ViewVC Help
View File | Revision Log | Show Annotations | Revision Graph | Root Listing
root/cebix/BasiliskII/src/Unix/video_vosf.h
(Generate patch)

Comparing BasiliskII/src/Unix/video_vosf.h (file contents):
Revision 1.18 by cebix, 2001-06-27T19:03:37Z vs.
Revision 1.50 by gbeauche, 2005-04-02T09:50:17Z

# Line 1 | Line 1
1   /*
2   *  video_vosf.h - Video/graphics emulation, video on SEGV signals support
3   *
4 < *  Basilisk II (C) 1997-2001 Christian Bauer
4 > *  Basilisk II (C) 1997-2005 Christian Bauer
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
# Line 21 | Line 21
21   #ifndef VIDEO_VOSF_H
22   #define VIDEO_VOSF_H
23  
24 < // Note: this file is #include'd in video_x.cpp
24 > // Note: this file must be #include'd only in video_x.cpp
25   #ifdef ENABLE_VOSF
26  
27 < /*
28 < *  Page-aligned memory allocation
29 < */
27 > #include "sigsegv.h"
28 > #include "vm_alloc.h"
29 > #ifdef _WIN32
30 > #include "util_windows.h"
31 > #endif
32 >
33 > // Glue for SDL and X11 support
34 > #ifdef USE_SDL_VIDEO
35 > #define MONITOR_INIT                    SDL_monitor_desc &monitor
36 > #define VIDEO_DRV_INIT                  driver_window *drv
37 > #define VIDEO_DRV_ROW_BYTES             drv->s->pitch
38 > #define VIDEO_DRV_LOCK_PIXELS   if (SDL_MUSTLOCK(drv->s)) SDL_LockSurface(drv->s)
39 > #define VIDEO_DRV_UNLOCK_PIXELS if (SDL_MUSTLOCK(drv->s)) SDL_UnlockSurface(drv->s)
40 > #else
41 > #ifdef SHEEPSHAVER
42 > #define MONITOR_INIT                    /* nothing */
43 > #define VIDEO_DRV_INIT                  /* nothing */
44 > #define VIDEO_DRV_WINDOW                the_win
45 > #define VIDEO_DRV_GC                    the_gc
46 > #define VIDEO_DRV_IMAGE                 img
47 > #define VIDEO_DRV_HAVE_SHM              have_shm
48 > #else
49 > #define MONITOR_INIT                    X11_monitor_desc &monitor
50 > #define VIDEO_DRV_INIT                  driver_window *drv
51 > #define VIDEO_DRV_WINDOW                drv->w
52 > #define VIDEO_DRV_GC                    drv->gc
53 > #define VIDEO_DRV_IMAGE                 drv->img
54 > #define VIDEO_DRV_HAVE_SHM              drv->have_shm
55 > #endif
56 > #define VIDEO_DRV_LOCK_PIXELS   /* nothing */
57 > #define VIDEO_DRV_UNLOCK_PIXELS /* nothing */
58 > #define VIDEO_DRV_ROW_BYTES             VIDEO_DRV_IMAGE->bytes_per_line
59 > #endif
60 >
61 > // Variables for Video on SEGV support
62 > static uint8 *the_host_buffer;  // Host frame buffer in VOSF mode
63 > static uint32 the_host_buffer_row_bytes; // Host frame buffer number of bytes per row
64 >
65 > struct ScreenPageInfo {
66 >    int top, bottom;                    // Mapping between this virtual page and Mac scanlines
67 > };
68 >
69 > struct ScreenInfo {
70 >    uintptr memStart;                   // Start address aligned to page boundary
71 >    uint32 memLength;                   // Length of the memory addressed by the screen pages
72 >    
73 >    uintptr pageSize;                   // Size of a page
74 >    int pageBits;                               // Shift count to get the page number
75 >    uint32 pageCount;                   // Number of pages allocated to the screen
76 >    
77 >        bool dirty;                                     // Flag: set if the frame buffer was touched
78 >        bool very_dirty;                        // Flag: set if the frame buffer was completely modified (e.g. colormap changes)
79 >    char * dirtyPages;                  // Table of flags set if page was altered
80 >    ScreenPageInfo * pageInfo;  // Table of mappings page -> Mac scanlines
81 > };
82 >
83 > static ScreenInfo mainBuffer;
84 >
85 > #define PFLAG_SET_VALUE                 0x00
86 > #define PFLAG_CLEAR_VALUE               0x01
87 > #define PFLAG_SET_VALUE_4               0x00000000
88 > #define PFLAG_CLEAR_VALUE_4             0x01010101
89 > #define PFLAG_SET(page)                 mainBuffer.dirtyPages[page] = PFLAG_SET_VALUE
90 > #define PFLAG_CLEAR(page)               mainBuffer.dirtyPages[page] = PFLAG_CLEAR_VALUE
91 > #define PFLAG_ISSET(page)               (mainBuffer.dirtyPages[page] == PFLAG_SET_VALUE)
92 > #define PFLAG_ISCLEAR(page)             (mainBuffer.dirtyPages[page] != PFLAG_SET_VALUE)
93 >
94 > #ifdef UNALIGNED_PROFITABLE
95 > # define PFLAG_ISSET_4(page)    (*((uint32 *)(mainBuffer.dirtyPages + (page))) == PFLAG_SET_VALUE_4)
96 > # define PFLAG_ISCLEAR_4(page)  (*((uint32 *)(mainBuffer.dirtyPages + (page))) == PFLAG_CLEAR_VALUE_4)
97 > #else
98 > # define PFLAG_ISSET_4(page) \
99 >                PFLAG_ISSET(page  ) && PFLAG_ISSET(page+1) \
100 >        &&      PFLAG_ISSET(page+2) && PFLAG_ISSET(page+3)
101 > # define PFLAG_ISCLEAR_4(page) \
102 >                PFLAG_ISCLEAR(page  ) && PFLAG_ISCLEAR(page+1) \
103 >        &&      PFLAG_ISCLEAR(page+2) && PFLAG_ISCLEAR(page+3)
104 > #endif
105 >
106 > // Set the selected page range [ first_page, last_page [ into the SET state
107 > #define PFLAG_SET_RANGE(first_page, last_page) \
108 >        memset(mainBuffer.dirtyPages + (first_page), PFLAG_SET_VALUE, \
109 >                (last_page) - (first_page))
110 >
111 > // Set the selected page range [ first_page, last_page [ into the CLEAR state
112 > #define PFLAG_CLEAR_RANGE(first_page, last_page) \
113 >        memset(mainBuffer.dirtyPages + (first_page), PFLAG_CLEAR_VALUE, \
114 >                (last_page) - (first_page))
115 >
116 > #define PFLAG_SET_ALL do { \
117 >        PFLAG_SET_RANGE(0, mainBuffer.pageCount); \
118 >        mainBuffer.dirty = true; \
119 > } while (0)
120 >
121 > #define PFLAG_CLEAR_ALL do { \
122 >        PFLAG_CLEAR_RANGE(0, mainBuffer.pageCount); \
123 >        mainBuffer.dirty = false; \
124 >        mainBuffer.very_dirty = false; \
125 > } while (0)
126 >
127 > #define PFLAG_SET_VERY_DIRTY do { \
128 >        mainBuffer.very_dirty = true; \
129 > } while (0)
130 >
131 > // Set the following macro definition to 1 if your system
132 > // provides a really fast strchr() implementation
133 > //#define HAVE_FAST_STRCHR 0
134 >
135 > static inline int find_next_page_set(int page)
136 > {
137 > #if HAVE_FAST_STRCHR
138 >        char *match = strchr(mainBuffer.dirtyPages + page, PFLAG_SET_VALUE);
139 >        return match ? match - mainBuffer.dirtyPages : mainBuffer.pageCount;
140 > #else
141 >        while (PFLAG_ISCLEAR_4(page))
142 >                page += 4;
143 >        while (PFLAG_ISCLEAR(page))
144 >                page++;
145 >        return page;
146 > #endif
147 > }
148 >
149 > static inline int find_next_page_clear(int page)
150 > {
151 > #if HAVE_FAST_STRCHR
152 >        char *match = strchr(mainBuffer.dirtyPages + page, PFLAG_CLEAR_VALUE);
153 >        return match ? match - mainBuffer.dirtyPages : mainBuffer.pageCount;
154 > #else
155 >        while (PFLAG_ISSET_4(page))
156 >                page += 4;
157 >        while (PFLAG_ISSET(page))
158 >                page++;
159 >        return page;
160 > #endif
161 > }
162 >
163 > #ifdef HAVE_SPINLOCKS
164 > static spinlock_t vosf_lock = SPIN_LOCK_UNLOCKED;                               // Mutex to protect frame buffer (dirtyPages in fact)
165 > #define LOCK_VOSF spin_lock(&vosf_lock)
166 > #define UNLOCK_VOSF spin_unlock(&vosf_lock)
167 > #elif defined(_WIN32)
168 > static mutex_t vosf_lock;                                                                               // Mutex to protect frame buffer (dirtyPages in fact)
169 > #define LOCK_VOSF vosf_lock.lock();
170 > #define UNLOCK_VOSF vosf_lock.unlock();
171 > #elif defined(HAVE_PTHREADS)
172 > static pthread_mutex_t vosf_lock = PTHREAD_MUTEX_INITIALIZER;   // Mutex to protect frame buffer (dirtyPages in fact)
173 > #define LOCK_VOSF pthread_mutex_lock(&vosf_lock);
174 > #define UNLOCK_VOSF pthread_mutex_unlock(&vosf_lock);
175 > #else
176 > #define LOCK_VOSF
177 > #define UNLOCK_VOSF
178 > #endif
179 >
180 > static int log_base_2(uint32 x)
181 > {
182 >        uint32 mask = 0x80000000;
183 >        int l = 31;
184 >        while (l >= 0 && (x & mask) == 0) {
185 >                mask >>= 1;
186 >                l--;
187 >        }
188 >        return l;
189 > }
190  
191   // Extend size to page boundary
192   static uint32 page_extend(uint32 size)
193   {
194 <        const uint32 page_size = getpagesize();
194 >        const uint32 page_size = vm_get_page_size();
195          const uint32 page_mask = page_size - 1;
196          return (size + page_mask) & ~page_mask;
197   }
198  
199 < // Screen fault handler
200 < static bool screen_fault_handler(sigsegv_address_t fault_address, sigsegv_address_t fault_instruction)
199 >
200 > /*
201 > *  Check if VOSF acceleration is profitable on this platform
202 > */
203 >
204 > const int VOSF_PROFITABLE_TRIES = 3;                    // Make 3 attempts for full screen update
205 > const int VOSF_PROFITABLE_THRESHOLD = 16667;    // 60 Hz
206 >
207 > static bool video_vosf_profitable(void)
208 > {
209 >        int64 durations[VOSF_PROFITABLE_TRIES];
210 >        int mean_duration = 0;
211 >
212 >        for (int i = 0; i < VOSF_PROFITABLE_TRIES; i++) {
213 >                uint64 start = GetTicks_usec();
214 >                for (int p = 0; p < mainBuffer.pageCount; p++) {
215 >                        uint8 *addr = (uint8 *)(mainBuffer.memStart + (p * mainBuffer.pageSize));
216 >                        addr[0] = 0; // Trigger Screen_fault_handler()
217 >                }
218 >                int64 duration = GetTicks_usec() - start;
219 >                mean_duration += duration;
220 >                durations[i] = duration;
221 >
222 >                PFLAG_CLEAR_ALL;
223 >                mainBuffer.dirty = false;
224 >                if (vm_protect((char *)mainBuffer.memStart, mainBuffer.memLength, VM_PAGE_READ) != 0)
225 >                        return false;
226 >        }
227 >
228 >        mean_duration /= VOSF_PROFITABLE_TRIES;
229 >        D(bug("Triggered %d screen faults in %ld usec on average\n", mainBuffer.pageCount, mean_duration));
230 >        return (mean_duration < (VOSF_PROFITABLE_THRESHOLD * (frame_skip ? frame_skip : 1)));
231 > }
232 >
233 >
234 > /*
235 > *  Initialize the VOSF system (mainBuffer structure, SIGSEGV handler)
236 > */
237 >
238 > static bool video_vosf_init(MONITOR_INIT)
239 > {
240 >        VIDEO_MODE_INIT_MONITOR;
241 >
242 >        const uintptr page_size = vm_get_page_size();
243 >        const uintptr page_mask = page_size - 1;
244 >        
245 >        // Round up frame buffer base to page boundary
246 >        mainBuffer.memStart = (((uintptr) the_buffer) + page_mask) & ~page_mask;
247 >        
248 >        // The frame buffer size shall already be aligned to page boundary (use page_extend)
249 >        mainBuffer.memLength = the_buffer_size;
250 >        
251 >        mainBuffer.pageSize = page_size;
252 >        mainBuffer.pageBits = log_base_2(mainBuffer.pageSize);
253 >        mainBuffer.pageCount =  (mainBuffer.memLength + page_mask)/mainBuffer.pageSize;
254 >        
255 >        // The "2" more bytes requested are a safety net to insure the
256 >        // loops in the update routines will terminate.
257 >        // See "How can we deal with array overrun conditions ?" hereunder for further details.
258 >        mainBuffer.dirtyPages = (char *) malloc(mainBuffer.pageCount + 2);
259 >        if (mainBuffer.dirtyPages == NULL)
260 >                return false;
261 >                
262 >        PFLAG_CLEAR_ALL;
263 >        PFLAG_CLEAR(mainBuffer.pageCount);
264 >        PFLAG_SET(mainBuffer.pageCount+1);
265 >        
266 >        // Allocate and fill in pageInfo with start and end (inclusive) row in number of bytes
267 >        mainBuffer.pageInfo = (ScreenPageInfo *) malloc(mainBuffer.pageCount * sizeof(ScreenPageInfo));
268 >        if (mainBuffer.pageInfo == NULL)
269 >                return false;
270 >        
271 >        uint32 a = 0;
272 >        for (unsigned i = 0; i < mainBuffer.pageCount; i++) {
273 >                unsigned y1 = a / VIDEO_MODE_ROW_BYTES;
274 >                if (y1 >= VIDEO_MODE_Y)
275 >                        y1 = VIDEO_MODE_Y - 1;
276 >
277 >                unsigned y2 = (a + mainBuffer.pageSize) / VIDEO_MODE_ROW_BYTES;
278 >                if (y2 >= VIDEO_MODE_Y)
279 >                        y2 = VIDEO_MODE_Y - 1;
280 >
281 >                mainBuffer.pageInfo[i].top = y1;
282 >                mainBuffer.pageInfo[i].bottom = y2;
283 >
284 >                a += mainBuffer.pageSize;
285 >                if (a > mainBuffer.memLength)
286 >                        a = mainBuffer.memLength;
287 >        }
288 >        
289 >        // We can now write-protect the frame buffer
290 >        if (vm_protect((char *)mainBuffer.memStart, mainBuffer.memLength, VM_PAGE_READ) != 0)
291 >                return false;
292 >        
293 >        // The frame buffer is sane, i.e. there is no write to it yet
294 >        mainBuffer.dirty = false;
295 >        return true;
296 > }
297 >
298 >
299 > /*
300 > * Deinitialize VOSF system
301 > */
302 >
303 > static void video_vosf_exit(void)
304 > {
305 >        if (mainBuffer.pageInfo) {
306 >                free(mainBuffer.pageInfo);
307 >                mainBuffer.pageInfo = NULL;
308 >        }
309 >        if (mainBuffer.dirtyPages) {
310 >                free(mainBuffer.dirtyPages);
311 >                mainBuffer.dirtyPages = NULL;
312 >        }
313 > }
314 >
315 >
316 > /*
317 > * Screen fault handler
318 > */
319 >
320 > bool Screen_fault_handler(sigsegv_address_t fault_address, sigsegv_address_t fault_instruction)
321   {
42        D(bug("screen_fault_handler: ADDR=0x%08X from IP=0x%08X\n", fault_address, fault_instruction));
322          const uintptr addr = (uintptr)fault_address;
323          
324          /* Someone attempted to write to the frame buffer. Make it writeable
325 <         * now so that the data could actually be written. It will be made
325 >         * now so that the data could actually be written to. It will be made
326           * read-only back in one of the screen update_*() functions.
327           */
328 <        if ((addr >= mainBuffer.memStart) && (addr < mainBuffer.memEnd)) {
329 <                const int page  = (addr - mainBuffer.memStart) >> mainBuffer.pageBits;
51 <                caddr_t page_ad = (caddr_t)(addr & -mainBuffer.pageSize);
328 >        if (((uintptr)addr - mainBuffer.memStart) < mainBuffer.memLength) {
329 >                const int page  = ((uintptr)addr - mainBuffer.memStart) >> mainBuffer.pageBits;
330                  LOCK_VOSF;
331                  PFLAG_SET(page);
332 <                vm_protect((char *)page_ad, mainBuffer.pageSize, VM_PAGE_READ | VM_PAGE_WRITE);
332 >                vm_protect((char *)(addr & -mainBuffer.pageSize), mainBuffer.pageSize, VM_PAGE_READ | VM_PAGE_WRITE);
333                  mainBuffer.dirty = true;
334                  UNLOCK_VOSF;
335                  return true;
336          }
337          
338          /* Otherwise, we don't know how to handle the fault, let it crash */
61        fprintf(stderr, "do_handle_screen_fault: unhandled address 0x%08X", addr);
62        if (fault_instruction != SIGSEGV_INVALID_PC)
63                fprintf(stderr, " [IP=0x%08X]", fault_instruction);
64        fprintf(stderr, "\n");
339          return false;
340   }
341  
342 +
343   /*
344   *      Update display for Windowed mode and VOSF
345   */
346  
72 // From video_blit.cpp
73 extern void (*Screen_blit)(uint8 * dest, const uint8 * source, uint32 length);
74 extern bool Screen_blitter_init(XVisualInfo * visual_info, bool native_byte_order);
75
347   /*      How can we deal with array overrun conditions ?
348          
349          The state of the framebuffer pages that have been touched are maintained
# Line 106 | Line 377 | There are two cases to check:
377          than pageCount.
378   */
379  
380 < static inline void update_display_window_vosf(void)
380 > static inline void update_display_window_vosf(VIDEO_DRV_INIT)
381   {
382 +        VIDEO_MODE_INIT;
383 +
384          int page = 0;
385          for (;;) {
386 <                const int first_page = find_next_page_set(page);
386 >                const unsigned first_page = find_next_page_set(page);
387                  if (first_page >= mainBuffer.pageCount)
388                          break;
389  
# Line 126 | Line 399 | static inline void update_display_window
399                  const int y1 = mainBuffer.pageInfo[first_page].top;
400                  const int y2 = mainBuffer.pageInfo[page - 1].bottom;
401                  const int height = y2 - y1 + 1;
129                
130                const int bytes_per_row = VideoMonitor.mode.bytes_per_row;
131                const int bytes_per_pixel = VideoMonitor.mode.bytes_per_row / VideoMonitor.mode.x;
132                int i = y1 * bytes_per_row, j;
133                
134                if (depth == 1) {
135
136                        // Update the_host_buffer and copy of the_buffer
137                        for (j = y1; j <= y2; j++) {
138                                Screen_blit(the_host_buffer + i, the_buffer + i, VideoMonitor.mode.x >> 3);
139                                i += bytes_per_row;
140                        }
402  
403 <                } else {
404 <
405 <                        // Update the_host_buffer and copy of the_buffer
406 <                        for (j = y1; j <= y2; j++) {
407 <                                Screen_blit(the_host_buffer + i, the_buffer + i, bytes_per_pixel * VideoMonitor.mode.x);
408 <                                i += bytes_per_row;
409 <                        }
403 >                // Update the_host_buffer
404 >                VIDEO_DRV_LOCK_PIXELS;
405 >                const int src_bytes_per_row = VIDEO_MODE_ROW_BYTES;
406 >                const int dst_bytes_per_row = VIDEO_DRV_ROW_BYTES;
407 >                int i1 = y1 * src_bytes_per_row, i2 = y1 * dst_bytes_per_row, j;
408 >                for (j = y1; j <= y2; j++) {
409 >                        Screen_blit(the_host_buffer + i2, the_buffer + i1, src_bytes_per_row);
410 >                        i1 += src_bytes_per_row;
411 >                        i2 += dst_bytes_per_row;
412                  }
413 +                VIDEO_DRV_UNLOCK_PIXELS;
414  
415 <                if (have_shm)
416 <                        XShmPutImage(x_display, the_win, the_gc, img, 0, y1, 0, y1, VideoMonitor.mode.x, height, 0);
415 > #ifdef USE_SDL_VIDEO
416 >                SDL_UpdateRect(drv->s, 0, y1, VIDEO_MODE_X, height);
417 > #else
418 >                if (VIDEO_DRV_HAVE_SHM)
419 >                        XShmPutImage(x_display, VIDEO_DRV_WINDOW, VIDEO_DRV_GC, VIDEO_DRV_IMAGE, 0, y1, 0, y1, VIDEO_MODE_X, height, 0);
420                  else
421 <                        XPutImage(x_display, the_win, the_gc, img, 0, y1, 0, y1, VideoMonitor.mode.x, height);
421 >                        XPutImage(x_display, VIDEO_DRV_WINDOW, VIDEO_DRV_GC, VIDEO_DRV_IMAGE, 0, y1, 0, y1, VIDEO_MODE_X, height);
422 > #endif
423          }
156
424          mainBuffer.dirty = false;
425   }
426  
427  
428   /*
429   *      Update display for DGA mode and VOSF
430 < *      (only in Direct Addressing mode)
430 > *      (only in Real or Direct Addressing mode)
431   */
432  
433   #if REAL_ADDRESSING || DIRECT_ADDRESSING
434   static inline void update_display_dga_vosf(void)
435   {
436 +        VIDEO_MODE_INIT;
437 +
438 +        if (mainBuffer.very_dirty) {
439 +                PFLAG_CLEAR_ALL;
440 +                vm_protect((char *)mainBuffer.memStart, mainBuffer.memLength, VM_PAGE_READ);
441 +                VIDEO_DRV_LOCK_PIXELS;
442 +                memcpy(the_buffer_copy, the_buffer, VIDEO_MODE_ROW_BYTES * VIDEO_MODE_Y);
443 +                Screen_blit(the_host_buffer, the_buffer, VIDEO_MODE_ROW_BYTES * VIDEO_MODE_Y);
444 +                VIDEO_DRV_UNLOCK_PIXELS;
445 +                return;
446 +        }
447 +
448          int page = 0;
449          for (;;) {
450 <                const int first_page = find_next_page_set(page);
450 >                const unsigned first_page = find_next_page_set(page);
451                  if (first_page >= mainBuffer.pageCount)
452                          break;
453  
# Line 180 | Line 459 | static inline void update_display_dga_vo
459                  const uint32 length = (page - first_page) << mainBuffer.pageBits;
460                  vm_protect((char *)mainBuffer.memStart + offset, length, VM_PAGE_READ);
461                  
462 <                // I am sure that y2 >= y1 and depth != 1
462 >                // There is at least one line to update
463                  const int y1 = mainBuffer.pageInfo[first_page].top;
464                  const int y2 = mainBuffer.pageInfo[page - 1].bottom;
465 <                
466 <                const int bytes_per_row = VideoMonitor.mode.bytes_per_row;
467 <                const int bytes_per_pixel = VideoMonitor.mode.bytes_per_row / VideoMonitor.mode.x;
465 >
466 > #ifndef USE_SDL_VIDEO
467 >                // Update the_host_buffer and copy of the_buffer (use 64 bytes chunks)
468 >                const int src_bytes_per_row = VIDEO_MODE_ROW_BYTES;
469 >                const int dst_bytes_per_row = the_host_buffer_row_bytes;
470 >                const int n_pixels = 64;
471 >                const int n_chunks = VIDEO_MODE_X / n_pixels;
472 >                const int src_chunk_size = src_bytes_per_row / n_chunks;
473 >                const int dst_chunk_size = dst_bytes_per_row / n_chunks;
474 >                const int src_chunk_size_left = src_bytes_per_row - (n_chunks * src_chunk_size);
475 >                const int dst_chunk_size_left = dst_bytes_per_row - (n_chunks * dst_chunk_size);
476 >                int i1 = y1 * src_bytes_per_row;
477 >                int i2 = y1 * dst_bytes_per_row;
478 >                VIDEO_DRV_LOCK_PIXELS;
479 >                for (int j = y1; j <= y2; j++) {
480 >                        for (int i = 0; i < n_chunks; i++) {
481 >                                if (memcmp(the_buffer_copy + i1, the_buffer + i1, src_chunk_size) != 0) {
482 >                                        memcpy(the_buffer_copy + i1, the_buffer + i1, src_chunk_size);
483 >                                        Screen_blit(the_host_buffer + i2, the_buffer + i1, src_chunk_size);
484 >                                }
485 >                                i1 += src_chunk_size;
486 >                                i2 += dst_chunk_size;
487 >                        }
488 >                        if (src_chunk_size_left && dst_chunk_size_left) {
489 >                                if (memcmp(the_buffer_copy + i1, the_buffer + i1, src_chunk_size_left) != 0) {
490 >                                        memcpy(the_buffer_copy + i1, the_buffer + i1, src_chunk_size_left);
491 >                                        Screen_blit(the_host_buffer + i2, the_buffer + i1, src_chunk_size_left);
492 >                                }
493 >                                i1 += src_chunk_size_left;
494 >                                i2 += dst_chunk_size_left;
495 >                        }
496 >                }
497 >                VIDEO_DRV_UNLOCK_PIXELS;
498 > #else
499 >                // Check for first chunk from left and first chunk from right that have changed
500 >                typedef uint64 chunk_t;
501 >                const int chunk_size = sizeof(chunk_t);
502 >                const int bytes_per_row = VIDEO_MODE_ROW_BYTES;
503 >
504                  int i, j;
505 <                
506 <                // Check for first column from left and first column
192 <                // from right that have changed
193 <                int x1 = VideoMonitor.mode.x * bytes_per_pixel - 1;
505 >                int b1 = bytes_per_row / chunk_size;
506 >                int b2 = 0;
507                  for (j = y1; j <= y2; j++) {
508 <                        uint8 * const p1 = &the_buffer[j * bytes_per_row];
509 <                        uint8 * const p2 = &the_buffer_copy[j * bytes_per_row];
510 <                        for (i = 0; i < x1; i++) {
508 >                        chunk_t * const p1 = (chunk_t *)(the_buffer + (j * bytes_per_row));
509 >                        chunk_t * const p2 = (chunk_t *)(the_buffer_copy + (j * bytes_per_row));
510 >                        for (i = 0; i < b1; i++) {
511                                  if (p1[i] != p2[i]) {
512 <                                        x1 = i;
512 >                                        b1 = i;
513                                          break;
514                                  }
515                          }
516 <                }
517 <                x1 /= bytes_per_pixel;
518 <                
206 <                int x2 = x1 * bytes_per_pixel;
207 <                for (j = y2; j >= y1; j--) {
208 <                        uint8 * const p1 = &the_buffer[j * bytes_per_row];
209 <                        uint8 * const p2 = &the_buffer_copy[j * bytes_per_row];
210 <                        for (i = VideoMonitor.mode.x * bytes_per_pixel - 1; i > x2; i--) {
516 >                        if (b1 > b2)
517 >                                b2 = b1;
518 >                        for (i = (bytes_per_row / chunk_size) - 1; i > b2; i--) {
519                                  if (p1[i] != p2[i]) {
520 <                                        x2 = i;
520 >                                        b2 = i;
521                                          break;
522                                  }
523                          }
524                  }
525 <                x2 /= bytes_per_pixel;
526 <                
525 >                b2++;
526 >
527 >                // Convert to pixel information
528 >                int x1, x2;
529 >                switch (VIDEO_MODE_DEPTH) {
530 >                case VIDEO_DEPTH_1BIT:  x1 = (b1 * chunk_size) << 3; x2 = (b2 * chunk_size) << 3;       break;
531 >                case VIDEO_DEPTH_2BIT:  x1 = (b1 * chunk_size) << 2; x2 = (b2 * chunk_size) << 2;       break;
532 >                case VIDEO_DEPTH_4BIT:  x1 = (b1 * chunk_size) << 1; x2 = (b2 * chunk_size) << 1;       break;
533 >                case VIDEO_DEPTH_8BIT:  x1 = b1 * chunk_size; x2 = b2 * chunk_size;                                     break;
534 >                case VIDEO_DEPTH_16BIT: x1 = (b1 * chunk_size) >> 1; x2 = (b2 * chunk_size) >> 1;       break;
535 >                case VIDEO_DEPTH_32BIT: x1 = (b1 * chunk_size) >> 2; x2 = (b2 * chunk_size) >> 2;       break;
536 >                }
537 >                const int width = x2 - x1;
538 >
539 >                // Normalize bounds for for the next blit
540 >                const int src_bytes_per_row = VIDEO_MODE_ROW_BYTES;
541 >                const int dst_bytes_per_row = the_host_buffer_row_bytes;
542 >                const int dst_bytes_per_pixel = dst_bytes_per_row / VIDEO_MODE_X;
543 >                int i2 = y1 * dst_bytes_per_row + x1 * dst_bytes_per_pixel;
544 >                int i1, n_bytes;
545 >                if ((int)VIDEO_MODE_DEPTH < VIDEO_DEPTH_8BIT) {
546 >                        const int src_pixels_per_byte = VIDEO_MODE_X / src_bytes_per_row;
547 >                        i1 = y1 * src_bytes_per_row + x1 / src_pixels_per_byte;
548 >                        n_bytes = width / src_pixels_per_byte;
549 >                } else {
550 >                        const int src_bytes_per_pixel = src_bytes_per_row / VIDEO_MODE_X;
551 >                        i1 = y1 * src_bytes_per_row + x1 * src_bytes_per_pixel;
552 >                        n_bytes = width * src_bytes_per_pixel;
553 >                }
554 >
555                  // Update the_host_buffer and copy of the_buffer
556 <                // There should be at least one pixel to copy
221 <                const int width = x2 - x1 + 1;
222 <                i = y1 * bytes_per_row + x1 * bytes_per_pixel;
556 >                VIDEO_DRV_LOCK_PIXELS;
557                  for (j = y1; j <= y2; j++) {
558 <                        Screen_blit(the_host_buffer + i, the_buffer + i, bytes_per_pixel * width);
559 <                        memcpy(the_buffer_copy + i, the_buffer + i, bytes_per_pixel * width);
560 <                        i += bytes_per_row;
558 >                        Screen_blit(the_host_buffer + i2, the_buffer + i1, n_bytes);
559 >                        memcpy(the_buffer_copy + i1, the_buffer + i1, n_bytes);
560 >                        i1 += src_bytes_per_row;
561 >                        i2 += dst_bytes_per_row;
562                  }
563 +                VIDEO_DRV_UNLOCK_PIXELS;
564 + #endif
565          }
566          mainBuffer.dirty = false;
567   }

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines