1 /**************************************************************************
3 * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **************************************************************************/
32 #include "i915_context.h"
35 #include "pipe/p_shader_tokens.h"
36 #include "util/u_math.h"
37 #include "util/u_memory.h"
38 #include "util/u_string.h"
39 #include "tgsi/tgsi_parse.h"
40 #include "tgsi/tgsi_dump.h"
42 #include "draw/draw_vertex.h"
45 #define M_PI 3.14159265358979323846
49 * Simple pass-through fragment shader to use when we don't have
50 * a real shader (or it fails to compile for some reason).
52 static unsigned passthrough_decl
[] =
54 _3DSTATE_PIXEL_SHADER_PROGRAM
| ((2*3)-1),
56 /* declare input color:
59 (REG_TYPE_T
<< D0_TYPE_SHIFT
) |
60 (T_DIFFUSE
<< D0_NR_SHIFT
) |
66 static unsigned passthrough_program
[] =
68 /* move to output color:
71 (REG_TYPE_OC
<< A0_DEST_TYPE_SHIFT
) |
73 (REG_TYPE_T
<< A0_SRC0_TYPE_SHIFT
) |
74 (T_DIFFUSE
<< A0_SRC0_NR_SHIFT
)),
75 0x01230000, /* .xyzw */
80 /* 1, -1/3!, 1/5!, -1/7! */
81 static const float scs_sin_constants
[4] = { 1.0,
83 1.0f
/ (5 * 4 * 3 * 2 * 1),
84 -1.0f
/ (7 * 6 * 5 * 4 * 3 * 2 * 1)
87 /* 1, -1/2!, 1/4!, -1/6! */
88 static const float scs_cos_constants
[4] = { 1.0,
90 1.0f
/ (4 * 3 * 2 * 1),
91 -1.0f
/ (6 * 5 * 4 * 3 * 2 * 1)
94 /* 2*pi, -(2*pi)^3/3!, (2*pi)^5/5!, -(2*pi)^7/7! */
95 static const float sin_constants
[4] = { 2.0 * M_PI
,
96 -8.0f
* M_PI
* M_PI
* M_PI
/ (3 * 2 * 1),
97 32.0f
* M_PI
* M_PI
* M_PI
* M_PI
* M_PI
/ (5 * 4 * 3 * 2 * 1),
98 -128.0f
* M_PI
* M_PI
* M_PI
* M_PI
* M_PI
* M_PI
* M_PI
/ (7 * 6 * 5 * 4 * 3 * 2 * 1)
101 /* 1, -(2*pi)^2/2!, (2*pi)^4/4!, -(2*pi)^6/6! */
102 static const float cos_constants
[4] = { 1.0,
103 -4.0f
* M_PI
* M_PI
/ (2 * 1),
104 16.0f
* M_PI
* M_PI
* M_PI
* M_PI
/ (4 * 3 * 2 * 1),
105 -64.0f
* M_PI
* M_PI
* M_PI
* M_PI
* M_PI
* M_PI
/ (6 * 5 * 4 * 3 * 2 * 1)
111 * component-wise negation of ureg
114 negate(int reg
, int x
, int y
, int z
, int w
)
116 /* Another neat thing about the UREG representation */
117 return reg
^ (((x
& 1) << UREG_CHANNEL_X_NEGATE_SHIFT
) |
118 ((y
& 1) << UREG_CHANNEL_Y_NEGATE_SHIFT
) |
119 ((z
& 1) << UREG_CHANNEL_Z_NEGATE_SHIFT
) |
120 ((w
& 1) << UREG_CHANNEL_W_NEGATE_SHIFT
));
125 * In the event of a translation failure, we'll generate a simple color
126 * pass-through program.
129 i915_use_passthrough_shader(struct i915_fragment_shader
*fs
)
131 fs
->program
= (uint
*) MALLOC(sizeof(passthrough_program
));
132 fs
->decl
= (uint
*) MALLOC(sizeof(passthrough_decl
));
134 memcpy(fs
->program
, passthrough_program
, sizeof(passthrough_program
));
135 memcpy(fs
->decl
, passthrough_decl
, sizeof(passthrough_decl
));
136 fs
->program_len
= Elements(passthrough_program
);
137 fs
->decl_len
= Elements(passthrough_decl
);
139 fs
->num_constants
= 0;
144 i915_program_error(struct i915_fp_compile
*p
, const char *msg
, ...)
149 debug_printf("i915_program_error: ");
150 va_start( args
, msg
);
151 util_vsnprintf( buffer
, sizeof(buffer
), msg
, args
);
153 debug_printf("%s", buffer
);
159 static uint
get_mapping(struct i915_fragment_shader
* fs
, int unit
)
162 for (i
= 0; i
< I915_TEX_UNITS
; i
++)
164 if (fs
->generic_mapping
[i
] == -1) {
165 fs
->generic_mapping
[i
] = unit
;
168 if (fs
->generic_mapping
[i
] == unit
)
171 debug_printf("Exceeded max generics\n");
176 * Construct a ureg for the given source register. Will emit
177 * constants, apply swizzling and negation as needed.
180 src_vector(struct i915_fp_compile
*p
,
181 const struct i915_full_src_register
*source
,
182 struct i915_fragment_shader
* fs
)
184 uint index
= source
->Register
.Index
;
185 uint src
= 0, sem_name
, sem_ind
;
187 switch (source
->Register
.File
) {
188 case TGSI_FILE_TEMPORARY
:
189 if (source
->Register
.Index
>= I915_MAX_TEMPORARY
) {
190 i915_program_error(p
, "Exceeded max temporary reg");
193 src
= UREG(REG_TYPE_R
, index
);
195 case TGSI_FILE_INPUT
:
196 /* XXX: Packing COL1, FOGC into a single attribute works for
197 * texenv programs, but will fail for real fragment programs
198 * that use these attributes and expect them to be a full 4
199 * components wide. Could use a texcoord to pass these
200 * attributes if necessary, but that won't work in the general
203 * We also use a texture coordinate to pass wpos when possible.
206 sem_name
= p
->shader
->info
.input_semantic_name
[index
];
207 sem_ind
= p
->shader
->info
.input_semantic_index
[index
];
210 case TGSI_SEMANTIC_POSITION
:
213 int real_tex_unit
= get_mapping(fs
, I915_SEMANTIC_POS
);
214 src
= i915_emit_decl(p
, REG_TYPE_T
, T_TEX0
+ real_tex_unit
, D0_CHANNEL_ALL
);
217 case TGSI_SEMANTIC_COLOR
:
219 src
= i915_emit_decl(p
, REG_TYPE_T
, T_DIFFUSE
, D0_CHANNEL_ALL
);
222 /* secondary color */
223 assert(sem_ind
== 1);
224 src
= i915_emit_decl(p
, REG_TYPE_T
, T_SPECULAR
, D0_CHANNEL_XYZ
);
225 src
= swizzle(src
, X
, Y
, Z
, ONE
);
228 case TGSI_SEMANTIC_FOG
:
229 src
= i915_emit_decl(p
, REG_TYPE_T
, T_FOG_W
, D0_CHANNEL_W
);
230 src
= swizzle(src
, W
, W
, W
, W
);
232 case TGSI_SEMANTIC_GENERIC
:
234 int real_tex_unit
= get_mapping(fs
, sem_ind
);
235 src
= i915_emit_decl(p
, REG_TYPE_T
, T_TEX0
+ real_tex_unit
, D0_CHANNEL_ALL
);
238 case TGSI_SEMANTIC_FACE
:
240 /* for back/front faces */
241 int real_tex_unit
= get_mapping(fs
, I915_SEMANTIC_FACE
);
242 src
= i915_emit_decl(p
, REG_TYPE_T
, T_TEX0
+ real_tex_unit
, D0_CHANNEL_X
);
246 i915_program_error(p
, "Bad source->Index");
251 case TGSI_FILE_IMMEDIATE
:
252 assert(index
< p
->num_immediates
);
253 index
= p
->immediates_map
[index
];
255 case TGSI_FILE_CONSTANT
:
256 src
= UREG(REG_TYPE_CONST
, index
);
260 i915_program_error(p
, "Bad source->File");
265 source
->Register
.SwizzleX
,
266 source
->Register
.SwizzleY
,
267 source
->Register
.SwizzleZ
,
268 source
->Register
.SwizzleW
);
270 /* There's both negate-all-components and per-component negation.
271 * Try to handle both here.
274 int n
= source
->Register
.Negate
;
275 src
= negate(src
, n
, n
, n
, n
);
280 /* XXX assertions disabled to allow arbfplight.c to run */
281 /* XXX enable these assertions, or fix things */
282 assert(!source
->Register
.Absolute
);
284 if (source
->Register
.Absolute
)
285 debug_printf("Unhandled absolute value\n");
292 * Construct a ureg for a destination register.
295 get_result_vector(struct i915_fp_compile
*p
,
296 const struct i915_full_dst_register
*dest
)
298 switch (dest
->Register
.File
) {
299 case TGSI_FILE_OUTPUT
:
301 uint sem_name
= p
->shader
->info
.output_semantic_name
[dest
->Register
.Index
];
303 case TGSI_SEMANTIC_POSITION
:
304 return UREG(REG_TYPE_OD
, 0);
305 case TGSI_SEMANTIC_COLOR
:
306 return UREG(REG_TYPE_OC
, 0);
308 i915_program_error(p
, "Bad inst->DstReg.Index/semantics");
312 case TGSI_FILE_TEMPORARY
:
313 return UREG(REG_TYPE_R
, dest
->Register
.Index
);
315 i915_program_error(p
, "Bad inst->DstReg.File");
322 * Compute flags for saturation and writemask.
325 get_result_flags(const struct i915_full_instruction
*inst
)
328 = inst
->Dst
[0].Register
.WriteMask
;
331 if (inst
->Instruction
.Saturate
== TGSI_SAT_ZERO_ONE
)
332 flags
|= A0_DEST_SATURATE
;
334 if (writeMask
& TGSI_WRITEMASK_X
)
335 flags
|= A0_DEST_CHANNEL_X
;
336 if (writeMask
& TGSI_WRITEMASK_Y
)
337 flags
|= A0_DEST_CHANNEL_Y
;
338 if (writeMask
& TGSI_WRITEMASK_Z
)
339 flags
|= A0_DEST_CHANNEL_Z
;
340 if (writeMask
& TGSI_WRITEMASK_W
)
341 flags
|= A0_DEST_CHANNEL_W
;
348 * Convert TGSI_TEXTURE_x token to DO_SAMPLE_TYPE_x token
351 translate_tex_src_target(struct i915_fp_compile
*p
, uint tex
)
354 case TGSI_TEXTURE_SHADOW1D
:
356 case TGSI_TEXTURE_1D
:
357 return D0_SAMPLE_TYPE_2D
;
359 case TGSI_TEXTURE_SHADOW2D
:
361 case TGSI_TEXTURE_2D
:
362 return D0_SAMPLE_TYPE_2D
;
364 case TGSI_TEXTURE_SHADOWRECT
:
366 case TGSI_TEXTURE_RECT
:
367 return D0_SAMPLE_TYPE_2D
;
369 case TGSI_TEXTURE_3D
:
370 return D0_SAMPLE_TYPE_VOLUME
;
372 case TGSI_TEXTURE_CUBE
:
373 return D0_SAMPLE_TYPE_CUBE
;
376 i915_program_error(p
, "TexSrc type");
382 * Return the number of coords needed to access a given TGSI_TEXTURE_*
385 texture_num_coords(struct i915_fp_compile
*p
, uint tex
)
388 case TGSI_TEXTURE_SHADOW1D
:
389 case TGSI_TEXTURE_1D
:
392 case TGSI_TEXTURE_SHADOW2D
:
393 case TGSI_TEXTURE_2D
:
394 case TGSI_TEXTURE_SHADOWRECT
:
395 case TGSI_TEXTURE_RECT
:
398 case TGSI_TEXTURE_3D
:
399 case TGSI_TEXTURE_CUBE
:
403 i915_program_error(p
, "Num coords");
410 * Generate texel lookup instruction.
413 emit_tex(struct i915_fp_compile
*p
,
414 const struct i915_full_instruction
*inst
,
416 struct i915_fragment_shader
* fs
)
418 uint texture
= inst
->Texture
.Texture
;
419 uint unit
= inst
->Src
[1].Register
.Index
;
420 uint tex
= translate_tex_src_target( p
, texture
);
421 uint sampler
= i915_emit_decl(p
, REG_TYPE_S
, unit
, tex
);
422 uint coord
= src_vector( p
, &inst
->Src
[0], fs
);
425 get_result_vector( p
, &inst
->Dst
[0] ),
426 get_result_flags( inst
),
430 texture_num_coords(p
, texture
) );
435 * Generate a simple arithmetic instruction
436 * \param opcode the i915 opcode
437 * \param numArgs the number of input/src arguments
440 emit_simple_arith(struct i915_fp_compile
*p
,
441 const struct i915_full_instruction
*inst
,
442 uint opcode
, uint numArgs
,
443 struct i915_fragment_shader
* fs
)
445 uint arg1
, arg2
, arg3
;
447 assert(numArgs
<= 3);
449 arg1
= (numArgs
< 1) ? 0 : src_vector( p
, &inst
->Src
[0], fs
);
450 arg2
= (numArgs
< 2) ? 0 : src_vector( p
, &inst
->Src
[1], fs
);
451 arg3
= (numArgs
< 3) ? 0 : src_vector( p
, &inst
->Src
[2], fs
);
455 get_result_vector( p
, &inst
->Dst
[0]),
456 get_result_flags( inst
), 0,
463 /** As above, but swap the first two src regs */
465 emit_simple_arith_swap2(struct i915_fp_compile
*p
,
466 const struct i915_full_instruction
*inst
,
467 uint opcode
, uint numArgs
,
468 struct i915_fragment_shader
* fs
)
470 struct i915_full_instruction inst2
;
472 assert(numArgs
== 2);
474 /* transpose first two registers */
476 inst2
.Src
[0] = inst
->Src
[1];
477 inst2
.Src
[1] = inst
->Src
[0];
479 emit_simple_arith(p
, &inst2
, opcode
, numArgs
, fs
);
483 * Translate TGSI instruction to i915 instruction.
487 * DDX, DDY -- return 0
488 * SIN, COS -- could use another taylor step?
489 * LIT -- results seem a little different to sw mesa
490 * LOG -- different to mesa on negative numbers, but this is conformant.
493 i915_translate_instruction(struct i915_fp_compile
*p
,
494 const struct i915_full_instruction
*inst
,
495 struct i915_fragment_shader
*fs
)
498 uint src0
, src1
, src2
, flags
;
501 switch (inst
->Instruction
.Opcode
) {
502 case TGSI_OPCODE_ABS
:
503 src0
= src_vector(p
, &inst
->Src
[0], fs
);
506 get_result_vector(p
, &inst
->Dst
[0]),
507 get_result_flags(inst
), 0,
508 src0
, negate(src0
, 1, 1, 1, 1), 0);
511 case TGSI_OPCODE_ADD
:
512 emit_simple_arith(p
, inst
, A0_ADD
, 2, fs
);
515 case TGSI_OPCODE_CEIL
:
516 src0
= src_vector(p
, &inst
->Src
[0], fs
);
517 tmp
= i915_get_utemp(p
);
518 flags
= get_result_flags(inst
);
522 flags
& A0_DEST_CHANNEL_ALL
, 0,
523 negate(src0
, 1, 1, 1, 1), 0, 0);
526 get_result_vector(p
, &inst
->Dst
[0]),
528 negate(tmp
, 1, 1, 1, 1), 0, 0);
531 case TGSI_OPCODE_CMP
:
532 src0
= src_vector(p
, &inst
->Src
[0], fs
);
533 src1
= src_vector(p
, &inst
->Src
[1], fs
);
534 src2
= src_vector(p
, &inst
->Src
[2], fs
);
535 i915_emit_arith(p
, A0_CMP
,
536 get_result_vector(p
, &inst
->Dst
[0]),
537 get_result_flags(inst
),
538 0, src0
, src2
, src1
); /* NOTE: order of src2, src1 */
541 case TGSI_OPCODE_COS
:
542 src0
= src_vector(p
, &inst
->Src
[0], fs
);
543 tmp
= i915_get_utemp(p
);
547 tmp
, A0_DEST_CHANNEL_X
, 0,
548 src0
, i915_emit_const1f(p
, 1.0f
/ (float) (M_PI
* 2.0)), 0);
550 i915_emit_arith(p
, A0_MOD
, tmp
, A0_DEST_CHANNEL_X
, 0, tmp
, 0, 0);
553 * t0.xy = MUL x.xx11, x.x111 ; x^2, x, 1, 1
554 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, 1
555 * t0 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1
556 * result = DP4 t0, cos_constants
560 tmp
, A0_DEST_CHANNEL_XY
, 0,
561 swizzle(tmp
, X
, X
, ONE
, ONE
),
562 swizzle(tmp
, X
, ONE
, ONE
, ONE
), 0);
566 tmp
, A0_DEST_CHANNEL_XYZ
, 0,
567 swizzle(tmp
, X
, Y
, X
, ONE
),
568 swizzle(tmp
, X
, X
, ONE
, ONE
), 0);
572 tmp
, A0_DEST_CHANNEL_XYZ
, 0,
573 swizzle(tmp
, X
, X
, Z
, ONE
),
574 swizzle(tmp
, Z
, ONE
, ONE
, ONE
), 0);
578 get_result_vector(p
, &inst
->Dst
[0]),
579 get_result_flags(inst
), 0,
580 swizzle(tmp
, ONE
, Z
, Y
, X
),
581 i915_emit_const4fv(p
, cos_constants
), 0);
584 case TGSI_OPCODE_DDX
:
585 case TGSI_OPCODE_DDY
:
586 /* XXX We just output 0 here */
587 debug_printf("Punting DDX/DDX\n");
588 src0
= get_result_vector(p
, &inst
->Dst
[0]);
591 get_result_vector(p
, &inst
->Dst
[0]),
592 get_result_flags(inst
), 0,
593 swizzle(src0
, ZERO
, ZERO
, ZERO
, ZERO
), 0, 0);
596 case TGSI_OPCODE_DP2
:
597 src0
= src_vector(p
, &inst
->Src
[0], fs
);
598 src1
= src_vector(p
, &inst
->Src
[1], fs
);
602 get_result_vector(p
, &inst
->Dst
[0]),
603 get_result_flags(inst
), 0,
604 swizzle(src0
, X
, Y
, ZERO
, ZERO
), src1
, 0);
607 case TGSI_OPCODE_DP3
:
608 emit_simple_arith(p
, inst
, A0_DP3
, 2, fs
);
611 case TGSI_OPCODE_DP4
:
612 emit_simple_arith(p
, inst
, A0_DP4
, 2, fs
);
615 case TGSI_OPCODE_DPH
:
616 src0
= src_vector(p
, &inst
->Src
[0], fs
);
617 src1
= src_vector(p
, &inst
->Src
[1], fs
);
621 get_result_vector(p
, &inst
->Dst
[0]),
622 get_result_flags(inst
), 0,
623 swizzle(src0
, X
, Y
, Z
, ONE
), src1
, 0);
626 case TGSI_OPCODE_DST
:
627 src0
= src_vector(p
, &inst
->Src
[0], fs
);
628 src1
= src_vector(p
, &inst
->Src
[1], fs
);
630 /* result[0] = 1 * 1;
631 * result[1] = a[1] * b[1];
632 * result[2] = a[2] * 1;
633 * result[3] = 1 * b[3];
637 get_result_vector(p
, &inst
->Dst
[0]),
638 get_result_flags(inst
), 0,
639 swizzle(src0
, ONE
, Y
, Z
, ONE
),
640 swizzle(src1
, ONE
, Y
, ONE
, W
), 0);
643 case TGSI_OPCODE_END
:
647 case TGSI_OPCODE_EX2
:
648 src0
= src_vector(p
, &inst
->Src
[0], fs
);
652 get_result_vector(p
, &inst
->Dst
[0]),
653 get_result_flags(inst
), 0,
654 swizzle(src0
, X
, X
, X
, X
), 0, 0);
657 case TGSI_OPCODE_FLR
:
658 emit_simple_arith(p
, inst
, A0_FLR
, 1, fs
);
661 case TGSI_OPCODE_FRC
:
662 emit_simple_arith(p
, inst
, A0_FRC
, 1, fs
);
665 case TGSI_OPCODE_KILL_IF
:
666 /* kill if src[0].x < 0 || src[0].y < 0 ... */
667 src0
= src_vector(p
, &inst
->Src
[0], fs
);
668 tmp
= i915_get_utemp(p
);
671 tmp
, /* dest reg: a dummy reg */
672 A0_DEST_CHANNEL_ALL
, /* dest writemask */
675 T0_TEXKILL
, /* opcode */
679 case TGSI_OPCODE_KILL
:
680 /* unconditional kill */
681 tmp
= i915_get_utemp(p
);
684 tmp
, /* dest reg: a dummy reg */
685 A0_DEST_CHANNEL_ALL
, /* dest writemask */
687 negate(swizzle(0, ONE
, ONE
, ONE
, ONE
), 1, 1, 1, 1), /* coord */
688 T0_TEXKILL
, /* opcode */
692 case TGSI_OPCODE_LG2
:
693 src0
= src_vector(p
, &inst
->Src
[0], fs
);
697 get_result_vector(p
, &inst
->Dst
[0]),
698 get_result_flags(inst
), 0,
699 swizzle(src0
, X
, X
, X
, X
), 0, 0);
702 case TGSI_OPCODE_LIT
:
703 src0
= src_vector(p
, &inst
->Src
[0], fs
);
704 tmp
= i915_get_utemp(p
);
706 /* tmp = max( a.xyzw, a.00zw )
707 * XXX: Clamp tmp.w to -128..128
709 * tmp.y = tmp.w * tmp.y
711 * result = cmp (a.11-x1, a.1x01, a.1xy1 )
713 i915_emit_arith(p
, A0_MAX
, tmp
, A0_DEST_CHANNEL_ALL
, 0,
714 src0
, swizzle(src0
, ZERO
, ZERO
, Z
, W
), 0);
716 i915_emit_arith(p
, A0_LOG
, tmp
, A0_DEST_CHANNEL_Y
, 0,
717 swizzle(tmp
, Y
, Y
, Y
, Y
), 0, 0);
719 i915_emit_arith(p
, A0_MUL
, tmp
, A0_DEST_CHANNEL_Y
, 0,
720 swizzle(tmp
, ZERO
, Y
, ZERO
, ZERO
),
721 swizzle(tmp
, ZERO
, W
, ZERO
, ZERO
), 0);
723 i915_emit_arith(p
, A0_EXP
, tmp
, A0_DEST_CHANNEL_Y
, 0,
724 swizzle(tmp
, Y
, Y
, Y
, Y
), 0, 0);
726 i915_emit_arith(p
, A0_CMP
,
727 get_result_vector(p
, &inst
->Dst
[0]),
728 get_result_flags(inst
), 0,
729 negate(swizzle(tmp
, ONE
, ONE
, X
, ONE
), 0, 0, 1, 0),
730 swizzle(tmp
, ONE
, X
, ZERO
, ONE
),
731 swizzle(tmp
, ONE
, X
, Y
, ONE
));
735 case TGSI_OPCODE_LRP
:
736 src0
= src_vector(p
, &inst
->Src
[0], fs
);
737 src1
= src_vector(p
, &inst
->Src
[1], fs
);
738 src2
= src_vector(p
, &inst
->Src
[2], fs
);
739 flags
= get_result_flags(inst
);
740 tmp
= i915_get_utemp(p
);
747 * result = (-c)*a + tmp
749 i915_emit_arith(p
, A0_MAD
, tmp
,
750 flags
& A0_DEST_CHANNEL_ALL
, 0, src1
, src0
, src2
);
752 i915_emit_arith(p
, A0_MAD
,
753 get_result_vector(p
, &inst
->Dst
[0]),
754 flags
, 0, negate(src2
, 1, 1, 1, 1), src0
, tmp
);
757 case TGSI_OPCODE_MAD
:
758 emit_simple_arith(p
, inst
, A0_MAD
, 3, fs
);
761 case TGSI_OPCODE_MAX
:
762 emit_simple_arith(p
, inst
, A0_MAX
, 2, fs
);
765 case TGSI_OPCODE_MIN
:
766 src0
= src_vector(p
, &inst
->Src
[0], fs
);
767 src1
= src_vector(p
, &inst
->Src
[1], fs
);
768 tmp
= i915_get_utemp(p
);
769 flags
= get_result_flags(inst
);
773 tmp
, flags
& A0_DEST_CHANNEL_ALL
, 0,
774 negate(src0
, 1, 1, 1, 1),
775 negate(src1
, 1, 1, 1, 1), 0);
779 get_result_vector(p
, &inst
->Dst
[0]),
780 flags
, 0, negate(tmp
, 1, 1, 1, 1), 0, 0);
783 case TGSI_OPCODE_MOV
:
784 emit_simple_arith(p
, inst
, A0_MOV
, 1, fs
);
787 case TGSI_OPCODE_MUL
:
788 emit_simple_arith(p
, inst
, A0_MUL
, 2, fs
);
791 case TGSI_OPCODE_NOP
:
794 case TGSI_OPCODE_POW
:
795 src0
= src_vector(p
, &inst
->Src
[0], fs
);
796 src1
= src_vector(p
, &inst
->Src
[1], fs
);
797 tmp
= i915_get_utemp(p
);
798 flags
= get_result_flags(inst
);
800 /* XXX: masking on intermediate values, here and elsewhere.
804 tmp
, A0_DEST_CHANNEL_X
, 0,
805 swizzle(src0
, X
, X
, X
, X
), 0, 0);
807 i915_emit_arith(p
, A0_MUL
, tmp
, A0_DEST_CHANNEL_X
, 0, tmp
, src1
, 0);
811 get_result_vector(p
, &inst
->Dst
[0]),
812 flags
, 0, swizzle(tmp
, X
, X
, X
, X
), 0, 0);
815 case TGSI_OPCODE_RET
:
819 case TGSI_OPCODE_RCP
:
820 src0
= src_vector(p
, &inst
->Src
[0], fs
);
824 get_result_vector(p
, &inst
->Dst
[0]),
825 get_result_flags(inst
), 0,
826 swizzle(src0
, X
, X
, X
, X
), 0, 0);
829 case TGSI_OPCODE_RSQ
:
830 src0
= src_vector(p
, &inst
->Src
[0], fs
);
834 get_result_vector(p
, &inst
->Dst
[0]),
835 get_result_flags(inst
), 0,
836 swizzle(src0
, X
, X
, X
, X
), 0, 0);
839 case TGSI_OPCODE_SCS
:
840 src0
= src_vector(p
, &inst
->Src
[0], fs
);
841 tmp
= i915_get_utemp(p
);
844 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
845 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
846 * t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x
847 * scs.x = DP4 t1, scs_sin_constants
848 * t1 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1
849 * scs.y = DP4 t1, scs_cos_constants
853 tmp
, A0_DEST_CHANNEL_XY
, 0,
854 swizzle(src0
, X
, X
, ONE
, ONE
),
855 swizzle(src0
, X
, ONE
, ONE
, ONE
), 0);
859 tmp
, A0_DEST_CHANNEL_ALL
, 0,
860 swizzle(tmp
, X
, Y
, X
, Y
),
861 swizzle(tmp
, X
, X
, ONE
, ONE
), 0);
863 writemask
= inst
->Dst
[0].Register
.WriteMask
;
865 if (writemask
& TGSI_WRITEMASK_Y
) {
868 if (writemask
& TGSI_WRITEMASK_X
)
869 tmp1
= i915_get_utemp(p
);
875 tmp1
, A0_DEST_CHANNEL_ALL
, 0,
876 swizzle(tmp
, X
, Y
, Y
, W
),
877 swizzle(tmp
, X
, Z
, ONE
, ONE
), 0);
881 get_result_vector(p
, &inst
->Dst
[0]),
882 A0_DEST_CHANNEL_Y
, 0,
883 swizzle(tmp1
, W
, Z
, Y
, X
),
884 i915_emit_const4fv(p
, scs_sin_constants
), 0);
887 if (writemask
& TGSI_WRITEMASK_X
) {
890 tmp
, A0_DEST_CHANNEL_XYZ
, 0,
891 swizzle(tmp
, X
, X
, Z
, ONE
),
892 swizzle(tmp
, Z
, ONE
, ONE
, ONE
), 0);
896 get_result_vector(p
, &inst
->Dst
[0]),
897 A0_DEST_CHANNEL_X
, 0,
898 swizzle(tmp
, ONE
, Z
, Y
, X
),
899 i915_emit_const4fv(p
, scs_cos_constants
), 0);
903 case TGSI_OPCODE_SEQ
:
904 /* if we're both >= and <= then we're == */
905 src0
= src_vector(p
, &inst
->Src
[0], fs
);
906 src1
= src_vector(p
, &inst
->Src
[1], fs
);
907 tmp
= i915_get_utemp(p
);
911 tmp
, A0_DEST_CHANNEL_ALL
, 0,
917 get_result_vector(p
, &inst
->Dst
[0]),
918 A0_DEST_CHANNEL_ALL
, 0,
924 get_result_vector(p
, &inst
->Dst
[0]),
925 A0_DEST_CHANNEL_ALL
, 0,
926 get_result_vector(p
, &inst
->Dst
[0]),
931 case TGSI_OPCODE_SGE
:
932 emit_simple_arith(p
, inst
, A0_SGE
, 2, fs
);
935 case TGSI_OPCODE_SIN
:
936 src0
= src_vector(p
, &inst
->Src
[0], fs
);
937 tmp
= i915_get_utemp(p
);
941 tmp
, A0_DEST_CHANNEL_X
, 0,
942 src0
, i915_emit_const1f(p
, 1.0f
/ (float) (M_PI
* 2.0)), 0);
944 i915_emit_arith(p
, A0_MOD
, tmp
, A0_DEST_CHANNEL_X
, 0, tmp
, 0, 0);
947 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
948 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
949 * t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x
950 * result = DP4 t1.wzyx, sin_constants
954 tmp
, A0_DEST_CHANNEL_XY
, 0,
955 swizzle(tmp
, X
, X
, ONE
, ONE
),
956 swizzle(tmp
, X
, ONE
, ONE
, ONE
), 0);
960 tmp
, A0_DEST_CHANNEL_ALL
, 0,
961 swizzle(tmp
, X
, Y
, X
, Y
),
962 swizzle(tmp
, X
, X
, ONE
, ONE
), 0);
966 tmp
, A0_DEST_CHANNEL_ALL
, 0,
967 swizzle(tmp
, X
, Y
, Y
, W
),
968 swizzle(tmp
, X
, Z
, ONE
, ONE
), 0);
972 get_result_vector(p
, &inst
->Dst
[0]),
973 get_result_flags(inst
), 0,
974 swizzle(tmp
, W
, Z
, Y
, X
),
975 i915_emit_const4fv(p
, sin_constants
), 0);
978 case TGSI_OPCODE_SLE
:
979 /* like SGE, but swap reg0, reg1 */
980 emit_simple_arith_swap2(p
, inst
, A0_SGE
, 2, fs
);
983 case TGSI_OPCODE_SLT
:
984 emit_simple_arith(p
, inst
, A0_SLT
, 2, fs
);
987 case TGSI_OPCODE_SGT
:
988 /* like SLT, but swap reg0, reg1 */
989 emit_simple_arith_swap2(p
, inst
, A0_SLT
, 2, fs
);
992 case TGSI_OPCODE_SNE
:
993 /* if we're < or > then we're != */
994 src0
= src_vector(p
, &inst
->Src
[0], fs
);
995 src1
= src_vector(p
, &inst
->Src
[1], fs
);
996 tmp
= i915_get_utemp(p
);
1001 A0_DEST_CHANNEL_ALL
, 0,
1007 get_result_vector(p
, &inst
->Dst
[0]),
1008 A0_DEST_CHANNEL_ALL
, 0,
1014 get_result_vector(p
, &inst
->Dst
[0]),
1015 A0_DEST_CHANNEL_ALL
, 0,
1016 get_result_vector(p
, &inst
->Dst
[0]),
1020 case TGSI_OPCODE_SSG
:
1021 /* compute (src>0) - (src<0) */
1022 src0
= src_vector(p
, &inst
->Src
[0], fs
);
1023 tmp
= i915_get_utemp(p
);
1028 A0_DEST_CHANNEL_ALL
, 0,
1030 swizzle(src0
, ZERO
, ZERO
, ZERO
, ZERO
), 0);
1034 get_result_vector(p
, &inst
->Dst
[0]),
1035 A0_DEST_CHANNEL_ALL
, 0,
1036 swizzle(src0
, ZERO
, ZERO
, ZERO
, ZERO
),
1041 get_result_vector(p
, &inst
->Dst
[0]),
1042 A0_DEST_CHANNEL_ALL
, 0,
1043 get_result_vector(p
, &inst
->Dst
[0]),
1044 negate(tmp
, 1, 1, 1, 1), 0);
1047 case TGSI_OPCODE_SUB
:
1048 src0
= src_vector(p
, &inst
->Src
[0], fs
);
1049 src1
= src_vector(p
, &inst
->Src
[1], fs
);
1053 get_result_vector(p
, &inst
->Dst
[0]),
1054 get_result_flags(inst
), 0,
1055 src0
, negate(src1
, 1, 1, 1, 1), 0);
1058 case TGSI_OPCODE_TEX
:
1059 emit_tex(p
, inst
, T0_TEXLD
, fs
);
1062 case TGSI_OPCODE_TRUNC
:
1063 emit_simple_arith(p
, inst
, A0_TRC
, 1, fs
);
1066 case TGSI_OPCODE_TXB
:
1067 emit_tex(p
, inst
, T0_TEXLDB
, fs
);
1070 case TGSI_OPCODE_TXP
:
1071 emit_tex(p
, inst
, T0_TEXLDP
, fs
);
1074 case TGSI_OPCODE_XPD
:
1076 * result.x = src0.y * src1.z - src0.z * src1.y;
1077 * result.y = src0.z * src1.x - src0.x * src1.z;
1078 * result.z = src0.x * src1.y - src0.y * src1.x;
1081 src0
= src_vector(p
, &inst
->Src
[0], fs
);
1082 src1
= src_vector(p
, &inst
->Src
[1], fs
);
1083 tmp
= i915_get_utemp(p
);
1087 tmp
, A0_DEST_CHANNEL_ALL
, 0,
1088 swizzle(src0
, Z
, X
, Y
, ONE
),
1089 swizzle(src1
, Y
, Z
, X
, ONE
), 0);
1093 get_result_vector(p
, &inst
->Dst
[0]),
1094 get_result_flags(inst
), 0,
1095 swizzle(src0
, Y
, Z
, X
, ONE
),
1096 swizzle(src1
, Z
, X
, Y
, ONE
),
1097 negate(tmp
, 1, 1, 1, 0));
1101 i915_program_error(p
, "bad opcode %d", inst
->Instruction
.Opcode
);
1106 i915_release_utemps(p
);
1110 static void i915_translate_token(struct i915_fp_compile
*p
,
1111 const union i915_full_token
* token
,
1112 struct i915_fragment_shader
*fs
)
1114 struct i915_fragment_shader
*ifs
= p
->shader
;
1115 switch( token
->Token
.Type
) {
1116 case TGSI_TOKEN_TYPE_PROPERTY
:
1118 * We only support one cbuf, but we still need to ignore the property
1119 * correctly so we don't hit the assert at the end of the switch case.
1121 assert(token
->FullProperty
.Property
.PropertyName
==
1122 TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS
);
1125 case TGSI_TOKEN_TYPE_DECLARATION
:
1126 if (token
->FullDeclaration
.Declaration
.File
1127 == TGSI_FILE_CONSTANT
) {
1129 for (i
= token
->FullDeclaration
.Range
.First
;
1130 i
<= token
->FullDeclaration
.Range
.Last
;
1132 assert(ifs
->constant_flags
[i
] == 0x0);
1133 ifs
->constant_flags
[i
] = I915_CONSTFLAG_USER
;
1134 ifs
->num_constants
= MAX2(ifs
->num_constants
, i
+ 1);
1137 else if (token
->FullDeclaration
.Declaration
.File
1138 == TGSI_FILE_TEMPORARY
) {
1140 for (i
= token
->FullDeclaration
.Range
.First
;
1141 i
<= token
->FullDeclaration
.Range
.Last
;
1143 if (i
>= I915_MAX_TEMPORARY
)
1144 debug_printf("Too many temps (%d)\n",i
);
1146 /* XXX just use shader->info->file_mask[TGSI_FILE_TEMPORARY] */
1147 p
->temp_flag
|= (1 << i
); /* mark temp as used */
1152 case TGSI_TOKEN_TYPE_IMMEDIATE
:
1154 const struct tgsi_full_immediate
*imm
1155 = &token
->FullImmediate
;
1156 const uint pos
= p
->num_immediates
++;
1158 assert( imm
->Immediate
.NrTokens
<= 4 + 1 );
1159 for (j
= 0; j
< imm
->Immediate
.NrTokens
- 1; j
++) {
1160 p
->immediates
[pos
][j
] = imm
->u
[j
].Float
;
1165 case TGSI_TOKEN_TYPE_INSTRUCTION
:
1166 if (p
->first_instruction
) {
1167 /* resolve location of immediates */
1169 for (i
= 0; i
< p
->num_immediates
; i
++) {
1170 /* find constant slot for this immediate */
1171 for (j
= 0; j
< I915_MAX_CONSTANT
; j
++) {
1172 if (ifs
->constant_flags
[j
] == 0x0) {
1173 memcpy(ifs
->constants
[j
],
1176 /*printf("immediate %d maps to const %d\n", i, j);*/
1177 ifs
->constant_flags
[j
] = 0xf; /* all four comps used */
1178 p
->immediates_map
[i
] = j
;
1179 ifs
->num_constants
= MAX2(ifs
->num_constants
, j
+ 1);
1185 p
->first_instruction
= FALSE
;
1188 i915_translate_instruction(p
, &token
->FullInstruction
, fs
);
1198 * Translate TGSI fragment shader into i915 hardware instructions.
1199 * \param p the translation state
1200 * \param tokens the TGSI token array
1203 i915_translate_instructions(struct i915_fp_compile
*p
,
1204 const struct i915_token_list
*tokens
,
1205 struct i915_fragment_shader
*fs
)
1208 for(i
= 0; i
<tokens
->NumTokens
; i
++) {
1209 i915_translate_token(p
, &tokens
->Tokens
[i
], fs
);
1214 static struct i915_fp_compile
*
1215 i915_init_compile(struct i915_context
*i915
,
1216 struct i915_fragment_shader
*ifs
)
1218 struct i915_fp_compile
*p
= CALLOC_STRUCT(i915_fp_compile
);
1223 /* Put new constants at end of const buffer, growing downward.
1224 * The problem is we don't know how many user-defined constants might
1225 * be specified with pipe->set_constant_buffer().
1226 * Should pre-scan the user's program to determine the highest-numbered
1227 * constant referenced.
1229 ifs
->num_constants
= 0;
1230 memset(ifs
->constant_flags
, 0, sizeof(ifs
->constant_flags
));
1232 memset(&p
->register_phases
, 0, sizeof(p
->register_phases
));
1234 for (i
= 0; i
< I915_TEX_UNITS
; i
++)
1235 ifs
->generic_mapping
[i
] = -1;
1237 p
->first_instruction
= TRUE
;
1239 p
->nr_tex_indirect
= 1; /* correct? */
1242 p
->nr_decl_insn
= 0;
1244 p
->csr
= p
->program
;
1245 p
->decl
= p
->declarations
;
1248 p
->temp_flag
= ~0x0 << I915_MAX_TEMPORARY
;
1249 p
->utemp_flag
= ~0x7;
1251 /* initialize the first program word */
1252 *(p
->decl
++) = _3DSTATE_PIXEL_SHADER_PROGRAM
;
1258 /* Copy compile results to the fragment program struct and destroy the
1259 * compilation context.
1262 i915_fini_compile(struct i915_context
*i915
, struct i915_fp_compile
*p
)
1264 struct i915_fragment_shader
*ifs
= p
->shader
;
1265 unsigned long program_size
= (unsigned long) (p
->csr
- p
->program
);
1266 unsigned long decl_size
= (unsigned long) (p
->decl
- p
->declarations
);
1268 if (p
->nr_tex_indirect
> I915_MAX_TEX_INDIRECT
)
1269 debug_printf("Exceeded max nr indirect texture lookups\n");
1271 if (p
->nr_tex_insn
> I915_MAX_TEX_INSN
)
1272 i915_program_error(p
, "Exceeded max TEX instructions");
1274 if (p
->nr_alu_insn
> I915_MAX_ALU_INSN
)
1275 i915_program_error(p
, "Exceeded max ALU instructions");
1277 if (p
->nr_decl_insn
> I915_MAX_DECL_INSN
)
1278 i915_program_error(p
, "Exceeded max DECL instructions");
1281 p
->NumNativeInstructions
= 0;
1282 p
->NumNativeAluInstructions
= 0;
1283 p
->NumNativeTexInstructions
= 0;
1284 p
->NumNativeTexIndirections
= 0;
1286 i915_use_passthrough_shader(ifs
);
1289 p
->NumNativeInstructions
1290 = p
->nr_alu_insn
+ p
->nr_tex_insn
+ p
->nr_decl_insn
;
1291 p
->NumNativeAluInstructions
= p
->nr_alu_insn
;
1292 p
->NumNativeTexInstructions
= p
->nr_tex_insn
;
1293 p
->NumNativeTexIndirections
= p
->nr_tex_indirect
;
1295 /* patch in the program length */
1296 p
->declarations
[0] |= program_size
+ decl_size
- 2;
1298 /* Copy compilation results to fragment program struct:
1301 assert(!ifs
->program
);
1304 = (uint
*) MALLOC(decl_size
* sizeof(uint
));
1306 = (uint
*) MALLOC(program_size
* sizeof(uint
));
1309 ifs
->decl_len
= decl_size
;
1313 decl_size
* sizeof(uint
));
1317 ifs
->program_len
= program_size
;
1319 memcpy(ifs
->program
,
1321 program_size
* sizeof(uint
));
1325 /* Release the compilation struct:
1335 * Rather than trying to intercept and jiggle depth writes during
1336 * emit, just move the value into its correct position at the end of
1340 i915_fixup_depth_write(struct i915_fp_compile
*p
)
1342 /* XXX assuming pos/depth is always in output[0] */
1343 if (p
->shader
->info
.output_semantic_name
[0] == TGSI_SEMANTIC_POSITION
) {
1344 const uint depth
= UREG(REG_TYPE_OD
, 0);
1347 A0_MOV
, /* opcode */
1348 depth
, /* dest reg */
1349 A0_DEST_CHANNEL_W
, /* write mask */
1351 swizzle(depth
, X
, Y
, Z
, Z
), /* src0 */
1352 0, 0 /* src1, src2 */);
1358 i915_translate_fragment_program( struct i915_context
*i915
,
1359 struct i915_fragment_shader
*fs
)
1361 struct i915_fp_compile
*p
;
1362 const struct tgsi_token
*tokens
= fs
->state
.tokens
;
1363 struct i915_token_list
* i_tokens
;
1366 tgsi_dump(tokens
, 0);
1369 /* hw doesn't seem to like empty frag programs, even when the depth write
1370 * fixup gets emitted below - may that one is fishy, too? */
1371 if (fs
->info
.num_instructions
== 1) {
1372 i915_use_passthrough_shader(fs
);
1377 p
= i915_init_compile(i915
, fs
);
1379 i_tokens
= i915_optimize(tokens
);
1380 i915_translate_instructions(p
, i_tokens
, fs
);
1381 i915_fixup_depth_write(p
);
1383 i915_fini_compile(i915
, p
);
1384 i915_optimize_free(i_tokens
);
1387 i915_disassemble_program(NULL
, fs
->program
, fs
->program_len
);