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 /* Write dirty L2 lines back to memory (shader and CP DMA stores), but don't
54 * invalidate L2. SI-CIK can't do it, so they will do complete invalidation. */
55 #define SI_CONTEXT_WRITEBACK_GLOBAL_L2 (R600_CONTEXT_PRIVATE_FLAG << 4)
56 /* Framebuffer caches. */
57 #define SI_CONTEXT_FLUSH_AND_INV_CB_META (R600_CONTEXT_PRIVATE_FLAG << 5)
58 #define SI_CONTEXT_FLUSH_AND_INV_DB_META (R600_CONTEXT_PRIVATE_FLAG << 6)
59 #define SI_CONTEXT_FLUSH_AND_INV_DB (R600_CONTEXT_PRIVATE_FLAG << 7)
60 #define SI_CONTEXT_FLUSH_AND_INV_CB (R600_CONTEXT_PRIVATE_FLAG << 8)
61 /* Engine synchronization. */
62 #define SI_CONTEXT_VS_PARTIAL_FLUSH (R600_CONTEXT_PRIVATE_FLAG << 9)
63 #define SI_CONTEXT_PS_PARTIAL_FLUSH (R600_CONTEXT_PRIVATE_FLAG << 10)
64 #define SI_CONTEXT_CS_PARTIAL_FLUSH (R600_CONTEXT_PRIVATE_FLAG << 11)
65 #define SI_CONTEXT_VGT_FLUSH (R600_CONTEXT_PRIVATE_FLAG << 12)
66 #define SI_CONTEXT_VGT_STREAMOUT_SYNC (R600_CONTEXT_PRIVATE_FLAG << 13)
68 #define SI_CONTEXT_FLUSH_AND_INV_FRAMEBUFFER (SI_CONTEXT_FLUSH_AND_INV_CB | \
69 SI_CONTEXT_FLUSH_AND_INV_CB_META | \
70 SI_CONTEXT_FLUSH_AND_INV_DB | \
71 SI_CONTEXT_FLUSH_AND_INV_DB_META)
73 #define SI_ENCODE_TRACE_POINT(id) (0xcafe0000 | ((id) & 0xffff))
74 #define SI_IS_TRACE_POINT(x) (((x) & 0xcafe0000) == 0xcafe0000)
75 #define SI_GET_TRACE_POINT_ID(x) ((x) & 0xffff)
77 #define SI_MAX_BORDER_COLORS 4096
81 struct u_suballocator
;
84 struct r600_common_screen b
;
85 unsigned gs_table_depth
;
86 unsigned tess_offchip_block_dw_size
;
87 bool has_distributed_tess
;
88 bool has_draw_indirect_multi
;
91 /* Whether shaders are monolithic (1-part) or separate (3-part). */
92 bool use_monolithic_shaders
;
95 pipe_mutex shader_parts_mutex
;
96 struct si_shader_part
*vs_prologs
;
97 struct si_shader_part
*vs_epilogs
;
98 struct si_shader_part
*tcs_epilogs
;
99 struct si_shader_part
*gs_prologs
;
100 struct si_shader_part
*ps_prologs
;
101 struct si_shader_part
*ps_epilogs
;
103 /* Shader cache in memory.
105 * Design & limitations:
106 * - The shader cache is per screen (= per process), never saved to
107 * disk, and skips redundant shader compilations from TGSI to bytecode.
108 * - It can only be used with one-variant-per-shader support, in which
109 * case only the main (typically middle) part of shaders is cached.
110 * - Only VS, TCS, TES, PS are cached, out of which only the hw VS
111 * variants of VS and TES are cached, so LS and ES aren't.
112 * - GS and CS aren't cached, but it's certainly possible to cache
115 pipe_mutex shader_cache_mutex
;
116 struct hash_table
*shader_cache
;
118 /* Shader compiler queue for multithreaded compilation. */
119 struct util_queue shader_compiler_queue
;
120 LLVMTargetMachineRef tm
[4]; /* used by the queue only */
123 struct si_blend_color
{
124 struct r600_atom atom
;
125 struct pipe_blend_color state
;
128 struct si_sampler_view
{
129 struct pipe_sampler_view base
;
130 /* [0..7] = image descriptor
131 * [4..7] = buffer descriptor */
133 uint32_t fmask_state
[8];
134 const struct radeon_surf_level
*base_level_info
;
136 unsigned block_width
;
137 bool is_stencil_sampler
;
140 #define SI_SAMPLER_STATE_MAGIC 0x34f1c35a
142 struct si_sampler_state
{
149 struct si_cs_shader_state
{
150 struct si_compute
*program
;
151 struct si_compute
*emitted_program
;
157 struct si_textures_info
{
158 struct si_sampler_views views
;
159 uint32_t depth_texture_mask
; /* which textures are depth */
160 uint32_t compressed_colortex_mask
;
163 struct si_images_info
{
164 struct pipe_image_view views
[SI_NUM_IMAGES
];
165 uint32_t compressed_colortex_mask
;
166 unsigned enabled_mask
;
169 struct si_framebuffer
{
170 struct r600_atom atom
;
171 struct pipe_framebuffer_state state
;
173 unsigned log_samples
;
174 unsigned compressed_cb_mask
;
175 unsigned colorbuf_enabled_4bit
;
176 unsigned spi_shader_col_format
;
177 unsigned spi_shader_col_format_alpha
;
178 unsigned spi_shader_col_format_blend
;
179 unsigned spi_shader_col_format_blend_alpha
;
180 unsigned color_is_int8
; /* bitmask */
181 unsigned dirty_cbufs
;
186 struct si_clip_state
{
187 struct r600_atom atom
;
188 struct pipe_clip_state state
;
191 struct si_sample_locs
{
192 struct r600_atom atom
;
196 struct si_sample_mask
{
197 struct r600_atom atom
;
198 uint16_t sample_mask
;
201 /* A shader state consists of the shader selector, which is a constant state
202 * object shared by multiple contexts and shouldn't be modified, and
203 * the current shader variant selected for this context.
205 struct si_shader_ctx_state
{
206 struct si_shader_selector
*cso
;
207 struct si_shader
*current
;
211 struct r600_common_context b
;
212 struct blitter_context
*blitter
;
213 void *custom_dsa_flush
;
214 void *custom_blend_resolve
;
215 void *custom_blend_decompress
;
216 void *custom_blend_fastclear
;
217 void *custom_blend_dcc_decompress
;
218 struct si_screen
*screen
;
220 struct radeon_winsys_cs
*ce_ib
;
221 struct radeon_winsys_cs
*ce_preamble_ib
;
222 bool ce_need_synchronization
;
223 struct u_suballocator
*ce_suballocator
;
225 struct si_shader_ctx_state fixed_func_tcs_shader
;
226 LLVMTargetMachineRef tm
; /* only non-threaded compilation */
227 bool gfx_flush_in_progress
;
228 bool compute_is_busy
;
230 /* Atoms (direct states). */
231 union si_state_atoms atoms
;
232 unsigned dirty_atoms
; /* mask */
233 /* PM4 states (precomputed immutable states) */
234 union si_state queued
;
235 union si_state emitted
;
237 /* Atom declarations. */
238 struct si_framebuffer framebuffer
;
239 struct si_sample_locs msaa_sample_locs
;
240 struct r600_atom db_render_state
;
241 struct r600_atom msaa_config
;
242 struct si_sample_mask sample_mask
;
243 struct r600_atom cb_render_state
;
244 struct si_blend_color blend_color
;
245 struct r600_atom clip_regs
;
246 struct si_clip_state clip_state
;
247 struct si_shader_data shader_userdata
;
248 struct si_stencil_ref stencil_ref
;
249 struct r600_atom spi_map
;
251 /* Precomputed states. */
252 struct si_pm4_state
*init_config
;
253 struct si_pm4_state
*init_config_gs_rings
;
254 bool init_config_has_vgt_flush
;
255 struct si_pm4_state
*vgt_shader_config
[4];
258 struct si_shader_ctx_state ps_shader
;
259 struct si_shader_ctx_state gs_shader
;
260 struct si_shader_ctx_state vs_shader
;
261 struct si_shader_ctx_state tcs_shader
;
262 struct si_shader_ctx_state tes_shader
;
263 struct si_cs_shader_state cs_shader_state
;
265 /* shader information */
266 struct si_vertex_element
*vertex_elements
;
267 unsigned sprite_coord_enable
;
269 bool do_update_shaders
;
271 /* shader descriptors */
272 struct si_descriptors vertex_buffers
;
273 struct si_descriptors descriptors
[SI_NUM_DESCS
];
274 unsigned descriptors_dirty
;
275 struct si_buffer_resources rw_buffers
;
276 struct si_buffer_resources const_buffers
[SI_NUM_SHADERS
];
277 struct si_buffer_resources shader_buffers
[SI_NUM_SHADERS
];
278 struct si_textures_info samplers
[SI_NUM_SHADERS
];
279 struct si_images_info images
[SI_NUM_SHADERS
];
281 /* other shader resources */
282 struct pipe_constant_buffer null_const_buf
; /* used for set_constant_buffer(NULL) on CIK */
283 struct pipe_resource
*esgs_ring
;
284 struct pipe_resource
*gsvs_ring
;
285 struct pipe_resource
*tf_ring
;
286 struct pipe_resource
*tess_offchip_ring
;
287 union pipe_color_union
*border_color_table
; /* in CPU memory, any endian */
288 struct r600_resource
*border_color_buffer
;
289 union pipe_color_union
*border_color_map
; /* in VRAM (slow access), little endian */
290 unsigned border_color_count
;
292 /* Vertex and index buffers. */
293 bool vertex_buffers_dirty
;
294 struct pipe_index_buffer index_buffer
;
295 struct pipe_vertex_buffer vertex_buffer
[SI_NUM_VERTEX_BUFFERS
];
297 /* MSAA config state. */
299 bool smoothing_enabled
;
301 /* DB render state. */
302 bool dbcb_depth_copy_enabled
;
303 bool dbcb_stencil_copy_enabled
;
304 unsigned dbcb_copy_sample
;
305 bool db_flush_depth_inplace
;
306 bool db_flush_stencil_inplace
;
308 bool db_depth_disable_expclear
;
309 bool db_stencil_clear
;
310 bool db_stencil_disable_expclear
;
311 unsigned ps_db_shader_control
;
312 bool occlusion_queries_disabled
;
314 /* Emitted draw state. */
316 int last_base_vertex
;
317 int last_start_instance
;
319 int last_sh_base_reg
;
320 int last_primitive_restart_en
;
321 int last_restart_index
;
322 int last_gs_out_prim
;
324 int last_multi_vgt_param
;
326 unsigned last_sc_line_stipple
;
327 int last_vtx_reuse_depth
;
328 int current_rast_prim
; /* primitive type after TES, GS */
329 bool gs_tri_strip_adj_fix
;
332 struct r600_resource
*scratch_buffer
;
333 bool emit_scratch_reloc
;
334 unsigned scratch_waves
;
335 unsigned spi_tmpring_size
;
337 struct r600_resource
*compute_scratch_buffer
;
339 /* Emitted derived tessellation state. */
340 struct si_shader
*last_ls
; /* local shader (VS) */
341 struct si_shader_selector
*last_tcs
;
342 int last_num_tcs_input_cp
;
343 int last_tes_sh_base
;
344 unsigned last_num_patches
;
348 struct radeon_saved_cs last_gfx
;
349 struct r600_resource
*last_trace_buf
;
350 struct r600_resource
*trace_buf
;
352 uint64_t dmesg_timestamp
;
353 unsigned apitrace_call_number
;
356 bool need_check_render_feedback
;
360 void cik_init_sdma_functions(struct si_context
*sctx
);
363 void si_init_blit_functions(struct si_context
*sctx
);
364 void si_decompress_graphics_textures(struct si_context
*sctx
);
365 void si_decompress_compute_textures(struct si_context
*sctx
);
366 void si_resource_copy_region(struct pipe_context
*ctx
,
367 struct pipe_resource
*dst
,
369 unsigned dstx
, unsigned dsty
, unsigned dstz
,
370 struct pipe_resource
*src
,
372 const struct pipe_box
*src_box
);
375 #define SI_CPDMA_SKIP_CHECK_CS_SPACE (1 << 0) /* don't call need_cs_space */
376 #define SI_CPDMA_SKIP_SYNC_AFTER (1 << 1) /* don't wait for DMA after the copy */
377 #define SI_CPDMA_SKIP_SYNC_BEFORE (1 << 2) /* don't wait for DMA before the copy (RAW hazards) */
378 #define SI_CPDMA_SKIP_GFX_SYNC (1 << 3) /* don't flush caches and don't wait for PS/CS */
380 void si_copy_buffer(struct si_context
*sctx
,
381 struct pipe_resource
*dst
, struct pipe_resource
*src
,
382 uint64_t dst_offset
, uint64_t src_offset
, unsigned size
,
383 unsigned user_flags
);
384 void si_init_cp_dma_functions(struct si_context
*sctx
);
387 void si_init_debug_functions(struct si_context
*sctx
);
388 void si_check_vm_faults(struct r600_common_context
*ctx
,
389 struct radeon_saved_cs
*saved
, enum ring_type ring
);
390 bool si_replace_shader(unsigned num
, struct radeon_shader_binary
*binary
);
393 void si_init_dma_functions(struct si_context
*sctx
);
395 /* si_hw_context.c */
396 void si_context_gfx_flush(void *context
, unsigned flags
,
397 struct pipe_fence_handle
**fence
);
398 void si_begin_new_cs(struct si_context
*ctx
);
399 void si_need_cs_space(struct si_context
*ctx
);
402 void si_init_compute_functions(struct si_context
*sctx
);
404 /* si_perfcounters.c */
405 void si_init_perfcounters(struct si_screen
*screen
);
408 struct pipe_video_codec
*si_uvd_create_decoder(struct pipe_context
*context
,
409 const struct pipe_video_codec
*templ
);
411 struct pipe_video_buffer
*si_video_buffer_create(struct pipe_context
*pipe
,
412 const struct pipe_video_buffer
*tmpl
);
419 si_invalidate_draw_sh_constants(struct si_context
*sctx
)
421 sctx
->last_base_vertex
= SI_BASE_VERTEX_UNKNOWN
;
425 si_set_atom_dirty(struct si_context
*sctx
,
426 struct r600_atom
*atom
, bool dirty
)
428 unsigned bit
= 1 << (atom
->id
- 1);
431 sctx
->dirty_atoms
|= bit
;
433 sctx
->dirty_atoms
&= ~bit
;
437 si_is_atom_dirty(struct si_context
*sctx
,
438 struct r600_atom
*atom
)
440 unsigned bit
= 1 << (atom
->id
- 1);
442 return sctx
->dirty_atoms
& bit
;
446 si_mark_atom_dirty(struct si_context
*sctx
,
447 struct r600_atom
*atom
)
449 si_set_atom_dirty(sctx
, atom
, true);
452 static inline struct tgsi_shader_info
*si_get_vs_info(struct si_context
*sctx
)
454 if (sctx
->gs_shader
.cso
)
455 return &sctx
->gs_shader
.cso
->info
;
456 else if (sctx
->tes_shader
.cso
)
457 return &sctx
->tes_shader
.cso
->info
;
458 else if (sctx
->vs_shader
.cso
)
459 return &sctx
->vs_shader
.cso
->info
;
464 static inline struct si_shader
* si_get_vs_state(struct si_context
*sctx
)
466 if (sctx
->gs_shader
.current
)
467 return sctx
->gs_shader
.cso
->gs_copy_shader
;
468 else if (sctx
->tes_shader
.current
)
469 return sctx
->tes_shader
.current
;
471 return sctx
->vs_shader
.current
;
474 static inline bool si_vs_exports_prim_id(struct si_shader
*shader
)
476 if (shader
->selector
->type
== PIPE_SHADER_VERTEX
)
477 return shader
->key
.part
.vs
.epilog
.export_prim_id
;
478 else if (shader
->selector
->type
== PIPE_SHADER_TESS_EVAL
)
479 return shader
->key
.part
.tes
.epilog
.export_prim_id
;