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
24 * Eric Anholt <eric@anholt.net>
30 #include <sys/types.h>
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"
45 #include "../glsl/glsl_types.h"
46 #include "../glsl/ir.h"
49 ARF
= BRW_ARCHITECTURE_REGISTER_FILE
,
50 GRF
= BRW_GENERAL_REGISTER_FILE
,
51 MRF
= BRW_MESSAGE_REGISTER_FILE
,
52 IMM
= BRW_IMMEDIATE_VALUE
,
53 FIXED_HW_REG
, /* a struct brw_reg */
54 UNIFORM
, /* prog_data->params[hw_reg] */
59 FS_OPCODE_FB_WRITE
= 256,
78 FS_OPCODE_DISCARD_NOT
,
79 FS_OPCODE_DISCARD_AND
,
82 FS_OPCODE_PULL_CONSTANT_LOAD
,
88 /* Callers of this ralloc-based new need not call delete. It's
89 * easier to just ralloc_free 'ctx' (or any of its ancestors). */
90 static void* operator new(size_t size
, void *ctx
)
94 node
= ralloc_size(ctx
, size
);
102 memset(this, 0, sizeof(*this));
107 /** Generic unset register constructor. */
111 this->file
= BAD_FILE
;
114 /** Immediate value constructor. */
119 this->type
= BRW_REGISTER_TYPE_F
;
123 /** Immediate value constructor. */
128 this->type
= BRW_REGISTER_TYPE_D
;
132 /** Immediate value constructor. */
137 this->type
= BRW_REGISTER_TYPE_UD
;
141 /** Fixed brw_reg Immediate value constructor. */
142 fs_reg(struct brw_reg fixed_hw_reg
)
145 this->file
= FIXED_HW_REG
;
146 this->fixed_hw_reg
= fixed_hw_reg
;
147 this->type
= fixed_hw_reg
.type
;
150 fs_reg(enum register_file file
, int hw_reg
);
151 fs_reg(enum register_file file
, int hw_reg
, uint32_t type
);
152 fs_reg(class fs_visitor
*v
, const struct glsl_type
*type
);
154 bool equals(fs_reg
*r
)
156 return (file
== r
->file
&&
158 reg_offset
== r
->reg_offset
&&
159 hw_reg
== r
->hw_reg
&&
161 negate
== r
->negate
&&
163 memcmp(&fixed_hw_reg
, &r
->fixed_hw_reg
,
164 sizeof(fixed_hw_reg
)) == 0 &&
169 /** Register file: ARF, GRF, MRF, IMM. */
170 enum register_file file
;
171 /** virtual register number. 0 = fixed hw reg */
173 /** Offset within the virtual register. */
175 /** HW register number. Generally unset until register allocation. */
177 /** Register type. BRW_REGISTER_TYPE_* */
181 struct brw_reg fixed_hw_reg
;
182 int smear
; /* -1, or a channel of the reg to smear to all channels. */
184 /** Value for file == BRW_IMMMEDIATE_FILE */
192 static const fs_reg reg_undef
;
193 static const fs_reg
reg_null_f(ARF
, BRW_ARF_NULL
, BRW_REGISTER_TYPE_F
);
194 static const fs_reg
reg_null_d(ARF
, BRW_ARF_NULL
, BRW_REGISTER_TYPE_D
);
196 class fs_inst
: public exec_node
{
198 /* Callers of this ralloc-based new need not call delete. It's
199 * easier to just ralloc_free 'ctx' (or any of its ancestors). */
200 static void* operator new(size_t size
, void *ctx
)
204 node
= rzalloc_size(ctx
, size
);
205 assert(node
!= NULL
);
212 memset(this, 0, sizeof(*this));
213 this->opcode
= BRW_OPCODE_NOP
;
214 this->conditional_mod
= BRW_CONDITIONAL_NONE
;
216 this->dst
= reg_undef
;
217 this->src
[0] = reg_undef
;
218 this->src
[1] = reg_undef
;
219 this->src
[2] = reg_undef
;
230 this->opcode
= opcode
;
233 fs_inst(int opcode
, fs_reg dst
)
236 this->opcode
= opcode
;
240 assert(dst
.reg_offset
>= 0);
243 fs_inst(int opcode
, fs_reg dst
, fs_reg src0
)
246 this->opcode
= opcode
;
251 assert(dst
.reg_offset
>= 0);
252 if (src
[0].file
== GRF
)
253 assert(src
[0].reg_offset
>= 0);
256 fs_inst(int opcode
, fs_reg dst
, fs_reg src0
, fs_reg src1
)
259 this->opcode
= opcode
;
265 assert(dst
.reg_offset
>= 0);
266 if (src
[0].file
== GRF
)
267 assert(src
[0].reg_offset
>= 0);
268 if (src
[1].file
== GRF
)
269 assert(src
[1].reg_offset
>= 0);
272 fs_inst(int opcode
, fs_reg dst
, fs_reg src0
, fs_reg src1
, fs_reg src2
)
275 this->opcode
= opcode
;
282 assert(dst
.reg_offset
>= 0);
283 if (src
[0].file
== GRF
)
284 assert(src
[0].reg_offset
>= 0);
285 if (src
[1].file
== GRF
)
286 assert(src
[1].reg_offset
>= 0);
287 if (src
[2].file
== GRF
)
288 assert(src
[2].reg_offset
>= 0);
291 bool equals(fs_inst
*inst
)
293 return (opcode
== inst
->opcode
&&
294 dst
.equals(&inst
->dst
) &&
295 src
[0].equals(&inst
->src
[0]) &&
296 src
[1].equals(&inst
->src
[1]) &&
297 src
[2].equals(&inst
->src
[2]) &&
298 saturate
== inst
->saturate
&&
299 predicated
== inst
->predicated
&&
300 conditional_mod
== inst
->conditional_mod
&&
301 mlen
== inst
->mlen
&&
302 base_mrf
== inst
->base_mrf
&&
303 sampler
== inst
->sampler
&&
304 target
== inst
->target
&&
306 header_present
== inst
->header_present
&&
307 shadow_compare
== inst
->shadow_compare
&&
308 offset
== inst
->offset
);
313 return (opcode
== FS_OPCODE_TEX
||
314 opcode
== FS_OPCODE_TXB
||
315 opcode
== FS_OPCODE_TXD
||
316 opcode
== FS_OPCODE_TXL
);
321 return (opcode
== FS_OPCODE_RCP
||
322 opcode
== FS_OPCODE_RSQ
||
323 opcode
== FS_OPCODE_SQRT
||
324 opcode
== FS_OPCODE_EXP2
||
325 opcode
== FS_OPCODE_LOG2
||
326 opcode
== FS_OPCODE_SIN
||
327 opcode
== FS_OPCODE_COS
||
328 opcode
== FS_OPCODE_POW
);
331 int opcode
; /* BRW_OPCODE_* or FS_OPCODE_* */
336 bool predicate_inverse
;
337 int conditional_mod
; /**< BRW_CONDITIONAL_* */
339 int mlen
; /**< SEND message length */
340 int base_mrf
; /**< First MRF in the SEND message, if mlen is nonzero. */
342 int target
; /**< MRT target. */
346 bool force_uncompressed
;
348 uint32_t offset
; /* spill/unspill offset */
351 * Annotation for the generated IR. One of the two can be set.
354 const char *annotation
;
358 class fs_visitor
: public ir_visitor
362 fs_visitor(struct brw_wm_compile
*c
, struct brw_shader
*shader
)
367 this->fp
= brw
->fragment_program
;
368 this->intel
= &brw
->intel
;
369 this->ctx
= &intel
->ctx
;
370 this->mem_ctx
= ralloc_context(NULL
);
371 this->shader
= shader
;
372 this->failed
= false;
373 this->variable_ht
= hash_table_ctor(0,
374 hash_table_pointer_hash
,
375 hash_table_pointer_compare
);
377 /* There's a question that appears to be left open in the spec:
378 * How do implicit dst conversions interact with the CMP
379 * instruction or conditional mods? On gen6, the instruction:
381 * CMP null<d> src0<f> src1<f>
383 * will do src1 - src0 and compare that result as if it was an
384 * integer. On gen4, it will do src1 - src0 as float, convert
385 * the result to int, and compare as int. In between, it
386 * appears that it does src1 - src0 and does the compare in the
387 * execution type so dst type doesn't matter.
389 if (this->intel
->gen
> 4)
390 this->reg_null_cmp
= reg_null_d
;
392 this->reg_null_cmp
= reg_null_f
;
394 this->frag_color
= NULL
;
395 this->frag_data
= NULL
;
396 this->frag_depth
= NULL
;
397 this->first_non_payload_grf
= 0;
399 this->current_annotation
= NULL
;
400 this->base_ir
= NULL
;
402 this->virtual_grf_sizes
= NULL
;
403 this->virtual_grf_next
= 1;
404 this->virtual_grf_array_size
= 0;
405 this->virtual_grf_def
= NULL
;
406 this->virtual_grf_use
= NULL
;
407 this->live_intervals_valid
= false;
409 this->kill_emitted
= false;
410 this->force_uncompressed_stack
= 0;
411 this->force_sechalf_stack
= 0;
416 ralloc_free(this->mem_ctx
);
417 hash_table_dtor(this->variable_ht
);
420 fs_reg
*variable_storage(ir_variable
*var
);
421 int virtual_grf_alloc(int size
);
423 void visit(ir_variable
*ir
);
424 void visit(ir_assignment
*ir
);
425 void visit(ir_dereference_variable
*ir
);
426 void visit(ir_dereference_record
*ir
);
427 void visit(ir_dereference_array
*ir
);
428 void visit(ir_expression
*ir
);
429 void visit(ir_texture
*ir
);
430 void visit(ir_if
*ir
);
431 void visit(ir_constant
*ir
);
432 void visit(ir_swizzle
*ir
);
433 void visit(ir_return
*ir
);
434 void visit(ir_loop
*ir
);
435 void visit(ir_loop_jump
*ir
);
436 void visit(ir_discard
*ir
);
437 void visit(ir_call
*ir
);
438 void visit(ir_function
*ir
);
439 void visit(ir_function_signature
*ir
);
441 fs_inst
*emit(fs_inst inst
);
443 fs_inst
*emit(int opcode
)
445 return emit(fs_inst(opcode
));
448 fs_inst
*emit(int opcode
, fs_reg dst
)
450 return emit(fs_inst(opcode
, dst
));
453 fs_inst
*emit(int opcode
, fs_reg dst
, fs_reg src0
)
455 return emit(fs_inst(opcode
, dst
, src0
));
458 fs_inst
*emit(int opcode
, fs_reg dst
, fs_reg src0
, fs_reg src1
)
460 return emit(fs_inst(opcode
, dst
, src0
, src1
));
463 fs_inst
*emit(int opcode
, fs_reg dst
, fs_reg src0
, fs_reg src1
, fs_reg src2
)
465 return emit(fs_inst(opcode
, dst
, src0
, src1
, src2
));
469 void setup_paramvalues_refs();
470 void assign_curb_setup();
471 void calculate_urb_setup();
472 void assign_urb_setup();
474 void assign_regs_trivial();
475 int choose_spill_reg(struct ra_graph
*g
);
476 void spill_reg(int spill_reg
);
477 void split_virtual_grfs();
478 void setup_pull_constants();
479 void calculate_live_intervals();
480 bool propagate_constants();
481 bool register_coalesce();
482 bool compute_to_mrf();
483 bool dead_code_eliminate();
484 bool remove_duplicate_mrf_writes();
485 bool virtual_grf_interferes(int a
, int b
);
486 void schedule_instructions();
487 void fail(const char *msg
, ...);
489 void push_force_uncompressed();
490 void pop_force_uncompressed();
491 void push_force_sechalf();
492 void pop_force_sechalf();
494 void generate_code();
495 void generate_fb_write(fs_inst
*inst
);
496 void generate_pixel_xy(struct brw_reg dst
, bool is_x
);
497 void generate_linterp(fs_inst
*inst
, struct brw_reg dst
,
498 struct brw_reg
*src
);
499 void generate_tex(fs_inst
*inst
, struct brw_reg dst
, struct brw_reg src
);
500 void generate_math(fs_inst
*inst
, struct brw_reg dst
, struct brw_reg
*src
);
501 void generate_discard_not(fs_inst
*inst
, struct brw_reg temp
);
502 void generate_discard_and(fs_inst
*inst
, struct brw_reg temp
);
503 void generate_ddx(fs_inst
*inst
, struct brw_reg dst
, struct brw_reg src
);
504 void generate_ddy(fs_inst
*inst
, struct brw_reg dst
, struct brw_reg src
);
505 void generate_spill(fs_inst
*inst
, struct brw_reg src
);
506 void generate_unspill(fs_inst
*inst
, struct brw_reg dst
);
507 void generate_pull_constant_load(fs_inst
*inst
, struct brw_reg dst
);
509 void emit_dummy_fs();
510 fs_reg
*emit_fragcoord_interpolation(ir_variable
*ir
);
511 fs_reg
*emit_frontfacing_interpolation(ir_variable
*ir
);
512 fs_reg
*emit_general_interpolation(ir_variable
*ir
);
513 void emit_interpolation_setup_gen4();
514 void emit_interpolation_setup_gen6();
515 fs_inst
*emit_texture_gen4(ir_texture
*ir
, fs_reg dst
, fs_reg coordinate
);
516 fs_inst
*emit_texture_gen5(ir_texture
*ir
, fs_reg dst
, fs_reg coordinate
);
517 fs_inst
*emit_math(fs_opcodes op
, fs_reg dst
, fs_reg src0
);
518 fs_inst
*emit_math(fs_opcodes op
, fs_reg dst
, fs_reg src0
, fs_reg src1
);
519 bool try_emit_saturate(ir_expression
*ir
);
520 void emit_bool_to_cond_code(ir_rvalue
*condition
);
521 void emit_if_gen6(ir_if
*ir
);
522 void emit_unspill(fs_inst
*inst
, fs_reg reg
, uint32_t spill_offset
);
524 void emit_fb_writes();
525 void emit_assignment_writes(fs_reg
&l
, fs_reg
&r
,
526 const glsl_type
*type
, bool predicated
);
528 struct brw_reg
interp_reg(int location
, int channel
);
529 int setup_uniform_values(int loc
, const glsl_type
*type
);
530 void setup_builtin_uniform_values(ir_variable
*ir
);
531 int implied_mrf_writes(fs_inst
*inst
);
533 struct brw_context
*brw
;
534 const struct gl_fragment_program
*fp
;
535 struct intel_context
*intel
;
536 struct gl_context
*ctx
;
537 struct brw_wm_compile
*c
;
538 struct brw_compile
*p
;
539 struct brw_shader
*shader
;
541 exec_list instructions
;
543 /* Delayed setup of c->prog_data.params[] due to realloc of
544 * ParamValues[] during compile.
546 int param_index
[MAX_UNIFORMS
* 4];
547 int param_offset
[MAX_UNIFORMS
* 4];
549 int *virtual_grf_sizes
;
550 int virtual_grf_next
;
551 int virtual_grf_array_size
;
552 int *virtual_grf_def
;
553 int *virtual_grf_use
;
554 bool live_intervals_valid
;
556 struct hash_table
*variable_ht
;
557 ir_variable
*frag_color
, *frag_data
, *frag_depth
;
558 int first_non_payload_grf
;
559 int urb_setup
[FRAG_ATTRIB_MAX
];
562 /** @{ debug annotation info */
563 const char *current_annotation
;
564 ir_instruction
*base_ir
;
569 /* Result of last visit() method. */
582 int force_uncompressed_stack
;
583 int force_sechalf_stack
;
586 GLboolean
brw_do_channel_expressions(struct exec_list
*instructions
);
587 GLboolean
brw_do_vector_splitting(struct exec_list
*instructions
);