2 * Copyright 2008 Ben Skeggs
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 shall be included in
12 * all copies or substantial portions of the Software.
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 BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
18 * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF
19 * OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
23 #include "pipe/p_context.h"
24 #include "pipe/p_defines.h"
25 #include "pipe/p_state.h"
26 #include "pipe/p_inlines.h"
28 #include "pipe/p_shader_tokens.h"
29 #include "tgsi/tgsi_parse.h"
30 #include "tgsi/tgsi_util.h"
32 #include "nv50_context.h"
34 #define NV50_SU_MAX_TEMP 64
35 //#define NV50_PROGRAM_DUMP
37 /* ARL - gallium craps itself on progs/vp/arl.txt
39 * MSB - Like MAD, but MUL+SUB
40 * - Fuck it off, introduce a way to negate args for ops that
43 * Look into inlining IMMD for ops other than MOV (make it general?)
44 * - Maybe even relax restrictions a bit, can't do P_RESULT + P_IMMD,
45 * but can emit to P_TEMP first - then MOV later. NVIDIA does this
47 * In ops such as ADD it's possible to construct a bad opcode in the !is_long()
48 * case, if the emit_src() causes the inst to suddenly become long.
50 * Verify half-insns work where expected - and force disable them where they
51 * don't work - MUL has it forcibly disabled atm as it fixes POW..
53 * FUCK! watch dst==src vectors, can overwrite components that are needed.
54 * ie. SUB R0, R0.yzxw, R0
56 * Things to check with renouveau:
57 * FP attr/result assignment - how?
59 * - 0x16bc maps vp output onto fp hpos
60 * - 0x16c0 maps vp output onto fp col0
64 * 0x16bc->0x16e8 --> some binding between vp/fp regs
65 * 0x16b8 --> VP output count
67 * 0x1298 --> "MOV rcol.x, fcol.y" "MOV depr, fcol.y" = 0x00000005
68 * "MOV rcol.x, fcol.y" = 0x00000004
69 * 0x19a8 --> as above but 0x00000100 and 0x00000000
70 * - 0x00100000 used when KIL used
71 * 0x196c --> as above but 0x00000011 and 0x00000000
73 * 0x1988 --> 0xXXNNNNNN
74 * - XX == FP high something
91 struct nv50_program
*p
;
94 struct nv50_reg
*r_temp
[NV50_SU_MAX_TEMP
];
97 struct nv50_reg
*temp
;
99 struct nv50_reg
*attr
;
101 struct nv50_reg
*result
;
103 struct nv50_reg
*param
;
105 struct nv50_reg
*immd
;
109 struct nv50_reg
*temp_temp
[16];
110 unsigned temp_temp_nr
;
114 alloc_reg(struct nv50_pc
*pc
, struct nv50_reg
*reg
)
118 if (reg
->type
== P_RESULT
) {
119 if (pc
->p
->cfg
.high_result
< (reg
->hw
+ 1))
120 pc
->p
->cfg
.high_result
= reg
->hw
+ 1;
123 if (reg
->type
!= P_TEMP
)
127 /*XXX: do this here too to catch FP temp-as-attr usage..
128 * not clean, but works */
129 if (pc
->p
->cfg
.high_temp
< (reg
->hw
+ 1))
130 pc
->p
->cfg
.high_temp
= reg
->hw
+ 1;
134 for (i
= 0; i
< NV50_SU_MAX_TEMP
; i
++) {
135 if (!(pc
->r_temp
[i
])) {
138 if (pc
->p
->cfg
.high_temp
< (i
+ 1))
139 pc
->p
->cfg
.high_temp
= i
+ 1;
147 static struct nv50_reg
*
148 alloc_temp(struct nv50_pc
*pc
, struct nv50_reg
*dst
)
153 if (dst
&& dst
->type
== P_TEMP
&& dst
->hw
== -1)
156 for (i
= 0; i
< NV50_SU_MAX_TEMP
; i
++) {
157 if (!pc
->r_temp
[i
]) {
158 r
= CALLOC_STRUCT(nv50_reg
);
172 free_temp(struct nv50_pc
*pc
, struct nv50_reg
*r
)
174 if (r
->index
== -1) {
177 FREE(pc
->r_temp
[hw
]);
178 pc
->r_temp
[hw
] = NULL
;
183 alloc_temp4(struct nv50_pc
*pc
, struct nv50_reg
*dst
[4], int idx
)
187 if ((idx
+ 4) >= NV50_SU_MAX_TEMP
)
190 if (pc
->r_temp
[idx
] || pc
->r_temp
[idx
+ 1] ||
191 pc
->r_temp
[idx
+ 2] || pc
->r_temp
[idx
+ 3])
192 return alloc_temp4(pc
, dst
, idx
+ 1);
194 for (i
= 0; i
< 4; i
++) {
195 dst
[i
] = CALLOC_STRUCT(nv50_reg
);
196 dst
[i
]->type
= P_TEMP
;
198 dst
[i
]->hw
= idx
+ i
;
199 pc
->r_temp
[idx
+ i
] = dst
[i
];
206 free_temp4(struct nv50_pc
*pc
, struct nv50_reg
*reg
[4])
210 for (i
= 0; i
< 4; i
++)
211 free_temp(pc
, reg
[i
]);
214 static struct nv50_reg
*
215 temp_temp(struct nv50_pc
*pc
)
217 if (pc
->temp_temp_nr
>= 16)
220 pc
->temp_temp
[pc
->temp_temp_nr
] = alloc_temp(pc
, NULL
);
221 return pc
->temp_temp
[pc
->temp_temp_nr
++];
225 kill_temp_temp(struct nv50_pc
*pc
)
229 for (i
= 0; i
< pc
->temp_temp_nr
; i
++)
230 free_temp(pc
, pc
->temp_temp
[i
]);
231 pc
->temp_temp_nr
= 0;
235 ctor_immd(struct nv50_pc
*pc
, float x
, float y
, float z
, float w
)
237 pc
->immd_buf
= REALLOC(pc
->immd_buf
, (pc
->immd_nr
* r
* sizeof(float)),
238 (pc
->immd_nr
+ 1) * 4 * sizeof(float));
239 pc
->immd_buf
[(pc
->immd_nr
* 4) + 0] = x
;
240 pc
->immd_buf
[(pc
->immd_nr
* 4) + 1] = y
;
241 pc
->immd_buf
[(pc
->immd_nr
* 4) + 2] = z
;
242 pc
->immd_buf
[(pc
->immd_nr
* 4) + 3] = w
;
244 return pc
->immd_nr
++;
247 static struct nv50_reg
*
248 alloc_immd(struct nv50_pc
*pc
, float f
)
250 struct nv50_reg
*r
= CALLOC_STRUCT(nv50_reg
);
253 for (hw
= 0; hw
< pc
->immd_nr
* 4; hw
++)
254 if (pc
->immd_buf
[hw
] == f
)
257 if (hw
== pc
->immd_nr
* 4)
258 hw
= ctor_immd(pc
, f
, -f
, 0.5 * f
, 0) * 4;
266 static struct nv50_program_exec
*
267 exec(struct nv50_pc
*pc
)
269 struct nv50_program_exec
*e
= CALLOC_STRUCT(nv50_program_exec
);
276 emit(struct nv50_pc
*pc
, struct nv50_program_exec
*e
)
278 struct nv50_program
*p
= pc
->p
;
281 p
->exec_tail
->next
= e
;
285 p
->exec_size
+= (e
->inst
[0] & 1) ? 2 : 1;
288 static INLINE
void set_long(struct nv50_pc
*, struct nv50_program_exec
*);
291 is_long(struct nv50_program_exec
*e
)
299 is_immd(struct nv50_program_exec
*e
)
301 if (is_long(e
) && (e
->inst
[1] & 3) == 3)
307 set_pred(struct nv50_pc
*pc
, unsigned pred
, unsigned idx
,
308 struct nv50_program_exec
*e
)
311 e
->inst
[1] &= ~((0x1f << 7) | (0x3 << 12));
312 e
->inst
[1] |= (pred
<< 7) | (idx
<< 12);
316 set_pred_wr(struct nv50_pc
*pc
, unsigned on
, unsigned idx
,
317 struct nv50_program_exec
*e
)
320 e
->inst
[1] &= ~((0x3 << 4) | (1 << 6));
321 e
->inst
[1] |= (idx
<< 4) | (on
<< 6);
325 set_long(struct nv50_pc
*pc
, struct nv50_program_exec
*e
)
331 set_pred(pc
, 0xf, 0, e
);
332 set_pred_wr(pc
, 0, 0, e
);
336 set_dst(struct nv50_pc
*pc
, struct nv50_reg
*dst
, struct nv50_program_exec
*e
)
338 if (dst
->type
== P_RESULT
) {
340 e
->inst
[1] |= 0x00000008;
344 e
->inst
[0] |= (dst
->hw
<< 2);
348 set_immd(struct nv50_pc
*pc
, struct nv50_reg
*imm
, struct nv50_program_exec
*e
)
350 unsigned val
= fui(pc
->immd_buf
[imm
->hw
]); /* XXX */
353 /*XXX: can't be predicated - bits overlap.. catch cases where both
354 * are required and avoid them. */
355 set_pred(pc
, 0, 0, e
);
356 set_pred_wr(pc
, 0, 0, e
);
358 e
->inst
[1] |= 0x00000002 | 0x00000001;
359 e
->inst
[0] |= (val
& 0x3f) << 16;
360 e
->inst
[1] |= (val
>> 6) << 2;
364 emit_interp(struct nv50_pc
*pc
, struct nv50_reg
*dst
,
365 struct nv50_reg
*src
, struct nv50_reg
*iv
)
367 struct nv50_program_exec
*e
= exec(pc
);
369 e
->inst
[0] |= 0x80000000;
372 e
->inst
[0] |= (src
->hw
<< 16);
374 e
->inst
[0] |= (1 << 25);
376 e
->inst
[0] |= (iv
->hw
<< 9);
383 set_data(struct nv50_pc
*pc
, struct nv50_reg
*src
, unsigned m
, unsigned s
,
384 struct nv50_program_exec
*e
)
388 e
->inst
[1] |= (1 << 22);
390 if (src
->type
== P_IMMD
) {
391 e
->inst
[1] |= (NV50_CB_PMISC
<< 22);
393 if (pc
->p
->type
== PIPE_SHADER_VERTEX
)
394 e
->inst
[1] |= (NV50_CB_PVP
<< 22);
396 e
->inst
[1] |= (NV50_CB_PFP
<< 22);
400 e
->param
.index
= src
->hw
;
402 e
->param
.mask
= m
<< (s
% 32);
406 emit_mov(struct nv50_pc
*pc
, struct nv50_reg
*dst
, struct nv50_reg
*src
)
408 struct nv50_program_exec
*e
= exec(pc
);
410 e
->inst
[0] |= 0x10000000;
414 if (0 && dst
->type
!= P_RESULT
&& src
->type
== P_IMMD
) {
415 set_immd(pc
, src
, e
);
416 /*XXX: 32-bit, but steals part of "half" reg space - need to
417 * catch and handle this case if/when we do half-regs
419 e
->inst
[0] |= 0x00008000;
421 if (src
->type
== P_IMMD
|| src
->type
== P_CONST
) {
423 set_data(pc
, src
, 0x7f, 9, e
);
424 e
->inst
[1] |= 0x20000000; /* src0 const? */
426 if (src
->type
== P_ATTR
) {
428 e
->inst
[1] |= 0x00200000;
432 e
->inst
[0] |= (src
->hw
<< 9);
435 /* We really should support "half" instructions here at some point,
436 * but I don't feel confident enough about them yet.
439 if (is_long(e
) && !is_immd(e
)) {
440 e
->inst
[1] |= 0x04000000; /* 32-bit */
441 e
->inst
[1] |= 0x0003c000; /* "subsubop" 0xf == mov */
448 emit_mov_immdval(struct nv50_pc
*pc
, struct nv50_reg
*dst
, float f
)
450 struct nv50_reg
*imm
= alloc_immd(pc
, f
);
451 emit_mov(pc
, dst
, imm
);
456 check_swap_src_0_1(struct nv50_pc
*pc
,
457 struct nv50_reg
**s0
, struct nv50_reg
**s1
)
459 struct nv50_reg
*src0
= *s0
, *src1
= *s1
;
461 if (src0
->type
== P_CONST
) {
462 if (src1
->type
!= P_CONST
) {
468 if (src1
->type
== P_ATTR
) {
469 if (src0
->type
!= P_ATTR
) {
480 set_src_0(struct nv50_pc
*pc
, struct nv50_reg
*src
, struct nv50_program_exec
*e
)
482 if (src
->type
== P_ATTR
) {
484 e
->inst
[1] |= 0x00200000;
486 if (src
->type
== P_CONST
|| src
->type
== P_IMMD
) {
487 struct nv50_reg
*temp
= temp_temp(pc
);
489 emit_mov(pc
, temp
, src
);
494 e
->inst
[0] |= (src
->hw
<< 9);
498 set_src_1(struct nv50_pc
*pc
, struct nv50_reg
*src
, struct nv50_program_exec
*e
)
500 if (src
->type
== P_ATTR
) {
501 struct nv50_reg
*temp
= temp_temp(pc
);
503 emit_mov(pc
, temp
, src
);
506 if (src
->type
== P_CONST
|| src
->type
== P_IMMD
) {
507 assert(!(e
->inst
[0] & 0x00800000));
508 if (e
->inst
[0] & 0x01000000) {
509 struct nv50_reg
*temp
= temp_temp(pc
);
511 emit_mov(pc
, temp
, src
);
514 set_data(pc
, src
, 0x7f, 16, e
);
515 e
->inst
[0] |= 0x00800000;
520 e
->inst
[0] |= (src
->hw
<< 16);
524 set_src_2(struct nv50_pc
*pc
, struct nv50_reg
*src
, struct nv50_program_exec
*e
)
528 if (src
->type
== P_ATTR
) {
529 struct nv50_reg
*temp
= temp_temp(pc
);
531 emit_mov(pc
, temp
, src
);
534 if (src
->type
== P_CONST
|| src
->type
== P_IMMD
) {
535 assert(!(e
->inst
[0] & 0x01000000));
536 if (e
->inst
[0] & 0x00800000) {
537 struct nv50_reg
*temp
= temp_temp(pc
);
539 emit_mov(pc
, temp
, src
);
542 set_data(pc
, src
, 0x7f, 32+14, e
);
543 e
->inst
[0] |= 0x01000000;
548 e
->inst
[1] |= (src
->hw
<< 14);
552 emit_mul(struct nv50_pc
*pc
, struct nv50_reg
*dst
, struct nv50_reg
*src0
,
553 struct nv50_reg
*src1
)
555 struct nv50_program_exec
*e
= exec(pc
);
557 e
->inst
[0] |= 0xc0000000;
560 check_swap_src_0_1(pc
, &src0
, &src1
);
562 set_src_0(pc
, src0
, e
);
563 set_src_1(pc
, src1
, e
);
569 emit_add(struct nv50_pc
*pc
, struct nv50_reg
*dst
,
570 struct nv50_reg
*src0
, struct nv50_reg
*src1
)
572 struct nv50_program_exec
*e
= exec(pc
);
574 e
->inst
[0] |= 0xb0000000;
576 check_swap_src_0_1(pc
, &src0
, &src1
);
578 set_src_0(pc
, src0
, e
);
580 set_src_2(pc
, src1
, e
);
582 set_src_1(pc
, src1
, e
);
588 emit_minmax(struct nv50_pc
*pc
, unsigned sub
, struct nv50_reg
*dst
,
589 struct nv50_reg
*src0
, struct nv50_reg
*src1
)
591 struct nv50_program_exec
*e
= exec(pc
);
594 e
->inst
[0] |= 0xb0000000;
595 e
->inst
[1] |= (sub
<< 29);
597 check_swap_src_0_1(pc
, &src0
, &src1
);
599 set_src_0(pc
, src0
, e
);
600 set_src_1(pc
, src1
, e
);
606 emit_sub(struct nv50_pc
*pc
, struct nv50_reg
*dst
, struct nv50_reg
*src0
,
607 struct nv50_reg
*src1
)
609 struct nv50_program_exec
*e
= exec(pc
);
611 e
->inst
[0] |= 0xb0000000;
614 if (check_swap_src_0_1(pc
, &src0
, &src1
))
615 e
->inst
[1] |= 0x04000000;
617 e
->inst
[1] |= 0x08000000;
620 set_src_0(pc
, src0
, e
);
621 set_src_2(pc
, src1
, e
);
627 emit_mad(struct nv50_pc
*pc
, struct nv50_reg
*dst
, struct nv50_reg
*src0
,
628 struct nv50_reg
*src1
, struct nv50_reg
*src2
)
630 struct nv50_program_exec
*e
= exec(pc
);
632 e
->inst
[0] |= 0xe0000000;
634 check_swap_src_0_1(pc
, &src0
, &src1
);
636 set_src_0(pc
, src0
, e
);
637 set_src_1(pc
, src1
, e
);
638 set_src_2(pc
, src2
, e
);
644 emit_msb(struct nv50_pc
*pc
, struct nv50_reg
*dst
, struct nv50_reg
*src0
,
645 struct nv50_reg
*src1
, struct nv50_reg
*src2
)
647 struct nv50_program_exec
*e
= exec(pc
);
649 e
->inst
[0] |= 0xe0000000;
651 e
->inst
[1] |= 0x08000000; /* src0 * src1 - src2 */
653 check_swap_src_0_1(pc
, &src0
, &src1
);
655 set_src_0(pc
, src0
, e
);
656 set_src_1(pc
, src1
, e
);
657 set_src_2(pc
, src2
, e
);
663 emit_flop(struct nv50_pc
*pc
, unsigned sub
,
664 struct nv50_reg
*dst
, struct nv50_reg
*src
)
666 struct nv50_program_exec
*e
= exec(pc
);
668 e
->inst
[0] |= 0x90000000;
671 e
->inst
[1] |= (sub
<< 29);
675 set_src_0(pc
, src
, e
);
681 emit_preex2(struct nv50_pc
*pc
, struct nv50_reg
*dst
, struct nv50_reg
*src
)
683 struct nv50_program_exec
*e
= exec(pc
);
685 e
->inst
[0] |= 0xb0000000;
688 set_src_0(pc
, src
, e
);
690 e
->inst
[1] |= (6 << 29) | 0x00004000;
696 emit_precossin(struct nv50_pc
*pc
, struct nv50_reg
*dst
, struct nv50_reg
*src
)
698 struct nv50_program_exec
*e
= exec(pc
);
700 e
->inst
[0] |= 0xb0000000;
703 set_src_0(pc
, src
, e
);
705 e
->inst
[1] |= (6 << 29);
711 emit_set(struct nv50_pc
*pc
, unsigned c_op
, struct nv50_reg
*dst
,
712 struct nv50_reg
*src0
, struct nv50_reg
*src1
)
714 struct nv50_program_exec
*e
= exec(pc
);
715 unsigned inv_cop
[8] = { 0, 4, 2, 6, 1, 5, 3, 7 };
716 struct nv50_reg
*rdst
;
719 if (check_swap_src_0_1(pc
, &src0
, &src1
))
720 c_op
= inv_cop
[c_op
];
723 if (dst
->type
!= P_TEMP
)
724 dst
= alloc_temp(pc
, NULL
);
728 e
->inst
[0] |= 0xb0000000;
729 e
->inst
[1] |= (3 << 29);
730 e
->inst
[1] |= (c_op
<< 14);
731 /*XXX: breaks things, .u32 by default?
732 * decuda will disasm as .u16 and use .lo/.hi regs, but this
733 * doesn't seem to match what the hw actually does.
734 inst[1] |= 0x04000000; << breaks things.. .u32 by default?
737 set_src_0(pc
, src0
, e
);
738 set_src_1(pc
, src1
, e
);
743 e
->inst
[0] = 0xa0000001;
744 e
->inst
[1] = 0x64014780;
745 set_dst(pc
, rdst
, e
);
746 set_src_0(pc
, dst
, e
);
754 emit_flr(struct nv50_pc
*pc
, struct nv50_reg
*dst
, struct nv50_reg
*src
)
756 struct nv50_program_exec
*e
= exec(pc
);
758 e
->inst
[0] = 0xa0000000; /* cvt */
760 e
->inst
[1] |= (6 << 29); /* cvt */
761 e
->inst
[1] |= 0x08000000; /* integer mode */
762 e
->inst
[1] |= 0x04000000; /* 32 bit */
763 e
->inst
[1] |= ((0x1 << 3)) << 14; /* .rn */
764 e
->inst
[1] |= (1 << 14); /* src .f32 */
766 set_src_0(pc
, src
, e
);
772 emit_pow(struct nv50_pc
*pc
, struct nv50_reg
*dst
,
773 struct nv50_reg
*v
, struct nv50_reg
*e
)
775 struct nv50_reg
*temp
= alloc_temp(pc
, NULL
);
777 emit_flop(pc
, 3, temp
, v
);
778 emit_mul(pc
, temp
, temp
, e
);
779 emit_preex2(pc
, temp
, temp
);
780 emit_flop(pc
, 6, dst
, temp
);
786 emit_abs(struct nv50_pc
*pc
, struct nv50_reg
*dst
, struct nv50_reg
*src
)
788 struct nv50_program_exec
*e
= exec(pc
);
790 e
->inst
[0] = 0xa0000000; /* cvt */
792 e
->inst
[1] |= (6 << 29); /* cvt */
793 e
->inst
[1] |= 0x04000000; /* 32 bit */
794 e
->inst
[1] |= (1 << 14); /* src .f32 */
795 e
->inst
[1] |= ((1 << 6) << 14); /* .abs */
797 set_src_0(pc
, src
, e
);
803 emit_lit(struct nv50_pc
*pc
, struct nv50_reg
**dst
, unsigned mask
,
804 struct nv50_reg
**src
)
806 struct nv50_reg
*one
= alloc_immd(pc
, 1.0);
807 struct nv50_reg
*zero
= alloc_immd(pc
, 0.0);
808 struct nv50_reg
*neg128
= alloc_immd(pc
, -127.999999);
809 struct nv50_reg
*pos128
= alloc_immd(pc
, 127.999999);
810 struct nv50_reg
*tmp
[4];
813 emit_mov(pc
, dst
[0], one
);
816 emit_mov(pc
, dst
[3], one
);
818 if (mask
& (3 << 1)) {
822 tmp
[0] = temp_temp(pc
);
823 emit_minmax(pc
, 4, tmp
[0], src
[0], zero
);
826 if (mask
& (1 << 2)) {
827 set_pred_wr(pc
, 1, 0, pc
->p
->exec_tail
);
829 tmp
[1] = temp_temp(pc
);
830 emit_minmax(pc
, 4, tmp
[1], src
[1], zero
);
832 tmp
[3] = temp_temp(pc
);
833 emit_minmax(pc
, 4, tmp
[3], src
[3], neg128
);
834 emit_minmax(pc
, 5, tmp
[3], tmp
[3], pos128
);
836 emit_pow(pc
, dst
[2], tmp
[1], tmp
[3]);
837 emit_mov(pc
, dst
[2], zero
);
838 set_pred(pc
, 3, 0, pc
->p
->exec_tail
);
848 emit_neg(struct nv50_pc
*pc
, struct nv50_reg
*dst
, struct nv50_reg
*src
)
850 struct nv50_program_exec
*e
= exec(pc
);
853 e
->inst
[0] |= 0xa0000000; /* delta */
854 e
->inst
[1] |= (7 << 29); /* delta */
855 e
->inst
[1] |= 0x04000000; /* negate arg0? probably not */
856 e
->inst
[1] |= (1 << 14); /* src .f32 */
858 set_src_0(pc
, src
, e
);
864 emit_kil(struct nv50_pc
*pc
, struct nv50_reg
*src
)
866 struct nv50_program_exec
*e
;
867 const int r_pred
= 1;
869 /* Sets predicate reg ? */
871 e
->inst
[0] = 0xa00001fd;
872 e
->inst
[1] = 0xc4014788;
873 set_src_0(pc
, src
, e
);
874 set_pred_wr(pc
, 1, r_pred
, e
);
877 /* This is probably KILP */
879 e
->inst
[0] = 0x000001fe;
881 set_pred(pc
, 1 /* LT? */, r_pred
, e
);
885 static struct nv50_reg
*
886 tgsi_dst(struct nv50_pc
*pc
, int c
, const struct tgsi_full_dst_register
*dst
)
888 switch (dst
->DstRegister
.File
) {
889 case TGSI_FILE_TEMPORARY
:
890 return &pc
->temp
[dst
->DstRegister
.Index
* 4 + c
];
891 case TGSI_FILE_OUTPUT
:
892 return &pc
->result
[dst
->DstRegister
.Index
* 4 + c
];
902 static struct nv50_reg
*
903 tgsi_src(struct nv50_pc
*pc
, int chan
, const struct tgsi_full_src_register
*src
)
905 struct nv50_reg
*r
= NULL
;
906 struct nv50_reg
*temp
;
909 sgn
= tgsi_util_get_full_src_register_sign_mode(src
, chan
);
911 c
= tgsi_util_get_full_src_register_extswizzle(src
, chan
);
913 case TGSI_EXTSWIZZLE_X
:
914 case TGSI_EXTSWIZZLE_Y
:
915 case TGSI_EXTSWIZZLE_Z
:
916 case TGSI_EXTSWIZZLE_W
:
917 switch (src
->SrcRegister
.File
) {
918 case TGSI_FILE_INPUT
:
919 r
= &pc
->attr
[src
->SrcRegister
.Index
* 4 + c
];
921 case TGSI_FILE_TEMPORARY
:
922 r
= &pc
->temp
[src
->SrcRegister
.Index
* 4 + c
];
924 case TGSI_FILE_CONSTANT
:
925 r
= &pc
->param
[src
->SrcRegister
.Index
* 4 + c
];
927 case TGSI_FILE_IMMEDIATE
:
928 r
= &pc
->immd
[src
->SrcRegister
.Index
* 4 + c
];
930 case TGSI_FILE_SAMPLER
:
937 case TGSI_EXTSWIZZLE_ZERO
:
938 r
= alloc_immd(pc
, 0.0);
940 case TGSI_EXTSWIZZLE_ONE
:
941 if (sgn
== TGSI_UTIL_SIGN_TOGGLE
|| sgn
== TGSI_UTIL_SIGN_SET
)
942 return alloc_immd(pc
, -1.0);
943 return alloc_immd(pc
, 1.0);
950 case TGSI_UTIL_SIGN_KEEP
:
952 case TGSI_UTIL_SIGN_CLEAR
:
953 temp
= temp_temp(pc
);
954 emit_abs(pc
, temp
, r
);
957 case TGSI_UTIL_SIGN_TOGGLE
:
958 temp
= temp_temp(pc
);
959 emit_neg(pc
, temp
, r
);
962 case TGSI_UTIL_SIGN_SET
:
963 temp
= temp_temp(pc
);
964 emit_abs(pc
, temp
, r
);
965 emit_neg(pc
, temp
, temp
);
977 nv50_program_tx_insn(struct nv50_pc
*pc
, const union tgsi_full_token
*tok
)
979 const struct tgsi_full_instruction
*inst
= &tok
->FullInstruction
;
980 struct nv50_reg
*rdst
[4], *dst
[4], *src
[3][4], *temp
;
981 unsigned mask
, sat
, unit
;
984 mask
= inst
->FullDstRegisters
[0].DstRegister
.WriteMask
;
985 sat
= inst
->Instruction
.Saturate
== TGSI_SAT_ZERO_ONE
;
987 for (c
= 0; c
< 4; c
++) {
989 dst
[c
] = tgsi_dst(pc
, c
, &inst
->FullDstRegisters
[0]);
998 for (i
= 0; i
< inst
->Instruction
.NumSrcRegs
; i
++) {
999 const struct tgsi_full_src_register
*fs
= &inst
->FullSrcRegisters
[i
];
1001 if (fs
->SrcRegister
.File
== TGSI_FILE_SAMPLER
)
1002 unit
= fs
->SrcRegister
.Index
;
1004 for (c
= 0; c
< 4; c
++)
1005 src
[i
][c
] = tgsi_src(pc
, c
, fs
);
1009 for (c
= 0; c
< 4; c
++) {
1011 dst
[c
] = temp_temp(pc
);
1015 switch (inst
->Instruction
.Opcode
) {
1016 case TGSI_OPCODE_ABS
:
1017 for (c
= 0; c
< 4; c
++) {
1018 if (!(mask
& (1 << c
)))
1020 emit_abs(pc
, dst
[c
], src
[0][c
]);
1023 case TGSI_OPCODE_ADD
:
1024 for (c
= 0; c
< 4; c
++) {
1025 if (!(mask
& (1 << c
)))
1027 emit_add(pc
, dst
[c
], src
[0][c
], src
[1][c
]);
1030 case TGSI_OPCODE_COS
:
1031 temp
= temp_temp(pc
);
1032 emit_precossin(pc
, temp
, src
[0][0]);
1033 emit_flop(pc
, 5, temp
, temp
);
1034 for (c
= 0; c
< 4; c
++) {
1035 if (!(mask
& (1 << c
)))
1037 emit_mov(pc
, dst
[c
], temp
);
1040 case TGSI_OPCODE_DP3
:
1041 temp
= temp_temp(pc
);
1042 emit_mul(pc
, temp
, src
[0][0], src
[1][0]);
1043 emit_mad(pc
, temp
, src
[0][1], src
[1][1], temp
);
1044 emit_mad(pc
, temp
, src
[0][2], src
[1][2], temp
);
1045 for (c
= 0; c
< 4; c
++) {
1046 if (!(mask
& (1 << c
)))
1048 emit_mov(pc
, dst
[c
], temp
);
1051 case TGSI_OPCODE_DP4
:
1052 temp
= temp_temp(pc
);
1053 emit_mul(pc
, temp
, src
[0][0], src
[1][0]);
1054 emit_mad(pc
, temp
, src
[0][1], src
[1][1], temp
);
1055 emit_mad(pc
, temp
, src
[0][2], src
[1][2], temp
);
1056 emit_mad(pc
, temp
, src
[0][3], src
[1][3], temp
);
1057 for (c
= 0; c
< 4; c
++) {
1058 if (!(mask
& (1 << c
)))
1060 emit_mov(pc
, dst
[c
], temp
);
1063 case TGSI_OPCODE_DPH
:
1064 temp
= temp_temp(pc
);
1065 emit_mul(pc
, temp
, src
[0][0], src
[1][0]);
1066 emit_mad(pc
, temp
, src
[0][1], src
[1][1], temp
);
1067 emit_mad(pc
, temp
, src
[0][2], src
[1][2], temp
);
1068 emit_add(pc
, temp
, src
[1][3], temp
);
1069 for (c
= 0; c
< 4; c
++) {
1070 if (!(mask
& (1 << c
)))
1072 emit_mov(pc
, dst
[c
], temp
);
1075 case TGSI_OPCODE_DST
:
1077 struct nv50_reg
*one
= alloc_immd(pc
, 1.0);
1078 if (mask
& (1 << 0))
1079 emit_mov(pc
, dst
[0], one
);
1080 if (mask
& (1 << 1))
1081 emit_mul(pc
, dst
[1], src
[0][1], src
[1][1]);
1082 if (mask
& (1 << 2))
1083 emit_mov(pc
, dst
[2], src
[0][2]);
1084 if (mask
& (1 << 3))
1085 emit_mov(pc
, dst
[3], src
[1][3]);
1089 case TGSI_OPCODE_EX2
:
1090 temp
= temp_temp(pc
);
1091 emit_preex2(pc
, temp
, src
[0][0]);
1092 emit_flop(pc
, 6, temp
, temp
);
1093 for (c
= 0; c
< 4; c
++) {
1094 if (!(mask
& (1 << c
)))
1096 emit_mov(pc
, dst
[c
], temp
);
1099 case TGSI_OPCODE_FLR
:
1100 for (c
= 0; c
< 4; c
++) {
1101 if (!(mask
& (1 << c
)))
1103 emit_flr(pc
, dst
[c
], src
[0][c
]);
1106 case TGSI_OPCODE_FRC
:
1107 temp
= temp_temp(pc
);
1108 for (c
= 0; c
< 4; c
++) {
1109 if (!(mask
& (1 << c
)))
1111 emit_flr(pc
, temp
, src
[0][c
]);
1112 emit_sub(pc
, dst
[c
], src
[0][c
], temp
);
1115 case TGSI_OPCODE_KIL
:
1116 emit_kil(pc
, src
[0][0]);
1117 emit_kil(pc
, src
[0][1]);
1118 emit_kil(pc
, src
[0][2]);
1119 emit_kil(pc
, src
[0][3]);
1121 case TGSI_OPCODE_LIT
:
1122 emit_lit(pc
, &dst
[0], mask
, &src
[0][0]);
1124 case TGSI_OPCODE_LG2
:
1125 temp
= temp_temp(pc
);
1126 emit_flop(pc
, 3, temp
, src
[0][0]);
1127 for (c
= 0; c
< 4; c
++) {
1128 if (!(mask
& (1 << c
)))
1130 emit_mov(pc
, dst
[c
], temp
);
1133 case TGSI_OPCODE_LRP
:
1134 temp
= temp_temp(pc
);
1135 for (c
= 0; c
< 4; c
++) {
1136 if (!(mask
& (1 << c
)))
1138 emit_sub(pc
, temp
, src
[1][c
], src
[2][c
]);
1139 emit_mad(pc
, dst
[c
], temp
, src
[0][c
], src
[2][c
]);
1142 case TGSI_OPCODE_MAD
:
1143 for (c
= 0; c
< 4; c
++) {
1144 if (!(mask
& (1 << c
)))
1146 emit_mad(pc
, dst
[c
], src
[0][c
], src
[1][c
], src
[2][c
]);
1149 case TGSI_OPCODE_MAX
:
1150 for (c
= 0; c
< 4; c
++) {
1151 if (!(mask
& (1 << c
)))
1153 emit_minmax(pc
, 4, dst
[c
], src
[0][c
], src
[1][c
]);
1156 case TGSI_OPCODE_MIN
:
1157 for (c
= 0; c
< 4; c
++) {
1158 if (!(mask
& (1 << c
)))
1160 emit_minmax(pc
, 5, dst
[c
], src
[0][c
], src
[1][c
]);
1163 case TGSI_OPCODE_MOV
:
1164 for (c
= 0; c
< 4; c
++) {
1165 if (!(mask
& (1 << c
)))
1167 emit_mov(pc
, dst
[c
], src
[0][c
]);
1170 case TGSI_OPCODE_MUL
:
1171 for (c
= 0; c
< 4; c
++) {
1172 if (!(mask
& (1 << c
)))
1174 emit_mul(pc
, dst
[c
], src
[0][c
], src
[1][c
]);
1177 case TGSI_OPCODE_POW
:
1178 temp
= temp_temp(pc
);
1179 emit_pow(pc
, temp
, src
[0][0], src
[1][0]);
1180 for (c
= 0; c
< 4; c
++) {
1181 if (!(mask
& (1 << c
)))
1183 emit_mov(pc
, dst
[c
], temp
);
1186 case TGSI_OPCODE_RCP
:
1187 for (c
= 0; c
< 4; c
++) {
1188 if (!(mask
& (1 << c
)))
1190 emit_flop(pc
, 0, dst
[c
], src
[0][0]);
1193 case TGSI_OPCODE_RSQ
:
1194 for (c
= 0; c
< 4; c
++) {
1195 if (!(mask
& (1 << c
)))
1197 emit_flop(pc
, 2, dst
[c
], src
[0][0]);
1200 case TGSI_OPCODE_SCS
:
1201 temp
= temp_temp(pc
);
1202 emit_precossin(pc
, temp
, src
[0][0]);
1203 if (mask
& (1 << 0))
1204 emit_flop(pc
, 5, dst
[0], temp
);
1205 if (mask
& (1 << 1))
1206 emit_flop(pc
, 4, dst
[1], temp
);
1207 if (mask
& (1 << 2))
1208 emit_mov_immdval(pc
, dst
[2], 0.0);
1209 if (mask
& (1 << 3))
1210 emit_mov_immdval(pc
, dst
[3], 1.0);
1212 case TGSI_OPCODE_SGE
:
1213 for (c
= 0; c
< 4; c
++) {
1214 if (!(mask
& (1 << c
)))
1216 emit_set(pc
, 6, dst
[c
], src
[0][c
], src
[1][c
]);
1219 case TGSI_OPCODE_SIN
:
1220 temp
= temp_temp(pc
);
1221 emit_precossin(pc
, temp
, src
[0][0]);
1222 emit_flop(pc
, 4, temp
, temp
);
1223 for (c
= 0; c
< 4; c
++) {
1224 if (!(mask
& (1 << c
)))
1226 emit_mov(pc
, dst
[c
], temp
);
1229 case TGSI_OPCODE_SLT
:
1230 for (c
= 0; c
< 4; c
++) {
1231 if (!(mask
& (1 << c
)))
1233 emit_set(pc
, 1, dst
[c
], src
[0][c
], src
[1][c
]);
1236 case TGSI_OPCODE_SUB
:
1237 for (c
= 0; c
< 4; c
++) {
1238 if (!(mask
& (1 << c
)))
1240 emit_sub(pc
, dst
[c
], src
[0][c
], src
[1][c
]);
1243 case TGSI_OPCODE_TEX
:
1244 case TGSI_OPCODE_TXP
:
1246 struct nv50_reg
*t
[4];
1247 struct nv50_program_exec
*e
;
1249 alloc_temp4(pc
, t
, 0);
1250 emit_mov(pc
, t
[0], src
[0][0]);
1251 emit_mov(pc
, t
[1], src
[0][1]);
1254 e
->inst
[0] = 0xf6400000;
1255 e
->inst
[0] |= (unit
<< 9);
1257 e
->inst
[1] |= 0x0000c004;
1258 set_dst(pc
, t
[0], e
);
1261 if (mask
& (1 << 0)) emit_mov(pc
, dst
[0], t
[0]);
1262 if (mask
& (1 << 1)) emit_mov(pc
, dst
[1], t
[1]);
1263 if (mask
& (1 << 2)) emit_mov(pc
, dst
[2], t
[2]);
1264 if (mask
& (1 << 3)) emit_mov(pc
, dst
[3], t
[3]);
1269 case TGSI_OPCODE_XPD
:
1270 temp
= temp_temp(pc
);
1271 if (mask
& (1 << 0)) {
1272 emit_mul(pc
, temp
, src
[0][2], src
[1][1]);
1273 emit_msb(pc
, dst
[0], src
[0][1], src
[1][2], temp
);
1275 if (mask
& (1 << 1)) {
1276 emit_mul(pc
, temp
, src
[0][0], src
[1][2]);
1277 emit_msb(pc
, dst
[1], src
[0][2], src
[1][0], temp
);
1279 if (mask
& (1 << 2)) {
1280 emit_mul(pc
, temp
, src
[0][1], src
[1][0]);
1281 emit_msb(pc
, dst
[2], src
[0][0], src
[1][1], temp
);
1283 if (mask
& (1 << 3))
1284 emit_mov_immdval(pc
, dst
[3], 1.0);
1286 case TGSI_OPCODE_END
:
1289 NOUVEAU_ERR("invalid opcode %d\n", inst
->Instruction
.Opcode
);
1294 for (c
= 0; c
< 4; c
++) {
1295 struct nv50_program_exec
*e
;
1297 if (!(mask
& (1 << c
)))
1301 e
->inst
[0] = 0xa0000000; /* cvt */
1303 e
->inst
[1] |= (6 << 29); /* cvt */
1304 e
->inst
[1] |= 0x04000000; /* 32 bit */
1305 e
->inst
[1] |= (1 << 14); /* src .f32 */
1306 e
->inst
[1] |= ((1 << 5) << 14); /* .sat */
1307 set_dst(pc
, rdst
[c
], e
);
1308 set_src_0(pc
, dst
[c
], e
);
1313 for (i
= 0; i
< inst
->Instruction
.NumSrcRegs
; i
++) {
1314 for (c
= 0; c
< 4; c
++) {
1317 if (src
[i
][c
]->index
== -1 && src
[i
][c
]->type
== P_IMMD
)
1327 nv50_program_tx_prep(struct nv50_pc
*pc
)
1329 struct tgsi_parse_context p
;
1330 boolean ret
= FALSE
;
1333 tgsi_parse_init(&p
, pc
->p
->pipe
.tokens
);
1334 while (!tgsi_parse_end_of_tokens(&p
)) {
1335 const union tgsi_full_token
*tok
= &p
.FullToken
;
1337 tgsi_parse_token(&p
);
1338 switch (tok
->Token
.Type
) {
1339 case TGSI_TOKEN_TYPE_IMMEDIATE
:
1341 const struct tgsi_full_immediate
*imm
=
1342 &p
.FullToken
.FullImmediate
;
1344 ctor_immd(pc
, imm
->u
.ImmediateFloat32
[0].Float
,
1345 imm
->u
.ImmediateFloat32
[1].Float
,
1346 imm
->u
.ImmediateFloat32
[2].Float
,
1347 imm
->u
.ImmediateFloat32
[3].Float
);
1350 case TGSI_TOKEN_TYPE_DECLARATION
:
1352 const struct tgsi_full_declaration
*d
;
1355 d
= &p
.FullToken
.FullDeclaration
;
1356 last
= d
->DeclarationRange
.Last
;
1358 switch (d
->Declaration
.File
) {
1359 case TGSI_FILE_TEMPORARY
:
1360 if (pc
->temp_nr
< (last
+ 1))
1361 pc
->temp_nr
= last
+ 1;
1363 case TGSI_FILE_OUTPUT
:
1364 if (pc
->result_nr
< (last
+ 1))
1365 pc
->result_nr
= last
+ 1;
1367 case TGSI_FILE_INPUT
:
1368 if (pc
->attr_nr
< (last
+ 1))
1369 pc
->attr_nr
= last
+ 1;
1371 case TGSI_FILE_CONSTANT
:
1372 if (pc
->param_nr
< (last
+ 1))
1373 pc
->param_nr
= last
+ 1;
1375 case TGSI_FILE_SAMPLER
:
1378 NOUVEAU_ERR("bad decl file %d\n",
1379 d
->Declaration
.File
);
1384 case TGSI_TOKEN_TYPE_INSTRUCTION
:
1392 pc
->temp
= CALLOC(pc
->temp_nr
* 4, sizeof(struct nv50_reg
));
1396 for (i
= 0; i
< pc
->temp_nr
; i
++) {
1397 for (c
= 0; c
< 4; c
++) {
1398 pc
->temp
[i
*4+c
].type
= P_TEMP
;
1399 pc
->temp
[i
*4+c
].hw
= -1;
1400 pc
->temp
[i
*4+c
].index
= i
;
1406 struct nv50_reg
*iv
= NULL
;
1409 pc
->attr
= CALLOC(pc
->attr_nr
* 4, sizeof(struct nv50_reg
));
1413 if (pc
->p
->type
== PIPE_SHADER_FRAGMENT
) {
1414 iv
= alloc_temp(pc
, NULL
);
1415 emit_interp(pc
, iv
, iv
, NULL
);
1416 emit_flop(pc
, 0, iv
, iv
);
1420 for (i
= 0; i
< pc
->attr_nr
; i
++) {
1421 struct nv50_reg
*a
= &pc
->attr
[i
*4];
1423 for (c
= 0; c
< 4; c
++) {
1424 if (pc
->p
->type
== PIPE_SHADER_FRAGMENT
) {
1425 struct nv50_reg
*at
=
1426 alloc_temp(pc
, NULL
);
1427 pc
->attr
[i
*4+c
].type
= at
->type
;
1428 pc
->attr
[i
*4+c
].hw
= at
->hw
;
1429 pc
->attr
[i
*4+c
].index
= at
->index
;
1431 pc
->p
->cfg
.vp
.attr
[aid
/32] |=
1433 pc
->attr
[i
*4+c
].type
= P_ATTR
;
1434 pc
->attr
[i
*4+c
].hw
= aid
++;
1435 pc
->attr
[i
*4+c
].index
= i
;
1439 if (pc
->p
->type
!= PIPE_SHADER_FRAGMENT
)
1442 emit_interp(pc
, &a
[0], &a
[0], iv
);
1443 emit_interp(pc
, &a
[1], &a
[1], iv
);
1444 emit_interp(pc
, &a
[2], &a
[2], iv
);
1445 emit_interp(pc
, &a
[3], &a
[3], iv
);
1452 if (pc
->result_nr
) {
1455 pc
->result
= CALLOC(pc
->result_nr
* 4, sizeof(struct nv50_reg
));
1459 for (i
= 0; i
< pc
->result_nr
; i
++) {
1460 for (c
= 0; c
< 4; c
++) {
1461 if (pc
->p
->type
== PIPE_SHADER_FRAGMENT
) {
1462 pc
->result
[i
*4+c
].type
= P_TEMP
;
1463 pc
->result
[i
*4+c
].hw
= -1;
1465 pc
->result
[i
*4+c
].type
= P_RESULT
;
1466 pc
->result
[i
*4+c
].hw
= rid
++;
1468 pc
->result
[i
*4+c
].index
= i
;
1476 pc
->param
= CALLOC(pc
->param_nr
* 4, sizeof(struct nv50_reg
));
1480 for (i
= 0; i
< pc
->param_nr
; i
++) {
1481 for (c
= 0; c
< 4; c
++) {
1482 pc
->param
[i
*4+c
].type
= P_CONST
;
1483 pc
->param
[i
*4+c
].hw
= rid
++;
1484 pc
->param
[i
*4+c
].index
= i
;
1490 int rid
= pc
->param_nr
* 4;
1492 pc
->immd
= CALLOC(pc
->immd_nr
* 4, sizeof(struct nv50_reg
));
1496 for (i
= 0; i
< pc
->immd_nr
; i
++) {
1497 for (c
= 0; c
< 4; c
++) {
1498 pc
->immd
[i
*4+c
].type
= P_IMMD
;
1499 pc
->immd
[i
*4+c
].hw
= rid
++;
1500 pc
->immd
[i
*4+c
].index
= i
;
1507 tgsi_parse_free(&p
);
1512 free_nv50_pc(struct nv50_pc
*pc
)
1527 for (i
= 0; i
< NV50_SU_MAX_TEMP
; i
++) {
1528 /* deallocate fragment program attributes */
1529 if (pc
->r_temp
[i
] && pc
->r_temp
[i
]->index
== -1)
1530 FREE(pc
->r_temp
[i
]);
1537 nv50_program_tx(struct nv50_program
*p
)
1539 struct tgsi_parse_context parse
;
1543 pc
= CALLOC_STRUCT(nv50_pc
);
1547 pc
->p
->cfg
.high_temp
= 4;
1549 ret
= nv50_program_tx_prep(pc
);
1553 tgsi_parse_init(&parse
, pc
->p
->pipe
.tokens
);
1554 while (!tgsi_parse_end_of_tokens(&parse
)) {
1555 const union tgsi_full_token
*tok
= &parse
.FullToken
;
1557 tgsi_parse_token(&parse
);
1559 switch (tok
->Token
.Type
) {
1560 case TGSI_TOKEN_TYPE_INSTRUCTION
:
1561 ret
= nv50_program_tx_insn(pc
, tok
);
1570 if (p
->type
== PIPE_SHADER_FRAGMENT
) {
1571 struct nv50_reg out
;
1574 for (out
.hw
= 0; out
.hw
< pc
->result_nr
* 4; out
.hw
++)
1575 emit_mov(pc
, &out
, &pc
->result
[out
.hw
]);
1578 assert(is_long(pc
->p
->exec_tail
) && !is_immd(pc
->p
->exec_head
));
1579 pc
->p
->exec_tail
->inst
[1] |= 0x00000001;
1581 p
->param_nr
= pc
->param_nr
* 4;
1582 p
->immd_nr
= pc
->immd_nr
* 4;
1583 p
->immd
= pc
->immd_buf
;
1586 tgsi_parse_free(&parse
);
1594 nv50_program_validate(struct nv50_context
*nv50
, struct nv50_program
*p
)
1596 if (nv50_program_tx(p
) == FALSE
)
1598 p
->translated
= TRUE
;
1602 nv50_program_upload_data(struct nv50_context
*nv50
, float *map
,
1603 unsigned start
, unsigned count
)
1605 struct nouveau_channel
*chan
= nv50
->screen
->nvws
->channel
;
1606 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
1609 unsigned nr
= count
> 2047 ? 2047 : count
;
1611 BEGIN_RING(chan
, tesla
, 0x00000f00, 1);
1612 OUT_RING (chan
, (NV50_CB_PMISC
<< 0) | (start
<< 8));
1613 BEGIN_RING(chan
, tesla
, 0x40000f04, nr
);
1614 OUT_RINGp (chan
, map
, nr
);
1623 nv50_program_validate_data(struct nv50_context
*nv50
, struct nv50_program
*p
)
1625 struct nouveau_winsys
*nvws
= nv50
->screen
->nvws
;
1626 struct pipe_winsys
*ws
= nv50
->pipe
.winsys
;
1627 unsigned nr
= p
->param_nr
+ p
->immd_nr
;
1629 if (!p
->data
&& nr
) {
1630 struct nouveau_resource
*heap
= nv50
->screen
->vp_data_heap
;
1632 if (nvws
->res_alloc(heap
, nr
, p
, &p
->data
)) {
1633 while (heap
->next
&& heap
->size
< nr
) {
1634 struct nv50_program
*evict
= heap
->next
->priv
;
1635 nvws
->res_free(&evict
->data
);
1638 if (nvws
->res_alloc(heap
, nr
, p
, &p
->data
))
1644 float *map
= ws
->buffer_map(ws
, nv50
->constbuf
[p
->type
],
1645 PIPE_BUFFER_USAGE_CPU_READ
);
1646 nv50_program_upload_data(nv50
, map
, p
->data
->start
,
1648 ws
->buffer_unmap(ws
, nv50
->constbuf
[p
->type
]);
1652 nv50_program_upload_data(nv50
, p
->immd
,
1653 p
->data
->start
+ p
->param_nr
,
1659 nv50_program_validate_code(struct nv50_context
*nv50
, struct nv50_program
*p
)
1661 struct nouveau_channel
*chan
= nv50
->screen
->nvws
->channel
;
1662 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
1663 struct pipe_screen
*screen
= nv50
->pipe
.screen
;
1664 struct nv50_program_exec
*e
;
1665 struct nouveau_stateobj
*so
;
1666 const unsigned flags
= NOUVEAU_BO_VRAM
| NOUVEAU_BO_WR
;
1667 unsigned start
, count
, *up
, *ptr
;
1668 boolean upload
= FALSE
;
1671 p
->buffer
= screen
->buffer_create(screen
, 0x100, 0, p
->exec_size
* 4);
1675 if (p
->data
&& p
->data
->start
!= p
->data_start
) {
1676 for (e
= p
->exec_head
; e
; e
= e
->next
) {
1679 if (e
->param
.index
< 0)
1681 ei
= e
->param
.shift
>> 5;
1682 ci
= e
->param
.index
+ p
->data
->start
;
1684 e
->inst
[ei
] &= ~e
->param
.mask
;
1685 e
->inst
[ei
] |= (ci
<< e
->param
.shift
);
1688 p
->data_start
= p
->data
->start
;
1695 #ifdef NV50_PROGRAM_DUMP
1696 NOUVEAU_ERR("-------\n");
1697 for (e
= p
->exec_head
; e
; e
= e
->next
) {
1698 NOUVEAU_ERR("0x%08x\n", e
->inst
[0]);
1700 NOUVEAU_ERR("0x%08x\n", e
->inst
[1]);
1704 up
= ptr
= MALLOC(p
->exec_size
* 4);
1705 for (e
= p
->exec_head
; e
; e
= e
->next
) {
1706 *(ptr
++) = e
->inst
[0];
1708 *(ptr
++) = e
->inst
[1];
1712 so_method(so
, nv50
->screen
->tesla
, 0x1280, 3);
1713 so_reloc (so
, p
->buffer
, 0, flags
| NOUVEAU_BO_HIGH
, 0, 0);
1714 so_reloc (so
, p
->buffer
, 0, flags
| NOUVEAU_BO_LOW
, 0, 0);
1715 so_data (so
, (NV50_CB_PUPLOAD
<< 16) | 0x0800); //(p->exec_size * 4));
1717 start
= 0; count
= p
->exec_size
;
1719 struct nouveau_winsys
*nvws
= nv50
->screen
->nvws
;
1724 nr
= MIN2(count
, 2047);
1725 nr
= MIN2(nvws
->channel
->pushbuf
->remaining
, nr
);
1726 if (nvws
->channel
->pushbuf
->remaining
< (nr
+ 3)) {
1731 BEGIN_RING(chan
, tesla
, 0x0f00, 1);
1732 OUT_RING (chan
, (start
<< 8) | NV50_CB_PUPLOAD
);
1733 BEGIN_RING(chan
, tesla
, 0x40000f04, nr
);
1734 OUT_RINGp (chan
, up
+ start
, nr
);
1745 nv50_vertprog_validate(struct nv50_context
*nv50
)
1747 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
1748 struct nv50_program
*p
= nv50
->vertprog
;
1749 struct nouveau_stateobj
*so
;
1751 if (!p
->translated
) {
1752 nv50_program_validate(nv50
, p
);
1757 nv50_program_validate_data(nv50
, p
);
1758 nv50_program_validate_code(nv50
, p
);
1761 so_method(so
, tesla
, NV50TCL_VP_ADDRESS_HIGH
, 2);
1762 so_reloc (so
, p
->buffer
, 0, NOUVEAU_BO_VRAM
| NOUVEAU_BO_RD
|
1763 NOUVEAU_BO_HIGH
, 0, 0);
1764 so_reloc (so
, p
->buffer
, 0, NOUVEAU_BO_VRAM
| NOUVEAU_BO_RD
|
1765 NOUVEAU_BO_LOW
, 0, 0);
1766 so_method(so
, tesla
, 0x1650, 2);
1767 so_data (so
, p
->cfg
.vp
.attr
[0]);
1768 so_data (so
, p
->cfg
.vp
.attr
[1]);
1769 so_method(so
, tesla
, 0x16b8, 1);
1770 so_data (so
, p
->cfg
.high_result
);
1771 so_method(so
, tesla
, 0x16ac, 2);
1772 so_data (so
, p
->cfg
.high_result
); //8);
1773 so_data (so
, p
->cfg
.high_temp
);
1774 so_method(so
, tesla
, 0x140c, 1);
1775 so_data (so
, 0); /* program start offset */
1776 so_ref(so
, &nv50
->state
.vertprog
);
1781 nv50_fragprog_validate(struct nv50_context
*nv50
)
1783 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
1784 struct nv50_program
*p
= nv50
->fragprog
;
1785 struct nouveau_stateobj
*so
;
1787 if (!p
->translated
) {
1788 nv50_program_validate(nv50
, p
);
1793 nv50_program_validate_data(nv50
, p
);
1794 nv50_program_validate_code(nv50
, p
);
1797 so_method(so
, tesla
, NV50TCL_FP_ADDRESS_HIGH
, 2);
1798 so_reloc (so
, p
->buffer
, 0, NOUVEAU_BO_VRAM
| NOUVEAU_BO_RD
|
1799 NOUVEAU_BO_HIGH
, 0, 0);
1800 so_reloc (so
, p
->buffer
, 0, NOUVEAU_BO_VRAM
| NOUVEAU_BO_RD
|
1801 NOUVEAU_BO_LOW
, 0, 0);
1802 so_method(so
, tesla
, 0x1904, 4);
1803 so_data (so
, 0x00040404); /* p: 0x01000404 */
1804 so_data (so
, 0x00000004);
1805 so_data (so
, 0x00000000);
1806 so_data (so
, 0x00000000);
1807 so_method(so
, tesla
, 0x16bc, 3); /*XXX: fixme */
1808 so_data (so
, 0x03020100);
1809 so_data (so
, 0x07060504);
1810 so_data (so
, 0x0b0a0908);
1811 so_method(so
, tesla
, 0x1988, 2);
1812 so_data (so
, 0x08080408); //0x08040404); /* p: 0x0f000401 */
1813 so_data (so
, p
->cfg
.high_temp
);
1814 so_method(so
, tesla
, 0x1414, 1);
1815 so_data (so
, 0); /* program start offset */
1816 so_ref(so
, &nv50
->state
.fragprog
);
1821 nv50_program_destroy(struct nv50_context
*nv50
, struct nv50_program
*p
)
1823 struct pipe_screen
*pscreen
= nv50
->pipe
.screen
;
1825 while (p
->exec_head
) {
1826 struct nv50_program_exec
*e
= p
->exec_head
;
1828 p
->exec_head
= e
->next
;
1831 p
->exec_tail
= NULL
;
1835 pipe_buffer_reference(&p
->buffer
, NULL
);
1837 nv50
->screen
->nvws
->res_free(&p
->data
);