i965: Calculate thread_count in brw_alloc_stage_scratch
[mesa.git] / src / mesa / drivers / dri / i965 / brw_nir_uniforms.cpp
1 /*
2 * Copyright © 2015 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 "compiler/brw_nir.h"
25 #include "compiler/glsl/ir_uniform.h"
26 #include "brw_program.h"
27
28 static void
29 brw_nir_setup_glsl_builtin_uniform(nir_variable *var,
30 const struct gl_program *prog,
31 struct brw_stage_prog_data *stage_prog_data,
32 bool is_scalar)
33 {
34 const nir_state_slot *const slots = var->state_slots;
35 assert(var->state_slots != NULL);
36
37 unsigned uniform_index = var->data.driver_location / 4;
38 for (unsigned int i = 0; i < var->num_state_slots; i++) {
39 /* This state reference has already been setup by ir_to_mesa, but we'll
40 * get the same index back here.
41 */
42 int index = _mesa_add_state_reference(prog->Parameters,
43 (gl_state_index *)slots[i].tokens);
44
45 /* Add each of the unique swizzles of the element as a parameter.
46 * This'll end up matching the expected layout of the
47 * array/matrix/structure we're trying to fill in.
48 */
49 int last_swiz = -1;
50 for (unsigned j = 0; j < 4; j++) {
51 int swiz = GET_SWZ(slots[i].swizzle, j);
52
53 /* If we hit a pair of identical swizzles, this means we've hit the
54 * end of the builtin variable. In scalar mode, we should just quit
55 * and move on to the next one. In vec4, we need to continue and pad
56 * it out to 4 components.
57 */
58 if (swiz == last_swiz && is_scalar)
59 break;
60
61 last_swiz = swiz;
62
63 stage_prog_data->param[uniform_index++] =
64 BRW_PARAM_PARAMETER(index, swiz);
65 }
66 }
67 }
68
69 static void
70 setup_vec4_image_param(uint32_t *params, uint32_t idx,
71 unsigned offset, unsigned n)
72 {
73 assert(offset % sizeof(uint32_t) == 0);
74 for (unsigned i = 0; i < n; ++i)
75 params[i] = BRW_PARAM_IMAGE(idx, offset / sizeof(uint32_t) + i);
76
77 for (unsigned i = n; i < 4; ++i)
78 params[i] = BRW_PARAM_BUILTIN_ZERO;
79 }
80
81 static void
82 brw_setup_image_uniform_values(gl_shader_stage stage,
83 struct brw_stage_prog_data *stage_prog_data,
84 unsigned param_start_index,
85 const gl_uniform_storage *storage)
86 {
87 uint32_t *param = &stage_prog_data->param[param_start_index];
88
89 for (unsigned i = 0; i < MAX2(storage->array_elements, 1); i++) {
90 const unsigned image_idx = storage->opaque[stage].index + i;
91
92 /* Upload the brw_image_param structure. The order is expected to match
93 * the BRW_IMAGE_PARAM_*_OFFSET defines.
94 */
95 setup_vec4_image_param(param + BRW_IMAGE_PARAM_SURFACE_IDX_OFFSET,
96 image_idx,
97 offsetof(brw_image_param, surface_idx), 1);
98 setup_vec4_image_param(param + BRW_IMAGE_PARAM_OFFSET_OFFSET,
99 image_idx,
100 offsetof(brw_image_param, offset), 2);
101 setup_vec4_image_param(param + BRW_IMAGE_PARAM_SIZE_OFFSET,
102 image_idx,
103 offsetof(brw_image_param, size), 3);
104 setup_vec4_image_param(param + BRW_IMAGE_PARAM_STRIDE_OFFSET,
105 image_idx,
106 offsetof(brw_image_param, stride), 4);
107 setup_vec4_image_param(param + BRW_IMAGE_PARAM_TILING_OFFSET,
108 image_idx,
109 offsetof(brw_image_param, tiling), 3);
110 setup_vec4_image_param(param + BRW_IMAGE_PARAM_SWIZZLING_OFFSET,
111 image_idx,
112 offsetof(brw_image_param, swizzling), 2);
113 param += BRW_IMAGE_PARAM_SIZE;
114
115 brw_mark_surface_used(
116 stage_prog_data,
117 stage_prog_data->binding_table.image_start + image_idx);
118 }
119 }
120
121 static void
122 brw_nir_setup_glsl_uniform(gl_shader_stage stage, nir_variable *var,
123 const struct gl_program *prog,
124 struct brw_stage_prog_data *stage_prog_data,
125 bool is_scalar)
126 {
127 int namelen = strlen(var->name);
128
129 /* The data for our (non-builtin) uniforms is stored in a series of
130 * gl_uniform_storage structs for each subcomponent that
131 * glGetUniformLocation() could name. We know it's been set up in the same
132 * order we'd walk the type, so walk the list of storage and find anything
133 * with our name, or the prefix of a component that starts with our name.
134 */
135 unsigned uniform_index = var->data.driver_location / 4;
136 for (unsigned u = 0; u < prog->sh.data->NumUniformStorage; u++) {
137 struct gl_uniform_storage *storage =
138 &prog->sh.data->UniformStorage[u];
139
140 if (storage->builtin || storage->type->is_sampler())
141 continue;
142
143 if (strncmp(var->name, storage->name, namelen) != 0 ||
144 (storage->name[namelen] != 0 &&
145 storage->name[namelen] != '.' &&
146 storage->name[namelen] != '[')) {
147 continue;
148 }
149
150 if (storage->type->is_image()) {
151 brw_setup_image_uniform_values(stage, stage_prog_data,
152 uniform_index, storage);
153 uniform_index +=
154 BRW_IMAGE_PARAM_SIZE * MAX2(storage->array_elements, 1);
155 } else {
156 gl_constant_value *components = storage->storage;
157 unsigned vector_count = (MAX2(storage->array_elements, 1) *
158 storage->type->matrix_columns);
159 unsigned vector_size = storage->type->vector_elements;
160 unsigned max_vector_size = 4;
161 if (storage->type->base_type == GLSL_TYPE_DOUBLE ||
162 storage->type->base_type == GLSL_TYPE_UINT64 ||
163 storage->type->base_type == GLSL_TYPE_INT64) {
164 vector_size *= 2;
165 if (vector_size > 4)
166 max_vector_size = 8;
167 }
168
169 for (unsigned s = 0; s < vector_count; s++) {
170 unsigned i;
171 for (i = 0; i < vector_size; i++) {
172 uint32_t idx = components - prog->sh.data->UniformDataSlots;
173 stage_prog_data->param[uniform_index++] = BRW_PARAM_UNIFORM(idx);
174 components++;
175 }
176
177 if (!is_scalar) {
178 /* Pad out with zeros if needed (only needed for vec4) */
179 for (; i < max_vector_size; i++) {
180 stage_prog_data->param[uniform_index++] =
181 BRW_PARAM_BUILTIN_ZERO;
182 }
183 }
184 }
185 }
186 }
187 }
188
189 void
190 brw_nir_setup_glsl_uniforms(void *mem_ctx, nir_shader *shader,
191 const struct gl_program *prog,
192 struct brw_stage_prog_data *stage_prog_data,
193 bool is_scalar)
194 {
195 unsigned nr_params = shader->num_uniforms / 4;
196 stage_prog_data->nr_params = nr_params;
197 stage_prog_data->param = rzalloc_array(mem_ctx, uint32_t, nr_params);
198
199 nir_foreach_variable(var, &shader->uniforms) {
200 /* UBO's, atomics and samplers don't take up space in the
201 uniform file */
202 if (var->interface_type != NULL || var->type->contains_atomic())
203 continue;
204
205 if (strncmp(var->name, "gl_", 3) == 0) {
206 brw_nir_setup_glsl_builtin_uniform(var, prog, stage_prog_data,
207 is_scalar);
208 } else {
209 brw_nir_setup_glsl_uniform(shader->info.stage, var, prog,
210 stage_prog_data, is_scalar);
211 }
212 }
213 }
214
215 void
216 brw_nir_setup_arb_uniforms(void *mem_ctx, nir_shader *shader,
217 struct gl_program *prog,
218 struct brw_stage_prog_data *stage_prog_data)
219 {
220 struct gl_program_parameter_list *plist = prog->Parameters;
221
222 unsigned nr_params = plist->NumParameters * 4;
223 stage_prog_data->nr_params = nr_params;
224 stage_prog_data->param = rzalloc_array(mem_ctx, uint32_t, nr_params);
225
226 /* For ARB programs, prog_to_nir generates a single "parameters" variable
227 * for all uniform data. nir_lower_wpos_ytransform may also create an
228 * additional variable.
229 */
230 assert(shader->uniforms.length() <= 2);
231
232 for (unsigned p = 0; p < plist->NumParameters; p++) {
233 /* Parameters should be either vec4 uniforms or single component
234 * constants; matrices and other larger types should have been broken
235 * down earlier.
236 */
237 assert(plist->Parameters[p].Size <= 4);
238
239 unsigned i;
240 for (i = 0; i < plist->Parameters[p].Size; i++)
241 stage_prog_data->param[4 * p + i] = BRW_PARAM_PARAMETER(p, i);
242 for (; i < 4; i++)
243 stage_prog_data->param[4 * p + i] = BRW_PARAM_BUILTIN_ZERO;
244 }
245 }