src/Unix/video_vosf.h

/*
 *  video_vosf.h - Video/graphics emulation, video on SEGV signals support
 *
 *  Basilisk II (C) 1997-2005 Christian Bauer
 *
 *  This program 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; either version 2 of the License, or
 *  (at your option) any later version.
 *
 *  This program 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.  See the
 *  GNU General Public License for more details.
 *
 *  You should have received a copy of the GNU General Public License
 *  along with this program; if not, write to the Free Software
 *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
 */

#ifndef VIDEO_VOSF_H
#define VIDEO_VOSF_H

// Note: this file must be #include'd only in video_x.cpp
#ifdef ENABLE_VOSF

#include "sigsegv.h"
#include "vm_alloc.h"
#ifdef _WIN32
#include "util_windows.h"
#endif

// Glue for SDL and X11 support
#ifdef USE_SDL_VIDEO
#define MONITOR_INIT                    SDL_monitor_desc &monitor
#define VIDEO_DRV_WIN_INIT              driver_window *drv
#define VIDEO_DRV_DGA_INIT              driver_fullscreen *drv
#define VIDEO_DRV_LOCK_PIXELS   if (SDL_MUSTLOCK(drv->s)) SDL_LockSurface(drv->s)
#define VIDEO_DRV_UNLOCK_PIXELS if (SDL_MUSTLOCK(drv->s)) SDL_UnlockSurface(drv->s)
#define VIDEO_DRV_DEPTH                 drv->s->format->BitsPerPixel
#define VIDEO_DRV_WIDTH                 drv->s->w
#define VIDEO_DRV_HEIGHT                drv->s->h
#define VIDEO_DRV_ROW_BYTES             drv->s->pitch
#else
#ifdef SHEEPSHAVER
#define MONITOR_INIT                    /* nothing */
#define VIDEO_DRV_WIN_INIT              /* nothing */
#define VIDEO_DRV_DGA_INIT              /* nothing */
#define VIDEO_DRV_WINDOW                the_win
#define VIDEO_DRV_GC                    the_gc
#define VIDEO_DRV_IMAGE                 img
#define VIDEO_DRV_HAVE_SHM              have_shm
#else
#define MONITOR_INIT                    X11_monitor_desc &monitor
#define VIDEO_DRV_WIN_INIT              driver_window *drv
#define VIDEO_DRV_DGA_INIT              driver_dga *drv
#define VIDEO_DRV_WINDOW                drv->w
#define VIDEO_DRV_GC                    drv->gc
#define VIDEO_DRV_IMAGE                 drv->img
#define VIDEO_DRV_HAVE_SHM              drv->have_shm
#endif
#define VIDEO_DRV_LOCK_PIXELS   /* nothing */
#define VIDEO_DRV_UNLOCK_PIXELS /* nothing */
#define VIDEO_DRV_DEPTH                 VIDEO_DRV_IMAGE->depth
#define VIDEO_DRV_WIDTH                 VIDEO_DRV_IMAGE->width
#define VIDEO_DRV_HEIGHT                VIDEO_DRV_IMAGE->height
#define VIDEO_DRV_ROW_BYTES             VIDEO_DRV_IMAGE->bytes_per_line
#endif

// Variables for Video on SEGV support
static uint8 *the_host_buffer;  // Host frame buffer in VOSF mode

struct ScreenPageInfo {
    int top, bottom;                    // Mapping between this virtual page and Mac scanlines
};

struct ScreenInfo {
    uintptr memStart;                   // Start address aligned to page boundary
    uint32 memLength;                   // Length of the memory addressed by the screen pages
    
    uintptr pageSize;                   // Size of a page
    int pageBits;                               // Shift count to get the page number
    uint32 pageCount;                   // Number of pages allocated to the screen
    
        bool dirty;                                     // Flag: set if the frame buffer was touched
        bool very_dirty;                        // Flag: set if the frame buffer was completely modified (e.g. colormap changes)
    char * dirtyPages;                  // Table of flags set if page was altered
    ScreenPageInfo * pageInfo;  // Table of mappings page -> Mac scanlines
};

static ScreenInfo mainBuffer;

#define PFLAG_SET_VALUE                 0x00
#define PFLAG_CLEAR_VALUE               0x01
#define PFLAG_SET_VALUE_4               0x00000000
#define PFLAG_CLEAR_VALUE_4             0x01010101
#define PFLAG_SET(page)                 mainBuffer.dirtyPages[page] = PFLAG_SET_VALUE
#define PFLAG_CLEAR(page)               mainBuffer.dirtyPages[page] = PFLAG_CLEAR_VALUE
#define PFLAG_ISSET(page)               (mainBuffer.dirtyPages[page] == PFLAG_SET_VALUE)
#define PFLAG_ISCLEAR(page)             (mainBuffer.dirtyPages[page] != PFLAG_SET_VALUE)

#ifdef UNALIGNED_PROFITABLE
# define PFLAG_ISSET_4(page)    (*((uint32 *)(mainBuffer.dirtyPages + (page))) == PFLAG_SET_VALUE_4)
# define PFLAG_ISCLEAR_4(page)  (*((uint32 *)(mainBuffer.dirtyPages + (page))) == PFLAG_CLEAR_VALUE_4)
#else
# define PFLAG_ISSET_4(page) \
                PFLAG_ISSET(page  ) && PFLAG_ISSET(page+1) \
        &&      PFLAG_ISSET(page+2) && PFLAG_ISSET(page+3)
# define PFLAG_ISCLEAR_4(page) \
                PFLAG_ISCLEAR(page  ) && PFLAG_ISCLEAR(page+1) \
        &&      PFLAG_ISCLEAR(page+2) && PFLAG_ISCLEAR(page+3)
#endif

// Set the selected page range [ first_page, last_page [ into the SET state
#define PFLAG_SET_RANGE(first_page, last_page) \
        memset(mainBuffer.dirtyPages + (first_page), PFLAG_SET_VALUE, \
                (last_page) - (first_page))

// Set the selected page range [ first_page, last_page [ into the CLEAR state
#define PFLAG_CLEAR_RANGE(first_page, last_page) \
        memset(mainBuffer.dirtyPages + (first_page), PFLAG_CLEAR_VALUE, \
                (last_page) - (first_page))

#define PFLAG_SET_ALL do { \
        PFLAG_SET_RANGE(0, mainBuffer.pageCount); \
        mainBuffer.dirty = true; \
} while (0)

#define PFLAG_CLEAR_ALL do { \
        PFLAG_CLEAR_RANGE(0, mainBuffer.pageCount); \
        mainBuffer.dirty = false; \
        mainBuffer.very_dirty = false; \
} while (0)

#define PFLAG_SET_VERY_DIRTY do { \
        mainBuffer.very_dirty = true; \
} while (0)

// Set the following macro definition to 1 if your system
// provides a really fast strchr() implementation
//#define HAVE_FAST_STRCHR 0

static inline int find_next_page_set(int page)
{
#if HAVE_FAST_STRCHR
        char *match = strchr(mainBuffer.dirtyPages + page, PFLAG_SET_VALUE);
        return match ? match - mainBuffer.dirtyPages : mainBuffer.pageCount;
#else
        while (PFLAG_ISCLEAR_4(page))
                page += 4;
        while (PFLAG_ISCLEAR(page))
                page++;
        return page;
#endif
}

static inline int find_next_page_clear(int page)
{
#if HAVE_FAST_STRCHR
        char *match = strchr(mainBuffer.dirtyPages + page, PFLAG_CLEAR_VALUE);
        return match ? match - mainBuffer.dirtyPages : mainBuffer.pageCount;
#else
        while (PFLAG_ISSET_4(page))
                page += 4;
        while (PFLAG_ISSET(page))
                page++;
        return page;
#endif
}

#if defined(HAVE_PTHREADS)
static pthread_mutex_t vosf_lock = PTHREAD_MUTEX_INITIALIZER;   // Mutex to protect frame buffer (dirtyPages in fact)
#define LOCK_VOSF pthread_mutex_lock(&vosf_lock);
#define UNLOCK_VOSF pthread_mutex_unlock(&vosf_lock);
#elif defined(_WIN32)
static mutex_t vosf_lock;                                                                               // Mutex to protect frame buffer (dirtyPages in fact)
#define LOCK_VOSF vosf_lock.lock();
#define UNLOCK_VOSF vosf_lock.unlock();
#elif defined(HAVE_SPINLOCKS)
static spinlock_t vosf_lock = SPIN_LOCK_UNLOCKED;                               // Mutex to protect frame buffer (dirtyPages in fact)
#define LOCK_VOSF spin_lock(&vosf_lock)
#define UNLOCK_VOSF spin_unlock(&vosf_lock)
#else
#define LOCK_VOSF
#define UNLOCK_VOSF
#endif

static int log_base_2(uint32 x)
{
        uint32 mask = 0x80000000;
        int l = 31;
        while (l >= 0 && (x & mask) == 0) {
                mask >>= 1;
                l--;
        }
        return l;
}

// Extend size to page boundary
static uint32 page_extend(uint32 size)
{
        const uint32 page_size = vm_get_page_size();
        const uint32 page_mask = page_size - 1;
        return (size + page_mask) & ~page_mask;
}


/*
 *  Check if VOSF acceleration is profitable on this platform
 */

const int VOSF_PROFITABLE_TRIES = 3;                    // Make 3 attempts for full screen update
const int VOSF_PROFITABLE_THRESHOLD = 16667;    // 60 Hz

static bool video_vosf_profitable(void)
{
        int64 durations[VOSF_PROFITABLE_TRIES];
        int mean_duration = 0;

        for (int i = 0; i < VOSF_PROFITABLE_TRIES; i++) {
                uint64 start = GetTicks_usec();
                for (int p = 0; p < mainBuffer.pageCount; p++) {
                        uint8 *addr = (uint8 *)(mainBuffer.memStart + (p * mainBuffer.pageSize));
                        addr[0] = 0; // Trigger Screen_fault_handler()
                }
                int64 duration = GetTicks_usec() - start;
                mean_duration += duration;
                durations[i] = duration;

                PFLAG_CLEAR_ALL;
                mainBuffer.dirty = false;
                if (vm_protect((char *)mainBuffer.memStart, mainBuffer.memLength, VM_PAGE_READ) != 0)
                        return false;
        }

        mean_duration /= VOSF_PROFITABLE_TRIES;
        D(bug("Triggered %d screen faults in %ld usec on average\n", mainBuffer.pageCount, mean_duration));
        return (mean_duration < (VOSF_PROFITABLE_THRESHOLD * (frame_skip ? frame_skip : 1)));
}


/*
 *  Initialize the VOSF system (mainBuffer structure, SIGSEGV handler)
 */

static bool video_vosf_init(MONITOR_INIT)
{
        VIDEO_MODE_INIT_MONITOR;

        const uintptr page_size = vm_get_page_size();
        const uintptr page_mask = page_size - 1;
        
        // Round up frame buffer base to page boundary
        mainBuffer.memStart = (((uintptr) the_buffer) + page_mask) & ~page_mask;
        
        // The frame buffer size shall already be aligned to page boundary (use page_extend)
        mainBuffer.memLength = the_buffer_size;
        
        mainBuffer.pageSize = page_size;
        mainBuffer.pageBits = log_base_2(mainBuffer.pageSize);
        mainBuffer.pageCount =  (mainBuffer.memLength + page_mask)/mainBuffer.pageSize;
        
        // The "2" more bytes requested are a safety net to insure the
        // loops in the update routines will terminate.
        // See "How can we deal with array overrun conditions ?" hereunder for further details.
        mainBuffer.dirtyPages = (char *) malloc(mainBuffer.pageCount + 2);
        if (mainBuffer.dirtyPages == NULL)
                return false;
                
        PFLAG_CLEAR_ALL;
        PFLAG_CLEAR(mainBuffer.pageCount);
        PFLAG_SET(mainBuffer.pageCount+1);
        
        // Allocate and fill in pageInfo with start and end (inclusive) row in number of bytes
        mainBuffer.pageInfo = (ScreenPageInfo *) malloc(mainBuffer.pageCount * sizeof(ScreenPageInfo));
        if (mainBuffer.pageInfo == NULL)
                return false;
        
        uint32 a = 0;
        for (unsigned i = 0; i < mainBuffer.pageCount; i++) {
                unsigned y1 = a / VIDEO_MODE_ROW_BYTES;
                if (y1 >= VIDEO_MODE_Y)
                        y1 = VIDEO_MODE_Y - 1;

                unsigned y2 = (a + mainBuffer.pageSize) / VIDEO_MODE_ROW_BYTES;
                if (y2 >= VIDEO_MODE_Y)
                        y2 = VIDEO_MODE_Y - 1;

                mainBuffer.pageInfo[i].top = y1;
                mainBuffer.pageInfo[i].bottom = y2;

                a += mainBuffer.pageSize;
                if (a > mainBuffer.memLength)
                        a = mainBuffer.memLength;
        }
        
        // We can now write-protect the frame buffer
        if (vm_protect((char *)mainBuffer.memStart, mainBuffer.memLength, VM_PAGE_READ) != 0)
                return false;
        
        // The frame buffer is sane, i.e. there is no write to it yet
        mainBuffer.dirty = false;
        return true;
}


/*
 * Deinitialize VOSF system
 */

static void video_vosf_exit(void)
{
        if (mainBuffer.pageInfo) {
                free(mainBuffer.pageInfo);
                mainBuffer.pageInfo = NULL;
        }
        if (mainBuffer.dirtyPages) {
                free(mainBuffer.dirtyPages);
                mainBuffer.dirtyPages = NULL;
        }
}


/*
 * Screen fault handler
 */

bool Screen_fault_handler(sigsegv_address_t fault_address, sigsegv_address_t fault_instruction)
{
        const uintptr addr = (uintptr)fault_address;
        
        /* Someone attempted to write to the frame buffer. Make it writeable
         * now so that the data could actually be written to. It will be made
         * read-only back in one of the screen update_*() functions.
         */
        if (((uintptr)addr - mainBuffer.memStart) < mainBuffer.memLength) {
                const int page  = ((uintptr)addr - mainBuffer.memStart) >> mainBuffer.pageBits;
                LOCK_VOSF;
                PFLAG_SET(page);
                vm_protect((char *)(addr & -mainBuffer.pageSize), mainBuffer.pageSize, VM_PAGE_READ | VM_PAGE_WRITE);
                mainBuffer.dirty = true;
                UNLOCK_VOSF;
                return true;
        }
        
        /* Otherwise, we don't know how to handle the fault, let it crash */
        return false;
}


/*
 *      Update display for Windowed mode and VOSF
 */

/*      How can we deal with array overrun conditions ?
        
        The state of the framebuffer pages that have been touched are maintained
        in the dirtyPages[] table. That table is (pageCount + 2) bytes long.

Terminology
        
        "Last Page" denotes the pageCount-nth page, i.e. dirtyPages[pageCount - 1].
        "CLEAR Page Guard" refers to the page following the Last Page but is always
        in the CLEAR state. "SET Page Guard" refers to the page following the CLEAR
        Page Guard but is always in the SET state.

Rough process
        
        The update routines must determine which pages have to be blitted to the
        screen. This job consists in finding the first_page that was touched.
        i.e. find the next page that is SET. Then, finding how many pages were
        touched starting from first_page. i.e. find the next page that is CLEAR.

There are two cases to check:

        - Last Page is CLEAR: find_next_page_set() will reach the SET Page Guard
        but it is beyond the valid pageCount value. Therefore, we exit from the
        update routine.
        
        - Last Page is SET: first_page equals (pageCount - 1) and
        find_next_page_clear() will reach the CLEAR Page Guard. We blit the last
        page to the screen. On the next iteration, page equals pageCount and
        find_next_page_set() will reach the SET Page Guard. We still safely exit
        from the update routine because the SET Page Guard position is greater
        than pageCount.
*/

static void update_display_window_vosf(VIDEO_DRV_WIN_INIT)
{
        VIDEO_MODE_INIT;

        int page = 0;
        for (;;) {
                const unsigned first_page = find_next_page_set(page);
                if (first_page >= mainBuffer.pageCount)
                        break;

                page = find_next_page_clear(first_page);
                PFLAG_CLEAR_RANGE(first_page, page);

                // Make the dirty pages read-only again
                const int32 offset  = first_page << mainBuffer.pageBits;
                const uint32 length = (page - first_page) << mainBuffer.pageBits;
                vm_protect((char *)mainBuffer.memStart + offset, length, VM_PAGE_READ);
                
                // There is at least one line to update
                const int y1 = mainBuffer.pageInfo[first_page].top;
                const int y2 = mainBuffer.pageInfo[page - 1].bottom;
                const int height = y2 - y1 + 1;

                // Update the_host_buffer
                VIDEO_DRV_LOCK_PIXELS;
                const int src_bytes_per_row = VIDEO_MODE_ROW_BYTES;
                const int dst_bytes_per_row = VIDEO_DRV_ROW_BYTES;
                int i1 = y1 * src_bytes_per_row, i2 = y1 * dst_bytes_per_row, j;
                for (j = y1; j <= y2; j++) {
                        Screen_blit(the_host_buffer + i2, the_buffer + i1, src_bytes_per_row);
                        i1 += src_bytes_per_row;
                        i2 += dst_bytes_per_row;
                }
                VIDEO_DRV_UNLOCK_PIXELS;

#ifdef USE_SDL_VIDEO
                SDL_UpdateRect(drv->s, 0, y1, VIDEO_MODE_X, height);
#else
                if (VIDEO_DRV_HAVE_SHM)
                        XShmPutImage(x_display, VIDEO_DRV_WINDOW, VIDEO_DRV_GC, VIDEO_DRV_IMAGE, 0, y1, 0, y1, VIDEO_MODE_X, height, 0);
                else
                        XPutImage(x_display, VIDEO_DRV_WINDOW, VIDEO_DRV_GC, VIDEO_DRV_IMAGE, 0, y1, 0, y1, VIDEO_MODE_X, height);
#endif
        }
        mainBuffer.dirty = false;
}


/*
 *      Update display for DGA mode and VOSF
 *      (only in Real or Direct Addressing mode)
 */

#if REAL_ADDRESSING || DIRECT_ADDRESSING
static void update_display_dga_vosf(VIDEO_DRV_DGA_INIT)
{
        VIDEO_MODE_INIT;

        // Compute number of bytes per row, take care to virtual screens
        const int src_bytes_per_row = VIDEO_MODE_ROW_BYTES;
        const int dst_bytes_per_row = TrivialBytesPerRow(VIDEO_MODE_X, DepthModeForPixelDepth(VIDEO_DRV_DEPTH));
        const int scr_bytes_per_row = VIDEO_DRV_ROW_BYTES;
        assert(dst_bytes_per_row <= scr_bytes_per_row);
        const int scr_bytes_left = scr_bytes_per_row - dst_bytes_per_row;

        // Full screen update requested?
        if (mainBuffer.very_dirty) {
                PFLAG_CLEAR_ALL;
                vm_protect((char *)mainBuffer.memStart, mainBuffer.memLength, VM_PAGE_READ);
                memcpy(the_buffer_copy, the_buffer, VIDEO_MODE_ROW_BYTES * VIDEO_MODE_Y);
                VIDEO_DRV_LOCK_PIXELS;
                int i1 = 0, i2 = 0;
                for (int j = 0;  j < VIDEO_MODE_Y; j++) {
                        Screen_blit(the_host_buffer + i2, the_buffer + i1, src_bytes_per_row);
                        i1 += src_bytes_per_row;
                        i2 += scr_bytes_per_row;
                }
#ifdef USE_SDL_VIDEO
                SDL_UpdateRect(drv->s, 0, 0, VIDEO_MODE_X, VIDEO_MODE_Y);
#endif
                VIDEO_DRV_UNLOCK_PIXELS;
                return;
        }

        // Setup partial blitter (use 64-pixel wide chunks)
        const int n_pixels = 64;
        const int n_chunks = VIDEO_MODE_X / n_pixels;
        const int n_pixels_left = VIDEO_MODE_X - (n_chunks * n_pixels);
        const int src_chunk_size = src_bytes_per_row / n_chunks;
        const int dst_chunk_size = dst_bytes_per_row / n_chunks;
        const int src_chunk_size_left = src_bytes_per_row - (n_chunks * src_chunk_size);
        const int dst_chunk_size_left = dst_bytes_per_row - (n_chunks * dst_chunk_size);

        int page = 0, last_scanline = -1;
        for (;;) {
                const unsigned first_page = find_next_page_set(page);
                if (first_page >= mainBuffer.pageCount)
                        break;

                page = find_next_page_clear(first_page);
                PFLAG_CLEAR_RANGE(first_page, page);

                // Make the dirty pages read-only again
                const int32 offset  = first_page << mainBuffer.pageBits;
                const uint32 length = (page - first_page) << mainBuffer.pageBits;
                vm_protect((char *)mainBuffer.memStart + offset, length, VM_PAGE_READ);

                // Optimized for scanlines, don't process overlapping lines again
                int y1 = mainBuffer.pageInfo[first_page].top;
                int y2 = mainBuffer.pageInfo[page - 1].bottom;
                if (y1 <= last_scanline && ++y1 >= VIDEO_MODE_Y)
                        continue;
                if (y2 <= last_scanline && ++y2 >= VIDEO_MODE_Y)
                        continue;
                last_scanline = y2;

                // Update the_host_buffer and copy of the_buffer, one line at a time
                int i1 = y1 * src_bytes_per_row;
                int i2 = y1 * scr_bytes_per_row;
#ifdef USE_SDL_VIDEO
                int bbi = 0;
                SDL_Rect bb[3] = {
                        { VIDEO_MODE_X, y1, 0, 0 },
                        { VIDEO_MODE_X, -1, 0, 0 },
                        { VIDEO_MODE_X, -1, 0, 0 }
                };
#endif
                VIDEO_DRV_LOCK_PIXELS;
                for (int j = y1; j <= y2; j++) {
                        for (int i = 0; i < n_chunks; i++) {
                                if (memcmp(the_buffer_copy + i1, the_buffer + i1, src_chunk_size) != 0) {
                                        memcpy(the_buffer_copy + i1, the_buffer + i1, src_chunk_size);
                                        Screen_blit(the_host_buffer + i2, the_buffer + i1, src_chunk_size);
#ifdef USE_SDL_VIDEO
                                        const int x = i * n_pixels;
                                        if (x < bb[bbi].x) {
                                                if (bb[bbi].w)
                                                        bb[bbi].w += bb[bbi].x - x;
                                                else
                                                        bb[bbi].w = n_pixels;
                                                bb[bbi].x = x;
                                        }
                                        else if (x >= bb[bbi].x + bb[bbi].w)
                                                bb[bbi].w = x + n_pixels - bb[bbi].x;
#endif
                                }
                                i1 += src_chunk_size;
                                i2 += dst_chunk_size;
                        }
                        if (src_chunk_size_left && dst_chunk_size_left) {
                                if (memcmp(the_buffer_copy + i1, the_buffer + i1, src_chunk_size_left) != 0) {
                                        memcpy(the_buffer_copy + i1, the_buffer + i1, src_chunk_size_left);
                                        Screen_blit(the_host_buffer + i2, the_buffer + i1, src_chunk_size_left);
                                }
                                i1 += src_chunk_size_left;
                                i2 += dst_chunk_size_left;
#ifdef USE_SDL_VIDEO
                                const int x = n_chunks * n_pixels;
                                if (x < bb[bbi].x) {
                                        if (bb[bbi].w)
                                                bb[bbi].w += bb[bbi].x - x;
                                        else
                                                bb[bbi].w = n_pixels_left;
                                        bb[bbi].x = x;
                                }
                                else if (x >= bb[bbi].x + bb[bbi].w)
                                        bb[bbi].w  = x + n_pixels_left - bb[bbi].x;
#endif
                        }
                        i2 += scr_bytes_left;
#ifdef USE_SDL_VIDEO
                        bb[bbi].h++;
                        if (bb[bbi].w && (j == y1 || j == y2 - 1 || j == y2)) {
                                bbi++;
                                assert(bbi <= 3);
                                if (j != y2)
                                        bb[bbi].y = j + 1;
                        }
#endif
                }
#ifdef USE_SDL_VIDEO
                SDL_UpdateRects(drv->s, bbi, bb);
#endif
                VIDEO_DRV_UNLOCK_PIXELS;
        }
        mainBuffer.dirty = false;
}
#endif

#endif /* ENABLE_VOSF */

#endif /* VIDEO_VOSF_H */
Revision:	1.54
Committed:	2005-06-25T11:40:29Z (19 years, 4 months ago) by gbeauche
Content type:	text/plain
Branch:	MAIN
Changes since 1.53:	+8 -8 lines
Log Message:	Use fast spinlocks only for small enough atomic operations. Otherwise, you run into some performance problems in e.g. video graphics experience because of busywaits in the current spin_lock() implementation.