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, the
30 * IsCentroid and IsSample bitfields, and the UsesDFdy flag.
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_expression
*);
62 virtual ir_visitor_status
visit_enter(ir_discard
*);
63 virtual ir_visitor_status
visit_enter(ir_texture
*);
64 virtual ir_visitor_status
visit(ir_dereference_variable
*);
67 void mark_whole_variable(ir_variable
*var
);
68 bool try_mark_partial_variable(ir_variable
*var
, ir_rvalue
*index
);
70 struct gl_program
*prog
;
71 gl_shader_stage shader_stage
;
74 } /* anonymous namespace */
77 is_shader_inout(ir_variable
*var
)
79 return var
->data
.mode
== ir_var_shader_in
||
80 var
->data
.mode
== ir_var_shader_out
||
81 var
->data
.mode
== ir_var_system_value
;
85 mark(struct gl_program
*prog
, ir_variable
*var
, int offset
, int len
,
86 gl_shader_stage stage
)
88 /* As of GLSL 1.20, varyings can only be floats, floating-point
89 * vectors or matrices, or arrays of them. For Mesa programs using
90 * InputsRead/OutputsWritten, everything but matrices uses one
91 * slot, while matrices use a slot per column. Presumably
92 * something doing a more clever packing would use something other
93 * than InputsRead/OutputsWritten.
96 for (int i
= 0; i
< len
; i
++) {
97 assert(var
->data
.location
!= -1);
99 int idx
= var
->data
.location
+ var
->data
.index
+ offset
+ i
;
100 bool is_patch_generic
= var
->data
.patch
&&
101 idx
!= VARYING_SLOT_TESS_LEVEL_INNER
&&
102 idx
!= VARYING_SLOT_TESS_LEVEL_OUTER
;
103 GLbitfield64 bitfield
;
105 if (is_patch_generic
) {
106 assert(idx
>= VARYING_SLOT_PATCH0
&& idx
< VARYING_SLOT_TESS_MAX
);
107 bitfield
= BITFIELD64_BIT(idx
- VARYING_SLOT_PATCH0
);
110 assert(idx
< VARYING_SLOT_MAX
);
111 bitfield
= BITFIELD64_BIT(idx
);
114 if (var
->data
.mode
== ir_var_shader_in
) {
115 if (is_patch_generic
)
116 prog
->PatchInputsRead
|= bitfield
;
118 prog
->InputsRead
|= bitfield
;
120 /* double inputs read is only for vertex inputs */
121 if (stage
== MESA_SHADER_VERTEX
&&
122 var
->type
->without_array()->is_dual_slot())
123 prog
->DoubleInputsRead
|= bitfield
;
125 if (stage
== MESA_SHADER_FRAGMENT
) {
126 gl_fragment_program
*fprog
= (gl_fragment_program
*) prog
;
127 fprog
->InterpQualifier
[idx
] =
128 (glsl_interp_mode
) var
->data
.interpolation
;
129 if (var
->data
.centroid
)
130 fprog
->IsCentroid
|= bitfield
;
131 if (var
->data
.sample
)
132 fprog
->IsSample
|= bitfield
;
134 } else if (var
->data
.mode
== ir_var_system_value
) {
135 prog
->SystemValuesRead
|= bitfield
;
137 assert(var
->data
.mode
== ir_var_shader_out
);
138 if (is_patch_generic
)
139 prog
->PatchOutputsWritten
|= bitfield
;
140 else if (!var
->data
.read_only
)
141 prog
->OutputsWritten
|= bitfield
;
142 if (var
->data
.fb_fetch_output
)
143 prog
->OutputsRead
|= bitfield
;
149 * Mark an entire variable as used. Caller must ensure that the variable
150 * represents a shader input or output.
153 ir_set_program_inouts_visitor::mark_whole_variable(ir_variable
*var
)
155 const glsl_type
*type
= var
->type
;
156 bool is_vertex_input
= false;
157 if (this->shader_stage
== MESA_SHADER_GEOMETRY
&&
158 var
->data
.mode
== ir_var_shader_in
&& type
->is_array()) {
159 type
= type
->fields
.array
;
162 if (this->shader_stage
== MESA_SHADER_TESS_CTRL
&&
163 var
->data
.mode
== ir_var_shader_in
) {
164 assert(type
->is_array());
165 type
= type
->fields
.array
;
168 if (this->shader_stage
== MESA_SHADER_TESS_CTRL
&&
169 var
->data
.mode
== ir_var_shader_out
&& !var
->data
.patch
) {
170 assert(type
->is_array());
171 type
= type
->fields
.array
;
174 if (this->shader_stage
== MESA_SHADER_TESS_EVAL
&&
175 var
->data
.mode
== ir_var_shader_in
&& !var
->data
.patch
) {
176 assert(type
->is_array());
177 type
= type
->fields
.array
;
180 if (this->shader_stage
== MESA_SHADER_VERTEX
&&
181 var
->data
.mode
== ir_var_shader_in
)
182 is_vertex_input
= true;
184 mark(this->prog
, var
, 0, type
->count_attribute_slots(is_vertex_input
),
188 /* Default handler: Mark all the locations in the variable as used. */
190 ir_set_program_inouts_visitor::visit(ir_dereference_variable
*ir
)
192 if (!is_shader_inout(ir
->var
))
193 return visit_continue
;
195 mark_whole_variable(ir
->var
);
197 return visit_continue
;
201 * Try to mark a portion of the given variable as used. Caller must ensure
202 * that the variable represents a shader input or output which can be indexed
203 * into in array fashion (an array or matrix). For the purpose of geometry
204 * shader inputs (which are always arrays*), this means that the array element
205 * must be something that can be indexed into in array fashion.
207 * *Except gl_PrimitiveIDIn, as noted below.
209 * For tessellation control shaders all inputs and non-patch outputs are
210 * arrays. For tessellation evaluation shaders non-patch inputs are arrays.
212 * If the index can't be interpreted as a constant, or some other problem
213 * occurs, then nothing will be marked and false will be returned.
216 ir_set_program_inouts_visitor::try_mark_partial_variable(ir_variable
*var
,
219 const glsl_type
*type
= var
->type
;
221 if (this->shader_stage
== MESA_SHADER_GEOMETRY
&&
222 var
->data
.mode
== ir_var_shader_in
) {
223 /* The only geometry shader input that is not an array is
224 * gl_PrimitiveIDIn, and in that case, this code will never be reached,
225 * because gl_PrimitiveIDIn can't be indexed into in array fashion.
227 assert(type
->is_array());
228 type
= type
->fields
.array
;
231 if (this->shader_stage
== MESA_SHADER_TESS_CTRL
&&
232 var
->data
.mode
== ir_var_shader_in
) {
233 assert(type
->is_array());
234 type
= type
->fields
.array
;
237 if (this->shader_stage
== MESA_SHADER_TESS_CTRL
&&
238 var
->data
.mode
== ir_var_shader_out
&& !var
->data
.patch
) {
239 assert(type
->is_array());
240 type
= type
->fields
.array
;
243 if (this->shader_stage
== MESA_SHADER_TESS_EVAL
&&
244 var
->data
.mode
== ir_var_shader_in
&& !var
->data
.patch
) {
245 assert(type
->is_array());
246 type
= type
->fields
.array
;
249 /* TODO: implement proper arrays of arrays support
250 * for now let the caller mark whole variable as used.
252 if (type
->is_array() && type
->fields
.array
->is_array())
255 /* The code below only handles:
257 * - Indexing into matrices
258 * - Indexing into arrays of (matrices, vectors, or scalars)
260 * All other possibilities are either prohibited by GLSL (vertex inputs and
261 * fragment outputs can't be structs) or should have been eliminated by
262 * lowering passes (do_vec_index_to_swizzle() gets rid of indexing into
263 * vectors, and lower_packed_varyings() gets rid of structs that occur in
266 * However, we don't use varying packing in all cases - tessellation
267 * shaders bypass it. This means we'll see varying structs and arrays
268 * of structs here. For now, we just give up so the caller marks the
269 * entire variable as used.
271 if (!(type
->is_matrix() ||
273 (type
->fields
.array
->is_numeric() ||
274 type
->fields
.array
->is_boolean())))) {
276 /* If we don't know how to handle this case, give up and let the
277 * caller mark the whole variable as used.
282 ir_constant
*index_as_constant
= index
->as_constant();
283 if (!index_as_constant
)
288 if (type
->is_array()) {
289 num_elems
= type
->length
;
290 if (type
->fields
.array
->is_matrix())
291 elem_width
= type
->fields
.array
->matrix_columns
;
295 num_elems
= type
->matrix_columns
;
299 if (index_as_constant
->value
.u
[0] >= num_elems
) {
300 /* Constant index outside the bounds of the matrix/array. This could
301 * arise as a result of constant folding of a legal GLSL program.
303 * Even though the spec says that indexing outside the bounds of a
304 * matrix/array results in undefined behaviour, we don't want to pass
305 * out-of-range values to mark() (since this could result in slots that
306 * don't exist being marked as used), so just let the caller mark the
307 * whole variable as used.
312 /* double element width for double types that takes two slots */
313 if (this->shader_stage
!= MESA_SHADER_VERTEX
||
314 var
->data
.mode
!= ir_var_shader_in
) {
315 if (type
->without_array()->is_dual_slot())
319 mark(this->prog
, var
, index_as_constant
->value
.u
[0] * elem_width
,
320 elem_width
, this->shader_stage
);
325 is_multiple_vertices(gl_shader_stage stage
, ir_variable
*var
)
330 if (var
->data
.mode
== ir_var_shader_in
)
331 return stage
== MESA_SHADER_GEOMETRY
||
332 stage
== MESA_SHADER_TESS_CTRL
||
333 stage
== MESA_SHADER_TESS_EVAL
;
334 if (var
->data
.mode
== ir_var_shader_out
)
335 return stage
== MESA_SHADER_TESS_CTRL
;
341 ir_set_program_inouts_visitor::visit_enter(ir_dereference_array
*ir
)
343 /* Note: for geometry shader inputs, lower_named_interface_blocks may
344 * create 2D arrays, so we need to be able to handle those. 2D arrays
345 * shouldn't be able to crop up for any other reason.
347 if (ir_dereference_array
* const inner_array
=
348 ir
->array
->as_dereference_array()) {
350 * inner_array => foo[i]
352 if (ir_dereference_variable
* const deref_var
=
353 inner_array
->array
->as_dereference_variable()) {
354 if (is_multiple_vertices(this->shader_stage
, deref_var
->var
)) {
355 /* foo is a geometry or tessellation shader input, so i is
356 * the vertex, and j the part of the input we're accessing.
358 if (try_mark_partial_variable(deref_var
->var
, ir
->array_index
))
360 /* We've now taken care of foo and j, but i might contain a
361 * subexpression that accesses shader inputs. So manually
362 * visit i and then continue with the parent.
364 inner_array
->array_index
->accept(this);
365 return visit_continue_with_parent
;
369 } else if (ir_dereference_variable
* const deref_var
=
370 ir
->array
->as_dereference_variable()) {
371 /* ir => foo[i], where foo is a variable. */
372 if (is_multiple_vertices(this->shader_stage
, deref_var
->var
)) {
373 /* foo is a geometry or tessellation shader input, so i is
374 * the vertex, and we're accessing the entire input.
376 mark_whole_variable(deref_var
->var
);
377 /* We've now taken care of foo, but i might contain a subexpression
378 * that accesses shader inputs. So manually visit i and then
379 * continue with the parent.
381 ir
->array_index
->accept(this);
382 return visit_continue_with_parent
;
383 } else if (is_shader_inout(deref_var
->var
)) {
384 /* foo is a shader input/output, but not a geometry shader input,
385 * so i is the part of the input we're accessing.
387 if (try_mark_partial_variable(deref_var
->var
, ir
->array_index
))
388 return visit_continue_with_parent
;
392 /* The expression is something we don't recognize. Just visit its
395 return visit_continue
;
399 ir_set_program_inouts_visitor::visit_enter(ir_function_signature
*ir
)
401 /* We don't want to descend into the function parameters and
402 * consider them as shader inputs or outputs.
404 visit_list_elements(this, &ir
->body
);
405 return visit_continue_with_parent
;
409 ir_set_program_inouts_visitor::visit_enter(ir_expression
*ir
)
411 if (this->shader_stage
== MESA_SHADER_FRAGMENT
&&
412 (ir
->operation
== ir_unop_dFdy
||
413 ir
->operation
== ir_unop_dFdy_coarse
||
414 ir
->operation
== ir_unop_dFdy_fine
)) {
415 gl_fragment_program
*fprog
= (gl_fragment_program
*) prog
;
416 fprog
->UsesDFdy
= true;
418 return visit_continue
;
422 ir_set_program_inouts_visitor::visit_enter(ir_discard
*)
424 /* discards are only allowed in fragment shaders. */
425 assert(this->shader_stage
== MESA_SHADER_FRAGMENT
);
427 gl_fragment_program
*fprog
= (gl_fragment_program
*) prog
;
428 fprog
->UsesKill
= true;
430 return visit_continue
;
434 ir_set_program_inouts_visitor::visit_enter(ir_texture
*ir
)
436 if (ir
->op
== ir_tg4
)
437 prog
->UsesGather
= true;
438 return visit_continue
;
442 do_set_program_inouts(exec_list
*instructions
, struct gl_program
*prog
,
443 gl_shader_stage shader_stage
)
445 ir_set_program_inouts_visitor
v(prog
, shader_stage
);
447 prog
->InputsRead
= 0;
448 prog
->OutputsWritten
= 0;
449 prog
->OutputsRead
= 0;
450 prog
->PatchInputsRead
= 0;
451 prog
->PatchOutputsWritten
= 0;
452 prog
->SystemValuesRead
= 0;
453 if (shader_stage
== MESA_SHADER_FRAGMENT
) {
454 gl_fragment_program
*fprog
= (gl_fragment_program
*) prog
;
455 memset(fprog
->InterpQualifier
, 0, sizeof(fprog
->InterpQualifier
));
456 fprog
->IsCentroid
= 0;
458 fprog
->UsesDFdy
= false;
459 fprog
->UsesKill
= false;
461 visit_list_elements(&v
, instructions
);