2 * Copyright © 2011 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
28 #include "brw_shader.h"
29 #include "main/compiler.h"
30 #include "program/hash_table.h"
34 #include "brw_context.h"
45 swizzle_for_size(int size
);
48 ARF
= BRW_ARCHITECTURE_REGISTER_FILE
,
49 GRF
= BRW_GENERAL_REGISTER_FILE
,
50 MRF
= BRW_MESSAGE_REGISTER_FILE
,
51 IMM
= BRW_IMMEDIATE_VALUE
,
52 HW_REG
, /* a struct brw_reg */
54 UNIFORM
, /* prog_data->params[hw_reg] */
61 /** Register file: ARF, GRF, MRF, IMM. */
62 enum register_file file
;
63 /** virtual register number. 0 = fixed hw reg */
65 /** Offset within the virtual register. */
67 /** Register type. BRW_REGISTER_TYPE_* */
69 struct brw_reg fixed_hw_reg
;
71 /** Value for file == BRW_IMMMEDIATE_FILE */
79 class src_reg
: public reg
82 /* Callers of this ralloc-based new need not call delete. It's
83 * easier to just ralloc_free 'ctx' (or any of its ancestors). */
84 static void* operator new(size_t size
, void *ctx
)
88 node
= ralloc_size(ctx
, size
);
96 src_reg(register_file file
, int reg
, const glsl_type
*type
);
102 bool equals(src_reg
*r
);
103 bool is_zero() const;
106 src_reg(class vec4_visitor
*v
, const struct glsl_type
*type
);
108 explicit src_reg(dst_reg reg
);
110 GLuint swizzle
; /**< SWIZZLE_XYZW swizzles from Mesa. */
117 class dst_reg
: public reg
120 /* Callers of this ralloc-based new need not call delete. It's
121 * easier to just ralloc_free 'ctx' (or any of its ancestors). */
122 static void* operator new(size_t size
, void *ctx
)
126 node
= ralloc_size(ctx
, size
);
127 assert(node
!= NULL
);
135 dst_reg(register_file file
, int reg
);
136 dst_reg(register_file file
, int reg
, const glsl_type
*type
, int writemask
);
137 dst_reg(struct brw_reg reg
);
138 dst_reg(class vec4_visitor
*v
, const struct glsl_type
*type
);
140 explicit dst_reg(src_reg reg
);
142 int writemask
; /**< Bitfield of WRITEMASK_[XYZW] */
147 class vec4_instruction
: public exec_node
{
149 /* Callers of this ralloc-based new need not call delete. It's
150 * easier to just ralloc_free 'ctx' (or any of its ancestors). */
151 static void* operator new(size_t size
, void *ctx
)
155 node
= rzalloc_size(ctx
, size
);
156 assert(node
!= NULL
);
161 vec4_instruction(vec4_visitor
*v
, enum opcode opcode
,
162 dst_reg dst
= dst_reg(),
163 src_reg src0
= src_reg(),
164 src_reg src1
= src_reg(),
165 src_reg src2
= src_reg());
167 struct brw_reg
get_dst(void);
168 struct brw_reg
get_src(int i
);
170 enum opcode opcode
; /* BRW_OPCODE_* or FS_OPCODE_* */
175 bool predicate_inverse
;
178 int conditional_mod
; /**< BRW_CONDITIONAL_* */
181 uint32_t texture_offset
; /**< Texture Offset bitfield */
182 int target
; /**< MRT target. */
187 int mlen
; /**< SEND message length */
188 int base_mrf
; /**< First MRF in the SEND message, if mlen is nonzero. */
190 uint32_t offset
; /* spill/unspill offset */
192 * Annotation for the generated IR. One of the two can be set.
195 const char *annotation
;
201 class vec4_visitor
: public ir_visitor
204 vec4_visitor(struct brw_vs_compile
*c
,
205 struct gl_shader_program
*prog
, struct brw_shader
*shader
);
210 return dst_reg(brw_null_reg());
215 return dst_reg(retype(brw_null_reg(), BRW_REGISTER_TYPE_D
));
218 struct brw_context
*brw
;
219 const struct gl_vertex_program
*vp
;
220 struct intel_context
*intel
;
221 struct gl_context
*ctx
;
222 struct brw_vs_compile
*c
;
223 struct brw_vs_prog_data
*prog_data
;
224 struct brw_compile
*p
;
225 struct brw_shader
*shader
;
226 struct gl_shader_program
*prog
;
228 exec_list instructions
;
234 * GLSL IR currently being processed, which is associated with our
235 * driver IR instructions for debugging purposes.
237 ir_instruction
*base_ir
;
238 const char *current_annotation
;
240 int *virtual_grf_sizes
;
241 int virtual_grf_count
;
242 int virtual_grf_array_size
;
243 int first_non_payload_grf
;
244 unsigned int max_grf
;
245 int *virtual_grf_def
;
246 int *virtual_grf_use
;
247 dst_reg userplane
[MAX_CLIP_PLANES
];
250 * This is the size to be used for an array with an element per
253 int virtual_grf_reg_count
;
254 /** Per-virtual-grf indices into an array of size virtual_grf_reg_count */
255 int *virtual_grf_reg_map
;
257 bool live_intervals_valid
;
259 dst_reg
*variable_storage(ir_variable
*var
);
261 void reladdr_to_temp(ir_instruction
*ir
, src_reg
*reg
, int *num_reladdr
);
263 src_reg
src_reg_for_float(float val
);
266 * \name Visit methods
268 * As typical for the visitor pattern, there must be one \c visit method for
269 * each concrete subclass of \c ir_instruction. Virtual base classes within
270 * the hierarchy should not have \c visit methods.
273 virtual void visit(ir_variable
*);
274 virtual void visit(ir_loop
*);
275 virtual void visit(ir_loop_jump
*);
276 virtual void visit(ir_function_signature
*);
277 virtual void visit(ir_function
*);
278 virtual void visit(ir_expression
*);
279 virtual void visit(ir_swizzle
*);
280 virtual void visit(ir_dereference_variable
*);
281 virtual void visit(ir_dereference_array
*);
282 virtual void visit(ir_dereference_record
*);
283 virtual void visit(ir_assignment
*);
284 virtual void visit(ir_constant
*);
285 virtual void visit(ir_call
*);
286 virtual void visit(ir_return
*);
287 virtual void visit(ir_discard
*);
288 virtual void visit(ir_texture
*);
289 virtual void visit(ir_if
*);
294 /* Regs for vertex results. Generated at ir_variable visiting time
295 * for the ir->location's used.
297 dst_reg output_reg
[BRW_VERT_RESULT_MAX
];
298 const char *output_reg_annotation
[BRW_VERT_RESULT_MAX
];
299 int uniform_size
[MAX_UNIFORMS
];
300 int uniform_vector_size
[MAX_UNIFORMS
];
303 struct hash_table
*variable_ht
;
306 void fail(const char *msg
, ...);
308 int virtual_grf_alloc(int size
);
309 void setup_uniform_clipplane_values();
310 int setup_uniform_values(int loc
, const glsl_type
*type
);
311 void setup_builtin_uniform_values(ir_variable
*ir
);
312 int setup_attributes(int payload_reg
);
313 int setup_uniforms(int payload_reg
);
314 void setup_payload();
315 bool reg_allocate_trivial();
317 void evaluate_spill_costs(float *spill_costs
, bool *no_spill
);
318 int choose_spill_reg(struct ra_graph
*g
);
319 void spill_reg(int spill_reg
);
320 void move_grf_array_access_to_scratch();
321 void move_uniform_array_access_to_pull_constants();
322 void move_push_constants_to_pull_constants();
323 void split_uniform_registers();
324 void pack_uniform_registers();
325 void calculate_live_intervals();
326 void split_virtual_grfs();
327 bool dead_code_eliminate();
328 bool virtual_grf_interferes(int a
, int b
);
329 bool opt_copy_propagation();
330 bool opt_algebraic();
331 bool opt_compute_to_mrf();
333 vec4_instruction
*emit(vec4_instruction
*inst
);
335 vec4_instruction
*emit(enum opcode opcode
);
337 vec4_instruction
*emit(enum opcode opcode
, dst_reg dst
, src_reg src0
);
339 vec4_instruction
*emit(enum opcode opcode
, dst_reg dst
,
340 src_reg src0
, src_reg src1
);
342 vec4_instruction
*emit(enum opcode opcode
, dst_reg dst
,
343 src_reg src0
, src_reg src1
, src_reg src2
);
345 vec4_instruction
*emit_before(vec4_instruction
*inst
,
346 vec4_instruction
*new_inst
);
348 vec4_instruction
*MOV(dst_reg dst
, src_reg src0
);
349 vec4_instruction
*NOT(dst_reg dst
, src_reg src0
);
350 vec4_instruction
*RNDD(dst_reg dst
, src_reg src0
);
351 vec4_instruction
*RNDE(dst_reg dst
, src_reg src0
);
352 vec4_instruction
*RNDZ(dst_reg dst
, src_reg src0
);
353 vec4_instruction
*FRC(dst_reg dst
, src_reg src0
);
354 vec4_instruction
*ADD(dst_reg dst
, src_reg src0
, src_reg src1
);
355 vec4_instruction
*MUL(dst_reg dst
, src_reg src0
, src_reg src1
);
356 vec4_instruction
*MACH(dst_reg dst
, src_reg src0
, src_reg src1
);
357 vec4_instruction
*MAC(dst_reg dst
, src_reg src0
, src_reg src1
);
358 vec4_instruction
*AND(dst_reg dst
, src_reg src0
, src_reg src1
);
359 vec4_instruction
*OR(dst_reg dst
, src_reg src0
, src_reg src1
);
360 vec4_instruction
*XOR(dst_reg dst
, src_reg src0
, src_reg src1
);
361 vec4_instruction
*DP3(dst_reg dst
, src_reg src0
, src_reg src1
);
362 vec4_instruction
*DP4(dst_reg dst
, src_reg src0
, src_reg src1
);
363 vec4_instruction
*CMP(dst_reg dst
, src_reg src0
, src_reg src1
,
365 vec4_instruction
*IF(src_reg src0
, src_reg src1
, uint32_t condition
);
366 vec4_instruction
*IF(uint32_t predicate
);
367 vec4_instruction
*PULL_CONSTANT_LOAD(dst_reg dst
, src_reg index
);
368 vec4_instruction
*SCRATCH_READ(dst_reg dst
, src_reg index
);
369 vec4_instruction
*SCRATCH_WRITE(dst_reg dst
, src_reg src
, src_reg index
);
371 int implied_mrf_writes(vec4_instruction
*inst
);
373 bool try_rewrite_rhs_to_dst(ir_assignment
*ir
,
376 vec4_instruction
*pre_rhs_inst
,
377 vec4_instruction
*last_rhs_inst
);
379 /** Walks an exec_list of ir_instruction and sends it through this visitor. */
380 void visit_instructions(const exec_list
*list
);
382 void emit_bool_to_cond_code(ir_rvalue
*ir
, uint32_t *predicate
);
383 void emit_bool_comparison(unsigned int op
, dst_reg dst
, src_reg src0
, src_reg src1
);
384 void emit_if_gen6(ir_if
*ir
);
386 void emit_block_move(dst_reg
*dst
, src_reg
*src
,
387 const struct glsl_type
*type
, uint32_t predicate
);
389 void emit_constant_values(dst_reg
*dst
, ir_constant
*value
);
392 * Emit the correct dot-product instruction for the type of arguments
394 void emit_dp(dst_reg dst
, src_reg src0
, src_reg src1
, unsigned elements
);
396 void emit_scalar(ir_instruction
*ir
, enum prog_opcode op
,
397 dst_reg dst
, src_reg src0
);
399 void emit_scalar(ir_instruction
*ir
, enum prog_opcode op
,
400 dst_reg dst
, src_reg src0
, src_reg src1
);
402 void emit_scs(ir_instruction
*ir
, enum prog_opcode op
,
403 dst_reg dst
, const src_reg
&src
);
405 void emit_math1_gen6(enum opcode opcode
, dst_reg dst
, src_reg src
);
406 void emit_math1_gen4(enum opcode opcode
, dst_reg dst
, src_reg src
);
407 void emit_math(enum opcode opcode
, dst_reg dst
, src_reg src
);
408 void emit_math2_gen6(enum opcode opcode
, dst_reg dst
, src_reg src0
, src_reg src1
);
409 void emit_math2_gen4(enum opcode opcode
, dst_reg dst
, src_reg src0
, src_reg src1
);
410 void emit_math(enum opcode opcode
, dst_reg dst
, src_reg src0
, src_reg src1
);
412 void swizzle_result(ir_texture
*ir
, src_reg orig_val
, int sampler
);
414 void emit_ndc_computation();
415 void emit_psiz_and_flags(struct brw_reg reg
);
416 void emit_clip_distances(struct brw_reg reg
, int offset
);
417 void emit_generic_urb_slot(dst_reg reg
, int vert_result
);
418 void emit_urb_slot(int mrf
, int vert_result
);
419 void emit_urb_writes(void);
421 src_reg
get_scratch_offset(vec4_instruction
*inst
,
422 src_reg
*reladdr
, int reg_offset
);
423 src_reg
get_pull_constant_offset(vec4_instruction
*inst
,
424 src_reg
*reladdr
, int reg_offset
);
425 void emit_scratch_read(vec4_instruction
*inst
,
429 void emit_scratch_write(vec4_instruction
*inst
,
433 void emit_pull_constant_load(vec4_instruction
*inst
,
438 bool try_emit_sat(ir_expression
*ir
);
439 void resolve_ud_negate(src_reg
*reg
);
441 bool process_move_condition(ir_rvalue
*ir
);
443 void generate_code();
444 void generate_vs_instruction(vec4_instruction
*inst
,
446 struct brw_reg
*src
);
448 void generate_math1_gen4(vec4_instruction
*inst
,
451 void generate_math1_gen6(vec4_instruction
*inst
,
454 void generate_math2_gen4(vec4_instruction
*inst
,
457 struct brw_reg src1
);
458 void generate_math2_gen6(vec4_instruction
*inst
,
461 struct brw_reg src1
);
462 void generate_math2_gen7(vec4_instruction
*inst
,
465 struct brw_reg src1
);
467 void generate_tex(vec4_instruction
*inst
,
471 void generate_urb_write(vec4_instruction
*inst
);
472 void generate_oword_dual_block_offsets(struct brw_reg m1
,
473 struct brw_reg index
);
474 void generate_scratch_write(vec4_instruction
*inst
,
477 struct brw_reg index
);
478 void generate_scratch_read(vec4_instruction
*inst
,
480 struct brw_reg index
);
481 void generate_pull_constant_load(vec4_instruction
*inst
,
483 struct brw_reg index
,
484 struct brw_reg offset
);
486 void dump_instruction(vec4_instruction
*inst
);
487 void dump_instructions();
490 } /* namespace brw */
492 #endif /* BRW_VEC4_H */