notaz.gp2x.de / pcsx_rearmed.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
WIP: Add ARM-assembly versions of lighting and blending
[pcsx_rearmed.git] / plugins / gpu_unai / gpu_inner.h
CommitLineData
86aad47b 1/***************************************************************************
2* Copyright (C) 2010 PCSX4ALL Team *
3* Copyright (C) 2010 Unai *
030d1121 4* Copyright (C) 2016 Senquack (dansilsby <AT> gmail <DOT> com) *
86aad47b 5* *
6* This program is free software; you can redistribute it and/or modify *
7* it under the terms of the GNU General Public License as published by *
8* the Free Software Foundation; either version 2 of the License, or *
9* (at your option) any later version. *
10* *
11* This program is distributed in the hope that it will be useful, *
12* but WITHOUT ANY WARRANTY; without even the implied warranty of *
13* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the *
14* GNU General Public License for more details. *
15* *
16* You should have received a copy of the GNU General Public License *
17* along with this program; if not, write to the *
18* Free Software Foundation, Inc., *
19* 51 Franklin Street, Fifth Floor, Boston, MA 02111-1307 USA. *
20***************************************************************************/
21
22///////////////////////////////////////////////////////////////////////////////
030d1121 23// Inner loop driver instantiation file
86aad47b 24
25///////////////////////////////////////////////////////////////////////////////
030d1121 26// Option Masks (CF template paramter)
27#define CF_LIGHT ((CF>> 0)&1) // Lighting
28#define CF_BLEND ((CF>> 1)&1) // Blending
29#define CF_MASKCHECK ((CF>> 2)&1) // Mask bit check
30#define CF_BLENDMODE ((CF>> 3)&3) // Blend mode 0..3
31#define CF_TEXTMODE ((CF>> 5)&3) // Texture mode 1..3 (0: texturing disabled)
32#define CF_GOURAUD ((CF>> 7)&1) // Gouraud shading
33#define CF_MASKSET ((CF>> 8)&1) // Mask bit set
34#define CF_DITHER ((CF>> 9)&1) // Dithering
35#define CF_BLITMASK ((CF>>10)&1) // blit_mask check (skip rendering pixels
36 // that wouldn't end up displayed on
37 // low-res screen using simple downscaler)
86aad47b 38
030d1121 39//#ifdef __arm__
40//#ifndef ENABLE_GPU_ARMV7
41/* ARMv5 */
42//#include "gpu_inner_blend_arm5.h"
43//#else
44/* ARMv7 optimized */
45//#include "gpu_inner_blend_arm7.h"
46//#endif
47//#else
48//#include "gpu_inner_blend.h"
49//#endif
86aad47b 50
51#include "gpu_inner_blend.h"
030d1121 52
788f5e89
JW		 53#ifdef __arm__
54#include "gpu_inner_blend_arm.h"
55#define gpuBlending gpuBlendingARM
56#else
57#define gpuBlending gpuBlendingGeneric
58#endif
59
030d1121 60#include "gpu_inner_quantization.h"
86aad47b 61#include "gpu_inner_light.h"
62
788f5e89
JW		 63#ifdef __arm__
64#include "gpu_inner_light_arm.h"
65#define gpuLightingRGB gpuLightingRGBARM
66#define gpuLightingTXT gpuLightingTXTARM
67#define gpuLightingTXTGouraud gpuLightingTXTGouraudARM
68#else
69#define gpuLightingRGB gpuLightingRGBGeneric
70#define gpuLightingTXT gpuLightingTXTGeneric
71#define gpuLightingTXTGouraud gpuLightingTXTGouraudGeneric
72#endif
73
74
030d1121 75// If defined, Gouraud colors are fixed-point 5.11, otherwise they are 8.16
76// This is only for debugging/verification of low-precision colors in C.
77// Low-precision Gouraud is intended for use by SIMD-optimized inner drivers
78// which get/use Gouraud colors in SIMD registers.
79//#define GPU_GOURAUD_LOW_PRECISION
80
81// How many bits of fixed-point precision GouraudColor uses
82#ifdef GPU_GOURAUD_LOW_PRECISION
83#define GPU_GOURAUD_FIXED_BITS 11
84#else
85#define GPU_GOURAUD_FIXED_BITS 16
86#endif
87
88// Used to pass Gouraud colors to gpuPixelSpanFn() (lines)
89struct GouraudColor {
90#ifdef GPU_GOURAUD_LOW_PRECISION
91 u16 r, g, b;
92 s16 r_incr, g_incr, b_incr;
93#else
94 u32 r, g, b;
95 s32 r_incr, g_incr, b_incr;
96#endif
97};
98
99static inline u16 gpuGouraudColor15bpp(u32 r, u32 g, u32 b)
100{
101 r >>= GPU_GOURAUD_FIXED_BITS;
102 g >>= GPU_GOURAUD_FIXED_BITS;
103 b >>= GPU_GOURAUD_FIXED_BITS;
104
105#ifndef GPU_GOURAUD_LOW_PRECISION
106 // High-precision Gouraud colors are 8-bit + fractional
107 r >>= 3; g >>= 3; b >>= 3;
108#endif
109
110 return r | (g << 5) | (b << 10);
111}
112
86aad47b 113///////////////////////////////////////////////////////////////////////////////
030d1121 114// GPU Pixel span operations generator gpuPixelSpanFn<>
115// Oct 2016: Created/adapted from old gpuPixelFn by senquack:
116// Original gpuPixelFn was used to draw lines one pixel at a time. I wrote
117// new line algorithms that draw lines using horizontal/vertical/diagonal
118// spans of pixels, necessitating new pixel-drawing function that could
119// not only render spans of pixels, but gouraud-shade them as well.
120// This speeds up line rendering and would allow tile-rendering (untextured
121// rectangles) to use the same set of functions. Since tiles are always
122// monochrome, they simply wouldn't use the extra set of 32 gouraud-shaded
123// gpuPixelSpanFn functions (TODO?).
124//
125// NOTE: While the PS1 framebuffer is 16 bit, we use 8-bit pointers here,
126// so that pDst can be incremented directly by 'incr' parameter
127// without having to shift it before use.
128template<int CF>
129static u8* gpuPixelSpanFn(u8* pDst, uintptr_t data, ptrdiff_t incr, size_t len)
86aad47b 130{
030d1121 131 // Blend func can save an operation if it knows uSrc MSB is
132 // unset. For untextured prims, this is always true.
133 const bool skip_uSrc_mask = true;
134
135 u16 col;
136 struct GouraudColor * gcPtr;
137 u32 r, g, b;
138 s32 r_incr, g_incr, b_incr;
139
140 if (CF_GOURAUD) {
141 gcPtr = (GouraudColor*)data;
142 r = gcPtr->r; r_incr = gcPtr->r_incr;
143 g = gcPtr->g; g_incr = gcPtr->g_incr;
144 b = gcPtr->b; b_incr = gcPtr->b_incr;
145 } else {
146 col = (u16)data;
86aad47b 147 }
030d1121 148
149 do {
150 if (!CF_GOURAUD)
151 { // NO GOURAUD
152 if (!CF_MASKCHECK && !CF_BLEND) {
153 if (CF_MASKSET) { *(u16*)pDst = col | 0x8000; }
154 else { *(u16*)pDst = col; }
155 } else if (CF_MASKCHECK && !CF_BLEND) {
156 if (!(*(u16*)pDst & 0x8000)) {
157 if (CF_MASKSET) { *(u16*)pDst = col | 0x8000; }
158 else { *(u16*)pDst = col; }
159 }
160 } else {
161 u16 uDst = *(u16*)pDst;
162 if (CF_MASKCHECK) { if (uDst & 0x8000) goto endpixel; }
163
164 u16 uSrc = col;
165
166 if (CF_BLEND)
167 uSrc = gpuBlending<CF_BLENDMODE, skip_uSrc_mask>(uSrc, uDst);
168
169 if (CF_MASKSET) { *(u16*)pDst = uSrc | 0x8000; }
170 else { *(u16*)pDst = uSrc; }
171 }
172
173 } else
174 { // GOURAUD
175
176 if (!CF_MASKCHECK && !CF_BLEND) {
177 col = gpuGouraudColor15bpp(r, g, b);
178 if (CF_MASKSET) { *(u16*)pDst = col | 0x8000; }
179 else { *(u16*)pDst = col; }
180 } else if (CF_MASKCHECK && !CF_BLEND) {
181 col = gpuGouraudColor15bpp(r, g, b);
182 if (!(*(u16*)pDst & 0x8000)) {
183 if (CF_MASKSET) { *(u16*)pDst = col | 0x8000; }
184 else { *(u16*)pDst = col; }
185 }
186 } else {
187 u16 uDst = *(u16*)pDst;
188 if (CF_MASKCHECK) { if (uDst & 0x8000) goto endpixel; }
189 col = gpuGouraudColor15bpp(r, g, b);
190
191 u16 uSrc = col;
192
193 // Blend func can save an operation if it knows uSrc MSB is
194 // unset. For untextured prims, this is always true.
195 const bool skip_uSrc_mask = true;
196
197 if (CF_BLEND)
198 uSrc = gpuBlending<CF_BLENDMODE, skip_uSrc_mask>(uSrc, uDst);
199
200 if (CF_MASKSET) { *(u16*)pDst = uSrc | 0x8000; }
201 else { *(u16*)pDst = uSrc; }
202 }
86aad47b 203 }
030d1121 204
205endpixel:
206 if (CF_GOURAUD) {
207 r += r_incr;
208 g += g_incr;
209 b += b_incr;
210 }
211 pDst += incr;
212 } while (len-- > 1);
213
214 // Note from senquack: Normally, I'd prefer to write a 'do {} while (--len)'
215 // loop, or even a for() loop, however, on MIPS platforms anything but the
216 // 'do {} while (len-- > 1)' tends to generate very unoptimal asm, with
217 // many unneeded MULs/ADDs/branches at the ends of these functions.
218 // If you change the loop structure above, be sure to compare the quality
219 // of the generated code!!
220
221 if (CF_GOURAUD) {
222 gcPtr->r = r;
223 gcPtr->g = g;
224 gcPtr->b = b;
86aad47b 225 }
030d1121 226 return pDst;
227}
228
229static u8* PixelSpanNULL(u8* pDst, uintptr_t data, ptrdiff_t incr, size_t len)
230{
231 #ifdef ENABLE_GPU_LOG_SUPPORT
232 fprintf(stdout,"PixelSpanNULL()\n");
233 #endif
234 return pDst;
86aad47b 235}
86aad47b 236
237///////////////////////////////////////////////////////////////////////////////
030d1121 238// PixelSpan (lines) innerloops driver
239typedef u8* (*PSD)(u8* dst, uintptr_t data, ptrdiff_t incr, size_t len);
240
241const PSD gpuPixelSpanDrivers[64] =
86aad47b 242{
030d1121 243 // Array index | 'CF' template field | Field value
244 // ------------+---------------------+----------------
245 // Bit 0 | CF_BLEND | off (0), on (1)
246 // Bit 1 | CF_MASKCHECK | off (0), on (1)
247 // Bit 3:2 | CF_BLENDMODE | 0..3
248 // Bit 4 | CF_MASKSET | off (0), on (1)
249 // Bit 5 | CF_GOURAUD | off (0), on (1)
250 //
251 // NULL entries are ones for which blending is disabled and blend-mode
252 // field is non-zero, which is obviously invalid.
253
254 // Flat-shaded
255 gpuPixelSpanFn<0x00<<1>, gpuPixelSpanFn<0x01<<1>, gpuPixelSpanFn<0x02<<1>, gpuPixelSpanFn<0x03<<1>,
256 PixelSpanNULL, gpuPixelSpanFn<0x05<<1>, PixelSpanNULL, gpuPixelSpanFn<0x07<<1>,
257 PixelSpanNULL, gpuPixelSpanFn<0x09<<1>, PixelSpanNULL, gpuPixelSpanFn<0x0B<<1>,
258 PixelSpanNULL, gpuPixelSpanFn<0x0D<<1>, PixelSpanNULL, gpuPixelSpanFn<0x0F<<1>,
259
260 // Flat-shaded + PixelMSB (CF_MASKSET)
261 gpuPixelSpanFn<(0x00<<1)|0x100>, gpuPixelSpanFn<(0x01<<1)|0x100>, gpuPixelSpanFn<(0x02<<1)|0x100>, gpuPixelSpanFn<(0x03<<1)|0x100>,
262 PixelSpanNULL, gpuPixelSpanFn<(0x05<<1)|0x100>, PixelSpanNULL, gpuPixelSpanFn<(0x07<<1)|0x100>,
263 PixelSpanNULL, gpuPixelSpanFn<(0x09<<1)|0x100>, PixelSpanNULL, gpuPixelSpanFn<(0x0B<<1)|0x100>,
264 PixelSpanNULL, gpuPixelSpanFn<(0x0D<<1)|0x100>, PixelSpanNULL, gpuPixelSpanFn<(0x0F<<1)|0x100>,
265
266 // Gouraud-shaded (CF_GOURAUD)
267 gpuPixelSpanFn<(0x00<<1)|0x80>, gpuPixelSpanFn<(0x01<<1)|0x80>, gpuPixelSpanFn<(0x02<<1)|0x80>, gpuPixelSpanFn<(0x03<<1)|0x80>,
268 PixelSpanNULL, gpuPixelSpanFn<(0x05<<1)|0x80>, PixelSpanNULL, gpuPixelSpanFn<(0x07<<1)|0x80>,
269 PixelSpanNULL, gpuPixelSpanFn<(0x09<<1)|0x80>, PixelSpanNULL, gpuPixelSpanFn<(0x0B<<1)|0x80>,
270 PixelSpanNULL, gpuPixelSpanFn<(0x0D<<1)|0x80>, PixelSpanNULL, gpuPixelSpanFn<(0x0F<<1)|0x80>,
271
272 // Gouraud-shaded (CF_GOURAUD) + PixelMSB (CF_MASKSET)
273 gpuPixelSpanFn<(0x00<<1)|0x180>, gpuPixelSpanFn<(0x01<<1)|0x180>, gpuPixelSpanFn<(0x02<<1)|0x180>, gpuPixelSpanFn<(0x03<<1)|0x180>,
274 PixelSpanNULL, gpuPixelSpanFn<(0x05<<1)|0x180>, PixelSpanNULL, gpuPixelSpanFn<(0x07<<1)|0x180>,
275 PixelSpanNULL, gpuPixelSpanFn<(0x09<<1)|0x180>, PixelSpanNULL, gpuPixelSpanFn<(0x0B<<1)|0x180>,
276 PixelSpanNULL, gpuPixelSpanFn<(0x0D<<1)|0x180>, PixelSpanNULL, gpuPixelSpanFn<(0x0F<<1)|0x180>
86aad47b 277};
278
279///////////////////////////////////////////////////////////////////////////////
280// GPU Tiles innerloops generator
281
030d1121 282template<int CF>
283static void gpuTileSpanFn(u16 *pDst, u32 count, u16 data)
86aad47b 284{
030d1121 285 if (!CF_MASKCHECK && !CF_BLEND) {
286 if (CF_MASKSET) { data = data | 0x8000; }
86aad47b 287 do { *pDst++ = data; } while (--count);
030d1121 288 } else if (CF_MASKCHECK && !CF_BLEND) {
289 if (CF_MASKSET) { data = data | 0x8000; }
86aad47b 290 do { if (!(*pDst&0x8000)) { *pDst = data; } pDst++; } while (--count);
030d1121 291 } else
86aad47b 292 {
030d1121 293 // Blend func can save an operation if it knows uSrc MSB is
294 // unset. For untextured prims, this is always true.
295 const bool skip_uSrc_mask = true;
296
297 u16 uSrc, uDst;
86aad47b 298 do
299 {
030d1121 300 if (CF_MASKCHECK || CF_BLEND) { uDst = *pDst; }
301 if (CF_MASKCHECK) { if (uDst&0x8000) goto endtile; }
302
86aad47b 303 uSrc = data;
304
030d1121 305 if (CF_BLEND)
306 uSrc = gpuBlending<CF_BLENDMODE, skip_uSrc_mask>(uSrc, uDst);
86aad47b 307
030d1121 308 if (CF_MASKSET) { *pDst = uSrc | 0x8000; }
309 else { *pDst = uSrc; }
310
311 //senquack - Did not apply "Silent Hill" mask-bit fix to here.
312 // It is hard to tell from scarce documentation available and
313 // lack of comments in code, but I believe the tile-span
314 // functions here should not bother to preserve any source MSB,
315 // as they are not drawing from a texture.
316endtile:
317 pDst++;
86aad47b 318 }
319 while (--count);
320 }
321}
322
030d1121 323static void TileNULL(u16 *pDst, u32 count, u16 data)
324{
325 #ifdef ENABLE_GPU_LOG_SUPPORT
326 fprintf(stdout,"TileNULL()\n");
327 #endif
328}
329
86aad47b 330///////////////////////////////////////////////////////////////////////////////
331// Tiles innerloops driver
332typedef void (*PT)(u16 *pDst, u32 count, u16 data);
86aad47b 333
030d1121 334// Template instantiation helper macros
335#define TI(cf) gpuTileSpanFn<(cf)>
336#define TN TileNULL
337#define TIBLOCK(ub) \
338 TI((ub)|0x00), TI((ub)|0x02), TI((ub)|0x04), TI((ub)|0x06), \
339 TN, TI((ub)|0x0a), TN, TI((ub)|0x0e), \
340 TN, TI((ub)|0x12), TN, TI((ub)|0x16), \
341 TN, TI((ub)|0x1a), TN, TI((ub)|0x1e)
342
343const PT gpuTileSpanDrivers[32] = {
344 TIBLOCK(0<<8), TIBLOCK(1<<8)
86aad47b 345};
346
030d1121 347#undef TI
348#undef TN
349#undef TIBLOCK
350
351
86aad47b 352///////////////////////////////////////////////////////////////////////////////
353// GPU Sprites innerloops generator
354
030d1121 355template<int CF>
356static void gpuSpriteSpanFn(u16 *pDst, u32 count, u8* pTxt, u32 u0)
86aad47b 357{
030d1121 358 // Blend func can save an operation if it knows uSrc MSB is unset.
359 // Untextured prims can always skip (source color always comes with MSB=0).
360 // For textured prims, lighting funcs always return it unset. (bonus!)
361 const bool skip_uSrc_mask = (!CF_TEXTMODE) || CF_LIGHT;
362
363 u16 uSrc, uDst, srcMSB;
364 u32 u0_mask = gpu_unai.TextureWindow[2];
365
366 u8 r5, g5, b5;
367 if (CF_LIGHT) {
368 r5 = gpu_unai.r5;
369 g5 = gpu_unai.g5;
370 b5 = gpu_unai.b5;
371 }
372
373 if (CF_TEXTMODE==3) {
374 // Texture is accessed byte-wise, so adjust mask if 16bpp
375 u0_mask <<= 1;
376 }
377
378 const u16 *CBA_; if (CF_TEXTMODE!=3) CBA_ = gpu_unai.CBA;
86aad47b 379
380 do
381 {
030d1121 382 if (CF_MASKCHECK || CF_BLEND) { uDst = *pDst; }
383 if (CF_MASKCHECK) if (uDst&0x8000) { goto endsprite; }
86aad47b 384
030d1121 385 if (CF_TEXTMODE==1) { // 4bpp (CLUT)
386 u8 rgb = pTxt[(u0 & u0_mask)>>1];
387 uSrc = CBA_[(rgb>>((u0&1)<<2))&0xf];
86aad47b 388 }
030d1121 389 if (CF_TEXTMODE==2) { // 8bpp (CLUT)
390 uSrc = CBA_[pTxt[u0 & u0_mask]];
391 }
392 if (CF_TEXTMODE==3) { // 16bpp
393 uSrc = *(u16*)(&pTxt[u0 & u0_mask]);
86aad47b 394 }
395
030d1121 396 if (!uSrc) goto endsprite;
397
398 //senquack - save source MSB, as blending or lighting macros will not
399 // (Silent Hill gray rectangles mask bit bug)
400 if (CF_BLEND || CF_LIGHT) srcMSB = uSrc & 0x8000;
86aad47b 401
030d1121 402 if (CF_LIGHT)
403 uSrc = gpuLightingTXT(uSrc, r5, g5, b5);
404
405 if (CF_BLEND && srcMSB)
406 uSrc = gpuBlending<CF_BLENDMODE, skip_uSrc_mask>(uSrc, uDst);
407
408 if (CF_MASKSET) { *pDst = uSrc | 0x8000; }
409 else if (CF_BLEND || CF_LIGHT) { *pDst = uSrc | srcMSB; }
410 else { *pDst = uSrc; }
411
412endsprite:
413 u0 += (CF_TEXTMODE==3) ? 2 : 1;
414 pDst++;
86aad47b 415 }
416 while (--count);
417}
030d1121 418
419static void SpriteNULL(u16 *pDst, u32 count, u8* pTxt, u32 u0)
420{
421 #ifdef ENABLE_GPU_LOG_SUPPORT
422 fprintf(stdout,"SpriteNULL()\n");
423 #endif
424}
425
86aad47b 426///////////////////////////////////////////////////////////////////////////////
427
428///////////////////////////////////////////////////////////////////////////////
429// Sprite innerloops driver
030d1121 430typedef void (*PS)(u16 *pDst, u32 count, u8* pTxt, u32 u0);
431
432// Template instantiation helper macros
433#define TI(cf) gpuSpriteSpanFn<(cf)>
434#define TN SpriteNULL
435#define TIBLOCK(ub) \
436 TN, TN, TN, TN, TN, TN, TN, TN, \
437 TN, TN, TN, TN, TN, TN, TN, TN, \
438 TN, TN, TN, TN, TN, TN, TN, TN, \
439 TN, TN, TN, TN, TN, TN, TN, TN, \
440 TI((ub)|0x20), TI((ub)|0x21), TI((ub)|0x22), TI((ub)|0x23), TI((ub)|0x24), TI((ub)|0x25), TI((ub)|0x26), TI((ub)|0x27), \
441 TN, TN, TI((ub)|0x2a), TI((ub)|0x2b), TN, TN, TI((ub)|0x2e), TI((ub)|0x2f), \
442 TN, TN, TI((ub)|0x32), TI((ub)|0x33), TN, TN, TI((ub)|0x36), TI((ub)|0x37), \
443 TN, TN, TI((ub)|0x3a), TI((ub)|0x3b), TN, TN, TI((ub)|0x3e), TI((ub)|0x3f), \
444 TI((ub)|0x40), TI((ub)|0x41), TI((ub)|0x42), TI((ub)|0x43), TI((ub)|0x44), TI((ub)|0x45), TI((ub)|0x46), TI((ub)|0x47), \
445 TN, TN, TI((ub)|0x4a), TI((ub)|0x4b), TN, TN, TI((ub)|0x4e), TI((ub)|0x4f), \
446 TN, TN, TI((ub)|0x52), TI((ub)|0x53), TN, TN, TI((ub)|0x56), TI((ub)|0x57), \
447 TN, TN, TI((ub)|0x5a), TI((ub)|0x5b), TN, TN, TI((ub)|0x5e), TI((ub)|0x5f), \
448 TI((ub)|0x60), TI((ub)|0x61), TI((ub)|0x62), TI((ub)|0x63), TI((ub)|0x64), TI((ub)|0x65), TI((ub)|0x66), TI((ub)|0x67), \
449 TN, TN, TI((ub)|0x6a), TI((ub)|0x6b), TN, TN, TI((ub)|0x6e), TI((ub)|0x6f), \
450 TN, TN, TI((ub)|0x72), TI((ub)|0x73), TN, TN, TI((ub)|0x76), TI((ub)|0x77), \
451 TN, TN, TI((ub)|0x7a), TI((ub)|0x7b), TN, TN, TI((ub)|0x7e), TI((ub)|0x7f)
452
453const PS gpuSpriteSpanDrivers[256] = {
454 TIBLOCK(0<<8), TIBLOCK(1<<8)
86aad47b 455};
456
030d1121 457#undef TI
458#undef TN
459#undef TIBLOCK
460
86aad47b 461///////////////////////////////////////////////////////////////////////////////
462// GPU Polygon innerloops generator
030d1121 463
464//senquack - Newer version with following changes:
465// * Adapted to work with new poly routings in gpu_raster_polygon.h
466// adapted from DrHell GPU. They are less glitchy and use 22.10
467// fixed-point instead of original UNAI's 16.16.
468// * Texture coordinates are no longer packed together into one
469// unsigned int. This seems to lose too much accuracy (they each
470// end up being only 8.7 fixed-point that way) and pixel-droupouts
471// were noticeable both with original code and current DrHell
472// adaptations. An example would be the sky in NFS3. Now, they are
473// stored in separate ints, using separate masks.
474// * Function is no longer INLINE, as it was always called
475// through a function pointer.
476// * Function now ensures the mask bit of source texture is preserved
477// across calls to blending functions (Silent Hill rectangles fix)
478// * November 2016: Large refactoring of blending/lighting when
479// JohnnyF added dithering. See gpu_inner_quantization.h and
480// relevant blend/light headers.
481// (see README_senquack.txt)
482template<int CF>
483static void gpuPolySpanFn(const gpu_unai_t &gpu_unai, u16 *pDst, u32 count)
86aad47b 484{
030d1121 485 // Blend func can save an operation if it knows uSrc MSB is unset.
486 // Untextured prims can always skip this (src color MSB is always 0).
487 // For textured prims, lighting funcs always return it unset. (bonus!)
488 const bool skip_uSrc_mask = (!CF_TEXTMODE) || CF_LIGHT;
489
490 u32 bMsk; if (CF_BLITMASK) bMsk = gpu_unai.blit_mask;
491
492 if (!CF_TEXTMODE)
493 {
494 if (!CF_GOURAUD)
86aad47b 495 {
030d1121 496 // UNTEXTURED, NO GOURAUD
497 const u16 pix15 = gpu_unai.PixelData;
498 do {
499 u16 uSrc, uDst;
500
501 // NOTE: Don't enable CF_BLITMASK pixel skipping (speed hack)
502 // on untextured polys. It seems to do more harm than good: see
503 // gravestone text at end of Medieval intro sequence. -senquack
504 //if (CF_BLITMASK) { if ((bMsk>>((((uintptr_t)pDst)>>1)&7))&1) { goto endpolynotextnogou; } }
505
506 if (CF_BLEND || CF_MASKCHECK) uDst = *pDst;
507 if (CF_MASKCHECK) { if (uDst&0x8000) { goto endpolynotextnogou; } }
508
509 uSrc = pix15;
510
511 if (CF_BLEND)
512 uSrc = gpuBlending<CF_BLENDMODE, skip_uSrc_mask>(uSrc, uDst);
513
514 if (CF_MASKSET) { *pDst = uSrc | 0x8000; }
515 else { *pDst = uSrc; }
516
517endpolynotextnogou:
518 pDst++;
519 } while(--count);
86aad47b 520 }
521 else
522 {
030d1121 523 // UNTEXTURED, GOURAUD
524 u32 l_gCol = gpu_unai.gCol;
525 u32 l_gInc = gpu_unai.gInc;
526
527 do {
528 u16 uDst, uSrc;
529
530 // See note in above loop regarding CF_BLITMASK
531 //if (CF_BLITMASK) { if ((bMsk>>((((uintptr_t)pDst)>>1)&7))&1) goto endpolynotextgou; }
532
533 if (CF_BLEND || CF_MASKCHECK) uDst = *pDst;
534 if (CF_MASKCHECK) { if (uDst&0x8000) goto endpolynotextgou; }
535
536 if (CF_DITHER) {
537 // GOURAUD, DITHER
538
539 u32 uSrc24 = gpuLightingRGB24(l_gCol);
540 if (CF_BLEND)
541 uSrc24 = gpuBlending24<CF_BLENDMODE>(uSrc24, uDst);
542 uSrc = gpuColorQuantization24<CF_DITHER>(uSrc24, pDst);
543 } else {
544 // GOURAUD, NO DITHER
545
546 uSrc = gpuLightingRGB(l_gCol);
547
548 if (CF_BLEND)
549 uSrc = gpuBlending<CF_BLENDMODE, skip_uSrc_mask>(uSrc, uDst);
86aad47b 550 }
030d1121 551
552 if (CF_MASKSET) { *pDst = uSrc | 0x8000; }
553 else { *pDst = uSrc; }
554
555endpolynotextgou:
556 pDst++;
557 l_gCol += l_gInc;
86aad47b 558 }
559 while (--count);
560 }
561 }
562 else
563 {
030d1121 564 // TEXTURED
565
566 u16 uDst, uSrc, srcMSB;
567
568 //senquack - note: original UNAI code had gpu_unai.{u4/v4} packed into
569 // one 32-bit unsigned int, but this proved to lose too much accuracy
570 // (pixel drouputs noticeable in NFS3 sky), so now are separate vars.
571 u32 l_u_msk = gpu_unai.u_msk; u32 l_v_msk = gpu_unai.v_msk;
572 u32 l_u = gpu_unai.u & l_u_msk; u32 l_v = gpu_unai.v & l_v_msk;
573 s32 l_u_inc = gpu_unai.u_inc; s32 l_v_inc = gpu_unai.v_inc;
574
575 const u16* TBA_ = gpu_unai.TBA;
576 const u16* CBA_; if (CF_TEXTMODE!=3) CBA_ = gpu_unai.CBA;
577
578 u8 r5, g5, b5;
579 u8 r8, g8, b8;
580
581 u32 l_gInc, l_gCol;
582
583 if (CF_LIGHT) {
584 if (CF_GOURAUD) {
585 l_gInc = gpu_unai.gInc;
586 l_gCol = gpu_unai.gCol;
587 } else {
588 if (CF_DITHER) {
589 r8 = gpu_unai.r8;
590 g8 = gpu_unai.g8;
591 b8 = gpu_unai.b8;
592 } else {
593 r5 = gpu_unai.r5;
594 g5 = gpu_unai.g5;
595 b5 = gpu_unai.b5;
596 }
597 }
598 }
599
86aad47b 600 do
601 {
030d1121 602 if (CF_BLITMASK) { if ((bMsk>>((((uintptr_t)pDst)>>1)&7))&1) goto endpolytext; }
603 if (CF_MASKCHECK || CF_BLEND) { uDst = *pDst; }
604 if (CF_MASKCHECK) if (uDst&0x8000) { goto endpolytext; }
605
606 //senquack - adapted to work with new 22.10 fixed point routines:
607 // (UNAI originally used 16.16)
608 if (CF_TEXTMODE==1) { // 4bpp (CLUT)
609 u32 tu=(l_u>>10);
610 u32 tv=(l_v<<1)&(0xff<<11);
611 u8 rgb=((u8*)TBA_)[tv+(tu>>1)];
612 uSrc=CBA_[(rgb>>((tu&1)<<2))&0xf];
613 if (!uSrc) goto endpolytext;
614 }
615 if (CF_TEXTMODE==2) { // 8bpp (CLUT)
616 uSrc = CBA_[(((u8*)TBA_)[(l_u>>10)+((l_v<<1)&(0xff<<11))])];
617 if (!uSrc) goto endpolytext;
86aad47b 618 }
030d1121 619 if (CF_TEXTMODE==3) { // 16bpp
620 uSrc = TBA_[(l_u>>10)+((l_v)&(0xff<<10))];
621 if (!uSrc) goto endpolytext;
622 }
623
624 // Save source MSB, as blending or lighting will not (Silent Hill)
625 if (CF_BLEND || CF_LIGHT) srcMSB = uSrc & 0x8000;
626
627 // When textured, only dither when LIGHT (texture blend) is enabled
628 // LIGHT && BLEND => dither
629 // LIGHT && !BLEND => dither
630 //!LIGHT && BLEND => no dither
631 //!LIGHT && !BLEND => no dither
632
633 if (CF_DITHER && CF_LIGHT) {
634 u32 uSrc24;
635 if ( CF_GOURAUD)
636 uSrc24 = gpuLightingTXT24Gouraud(uSrc, l_gCol);
637 if (!CF_GOURAUD)
638 uSrc24 = gpuLightingTXT24(uSrc, r8, g8, b8);
639
640 if (CF_BLEND && srcMSB)
641 uSrc24 = gpuBlending24<CF_BLENDMODE>(uSrc24, uDst);
642
643 uSrc = gpuColorQuantization24<CF_DITHER>(uSrc24, pDst);
644 } else
86aad47b 645 {
030d1121 646 if (CF_LIGHT) {
647 if ( CF_GOURAUD)
648 uSrc = gpuLightingTXTGouraud(uSrc, l_gCol);
649 if (!CF_GOURAUD)
650 uSrc = gpuLightingTXT(uSrc, r5, g5, b5);
651 }
652
653 if (CF_BLEND && srcMSB)
654 uSrc = gpuBlending<CF_BLENDMODE, skip_uSrc_mask>(uSrc, uDst);
86aad47b 655 }
030d1121 656
657 if (CF_MASKSET) { *pDst = uSrc | 0x8000; }
658 else if (CF_BLEND || CF_LIGHT) { *pDst = uSrc | srcMSB; }
659 else { *pDst = uSrc; }
660endpolytext:
661 pDst++;
662 l_u = (l_u + l_u_inc) & l_u_msk;
663 l_v = (l_v + l_v_inc) & l_v_msk;
664 if (CF_LIGHT && CF_GOURAUD) l_gCol += l_gInc;
86aad47b 665 }
666 while (--count);
667 }
668}
faf8bb13 669
030d1121 670static void PolyNULL(const gpu_unai_t &gpu_unai, u16 *pDst, u32 count)
faf8bb13 671{
030d1121 672 #ifdef ENABLE_GPU_LOG_SUPPORT
673 fprintf(stdout,"PolyNULL()\n");
674 #endif
faf8bb13 675}
676
86aad47b 677///////////////////////////////////////////////////////////////////////////////
678// Polygon innerloops driver
030d1121 679typedef void (*PP)(const gpu_unai_t &gpu_unai, u16 *pDst, u32 count);
680
681// Template instantiation helper macros
682#define TI(cf) gpuPolySpanFn<(cf)>
683#define TN PolyNULL
684#define TIBLOCK(ub) \
685 TI((ub)|0x00), TI((ub)|0x01), TI((ub)|0x02), TI((ub)|0x03), TI((ub)|0x04), TI((ub)|0x05), TI((ub)|0x06), TI((ub)|0x07), \
686 TN, TN, TI((ub)|0x0a), TI((ub)|0x0b), TN, TN, TI((ub)|0x0e), TI((ub)|0x0f), \
687 TN, TN, TI((ub)|0x12), TI((ub)|0x13), TN, TN, TI((ub)|0x16), TI((ub)|0x17), \
688 TN, TN, TI((ub)|0x1a), TI((ub)|0x1b), TN, TN, TI((ub)|0x1e), TI((ub)|0x1f), \
689 TI((ub)|0x20), TI((ub)|0x21), TI((ub)|0x22), TI((ub)|0x23), TI((ub)|0x24), TI((ub)|0x25), TI((ub)|0x26), TI((ub)|0x27), \
690 TN, TN, TI((ub)|0x2a), TI((ub)|0x2b), TN, TN, TI((ub)|0x2e), TI((ub)|0x2f), \
691 TN, TN, TI((ub)|0x32), TI((ub)|0x33), TN, TN, TI((ub)|0x36), TI((ub)|0x37), \
692 TN, TN, TI((ub)|0x3a), TI((ub)|0x3b), TN, TN, TI((ub)|0x3e), TI((ub)|0x3f), \
693 TI((ub)|0x40), TI((ub)|0x41), TI((ub)|0x42), TI((ub)|0x43), TI((ub)|0x44), TI((ub)|0x45), TI((ub)|0x46), TI((ub)|0x47), \
694 TN, TN, TI((ub)|0x4a), TI((ub)|0x4b), TN, TN, TI((ub)|0x4e), TI((ub)|0x4f), \
695 TN, TN, TI((ub)|0x52), TI((ub)|0x53), TN, TN, TI((ub)|0x56), TI((ub)|0x57), \
696 TN, TN, TI((ub)|0x5a), TI((ub)|0x5b), TN, TN, TI((ub)|0x5e), TI((ub)|0x5f), \
697 TI((ub)|0x60), TI((ub)|0x61), TI((ub)|0x62), TI((ub)|0x63), TI((ub)|0x64), TI((ub)|0x65), TI((ub)|0x66), TI((ub)|0x67), \
698 TN, TN, TI((ub)|0x6a), TI((ub)|0x6b), TN, TN, TI((ub)|0x6e), TI((ub)|0x6f), \
699 TN, TN, TI((ub)|0x72), TI((ub)|0x73), TN, TN, TI((ub)|0x76), TI((ub)|0x77), \
700 TN, TN, TI((ub)|0x7a), TI((ub)|0x7b), TN, TN, TI((ub)|0x7e), TI((ub)|0x7f), \
701 TN, TI((ub)|0x81), TN, TI((ub)|0x83), TN, TI((ub)|0x85), TN, TI((ub)|0x87), \
702 TN, TN, TN, TI((ub)|0x8b), TN, TN, TN, TI((ub)|0x8f), \
703 TN, TN, TN, TI((ub)|0x93), TN, TN, TN, TI((ub)|0x97), \
704 TN, TN, TN, TI((ub)|0x9b), TN, TN, TN, TI((ub)|0x9f), \
705 TN, TI((ub)|0xa1), TN, TI((ub)|0xa3), TN, TI((ub)|0xa5), TN, TI((ub)|0xa7), \
706 TN, TN, TN, TI((ub)|0xab), TN, TN, TN, TI((ub)|0xaf), \
707 TN, TN, TN, TI((ub)|0xb3), TN, TN, TN, TI((ub)|0xb7), \
708 TN, TN, TN, TI((ub)|0xbb), TN, TN, TN, TI((ub)|0xbf), \
709 TN, TI((ub)|0xc1), TN, TI((ub)|0xc3), TN, TI((ub)|0xc5), TN, TI((ub)|0xc7), \
710 TN, TN, TN, TI((ub)|0xcb), TN, TN, TN, TI((ub)|0xcf), \
711 TN, TN, TN, TI((ub)|0xd3), TN, TN, TN, TI((ub)|0xd7), \
712 TN, TN, TN, TI((ub)|0xdb), TN, TN, TN, TI((ub)|0xdf), \
713 TN, TI((ub)|0xe1), TN, TI((ub)|0xe3), TN, TI((ub)|0xe5), TN, TI((ub)|0xe7), \
714 TN, TN, TN, TI((ub)|0xeb), TN, TN, TN, TI((ub)|0xef), \
715 TN, TN, TN, TI((ub)|0xf3), TN, TN, TN, TI((ub)|0xf7), \
716 TN, TN, TN, TI((ub)|0xfb), TN, TN, TN, TI((ub)|0xff)
717
718const PP gpuPolySpanDrivers[2048] = {
719 TIBLOCK(0<<8), TIBLOCK(1<<8), TIBLOCK(2<<8), TIBLOCK(3<<8),
720 TIBLOCK(4<<8), TIBLOCK(5<<8), TIBLOCK(6<<8), TIBLOCK(7<<8)
86aad47b 721};
030d1121 722
723#undef TI
724#undef TN
725#undef TIBLOCK
ARM optimized PCSX fork