1/*
2 * jdmerge.c
3 *
4 * This file was part of the Independent JPEG Group's software:
5 * Copyright (C) 1994-1996, Thomas G. Lane.
6 * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
7 * libjpeg-turbo Modifications:
8 * Copyright (C) 2009, 2011, D. R. Commander.
9 * For conditions of distribution and use, see the accompanying README file.
10 *
11 * This file contains code for merged upsampling/color conversion.
12 *
13 * This file combines functions from jdsample.c and jdcolor.c;
14 * read those files first to understand what's going on.
15 *
16 * When the chroma components are to be upsampled by simple replication
17 * (ie, box filtering), we can save some work in color conversion by
18 * calculating all the output pixels corresponding to a pair of chroma
19 * samples at one time.  In the conversion equations
20 *	R = Y           + K1 * Cr
21 *	G = Y + K2 * Cb + K3 * Cr
22 *	B = Y + K4 * Cb
23 * only the Y term varies among the group of pixels corresponding to a pair
24 * of chroma samples, so the rest of the terms can be calculated just once.
25 * At typical sampling ratios, this eliminates half or three-quarters of the
26 * multiplications needed for color conversion.
27 *
28 * This file currently provides implementations for the following cases:
29 *	YCbCr => RGB color conversion only.
30 *	Sampling ratios of 2h1v or 2h2v.
31 *	No scaling needed at upsample time.
32 *	Corner-aligned (non-CCIR601) sampling alignment.
33 * Other special cases could be added, but in most applications these are
34 * the only common cases.  (For uncommon cases we fall back on the more
35 * general code in jdsample.c and jdcolor.c.)
36 */
37
38#define JPEG_INTERNALS
39#include "jinclude.h"
40#include "jpeglib.h"
41#include "jsimd.h"
42#include "config.h"
43
44#ifdef UPSAMPLE_MERGING_SUPPORTED
45
46
47/* Private subobject */
48
49typedef struct {
50  struct jpeg_upsampler pub;	/* public fields */
51
52  /* Pointer to routine to do actual upsampling/conversion of one row group */
53  JMETHOD(void, upmethod, (j_decompress_ptr cinfo,
54			   JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr,
55			   JSAMPARRAY output_buf));
56
57  /* Private state for YCC->RGB conversion */
58  int * Cr_r_tab;		/* => table for Cr to R conversion */
59  int * Cb_b_tab;		/* => table for Cb to B conversion */
60  INT32 * Cr_g_tab;		/* => table for Cr to G conversion */
61  INT32 * Cb_g_tab;		/* => table for Cb to G conversion */
62
63  /* For 2:1 vertical sampling, we produce two output rows at a time.
64   * We need a "spare" row buffer to hold the second output row if the
65   * application provides just a one-row buffer; we also use the spare
66   * to discard the dummy last row if the image height is odd.
67   */
68  JSAMPROW spare_row;
69  boolean spare_full;		/* T if spare buffer is occupied */
70
71  JDIMENSION out_row_width;	/* samples per output row */
72  JDIMENSION rows_to_go;	/* counts rows remaining in image */
73} my_upsampler;
74
75typedef my_upsampler * my_upsample_ptr;
76
77#define SCALEBITS	16	/* speediest right-shift on some machines */
78#define ONE_HALF	((INT32) 1 << (SCALEBITS-1))
79#define FIX(x)		((INT32) ((x) * (1L<<SCALEBITS) + 0.5))
80
81
82/* Include inline routines for colorspace extensions */
83
84#include "jdmrgext.c"
85#undef RGB_RED
86#undef RGB_GREEN
87#undef RGB_BLUE
88#undef RGB_PIXELSIZE
89
90#define RGB_RED EXT_RGB_RED
91#define RGB_GREEN EXT_RGB_GREEN
92#define RGB_BLUE EXT_RGB_BLUE
93#define RGB_PIXELSIZE EXT_RGB_PIXELSIZE
94#define h2v1_merged_upsample_internal extrgb_h2v1_merged_upsample_internal
95#define h2v2_merged_upsample_internal extrgb_h2v2_merged_upsample_internal
96#include "jdmrgext.c"
97#undef RGB_RED
98#undef RGB_GREEN
99#undef RGB_BLUE
100#undef RGB_PIXELSIZE
101#undef h2v1_merged_upsample_internal
102#undef h2v2_merged_upsample_internal
103
104#define RGB_RED EXT_RGBX_RED
105#define RGB_GREEN EXT_RGBX_GREEN
106#define RGB_BLUE EXT_RGBX_BLUE
107#define RGB_ALPHA 3
108#define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE
109#define h2v1_merged_upsample_internal extrgbx_h2v1_merged_upsample_internal
110#define h2v2_merged_upsample_internal extrgbx_h2v2_merged_upsample_internal
111#include "jdmrgext.c"
112#undef RGB_RED
113#undef RGB_GREEN
114#undef RGB_BLUE
115#undef RGB_ALPHA
116#undef RGB_PIXELSIZE
117#undef h2v1_merged_upsample_internal
118#undef h2v2_merged_upsample_internal
119
120#define RGB_RED EXT_BGR_RED
121#define RGB_GREEN EXT_BGR_GREEN
122#define RGB_BLUE EXT_BGR_BLUE
123#define RGB_PIXELSIZE EXT_BGR_PIXELSIZE
124#define h2v1_merged_upsample_internal extbgr_h2v1_merged_upsample_internal
125#define h2v2_merged_upsample_internal extbgr_h2v2_merged_upsample_internal
126#include "jdmrgext.c"
127#undef RGB_RED
128#undef RGB_GREEN
129#undef RGB_BLUE
130#undef RGB_PIXELSIZE
131#undef h2v1_merged_upsample_internal
132#undef h2v2_merged_upsample_internal
133
134#define RGB_RED EXT_BGRX_RED
135#define RGB_GREEN EXT_BGRX_GREEN
136#define RGB_BLUE EXT_BGRX_BLUE
137#define RGB_ALPHA 3
138#define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE
139#define h2v1_merged_upsample_internal extbgrx_h2v1_merged_upsample_internal
140#define h2v2_merged_upsample_internal extbgrx_h2v2_merged_upsample_internal
141#include "jdmrgext.c"
142#undef RGB_RED
143#undef RGB_GREEN
144#undef RGB_BLUE
145#undef RGB_ALPHA
146#undef RGB_PIXELSIZE
147#undef h2v1_merged_upsample_internal
148#undef h2v2_merged_upsample_internal
149
150#define RGB_RED EXT_XBGR_RED
151#define RGB_GREEN EXT_XBGR_GREEN
152#define RGB_BLUE EXT_XBGR_BLUE
153#define RGB_ALPHA 0
154#define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE
155#define h2v1_merged_upsample_internal extxbgr_h2v1_merged_upsample_internal
156#define h2v2_merged_upsample_internal extxbgr_h2v2_merged_upsample_internal
157#include "jdmrgext.c"
158#undef RGB_RED
159#undef RGB_GREEN
160#undef RGB_BLUE
161#undef RGB_ALPHA
162#undef RGB_PIXELSIZE
163#undef h2v1_merged_upsample_internal
164#undef h2v2_merged_upsample_internal
165
166#define RGB_RED EXT_XRGB_RED
167#define RGB_GREEN EXT_XRGB_GREEN
168#define RGB_BLUE EXT_XRGB_BLUE
169#define RGB_ALPHA 0
170#define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE
171#define h2v1_merged_upsample_internal extxrgb_h2v1_merged_upsample_internal
172#define h2v2_merged_upsample_internal extxrgb_h2v2_merged_upsample_internal
173#include "jdmrgext.c"
174#undef RGB_RED
175#undef RGB_GREEN
176#undef RGB_BLUE
177#undef RGB_ALPHA
178#undef RGB_PIXELSIZE
179#undef h2v1_merged_upsample_internal
180#undef h2v2_merged_upsample_internal
181
182
183/*
184 * Initialize tables for YCC->RGB colorspace conversion.
185 * This is taken directly from jdcolor.c; see that file for more info.
186 */
187
188LOCAL(void)
189build_ycc_rgb_table (j_decompress_ptr cinfo)
190{
191  my_upsample_ptr upsample = (my_upsample_ptr) cinfo->upsample;
192  int i;
193  INT32 x;
194  SHIFT_TEMPS
195
196  upsample->Cr_r_tab = (int *)
197    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
198				(MAXJSAMPLE+1) * SIZEOF(int));
199  upsample->Cb_b_tab = (int *)
200    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
201				(MAXJSAMPLE+1) * SIZEOF(int));
202  upsample->Cr_g_tab = (INT32 *)
203    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
204				(MAXJSAMPLE+1) * SIZEOF(INT32));
205  upsample->Cb_g_tab = (INT32 *)
206    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
207				(MAXJSAMPLE+1) * SIZEOF(INT32));
208
209  for (i = 0, x = -CENTERJSAMPLE; i <= MAXJSAMPLE; i++, x++) {
210    /* i is the actual input pixel value, in the range 0..MAXJSAMPLE */
211    /* The Cb or Cr value we are thinking of is x = i - CENTERJSAMPLE */
212    /* Cr=>R value is nearest int to 1.40200 * x */
213    upsample->Cr_r_tab[i] = (int)
214		    RIGHT_SHIFT(FIX(1.40200) * x + ONE_HALF, SCALEBITS);
215    /* Cb=>B value is nearest int to 1.77200 * x */
216    upsample->Cb_b_tab[i] = (int)
217		    RIGHT_SHIFT(FIX(1.77200) * x + ONE_HALF, SCALEBITS);
218    /* Cr=>G value is scaled-up -0.71414 * x */
219    upsample->Cr_g_tab[i] = (- FIX(0.71414)) * x;
220    /* Cb=>G value is scaled-up -0.34414 * x */
221    /* We also add in ONE_HALF so that need not do it in inner loop */
222    upsample->Cb_g_tab[i] = (- FIX(0.34414)) * x + ONE_HALF;
223  }
224}
225
226
227/*
228 * Initialize for an upsampling pass.
229 */
230
231METHODDEF(void)
232start_pass_merged_upsample (j_decompress_ptr cinfo)
233{
234  my_upsample_ptr upsample = (my_upsample_ptr) cinfo->upsample;
235
236  /* Mark the spare buffer empty */
237  upsample->spare_full = FALSE;
238  /* Initialize total-height counter for detecting bottom of image */
239  upsample->rows_to_go = cinfo->output_height;
240}
241
242
243/*
244 * Control routine to do upsampling (and color conversion).
245 *
246 * The control routine just handles the row buffering considerations.
247 */
248
249METHODDEF(void)
250merged_2v_upsample (j_decompress_ptr cinfo,
251		    JSAMPIMAGE input_buf, JDIMENSION *in_row_group_ctr,
252		    JDIMENSION in_row_groups_avail,
253		    JSAMPARRAY output_buf, JDIMENSION *out_row_ctr,
254		    JDIMENSION out_rows_avail)
255/* 2:1 vertical sampling case: may need a spare row. */
256{
257  my_upsample_ptr upsample = (my_upsample_ptr) cinfo->upsample;
258  JSAMPROW work_ptrs[2];
259  JDIMENSION num_rows;		/* number of rows returned to caller */
260
261  if (upsample->spare_full) {
262    /* If we have a spare row saved from a previous cycle, just return it. */
263    jcopy_sample_rows(& upsample->spare_row, 0, output_buf + *out_row_ctr, 0,
264		      1, upsample->out_row_width);
265    num_rows = 1;
266    upsample->spare_full = FALSE;
267  } else {
268    /* Figure number of rows to return to caller. */
269    num_rows = 2;
270    /* Not more than the distance to the end of the image. */
271    if (num_rows > upsample->rows_to_go)
272      num_rows = upsample->rows_to_go;
273    /* And not more than what the client can accept: */
274    out_rows_avail -= *out_row_ctr;
275    if (num_rows > out_rows_avail)
276      num_rows = out_rows_avail;
277    /* Create output pointer array for upsampler. */
278    work_ptrs[0] = output_buf[*out_row_ctr];
279    if (num_rows > 1) {
280      work_ptrs[1] = output_buf[*out_row_ctr + 1];
281    } else {
282      work_ptrs[1] = upsample->spare_row;
283      upsample->spare_full = TRUE;
284    }
285    /* Now do the upsampling. */
286    (*upsample->upmethod) (cinfo, input_buf, *in_row_group_ctr, work_ptrs);
287  }
288
289  /* Adjust counts */
290  *out_row_ctr += num_rows;
291  upsample->rows_to_go -= num_rows;
292  /* When the buffer is emptied, declare this input row group consumed */
293  if (! upsample->spare_full)
294    (*in_row_group_ctr)++;
295}
296
297
298METHODDEF(void)
299merged_1v_upsample (j_decompress_ptr cinfo,
300		    JSAMPIMAGE input_buf, JDIMENSION *in_row_group_ctr,
301		    JDIMENSION in_row_groups_avail,
302		    JSAMPARRAY output_buf, JDIMENSION *out_row_ctr,
303		    JDIMENSION out_rows_avail)
304/* 1:1 vertical sampling case: much easier, never need a spare row. */
305{
306  my_upsample_ptr upsample = (my_upsample_ptr) cinfo->upsample;
307
308  /* Just do the upsampling. */
309  (*upsample->upmethod) (cinfo, input_buf, *in_row_group_ctr,
310			 output_buf + *out_row_ctr);
311  /* Adjust counts */
312  (*out_row_ctr)++;
313  (*in_row_group_ctr)++;
314}
315
316
317/*
318 * These are the routines invoked by the control routines to do
319 * the actual upsampling/conversion.  One row group is processed per call.
320 *
321 * Note: since we may be writing directly into application-supplied buffers,
322 * we have to be honest about the output width; we can't assume the buffer
323 * has been rounded up to an even width.
324 */
325
326
327/*
328 * Upsample and color convert for the case of 2:1 horizontal and 1:1 vertical.
329 */
330
331METHODDEF(void)
332h2v1_merged_upsample (j_decompress_ptr cinfo,
333		      JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr,
334		      JSAMPARRAY output_buf)
335{
336  switch (cinfo->out_color_space) {
337    case JCS_EXT_RGB:
338      extrgb_h2v1_merged_upsample_internal(cinfo, input_buf, in_row_group_ctr,
339                                           output_buf);
340      break;
341    case JCS_EXT_RGBX:
342    case JCS_EXT_RGBA:
343      extrgbx_h2v1_merged_upsample_internal(cinfo, input_buf, in_row_group_ctr,
344                                            output_buf);
345      break;
346    case JCS_EXT_BGR:
347      extbgr_h2v1_merged_upsample_internal(cinfo, input_buf, in_row_group_ctr,
348                                           output_buf);
349      break;
350    case JCS_EXT_BGRX:
351    case JCS_EXT_BGRA:
352      extbgrx_h2v1_merged_upsample_internal(cinfo, input_buf, in_row_group_ctr,
353                                            output_buf);
354      break;
355    case JCS_EXT_XBGR:
356    case JCS_EXT_ABGR:
357      extxbgr_h2v1_merged_upsample_internal(cinfo, input_buf, in_row_group_ctr,
358                                            output_buf);
359      break;
360    case JCS_EXT_XRGB:
361    case JCS_EXT_ARGB:
362      extxrgb_h2v1_merged_upsample_internal(cinfo, input_buf, in_row_group_ctr,
363                                            output_buf);
364      break;
365    default:
366      h2v1_merged_upsample_internal(cinfo, input_buf, in_row_group_ctr,
367                                    output_buf);
368      break;
369  }
370}
371
372
373/*
374 * Upsample and color convert for the case of 2:1 horizontal and 2:1 vertical.
375 */
376
377METHODDEF(void)
378h2v2_merged_upsample (j_decompress_ptr cinfo,
379		      JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr,
380		      JSAMPARRAY output_buf)
381{
382  switch (cinfo->out_color_space) {
383    case JCS_EXT_RGB:
384      extrgb_h2v2_merged_upsample_internal(cinfo, input_buf, in_row_group_ctr,
385                                           output_buf);
386      break;
387    case JCS_EXT_RGBX:
388    case JCS_EXT_RGBA:
389      extrgbx_h2v2_merged_upsample_internal(cinfo, input_buf, in_row_group_ctr,
390                                            output_buf);
391      break;
392    case JCS_EXT_BGR:
393      extbgr_h2v2_merged_upsample_internal(cinfo, input_buf, in_row_group_ctr,
394                                           output_buf);
395      break;
396    case JCS_EXT_BGRX:
397    case JCS_EXT_BGRA:
398      extbgrx_h2v2_merged_upsample_internal(cinfo, input_buf, in_row_group_ctr,
399                                            output_buf);
400      break;
401    case JCS_EXT_XBGR:
402    case JCS_EXT_ABGR:
403      extxbgr_h2v2_merged_upsample_internal(cinfo, input_buf, in_row_group_ctr,
404                                            output_buf);
405      break;
406    case JCS_EXT_XRGB:
407    case JCS_EXT_ARGB:
408      extxrgb_h2v2_merged_upsample_internal(cinfo, input_buf, in_row_group_ctr,
409                                            output_buf);
410      break;
411    default:
412      h2v2_merged_upsample_internal(cinfo, input_buf, in_row_group_ctr,
413                                    output_buf);
414      break;
415  }
416}
417
418
419/*
420 * Module initialization routine for merged upsampling/color conversion.
421 *
422 * NB: this is called under the conditions determined by use_merged_upsample()
423 * in jdmaster.c.  That routine MUST correspond to the actual capabilities
424 * of this module; no safety checks are made here.
425 */
426
427GLOBAL(void)
428jinit_merged_upsampler (j_decompress_ptr cinfo)
429{
430  my_upsample_ptr upsample;
431
432  upsample = (my_upsample_ptr)
433    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
434				SIZEOF(my_upsampler));
435  cinfo->upsample = (struct jpeg_upsampler *) upsample;
436  upsample->pub.start_pass = start_pass_merged_upsample;
437  upsample->pub.need_context_rows = FALSE;
438
439  upsample->out_row_width = cinfo->output_width * cinfo->out_color_components;
440
441  if (cinfo->max_v_samp_factor == 2) {
442    upsample->pub.upsample = merged_2v_upsample;
443    if (jsimd_can_h2v2_merged_upsample())
444      upsample->upmethod = jsimd_h2v2_merged_upsample;
445    else
446      upsample->upmethod = h2v2_merged_upsample;
447    /* Allocate a spare row buffer */
448    upsample->spare_row = (JSAMPROW)
449      (*cinfo->mem->alloc_large) ((j_common_ptr) cinfo, JPOOL_IMAGE,
450		(size_t) (upsample->out_row_width * SIZEOF(JSAMPLE)));
451  } else {
452    upsample->pub.upsample = merged_1v_upsample;
453    if (jsimd_can_h2v1_merged_upsample())
454      upsample->upmethod = jsimd_h2v1_merged_upsample;
455    else
456      upsample->upmethod = h2v1_merged_upsample;
457    /* No spare row needed */
458    upsample->spare_row = NULL;
459  }
460
461  build_ycc_rgb_table(cinfo);
462}
463
464#endif /* UPSAMPLE_MERGING_SUPPORTED */
465