vc4: Don't look up the compiled shaders unless state has changed.
[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_format.h"
26 #include "util/u_pack_color.h"
27 #include "indices/u_primconvert.h"
28
29 #include "vc4_context.h"
30 #include "vc4_resource.h"
31
32 /**
33 * Does the initial bining command list setup for drawing to a given FBO.
34 */
35 static void
36 vc4_start_draw(struct vc4_context *vc4)
37 {
38 if (vc4->needs_flush)
39 return;
40
41 uint32_t width = vc4->framebuffer.width;
42 uint32_t height = vc4->framebuffer.height;
43 uint32_t tilew = align(width, 64) / 64;
44 uint32_t tileh = align(height, 64) / 64;
45
46 /* Tile alloc memory setup: We use an initial alloc size of 32b. The
47 * hardware then aligns that to 256b (we use 4096, because all of our
48 * BO allocations align to that anyway), then for some reason the
49 * simulator wants an extra page available, even if you have overflow
50 * memory set up.
51 */
52 uint32_t tile_alloc_size = 32 * tilew * tileh;
53 tile_alloc_size = align(tile_alloc_size, 4096);
54 tile_alloc_size += 4096;
55 uint32_t tile_state_size = 48 * tilew * tileh;
56 if (!vc4->tile_alloc || vc4->tile_alloc->size < tile_alloc_size) {
57 vc4_bo_unreference(&vc4->tile_alloc);
58 vc4->tile_alloc = vc4_bo_alloc(vc4->screen, tile_alloc_size,
59 "tile_alloc");
60 }
61 if (!vc4->tile_state || vc4->tile_state->size < tile_state_size) {
62 vc4_bo_unreference(&vc4->tile_state);
63 vc4->tile_state = vc4_bo_alloc(vc4->screen, tile_state_size,
64 "tile_state");
65 }
66
67 // Tile state data is 48 bytes per tile, I think it can be thrown away
68 // as soon as binning is finished.
69 cl_start_reloc(&vc4->bcl, 2);
70 cl_u8(&vc4->bcl, VC4_PACKET_TILE_BINNING_MODE_CONFIG);
71 cl_reloc(vc4, &vc4->bcl, vc4->tile_alloc, 0);
72 cl_u32(&vc4->bcl, vc4->tile_alloc->size);
73 cl_reloc(vc4, &vc4->bcl, vc4->tile_state, 0);
74 cl_u8(&vc4->bcl, tilew);
75 cl_u8(&vc4->bcl, tileh);
76 cl_u8(&vc4->bcl,
77 VC4_BIN_CONFIG_AUTO_INIT_TSDA |
78 VC4_BIN_CONFIG_ALLOC_BLOCK_SIZE_32 |
79 VC4_BIN_CONFIG_ALLOC_INIT_BLOCK_SIZE_32);
80
81 cl_u8(&vc4->bcl, VC4_PACKET_START_TILE_BINNING);
82
83 vc4->needs_flush = true;
84 vc4->draw_call_queued = true;
85 }
86
87 static void
88 vc4_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
89 {
90 struct vc4_context *vc4 = vc4_context(pctx);
91
92 if (info->mode >= PIPE_PRIM_QUADS) {
93 util_primconvert_save_index_buffer(vc4->primconvert, &vc4->indexbuf);
94 util_primconvert_save_rasterizer_state(vc4->primconvert, &vc4->rasterizer->base);
95 util_primconvert_draw_vbo(vc4->primconvert, info);
96 return;
97 }
98
99 struct vc4_vertex_stateobj *vtx = vc4->vtx;
100 struct vc4_vertexbuf_stateobj *vertexbuf = &vc4->vertexbuf;
101
102 if (vc4->prim_mode != info->mode) {
103 vc4->prim_mode = info->mode;
104 vc4->dirty |= VC4_DIRTY_PRIM_MODE;
105 }
106
107 vc4_start_draw(vc4);
108 vc4_update_compiled_shaders(vc4, info->mode);
109
110 vc4_emit_state(pctx);
111 vc4->dirty = 0;
112
113 vc4_write_uniforms(vc4, vc4->prog.fs,
114 &vc4->constbuf[PIPE_SHADER_FRAGMENT],
115 &vc4->fragtex);
116 vc4_write_uniforms(vc4, vc4->prog.vs,
117 &vc4->constbuf[PIPE_SHADER_VERTEX],
118 &vc4->verttex);
119 vc4_write_uniforms(vc4, vc4->prog.cs,
120 &vc4->constbuf[PIPE_SHADER_VERTEX],
121 &vc4->verttex);
122
123 /* Emit the shader record. */
124 cl_start_shader_reloc(&vc4->shader_rec, 3 + vtx->num_elements);
125 cl_u16(&vc4->shader_rec,
126 VC4_SHADER_FLAG_ENABLE_CLIPPING |
127 ((info->mode == PIPE_PRIM_POINTS &&
128 vc4->rasterizer->base.point_size_per_vertex) ?
129 VC4_SHADER_FLAG_VS_POINT_SIZE : 0));
130 cl_u8(&vc4->shader_rec, 0); /* fs num uniforms (unused) */
131 cl_u8(&vc4->shader_rec, vc4->prog.fs->num_inputs);
132 cl_reloc(vc4, &vc4->shader_rec, vc4->prog.fs->bo, 0);
133 cl_u32(&vc4->shader_rec, 0); /* UBO offset written by kernel */
134
135 cl_u16(&vc4->shader_rec, 0); /* vs num uniforms */
136 cl_u8(&vc4->shader_rec, (1 << vtx->num_elements) - 1); /* vs attribute array bitfield */
137 cl_u8(&vc4->shader_rec, 16 * vtx->num_elements); /* vs total attribute size */
138 cl_reloc(vc4, &vc4->shader_rec, vc4->prog.vs->bo, 0);
139 cl_u32(&vc4->shader_rec, 0); /* UBO offset written by kernel */
140
141 cl_u16(&vc4->shader_rec, 0); /* cs num uniforms */
142 cl_u8(&vc4->shader_rec, (1 << vtx->num_elements) - 1); /* cs attribute array bitfield */
143 cl_u8(&vc4->shader_rec, 16 * vtx->num_elements); /* cs total attribute size */
144 cl_reloc(vc4, &vc4->shader_rec, vc4->prog.cs->bo, 0);
145 cl_u32(&vc4->shader_rec, 0); /* UBO offset written by kernel */
146
147 uint32_t max_index = 0xffff;
148 for (int i = 0; i < vtx->num_elements; i++) {
149 struct pipe_vertex_element *elem = &vtx->pipe[i];
150 struct pipe_vertex_buffer *vb =
151 &vertexbuf->vb[elem->vertex_buffer_index];
152 struct vc4_resource *rsc = vc4_resource(vb->buffer);
153 uint32_t offset = vb->buffer_offset + elem->src_offset;
154 uint32_t vb_size = rsc->bo->size - offset;
155 uint32_t elem_size =
156 util_format_get_blocksize(elem->src_format);
157
158 cl_reloc(vc4, &vc4->shader_rec, rsc->bo, offset);
159 cl_u8(&vc4->shader_rec, elem_size - 1);
160 cl_u8(&vc4->shader_rec, vb->stride);
161 cl_u8(&vc4->shader_rec, i * 16); /* VS VPM offset */
162 cl_u8(&vc4->shader_rec, i * 16); /* CS VPM offset */
163
164 if (vb->stride > 0) {
165 max_index = MIN2(max_index,
166 (vb_size - elem_size) / vb->stride);
167 }
168 }
169
170 /* the actual draw call. */
171 cl_u8(&vc4->bcl, VC4_PACKET_GL_SHADER_STATE);
172 assert(vtx->num_elements <= 8);
173 /* Note that number of attributes == 0 in the packet means 8
174 * attributes. This field also contains the offset into shader_rec.
175 */
176 cl_u32(&vc4->bcl, vtx->num_elements & 0x7);
177
178 /* Note that the primitive type fields match with OpenGL/gallium
179 * definitions, up to but not including QUADS.
180 */
181 if (info->indexed) {
182 struct vc4_resource *rsc = vc4_resource(vc4->indexbuf.buffer);
183
184 assert(vc4->indexbuf.index_size == 1 ||
185 vc4->indexbuf.index_size == 2);
186
187 cl_start_reloc(&vc4->bcl, 1);
188 cl_u8(&vc4->bcl, VC4_PACKET_GL_INDEXED_PRIMITIVE);
189 cl_u8(&vc4->bcl,
190 info->mode |
191 (vc4->indexbuf.index_size == 2 ?
192 VC4_INDEX_BUFFER_U16:
193 VC4_INDEX_BUFFER_U8));
194 cl_u32(&vc4->bcl, info->count);
195 cl_reloc(vc4, &vc4->bcl, rsc->bo, vc4->indexbuf.offset);
196 cl_u32(&vc4->bcl, max_index);
197 } else {
198 cl_u8(&vc4->bcl, VC4_PACKET_GL_ARRAY_PRIMITIVE);
199 cl_u8(&vc4->bcl, info->mode);
200 cl_u32(&vc4->bcl, info->count);
201 cl_u32(&vc4->bcl, info->start);
202 }
203
204 if (vc4->zsa && vc4->zsa->base.depth.enabled) {
205 vc4->resolve |= PIPE_CLEAR_DEPTH;
206 }
207 if (vc4->zsa && vc4->zsa->base.stencil[0].enabled)
208 vc4->resolve |= PIPE_CLEAR_STENCIL;
209 vc4->resolve |= PIPE_CLEAR_COLOR0;
210
211 vc4->shader_rec_count++;
212
213 if (vc4_debug & VC4_DEBUG_ALWAYS_FLUSH)
214 vc4_flush(pctx);
215 }
216
217 static uint32_t
218 pack_rgba(enum pipe_format format, const float *rgba)
219 {
220 union util_color uc;
221 util_pack_color(rgba, format, &uc);
222 return uc.ui[0];
223 }
224
225 static void
226 vc4_clear(struct pipe_context *pctx, unsigned buffers,
227 const union pipe_color_union *color, double depth, unsigned stencil)
228 {
229 struct vc4_context *vc4 = vc4_context(pctx);
230
231 /* We can't flag new buffers for clearing once we've queued draws. We
232 * could avoid this by using the 3d engine to clear.
233 */
234 if (vc4->draw_call_queued)
235 vc4_flush(pctx);
236
237 if (buffers & PIPE_CLEAR_COLOR0) {
238 vc4->clear_color[0] = vc4->clear_color[1] =
239 pack_rgba(vc4->framebuffer.cbufs[0]->format,
240 color->f);
241 }
242
243 if (buffers & PIPE_CLEAR_DEPTH) {
244 /* Though the depth buffer is stored with Z in the high 24,
245 * for this field we just need to store it in the low 24.
246 */
247 vc4->clear_depth = util_pack_z(PIPE_FORMAT_Z24X8_UNORM, depth);
248 }
249
250 if (buffers & PIPE_CLEAR_STENCIL)
251 vc4->clear_stencil = stencil;
252
253 vc4->cleared |= buffers;
254 vc4->resolve |= buffers;
255
256 vc4_start_draw(vc4);
257 }
258
259 static void
260 vc4_clear_render_target(struct pipe_context *pctx, struct pipe_surface *ps,
261 const union pipe_color_union *color,
262 unsigned x, unsigned y, unsigned w, unsigned h)
263 {
264 fprintf(stderr, "unimpl: clear RT\n");
265 }
266
267 static void
268 vc4_clear_depth_stencil(struct pipe_context *pctx, struct pipe_surface *ps,
269 unsigned buffers, double depth, unsigned stencil,
270 unsigned x, unsigned y, unsigned w, unsigned h)
271 {
272 fprintf(stderr, "unimpl: clear DS\n");
273 }
274
275 void
276 vc4_draw_init(struct pipe_context *pctx)
277 {
278 pctx->draw_vbo = vc4_draw_vbo;
279 pctx->clear = vc4_clear;
280 pctx->clear_render_target = vc4_clear_render_target;
281 pctx->clear_depth_stencil = vc4_clear_depth_stencil;
282 }