radeon: this code forgot to add dPriv x/y back to spans
[mesa.git] / src / mesa / drivers / dri / radeon / common_context.h
1
2 #ifndef COMMON_CONTEXT_H
3 #define COMMON_CONTEXT_H
4
5 #include "main/mm.h"
6 #include "math/m_vector.h"
7 #include "texmem.h"
8 #include "tnl/t_context.h"
9 #include "main/colormac.h"
10
11 #include "radeon_screen.h"
12 #include "radeon_drm.h"
13 #include "dri_util.h"
14 #include "tnl/t_vertex.h"
15
16 /* This union is used to avoid warnings/miscompilation
17 with float to uint32_t casts due to strict-aliasing */
18 typedef union { GLfloat f; uint32_t ui32; } float_ui32_type;
19
20 struct radeon_context;
21 typedef struct radeon_context radeonContextRec;
22 typedef struct radeon_context *radeonContextPtr;
23
24
25 #define TEX_0 0x1
26 #define TEX_1 0x2
27 #define TEX_2 0x4
28 #define TEX_3 0x8
29 #define TEX_4 0x10
30 #define TEX_5 0x20
31
32 /* Rasterizing fallbacks */
33 /* See correponding strings in r200_swtcl.c */
34 #define RADEON_FALLBACK_TEXTURE 0x0001
35 #define RADEON_FALLBACK_DRAW_BUFFER 0x0002
36 #define RADEON_FALLBACK_STENCIL 0x0004
37 #define RADEON_FALLBACK_RENDER_MODE 0x0008
38 #define RADEON_FALLBACK_BLEND_EQ 0x0010
39 #define RADEON_FALLBACK_BLEND_FUNC 0x0020
40 #define RADEON_FALLBACK_DISABLE 0x0040
41 #define RADEON_FALLBACK_BORDER_MODE 0x0080
42
43 #define R200_FALLBACK_TEXTURE 0x01
44 #define R200_FALLBACK_DRAW_BUFFER 0x02
45 #define R200_FALLBACK_STENCIL 0x04
46 #define R200_FALLBACK_RENDER_MODE 0x08
47 #define R200_FALLBACK_DISABLE 0x10
48 #define R200_FALLBACK_BORDER_MODE 0x20
49
50 #define RADEON_TCL_FALLBACK_RASTER 0x1 /* rasterization */
51 #define RADEON_TCL_FALLBACK_UNFILLED 0x2 /* unfilled tris */
52 #define RADEON_TCL_FALLBACK_LIGHT_TWOSIDE 0x4 /* twoside tris */
53 #define RADEON_TCL_FALLBACK_MATERIAL 0x8 /* material in vb */
54 #define RADEON_TCL_FALLBACK_TEXGEN_0 0x10 /* texgen, unit 0 */
55 #define RADEON_TCL_FALLBACK_TEXGEN_1 0x20 /* texgen, unit 1 */
56 #define RADEON_TCL_FALLBACK_TEXGEN_2 0x40 /* texgen, unit 2 */
57 #define RADEON_TCL_FALLBACK_TCL_DISABLE 0x80 /* user disable */
58 #define RADEON_TCL_FALLBACK_FOGCOORDSPEC 0x100 /* fogcoord, sep. spec light */
59
60 /* The blit width for texture uploads
61 */
62 #define BLIT_WIDTH_BYTES 1024
63
64 /* Use the templated vertex format:
65 */
66 #define COLOR_IS_RGBA
67 #define TAG(x) radeon##x
68 #include "tnl_dd/t_dd_vertex.h"
69 #undef TAG
70
71 struct radeon_colorbuffer_state {
72 GLuint clear;
73 int roundEnable;
74 struct radeon_renderbuffer *rrb;
75 };
76
77 struct radeon_depthbuffer_state {
78 GLuint clear;
79 GLfloat scale;
80 struct radeon_renderbuffer *rrb;
81 };
82
83 struct radeon_scissor_state {
84 drm_clip_rect_t rect;
85 GLboolean enabled;
86
87 GLuint numClipRects; /* Cliprects active */
88 GLuint numAllocedClipRects; /* Cliprects available */
89 drm_clip_rect_t *pClipRects;
90 };
91
92 struct radeon_stencilbuffer_state {
93 GLboolean hwBuffer;
94 GLuint clear; /* rb3d_stencilrefmask value */
95 };
96
97 struct radeon_stipple_state {
98 GLuint mask[32];
99 };
100
101 struct radeon_state_atom {
102 struct radeon_state_atom *next, *prev;
103 const char *name; /* for debug */
104 int cmd_size; /* size in bytes */
105 GLuint idx;
106 GLuint is_tcl;
107 GLuint *cmd; /* one or more cmd's */
108 GLuint *lastcmd; /* one or more cmd's */
109 GLboolean dirty; /* dirty-mark in emit_state_list */
110 int (*check) (GLcontext *, struct radeon_state_atom *atom); /* is this state active? */
111 void (*emit) (GLcontext *, struct radeon_state_atom *atom);
112 };
113
114
115 /* Texture related */
116 typedef struct _radeon_texture_image radeon_texture_image;
117
118 struct _radeon_texture_image {
119 struct gl_texture_image base;
120
121 /**
122 * If mt != 0, the image is stored in hardware format in the
123 * given mipmap tree. In this case, base.Data may point into the
124 * mapping of the buffer object that contains the mipmap tree.
125 *
126 * If mt == 0, the image is stored in normal memory pointed to
127 * by base.Data.
128 */
129 struct _radeon_mipmap_tree *mt;
130 struct radeon_bo *bo;
131
132 int mtlevel; /** if mt != 0, this is the image's level in the mipmap tree */
133 int mtface; /** if mt != 0, this is the image's face in the mipmap tree */
134 };
135
136
137 static INLINE radeon_texture_image *get_radeon_texture_image(struct gl_texture_image *image)
138 {
139 return (radeon_texture_image*)image;
140 }
141
142
143 typedef struct radeon_tex_obj radeonTexObj, *radeonTexObjPtr;
144
145 #define RADEON_TXO_MICRO_TILE (1 << 3)
146
147 /* Texture object in locally shared texture space.
148 */
149 struct radeon_tex_obj {
150 // driTextureObject base;
151 struct gl_texture_object base;
152 struct _radeon_mipmap_tree *mt;
153
154 /**
155 * This is true if we've verified that the mipmap tree above is complete
156 * and so on.
157 */
158 GLboolean validated;
159
160 GLuint override_offset;
161 GLboolean image_override; /* Image overridden by GLX_EXT_tfp */
162 GLuint tile_bits; /* hw texture tile bits used on this texture */
163 struct radeon_bo *bo;
164
165 GLuint bufAddr; /* Offset to start of locally
166 shared texture block */
167
168 GLuint dirty_state; /* Flags (1 per texunit) for
169 whether or not this texobj
170 has dirty hardware state
171 (pp_*) that needs to be
172 brought into the
173 texunit. */
174
175 drm_radeon_tex_image_t image[6][RADEON_MAX_TEXTURE_LEVELS];
176 /* Six, for the cube faces */
177
178
179
180 GLuint pp_txfilter; /* hardware register values */
181 GLuint pp_txformat;
182 GLuint pp_txformat_x;
183 GLuint pp_txoffset; /* Image location in texmem.
184 All cube faces follow. */
185 GLuint pp_txsize; /* npot only */
186 GLuint pp_txpitch; /* npot only */
187 GLuint pp_border_color;
188 GLuint pp_cubic_faces; /* cube face 1,2,3,4 log2 sizes */
189
190 GLuint pp_txfilter_1; /* r300 */
191
192 GLboolean border_fallback;
193
194
195 };
196
197 static INLINE radeonTexObj* radeon_tex_obj(struct gl_texture_object *texObj)
198 {
199 return (radeonTexObj*)texObj;
200 }
201
202 /* Need refcounting on dma buffers:
203 */
204 struct radeon_dma_buffer {
205 int refcount; /* the number of retained regions in buf */
206 drmBufPtr buf;
207 };
208
209 /* A retained region, eg vertices for indexed vertices.
210 */
211 struct radeon_dma_region {
212 struct radeon_dma_buffer *buf;
213 char *address; /* == buf->address */
214 int start, end, ptr; /* offsets from start of buf */
215 int aos_start;
216 int aos_stride;
217 int aos_size;
218 };
219
220 struct radeon_aos {
221 struct radeon_bo *bo; /** Buffer object where vertex data is stored */
222 int offset; /** Offset into buffer object, in bytes */
223 int components; /** Number of components per vertex */
224 int stride; /** Stride in dwords (may be 0 for repeating) */
225 int count; /** Number of vertices */
226 };
227
228 struct radeon_dma {
229 /* Active dma region. Allocations for vertices and retained
230 * regions come from here. Also used for emitting random vertices,
231 * these may be flushed by calling flush_current();
232 */
233 struct radeon_bo *current; /** Buffer that DMA memory is allocated from */
234 int current_used; /** Number of bytes allocated and forgotten about */
235 int current_vertexptr; /** End of active vertex region */
236
237 /**
238 * If current_vertexptr != current_used then flush must be non-zero.
239 * flush must be called before non-active vertex allocations can be
240 * performed.
241 */
242 void (*flush) (GLcontext *);
243
244 /* Number of "in-flight" DMA buffers, i.e. the number of buffers
245 * for which a DISCARD command is currently queued in the command buffer
246 .
247 */
248 GLuint nr_released_bufs;
249 };
250
251 /* radeon_swtcl.c
252 */
253 struct radeon_swtcl_info {
254
255 GLuint RenderIndex;
256 GLuint vertex_size;
257 GLubyte *verts;
258
259 /* Fallback rasterization functions
260 */
261 GLuint hw_primitive;
262 GLenum render_primitive;
263 GLuint numverts;
264
265 struct tnl_attr_map vertex_attrs[VERT_ATTRIB_MAX];
266 GLuint vertex_attr_count;
267
268 };
269
270 struct radeon_ioctl {
271 GLuint vertex_offset;
272 GLuint vertex_size;
273 };
274
275 #define RADEON_MAX_PRIMS 64
276
277 struct radeon_prim {
278 GLuint start;
279 GLuint end;
280 GLuint prim;
281 };
282
283 static INLINE GLuint radeonPackColor(GLuint cpp,
284 GLubyte r, GLubyte g,
285 GLubyte b, GLubyte a)
286 {
287 switch (cpp) {
288 case 2:
289 return PACK_COLOR_565(r, g, b);
290 case 4:
291 return PACK_COLOR_8888(a, r, g, b);
292 default:
293 return 0;
294 }
295 }
296
297 #define MAX_CMD_BUF_SZ (16*1024)
298
299 #define MAX_DMA_BUF_SZ (64*1024)
300
301 struct radeon_store {
302 GLuint statenr;
303 GLuint primnr;
304 char cmd_buf[MAX_CMD_BUF_SZ];
305 int cmd_used;
306 int elts_start;
307 };
308
309 struct radeon_dri_mirror {
310 __DRIcontextPrivate *context; /* DRI context */
311 __DRIscreenPrivate *screen; /* DRI screen */
312
313 /**
314 * DRI drawable bound to this context for drawing.
315 */
316 __DRIdrawablePrivate *drawable;
317
318 /**
319 * DRI drawable bound to this context for reading.
320 */
321 __DRIdrawablePrivate *readable;
322
323 drm_context_t hwContext;
324 drm_hw_lock_t *hwLock;
325 int fd;
326 int drmMinor;
327 };
328
329 #define DEBUG_TEXTURE 0x001
330 #define DEBUG_STATE 0x002
331 #define DEBUG_IOCTL 0x004
332 #define DEBUG_PRIMS 0x008
333 #define DEBUG_VERTS 0x010
334 #define DEBUG_FALLBACKS 0x020
335 #define DEBUG_VFMT 0x040
336 #define DEBUG_CODEGEN 0x080
337 #define DEBUG_VERBOSE 0x100
338 #define DEBUG_DRI 0x200
339 #define DEBUG_DMA 0x400
340 #define DEBUG_SANITY 0x800
341 #define DEBUG_SYNC 0x1000
342 #define DEBUG_PIXEL 0x2000
343 #define DEBUG_MEMORY 0x4000
344
345
346
347 typedef void (*radeon_tri_func) (radeonContextPtr,
348 radeonVertex *,
349 radeonVertex *, radeonVertex *);
350
351 typedef void (*radeon_line_func) (radeonContextPtr,
352 radeonVertex *, radeonVertex *);
353
354 typedef void (*radeon_point_func) (radeonContextPtr, radeonVertex *);
355
356 struct radeon_state {
357 struct radeon_colorbuffer_state color;
358 struct radeon_depthbuffer_state depth;
359 struct radeon_scissor_state scissor;
360 struct radeon_stencilbuffer_state stencil;
361 };
362
363 /**
364 * This structure holds the command buffer while it is being constructed.
365 *
366 * The first batch of commands in the buffer is always the state that needs
367 * to be re-emitted when the context is lost. This batch can be skipped
368 * otherwise.
369 */
370 struct radeon_cmdbuf {
371 struct radeon_cs_manager *csm;
372 struct radeon_cs *cs;
373 int size; /** # of dwords total */
374 unsigned int flushing:1; /** whether we're currently in FlushCmdBufLocked */
375 };
376
377 struct radeon_context {
378 GLcontext *glCtx;
379 radeonScreenPtr radeonScreen; /* Screen private DRI data */
380
381 /* Texture object bookkeeping
382 */
383 unsigned nr_heaps;
384 driTexHeap * texture_heaps[ RADEON_NR_TEX_HEAPS ];
385 driTextureObject swapped;
386 int texture_depth;
387 float initialMaxAnisotropy;
388
389 struct radeon_dma dma;
390 /* Rasterization and vertex state:
391 */
392 GLuint TclFallback;
393 GLuint Fallback;
394 GLuint NewGLState;
395 DECLARE_RENDERINPUTS(tnl_index_bitset); /* index of bits for last tnl_install_attrs */
396
397 /* Page flipping */
398 GLuint doPageFlip;
399
400 /* Drawable, cliprect and scissor information */
401 GLuint numClipRects; /* Cliprects for the draw buffer */
402 drm_clip_rect_t *pClipRects;
403 unsigned int lastStamp;
404 GLboolean lost_context;
405 drm_radeon_sarea_t *sarea; /* Private SAREA data */
406
407 /* Mirrors of some DRI state */
408 struct radeon_dri_mirror dri;
409
410 /* Busy waiting */
411 GLuint do_usleeps;
412 GLuint do_irqs;
413 GLuint irqsEmitted;
414 drm_radeon_irq_wait_t iw;
415
416 /* buffer swap */
417 int64_t swap_ust;
418 int64_t swap_missed_ust;
419
420 GLuint swap_count;
421 GLuint swap_missed_count;
422
423 /* Derived state - for r300 only */
424 struct radeon_state state;
425
426 struct radeon_swtcl_info swtcl;
427 /* Configuration cache
428 */
429 driOptionCache optionCache;
430
431 struct radeon_cmdbuf cmdbuf;
432
433 struct {
434 void (*get_lock)(radeonContextPtr radeon);
435 void (*update_viewport_offset)(GLcontext *ctx);
436 void (*flush)(GLcontext *ctx);
437 void (*set_all_dirty)(GLcontext *ctx);
438 void (*update_draw_buffer)(GLcontext *ctx);
439 void (*emit_cs_header)(struct radeon_cs *cs, radeonContextPtr rmesa);
440 void (*emit_state)(radeonContextPtr rmesa);
441 void (*flush_vertices)(radeonContextPtr rmesa);
442 void (*swtcl_flush)(GLcontext *ctx, uint32_t offset);
443 } vtbl;
444 };
445
446 #define RADEON_CONTEXT(glctx) ((radeonContextPtr)(ctx->DriverCtx))
447
448 /**
449 * This function takes a float and packs it into a uint32_t
450 */
451 static INLINE uint32_t radeonPackFloat32(float fl)
452 {
453 union {
454 float fl;
455 uint32_t u;
456 } u;
457
458 u.fl = fl;
459 return u.u;
460 }
461
462 /* This is probably wrong for some values, I need to test this
463 * some more. Range checking would be a good idea also..
464 *
465 * But it works for most things. I'll fix it later if someone
466 * else with a better clue doesn't
467 */
468 static INLINE uint32_t radeonPackFloat24(float f)
469 {
470 float mantissa;
471 int exponent;
472 uint32_t float24 = 0;
473
474 if (f == 0.0)
475 return 0;
476
477 mantissa = frexpf(f, &exponent);
478
479 /* Handle -ve */
480 if (mantissa < 0) {
481 float24 |= (1 << 23);
482 mantissa = mantissa * -1.0;
483 }
484 /* Handle exponent, bias of 63 */
485 exponent += 62;
486 float24 |= (exponent << 16);
487 /* Kill 7 LSB of mantissa */
488 float24 |= (radeonPackFloat32(mantissa) & 0x7FFFFF) >> 7;
489
490 return float24;
491 }
492
493 /* ================================================================
494 * Debugging:
495 */
496 #define DO_DEBUG 1
497
498 #if DO_DEBUG
499 extern int RADEON_DEBUG;
500 #else
501 #define RADEON_DEBUG 0
502 #endif
503
504 #endif