src/gallium/drivers/llvmpipe/lp_rast.h

   1 /**************************************************************************
   2  *
   3  * Copyright 2009 VMware, Inc.
   4  * All Rights Reserved.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the
   8  * "Software"), to deal in the Software without restriction, including
   9  * without limitation the rights to use, copy, modify, merge, publish,
  10  * distribute, sub license, and/or sell copies of the Software, and to
  11  * permit persons to whom the Software is furnished to do so, subject to
  12  * the following conditions:
  13  *
  14  * The above copyright notice and this permission notice (including the
  15  * next paragraph) shall be included in all copies or substantial portions
  16  * of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  19  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  21  * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
  22  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  23  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  24  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25  *
  26  **************************************************************************/
  27
  28 /**
  29  * The rast code is concerned with rasterization of command bins.
  30  * Each screen tile has a bin associated with it.  To render the
  31  * scene we iterate over the tile bins and execute the commands
  32  * in each bin.
  33  * We'll do that with multiple threads...
  34  */
  35
  36
  37 #ifndef LP_RAST_H
  38 #define LP_RAST_H
  39
  40 #include "pipe/p_compiler.h"
  41 #include "util/u_pack_color.h"
  42 #include "lp_jit.h"
  43
  44
  45 struct lp_rasterizer;
  46 struct lp_scene;
  47 struct lp_fence;
  48 struct cmd_bin;
  49
  50 #define FIXED_TYPE_WIDTH 64
  51 /** For sub-pixel positioning */
  52 #define FIXED_ORDER 8
  53 #define FIXED_ONE (1<<FIXED_ORDER)
  54 #define FIXED_SHIFT (FIXED_TYPE_WIDTH - 1)
  55 /** Maximum length of an edge in a primitive in pixels.
  56  *  If the framebuffer is large we have to think about fixed-point
  57  *  integer overflow. Coordinates need ((FIXED_TYPE_WIDTH/2) - 1) bits
  58  *  to be able to fit product of two such coordinates inside
  59  *  FIXED_TYPE_WIDTH, any larger and we could overflow a
  60  *  FIXED_TYPE_WIDTH_-bit int.
  61  */
  62 #define MAX_FIXED_LENGTH (1 << (((FIXED_TYPE_WIDTH/2) - 1) - FIXED_ORDER))
  63
  64 #define MAX_FIXED_LENGTH32 (1 << (((32/2) - 1) - FIXED_ORDER))
  65
  66 /* Rasterizer output size going to jit fs, width/height */
  67 #define LP_RASTER_BLOCK_SIZE 4
  68
  69 #define LP_MAX_ACTIVE_BINNED_QUERIES 64
  70
  71 #define IMUL64(a, b) (((int64_t)(a)) * ((int64_t)(b)))
  72
  73 struct lp_rasterizer_task;
  74
  75 extern const float lp_sample_pos_4x[4][2];
  76
  77 /**
  78  * Rasterization state.
  79  * Objects of this type are put into the shared data bin and pointed
  80  * to by commands in the per-tile bins.
  81  */
  82 struct lp_rast_state {
  83    /* State for the shader.  This also contains state which feeds into
  84     * the fragment shader, such as blend color and alpha ref value.
  85     */
  86    struct lp_jit_context jit_context;
  87
  88    /* The shader itself.  Probably we also need to pass a pointer to
  89     * the tile color/z/stencil data somehow
  90      */
  91    struct lp_fragment_shader_variant *variant;
  92 };
  93
  94
  95 /**
  96  * Coefficients necessary to run the shader at a given location.
  97  * First coefficient is position.
  98  * These pointers point into the bin data buffer.
  99  */
 100 struct lp_rast_shader_inputs {
 101    unsigned frontfacing:1;      /** True for front-facing */
 102    unsigned disable:1;          /** Partially binned, disable this command */
 103    unsigned opaque:1;           /** Is opaque */
 104    unsigned pad0:29;            /* wasted space */
 105    unsigned stride;             /* how much to advance data between a0, dadx, dady */
 106    unsigned layer;              /* the layer to render to (from gs, already clamped) */
 107    unsigned viewport_index;     /* the active viewport index (from gs, already clamped) */
 108    /* followed by a0, dadx, dady and planes[] */
 109 };
 110
 111 struct lp_rast_plane {
 112    /* edge function values at minx,miny ?? */
 113    int64_t c;
 114
 115    int32_t dcdx;
 116    int32_t dcdy;
 117
 118    /* one-pixel sized trivial reject offsets for each plane */
 119    uint32_t eo;
 120    /*
 121     * We rely on this struct being 64bit aligned (ideally it would be 128bit
 122     * but that's quite the waste) and therefore on 32bit we need padding
 123     * since otherwise (even with the 64bit number in there) it wouldn't be.
 124     */
 125    uint32_t pad;
 126 };
 127
 128 /**
 129  * Rasterization information for a triangle known to be in this bin,
 130  * plus inputs to run the shader:
 131  * These fields are tile- and bin-independent.
 132  * Objects of this type are put into the lp_setup_context::data buffer.
 133  */
 134 struct lp_rast_triangle {
 135 #ifdef DEBUG
 136    float v[3][2];
 137    float pad0;
 138    float pad1;
 139 #endif
 140
 141    /* inputs for the shader */
 142    struct lp_rast_shader_inputs inputs;
 143    /* planes are also allocated here */
 144 };
 145
 146
 147 struct lp_rast_clear_rb {
 148    union util_color color_val;
 149    unsigned cbuf;
 150 };
 151
 152
 153 #define GET_A0(inputs) ((float (*)[4])((inputs)+1))
 154 #define GET_DADX(inputs) ((float (*)[4])((char *)((inputs) + 1) + (inputs)->stride))
 155 #define GET_DADY(inputs) ((float (*)[4])((char *)((inputs) + 1) + 2 * (inputs)->stride))
 156 #define GET_PLANES(tri) ((struct lp_rast_plane *)((char *)(&(tri)->inputs + 1) + 3 * (tri)->inputs.stride))
 157
 158
 159
 160 struct lp_rasterizer *
 161 lp_rast_create( unsigned num_threads );
 162
 163 void
 164 lp_rast_destroy( struct lp_rasterizer * );
 165
 166 void
 167 lp_rast_queue_scene( struct lp_rasterizer *rast,
 168                      struct lp_scene *scene );
 169
 170 void
 171 lp_rast_finish( struct lp_rasterizer *rast );
 172
 173
 174 union lp_rast_cmd_arg {
 175    const struct lp_rast_shader_inputs *shade_tile;
 176    struct {
 177       const struct lp_rast_triangle *tri;
 178       unsigned plane_mask;
 179    } triangle;
 180    const struct lp_rast_state *set_state;
 181    const struct lp_rast_clear_rb *clear_rb;
 182    struct {
 183       uint64_t value;
 184       uint64_t mask;
 185    } clear_zstencil;
 186    const struct lp_rast_state *state;
 187    struct lp_fence *fence;
 188    struct llvmpipe_query *query_obj;
 189 };
 190
 191
 192 /* Cast wrappers.  Hopefully these compile to noops!
 193  */
 194 static inline union lp_rast_cmd_arg
 195 lp_rast_arg_inputs( const struct lp_rast_shader_inputs *shade_tile )
 196 {
 197    union lp_rast_cmd_arg arg;
 198    arg.shade_tile = shade_tile;
 199    return arg;
 200 }
 201
 202 static inline union lp_rast_cmd_arg
 203 lp_rast_arg_triangle( const struct lp_rast_triangle *triangle,
 204                       unsigned plane_mask)
 205 {
 206    union lp_rast_cmd_arg arg;
 207    arg.triangle.tri = triangle;
 208    arg.triangle.plane_mask = plane_mask;
 209    return arg;
 210 }
 211
 212 /**
 213  * Build argument for a contained triangle.
 214  *
 215  * All planes are enabled, so instead of the plane mask we pass the upper
 216  * left coordinates of the a block that fully encloses the triangle.
 217  */
 218 static inline union lp_rast_cmd_arg
 219 lp_rast_arg_triangle_contained( const struct lp_rast_triangle *triangle,
 220                                 unsigned x, unsigned y)
 221 {
 222    union lp_rast_cmd_arg arg;
 223    arg.triangle.tri = triangle;
 224    arg.triangle.plane_mask = x | (y << 8);
 225    return arg;
 226 }
 227
 228 static inline union lp_rast_cmd_arg
 229 lp_rast_arg_state( const struct lp_rast_state *state )
 230 {
 231    union lp_rast_cmd_arg arg;
 232    arg.set_state = state;
 233    return arg;
 234 }
 235
 236 static inline union lp_rast_cmd_arg
 237 lp_rast_arg_fence( struct lp_fence *fence )
 238 {
 239    union lp_rast_cmd_arg arg;
 240    arg.fence = fence;
 241    return arg;
 242 }
 243
 244
 245 static inline union lp_rast_cmd_arg
 246 lp_rast_arg_clearzs( uint64_t value, uint64_t mask )
 247 {
 248    union lp_rast_cmd_arg arg;
 249    arg.clear_zstencil.value = value;
 250    arg.clear_zstencil.mask = mask;
 251    return arg;
 252 }
 253
 254
 255 static inline union lp_rast_cmd_arg
 256 lp_rast_arg_query( struct llvmpipe_query *pq )
 257 {
 258    union lp_rast_cmd_arg arg;
 259    arg.query_obj = pq;
 260    return arg;
 261 }
 262
 263 static inline union lp_rast_cmd_arg
 264 lp_rast_arg_null( void )
 265 {
 266    union lp_rast_cmd_arg arg;
 267    arg.set_state = NULL;
 268    return arg;
 269 }
 270
 271
 272 /**
 273  * Binnable Commands.
 274  * These get put into bins by the setup code and are called when
 275  * the bins are executed.
 276  */
 277 #define LP_RAST_OP_CLEAR_COLOR       0x0
 278 #define LP_RAST_OP_CLEAR_ZSTENCIL    0x1
 279 #define LP_RAST_OP_TRIANGLE_1        0x2
 280 #define LP_RAST_OP_TRIANGLE_2        0x3
 281 #define LP_RAST_OP_TRIANGLE_3        0x4
 282 #define LP_RAST_OP_TRIANGLE_4        0x5
 283 #define LP_RAST_OP_TRIANGLE_5        0x6
 284 #define LP_RAST_OP_TRIANGLE_6        0x7
 285 #define LP_RAST_OP_TRIANGLE_7        0x8
 286 #define LP_RAST_OP_TRIANGLE_8        0x9
 287 #define LP_RAST_OP_TRIANGLE_3_4      0xa
 288 #define LP_RAST_OP_TRIANGLE_3_16     0xb
 289 #define LP_RAST_OP_TRIANGLE_4_16     0xc
 290 #define LP_RAST_OP_SHADE_TILE        0xd
 291 #define LP_RAST_OP_SHADE_TILE_OPAQUE 0xe
 292 #define LP_RAST_OP_BEGIN_QUERY       0xf
 293 #define LP_RAST_OP_END_QUERY         0x10
 294 #define LP_RAST_OP_SET_STATE         0x11
 295 #define LP_RAST_OP_TRIANGLE_32_1     0x12
 296 #define LP_RAST_OP_TRIANGLE_32_2     0x13
 297 #define LP_RAST_OP_TRIANGLE_32_3     0x14
 298 #define LP_RAST_OP_TRIANGLE_32_4     0x15
 299 #define LP_RAST_OP_TRIANGLE_32_5     0x16
 300 #define LP_RAST_OP_TRIANGLE_32_6     0x17
 301 #define LP_RAST_OP_TRIANGLE_32_7     0x18
 302 #define LP_RAST_OP_TRIANGLE_32_8     0x19
 303 #define LP_RAST_OP_TRIANGLE_32_3_4   0x1a
 304 #define LP_RAST_OP_TRIANGLE_32_3_16  0x1b
 305 #define LP_RAST_OP_TRIANGLE_32_4_16  0x1c
 306
 307 #define LP_RAST_OP_MS_TRIANGLE_1     0x1d
 308 #define LP_RAST_OP_MS_TRIANGLE_2     0x1e
 309 #define LP_RAST_OP_MS_TRIANGLE_3     0x1f
 310 #define LP_RAST_OP_MS_TRIANGLE_4     0x20
 311 #define LP_RAST_OP_MS_TRIANGLE_5     0x21
 312 #define LP_RAST_OP_MS_TRIANGLE_6     0x22
 313 #define LP_RAST_OP_MS_TRIANGLE_7     0x23
 314 #define LP_RAST_OP_MS_TRIANGLE_8     0x24
 315 #define LP_RAST_OP_MS_TRIANGLE_3_4   0x25
 316 #define LP_RAST_OP_MS_TRIANGLE_3_16  0x26
 317 #define LP_RAST_OP_MS_TRIANGLE_4_16  0x27
 318 #define LP_RAST_OP_MAX               0x28
 319 #define LP_RAST_OP_MASK              0xff
 320
 321 void
 322 lp_debug_bins( struct lp_scene *scene );
 323 void
 324 lp_debug_draw_bins_by_cmd_length( struct lp_scene *scene );
 325 void
 326 lp_debug_draw_bins_by_coverage( struct lp_scene *scene );
 327
 328
 329 #endif