radeonsi: don't execute LDS stores for TCS outputs that are never read
[mesa.git] / 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 #include "pan_blend.h"
35 #include "pan_encoder.h"
36 #include "pan_texture.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
48 #include "midgard/midgard_compile.h"
49 #include "compiler/shader_enums.h"
50
51 /* Forward declare to avoid extra header dep */
52 struct prim_convert_context;
53
54 #define MAX_VARYINGS 4096
55
56 #define SET_BIT(lval, bit, cond) \
57 if (cond) \
58 lval |= (bit); \
59 else \
60 lval &= ~(bit);
61
62 struct panfrost_constant_buffer {
63 struct pipe_constant_buffer cb[PIPE_MAX_CONSTANT_BUFFERS];
64 uint32_t enabled_mask;
65 uint32_t dirty_mask;
66 };
67
68 struct panfrost_query {
69 /* Passthrough from Gallium */
70 unsigned type;
71 unsigned index;
72
73 /* For computed queries. 64-bit to prevent overflow */
74 struct {
75 uint64_t start;
76 uint64_t end;
77 };
78
79 /* Memory for the GPU to writeback the value of the query */
80 struct panfrost_bo *bo;
81 };
82
83 struct panfrost_fence {
84 struct pipe_reference reference;
85 uint32_t syncobj;
86 bool signaled;
87 };
88
89 struct panfrost_streamout_target {
90 struct pipe_stream_output_target base;
91 uint32_t offset;
92 };
93
94 struct panfrost_streamout {
95 struct pipe_stream_output_target *targets[PIPE_MAX_SO_BUFFERS];
96 unsigned num_targets;
97 };
98
99 struct panfrost_context {
100 /* Gallium context */
101 struct pipe_context base;
102
103 /* Upload manager for small resident GPU-internal data structures, like
104 * sampler descriptors. We use an upload manager since the minimum BO
105 * size from the kernel is 4kb */
106 struct u_upload_mgr *state_uploader;
107
108 /* Bound job batch and map of panfrost_batch_key to job batches */
109 struct panfrost_batch *batch;
110 struct hash_table *batches;
111
112 /* panfrost_bo -> panfrost_bo_access */
113 struct hash_table *accessed_bos;
114
115 /* Within a launch_grid call.. */
116 const struct pipe_grid_info *compute_grid;
117
118 /* Bit mask for supported PIPE_DRAW for this hardware */
119 unsigned draw_modes;
120
121 struct pipe_framebuffer_state pipe_framebuffer;
122 struct panfrost_streamout streamout;
123
124 bool active_queries;
125 uint64_t prims_generated;
126 uint64_t tf_prims_generated;
127 struct panfrost_query *occlusion_query;
128
129 unsigned vertex_count;
130 unsigned instance_count;
131 unsigned offset_start;
132 enum pipe_prim_type active_prim;
133
134 /* If instancing is enabled, vertex count padded for instance; if
135 * it is disabled, just equal to plain vertex count */
136 unsigned padded_count;
137
138 /* TODO: Multiple uniform buffers (index =/= 0), finer updates? */
139
140 struct panfrost_constant_buffer constant_buffer[PIPE_SHADER_TYPES];
141
142 struct panfrost_rasterizer *rasterizer;
143 struct panfrost_shader_variants *shader[PIPE_SHADER_TYPES];
144 struct panfrost_vertex_state *vertex;
145
146 struct pipe_vertex_buffer vertex_buffers[PIPE_MAX_ATTRIBS];
147 uint32_t vb_mask;
148
149 struct pipe_shader_buffer ssbo[PIPE_SHADER_TYPES][PIPE_MAX_SHADER_BUFFERS];
150 uint32_t ssbo_mask[PIPE_SHADER_TYPES];
151
152 struct panfrost_sampler_state *samplers[PIPE_SHADER_TYPES][PIPE_MAX_SAMPLERS];
153 unsigned sampler_count[PIPE_SHADER_TYPES];
154
155 struct panfrost_sampler_view *sampler_views[PIPE_SHADER_TYPES][PIPE_MAX_SHADER_SAMPLER_VIEWS];
156 unsigned sampler_view_count[PIPE_SHADER_TYPES];
157
158 struct primconvert_context *primconvert;
159 struct blitter_context *blitter;
160
161 /* Blitting the wallpaper (the old contents of the framebuffer back to
162 * itself) uses a dedicated u_blitter instance versus general blit()
163 * callbacks from Gallium, as the blit() callback can trigger
164 * wallpapering without Gallium realising, which in turns u_blitter
165 * errors due to unsupported reucrsion */
166
167 struct blitter_context *blitter_wallpaper;
168 struct panfrost_batch *wallpaper_batch;
169
170 struct panfrost_blend_state *blend;
171
172 struct pipe_viewport_state pipe_viewport;
173 struct pipe_scissor_state scissor;
174 struct pipe_blend_color blend_color;
175 struct panfrost_zsa_state *depth_stencil;
176 struct pipe_stencil_ref stencil_ref;
177 unsigned sample_mask;
178 unsigned min_samples;
179
180 struct panfrost_blend_state blit_blend;
181 };
182
183 /* Corresponds to the CSO */
184
185 struct panfrost_rasterizer {
186 struct pipe_rasterizer_state base;
187 };
188
189 /* Variants bundle together to form the backing CSO, bundling multiple
190 * shaders with varying emulated features baked in */
191
192 /* A shader state corresponds to the actual, current variant of the shader */
193 struct panfrost_shader_state {
194 /* Compiled, mapped descriptor, ready for the hardware */
195 bool compiled;
196
197 /* Uploaded shader descriptor (TODO: maybe stuff the packed unuploaded
198 * bits in a union to save some memory?) */
199
200 struct {
201 struct pipe_resource *rsrc;
202 uint32_t offset;
203 } upload;
204
205 struct mali_shader_packed shader;
206 struct mali_midgard_properties_packed properties;
207 struct mali_preload_packed preload;
208
209 /* Non-descript information */
210 unsigned uniform_count;
211 unsigned work_reg_count;
212 bool can_discard;
213 bool writes_point_size;
214 bool writes_depth;
215 bool writes_stencil;
216 bool reads_point_coord;
217 bool reads_face;
218 bool reads_frag_coord;
219 bool writes_global;
220 unsigned stack_size;
221 unsigned shared_size;
222
223 /* Does the fragment shader have side effects? In particular, if output
224 * is masked out, is it legal to skip shader execution? */
225 bool fs_sidefx;
226
227 /* For Bifrost - output type for each RT */
228 enum bifrost_shader_type blend_types[BIFROST_MAX_RENDER_TARGET_COUNT];
229
230 unsigned attribute_count, varying_count, ubo_count;
231 enum mali_format varyings[PIPE_MAX_ATTRIBS];
232 gl_varying_slot varyings_loc[PIPE_MAX_ATTRIBS];
233 struct pipe_stream_output_info stream_output;
234 uint64_t so_mask;
235
236 unsigned sysval_count;
237 unsigned sysval[MAX_SYSVAL_COUNT];
238
239 /* Should we enable helper invocations */
240 bool helper_invocations;
241
242 /* GPU-executable memory */
243 struct panfrost_bo *bo;
244
245 BITSET_WORD outputs_read;
246 enum pipe_format rt_formats[8];
247 };
248
249 /* A collection of varyings (the CSO) */
250 struct panfrost_shader_variants {
251 /* A panfrost_shader_variants can represent a shader for
252 * either graphics or compute */
253
254 bool is_compute;
255
256 union {
257 struct pipe_shader_state base;
258 struct pipe_compute_state cbase;
259 };
260
261 struct panfrost_shader_state *variants;
262 unsigned variant_space;
263
264 unsigned variant_count;
265
266 /* The current active variant */
267 unsigned active_variant;
268 };
269
270 struct panfrost_vertex_state {
271 unsigned num_elements;
272
273 struct pipe_vertex_element pipe[PIPE_MAX_ATTRIBS];
274 unsigned formats[PIPE_MAX_ATTRIBS];
275 };
276
277 struct panfrost_zsa_state {
278 struct pipe_depth_stencil_alpha_state base;
279
280 /* Precomputed stencil state */
281 struct mali_stencil_packed stencil_front;
282 struct mali_stencil_packed stencil_back;
283 u8 stencil_mask_front;
284 u8 stencil_mask_back;
285 };
286
287 struct panfrost_sampler_state {
288 struct pipe_sampler_state base;
289 struct mali_midgard_sampler_packed hw;
290 };
291
292 /* Misnomer: Sampler view corresponds to textures, not samplers */
293
294 struct panfrost_sampler_view {
295 struct pipe_sampler_view base;
296 struct panfrost_bo *bo;
297 struct mali_bifrost_texture_packed bifrost_descriptor;
298 mali_ptr texture_bo;
299 uint64_t modifier;
300 };
301
302 static inline struct panfrost_context *
303 pan_context(struct pipe_context *pcontext)
304 {
305 return (struct panfrost_context *) pcontext;
306 }
307
308 static inline struct panfrost_streamout_target *
309 pan_so_target(struct pipe_stream_output_target *target)
310 {
311 return (struct panfrost_streamout_target *)target;
312 }
313
314 static inline struct panfrost_shader_state *
315 panfrost_get_shader_state(struct panfrost_context *ctx,
316 enum pipe_shader_type st)
317 {
318 struct panfrost_shader_variants *all = ctx->shader[st];
319
320 if (!all)
321 return NULL;
322
323 return &all->variants[all->active_variant];
324 }
325
326 struct pipe_context *
327 panfrost_create_context(struct pipe_screen *screen, void *priv, unsigned flags);
328
329 bool
330 panfrost_writes_point_size(struct panfrost_context *ctx);
331
332 struct panfrost_transfer
333 panfrost_vertex_tiler_job(struct panfrost_context *ctx, bool is_tiler);
334
335 void
336 panfrost_flush(
337 struct pipe_context *pipe,
338 struct pipe_fence_handle **fence,
339 unsigned flags);
340
341 mali_ptr panfrost_sfbd_fragment(struct panfrost_batch *batch, bool has_draws);
342 mali_ptr panfrost_mfbd_fragment(struct panfrost_batch *batch, bool has_draws);
343
344 void
345 panfrost_attach_mfbd(struct panfrost_batch *batch, unsigned vertex_count);
346
347 void
348 panfrost_attach_sfbd(struct panfrost_batch *batch, unsigned vertex_count);
349
350 struct midgard_tiler_descriptor
351 panfrost_emit_midg_tiler(struct panfrost_batch *batch, unsigned vertex_count);
352
353 mali_ptr
354 panfrost_fragment_job(struct panfrost_batch *batch, bool has_draws);
355
356 void
357 panfrost_shader_compile(struct panfrost_context *ctx,
358 enum pipe_shader_ir ir_type,
359 const void *ir,
360 gl_shader_stage stage,
361 struct panfrost_shader_state *state,
362 uint64_t *outputs_written);
363
364 void
365 panfrost_create_sampler_view_bo(struct panfrost_sampler_view *so,
366 struct pipe_context *pctx,
367 struct pipe_resource *texture);
368
369 /* Instancing */
370
371 mali_ptr
372 panfrost_vertex_buffer_address(struct panfrost_context *ctx, unsigned i);
373
374 /* Compute */
375
376 void
377 panfrost_compute_context_init(struct pipe_context *pctx);
378
379
380 #endif