2 * Copyright 2016 Paul Gofman
4 * This library is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU Lesser General Public
6 * License as published by the Free Software Foundation; either
7 * version 2.1 of the License, or (at your option) any later version.
9 * This library is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 * Lesser General Public License for more details.
14 * You should have received a copy of the GNU Lesser General Public
15 * License along with this library; if not, write to the Free Software
16 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
20 #include "d3dx9_private.h"
26 WINE_DEFAULT_DEBUG_CHANNEL(d3dx
);
53 PRESHADER_OP_DOTSWIZ6
,
54 PRESHADER_OP_DOTSWIZ8
,
57 typedef double (*pres_op_func
)(double *args
, int n
);
59 static double to_signed_nan(double v
)
63 ULONG64 ulong64_value
;
71 return isnan(v
) ? signed_nan
.double_value
: v
;
74 static double pres_mov(double *args
, int n
) {return args
[0];}
75 static double pres_add(double *args
, int n
) {return args
[0] + args
[1];}
76 static double pres_mul(double *args
, int n
) {return args
[0] * args
[1];}
77 static double pres_dot(double *args
, int n
)
83 for (i
= 0; i
< n
; ++i
)
84 sum
+= args
[i
] * args
[i
+ n
];
88 static double pres_dotswiz6(double *args
, int n
)
90 return pres_dot(args
, 3);
93 static double pres_dotswiz8(double *args
, int n
)
95 return pres_dot(args
, 4);
98 static double pres_neg(double *args
, int n
) {return -args
[0];}
99 static double pres_rcp(double *args
, int n
) {return 1.0 / args
[0];}
100 static double pres_lt(double *args
, int n
) {return args
[0] < args
[1] ? 1.0 : 0.0;}
101 static double pres_ge(double *args
, int n
) {return args
[0] >= args
[1] ? 1.0 : 0.0;}
102 static double pres_frc(double *args
, int n
) {return args
[0] - floor(args
[0]);}
103 static double pres_min(double *args
, int n
) {return fmin(args
[0], args
[1]);}
104 static double pres_max(double *args
, int n
) {return fmax(args
[0], args
[1]);}
105 static double pres_cmp(double *args
, int n
) {return args
[0] >= 0.0 ? args
[1] : args
[2];}
106 static double pres_sin(double *args
, int n
) {return sin(args
[0]);}
107 static double pres_cos(double *args
, int n
) {return cos(args
[0]);}
108 static double pres_rsq(double *args
, int n
)
116 return 1.0 / sqrt(v
);
118 static double pres_exp(double *args
, int n
) {return pow(2.0, args
[0]);}
119 static double pres_log(double *args
, int n
)
129 static double pres_asin(double *args
, int n
) {return to_signed_nan(asin(args
[0]));}
130 static double pres_acos(double *args
, int n
) {return to_signed_nan(acos(args
[0]));}
131 static double pres_atan(double *args
, int n
) {return atan(args
[0]);}
132 static double pres_atan2(double *args
, int n
) {return atan2(args
[0], args
[1]);}
134 /* According to the test results 'div' operation always returns 0. Compiler does not seem to ever
135 * generate it, using rcp + mul instead, so probably it is not implemented in native d3dx. */
136 static double pres_div(double *args
, int n
) {return 0.0;}
138 #define PRES_OPCODE_MASK 0x7ff00000
139 #define PRES_OPCODE_SHIFT 20
140 #define PRES_SCALAR_FLAG 0x80000000
141 #define PRES_NCOMP_MASK 0x0000ffff
143 #define FOURCC_PRES 0x53455250
144 #define FOURCC_CLIT 0x54494c43
145 #define FOURCC_FXLC 0x434c5846
146 #define FOURCC_PRSI 0x49535250
147 #define PRES_SIGN 0x46580000
153 unsigned int input_count
;
158 static const struct op_info pres_op_info
[] =
160 {0x000, "nop", 0, 0, NULL
}, /* PRESHADER_OP_NOP */
161 {0x100, "mov", 1, 0, pres_mov
}, /* PRESHADER_OP_MOV */
162 {0x101, "neg", 1, 0, pres_neg
}, /* PRESHADER_OP_NEG */
163 {0x103, "rcp", 1, 0, pres_rcp
}, /* PRESHADER_OP_RCP */
164 {0x104, "frc", 1, 0, pres_frc
}, /* PRESHADER_OP_FRC */
165 {0x105, "exp", 1, 0, pres_exp
}, /* PRESHADER_OP_EXP */
166 {0x106, "log", 1, 0, pres_log
}, /* PRESHADER_OP_LOG */
167 {0x107, "rsq", 1, 0, pres_rsq
}, /* PRESHADER_OP_RSQ */
168 {0x108, "sin", 1, 0, pres_sin
}, /* PRESHADER_OP_SIN */
169 {0x109, "cos", 1, 0, pres_cos
}, /* PRESHADER_OP_COS */
170 {0x10a, "asin", 1, 0, pres_asin
}, /* PRESHADER_OP_ASIN */
171 {0x10b, "acos", 1, 0, pres_acos
}, /* PRESHADER_OP_ACOS */
172 {0x10c, "atan", 1, 0, pres_atan
}, /* PRESHADER_OP_ATAN */
173 {0x200, "min", 2, 0, pres_min
}, /* PRESHADER_OP_MIN */
174 {0x201, "max", 2, 0, pres_max
}, /* PRESHADER_OP_MAX */
175 {0x202, "lt", 2, 0, pres_lt
}, /* PRESHADER_OP_LT */
176 {0x203, "ge", 2, 0, pres_ge
}, /* PRESHADER_OP_GE */
177 {0x204, "add", 2, 0, pres_add
}, /* PRESHADER_OP_ADD */
178 {0x205, "mul", 2, 0, pres_mul
}, /* PRESHADER_OP_MUL */
179 {0x206, "atan2", 2, 0, pres_atan2
}, /* PRESHADER_OP_ATAN2 */
180 {0x208, "div", 2, 0, pres_div
}, /* PRESHADER_OP_DIV */
181 {0x300, "cmp", 3, 0, pres_cmp
}, /* PRESHADER_OP_CMP */
182 {0x500, "dot", 2, 1, pres_dot
}, /* PRESHADER_OP_DOT */
183 {0x70e, "d3ds_dotswiz", 6, 0, pres_dotswiz6
}, /* PRESHADER_OP_DOTSWIZ6 */
184 {0x70e, "d3ds_dotswiz", 8, 0, pres_dotswiz8
}, /* PRESHADER_OP_DOTSWIZ8 */
198 unsigned int component_size
;
199 enum pres_value_type type
;
203 {sizeof(double), PRES_VT_DOUBLE
}, /* PRES_REGTAB_IMMED */
204 {sizeof(float), PRES_VT_FLOAT
}, /* PRES_REGTAB_CONST */
205 {sizeof(float), PRES_VT_FLOAT
}, /* PRES_REGTAB_INPUT */
206 {sizeof(float), PRES_VT_FLOAT
}, /* PRES_REGTAB_OCONST */
207 {sizeof(BOOL
), PRES_VT_BOOL
}, /* PRES_REGTAB_OBCONST */
208 {sizeof(int), PRES_VT_INT
, }, /* PRES_REGTAB_OICONST */
209 /* TODO: use double precision for 64 bit */
210 {sizeof(float), PRES_VT_FLOAT
} /* PRES_REGTAB_TEMP */
213 static const char *table_symbol
[] =
215 "imm", "c", "v", "oc", "ob", "oi", "r", "(null)",
218 static const enum pres_reg_tables pres_regset2table
[] =
220 PRES_REGTAB_OBCONST
, /* D3DXRS_BOOL */
221 PRES_REGTAB_OICONST
, /* D3DXRS_INT4 */
222 PRES_REGTAB_CONST
, /* D3DXRS_FLOAT4 */
223 PRES_REGTAB_COUNT
, /* D3DXRS_SAMPLER */
226 static const enum pres_reg_tables shad_regset2table
[] =
228 PRES_REGTAB_OBCONST
, /* D3DXRS_BOOL */
229 PRES_REGTAB_OICONST
, /* D3DXRS_INT4 */
230 PRES_REGTAB_OCONST
, /* D3DXRS_FLOAT4 */
231 PRES_REGTAB_COUNT
, /* D3DXRS_SAMPLER */
236 enum pres_reg_tables table
;
237 /* offset is component index, not register index, e. g.
238 offset for component c3.y is 13 (3 * 4 + 1) */
242 struct d3dx_pres_operand
244 struct d3dx_pres_reg reg
;
245 struct d3dx_pres_reg index_reg
;
248 #define MAX_INPUTS_COUNT 8
253 /* first input argument is scalar,
254 scalar component is propagated */
256 unsigned int component_count
;
257 struct d3dx_pres_operand inputs
[MAX_INPUTS_COUNT
];
258 struct d3dx_pres_operand output
;
261 struct const_upload_info
264 unsigned int major
, minor
;
265 unsigned int major_stride
;
266 unsigned int major_count
;
268 unsigned int minor_remainder
;
271 static enum pres_value_type
table_type_from_param_type(D3DXPARAMETER_TYPE type
)
276 return PRES_VT_FLOAT
;
282 FIXME("Unsupported type %u.\n", type
);
283 return PRES_VT_COUNT
;
287 static unsigned int get_reg_offset(unsigned int table
, unsigned int offset
)
289 return table
== PRES_REGTAB_OBCONST
? offset
: offset
>> 2;
292 static unsigned int get_offset_reg(unsigned int table
, unsigned int reg_idx
)
294 return table
== PRES_REGTAB_OBCONST
? reg_idx
: reg_idx
<< 2;
297 static unsigned int get_reg_components(unsigned int table
)
299 return get_offset_reg(table
, 1);
302 #define PRES_BITMASK_BLOCK_SIZE (sizeof(unsigned int) * 8)
304 static HRESULT
regstore_alloc_table(struct d3dx_regstore
*rs
, unsigned int table
)
308 size
= get_offset_reg(table
, rs
->table_sizes
[table
]) * table_info
[table
].component_size
;
311 rs
->tables
[table
] = calloc(1, size
);
312 if (!rs
->tables
[table
])
313 return E_OUTOFMEMORY
;
318 static void regstore_free_tables(struct d3dx_regstore
*rs
)
322 for (i
= 0; i
< PRES_REGTAB_COUNT
; ++i
)
328 static void regstore_set_values(struct d3dx_regstore
*rs
, unsigned int table
, const void *data
,
329 unsigned int start_offset
, unsigned int count
)
331 BYTE
*dst
= rs
->tables
[table
];
332 const BYTE
*src
= data
;
335 dst
+= start_offset
* table_info
[table
].component_size
;
336 size
= count
* table_info
[table
].component_size
;
337 assert((src
< dst
&& size
<= dst
- src
) || (src
> dst
&& size
<= src
- dst
));
338 memcpy(dst
, src
, size
);
341 static double regstore_get_double(struct d3dx_regstore
*rs
, unsigned int table
, unsigned int offset
)
345 p
= (BYTE
*)rs
->tables
[table
] + table_info
[table
].component_size
* offset
;
346 switch (table_info
[table
].type
)
353 FIXME("Unexpected preshader input from table %u.\n", table
);
358 static void regstore_set_double(struct d3dx_regstore
*rs
, unsigned int table
, unsigned int offset
, double v
)
362 p
= (BYTE
*)rs
->tables
[table
] + table_info
[table
].component_size
* offset
;
363 switch (table_info
[table
].type
)
365 case PRES_VT_FLOAT
: *(float *)p
= v
; break;
366 case PRES_VT_DOUBLE
: *(double *)p
= v
; break;
367 case PRES_VT_INT
: *(int *)p
= lrint(v
); break;
368 case PRES_VT_BOOL
: *(BOOL
*)p
= !!v
; break;
370 FIXME("Bad type %u.\n", table_info
[table
].type
);
375 static void dump_bytecode(void *data
, unsigned int size
)
377 unsigned int *bytecode
= (unsigned int *)data
;
378 unsigned int i
, j
, n
;
380 size
/= sizeof(*bytecode
);
384 n
= min(size
- i
, 8);
385 for (j
= 0; j
< n
; ++j
)
386 TRACE("0x%08x,", bytecode
[i
+ j
]);
392 static unsigned int *find_bytecode_comment(unsigned int *ptr
, unsigned int count
,
393 unsigned int fourcc
, unsigned int *size
)
395 /* Provide at least one value in comment section on non-NULL return. */
396 while (count
> 2 && (*ptr
& 0xffff) == 0xfffe)
398 unsigned int section_size
;
400 section_size
= (*ptr
>> 16);
401 if (!section_size
|| section_size
+ 1 > count
)
403 if (*(ptr
+ 1) == fourcc
)
405 *size
= section_size
;
408 count
-= section_size
+ 1;
409 ptr
+= section_size
+ 1;
414 static unsigned int *parse_pres_reg(unsigned int *ptr
, struct d3dx_pres_reg
*reg
)
416 static const enum pres_reg_tables reg_table
[8] =
418 PRES_REGTAB_COUNT
, PRES_REGTAB_IMMED
, PRES_REGTAB_CONST
, PRES_REGTAB_INPUT
,
419 PRES_REGTAB_OCONST
, PRES_REGTAB_OBCONST
, PRES_REGTAB_OICONST
, PRES_REGTAB_TEMP
422 if (*ptr
>= ARRAY_SIZE(reg_table
) || reg_table
[*ptr
] == PRES_REGTAB_COUNT
)
424 FIXME("Unsupported register table %#x.\n", *ptr
);
428 reg
->table
= reg_table
[*ptr
++];
429 reg
->offset
= *ptr
++;
433 static unsigned int *parse_pres_arg(unsigned int *ptr
, unsigned int count
, struct d3dx_pres_operand
*opr
)
435 if (count
< 3 || (*ptr
&& count
< 5))
437 WARN("Byte code buffer ends unexpectedly, count %u.\n", count
);
445 FIXME("Unknown relative addressing flag, word %#x.\n", *ptr
);
448 ptr
= parse_pres_reg(ptr
+ 1, &opr
->index_reg
);
454 opr
->index_reg
.table
= PRES_REGTAB_COUNT
;
458 ptr
= parse_pres_reg(ptr
, &opr
->reg
);
460 if (opr
->reg
.table
== PRES_REGTAB_OBCONST
)
461 opr
->reg
.offset
/= 4;
465 static unsigned int *parse_pres_ins(unsigned int *ptr
, unsigned int count
, struct d3dx_pres_ins
*ins
)
467 unsigned int ins_code
, ins_raw
;
468 unsigned int input_count
;
473 WARN("Byte code buffer ends unexpectedly.\n");
478 ins_code
= (ins_raw
& PRES_OPCODE_MASK
) >> PRES_OPCODE_SHIFT
;
479 ins
->component_count
= ins_raw
& PRES_NCOMP_MASK
;
480 ins
->scalar_op
= !!(ins_raw
& PRES_SCALAR_FLAG
);
482 if (ins
->component_count
< 1 || ins
->component_count
> 4)
484 FIXME("Unsupported number of components %u.\n", ins
->component_count
);
487 input_count
= *ptr
++;
489 for (i
= 0; i
< ARRAY_SIZE(pres_op_info
); ++i
)
490 if (ins_code
== pres_op_info
[i
].opcode
&& input_count
== pres_op_info
[i
].input_count
)
492 if (i
== ARRAY_SIZE(pres_op_info
))
494 FIXME("Unknown opcode %#x, input_count %u, raw %#x.\n", ins_code
, input_count
, ins_raw
);
498 if (input_count
> ARRAY_SIZE(ins
->inputs
))
500 FIXME("Actual input args count %u exceeds inputs array size, instruction %s.\n", input_count
,
501 pres_op_info
[i
].mnem
);
504 for (i
= 0; i
< input_count
; ++i
)
508 p
= parse_pres_arg(ptr
, count
, &ins
->inputs
[i
]);
514 ptr
= parse_pres_arg(ptr
, count
, &ins
->output
);
515 if (ins
->output
.index_reg
.table
!= PRES_REGTAB_COUNT
)
517 FIXME("Relative addressing in output register not supported.\n");
520 if (get_reg_offset(ins
->output
.reg
.table
, ins
->output
.reg
.offset
521 + (pres_op_info
[ins
->op
].func_all_comps
? 0 : ins
->component_count
- 1))
522 != get_reg_offset(ins
->output
.reg
.table
, ins
->output
.reg
.offset
))
524 FIXME("Instructions outputting multiple registers are not supported.\n");
530 static HRESULT
get_ctab_constant_desc(ID3DXConstantTable
*ctab
, D3DXHANDLE hc
, D3DXCONSTANT_DESC
*desc
,
531 WORD
*constantinfo_reserved
)
533 const struct ctab_constant
*constant
= d3dx_shader_get_ctab_constant(ctab
, hc
);
537 FIXME("Could not get constant desc.\n");
538 if (constantinfo_reserved
)
539 *constantinfo_reserved
= 0;
540 return D3DERR_INVALIDCALL
;
542 *desc
= constant
->desc
;
543 if (constantinfo_reserved
)
544 *constantinfo_reserved
= constant
->constantinfo_reserved
;
548 static void get_const_upload_info(struct d3dx_const_param_eval_output
*const_set
,
549 struct const_upload_info
*info
)
551 struct d3dx_parameter
*param
= const_set
->param
;
552 unsigned int table
= const_set
->table
;
554 info
->transpose
= (const_set
->constant_class
== D3DXPC_MATRIX_COLUMNS
&& param
->class == D3DXPC_MATRIX_ROWS
)
555 || (param
->class == D3DXPC_MATRIX_COLUMNS
&& const_set
->constant_class
== D3DXPC_MATRIX_ROWS
);
556 if (const_set
->constant_class
== D3DXPC_MATRIX_COLUMNS
)
558 info
->major
= param
->columns
;
559 info
->minor
= param
->rows
;
563 info
->major
= param
->rows
;
564 info
->minor
= param
->columns
;
567 if (get_reg_components(table
) == 1)
569 unsigned int const_length
= get_offset_reg(table
, const_set
->register_count
);
571 info
->major_stride
= info
->minor
;
572 info
->major_count
= const_length
/ info
->major_stride
;
573 info
->minor_remainder
= const_length
% info
->major_stride
;
577 info
->major_stride
= get_reg_components(table
);
578 info
->major_count
= const_set
->register_count
;
579 info
->minor_remainder
= 0;
581 info
->count
= info
->major_count
* info
->minor
+ info
->minor_remainder
;
584 #define INITIAL_CONST_SET_SIZE 16
586 static HRESULT
append_const_set(struct d3dx_const_tab
*const_tab
, struct d3dx_const_param_eval_output
*set
)
588 if (const_tab
->const_set_count
>= const_tab
->const_set_size
)
590 unsigned int new_size
;
591 struct d3dx_const_param_eval_output
*new_alloc
;
593 if (!const_tab
->const_set_size
)
595 new_size
= INITIAL_CONST_SET_SIZE
;
596 new_alloc
= malloc(sizeof(*const_tab
->const_set
) * new_size
);
599 ERR("Out of memory.\n");
600 return E_OUTOFMEMORY
;
605 new_size
= const_tab
->const_set_size
* 2;
606 new_alloc
= realloc(const_tab
->const_set
, sizeof(*const_tab
->const_set
) * new_size
);
609 ERR("Out of memory.\n");
610 return E_OUTOFMEMORY
;
613 const_tab
->const_set
= new_alloc
;
614 const_tab
->const_set_size
= new_size
;
616 const_tab
->const_set
[const_tab
->const_set_count
++] = *set
;
620 static void append_pres_const_sets_for_shader_input(struct d3dx_const_tab
*const_tab
,
621 struct d3dx_preshader
*pres
)
624 struct d3dx_const_param_eval_output const_set
= {NULL
};
626 for (i
= 0; i
< pres
->ins_count
; ++i
)
628 const struct d3dx_pres_ins
*ins
= &pres
->ins
[i
];
629 const struct d3dx_pres_reg
*reg
= &ins
->output
.reg
;
631 if (reg
->table
== PRES_REGTAB_TEMP
)
634 const_set
.register_index
= get_reg_offset(reg
->table
, reg
->offset
);
635 const_set
.register_count
= 1;
636 const_set
.table
= reg
->table
;
637 const_set
.constant_class
= D3DXPC_FORCE_DWORD
;
638 const_set
.element_count
= 1;
639 append_const_set(const_tab
, &const_set
);
643 static int __cdecl
compare_const_set(const void *a
, const void *b
)
645 const struct d3dx_const_param_eval_output
*r1
= a
;
646 const struct d3dx_const_param_eval_output
*r2
= b
;
648 if (r1
->table
!= r2
->table
)
649 return r1
->table
- r2
->table
;
650 return r1
->register_index
- r2
->register_index
;
653 static HRESULT
merge_const_set_entries(struct d3dx_const_tab
*const_tab
,
654 struct d3dx_parameter
*param
, unsigned int index
)
656 unsigned int i
, start_index
= index
;
658 enum pres_reg_tables current_table
;
659 unsigned int current_start_offset
, element_count
;
660 struct d3dx_const_param_eval_output
*first_const
;
662 if (!const_tab
->const_set_count
)
665 while (index
< const_tab
->const_set_count
- 1)
667 first_const
= &const_tab
->const_set
[index
];
668 current_data
= first_const
->param
->data
;
669 current_table
= first_const
->table
;
670 current_start_offset
= get_offset_reg(current_table
, first_const
->register_index
);
672 for (i
= index
; i
< const_tab
->const_set_count
; ++i
)
674 struct d3dx_const_param_eval_output
*const_set
= &const_tab
->const_set
[i
];
675 unsigned int count
= get_offset_reg(const_set
->table
,
676 const_set
->register_count
* const_set
->element_count
);
677 unsigned int start_offset
= get_offset_reg(const_set
->table
, const_set
->register_index
);
679 if (!(const_set
->table
== current_table
&& current_start_offset
== start_offset
680 && const_set
->direct_copy
== first_const
->direct_copy
681 && current_data
== const_set
->param
->data
682 && (const_set
->direct_copy
|| (first_const
->param
->type
== const_set
->param
->type
683 && first_const
->param
->class == const_set
->param
->class
684 && first_const
->param
->columns
== const_set
->param
->columns
685 && first_const
->param
->rows
== const_set
->param
->rows
686 && first_const
->register_count
== const_set
->register_count
687 && (i
== const_tab
->const_set_count
- 1
688 || first_const
->param
->element_count
== const_set
->param
->element_count
)))))
691 current_start_offset
+= count
;
692 current_data
+= const_set
->direct_copy
? count
: const_set
->param
->rows
693 * const_set
->param
->columns
* const_set
->element_count
;
694 element_count
+= const_set
->element_count
;
699 TRACE("Merging %u child parameters for %s, not merging %u, direct_copy %#x.\n", i
- index
,
700 debugstr_a(param
->name
), const_tab
->const_set_count
- i
, first_const
->direct_copy
);
702 first_const
->element_count
= element_count
;
703 if (first_const
->direct_copy
)
705 first_const
->element_count
= 1;
706 if (index
== start_index
707 && !(param
->type
== D3DXPT_VOID
&& param
->class == D3DXPC_STRUCT
))
709 if (table_type_from_param_type(param
->type
) == PRES_VT_COUNT
)
710 return D3DERR_INVALIDCALL
;
711 first_const
->param
= param
;
713 first_const
->register_count
= get_reg_offset(current_table
, current_start_offset
)
714 - first_const
->register_index
;
716 memmove(&const_tab
->const_set
[index
+ 1], &const_tab
->const_set
[i
],
717 sizeof(*const_tab
->const_set
) * (const_tab
->const_set_count
- i
));
718 const_tab
->const_set_count
-= i
- index
- 1;
722 TRACE("Not merging %u child parameters for %s, direct_copy %#x.\n",
723 const_tab
->const_set_count
- i
, debugstr_a(param
->name
), first_const
->direct_copy
);
730 static HRESULT
init_set_constants_param(struct d3dx_const_tab
*const_tab
, ID3DXConstantTable
*ctab
,
731 D3DXHANDLE hc
, struct d3dx_parameter
*param
)
733 D3DXCONSTANT_DESC desc
;
734 unsigned int const_count
, param_count
, i
;
736 struct d3dx_const_param_eval_output const_set
;
737 struct const_upload_info info
;
738 enum pres_value_type table_type
;
741 if (FAILED(get_ctab_constant_desc(ctab
, hc
, &desc
, NULL
)))
742 return D3DERR_INVALIDCALL
;
744 if (param
->element_count
)
746 param_count
= param
->element_count
;
747 const_count
= desc
.Elements
;
752 if (desc
.Elements
> 1)
754 FIXME("Unexpected number of constant elements %u.\n", desc
.Elements
);
755 return D3DERR_INVALIDCALL
;
757 param_count
= param
->member_count
;
758 const_count
= desc
.StructMembers
;
761 if (const_count
!= param_count
)
763 FIXME("Number of elements or struct members differs between parameter (%u) and constant (%u).\n",
764 param_count
, const_count
);
765 return D3DERR_INVALIDCALL
;
769 HRESULT ret
= D3D_OK
;
770 D3DXHANDLE hc_element
;
771 unsigned int index
= const_tab
->const_set_count
;
773 for (i
= 0; i
< const_count
; ++i
)
776 hc_element
= ID3DXConstantTable_GetConstantElement(ctab
, hc
, i
);
778 hc_element
= ID3DXConstantTable_GetConstant(ctab
, hc
, i
);
781 FIXME("Could not get constant.\n");
782 hr
= D3DERR_INVALIDCALL
;
786 hr
= init_set_constants_param(const_tab
, ctab
, hc_element
, ¶m
->members
[i
]);
793 return merge_const_set_entries(const_tab
, param
, index
);
796 TRACE("Constant %s, rows %u, columns %u, class %u, bytes %u.\n",
797 debugstr_a(desc
.Name
), desc
.Rows
, desc
.Columns
, desc
.Class
, desc
.Bytes
);
798 TRACE("Parameter %s, rows %u, columns %u, class %u, flags %#x, bytes %u.\n",
799 debugstr_a(param
->name
), param
->rows
, param
->columns
, param
->class,
800 param
->flags
, param
->bytes
);
802 const_set
.element_count
= 1;
803 const_set
.param
= param
;
804 const_set
.constant_class
= desc
.Class
;
805 if (desc
.RegisterSet
>= ARRAY_SIZE(shad_regset2table
))
807 FIXME("Unknown register set %u.\n", desc
.RegisterSet
);
808 return D3DERR_INVALIDCALL
;
810 const_set
.register_index
= desc
.RegisterIndex
;
811 const_set
.table
= const_tab
->regset2table
[desc
.RegisterSet
];
812 if (const_set
.table
>= PRES_REGTAB_COUNT
)
814 ERR("Unexpected register set %u.\n", desc
.RegisterSet
);
815 return D3DERR_INVALIDCALL
;
817 assert(table_info
[const_set
.table
].component_size
== sizeof(unsigned int));
818 assert(param
->bytes
/ (param
->rows
* param
->columns
) == sizeof(unsigned int));
819 const_set
.register_count
= desc
.RegisterCount
;
820 table_type
= table_info
[const_set
.table
].type
;
821 get_const_upload_info(&const_set
, &info
);
824 TRACE("%s has zero count, skipping.\n", debugstr_a(param
->name
));
828 if (table_type_from_param_type(param
->type
) == PRES_VT_COUNT
)
829 return D3DERR_INVALIDCALL
;
831 const_set
.direct_copy
= table_type_from_param_type(param
->type
) == table_type
832 && !info
.transpose
&& info
.minor
== info
.major_stride
833 && info
.count
== get_offset_reg(const_set
.table
, const_set
.register_count
)
834 && info
.count
* sizeof(unsigned int) <= param
->bytes
;
835 if (info
.minor_remainder
&& !const_set
.direct_copy
&& !info
.transpose
)
836 FIXME("Incomplete last row for not transposed matrix which cannot be directly copied, parameter %s.\n",
837 debugstr_a(param
->name
));
839 if (info
.major_count
> info
.major
840 || (info
.major_count
== info
.major
&& info
.minor_remainder
))
842 WARN("Constant dimensions exceed parameter size.\n");
843 return D3DERR_INVALIDCALL
;
846 if (FAILED(hr
= append_const_set(const_tab
, &const_set
)))
852 static HRESULT
get_constants_desc(unsigned int *byte_code
, struct d3dx_const_tab
*out
,
853 struct d3dx_parameters_store
*parameters
, const char **skip_constants
,
854 unsigned int skip_constants_count
, struct d3dx_preshader
*pres
)
856 ID3DXConstantTable
*ctab
;
857 D3DXCONSTANT_DESC
*cdesc
;
858 struct d3dx_parameter
**inputs_param
;
859 D3DXCONSTANTTABLE_DESC desc
;
864 hr
= D3DXGetShaderConstantTable((DWORD
*)byte_code
, &ctab
);
865 if (FAILED(hr
) || !ctab
)
867 TRACE("Could not get CTAB data, hr %#lx.\n", hr
);
868 /* returning OK, shaders and preshaders without CTAB are valid */
871 if (FAILED(hr
= ID3DXConstantTable_GetDesc(ctab
, &desc
)))
873 FIXME("Could not get CTAB desc, hr %#lx.\n", hr
);
877 out
->inputs
= cdesc
= malloc(sizeof(*cdesc
) * desc
.Constants
);
878 out
->inputs_param
= inputs_param
= malloc(sizeof(*inputs_param
) * desc
.Constants
);
879 if (!cdesc
|| !inputs_param
)
885 for (i
= 0; i
< desc
.Constants
; ++i
)
887 unsigned int index
= out
->input_count
;
888 WORD constantinfo_reserved
;
890 hc
= ID3DXConstantTable_GetConstant(ctab
, NULL
, i
);
893 FIXME("Null constant handle.\n");
896 if (FAILED(hr
= get_ctab_constant_desc(ctab
, hc
, &cdesc
[index
], &constantinfo_reserved
)))
898 inputs_param
[index
] = get_parameter_by_name(parameters
, NULL
, cdesc
[index
].Name
);
899 if (!inputs_param
[index
])
901 WARN("Could not find parameter %s in effect.\n", cdesc
[index
].Name
);
904 if (cdesc
[index
].Class
== D3DXPC_OBJECT
)
906 TRACE("Object %s, parameter %p.\n", cdesc
[index
].Name
, inputs_param
[index
]);
907 if (cdesc
[index
].RegisterSet
!= D3DXRS_SAMPLER
|| inputs_param
[index
]->class != D3DXPC_OBJECT
908 || !is_param_type_sampler(inputs_param
[index
]->type
))
910 WARN("Unexpected object type, constant %s.\n", debugstr_a(cdesc
[index
].Name
));
911 hr
= D3DERR_INVALIDCALL
;
914 if (max(inputs_param
[index
]->element_count
, 1) < cdesc
[index
].RegisterCount
)
916 WARN("Register count exceeds parameter size, constant %s.\n", debugstr_a(cdesc
[index
].Name
));
917 hr
= D3DERR_INVALIDCALL
;
921 if (!is_top_level_parameter(inputs_param
[index
]))
923 WARN("Expected top level parameter '%s'.\n", debugstr_a(cdesc
[index
].Name
));
928 for (j
= 0; j
< skip_constants_count
; ++j
)
930 if (!strcmp(cdesc
[index
].Name
, skip_constants
[j
]))
932 if (!constantinfo_reserved
)
934 WARN("skip_constants parameter %s is not register bound.\n",
936 hr
= D3DERR_INVALIDCALL
;
939 TRACE("Skipping constant %s.\n", cdesc
[index
].Name
);
943 if (j
< skip_constants_count
)
946 if (inputs_param
[index
]->class == D3DXPC_OBJECT
)
948 if (FAILED(hr
= init_set_constants_param(out
, ctab
, hc
, inputs_param
[index
])))
952 append_pres_const_sets_for_shader_input(out
, pres
);
953 if (out
->const_set_count
)
955 struct d3dx_const_param_eval_output
*new_alloc
;
957 qsort(out
->const_set
, out
->const_set_count
, sizeof(*out
->const_set
), compare_const_set
);
960 while (i
< out
->const_set_count
- 1)
962 if (out
->const_set
[i
].constant_class
== D3DXPC_FORCE_DWORD
963 && out
->const_set
[i
+ 1].constant_class
== D3DXPC_FORCE_DWORD
964 && out
->const_set
[i
].table
== out
->const_set
[i
+ 1].table
965 && out
->const_set
[i
].register_index
+ out
->const_set
[i
].register_count
966 >= out
->const_set
[i
+ 1].register_index
)
968 assert(out
->const_set
[i
].register_index
+ out
->const_set
[i
].register_count
969 <= out
->const_set
[i
+ 1].register_index
+ 1);
970 out
->const_set
[i
].register_count
= out
->const_set
[i
+ 1].register_index
+ 1
971 - out
->const_set
[i
].register_index
;
972 memmove(&out
->const_set
[i
+ 1], &out
->const_set
[i
+ 2], sizeof(out
->const_set
[i
])
973 * (out
->const_set_count
- i
- 2));
974 --out
->const_set_count
;
982 new_alloc
= realloc(out
->const_set
, sizeof(*out
->const_set
) * out
->const_set_count
);
985 out
->const_set
= new_alloc
;
986 out
->const_set_size
= out
->const_set_count
;
990 WARN("Out of memory.\n");
994 ID3DXConstantTable_Release(ctab
);
998 static void update_table_size(unsigned int *table_sizes
, unsigned int table
, unsigned int max_register
)
1000 if (table
< PRES_REGTAB_COUNT
)
1001 table_sizes
[table
] = max(table_sizes
[table
], max_register
+ 1);
1004 static void update_table_sizes_consts(unsigned int *table_sizes
, struct d3dx_const_tab
*ctab
)
1006 unsigned int i
, table
, max_register
;
1008 for (i
= 0; i
< ctab
->input_count
; ++i
)
1010 if (!ctab
->inputs
[i
].RegisterCount
)
1012 max_register
= ctab
->inputs
[i
].RegisterIndex
+ ctab
->inputs
[i
].RegisterCount
- 1;
1013 table
= ctab
->regset2table
[ctab
->inputs
[i
].RegisterSet
];
1014 update_table_size(table_sizes
, table
, max_register
);
1018 static void dump_arg(struct d3dx_regstore
*rs
, const struct d3dx_pres_operand
*arg
, int component_count
)
1020 static const char *xyzw_str
= "xyzw";
1021 unsigned int i
, table
, reg_offset
;
1023 table
= arg
->reg
.table
;
1024 if (table
== PRES_REGTAB_IMMED
&& arg
->index_reg
.table
== PRES_REGTAB_COUNT
)
1027 for (i
= 0; i
< component_count
; ++i
)
1028 TRACE(i
< component_count
- 1 ? "%.16e, " : "%.16e",
1029 ((double *)rs
->tables
[PRES_REGTAB_IMMED
])[arg
->reg
.offset
+ i
]);
1034 reg_offset
= get_reg_offset(table
, arg
->reg
.offset
);
1036 if (arg
->index_reg
.table
== PRES_REGTAB_COUNT
)
1038 if (table
== PRES_REGTAB_INPUT
&& reg_offset
< 2)
1039 TRACE("%s%s.", table_symbol
[table
], reg_offset
? "PSize" : "Pos");
1041 TRACE("%s%u.", table_symbol
[table
], reg_offset
);
1045 unsigned int index_reg
;
1047 index_reg
= get_reg_offset(arg
->index_reg
.table
, arg
->index_reg
.offset
);
1048 TRACE("%s[%u + %s%u.%c].", table_symbol
[table
], reg_offset
,
1049 table_symbol
[arg
->index_reg
.table
], index_reg
,
1050 xyzw_str
[arg
->index_reg
.offset
- get_offset_reg(arg
->index_reg
.table
, index_reg
)]);
1052 for (i
= 0; i
< component_count
; ++i
)
1053 TRACE("%c", xyzw_str
[(arg
->reg
.offset
+ i
) % 4]);
1057 static void dump_registers(struct d3dx_const_tab
*ctab
)
1059 unsigned int table
, i
;
1061 for (i
= 0; i
< ctab
->input_count
; ++i
)
1063 table
= ctab
->regset2table
[ctab
->inputs
[i
].RegisterSet
];
1064 TRACE("// %-12s %s%-4u %u\n", ctab
->inputs_param
[i
] ? ctab
->inputs_param
[i
]->name
: "(nil)",
1065 table_symbol
[table
], ctab
->inputs
[i
].RegisterIndex
, ctab
->inputs
[i
].RegisterCount
);
1069 static void dump_ins(struct d3dx_regstore
*rs
, const struct d3dx_pres_ins
*ins
)
1073 TRACE("%s ", pres_op_info
[ins
->op
].mnem
);
1074 dump_arg(rs
, &ins
->output
, pres_op_info
[ins
->op
].func_all_comps
? 1 : ins
->component_count
);
1075 for (i
= 0; i
< pres_op_info
[ins
->op
].input_count
; ++i
)
1078 dump_arg(rs
, &ins
->inputs
[i
], ins
->scalar_op
&& !i
? 1 : ins
->component_count
);
1083 static void dump_preshader(struct d3dx_preshader
*pres
)
1085 unsigned int i
, immediate_count
= pres
->regs
.table_sizes
[PRES_REGTAB_IMMED
] * 4;
1086 const double *immediates
= pres
->regs
.tables
[PRES_REGTAB_IMMED
];
1088 if (immediate_count
)
1089 TRACE("// Immediates:\n");
1090 for (i
= 0; i
< immediate_count
; ++i
)
1094 TRACE("%.8e", immediates
[i
]);
1100 TRACE("// Preshader registers:\n");
1101 dump_registers(&pres
->inputs
);
1102 TRACE("preshader\n");
1103 for (i
= 0; i
< pres
->ins_count
; ++i
)
1104 dump_ins(&pres
->regs
, &pres
->ins
[i
]);
1107 static HRESULT
parse_preshader(struct d3dx_preshader
*pres
, unsigned int *ptr
, unsigned int count
,
1108 struct d3dx_parameters_store
*parameters
)
1111 unsigned int i
, j
, const_count
, magic
;
1114 unsigned int saved_word
;
1115 unsigned int section_size
;
1119 TRACE("Preshader version %#x.\n", *ptr
);
1123 WARN("Unexpected end of byte code buffer.\n");
1124 return D3DXERR_INVALIDDATA
;
1127 p
= find_bytecode_comment(ptr
+ 1, count
- 1, FOURCC_CLIT
, §ion_size
);
1131 if (const_count
> (section_size
- 1) / (sizeof(double) / sizeof(unsigned int)))
1133 WARN("Byte code buffer ends unexpectedly.\n");
1134 return D3DXERR_INVALIDDATA
;
1136 dconst
= (double *)p
;
1143 TRACE("%u double constants.\n", const_count
);
1145 p
= find_bytecode_comment(ptr
+ 1, count
- 1, FOURCC_FXLC
, §ion_size
);
1148 WARN("Could not find preshader code.\n");
1151 pres
->ins_count
= *p
++;
1153 if (pres
->ins_count
> UINT_MAX
/ sizeof(*pres
->ins
))
1155 WARN("Invalid instruction count %u.\n", pres
->ins_count
);
1156 return D3DXERR_INVALIDDATA
;
1158 TRACE("%u instructions.\n", pres
->ins_count
);
1159 pres
->ins
= calloc(pres
->ins_count
, sizeof(*pres
->ins
));
1161 return E_OUTOFMEMORY
;
1162 for (i
= 0; i
< pres
->ins_count
; ++i
)
1164 unsigned int *ptr_next
;
1166 ptr_next
= parse_pres_ins(p
, section_size
, &pres
->ins
[i
]);
1168 return D3DXERR_INVALIDDATA
;
1169 section_size
-= ptr_next
- p
;
1173 pres
->inputs
.regset2table
= pres_regset2table
;
1177 hr
= get_constants_desc(ptr
, &pres
->inputs
, parameters
, NULL
, 0, NULL
);
1182 if (const_count
% get_reg_components(PRES_REGTAB_IMMED
))
1184 FIXME("const_count %u is not a multiple of %u.\n", const_count
,
1185 get_reg_components(PRES_REGTAB_IMMED
));
1186 return D3DXERR_INVALIDDATA
;
1188 pres
->regs
.table_sizes
[PRES_REGTAB_IMMED
] = get_reg_offset(PRES_REGTAB_IMMED
, const_count
);
1189 if (magic
== FOURCC_TX_1
)
1190 pres
->regs
.table_sizes
[PRES_REGTAB_INPUT
] = 2;
1192 update_table_sizes_consts(pres
->regs
.table_sizes
, &pres
->inputs
);
1193 for (i
= 0; i
< pres
->ins_count
; ++i
)
1195 for (j
= 0; j
< pres_op_info
[pres
->ins
[i
].op
].input_count
; ++j
)
1197 enum pres_reg_tables table
;
1198 unsigned int reg_idx
;
1200 if (pres
->ins
[i
].inputs
[j
].index_reg
.table
== PRES_REGTAB_COUNT
)
1202 unsigned int last_component_index
= pres
->ins
[i
].scalar_op
&& !j
? 0
1203 : pres
->ins
[i
].component_count
- 1;
1205 table
= pres
->ins
[i
].inputs
[j
].reg
.table
;
1206 reg_idx
= get_reg_offset(table
, pres
->ins
[i
].inputs
[j
].reg
.offset
1207 + last_component_index
);
1211 table
= pres
->ins
[i
].inputs
[j
].index_reg
.table
;
1212 reg_idx
= get_reg_offset(table
, pres
->ins
[i
].inputs
[j
].index_reg
.offset
);
1214 if (reg_idx
>= pres
->regs
.table_sizes
[table
])
1216 /* Native accepts these broken preshaders. */
1217 FIXME("Out of bounds register index, i %u, j %u, table %u, reg_idx %u, preshader parsing failed.\n",
1218 i
, j
, table
, reg_idx
);
1219 return D3DXERR_INVALIDDATA
;
1222 update_table_size(pres
->regs
.table_sizes
, pres
->ins
[i
].output
.reg
.table
,
1223 get_reg_offset(pres
->ins
[i
].output
.reg
.table
, pres
->ins
[i
].output
.reg
.offset
));
1225 if (FAILED(regstore_alloc_table(&pres
->regs
, PRES_REGTAB_IMMED
)))
1226 return E_OUTOFMEMORY
;
1227 regstore_set_values(&pres
->regs
, PRES_REGTAB_IMMED
, dconst
, 0, const_count
);
1232 HRESULT
d3dx_create_param_eval(struct d3dx_parameters_store
*parameters
, void *byte_code
, unsigned int byte_code_size
,
1233 D3DXPARAMETER_TYPE type
, struct d3dx_param_eval
**peval_out
, ULONG64
*version_counter
,
1234 const char **skip_constants
, unsigned int skip_constants_count
)
1236 struct d3dx_param_eval
*peval
;
1237 unsigned int *ptr
, *shader_ptr
= NULL
;
1240 unsigned int count
, pres_size
;
1243 TRACE("parameters %p, byte_code %p, byte_code_size %u, type %u, peval_out %p.\n",
1244 parameters
, byte_code
, byte_code_size
, type
, peval_out
);
1246 count
= byte_code_size
/ sizeof(unsigned int);
1247 if (!byte_code
|| !count
)
1253 peval
= calloc(1, sizeof(*peval
));
1256 ret
= E_OUTOFMEMORY
;
1259 peval
->version_counter
= version_counter
;
1261 peval
->param_type
= type
;
1264 case D3DXPT_VERTEXSHADER
:
1265 case D3DXPT_PIXELSHADER
:
1272 peval
->shader_inputs
.regset2table
= shad_regset2table
;
1274 ptr
= (unsigned int *)byte_code
;
1277 if ((*ptr
& 0xfffe0000) != 0xfffe0000)
1279 FIXME("Invalid shader signature %#x.\n", *ptr
);
1280 ret
= D3DXERR_INVALIDDATA
;
1283 TRACE("Shader version %#x.\n", *ptr
& 0xffff);
1285 ptr
= find_bytecode_comment(ptr
+ 1, count
- 1, FOURCC_PRES
, &pres_size
);
1287 TRACE("No preshader found.\n");
1294 if (ptr
&& FAILED(ret
= parse_preshader(&peval
->pres
, ptr
, pres_size
, parameters
)))
1296 FIXME("Failed parsing preshader, byte code for analysis follows.\n");
1297 dump_bytecode(byte_code
, byte_code_size
);
1303 if (FAILED(ret
= get_constants_desc(shader_ptr
, &peval
->shader_inputs
, parameters
,
1304 skip_constants
, skip_constants_count
, &peval
->pres
)))
1306 TRACE("Could not get shader constant table, hr %#lx.\n", ret
);
1309 update_table_sizes_consts(peval
->pres
.regs
.table_sizes
, &peval
->shader_inputs
);
1312 for (i
= PRES_REGTAB_FIRST_SHADER
; i
< PRES_REGTAB_COUNT
; ++i
)
1314 if (FAILED(ret
= regstore_alloc_table(&peval
->pres
.regs
, i
)))
1320 dump_bytecode(byte_code
, byte_code_size
);
1321 dump_preshader(&peval
->pres
);
1324 TRACE("// Shader registers:\n");
1325 dump_registers(&peval
->shader_inputs
);
1329 TRACE("Created parameter evaluator %p.\n", *peval_out
);
1333 WARN("Error creating parameter evaluator.\n");
1335 dump_bytecode(byte_code
, byte_code_size
);
1337 d3dx_free_param_eval(peval
);
1342 static void d3dx_free_const_tab(struct d3dx_const_tab
*ctab
)
1345 free(ctab
->inputs_param
);
1346 free(ctab
->const_set
);
1349 static void d3dx_free_preshader(struct d3dx_preshader
*pres
)
1353 regstore_free_tables(&pres
->regs
);
1354 d3dx_free_const_tab(&pres
->inputs
);
1357 void d3dx_free_param_eval(struct d3dx_param_eval
*peval
)
1359 TRACE("peval %p.\n", peval
);
1364 d3dx_free_preshader(&peval
->pres
);
1365 d3dx_free_const_tab(&peval
->shader_inputs
);
1369 static void pres_int_from_float(void *out
, const void *in
, unsigned int count
)
1372 const float *in_float
= in
;
1375 for (i
= 0; i
< count
; ++i
)
1376 out_int
[i
] = in_float
[i
];
1379 static void pres_bool_from_value(void *out
, const void *in
, unsigned int count
)
1382 const DWORD
*in_dword
= in
;
1383 BOOL
*out_bool
= out
;
1385 for (i
= 0; i
< count
; ++i
)
1386 out_bool
[i
] = !!in_dword
[i
];
1389 static void pres_float_from_int(void *out
, const void *in
, unsigned int count
)
1392 const int *in_int
= in
;
1393 float *out_float
= out
;
1395 for (i
= 0; i
< count
; ++i
)
1396 out_float
[i
] = in_int
[i
];
1399 static void pres_float_from_bool(void *out
, const void *in
, unsigned int count
)
1402 const BOOL
*in_bool
= in
;
1403 float *out_float
= out
;
1405 for (i
= 0; i
< count
; ++i
)
1406 out_float
[i
] = !!in_bool
[i
];
1409 static void pres_int_from_bool(void *out
, const void *in
, unsigned int count
)
1412 const float *in_bool
= in
;
1415 for (i
= 0; i
< count
; ++i
)
1416 out_int
[i
] = !!in_bool
[i
];
1419 static void regstore_set_data(struct d3dx_regstore
*rs
, unsigned int table
,
1420 unsigned int offset
, const unsigned int *in
, unsigned int count
, enum pres_value_type param_type
)
1422 typedef void (*conv_func
)(void *out
, const void *in
, unsigned int count
);
1423 static const conv_func set_const_funcs
[PRES_VT_COUNT
][PRES_VT_COUNT
] =
1425 {NULL
, NULL
, pres_int_from_float
, pres_bool_from_value
},
1426 {NULL
, NULL
, NULL
, NULL
},
1427 {pres_float_from_int
, NULL
, NULL
, pres_bool_from_value
},
1428 {pres_float_from_bool
, NULL
, pres_int_from_bool
, NULL
}
1430 enum pres_value_type table_type
= table_info
[table
].type
;
1432 if (param_type
== table_type
)
1434 regstore_set_values(rs
, table
, in
, offset
, count
);
1438 set_const_funcs
[param_type
][table_type
]((unsigned int *)rs
->tables
[table
] + offset
, in
, count
);
1441 static HRESULT
set_constants_device(ID3DXEffectStateManager
*manager
, struct IDirect3DDevice9
*device
,
1442 D3DXPARAMETER_TYPE type
, enum pres_reg_tables table
, void *ptr
,
1443 unsigned int start
, unsigned int count
)
1445 if (type
== D3DXPT_VERTEXSHADER
)
1449 case PRES_REGTAB_OCONST
:
1450 return SET_D3D_STATE_(manager
, device
, SetVertexShaderConstantF
, start
, ptr
, count
);
1451 case PRES_REGTAB_OICONST
:
1452 return SET_D3D_STATE_(manager
, device
, SetVertexShaderConstantI
, start
, ptr
, count
);
1453 case PRES_REGTAB_OBCONST
:
1454 return SET_D3D_STATE_(manager
, device
, SetVertexShaderConstantB
, start
, ptr
, count
);
1456 FIXME("Unexpected register table %u.\n", table
);
1457 return D3DERR_INVALIDCALL
;
1460 else if (type
== D3DXPT_PIXELSHADER
)
1464 case PRES_REGTAB_OCONST
:
1465 return SET_D3D_STATE_(manager
, device
, SetPixelShaderConstantF
, start
, ptr
, count
);
1466 case PRES_REGTAB_OICONST
:
1467 return SET_D3D_STATE_(manager
, device
, SetPixelShaderConstantI
, start
, ptr
, count
);
1468 case PRES_REGTAB_OBCONST
:
1469 return SET_D3D_STATE_(manager
, device
, SetPixelShaderConstantB
, start
, ptr
, count
);
1471 FIXME("Unexpected register table %u.\n", table
);
1472 return D3DERR_INVALIDCALL
;
1477 FIXME("Unexpected parameter type %u.\n", type
);
1478 return D3DERR_INVALIDCALL
;
1482 static HRESULT
set_constants(struct d3dx_regstore
*rs
, struct d3dx_const_tab
*const_tab
,
1483 ULONG64 new_update_version
, ID3DXEffectStateManager
*manager
, struct IDirect3DDevice9
*device
,
1484 D3DXPARAMETER_TYPE type
, BOOL device_update_all
, BOOL pres_dirty
)
1486 unsigned int const_idx
;
1487 unsigned int current_start
= 0, current_count
= 0;
1488 enum pres_reg_tables current_table
= PRES_REGTAB_COUNT
;
1489 BOOL update_device
= manager
|| device
;
1490 HRESULT hr
, result
= D3D_OK
;
1491 ULONG64 update_version
= const_tab
->update_version
;
1493 for (const_idx
= 0; const_idx
< const_tab
->const_set_count
; ++const_idx
)
1495 struct d3dx_const_param_eval_output
*const_set
= &const_tab
->const_set
[const_idx
];
1496 enum pres_reg_tables table
= const_set
->table
;
1497 struct d3dx_parameter
*param
= const_set
->param
;
1498 unsigned int element
, i
, j
, start_offset
;
1499 struct const_upload_info info
;
1501 enum pres_value_type param_type
;
1503 if (!(param
&& is_param_dirty(param
, update_version
)))
1507 start_offset
= get_offset_reg(table
, const_set
->register_index
);
1508 if (const_set
->direct_copy
)
1510 regstore_set_values(rs
, table
, data
, start_offset
,
1511 get_offset_reg(table
, const_set
->register_count
));
1514 param_type
= table_type_from_param_type(param
->type
);
1515 if (const_set
->constant_class
== D3DXPC_SCALAR
|| const_set
->constant_class
== D3DXPC_VECTOR
)
1517 unsigned int count
= max(param
->rows
, param
->columns
);
1519 if (count
>= get_reg_components(table
))
1521 regstore_set_data(rs
, table
, start_offset
, data
,
1522 count
* const_set
->element_count
, param_type
);
1526 for (element
= 0; element
< const_set
->element_count
; ++element
)
1527 regstore_set_data(rs
, table
, start_offset
+ get_offset_reg(table
, element
),
1528 &data
[element
* count
], count
, param_type
);
1532 get_const_upload_info(const_set
, &info
);
1533 for (element
= 0; element
< const_set
->element_count
; ++element
)
1535 unsigned int *out
= (unsigned int *)rs
->tables
[table
] + start_offset
;
1537 /* Store reshaped but (possibly) not converted yet data temporarily in the same constants buffer.
1538 * All the supported types of parameters and table values have the same size. */
1541 for (i
= 0; i
< info
.major_count
; ++i
)
1542 for (j
= 0; j
< info
.minor
; ++j
)
1543 out
[i
* info
.major_stride
+ j
] = data
[i
+ j
* info
.major
];
1545 for (j
= 0; j
< info
.minor_remainder
; ++j
)
1546 out
[i
* info
.major_stride
+ j
] = data
[i
+ j
* info
.major
];
1550 for (i
= 0; i
< info
.major_count
; ++i
)
1551 for (j
= 0; j
< info
.minor
; ++j
)
1552 out
[i
* info
.major_stride
+ j
] = data
[i
* info
.minor
+ j
];
1554 start_offset
+= get_offset_reg(table
, const_set
->register_count
);
1555 data
+= param
->rows
* param
->columns
;
1557 start_offset
= get_offset_reg(table
, const_set
->register_index
);
1558 if (table_info
[table
].type
!= param_type
)
1559 regstore_set_data(rs
, table
, start_offset
, (unsigned int *)rs
->tables
[table
] + start_offset
,
1560 get_offset_reg(table
, const_set
->register_count
) * const_set
->element_count
, param_type
);
1562 const_tab
->update_version
= new_update_version
;
1566 for (const_idx
= 0; const_idx
< const_tab
->const_set_count
; ++const_idx
)
1568 struct d3dx_const_param_eval_output
*const_set
= &const_tab
->const_set
[const_idx
];
1570 if (device_update_all
|| (const_set
->param
1571 ? is_param_dirty(const_set
->param
, update_version
) : pres_dirty
))
1573 enum pres_reg_tables table
= const_set
->table
;
1575 if (table
== current_table
&& current_start
+ current_count
== const_set
->register_index
)
1577 current_count
+= const_set
->register_count
* const_set
->element_count
;
1583 if (FAILED(hr
= set_constants_device(manager
, device
, type
, current_table
,
1584 (DWORD
*)rs
->tables
[current_table
]
1585 + get_offset_reg(current_table
, current_start
), current_start
, current_count
)))
1588 current_table
= table
;
1589 current_start
= const_set
->register_index
;
1590 current_count
= const_set
->register_count
* const_set
->element_count
;
1596 if (FAILED(hr
= set_constants_device(manager
, device
, type
, current_table
,
1597 (DWORD
*)rs
->tables
[current_table
]
1598 + get_offset_reg(current_table
, current_start
), current_start
, current_count
)))
1604 static double exec_get_reg_value(struct d3dx_regstore
*rs
, enum pres_reg_tables table
, unsigned int offset
)
1606 return regstore_get_double(rs
, table
, offset
);
1609 static double exec_get_arg(struct d3dx_regstore
*rs
, const struct d3dx_pres_operand
*opr
, unsigned int comp
)
1611 unsigned int offset
, base_index
, reg_index
, table
;
1613 table
= opr
->reg
.table
;
1615 if (opr
->index_reg
.table
== PRES_REGTAB_COUNT
)
1618 base_index
= lrint(exec_get_reg_value(rs
, opr
->index_reg
.table
, opr
->index_reg
.offset
));
1620 offset
= get_offset_reg(table
, base_index
) + opr
->reg
.offset
+ comp
;
1621 reg_index
= get_reg_offset(table
, offset
);
1623 if (reg_index
>= rs
->table_sizes
[table
])
1625 unsigned int wrap_size
;
1627 if (table
== PRES_REGTAB_CONST
)
1629 /* As it can be guessed from tests, offset into floating constant table is wrapped
1630 * to the nearest power of 2 and not to the actual table size. */
1631 for (wrap_size
= 1; wrap_size
< rs
->table_sizes
[table
]; wrap_size
<<= 1)
1636 wrap_size
= rs
->table_sizes
[table
];
1638 WARN("Wrapping register index %u, table %u, wrap_size %u, table size %u.\n",
1639 reg_index
, table
, wrap_size
, rs
->table_sizes
[table
]);
1640 reg_index
%= wrap_size
;
1642 if (reg_index
>= rs
->table_sizes
[table
])
1645 offset
= get_offset_reg(table
, reg_index
) + offset
% get_reg_components(table
);
1648 return exec_get_reg_value(rs
, table
, offset
);
1651 static void exec_set_arg(struct d3dx_regstore
*rs
, const struct d3dx_pres_reg
*reg
,
1652 unsigned int comp
, double res
)
1654 regstore_set_double(rs
, reg
->table
, reg
->offset
+ comp
, res
);
1657 #define ARGS_ARRAY_SIZE 8
1658 static HRESULT
execute_preshader(struct d3dx_preshader
*pres
)
1660 unsigned int i
, j
, k
;
1661 double args
[ARGS_ARRAY_SIZE
];
1664 for (i
= 0; i
< pres
->ins_count
; ++i
)
1666 const struct d3dx_pres_ins
*ins
;
1667 const struct op_info
*oi
;
1669 ins
= &pres
->ins
[i
];
1670 oi
= &pres_op_info
[ins
->op
];
1671 if (oi
->func_all_comps
)
1673 if (oi
->input_count
* ins
->component_count
> ARGS_ARRAY_SIZE
)
1675 FIXME("Too many arguments (%u) for one instruction.\n", oi
->input_count
* ins
->component_count
);
1678 for (k
= 0; k
< oi
->input_count
; ++k
)
1679 for (j
= 0; j
< ins
->component_count
; ++j
)
1680 args
[k
* ins
->component_count
+ j
] = exec_get_arg(&pres
->regs
, &ins
->inputs
[k
],
1681 ins
->scalar_op
&& !k
? 0 : j
);
1682 res
= oi
->func(args
, ins
->component_count
);
1684 /* only 'dot' instruction currently falls here */
1685 exec_set_arg(&pres
->regs
, &ins
->output
.reg
, 0, res
);
1689 for (j
= 0; j
< ins
->component_count
; ++j
)
1691 for (k
= 0; k
< oi
->input_count
; ++k
)
1692 args
[k
] = exec_get_arg(&pres
->regs
, &ins
->inputs
[k
], ins
->scalar_op
&& !k
? 0 : j
);
1693 res
= oi
->func(args
, ins
->component_count
);
1694 exec_set_arg(&pres
->regs
, &ins
->output
.reg
, j
, res
);
1701 static BOOL
is_const_tab_input_dirty(struct d3dx_const_tab
*ctab
, ULONG64 update_version
)
1705 if (update_version
== ULONG64_MAX
)
1706 update_version
= ctab
->update_version
;
1707 for (i
= 0; i
< ctab
->input_count
; ++i
)
1709 if (is_top_level_param_dirty(top_level_parameter_from_parameter(ctab
->inputs_param
[i
]),
1716 BOOL
is_param_eval_input_dirty(struct d3dx_param_eval
*peval
, ULONG64 update_version
)
1718 return is_const_tab_input_dirty(&peval
->pres
.inputs
, update_version
)
1719 || is_const_tab_input_dirty(&peval
->shader_inputs
, update_version
);
1722 HRESULT
d3dx_evaluate_parameter(struct d3dx_param_eval
*peval
, const struct d3dx_parameter
*param
,
1727 unsigned int elements
, elements_param
, elements_table
;
1731 TRACE("peval %p, param %p, param_value %p.\n", peval
, param
, param_value
);
1733 if ((is_dirty
= is_const_tab_input_dirty(&peval
->pres
.inputs
, ULONG64_MAX
)))
1735 set_constants(&peval
->pres
.regs
, &peval
->pres
.inputs
,
1736 next_update_version(peval
->version_counter
), NULL
, NULL
,
1737 peval
->param_type
, FALSE
, FALSE
);
1740 if (is_dirty
|| peval
->pres
.regs
.table_sizes
[PRES_REGTAB_INPUT
])
1742 if (FAILED(hr
= execute_preshader(&peval
->pres
)))
1746 elements_table
= get_offset_reg(PRES_REGTAB_OCONST
, peval
->pres
.regs
.table_sizes
[PRES_REGTAB_OCONST
]);
1747 elements_param
= param
->bytes
/ sizeof(unsigned int);
1748 elements
= min(elements_table
, elements_param
);
1749 oc
= (float *)peval
->pres
.regs
.tables
[PRES_REGTAB_OCONST
];
1750 for (i
= 0; i
< elements
; ++i
)
1751 set_number((unsigned int *)param_value
+ i
, param
->type
, oc
+ i
, D3DXPT_FLOAT
);
1755 HRESULT
d3dx_param_eval_set_shader_constants(ID3DXEffectStateManager
*manager
, struct IDirect3DDevice9
*device
,
1756 struct d3dx_param_eval
*peval
, BOOL update_all
)
1759 struct d3dx_preshader
*pres
= &peval
->pres
;
1760 struct d3dx_regstore
*rs
= &pres
->regs
;
1761 ULONG64 new_update_version
= next_update_version(peval
->version_counter
);
1762 BOOL pres_dirty
= FALSE
;
1764 TRACE("device %p, peval %p, param_type %u.\n", device
, peval
, peval
->param_type
);
1766 if (is_const_tab_input_dirty(&pres
->inputs
, ULONG64_MAX
))
1768 set_constants(rs
, &pres
->inputs
, new_update_version
,
1769 NULL
, NULL
, peval
->param_type
, FALSE
, FALSE
);
1770 if (FAILED(hr
= execute_preshader(pres
)))
1775 return set_constants(rs
, &peval
->shader_inputs
, new_update_version
,
1776 manager
, device
, peval
->param_type
, update_all
, pres_dirty
);