1 /**************************************************************************
3 * Copyright 2007-2008 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 **************************************************************************/
29 * TGSI interpreter/executor.
31 * Flow control information:
33 * Since we operate on 'quads' (4 pixels or 4 vertices in parallel)
34 * flow control statements (IF/ELSE/ENDIF, LOOP/ENDLOOP) require special
35 * care since a condition may be true for some quad components but false
36 * for other components.
38 * We basically execute all statements (even if they're in the part of
39 * an IF/ELSE clause that's "not taken") and use a special mask to
40 * control writing to destination registers. This is the ExecMask.
43 * The ExecMask is computed from three other masks (CondMask, LoopMask and
44 * ContMask) which are controlled by the flow control instructions (namely:
45 * (IF/ELSE/ENDIF, LOOP/ENDLOOP and CONT).
53 #include "pipe/p_compiler.h"
54 #include "pipe/p_state.h"
55 #include "pipe/p_shader_tokens.h"
56 #include "tgsi/tgsi_parse.h"
57 #include "tgsi/tgsi_util.h"
58 #include "tgsi_exec.h"
59 #include "util/u_memory.h"
60 #include "util/u_math.h"
64 #define TILE_TOP_LEFT 0
65 #define TILE_TOP_RIGHT 1
66 #define TILE_BOTTOM_LEFT 2
67 #define TILE_BOTTOM_RIGHT 3
75 * Shorthand locations of various utility registers (_I = Index, _C = Channel)
77 #define TEMP_0_I TGSI_EXEC_TEMP_00000000_I
78 #define TEMP_0_C TGSI_EXEC_TEMP_00000000_C
79 #define TEMP_7F_I TGSI_EXEC_TEMP_7FFFFFFF_I
80 #define TEMP_7F_C TGSI_EXEC_TEMP_7FFFFFFF_C
81 #define TEMP_80_I TGSI_EXEC_TEMP_80000000_I
82 #define TEMP_80_C TGSI_EXEC_TEMP_80000000_C
83 #define TEMP_FF_I TGSI_EXEC_TEMP_FFFFFFFF_I
84 #define TEMP_FF_C TGSI_EXEC_TEMP_FFFFFFFF_C
85 #define TEMP_1_I TGSI_EXEC_TEMP_ONE_I
86 #define TEMP_1_C TGSI_EXEC_TEMP_ONE_C
87 #define TEMP_2_I TGSI_EXEC_TEMP_TWO_I
88 #define TEMP_2_C TGSI_EXEC_TEMP_TWO_C
89 #define TEMP_128_I TGSI_EXEC_TEMP_128_I
90 #define TEMP_128_C TGSI_EXEC_TEMP_128_C
91 #define TEMP_M128_I TGSI_EXEC_TEMP_MINUS_128_I
92 #define TEMP_M128_C TGSI_EXEC_TEMP_MINUS_128_C
93 #define TEMP_KILMASK_I TGSI_EXEC_TEMP_KILMASK_I
94 #define TEMP_KILMASK_C TGSI_EXEC_TEMP_KILMASK_C
95 #define TEMP_OUTPUT_I TGSI_EXEC_TEMP_OUTPUT_I
96 #define TEMP_OUTPUT_C TGSI_EXEC_TEMP_OUTPUT_C
97 #define TEMP_PRIMITIVE_I TGSI_EXEC_TEMP_PRIMITIVE_I
98 #define TEMP_PRIMITIVE_C TGSI_EXEC_TEMP_PRIMITIVE_C
99 #define TEMP_CC_I TGSI_EXEC_TEMP_CC_I
100 #define TEMP_CC_C TGSI_EXEC_TEMP_CC_C
101 #define TEMP_3_I TGSI_EXEC_TEMP_THREE_I
102 #define TEMP_3_C TGSI_EXEC_TEMP_THREE_C
103 #define TEMP_HALF_I TGSI_EXEC_TEMP_HALF_I
104 #define TEMP_HALF_C TGSI_EXEC_TEMP_HALF_C
105 #define TEMP_R0 TGSI_EXEC_TEMP_R0
107 #define IS_CHANNEL_ENABLED(INST, CHAN)\
108 ((INST).FullDstRegisters[0].DstRegister.WriteMask & (1 << (CHAN)))
110 #define IS_CHANNEL_ENABLED2(INST, CHAN)\
111 ((INST).FullDstRegisters[1].DstRegister.WriteMask & (1 << (CHAN)))
113 #define FOR_EACH_ENABLED_CHANNEL(INST, CHAN)\
114 for (CHAN = 0; CHAN < NUM_CHANNELS; CHAN++)\
115 if (IS_CHANNEL_ENABLED( INST, CHAN ))
117 #define FOR_EACH_ENABLED_CHANNEL2(INST, CHAN)\
118 for (CHAN = 0; CHAN < NUM_CHANNELS; CHAN++)\
119 if (IS_CHANNEL_ENABLED2( INST, CHAN ))
122 /** The execution mask depends on the conditional mask and the loop mask */
123 #define UPDATE_EXEC_MASK(MACH) \
124 MACH->ExecMask = MACH->CondMask & MACH->LoopMask & MACH->ContMask & MACH->FuncMask
127 * Initialize machine state by expanding tokens to full instructions,
128 * allocating temporary storage, setting up constants, etc.
129 * After this, we can call tgsi_exec_machine_run() many times.
132 tgsi_exec_machine_bind_shader(
133 struct tgsi_exec_machine
*mach
,
134 const struct tgsi_token
*tokens
,
136 struct tgsi_sampler
**samplers
)
139 struct tgsi_parse_context parse
;
140 struct tgsi_exec_labels
*labels
= &mach
->Labels
;
141 struct tgsi_full_instruction
*instructions
;
142 struct tgsi_full_declaration
*declarations
;
143 uint maxInstructions
= 10, numInstructions
= 0;
144 uint maxDeclarations
= 10, numDeclarations
= 0;
148 tgsi_dump(tokens
, 0);
153 mach
->Tokens
= tokens
;
154 mach
->Samplers
= samplers
;
156 k
= tgsi_parse_init (&parse
, mach
->Tokens
);
157 if (k
!= TGSI_PARSE_OK
) {
158 debug_printf( "Problem parsing!\n" );
162 mach
->Processor
= parse
.FullHeader
.Processor
.Processor
;
166 declarations
= (struct tgsi_full_declaration
*)
167 MALLOC( maxDeclarations
* sizeof(struct tgsi_full_declaration
) );
173 instructions
= (struct tgsi_full_instruction
*)
174 MALLOC( maxInstructions
* sizeof(struct tgsi_full_instruction
) );
177 FREE( declarations
);
181 while( !tgsi_parse_end_of_tokens( &parse
) ) {
182 uint pointer
= parse
.Position
;
185 tgsi_parse_token( &parse
);
186 switch( parse
.FullToken
.Token
.Type
) {
187 case TGSI_TOKEN_TYPE_DECLARATION
:
188 /* save expanded declaration */
189 if (numDeclarations
== maxDeclarations
) {
190 declarations
= REALLOC(declarations
,
192 * sizeof(struct tgsi_full_declaration
),
193 (maxDeclarations
+ 10)
194 * sizeof(struct tgsi_full_declaration
));
195 maxDeclarations
+= 10;
197 memcpy(declarations
+ numDeclarations
,
198 &parse
.FullToken
.FullDeclaration
,
199 sizeof(declarations
[0]));
203 case TGSI_TOKEN_TYPE_IMMEDIATE
:
205 uint size
= parse
.FullToken
.FullImmediate
.Immediate
.NrTokens
- 1;
206 assert( size
% 4 == 0 );
207 assert( mach
->ImmLimit
+ size
/ 4 <= TGSI_EXEC_NUM_IMMEDIATES
);
209 for( i
= 0; i
< size
; i
++ ) {
210 mach
->Imms
[mach
->ImmLimit
+ i
/ 4][i
% 4] =
211 parse
.FullToken
.FullImmediate
.u
.ImmediateFloat32
[i
].Float
;
213 mach
->ImmLimit
+= size
/ 4;
217 case TGSI_TOKEN_TYPE_INSTRUCTION
:
218 assert( labels
->count
< MAX_LABELS
);
220 labels
->labels
[labels
->count
][0] = instno
;
221 labels
->labels
[labels
->count
][1] = pointer
;
224 /* save expanded instruction */
225 if (numInstructions
== maxInstructions
) {
226 instructions
= REALLOC(instructions
,
228 * sizeof(struct tgsi_full_instruction
),
229 (maxInstructions
+ 10)
230 * sizeof(struct tgsi_full_instruction
));
231 maxInstructions
+= 10;
233 memcpy(instructions
+ numInstructions
,
234 &parse
.FullToken
.FullInstruction
,
235 sizeof(instructions
[0]));
243 tgsi_parse_free (&parse
);
245 if (mach
->Declarations
) {
246 FREE( mach
->Declarations
);
248 mach
->Declarations
= declarations
;
249 mach
->NumDeclarations
= numDeclarations
;
251 if (mach
->Instructions
) {
252 FREE( mach
->Instructions
);
254 mach
->Instructions
= instructions
;
255 mach
->NumInstructions
= numInstructions
;
260 tgsi_exec_machine_init(
261 struct tgsi_exec_machine
*mach
)
265 mach
->Temps
= (struct tgsi_exec_vector
*) tgsi_align_128bit( mach
->_Temps
);
266 mach
->Addrs
= &mach
->Temps
[TGSI_EXEC_TEMP_ADDR
];
268 /* Setup constants. */
269 for( i
= 0; i
< 4; i
++ ) {
270 mach
->Temps
[TEMP_0_I
].xyzw
[TEMP_0_C
].u
[i
] = 0x00000000;
271 mach
->Temps
[TEMP_7F_I
].xyzw
[TEMP_7F_C
].u
[i
] = 0x7FFFFFFF;
272 mach
->Temps
[TEMP_80_I
].xyzw
[TEMP_80_C
].u
[i
] = 0x80000000;
273 mach
->Temps
[TEMP_FF_I
].xyzw
[TEMP_FF_C
].u
[i
] = 0xFFFFFFFF;
274 mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
].f
[i
] = 1.0f
;
275 mach
->Temps
[TEMP_2_I
].xyzw
[TEMP_2_C
].f
[i
] = 2.0f
;
276 mach
->Temps
[TEMP_128_I
].xyzw
[TEMP_128_C
].f
[i
] = 128.0f
;
277 mach
->Temps
[TEMP_M128_I
].xyzw
[TEMP_M128_C
].f
[i
] = -128.0f
;
278 mach
->Temps
[TEMP_3_I
].xyzw
[TEMP_3_C
].f
[i
] = 3.0f
;
279 mach
->Temps
[TEMP_HALF_I
].xyzw
[TEMP_HALF_C
].f
[i
] = 0.5f
;
285 tgsi_exec_machine_free_data(struct tgsi_exec_machine
*mach
)
287 if (mach
->Instructions
) {
288 FREE(mach
->Instructions
);
289 mach
->Instructions
= NULL
;
290 mach
->NumInstructions
= 0;
292 if (mach
->Declarations
) {
293 FREE(mach
->Declarations
);
294 mach
->Declarations
= NULL
;
295 mach
->NumDeclarations
= 0;
302 union tgsi_exec_channel
*dst
,
303 const union tgsi_exec_channel
*src
)
305 dst
->f
[0] = fabsf( src
->f
[0] );
306 dst
->f
[1] = fabsf( src
->f
[1] );
307 dst
->f
[2] = fabsf( src
->f
[2] );
308 dst
->f
[3] = fabsf( src
->f
[3] );
313 union tgsi_exec_channel
*dst
,
314 const union tgsi_exec_channel
*src0
,
315 const union tgsi_exec_channel
*src1
)
317 dst
->f
[0] = src0
->f
[0] + src1
->f
[0];
318 dst
->f
[1] = src0
->f
[1] + src1
->f
[1];
319 dst
->f
[2] = src0
->f
[2] + src1
->f
[2];
320 dst
->f
[3] = src0
->f
[3] + src1
->f
[3];
326 union tgsi_exec_channel
*dst
,
327 const union tgsi_exec_channel
*src0
,
328 const union tgsi_exec_channel
*src1
)
330 dst
->i
[0] = src0
->i
[0] + src1
->i
[0];
331 dst
->i
[1] = src0
->i
[1] + src1
->i
[1];
332 dst
->i
[2] = src0
->i
[2] + src1
->i
[2];
333 dst
->i
[3] = src0
->i
[3] + src1
->i
[3];
339 union tgsi_exec_channel
*dst
,
340 const union tgsi_exec_channel
*src0
,
341 const union tgsi_exec_channel
*src1
)
343 dst
->u
[0] = src0
->u
[0] & src1
->u
[0];
344 dst
->u
[1] = src0
->u
[1] & src1
->u
[1];
345 dst
->u
[2] = src0
->u
[2] & src1
->u
[2];
346 dst
->u
[3] = src0
->u
[3] & src1
->u
[3];
351 union tgsi_exec_channel
*dst
,
352 const union tgsi_exec_channel
*src
)
354 dst
->f
[0] = ceilf( src
->f
[0] );
355 dst
->f
[1] = ceilf( src
->f
[1] );
356 dst
->f
[2] = ceilf( src
->f
[2] );
357 dst
->f
[3] = ceilf( src
->f
[3] );
362 union tgsi_exec_channel
*dst
,
363 const union tgsi_exec_channel
*src
)
365 dst
->f
[0] = cosf( src
->f
[0] );
366 dst
->f
[1] = cosf( src
->f
[1] );
367 dst
->f
[2] = cosf( src
->f
[2] );
368 dst
->f
[3] = cosf( src
->f
[3] );
373 union tgsi_exec_channel
*dst
,
374 const union tgsi_exec_channel
*src
)
379 dst
->f
[3] = src
->f
[TILE_BOTTOM_RIGHT
] - src
->f
[TILE_BOTTOM_LEFT
];
384 union tgsi_exec_channel
*dst
,
385 const union tgsi_exec_channel
*src
)
390 dst
->f
[3] = src
->f
[TILE_TOP_LEFT
] - src
->f
[TILE_BOTTOM_LEFT
];
395 union tgsi_exec_channel
*dst
,
396 const union tgsi_exec_channel
*src0
,
397 const union tgsi_exec_channel
*src1
)
399 if (src1
->f
[0] != 0) {
400 dst
->f
[0] = src0
->f
[0] / src1
->f
[0];
402 if (src1
->f
[1] != 0) {
403 dst
->f
[1] = src0
->f
[1] / src1
->f
[1];
405 if (src1
->f
[2] != 0) {
406 dst
->f
[2] = src0
->f
[2] / src1
->f
[2];
408 if (src1
->f
[3] != 0) {
409 dst
->f
[3] = src0
->f
[3] / src1
->f
[3];
416 union tgsi_exec_channel
*dst
,
417 const union tgsi_exec_channel
*src0
,
418 const union tgsi_exec_channel
*src1
)
420 dst
->u
[0] = src0
->u
[0] / src1
->u
[0];
421 dst
->u
[1] = src0
->u
[1] / src1
->u
[1];
422 dst
->u
[2] = src0
->u
[2] / src1
->u
[2];
423 dst
->u
[3] = src0
->u
[3] / src1
->u
[3];
429 union tgsi_exec_channel
*dst
,
430 const union tgsi_exec_channel
*src0
,
431 const union tgsi_exec_channel
*src1
,
432 const union tgsi_exec_channel
*src2
,
433 const union tgsi_exec_channel
*src3
)
435 dst
->f
[0] = src0
->f
[0] == src1
->f
[0] ? src2
->f
[0] : src3
->f
[0];
436 dst
->f
[1] = src0
->f
[1] == src1
->f
[1] ? src2
->f
[1] : src3
->f
[1];
437 dst
->f
[2] = src0
->f
[2] == src1
->f
[2] ? src2
->f
[2] : src3
->f
[2];
438 dst
->f
[3] = src0
->f
[3] == src1
->f
[3] ? src2
->f
[3] : src3
->f
[3];
444 union tgsi_exec_channel
*dst
,
445 const union tgsi_exec_channel
*src0
,
446 const union tgsi_exec_channel
*src1
,
447 const union tgsi_exec_channel
*src2
,
448 const union tgsi_exec_channel
*src3
)
450 dst
->i
[0] = src0
->i
[0] == src1
->i
[0] ? src2
->i
[0] : src3
->i
[0];
451 dst
->i
[1] = src0
->i
[1] == src1
->i
[1] ? src2
->i
[1] : src3
->i
[1];
452 dst
->i
[2] = src0
->i
[2] == src1
->i
[2] ? src2
->i
[2] : src3
->i
[2];
453 dst
->i
[3] = src0
->i
[3] == src1
->i
[3] ? src2
->i
[3] : src3
->i
[3];
459 union tgsi_exec_channel
*dst
,
460 const union tgsi_exec_channel
*src
)
463 dst
->f
[0] = util_fast_exp2( src
->f
[0] );
464 dst
->f
[1] = util_fast_exp2( src
->f
[1] );
465 dst
->f
[2] = util_fast_exp2( src
->f
[2] );
466 dst
->f
[3] = util_fast_exp2( src
->f
[3] );
468 dst
->f
[0] = powf( 2.0f
, src
->f
[0] );
469 dst
->f
[1] = powf( 2.0f
, src
->f
[1] );
470 dst
->f
[2] = powf( 2.0f
, src
->f
[2] );
471 dst
->f
[3] = powf( 2.0f
, src
->f
[3] );
478 union tgsi_exec_channel
*dst
,
479 const union tgsi_exec_channel
*src
)
481 dst
->u
[0] = (uint
) src
->f
[0];
482 dst
->u
[1] = (uint
) src
->f
[1];
483 dst
->u
[2] = (uint
) src
->f
[2];
484 dst
->u
[3] = (uint
) src
->f
[3];
490 union tgsi_exec_channel
*dst
,
491 const union tgsi_exec_channel
*src
)
493 dst
->f
[0] = floorf( src
->f
[0] );
494 dst
->f
[1] = floorf( src
->f
[1] );
495 dst
->f
[2] = floorf( src
->f
[2] );
496 dst
->f
[3] = floorf( src
->f
[3] );
501 union tgsi_exec_channel
*dst
,
502 const union tgsi_exec_channel
*src
)
504 dst
->f
[0] = src
->f
[0] - floorf( src
->f
[0] );
505 dst
->f
[1] = src
->f
[1] - floorf( src
->f
[1] );
506 dst
->f
[2] = src
->f
[2] - floorf( src
->f
[2] );
507 dst
->f
[3] = src
->f
[3] - floorf( src
->f
[3] );
512 union tgsi_exec_channel
*dst
,
513 const union tgsi_exec_channel
*src0
,
514 const union tgsi_exec_channel
*src1
,
515 const union tgsi_exec_channel
*src2
,
516 const union tgsi_exec_channel
*src3
)
518 dst
->f
[0] = src0
->f
[0] >= src1
->f
[0] ? src2
->f
[0] : src3
->f
[0];
519 dst
->f
[1] = src0
->f
[1] >= src1
->f
[1] ? src2
->f
[1] : src3
->f
[1];
520 dst
->f
[2] = src0
->f
[2] >= src1
->f
[2] ? src2
->f
[2] : src3
->f
[2];
521 dst
->f
[3] = src0
->f
[3] >= src1
->f
[3] ? src2
->f
[3] : src3
->f
[3];
526 union tgsi_exec_channel
*dst
,
527 const union tgsi_exec_channel
*src
)
529 dst
->f
[0] = (float) src
->i
[0];
530 dst
->f
[1] = (float) src
->i
[1];
531 dst
->f
[2] = (float) src
->i
[2];
532 dst
->f
[3] = (float) src
->i
[3];
537 union tgsi_exec_channel
*dst
,
538 const union tgsi_exec_channel
*src
)
541 dst
->f
[0] = util_fast_log2( src
->f
[0] );
542 dst
->f
[1] = util_fast_log2( src
->f
[1] );
543 dst
->f
[2] = util_fast_log2( src
->f
[2] );
544 dst
->f
[3] = util_fast_log2( src
->f
[3] );
546 dst
->f
[0] = logf( src
->f
[0] ) * 1.442695f
;
547 dst
->f
[1] = logf( src
->f
[1] ) * 1.442695f
;
548 dst
->f
[2] = logf( src
->f
[2] ) * 1.442695f
;
549 dst
->f
[3] = logf( src
->f
[3] ) * 1.442695f
;
555 union tgsi_exec_channel
*dst
,
556 const union tgsi_exec_channel
*src0
,
557 const union tgsi_exec_channel
*src1
,
558 const union tgsi_exec_channel
*src2
,
559 const union tgsi_exec_channel
*src3
)
561 dst
->f
[0] = src0
->f
[0] <= src1
->f
[0] ? src2
->f
[0] : src3
->f
[0];
562 dst
->f
[1] = src0
->f
[1] <= src1
->f
[1] ? src2
->f
[1] : src3
->f
[1];
563 dst
->f
[2] = src0
->f
[2] <= src1
->f
[2] ? src2
->f
[2] : src3
->f
[2];
564 dst
->f
[3] = src0
->f
[3] <= src1
->f
[3] ? src2
->f
[3] : src3
->f
[3];
569 union tgsi_exec_channel
*dst
,
570 const union tgsi_exec_channel
*src0
,
571 const union tgsi_exec_channel
*src1
,
572 const union tgsi_exec_channel
*src2
,
573 const union tgsi_exec_channel
*src3
)
575 dst
->f
[0] = src0
->f
[0] < src1
->f
[0] ? src2
->f
[0] : src3
->f
[0];
576 dst
->f
[1] = src0
->f
[1] < src1
->f
[1] ? src2
->f
[1] : src3
->f
[1];
577 dst
->f
[2] = src0
->f
[2] < src1
->f
[2] ? src2
->f
[2] : src3
->f
[2];
578 dst
->f
[3] = src0
->f
[3] < src1
->f
[3] ? src2
->f
[3] : src3
->f
[3];
584 union tgsi_exec_channel
*dst
,
585 const union tgsi_exec_channel
*src0
,
586 const union tgsi_exec_channel
*src1
,
587 const union tgsi_exec_channel
*src2
,
588 const union tgsi_exec_channel
*src3
)
590 dst
->i
[0] = src0
->i
[0] < src1
->i
[0] ? src2
->i
[0] : src3
->i
[0];
591 dst
->i
[1] = src0
->i
[1] < src1
->i
[1] ? src2
->i
[1] : src3
->i
[1];
592 dst
->i
[2] = src0
->i
[2] < src1
->i
[2] ? src2
->i
[2] : src3
->i
[2];
593 dst
->i
[3] = src0
->i
[3] < src1
->i
[3] ? src2
->i
[3] : src3
->i
[3];
600 union tgsi_exec_channel
*dst
,
601 const union tgsi_exec_channel
*src0
,
602 const union tgsi_exec_channel
*src1
,
603 const union tgsi_exec_channel
*src2
,
604 const union tgsi_exec_channel
*src3
)
606 dst
->u
[0] = src0
->u
[0] < src1
->u
[0] ? src2
->u
[0] : src3
->u
[0];
607 dst
->u
[1] = src0
->u
[1] < src1
->u
[1] ? src2
->u
[1] : src3
->u
[1];
608 dst
->u
[2] = src0
->u
[2] < src1
->u
[2] ? src2
->u
[2] : src3
->u
[2];
609 dst
->u
[3] = src0
->u
[3] < src1
->u
[3] ? src2
->u
[3] : src3
->u
[3];
615 union tgsi_exec_channel
*dst
,
616 const union tgsi_exec_channel
*src0
,
617 const union tgsi_exec_channel
*src1
)
619 dst
->f
[0] = src0
->f
[0] > src1
->f
[0] ? src0
->f
[0] : src1
->f
[0];
620 dst
->f
[1] = src0
->f
[1] > src1
->f
[1] ? src0
->f
[1] : src1
->f
[1];
621 dst
->f
[2] = src0
->f
[2] > src1
->f
[2] ? src0
->f
[2] : src1
->f
[2];
622 dst
->f
[3] = src0
->f
[3] > src1
->f
[3] ? src0
->f
[3] : src1
->f
[3];
628 union tgsi_exec_channel
*dst
,
629 const union tgsi_exec_channel
*src0
,
630 const union tgsi_exec_channel
*src1
)
632 dst
->i
[0] = src0
->i
[0] > src1
->i
[0] ? src0
->i
[0] : src1
->i
[0];
633 dst
->i
[1] = src0
->i
[1] > src1
->i
[1] ? src0
->i
[1] : src1
->i
[1];
634 dst
->i
[2] = src0
->i
[2] > src1
->i
[2] ? src0
->i
[2] : src1
->i
[2];
635 dst
->i
[3] = src0
->i
[3] > src1
->i
[3] ? src0
->i
[3] : src1
->i
[3];
642 union tgsi_exec_channel
*dst
,
643 const union tgsi_exec_channel
*src0
,
644 const union tgsi_exec_channel
*src1
)
646 dst
->u
[0] = src0
->u
[0] > src1
->u
[0] ? src0
->u
[0] : src1
->u
[0];
647 dst
->u
[1] = src0
->u
[1] > src1
->u
[1] ? src0
->u
[1] : src1
->u
[1];
648 dst
->u
[2] = src0
->u
[2] > src1
->u
[2] ? src0
->u
[2] : src1
->u
[2];
649 dst
->u
[3] = src0
->u
[3] > src1
->u
[3] ? src0
->u
[3] : src1
->u
[3];
655 union tgsi_exec_channel
*dst
,
656 const union tgsi_exec_channel
*src0
,
657 const union tgsi_exec_channel
*src1
)
659 dst
->f
[0] = src0
->f
[0] < src1
->f
[0] ? src0
->f
[0] : src1
->f
[0];
660 dst
->f
[1] = src0
->f
[1] < src1
->f
[1] ? src0
->f
[1] : src1
->f
[1];
661 dst
->f
[2] = src0
->f
[2] < src1
->f
[2] ? src0
->f
[2] : src1
->f
[2];
662 dst
->f
[3] = src0
->f
[3] < src1
->f
[3] ? src0
->f
[3] : src1
->f
[3];
668 union tgsi_exec_channel
*dst
,
669 const union tgsi_exec_channel
*src0
,
670 const union tgsi_exec_channel
*src1
)
672 dst
->i
[0] = src0
->i
[0] < src1
->i
[0] ? src0
->i
[0] : src1
->i
[0];
673 dst
->i
[1] = src0
->i
[1] < src1
->i
[1] ? src0
->i
[1] : src1
->i
[1];
674 dst
->i
[2] = src0
->i
[2] < src1
->i
[2] ? src0
->i
[2] : src1
->i
[2];
675 dst
->i
[3] = src0
->i
[3] < src1
->i
[3] ? src0
->i
[3] : src1
->i
[3];
682 union tgsi_exec_channel
*dst
,
683 const union tgsi_exec_channel
*src0
,
684 const union tgsi_exec_channel
*src1
)
686 dst
->u
[0] = src0
->u
[0] < src1
->u
[0] ? src0
->u
[0] : src1
->u
[0];
687 dst
->u
[1] = src0
->u
[1] < src1
->u
[1] ? src0
->u
[1] : src1
->u
[1];
688 dst
->u
[2] = src0
->u
[2] < src1
->u
[2] ? src0
->u
[2] : src1
->u
[2];
689 dst
->u
[3] = src0
->u
[3] < src1
->u
[3] ? src0
->u
[3] : src1
->u
[3];
696 union tgsi_exec_channel
*dst
,
697 const union tgsi_exec_channel
*src0
,
698 const union tgsi_exec_channel
*src1
)
700 dst
->u
[0] = src0
->u
[0] % src1
->u
[0];
701 dst
->u
[1] = src0
->u
[1] % src1
->u
[1];
702 dst
->u
[2] = src0
->u
[2] % src1
->u
[2];
703 dst
->u
[3] = src0
->u
[3] % src1
->u
[3];
709 union tgsi_exec_channel
*dst
,
710 const union tgsi_exec_channel
*src0
,
711 const union tgsi_exec_channel
*src1
)
713 dst
->f
[0] = src0
->f
[0] * src1
->f
[0];
714 dst
->f
[1] = src0
->f
[1] * src1
->f
[1];
715 dst
->f
[2] = src0
->f
[2] * src1
->f
[2];
716 dst
->f
[3] = src0
->f
[3] * src1
->f
[3];
722 union tgsi_exec_channel
*dst
,
723 const union tgsi_exec_channel
*src0
,
724 const union tgsi_exec_channel
*src1
)
726 dst
->i
[0] = src0
->i
[0] * src1
->i
[0];
727 dst
->i
[1] = src0
->i
[1] * src1
->i
[1];
728 dst
->i
[2] = src0
->i
[2] * src1
->i
[2];
729 dst
->i
[3] = src0
->i
[3] * src1
->i
[3];
736 union tgsi_exec_channel
*dst0
,
737 union tgsi_exec_channel
*dst1
,
738 const union tgsi_exec_channel
*src0
,
739 const union tgsi_exec_channel
*src1
)
741 dst1
->i
[0] = src0
->i
[0] * src1
->i
[0];
742 dst1
->i
[1] = src0
->i
[1] * src1
->i
[1];
743 dst1
->i
[2] = src0
->i
[2] * src1
->i
[2];
744 dst1
->i
[3] = src0
->i
[3] * src1
->i
[3];
755 union tgsi_exec_channel
*dst0
,
756 union tgsi_exec_channel
*dst1
,
757 const union tgsi_exec_channel
*src0
,
758 const union tgsi_exec_channel
*src1
)
760 dst1
->u
[0] = src0
->u
[0] * src1
->u
[0];
761 dst1
->u
[1] = src0
->u
[1] * src1
->u
[1];
762 dst1
->u
[2] = src0
->u
[2] * src1
->u
[2];
763 dst1
->u
[3] = src0
->u
[3] * src1
->u
[3];
775 union tgsi_exec_channel
*dst
,
776 const union tgsi_exec_channel
*src0
,
777 const union tgsi_exec_channel
*src1
,
778 const union tgsi_exec_channel
*src2
)
780 dst
->u
[0] = src0
->u
[0] ? src1
->u
[0] : src2
->u
[0];
781 dst
->u
[1] = src0
->u
[1] ? src1
->u
[1] : src2
->u
[1];
782 dst
->u
[2] = src0
->u
[2] ? src1
->u
[2] : src2
->u
[2];
783 dst
->u
[3] = src0
->u
[3] ? src1
->u
[3] : src2
->u
[3];
789 union tgsi_exec_channel
*dst
,
790 const union tgsi_exec_channel
*src
)
792 dst
->f
[0] = -src
->f
[0];
793 dst
->f
[1] = -src
->f
[1];
794 dst
->f
[2] = -src
->f
[2];
795 dst
->f
[3] = -src
->f
[3];
801 union tgsi_exec_channel
*dst
,
802 const union tgsi_exec_channel
*src
)
804 dst
->i
[0] = -src
->i
[0];
805 dst
->i
[1] = -src
->i
[1];
806 dst
->i
[2] = -src
->i
[2];
807 dst
->i
[3] = -src
->i
[3];
813 union tgsi_exec_channel
*dst
,
814 const union tgsi_exec_channel
*src
)
816 dst
->u
[0] = ~src
->u
[0];
817 dst
->u
[1] = ~src
->u
[1];
818 dst
->u
[2] = ~src
->u
[2];
819 dst
->u
[3] = ~src
->u
[3];
824 union tgsi_exec_channel
*dst
,
825 const union tgsi_exec_channel
*src0
,
826 const union tgsi_exec_channel
*src1
)
828 dst
->u
[0] = src0
->u
[0] | src1
->u
[0];
829 dst
->u
[1] = src0
->u
[1] | src1
->u
[1];
830 dst
->u
[2] = src0
->u
[2] | src1
->u
[2];
831 dst
->u
[3] = src0
->u
[3] | src1
->u
[3];
836 union tgsi_exec_channel
*dst
,
837 const union tgsi_exec_channel
*src0
,
838 const union tgsi_exec_channel
*src1
)
841 dst
->f
[0] = util_fast_pow( src0
->f
[0], src1
->f
[0] );
842 dst
->f
[1] = util_fast_pow( src0
->f
[1], src1
->f
[1] );
843 dst
->f
[2] = util_fast_pow( src0
->f
[2], src1
->f
[2] );
844 dst
->f
[3] = util_fast_pow( src0
->f
[3], src1
->f
[3] );
846 dst
->f
[0] = powf( src0
->f
[0], src1
->f
[0] );
847 dst
->f
[1] = powf( src0
->f
[1], src1
->f
[1] );
848 dst
->f
[2] = powf( src0
->f
[2], src1
->f
[2] );
849 dst
->f
[3] = powf( src0
->f
[3], src1
->f
[3] );
855 union tgsi_exec_channel
*dst
,
856 const union tgsi_exec_channel
*src
)
858 dst
->f
[0] = floorf( src
->f
[0] + 0.5f
);
859 dst
->f
[1] = floorf( src
->f
[1] + 0.5f
);
860 dst
->f
[2] = floorf( src
->f
[2] + 0.5f
);
861 dst
->f
[3] = floorf( src
->f
[3] + 0.5f
);
866 union tgsi_exec_channel
*dst
,
867 const union tgsi_exec_channel
*src
)
869 dst
->f
[0] = src
->f
[0] < 0.0f
? -1.0f
: src
->f
[0] > 0.0f
? 1.0f
: 0.0f
;
870 dst
->f
[1] = src
->f
[1] < 0.0f
? -1.0f
: src
->f
[1] > 0.0f
? 1.0f
: 0.0f
;
871 dst
->f
[2] = src
->f
[2] < 0.0f
? -1.0f
: src
->f
[2] > 0.0f
? 1.0f
: 0.0f
;
872 dst
->f
[3] = src
->f
[3] < 0.0f
? -1.0f
: src
->f
[3] > 0.0f
? 1.0f
: 0.0f
;
877 union tgsi_exec_channel
*dst
,
878 const union tgsi_exec_channel
*src0
,
879 const union tgsi_exec_channel
*src1
)
881 dst
->i
[0] = src0
->i
[0] << src1
->i
[0];
882 dst
->i
[1] = src0
->i
[1] << src1
->i
[1];
883 dst
->i
[2] = src0
->i
[2] << src1
->i
[2];
884 dst
->i
[3] = src0
->i
[3] << src1
->i
[3];
889 union tgsi_exec_channel
*dst
,
890 const union tgsi_exec_channel
*src0
,
891 const union tgsi_exec_channel
*src1
)
893 dst
->i
[0] = src0
->i
[0] >> src1
->i
[0];
894 dst
->i
[1] = src0
->i
[1] >> src1
->i
[1];
895 dst
->i
[2] = src0
->i
[2] >> src1
->i
[2];
896 dst
->i
[3] = src0
->i
[3] >> src1
->i
[3];
901 union tgsi_exec_channel
*dst
,
902 const union tgsi_exec_channel
*src0
)
904 dst
->f
[0] = (float) (int) src0
->f
[0];
905 dst
->f
[1] = (float) (int) src0
->f
[1];
906 dst
->f
[2] = (float) (int) src0
->f
[2];
907 dst
->f
[3] = (float) (int) src0
->f
[3];
913 union tgsi_exec_channel
*dst
,
914 const union tgsi_exec_channel
*src0
,
915 const union tgsi_exec_channel
*src1
)
917 dst
->u
[0] = src0
->u
[0] >> src1
->u
[0];
918 dst
->u
[1] = src0
->u
[1] >> src1
->u
[1];
919 dst
->u
[2] = src0
->u
[2] >> src1
->u
[2];
920 dst
->u
[3] = src0
->u
[3] >> src1
->u
[3];
926 union tgsi_exec_channel
*dst
,
927 const union tgsi_exec_channel
*src
)
929 dst
->f
[0] = sinf( src
->f
[0] );
930 dst
->f
[1] = sinf( src
->f
[1] );
931 dst
->f
[2] = sinf( src
->f
[2] );
932 dst
->f
[3] = sinf( src
->f
[3] );
936 micro_sqrt( union tgsi_exec_channel
*dst
,
937 const union tgsi_exec_channel
*src
)
939 dst
->f
[0] = sqrtf( src
->f
[0] );
940 dst
->f
[1] = sqrtf( src
->f
[1] );
941 dst
->f
[2] = sqrtf( src
->f
[2] );
942 dst
->f
[3] = sqrtf( src
->f
[3] );
947 union tgsi_exec_channel
*dst
,
948 const union tgsi_exec_channel
*src0
,
949 const union tgsi_exec_channel
*src1
)
951 dst
->f
[0] = src0
->f
[0] - src1
->f
[0];
952 dst
->f
[1] = src0
->f
[1] - src1
->f
[1];
953 dst
->f
[2] = src0
->f
[2] - src1
->f
[2];
954 dst
->f
[3] = src0
->f
[3] - src1
->f
[3];
960 union tgsi_exec_channel
*dst
,
961 const union tgsi_exec_channel
*src
)
963 dst
->f
[0] = (float) src
->u
[0];
964 dst
->f
[1] = (float) src
->u
[1];
965 dst
->f
[2] = (float) src
->u
[2];
966 dst
->f
[3] = (float) src
->u
[3];
972 union tgsi_exec_channel
*dst
,
973 const union tgsi_exec_channel
*src0
,
974 const union tgsi_exec_channel
*src1
)
976 dst
->u
[0] = src0
->u
[0] ^ src1
->u
[0];
977 dst
->u
[1] = src0
->u
[1] ^ src1
->u
[1];
978 dst
->u
[2] = src0
->u
[2] ^ src1
->u
[2];
979 dst
->u
[3] = src0
->u
[3] ^ src1
->u
[3];
983 fetch_src_file_channel(
984 const struct tgsi_exec_machine
*mach
,
987 const union tgsi_exec_channel
*index
,
988 union tgsi_exec_channel
*chan
)
991 case TGSI_EXTSWIZZLE_X
:
992 case TGSI_EXTSWIZZLE_Y
:
993 case TGSI_EXTSWIZZLE_Z
:
994 case TGSI_EXTSWIZZLE_W
:
996 case TGSI_FILE_CONSTANT
:
997 assert(mach
->Consts
);
1001 chan
->f
[0] = mach
->Consts
[index
->i
[0]][swizzle
];
1002 if (index
->i
[1] < 0)
1005 chan
->f
[1] = mach
->Consts
[index
->i
[1]][swizzle
];
1006 if (index
->i
[2] < 0)
1009 chan
->f
[2] = mach
->Consts
[index
->i
[2]][swizzle
];
1010 if (index
->i
[3] < 0)
1013 chan
->f
[3] = mach
->Consts
[index
->i
[3]][swizzle
];
1016 case TGSI_FILE_INPUT
:
1017 chan
->u
[0] = mach
->Inputs
[index
->i
[0]].xyzw
[swizzle
].u
[0];
1018 chan
->u
[1] = mach
->Inputs
[index
->i
[1]].xyzw
[swizzle
].u
[1];
1019 chan
->u
[2] = mach
->Inputs
[index
->i
[2]].xyzw
[swizzle
].u
[2];
1020 chan
->u
[3] = mach
->Inputs
[index
->i
[3]].xyzw
[swizzle
].u
[3];
1023 case TGSI_FILE_TEMPORARY
:
1024 assert(index
->i
[0] < TGSI_EXEC_NUM_TEMPS
);
1025 chan
->u
[0] = mach
->Temps
[index
->i
[0]].xyzw
[swizzle
].u
[0];
1026 chan
->u
[1] = mach
->Temps
[index
->i
[1]].xyzw
[swizzle
].u
[1];
1027 chan
->u
[2] = mach
->Temps
[index
->i
[2]].xyzw
[swizzle
].u
[2];
1028 chan
->u
[3] = mach
->Temps
[index
->i
[3]].xyzw
[swizzle
].u
[3];
1031 case TGSI_FILE_IMMEDIATE
:
1032 assert( index
->i
[0] < (int) mach
->ImmLimit
);
1033 chan
->f
[0] = mach
->Imms
[index
->i
[0]][swizzle
];
1034 assert( index
->i
[1] < (int) mach
->ImmLimit
);
1035 chan
->f
[1] = mach
->Imms
[index
->i
[1]][swizzle
];
1036 assert( index
->i
[2] < (int) mach
->ImmLimit
);
1037 chan
->f
[2] = mach
->Imms
[index
->i
[2]][swizzle
];
1038 assert( index
->i
[3] < (int) mach
->ImmLimit
);
1039 chan
->f
[3] = mach
->Imms
[index
->i
[3]][swizzle
];
1042 case TGSI_FILE_ADDRESS
:
1043 chan
->u
[0] = mach
->Addrs
[index
->i
[0]].xyzw
[swizzle
].u
[0];
1044 chan
->u
[1] = mach
->Addrs
[index
->i
[1]].xyzw
[swizzle
].u
[1];
1045 chan
->u
[2] = mach
->Addrs
[index
->i
[2]].xyzw
[swizzle
].u
[2];
1046 chan
->u
[3] = mach
->Addrs
[index
->i
[3]].xyzw
[swizzle
].u
[3];
1049 case TGSI_FILE_OUTPUT
:
1050 /* vertex/fragment output vars can be read too */
1051 chan
->u
[0] = mach
->Outputs
[index
->i
[0]].xyzw
[swizzle
].u
[0];
1052 chan
->u
[1] = mach
->Outputs
[index
->i
[1]].xyzw
[swizzle
].u
[1];
1053 chan
->u
[2] = mach
->Outputs
[index
->i
[2]].xyzw
[swizzle
].u
[2];
1054 chan
->u
[3] = mach
->Outputs
[index
->i
[3]].xyzw
[swizzle
].u
[3];
1062 case TGSI_EXTSWIZZLE_ZERO
:
1063 *chan
= mach
->Temps
[TEMP_0_I
].xyzw
[TEMP_0_C
];
1066 case TGSI_EXTSWIZZLE_ONE
:
1067 *chan
= mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
];
1077 const struct tgsi_exec_machine
*mach
,
1078 union tgsi_exec_channel
*chan
,
1079 const struct tgsi_full_src_register
*reg
,
1080 const uint chan_index
)
1082 union tgsi_exec_channel index
;
1085 /* We start with a direct index into a register file.
1089 * file = SrcRegister.File
1090 * [1] = SrcRegister.Index
1095 index
.i
[3] = reg
->SrcRegister
.Index
;
1097 /* There is an extra source register that indirectly subscripts
1098 * a register file. The direct index now becomes an offset
1099 * that is being added to the indirect register.
1103 * ind = SrcRegisterInd.File
1104 * [2] = SrcRegisterInd.Index
1105 * .x = SrcRegisterInd.SwizzleX
1107 if (reg
->SrcRegister
.Indirect
) {
1108 union tgsi_exec_channel index2
;
1109 union tgsi_exec_channel indir_index
;
1110 const uint execmask
= mach
->ExecMask
;
1113 /* which address register (always zero now) */
1117 index2
.i
[3] = reg
->SrcRegisterInd
.Index
;
1119 /* get current value of address register[swizzle] */
1120 swizzle
= tgsi_util_get_src_register_swizzle( ®
->SrcRegisterInd
, CHAN_X
);
1121 fetch_src_file_channel(
1123 reg
->SrcRegisterInd
.File
,
1128 /* add value of address register to the offset */
1129 index
.i
[0] += (int) indir_index
.f
[0];
1130 index
.i
[1] += (int) indir_index
.f
[1];
1131 index
.i
[2] += (int) indir_index
.f
[2];
1132 index
.i
[3] += (int) indir_index
.f
[3];
1134 /* for disabled execution channels, zero-out the index to
1135 * avoid using a potential garbage value.
1137 for (i
= 0; i
< QUAD_SIZE
; i
++) {
1138 if ((execmask
& (1 << i
)) == 0)
1143 /* There is an extra source register that is a second
1144 * subscript to a register file. Effectively it means that
1145 * the register file is actually a 2D array of registers.
1147 * file[1][3] == file[1*sizeof(file[1])+3],
1149 * [3] = SrcRegisterDim.Index
1151 if (reg
->SrcRegister
.Dimension
) {
1152 /* The size of the first-order array depends on the register file type.
1153 * We need to multiply the index to the first array to get an effective,
1154 * "flat" index that points to the beginning of the second-order array.
1156 switch (reg
->SrcRegister
.File
) {
1157 case TGSI_FILE_INPUT
:
1158 index
.i
[0] *= TGSI_EXEC_MAX_INPUT_ATTRIBS
;
1159 index
.i
[1] *= TGSI_EXEC_MAX_INPUT_ATTRIBS
;
1160 index
.i
[2] *= TGSI_EXEC_MAX_INPUT_ATTRIBS
;
1161 index
.i
[3] *= TGSI_EXEC_MAX_INPUT_ATTRIBS
;
1163 case TGSI_FILE_CONSTANT
:
1164 index
.i
[0] *= TGSI_EXEC_MAX_CONST_BUFFER
;
1165 index
.i
[1] *= TGSI_EXEC_MAX_CONST_BUFFER
;
1166 index
.i
[2] *= TGSI_EXEC_MAX_CONST_BUFFER
;
1167 index
.i
[3] *= TGSI_EXEC_MAX_CONST_BUFFER
;
1173 index
.i
[0] += reg
->SrcRegisterDim
.Index
;
1174 index
.i
[1] += reg
->SrcRegisterDim
.Index
;
1175 index
.i
[2] += reg
->SrcRegisterDim
.Index
;
1176 index
.i
[3] += reg
->SrcRegisterDim
.Index
;
1178 /* Again, the second subscript index can be addressed indirectly
1179 * identically to the first one.
1180 * Nothing stops us from indirectly addressing the indirect register,
1181 * but there is no need for that, so we won't exercise it.
1183 * file[1][ind[4].y+3],
1185 * ind = SrcRegisterDimInd.File
1186 * [4] = SrcRegisterDimInd.Index
1187 * .y = SrcRegisterDimInd.SwizzleX
1189 if (reg
->SrcRegisterDim
.Indirect
) {
1190 union tgsi_exec_channel index2
;
1191 union tgsi_exec_channel indir_index
;
1192 const uint execmask
= mach
->ExecMask
;
1198 index2
.i
[3] = reg
->SrcRegisterDimInd
.Index
;
1200 swizzle
= tgsi_util_get_src_register_swizzle( ®
->SrcRegisterDimInd
, CHAN_X
);
1201 fetch_src_file_channel(
1203 reg
->SrcRegisterDimInd
.File
,
1208 index
.i
[0] += (int) indir_index
.f
[0];
1209 index
.i
[1] += (int) indir_index
.f
[1];
1210 index
.i
[2] += (int) indir_index
.f
[2];
1211 index
.i
[3] += (int) indir_index
.f
[3];
1213 /* for disabled execution channels, zero-out the index to
1214 * avoid using a potential garbage value.
1216 for (i
= 0; i
< QUAD_SIZE
; i
++) {
1217 if ((execmask
& (1 << i
)) == 0)
1222 /* If by any chance there was a need for a 3D array of register
1223 * files, we would have to check whether SrcRegisterDim is followed
1224 * by a dimension register and continue the saga.
1228 swizzle
= tgsi_util_get_full_src_register_extswizzle( reg
, chan_index
);
1229 fetch_src_file_channel(
1231 reg
->SrcRegister
.File
,
1236 switch (tgsi_util_get_full_src_register_sign_mode( reg
, chan_index
)) {
1237 case TGSI_UTIL_SIGN_CLEAR
:
1238 micro_abs( chan
, chan
);
1241 case TGSI_UTIL_SIGN_SET
:
1242 micro_abs( chan
, chan
);
1243 micro_neg( chan
, chan
);
1246 case TGSI_UTIL_SIGN_TOGGLE
:
1247 micro_neg( chan
, chan
);
1250 case TGSI_UTIL_SIGN_KEEP
:
1254 if (reg
->SrcRegisterExtMod
.Complement
) {
1255 micro_sub( chan
, &mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], chan
);
1261 struct tgsi_exec_machine
*mach
,
1262 const union tgsi_exec_channel
*chan
,
1263 const struct tgsi_full_dst_register
*reg
,
1264 const struct tgsi_full_instruction
*inst
,
1268 union tgsi_exec_channel null
;
1269 union tgsi_exec_channel
*dst
;
1270 uint execmask
= mach
->ExecMask
;
1272 switch (reg
->DstRegister
.File
) {
1273 case TGSI_FILE_NULL
:
1277 case TGSI_FILE_OUTPUT
:
1278 dst
= &mach
->Outputs
[mach
->Temps
[TEMP_OUTPUT_I
].xyzw
[TEMP_OUTPUT_C
].u
[0]
1279 + reg
->DstRegister
.Index
].xyzw
[chan_index
];
1282 case TGSI_FILE_TEMPORARY
:
1283 assert( reg
->DstRegister
.Index
< TGSI_EXEC_NUM_TEMPS
);
1284 dst
= &mach
->Temps
[reg
->DstRegister
.Index
].xyzw
[chan_index
];
1287 case TGSI_FILE_ADDRESS
:
1288 dst
= &mach
->Addrs
[reg
->DstRegister
.Index
].xyzw
[chan_index
];
1296 if (inst
->InstructionExtNv
.CondFlowEnable
) {
1297 union tgsi_exec_channel
*cc
= &mach
->Temps
[TEMP_CC_I
].xyzw
[TEMP_CC_C
];
1303 /* Only CC0 supported.
1305 assert( inst
->InstructionExtNv
.CondFlowIndex
< 1 );
1307 switch (chan_index
) {
1309 swizzle
= inst
->InstructionExtNv
.CondSwizzleX
;
1312 swizzle
= inst
->InstructionExtNv
.CondSwizzleY
;
1315 swizzle
= inst
->InstructionExtNv
.CondSwizzleZ
;
1318 swizzle
= inst
->InstructionExtNv
.CondSwizzleW
;
1326 case TGSI_SWIZZLE_X
:
1327 shift
= TGSI_EXEC_CC_X_SHIFT
;
1328 mask
= TGSI_EXEC_CC_X_MASK
;
1330 case TGSI_SWIZZLE_Y
:
1331 shift
= TGSI_EXEC_CC_Y_SHIFT
;
1332 mask
= TGSI_EXEC_CC_Y_MASK
;
1334 case TGSI_SWIZZLE_Z
:
1335 shift
= TGSI_EXEC_CC_Z_SHIFT
;
1336 mask
= TGSI_EXEC_CC_Z_MASK
;
1338 case TGSI_SWIZZLE_W
:
1339 shift
= TGSI_EXEC_CC_W_SHIFT
;
1340 mask
= TGSI_EXEC_CC_W_MASK
;
1347 switch (inst
->InstructionExtNv
.CondMask
) {
1349 test
= ~(TGSI_EXEC_CC_GT
<< shift
) & mask
;
1350 for (i
= 0; i
< QUAD_SIZE
; i
++)
1351 if (cc
->u
[i
] & test
)
1352 execmask
&= ~(1 << i
);
1356 test
= ~(TGSI_EXEC_CC_EQ
<< shift
) & mask
;
1357 for (i
= 0; i
< QUAD_SIZE
; i
++)
1358 if (cc
->u
[i
] & test
)
1359 execmask
&= ~(1 << i
);
1363 test
= ~(TGSI_EXEC_CC_LT
<< shift
) & mask
;
1364 for (i
= 0; i
< QUAD_SIZE
; i
++)
1365 if (cc
->u
[i
] & test
)
1366 execmask
&= ~(1 << i
);
1370 test
= ~((TGSI_EXEC_CC_GT
| TGSI_EXEC_CC_EQ
) << shift
) & mask
;
1371 for (i
= 0; i
< QUAD_SIZE
; i
++)
1372 if (cc
->u
[i
] & test
)
1373 execmask
&= ~(1 << i
);
1377 test
= ~((TGSI_EXEC_CC_LT
| TGSI_EXEC_CC_EQ
) << shift
) & mask
;
1378 for (i
= 0; i
< QUAD_SIZE
; i
++)
1379 if (cc
->u
[i
] & test
)
1380 execmask
&= ~(1 << i
);
1384 test
= ~((TGSI_EXEC_CC_GT
| TGSI_EXEC_CC_LT
| TGSI_EXEC_CC_UN
) << shift
) & mask
;
1385 for (i
= 0; i
< QUAD_SIZE
; i
++)
1386 if (cc
->u
[i
] & test
)
1387 execmask
&= ~(1 << i
);
1394 for (i
= 0; i
< QUAD_SIZE
; i
++)
1395 execmask
&= ~(1 << i
);
1404 switch (inst
->Instruction
.Saturate
) {
1406 for (i
= 0; i
< QUAD_SIZE
; i
++)
1407 if (execmask
& (1 << i
))
1408 dst
->i
[i
] = chan
->i
[i
];
1411 case TGSI_SAT_ZERO_ONE
:
1412 for (i
= 0; i
< QUAD_SIZE
; i
++)
1413 if (execmask
& (1 << i
)) {
1414 if (chan
->f
[i
] < 0.0f
)
1416 else if (chan
->f
[i
] > 1.0f
)
1419 dst
->i
[i
] = chan
->i
[i
];
1423 case TGSI_SAT_MINUS_PLUS_ONE
:
1424 for (i
= 0; i
< QUAD_SIZE
; i
++)
1425 if (execmask
& (1 << i
)) {
1426 if (chan
->f
[i
] < -1.0f
)
1428 else if (chan
->f
[i
] > 1.0f
)
1431 dst
->i
[i
] = chan
->i
[i
];
1439 if (inst
->InstructionExtNv
.CondDstUpdate
) {
1440 union tgsi_exec_channel
*cc
= &mach
->Temps
[TEMP_CC_I
].xyzw
[TEMP_CC_C
];
1444 /* Only CC0 supported.
1446 assert( inst
->InstructionExtNv
.CondDstIndex
< 1 );
1448 switch (chan_index
) {
1450 shift
= TGSI_EXEC_CC_X_SHIFT
;
1451 mask
= ~TGSI_EXEC_CC_X_MASK
;
1454 shift
= TGSI_EXEC_CC_Y_SHIFT
;
1455 mask
= ~TGSI_EXEC_CC_Y_MASK
;
1458 shift
= TGSI_EXEC_CC_Z_SHIFT
;
1459 mask
= ~TGSI_EXEC_CC_Z_MASK
;
1462 shift
= TGSI_EXEC_CC_W_SHIFT
;
1463 mask
= ~TGSI_EXEC_CC_W_MASK
;
1470 for (i
= 0; i
< QUAD_SIZE
; i
++)
1471 if (execmask
& (1 << i
)) {
1473 if (dst
->f
[i
] < 0.0f
)
1474 cc
->u
[i
] |= TGSI_EXEC_CC_LT
<< shift
;
1475 else if (dst
->f
[i
] > 0.0f
)
1476 cc
->u
[i
] |= TGSI_EXEC_CC_GT
<< shift
;
1477 else if (dst
->f
[i
] == 0.0f
)
1478 cc
->u
[i
] |= TGSI_EXEC_CC_EQ
<< shift
;
1480 cc
->u
[i
] |= TGSI_EXEC_CC_UN
<< shift
;
1485 #define FETCH(VAL,INDEX,CHAN)\
1486 fetch_source (mach, VAL, &inst->FullSrcRegisters[INDEX], CHAN)
1488 #define STORE(VAL,INDEX,CHAN)\
1489 store_dest (mach, VAL, &inst->FullDstRegisters[INDEX], inst, CHAN )
1493 * Execute ARB-style KIL which is predicated by a src register.
1494 * Kill fragment if any of the four values is less than zero.
1497 exec_kil(struct tgsi_exec_machine
*mach
,
1498 const struct tgsi_full_instruction
*inst
)
1502 uint kilmask
= 0; /* bit 0 = pixel 0, bit 1 = pixel 1, etc */
1503 union tgsi_exec_channel r
[1];
1505 /* This mask stores component bits that were already tested. Note that
1506 * we test if the value is less than zero, so 1.0 and 0.0 need not to be
1508 uniquemask
= (1 << TGSI_EXTSWIZZLE_ZERO
) | (1 << TGSI_EXTSWIZZLE_ONE
);
1510 for (chan_index
= 0; chan_index
< 4; chan_index
++)
1515 /* unswizzle channel */
1516 swizzle
= tgsi_util_get_full_src_register_extswizzle (
1517 &inst
->FullSrcRegisters
[0],
1520 /* check if the component has not been already tested */
1521 if (uniquemask
& (1 << swizzle
))
1523 uniquemask
|= 1 << swizzle
;
1525 FETCH(&r
[0], 0, chan_index
);
1526 for (i
= 0; i
< 4; i
++)
1527 if (r
[0].f
[i
] < 0.0f
)
1531 mach
->Temps
[TEMP_KILMASK_I
].xyzw
[TEMP_KILMASK_C
].u
[0] |= kilmask
;
1535 * Execute NVIDIA-style KIL which is predicated by a condition code.
1536 * Kill fragment if the condition code is TRUE.
1539 exec_kilp(struct tgsi_exec_machine
*mach
,
1540 const struct tgsi_full_instruction
*inst
)
1542 uint kilmask
; /* bit 0 = pixel 0, bit 1 = pixel 1, etc */
1544 if (inst
->InstructionExtNv
.CondFlowEnable
) {
1550 swizzle
[0] = inst
->InstructionExtNv
.CondSwizzleX
;
1551 swizzle
[1] = inst
->InstructionExtNv
.CondSwizzleY
;
1552 swizzle
[2] = inst
->InstructionExtNv
.CondSwizzleZ
;
1553 swizzle
[3] = inst
->InstructionExtNv
.CondSwizzleW
;
1555 for (chan_index
= 0; chan_index
< 4; chan_index
++)
1559 for (i
= 0; i
< 4; i
++) {
1560 /* TODO: evaluate the condition code */
1567 /* "unconditional" kil */
1568 kilmask
= mach
->ExecMask
;
1570 mach
->Temps
[TEMP_KILMASK_I
].xyzw
[TEMP_KILMASK_C
].u
[0] |= kilmask
;
1575 * Fetch a four texture samples using STR texture coordinates.
1578 fetch_texel( struct tgsi_sampler
*sampler
,
1579 const union tgsi_exec_channel
*s
,
1580 const union tgsi_exec_channel
*t
,
1581 const union tgsi_exec_channel
*p
,
1582 float lodbias
, /* XXX should be float[4] */
1583 union tgsi_exec_channel
*r
,
1584 union tgsi_exec_channel
*g
,
1585 union tgsi_exec_channel
*b
,
1586 union tgsi_exec_channel
*a
)
1589 float rgba
[NUM_CHANNELS
][QUAD_SIZE
];
1591 sampler
->get_samples(sampler
, s
->f
, t
->f
, p
->f
, lodbias
, rgba
);
1593 for (j
= 0; j
< 4; j
++) {
1594 r
->f
[j
] = rgba
[0][j
];
1595 g
->f
[j
] = rgba
[1][j
];
1596 b
->f
[j
] = rgba
[2][j
];
1597 a
->f
[j
] = rgba
[3][j
];
1603 exec_tex(struct tgsi_exec_machine
*mach
,
1604 const struct tgsi_full_instruction
*inst
,
1608 const uint unit
= inst
->FullSrcRegisters
[1].SrcRegister
.Index
;
1609 union tgsi_exec_channel r
[4];
1613 /* debug_printf("Sampler %u unit %u\n", sampler, unit); */
1615 switch (inst
->InstructionExtTexture
.Texture
) {
1616 case TGSI_TEXTURE_1D
:
1617 case TGSI_TEXTURE_SHADOW1D
:
1619 FETCH(&r
[0], 0, CHAN_X
);
1622 FETCH(&r
[1], 0, CHAN_W
);
1623 micro_div( &r
[0], &r
[0], &r
[1] );
1627 FETCH(&r
[1], 0, CHAN_W
);
1628 lodBias
= r
[2].f
[0];
1633 fetch_texel(mach
->Samplers
[unit
],
1634 &r
[0], NULL
, NULL
, lodBias
, /* S, T, P, BIAS */
1635 &r
[0], &r
[1], &r
[2], &r
[3]); /* R, G, B, A */
1638 case TGSI_TEXTURE_2D
:
1639 case TGSI_TEXTURE_RECT
:
1640 case TGSI_TEXTURE_SHADOW2D
:
1641 case TGSI_TEXTURE_SHADOWRECT
:
1643 FETCH(&r
[0], 0, CHAN_X
);
1644 FETCH(&r
[1], 0, CHAN_Y
);
1645 FETCH(&r
[2], 0, CHAN_Z
);
1648 FETCH(&r
[3], 0, CHAN_W
);
1649 micro_div( &r
[0], &r
[0], &r
[3] );
1650 micro_div( &r
[1], &r
[1], &r
[3] );
1651 micro_div( &r
[2], &r
[2], &r
[3] );
1655 FETCH(&r
[3], 0, CHAN_W
);
1656 lodBias
= r
[3].f
[0];
1661 fetch_texel(mach
->Samplers
[unit
],
1662 &r
[0], &r
[1], &r
[2], lodBias
, /* inputs */
1663 &r
[0], &r
[1], &r
[2], &r
[3]); /* outputs */
1666 case TGSI_TEXTURE_3D
:
1667 case TGSI_TEXTURE_CUBE
:
1669 FETCH(&r
[0], 0, CHAN_X
);
1670 FETCH(&r
[1], 0, CHAN_Y
);
1671 FETCH(&r
[2], 0, CHAN_Z
);
1674 FETCH(&r
[3], 0, CHAN_W
);
1675 micro_div( &r
[0], &r
[0], &r
[3] );
1676 micro_div( &r
[1], &r
[1], &r
[3] );
1677 micro_div( &r
[2], &r
[2], &r
[3] );
1681 FETCH(&r
[3], 0, CHAN_W
);
1682 lodBias
= r
[3].f
[0];
1687 fetch_texel(mach
->Samplers
[unit
],
1688 &r
[0], &r
[1], &r
[2], lodBias
,
1689 &r
[0], &r
[1], &r
[2], &r
[3]);
1696 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
1697 STORE( &r
[chan_index
], 0, chan_index
);
1703 * Evaluate a constant-valued coefficient at the position of the
1708 struct tgsi_exec_machine
*mach
,
1714 for( i
= 0; i
< QUAD_SIZE
; i
++ ) {
1715 mach
->Inputs
[attrib
].xyzw
[chan
].f
[i
] = mach
->InterpCoefs
[attrib
].a0
[chan
];
1720 * Evaluate a linear-valued coefficient at the position of the
1725 struct tgsi_exec_machine
*mach
,
1729 const float x
= mach
->QuadPos
.xyzw
[0].f
[0];
1730 const float y
= mach
->QuadPos
.xyzw
[1].f
[0];
1731 const float dadx
= mach
->InterpCoefs
[attrib
].dadx
[chan
];
1732 const float dady
= mach
->InterpCoefs
[attrib
].dady
[chan
];
1733 const float a0
= mach
->InterpCoefs
[attrib
].a0
[chan
] + dadx
* x
+ dady
* y
;
1734 mach
->Inputs
[attrib
].xyzw
[chan
].f
[0] = a0
;
1735 mach
->Inputs
[attrib
].xyzw
[chan
].f
[1] = a0
+ dadx
;
1736 mach
->Inputs
[attrib
].xyzw
[chan
].f
[2] = a0
+ dady
;
1737 mach
->Inputs
[attrib
].xyzw
[chan
].f
[3] = a0
+ dadx
+ dady
;
1741 * Evaluate a perspective-valued coefficient at the position of the
1745 eval_perspective_coef(
1746 struct tgsi_exec_machine
*mach
,
1750 const float x
= mach
->QuadPos
.xyzw
[0].f
[0];
1751 const float y
= mach
->QuadPos
.xyzw
[1].f
[0];
1752 const float dadx
= mach
->InterpCoefs
[attrib
].dadx
[chan
];
1753 const float dady
= mach
->InterpCoefs
[attrib
].dady
[chan
];
1754 const float a0
= mach
->InterpCoefs
[attrib
].a0
[chan
] + dadx
* x
+ dady
* y
;
1755 const float *w
= mach
->QuadPos
.xyzw
[3].f
;
1756 /* divide by W here */
1757 mach
->Inputs
[attrib
].xyzw
[chan
].f
[0] = a0
/ w
[0];
1758 mach
->Inputs
[attrib
].xyzw
[chan
].f
[1] = (a0
+ dadx
) / w
[1];
1759 mach
->Inputs
[attrib
].xyzw
[chan
].f
[2] = (a0
+ dady
) / w
[2];
1760 mach
->Inputs
[attrib
].xyzw
[chan
].f
[3] = (a0
+ dadx
+ dady
) / w
[3];
1764 typedef void (* eval_coef_func
)(
1765 struct tgsi_exec_machine
*mach
,
1771 struct tgsi_exec_machine
*mach
,
1772 const struct tgsi_full_declaration
*decl
)
1774 if( mach
->Processor
== TGSI_PROCESSOR_FRAGMENT
) {
1775 if( decl
->Declaration
.File
== TGSI_FILE_INPUT
) {
1776 unsigned first
, last
, mask
;
1777 eval_coef_func eval
;
1779 first
= decl
->DeclarationRange
.First
;
1780 last
= decl
->DeclarationRange
.Last
;
1781 mask
= decl
->Declaration
.UsageMask
;
1783 switch( decl
->Declaration
.Interpolate
) {
1784 case TGSI_INTERPOLATE_CONSTANT
:
1785 eval
= eval_constant_coef
;
1788 case TGSI_INTERPOLATE_LINEAR
:
1789 eval
= eval_linear_coef
;
1792 case TGSI_INTERPOLATE_PERSPECTIVE
:
1793 eval
= eval_perspective_coef
;
1801 if( mask
== TGSI_WRITEMASK_XYZW
) {
1804 for( i
= first
; i
<= last
; i
++ ) {
1805 for( j
= 0; j
< NUM_CHANNELS
; j
++ ) {
1813 for( j
= 0; j
< NUM_CHANNELS
; j
++ ) {
1814 if( mask
& (1 << j
) ) {
1815 for( i
= first
; i
<= last
; i
++ ) {
1827 struct tgsi_exec_machine
*mach
,
1828 const struct tgsi_full_instruction
*inst
,
1832 union tgsi_exec_channel r
[8];
1836 switch (inst
->Instruction
.Opcode
) {
1837 case TGSI_OPCODE_ARL
:
1838 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
1839 FETCH( &r
[0], 0, chan_index
);
1840 micro_flr( &r
[0], &r
[0] );
1841 STORE( &r
[0], 0, chan_index
);
1845 case TGSI_OPCODE_MOV
:
1846 case TGSI_OPCODE_SWZ
:
1847 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
1848 FETCH( &r
[0], 0, chan_index
);
1849 STORE( &r
[0], 0, chan_index
);
1853 case TGSI_OPCODE_LIT
:
1854 if (IS_CHANNEL_ENABLED( *inst
, CHAN_X
)) {
1855 STORE( &mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], 0, CHAN_X
);
1858 if (IS_CHANNEL_ENABLED( *inst
, CHAN_Y
) || IS_CHANNEL_ENABLED( *inst
, CHAN_Z
)) {
1859 FETCH( &r
[0], 0, CHAN_X
);
1860 if (IS_CHANNEL_ENABLED( *inst
, CHAN_Y
)) {
1861 micro_max( &r
[0], &r
[0], &mach
->Temps
[TEMP_0_I
].xyzw
[TEMP_0_C
] );
1862 STORE( &r
[0], 0, CHAN_Y
);
1865 if (IS_CHANNEL_ENABLED( *inst
, CHAN_Z
)) {
1866 FETCH( &r
[1], 0, CHAN_Y
);
1867 micro_max( &r
[1], &r
[1], &mach
->Temps
[TEMP_0_I
].xyzw
[TEMP_0_C
] );
1869 FETCH( &r
[2], 0, CHAN_W
);
1870 micro_min( &r
[2], &r
[2], &mach
->Temps
[TEMP_128_I
].xyzw
[TEMP_128_C
] );
1871 micro_max( &r
[2], &r
[2], &mach
->Temps
[TEMP_M128_I
].xyzw
[TEMP_M128_C
] );
1872 micro_pow( &r
[1], &r
[1], &r
[2] );
1873 micro_lt( &r
[0], &mach
->Temps
[TEMP_0_I
].xyzw
[TEMP_0_C
], &r
[0], &r
[1], &mach
->Temps
[TEMP_0_I
].xyzw
[TEMP_0_C
] );
1874 STORE( &r
[0], 0, CHAN_Z
);
1878 if (IS_CHANNEL_ENABLED( *inst
, CHAN_W
)) {
1879 STORE( &mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], 0, CHAN_W
);
1883 case TGSI_OPCODE_RCP
:
1884 /* TGSI_OPCODE_RECIP */
1885 FETCH( &r
[0], 0, CHAN_X
);
1886 micro_div( &r
[0], &mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], &r
[0] );
1887 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
1888 STORE( &r
[0], 0, chan_index
);
1892 case TGSI_OPCODE_RSQ
:
1893 /* TGSI_OPCODE_RECIPSQRT */
1894 FETCH( &r
[0], 0, CHAN_X
);
1895 micro_abs( &r
[0], &r
[0] );
1896 micro_sqrt( &r
[0], &r
[0] );
1897 micro_div( &r
[0], &mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], &r
[0] );
1898 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
1899 STORE( &r
[0], 0, chan_index
);
1903 case TGSI_OPCODE_EXP
:
1904 FETCH( &r
[0], 0, CHAN_X
);
1905 micro_flr( &r
[1], &r
[0] ); /* r1 = floor(r0) */
1906 if (IS_CHANNEL_ENABLED( *inst
, CHAN_X
)) {
1907 micro_exp2( &r
[2], &r
[1] ); /* r2 = 2 ^ r1 */
1908 STORE( &r
[2], 0, CHAN_X
); /* store r2 */
1910 if (IS_CHANNEL_ENABLED( *inst
, CHAN_Y
)) {
1911 micro_sub( &r
[2], &r
[0], &r
[1] ); /* r2 = r0 - r1 */
1912 STORE( &r
[2], 0, CHAN_Y
); /* store r2 */
1914 if (IS_CHANNEL_ENABLED( *inst
, CHAN_Z
)) {
1915 micro_exp2( &r
[2], &r
[0] ); /* r2 = 2 ^ r0 */
1916 STORE( &r
[2], 0, CHAN_Z
); /* store r2 */
1918 if (IS_CHANNEL_ENABLED( *inst
, CHAN_W
)) {
1919 STORE( &mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], 0, CHAN_W
);
1923 case TGSI_OPCODE_LOG
:
1924 FETCH( &r
[0], 0, CHAN_X
);
1925 micro_abs( &r
[2], &r
[0] ); /* r2 = abs(r0) */
1926 micro_lg2( &r
[1], &r
[2] ); /* r1 = lg2(r2) */
1927 micro_flr( &r
[0], &r
[1] ); /* r0 = floor(r1) */
1928 if (IS_CHANNEL_ENABLED( *inst
, CHAN_X
)) {
1929 STORE( &r
[0], 0, CHAN_X
);
1931 if (IS_CHANNEL_ENABLED( *inst
, CHAN_Y
)) {
1932 micro_exp2( &r
[0], &r
[0] ); /* r0 = 2 ^ r0 */
1933 micro_div( &r
[0], &r
[2], &r
[0] ); /* r0 = r2 / r0 */
1934 STORE( &r
[0], 0, CHAN_Y
);
1936 if (IS_CHANNEL_ENABLED( *inst
, CHAN_Z
)) {
1937 STORE( &r
[1], 0, CHAN_Z
);
1939 if (IS_CHANNEL_ENABLED( *inst
, CHAN_W
)) {
1940 STORE( &mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], 0, CHAN_W
);
1944 case TGSI_OPCODE_MUL
:
1945 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
)
1947 FETCH(&r
[0], 0, chan_index
);
1948 FETCH(&r
[1], 1, chan_index
);
1950 micro_mul( &r
[0], &r
[0], &r
[1] );
1952 STORE(&r
[0], 0, chan_index
);
1956 case TGSI_OPCODE_ADD
:
1957 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
1958 FETCH( &r
[0], 0, chan_index
);
1959 FETCH( &r
[1], 1, chan_index
);
1960 micro_add( &r
[0], &r
[0], &r
[1] );
1961 STORE( &r
[0], 0, chan_index
);
1965 case TGSI_OPCODE_DP3
:
1966 /* TGSI_OPCODE_DOT3 */
1967 FETCH( &r
[0], 0, CHAN_X
);
1968 FETCH( &r
[1], 1, CHAN_X
);
1969 micro_mul( &r
[0], &r
[0], &r
[1] );
1971 FETCH( &r
[1], 0, CHAN_Y
);
1972 FETCH( &r
[2], 1, CHAN_Y
);
1973 micro_mul( &r
[1], &r
[1], &r
[2] );
1974 micro_add( &r
[0], &r
[0], &r
[1] );
1976 FETCH( &r
[1], 0, CHAN_Z
);
1977 FETCH( &r
[2], 1, CHAN_Z
);
1978 micro_mul( &r
[1], &r
[1], &r
[2] );
1979 micro_add( &r
[0], &r
[0], &r
[1] );
1981 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
1982 STORE( &r
[0], 0, chan_index
);
1986 case TGSI_OPCODE_DP4
:
1987 /* TGSI_OPCODE_DOT4 */
1988 FETCH(&r
[0], 0, CHAN_X
);
1989 FETCH(&r
[1], 1, CHAN_X
);
1991 micro_mul( &r
[0], &r
[0], &r
[1] );
1993 FETCH(&r
[1], 0, CHAN_Y
);
1994 FETCH(&r
[2], 1, CHAN_Y
);
1996 micro_mul( &r
[1], &r
[1], &r
[2] );
1997 micro_add( &r
[0], &r
[0], &r
[1] );
1999 FETCH(&r
[1], 0, CHAN_Z
);
2000 FETCH(&r
[2], 1, CHAN_Z
);
2002 micro_mul( &r
[1], &r
[1], &r
[2] );
2003 micro_add( &r
[0], &r
[0], &r
[1] );
2005 FETCH(&r
[1], 0, CHAN_W
);
2006 FETCH(&r
[2], 1, CHAN_W
);
2008 micro_mul( &r
[1], &r
[1], &r
[2] );
2009 micro_add( &r
[0], &r
[0], &r
[1] );
2011 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2012 STORE( &r
[0], 0, chan_index
);
2016 case TGSI_OPCODE_DST
:
2017 if (IS_CHANNEL_ENABLED( *inst
, CHAN_X
)) {
2018 STORE( &mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], 0, CHAN_X
);
2021 if (IS_CHANNEL_ENABLED( *inst
, CHAN_Y
)) {
2022 FETCH( &r
[0], 0, CHAN_Y
);
2023 FETCH( &r
[1], 1, CHAN_Y
);
2024 micro_mul( &r
[0], &r
[0], &r
[1] );
2025 STORE( &r
[0], 0, CHAN_Y
);
2028 if (IS_CHANNEL_ENABLED( *inst
, CHAN_Z
)) {
2029 FETCH( &r
[0], 0, CHAN_Z
);
2030 STORE( &r
[0], 0, CHAN_Z
);
2033 if (IS_CHANNEL_ENABLED( *inst
, CHAN_W
)) {
2034 FETCH( &r
[0], 1, CHAN_W
);
2035 STORE( &r
[0], 0, CHAN_W
);
2039 case TGSI_OPCODE_MIN
:
2040 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2041 FETCH(&r
[0], 0, chan_index
);
2042 FETCH(&r
[1], 1, chan_index
);
2044 /* XXX use micro_min()?? */
2045 micro_lt( &r
[0], &r
[0], &r
[1], &r
[0], &r
[1] );
2047 STORE(&r
[0], 0, chan_index
);
2051 case TGSI_OPCODE_MAX
:
2052 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2053 FETCH(&r
[0], 0, chan_index
);
2054 FETCH(&r
[1], 1, chan_index
);
2056 /* XXX use micro_max()?? */
2057 micro_lt( &r
[0], &r
[0], &r
[1], &r
[1], &r
[0] );
2059 STORE(&r
[0], 0, chan_index
);
2063 case TGSI_OPCODE_SLT
:
2064 /* TGSI_OPCODE_SETLT */
2065 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2066 FETCH( &r
[0], 0, chan_index
);
2067 FETCH( &r
[1], 1, chan_index
);
2068 micro_lt( &r
[0], &r
[0], &r
[1], &mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], &mach
->Temps
[TEMP_0_I
].xyzw
[TEMP_0_C
] );
2069 STORE( &r
[0], 0, chan_index
);
2073 case TGSI_OPCODE_SGE
:
2074 /* TGSI_OPCODE_SETGE */
2075 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2076 FETCH( &r
[0], 0, chan_index
);
2077 FETCH( &r
[1], 1, chan_index
);
2078 micro_ge( &r
[0], &r
[0], &r
[1], &mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], &mach
->Temps
[TEMP_0_I
].xyzw
[TEMP_0_C
] );
2079 STORE( &r
[0], 0, chan_index
);
2083 case TGSI_OPCODE_MAD
:
2084 /* TGSI_OPCODE_MADD */
2085 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2086 FETCH( &r
[0], 0, chan_index
);
2087 FETCH( &r
[1], 1, chan_index
);
2088 micro_mul( &r
[0], &r
[0], &r
[1] );
2089 FETCH( &r
[1], 2, chan_index
);
2090 micro_add( &r
[0], &r
[0], &r
[1] );
2091 STORE( &r
[0], 0, chan_index
);
2095 case TGSI_OPCODE_SUB
:
2096 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2097 FETCH(&r
[0], 0, chan_index
);
2098 FETCH(&r
[1], 1, chan_index
);
2100 micro_sub( &r
[0], &r
[0], &r
[1] );
2102 STORE(&r
[0], 0, chan_index
);
2106 case TGSI_OPCODE_LERP
:
2107 /* TGSI_OPCODE_LRP */
2108 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2109 FETCH(&r
[0], 0, chan_index
);
2110 FETCH(&r
[1], 1, chan_index
);
2111 FETCH(&r
[2], 2, chan_index
);
2113 micro_sub( &r
[1], &r
[1], &r
[2] );
2114 micro_mul( &r
[0], &r
[0], &r
[1] );
2115 micro_add( &r
[0], &r
[0], &r
[2] );
2117 STORE(&r
[0], 0, chan_index
);
2121 case TGSI_OPCODE_CND
:
2125 case TGSI_OPCODE_CND0
:
2129 case TGSI_OPCODE_DOT2ADD
:
2130 /* TGSI_OPCODE_DP2A */
2131 FETCH( &r
[0], 0, CHAN_X
);
2132 FETCH( &r
[1], 1, CHAN_X
);
2133 micro_mul( &r
[0], &r
[0], &r
[1] );
2135 FETCH( &r
[1], 0, CHAN_Y
);
2136 FETCH( &r
[2], 1, CHAN_Y
);
2137 micro_mul( &r
[1], &r
[1], &r
[2] );
2138 micro_add( &r
[0], &r
[0], &r
[1] );
2140 FETCH( &r
[2], 2, CHAN_X
);
2141 micro_add( &r
[0], &r
[0], &r
[2] );
2143 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2144 STORE( &r
[0], 0, chan_index
);
2148 case TGSI_OPCODE_INDEX
:
2152 case TGSI_OPCODE_NEGATE
:
2156 case TGSI_OPCODE_FRAC
:
2157 /* TGSI_OPCODE_FRC */
2158 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2159 FETCH( &r
[0], 0, chan_index
);
2160 micro_frc( &r
[0], &r
[0] );
2161 STORE( &r
[0], 0, chan_index
);
2165 case TGSI_OPCODE_CLAMP
:
2169 case TGSI_OPCODE_FLOOR
:
2170 /* TGSI_OPCODE_FLR */
2171 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2172 FETCH( &r
[0], 0, chan_index
);
2173 micro_flr( &r
[0], &r
[0] );
2174 STORE( &r
[0], 0, chan_index
);
2178 case TGSI_OPCODE_ROUND
:
2179 case TGSI_OPCODE_ARR
:
2180 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2181 FETCH( &r
[0], 0, chan_index
);
2182 micro_rnd( &r
[0], &r
[0] );
2183 STORE( &r
[0], 0, chan_index
);
2187 case TGSI_OPCODE_EXPBASE2
:
2188 /* TGSI_OPCODE_EX2 */
2189 FETCH(&r
[0], 0, CHAN_X
);
2192 micro_exp2( &r
[0], &r
[0] );
2194 micro_pow( &r
[0], &mach
->Temps
[TEMP_2_I
].xyzw
[TEMP_2_C
], &r
[0] );
2197 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2198 STORE( &r
[0], 0, chan_index
);
2202 case TGSI_OPCODE_LOGBASE2
:
2203 /* TGSI_OPCODE_LG2 */
2204 FETCH( &r
[0], 0, CHAN_X
);
2205 micro_lg2( &r
[0], &r
[0] );
2206 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2207 STORE( &r
[0], 0, chan_index
);
2211 case TGSI_OPCODE_POWER
:
2212 /* TGSI_OPCODE_POW */
2213 FETCH(&r
[0], 0, CHAN_X
);
2214 FETCH(&r
[1], 1, CHAN_X
);
2216 micro_pow( &r
[0], &r
[0], &r
[1] );
2218 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2219 STORE( &r
[0], 0, chan_index
);
2223 case TGSI_OPCODE_CROSSPRODUCT
:
2224 /* TGSI_OPCODE_XPD */
2225 FETCH(&r
[0], 0, CHAN_Y
);
2226 FETCH(&r
[1], 1, CHAN_Z
);
2228 micro_mul( &r
[2], &r
[0], &r
[1] );
2230 FETCH(&r
[3], 0, CHAN_Z
);
2231 FETCH(&r
[4], 1, CHAN_Y
);
2233 micro_mul( &r
[5], &r
[3], &r
[4] );
2234 micro_sub( &r
[2], &r
[2], &r
[5] );
2236 if (IS_CHANNEL_ENABLED( *inst
, CHAN_X
)) {
2237 STORE( &r
[2], 0, CHAN_X
);
2240 FETCH(&r
[2], 1, CHAN_X
);
2242 micro_mul( &r
[3], &r
[3], &r
[2] );
2244 FETCH(&r
[5], 0, CHAN_X
);
2246 micro_mul( &r
[1], &r
[1], &r
[5] );
2247 micro_sub( &r
[3], &r
[3], &r
[1] );
2249 if (IS_CHANNEL_ENABLED( *inst
, CHAN_Y
)) {
2250 STORE( &r
[3], 0, CHAN_Y
);
2253 micro_mul( &r
[5], &r
[5], &r
[4] );
2254 micro_mul( &r
[0], &r
[0], &r
[2] );
2255 micro_sub( &r
[5], &r
[5], &r
[0] );
2257 if (IS_CHANNEL_ENABLED( *inst
, CHAN_Z
)) {
2258 STORE( &r
[5], 0, CHAN_Z
);
2261 if (IS_CHANNEL_ENABLED( *inst
, CHAN_W
)) {
2262 STORE( &mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], 0, CHAN_W
);
2266 case TGSI_OPCODE_MULTIPLYMATRIX
:
2270 case TGSI_OPCODE_ABS
:
2271 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2272 FETCH(&r
[0], 0, chan_index
);
2274 micro_abs( &r
[0], &r
[0] );
2276 STORE(&r
[0], 0, chan_index
);
2280 case TGSI_OPCODE_RCC
:
2284 case TGSI_OPCODE_DPH
:
2285 FETCH(&r
[0], 0, CHAN_X
);
2286 FETCH(&r
[1], 1, CHAN_X
);
2288 micro_mul( &r
[0], &r
[0], &r
[1] );
2290 FETCH(&r
[1], 0, CHAN_Y
);
2291 FETCH(&r
[2], 1, CHAN_Y
);
2293 micro_mul( &r
[1], &r
[1], &r
[2] );
2294 micro_add( &r
[0], &r
[0], &r
[1] );
2296 FETCH(&r
[1], 0, CHAN_Z
);
2297 FETCH(&r
[2], 1, CHAN_Z
);
2299 micro_mul( &r
[1], &r
[1], &r
[2] );
2300 micro_add( &r
[0], &r
[0], &r
[1] );
2302 FETCH(&r
[1], 1, CHAN_W
);
2304 micro_add( &r
[0], &r
[0], &r
[1] );
2306 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2307 STORE( &r
[0], 0, chan_index
);
2311 case TGSI_OPCODE_COS
:
2312 FETCH(&r
[0], 0, CHAN_X
);
2314 micro_cos( &r
[0], &r
[0] );
2316 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2317 STORE( &r
[0], 0, chan_index
);
2321 case TGSI_OPCODE_DDX
:
2322 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2323 FETCH( &r
[0], 0, chan_index
);
2324 micro_ddx( &r
[0], &r
[0] );
2325 STORE( &r
[0], 0, chan_index
);
2329 case TGSI_OPCODE_DDY
:
2330 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2331 FETCH( &r
[0], 0, chan_index
);
2332 micro_ddy( &r
[0], &r
[0] );
2333 STORE( &r
[0], 0, chan_index
);
2337 case TGSI_OPCODE_KILP
:
2338 exec_kilp (mach
, inst
);
2341 case TGSI_OPCODE_KIL
:
2342 exec_kil (mach
, inst
);
2345 case TGSI_OPCODE_PK2H
:
2349 case TGSI_OPCODE_PK2US
:
2353 case TGSI_OPCODE_PK4B
:
2357 case TGSI_OPCODE_PK4UB
:
2361 case TGSI_OPCODE_RFL
:
2365 case TGSI_OPCODE_SEQ
:
2366 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2367 FETCH( &r
[0], 0, chan_index
);
2368 FETCH( &r
[1], 1, chan_index
);
2369 micro_eq( &r
[0], &r
[0], &r
[1],
2370 &mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
],
2371 &mach
->Temps
[TEMP_0_I
].xyzw
[TEMP_0_C
] );
2372 STORE( &r
[0], 0, chan_index
);
2376 case TGSI_OPCODE_SFL
:
2380 case TGSI_OPCODE_SGT
:
2381 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2382 FETCH( &r
[0], 0, chan_index
);
2383 FETCH( &r
[1], 1, chan_index
);
2384 micro_le( &r
[0], &r
[0], &r
[1], &mach
->Temps
[TEMP_0_I
].xyzw
[TEMP_0_C
], &mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
] );
2385 STORE( &r
[0], 0, chan_index
);
2389 case TGSI_OPCODE_SIN
:
2390 FETCH( &r
[0], 0, CHAN_X
);
2391 micro_sin( &r
[0], &r
[0] );
2392 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2393 STORE( &r
[0], 0, chan_index
);
2397 case TGSI_OPCODE_SLE
:
2398 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2399 FETCH( &r
[0], 0, chan_index
);
2400 FETCH( &r
[1], 1, chan_index
);
2401 micro_le( &r
[0], &r
[0], &r
[1], &mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], &mach
->Temps
[TEMP_0_I
].xyzw
[TEMP_0_C
] );
2402 STORE( &r
[0], 0, chan_index
);
2406 case TGSI_OPCODE_SNE
:
2407 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2408 FETCH( &r
[0], 0, chan_index
);
2409 FETCH( &r
[1], 1, chan_index
);
2410 micro_eq( &r
[0], &r
[0], &r
[1], &mach
->Temps
[TEMP_0_I
].xyzw
[TEMP_0_C
], &mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
] );
2411 STORE( &r
[0], 0, chan_index
);
2415 case TGSI_OPCODE_STR
:
2419 case TGSI_OPCODE_TEX
:
2420 /* simple texture lookup */
2421 /* src[0] = texcoord */
2422 /* src[1] = sampler unit */
2423 exec_tex(mach
, inst
, FALSE
, FALSE
);
2426 case TGSI_OPCODE_TXB
:
2427 /* Texture lookup with lod bias */
2428 /* src[0] = texcoord (src[0].w = LOD bias) */
2429 /* src[1] = sampler unit */
2430 exec_tex(mach
, inst
, TRUE
, FALSE
);
2433 case TGSI_OPCODE_TXD
:
2434 /* Texture lookup with explict partial derivatives */
2435 /* src[0] = texcoord */
2436 /* src[1] = d[strq]/dx */
2437 /* src[2] = d[strq]/dy */
2438 /* src[3] = sampler unit */
2442 case TGSI_OPCODE_TXL
:
2443 /* Texture lookup with explit LOD */
2444 /* src[0] = texcoord (src[0].w = LOD) */
2445 /* src[1] = sampler unit */
2446 exec_tex(mach
, inst
, TRUE
, FALSE
);
2449 case TGSI_OPCODE_TXP
:
2450 /* Texture lookup with projection */
2451 /* src[0] = texcoord (src[0].w = projection) */
2452 /* src[1] = sampler unit */
2453 exec_tex(mach
, inst
, FALSE
, TRUE
);
2456 case TGSI_OPCODE_UP2H
:
2460 case TGSI_OPCODE_UP2US
:
2464 case TGSI_OPCODE_UP4B
:
2468 case TGSI_OPCODE_UP4UB
:
2472 case TGSI_OPCODE_X2D
:
2476 case TGSI_OPCODE_ARA
:
2480 case TGSI_OPCODE_BRA
:
2484 case TGSI_OPCODE_CAL
:
2485 /* skip the call if no execution channels are enabled */
2486 if (mach
->ExecMask
) {
2489 /* push the Cond, Loop, Cont stacks */
2490 assert(mach
->CondStackTop
< TGSI_EXEC_MAX_COND_NESTING
);
2491 mach
->CondStack
[mach
->CondStackTop
++] = mach
->CondMask
;
2492 assert(mach
->LoopStackTop
< TGSI_EXEC_MAX_LOOP_NESTING
);
2493 mach
->LoopStack
[mach
->LoopStackTop
++] = mach
->LoopMask
;
2494 assert(mach
->ContStackTop
< TGSI_EXEC_MAX_LOOP_NESTING
);
2495 mach
->ContStack
[mach
->ContStackTop
++] = mach
->ContMask
;
2497 assert(mach
->FuncStackTop
< TGSI_EXEC_MAX_CALL_NESTING
);
2498 mach
->FuncStack
[mach
->FuncStackTop
++] = mach
->FuncMask
;
2500 /* note that PC was already incremented above */
2501 mach
->CallStack
[mach
->CallStackTop
++] = *pc
;
2502 *pc
= inst
->InstructionExtLabel
.Label
;
2506 case TGSI_OPCODE_RET
:
2507 mach
->FuncMask
&= ~mach
->ExecMask
;
2508 UPDATE_EXEC_MASK(mach
);
2510 if (mach
->FuncMask
== 0x0) {
2511 /* really return now (otherwise, keep executing */
2513 if (mach
->CallStackTop
== 0) {
2514 /* returning from main() */
2518 *pc
= mach
->CallStack
[--mach
->CallStackTop
];
2520 /* pop the Cond, Loop, Cont stacks */
2521 assert(mach
->CondStackTop
> 0);
2522 mach
->CondMask
= mach
->CondStack
[--mach
->CondStackTop
];
2523 assert(mach
->LoopStackTop
> 0);
2524 mach
->LoopMask
= mach
->LoopStack
[--mach
->LoopStackTop
];
2525 assert(mach
->ContStackTop
> 0);
2526 mach
->ContMask
= mach
->ContStack
[--mach
->ContStackTop
];
2527 assert(mach
->FuncStackTop
> 0);
2528 mach
->FuncMask
= mach
->FuncStack
[--mach
->FuncStackTop
];
2530 UPDATE_EXEC_MASK(mach
);
2534 case TGSI_OPCODE_SSG
:
2535 /* TGSI_OPCODE_SGN */
2536 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2537 FETCH( &r
[0], 0, chan_index
);
2538 micro_sgn( &r
[0], &r
[0] );
2539 STORE( &r
[0], 0, chan_index
);
2543 case TGSI_OPCODE_CMP
:
2544 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2545 FETCH(&r
[0], 0, chan_index
);
2546 FETCH(&r
[1], 1, chan_index
);
2547 FETCH(&r
[2], 2, chan_index
);
2549 micro_lt( &r
[0], &r
[0], &mach
->Temps
[TEMP_0_I
].xyzw
[TEMP_0_C
], &r
[1], &r
[2] );
2551 STORE(&r
[0], 0, chan_index
);
2555 case TGSI_OPCODE_SCS
:
2556 if( IS_CHANNEL_ENABLED( *inst
, CHAN_X
) || IS_CHANNEL_ENABLED( *inst
, CHAN_Y
) ) {
2557 FETCH( &r
[0], 0, CHAN_X
);
2559 if( IS_CHANNEL_ENABLED( *inst
, CHAN_X
) ) {
2560 micro_cos( &r
[1], &r
[0] );
2561 STORE( &r
[1], 0, CHAN_X
);
2563 if( IS_CHANNEL_ENABLED( *inst
, CHAN_Y
) ) {
2564 micro_sin( &r
[1], &r
[0] );
2565 STORE( &r
[1], 0, CHAN_Y
);
2567 if( IS_CHANNEL_ENABLED( *inst
, CHAN_Z
) ) {
2568 STORE( &mach
->Temps
[TEMP_0_I
].xyzw
[TEMP_0_C
], 0, CHAN_Z
);
2570 if( IS_CHANNEL_ENABLED( *inst
, CHAN_W
) ) {
2571 STORE( &mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], 0, CHAN_W
);
2575 case TGSI_OPCODE_NRM
:
2576 /* 3-component vector normalize */
2578 union tgsi_exec_channel tmp
, dot
;
2580 /* tmp = dp3(src0, src0): */
2581 FETCH( &r
[0], 0, CHAN_X
);
2582 micro_mul( &tmp
, &r
[0], &r
[0] );
2584 FETCH( &r
[1], 0, CHAN_Y
);
2585 micro_mul( &dot
, &r
[1], &r
[1] );
2586 micro_add( &tmp
, &tmp
, &dot
);
2588 FETCH( &r
[2], 0, CHAN_Z
);
2589 micro_mul( &dot
, &r
[2], &r
[2] );
2590 micro_add( &tmp
, &tmp
, &dot
);
2592 /* tmp = 1 / sqrt(tmp) */
2593 micro_sqrt( &tmp
, &tmp
);
2594 micro_div( &tmp
, &mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], &tmp
);
2596 /* note: w channel is undefined */
2597 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2598 /* chan = chan * tmp */
2599 micro_mul( &r
[chan_index
], &tmp
, &r
[chan_index
] );
2600 STORE( &r
[chan_index
], 0, chan_index
);
2605 case TGSI_OPCODE_NRM4
:
2606 /* 4-component vector normalize */
2608 union tgsi_exec_channel tmp
, dot
;
2610 /* tmp = dp4(src0, src0): */
2611 FETCH( &r
[0], 0, CHAN_X
);
2612 micro_mul( &tmp
, &r
[0], &r
[0] );
2614 FETCH( &r
[1], 0, CHAN_Y
);
2615 micro_mul( &dot
, &r
[1], &r
[1] );
2616 micro_add( &tmp
, &tmp
, &dot
);
2618 FETCH( &r
[2], 0, CHAN_Z
);
2619 micro_mul( &dot
, &r
[2], &r
[2] );
2620 micro_add( &tmp
, &tmp
, &dot
);
2622 FETCH( &r
[3], 0, CHAN_W
);
2623 micro_mul( &dot
, &r
[3], &r
[3] );
2624 micro_add( &tmp
, &tmp
, &dot
);
2626 /* tmp = 1 / sqrt(tmp) */
2627 micro_sqrt( &tmp
, &tmp
);
2628 micro_div( &tmp
, &mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], &tmp
);
2630 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2631 /* chan = chan * tmp */
2632 micro_mul( &r
[chan_index
], &tmp
, &r
[chan_index
] );
2633 STORE( &r
[chan_index
], 0, chan_index
);
2638 case TGSI_OPCODE_DIV
:
2642 case TGSI_OPCODE_DP2
:
2643 FETCH( &r
[0], 0, CHAN_X
);
2644 FETCH( &r
[1], 1, CHAN_X
);
2645 micro_mul( &r
[0], &r
[0], &r
[1] );
2647 FETCH( &r
[1], 0, CHAN_Y
);
2648 FETCH( &r
[2], 1, CHAN_Y
);
2649 micro_mul( &r
[1], &r
[1], &r
[2] );
2650 micro_add( &r
[0], &r
[0], &r
[1] );
2652 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2653 STORE( &r
[0], 0, chan_index
);
2657 case TGSI_OPCODE_IF
:
2659 assert(mach
->CondStackTop
< TGSI_EXEC_MAX_COND_NESTING
);
2660 mach
->CondStack
[mach
->CondStackTop
++] = mach
->CondMask
;
2661 FETCH( &r
[0], 0, CHAN_X
);
2662 /* update CondMask */
2664 mach
->CondMask
&= ~0x1;
2667 mach
->CondMask
&= ~0x2;
2670 mach
->CondMask
&= ~0x4;
2673 mach
->CondMask
&= ~0x8;
2675 UPDATE_EXEC_MASK(mach
);
2676 /* Todo: If CondMask==0, jump to ELSE */
2679 case TGSI_OPCODE_ELSE
:
2680 /* invert CondMask wrt previous mask */
2683 assert(mach
->CondStackTop
> 0);
2684 prevMask
= mach
->CondStack
[mach
->CondStackTop
- 1];
2685 mach
->CondMask
= ~mach
->CondMask
& prevMask
;
2686 UPDATE_EXEC_MASK(mach
);
2687 /* Todo: If CondMask==0, jump to ENDIF */
2691 case TGSI_OPCODE_ENDIF
:
2693 assert(mach
->CondStackTop
> 0);
2694 mach
->CondMask
= mach
->CondStack
[--mach
->CondStackTop
];
2695 UPDATE_EXEC_MASK(mach
);
2698 case TGSI_OPCODE_END
:
2699 /* halt execution */
2703 case TGSI_OPCODE_REP
:
2707 case TGSI_OPCODE_ENDREP
:
2711 case TGSI_OPCODE_PUSHA
:
2715 case TGSI_OPCODE_POPA
:
2719 case TGSI_OPCODE_CEIL
:
2720 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2721 FETCH( &r
[0], 0, chan_index
);
2722 micro_ceil( &r
[0], &r
[0] );
2723 STORE( &r
[0], 0, chan_index
);
2727 case TGSI_OPCODE_I2F
:
2728 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2729 FETCH( &r
[0], 0, chan_index
);
2730 micro_i2f( &r
[0], &r
[0] );
2731 STORE( &r
[0], 0, chan_index
);
2735 case TGSI_OPCODE_NOT
:
2736 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2737 FETCH( &r
[0], 0, chan_index
);
2738 micro_not( &r
[0], &r
[0] );
2739 STORE( &r
[0], 0, chan_index
);
2743 case TGSI_OPCODE_TRUNC
:
2744 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2745 FETCH( &r
[0], 0, chan_index
);
2746 micro_trunc( &r
[0], &r
[0] );
2747 STORE( &r
[0], 0, chan_index
);
2751 case TGSI_OPCODE_SHL
:
2752 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2753 FETCH( &r
[0], 0, chan_index
);
2754 FETCH( &r
[1], 1, chan_index
);
2755 micro_shl( &r
[0], &r
[0], &r
[1] );
2756 STORE( &r
[0], 0, chan_index
);
2760 case TGSI_OPCODE_SHR
:
2761 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2762 FETCH( &r
[0], 0, chan_index
);
2763 FETCH( &r
[1], 1, chan_index
);
2764 micro_ishr( &r
[0], &r
[0], &r
[1] );
2765 STORE( &r
[0], 0, chan_index
);
2769 case TGSI_OPCODE_AND
:
2770 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2771 FETCH( &r
[0], 0, chan_index
);
2772 FETCH( &r
[1], 1, chan_index
);
2773 micro_and( &r
[0], &r
[0], &r
[1] );
2774 STORE( &r
[0], 0, chan_index
);
2778 case TGSI_OPCODE_OR
:
2779 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2780 FETCH( &r
[0], 0, chan_index
);
2781 FETCH( &r
[1], 1, chan_index
);
2782 micro_or( &r
[0], &r
[0], &r
[1] );
2783 STORE( &r
[0], 0, chan_index
);
2787 case TGSI_OPCODE_MOD
:
2791 case TGSI_OPCODE_XOR
:
2792 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2793 FETCH( &r
[0], 0, chan_index
);
2794 FETCH( &r
[1], 1, chan_index
);
2795 micro_xor( &r
[0], &r
[0], &r
[1] );
2796 STORE( &r
[0], 0, chan_index
);
2800 case TGSI_OPCODE_SAD
:
2804 case TGSI_OPCODE_TXF
:
2808 case TGSI_OPCODE_TXQ
:
2812 case TGSI_OPCODE_EMIT
:
2813 mach
->Temps
[TEMP_OUTPUT_I
].xyzw
[TEMP_OUTPUT_C
].u
[0] += 16;
2814 mach
->Primitives
[mach
->Temps
[TEMP_PRIMITIVE_I
].xyzw
[TEMP_PRIMITIVE_C
].u
[0]]++;
2817 case TGSI_OPCODE_ENDPRIM
:
2818 mach
->Temps
[TEMP_PRIMITIVE_I
].xyzw
[TEMP_PRIMITIVE_C
].u
[0]++;
2819 mach
->Primitives
[mach
->Temps
[TEMP_PRIMITIVE_I
].xyzw
[TEMP_PRIMITIVE_C
].u
[0]] = 0;
2822 case TGSI_OPCODE_LOOP
:
2823 /* fall-through (for now) */
2824 case TGSI_OPCODE_BGNLOOP2
:
2825 /* push LoopMask and ContMasks */
2826 assert(mach
->LoopStackTop
< TGSI_EXEC_MAX_LOOP_NESTING
);
2827 mach
->LoopStack
[mach
->LoopStackTop
++] = mach
->LoopMask
;
2828 assert(mach
->ContStackTop
< TGSI_EXEC_MAX_LOOP_NESTING
);
2829 mach
->ContStack
[mach
->ContStackTop
++] = mach
->ContMask
;
2832 case TGSI_OPCODE_ENDLOOP
:
2833 /* fall-through (for now at least) */
2834 case TGSI_OPCODE_ENDLOOP2
:
2835 /* Restore ContMask, but don't pop */
2836 assert(mach
->ContStackTop
> 0);
2837 mach
->ContMask
= mach
->ContStack
[mach
->ContStackTop
- 1];
2838 UPDATE_EXEC_MASK(mach
);
2839 if (mach
->ExecMask
) {
2840 /* repeat loop: jump to instruction just past BGNLOOP */
2841 *pc
= inst
->InstructionExtLabel
.Label
+ 1;
2844 /* exit loop: pop LoopMask */
2845 assert(mach
->LoopStackTop
> 0);
2846 mach
->LoopMask
= mach
->LoopStack
[--mach
->LoopStackTop
];
2848 assert(mach
->ContStackTop
> 0);
2849 mach
->ContMask
= mach
->ContStack
[--mach
->ContStackTop
];
2851 UPDATE_EXEC_MASK(mach
);
2854 case TGSI_OPCODE_BRK
:
2855 /* turn off loop channels for each enabled exec channel */
2856 mach
->LoopMask
&= ~mach
->ExecMask
;
2857 /* Todo: if mach->LoopMask == 0, jump to end of loop */
2858 UPDATE_EXEC_MASK(mach
);
2861 case TGSI_OPCODE_CONT
:
2862 /* turn off cont channels for each enabled exec channel */
2863 mach
->ContMask
&= ~mach
->ExecMask
;
2864 /* Todo: if mach->LoopMask == 0, jump to end of loop */
2865 UPDATE_EXEC_MASK(mach
);
2868 case TGSI_OPCODE_BGNSUB
:
2872 case TGSI_OPCODE_ENDSUB
:
2876 case TGSI_OPCODE_NOISE1
:
2880 case TGSI_OPCODE_NOISE2
:
2884 case TGSI_OPCODE_NOISE3
:
2888 case TGSI_OPCODE_NOISE4
:
2892 case TGSI_OPCODE_NOP
:
2902 * Run TGSI interpreter.
2903 * \return bitmask of "alive" quad components
2906 tgsi_exec_machine_run( struct tgsi_exec_machine
*mach
)
2911 mach
->CondMask
= 0xf;
2912 mach
->LoopMask
= 0xf;
2913 mach
->ContMask
= 0xf;
2914 mach
->FuncMask
= 0xf;
2915 mach
->ExecMask
= 0xf;
2917 mach
->CondStackTop
= 0; /* temporarily subvert this assertion */
2918 assert(mach
->CondStackTop
== 0);
2919 assert(mach
->LoopStackTop
== 0);
2920 assert(mach
->ContStackTop
== 0);
2921 assert(mach
->CallStackTop
== 0);
2923 mach
->Temps
[TEMP_KILMASK_I
].xyzw
[TEMP_KILMASK_C
].u
[0] = 0;
2924 mach
->Temps
[TEMP_OUTPUT_I
].xyzw
[TEMP_OUTPUT_C
].u
[0] = 0;
2926 if( mach
->Processor
== TGSI_PROCESSOR_GEOMETRY
) {
2927 mach
->Temps
[TEMP_PRIMITIVE_I
].xyzw
[TEMP_PRIMITIVE_C
].u
[0] = 0;
2928 mach
->Primitives
[0] = 0;
2931 for (i
= 0; i
< QUAD_SIZE
; i
++) {
2932 mach
->Temps
[TEMP_CC_I
].xyzw
[TEMP_CC_C
].u
[i
] =
2933 (TGSI_EXEC_CC_EQ
<< TGSI_EXEC_CC_X_SHIFT
) |
2934 (TGSI_EXEC_CC_EQ
<< TGSI_EXEC_CC_Y_SHIFT
) |
2935 (TGSI_EXEC_CC_EQ
<< TGSI_EXEC_CC_Z_SHIFT
) |
2936 (TGSI_EXEC_CC_EQ
<< TGSI_EXEC_CC_W_SHIFT
);
2939 /* execute declarations (interpolants) */
2940 for (i
= 0; i
< mach
->NumDeclarations
; i
++) {
2941 exec_declaration( mach
, mach
->Declarations
+i
);
2944 /* execute instructions, until pc is set to -1 */
2946 assert(pc
< (int) mach
->NumInstructions
);
2947 exec_instruction( mach
, mach
->Instructions
+ pc
, &pc
);
2951 /* we scale from floats in [0,1] to Zbuffer ints in sp_quad_depth_test.c */
2952 if (mach
->Processor
== TGSI_PROCESSOR_FRAGMENT
) {
2954 * Scale back depth component.
2956 for (i
= 0; i
< 4; i
++)
2957 mach
->Outputs
[0].xyzw
[2].f
[i
] *= ctx
->DrawBuffer
->_DepthMaxF
;
2961 return ~mach
->Temps
[TEMP_KILMASK_I
].xyzw
[TEMP_KILMASK_C
].u
[0];