2 * Copyright © 2015 Red Hat
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 FROM,
20 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
26 #include "pipe/p_defines.h"
27 #include "pipe/p_screen.h"
28 #include "pipe/p_context.h"
30 #include "program/program.h"
31 #include "program/prog_statevars.h"
32 #include "program/prog_parameter.h"
33 #include "program/ir_to_mesa.h"
34 #include "program/hash_table.h"
35 #include "main/mtypes.h"
36 #include "main/errors.h"
37 #include "main/shaderapi.h"
38 #include "main/uniforms.h"
40 #include "st_context.h"
41 #include "st_program.h"
42 #include "st_glsl_types.h"
44 #include "compiler/nir/nir.h"
45 #include "compiler/glsl_types.h"
46 #include "compiler/glsl/glsl_to_nir.h"
47 #include "compiler/glsl/ir.h"
50 /* Depending on PIPE_CAP_TGSI_TEXCOORD (st->needs_texcoord_semantic) we
51 * may need to fix up varying slots so the glsl->nir path is aligned
52 * with the anything->tgsi->nir path.
55 st_nir_fixup_varying_slots(struct st_context
*st
, struct exec_list
*var_list
)
57 if (st
->needs_texcoord_semantic
)
60 nir_foreach_variable(var
, var_list
) {
61 if (var
->data
.location
>= VARYING_SLOT_VAR0
) {
62 var
->data
.location
+= 9;
63 } else if ((var
->data
.location
>= VARYING_SLOT_TEX0
) &&
64 (var
->data
.location
<= VARYING_SLOT_TEX7
)) {
65 var
->data
.location
+= VARYING_SLOT_VAR0
- VARYING_SLOT_TEX0
;
70 /* input location assignment for VS inputs must be handled specially, so
71 * that it is aligned w/ st's vbo state.
72 * (This isn't the case with, for ex, FS inputs, which only need to agree
73 * on varying-slot w/ the VS outputs)
76 st_nir_assign_vs_in_locations(struct gl_program
*prog
,
77 struct exec_list
*var_list
, unsigned *size
)
79 unsigned attr
, num_inputs
= 0;
80 unsigned input_to_index
[VERT_ATTRIB_MAX
] = {0};
82 /* TODO de-duplicate w/ similar code in st_translate_vertex_program()? */
83 for (attr
= 0; attr
< VERT_ATTRIB_MAX
; attr
++) {
84 if ((prog
->InputsRead
& BITFIELD64_BIT(attr
)) != 0) {
85 input_to_index
[attr
] = num_inputs
;
87 if ((prog
->DoubleInputsRead
& BITFIELD64_BIT(attr
)) != 0) {
88 /* add placeholder for second part of a double attribute */
95 nir_foreach_variable(var
, var_list
) {
96 attr
= var
->data
.location
;
97 assert(attr
< ARRAY_SIZE(input_to_index
));
98 var
->data
.driver_location
= input_to_index
[attr
];
104 st_nir_lookup_parameter_index(const struct gl_program_parameter_list
*params
,
107 int loc
= _mesa_lookup_parameter_index(params
, name
);
109 /* is there a better way to do this? If we have something like:
117 * Then what we get in prog->Parameters looks like:
119 * 0: Name=color.f, Type=6, DataType=1406, Size=1
120 * 1: Name=color.v, Type=6, DataType=8b52, Size=4
122 * So the name doesn't match up and _mesa_lookup_parameter_index()
123 * fails. In this case just find the first matching "color.*"..
125 * Note for arrays you could end up w/ color[n].f, for example.
127 * glsl_to_tgsi works slightly differently in this regard. It is
128 * emitting something more low level, so it just translates the
129 * params list 1:1 to CONST[] regs. Going from GLSL IR to TGSI,
130 * it just calculates the additional offset of struct field members
131 * in glsl_to_tgsi_visitor::visit(ir_dereference_record *ir) or
132 * glsl_to_tgsi_visitor::visit(ir_dereference_array *ir). It never
133 * needs to work backwards to get base var loc from the param-list
134 * which already has them separated out.
137 int namelen
= strlen(name
);
138 for (unsigned i
= 0; i
< params
->NumParameters
; i
++) {
139 struct gl_program_parameter
*p
= ¶ms
->Parameters
[i
];
140 if ((strncmp(p
->Name
, name
, namelen
) == 0) &&
141 ((p
->Name
[namelen
] == '.') || (p
->Name
[namelen
] == '['))) {
152 st_nir_assign_uniform_locations(struct gl_program
*prog
,
153 struct gl_shader_program
*shader_program
,
154 struct exec_list
*uniform_list
, unsigned *size
)
159 nir_foreach_variable(uniform
, uniform_list
) {
163 * UBO's have their own address spaces, so don't count them towards the
164 * number of global uniforms
166 if ((uniform
->data
.mode
== nir_var_uniform
|| uniform
->data
.mode
== nir_var_shader_storage
) &&
167 uniform
->interface_type
!= NULL
)
170 if (uniform
->type
->is_sampler()) {
172 bool found
= shader_program
->UniformHash
->get(val
, uniform
->name
);
175 /* this ensure that nir_lower_samplers looks at the correct
176 * shader_program->UniformStorage[location]:
178 uniform
->data
.location
= val
;
179 } else if (strncmp(uniform
->name
, "gl_", 3) == 0) {
180 const gl_state_index
*const stateTokens
= (gl_state_index
*)uniform
->state_slots
[0].tokens
;
181 /* This state reference has already been setup by ir_to_mesa, but we'll
182 * get the same index back here.
184 loc
= _mesa_add_state_reference(prog
->Parameters
, stateTokens
);
186 loc
= st_nir_lookup_parameter_index(prog
->Parameters
, uniform
->name
);
189 uniform
->data
.driver_location
= loc
;
191 max
= MAX2(max
, loc
+ st_glsl_type_size(uniform
->type
));
198 /* First half of converting glsl_to_nir.. this leaves things in a pre-
199 * nir_lower_io state, so that shader variants can more easily insert/
200 * replace variables, etc.
203 st_glsl_to_nir(struct st_context
*st
, struct gl_program
*prog
,
204 struct gl_shader_program
*shader_program
,
205 gl_shader_stage stage
)
207 struct pipe_screen
*pscreen
= st
->pipe
->screen
;
208 unsigned ptarget
= st_shader_stage_to_ptarget(stage
);
209 const nir_shader_compiler_options
*options
;
212 assert(pscreen
->get_compiler_options
); /* drivers using NIR must implement this */
214 options
= (const nir_shader_compiler_options
*)
215 pscreen
->get_compiler_options(pscreen
, PIPE_SHADER_IR_NIR
, ptarget
);
221 nir
= glsl_to_nir(shader_program
, stage
, options
);
224 NIR_PASS_V(nir
, nir_lower_io_to_temporaries
,
225 nir_shader_get_entrypoint(nir
),
227 NIR_PASS_V(nir
, nir_lower_global_vars_to_local
);
228 NIR_PASS_V(nir
, nir_split_var_copies
);
229 NIR_PASS_V(nir
, nir_lower_var_copies
);
230 NIR_PASS_V(nir
, st_nir_lower_builtin
);
232 /* fragment shaders may need : */
233 if (stage
== MESA_SHADER_FRAGMENT
) {
234 static const gl_state_index wposTransformState
[STATE_LENGTH
] = {
235 STATE_INTERNAL
, STATE_FB_WPOS_Y_TRANSFORM
237 nir_lower_wpos_ytransform_options wpos_options
= {0};
238 struct pipe_screen
*pscreen
= st
->pipe
->screen
;
240 memcpy(wpos_options
.state_tokens
, wposTransformState
,
241 sizeof(wpos_options
.state_tokens
));
242 wpos_options
.fs_coord_origin_upper_left
=
243 pscreen
->get_param(pscreen
, PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT
);
244 wpos_options
.fs_coord_origin_lower_left
=
245 pscreen
->get_param(pscreen
, PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT
);
246 wpos_options
.fs_coord_pixel_center_integer
=
247 pscreen
->get_param(pscreen
, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER
);
248 wpos_options
.fs_coord_pixel_center_half_integer
=
249 pscreen
->get_param(pscreen
, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER
);
251 if (nir_lower_wpos_ytransform(nir
, &wpos_options
)) {
252 nir_validate_shader(nir
);
253 _mesa_add_state_reference(prog
->Parameters
, wposTransformState
);
257 if (st
->ctx
->_Shader
->Flags
& GLSL_DUMP
) {
259 _mesa_log("NIR IR for linked %s program %d:\n",
260 _mesa_shader_stage_to_string(stage
),
261 shader_program
->Name
);
262 nir_print_shader(nir
, _mesa_get_log_file());
269 /* TODO any better helper somewhere to sort a list? */
272 insert_sorted(struct exec_list
*var_list
, nir_variable
*new_var
)
274 nir_foreach_variable(var
, var_list
) {
275 if (var
->data
.location
> new_var
->data
.location
) {
276 exec_node_insert_node_before(&var
->node
, &new_var
->node
);
280 exec_list_push_tail(var_list
, &new_var
->node
);
284 sort_varyings(struct exec_list
*var_list
)
286 struct exec_list new_list
;
287 exec_list_make_empty(&new_list
);
288 nir_foreach_variable_safe(var
, var_list
) {
289 exec_node_remove(&var
->node
);
290 insert_sorted(&new_list
, var
);
292 exec_list_move_nodes_to(&new_list
, var_list
);
295 /* Second half of preparing nir from glsl, which happens after shader
299 st_finalize_nir(struct st_context
*st
, struct gl_program
*prog
, nir_shader
*nir
)
301 NIR_PASS_V(nir
, nir_split_var_copies
);
302 NIR_PASS_V(nir
, nir_lower_var_copies
);
303 NIR_PASS_V(nir
, nir_lower_io_types
);
305 if (nir
->stage
== MESA_SHADER_VERTEX
) {
306 /* Needs special handling so drvloc matches the vbo state: */
307 st_nir_assign_vs_in_locations(prog
, &nir
->inputs
, &nir
->num_inputs
);
308 sort_varyings(&nir
->outputs
);
309 nir_assign_var_locations(&nir
->outputs
,
312 st_nir_fixup_varying_slots(st
, &nir
->outputs
);
313 } else if (nir
->stage
== MESA_SHADER_FRAGMENT
) {
314 sort_varyings(&nir
->inputs
);
315 nir_assign_var_locations(&nir
->inputs
,
318 st_nir_fixup_varying_slots(st
, &nir
->inputs
);
319 nir_assign_var_locations(&nir
->outputs
,
323 unreachable("invalid shader type for tgsi bypass\n");
326 struct gl_shader_program
*shader_program
;
327 switch (nir
->stage
) {
328 case MESA_SHADER_VERTEX
:
329 shader_program
= ((struct st_vertex_program
*)prog
)->shader_program
;
331 case MESA_SHADER_FRAGMENT
:
332 shader_program
= ((struct st_fragment_program
*)prog
)->shader_program
;
335 assert(!"should not be reached");
339 st_nir_assign_uniform_locations(prog
, shader_program
,
340 &nir
->uniforms
, &nir
->num_uniforms
);
342 NIR_PASS_V(nir
, nir_lower_system_values
);
343 NIR_PASS_V(nir
, nir_lower_io
, nir_var_all
, st_glsl_type_size
);
344 NIR_PASS_V(nir
, nir_lower_samplers
, shader_program
);
348 st_nir_get_mesa_program(struct gl_context
*ctx
,
349 struct gl_shader_program
*shader_program
,
350 struct gl_shader
*shader
)
352 struct gl_program
*prog
;
353 GLenum target
= _mesa_shader_stage_to_program(shader
->Stage
);
355 validate_ir_tree(shader
->ir
);
357 prog
= ctx
->Driver
.NewProgram(ctx
, target
, shader_program
->Name
);
361 prog
->Parameters
= _mesa_new_parameter_list();
363 _mesa_copy_linked_program_data(shader
->Stage
, shader_program
, prog
);
364 _mesa_generate_parameters_list_for_uniforms(shader_program
, shader
,
367 /* Make a pass over the IR to add state references for any built-in
368 * uniforms that are used. This has to be done now (during linking).
369 * Code generation doesn't happen until the first time this shader is
370 * used for rendering. Waiting until then to generate the parameters is
371 * too late. At that point, the values for the built-in uniforms won't
372 * get sent to the shader.
374 foreach_in_list(ir_instruction
, node
, shader
->ir
) {
375 ir_variable
*var
= node
->as_variable();
377 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_uniform
) ||
378 (strncmp(var
->name
, "gl_", 3) != 0))
381 const ir_state_slot
*const slots
= var
->get_state_slots();
382 assert(slots
!= NULL
);
384 for (unsigned int i
= 0; i
< var
->get_num_state_slots(); i
++) {
385 _mesa_add_state_reference(prog
->Parameters
,
386 (gl_state_index
*) slots
[i
].tokens
);
390 if (ctx
->_Shader
->Flags
& GLSL_DUMP
) {
392 _mesa_log("GLSL IR for linked %s program %d:\n",
393 _mesa_shader_stage_to_string(shader
->Stage
),
394 shader_program
->Name
);
395 _mesa_print_ir(_mesa_get_log_file(), shader
->ir
, NULL
);
399 prog
->Instructions
= NULL
;
400 prog
->NumInstructions
= 0;
402 do_set_program_inouts(shader
->ir
, prog
, shader
->Stage
);
404 prog
->SamplersUsed
= shader
->active_samplers
;
405 prog
->ShadowSamplers
= shader
->shadow_samplers
;
406 _mesa_update_shader_textures_used(shader_program
, prog
);
408 _mesa_reference_program(ctx
, &shader
->Program
, prog
);
410 /* Avoid reallocation of the program parameter list, because the uniform
411 * storage is only associated with the original parameter list.
412 * This should be enough for Bitmap and DrawPixels constants.
414 _mesa_reserve_parameter_storage(prog
->Parameters
, 8);
416 /* This has to be done last. Any operation the can cause
417 * prog->ParameterValues to get reallocated (e.g., anything that adds a
418 * program constant) has to happen before creating this linkage.
420 _mesa_associate_uniform_storage(ctx
, shader_program
, prog
->Parameters
);
422 struct st_vertex_program
*stvp
;
423 struct st_fragment_program
*stfp
;
425 switch (shader
->Stage
) {
426 case MESA_SHADER_VERTEX
:
427 stvp
= (struct st_vertex_program
*)prog
;
428 stvp
->shader_program
= shader_program
;
430 case MESA_SHADER_FRAGMENT
:
431 stfp
= (struct st_fragment_program
*)prog
;
432 stfp
->shader_program
= shader_program
;
435 assert(!"should not be reached");