1 /*
2 * Copyright © 2010 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 extern "C" {
25 #include "main/macros.h"
26 #include "brw_context.h"
27 #include "brw_vs.h"
28 }
29 #include "brw_fs.h"
30 #include "glsl/ir_optimization.h"
31 #include "glsl/ir_print_visitor.h"
32
33 struct gl_shader *
brw_new_shader(struct gl_context * ctx,GLuint name,GLuint type)34 brw_new_shader(struct gl_context *ctx, GLuint name, GLuint type)
35 {
36 struct brw_shader *shader;
37
38 shader = rzalloc(NULL, struct brw_shader);
39 if (shader) {
40 shader->base.Type = type;
41 shader->base.Name = name;
42 _mesa_init_shader(ctx, &shader->base);
43 }
44
45 return &shader->base;
46 }
47
48 struct gl_shader_program *
brw_new_shader_program(struct gl_context * ctx,GLuint name)49 brw_new_shader_program(struct gl_context *ctx, GLuint name)
50 {
51 struct brw_shader_program *prog;
52 prog = rzalloc(NULL, struct brw_shader_program);
53 if (prog) {
54 prog->base.Name = name;
55 _mesa_init_shader_program(ctx, &prog->base);
56 }
57 return &prog->base;
58 }
59
60 /**
61 * Performs a compile of the shader stages even when we don't know
62 * what non-orthogonal state will be set, in the hope that it reflects
63 * the eventual NOS used, and thus allows us to produce link failures.
64 */
65 bool
brw_shader_precompile(struct gl_context * ctx,struct gl_shader_program * prog)66 brw_shader_precompile(struct gl_context *ctx, struct gl_shader_program *prog)
67 {
68 struct brw_context *brw = brw_context(ctx);
69
70 if (brw->precompile && !brw_fs_precompile(ctx, prog))
71 return false;
72
73 if (brw->precompile && !brw_vs_precompile(ctx, prog))
74 return false;
75
76 return true;
77 }
78
79 GLboolean
brw_link_shader(struct gl_context * ctx,struct gl_shader_program * shProg)80 brw_link_shader(struct gl_context *ctx, struct gl_shader_program *shProg)
81 {
82 struct brw_context *brw = brw_context(ctx);
83 struct intel_context *intel = &brw->intel;
84 unsigned int stage;
85
86 for (stage = 0; stage < ARRAY_SIZE(shProg->_LinkedShaders); stage++) {
87 struct brw_shader *shader =
88 (struct brw_shader *)shProg->_LinkedShaders[stage];
89 static const GLenum targets[] = {
90 GL_VERTEX_PROGRAM_ARB,
91 GL_FRAGMENT_PROGRAM_ARB,
92 GL_GEOMETRY_PROGRAM_NV
93 };
94
95 if (!shader)
96 continue;
97
98 struct gl_program *prog =
99 ctx->Driver.NewProgram(ctx, targets[stage], shader->base.Name);
100 if (!prog)
101 return false;
102 prog->Parameters = _mesa_new_parameter_list();
103
104 _mesa_generate_parameters_list_for_uniforms(shProg, &shader->base,
105 prog->Parameters);
106
107 if (stage == 0) {
108 struct gl_vertex_program *vp = (struct gl_vertex_program *) prog;
109 vp->UsesClipDistance = shProg->Vert.UsesClipDistance;
110 }
111
112 void *mem_ctx = ralloc_context(NULL);
113 bool progress;
114
115 if (shader->ir)
116 ralloc_free(shader->ir);
117 shader->ir = new(shader) exec_list;
118 clone_ir_list(mem_ctx, shader->ir, shader->base.ir);
119
120 do_mat_op_to_vec(shader->ir);
121 lower_instructions(shader->ir,
122 MOD_TO_FRACT |
123 DIV_TO_MUL_RCP |
124 SUB_TO_ADD_NEG |
125 EXP_TO_EXP2 |
126 LOG_TO_LOG2);
127
128 /* Pre-gen6 HW can only nest if-statements 16 deep. Beyond this,
129 * if-statements need to be flattened.
130 */
131 if (intel->gen < 6)
132 lower_if_to_cond_assign(shader->ir, 16);
133
134 do_lower_texture_projection(shader->ir);
135 if (intel->gen < 8 && !intel->is_haswell)
136 brw_lower_texture_gradients(shader->ir);
137 do_vec_index_to_cond_assign(shader->ir);
138 brw_do_cubemap_normalize(shader->ir);
139 lower_noise(shader->ir);
140 lower_quadop_vector(shader->ir, false);
141
142 bool input = true;
143 bool output = stage == MESA_SHADER_FRAGMENT;
144 bool temp = stage == MESA_SHADER_FRAGMENT;
145 bool uniform = stage == MESA_SHADER_FRAGMENT;
146
147 lower_variable_index_to_cond_assign(shader->ir,
148 input, output, temp, uniform);
149
150 /* FINISHME: Do this before the variable index lowering. */
151 lower_ubo_reference(&shader->base, shader->ir);
152
153 do {
154 progress = false;
155
156 if (stage == MESA_SHADER_FRAGMENT) {
157 brw_do_channel_expressions(shader->ir);
158 brw_do_vector_splitting(shader->ir);
159 }
160
161 progress = do_lower_jumps(shader->ir, true, true,
162 true, /* main return */
163 false, /* continue */
164 false /* loops */
165 ) || progress;
166
167 progress = do_common_optimization(shader->ir, true, true, 32)
168 || progress;
169 } while (progress);
170
171 /* Make a pass over the IR to add state references for any built-in
172 * uniforms that are used. This has to be done now (during linking).
173 * Code generation doesn't happen until the first time this shader is
174 * used for rendering. Waiting until then to generate the parameters is
175 * too late. At that point, the values for the built-in informs won't
176 * get sent to the shader.
177 */
178 foreach_list(node, shader->ir) {
179 ir_variable *var = ((ir_instruction *) node)->as_variable();
180
181 if ((var == NULL) || (var->mode != ir_var_uniform)
182 || (strncmp(var->name, "gl_", 3) != 0))
183 continue;
184
185 const ir_state_slot *const slots = var->state_slots;
186 assert(var->state_slots != NULL);
187
188 for (unsigned int i = 0; i < var->num_state_slots; i++) {
189 _mesa_add_state_reference(prog->Parameters,
190 (gl_state_index *) slots[i].tokens);
191 }
192 }
193
194 validate_ir_tree(shader->ir);
195
196 reparent_ir(shader->ir, shader->ir);
197 ralloc_free(mem_ctx);
198
199 do_set_program_inouts(shader->ir, prog,
200 shader->base.Type == GL_FRAGMENT_SHADER);
201
202 prog->SamplersUsed = shader->base.active_samplers;
203 _mesa_update_shader_textures_used(shProg, prog);
204
205 _mesa_reference_program(ctx, &shader->base.Program, prog);
206
207 /* This has to be done last. Any operation that can cause
208 * prog->ParameterValues to get reallocated (e.g., anything that adds a
209 * program constant) has to happen before creating this linkage.
210 */
211 _mesa_associate_uniform_storage(ctx, shProg, prog->Parameters);
212
213 _mesa_reference_program(ctx, &prog, NULL);
214 }
215
216 if (!brw_shader_precompile(ctx, shProg))
217 return false;
218
219 return true;
220 }
221
222
223 int
brw_type_for_base_type(const struct glsl_type * type)224 brw_type_for_base_type(const struct glsl_type *type)
225 {
226 switch (type->base_type) {
227 case GLSL_TYPE_FLOAT:
228 return BRW_REGISTER_TYPE_F;
229 case GLSL_TYPE_INT:
230 case GLSL_TYPE_BOOL:
231 return BRW_REGISTER_TYPE_D;
232 case GLSL_TYPE_UINT:
233 return BRW_REGISTER_TYPE_UD;
234 case GLSL_TYPE_ARRAY:
235 return brw_type_for_base_type(type->fields.array);
236 case GLSL_TYPE_STRUCT:
237 case GLSL_TYPE_SAMPLER:
238 /* These should be overridden with the type of the member when
239 * dereferenced into. BRW_REGISTER_TYPE_UD seems like a likely
240 * way to trip up if we don't.
241 */
242 return BRW_REGISTER_TYPE_UD;
243 default:
244 assert(!"not reached");
245 return BRW_REGISTER_TYPE_F;
246 }
247 }
248
249 uint32_t
brw_conditional_for_comparison(unsigned int op)250 brw_conditional_for_comparison(unsigned int op)
251 {
252 switch (op) {
253 case ir_binop_less:
254 return BRW_CONDITIONAL_L;
255 case ir_binop_greater:
256 return BRW_CONDITIONAL_G;
257 case ir_binop_lequal:
258 return BRW_CONDITIONAL_LE;
259 case ir_binop_gequal:
260 return BRW_CONDITIONAL_GE;
261 case ir_binop_equal:
262 case ir_binop_all_equal: /* same as equal for scalars */
263 return BRW_CONDITIONAL_Z;
264 case ir_binop_nequal:
265 case ir_binop_any_nequal: /* same as nequal for scalars */
266 return BRW_CONDITIONAL_NZ;
267 default:
268 assert(!"not reached: bad operation for comparison");
269 return BRW_CONDITIONAL_NZ;
270 }
271 }
272
273 uint32_t
brw_math_function(enum opcode op)274 brw_math_function(enum opcode op)
275 {
276 switch (op) {
277 case SHADER_OPCODE_RCP:
278 return BRW_MATH_FUNCTION_INV;
279 case SHADER_OPCODE_RSQ:
280 return BRW_MATH_FUNCTION_RSQ;
281 case SHADER_OPCODE_SQRT:
282 return BRW_MATH_FUNCTION_SQRT;
283 case SHADER_OPCODE_EXP2:
284 return BRW_MATH_FUNCTION_EXP;
285 case SHADER_OPCODE_LOG2:
286 return BRW_MATH_FUNCTION_LOG;
287 case SHADER_OPCODE_POW:
288 return BRW_MATH_FUNCTION_POW;
289 case SHADER_OPCODE_SIN:
290 return BRW_MATH_FUNCTION_SIN;
291 case SHADER_OPCODE_COS:
292 return BRW_MATH_FUNCTION_COS;
293 case SHADER_OPCODE_INT_QUOTIENT:
294 return BRW_MATH_FUNCTION_INT_DIV_QUOTIENT;
295 case SHADER_OPCODE_INT_REMAINDER:
296 return BRW_MATH_FUNCTION_INT_DIV_REMAINDER;
297 default:
298 assert(!"not reached: unknown math function");
299 return 0;
300 }
301 }
302
303 uint32_t
brw_texture_offset(ir_constant * offset)304 brw_texture_offset(ir_constant *offset)
305 {
306 assert(offset != NULL);
307
308 signed char offsets[3];
309 for (unsigned i = 0; i < offset->type->vector_elements; i++)
310 offsets[i] = (signed char) offset->value.i[i];
311
312 /* Combine all three offsets into a single unsigned dword:
313 *
314 * bits 11:8 - U Offset (X component)
315 * bits 7:4 - V Offset (Y component)
316 * bits 3:0 - R Offset (Z component)
317 */
318 unsigned offset_bits = 0;
319 for (unsigned i = 0; i < offset->type->vector_elements; i++) {
320 const unsigned shift = 4 * (2 - i);
321 offset_bits |= (offsets[i] << shift) & (0xF << shift);
322 }
323 return offset_bits;
324 }
325