• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * jsimd_i386.c
3  *
4  * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
5  * Copyright (C) 2009-2011, 2013-2014, 2016, 2018, D. R. Commander.
6  * Copyright (C) 2015-2016, 2018, Matthieu Darbois.
7  *
8  * Based on the x86 SIMD extension for IJG JPEG library,
9  * Copyright (C) 1999-2006, MIYASAKA Masaru.
10  * For conditions of distribution and use, see copyright notice in jsimdext.inc
11  *
12  * This file contains the interface between the "normal" portions
13  * of the library and the SIMD implementations when running on a
14  * 32-bit x86 architecture.
15  */
16 
17 #define JPEG_INTERNALS
18 #include "../../jinclude.h"
19 #include "../../jpeglib.h"
20 #include "../../jsimd.h"
21 #include "../../jdct.h"
22 #include "../../jsimddct.h"
23 #include "../jsimd.h"
24 #include "jconfigint.h"
25 
26 /*
27  * In the PIC cases, we have no guarantee that constants will keep
28  * their alignment. This macro allows us to verify it at runtime.
29  */
30 #define IS_ALIGNED(ptr, order)  (((unsigned)ptr & ((1 << order) - 1)) == 0)
31 
32 #define IS_ALIGNED_SSE(ptr)  (IS_ALIGNED(ptr, 4)) /* 16 byte alignment */
33 #define IS_ALIGNED_AVX(ptr)  (IS_ALIGNED(ptr, 5)) /* 32 byte alignment */
34 
35 static unsigned int simd_support = (unsigned int)(~0);
36 static unsigned int simd_huffman = 1;
37 
38 /*
39  * Check what SIMD accelerations are supported.
40  *
41  * FIXME: This code is racy under a multi-threaded environment.
42  */
43 LOCAL(void)
init_simd(void)44 init_simd(void)
45 {
46 #ifndef NO_GETENV
47   char *env = NULL;
48 #endif
49 
50   if (simd_support != ~0U)
51     return;
52 
53   simd_support = jpeg_simd_cpu_support();
54 
55 #ifndef NO_GETENV
56   /* Force different settings through environment variables */
57   env = getenv("JSIMD_FORCEMMX");
58   if ((env != NULL) && (strcmp(env, "1") == 0))
59     simd_support &= JSIMD_MMX;
60   env = getenv("JSIMD_FORCE3DNOW");
61   if ((env != NULL) && (strcmp(env, "1") == 0))
62     simd_support &= JSIMD_3DNOW | JSIMD_MMX;
63   env = getenv("JSIMD_FORCESSE");
64   if ((env != NULL) && (strcmp(env, "1") == 0))
65     simd_support &= JSIMD_SSE | JSIMD_MMX;
66   env = getenv("JSIMD_FORCESSE2");
67   if ((env != NULL) && (strcmp(env, "1") == 0))
68     simd_support &= JSIMD_SSE2;
69   env = getenv("JSIMD_FORCEAVX2");
70   if ((env != NULL) && (strcmp(env, "1") == 0))
71     simd_support &= JSIMD_AVX2;
72   env = getenv("JSIMD_FORCENONE");
73   if ((env != NULL) && (strcmp(env, "1") == 0))
74     simd_support = 0;
75   env = getenv("JSIMD_NOHUFFENC");
76   if ((env != NULL) && (strcmp(env, "1") == 0))
77     simd_huffman = 0;
78 #endif
79 }
80 
81 GLOBAL(int)
jsimd_can_rgb_ycc(void)82 jsimd_can_rgb_ycc(void)
83 {
84   init_simd();
85 
86   /* The code is optimised for these values only */
87   if (BITS_IN_JSAMPLE != 8)
88     return 0;
89   if (sizeof(JDIMENSION) != 4)
90     return 0;
91   if ((RGB_PIXELSIZE != 3) && (RGB_PIXELSIZE != 4))
92     return 0;
93 
94   if ((simd_support & JSIMD_AVX2) &&
95       IS_ALIGNED_AVX(jconst_rgb_ycc_convert_avx2))
96     return 1;
97   if ((simd_support & JSIMD_SSE2) &&
98       IS_ALIGNED_SSE(jconst_rgb_ycc_convert_sse2))
99     return 1;
100   if (simd_support & JSIMD_MMX)
101     return 1;
102 
103   return 0;
104 }
105 
106 GLOBAL(int)
jsimd_can_rgb_gray(void)107 jsimd_can_rgb_gray(void)
108 {
109   init_simd();
110 
111   /* The code is optimised for these values only */
112   if (BITS_IN_JSAMPLE != 8)
113     return 0;
114   if (sizeof(JDIMENSION) != 4)
115     return 0;
116   if ((RGB_PIXELSIZE != 3) && (RGB_PIXELSIZE != 4))
117     return 0;
118 
119   if ((simd_support & JSIMD_AVX2) &&
120       IS_ALIGNED_AVX(jconst_rgb_gray_convert_avx2))
121     return 1;
122   if ((simd_support & JSIMD_SSE2) &&
123       IS_ALIGNED_SSE(jconst_rgb_gray_convert_sse2))
124     return 1;
125   if (simd_support & JSIMD_MMX)
126     return 1;
127 
128   return 0;
129 }
130 
131 GLOBAL(int)
jsimd_can_ycc_rgb(void)132 jsimd_can_ycc_rgb(void)
133 {
134   init_simd();
135 
136   /* The code is optimised for these values only */
137   if (BITS_IN_JSAMPLE != 8)
138     return 0;
139   if (sizeof(JDIMENSION) != 4)
140     return 0;
141   if ((RGB_PIXELSIZE != 3) && (RGB_PIXELSIZE != 4))
142     return 0;
143 
144   if ((simd_support & JSIMD_AVX2) &&
145       IS_ALIGNED_AVX(jconst_ycc_rgb_convert_avx2))
146     return 1;
147   if ((simd_support & JSIMD_SSE2) &&
148       IS_ALIGNED_SSE(jconst_ycc_rgb_convert_sse2))
149     return 1;
150   if (simd_support & JSIMD_MMX)
151     return 1;
152 
153   return 0;
154 }
155 
156 GLOBAL(int)
jsimd_can_ycc_rgb565(void)157 jsimd_can_ycc_rgb565(void)
158 {
159   return 0;
160 }
161 
162 GLOBAL(void)
jsimd_rgb_ycc_convert(j_compress_ptr cinfo,JSAMPARRAY input_buf,JSAMPIMAGE output_buf,JDIMENSION output_row,int num_rows)163 jsimd_rgb_ycc_convert(j_compress_ptr cinfo, JSAMPARRAY input_buf,
164                       JSAMPIMAGE output_buf, JDIMENSION output_row,
165                       int num_rows)
166 {
167   void (*avx2fct) (JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int);
168   void (*sse2fct) (JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int);
169   void (*mmxfct) (JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int);
170 
171   switch (cinfo->in_color_space) {
172   case JCS_EXT_RGB:
173     avx2fct = jsimd_extrgb_ycc_convert_avx2;
174     sse2fct = jsimd_extrgb_ycc_convert_sse2;
175     mmxfct = jsimd_extrgb_ycc_convert_mmx;
176     break;
177   case JCS_EXT_RGBX:
178   case JCS_EXT_RGBA:
179     avx2fct = jsimd_extrgbx_ycc_convert_avx2;
180     sse2fct = jsimd_extrgbx_ycc_convert_sse2;
181     mmxfct = jsimd_extrgbx_ycc_convert_mmx;
182     break;
183   case JCS_EXT_BGR:
184     avx2fct = jsimd_extbgr_ycc_convert_avx2;
185     sse2fct = jsimd_extbgr_ycc_convert_sse2;
186     mmxfct = jsimd_extbgr_ycc_convert_mmx;
187     break;
188   case JCS_EXT_BGRX:
189   case JCS_EXT_BGRA:
190     avx2fct = jsimd_extbgrx_ycc_convert_avx2;
191     sse2fct = jsimd_extbgrx_ycc_convert_sse2;
192     mmxfct = jsimd_extbgrx_ycc_convert_mmx;
193     break;
194   case JCS_EXT_XBGR:
195   case JCS_EXT_ABGR:
196     avx2fct = jsimd_extxbgr_ycc_convert_avx2;
197     sse2fct = jsimd_extxbgr_ycc_convert_sse2;
198     mmxfct = jsimd_extxbgr_ycc_convert_mmx;
199     break;
200   case JCS_EXT_XRGB:
201   case JCS_EXT_ARGB:
202     avx2fct = jsimd_extxrgb_ycc_convert_avx2;
203     sse2fct = jsimd_extxrgb_ycc_convert_sse2;
204     mmxfct = jsimd_extxrgb_ycc_convert_mmx;
205     break;
206   default:
207     avx2fct = jsimd_rgb_ycc_convert_avx2;
208     sse2fct = jsimd_rgb_ycc_convert_sse2;
209     mmxfct = jsimd_rgb_ycc_convert_mmx;
210     break;
211   }
212 
213   if (simd_support & JSIMD_AVX2)
214     avx2fct(cinfo->image_width, input_buf, output_buf, output_row, num_rows);
215   else if (simd_support & JSIMD_SSE2)
216     sse2fct(cinfo->image_width, input_buf, output_buf, output_row, num_rows);
217   else
218     mmxfct(cinfo->image_width, input_buf, output_buf, output_row, num_rows);
219 }
220 
221 GLOBAL(void)
jsimd_rgb_gray_convert(j_compress_ptr cinfo,JSAMPARRAY input_buf,JSAMPIMAGE output_buf,JDIMENSION output_row,int num_rows)222 jsimd_rgb_gray_convert(j_compress_ptr cinfo, JSAMPARRAY input_buf,
223                        JSAMPIMAGE output_buf, JDIMENSION output_row,
224                        int num_rows)
225 {
226   void (*avx2fct) (JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int);
227   void (*sse2fct) (JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int);
228   void (*mmxfct) (JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int);
229 
230   switch (cinfo->in_color_space) {
231   case JCS_EXT_RGB:
232     avx2fct = jsimd_extrgb_gray_convert_avx2;
233     sse2fct = jsimd_extrgb_gray_convert_sse2;
234     mmxfct = jsimd_extrgb_gray_convert_mmx;
235     break;
236   case JCS_EXT_RGBX:
237   case JCS_EXT_RGBA:
238     avx2fct = jsimd_extrgbx_gray_convert_avx2;
239     sse2fct = jsimd_extrgbx_gray_convert_sse2;
240     mmxfct = jsimd_extrgbx_gray_convert_mmx;
241     break;
242   case JCS_EXT_BGR:
243     avx2fct = jsimd_extbgr_gray_convert_avx2;
244     sse2fct = jsimd_extbgr_gray_convert_sse2;
245     mmxfct = jsimd_extbgr_gray_convert_mmx;
246     break;
247   case JCS_EXT_BGRX:
248   case JCS_EXT_BGRA:
249     avx2fct = jsimd_extbgrx_gray_convert_avx2;
250     sse2fct = jsimd_extbgrx_gray_convert_sse2;
251     mmxfct = jsimd_extbgrx_gray_convert_mmx;
252     break;
253   case JCS_EXT_XBGR:
254   case JCS_EXT_ABGR:
255     avx2fct = jsimd_extxbgr_gray_convert_avx2;
256     sse2fct = jsimd_extxbgr_gray_convert_sse2;
257     mmxfct = jsimd_extxbgr_gray_convert_mmx;
258     break;
259   case JCS_EXT_XRGB:
260   case JCS_EXT_ARGB:
261     avx2fct = jsimd_extxrgb_gray_convert_avx2;
262     sse2fct = jsimd_extxrgb_gray_convert_sse2;
263     mmxfct = jsimd_extxrgb_gray_convert_mmx;
264     break;
265   default:
266     avx2fct = jsimd_rgb_gray_convert_avx2;
267     sse2fct = jsimd_rgb_gray_convert_sse2;
268     mmxfct = jsimd_rgb_gray_convert_mmx;
269     break;
270   }
271 
272   if (simd_support & JSIMD_AVX2)
273     avx2fct(cinfo->image_width, input_buf, output_buf, output_row, num_rows);
274   else if (simd_support & JSIMD_SSE2)
275     sse2fct(cinfo->image_width, input_buf, output_buf, output_row, num_rows);
276   else
277     mmxfct(cinfo->image_width, input_buf, output_buf, output_row, num_rows);
278 }
279 
280 GLOBAL(void)
jsimd_ycc_rgb_convert(j_decompress_ptr cinfo,JSAMPIMAGE input_buf,JDIMENSION input_row,JSAMPARRAY output_buf,int num_rows)281 jsimd_ycc_rgb_convert(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
282                       JDIMENSION input_row, JSAMPARRAY output_buf,
283                       int num_rows)
284 {
285   void (*avx2fct) (JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY, int);
286   void (*sse2fct) (JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY, int);
287   void (*mmxfct) (JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY, int);
288 
289   switch (cinfo->out_color_space) {
290   case JCS_EXT_RGB:
291     avx2fct = jsimd_ycc_extrgb_convert_avx2;
292     sse2fct = jsimd_ycc_extrgb_convert_sse2;
293     mmxfct = jsimd_ycc_extrgb_convert_mmx;
294     break;
295   case JCS_EXT_RGBX:
296   case JCS_EXT_RGBA:
297     avx2fct = jsimd_ycc_extrgbx_convert_avx2;
298     sse2fct = jsimd_ycc_extrgbx_convert_sse2;
299     mmxfct = jsimd_ycc_extrgbx_convert_mmx;
300     break;
301   case JCS_EXT_BGR:
302     avx2fct = jsimd_ycc_extbgr_convert_avx2;
303     sse2fct = jsimd_ycc_extbgr_convert_sse2;
304     mmxfct = jsimd_ycc_extbgr_convert_mmx;
305     break;
306   case JCS_EXT_BGRX:
307   case JCS_EXT_BGRA:
308     avx2fct = jsimd_ycc_extbgrx_convert_avx2;
309     sse2fct = jsimd_ycc_extbgrx_convert_sse2;
310     mmxfct = jsimd_ycc_extbgrx_convert_mmx;
311     break;
312   case JCS_EXT_XBGR:
313   case JCS_EXT_ABGR:
314     avx2fct = jsimd_ycc_extxbgr_convert_avx2;
315     sse2fct = jsimd_ycc_extxbgr_convert_sse2;
316     mmxfct = jsimd_ycc_extxbgr_convert_mmx;
317     break;
318   case JCS_EXT_XRGB:
319   case JCS_EXT_ARGB:
320     avx2fct = jsimd_ycc_extxrgb_convert_avx2;
321     sse2fct = jsimd_ycc_extxrgb_convert_sse2;
322     mmxfct = jsimd_ycc_extxrgb_convert_mmx;
323     break;
324   default:
325     avx2fct = jsimd_ycc_rgb_convert_avx2;
326     sse2fct = jsimd_ycc_rgb_convert_sse2;
327     mmxfct = jsimd_ycc_rgb_convert_mmx;
328     break;
329   }
330 
331   if (simd_support & JSIMD_AVX2)
332     avx2fct(cinfo->output_width, input_buf, input_row, output_buf, num_rows);
333   else if (simd_support & JSIMD_SSE2)
334     sse2fct(cinfo->output_width, input_buf, input_row, output_buf, num_rows);
335   else
336     mmxfct(cinfo->output_width, input_buf, input_row, output_buf, num_rows);
337 }
338 
339 GLOBAL(void)
jsimd_ycc_rgb565_convert(j_decompress_ptr cinfo,JSAMPIMAGE input_buf,JDIMENSION input_row,JSAMPARRAY output_buf,int num_rows)340 jsimd_ycc_rgb565_convert(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
341                          JDIMENSION input_row, JSAMPARRAY output_buf,
342                          int num_rows)
343 {
344 }
345 
346 GLOBAL(int)
jsimd_can_h2v2_downsample(void)347 jsimd_can_h2v2_downsample(void)
348 {
349   init_simd();
350 
351   /* The code is optimised for these values only */
352   if (BITS_IN_JSAMPLE != 8)
353     return 0;
354   if (sizeof(JDIMENSION) != 4)
355     return 0;
356 
357   if (simd_support & JSIMD_AVX2)
358     return 1;
359   if (simd_support & JSIMD_SSE2)
360     return 1;
361   if (simd_support & JSIMD_MMX)
362     return 1;
363 
364   return 0;
365 }
366 
367 GLOBAL(int)
jsimd_can_h2v1_downsample(void)368 jsimd_can_h2v1_downsample(void)
369 {
370   init_simd();
371 
372   /* The code is optimised for these values only */
373   if (BITS_IN_JSAMPLE != 8)
374     return 0;
375   if (sizeof(JDIMENSION) != 4)
376     return 0;
377 
378   if (simd_support & JSIMD_AVX2)
379     return 1;
380   if (simd_support & JSIMD_SSE2)
381     return 1;
382   if (simd_support & JSIMD_MMX)
383     return 1;
384 
385   return 0;
386 }
387 
388 GLOBAL(void)
jsimd_h2v2_downsample(j_compress_ptr cinfo,jpeg_component_info * compptr,JSAMPARRAY input_data,JSAMPARRAY output_data)389 jsimd_h2v2_downsample(j_compress_ptr cinfo, jpeg_component_info *compptr,
390                       JSAMPARRAY input_data, JSAMPARRAY output_data)
391 {
392   if (simd_support & JSIMD_AVX2)
393     jsimd_h2v2_downsample_avx2(cinfo->image_width, cinfo->max_v_samp_factor,
394                                compptr->v_samp_factor,
395                                compptr->width_in_blocks, input_data,
396                                output_data);
397   else if (simd_support & JSIMD_SSE2)
398     jsimd_h2v2_downsample_sse2(cinfo->image_width, cinfo->max_v_samp_factor,
399                                compptr->v_samp_factor,
400                                compptr->width_in_blocks, input_data,
401                                output_data);
402   else
403     jsimd_h2v2_downsample_mmx(cinfo->image_width, cinfo->max_v_samp_factor,
404                               compptr->v_samp_factor, compptr->width_in_blocks,
405                               input_data, output_data);
406 }
407 
408 GLOBAL(void)
jsimd_h2v1_downsample(j_compress_ptr cinfo,jpeg_component_info * compptr,JSAMPARRAY input_data,JSAMPARRAY output_data)409 jsimd_h2v1_downsample(j_compress_ptr cinfo, jpeg_component_info *compptr,
410                       JSAMPARRAY input_data, JSAMPARRAY output_data)
411 {
412   if (simd_support & JSIMD_AVX2)
413     jsimd_h2v1_downsample_avx2(cinfo->image_width, cinfo->max_v_samp_factor,
414                                compptr->v_samp_factor,
415                                compptr->width_in_blocks, input_data,
416                                output_data);
417   else if (simd_support & JSIMD_SSE2)
418     jsimd_h2v1_downsample_sse2(cinfo->image_width, cinfo->max_v_samp_factor,
419                                compptr->v_samp_factor,
420                                compptr->width_in_blocks, input_data,
421                                output_data);
422   else
423     jsimd_h2v1_downsample_mmx(cinfo->image_width, cinfo->max_v_samp_factor,
424                               compptr->v_samp_factor, compptr->width_in_blocks,
425                               input_data, output_data);
426 }
427 
428 GLOBAL(int)
jsimd_can_h2v2_upsample(void)429 jsimd_can_h2v2_upsample(void)
430 {
431   init_simd();
432 
433   /* The code is optimised for these values only */
434   if (BITS_IN_JSAMPLE != 8)
435     return 0;
436   if (sizeof(JDIMENSION) != 4)
437     return 0;
438 
439   if (simd_support & JSIMD_AVX2)
440     return 1;
441   if (simd_support & JSIMD_SSE2)
442     return 1;
443   if (simd_support & JSIMD_MMX)
444     return 1;
445 
446   return 0;
447 }
448 
449 GLOBAL(int)
jsimd_can_h2v1_upsample(void)450 jsimd_can_h2v1_upsample(void)
451 {
452   init_simd();
453 
454   /* The code is optimised for these values only */
455   if (BITS_IN_JSAMPLE != 8)
456     return 0;
457   if (sizeof(JDIMENSION) != 4)
458     return 0;
459 
460   if (simd_support & JSIMD_AVX2)
461     return 1;
462   if (simd_support & JSIMD_SSE2)
463     return 1;
464   if (simd_support & JSIMD_MMX)
465     return 1;
466 
467   return 0;
468 }
469 
470 GLOBAL(void)
jsimd_h2v2_upsample(j_decompress_ptr cinfo,jpeg_component_info * compptr,JSAMPARRAY input_data,JSAMPARRAY * output_data_ptr)471 jsimd_h2v2_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr,
472                     JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr)
473 {
474   if (simd_support & JSIMD_AVX2)
475     jsimd_h2v2_upsample_avx2(cinfo->max_v_samp_factor, cinfo->output_width,
476                              input_data, output_data_ptr);
477   else if (simd_support & JSIMD_SSE2)
478     jsimd_h2v2_upsample_sse2(cinfo->max_v_samp_factor, cinfo->output_width,
479                              input_data, output_data_ptr);
480   else
481     jsimd_h2v2_upsample_mmx(cinfo->max_v_samp_factor, cinfo->output_width,
482                             input_data, output_data_ptr);
483 }
484 
485 GLOBAL(void)
jsimd_h2v1_upsample(j_decompress_ptr cinfo,jpeg_component_info * compptr,JSAMPARRAY input_data,JSAMPARRAY * output_data_ptr)486 jsimd_h2v1_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr,
487                     JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr)
488 {
489   if (simd_support & JSIMD_AVX2)
490     jsimd_h2v1_upsample_avx2(cinfo->max_v_samp_factor, cinfo->output_width,
491                              input_data, output_data_ptr);
492   else if (simd_support & JSIMD_SSE2)
493     jsimd_h2v1_upsample_sse2(cinfo->max_v_samp_factor, cinfo->output_width,
494                              input_data, output_data_ptr);
495   else
496     jsimd_h2v1_upsample_mmx(cinfo->max_v_samp_factor, cinfo->output_width,
497                             input_data, output_data_ptr);
498 }
499 
500 GLOBAL(int)
jsimd_can_h2v2_fancy_upsample(void)501 jsimd_can_h2v2_fancy_upsample(void)
502 {
503   init_simd();
504 
505   /* The code is optimised for these values only */
506   if (BITS_IN_JSAMPLE != 8)
507     return 0;
508   if (sizeof(JDIMENSION) != 4)
509     return 0;
510 
511   if ((simd_support & JSIMD_AVX2) &&
512       IS_ALIGNED_AVX(jconst_fancy_upsample_avx2))
513     return 1;
514   if ((simd_support & JSIMD_SSE2) &&
515       IS_ALIGNED_SSE(jconst_fancy_upsample_sse2))
516     return 1;
517   if (simd_support & JSIMD_MMX)
518     return 1;
519 
520   return 0;
521 }
522 
523 GLOBAL(int)
jsimd_can_h2v1_fancy_upsample(void)524 jsimd_can_h2v1_fancy_upsample(void)
525 {
526   init_simd();
527 
528   /* The code is optimised for these values only */
529   if (BITS_IN_JSAMPLE != 8)
530     return 0;
531   if (sizeof(JDIMENSION) != 4)
532     return 0;
533 
534   if ((simd_support & JSIMD_AVX2) &&
535       IS_ALIGNED_AVX(jconst_fancy_upsample_avx2))
536     return 1;
537   if ((simd_support & JSIMD_SSE2) &&
538       IS_ALIGNED_SSE(jconst_fancy_upsample_sse2))
539     return 1;
540   if (simd_support & JSIMD_MMX)
541     return 1;
542 
543   return 0;
544 }
545 
546 GLOBAL(void)
jsimd_h2v2_fancy_upsample(j_decompress_ptr cinfo,jpeg_component_info * compptr,JSAMPARRAY input_data,JSAMPARRAY * output_data_ptr)547 jsimd_h2v2_fancy_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr,
548                           JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr)
549 {
550   if (simd_support & JSIMD_AVX2)
551     jsimd_h2v2_fancy_upsample_avx2(cinfo->max_v_samp_factor,
552                                    compptr->downsampled_width, input_data,
553                                    output_data_ptr);
554   else if (simd_support & JSIMD_SSE2)
555     jsimd_h2v2_fancy_upsample_sse2(cinfo->max_v_samp_factor,
556                                    compptr->downsampled_width, input_data,
557                                    output_data_ptr);
558   else
559     jsimd_h2v2_fancy_upsample_mmx(cinfo->max_v_samp_factor,
560                                   compptr->downsampled_width, input_data,
561                                   output_data_ptr);
562 }
563 
564 GLOBAL(void)
jsimd_h2v1_fancy_upsample(j_decompress_ptr cinfo,jpeg_component_info * compptr,JSAMPARRAY input_data,JSAMPARRAY * output_data_ptr)565 jsimd_h2v1_fancy_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr,
566                           JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr)
567 {
568   if (simd_support & JSIMD_AVX2)
569     jsimd_h2v1_fancy_upsample_avx2(cinfo->max_v_samp_factor,
570                                    compptr->downsampled_width, input_data,
571                                    output_data_ptr);
572   else if (simd_support & JSIMD_SSE2)
573     jsimd_h2v1_fancy_upsample_sse2(cinfo->max_v_samp_factor,
574                                    compptr->downsampled_width, input_data,
575                                    output_data_ptr);
576   else
577     jsimd_h2v1_fancy_upsample_mmx(cinfo->max_v_samp_factor,
578                                   compptr->downsampled_width, input_data,
579                                   output_data_ptr);
580 }
581 
582 GLOBAL(int)
jsimd_can_h2v2_merged_upsample(void)583 jsimd_can_h2v2_merged_upsample(void)
584 {
585   init_simd();
586 
587   /* The code is optimised for these values only */
588   if (BITS_IN_JSAMPLE != 8)
589     return 0;
590   if (sizeof(JDIMENSION) != 4)
591     return 0;
592 
593   if ((simd_support & JSIMD_AVX2) &&
594       IS_ALIGNED_AVX(jconst_merged_upsample_avx2))
595     return 1;
596   if ((simd_support & JSIMD_SSE2) &&
597       IS_ALIGNED_SSE(jconst_merged_upsample_sse2))
598     return 1;
599   if (simd_support & JSIMD_MMX)
600     return 1;
601 
602   return 0;
603 }
604 
605 GLOBAL(int)
jsimd_can_h2v1_merged_upsample(void)606 jsimd_can_h2v1_merged_upsample(void)
607 {
608   init_simd();
609 
610   /* The code is optimised for these values only */
611   if (BITS_IN_JSAMPLE != 8)
612     return 0;
613   if (sizeof(JDIMENSION) != 4)
614     return 0;
615 
616   if ((simd_support & JSIMD_AVX2) &&
617       IS_ALIGNED_AVX(jconst_merged_upsample_avx2))
618     return 1;
619   if ((simd_support & JSIMD_SSE2) &&
620       IS_ALIGNED_SSE(jconst_merged_upsample_sse2))
621     return 1;
622   if (simd_support & JSIMD_MMX)
623     return 1;
624 
625   return 0;
626 }
627 
628 GLOBAL(void)
jsimd_h2v2_merged_upsample(j_decompress_ptr cinfo,JSAMPIMAGE input_buf,JDIMENSION in_row_group_ctr,JSAMPARRAY output_buf)629 jsimd_h2v2_merged_upsample(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
630                            JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf)
631 {
632   void (*avx2fct) (JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY);
633   void (*sse2fct) (JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY);
634   void (*mmxfct) (JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY);
635 
636   switch (cinfo->out_color_space) {
637   case JCS_EXT_RGB:
638     avx2fct = jsimd_h2v2_extrgb_merged_upsample_avx2;
639     sse2fct = jsimd_h2v2_extrgb_merged_upsample_sse2;
640     mmxfct = jsimd_h2v2_extrgb_merged_upsample_mmx;
641     break;
642   case JCS_EXT_RGBX:
643   case JCS_EXT_RGBA:
644     avx2fct = jsimd_h2v2_extrgbx_merged_upsample_avx2;
645     sse2fct = jsimd_h2v2_extrgbx_merged_upsample_sse2;
646     mmxfct = jsimd_h2v2_extrgbx_merged_upsample_mmx;
647     break;
648   case JCS_EXT_BGR:
649     avx2fct = jsimd_h2v2_extbgr_merged_upsample_avx2;
650     sse2fct = jsimd_h2v2_extbgr_merged_upsample_sse2;
651     mmxfct = jsimd_h2v2_extbgr_merged_upsample_mmx;
652     break;
653   case JCS_EXT_BGRX:
654   case JCS_EXT_BGRA:
655     avx2fct = jsimd_h2v2_extbgrx_merged_upsample_avx2;
656     sse2fct = jsimd_h2v2_extbgrx_merged_upsample_sse2;
657     mmxfct = jsimd_h2v2_extbgrx_merged_upsample_mmx;
658     break;
659   case JCS_EXT_XBGR:
660   case JCS_EXT_ABGR:
661     avx2fct = jsimd_h2v2_extxbgr_merged_upsample_avx2;
662     sse2fct = jsimd_h2v2_extxbgr_merged_upsample_sse2;
663     mmxfct = jsimd_h2v2_extxbgr_merged_upsample_mmx;
664     break;
665   case JCS_EXT_XRGB:
666   case JCS_EXT_ARGB:
667     avx2fct = jsimd_h2v2_extxrgb_merged_upsample_avx2;
668     sse2fct = jsimd_h2v2_extxrgb_merged_upsample_sse2;
669     mmxfct = jsimd_h2v2_extxrgb_merged_upsample_mmx;
670     break;
671   default:
672     avx2fct = jsimd_h2v2_merged_upsample_avx2;
673     sse2fct = jsimd_h2v2_merged_upsample_sse2;
674     mmxfct = jsimd_h2v2_merged_upsample_mmx;
675     break;
676   }
677 
678   if (simd_support & JSIMD_AVX2)
679     avx2fct(cinfo->output_width, input_buf, in_row_group_ctr, output_buf);
680   else if (simd_support & JSIMD_SSE2)
681     sse2fct(cinfo->output_width, input_buf, in_row_group_ctr, output_buf);
682   else
683     mmxfct(cinfo->output_width, input_buf, in_row_group_ctr, output_buf);
684 }
685 
686 GLOBAL(void)
jsimd_h2v1_merged_upsample(j_decompress_ptr cinfo,JSAMPIMAGE input_buf,JDIMENSION in_row_group_ctr,JSAMPARRAY output_buf)687 jsimd_h2v1_merged_upsample(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
688                            JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf)
689 {
690   void (*avx2fct) (JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY);
691   void (*sse2fct) (JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY);
692   void (*mmxfct) (JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY);
693 
694   switch (cinfo->out_color_space) {
695   case JCS_EXT_RGB:
696     avx2fct = jsimd_h2v1_extrgb_merged_upsample_avx2;
697     sse2fct = jsimd_h2v1_extrgb_merged_upsample_sse2;
698     mmxfct = jsimd_h2v1_extrgb_merged_upsample_mmx;
699     break;
700   case JCS_EXT_RGBX:
701   case JCS_EXT_RGBA:
702     avx2fct = jsimd_h2v1_extrgbx_merged_upsample_avx2;
703     sse2fct = jsimd_h2v1_extrgbx_merged_upsample_sse2;
704     mmxfct = jsimd_h2v1_extrgbx_merged_upsample_mmx;
705     break;
706   case JCS_EXT_BGR:
707     avx2fct = jsimd_h2v1_extbgr_merged_upsample_avx2;
708     sse2fct = jsimd_h2v1_extbgr_merged_upsample_sse2;
709     mmxfct = jsimd_h2v1_extbgr_merged_upsample_mmx;
710     break;
711   case JCS_EXT_BGRX:
712   case JCS_EXT_BGRA:
713     avx2fct = jsimd_h2v1_extbgrx_merged_upsample_avx2;
714     sse2fct = jsimd_h2v1_extbgrx_merged_upsample_sse2;
715     mmxfct = jsimd_h2v1_extbgrx_merged_upsample_mmx;
716     break;
717   case JCS_EXT_XBGR:
718   case JCS_EXT_ABGR:
719     avx2fct = jsimd_h2v1_extxbgr_merged_upsample_avx2;
720     sse2fct = jsimd_h2v1_extxbgr_merged_upsample_sse2;
721     mmxfct = jsimd_h2v1_extxbgr_merged_upsample_mmx;
722     break;
723   case JCS_EXT_XRGB:
724   case JCS_EXT_ARGB:
725     avx2fct = jsimd_h2v1_extxrgb_merged_upsample_avx2;
726     sse2fct = jsimd_h2v1_extxrgb_merged_upsample_sse2;
727     mmxfct = jsimd_h2v1_extxrgb_merged_upsample_mmx;
728     break;
729   default:
730     avx2fct = jsimd_h2v1_merged_upsample_avx2;
731     sse2fct = jsimd_h2v1_merged_upsample_sse2;
732     mmxfct = jsimd_h2v1_merged_upsample_mmx;
733     break;
734   }
735 
736   if (simd_support & JSIMD_AVX2)
737     avx2fct(cinfo->output_width, input_buf, in_row_group_ctr, output_buf);
738   else if (simd_support & JSIMD_SSE2)
739     sse2fct(cinfo->output_width, input_buf, in_row_group_ctr, output_buf);
740   else
741     mmxfct(cinfo->output_width, input_buf, in_row_group_ctr, output_buf);
742 }
743 
744 GLOBAL(int)
jsimd_can_convsamp(void)745 jsimd_can_convsamp(void)
746 {
747   init_simd();
748 
749   /* The code is optimised for these values only */
750   if (DCTSIZE != 8)
751     return 0;
752   if (BITS_IN_JSAMPLE != 8)
753     return 0;
754   if (sizeof(JDIMENSION) != 4)
755     return 0;
756   if (sizeof(DCTELEM) != 2)
757     return 0;
758 
759   if (simd_support & JSIMD_AVX2)
760     return 1;
761   if (simd_support & JSIMD_SSE2)
762     return 1;
763   if (simd_support & JSIMD_MMX)
764     return 1;
765 
766   return 0;
767 }
768 
769 GLOBAL(int)
jsimd_can_convsamp_float(void)770 jsimd_can_convsamp_float(void)
771 {
772   init_simd();
773 
774   /* The code is optimised for these values only */
775   if (DCTSIZE != 8)
776     return 0;
777   if (BITS_IN_JSAMPLE != 8)
778     return 0;
779   if (sizeof(JDIMENSION) != 4)
780     return 0;
781   if (sizeof(FAST_FLOAT) != 4)
782     return 0;
783 
784   if (simd_support & JSIMD_SSE2)
785     return 1;
786   if (simd_support & JSIMD_SSE)
787     return 1;
788   if (simd_support & JSIMD_3DNOW)
789     return 1;
790 
791   return 0;
792 }
793 
794 GLOBAL(void)
jsimd_convsamp(JSAMPARRAY sample_data,JDIMENSION start_col,DCTELEM * workspace)795 jsimd_convsamp(JSAMPARRAY sample_data, JDIMENSION start_col,
796                DCTELEM *workspace)
797 {
798   if (simd_support & JSIMD_AVX2)
799     jsimd_convsamp_avx2(sample_data, start_col, workspace);
800   else if (simd_support & JSIMD_SSE2)
801     jsimd_convsamp_sse2(sample_data, start_col, workspace);
802   else
803     jsimd_convsamp_mmx(sample_data, start_col, workspace);
804 }
805 
806 GLOBAL(void)
jsimd_convsamp_float(JSAMPARRAY sample_data,JDIMENSION start_col,FAST_FLOAT * workspace)807 jsimd_convsamp_float(JSAMPARRAY sample_data, JDIMENSION start_col,
808                      FAST_FLOAT *workspace)
809 {
810   if (simd_support & JSIMD_SSE2)
811     jsimd_convsamp_float_sse2(sample_data, start_col, workspace);
812   else if (simd_support & JSIMD_SSE)
813     jsimd_convsamp_float_sse(sample_data, start_col, workspace);
814   else
815     jsimd_convsamp_float_3dnow(sample_data, start_col, workspace);
816 }
817 
818 GLOBAL(int)
jsimd_can_fdct_islow(void)819 jsimd_can_fdct_islow(void)
820 {
821   init_simd();
822 
823   /* The code is optimised for these values only */
824   if (DCTSIZE != 8)
825     return 0;
826   if (sizeof(DCTELEM) != 2)
827     return 0;
828 
829   if ((simd_support & JSIMD_AVX2) && IS_ALIGNED_AVX(jconst_fdct_islow_avx2))
830     return 1;
831   if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_fdct_islow_sse2))
832     return 1;
833   if (simd_support & JSIMD_MMX)
834     return 1;
835 
836   return 0;
837 }
838 
839 GLOBAL(int)
jsimd_can_fdct_ifast(void)840 jsimd_can_fdct_ifast(void)
841 {
842   init_simd();
843 
844   /* The code is optimised for these values only */
845   if (DCTSIZE != 8)
846     return 0;
847   if (sizeof(DCTELEM) != 2)
848     return 0;
849 
850   if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_fdct_ifast_sse2))
851     return 1;
852   if (simd_support & JSIMD_MMX)
853     return 1;
854 
855   return 0;
856 }
857 
858 GLOBAL(int)
jsimd_can_fdct_float(void)859 jsimd_can_fdct_float(void)
860 {
861   init_simd();
862 
863   /* The code is optimised for these values only */
864   if (DCTSIZE != 8)
865     return 0;
866   if (sizeof(FAST_FLOAT) != 4)
867     return 0;
868 
869   if ((simd_support & JSIMD_SSE) && IS_ALIGNED_SSE(jconst_fdct_float_sse))
870     return 1;
871   if (simd_support & JSIMD_3DNOW)
872     return 1;
873 
874   return 0;
875 }
876 
877 GLOBAL(void)
jsimd_fdct_islow(DCTELEM * data)878 jsimd_fdct_islow(DCTELEM *data)
879 {
880   if (simd_support & JSIMD_AVX2)
881     jsimd_fdct_islow_avx2(data);
882   else if (simd_support & JSIMD_SSE2)
883     jsimd_fdct_islow_sse2(data);
884   else
885     jsimd_fdct_islow_mmx(data);
886 }
887 
888 GLOBAL(void)
jsimd_fdct_ifast(DCTELEM * data)889 jsimd_fdct_ifast(DCTELEM *data)
890 {
891   if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_fdct_islow_sse2))
892     jsimd_fdct_ifast_sse2(data);
893   else
894     jsimd_fdct_ifast_mmx(data);
895 }
896 
897 GLOBAL(void)
jsimd_fdct_float(FAST_FLOAT * data)898 jsimd_fdct_float(FAST_FLOAT *data)
899 {
900   if ((simd_support & JSIMD_SSE) && IS_ALIGNED_SSE(jconst_fdct_float_sse))
901     jsimd_fdct_float_sse(data);
902   else if (simd_support & JSIMD_3DNOW)
903     jsimd_fdct_float_3dnow(data);
904 }
905 
906 GLOBAL(int)
jsimd_can_quantize(void)907 jsimd_can_quantize(void)
908 {
909   init_simd();
910 
911   /* The code is optimised for these values only */
912   if (DCTSIZE != 8)
913     return 0;
914   if (sizeof(JCOEF) != 2)
915     return 0;
916   if (sizeof(DCTELEM) != 2)
917     return 0;
918 
919   if (simd_support & JSIMD_AVX2)
920     return 1;
921   if (simd_support & JSIMD_SSE2)
922     return 1;
923   if (simd_support & JSIMD_MMX)
924     return 1;
925 
926   return 0;
927 }
928 
929 GLOBAL(int)
jsimd_can_quantize_float(void)930 jsimd_can_quantize_float(void)
931 {
932   init_simd();
933 
934   /* The code is optimised for these values only */
935   if (DCTSIZE != 8)
936     return 0;
937   if (sizeof(JCOEF) != 2)
938     return 0;
939   if (sizeof(FAST_FLOAT) != 4)
940     return 0;
941 
942   if (simd_support & JSIMD_SSE2)
943     return 1;
944   if (simd_support & JSIMD_SSE)
945     return 1;
946   if (simd_support & JSIMD_3DNOW)
947     return 1;
948 
949   return 0;
950 }
951 
952 GLOBAL(void)
jsimd_quantize(JCOEFPTR coef_block,DCTELEM * divisors,DCTELEM * workspace)953 jsimd_quantize(JCOEFPTR coef_block, DCTELEM *divisors, DCTELEM *workspace)
954 {
955   if (simd_support & JSIMD_AVX2)
956     jsimd_quantize_avx2(coef_block, divisors, workspace);
957   else if (simd_support & JSIMD_SSE2)
958     jsimd_quantize_sse2(coef_block, divisors, workspace);
959   else
960     jsimd_quantize_mmx(coef_block, divisors, workspace);
961 }
962 
963 GLOBAL(void)
jsimd_quantize_float(JCOEFPTR coef_block,FAST_FLOAT * divisors,FAST_FLOAT * workspace)964 jsimd_quantize_float(JCOEFPTR coef_block, FAST_FLOAT *divisors,
965                      FAST_FLOAT *workspace)
966 {
967   if (simd_support & JSIMD_SSE2)
968     jsimd_quantize_float_sse2(coef_block, divisors, workspace);
969   else if (simd_support & JSIMD_SSE)
970     jsimd_quantize_float_sse(coef_block, divisors, workspace);
971   else
972     jsimd_quantize_float_3dnow(coef_block, divisors, workspace);
973 }
974 
975 GLOBAL(int)
jsimd_can_idct_2x2(void)976 jsimd_can_idct_2x2(void)
977 {
978   init_simd();
979 
980   /* The code is optimised for these values only */
981   if (DCTSIZE != 8)
982     return 0;
983   if (sizeof(JCOEF) != 2)
984     return 0;
985   if (BITS_IN_JSAMPLE != 8)
986     return 0;
987   if (sizeof(JDIMENSION) != 4)
988     return 0;
989   if (sizeof(ISLOW_MULT_TYPE) != 2)
990     return 0;
991 
992   if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_red_sse2))
993     return 1;
994   if (simd_support & JSIMD_MMX)
995     return 1;
996 
997   return 0;
998 }
999 
1000 GLOBAL(int)
jsimd_can_idct_4x4(void)1001 jsimd_can_idct_4x4(void)
1002 {
1003   init_simd();
1004 
1005   /* The code is optimised for these values only */
1006   if (DCTSIZE != 8)
1007     return 0;
1008   if (sizeof(JCOEF) != 2)
1009     return 0;
1010   if (BITS_IN_JSAMPLE != 8)
1011     return 0;
1012   if (sizeof(JDIMENSION) != 4)
1013     return 0;
1014   if (sizeof(ISLOW_MULT_TYPE) != 2)
1015     return 0;
1016 
1017   if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_red_sse2))
1018     return 1;
1019   if (simd_support & JSIMD_MMX)
1020     return 1;
1021 
1022   return 0;
1023 }
1024 
1025 GLOBAL(void)
jsimd_idct_2x2(j_decompress_ptr cinfo,jpeg_component_info * compptr,JCOEFPTR coef_block,JSAMPARRAY output_buf,JDIMENSION output_col)1026 jsimd_idct_2x2(j_decompress_ptr cinfo, jpeg_component_info *compptr,
1027                JCOEFPTR coef_block, JSAMPARRAY output_buf,
1028                JDIMENSION output_col)
1029 {
1030   if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_red_sse2))
1031     jsimd_idct_2x2_sse2(compptr->dct_table, coef_block, output_buf,
1032                         output_col);
1033   else
1034     jsimd_idct_2x2_mmx(compptr->dct_table, coef_block, output_buf, output_col);
1035 }
1036 
1037 GLOBAL(void)
jsimd_idct_4x4(j_decompress_ptr cinfo,jpeg_component_info * compptr,JCOEFPTR coef_block,JSAMPARRAY output_buf,JDIMENSION output_col)1038 jsimd_idct_4x4(j_decompress_ptr cinfo, jpeg_component_info *compptr,
1039                JCOEFPTR coef_block, JSAMPARRAY output_buf,
1040                JDIMENSION output_col)
1041 {
1042   if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_red_sse2))
1043     jsimd_idct_4x4_sse2(compptr->dct_table, coef_block, output_buf,
1044                         output_col);
1045   else
1046     jsimd_idct_4x4_mmx(compptr->dct_table, coef_block, output_buf, output_col);
1047 }
1048 
1049 GLOBAL(int)
jsimd_can_idct_islow(void)1050 jsimd_can_idct_islow(void)
1051 {
1052   init_simd();
1053 
1054   /* The code is optimised for these values only */
1055   if (DCTSIZE != 8)
1056     return 0;
1057   if (sizeof(JCOEF) != 2)
1058     return 0;
1059   if (BITS_IN_JSAMPLE != 8)
1060     return 0;
1061   if (sizeof(JDIMENSION) != 4)
1062     return 0;
1063   if (sizeof(ISLOW_MULT_TYPE) != 2)
1064     return 0;
1065 
1066   if ((simd_support & JSIMD_AVX2) && IS_ALIGNED_AVX(jconst_idct_islow_avx2))
1067     return 1;
1068   if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_islow_sse2))
1069     return 1;
1070   if (simd_support & JSIMD_MMX)
1071     return 1;
1072 
1073   return 0;
1074 }
1075 
1076 GLOBAL(int)
jsimd_can_idct_ifast(void)1077 jsimd_can_idct_ifast(void)
1078 {
1079   init_simd();
1080 
1081   /* The code is optimised for these values only */
1082   if (DCTSIZE != 8)
1083     return 0;
1084   if (sizeof(JCOEF) != 2)
1085     return 0;
1086   if (BITS_IN_JSAMPLE != 8)
1087     return 0;
1088   if (sizeof(JDIMENSION) != 4)
1089     return 0;
1090   if (sizeof(IFAST_MULT_TYPE) != 2)
1091     return 0;
1092   if (IFAST_SCALE_BITS != 2)
1093     return 0;
1094 
1095   if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_ifast_sse2))
1096     return 1;
1097   if (simd_support & JSIMD_MMX)
1098     return 1;
1099 
1100   return 0;
1101 }
1102 
1103 GLOBAL(int)
jsimd_can_idct_float(void)1104 jsimd_can_idct_float(void)
1105 {
1106   init_simd();
1107 
1108   if (DCTSIZE != 8)
1109     return 0;
1110   if (sizeof(JCOEF) != 2)
1111     return 0;
1112   if (BITS_IN_JSAMPLE != 8)
1113     return 0;
1114   if (sizeof(JDIMENSION) != 4)
1115     return 0;
1116   if (sizeof(FAST_FLOAT) != 4)
1117     return 0;
1118   if (sizeof(FLOAT_MULT_TYPE) != 4)
1119     return 0;
1120 
1121   if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_float_sse2))
1122     return 1;
1123   if ((simd_support & JSIMD_SSE) && IS_ALIGNED_SSE(jconst_idct_float_sse))
1124     return 1;
1125   if (simd_support & JSIMD_3DNOW)
1126     return 1;
1127 
1128   return 0;
1129 }
1130 
1131 GLOBAL(void)
jsimd_idct_islow(j_decompress_ptr cinfo,jpeg_component_info * compptr,JCOEFPTR coef_block,JSAMPARRAY output_buf,JDIMENSION output_col)1132 jsimd_idct_islow(j_decompress_ptr cinfo, jpeg_component_info *compptr,
1133                  JCOEFPTR coef_block, JSAMPARRAY output_buf,
1134                  JDIMENSION output_col)
1135 {
1136   if (simd_support & JSIMD_AVX2)
1137     jsimd_idct_islow_avx2(compptr->dct_table, coef_block, output_buf,
1138                           output_col);
1139   else if (simd_support & JSIMD_SSE2)
1140     jsimd_idct_islow_sse2(compptr->dct_table, coef_block, output_buf,
1141                           output_col);
1142   else
1143     jsimd_idct_islow_mmx(compptr->dct_table, coef_block, output_buf,
1144                          output_col);
1145 }
1146 
1147 GLOBAL(void)
jsimd_idct_ifast(j_decompress_ptr cinfo,jpeg_component_info * compptr,JCOEFPTR coef_block,JSAMPARRAY output_buf,JDIMENSION output_col)1148 jsimd_idct_ifast(j_decompress_ptr cinfo, jpeg_component_info *compptr,
1149                  JCOEFPTR coef_block, JSAMPARRAY output_buf,
1150                  JDIMENSION output_col)
1151 {
1152   if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_ifast_sse2))
1153     jsimd_idct_ifast_sse2(compptr->dct_table, coef_block, output_buf,
1154                           output_col);
1155   else
1156     jsimd_idct_ifast_mmx(compptr->dct_table, coef_block, output_buf,
1157                          output_col);
1158 }
1159 
1160 GLOBAL(void)
jsimd_idct_float(j_decompress_ptr cinfo,jpeg_component_info * compptr,JCOEFPTR coef_block,JSAMPARRAY output_buf,JDIMENSION output_col)1161 jsimd_idct_float(j_decompress_ptr cinfo, jpeg_component_info *compptr,
1162                  JCOEFPTR coef_block, JSAMPARRAY output_buf,
1163                  JDIMENSION output_col)
1164 {
1165   if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_float_sse2))
1166     jsimd_idct_float_sse2(compptr->dct_table, coef_block, output_buf,
1167                           output_col);
1168   else if ((simd_support & JSIMD_SSE) && IS_ALIGNED_SSE(jconst_idct_float_sse))
1169     jsimd_idct_float_sse(compptr->dct_table, coef_block, output_buf,
1170                          output_col);
1171   else
1172     jsimd_idct_float_3dnow(compptr->dct_table, coef_block, output_buf,
1173                            output_col);
1174 }
1175 
1176 GLOBAL(int)
jsimd_can_huff_encode_one_block(void)1177 jsimd_can_huff_encode_one_block(void)
1178 {
1179   init_simd();
1180 
1181   if (DCTSIZE != 8)
1182     return 0;
1183   if (sizeof(JCOEF) != 2)
1184     return 0;
1185 
1186   if ((simd_support & JSIMD_SSE2) && simd_huffman &&
1187       IS_ALIGNED_SSE(jconst_huff_encode_one_block))
1188     return 1;
1189 
1190   return 0;
1191 }
1192 
1193 GLOBAL(JOCTET *)
jsimd_huff_encode_one_block(void * state,JOCTET * buffer,JCOEFPTR block,int last_dc_val,c_derived_tbl * dctbl,c_derived_tbl * actbl)1194 jsimd_huff_encode_one_block(void *state, JOCTET *buffer, JCOEFPTR block,
1195                             int last_dc_val, c_derived_tbl *dctbl,
1196                             c_derived_tbl *actbl)
1197 {
1198   return jsimd_huff_encode_one_block_sse2(state, buffer, block, last_dc_val,
1199                                           dctbl, actbl);
1200 }
1201 
1202 GLOBAL(int)
jsimd_can_encode_mcu_AC_first_prepare(void)1203 jsimd_can_encode_mcu_AC_first_prepare(void)
1204 {
1205   init_simd();
1206 
1207   if (DCTSIZE != 8)
1208     return 0;
1209   if (sizeof(JCOEF) != 2)
1210     return 0;
1211   if (SIZEOF_SIZE_T != 4)
1212     return 0;
1213   if (simd_support & JSIMD_SSE2)
1214     return 1;
1215 
1216   return 0;
1217 }
1218 
1219 GLOBAL(void)
jsimd_encode_mcu_AC_first_prepare(const JCOEF * block,const int * jpeg_natural_order_start,int Sl,int Al,JCOEF * values,size_t * zerobits)1220 jsimd_encode_mcu_AC_first_prepare(const JCOEF *block,
1221                                   const int *jpeg_natural_order_start, int Sl,
1222                                   int Al, JCOEF *values, size_t *zerobits)
1223 {
1224   jsimd_encode_mcu_AC_first_prepare_sse2(block, jpeg_natural_order_start,
1225                                          Sl, Al, values, zerobits);
1226 }
1227 
1228 GLOBAL(int)
jsimd_can_encode_mcu_AC_refine_prepare(void)1229 jsimd_can_encode_mcu_AC_refine_prepare(void)
1230 {
1231   init_simd();
1232 
1233   if (DCTSIZE != 8)
1234     return 0;
1235   if (sizeof(JCOEF) != 2)
1236     return 0;
1237   if (SIZEOF_SIZE_T != 4)
1238     return 0;
1239   if (simd_support & JSIMD_SSE2)
1240     return 1;
1241 
1242   return 0;
1243 }
1244 
1245 GLOBAL(int)
jsimd_encode_mcu_AC_refine_prepare(const JCOEF * block,const int * jpeg_natural_order_start,int Sl,int Al,JCOEF * absvalues,size_t * bits)1246 jsimd_encode_mcu_AC_refine_prepare(const JCOEF *block,
1247                                    const int *jpeg_natural_order_start, int Sl,
1248                                    int Al, JCOEF *absvalues, size_t *bits)
1249 {
1250   return jsimd_encode_mcu_AC_refine_prepare_sse2(block,
1251                                                  jpeg_natural_order_start,
1252                                                  Sl, Al, absvalues, bits);
1253 }
1254