radeon_span.c revision e4b2356c07d31fbeeabb13b2fb47db703b473080
1/* $XFree86: xc/lib/GL/mesa/src/drv/radeon/radeon_span.c,v 1.6 2002/10/30 12:51:56 alanh Exp $ */ 2/************************************************************************** 3 4Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and 5 VA Linux Systems Inc., Fremont, California. 6 7All Rights Reserved. 8 9Permission is hereby granted, free of charge, to any person obtaining 10a copy of this software and associated documentation files (the 11"Software"), to deal in the Software without restriction, including 12without limitation the rights to use, copy, modify, merge, publish, 13distribute, sublicense, and/or sell copies of the Software, and to 14permit persons to whom the Software is furnished to do so, subject to 15the following conditions: 16 17The above copyright notice and this permission notice (including the 18next paragraph) shall be included in all copies or substantial 19portions of the Software. 20 21THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 22EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 23MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. 24IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE 25LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 26OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 27WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. 28 29**************************************************************************/ 30 31/* 32 * Authors: 33 * Kevin E. Martin <martin@valinux.com> 34 * Gareth Hughes <gareth@valinux.com> 35 * Keith Whitwell <keith@tungstengraphics.com> 36 * 37 */ 38 39#include "glheader.h" 40#include "swrast/swrast.h" 41 42#include "radeon_context.h" 43#include "radeon_ioctl.h" 44#include "radeon_state.h" 45#include "radeon_span.h" 46#include "radeon_tex.h" 47 48#define DBG 0 49 50#define LOCAL_VARS \ 51 radeonContextPtr rmesa = RADEON_CONTEXT(ctx); \ 52 radeonScreenPtr radeonScreen = rmesa->radeonScreen; \ 53 __DRIscreenPrivate *sPriv = rmesa->dri.screen; \ 54 __DRIdrawablePrivate *dPriv = rmesa->dri.drawable; \ 55 GLuint pitch = radeonScreen->frontPitch * radeonScreen->cpp; \ 56 GLuint height = dPriv->h; \ 57 char *buf = (char *)(sPriv->pFB + \ 58 rmesa->state.color.drawOffset + \ 59 (dPriv->x * radeonScreen->cpp) + \ 60 (dPriv->y * pitch)); \ 61 char *read_buf = (char *)(sPriv->pFB + \ 62 rmesa->state.pixel.readOffset + \ 63 (dPriv->x * radeonScreen->cpp) + \ 64 (dPriv->y * pitch)); \ 65 GLuint p; \ 66 (void) read_buf; (void) buf; (void) p 67 68#define LOCAL_DEPTH_VARS \ 69 radeonContextPtr rmesa = RADEON_CONTEXT(ctx); \ 70 radeonScreenPtr radeonScreen = rmesa->radeonScreen; \ 71 __DRIscreenPrivate *sPriv = rmesa->dri.screen; \ 72 __DRIdrawablePrivate *dPriv = rmesa->dri.drawable; \ 73 GLuint height = dPriv->h; \ 74 GLuint xo = dPriv->x; \ 75 GLuint yo = dPriv->y; \ 76 char *buf = (char *)(sPriv->pFB + radeonScreen->depthOffset); \ 77 (void) buf 78 79#define LOCAL_STENCIL_VARS LOCAL_DEPTH_VARS 80 81 82#define CLIPPIXEL( _x, _y ) \ 83 ((_x >= minx) && (_x < maxx) && (_y >= miny) && (_y < maxy)) 84 85 86#define CLIPSPAN( _x, _y, _n, _x1, _n1, _i ) \ 87 if ( _y < miny || _y >= maxy ) { \ 88 _n1 = 0, _x1 = x; \ 89 } else { \ 90 _n1 = _n; \ 91 _x1 = _x; \ 92 if ( _x1 < minx ) _i += (minx-_x1), n1 -= (minx-_x1), _x1 = minx; \ 93 if ( _x1 + _n1 >= maxx ) n1 -= (_x1 + n1 - maxx); \ 94 } 95 96#define Y_FLIP( _y ) (height - _y - 1) 97 98 99#define HW_LOCK() 100 101#define HW_CLIPLOOP() \ 102 do { \ 103 __DRIdrawablePrivate *dPriv = rmesa->dri.drawable; \ 104 int _nc = dPriv->numClipRects; \ 105 \ 106 while ( _nc-- ) { \ 107 int minx = dPriv->pClipRects[_nc].x1 - dPriv->x; \ 108 int miny = dPriv->pClipRects[_nc].y1 - dPriv->y; \ 109 int maxx = dPriv->pClipRects[_nc].x2 - dPriv->x; \ 110 int maxy = dPriv->pClipRects[_nc].y2 - dPriv->y; 111 112#define HW_ENDCLIPLOOP() \ 113 } \ 114 } while (0) 115 116#define HW_UNLOCK() 117 118 119 120/* ================================================================ 121 * Color buffer 122 */ 123 124/* 16 bit, RGB565 color spanline and pixel functions 125 */ 126#define INIT_MONO_PIXEL(p, color) \ 127 p = PACK_COLOR_565( color[0], color[1], color[2] ) 128 129#define WRITE_RGBA( _x, _y, r, g, b, a ) \ 130 *(GLushort *)(buf + _x*2 + _y*pitch) = ((((int)r & 0xf8) << 8) | \ 131 (((int)g & 0xfc) << 3) | \ 132 (((int)b & 0xf8) >> 3)) 133 134#define WRITE_PIXEL( _x, _y, p ) \ 135 *(GLushort *)(buf + _x*2 + _y*pitch) = p 136 137#define READ_RGBA( rgba, _x, _y ) \ 138 do { \ 139 GLushort p = *(GLushort *)(read_buf + _x*2 + _y*pitch); \ 140 rgba[0] = ((p >> 8) & 0xf8) * 255 / 0xf8; \ 141 rgba[1] = ((p >> 3) & 0xfc) * 255 / 0xfc; \ 142 rgba[2] = ((p << 3) & 0xf8) * 255 / 0xf8; \ 143 rgba[3] = 0xff; \ 144 } while (0) 145 146#define TAG(x) radeon##x##_RGB565 147#include "spantmp.h" 148 149/* 32 bit, ARGB8888 color spanline and pixel functions 150 */ 151#undef INIT_MONO_PIXEL 152#define INIT_MONO_PIXEL(p, color) \ 153 p = PACK_COLOR_8888( color[3], color[0], color[1], color[2] ) 154 155#define WRITE_RGBA( _x, _y, r, g, b, a ) \ 156do { \ 157 *(GLuint *)(buf + _x*4 + _y*pitch) = ((b << 0) | \ 158 (g << 8) | \ 159 (r << 16) | \ 160 (a << 24) ); \ 161} while (0) 162 163#define WRITE_PIXEL( _x, _y, p ) \ 164do { \ 165 *(GLuint *)(buf + _x*4 + _y*pitch) = p; \ 166} while (0) 167 168#define READ_RGBA( rgba, _x, _y ) \ 169do { \ 170 volatile GLuint *ptr = (volatile GLuint *)(read_buf + _x*4 + _y*pitch); \ 171 GLuint p = *ptr; \ 172 rgba[0] = (p >> 16) & 0xff; \ 173 rgba[1] = (p >> 8) & 0xff; \ 174 rgba[2] = (p >> 0) & 0xff; \ 175 rgba[3] = (p >> 24) & 0xff; \ 176} while (0) 177 178#define TAG(x) radeon##x##_ARGB8888 179#include "spantmp.h" 180 181 182 183/* ================================================================ 184 * Depth buffer 185 */ 186 187/* The Radeon family has depth tiling on all the time, so we have to convert 188 * the x,y coordinates into the memory bus address (mba) in the same 189 * manner as the engine. In each case, the linear block address (ba) 190 * is calculated, and then wired with x and y to produce the final 191 * memory address. 192 * The chip will do address translation on its own if the surface registers 193 * are set up correctly. It is not quite enough to get it working with hyperz too... 194 */ 195 196static GLuint radeon_mba_z32( radeonContextPtr rmesa, 197 GLint x, GLint y ) 198{ 199 GLuint pitch = rmesa->radeonScreen->frontPitch; 200 if (rmesa->radeonScreen->depthHasSurface) { 201 return 4*(x + y*pitch); 202 } 203 else { 204 GLuint ba, address = 0; /* a[0..1] = 0 */ 205 206 ba = (y / 16) * (pitch / 16) + (x / 16); 207 208 address |= (x & 0x7) << 2; /* a[2..4] = x[0..2] */ 209 address |= (y & 0x3) << 5; /* a[5..6] = y[0..1] */ 210 address |= 211 (((x & 0x10) >> 2) ^ (y & 0x4)) << 5; /* a[7] = x[4] ^ y[2] */ 212 address |= (ba & 0x3) << 8; /* a[8..9] = ba[0..1] */ 213 214 address |= (y & 0x8) << 7; /* a[10] = y[3] */ 215 address |= 216 (((x & 0x8) << 1) ^ (y & 0x10)) << 7; /* a[11] = x[3] ^ y[4] */ 217 address |= (ba & ~0x3) << 10; /* a[12..] = ba[2..] */ 218 219 return address; 220 } 221} 222 223static __inline GLuint radeon_mba_z16( radeonContextPtr rmesa, GLint x, GLint y ) 224{ 225 GLuint pitch = rmesa->radeonScreen->frontPitch; 226 if (rmesa->radeonScreen->depthHasSurface) { 227 return 2*(x + y*pitch); 228 } 229 else { 230 GLuint ba, address = 0; /* a[0] = 0 */ 231 232 ba = (y / 16) * (pitch / 32) + (x / 32); 233 234 address |= (x & 0x7) << 1; /* a[1..3] = x[0..2] */ 235 address |= (y & 0x7) << 4; /* a[4..6] = y[0..2] */ 236 address |= (x & 0x8) << 4; /* a[7] = x[3] */ 237 address |= (ba & 0x3) << 8; /* a[8..9] = ba[0..1] */ 238 address |= (y & 0x8) << 7; /* a[10] = y[3] */ 239 address |= ((x & 0x10) ^ (y & 0x10)) << 7; /* a[11] = x[4] ^ y[4] */ 240 address |= (ba & ~0x3) << 10; /* a[12..] = ba[2..] */ 241 242 return address; 243 } 244} 245 246 247/* 16-bit depth buffer functions 248 */ 249#define WRITE_DEPTH( _x, _y, d ) \ 250 *(GLushort *)(buf + radeon_mba_z16( rmesa, _x + xo, _y + yo )) = d; 251 252#define READ_DEPTH( d, _x, _y ) \ 253 d = *(GLushort *)(buf + radeon_mba_z16( rmesa, _x + xo, _y + yo )); 254 255#define TAG(x) radeon##x##_16 256#include "depthtmp.h" 257 258/* 24 bit depth, 8 bit stencil depthbuffer functions 259 */ 260#define WRITE_DEPTH( _x, _y, d ) \ 261do { \ 262 GLuint offset = radeon_mba_z32( rmesa, _x + xo, _y + yo ); \ 263 GLuint tmp = *(GLuint *)(buf + offset); \ 264 tmp &= 0xff000000; \ 265 tmp |= ((d) & 0x00ffffff); \ 266 *(GLuint *)(buf + offset) = tmp; \ 267} while (0) 268 269#define READ_DEPTH( d, _x, _y ) \ 270 d = *(GLuint *)(buf + radeon_mba_z32( rmesa, _x + xo, \ 271 _y + yo )) & 0x00ffffff; 272 273#define TAG(x) radeon##x##_24_8 274#include "depthtmp.h" 275 276 277/* ================================================================ 278 * Stencil buffer 279 */ 280 281/* 24 bit depth, 8 bit stencil depthbuffer functions 282 */ 283#define WRITE_STENCIL( _x, _y, d ) \ 284do { \ 285 GLuint offset = radeon_mba_z32( rmesa, _x + xo, _y + yo ); \ 286 GLuint tmp = *(GLuint *)(buf + offset); \ 287 tmp &= 0x00ffffff; \ 288 tmp |= (((d) & 0xff) << 24); \ 289 *(GLuint *)(buf + offset) = tmp; \ 290} while (0) 291 292#define READ_STENCIL( d, _x, _y ) \ 293do { \ 294 GLuint offset = radeon_mba_z32( rmesa, _x + xo, _y + yo ); \ 295 GLuint tmp = *(GLuint *)(buf + offset); \ 296 tmp &= 0xff000000; \ 297 d = tmp >> 24; \ 298} while (0) 299 300#define TAG(x) radeon##x##_24_8 301#include "stenciltmp.h" 302 303 304/* 305 * This function is called to specify which buffer to read and write 306 * for software rasterization (swrast) fallbacks. This doesn't necessarily 307 * correspond to glDrawBuffer() or glReadBuffer() calls. 308 */ 309static void radeonSetBuffer( GLcontext *ctx, 310 GLframebuffer *colorBuffer, 311 GLuint bufferBit ) 312{ 313 radeonContextPtr rmesa = RADEON_CONTEXT(ctx); 314 315 switch ( bufferBit ) { 316 case BUFFER_BIT_FRONT_LEFT: 317 if ( rmesa->sarea->pfCurrentPage == 1 ) { 318 rmesa->state.pixel.readOffset = rmesa->radeonScreen->backOffset; 319 rmesa->state.pixel.readPitch = rmesa->radeonScreen->backPitch; 320 rmesa->state.color.drawOffset = rmesa->radeonScreen->backOffset; 321 rmesa->state.color.drawPitch = rmesa->radeonScreen->backPitch; 322 } else { 323 rmesa->state.pixel.readOffset = rmesa->radeonScreen->frontOffset; 324 rmesa->state.pixel.readPitch = rmesa->radeonScreen->frontPitch; 325 rmesa->state.color.drawOffset = rmesa->radeonScreen->frontOffset; 326 rmesa->state.color.drawPitch = rmesa->radeonScreen->frontPitch; 327 } 328 break; 329 case BUFFER_BIT_BACK_LEFT: 330 if ( rmesa->sarea->pfCurrentPage == 1 ) { 331 rmesa->state.pixel.readOffset = rmesa->radeonScreen->frontOffset; 332 rmesa->state.pixel.readPitch = rmesa->radeonScreen->frontPitch; 333 rmesa->state.color.drawOffset = rmesa->radeonScreen->frontOffset; 334 rmesa->state.color.drawPitch = rmesa->radeonScreen->frontPitch; 335 } else { 336 rmesa->state.pixel.readOffset = rmesa->radeonScreen->backOffset; 337 rmesa->state.pixel.readPitch = rmesa->radeonScreen->backPitch; 338 rmesa->state.color.drawOffset = rmesa->radeonScreen->backOffset; 339 rmesa->state.color.drawPitch = rmesa->radeonScreen->backPitch; 340 } 341 break; 342 default: 343 assert(0); 344 break; 345 } 346} 347 348/* Move locking out to get reasonable span performance (10x better 349 * than doing this in HW_LOCK above). WaitForIdle() is the main 350 * culprit. 351 */ 352 353static void radeonSpanRenderStart( GLcontext *ctx ) 354{ 355 radeonContextPtr rmesa = RADEON_CONTEXT( ctx ); 356 357 RADEON_FIREVERTICES( rmesa ); 358 LOCK_HARDWARE( rmesa ); 359 radeonWaitForIdleLocked( rmesa ); 360} 361 362static void radeonSpanRenderFinish( GLcontext *ctx ) 363{ 364 radeonContextPtr rmesa = RADEON_CONTEXT( ctx ); 365 _swrast_flush( ctx ); 366 UNLOCK_HARDWARE( rmesa ); 367} 368 369void radeonInitSpanFuncs( GLcontext *ctx ) 370{ 371 radeonContextPtr rmesa = RADEON_CONTEXT(ctx); 372 struct swrast_device_driver *swdd = _swrast_GetDeviceDriverReference(ctx); 373 374 swdd->SetBuffer = radeonSetBuffer; 375 376 switch ( rmesa->radeonScreen->cpp ) { 377 case 2: 378#if 0 379 swdd->WriteRGBASpan = radeonWriteRGBASpan_RGB565; 380 swdd->WriteRGBSpan = radeonWriteRGBSpan_RGB565; 381 swdd->WriteMonoRGBASpan = radeonWriteMonoRGBASpan_RGB565; 382 swdd->WriteRGBAPixels = radeonWriteRGBAPixels_RGB565; 383 swdd->WriteMonoRGBAPixels = radeonWriteMonoRGBAPixels_RGB565; 384 swdd->ReadRGBASpan = radeonReadRGBASpan_RGB565; 385 swdd->ReadRGBAPixels = radeonReadRGBAPixels_RGB565; 386#endif 387 break; 388 389 case 4: 390#if 0 391 swdd->WriteRGBASpan = radeonWriteRGBASpan_ARGB8888; 392 swdd->WriteRGBSpan = radeonWriteRGBSpan_ARGB8888; 393 swdd->WriteMonoRGBASpan = radeonWriteMonoRGBASpan_ARGB8888; 394 swdd->WriteRGBAPixels = radeonWriteRGBAPixels_ARGB8888; 395 swdd->WriteMonoRGBAPixels = radeonWriteMonoRGBAPixels_ARGB8888; 396 swdd->ReadRGBASpan = radeonReadRGBASpan_ARGB8888; 397 swdd->ReadRGBAPixels = radeonReadRGBAPixels_ARGB8888; 398#endif 399 break; 400 401 default: 402 break; 403 } 404 405 switch ( rmesa->glCtx->Visual.depthBits ) { 406 case 16: 407#if 0 408 swdd->ReadDepthSpan = radeonReadDepthSpan_16; 409 swdd->WriteDepthSpan = radeonWriteDepthSpan_16; 410 swdd->ReadDepthPixels = radeonReadDepthPixels_16; 411 swdd->WriteDepthPixels = radeonWriteDepthPixels_16; 412#endif 413 break; 414 415 case 24: 416#if 0 417 swdd->ReadDepthSpan = radeonReadDepthSpan_24_8; 418 swdd->WriteDepthSpan = radeonWriteDepthSpan_24_8; 419 swdd->ReadDepthPixels = radeonReadDepthPixels_24_8; 420 swdd->WriteDepthPixels = radeonWriteDepthPixels_24_8; 421 422 swdd->ReadStencilSpan = radeonReadStencilSpan_24_8; 423 swdd->WriteStencilSpan = radeonWriteStencilSpan_24_8; 424 swdd->ReadStencilPixels = radeonReadStencilPixels_24_8; 425 swdd->WriteStencilPixels = radeonWriteStencilPixels_24_8; 426#endif 427 break; 428 429 default: 430 break; 431 } 432 433 swdd->SpanRenderStart = radeonSpanRenderStart; 434 swdd->SpanRenderFinish = radeonSpanRenderFinish; 435} 436 437 438/** 439 * Plug in the Get/Put routines for the given driRenderbuffer. 440 */ 441void 442radeonSetSpanFunctions(driRenderbuffer *drb, const GLvisual *vis) 443{ 444 if (drb->Base.InternalFormat == GL_RGBA) { 445 if (vis->redBits == 5 && vis->greenBits == 6 && vis->blueBits == 5) { 446 drb->Base.GetRow = radeonReadRGBASpan_RGB565; 447 drb->Base.GetValues = radeonReadRGBAPixels_RGB565; 448 drb->Base.PutRow = radeonWriteRGBASpan_RGB565; 449 drb->Base.PutRowRGB = radeonWriteRGBSpan_RGB565; 450 drb->Base.PutMonoRow = radeonWriteMonoRGBASpan_RGB565; 451 drb->Base.PutValues = radeonWriteRGBAPixels_RGB565; 452 drb->Base.PutMonoValues = radeonWriteMonoRGBAPixels_RGB565; 453 } 454 else { 455 drb->Base.GetRow = radeonReadRGBASpan_ARGB8888; 456 drb->Base.GetValues = radeonReadRGBAPixels_ARGB8888; 457 drb->Base.PutRow = radeonWriteRGBASpan_ARGB8888; 458 drb->Base.PutRowRGB = radeonWriteRGBSpan_ARGB8888; 459 drb->Base.PutMonoRow = radeonWriteMonoRGBASpan_ARGB8888; 460 drb->Base.PutValues = radeonWriteRGBAPixels_ARGB8888; 461 drb->Base.PutMonoValues = radeonWriteMonoRGBAPixels_ARGB8888; 462 } 463 } 464 else if (drb->Base.InternalFormat == GL_DEPTH_COMPONENT16) { 465 drb->Base.GetRow = radeonReadDepthSpan_16; 466 drb->Base.GetValues = radeonReadDepthPixels_16; 467 drb->Base.PutRow = radeonWriteDepthSpan_16; 468 drb->Base.PutMonoRow = radeonWriteMonoDepthSpan_16; 469 drb->Base.PutValues = radeonWriteDepthPixels_16; 470 drb->Base.PutMonoValues = NULL; 471 } 472 else if (drb->Base.InternalFormat == GL_DEPTH_COMPONENT24) { 473 drb->Base.GetRow = radeonReadDepthSpan_24_8; 474 drb->Base.GetValues = radeonReadDepthPixels_24_8; 475 drb->Base.PutRow = radeonWriteDepthSpan_24_8; 476 drb->Base.PutMonoRow = radeonWriteMonoDepthSpan_24_8; 477 drb->Base.PutValues = radeonWriteDepthPixels_24_8; 478 drb->Base.PutMonoValues = NULL; 479 } 480 else if (drb->Base.InternalFormat == GL_STENCIL_INDEX8_EXT) { 481 drb->Base.GetRow = radeonReadStencilSpan_24_8; 482 drb->Base.GetValues = radeonReadStencilPixels_24_8; 483 drb->Base.PutRow = radeonWriteStencilSpan_24_8; 484 drb->Base.PutMonoRow = radeonWriteMonoStencilSpan_24_8; 485 drb->Base.PutValues = radeonWriteStencilPixels_24_8; 486 drb->Base.PutMonoValues = NULL; 487 } 488} 489