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 DEALINGS
25 #include "main/macros.h"
26 #include "brw_context.h"
30 #include "glsl/ir_optimization.h"
31 #include "glsl/ir_print_visitor.h"
34 brw_new_shader(struct gl_context *ctx, GLuint name, GLuint type)
36 struct brw_shader *shader;
38 shader = rzalloc(NULL, struct brw_shader);
40 shader->base.Type = type;
41 shader->base.Name = name;
42 _mesa_init_shader(ctx, &shader->base);
48 struct gl_shader_program *
49 brw_new_shader_program(struct gl_context *ctx, GLuint name)
51 struct brw_shader_program *prog;
52 prog = rzalloc(NULL, struct brw_shader_program);
54 prog->base.Name = name;
55 _mesa_init_shader_program(ctx, &prog->base);
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.
66 brw_shader_precompile(struct gl_context *ctx, struct gl_shader_program *prog)
68 struct brw_context *brw = brw_context(ctx);
70 if (brw->precompile && !brw_fs_precompile(ctx, prog))
73 if (!brw_vs_precompile(ctx, prog))
80 brw_link_shader(struct gl_context *ctx, struct gl_shader_program *shProg)
82 struct brw_context *brw = brw_context(ctx);
83 struct intel_context *intel = &brw->intel;
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
98 struct gl_program *prog =
99 ctx->Driver.NewProgram(ctx, targets[stage], shader->base.Name);
102 prog->Parameters = _mesa_new_parameter_list();
104 _mesa_generate_parameters_list_for_uniforms(shProg, &shader->base,
108 struct gl_vertex_program *vp = (struct gl_vertex_program *) prog;
109 vp->UsesClipDistance = shProg->Vert.UsesClipDistance;
113 class uses_kill_visitor : public ir_hierarchical_visitor {
115 uses_kill_visitor() : uses_kill(false)
120 virtual ir_visitor_status visit_enter(class ir_discard *ir)
122 this->uses_kill = true;
131 v.run(shader->base.ir);
133 struct gl_fragment_program *fp = (struct gl_fragment_program *) prog;
134 fp->UsesKill = v.uses_kill;
137 void *mem_ctx = ralloc_context(NULL);
141 ralloc_free(shader->ir);
142 shader->ir = new(shader) exec_list;
143 clone_ir_list(mem_ctx, shader->ir, shader->base.ir);
145 do_mat_op_to_vec(shader->ir);
146 lower_instructions(shader->ir,
153 /* Pre-gen6 HW can only nest if-statements 16 deep. Beyond this,
154 * if-statements need to be flattened.
157 lower_if_to_cond_assign(shader->ir, 16);
159 do_lower_texture_projection(shader->ir);
160 do_vec_index_to_cond_assign(shader->ir);
161 brw_do_cubemap_normalize(shader->ir);
162 lower_noise(shader->ir);
163 lower_quadop_vector(shader->ir, false);
166 bool output = stage == MESA_SHADER_FRAGMENT;
167 bool temp = stage == MESA_SHADER_FRAGMENT;
168 bool uniform = stage == MESA_SHADER_FRAGMENT;
170 lower_variable_index_to_cond_assign(shader->ir,
171 input, output, temp, uniform);
176 if (stage == MESA_SHADER_FRAGMENT) {
177 brw_do_channel_expressions(shader->ir);
178 brw_do_vector_splitting(shader->ir);
181 progress = do_lower_jumps(shader->ir, true, true,
182 true, /* main return */
183 false, /* continue */
187 progress = do_common_optimization(shader->ir, true, true, 32)
191 /* Make a pass over the IR to add state references for any built-in
192 * uniforms that are used. This has to be done now (during linking).
193 * Code generation doesn't happen until the first time this shader is
194 * used for rendering. Waiting until then to generate the parameters is
195 * too late. At that point, the values for the built-in informs won't
196 * get sent to the shader.
198 foreach_list(node, shader->ir) {
199 ir_variable *var = ((ir_instruction *) node)->as_variable();
201 if ((var == NULL) || (var->mode != ir_var_uniform)
202 || (strncmp(var->name, "gl_", 3) != 0))
205 const ir_state_slot *const slots = var->state_slots;
206 assert(var->state_slots != NULL);
208 for (unsigned int i = 0; i < var->num_state_slots; i++) {
209 _mesa_add_state_reference(prog->Parameters,
210 (gl_state_index *) slots[i].tokens);
214 validate_ir_tree(shader->ir);
216 reparent_ir(shader->ir, shader->ir);
217 ralloc_free(mem_ctx);
219 do_set_program_inouts(shader->ir, prog,
220 shader->base.Type == GL_FRAGMENT_SHADER);
222 prog->SamplersUsed = shader->base.active_samplers;
223 _mesa_update_shader_textures_used(shProg, prog);
225 _mesa_reference_program(ctx, &shader->base.Program, prog);
227 /* This has to be done last. Any operation that can cause
228 * prog->ParameterValues to get reallocated (e.g., anything that adds a
229 * program constant) has to happen before creating this linkage.
231 _mesa_associate_uniform_storage(ctx, shProg, prog->Parameters);
233 _mesa_reference_program(ctx, &prog, NULL);
236 if (!brw_shader_precompile(ctx, shProg))
244 brw_type_for_base_type(const struct glsl_type *type)
246 switch (type->base_type) {
247 case GLSL_TYPE_FLOAT:
248 return BRW_REGISTER_TYPE_F;
251 return BRW_REGISTER_TYPE_D;
253 return BRW_REGISTER_TYPE_UD;
254 case GLSL_TYPE_ARRAY:
255 return brw_type_for_base_type(type->fields.array);
256 case GLSL_TYPE_STRUCT:
257 case GLSL_TYPE_SAMPLER:
258 /* These should be overridden with the type of the member when
259 * dereferenced into. BRW_REGISTER_TYPE_UD seems like a likely
260 * way to trip up if we don't.
262 return BRW_REGISTER_TYPE_UD;
264 assert(!"not reached");
265 return BRW_REGISTER_TYPE_F;
270 brw_conditional_for_comparison(unsigned int op)
274 return BRW_CONDITIONAL_L;
275 case ir_binop_greater:
276 return BRW_CONDITIONAL_G;
277 case ir_binop_lequal:
278 return BRW_CONDITIONAL_LE;
279 case ir_binop_gequal:
280 return BRW_CONDITIONAL_GE;
282 case ir_binop_all_equal: /* same as equal for scalars */
283 return BRW_CONDITIONAL_Z;
284 case ir_binop_nequal:
285 case ir_binop_any_nequal: /* same as nequal for scalars */
286 return BRW_CONDITIONAL_NZ;
288 assert(!"not reached: bad operation for comparison");
289 return BRW_CONDITIONAL_NZ;
294 brw_math_function(enum opcode op)
297 case SHADER_OPCODE_RCP:
298 return BRW_MATH_FUNCTION_INV;
299 case SHADER_OPCODE_RSQ:
300 return BRW_MATH_FUNCTION_RSQ;
301 case SHADER_OPCODE_SQRT:
302 return BRW_MATH_FUNCTION_SQRT;
303 case SHADER_OPCODE_EXP2:
304 return BRW_MATH_FUNCTION_EXP;
305 case SHADER_OPCODE_LOG2:
306 return BRW_MATH_FUNCTION_LOG;
307 case SHADER_OPCODE_POW:
308 return BRW_MATH_FUNCTION_POW;
309 case SHADER_OPCODE_SIN:
310 return BRW_MATH_FUNCTION_SIN;
311 case SHADER_OPCODE_COS:
312 return BRW_MATH_FUNCTION_COS;
313 case SHADER_OPCODE_INT_QUOTIENT:
314 return BRW_MATH_FUNCTION_INT_DIV_QUOTIENT;
315 case SHADER_OPCODE_INT_REMAINDER:
316 return BRW_MATH_FUNCTION_INT_DIV_REMAINDER;
318 assert(!"not reached: unknown math function");
324 brw_texture_offset(ir_constant *offset)
326 assert(offset != NULL);
328 signed char offsets[3];
329 for (unsigned i = 0; i < offset->type->vector_elements; i++)
330 offsets[i] = (signed char) offset->value.i[i];
332 /* Combine all three offsets into a single unsigned dword:
334 * bits 11:8 - U Offset (X component)
335 * bits 7:4 - V Offset (Y component)
336 * bits 3:0 - R Offset (Z component)
338 unsigned offset_bits = 0;
339 for (unsigned i = 0; i < offset->type->vector_elements; i++) {
340 const unsigned shift = 4 * (2 - i);
341 offset_bits |= (offsets[i] << shift) & (0xF << shift);