src/gallium/drivers/panfrost/pan_context.h

   1 /*
   2  * © Copyright 2018 Alyssa Rosenzweig
   3  *
   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:
  10  *
  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
  13  * Software.
  14  *
  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 FROM,
  20  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  21  * SOFTWARE.
  22  *
  23  */
  24
  25 #ifndef __BUILDER_H__
  26 #define __BUILDER_H__
  27
  28 #define _LARGEFILE64_SOURCE 1
  29 #define CACHE_LINE_SIZE 1024 /* TODO */
  30 #include <sys/mman.h>
  31 #include <assert.h>
  32 #include "pan_resource.h"
  33 #include "pan_job.h"
  34
  35 #include "pipe/p_compiler.h"
  36 #include "pipe/p_config.h"
  37 #include "pipe/p_context.h"
  38 #include "pipe/p_defines.h"
  39 #include "pipe/p_format.h"
  40 #include "pipe/p_screen.h"
  41 #include "pipe/p_state.h"
  42 #include "util/u_blitter.h"
  43 #include "util/hash_table.h"
  44
  45 /* Forward declare to avoid extra header dep */
  46 struct prim_convert_context;
  47
  48 #define MAX_DRAW_CALLS 4096
  49 #define MAX_VARYINGS   4096
  50
  51 //#define PAN_DIRTY_CLEAR            (1 << 0)
  52 #define PAN_DIRTY_RASTERIZER (1 << 2)
  53 #define PAN_DIRTY_FS         (1 << 3)
  54 #define PAN_DIRTY_FRAG_CORE  (PAN_DIRTY_FS) /* Dirty writes are tied */
  55 #define PAN_DIRTY_VS         (1 << 4)
  56 #define PAN_DIRTY_VERTEX     (1 << 5)
  57 #define PAN_DIRTY_VERT_BUF   (1 << 6)
  58 //#define PAN_DIRTY_VIEWPORT   (1 << 7)
  59 #define PAN_DIRTY_SAMPLERS   (1 << 8)
  60 #define PAN_DIRTY_TEXTURES   (1 << 9)
  61
  62 struct panfrost_constant_buffer {
  63         bool dirty;
  64         size_t size;
  65         void *buffer;
  66 };
  67
  68 struct panfrost_query {
  69         /* Passthrough from Gallium */
  70         unsigned type;
  71         unsigned index;
  72
  73         /* Memory for the GPU to writeback the value of the query */
  74         struct panfrost_transfer transfer;
  75 };
  76
  77 #define PANFROST_MAX_TRANSIENT_ENTRIES 64
  78
  79 struct panfrost_transient_pool {
  80         /* Memory blocks in the pool */
  81         struct panfrost_memory_entry *entries[PANFROST_MAX_TRANSIENT_ENTRIES];
  82
  83         /* Number of entries we own */
  84         unsigned entry_count;
  85
  86         /* Current entry that we are writing to, zero-indexed, strictly less than entry_count */
  87         unsigned entry_index;
  88
  89         /* Number of bytes into the current entry we are */
  90         off_t entry_offset;
  91
  92         /* Entry size (all entries must be homogenous) */
  93         size_t entry_size;
  94 };
  95
  96 struct panfrost_context {
  97         /* Gallium context */
  98         struct pipe_context base;
  99
 100         /* Bound job and map of panfrost_job_key to jobs */
 101         struct panfrost_job *job;
 102         struct hash_table *jobs;
 103
 104         /* Bit mask for supported PIPE_DRAW for this hardware */
 105         unsigned draw_modes;
 106
 107         struct pipe_framebuffer_state pipe_framebuffer;
 108
 109         /* The number of concurrent FBOs allowed depends on the number of pools
 110          * used; pools are ringed for parallelism opportunities */
 111
 112         struct panfrost_transient_pool transient_pools[2];
 113         int cmdstream_i;
 114
 115         struct panfrost_memory cmdstream_persistent;
 116         struct panfrost_memory shaders;
 117         struct panfrost_memory scratchpad;
 118         struct panfrost_memory tiler_heap;
 119         struct panfrost_memory varying_mem;
 120         struct panfrost_memory misc_0;
 121         struct panfrost_memory misc_1;
 122         struct panfrost_memory depth_stencil_buffer;
 123
 124         struct {
 125                 unsigned buffers;
 126                 const union pipe_color_union *color;
 127                 double depth;
 128                 unsigned stencil;
 129         } last_clear;
 130
 131         struct panfrost_query *occlusion_query;
 132
 133         /* Each render job has multiple framebuffer descriptors associated with
 134          * it, used for various purposes with more or less the same format. The
 135          * most obvious is the fragment framebuffer descriptor, which carries
 136          * e.g. clearing information */
 137
 138         union {
 139                 struct mali_single_framebuffer fragment_sfbd;
 140                 struct {
 141                         struct bifrost_framebuffer fragment_mfbd;
 142                         struct bifrost_fb_extra fragment_extra;
 143                         struct bifrost_render_target fragment_rts[4];
 144                 };
 145         };
 146
 147         /* Each draw has corresponding vertex and tiler payloads */
 148         struct midgard_payload_vertex_tiler payload_vertex;
 149         struct midgard_payload_vertex_tiler payload_tiler;
 150
 151         /* The fragment shader binary itself is pointed here (for the tripipe) but
 152          * also everything else in the shader core, including blending, the
 153          * stencil/depth tests, etc. Refer to the presentations. */
 154
 155         struct mali_shader_meta fragment_shader_core;
 156
 157         /* A frame is composed of a starting set value job, a number of vertex
 158          * and tiler jobs, linked to the fragment job at the end. See the
 159          * presentations for more information how this works */
 160
 161         unsigned draw_count;
 162
 163         mali_ptr set_value_job;
 164         mali_ptr vertex_jobs[MAX_DRAW_CALLS];
 165         mali_ptr tiler_jobs[MAX_DRAW_CALLS];
 166
 167         struct mali_job_descriptor_header *u_set_value_job;
 168         struct mali_job_descriptor_header *u_vertex_jobs[MAX_DRAW_CALLS];
 169         struct mali_job_descriptor_header *u_tiler_jobs[MAX_DRAW_CALLS];
 170
 171         unsigned vertex_job_count;
 172         unsigned tiler_job_count;
 173
 174         /* Per-draw Dirty flags are setup like any other driver */
 175         int dirty;
 176
 177         /* Per frame dirty flag - whether there was a clear. If not, we need to do a partial update, maybe */
 178         bool frame_cleared;
 179
 180         unsigned vertex_count;
 181
 182         union mali_attr attributes[PIPE_MAX_ATTRIBS];
 183
 184         unsigned varying_height;
 185
 186         struct mali_viewport *viewport;
 187         struct mali_single_framebuffer vt_framebuffer_sfbd;
 188         struct bifrost_framebuffer vt_framebuffer_mfbd;
 189
 190         /* TODO: Multiple uniform buffers (index =/= 0), finer updates? */
 191
 192         struct panfrost_constant_buffer constant_buffer[PIPE_SHADER_TYPES];
 193
 194         /* CSOs */
 195         struct panfrost_rasterizer *rasterizer;
 196
 197         struct panfrost_shader_variants *vs;
 198         struct panfrost_shader_variants *fs;
 199
 200         struct panfrost_vertex_state *vertex;
 201
 202         struct pipe_vertex_buffer *vertex_buffers;
 203         unsigned vertex_buffer_count;
 204
 205         struct panfrost_sampler_state *samplers[PIPE_SHADER_TYPES][PIPE_MAX_SAMPLERS];
 206         unsigned sampler_count[PIPE_SHADER_TYPES];
 207
 208         struct panfrost_sampler_view *sampler_views[PIPE_SHADER_TYPES][PIPE_MAX_SHADER_SAMPLER_VIEWS];
 209         unsigned sampler_view_count[PIPE_SHADER_TYPES];
 210
 211         struct primconvert_context *primconvert;
 212         struct blitter_context *blitter;
 213
 214         struct panfrost_blend_state *blend;
 215
 216         struct pipe_viewport_state pipe_viewport;
 217         struct pipe_scissor_state scissor;
 218         struct pipe_blend_color blend_color;
 219         struct pipe_depth_stencil_alpha_state *depth_stencil;
 220         struct pipe_stencil_ref stencil_ref;
 221 };
 222
 223 /* Corresponds to the CSO */
 224
 225 struct panfrost_rasterizer {
 226         struct pipe_rasterizer_state base;
 227
 228         /* Bitmask of front face, etc */
 229         unsigned tiler_gl_enables;
 230 };
 231
 232 struct panfrost_blend_state {
 233         struct pipe_blend_state base;
 234
 235         /* Whether a blend shader is in use */
 236         bool has_blend_shader;
 237
 238         /* Compiled fixed function command */
 239         struct mali_blend_equation equation;
 240
 241         /* Compiled blend shader */
 242         mali_ptr blend_shader;
 243         int blend_work_count;
 244 };
 245
 246 /* Internal varyings descriptor */
 247 struct panfrost_varyings {
 248         /* Varyings information: stride of each chunk of memory used for
 249          * varyings (similar structure with attributes). Count is just the
 250          * number of vec4's. Buffer count is the number of varying chunks (<=
 251          * count). Height is used to calculate gl_Position's position ("it's
 252          * not a pun, Alyssa!"). Vertex-only varyings == descriptor for
 253          * gl_Position and something else apparently occupying the same space.
 254          * Varyings == main varyings descriptors following typical mali_attr
 255          * conventions. */
 256
 257         unsigned varyings_stride[MAX_VARYINGS];
 258         unsigned varying_count;
 259         unsigned varying_buffer_count;
 260
 261         /* Map of the actual varyings buffer */
 262         uint8_t *varyings_buffer_cpu;
 263         mali_ptr varyings_descriptor;
 264         mali_ptr varyings_descriptor_fragment;
 265 };
 266
 267 /* Variants bundle together to form the backing CSO, bundling multiple
 268  * shaders with varying emulated features baked in (alpha test
 269  * parameters, etc) */
 270 #define MAX_SHADER_VARIANTS 8
 271
 272 /* A shader state corresponds to the actual, current variant of the shader */
 273 struct panfrost_shader_state {
 274         struct pipe_shader_state *base;
 275
 276         /* Compiled, mapped descriptor, ready for the hardware */
 277         bool compiled;
 278         struct mali_shader_meta *tripipe;
 279         mali_ptr tripipe_gpu;
 280
 281         /* Non-descript information */
 282         int uniform_count;
 283         bool can_discard;
 284         bool writes_point_size;
 285
 286         /* Valid for vertex shaders only due to when this is calculated */
 287         struct panfrost_varyings varyings;
 288
 289         /* Information on this particular shader variant */
 290         struct pipe_alpha_state alpha_state;
 291 };
 292
 293 /* A collection of varyings (the CSO) */
 294 struct panfrost_shader_variants {
 295         struct pipe_shader_state base;
 296
 297         struct panfrost_shader_state variants[MAX_SHADER_VARIANTS];
 298         unsigned variant_count;
 299
 300         /* The current active variant */
 301         unsigned active_variant;
 302 };
 303
 304 struct panfrost_vertex_state {
 305         unsigned num_elements;
 306
 307         struct pipe_vertex_element pipe[PIPE_MAX_ATTRIBS];
 308         int nr_components[PIPE_MAX_ATTRIBS];
 309
 310         /* The actual attribute meta, prebaked and GPU mapped. TODO: Free memory */
 311         struct mali_attr_meta *hw;
 312         mali_ptr descriptor_ptr;
 313 };
 314
 315 struct panfrost_sampler_state {
 316         struct pipe_sampler_state base;
 317         struct mali_sampler_descriptor hw;
 318 };
 319
 320 /* Misnomer: Sampler view corresponds to textures, not samplers */
 321
 322 struct panfrost_sampler_view {
 323         struct pipe_sampler_view base;
 324         struct mali_texture_descriptor hw;
 325 };
 326
 327 static inline struct panfrost_context *
 328 pan_context(struct pipe_context *pcontext)
 329 {
 330         return (struct panfrost_context *) pcontext;
 331 }
 332
 333 static inline struct panfrost_screen *
 334 pan_screen(struct pipe_screen *p)
 335 {
 336    return (struct panfrost_screen *)p;
 337 }
 338
 339 struct pipe_context *
 340 panfrost_create_context(struct pipe_screen *screen, void *priv, unsigned flags);
 341
 342 void
 343 panfrost_emit_for_draw(struct panfrost_context *ctx, bool with_vertex_data);
 344
 345 struct panfrost_transfer
 346 panfrost_vertex_tiler_job(struct panfrost_context *ctx, bool is_tiler, bool is_elided_tiler);
 347
 348 unsigned
 349 panfrost_get_default_swizzle(unsigned components);
 350
 351 void
 352 panfrost_flush(
 353         struct pipe_context *pipe,
 354         struct pipe_fence_handle **fence,
 355         unsigned flags);
 356
 357 mali_ptr
 358 panfrost_fragment_job(struct panfrost_context *ctx);
 359
 360 void
 361 panfrost_shader_compile(struct panfrost_context *ctx, struct mali_shader_meta *meta, const char *src, int type, struct panfrost_shader_state *state);
 362
 363 #endif