1 /* Copyright © 2011 Intel Corporation
3 * Permission is hereby granted, free of charge, to any person obtaining a
4 * copy of this software and associated documentation files (the "Software"),
5 * to deal in the Software without restriction, including without limitation
6 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
7 * and/or sell copies of the Software, and to permit persons to whom the
8 * Software is furnished to do so, subject to the following conditions:
10 * The above copyright notice and this permission notice (including the next
11 * paragraph) shall be included in all copies or substantial portions of the
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
18 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
19 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
27 #include "main/macros.h"
28 #include "program/prog_print.h"
29 #include "program/prog_parameter.h"
35 vec4_instruction::get_dst(void)
37 struct brw_reg brw_reg
;
41 brw_reg
= brw_vec8_grf(dst
.reg
+ dst
.reg_offset
, 0);
42 brw_reg
= retype(brw_reg
, dst
.type
);
43 brw_reg
.dw1
.bits
.writemask
= dst
.writemask
;
47 brw_reg
= brw_message_reg(dst
.reg
+ dst
.reg_offset
);
48 brw_reg
= retype(brw_reg
, dst
.type
);
49 brw_reg
.dw1
.bits
.writemask
= dst
.writemask
;
53 brw_reg
= dst
.fixed_hw_reg
;
57 brw_reg
= brw_null_reg();
61 assert(!"not reached");
62 brw_reg
= brw_null_reg();
69 vec4_instruction::get_src(int i
)
71 struct brw_reg brw_reg
;
73 switch (src
[i
].file
) {
75 brw_reg
= brw_vec8_grf(src
[i
].reg
+ src
[i
].reg_offset
, 0);
76 brw_reg
= retype(brw_reg
, src
[i
].type
);
77 brw_reg
.dw1
.bits
.swizzle
= src
[i
].swizzle
;
79 brw_reg
= brw_abs(brw_reg
);
81 brw_reg
= negate(brw_reg
);
85 switch (src
[i
].type
) {
86 case BRW_REGISTER_TYPE_F
:
87 brw_reg
= brw_imm_f(src
[i
].imm
.f
);
89 case BRW_REGISTER_TYPE_D
:
90 brw_reg
= brw_imm_d(src
[i
].imm
.i
);
92 case BRW_REGISTER_TYPE_UD
:
93 brw_reg
= brw_imm_ud(src
[i
].imm
.u
);
96 assert(!"not reached");
97 brw_reg
= brw_null_reg();
103 brw_reg
= stride(brw_vec4_grf(1 + (src
[i
].reg
+ src
[i
].reg_offset
) / 2,
104 ((src
[i
].reg
+ src
[i
].reg_offset
) % 2) * 4),
106 brw_reg
= retype(brw_reg
, src
[i
].type
);
107 brw_reg
.dw1
.bits
.swizzle
= src
[i
].swizzle
;
109 brw_reg
= brw_abs(brw_reg
);
111 brw_reg
= negate(brw_reg
);
113 /* This should have been moved to pull constants. */
114 assert(!src
[i
].reladdr
);
118 brw_reg
= src
[i
].fixed_hw_reg
;
122 /* Probably unused. */
123 brw_reg
= brw_null_reg();
127 assert(!"not reached");
128 brw_reg
= brw_null_reg();
135 vec4_generator::vec4_generator(struct brw_context
*brw
,
136 struct brw_vs_compile
*c
,
137 struct gl_shader_program
*prog
,
139 : brw(brw
), c(c
), prog(prog
), mem_ctx(mem_ctx
)
142 vp
= &c
->vp
->program
;
144 shader
= prog
? prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] : NULL
;
146 p
= rzalloc(mem_ctx
, struct brw_compile
);
147 brw_init_compile(brw
, p
, mem_ctx
);
150 vec4_generator::~vec4_generator()
155 vec4_generator::generate_math1_gen4(vec4_instruction
*inst
,
161 brw_math_function(inst
->opcode
),
164 BRW_MATH_DATA_VECTOR
,
165 BRW_MATH_PRECISION_FULL
);
169 check_gen6_math_src_arg(struct brw_reg src
)
171 /* Source swizzles are ignored. */
174 assert(src
.dw1
.bits
.swizzle
== BRW_SWIZZLE_XYZW
);
178 vec4_generator::generate_math1_gen6(vec4_instruction
*inst
,
182 /* Can't do writemask because math can't be align16. */
183 assert(dst
.dw1
.bits
.writemask
== WRITEMASK_XYZW
);
184 check_gen6_math_src_arg(src
);
186 brw_set_access_mode(p
, BRW_ALIGN_1
);
189 brw_math_function(inst
->opcode
),
192 BRW_MATH_DATA_SCALAR
,
193 BRW_MATH_PRECISION_FULL
);
194 brw_set_access_mode(p
, BRW_ALIGN_16
);
198 vec4_generator::generate_math2_gen7(vec4_instruction
*inst
,
205 brw_math_function(inst
->opcode
),
210 vec4_generator::generate_math2_gen6(vec4_instruction
*inst
,
215 /* Can't do writemask because math can't be align16. */
216 assert(dst
.dw1
.bits
.writemask
== WRITEMASK_XYZW
);
217 /* Source swizzles are ignored. */
218 check_gen6_math_src_arg(src0
);
219 check_gen6_math_src_arg(src1
);
221 brw_set_access_mode(p
, BRW_ALIGN_1
);
224 brw_math_function(inst
->opcode
),
226 brw_set_access_mode(p
, BRW_ALIGN_16
);
230 vec4_generator::generate_math2_gen4(vec4_instruction
*inst
,
235 /* From the Ironlake PRM, Volume 4, Part 1, Section 6.1.13
238 * "Operand0[7]. For the INT DIV functions, this operand is the
241 * "Operand1[7]. For the INT DIV functions, this operand is the
244 bool is_int_div
= inst
->opcode
!= SHADER_OPCODE_POW
;
245 struct brw_reg
&op0
= is_int_div
? src1
: src0
;
246 struct brw_reg
&op1
= is_int_div
? src0
: src1
;
248 brw_push_insn_state(p
);
249 brw_set_saturate(p
, false);
250 brw_set_predicate_control(p
, BRW_PREDICATE_NONE
);
251 brw_MOV(p
, retype(brw_message_reg(inst
->base_mrf
+ 1), op1
.type
), op1
);
252 brw_pop_insn_state(p
);
256 brw_math_function(inst
->opcode
),
259 BRW_MATH_DATA_VECTOR
,
260 BRW_MATH_PRECISION_FULL
);
264 vec4_generator::generate_tex(vec4_instruction
*inst
,
270 if (intel
->gen
>= 5) {
271 switch (inst
->opcode
) {
272 case SHADER_OPCODE_TEX
:
273 case SHADER_OPCODE_TXL
:
274 if (inst
->shadow_compare
) {
275 msg_type
= GEN5_SAMPLER_MESSAGE_SAMPLE_LOD_COMPARE
;
277 msg_type
= GEN5_SAMPLER_MESSAGE_SAMPLE_LOD
;
280 case SHADER_OPCODE_TXD
:
281 if (inst
->shadow_compare
) {
282 /* Gen7.5+. Otherwise, lowered by brw_lower_texture_gradients(). */
283 assert(intel
->is_haswell
);
284 msg_type
= HSW_SAMPLER_MESSAGE_SAMPLE_DERIV_COMPARE
;
286 msg_type
= GEN5_SAMPLER_MESSAGE_SAMPLE_DERIVS
;
289 case SHADER_OPCODE_TXF
:
290 msg_type
= GEN5_SAMPLER_MESSAGE_SAMPLE_LD
;
292 case SHADER_OPCODE_TXF_MS
:
294 msg_type
= GEN7_SAMPLER_MESSAGE_SAMPLE_LD2DMS
;
296 msg_type
= GEN5_SAMPLER_MESSAGE_SAMPLE_LD
;
298 case SHADER_OPCODE_TXS
:
299 msg_type
= GEN5_SAMPLER_MESSAGE_SAMPLE_RESINFO
;
302 assert(!"should not get here: invalid VS texture opcode");
306 switch (inst
->opcode
) {
307 case SHADER_OPCODE_TEX
:
308 case SHADER_OPCODE_TXL
:
309 if (inst
->shadow_compare
) {
310 msg_type
= BRW_SAMPLER_MESSAGE_SIMD4X2_SAMPLE_LOD_COMPARE
;
311 assert(inst
->mlen
== 3);
313 msg_type
= BRW_SAMPLER_MESSAGE_SIMD4X2_SAMPLE_LOD
;
314 assert(inst
->mlen
== 2);
317 case SHADER_OPCODE_TXD
:
318 /* There is no sample_d_c message; comparisons are done manually. */
319 msg_type
= BRW_SAMPLER_MESSAGE_SIMD4X2_SAMPLE_GRADIENTS
;
320 assert(inst
->mlen
== 4);
322 case SHADER_OPCODE_TXF
:
323 msg_type
= BRW_SAMPLER_MESSAGE_SIMD4X2_LD
;
324 assert(inst
->mlen
== 2);
326 case SHADER_OPCODE_TXS
:
327 msg_type
= BRW_SAMPLER_MESSAGE_SIMD4X2_RESINFO
;
328 assert(inst
->mlen
== 2);
331 assert(!"should not get here: invalid VS texture opcode");
336 assert(msg_type
!= -1);
338 /* Load the message header if present. If there's a texture offset, we need
339 * to set it up explicitly and load the offset bitfield. Otherwise, we can
340 * use an implied move from g0 to the first message register.
342 if (inst
->texture_offset
) {
343 /* Explicitly set up the message header by copying g0 to the MRF. */
344 brw_push_insn_state(p
);
345 brw_set_mask_control(p
, BRW_MASK_DISABLE
);
346 brw_MOV(p
, retype(brw_message_reg(inst
->base_mrf
), BRW_REGISTER_TYPE_UD
),
347 retype(brw_vec8_grf(0, 0), BRW_REGISTER_TYPE_UD
));
349 /* Then set the offset bits in DWord 2. */
350 brw_set_access_mode(p
, BRW_ALIGN_1
);
352 retype(brw_vec1_reg(BRW_MESSAGE_REGISTER_FILE
, inst
->base_mrf
, 2),
353 BRW_REGISTER_TYPE_UD
),
354 brw_imm_uw(inst
->texture_offset
));
355 brw_pop_insn_state(p
);
356 } else if (inst
->header_present
) {
357 /* Set up an implied move from g0 to the MRF. */
358 src
= brw_vec8_grf(0, 0);
361 uint32_t return_format
;
364 case BRW_REGISTER_TYPE_D
:
365 return_format
= BRW_SAMPLER_RETURN_FORMAT_SINT32
;
367 case BRW_REGISTER_TYPE_UD
:
368 return_format
= BRW_SAMPLER_RETURN_FORMAT_UINT32
;
371 return_format
= BRW_SAMPLER_RETURN_FORMAT_FLOAT32
;
379 SURF_INDEX_VS_TEXTURE(inst
->sampler
),
382 1, /* response length */
384 inst
->header_present
,
385 BRW_SAMPLER_SIMD_MODE_SIMD4X2
,
390 vec4_generator::generate_urb_write(vec4_instruction
*inst
)
393 brw_null_reg(), /* dest */
394 inst
->base_mrf
, /* starting mrf reg nr */
395 brw_vec8_grf(0, 0), /* src */
396 false, /* allocate */
399 0, /* response len */
401 inst
->eot
, /* writes complete */
402 inst
->offset
, /* urb destination offset */
403 BRW_URB_SWIZZLE_INTERLEAVE
);
407 vec4_generator::generate_oword_dual_block_offsets(struct brw_reg m1
,
408 struct brw_reg index
)
410 int second_vertex_offset
;
413 second_vertex_offset
= 1;
415 second_vertex_offset
= 16;
417 m1
= retype(m1
, BRW_REGISTER_TYPE_D
);
419 /* Set up M1 (message payload). Only the block offsets in M1.0 and
420 * M1.4 are used, and the rest are ignored.
422 struct brw_reg m1_0
= suboffset(vec1(m1
), 0);
423 struct brw_reg m1_4
= suboffset(vec1(m1
), 4);
424 struct brw_reg index_0
= suboffset(vec1(index
), 0);
425 struct brw_reg index_4
= suboffset(vec1(index
), 4);
427 brw_push_insn_state(p
);
428 brw_set_mask_control(p
, BRW_MASK_DISABLE
);
429 brw_set_access_mode(p
, BRW_ALIGN_1
);
431 brw_MOV(p
, m1_0
, index_0
);
433 if (index
.file
== BRW_IMMEDIATE_VALUE
) {
434 index_4
.dw1
.ud
+= second_vertex_offset
;
435 brw_MOV(p
, m1_4
, index_4
);
437 brw_ADD(p
, m1_4
, index_4
, brw_imm_d(second_vertex_offset
));
440 brw_pop_insn_state(p
);
444 vec4_generator::generate_scratch_read(vec4_instruction
*inst
,
446 struct brw_reg index
)
448 struct brw_reg header
= brw_vec8_grf(0, 0);
450 gen6_resolve_implied_move(p
, &header
, inst
->base_mrf
);
452 generate_oword_dual_block_offsets(brw_message_reg(inst
->base_mrf
+ 1),
458 msg_type
= GEN6_DATAPORT_READ_MESSAGE_OWORD_DUAL_BLOCK_READ
;
459 else if (intel
->gen
== 5 || intel
->is_g4x
)
460 msg_type
= G45_DATAPORT_READ_MESSAGE_OWORD_DUAL_BLOCK_READ
;
462 msg_type
= BRW_DATAPORT_READ_MESSAGE_OWORD_DUAL_BLOCK_READ
;
464 /* Each of the 8 channel enables is considered for whether each
467 struct brw_instruction
*send
= brw_next_insn(p
, BRW_OPCODE_SEND
);
468 brw_set_dest(p
, send
, dst
);
469 brw_set_src0(p
, send
, header
);
471 send
->header
.destreg__conditionalmod
= inst
->base_mrf
;
472 brw_set_dp_read_message(p
, send
,
473 255, /* binding table index: stateless access */
474 BRW_DATAPORT_OWORD_DUAL_BLOCK_1OWORD
,
476 BRW_DATAPORT_READ_TARGET_RENDER_CACHE
,
478 true, /* header_present */
483 vec4_generator::generate_scratch_write(vec4_instruction
*inst
,
486 struct brw_reg index
)
488 struct brw_reg header
= brw_vec8_grf(0, 0);
491 /* If the instruction is predicated, we'll predicate the send, not
494 brw_set_predicate_control(p
, false);
496 gen6_resolve_implied_move(p
, &header
, inst
->base_mrf
);
498 generate_oword_dual_block_offsets(brw_message_reg(inst
->base_mrf
+ 1),
502 retype(brw_message_reg(inst
->base_mrf
+ 2), BRW_REGISTER_TYPE_D
),
503 retype(src
, BRW_REGISTER_TYPE_D
));
508 msg_type
= GEN7_DATAPORT_WRITE_MESSAGE_OWORD_DUAL_BLOCK_WRITE
;
509 else if (intel
->gen
== 6)
510 msg_type
= GEN6_DATAPORT_WRITE_MESSAGE_OWORD_DUAL_BLOCK_WRITE
;
512 msg_type
= BRW_DATAPORT_WRITE_MESSAGE_OWORD_DUAL_BLOCK_WRITE
;
514 brw_set_predicate_control(p
, inst
->predicate
);
516 /* Pre-gen6, we have to specify write commits to ensure ordering
517 * between reads and writes within a thread. Afterwards, that's
518 * guaranteed and write commits only matter for inter-thread
521 if (intel
->gen
>= 6) {
522 write_commit
= false;
524 /* The visitor set up our destination register to be g0. This
525 * means that when the next read comes along, we will end up
526 * reading from g0 and causing a block on the write commit. For
527 * write-after-read, we are relying on the value of the previous
528 * read being used (and thus blocking on completion) before our
529 * write is executed. This means we have to be careful in
530 * instruction scheduling to not violate this assumption.
535 /* Each of the 8 channel enables is considered for whether each
538 struct brw_instruction
*send
= brw_next_insn(p
, BRW_OPCODE_SEND
);
539 brw_set_dest(p
, send
, dst
);
540 brw_set_src0(p
, send
, header
);
542 send
->header
.destreg__conditionalmod
= inst
->base_mrf
;
543 brw_set_dp_write_message(p
, send
,
544 255, /* binding table index: stateless access */
545 BRW_DATAPORT_OWORD_DUAL_BLOCK_1OWORD
,
548 true, /* header present */
549 false, /* not a render target write */
550 write_commit
, /* rlen */
556 vec4_generator::generate_pull_constant_load(vec4_instruction
*inst
,
558 struct brw_reg index
,
559 struct brw_reg offset
)
561 assert(intel
->gen
<= 7);
562 assert(index
.file
== BRW_IMMEDIATE_VALUE
&&
563 index
.type
== BRW_REGISTER_TYPE_UD
);
564 uint32_t surf_index
= index
.dw1
.ud
;
566 struct brw_reg header
= brw_vec8_grf(0, 0);
568 gen6_resolve_implied_move(p
, &header
, inst
->base_mrf
);
570 brw_MOV(p
, retype(brw_message_reg(inst
->base_mrf
+ 1), BRW_REGISTER_TYPE_D
),
576 msg_type
= GEN6_DATAPORT_READ_MESSAGE_OWORD_DUAL_BLOCK_READ
;
577 else if (intel
->gen
== 5 || intel
->is_g4x
)
578 msg_type
= G45_DATAPORT_READ_MESSAGE_OWORD_DUAL_BLOCK_READ
;
580 msg_type
= BRW_DATAPORT_READ_MESSAGE_OWORD_DUAL_BLOCK_READ
;
582 /* Each of the 8 channel enables is considered for whether each
585 struct brw_instruction
*send
= brw_next_insn(p
, BRW_OPCODE_SEND
);
586 brw_set_dest(p
, send
, dst
);
587 brw_set_src0(p
, send
, header
);
589 send
->header
.destreg__conditionalmod
= inst
->base_mrf
;
590 brw_set_dp_read_message(p
, send
,
592 BRW_DATAPORT_OWORD_DUAL_BLOCK_1OWORD
,
594 BRW_DATAPORT_READ_TARGET_DATA_CACHE
,
596 true, /* header_present */
601 vec4_generator::generate_pull_constant_load_gen7(vec4_instruction
*inst
,
603 struct brw_reg surf_index
,
604 struct brw_reg offset
)
606 assert(surf_index
.file
== BRW_IMMEDIATE_VALUE
&&
607 surf_index
.type
== BRW_REGISTER_TYPE_UD
);
609 brw_instruction
*insn
= brw_next_insn(p
, BRW_OPCODE_SEND
);
610 brw_set_dest(p
, insn
, dst
);
611 brw_set_src0(p
, insn
, offset
);
612 brw_set_sampler_message(p
, insn
,
614 0, /* LD message ignores sampler unit */
615 GEN5_SAMPLER_MESSAGE_SAMPLE_LD
,
618 false, /* no header */
619 BRW_SAMPLER_SIMD_MODE_SIMD4X2
,
624 vec4_generator::generate_vs_instruction(vec4_instruction
*instruction
,
628 vec4_instruction
*inst
= (vec4_instruction
*)instruction
;
630 switch (inst
->opcode
) {
631 case SHADER_OPCODE_RCP
:
632 case SHADER_OPCODE_RSQ
:
633 case SHADER_OPCODE_SQRT
:
634 case SHADER_OPCODE_EXP2
:
635 case SHADER_OPCODE_LOG2
:
636 case SHADER_OPCODE_SIN
:
637 case SHADER_OPCODE_COS
:
638 if (intel
->gen
== 6) {
639 generate_math1_gen6(inst
, dst
, src
[0]);
641 /* Also works for Gen7. */
642 generate_math1_gen4(inst
, dst
, src
[0]);
646 case SHADER_OPCODE_POW
:
647 case SHADER_OPCODE_INT_QUOTIENT
:
648 case SHADER_OPCODE_INT_REMAINDER
:
649 if (intel
->gen
>= 7) {
650 generate_math2_gen7(inst
, dst
, src
[0], src
[1]);
651 } else if (intel
->gen
== 6) {
652 generate_math2_gen6(inst
, dst
, src
[0], src
[1]);
654 generate_math2_gen4(inst
, dst
, src
[0], src
[1]);
658 case SHADER_OPCODE_TEX
:
659 case SHADER_OPCODE_TXD
:
660 case SHADER_OPCODE_TXF
:
661 case SHADER_OPCODE_TXF_MS
:
662 case SHADER_OPCODE_TXL
:
663 case SHADER_OPCODE_TXS
:
664 generate_tex(inst
, dst
, src
[0]);
667 case VS_OPCODE_URB_WRITE
:
668 generate_urb_write(inst
);
671 case VS_OPCODE_SCRATCH_READ
:
672 generate_scratch_read(inst
, dst
, src
[0]);
675 case VS_OPCODE_SCRATCH_WRITE
:
676 generate_scratch_write(inst
, dst
, src
[0], src
[1]);
679 case VS_OPCODE_PULL_CONSTANT_LOAD
:
680 generate_pull_constant_load(inst
, dst
, src
[0], src
[1]);
683 case VS_OPCODE_PULL_CONSTANT_LOAD_GEN7
:
684 generate_pull_constant_load_gen7(inst
, dst
, src
[0], src
[1]);
687 case SHADER_OPCODE_SHADER_TIME_ADD
:
688 brw_shader_time_add(p
, src
[0], SURF_INDEX_VS_SHADER_TIME
);
692 if (inst
->opcode
< (int) ARRAY_SIZE(opcode_descs
)) {
693 _mesa_problem(ctx
, "Unsupported opcode in `%s' in VS\n",
694 opcode_descs
[inst
->opcode
].name
);
696 _mesa_problem(ctx
, "Unsupported opcode %d in VS", inst
->opcode
);
703 vec4_generator::generate_code(exec_list
*instructions
)
705 int last_native_insn_offset
= 0;
706 const char *last_annotation_string
= NULL
;
707 const void *last_annotation_ir
= NULL
;
709 if (unlikely(INTEL_DEBUG
& DEBUG_VS
)) {
711 printf("Native code for vertex shader %d:\n", prog
->Name
);
713 printf("Native code for vertex program %d:\n", c
->vp
->program
.Base
.Id
);
717 foreach_list(node
, instructions
) {
718 vec4_instruction
*inst
= (vec4_instruction
*)node
;
719 struct brw_reg src
[3], dst
;
721 if (unlikely(INTEL_DEBUG
& DEBUG_VS
)) {
722 if (last_annotation_ir
!= inst
->ir
) {
723 last_annotation_ir
= inst
->ir
;
724 if (last_annotation_ir
) {
727 ((ir_instruction
*) last_annotation_ir
)->print();
729 const prog_instruction
*vpi
;
730 vpi
= (const prog_instruction
*) inst
->ir
;
731 printf("%d: ", (int)(vpi
- vp
->Base
.Instructions
));
732 _mesa_fprint_instruction_opt(stdout
, vpi
, 0,
733 PROG_PRINT_DEBUG
, NULL
);
738 if (last_annotation_string
!= inst
->annotation
) {
739 last_annotation_string
= inst
->annotation
;
740 if (last_annotation_string
)
741 printf(" %s\n", last_annotation_string
);
745 for (unsigned int i
= 0; i
< 3; i
++) {
746 src
[i
] = inst
->get_src(i
);
748 dst
= inst
->get_dst();
750 brw_set_conditionalmod(p
, inst
->conditional_mod
);
751 brw_set_predicate_control(p
, inst
->predicate
);
752 brw_set_predicate_inverse(p
, inst
->predicate_inverse
);
753 brw_set_saturate(p
, inst
->saturate
);
754 brw_set_mask_control(p
, inst
->force_writemask_all
);
756 unsigned pre_emit_nr_insn
= p
->nr_insn
;
758 switch (inst
->opcode
) {
760 brw_MOV(p
, dst
, src
[0]);
763 brw_ADD(p
, dst
, src
[0], src
[1]);
766 brw_MUL(p
, dst
, src
[0], src
[1]);
768 case BRW_OPCODE_MACH
:
769 brw_set_acc_write_control(p
, 1);
770 brw_MACH(p
, dst
, src
[0], src
[1]);
771 brw_set_acc_write_control(p
, 0);
775 brw_FRC(p
, dst
, src
[0]);
777 case BRW_OPCODE_RNDD
:
778 brw_RNDD(p
, dst
, src
[0]);
780 case BRW_OPCODE_RNDE
:
781 brw_RNDE(p
, dst
, src
[0]);
783 case BRW_OPCODE_RNDZ
:
784 brw_RNDZ(p
, dst
, src
[0]);
788 brw_AND(p
, dst
, src
[0], src
[1]);
791 brw_OR(p
, dst
, src
[0], src
[1]);
794 brw_XOR(p
, dst
, src
[0], src
[1]);
797 brw_NOT(p
, dst
, src
[0]);
800 brw_ASR(p
, dst
, src
[0], src
[1]);
803 brw_SHR(p
, dst
, src
[0], src
[1]);
806 brw_SHL(p
, dst
, src
[0], src
[1]);
810 brw_CMP(p
, dst
, inst
->conditional_mod
, src
[0], src
[1]);
813 brw_SEL(p
, dst
, src
[0], src
[1]);
817 brw_DPH(p
, dst
, src
[0], src
[1]);
821 brw_DP4(p
, dst
, src
[0], src
[1]);
825 brw_DP3(p
, dst
, src
[0], src
[1]);
829 brw_DP2(p
, dst
, src
[0], src
[1]);
832 case BRW_OPCODE_F32TO16
:
833 brw_F32TO16(p
, dst
, src
[0]);
836 case BRW_OPCODE_F16TO32
:
837 brw_F16TO32(p
, dst
, src
[0]);
841 if (inst
->src
[0].file
!= BAD_FILE
) {
842 /* The instruction has an embedded compare (only allowed on gen6) */
843 assert(intel
->gen
== 6);
844 gen6_IF(p
, inst
->conditional_mod
, src
[0], src
[1]);
846 struct brw_instruction
*brw_inst
= brw_IF(p
, BRW_EXECUTE_8
);
847 brw_inst
->header
.predicate_control
= inst
->predicate
;
851 case BRW_OPCODE_ELSE
:
854 case BRW_OPCODE_ENDIF
:
859 brw_DO(p
, BRW_EXECUTE_8
);
862 case BRW_OPCODE_BREAK
:
864 brw_set_predicate_control(p
, BRW_PREDICATE_NONE
);
866 case BRW_OPCODE_CONTINUE
:
867 /* FINISHME: We need to write the loop instruction support still. */
872 brw_set_predicate_control(p
, BRW_PREDICATE_NONE
);
875 case BRW_OPCODE_WHILE
:
880 generate_vs_instruction(inst
, dst
, src
);
884 if (inst
->no_dd_clear
|| inst
->no_dd_check
) {
885 assert(p
->nr_insn
== pre_emit_nr_insn
+ 1 ||
886 !"no_dd_check or no_dd_clear set for IR emitting more "
887 "than 1 instruction");
889 struct brw_instruction
*last
= &p
->store
[pre_emit_nr_insn
];
891 if (inst
->no_dd_clear
)
892 last
->header
.dependency_control
|= BRW_DEPENDENCY_NOTCLEARED
;
893 if (inst
->no_dd_check
)
894 last
->header
.dependency_control
|= BRW_DEPENDENCY_NOTCHECKED
;
897 if (unlikely(INTEL_DEBUG
& DEBUG_VS
)) {
898 brw_dump_compile(p
, stdout
,
899 last_native_insn_offset
, p
->next_insn_offset
);
902 last_native_insn_offset
= p
->next_insn_offset
;
905 if (unlikely(INTEL_DEBUG
& DEBUG_VS
)) {
911 /* OK, while the INTEL_DEBUG=vs above is very nice for debugging VS
912 * emit issues, it doesn't get the jump distances into the output,
913 * which is often something we want to debug. So this is here in
914 * case you're doing that.
916 if (0 && unlikely(INTEL_DEBUG
& DEBUG_VS
)) {
917 brw_dump_compile(p
, stdout
, 0, p
->next_insn_offset
);
922 vec4_generator::generate_assembly(exec_list
*instructions
,
923 unsigned *assembly_size
)
925 brw_set_access_mode(p
, BRW_ALIGN_16
);
926 generate_code(instructions
);
927 return brw_get_program(p
, assembly_size
);
930 } /* namespace brw */