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 gl_shader_program *prog = rzalloc(NULL, struct gl_shader_program);
54 _mesa_init_shader_program(ctx, prog);
60 * Performs a compile of the shader stages even when we don't know
61 * what non-orthogonal state will be set, in the hope that it reflects
62 * the eventual NOS used, and thus allows us to produce link failures.
65 brw_shader_precompile(struct gl_context *ctx, struct gl_shader_program *prog)
67 struct brw_context *brw = brw_context(ctx);
69 if (brw->precompile && !brw_fs_precompile(ctx, prog))
72 if (brw->precompile && !brw_vs_precompile(ctx, prog))
79 brw_link_shader(struct gl_context *ctx, struct gl_shader_program *shProg)
81 struct brw_context *brw = brw_context(ctx);
82 struct intel_context *intel = &brw->intel;
85 for (stage = 0; stage < ARRAY_SIZE(shProg->_LinkedShaders); stage++) {
86 struct brw_shader *shader =
87 (struct brw_shader *)shProg->_LinkedShaders[stage];
88 static const GLenum targets[] = {
89 GL_VERTEX_PROGRAM_ARB,
90 GL_FRAGMENT_PROGRAM_ARB,
91 GL_GEOMETRY_PROGRAM_NV
97 struct gl_program *prog =
98 ctx->Driver.NewProgram(ctx, targets[stage], shader->base.Name);
101 prog->Parameters = _mesa_new_parameter_list();
104 struct gl_vertex_program *vp = (struct gl_vertex_program *) prog;
105 vp->UsesClipDistance = shProg->Vert.UsesClipDistance;
108 void *mem_ctx = ralloc_context(NULL);
112 ralloc_free(shader->ir);
113 shader->ir = new(shader) exec_list;
114 clone_ir_list(mem_ctx, shader->ir, shader->base.ir);
116 do_mat_op_to_vec(shader->ir);
117 lower_instructions(shader->ir,
124 /* Pre-gen6 HW can only nest if-statements 16 deep. Beyond this,
125 * if-statements need to be flattened.
128 lower_if_to_cond_assign(shader->ir, 16);
130 do_lower_texture_projection(shader->ir);
131 brw_lower_texture_gradients(shader->ir);
132 do_vec_index_to_cond_assign(shader->ir);
133 brw_do_cubemap_normalize(shader->ir);
134 lower_noise(shader->ir);
135 lower_quadop_vector(shader->ir, false);
138 bool output = stage == MESA_SHADER_FRAGMENT;
139 bool temp = stage == MESA_SHADER_FRAGMENT;
140 bool uniform = false;
142 bool lowered_variable_indexing =
143 lower_variable_index_to_cond_assign(shader->ir,
144 input, output, temp, uniform);
146 if (unlikely((INTEL_DEBUG & DEBUG_PERF) && lowered_variable_indexing)) {
147 perf_debug("Unsupported form of variable indexing in FS; falling "
148 "back to very inefficient code generation\n");
151 /* FINISHME: Do this before the variable index lowering. */
152 lower_ubo_reference(&shader->base, shader->ir);
157 if (stage == MESA_SHADER_FRAGMENT) {
158 brw_do_channel_expressions(shader->ir);
159 brw_do_vector_splitting(shader->ir);
162 progress = do_lower_jumps(shader->ir, true, true,
163 true, /* main return */
164 false, /* continue */
168 progress = do_common_optimization(shader->ir, true, true, 32)
172 /* Make a pass over the IR to add state references for any built-in
173 * uniforms that are used. This has to be done now (during linking).
174 * Code generation doesn't happen until the first time this shader is
175 * used for rendering. Waiting until then to generate the parameters is
176 * too late. At that point, the values for the built-in uniforms won't
177 * get sent to the shader.
179 foreach_list(node, shader->ir) {
180 ir_variable *var = ((ir_instruction *) node)->as_variable();
182 if ((var == NULL) || (var->mode != ir_var_uniform)
183 || (strncmp(var->name, "gl_", 3) != 0))
186 const ir_state_slot *const slots = var->state_slots;
187 assert(var->state_slots != NULL);
189 for (unsigned int i = 0; i < var->num_state_slots; i++) {
190 _mesa_add_state_reference(prog->Parameters,
191 (gl_state_index *) slots[i].tokens);
195 validate_ir_tree(shader->ir);
197 reparent_ir(shader->ir, shader->ir);
198 ralloc_free(mem_ctx);
200 do_set_program_inouts(shader->ir, prog,
201 shader->base.Type == GL_FRAGMENT_SHADER);
203 prog->SamplersUsed = shader->base.active_samplers;
204 _mesa_update_shader_textures_used(shProg, prog);
206 _mesa_reference_program(ctx, &shader->base.Program, prog);
208 brw_add_texrect_params(prog);
210 /* This has to be done last. Any operation that can cause
211 * prog->ParameterValues to get reallocated (e.g., anything that adds a
212 * program constant) has to happen before creating this linkage.
214 _mesa_associate_uniform_storage(ctx, shProg, prog->Parameters);
216 _mesa_reference_program(ctx, &prog, NULL);
218 if (ctx->Shader.Flags & GLSL_DUMP) {
219 static const char *target_strings[]
220 = { "vertex", "fragment", "geometry" };
222 printf("GLSL IR for linked %s program %d:\n", target_strings[stage],
224 _mesa_print_ir(shader->base.ir, NULL);
228 if (!brw_shader_precompile(ctx, shProg))
236 brw_type_for_base_type(const struct glsl_type *type)
238 switch (type->base_type) {
239 case GLSL_TYPE_FLOAT:
240 return BRW_REGISTER_TYPE_F;
243 return BRW_REGISTER_TYPE_D;
245 return BRW_REGISTER_TYPE_UD;
246 case GLSL_TYPE_ARRAY:
247 return brw_type_for_base_type(type->fields.array);
248 case GLSL_TYPE_STRUCT:
249 case GLSL_TYPE_SAMPLER:
250 /* These should be overridden with the type of the member when
251 * dereferenced into. BRW_REGISTER_TYPE_UD seems like a likely
252 * way to trip up if we don't.
254 return BRW_REGISTER_TYPE_UD;
256 assert(!"not reached");
257 return BRW_REGISTER_TYPE_F;
262 brw_conditional_for_comparison(unsigned int op)
266 return BRW_CONDITIONAL_L;
267 case ir_binop_greater:
268 return BRW_CONDITIONAL_G;
269 case ir_binop_lequal:
270 return BRW_CONDITIONAL_LE;
271 case ir_binop_gequal:
272 return BRW_CONDITIONAL_GE;
274 case ir_binop_all_equal: /* same as equal for scalars */
275 return BRW_CONDITIONAL_Z;
276 case ir_binop_nequal:
277 case ir_binop_any_nequal: /* same as nequal for scalars */
278 return BRW_CONDITIONAL_NZ;
280 assert(!"not reached: bad operation for comparison");
281 return BRW_CONDITIONAL_NZ;
286 brw_math_function(enum opcode op)
289 case SHADER_OPCODE_RCP:
290 return BRW_MATH_FUNCTION_INV;
291 case SHADER_OPCODE_RSQ:
292 return BRW_MATH_FUNCTION_RSQ;
293 case SHADER_OPCODE_SQRT:
294 return BRW_MATH_FUNCTION_SQRT;
295 case SHADER_OPCODE_EXP2:
296 return BRW_MATH_FUNCTION_EXP;
297 case SHADER_OPCODE_LOG2:
298 return BRW_MATH_FUNCTION_LOG;
299 case SHADER_OPCODE_POW:
300 return BRW_MATH_FUNCTION_POW;
301 case SHADER_OPCODE_SIN:
302 return BRW_MATH_FUNCTION_SIN;
303 case SHADER_OPCODE_COS:
304 return BRW_MATH_FUNCTION_COS;
305 case SHADER_OPCODE_INT_QUOTIENT:
306 return BRW_MATH_FUNCTION_INT_DIV_QUOTIENT;
307 case SHADER_OPCODE_INT_REMAINDER:
308 return BRW_MATH_FUNCTION_INT_DIV_REMAINDER;
310 assert(!"not reached: unknown math function");
316 brw_texture_offset(ir_constant *offset)
318 assert(offset != NULL);
320 signed char offsets[3];
321 for (unsigned i = 0; i < offset->type->vector_elements; i++)
322 offsets[i] = (signed char) offset->value.i[i];
324 /* Combine all three offsets into a single unsigned dword:
326 * bits 11:8 - U Offset (X component)
327 * bits 7:4 - V Offset (Y component)
328 * bits 3:0 - R Offset (Z component)
330 unsigned offset_bits = 0;
331 for (unsigned i = 0; i < offset->type->vector_elements; i++) {
332 const unsigned shift = 4 * (2 - i);
333 offset_bits |= (offsets[i] << shift) & (0xF << shift);