radeonsi/gfx9: use CE RAM optimally
[mesa.git] / src / gallium / drivers / radeonsi / si_state.h
1 /*
2 * Copyright 2012 Advanced Micro Devices, Inc.
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 * on the rights to use, copy, modify, merge, publish, distribute, sub
8 * license, and/or sell copies of the Software, and to permit persons to whom
9 * the 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 NON-INFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
19 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
20 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
21 * USE OR OTHER DEALINGS IN THE SOFTWARE.
22 *
23 * Authors:
24 * Christian König <christian.koenig@amd.com>
25 */
26
27 #ifndef SI_STATE_H
28 #define SI_STATE_H
29
30 #include "si_pm4.h"
31 #include "radeon/r600_pipe_common.h"
32
33 #define SI_NUM_GRAPHICS_SHADERS (PIPE_SHADER_TESS_EVAL+1)
34 #define SI_NUM_SHADERS (PIPE_SHADER_COMPUTE+1)
35
36 #define SI_MAX_ATTRIBS 16
37 #define SI_NUM_VERTEX_BUFFERS SI_MAX_ATTRIBS
38 #define SI_NUM_SAMPLERS 32 /* OpenGL textures units per shader */
39 #define SI_NUM_CONST_BUFFERS 16
40 #define SI_NUM_IMAGES 16
41 #define SI_NUM_SHADER_BUFFERS 16
42
43 struct si_screen;
44 struct si_shader;
45 struct si_shader_selector;
46
47 struct si_state_blend {
48 struct si_pm4_state pm4;
49 uint32_t cb_target_mask;
50 bool alpha_to_coverage;
51 bool alpha_to_one;
52 bool dual_src_blend;
53 /* Set 0xf or 0x0 (4 bits) per render target if the following is
54 * true. ANDed with spi_shader_col_format.
55 */
56 unsigned blend_enable_4bit;
57 unsigned need_src_alpha_4bit;
58 };
59
60 struct si_state_rasterizer {
61 struct si_pm4_state pm4;
62 /* poly offset states for 16-bit, 24-bit, and 32-bit zbuffers */
63 struct si_pm4_state pm4_poly_offset[3];
64 bool flatshade;
65 bool two_side;
66 bool multisample_enable;
67 bool force_persample_interp;
68 bool line_stipple_enable;
69 unsigned sprite_coord_enable;
70 unsigned pa_sc_line_stipple;
71 unsigned pa_cl_clip_cntl;
72 unsigned clip_plane_enable;
73 bool poly_stipple_enable;
74 bool line_smooth;
75 bool poly_smooth;
76 bool uses_poly_offset;
77 bool clamp_fragment_color;
78 bool clamp_vertex_color;
79 bool rasterizer_discard;
80 bool scissor_enable;
81 bool clip_halfz;
82 };
83
84 struct si_dsa_stencil_ref_part {
85 uint8_t valuemask[2];
86 uint8_t writemask[2];
87 };
88
89 struct si_state_dsa {
90 struct si_pm4_state pm4;
91 unsigned alpha_func;
92 struct si_dsa_stencil_ref_part stencil_ref;
93 };
94
95 struct si_stencil_ref {
96 struct r600_atom atom;
97 struct pipe_stencil_ref state;
98 struct si_dsa_stencil_ref_part dsa_part;
99 };
100
101 struct si_vertex_element
102 {
103 unsigned count;
104 unsigned first_vb_use_mask;
105 /* Vertex buffer descriptor list size aligned for optimal prefetch. */
106 unsigned desc_list_byte_size;
107
108 uint8_t fix_fetch[SI_MAX_ATTRIBS];
109 uint32_t rsrc_word3[SI_MAX_ATTRIBS];
110 uint32_t format_size[SI_MAX_ATTRIBS];
111 struct pipe_vertex_element elements[SI_MAX_ATTRIBS];
112 };
113
114 union si_state {
115 struct {
116 struct si_state_blend *blend;
117 struct si_state_rasterizer *rasterizer;
118 struct si_state_dsa *dsa;
119 struct si_pm4_state *poly_offset;
120 struct si_pm4_state *ls;
121 struct si_pm4_state *hs;
122 struct si_pm4_state *es;
123 struct si_pm4_state *gs;
124 struct si_pm4_state *vgt_shader_config;
125 struct si_pm4_state *vs;
126 struct si_pm4_state *ps;
127 } named;
128 struct si_pm4_state *array[0];
129 };
130
131 #define SI_NUM_STATES (sizeof(union si_state) / sizeof(struct si_pm4_state *))
132
133 union si_state_atoms {
134 struct {
135 /* The order matters. */
136 struct r600_atom *prefetch_L2;
137 struct r600_atom *render_cond;
138 struct r600_atom *streamout_begin;
139 struct r600_atom *streamout_enable; /* must be after streamout_begin */
140 struct r600_atom *framebuffer;
141 struct r600_atom *msaa_sample_locs;
142 struct r600_atom *db_render_state;
143 struct r600_atom *msaa_config;
144 struct r600_atom *sample_mask;
145 struct r600_atom *cb_render_state;
146 struct r600_atom *blend_color;
147 struct r600_atom *clip_regs;
148 struct r600_atom *clip_state;
149 struct r600_atom *shader_userdata;
150 struct r600_atom *scissors;
151 struct r600_atom *viewports;
152 struct r600_atom *stencil_ref;
153 struct r600_atom *spi_map;
154 struct r600_atom *scratch_state;
155 } s;
156 struct r600_atom *array[0];
157 };
158
159 #define SI_NUM_ATOMS (sizeof(union si_state_atoms)/sizeof(struct r600_atom*))
160
161 struct si_shader_data {
162 struct r600_atom atom;
163 uint32_t sh_base[SI_NUM_SHADERS];
164 };
165
166 /* Private read-write buffer slots. */
167 enum {
168 SI_ES_RING_ESGS,
169 SI_GS_RING_ESGS,
170
171 SI_RING_GSVS,
172
173 SI_VS_STREAMOUT_BUF0,
174 SI_VS_STREAMOUT_BUF1,
175 SI_VS_STREAMOUT_BUF2,
176 SI_VS_STREAMOUT_BUF3,
177
178 SI_HS_CONST_DEFAULT_TESS_LEVELS,
179 SI_VS_CONST_CLIP_PLANES,
180 SI_PS_CONST_POLY_STIPPLE,
181 SI_PS_CONST_SAMPLE_POSITIONS,
182
183 SI_NUM_RW_BUFFERS,
184 };
185
186 /* Indices into sctx->descriptors, laid out so that gfx and compute pipelines
187 * are contiguous:
188 *
189 * 0 - rw buffers
190 * 1 - vertex const buffers
191 * 2 - vertex shader buffers
192 * ...
193 * 5 - fragment const buffers
194 * ...
195 * 21 - compute const buffers
196 * ...
197 */
198 enum {
199 SI_SHADER_DESCS_CONST_AND_SHADER_BUFFERS,
200 SI_SHADER_DESCS_SAMPLERS_AND_IMAGES,
201 SI_NUM_SHADER_DESCS,
202 };
203
204 #define SI_DESCS_RW_BUFFERS 0
205 #define SI_DESCS_FIRST_SHADER 1
206 #define SI_DESCS_FIRST_COMPUTE (SI_DESCS_FIRST_SHADER + \
207 PIPE_SHADER_COMPUTE * SI_NUM_SHADER_DESCS)
208 #define SI_NUM_DESCS (SI_DESCS_FIRST_SHADER + \
209 SI_NUM_SHADERS * SI_NUM_SHADER_DESCS)
210
211 /* This represents descriptors in memory, such as buffer resources,
212 * image resources, and sampler states.
213 */
214 struct si_descriptors {
215 /* The list of descriptors in malloc'd memory. */
216 uint32_t *list;
217 /* The list in mapped GPU memory. */
218 uint32_t *gpu_list;
219 /* The size of one descriptor. */
220 unsigned element_dw_size;
221 /* The maximum number of descriptors. */
222 unsigned num_elements;
223
224 /* The buffer where the descriptors have been uploaded. */
225 struct r600_resource *buffer;
226 int buffer_offset; /* can be negative if not using lower slots */
227
228 /* Offset in CE RAM */
229 unsigned ce_offset;
230
231 /* Slots allocated in CE RAM. If we get active slots outside of this
232 * range, direct uploads to memory will be used instead. This basically
233 * governs switching between onchip (CE) and offchip (upload) modes.
234 */
235 unsigned first_ce_slot;
236 unsigned num_ce_slots;
237
238 /* Slots that are used by currently-bound shaders.
239 * With CE: It determines which slots are dumped to L2.
240 * It doesn't skip uploads to CE RAM.
241 * Without CE: It determines which slots are uploaded.
242 */
243 unsigned first_active_slot;
244 unsigned num_active_slots;
245
246 /* Slots that have been changed and need to be uploaded. */
247 uint64_t dirty_mask;
248
249 /* Whether CE is used to upload this descriptor array. */
250 bool uses_ce;
251
252 /* The shader userdata offset within a shader where the 64-bit pointer to the descriptor
253 * array will be stored. */
254 unsigned shader_userdata_offset;
255 };
256
257 struct si_sampler_views {
258 struct pipe_sampler_view *views[SI_NUM_SAMPLERS];
259 struct si_sampler_state *sampler_states[SI_NUM_SAMPLERS];
260
261 /* The i-th bit is set if that element is enabled (non-NULL resource). */
262 unsigned enabled_mask;
263 };
264
265 struct si_buffer_resources {
266 enum radeon_bo_usage shader_usage; /* READ, WRITE, or READWRITE */
267 enum radeon_bo_usage shader_usage_constbuf;
268 enum radeon_bo_priority priority;
269 enum radeon_bo_priority priority_constbuf;
270 struct pipe_resource **buffers; /* this has num_buffers elements */
271
272 /* The i-th bit is set if that element is enabled (non-NULL resource). */
273 unsigned enabled_mask;
274 };
275
276 #define si_pm4_block_idx(member) \
277 (offsetof(union si_state, named.member) / sizeof(struct si_pm4_state *))
278
279 #define si_pm4_state_changed(sctx, member) \
280 ((sctx)->queued.named.member != (sctx)->emitted.named.member)
281
282 #define si_pm4_bind_state(sctx, member, value) \
283 do { \
284 (sctx)->queued.named.member = (value); \
285 (sctx)->dirty_states |= 1 << si_pm4_block_idx(member); \
286 } while(0)
287
288 #define si_pm4_delete_state(sctx, member, value) \
289 do { \
290 if ((sctx)->queued.named.member == (value)) { \
291 (sctx)->queued.named.member = NULL; \
292 } \
293 si_pm4_free_state(sctx, (struct si_pm4_state *)(value), \
294 si_pm4_block_idx(member)); \
295 } while(0)
296
297 /* si_descriptors.c */
298 void si_ce_save_all_descriptors_at_ib_end(struct si_context* sctx);
299 void si_ce_restore_all_descriptors_at_ib_start(struct si_context *sctx);
300 void si_ce_enable_loads(struct radeon_winsys_cs *ib);
301 void si_set_mutable_tex_desc_fields(struct si_screen *sscreen,
302 struct r600_texture *tex,
303 const struct legacy_surf_level *base_level_info,
304 unsigned base_level, unsigned first_level,
305 unsigned block_width, bool is_stencil,
306 uint32_t *state);
307 void si_get_pipe_constant_buffer(struct si_context *sctx, uint shader,
308 uint slot, struct pipe_constant_buffer *cbuf);
309 void si_get_shader_buffers(struct si_context *sctx,
310 enum pipe_shader_type shader,
311 uint start_slot, uint count,
312 struct pipe_shader_buffer *sbuf);
313 void si_set_ring_buffer(struct pipe_context *ctx, uint slot,
314 struct pipe_resource *buffer,
315 unsigned stride, unsigned num_records,
316 bool add_tid, bool swizzle,
317 unsigned element_size, unsigned index_stride, uint64_t offset);
318 void si_init_all_descriptors(struct si_context *sctx);
319 bool si_upload_vertex_buffer_descriptors(struct si_context *sctx);
320 bool si_upload_graphics_shader_descriptors(struct si_context *sctx);
321 bool si_upload_compute_shader_descriptors(struct si_context *sctx);
322 void si_release_all_descriptors(struct si_context *sctx);
323 void si_all_descriptors_begin_new_cs(struct si_context *sctx);
324 void si_upload_const_buffer(struct si_context *sctx, struct r600_resource **rbuffer,
325 const uint8_t *ptr, unsigned size, uint32_t *const_offset);
326 void si_update_all_texture_descriptors(struct si_context *sctx);
327 void si_shader_change_notify(struct si_context *sctx);
328 void si_update_compressed_colortex_masks(struct si_context *sctx);
329 void si_emit_graphics_shader_userdata(struct si_context *sctx,
330 struct r600_atom *atom);
331 void si_emit_compute_shader_userdata(struct si_context *sctx);
332 void si_set_rw_buffer(struct si_context *sctx,
333 uint slot, const struct pipe_constant_buffer *input);
334 void si_set_active_descriptors(struct si_context *sctx, unsigned desc_idx,
335 uint64_t new_active_mask);
336 void si_set_active_descriptors_for_shader(struct si_context *sctx,
337 struct si_shader_selector *sel);
338
339 /* si_state.c */
340 struct si_shader_selector;
341
342 void si_init_atom(struct si_context *sctx, struct r600_atom *atom,
343 struct r600_atom **list_elem,
344 void (*emit_func)(struct si_context *ctx, struct r600_atom *state));
345 void si_init_state_functions(struct si_context *sctx);
346 void si_init_screen_state_functions(struct si_screen *sscreen);
347 void
348 si_make_buffer_descriptor(struct si_screen *screen, struct r600_resource *buf,
349 enum pipe_format format,
350 unsigned offset, unsigned size,
351 uint32_t *state);
352 void
353 si_make_texture_descriptor(struct si_screen *screen,
354 struct r600_texture *tex,
355 bool sampler,
356 enum pipe_texture_target target,
357 enum pipe_format pipe_format,
358 const unsigned char state_swizzle[4],
359 unsigned first_level, unsigned last_level,
360 unsigned first_layer, unsigned last_layer,
361 unsigned width, unsigned height, unsigned depth,
362 uint32_t *state,
363 uint32_t *fmask_state);
364 struct pipe_sampler_view *
365 si_create_sampler_view_custom(struct pipe_context *ctx,
366 struct pipe_resource *texture,
367 const struct pipe_sampler_view *state,
368 unsigned width0, unsigned height0,
369 unsigned force_level);
370
371 /* si_state_shader.c */
372 bool si_update_shaders(struct si_context *sctx);
373 void si_init_shader_functions(struct si_context *sctx);
374 bool si_init_shader_cache(struct si_screen *sscreen);
375 void si_destroy_shader_cache(struct si_screen *sscreen);
376 void si_init_shader_selector_async(void *job, int thread_index);
377 void si_get_active_slot_masks(const struct tgsi_shader_info *info,
378 uint32_t *const_and_shader_buffers,
379 uint64_t *samplers_and_images);
380
381 /* si_state_draw.c */
382 void si_init_ia_multi_vgt_param_table(struct si_context *sctx);
383 void si_emit_cache_flush(struct si_context *sctx);
384 void si_ce_pre_draw_synchronization(struct si_context *sctx);
385 void si_ce_post_draw_synchronization(struct si_context *sctx);
386 void si_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *dinfo);
387 void si_trace_emit(struct si_context *sctx);
388
389
390 static inline unsigned
391 si_tile_mode_index(struct r600_texture *rtex, unsigned level, bool stencil)
392 {
393 if (stencil)
394 return rtex->surface.u.legacy.stencil_tiling_index[level];
395 else
396 return rtex->surface.u.legacy.tiling_index[level];
397 }
398
399 static inline unsigned si_get_constbuf_slot(unsigned slot)
400 {
401 /* Constant buffers are in slots [16..31], ascending */
402 return SI_NUM_SHADER_BUFFERS + slot;
403 }
404
405 static inline unsigned si_get_shaderbuf_slot(unsigned slot)
406 {
407 /* shader buffers are in slots [15..0], descending */
408 return SI_NUM_SHADER_BUFFERS - 1 - slot;
409 }
410
411 static inline unsigned si_get_sampler_slot(unsigned slot)
412 {
413 /* samplers are in slots [8..39], ascending */
414 return SI_NUM_IMAGES / 2 + slot;
415 }
416
417 static inline unsigned si_get_image_slot(unsigned slot)
418 {
419 /* images are in slots [15..0] (sampler slots [7..0]), descending */
420 return SI_NUM_IMAGES - 1 - slot;
421 }
422
423 #endif