freedreno/a4xx: format updates
[mesa.git] / src / gallium / drivers / vc4 / vc4_draw.c
1 /*
2 * Copyright (c) 2014 Scott Mansell
3 * Copyright © 2014 Broadcom
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
14 * Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
21 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
22 * IN THE SOFTWARE.
23 */
24
25 #include "util/u_prim.h"
26 #include "util/u_format.h"
27 #include "util/u_pack_color.h"
28 #include "indices/u_primconvert.h"
29
30 #include "vc4_context.h"
31 #include "vc4_resource.h"
32
33 static void
34 vc4_get_draw_cl_space(struct vc4_context *vc4)
35 {
36 /* Binner gets our packet state -- vc4_emit.c contents,
37 * and the primitive itself.
38 */
39 cl_ensure_space(&vc4->bcl, 256);
40
41 /* Nothing for rcl -- that's covered by vc4_context.c */
42
43 /* shader_rec gets up to 12 dwords of reloc handles plus a maximally
44 * sized shader_rec (104 bytes base for 8 vattrs plus 32 bytes of
45 * vattr stride).
46 */
47 cl_ensure_space(&vc4->shader_rec, 12 * sizeof(uint32_t) + 104 + 8 * 32);
48
49 /* Uniforms are covered by vc4_write_uniforms(). */
50
51 /* There could be up to 16 textures per stage, plus misc other
52 * pointers.
53 */
54 cl_ensure_space(&vc4->bo_handles, (2 * 16 + 20) * sizeof(uint32_t));
55 cl_ensure_space(&vc4->bo_pointers,
56 (2 * 16 + 20) * sizeof(struct vc4_bo *));
57 }
58
59 /**
60 * Does the initial bining command list setup for drawing to a given FBO.
61 */
62 static void
63 vc4_start_draw(struct vc4_context *vc4)
64 {
65 if (vc4->needs_flush)
66 return;
67
68 vc4_get_draw_cl_space(vc4);
69
70 uint32_t width = vc4->framebuffer.width;
71 uint32_t height = vc4->framebuffer.height;
72 uint32_t tilew = align(width, 64) / 64;
73 uint32_t tileh = align(height, 64) / 64;
74 struct vc4_cl_out *bcl = cl_start(&vc4->bcl);
75
76 // Tile state data is 48 bytes per tile, I think it can be thrown away
77 // as soon as binning is finished.
78 cl_u8(&bcl, VC4_PACKET_TILE_BINNING_MODE_CONFIG);
79 cl_u32(&bcl, 0); /* tile alloc addr, filled by kernel */
80 cl_u32(&bcl, 0); /* tile alloc size, filled by kernel */
81 cl_u32(&bcl, 0); /* tile state addr, filled by kernel */
82 cl_u8(&bcl, tilew);
83 cl_u8(&bcl, tileh);
84 cl_u8(&bcl, 0); /* flags, filled by kernel. */
85
86 /* START_TILE_BINNING resets the statechange counters in the hardware,
87 * which are what is used when a primitive is binned to a tile to
88 * figure out what new state packets need to be written to that tile's
89 * command list.
90 */
91 cl_u8(&bcl, VC4_PACKET_START_TILE_BINNING);
92
93 /* Reset the current compressed primitives format. This gets modified
94 * by VC4_PACKET_GL_INDEXED_PRIMITIVE and
95 * VC4_PACKET_GL_ARRAY_PRIMITIVE, so it needs to be reset at the start
96 * of every tile.
97 */
98 cl_u8(&bcl, VC4_PACKET_PRIMITIVE_LIST_FORMAT);
99 cl_u8(&bcl, (VC4_PRIMITIVE_LIST_FORMAT_16_INDEX |
100 VC4_PRIMITIVE_LIST_FORMAT_TYPE_TRIANGLES));
101
102 vc4->needs_flush = true;
103 vc4->draw_call_queued = true;
104 vc4->draw_width = width;
105 vc4->draw_height = height;
106
107 cl_end(&vc4->bcl, bcl);
108 }
109
110 static void
111 vc4_update_shadow_textures(struct pipe_context *pctx,
112 struct vc4_texture_stateobj *stage_tex)
113 {
114 for (int i = 0; i < stage_tex->num_textures; i++) {
115 struct pipe_sampler_view *view = stage_tex->textures[i];
116 if (!view)
117 continue;
118 struct vc4_resource *rsc = vc4_resource(view->texture);
119 if (rsc->shadow_parent)
120 vc4_update_shadow_baselevel_texture(pctx, view);
121 }
122 }
123
124 static void
125 vc4_emit_gl_shader_state(struct vc4_context *vc4, const struct pipe_draw_info *info)
126 {
127 /* VC4_DIRTY_VTXSTATE */
128 struct vc4_vertex_stateobj *vtx = vc4->vtx;
129 /* VC4_DIRTY_VTXBUF */
130 struct vc4_vertexbuf_stateobj *vertexbuf = &vc4->vertexbuf;
131
132 /* The simulator throws a fit if VS or CS don't read an attribute, so
133 * we emit a dummy read.
134 */
135 uint32_t num_elements_emit = MAX2(vtx->num_elements, 1);
136 /* Emit the shader record. */
137 struct vc4_cl_out *shader_rec =
138 cl_start_shader_reloc(&vc4->shader_rec, 3 + num_elements_emit);
139 /* VC4_DIRTY_PRIM_MODE | VC4_DIRTY_RASTERIZER */
140 cl_u16(&shader_rec,
141 VC4_SHADER_FLAG_ENABLE_CLIPPING |
142 VC4_SHADER_FLAG_FS_SINGLE_THREAD |
143 ((info->mode == PIPE_PRIM_POINTS &&
144 vc4->rasterizer->base.point_size_per_vertex) ?
145 VC4_SHADER_FLAG_VS_POINT_SIZE : 0));
146
147 /* VC4_DIRTY_COMPILED_FS */
148 cl_u8(&shader_rec, 0); /* fs num uniforms (unused) */
149 cl_u8(&shader_rec, vc4->prog.fs->num_inputs);
150 cl_reloc(vc4, &vc4->shader_rec, &shader_rec, vc4->prog.fs->bo, 0);
151 cl_u32(&shader_rec, 0); /* UBO offset written by kernel */
152
153 /* VC4_DIRTY_COMPILED_VS */
154 cl_u16(&shader_rec, 0); /* vs num uniforms */
155 cl_u8(&shader_rec, vc4->prog.vs->vattrs_live);
156 cl_u8(&shader_rec, vc4->prog.vs->vattr_offsets[8]);
157 cl_reloc(vc4, &vc4->shader_rec, &shader_rec, vc4->prog.vs->bo, 0);
158 cl_u32(&shader_rec, 0); /* UBO offset written by kernel */
159
160 /* VC4_DIRTY_COMPILED_CS */
161 cl_u16(&shader_rec, 0); /* cs num uniforms */
162 cl_u8(&shader_rec, vc4->prog.cs->vattrs_live);
163 cl_u8(&shader_rec, vc4->prog.cs->vattr_offsets[8]);
164 cl_reloc(vc4, &vc4->shader_rec, &shader_rec, vc4->prog.cs->bo, 0);
165 cl_u32(&shader_rec, 0); /* UBO offset written by kernel */
166
167 uint32_t max_index = 0xffff;
168 for (int i = 0; i < vtx->num_elements; i++) {
169 struct pipe_vertex_element *elem = &vtx->pipe[i];
170 struct pipe_vertex_buffer *vb =
171 &vertexbuf->vb[elem->vertex_buffer_index];
172 struct vc4_resource *rsc = vc4_resource(vb->buffer);
173 /* not vc4->dirty tracked: vc4->last_index_bias */
174 uint32_t offset = (vb->buffer_offset +
175 elem->src_offset +
176 vb->stride * info->index_bias);
177 uint32_t vb_size = rsc->bo->size - offset;
178 uint32_t elem_size =
179 util_format_get_blocksize(elem->src_format);
180
181 cl_reloc(vc4, &vc4->shader_rec, &shader_rec, rsc->bo, offset);
182 cl_u8(&shader_rec, elem_size - 1);
183 cl_u8(&shader_rec, vb->stride);
184 cl_u8(&shader_rec, vc4->prog.vs->vattr_offsets[i]);
185 cl_u8(&shader_rec, vc4->prog.cs->vattr_offsets[i]);
186
187 if (vb->stride > 0) {
188 max_index = MIN2(max_index,
189 (vb_size - elem_size) / vb->stride);
190 }
191 }
192
193 if (vtx->num_elements == 0) {
194 assert(num_elements_emit == 1);
195 struct vc4_bo *bo = vc4_bo_alloc(vc4->screen, 4096, "scratch VBO");
196 cl_reloc(vc4, &vc4->shader_rec, &shader_rec, bo, 0);
197 cl_u8(&shader_rec, 16 - 1); /* element size */
198 cl_u8(&shader_rec, 0); /* stride */
199 cl_u8(&shader_rec, 0); /* VS VPM offset */
200 cl_u8(&shader_rec, 0); /* CS VPM offset */
201 vc4_bo_unreference(&bo);
202 }
203 cl_end(&vc4->shader_rec, shader_rec);
204
205 struct vc4_cl_out *bcl = cl_start(&vc4->bcl);
206 /* the actual draw call. */
207 cl_u8(&bcl, VC4_PACKET_GL_SHADER_STATE);
208 assert(vtx->num_elements <= 8);
209 /* Note that number of attributes == 0 in the packet means 8
210 * attributes. This field also contains the offset into shader_rec.
211 */
212 cl_u32(&bcl, num_elements_emit & 0x7);
213 cl_end(&vc4->bcl, bcl);
214
215 vc4_write_uniforms(vc4, vc4->prog.fs,
216 &vc4->constbuf[PIPE_SHADER_FRAGMENT],
217 &vc4->fragtex);
218 vc4_write_uniforms(vc4, vc4->prog.vs,
219 &vc4->constbuf[PIPE_SHADER_VERTEX],
220 &vc4->verttex);
221 vc4_write_uniforms(vc4, vc4->prog.cs,
222 &vc4->constbuf[PIPE_SHADER_VERTEX],
223 &vc4->verttex);
224
225 vc4->last_index_bias = info->index_bias;
226 vc4->max_index = max_index;
227 }
228
229 static void
230 vc4_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
231 {
232 struct vc4_context *vc4 = vc4_context(pctx);
233
234 if (info->mode >= PIPE_PRIM_QUADS) {
235 util_primconvert_save_index_buffer(vc4->primconvert, &vc4->indexbuf);
236 util_primconvert_save_rasterizer_state(vc4->primconvert, &vc4->rasterizer->base);
237 util_primconvert_draw_vbo(vc4->primconvert, info);
238 perf_debug("Fallback conversion for %d %s vertices\n",
239 info->count, u_prim_name(info->mode));
240 return;
241 }
242
243 /* Before setting up the draw, do any fixup blits necessary. */
244 vc4_update_shadow_textures(pctx, &vc4->verttex);
245 vc4_update_shadow_textures(pctx, &vc4->fragtex);
246
247 vc4_get_draw_cl_space(vc4);
248
249 if (vc4->prim_mode != info->mode) {
250 vc4->prim_mode = info->mode;
251 vc4->dirty |= VC4_DIRTY_PRIM_MODE;
252 }
253
254 vc4_start_draw(vc4);
255 vc4_update_compiled_shaders(vc4, info->mode);
256
257 vc4_emit_state(pctx);
258
259 if ((vc4->dirty & (VC4_DIRTY_VTXBUF |
260 VC4_DIRTY_VTXSTATE |
261 VC4_DIRTY_PRIM_MODE |
262 VC4_DIRTY_RASTERIZER |
263 VC4_DIRTY_COMPILED_CS |
264 VC4_DIRTY_COMPILED_VS |
265 VC4_DIRTY_COMPILED_FS |
266 vc4->prog.cs->uniform_dirty_bits |
267 vc4->prog.vs->uniform_dirty_bits |
268 vc4->prog.fs->uniform_dirty_bits)) ||
269 vc4->last_index_bias != info->index_bias) {
270 vc4_emit_gl_shader_state(vc4, info);
271 }
272
273 vc4->dirty = 0;
274
275 /* Note that the primitive type fields match with OpenGL/gallium
276 * definitions, up to but not including QUADS.
277 */
278 struct vc4_cl_out *bcl = cl_start(&vc4->bcl);
279 if (info->indexed) {
280 uint32_t offset = vc4->indexbuf.offset;
281 uint32_t index_size = vc4->indexbuf.index_size;
282 struct pipe_resource *prsc;
283 if (vc4->indexbuf.index_size == 4) {
284 prsc = vc4_get_shadow_index_buffer(pctx, &vc4->indexbuf,
285 info->count, &offset);
286 index_size = 2;
287 } else {
288 prsc = vc4->indexbuf.buffer;
289 }
290 struct vc4_resource *rsc = vc4_resource(prsc);
291
292 cl_start_reloc(&vc4->bcl, &bcl, 1);
293 cl_u8(&bcl, VC4_PACKET_GL_INDEXED_PRIMITIVE);
294 cl_u8(&bcl,
295 info->mode |
296 (index_size == 2 ?
297 VC4_INDEX_BUFFER_U16:
298 VC4_INDEX_BUFFER_U8));
299 cl_u32(&bcl, info->count);
300 cl_reloc(vc4, &vc4->bcl, &bcl, rsc->bo, offset);
301 cl_u32(&bcl, vc4->max_index);
302
303 if (vc4->indexbuf.index_size == 4)
304 pipe_resource_reference(&prsc, NULL);
305 } else {
306 cl_u8(&bcl, VC4_PACKET_GL_ARRAY_PRIMITIVE);
307 cl_u8(&bcl, info->mode);
308 cl_u32(&bcl, info->count);
309 cl_u32(&bcl, info->start);
310 }
311 cl_end(&vc4->bcl, bcl);
312
313 if (vc4->zsa && vc4->zsa->base.depth.enabled) {
314 vc4->resolve |= PIPE_CLEAR_DEPTH;
315 }
316 if (vc4->zsa && vc4->zsa->base.stencil[0].enabled)
317 vc4->resolve |= PIPE_CLEAR_STENCIL;
318 vc4->resolve |= PIPE_CLEAR_COLOR0;
319
320 vc4->shader_rec_count++;
321
322 if (vc4_debug & VC4_DEBUG_ALWAYS_FLUSH)
323 vc4_flush(pctx);
324 }
325
326 static uint32_t
327 pack_rgba(enum pipe_format format, const float *rgba)
328 {
329 union util_color uc;
330 util_pack_color(rgba, format, &uc);
331 if (util_format_get_blocksize(format) == 2)
332 return uc.us;
333 else
334 return uc.ui[0];
335 }
336
337 static void
338 vc4_clear(struct pipe_context *pctx, unsigned buffers,
339 const union pipe_color_union *color, double depth, unsigned stencil)
340 {
341 struct vc4_context *vc4 = vc4_context(pctx);
342
343 /* We can't flag new buffers for clearing once we've queued draws. We
344 * could avoid this by using the 3d engine to clear.
345 */
346 if (vc4->draw_call_queued) {
347 perf_debug("Flushing rendering to process new clear.");
348 vc4_flush(pctx);
349 }
350
351 if (buffers & PIPE_CLEAR_COLOR0) {
352 vc4->clear_color[0] = vc4->clear_color[1] =
353 pack_rgba(vc4->framebuffer.cbufs[0]->format,
354 color->f);
355 }
356
357 if (buffers & PIPE_CLEAR_DEPTH) {
358 /* Though the depth buffer is stored with Z in the high 24,
359 * for this field we just need to store it in the low 24.
360 */
361 vc4->clear_depth = util_pack_z(PIPE_FORMAT_Z24X8_UNORM, depth);
362 }
363
364 if (buffers & PIPE_CLEAR_STENCIL)
365 vc4->clear_stencil = stencil;
366
367 vc4->draw_min_x = 0;
368 vc4->draw_min_y = 0;
369 vc4->draw_max_x = vc4->framebuffer.width;
370 vc4->draw_max_y = vc4->framebuffer.height;
371 vc4->cleared |= buffers;
372 vc4->resolve |= buffers;
373
374 vc4_start_draw(vc4);
375 }
376
377 static void
378 vc4_clear_render_target(struct pipe_context *pctx, struct pipe_surface *ps,
379 const union pipe_color_union *color,
380 unsigned x, unsigned y, unsigned w, unsigned h)
381 {
382 fprintf(stderr, "unimpl: clear RT\n");
383 }
384
385 static void
386 vc4_clear_depth_stencil(struct pipe_context *pctx, struct pipe_surface *ps,
387 unsigned buffers, double depth, unsigned stencil,
388 unsigned x, unsigned y, unsigned w, unsigned h)
389 {
390 fprintf(stderr, "unimpl: clear DS\n");
391 }
392
393 void
394 vc4_draw_init(struct pipe_context *pctx)
395 {
396 pctx->draw_vbo = vc4_draw_vbo;
397 pctx->clear = vc4_clear;
398 pctx->clear_render_target = vc4_clear_render_target;
399 pctx->clear_depth_stencil = vc4_clear_depth_stencil;
400 }