) s; // using '1' causes warning on some compilers
6362
6363 default:
6364 return stbi__errpuc("unknown code", "Corrupt GIF");
6365 }
6366 }
6367
6368 STBI_NOTUSED(req_comp);
6369}
6370
6371static void *stbi__gif_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
6372{
6373 stbi_uc *u = 0;
6374 stbi__gif* g = (stbi__gif*) stbi__malloc(sizeof(stbi__gif));
6375 memset(g, 0, sizeof(*g));
6376 STBI_NOTUSED(ri);
6377
6378 u = stbi__gif_load_next(s, g, comp, req_comp);
6379 if (u == (stbi_uc *) s) u = 0; // end of animated gif marker
6380 if (u) {
6381 *x = g->w;
6382 *y = g->h;
6383 if (req_comp && req_comp != 4)
6384 u = stbi__convert_format(u, 4, req_comp, g->w, g->h);
6385 }
6386 else if (g->out)
6387 STBI_FREE(g->out);
6388 STBI_FREE(g);
6389 return u;
6390}
6391
6392static int stbi__gif_info(stbi__context *s, int *x, int *y, int *comp)
6393{
6394 return stbi__gif_info_raw(s,x,y,comp);
6395}
6396#endif
6397
6398// *************************************************************************************************
6399// Radiance RGBE HDR loader
6400// originally by Nicolas Schulz
6401#ifndef STBI_NO_HDR
6402static int stbi__hdr_test_core(stbi__context *s, const char *signature)
6403{
6404 int i;
6405 for (i=0; signature[i]; ++i)
6406 if (stbi__get8(s) != signature[i])
6407 return 0;
6408 stbi__rewind(s);
6409 return 1;
6410}
6411
6412static int stbi__hdr_test(stbi__context* s)
6413{
6414 int r = stbi__hdr_test_core(s, "#?RADIANCE\n");
6415 stbi__rewind(s);
6416 if(!r) {
6417 r = stbi__hdr_test_core(s, "#?RGBE\n");
6418 stbi__rewind(s);
6419 }
6420 return r;
6421}
6422
6423#define STBI__HDR_BUFLEN 1024
6424static char *stbi__hdr_gettoken(stbi__context *z, char *buffer)
6425{
6426 int len=0;
6427 char c = '\0';
6428
6429 c = (char) stbi__get8(z);
6430
6431 while (!stbi__at_eof(z) && c != '\n') {
6432 buffer[len++] = c;
6433 if (len == STBI__HDR_BUFLEN-1) {
6434 // flush to end of line
6435 while (!stbi__at_eof(z) && stbi__get8(z) != '\n')
6436 ;
6437 break;
6438 }
6439 c = (char) stbi__get8(z);
6440 }
6441
6442 buffer[len] = 0;
6443 return buffer;
6444}
6445
6446static void stbi__hdr_convert(float *output, stbi_uc *input, int req_comp)
6447{
6448 if ( input[3] != 0 ) {
6449 float f1;
6450 // Exponent
6451 f1 = (float) ldexp(1.0f, input[3] - (int)(128 + 8));
6452 if (req_comp <= 2)
6453 output[0] = (input[0] + input[1] + input[2]) * f1 / 3;
6454 else {
6455 output[0] = input[0] * f1;
6456 output[1] = input[1] * f1;
6457 output[2] = input[2] * f1;
6458 }
6459 if (req_comp == 2) output[1] = 1;
6460 if (req_comp == 4) output[3] = 1;
6461 } else {
6462 switch (req_comp) {
6463 case 4: output[3] = 1; /* fallthrough */
6464 case 3: output[0] = output[1] = output[2] = 0;
6465 break;
6466 case 2: output[1] = 1; /* fallthrough */
6467 case 1: output[0] = 0;
6468 break;
6469 }
6470 }
6471}
6472
6473static float *stbi__hdr_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
6474{
6475 char buffer[STBI__HDR_BUFLEN];
6476 char *token;
6477 int valid = 0;
6478 int width, height;
6479 stbi_uc *scanline;
6480 float *hdr_data;
6481 int len;
6482 unsigned char count, value;
6483 int i, j, k, c1,c2, z;
6484 const char *headerToken;
6485 STBI_NOTUSED(ri);
6486
6487 // Check identifier
6488 headerToken = stbi__hdr_gettoken(s,buffer);
6489 if (strcmp(headerToken, "#?RADIANCE") != 0 && strcmp(headerToken, "#?RGBE") != 0)
6490 return stbi__errpf("not HDR", "Corrupt HDR image");
6491
6492 // Parse header
6493 for(;;) {
6494 token = stbi__hdr_gettoken(s,buffer);
6495 if (token[0] == 0) break;
6496 if (strcmp(token, "FORMAT=32-bit_rle_rgbe") == 0) valid = 1;
6497 }
6498
6499 if (!valid) return stbi__errpf("unsupported format", "Unsupported HDR format");
6500
6501 // Parse width and height
6502 // can't use sscanf() if we're not using stdio!
6503 token = stbi__hdr_gettoken(s,buffer);
6504 if (strncmp(token, "-Y ", 3)) return stbi__errpf("unsupported data layout", "Unsupported HDR format");
6505 token += 3;
6506 height = (int) strtol(token, &token, 10);
6507 while (*token == ' ') ++token;
6508 if (strncmp(token, "+X ", 3)) return stbi__errpf("unsupported data layout", "Unsupported HDR format");
6509 token += 3;
6510 width = (int) strtol(token, NULL, 10);
6511
6512 *x = width;
6513 *y = height;
6514
6515 if (comp) *comp = 3;
6516 if (req_comp == 0) req_comp = 3;
6517
6518 if (!stbi__mad4sizes_valid(width, height, req_comp, sizeof(float), 0))
6519 return stbi__errpf("too large", "HDR image is too large");
6520
6521 // Read data
6522 hdr_data = (float *) stbi__malloc_mad4(width, height, req_comp, sizeof(float), 0);
6523 if (!hdr_data)
6524 return stbi__errpf("outofmem", "Out of memory");
6525
6526 // Load image data
6527 // image data is stored as some number of sca
6528 if ( width < 8 || width >= 32768) {
6529 // Read flat data
6530 for (j=0; j < height; ++j) {
6531 for (i=0; i < width; ++i) {
6532 stbi_uc rgbe[4];
6533 main_decode_loop:
6534 stbi__getn(s, rgbe, 4);
6535 stbi__hdr_convert(hdr_data + j * width * req_comp + i * req_comp, rgbe, req_comp);
6536 }
6537 }
6538 } else {
6539 // Read RLE-encoded data
6540 scanline = NULL;
6541
6542 for (j = 0; j < height; ++j) {
6543 c1 = stbi__get8(s);
6544 c2 = stbi__get8(s);
6545 len = stbi__get8(s);
6546 if (c1 != 2 || c2 != 2 || (len & 0x80)) {
6547 // not run-length encoded, so we have to actually use THIS data as a decoded
6548 // pixel (note this can't be a valid pixel--one of RGB must be >= 128)
6549 stbi_uc rgbe[4];
6550 rgbe[0] = (stbi_uc) c1;
6551 rgbe[1] = (stbi_uc) c2;
6552 rgbe[2] = (stbi_uc) len;
6553 rgbe[3] = (stbi_uc) stbi__get8(s);
6554 stbi__hdr_convert(hdr_data, rgbe, req_comp);
6555 i = 1;
6556 j = 0;
6557 STBI_FREE(scanline);
6558 goto main_decode_loop; // yes, this makes no sense
6559 }
6560 len <<= 8;
6561 len |= stbi__get8(s);
6562 if (len != width) { STBI_FREE(hdr_data); STBI_FREE(scanline); return stbi__errpf("invalid decoded scanline length", "corrupt HDR"); }
6563 if (scanline == NULL) {
6564 scanline = (stbi_uc *) stbi__malloc_mad2(width, 4, 0);
6565 if (!scanline) {
6566 STBI_FREE(hdr_data);
6567 return stbi__errpf("outofmem", "Out of memory");
6568 }
6569 }
6570
6571 for (k = 0; k < 4; ++k) {
6572 int nleft;
6573 i = 0;
6574 while ((nleft = width - i) > 0) {
6575 count = stbi__get8(s);
6576 if (count > 128) {
6577 // Run
6578 value = stbi__get8(s);
6579 count -= 128;
6580 if (count > nleft) { STBI_FREE(hdr_data); STBI_FREE(scanline); return stbi__errpf("corrupt", "bad RLE data in HDR"); }
6581 for (z = 0; z < count; ++z)
6582 scanline[i++ * 4 + k] = value;
6583 } else {
6584 // Dump
6585 if (count > nleft) { STBI_FREE(hdr_data); STBI_FREE(scanline); return stbi__errpf("corrupt", "bad RLE data in HDR"); }
6586 for (z = 0; z < count; ++z)
6587 scanline[i++ * 4 + k] = stbi__get8(s);
6588 }
6589 }
6590 }
6591 for (i=0; i < width; ++i)
6592 stbi__hdr_convert(hdr_data+(j*width + i)*req_comp, scanline + i*4, req_comp);
6593 }
6594 if (scanline)
6595 STBI_FREE(scanline);
6596 }
6597
6598 return hdr_data;
6599}
6600
6601static int stbi__hdr_info(stbi__context *s, int *x, int *y, int *comp)
6602{
6603 char buffer[STBI__HDR_BUFLEN];
6604 char *token;
6605 int valid = 0;
6606 int dummy;
6607
6608 if (!x) x = &dummy;
6609 if (!y) y = &dummy;
6610 if (!comp) comp = &dummy;
6611
6612 if (stbi__hdr_test(s) == 0) {
6613 stbi__rewind( s );
6614 return 0;
6615 }
6616
6617 for(;;) {
6618 token = stbi__hdr_gettoken(s,buffer);
6619 if (token[0] == 0) break;
6620 if (strcmp(token, "FORMAT=32-bit_rle_rgbe") == 0) valid = 1;
6621 }
6622
6623 if (!valid) {
6624 stbi__rewind( s );
6625 return 0;
6626 }
6627 token = stbi__hdr_gettoken(s,buffer);
6628 if (strncmp(token, "-Y ", 3)) {
6629 stbi__rewind( s );
6630 return 0;
6631 }
6632 token += 3;
6633 *y = (int) strtol(token, &token, 10);
6634 while (*token == ' ') ++token;
6635 if (strncmp(token, "+X ", 3)) {
6636 stbi__rewind( s );
6637 return 0;
6638 }
6639 token += 3;
6640 *x = (int) strtol(token, NULL, 10);
6641 *comp = 3;
6642 return 1;
6643}
6644#endif // STBI_NO_HDR
6645
6646#ifndef STBI_NO_BMP
6647static int stbi__bmp_info(stbi__context *s, int *x, int *y, int *comp)
6648{
6649 void *p;
6650 stbi__bmp_data info;
6651
6652 info.all_a = 255;
6653 p = stbi__bmp_parse_header(s, &info);
6654 stbi__rewind( s );
6655 if (p == NULL)
6656 return 0;
6657 if (x) *x = s->img_x;
6658 if (y) *y = s->img_y;
6659 if (comp) *comp = info.ma ? 4 : 3;
6660 return 1;
6661}
6662#endif
6663
6664#ifndef STBI_NO_PSD
6665static int stbi__psd_info(stbi__context *s, int *x, int *y, int *comp)
6666{
6667 int channelCount, dummy;
6668 if (!x) x = &dummy;
6669 if (!y) y = &dummy;
6670 if (!comp) comp = &dummy;
6671 if (stbi__get32be(s) != 0x38425053) {
6672 stbi__rewind( s );
6673 return 0;
6674 }
6675 if (stbi__get16be(s) != 1) {
6676 stbi__rewind( s );
6677 return 0;
6678 }
6679 stbi__skip(s, 6);
6680 channelCount = stbi__get16be(s);
6681 if (channelCount < 0 || channelCount > 16) {
6682 stbi__rewind( s );
6683 return 0;
6684 }
6685 *y = stbi__get32be(s);
6686 *x = stbi__get32be(s);
6687 if (stbi__get16be(s) != 8) {
6688 stbi__rewind( s );
6689 return 0;
6690 }
6691 if (stbi__get16be(s) != 3) {
6692 stbi__rewind( s );
6693 return 0;
6694 }
6695 *comp = 4;
6696 return 1;
6697}
6698#endif
6699
6700#ifndef STBI_NO_PIC
6701static int stbi__pic_info(stbi__context *s, int *x, int *y, int *comp)
6702{
6703 int act_comp=0,num_packets=0,chained,dummy;
6704 stbi__pic_packet packets[10];
6705
6706 if (!x) x = &dummy;
6707 if (!y) y = &dummy;
6708 if (!comp) comp = &dummy;
6709
6710 if (!stbi__pic_is4(s,"\x53\x80\xF6\x34")) {
6711 stbi__rewind(s);
6712 return 0;
6713 }
6714
6715 stbi__skip(s, 88);
6716
6717 *x = stbi__get16be(s);
6718 *y = stbi__get16be(s);
6719 if (stbi__at_eof(s)) {
6720 stbi__rewind( s);
6721 return 0;
6722 }
6723 if ( (*x) != 0 && (1 << 28) / (*x) < (*y)) {
6724 stbi__rewind( s );
6725 return 0;
6726 }
6727
6728 stbi__skip(s, 8);
6729
6730 do {
6731 stbi__pic_packet *packet;
6732
6733 if (num_packets==sizeof(packets)/sizeof(packets[0]))
6734 return 0;
6735
6736 packet = &packets[num_packets++];
6737 chained = stbi__get8(s);
6738 packet->size = stbi__get8(s);
6739 packet->type = stbi__get8(s);
6740 packet->channel = stbi__get8(s);
6741 act_comp |= packet->channel;
6742
6743 if (stbi__at_eof(s)) {
6744 stbi__rewind( s );
6745 return 0;
6746 }
6747 if (packet->size != 8) {
6748 stbi__rewind( s );
6749 return 0;
6750 }
6751 } while (chained);
6752
6753 *comp = (act_comp & 0x10 ? 4 : 3);
6754
6755 return 1;
6756}
6757#endif
6758
6759// *************************************************************************************************
6760// Portable Gray Map and Portable Pixel Map loader
6761// by Ken Miller
6762//
6763// PGM: http://netpbm.sourceforge.net/doc/pgm.html
6764// PPM: http://netpbm.sourceforge.net/doc/ppm.html
6765//
6766// Known limitations:
6767// Does not support comments in the header section
6768// Does not support ASCII image data (formats P2 and P3)
6769// Does not support 16-bit-per-channel
6770
6771#ifndef STBI_NO_PNM
6772
6773static int stbi__pnm_test(stbi__context *s)
6774{
6775 char p, t;
6776 p = (char) stbi__get8(s);
6777 t = (char) stbi__get8(s);
6778 if (p != 'P' || (t != '5' && t != '6')) {
6779 stbi__rewind( s );
6780 return 0;
6781 }
6782 return 1;
6783}
6784
6785static void *stbi__pnm_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
6786{
6787 stbi_uc *out;
6788 STBI_NOTUSED(ri);
6789
6790 if (!stbi__pnm_info(s, (int *)&s->img_x, (int *)&s->img_y, (int *)&s->img_n))
6791 return 0;
6792
6793 *x = s->img_x;
6794 *y = s->img_y;
6795 if (comp) *comp = s->img_n;
6796
6797 if (!stbi__mad3sizes_valid(s->img_n, s->img_x, s->img_y, 0))
6798 return stbi__errpuc("too large", "PNM too large");
6799
6800 out = (stbi_uc *) stbi__malloc_mad3(s->img_n, s->img_x, s->img_y, 0);
6801 if (!out) return stbi__errpuc("outofmem", "Out of memory");
6802 stbi__getn(s, out, s->img_n * s->img_x * s->img_y);
6803
6804 if (req_comp && req_comp != s->img_n) {
6805 out = stbi__convert_format(out, s->img_n, req_comp, s->img_x, s->img_y);
6806 if (out == NULL) return out; // stbi__convert_format frees input on failure
6807 }
6808 return out;
6809}
6810
6811static int stbi__pnm_isspace(char c)
6812{
6813 return c == ' ' || c == '\t' || c == '\n' || c == '\v' || c == '\f' || c == '\r';
6814}
6815
6816static void stbi__pnm_skip_whitespace(stbi__context *s, char *c)
6817{
6818 for (;;) {
6819 while (!stbi__at_eof(s) && stbi__pnm_isspace(*c))
6820 *c = (char) stbi__get8(s);
6821
6822 if (stbi__at_eof(s) || *c != '#')
6823 break;
6824
6825 while (!stbi__at_eof(s) && *c != '\n' && *c != '\r' )
6826 *c = (char) stbi__get8(s);
6827 }
6828}
6829
6830static int stbi__pnm_isdigit(char c)
6831{
6832 return c >= '0' && c <= '9';
6833}
6834
6835static int stbi__pnm_getinteger(stbi__context *s, char *c)
6836{
6837 int value = 0;
6838
6839 while (!stbi__at_eof(s) && stbi__pnm_isdigit(*c)) {
6840 value = value*10 + (*c - '0');
6841 *c = (char) stbi__get8(s);
6842 }
6843
6844 return value;
6845}
6846
6847static int stbi__pnm_info(stbi__context *s, int *x, int *y, int *comp)
6848{
6849 int maxv, dummy;
6850 char c, p, t;
6851
6852 if (!x) x = &dummy;
6853 if (!y) y = &dummy;
6854 if (!comp) comp = &dummy;
6855
6856 stbi__rewind(s);
6857
6858 // Get identifier
6859 p = (char) stbi__get8(s);
6860 t = (char) stbi__get8(s);
6861 if (p != 'P' || (t != '5' && t != '6')) {
6862 stbi__rewind(s);
6863 return 0;
6864 }
6865
6866 *comp = (t == '6') ? 3 : 1; // '5' is 1-component .pgm; '6' is 3-component .ppm
6867
6868 c = (char) stbi__get8(s);
6869 stbi__pnm_skip_whitespace(s, &c);
6870
6871 *x = stbi__pnm_getinteger(s, &c); // read width
6872 stbi__pnm_skip_whitespace(s, &c);
6873
6874 *y = stbi__pnm_getinteger(s, &c); // read height
6875 stbi__pnm_skip_whitespace(s, &c);
6876
6877 maxv = stbi__pnm_getinteger(s, &c); // read max value
6878
6879 if (maxv > 255)
6880 return stbi__err("max value > 255", "PPM image not 8-bit");
6881 else
6882 return 1;
6883}
6884#endif
6885
6886static int stbi__info_main(stbi__context *s, int *x, int *y, int *comp)
6887{
6888 #ifndef STBI_NO_JPEG
6889 if (stbi__jpeg_info(s, x, y, comp)) return 1;
6890 #endif
6891
6892 #ifndef STBI_NO_PNG
6893 if (stbi__png_info(s, x, y, comp)) return 1;
6894 #endif
6895
6896 #ifndef STBI_NO_GIF
6897 if (stbi__gif_info(s, x, y, comp)) return 1;
6898 #endif
6899
6900 #ifndef STBI_NO_BMP
6901 if (stbi__bmp_info(s, x, y, comp)) return 1;
6902 #endif
6903
6904 #ifndef STBI_NO_PSD
6905 if (stbi__psd_info(s, x, y, comp)) return 1;
6906 #endif
6907
6908 #ifndef STBI_NO_PIC
6909 if (stbi__pic_info(s, x, y, comp)) return 1;
6910 #endif
6911
6912 #ifndef STBI_NO_PNM
6913 if (stbi__pnm_info(s, x, y, comp)) return 1;
6914 #endif
6915
6916 #ifndef STBI_NO_HDR
6917 if (stbi__hdr_info(s, x, y, comp)) return 1;
6918 #endif
6919
6920 // test tga last because it's a crappy test!
6921 #ifndef STBI_NO_TGA
6922 if (stbi__tga_info(s, x, y, comp))
6923 return 1;
6924 #endif
6925 return stbi__err("unknown image type", "Image not of any known type, or corrupt");
6926}
6927
6928#ifndef STBI_NO_STDIO
6929STBIDEF int stbi_info(char const *filename, int *x, int *y, int *comp)
6930{
6931 FILE *f = stbi__fopen(filename, "rb");
6932 int result;
6933 if (!f) return stbi__err("can't fopen", "Unable to open file");
6934 result = stbi_info_from_file(f, x, y, comp);
6935 fclose(f);
6936 return result;
6937}
6938
6939STBIDEF int stbi_info_from_file(FILE *f, int *x, int *y, int *comp)
6940{
6941 int r;
6942 stbi__context s;
6943 long pos = ftell(f);
6944 stbi__start_file(&s, f);
6945 r = stbi__info_main(&s,x,y,comp);
6946 fseek(f,pos,SEEK_SET);
6947 return r;
6948}
6949#endif // !STBI_NO_STDIO
6950
6951STBIDEF int stbi_info_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp)
6952{
6953 stbi__context s;
6954 stbi__start_mem(&s,buffer,len);
6955 return stbi__info_main(&s,x,y,comp);
6956}
6957
6958STBIDEF int stbi_info_from_callbacks(stbi_io_callbacks const *c, void *user, int *x, int *y, int *comp)
6959{
6960 stbi__context s;
6961 stbi__start_callbacks(&s, (stbi_io_callbacks *) c, user);
6962 return stbi__info_main(&s,x,y,comp);
6963}
6964
6965#endif // STB_IMAGE_IMPLEMENTATION
6966
6967/*
6968 revision history:
6969 2.15 (2017-03-18) fix png-1,2,4 bug; now all Imagenet JPGs decode;
6970 warning fixes; disable run-time SSE detection on gcc;
6971 uniform handling of optional "return" values;
6972 thread-safe initialization of zlib tables
6973 2.14 (2017-03-03) remove deprecated STBI_JPEG_OLD; fixes for Imagenet JPGs
6974 2.13 (2016-11-29) add 16-bit API, only supported for PNG right now
6975 2.12 (2016-04-02) fix typo in 2.11 PSD fix that caused crashes
6976 2.11 (2016-04-02) allocate large structures on the stack
6977 remove white matting for transparent PSD
6978 fix reported channel count for PNG & BMP
6979 re-enable SSE2 in non-gcc 64-bit
6980 support RGB-formatted JPEG
6981 read 16-bit PNGs (only as 8-bit)
6982 2.10 (2016-01-22) avoid warning introduced in 2.09 by STBI_REALLOC_SIZED
6983 2.09 (2016-01-16) allow comments in PNM files
6984 16-bit-per-pixel TGA (not bit-per-component)
6985 info() for TGA could break due to .hdr handling
6986 info() for BMP to shares code instead of sloppy parse
6987 can use STBI_REALLOC_SIZED if allocator doesn't support realloc
6988 code cleanup
6989 2.08 (2015-09-13) fix to 2.07 cleanup, reading RGB PSD as RGBA
6990 2.07 (2015-09-13) fix compiler warnings
6991 partial animated GIF support
6992 limited 16-bpc PSD support
6993 #ifdef unused functions
6994 bug with < 92 byte PIC,PNM,HDR,TGA
6995 2.06 (2015-04-19) fix bug where PSD returns wrong '*comp' value
6996 2.05 (2015-04-19) fix bug in progressive JPEG handling, fix warning
6997 2.04 (2015-04-15) try to re-enable SIMD on MinGW 64-bit
6998 2.03 (2015-04-12) extra corruption checking (mmozeiko)
6999 stbi_set_flip_vertically_on_load (nguillemot)
7000 fix NEON support; fix mingw support
7001 2.02 (2015-01-19) fix incorrect assert, fix warning
7002 2.01 (2015-01-17) fix various warnings; suppress SIMD on gcc 32-bit without -msse2
7003 2.00b (2014-12-25) fix STBI_MALLOC in progressive JPEG
7004 2.00 (2014-12-25) optimize JPG, including x86 SSE2 & NEON SIMD (ryg)
7005 progressive JPEG (stb)
7006 PGM/PPM support (Ken Miller)
7007 STBI_MALLOC,STBI_REALLOC,STBI_FREE
7008 GIF bugfix -- seemingly never worked
7009 STBI_NO_*, STBI_ONLY_*
7010 1.48 (2014-12-14) fix incorrectly-named assert()
7011 1.47 (2014-12-14) 1/2/4-bit PNG support, both direct and paletted (Omar Cornut & stb)
7012 optimize PNG (ryg)
7013 fix bug in interlaced PNG with user-specified channel count (stb)
7014 1.46 (2014-08-26)
7015 fix broken tRNS chunk (colorkey-style transparency) in non-paletted PNG
7016 1.45 (2014-08-16)
7017 fix MSVC-ARM internal compiler error by wrapping malloc
7018 1.44 (2014-08-07)
7019 various warning fixes from Ronny Chevalier
7020 1.43 (2014-07-15)
7021 fix MSVC-only compiler problem in code changed in 1.42
7022 1.42 (2014-07-09)
7023 don't define _CRT_SECURE_NO_WARNINGS (affects user code)
7024 fixes to stbi__cleanup_jpeg path
7025 added STBI_ASSERT to avoid requiring assert.h
7026 1.41 (2014-06-25)
7027 fix search&replace from 1.36 that messed up comments/error messages
7028 1.40 (2014-06-22)
7029 fix gcc struct-initialization warning
7030 1.39 (2014-06-15)
7031 fix to TGA optimization when req_comp != number of components in TGA;
7032 fix to GIF loading because BMP wasn't rewinding (whoops, no GIFs in my test suite)
7033 add support for BMP version 5 (more ignored fields)
7034 1.38 (2014-06-06)
7035 suppress MSVC warnings on integer casts truncating values
7036 fix accidental rename of 'skip' field of I/O
7037 1.37 (2014-06-04)
7038 remove duplicate typedef
7039 1.36 (2014-06-03)
7040 convert to header file single-file library
7041 if de-iphone isn't set, load iphone images color-swapped instead of returning NULL
7042 1.35 (2014-05-27)
7043 various warnings
7044 fix broken STBI_SIMD path
7045 fix bug where stbi_load_from_file no longer left file pointer in correct place
7046 fix broken non-easy path for 32-bit BMP (possibly never used)
7047 TGA optimization by Arseny Kapoulkine
7048 1.34 (unknown)
7049 use STBI_NOTUSED in stbi__resample_row_generic(), fix one more leak in tga failure case
7050 1.33 (2011-07-14)
7051 make stbi_is_hdr work in STBI_NO_HDR (as specified), minor compiler-friendly improvements
7052 1.32 (2011-07-13)
7053 support for "info" function for all supported filetypes (SpartanJ)
7054 1.31 (2011-06-20)
7055 a few more leak fixes, bug in PNG handling (SpartanJ)
7056 1.30 (2011-06-11)
7057 added ability to load files via callbacks to accomidate custom input streams (Ben Wenger)
7058 removed deprecated format-specific test/load functions
7059 removed support for installable file formats (stbi_loader) -- would have been broken for IO callbacks anyway
7060 error cases in bmp and tga give messages and don't leak (Raymond Barbiero, grisha)
7061 fix inefficiency in decoding 32-bit BMP (David Woo)
7062 1.29 (2010-08-16)
7063 various warning fixes from Aurelien Pocheville
7064 1.28 (2010-08-01)
7065 fix bug in GIF palette transparency (SpartanJ)
7066 1.27 (2010-08-01)
7067 cast-to-stbi_uc to fix warnings
7068 1.26 (2010-07-24)
7069 fix bug in file buffering for PNG reported by SpartanJ
7070 1.25 (2010-07-17)
7071 refix trans_data warning (Won Chun)
7072 1.24 (2010-07-12)
7073 perf improvements reading from files on platforms with lock-heavy fgetc()
7074 minor perf improvements for jpeg
7075 deprecated type-specific functions so we'll get feedback if they're needed
7076 attempt to fix trans_data warning (Won Chun)
7077 1.23 fixed bug in iPhone support
7078 1.22 (2010-07-10)
7079 removed image *writing* support
7080 stbi_info support from Jetro Lauha
7081 GIF support from Jean-Marc Lienher
7082 iPhone PNG-extensions from James Brown
7083 warning-fixes from Nicolas Schulz and Janez Zemva (i.stbi__err. Janez (U+017D)emva)
7084 1.21 fix use of 'stbi_uc' in header (reported by jon blow)
7085 1.20 added support for Softimage PIC, by Tom Seddon
7086 1.19 bug in interlaced PNG corruption check (found by ryg)
7087 1.18 (2008-08-02)
7088 fix a threading bug (local mutable static)
7089 1.17 support interlaced PNG
7090 1.16 major bugfix - stbi__convert_format converted one too many pixels
7091 1.15 initialize some fields for thread safety
7092 1.14 fix threadsafe conversion bug
7093 header-file-only version (#define STBI_HEADER_FILE_ONLY before including)
7094 1.13 threadsafe
7095 1.12 const qualifiers in the API
7096 1.11 Support installable IDCT, colorspace conversion routines
7097 1.10 Fixes for 64-bit (don't use "unsigned long")
7098 optimized upsampling by Fabian "ryg" Giesen
7099 1.09 Fix format-conversion for PSD code (bad global variables!)
7100 1.08 Thatcher Ulrich's PSD code integrated by Nicolas Schulz
7101 1.07 attempt to fix C++ warning/errors again
7102 1.06 attempt to fix C++ warning/errors again
7103 1.05 fix TGA loading to return correct *comp and use good luminance calc
7104 1.04 default float alpha is 1, not 255; use 'void *' for stbi_image_free
7105 1.03 bugfixes to STBI_NO_STDIO, STBI_NO_HDR
7106 1.02 support for (subset of) HDR files, float interface for preferred access to them
7107 1.01 fix bug: possible bug in handling right-side up bmps... not sure
7108 fix bug: the stbi__bmp_load() and stbi__tga_load() functions didn't work at all
7109 1.00 interface to zlib that skips zlib header
7110 0.99 correct handling of alpha in palette
7111 0.98 TGA loader by lonesock; dynamically add loaders (untested)
7112 0.97 jpeg errors on too large a file; also catch another malloc failure
7113 0.96 fix detection of invalid v value - particleman@mollyrocket forum
7114 0.95 during header scan, seek to markers in case of padding
7115 0.94 STBI_NO_STDIO to disable stdio usage; rename all #defines the same
7116 0.93 handle jpegtran output; verbose errors
7117 0.92 read 4,8,16,24,32-bit BMP files of several formats
7118 0.91 output 24-bit Windows 3.0 BMP files
7119 0.90 fix a few more warnings; bump version number to approach 1.0
7120 0.61 bugfixes due to Marc LeBlanc, Christopher Lloyd
7121 0.60 fix compiling as c++
7122 0.59 fix warnings: merge Dave Moore's -Wall fixes
7123 0.58 fix bug: zlib uncompressed mode len/nlen was wrong endian
7124 0.57 fix bug: jpg last huffman symbol before marker was >9 bits but less than 16 available
7125 0.56 fix bug: zlib uncompressed mode len vs. nlen
7126 0.55 fix bug: restart_interval not initialized to 0
7127 0.54 allow NULL for 'int *comp'
7128 0.53 fix bug in png 3->4; speedup png decoding
7129 0.52 png handles req_comp=3,4 directly; minor cleanup; jpeg comments
7130 0.51 obey req_comp requests, 1-component jpegs return as 1-component,
7131 on 'test' only check type, not whether we support this variant
7132 0.50 (2006-11-19)
7133 first released version
7134*/