nir: Silence a couple of warnings in release builds
[mesa.git] / src / compiler / nir / nir_gather_xfb_info.c
1 /*
2 * Copyright © 2018 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_xfb_info.h"
25
26 #include <util/u_math.h>
27
28 static void
29 add_var_xfb_outputs(nir_xfb_info *xfb,
30 nir_variable *var,
31 unsigned buffer,
32 unsigned *location,
33 unsigned *offset,
34 const struct glsl_type *type)
35 {
36 if (glsl_type_is_array(type) || glsl_type_is_matrix(type)) {
37 unsigned length = glsl_get_length(type);
38 const struct glsl_type *child_type = glsl_get_array_element(type);
39 for (unsigned i = 0; i < length; i++)
40 add_var_xfb_outputs(xfb, var, buffer, location, offset, child_type);
41 } else if (glsl_type_is_struct(type)) {
42 unsigned length = glsl_get_length(type);
43 for (unsigned i = 0; i < length; i++) {
44 const struct glsl_type *child_type = glsl_get_struct_field(type, i);
45 add_var_xfb_outputs(xfb, var, buffer, location, offset, child_type);
46 }
47 } else {
48 assert(buffer < NIR_MAX_XFB_BUFFERS);
49 if (xfb->buffers_written & (1 << buffer)) {
50 assert(xfb->strides[buffer] == var->data.xfb_stride);
51 assert(xfb->buffer_to_stream[buffer] == var->data.stream);
52 } else {
53 xfb->buffers_written |= (1 << buffer);
54 xfb->strides[buffer] = var->data.xfb_stride;
55 xfb->buffer_to_stream[buffer] = var->data.stream;
56 }
57
58 assert(var->data.stream < NIR_MAX_XFB_STREAMS);
59 xfb->streams_written |= (1 << var->data.stream);
60
61 unsigned comp_slots = glsl_get_component_slots(type);
62 unsigned attrib_slots = DIV_ROUND_UP(comp_slots, 4);
63 assert(attrib_slots == glsl_count_attribute_slots(type, false));
64
65 /* Ensure that we don't have, for instance, a dvec2 with a location_frac
66 * of 2 which would make it crass a location boundary even though it
67 * fits in a single slot. However, you can have a dvec3 which crosses
68 * the slot boundary with a location_frac of 2.
69 */
70 assert(DIV_ROUND_UP(var->data.location_frac + comp_slots, 4) == attrib_slots);
71
72 assert(var->data.location_frac + comp_slots <= 8);
73 uint8_t comp_mask = ((1 << comp_slots) - 1) << var->data.location_frac;
74
75 assert(attrib_slots <= 2);
76 for (unsigned s = 0; s < attrib_slots; s++) {
77 nir_xfb_output_info *output = &xfb->outputs[xfb->output_count++];
78
79 output->buffer = buffer;
80 output->offset = *offset + s * 16;
81 output->location = *location;
82 output->component_mask = (comp_mask >> (s * 4)) & 0xf;
83
84 (*location)++;
85 }
86 *offset += comp_slots * 4;
87 }
88 }
89
90 static int
91 compare_xfb_output_offsets(const void *_a, const void *_b)
92 {
93 const nir_xfb_output_info *a = _a, *b = _b;
94 return a->offset - b->offset;
95 }
96
97 nir_xfb_info *
98 nir_gather_xfb_info(const nir_shader *shader, void *mem_ctx)
99 {
100 assert(shader->info.stage == MESA_SHADER_VERTEX ||
101 shader->info.stage == MESA_SHADER_TESS_EVAL ||
102 shader->info.stage == MESA_SHADER_GEOMETRY);
103
104 /* Compute the number of outputs we have. This is simply the number of
105 * cumulative locations consumed by all the variables. If a location is
106 * represented by multiple variables, then they each count separately in
107 * number of outputs. This is only an estimate as some variables may have
108 * an xfb_buffer but not an output so it may end up larger than we need but
109 * it should be good enough for allocation.
110 */
111 unsigned num_outputs = 0;
112 nir_foreach_variable(var, &shader->outputs) {
113 if (var->data.explicit_xfb_buffer)
114 num_outputs += glsl_count_attribute_slots(var->type, false);
115 }
116 if (num_outputs == 0)
117 return NULL;
118
119 nir_xfb_info *xfb = rzalloc_size(mem_ctx, nir_xfb_info_size(num_outputs));
120
121 /* Walk the list of outputs and add them to the array */
122 nir_foreach_variable(var, &shader->outputs) {
123 if (!var->data.explicit_xfb_buffer)
124 continue;
125
126 unsigned location = var->data.location;
127
128 /* In order to know if we have a array of blocks can't be done just by
129 * checking if we have an interface type and is an array, because due
130 * splitting we could end on a case were we received a split struct
131 * that contains an array.
132 */
133 bool is_array_block = var->interface_type != NULL &&
134 glsl_type_is_array(var->type) &&
135 glsl_without_array(var->type) == glsl_get_bare_type(var->interface_type);
136
137 if (var->data.explicit_offset && !is_array_block) {
138 unsigned offset = var->data.offset;
139 add_var_xfb_outputs(xfb, var, var->data.xfb_buffer,
140 &location, &offset, var->type);
141 } else if (is_array_block) {
142 assert(glsl_type_is_struct(var->interface_type));
143
144 unsigned aoa_size = glsl_get_aoa_size(var->type);
145 const struct glsl_type *itype = var->interface_type;
146 unsigned nfields = glsl_get_length(itype);
147 for (unsigned b = 0; b < aoa_size; b++) {
148 for (unsigned f = 0; f < nfields; f++) {
149 int foffset = glsl_get_struct_field_offset(itype, f);
150 const struct glsl_type *ftype = glsl_get_struct_field(itype, f);
151 if (foffset < 0) {
152 location += glsl_count_attribute_slots(ftype, false);
153 continue;
154 }
155
156 unsigned offset = foffset;
157 add_var_xfb_outputs(xfb, var, var->data.xfb_buffer + b,
158 &location, &offset, ftype);
159 }
160 }
161 }
162 }
163
164 /* Everything is easier in the state setup code if the list is sorted in
165 * order of output offset.
166 */
167 qsort(xfb->outputs, xfb->output_count, sizeof(xfb->outputs[0]),
168 compare_xfb_output_offsets);
169
170 #ifndef NDEBUG
171 /* Finally, do a sanity check */
172 unsigned max_offset[NIR_MAX_XFB_BUFFERS] = {0};
173 for (unsigned i = 0; i < xfb->output_count; i++) {
174 assert(xfb->outputs[i].offset >= max_offset[xfb->outputs[i].buffer]);
175 assert(xfb->outputs[i].component_mask != 0);
176 unsigned slots = util_bitcount(xfb->outputs[i].component_mask);
177 max_offset[xfb->outputs[i].buffer] = xfb->outputs[i].offset + slots * 4;
178 }
179 #endif
180
181 return xfb;
182 }