1 /*
2 * Copyright © 2014 Intel Corporation
3 *
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:
10 *
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
13 * Software.
14 *
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
21 * IN THE SOFTWARE.
22 */
23
24 #include "nir.h"
25 #include "nir_builder.h"
26
27 /*
28 * Implements "copy splitting" which is similar to structure splitting only
29 * it works on copy operations rather than the datatypes themselves. The
30 * GLSL language allows you to copy one variable to another an entire
31 * structure (which may contain arrays or other structures) at a time.
32 * Normally, in a language such as C this would be handled by a "structure
33 * splitting" pass that breaks up the structures. Unfortunately for us,
34 * structures used in inputs or outputs can't be split. Therefore,
35 * regardlesss of what we do, we have to be able to copy to/from
36 * structures.
37 *
38 * The primary purpose of structure splitting is to allow you to better
39 * optimize variable access and lower things to registers where you can.
40 * The primary issue here is that, if you lower the copy to a bunch of
41 * loads and stores, you loose a lot of information about the copy
42 * operation that you would like to keep around. To solve this problem, we
43 * have a "copy splitting" pass that, instead of splitting the structures
44 * or lowering the copy into loads and storres, splits the copy operation
45 * into a bunch of copy operations one for each leaf of the structure tree.
46 * If an intermediate array is encountered, it is referenced with a
47 * wildcard reference to indicate that the entire array is to be copied.
48 *
49 * As things become direct, array copies may be able to be losslessly
50 * lowered to having fewer and fewer wildcards. However, until that
51 * happens we want to keep the information about the arrays intact.
52 *
53 * Prior to the copy splitting pass, there are no wildcard references but
54 * there may be incomplete references where the tail of the deref chain is
55 * an array or a structure and not a specific element. After the copy
56 * splitting pass has completed, every variable deref will be a full-length
57 * dereference pointing to a single leaf in the structure type tree with
58 * possibly a few wildcard array dereferences.
59 */
60
61 static void
split_deref_copy_instr(nir_builder * b,nir_deref_instr * dst,nir_deref_instr * src,enum gl_access_qualifier dst_access,enum gl_access_qualifier src_access)62 split_deref_copy_instr(nir_builder *b,
63 nir_deref_instr *dst, nir_deref_instr *src,
64 enum gl_access_qualifier dst_access,
65 enum gl_access_qualifier src_access)
66 {
67 assert(glsl_get_bare_type(dst->type) ==
68 glsl_get_bare_type(src->type));
69 if (glsl_type_is_vector_or_scalar(src->type)) {
70 nir_copy_deref_with_access(b, dst, src, dst_access, src_access);
71 } else if (glsl_type_is_struct_or_ifc(src->type)) {
72 for (unsigned i = 0; i < glsl_get_length(src->type); i++) {
73 split_deref_copy_instr(b, nir_build_deref_struct(b, dst, i),
74 nir_build_deref_struct(b, src, i),
75 dst_access, src_access);
76 }
77 } else {
78 assert(glsl_type_is_matrix(src->type) || glsl_type_is_array(src->type));
79 split_deref_copy_instr(b, nir_build_deref_array_wildcard(b, dst),
80 nir_build_deref_array_wildcard(b, src),
81 dst_access, src_access);
82 }
83 }
84
85 static bool
split_var_copies_instr(nir_builder * b,nir_intrinsic_instr * copy,UNUSED void * cb_data)86 split_var_copies_instr(nir_builder *b, nir_intrinsic_instr *copy,
87 UNUSED void *cb_data)
88 {
89 if (copy->intrinsic != nir_intrinsic_copy_deref)
90 return false;
91
92 b->cursor = nir_instr_remove(©->instr);
93
94 nir_deref_instr *dst = nir_instr_as_deref(copy->src[0].ssa->parent_instr);
95 nir_deref_instr *src = nir_instr_as_deref(copy->src[1].ssa->parent_instr);
96 split_deref_copy_instr(b, dst, src,
97 nir_intrinsic_dst_access(copy),
98 nir_intrinsic_src_access(copy));
99
100 return true;
101 }
102
103 bool
nir_split_var_copies(nir_shader * shader)104 nir_split_var_copies(nir_shader *shader)
105 {
106 return nir_shader_intrinsics_pass(shader, split_var_copies_instr,
107 nir_metadata_control_flow,
108 NULL);
109 }
110