2 * Copyright 2011 Joakim Sindholt <opensource@zhasha.com>
3 * Copyright 2013 Christoph Bumiller
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * on the rights to use, copy, modify, merge, publish, distribute, sub
9 * license, and/or sell copies of the Software, and to permit persons to whom
10 * the Software is furnished to do so, subject to the following conditions:
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
20 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
21 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
22 * USE OR OTHER DEALINGS IN THE SOFTWARE. */
24 #include "nine_shader.h"
27 #include "nine_debug.h"
28 #include "nine_state.h"
30 #include "util/u_memory.h"
31 #include "util/u_inlines.h"
32 #include "pipe/p_shader_tokens.h"
33 #include "tgsi/tgsi_ureg.h"
34 #include "tgsi/tgsi_dump.h"
36 #define DBG_CHANNEL DBG_SHADER
38 #define DUMP(args...) _nine_debug_printf(DBG_CHANNEL, NULL, args)
41 struct shader_translator
;
43 typedef HRESULT (*translate_instruction_func
)(struct shader_translator
*);
45 static INLINE
const char *d3dsio_to_string(unsigned opcode
);
48 #define NINED3D_SM1_VS 0xfffe
49 #define NINED3D_SM1_PS 0xffff
51 #define NINE_MAX_COND_DEPTH 64
52 #define NINE_MAX_LOOP_DEPTH 64
54 #define NINED3DSP_END 0x0000ffff
56 #define NINED3DSPTYPE_FLOAT4 0
57 #define NINED3DSPTYPE_INT4 1
58 #define NINED3DSPTYPE_BOOL 2
60 #define NINED3DSPR_IMMEDIATE (D3DSPR_PREDICATE + 1)
62 #define NINED3DSP_WRITEMASK_MASK D3DSP_WRITEMASK_ALL
63 #define NINED3DSP_WRITEMASK_SHIFT 16
65 #define NINED3DSHADER_INST_PREDICATED (1 << 28)
67 #define NINED3DSHADER_REL_OP_GT 1
68 #define NINED3DSHADER_REL_OP_EQ 2
69 #define NINED3DSHADER_REL_OP_GE 3
70 #define NINED3DSHADER_REL_OP_LT 4
71 #define NINED3DSHADER_REL_OP_NE 5
72 #define NINED3DSHADER_REL_OP_LE 6
74 #define NINED3DSIO_OPCODE_FLAGS_SHIFT 16
75 #define NINED3DSIO_OPCODE_FLAGS_MASK (0xff << NINED3DSIO_OPCODE_FLAGS_SHIFT)
77 #define NINED3DSI_TEXLD_PROJECT 0x1
78 #define NINED3DSI_TEXLD_BIAS 0x2
80 #define NINED3DSP_WRITEMASK_0 0x1
81 #define NINED3DSP_WRITEMASK_1 0x2
82 #define NINED3DSP_WRITEMASK_2 0x4
83 #define NINED3DSP_WRITEMASK_3 0x8
84 #define NINED3DSP_WRITEMASK_ALL 0xf
86 #define NINED3DSP_NOSWIZZLE ((0 << 0) | (1 << 2) | (2 << 4) | (3 << 6))
88 #define NINE_SWIZZLE4(x,y,z,w) \
89 TGSI_SWIZZLE_##x, TGSI_SWIZZLE_##y, TGSI_SWIZZLE_##z, TGSI_SWIZZLE_##w
91 #define NINED3DSPDM_SATURATE (D3DSPDM_SATURATE >> D3DSP_DSTMOD_SHIFT)
92 #define NINED3DSPDM_PARTIALP (D3DSPDM_PARTIALPRECISION >> D3DSP_DSTMOD_SHIFT)
93 #define NINED3DSPDM_CENTROID (D3DSPDM_MSAMPCENTROID >> D3DSP_DSTMOD_SHIFT)
96 * NEG all, not ps: m3x2, m3x3, m3x4, m4x3, m4x4
97 * BIAS <= PS 1.4 (x-0.5)
98 * BIASNEG <= PS 1.4 (-(x-0.5))
99 * SIGN <= PS 1.4 (2(x-0.5))
100 * SIGNNEG <= PS 1.4 (-2(x-0.5))
101 * COMP <= PS 1.4 (1-x)
103 * X2NEG = PS 1.4 (-2x)
104 * DZ <= PS 1.4, tex{ld,crd} (.xy/.z), z=0 => .11
105 * DW <= PS 1.4, tex{ld,crd} (.xy/.w), w=0 => .11
106 * ABS >= SM 3.0 (abs(x))
107 * ABSNEG >= SM 3.0 (-abs(x))
108 * NOT >= SM 2.0 pedication only
110 #define NINED3DSPSM_NONE (D3DSPSM_NONE >> D3DSP_SRCMOD_SHIFT)
111 #define NINED3DSPSM_NEG (D3DSPSM_NEG >> D3DSP_SRCMOD_SHIFT)
112 #define NINED3DSPSM_BIAS (D3DSPSM_BIAS >> D3DSP_SRCMOD_SHIFT)
113 #define NINED3DSPSM_BIASNEG (D3DSPSM_BIASNEG >> D3DSP_SRCMOD_SHIFT)
114 #define NINED3DSPSM_SIGN (D3DSPSM_SIGN >> D3DSP_SRCMOD_SHIFT)
115 #define NINED3DSPSM_SIGNNEG (D3DSPSM_SIGNNEG >> D3DSP_SRCMOD_SHIFT)
116 #define NINED3DSPSM_COMP (D3DSPSM_COMP >> D3DSP_SRCMOD_SHIFT)
117 #define NINED3DSPSM_X2 (D3DSPSM_X2 >> D3DSP_SRCMOD_SHIFT)
118 #define NINED3DSPSM_X2NEG (D3DSPSM_X2NEG >> D3DSP_SRCMOD_SHIFT)
119 #define NINED3DSPSM_DZ (D3DSPSM_DZ >> D3DSP_SRCMOD_SHIFT)
120 #define NINED3DSPSM_DW (D3DSPSM_DW >> D3DSP_SRCMOD_SHIFT)
121 #define NINED3DSPSM_ABS (D3DSPSM_ABS >> D3DSP_SRCMOD_SHIFT)
122 #define NINED3DSPSM_ABSNEG (D3DSPSM_ABSNEG >> D3DSP_SRCMOD_SHIFT)
123 #define NINED3DSPSM_NOT (D3DSPSM_NOT >> D3DSP_SRCMOD_SHIFT)
125 static const char *sm1_mod_str
[] =
127 [NINED3DSPSM_NONE
] = "",
128 [NINED3DSPSM_NEG
] = "-",
129 [NINED3DSPSM_BIAS
] = "bias",
130 [NINED3DSPSM_BIASNEG
] = "biasneg",
131 [NINED3DSPSM_SIGN
] = "sign",
132 [NINED3DSPSM_SIGNNEG
] = "signneg",
133 [NINED3DSPSM_COMP
] = "comp",
134 [NINED3DSPSM_X2
] = "x2",
135 [NINED3DSPSM_X2NEG
] = "x2neg",
136 [NINED3DSPSM_DZ
] = "dz",
137 [NINED3DSPSM_DW
] = "dw",
138 [NINED3DSPSM_ABS
] = "abs",
139 [NINED3DSPSM_ABSNEG
] = "-abs",
140 [NINED3DSPSM_NOT
] = "not"
144 sm1_dump_writemask(BYTE mask
)
146 if (mask
& 1) DUMP("x"); else DUMP("_");
147 if (mask
& 2) DUMP("y"); else DUMP("_");
148 if (mask
& 4) DUMP("z"); else DUMP("_");
149 if (mask
& 8) DUMP("w"); else DUMP("_");
153 sm1_dump_swizzle(BYTE s
)
155 char c
[4] = { 'x', 'y', 'z', 'w' };
157 c
[(s
>> 0) & 3], c
[(s
>> 2) & 3], c
[(s
>> 4) & 3], c
[(s
>> 6) & 3]);
160 static const char sm1_file_char
[] =
163 [D3DSPR_INPUT
] = 'v',
164 [D3DSPR_CONST
] = 'c',
166 [D3DSPR_RASTOUT
] = 'R',
167 [D3DSPR_ATTROUT
] = 'D',
168 [D3DSPR_OUTPUT
] = 'o',
169 [D3DSPR_CONSTINT
] = 'I',
170 [D3DSPR_COLOROUT
] = 'C',
171 [D3DSPR_DEPTHOUT
] = 'D',
172 [D3DSPR_SAMPLER
] = 's',
173 [D3DSPR_CONST2
] = 'c',
174 [D3DSPR_CONST3
] = 'c',
175 [D3DSPR_CONST4
] = 'c',
176 [D3DSPR_CONSTBOOL
] = 'B',
178 [D3DSPR_TEMPFLOAT16
] = 'h',
179 [D3DSPR_MISCTYPE
] = 'M',
180 [D3DSPR_LABEL
] = 'X',
181 [D3DSPR_PREDICATE
] = 'p'
185 sm1_dump_reg(BYTE file
, INT index
)
191 case D3DSPR_COLOROUT
:
194 case D3DSPR_DEPTHOUT
:
198 DUMP("oRast%i", index
);
200 case D3DSPR_CONSTINT
:
201 DUMP("iconst[%i]", index
);
203 case D3DSPR_CONSTBOOL
:
204 DUMP("bconst[%i]", index
);
207 DUMP("%c%i", sm1_file_char
[file
], index
);
215 struct sm1_src_param
*rel
;
228 sm1_parse_immediate(struct shader_translator
*, struct sm1_src_param
*);
233 struct sm1_src_param
*rel
;
237 int8_t shift
; /* sint4 */
242 assert_replicate_swizzle(const struct ureg_src
*reg
)
244 assert(reg
->SwizzleY
== reg
->SwizzleX
&&
245 reg
->SwizzleZ
== reg
->SwizzleX
&&
246 reg
->SwizzleW
== reg
->SwizzleX
);
250 sm1_dump_immediate(const struct sm1_src_param
*param
)
252 switch (param
->type
) {
253 case NINED3DSPTYPE_FLOAT4
:
254 DUMP("{ %f %f %f %f }",
255 param
->imm
.f
[0], param
->imm
.f
[1],
256 param
->imm
.f
[2], param
->imm
.f
[3]);
258 case NINED3DSPTYPE_INT4
:
259 DUMP("{ %i %i %i %i }",
260 param
->imm
.i
[0], param
->imm
.i
[1],
261 param
->imm
.i
[2], param
->imm
.i
[3]);
263 case NINED3DSPTYPE_BOOL
:
264 DUMP("%s", param
->imm
.b
? "TRUE" : "FALSE");
273 sm1_dump_src_param(const struct sm1_src_param
*param
)
275 if (param
->file
== NINED3DSPR_IMMEDIATE
) {
276 assert(!param
->mod
&&
278 param
->swizzle
== NINED3DSP_NOSWIZZLE
);
279 sm1_dump_immediate(param
);
284 DUMP("%s(", sm1_mod_str
[param
->mod
]);
286 DUMP("%c[", sm1_file_char
[param
->file
]);
287 sm1_dump_src_param(param
->rel
);
288 DUMP("+%i]", param
->idx
);
290 sm1_dump_reg(param
->file
, param
->idx
);
294 if (param
->swizzle
!= NINED3DSP_NOSWIZZLE
) {
296 sm1_dump_swizzle(param
->swizzle
);
301 sm1_dump_dst_param(const struct sm1_dst_param
*param
)
303 if (param
->mod
& NINED3DSPDM_SATURATE
)
305 if (param
->mod
& NINED3DSPDM_PARTIALP
)
307 if (param
->mod
& NINED3DSPDM_CENTROID
)
309 if (param
->shift
< 0)
310 DUMP("/%u ", 1 << -param
->shift
);
311 if (param
->shift
> 0)
312 DUMP("*%u ", 1 << param
->shift
);
315 DUMP("%c[", sm1_file_char
[param
->file
]);
316 sm1_dump_src_param(param
->rel
);
317 DUMP("+%i]", param
->idx
);
319 sm1_dump_reg(param
->file
, param
->idx
);
321 if (param
->mask
!= NINED3DSP_WRITEMASK_ALL
) {
323 sm1_dump_writemask(param
->mask
);
329 struct sm1_dst_param reg
;
337 /* NOTE: 0 is a valid TGSI opcode, but if handler is set, this parameter
338 * should be ignored completely */
340 unsigned opcode
; /* TGSI_OPCODE_x */
342 /* versions are still set even handler is set */
346 } vert_version
, frag_version
;
348 /* number of regs parsed outside of special handler */
352 /* some instructions don't map perfectly, so use a special handler */
353 translate_instruction_func handler
;
356 struct sm1_instruction
358 D3DSHADER_INSTRUCTION_OPCODE_TYPE opcode
;
364 struct sm1_src_param src
[4];
365 struct sm1_src_param src_rel
[4];
366 struct sm1_src_param pred
;
367 struct sm1_src_param dst_rel
[1];
368 struct sm1_dst_param dst
[1];
370 struct sm1_op_info
*info
;
374 sm1_dump_instruction(struct sm1_instruction
*insn
, unsigned indent
)
378 /* no info stored for these: */
379 if (insn
->opcode
== D3DSIO_DCL
)
381 for (i
= 0; i
< indent
; ++i
)
384 if (insn
->predicated
) {
386 sm1_dump_src_param(&insn
->pred
);
389 DUMP("%s", d3dsio_to_string(insn
->opcode
));
391 switch (insn
->opcode
) {
393 DUMP(insn
->flags
== NINED3DSI_TEXLD_PROJECT
? "p" : "b");
396 DUMP("_%x", insn
->flags
);
404 for (i
= 0; i
< insn
->ndst
&& i
< Elements(insn
->dst
); ++i
) {
405 sm1_dump_dst_param(&insn
->dst
[i
]);
409 for (i
= 0; i
< insn
->nsrc
&& i
< Elements(insn
->src
); ++i
) {
410 sm1_dump_src_param(&insn
->src
[i
]);
413 if (insn
->opcode
== D3DSIO_DEF
||
414 insn
->opcode
== D3DSIO_DEFI
||
415 insn
->opcode
== D3DSIO_DEFB
)
416 sm1_dump_immediate(&insn
->src
[0]);
421 struct sm1_local_const
432 struct shader_translator
434 const DWORD
*byte_code
;
436 const DWORD
*parse_next
;
438 struct ureg_program
*ureg
;
445 unsigned processor
; /* TGSI_PROCESSOR_VERTEX/FRAMGENT */
447 boolean native_integers
;
448 boolean inline_subroutines
;
450 boolean want_texcoord
;
452 unsigned texcoord_sn
;
454 struct sm1_instruction insn
; /* current instruction */
458 struct ureg_dst oPos
;
459 struct ureg_dst oFog
;
460 struct ureg_dst oPts
;
461 struct ureg_dst oCol
[4];
462 struct ureg_dst o
[PIPE_MAX_SHADER_OUTPUTS
];
463 struct ureg_dst oDepth
;
464 struct ureg_src v
[PIPE_MAX_SHADER_INPUTS
];
465 struct ureg_src vPos
;
466 struct ureg_src vFace
;
469 struct ureg_dst address
;
471 struct ureg_dst tS
[8]; /* texture stage registers */
472 struct ureg_dst tdst
; /* scratch dst if we need extra modifiers */
473 struct ureg_dst t
[5]; /* scratch TEMPs */
474 struct ureg_src vC
[2]; /* PS color in */
475 struct ureg_src vT
[8]; /* PS texcoord in */
476 struct ureg_dst rL
[NINE_MAX_LOOP_DEPTH
]; /* loop ctr */
478 unsigned num_temp
; /* Elements(regs.r) */
479 unsigned num_scratch
;
481 unsigned loop_depth_max
;
483 unsigned loop_labels
[NINE_MAX_LOOP_DEPTH
];
484 unsigned cond_labels
[NINE_MAX_COND_DEPTH
];
485 boolean loop_or_rep
[NINE_MAX_LOOP_DEPTH
]; /* true: loop, false: rep */
487 unsigned *inst_labels
; /* LABEL op */
488 unsigned num_inst_labels
;
490 unsigned sampler_targets
[NINE_MAX_SAMPLERS
]; /* TGSI_TEXTURE_x */
492 struct sm1_local_const
*lconstf
;
493 unsigned num_lconstf
;
494 struct sm1_local_const lconsti
[NINE_MAX_CONST_I
];
495 struct sm1_local_const lconstb
[NINE_MAX_CONST_B
];
497 boolean indirect_const_access
;
499 struct nine_shader_info
*info
;
501 int16_t op_info_map
[D3DSIO_BREAKP
+ 1];
504 #define IS_VS (tx->processor == TGSI_PROCESSOR_VERTEX)
505 #define IS_PS (tx->processor == TGSI_PROCESSOR_FRAGMENT)
508 sm1_read_semantic(struct shader_translator
*, struct sm1_semantic
*);
511 sm1_instruction_check(const struct sm1_instruction
*insn
)
513 if (insn
->opcode
== D3DSIO_CRS
)
515 if (insn
->dst
[0].mask
& NINED3DSP_WRITEMASK_3
)
523 tx_lconstf(struct shader_translator
*tx
, struct ureg_src
*src
, INT index
)
526 assert(index
>= 0 && index
< (NINE_MAX_CONST_F
* 2));
527 for (i
= 0; i
< tx
->num_lconstf
; ++i
) {
528 if (tx
->lconstf
[i
].idx
== index
) {
529 *src
= tx
->lconstf
[i
].reg
;
536 tx_lconsti(struct shader_translator
*tx
, struct ureg_src
*src
, INT index
)
538 assert(index
>= 0 && index
< NINE_MAX_CONST_I
);
539 if (tx
->lconsti
[index
].idx
== index
)
540 *src
= tx
->lconsti
[index
].reg
;
541 return tx
->lconsti
[index
].idx
== index
;
544 tx_lconstb(struct shader_translator
*tx
, struct ureg_src
*src
, INT index
)
546 assert(index
>= 0 && index
< NINE_MAX_CONST_B
);
547 if (tx
->lconstb
[index
].idx
== index
)
548 *src
= tx
->lconstb
[index
].reg
;
549 return tx
->lconstb
[index
].idx
== index
;
553 tx_set_lconstf(struct shader_translator
*tx
, INT index
, float f
[4])
557 /* Anno1404 sets out of range constants. */
558 assert(index
>= 0 && index
< (NINE_MAX_CONST_F
* 2));
559 if (index
>= NINE_MAX_CONST_F
)
560 WARN("lconstf index %i too high, indirect access won't work\n", index
);
562 for (n
= 0; n
< tx
->num_lconstf
; ++n
)
563 if (tx
->lconstf
[n
].idx
== index
)
565 if (n
== tx
->num_lconstf
) {
567 tx
->lconstf
= REALLOC(tx
->lconstf
,
568 (n
+ 0) * sizeof(tx
->lconstf
[0]),
569 (n
+ 8) * sizeof(tx
->lconstf
[0]));
574 tx
->lconstf
[n
].idx
= index
;
575 tx
->lconstf
[n
].reg
= ureg_imm4f(tx
->ureg
, f
[0], f
[1], f
[2], f
[3]);
577 memcpy(tx
->lconstf
[n
].imm
.f
, f
, sizeof(tx
->lconstf
[n
].imm
.f
));
580 tx_set_lconsti(struct shader_translator
*tx
, INT index
, int i
[4])
582 assert(index
>= 0 && index
< NINE_MAX_CONST_I
);
583 tx
->lconsti
[index
].idx
= index
;
584 tx
->lconsti
[index
].reg
= tx
->native_integers
?
585 ureg_imm4i(tx
->ureg
, i
[0], i
[1], i
[2], i
[3]) :
586 ureg_imm4f(tx
->ureg
, i
[0], i
[1], i
[2], i
[3]);
589 tx_set_lconstb(struct shader_translator
*tx
, INT index
, BOOL b
)
591 assert(index
>= 0 && index
< NINE_MAX_CONST_B
);
592 tx
->lconstb
[index
].idx
= index
;
593 tx
->lconstb
[index
].reg
= tx
->native_integers
?
594 ureg_imm1u(tx
->ureg
, b
? 0xffffffff : 0) :
595 ureg_imm1f(tx
->ureg
, b
? 1.0f
: 0.0f
);
598 static INLINE
struct ureg_dst
599 tx_scratch(struct shader_translator
*tx
)
601 assert(tx
->num_scratch
< Elements(tx
->regs
.t
));
602 if (ureg_dst_is_undef(tx
->regs
.t
[tx
->num_scratch
]))
603 tx
->regs
.t
[tx
->num_scratch
] = ureg_DECL_local_temporary(tx
->ureg
);
604 return tx
->regs
.t
[tx
->num_scratch
++];
607 static INLINE
struct ureg_dst
608 tx_scratch_scalar(struct shader_translator
*tx
)
610 return ureg_writemask(tx_scratch(tx
), TGSI_WRITEMASK_X
);
613 static INLINE
struct ureg_src
614 tx_src_scalar(struct ureg_dst dst
)
616 struct ureg_src src
= ureg_src(dst
);
617 int c
= ffs(dst
.WriteMask
) - 1;
618 if (dst
.WriteMask
== (1 << c
))
619 src
= ureg_scalar(src
, c
);
623 /* Need to declare all constants if indirect addressing is used,
624 * otherwise we could scan the shader to determine the maximum.
625 * TODO: It doesn't really matter for nv50 so I won't do the scan,
626 * but radeon drivers might care, if they don't infer it from TGSI.
629 tx_decl_constants(struct shader_translator
*tx
)
633 for (i
= 0; i
< NINE_MAX_CONST_F
; ++i
)
634 ureg_DECL_constant(tx
->ureg
, n
++);
635 for (i
= 0; i
< NINE_MAX_CONST_I
; ++i
)
636 ureg_DECL_constant(tx
->ureg
, n
++);
637 for (i
= 0; i
< (NINE_MAX_CONST_B
/ 4); ++i
)
638 ureg_DECL_constant(tx
->ureg
, n
++);
642 tx_temp_alloc(struct shader_translator
*tx
, INT idx
)
645 if (idx
>= tx
->num_temp
) {
646 unsigned k
= tx
->num_temp
;
647 unsigned n
= idx
+ 1;
648 tx
->regs
.r
= REALLOC(tx
->regs
.r
,
649 k
* sizeof(tx
->regs
.r
[0]),
650 n
* sizeof(tx
->regs
.r
[0]));
652 tx
->regs
.r
[k
] = ureg_dst_undef();
655 if (ureg_dst_is_undef(tx
->regs
.r
[idx
]))
656 tx
->regs
.r
[idx
] = ureg_DECL_temporary(tx
->ureg
);
660 tx_addr_alloc(struct shader_translator
*tx
, INT idx
)
663 if (ureg_dst_is_undef(tx
->regs
.address
))
664 tx
->regs
.address
= ureg_DECL_address(tx
->ureg
);
665 if (ureg_dst_is_undef(tx
->regs
.a0
))
666 tx
->regs
.a0
= ureg_DECL_temporary(tx
->ureg
);
670 tx_pred_alloc(struct shader_translator
*tx
, INT idx
)
673 if (ureg_dst_is_undef(tx
->regs
.p
))
674 tx
->regs
.p
= ureg_DECL_predicate(tx
->ureg
);
678 tx_texcoord_alloc(struct shader_translator
*tx
, INT idx
)
681 assert(idx
>= 0 && idx
< Elements(tx
->regs
.vT
));
682 if (ureg_src_is_undef(tx
->regs
.vT
[idx
]))
683 tx
->regs
.vT
[idx
] = ureg_DECL_fs_input(tx
->ureg
, tx
->texcoord_sn
, idx
,
684 TGSI_INTERPOLATE_PERSPECTIVE
);
687 static INLINE
unsigned *
688 tx_bgnloop(struct shader_translator
*tx
)
691 if (tx
->loop_depth_max
< tx
->loop_depth
)
692 tx
->loop_depth_max
= tx
->loop_depth
;
693 assert(tx
->loop_depth
< NINE_MAX_LOOP_DEPTH
);
694 return &tx
->loop_labels
[tx
->loop_depth
- 1];
697 static INLINE
unsigned *
698 tx_endloop(struct shader_translator
*tx
)
700 assert(tx
->loop_depth
);
702 ureg_fixup_label(tx
->ureg
, tx
->loop_labels
[tx
->loop_depth
],
703 ureg_get_instruction_number(tx
->ureg
));
704 return &tx
->loop_labels
[tx
->loop_depth
];
707 static struct ureg_dst
708 tx_get_loopctr(struct shader_translator
*tx
, boolean loop_or_rep
)
710 const unsigned l
= tx
->loop_depth
- 1;
714 DBG("loop counter requested outside of loop\n");
715 return ureg_dst_undef();
718 if (ureg_dst_is_undef(tx
->regs
.rL
[l
])) {
719 /* loop or rep ctr creation */
720 tx
->regs
.rL
[l
] = ureg_DECL_local_temporary(tx
->ureg
);
721 tx
->loop_or_rep
[l
] = loop_or_rep
;
723 /* loop - rep - endloop - endrep not allowed */
724 assert(tx
->loop_or_rep
[l
] == loop_or_rep
);
726 return tx
->regs
.rL
[l
];
729 static struct ureg_src
730 tx_get_loopal(struct shader_translator
*tx
)
732 int loop_level
= tx
->loop_depth
- 1;
734 while (loop_level
>= 0) {
735 /* handle loop - rep - endrep - endloop case */
736 if (tx
->loop_or_rep
[loop_level
])
737 /* the value is in the loop counter y component (nine implementation) */
738 return ureg_scalar(ureg_src(tx
->regs
.rL
[loop_level
]), TGSI_SWIZZLE_Y
);
742 DBG("aL counter requested outside of loop\n");
743 return ureg_src_undef();
746 static INLINE
unsigned *
747 tx_cond(struct shader_translator
*tx
)
749 assert(tx
->cond_depth
<= NINE_MAX_COND_DEPTH
);
751 return &tx
->cond_labels
[tx
->cond_depth
- 1];
754 static INLINE
unsigned *
755 tx_elsecond(struct shader_translator
*tx
)
757 assert(tx
->cond_depth
);
758 return &tx
->cond_labels
[tx
->cond_depth
- 1];
762 tx_endcond(struct shader_translator
*tx
)
764 assert(tx
->cond_depth
);
766 ureg_fixup_label(tx
->ureg
, tx
->cond_labels
[tx
->cond_depth
],
767 ureg_get_instruction_number(tx
->ureg
));
770 static INLINE
struct ureg_dst
771 nine_ureg_dst_register(unsigned file
, int index
)
773 return ureg_dst(ureg_src_register(file
, index
));
776 static struct ureg_src
777 tx_src_param(struct shader_translator
*tx
, const struct sm1_src_param
*param
)
779 struct ureg_program
*ureg
= tx
->ureg
;
787 tx_temp_alloc(tx
, param
->idx
);
788 src
= ureg_src(tx
->regs
.r
[param
->idx
]);
790 /* case D3DSPR_TEXTURE: == D3DSPR_ADDR */
794 assert(param
->idx
== 0);
795 /* the address register (vs only) must be
796 * assigned before use */
797 assert(!ureg_dst_is_undef(tx
->regs
.a0
));
798 ureg_ARR(ureg
, tx
->regs
.address
, ureg_src(tx
->regs
.a0
));
799 src
= ureg_src(tx
->regs
.address
);
801 if (tx
->version
.major
< 2 && tx
->version
.minor
< 4) {
802 /* no subroutines, so should be defined */
803 src
= ureg_src(tx
->regs
.tS
[param
->idx
]);
805 tx_texcoord_alloc(tx
, param
->idx
);
806 src
= tx
->regs
.vT
[param
->idx
];
812 src
= ureg_src_register(TGSI_FILE_INPUT
, param
->idx
);
814 if (tx
->version
.major
< 3) {
816 src
= ureg_DECL_fs_input(tx
->ureg
, TGSI_SEMANTIC_COLOR
,
818 TGSI_INTERPOLATE_PERSPECTIVE
);
820 assert(!param
->rel
); /* TODO */
821 assert(param
->idx
< Elements(tx
->regs
.v
));
822 src
= tx
->regs
.v
[param
->idx
];
826 case D3DSPR_PREDICATE
:
828 tx_pred_alloc(tx
, param
->idx
);
829 src
= ureg_src(tx
->regs
.p
);
832 assert(param
->mod
== NINED3DSPSM_NONE
);
833 assert(param
->swizzle
== NINED3DSP_NOSWIZZLE
);
835 src
= ureg_src_register(TGSI_FILE_SAMPLER
, param
->idx
);
839 tx
->indirect_const_access
= TRUE
;
840 if (param
->rel
|| !tx_lconstf(tx
, &src
, param
->idx
)) {
842 nine_info_mark_const_f_used(tx
->info
, param
->idx
);
843 src
= ureg_src_register(TGSI_FILE_CONSTANT
, param
->idx
);
845 if (!IS_VS
&& tx
->version
.major
< 2) {
846 /* ps 1.X clamps constants */
847 tmp
= tx_scratch(tx
);
848 ureg_MIN(ureg
, tmp
, src
, ureg_imm1f(ureg
, 1.0f
));
849 ureg_MAX(ureg
, tmp
, ureg_src(tmp
), ureg_imm1f(ureg
, -1.0f
));
856 DBG("CONST2/3/4 should have been collapsed into D3DSPR_CONST !\n");
857 assert(!"CONST2/3/4");
858 src
= ureg_imm1f(ureg
, 0.0f
);
860 case D3DSPR_CONSTINT
:
861 if (param
->rel
|| !tx_lconsti(tx
, &src
, param
->idx
)) {
863 nine_info_mark_const_i_used(tx
->info
, param
->idx
);
864 src
= ureg_src_register(TGSI_FILE_CONSTANT
,
865 tx
->info
->const_i_base
+ param
->idx
);
868 case D3DSPR_CONSTBOOL
:
869 if (param
->rel
|| !tx_lconstb(tx
, &src
, param
->idx
)) {
870 char r
= param
->idx
/ 4;
871 char s
= param
->idx
& 3;
873 nine_info_mark_const_b_used(tx
->info
, param
->idx
);
874 src
= ureg_src_register(TGSI_FILE_CONSTANT
,
875 tx
->info
->const_b_base
+ r
);
876 src
= ureg_swizzle(src
, s
, s
, s
, s
);
880 if (ureg_dst_is_undef(tx
->regs
.address
))
881 tx
->regs
.address
= ureg_DECL_address(ureg
);
882 if (!tx
->native_integers
)
883 ureg_ARR(ureg
, tx
->regs
.address
, tx_get_loopal(tx
));
885 ureg_UARL(ureg
, tx
->regs
.address
, tx_get_loopal(tx
));
886 src
= ureg_src(tx
->regs
.address
);
888 case D3DSPR_MISCTYPE
:
889 switch (param
->idx
) {
890 case D3DSMO_POSITION
:
891 if (ureg_src_is_undef(tx
->regs
.vPos
))
892 tx
->regs
.vPos
= ureg_DECL_fs_input(ureg
,
893 TGSI_SEMANTIC_POSITION
, 0,
894 TGSI_INTERPOLATE_LINEAR
);
895 if (tx
->shift_wpos
) {
896 /* TODO: do this only once */
897 struct ureg_dst wpos
= tx_scratch(tx
);
898 ureg_SUB(ureg
, wpos
, tx
->regs
.vPos
,
899 ureg_imm4f(ureg
, 0.5f
, 0.5f
, 0.0f
, 0.0f
));
900 src
= ureg_src(wpos
);
906 if (ureg_src_is_undef(tx
->regs
.vFace
)) {
907 tx
->regs
.vFace
= ureg_DECL_fs_input(ureg
,
908 TGSI_SEMANTIC_FACE
, 0,
909 TGSI_INTERPOLATE_CONSTANT
);
910 tx
->regs
.vFace
= ureg_scalar(tx
->regs
.vFace
, TGSI_SWIZZLE_X
);
912 src
= tx
->regs
.vFace
;
915 assert(!"invalid src D3DSMO");
920 case D3DSPR_TEMPFLOAT16
:
923 assert(!"invalid src D3DSPR");
926 src
= ureg_src_indirect(src
, tx_src_param(tx
, param
->rel
));
928 if (param
->swizzle
!= NINED3DSP_NOSWIZZLE
)
929 src
= ureg_swizzle(src
,
930 (param
->swizzle
>> 0) & 0x3,
931 (param
->swizzle
>> 2) & 0x3,
932 (param
->swizzle
>> 4) & 0x3,
933 (param
->swizzle
>> 6) & 0x3);
935 switch (param
->mod
) {
936 case NINED3DSPSM_ABS
:
939 case NINED3DSPSM_ABSNEG
:
940 src
= ureg_negate(ureg_abs(src
));
942 case NINED3DSPSM_NEG
:
943 src
= ureg_negate(src
);
945 case NINED3DSPSM_BIAS
:
946 tmp
= tx_scratch(tx
);
947 ureg_SUB(ureg
, tmp
, src
, ureg_imm1f(ureg
, 0.5f
));
950 case NINED3DSPSM_BIASNEG
:
951 tmp
= tx_scratch(tx
);
952 ureg_SUB(ureg
, tmp
, ureg_imm1f(ureg
, 0.5f
), src
);
955 case NINED3DSPSM_NOT
:
956 if (tx
->native_integers
) {
957 tmp
= tx_scratch(tx
);
958 ureg_NOT(ureg
, tmp
, src
);
963 case NINED3DSPSM_COMP
:
964 tmp
= tx_scratch(tx
);
965 ureg_SUB(ureg
, tmp
, ureg_imm1f(ureg
, 1.0f
), src
);
970 /* handled in instruction */
972 case NINED3DSPSM_SIGN
:
973 tmp
= tx_scratch(tx
);
974 ureg_MAD(ureg
, tmp
, src
, ureg_imm1f(ureg
, 2.0f
), ureg_imm1f(ureg
, -1.0f
));
977 case NINED3DSPSM_SIGNNEG
:
978 tmp
= tx_scratch(tx
);
979 ureg_MAD(ureg
, tmp
, src
, ureg_imm1f(ureg
, -2.0f
), ureg_imm1f(ureg
, 1.0f
));
983 tmp
= tx_scratch(tx
);
984 ureg_ADD(ureg
, tmp
, src
, src
);
987 case NINED3DSPSM_X2NEG
:
988 tmp
= tx_scratch(tx
);
989 ureg_ADD(ureg
, tmp
, src
, src
);
990 src
= ureg_negate(ureg_src(tmp
));
993 assert(param
->mod
== NINED3DSPSM_NONE
);
1000 static struct ureg_dst
1001 _tx_dst_param(struct shader_translator
*tx
, const struct sm1_dst_param
*param
)
1003 struct ureg_dst dst
;
1005 switch (param
->file
)
1008 assert(!param
->rel
);
1009 tx_temp_alloc(tx
, param
->idx
);
1010 dst
= tx
->regs
.r
[param
->idx
];
1012 /* case D3DSPR_TEXTURE: == D3DSPR_ADDR */
1014 assert(!param
->rel
);
1015 if (tx
->version
.major
< 2 && !IS_VS
) {
1016 if (ureg_dst_is_undef(tx
->regs
.tS
[param
->idx
]))
1017 tx
->regs
.tS
[param
->idx
] = ureg_DECL_temporary(tx
->ureg
);
1018 dst
= tx
->regs
.tS
[param
->idx
];
1020 if (!IS_VS
&& tx
->insn
.opcode
== D3DSIO_TEXKILL
) { /* maybe others, too */
1021 tx_texcoord_alloc(tx
, param
->idx
);
1022 dst
= ureg_dst(tx
->regs
.vT
[param
->idx
]);
1024 tx_addr_alloc(tx
, param
->idx
);
1028 case D3DSPR_RASTOUT
:
1029 assert(!param
->rel
);
1030 switch (param
->idx
) {
1032 if (ureg_dst_is_undef(tx
->regs
.oPos
))
1034 ureg_DECL_output(tx
->ureg
, TGSI_SEMANTIC_POSITION
, 0);
1035 dst
= tx
->regs
.oPos
;
1038 if (ureg_dst_is_undef(tx
->regs
.oFog
))
1040 ureg_saturate(ureg_DECL_output(tx
->ureg
, TGSI_SEMANTIC_FOG
, 0));
1041 dst
= tx
->regs
.oFog
;
1044 if (ureg_dst_is_undef(tx
->regs
.oPts
))
1046 ureg_saturate(ureg_DECL_output(tx
->ureg
, TGSI_SEMANTIC_PSIZE
, 0));
1047 dst
= tx
->regs
.oPts
;
1054 /* case D3DSPR_TEXCRDOUT: == D3DSPR_OUTPUT */
1056 if (tx
->version
.major
< 3) {
1057 assert(!param
->rel
);
1058 dst
= ureg_DECL_output(tx
->ureg
, tx
->texcoord_sn
, param
->idx
);
1060 assert(!param
->rel
); /* TODO */
1061 assert(param
->idx
< Elements(tx
->regs
.o
));
1062 dst
= tx
->regs
.o
[param
->idx
];
1065 case D3DSPR_ATTROUT
: /* VS */
1066 case D3DSPR_COLOROUT
: /* PS */
1067 assert(param
->idx
>= 0 && param
->idx
< 4);
1068 assert(!param
->rel
);
1069 tx
->info
->rt_mask
|= 1 << param
->idx
;
1070 if (ureg_dst_is_undef(tx
->regs
.oCol
[param
->idx
]))
1071 tx
->regs
.oCol
[param
->idx
] =
1072 ureg_DECL_output(tx
->ureg
, TGSI_SEMANTIC_COLOR
, param
->idx
);
1073 dst
= tx
->regs
.oCol
[param
->idx
];
1074 if (IS_VS
&& tx
->version
.major
< 3)
1075 dst
= ureg_saturate(dst
);
1077 case D3DSPR_DEPTHOUT
:
1078 assert(!param
->rel
);
1079 if (ureg_dst_is_undef(tx
->regs
.oDepth
))
1081 ureg_DECL_output_masked(tx
->ureg
, TGSI_SEMANTIC_POSITION
, 0,
1083 dst
= tx
->regs
.oDepth
; /* XXX: must write .z component */
1085 case D3DSPR_PREDICATE
:
1086 assert(!param
->rel
);
1087 tx_pred_alloc(tx
, param
->idx
);
1090 case D3DSPR_TEMPFLOAT16
:
1091 DBG("unhandled D3DSPR: %u\n", param
->file
);
1094 assert(!"invalid dst D3DSPR");
1098 dst
= ureg_dst_indirect(dst
, tx_src_param(tx
, param
->rel
));
1100 if (param
->mask
!= NINED3DSP_WRITEMASK_ALL
)
1101 dst
= ureg_writemask(dst
, param
->mask
);
1102 if (param
->mod
& NINED3DSPDM_SATURATE
)
1103 dst
= ureg_saturate(dst
);
1108 static struct ureg_dst
1109 tx_dst_param(struct shader_translator
*tx
, const struct sm1_dst_param
*param
)
1112 tx
->regs
.tdst
= ureg_writemask(tx_scratch(tx
), param
->mask
);
1113 return tx
->regs
.tdst
;
1115 return _tx_dst_param(tx
, param
);
1119 tx_apply_dst0_modifiers(struct shader_translator
*tx
)
1121 struct ureg_dst rdst
;
1124 if (!tx
->insn
.ndst
|| !tx
->insn
.dst
[0].shift
|| tx
->insn
.opcode
== D3DSIO_TEXKILL
)
1126 rdst
= _tx_dst_param(tx
, &tx
->insn
.dst
[0]);
1128 assert(rdst
.File
!= TGSI_FILE_ADDRESS
); /* this probably isn't possible */
1130 if (tx
->insn
.dst
[0].shift
< 0)
1131 f
= 1.0f
/ (1 << -tx
->insn
.dst
[0].shift
);
1133 f
= 1 << tx
->insn
.dst
[0].shift
;
1135 ureg_MUL(tx
->ureg
, rdst
, ureg_src(tx
->regs
.tdst
), ureg_imm1f(tx
->ureg
, f
));
1138 static struct ureg_src
1139 tx_dst_param_as_src(struct shader_translator
*tx
, const struct sm1_dst_param
*param
)
1141 struct ureg_src src
;
1143 assert(!param
->shift
);
1144 assert(!(param
->mod
& NINED3DSPDM_SATURATE
));
1146 switch (param
->file
) {
1149 src
= ureg_src_register(TGSI_FILE_INPUT
, param
->idx
);
1151 assert(!param
->rel
);
1152 assert(param
->idx
< Elements(tx
->regs
.v
));
1153 src
= tx
->regs
.v
[param
->idx
];
1157 src
= ureg_src(tx_dst_param(tx
, param
));
1161 src
= ureg_src_indirect(src
, tx_src_param(tx
, param
->rel
));
1164 WARN("mask is 0, using identity swizzle\n");
1166 if (param
->mask
&& param
->mask
!= NINED3DSP_WRITEMASK_ALL
) {
1170 for (n
= 0, c
= 0; c
< 4; ++c
)
1171 if (param
->mask
& (1 << c
))
1174 for (c
= n
; c
< 4; ++c
)
1176 src
= ureg_swizzle(src
, s
[0], s
[1], s
[2], s
[3]);
1182 NineTranslateInstruction_Mkxn(struct shader_translator
*tx
, const unsigned k
, const unsigned n
)
1184 struct ureg_program
*ureg
= tx
->ureg
;
1185 struct ureg_dst dst
;
1186 struct ureg_src src
[2];
1187 struct sm1_src_param
*src_mat
= &tx
->insn
.src
[1];
1190 dst
= tx_dst_param(tx
, &tx
->insn
.dst
[0]);
1191 src
[0] = tx_src_param(tx
, &tx
->insn
.src
[0]);
1193 for (i
= 0; i
< n
; i
++)
1195 const unsigned m
= (1 << i
);
1197 src
[1] = tx_src_param(tx
, src_mat
);
1200 if (!(dst
.WriteMask
& m
))
1203 /* XXX: src == dst case ? */
1207 ureg_DP3(ureg
, ureg_writemask(dst
, m
), src
[0], src
[1]);
1210 ureg_DP4(ureg
, ureg_writemask(dst
, m
), src
[0], src
[1]);
1213 DBG("invalid operation: M%ux%u\n", m
, n
);
1221 #define VNOTSUPPORTED 0, 0
1222 #define V(maj, min) (((maj) << 8) | (min))
1224 static INLINE
const char *
1225 d3dsio_to_string( unsigned opcode
)
1227 static const char *names
[] = {
1327 if (opcode
< Elements(names
)) return names
[opcode
];
1330 case D3DSIO_PHASE
: return "PHASE";
1331 case D3DSIO_COMMENT
: return "COMMENT";
1332 case D3DSIO_END
: return "END";
1338 #define NULL_INSTRUCTION { 0, { 0, 0 }, { 0, 0 }, 0, 0, NULL }
1339 #define IS_VALID_INSTRUCTION(inst) ((inst).vert_version.min | \
1340 (inst).vert_version.max | \
1341 (inst).frag_version.min | \
1342 (inst).frag_version.max)
1344 #define SPECIAL(name) \
1345 NineTranslateInstruction_##name
1347 #define DECL_SPECIAL(name) \
1349 NineTranslateInstruction_##name( struct shader_translator *tx )
1352 NineTranslateInstruction_Generic(struct shader_translator
*);
1356 return NineTranslateInstruction_Mkxn(tx
, 4, 4);
1361 return NineTranslateInstruction_Mkxn(tx
, 4, 3);
1366 return NineTranslateInstruction_Mkxn(tx
, 3, 4);
1371 return NineTranslateInstruction_Mkxn(tx
, 3, 3);
1376 return NineTranslateInstruction_Mkxn(tx
, 3, 2);
1381 ureg_CMP(tx
->ureg
, tx_dst_param(tx
, &tx
->insn
.dst
[0]),
1382 tx_src_param(tx
, &tx
->insn
.src
[0]),
1383 tx_src_param(tx
, &tx
->insn
.src
[2]),
1384 tx_src_param(tx
, &tx
->insn
.src
[1]));
1390 struct ureg_dst dst
= tx_dst_param(tx
, &tx
->insn
.dst
[0]);
1391 struct ureg_dst cgt
;
1392 struct ureg_src cnd
;
1394 /* the coissue flag was a tip for compilers to advise to
1395 * execute two operations at the same time, in cases
1396 * the two executions had same dst with different channels.
1397 * It has no effect on current hw. However it seems CND
1398 * is affected. The handling of this very specific case
1399 * handled below mimick wine behaviour */
1400 if (tx
->insn
.coissue
&& tx
->version
.major
== 1 && tx
->version
.minor
< 4 && tx
->insn
.dst
[0].mask
!= NINED3DSP_WRITEMASK_3
) {
1402 dst
, tx_src_param(tx
, &tx
->insn
.src
[1]));
1406 cnd
= tx_src_param(tx
, &tx
->insn
.src
[0]);
1407 cgt
= tx_scratch(tx
);
1409 if (tx
->version
.major
== 1 && tx
->version
.minor
< 4)
1410 cnd
= ureg_scalar(cnd
, TGSI_SWIZZLE_W
);
1412 ureg_SGT(tx
->ureg
, cgt
, cnd
, ureg_imm1f(tx
->ureg
, 0.5f
));
1414 ureg_CMP(tx
->ureg
, dst
, ureg_negate(ureg_src(cgt
)),
1415 tx_src_param(tx
, &tx
->insn
.src
[1]),
1416 tx_src_param(tx
, &tx
->insn
.src
[2]));
1422 assert(tx
->insn
.src
[0].idx
< tx
->num_inst_labels
);
1423 ureg_CAL(tx
->ureg
, &tx
->inst_labels
[tx
->insn
.src
[0].idx
]);
1427 DECL_SPECIAL(CALLNZ
)
1429 struct ureg_program
*ureg
= tx
->ureg
;
1430 struct ureg_dst tmp
= tx_scratch_scalar(tx
);
1431 struct ureg_src src
= tx_src_param(tx
, &tx
->insn
.src
[1]);
1433 /* NOTE: source should be const bool, so we can use NOT/SUB instead of [U]SNE 0 */
1434 if (!tx
->insn
.flags
) {
1435 if (tx
->native_integers
)
1436 ureg_NOT(ureg
, tmp
, src
);
1438 ureg_SUB(ureg
, tmp
, ureg_imm1f(ureg
, 1.0f
), src
);
1440 ureg_IF(ureg
, tx
->insn
.flags
? src
: tx_src_scalar(tmp
), tx_cond(tx
));
1441 ureg_CAL(ureg
, &tx
->inst_labels
[tx
->insn
.src
[0].idx
]);
1447 DECL_SPECIAL(MOV_vs1x
)
1449 if (tx
->insn
.dst
[0].file
== D3DSPR_ADDR
) {
1450 /* Implementation note: We don't write directly
1451 * to the addr register, but to an intermediate
1453 * Contrary to the doc, when writing to ADDR here,
1454 * the rounding is not to nearest, but to lowest
1456 * Since we use ARR next, substract 0.5. */
1458 tx_dst_param(tx
, &tx
->insn
.dst
[0]),
1459 tx_src_param(tx
, &tx
->insn
.src
[0]),
1460 ureg_imm1f(tx
->ureg
, 0.5f
));
1463 return NineTranslateInstruction_Generic(tx
);
1468 struct ureg_program
*ureg
= tx
->ureg
;
1470 struct ureg_src src
= tx_src_param(tx
, &tx
->insn
.src
[1]);
1471 struct ureg_dst ctr
;
1472 struct ureg_dst tmp
;
1473 struct ureg_src ctrx
;
1475 label
= tx_bgnloop(tx
);
1476 ctr
= tx_get_loopctr(tx
, TRUE
);
1477 ctrx
= ureg_scalar(ureg_src(ctr
), TGSI_SWIZZLE_X
);
1479 /* src: num_iterations - start_value of al - step for al - 0 */
1480 ureg_MOV(ureg
, ctr
, src
);
1481 ureg_BGNLOOP(tx
->ureg
, label
);
1482 tmp
= tx_scratch_scalar(tx
);
1483 /* Initially ctr.x contains the number of iterations.
1484 * ctr.y will contain the updated value of al.
1485 * We decrease ctr.x at the end of every iteration,
1486 * and stop when it reaches 0. */
1488 if (!tx
->native_integers
) {
1489 /* case src and ctr contain floats */
1490 /* to avoid precision issue, we stop when ctr <= 0.5 */
1491 ureg_SGE(ureg
, tmp
, ureg_imm1f(ureg
, 0.5f
), ctrx
);
1492 ureg_IF(ureg
, tx_src_scalar(tmp
), tx_cond(tx
));
1494 /* case src and ctr contain integers */
1495 ureg_ISGE(ureg
, tmp
, ureg_imm1i(ureg
, 0), ctrx
);
1496 ureg_UIF(ureg
, tx_src_scalar(tmp
), tx_cond(tx
));
1510 DECL_SPECIAL(ENDLOOP
)
1512 struct ureg_program
*ureg
= tx
->ureg
;
1513 struct ureg_dst ctr
= tx_get_loopctr(tx
, TRUE
);
1514 struct ureg_dst dst_ctrx
, dst_al
;
1515 struct ureg_src src_ctr
, al_counter
;
1517 dst_ctrx
= ureg_writemask(ctr
, NINED3DSP_WRITEMASK_0
);
1518 dst_al
= ureg_writemask(ctr
, NINED3DSP_WRITEMASK_1
);
1519 src_ctr
= ureg_src(ctr
);
1520 al_counter
= ureg_scalar(src_ctr
, TGSI_SWIZZLE_Z
);
1523 * ctr.y (aL) += step */
1524 if (!tx
->native_integers
) {
1525 ureg_ADD(ureg
, dst_ctrx
, src_ctr
, ureg_imm1f(ureg
, -1.0f
));
1526 ureg_ADD(ureg
, dst_al
, src_ctr
, al_counter
);
1528 ureg_UADD(ureg
, dst_ctrx
, src_ctr
, ureg_imm1i(ureg
, -1));
1529 ureg_UADD(ureg
, dst_al
, src_ctr
, al_counter
);
1531 ureg_ENDLOOP(tx
->ureg
, tx_endloop(tx
));
1537 unsigned k
= tx
->num_inst_labels
;
1538 unsigned n
= tx
->insn
.src
[0].idx
;
1541 tx
->inst_labels
= REALLOC(tx
->inst_labels
,
1542 k
* sizeof(tx
->inst_labels
[0]),
1543 n
* sizeof(tx
->inst_labels
[0]));
1545 tx
->inst_labels
[n
] = ureg_get_instruction_number(tx
->ureg
);
1549 DECL_SPECIAL(SINCOS
)
1551 struct ureg_dst dst
= tx_dst_param(tx
, &tx
->insn
.dst
[0]);
1552 struct ureg_src src
= tx_src_param(tx
, &tx
->insn
.src
[0]);
1554 assert(!(dst
.WriteMask
& 0xc));
1556 dst
.WriteMask
&= TGSI_WRITEMASK_XY
; /* z undefined, w untouched */
1557 ureg_SCS(tx
->ureg
, dst
, src
);
1564 tx_dst_param(tx
, &tx
->insn
.dst
[0]),
1565 tx_src_param(tx
, &tx
->insn
.src
[0]));
1571 struct ureg_program
*ureg
= tx
->ureg
;
1573 struct ureg_src rep
= tx_src_param(tx
, &tx
->insn
.src
[0]);
1574 struct ureg_dst ctr
;
1575 struct ureg_dst tmp
;
1576 struct ureg_src ctrx
;
1578 label
= tx_bgnloop(tx
);
1579 ctr
= ureg_writemask(tx_get_loopctr(tx
, FALSE
), NINED3DSP_WRITEMASK_0
);
1580 ctrx
= ureg_scalar(ureg_src(ctr
), TGSI_SWIZZLE_X
);
1582 /* NOTE: rep must be constant, so we don't have to save the count */
1583 assert(rep
.File
== TGSI_FILE_CONSTANT
|| rep
.File
== TGSI_FILE_IMMEDIATE
);
1585 /* rep: num_iterations - 0 - 0 - 0 */
1586 ureg_MOV(ureg
, ctr
, rep
);
1587 ureg_BGNLOOP(ureg
, label
);
1588 tmp
= tx_scratch_scalar(tx
);
1589 /* Initially ctr.x contains the number of iterations.
1590 * We decrease ctr.x at the end of every iteration,
1591 * and stop when it reaches 0. */
1593 if (!tx
->native_integers
) {
1594 /* case src and ctr contain floats */
1595 /* to avoid precision issue, we stop when ctr <= 0.5 */
1596 ureg_SGE(ureg
, tmp
, ureg_imm1f(ureg
, 0.5f
), ctrx
);
1597 ureg_IF(ureg
, tx_src_scalar(tmp
), tx_cond(tx
));
1599 /* case src and ctr contain integers */
1600 ureg_ISGE(ureg
, tmp
, ureg_imm1i(ureg
, 0), ctrx
);
1601 ureg_UIF(ureg
, tx_src_scalar(tmp
), tx_cond(tx
));
1610 DECL_SPECIAL(ENDREP
)
1612 struct ureg_program
*ureg
= tx
->ureg
;
1613 struct ureg_dst ctr
= tx_get_loopctr(tx
, FALSE
);
1614 struct ureg_dst dst_ctrx
= ureg_writemask(ctr
, NINED3DSP_WRITEMASK_0
);
1615 struct ureg_src src_ctr
= ureg_src(ctr
);
1618 if (!tx
->native_integers
)
1619 ureg_ADD(ureg
, dst_ctrx
, src_ctr
, ureg_imm1f(ureg
, -1.0f
));
1621 ureg_UADD(ureg
, dst_ctrx
, src_ctr
, ureg_imm1i(ureg
, -1));
1623 ureg_ENDLOOP(tx
->ureg
, tx_endloop(tx
));
1630 ureg_ENDIF(tx
->ureg
);
1636 struct ureg_src src
= tx_src_param(tx
, &tx
->insn
.src
[0]);
1638 if (tx
->native_integers
&& tx
->insn
.src
[0].file
== D3DSPR_CONSTBOOL
)
1639 ureg_UIF(tx
->ureg
, src
, tx_cond(tx
));
1641 ureg_IF(tx
->ureg
, src
, tx_cond(tx
));
1646 static INLINE
unsigned
1647 sm1_insn_flags_to_tgsi_setop(BYTE flags
)
1650 case NINED3DSHADER_REL_OP_GT
: return TGSI_OPCODE_SGT
;
1651 case NINED3DSHADER_REL_OP_EQ
: return TGSI_OPCODE_SEQ
;
1652 case NINED3DSHADER_REL_OP_GE
: return TGSI_OPCODE_SGE
;
1653 case NINED3DSHADER_REL_OP_LT
: return TGSI_OPCODE_SLT
;
1654 case NINED3DSHADER_REL_OP_NE
: return TGSI_OPCODE_SNE
;
1655 case NINED3DSHADER_REL_OP_LE
: return TGSI_OPCODE_SLE
;
1657 assert(!"invalid comparison flags");
1658 return TGSI_OPCODE_SGT
;
1664 const unsigned cmp_op
= sm1_insn_flags_to_tgsi_setop(tx
->insn
.flags
);
1665 struct ureg_src src
[2];
1666 struct ureg_dst tmp
= ureg_writemask(tx_scratch(tx
), TGSI_WRITEMASK_X
);
1667 src
[0] = tx_src_param(tx
, &tx
->insn
.src
[0]);
1668 src
[1] = tx_src_param(tx
, &tx
->insn
.src
[1]);
1669 ureg_insn(tx
->ureg
, cmp_op
, &tmp
, 1, src
, 2);
1670 ureg_IF(tx
->ureg
, ureg_scalar(ureg_src(tmp
), TGSI_SWIZZLE_X
), tx_cond(tx
));
1676 ureg_ELSE(tx
->ureg
, tx_elsecond(tx
));
1680 DECL_SPECIAL(BREAKC
)
1682 const unsigned cmp_op
= sm1_insn_flags_to_tgsi_setop(tx
->insn
.flags
);
1683 struct ureg_src src
[2];
1684 struct ureg_dst tmp
= ureg_writemask(tx_scratch(tx
), TGSI_WRITEMASK_X
);
1685 src
[0] = tx_src_param(tx
, &tx
->insn
.src
[0]);
1686 src
[1] = tx_src_param(tx
, &tx
->insn
.src
[1]);
1687 ureg_insn(tx
->ureg
, cmp_op
, &tmp
, 1, src
, 2);
1688 ureg_IF(tx
->ureg
, ureg_scalar(ureg_src(tmp
), TGSI_SWIZZLE_X
), tx_cond(tx
));
1691 ureg_ENDIF(tx
->ureg
);
1695 static const char *sm1_declusage_names
[] =
1697 [D3DDECLUSAGE_POSITION
] = "POSITION",
1698 [D3DDECLUSAGE_BLENDWEIGHT
] = "BLENDWEIGHT",
1699 [D3DDECLUSAGE_BLENDINDICES
] = "BLENDINDICES",
1700 [D3DDECLUSAGE_NORMAL
] = "NORMAL",
1701 [D3DDECLUSAGE_PSIZE
] = "PSIZE",
1702 [D3DDECLUSAGE_TEXCOORD
] = "TEXCOORD",
1703 [D3DDECLUSAGE_TANGENT
] = "TANGENT",
1704 [D3DDECLUSAGE_BINORMAL
] = "BINORMAL",
1705 [D3DDECLUSAGE_TESSFACTOR
] = "TESSFACTOR",
1706 [D3DDECLUSAGE_POSITIONT
] = "POSITIONT",
1707 [D3DDECLUSAGE_COLOR
] = "COLOR",
1708 [D3DDECLUSAGE_FOG
] = "FOG",
1709 [D3DDECLUSAGE_DEPTH
] = "DEPTH",
1710 [D3DDECLUSAGE_SAMPLE
] = "SAMPLE"
1713 static INLINE
unsigned
1714 sm1_to_nine_declusage(struct sm1_semantic
*dcl
)
1716 return nine_d3d9_to_nine_declusage(dcl
->usage
, dcl
->usage_idx
);
1720 sm1_declusage_to_tgsi(struct tgsi_declaration_semantic
*sem
,
1722 struct sm1_semantic
*dcl
)
1724 BYTE index
= dcl
->usage_idx
;
1726 /* For everything that is not matching to a TGSI_SEMANTIC_****,
1727 * we match to a TGSI_SEMANTIC_GENERIC with index.
1729 * The index can be anything UINT16 and usage_idx is BYTE,
1730 * so we can fit everything. It doesn't matter if indices
1731 * are close together or low.
1734 * POSITION >= 1: 10 * index + 6
1735 * COLOR >= 2: 10 * (index-1) + 7
1736 * TEXCOORD[0..15]: index
1737 * BLENDWEIGHT: 10 * index + 18
1738 * BLENDINDICES: 10 * index + 19
1739 * NORMAL: 10 * index + 20
1740 * TANGENT: 10 * index + 21
1741 * BINORMAL: 10 * index + 22
1742 * TESSFACTOR: 10 * index + 23
1745 switch (dcl
->usage
) {
1746 case D3DDECLUSAGE_POSITION
:
1747 case D3DDECLUSAGE_POSITIONT
:
1748 case D3DDECLUSAGE_DEPTH
:
1750 sem
->Name
= TGSI_SEMANTIC_POSITION
;
1753 sem
->Name
= TGSI_SEMANTIC_GENERIC
;
1754 sem
->Index
= 10 * index
+ 6;
1757 case D3DDECLUSAGE_COLOR
:
1759 sem
->Name
= TGSI_SEMANTIC_COLOR
;
1762 sem
->Name
= TGSI_SEMANTIC_GENERIC
;
1763 sem
->Index
= 10 * (index
-1) + 7;
1766 case D3DDECLUSAGE_FOG
:
1768 sem
->Name
= TGSI_SEMANTIC_FOG
;
1771 case D3DDECLUSAGE_PSIZE
:
1773 sem
->Name
= TGSI_SEMANTIC_PSIZE
;
1776 case D3DDECLUSAGE_TEXCOORD
:
1778 if (index
< 8 && tc
)
1779 sem
->Name
= TGSI_SEMANTIC_TEXCOORD
;
1781 sem
->Name
= TGSI_SEMANTIC_GENERIC
;
1784 case D3DDECLUSAGE_BLENDWEIGHT
:
1785 sem
->Name
= TGSI_SEMANTIC_GENERIC
;
1786 sem
->Index
= 10 * index
+ 18;
1788 case D3DDECLUSAGE_BLENDINDICES
:
1789 sem
->Name
= TGSI_SEMANTIC_GENERIC
;
1790 sem
->Index
= 10 * index
+ 19;
1792 case D3DDECLUSAGE_NORMAL
:
1793 sem
->Name
= TGSI_SEMANTIC_GENERIC
;
1794 sem
->Index
= 10 * index
+ 20;
1796 case D3DDECLUSAGE_TANGENT
:
1797 sem
->Name
= TGSI_SEMANTIC_GENERIC
;
1798 sem
->Index
= 10 * index
+ 21;
1800 case D3DDECLUSAGE_BINORMAL
:
1801 sem
->Name
= TGSI_SEMANTIC_GENERIC
;
1802 sem
->Index
= 10 * index
+ 22;
1804 case D3DDECLUSAGE_TESSFACTOR
:
1805 sem
->Name
= TGSI_SEMANTIC_GENERIC
;
1806 sem
->Index
= 10 * index
+ 23;
1808 case D3DDECLUSAGE_SAMPLE
:
1809 sem
->Name
= TGSI_SEMANTIC_COUNT
;
1813 assert(!"Invalid DECLUSAGE.");
1818 #define NINED3DSTT_1D (D3DSTT_1D >> D3DSP_TEXTURETYPE_SHIFT)
1819 #define NINED3DSTT_2D (D3DSTT_2D >> D3DSP_TEXTURETYPE_SHIFT)
1820 #define NINED3DSTT_VOLUME (D3DSTT_VOLUME >> D3DSP_TEXTURETYPE_SHIFT)
1821 #define NINED3DSTT_CUBE (D3DSTT_CUBE >> D3DSP_TEXTURETYPE_SHIFT)
1822 static INLINE
unsigned
1823 d3dstt_to_tgsi_tex(BYTE sampler_type
)
1825 switch (sampler_type
) {
1826 case NINED3DSTT_1D
: return TGSI_TEXTURE_1D
;
1827 case NINED3DSTT_2D
: return TGSI_TEXTURE_2D
;
1828 case NINED3DSTT_VOLUME
: return TGSI_TEXTURE_3D
;
1829 case NINED3DSTT_CUBE
: return TGSI_TEXTURE_CUBE
;
1832 return TGSI_TEXTURE_UNKNOWN
;
1835 static INLINE
unsigned
1836 d3dstt_to_tgsi_tex_shadow(BYTE sampler_type
)
1838 switch (sampler_type
) {
1839 case NINED3DSTT_1D
: return TGSI_TEXTURE_SHADOW1D
;
1840 case NINED3DSTT_2D
: return TGSI_TEXTURE_SHADOW2D
;
1841 case NINED3DSTT_VOLUME
:
1842 case NINED3DSTT_CUBE
:
1845 return TGSI_TEXTURE_UNKNOWN
;
1848 static INLINE
unsigned
1849 ps1x_sampler_type(const struct nine_shader_info
*info
, unsigned stage
)
1851 switch ((info
->sampler_ps1xtypes
>> (stage
* 2)) & 0x3) {
1852 case 1: return TGSI_TEXTURE_1D
;
1853 case 0: return TGSI_TEXTURE_2D
;
1854 case 3: return TGSI_TEXTURE_3D
;
1856 return TGSI_TEXTURE_CUBE
;
1861 sm1_sampler_type_name(BYTE sampler_type
)
1863 switch (sampler_type
) {
1864 case NINED3DSTT_1D
: return "1D";
1865 case NINED3DSTT_2D
: return "2D";
1866 case NINED3DSTT_VOLUME
: return "VOLUME";
1867 case NINED3DSTT_CUBE
: return "CUBE";
1869 return "(D3DSTT_?)";
1873 static INLINE
unsigned
1874 nine_tgsi_to_interp_mode(struct tgsi_declaration_semantic
*sem
)
1876 switch (sem
->Name
) {
1877 case TGSI_SEMANTIC_POSITION
:
1878 case TGSI_SEMANTIC_NORMAL
:
1879 return TGSI_INTERPOLATE_LINEAR
;
1880 case TGSI_SEMANTIC_BCOLOR
:
1881 case TGSI_SEMANTIC_COLOR
:
1882 case TGSI_SEMANTIC_FOG
:
1883 case TGSI_SEMANTIC_GENERIC
:
1884 case TGSI_SEMANTIC_TEXCOORD
:
1885 case TGSI_SEMANTIC_CLIPDIST
:
1886 case TGSI_SEMANTIC_CLIPVERTEX
:
1887 return TGSI_INTERPOLATE_PERSPECTIVE
;
1888 case TGSI_SEMANTIC_EDGEFLAG
:
1889 case TGSI_SEMANTIC_FACE
:
1890 case TGSI_SEMANTIC_INSTANCEID
:
1891 case TGSI_SEMANTIC_PCOORD
:
1892 case TGSI_SEMANTIC_PRIMID
:
1893 case TGSI_SEMANTIC_PSIZE
:
1894 case TGSI_SEMANTIC_VERTEXID
:
1895 return TGSI_INTERPOLATE_CONSTANT
;
1898 return TGSI_INTERPOLATE_CONSTANT
;
1904 struct ureg_program
*ureg
= tx
->ureg
;
1907 struct tgsi_declaration_semantic tgsi
;
1908 struct sm1_semantic sem
;
1909 sm1_read_semantic(tx
, &sem
);
1911 is_input
= sem
.reg
.file
== D3DSPR_INPUT
;
1913 sem
.usage
== D3DDECLUSAGE_SAMPLE
|| sem
.reg
.file
== D3DSPR_SAMPLER
;
1916 sm1_dump_dst_param(&sem
.reg
);
1918 DUMP(" %s\n", sm1_sampler_type_name(sem
.sampler_type
));
1920 if (tx
->version
.major
>= 3)
1921 DUMP(" %s%i\n", sm1_declusage_names
[sem
.usage
], sem
.usage_idx
);
1923 if (sem
.usage
| sem
.usage_idx
)
1924 DUMP(" %u[%u]\n", sem
.usage
, sem
.usage_idx
);
1929 const unsigned m
= 1 << sem
.reg
.idx
;
1930 ureg_DECL_sampler(ureg
, sem
.reg
.idx
);
1931 tx
->info
->sampler_mask
|= m
;
1932 tx
->sampler_targets
[sem
.reg
.idx
] = (tx
->info
->sampler_mask_shadow
& m
) ?
1933 d3dstt_to_tgsi_tex_shadow(sem
.sampler_type
) :
1934 d3dstt_to_tgsi_tex(sem
.sampler_type
);
1938 sm1_declusage_to_tgsi(&tgsi
, tx
->want_texcoord
, &sem
);
1941 /* linkage outside of shader with vertex declaration */
1942 ureg_DECL_vs_input(ureg
, sem
.reg
.idx
);
1943 assert(sem
.reg
.idx
< Elements(tx
->info
->input_map
));
1944 tx
->info
->input_map
[sem
.reg
.idx
] = sm1_to_nine_declusage(&sem
);
1945 tx
->info
->num_inputs
= sem
.reg
.idx
+ 1;
1946 /* NOTE: preserving order in case of indirect access */
1948 if (tx
->version
.major
>= 3) {
1949 /* SM2 output semantic determined by file */
1950 assert(sem
.reg
.mask
!= 0);
1951 if (sem
.usage
== D3DDECLUSAGE_POSITIONT
)
1952 tx
->info
->position_t
= TRUE
;
1953 assert(sem
.reg
.idx
< Elements(tx
->regs
.o
));
1954 tx
->regs
.o
[sem
.reg
.idx
] = ureg_DECL_output_masked(
1955 ureg
, tgsi
.Name
, tgsi
.Index
, sem
.reg
.mask
);
1957 if (tgsi
.Name
== TGSI_SEMANTIC_PSIZE
)
1958 tx
->regs
.oPts
= tx
->regs
.o
[sem
.reg
.idx
];
1961 if (is_input
&& tx
->version
.major
>= 3) {
1962 /* SM3 only, SM2 input semantic determined by file */
1963 assert(sem
.reg
.idx
< Elements(tx
->regs
.v
));
1964 tx
->regs
.v
[sem
.reg
.idx
] = ureg_DECL_fs_input_cyl_centroid(
1965 ureg
, tgsi
.Name
, tgsi
.Index
,
1966 nine_tgsi_to_interp_mode(&tgsi
),
1968 sem
.reg
.mod
& NINED3DSPDM_CENTROID
);
1970 if (!is_input
&& 0) { /* declare in COLOROUT/DEPTHOUT case */
1971 /* FragColor or FragDepth */
1972 assert(sem
.reg
.mask
!= 0);
1973 ureg_DECL_output_masked(ureg
, tgsi
.Name
, tgsi
.Index
, sem
.reg
.mask
);
1981 tx_set_lconstf(tx
, tx
->insn
.dst
[0].idx
, tx
->insn
.src
[0].imm
.f
);
1987 tx_set_lconstb(tx
, tx
->insn
.dst
[0].idx
, tx
->insn
.src
[0].imm
.b
);
1993 tx_set_lconsti(tx
, tx
->insn
.dst
[0].idx
, tx
->insn
.src
[0].imm
.i
);
1999 struct ureg_dst dst
= tx_dst_param(tx
, &tx
->insn
.dst
[0]);
2000 struct ureg_src src
[2] = {
2001 tx_src_param(tx
, &tx
->insn
.src
[0]),
2002 tx_src_param(tx
, &tx
->insn
.src
[1])
2004 ureg_POW(tx
->ureg
, dst
, ureg_abs(src
[0]), src
[1]);
2010 struct ureg_program
*ureg
= tx
->ureg
;
2011 struct ureg_dst dst
= tx_dst_param(tx
, &tx
->insn
.dst
[0]);
2012 struct ureg_src src
= tx_src_param(tx
, &tx
->insn
.src
[0]);
2013 struct ureg_dst tmp
= tx_scratch(tx
);
2014 ureg_RSQ(ureg
, tmp
, ureg_abs(src
));
2015 ureg_MIN(ureg
, dst
, ureg_imm1f(ureg
, FLT_MAX
), ureg_src(tmp
));
2021 struct ureg_program
*ureg
= tx
->ureg
;
2022 struct ureg_dst tmp
= tx_scratch_scalar(tx
);
2023 struct ureg_dst dst
= tx_dst_param(tx
, &tx
->insn
.dst
[0]);
2024 struct ureg_src src
= tx_src_param(tx
, &tx
->insn
.src
[0]);
2025 ureg_LG2(ureg
, tmp
, ureg_abs(src
));
2026 ureg_MAX(ureg
, dst
, ureg_imm1f(ureg
, -FLT_MAX
), tx_src_scalar(tmp
));
2032 struct ureg_program
*ureg
= tx
->ureg
;
2033 struct ureg_dst tmp
= tx_scratch_scalar(tx
);
2034 struct ureg_src nrm
= tx_src_scalar(tmp
);
2035 struct ureg_dst dst
= tx_dst_param(tx
, &tx
->insn
.dst
[0]);
2036 struct ureg_src src
= tx_src_param(tx
, &tx
->insn
.src
[0]);
2037 ureg_DP3(ureg
, tmp
, src
, src
);
2038 ureg_RSQ(ureg
, tmp
, nrm
);
2039 ureg_MIN(ureg
, tmp
, ureg_imm1f(ureg
, FLT_MAX
), nrm
);
2040 ureg_MUL(ureg
, dst
, src
, nrm
);
2044 DECL_SPECIAL(DP2ADD
)
2046 struct ureg_dst tmp
= tx_scratch_scalar(tx
);
2047 struct ureg_src dp2
= tx_src_scalar(tmp
);
2048 struct ureg_dst dst
= tx_dst_param(tx
, &tx
->insn
.dst
[0]);
2049 struct ureg_src src
[3];
2051 for (i
= 0; i
< 3; ++i
)
2052 src
[i
] = tx_src_param(tx
, &tx
->insn
.src
[i
]);
2053 assert_replicate_swizzle(&src
[2]);
2055 ureg_DP2(tx
->ureg
, tmp
, src
[0], src
[1]);
2056 ureg_ADD(tx
->ureg
, dst
, src
[2], dp2
);
2061 DECL_SPECIAL(TEXCOORD
)
2063 struct ureg_program
*ureg
= tx
->ureg
;
2064 const unsigned s
= tx
->insn
.dst
[0].idx
;
2065 struct ureg_dst dst
= tx_dst_param(tx
, &tx
->insn
.dst
[0]);
2067 tx_texcoord_alloc(tx
, s
);
2068 ureg_MOV(ureg
, dst
, tx
->regs
.vT
[s
]); /* XXX is this sufficient ? */
2073 DECL_SPECIAL(TEXCOORD_ps14
)
2075 struct ureg_program
*ureg
= tx
->ureg
;
2076 const unsigned s
= tx
->insn
.src
[0].idx
;
2077 struct ureg_dst dst
= tx_dst_param(tx
, &tx
->insn
.dst
[0]);
2079 tx_texcoord_alloc(tx
, s
);
2080 ureg_MOV(ureg
, dst
, tx
->regs
.vT
[s
]); /* XXX is this sufficient ? */
2085 DECL_SPECIAL(TEXKILL
)
2087 struct ureg_src reg
;
2089 if (tx
->version
.major
> 1 || tx
->version
.minor
> 3) {
2090 reg
= tx_dst_param_as_src(tx
, &tx
->insn
.dst
[0]);
2092 tx_texcoord_alloc(tx
, tx
->insn
.dst
[0].idx
);
2093 reg
= tx
->regs
.vT
[tx
->insn
.dst
[0].idx
];
2095 if (tx
->version
.major
< 2)
2096 reg
= ureg_swizzle(reg
, NINE_SWIZZLE4(X
,Y
,Z
,Z
));
2097 ureg_KILL_IF(tx
->ureg
, reg
);
2102 DECL_SPECIAL(TEXBEM
)
2104 STUB(D3DERR_INVALIDCALL
);
2107 DECL_SPECIAL(TEXBEML
)
2109 STUB(D3DERR_INVALIDCALL
);
2112 DECL_SPECIAL(TEXREG2AR
)
2114 STUB(D3DERR_INVALIDCALL
);
2117 DECL_SPECIAL(TEXREG2GB
)
2119 STUB(D3DERR_INVALIDCALL
);
2122 DECL_SPECIAL(TEXM3x2PAD
)
2124 STUB(D3DERR_INVALIDCALL
);
2127 DECL_SPECIAL(TEXM3x2TEX
)
2129 STUB(D3DERR_INVALIDCALL
);
2132 DECL_SPECIAL(TEXM3x3PAD
)
2134 return D3D_OK
; /* this is just padding */
2137 DECL_SPECIAL(TEXM3x3SPEC
)
2139 STUB(D3DERR_INVALIDCALL
);
2142 DECL_SPECIAL(TEXM3x3VSPEC
)
2144 STUB(D3DERR_INVALIDCALL
);
2147 DECL_SPECIAL(TEXREG2RGB
)
2149 STUB(D3DERR_INVALIDCALL
);
2152 DECL_SPECIAL(TEXDP3TEX
)
2154 STUB(D3DERR_INVALIDCALL
);
2157 DECL_SPECIAL(TEXM3x2DEPTH
)
2159 STUB(D3DERR_INVALIDCALL
);
2162 DECL_SPECIAL(TEXDP3
)
2164 STUB(D3DERR_INVALIDCALL
);
2167 DECL_SPECIAL(TEXM3x3
)
2169 struct ureg_program
*ureg
= tx
->ureg
;
2170 struct ureg_dst dst
= tx_dst_param(tx
, &tx
->insn
.dst
[0]);
2171 struct ureg_src src
[4];
2173 const int m
= tx
->insn
.dst
[0].idx
- 2;
2174 const int n
= tx
->insn
.src
[0].idx
;
2175 assert(m
>= 0 && m
> n
);
2177 for (s
= m
; s
<= (m
+ 2); ++s
) {
2178 tx_texcoord_alloc(tx
, s
);
2179 src
[s
] = tx
->regs
.vT
[s
];
2181 ureg_DP3(ureg
, ureg_writemask(dst
, TGSI_WRITEMASK_X
), src
[0], ureg_src(tx
->regs
.tS
[n
]));
2182 ureg_DP3(ureg
, ureg_writemask(dst
, TGSI_WRITEMASK_Y
), src
[1], ureg_src(tx
->regs
.tS
[n
]));
2183 ureg_DP3(ureg
, ureg_writemask(dst
, TGSI_WRITEMASK_Z
), src
[2], ureg_src(tx
->regs
.tS
[n
]));
2185 switch (tx
->insn
.opcode
) {
2186 case D3DSIO_TEXM3x3
:
2187 ureg_MOV(ureg
, ureg_writemask(dst
, TGSI_WRITEMASK_W
), ureg_imm1f(ureg
, 1.0f
));
2189 case D3DSIO_TEXM3x3TEX
:
2190 src
[3] = ureg_DECL_sampler(ureg
, m
+ 2);
2191 tx
->info
->sampler_mask
|= 1 << (m
+ 2);
2192 ureg_TEX(ureg
, dst
, ps1x_sampler_type(tx
->info
, m
+ 2), ureg_src(dst
), src
[3]);
2195 return D3DERR_INVALIDCALL
;
2200 DECL_SPECIAL(TEXDEPTH
)
2202 STUB(D3DERR_INVALIDCALL
);
2207 STUB(D3DERR_INVALIDCALL
);
2212 struct ureg_program
*ureg
= tx
->ureg
;
2214 struct ureg_dst dst
= tx_dst_param(tx
, &tx
->insn
.dst
[0]);
2215 struct ureg_src src
[2] = {
2216 tx_src_param(tx
, &tx
->insn
.src
[0]),
2217 tx_src_param(tx
, &tx
->insn
.src
[1])
2219 assert(tx
->insn
.src
[1].idx
>= 0 &&
2220 tx
->insn
.src
[1].idx
< Elements(tx
->sampler_targets
));
2221 target
= tx
->sampler_targets
[tx
->insn
.src
[1].idx
];
2223 switch (tx
->insn
.flags
) {
2225 ureg_TEX(ureg
, dst
, target
, src
[0], src
[1]);
2227 case NINED3DSI_TEXLD_PROJECT
:
2228 ureg_TXP(ureg
, dst
, target
, src
[0], src
[1]);
2230 case NINED3DSI_TEXLD_BIAS
:
2231 ureg_TXB(ureg
, dst
, target
, src
[0], src
[1]);
2235 return D3DERR_INVALIDCALL
;
2240 DECL_SPECIAL(TEXLD_14
)
2242 struct ureg_program
*ureg
= tx
->ureg
;
2243 struct ureg_dst dst
= tx_dst_param(tx
, &tx
->insn
.dst
[0]);
2244 struct ureg_src src
= tx_src_param(tx
, &tx
->insn
.src
[0]);
2245 const unsigned s
= tx
->insn
.dst
[0].idx
;
2246 const unsigned t
= ps1x_sampler_type(tx
->info
, s
);
2248 tx
->info
->sampler_mask
|= 1 << s
;
2249 ureg_TEX(ureg
, dst
, t
, src
, ureg_DECL_sampler(ureg
, s
));
2256 struct ureg_program
*ureg
= tx
->ureg
;
2257 const unsigned s
= tx
->insn
.dst
[0].idx
;
2258 const unsigned t
= ps1x_sampler_type(tx
->info
, s
);
2259 struct ureg_dst dst
= tx_dst_param(tx
, &tx
->insn
.dst
[0]);
2260 struct ureg_src src
[2];
2262 tx_texcoord_alloc(tx
, s
);
2264 src
[0] = tx
->regs
.vT
[s
];
2265 src
[1] = ureg_DECL_sampler(ureg
, s
);
2266 tx
->info
->sampler_mask
|= 1 << s
;
2268 ureg_TEX(ureg
, dst
, t
, src
[0], src
[1]);
2273 DECL_SPECIAL(TEXLDD
)
2276 struct ureg_dst dst
= tx_dst_param(tx
, &tx
->insn
.dst
[0]);
2277 struct ureg_src src
[4] = {
2278 tx_src_param(tx
, &tx
->insn
.src
[0]),
2279 tx_src_param(tx
, &tx
->insn
.src
[1]),
2280 tx_src_param(tx
, &tx
->insn
.src
[2]),
2281 tx_src_param(tx
, &tx
->insn
.src
[3])
2283 assert(tx
->insn
.src
[3].idx
>= 0 &&
2284 tx
->insn
.src
[3].idx
< Elements(tx
->sampler_targets
));
2285 target
= tx
->sampler_targets
[tx
->insn
.src
[1].idx
];
2287 ureg_TXD(tx
->ureg
, dst
, target
, src
[0], src
[2], src
[3], src
[1]);
2291 DECL_SPECIAL(TEXLDL
)
2294 struct ureg_dst dst
= tx_dst_param(tx
, &tx
->insn
.dst
[0]);
2295 struct ureg_src src
[2] = {
2296 tx_src_param(tx
, &tx
->insn
.src
[0]),
2297 tx_src_param(tx
, &tx
->insn
.src
[1])
2299 assert(tx
->insn
.src
[3].idx
>= 0 &&
2300 tx
->insn
.src
[3].idx
< Elements(tx
->sampler_targets
));
2301 target
= tx
->sampler_targets
[tx
->insn
.src
[1].idx
];
2303 ureg_TXL(tx
->ureg
, dst
, target
, src
[0], src
[1]);
2309 STUB(D3DERR_INVALIDCALL
);
2312 DECL_SPECIAL(BREAKP
)
2314 STUB(D3DERR_INVALIDCALL
);
2319 return D3D_OK
; /* we don't care about phase */
2322 DECL_SPECIAL(COMMENT
)
2324 return D3D_OK
; /* nothing to do */
2328 #define _OPI(o,t,vv1,vv2,pv1,pv2,d,s,h) \
2329 { D3DSIO_##o, TGSI_OPCODE_##t, { vv1, vv2 }, { pv1, pv2, }, d, s, h }
2331 struct sm1_op_info inst_table
[] =
2333 _OPI(NOP
, NOP
, V(0,0), V(3,0), V(0,0), V(3,0), 0, 0, NULL
), /* 0 */
2334 _OPI(MOV
, MOV
, V(0,0), V(1,1), V(0,0), V(0,0), 1, 1, SPECIAL(MOV_vs1x
)),
2335 _OPI(MOV
, MOV
, V(2,0), V(3,0), V(0,0), V(3,0), 1, 1, NULL
),
2336 _OPI(ADD
, ADD
, V(0,0), V(3,0), V(0,0), V(3,0), 1, 2, NULL
), /* 2 */
2337 _OPI(SUB
, SUB
, V(0,0), V(3,0), V(0,0), V(3,0), 1, 2, NULL
), /* 3 */
2338 _OPI(MAD
, MAD
, V(0,0), V(3,0), V(0,0), V(3,0), 1, 3, NULL
), /* 4 */
2339 _OPI(MUL
, MUL
, V(0,0), V(3,0), V(0,0), V(3,0), 1, 2, NULL
), /* 5 */
2340 _OPI(RCP
, RCP
, V(0,0), V(3,0), V(0,0), V(3,0), 1, 1, NULL
), /* 6 */
2341 _OPI(RSQ
, RSQ
, V(0,0), V(3,0), V(0,0), V(3,0), 1, 1, SPECIAL(RSQ
)), /* 7 */
2342 _OPI(DP3
, DP3
, V(0,0), V(3,0), V(0,0), V(3,0), 1, 2, NULL
), /* 8 */
2343 _OPI(DP4
, DP4
, V(0,0), V(3,0), V(0,0), V(3,0), 1, 2, NULL
), /* 9 */
2344 _OPI(MIN
, MIN
, V(0,0), V(3,0), V(0,0), V(3,0), 1, 2, NULL
), /* 10 */
2345 _OPI(MAX
, MAX
, V(0,0), V(3,0), V(0,0), V(3,0), 1, 2, NULL
), /* 11 */
2346 _OPI(SLT
, SLT
, V(0,0), V(3,0), V(0,0), V(3,0), 1, 2, NULL
), /* 12 */
2347 _OPI(SGE
, SGE
, V(0,0), V(3,0), V(0,0), V(3,0), 1, 2, NULL
), /* 13 */
2348 _OPI(EXP
, EX2
, V(0,0), V(3,0), V(0,0), V(3,0), 1, 1, NULL
), /* 14 */
2349 _OPI(LOG
, LG2
, V(0,0), V(3,0), V(0,0), V(3,0), 1, 1, SPECIAL(LOG
)), /* 15 */
2350 _OPI(LIT
, LIT
, V(0,0), V(3,0), V(0,0), V(0,0), 1, 1, NULL
), /* 16 */
2351 _OPI(DST
, DST
, V(0,0), V(3,0), V(0,0), V(3,0), 1, 2, NULL
), /* 17 */
2352 _OPI(LRP
, LRP
, V(0,0), V(3,0), V(0,0), V(3,0), 1, 3, NULL
), /* 18 */
2353 _OPI(FRC
, FRC
, V(0,0), V(3,0), V(0,0), V(3,0), 1, 1, NULL
), /* 19 */
2355 _OPI(M4x4
, NOP
, V(0,0), V(3,0), V(0,0), V(3,0), 1, 2, SPECIAL(M4x4
)),
2356 _OPI(M4x3
, NOP
, V(0,0), V(3,0), V(0,0), V(3,0), 1, 2, SPECIAL(M4x3
)),
2357 _OPI(M3x4
, NOP
, V(0,0), V(3,0), V(0,0), V(3,0), 1, 2, SPECIAL(M3x4
)),
2358 _OPI(M3x3
, NOP
, V(0,0), V(3,0), V(0,0), V(3,0), 1, 2, SPECIAL(M3x3
)),
2359 _OPI(M3x2
, NOP
, V(0,0), V(3,0), V(0,0), V(3,0), 1, 2, SPECIAL(M3x2
)),
2361 _OPI(CALL
, CAL
, V(2,0), V(3,0), V(2,1), V(3,0), 0, 0, SPECIAL(CALL
)),
2362 _OPI(CALLNZ
, CAL
, V(2,0), V(3,0), V(2,1), V(3,0), 0, 0, SPECIAL(CALLNZ
)),
2363 _OPI(LOOP
, BGNLOOP
, V(2,0), V(3,0), V(3,0), V(3,0), 0, 2, SPECIAL(LOOP
)),
2364 _OPI(RET
, RET
, V(2,0), V(3,0), V(2,1), V(3,0), 0, 0, SPECIAL(RET
)),
2365 _OPI(ENDLOOP
, ENDLOOP
, V(2,0), V(3,0), V(3,0), V(3,0), 0, 0, SPECIAL(ENDLOOP
)),
2366 _OPI(LABEL
, NOP
, V(2,0), V(3,0), V(2,1), V(3,0), 0, 0, SPECIAL(LABEL
)),
2368 _OPI(DCL
, NOP
, V(0,0), V(3,0), V(0,0), V(3,0), 0, 0, SPECIAL(DCL
)),
2370 _OPI(POW
, POW
, V(0,0), V(3,0), V(0,0), V(3,0), 1, 2, SPECIAL(POW
)),
2371 _OPI(CRS
, XPD
, V(0,0), V(3,0), V(0,0), V(3,0), 1, 2, NULL
), /* XXX: .w */
2372 _OPI(SGN
, SSG
, V(2,0), V(3,0), V(0,0), V(0,0), 1, 3, SPECIAL(SGN
)), /* ignore src1,2 */
2373 _OPI(ABS
, ABS
, V(0,0), V(3,0), V(0,0), V(3,0), 1, 1, NULL
),
2374 _OPI(NRM
, NOP
, V(0,0), V(3,0), V(0,0), V(3,0), 1, 1, SPECIAL(NRM
)), /* NRM doesn't fit */
2376 _OPI(SINCOS
, SCS
, V(2,0), V(2,1), V(2,0), V(2,1), 1, 3, SPECIAL(SINCOS
)),
2377 _OPI(SINCOS
, SCS
, V(3,0), V(3,0), V(3,0), V(3,0), 1, 1, SPECIAL(SINCOS
)),
2379 /* More flow control */
2380 _OPI(REP
, NOP
, V(2,0), V(3,0), V(2,1), V(3,0), 0, 1, SPECIAL(REP
)),
2381 _OPI(ENDREP
, NOP
, V(2,0), V(3,0), V(2,1), V(3,0), 0, 0, SPECIAL(ENDREP
)),
2382 _OPI(IF
, IF
, V(2,0), V(3,0), V(2,1), V(3,0), 0, 1, SPECIAL(IF
)),
2383 _OPI(IFC
, IF
, V(2,1), V(3,0), V(2,1), V(3,0), 0, 2, SPECIAL(IFC
)),
2384 _OPI(ELSE
, ELSE
, V(2,0), V(3,0), V(2,1), V(3,0), 0, 0, SPECIAL(ELSE
)),
2385 _OPI(ENDIF
, ENDIF
, V(2,0), V(3,0), V(2,1), V(3,0), 0, 0, SPECIAL(ENDIF
)),
2386 _OPI(BREAK
, BRK
, V(2,1), V(3,0), V(2,1), V(3,0), 0, 0, NULL
),
2387 _OPI(BREAKC
, BREAKC
, V(2,1), V(3,0), V(2,1), V(3,0), 0, 2, SPECIAL(BREAKC
)),
2388 /* we don't write to the address register, but a normal register (copied
2389 * when needed to the address register), thus we don't use ARR */
2390 _OPI(MOVA
, MOV
, V(2,0), V(3,0), V(0,0), V(0,0), 1, 1, NULL
),
2392 _OPI(DEFB
, NOP
, V(0,0), V(3,0) , V(0,0), V(3,0) , 1, 0, SPECIAL(DEFB
)),
2393 _OPI(DEFI
, NOP
, V(0,0), V(3,0) , V(0,0), V(3,0) , 1, 0, SPECIAL(DEFI
)),
2395 _OPI(TEXCOORD
, NOP
, V(0,0), V(0,0), V(0,0), V(1,3), 1, 0, SPECIAL(TEXCOORD
)),
2396 _OPI(TEXCOORD
, MOV
, V(0,0), V(0,0), V(1,4), V(1,4), 1, 1, SPECIAL(TEXCOORD_ps14
)),
2397 _OPI(TEXKILL
, KILL_IF
, V(0,0), V(0,0), V(0,0), V(3,0), 1, 0, SPECIAL(TEXKILL
)),
2398 _OPI(TEX
, TEX
, V(0,0), V(0,0), V(0,0), V(1,3), 1, 0, SPECIAL(TEX
)),
2399 _OPI(TEX
, TEX
, V(0,0), V(0,0), V(1,4), V(1,4), 1, 1, SPECIAL(TEXLD_14
)),
2400 _OPI(TEX
, TEX
, V(0,0), V(0,0), V(2,0), V(3,0), 1, 2, SPECIAL(TEXLD
)),
2401 _OPI(TEXBEM
, TEX
, V(0,0), V(0,0), V(0,0), V(1,3), 0, 0, SPECIAL(TEXBEM
)),
2402 _OPI(TEXBEML
, TEX
, V(0,0), V(0,0), V(0,0), V(1,3), 0, 0, SPECIAL(TEXBEML
)),
2403 _OPI(TEXREG2AR
, TEX
, V(0,0), V(0,0), V(0,0), V(1,3), 0, 0, SPECIAL(TEXREG2AR
)),
2404 _OPI(TEXREG2GB
, TEX
, V(0,0), V(0,0), V(0,0), V(1,3), 0, 0, SPECIAL(TEXREG2GB
)),
2405 _OPI(TEXM3x2PAD
, TEX
, V(0,0), V(0,0), V(0,0), V(1,3), 0, 0, SPECIAL(TEXM3x2PAD
)),
2406 _OPI(TEXM3x2TEX
, TEX
, V(0,0), V(0,0), V(0,0), V(1,3), 0, 0, SPECIAL(TEXM3x2TEX
)),
2407 _OPI(TEXM3x3PAD
, TEX
, V(0,0), V(0,0), V(0,0), V(1,3), 0, 0, SPECIAL(TEXM3x3PAD
)),
2408 _OPI(TEXM3x3TEX
, TEX
, V(0,0), V(0,0), V(0,0), V(1,3), 0, 0, SPECIAL(TEXM3x3
)),
2409 _OPI(TEXM3x3SPEC
, TEX
, V(0,0), V(0,0), V(0,0), V(1,3), 0, 0, SPECIAL(TEXM3x3SPEC
)),
2410 _OPI(TEXM3x3VSPEC
, TEX
, V(0,0), V(0,0), V(0,0), V(1,3), 0, 0, SPECIAL(TEXM3x3VSPEC
)),
2412 _OPI(EXPP
, EXP
, V(0,0), V(1,1), V(0,0), V(0,0), 1, 1, NULL
),
2413 _OPI(EXPP
, EX2
, V(2,0), V(3,0), V(0,0), V(0,0), 1, 1, NULL
),
2414 _OPI(LOGP
, LG2
, V(0,0), V(3,0), V(0,0), V(0,0), 1, 1, SPECIAL(LOG
)),
2415 _OPI(CND
, NOP
, V(0,0), V(0,0), V(0,0), V(1,4), 1, 3, SPECIAL(CND
)),
2417 _OPI(DEF
, NOP
, V(0,0), V(3,0), V(0,0), V(3,0), 1, 0, SPECIAL(DEF
)),
2419 /* More tex stuff */
2420 _OPI(TEXREG2RGB
, TEX
, V(0,0), V(0,0), V(1,2), V(1,3), 0, 0, SPECIAL(TEXREG2RGB
)),
2421 _OPI(TEXDP3TEX
, TEX
, V(0,0), V(0,0), V(1,2), V(1,3), 0, 0, SPECIAL(TEXDP3TEX
)),
2422 _OPI(TEXM3x2DEPTH
, TEX
, V(0,0), V(0,0), V(1,3), V(1,3), 0, 0, SPECIAL(TEXM3x2DEPTH
)),
2423 _OPI(TEXDP3
, TEX
, V(0,0), V(0,0), V(1,2), V(1,3), 0, 0, SPECIAL(TEXDP3
)),
2424 _OPI(TEXM3x3
, TEX
, V(0,0), V(0,0), V(1,2), V(1,3), 0, 0, SPECIAL(TEXM3x3
)),
2425 _OPI(TEXDEPTH
, TEX
, V(0,0), V(0,0), V(1,4), V(1,4), 0, 0, SPECIAL(TEXDEPTH
)),
2428 _OPI(CMP
, CMP
, V(0,0), V(0,0), V(1,2), V(3,0), 1, 3, SPECIAL(CMP
)), /* reversed */
2429 _OPI(BEM
, NOP
, V(0,0), V(0,0), V(1,4), V(1,4), 0, 0, SPECIAL(BEM
)),
2430 _OPI(DP2ADD
, NOP
, V(0,0), V(0,0), V(2,0), V(3,0), 1, 3, SPECIAL(DP2ADD
)),
2431 _OPI(DSX
, DDX
, V(0,0), V(0,0), V(2,1), V(3,0), 1, 1, NULL
),
2432 _OPI(DSY
, DDY
, V(0,0), V(0,0), V(2,1), V(3,0), 1, 1, NULL
),
2433 _OPI(TEXLDD
, TXD
, V(0,0), V(0,0), V(2,1), V(3,0), 1, 4, SPECIAL(TEXLDD
)),
2434 _OPI(SETP
, NOP
, V(0,0), V(3,0), V(2,1), V(3,0), 0, 0, SPECIAL(SETP
)),
2435 _OPI(TEXLDL
, TXL
, V(3,0), V(3,0), V(3,0), V(3,0), 1, 2, SPECIAL(TEXLDL
)),
2436 _OPI(BREAKP
, BRK
, V(0,0), V(3,0), V(2,1), V(3,0), 0, 0, SPECIAL(BREAKP
))
2439 struct sm1_op_info inst_phase
=
2440 _OPI(PHASE
, NOP
, V(0,0), V(0,0), V(1,4), V(1,4), 0, 0, SPECIAL(PHASE
));
2442 struct sm1_op_info inst_comment
=
2443 _OPI(COMMENT
, NOP
, V(0,0), V(3,0), V(0,0), V(3,0), 0, 0, SPECIAL(COMMENT
));
2446 create_op_info_map(struct shader_translator
*tx
)
2448 const unsigned version
= (tx
->version
.major
<< 8) | tx
->version
.minor
;
2451 for (i
= 0; i
< Elements(tx
->op_info_map
); ++i
)
2452 tx
->op_info_map
[i
] = -1;
2454 if (tx
->processor
== TGSI_PROCESSOR_VERTEX
) {
2455 for (i
= 0; i
< Elements(inst_table
); ++i
) {
2456 assert(inst_table
[i
].sio
< Elements(tx
->op_info_map
));
2457 if (inst_table
[i
].vert_version
.min
<= version
&&
2458 inst_table
[i
].vert_version
.max
>= version
)
2459 tx
->op_info_map
[inst_table
[i
].sio
] = i
;
2462 for (i
= 0; i
< Elements(inst_table
); ++i
) {
2463 assert(inst_table
[i
].sio
< Elements(tx
->op_info_map
));
2464 if (inst_table
[i
].frag_version
.min
<= version
&&
2465 inst_table
[i
].frag_version
.max
>= version
)
2466 tx
->op_info_map
[inst_table
[i
].sio
] = i
;
2471 static INLINE HRESULT
2472 NineTranslateInstruction_Generic(struct shader_translator
*tx
)
2474 struct ureg_dst dst
[1];
2475 struct ureg_src src
[4];
2478 for (i
= 0; i
< tx
->insn
.ndst
&& i
< Elements(dst
); ++i
)
2479 dst
[i
] = tx_dst_param(tx
, &tx
->insn
.dst
[i
]);
2480 for (i
= 0; i
< tx
->insn
.nsrc
&& i
< Elements(src
); ++i
)
2481 src
[i
] = tx_src_param(tx
, &tx
->insn
.src
[i
]);
2483 ureg_insn(tx
->ureg
, tx
->insn
.info
->opcode
,
2485 src
, tx
->insn
.nsrc
);
2490 TOKEN_PEEK(struct shader_translator
*tx
)
2492 return *(tx
->parse
);
2496 TOKEN_NEXT(struct shader_translator
*tx
)
2498 return *(tx
->parse
)++;
2502 TOKEN_JUMP(struct shader_translator
*tx
)
2504 if (tx
->parse_next
&& tx
->parse
!= tx
->parse_next
) {
2505 WARN("parse(%p) != parse_next(%p) !\n", tx
->parse
, tx
->parse_next
);
2506 tx
->parse
= tx
->parse_next
;
2510 static INLINE boolean
2511 sm1_parse_eof(struct shader_translator
*tx
)
2513 return TOKEN_PEEK(tx
) == NINED3DSP_END
;
2517 sm1_read_version(struct shader_translator
*tx
)
2519 const DWORD tok
= TOKEN_NEXT(tx
);
2521 tx
->version
.major
= D3DSHADER_VERSION_MAJOR(tok
);
2522 tx
->version
.minor
= D3DSHADER_VERSION_MINOR(tok
);
2524 switch (tok
>> 16) {
2525 case NINED3D_SM1_VS
: tx
->processor
= TGSI_PROCESSOR_VERTEX
; break;
2526 case NINED3D_SM1_PS
: tx
->processor
= TGSI_PROCESSOR_FRAGMENT
; break;
2528 DBG("Invalid shader type: %x\n", tok
);
2534 /* This is just to check if we parsed the instruction properly. */
2536 sm1_parse_get_skip(struct shader_translator
*tx
)
2538 const DWORD tok
= TOKEN_PEEK(tx
);
2540 if (tx
->version
.major
>= 2) {
2541 tx
->parse_next
= tx
->parse
+ 1 /* this */ +
2542 ((tok
& D3DSI_INSTLENGTH_MASK
) >> D3DSI_INSTLENGTH_SHIFT
);
2544 tx
->parse_next
= NULL
; /* TODO: determine from param count */
2549 sm1_print_comment(const char *comment
, UINT size
)
2557 sm1_parse_comments(struct shader_translator
*tx
, BOOL print
)
2559 DWORD tok
= TOKEN_PEEK(tx
);
2561 while ((tok
& D3DSI_OPCODE_MASK
) == D3DSIO_COMMENT
)
2563 const char *comment
= "";
2564 UINT size
= (tok
& D3DSI_COMMENTSIZE_MASK
) >> D3DSI_COMMENTSIZE_SHIFT
;
2565 tx
->parse
+= size
+ 1;
2568 sm1_print_comment(comment
, size
);
2570 tok
= TOKEN_PEEK(tx
);
2575 sm1_parse_get_param(struct shader_translator
*tx
, DWORD
*reg
, DWORD
*rel
)
2577 *reg
= TOKEN_NEXT(tx
);
2579 if (*reg
& D3DSHADER_ADDRMODE_RELATIVE
)
2581 if (tx
->version
.major
< 2)
2583 ((D3DSPR_ADDR
<< D3DSP_REGTYPE_SHIFT2
) & D3DSP_REGTYPE_MASK2
) |
2584 ((D3DSPR_ADDR
<< D3DSP_REGTYPE_SHIFT
) & D3DSP_REGTYPE_MASK
) |
2585 (D3DSP_NOSWIZZLE
<< D3DSP_SWIZZLE_SHIFT
);
2587 *rel
= TOKEN_NEXT(tx
);
2592 sm1_parse_dst_param(struct sm1_dst_param
*dst
, DWORD tok
)
2596 (tok
& D3DSP_REGTYPE_MASK
) >> D3DSP_REGTYPE_SHIFT
|
2597 (tok
& D3DSP_REGTYPE_MASK2
) >> D3DSP_REGTYPE_SHIFT2
;
2598 dst
->type
= TGSI_RETURN_TYPE_FLOAT
;
2599 dst
->idx
= tok
& D3DSP_REGNUM_MASK
;
2601 dst
->mask
= (tok
& NINED3DSP_WRITEMASK_MASK
) >> NINED3DSP_WRITEMASK_SHIFT
;
2602 dst
->mod
= (tok
& D3DSP_DSTMOD_MASK
) >> D3DSP_DSTMOD_SHIFT
;
2603 shift
= (tok
& D3DSP_DSTSHIFT_MASK
) >> D3DSP_DSTSHIFT_SHIFT
;
2604 dst
->shift
= (shift
& 0x8) ? -(shift
& 0x7) : shift
& 0x7;
2608 sm1_parse_src_param(struct sm1_src_param
*src
, DWORD tok
)
2611 ((tok
& D3DSP_REGTYPE_MASK
) >> D3DSP_REGTYPE_SHIFT
) |
2612 ((tok
& D3DSP_REGTYPE_MASK2
) >> D3DSP_REGTYPE_SHIFT2
);
2613 src
->type
= TGSI_RETURN_TYPE_FLOAT
;
2614 src
->idx
= tok
& D3DSP_REGNUM_MASK
;
2616 src
->swizzle
= (tok
& D3DSP_SWIZZLE_MASK
) >> D3DSP_SWIZZLE_SHIFT
;
2617 src
->mod
= (tok
& D3DSP_SRCMOD_MASK
) >> D3DSP_SRCMOD_SHIFT
;
2619 switch (src
->file
) {
2620 case D3DSPR_CONST2
: src
->file
= D3DSPR_CONST
; src
->idx
+= 2048; break;
2621 case D3DSPR_CONST3
: src
->file
= D3DSPR_CONST
; src
->idx
+= 4096; break;
2622 case D3DSPR_CONST4
: src
->file
= D3DSPR_CONST
; src
->idx
+= 6144; break;
2629 sm1_parse_immediate(struct shader_translator
*tx
,
2630 struct sm1_src_param
*imm
)
2632 imm
->file
= NINED3DSPR_IMMEDIATE
;
2635 imm
->swizzle
= NINED3DSP_NOSWIZZLE
;
2637 switch (tx
->insn
.opcode
) {
2639 imm
->type
= NINED3DSPTYPE_FLOAT4
;
2640 memcpy(&imm
->imm
.d
[0], tx
->parse
, 4 * sizeof(DWORD
));
2644 imm
->type
= NINED3DSPTYPE_INT4
;
2645 memcpy(&imm
->imm
.d
[0], tx
->parse
, 4 * sizeof(DWORD
));
2649 imm
->type
= NINED3DSPTYPE_BOOL
;
2650 memcpy(&imm
->imm
.d
[0], tx
->parse
, 1 * sizeof(DWORD
));
2660 sm1_read_dst_param(struct shader_translator
*tx
,
2661 struct sm1_dst_param
*dst
,
2662 struct sm1_src_param
*rel
)
2664 DWORD tok_dst
, tok_rel
= 0;
2666 sm1_parse_get_param(tx
, &tok_dst
, &tok_rel
);
2667 sm1_parse_dst_param(dst
, tok_dst
);
2668 if (tok_dst
& D3DSHADER_ADDRMODE_RELATIVE
) {
2669 sm1_parse_src_param(rel
, tok_rel
);
2675 sm1_read_src_param(struct shader_translator
*tx
,
2676 struct sm1_src_param
*src
,
2677 struct sm1_src_param
*rel
)
2679 DWORD tok_src
, tok_rel
= 0;
2681 sm1_parse_get_param(tx
, &tok_src
, &tok_rel
);
2682 sm1_parse_src_param(src
, tok_src
);
2683 if (tok_src
& D3DSHADER_ADDRMODE_RELATIVE
) {
2685 sm1_parse_src_param(rel
, tok_rel
);
2691 sm1_read_semantic(struct shader_translator
*tx
,
2692 struct sm1_semantic
*sem
)
2694 const DWORD tok_usg
= TOKEN_NEXT(tx
);
2695 const DWORD tok_dst
= TOKEN_NEXT(tx
);
2697 sem
->sampler_type
= (tok_usg
& D3DSP_TEXTURETYPE_MASK
) >> D3DSP_TEXTURETYPE_SHIFT
;
2698 sem
->usage
= (tok_usg
& D3DSP_DCL_USAGE_MASK
) >> D3DSP_DCL_USAGE_SHIFT
;
2699 sem
->usage_idx
= (tok_usg
& D3DSP_DCL_USAGEINDEX_MASK
) >> D3DSP_DCL_USAGEINDEX_SHIFT
;
2701 sm1_parse_dst_param(&sem
->reg
, tok_dst
);
2705 sm1_parse_instruction(struct shader_translator
*tx
)
2707 struct sm1_instruction
*insn
= &tx
->insn
;
2709 struct sm1_op_info
*info
= NULL
;
2712 sm1_parse_comments(tx
, TRUE
);
2713 sm1_parse_get_skip(tx
);
2715 tok
= TOKEN_NEXT(tx
);
2717 insn
->opcode
= tok
& D3DSI_OPCODE_MASK
;
2718 insn
->flags
= (tok
& NINED3DSIO_OPCODE_FLAGS_MASK
) >> NINED3DSIO_OPCODE_FLAGS_SHIFT
;
2719 insn
->coissue
= !!(tok
& D3DSI_COISSUE
);
2720 insn
->predicated
= !!(tok
& NINED3DSHADER_INST_PREDICATED
);
2722 if (insn
->opcode
< Elements(tx
->op_info_map
)) {
2723 int k
= tx
->op_info_map
[insn
->opcode
];
2725 assert(k
< Elements(inst_table
));
2726 info
= &inst_table
[k
];
2729 if (insn
->opcode
== D3DSIO_PHASE
) info
= &inst_phase
;
2730 if (insn
->opcode
== D3DSIO_COMMENT
) info
= &inst_comment
;
2733 DBG("illegal or unhandled opcode: %08x\n", insn
->opcode
);
2738 insn
->ndst
= info
->ndst
;
2739 insn
->nsrc
= info
->nsrc
;
2741 assert(!insn
->predicated
&& "TODO: predicated instructions");
2745 unsigned min
= IS_VS
? info
->vert_version
.min
: info
->frag_version
.min
;
2746 unsigned max
= IS_VS
? info
->vert_version
.max
: info
->frag_version
.max
;
2747 unsigned ver
= (tx
->version
.major
<< 8) | tx
->version
.minor
;
2748 if (ver
< min
|| ver
> max
) {
2749 DBG("opcode not supported in this shader version: %x <= %x <= %x\n",
2755 for (i
= 0; i
< insn
->ndst
; ++i
)
2756 sm1_read_dst_param(tx
, &insn
->dst
[i
], &insn
->dst_rel
[i
]);
2757 if (insn
->predicated
)
2758 sm1_read_src_param(tx
, &insn
->pred
, NULL
);
2759 for (i
= 0; i
< insn
->nsrc
; ++i
)
2760 sm1_read_src_param(tx
, &insn
->src
[i
], &insn
->src_rel
[i
]);
2762 /* parse here so we can dump them before processing */
2763 if (insn
->opcode
== D3DSIO_DEF
||
2764 insn
->opcode
== D3DSIO_DEFI
||
2765 insn
->opcode
== D3DSIO_DEFB
)
2766 sm1_parse_immediate(tx
, &tx
->insn
.src
[0]);
2768 sm1_dump_instruction(insn
, tx
->cond_depth
+ tx
->loop_depth
);
2769 sm1_instruction_check(insn
);
2774 NineTranslateInstruction_Generic(tx
);
2775 tx_apply_dst0_modifiers(tx
);
2777 tx
->num_scratch
= 0; /* reset */
2783 tx_ctor(struct shader_translator
*tx
, struct nine_shader_info
*info
)
2789 tx
->byte_code
= info
->byte_code
;
2790 tx
->parse
= info
->byte_code
;
2792 for (i
= 0; i
< Elements(info
->input_map
); ++i
)
2793 info
->input_map
[i
] = NINE_DECLUSAGE_NONE
;
2794 info
->num_inputs
= 0;
2796 info
->position_t
= FALSE
;
2797 info
->point_size
= FALSE
;
2799 tx
->info
->const_used_size
= 0;
2801 info
->sampler_mask
= 0x0;
2802 info
->rt_mask
= 0x0;
2804 info
->lconstf
.data
= NULL
;
2805 info
->lconstf
.ranges
= NULL
;
2807 for (i
= 0; i
< Elements(tx
->regs
.rL
); ++i
) {
2808 tx
->regs
.rL
[i
] = ureg_dst_undef();
2810 tx
->regs
.address
= ureg_dst_undef();
2811 tx
->regs
.a0
= ureg_dst_undef();
2812 tx
->regs
.p
= ureg_dst_undef();
2813 tx
->regs
.oDepth
= ureg_dst_undef();
2814 tx
->regs
.vPos
= ureg_src_undef();
2815 tx
->regs
.vFace
= ureg_src_undef();
2816 for (i
= 0; i
< Elements(tx
->regs
.o
); ++i
)
2817 tx
->regs
.o
[i
] = ureg_dst_undef();
2818 for (i
= 0; i
< Elements(tx
->regs
.oCol
); ++i
)
2819 tx
->regs
.oCol
[i
] = ureg_dst_undef();
2820 for (i
= 0; i
< Elements(tx
->regs
.vC
); ++i
)
2821 tx
->regs
.vC
[i
] = ureg_src_undef();
2822 for (i
= 0; i
< Elements(tx
->regs
.vT
); ++i
)
2823 tx
->regs
.vT
[i
] = ureg_src_undef();
2825 for (i
= 0; i
< Elements(tx
->lconsti
); ++i
)
2826 tx
->lconsti
[i
].idx
= -1;
2827 for (i
= 0; i
< Elements(tx
->lconstb
); ++i
)
2828 tx
->lconstb
[i
].idx
= -1;
2830 sm1_read_version(tx
);
2832 info
->version
= (tx
->version
.major
<< 4) | tx
->version
.minor
;
2834 create_op_info_map(tx
);
2838 tx_dtor(struct shader_translator
*tx
)
2840 if (tx
->num_inst_labels
)
2841 FREE(tx
->inst_labels
);
2847 static INLINE
unsigned
2848 tgsi_processor_from_type(unsigned shader_type
)
2850 switch (shader_type
) {
2851 case PIPE_SHADER_VERTEX
: return TGSI_PROCESSOR_VERTEX
;
2852 case PIPE_SHADER_FRAGMENT
: return TGSI_PROCESSOR_FRAGMENT
;
2858 #define GET_CAP(n) device->screen->get_param( \
2859 device->screen, PIPE_CAP_##n)
2860 #define GET_SHADER_CAP(n) device->screen->get_shader_param( \
2861 device->screen, info->type, PIPE_SHADER_CAP_##n)
2864 nine_translate_shader(struct NineDevice9
*device
, struct nine_shader_info
*info
)
2866 struct shader_translator
*tx
;
2867 HRESULT hr
= D3D_OK
;
2868 const unsigned processor
= tgsi_processor_from_type(info
->type
);
2870 user_assert(processor
!= ~0, D3DERR_INVALIDCALL
);
2872 tx
= CALLOC_STRUCT(shader_translator
);
2874 return E_OUTOFMEMORY
;
2877 if (((tx
->version
.major
<< 16) | tx
->version
.minor
) > 0x00030000) {
2878 hr
= D3DERR_INVALIDCALL
;
2879 DBG("Unsupported shader version: %u.%u !\n",
2880 tx
->version
.major
, tx
->version
.minor
);
2883 if (tx
->processor
!= processor
) {
2884 hr
= D3DERR_INVALIDCALL
;
2885 DBG("Shader type mismatch: %u / %u !\n", tx
->processor
, processor
);
2888 DUMP("%s%u.%u\n", processor
== TGSI_PROCESSOR_VERTEX
? "VS" : "PS",
2889 tx
->version
.major
, tx
->version
.minor
);
2891 tx
->ureg
= ureg_create(processor
);
2896 tx_decl_constants(tx
);
2898 tx
->native_integers
= GET_SHADER_CAP(INTEGERS
);
2899 tx
->inline_subroutines
= !GET_SHADER_CAP(SUBROUTINES
);
2900 tx
->lower_preds
= !GET_SHADER_CAP(MAX_PREDS
);
2901 tx
->want_texcoord
= GET_CAP(TGSI_TEXCOORD
);
2902 tx
->shift_wpos
= !GET_CAP(TGSI_FS_COORD_PIXEL_CENTER_INTEGER
);
2903 tx
->texcoord_sn
= tx
->want_texcoord
?
2904 TGSI_SEMANTIC_TEXCOORD
: TGSI_SEMANTIC_GENERIC
;
2906 /* VS must always write position. Declare it here to make it the 1st output.
2907 * (Some drivers like nv50 are buggy and rely on that.)
2910 tx
->regs
.oPos
= ureg_DECL_output(tx
->ureg
, TGSI_SEMANTIC_POSITION
, 0);
2912 ureg_property(tx
->ureg
, TGSI_PROPERTY_FS_COORD_ORIGIN
, TGSI_FS_COORD_ORIGIN_UPPER_LEFT
);
2913 if (!tx
->shift_wpos
)
2914 ureg_property(tx
->ureg
, TGSI_PROPERTY_FS_COORD_PIXEL_CENTER
, TGSI_FS_COORD_PIXEL_CENTER_INTEGER
);
2917 while (!sm1_parse_eof(tx
))
2918 sm1_parse_instruction(tx
);
2919 tx
->parse
++; /* for byte_size */
2921 if (IS_PS
&& (tx
->version
.major
< 2) && tx
->num_temp
) {
2922 ureg_MOV(tx
->ureg
, ureg_DECL_output(tx
->ureg
, TGSI_SEMANTIC_COLOR
, 0),
2923 ureg_src(tx
->regs
.r
[0]));
2924 info
->rt_mask
|= 0x1;
2927 if (info
->position_t
)
2928 ureg_property(tx
->ureg
, TGSI_PROPERTY_VS_WINDOW_SPACE_POSITION
, TRUE
);
2932 if (IS_VS
&& !ureg_dst_is_undef(tx
->regs
.oPts
))
2933 info
->point_size
= TRUE
;
2935 if (debug_get_bool_option("NINE_TGSI_DUMP", FALSE
)) {
2937 const struct tgsi_token
*toks
= ureg_get_tokens(tx
->ureg
, &count
);
2939 ureg_free_tokens(toks
);
2942 /* record local constants */
2943 if (tx
->num_lconstf
&& tx
->indirect_const_access
) {
2944 struct nine_range
*ranges
;
2951 data
= MALLOC(tx
->num_lconstf
* 4 * sizeof(float));
2954 info
->lconstf
.data
= data
;
2956 indices
= MALLOC(tx
->num_lconstf
* sizeof(indices
[0]));
2960 /* lazy sort, num_lconstf should be small */
2961 for (n
= 0; n
< tx
->num_lconstf
; ++n
) {
2962 for (k
= 0, i
= 0; i
< tx
->num_lconstf
; ++i
) {
2963 if (tx
->lconstf
[i
].idx
< tx
->lconstf
[k
].idx
)
2966 indices
[n
] = tx
->lconstf
[k
].idx
;
2967 memcpy(&data
[n
* 4], &tx
->lconstf
[k
].imm
.f
[0], 4 * sizeof(float));
2968 tx
->lconstf
[k
].idx
= INT_MAX
;
2972 for (n
= 1, i
= 1; i
< tx
->num_lconstf
; ++i
)
2973 if (indices
[i
] != indices
[i
- 1] + 1)
2975 ranges
= MALLOC(n
* sizeof(ranges
[0]));
2980 info
->lconstf
.ranges
= ranges
;
2983 ranges
[k
].bgn
= indices
[0];
2984 for (i
= 1; i
< tx
->num_lconstf
; ++i
) {
2985 if (indices
[i
] != indices
[i
- 1] + 1) {
2986 ranges
[k
].next
= &ranges
[k
+ 1];
2987 ranges
[k
].end
= indices
[i
- 1] + 1;
2989 ranges
[k
].bgn
= indices
[i
];
2992 ranges
[k
].end
= indices
[i
- 1] + 1;
2993 ranges
[k
].next
= NULL
;
2994 assert(n
== (k
+ 1));
3000 if (tx
->indirect_const_access
)
3001 info
->const_used_size
= ~0;
3003 info
->cso
= ureg_create_shader_and_destroy(tx
->ureg
, device
->pipe
);
3005 hr
= D3DERR_DRIVERINTERNALERROR
;
3006 FREE(info
->lconstf
.data
);
3007 FREE(info
->lconstf
.ranges
);
3011 info
->byte_size
= (tx
->parse
- tx
->byte_code
) * sizeof(DWORD
);