1 #include "pipe/p_context.h"
2 #include "pipe/p_defines.h"
3 #include "pipe/p_state.h"
4 #include "pipe/p_inlines.h"
6 #include "pipe/p_shader_tokens.h"
7 #include "tgsi/tgsi_parse.h"
8 #include "tgsi/tgsi_util.h"
10 #include "nv40_context.h"
20 #define MASK_ALL (MASK_X|MASK_Y|MASK_Z|MASK_W)
21 #define DEF_SCALE NV40_FP_OP_DST_SCALE_1X
22 #define DEF_CTEST NV40_FP_OP_COND_TR
23 #include "nv40_shader.h"
25 #define swz(s,x,y,z,w) nv40_sr_swz((s), SWZ_##x, SWZ_##y, SWZ_##z, SWZ_##w)
26 #define neg(s) nv40_sr_neg((s))
27 #define abs(s) nv40_sr_abs((s))
28 #define scale(s,v) nv40_sr_scale((s), NV40_FP_OP_DST_SCALE_##v)
30 #define MAX_CONSTS 128
33 struct nv40_fragment_program
*fp
;
35 uint attrib_map
[PIPE_MAX_SHADER_INPUTS
];
38 unsigned r_temps_discard
;
39 struct nv40_sreg r_result
[PIPE_MAX_SHADER_OUTPUTS
];
40 struct nv40_sreg
*r_temp
;
53 struct nv40_sreg imm
[MAX_IMM
];
57 static INLINE
struct nv40_sreg
58 temp(struct nv40_fpc
*fpc
)
60 int idx
= ffs(~fpc
->r_temps
) - 1;
63 NOUVEAU_ERR("out of temps!!\n");
65 return nv40_sr(NV40SR_TEMP
, 0);
68 fpc
->r_temps
|= (1 << idx
);
69 fpc
->r_temps_discard
|= (1 << idx
);
70 return nv40_sr(NV40SR_TEMP
, idx
);
74 release_temps(struct nv40_fpc
*fpc
)
76 fpc
->r_temps
&= ~fpc
->r_temps_discard
;
77 fpc
->r_temps_discard
= 0;
80 static INLINE
struct nv40_sreg
81 constant(struct nv40_fpc
*fpc
, int pipe
, float vals
[4])
85 if (fpc
->nr_consts
== MAX_CONSTS
)
87 idx
= fpc
->nr_consts
++;
89 fpc
->consts
[idx
].pipe
= pipe
;
91 memcpy(fpc
->consts
[idx
].vals
, vals
, 4 * sizeof(float));
92 return nv40_sr(NV40SR_CONST
, idx
);
95 #define arith(cc,s,o,d,m,s0,s1,s2) \
96 nv40_fp_arith((cc), (s), NV40_FP_OP_OPCODE_##o, \
97 (d), (m), (s0), (s1), (s2))
98 #define tex(cc,s,o,u,d,m,s0,s1,s2) \
99 nv40_fp_tex((cc), (s), NV40_FP_OP_OPCODE_##o, (u), \
100 (d), (m), (s0), none, none)
103 grow_insns(struct nv40_fpc
*fpc
, int size
)
105 struct nv40_fragment_program
*fp
= fpc
->fp
;
107 fp
->insn_len
+= size
;
108 fp
->insn
= realloc(fp
->insn
, sizeof(uint32_t) * fp
->insn_len
);
112 emit_src(struct nv40_fpc
*fpc
, int pos
, struct nv40_sreg src
)
114 struct nv40_fragment_program
*fp
= fpc
->fp
;
115 uint32_t *hw
= &fp
->insn
[fpc
->inst_offset
];
120 sr
|= (NV40_FP_REG_TYPE_INPUT
<< NV40_FP_REG_TYPE_SHIFT
);
121 hw
[0] |= (src
.index
<< NV40_FP_OP_INPUT_SRC_SHIFT
);
124 sr
|= NV40_FP_REG_SRC_HALF
;
127 sr
|= (NV40_FP_REG_TYPE_TEMP
<< NV40_FP_REG_TYPE_SHIFT
);
128 sr
|= (src
.index
<< NV40_FP_REG_SRC_SHIFT
);
131 if (!fpc
->have_const
) {
136 hw
= &fp
->insn
[fpc
->inst_offset
];
137 if (fpc
->consts
[src
.index
].pipe
>= 0) {
138 struct nv40_fragment_program_data
*fpd
;
140 fp
->consts
= realloc(fp
->consts
, ++fp
->nr_consts
*
142 fpd
= &fp
->consts
[fp
->nr_consts
- 1];
143 fpd
->offset
= fpc
->inst_offset
+ 4;
144 fpd
->index
= fpc
->consts
[src
.index
].pipe
;
145 memset(&fp
->insn
[fpd
->offset
], 0, sizeof(uint32_t) * 4);
147 memcpy(&fp
->insn
[fpc
->inst_offset
+ 4],
148 fpc
->consts
[src
.index
].vals
,
149 sizeof(uint32_t) * 4);
152 sr
|= (NV40_FP_REG_TYPE_CONST
<< NV40_FP_REG_TYPE_SHIFT
);
155 sr
|= (NV40_FP_REG_TYPE_INPUT
<< NV40_FP_REG_TYPE_SHIFT
);
162 sr
|= NV40_FP_REG_NEGATE
;
165 hw
[1] |= (1 << (29 + pos
));
167 sr
|= ((src
.swz
[0] << NV40_FP_REG_SWZ_X_SHIFT
) |
168 (src
.swz
[1] << NV40_FP_REG_SWZ_Y_SHIFT
) |
169 (src
.swz
[2] << NV40_FP_REG_SWZ_Z_SHIFT
) |
170 (src
.swz
[3] << NV40_FP_REG_SWZ_W_SHIFT
));
176 emit_dst(struct nv40_fpc
*fpc
, struct nv40_sreg dst
)
178 struct nv40_fragment_program
*fp
= fpc
->fp
;
179 uint32_t *hw
= &fp
->insn
[fpc
->inst_offset
];
183 if (fpc
->num_regs
< (dst
.index
+ 1))
184 fpc
->num_regs
= dst
.index
+ 1;
187 if (dst
.index
== 1) {
188 fp
->fp_control
|= 0xe;
190 hw
[0] |= NV40_FP_OP_OUT_REG_HALF
;
200 hw
[0] |= (dst
.index
<< NV40_FP_OP_OUT_REG_SHIFT
);
204 nv40_fp_arith(struct nv40_fpc
*fpc
, int sat
, int op
,
205 struct nv40_sreg dst
, int mask
,
206 struct nv40_sreg s0
, struct nv40_sreg s1
, struct nv40_sreg s2
)
208 struct nv40_fragment_program
*fp
= fpc
->fp
;
211 fpc
->inst_offset
= fp
->insn_len
;
214 hw
= &fp
->insn
[fpc
->inst_offset
];
215 memset(hw
, 0, sizeof(uint32_t) * 4);
217 if (op
== NV40_FP_OP_OPCODE_KIL
)
218 fp
->fp_control
|= NV40TCL_FP_CONTROL_KIL
;
219 hw
[0] |= (op
<< NV40_FP_OP_OPCODE_SHIFT
);
220 hw
[0] |= (mask
<< NV40_FP_OP_OUTMASK_SHIFT
);
221 hw
[2] |= (dst
.dst_scale
<< NV40_FP_OP_DST_SCALE_SHIFT
);
224 hw
[0] |= NV40_FP_OP_OUT_SAT
;
227 hw
[0] |= NV40_FP_OP_COND_WRITE_ENABLE
;
228 hw
[1] |= (dst
.cc_test
<< NV40_FP_OP_COND_SHIFT
);
229 hw
[1] |= ((dst
.cc_swz
[0] << NV40_FP_OP_COND_SWZ_X_SHIFT
) |
230 (dst
.cc_swz
[1] << NV40_FP_OP_COND_SWZ_Y_SHIFT
) |
231 (dst
.cc_swz
[2] << NV40_FP_OP_COND_SWZ_Z_SHIFT
) |
232 (dst
.cc_swz
[3] << NV40_FP_OP_COND_SWZ_W_SHIFT
));
235 emit_src(fpc
, 0, s0
);
236 emit_src(fpc
, 1, s1
);
237 emit_src(fpc
, 2, s2
);
241 nv40_fp_tex(struct nv40_fpc
*fpc
, int sat
, int op
, int unit
,
242 struct nv40_sreg dst
, int mask
,
243 struct nv40_sreg s0
, struct nv40_sreg s1
, struct nv40_sreg s2
)
245 struct nv40_fragment_program
*fp
= fpc
->fp
;
247 nv40_fp_arith(fpc
, sat
, op
, dst
, mask
, s0
, s1
, s2
);
249 fp
->insn
[fpc
->inst_offset
] |= (unit
<< NV40_FP_OP_TEX_UNIT_SHIFT
);
250 fp
->samplers
|= (1 << unit
);
253 static INLINE
struct nv40_sreg
254 tgsi_src(struct nv40_fpc
*fpc
, const struct tgsi_full_src_register
*fsrc
)
256 struct nv40_sreg src
;
258 switch (fsrc
->SrcRegister
.File
) {
259 case TGSI_FILE_INPUT
:
260 src
= nv40_sr(NV40SR_INPUT
,
261 fpc
->attrib_map
[fsrc
->SrcRegister
.Index
]);
263 case TGSI_FILE_CONSTANT
:
264 src
= constant(fpc
, fsrc
->SrcRegister
.Index
, NULL
);
266 case TGSI_FILE_IMMEDIATE
:
267 assert(fsrc
->SrcRegister
.Index
< fpc
->nr_imm
);
268 src
= fpc
->imm
[fsrc
->SrcRegister
.Index
];
270 case TGSI_FILE_TEMPORARY
:
271 src
= fpc
->r_temp
[fsrc
->SrcRegister
.Index
];
273 /* NV40 fragprog result regs are just temps, so this is simple */
274 case TGSI_FILE_OUTPUT
:
275 src
= fpc
->r_result
[fsrc
->SrcRegister
.Index
];
278 NOUVEAU_ERR("bad src file\n");
282 src
.abs
= fsrc
->SrcRegisterExtMod
.Absolute
;
283 src
.negate
= fsrc
->SrcRegister
.Negate
;
284 src
.swz
[0] = fsrc
->SrcRegister
.SwizzleX
;
285 src
.swz
[1] = fsrc
->SrcRegister
.SwizzleY
;
286 src
.swz
[2] = fsrc
->SrcRegister
.SwizzleZ
;
287 src
.swz
[3] = fsrc
->SrcRegister
.SwizzleW
;
291 static INLINE
struct nv40_sreg
292 tgsi_dst(struct nv40_fpc
*fpc
, const struct tgsi_full_dst_register
*fdst
) {
293 switch (fdst
->DstRegister
.File
) {
294 case TGSI_FILE_OUTPUT
:
295 return fpc
->r_result
[fdst
->DstRegister
.Index
];
296 case TGSI_FILE_TEMPORARY
:
297 return fpc
->r_temp
[fdst
->DstRegister
.Index
];
299 return nv40_sr(NV40SR_NONE
, 0);
301 NOUVEAU_ERR("bad dst file %d\n", fdst
->DstRegister
.File
);
302 return nv40_sr(NV40SR_NONE
, 0);
311 if (tgsi
& TGSI_WRITEMASK_X
) mask
|= MASK_X
;
312 if (tgsi
& TGSI_WRITEMASK_Y
) mask
|= MASK_Y
;
313 if (tgsi
& TGSI_WRITEMASK_Z
) mask
|= MASK_Z
;
314 if (tgsi
& TGSI_WRITEMASK_W
) mask
|= MASK_W
;
319 src_native_swz(struct nv40_fpc
*fpc
, const struct tgsi_full_src_register
*fsrc
,
320 struct nv40_sreg
*src
)
322 const struct nv40_sreg none
= nv40_sr(NV40SR_NONE
, 0);
323 struct nv40_sreg tgsi
= tgsi_src(fpc
, fsrc
);
324 uint mask
= 0, zero_mask
= 0, one_mask
= 0, neg_mask
= 0;
325 uint neg
[4] = { fsrc
->SrcRegisterExtSwz
.NegateX
,
326 fsrc
->SrcRegisterExtSwz
.NegateY
,
327 fsrc
->SrcRegisterExtSwz
.NegateZ
,
328 fsrc
->SrcRegisterExtSwz
.NegateW
};
331 for (c
= 0; c
< 4; c
++) {
332 switch (tgsi_util_get_full_src_register_swizzle(fsrc
, c
)) {
343 if (!tgsi
.negate
&& neg
[c
])
344 neg_mask
|= (1 << c
);
347 if (mask
== MASK_ALL
&& !neg_mask
)
353 arith(fpc
, 0, MOV
, *src
, mask
, tgsi
, none
, none
);
356 arith(fpc
, 0, SFL
, *src
, zero_mask
, *src
, none
, none
);
359 arith(fpc
, 0, STR
, *src
, one_mask
, *src
, none
, none
);
362 struct nv40_sreg one
= temp(fpc
);
363 arith(fpc
, 0, STR
, one
, neg_mask
, one
, none
, none
);
364 arith(fpc
, 0, MUL
, *src
, neg_mask
, *src
, neg(one
), none
);
371 nv40_fragprog_parse_instruction(struct nv40_fpc
*fpc
,
372 const struct tgsi_full_instruction
*finst
)
374 const struct nv40_sreg none
= nv40_sr(NV40SR_NONE
, 0);
375 struct nv40_sreg src
[3], dst
, tmp
;
377 int ai
= -1, ci
= -1, ii
= -1;
380 if (finst
->Instruction
.Opcode
== TGSI_OPCODE_END
)
383 for (i
= 0; i
< finst
->Instruction
.NumSrcRegs
; i
++) {
384 const struct tgsi_full_src_register
*fsrc
;
386 fsrc
= &finst
->FullSrcRegisters
[i
];
387 if (fsrc
->SrcRegister
.File
== TGSI_FILE_TEMPORARY
) {
388 src
[i
] = tgsi_src(fpc
, fsrc
);
392 for (i
= 0; i
< finst
->Instruction
.NumSrcRegs
; i
++) {
393 const struct tgsi_full_src_register
*fsrc
;
395 fsrc
= &finst
->FullSrcRegisters
[i
];
397 switch (fsrc
->SrcRegister
.File
) {
398 case TGSI_FILE_INPUT
:
399 case TGSI_FILE_CONSTANT
:
400 case TGSI_FILE_TEMPORARY
:
401 if (!src_native_swz(fpc
, fsrc
, &src
[i
]))
408 switch (fsrc
->SrcRegister
.File
) {
409 case TGSI_FILE_INPUT
:
410 if (ai
== -1 || ai
== fsrc
->SrcRegister
.Index
) {
411 ai
= fsrc
->SrcRegister
.Index
;
412 src
[i
] = tgsi_src(fpc
, fsrc
);
415 arith(fpc
, 0, MOV
, src
[i
], MASK_ALL
,
416 tgsi_src(fpc
, fsrc
), none
, none
);
419 case TGSI_FILE_CONSTANT
:
420 if ((ci
== -1 && ii
== -1) ||
421 ci
== fsrc
->SrcRegister
.Index
) {
422 ci
= fsrc
->SrcRegister
.Index
;
423 src
[i
] = tgsi_src(fpc
, fsrc
);
426 arith(fpc
, 0, MOV
, src
[i
], MASK_ALL
,
427 tgsi_src(fpc
, fsrc
), none
, none
);
430 case TGSI_FILE_IMMEDIATE
:
431 if ((ci
== -1 && ii
== -1) ||
432 ii
== fsrc
->SrcRegister
.Index
) {
433 ii
= fsrc
->SrcRegister
.Index
;
434 src
[i
] = tgsi_src(fpc
, fsrc
);
437 arith(fpc
, 0, MOV
, src
[i
], MASK_ALL
,
438 tgsi_src(fpc
, fsrc
), none
, none
);
441 case TGSI_FILE_TEMPORARY
:
444 case TGSI_FILE_SAMPLER
:
445 unit
= fsrc
->SrcRegister
.Index
;
447 case TGSI_FILE_OUTPUT
:
450 NOUVEAU_ERR("bad src file\n");
455 dst
= tgsi_dst(fpc
, &finst
->FullDstRegisters
[0]);
456 mask
= tgsi_mask(finst
->FullDstRegisters
[0].DstRegister
.WriteMask
);
457 sat
= (finst
->Instruction
.Saturate
== TGSI_SAT_ZERO_ONE
);
459 switch (finst
->Instruction
.Opcode
) {
460 case TGSI_OPCODE_ABS
:
461 arith(fpc
, sat
, MOV
, dst
, mask
, abs(src
[0]), none
, none
);
463 case TGSI_OPCODE_ADD
:
464 arith(fpc
, sat
, ADD
, dst
, mask
, src
[0], src
[1], none
);
466 case TGSI_OPCODE_CMP
:
468 arith(fpc
, sat
, MOV
, dst
, mask
, src
[2], none
, none
);
470 arith(fpc
, 0, MOV
, tmp
, 0xf, src
[0], none
, none
);
471 dst
.cc_test
= NV40_VP_INST_COND_LT
;
472 arith(fpc
, sat
, MOV
, dst
, mask
, src
[1], none
, none
);
474 case TGSI_OPCODE_COS
:
475 arith(fpc
, sat
, COS
, dst
, mask
, src
[0], none
, none
);
477 case TGSI_OPCODE_DDX
:
478 if (mask
& (MASK_Z
| MASK_W
)) {
480 arith(fpc
, sat
, DDX
, tmp
, MASK_X
| MASK_Y
,
481 swz(src
[0], Z
, W
, Z
, W
), none
, none
);
482 arith(fpc
, 0, MOV
, tmp
, MASK_Z
| MASK_W
,
483 swz(tmp
, X
, Y
, X
, Y
), none
, none
);
484 arith(fpc
, sat
, DDX
, tmp
, MASK_X
| MASK_Y
, src
[0],
486 arith(fpc
, 0, MOV
, dst
, mask
, tmp
, none
, none
);
488 arith(fpc
, sat
, DDX
, dst
, mask
, src
[0], none
, none
);
491 case TGSI_OPCODE_DDY
:
492 if (mask
& (MASK_Z
| MASK_W
)) {
494 arith(fpc
, sat
, DDY
, tmp
, MASK_X
| MASK_Y
,
495 swz(src
[0], Z
, W
, Z
, W
), none
, none
);
496 arith(fpc
, 0, MOV
, tmp
, MASK_Z
| MASK_W
,
497 swz(tmp
, X
, Y
, X
, Y
), none
, none
);
498 arith(fpc
, sat
, DDY
, tmp
, MASK_X
| MASK_Y
, src
[0],
500 arith(fpc
, 0, MOV
, dst
, mask
, tmp
, none
, none
);
502 arith(fpc
, sat
, DDY
, dst
, mask
, src
[0], none
, none
);
505 case TGSI_OPCODE_DP3
:
506 arith(fpc
, sat
, DP3
, dst
, mask
, src
[0], src
[1], none
);
508 case TGSI_OPCODE_DP4
:
509 arith(fpc
, sat
, DP4
, dst
, mask
, src
[0], src
[1], none
);
511 case TGSI_OPCODE_DPH
:
513 arith(fpc
, 0, DP3
, tmp
, MASK_X
, src
[0], src
[1], none
);
514 arith(fpc
, sat
, ADD
, dst
, mask
, swz(tmp
, X
, X
, X
, X
),
515 swz(src
[1], W
, W
, W
, W
), none
);
517 case TGSI_OPCODE_DST
:
518 arith(fpc
, sat
, DST
, dst
, mask
, src
[0], src
[1], none
);
520 case TGSI_OPCODE_EX2
:
521 arith(fpc
, sat
, EX2
, dst
, mask
, src
[0], none
, none
);
523 case TGSI_OPCODE_FLR
:
524 arith(fpc
, sat
, FLR
, dst
, mask
, src
[0], none
, none
);
526 case TGSI_OPCODE_FRC
:
527 arith(fpc
, sat
, FRC
, dst
, mask
, src
[0], none
, none
);
529 case TGSI_OPCODE_KILP
:
530 arith(fpc
, 0, KIL
, none
, 0, none
, none
, none
);
532 case TGSI_OPCODE_KIL
:
533 dst
= nv40_sr(NV40SR_NONE
, 0);
535 arith(fpc
, 0, MOV
, dst
, MASK_ALL
, src
[0], none
, none
);
536 dst
.cc_update
= 0; dst
.cc_test
= NV40_FP_OP_COND_LT
;
537 arith(fpc
, 0, KIL
, dst
, 0, none
, none
, none
);
539 case TGSI_OPCODE_LG2
:
540 arith(fpc
, sat
, LG2
, dst
, mask
, src
[0], none
, none
);
542 // case TGSI_OPCODE_LIT:
543 case TGSI_OPCODE_LRP
:
545 arith(fpc
, 0, MAD
, tmp
, mask
, neg(src
[0]), src
[2], src
[2]);
546 arith(fpc
, sat
, MAD
, dst
, mask
, src
[0], src
[1], tmp
);
548 case TGSI_OPCODE_MAD
:
549 arith(fpc
, sat
, MAD
, dst
, mask
, src
[0], src
[1], src
[2]);
551 case TGSI_OPCODE_MAX
:
552 arith(fpc
, sat
, MAX
, dst
, mask
, src
[0], src
[1], none
);
554 case TGSI_OPCODE_MIN
:
555 arith(fpc
, sat
, MIN
, dst
, mask
, src
[0], src
[1], none
);
557 case TGSI_OPCODE_MOV
:
558 arith(fpc
, sat
, MOV
, dst
, mask
, src
[0], none
, none
);
560 case TGSI_OPCODE_MUL
:
561 arith(fpc
, sat
, MUL
, dst
, mask
, src
[0], src
[1], none
);
563 case TGSI_OPCODE_POW
:
565 arith(fpc
, 0, LG2
, tmp
, MASK_X
,
566 swz(src
[0], X
, X
, X
, X
), none
, none
);
567 arith(fpc
, 0, MUL
, tmp
, MASK_X
, swz(tmp
, X
, X
, X
, X
),
568 swz(src
[1], X
, X
, X
, X
), none
);
569 arith(fpc
, sat
, EX2
, dst
, mask
,
570 swz(tmp
, X
, X
, X
, X
), none
, none
);
572 case TGSI_OPCODE_RCP
:
573 arith(fpc
, sat
, RCP
, dst
, mask
, src
[0], none
, none
);
575 case TGSI_OPCODE_RET
:
578 case TGSI_OPCODE_RFL
:
580 arith(fpc
, 0, DP3
, tmp
, MASK_X
, src
[0], src
[0], none
);
581 arith(fpc
, 0, DP3
, tmp
, MASK_Y
, src
[0], src
[1], none
);
582 arith(fpc
, 0, DIV
, scale(tmp
, 2X
), MASK_Z
,
583 swz(tmp
, Y
, Y
, Y
, Y
), swz(tmp
, X
, X
, X
, X
), none
);
584 arith(fpc
, sat
, MAD
, dst
, mask
,
585 swz(tmp
, Z
, Z
, Z
, Z
), src
[0], neg(src
[1]));
587 case TGSI_OPCODE_RSQ
:
589 arith(fpc
, 0, LG2
, scale(tmp
, INV_2X
), MASK_X
,
590 abs(swz(src
[0], X
, X
, X
, X
)), none
, none
);
591 arith(fpc
, sat
, EX2
, dst
, mask
,
592 neg(swz(tmp
, X
, X
, X
, X
)), none
, none
);
594 case TGSI_OPCODE_SCS
:
596 arith(fpc
, sat
, COS
, dst
, MASK_X
,
597 swz(src
[0], X
, X
, X
, X
), none
, none
);
600 arith(fpc
, sat
, SIN
, dst
, MASK_Y
,
601 swz(src
[0], X
, X
, X
, X
), none
, none
);
604 case TGSI_OPCODE_SEQ
:
605 arith(fpc
, sat
, SEQ
, dst
, mask
, src
[0], src
[1], none
);
607 case TGSI_OPCODE_SFL
:
608 arith(fpc
, sat
, SFL
, dst
, mask
, src
[0], src
[1], none
);
610 case TGSI_OPCODE_SGE
:
611 arith(fpc
, sat
, SGE
, dst
, mask
, src
[0], src
[1], none
);
613 case TGSI_OPCODE_SGT
:
614 arith(fpc
, sat
, SGT
, dst
, mask
, src
[0], src
[1], none
);
616 case TGSI_OPCODE_SIN
:
617 arith(fpc
, sat
, SIN
, dst
, mask
, src
[0], none
, none
);
619 case TGSI_OPCODE_SLE
:
620 arith(fpc
, sat
, SLE
, dst
, mask
, src
[0], src
[1], none
);
622 case TGSI_OPCODE_SLT
:
623 arith(fpc
, sat
, SLT
, dst
, mask
, src
[0], src
[1], none
);
625 case TGSI_OPCODE_SNE
:
626 arith(fpc
, sat
, SNE
, dst
, mask
, src
[0], src
[1], none
);
628 case TGSI_OPCODE_STR
:
629 arith(fpc
, sat
, STR
, dst
, mask
, src
[0], src
[1], none
);
631 case TGSI_OPCODE_SUB
:
632 arith(fpc
, sat
, ADD
, dst
, mask
, src
[0], neg(src
[1]), none
);
634 case TGSI_OPCODE_TEX
:
635 tex(fpc
, sat
, TEX
, unit
, dst
, mask
, src
[0], none
, none
);
637 case TGSI_OPCODE_TXB
:
638 tex(fpc
, sat
, TXB
, unit
, dst
, mask
, src
[0], none
, none
);
640 case TGSI_OPCODE_TXP
:
641 tex(fpc
, sat
, TXP
, unit
, dst
, mask
, src
[0], none
, none
);
643 case TGSI_OPCODE_XPD
:
645 arith(fpc
, 0, MUL
, tmp
, mask
,
646 swz(src
[0], Z
, X
, Y
, Y
), swz(src
[1], Y
, Z
, X
, X
), none
);
647 arith(fpc
, sat
, MAD
, dst
, (mask
& ~MASK_W
),
648 swz(src
[0], Y
, Z
, X
, X
), swz(src
[1], Z
, X
, Y
, Y
),
652 NOUVEAU_ERR("invalid opcode %d\n", finst
->Instruction
.Opcode
);
661 nv40_fragprog_parse_decl_attrib(struct nv40_fpc
*fpc
,
662 const struct tgsi_full_declaration
*fdec
)
666 switch (fdec
->Semantic
.SemanticName
) {
667 case TGSI_SEMANTIC_POSITION
:
668 hw
= NV40_FP_OP_INPUT_SRC_POSITION
;
670 case TGSI_SEMANTIC_COLOR
:
671 if (fdec
->Semantic
.SemanticIndex
== 0) {
672 hw
= NV40_FP_OP_INPUT_SRC_COL0
;
674 if (fdec
->Semantic
.SemanticIndex
== 1) {
675 hw
= NV40_FP_OP_INPUT_SRC_COL1
;
677 NOUVEAU_ERR("bad colour semantic index\n");
681 case TGSI_SEMANTIC_FOG
:
682 hw
= NV40_FP_OP_INPUT_SRC_FOGC
;
684 case TGSI_SEMANTIC_GENERIC
:
685 if (fdec
->Semantic
.SemanticIndex
<= 7) {
686 hw
= NV40_FP_OP_INPUT_SRC_TC(fdec
->Semantic
.
689 NOUVEAU_ERR("bad generic semantic index\n");
694 NOUVEAU_ERR("bad input semantic\n");
698 fpc
->attrib_map
[fdec
->DeclarationRange
.First
] = hw
;
703 nv40_fragprog_parse_decl_output(struct nv40_fpc
*fpc
,
704 const struct tgsi_full_declaration
*fdec
)
706 unsigned idx
= fdec
->DeclarationRange
.First
;
709 switch (fdec
->Semantic
.SemanticName
) {
710 case TGSI_SEMANTIC_POSITION
:
713 case TGSI_SEMANTIC_COLOR
:
714 switch (fdec
->Semantic
.SemanticIndex
) {
715 case 0: hw
= 0; break;
716 case 1: hw
= 2; break;
717 case 2: hw
= 3; break;
718 case 3: hw
= 4; break;
720 NOUVEAU_ERR("bad rcol index\n");
725 NOUVEAU_ERR("bad output semantic\n");
729 fpc
->r_result
[idx
] = nv40_sr(NV40SR_OUTPUT
, hw
);
730 fpc
->r_temps
|= (1 << hw
);
735 nv40_fragprog_prepare(struct nv40_fpc
*fpc
)
737 struct tgsi_parse_context p
;
738 int high_temp
= -1, i
;
740 tgsi_parse_init(&p
, fpc
->fp
->pipe
.tokens
);
741 while (!tgsi_parse_end_of_tokens(&p
)) {
742 const union tgsi_full_token
*tok
= &p
.FullToken
;
744 tgsi_parse_token(&p
);
745 switch(tok
->Token
.Type
) {
746 case TGSI_TOKEN_TYPE_DECLARATION
:
748 const struct tgsi_full_declaration
*fdec
;
749 fdec
= &p
.FullToken
.FullDeclaration
;
750 switch (fdec
->Declaration
.File
) {
751 case TGSI_FILE_INPUT
:
752 if (!nv40_fragprog_parse_decl_attrib(fpc
, fdec
))
755 case TGSI_FILE_OUTPUT
:
756 if (!nv40_fragprog_parse_decl_output(fpc
, fdec
))
759 case TGSI_FILE_TEMPORARY
:
760 if (fdec
->DeclarationRange
.Last
> high_temp
) {
762 fdec
->DeclarationRange
.Last
;
770 case TGSI_TOKEN_TYPE_IMMEDIATE
:
772 struct tgsi_full_immediate
*imm
;
775 imm
= &p
.FullToken
.FullImmediate
;
776 assert(imm
->Immediate
.DataType
== TGSI_IMM_FLOAT32
);
777 assert(fpc
->nr_imm
< MAX_IMM
);
779 vals
[0] = imm
->u
[0].Float
;
780 vals
[1] = imm
->u
[1].Float
;
781 vals
[2] = imm
->u
[2].Float
;
782 vals
[3] = imm
->u
[3].Float
;
783 fpc
->imm
[fpc
->nr_imm
++] = constant(fpc
, -1, vals
);
793 fpc
->r_temp
= CALLOC(high_temp
, sizeof(struct nv40_sreg
));
794 for (i
= 0; i
< high_temp
; i
++)
795 fpc
->r_temp
[i
] = temp(fpc
);
796 fpc
->r_temps_discard
= 0;
809 nv40_fragprog_translate(struct nv40_context
*nv40
,
810 struct nv40_fragment_program
*fp
)
812 struct tgsi_parse_context parse
;
813 struct nv40_fpc
*fpc
= NULL
;
815 fpc
= CALLOC(1, sizeof(struct nv40_fpc
));
821 if (!nv40_fragprog_prepare(fpc
)) {
826 tgsi_parse_init(&parse
, fp
->pipe
.tokens
);
828 while (!tgsi_parse_end_of_tokens(&parse
)) {
829 tgsi_parse_token(&parse
);
831 switch (parse
.FullToken
.Token
.Type
) {
832 case TGSI_TOKEN_TYPE_INSTRUCTION
:
834 const struct tgsi_full_instruction
*finst
;
836 finst
= &parse
.FullToken
.FullInstruction
;
837 if (!nv40_fragprog_parse_instruction(fpc
, finst
))
846 fp
->fp_control
|= fpc
->num_regs
<< NV40TCL_FP_CONTROL_TEMP_COUNT_SHIFT
;
848 /* Terminate final instruction */
849 fp
->insn
[fpc
->inst_offset
] |= 0x00000001;
851 /* Append NOP + END instruction, may or may not be necessary. */
852 fpc
->inst_offset
= fp
->insn_len
;
854 fp
->insn
[fpc
->inst_offset
+ 0] = 0x00000001;
855 fp
->insn
[fpc
->inst_offset
+ 1] = 0x00000000;
856 fp
->insn
[fpc
->inst_offset
+ 2] = 0x00000000;
857 fp
->insn
[fpc
->inst_offset
+ 3] = 0x00000000;
859 fp
->translated
= TRUE
;
861 tgsi_parse_free(&parse
);
868 nv40_fragprog_upload(struct nv40_context
*nv40
,
869 struct nv40_fragment_program
*fp
)
871 struct pipe_screen
*pscreen
= nv40
->pipe
.screen
;
872 const uint32_t le
= 1;
876 map
= pipe_buffer_map(pscreen
, fp
->buffer
, PIPE_BUFFER_USAGE_CPU_WRITE
);
879 for (i
= 0; i
< fp
->insn_len
; i
++) {
880 fflush(stdout
); fflush(stderr
);
881 NOUVEAU_ERR("%d 0x%08x\n", i
, fp
->insn
[i
]);
882 fflush(stdout
); fflush(stderr
);
886 if ((*(const uint8_t *)&le
)) {
887 for (i
= 0; i
< fp
->insn_len
; i
++) {
888 map
[i
] = fp
->insn
[i
];
891 /* Weird swapping for big-endian chips */
892 for (i
= 0; i
< fp
->insn_len
; i
++) {
893 map
[i
] = ((fp
->insn
[i
] & 0xffff) << 16) |
894 ((fp
->insn
[i
] >> 16) & 0xffff);
898 pipe_buffer_unmap(pscreen
, fp
->buffer
);
902 nv40_fragprog_validate(struct nv40_context
*nv40
)
904 struct nv40_fragment_program
*fp
= nv40
->fragprog
;
905 struct pipe_buffer
*constbuf
=
906 nv40
->constbuf
[PIPE_SHADER_FRAGMENT
];
907 struct pipe_screen
*pscreen
= nv40
->pipe
.screen
;
908 struct nouveau_stateobj
*so
;
909 boolean new_consts
= FALSE
;
913 goto update_constants
;
915 nv40
->fallback_swrast
&= ~NV40_NEW_FRAGPROG
;
916 nv40_fragprog_translate(nv40
, fp
);
917 if (!fp
->translated
) {
918 nv40
->fallback_swrast
|= NV40_NEW_FRAGPROG
;
922 fp
->buffer
= pscreen
->buffer_create(pscreen
, 0x100, 0, fp
->insn_len
* 4);
923 nv40_fragprog_upload(nv40
, fp
);
926 so_method(so
, nv40
->screen
->curie
, NV40TCL_FP_ADDRESS
, 1);
927 so_reloc (so
, nouveau_bo(fp
->buffer
), 0, NOUVEAU_BO_VRAM
|
928 NOUVEAU_BO_GART
| NOUVEAU_BO_RD
| NOUVEAU_BO_LOW
|
929 NOUVEAU_BO_OR
, NV40TCL_FP_ADDRESS_DMA0
,
930 NV40TCL_FP_ADDRESS_DMA1
);
931 so_method(so
, nv40
->screen
->curie
, NV40TCL_FP_CONTROL
, 1);
932 so_data (so
, fp
->fp_control
);
940 map
= pipe_buffer_map(pscreen
, constbuf
,
941 PIPE_BUFFER_USAGE_CPU_READ
);
942 for (i
= 0; i
< fp
->nr_consts
; i
++) {
943 struct nv40_fragment_program_data
*fpd
= &fp
->consts
[i
];
944 uint32_t *p
= &fp
->insn
[fpd
->offset
];
945 uint32_t *cb
= (uint32_t *)&map
[fpd
->index
* 4];
947 if (!memcmp(p
, cb
, 4 * sizeof(float)))
949 memcpy(p
, cb
, 4 * sizeof(float));
952 pipe_buffer_unmap(pscreen
, constbuf
);
955 nv40_fragprog_upload(nv40
, fp
);
958 if (new_consts
|| fp
->so
!= nv40
->state
.hw
[NV40_STATE_FRAGPROG
]) {
959 so_ref(fp
->so
, &nv40
->state
.hw
[NV40_STATE_FRAGPROG
]);
967 nv40_fragprog_destroy(struct nv40_context
*nv40
,
968 struct nv40_fragment_program
*fp
)
974 struct nv40_state_entry nv40_state_fragprog
= {
975 .validate
= nv40_fragprog_validate
,
977 .pipe
= NV40_NEW_FRAGPROG
,
978 .hw
= NV40_STATE_FRAGPROG