tdfx_span.c

Mesa is an open-source implementation of the OpenGL specification - a system for rendering interacti

/* -*- mode: c; c-basic-offset: 3 -*-
 *
 * Copyright 2000 VA Linux Systems Inc., Fremont, California.
 *
 * All Rights Reserved.
 *
 * 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 and this permission notice (including the next
 * paragraph) 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
 * VA LINUX SYSTEMS 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.
 */
/* $XFree86: xc/lib/GL/mesa/src/drv/tdfx/tdfx_span.c,v 1.7 2002/10/30 12:52:00 alanh Exp $ */

/*
 * Original rewrite:
 *	Gareth Hughes <[e-mail]>, 29 Sep - 1 Oct 2000
 *
 * Authors:
 *	Gareth Hughes <[e-mail]>
 *	Brian Paul <[e-mail]>
 *	Keith Whitwell <[e-mail]>
 *
 */

#include "tdfx_context.h"
#include "tdfx_lock.h"
#include "tdfx_span.h"
#include "tdfx_render.h"
#include "swrast/swrast.h"


#define DBG 0


#define LOCAL_VARS							\
   driRenderbuffer *drb = (driRenderbuffer *) rb;			\
   __DRIdrawablePrivate *const dPriv = drb->dPriv;			\
   GLuint pitch = drb->backBuffer ? info.strideInBytes			\
     : (drb->pitch * drb->cpp);						\
   const GLuint bottom = dPriv->h - 1;					\
   char *buf = (char *)((char *)info.lfbPtr +				\
			 (dPriv->x * drb->cpp) +			\
			 (dPriv->y * pitch));				\
   GLuint p;								\
   (void) buf; (void) p;


#define Y_FLIP(_y)		(bottom - _y)


#define HW_WRITE_LOCK()							\
   tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);				\
   GrLfbInfo_t info;							\
   FLUSH_BATCH( fxMesa );						\
   UNLOCK_HARDWARE( fxMesa );						\
   LOCK_HARDWARE( fxMesa );						\
   info.size = sizeof(GrLfbInfo_t);					\
   if (fxMesa->Glide.grLfbLock(GR_LFB_WRITE_ONLY, fxMesa->DrawBuffer,	\
			       LFB_MODE, GR_ORIGIN_UPPER_LEFT, FXFALSE,	\
			       &info)) {

#define HW_WRITE_UNLOCK()						\
      fxMesa->Glide.grLfbUnlock( GR_LFB_WRITE_ONLY, fxMesa->DrawBuffer );\
   }


#define HW_READ_LOCK()							\
   tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);				\
   GrLfbInfo_t info;							\
   FLUSH_BATCH( fxMesa );						\
   UNLOCK_HARDWARE( fxMesa );						\
   LOCK_HARDWARE( fxMesa );						\
   info.size = sizeof(GrLfbInfo_t);					\
   if ( fxMesa->Glide.grLfbLock( GR_LFB_READ_ONLY, fxMesa->ReadBuffer,	\
                   LFB_MODE, GR_ORIGIN_UPPER_LEFT, FXFALSE, &info ) )	\
   {

#define HW_READ_UNLOCK()						\
      fxMesa->Glide.grLfbUnlock( GR_LFB_READ_ONLY, fxMesa->ReadBuffer );\
   }


#define HW_WRITE_CLIPLOOP()						\
      do {								\
         int _nc = fxMesa->numClipRects;				\
         while (_nc--) {						\
            int minx = fxMesa->pClipRects[_nc].x1 - fxMesa->x_offset;	\
	    int miny = fxMesa->pClipRects[_nc].y1 - fxMesa->y_offset;	\
	    int maxx = fxMesa->pClipRects[_nc].x2 - fxMesa->x_offset;	\
	    int maxy = fxMesa->pClipRects[_nc].y2 - fxMesa->y_offset;

#define HW_READ_CLIPLOOP()						\
      do {								\
         const __DRIdrawablePrivate *dPriv = fxMesa->driDrawable;	\
         drm_clip_rect_t *rect = dPriv->pClipRects;			\
         int _nc = dPriv->numClipRects;					\
         while (_nc--) {						\
            const int minx = rect->x1 - fxMesa->x_offset;		\
            const int miny = rect->y1 - fxMesa->y_offset;		\
            const int maxx = rect->x2 - fxMesa->x_offset;		\
            const int maxy = rect->y2 - fxMesa->y_offset;		\
            rect++;

#define HW_ENDCLIPLOOP()						\
	 }								\
      } while (0)



#define LFB_MODE	GR_LFBWRITEMODE_565


/* 16 bit, RGB565 color spanline and pixel functions */			\

#undef INIT_MONO_PIXEL
#define INIT_MONO_PIXEL(p, color) \
  p = TDFXPACKCOLOR565( color[0], color[1], color[2] )


#define WRITE_RGBA( _x, _y, r, g, b, a )				\
   *(GLushort *)(buf + _x*2 + _y*pitch) = ((((int)r & 0xf8) << 8) |	\
					   (((int)g & 0xfc) << 3) |	\
					   (((int)b & 0xf8) >> 3))

#define WRITE_PIXEL( _x, _y, p )					\
    *(GLushort *)(buf + _x*2 + _y*pitch) = p

#define READ_RGBA( rgba, _x, _y )					\
    do {								\
	GLushort p = *(GLushort *)(buf + _x*2 + _y*pitch);		\
	rgba[0] = (((p >> 11) & 0x1f) * 255) / 31;			\
	rgba[1] = (((p >>  5) & 0x3f) * 255) / 63;			\
	rgba[2] = (((p >>  0) & 0x1f) * 255) / 31;			\
	rgba[3] = 0xff;							\
    } while (0)

#define TAG(x) tdfx##x##_RGB565
#define BYTESPERPIXEL 2
#include "spantmp.h"
#undef BYTESPERPIXEL


/* 16 bit, BGR565 color spanline and pixel functions */			\
#if 0

#define WRITE_RGBA( _x, _y, r, g, b, a )				\
   *(GLushort *)(buf + _x*2 + _y*pitch) = ((((int)b & 0xf8) << 8) |	\
					   (((int)g & 0xfc) << 3) |	\
					   (((int)r & 0xf8) >> 3))

#define WRITE_PIXEL( _x, _y, p )					\
    *(GLushort *)(buf + _x*2 + _y*pitch) = p

#define READ_RGBA( rgba, _x, _y )					\
    do {								\
	GLushort p = *(GLushort *)(buf + _x*2 + _y*pitch);		\
	rgba[0] = (p << 3) & 0xf8;					\
	rgba[1] = (p >> 3) & 0xfc;					\
	rgba[2] = (p >> 8) & 0xf8;					\
	rgba[3] = 0xff;							\
    } while (0)

#define TAG(x) tdfx##x##_BGR565
#define BYTESPERPIXEL 2
#include "spantmp.h"
#undef BYTESPERPIXEL
#endif


#undef LFB_MODE
#define LFB_MODE	GR_LFBWRITEMODE_888


/* 24 bit, RGB888 color spanline and pixel functions */
#undef INIT_MONO_PIXEL
#define INIT_MONO_PIXEL(p, color) \
  p = TDFXPACKCOLOR888( color[0], color[1], color[2] )

#define WRITE_RGBA( _x, _y, r, g, b, a )				\
   *(GLuint *)(buf + _x*3 + _y*pitch) = ((b << 0) |			\
					 (g << 8) |			\
					 (r << 16))

#define WRITE_PIXEL( _x, _y, p )					\
   *(GLuint *)(buf + _x*3 + _y*pitch) = p

#define READ_RGBA( rgba, _x, _y )					\
do {									\
   GLuint p = *(GLuint *)(buf + _x*3 + _y*pitch);			\
   rgba[0] = (p >> 16) & 0xff;						\
   rgba[1] = (p >> 8)  & 0xff;						\
   rgba[2] = (p >> 0)  & 0xff;						\
   rgba[3] = 0xff;							\
} while (0)

#define TAG(x) tdfx##x##_RGB888
#define BYTESPERPIXEL 4
#include "spantmp.h"
#undef BYTESPERPIXEL


#undef LFB_MODE
#define LFB_MODE	GR_LFBWRITEMODE_8888


/* 32 bit, ARGB8888 color spanline and pixel functions */
#undef INIT_MONO_PIXEL
#define INIT_MONO_PIXEL(p, color) \
  p = TDFXPACKCOLOR8888( color[0], color[1], color[2], color[3] )

#define WRITE_RGBA( _x, _y, r, g, b, a )				\
   *(GLuint *)(buf + _x*4 + _y*pitch) = ((b <<  0) |			\
					 (g <<  8) |			\
					 (r << 16) |			\
					 (a << 24) )

#define WRITE_PIXEL( _x, _y, p )					\
   *(GLuint *)(buf + _x*4 + _y*pitch) = p

#define READ_RGBA( rgba, _x, _y )					\
do {									\
   GLuint p = *(GLuint *)(buf + _x*4 + _y*pitch);			\
   rgba[0] = (p >> 16) & 0xff;						\
   rgba[1] = (p >>  8) & 0xff;						\
   rgba[2] = (p >>  0) & 0xff;						\
   rgba[3] = (p >> 24) & 0xff;						\
} while (0)

#define TAG(x) tdfx##x##_ARGB8888
#define BYTESPERPIXEL 4
#include "spantmp.h"
#undef BYTESPERPIXEL



/* ================================================================
 * Old span functions below...
 */


/*
 * Examine the cliprects to generate an array of flags to indicate
 * which pixels in a span are visible.  Note: (x,y) is a screen
 * coordinate.
 */
static void
generate_vismask(const tdfxContextPtr fxMesa, GLint x, GLint y, GLint n,
                 GLubyte vismask[])
{
   GLboolean initialized = GL_FALSE;
   GLint i, j;

   /* Ensure we clear the visual mask */
   MEMSET(vismask, 0, n);

   /* turn on flags for all visible pixels */
   for (i = 0; i < fxMesa->numClipRects; i++) {
      const drm_clip_rect_t *rect = &fxMesa->pClipRects[i];

      if (y >= rect->y1 && y < rect->y2) {
	 if (x >= rect->x1 && x + n <= rect->x2) {
	    /* common case, whole span inside cliprect */
	    MEMSET(vismask, 1, n);
	    return;
	 }
	 if (x < rect->x2 && x + n >= rect->x1) {
	    /* some of the span is inside the rect */
	    GLint start, end;
	    if (!initialized) {
	       MEMSET(vismask, 0, n);
	       initialized = GL_TRUE;
	    }
	    if (x < rect->x1)
	       start = rect->x1 - x;
	    else
	       start = 0;
	    if (x + n > rect->x2)
	       end = rect->x2 - x;
	    else
	       end = n;
	    assert(start >= 0);
	    assert(end <= n);
	    for (j = start; j < end; j++)
	       vismask[j] = 1;
	 }
      }
   }
}

/*
 * Examine cliprects and determine if the given screen pixel is visible.
 */
static GLboolean
visible_pixel(const tdfxContextPtr fxMesa, int scrX, int scrY)
{
   int i;
   for (i = 0; i < fxMesa->numClipRects; i++) {
      const drm_clip_rect_t *rect = &fxMesa->pClipRects[i];
      if (scrX >= rect->x1 &&
	  scrX < rect->x2 &&
	  scrY >= rect->y1 && scrY < rect->y2) return GL_TRUE;
   }
   return GL_FALSE;
}



/*
 * Depth buffer read/write functions.
 */
/*
 * To read the frame buffer, we need to lock and unlock it.  The
 * four macros {READ,WRITE}_FB_SPAN_{LOCK,UNLOCK}
 * do this for us.
 *
 * Note that the lock must be matched with an unlock.  These
 * macros include a spare curly brace, so they must
 * be syntactically matched.
 *
 * Note, also, that you can't lock a buffer twice with different
 * modes.  That is to say, you can't lock a buffer in both read
 * and write modes.  The strideInBytes and LFB pointer will be
 * the same with read and write locks, so you can use either.
 * o The HW has different state for reads and writes, so
 *   locking it twice may give screwy results.
 * o The DRM won't let you lock twice.  It hangs.  This is probably
 *   because of the LOCK_HARDWARE IN THE *_FB_SPAN_LOCK macros,
 *   and could be eliminated with nonlocking lock routines.  But
 *   what's the point after all.
 */
#define READ_FB_SPAN_LOCK(fxMesa, info, target_buffer)              \
  UNLOCK_HARDWARE(fxMesa);                                          \
  LOCK_HARDWARE(fxMesa);                                            \
  (info).size=sizeof(info);                                         \
  if (fxMesa->Glide.grLfbLock(GR_LFB_READ_ONLY,                     \
                target_buffer,                                      \
                GR_LFBWRITEMODE_ANY,                                \
                GR_ORIGIN_UPPER_LEFT,                               \
                FXFALSE,                                            \
                &(info))) {

#define READ_FB_SPAN_UNLOCK(fxMesa, target_buffer)                  \
    fxMesa->Glide.grLfbUnlock(GR_LFB_READ_ONLY, target_buffer);     \
  } else {                                                          \
    fprintf(stderr, "tdfxDriver: Can't get %s (%d) read lock\n",    \
            (target_buffer == GR_BUFFER_BACKBUFFER)                 \
                ? "back buffer"                                     \
            : ((target_buffer == GR_BUFFER_AUXBUFFER)               \
                ? "depth buffer"                                    \
               : "unknown buffer"),                                 \
            target_buffer);                                         \
  }

#define WRITE_FB_SPAN_LOCK(fxMesa, info, target_buffer, write_mode) \
  UNLOCK_HARDWARE(fxMesa);                                          \
  LOCK_HARDWARE(fxMesa);                                            \
  info.size=sizeof(info);                                           \
  if (fxMesa->Glide.grLfbLock(GR_LFB_WRITE_ONLY,                    \
                target_buffer,                                      \
                write_mode,                                         \
                GR_ORIGIN_UPPER_LEFT,                               \
                FXFALSE,                                            \
                &info)) {

#define WRITE_FB_SPAN_UNLOCK(fxMesa, target_buffer)                 \
    fxMesa->Glide.grLfbUnlock(GR_LFB_WRITE_ONLY, target_buffer);    \
  } else {                                                          \
    fprintf(stderr, "tdfxDriver: Can't get %s (%d) write lock\n",   \
            (target_buffer == GR_BUFFER_BACKBUFFER)                 \
                ? "back buffer"                                     \
            : ((target_buffer == GR_BUFFER_AUXBUFFER)               \
                ? "depth buffer"                                    \
               : "unknown buffer"),                                 \
            target_buffer);                                         \
  }

/*
 * Because the Linear Frame Buffer is not necessarily aligned
 * with the depth buffer, we have to do some fiddling
 * around to get the right addresses.
 *
 * Perhaps a picture is in order.  The Linear Frame Buffer
 * looks like this:
 *
 *   |<----------------------info.strideInBytes------------->|
 *   |<-----physicalStrideInBytes------->|
 *   +-----------------------------------+xxxxxxxxxxxxxxxxxxx+
 *   |                                   |                   |
 *   |          Legal Memory             |  Forbidden Zone   |
 *   |                                   |                   |
 *   +-----------------------------------+xxxxxxxxxxxxxxxxxxx+
 *
 * You can only reliably read and write legal locations.  Reads
 * and writes from the Forbidden Zone will return undefined values,
 * and may cause segmentation faults.
 *
 * Now, the depth buffer may not end up in a location such each
 * scan line is an LFB line.  For example, the depth buffer may
 * look like this:
 *
 *    wrapped               ordinary.
 *   +-----------------------------------+xxxxxxxxxxxxxxxxxxx+
 *   |0000000000000000000000             |                   | back
 *   |1111111111111111111111             |                   | buffer
 *   |2222222222222222222222             |                   |
 *   |4096b align. padxx00000000000000000|  Forbidden Zone   | depth
 *   |0000              11111111111111111|                   | buffer
 *   |1111              22222222222222222|                   |
 *   |2222                               |                   |
 *   +-----------------------------------+xxxxxxxxxxxxxxxxxxx+
 * where each number is the scan line number.  We know it will
 * be aligned on 128 byte boundaries, at least.  Aligning this
 * on a scanline boundary causes the back and depth buffers to
 * thrash in the SST1 cache.  (Note that the back buffer is always
 * allocated at the beginning of LFB memory, and so it is always
 * properly aligned with the LFB stride.)
 *
 * We call the beginning of the line (which is the rightmost
 * part of the depth line in the picture above) the *ordinary* part
 * of the scanline, and the end of the line (which is the
 * leftmost part, one line below) the *wrapped* part of the scanline.
 * a.) We need to know what x value to subtract from the screen
 *     x coordinate to index into the wrapped part.
 * b.) We also need to figure out if we need to read from the ordinary
 *     part scan line, or from the wrapped part of the scan line.
 *
 * [ad a]
 * The first wrapped x coordinate is that coordinate such that
 *           depthBufferOffset&(info.strideInBytes) + x*elmentSize  {*}
 *                            > physicalStrideInBytes
 *     where depthBufferOffset is the LFB distance in bytes
 *     from the back buffer to the depth buffer.  The expression
 *           depthBufferOffset&(info.strideInBytes)
 *     is then the offset (in bytes) from the beginining of (any)
 *     depth buffer line to first element in the line.
 * Simplifying inequation {*} above we see that x is the smallest
 * value such that
 *         x*elementSize > physicalStrideInBytes                      {**}
 *                            - depthBufferOffset&(info.strideInBytes)
 * Now, we know that both the summands on the right are multiples of
 * 128, and elementSize <= 4, so if equality holds in {**}, x would
 * be a multiple of 32.  Thus we can set x to
 *         xwrapped = (physicalStrideInBytes
 *                      - depthBufferOffset&(info.strideInBytes))/elementSize
 *                      + 1
 *
 * [ad b]
 * Question b is now simple.  We read from the wrapped scan line if
 * x is greater than xwrapped.