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"
42 #include "gen8_generator.h"
48 struct brw_vec4_compile
{
49 GLuint last_scratch
; /**< measured in 32-byte (register size) units */
53 struct brw_vec4_prog_key
{
54 GLuint program_string_id
;
57 * True if at least one clip flag is enabled, regardless of whether the
58 * shader uses clip planes or gl_ClipDistance.
60 GLuint userclip_active
:1;
63 * How many user clipping planes are being uploaded to the vertex shader as
66 GLuint nr_userclip_plane_consts
:4;
68 GLuint clamp_vertex_color
:1;
70 struct brw_sampler_prog_key_data tex
;
79 brw_vec4_setup_prog_key_for_precompile(struct gl_context
*ctx
,
80 struct brw_vec4_prog_key
*key
,
81 GLuint id
, struct gl_program
*prog
);
91 swizzle_for_size(int size
);
96 /** Register file: GRF, MRF, IMM. */
97 enum register_file file
;
98 /** virtual register number. 0 = fixed hw reg */
100 /** Offset within the virtual register. */
102 /** Register type. BRW_REGISTER_TYPE_* */
104 struct brw_reg fixed_hw_reg
;
106 /** Value for file == BRW_IMMMEDIATE_FILE */
114 class src_reg
: public reg
117 DECLARE_RALLOC_CXX_OPERATORS(src_reg
)
121 src_reg(register_file file
, int reg
, const glsl_type
*type
);
126 src_reg(struct brw_reg reg
);
128 bool equals(src_reg
*r
);
129 bool is_zero() const;
132 src_reg(class vec4_visitor
*v
, const struct glsl_type
*type
);
134 explicit src_reg(dst_reg reg
);
136 GLuint swizzle
; /**< BRW_SWIZZLE_XYZW macros from brw_reg.h. */
143 class dst_reg
: public reg
146 DECLARE_RALLOC_CXX_OPERATORS(dst_reg
)
151 dst_reg(register_file file
, int reg
);
152 dst_reg(register_file file
, int reg
, const glsl_type
*type
, int writemask
);
153 dst_reg(struct brw_reg reg
);
154 dst_reg(class vec4_visitor
*v
, const struct glsl_type
*type
);
156 explicit dst_reg(src_reg reg
);
158 int writemask
; /**< Bitfield of WRITEMASK_[XYZW] */
164 with_writemask(dst_reg
const &r
, int mask
);
166 class vec4_instruction
: public backend_instruction
{
168 DECLARE_RALLOC_CXX_OPERATORS(vec4_instruction
)
170 vec4_instruction(vec4_visitor
*v
, enum opcode opcode
,
171 dst_reg dst
= dst_reg(),
172 src_reg src0
= src_reg(),
173 src_reg src1
= src_reg(),
174 src_reg src2
= src_reg());
176 struct brw_reg
get_dst(void);
177 struct brw_reg
get_src(const struct brw_vec4_prog_data
*prog_data
, int i
);
183 bool force_writemask_all
;
184 bool no_dd_clear
, no_dd_check
;
186 int conditional_mod
; /**< BRW_CONDITIONAL_* */
189 uint32_t texture_offset
; /**< Texture Offset bitfield */
190 int target
; /**< MRT target. */
193 enum brw_urb_write_flags urb_write_flags
;
195 int mlen
; /**< SEND message length */
196 int base_mrf
; /**< First MRF in the SEND message, if mlen is nonzero. */
198 uint32_t offset
; /* spill/unspill offset */
200 * Annotation for the generated IR. One of the two can be set.
203 const char *annotation
;
205 bool is_send_from_grf();
206 bool can_reswizzle_dst(int dst_writemask
, int swizzle
, int swizzle_mask
);
207 void reswizzle_dst(int dst_writemask
, int swizzle
);
209 bool depends_on_flags()
211 return predicate
|| opcode
== VS_OPCODE_UNPACK_FLAGS_SIMD4X2
;
216 * The vertex shader front-end.
218 * Translates either GLSL IR or Mesa IR (for ARB_vertex_program and
219 * fixed-function) into VS IR.
221 class vec4_visitor
: public backend_visitor
224 vec4_visitor(struct brw_context
*brw
,
225 struct brw_vec4_compile
*c
,
226 struct gl_program
*prog
,
227 const struct brw_vec4_prog_key
*key
,
228 struct brw_vec4_prog_data
*prog_data
,
229 struct gl_shader_program
*shader_prog
,
230 struct brw_shader
*shader
,
234 shader_time_shader_type st_base
,
235 shader_time_shader_type st_written
,
236 shader_time_shader_type st_reset
);
241 return dst_reg(brw_null_reg());
246 return dst_reg(retype(brw_null_reg(), BRW_REGISTER_TYPE_D
));
249 dst_reg
dst_null_ud()
251 return dst_reg(retype(brw_null_reg(), BRW_REGISTER_TYPE_UD
));
254 struct brw_vec4_compile
*c
;
255 const struct brw_vec4_prog_key
*key
;
256 struct brw_vec4_prog_data
*prog_data
;
257 unsigned int sanity_param_count
;
263 * GLSL IR currently being processed, which is associated with our
264 * driver IR instructions for debugging purposes.
267 const char *current_annotation
;
269 int *virtual_grf_sizes
;
270 int virtual_grf_count
;
271 int virtual_grf_array_size
;
272 int first_non_payload_grf
;
273 unsigned int max_grf
;
274 int *virtual_grf_start
;
275 int *virtual_grf_end
;
276 dst_reg userplane
[MAX_CLIP_PLANES
];
279 * This is the size to be used for an array with an element per
282 int virtual_grf_reg_count
;
283 /** Per-virtual-grf indices into an array of size virtual_grf_reg_count */
284 int *virtual_grf_reg_map
;
286 bool live_intervals_valid
;
288 dst_reg
*variable_storage(ir_variable
*var
);
290 void reladdr_to_temp(ir_instruction
*ir
, src_reg
*reg
, int *num_reladdr
);
292 bool need_all_constants_in_pull_buffer
;
295 * \name Visit methods
297 * As typical for the visitor pattern, there must be one \c visit method for
298 * each concrete subclass of \c ir_instruction. Virtual base classes within
299 * the hierarchy should not have \c visit methods.
302 virtual void visit(ir_variable
*);
303 virtual void visit(ir_loop
*);
304 virtual void visit(ir_loop_jump
*);
305 virtual void visit(ir_function_signature
*);
306 virtual void visit(ir_function
*);
307 virtual void visit(ir_expression
*);
308 virtual void visit(ir_swizzle
*);
309 virtual void visit(ir_dereference_variable
*);
310 virtual void visit(ir_dereference_array
*);
311 virtual void visit(ir_dereference_record
*);
312 virtual void visit(ir_assignment
*);
313 virtual void visit(ir_constant
*);
314 virtual void visit(ir_call
*);
315 virtual void visit(ir_return
*);
316 virtual void visit(ir_discard
*);
317 virtual void visit(ir_texture
*);
318 virtual void visit(ir_if
*);
319 virtual void visit(ir_emit_vertex
*);
320 virtual void visit(ir_end_primitive
*);
325 /* Regs for vertex results. Generated at ir_variable visiting time
326 * for the ir->location's used.
328 dst_reg output_reg
[BRW_VARYING_SLOT_COUNT
];
329 const char *output_reg_annotation
[BRW_VARYING_SLOT_COUNT
];
330 int uniform_size
[MAX_UNIFORMS
];
331 int uniform_vector_size
[MAX_UNIFORMS
];
334 src_reg shader_start_time
;
336 struct hash_table
*variable_ht
;
339 void fail(const char *msg
, ...);
341 int virtual_grf_alloc(int size
);
342 void setup_uniform_clipplane_values();
343 void setup_uniform_values(ir_variable
*ir
);
344 void setup_builtin_uniform_values(ir_variable
*ir
);
345 int setup_uniforms(int payload_reg
);
346 bool reg_allocate_trivial();
348 void evaluate_spill_costs(float *spill_costs
, bool *no_spill
);
349 int choose_spill_reg(struct ra_graph
*g
);
350 void spill_reg(int spill_reg
);
351 void move_grf_array_access_to_scratch();
352 void move_uniform_array_access_to_pull_constants();
353 void move_push_constants_to_pull_constants();
354 void split_uniform_registers();
355 void pack_uniform_registers();
356 void calculate_live_intervals();
357 void invalidate_live_intervals();
358 void split_virtual_grfs();
359 bool dead_code_eliminate();
360 bool virtual_grf_interferes(int a
, int b
);
361 bool opt_copy_propagation();
362 bool opt_algebraic();
363 bool opt_register_coalesce();
364 void opt_set_dependency_control();
365 void opt_schedule_instructions();
367 bool can_do_source_mods(vec4_instruction
*inst
);
369 vec4_instruction
*emit(vec4_instruction
*inst
);
371 vec4_instruction
*emit(enum opcode opcode
);
373 vec4_instruction
*emit(enum opcode opcode
, dst_reg dst
);
375 vec4_instruction
*emit(enum opcode opcode
, dst_reg dst
, src_reg src0
);
377 vec4_instruction
*emit(enum opcode opcode
, dst_reg dst
,
378 src_reg src0
, src_reg src1
);
380 vec4_instruction
*emit(enum opcode opcode
, dst_reg dst
,
381 src_reg src0
, src_reg src1
, src_reg src2
);
383 vec4_instruction
*emit_before(vec4_instruction
*inst
,
384 vec4_instruction
*new_inst
);
386 vec4_instruction
*MOV(dst_reg dst
, src_reg src0
);
387 vec4_instruction
*NOT(dst_reg dst
, src_reg src0
);
388 vec4_instruction
*RNDD(dst_reg dst
, src_reg src0
);
389 vec4_instruction
*RNDE(dst_reg dst
, src_reg src0
);
390 vec4_instruction
*RNDZ(dst_reg dst
, src_reg src0
);
391 vec4_instruction
*FRC(dst_reg dst
, src_reg src0
);
392 vec4_instruction
*F32TO16(dst_reg dst
, src_reg src0
);
393 vec4_instruction
*F16TO32(dst_reg dst
, src_reg src0
);
394 vec4_instruction
*ADD(dst_reg dst
, src_reg src0
, src_reg src1
);
395 vec4_instruction
*MUL(dst_reg dst
, src_reg src0
, src_reg src1
);
396 vec4_instruction
*MACH(dst_reg dst
, src_reg src0
, src_reg src1
);
397 vec4_instruction
*MAC(dst_reg dst
, src_reg src0
, src_reg src1
);
398 vec4_instruction
*AND(dst_reg dst
, src_reg src0
, src_reg src1
);
399 vec4_instruction
*OR(dst_reg dst
, src_reg src0
, src_reg src1
);
400 vec4_instruction
*XOR(dst_reg dst
, src_reg src0
, src_reg src1
);
401 vec4_instruction
*DP3(dst_reg dst
, src_reg src0
, src_reg src1
);
402 vec4_instruction
*DP4(dst_reg dst
, src_reg src0
, src_reg src1
);
403 vec4_instruction
*DPH(dst_reg dst
, src_reg src0
, src_reg src1
);
404 vec4_instruction
*SHL(dst_reg dst
, src_reg src0
, src_reg src1
);
405 vec4_instruction
*SHR(dst_reg dst
, src_reg src0
, src_reg src1
);
406 vec4_instruction
*ASR(dst_reg dst
, src_reg src0
, src_reg src1
);
407 vec4_instruction
*CMP(dst_reg dst
, src_reg src0
, src_reg src1
,
409 vec4_instruction
*IF(src_reg src0
, src_reg src1
, uint32_t condition
);
410 vec4_instruction
*IF(uint32_t predicate
);
411 vec4_instruction
*PULL_CONSTANT_LOAD(dst_reg dst
, src_reg index
);
412 vec4_instruction
*SCRATCH_READ(dst_reg dst
, src_reg index
);
413 vec4_instruction
*SCRATCH_WRITE(dst_reg dst
, src_reg src
, src_reg index
);
414 vec4_instruction
*LRP(dst_reg dst
, src_reg a
, src_reg y
, src_reg x
);
415 vec4_instruction
*BFREV(dst_reg dst
, src_reg value
);
416 vec4_instruction
*BFE(dst_reg dst
, src_reg bits
, src_reg offset
, src_reg value
);
417 vec4_instruction
*BFI1(dst_reg dst
, src_reg bits
, src_reg offset
);
418 vec4_instruction
*BFI2(dst_reg dst
, src_reg bfi1_dst
, src_reg insert
, src_reg base
);
419 vec4_instruction
*FBH(dst_reg dst
, src_reg value
);
420 vec4_instruction
*FBL(dst_reg dst
, src_reg value
);
421 vec4_instruction
*CBIT(dst_reg dst
, src_reg value
);
422 vec4_instruction
*MAD(dst_reg dst
, src_reg c
, src_reg b
, src_reg a
);
423 vec4_instruction
*ADDC(dst_reg dst
, src_reg src0
, src_reg src1
);
424 vec4_instruction
*SUBB(dst_reg dst
, src_reg src0
, src_reg src1
);
426 int implied_mrf_writes(vec4_instruction
*inst
);
428 bool try_rewrite_rhs_to_dst(ir_assignment
*ir
,
431 vec4_instruction
*pre_rhs_inst
,
432 vec4_instruction
*last_rhs_inst
);
434 bool try_copy_propagation(vec4_instruction
*inst
, int arg
,
437 /** Walks an exec_list of ir_instruction and sends it through this visitor. */
438 void visit_instructions(const exec_list
*list
);
440 void emit_vp_sop(uint32_t condmod
, dst_reg dst
,
441 src_reg src0
, src_reg src1
, src_reg one
);
443 void emit_bool_to_cond_code(ir_rvalue
*ir
, uint32_t *predicate
);
444 void emit_bool_comparison(unsigned int op
, dst_reg dst
, src_reg src0
, src_reg src1
);
445 void emit_if_gen6(ir_if
*ir
);
447 void emit_minmax(uint32_t condmod
, dst_reg dst
, src_reg src0
, src_reg src1
);
449 void emit_block_move(dst_reg
*dst
, src_reg
*src
,
450 const struct glsl_type
*type
, uint32_t predicate
);
452 void emit_constant_values(dst_reg
*dst
, ir_constant
*value
);
455 * Emit the correct dot-product instruction for the type of arguments
457 void emit_dp(dst_reg dst
, src_reg src0
, src_reg src1
, unsigned elements
);
459 void emit_scalar(ir_instruction
*ir
, enum prog_opcode op
,
460 dst_reg dst
, src_reg src0
);
462 void emit_scalar(ir_instruction
*ir
, enum prog_opcode op
,
463 dst_reg dst
, src_reg src0
, src_reg src1
);
465 void emit_scs(ir_instruction
*ir
, enum prog_opcode op
,
466 dst_reg dst
, const src_reg
&src
);
468 src_reg
fix_3src_operand(src_reg src
);
470 void emit_math1_gen6(enum opcode opcode
, dst_reg dst
, src_reg src
);
471 void emit_math1_gen4(enum opcode opcode
, dst_reg dst
, src_reg src
);
472 void emit_math(enum opcode opcode
, dst_reg dst
, src_reg src
);
473 void emit_math2_gen6(enum opcode opcode
, dst_reg dst
, src_reg src0
, src_reg src1
);
474 void emit_math2_gen4(enum opcode opcode
, dst_reg dst
, src_reg src0
, src_reg src1
);
475 void emit_math(enum opcode opcode
, dst_reg dst
, src_reg src0
, src_reg src1
);
476 src_reg
fix_math_operand(src_reg src
);
478 void emit_pack_half_2x16(dst_reg dst
, src_reg src0
);
479 void emit_unpack_half_2x16(dst_reg dst
, src_reg src0
);
481 uint32_t gather_channel(ir_texture
*ir
, int sampler
);
482 src_reg
emit_mcs_fetch(ir_texture
*ir
, src_reg coordinate
, int sampler
);
483 void emit_gen6_gather_wa(uint8_t wa
, dst_reg dst
);
484 void swizzle_result(ir_texture
*ir
, src_reg orig_val
, int sampler
);
486 void emit_ndc_computation();
487 void emit_psiz_and_flags(struct brw_reg reg
);
488 void emit_clip_distances(dst_reg reg
, int offset
);
489 void emit_generic_urb_slot(dst_reg reg
, int varying
);
490 void emit_urb_slot(int mrf
, int varying
);
492 void emit_shader_time_begin();
493 void emit_shader_time_end();
494 void emit_shader_time_write(enum shader_time_shader_type type
,
497 void emit_untyped_atomic(unsigned atomic_op
, unsigned surf_index
,
498 dst_reg dst
, src_reg offset
, src_reg src0
,
501 void emit_untyped_surface_read(unsigned surf_index
, dst_reg dst
,
504 src_reg
get_scratch_offset(vec4_instruction
*inst
,
505 src_reg
*reladdr
, int reg_offset
);
506 src_reg
get_pull_constant_offset(vec4_instruction
*inst
,
507 src_reg
*reladdr
, int reg_offset
);
508 void emit_scratch_read(vec4_instruction
*inst
,
512 void emit_scratch_write(vec4_instruction
*inst
,
514 void emit_pull_constant_load(vec4_instruction
*inst
,
519 bool try_emit_sat(ir_expression
*ir
);
520 bool try_emit_mad(ir_expression
*ir
, int mul_arg
);
521 void resolve_ud_negate(src_reg
*reg
);
523 src_reg
get_timestamp();
525 bool process_move_condition(ir_rvalue
*ir
);
527 void dump_instruction(backend_instruction
*inst
);
529 void visit_atomic_counter_intrinsic(ir_call
*ir
);
533 void lower_attributes_to_hw_regs(const int *attribute_map
,
535 void setup_payload_interference(struct ra_graph
*g
, int first_payload_node
,
537 virtual dst_reg
*make_reg_for_system_value(ir_variable
*ir
) = 0;
538 virtual void setup_payload() = 0;
539 virtual void emit_prolog() = 0;
540 virtual void emit_program_code() = 0;
541 virtual void emit_thread_end() = 0;
542 virtual void emit_urb_write_header(int mrf
) = 0;
543 virtual vec4_instruction
*emit_urb_write_opcode(bool complete
) = 0;
544 virtual int compute_array_stride(ir_dereference_array
*ir
);
546 const bool debug_flag
;
550 * If true, then register allocation should fail instead of spilling.
552 const bool no_spills
;
554 const shader_time_shader_type st_base
;
555 const shader_time_shader_type st_written
;
556 const shader_time_shader_type st_reset
;
561 * The vertex shader code generator.
563 * Translates VS IR to actual i965 assembly code.
568 vec4_generator(struct brw_context
*brw
,
569 struct gl_shader_program
*shader_prog
,
570 struct gl_program
*prog
,
571 struct brw_vec4_prog_data
*prog_data
,
576 const unsigned *generate_assembly(exec_list
*insts
, unsigned *asm_size
);
579 void generate_code(exec_list
*instructions
);
580 void generate_vec4_instruction(vec4_instruction
*inst
,
582 struct brw_reg
*src
);
584 void generate_math1_gen4(vec4_instruction
*inst
,
587 void generate_math1_gen6(vec4_instruction
*inst
,
590 void generate_math2_gen4(vec4_instruction
*inst
,
593 struct brw_reg src1
);
594 void generate_math2_gen6(vec4_instruction
*inst
,
597 struct brw_reg src1
);
598 void generate_math2_gen7(vec4_instruction
*inst
,
601 struct brw_reg src1
);
603 void generate_tex(vec4_instruction
*inst
,
607 void generate_vs_urb_write(vec4_instruction
*inst
);
608 void generate_gs_urb_write(vec4_instruction
*inst
);
609 void generate_gs_thread_end(vec4_instruction
*inst
);
610 void generate_gs_set_write_offset(struct brw_reg dst
,
612 struct brw_reg src1
);
613 void generate_gs_set_vertex_count(struct brw_reg dst
,
615 void generate_gs_set_dword_2_immed(struct brw_reg dst
, struct brw_reg src
);
616 void generate_gs_prepare_channel_masks(struct brw_reg dst
);
617 void generate_gs_set_channel_masks(struct brw_reg dst
, struct brw_reg src
);
618 void generate_oword_dual_block_offsets(struct brw_reg m1
,
619 struct brw_reg index
);
620 void generate_scratch_write(vec4_instruction
*inst
,
623 struct brw_reg index
);
624 void generate_scratch_read(vec4_instruction
*inst
,
626 struct brw_reg index
);
627 void generate_pull_constant_load(vec4_instruction
*inst
,
629 struct brw_reg index
,
630 struct brw_reg offset
);
631 void generate_pull_constant_load_gen7(vec4_instruction
*inst
,
633 struct brw_reg surf_index
,
634 struct brw_reg offset
);
635 void generate_unpack_flags(vec4_instruction
*inst
,
638 void generate_untyped_atomic(vec4_instruction
*inst
,
640 struct brw_reg atomic_op
,
641 struct brw_reg surf_index
);
643 void generate_untyped_surface_read(vec4_instruction
*inst
,
645 struct brw_reg surf_index
);
647 void mark_surface_used(unsigned surf_index
);
649 struct brw_context
*brw
;
651 struct brw_compile
*p
;
653 struct gl_shader_program
*shader_prog
;
654 const struct gl_program
*prog
;
656 struct brw_vec4_prog_data
*prog_data
;
659 const bool debug_flag
;
663 * The vertex shader code generator.
665 * Translates VS IR to actual i965 assembly code.
667 class gen8_vec4_generator
: public gen8_generator
670 gen8_vec4_generator(struct brw_context
*brw
,
671 struct gl_shader_program
*shader_prog
,
672 struct gl_program
*prog
,
673 struct brw_vec4_prog_data
*prog_data
,
676 ~gen8_vec4_generator();
678 const unsigned *generate_assembly(exec_list
*insts
, unsigned *asm_size
);
681 void generate_code(exec_list
*instructions
);
682 void generate_vec4_instruction(vec4_instruction
*inst
,
684 struct brw_reg
*src
);
686 void generate_tex(vec4_instruction
*inst
,
689 void generate_urb_write(vec4_instruction
*ir
, bool copy_g0
);
690 void generate_gs_thread_end(vec4_instruction
*ir
);
691 void generate_gs_set_write_offset(struct brw_reg dst
,
693 struct brw_reg src1
);
694 void generate_gs_set_vertex_count(struct brw_reg dst
,
696 void generate_gs_set_dword_2_immed(struct brw_reg dst
, struct brw_reg src
);
697 void generate_gs_prepare_channel_masks(struct brw_reg dst
);
698 void generate_gs_set_channel_masks(struct brw_reg dst
, struct brw_reg src
);
700 void generate_oword_dual_block_offsets(struct brw_reg m1
,
701 struct brw_reg index
);
702 void generate_scratch_write(vec4_instruction
*inst
,
705 struct brw_reg index
);
706 void generate_scratch_read(vec4_instruction
*inst
,
708 struct brw_reg index
);
709 void generate_pull_constant_load(vec4_instruction
*inst
,
711 struct brw_reg index
,
712 struct brw_reg offset
);
714 void mark_surface_used(unsigned surf_index
);
716 struct brw_vec4_prog_data
*prog_data
;
718 const bool debug_flag
;
722 } /* namespace brw */
723 #endif /* __cplusplus */
725 #endif /* BRW_VEC4_H */