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#include <sys/mman.h>
30#include <assert.h>
31#include "pan_resource.h"
32#include "pan_job.h"
33#include "pan_blend_cso.h"
34#include "pan_encoder.h"
35#include "pan_texture.h"
36#include "pan_earlyzs.h"
37
38#include "pipe/p_compiler.h"
39#include "pipe/p_config.h"
40#include "pipe/p_context.h"
41#include "pipe/p_defines.h"
42#include "pipe/p_format.h"
43#include "pipe/p_screen.h"
44#include "pipe/p_state.h"
45#include "util/u_blitter.h"
46#include "util/hash_table.h"
47#include "util/simple_mtx.h"
48
49#include "midgard/midgard_compile.h"
50#include "compiler/shader_enums.h"
51
52#define SET_BIT(lval, bit, cond) \
53	if (cond) \
54		lval |= (bit); \
55	else \
56		lval &= ~(bit);
57
58/* Dirty tracking flags. 3D is for general 3D state. Shader flags are
59 * per-stage. Renderer refers to Renderer State Descriptors. Vertex refers to
60 * vertex attributes/elements. */
61
62enum pan_dirty_3d {
63        PAN_DIRTY_VIEWPORT       = BITFIELD_BIT(0),
64        PAN_DIRTY_SCISSOR        = BITFIELD_BIT(1),
65        PAN_DIRTY_VERTEX         = BITFIELD_BIT(2),
66        PAN_DIRTY_PARAMS         = BITFIELD_BIT(3),
67        PAN_DIRTY_DRAWID         = BITFIELD_BIT(4),
68        PAN_DIRTY_TLS_SIZE       = BITFIELD_BIT(5),
69        PAN_DIRTY_ZS             = BITFIELD_BIT(6),
70        PAN_DIRTY_BLEND          = BITFIELD_BIT(7),
71        PAN_DIRTY_MSAA           = BITFIELD_BIT(8),
72        PAN_DIRTY_OQ             = BITFIELD_BIT(9),
73        PAN_DIRTY_RASTERIZER     = BITFIELD_BIT(10),
74        PAN_DIRTY_POINTS         = BITFIELD_BIT(11),
75        PAN_DIRTY_SO             = BITFIELD_BIT(12),
76};
77
78enum pan_dirty_shader {
79        PAN_DIRTY_STAGE_SHADER   = BITFIELD_BIT(0),
80        PAN_DIRTY_STAGE_TEXTURE  = BITFIELD_BIT(1),
81        PAN_DIRTY_STAGE_SAMPLER  = BITFIELD_BIT(2),
82        PAN_DIRTY_STAGE_IMAGE    = BITFIELD_BIT(3),
83        PAN_DIRTY_STAGE_CONST    = BITFIELD_BIT(4),
84        PAN_DIRTY_STAGE_SSBO     = BITFIELD_BIT(5),
85};
86
87struct panfrost_constant_buffer {
88        struct pipe_constant_buffer cb[PIPE_MAX_CONSTANT_BUFFERS];
89        uint32_t enabled_mask;
90};
91
92struct panfrost_query {
93        /* Passthrough from Gallium */
94        unsigned type;
95        unsigned index;
96
97        /* For computed queries. 64-bit to prevent overflow */
98        struct {
99                uint64_t start;
100                uint64_t end;
101        };
102
103        /* Memory for the GPU to writeback the value of the query */
104        struct pipe_resource *rsrc;
105
106        /* Whether an occlusion query is for a MSAA framebuffer */
107        bool msaa;
108};
109
110struct pipe_fence_handle {
111        struct pipe_reference reference;
112        uint32_t syncobj;
113        bool signaled;
114};
115
116struct panfrost_streamout_target {
117        struct pipe_stream_output_target base;
118        uint32_t offset;
119};
120
121struct panfrost_streamout {
122        struct pipe_stream_output_target *targets[PIPE_MAX_SO_BUFFERS];
123        unsigned num_targets;
124};
125
126struct panfrost_context {
127        /* Gallium context */
128        struct pipe_context base;
129
130        /* Dirty global state */
131        enum pan_dirty_3d dirty;
132
133        /* Per shader stage dirty state */
134        enum pan_dirty_shader dirty_shader[PIPE_SHADER_TYPES];
135
136        /* Unowned pools, so manage yourself. */
137        struct panfrost_pool descs, shaders;
138
139        /* Sync obj used to keep track of in-flight jobs. */
140        uint32_t syncobj;
141
142        /* Set of 32 batches. When the set is full, the LRU entry (the batch
143         * with the smallest seqnum) is flushed to free a slot.
144         */
145        struct {
146                uint64_t seqnum;
147                struct panfrost_batch slots[PAN_MAX_BATCHES];
148
149                /** Set of active batches for faster traversal */
150                BITSET_DECLARE(active, PAN_MAX_BATCHES);
151        } batches;
152
153        /* Map from resources to panfrost_batches */
154        struct hash_table *writers;
155
156        /* Bound job batch */
157        struct panfrost_batch *batch;
158
159        /* Within a launch_grid call.. */
160        const struct pipe_grid_info *compute_grid;
161
162        struct pipe_framebuffer_state pipe_framebuffer;
163        struct panfrost_streamout streamout;
164
165        bool active_queries;
166        uint64_t prims_generated;
167        uint64_t tf_prims_generated;
168        struct panfrost_query *occlusion_query;
169
170        bool indirect_draw;
171        unsigned drawid;
172        unsigned vertex_count;
173        unsigned instance_count;
174        unsigned offset_start;
175        unsigned base_vertex;
176        unsigned base_instance;
177        mali_ptr first_vertex_sysval_ptr;
178        mali_ptr base_vertex_sysval_ptr;
179        mali_ptr base_instance_sysval_ptr;
180        enum pipe_prim_type active_prim;
181
182        /* If instancing is enabled, vertex count padded for instance; if
183         * it is disabled, just equal to plain vertex count */
184        unsigned padded_count;
185
186        struct panfrost_constant_buffer constant_buffer[PIPE_SHADER_TYPES];
187        struct panfrost_rasterizer *rasterizer;
188        struct panfrost_shader_variants *shader[PIPE_SHADER_TYPES];
189        struct panfrost_vertex_state *vertex;
190
191        struct pipe_vertex_buffer vertex_buffers[PIPE_MAX_ATTRIBS];
192        uint32_t vb_mask;
193
194        struct pipe_shader_buffer ssbo[PIPE_SHADER_TYPES][PIPE_MAX_SHADER_BUFFERS];
195        uint32_t ssbo_mask[PIPE_SHADER_TYPES];
196
197        struct pipe_image_view images[PIPE_SHADER_TYPES][PIPE_MAX_SHADER_IMAGES];
198        uint32_t image_mask[PIPE_SHADER_TYPES];
199
200        struct panfrost_sampler_state *samplers[PIPE_SHADER_TYPES][PIPE_MAX_SAMPLERS];
201        unsigned sampler_count[PIPE_SHADER_TYPES];
202
203        struct panfrost_sampler_view *sampler_views[PIPE_SHADER_TYPES][PIPE_MAX_SHADER_SAMPLER_VIEWS];
204        unsigned sampler_view_count[PIPE_SHADER_TYPES];
205
206        struct blitter_context *blitter;
207
208        struct panfrost_blend_state *blend;
209
210        /* On Valhall, does the current blend state use a blend shader for any
211         * output? We need this information in a hot path to decide if
212         * per-sample shading should be enabled.
213         */
214        bool valhall_has_blend_shader;
215
216        struct pipe_viewport_state pipe_viewport;
217        struct pipe_scissor_state scissor;
218        struct pipe_blend_color blend_color;
219        struct panfrost_zsa_state *depth_stencil;
220        struct pipe_stencil_ref stencil_ref;
221        uint16_t sample_mask;
222        unsigned min_samples;
223
224        struct panfrost_query *cond_query;
225        bool cond_cond;
226        enum pipe_render_cond_flag cond_mode;
227
228        bool is_noop;
229
230        /* Mask of active render targets */
231        uint8_t fb_rt_mask;
232};
233
234/* Corresponds to the CSO */
235
236struct panfrost_rasterizer;
237
238/* Linked varyings */
239struct pan_linkage {
240        /* If the upload is owned by the CSO instead
241         * of the pool, the referenced BO. Else,
242         * NULL. */
243        struct panfrost_bo *bo;
244
245        /* Uploaded attribute descriptors */
246        mali_ptr producer, consumer;
247
248        /* Varyings buffers required */
249        uint32_t present;
250
251        /* Per-vertex stride for general varying buffer */
252        uint32_t stride;
253};
254
255#define RSD_WORDS 16
256
257/* Variants bundle together to form the backing CSO, bundling multiple
258 * shaders with varying emulated features baked in
259 */
260struct panfrost_fs_key {
261        /* Number of colour buffers */
262        unsigned nr_cbufs;
263
264        /* Midgard shaders that read the tilebuffer must be keyed for
265         * non-blendable formats
266         */
267        enum pipe_format rt_formats[8];
268
269        /* From rasterize state, to lower point sprites */
270        uint16_t sprite_coord_enable;
271
272        /* User clip plane lowering */
273        uint8_t clip_plane_enable;
274};
275
276struct panfrost_shader_key {
277        /* Valhall needs special handling for desktop GL varyings */
278        uint32_t fixed_varying_mask;
279
280        /* If we need vertex shader keys, union it in */
281        struct panfrost_fs_key fs;
282};
283
284/* A shader state corresponds to the actual, current variant of the shader */
285struct panfrost_shader_state {
286        /* Respectively, shader binary and Renderer State Descriptor */
287        struct panfrost_pool_ref bin, state;
288
289        /* For fragment shaders, a prepared (but not uploaded RSD) */
290        uint32_t partial_rsd[RSD_WORDS];
291
292        struct pan_shader_info info;
293
294        struct pan_earlyzs_lut earlyzs;
295
296        /* Attached transform feedback program, if one exists */
297        struct panfrost_shader_state *xfb;
298
299        /* Linked varyings, for non-separable programs */
300        struct pan_linkage linkage;
301
302        struct pipe_stream_output_info stream_output;
303        uint64_t so_mask;
304
305        struct panfrost_shader_key key;
306
307        /* Mask of state that dirties the sysvals */
308        unsigned dirty_3d, dirty_shader;
309};
310
311/* A collection of varyings (the CSO) */
312struct panfrost_shader_variants {
313        nir_shader *nir;
314
315        union {
316                struct pipe_stream_output_info stream_output;
317                unsigned req_input_mem;
318        };
319
320        /** Lock for the variants array */
321        simple_mtx_t lock;
322
323        struct panfrost_shader_state *variants;
324        unsigned variant_space;
325
326        unsigned variant_count;
327
328        /* On vertex shaders, bit mask of special desktop-only varyings to link
329         * with the fragment shader. Used on Valhall to implement separable
330         * shaders for desktop GL.
331         */
332        uint32_t fixed_varying_mask;
333
334        /* The current active variant */
335        unsigned active_variant;
336};
337
338/** (Vertex buffer index, divisor) tuple that will become an Attribute Buffer
339 * Descriptor at draw-time on Midgard
340 */
341struct pan_vertex_buffer {
342        unsigned vbi;
343        unsigned divisor;
344};
345
346unsigned
347pan_assign_vertex_buffer(struct pan_vertex_buffer *buffers,
348                         unsigned *nr_bufs,
349                         unsigned vbi,
350                         unsigned divisor);
351
352struct panfrost_zsa_state;
353struct panfrost_sampler_state;
354struct panfrost_sampler_view;
355struct panfrost_vertex_state;
356
357static inline struct panfrost_context *
358pan_context(struct pipe_context *pcontext)
359{
360        return (struct panfrost_context *) pcontext;
361}
362
363static inline struct panfrost_streamout_target *
364pan_so_target(struct pipe_stream_output_target *target)
365{
366        return (struct panfrost_streamout_target *)target;
367}
368
369static inline struct panfrost_shader_state *
370panfrost_get_shader_state(struct panfrost_context *ctx,
371                          enum pipe_shader_type st)
372{
373        struct panfrost_shader_variants *all = ctx->shader[st];
374
375        if (!all)
376                return NULL;
377
378        return &all->variants[all->active_variant];
379}
380
381struct pipe_context *
382panfrost_create_context(struct pipe_screen *screen, void *priv, unsigned flags);
383
384bool
385panfrost_writes_point_size(struct panfrost_context *ctx);
386
387struct panfrost_ptr
388panfrost_vertex_tiler_job(struct panfrost_context *ctx, bool is_tiler);
389
390void
391panfrost_flush(
392        struct pipe_context *pipe,
393        struct pipe_fence_handle **fence,
394        unsigned flags);
395
396bool
397panfrost_render_condition_check(struct panfrost_context *ctx);
398
399void
400panfrost_update_shader_variant(struct panfrost_context *ctx,
401                               enum pipe_shader_type type);
402void
403panfrost_shader_compile(struct pipe_screen *pscreen,
404                        struct panfrost_pool *shader_pool,
405                        struct panfrost_pool *desc_pool,
406                        const nir_shader *ir,
407                        struct panfrost_shader_state *state);
408
409void
410panfrost_analyze_sysvals(struct panfrost_shader_state *ss);
411
412mali_ptr
413panfrost_get_index_buffer(struct panfrost_batch *batch,
414                          const struct pipe_draw_info *info,
415                          const struct pipe_draw_start_count_bias *draw);
416
417mali_ptr
418panfrost_get_index_buffer_bounded(struct panfrost_batch *batch,
419                                  const struct pipe_draw_info *info,
420                                  const struct pipe_draw_start_count_bias *draw,
421                                  unsigned *min_index, unsigned *max_index);
422
423/* Instancing */
424
425mali_ptr
426panfrost_vertex_buffer_address(struct panfrost_context *ctx, unsigned i);
427
428/* Compute */
429
430void
431panfrost_compute_context_init(struct pipe_context *pctx);
432
433static inline void
434panfrost_dirty_state_all(struct panfrost_context *ctx)
435{
436        ctx->dirty = ~0;
437
438        for (unsigned i = 0; i < PIPE_SHADER_TYPES; ++i)
439                ctx->dirty_shader[i] = ~0;
440}
441
442static inline void
443panfrost_clean_state_3d(struct panfrost_context *ctx)
444{
445        ctx->dirty = 0;
446
447        for (unsigned i = 0; i < PIPE_SHADER_TYPES; ++i) {
448                if (i != PIPE_SHADER_COMPUTE)
449                        ctx->dirty_shader[i] = 0;
450        }
451}
452
453void
454panfrost_set_batch_masks_blend(struct panfrost_batch *batch);
455
456void
457panfrost_set_batch_masks_zs(struct panfrost_batch *batch);
458
459void
460panfrost_track_image_access(struct panfrost_batch *batch,
461                            enum pipe_shader_type stage,
462                            struct pipe_image_view *image);
463
464#endif
465