/* grab-ximage.c --- grab the screen to an XImage for use with OpenGL. * xscreensaver, Copyright (c) 2001-2021 Jamie Zawinski * * Modified by Richard Weeks Copyright (c) 2020 * * Permission to use, copy, modify, distribute, and sell this software and its * documentation for any purpose is hereby granted without fee, provided that * the above copyright notice appear in all copies and that both that * copyright notice and this permission notice appear in supporting * documentation. No representations are made about the suitability of this * software for any purpose. It is provided "as is" without express or * implied warranty. */ #include "xlockmoreI.h" #include "grab-ximage.h" #include "grabscreen.h" #include "pow2.h" #include "visual.h" #include "xshm.h" #include #include #include #include /* If REFORMAT_IMAGE_DATA is defined, then we convert Pixmaps to textures like this: - get Pixmap as an XImage in whatever form the server hands us; - convert that XImage to 32-bit RGBA in client-local endianness; - make the texture using RGBA, UNSIGNED_BYTE. If undefined, we do this: - get Pixmap as an XImage in whatever form the server hands us; - figure out what OpenGL texture packing parameters correspond to the image data that the server sent us and use that, e.g., BGRA, INT_8_8_8_8_REV. You might expect the second method to be faster, since we're not making a second copy of the data and iterating each pixel before we hand it to GL. But, you'd be wrong. The first method is almost 6x faster. I guess GL is reformatting it *again*, and doing it very inefficiently! */ #define REFORMAT_IMAGE_DATA #undef MAX #define MAX(a,b) ((a)>(b)?(a):(b)) static int debug_p = 0; static double perf = 32768000.0; /* initially assume 2^19 pixels in 1/60/sec */ static Bool bigendian (void) { union { int i; char c[sizeof(int)]; } u; u.i = 1; return !u.c[0]; } #if defined(HAVE_IPHONE) || defined(HAVE_ANDROID) # undef GENERATE_MIPMAPS #else # define GENERATE_MIPMAPS #endif #ifdef REFORMAT_IMAGE_DATA /* Given a bitmask, returns the position and width of the field. */ static void decode_mask (unsigned long mask, unsigned long *pos_ret, unsigned long *size_ret) { int i; for (i = 0; i < 32; i++) if (mask & (1L << i)) { int j = 0; *pos_ret = i; for (; i < 32; i++, j++) if (! (mask & (1L << i))) break; *size_ret = j; return; } } /* Given a value and a field-width, expands the field to fill out 8 bits. */ static unsigned char spread_bits (unsigned char value, unsigned char width) { switch (width) { case 8: return value; case 7: return (value << 1) | (value >> 6); case 6: return (value << 2) | (value >> 4); case 5: return (value << 3) | (value >> 2); case 4: return (value << 4) | (value); case 3: return (value << 5) | (value << 2) | (value >> 2); case 2: return (value << 6) | (value << 4) | (value); default: abort(); break; } } static XImage * convert_ximage_to_rgba32 (Screen *screen, XImage *image) { Display *dpy = DisplayOfScreen (screen); Visual *visual = DefaultVisualOfScreen (screen); int x, y; unsigned long crpos=0, cgpos=0, cbpos=0, capos=0; /* bitfield positions */ unsigned long srpos=0, sgpos=0, sbpos=0; unsigned long srmsk=0, sgmsk=0, sbmsk=0; unsigned long srsiz=0, sgsiz=0, sbsiz=0; XColor *colors = 0; unsigned char spread_map[3][256]; /* Note: height+2 in "to" to work around an array bounds overrun in gluBuild2DMipmaps / gluScaleImage. */ XImage *from = image; XImage *to = XCreateImage (dpy, visual, 32, /* depth */ ZPixmap, 0, 0, from->width, from->height, 32, /* bitmap pad */ 0); to->data = (char *) calloc (to->height + 2, to->bytes_per_line); /* Set the bit order in the XImage structure to whatever the local host's native bit order is. */ to->bitmap_bit_order = to->byte_order = (bigendian() ? MSBFirst : LSBFirst); if (visual_class (screen, visual) == PseudoColor || visual_class (screen, visual) == GrayScale) { Colormap cmap = DefaultColormapOfScreen (screen); int ncolors = visual_cells (screen, visual); int i; colors = (XColor *) calloc (sizeof (*colors), ncolors+1); for (i = 0; i < ncolors; i++) colors[i].pixel = i; XQueryColors (dpy, cmap, colors, ncolors); } if (colors == 0) /* truecolor */ { srmsk = to->red_mask; sgmsk = to->green_mask; sbmsk = to->blue_mask; decode_mask (srmsk, &srpos, &srsiz); decode_mask (sgmsk, &sgpos, &sgsiz); decode_mask (sbmsk, &sbpos, &sbsiz); } /* Pack things in "RGBA" order in client endianness. */ if (bigendian()) crpos = 24, cgpos = 16, cbpos = 8, capos = 0; else crpos = 0, cgpos = 8, cbpos = 16, capos = 24; if (colors == 0) /* truecolor */ { int i; for (i = 0; i < 256; i++) { spread_map[0][i] = spread_bits (i, srsiz); spread_map[1][i] = spread_bits (i, sgsiz); spread_map[2][i] = spread_bits (i, sbsiz); } } /* trying to track down an intermittent crash in ximage_putpixel_32 */ if (to->width < from->width) abort(); if (to->height < from->height) abort(); for (y = 0; y < from->height; y++) for (x = 0; x < from->width; x++) { unsigned long sp = XGetPixel (from, x, y); unsigned char sr, sg, sb; unsigned long cp; if (colors) { sr = colors[sp].red & 0xFF; sg = colors[sp].green & 0xFF; sb = colors[sp].blue & 0xFF; } else { sr = (sp & srmsk) >> srpos; sg = (sp & sgmsk) >> sgpos; sb = (sp & sbmsk) >> sbpos; sr = spread_map[0][sr]; sg = spread_map[1][sg]; sb = spread_map[2][sb]; } cp = ((sr << crpos) | (sg << cgpos) | (sb << cbpos) | (0xFF << capos)); XPutPixel (to, x, y, cp); } if (colors) free (colors); return to; } #endif /* REFORMAT_IMAGE_DATA */ /* Shrinks the XImage by a factor of two. We use this when mipmapping fails on large textures. */ static void halve_image (XImage *ximage, XRectangle *geom) { int w2 = ximage->width/2; int h2 = ximage->height/2; int x, y; XImage *ximage2; if (w2 <= 32 || h2 <= 32) /* let's not go crazy here, man. */ return; if (debug_p) fprintf (stderr, "%s: shrinking image %dx%d -> %dx%d\n", progname, ximage->width, ximage->height, w2, h2); ximage2 = (XImage *) calloc (1, sizeof (*ximage2)); *ximage2 = *ximage; ximage2->width = w2; ximage2->height = h2; ximage2->bytes_per_line = 0; ximage2->data = 0; XInitImage (ximage2); ximage2->data = (char *) calloc (h2, ximage2->bytes_per_line); if (!ximage2->data) { fprintf (stderr, "%s: out of memory (scaling %dx%d image to %dx%d)\n", progname, ximage->width, ximage->height, w2, h2); exit (1); } for (y = 0; y < h2; y++) for (x = 0; x < w2; x++) XPutPixel (ximage2, x, y, XGetPixel (ximage, x*2, y*2)); free (ximage->data); *ximage = *ximage2; ximage2->data = 0; XFree (ximage2); if (geom) { geom->x /= 2; geom->y /= 2; geom->width /= 2; geom->height /= 2; } } #ifdef REFORMAT_IMAGE_DATA /* Pulls the Pixmap bits from the server and returns an XImage in some format acceptable to OpenGL. */ static XImage * pixmap_to_gl_ximage (Screen *screen, Window window, Pixmap pixmap) { Display *dpy = DisplayOfScreen (screen); Visual *visual = DefaultVisualOfScreen (screen); unsigned int width, height, depth; XShmSegmentInfo shm_info; XImage *server_ximage = 0; XImage *client_ximage = 0; { Window root; int x, y; unsigned int bw; XGetGeometry (dpy, pixmap, &root, &x, &y, &width, &height, &bw, &depth); } if (width < 5 || height < 5) /* something's gone wrong somewhere... */ return 0; /* Convert the server-side Pixmap to a client-side GL-ordered XImage. */ server_ximage = create_xshm_image (dpy, visual, depth, ZPixmap, &shm_info, width, height); get_xshm_image (dpy, pixmap, server_ximage, 0, 0, ~0L, &shm_info); client_ximage = convert_ximage_to_rgba32 (screen, server_ximage); destroy_xshm_image (dpy, server_ximage, &shm_info); return client_ximage; } # else /* ! REFORMAT_IMAGE_DATA */ typedef struct { unsigned int depth, red_mask, green_mask, blue_mask; /* when this... */ GLint type, format; /* ...use this. */ } conversion_table; /* Abbreviate these so that the table entries all fit on one line... */ #define BYTE GL_UNSIGNED_BYTE #define BYTE_2_3_3_REV GL_UNSIGNED_BYTE_2_3_3_REV #define BYTE_3_3_2 GL_UNSIGNED_BYTE_3_3_2 #define INT_10_10_10_2 GL_UNSIGNED_INT_10_10_10_2 #define INT_2_10_10_10_REV GL_UNSIGNED_INT_2_10_10_10_REV #define INT_8_8_8_8 GL_UNSIGNED_INT_8_8_8_8 #define INT_8_8_8_8_REV GL_UNSIGNED_INT_8_8_8_8_REV #define SHORT_1_5_5_5_REV GL_UNSIGNED_SHORT_1_5_5_5_REV #define SHORT_4_4_4_4 GL_UNSIGNED_SHORT_4_4_4_4 #define SHORT_4_4_4_4_REV GL_UNSIGNED_SHORT_4_4_4_4_REV #define SHORT_5_5_5_1 GL_UNSIGNED_SHORT_5_5_5_1 #define SHORT_5_6_5 GL_UNSIGNED_SHORT_5_6_5 #define SHORT_5_6_5_REV GL_UNSIGNED_SHORT_5_6_5_REV static const conversion_table ctable[] = { { 8, 0x000000E0, 0x0000001C, 0x00000003, BYTE_3_3_2, GL_RGB }, { 8, 0x00000007, 0x00000038, 0x000000C0, BYTE_2_3_3_REV, GL_RGB }, { 16, 0x0000F800, 0x000007E0, 0x0000001F, SHORT_5_6_5, GL_RGB }, { 16, 0x0000001F, 0x000007E0, 0x0000F800, SHORT_5_6_5_REV, GL_RGB }, { 16, 0x0000F000, 0x00000F00, 0x000000F0, SHORT_4_4_4_4, GL_RGBA }, { 16, 0x000000F0, 0x00000F00, 0x0000F000, SHORT_4_4_4_4, GL_BGRA }, { 16, 0x0000000F, 0x000000F0, 0x00000F00, SHORT_4_4_4_4, GL_ABGR_EXT }, { 16, 0x0000000F, 0x000000F0, 0x00000F00, SHORT_4_4_4_4_REV, GL_RGBA }, { 16, 0x00000F00, 0x000000F0, 0x0000000F, SHORT_4_4_4_4_REV, GL_BGRA }, { 16, 0x0000F800, 0x000007C0, 0x0000003E, SHORT_5_5_5_1, GL_RGBA }, { 16, 0x0000003E, 0x000007C0, 0x0000F800, SHORT_5_5_5_1, GL_BGRA }, { 16, 0x00000001, 0x0000003E, 0x000007C0, SHORT_5_5_5_1, GL_ABGR_EXT }, { 16, 0x0000001F, 0x000003E0, 0x00007C00, SHORT_1_5_5_5_REV, GL_RGBA }, { 16, 0x00007C00, 0x000003E0, 0x0000001F, SHORT_1_5_5_5_REV, GL_BGRA }, { 32, 0xFF000000, 0x00FF0000, 0x0000FF00, INT_8_8_8_8, GL_RGBA }, { 32, 0x0000FF00, 0x00FF0000, 0xFF000000, INT_8_8_8_8, GL_BGRA }, { 32, 0x000000FF, 0x0000FF00, 0x00FF0000, INT_8_8_8_8, GL_ABGR_EXT }, { 32, 0x000000FF, 0x0000FF00, 0x00FF0000, INT_8_8_8_8_REV, GL_RGBA }, { 32, 0x00FF0000, 0x0000FF00, 0x000000FF, INT_8_8_8_8_REV, GL_BGRA }, { 32, 0xFFC00000, 0x003FF000, 0x00000FFC, INT_10_10_10_2, GL_RGBA }, { 32, 0x00000FFC, 0x003FF000, 0xFFC00000, INT_10_10_10_2, GL_BGRA }, { 32, 0x00000003, 0x00000FFC, 0x003FF000, INT_10_10_10_2, GL_ABGR_EXT }, { 32, 0x000003FF, 0x000FFC00, 0x3FF00000, INT_2_10_10_10_REV, GL_RGBA }, { 32, 0x3FF00000, 0x000FFC00, 0x000003FF, INT_2_10_10_10_REV, GL_BGRA }, { 24, 0x000000FF, 0x0000FF00, 0x00FF0000, BYTE, GL_RGB }, { 24, 0x00FF0000, 0x0000FF00, 0x000000FF, BYTE, GL_BGR }, }; /* Given an XImage, returns the GL settings to use its data as a texture. */ static void gl_settings_for_ximage (XImage *image, GLint *type_ret, GLint *format_ret, GLint *swap_ret) { int i; for (i = 0; i < countof(ctable); ++i) { if (image->bits_per_pixel == ctable[i].depth && image->red_mask == ctable[i].red_mask && image->green_mask == ctable[i].green_mask && image->blue_mask == ctable[i].blue_mask) { *type_ret = ctable[i].type; *format_ret = ctable[i].format; if (image->bits_per_pixel == 24) { /* don't know how to test this... */ *type_ret = (ctable[i].type == GL_RGB) ? GL_BGR : GL_RGB; *swap_ret = 0; } else { *swap_ret = !!(image->byte_order == MSBFirst) ^ !!bigendian(); } if (debug_p) { fprintf (stderr, "%s: using %s %s %d for %d %08lX %08lX %08lX\n", progname, (*format_ret == GL_RGB ? "RGB" : *format_ret == GL_BGR ? "BGR" : *format_ret == GL_RGBA ? "RGBA" : *format_ret == GL_BGRA ? "BGRA" : *format_ret == GL_ABGR_EXT ? "ABGR_EXT" : "???"), (*type_ret == BYTE ? "BYTE" : *type_ret == BYTE_3_3_2 ? "BYTE_3_3_2" : *type_ret == BYTE_2_3_3_REV ? "BYTE_2_3_3_REV" : *type_ret == INT_8_8_8_8 ? "INT_8_8_8_8" : *type_ret == INT_8_8_8_8_REV ? "INT_8_8_8_8_REV" : *type_ret == INT_10_10_10_2 ? "INT_10_10_10_2" : *type_ret == INT_2_10_10_10_REV ? "INT_2_10_10_10_REV": *type_ret == SHORT_4_4_4_4 ? "SHORT_4_4_4_4" : *type_ret == SHORT_4_4_4_4_REV ? "SHORT_4_4_4_4_REV" : *type_ret == SHORT_5_5_5_1 ? "SHORT_5_5_5_1" : *type_ret == SHORT_1_5_5_5_REV ? "SHORT_1_5_5_5_REV" : *type_ret == SHORT_5_6_5 ? "SHORT_5_6_5" : *type_ret == SHORT_5_6_5_REV ? "SHORT_5_6_5_REV" : "???"), *swap_ret, image->bits_per_pixel, image->red_mask, image->green_mask, image->blue_mask); } return; } } /* Unknown RGB fields? */ abort(); } #endif /* ! REFORMAT_IMAGE_DATA */ typedef struct { GLXContext glx_context; Pixmap pixmap; int pix_width, pix_height, pix_depth; int texid; Bool mipmap_p; double load_time; /* Used in async mode */ void (*callback) (const char *filename, XRectangle *geometry, int iw, int ih, int tw, int th, void *closure); void *closure; /* Used in sync mode */ char **filename_return; XRectangle *geometry_return; int *image_width_return; int *image_height_return; int *texture_width_return; int *texture_height_return; } img_closure; typedef enum { TLP_LOADING = 0, TLP_IMPORTING, TLP_COMPLETE, TLP_ERROR } texture_loader_phase; struct texture_loader_t { texture_loader_phase phase; Screen *screen; Window window; XShmSegmentInfo shm_info; XImage *ximage; img_closure load_closure; Bool pixmap_valid_p; int img_width, img_height, tex_width, tex_height; XRectangle geometry; int y; unsigned int stripe_height; char *name; /* debugging */ int steps; /* number of calls to step_texture_loader() that loaded part of the texture */ int stripes; /* number of stripes put into the texture so far */ double loaded_time, work_seconds; }; /* Returns the current time in seconds as a double. */ static double double_time (void) { struct timeval now; # ifdef GETTIMEOFDAY_TWO_ARGS struct timezone tzp; gettimeofday(&now, &tzp); # else gettimeofday(&now); # endif return (now.tv_sec + ((double) now.tv_usec * 0.000001)); } /* Loads the given XImage into GL's texture memory. The image may be of any size. If mipmap_p is true, then make mipmaps instead of just a single texture. Writes to stderr and returns False on error. */ static Bool ximage_to_texture (XImage *ximage, GLint type, GLint format, int *width_return, int *height_return, XRectangle *geometry, Bool mipmap_p) { int max_reduction = 7; int err_count = 0; GLenum err = 0; int orig_width = ximage->width; int orig_height = ximage->height; GLsizei tex_width = 0; GLsizei tex_height = 0; AGAIN: if (mipmap_p) { /* gluBuild2DMipmaps doesn't require textures to be a power of 2. */ tex_width = ximage->width; tex_height = ximage->height; if (debug_p) fprintf (stderr, "%s: mipmap %d x %d\n", progname, ximage->width, ximage->height); gluBuild2DMipmaps (GL_TEXTURE_2D, 3, ximage->width, ximage->height, format, type, ximage->data); err = glGetError(); } else { /* glTexImage2D() requires the texture sizes to be powers of 2. So first, create a texture of that size (but don't write any data into it.) */ tex_width = (GLsizei) to_pow2 (ximage->width); tex_height = (GLsizei) to_pow2 (ximage->height); if (debug_p) fprintf (stderr, "%s: texture %d x %d (%d x %d)\n", progname, ximage->width, ximage->height, tex_width, tex_height); glTexImage2D (GL_TEXTURE_2D, 0, GL_RGBA, tex_width, tex_height, 0, format, type, 0); err = glGetError(); /* Now load our non-power-of-2 image data into the existing texture. */ if (!err) { glTexSubImage2D (GL_TEXTURE_2D, 0, 0, 0, ximage->width, ximage->height, GL_RGBA, GL_UNSIGNED_BYTE, ximage->data); err = glGetError(); } } if (err) { char buf[100]; const char *s = (char *) gluErrorString (err); if (!s || !*s) { sprintf (buf, "unknown error %d", (int) err); s = buf; } while (glGetError() != GL_NO_ERROR) ; /* clear any lingering errors */ if (++err_count > max_reduction) { fprintf (stderr, "\n" "%s: %dx%d texture failed, even after reducing to %dx%d:\n" "%s: The error was: \"%s\".\n" "%s: probably this means " "\"your video card is worthless and weak\"?\n\n", progname, orig_width, orig_height, ximage->width, ximage->height, progname, s, progname); return False; } else { if (debug_p) fprintf (stderr, "%s: mipmap error (%dx%d): %s\n", progname, ximage->width, ximage->height, s); halve_image (ximage, geometry); goto AGAIN; } } if (width_return) *width_return = tex_width; if (height_return) *height_return = tex_height; return True; } static void load_texture_async_cb (Screen *screen, Window window, Drawable drawable, const char *name, XRectangle *geometry, void *closure); /* Grabs an image of the desktop (or another random image file) and loads the image into GL's texture memory. When the callback is called, the image data will have been loaded into texture number `texid' (via glBindTexture.) If an error occurred, width/height will be 0. */ void load_texture_async (Screen *screen, Window window, GLXContext glx_context, int desired_width, int desired_height, Bool mipmap_p, GLuint texid, void (*callback) (const char *filename, XRectangle *geometry, int image_width, int image_height, int texture_width, int texture_height, void *closure), void *closure) { Display *dpy = DisplayOfScreen (screen); XWindowAttributes xgwa; img_closure *data = (img_closure *) calloc (1, sizeof(*data)); if (debug_p) data->load_time = double_time(); data->texid = texid; data->mipmap_p = mipmap_p; data->glx_context = glx_context; data->callback = callback; data->closure = closure; XGetWindowAttributes (dpy, window, &xgwa); data->pix_width = xgwa.width; data->pix_height = xgwa.height; data->pix_depth = xgwa.depth; /* Allow the pixmap to be larger than the window. Esper wants this. */ if (desired_width /* && desired_width < xgwa.width */) data->pix_width = desired_width; if (desired_height /* && desired_height < xgwa.height */) data->pix_height = desired_height; data->pixmap = XCreatePixmap (dpy, window, data->pix_width, data->pix_height, data->pix_depth); load_image_async (screen, window, data->pixmap, load_texture_async_cb, data); } static void incremental_load_texture_async_cb (Screen *screen, Window window, Drawable drawable, const char *name, XRectangle *geometry, void *closure); /* Allocate a texture loader to grab the image of a Window and load the image into GL's texture memory. The returned texture loader should be stepped with step_texture_loader until that function calls the callback function passed to it. */ texture_loader_t* alloc_texture_loader (Screen *screen, Window window, GLXContext glx_context, int desired_width, int desired_height, Bool mipmap_p, GLuint texid) { texture_loader_t *loader = (texture_loader_t*) calloc (1, sizeof(*loader)); Display *dpy = DisplayOfScreen (screen); XWindowAttributes xgwa; img_closure *data = &loader->load_closure; if (debug_p) data->load_time = double_time(); data->texid = texid; data->mipmap_p = mipmap_p; data->glx_context = glx_context; loader->screen = screen; loader->window = window; XGetWindowAttributes (dpy, window, &xgwa); data->pix_width = xgwa.width; data->pix_height = xgwa.height; data->pix_depth = xgwa.depth; /* Allow the pixmap to be larger than the window. Esper wants this. */ if (desired_width /* && desired_width < xgwa.width */) data->pix_width = desired_width; if (desired_height /* && desired_height < xgwa.height */) data->pix_height = desired_height; data->pixmap = XCreatePixmap (dpy, window, data->pix_width, data->pix_height, data->pix_depth); loader->pixmap_valid_p = True; load_image_async (screen, window, data->pixmap, incremental_load_texture_async_cb, loader); return loader; } /* Free the texture loader */ void free_texture_loader (texture_loader_t *loader) { Display *dpy = DisplayOfScreen(loader->screen); /* If the loader is still awaiting asynchronous completion of loading the XImage, there is no way to recover from this request. */ if (loader->phase == TLP_LOADING) abort(); if (loader->ximage) { XImage *ximage = loader->ximage; loader->ximage = 0; destroy_xshm_image (dpy, ximage, &loader->shm_info); } if (loader->pixmap_valid_p) { loader->pixmap_valid_p = False; XFreePixmap (dpy, loader->load_closure.pixmap); } free ((char *) loader->name); loader->name = 0; free (loader); } /* Once we have an XImage, this loads it into GL. This is used in both synchronous and asynchronous mode. */ static void load_texture_async_cb (Screen *screen, Window window, Drawable drawable, const char *name, XRectangle *geometry, void *closure) { Display *dpy = DisplayOfScreen (screen); Bool ok; XImage *ximage; GLint type, format; int iw=0, ih=0, tw=0, th=0; double cvt_time=0, tex_time=0, done_time=0; img_closure *data = (img_closure *) closure; /* copy closure data to stack and free the original before running cb */ img_closure dd = *data; memset (data, 0, sizeof (*data)); free (data); data = 0; if (dd.glx_context) glXMakeCurrent (dpy, window, dd.glx_context); if (geometry->width <= 0 || geometry->height <= 0) { /* This can happen if an old version of xscreensaver-getimage is installed. Or if we have no image (checkerboard). */ geometry->x = 0; geometry->y = 0; geometry->width = dd.pix_width; geometry->height = dd.pix_height; } if (geometry->width <= 0 || geometry->height <= 0) abort(); if (debug_p) cvt_time = double_time(); # ifdef REFORMAT_IMAGE_DATA ximage = pixmap_to_gl_ximage (screen, window, dd.pixmap); format = GL_RGBA; type = GL_UNSIGNED_BYTE; #else /* ! REFORMAT_IMAGE_DATA */ { Visual *visual = DefaultVisualOfScreen (screen); GLint swap; ximage = XCreateImage (dpy, visual, dd.pix_depth, ZPixmap, 0, 0, dd.pix_width, dd.pix_height, 32, 0); /* Note: height+2 in "to" to be to work around an array bounds overrun in gluBuild2DMipmaps / gluScaleImage. */ ximage->data = (char *) calloc (ximage->height+2, ximage->bytes_per_line); if (!ximage->data || !XGetSubImage (dpy, dd.pixmap, 0, 0, ximage->width, ximage->height, ~0L, ximage->format, ximage, 0, 0)) { XDestroyImage (ximage); ximage = 0; } gl_settings_for_ximage (ximage, &type, &format, &swap); glPixelStorei (GL_UNPACK_SWAP_BYTES, !swap); } #endif /* REFORMAT_IMAGE_DATA */ XFreePixmap (dpy, dd.pixmap); dd.pixmap = 0; if (debug_p) tex_time = double_time(); if (! ximage) ok = False; else { iw = ximage->width; ih = ximage->height; if (dd.texid != -1) glBindTexture (GL_TEXTURE_2D, dd.texid); glPixelStorei (GL_UNPACK_ALIGNMENT, ximage->bitmap_pad / 8); ok = ximage_to_texture (ximage, type, format, &tw, &th, geometry, dd.mipmap_p); if (ok) { iw = ximage->width; /* in case the image was shrunk */ ih = ximage->height; } } if (ximage) XDestroyImage (ximage); if (! ok) iw = ih = tw = th = 0; if (debug_p) done_time = double_time(); if (debug_p) fprintf (stderr, /* prints: A + B + C = D A = file I/O time (happens in background) B = time to pull bits from server (this process) C = time to convert bits to GL textures (this process) D = total elapsed time from "want image" to "see image" B+C is responsible for any frame-rate glitches. */ "%s: loading elapsed: %.2f + %.2f + %.2f = %.2f sec\n", progname, cvt_time - dd.load_time, tex_time - cvt_time, done_time - tex_time, done_time - dd.load_time); if (dd.callback) /* asynchronous mode */ dd.callback (name, geometry, iw, ih, tw, th, dd.closure); else { /* synchronous mode */ if (dd.filename_return) *dd.filename_return = (char *) name; if (dd.geometry_return) *dd.geometry_return = *geometry; if (dd.image_width_return) *dd.image_width_return = iw; if (dd.image_height_return) *dd.image_height_return = ih; if (dd.texture_width_return) *dd.texture_width_return = tw; if (dd.texture_height_return) *dd.texture_height_return = th; } } /* Once we have a pixmap, this sets us up to step-load it into a GL texture. */ static void incremental_load_texture_async_cb (Screen *screen, Window window, Drawable drawable, const char *name, XRectangle *geometry, void *closure) { texture_loader_t *loader = (texture_loader_t *) closure; Display *dpy = DisplayOfScreen (screen); img_closure dd = loader->load_closure; GLenum err = 0; GLsizei tex_width = 0, tex_height = 0; loader->load_closure.texid = dd.texid; loader->phase = TLP_IMPORTING; if (debug_p) loader->loaded_time = double_time(); /* Like load_texture_async_cb() until it calls pixmap_to_gl_ximage() */ if (dd.glx_context) glXMakeCurrent (dpy, window, dd.glx_context); if (geometry->width <= 0 || geometry->height <= 0) { geometry->x = 0; geometry->y = 0; geometry->width = dd.pix_width; geometry->height = dd.pix_height; } if (geometry->width <= 0 || geometry->height <= 0) abort(); loader->geometry = *geometry; /* Like pixmap_to_gl_ximage(): */ { Visual *visual = DefaultVisualOfScreen (screen); unsigned int width, height, depth; { Window root; int x, y; unsigned int bw; XGetGeometry (dpy, dd.pixmap, &root, &x, &y, &width, &height, &bw, &depth); } if (width < 5 || height < 5) { loader->phase = TLP_ERROR; return; } loader->ximage = create_xshm_image (dpy, visual, depth, ZPixmap, &loader->shm_info, width, height); get_xshm_image (dpy, dd.pixmap, loader->ximage, 0, 0, ~0L, &loader->shm_info); } loader->img_width = loader->ximage->width; loader->img_height = loader->ximage->height; loader->stripe_height = (1 << 19) / loader->img_width; if (dd.texid != -1) glBindTexture (GL_TEXTURE_2D, dd.texid); /* as much of ximage_to_texture() functionality as we can precompute */ tex_width = (GLsizei) to_pow2 (loader->ximage->width); tex_height = (GLsizei) to_pow2 (loader->ximage->height); /* glTexImage2D() to allocate OpenGL texture */ glTexImage2D (GL_TEXTURE_2D, 0, GL_RGBA, tex_width, tex_height, 0, GL_RGBA, GL_UNSIGNED_BYTE, 0); err = glGetError(); if (err) { loader->phase = TLP_ERROR; loader->img_width = loader->img_height = 0; return; } /* Capture texture dimensions and name in loader */ loader->tex_width = tex_width; loader->tex_height = tex_height; loader->name = name ? strdup(name) : 0; } static void advance_texture_loader (texture_loader_t *loader, double allowed_seconds); static void complete_texture_load (texture_loader_t *loader, void (*callback) (const char *filename, XRectangle *geometry, int image_width, int image_height, int texture_width, int texture_height, void *closure), void *closure); void step_texture_loader (texture_loader_t *loader, double allowed_seconds, void (*callback) (const char *filename, XRectangle *geometry, int image_width, int image_height, int texture_width, int texture_height, void *closure), void *closure) { if (loader->phase != TLP_IMPORTING) return; if (loader->y < loader->ximage->height) { loader->steps++; if (loader->steps == 1) /* Initial tune on the loader for the number of allowed_seconds */ loader->stripe_height = ((unsigned int) (perf * allowed_seconds / loader->ximage->width)) / 8 + 1; else advance_texture_loader (loader, allowed_seconds); } else { loader->phase = TLP_COMPLETE; complete_texture_load (loader, callback, closure); } } static unsigned int texture_loader_next_stripe_height (texture_loader_t *loader) { if (loader->y + loader->stripe_height <= loader->img_height) return loader->stripe_height; else return loader->img_height - loader->y; } static void advance_texture_loader (texture_loader_t *loader, double allowed_seconds) { Display *dpy = loader->screen ? DisplayOfScreen (loader->screen) : 0; double start_time = double_time(), elapsed_seconds = 0; double step_end = start_time + allowed_seconds; int iter_count = 0; unsigned int lines_processed = 0; if (loader->load_closure.glx_context) glXMakeCurrent (dpy, loader->window, loader->load_closure.glx_context); for ( ; (double_time() < step_end) && (loader->y < loader->img_height); loader->y += loader->stripe_height, ++iter_count ) { /* * Use XSubImage() to extract a section of the shared image loader->stripe_height tall * Use convert_ximage_to_rgba32() to convert that to the desired format * Use XDestroyImage() to free the extracted section of the shared image * glBindTexture(GL_TEXTURE_2D, loader->load_closure.texid) * Import the data to the next stripe of the texture with glTexSubImage2D() * XDestroyImage() to destroy the converted image * Increment loader->y by loader->stripe_height */ unsigned int patch_height = texture_loader_next_stripe_height (loader); XImage* patch = XSubImage (loader->ximage, 0, loader->y, loader->img_width, patch_height); XImage* cvt_patch = convert_ximage_to_rgba32 (loader->screen, patch); Bool use_old_mipmap_p = False; # ifdef GENERATE_MIPMAPS use_old_mipmap_p = (loader->load_closure.mipmap_p && (loader->y + loader->stripe_height >= loader->ximage->height)); # endif loader->stripes++; XDestroyImage (patch); glBindTexture (GL_TEXTURE_2D, loader->load_closure.texid); glPixelStorei (GL_UNPACK_ALIGNMENT, cvt_patch->bitmap_pad / 8); if (use_old_mipmap_p) /* Use old GL_GENERATE_MIPMAP (if before OpenGL 3.0) */ glTexParameteri (GL_TEXTURE_2D, GL_GENERATE_MIPMAP, GL_TRUE); /* CHECK: loader->y or -loader->y? */ glTexSubImage2D (GL_TEXTURE_2D, 0, 0, loader->y, cvt_patch->width, cvt_patch->height, GL_RGBA, GL_UNSIGNED_BYTE, cvt_patch->data); if (use_old_mipmap_p) /* Turn off GL_GENERATE_MIPMAP if we turned it on */ glTexParameteri (GL_TEXTURE_2D, GL_GENERATE_MIPMAP, GL_FALSE); XDestroyImage (cvt_patch); lines_processed += patch_height; } if (iter_count == 1 && loader->y < loader->img_height && loader->stripe_height > 1) { loader->stripe_height >>= 1; } elapsed_seconds = double_time() - start_time; /* monitor perf for use setting the next initial stripe_height (px/s) */ if (elapsed_seconds > 0.001) { perf = lines_processed * (double) loader->ximage->width / elapsed_seconds; if (perf > 2e8) perf = 2e8; else if (perf < 1e6) perf = 1e6; } if (debug_p) loader->work_seconds += elapsed_seconds; } static void complete_texture_load (texture_loader_t *loader, void (*callback) (const char *filename, XRectangle *geometry, int image_width, int image_height, int texture_width, int texture_height, void *closure), void *closure) { XImage *ximage = loader->ximage; Display *dpy = loader->screen ? DisplayOfScreen (loader->screen) : 0; char *name = loader->name; XRectangle geometry = loader->geometry; loader->name = 0; if (loader->load_closure.glx_context) glXMakeCurrent (dpy, loader->window, loader->load_closure.glx_context); loader->ximage = 0; destroy_xshm_image (dpy, ximage, &loader->shm_info); loader->pixmap_valid_p = False; XFreePixmap (dpy, loader->load_closure.pixmap); if (loader->load_closure.mipmap_p) { glBindTexture (GL_TEXTURE_2D, loader->load_closure.texid); # ifdef GENERATE_MIPMAPS glGenerateMipmap (GL_TEXTURE_2D); # endif } if (debug_p) { double done_time = double_time(); fprintf (stderr, "%s: texture loading: [load %.2f sec] [import %.2f sec, %d stripes, %d steps] [elapsed %2.f sec]\n", progname, loader->loaded_time - loader->load_closure.load_time, loader->work_seconds, loader->stripes, loader->steps, done_time - loader->load_closure.load_time); } callback (name, &geometry, loader->img_width, loader->img_height, loader->tex_width, loader->tex_height, closure); free (name); } Bool texture_loader_failed (texture_loader_t *loader) { return loader->phase == TLP_ERROR; }