• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright 2010 Jerome Glisse <glisse@freedesktop.org>
3  * Copyright 2018 Advanced Micro Devices, Inc.
4  * All Rights Reserved.
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a
7  * copy of this software and associated documentation files (the "Software"),
8  * to deal in the Software without restriction, including without limitation
9  * on the rights to use, copy, modify, merge, publish, distribute, sub
10  * license, and/or sell copies of the Software, and to permit persons to whom
11  * the Software is furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice (including the next
14  * paragraph) shall be included in all copies or substantial portions of the
15  * Software.
16  *
17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
20  * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
21  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
22  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
23  * USE OR OTHER DEALINGS IN THE SOFTWARE.
24  */
25 #ifndef SI_PIPE_H
26 #define SI_PIPE_H
27 
28 #include "si_shader.h"
29 #include "si_state.h"
30 #include "util/u_dynarray.h"
31 #include "util/u_idalloc.h"
32 #include "util/u_suballoc.h"
33 #include "util/u_threaded_context.h"
34 #include "util/u_vertex_state_cache.h"
35 #include "ac_sqtt.h"
36 #include "ac_spm.h"
37 
38 #ifdef __cplusplus
39 extern "C" {
40 #endif
41 
42 #if UTIL_ARCH_BIG_ENDIAN
43 #define SI_BIG_ENDIAN 1
44 #else
45 #define SI_BIG_ENDIAN 0
46 #endif
47 
48 #define ATI_VENDOR_ID         0x1002
49 #define SI_NOT_QUERY          0xffffffff
50 
51 /* The base vertex and primitive restart can be any number, but we must pick
52  * one which will mean "unknown" for the purpose of state tracking and
53  * the number shouldn't be a commonly-used one. */
54 #define SI_BASE_VERTEX_UNKNOWN    INT_MIN
55 #define SI_START_INSTANCE_UNKNOWN ((unsigned)INT_MIN)
56 #define SI_DRAW_ID_UNKNOWN        ((unsigned)INT_MIN)
57 #define SI_RESTART_INDEX_UNKNOWN  ((unsigned)INT_MIN)
58 #define SI_INSTANCE_COUNT_UNKNOWN ((unsigned)INT_MIN)
59 #define SI_NUM_SMOOTH_AA_SAMPLES  4
60 #define SI_MAX_POINT_SIZE         2048
61 #define SI_GS_PER_ES              128
62 /* Alignment for optimal CP DMA performance. */
63 #define SI_CPDMA_ALIGNMENT 32
64 
65 /* Tunables for compute-based clear_buffer and copy_buffer: */
66 #define SI_COMPUTE_CLEAR_DW_PER_THREAD 4
67 #define SI_COMPUTE_COPY_DW_PER_THREAD  4
68 /* L2 LRU is recommended because the compute shader can finish sooner due to fewer L2 evictions. */
69 #define SI_COMPUTE_DST_CACHE_POLICY    L2_LRU
70 
71 /* Pipeline & streamout query controls. */
72 #define SI_CONTEXT_START_PIPELINE_STATS  (1 << 0)
73 #define SI_CONTEXT_STOP_PIPELINE_STATS   (1 << 1)
74 #define SI_CONTEXT_FLUSH_FOR_RENDER_COND (1 << 2)
75 /* Instruction cache. */
76 #define SI_CONTEXT_INV_ICACHE (1 << 3)
77 /* Scalar cache. (GFX6-9: scalar L1; GFX10: scalar L0)
78  * GFX10: This also invalidates the L1 shader array cache. */
79 #define SI_CONTEXT_INV_SCACHE (1 << 4)
80 /* Vector cache. (GFX6-9: vector L1; GFX10: vector L0)
81  * GFX10: This also invalidates the L1 shader array cache. */
82 #define SI_CONTEXT_INV_VCACHE (1 << 5)
83 /* L2 cache + L2 metadata cache writeback & invalidate.
84  * GFX6-8: Used by shaders only. GFX9-10: Used by everything. */
85 #define SI_CONTEXT_INV_L2 (1 << 6)
86 /* L2 writeback (write dirty L2 lines to memory for non-L2 clients).
87  * Only used for coherency with non-L2 clients like CB, DB, CP on GFX6-8.
88  * GFX6-7 will do complete invalidation, because the writeback is unsupported. */
89 #define SI_CONTEXT_WB_L2 (1 << 7)
90 /* Writeback & invalidate the L2 metadata cache only. It can only be coupled with
91  * a CB or DB flush. */
92 #define SI_CONTEXT_INV_L2_METADATA (1 << 8)
93 /* Framebuffer caches. */
94 #define SI_CONTEXT_FLUSH_AND_INV_DB      (1 << 9)
95 #define SI_CONTEXT_FLUSH_AND_INV_DB_META (1 << 10)
96 #define SI_CONTEXT_FLUSH_AND_INV_CB      (1 << 11)
97 /* Engine synchronization. */
98 #define SI_CONTEXT_VS_PARTIAL_FLUSH   (1 << 12)
99 #define SI_CONTEXT_PS_PARTIAL_FLUSH   (1 << 13)
100 #define SI_CONTEXT_CS_PARTIAL_FLUSH   (1 << 14)
101 #define SI_CONTEXT_VGT_FLUSH          (1 << 15)
102 #define SI_CONTEXT_VGT_STREAMOUT_SYNC (1 << 16)
103 /* PFP waits for ME to finish. Used to sync for index and indirect buffers and render
104  * condition. It's typically set when doing a VS/PS/CS partial flush for buffers. */
105 #define SI_CONTEXT_PFP_SYNC_ME        (1 << 17)
106 
107 #define SI_PREFETCH_LS              (1 << 1)
108 #define SI_PREFETCH_HS              (1 << 2)
109 #define SI_PREFETCH_ES              (1 << 3)
110 #define SI_PREFETCH_GS              (1 << 4)
111 #define SI_PREFETCH_VS              (1 << 5)
112 #define SI_PREFETCH_PS              (1 << 6)
113 
114 #define SI_MAX_BORDER_COLORS              4096
115 #define SI_MAX_VIEWPORTS                  16
116 #define SI_MAP_BUFFER_ALIGNMENT           64
117 /* We only support the minimum allowed value (512), so that we can pack a 3D block size
118  * in 1 SGPR. */
119 #define SI_MAX_VARIABLE_THREADS_PER_BLOCK 512
120 
121 #define SI_CONTEXT_FLAG_AUX               (1u << 31)
122 
123 #define SI_RESOURCE_FLAG_FORCE_LINEAR      (PIPE_RESOURCE_FLAG_DRV_PRIV << 0)
124 #define SI_RESOURCE_FLAG_FLUSHED_DEPTH     (PIPE_RESOURCE_FLAG_DRV_PRIV << 1)
125 #define SI_RESOURCE_FLAG_FORCE_MSAA_TILING (PIPE_RESOURCE_FLAG_DRV_PRIV << 2)
126 #define SI_RESOURCE_FLAG_DISABLE_DCC       (PIPE_RESOURCE_FLAG_DRV_PRIV << 3)
127 #define SI_RESOURCE_FLAG_DRIVER_INTERNAL   (PIPE_RESOURCE_FLAG_DRV_PRIV << 4)
128 #define SI_RESOURCE_FLAG_READ_ONLY         (PIPE_RESOURCE_FLAG_DRV_PRIV << 5)
129 #define SI_RESOURCE_FLAG_32BIT             (PIPE_RESOURCE_FLAG_DRV_PRIV << 6)
130 #define SI_RESOURCE_FLAG_CLEAR             (PIPE_RESOURCE_FLAG_DRV_PRIV << 7)
131 #define SI_RESOURCE_AUX_PLANE              (PIPE_RESOURCE_FLAG_DRV_PRIV << 8)
132 /* Set a micro tile mode: */
133 #define SI_RESOURCE_FLAG_FORCE_MICRO_TILE_MODE (PIPE_RESOURCE_FLAG_DRV_PRIV << 9)
134 #define SI_RESOURCE_FLAG_MICRO_TILE_MODE_SHIFT (util_logbase2(PIPE_RESOURCE_FLAG_DRV_PRIV) + 10)
135 #define SI_RESOURCE_FLAG_MICRO_TILE_MODE_SET(x)                                                    \
136    (((x)&0x3) << SI_RESOURCE_FLAG_MICRO_TILE_MODE_SHIFT)
137 #define SI_RESOURCE_FLAG_MICRO_TILE_MODE_GET(x)                                                    \
138    (((x) >> SI_RESOURCE_FLAG_MICRO_TILE_MODE_SHIFT) & 0x3)
139 #define SI_RESOURCE_FLAG_GL2_BYPASS        (PIPE_RESOURCE_FLAG_DRV_PRIV << 12)
140 /* Discard instead of evict. */
141 #define SI_RESOURCE_FLAG_DISCARDABLE       (PIPE_RESOURCE_FLAG_DRV_PRIV << 13)
142 
143 enum si_has_gs {
144    GS_OFF,
145    GS_ON,
146 };
147 
148 enum si_has_tess {
149    TESS_OFF,
150    TESS_ON,
151 };
152 
153 enum si_has_ngg {
154    NGG_OFF,
155    NGG_ON,
156 };
157 
158 #define DCC_CODE(x) (((x) << 24) | ((x) << 16) | ((x) << 8) | (x))
159 
160 enum si_clear_code
161 {
162    /* Common clear codes. */
163    DCC_CLEAR_0000    = DCC_CODE(0x00), /* all bits are 0 */
164    DCC_UNCOMPRESSED  = DCC_CODE(0xFF),
165 
166    GFX8_DCC_CLEAR_0000     = DCC_CLEAR_0000,
167    GFX8_DCC_CLEAR_0001     = DCC_CODE(0x40),
168    GFX8_DCC_CLEAR_1110     = DCC_CODE(0x80),
169    GFX8_DCC_CLEAR_1111     = DCC_CODE(0xC0),
170    GFX8_DCC_CLEAR_REG      = DCC_CODE(0x20),
171    GFX9_DCC_CLEAR_SINGLE   = DCC_CODE(0x10),
172 
173    GFX11_DCC_CLEAR_SINGLE     = DCC_CODE(0x01),
174    GFX11_DCC_CLEAR_0000       = DCC_CLEAR_0000, /* all bits are 0 */
175    GFX11_DCC_CLEAR_1111_UNORM = DCC_CODE(0x02), /* all bits are 1 */
176    GFX11_DCC_CLEAR_1111_FP16  = DCC_CODE(0x04), /* all 16-bit words are 0x3c00, max 64bpp */
177    GFX11_DCC_CLEAR_1111_FP32  = DCC_CODE(0x06), /* all 32-bit words are 0x3f800000 */
178    /* Color bits are 0, alpha bits are 1; only 88, 8888, 16161616 with alpha_on_msb=1 */
179    GFX11_DCC_CLEAR_0001_UNORM = DCC_CODE(0x08),
180    /* Color bits are 1, alpha bits are 0, only 88, 8888, 16161616 with alpha_on_msb=1 */
181    GFX11_DCC_CLEAR_1110_UNORM = DCC_CODE(0x0A),
182 };
183 
184 #define SI_IMAGE_ACCESS_DCC_OFF              (1 << 8)
185 #define SI_IMAGE_ACCESS_ALLOW_DCC_STORE      (1 << 9)
186 #define SI_IMAGE_ACCESS_BLOCK_FORMAT_AS_UINT (1 << 10) /* for compressed/subsampled images */
187 
188 /* Debug flags. */
189 enum
190 {
191    /* Shader logging options: */
192    DBG_VS = MESA_SHADER_VERTEX,
193    DBG_TCS = MESA_SHADER_TESS_CTRL,
194    DBG_TES = MESA_SHADER_TESS_EVAL,
195    DBG_GS = MESA_SHADER_GEOMETRY,
196    DBG_PS = MESA_SHADER_FRAGMENT,
197    DBG_CS = MESA_SHADER_COMPUTE,
198    DBG_NO_IR,
199    DBG_NO_NIR,
200    DBG_NO_ASM,
201    DBG_PREOPT_IR,
202 
203    /* Shader compiler options the shader cache should be aware of: */
204    DBG_FS_CORRECT_DERIVS_AFTER_KILL,
205    DBG_W32_GE,
206    DBG_W32_PS,
207    DBG_W32_PS_DISCARD,
208    DBG_W32_CS,
209    DBG_W64_GE,
210    DBG_W64_PS,
211    DBG_W64_CS,
212 
213    /* Shader compiler options (with no effect on the shader cache): */
214    DBG_CHECK_IR,
215    DBG_MONOLITHIC_SHADERS,
216    DBG_NO_OPT_VARIANT,
217 
218    /* Information logging options: */
219    DBG_INFO,
220    DBG_TEX,
221    DBG_COMPUTE,
222    DBG_VM,
223    DBG_CACHE_STATS,
224    DBG_IB,
225 
226    /* Driver options: */
227    DBG_NO_WC,
228    DBG_CHECK_VM,
229    DBG_RESERVE_VMID,
230    DBG_SHADOW_REGS,
231    DBG_NO_FAST_DISPLAY_LIST,
232 
233    /* Multimedia options: */
234    DBG_NO_EFC,
235 
236    /* 3D engine options: */
237    DBG_NO_GFX,
238    DBG_NO_NGG,
239    DBG_ALWAYS_NGG_CULLING_ALL,
240    DBG_NO_NGG_CULLING,
241    DBG_SWITCH_ON_EOP,
242    DBG_NO_OUT_OF_ORDER,
243    DBG_NO_DPBB,
244    DBG_DPBB,
245    DBG_NO_HYPERZ,
246    DBG_NO_2D_TILING,
247    DBG_NO_TILING,
248    DBG_NO_DISPLAY_TILING,
249    DBG_NO_DISPLAY_DCC,
250    DBG_NO_EXPORTED_DCC,
251    DBG_NO_DCC,
252    DBG_NO_DCC_CLEAR,
253    DBG_NO_DCC_STORE,
254    DBG_DCC_STORE,
255    DBG_NO_DCC_MSAA,
256    DBG_NO_FMASK,
257    DBG_NO_DMA,
258 
259    DBG_TMZ,
260    DBG_SQTT,
261 
262    DBG_COUNT
263 };
264 
265 enum
266 {
267    /* Tests: */
268    DBG_TEST_IMAGE_COPY,
269    DBG_TEST_CB_RESOLVE,
270    DBG_TEST_COMPUTE_BLIT,
271    DBG_TEST_VMFAULT_CP,
272    DBG_TEST_VMFAULT_SHADER,
273    DBG_TEST_DMA_PERF,
274    DBG_TEST_GDS,
275    DBG_TEST_GDS_MM,
276    DBG_TEST_GDS_OA_MM,
277 };
278 
279 #define DBG_ALL_SHADERS (((1 << (DBG_CS + 1)) - 1))
280 #define DBG(name)       (1ull << DBG_##name)
281 
282 enum si_cache_policy
283 {
284    L2_BYPASS,
285    L2_STREAM, /* same as SLC=1 */
286    L2_LRU,    /* same as SLC=0 */
287 };
288 
289 enum si_coherency
290 {
291    SI_COHERENCY_NONE, /* no cache flushes needed */
292    SI_COHERENCY_SHADER,
293    SI_COHERENCY_CB_META,
294    SI_COHERENCY_DB_META,
295    SI_COHERENCY_CP,
296 };
297 
298 #define SI_BIND_CONSTANT_BUFFER_SHIFT     0
299 #define SI_BIND_SHADER_BUFFER_SHIFT       6
300 #define SI_BIND_IMAGE_BUFFER_SHIFT        12
301 #define SI_BIND_SAMPLER_BUFFER_SHIFT      18
302 #define SI_BIND_OTHER_BUFFER_SHIFT        24
303 
304 /* Bind masks for all 6 shader stages. */
305 #define SI_BIND_CONSTANT_BUFFER_ALL       (0x3f << SI_BIND_CONSTANT_BUFFER_SHIFT)
306 #define SI_BIND_SHADER_BUFFER_ALL         (0x3f << SI_BIND_SHADER_BUFFER_SHIFT)
307 #define SI_BIND_IMAGE_BUFFER_ALL          (0x3f << SI_BIND_IMAGE_BUFFER_SHIFT)
308 #define SI_BIND_SAMPLER_BUFFER_ALL        (0x3f << SI_BIND_SAMPLER_BUFFER_SHIFT)
309 
310 #define SI_BIND_CONSTANT_BUFFER(shader)   ((1 << (shader)) << SI_BIND_CONSTANT_BUFFER_SHIFT)
311 #define SI_BIND_SHADER_BUFFER(shader)     ((1 << (shader)) << SI_BIND_SHADER_BUFFER_SHIFT)
312 #define SI_BIND_IMAGE_BUFFER(shader)      ((1 << (shader)) << SI_BIND_IMAGE_BUFFER_SHIFT)
313 #define SI_BIND_SAMPLER_BUFFER(shader)    ((1 << (shader)) << SI_BIND_SAMPLER_BUFFER_SHIFT)
314 #define SI_BIND_VERTEX_BUFFER             (1 << (SI_BIND_OTHER_BUFFER_SHIFT + 0))
315 #define SI_BIND_STREAMOUT_BUFFER          (1 << (SI_BIND_OTHER_BUFFER_SHIFT + 1))
316 
317 struct si_compute;
318 struct si_shader_context;
319 struct hash_table;
320 
321 /* Only 32-bit buffer allocations are supported, gallium doesn't support more
322  * at the moment.
323  */
324 struct si_resource {
325    struct threaded_resource b;
326 
327    /* Winsys objects. */
328    struct pb_buffer *buf;
329    uint64_t gpu_address;
330    /* Memory usage if the buffer placement is optimal. */
331    uint32_t memory_usage_kb;
332 
333    /* Resource properties. */
334    uint64_t bo_size;
335    uint8_t bo_alignment_log2;
336    enum radeon_bo_domain domains:8;
337    enum radeon_bo_flag flags:16;
338    unsigned bind_history; /* bitmask of SI_BIND_xxx_BUFFER */
339 
340    /* The buffer range which is initialized (with a write transfer,
341     * streamout, DMA, or as a random access target). The rest of
342     * the buffer is considered invalid and can be mapped unsynchronized.
343     *
344     * This allows unsynchronized mapping of a buffer range which hasn't
345     * been used yet. It's for applications which forget to use
346     * the unsynchronized map flag and expect the driver to figure it out.
347     */
348    struct util_range valid_buffer_range;
349 
350    /* For buffers only. This indicates that a write operation has been
351     * performed by TC L2, but the cache hasn't been flushed.
352     * Any hw block which doesn't use or bypasses TC L2 should check this
353     * flag and flush the cache before using the buffer.
354     *
355     * For example, TC L2 must be flushed if a buffer which has been
356     * modified by a shader store instruction is about to be used as
357     * an index buffer. The reason is that VGT DMA index fetching doesn't
358     * use TC L2.
359     */
360    bool TC_L2_dirty;
361 
362    /* Whether this resource is referenced by bindless handles. */
363    bool texture_handle_allocated;
364    bool image_handle_allocated;
365 
366    /* Whether the resource has been exported via resource_get_handle. */
367    uint8_t external_usage; /* PIPE_HANDLE_USAGE_* */
368 };
369 
370 struct si_transfer {
371    struct threaded_transfer b;
372    struct si_resource *staging;
373 };
374 
375 struct si_texture {
376    struct si_resource buffer;
377 
378    struct radeon_surf surface;
379    struct si_texture *flushed_depth_texture;
380 
381    /* One texture allocation can contain these buffers:
382     * - image (pixel data)
383     * - FMASK buffer (MSAA compression)
384     * - CMASK buffer (MSAA compression and/or legacy fast color clear)
385     * - HTILE buffer (Z/S compression and fast Z/S clear)
386     * - DCC buffer (color compression and new fast color clear)
387     * - displayable DCC buffer (if the DCC buffer is not displayable)
388     */
389    uint64_t cmask_base_address_reg;
390    struct si_resource *cmask_buffer;
391    unsigned cb_color_info; /* fast clear enable bit */
392    unsigned color_clear_value[2]; /* not on gfx11 */
393    unsigned last_msaa_resolve_target_micro_mode;
394    bool swap_rgb_to_bgr_on_next_clear;
395    bool swap_rgb_to_bgr;
396    unsigned num_level0_transfers;
397    unsigned plane_index; /* other planes are different pipe_resources */
398    unsigned num_planes;
399 
400    /* Depth buffer compression and fast clear. */
401    float depth_clear_value[RADEON_SURF_MAX_LEVELS];
402    uint8_t stencil_clear_value[RADEON_SURF_MAX_LEVELS];
403    uint16_t depth_cleared_level_mask_once; /* if it was cleared at least once */
404    uint16_t depth_cleared_level_mask;     /* track if it's cleared (can be false negative) */
405    uint16_t stencil_cleared_level_mask_once; /* if it was cleared at least once */
406    uint16_t dirty_level_mask;         /* each bit says if that mipmap is compressed */
407    uint16_t stencil_dirty_level_mask; /* each bit says if that mipmap is compressed */
408    enum pipe_format db_render_format : 16;
409    bool fmask_is_identity : 1;
410    bool tc_compatible_htile : 1;
411    bool enable_tc_compatible_htile_next_clear : 1;
412    bool htile_stencil_disabled : 1;
413    bool upgraded_depth : 1;  /* upgraded from unorm to Z32_FLOAT */
414    bool is_depth : 1;
415    bool db_compatible : 1;
416    bool can_sample_z : 1;
417    bool can_sample_s : 1;
418    bool need_flush_after_depth_decompression: 1;
419 
420    /* We need to track DCC dirtiness, because st/dri usually calls
421     * flush_resource twice per frame (not a bug) and we don't wanna
422     * decompress DCC twice.
423     */
424    bool displayable_dcc_dirty : 1;
425 
426    /* Counter that should be non-zero if the texture is bound to a
427     * framebuffer.
428     */
429    unsigned framebuffers_bound;
430 };
431 
432 /* State trackers create separate textures in a next-chain for extra planes
433  * even if those are planes created purely for modifiers. Because the linking
434  * of the chain happens outside of the driver, and NULL is interpreted as
435  * failure, let's create some dummy texture structs. We could use these
436  * later to use the offsets for linking if we really wanted to.
437  *
438  * For now just create a dummy struct and completely ignore it.
439  *
440  * Potentially in the future we could store stride/offset and use it during
441  * creation, though we might want to change how linking is done first.
442  */
443 struct si_auxiliary_texture {
444    struct threaded_resource b;
445    struct pb_buffer *buffer;
446    uint32_t offset;
447    uint32_t stride;
448 };
449 
450 struct si_surface {
451    struct pipe_surface base;
452 
453    /* These can vary with block-compressed textures. */
454    uint16_t width0;
455    uint16_t height0;
456 
457    bool color_initialized : 1;
458    bool depth_initialized : 1;
459 
460    /* Misc. color flags. */
461    bool color_is_int8 : 1;
462    bool color_is_int10 : 1;
463    bool dcc_incompatible : 1;
464 
465    /* Color registers. */
466    unsigned cb_color_info;
467    unsigned cb_color_view;
468    unsigned cb_color_attrib;
469    unsigned cb_color_attrib2;                      /* GFX9 and later */
470    unsigned cb_color_attrib3;                      /* GFX10 and later */
471    unsigned cb_dcc_control;                        /* GFX8 and later */
472    unsigned spi_shader_col_format : 8;             /* no blending, no alpha-to-coverage. */
473    unsigned spi_shader_col_format_alpha : 8;       /* alpha-to-coverage */
474    unsigned spi_shader_col_format_blend : 8;       /* blending without alpha. */
475    unsigned spi_shader_col_format_blend_alpha : 8; /* blending with alpha. */
476 
477    /* DB registers. */
478    uint64_t db_depth_base; /* DB_Z_READ/WRITE_BASE */
479    uint64_t db_stencil_base;
480    uint64_t db_htile_data_base;
481    unsigned db_depth_info;
482    unsigned db_z_info;
483    unsigned db_z_info2; /* GFX9 only */
484    unsigned db_depth_view;
485    unsigned db_depth_size;
486    unsigned db_depth_slice;
487    unsigned db_stencil_info;
488    unsigned db_stencil_info2; /* GFX9 only */
489    unsigned db_htile_surface;
490 };
491 
492 struct si_mmio_counter {
493    unsigned busy;
494    unsigned idle;
495 };
496 
497 union si_mmio_counters {
498    struct si_mmio_counters_named {
499       /* For global GPU load including SDMA. */
500       struct si_mmio_counter gpu;
501 
502       /* GRBM_STATUS */
503       struct si_mmio_counter spi;
504       struct si_mmio_counter gui;
505       struct si_mmio_counter ta;
506       struct si_mmio_counter gds;
507       struct si_mmio_counter vgt;
508       struct si_mmio_counter ia;
509       struct si_mmio_counter sx;
510       struct si_mmio_counter wd;
511       struct si_mmio_counter bci;
512       struct si_mmio_counter sc;
513       struct si_mmio_counter pa;
514       struct si_mmio_counter db;
515       struct si_mmio_counter cp;
516       struct si_mmio_counter cb;
517 
518       /* SRBM_STATUS2 */
519       struct si_mmio_counter sdma;
520 
521       /* CP_STAT */
522       struct si_mmio_counter pfp;
523       struct si_mmio_counter meq;
524       struct si_mmio_counter me;
525       struct si_mmio_counter surf_sync;
526       struct si_mmio_counter cp_dma;
527       struct si_mmio_counter scratch_ram;
528    } named;
529 
530    unsigned array[sizeof(struct si_mmio_counters_named) / sizeof(unsigned)];
531 };
532 
533 struct si_memory_object {
534    struct pipe_memory_object b;
535    struct pb_buffer *buf;
536    uint32_t stride;
537 };
538 
539 /* Saved CS data for debugging features. */
540 struct radeon_saved_cs {
541    uint32_t *ib;
542    unsigned num_dw;
543 
544    struct radeon_bo_list_item *bo_list;
545    unsigned bo_count;
546 };
547 
548 struct si_screen {
549    struct pipe_screen b;
550    struct radeon_winsys *ws;
551    struct disk_cache *disk_shader_cache;
552 
553    struct radeon_info info;
554    struct nir_shader_compiler_options nir_options;
555    uint64_t debug_flags;
556    char renderer_string[183];
557 
558    void (*make_texture_descriptor)(struct si_screen *screen, struct si_texture *tex, bool sampler,
559                                    enum pipe_texture_target target, enum pipe_format pipe_format,
560                                    const unsigned char state_swizzle[4], unsigned first_level,
561                                    unsigned last_level, unsigned first_layer, unsigned last_layer,
562                                    unsigned width, unsigned height, unsigned depth, uint32_t *state,
563                                    uint32_t *fmask_state);
564 
565    unsigned max_memory_usage_kb;
566    unsigned pa_sc_raster_config;
567    unsigned pa_sc_raster_config_1;
568    unsigned se_tile_repeat;
569    unsigned gs_table_depth;
570    struct ac_hs_info hs;
571    unsigned eqaa_force_coverage_samples;
572    unsigned eqaa_force_z_samples;
573    unsigned eqaa_force_color_samples;
574    unsigned pbb_context_states_per_bin;
575    unsigned pbb_persistent_states_per_bin;
576    bool has_draw_indirect_multi;
577    bool has_out_of_order_rast;
578    bool dpbb_allowed;
579    bool use_ngg;
580    bool use_ngg_culling;
581    bool use_ngg_streamout;
582    bool allow_dcc_msaa_clear_to_reg_for_bpp[5]; /* indexed by log2(Bpp) */
583    bool always_allow_dcc_stores;
584 
585    struct {
586 #define OPT_BOOL(name, dflt, description) bool name : 1;
587 #define OPT_INT(name, dflt, description) int name;
588 #include "si_debug_options.h"
589    } options;
590 
591    /* Whether shaders are monolithic (1-part) or separate (3-part). */
592    bool use_monolithic_shaders;
593    bool record_llvm_ir;
594 
595    struct slab_parent_pool pool_transfers;
596 
597    /* Texture filter settings. */
598    int force_aniso; /* -1 = disabled */
599 
600    unsigned max_texel_buffer_elements;
601 
602    /* Auxiliary context. Mainly used to initialize resources. */
603    void *aux_context;
604    mtx_t aux_context_lock;
605 
606    /* Async compute context for DRI_PRIME copies. */
607    struct pipe_context *async_compute_context;
608    simple_mtx_t async_compute_context_lock;
609 
610    /* This must be in the screen, because UE4 uses one context for
611     * compilation and another one for rendering.
612     */
613    unsigned num_compilations;
614    /* Along with ST_DEBUG=precompile, this should show if applications
615     * are loading shaders on demand. This is a monotonic counter.
616     */
617    unsigned num_shaders_created;
618    unsigned num_memory_shader_cache_hits;
619    unsigned num_memory_shader_cache_misses;
620    unsigned num_disk_shader_cache_hits;
621    unsigned num_disk_shader_cache_misses;
622 
623    /* GPU load thread. */
624    simple_mtx_t gpu_load_mutex;
625    thrd_t gpu_load_thread;
626    bool gpu_load_thread_created;
627    union si_mmio_counters mmio_counters;
628    volatile unsigned gpu_load_stop_thread; /* bool */
629 
630    /* Performance counters. */
631    struct si_perfcounters *perfcounters;
632 
633    /* If pipe_screen wants to recompute and re-emit the framebuffer,
634     * sampler, and image states of all contexts, it should atomically
635     * increment this.
636     *
637     * Each context will compare this with its own last known value of
638     * the counter before drawing and re-emit the states accordingly.
639     */
640    unsigned dirty_tex_counter;
641    unsigned dirty_buf_counter;
642 
643    /* Atomically increment this counter when an existing texture's
644     * metadata is enabled or disabled in a way that requires changing
645     * contexts' compressed texture binding masks.
646     */
647    unsigned compressed_colortex_counter;
648 
649    struct {
650       /* Context flags to set so that all writes from earlier jobs
651        * in the CP are seen by L2 clients.
652        */
653       unsigned cp_to_L2;
654 
655       /* Context flags to set so that all writes from earlier jobs
656        * that end in L2 are seen by CP.
657        */
658       unsigned L2_to_cp;
659    } barrier_flags;
660 
661    simple_mtx_t shader_parts_mutex;
662    struct si_shader_part *vs_prologs;
663    struct si_shader_part *tcs_epilogs;
664    struct si_shader_part *ps_prologs;
665    struct si_shader_part *ps_epilogs;
666 
667    /* Shader cache in memory.
668     *
669     * Design & limitations:
670     * - The shader cache is per screen (= per process), never saved to
671     *   disk, and skips redundant shader compilations from NIR to bytecode.
672     * - It can only be used with one-variant-per-shader support, in which
673     *   case only the main (typically middle) part of shaders is cached.
674     * - Only VS, TCS, TES, PS are cached, out of which only the hw VS
675     *   variants of VS and TES are cached, so LS and ES aren't.
676     * - GS and CS aren't cached, but it's certainly possible to cache
677     *   those as well.
678     */
679    simple_mtx_t shader_cache_mutex;
680    struct hash_table *shader_cache;
681    /* Maximum and current size */
682    uint32_t shader_cache_size;
683    uint32_t shader_cache_max_size;
684 
685    /* Shader cache of live shaders. */
686    struct util_live_shader_cache live_shader_cache;
687 
688    /* Shader compiler queue for multithreaded compilation. */
689    struct util_queue shader_compiler_queue;
690    /* Use at most 3 normal compiler threads on quadcore and better.
691     * Hyperthreaded CPUs report the number of threads, but we want
692     * the number of cores. We only need this many threads for shader-db. */
693    struct ac_llvm_compiler compiler[24]; /* used by the queue only */
694 
695    struct util_queue shader_compiler_queue_low_priority;
696    /* Use at most 2 low priority threads on quadcore and better.
697     * We want to minimize the impact on multithreaded Mesa. */
698    struct ac_llvm_compiler compiler_lowp[10];
699 
700    unsigned ngg_subgroup_size;
701 
702    struct util_idalloc_mt buffer_ids;
703    struct util_vertex_state_cache vertex_state_cache;
704 
705    struct si_resource *attribute_ring;
706 
707    /* NGG streamout. */
708    simple_mtx_t gds_mutex;
709    struct pb_buffer *gds;
710    struct pb_buffer *gds_oa;
711 };
712 
713 struct si_sampler_view {
714    struct pipe_sampler_view base;
715    /* [0..7] = image descriptor
716     * [4..7] = buffer descriptor */
717    uint32_t state[8];
718    uint32_t fmask_state[8];
719    const struct legacy_surf_level *base_level_info;
720    ubyte block_width;
721    bool is_stencil_sampler;
722    bool dcc_incompatible;
723 };
724 
725 #define SI_SAMPLER_STATE_MAGIC 0x34f1c35a
726 
727 struct si_sampler_state {
728 #ifndef NDEBUG
729    unsigned magic;
730 #endif
731    uint32_t val[4];
732    uint32_t upgraded_depth_val[4];
733 };
734 
735 struct si_cs_shader_state {
736    struct si_compute *program;
737    struct si_compute *emitted_program;
738    unsigned offset;
739    bool initialized;
740 };
741 
742 struct si_samplers {
743    struct pipe_sampler_view *views[SI_NUM_SAMPLERS];
744    struct si_sampler_state *sampler_states[SI_NUM_SAMPLERS];
745 
746    /* The i-th bit is set if that element is enabled (non-NULL resource). */
747    unsigned enabled_mask;
748    uint32_t has_depth_tex_mask;
749    uint32_t needs_depth_decompress_mask;
750    uint32_t needs_color_decompress_mask;
751 };
752 
753 struct si_images {
754    struct pipe_image_view views[SI_NUM_IMAGES];
755    uint32_t needs_color_decompress_mask;
756    unsigned enabled_mask;
757    unsigned display_dcc_store_mask;
758 };
759 
760 struct si_framebuffer {
761    struct pipe_framebuffer_state state;
762    unsigned colorbuf_enabled_4bit;
763    unsigned spi_shader_col_format;
764    unsigned spi_shader_col_format_alpha;
765    unsigned spi_shader_col_format_blend;
766    unsigned spi_shader_col_format_blend_alpha;
767    ubyte nr_samples : 5;   /* at most 16xAA */
768    ubyte log_samples : 3;  /* at most 4 = 16xAA */
769    ubyte nr_color_samples; /* at most 8xAA */
770    ubyte compressed_cb_mask;
771    ubyte uncompressed_cb_mask;
772    ubyte color_is_int8;
773    ubyte color_is_int10;
774    ubyte dirty_cbufs;
775    ubyte min_bytes_per_pixel;
776    bool dirty_zsbuf;
777    bool any_dst_linear;
778    bool CB_has_shader_readable_metadata;
779    bool DB_has_shader_readable_metadata;
780    bool all_DCC_pipe_aligned;
781    bool has_dcc_msaa;
782 };
783 
784 enum si_quant_mode
785 {
786    /* This is the list we want to support. */
787    SI_QUANT_MODE_16_8_FIXED_POINT_1_256TH,
788    SI_QUANT_MODE_14_10_FIXED_POINT_1_1024TH,
789    SI_QUANT_MODE_12_12_FIXED_POINT_1_4096TH,
790 };
791 
792 struct si_signed_scissor {
793    int minx;
794    int miny;
795    int maxx;
796    int maxy;
797    enum si_quant_mode quant_mode;
798 };
799 
800 struct si_viewports {
801    struct pipe_viewport_state states[SI_MAX_VIEWPORTS];
802    struct si_signed_scissor as_scissor[SI_MAX_VIEWPORTS];
803 };
804 
805 struct si_streamout_target {
806    struct pipe_stream_output_target b;
807 
808    /* The buffer where BUFFER_FILLED_SIZE is stored. */
809    struct si_resource *buf_filled_size;
810    unsigned buf_filled_size_offset;
811    bool buf_filled_size_valid;
812 
813    unsigned stride_in_dw;
814 };
815 
816 struct si_streamout {
817    bool begin_emitted;
818 
819    unsigned enabled_mask;
820    unsigned num_targets;
821    struct si_streamout_target *targets[PIPE_MAX_SO_BUFFERS];
822 
823    unsigned append_bitmask;
824    bool suspended;
825 
826    /* External state which comes from the vertex shader,
827     * it must be set explicitly when binding a shader. */
828    uint8_t *stride_in_dw;
829    unsigned enabled_stream_buffers_mask; /* stream0 buffers0-3 in 4 LSB */
830 
831    /* The state of VGT_STRMOUT_BUFFER_(CONFIG|EN). */
832    unsigned hw_enabled_mask;
833 
834    /* The state of VGT_STRMOUT_(CONFIG|EN). */
835    bool streamout_enabled;
836    bool prims_gen_query_enabled;
837    int num_prims_gen_queries;
838 };
839 
840 /* A shader state consists of the shader selector, which is a constant state
841  * object shared by multiple contexts and shouldn't be modified, and
842  * the current shader variant selected for this context.
843  */
844 struct si_shader_ctx_state {
845    struct si_shader_selector *cso;
846    struct si_shader *current;
847    /* The shader variant key representing the current state. */
848    union si_shader_key key;
849 };
850 
851 #define SI_NUM_VGT_PARAM_KEY_BITS 12
852 #define SI_NUM_VGT_PARAM_STATES   (1 << SI_NUM_VGT_PARAM_KEY_BITS)
853 
854 /* The IA_MULTI_VGT_PARAM key used to index the table of precomputed values.
855  * Some fields are set by state-change calls, most are set by draw_vbo.
856  */
857 union si_vgt_param_key {
858    struct {
859 #if UTIL_ARCH_LITTLE_ENDIAN
860       uint16_t prim : 4;
861       uint16_t uses_instancing : 1;
862       uint16_t multi_instances_smaller_than_primgroup : 1;
863       uint16_t primitive_restart : 1;
864       uint16_t count_from_stream_output : 1;
865       uint16_t line_stipple_enabled : 1;
866       uint16_t uses_tess : 1;
867       uint16_t tess_uses_prim_id : 1;
868       uint16_t uses_gs : 1;
869       uint16_t _pad : 16 - SI_NUM_VGT_PARAM_KEY_BITS;
870 #else /* UTIL_ARCH_BIG_ENDIAN */
871       uint16_t _pad : 16 - SI_NUM_VGT_PARAM_KEY_BITS;
872       uint16_t uses_gs : 1;
873       uint16_t tess_uses_prim_id : 1;
874       uint16_t uses_tess : 1;
875       uint16_t line_stipple_enabled : 1;
876       uint16_t count_from_stream_output : 1;
877       uint16_t primitive_restart : 1;
878       uint16_t multi_instances_smaller_than_primgroup : 1;
879       uint16_t uses_instancing : 1;
880       uint16_t prim : 4;
881 #endif
882    } u;
883    uint16_t index;
884 };
885 
886 struct si_texture_handle {
887    unsigned desc_slot;
888    bool desc_dirty;
889    struct pipe_sampler_view *view;
890    struct si_sampler_state sstate;
891 };
892 
893 struct si_image_handle {
894    unsigned desc_slot;
895    bool desc_dirty;
896    struct pipe_image_view view;
897 };
898 
899 struct si_saved_cs {
900    struct pipe_reference reference;
901    struct si_context *ctx;
902    struct radeon_saved_cs gfx;
903    struct radeon_saved_cs compute;
904    struct si_resource *trace_buf;
905    unsigned trace_id;
906 
907    unsigned gfx_last_dw;
908    bool flushed;
909    int64_t time_flush;
910 };
911 
912 struct si_small_prim_cull_info {
913    float scale[2], translate[2];
914    float scale_no_aa[2], translate_no_aa[2];
915    float clip_half_line_width[2];      /* line_width * 0.5 in clip space in X and Y directions */
916 };
917 
918 struct si_vertex_state {
919    struct pipe_vertex_state b;
920    struct si_vertex_elements velems;
921    uint32_t descriptors[4 * SI_MAX_ATTRIBS];
922 };
923 
924 typedef void (*pipe_draw_vbo_func)(struct pipe_context *pipe,
925                                    const struct pipe_draw_info *info,
926                                    unsigned drawid_offset,
927                                    const struct pipe_draw_indirect_info *indirect,
928                                    const struct pipe_draw_start_count_bias *draws,
929                                    unsigned num_draws);
930 typedef void (*pipe_draw_vertex_state_func)(struct pipe_context *ctx,
931                                             struct pipe_vertex_state *vstate,
932                                             uint32_t partial_velem_mask,
933                                             struct pipe_draw_vertex_state_info info,
934                                             const struct pipe_draw_start_count_bias *draws,
935                                             unsigned num_draws);
936 
937 struct si_context {
938    struct pipe_context b; /* base class */
939 
940    enum radeon_family family;
941    enum amd_gfx_level gfx_level;
942 
943    struct radeon_winsys *ws;
944    struct radeon_winsys_ctx *ctx;
945    struct radeon_cmdbuf gfx_cs; /* compute IB if graphics is disabled */
946    struct radeon_cmdbuf *sdma_cs;
947    struct pipe_fence_handle *last_gfx_fence;
948    struct si_resource *eop_bug_scratch;
949    struct si_resource *eop_bug_scratch_tmz;
950    struct u_upload_mgr *cached_gtt_allocator;
951    struct threaded_context *tc;
952    struct u_suballocator allocator_zeroed_memory;
953    struct slab_child_pool pool_transfers;
954    struct slab_child_pool pool_transfers_unsync; /* for threaded_context */
955    struct pipe_device_reset_callback device_reset_callback;
956    struct u_log_context *log;
957    void *query_result_shader;
958    void *sh_query_result_shader;
959    struct si_resource *shadowed_regs;
960 
961    void (*emit_cache_flush)(struct si_context *ctx, struct radeon_cmdbuf *cs);
962 
963    struct blitter_context *blitter;
964    void *noop_blend;
965    void *noop_dsa;
966    void *no_velems_state;
967    void *discard_rasterizer_state;
968    void *custom_dsa_flush;
969    void *custom_blend_resolve;
970    void *custom_blend_fmask_decompress;
971    void *custom_blend_eliminate_fastclear;
972    void *custom_blend_dcc_decompress;
973    void *vs_blit_pos;
974    void *vs_blit_pos_layered;
975    void *vs_blit_color;
976    void *vs_blit_color_layered;
977    void *vs_blit_texcoord;
978    void *cs_clear_buffer;
979    void *cs_clear_buffer_rmw;
980    void *cs_copy_buffer;
981    void *cs_copy_image[2][2]; /* [src_is_1d][dst_is_1d] */
982    void *cs_clear_render_target;
983    void *cs_clear_render_target_1d_array;
984    void *cs_clear_12bytes_buffer;
985    void *cs_dcc_retile[32];
986    void *cs_fmask_expand[3][2]; /* [log2(samples)-1][is_array] */
987    struct si_screen *screen;
988    struct util_debug_callback debug;
989    struct ac_llvm_compiler compiler; /* only non-threaded compilation */
990    struct hash_table *fixed_func_tcs_shader_cache;
991    struct si_resource *wait_mem_scratch;
992    struct si_resource *wait_mem_scratch_tmz;
993    unsigned wait_mem_number;
994    uint16_t prefetch_L2_mask;
995 
996    bool blitter_running;
997    bool in_update_ps_colorbuf0_slot;
998    bool is_noop:1;
999    bool has_graphics:1;
1000    bool gfx_flush_in_progress : 1;
1001    bool gfx_last_ib_is_busy : 1;
1002    bool compute_is_busy : 1;
1003    int8_t pipeline_stats_enabled; /* -1 = unknown, 0 = disabled, 1 = enabled */
1004 
1005    unsigned num_gfx_cs_flushes;
1006    unsigned initial_gfx_cs_size;
1007    unsigned last_dirty_tex_counter;
1008    unsigned last_dirty_buf_counter;
1009    unsigned last_compressed_colortex_counter;
1010    unsigned last_num_draw_calls;
1011    unsigned flags; /* flush flags */
1012    /* Current unaccounted memory usage. */
1013    uint32_t memory_usage_kb;
1014 
1015    /* Atoms (direct states). */
1016    union si_state_atoms atoms;
1017    unsigned dirty_atoms; /* mask */
1018    /* PM4 states (precomputed immutable states) */
1019    unsigned dirty_states;
1020    union si_state queued;
1021    union si_state emitted;
1022 
1023    /* Atom declarations. */
1024    struct si_framebuffer framebuffer;
1025    unsigned sample_locs_num_samples;
1026    uint16_t sample_mask;
1027    unsigned last_cb_target_mask;
1028    struct pipe_blend_color blend_color;
1029    struct pipe_clip_state clip_state;
1030    struct si_shader_data shader_pointers;
1031    struct si_stencil_ref stencil_ref;
1032    bool blend_color_any_nonzeros:1;
1033    bool clip_state_any_nonzeros:1;
1034    bool viewport0_y_inverted;
1035    struct pipe_scissor_state scissors[SI_MAX_VIEWPORTS];
1036    struct si_streamout streamout;
1037    struct si_viewports viewports;
1038    unsigned num_window_rectangles;
1039    bool window_rectangles_include;
1040    struct pipe_scissor_state window_rectangles[4];
1041 
1042    /* Precomputed states. */
1043    struct si_pm4_state *last_preamble;
1044    struct si_pm4_state *cs_preamble_state;
1045    struct si_pm4_state *cs_preamble_state_tmz;
1046    uint16_t gs_ring_state_dw_offset;
1047    uint16_t gs_ring_state_dw_offset_tmz;
1048    bool cs_preamble_has_vgt_flush;
1049    bool cs_preamble_has_vgt_flush_tmz;
1050 
1051    struct si_pm4_state *vgt_shader_config[SI_NUM_VGT_STAGES_STATES];
1052 
1053    /* shaders */
1054    union {
1055       struct {
1056          struct si_shader_ctx_state vs;
1057          struct si_shader_ctx_state ps;
1058          struct si_shader_ctx_state gs;
1059          struct si_shader_ctx_state tcs;
1060          struct si_shader_ctx_state tes;
1061       } shader;
1062       /* indexed access using pipe_shader_type (not by MESA_SHADER_*) */
1063       struct si_shader_ctx_state shaders[SI_NUM_GRAPHICS_SHADERS];
1064    };
1065    struct si_cs_shader_state cs_shader_state;
1066    /* if current tcs set by user */
1067    bool is_user_tcs;
1068 
1069    /* shader information */
1070    uint64_t ps_inputs_read_or_disabled;
1071    struct si_vertex_elements *vertex_elements;
1072    unsigned num_vertex_elements;
1073    unsigned cs_max_waves_per_sh;
1074    bool uses_nontrivial_vs_prolog;
1075    bool force_trivial_vs_prolog;
1076    bool do_update_shaders;
1077    bool compute_shaderbuf_sgprs_dirty;
1078    bool compute_image_sgprs_dirty;
1079    bool vs_uses_base_instance;
1080    bool vs_uses_draw_id;
1081    uint8_t patch_vertices;
1082 
1083    /* shader descriptors */
1084    struct si_descriptors descriptors[SI_NUM_DESCS];
1085    unsigned descriptors_dirty;
1086    unsigned shader_pointers_dirty;
1087    unsigned shader_needs_decompress_mask;
1088    unsigned shader_has_depth_tex;
1089    struct si_buffer_resources internal_bindings;
1090    struct si_buffer_resources const_and_shader_buffers[SI_NUM_SHADERS];
1091    struct si_samplers samplers[SI_NUM_SHADERS];
1092    struct si_images images[SI_NUM_SHADERS];
1093    bool bo_list_add_all_resident_resources;
1094    bool bo_list_add_all_gfx_resources;
1095    bool bo_list_add_all_compute_resources;
1096 
1097    /* other shader resources */
1098    struct pipe_constant_buffer null_const_buf; /* used for set_constant_buffer(NULL) on GFX7 */
1099    struct pipe_resource *esgs_ring;
1100    struct pipe_resource *gsvs_ring;
1101    struct pipe_resource *tess_rings;
1102    struct pipe_resource *tess_rings_tmz;
1103    union pipe_color_union *border_color_table; /* in CPU memory, any endian */
1104    struct si_resource *border_color_buffer;
1105    union pipe_color_union *border_color_map; /* in VRAM (slow access), little endian */
1106    unsigned border_color_count;
1107    unsigned num_vs_blit_sgprs;
1108    uint32_t vs_blit_sh_data[SI_VS_BLIT_SGPRS_POS_TEXCOORD];
1109    uint32_t cs_user_data[4];
1110 
1111    /* Vertex buffers. */
1112    bool vertex_buffers_dirty;
1113    bool vertex_buffer_pointer_dirty;
1114    bool vertex_buffer_user_sgprs_dirty;
1115    struct pipe_vertex_buffer vertex_buffer[SI_NUM_VERTEX_BUFFERS];
1116    uint16_t vertex_buffer_unaligned; /* bitmask of not dword-aligned buffers */
1117    uint32_t *vb_descriptors_gpu_list;
1118    struct si_resource *vb_descriptors_buffer;
1119    unsigned vb_descriptors_offset;
1120    unsigned vb_descriptor_user_sgprs[5 * 4];
1121 
1122    /* MSAA config state. */
1123    int ps_iter_samples;
1124    bool ps_uses_fbfetch;
1125    bool smoothing_enabled;
1126 
1127    /* point smoothing state.*/
1128    bool point_smoothing_enabled;
1129 
1130    /* DB render state. */
1131    unsigned ps_db_shader_control;
1132    unsigned dbcb_copy_sample;
1133    bool dbcb_depth_copy_enabled : 1;
1134    bool dbcb_stencil_copy_enabled : 1;
1135    bool db_flush_depth_inplace : 1;
1136    bool db_flush_stencil_inplace : 1;
1137    bool db_depth_clear : 1;
1138    bool db_depth_disable_expclear : 1;
1139    bool db_stencil_clear : 1;
1140    bool db_stencil_disable_expclear : 1;
1141    bool occlusion_queries_disabled : 1;
1142    bool generate_mipmap_for_depth : 1;
1143    bool allow_flat_shading : 1;
1144 
1145    /* Emitted draw state. */
1146    bool ngg : 1;
1147    bool disable_instance_packing : 1;
1148    uint16_t ngg_culling;
1149    unsigned last_index_size;
1150    int last_base_vertex;
1151    unsigned last_start_instance;
1152    unsigned last_instance_count;
1153    unsigned last_drawid;
1154    unsigned last_sh_base_reg;
1155    int last_primitive_restart_en;
1156    unsigned last_restart_index;
1157    unsigned last_prim;
1158    unsigned last_multi_vgt_param;
1159    unsigned last_gs_out_prim;
1160    unsigned current_vs_state; /* all VS bits including LS bits */
1161    unsigned current_gs_state; /* only GS and NGG bits */
1162    unsigned last_vs_state;
1163    unsigned last_gs_state;
1164    enum pipe_prim_type current_rast_prim; /* primitive type after TES, GS */
1165 
1166    struct si_small_prim_cull_info last_small_prim_cull_info;
1167    struct si_resource *small_prim_cull_info_buf;
1168    uint64_t small_prim_cull_info_address;
1169 
1170    /* Scratch buffer */
1171    struct si_resource *scratch_buffer;
1172    unsigned spi_tmpring_size;
1173    unsigned max_seen_scratch_bytes_per_wave;
1174    unsigned max_seen_compute_scratch_bytes_per_wave;
1175 
1176    struct si_resource *compute_scratch_buffer;
1177 
1178    /* Emitted derived tessellation state. */
1179    /* Local shader (VS), or HS if LS-HS are merged. */
1180    struct si_shader *last_ls;
1181    struct si_shader_selector *last_tcs;
1182    unsigned last_num_tcs_input_cp;
1183    unsigned last_tes_sh_base;
1184    bool last_tess_uses_primid;
1185    unsigned last_num_patches;
1186    unsigned last_ls_hs_config;
1187 
1188    /* Debug state. */
1189    bool is_debug;
1190    struct si_saved_cs *current_saved_cs;
1191    uint64_t dmesg_timestamp;
1192    unsigned apitrace_call_number;
1193 
1194    /* Other state */
1195    bool need_check_render_feedback;
1196    bool decompression_enabled;
1197    bool dpbb_force_off;
1198    bool dpbb_force_off_profile_vs;
1199    bool dpbb_force_off_profile_ps;
1200    bool vs_writes_viewport_index;
1201    bool vs_disables_clipping_viewport;
1202 
1203    /* Precomputed IA_MULTI_VGT_PARAM */
1204    union si_vgt_param_key ia_multi_vgt_param_key;
1205    unsigned ia_multi_vgt_param[SI_NUM_VGT_PARAM_STATES];
1206 
1207    /* Bindless descriptors. */
1208    struct si_descriptors bindless_descriptors;
1209    struct util_idalloc bindless_used_slots;
1210    unsigned num_bindless_descriptors;
1211    bool bindless_descriptors_dirty;
1212    bool graphics_bindless_pointer_dirty;
1213    bool compute_bindless_pointer_dirty;
1214    bool gs_attribute_ring_pointer_dirty;
1215 
1216    /* Allocated bindless handles */
1217    struct hash_table *tex_handles;
1218    struct hash_table *img_handles;
1219 
1220    /* Resident bindless handles */
1221    struct util_dynarray resident_tex_handles;
1222    struct util_dynarray resident_img_handles;
1223 
1224    /* Resident bindless handles which need decompression */
1225    struct util_dynarray resident_tex_needs_color_decompress;
1226    struct util_dynarray resident_img_needs_color_decompress;
1227    struct util_dynarray resident_tex_needs_depth_decompress;
1228 
1229    /* Bindless state */
1230    bool uses_bindless_samplers;
1231    bool uses_bindless_images;
1232 
1233    /* MSAA sample locations.
1234     * The first index is the sample index.
1235     * The second index is the coordinate: X, Y. */
1236    struct {
1237       float x1[1][2];
1238       float x2[2][2];
1239       float x4[4][2];
1240       float x8[8][2];
1241       float x16[16][2];
1242    } sample_positions;
1243    struct pipe_resource *sample_pos_buffer;
1244 
1245    /* Misc stats. */
1246    unsigned num_draw_calls;
1247    unsigned num_decompress_calls;
1248    unsigned num_prim_restart_calls;
1249    unsigned num_compute_calls;
1250    unsigned num_cp_dma_calls;
1251    unsigned num_vs_flushes;
1252    unsigned num_ps_flushes;
1253    unsigned num_cs_flushes;
1254    unsigned num_cb_cache_flushes;
1255    unsigned num_db_cache_flushes;
1256    unsigned num_L2_invalidates;
1257    unsigned num_L2_writebacks;
1258    unsigned num_resident_handles;
1259    uint64_t num_alloc_tex_transfer_bytes;
1260    unsigned last_tex_ps_draw_ratio; /* for query */
1261    unsigned context_roll;
1262 
1263    /* Queries. */
1264    /* Maintain the list of active queries for pausing between IBs. */
1265    int num_occlusion_queries;
1266    int num_perfect_occlusion_queries;
1267    int num_pipeline_stat_queries;
1268    int num_pipeline_stat_emulated_queries;
1269    struct list_head active_queries;
1270    unsigned num_cs_dw_queries_suspend;
1271    /* Shared buffer for pipeline stats queries implemented with an atomic op */
1272    struct si_resource *pipeline_stats_query_buf;
1273 
1274    /* Render condition. */
1275    struct pipe_query *render_cond;
1276    unsigned render_cond_mode;
1277    bool render_cond_invert;
1278    bool render_cond_enabled; /* for u_blitter */
1279 
1280    /* Shader-based queries. */
1281    struct list_head shader_query_buffers;
1282    unsigned num_active_shader_queries;
1283 
1284    bool force_cb_shader_coherent;
1285 
1286    struct si_tracked_regs tracked_regs;
1287 
1288    /* Resources that need to be flushed, but will not get an explicit
1289     * flush_resource from the frontend and that will need to get flushed during
1290     * a context flush.
1291     */
1292    struct hash_table *dirty_implicit_resources;
1293 
1294    pipe_draw_vbo_func draw_vbo[2][2][2];
1295    pipe_draw_vertex_state_func draw_vertex_state[2][2][2];
1296    /* When b.draw_vbo is a wrapper, real_draw_vbo is the real draw_vbo function */
1297    pipe_draw_vbo_func real_draw_vbo;
1298    pipe_draw_vertex_state_func real_draw_vertex_state;
1299    void (*emit_spi_map[33])(struct si_context *sctx);
1300 
1301    /* SQTT */
1302    struct ac_thread_trace_data *thread_trace;
1303    struct ac_spm_trace_data spm_trace;
1304    struct pipe_fence_handle *last_sqtt_fence;
1305    enum rgp_sqtt_marker_event_type sqtt_next_event;
1306    bool thread_trace_enabled;
1307 
1308    unsigned context_flags;
1309 
1310    /* Shaders. */
1311    /* TODO: move other shaders here too */
1312    /* Only used for DCC MSAA clears with 4-8 fragments and 4-16 samples. */
1313    void *cs_clear_dcc_msaa[32][5][2][3][2]; /* [swizzle_mode][log2(bpe)][fragments == 8][log2(samples)-2][is_array] */
1314 };
1315 
1316 /* si_blit.c */
1317 enum si_blitter_op /* bitmask */
1318 {
1319    SI_SAVE_TEXTURES = 1,
1320    SI_SAVE_FRAMEBUFFER = 2,
1321    SI_SAVE_FRAGMENT_STATE = 4,
1322    SI_DISABLE_RENDER_COND = 8,
1323 };
1324 
1325 void si_blitter_begin(struct si_context *sctx, enum si_blitter_op op);
1326 void si_blitter_end(struct si_context *sctx);
1327 void si_init_blit_functions(struct si_context *sctx);
1328 void si_decompress_textures(struct si_context *sctx, unsigned shader_mask);
1329 void si_decompress_subresource(struct pipe_context *ctx, struct pipe_resource *tex, unsigned planes,
1330                                unsigned level, unsigned first_layer, unsigned last_layer,
1331                                bool need_fmask_expand);
1332 void si_resource_copy_region(struct pipe_context *ctx, struct pipe_resource *dst,
1333                              unsigned dst_level, unsigned dstx, unsigned dsty, unsigned dstz,
1334                              struct pipe_resource *src, unsigned src_level,
1335                              const struct pipe_box *src_box);
1336 void si_decompress_dcc(struct si_context *sctx, struct si_texture *tex);
1337 void si_flush_implicit_resources(struct si_context *sctx);
1338 bool si_msaa_resolve_blit_via_CB(struct pipe_context *ctx, const struct pipe_blit_info *info);
1339 void si_gfx_blit(struct pipe_context *ctx, const struct pipe_blit_info *info);
1340 
1341 /* si_nir_optim.c */
1342 bool si_nir_is_output_const_if_tex_is_const(nir_shader *shader, float *in, float *out, int *texunit);
1343 
1344 /* si_buffer.c */
1345 bool si_cs_is_buffer_referenced(struct si_context *sctx, struct pb_buffer *buf,
1346                                 unsigned usage);
1347 void *si_buffer_map(struct si_context *sctx, struct si_resource *resource,
1348                     unsigned usage);
1349 void si_init_resource_fields(struct si_screen *sscreen, struct si_resource *res, uint64_t size,
1350                              unsigned alignment);
1351 bool si_alloc_resource(struct si_screen *sscreen, struct si_resource *res);
1352 struct pipe_resource *pipe_aligned_buffer_create(struct pipe_screen *screen, unsigned flags,
1353                                                  unsigned usage, unsigned size, unsigned alignment);
1354 struct si_resource *si_aligned_buffer_create(struct pipe_screen *screen, unsigned flags,
1355                                              unsigned usage, unsigned size, unsigned alignment);
1356 struct pipe_resource *si_buffer_from_winsys_buffer(struct pipe_screen *screen,
1357                                                    const struct pipe_resource *templ,
1358                                                    struct pb_buffer *imported_buf,
1359                                                    uint64_t offset);
1360 void si_replace_buffer_storage(struct pipe_context *ctx, struct pipe_resource *dst,
1361                                struct pipe_resource *src, unsigned num_rebinds,
1362                                uint32_t rebind_mask, uint32_t delete_buffer_id);
1363 void si_init_screen_buffer_functions(struct si_screen *sscreen);
1364 void si_init_buffer_functions(struct si_context *sctx);
1365 
1366 /* si_clear.c */
1367 #define SI_CLEAR_TYPE_CMASK  (1 << 0)
1368 #define SI_CLEAR_TYPE_DCC    (1 << 1)
1369 #define SI_CLEAR_TYPE_HTILE  (1 << 2)
1370 
1371 struct si_clear_info {
1372    struct pipe_resource *resource;
1373    uint64_t offset;
1374    uint32_t size;
1375    uint32_t clear_value;
1376    uint32_t writemask;
1377    bool is_dcc_msaa; /* Clear it as a DCC MSAA image. */
1378 };
1379 
1380 enum pipe_format si_simplify_cb_format(enum pipe_format format);
1381 bool vi_alpha_is_on_msb(struct si_screen *sscreen, enum pipe_format format);
1382 bool vi_dcc_get_clear_info(struct si_context *sctx, struct si_texture *tex, unsigned level,
1383                            unsigned clear_value, struct si_clear_info *out);
1384 void si_init_buffer_clear(struct si_clear_info *info,
1385                           struct pipe_resource *resource, uint64_t offset,
1386                           uint32_t size, uint32_t clear_value);
1387 void si_execute_clears(struct si_context *sctx, struct si_clear_info *info,
1388                        unsigned num_clears, unsigned types);
1389 void si_init_clear_functions(struct si_context *sctx);
1390 
1391 /* si_compute_blit.c */
1392 #define SI_OP_SYNC_CS_BEFORE              (1 << 0)
1393 #define SI_OP_SYNC_PS_BEFORE              (1 << 1)
1394 #define SI_OP_SYNC_CPDMA_BEFORE           (1 << 2) /* only affects CP DMA calls */
1395 #define SI_OP_SYNC_BEFORE                 (SI_OP_SYNC_CS_BEFORE | SI_OP_SYNC_PS_BEFORE | SI_OP_SYNC_CPDMA_BEFORE)
1396 #define SI_OP_SYNC_AFTER                  (1 << 3)
1397 #define SI_OP_SYNC_BEFORE_AFTER           (SI_OP_SYNC_BEFORE | SI_OP_SYNC_AFTER)
1398 #define SI_OP_SKIP_CACHE_INV_BEFORE       (1 << 4) /* don't invalidate caches */
1399 #define SI_OP_CS_IMAGE                    (1 << 5)
1400 #define SI_OP_CS_RENDER_COND_ENABLE       (1 << 6)
1401 #define SI_OP_CPDMA_SKIP_CHECK_CS_SPACE   (1 << 7) /* don't call need_cs_space */
1402 #define SI_OP_SYNC_GE_BEFORE              (1 << 8) /* only sync VS, TCS, TES, GS */
1403 
1404 unsigned si_get_flush_flags(struct si_context *sctx, enum si_coherency coher,
1405                             enum si_cache_policy cache_policy);
1406 void si_launch_grid_internal_ssbos(struct si_context *sctx, struct pipe_grid_info *info,
1407                                    void *shader, unsigned flags, enum si_coherency coher,
1408                                    unsigned num_buffers, const struct pipe_shader_buffer *buffers,
1409                                    unsigned writeable_bitmask);
1410 enum si_clear_method {
1411   SI_CP_DMA_CLEAR_METHOD,
1412   SI_COMPUTE_CLEAR_METHOD,
1413   SI_AUTO_SELECT_CLEAR_METHOD
1414 };
1415 void si_clear_buffer(struct si_context *sctx, struct pipe_resource *dst,
1416                      uint64_t offset, uint64_t size, uint32_t *clear_value,
1417                      uint32_t clear_value_size, unsigned flags,
1418                      enum si_coherency coher, enum si_clear_method method);
1419 void si_compute_clear_buffer_rmw(struct si_context *sctx, struct pipe_resource *dst,
1420                                  unsigned dst_offset, unsigned size,
1421                                  uint32_t clear_value, uint32_t writebitmask,
1422                                  unsigned flags, enum si_coherency coher);
1423 void si_screen_clear_buffer(struct si_screen *sscreen, struct pipe_resource *dst, uint64_t offset,
1424                             uint64_t size, unsigned value, unsigned flags);
1425 void si_copy_buffer(struct si_context *sctx, struct pipe_resource *dst, struct pipe_resource *src,
1426                     uint64_t dst_offset, uint64_t src_offset, unsigned size, unsigned flags);
1427 bool si_compute_copy_image(struct si_context *sctx, struct pipe_resource *dst, unsigned dst_level,
1428                            struct pipe_resource *src, unsigned src_level, unsigned dstx,
1429                            unsigned dsty, unsigned dstz, const struct pipe_box *src_box,
1430                            unsigned flags);
1431 void si_compute_clear_render_target(struct pipe_context *ctx, struct pipe_surface *dstsurf,
1432                                     const union pipe_color_union *color, unsigned dstx,
1433                                     unsigned dsty, unsigned width, unsigned height,
1434                                     bool render_condition_enabled);
1435 void si_retile_dcc(struct si_context *sctx, struct si_texture *tex);
1436 void gfx9_clear_dcc_msaa(struct si_context *sctx, struct pipe_resource *res, uint32_t clear_value,
1437                          unsigned flags, enum si_coherency coher);
1438 void si_compute_expand_fmask(struct pipe_context *ctx, struct pipe_resource *tex);
1439 void si_init_compute_blit_functions(struct si_context *sctx);
1440 
1441 /* si_cp_dma.c */
1442 void si_cp_dma_wait_for_idle(struct si_context *sctx, struct radeon_cmdbuf *cs);
1443 void si_cp_dma_clear_buffer(struct si_context *sctx, struct radeon_cmdbuf *cs,
1444                             struct pipe_resource *dst, uint64_t offset, uint64_t size,
1445                             unsigned value, unsigned user_flags, enum si_coherency coher,
1446                             enum si_cache_policy cache_policy);
1447 void si_cp_dma_copy_buffer(struct si_context *sctx, struct pipe_resource *dst,
1448                            struct pipe_resource *src, uint64_t dst_offset, uint64_t src_offset,
1449                            unsigned size, unsigned user_flags, enum si_coherency coher,
1450                            enum si_cache_policy cache_policy);
1451 void si_test_gds(struct si_context *sctx);
1452 void si_cp_write_data(struct si_context *sctx, struct si_resource *buf, unsigned offset,
1453                       unsigned size, unsigned dst_sel, unsigned engine, const void *data);
1454 void si_cp_copy_data(struct si_context *sctx, struct radeon_cmdbuf *cs, unsigned dst_sel,
1455                      struct si_resource *dst, unsigned dst_offset, unsigned src_sel,
1456                      struct si_resource *src, unsigned src_offset);
1457 
1458 /* si_cp_reg_shadowing.c */
1459 void si_init_cp_reg_shadowing(struct si_context *sctx);
1460 
1461 /* si_debug.c */
1462 void si_save_cs(struct radeon_winsys *ws, struct radeon_cmdbuf *cs, struct radeon_saved_cs *saved,
1463                 bool get_buffer_list);
1464 void si_clear_saved_cs(struct radeon_saved_cs *saved);
1465 void si_destroy_saved_cs(struct si_saved_cs *scs);
1466 void si_auto_log_cs(void *data, struct u_log_context *log);
1467 void si_log_hw_flush(struct si_context *sctx);
1468 void si_log_draw_state(struct si_context *sctx, struct u_log_context *log);
1469 void si_log_compute_state(struct si_context *sctx, struct u_log_context *log);
1470 void si_init_debug_functions(struct si_context *sctx);
1471 void si_check_vm_faults(struct si_context *sctx, struct radeon_saved_cs *saved,
1472                         enum amd_ip_type ring);
1473 bool si_replace_shader(unsigned num, struct si_shader_binary *binary);
1474 void si_print_current_ib(struct si_context *sctx, FILE *f);
1475 
1476 /* si_fence.c */
1477 void si_cp_release_mem(struct si_context *ctx, struct radeon_cmdbuf *cs, unsigned event,
1478                        unsigned event_flags, unsigned dst_sel, unsigned int_sel, unsigned data_sel,
1479                        struct si_resource *buf, uint64_t va, uint32_t new_fence,
1480                        unsigned query_type);
1481 unsigned si_cp_write_fence_dwords(struct si_screen *screen);
1482 void si_cp_wait_mem(struct si_context *ctx, struct radeon_cmdbuf *cs, uint64_t va, uint32_t ref,
1483                     uint32_t mask, unsigned flags);
1484 void si_init_fence_functions(struct si_context *ctx);
1485 void si_init_screen_fence_functions(struct si_screen *screen);
1486 struct pipe_fence_handle *si_create_fence(struct pipe_context *ctx,
1487                                           struct tc_unflushed_batch_token *tc_token);
1488 
1489 /* si_get.c */
1490 void si_init_screen_get_functions(struct si_screen *sscreen);
1491 
1492 bool si_sdma_copy_image(struct si_context *ctx, struct si_texture *dst, struct si_texture *src);
1493 
1494 /* si_gfx_cs.c */
1495 void si_flush_gfx_cs(struct si_context *ctx, unsigned flags, struct pipe_fence_handle **fence);
1496 void si_allocate_gds(struct si_context *ctx);
1497 void si_set_tracked_regs_to_clear_state(struct si_context *ctx);
1498 void si_begin_new_gfx_cs(struct si_context *ctx, bool first_cs);
1499 void si_trace_emit(struct si_context *sctx);
1500 void si_emit_surface_sync(struct si_context *sctx, struct radeon_cmdbuf *cs,
1501                           unsigned cp_coher_cntl);
1502 void gfx10_emit_cache_flush(struct si_context *sctx, struct radeon_cmdbuf *cs);
1503 void si_emit_cache_flush(struct si_context *sctx, struct radeon_cmdbuf *cs);
1504 /* Replace the sctx->b.draw_vbo function with a wrapper. This can be use to implement
1505  * optimizations without affecting the normal draw_vbo functions perf.
1506  */
1507 void si_install_draw_wrapper(struct si_context *sctx, pipe_draw_vbo_func wrapper,
1508                              pipe_draw_vertex_state_func vstate_wrapper);
1509 
1510 /* si_gpu_load.c */
1511 void si_gpu_load_kill_thread(struct si_screen *sscreen);
1512 uint64_t si_begin_counter(struct si_screen *sscreen, unsigned type);
1513 unsigned si_end_counter(struct si_screen *sscreen, unsigned type, uint64_t begin);
1514 
1515 /* si_compute.c */
1516 void si_emit_initial_compute_regs(struct si_context *sctx, struct radeon_cmdbuf *cs);
1517 void si_init_compute_functions(struct si_context *sctx);
1518 
1519 /* si_pipe.c */
1520 bool si_init_compiler(struct si_screen *sscreen, struct ac_llvm_compiler *compiler);
1521 void si_init_aux_async_compute_ctx(struct si_screen *sscreen);
1522 struct si_context* si_get_aux_context(struct si_screen *sscreen);
1523 void si_put_aux_context_flush(struct si_screen *sscreen);
1524 
1525 /* si_perfcounters.c */
1526 void si_init_perfcounters(struct si_screen *screen);
1527 void si_destroy_perfcounters(struct si_screen *screen);
1528 void si_inhibit_clockgating(struct si_context *sctx, struct radeon_cmdbuf *cs, bool inhibit);
1529 void si_pc_emit_shaders(struct radeon_cmdbuf *cs, unsigned shaders);
1530 void si_pc_emit_spm_start(struct radeon_cmdbuf *cs);
1531 void si_pc_emit_spm_stop(struct radeon_cmdbuf *cs, bool never_stop_sq_perf_counters,
1532                          bool never_send_perfcounter_stop);
1533 void si_pc_emit_spm_reset(struct radeon_cmdbuf *cs);
1534 void si_emit_spm_setup(struct si_context *sctx, struct radeon_cmdbuf *cs);
1535 bool si_spm_init(struct si_context *sctx);
1536 void si_spm_finish(struct si_context *sctx);
1537 
1538 /* si_query.c */
1539 void si_init_screen_query_functions(struct si_screen *sscreen);
1540 void si_init_query_functions(struct si_context *sctx);
1541 void si_suspend_queries(struct si_context *sctx);
1542 void si_resume_queries(struct si_context *sctx);
1543 
1544 /* si_shaderlib_nir.c */
1545 void *si_create_copy_image_cs(struct si_context *sctx, bool src_is_1d_array, bool dst_is_1d_array);
1546 void *si_create_dcc_retile_cs(struct si_context *sctx, struct radeon_surf *surf);
1547 void *gfx9_create_clear_dcc_msaa_cs(struct si_context *sctx, struct si_texture *tex);
1548 void *si_create_passthrough_tcs(struct si_context *sctx);
1549 
1550 /* si_shaderlib_tgsi.c */
1551 void *si_get_blitter_vs(struct si_context *sctx, enum blitter_attrib_type type,
1552                         unsigned num_layers);
1553 void *si_create_dma_compute_shader(struct pipe_context *ctx, unsigned num_dwords_per_thread,
1554                                    bool dst_stream_cache_policy, bool is_copy);
1555 void *si_create_clear_buffer_rmw_cs(struct si_context *sctx);
1556 void *si_clear_render_target_shader(struct pipe_context *ctx);
1557 void *si_clear_render_target_shader_1d_array(struct pipe_context *ctx);
1558 void *si_clear_12bytes_buffer_shader(struct pipe_context *ctx);
1559 void *si_create_fmask_expand_cs(struct pipe_context *ctx, unsigned num_samples, bool is_array);
1560 void *si_create_query_result_cs(struct si_context *sctx);
1561 void *gfx10_create_sh_query_result_cs(struct si_context *sctx);
1562 
1563 /* gfx10_query.c */
1564 void gfx10_init_query(struct si_context *sctx);
1565 void gfx10_destroy_query(struct si_context *sctx);
1566 
1567 /* si_test_image_copy_region.c */
1568 void si_test_image_copy_region(struct si_screen *sscreen);
1569 void si_test_blit(struct si_screen *sscreen, unsigned test_flags);
1570 
1571 /* si_test_clearbuffer.c */
1572 void si_test_dma_perf(struct si_screen *sscreen);
1573 
1574 /* si_uvd.c */
1575 struct pipe_video_codec *si_uvd_create_decoder(struct pipe_context *context,
1576                                                const struct pipe_video_codec *templ);
1577 
1578 struct pipe_video_buffer *si_video_buffer_create(struct pipe_context *pipe,
1579                                                  const struct pipe_video_buffer *tmpl);
1580 struct pipe_video_buffer *si_video_buffer_create_with_modifiers(struct pipe_context *pipe,
1581                                                                 const struct pipe_video_buffer *tmpl,
1582                                                                 const uint64_t *modifiers,
1583                                                                 unsigned int modifiers_count);
1584 
1585 /* si_state_viewport.c */
1586 void si_update_vs_viewport_state(struct si_context *ctx);
1587 void si_init_viewport_functions(struct si_context *ctx);
1588 
1589 /* si_texture.c */
1590 void si_eliminate_fast_color_clear(struct si_context *sctx, struct si_texture *tex,
1591                                    bool *ctx_flushed);
1592 void si_texture_discard_cmask(struct si_screen *sscreen, struct si_texture *tex);
1593 bool si_init_flushed_depth_texture(struct pipe_context *ctx, struct pipe_resource *texture);
1594 void si_print_texture_info(struct si_screen *sscreen, struct si_texture *tex,
1595                            struct u_log_context *log);
1596 struct pipe_resource *si_texture_create(struct pipe_screen *screen,
1597                                         const struct pipe_resource *templ);
1598 bool si_texture_commit(struct si_context *ctx, struct si_resource *res, unsigned level,
1599                        struct pipe_box *box, bool commit);
1600 bool vi_dcc_formats_compatible(struct si_screen *sscreen, enum pipe_format format1,
1601                                enum pipe_format format2);
1602 bool vi_dcc_formats_are_incompatible(struct pipe_resource *tex, unsigned level,
1603                                      enum pipe_format view_format);
1604 void vi_disable_dcc_if_incompatible_format(struct si_context *sctx, struct pipe_resource *tex,
1605                                            unsigned level, enum pipe_format view_format);
1606 unsigned si_translate_colorswap(enum amd_gfx_level gfx_level, enum pipe_format format,
1607                                 bool do_endian_swap);
1608 bool si_texture_disable_dcc(struct si_context *sctx, struct si_texture *tex);
1609 void si_init_screen_texture_functions(struct si_screen *sscreen);
1610 void si_init_context_texture_functions(struct si_context *sctx);
1611 
1612 /* si_sqtt.c */
1613 void si_sqtt_write_event_marker(struct si_context* sctx, struct radeon_cmdbuf *rcs,
1614                                 enum rgp_sqtt_marker_event_type api_type,
1615                                 uint32_t vertex_offset_user_data,
1616                                 uint32_t instance_offset_user_data,
1617                                 uint32_t draw_index_user_data);
1618 bool si_sqtt_register_pipeline(struct si_context* sctx, uint64_t pipeline_hash, uint64_t base_address, bool is_compute);
1619 bool si_sqtt_pipeline_is_registered(struct ac_thread_trace_data *thread_trace_data,
1620                                     uint64_t pipeline_hash);
1621 void si_sqtt_describe_pipeline_bind(struct si_context* sctx, uint64_t pipeline_hash, int bind_point);
1622 void
1623 si_write_event_with_dims_marker(struct si_context* sctx, struct radeon_cmdbuf *rcs,
1624                                 enum rgp_sqtt_marker_event_type api_type,
1625                                 uint32_t x, uint32_t y, uint32_t z);
1626 void
1627 si_write_user_event(struct si_context* sctx, struct radeon_cmdbuf *rcs,
1628                     enum rgp_sqtt_marker_user_event_type type,
1629                     const char *str, int len);
1630 void
1631 si_sqtt_describe_barrier_start(struct si_context* sctx, struct radeon_cmdbuf *rcs);
1632 void
1633 si_sqtt_describe_barrier_end(struct si_context* sctx, struct radeon_cmdbuf *rcs, unsigned flags);
1634 bool si_init_thread_trace(struct si_context *sctx);
1635 void si_destroy_thread_trace(struct si_context *sctx);
1636 void si_handle_thread_trace(struct si_context *sctx, struct radeon_cmdbuf *rcs);
1637 
1638 /*
1639  * common helpers
1640  */
1641 
si_resource(struct pipe_resource * r)1642 static inline struct si_resource *si_resource(struct pipe_resource *r)
1643 {
1644    return (struct si_resource *)r;
1645 }
1646 
si_resource_reference(struct si_resource ** ptr,struct si_resource * res)1647 static inline void si_resource_reference(struct si_resource **ptr, struct si_resource *res)
1648 {
1649    pipe_resource_reference((struct pipe_resource **)ptr, (struct pipe_resource *)res);
1650 }
1651 
si_texture_reference(struct si_texture ** ptr,struct si_texture * res)1652 static inline void si_texture_reference(struct si_texture **ptr, struct si_texture *res)
1653 {
1654    pipe_resource_reference((struct pipe_resource **)ptr, &res->buffer.b.b);
1655 }
1656 
1657 static inline void
si_shader_selector_reference(struct si_context * sctx,struct si_shader_selector ** dst,struct si_shader_selector * src)1658 si_shader_selector_reference(struct si_context *sctx, /* sctx can optionally be NULL */
1659                              struct si_shader_selector **dst, struct si_shader_selector *src)
1660 {
1661    if (*dst == src)
1662       return;
1663 
1664    struct si_screen *sscreen = src ? src->screen : (*dst)->screen;
1665    util_shader_reference(&sctx->b, &sscreen->live_shader_cache, (void **)dst, src);
1666 }
1667 
vi_dcc_enabled(struct si_texture * tex,unsigned level)1668 static inline bool vi_dcc_enabled(struct si_texture *tex, unsigned level)
1669 {
1670    return !tex->is_depth && tex->surface.meta_offset && level < tex->surface.num_meta_levels;
1671 }
1672 
si_tile_mode_index(struct si_texture * tex,unsigned level,bool stencil)1673 static inline unsigned si_tile_mode_index(struct si_texture *tex, unsigned level, bool stencil)
1674 {
1675    if (stencil)
1676       return tex->surface.u.legacy.zs.stencil_tiling_index[level];
1677    else
1678       return tex->surface.u.legacy.tiling_index[level];
1679 }
1680 
si_get_minimum_num_gfx_cs_dwords(struct si_context * sctx,unsigned num_draws)1681 static inline unsigned si_get_minimum_num_gfx_cs_dwords(struct si_context *sctx,
1682                                                         unsigned num_draws)
1683 {
1684    /* Don't count the needed CS space exactly and just use an upper bound.
1685     *
1686     * Also reserve space for stopping queries at the end of IB, because
1687     * the number of active queries is unlimited in theory.
1688     */
1689    return 2048 + sctx->num_cs_dw_queries_suspend + num_draws * 10;
1690 }
1691 
si_context_add_resource_size(struct si_context * sctx,struct pipe_resource * r)1692 static inline void si_context_add_resource_size(struct si_context *sctx, struct pipe_resource *r)
1693 {
1694    if (r) {
1695       /* Add memory usage for need_gfx_cs_space */
1696       sctx->memory_usage_kb += si_resource(r)->memory_usage_kb;
1697    }
1698 }
1699 
si_invalidate_draw_sh_constants(struct si_context * sctx)1700 static inline void si_invalidate_draw_sh_constants(struct si_context *sctx)
1701 {
1702    sctx->last_base_vertex = SI_BASE_VERTEX_UNKNOWN;
1703    sctx->last_start_instance = SI_START_INSTANCE_UNKNOWN;
1704    sctx->last_drawid = SI_DRAW_ID_UNKNOWN;
1705 }
1706 
si_invalidate_draw_constants(struct si_context * sctx)1707 static inline void si_invalidate_draw_constants(struct si_context *sctx)
1708 {
1709    si_invalidate_draw_sh_constants(sctx);
1710    sctx->last_instance_count = SI_INSTANCE_COUNT_UNKNOWN;
1711 }
1712 
si_get_atom_bit(struct si_context * sctx,struct si_atom * atom)1713 static inline unsigned si_get_atom_bit(struct si_context *sctx, struct si_atom *atom)
1714 {
1715    return 1 << (atom - sctx->atoms.array);
1716 }
1717 
si_set_atom_dirty(struct si_context * sctx,struct si_atom * atom,bool dirty)1718 static inline void si_set_atom_dirty(struct si_context *sctx, struct si_atom *atom, bool dirty)
1719 {
1720    unsigned bit = si_get_atom_bit(sctx, atom);
1721 
1722    if (dirty)
1723       sctx->dirty_atoms |= bit;
1724    else
1725       sctx->dirty_atoms &= ~bit;
1726 }
1727 
si_is_atom_dirty(struct si_context * sctx,struct si_atom * atom)1728 static inline bool si_is_atom_dirty(struct si_context *sctx, struct si_atom *atom)
1729 {
1730    return (sctx->dirty_atoms & si_get_atom_bit(sctx, atom)) != 0;
1731 }
1732 
si_mark_atom_dirty(struct si_context * sctx,struct si_atom * atom)1733 static inline void si_mark_atom_dirty(struct si_context *sctx, struct si_atom *atom)
1734 {
1735    si_set_atom_dirty(sctx, atom, true);
1736 }
1737 
1738 /* This should be evaluated at compile time if all parameters except sctx are constants. */
1739 static ALWAYS_INLINE struct si_shader_ctx_state *
si_get_vs_inline(struct si_context * sctx,enum si_has_tess has_tess,enum si_has_gs has_gs)1740 si_get_vs_inline(struct si_context *sctx, enum si_has_tess has_tess, enum si_has_gs has_gs)
1741 {
1742    if (has_gs)
1743       return &sctx->shader.gs;
1744    if (has_tess)
1745       return &sctx->shader.tes;
1746 
1747    return &sctx->shader.vs;
1748 }
1749 
si_get_vs(struct si_context * sctx)1750 static inline struct si_shader_ctx_state *si_get_vs(struct si_context *sctx)
1751 {
1752    return si_get_vs_inline(sctx, sctx->shader.tes.cso ? TESS_ON : TESS_OFF,
1753                            sctx->shader.gs.cso ? GS_ON : GS_OFF);
1754 }
1755 
si_can_dump_shader(struct si_screen * sscreen,gl_shader_stage stage)1756 static inline bool si_can_dump_shader(struct si_screen *sscreen, gl_shader_stage stage)
1757 {
1758    return sscreen->debug_flags & (1 << stage);
1759 }
1760 
si_get_strmout_en(struct si_context * sctx)1761 static inline bool si_get_strmout_en(struct si_context *sctx)
1762 {
1763    return sctx->streamout.streamout_enabled || sctx->streamout.prims_gen_query_enabled;
1764 }
1765 
si_optimal_tcc_alignment(struct si_context * sctx,unsigned upload_size)1766 static inline unsigned si_optimal_tcc_alignment(struct si_context *sctx, unsigned upload_size)
1767 {
1768    unsigned alignment, tcc_cache_line_size;
1769 
1770    /* If the upload size is less than the cache line size (e.g. 16, 32),
1771     * the whole thing will fit into a cache line if we align it to its size.
1772     * The idea is that multiple small uploads can share a cache line.
1773     * If the upload size is greater, align it to the cache line size.
1774     */
1775    alignment = util_next_power_of_two(upload_size);
1776    tcc_cache_line_size = sctx->screen->info.tcc_cache_line_size;
1777    return MIN2(alignment, tcc_cache_line_size);
1778 }
1779 
si_saved_cs_reference(struct si_saved_cs ** dst,struct si_saved_cs * src)1780 static inline void si_saved_cs_reference(struct si_saved_cs **dst, struct si_saved_cs *src)
1781 {
1782    if (pipe_reference(&(*dst)->reference, &src->reference))
1783       si_destroy_saved_cs(*dst);
1784 
1785    *dst = src;
1786 }
1787 
si_make_CB_shader_coherent(struct si_context * sctx,unsigned num_samples,bool shaders_read_metadata,bool dcc_pipe_aligned)1788 static inline void si_make_CB_shader_coherent(struct si_context *sctx, unsigned num_samples,
1789                                               bool shaders_read_metadata, bool dcc_pipe_aligned)
1790 {
1791    sctx->flags |= SI_CONTEXT_FLUSH_AND_INV_CB | SI_CONTEXT_INV_VCACHE;
1792    sctx->force_cb_shader_coherent = false;
1793 
1794    if (sctx->gfx_level >= GFX10) {
1795       if (sctx->screen->info.tcc_rb_non_coherent)
1796          sctx->flags |= SI_CONTEXT_INV_L2;
1797       else if (shaders_read_metadata)
1798          sctx->flags |= SI_CONTEXT_INV_L2_METADATA;
1799    } else if (sctx->gfx_level == GFX9) {
1800       /* Single-sample color is coherent with shaders on GFX9, but
1801        * L2 metadata must be flushed if shaders read metadata.
1802        * (DCC, CMASK).
1803        */
1804       if (num_samples >= 2 || (shaders_read_metadata && !dcc_pipe_aligned))
1805          sctx->flags |= SI_CONTEXT_INV_L2;
1806       else if (shaders_read_metadata)
1807          sctx->flags |= SI_CONTEXT_INV_L2_METADATA;
1808    } else {
1809       /* GFX6-GFX8 */
1810       sctx->flags |= SI_CONTEXT_INV_L2;
1811    }
1812 }
1813 
si_make_DB_shader_coherent(struct si_context * sctx,unsigned num_samples,bool include_stencil,bool shaders_read_metadata)1814 static inline void si_make_DB_shader_coherent(struct si_context *sctx, unsigned num_samples,
1815                                               bool include_stencil, bool shaders_read_metadata)
1816 {
1817    sctx->flags |= SI_CONTEXT_FLUSH_AND_INV_DB | SI_CONTEXT_INV_VCACHE;
1818 
1819    if (sctx->gfx_level >= GFX10) {
1820       if (sctx->screen->info.tcc_rb_non_coherent)
1821          sctx->flags |= SI_CONTEXT_INV_L2;
1822       else if (shaders_read_metadata)
1823          sctx->flags |= SI_CONTEXT_INV_L2_METADATA;
1824    } else if (sctx->gfx_level == GFX9) {
1825       /* Single-sample depth (not stencil) is coherent with shaders
1826        * on GFX9, but L2 metadata must be flushed if shaders read
1827        * metadata.
1828        */
1829       if (num_samples >= 2 || include_stencil)
1830          sctx->flags |= SI_CONTEXT_INV_L2;
1831       else if (shaders_read_metadata)
1832          sctx->flags |= SI_CONTEXT_INV_L2_METADATA;
1833    } else {
1834       /* GFX6-GFX8 */
1835       sctx->flags |= SI_CONTEXT_INV_L2;
1836    }
1837 }
1838 
si_can_sample_zs(struct si_texture * tex,bool stencil_sampler)1839 static inline bool si_can_sample_zs(struct si_texture *tex, bool stencil_sampler)
1840 {
1841    return (stencil_sampler && tex->can_sample_s) || (!stencil_sampler && tex->can_sample_z);
1842 }
1843 
si_htile_enabled(struct si_texture * tex,unsigned level,unsigned zs_mask)1844 static inline bool si_htile_enabled(struct si_texture *tex, unsigned level, unsigned zs_mask)
1845 {
1846    if (zs_mask == PIPE_MASK_S && (tex->htile_stencil_disabled || !tex->surface.has_stencil))
1847       return false;
1848 
1849    if (!tex->is_depth || !tex->surface.meta_offset)
1850       return false;
1851 
1852    struct si_screen *sscreen = (struct si_screen *)tex->buffer.b.b.screen;
1853    if (sscreen->info.gfx_level >= GFX8) {
1854       return level < tex->surface.num_meta_levels;
1855    } else {
1856       /* GFX6-7 don't have TC-compatible HTILE, which means they have to run
1857        * a decompression pass for every mipmap level before texturing, so compress
1858        * only one level to reduce the number of decompression passes to a minimum.
1859        */
1860       return level == 0;
1861    }
1862 }
1863 
vi_tc_compat_htile_enabled(struct si_texture * tex,unsigned level,unsigned zs_mask)1864 static inline bool vi_tc_compat_htile_enabled(struct si_texture *tex, unsigned level,
1865                                               unsigned zs_mask)
1866 {
1867    assert(!tex->tc_compatible_htile || tex->surface.meta_offset);
1868    return tex->tc_compatible_htile && si_htile_enabled(tex, level, zs_mask);
1869 }
1870 
si_get_ps_iter_samples(struct si_context * sctx)1871 static inline unsigned si_get_ps_iter_samples(struct si_context *sctx)
1872 {
1873    if (sctx->ps_uses_fbfetch)
1874       return sctx->framebuffer.nr_color_samples;
1875 
1876    return MIN2(sctx->ps_iter_samples, sctx->framebuffer.nr_color_samples);
1877 }
1878 
si_get_total_colormask(struct si_context * sctx)1879 static inline unsigned si_get_total_colormask(struct si_context *sctx)
1880 {
1881    if (sctx->queued.named.rasterizer->rasterizer_discard)
1882       return 0;
1883 
1884    struct si_shader_selector *ps = sctx->shader.ps.cso;
1885    if (!ps)
1886       return 0;
1887 
1888    unsigned colormask =
1889       sctx->framebuffer.colorbuf_enabled_4bit & sctx->queued.named.blend->cb_target_mask;
1890 
1891    if (!ps->info.color0_writes_all_cbufs)
1892       colormask &= ps->info.colors_written_4bit;
1893    else if (!ps->info.colors_written_4bit)
1894       colormask = 0; /* color0 writes all cbufs, but it's not written */
1895 
1896    return colormask;
1897 }
1898 
1899 #define UTIL_ALL_PRIM_LINE_MODES                                                                   \
1900    ((1 << PIPE_PRIM_LINES) | (1 << PIPE_PRIM_LINE_LOOP) | (1 << PIPE_PRIM_LINE_STRIP) |            \
1901     (1 << PIPE_PRIM_LINES_ADJACENCY) | (1 << PIPE_PRIM_LINE_STRIP_ADJACENCY))
1902 
1903 #define UTIL_ALL_PRIM_TRIANGLE_MODES \
1904    ((1 << PIPE_PRIM_TRIANGLES) | (1 << PIPE_PRIM_TRIANGLE_STRIP) | \
1905     (1 << PIPE_PRIM_TRIANGLE_FAN) | (1 << PIPE_PRIM_QUADS) | (1 << PIPE_PRIM_QUAD_STRIP) | \
1906     (1 << PIPE_PRIM_POLYGON) | (1 << PIPE_PRIM_TRIANGLES_ADJACENCY) | \
1907     (1 << PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY))
1908 
util_prim_is_lines(unsigned prim)1909 static inline bool util_prim_is_lines(unsigned prim)
1910 {
1911    return ((1 << prim) & UTIL_ALL_PRIM_LINE_MODES) != 0;
1912 }
1913 
util_prim_is_points_or_lines(unsigned prim)1914 static inline bool util_prim_is_points_or_lines(unsigned prim)
1915 {
1916    return ((1 << prim) & (UTIL_ALL_PRIM_LINE_MODES | (1 << PIPE_PRIM_POINTS))) != 0;
1917 }
1918 
util_rast_prim_is_triangles(unsigned prim)1919 static inline bool util_rast_prim_is_triangles(unsigned prim)
1920 {
1921    return ((1 << prim) & UTIL_ALL_PRIM_TRIANGLE_MODES) != 0;
1922 }
1923 
util_rast_prim_is_lines_or_triangles(unsigned prim)1924 static inline bool util_rast_prim_is_lines_or_triangles(unsigned prim)
1925 {
1926    return ((1 << prim) & (UTIL_ALL_PRIM_LINE_MODES | UTIL_ALL_PRIM_TRIANGLE_MODES)) != 0;
1927 }
1928 
1929 /**
1930  * Return true if there is enough memory in VRAM and GTT for the buffers
1931  * added so far.
1932  *
1933  * \param vram      VRAM memory size not added to the buffer list yet
1934  * \param gtt       GTT memory size not added to the buffer list yet
1935  */
radeon_cs_memory_below_limit(struct si_screen * screen,struct radeon_cmdbuf * cs,uint32_t kb)1936 static inline bool radeon_cs_memory_below_limit(struct si_screen *screen, struct radeon_cmdbuf *cs,
1937                                                 uint32_t kb)
1938 {
1939    return kb + cs->used_vram_kb + cs->used_gart_kb < screen->max_memory_usage_kb;
1940 }
1941 
si_need_gfx_cs_space(struct si_context * ctx,unsigned num_draws)1942 static inline void si_need_gfx_cs_space(struct si_context *ctx, unsigned num_draws)
1943 {
1944    struct radeon_cmdbuf *cs = &ctx->gfx_cs;
1945 
1946    /* There are two memory usage counters in the winsys for all buffers
1947     * that have been added (cs_add_buffer) and one counter in the pipe
1948     * driver for those that haven't been added yet.
1949     */
1950    uint32_t kb = ctx->memory_usage_kb;
1951    ctx->memory_usage_kb = 0;
1952 
1953    if (radeon_cs_memory_below_limit(ctx->screen, &ctx->gfx_cs, kb) &&
1954        ctx->ws->cs_check_space(cs, si_get_minimum_num_gfx_cs_dwords(ctx, num_draws)))
1955       return;
1956 
1957    si_flush_gfx_cs(ctx, RADEON_FLUSH_ASYNC_START_NEXT_GFX_IB_NOW, NULL);
1958 }
1959 
1960 /**
1961  * Add a buffer to the buffer list for the given command stream (CS).
1962  *
1963  * All buffers used by a CS must be added to the list. This tells the kernel
1964  * driver which buffers are used by GPU commands. Other buffers can
1965  * be swapped out (not accessible) during execution.
1966  *
1967  * The buffer list becomes empty after every context flush and must be
1968  * rebuilt.
1969  */
radeon_add_to_buffer_list(struct si_context * sctx,struct radeon_cmdbuf * cs,struct si_resource * bo,unsigned usage)1970 static inline void radeon_add_to_buffer_list(struct si_context *sctx, struct radeon_cmdbuf *cs,
1971                                              struct si_resource *bo, unsigned usage)
1972 {
1973    assert(usage);
1974    sctx->ws->cs_add_buffer(cs, bo->buf, usage | RADEON_USAGE_SYNCHRONIZED,
1975                            bo->domains);
1976 }
1977 
1978 /**
1979  * Same as above, but also checks memory usage and flushes the context
1980  * accordingly.
1981  *
1982  * When this SHOULD NOT be used:
1983  *
1984  * - if si_context_add_resource_size has been called for the buffer
1985  *   followed by *_need_cs_space for checking the memory usage
1986  *
1987  * - when emitting state packets and draw packets (because preceding packets
1988  *   can't be re-emitted at that point)
1989  *
1990  * - if shader resource "enabled_mask" is not up-to-date or there is
1991  *   a different constraint disallowing a context flush
1992  */
radeon_add_to_gfx_buffer_list_check_mem(struct si_context * sctx,struct si_resource * bo,unsigned usage,bool check_mem)1993 static inline void radeon_add_to_gfx_buffer_list_check_mem(struct si_context *sctx,
1994                                                            struct si_resource *bo,
1995                                                            unsigned usage,
1996                                                            bool check_mem)
1997 {
1998    if (check_mem &&
1999        !radeon_cs_memory_below_limit(sctx->screen, &sctx->gfx_cs, sctx->memory_usage_kb + bo->memory_usage_kb))
2000       si_flush_gfx_cs(sctx, RADEON_FLUSH_ASYNC_START_NEXT_GFX_IB_NOW, NULL);
2001 
2002    radeon_add_to_buffer_list(sctx, &sctx->gfx_cs, bo, usage);
2003 }
2004 
si_select_draw_vbo(struct si_context * sctx)2005 static inline void si_select_draw_vbo(struct si_context *sctx)
2006 {
2007    pipe_draw_vbo_func draw_vbo = sctx->draw_vbo[!!sctx->shader.tes.cso]
2008                                                [!!sctx->shader.gs.cso]
2009                                                [sctx->ngg];
2010    pipe_draw_vertex_state_func draw_vertex_state =
2011       sctx->draw_vertex_state[!!sctx->shader.tes.cso]
2012                              [!!sctx->shader.gs.cso]
2013                              [sctx->ngg];
2014    assert(draw_vbo);
2015    assert(draw_vertex_state);
2016 
2017    if (unlikely(sctx->real_draw_vbo)) {
2018       assert(sctx->real_draw_vertex_state);
2019       sctx->real_draw_vbo = draw_vbo;
2020       sctx->real_draw_vertex_state = draw_vertex_state;
2021    } else {
2022       assert(!sctx->real_draw_vertex_state);
2023       sctx->b.draw_vbo = draw_vbo;
2024       sctx->b.draw_vertex_state = draw_vertex_state;
2025    }
2026 }
2027 
2028 /* Return the number of samples that the rasterizer uses. */
si_get_num_coverage_samples(struct si_context * sctx)2029 static inline unsigned si_get_num_coverage_samples(struct si_context *sctx)
2030 {
2031    if (sctx->framebuffer.nr_samples > 1 &&
2032        sctx->queued.named.rasterizer->multisample_enable)
2033       return sctx->framebuffer.nr_samples;
2034 
2035    /* Note that smoothing_enabled is set by si_update_shaders. */
2036    if (sctx->smoothing_enabled)
2037       return SI_NUM_SMOOTH_AA_SAMPLES;
2038 
2039    return 1;
2040 }
2041 
2042 static unsigned ALWAYS_INLINE
si_num_vbos_in_user_sgprs_inline(enum amd_gfx_level gfx_level)2043 si_num_vbos_in_user_sgprs_inline(enum amd_gfx_level gfx_level)
2044 {
2045    /* This decreases CPU overhead if all descriptors are in user SGPRs because we don't
2046     * have to allocate and count references for the upload buffer.
2047     */
2048    return gfx_level >= GFX9 ? 5 : 1;
2049 }
2050 
si_num_vbos_in_user_sgprs(struct si_screen * sscreen)2051 static inline unsigned si_num_vbos_in_user_sgprs(struct si_screen *sscreen)
2052 {
2053    return si_num_vbos_in_user_sgprs_inline(sscreen->info.gfx_level);
2054 }
2055 
2056 static inline
si_check_dirty_buffers_textures(struct si_context * sctx)2057 void si_check_dirty_buffers_textures(struct si_context *sctx)
2058 {
2059    /* Recompute and re-emit the texture resource states if needed. */
2060    unsigned dirty_tex_counter = p_atomic_read(&sctx->screen->dirty_tex_counter);
2061    if (unlikely(dirty_tex_counter != sctx->last_dirty_tex_counter)) {
2062       sctx->last_dirty_tex_counter = dirty_tex_counter;
2063       sctx->framebuffer.dirty_cbufs |= ((1 << sctx->framebuffer.state.nr_cbufs) - 1);
2064       sctx->framebuffer.dirty_zsbuf = true;
2065       si_mark_atom_dirty(sctx, &sctx->atoms.s.framebuffer);
2066       si_update_all_texture_descriptors(sctx);
2067    }
2068 
2069    unsigned dirty_buf_counter = p_atomic_read(&sctx->screen->dirty_buf_counter);
2070    if (unlikely(dirty_buf_counter != sctx->last_dirty_buf_counter)) {
2071       sctx->last_dirty_buf_counter = dirty_buf_counter;
2072       /* Rebind all buffers unconditionally. */
2073       si_rebind_buffer(sctx, NULL);
2074    }
2075 }
2076 
2077 
2078 #define PRINT_ERR(fmt, args...)                                                                    \
2079    fprintf(stderr, "EE %s:%d %s - " fmt, __FILE__, __LINE__, __func__, ##args)
2080 
2081 #ifdef __cplusplus
2082 }
2083 #endif
2084 
2085 #endif
2086