2 * Copyright © 2014-2017 Broadcom
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
24 #include "util/u_blitter.h"
25 #include "util/u_prim.h"
26 #include "util/u_format.h"
27 #include "util/u_pack_color.h"
28 #include "util/u_prim_restart.h"
29 #include "util/u_upload_mgr.h"
30 #include "indices/u_primconvert.h"
32 #include "vc5_context.h"
33 #include "vc5_resource.h"
35 #include "broadcom/cle/v3d_packet_v33_pack.h"
36 #include "broadcom/compiler/v3d_compiler.h"
39 * Does the initial bining command list setup for drawing to a given FBO.
42 vc5_start_draw(struct vc5_context
*vc5
)
44 struct vc5_job
*job
= vc5
->job
;
49 /* Get space to emit our BCL state, using a branch to jump to a new BO
52 vc5_cl_ensure_space_with_branch(&job
->bcl
, 256 /* XXX */);
54 job
->submit
.bcl_start
= job
->bcl
.bo
->offset
;
55 vc5_job_add_bo(job
, job
->bcl
.bo
);
57 job
->tile_alloc
= vc5_bo_alloc(vc5
->screen
, 1024 * 1024, "tile alloc");
58 struct vc5_bo
*tsda
= vc5_bo_alloc(vc5
->screen
,
64 /* "Binning mode lists start with a Tile Binning Mode Configuration
67 * Part1 signals the end of binning config setup.
69 cl_emit(&job
->bcl
, TILE_BINNING_MODE_CONFIGURATION_PART2
, config
) {
70 config
.tile_allocation_memory_address
=
71 cl_address(job
->tile_alloc
, 0);
72 config
.tile_allocation_memory_size
= job
->tile_alloc
->size
;
75 cl_emit(&job
->bcl
, TILE_BINNING_MODE_CONFIGURATION_PART1
, config
) {
76 config
.tile_state_data_array_base_address
=
79 config
.width_in_tiles
= job
->draw_tiles_x
;
80 config
.height_in_tiles
= job
->draw_tiles_y
;
83 config
.number_of_render_targets
=
84 MAX2(vc5
->framebuffer
.nr_cbufs
, 1);
86 config
.multisample_mode_4x
= job
->msaa
;
88 config
.maximum_bpp_of_all_render_targets
= job
->internal_bpp
;
91 vc5_bo_unreference(&tsda
);
93 /* There's definitely nothing in the VCD cache we want. */
94 cl_emit(&job
->bcl
, FLUSH_VCD_CACHE
, bin
);
96 /* "Binning mode lists must have a Start Tile Binning item (6) after
97 * any prefix state data before the binning list proper starts."
99 cl_emit(&job
->bcl
, START_TILE_BINNING
, bin
);
101 cl_emit(&job
->bcl
, PRIMITIVE_LIST_FORMAT
, fmt
) {
102 fmt
.data_type
= LIST_INDEXED
;
103 fmt
.primitive_type
= LIST_TRIANGLES
;
106 job
->needs_flush
= true;
107 job
->draw_width
= vc5
->framebuffer
.width
;
108 job
->draw_height
= vc5
->framebuffer
.height
;
112 vc5_predraw_check_textures(struct pipe_context
*pctx
,
113 struct vc5_texture_stateobj
*stage_tex
)
115 struct vc5_context
*vc5
= vc5_context(pctx
);
117 for (int i
= 0; i
< stage_tex
->num_textures
; i
++) {
118 struct pipe_sampler_view
*view
= stage_tex
->textures
[i
];
122 vc5_flush_jobs_writing_resource(vc5
, view
->texture
);
127 vc5_emit_gl_shader_state(struct vc5_context
*vc5
,
128 const struct pipe_draw_info
*info
)
130 struct vc5_job
*job
= vc5
->job
;
131 /* VC5_DIRTY_VTXSTATE */
132 struct vc5_vertex_stateobj
*vtx
= vc5
->vtx
;
133 /* VC5_DIRTY_VTXBUF */
134 struct vc5_vertexbuf_stateobj
*vertexbuf
= &vc5
->vertexbuf
;
136 /* Upload the uniforms to the indirect CL first */
137 struct vc5_cl_reloc fs_uniforms
=
138 vc5_write_uniforms(vc5
, vc5
->prog
.fs
,
139 &vc5
->constbuf
[PIPE_SHADER_FRAGMENT
],
141 struct vc5_cl_reloc vs_uniforms
=
142 vc5_write_uniforms(vc5
, vc5
->prog
.vs
,
143 &vc5
->constbuf
[PIPE_SHADER_VERTEX
],
145 struct vc5_cl_reloc cs_uniforms
=
146 vc5_write_uniforms(vc5
, vc5
->prog
.cs
,
147 &vc5
->constbuf
[PIPE_SHADER_VERTEX
],
150 /* See GFXH-930 workaround below */
151 uint32_t num_elements_to_emit
= MAX2(vtx
->num_elements
, 1);
152 uint32_t shader_rec_offset
=
153 vc5_cl_ensure_space(&job
->indirect
,
154 cl_packet_length(GL_SHADER_STATE_RECORD
) +
155 num_elements_to_emit
*
156 cl_packet_length(GL_SHADER_STATE_ATTRIBUTE_RECORD
),
159 cl_emit(&job
->indirect
, GL_SHADER_STATE_RECORD
, shader
) {
160 shader
.enable_clipping
= true;
161 /* VC5_DIRTY_PRIM_MODE | VC5_DIRTY_RASTERIZER */
162 shader
.point_size_in_shaded_vertex_data
=
163 (info
->mode
== PIPE_PRIM_POINTS
&&
164 vc5
->rasterizer
->base
.point_size_per_vertex
);
166 /* Must be set if the shader modifies Z, discards, or modifies
167 * the sample mask. For any of these cases, the fragment
168 * shader needs to write the Z value (even just discards).
170 shader
.fragment_shader_does_z_writes
=
171 (vc5
->prog
.fs
->prog_data
.fs
->writes_z
||
172 vc5
->prog
.fs
->prog_data
.fs
->discard
);
174 shader
.number_of_varyings_in_fragment_shader
=
175 vc5
->prog
.fs
->prog_data
.base
->num_inputs
;
177 shader
.propagate_nans
= true;
179 shader
.coordinate_shader_code_address
=
180 cl_address(vc5
->prog
.cs
->bo
, 0);
181 shader
.vertex_shader_code_address
=
182 cl_address(vc5
->prog
.vs
->bo
, 0);
183 shader
.fragment_shader_code_address
=
184 cl_address(vc5
->prog
.fs
->bo
, 0);
186 /* XXX: Use combined input/output size flag in the common
189 shader
.coordinate_shader_has_separate_input_and_output_vpm_blocks
= true;
190 shader
.vertex_shader_has_separate_input_and_output_vpm_blocks
= true;
191 shader
.coordinate_shader_input_vpm_segment_size
=
192 MAX2(vc5
->prog
.cs
->prog_data
.vs
->vpm_input_size
, 1);
193 shader
.vertex_shader_input_vpm_segment_size
=
194 MAX2(vc5
->prog
.vs
->prog_data
.vs
->vpm_input_size
, 1);
196 shader
.coordinate_shader_output_vpm_segment_size
=
197 vc5
->prog
.cs
->prog_data
.vs
->vpm_output_size
;
198 shader
.vertex_shader_output_vpm_segment_size
=
199 vc5
->prog
.vs
->prog_data
.vs
->vpm_output_size
;
201 shader
.coordinate_shader_uniforms_address
= cs_uniforms
;
202 shader
.vertex_shader_uniforms_address
= vs_uniforms
;
203 shader
.fragment_shader_uniforms_address
= fs_uniforms
;
205 shader
.vertex_id_read_by_coordinate_shader
=
206 vc5
->prog
.cs
->prog_data
.vs
->uses_vid
;
207 shader
.instance_id_read_by_coordinate_shader
=
208 vc5
->prog
.cs
->prog_data
.vs
->uses_iid
;
209 shader
.vertex_id_read_by_vertex_shader
=
210 vc5
->prog
.vs
->prog_data
.vs
->uses_vid
;
211 shader
.instance_id_read_by_vertex_shader
=
212 vc5
->prog
.vs
->prog_data
.vs
->uses_iid
;
214 shader
.address_of_default_attribute_values
=
215 cl_address(vtx
->default_attribute_values
, 0);
218 for (int i
= 0; i
< vtx
->num_elements
; i
++) {
219 struct pipe_vertex_element
*elem
= &vtx
->pipe
[i
];
220 struct pipe_vertex_buffer
*vb
=
221 &vertexbuf
->vb
[elem
->vertex_buffer_index
];
222 struct vc5_resource
*rsc
= vc5_resource(vb
->buffer
.resource
);
224 const uint32_t size
=
225 cl_packet_length(GL_SHADER_STATE_ATTRIBUTE_RECORD
);
226 cl_emit_with_prepacked(&job
->indirect
,
227 GL_SHADER_STATE_ATTRIBUTE_RECORD
,
228 &vtx
->attrs
[i
* size
], attr
) {
229 attr
.stride
= vb
->stride
;
230 attr
.address
= cl_address(rsc
->bo
,
233 attr
.number_of_values_read_by_coordinate_shader
=
234 vc5
->prog
.cs
->prog_data
.vs
->vattr_sizes
[i
];
235 attr
.number_of_values_read_by_vertex_shader
=
236 vc5
->prog
.vs
->prog_data
.vs
->vattr_sizes
[i
];
240 if (vtx
->num_elements
== 0) {
241 /* GFXH-930: At least one attribute must be enabled and read
242 * by CS and VS. If we have no attributes being consumed by
243 * the shader, set up a dummy to be loaded into the VPM.
245 cl_emit(&job
->indirect
, GL_SHADER_STATE_ATTRIBUTE_RECORD
, attr
) {
246 /* Valid address of data whose value will be unused. */
247 attr
.address
= cl_address(job
->indirect
.bo
, 0);
249 attr
.type
= ATTRIBUTE_FLOAT
;
253 attr
.number_of_values_read_by_coordinate_shader
= 1;
254 attr
.number_of_values_read_by_vertex_shader
= 1;
258 cl_emit(&job
->bcl
, GL_SHADER_STATE
, state
) {
259 state
.address
= cl_address(job
->indirect
.bo
, shader_rec_offset
);
260 state
.number_of_attribute_arrays
= num_elements_to_emit
;
263 vc5_bo_unreference(&cs_uniforms
.bo
);
264 vc5_bo_unreference(&vs_uniforms
.bo
);
265 vc5_bo_unreference(&fs_uniforms
.bo
);
267 job
->shader_rec_count
++;
271 vc5_draw_vbo(struct pipe_context
*pctx
, const struct pipe_draw_info
*info
)
273 struct vc5_context
*vc5
= vc5_context(pctx
);
275 if (!info
->count_from_stream_output
&& !info
->indirect
&&
276 !info
->primitive_restart
&&
277 !u_trim_pipe_prim(info
->mode
, (unsigned*)&info
->count
))
280 /* Fall back for weird desktop GL primitive restart values. */
281 if (info
->primitive_restart
&&
285 switch (info
->index_size
) {
294 if (info
->restart_index
!= mask
) {
295 util_draw_vbo_without_prim_restart(pctx
, info
);
300 if (info
->mode
>= PIPE_PRIM_QUADS
) {
301 util_primconvert_save_rasterizer_state(vc5
->primconvert
, &vc5
->rasterizer
->base
);
302 util_primconvert_draw_vbo(vc5
->primconvert
, info
);
303 perf_debug("Fallback conversion for %d %s vertices\n",
304 info
->count
, u_prim_name(info
->mode
));
308 /* Before setting up the draw, flush anything writing to the textures
311 vc5_predraw_check_textures(pctx
, &vc5
->verttex
);
312 vc5_predraw_check_textures(pctx
, &vc5
->fragtex
);
314 struct vc5_job
*job
= vc5_get_job_for_fbo(vc5
);
316 /* Get space to emit our draw call into the BCL, using a branch to
317 * jump to a new BO if necessary.
319 vc5_cl_ensure_space_with_branch(&job
->bcl
, 256 /* XXX */);
321 if (vc5
->prim_mode
!= info
->mode
) {
322 vc5
->prim_mode
= info
->mode
;
323 vc5
->dirty
|= VC5_DIRTY_PRIM_MODE
;
327 vc5_update_compiled_shaders(vc5
, info
->mode
);
329 vc5_emit_state(pctx
);
331 if (vc5
->dirty
& (VC5_DIRTY_VTXBUF
|
333 VC5_DIRTY_PRIM_MODE
|
334 VC5_DIRTY_RASTERIZER
|
335 VC5_DIRTY_COMPILED_CS
|
336 VC5_DIRTY_COMPILED_VS
|
337 VC5_DIRTY_COMPILED_FS
|
338 vc5
->prog
.cs
->uniform_dirty_bits
|
339 vc5
->prog
.vs
->uniform_dirty_bits
|
340 vc5
->prog
.fs
->uniform_dirty_bits
)) {
341 vc5_emit_gl_shader_state(vc5
, info
);
346 /* The Base Vertex/Base Instance packet sets those values to nonzero
347 * for the next draw call only.
349 if (info
->index_bias
|| info
->start_instance
) {
350 cl_emit(&job
->bcl
, BASE_VERTEX_BASE_INSTANCE
, base
) {
351 base
.base_instance
= info
->start_instance
;
352 base
.base_vertex
= info
->index_bias
;
356 /* The HW only processes transform feedback on primitives with the
359 uint32_t prim_tf_enable
= 0;
360 if (vc5
->prog
.bind_vs
->num_tf_outputs
)
361 prim_tf_enable
= (V3D_PRIM_POINTS_TF
- V3D_PRIM_POINTS
);
363 /* Note that the primitive type fields match with OpenGL/gallium
364 * definitions, up to but not including QUADS.
366 if (info
->index_size
) {
367 uint32_t index_size
= info
->index_size
;
368 uint32_t offset
= info
->start
* index_size
;
369 struct pipe_resource
*prsc
;
370 if (info
->has_user_indices
) {
372 u_upload_data(vc5
->uploader
, 0,
373 info
->count
* info
->index_size
, 4,
377 prsc
= info
->index
.resource
;
379 struct vc5_resource
*rsc
= vc5_resource(prsc
);
381 if (info
->instance_count
> 1) {
382 cl_emit(&job
->bcl
, INDEXED_INSTANCED_PRIMITIVE_LIST
, prim
) {
383 prim
.index_type
= ffs(info
->index_size
) - 1;
384 prim
.maximum_index
= (1u << 31) - 1; /* XXX */
385 prim
.address_of_indices_list
=
386 cl_address(rsc
->bo
, offset
);
387 prim
.mode
= info
->mode
| prim_tf_enable
;
388 prim
.enable_primitive_restarts
= info
->primitive_restart
;
390 prim
.number_of_instances
= info
->instance_count
;
391 prim
.instance_length
= info
->count
;
394 cl_emit(&job
->bcl
, INDEXED_PRIMITIVE_LIST
, prim
) {
395 prim
.index_type
= ffs(info
->index_size
) - 1;
396 prim
.length
= info
->count
;
397 prim
.maximum_index
= (1u << 31) - 1; /* XXX */
398 prim
.address_of_indices_list
=
399 cl_address(rsc
->bo
, offset
);
400 prim
.mode
= info
->mode
| prim_tf_enable
;
401 prim
.enable_primitive_restarts
= info
->primitive_restart
;
405 job
->draw_calls_queued
++;
407 if (info
->has_user_indices
)
408 pipe_resource_reference(&prsc
, NULL
);
410 if (info
->instance_count
> 1) {
411 cl_emit(&job
->bcl
, VERTEX_ARRAY_INSTANCED_PRIMITIVES
, prim
) {
412 prim
.mode
= info
->mode
| prim_tf_enable
;
413 prim
.index_of_first_vertex
= info
->start
;
414 prim
.number_of_instances
= info
->instance_count
;
415 prim
.instance_length
= info
->count
;
418 cl_emit(&job
->bcl
, VERTEX_ARRAY_PRIMITIVES
, prim
) {
419 prim
.mode
= info
->mode
| prim_tf_enable
;
420 prim
.length
= info
->count
;
421 prim
.index_of_first_vertex
= info
->start
;
425 job
->draw_calls_queued
++;
427 if (vc5
->zsa
&& job
->zsbuf
&&
428 (vc5
->zsa
->base
.depth
.enabled
||
429 vc5
->zsa
->base
.stencil
[0].enabled
)) {
430 struct vc5_resource
*rsc
= vc5_resource(job
->zsbuf
->texture
);
431 vc5_job_add_bo(job
, rsc
->bo
);
433 if (vc5
->zsa
->base
.depth
.enabled
) {
434 job
->resolve
|= PIPE_CLEAR_DEPTH
;
435 rsc
->initialized_buffers
= PIPE_CLEAR_DEPTH
;
437 if (vc5
->zsa
->early_z_enable
)
438 job
->uses_early_z
= true;
441 if (vc5
->zsa
->base
.stencil
[0].enabled
) {
442 job
->resolve
|= PIPE_CLEAR_STENCIL
;
443 rsc
->initialized_buffers
|= PIPE_CLEAR_STENCIL
;
447 for (int i
= 0; i
< VC5_MAX_DRAW_BUFFERS
; i
++) {
448 uint32_t bit
= PIPE_CLEAR_COLOR0
<< i
;
450 if (job
->resolve
& bit
|| !job
->cbufs
[i
])
452 struct vc5_resource
*rsc
= vc5_resource(job
->cbufs
[i
]->texture
);
455 vc5_job_add_bo(job
, rsc
->bo
);
458 if (V3D_DEBUG
& V3D_DEBUG_ALWAYS_FLUSH
)
463 vc5_clear(struct pipe_context
*pctx
, unsigned buffers
,
464 const union pipe_color_union
*color
, double depth
, unsigned stencil
)
466 struct vc5_context
*vc5
= vc5_context(pctx
);
467 struct vc5_job
*job
= vc5_get_job_for_fbo(vc5
);
469 /* We can't flag new buffers for clearing once we've queued draws. We
470 * could avoid this by using the 3d engine to clear.
472 if (job
->draw_calls_queued
) {
473 perf_debug("Flushing rendering to process new clear.\n");
474 vc5_job_submit(vc5
, job
);
475 job
= vc5_get_job_for_fbo(vc5
);
478 for (int i
= 0; i
< VC5_MAX_DRAW_BUFFERS
; i
++) {
479 uint32_t bit
= PIPE_CLEAR_COLOR0
<< i
;
480 if (!(buffers
& bit
))
483 struct pipe_surface
*psurf
= vc5
->framebuffer
.cbufs
[i
];
484 struct vc5_surface
*surf
= vc5_surface(psurf
);
485 struct vc5_resource
*rsc
= vc5_resource(psurf
->texture
);
488 uint32_t internal_size
= 4 << surf
->internal_bpp
;
490 switch (surf
->internal_type
) {
491 case INTERNAL_TYPE_8
:
492 if (surf
->format
== PIPE_FORMAT_B4G4R4A4_UNORM
||
493 surf
->format
== PIPE_FORMAT_B4G4R4A4_UNORM
) {
494 /* Our actual hardware layout is ABGR4444, but
495 * we apply a swizzle when texturing to flip
496 * things back around.
498 util_pack_color(color
->f
, PIPE_FORMAT_A8R8G8B8_UNORM
,
501 util_pack_color(color
->f
, PIPE_FORMAT_R8G8B8A8_UNORM
,
504 memcpy(job
->clear_color
[i
], uc
.ui
, internal_size
);
506 case INTERNAL_TYPE_8I
:
507 case INTERNAL_TYPE_8UI
:
508 job
->clear_color
[i
][0] = ((uc
.ui
[0] & 0xff) |
509 (uc
.ui
[1] & 0xff) << 8 |
510 (uc
.ui
[2] & 0xff) << 16 |
511 (uc
.ui
[3] & 0xff) << 24);
513 case INTERNAL_TYPE_16F
:
514 util_pack_color(color
->f
, PIPE_FORMAT_R16G16B16A16_FLOAT
,
516 memcpy(job
->clear_color
[i
], uc
.ui
, internal_size
);
518 case INTERNAL_TYPE_16I
:
519 case INTERNAL_TYPE_16UI
:
520 job
->clear_color
[i
][0] = ((uc
.ui
[0] & 0xffff) |
522 job
->clear_color
[i
][1] = ((uc
.ui
[2] & 0xffff) |
525 case INTERNAL_TYPE_32F
:
526 case INTERNAL_TYPE_32I
:
527 case INTERNAL_TYPE_32UI
:
528 memcpy(job
->clear_color
[i
], color
->ui
, internal_size
);
532 rsc
->initialized_buffers
|= bit
;
535 unsigned zsclear
= buffers
& PIPE_CLEAR_DEPTHSTENCIL
;
537 struct vc5_resource
*rsc
=
538 vc5_resource(vc5
->framebuffer
.zsbuf
->texture
);
540 if (zsclear
& PIPE_CLEAR_DEPTH
)
541 job
->clear_z
= depth
;
542 if (zsclear
& PIPE_CLEAR_STENCIL
)
543 job
->clear_s
= stencil
;
545 rsc
->initialized_buffers
|= zsclear
;
550 job
->draw_max_x
= vc5
->framebuffer
.width
;
551 job
->draw_max_y
= vc5
->framebuffer
.height
;
552 job
->cleared
|= buffers
;
553 job
->resolve
|= buffers
;
559 vc5_clear_render_target(struct pipe_context
*pctx
, struct pipe_surface
*ps
,
560 const union pipe_color_union
*color
,
561 unsigned x
, unsigned y
, unsigned w
, unsigned h
,
562 bool render_condition_enabled
)
564 fprintf(stderr
, "unimpl: clear RT\n");
568 vc5_clear_depth_stencil(struct pipe_context
*pctx
, struct pipe_surface
*ps
,
569 unsigned buffers
, double depth
, unsigned stencil
,
570 unsigned x
, unsigned y
, unsigned w
, unsigned h
,
571 bool render_condition_enabled
)
573 fprintf(stderr
, "unimpl: clear DS\n");
577 vc5_draw_init(struct pipe_context
*pctx
)
579 pctx
->draw_vbo
= vc5_draw_vbo
;
580 pctx
->clear
= vc5_clear
;
581 pctx
->clear_render_target
= vc5_clear_render_target
;
582 pctx
->clear_depth_stencil
= vc5_clear_depth_stencil
;