2 * Copyright 2012 Advanced Micro Devices, Inc.
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.
24 * Tom Stellard <thomas.stellard@amd.com>
25 * Michel Dänzer <michel.daenzer@amd.com>
26 * Christian König <christian.koenig@amd.com>
29 /* How linking shader inputs and outputs between vertex, tessellation, and
30 * geometry shaders works.
32 * Inputs and outputs between shaders are stored in a buffer. This buffer
33 * lives in LDS (typical case for tessellation), but it can also live
34 * in memory (ESGS). Each input or output has a fixed location within a vertex.
35 * The highest used input or output determines the stride between vertices.
37 * Since GS and tessellation are only possible in the OpenGL core profile,
38 * only these semantics are valid for per-vertex data:
45 * CULLDIST0..1 (not implemented)
48 * For example, a shader only writing GENERIC0 has the output stride of 5.
50 * Only these semantics are valid for per-patch data:
58 * That's how independent shaders agree on input and output locations.
59 * The si_shader_io_get_unique_index function assigns the locations.
61 * For tessellation, other required information for calculating the input and
62 * output addresses like the vertex stride, the patch stride, and the offsets
63 * where per-vertex and per-patch data start, is passed to the shader via
64 * user data SGPRs. The offsets and strides are calculated at draw time and
65 * aren't available at compile time.
71 #include <llvm-c/Core.h> /* LLVMModuleRef */
72 #include "tgsi/tgsi_scan.h"
75 struct radeon_shader_binary
;
76 struct radeon_shader_reloc
;
78 #define SI_SGPR_RW_BUFFERS 0 /* rings (& stream-out, VS only) */
79 #define SI_SGPR_CONST_BUFFERS 2
80 #define SI_SGPR_SAMPLER_STATES 4
81 #define SI_SGPR_SAMPLER_VIEWS 6
82 #define SI_SGPR_VERTEX_BUFFERS 8 /* VS only */
83 #define SI_SGPR_BASE_VERTEX 10 /* VS only */
84 #define SI_SGPR_START_INSTANCE 11 /* VS only */
85 #define SI_SGPR_VS_STATE_BITS 12 /* VS(VS) only */
86 #define SI_SGPR_LS_OUT_LAYOUT 12 /* VS(LS) only */
87 #define SI_SGPR_TCS_OUT_OFFSETS 8 /* TCS & TES only */
88 #define SI_SGPR_TCS_OUT_LAYOUT 9 /* TCS & TES only */
89 #define SI_SGPR_TCS_IN_LAYOUT 10 /* TCS only */
90 #define SI_SGPR_ALPHA_REF 8 /* PS only */
91 #define SI_SGPR_PS_STATE_BITS 9 /* PS only */
93 #define SI_VS_NUM_USER_SGPR 13 /* API VS */
94 #define SI_ES_NUM_USER_SGPR 12 /* API VS */
95 #define SI_LS_NUM_USER_SGPR 13 /* API VS */
96 #define SI_TCS_NUM_USER_SGPR 11
97 #define SI_TES_NUM_USER_SGPR 10
98 #define SI_GS_NUM_USER_SGPR 8
99 #define SI_GSCOPY_NUM_USER_SGPR 4
100 #define SI_PS_NUM_USER_SGPR 10
102 /* LLVM function parameter indices */
103 #define SI_PARAM_RW_BUFFERS 0
104 #define SI_PARAM_CONST_BUFFERS 1
105 #define SI_PARAM_SAMPLER_STATES 2
106 #define SI_PARAM_SAMPLER_VIEWS 3
108 /* VS only parameters */
109 #define SI_PARAM_VERTEX_BUFFERS 4
110 #define SI_PARAM_BASE_VERTEX 5
111 #define SI_PARAM_START_INSTANCE 6
112 /* [0] = clamp vertex color */
113 #define SI_PARAM_VS_STATE_BITS 7
114 /* the other VS parameters are assigned dynamically */
116 /* Offsets where TCS outputs and TCS patch outputs live in LDS:
117 * [0:15] = TCS output patch0 offset / 16, max = NUM_PATCHES * 32 * 32
118 * [16:31] = TCS output patch0 offset for per-patch / 16, max = NUM_PATCHES*32*32* + 32*32
120 #define SI_PARAM_TCS_OUT_OFFSETS 4 /* for TCS & TES */
122 /* Layout of TCS outputs / TES inputs:
123 * [0:12] = stride between output patches in dwords, num_outputs * num_vertices * 4, max = 32*32*4
124 * [13:20] = stride between output vertices in dwords = num_inputs * 4, max = 32*4
125 * [26:31] = gl_PatchVerticesIn, max = 32
127 #define SI_PARAM_TCS_OUT_LAYOUT 5 /* for TCS & TES */
129 /* Layout of LS outputs / TCS inputs
130 * [0:12] = stride between patches in dwords = num_inputs * num_vertices * 4, max = 32*32*4
131 * [13:20] = stride between vertices in dwords = num_inputs * 4, max = 32*4
133 #define SI_PARAM_TCS_IN_LAYOUT 6 /* TCS only */
134 #define SI_PARAM_LS_OUT_LAYOUT 7 /* same value as TCS_IN_LAYOUT, LS only */
136 /* TCS only parameters. */
137 #define SI_PARAM_TESS_FACTOR_OFFSET 7
138 #define SI_PARAM_PATCH_ID 8
139 #define SI_PARAM_REL_IDS 9
141 /* GS only parameters */
142 #define SI_PARAM_GS2VS_OFFSET 4
143 #define SI_PARAM_GS_WAVE_ID 5
144 #define SI_PARAM_VTX0_OFFSET 6
145 #define SI_PARAM_VTX1_OFFSET 7
146 #define SI_PARAM_PRIMITIVE_ID 8
147 #define SI_PARAM_VTX2_OFFSET 9
148 #define SI_PARAM_VTX3_OFFSET 10
149 #define SI_PARAM_VTX4_OFFSET 11
150 #define SI_PARAM_VTX5_OFFSET 12
151 #define SI_PARAM_GS_INSTANCE_ID 13
153 /* PS only parameters */
154 #define SI_PARAM_ALPHA_REF 4
156 * 0: force_persample_interp
158 #define SI_PARAM_PS_STATE_BITS 5
159 #define SI_PARAM_PRIM_MASK 6
160 #define SI_PARAM_PERSP_SAMPLE 7
161 #define SI_PARAM_PERSP_CENTER 8
162 #define SI_PARAM_PERSP_CENTROID 9
163 #define SI_PARAM_PERSP_PULL_MODEL 10
164 #define SI_PARAM_LINEAR_SAMPLE 11
165 #define SI_PARAM_LINEAR_CENTER 12
166 #define SI_PARAM_LINEAR_CENTROID 13
167 #define SI_PARAM_LINE_STIPPLE_TEX 14
168 #define SI_PARAM_POS_X_FLOAT 15
169 #define SI_PARAM_POS_Y_FLOAT 16
170 #define SI_PARAM_POS_Z_FLOAT 17
171 #define SI_PARAM_POS_W_FLOAT 18
172 #define SI_PARAM_FRONT_FACE 19
173 #define SI_PARAM_ANCILLARY 20
174 #define SI_PARAM_SAMPLE_COVERAGE 21
175 #define SI_PARAM_POS_FIXED_PT 22
177 #define SI_NUM_PARAMS (SI_PARAM_POS_FIXED_PT + 1)
181 /* A shader selector is a gallium CSO and contains shader variants and
182 * binaries for one TGSI program. This can be shared by multiple contexts.
184 struct si_shader_selector
{
186 struct si_shader
*first_variant
; /* immutable after the first variant */
187 struct si_shader
*last_variant
; /* mutable */
189 struct tgsi_token
*tokens
;
190 struct pipe_stream_output_info so
;
191 struct tgsi_shader_info info
;
193 /* PIPE_SHADER_[VERTEX|FRAGMENT|...] */
196 /* Whether the shader has to use a conditional assignment to
197 * choose between weights when emulating
198 * pipe_rasterizer_state::force_persample_interp.
199 * If false, "si_emit_spi_ps_input" will take care of it instead.
201 bool forces_persample_interp_for_persp
;
202 bool forces_persample_interp_for_linear
;
204 unsigned esgs_itemsize
;
205 unsigned gs_input_verts_per_prim
;
206 unsigned gs_output_prim
;
207 unsigned gs_max_out_vertices
;
208 unsigned gs_num_invocations
;
209 unsigned max_gs_stream
; /* count - 1 */
210 unsigned gsvs_vertex_size
;
211 unsigned max_gsvs_emit_size
;
213 /* masks of "get_unique_index" bits */
214 uint64_t outputs_written
;
215 uint32_t patch_outputs_written
;
218 /* Valid shader configurations:
220 * API shaders VS | TCS | TES | GS |pass| PS
221 * are compiled as: | | | |thru|
223 * Only VS & PS: VS | -- | -- | -- | -- | PS
224 * With GS: ES | -- | -- | GS | VS | PS
225 * With Tessel.: LS | HS | VS | -- | -- | PS
226 * With both: LS | HS | ES | GS | VS | PS
229 union si_shader_key
{
231 unsigned export_16bpc
:8;
232 unsigned last_cbuf
:3;
233 unsigned color_two_side
:1;
234 unsigned alpha_func
:3;
235 unsigned alpha_to_one
:1;
236 unsigned poly_stipple
:1;
237 unsigned poly_line_smoothing
:1;
238 unsigned clamp_color
:1;
241 unsigned instance_divisors
[SI_NUM_VERTEX_BUFFERS
];
242 /* Mask of "get_unique_index" bits - which outputs are read
243 * by the next stage (needed by ES).
244 * This describes how outputs are laid out in memory. */
245 unsigned as_es
:1; /* export shader */
246 unsigned as_ls
:1; /* local shader */
247 unsigned export_prim_id
:1; /* when PS needs it and GS is disabled */
250 unsigned prim_mode
:3;
251 } tcs
; /* tessellation control shader */
253 /* Mask of "get_unique_index" bits - which outputs are read
254 * by the next stage (needed by ES).
255 * This describes how outputs are laid out in memory. */
256 unsigned as_es
:1; /* export shader */
257 unsigned export_prim_id
:1; /* when PS needs it and GS is disabled */
258 } tes
; /* tessellation evaluation shader */
262 struct si_shader_selector
*selector
;
263 struct si_shader
*next_variant
;
265 struct si_shader
*gs_copy_shader
;
266 struct si_pm4_state
*pm4
;
267 struct r600_resource
*bo
;
268 struct r600_resource
*scratch_bo
;
269 struct radeon_shader_binary binary
;
273 unsigned spi_ps_input_ena
;
275 unsigned scratch_bytes_per_wave
;
276 unsigned spi_shader_col_format
;
277 unsigned spi_shader_z_format
;
278 unsigned db_shader_control
;
279 unsigned cb_shader_mask
;
280 union si_shader_key key
;
283 unsigned vs_output_param_offset
[PIPE_MAX_SHADER_OUTPUTS
];
284 unsigned ps_input_param_offset
[PIPE_MAX_SHADER_INPUTS
];
285 unsigned ps_input_interpolate
[PIPE_MAX_SHADER_INPUTS
];
286 bool uses_instanceid
;
287 unsigned nr_pos_exports
;
288 unsigned nr_param_exports
;
289 bool is_gs_copy_shader
;
290 bool dx10_clamp_mode
; /* convert NaNs to 0 */
296 static inline struct tgsi_shader_info
*si_get_vs_info(struct si_context
*sctx
)
298 if (sctx
->gs_shader
.cso
)
299 return &sctx
->gs_shader
.cso
->info
;
300 else if (sctx
->tes_shader
.cso
)
301 return &sctx
->tes_shader
.cso
->info
;
302 else if (sctx
->vs_shader
.cso
)
303 return &sctx
->vs_shader
.cso
->info
;
308 static inline struct si_shader
* si_get_vs_state(struct si_context
*sctx
)
310 if (sctx
->gs_shader
.current
)
311 return sctx
->gs_shader
.current
->gs_copy_shader
;
312 else if (sctx
->tes_shader
.current
)
313 return sctx
->tes_shader
.current
;
315 return sctx
->vs_shader
.current
;
318 static inline bool si_vs_exports_prim_id(struct si_shader
*shader
)
320 if (shader
->selector
->type
== PIPE_SHADER_VERTEX
)
321 return shader
->key
.vs
.export_prim_id
;
322 else if (shader
->selector
->type
== PIPE_SHADER_TESS_EVAL
)
323 return shader
->key
.tes
.export_prim_id
;
328 /* radeonsi_shader.c */
329 int si_shader_create(struct si_screen
*sscreen
, LLVMTargetMachineRef tm
,
330 struct si_shader
*shader
,
331 struct pipe_debug_callback
*debug
);
332 void si_dump_shader_key(unsigned shader
, union si_shader_key
*key
, FILE *f
);
333 int si_compile_llvm(struct si_screen
*sscreen
, struct si_shader
*shader
,
334 LLVMTargetMachineRef tm
, LLVMModuleRef mod
,
335 struct pipe_debug_callback
*debug
, unsigned processor
);
336 void si_shader_destroy(struct si_shader
*shader
);
337 unsigned si_shader_io_get_unique_index(unsigned semantic_name
, unsigned index
);
338 int si_shader_binary_upload(struct si_screen
*sscreen
, struct si_shader
*shader
);
339 void si_shader_binary_read(struct si_screen
*sscreen
, struct si_shader
*shader
,
340 struct pipe_debug_callback
*debug
, unsigned processor
);
341 void si_shader_apply_scratch_relocs(struct si_context
*sctx
,
342 struct si_shader
*shader
,
343 uint64_t scratch_va
);
344 void si_shader_binary_read_config(const struct si_screen
*sscreen
,
345 struct si_shader
*shader
,
346 unsigned symbol_offset
);