1 /**************************************************************************
3 * Copyright 2009 VMware, Inc.
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:
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
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.
26 **************************************************************************/
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
33 * We'll do that with multiple threads...
40 #include "pipe/p_compiler.h"
49 #define FIXED_TYPE_WIDTH 64
50 /** For sub-pixel positioning */
52 #define FIXED_ONE (1<<FIXED_ORDER)
53 #define FIXED_SHIFT (FIXED_TYPE_WIDTH - 1)
54 /** Maximum length of an edge in a primitive in pixels.
55 * If the framebuffer is large we have to think about fixed-point
56 * integer overflow. Coordinates need ((FIXED_TYPE_WIDTH/2) - 1) bits
57 * to be able to fit product of two such coordinates inside
58 * FIXED_TYPE_WIDTH, any larger and we could overflow a
59 * FIXED_TYPE_WIDTH_-bit int.
61 #define MAX_FIXED_LENGTH (1 << (((FIXED_TYPE_WIDTH/2) - 1) - FIXED_ORDER))
63 #define MAX_FIXED_LENGTH32 (1 << (((32/2) - 1) - FIXED_ORDER))
65 /* Rasterizer output size going to jit fs, width/height */
66 #define LP_RASTER_BLOCK_SIZE 4
68 #define LP_MAX_ACTIVE_BINNED_QUERIES 16
70 #define IMUL64(a, b) (((int64_t)(a)) * ((int64_t)(b)))
72 struct lp_rasterizer_task
;
76 * Rasterization state.
77 * Objects of this type are put into the shared data bin and pointed
78 * to by commands in the per-tile bins.
80 struct lp_rast_state
{
81 /* State for the shader. This also contains state which feeds into
82 * the fragment shader, such as blend color and alpha ref value.
84 struct lp_jit_context jit_context
;
86 /* The shader itself. Probably we also need to pass a pointer to
87 * the tile color/z/stencil data somehow
89 struct lp_fragment_shader_variant
*variant
;
94 * Coefficients necessary to run the shader at a given location.
95 * First coefficient is position.
96 * These pointers point into the bin data buffer.
98 struct lp_rast_shader_inputs
{
99 unsigned frontfacing
:1; /** True for front-facing */
100 unsigned disable
:1; /** Partially binned, disable this command */
101 unsigned opaque
:1; /** Is opaque */
102 unsigned pad0
:29; /* wasted space */
103 unsigned stride
; /* how much to advance data between a0, dadx, dady */
104 unsigned layer
; /* the layer to render to (from gs, already clamped) */
105 unsigned pad2
; /* wasted space */
106 /* followed by a0, dadx, dady and planes[] */
109 struct lp_rast_plane
{
110 /* edge function values at minx,miny ?? */
116 /* one-pixel sized trivial reject offsets for each plane */
121 * Rasterization information for a triangle known to be in this bin,
122 * plus inputs to run the shader:
123 * These fields are tile- and bin-independent.
124 * Objects of this type are put into the lp_setup_context::data buffer.
126 struct lp_rast_triangle
{
133 /* inputs for the shader */
134 struct lp_rast_shader_inputs inputs
;
135 /* planes are also allocated here */
139 #define GET_A0(inputs) ((float (*)[4])((inputs)+1))
140 #define GET_DADX(inputs) ((float (*)[4])((char *)((inputs) + 1) + (inputs)->stride))
141 #define GET_DADY(inputs) ((float (*)[4])((char *)((inputs) + 1) + 2 * (inputs)->stride))
142 #define GET_PLANES(tri) ((struct lp_rast_plane *)((char *)(&(tri)->inputs + 1) + 3 * (tri)->inputs.stride))
146 struct lp_rasterizer
*
147 lp_rast_create( unsigned num_threads
);
150 lp_rast_destroy( struct lp_rasterizer
* );
153 lp_rast_queue_scene( struct lp_rasterizer
*rast
,
154 struct lp_scene
*scene
);
157 lp_rast_finish( struct lp_rasterizer
*rast
);
160 union lp_rast_cmd_arg
{
161 const struct lp_rast_shader_inputs
*shade_tile
;
163 const struct lp_rast_triangle
*tri
;
166 const struct lp_rast_state
*set_state
;
167 union pipe_color_union clear_color
;
172 const struct lp_rast_state
*state
;
173 struct lp_fence
*fence
;
174 struct llvmpipe_query
*query_obj
;
178 /* Cast wrappers. Hopefully these compile to noops!
180 static INLINE
union lp_rast_cmd_arg
181 lp_rast_arg_inputs( const struct lp_rast_shader_inputs
*shade_tile
)
183 union lp_rast_cmd_arg arg
;
184 arg
.shade_tile
= shade_tile
;
188 static INLINE
union lp_rast_cmd_arg
189 lp_rast_arg_triangle( const struct lp_rast_triangle
*triangle
,
192 union lp_rast_cmd_arg arg
;
193 arg
.triangle
.tri
= triangle
;
194 arg
.triangle
.plane_mask
= plane_mask
;
199 * Build argument for a contained triangle.
201 * All planes are enabled, so instead of the plane mask we pass the upper
202 * left coordinates of the a block that fully encloses the triangle.
204 static INLINE
union lp_rast_cmd_arg
205 lp_rast_arg_triangle_contained( const struct lp_rast_triangle
*triangle
,
206 unsigned x
, unsigned y
)
208 union lp_rast_cmd_arg arg
;
209 arg
.triangle
.tri
= triangle
;
210 arg
.triangle
.plane_mask
= x
| (y
<< 8);
214 static INLINE
union lp_rast_cmd_arg
215 lp_rast_arg_state( const struct lp_rast_state
*state
)
217 union lp_rast_cmd_arg arg
;
218 arg
.set_state
= state
;
222 static INLINE
union lp_rast_cmd_arg
223 lp_rast_arg_fence( struct lp_fence
*fence
)
225 union lp_rast_cmd_arg arg
;
231 static INLINE
union lp_rast_cmd_arg
232 lp_rast_arg_clearzs( uint64_t value
, uint64_t mask
)
234 union lp_rast_cmd_arg arg
;
235 arg
.clear_zstencil
.value
= value
;
236 arg
.clear_zstencil
.mask
= mask
;
241 static INLINE
union lp_rast_cmd_arg
242 lp_rast_arg_query( struct llvmpipe_query
*pq
)
244 union lp_rast_cmd_arg arg
;
249 static INLINE
union lp_rast_cmd_arg
250 lp_rast_arg_null( void )
252 union lp_rast_cmd_arg arg
;
253 arg
.set_state
= NULL
;
260 * These get put into bins by the setup code and are called when
261 * the bins are executed.
263 #define LP_RAST_OP_CLEAR_COLOR 0x0
264 #define LP_RAST_OP_CLEAR_ZSTENCIL 0x1
265 #define LP_RAST_OP_TRIANGLE_1 0x2
266 #define LP_RAST_OP_TRIANGLE_2 0x3
267 #define LP_RAST_OP_TRIANGLE_3 0x4
268 #define LP_RAST_OP_TRIANGLE_4 0x5
269 #define LP_RAST_OP_TRIANGLE_5 0x6
270 #define LP_RAST_OP_TRIANGLE_6 0x7
271 #define LP_RAST_OP_TRIANGLE_7 0x8
272 #define LP_RAST_OP_TRIANGLE_8 0x9
273 #define LP_RAST_OP_TRIANGLE_3_4 0xa
274 #define LP_RAST_OP_TRIANGLE_3_16 0xb
275 #define LP_RAST_OP_TRIANGLE_4_16 0xc
276 #define LP_RAST_OP_SHADE_TILE 0xd
277 #define LP_RAST_OP_SHADE_TILE_OPAQUE 0xe
278 #define LP_RAST_OP_BEGIN_QUERY 0xf
279 #define LP_RAST_OP_END_QUERY 0x10
280 #define LP_RAST_OP_SET_STATE 0x11
281 #define LP_RAST_OP_TRIANGLE_32_1 0x12
282 #define LP_RAST_OP_TRIANGLE_32_2 0x13
283 #define LP_RAST_OP_TRIANGLE_32_3 0x14
284 #define LP_RAST_OP_TRIANGLE_32_4 0x15
285 #define LP_RAST_OP_TRIANGLE_32_5 0x16
286 #define LP_RAST_OP_TRIANGLE_32_6 0x17
287 #define LP_RAST_OP_TRIANGLE_32_7 0x18
288 #define LP_RAST_OP_TRIANGLE_32_8 0x19
289 #define LP_RAST_OP_TRIANGLE_32_3_4 0x1a
290 #define LP_RAST_OP_TRIANGLE_32_3_16 0x1b
291 #define LP_RAST_OP_TRIANGLE_32_4_16 0x1c
293 #define LP_RAST_OP_MAX 0x1d
294 #define LP_RAST_OP_MASK 0xff
297 lp_debug_bins( struct lp_scene
*scene
);
299 lp_debug_draw_bins_by_cmd_length( struct lp_scene
*scene
);
301 lp_debug_draw_bins_by_coverage( struct lp_scene
*scene
);
305 #include <emmintrin.h>
306 #include "util/u_sse.h"
308 static INLINE __m128i
309 lp_plane_to_m128i(const struct lp_rast_plane
*plane
)
311 return _mm_setr_epi32((int32_t)plane
->c
, (int32_t)plane
->dcdx
,
312 (int32_t)plane
->dcdy
, (int32_t)plane
->eo
);