2
* Copyright (c) 2014 Scott Mansell
3
* Copyright Ā© 2014 Broadcom
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:
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
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
27
#include "util/u_format.h"
28
#include "util/u_pack_color.h"
29
#include "indices/u_primconvert.h"
31
#include "vc4_context.h"
32
#include "vc4_resource.h"
35
* Does the initial bining command list setup for drawing to a given FBO.
38
vc4_start_draw(struct vc4_context *vc4)
43
uint32_t width = vc4->framebuffer.width;
44
uint32_t height = vc4->framebuffer.height;
45
uint32_t tilew = align(width, 64) / 64;
46
uint32_t tileh = align(height, 64) / 64;
48
/* Tile alloc memory setup: We use an initial alloc size of 32b. The
49
* hardware then aligns that to 256b (we use 4096, because all of our
50
* BO allocations align to that anyway), then for some reason the
51
* simulator wants an extra page available, even if you have overflow
54
uint32_t tile_alloc_size = 32 * tilew * tileh;
55
tile_alloc_size = align(tile_alloc_size, 4096);
56
tile_alloc_size += 4096;
57
uint32_t tile_state_size = 48 * tilew * tileh;
58
if (!vc4->tile_alloc || vc4->tile_alloc->size < tile_alloc_size) {
59
vc4_bo_unreference(&vc4->tile_alloc);
60
vc4->tile_alloc = vc4_bo_alloc(vc4->screen, tile_alloc_size,
63
if (!vc4->tile_state || vc4->tile_state->size < tile_state_size) {
64
vc4_bo_unreference(&vc4->tile_state);
65
vc4->tile_state = vc4_bo_alloc(vc4->screen, tile_state_size,
69
// Tile state data is 48 bytes per tile, I think it can be thrown away
70
// as soon as binning is finished.
71
cl_start_reloc(&vc4->bcl, 2);
72
cl_u8(&vc4->bcl, VC4_PACKET_TILE_BINNING_MODE_CONFIG);
73
cl_reloc(vc4, &vc4->bcl, vc4->tile_alloc, 0);
74
cl_u32(&vc4->bcl, vc4->tile_alloc->size);
75
cl_reloc(vc4, &vc4->bcl, vc4->tile_state, 0);
76
cl_u8(&vc4->bcl, tilew);
77
cl_u8(&vc4->bcl, tileh);
79
VC4_BIN_CONFIG_AUTO_INIT_TSDA |
80
VC4_BIN_CONFIG_ALLOC_BLOCK_SIZE_32 |
81
VC4_BIN_CONFIG_ALLOC_INIT_BLOCK_SIZE_32);
83
cl_u8(&vc4->bcl, VC4_PACKET_START_TILE_BINNING);
85
vc4->needs_flush = true;
86
vc4->draw_call_queued = true;
90
vc4_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
92
struct vc4_context *vc4 = vc4_context(pctx);
94
if (info->mode >= PIPE_PRIM_QUADS) {
95
util_primconvert_save_index_buffer(vc4->primconvert, &vc4->indexbuf);
96
util_primconvert_save_rasterizer_state(vc4->primconvert, &vc4->rasterizer->base);
97
util_primconvert_draw_vbo(vc4->primconvert, info);
102
vc4_update_compiled_shaders(vc4, info->mode);
104
vc4_emit_state(pctx);
106
/* the actual draw call. */
107
struct vc4_vertex_stateobj *vtx = vc4->vtx;
108
struct vc4_vertexbuf_stateobj *vertexbuf = &vc4->vertexbuf;
109
cl_u8(&vc4->bcl, VC4_PACKET_GL_SHADER_STATE);
110
assert(vtx->num_elements <= 8);
111
/* Note that number of attributes == 0 in the packet means 8
112
* attributes. This field also contains the offset into shader_rec.
114
cl_u32(&vc4->bcl, vtx->num_elements & 0x7);
116
/* Note that the primitive type fields match with OpenGL/gallium
117
* definitions, up to but not including QUADS.
120
struct vc4_resource *rsc = vc4_resource(vc4->indexbuf.buffer);
122
assert(vc4->indexbuf.index_size == 1 ||
123
vc4->indexbuf.index_size == 2);
125
cl_start_reloc(&vc4->bcl, 1);
126
cl_u8(&vc4->bcl, VC4_PACKET_GL_INDEXED_PRIMITIVE);
129
(vc4->indexbuf.index_size == 2 ?
130
VC4_INDEX_BUFFER_U16:
131
VC4_INDEX_BUFFER_U8));
132
cl_u32(&vc4->bcl, info->count);
133
cl_reloc(vc4, &vc4->bcl, rsc->bo, vc4->indexbuf.offset);
134
cl_u32(&vc4->bcl, info->max_index);
136
cl_u8(&vc4->bcl, VC4_PACKET_GL_ARRAY_PRIMITIVE);
137
cl_u8(&vc4->bcl, info->mode);
138
cl_u32(&vc4->bcl, info->count);
139
cl_u32(&vc4->bcl, info->start);
144
vc4_write_uniforms(vc4, vc4->prog.fs,
145
&vc4->constbuf[PIPE_SHADER_FRAGMENT],
148
vc4_write_uniforms(vc4, vc4->prog.vs,
149
&vc4->constbuf[PIPE_SHADER_VERTEX],
152
vc4_write_uniforms(vc4, vc4->prog.vs,
153
&vc4->constbuf[PIPE_SHADER_VERTEX],
157
cl_start_shader_reloc(&vc4->shader_rec, 3 + vtx->num_elements);
158
cl_u16(&vc4->shader_rec, VC4_SHADER_FLAG_ENABLE_CLIPPING);
159
cl_u8(&vc4->shader_rec, 0); /* fs num uniforms (unused) */
160
cl_u8(&vc4->shader_rec, vc4->prog.fs->num_inputs);
161
cl_reloc(vc4, &vc4->shader_rec, vc4->prog.fs->bo, 0);
162
cl_u32(&vc4->shader_rec, 0); /* UBO offset written by kernel */
164
cl_u16(&vc4->shader_rec, 0); /* vs num uniforms */
165
cl_u8(&vc4->shader_rec, (1 << vtx->num_elements) - 1); /* vs attribute array bitfield */
166
cl_u8(&vc4->shader_rec, 16 * vtx->num_elements); /* vs total attribute size */
167
cl_reloc(vc4, &vc4->shader_rec, vc4->prog.vs->bo, 0);
168
cl_u32(&vc4->shader_rec, 0); /* UBO offset written by kernel */
170
cl_u16(&vc4->shader_rec, 0); /* cs num uniforms */
171
cl_u8(&vc4->shader_rec, (1 << vtx->num_elements) - 1); /* cs attribute array bitfield */
172
cl_u8(&vc4->shader_rec, 16 * vtx->num_elements); /* vs total attribute size */
173
cl_reloc(vc4, &vc4->shader_rec, vc4->prog.vs->bo,
174
vc4->prog.vs->coord_shader_offset);
175
cl_u32(&vc4->shader_rec, 0); /* UBO offset written by kernel */
177
for (int i = 0; i < vtx->num_elements; i++) {
178
struct pipe_vertex_element *elem = &vtx->pipe[i];
179
struct pipe_vertex_buffer *vb =
180
&vertexbuf->vb[elem->vertex_buffer_index];
181
struct vc4_resource *rsc = vc4_resource(vb->buffer);
183
cl_reloc(vc4, &vc4->shader_rec, rsc->bo,
184
vb->buffer_offset + elem->src_offset);
185
cl_u8(&vc4->shader_rec,
186
util_format_get_blocksize(elem->src_format) - 1);
187
cl_u8(&vc4->shader_rec, vb->stride);
188
cl_u8(&vc4->shader_rec, i * 16); /* VS VPM offset */
189
cl_u8(&vc4->shader_rec, i * 16); /* CS VPM offset */
192
if (vc4->zsa && vc4->zsa->base.depth.enabled) {
193
vc4->resolve |= PIPE_CLEAR_DEPTH;
195
vc4->resolve |= PIPE_CLEAR_COLOR0;
197
vc4->shader_rec_count++;
201
pack_rgba(enum pipe_format format, const float *rgba)
204
util_pack_color(rgba, format, &uc);
209
vc4_clear(struct pipe_context *pctx, unsigned buffers,
210
const union pipe_color_union *color, double depth, unsigned stencil)
212
struct vc4_context *vc4 = vc4_context(pctx);
214
/* We can't flag new buffers for clearing once we've queued draws. We
215
* could avoid this by using the 3d engine to clear.
217
if (vc4->draw_call_queued)
220
if (buffers & PIPE_CLEAR_COLOR0) {
221
vc4->clear_color[0] = vc4->clear_color[1] =
222
pack_rgba(vc4->framebuffer.cbufs[0]->format,
226
if (buffers & PIPE_CLEAR_DEPTH)
227
vc4->clear_depth = util_pack_z(PIPE_FORMAT_Z24X8_UNORM, depth);
229
vc4->cleared |= buffers;
230
vc4->resolve |= buffers;
236
vc4_clear_render_target(struct pipe_context *pctx, struct pipe_surface *ps,
237
const union pipe_color_union *color,
238
unsigned x, unsigned y, unsigned w, unsigned h)
240
fprintf(stderr, "unimpl: clear RT\n");
244
vc4_clear_depth_stencil(struct pipe_context *pctx, struct pipe_surface *ps,
245
unsigned buffers, double depth, unsigned stencil,
246
unsigned x, unsigned y, unsigned w, unsigned h)
248
fprintf(stderr, "unimpl: clear DS\n");
252
vc4_draw_init(struct pipe_context *pctx)
254
pctx->draw_vbo = vc4_draw_vbo;
255
pctx->clear = vc4_clear;
256
pctx->clear_render_target = vc4_clear_render_target;
257
pctx->clear_depth_stencil = vc4_clear_depth_stencil;