src/mesa/drivers/dri/radeon/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_debug.h"
  12 #include "radeon_screen.h"
  13 #include "radeon_drm.h"
  14 #include "dri_util.h"
  15 #include "tnl/t_vertex.h"
  16
  17 #include "dri_metaops.h"
  18 struct radeon_context;
  19
  20 #include "radeon_bocs_wrapper.h"
  21
  22 /* This union is used to avoid warnings/miscompilation
  23    with float to uint32_t casts due to strict-aliasing */
  24 typedef union { GLfloat f; uint32_t ui32; } float_ui32_type;
  25
  26 struct radeon_context;
  27 typedef struct radeon_context radeonContextRec;
  28 typedef struct radeon_context *radeonContextPtr;
  29
  30
  31 #define TEX_0   0x1
  32 #define TEX_1   0x2
  33 #define TEX_2   0x4
  34 #define TEX_3   0x8
  35 #define TEX_4   0x10
  36 #define TEX_5   0x20
  37
  38 /* Rasterizing fallbacks */
  39 /* See correponding strings in r200_swtcl.c */
  40 #define RADEON_FALLBACK_TEXTURE         0x0001
  41 #define RADEON_FALLBACK_DRAW_BUFFER     0x0002
  42 #define RADEON_FALLBACK_STENCIL         0x0004
  43 #define RADEON_FALLBACK_RENDER_MODE     0x0008
  44 #define RADEON_FALLBACK_BLEND_EQ        0x0010
  45 #define RADEON_FALLBACK_BLEND_FUNC      0x0020
  46 #define RADEON_FALLBACK_DISABLE         0x0040
  47 #define RADEON_FALLBACK_BORDER_MODE     0x0080
  48 #define RADEON_FALLBACK_DEPTH_BUFFER    0x0100
  49 #define RADEON_FALLBACK_STENCIL_BUFFER  0x0200
  50
  51 #define R200_FALLBACK_TEXTURE           0x01
  52 #define R200_FALLBACK_DRAW_BUFFER       0x02
  53 #define R200_FALLBACK_STENCIL           0x04
  54 #define R200_FALLBACK_RENDER_MODE       0x08
  55 #define R200_FALLBACK_DISABLE           0x10
  56 #define R200_FALLBACK_BORDER_MODE       0x20
  57
  58 #define RADEON_TCL_FALLBACK_RASTER            0x1 /* rasterization */
  59 #define RADEON_TCL_FALLBACK_UNFILLED          0x2 /* unfilled tris */
  60 #define RADEON_TCL_FALLBACK_LIGHT_TWOSIDE     0x4 /* twoside tris */
  61 #define RADEON_TCL_FALLBACK_MATERIAL          0x8 /* material in vb */
  62 #define RADEON_TCL_FALLBACK_TEXGEN_0          0x10 /* texgen, unit 0 */
  63 #define RADEON_TCL_FALLBACK_TEXGEN_1          0x20 /* texgen, unit 1 */
  64 #define RADEON_TCL_FALLBACK_TEXGEN_2          0x40 /* texgen, unit 2 */
  65 #define RADEON_TCL_FALLBACK_TCL_DISABLE       0x80 /* user disable */
  66 #define RADEON_TCL_FALLBACK_FOGCOORDSPEC      0x100 /* fogcoord, sep. spec light */
  67
  68 /* The blit width for texture uploads
  69  */
  70 #define BLIT_WIDTH_BYTES 1024
  71
  72 /* Use the templated vertex format:
  73  */
  74 #define COLOR_IS_RGBA
  75 #define TAG(x) radeon##x
  76 #include "tnl_dd/t_dd_vertex.h"
  77 #undef TAG
  78
  79 #define RADEON_RB_CLASS 0xdeadbeef
  80
  81 struct radeon_renderbuffer
  82 {
  83         struct gl_renderbuffer base;
  84         struct radeon_bo *bo;
  85         unsigned int cpp;
  86         /* unsigned int offset; */
  87         unsigned int pitch;
  88
  89         uint32_t draw_offset; /* FBO */
  90         /* boo Xorg 6.8.2 compat */
  91         int has_surface;
  92
  93         GLuint pf_pending;  /**< sequence number of pending flip */
  94         GLuint vbl_pending;   /**< vblank sequence number of pending flip */
  95         __DRIdrawablePrivate *dPriv;
  96 };
  97
  98 struct radeon_framebuffer
  99 {
 100         struct gl_framebuffer base;
 101
 102         struct radeon_renderbuffer *color_rb[2];
 103
 104         GLuint vbl_waited;
 105
 106         /* buffer swap */
 107         int64_t swap_ust;
 108         int64_t swap_missed_ust;
 109
 110         GLuint swap_count;
 111         GLuint swap_missed_count;
 112
 113         /* Drawable page flipping state */
 114         GLboolean pf_active;
 115         GLint pf_current_page;
 116         GLint pf_num_pages;
 117
 118 };
 119
 120
 121 struct radeon_colorbuffer_state {
 122         GLuint clear;
 123         int roundEnable;
 124         struct gl_renderbuffer *rb;
 125         uint32_t draw_offset; /* offset into color renderbuffer - FBOs */
 126 };
 127
 128 struct radeon_depthbuffer_state {
 129         GLuint clear;
 130         struct gl_renderbuffer *rb;
 131 };
 132
 133 struct radeon_scissor_state {
 134         drm_clip_rect_t rect;
 135         GLboolean enabled;
 136
 137         GLuint numClipRects;    /* Cliprects active */
 138         GLuint numAllocedClipRects;     /* Cliprects available */
 139         drm_clip_rect_t *pClipRects;
 140 };
 141
 142 struct radeon_stencilbuffer_state {
 143         GLuint clear;           /* rb3d_stencilrefmask value */
 144 };
 145
 146 struct radeon_state_atom {
 147         struct radeon_state_atom *next, *prev;
 148         const char *name;       /* for debug */
 149         int cmd_size;           /* size in bytes */
 150         GLuint idx;
 151         GLuint is_tcl;
 152         GLuint *cmd;            /* one or more cmd's */
 153         GLuint *lastcmd;                /* one or more cmd's */
 154         GLboolean dirty;        /* dirty-mark in emit_state_list */
 155         int (*check) (GLcontext *, struct radeon_state_atom *atom); /* is this state active? */
 156         void (*emit) (GLcontext *, struct radeon_state_atom *atom);
 157 };
 158
 159 struct radeon_hw_state {
 160         /* Head of the linked list of state atoms. */
 161         struct radeon_state_atom atomlist;
 162         int max_state_size;     /* Number of bytes necessary for a full state emit. */
 163         int max_post_flush_size; /* Number of bytes necessary for post flushing emits */
 164         GLboolean is_dirty, all_dirty;
 165 };
 166
 167
 168 /* Texture related */
 169 typedef struct _radeon_texture_image radeon_texture_image;
 170
 171 struct _radeon_texture_image {
 172         struct gl_texture_image base;
 173
 174         /**
 175          * If mt != 0, the image is stored in hardware format in the
 176          * given mipmap tree. In this case, base.Data may point into the
 177          * mapping of the buffer object that contains the mipmap tree.
 178          *
 179          * If mt == 0, the image is stored in normal memory pointed to
 180          * by base.Data.
 181          */
 182         struct _radeon_mipmap_tree *mt;
 183         struct radeon_bo *bo;
 184
 185         int mtlevel; /** if mt != 0, this is the image's level in the mipmap tree */
 186         int mtface; /** if mt != 0, this is the image's face in the mipmap tree */
 187 };
 188
 189
 190 static INLINE radeon_texture_image *get_radeon_texture_image(struct gl_texture_image *image)
 191 {
 192         return (radeon_texture_image*)image;
 193 }
 194
 195
 196 typedef struct radeon_tex_obj radeonTexObj, *radeonTexObjPtr;
 197
 198 #define RADEON_TXO_MICRO_TILE               (1 << 3)
 199
 200 /* Texture object in locally shared texture space.
 201  */
 202 struct radeon_tex_obj {
 203         struct gl_texture_object base;
 204         struct _radeon_mipmap_tree *mt;
 205
 206         /**
 207          * This is true if we've verified that the mipmap tree above is complete
 208          * and so on.
 209          */
 210         GLboolean validated;
 211         /* Minimum LOD to be used during rendering */
 212         unsigned minLod;
 213         /* Miximum LOD to be used during rendering */
 214         unsigned maxLod;
 215
 216         GLuint override_offset;
 217         GLboolean image_override; /* Image overridden by GLX_EXT_tfp */
 218         GLuint tile_bits;       /* hw texture tile bits used on this texture */
 219         struct radeon_bo *bo;
 220
 221         GLuint pp_txfilter;     /* hardware register values */
 222         GLuint pp_txformat;
 223         GLuint pp_txformat_x;
 224         GLuint pp_txsize;       /* npot only */
 225         GLuint pp_txpitch;      /* npot only */
 226         GLuint pp_border_color;
 227         GLuint pp_cubic_faces;  /* cube face 1,2,3,4 log2 sizes */
 228
 229         GLuint pp_txfilter_1;   /*  r300 */
 230
 231         /* r700 texture states */
 232         GLuint SQ_TEX_RESOURCE0;
 233         GLuint SQ_TEX_RESOURCE1;
 234         GLuint SQ_TEX_RESOURCE2;
 235         GLuint SQ_TEX_RESOURCE3;
 236         GLuint SQ_TEX_RESOURCE4;
 237         GLuint SQ_TEX_RESOURCE5;
 238         GLuint SQ_TEX_RESOURCE6;
 239
 240         GLuint SQ_TEX_SAMPLER0;
 241         GLuint SQ_TEX_SAMPLER1;
 242         GLuint SQ_TEX_SAMPLER2;
 243
 244         GLuint TD_PS_SAMPLER0_BORDER_RED;
 245         GLuint TD_PS_SAMPLER0_BORDER_GREEN;
 246         GLuint TD_PS_SAMPLER0_BORDER_BLUE;
 247         GLuint TD_PS_SAMPLER0_BORDER_ALPHA;
 248
 249         GLboolean border_fallback;
 250
 251
 252 };
 253
 254 static INLINE radeonTexObj* radeon_tex_obj(struct gl_texture_object *texObj)
 255 {
 256         return (radeonTexObj*)texObj;
 257 }
 258
 259 /* occlusion query */
 260 struct radeon_query_object {
 261         struct gl_query_object Base;
 262         struct radeon_bo *bo;
 263         int curr_offset;
 264         GLboolean emitted_begin;
 265
 266         /* Double linked list of not flushed query objects */
 267         struct radeon_query_object *prev, *next;
 268 };
 269
 270 /* Need refcounting on dma buffers:
 271  */
 272 struct radeon_dma_buffer {
 273         int refcount;           /* the number of retained regions in buf */
 274         drmBufPtr buf;
 275 };
 276
 277 struct radeon_aos {
 278         struct radeon_bo *bo; /** Buffer object where vertex data is stored */
 279         int offset; /** Offset into buffer object, in bytes */
 280         int components; /** Number of components per vertex */
 281         int stride; /** Stride in dwords (may be 0 for repeating) */
 282         int count; /** Number of vertices */
 283 };
 284
 285 #define DMA_BO_FREE_TIME 100
 286
 287 struct radeon_dma_bo {
 288   struct radeon_dma_bo *next, *prev;
 289   struct radeon_bo *bo;
 290   int expire_counter;
 291 };
 292
 293 struct radeon_dma {
 294         /* Active dma region.  Allocations for vertices and retained
 295          * regions come from here.  Also used for emitting random vertices,
 296          * these may be flushed by calling flush_current();
 297          */
 298         struct radeon_dma_bo free;
 299         struct radeon_dma_bo wait;
 300         struct radeon_dma_bo reserved;
 301         size_t current_used; /** Number of bytes allocated and forgotten about */
 302         size_t current_vertexptr; /** End of active vertex region */
 303         size_t minimum_size;
 304
 305         /**
 306          * If current_vertexptr != current_used then flush must be non-zero.
 307          * flush must be called before non-active vertex allocations can be
 308          * performed.
 309          */
 310         void (*flush) (GLcontext *);
 311 };
 312
 313 /* radeon_swtcl.c
 314  */
 315 struct radeon_swtcl_info {
 316
 317         GLuint RenderIndex;
 318         GLuint vertex_size;
 319         GLubyte *verts;
 320
 321         /* Fallback rasterization functions
 322          */
 323         GLuint hw_primitive;
 324         GLenum render_primitive;
 325         GLuint numverts;
 326
 327         struct tnl_attr_map vertex_attrs[VERT_ATTRIB_MAX];
 328         GLuint vertex_attr_count;
 329
 330         GLuint emit_prediction;
 331         struct radeon_bo *bo;
 332 };
 333
 334 #define RADEON_MAX_AOS_ARRAYS           16
 335 struct radeon_tcl_info {
 336         struct radeon_aos aos[RADEON_MAX_AOS_ARRAYS];
 337         GLuint aos_count;
 338         struct radeon_bo *elt_dma_bo; /** Buffer object that contains element indices */
 339         int elt_dma_offset; /** Offset into this buffer object, in bytes */
 340 };
 341
 342 struct radeon_ioctl {
 343         GLuint vertex_offset;
 344         GLuint vertex_max;
 345         struct radeon_bo *bo;
 346         GLuint vertex_size;
 347 };
 348
 349 #define RADEON_MAX_PRIMS 64
 350
 351 struct radeon_prim {
 352         GLuint start;
 353         GLuint end;
 354         GLuint prim;
 355 };
 356
 357 static INLINE GLuint radeonPackColor(GLuint cpp,
 358                                      GLubyte r, GLubyte g,
 359                                      GLubyte b, GLubyte a)
 360 {
 361         switch (cpp) {
 362         case 2:
 363                 return PACK_COLOR_565(r, g, b);
 364         case 4:
 365                 return PACK_COLOR_8888(a, r, g, b);
 366         default:
 367                 return 0;
 368         }
 369 }
 370
 371 #define MAX_CMD_BUF_SZ (16*1024)
 372
 373 #define MAX_DMA_BUF_SZ (64*1024)
 374
 375 struct radeon_store {
 376         GLuint statenr;
 377         GLuint primnr;
 378         char cmd_buf[MAX_CMD_BUF_SZ];
 379         int cmd_used;
 380         int elts_start;
 381 };
 382
 383 struct radeon_dri_mirror {
 384         __DRIcontextPrivate *context;   /* DRI context */
 385         __DRIscreenPrivate *screen;     /* DRI screen */
 386
 387         drm_context_t hwContext;
 388         drm_hw_lock_t *hwLock;
 389         int hwLockCount;
 390         int fd;
 391         int drmMinor;
 392 };
 393
 394 typedef void (*radeon_tri_func) (radeonContextPtr,
 395                                  radeonVertex *,
 396                                  radeonVertex *, radeonVertex *);
 397
 398 typedef void (*radeon_line_func) (radeonContextPtr,
 399                                   radeonVertex *, radeonVertex *);
 400
 401 typedef void (*radeon_point_func) (radeonContextPtr, radeonVertex *);
 402
 403 #define RADEON_MAX_BOS 32
 404 struct radeon_state {
 405         struct radeon_colorbuffer_state color;
 406         struct radeon_depthbuffer_state depth;
 407         struct radeon_scissor_state scissor;
 408         struct radeon_stencilbuffer_state stencil;
 409
 410         struct radeon_cs_space_check bos[RADEON_MAX_BOS];
 411         int validated_bo_count;
 412 };
 413
 414 /**
 415  * This structure holds the command buffer while it is being constructed.
 416  *
 417  * The first batch of commands in the buffer is always the state that needs
 418  * to be re-emitted when the context is lost. This batch can be skipped
 419  * otherwise.
 420  */
 421 struct radeon_cmdbuf {
 422         struct radeon_cs_manager    *csm;
 423         struct radeon_cs            *cs;
 424         int size; /** # of dwords total */
 425         unsigned int flushing:1; /** whether we're currently in FlushCmdBufLocked */
 426 };
 427
 428 struct radeon_context {
 429    GLcontext *glCtx;
 430    radeonScreenPtr radeonScreen;        /* Screen private DRI data */
 431
 432    /* Texture object bookkeeping
 433     */
 434    int                   texture_depth;
 435    float                 initialMaxAnisotropy;
 436    uint32_t              texture_row_align;
 437    uint32_t              texture_rect_row_align;
 438    uint32_t              texture_compressed_row_align;
 439
 440   struct radeon_dma dma;
 441   struct radeon_hw_state hw;
 442    /* Rasterization and vertex state:
 443     */
 444    GLuint TclFallback;
 445    GLuint Fallback;
 446    GLuint NewGLState;
 447    DECLARE_RENDERINPUTS(tnl_index_bitset);      /* index of bits for last tnl_install_attrs */
 448
 449    /* Drawable, cliprect and scissor information */
 450    GLuint numClipRects; /* Cliprects for the draw buffer */
 451    drm_clip_rect_t *pClipRects;
 452    unsigned int lastStamp;
 453    drm_radeon_sarea_t *sarea;   /* Private SAREA data */
 454
 455    /* Mirrors of some DRI state */
 456    struct radeon_dri_mirror dri;
 457
 458    /* Busy waiting */
 459    GLuint do_usleeps;
 460    GLuint do_irqs;
 461    GLuint irqsEmitted;
 462    drm_radeon_irq_wait_t iw;
 463
 464    /* Derived state - for r300 only */
 465    struct radeon_state state;
 466
 467    struct radeon_swtcl_info swtcl;
 468    struct radeon_tcl_info tcl;
 469    /* Configuration cache
 470     */
 471    driOptionCache optionCache;
 472
 473    struct radeon_cmdbuf cmdbuf;
 474
 475    struct radeon_debug debug;
 476
 477   drm_clip_rect_t fboRect;
 478   GLboolean constant_cliprect; /* use for FBO or DRI2 rendering */
 479   GLboolean front_cliprects;
 480
 481    /**
 482     * Set if rendering has occured to the drawable's front buffer.
 483     *
 484     * This is used in the DRI2 case to detect that glFlush should also copy
 485     * the contents of the fake front buffer to the real front buffer.
 486     */
 487    GLboolean front_buffer_dirty;
 488
 489    /**
 490     * Track whether front-buffer rendering is currently enabled
 491     *
 492     * A separate flag is used to track this in order to support MRT more
 493     * easily.
 494     */
 495    GLboolean is_front_buffer_rendering;
 496
 497    /**
 498     * Track whether front-buffer is the current read target.
 499     *
 500     * This is closely associated with is_front_buffer_rendering, but may
 501     * be set separately.  The DRI2 fake front buffer must be referenced
 502     * either way.
 503     */
 504    GLboolean is_front_buffer_reading;
 505
 506    struct dri_metaops meta;
 507
 508    struct {
 509         struct radeon_query_object *current;
 510         struct radeon_state_atom queryobj;
 511    } query;
 512
 513    struct {
 514            void (*get_lock)(radeonContextPtr radeon);
 515            void (*update_viewport_offset)(GLcontext *ctx);
 516            void (*emit_cs_header)(struct radeon_cs *cs, radeonContextPtr rmesa);
 517            void (*swtcl_flush)(GLcontext *ctx, uint32_t offset);
 518            void (*pre_emit_atoms)(radeonContextPtr rmesa);
 519            void (*pre_emit_state)(radeonContextPtr rmesa);
 520            void (*fallback)(GLcontext *ctx, GLuint bit, GLboolean mode);
 521            void (*free_context)(GLcontext *ctx);
 522            void (*emit_query_finish)(radeonContextPtr radeon);
 523            void (*update_scissor)(GLcontext *ctx);
 524    } vtbl;
 525 };
 526
 527 #define RADEON_CONTEXT(glctx) ((radeonContextPtr)(ctx->DriverCtx))
 528
 529 static inline __DRIdrawablePrivate* radeon_get_drawable(radeonContextPtr radeon)
 530 {
 531         return radeon->dri.context->driDrawablePriv;
 532 }
 533
 534 static inline __DRIdrawablePrivate* radeon_get_readable(radeonContextPtr radeon)
 535 {
 536         return radeon->dri.context->driReadablePriv;
 537 }
 538
 539 /**
 540  * This function takes a float and packs it into a uint32_t
 541  */
 542 static INLINE uint32_t radeonPackFloat32(float fl)
 543 {
 544         union {
 545                 float fl;
 546                 uint32_t u;
 547         } u;
 548
 549         u.fl = fl;
 550         return u.u;
 551 }
 552
 553 /* This is probably wrong for some values, I need to test this
 554  * some more.  Range checking would be a good idea also..
 555  *
 556  * But it works for most things.  I'll fix it later if someone
 557  * else with a better clue doesn't
 558  */
 559 static INLINE uint32_t radeonPackFloat24(float f)
 560 {
 561         float mantissa;
 562         int exponent;
 563         uint32_t float24 = 0;
 564
 565         if (f == 0.0)
 566                 return 0;
 567
 568         mantissa = frexpf(f, &exponent);
 569
 570         /* Handle -ve */
 571         if (mantissa < 0) {
 572                 float24 |= (1 << 23);
 573                 mantissa = mantissa * -1.0;
 574         }
 575         /* Handle exponent, bias of 63 */
 576         exponent += 62;
 577         float24 |= (exponent << 16);
 578         /* Kill 7 LSB of mantissa */
 579         float24 |= (radeonPackFloat32(mantissa) & 0x7FFFFF) >> 7;
 580
 581         return float24;
 582 }
 583
 584 GLboolean radeonInitContext(radeonContextPtr radeon,
 585                             struct dd_function_table* functions,
 586                             const __GLcontextModes * glVisual,
 587                             __DRIcontextPrivate * driContextPriv,
 588                             void *sharedContextPrivate);
 589
 590 void radeonCleanupContext(radeonContextPtr radeon);
 591 GLboolean radeonUnbindContext(__DRIcontextPrivate * driContextPriv);
 592 void radeon_update_renderbuffers(__DRIcontext *context, __DRIdrawable *drawable,
 593                                  GLboolean front_only);
 594 GLboolean radeonMakeCurrent(__DRIcontextPrivate * driContextPriv,
 595                             __DRIdrawablePrivate * driDrawPriv,
 596                             __DRIdrawablePrivate * driReadPriv);
 597 extern void radeonDestroyContext(__DRIcontextPrivate * driContextPriv);
 598
 599 #endif