rescaler.c revision 33f74dabbc7920a65ed435d7417987589febdc16
1// Copyright 2012 Google Inc. All Rights Reserved.
2//
3// Use of this source code is governed by a BSD-style license
4// that can be found in the COPYING file in the root of the source
5// tree. An additional intellectual property rights grant can be found
6// in the file PATENTS. All contributing project authors may
7// be found in the AUTHORS file in the root of the source tree.
8// -----------------------------------------------------------------------------
9//
10// Rescaling functions
11//
12// Author: Skal (pascal.massimino@gmail.com)
13
14#include <assert.h>
15#include <stdlib.h>
16#include "./rescaler.h"
17#include "../dsp/dsp.h"
18
19//------------------------------------------------------------------------------
20// Implementations of critical functions ImportRow / ExportRow
21
22void (*WebPRescalerImportRow)(WebPRescaler* const wrk,
23                              const uint8_t* const src, int channel) = NULL;
24void (*WebPRescalerExportRow)(WebPRescaler* const wrk, int x_out) = NULL;
25
26#define RFIX 30
27#define MULT_FIX(x, y) (((int64_t)(x) * (y) + (1 << (RFIX - 1))) >> RFIX)
28
29static void ImportRowC(WebPRescaler* const wrk,
30                       const uint8_t* const src, int channel) {
31  const int x_stride = wrk->num_channels;
32  const int x_out_max = wrk->dst_width * wrk->num_channels;
33  int x_in = channel;
34  int x_out;
35  int accum = 0;
36  if (!wrk->x_expand) {
37    int sum = 0;
38    for (x_out = channel; x_out < x_out_max; x_out += x_stride) {
39      accum += wrk->x_add;
40      for (; accum > 0; accum -= wrk->x_sub) {
41        sum += src[x_in];
42        x_in += x_stride;
43      }
44      {        // Emit next horizontal pixel.
45        const int32_t base = src[x_in];
46        const int32_t frac = base * (-accum);
47        x_in += x_stride;
48        wrk->frow[x_out] = (sum + base) * wrk->x_sub - frac;
49        // fresh fractional start for next pixel
50        sum = (int)MULT_FIX(frac, wrk->fx_scale);
51      }
52    }
53  } else {        // simple bilinear interpolation
54    int left = src[channel], right = src[channel];
55    for (x_out = channel; x_out < x_out_max; x_out += x_stride) {
56      if (accum < 0) {
57        left = right;
58        x_in += x_stride;
59        right = src[x_in];
60        accum += wrk->x_add;
61      }
62      wrk->frow[x_out] = right * wrk->x_add + (left - right) * accum;
63      accum -= wrk->x_sub;
64    }
65  }
66  // Accumulate the contribution of the new row.
67  for (x_out = channel; x_out < x_out_max; x_out += x_stride) {
68    wrk->irow[x_out] += wrk->frow[x_out];
69  }
70}
71
72static void ExportRowC(WebPRescaler* const wrk, int x_out) {
73  if (wrk->y_accum <= 0) {
74    uint8_t* const dst = wrk->dst;
75    int32_t* const irow = wrk->irow;
76    const int32_t* const frow = wrk->frow;
77    const int yscale = wrk->fy_scale * (-wrk->y_accum);
78    const int x_out_max = wrk->dst_width * wrk->num_channels;
79    for (; x_out < x_out_max; ++x_out) {
80      const int frac = (int)MULT_FIX(frow[x_out], yscale);
81      const int v = (int)MULT_FIX(irow[x_out] - frac, wrk->fxy_scale);
82      dst[x_out] = (!(v & ~0xff)) ? v : (v < 0) ? 0 : 255;
83      irow[x_out] = frac;   // new fractional start
84    }
85    wrk->y_accum += wrk->y_add;
86    wrk->dst += wrk->dst_stride;
87  }
88}
89
90//------------------------------------------------------------------------------
91// MIPS version
92
93#if defined(WEBP_USE_MIPS32)
94
95static void ImportRowMIPS(WebPRescaler* const wrk,
96                          const uint8_t* const src, int channel) {
97  const int x_stride = wrk->num_channels;
98  const int x_out_max = wrk->dst_width * wrk->num_channels;
99  const int fx_scale = wrk->fx_scale;
100  const int x_add = wrk->x_add;
101  const int x_sub = wrk->x_sub;
102  int* frow = wrk->frow + channel;
103  int* irow = wrk->irow + channel;
104  const uint8_t* src1 = src + channel;
105  int temp1, temp2, temp3;
106  int base, frac, sum;
107  int accum, accum1;
108  const int x_stride1 = x_stride << 2;
109  int loop_c = x_out_max - channel;
110
111  if (!wrk->x_expand) {
112    __asm__ volatile (
113      "li     %[temp1],   0x8000                    \n\t"
114      "li     %[temp2],   0x10000                   \n\t"
115      "li     %[sum],     0                         \n\t"
116      "li     %[accum],   0                         \n\t"
117    "1:                                             \n\t"
118      "addu   %[accum],   %[accum],   %[x_add]      \n\t"
119      "blez   %[accum],   3f                        \n\t"
120    "2:                                             \n\t"
121      "lbu    %[temp3],   0(%[src1])                \n\t"
122      "subu   %[accum],   %[accum],   %[x_sub]      \n\t"
123      "addu   %[src1],    %[src1],    %[x_stride]   \n\t"
124      "addu   %[sum],     %[sum],     %[temp3]      \n\t"
125      "bgtz   %[accum],   2b                        \n\t"
126    "3:                                             \n\t"
127      "lbu    %[base],    0(%[src1])                \n\t"
128      "addu   %[src1],    %[src1],    %[x_stride]   \n\t"
129      "negu   %[accum1],  %[accum]                  \n\t"
130      "mul    %[frac],    %[base],    %[accum1]     \n\t"
131      "addu   %[temp3],   %[sum],     %[base]       \n\t"
132      "mul    %[temp3],   %[temp3],   %[x_sub]      \n\t"
133      "lw     %[base],    0(%[irow])                \n\t"
134      "subu   %[loop_c],  %[loop_c],  %[x_stride]   \n\t"
135      "sll    %[accum1],  %[frac],    2             \n\t"
136      "mult   %[temp1],   %[temp2]                  \n\t"
137      "madd   %[accum1],  %[fx_scale]               \n\t"
138      "mfhi   %[sum]                                \n\t"
139      "subu   %[temp3],   %[temp3],   %[frac]       \n\t"
140      "sw     %[temp3],   0(%[frow])                \n\t"
141      "add    %[base],    %[base],    %[temp3]      \n\t"
142      "sw     %[base],    0(%[irow])                \n\t"
143      "addu   %[irow],    %[irow],    %[x_stride1]  \n\t"
144      "addu   %[frow],    %[frow],    %[x_stride1]  \n\t"
145      "bgtz   %[loop_c],  1b                        \n\t"
146
147      : [accum] "=&r" (accum), [src1] "+r" (src1), [temp3] "=&r" (temp3),
148        [sum] "=&r" (sum), [base] "=&r" (base), [frac] "=&r" (frac),
149        [frow] "+r" (frow), [irow] "+r" (irow), [accum1] "=&r" (accum1),
150        [temp2] "=&r" (temp2), [temp1] "=&r" (temp1)
151      : [x_stride] "r" (x_stride), [fx_scale] "r" (fx_scale),
152        [x_sub] "r" (x_sub), [x_add] "r" (x_add),
153        [loop_c] "r" (loop_c), [x_stride1] "r" (x_stride1)
154      : "memory", "hi", "lo"
155    );
156  } else {
157    __asm__ volatile (
158      "lbu    %[temp1],   0(%[src1])                \n\t"
159      "move   %[temp2],   %[temp1]                  \n\t"
160      "li     %[accum],   0                         \n\t"
161    "1:                                             \n\t"
162      "bgez   %[accum],   2f                        \n\t"
163      "move   %[temp2],   %[temp1]                  \n\t"
164      "addu   %[src1],    %[x_stride]               \n\t"
165      "lbu    %[temp1],   0(%[src1])                \n\t"
166      "addu   %[accum],   %[x_add]                  \n\t"
167    "2:                                             \n\t"
168      "subu   %[temp3],   %[temp2],   %[temp1]      \n\t"
169      "mul    %[temp3],   %[temp3],   %[accum]      \n\t"
170      "mul    %[base],    %[temp1],   %[x_add]      \n\t"
171      "subu   %[accum],   %[accum],   %[x_sub]      \n\t"
172      "lw     %[frac],    0(%[irow])                \n\t"
173      "subu   %[loop_c],  %[loop_c],  %[x_stride]   \n\t"
174      "addu   %[temp3],   %[base],    %[temp3]      \n\t"
175      "sw     %[temp3],   0(%[frow])                \n\t"
176      "addu   %[frow],    %[x_stride1]              \n\t"
177      "addu   %[frac],    %[temp3]                  \n\t"
178      "sw     %[frac],    0(%[irow])                \n\t"
179      "addu   %[irow],    %[x_stride1]              \n\t"
180      "bgtz   %[loop_c],  1b                        \n\t"
181
182      : [src1] "+r" (src1), [accum] "=&r" (accum), [temp1] "=&r" (temp1),
183        [temp2] "=&r" (temp2), [temp3] "=&r" (temp3), [base] "=&r" (base),
184        [frac] "=&r" (frac), [frow] "+r" (frow), [irow] "+r" (irow)
185      : [x_stride] "r" (x_stride), [x_add] "r" (x_add), [x_sub] "r" (x_sub),
186        [x_stride1] "r" (x_stride1), [loop_c] "r" (loop_c)
187      : "memory", "hi", "lo"
188    );
189  }
190}
191
192static void ExportRowMIPS(WebPRescaler* const wrk, int x_out) {
193  if (wrk->y_accum <= 0) {
194    uint8_t* const dst = wrk->dst;
195    int32_t* const irow = wrk->irow;
196    const int32_t* const frow = wrk->frow;
197    const int yscale = wrk->fy_scale * (-wrk->y_accum);
198    const int x_out_max = wrk->dst_width * wrk->num_channels;
199    // if wrk->fxy_scale can fit into 32 bits use optimized code,
200    // otherwise use C code
201    if ((wrk->fxy_scale >> 32) == 0) {
202      int temp0, temp1, temp3, temp4, temp5, temp6, temp7, loop_end;
203      const int temp2 = (int)(wrk->fxy_scale);
204      const int temp8 = x_out_max << 2;
205      uint8_t* dst_t = (uint8_t*)dst;
206      int32_t* irow_t = (int32_t*)irow;
207      const int32_t* frow_t = (const int32_t*)frow;
208
209      __asm__ volatile(
210        "addiu    %[temp6],    $zero,       -256          \n\t"
211        "addiu    %[temp7],    $zero,       255           \n\t"
212        "li       %[temp3],    0x10000                    \n\t"
213        "li       %[temp4],    0x8000                     \n\t"
214        "addu     %[loop_end], %[frow_t],   %[temp8]      \n\t"
215      "1:                                                 \n\t"
216        "lw       %[temp0],    0(%[frow_t])               \n\t"
217        "mult     %[temp3],    %[temp4]                   \n\t"
218        "addiu    %[frow_t],   %[frow_t],   4             \n\t"
219        "sll      %[temp0],    %[temp0],    2             \n\t"
220        "madd     %[temp0],    %[yscale]                  \n\t"
221        "mfhi     %[temp1]                                \n\t"
222        "lw       %[temp0],    0(%[irow_t])               \n\t"
223        "addiu    %[dst_t],    %[dst_t],    1             \n\t"
224        "addiu    %[irow_t],   %[irow_t],   4             \n\t"
225        "subu     %[temp0],    %[temp0],    %[temp1]      \n\t"
226        "mult     %[temp3],    %[temp4]                   \n\t"
227        "sll      %[temp0],    %[temp0],    2             \n\t"
228        "madd     %[temp0],    %[temp2]                   \n\t"
229        "mfhi     %[temp5]                                \n\t"
230        "sw       %[temp1],    -4(%[irow_t])              \n\t"
231        "and      %[temp0],    %[temp5],    %[temp6]      \n\t"
232        "slti     %[temp1],    %[temp5],    0             \n\t"
233        "beqz     %[temp0],    2f                         \n\t"
234        "xor      %[temp5],    %[temp5],    %[temp5]      \n\t"
235        "movz     %[temp5],    %[temp7],    %[temp1]      \n\t"
236      "2:                                                 \n\t"
237        "sb       %[temp5],    -1(%[dst_t])               \n\t"
238        "bne      %[frow_t],   %[loop_end], 1b            \n\t"
239
240        : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp3]"=&r"(temp3),
241          [temp4]"=&r"(temp4), [temp5]"=&r"(temp5), [temp6]"=&r"(temp6),
242          [temp7]"=&r"(temp7), [frow_t]"+r"(frow_t), [irow_t]"+r"(irow_t),
243          [dst_t]"+r"(dst_t), [loop_end]"=&r"(loop_end)
244        : [temp2]"r"(temp2), [yscale]"r"(yscale), [temp8]"r"(temp8)
245        : "memory", "hi", "lo"
246      );
247      wrk->y_accum += wrk->y_add;
248      wrk->dst += wrk->dst_stride;
249    } else {
250      ExportRowC(wrk, x_out);
251    }
252  }
253}
254#endif   // WEBP_USE_MIPS32
255
256//------------------------------------------------------------------------------
257
258void WebPRescalerInit(WebPRescaler* const wrk, int src_width, int src_height,
259                      uint8_t* const dst, int dst_width, int dst_height,
260                      int dst_stride, int num_channels, int x_add, int x_sub,
261                      int y_add, int y_sub, int32_t* const work) {
262  wrk->x_expand = (src_width < dst_width);
263  wrk->src_width = src_width;
264  wrk->src_height = src_height;
265  wrk->dst_width = dst_width;
266  wrk->dst_height = dst_height;
267  wrk->dst = dst;
268  wrk->dst_stride = dst_stride;
269  wrk->num_channels = num_channels;
270  // for 'x_expand', we use bilinear interpolation
271  wrk->x_add = wrk->x_expand ? (x_sub - 1) : x_add - x_sub;
272  wrk->x_sub = wrk->x_expand ? (x_add - 1) : x_sub;
273  wrk->y_accum = y_add;
274  wrk->y_add = y_add;
275  wrk->y_sub = y_sub;
276  wrk->fx_scale = (1 << RFIX) / x_sub;
277  wrk->fy_scale = (1 << RFIX) / y_sub;
278  wrk->fxy_scale = wrk->x_expand ?
279      ((int64_t)dst_height << RFIX) / (x_sub * src_height) :
280      ((int64_t)dst_height << RFIX) / (x_add * src_height);
281  wrk->irow = work;
282  wrk->frow = work + num_channels * dst_width;
283
284  if (WebPRescalerImportRow == NULL) {
285    WebPRescalerImportRow = ImportRowC;
286    WebPRescalerExportRow = ExportRowC;
287    if (VP8GetCPUInfo != NULL) {
288#if defined(WEBP_USE_MIPS32)
289      if (VP8GetCPUInfo(kMIPS32)) {
290        WebPRescalerImportRow = ImportRowMIPS;
291        WebPRescalerExportRow = ExportRowMIPS;
292      }
293#endif
294    }
295  }
296}
297
298#undef MULT_FIX
299#undef RFIX
300
301//------------------------------------------------------------------------------
302// all-in-one calls
303
304int WebPRescaleNeededLines(const WebPRescaler* const wrk, int max_num_lines) {
305  const int num_lines = (wrk->y_accum + wrk->y_sub - 1) / wrk->y_sub;
306  return (num_lines > max_num_lines) ? max_num_lines : num_lines;
307}
308
309int WebPRescalerImport(WebPRescaler* const wrk, int num_lines,
310                       const uint8_t* src, int src_stride) {
311  int total_imported = 0;
312  while (total_imported < num_lines && wrk->y_accum > 0) {
313    int channel;
314    for (channel = 0; channel < wrk->num_channels; ++channel) {
315      WebPRescalerImportRow(wrk, src, channel);
316    }
317    src += src_stride;
318    ++total_imported;
319    wrk->y_accum -= wrk->y_sub;
320  }
321  return total_imported;
322}
323
324int WebPRescalerExport(WebPRescaler* const rescaler) {
325  int total_exported = 0;
326  while (WebPRescalerHasPendingOutput(rescaler)) {
327    WebPRescalerExportRow(rescaler, 0);
328    ++total_exported;
329  }
330  return total_exported;
331}
332
333//------------------------------------------------------------------------------
334