src/mesa/drivers/dri/r200/r200_span.c

   1 /* $XFree86: xc/lib/GL/mesa/src/drv/r200/r200_span.c,v 1.1 2002/10/30 12:51:52 alanh Exp $ */
   2 /*
   3 Copyright (C) The Weather Channel, Inc.  2002.  All Rights Reserved.
   4
   5 The Weather Channel (TM) funded Tungsten Graphics to develop the
   6 initial release of the Radeon 8500 driver under the XFree86 license.
   7 This notice must be preserved.
   8
   9 Permission is hereby granted, free of charge, to any person obtaining
  10 a copy of this software and associated documentation files (the
  11 "Software"), to deal in the Software without restriction, including
  12 without limitation the rights to use, copy, modify, merge, publish,
  13 distribute, sublicense, and/or sell copies of the Software, and to
  14 permit persons to whom the Software is furnished to do so, subject to
  15 the following conditions:
  16
  17 The above copyright notice and this permission notice (including the
  18 next paragraph) shall be included in all copies or substantial
  19 portions of the Software.
  20
  21 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  22 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  23 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
  24 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
  25 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
  26 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
  27 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  28
  29 **************************************************************************/
  30
  31 /*
  32  * Authors:
  33  *   Keith Whitwell <keith@tungstengraphics.com>
  34  */
  35
  36 #include "glheader.h"
  37 #include "imports.h"
  38 #include "swrast/swrast.h"
  39 #include "colormac.h"
  40
  41 #include "r200_context.h"
  42 #include "r200_ioctl.h"
  43 #include "r200_state.h"
  44 #include "r200_span.h"
  45 #include "r200_tex.h"
  46
  47 #define DBG 0
  48
  49 #define LOCAL_VARS                                                      \
  50    r200ContextPtr rmesa = R200_CONTEXT(ctx);                    \
  51    r200ScreenPtr r200Screen = rmesa->r200Screen;                        \
  52    __DRIscreenPrivate *sPriv = rmesa->dri.screen;                       \
  53    __DRIdrawablePrivate *dPriv = rmesa->dri.drawable;                   \
  54    GLuint pitch = r200Screen->frontPitch * r200Screen->cpp;             \
  55    GLuint height = dPriv->h;                                            \
  56    char *buf = (char *)(sPriv->pFB +                                    \
  57                         rmesa->state.color.drawOffset +                 \
  58                         (dPriv->x * r200Screen->cpp) +          \
  59                         (dPriv->y * pitch));                            \
  60    char *read_buf = (char *)(sPriv->pFB +                               \
  61                              rmesa->state.pixel.readOffset +            \
  62                              (dPriv->x * r200Screen->cpp) +             \
  63                              (dPriv->y * pitch));                       \
  64    GLuint p;                                                            \
  65    (void) read_buf; (void) buf; (void) p
  66
  67 #define LOCAL_DEPTH_VARS                                                \
  68    r200ContextPtr rmesa = R200_CONTEXT(ctx);                    \
  69    r200ScreenPtr r200Screen = rmesa->r200Screen;                        \
  70    __DRIscreenPrivate *sPriv = rmesa->dri.screen;                       \
  71    __DRIdrawablePrivate *dPriv = rmesa->dri.drawable;                   \
  72    GLuint height = dPriv->h;                                            \
  73    GLuint xo = dPriv->x;                                                \
  74    GLuint yo = dPriv->y;                                                \
  75    char *buf = (char *)(sPriv->pFB + r200Screen->depthOffset);  \
  76    (void) buf
  77
  78 #define LOCAL_STENCIL_VARS      LOCAL_DEPTH_VARS
  79
  80
  81 #define CLIPPIXEL( _x, _y )                                             \
  82    ((_x >= minx) && (_x < maxx) && (_y >= miny) && (_y < maxy))
  83
  84
  85 #define CLIPSPAN( _x, _y, _n, _x1, _n1, _i )                            \
  86    if ( _y < miny || _y >= maxy ) {                                     \
  87       _n1 = 0, _x1 = x;                                                 \
  88    } else {                                                             \
  89       _n1 = _n;                                                         \
  90       _x1 = _x;                                                         \
  91       if ( _x1 < minx ) _i += (minx-_x1), n1 -= (minx-_x1), _x1 = minx; \
  92       if ( _x1 + _n1 >= maxx ) n1 -= (_x1 + n1 - maxx);                 \
  93    }
  94
  95 #define Y_FLIP( _y )            (height - _y - 1)
  96
  97
  98 #define HW_LOCK()
  99
 100 #define HW_CLIPLOOP()                                                   \
 101    do {                                                                 \
 102       __DRIdrawablePrivate *dPriv = rmesa->dri.drawable;                \
 103       int _nc = dPriv->numClipRects;                                    \
 104                                                                         \
 105       while ( _nc-- ) {                                                 \
 106          int minx = dPriv->pClipRects[_nc].x1 - dPriv->x;               \
 107          int miny = dPriv->pClipRects[_nc].y1 - dPriv->y;               \
 108          int maxx = dPriv->pClipRects[_nc].x2 - dPriv->x;               \
 109          int maxy = dPriv->pClipRects[_nc].y2 - dPriv->y;
 110
 111 #define HW_ENDCLIPLOOP()                                                \
 112       }                                                                 \
 113    } while (0)
 114
 115 #define HW_UNLOCK()
 116
 117
 118
 119 /* ================================================================
 120  * Color buffer
 121  */
 122
 123 /* 16 bit, RGB565 color spanline and pixel functions
 124  */
 125
 126 #define GET_SRC_PTR(_x, _y) (read_buf + _x * 2 + _y * pitch)
 127 #define GET_DST_PTR(_x, _y) (     buf + _x * 2 + _y * pitch)
 128 #define SPANTMP_PIXEL_FMT GL_RGB
 129 #define SPANTMP_PIXEL_TYPE GL_UNSIGNED_SHORT_5_6_5
 130
 131 #define TAG(x)    r200##x##_RGB565
 132 #define TAG2(x,y) r200##x##_RGB565##y
 133 #include "spantmp2.h"
 134
 135 /* 32 bit, ARGB8888 color spanline and pixel functions
 136  */
 137
 138 #define GET_SRC_PTR(_x, _y) (read_buf + _x * 4 + _y * pitch)
 139 #define GET_DST_PTR(_x, _y) (     buf + _x * 4 + _y * pitch)
 140 #define SPANTMP_PIXEL_FMT GL_BGRA
 141 #define SPANTMP_PIXEL_TYPE GL_UNSIGNED_INT_8_8_8_8_REV
 142
 143 #define TAG(x)    r200##x##_ARGB8888
 144 #define TAG2(x,y) r200##x##_ARGB8888##y
 145 #include "spantmp2.h"
 146
 147
 148 /* ================================================================
 149  * Depth buffer
 150  */
 151
 152 /* The Radeon family has depth tiling on all the time, so we have to convert
 153  * the x,y coordinates into the memory bus address (mba) in the same
 154  * manner as the engine.  In each case, the linear block address (ba)
 155  * is calculated, and then wired with x and y to produce the final
 156  * memory address.
 157  */
 158
 159 #define BIT(x,b) ((x & (1<<b))>>b)
 160 static GLuint r200_mba_z32( r200ContextPtr rmesa,
 161                                        GLint x, GLint y )
 162 {
 163    GLuint pitch = rmesa->r200Screen->frontPitch;
 164    GLuint b = ((y & 0x3FF) >> 4) * ((pitch & 0xFFF) >> 5) + ((x & 0x3FF) >> 5);
 165    GLuint a =
 166       (BIT(x,0) << 2) |
 167       (BIT(y,0) << 3) |
 168       (BIT(x,1) << 4) |
 169       (BIT(y,1) << 5) |
 170       (BIT(x,3) << 6) |
 171       (BIT(x,4) << 7) |
 172       (BIT(x,2) << 8) |
 173       (BIT(y,2) << 9) |
 174       (BIT(y,3) << 10) |
 175       (((pitch & 0x20) ? (b & 0x01) : ((b & 0x01) ^ (BIT(y,4)))) << 11) |
 176       ((b >> 1) << 12);
 177    return a;
 178 }
 179
 180 static GLuint r200_mba_z16( r200ContextPtr rmesa, GLint x, GLint y )
 181 {
 182    GLuint pitch = rmesa->r200Screen->frontPitch;
 183    GLuint b = ((y & 0x3FF) >> 4) * ((pitch & 0xFFF) >> 6) + ((x & 0x3FF) >> 6);
 184    GLuint a =
 185       (BIT(x,0) << 1) |
 186       (BIT(y,0) << 2) |
 187       (BIT(x,1) << 3) |
 188       (BIT(y,1) << 4) |
 189       (BIT(x,2) << 5) |
 190       (BIT(x,4) << 6) |
 191       (BIT(x,5) << 7) |
 192       (BIT(x,3) << 8) |
 193       (BIT(y,2) << 9) |
 194       (BIT(y,3) << 10) |
 195       (((pitch & 0x40) ? (b & 0x01) : ((b & 0x01) ^ (BIT(y,4)))) << 11) |
 196       ((b >> 1) << 12);
 197    return a;
 198 }
 199
 200
 201 /* 16-bit depth buffer functions
 202  */
 203 #define WRITE_DEPTH( _x, _y, d )                                        \
 204    *(GLushort *)(buf + r200_mba_z16( rmesa, _x + xo, _y + yo )) = d;
 205
 206 #define READ_DEPTH( d, _x, _y )                                         \
 207    d = *(GLushort *)(buf + r200_mba_z16( rmesa, _x + xo, _y + yo ));
 208
 209 #define TAG(x) r200##x##_16
 210 #include "depthtmp.h"
 211
 212 /* 24 bit depth, 8 bit stencil depthbuffer functions
 213  */
 214 #define WRITE_DEPTH( _x, _y, d )                                        \
 215 do {                                                                    \
 216    GLuint offset = r200_mba_z32( rmesa, _x + xo, _y + yo );             \
 217    GLuint tmp = *(GLuint *)(buf + offset);                              \
 218    tmp &= 0xff000000;                                                   \
 219    tmp |= ((d) & 0x00ffffff);                                           \
 220    *(GLuint *)(buf + offset) = tmp;                                     \
 221 } while (0)
 222
 223 #define READ_DEPTH( d, _x, _y )                                         \
 224    d = *(GLuint *)(buf + r200_mba_z32( rmesa, _x + xo,          \
 225                                          _y + yo )) & 0x00ffffff;
 226
 227 #define TAG(x) r200##x##_24_8
 228 #include "depthtmp.h"
 229
 230
 231 /* ================================================================
 232  * Stencil buffer
 233  */
 234
 235 /* 24 bit depth, 8 bit stencil depthbuffer functions
 236  */
 237 #define WRITE_STENCIL( _x, _y, d )                                      \
 238 do {                                                                    \
 239    GLuint offset = r200_mba_z32( rmesa, _x + xo, _y + yo );             \
 240    GLuint tmp = *(GLuint *)(buf + offset);                              \
 241    tmp &= 0x00ffffff;                                                   \
 242    tmp |= (((d) & 0xff) << 24);                                         \
 243    *(GLuint *)(buf + offset) = tmp;                                     \
 244 } while (0)
 245
 246 #define READ_STENCIL( d, _x, _y )                                       \
 247 do {                                                                    \
 248    GLuint offset = r200_mba_z32( rmesa, _x + xo, _y + yo );             \
 249    GLuint tmp = *(GLuint *)(buf + offset);                              \
 250    tmp &= 0xff000000;                                                   \
 251    d = tmp >> 24;                                                       \
 252 } while (0)
 253
 254 #define TAG(x) r200##x##_24_8
 255 #include "stenciltmp.h"
 256
 257
 258 /*
 259  * This function is called to specify which buffer to read and write
 260  * for software rasterization (swrast) fallbacks.  This doesn't necessarily
 261  * correspond to glDrawBuffer() or glReadBuffer() calls.
 262  */
 263 static void r200SetBuffer( GLcontext *ctx,
 264                            GLframebuffer *colorBuffer,
 265                            GLuint bufferBit )
 266 {
 267    r200ContextPtr rmesa = R200_CONTEXT(ctx);
 268
 269    switch ( bufferBit ) {
 270    case DD_FRONT_LEFT_BIT:
 271       if ( rmesa->doPageFlip && rmesa->sarea->pfCurrentPage == 1 ) {
 272         rmesa->state.pixel.readOffset = rmesa->r200Screen->backOffset;
 273         rmesa->state.pixel.readPitch  = rmesa->r200Screen->backPitch;
 274         rmesa->state.color.drawOffset = rmesa->r200Screen->backOffset;
 275         rmesa->state.color.drawPitch  = rmesa->r200Screen->backPitch;
 276       } else {
 277         rmesa->state.pixel.readOffset = rmesa->r200Screen->frontOffset;
 278         rmesa->state.pixel.readPitch  = rmesa->r200Screen->frontPitch;
 279         rmesa->state.color.drawOffset = rmesa->r200Screen->frontOffset;
 280         rmesa->state.color.drawPitch  = rmesa->r200Screen->frontPitch;
 281       }
 282       break;
 283    case DD_BACK_LEFT_BIT:
 284       if ( rmesa->doPageFlip && rmesa->sarea->pfCurrentPage == 1 ) {
 285         rmesa->state.pixel.readOffset = rmesa->r200Screen->frontOffset;
 286         rmesa->state.pixel.readPitch  = rmesa->r200Screen->frontPitch;
 287         rmesa->state.color.drawOffset = rmesa->r200Screen->frontOffset;
 288         rmesa->state.color.drawPitch  = rmesa->r200Screen->frontPitch;
 289       } else {
 290         rmesa->state.pixel.readOffset = rmesa->r200Screen->backOffset;
 291         rmesa->state.pixel.readPitch  = rmesa->r200Screen->backPitch;
 292         rmesa->state.color.drawOffset = rmesa->r200Screen->backOffset;
 293         rmesa->state.color.drawPitch  = rmesa->r200Screen->backPitch;
 294       }
 295       break;
 296    default:
 297       _mesa_problem(ctx, "Bad bufferBit in %s", __FUNCTION__);
 298       break;
 299    }
 300 }
 301
 302 /* Move locking out to get reasonable span performance (10x better
 303  * than doing this in HW_LOCK above).  WaitForIdle() is the main
 304  * culprit.
 305  */
 306
 307 static void r200SpanRenderStart( GLcontext *ctx )
 308 {
 309    r200ContextPtr rmesa = R200_CONTEXT( ctx );
 310
 311    R200_FIREVERTICES( rmesa );
 312    LOCK_HARDWARE( rmesa );
 313    r200WaitForIdleLocked( rmesa );
 314
 315    /* Read & rewrite the first pixel in the frame buffer.  This should
 316     * be a noop, right?  In fact without this conform fails as reading
 317     * from the framebuffer sometimes produces old results -- the
 318     * on-card read cache gets mixed up and doesn't notice that the
 319     * framebuffer has been updated.
 320     *
 321     * In the worst case this is buggy too as p might get the wrong
 322     * value first time, so really need a hidden pixel somewhere for this.
 323     */
 324    {
 325       int p;
 326       volatile int *read_buf = (volatile int *)(rmesa->dri.screen->pFB +
 327                                                 rmesa->state.pixel.readOffset);
 328       p = *read_buf;
 329       *read_buf = p;
 330    }
 331 }
 332
 333 static void r200SpanRenderFinish( GLcontext *ctx )
 334 {
 335    r200ContextPtr rmesa = R200_CONTEXT( ctx );
 336    _swrast_flush( ctx );
 337    UNLOCK_HARDWARE( rmesa );
 338 }
 339
 340 void r200InitSpanFuncs( GLcontext *ctx )
 341 {
 342    r200ContextPtr rmesa = R200_CONTEXT(ctx);
 343    struct swrast_device_driver *swdd = _swrast_GetDeviceDriverReference(ctx);
 344
 345    swdd->SetBuffer = r200SetBuffer;
 346
 347    switch ( rmesa->r200Screen->cpp ) {
 348    case 2:
 349       r200InitPointers_RGB565( swdd );
 350       break;
 351
 352    case 4:
 353       r200InitPointers_ARGB8888( swdd );
 354       break;
 355
 356    default:
 357       break;
 358    }
 359
 360    switch ( rmesa->glCtx->Visual.depthBits ) {
 361    case 16:
 362       swdd->ReadDepthSpan       = r200ReadDepthSpan_16;
 363       swdd->WriteDepthSpan      = r200WriteDepthSpan_16;
 364       swdd->ReadDepthPixels     = r200ReadDepthPixels_16;
 365       swdd->WriteDepthPixels    = r200WriteDepthPixels_16;
 366       break;
 367
 368    case 24:
 369       swdd->ReadDepthSpan       = r200ReadDepthSpan_24_8;
 370       swdd->WriteDepthSpan      = r200WriteDepthSpan_24_8;
 371       swdd->ReadDepthPixels     = r200ReadDepthPixels_24_8;
 372       swdd->WriteDepthPixels    = r200WriteDepthPixels_24_8;
 373
 374       swdd->ReadStencilSpan     = r200ReadStencilSpan_24_8;
 375       swdd->WriteStencilSpan    = r200WriteStencilSpan_24_8;
 376       swdd->ReadStencilPixels   = r200ReadStencilPixels_24_8;
 377       swdd->WriteStencilPixels  = r200WriteStencilPixels_24_8;
 378       break;
 379
 380    default:
 381       break;
 382    }
 383
 384    swdd->SpanRenderStart          = r200SpanRenderStart;
 385    swdd->SpanRenderFinish         = r200SpanRenderFinish;
 386 }