2 * Copyright (c) 2010 The WebM project authors. All Rights Reserved.
4 * Use of this source code is governed by a BSD-style license
5 * that can be found in the LICENSE file in the root of the source
6 * tree. An additional intellectual property rights grant can be found
7 * in the file PATENTS. All contributing project authors may
8 * be found in the AUTHORS file in the root of the source tree.
14 #include "vpx_ports/mem.h"
16 DECLARE_ALIGNED(16, const short, vp8_bilinear_filters
[8][2]) =
28 DECLARE_ALIGNED(16, const short, vp8_sub_pel_filters
[8][6]) =
31 { 0, 0, 128, 0, 0, 0 }, /* note that 1/8 pel positions are just as per alpha -0.5 bicubic */
32 { 0, -6, 123, 12, -1, 0 },
33 { 2, -11, 108, 36, -8, 1 }, /* New 1/4 pel 6 tap filter */
34 { 0, -9, 93, 50, -6, 0 },
35 { 3, -16, 77, 77, -16, 3 }, /* New 1/2 pel 6 tap filter */
36 { 0, -6, 50, 93, -9, 0 },
37 { 1, -8, 36, 108, -11, 2 }, /* New 1/4 pel 6 tap filter */
38 { 0, -1, 12, 123, -6, 0 },
41 void vp8_filter_block2d_first_pass
43 unsigned char *src_ptr
,
45 unsigned int src_pixels_per_line
,
46 unsigned int pixel_step
,
47 unsigned int output_height
,
48 unsigned int output_width
,
49 const short *vp8_filter
55 for (i
= 0; i
< output_height
; i
++)
57 for (j
= 0; j
< output_width
; j
++)
59 Temp
= ((int)src_ptr
[-2 * (int)pixel_step
] * vp8_filter
[0]) +
60 ((int)src_ptr
[-1 * (int)pixel_step
] * vp8_filter
[1]) +
61 ((int)src_ptr
[0] * vp8_filter
[2]) +
62 ((int)src_ptr
[pixel_step
] * vp8_filter
[3]) +
63 ((int)src_ptr
[2*pixel_step
] * vp8_filter
[4]) +
64 ((int)src_ptr
[3*pixel_step
] * vp8_filter
[5]) +
65 (VP8_FILTER_WEIGHT
>> 1); /* Rounding */
67 /* Normalize back to 0-255 */
68 Temp
= Temp
>> VP8_FILTER_SHIFT
;
80 src_ptr
+= src_pixels_per_line
- output_width
;
81 output_ptr
+= output_width
;
85 void vp8_filter_block2d_second_pass
88 unsigned char *output_ptr
,
90 unsigned int src_pixels_per_line
,
91 unsigned int pixel_step
,
92 unsigned int output_height
,
93 unsigned int output_width
,
94 const short *vp8_filter
100 for (i
= 0; i
< output_height
; i
++)
102 for (j
= 0; j
< output_width
; j
++)
105 Temp
= ((int)src_ptr
[-2 * (int)pixel_step
] * vp8_filter
[0]) +
106 ((int)src_ptr
[-1 * (int)pixel_step
] * vp8_filter
[1]) +
107 ((int)src_ptr
[0] * vp8_filter
[2]) +
108 ((int)src_ptr
[pixel_step
] * vp8_filter
[3]) +
109 ((int)src_ptr
[2*pixel_step
] * vp8_filter
[4]) +
110 ((int)src_ptr
[3*pixel_step
] * vp8_filter
[5]) +
111 (VP8_FILTER_WEIGHT
>> 1); /* Rounding */
113 /* Normalize back to 0-255 */
114 Temp
= Temp
>> VP8_FILTER_SHIFT
;
121 output_ptr
[j
] = (unsigned char)Temp
;
126 src_ptr
+= src_pixels_per_line
- output_width
;
127 output_ptr
+= output_pitch
;
132 void vp8_filter_block2d
134 unsigned char *src_ptr
,
135 unsigned char *output_ptr
,
136 unsigned int src_pixels_per_line
,
138 const short *HFilter
,
142 int FData
[9*4]; /* Temp data buffer used in filtering */
144 /* First filter 1-D horizontally... */
145 vp8_filter_block2d_first_pass(src_ptr
- (2 * src_pixels_per_line
), FData
, src_pixels_per_line
, 1, 9, 4, HFilter
);
147 /* then filter verticaly... */
148 vp8_filter_block2d_second_pass(FData
+ 8, output_ptr
, output_pitch
, 4, 4, 4, 4, VFilter
);
152 void vp8_block_variation_c
154 unsigned char *src_ptr
,
155 int src_pixels_per_line
,
161 unsigned char *Ptr
= src_ptr
;
163 for (i
= 0; i
< 4; i
++)
165 for (j
= 0; j
< 4; j
++)
167 *HVar
+= abs((int)Ptr
[j
] - (int)Ptr
[j
+1]);
168 *VVar
+= abs((int)Ptr
[j
] - (int)Ptr
[j
+src_pixels_per_line
]);
171 Ptr
+= src_pixels_per_line
;
178 void vp8_sixtap_predict_c
180 unsigned char *src_ptr
,
181 int src_pixels_per_line
,
184 unsigned char *dst_ptr
,
188 const short *HFilter
;
189 const short *VFilter
;
191 HFilter
= vp8_sub_pel_filters
[xoffset
]; /* 6 tap */
192 VFilter
= vp8_sub_pel_filters
[yoffset
]; /* 6 tap */
194 vp8_filter_block2d(src_ptr
, dst_ptr
, src_pixels_per_line
, dst_pitch
, HFilter
, VFilter
);
196 void vp8_sixtap_predict8x8_c
198 unsigned char *src_ptr
,
199 int src_pixels_per_line
,
202 unsigned char *dst_ptr
,
206 const short *HFilter
;
207 const short *VFilter
;
208 int FData
[13*16]; /* Temp data buffer used in filtering */
210 HFilter
= vp8_sub_pel_filters
[xoffset
]; /* 6 tap */
211 VFilter
= vp8_sub_pel_filters
[yoffset
]; /* 6 tap */
213 /* First filter 1-D horizontally... */
214 vp8_filter_block2d_first_pass(src_ptr
- (2 * src_pixels_per_line
), FData
, src_pixels_per_line
, 1, 13, 8, HFilter
);
217 /* then filter verticaly... */
218 vp8_filter_block2d_second_pass(FData
+ 16, dst_ptr
, dst_pitch
, 8, 8, 8, 8, VFilter
);
222 void vp8_sixtap_predict8x4_c
224 unsigned char *src_ptr
,
225 int src_pixels_per_line
,
228 unsigned char *dst_ptr
,
232 const short *HFilter
;
233 const short *VFilter
;
234 int FData
[13*16]; /* Temp data buffer used in filtering */
236 HFilter
= vp8_sub_pel_filters
[xoffset
]; /* 6 tap */
237 VFilter
= vp8_sub_pel_filters
[yoffset
]; /* 6 tap */
239 /* First filter 1-D horizontally... */
240 vp8_filter_block2d_first_pass(src_ptr
- (2 * src_pixels_per_line
), FData
, src_pixels_per_line
, 1, 9, 8, HFilter
);
243 /* then filter verticaly... */
244 vp8_filter_block2d_second_pass(FData
+ 16, dst_ptr
, dst_pitch
, 8, 8, 4, 8, VFilter
);
248 void vp8_sixtap_predict16x16_c
250 unsigned char *src_ptr
,
251 int src_pixels_per_line
,
254 unsigned char *dst_ptr
,
258 const short *HFilter
;
259 const short *VFilter
;
260 int FData
[21*24]; /* Temp data buffer used in filtering */
263 HFilter
= vp8_sub_pel_filters
[xoffset
]; /* 6 tap */
264 VFilter
= vp8_sub_pel_filters
[yoffset
]; /* 6 tap */
266 /* First filter 1-D horizontally... */
267 vp8_filter_block2d_first_pass(src_ptr
- (2 * src_pixels_per_line
), FData
, src_pixels_per_line
, 1, 21, 16, HFilter
);
269 /* then filter verticaly... */
270 vp8_filter_block2d_second_pass(FData
+ 32, dst_ptr
, dst_pitch
, 16, 16, 16, 16, VFilter
);
275 /****************************************************************************
277 * ROUTINE : filter_block2d_bil_first_pass
279 * INPUTS : UINT8 *src_ptr : Pointer to source block.
280 * UINT32 src_stride : Stride of source block.
281 * UINT32 height : Block height.
282 * UINT32 width : Block width.
283 * INT32 *vp8_filter : Array of 2 bi-linear filter taps.
285 * OUTPUTS : INT32 *dst_ptr : Pointer to filtered block.
289 * FUNCTION : Applies a 1-D 2-tap bi-linear filter to the source block
290 * in the horizontal direction to produce the filtered output
291 * block. Used to implement first-pass of 2-D separable filter.
293 * SPECIAL NOTES : Produces INT32 output to retain precision for next pass.
294 * Two filter taps should sum to VP8_FILTER_WEIGHT.
296 ****************************************************************************/
297 void vp8_filter_block2d_bil_first_pass
299 unsigned char *src_ptr
,
300 unsigned short *dst_ptr
,
301 unsigned int src_stride
,
304 const short *vp8_filter
309 for (i
= 0; i
< height
; i
++)
311 for (j
= 0; j
< width
; j
++)
313 /* Apply bilinear filter */
314 dst_ptr
[j
] = (((int)src_ptr
[0] * vp8_filter
[0]) +
315 ((int)src_ptr
[1] * vp8_filter
[1]) +
316 (VP8_FILTER_WEIGHT
/ 2)) >> VP8_FILTER_SHIFT
;
321 src_ptr
+= src_stride
- width
;
326 /****************************************************************************
328 * ROUTINE : filter_block2d_bil_second_pass
330 * INPUTS : INT32 *src_ptr : Pointer to source block.
331 * UINT32 dst_pitch : Destination block pitch.
332 * UINT32 height : Block height.
333 * UINT32 width : Block width.
334 * INT32 *vp8_filter : Array of 2 bi-linear filter taps.
336 * OUTPUTS : UINT16 *dst_ptr : Pointer to filtered block.
340 * FUNCTION : Applies a 1-D 2-tap bi-linear filter to the source block
341 * in the vertical direction to produce the filtered output
342 * block. Used to implement second-pass of 2-D separable filter.
344 * SPECIAL NOTES : Requires 32-bit input as produced by filter_block2d_bil_first_pass.
345 * Two filter taps should sum to VP8_FILTER_WEIGHT.
347 ****************************************************************************/
348 void vp8_filter_block2d_bil_second_pass
350 unsigned short *src_ptr
,
351 unsigned char *dst_ptr
,
355 const short *vp8_filter
361 for (i
= 0; i
< height
; i
++)
363 for (j
= 0; j
< width
; j
++)
366 Temp
= ((int)src_ptr
[0] * vp8_filter
[0]) +
367 ((int)src_ptr
[width
] * vp8_filter
[1]) +
368 (VP8_FILTER_WEIGHT
/ 2);
369 dst_ptr
[j
] = (unsigned int)(Temp
>> VP8_FILTER_SHIFT
);
374 dst_ptr
+= dst_pitch
;
379 /****************************************************************************
381 * ROUTINE : filter_block2d_bil
383 * INPUTS : UINT8 *src_ptr : Pointer to source block.
384 * UINT32 src_pitch : Stride of source block.
385 * UINT32 dst_pitch : Stride of destination block.
386 * INT32 *HFilter : Array of 2 horizontal filter taps.
387 * INT32 *VFilter : Array of 2 vertical filter taps.
388 * INT32 Width : Block width
389 * INT32 Height : Block height
391 * OUTPUTS : UINT16 *dst_ptr : Pointer to filtered block.
395 * FUNCTION : 2-D filters an input block by applying a 2-tap
396 * bi-linear filter horizontally followed by a 2-tap
397 * bi-linear filter vertically on the result.
399 * SPECIAL NOTES : The largest block size can be handled here is 16x16
401 ****************************************************************************/
402 void vp8_filter_block2d_bil
404 unsigned char *src_ptr
,
405 unsigned char *dst_ptr
,
406 unsigned int src_pitch
,
407 unsigned int dst_pitch
,
408 const short *HFilter
,
409 const short *VFilter
,
415 unsigned short FData
[17*16]; /* Temp data buffer used in filtering */
417 /* First filter 1-D horizontally... */
418 vp8_filter_block2d_bil_first_pass(src_ptr
, FData
, src_pitch
, Height
+ 1, Width
, HFilter
);
420 /* then 1-D vertically... */
421 vp8_filter_block2d_bil_second_pass(FData
, dst_ptr
, dst_pitch
, Height
, Width
, VFilter
);
425 void vp8_bilinear_predict4x4_c
427 unsigned char *src_ptr
,
428 int src_pixels_per_line
,
431 unsigned char *dst_ptr
,
435 const short *HFilter
;
436 const short *VFilter
;
438 HFilter
= vp8_bilinear_filters
[xoffset
];
439 VFilter
= vp8_bilinear_filters
[yoffset
];
443 unsigned char temp1
[16];
444 unsigned char temp2
[16];
446 bilinear_predict4x4_mmx(src_ptr
, src_pixels_per_line
, xoffset
, yoffset
, temp1
, 4);
447 vp8_filter_block2d_bil(src_ptr
, temp2
, src_pixels_per_line
, 4, HFilter
, VFilter
, 4, 4);
449 for (i
= 0; i
< 16; i
++)
451 if (temp1
[i
] != temp2
[i
])
453 bilinear_predict4x4_mmx(src_ptr
, src_pixels_per_line
, xoffset
, yoffset
, temp1
, 4);
454 vp8_filter_block2d_bil(src_ptr
, temp2
, src_pixels_per_line
, 4, HFilter
, VFilter
, 4, 4);
459 vp8_filter_block2d_bil(src_ptr
, dst_ptr
, src_pixels_per_line
, dst_pitch
, HFilter
, VFilter
, 4, 4);
463 void vp8_bilinear_predict8x8_c
465 unsigned char *src_ptr
,
466 int src_pixels_per_line
,
469 unsigned char *dst_ptr
,
473 const short *HFilter
;
474 const short *VFilter
;
476 HFilter
= vp8_bilinear_filters
[xoffset
];
477 VFilter
= vp8_bilinear_filters
[yoffset
];
479 vp8_filter_block2d_bil(src_ptr
, dst_ptr
, src_pixels_per_line
, dst_pitch
, HFilter
, VFilter
, 8, 8);
483 void vp8_bilinear_predict8x4_c
485 unsigned char *src_ptr
,
486 int src_pixels_per_line
,
489 unsigned char *dst_ptr
,
493 const short *HFilter
;
494 const short *VFilter
;
496 HFilter
= vp8_bilinear_filters
[xoffset
];
497 VFilter
= vp8_bilinear_filters
[yoffset
];
499 vp8_filter_block2d_bil(src_ptr
, dst_ptr
, src_pixels_per_line
, dst_pitch
, HFilter
, VFilter
, 8, 4);
503 void vp8_bilinear_predict16x16_c
505 unsigned char *src_ptr
,
506 int src_pixels_per_line
,
509 unsigned char *dst_ptr
,
513 const short *HFilter
;
514 const short *VFilter
;
516 HFilter
= vp8_bilinear_filters
[xoffset
];
517 VFilter
= vp8_bilinear_filters
[yoffset
];
519 vp8_filter_block2d_bil(src_ptr
, dst_ptr
, src_pixels_per_line
, dst_pitch
, HFilter
, VFilter
, 16, 16);