2 * Copyright 2010 Jerome Glisse <glisse@freedesktop.org>
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 * on the rights to use, copy, modify, merge, publish, distribute, sub
8 * license, and/or sell copies of the Software, and to permit persons to whom
9 * the 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 NON-INFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
19 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
20 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
21 * USE OR OTHER DEALINGS IN THE SOFTWARE.
24 #include "r600_context.h"
25 #include "util/u_memory.h"
27 #include "r600_opcodes.h"
31 static inline unsigned int r600_bc_get_num_operands(struct r600_bc_alu
*alu
)
37 case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
:
39 case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_ADD
:
40 case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_KILLE
:
41 case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_KILLGT
:
42 case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_KILLGE
:
43 case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_KILLNE
:
44 case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MUL
:
45 case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MAX
:
46 case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MIN
:
47 case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETE
:
48 case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETNE
:
49 case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETGT
:
50 case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETGE
:
51 case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_PRED_SETE
:
52 case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_PRED_SETGT
:
53 case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_PRED_SETGE
:
54 case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_PRED_SETNE
:
55 case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_DOT4
:
56 case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_DOT4_IEEE
:
57 case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_CUBE
:
60 case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV
:
61 case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOVA_FLOOR
:
62 case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FRACT
:
63 case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLOOR
:
64 case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_TRUNC
:
65 case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_EXP_IEEE
:
66 case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_LOG_CLAMPED
:
67 case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_LOG_IEEE
:
68 case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_RECIP_IEEE
:
69 case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_RECIPSQRT_IEEE
:
70 case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLT_TO_INT
:
71 case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SIN
:
72 case V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_COS
:
76 "Need instruction operand number for 0x%x.\n", alu
->inst
);
82 int r700_bc_alu_build(struct r600_bc
*bc
, struct r600_bc_alu
*alu
, unsigned id
);
84 static struct r600_bc_cf
*r600_bc_cf(void)
86 struct r600_bc_cf
*cf
= CALLOC_STRUCT(r600_bc_cf
);
90 LIST_INITHEAD(&cf
->list
);
91 LIST_INITHEAD(&cf
->alu
);
92 LIST_INITHEAD(&cf
->vtx
);
93 LIST_INITHEAD(&cf
->tex
);
97 static struct r600_bc_alu
*r600_bc_alu(void)
99 struct r600_bc_alu
*alu
= CALLOC_STRUCT(r600_bc_alu
);
103 LIST_INITHEAD(&alu
->list
);
104 LIST_INITHEAD(&alu
->bs_list
);
108 static struct r600_bc_vtx
*r600_bc_vtx(void)
110 struct r600_bc_vtx
*vtx
= CALLOC_STRUCT(r600_bc_vtx
);
114 LIST_INITHEAD(&vtx
->list
);
118 static struct r600_bc_tex
*r600_bc_tex(void)
120 struct r600_bc_tex
*tex
= CALLOC_STRUCT(r600_bc_tex
);
124 LIST_INITHEAD(&tex
->list
);
128 int r600_bc_init(struct r600_bc
*bc
, enum radeon_family family
)
130 LIST_INITHEAD(&bc
->cf
);
132 switch (bc
->family
) {
157 R600_ERR("unknown family %d\n", bc
->family
);
163 static int r600_bc_add_cf(struct r600_bc
*bc
)
165 struct r600_bc_cf
*cf
= r600_bc_cf();
169 LIST_ADDTAIL(&cf
->list
, &bc
->cf
);
171 cf
->id
= bc
->cf_last
->id
+ 2;
175 bc
->force_add_cf
= 0;
179 int r600_bc_add_output(struct r600_bc
*bc
, const struct r600_bc_output
*output
)
183 r
= r600_bc_add_cf(bc
);
186 bc
->cf_last
->inst
= output
->inst
;
187 memcpy(&bc
->cf_last
->output
, output
, sizeof(struct r600_bc_output
));
191 const unsigned bank_swizzle_vec
[8] = {SQ_ALU_VEC_210
, //000
192 SQ_ALU_VEC_120
, //001
193 SQ_ALU_VEC_102
, //010
195 SQ_ALU_VEC_201
, //011
196 SQ_ALU_VEC_012
, //100
197 SQ_ALU_VEC_021
, //101
199 SQ_ALU_VEC_012
, //110
200 SQ_ALU_VEC_012
}; //111
202 const unsigned bank_swizzle_scl
[8] = {SQ_ALU_SCL_210
, //000
203 SQ_ALU_SCL_122
, //001
204 SQ_ALU_SCL_122
, //010
206 SQ_ALU_SCL_221
, //011
207 SQ_ALU_SCL_212
, //100
208 SQ_ALU_SCL_122
, //101
210 SQ_ALU_SCL_122
, //110
211 SQ_ALU_SCL_122
}; //111
213 static int init_gpr(struct r600_bc_alu
*alu
)
215 int cycle
, component
;
217 for (cycle
= 0; cycle
< NUM_OF_CYCLES
; cycle
++)
218 for (component
= 0; component
< NUM_OF_COMPONENTS
; component
++)
219 alu
->hw_gpr
[cycle
][component
] = -1;
223 static int reserve_gpr(struct r600_bc_alu
*alu
, unsigned sel
, unsigned chan
, unsigned cycle
)
225 if (alu
->hw_gpr
[cycle
][chan
] < 0)
226 alu
->hw_gpr
[cycle
][chan
] = sel
;
227 else if (alu
->hw_gpr
[cycle
][chan
] != (int)sel
) {
228 R600_ERR("Another scalar operation has already used GPR read port for channel\n");
234 static int cycle_for_scalar_bank_swizzle(const int swiz
, const int sel
, unsigned *p_cycle
)
240 table
[0] = 2; table
[1] = 1; table
[2] = 0;
241 *p_cycle
= table
[sel
];
244 table
[0] = 1; table
[1] = 2; table
[2] = 2;
245 *p_cycle
= table
[sel
];
248 table
[0] = 2; table
[1] = 1; table
[2] = 2;
249 *p_cycle
= table
[sel
];
252 table
[0] = 2; table
[1] = 2; table
[2] = 1;
253 *p_cycle
= table
[sel
];
257 R600_ERR("bad scalar bank swizzle value\n");
264 static int cycle_for_vector_bank_swizzle(const int swiz
, const int sel
, unsigned *p_cycle
)
271 table
[0] = 0; table
[1] = 1; table
[2] = 2;
272 *p_cycle
= table
[sel
];
275 table
[0] = 0; table
[1] = 2; table
[2] = 1;
276 *p_cycle
= table
[sel
];
279 table
[0] = 1; table
[1] = 2; table
[2] = 0;
280 *p_cycle
= table
[sel
];
283 table
[0] = 1; table
[1] = 0; table
[2] = 2;
284 *p_cycle
= table
[sel
];
287 table
[0] = 2; table
[1] = 0; table
[2] = 1;
288 *p_cycle
= table
[sel
];
291 table
[0] = 2; table
[1] = 1; table
[2] = 0;
292 *p_cycle
= table
[sel
];
295 R600_ERR("bad vector bank swizzle value\n");
302 static int is_const(int sel
)
304 if (sel
> 255 && sel
< 512)
306 if (sel
>= V_SQ_ALU_SRC_0
&& sel
<= V_SQ_ALU_SRC_LITERAL
)
311 static void update_chan_counter(struct r600_bc_alu
*alu
, int *chan_counter
)
317 num_src
= r600_bc_get_num_operands(alu
);
319 for (i
= 0; i
< num_src
; i
++) {
320 channel_swizzle
= alu
->src
[i
].chan
;
321 if ((alu
->src
[i
].sel
> 0 && alu
->src
[i
].sel
< 128) && channel_swizzle
<= 3)
322 chan_counter
[channel_swizzle
]++;
327 /* we need something like this I think - but this is bogus */
328 int check_read_slots(struct r600_bc
*bc
, struct r600_bc_alu
*alu_first
)
330 struct r600_bc_alu
*alu
;
331 int chan_counter
[4] = { 0 };
333 update_chan_counter(alu_first
, chan_counter
);
335 LIST_FOR_EACH_ENTRY(alu
, &alu_first
->bs_list
, bs_list
) {
336 update_chan_counter(alu
, chan_counter
);
339 if (chan_counter
[0] > 3 ||
340 chan_counter
[1] > 3 ||
341 chan_counter
[2] > 3 ||
342 chan_counter
[3] > 3) {
343 R600_ERR("needed to split instruction for input ran out of banks %x %d %d %d %d\n",
344 alu_first
->inst
, chan_counter
[0], chan_counter
[1], chan_counter
[2], chan_counter
[3]);
351 static int check_scalar(struct r600_bc
*bc
, struct r600_bc_alu
*alu
)
353 unsigned swizzle_key
;
355 if (alu
->bank_swizzle_force
) {
356 alu
->bank_swizzle
= alu
->bank_swizzle_force
;
359 swizzle_key
= (is_const(alu
->src
[0].sel
) ? 4 : 0 ) +
360 (is_const(alu
->src
[1].sel
) ? 2 : 0 ) +
361 (is_const(alu
->src
[2].sel
) ? 1 : 0 );
363 alu
->bank_swizzle
= bank_swizzle_scl
[swizzle_key
];
367 static int check_vector(struct r600_bc
*bc
, struct r600_bc_alu
*alu
)
369 unsigned swizzle_key
;
371 if (alu
->bank_swizzle_force
) {
372 alu
->bank_swizzle
= alu
->bank_swizzle_force
;
375 swizzle_key
= (is_const(alu
->src
[0].sel
) ? 4 : 0 ) +
376 (is_const(alu
->src
[1].sel
) ? 2 : 0 ) +
377 (is_const(alu
->src
[2].sel
) ? 1 : 0 );
379 alu
->bank_swizzle
= bank_swizzle_vec
[swizzle_key
];
383 static int check_and_set_bank_swizzle(struct r600_bc
*bc
, struct r600_bc_alu
*alu_first
)
385 struct r600_bc_alu
*alu
;
390 LIST_FOR_EACH_ENTRY(alu
, &alu_first
->bs_list
, bs_list
) {
394 if (num_instr
== 1) {
395 check_scalar(bc
, alu_first
);
398 /* check_read_slots(bc, bc->cf_last->curr_bs_head);*/
399 check_vector(bc
, alu_first
);
400 LIST_FOR_EACH_ENTRY(alu
, &alu_first
->bs_list
, bs_list
) {
401 check_vector(bc
, alu
);
407 int r600_bc_add_alu_type(struct r600_bc
*bc
, const struct r600_bc_alu
*alu
, int type
)
409 struct r600_bc_alu
*nalu
= r600_bc_alu();
410 struct r600_bc_alu
*lalu
;
411 struct r600_bc_alu
*curr_bs_head
;
416 memcpy(nalu
, alu
, sizeof(struct r600_bc_alu
));
419 /* cf can contains only alu or only vtx or only tex */
420 if (bc
->cf_last
== NULL
|| bc
->cf_last
->inst
!= (type
<< 3) ||
422 /* at most 128 slots, one add alu can add 4 slots + 4 constant worst case */
423 r
= r600_bc_add_cf(bc
);
428 bc
->cf_last
->inst
= (type
<< 3);
430 if (!bc
->cf_last
->curr_bs_head
) {
431 bc
->cf_last
->curr_bs_head
= nalu
;
432 LIST_INITHEAD(&nalu
->bs_list
);
434 LIST_ADDTAIL(&nalu
->bs_list
, &bc
->cf_last
->curr_bs_head
->bs_list
);
436 if (alu
->last
&& (bc
->cf_last
->ndw
>> 1) >= 124) {
437 bc
->force_add_cf
= 1;
439 /* number of gpr == the last gpr used in any alu */
440 for (i
= 0; i
< 3; i
++) {
441 if (alu
->src
[i
].sel
>= bc
->ngpr
&& alu
->src
[i
].sel
< 128) {
442 bc
->ngpr
= alu
->src
[i
].sel
+ 1;
444 /* compute how many literal are needed
445 * either 2 or 4 literals
447 if (alu
->src
[i
].sel
== 253) {
448 if (((alu
->src
[i
].chan
+ 2) & 0x6) > nalu
->nliteral
) {
449 nalu
->nliteral
= (alu
->src
[i
].chan
+ 2) & 0x6;
453 if (!LIST_IS_EMPTY(&bc
->cf_last
->alu
)) {
454 lalu
= LIST_ENTRY(struct r600_bc_alu
, bc
->cf_last
->alu
.prev
, list
);
455 if (!lalu
->last
&& lalu
->nliteral
> nalu
->nliteral
) {
456 nalu
->nliteral
= lalu
->nliteral
;
459 if (alu
->dst
.sel
>= bc
->ngpr
) {
460 bc
->ngpr
= alu
->dst
.sel
+ 1;
462 LIST_ADDTAIL(&nalu
->list
, &bc
->cf_last
->alu
);
463 /* each alu use 2 dwords */
464 bc
->cf_last
->ndw
+= 2;
467 if (bc
->use_mem_constant
)
468 bc
->cf_last
->kcache0_mode
= 2;
470 /* process cur ALU instructions for bank swizzle */
472 check_and_set_bank_swizzle(bc
, bc
->cf_last
->curr_bs_head
);
473 bc
->cf_last
->curr_bs_head
= NULL
;
478 int r600_bc_add_alu(struct r600_bc
*bc
, const struct r600_bc_alu
*alu
)
480 return r600_bc_add_alu_type(bc
, alu
, BC_INST(bc
, V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU
));
483 int r600_bc_add_literal(struct r600_bc
*bc
, const u32
*value
)
485 struct r600_bc_alu
*alu
;
487 if (bc
->cf_last
== NULL
) {
490 if (bc
->cf_last
->inst
== V_SQ_CF_WORD1_SQ_CF_INST_TEX
) {
494 if (bc
->cf_last
->inst
== V_SQ_CF_WORD1_SQ_CF_INST_JUMP
||
495 bc
->cf_last
->inst
== V_SQ_CF_WORD1_SQ_CF_INST_ELSE
||
496 bc
->cf_last
->inst
== V_SQ_CF_WORD1_SQ_CF_INST_LOOP_START_NO_AL
||
497 bc
->cf_last
->inst
== V_SQ_CF_WORD1_SQ_CF_INST_LOOP_BREAK
||
498 bc
->cf_last
->inst
== V_SQ_CF_WORD1_SQ_CF_INST_LOOP_CONTINUE
||
499 bc
->cf_last
->inst
== V_SQ_CF_WORD1_SQ_CF_INST_LOOP_END
||
500 bc
->cf_last
->inst
== V_SQ_CF_WORD1_SQ_CF_INST_POP
) {
504 if (((bc
->cf_last
->inst
!= (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU
<< 3)) &&
505 (bc
->cf_last
->inst
!= (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_PUSH_BEFORE
<< 3))) ||
506 LIST_IS_EMPTY(&bc
->cf_last
->alu
)) {
507 R600_ERR("last CF is not ALU (%p)\n", bc
->cf_last
);
510 alu
= LIST_ENTRY(struct r600_bc_alu
, bc
->cf_last
->alu
.prev
, list
);
511 if (!alu
->last
|| !alu
->nliteral
|| alu
->literal_added
) {
514 memcpy(alu
->value
, value
, 4 * 4);
515 bc
->cf_last
->ndw
+= alu
->nliteral
;
516 bc
->ndw
+= alu
->nliteral
;
517 alu
->literal_added
= 1;
521 int r600_bc_add_vtx(struct r600_bc
*bc
, const struct r600_bc_vtx
*vtx
)
523 struct r600_bc_vtx
*nvtx
= r600_bc_vtx();
528 memcpy(nvtx
, vtx
, sizeof(struct r600_bc_vtx
));
530 /* cf can contains only alu or only vtx or only tex */
531 if (bc
->cf_last
== NULL
||
532 (bc
->cf_last
->inst
!= V_SQ_CF_WORD1_SQ_CF_INST_VTX
&&
533 bc
->cf_last
->inst
!= V_SQ_CF_WORD1_SQ_CF_INST_VTX_TC
)) {
534 r
= r600_bc_add_cf(bc
);
539 bc
->cf_last
->inst
= V_SQ_CF_WORD1_SQ_CF_INST_VTX
;
541 LIST_ADDTAIL(&nvtx
->list
, &bc
->cf_last
->vtx
);
542 /* each fetch use 4 dwords */
543 bc
->cf_last
->ndw
+= 4;
548 int r600_bc_add_tex(struct r600_bc
*bc
, const struct r600_bc_tex
*tex
)
550 struct r600_bc_tex
*ntex
= r600_bc_tex();
555 memcpy(ntex
, tex
, sizeof(struct r600_bc_tex
));
557 /* cf can contains only alu or only vtx or only tex */
558 if (bc
->cf_last
== NULL
||
559 bc
->cf_last
->inst
!= V_SQ_CF_WORD1_SQ_CF_INST_TEX
) {
560 r
= r600_bc_add_cf(bc
);
565 bc
->cf_last
->inst
= V_SQ_CF_WORD1_SQ_CF_INST_TEX
;
567 LIST_ADDTAIL(&ntex
->list
, &bc
->cf_last
->tex
);
568 /* each texture fetch use 4 dwords */
569 bc
->cf_last
->ndw
+= 4;
574 int r600_bc_add_cfinst(struct r600_bc
*bc
, int inst
)
577 r
= r600_bc_add_cf(bc
);
581 bc
->cf_last
->cond
= V_SQ_CF_COND_ACTIVE
;
582 bc
->cf_last
->inst
= inst
;
586 /* common to all 3 families */
587 static int r600_bc_vtx_build(struct r600_bc
*bc
, struct r600_bc_vtx
*vtx
, unsigned id
)
589 bc
->bytecode
[id
++] = S_SQ_VTX_WORD0_BUFFER_ID(vtx
->buffer_id
) |
590 S_SQ_VTX_WORD0_SRC_GPR(vtx
->src_gpr
) |
591 S_SQ_VTX_WORD0_SRC_SEL_X(vtx
->src_sel_x
) |
592 S_SQ_VTX_WORD0_MEGA_FETCH_COUNT(vtx
->mega_fetch_count
);
593 bc
->bytecode
[id
++] = S_SQ_VTX_WORD1_DST_SEL_X(vtx
->dst_sel_x
) |
594 S_SQ_VTX_WORD1_DST_SEL_Y(vtx
->dst_sel_y
) |
595 S_SQ_VTX_WORD1_DST_SEL_Z(vtx
->dst_sel_z
) |
596 S_SQ_VTX_WORD1_DST_SEL_W(vtx
->dst_sel_w
) |
597 S_SQ_VTX_WORD1_USE_CONST_FIELDS(1) |
598 S_SQ_VTX_WORD1_GPR_DST_GPR(vtx
->dst_gpr
);
599 bc
->bytecode
[id
++] = S_SQ_VTX_WORD2_MEGA_FETCH(1);
600 bc
->bytecode
[id
++] = 0;
604 /* common to all 3 families */
605 static int r600_bc_tex_build(struct r600_bc
*bc
, struct r600_bc_tex
*tex
, unsigned id
)
607 bc
->bytecode
[id
++] = S_SQ_TEX_WORD0_TEX_INST(tex
->inst
) |
608 S_SQ_TEX_WORD0_RESOURCE_ID(tex
->resource_id
) |
609 S_SQ_TEX_WORD0_SRC_GPR(tex
->src_gpr
) |
610 S_SQ_TEX_WORD0_SRC_REL(tex
->src_rel
);
611 bc
->bytecode
[id
++] = S_SQ_TEX_WORD1_DST_GPR(tex
->dst_gpr
) |
612 S_SQ_TEX_WORD1_DST_REL(tex
->dst_rel
) |
613 S_SQ_TEX_WORD1_DST_SEL_X(tex
->dst_sel_x
) |
614 S_SQ_TEX_WORD1_DST_SEL_Y(tex
->dst_sel_y
) |
615 S_SQ_TEX_WORD1_DST_SEL_Z(tex
->dst_sel_z
) |
616 S_SQ_TEX_WORD1_DST_SEL_W(tex
->dst_sel_w
) |
617 S_SQ_TEX_WORD1_LOD_BIAS(tex
->lod_bias
) |
618 S_SQ_TEX_WORD1_COORD_TYPE_X(tex
->coord_type_x
) |
619 S_SQ_TEX_WORD1_COORD_TYPE_Y(tex
->coord_type_y
) |
620 S_SQ_TEX_WORD1_COORD_TYPE_Z(tex
->coord_type_z
) |
621 S_SQ_TEX_WORD1_COORD_TYPE_W(tex
->coord_type_w
);
622 bc
->bytecode
[id
++] = S_SQ_TEX_WORD2_OFFSET_X(tex
->offset_x
) |
623 S_SQ_TEX_WORD2_OFFSET_Y(tex
->offset_y
) |
624 S_SQ_TEX_WORD2_OFFSET_Z(tex
->offset_z
) |
625 S_SQ_TEX_WORD2_SAMPLER_ID(tex
->sampler_id
) |
626 S_SQ_TEX_WORD2_SRC_SEL_X(tex
->src_sel_x
) |
627 S_SQ_TEX_WORD2_SRC_SEL_Y(tex
->src_sel_y
) |
628 S_SQ_TEX_WORD2_SRC_SEL_Z(tex
->src_sel_z
) |
629 S_SQ_TEX_WORD2_SRC_SEL_W(tex
->src_sel_w
);
630 bc
->bytecode
[id
++] = 0;
634 /* r600 only, r700/eg bits in r700_asm.c */
635 static int r600_bc_alu_build(struct r600_bc
*bc
, struct r600_bc_alu
*alu
, unsigned id
)
639 /* don't replace gpr by pv or ps for destination register */
640 bc
->bytecode
[id
++] = S_SQ_ALU_WORD0_SRC0_SEL(alu
->src
[0].sel
) |
641 S_SQ_ALU_WORD0_SRC0_REL(alu
->src
[0].rel
) |
642 S_SQ_ALU_WORD0_SRC0_CHAN(alu
->src
[0].chan
) |
643 S_SQ_ALU_WORD0_SRC0_NEG(alu
->src
[0].neg
) |
644 S_SQ_ALU_WORD0_SRC1_SEL(alu
->src
[1].sel
) |
645 S_SQ_ALU_WORD0_SRC1_REL(alu
->src
[1].rel
) |
646 S_SQ_ALU_WORD0_SRC1_CHAN(alu
->src
[1].chan
) |
647 S_SQ_ALU_WORD0_SRC1_NEG(alu
->src
[1].neg
) |
648 S_SQ_ALU_WORD0_LAST(alu
->last
);
651 bc
->bytecode
[id
++] = S_SQ_ALU_WORD1_DST_GPR(alu
->dst
.sel
) |
652 S_SQ_ALU_WORD1_DST_CHAN(alu
->dst
.chan
) |
653 S_SQ_ALU_WORD1_DST_REL(alu
->dst
.rel
) |
654 S_SQ_ALU_WORD1_CLAMP(alu
->dst
.clamp
) |
655 S_SQ_ALU_WORD1_OP3_SRC2_SEL(alu
->src
[2].sel
) |
656 S_SQ_ALU_WORD1_OP3_SRC2_REL(alu
->src
[2].rel
) |
657 S_SQ_ALU_WORD1_OP3_SRC2_CHAN(alu
->src
[2].chan
) |
658 S_SQ_ALU_WORD1_OP3_SRC2_NEG(alu
->src
[2].neg
) |
659 S_SQ_ALU_WORD1_OP3_ALU_INST(alu
->inst
) |
660 S_SQ_ALU_WORD1_BANK_SWIZZLE(alu
->bank_swizzle
);
662 bc
->bytecode
[id
++] = S_SQ_ALU_WORD1_DST_GPR(alu
->dst
.sel
) |
663 S_SQ_ALU_WORD1_DST_CHAN(alu
->dst
.chan
) |
664 S_SQ_ALU_WORD1_DST_REL(alu
->dst
.rel
) |
665 S_SQ_ALU_WORD1_CLAMP(alu
->dst
.clamp
) |
666 S_SQ_ALU_WORD1_OP2_SRC0_ABS(alu
->src
[0].abs
) |
667 S_SQ_ALU_WORD1_OP2_SRC1_ABS(alu
->src
[1].abs
) |
668 S_SQ_ALU_WORD1_OP2_WRITE_MASK(alu
->dst
.write
) |
669 S_SQ_ALU_WORD1_OP2_ALU_INST(alu
->inst
) |
670 S_SQ_ALU_WORD1_BANK_SWIZZLE(alu
->bank_swizzle
) |
671 S_SQ_ALU_WORD1_OP2_UPDATE_EXECUTE_MASK(alu
->predicate
) |
672 S_SQ_ALU_WORD1_OP2_UPDATE_PRED(alu
->predicate
);
675 if (alu
->nliteral
&& !alu
->literal_added
) {
676 R600_ERR("Bug in ALU processing for instruction 0x%08x, literal not added correctly\n", alu
->inst
);
678 for (i
= 0; i
< alu
->nliteral
; i
++) {
679 bc
->bytecode
[id
++] = alu
->value
[i
];
685 /* common for r600/r700 - eg in eg_asm.c */
686 static int r600_bc_cf_build(struct r600_bc
*bc
, struct r600_bc_cf
*cf
)
688 unsigned id
= cf
->id
;
691 case (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU
<< 3):
692 case (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_PUSH_BEFORE
<< 3):
693 bc
->bytecode
[id
++] = S_SQ_CF_ALU_WORD0_ADDR(cf
->addr
>> 1) |
694 S_SQ_CF_ALU_WORD0_KCACHE_MODE0(cf
->kcache0_mode
);
696 bc
->bytecode
[id
++] = S_SQ_CF_ALU_WORD1_CF_INST(cf
->inst
>> 3) |
697 S_SQ_CF_ALU_WORD1_BARRIER(1) |
698 S_SQ_CF_ALU_WORD1_COUNT((cf
->ndw
/ 2) - 1);
700 case V_SQ_CF_WORD1_SQ_CF_INST_TEX
:
701 case V_SQ_CF_WORD1_SQ_CF_INST_VTX
:
702 case V_SQ_CF_WORD1_SQ_CF_INST_VTX_TC
:
703 bc
->bytecode
[id
++] = S_SQ_CF_WORD0_ADDR(cf
->addr
>> 1);
704 bc
->bytecode
[id
++] = S_SQ_CF_WORD1_CF_INST(cf
->inst
) |
705 S_SQ_CF_WORD1_BARRIER(1) |
706 S_SQ_CF_WORD1_COUNT((cf
->ndw
/ 4) - 1);
708 case V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_EXPORT
:
709 case V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_EXPORT_DONE
:
710 bc
->bytecode
[id
++] = S_SQ_CF_ALLOC_EXPORT_WORD0_RW_GPR(cf
->output
.gpr
) |
711 S_SQ_CF_ALLOC_EXPORT_WORD0_ELEM_SIZE(cf
->output
.elem_size
) |
712 S_SQ_CF_ALLOC_EXPORT_WORD0_ARRAY_BASE(cf
->output
.array_base
) |
713 S_SQ_CF_ALLOC_EXPORT_WORD0_TYPE(cf
->output
.type
);
714 bc
->bytecode
[id
++] = S_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_X(cf
->output
.swizzle_x
) |
715 S_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_Y(cf
->output
.swizzle_y
) |
716 S_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_Z(cf
->output
.swizzle_z
) |
717 S_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_W(cf
->output
.swizzle_w
) |
718 S_SQ_CF_ALLOC_EXPORT_WORD1_BARRIER(cf
->output
.barrier
) |
719 S_SQ_CF_ALLOC_EXPORT_WORD1_CF_INST(cf
->output
.inst
) |
720 S_SQ_CF_ALLOC_EXPORT_WORD1_END_OF_PROGRAM(cf
->output
.end_of_program
);
722 case V_SQ_CF_WORD1_SQ_CF_INST_JUMP
:
723 case V_SQ_CF_WORD1_SQ_CF_INST_ELSE
:
724 case V_SQ_CF_WORD1_SQ_CF_INST_POP
:
725 case V_SQ_CF_WORD1_SQ_CF_INST_LOOP_START_NO_AL
:
726 case V_SQ_CF_WORD1_SQ_CF_INST_LOOP_END
:
727 case V_SQ_CF_WORD1_SQ_CF_INST_LOOP_CONTINUE
:
728 case V_SQ_CF_WORD1_SQ_CF_INST_LOOP_BREAK
:
729 bc
->bytecode
[id
++] = S_SQ_CF_WORD0_ADDR(cf
->cf_addr
>> 1);
730 bc
->bytecode
[id
++] = S_SQ_CF_WORD1_CF_INST(cf
->inst
) |
731 S_SQ_CF_WORD1_BARRIER(1) |
732 S_SQ_CF_WORD1_COND(cf
->cond
) |
733 S_SQ_CF_WORD1_POP_COUNT(cf
->pop_count
);
737 R600_ERR("unsupported CF instruction (0x%X)\n", cf
->inst
);
743 int r600_bc_build(struct r600_bc
*bc
)
745 struct r600_bc_cf
*cf
;
746 struct r600_bc_alu
*alu
;
747 struct r600_bc_vtx
*vtx
;
748 struct r600_bc_tex
*tex
;
752 if (bc
->callstack
[0].max
> 0)
753 bc
->nstack
= ((bc
->callstack
[0].max
+ 3) >> 2) + 2;
755 /* first path compute addr of each CF block */
756 /* addr start after all the CF instructions */
757 addr
= bc
->cf_last
->id
+ 2;
758 LIST_FOR_EACH_ENTRY(cf
, &bc
->cf
, list
) {
760 case (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU
<< 3):
761 case (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_PUSH_BEFORE
<< 3):
763 case V_SQ_CF_WORD1_SQ_CF_INST_TEX
:
764 case V_SQ_CF_WORD1_SQ_CF_INST_VTX
:
765 case V_SQ_CF_WORD1_SQ_CF_INST_VTX_TC
:
766 /* fetch node need to be 16 bytes aligned*/
768 addr
&= 0xFFFFFFFCUL
;
770 case V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_EXPORT
:
771 case V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_EXPORT_DONE
:
772 case EG_V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_EXPORT
:
773 case EG_V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_EXPORT_DONE
:
775 case V_SQ_CF_WORD1_SQ_CF_INST_JUMP
:
776 case V_SQ_CF_WORD1_SQ_CF_INST_ELSE
:
777 case V_SQ_CF_WORD1_SQ_CF_INST_POP
:
778 case V_SQ_CF_WORD1_SQ_CF_INST_LOOP_START_NO_AL
:
779 case V_SQ_CF_WORD1_SQ_CF_INST_LOOP_END
:
780 case V_SQ_CF_WORD1_SQ_CF_INST_LOOP_CONTINUE
:
781 case V_SQ_CF_WORD1_SQ_CF_INST_LOOP_BREAK
:
784 R600_ERR("unsupported CF instruction (0x%X)\n", cf
->inst
);
789 bc
->ndw
= cf
->addr
+ cf
->ndw
;
792 bc
->bytecode
= calloc(1, bc
->ndw
* 4);
793 if (bc
->bytecode
== NULL
)
795 LIST_FOR_EACH_ENTRY(cf
, &bc
->cf
, list
) {
797 if (bc
->chiprev
== 2)
798 r
= eg_bc_cf_build(bc
, cf
);
800 r
= r600_bc_cf_build(bc
, cf
);
804 case (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU
<< 3):
805 case (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_PUSH_BEFORE
<< 3):
806 LIST_FOR_EACH_ENTRY(alu
, &cf
->alu
, list
) {
807 switch(bc
->chiprev
) {
809 r
= r600_bc_alu_build(bc
, alu
, addr
);
812 case 2: /* eg alu is same encoding as r700 */
813 r
= r700_bc_alu_build(bc
, alu
, addr
);
816 R600_ERR("unknown family %d\n", bc
->family
);
823 addr
+= alu
->nliteral
;
827 case V_SQ_CF_WORD1_SQ_CF_INST_VTX
:
828 case V_SQ_CF_WORD1_SQ_CF_INST_VTX_TC
:
829 LIST_FOR_EACH_ENTRY(vtx
, &cf
->vtx
, list
) {
830 r
= r600_bc_vtx_build(bc
, vtx
, addr
);
836 case V_SQ_CF_WORD1_SQ_CF_INST_TEX
:
837 LIST_FOR_EACH_ENTRY(tex
, &cf
->tex
, list
) {
838 r
= r600_bc_tex_build(bc
, tex
, addr
);
844 case V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_EXPORT
:
845 case V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_EXPORT_DONE
:
846 case EG_V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_EXPORT
:
847 case EG_V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_EXPORT_DONE
:
848 case V_SQ_CF_WORD1_SQ_CF_INST_LOOP_START_NO_AL
:
849 case V_SQ_CF_WORD1_SQ_CF_INST_LOOP_END
:
850 case V_SQ_CF_WORD1_SQ_CF_INST_LOOP_CONTINUE
:
851 case V_SQ_CF_WORD1_SQ_CF_INST_LOOP_BREAK
:
852 case V_SQ_CF_WORD1_SQ_CF_INST_JUMP
:
853 case V_SQ_CF_WORD1_SQ_CF_INST_ELSE
:
854 case V_SQ_CF_WORD1_SQ_CF_INST_POP
:
857 R600_ERR("unsupported CF instruction (0x%X)\n", cf
->inst
);