OpenTTD
32bpp_anim_sse4.cpp
1 /* $Id: 32bpp_anim_sse4.cpp 26969 2014-10-06 18:45:51Z rubidium $ */
2 
3 /*
4  * This file is part of OpenTTD.
5  * OpenTTD is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, version 2.
6  * OpenTTD is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
7  * See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with OpenTTD. If not, see <http://www.gnu.org/licenses/>.
8  */
9 
12 #ifdef WITH_SSE
13 
14 #include "../stdafx.h"
15 #include "../video/video_driver.hpp"
16 #include "../table/sprites.h"
17 #include "32bpp_anim_sse4.hpp"
18 #include "32bpp_sse_func.hpp"
19 
20 #include "../safeguards.h"
21 
23 static FBlitter_32bppSSE4_Anim iFBlitter_32bppSSE4_Anim;
24 
32 IGNORE_UNINITIALIZED_WARNING_START
33 template <BlitterMode mode, Blitter_32bppSSE2::ReadMode read_mode, Blitter_32bppSSE2::BlockType bt_last, bool translucent, bool animated>
34 inline void Blitter_32bppSSE4_Anim::Draw(const Blitter::BlitterParams *bp, ZoomLevel zoom)
35 {
36  const byte * const remap = bp->remap;
37  Colour *dst_line = (Colour *) bp->dst + bp->top * bp->pitch + bp->left;
38  uint16 *anim_line = this->anim_buf + ((uint32 *)bp->dst - (uint32 *)_screen.dst_ptr) + bp->top * this->anim_buf_width + bp->left;
39  int effective_width = bp->width;
40 
41  /* Find where to start reading in the source sprite. */
42  const Blitter_32bppSSE_Base::SpriteData * const sd = (const Blitter_32bppSSE_Base::SpriteData *) bp->sprite;
43  const SpriteInfo * const si = &sd->infos[zoom];
44  const MapValue *src_mv_line = (const MapValue *) &sd->data[si->mv_offset] + bp->skip_top * si->sprite_width;
45  const Colour *src_rgba_line = (const Colour *) ((const byte *) &sd->data[si->sprite_offset] + bp->skip_top * si->sprite_line_size);
46 
47  if (read_mode != RM_WITH_MARGIN) {
48  src_rgba_line += bp->skip_left;
49  src_mv_line += bp->skip_left;
50  }
51  const MapValue *src_mv = src_mv_line;
52 
53  /* Load these variables into register before loop. */
54  const __m128i a_cm = ALPHA_CONTROL_MASK;
55  const __m128i pack_low_cm = PACK_LOW_CONTROL_MASK;
56  const __m128i tr_nom_base = TRANSPARENT_NOM_BASE;
57 
58  for (int y = bp->height; y != 0; y--) {
59  Colour *dst = dst_line;
60  const Colour *src = src_rgba_line + META_LENGTH;
61  if (mode != BM_TRANSPARENT) src_mv = src_mv_line;
62  uint16 *anim = anim_line;
63 
64  if (read_mode == RM_WITH_MARGIN) {
65  assert(bt_last == BT_NONE); // or you must ensure block type is preserved
66  anim += src_rgba_line[0].data;
67  src += src_rgba_line[0].data;
68  dst += src_rgba_line[0].data;
69  if (mode != BM_TRANSPARENT) src_mv += src_rgba_line[0].data;
70  const int width_diff = si->sprite_width - bp->width;
71  effective_width = bp->width - (int) src_rgba_line[0].data;
72  const int delta_diff = (int) src_rgba_line[1].data - width_diff;
73  const int new_width = effective_width - delta_diff;
74  effective_width = delta_diff > 0 ? new_width : effective_width;
75  if (effective_width <= 0) goto next_line;
76  }
77 
78  switch (mode) {
79  default:
80  if (!translucent) {
81  for (uint x = (uint) effective_width; x > 0; x--) {
82  if (src->a) {
83  if (animated) {
84  *anim = *(const uint16*) src_mv;
85  *dst = (src_mv->m >= PALETTE_ANIM_START) ? AdjustBrightneSSE(this->LookupColourInPalette(src_mv->m), src_mv->v) : src->data;
86  } else {
87  *anim = 0;
88  *dst = *src;
89  }
90  }
91  if (animated) src_mv++;
92  anim++;
93  src++;
94  dst++;
95  }
96  break;
97  }
98 
99  for (uint x = (uint) effective_width/2; x != 0; x--) {
100  uint32 mvX2 = *((uint32 *) const_cast<MapValue *>(src_mv));
101  __m128i srcABCD = _mm_loadl_epi64((const __m128i*) src);
102  __m128i dstABCD = _mm_loadl_epi64((__m128i*) dst);
103 
104  if (animated) {
105  /* Remap colours. */
106  const byte m0 = mvX2;
107  if (m0 >= PALETTE_ANIM_START) {
108  const Colour c0 = (this->LookupColourInPalette(m0).data & 0x00FFFFFF) | (src[0].data & 0xFF000000);
109  InsertFirstUint32(AdjustBrightneSSE(c0, (byte) (mvX2 >> 8)).data, srcABCD);
110  }
111  const byte m1 = mvX2 >> 16;
112  if (m1 >= PALETTE_ANIM_START) {
113  const Colour c1 = (this->LookupColourInPalette(m1).data & 0x00FFFFFF) | (src[1].data & 0xFF000000);
114  InsertSecondUint32(AdjustBrightneSSE(c1, (byte) (mvX2 >> 24)).data, srcABCD);
115  }
116 
117  /* Update anim buffer. */
118  const byte a0 = src[0].a;
119  const byte a1 = src[1].a;
120  uint32 anim01 = 0;
121  if (a0 == 255) {
122  if (a1 == 255) {
123  *(uint32*) anim = mvX2;
124  goto bmno_full_opacity;
125  }
126  anim01 = (uint16) mvX2;
127  } else if (a0 == 0) {
128  if (a1 == 0) {
129  goto bmno_full_transparency;
130  } else {
131  if (a1 == 255) anim[1] = (uint16) (mvX2 >> 16);
132  goto bmno_alpha_blend;
133  }
134  }
135  if (a1 > 0) {
136  if (a1 == 255) anim01 |= mvX2 & 0xFFFF0000;
137  *(uint32*) anim = anim01;
138  } else {
139  anim[0] = (uint16) anim01;
140  }
141  } else {
142  if (src[0].a) anim[0] = 0;
143  if (src[1].a) anim[1] = 0;
144  }
145 
146  /* Blend colours. */
147 bmno_alpha_blend:
148  srcABCD = AlphaBlendTwoPixels(srcABCD, dstABCD, a_cm, pack_low_cm);
149 bmno_full_opacity:
150  _mm_storel_epi64((__m128i *) dst, srcABCD);
151 bmno_full_transparency:
152  src_mv += 2;
153  src += 2;
154  anim += 2;
155  dst += 2;
156  }
157 
158  if ((bt_last == BT_NONE && effective_width & 1) || bt_last == BT_ODD) {
159  if (src->a == 0) {
160  } else if (src->a == 255) {
161  *anim = *(const uint16*) src_mv;
162  *dst = (src_mv->m >= PALETTE_ANIM_START) ? AdjustBrightneSSE(LookupColourInPalette(src_mv->m), src_mv->v) : *src;
163  } else {
164  *anim = 0;
165  __m128i srcABCD;
166  __m128i dstABCD = _mm_cvtsi32_si128(dst->data);
167  if (src_mv->m >= PALETTE_ANIM_START) {
168  Colour colour = AdjustBrightneSSE(LookupColourInPalette(src_mv->m), src_mv->v);
169  colour.a = src->a;
170  srcABCD = _mm_cvtsi32_si128(colour.data);
171  } else {
172  srcABCD = _mm_cvtsi32_si128(src->data);
173  }
174  dst->data = _mm_cvtsi128_si32(AlphaBlendTwoPixels(srcABCD, dstABCD, a_cm, pack_low_cm));
175  }
176  }
177  break;
178 
179  case BM_COLOUR_REMAP:
180  for (uint x = (uint) effective_width / 2; x != 0; x--) {
181  uint32 mvX2 = *((uint32 *) const_cast<MapValue *>(src_mv));
182  __m128i srcABCD = _mm_loadl_epi64((const __m128i*) src);
183  __m128i dstABCD = _mm_loadl_epi64((__m128i*) dst);
184 
185  /* Remap colours. */
186  const uint m0 = (byte) mvX2;
187  const uint r0 = remap[m0];
188  const uint m1 = (byte) (mvX2 >> 16);
189  const uint r1 = remap[m1];
190  if (mvX2 & 0x00FF00FF) {
191  #define CMOV_REMAP(m_colour, m_colour_init, m_src, m_m) \
192  /* Written so the compiler uses CMOV. */ \
193  Colour m_colour = m_colour_init; \
194  { \
195  const Colour srcm = (Colour) (m_src); \
196  const uint m = (byte) (m_m); \
197  const uint r = remap[m]; \
198  const Colour cmap = (this->LookupColourInPalette(r).data & 0x00FFFFFF) | (srcm.data & 0xFF000000); \
199  m_colour = r == 0 ? m_colour : cmap; \
200  m_colour = m != 0 ? m_colour : srcm; \
201  }
202 #ifdef _SQ64
203  uint64 srcs = _mm_cvtsi128_si64(srcABCD);
204  uint64 dsts;
205  if (animated) dsts = _mm_cvtsi128_si64(dstABCD);
206  uint64 remapped_src = 0;
207  CMOV_REMAP(c0, animated ? dsts : 0, srcs, mvX2);
208  remapped_src = c0.data;
209  CMOV_REMAP(c1, animated ? dsts >> 32 : 0, srcs >> 32, mvX2 >> 16);
210  remapped_src |= (uint64) c1.data << 32;
211  srcABCD = _mm_cvtsi64_si128(remapped_src);
212 #else
213  Colour remapped_src[2];
214  CMOV_REMAP(c0, animated ? _mm_cvtsi128_si32(dstABCD) : 0, _mm_cvtsi128_si32(srcABCD), mvX2);
215  remapped_src[0] = c0.data;
216  CMOV_REMAP(c1, animated ? dst[1] : 0, src[1], mvX2 >> 16);
217  remapped_src[1] = c1.data;
218  srcABCD = _mm_loadl_epi64((__m128i*) &remapped_src);
219 #endif
220 
221  if ((mvX2 & 0xFF00FF00) != 0x80008000) srcABCD = AdjustBrightnessOfTwoPixels(srcABCD, mvX2);
222  }
223 
224  /* Update anim buffer. */
225  if (animated) {
226  const byte a0 = src[0].a;
227  const byte a1 = src[1].a;
228  uint32 anim01 = mvX2 & 0xFF00FF00;
229  if (a0 == 255) {
230  anim01 |= r0;
231  if (a1 == 255) {
232  *(uint32*) anim = anim01 | (r1 << 16);
233  goto bmcr_full_opacity;
234  }
235  } else if (a0 == 0) {
236  if (a1 == 0) {
237  goto bmcr_full_transparency;
238  } else {
239  if (a1 == 255) {
240  anim[1] = r1 | (anim01 >> 16);
241  }
242  goto bmcr_alpha_blend;
243  }
244  }
245  if (a1 > 0) {
246  if (a1 == 255) anim01 |= r1 << 16;
247  *(uint32*) anim = anim01;
248  } else {
249  anim[0] = (uint16) anim01;
250  }
251  } else {
252  if (src[0].a) anim[0] = 0;
253  if (src[1].a) anim[1] = 0;
254  }
255 
256  /* Blend colours. */
257 bmcr_alpha_blend:
258  srcABCD = AlphaBlendTwoPixels(srcABCD, dstABCD, a_cm, pack_low_cm);
259 bmcr_full_opacity:
260  _mm_storel_epi64((__m128i *) dst, srcABCD);
261 bmcr_full_transparency:
262  src_mv += 2;
263  dst += 2;
264  src += 2;
265  anim += 2;
266  }
267 
268  if ((bt_last == BT_NONE && effective_width & 1) || bt_last == BT_ODD) {
269  /* In case the m-channel is zero, do not remap this pixel in any way. */
270  __m128i srcABCD;
271  if (src->a == 0) break;
272  if (src_mv->m) {
273  const uint r = remap[src_mv->m];
274  *anim = (animated && src->a == 255) ? r | ((uint16) src_mv->v << 8 ) : 0;
275  if (r != 0) {
276  Colour remapped_colour = AdjustBrightneSSE(this->LookupColourInPalette(r), src_mv->v);
277  if (src->a == 255) {
278  *dst = remapped_colour;
279  } else {
280  remapped_colour.a = src->a;
281  srcABCD = _mm_cvtsi32_si128(remapped_colour.data);
282  goto bmcr_alpha_blend_single;
283  }
284  }
285  } else {
286  *anim = 0;
287  srcABCD = _mm_cvtsi32_si128(src->data);
288  if (src->a < 255) {
289 bmcr_alpha_blend_single:
290  __m128i dstABCD = _mm_cvtsi32_si128(dst->data);
291  srcABCD = AlphaBlendTwoPixels(srcABCD, dstABCD, a_cm, pack_low_cm);
292  }
293  dst->data = _mm_cvtsi128_si32(srcABCD);
294  }
295  }
296  break;
297 
298  case BM_TRANSPARENT:
299  /* Make the current colour a bit more black, so it looks like this image is transparent. */
300  for (uint x = (uint) bp->width / 2; x > 0; x--) {
301  __m128i srcABCD = _mm_loadl_epi64((const __m128i*) src);
302  __m128i dstABCD = _mm_loadl_epi64((__m128i*) dst);
303  _mm_storel_epi64((__m128i *) dst, DarkenTwoPixels(srcABCD, dstABCD, a_cm, tr_nom_base));
304  src += 2;
305  dst += 2;
306  anim += 2;
307  if (src[-2].a) anim[-2] = 0;
308  if (src[-1].a) anim[-1] = 0;
309  }
310 
311  if ((bt_last == BT_NONE && bp->width & 1) || bt_last == BT_ODD) {
312  __m128i srcABCD = _mm_cvtsi32_si128(src->data);
313  __m128i dstABCD = _mm_cvtsi32_si128(dst->data);
314  dst->data = _mm_cvtsi128_si32(DarkenTwoPixels(srcABCD, dstABCD, a_cm, tr_nom_base));
315  if (src[0].a) anim[0] = 0;
316  }
317  break;
318 
319  case BM_CRASH_REMAP:
320  for (uint x = (uint) bp->width; x > 0; x--) {
321  if (src_mv->m == 0) {
322  if (src->a != 0) {
323  uint8 g = MakeDark(src->r, src->g, src->b);
324  *dst = ComposeColourRGBA(g, g, g, src->a, *dst);
325  *anim = 0;
326  }
327  } else {
328  uint r = remap[src_mv->m];
329  if (r != 0) *dst = ComposeColourPANoCheck(this->AdjustBrightness(this->LookupColourInPalette(r), src_mv->v), src->a, *dst);
330  }
331  src_mv++;
332  dst++;
333  src++;
334  anim++;
335  }
336  break;
337 
338  case BM_BLACK_REMAP:
339  for (uint x = (uint) bp->width; x > 0; x--) {
340  if (src->a != 0) {
341  *dst = Colour(0, 0, 0);
342  *anim = 0;
343  }
344  src_mv++;
345  dst++;
346  src++;
347  anim++;
348  }
349  break;
350  }
351 
352 next_line:
353  if (mode != BM_TRANSPARENT) src_mv_line += si->sprite_width;
354  src_rgba_line = (const Colour*) ((const byte*) src_rgba_line + si->sprite_line_size);
355  dst_line += bp->pitch;
356  anim_line += this->anim_buf_width;
357  }
358 }
359 IGNORE_UNINITIALIZED_WARNING_STOP
360 
368 void Blitter_32bppSSE4_Anim::Draw(Blitter::BlitterParams *bp, BlitterMode mode, ZoomLevel zoom)
369 {
370  const Blitter_32bppSSE_Base::SpriteFlags sprite_flags = ((const Blitter_32bppSSE_Base::SpriteData *) bp->sprite)->flags;
371  switch (mode) {
372  default: {
373 bm_normal:
374  if (bp->skip_left != 0 || bp->width <= MARGIN_NORMAL_THRESHOLD) {
375  const BlockType bt_last = (BlockType) (bp->width & 1);
376  if (bt_last == BT_EVEN) {
377  if (sprite_flags & SF_NO_ANIM) Draw<BM_NORMAL, RM_WITH_SKIP, BT_EVEN, true, false>(bp, zoom);
378  else Draw<BM_NORMAL, RM_WITH_SKIP, BT_EVEN, true, true>(bp, zoom);
379  } else {
380  if (sprite_flags & SF_NO_ANIM) Draw<BM_NORMAL, RM_WITH_SKIP, BT_ODD, true, false>(bp, zoom);
381  else Draw<BM_NORMAL, RM_WITH_SKIP, BT_ODD, true, true>(bp, zoom);
382  }
383  } else {
384 #ifdef _SQ64
385  if (sprite_flags & SF_TRANSLUCENT) {
386  if (sprite_flags & SF_NO_ANIM) Draw<BM_NORMAL, RM_WITH_MARGIN, BT_NONE, true, false>(bp, zoom);
387  else Draw<BM_NORMAL, RM_WITH_MARGIN, BT_NONE, true, true>(bp, zoom);
388  } else {
389  if (sprite_flags & SF_NO_ANIM) Draw<BM_NORMAL, RM_WITH_MARGIN, BT_NONE, false, false>(bp, zoom);
390  else Draw<BM_NORMAL, RM_WITH_MARGIN, BT_NONE, false, true>(bp, zoom);
391  }
392 #else
393  if (sprite_flags & SF_NO_ANIM) Draw<BM_NORMAL, RM_WITH_MARGIN, BT_NONE, true, false>(bp, zoom);
394  else Draw<BM_NORMAL, RM_WITH_MARGIN, BT_NONE, true, true>(bp, zoom);
395 #endif
396  }
397  break;
398  }
399  case BM_COLOUR_REMAP:
400  if (sprite_flags & SF_NO_REMAP) goto bm_normal;
401  if (bp->skip_left != 0 || bp->width <= MARGIN_REMAP_THRESHOLD) {
402  if (sprite_flags & SF_NO_ANIM) Draw<BM_COLOUR_REMAP, RM_WITH_SKIP, BT_NONE, true, false>(bp, zoom);
403  else Draw<BM_COLOUR_REMAP, RM_WITH_SKIP, BT_NONE, true, true>(bp, zoom);
404  } else {
405  if (sprite_flags & SF_NO_ANIM) Draw<BM_COLOUR_REMAP, RM_WITH_MARGIN, BT_NONE, true, false>(bp, zoom);
406  else Draw<BM_COLOUR_REMAP, RM_WITH_MARGIN, BT_NONE, true, true>(bp, zoom);
407  }
408  break;
409  case BM_TRANSPARENT: Draw<BM_TRANSPARENT, RM_NONE, BT_NONE, true, true>(bp, zoom); return;
410  case BM_CRASH_REMAP: Draw<BM_CRASH_REMAP, RM_NONE, BT_NONE, true, true>(bp, zoom); return;
411  case BM_BLACK_REMAP: Draw<BM_BLACK_REMAP, RM_NONE, BT_NONE, true, true>(bp, zoom); return;
412  }
413 }
414 
415 #endif /* WITH_SSE */