glxext.c   [plain text]

/*
 * SGI FREE SOFTWARE LICENSE B (Version 2.0, Sept. 18, 2008)
 * Copyright (C) 1991-2000 Silicon Graphics, Inc. All Rights Reserved.
 * Copyright (C) 2009 Apple Inc.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice including the dates of first publication and
 * either this permission notice or a reference to
 * http://oss.sgi.com/projects/FreeB/
 * shall be included in all copies or substantial portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
 * SILICON GRAPHICS, INC. BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
 * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF
 * OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 *
 * Except as contained in this notice, the name of Silicon Graphics, Inc.
 * shall not be used in advertising or otherwise to promote the sale, use or
 * other dealings in this Software without prior written authorization from
 * Silicon Graphics, Inc.
 */

/**
 * \file glxext.c
 * GLX protocol interface boot-strap code.
 *
 * Direct rendering support added by Precision Insight, Inc.
 *
 * \author Kevin E. Martin <kevin@precisioninsight.com>
 */     

#include <assert.h>
#include "glxclient.h"
#include <X11/extensions/Xext.h>
#include <X11/extensions/extutil.h>
//#include "glapi.h"
#include "glxextensions.h"
#include "glcontextmodes.h"
//#include "glheader.h"

#include "apple_glx.h"
#include "apple_visual.h"

#ifdef USE_XCB
#include <X11/Xlib-xcb.h>
#include <xcb/xcb.h>
#include <xcb/glx.h>
#endif


#ifdef DEBUG
void __glXDumpDrawBuffer(__GLXcontext *ctx);
#endif

#ifdef USE_SPARC_ASM
static void _glx_mesa_init_sparc_glapi_relocs(void);
static int _mesa_sparc_needs_init = 1;
#define INIT_MESA_SPARC { \
    if(_mesa_sparc_needs_init) { \
      _glx_mesa_init_sparc_glapi_relocs(); \
      _mesa_sparc_needs_init = 0; \
  } \
}
#else
#define INIT_MESA_SPARC
#endif

/*
** You can set this cell to 1 to force the gl drawing stuff to be
** one command per packet
*/
_X_HIDDEN int __glXDebug = 0;

/* Extension required boiler plate */

static char *__glXExtensionName = GLX_EXTENSION_NAME;

static XExtensionInfo glxext_info_data;
static XExtensionInfo *glXExtensionInfo = &glxext_info_data;

static /* const */ char *error_list[] = {
    "GLXBadContext",
    "GLXBadContextState",
    "GLXBadDrawable",
    "GLXBadPixmap",
    "GLXBadContextTag",
    "GLXBadCurrentWindow",
    "GLXBadRenderRequest",
    "GLXBadLargeRequest",
    "GLXUnsupportedPrivateRequest",
    "GLXBadFBConfig",
    "GLXBadPbuffer",
    "GLXBadCurrentDrawable",
    "GLXBadWindow",
};

static int __glXCloseDisplay(Display *dpy, XExtCodes *codes)
{
  GLXContext gc;

  gc = __glXGetCurrentContext();
  if (dpy == gc->currentDpy) {
    __glXSetCurrentContextNull();
    __glXFreeContext(gc);
  }

  return XextRemoveDisplay(glXExtensionInfo, dpy);
}

static char *__glXErrorString(Display *dpy, int code, XExtCodes *codes, 
			  char *buf, int n);

static /* const */ XExtensionHooks __glXExtensionHooks = {
    NULL,				/* create_gc */
    NULL,				/* copy_gc */
    NULL,				/* flush_gc */
    NULL,				/* free_gc */
    NULL,				/* create_font */
    NULL,				/* free_font */
    __glXCloseDisplay,			/* close_display */
    NULL,				/* wire_to_event */
    NULL,				/* event_to_wire */
    NULL,				/* error */
    __glXErrorString,			/* error_string */
};


XEXT_GENERATE_FIND_DISPLAY(__glXFindDisplay, glXExtensionInfo,
			   __glXExtensionName, &__glXExtensionHooks,
			   __GLX_NUMBER_EVENTS, NULL)

static XEXT_GENERATE_ERROR_STRING(__glXErrorString, __glXExtensionName,
				  __GLX_NUMBER_ERRORS, error_list)




/************************************************************************/

/*
** Free the per screen configs data as well as the array of
** __glXScreenConfigs.
*/
static void FreeScreenConfigs(__GLXdisplayPrivate *priv)
{
    __GLXscreenConfigs *psc;
    GLint i, screens;

    /* Free screen configuration information */
    psc = priv->screenConfigs;
    screens = ScreenCount(priv->dpy);
    for (i = 0; i < screens; i++, psc++) {
	if (psc->configs) {
	    _gl_context_modes_destroy( psc->configs );
	    if (psc->effectiveGLXexts)
		Xfree(psc->effectiveGLXexts);
	    psc->configs = NULL;	/* NOTE: just for paranoia */
	}
        if (psc->visuals) {
            _gl_context_modes_destroy( psc->visuals );
            psc->visuals = NULL;  /* NOTE: just for paranoia */
        }
 	Xfree((char*) psc->serverGLXexts);

#ifdef GLX_DIRECT_RENDERING
	if (psc->driScreen) {
	    psc->driScreen->destroyScreen(psc);
	    __glxHashDestroy(psc->drawHash);
            XFree(psc->driScreen);
            psc->driScreen = NULL;
	}
#endif
    }
    XFree((char*) priv->screenConfigs);
    priv->screenConfigs = NULL;
}

/*
** Release the private memory referred to in a display private
** structure.  The caller will free the extension structure.
*/
static int __glXFreeDisplayPrivate(XExtData *extension)
{
    __GLXdisplayPrivate *priv;

    priv = (__GLXdisplayPrivate*) extension->private_data;
    FreeScreenConfigs(priv);
    if(priv->serverGLXvendor) {
	Xfree((char*)priv->serverGLXvendor);
	priv->serverGLXvendor = 0x0; /* to protect against double free's */
    }
    if(priv->serverGLXversion) {
	Xfree((char*)priv->serverGLXversion);
	priv->serverGLXversion = 0x0; /* to protect against double free's */
    }

#ifdef GLX_DIRECT_RENDERING
    /* Free the direct rendering per display data */
    if (priv->driswDisplay)
	(*priv->driswDisplay->destroyDisplay)(priv->driswDisplay);
    priv->driswDisplay = NULL;

    if (priv->driDisplay)
	(*priv->driDisplay->destroyDisplay)(priv->driDisplay);
    priv->driDisplay = NULL;
#endif

    Xfree((char*) priv);
    return 0;
}

/************************************************************************/

/*
** Query the version of the GLX extension.  This procedure works even if
** the client extension is not completely set up.
*/
static Bool QueryVersion(Display *dpy, int opcode, int *major, int *minor)
{
    xGLXQueryVersionReq *req;
    xGLXQueryVersionReply reply;

    /* Send the glXQueryVersion request */
    LockDisplay(dpy);
    GetReq(GLXQueryVersion,req);
    req->reqType = opcode;
    req->glxCode = X_GLXQueryVersion;
    req->majorVersion = GLX_MAJOR_VERSION;
    req->minorVersion = GLX_MINOR_VERSION;
    _XReply(dpy, (xReply*) &reply, 0, False);
    UnlockDisplay(dpy);
    SyncHandle();

    if (reply.majorVersion != GLX_MAJOR_VERSION) {
	/*
	** The server does not support the same major release as this
	** client.
	*/
	return GL_FALSE;
    }
    *major = reply.majorVersion;
    *minor = GLX_MINOR_VERSION;
    // *minor = min(reply.minorVersion, GLX_MINOR_VERSION);
    return GL_TRUE;
}


/* 
 * We don't want to enable this GLX_OML_swap_method in glxext.h, 
 * because we can't support it.  The X server writes it out though,
 * so we should handle it somehow, to avoid false warnings.
 */
enum {
    IGNORE_GLX_SWAP_METHOD_OML = 0x8060
};


/*
 * getVisualConfigs uses the !tagged_only path.
 * getFBConfigs uses the tagged_only path.
 */
_X_HIDDEN void 
__glXInitializeVisualConfigFromTags( __GLcontextModes *config, int count, 
				     const INT32 *bp, Bool tagged_only,
				     Bool fbconfig_style_tags )
{
    int i;
    long int tag, tagvalue;
    
    if (!tagged_only) {
	/*
	 * Copy in the first set of properties.
	 * There should only be 18 initial properties.
	 */
	config->visualID = *bp++;

	config->visualType = _gl_convert_from_x_visual_type( *bp++ );

	config->rgbMode = *bp++;

	config->redBits = *bp++;
	config->greenBits = *bp++;
	config->blueBits = *bp++;
	config->alphaBits = *bp++;
	config->accumRedBits = *bp++;
	config->accumGreenBits = *bp++;
	config->accumBlueBits = *bp++;
	config->accumAlphaBits = *bp++;

	config->doubleBufferMode = *bp++;
	config->stereoMode = *bp++;

	config->rgbBits = *bp++;
	config->depthBits = *bp++;
	config->stencilBits = *bp++;
	config->numAuxBuffers = *bp++;
	config->level = *bp++;

	/* AppleSGLX supports pixmap and pbuffers with all config. */
	config->drawableType = GLX_WINDOW_BIT | GLX_PIXMAP_BIT | GLX_PBUFFER_BIT;
	/* Unfortunately this can create an ABI compatibility problem. */
	count -= 18;
    }

    /*
    ** Additional properties may be in a list at the end
    ** of the reply.  They are in pairs of property type
    ** and property value.
    */

#define FETCH_OR_SET(tag) \
    config-> tag = ( fbconfig_style_tags ) ? *bp++ : 1

    for (i = 0; i < count; i += 2 ) {
	tag = *bp++;

	switch(tag) {
	  case GLX_RGBA:
	    FETCH_OR_SET( rgbMode );
	    break;
	  case GLX_BUFFER_SIZE:
	    config->rgbBits = *bp++;
	    break;
	  case GLX_LEVEL:
	    config->level = *bp++;
	    break;
	  case GLX_DOUBLEBUFFER:
	    FETCH_OR_SET( doubleBufferMode );
	    break;
	  case GLX_STEREO:
	    FETCH_OR_SET( stereoMode );
	    break;
	  case GLX_AUX_BUFFERS:
	    config->numAuxBuffers = *bp++;
	    break;
	  case GLX_RED_SIZE:
	    config->redBits = *bp++;
	    break;
	  case GLX_GREEN_SIZE:
	    config->greenBits = *bp++;
	    break;
	  case GLX_BLUE_SIZE:
	    config->blueBits = *bp++;
	    break;
	  case GLX_ALPHA_SIZE:
	    config->alphaBits = *bp++;
	    break;
	  case GLX_DEPTH_SIZE:
	    config->depthBits = *bp++;
	    break;
	  case GLX_STENCIL_SIZE:
	    config->stencilBits = *bp++;
	    break;
	  case GLX_ACCUM_RED_SIZE:
	    config->accumRedBits = *bp++;
	    break;
	  case GLX_ACCUM_GREEN_SIZE:
	    config->accumGreenBits = *bp++;
	    break;
	  case GLX_ACCUM_BLUE_SIZE:
	    config->accumBlueBits = *bp++;
	    break;
	  case GLX_ACCUM_ALPHA_SIZE:
	    config->accumAlphaBits = *bp++;
	    break;
	  case GLX_VISUAL_CAVEAT_EXT:
	    config->visualRating = *bp++;    
	    break;
	  case GLX_X_VISUAL_TYPE:
	    config->visualType = *bp++;
	    break;
	  case GLX_TRANSPARENT_TYPE:
	    config->transparentPixel = *bp++;    
	    break;
	  case GLX_TRANSPARENT_INDEX_VALUE:
	    config->transparentIndex = *bp++;    
	    break;
	  case GLX_TRANSPARENT_RED_VALUE:
	    config->transparentRed = *bp++;    
	    break;
	  case GLX_TRANSPARENT_GREEN_VALUE:
	    config->transparentGreen = *bp++;    
	    break;
	  case GLX_TRANSPARENT_BLUE_VALUE:
	    config->transparentBlue = *bp++;    
	    break;
	  case GLX_TRANSPARENT_ALPHA_VALUE:
	    config->transparentAlpha = *bp++;    
	    break;
	  case GLX_VISUAL_ID:
	    config->visualID = *bp++;
	    break;
	  case GLX_DRAWABLE_TYPE:
	    config->drawableType = *bp++;

	    /* AppleSGLX supports pixmap and pbuffers with all config. */
	    config->drawableType |= GLX_WINDOW_BIT | GLX_PIXMAP_BIT 
		| GLX_PBUFFER_BIT;
	    break;
	  case GLX_RENDER_TYPE:
	    config->renderType = *bp++;
	    break;
	  case GLX_X_RENDERABLE:
	    config->xRenderable = *bp++;
	    break;
	  case GLX_FBCONFIG_ID:
	    config->fbconfigID = *bp++;
	    break;
	  case GLX_MAX_PBUFFER_WIDTH:
	    config->maxPbufferWidth = *bp++;
	    break;
	  case GLX_MAX_PBUFFER_HEIGHT:
	    config->maxPbufferHeight = *bp++;
	    break;
	  case GLX_MAX_PBUFFER_PIXELS:
	    config->maxPbufferPixels = *bp++;
	    break;
#if 0
	    /* We can't currently support these with CGL. */
	  case GLX_OPTIMAL_PBUFFER_WIDTH_SGIX:
	    config->optimalPbufferWidth = *bp++;
	    break;
	  case GLX_OPTIMAL_PBUFFER_HEIGHT_SGIX:
	    config->optimalPbufferHeight = *bp++;
	    break;
#endif
#if 0 
	    /* Not supported. */
          case GLX_VISUAL_SELECT_GROUP_SGIX:
	    config->visualSelectGroup = *bp++;
	    break;
#endif
#if 0
	    /* Changing the swap method is not supported by Xplugin. */
	  case GLX_SWAP_METHOD_OML:
	    config->swapMethod = *bp++;
	    break;
#endif
	    /* These 2 have the same values as the SGIS versions, and ARB. */
	  case GLX_SAMPLE_BUFFERS:
	    config->sampleBuffers = *bp++;
	    break;
	  case GLX_SAMPLES:
	    config->samples = *bp++;
	    break;
#if 0
	    /*These are part of GLX_EXT_texture_from_pixmap */
	  case GLX_BIND_TO_TEXTURE_RGB_EXT:
	    config->bindToTextureRgb = *bp++;
	    break;
	  case GLX_BIND_TO_TEXTURE_RGBA_EXT:
	    config->bindToTextureRgba = *bp++;
	    break;
	  case GLX_BIND_TO_MIPMAP_TEXTURE_EXT:
	    config->bindToMipmapTexture = *bp++;
	    break;
	  case GLX_BIND_TO_TEXTURE_TARGETS_EXT:
	    config->bindToTextureTargets = *bp++;
	    break;
	  case GLX_Y_INVERTED_EXT:
	    config->yInverted = *bp++;
	    break;
#endif
	case IGNORE_GLX_SWAP_METHOD_OML:
	    /* We ignore this tag.  See the comment above this function. */
	    ++bp;
	    break;
		
	  case None:
	    i = count;
	    break;
	  default:
	    if(getenv("LIBGL_DIAGNOSTIC")) {
		tagvalue = *bp++;
		fprintf(stderr, "WARNING: unknown GLX tag from server: "
			"tag 0x%lx value 0x%lx\n",
			tag, tagvalue);
	    }
	    break;
	}
    }

    config->renderType = (config->rgbMode) ? GLX_RGBA_BIT : GLX_COLOR_INDEX_BIT;

    config->haveAccumBuffer = ((config->accumRedBits +
			       config->accumGreenBits +
			       config->accumBlueBits +
			       config->accumAlphaBits) > 0);
    config->haveDepthBuffer = (config->depthBits > 0);
    config->haveStencilBuffer = (config->stencilBits > 0);
}

static __GLcontextModes *
createConfigsFromProperties(Display *dpy, int nvisuals, int nprops,
			    int screen, GLboolean tagged_only)
{
    INT32 buf[__GLX_TOTAL_CONFIG], *props;
    unsigned prop_size;
    __GLcontextModes *modes, *m;
    int i;

    if (nprops == 0)
	return NULL;

    /* Check number of properties */
    if (nprops < __GLX_MIN_CONFIG_PROPS || nprops > __GLX_MAX_CONFIG_PROPS)
	return NULL;

    /* Allocate memory for our config structure */
    modes = _gl_context_modes_create(nvisuals, sizeof(__GLcontextModes));
    if (!modes)
	return NULL;

    prop_size = nprops * __GLX_SIZE_INT32;
    if (prop_size <= sizeof(buf))
	props = buf;
    else
	props = Xmalloc(prop_size);

    /* Read each config structure and convert it into our format */
    m = modes;
    for (i = 0; i < nvisuals; i++) {
	_XRead(dpy, (char *)props, prop_size);
	
	/* 
	 * The XQuartz 2.3.2.1 X server doesn't set this properly, so
	 * set the proper bits here.
	 * AppleSGLX supports windows, pixmaps, and pbuffers with all config.
	 */
	m->drawableType = GLX_WINDOW_BIT | GLX_PIXMAP_BIT | GLX_PBUFFER_BIT;

	__glXInitializeVisualConfigFromTags(m, nprops, props,
					    tagged_only, GL_TRUE);
	m->screen = screen;
	m = m->next;
    }

    if (props != buf)
	Xfree(props);

    return modes;
}

static GLboolean
getVisualConfigs(Display *dpy, __GLXdisplayPrivate *priv, int screen)
{
    xGLXGetVisualConfigsReq *req;
    __GLXscreenConfigs *psc;
    xGLXGetVisualConfigsReply reply;
    
    LockDisplay(dpy);

    psc = priv->screenConfigs + screen;
    psc->visuals = NULL;
    GetReq(GLXGetVisualConfigs, req);
    req->reqType = priv->majorOpcode;
    req->glxCode = X_GLXGetVisualConfigs;
    req->screen = screen;

    if (!_XReply(dpy, (xReply*) &reply, 0, False))
	goto out;

    psc->visuals = createConfigsFromProperties(dpy,
					       reply.numVisuals,
					       reply.numProps,
					       screen, GL_FALSE);

 out:
    UnlockDisplay(dpy);
    return psc->visuals != NULL;
}

static GLboolean
getFBConfigs(Display *dpy, __GLXdisplayPrivate *priv, int screen)
{
    xGLXGetFBConfigsReq *fb_req;
    xGLXGetFBConfigsSGIXReq *sgi_req;
    xGLXVendorPrivateWithReplyReq *vpreq;
    xGLXGetFBConfigsReply reply;
    __GLXscreenConfigs *psc;

    psc = priv->screenConfigs + screen;
    psc->serverGLXexts = __glXGetStringFromServer(dpy, priv->majorOpcode,
						  X_GLXQueryServerString,
						  screen, GLX_EXTENSIONS);

    LockDisplay(dpy);

    psc->configs = NULL;
    if (atof(priv->serverGLXversion) >= 1.3) {
	GetReq(GLXGetFBConfigs, fb_req);
	fb_req->reqType = priv->majorOpcode;
	fb_req->glxCode = X_GLXGetFBConfigs;
	fb_req->screen = screen;
    } else if (strstr(psc->serverGLXexts, "GLX_SGIX_fbconfig") != NULL) {
	GetReqExtra(GLXVendorPrivateWithReply,
		    sz_xGLXGetFBConfigsSGIXReq +
		    sz_xGLXVendorPrivateWithReplyReq, vpreq);
	sgi_req = (xGLXGetFBConfigsSGIXReq *) vpreq;
	sgi_req->reqType = priv->majorOpcode;
	sgi_req->glxCode = X_GLXVendorPrivateWithReply;
	sgi_req->vendorCode = X_GLXvop_GetFBConfigsSGIX;
	sgi_req->screen = screen;
    } else
	goto out;

    if (!_XReply(dpy, (xReply*) &reply, 0, False))
	goto out;

    psc->configs = createConfigsFromProperties(dpy,
					       reply.numFBConfigs,
					       reply.numAttribs * 2,
					       screen, GL_TRUE);

 out:
    UnlockDisplay(dpy);
    return psc->configs != NULL;
}

/*
** Allocate the memory for the per screen configs for each screen.
** If that works then fetch the per screen configs data.
*/
static Bool AllocAndFetchScreenConfigs(Display *dpy, __GLXdisplayPrivate *priv)
{
    __GLXscreenConfigs *psc;
    GLint i, screens;

    /*
    ** First allocate memory for the array of per screen configs.
    */
    screens = ScreenCount(dpy);
    psc = (__GLXscreenConfigs*) Xmalloc(screens * sizeof(__GLXscreenConfigs));
    if (!psc) {
	return GL_FALSE;
    }
    memset(psc, 0, screens * sizeof(__GLXscreenConfigs));
    priv->screenConfigs = psc;
    
    priv->serverGLXversion = __glXGetStringFromServer(dpy, priv->majorOpcode,
						      X_GLXQueryServerString,
						      0, GLX_VERSION);
    if ( priv->serverGLXversion == NULL ) {
	FreeScreenConfigs(priv);
	return GL_FALSE;
    }

    for (i = 0; i < screens; i++, psc++) {
	getVisualConfigs(dpy, priv, i);
	getFBConfigs(dpy, priv, i);
    }

    SyncHandle();
    return GL_TRUE;
}

/*
** Initialize the client side extension code.
*/
_X_HIDDEN __GLXdisplayPrivate *__glXInitialize(Display* dpy)
{
    XExtDisplayInfo *info = __glXFindDisplay(dpy);
    XExtData **privList, *private, *found;
    __GLXdisplayPrivate *dpyPriv;
    XEDataObject dataObj;
    int major, minor;
#ifdef GLX_DIRECT_RENDERING
    Bool glx_direct, glx_accel;
#endif
 
    
#if defined(USE_XTHREADS)
    {
        static int firstCall = 1;
        if (firstCall) {
            /* initialize the GLX mutexes */
            xmutex_init(&__glXmutex);
            firstCall = 0;
        }
    }
#endif

    INIT_MESA_SPARC
    /* The one and only long long lock */
    __glXLock();

    if (!XextHasExtension(info)) {
	/* No GLX extension supported by this server. Oh well. */
	__glXUnlock();
	XMissingExtension(dpy, __glXExtensionName);
	return 0;
    }

    /* See if a display private already exists.  If so, return it */
    dataObj.display = dpy;
    privList = XEHeadOfExtensionList(dataObj);
    found = XFindOnExtensionList(privList, info->codes->extension);
    if (found) {
	__glXUnlock();
	return (__GLXdisplayPrivate *) found->private_data;
    }

    /* See if the versions are compatible */
    if (!QueryVersion(dpy, info->codes->major_opcode, &major, &minor)) {
	/* The client and server do not agree on versions.  Punt. */
	__glXUnlock();
	return 0;
    }

    /*
    ** Allocate memory for all the pieces needed for this buffer.
    */
    private = (XExtData *) Xmalloc(sizeof(XExtData));
    if (!private) {
	__glXUnlock();
	return 0;
    }
    dpyPriv = (__GLXdisplayPrivate *) Xcalloc(1, sizeof(__GLXdisplayPrivate));
    if (!dpyPriv) {
	__glXUnlock();
	Xfree((char*) private);
	return 0;
    }

    /*
    ** Init the display private and then read in the screen config
    ** structures from the server.
    */
    dpyPriv->majorOpcode = info->codes->major_opcode;
    dpyPriv->majorVersion = major;
    dpyPriv->minorVersion = minor;
    dpyPriv->dpy = dpy;

    dpyPriv->serverGLXvendor = 0x0; 
    dpyPriv->serverGLXversion = 0x0;

#ifdef GLX_DIRECT_RENDERING
    glx_direct = (getenv("LIBGL_ALWAYS_INDIRECT") == NULL);
    glx_accel = (getenv("LIBGL_ALWAYS_SOFTWARE") == NULL);

    /*
    ** Initialize the direct rendering per display data and functions.
    ** Note: This _must_ be done before calling any other DRI routines
    ** (e.g., those called in AllocAndFetchScreenConfigs).
    */
    if (glx_direct && glx_accel) {
	dpyPriv->driDisplay = driCreateDisplay(dpy);
    }
    if (glx_direct)
	dpyPriv->driswDisplay = driswCreateDisplay(dpy);
#endif
    apple_init_glx(dpy);

    if (!AllocAndFetchScreenConfigs(dpy, dpyPriv)) {
	__glXUnlock();
	Xfree((char*) dpyPriv);
	Xfree((char*) private);
	return 0;
    }

    /*
    ** Fill in the private structure.  This is the actual structure that
    ** hangs off of the Display structure.  Our private structure is
    ** referred to by this structure.  Got that?
    */
    private->number = info->codes->extension;
    private->next = 0;
    private->free_private = __glXFreeDisplayPrivate;
    private->private_data = (char *) dpyPriv;
    XAddToExtensionList(privList, private);

    if (dpyPriv->majorVersion == 1 && dpyPriv->minorVersion >= 1) {
        __glXClientInfo(dpy, dpyPriv->majorOpcode);
    }
    __glXUnlock();

    return dpyPriv;
}

/*
** Setup for sending a GLX command on dpy.  Make sure the extension is
** initialized.  Try to avoid calling __glXInitialize as its kinda slow.
*/
_X_HIDDEN CARD8 __glXSetupForCommand(Display *dpy)
{
    GLXContext gc;
    __GLXdisplayPrivate *priv;

    /* If this thread has a current context, flush its rendering commands */
    gc = __glXGetCurrentContext();
    if (gc->currentDpy) {
	/* Flush rendering buffer of the current context, if any */
	(void) __glXFlushRenderBuffer(gc, gc->pc);

	if (gc->currentDpy == dpy) {
	    /* Use opcode from gc because its right */
	    INIT_MESA_SPARC
	    return gc->majorOpcode;
	} else {
	    /*
	    ** Have to get info about argument dpy because it might be to
	    ** a different server
	    */
	}
    }

    /* Forced to lookup extension via the slow initialize route */
    priv = __glXInitialize(dpy);
    if (!priv) {
	return 0;
    }
    return priv->majorOpcode;
}

/**
 * Flush the drawing command transport buffer.
 * 
 * \param ctx  Context whose transport buffer is to be flushed.
 * \param pc   Pointer to first unused buffer location.
 * 
 * \todo
 * Modify this function to use \c ctx->pc instead of the explicit
 * \c pc parameter.
 */
_X_HIDDEN GLubyte *__glXFlushRenderBuffer(__GLXcontext *ctx, GLubyte *pc)
{
    Display * const dpy = ctx->currentDpy;
#ifdef USE_XCB
    xcb_connection_t *c = XGetXCBConnection(dpy);
#else
    xGLXRenderReq *req;
#endif /* USE_XCB */
    const GLint size = pc - ctx->buf;

    if ( (dpy != NULL) && (size > 0) ) {
#ifdef USE_XCB
	xcb_glx_render(c, ctx->currentContextTag, size,
		       (const uint8_t *)ctx->buf);
#else
	/* Send the entire buffer as an X request */
	LockDisplay(dpy);
	GetReq(GLXRender,req); 
	req->reqType = ctx->majorOpcode;
	req->glxCode = X_GLXRender; 
	req->contextTag = ctx->currentContextTag;
	req->length += (size + 3) >> 2;
	_XSend(dpy, (char *)ctx->buf, size);
	UnlockDisplay(dpy);
	SyncHandle();
#endif
    }

    /* Reset pointer and return it */
    ctx->pc = ctx->buf;
    return ctx->pc;
}


/**
 * Send a portion of a GLXRenderLarge command to the server.  The advantage of
 * this function over \c __glXSendLargeCommand is that callers can use the
 * data buffer in the GLX context and may be able to avoid allocating an
 * extra buffer.  The disadvantage is the clients will have to do more
 * GLX protocol work (i.e., calculating \c totalRequests, etc.).
 *
 * \sa __glXSendLargeCommand
 *
 * \param gc             GLX context
 * \param requestNumber  Which part of the whole command is this?  The first
 *                       request is 1.
 * \param totalRequests  How many requests will there be?
 * \param data           Command data.
 * \param dataLen        Size, in bytes, of the command data.
 */
_X_HIDDEN void __glXSendLargeChunk(__GLXcontext *gc, GLint requestNumber, 
				   GLint totalRequests,
				   const GLvoid * data, GLint dataLen)
{
    Display *dpy = gc->currentDpy;
#ifdef USE_XCB
    xcb_connection_t *c = XGetXCBConnection(dpy);
    xcb_glx_render_large(c, gc->currentContextTag, requestNumber, totalRequests, dataLen, data);
#else
    xGLXRenderLargeReq *req;
    
    if ( requestNumber == 1 ) {
	LockDisplay(dpy);
    }

    GetReq(GLXRenderLarge,req); 
    req->reqType = gc->majorOpcode;
    req->glxCode = X_GLXRenderLarge; 
    req->contextTag = gc->currentContextTag;
    req->length += (dataLen + 3) >> 2;
    req->requestNumber = requestNumber;
    req->requestTotal = totalRequests;
    req->dataBytes = dataLen;
    Data(dpy, data, dataLen);

    if ( requestNumber == totalRequests ) {
	UnlockDisplay(dpy);
	SyncHandle();
    }
#endif /* USE_XCB */
}


/**
 * Send a command that is too large for the GLXRender protocol request.
 * 
 * Send a large command, one that is too large for some reason to
 * send using the GLXRender protocol request.  One reason to send
 * a large command is to avoid copying the data.
 * 
 * \param ctx        GLX context
 * \param header     Header data.
 * \param headerLen  Size, in bytes, of the header data.  It is assumed that
 *                   the header data will always be small enough to fit in
 *                   a single X protocol packet.
 * \param data       Command data.
 * \param dataLen    Size, in bytes, of the command data.
 */
_X_HIDDEN void __glXSendLargeCommand(__GLXcontext *ctx,
				     const GLvoid *header, GLint headerLen,
				     const GLvoid *data, GLint dataLen)
{
    GLint maxSize;
    GLint totalRequests, requestNumber;

    /*
    ** Calculate the maximum amount of data can be stuffed into a single
    ** packet.  sz_xGLXRenderReq is added because bufSize is the maximum
    ** packet size minus sz_xGLXRenderReq.
    */
    maxSize = (ctx->bufSize + sz_xGLXRenderReq) - sz_xGLXRenderLargeReq;
    totalRequests = 1 + (dataLen / maxSize);
    if (dataLen % maxSize) totalRequests++;

    /*
    ** Send all of the command, except the large array, as one request.
    */
    assert( headerLen <= maxSize );
    __glXSendLargeChunk(ctx, 1, totalRequests, header, headerLen);

    /*
    ** Send enough requests until the whole array is sent.
    */
    for ( requestNumber = 2 ; requestNumber <= (totalRequests - 1) ; requestNumber++ ) {
	__glXSendLargeChunk(ctx, requestNumber, totalRequests, data, maxSize);
	data = (const GLvoid *) (((const GLubyte *) data) + maxSize);
	dataLen -= maxSize;
	assert( dataLen > 0 );
    }

    assert( dataLen <= maxSize );
    __glXSendLargeChunk(ctx, requestNumber, totalRequests, data, dataLen);
}

/************************************************************************/

#ifdef DEBUG
_X_HIDDEN void __glXDumpDrawBuffer(__GLXcontext *ctx)
{
    GLubyte *p = ctx->buf;
    GLubyte *end = ctx->pc;
    GLushort opcode, length;

    while (p < end) {
	/* Fetch opcode */
	opcode = *((GLushort*) p);
	length = *((GLushort*) (p + 2));
	printf("%2x: %5d: ", opcode, length);
	length -= 4;
	p += 4;
	while (length > 0) {
	    printf("%08x ", *((unsigned *) p));
	    p += 4;
	    length -= 4;
	}
	printf("\n");
    }	    
}
#endif

#ifdef  USE_SPARC_ASM
/*
 * This is where our dispatch table's bounds are.
 * And the static mesa_init is taken directly from
 * Mesa's 'sparc.c' initializer.
 *
 * We need something like this here, because this version
 * of openGL/glx never initializes a Mesa context, and so
 * the address of the dispatch table pointer never gets stuffed
 * into the dispatch jump table otherwise.
 *
 * It matters only on SPARC, and only if you are using assembler
 * code instead of C-code indirect dispatch.
 *
 * -- FEM, 04.xii.03
 */
extern unsigned int _mesa_sparc_glapi_begin;
extern unsigned int _mesa_sparc_glapi_end;
extern void __glapi_sparc_icache_flush(unsigned int *);

static void
_glx_mesa_init_sparc_glapi_relocs(void)
{
	unsigned int *insn_ptr, *end_ptr;
	unsigned long disp_addr;

	insn_ptr = &_mesa_sparc_glapi_begin;
	end_ptr = &_mesa_sparc_glapi_end;
	disp_addr = (unsigned long) &_glapi_Dispatch;

	/*
         * Verbatim from Mesa sparc.c.  It's needed because there doesn't
         * seem to be a better way to do this:
         *
         * UNCONDITIONAL_JUMP ( (*_glapi_Dispatch) + entry_offset )
         *
         * This code is patching in the ADDRESS of the pointer to the
         * dispatch table.  Hence, it must be called exactly once, because
         * that address is not going to change.
         *
         * What it points to can change, but Mesa (and hence, we) assume
         * that there is only one pointer.
         *
	 */
	while (insn_ptr < end_ptr) {
#if ( defined(__sparc_v9__) && ( !defined(__linux__) || defined(__linux_64__) ) )	
/*
	This code patches for 64-bit addresses.  This had better
	not happen for Sparc/Linux, no matter what architecture we
	are building for.  So, don't do this.

        The 'defined(__linux_64__)' is used here as a placeholder for
        when we do do 64-bit usermode on sparc linux.
	*/
		insn_ptr[0] |= (disp_addr >> (32 + 10));
		insn_ptr[1] |= ((disp_addr & 0xffffffff) >> 10);
		__glapi_sparc_icache_flush(&insn_ptr[0]);
		insn_ptr[2] |= ((disp_addr >> 32) & ((1 << 10) - 1));
		insn_ptr[3] |= (disp_addr & ((1 << 10) - 1));
		__glapi_sparc_icache_flush(&insn_ptr[2]);
		insn_ptr += 11;
#else
		insn_ptr[0] |= (disp_addr >> 10);
		insn_ptr[1] |= (disp_addr & ((1 << 10) - 1));
		__glapi_sparc_icache_flush(&insn_ptr[0]);
		insn_ptr += 5;
#endif
	}
}
#endif  /* sparc ASM in use */