cell: implement fencing for texture buffers
[mesa.git] / src / gallium / drivers / cell / ppu / cell_vbuf.c
1 /**************************************************************************
2 *
3 * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
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 TUNGSTEN GRAPHICS 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 * Vertex buffer code. The draw module transforms vertices to window
30 * coords, etc. and emits the vertices into buffer supplied by this module.
31 * When a vertex buffer is full, or we flush, we'll send the vertex data
32 * to the SPUs.
33 *
34 * Authors
35 * Brian Paul
36 */
37
38
39 #include "cell_batch.h"
40 #include "cell_context.h"
41 #include "cell_fence.h"
42 #include "cell_flush.h"
43 #include "cell_spu.h"
44 #include "cell_vbuf.h"
45 #include "draw/draw_vbuf.h"
46 #include "util/u_memory.h"
47
48
49 /** Allow vertex data to be inlined after RENDER command */
50 #define ALLOW_INLINE_VERTS 1
51
52
53 /**
54 * Subclass of vbuf_render because we need a cell_context pointer in
55 * a few places.
56 */
57 struct cell_vbuf_render
58 {
59 struct vbuf_render base;
60 struct cell_context *cell;
61 uint prim; /**< PIPE_PRIM_x */
62 uint vertex_size; /**< in bytes */
63 void *vertex_buffer; /**< just for debug, really */
64 uint vertex_buf; /**< in [0, CELL_NUM_BUFFERS-1] */
65 };
66
67
68 /** cast wrapper */
69 static struct cell_vbuf_render *
70 cell_vbuf_render(struct vbuf_render *vbr)
71 {
72 return (struct cell_vbuf_render *) vbr;
73 }
74
75
76
77 static const struct vertex_info *
78 cell_vbuf_get_vertex_info(struct vbuf_render *vbr)
79 {
80 struct cell_vbuf_render *cvbr = cell_vbuf_render(vbr);
81 return &cvbr->cell->vertex_info;
82 }
83
84
85 static void *
86 cell_vbuf_allocate_vertices(struct vbuf_render *vbr,
87 ushort vertex_size, ushort nr_vertices)
88 {
89 struct cell_vbuf_render *cvbr = cell_vbuf_render(vbr);
90 /*printf("Alloc verts %u * %u\n", vertex_size, nr_vertices);*/
91
92 assert(cvbr->vertex_buf == ~0);
93 cvbr->vertex_buf = cell_get_empty_buffer(cvbr->cell);
94 cvbr->vertex_buffer = cvbr->cell->buffer[cvbr->vertex_buf];
95 cvbr->vertex_size = vertex_size;
96 return cvbr->vertex_buffer;
97 }
98
99
100 static void
101 cell_vbuf_release_vertices(struct vbuf_render *vbr, void *vertices,
102 unsigned vertex_size, unsigned vertices_used)
103 {
104 struct cell_vbuf_render *cvbr = cell_vbuf_render(vbr);
105 struct cell_context *cell = cvbr->cell;
106
107 /*
108 printf("%s vertex_buf = %u count = %u\n",
109 __FUNCTION__, cvbr->vertex_buf, vertices_used);
110 */
111
112 /* Make sure texture buffers aren't released until we're done rendering
113 * with them.
114 */
115 cell_add_fenced_textures(cell);
116
117 /* Tell SPUs they can release the vert buf */
118 if (cvbr->vertex_buf != ~0U) {
119 struct cell_command_release_verts *release
120 = (struct cell_command_release_verts *)
121 cell_batch_alloc(cell, sizeof(struct cell_command_release_verts));
122 release->opcode = CELL_CMD_RELEASE_VERTS;
123 release->vertex_buf = cvbr->vertex_buf;
124 }
125
126 cvbr->vertex_buf = ~0;
127 cell_flush_int(cell, 0x0);
128
129 assert(vertices == cvbr->vertex_buffer);
130 cvbr->vertex_buffer = NULL;
131 }
132
133
134
135 static boolean
136 cell_vbuf_set_primitive(struct vbuf_render *vbr, unsigned prim)
137 {
138 struct cell_vbuf_render *cvbr = cell_vbuf_render(vbr);
139 cvbr->prim = prim;
140 /*printf("cell_set_prim %u\n", prim);*/
141 return TRUE;
142 }
143
144
145 static void
146 cell_vbuf_draw(struct vbuf_render *vbr,
147 const ushort *indices,
148 uint nr_indices)
149 {
150 struct cell_vbuf_render *cvbr = cell_vbuf_render(vbr);
151 struct cell_context *cell = cvbr->cell;
152 float xmin, ymin, xmax, ymax;
153 uint i;
154 uint nr_vertices = 0, min_index = ~0;
155 const void *vertices = cvbr->vertex_buffer;
156 const uint vertex_size = cvbr->vertex_size;
157
158 for (i = 0; i < nr_indices; i++) {
159 if (indices[i] > nr_vertices)
160 nr_vertices = indices[i];
161 if (indices[i] < min_index)
162 min_index = indices[i];
163 }
164 nr_vertices++;
165
166 #if 0
167 /*if (min_index > 0)*/
168 printf("%s min_index = %u\n", __FUNCTION__, min_index);
169 #endif
170
171 #if 0
172 printf("cell_vbuf_draw() nr_indices = %u nr_verts = %u\n",
173 nr_indices, nr_vertices);
174 printf(" ");
175 for (i = 0; i < nr_indices; i += 3) {
176 printf("%u %u %u, ", indices[i+0], indices[i+1], indices[i+2]);
177 }
178 printf("\n");
179 #elif 0
180 printf("cell_vbuf_draw() nr_indices = %u nr_verts = %u indexes = [%u %u %u ...]\n",
181 nr_indices, nr_vertices,
182 indices[0], indices[1], indices[2]);
183 printf("ind space = %u, vert space = %u, space = %u\n",
184 nr_indices * 2,
185 nr_vertices * 4 * cell->vertex_info.size,
186 cell_batch_free_space(cell));
187 #endif
188
189 /* compute x/y bounding box */
190 xmin = ymin = 1e50;
191 xmax = ymax = -1e50;
192 for (i = min_index; i < nr_vertices; i++) {
193 const float *v = (float *) ((ubyte *) vertices + i * vertex_size);
194 if (v[0] < xmin)
195 xmin = v[0];
196 if (v[0] > xmax)
197 xmax = v[0];
198 if (v[1] < ymin)
199 ymin = v[1];
200 if (v[1] > ymax)
201 ymax = v[1];
202 }
203 #if 0
204 printf("PPU Bounds %g, %g .. %g, %g\n", xmin, ymin, xmax, ymax);
205 fflush(stdout);
206 #endif
207
208 if (cvbr->prim != PIPE_PRIM_TRIANGLES)
209 return; /* only render tris for now */
210
211 /* build/insert batch RENDER command */
212 {
213 const uint index_bytes = ROUNDUP8(nr_indices * 2);
214 const uint vertex_bytes = nr_vertices * 4 * cell->vertex_info.size;
215 const uint batch_size = sizeof(struct cell_command_render) + index_bytes;
216
217 struct cell_command_render *render
218 = (struct cell_command_render *)
219 cell_batch_alloc(cell, batch_size);
220
221 render->opcode = CELL_CMD_RENDER;
222 render->prim_type = cvbr->prim;
223
224 render->num_indexes = nr_indices;
225 render->min_index = min_index;
226
227 /* append indices after render command */
228 memcpy(render + 1, indices, nr_indices * 2);
229
230 /* if there's room, append vertices after the indices, else leave
231 * vertices in the original/separate buffer.
232 */
233 render->vertex_size = 4 * cell->vertex_info.size;
234 render->num_verts = nr_vertices;
235 if (ALLOW_INLINE_VERTS &&
236 min_index == 0 &&
237 vertex_bytes + 16 <= cell_batch_free_space(cell)) {
238 /* vertex data inlined, after indices, at 16-byte boundary */
239 void *dst = cell_batch_alloc_aligned(cell, vertex_bytes, 16);
240 memcpy(dst, vertices, vertex_bytes);
241 render->inline_verts = TRUE;
242 render->vertex_buf = ~0;
243 }
244 else {
245 /* vertex data in separate buffer */
246 render->inline_verts = FALSE;
247 ASSERT(cvbr->vertex_buf >= 0);
248 render->vertex_buf = cvbr->vertex_buf;
249 }
250
251 render->xmin = xmin;
252 render->ymin = ymin;
253 render->xmax = xmax;
254 render->ymax = ymax;
255 }
256
257 #if 0
258 /* helpful for debug */
259 cell_flush_int(cell, CELL_FLUSH_WAIT);
260 #endif
261 }
262
263
264 static void
265 cell_vbuf_destroy(struct vbuf_render *vbr)
266 {
267 struct cell_vbuf_render *cvbr = cell_vbuf_render(vbr);
268 cvbr->cell->vbuf_render = NULL;
269 FREE(cvbr);
270 }
271
272
273 /**
274 * Initialize the post-transform vertex buffer information for the given
275 * context.
276 */
277 void
278 cell_init_vbuf(struct cell_context *cell)
279 {
280 assert(cell->draw);
281
282 cell->vbuf_render = CALLOC_STRUCT(cell_vbuf_render);
283
284 /* The max number of indexes is what can fix into a batch buffer,
285 * minus the render and release-verts commands.
286 */
287 cell->vbuf_render->base.max_indices
288 = (CELL_BUFFER_SIZE
289 - sizeof(struct cell_command_render)
290 - sizeof(struct cell_command_release_verts))
291 / sizeof(ushort);
292 cell->vbuf_render->base.max_vertex_buffer_bytes = CELL_BUFFER_SIZE;
293
294 cell->vbuf_render->base.get_vertex_info = cell_vbuf_get_vertex_info;
295 cell->vbuf_render->base.allocate_vertices = cell_vbuf_allocate_vertices;
296 cell->vbuf_render->base.set_primitive = cell_vbuf_set_primitive;
297 cell->vbuf_render->base.draw = cell_vbuf_draw;
298 cell->vbuf_render->base.release_vertices = cell_vbuf_release_vertices;
299 cell->vbuf_render->base.destroy = cell_vbuf_destroy;
300
301 cell->vbuf_render->cell = cell;
302 #if 1
303 cell->vbuf_render->vertex_buf = ~0;
304 #endif
305
306 cell->vbuf = draw_vbuf_stage(cell->draw, &cell->vbuf_render->base);
307 }