2 * Copyright © 2014-2017 Broadcom
3 * Copyright (C) 2012 Rob Clark <robclark@freedesktop.org>
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
29 #include "broadcom/common/v3d_macros.h"
34 #include "pipe/p_context.h"
35 #include "pipe/p_state.h"
36 #include "util/bitset.h"
37 #include "util/slab.h"
39 #include "drm-uapi/v3d_drm.h"
40 #include "v3d_screen.h"
41 #include "broadcom/common/v3d_limits.h"
45 void v3d_job_add_bo(struct v3d_job
*job
, struct v3d_bo
*bo
);
47 #include "v3d_bufmgr.h"
48 #include "v3d_resource.h"
51 #ifdef USE_V3D_SIMULATOR
52 #define using_v3d_simulator true
54 #define using_v3d_simulator false
57 #define VC5_DIRTY_BLEND (1 << 0)
58 #define VC5_DIRTY_RASTERIZER (1 << 1)
59 #define VC5_DIRTY_ZSA (1 << 2)
60 #define VC5_DIRTY_FRAGTEX (1 << 3)
61 #define VC5_DIRTY_VERTTEX (1 << 4)
62 #define VC5_DIRTY_SHADER_IMAGE (1 << 5)
64 #define VC5_DIRTY_BLEND_COLOR (1 << 7)
65 #define VC5_DIRTY_STENCIL_REF (1 << 8)
66 #define VC5_DIRTY_SAMPLE_STATE (1 << 9)
67 #define VC5_DIRTY_FRAMEBUFFER (1 << 10)
68 #define VC5_DIRTY_STIPPLE (1 << 11)
69 #define VC5_DIRTY_VIEWPORT (1 << 12)
70 #define VC5_DIRTY_CONSTBUF (1 << 13)
71 #define VC5_DIRTY_VTXSTATE (1 << 14)
72 #define VC5_DIRTY_VTXBUF (1 << 15)
73 #define VC5_DIRTY_SCISSOR (1 << 17)
74 #define VC5_DIRTY_FLAT_SHADE_FLAGS (1 << 18)
75 #define VC5_DIRTY_PRIM_MODE (1 << 19)
76 #define VC5_DIRTY_CLIP (1 << 20)
77 #define VC5_DIRTY_UNCOMPILED_VS (1 << 21)
78 #define VC5_DIRTY_UNCOMPILED_FS (1 << 22)
79 #define VC5_DIRTY_COMPILED_CS (1 << 23)
80 #define VC5_DIRTY_COMPILED_VS (1 << 24)
81 #define VC5_DIRTY_COMPILED_FS (1 << 25)
82 #define VC5_DIRTY_FS_INPUTS (1 << 26)
83 #define VC5_DIRTY_STREAMOUT (1 << 27)
84 #define VC5_DIRTY_OQ (1 << 28)
85 #define VC5_DIRTY_CENTROID_FLAGS (1 << 29)
86 #define VC5_DIRTY_NOPERSPECTIVE_FLAGS (1 << 30)
87 #define VC5_DIRTY_SSBO (1 << 31)
89 #define VC5_MAX_FS_INPUTS 64
91 enum v3d_sampler_state_variant
{
92 V3D_SAMPLER_STATE_BORDER_0
,
93 V3D_SAMPLER_STATE_F16
,
94 V3D_SAMPLER_STATE_F16_UNORM
,
95 V3D_SAMPLER_STATE_F16_SNORM
,
96 V3D_SAMPLER_STATE_F16_BGRA
,
97 V3D_SAMPLER_STATE_F16_BGRA_UNORM
,
98 V3D_SAMPLER_STATE_F16_BGRA_SNORM
,
99 V3D_SAMPLER_STATE_F16_A
,
100 V3D_SAMPLER_STATE_F16_A_SNORM
,
101 V3D_SAMPLER_STATE_F16_A_UNORM
,
102 V3D_SAMPLER_STATE_F16_LA
,
103 V3D_SAMPLER_STATE_F16_LA_UNORM
,
104 V3D_SAMPLER_STATE_F16_LA_SNORM
,
105 V3D_SAMPLER_STATE_32
,
106 V3D_SAMPLER_STATE_32_UNORM
,
107 V3D_SAMPLER_STATE_32_SNORM
,
108 V3D_SAMPLER_STATE_32_A
,
109 V3D_SAMPLER_STATE_32_A_UNORM
,
110 V3D_SAMPLER_STATE_32_A_SNORM
,
111 V3D_SAMPLER_STATE_1010102U
,
112 V3D_SAMPLER_STATE_16U
,
113 V3D_SAMPLER_STATE_16I
,
114 V3D_SAMPLER_STATE_8I
,
115 V3D_SAMPLER_STATE_8U
,
117 V3D_SAMPLER_STATE_VARIANT_COUNT
,
120 enum v3d_flush_cond
{
121 /* Flush job unless we are flushing for transform feedback, where we
122 * handle flushing in the driver via the 'Wait for TF' packet.
125 /* Always flush the job, even for cases where we would normally not
126 * do it, such as transform feedback.
129 /* Flush job if it is not the current FBO job. This is intended to
130 * skip automatic flushes of the current job for resources that we
131 * expect to be externally synchronized by the application using
132 * glMemoryBarrier(), such as SSBOs and shader images.
134 V3D_FLUSH_NOT_CURRENT_JOB
,
137 struct v3d_sampler_view
{
138 struct pipe_sampler_view base
;
141 /* Precomputed swizzles to pass in to the shader key. */
144 uint8_t texture_shader_state
[32];
145 /* V3D 4.x: Texture state struct. */
148 enum v3d_sampler_state_variant sampler_variant
;
150 /* Actual texture to be read by this sampler view. May be different
151 * from base.texture in the case of having a shadow tiled copy of a
154 struct pipe_resource
*texture
;
157 struct v3d_sampler_state
{
158 struct pipe_sampler_state base
;
162 /* V3D 3.x: Packed texture state. */
163 uint8_t texture_shader_state
[32];
164 /* V3D 4.x: Sampler state struct. */
165 struct pipe_resource
*sampler_state
;
166 uint32_t sampler_state_offset
[V3D_SAMPLER_STATE_VARIANT_COUNT
];
168 bool border_color_variants
;
171 struct v3d_texture_stateobj
{
172 struct pipe_sampler_view
*textures
[V3D_MAX_TEXTURE_SAMPLERS
];
173 unsigned num_textures
;
174 struct pipe_sampler_state
*samplers
[V3D_MAX_TEXTURE_SAMPLERS
];
175 unsigned num_samplers
;
176 struct v3d_cl_reloc texture_state
[V3D_MAX_TEXTURE_SAMPLERS
];
179 struct v3d_shader_uniform_info
{
180 enum quniform_contents
*contents
;
185 struct v3d_uncompiled_shader
{
186 /** A name for this program, so you can track it in shader-db output. */
188 /** How many variants of this program were compiled, for shader-db. */
189 uint32_t compiled_variant_count
;
190 struct pipe_shader_state base
;
191 uint32_t num_tf_outputs
;
192 struct v3d_varying_slot
*tf_outputs
;
193 uint16_t tf_specs
[16];
194 uint16_t tf_specs_psiz
[16];
195 uint32_t num_tf_specs
;
198 struct v3d_compiled_shader
{
199 struct pipe_resource
*resource
;
203 struct v3d_prog_data
*base
;
204 struct v3d_vs_prog_data
*vs
;
205 struct v3d_fs_prog_data
*fs
;
206 struct v3d_compute_prog_data
*compute
;
210 * VC5_DIRTY_* flags that, when set in v3d->dirty, mean that the
211 * uniforms have to be rewritten (and therefore the shader state
214 uint32_t uniform_dirty_bits
;
217 struct v3d_program_stateobj
{
218 struct v3d_uncompiled_shader
*bind_vs
, *bind_fs
, *bind_compute
;
219 struct v3d_compiled_shader
*cs
, *vs
, *fs
, *compute
;
221 struct hash_table
*cache
[MESA_SHADER_STAGES
];
223 struct v3d_bo
*spill_bo
;
224 int spill_size_per_thread
;
227 struct v3d_constbuf_stateobj
{
228 struct pipe_constant_buffer cb
[PIPE_MAX_CONSTANT_BUFFERS
];
229 uint32_t enabled_mask
;
233 struct v3d_vertexbuf_stateobj
{
234 struct pipe_vertex_buffer vb
[PIPE_MAX_ATTRIBS
];
236 uint32_t enabled_mask
;
240 struct v3d_vertex_stateobj
{
241 struct pipe_vertex_element pipe
[V3D_MAX_VS_INPUTS
/ 4];
242 unsigned num_elements
;
244 uint8_t attrs
[16 * (V3D_MAX_VS_INPUTS
/ 4)];
245 struct pipe_resource
*defaults
;
246 uint32_t defaults_offset
;
249 struct v3d_stream_output_target
{
250 struct pipe_stream_output_target base
;
251 /* Number of transform feedback vertices written to this target */
252 uint32_t recorded_vertex_count
;
255 struct v3d_streamout_stateobj
{
256 struct pipe_stream_output_target
*targets
[PIPE_MAX_SO_BUFFERS
];
257 /* Number of vertices we've written into the buffer so far. */
258 uint32_t offsets
[PIPE_MAX_SO_BUFFERS
];
259 unsigned num_targets
;
262 struct v3d_ssbo_stateobj
{
263 struct pipe_shader_buffer sb
[PIPE_MAX_SHADER_BUFFERS
];
264 uint32_t enabled_mask
;
267 /* Hash table key for v3d->jobs */
269 struct pipe_surface
*cbufs
[4];
270 struct pipe_surface
*zsbuf
;
274 VC5_EZ_UNDECIDED
= 0,
280 struct v3d_image_view
{
281 struct pipe_image_view base
;
282 /* V3D 4.x texture shader state struct */
283 struct pipe_resource
*tex_state
;
284 uint32_t tex_state_offset
;
287 struct v3d_shaderimg_stateobj
{
288 struct v3d_image_view si
[PIPE_MAX_SHADER_IMAGES
];
289 uint32_t enabled_mask
;
293 * A complete bin/render job.
295 * This is all of the state necessary to submit a bin/render to the kernel.
296 * We want to be able to have multiple in progress at a time, so that we don't
297 * need to flush an existing CL just to switch to rendering to a new render
298 * target (which would mean reading back from the old render target when
299 * starting to render to it again).
302 struct v3d_context
*v3d
;
305 struct v3d_cl indirect
;
306 struct v3d_bo
*tile_alloc
;
307 struct v3d_bo
*tile_state
;
308 uint32_t shader_rec_count
;
310 struct drm_v3d_submit_cl submit
;
313 * Set of all BOs referenced by the job. This will be used for making
314 * the list of BOs that the kernel will need to have paged in to
319 /** Sum of the sizes of the BOs referenced by the job. */
320 uint32_t referenced_size
;
322 struct set
*write_prscs
;
323 struct set
*tf_write_prscs
;
325 /* Size of the submit.bo_handles array. */
326 uint32_t bo_handles_size
;
328 /** @{ Surfaces to submit rendering for. */
329 struct pipe_surface
*cbufs
[4];
330 struct pipe_surface
*zsbuf
;
333 * Bounding box of the scissor across all queued drawing.
335 * Note that the max values are exclusive.
343 * Width/height of the color framebuffer being rendered to,
344 * for VC5_TILE_RENDERING_MODE_CONFIG.
347 uint32_t draw_height
;
349 /** @{ Tile information, depending on MSAA and float color buffer. */
350 uint32_t draw_tiles_x
; /** @< Number of tiles wide for framebuffer. */
351 uint32_t draw_tiles_y
; /** @< Number of tiles high for framebuffer. */
353 uint32_t tile_width
; /** @< Width of a tile. */
354 uint32_t tile_height
; /** @< Height of a tile. */
355 /** maximum internal_bpp of all color render targets. */
356 uint32_t internal_bpp
;
358 /** Whether the current rendering is in a 4X MSAA tile buffer. */
362 /* Bitmask of PIPE_CLEAR_* of buffers that were cleared before the
366 /* Bitmask of PIPE_CLEAR_* of buffers that have been read by a draw
367 * call without having been cleared first.
370 /* Bitmask of PIPE_CLEAR_* of buffers that have been rendered to
371 * (either clears or draws) and should be stored.
374 uint32_t clear_color
[4][4];
379 * Set if some drawing (triangles, blits, or just a glClear()) has
380 * been done to the FBO, meaning that we need to
381 * DRM_IOCTL_VC5_SUBMIT_CL.
385 /* Set if any shader has dirtied cachelines in the TMU that need to be
386 * flushed before job end.
391 * Set if a packet enabling TF has been emitted in the job (V3D 4.x).
396 * Current EZ state for drawing. Updated at the start of draw after
397 * we've decided on the shader being rendered.
399 enum v3d_ez_state ez_state
;
401 * The first EZ state that was used for drawing with a decided EZ
402 * direction (so either UNDECIDED, GT, or LT).
404 enum v3d_ez_state first_ez_state
;
407 * Number of draw calls (not counting full buffer clears) queued in
410 uint32_t draw_calls_queued
;
412 struct v3d_job_key key
;
416 struct pipe_context base
;
419 struct v3d_screen
*screen
;
421 /** The 3D rendering job for the currently bound FBO. */
424 /* Map from struct v3d_job_key to the job for that FBO.
426 struct hash_table
*jobs
;
429 * Map from v3d_resource to a job writing to that resource.
431 * Primarily for flushing jobs rendering to textures that are now
434 struct hash_table
*write_jobs
;
436 struct slab_child_pool transfer_pool
;
437 struct blitter_context
*blitter
;
439 /** bitfield of VC5_DIRTY_* */
442 struct primconvert_context
*primconvert
;
444 uint32_t next_uncompiled_program_id
;
445 uint64_t next_compiled_program_id
;
447 struct v3d_compiler_state
*compiler_state
;
451 /** Maximum index buffer valid for the current shader_rec. */
454 /** Sync object that our RCL or TFU job will update as its out_sync. */
457 /* Stream uploader used by gallium internals. This could also be used
458 * by driver internals, but we tend to use the v3d_cl.h interfaces
461 struct u_upload_mgr
*uploader
;
462 /* State uploader used inside the driver. This is for packing bits of
463 * long-term state inside buffers, since the kernel interfaces
464 * allocate a page at a time.
466 struct u_upload_mgr
*state_uploader
;
468 /** @{ Current pipeline state objects */
469 struct pipe_scissor_state scissor
;
470 struct v3d_blend_state
*blend
;
471 struct v3d_rasterizer_state
*rasterizer
;
472 struct v3d_depth_stencil_alpha_state
*zsa
;
474 struct v3d_program_stateobj prog
;
475 uint32_t compute_num_workgroups
[3];
476 struct v3d_bo
*compute_shared_memory
;
478 struct v3d_vertex_stateobj
*vtx
;
481 struct pipe_blend_color f
;
484 struct pipe_stencil_ref stencil_ref
;
485 unsigned sample_mask
;
486 struct pipe_framebuffer_state framebuffer
;
488 /* Per render target, whether we should swap the R and B fields in the
489 * shader's color output and in blending. If render targets disagree
490 * on the R/B swap and use the constant color, then we would need to
491 * fall back to in-shader blending.
493 uint8_t swap_color_rb
;
495 /* Per render target, whether we should treat the dst alpha values as
498 * For RGBX formats, the tile buffer's alpha channel will be
501 uint8_t blend_dst_alpha_one
;
505 uint32_t tf_prims_generated
;
506 uint32_t prims_generated
;
508 struct pipe_poly_stipple stipple
;
509 struct pipe_clip_state clip
;
510 struct pipe_viewport_state viewport
;
511 struct v3d_ssbo_stateobj ssbo
[PIPE_SHADER_TYPES
];
512 struct v3d_shaderimg_stateobj shaderimg
[PIPE_SHADER_TYPES
];
513 struct v3d_constbuf_stateobj constbuf
[PIPE_SHADER_TYPES
];
514 struct v3d_texture_stateobj tex
[PIPE_SHADER_TYPES
];
515 struct v3d_vertexbuf_stateobj vertexbuf
;
516 struct v3d_streamout_stateobj streamout
;
517 struct v3d_bo
*current_oq
;
518 struct pipe_resource
*prim_counts
;
519 uint32_t prim_counts_offset
;
520 struct pipe_debug_callback debug
;
524 struct v3d_rasterizer_state
{
525 struct pipe_rasterizer_state base
;
529 uint8_t depth_offset
[9];
530 uint8_t depth_offset_z16
[9];
533 struct v3d_depth_stencil_alpha_state
{
534 struct pipe_depth_stencil_alpha_state base
;
536 enum v3d_ez_state ez_state
;
538 uint8_t stencil_front
[6];
539 uint8_t stencil_back
[6];
542 struct v3d_blend_state
{
543 struct pipe_blend_state base
;
545 /* Per-RT mask of whether blending is enabled. */
546 uint8_t blend_enables
;
549 #define perf_debug(...) do { \
550 if (unlikely(V3D_DEBUG & V3D_DEBUG_PERF)) \
551 fprintf(stderr, __VA_ARGS__); \
552 if (unlikely(v3d->debug.debug_message)) \
553 pipe_debug_message(&v3d->debug, PERF_INFO, __VA_ARGS__); \
556 #define foreach_bit(b, mask) \
557 for (uint32_t _m = (mask), b; _m && ({(b) = u_bit_scan(&_m); 1;});)
559 static inline struct v3d_context
*
560 v3d_context(struct pipe_context
*pcontext
)
562 return (struct v3d_context
*)pcontext
;
565 static inline struct v3d_sampler_view
*
566 v3d_sampler_view(struct pipe_sampler_view
*psview
)
568 return (struct v3d_sampler_view
*)psview
;
571 static inline struct v3d_sampler_state
*
572 v3d_sampler_state(struct pipe_sampler_state
*psampler
)
574 return (struct v3d_sampler_state
*)psampler
;
577 static inline struct v3d_stream_output_target
*
578 v3d_stream_output_target(struct pipe_stream_output_target
*ptarget
)
580 return (struct v3d_stream_output_target
*)ptarget
;
583 static inline uint32_t
584 v3d_stream_output_target_get_vertex_count(struct pipe_stream_output_target
*ptarget
)
586 return v3d_stream_output_target(ptarget
)->recorded_vertex_count
;
589 struct pipe_context
*v3d_context_create(struct pipe_screen
*pscreen
,
590 void *priv
, unsigned flags
);
591 void v3d_program_init(struct pipe_context
*pctx
);
592 void v3d_program_fini(struct pipe_context
*pctx
);
593 void v3d_query_init(struct pipe_context
*pctx
);
595 void v3d_simulator_init(struct v3d_screen
*screen
);
596 void v3d_simulator_destroy(struct v3d_screen
*screen
);
597 uint32_t v3d_simulator_get_spill(uint32_t spill_size
);
598 int v3d_simulator_ioctl(int fd
, unsigned long request
, void *arg
);
599 void v3d_simulator_open_from_handle(int fd
, int handle
, uint32_t size
);
602 v3d_ioctl(int fd
, unsigned long request
, void *arg
)
604 if (using_v3d_simulator
)
605 return v3d_simulator_ioctl(fd
, request
, arg
);
607 return drmIoctl(fd
, request
, arg
);
611 v3d_transform_feedback_enabled(struct v3d_context
*v3d
)
613 return v3d
->prog
.bind_vs
->num_tf_specs
!= 0 &&
617 void v3d_set_shader_uniform_dirty_flags(struct v3d_compiled_shader
*shader
);
618 struct v3d_cl_reloc
v3d_write_uniforms(struct v3d_context
*v3d
,
619 struct v3d_compiled_shader
*shader
,
620 enum pipe_shader_type stage
);
622 void v3d_flush(struct pipe_context
*pctx
);
623 void v3d_job_init(struct v3d_context
*v3d
);
624 struct v3d_job
*v3d_get_job(struct v3d_context
*v3d
,
625 struct pipe_surface
**cbufs
,
626 struct pipe_surface
*zsbuf
);
627 struct v3d_job
*v3d_get_job_for_fbo(struct v3d_context
*v3d
);
628 void v3d_job_add_bo(struct v3d_job
*job
, struct v3d_bo
*bo
);
629 void v3d_job_add_write_resource(struct v3d_job
*job
, struct pipe_resource
*prsc
);
630 void v3d_job_add_tf_write_resource(struct v3d_job
*job
, struct pipe_resource
*prsc
);
631 void v3d_job_submit(struct v3d_context
*v3d
, struct v3d_job
*job
);
632 void v3d_flush_jobs_using_bo(struct v3d_context
*v3d
, struct v3d_bo
*bo
);
633 void v3d_flush_jobs_writing_resource(struct v3d_context
*v3d
,
634 struct pipe_resource
*prsc
,
635 enum v3d_flush_cond flush_cond
);
636 void v3d_flush_jobs_reading_resource(struct v3d_context
*v3d
,
637 struct pipe_resource
*prsc
,
638 enum v3d_flush_cond flush_cond
);
639 void v3d_update_compiled_shaders(struct v3d_context
*v3d
, uint8_t prim_mode
);
640 void v3d_update_compiled_cs(struct v3d_context
*v3d
);
642 bool v3d_rt_format_supported(const struct v3d_device_info
*devinfo
,
644 bool v3d_tex_format_supported(const struct v3d_device_info
*devinfo
,
646 uint8_t v3d_get_rt_format(const struct v3d_device_info
*devinfo
, enum pipe_format f
);
647 uint8_t v3d_get_tex_format(const struct v3d_device_info
*devinfo
, enum pipe_format f
);
648 uint8_t v3d_get_tex_return_size(const struct v3d_device_info
*devinfo
,
650 enum pipe_tex_compare compare
);
651 uint8_t v3d_get_tex_return_channels(const struct v3d_device_info
*devinfo
,
653 const uint8_t *v3d_get_format_swizzle(const struct v3d_device_info
*devinfo
,
655 void v3d_get_internal_type_bpp_for_output_format(const struct v3d_device_info
*devinfo
,
659 bool v3d_tfu_supports_tex_format(const struct v3d_device_info
*devinfo
,
660 uint32_t tex_format
);
662 void v3d_init_query_functions(struct v3d_context
*v3d
);
663 void v3d_blit(struct pipe_context
*pctx
, const struct pipe_blit_info
*blit_info
);
664 void v3d_blitter_save(struct v3d_context
*v3d
);
665 bool v3d_generate_mipmap(struct pipe_context
*pctx
,
666 struct pipe_resource
*prsc
,
667 enum pipe_format format
,
668 unsigned int base_level
,
669 unsigned int last_level
,
670 unsigned int first_layer
,
671 unsigned int last_layer
);
673 struct v3d_fence
*v3d_fence_create(struct v3d_context
*v3d
);
675 void v3d_tf_update_counters(struct v3d_context
*v3d
);
678 # include "v3dx_context.h"
680 # define v3dX(x) v3d33_##x
681 # include "v3dx_context.h"
684 # define v3dX(x) v3d41_##x
685 # include "v3dx_context.h"
689 #endif /* VC5_CONTEXT_H */