2 * Copyright © 2010 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
21 * DEALINGS IN THE SOFTWARE.
25 * \file ir_set_program_inouts.cpp
27 * Sets the InputsRead and OutputsWritten of Mesa programs.
29 * Additionally, for fragment shaders, sets the InterpQualifier array, and the
30 * IsCentroid and IsSample bitfields.
32 * Mesa programs (gl_program, not gl_shader_program) have a set of
33 * flags indicating which varyings are read and written. Computing
34 * which are actually read from some sort of backend code can be
35 * tricky when variable array indexing involved. So this pass
36 * provides support for setting InputsRead and OutputsWritten right
40 #include "main/core.h" /* for struct gl_program */
42 #include "ir_visitor.h"
43 #include "compiler/glsl_types.h"
47 class ir_set_program_inouts_visitor
: public ir_hierarchical_visitor
{
49 ir_set_program_inouts_visitor(struct gl_program
*prog
,
50 gl_shader_stage shader_stage
)
53 this->shader_stage
= shader_stage
;
55 ~ir_set_program_inouts_visitor()
59 virtual ir_visitor_status
visit_enter(ir_dereference_array
*);
60 virtual ir_visitor_status
visit_enter(ir_function_signature
*);
61 virtual ir_visitor_status
visit_enter(ir_discard
*);
62 virtual ir_visitor_status
visit_enter(ir_texture
*);
63 virtual ir_visitor_status
visit(ir_dereference_variable
*);
66 void mark_whole_variable(ir_variable
*var
);
67 bool try_mark_partial_variable(ir_variable
*var
, ir_rvalue
*index
);
69 struct gl_program
*prog
;
70 gl_shader_stage shader_stage
;
73 } /* anonymous namespace */
76 is_shader_inout(ir_variable
*var
)
78 return var
->data
.mode
== ir_var_shader_in
||
79 var
->data
.mode
== ir_var_shader_out
||
80 var
->data
.mode
== ir_var_system_value
;
84 mark(struct gl_program
*prog
, ir_variable
*var
, int offset
, int len
,
85 gl_shader_stage stage
)
87 /* As of GLSL 1.20, varyings can only be floats, floating-point
88 * vectors or matrices, or arrays of them. For Mesa programs using
89 * InputsRead/OutputsWritten, everything but matrices uses one
90 * slot, while matrices use a slot per column. Presumably
91 * something doing a more clever packing would use something other
92 * than InputsRead/OutputsWritten.
95 for (int i
= 0; i
< len
; i
++) {
96 assert(var
->data
.location
!= -1);
98 int idx
= var
->data
.location
+ offset
+ i
;
99 bool is_patch_generic
= var
->data
.patch
&&
100 idx
!= VARYING_SLOT_TESS_LEVEL_INNER
&&
101 idx
!= VARYING_SLOT_TESS_LEVEL_OUTER
&&
102 idx
!= VARYING_SLOT_BOUNDING_BOX0
&&
103 idx
!= VARYING_SLOT_BOUNDING_BOX1
;
104 GLbitfield64 bitfield
;
106 if (is_patch_generic
) {
107 assert(idx
>= VARYING_SLOT_PATCH0
&& idx
< VARYING_SLOT_TESS_MAX
);
108 bitfield
= BITFIELD64_BIT(idx
- VARYING_SLOT_PATCH0
);
111 assert(idx
< VARYING_SLOT_MAX
);
112 bitfield
= BITFIELD64_BIT(idx
);
115 if (var
->data
.mode
== ir_var_shader_in
) {
116 if (is_patch_generic
)
117 prog
->PatchInputsRead
|= bitfield
;
119 prog
->InputsRead
|= bitfield
;
121 /* double inputs read is only for vertex inputs */
122 if (stage
== MESA_SHADER_VERTEX
&&
123 var
->type
->without_array()->is_dual_slot())
124 prog
->DoubleInputsRead
|= bitfield
;
126 if (stage
== MESA_SHADER_FRAGMENT
) {
127 gl_fragment_program
*fprog
= (gl_fragment_program
*) prog
;
128 fprog
->InterpQualifier
[idx
] =
129 (glsl_interp_mode
) var
->data
.interpolation
;
130 if (var
->data
.centroid
)
131 fprog
->IsCentroid
|= bitfield
;
132 if (var
->data
.sample
)
133 fprog
->IsSample
|= bitfield
;
135 } else if (var
->data
.mode
== ir_var_system_value
) {
136 prog
->SystemValuesRead
|= bitfield
;
138 assert(var
->data
.mode
== ir_var_shader_out
);
139 if (is_patch_generic
) {
140 prog
->PatchOutputsWritten
|= bitfield
;
141 } else if (!var
->data
.read_only
) {
142 prog
->OutputsWritten
|= bitfield
;
143 if (var
->data
.index
> 0)
144 prog
->SecondaryOutputsWritten
|= bitfield
;
147 if (var
->data
.fb_fetch_output
)
148 prog
->OutputsRead
|= bitfield
;
154 * Mark an entire variable as used. Caller must ensure that the variable
155 * represents a shader input or output.
158 ir_set_program_inouts_visitor::mark_whole_variable(ir_variable
*var
)
160 const glsl_type
*type
= var
->type
;
161 bool is_vertex_input
= false;
162 if (this->shader_stage
== MESA_SHADER_GEOMETRY
&&
163 var
->data
.mode
== ir_var_shader_in
&& type
->is_array()) {
164 type
= type
->fields
.array
;
167 if (this->shader_stage
== MESA_SHADER_TESS_CTRL
&&
168 var
->data
.mode
== ir_var_shader_in
) {
169 assert(type
->is_array());
170 type
= type
->fields
.array
;
173 if (this->shader_stage
== MESA_SHADER_TESS_CTRL
&&
174 var
->data
.mode
== ir_var_shader_out
&& !var
->data
.patch
) {
175 assert(type
->is_array());
176 type
= type
->fields
.array
;
179 if (this->shader_stage
== MESA_SHADER_TESS_EVAL
&&
180 var
->data
.mode
== ir_var_shader_in
&& !var
->data
.patch
) {
181 assert(type
->is_array());
182 type
= type
->fields
.array
;
185 if (this->shader_stage
== MESA_SHADER_VERTEX
&&
186 var
->data
.mode
== ir_var_shader_in
)
187 is_vertex_input
= true;
189 mark(this->prog
, var
, 0, type
->count_attribute_slots(is_vertex_input
),
193 /* Default handler: Mark all the locations in the variable as used. */
195 ir_set_program_inouts_visitor::visit(ir_dereference_variable
*ir
)
197 if (!is_shader_inout(ir
->var
))
198 return visit_continue
;
200 mark_whole_variable(ir
->var
);
202 return visit_continue
;
206 * Try to mark a portion of the given variable as used. Caller must ensure
207 * that the variable represents a shader input or output which can be indexed
208 * into in array fashion (an array or matrix). For the purpose of geometry
209 * shader inputs (which are always arrays*), this means that the array element
210 * must be something that can be indexed into in array fashion.
212 * *Except gl_PrimitiveIDIn, as noted below.
214 * For tessellation control shaders all inputs and non-patch outputs are
215 * arrays. For tessellation evaluation shaders non-patch inputs are arrays.
217 * If the index can't be interpreted as a constant, or some other problem
218 * occurs, then nothing will be marked and false will be returned.
221 ir_set_program_inouts_visitor::try_mark_partial_variable(ir_variable
*var
,
224 const glsl_type
*type
= var
->type
;
226 if (this->shader_stage
== MESA_SHADER_GEOMETRY
&&
227 var
->data
.mode
== ir_var_shader_in
) {
228 /* The only geometry shader input that is not an array is
229 * gl_PrimitiveIDIn, and in that case, this code will never be reached,
230 * because gl_PrimitiveIDIn can't be indexed into in array fashion.
232 assert(type
->is_array());
233 type
= type
->fields
.array
;
236 if (this->shader_stage
== MESA_SHADER_TESS_CTRL
&&
237 var
->data
.mode
== ir_var_shader_in
) {
238 assert(type
->is_array());
239 type
= type
->fields
.array
;
242 if (this->shader_stage
== MESA_SHADER_TESS_CTRL
&&
243 var
->data
.mode
== ir_var_shader_out
&& !var
->data
.patch
) {
244 assert(type
->is_array());
245 type
= type
->fields
.array
;
248 if (this->shader_stage
== MESA_SHADER_TESS_EVAL
&&
249 var
->data
.mode
== ir_var_shader_in
&& !var
->data
.patch
) {
250 assert(type
->is_array());
251 type
= type
->fields
.array
;
254 /* TODO: implement proper arrays of arrays support
255 * for now let the caller mark whole variable as used.
257 if (type
->is_array() && type
->fields
.array
->is_array())
260 /* The code below only handles:
262 * - Indexing into matrices
263 * - Indexing into arrays of (matrices, vectors, or scalars)
265 * All other possibilities are either prohibited by GLSL (vertex inputs and
266 * fragment outputs can't be structs) or should have been eliminated by
267 * lowering passes (do_vec_index_to_swizzle() gets rid of indexing into
268 * vectors, and lower_packed_varyings() gets rid of structs that occur in
271 * However, we don't use varying packing in all cases - tessellation
272 * shaders bypass it. This means we'll see varying structs and arrays
273 * of structs here. For now, we just give up so the caller marks the
274 * entire variable as used.
276 if (!(type
->is_matrix() ||
278 (type
->fields
.array
->is_numeric() ||
279 type
->fields
.array
->is_boolean())))) {
281 /* If we don't know how to handle this case, give up and let the
282 * caller mark the whole variable as used.
287 ir_constant
*index_as_constant
= index
->as_constant();
288 if (!index_as_constant
)
293 if (type
->is_array()) {
294 num_elems
= type
->length
;
295 if (type
->fields
.array
->is_matrix())
296 elem_width
= type
->fields
.array
->matrix_columns
;
300 num_elems
= type
->matrix_columns
;
304 if (index_as_constant
->value
.u
[0] >= num_elems
) {
305 /* Constant index outside the bounds of the matrix/array. This could
306 * arise as a result of constant folding of a legal GLSL program.
308 * Even though the spec says that indexing outside the bounds of a
309 * matrix/array results in undefined behaviour, we don't want to pass
310 * out-of-range values to mark() (since this could result in slots that
311 * don't exist being marked as used), so just let the caller mark the
312 * whole variable as used.
317 /* double element width for double types that takes two slots */
318 if (this->shader_stage
!= MESA_SHADER_VERTEX
||
319 var
->data
.mode
!= ir_var_shader_in
) {
320 if (type
->without_array()->is_dual_slot())
324 mark(this->prog
, var
, index_as_constant
->value
.u
[0] * elem_width
,
325 elem_width
, this->shader_stage
);
330 is_multiple_vertices(gl_shader_stage stage
, ir_variable
*var
)
335 if (var
->data
.mode
== ir_var_shader_in
)
336 return stage
== MESA_SHADER_GEOMETRY
||
337 stage
== MESA_SHADER_TESS_CTRL
||
338 stage
== MESA_SHADER_TESS_EVAL
;
339 if (var
->data
.mode
== ir_var_shader_out
)
340 return stage
== MESA_SHADER_TESS_CTRL
;
346 ir_set_program_inouts_visitor::visit_enter(ir_dereference_array
*ir
)
348 /* Note: for geometry shader inputs, lower_named_interface_blocks may
349 * create 2D arrays, so we need to be able to handle those. 2D arrays
350 * shouldn't be able to crop up for any other reason.
352 if (ir_dereference_array
* const inner_array
=
353 ir
->array
->as_dereference_array()) {
355 * inner_array => foo[i]
357 if (ir_dereference_variable
* const deref_var
=
358 inner_array
->array
->as_dereference_variable()) {
359 if (is_multiple_vertices(this->shader_stage
, deref_var
->var
)) {
360 /* foo is a geometry or tessellation shader input, so i is
361 * the vertex, and j the part of the input we're accessing.
363 if (try_mark_partial_variable(deref_var
->var
, ir
->array_index
))
365 /* We've now taken care of foo and j, but i might contain a
366 * subexpression that accesses shader inputs. So manually
367 * visit i and then continue with the parent.
369 inner_array
->array_index
->accept(this);
370 return visit_continue_with_parent
;
374 } else if (ir_dereference_variable
* const deref_var
=
375 ir
->array
->as_dereference_variable()) {
376 /* ir => foo[i], where foo is a variable. */
377 if (is_multiple_vertices(this->shader_stage
, deref_var
->var
)) {
378 /* foo is a geometry or tessellation shader input, so i is
379 * the vertex, and we're accessing the entire input.
381 mark_whole_variable(deref_var
->var
);
382 /* We've now taken care of foo, but i might contain a subexpression
383 * that accesses shader inputs. So manually visit i and then
384 * continue with the parent.
386 ir
->array_index
->accept(this);
387 return visit_continue_with_parent
;
388 } else if (is_shader_inout(deref_var
->var
)) {
389 /* foo is a shader input/output, but not a geometry shader input,
390 * so i is the part of the input we're accessing.
392 if (try_mark_partial_variable(deref_var
->var
, ir
->array_index
))
393 return visit_continue_with_parent
;
397 /* The expression is something we don't recognize. Just visit its
400 return visit_continue
;
404 ir_set_program_inouts_visitor::visit_enter(ir_function_signature
*ir
)
406 /* We don't want to descend into the function parameters and
407 * consider them as shader inputs or outputs.
409 visit_list_elements(this, &ir
->body
);
410 return visit_continue_with_parent
;
414 ir_set_program_inouts_visitor::visit_enter(ir_discard
*)
416 /* discards are only allowed in fragment shaders. */
417 assert(this->shader_stage
== MESA_SHADER_FRAGMENT
);
419 gl_fragment_program
*fprog
= (gl_fragment_program
*) prog
;
420 fprog
->UsesKill
= true;
422 return visit_continue
;
426 ir_set_program_inouts_visitor::visit_enter(ir_texture
*ir
)
428 if (ir
->op
== ir_tg4
)
429 prog
->UsesGather
= true;
430 return visit_continue
;
434 do_set_program_inouts(exec_list
*instructions
, struct gl_program
*prog
,
435 gl_shader_stage shader_stage
)
437 ir_set_program_inouts_visitor
v(prog
, shader_stage
);
439 prog
->InputsRead
= 0;
440 prog
->OutputsWritten
= 0;
441 prog
->SecondaryOutputsWritten
= 0;
442 prog
->OutputsRead
= 0;
443 prog
->PatchInputsRead
= 0;
444 prog
->PatchOutputsWritten
= 0;
445 prog
->SystemValuesRead
= 0;
446 if (shader_stage
== MESA_SHADER_FRAGMENT
) {
447 gl_fragment_program
*fprog
= (gl_fragment_program
*) prog
;
448 memset(fprog
->InterpQualifier
, 0, sizeof(fprog
->InterpQualifier
));
449 fprog
->IsCentroid
= 0;
451 fprog
->UsesKill
= false;
453 visit_list_elements(&v
, instructions
);