2 * Copyright 2010 Jerome Glisse <glisse@freedesktop.org>
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:
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
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.
29 #include "si_shader.h"
31 #ifdef PIPE_ARCH_BIG_ENDIAN
32 #define SI_BIG_ENDIAN 1
34 #define SI_BIG_ENDIAN 0
37 /* The base vertex and primitive restart can be any number, but we must pick
38 * one which will mean "unknown" for the purpose of state tracking and
39 * the number shouldn't be a commonly-used one. */
40 #define SI_BASE_VERTEX_UNKNOWN INT_MIN
41 #define SI_RESTART_INDEX_UNKNOWN INT_MIN
42 #define SI_NUM_SMOOTH_AA_SAMPLES 8
43 #define SI_GS_PER_ES 128
45 /* Instruction cache. */
46 #define SI_CONTEXT_INV_ICACHE (R600_CONTEXT_PRIVATE_FLAG << 0)
47 /* SMEM L1, other names: KCACHE, constant cache, DCACHE, data cache */
48 #define SI_CONTEXT_INV_SMEM_L1 (R600_CONTEXT_PRIVATE_FLAG << 1)
49 /* VMEM L1 can optionally be bypassed (GLC=1). Other names: TC L1 */
50 #define SI_CONTEXT_INV_VMEM_L1 (R600_CONTEXT_PRIVATE_FLAG << 2)
51 /* Used by everything except CB/DB, can be bypassed (SLC=1). Other names: TC L2 */
52 #define SI_CONTEXT_INV_GLOBAL_L2 (R600_CONTEXT_PRIVATE_FLAG << 3)
53 /* Framebuffer caches. */
54 #define SI_CONTEXT_FLUSH_AND_INV_CB_META (R600_CONTEXT_PRIVATE_FLAG << 4)
55 #define SI_CONTEXT_FLUSH_AND_INV_DB_META (R600_CONTEXT_PRIVATE_FLAG << 5)
56 #define SI_CONTEXT_FLUSH_AND_INV_DB (R600_CONTEXT_PRIVATE_FLAG << 6)
57 #define SI_CONTEXT_FLUSH_AND_INV_CB (R600_CONTEXT_PRIVATE_FLAG << 7)
58 /* Engine synchronization. */
59 #define SI_CONTEXT_VS_PARTIAL_FLUSH (R600_CONTEXT_PRIVATE_FLAG << 8)
60 #define SI_CONTEXT_PS_PARTIAL_FLUSH (R600_CONTEXT_PRIVATE_FLAG << 9)
61 #define SI_CONTEXT_CS_PARTIAL_FLUSH (R600_CONTEXT_PRIVATE_FLAG << 10)
62 #define SI_CONTEXT_VGT_FLUSH (R600_CONTEXT_PRIVATE_FLAG << 11)
63 #define SI_CONTEXT_VGT_STREAMOUT_SYNC (R600_CONTEXT_PRIVATE_FLAG << 12)
65 #define SI_CONTEXT_FLUSH_AND_INV_FRAMEBUFFER (SI_CONTEXT_FLUSH_AND_INV_CB | \
66 SI_CONTEXT_FLUSH_AND_INV_CB_META | \
67 SI_CONTEXT_FLUSH_AND_INV_DB | \
68 SI_CONTEXT_FLUSH_AND_INV_DB_META)
70 #define SI_ENCODE_TRACE_POINT(id) (0xcafe0000 | ((id) & 0xffff))
71 #define SI_IS_TRACE_POINT(x) (((x) & 0xcafe0000) == 0xcafe0000)
72 #define SI_GET_TRACE_POINT_ID(x) ((x) & 0xffff)
74 #define SI_MAX_BORDER_COLORS 4096
78 struct u_suballocator
;
81 struct r600_common_screen b
;
82 unsigned gs_table_depth
;
83 unsigned tess_offchip_block_dw_size
;
84 bool has_distributed_tess
;
85 bool has_draw_indirect_multi
;
88 /* Whether shaders are monolithic (1-part) or separate (3-part). */
89 bool use_monolithic_shaders
;
92 pipe_mutex shader_parts_mutex
;
93 struct si_shader_part
*vs_prologs
;
94 struct si_shader_part
*vs_epilogs
;
95 struct si_shader_part
*tcs_epilogs
;
96 struct si_shader_part
*ps_prologs
;
97 struct si_shader_part
*ps_epilogs
;
99 /* Shader cache in memory.
101 * Design & limitations:
102 * - The shader cache is per screen (= per process), never saved to
103 * disk, and skips redundant shader compilations from TGSI to bytecode.
104 * - It can only be used with one-variant-per-shader support, in which
105 * case only the main (typically middle) part of shaders is cached.
106 * - Only VS, TCS, TES, PS are cached, out of which only the hw VS
107 * variants of VS and TES are cached, so LS and ES aren't.
108 * - GS and CS aren't cached, but it's certainly possible to cache
111 pipe_mutex shader_cache_mutex
;
112 struct hash_table
*shader_cache
;
114 /* Shader compiler queue for multithreaded compilation. */
115 struct util_queue shader_compiler_queue
;
116 LLVMTargetMachineRef tm
[4]; /* used by the queue only */
119 struct si_blend_color
{
120 struct r600_atom atom
;
121 struct pipe_blend_color state
;
124 struct si_sampler_view
{
125 struct pipe_sampler_view base
;
126 /* [0..7] = image descriptor
127 * [4..7] = buffer descriptor */
129 uint32_t fmask_state
[8];
130 const struct radeon_surf_level
*base_level_info
;
132 unsigned block_width
;
133 bool is_stencil_sampler
;
136 struct si_sampler_state
{
140 struct si_cs_shader_state
{
141 struct si_compute
*program
;
142 struct si_compute
*emitted_program
;
148 struct si_textures_info
{
149 struct si_sampler_views views
;
150 uint32_t depth_texture_mask
; /* which textures are depth */
151 uint32_t compressed_colortex_mask
;
154 struct si_images_info
{
155 struct pipe_image_view views
[SI_NUM_IMAGES
];
156 uint32_t compressed_colortex_mask
;
157 unsigned enabled_mask
;
160 struct si_framebuffer
{
161 struct r600_atom atom
;
162 struct pipe_framebuffer_state state
;
164 unsigned log_samples
;
165 unsigned cb0_is_integer
;
166 unsigned compressed_cb_mask
;
167 unsigned spi_shader_col_format
;
168 unsigned spi_shader_col_format_alpha
;
169 unsigned spi_shader_col_format_blend
;
170 unsigned spi_shader_col_format_blend_alpha
;
171 unsigned color_is_int8
; /* bitmask */
172 unsigned dirty_cbufs
;
177 struct si_clip_state
{
178 struct r600_atom atom
;
179 struct pipe_clip_state state
;
182 struct si_sample_locs
{
183 struct r600_atom atom
;
187 struct si_sample_mask
{
188 struct r600_atom atom
;
189 uint16_t sample_mask
;
192 /* A shader state consists of the shader selector, which is a constant state
193 * object shared by multiple contexts and shouldn't be modified, and
194 * the current shader variant selected for this context.
196 struct si_shader_ctx_state
{
197 struct si_shader_selector
*cso
;
198 struct si_shader
*current
;
202 struct r600_common_context b
;
203 struct blitter_context
*blitter
;
204 void *custom_dsa_flush
;
205 void *custom_blend_resolve
;
206 void *custom_blend_decompress
;
207 void *custom_blend_fastclear
;
208 void *custom_blend_dcc_decompress
;
209 struct si_screen
*screen
;
211 struct radeon_winsys_cs
*ce_ib
;
212 struct radeon_winsys_cs
*ce_preamble_ib
;
213 bool ce_need_synchronization
;
214 struct u_suballocator
*ce_suballocator
;
216 struct si_shader_ctx_state fixed_func_tcs_shader
;
217 LLVMTargetMachineRef tm
; /* only non-threaded compilation */
218 bool gfx_flush_in_progress
;
219 bool compute_is_busy
;
221 /* Atoms (direct states). */
222 union si_state_atoms atoms
;
223 unsigned dirty_atoms
; /* mask */
224 /* PM4 states (precomputed immutable states) */
225 union si_state queued
;
226 union si_state emitted
;
228 /* Atom declarations. */
229 struct si_framebuffer framebuffer
;
230 struct si_sample_locs msaa_sample_locs
;
231 struct r600_atom db_render_state
;
232 struct r600_atom msaa_config
;
233 struct si_sample_mask sample_mask
;
234 struct r600_atom cb_render_state
;
235 struct si_blend_color blend_color
;
236 struct r600_atom clip_regs
;
237 struct si_clip_state clip_state
;
238 struct si_shader_data shader_userdata
;
239 struct si_stencil_ref stencil_ref
;
240 struct r600_atom spi_map
;
242 /* Precomputed states. */
243 struct si_pm4_state
*init_config
;
244 struct si_pm4_state
*init_config_gs_rings
;
245 bool init_config_has_vgt_flush
;
246 struct si_pm4_state
*vgt_shader_config
[4];
249 struct si_shader_ctx_state ps_shader
;
250 struct si_shader_ctx_state gs_shader
;
251 struct si_shader_ctx_state vs_shader
;
252 struct si_shader_ctx_state tcs_shader
;
253 struct si_shader_ctx_state tes_shader
;
254 struct si_cs_shader_state cs_shader_state
;
256 /* shader information */
257 struct si_vertex_element
*vertex_elements
;
258 unsigned sprite_coord_enable
;
260 bool do_update_shaders
;
262 /* shader descriptors */
263 struct si_descriptors vertex_buffers
;
264 struct si_descriptors descriptors
[SI_NUM_DESCS
];
265 unsigned descriptors_dirty
;
266 struct si_buffer_resources rw_buffers
;
267 struct si_buffer_resources const_buffers
[SI_NUM_SHADERS
];
268 struct si_buffer_resources shader_buffers
[SI_NUM_SHADERS
];
269 struct si_textures_info samplers
[SI_NUM_SHADERS
];
270 struct si_images_info images
[SI_NUM_SHADERS
];
272 /* other shader resources */
273 struct pipe_constant_buffer null_const_buf
; /* used for set_constant_buffer(NULL) on CIK */
274 struct pipe_resource
*esgs_ring
;
275 struct pipe_resource
*gsvs_ring
;
276 struct pipe_resource
*tf_ring
;
277 struct pipe_resource
*tess_offchip_ring
;
278 union pipe_color_union
*border_color_table
; /* in CPU memory, any endian */
279 struct r600_resource
*border_color_buffer
;
280 union pipe_color_union
*border_color_map
; /* in VRAM (slow access), little endian */
281 unsigned border_color_count
;
283 /* Vertex and index buffers. */
284 bool vertex_buffers_dirty
;
285 struct pipe_index_buffer index_buffer
;
286 struct pipe_vertex_buffer vertex_buffer
[SI_NUM_VERTEX_BUFFERS
];
288 /* MSAA config state. */
290 bool smoothing_enabled
;
292 /* DB render state. */
293 bool dbcb_depth_copy_enabled
;
294 bool dbcb_stencil_copy_enabled
;
295 unsigned dbcb_copy_sample
;
296 bool db_flush_depth_inplace
;
297 bool db_flush_stencil_inplace
;
299 bool db_depth_disable_expclear
;
300 bool db_stencil_clear
;
301 bool db_stencil_disable_expclear
;
302 unsigned ps_db_shader_control
;
303 bool occlusion_queries_disabled
;
305 /* Emitted draw state. */
307 int last_base_vertex
;
308 int last_start_instance
;
310 int last_sh_base_reg
;
311 int last_primitive_restart_en
;
312 int last_restart_index
;
313 int last_gs_out_prim
;
315 int last_multi_vgt_param
;
317 unsigned last_sc_line_stipple
;
318 int last_vtx_reuse_depth
;
319 int current_rast_prim
; /* primitive type after TES, GS */
320 unsigned last_gsvs_itemsize
;
323 struct r600_resource
*scratch_buffer
;
324 bool emit_scratch_reloc
;
325 unsigned scratch_waves
;
326 unsigned spi_tmpring_size
;
328 struct r600_resource
*compute_scratch_buffer
;
330 /* Emitted derived tessellation state. */
331 struct si_shader
*last_ls
; /* local shader (VS) */
332 struct si_shader_selector
*last_tcs
;
333 int last_num_tcs_input_cp
;
334 int last_tes_sh_base
;
338 struct radeon_saved_cs last_gfx
;
339 struct r600_resource
*last_trace_buf
;
340 struct r600_resource
*trace_buf
;
342 uint64_t dmesg_timestamp
;
343 unsigned apitrace_call_number
;
346 bool need_check_render_feedback
;
350 void cik_init_sdma_functions(struct si_context
*sctx
);
353 void si_init_blit_functions(struct si_context
*sctx
);
354 void si_decompress_graphics_textures(struct si_context
*sctx
);
355 void si_decompress_compute_textures(struct si_context
*sctx
);
356 void si_resource_copy_region(struct pipe_context
*ctx
,
357 struct pipe_resource
*dst
,
359 unsigned dstx
, unsigned dsty
, unsigned dstz
,
360 struct pipe_resource
*src
,
362 const struct pipe_box
*src_box
);
365 void si_copy_buffer(struct si_context
*sctx
,
366 struct pipe_resource
*dst
, struct pipe_resource
*src
,
367 uint64_t dst_offset
, uint64_t src_offset
, unsigned size
);
368 void si_init_cp_dma_functions(struct si_context
*sctx
);
371 void si_init_debug_functions(struct si_context
*sctx
);
372 void si_check_vm_faults(struct r600_common_context
*ctx
,
373 struct radeon_saved_cs
*saved
, enum ring_type ring
);
374 bool si_replace_shader(unsigned num
, struct radeon_shader_binary
*binary
);
377 void si_init_dma_functions(struct si_context
*sctx
);
379 /* si_hw_context.c */
380 void si_context_gfx_flush(void *context
, unsigned flags
,
381 struct pipe_fence_handle
**fence
);
382 void si_begin_new_cs(struct si_context
*ctx
);
383 void si_need_cs_space(struct si_context
*ctx
);
386 void si_init_compute_functions(struct si_context
*sctx
);
388 /* si_perfcounters.c */
389 void si_init_perfcounters(struct si_screen
*screen
);
392 struct pipe_video_codec
*si_uvd_create_decoder(struct pipe_context
*context
,
393 const struct pipe_video_codec
*templ
);
395 struct pipe_video_buffer
*si_video_buffer_create(struct pipe_context
*pipe
,
396 const struct pipe_video_buffer
*tmpl
);
402 static inline struct r600_resource
*
403 si_resource_create_custom(struct pipe_screen
*screen
,
404 unsigned usage
, unsigned size
)
407 return r600_resource(pipe_buffer_create(screen
,
408 PIPE_BIND_CUSTOM
, usage
, size
));
412 si_invalidate_draw_sh_constants(struct si_context
*sctx
)
414 sctx
->last_base_vertex
= SI_BASE_VERTEX_UNKNOWN
;
418 si_set_atom_dirty(struct si_context
*sctx
,
419 struct r600_atom
*atom
, bool dirty
)
421 unsigned bit
= 1 << (atom
->id
- 1);
424 sctx
->dirty_atoms
|= bit
;
426 sctx
->dirty_atoms
&= ~bit
;
430 si_is_atom_dirty(struct si_context
*sctx
,
431 struct r600_atom
*atom
)
433 unsigned bit
= 1 << (atom
->id
- 1);
435 return sctx
->dirty_atoms
& bit
;
439 si_mark_atom_dirty(struct si_context
*sctx
,
440 struct r600_atom
*atom
)
442 si_set_atom_dirty(sctx
, atom
, true);
445 static inline struct tgsi_shader_info
*si_get_vs_info(struct si_context
*sctx
)
447 if (sctx
->gs_shader
.cso
)
448 return &sctx
->gs_shader
.cso
->info
;
449 else if (sctx
->tes_shader
.cso
)
450 return &sctx
->tes_shader
.cso
->info
;
451 else if (sctx
->vs_shader
.cso
)
452 return &sctx
->vs_shader
.cso
->info
;
457 static inline struct si_shader
* si_get_vs_state(struct si_context
*sctx
)
459 if (sctx
->gs_shader
.current
)
460 return sctx
->gs_shader
.current
->gs_copy_shader
;
461 else if (sctx
->tes_shader
.current
)
462 return sctx
->tes_shader
.current
;
464 return sctx
->vs_shader
.current
;
467 static inline bool si_vs_exports_prim_id(struct si_shader
*shader
)
469 if (shader
->selector
->type
== PIPE_SHADER_VERTEX
)
470 return shader
->key
.vs
.epilog
.export_prim_id
;
471 else if (shader
->selector
->type
== PIPE_SHADER_TESS_EVAL
)
472 return shader
->key
.tes
.epilog
.export_prim_id
;