2 * Copyright © 2015 Intel Corporation
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 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * 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 NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
24 #include "brw_context.h"
25 #include "brw_state.h"
26 #include "brw_defines.h"
27 #include "brw_program.h"
28 #include "intel_batchbuffer.h"
29 #include "intel_buffer_objects.h"
30 #include "program/prog_parameter.h"
31 #include "main/shaderapi.h"
36 union fi fi
= { .f
= f
};
41 brw_param_value(struct brw_context
*brw
,
42 const struct gl_program
*prog
,
43 const struct brw_stage_state
*stage_state
,
46 struct gl_context
*ctx
= &brw
->ctx
;
48 switch (BRW_PARAM_DOMAIN(param
)) {
49 case BRW_PARAM_DOMAIN_BUILTIN
:
50 if (param
== BRW_PARAM_BUILTIN_ZERO
) {
52 } else if (BRW_PARAM_BUILTIN_IS_CLIP_PLANE(param
)) {
53 gl_clip_plane
*clip_planes
= brw_select_clip_planes(ctx
);
54 unsigned idx
= BRW_PARAM_BUILTIN_CLIP_PLANE_IDX(param
);
55 unsigned comp
= BRW_PARAM_BUILTIN_CLIP_PLANE_COMP(param
);
56 return ((uint32_t *)clip_planes
[idx
])[comp
];
57 } else if (param
>= BRW_PARAM_BUILTIN_TESS_LEVEL_OUTER_X
&&
58 param
<= BRW_PARAM_BUILTIN_TESS_LEVEL_OUTER_W
) {
59 unsigned i
= param
- BRW_PARAM_BUILTIN_TESS_LEVEL_OUTER_X
;
60 return f_as_u32(ctx
->TessCtrlProgram
.patch_default_outer_level
[i
]);
61 } else if (param
== BRW_PARAM_BUILTIN_TESS_LEVEL_INNER_X
) {
62 return f_as_u32(ctx
->TessCtrlProgram
.patch_default_inner_level
[0]);
63 } else if (param
== BRW_PARAM_BUILTIN_TESS_LEVEL_INNER_Y
) {
64 return f_as_u32(ctx
->TessCtrlProgram
.patch_default_inner_level
[1]);
66 unreachable("Invalid param builtin");
69 case BRW_PARAM_DOMAIN_PARAMETER
: {
70 unsigned idx
= BRW_PARAM_PARAMETER_IDX(param
);
71 unsigned comp
= BRW_PARAM_PARAMETER_COMP(param
);
72 assert(idx
< prog
->Parameters
->NumParameters
);
73 return prog
->Parameters
->ParameterValues
[idx
][comp
].u
;
76 case BRW_PARAM_DOMAIN_UNIFORM
: {
77 unsigned idx
= BRW_PARAM_UNIFORM_IDX(param
);
78 assert(idx
< prog
->sh
.data
->NumUniformDataSlots
);
79 return prog
->sh
.data
->UniformDataSlots
[idx
].u
;
82 case BRW_PARAM_DOMAIN_IMAGE
: {
83 unsigned idx
= BRW_PARAM_IMAGE_IDX(param
);
84 unsigned offset
= BRW_PARAM_IMAGE_OFFSET(param
);
85 assert(offset
< ARRAY_SIZE(stage_state
->image_param
));
86 return ((uint32_t *)&stage_state
->image_param
[idx
])[offset
];
90 unreachable("Invalid param domain");
96 brw_populate_constant_data(struct brw_context
*brw
,
97 const struct gl_program
*prog
,
98 const struct brw_stage_state
*stage_state
,
100 const uint32_t *param
,
103 uint32_t *dst
= void_dst
;
104 for (unsigned i
= 0; i
< nr_params
; i
++)
105 dst
[i
] = brw_param_value(brw
, prog
, stage_state
, param
[i
]);
110 * Creates a streamed BO containing the push constants for the VS or GS on
113 * Push constants are constant values (such as GLSL uniforms) that are
114 * pre-loaded into a shader stage's register space at thread spawn time.
116 * Not all GLSL uniforms will be uploaded as push constants: The hardware has
117 * a limitation of 32 or 64 EU registers (256 or 512 floats) per stage to be
118 * uploaded as push constants, while GL 4.4 requires at least 1024 components
119 * to be usable for the VS. Plus, currently we always use pull constants
120 * instead of push constants when doing variable-index array access.
122 * See brw_curbe.c for the equivalent gen4/5 code.
125 gen6_upload_push_constants(struct brw_context
*brw
,
126 const struct gl_program
*prog
,
127 const struct brw_stage_prog_data
*prog_data
,
128 struct brw_stage_state
*stage_state
)
130 const struct gen_device_info
*devinfo
= &brw
->screen
->devinfo
;
131 struct gl_context
*ctx
= &brw
->ctx
;
133 _mesa_shader_write_subroutine_indices(ctx
, stage_state
->stage
);
135 if (prog_data
->nr_params
== 0) {
136 stage_state
->push_const_size
= 0;
138 /* Updates the ParamaterValues[i] pointers for all parameters of the
139 * basic type of PROGRAM_STATE_VAR.
141 /* XXX: Should this happen somewhere before to get our state flag set? */
143 _mesa_load_state_parameters(ctx
, prog
->Parameters
);
146 const int size
= prog_data
->nr_params
* sizeof(gl_constant_value
);
147 gl_constant_value
*param
;
148 if (devinfo
->gen
>= 8 || devinfo
->is_haswell
) {
149 param
= intel_upload_space(brw
, size
, 32,
150 &stage_state
->push_const_bo
,
151 &stage_state
->push_const_offset
);
153 param
= brw_state_batch(brw
, size
, 32,
154 &stage_state
->push_const_offset
);
157 STATIC_ASSERT(sizeof(gl_constant_value
) == sizeof(float));
159 /* _NEW_PROGRAM_CONSTANTS
161 * Also _NEW_TRANSFORM -- we may reference clip planes other than as a
162 * side effect of dereferencing uniforms, so _NEW_PROGRAM_CONSTANTS
163 * wouldn't be set for them.
165 brw_populate_constant_data(brw
, prog
, stage_state
, param
,
167 prog_data
->nr_params
);
170 fprintf(stderr
, "%s constants:\n",
171 _mesa_shader_stage_to_string(stage_state
->stage
));
172 for (i
= 0; i
< prog_data
->nr_params
; i
++) {
174 fprintf(stderr
, "g%d: ",
175 prog_data
->dispatch_grf_start_reg
+ i
/ 8);
176 fprintf(stderr
, "%8f ", param
[i
].f
);
178 fprintf(stderr
, "\n");
181 fprintf(stderr
, "\n");
182 fprintf(stderr
, "\n");
185 stage_state
->push_const_size
= ALIGN(prog_data
->nr_params
, 8) / 8;
186 /* We can only push 32 registers of constants at a time. */
188 /* From the SNB PRM (vol2, part 1, section 3.2.1.4: 3DSTATE_CONSTANT_VS:
190 * "The sum of all four read length fields (each incremented to
191 * represent the actual read length) must be less than or equal to
194 * From the IVB PRM (vol2, part 1, section 3.2.1.3: 3DSTATE_CONSTANT_VS:
196 * "The sum of all four read length fields must be less than or
197 * equal to the size of 64"
199 * The other shader stages all match the VS's limits.
201 assert(stage_state
->push_const_size
<= 32);
204 stage_state
->push_constants_dirty
= true;
209 * Creates a temporary BO containing the pull constant data for the shader
210 * stage, and the SURFACE_STATE struct that points at it.
212 * Pull constants are GLSL uniforms (and other constant data) beyond what we
213 * could fit as push constants, or that have variable-index array access
214 * (which is easiest to support using pull constants, and avoids filling
215 * register space with mostly-unused data).
217 * Compare this path to brw_curbe.c for gen4/5 push constants, and
218 * gen6_vs_state.c for gen6+ push constants.
221 brw_upload_pull_constants(struct brw_context
*brw
,
222 GLbitfield64 brw_new_constbuf
,
223 const struct gl_program
*prog
,
224 struct brw_stage_state
*stage_state
,
225 const struct brw_stage_prog_data
*prog_data
)
228 uint32_t surf_index
= prog_data
->binding_table
.pull_constants_start
;
230 if (!prog_data
->nr_pull_params
) {
231 if (stage_state
->surf_offset
[surf_index
]) {
232 stage_state
->surf_offset
[surf_index
] = 0;
233 brw
->ctx
.NewDriverState
|= brw_new_constbuf
;
238 /* Updates the ParamaterValues[i] pointers for all parameters of the
239 * basic type of PROGRAM_STATE_VAR.
241 _mesa_load_state_parameters(&brw
->ctx
, prog
->Parameters
);
243 /* BRW_NEW_*_PROG_DATA | _NEW_PROGRAM_CONSTANTS */
244 uint32_t size
= prog_data
->nr_pull_params
* 4;
245 struct brw_bo
*const_bo
= NULL
;
246 uint32_t const_offset
;
247 gl_constant_value
*constants
= intel_upload_space(brw
, size
, 64,
248 &const_bo
, &const_offset
);
250 STATIC_ASSERT(sizeof(gl_constant_value
) == sizeof(float));
252 brw_populate_constant_data(brw
, prog
, stage_state
, constants
,
253 prog_data
->pull_param
,
254 prog_data
->nr_pull_params
);
257 for (i
= 0; i
< ALIGN(prog_data
->nr_pull_params
, 4) / 4; i
++) {
258 const gl_constant_value
*row
= &constants
[i
* 4];
259 fprintf(stderr
, "const surface %3d: %4.3f %4.3f %4.3f %4.3f\n",
260 i
, row
[0].f
, row
[1].f
, row
[2].f
, row
[3].f
);
264 brw_create_constant_surface(brw
, const_bo
, const_offset
, size
,
265 &stage_state
->surf_offset
[surf_index
]);
266 brw_bo_unreference(const_bo
);
268 brw
->ctx
.NewDriverState
|= brw_new_constbuf
;
272 * Creates a region containing the push constants for the CS on gen7+.
274 * Push constants are constant values (such as GLSL uniforms) that are
275 * pre-loaded into a shader stage's register space at thread spawn time.
277 * For other stages, see brw_curbe.c:brw_upload_constant_buffer for the
278 * equivalent gen4/5 code and gen6_vs_state.c:gen6_upload_push_constants for
282 brw_upload_cs_push_constants(struct brw_context
*brw
,
283 const struct gl_program
*prog
,
284 const struct brw_cs_prog_data
*cs_prog_data
,
285 struct brw_stage_state
*stage_state
)
287 struct gl_context
*ctx
= &brw
->ctx
;
288 const struct brw_stage_prog_data
*prog_data
=
289 (struct brw_stage_prog_data
*) cs_prog_data
;
291 /* Updates the ParamaterValues[i] pointers for all parameters of the
292 * basic type of PROGRAM_STATE_VAR.
294 /* XXX: Should this happen somewhere before to get our state flag set? */
295 _mesa_load_state_parameters(ctx
, prog
->Parameters
);
297 if (cs_prog_data
->push
.total
.size
== 0) {
298 stage_state
->push_const_size
= 0;
304 brw_state_batch(brw
, ALIGN(cs_prog_data
->push
.total
.size
, 64),
305 64, &stage_state
->push_const_offset
);
308 STATIC_ASSERT(sizeof(gl_constant_value
) == sizeof(float));
310 if (cs_prog_data
->push
.cross_thread
.size
> 0) {
311 uint32_t *param_copy
= param
;
313 i
< cs_prog_data
->push
.cross_thread
.dwords
;
315 assert(prog_data
->param
[i
] != BRW_PARAM_BUILTIN_THREAD_LOCAL_ID
);
316 param_copy
[i
] = brw_param_value(brw
, prog
, stage_state
,
317 prog_data
->param
[i
]);
321 if (cs_prog_data
->push
.per_thread
.size
> 0) {
322 for (unsigned t
= 0; t
< cs_prog_data
->threads
; t
++) {
324 8 * (cs_prog_data
->push
.per_thread
.regs
* t
+
325 cs_prog_data
->push
.cross_thread
.regs
);
326 unsigned src
= cs_prog_data
->push
.cross_thread
.dwords
;
327 for ( ; src
< prog_data
->nr_params
; src
++, dst
++) {
328 if (prog_data
->param
[src
] == BRW_PARAM_BUILTIN_THREAD_LOCAL_ID
) {
329 param
[dst
] = t
* cs_prog_data
->simd_size
;
331 param
[dst
] = brw_param_value(brw
, prog
, stage_state
,
332 prog_data
->param
[src
]);
338 stage_state
->push_const_size
=
339 cs_prog_data
->push
.cross_thread
.regs
+
340 cs_prog_data
->push
.per_thread
.regs
;