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"
31 #include "brw_program.h"
37 #include "brw_context.h"
47 struct brw_vec4_compile
{
48 GLuint last_scratch
; /**< measured in 32-byte (register size) units */
52 struct brw_vec4_prog_key
{
53 GLuint program_string_id
;
56 * True if at least one clip flag is enabled, regardless of whether the
57 * shader uses clip planes or gl_ClipDistance.
59 GLuint userclip_active
:1;
62 * How many user clipping planes are being uploaded to the vertex shader as
65 GLuint nr_userclip_plane_consts
:4;
67 GLuint clamp_vertex_color
:1;
69 struct brw_sampler_prog_key_data tex
;
78 brw_vec4_setup_prog_key_for_precompile(struct gl_context
*ctx
,
79 struct brw_vec4_prog_key
*key
,
80 GLuint id
, struct gl_program
*prog
);
81 bool brw_vec4_prog_data_compare(const struct brw_vec4_prog_data
*a
,
82 const struct brw_vec4_prog_data
*b
);
83 void brw_vec4_prog_data_free(const struct brw_vec4_prog_data
*prog_data
);
93 swizzle_for_size(int size
);
98 /** Register file: GRF, MRF, IMM. */
99 enum register_file file
;
100 /** virtual register number. 0 = fixed hw reg */
102 /** Offset within the virtual register. */
104 /** Register type. BRW_REGISTER_TYPE_* */
106 struct brw_reg fixed_hw_reg
;
108 /** Value for file == BRW_IMMMEDIATE_FILE */
116 class src_reg
: public reg
119 DECLARE_RALLOC_CXX_OPERATORS(src_reg
)
123 src_reg(register_file file
, int reg
, const glsl_type
*type
);
129 bool equals(src_reg
*r
);
130 bool is_zero() const;
133 src_reg(class vec4_visitor
*v
, const struct glsl_type
*type
);
135 explicit src_reg(dst_reg reg
);
137 GLuint swizzle
; /**< SWIZZLE_XYZW swizzles from Mesa. */
144 class dst_reg
: public reg
147 DECLARE_RALLOC_CXX_OPERATORS(dst_reg
)
152 dst_reg(register_file file
, int reg
);
153 dst_reg(register_file file
, int reg
, const glsl_type
*type
, int writemask
);
154 dst_reg(struct brw_reg reg
);
155 dst_reg(class vec4_visitor
*v
, const struct glsl_type
*type
);
157 explicit dst_reg(src_reg reg
);
159 int writemask
; /**< Bitfield of WRITEMASK_[XYZW] */
165 with_writemask(dst_reg
const &r
, int mask
);
167 class vec4_instruction
: public backend_instruction
{
169 DECLARE_RALLOC_CXX_OPERATORS(vec4_instruction
)
171 vec4_instruction(vec4_visitor
*v
, enum opcode opcode
,
172 dst_reg dst
= dst_reg(),
173 src_reg src0
= src_reg(),
174 src_reg src1
= src_reg(),
175 src_reg src2
= src_reg());
177 struct brw_reg
get_dst(void);
178 struct brw_reg
get_src(const struct brw_vec4_prog_data
*prog_data
, int i
);
184 bool force_writemask_all
;
185 bool no_dd_clear
, no_dd_check
;
187 int conditional_mod
; /**< BRW_CONDITIONAL_* */
190 uint32_t texture_offset
; /**< Texture Offset bitfield */
191 int target
; /**< MRT target. */
194 enum brw_urb_write_flags urb_write_flags
;
196 int mlen
; /**< SEND message length */
197 int base_mrf
; /**< First MRF in the SEND message, if mlen is nonzero. */
199 uint32_t offset
; /* spill/unspill offset */
201 * Annotation for the generated IR. One of the two can be set.
204 const char *annotation
;
206 bool is_send_from_grf();
207 bool can_reswizzle_dst(int dst_writemask
, int swizzle
, int swizzle_mask
);
208 void reswizzle_dst(int dst_writemask
, int swizzle
);
210 bool depends_on_flags()
212 return predicate
|| opcode
== VS_OPCODE_UNPACK_FLAGS_SIMD4X2
;
217 * The vertex shader front-end.
219 * Translates either GLSL IR or Mesa IR (for ARB_vertex_program and
220 * fixed-function) into VS IR.
222 class vec4_visitor
: public backend_visitor
225 vec4_visitor(struct brw_context
*brw
,
226 struct brw_vec4_compile
*c
,
227 struct gl_program
*prog
,
228 const struct brw_vec4_prog_key
*key
,
229 struct brw_vec4_prog_data
*prog_data
,
230 struct gl_shader_program
*shader_prog
,
231 struct brw_shader
*shader
,
239 return dst_reg(brw_null_reg());
244 return dst_reg(retype(brw_null_reg(), BRW_REGISTER_TYPE_D
));
247 dst_reg
dst_null_ud()
249 return dst_reg(retype(brw_null_reg(), BRW_REGISTER_TYPE_UD
));
252 struct brw_vec4_compile
*c
;
253 const struct brw_vec4_prog_key
*key
;
254 struct brw_vec4_prog_data
*prog_data
;
255 unsigned int sanity_param_count
;
261 * GLSL IR currently being processed, which is associated with our
262 * driver IR instructions for debugging purposes.
265 const char *current_annotation
;
267 int *virtual_grf_sizes
;
268 int virtual_grf_count
;
269 int virtual_grf_array_size
;
270 int first_non_payload_grf
;
271 unsigned int max_grf
;
272 int *virtual_grf_start
;
273 int *virtual_grf_end
;
274 dst_reg userplane
[MAX_CLIP_PLANES
];
277 * This is the size to be used for an array with an element per
280 int virtual_grf_reg_count
;
281 /** Per-virtual-grf indices into an array of size virtual_grf_reg_count */
282 int *virtual_grf_reg_map
;
284 bool live_intervals_valid
;
286 dst_reg
*variable_storage(ir_variable
*var
);
288 void reladdr_to_temp(ir_instruction
*ir
, src_reg
*reg
, int *num_reladdr
);
290 bool need_all_constants_in_pull_buffer
;
293 * \name Visit methods
295 * As typical for the visitor pattern, there must be one \c visit method for
296 * each concrete subclass of \c ir_instruction. Virtual base classes within
297 * the hierarchy should not have \c visit methods.
300 virtual void visit(ir_variable
*);
301 virtual void visit(ir_loop
*);
302 virtual void visit(ir_loop_jump
*);
303 virtual void visit(ir_function_signature
*);
304 virtual void visit(ir_function
*);
305 virtual void visit(ir_expression
*);
306 virtual void visit(ir_swizzle
*);
307 virtual void visit(ir_dereference_variable
*);
308 virtual void visit(ir_dereference_array
*);
309 virtual void visit(ir_dereference_record
*);
310 virtual void visit(ir_assignment
*);
311 virtual void visit(ir_constant
*);
312 virtual void visit(ir_call
*);
313 virtual void visit(ir_return
*);
314 virtual void visit(ir_discard
*);
315 virtual void visit(ir_texture
*);
316 virtual void visit(ir_if
*);
317 virtual void visit(ir_emit_vertex
*);
318 virtual void visit(ir_end_primitive
*);
323 /* Regs for vertex results. Generated at ir_variable visiting time
324 * for the ir->location's used.
326 dst_reg output_reg
[BRW_VARYING_SLOT_COUNT
];
327 const char *output_reg_annotation
[BRW_VARYING_SLOT_COUNT
];
328 int uniform_size
[MAX_UNIFORMS
];
329 int uniform_vector_size
[MAX_UNIFORMS
];
332 src_reg shader_start_time
;
334 struct hash_table
*variable_ht
;
337 void fail(const char *msg
, ...);
339 int virtual_grf_alloc(int size
);
340 void setup_uniform_clipplane_values();
341 void setup_uniform_values(ir_variable
*ir
);
342 void setup_builtin_uniform_values(ir_variable
*ir
);
343 int setup_uniforms(int payload_reg
);
344 bool reg_allocate_trivial();
346 void evaluate_spill_costs(float *spill_costs
, bool *no_spill
);
347 int choose_spill_reg(struct ra_graph
*g
);
348 void spill_reg(int spill_reg
);
349 void move_grf_array_access_to_scratch();
350 void move_uniform_array_access_to_pull_constants();
351 void move_push_constants_to_pull_constants();
352 void split_uniform_registers();
353 void pack_uniform_registers();
354 void calculate_live_intervals();
355 void split_virtual_grfs();
356 bool dead_code_eliminate();
357 bool virtual_grf_interferes(int a
, int b
);
358 bool opt_copy_propagation();
359 bool opt_algebraic();
360 bool opt_register_coalesce();
361 void opt_set_dependency_control();
362 void opt_schedule_instructions();
364 bool can_do_source_mods(vec4_instruction
*inst
);
366 vec4_instruction
*emit(vec4_instruction
*inst
);
368 vec4_instruction
*emit(enum opcode opcode
);
370 vec4_instruction
*emit(enum opcode opcode
, dst_reg dst
);
372 vec4_instruction
*emit(enum opcode opcode
, dst_reg dst
, src_reg src0
);
374 vec4_instruction
*emit(enum opcode opcode
, dst_reg dst
,
375 src_reg src0
, src_reg src1
);
377 vec4_instruction
*emit(enum opcode opcode
, dst_reg dst
,
378 src_reg src0
, src_reg src1
, src_reg src2
);
380 vec4_instruction
*emit_before(vec4_instruction
*inst
,
381 vec4_instruction
*new_inst
);
383 vec4_instruction
*MOV(dst_reg dst
, src_reg src0
);
384 vec4_instruction
*NOT(dst_reg dst
, src_reg src0
);
385 vec4_instruction
*RNDD(dst_reg dst
, src_reg src0
);
386 vec4_instruction
*RNDE(dst_reg dst
, src_reg src0
);
387 vec4_instruction
*RNDZ(dst_reg dst
, src_reg src0
);
388 vec4_instruction
*FRC(dst_reg dst
, src_reg src0
);
389 vec4_instruction
*F32TO16(dst_reg dst
, src_reg src0
);
390 vec4_instruction
*F16TO32(dst_reg dst
, src_reg src0
);
391 vec4_instruction
*ADD(dst_reg dst
, src_reg src0
, src_reg src1
);
392 vec4_instruction
*MUL(dst_reg dst
, src_reg src0
, src_reg src1
);
393 vec4_instruction
*MACH(dst_reg dst
, src_reg src0
, src_reg src1
);
394 vec4_instruction
*MAC(dst_reg dst
, src_reg src0
, src_reg src1
);
395 vec4_instruction
*AND(dst_reg dst
, src_reg src0
, src_reg src1
);
396 vec4_instruction
*OR(dst_reg dst
, src_reg src0
, src_reg src1
);
397 vec4_instruction
*XOR(dst_reg dst
, src_reg src0
, src_reg src1
);
398 vec4_instruction
*DP3(dst_reg dst
, src_reg src0
, src_reg src1
);
399 vec4_instruction
*DP4(dst_reg dst
, src_reg src0
, src_reg src1
);
400 vec4_instruction
*DPH(dst_reg dst
, src_reg src0
, src_reg src1
);
401 vec4_instruction
*SHL(dst_reg dst
, src_reg src0
, src_reg src1
);
402 vec4_instruction
*SHR(dst_reg dst
, src_reg src0
, src_reg src1
);
403 vec4_instruction
*ASR(dst_reg dst
, src_reg src0
, src_reg src1
);
404 vec4_instruction
*CMP(dst_reg dst
, src_reg src0
, src_reg src1
,
406 vec4_instruction
*IF(src_reg src0
, src_reg src1
, uint32_t condition
);
407 vec4_instruction
*IF(uint32_t predicate
);
408 vec4_instruction
*PULL_CONSTANT_LOAD(dst_reg dst
, src_reg index
);
409 vec4_instruction
*SCRATCH_READ(dst_reg dst
, src_reg index
);
410 vec4_instruction
*SCRATCH_WRITE(dst_reg dst
, src_reg src
, src_reg index
);
411 vec4_instruction
*LRP(dst_reg dst
, src_reg a
, src_reg y
, src_reg x
);
412 vec4_instruction
*BFREV(dst_reg dst
, src_reg value
);
413 vec4_instruction
*BFE(dst_reg dst
, src_reg bits
, src_reg offset
, src_reg value
);
414 vec4_instruction
*BFI1(dst_reg dst
, src_reg bits
, src_reg offset
);
415 vec4_instruction
*BFI2(dst_reg dst
, src_reg bfi1_dst
, src_reg insert
, src_reg base
);
416 vec4_instruction
*FBH(dst_reg dst
, src_reg value
);
417 vec4_instruction
*FBL(dst_reg dst
, src_reg value
);
418 vec4_instruction
*CBIT(dst_reg dst
, src_reg value
);
419 vec4_instruction
*MAD(dst_reg dst
, src_reg c
, src_reg b
, src_reg a
);
420 vec4_instruction
*ADDC(dst_reg dst
, src_reg src0
, src_reg src1
);
421 vec4_instruction
*SUBB(dst_reg dst
, src_reg src0
, src_reg src1
);
423 int implied_mrf_writes(vec4_instruction
*inst
);
425 bool try_rewrite_rhs_to_dst(ir_assignment
*ir
,
428 vec4_instruction
*pre_rhs_inst
,
429 vec4_instruction
*last_rhs_inst
);
431 bool try_copy_propagation(vec4_instruction
*inst
, int arg
,
434 /** Walks an exec_list of ir_instruction and sends it through this visitor. */
435 void visit_instructions(const exec_list
*list
);
437 void emit_vp_sop(uint32_t condmod
, dst_reg dst
,
438 src_reg src0
, src_reg src1
, src_reg one
);
440 void emit_bool_to_cond_code(ir_rvalue
*ir
, uint32_t *predicate
);
441 void emit_bool_comparison(unsigned int op
, dst_reg dst
, src_reg src0
, src_reg src1
);
442 void emit_if_gen6(ir_if
*ir
);
444 void emit_minmax(uint32_t condmod
, dst_reg dst
, src_reg src0
, src_reg src1
);
446 void emit_block_move(dst_reg
*dst
, src_reg
*src
,
447 const struct glsl_type
*type
, uint32_t predicate
);
449 void emit_constant_values(dst_reg
*dst
, ir_constant
*value
);
452 * Emit the correct dot-product instruction for the type of arguments
454 void emit_dp(dst_reg dst
, src_reg src0
, src_reg src1
, unsigned elements
);
456 void emit_scalar(ir_instruction
*ir
, enum prog_opcode op
,
457 dst_reg dst
, src_reg src0
);
459 void emit_scalar(ir_instruction
*ir
, enum prog_opcode op
,
460 dst_reg dst
, src_reg src0
, src_reg src1
);
462 void emit_scs(ir_instruction
*ir
, enum prog_opcode op
,
463 dst_reg dst
, const src_reg
&src
);
465 src_reg
fix_3src_operand(src_reg src
);
467 void emit_math1_gen6(enum opcode opcode
, dst_reg dst
, src_reg src
);
468 void emit_math1_gen4(enum opcode opcode
, dst_reg dst
, src_reg src
);
469 void emit_math(enum opcode opcode
, dst_reg dst
, src_reg src
);
470 void emit_math2_gen6(enum opcode opcode
, dst_reg dst
, src_reg src0
, src_reg src1
);
471 void emit_math2_gen4(enum opcode opcode
, dst_reg dst
, src_reg src0
, src_reg src1
);
472 void emit_math(enum opcode opcode
, dst_reg dst
, src_reg src0
, src_reg src1
);
473 src_reg
fix_math_operand(src_reg src
);
475 void emit_pack_half_2x16(dst_reg dst
, src_reg src0
);
476 void emit_unpack_half_2x16(dst_reg dst
, src_reg src0
);
478 uint32_t gather_channel(ir_texture
*ir
, int sampler
);
479 void swizzle_result(ir_texture
*ir
, src_reg orig_val
, int sampler
);
481 void emit_ndc_computation();
482 void emit_psiz_and_flags(struct brw_reg reg
);
483 void emit_clip_distances(dst_reg reg
, int offset
);
484 void emit_generic_urb_slot(dst_reg reg
, int varying
);
485 void emit_urb_slot(int mrf
, int varying
);
487 void emit_shader_time_begin();
488 void emit_shader_time_end();
489 void emit_shader_time_write(enum shader_time_shader_type type
,
492 src_reg
get_scratch_offset(vec4_instruction
*inst
,
493 src_reg
*reladdr
, int reg_offset
);
494 src_reg
get_pull_constant_offset(vec4_instruction
*inst
,
495 src_reg
*reladdr
, int reg_offset
);
496 void emit_scratch_read(vec4_instruction
*inst
,
500 void emit_scratch_write(vec4_instruction
*inst
,
502 void emit_pull_constant_load(vec4_instruction
*inst
,
507 bool try_emit_sat(ir_expression
*ir
);
508 bool try_emit_mad(ir_expression
*ir
, int mul_arg
);
509 void resolve_ud_negate(src_reg
*reg
);
511 src_reg
get_timestamp();
513 bool process_move_condition(ir_rvalue
*ir
);
515 void dump_instruction(backend_instruction
*inst
);
519 void lower_attributes_to_hw_regs(const int *attribute_map
,
521 void setup_payload_interference(struct ra_graph
*g
, int first_payload_node
,
523 virtual dst_reg
*make_reg_for_system_value(ir_variable
*ir
) = 0;
524 virtual void setup_payload() = 0;
525 virtual void emit_prolog() = 0;
526 virtual void emit_program_code() = 0;
527 virtual void emit_thread_end() = 0;
528 virtual void emit_urb_write_header(int mrf
) = 0;
529 virtual vec4_instruction
*emit_urb_write_opcode(bool complete
) = 0;
530 virtual int compute_array_stride(ir_dereference_array
*ir
);
532 const bool debug_flag
;
536 * If true, then register allocation should fail instead of spilling.
538 const bool no_spills
;
543 * The vertex shader code generator.
545 * Translates VS IR to actual i965 assembly code.
550 vec4_generator(struct brw_context
*brw
,
551 struct gl_shader_program
*shader_prog
,
552 struct gl_program
*prog
,
553 struct brw_vec4_prog_data
*prog_data
,
558 const unsigned *generate_assembly(exec_list
*insts
, unsigned *asm_size
);
561 void generate_code(exec_list
*instructions
);
562 void generate_vec4_instruction(vec4_instruction
*inst
,
564 struct brw_reg
*src
);
566 void generate_math1_gen4(vec4_instruction
*inst
,
569 void generate_math1_gen6(vec4_instruction
*inst
,
572 void generate_math2_gen4(vec4_instruction
*inst
,
575 struct brw_reg src1
);
576 void generate_math2_gen6(vec4_instruction
*inst
,
579 struct brw_reg src1
);
580 void generate_math2_gen7(vec4_instruction
*inst
,
583 struct brw_reg src1
);
585 void generate_tex(vec4_instruction
*inst
,
589 void generate_vs_urb_write(vec4_instruction
*inst
);
590 void generate_gs_urb_write(vec4_instruction
*inst
);
591 void generate_gs_thread_end(vec4_instruction
*inst
);
592 void generate_gs_set_write_offset(struct brw_reg dst
,
594 struct brw_reg src1
);
595 void generate_gs_set_vertex_count(struct brw_reg dst
,
597 void generate_gs_set_dword_2_immed(struct brw_reg dst
, struct brw_reg src
);
598 void generate_gs_prepare_channel_masks(struct brw_reg dst
);
599 void generate_gs_set_channel_masks(struct brw_reg dst
, struct brw_reg src
);
600 void generate_oword_dual_block_offsets(struct brw_reg m1
,
601 struct brw_reg index
);
602 void generate_scratch_write(vec4_instruction
*inst
,
605 struct brw_reg index
);
606 void generate_scratch_read(vec4_instruction
*inst
,
608 struct brw_reg index
);
609 void generate_pull_constant_load(vec4_instruction
*inst
,
611 struct brw_reg index
,
612 struct brw_reg offset
);
613 void generate_pull_constant_load_gen7(vec4_instruction
*inst
,
615 struct brw_reg surf_index
,
616 struct brw_reg offset
);
617 void generate_unpack_flags(vec4_instruction
*inst
,
620 void generate_untyped_atomic(vec4_instruction
*inst
,
622 struct brw_reg atomic_op
,
623 struct brw_reg surf_index
);
625 void generate_untyped_surface_read(vec4_instruction
*inst
,
627 struct brw_reg surf_index
);
629 void mark_surface_used(unsigned surf_index
);
631 struct brw_context
*brw
;
633 struct brw_compile
*p
;
635 struct gl_shader_program
*shader_prog
;
636 struct gl_shader
*shader
;
637 const struct gl_program
*prog
;
639 struct brw_vec4_prog_data
*prog_data
;
642 const bool debug_flag
;
645 } /* namespace brw */
646 #endif /* __cplusplus */
648 #endif /* BRW_VEC4_H */