brw_fs.h revision f9995b30756140724f41daf963fa06167912be7f 
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 * Authors:
24 *    Eric Anholt <eric@anholt.net>
25 *
26 */
27
28extern "C" {
29
30#include <sys/types.h>
31
32#include "main/macros.h"
33#include "main/shaderobj.h"
34#include "main/uniforms.h"
35#include "program/prog_parameter.h"
36#include "program/prog_print.h"
37#include "program/prog_optimize.h"
38#include "program/register_allocate.h"
39#include "program/sampler.h"
40#include "program/hash_table.h"
41#include "brw_context.h"
42#include "brw_eu.h"
43#include "brw_wm.h"
44#include "talloc.h"
45}
46#include "../glsl/glsl_types.h"
47#include "../glsl/ir.h"
48
49enum register_file {
50   ARF = BRW_ARCHITECTURE_REGISTER_FILE,
51   GRF = BRW_GENERAL_REGISTER_FILE,
52   MRF = BRW_MESSAGE_REGISTER_FILE,
53   IMM = BRW_IMMEDIATE_VALUE,
54   FIXED_HW_REG, /* a struct brw_reg */
55   UNIFORM, /* prog_data->params[hw_reg] */
56   BAD_FILE
57};
58
59enum fs_opcodes {
60   FS_OPCODE_FB_WRITE = 256,
61   FS_OPCODE_RCP,
62   FS_OPCODE_RSQ,
63   FS_OPCODE_SQRT,
64   FS_OPCODE_EXP2,
65   FS_OPCODE_LOG2,
66   FS_OPCODE_POW,
67   FS_OPCODE_SIN,
68   FS_OPCODE_COS,
69   FS_OPCODE_DDX,
70   FS_OPCODE_DDY,
71   FS_OPCODE_LINTERP,
72   FS_OPCODE_TEX,
73   FS_OPCODE_TXB,
74   FS_OPCODE_TXL,
75   FS_OPCODE_DISCARD_NOT,
76   FS_OPCODE_DISCARD_AND,
77};
78
79
80class fs_reg {
81public:
82   /* Callers of this talloc-based new need not call delete. It's
83    * easier to just talloc_free 'ctx' (or any of its ancestors). */
84   static void* operator new(size_t size, void *ctx)
85   {
86      void *node;
87
88      node = talloc_size(ctx, size);
89      assert(node != NULL);
90
91      return node;
92   }
93
94   void init()
95   {
96      this->reg = 0;
97      this->reg_offset = 0;
98      this->negate = 0;
99      this->abs = 0;
100      this->hw_reg = -1;
101   }
102
103   /** Generic unset register constructor. */
104   fs_reg()
105   {
106      init();
107      this->file = BAD_FILE;
108   }
109
110   /** Immediate value constructor. */
111   fs_reg(float f)
112   {
113      init();
114      this->file = IMM;
115      this->type = BRW_REGISTER_TYPE_F;
116      this->imm.f = f;
117   }
118
119   /** Immediate value constructor. */
120   fs_reg(int32_t i)
121   {
122      init();
123      this->file = IMM;
124      this->type = BRW_REGISTER_TYPE_D;
125      this->imm.i = i;
126   }
127
128   /** Immediate value constructor. */
129   fs_reg(uint32_t u)
130   {
131      init();
132      this->file = IMM;
133      this->type = BRW_REGISTER_TYPE_UD;
134      this->imm.u = u;
135   }
136
137   /** Fixed brw_reg Immediate value constructor. */
138   fs_reg(struct brw_reg fixed_hw_reg)
139   {
140      init();
141      this->file = FIXED_HW_REG;
142      this->fixed_hw_reg = fixed_hw_reg;
143      this->type = fixed_hw_reg.type;
144   }
145
146   fs_reg(enum register_file file, int hw_reg);
147   fs_reg(class fs_visitor *v, const struct glsl_type *type);
148
149   /** Register file: ARF, GRF, MRF, IMM. */
150   enum register_file file;
151   /** virtual register number.  0 = fixed hw reg */
152   int reg;
153   /** Offset within the virtual register. */
154   int reg_offset;
155   /** HW register number.  Generally unset until register allocation. */
156   int hw_reg;
157   /** Register type.  BRW_REGISTER_TYPE_* */
158   int type;
159   bool negate;
160   bool abs;
161   struct brw_reg fixed_hw_reg;
162
163   /** Value for file == BRW_IMMMEDIATE_FILE */
164   union {
165      int32_t i;
166      uint32_t u;
167      float f;
168   } imm;
169};
170
171class fs_inst : public exec_node {
172public:
173   /* Callers of this talloc-based new need not call delete. It's
174    * easier to just talloc_free 'ctx' (or any of its ancestors). */
175   static void* operator new(size_t size, void *ctx)
176   {
177      void *node;
178
179      node = talloc_zero_size(ctx, size);
180      assert(node != NULL);
181
182      return node;
183   }
184
185   void init()
186   {
187      this->opcode = BRW_OPCODE_NOP;
188      this->saturate = false;
189      this->conditional_mod = BRW_CONDITIONAL_NONE;
190      this->predicated = false;
191      this->sampler = 0;
192      this->target = 0;
193      this->eot = false;
194      this->header_present = false;
195      this->shadow_compare = false;
196      this->mlen = 0;
197      this->base_mrf = 0;
198   }
199
200   fs_inst()
201   {
202      init();
203   }
204
205   fs_inst(int opcode)
206   {
207      init();
208      this->opcode = opcode;
209   }
210
211   fs_inst(int opcode, fs_reg dst)
212   {
213      init();
214      this->opcode = opcode;
215      this->dst = dst;
216   }
217
218   fs_inst(int opcode, fs_reg dst, fs_reg src0)
219   {
220      init();
221      this->opcode = opcode;
222      this->dst = dst;
223      this->src[0] = src0;
224   }
225
226   fs_inst(int opcode, fs_reg dst, fs_reg src0, fs_reg src1)
227   {
228      init();
229      this->opcode = opcode;
230      this->dst = dst;
231      this->src[0] = src0;
232      this->src[1] = src1;
233   }
234
235   fs_inst(int opcode, fs_reg dst, fs_reg src0, fs_reg src1, fs_reg src2)
236   {
237      init();
238      this->opcode = opcode;
239      this->dst = dst;
240      this->src[0] = src0;
241      this->src[1] = src1;
242      this->src[2] = src2;
243   }
244
245   int opcode; /* BRW_OPCODE_* or FS_OPCODE_* */
246   fs_reg dst;
247   fs_reg src[3];
248   bool saturate;
249   bool predicated;
250   int conditional_mod; /**< BRW_CONDITIONAL_* */
251
252   int mlen; /**< SEND message length */
253   int base_mrf; /**< First MRF in the SEND message, if mlen is nonzero. */
254   int sampler;
255   int target; /**< MRT target. */
256   bool eot;
257   bool header_present;
258   bool shadow_compare;
259
260   /** @{
261    * Annotation for the generated IR.  One of the two can be set.
262    */
263   ir_instruction *ir;
264   const char *annotation;
265   /** @} */
266};
267
268class fs_visitor : public ir_visitor
269{
270public:
271
272   fs_visitor(struct brw_wm_compile *c, struct brw_shader *shader)
273   {
274      this->c = c;
275      this->p = &c->func;
276      this->brw = p->brw;
277      this->fp = brw->fragment_program;
278      this->intel = &brw->intel;
279      this->ctx = &intel->ctx;
280      this->mem_ctx = talloc_new(NULL);
281      this->shader = shader;
282      this->fail = false;
283      this->variable_ht = hash_table_ctor(0,
284					  hash_table_pointer_hash,
285					  hash_table_pointer_compare);
286
287      this->frag_color = NULL;
288      this->frag_data = NULL;
289      this->frag_depth = NULL;
290      this->first_non_payload_grf = 0;
291
292      this->current_annotation = NULL;
293      this->annotation_string = NULL;
294      this->annotation_ir = NULL;
295      this->base_ir = NULL;
296
297      this->virtual_grf_sizes = NULL;
298      this->virtual_grf_next = 1;
299      this->virtual_grf_array_size = 0;
300      this->virtual_grf_def = NULL;
301      this->virtual_grf_use = NULL;
302
303      this->kill_emitted = false;
304   }
305
306   ~fs_visitor()
307   {
308      talloc_free(this->mem_ctx);
309      hash_table_dtor(this->variable_ht);
310   }
311
312   fs_reg *variable_storage(ir_variable *var);
313   int virtual_grf_alloc(int size);
314
315   void visit(ir_variable *ir);
316   void visit(ir_assignment *ir);
317   void visit(ir_dereference_variable *ir);
318   void visit(ir_dereference_record *ir);
319   void visit(ir_dereference_array *ir);
320   void visit(ir_expression *ir);
321   void visit(ir_texture *ir);
322   void visit(ir_if *ir);
323   void visit(ir_constant *ir);
324   void visit(ir_swizzle *ir);
325   void visit(ir_return *ir);
326   void visit(ir_loop *ir);
327   void visit(ir_loop_jump *ir);
328   void visit(ir_discard *ir);
329   void visit(ir_call *ir);
330   void visit(ir_function *ir);
331   void visit(ir_function_signature *ir);
332
333   fs_inst *emit(fs_inst inst);
334   void assign_curb_setup();
335   void calculate_urb_setup();
336   void assign_urb_setup();
337   void assign_regs();
338   void assign_regs_trivial();
339   void calculate_live_intervals();
340   bool propagate_constants();
341   bool register_coalesce();
342   bool compute_to_mrf();
343   bool dead_code_eliminate();
344   bool virtual_grf_interferes(int a, int b);
345   void generate_code();
346   void generate_fb_write(fs_inst *inst);
347   void generate_linterp(fs_inst *inst, struct brw_reg dst,
348			 struct brw_reg *src);
349   void generate_tex(fs_inst *inst, struct brw_reg dst);
350   void generate_math(fs_inst *inst, struct brw_reg dst, struct brw_reg *src);
351   void generate_discard_not(fs_inst *inst, struct brw_reg temp);
352   void generate_discard_and(fs_inst *inst, struct brw_reg temp);
353   void generate_ddx(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
354   void generate_ddy(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
355
356   void emit_dummy_fs();
357   fs_reg *emit_fragcoord_interpolation(ir_variable *ir);
358   fs_reg *emit_frontfacing_interpolation(ir_variable *ir);
359   fs_reg *emit_general_interpolation(ir_variable *ir);
360   void emit_interpolation_setup_gen4();
361   void emit_interpolation_setup_gen6();
362   fs_inst *emit_texture_gen4(ir_texture *ir, fs_reg dst, fs_reg coordinate);
363   fs_inst *emit_texture_gen5(ir_texture *ir, fs_reg dst, fs_reg coordinate);
364   fs_inst *emit_math(fs_opcodes op, fs_reg dst, fs_reg src0);
365   fs_inst *emit_math(fs_opcodes op, fs_reg dst, fs_reg src0, fs_reg src1);
366
367   void emit_fb_writes();
368   void emit_assignment_writes(fs_reg &l, fs_reg &r,
369			       const glsl_type *type, bool predicated);
370
371   struct brw_reg interp_reg(int location, int channel);
372   int setup_uniform_values(int loc, const glsl_type *type);
373   void setup_builtin_uniform_values(ir_variable *ir);
374
375   struct brw_context *brw;
376   const struct gl_fragment_program *fp;
377   struct intel_context *intel;
378   struct gl_context *ctx;
379   struct brw_wm_compile *c;
380   struct brw_compile *p;
381   struct brw_shader *shader;
382   void *mem_ctx;
383   exec_list instructions;
384
385   int *virtual_grf_sizes;
386   int virtual_grf_next;
387   int virtual_grf_array_size;
388   int *virtual_grf_def;
389   int *virtual_grf_use;
390
391   struct hash_table *variable_ht;
392   ir_variable *frag_color, *frag_data, *frag_depth;
393   int first_non_payload_grf;
394   int urb_setup[FRAG_ATTRIB_MAX];
395   bool kill_emitted;
396
397   /** @{ debug annotation info */
398   const char *current_annotation;
399   ir_instruction *base_ir;
400   const char **annotation_string;
401   ir_instruction **annotation_ir;
402   /** @} */
403
404   bool fail;
405
406   /* Result of last visit() method. */
407   fs_reg result;
408
409   fs_reg pixel_x;
410   fs_reg pixel_y;
411   fs_reg wpos_w;
412   fs_reg pixel_w;
413   fs_reg delta_x;
414   fs_reg delta_y;
415
416   int grf_used;
417};
418
419GLboolean brw_do_channel_expressions(struct exec_list *instructions);
420GLboolean brw_do_vector_splitting(struct exec_list *instructions);
421