1 /**************************************************************************
3 * Copyright 2007-2008 Tungsten Graphics, Inc., Cedar Park, Texas.
5 * Copyright 2009-2010 VMware, Inc. All rights Reserved.
7 * Permission is hereby granted, free of charge, to any person obtaining a
8 * copy of this software and associated documentation files (the
9 * "Software"), to deal in the Software without restriction, including
10 * without limitation the rights to use, copy, modify, merge, publish,
11 * distribute, sub license, and/or sell copies of the Software, and to
12 * permit persons to whom the Software is furnished to do so, subject to
13 * the following conditions:
15 * The above copyright notice and this permission notice (including the
16 * next paragraph) shall be included in all copies or substantial portions
19 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
20 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
21 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
22 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
23 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
24 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
25 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
27 **************************************************************************/
30 * TGSI interpreter/executor.
32 * Flow control information:
34 * Since we operate on 'quads' (4 pixels or 4 vertices in parallel)
35 * flow control statements (IF/ELSE/ENDIF, LOOP/ENDLOOP) require special
36 * care since a condition may be true for some quad components but false
37 * for other components.
39 * We basically execute all statements (even if they're in the part of
40 * an IF/ELSE clause that's "not taken") and use a special mask to
41 * control writing to destination registers. This is the ExecMask.
44 * The ExecMask is computed from three other masks (CondMask, LoopMask and
45 * ContMask) which are controlled by the flow control instructions (namely:
46 * (IF/ELSE/ENDIF, LOOP/ENDLOOP and CONT).
54 #include "pipe/p_compiler.h"
55 #include "pipe/p_state.h"
56 #include "pipe/p_shader_tokens.h"
57 #include "tgsi/tgsi_dump.h"
58 #include "tgsi/tgsi_parse.h"
59 #include "tgsi/tgsi_util.h"
60 #include "tgsi_exec.h"
61 #include "util/u_memory.h"
62 #include "util/u_math.h"
67 #define TILE_TOP_LEFT 0
68 #define TILE_TOP_RIGHT 1
69 #define TILE_BOTTOM_LEFT 2
70 #define TILE_BOTTOM_RIGHT 3
73 micro_abs(union tgsi_exec_channel
*dst
,
74 const union tgsi_exec_channel
*src
)
76 dst
->f
[0] = fabsf(src
->f
[0]);
77 dst
->f
[1] = fabsf(src
->f
[1]);
78 dst
->f
[2] = fabsf(src
->f
[2]);
79 dst
->f
[3] = fabsf(src
->f
[3]);
83 micro_arl(union tgsi_exec_channel
*dst
,
84 const union tgsi_exec_channel
*src
)
86 dst
->i
[0] = (int)floorf(src
->f
[0]);
87 dst
->i
[1] = (int)floorf(src
->f
[1]);
88 dst
->i
[2] = (int)floorf(src
->f
[2]);
89 dst
->i
[3] = (int)floorf(src
->f
[3]);
93 micro_arr(union tgsi_exec_channel
*dst
,
94 const union tgsi_exec_channel
*src
)
96 dst
->i
[0] = (int)floorf(src
->f
[0] + 0.5f
);
97 dst
->i
[1] = (int)floorf(src
->f
[1] + 0.5f
);
98 dst
->i
[2] = (int)floorf(src
->f
[2] + 0.5f
);
99 dst
->i
[3] = (int)floorf(src
->f
[3] + 0.5f
);
103 micro_ceil(union tgsi_exec_channel
*dst
,
104 const union tgsi_exec_channel
*src
)
106 dst
->f
[0] = ceilf(src
->f
[0]);
107 dst
->f
[1] = ceilf(src
->f
[1]);
108 dst
->f
[2] = ceilf(src
->f
[2]);
109 dst
->f
[3] = ceilf(src
->f
[3]);
113 micro_cos(union tgsi_exec_channel
*dst
,
114 const union tgsi_exec_channel
*src
)
116 dst
->f
[0] = cosf(src
->f
[0]);
117 dst
->f
[1] = cosf(src
->f
[1]);
118 dst
->f
[2] = cosf(src
->f
[2]);
119 dst
->f
[3] = cosf(src
->f
[3]);
123 micro_ddx(union tgsi_exec_channel
*dst
,
124 const union tgsi_exec_channel
*src
)
129 dst
->f
[3] = src
->f
[TILE_BOTTOM_RIGHT
] - src
->f
[TILE_BOTTOM_LEFT
];
133 micro_ddy(union tgsi_exec_channel
*dst
,
134 const union tgsi_exec_channel
*src
)
139 dst
->f
[3] = src
->f
[TILE_BOTTOM_LEFT
] - src
->f
[TILE_TOP_LEFT
];
143 micro_exp2(union tgsi_exec_channel
*dst
,
144 const union tgsi_exec_channel
*src
)
147 dst
->f
[0] = util_fast_exp2(src
->f
[0]);
148 dst
->f
[1] = util_fast_exp2(src
->f
[1]);
149 dst
->f
[2] = util_fast_exp2(src
->f
[2]);
150 dst
->f
[3] = util_fast_exp2(src
->f
[3]);
153 /* Inf is okay for this instruction, so clamp it to silence assertions. */
155 union tgsi_exec_channel clamped
;
157 for (i
= 0; i
< 4; i
++) {
158 if (src
->f
[i
] > 127.99999f
) {
159 clamped
.f
[i
] = 127.99999f
;
160 } else if (src
->f
[i
] < -126.99999f
) {
161 clamped
.f
[i
] = -126.99999f
;
163 clamped
.f
[i
] = src
->f
[i
];
169 dst
->f
[0] = powf(2.0f
, src
->f
[0]);
170 dst
->f
[1] = powf(2.0f
, src
->f
[1]);
171 dst
->f
[2] = powf(2.0f
, src
->f
[2]);
172 dst
->f
[3] = powf(2.0f
, src
->f
[3]);
173 #endif /* FAST_MATH */
177 micro_flr(union tgsi_exec_channel
*dst
,
178 const union tgsi_exec_channel
*src
)
180 dst
->f
[0] = floorf(src
->f
[0]);
181 dst
->f
[1] = floorf(src
->f
[1]);
182 dst
->f
[2] = floorf(src
->f
[2]);
183 dst
->f
[3] = floorf(src
->f
[3]);
187 micro_frc(union tgsi_exec_channel
*dst
,
188 const union tgsi_exec_channel
*src
)
190 dst
->f
[0] = src
->f
[0] - floorf(src
->f
[0]);
191 dst
->f
[1] = src
->f
[1] - floorf(src
->f
[1]);
192 dst
->f
[2] = src
->f
[2] - floorf(src
->f
[2]);
193 dst
->f
[3] = src
->f
[3] - floorf(src
->f
[3]);
197 micro_iabs(union tgsi_exec_channel
*dst
,
198 const union tgsi_exec_channel
*src
)
200 dst
->i
[0] = src
->i
[0] >= 0 ? src
->i
[0] : -src
->i
[0];
201 dst
->i
[1] = src
->i
[1] >= 0 ? src
->i
[1] : -src
->i
[1];
202 dst
->i
[2] = src
->i
[2] >= 0 ? src
->i
[2] : -src
->i
[2];
203 dst
->i
[3] = src
->i
[3] >= 0 ? src
->i
[3] : -src
->i
[3];
207 micro_ineg(union tgsi_exec_channel
*dst
,
208 const union tgsi_exec_channel
*src
)
210 dst
->i
[0] = -src
->i
[0];
211 dst
->i
[1] = -src
->i
[1];
212 dst
->i
[2] = -src
->i
[2];
213 dst
->i
[3] = -src
->i
[3];
217 micro_lg2(union tgsi_exec_channel
*dst
,
218 const union tgsi_exec_channel
*src
)
221 dst
->f
[0] = util_fast_log2(src
->f
[0]);
222 dst
->f
[1] = util_fast_log2(src
->f
[1]);
223 dst
->f
[2] = util_fast_log2(src
->f
[2]);
224 dst
->f
[3] = util_fast_log2(src
->f
[3]);
226 dst
->f
[0] = logf(src
->f
[0]) * 1.442695f
;
227 dst
->f
[1] = logf(src
->f
[1]) * 1.442695f
;
228 dst
->f
[2] = logf(src
->f
[2]) * 1.442695f
;
229 dst
->f
[3] = logf(src
->f
[3]) * 1.442695f
;
234 micro_lrp(union tgsi_exec_channel
*dst
,
235 const union tgsi_exec_channel
*src
)
237 dst
->f
[0] = src
[0].f
[0] * (src
[1].f
[0] - src
[2].f
[0]) + src
[2].f
[0];
238 dst
->f
[1] = src
[0].f
[1] * (src
[1].f
[1] - src
[2].f
[1]) + src
[2].f
[1];
239 dst
->f
[2] = src
[0].f
[2] * (src
[1].f
[2] - src
[2].f
[2]) + src
[2].f
[2];
240 dst
->f
[3] = src
[0].f
[3] * (src
[1].f
[3] - src
[2].f
[3]) + src
[2].f
[3];
244 micro_mad(union tgsi_exec_channel
*dst
,
245 const union tgsi_exec_channel
*src
)
247 dst
->f
[0] = src
[0].f
[0] * src
[1].f
[0] + src
[2].f
[0];
248 dst
->f
[1] = src
[0].f
[1] * src
[1].f
[1] + src
[2].f
[1];
249 dst
->f
[2] = src
[0].f
[2] * src
[1].f
[2] + src
[2].f
[2];
250 dst
->f
[3] = src
[0].f
[3] * src
[1].f
[3] + src
[2].f
[3];
254 micro_mov(union tgsi_exec_channel
*dst
,
255 const union tgsi_exec_channel
*src
)
257 dst
->u
[0] = src
->u
[0];
258 dst
->u
[1] = src
->u
[1];
259 dst
->u
[2] = src
->u
[2];
260 dst
->u
[3] = src
->u
[3];
264 micro_rcp(union tgsi_exec_channel
*dst
,
265 const union tgsi_exec_channel
*src
)
267 dst
->f
[0] = 1.0f
/ src
->f
[0];
268 dst
->f
[1] = 1.0f
/ src
->f
[1];
269 dst
->f
[2] = 1.0f
/ src
->f
[2];
270 dst
->f
[3] = 1.0f
/ src
->f
[3];
274 micro_rnd(union tgsi_exec_channel
*dst
,
275 const union tgsi_exec_channel
*src
)
277 dst
->f
[0] = floorf(src
->f
[0] + 0.5f
);
278 dst
->f
[1] = floorf(src
->f
[1] + 0.5f
);
279 dst
->f
[2] = floorf(src
->f
[2] + 0.5f
);
280 dst
->f
[3] = floorf(src
->f
[3] + 0.5f
);
284 micro_rsq(union tgsi_exec_channel
*dst
,
285 const union tgsi_exec_channel
*src
)
287 dst
->f
[0] = 1.0f
/ sqrtf(fabsf(src
->f
[0]));
288 dst
->f
[1] = 1.0f
/ sqrtf(fabsf(src
->f
[1]));
289 dst
->f
[2] = 1.0f
/ sqrtf(fabsf(src
->f
[2]));
290 dst
->f
[3] = 1.0f
/ sqrtf(fabsf(src
->f
[3]));
294 micro_seq(union tgsi_exec_channel
*dst
,
295 const union tgsi_exec_channel
*src
)
297 dst
->f
[0] = src
[0].f
[0] == src
[1].f
[0] ? 1.0f
: 0.0f
;
298 dst
->f
[1] = src
[0].f
[1] == src
[1].f
[1] ? 1.0f
: 0.0f
;
299 dst
->f
[2] = src
[0].f
[2] == src
[1].f
[2] ? 1.0f
: 0.0f
;
300 dst
->f
[3] = src
[0].f
[3] == src
[1].f
[3] ? 1.0f
: 0.0f
;
304 micro_sge(union tgsi_exec_channel
*dst
,
305 const union tgsi_exec_channel
*src
)
307 dst
->f
[0] = src
[0].f
[0] >= src
[1].f
[0] ? 1.0f
: 0.0f
;
308 dst
->f
[1] = src
[0].f
[1] >= src
[1].f
[1] ? 1.0f
: 0.0f
;
309 dst
->f
[2] = src
[0].f
[2] >= src
[1].f
[2] ? 1.0f
: 0.0f
;
310 dst
->f
[3] = src
[0].f
[3] >= src
[1].f
[3] ? 1.0f
: 0.0f
;
314 micro_sgn(union tgsi_exec_channel
*dst
,
315 const union tgsi_exec_channel
*src
)
317 dst
->f
[0] = src
->f
[0] < 0.0f
? -1.0f
: src
->f
[0] > 0.0f
? 1.0f
: 0.0f
;
318 dst
->f
[1] = src
->f
[1] < 0.0f
? -1.0f
: src
->f
[1] > 0.0f
? 1.0f
: 0.0f
;
319 dst
->f
[2] = src
->f
[2] < 0.0f
? -1.0f
: src
->f
[2] > 0.0f
? 1.0f
: 0.0f
;
320 dst
->f
[3] = src
->f
[3] < 0.0f
? -1.0f
: src
->f
[3] > 0.0f
? 1.0f
: 0.0f
;
324 micro_sgt(union tgsi_exec_channel
*dst
,
325 const union tgsi_exec_channel
*src
)
327 dst
->f
[0] = src
[0].f
[0] > src
[1].f
[0] ? 1.0f
: 0.0f
;
328 dst
->f
[1] = src
[0].f
[1] > src
[1].f
[1] ? 1.0f
: 0.0f
;
329 dst
->f
[2] = src
[0].f
[2] > src
[1].f
[2] ? 1.0f
: 0.0f
;
330 dst
->f
[3] = src
[0].f
[3] > src
[1].f
[3] ? 1.0f
: 0.0f
;
334 micro_sin(union tgsi_exec_channel
*dst
,
335 const union tgsi_exec_channel
*src
)
337 dst
->f
[0] = sinf(src
->f
[0]);
338 dst
->f
[1] = sinf(src
->f
[1]);
339 dst
->f
[2] = sinf(src
->f
[2]);
340 dst
->f
[3] = sinf(src
->f
[3]);
344 micro_sle(union tgsi_exec_channel
*dst
,
345 const union tgsi_exec_channel
*src
)
347 dst
->f
[0] = src
[0].f
[0] <= src
[1].f
[0] ? 1.0f
: 0.0f
;
348 dst
->f
[1] = src
[0].f
[1] <= src
[1].f
[1] ? 1.0f
: 0.0f
;
349 dst
->f
[2] = src
[0].f
[2] <= src
[1].f
[2] ? 1.0f
: 0.0f
;
350 dst
->f
[3] = src
[0].f
[3] <= src
[1].f
[3] ? 1.0f
: 0.0f
;
354 micro_slt(union tgsi_exec_channel
*dst
,
355 const union tgsi_exec_channel
*src
)
357 dst
->f
[0] = src
[0].f
[0] < src
[1].f
[0] ? 1.0f
: 0.0f
;
358 dst
->f
[1] = src
[0].f
[1] < src
[1].f
[1] ? 1.0f
: 0.0f
;
359 dst
->f
[2] = src
[0].f
[2] < src
[1].f
[2] ? 1.0f
: 0.0f
;
360 dst
->f
[3] = src
[0].f
[3] < src
[1].f
[3] ? 1.0f
: 0.0f
;
364 micro_sne(union tgsi_exec_channel
*dst
,
365 const union tgsi_exec_channel
*src
)
367 dst
->f
[0] = src
[0].f
[0] != src
[1].f
[0] ? 1.0f
: 0.0f
;
368 dst
->f
[1] = src
[0].f
[1] != src
[1].f
[1] ? 1.0f
: 0.0f
;
369 dst
->f
[2] = src
[0].f
[2] != src
[1].f
[2] ? 1.0f
: 0.0f
;
370 dst
->f
[3] = src
[0].f
[3] != src
[1].f
[3] ? 1.0f
: 0.0f
;
374 micro_trunc(union tgsi_exec_channel
*dst
,
375 const union tgsi_exec_channel
*src
)
377 dst
->f
[0] = (float)(int)src
->f
[0];
378 dst
->f
[1] = (float)(int)src
->f
[1];
379 dst
->f
[2] = (float)(int)src
->f
[2];
380 dst
->f
[3] = (float)(int)src
->f
[3];
389 enum tgsi_exec_datatype
{
390 TGSI_EXEC_DATA_FLOAT
,
396 * Shorthand locations of various utility registers (_I = Index, _C = Channel)
398 #define TEMP_0_I TGSI_EXEC_TEMP_00000000_I
399 #define TEMP_0_C TGSI_EXEC_TEMP_00000000_C
400 #define TEMP_7F_I TGSI_EXEC_TEMP_7FFFFFFF_I
401 #define TEMP_7F_C TGSI_EXEC_TEMP_7FFFFFFF_C
402 #define TEMP_80_I TGSI_EXEC_TEMP_80000000_I
403 #define TEMP_80_C TGSI_EXEC_TEMP_80000000_C
404 #define TEMP_FF_I TGSI_EXEC_TEMP_FFFFFFFF_I
405 #define TEMP_FF_C TGSI_EXEC_TEMP_FFFFFFFF_C
406 #define TEMP_1_I TGSI_EXEC_TEMP_ONE_I
407 #define TEMP_1_C TGSI_EXEC_TEMP_ONE_C
408 #define TEMP_2_I TGSI_EXEC_TEMP_TWO_I
409 #define TEMP_2_C TGSI_EXEC_TEMP_TWO_C
410 #define TEMP_128_I TGSI_EXEC_TEMP_128_I
411 #define TEMP_128_C TGSI_EXEC_TEMP_128_C
412 #define TEMP_M128_I TGSI_EXEC_TEMP_MINUS_128_I
413 #define TEMP_M128_C TGSI_EXEC_TEMP_MINUS_128_C
414 #define TEMP_KILMASK_I TGSI_EXEC_TEMP_KILMASK_I
415 #define TEMP_KILMASK_C TGSI_EXEC_TEMP_KILMASK_C
416 #define TEMP_OUTPUT_I TGSI_EXEC_TEMP_OUTPUT_I
417 #define TEMP_OUTPUT_C TGSI_EXEC_TEMP_OUTPUT_C
418 #define TEMP_PRIMITIVE_I TGSI_EXEC_TEMP_PRIMITIVE_I
419 #define TEMP_PRIMITIVE_C TGSI_EXEC_TEMP_PRIMITIVE_C
420 #define TEMP_CC_I TGSI_EXEC_TEMP_CC_I
421 #define TEMP_CC_C TGSI_EXEC_TEMP_CC_C
422 #define TEMP_3_I TGSI_EXEC_TEMP_THREE_I
423 #define TEMP_3_C TGSI_EXEC_TEMP_THREE_C
424 #define TEMP_HALF_I TGSI_EXEC_TEMP_HALF_I
425 #define TEMP_HALF_C TGSI_EXEC_TEMP_HALF_C
426 #define TEMP_R0 TGSI_EXEC_TEMP_R0
427 #define TEMP_P0 TGSI_EXEC_TEMP_P0
429 #define IS_CHANNEL_ENABLED(INST, CHAN)\
430 ((INST).Dst[0].Register.WriteMask & (1 << (CHAN)))
432 #define IS_CHANNEL_ENABLED2(INST, CHAN)\
433 ((INST).Dst[1].Register.WriteMask & (1 << (CHAN)))
435 #define FOR_EACH_ENABLED_CHANNEL(INST, CHAN)\
436 for (CHAN = 0; CHAN < NUM_CHANNELS; CHAN++)\
437 if (IS_CHANNEL_ENABLED( INST, CHAN ))
439 #define FOR_EACH_ENABLED_CHANNEL2(INST, CHAN)\
440 for (CHAN = 0; CHAN < NUM_CHANNELS; CHAN++)\
441 if (IS_CHANNEL_ENABLED2( INST, CHAN ))
444 /** The execution mask depends on the conditional mask and the loop mask */
445 #define UPDATE_EXEC_MASK(MACH) \
446 MACH->ExecMask = MACH->CondMask & MACH->LoopMask & MACH->ContMask & MACH->Switch.mask & MACH->FuncMask
449 static const union tgsi_exec_channel ZeroVec
=
450 { { 0.0, 0.0, 0.0, 0.0 } };
453 #define CHECK_INF_OR_NAN(chan) do {\
454 assert(!util_is_inf_or_nan((chan)->f[0]));\
455 assert(!util_is_inf_or_nan((chan)->f[1]));\
456 assert(!util_is_inf_or_nan((chan)->f[2]));\
457 assert(!util_is_inf_or_nan((chan)->f[3]));\
463 print_chan(const char *msg
, const union tgsi_exec_channel
*chan
)
465 debug_printf("%s = {%f, %f, %f, %f}\n",
466 msg
, chan
->f
[0], chan
->f
[1], chan
->f
[2], chan
->f
[3]);
473 print_temp(const struct tgsi_exec_machine
*mach
, uint index
)
475 const struct tgsi_exec_vector
*tmp
= &mach
->Temps
[index
];
477 debug_printf("Temp[%u] =\n", index
);
478 for (i
= 0; i
< 4; i
++) {
479 debug_printf(" %c: { %f, %f, %f, %f }\n",
491 * Check if there's a potential src/dst register data dependency when
492 * using SOA execution.
495 * This would expand into:
500 * The second instruction will have the wrong value for t0 if executed as-is.
503 tgsi_check_soa_dependencies(const struct tgsi_full_instruction
*inst
)
507 uint writemask
= inst
->Dst
[0].Register
.WriteMask
;
508 if (writemask
== TGSI_WRITEMASK_X
||
509 writemask
== TGSI_WRITEMASK_Y
||
510 writemask
== TGSI_WRITEMASK_Z
||
511 writemask
== TGSI_WRITEMASK_W
||
512 writemask
== TGSI_WRITEMASK_NONE
) {
513 /* no chance of data dependency */
517 /* loop over src regs */
518 for (i
= 0; i
< inst
->Instruction
.NumSrcRegs
; i
++) {
519 if ((inst
->Src
[i
].Register
.File
==
520 inst
->Dst
[0].Register
.File
) &&
521 (inst
->Src
[i
].Register
.Index
==
522 inst
->Dst
[0].Register
.Index
)) {
523 /* loop over dest channels */
524 uint channelsWritten
= 0x0;
525 FOR_EACH_ENABLED_CHANNEL(*inst
, chan
) {
526 /* check if we're reading a channel that's been written */
527 uint swizzle
= tgsi_util_get_full_src_register_swizzle(&inst
->Src
[i
], chan
);
528 if (channelsWritten
& (1 << swizzle
)) {
532 channelsWritten
|= (1 << chan
);
541 * Initialize machine state by expanding tokens to full instructions,
542 * allocating temporary storage, setting up constants, etc.
543 * After this, we can call tgsi_exec_machine_run() many times.
546 tgsi_exec_machine_bind_shader(
547 struct tgsi_exec_machine
*mach
,
548 const struct tgsi_token
*tokens
,
550 struct tgsi_sampler
**samplers
)
553 struct tgsi_parse_context parse
;
554 struct tgsi_exec_labels
*labels
= &mach
->Labels
;
555 struct tgsi_full_instruction
*instructions
;
556 struct tgsi_full_declaration
*declarations
;
557 uint maxInstructions
= 10, numInstructions
= 0;
558 uint maxDeclarations
= 10, numDeclarations
= 0;
562 tgsi_dump(tokens
, 0);
567 mach
->Tokens
= tokens
;
568 mach
->Samplers
= samplers
;
570 k
= tgsi_parse_init (&parse
, mach
->Tokens
);
571 if (k
!= TGSI_PARSE_OK
) {
572 debug_printf( "Problem parsing!\n" );
576 mach
->Processor
= parse
.FullHeader
.Processor
.Processor
;
580 declarations
= (struct tgsi_full_declaration
*)
581 MALLOC( maxDeclarations
* sizeof(struct tgsi_full_declaration
) );
587 instructions
= (struct tgsi_full_instruction
*)
588 MALLOC( maxInstructions
* sizeof(struct tgsi_full_instruction
) );
591 FREE( declarations
);
595 while( !tgsi_parse_end_of_tokens( &parse
) ) {
596 uint pointer
= parse
.Position
;
599 tgsi_parse_token( &parse
);
600 switch( parse
.FullToken
.Token
.Type
) {
601 case TGSI_TOKEN_TYPE_DECLARATION
:
602 /* save expanded declaration */
603 if (numDeclarations
== maxDeclarations
) {
604 declarations
= REALLOC(declarations
,
606 * sizeof(struct tgsi_full_declaration
),
607 (maxDeclarations
+ 10)
608 * sizeof(struct tgsi_full_declaration
));
609 maxDeclarations
+= 10;
611 if (parse
.FullToken
.FullDeclaration
.Declaration
.File
== TGSI_FILE_OUTPUT
) {
613 for (reg
= parse
.FullToken
.FullDeclaration
.Range
.First
;
614 reg
<= parse
.FullToken
.FullDeclaration
.Range
.Last
;
619 memcpy(declarations
+ numDeclarations
,
620 &parse
.FullToken
.FullDeclaration
,
621 sizeof(declarations
[0]));
625 case TGSI_TOKEN_TYPE_IMMEDIATE
:
627 uint size
= parse
.FullToken
.FullImmediate
.Immediate
.NrTokens
- 1;
629 assert( mach
->ImmLimit
+ 1 <= TGSI_EXEC_NUM_IMMEDIATES
);
631 for( i
= 0; i
< size
; i
++ ) {
632 mach
->Imms
[mach
->ImmLimit
][i
] =
633 parse
.FullToken
.FullImmediate
.u
[i
].Float
;
639 case TGSI_TOKEN_TYPE_INSTRUCTION
:
640 assert( labels
->count
< MAX_LABELS
);
642 labels
->labels
[labels
->count
][0] = instno
;
643 labels
->labels
[labels
->count
][1] = pointer
;
646 /* save expanded instruction */
647 if (numInstructions
== maxInstructions
) {
648 instructions
= REALLOC(instructions
,
650 * sizeof(struct tgsi_full_instruction
),
651 (maxInstructions
+ 10)
652 * sizeof(struct tgsi_full_instruction
));
653 maxInstructions
+= 10;
656 memcpy(instructions
+ numInstructions
,
657 &parse
.FullToken
.FullInstruction
,
658 sizeof(instructions
[0]));
663 case TGSI_TOKEN_TYPE_PROPERTY
:
670 tgsi_parse_free (&parse
);
672 if (mach
->Declarations
) {
673 FREE( mach
->Declarations
);
675 mach
->Declarations
= declarations
;
676 mach
->NumDeclarations
= numDeclarations
;
678 if (mach
->Instructions
) {
679 FREE( mach
->Instructions
);
681 mach
->Instructions
= instructions
;
682 mach
->NumInstructions
= numInstructions
;
686 struct tgsi_exec_machine
*
687 tgsi_exec_machine_create( void )
689 struct tgsi_exec_machine
*mach
;
692 mach
= align_malloc( sizeof *mach
, 16 );
696 memset(mach
, 0, sizeof(*mach
));
698 mach
->Addrs
= &mach
->Temps
[TGSI_EXEC_TEMP_ADDR
];
699 mach
->MaxGeometryShaderOutputs
= TGSI_MAX_TOTAL_VERTICES
;
700 mach
->Predicates
= &mach
->Temps
[TGSI_EXEC_TEMP_P0
];
702 /* Setup constants. */
703 for( i
= 0; i
< 4; i
++ ) {
704 mach
->Temps
[TEMP_0_I
].xyzw
[TEMP_0_C
].u
[i
] = 0x00000000;
705 mach
->Temps
[TEMP_7F_I
].xyzw
[TEMP_7F_C
].u
[i
] = 0x7FFFFFFF;
706 mach
->Temps
[TEMP_80_I
].xyzw
[TEMP_80_C
].u
[i
] = 0x80000000;
707 mach
->Temps
[TEMP_FF_I
].xyzw
[TEMP_FF_C
].u
[i
] = 0xFFFFFFFF;
708 mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
].f
[i
] = 1.0f
;
709 mach
->Temps
[TEMP_2_I
].xyzw
[TEMP_2_C
].f
[i
] = 2.0f
;
710 mach
->Temps
[TEMP_128_I
].xyzw
[TEMP_128_C
].f
[i
] = 128.0f
;
711 mach
->Temps
[TEMP_M128_I
].xyzw
[TEMP_M128_C
].f
[i
] = -128.0f
;
712 mach
->Temps
[TEMP_3_I
].xyzw
[TEMP_3_C
].f
[i
] = 3.0f
;
713 mach
->Temps
[TEMP_HALF_I
].xyzw
[TEMP_HALF_C
].f
[i
] = 0.5f
;
717 /* silence warnings */
731 tgsi_exec_machine_destroy(struct tgsi_exec_machine
*mach
)
734 FREE(mach
->Instructions
);
735 FREE(mach
->Declarations
);
743 union tgsi_exec_channel
*dst
,
744 const union tgsi_exec_channel
*src0
,
745 const union tgsi_exec_channel
*src1
)
747 dst
->f
[0] = src0
->f
[0] + src1
->f
[0];
748 dst
->f
[1] = src0
->f
[1] + src1
->f
[1];
749 dst
->f
[2] = src0
->f
[2] + src1
->f
[2];
750 dst
->f
[3] = src0
->f
[3] + src1
->f
[3];
755 union tgsi_exec_channel
*dst
,
756 const union tgsi_exec_channel
*src0
,
757 const union tgsi_exec_channel
*src1
)
759 if (src1
->f
[0] != 0) {
760 dst
->f
[0] = src0
->f
[0] / src1
->f
[0];
762 if (src1
->f
[1] != 0) {
763 dst
->f
[1] = src0
->f
[1] / src1
->f
[1];
765 if (src1
->f
[2] != 0) {
766 dst
->f
[2] = src0
->f
[2] / src1
->f
[2];
768 if (src1
->f
[3] != 0) {
769 dst
->f
[3] = src0
->f
[3] / src1
->f
[3];
774 micro_float_clamp(union tgsi_exec_channel
*dst
,
775 const union tgsi_exec_channel
*src
)
779 for (i
= 0; i
< 4; i
++) {
780 if (src
->f
[i
] > 0.0f
) {
781 if (src
->f
[i
] > 1.884467e+019f
)
782 dst
->f
[i
] = 1.884467e+019f
;
783 else if (src
->f
[i
] < 5.42101e-020f
)
784 dst
->f
[i
] = 5.42101e-020f
;
786 dst
->f
[i
] = src
->f
[i
];
789 if (src
->f
[i
] < -1.884467e+019f
)
790 dst
->f
[i
] = -1.884467e+019f
;
791 else if (src
->f
[i
] > -5.42101e-020f
)
792 dst
->f
[i
] = -5.42101e-020f
;
794 dst
->f
[i
] = src
->f
[i
];
801 union tgsi_exec_channel
*dst
,
802 const union tgsi_exec_channel
*src0
,
803 const union tgsi_exec_channel
*src1
,
804 const union tgsi_exec_channel
*src2
,
805 const union tgsi_exec_channel
*src3
)
807 dst
->f
[0] = src0
->f
[0] < src1
->f
[0] ? src2
->f
[0] : src3
->f
[0];
808 dst
->f
[1] = src0
->f
[1] < src1
->f
[1] ? src2
->f
[1] : src3
->f
[1];
809 dst
->f
[2] = src0
->f
[2] < src1
->f
[2] ? src2
->f
[2] : src3
->f
[2];
810 dst
->f
[3] = src0
->f
[3] < src1
->f
[3] ? src2
->f
[3] : src3
->f
[3];
815 union tgsi_exec_channel
*dst
,
816 const union tgsi_exec_channel
*src0
,
817 const union tgsi_exec_channel
*src1
)
819 dst
->f
[0] = src0
->f
[0] > src1
->f
[0] ? src0
->f
[0] : src1
->f
[0];
820 dst
->f
[1] = src0
->f
[1] > src1
->f
[1] ? src0
->f
[1] : src1
->f
[1];
821 dst
->f
[2] = src0
->f
[2] > src1
->f
[2] ? src0
->f
[2] : src1
->f
[2];
822 dst
->f
[3] = src0
->f
[3] > src1
->f
[3] ? src0
->f
[3] : src1
->f
[3];
827 union tgsi_exec_channel
*dst
,
828 const union tgsi_exec_channel
*src0
,
829 const union tgsi_exec_channel
*src1
)
831 dst
->f
[0] = src0
->f
[0] < src1
->f
[0] ? src0
->f
[0] : src1
->f
[0];
832 dst
->f
[1] = src0
->f
[1] < src1
->f
[1] ? src0
->f
[1] : src1
->f
[1];
833 dst
->f
[2] = src0
->f
[2] < src1
->f
[2] ? src0
->f
[2] : src1
->f
[2];
834 dst
->f
[3] = src0
->f
[3] < src1
->f
[3] ? src0
->f
[3] : src1
->f
[3];
839 union tgsi_exec_channel
*dst
,
840 const union tgsi_exec_channel
*src0
,
841 const union tgsi_exec_channel
*src1
)
843 dst
->f
[0] = src0
->f
[0] * src1
->f
[0];
844 dst
->f
[1] = src0
->f
[1] * src1
->f
[1];
845 dst
->f
[2] = src0
->f
[2] * src1
->f
[2];
846 dst
->f
[3] = src0
->f
[3] * src1
->f
[3];
852 union tgsi_exec_channel
*dst0
,
853 union tgsi_exec_channel
*dst1
,
854 const union tgsi_exec_channel
*src0
,
855 const union tgsi_exec_channel
*src1
)
857 dst1
->i
[0] = src0
->i
[0] * src1
->i
[0];
858 dst1
->i
[1] = src0
->i
[1] * src1
->i
[1];
859 dst1
->i
[2] = src0
->i
[2] * src1
->i
[2];
860 dst1
->i
[3] = src0
->i
[3] * src1
->i
[3];
871 union tgsi_exec_channel
*dst0
,
872 union tgsi_exec_channel
*dst1
,
873 const union tgsi_exec_channel
*src0
,
874 const union tgsi_exec_channel
*src1
)
876 dst1
->u
[0] = src0
->u
[0] * src1
->u
[0];
877 dst1
->u
[1] = src0
->u
[1] * src1
->u
[1];
878 dst1
->u
[2] = src0
->u
[2] * src1
->u
[2];
879 dst1
->u
[3] = src0
->u
[3] * src1
->u
[3];
891 union tgsi_exec_channel
*dst
,
892 const union tgsi_exec_channel
*src0
,
893 const union tgsi_exec_channel
*src1
,
894 const union tgsi_exec_channel
*src2
)
896 dst
->u
[0] = src0
->u
[0] ? src1
->u
[0] : src2
->u
[0];
897 dst
->u
[1] = src0
->u
[1] ? src1
->u
[1] : src2
->u
[1];
898 dst
->u
[2] = src0
->u
[2] ? src1
->u
[2] : src2
->u
[2];
899 dst
->u
[3] = src0
->u
[3] ? src1
->u
[3] : src2
->u
[3];
905 union tgsi_exec_channel
*dst
,
906 const union tgsi_exec_channel
*src
)
908 dst
->f
[0] = -src
->f
[0];
909 dst
->f
[1] = -src
->f
[1];
910 dst
->f
[2] = -src
->f
[2];
911 dst
->f
[3] = -src
->f
[3];
916 union tgsi_exec_channel
*dst
,
917 const union tgsi_exec_channel
*src0
,
918 const union tgsi_exec_channel
*src1
)
921 dst
->f
[0] = util_fast_pow( src0
->f
[0], src1
->f
[0] );
922 dst
->f
[1] = util_fast_pow( src0
->f
[1], src1
->f
[1] );
923 dst
->f
[2] = util_fast_pow( src0
->f
[2], src1
->f
[2] );
924 dst
->f
[3] = util_fast_pow( src0
->f
[3], src1
->f
[3] );
926 dst
->f
[0] = powf( src0
->f
[0], src1
->f
[0] );
927 dst
->f
[1] = powf( src0
->f
[1], src1
->f
[1] );
928 dst
->f
[2] = powf( src0
->f
[2], src1
->f
[2] );
929 dst
->f
[3] = powf( src0
->f
[3], src1
->f
[3] );
934 micro_sqrt( union tgsi_exec_channel
*dst
,
935 const union tgsi_exec_channel
*src
)
937 dst
->f
[0] = sqrtf( src
->f
[0] );
938 dst
->f
[1] = sqrtf( src
->f
[1] );
939 dst
->f
[2] = sqrtf( src
->f
[2] );
940 dst
->f
[3] = sqrtf( src
->f
[3] );
945 union tgsi_exec_channel
*dst
,
946 const union tgsi_exec_channel
*src0
,
947 const union tgsi_exec_channel
*src1
)
949 dst
->f
[0] = src0
->f
[0] - src1
->f
[0];
950 dst
->f
[1] = src0
->f
[1] - src1
->f
[1];
951 dst
->f
[2] = src0
->f
[2] - src1
->f
[2];
952 dst
->f
[3] = src0
->f
[3] - src1
->f
[3];
956 fetch_src_file_channel(
957 const struct tgsi_exec_machine
*mach
,
960 const union tgsi_exec_channel
*index
,
961 union tgsi_exec_channel
*chan
)
969 case TGSI_FILE_CONSTANT
:
970 assert(mach
->Consts
);
974 chan
->f
[0] = mach
->Consts
[index
->i
[0]][swizzle
];
978 chan
->f
[1] = mach
->Consts
[index
->i
[1]][swizzle
];
982 chan
->f
[2] = mach
->Consts
[index
->i
[2]][swizzle
];
986 chan
->f
[3] = mach
->Consts
[index
->i
[3]][swizzle
];
989 case TGSI_FILE_INPUT
:
990 case TGSI_FILE_SYSTEM_VALUE
:
991 chan
->u
[0] = mach
->Inputs
[index
->i
[0]].xyzw
[swizzle
].u
[0];
992 chan
->u
[1] = mach
->Inputs
[index
->i
[1]].xyzw
[swizzle
].u
[1];
993 chan
->u
[2] = mach
->Inputs
[index
->i
[2]].xyzw
[swizzle
].u
[2];
994 chan
->u
[3] = mach
->Inputs
[index
->i
[3]].xyzw
[swizzle
].u
[3];
997 case TGSI_FILE_TEMPORARY
:
998 assert(index
->i
[0] < TGSI_EXEC_NUM_TEMPS
);
999 chan
->u
[0] = mach
->Temps
[index
->i
[0]].xyzw
[swizzle
].u
[0];
1000 chan
->u
[1] = mach
->Temps
[index
->i
[1]].xyzw
[swizzle
].u
[1];
1001 chan
->u
[2] = mach
->Temps
[index
->i
[2]].xyzw
[swizzle
].u
[2];
1002 chan
->u
[3] = mach
->Temps
[index
->i
[3]].xyzw
[swizzle
].u
[3];
1005 case TGSI_FILE_IMMEDIATE
:
1006 assert( index
->i
[0] < (int) mach
->ImmLimit
);
1007 chan
->f
[0] = mach
->Imms
[index
->i
[0]][swizzle
];
1008 assert( index
->i
[1] < (int) mach
->ImmLimit
);
1009 chan
->f
[1] = mach
->Imms
[index
->i
[1]][swizzle
];
1010 assert( index
->i
[2] < (int) mach
->ImmLimit
);
1011 chan
->f
[2] = mach
->Imms
[index
->i
[2]][swizzle
];
1012 assert( index
->i
[3] < (int) mach
->ImmLimit
);
1013 chan
->f
[3] = mach
->Imms
[index
->i
[3]][swizzle
];
1016 case TGSI_FILE_ADDRESS
:
1017 chan
->u
[0] = mach
->Addrs
[index
->i
[0]].xyzw
[swizzle
].u
[0];
1018 chan
->u
[1] = mach
->Addrs
[index
->i
[1]].xyzw
[swizzle
].u
[1];
1019 chan
->u
[2] = mach
->Addrs
[index
->i
[2]].xyzw
[swizzle
].u
[2];
1020 chan
->u
[3] = mach
->Addrs
[index
->i
[3]].xyzw
[swizzle
].u
[3];
1023 case TGSI_FILE_PREDICATE
:
1024 assert(index
->i
[0] < TGSI_EXEC_NUM_PREDS
);
1025 assert(index
->i
[1] < TGSI_EXEC_NUM_PREDS
);
1026 assert(index
->i
[2] < TGSI_EXEC_NUM_PREDS
);
1027 assert(index
->i
[3] < TGSI_EXEC_NUM_PREDS
);
1028 chan
->u
[0] = mach
->Predicates
[0].xyzw
[swizzle
].u
[0];
1029 chan
->u
[1] = mach
->Predicates
[0].xyzw
[swizzle
].u
[1];
1030 chan
->u
[2] = mach
->Predicates
[0].xyzw
[swizzle
].u
[2];
1031 chan
->u
[3] = mach
->Predicates
[0].xyzw
[swizzle
].u
[3];
1034 case TGSI_FILE_OUTPUT
:
1035 /* vertex/fragment output vars can be read too */
1036 chan
->u
[0] = mach
->Outputs
[index
->i
[0]].xyzw
[swizzle
].u
[0];
1037 chan
->u
[1] = mach
->Outputs
[index
->i
[1]].xyzw
[swizzle
].u
[1];
1038 chan
->u
[2] = mach
->Outputs
[index
->i
[2]].xyzw
[swizzle
].u
[2];
1039 chan
->u
[3] = mach
->Outputs
[index
->i
[3]].xyzw
[swizzle
].u
[3];
1053 fetch_source(const struct tgsi_exec_machine
*mach
,
1054 union tgsi_exec_channel
*chan
,
1055 const struct tgsi_full_src_register
*reg
,
1056 const uint chan_index
,
1057 enum tgsi_exec_datatype src_datatype
)
1059 union tgsi_exec_channel index
;
1062 /* We start with a direct index into a register file.
1066 * file = Register.File
1067 * [1] = Register.Index
1072 index
.i
[3] = reg
->Register
.Index
;
1074 /* There is an extra source register that indirectly subscripts
1075 * a register file. The direct index now becomes an offset
1076 * that is being added to the indirect register.
1080 * ind = Indirect.File
1081 * [2] = Indirect.Index
1082 * .x = Indirect.SwizzleX
1084 if (reg
->Register
.Indirect
) {
1085 union tgsi_exec_channel index2
;
1086 union tgsi_exec_channel indir_index
;
1087 const uint execmask
= mach
->ExecMask
;
1090 /* which address register (always zero now) */
1094 index2
.i
[3] = reg
->Indirect
.Index
;
1096 /* get current value of address register[swizzle] */
1097 swizzle
= tgsi_util_get_src_register_swizzle( ®
->Indirect
, CHAN_X
);
1098 fetch_src_file_channel(
1105 /* add value of address register to the offset */
1106 index
.i
[0] += indir_index
.i
[0];
1107 index
.i
[1] += indir_index
.i
[1];
1108 index
.i
[2] += indir_index
.i
[2];
1109 index
.i
[3] += indir_index
.i
[3];
1111 /* for disabled execution channels, zero-out the index to
1112 * avoid using a potential garbage value.
1114 for (i
= 0; i
< QUAD_SIZE
; i
++) {
1115 if ((execmask
& (1 << i
)) == 0)
1120 /* There is an extra source register that is a second
1121 * subscript to a register file. Effectively it means that
1122 * the register file is actually a 2D array of registers.
1124 * file[1][3] == file[1*sizeof(file[1])+3],
1126 * [3] = Dimension.Index
1128 if (reg
->Register
.Dimension
) {
1129 /* The size of the first-order array depends on the register file type.
1130 * We need to multiply the index to the first array to get an effective,
1131 * "flat" index that points to the beginning of the second-order array.
1133 switch (reg
->Register
.File
) {
1134 case TGSI_FILE_INPUT
:
1135 case TGSI_FILE_SYSTEM_VALUE
:
1136 index
.i
[0] *= TGSI_EXEC_MAX_INPUT_ATTRIBS
;
1137 index
.i
[1] *= TGSI_EXEC_MAX_INPUT_ATTRIBS
;
1138 index
.i
[2] *= TGSI_EXEC_MAX_INPUT_ATTRIBS
;
1139 index
.i
[3] *= TGSI_EXEC_MAX_INPUT_ATTRIBS
;
1141 case TGSI_FILE_CONSTANT
:
1142 index
.i
[0] *= TGSI_EXEC_MAX_CONST_BUFFER
;
1143 index
.i
[1] *= TGSI_EXEC_MAX_CONST_BUFFER
;
1144 index
.i
[2] *= TGSI_EXEC_MAX_CONST_BUFFER
;
1145 index
.i
[3] *= TGSI_EXEC_MAX_CONST_BUFFER
;
1151 index
.i
[0] += reg
->Dimension
.Index
;
1152 index
.i
[1] += reg
->Dimension
.Index
;
1153 index
.i
[2] += reg
->Dimension
.Index
;
1154 index
.i
[3] += reg
->Dimension
.Index
;
1156 /* Again, the second subscript index can be addressed indirectly
1157 * identically to the first one.
1158 * Nothing stops us from indirectly addressing the indirect register,
1159 * but there is no need for that, so we won't exercise it.
1161 * file[1][ind[4].y+3],
1163 * ind = DimIndirect.File
1164 * [4] = DimIndirect.Index
1165 * .y = DimIndirect.SwizzleX
1167 if (reg
->Dimension
.Indirect
) {
1168 union tgsi_exec_channel index2
;
1169 union tgsi_exec_channel indir_index
;
1170 const uint execmask
= mach
->ExecMask
;
1176 index2
.i
[3] = reg
->DimIndirect
.Index
;
1178 swizzle
= tgsi_util_get_src_register_swizzle( ®
->DimIndirect
, CHAN_X
);
1179 fetch_src_file_channel(
1181 reg
->DimIndirect
.File
,
1186 index
.i
[0] += indir_index
.i
[0];
1187 index
.i
[1] += indir_index
.i
[1];
1188 index
.i
[2] += indir_index
.i
[2];
1189 index
.i
[3] += indir_index
.i
[3];
1191 /* for disabled execution channels, zero-out the index to
1192 * avoid using a potential garbage value.
1194 for (i
= 0; i
< QUAD_SIZE
; i
++) {
1195 if ((execmask
& (1 << i
)) == 0)
1200 /* If by any chance there was a need for a 3D array of register
1201 * files, we would have to check whether Dimension is followed
1202 * by a dimension register and continue the saga.
1206 swizzle
= tgsi_util_get_full_src_register_swizzle( reg
, chan_index
);
1207 fetch_src_file_channel(
1214 if (reg
->Register
.Absolute
) {
1215 if (src_datatype
== TGSI_EXEC_DATA_FLOAT
) {
1216 micro_abs(chan
, chan
);
1218 micro_iabs(chan
, chan
);
1222 if (reg
->Register
.Negate
) {
1223 if (src_datatype
== TGSI_EXEC_DATA_FLOAT
) {
1224 micro_neg(chan
, chan
);
1226 micro_ineg(chan
, chan
);
1232 store_dest(struct tgsi_exec_machine
*mach
,
1233 const union tgsi_exec_channel
*chan
,
1234 const struct tgsi_full_dst_register
*reg
,
1235 const struct tgsi_full_instruction
*inst
,
1237 enum tgsi_exec_datatype dst_datatype
)
1240 union tgsi_exec_channel null
;
1241 union tgsi_exec_channel
*dst
;
1242 uint execmask
= mach
->ExecMask
;
1243 int offset
= 0; /* indirection offset */
1246 if (dst_datatype
== TGSI_EXEC_DATA_FLOAT
) {
1247 CHECK_INF_OR_NAN(chan
);
1250 /* There is an extra source register that indirectly subscripts
1251 * a register file. The direct index now becomes an offset
1252 * that is being added to the indirect register.
1256 * ind = Indirect.File
1257 * [2] = Indirect.Index
1258 * .x = Indirect.SwizzleX
1260 if (reg
->Register
.Indirect
) {
1261 union tgsi_exec_channel index
;
1262 union tgsi_exec_channel indir_index
;
1265 /* which address register (always zero for now) */
1269 index
.i
[3] = reg
->Indirect
.Index
;
1271 /* get current value of address register[swizzle] */
1272 swizzle
= tgsi_util_get_src_register_swizzle( ®
->Indirect
, CHAN_X
);
1274 /* fetch values from the address/indirection register */
1275 fetch_src_file_channel(
1282 /* save indirection offset */
1283 offset
= indir_index
.i
[0];
1286 switch (reg
->Register
.File
) {
1287 case TGSI_FILE_NULL
:
1291 case TGSI_FILE_OUTPUT
:
1292 index
= mach
->Temps
[TEMP_OUTPUT_I
].xyzw
[TEMP_OUTPUT_C
].u
[0]
1293 + reg
->Register
.Index
;
1294 dst
= &mach
->Outputs
[offset
+ index
].xyzw
[chan_index
];
1296 if (TGSI_PROCESSOR_GEOMETRY
== mach
->Processor
) {
1297 fprintf(stderr
, "STORING OUT[%d] mask(%d), = (", offset
+ index
, execmask
);
1298 for (i
= 0; i
< QUAD_SIZE
; i
++)
1299 if (execmask
& (1 << i
))
1300 fprintf(stderr
, "%f, ", chan
->f
[i
]);
1301 fprintf(stderr
, ")\n");
1306 case TGSI_FILE_TEMPORARY
:
1307 index
= reg
->Register
.Index
;
1308 assert( index
< TGSI_EXEC_NUM_TEMPS
);
1309 dst
= &mach
->Temps
[offset
+ index
].xyzw
[chan_index
];
1312 case TGSI_FILE_ADDRESS
:
1313 index
= reg
->Register
.Index
;
1314 dst
= &mach
->Addrs
[index
].xyzw
[chan_index
];
1317 case TGSI_FILE_LOOP
:
1318 assert(reg
->Register
.Index
== 0);
1319 assert(mach
->LoopCounterStackTop
> 0);
1320 assert(chan_index
== CHAN_X
);
1321 dst
= &mach
->LoopCounterStack
[mach
->LoopCounterStackTop
- 1].xyzw
[chan_index
];
1324 case TGSI_FILE_PREDICATE
:
1325 index
= reg
->Register
.Index
;
1326 assert(index
< TGSI_EXEC_NUM_PREDS
);
1327 dst
= &mach
->Predicates
[index
].xyzw
[chan_index
];
1335 if (inst
->Instruction
.Predicate
) {
1337 union tgsi_exec_channel
*pred
;
1339 switch (chan_index
) {
1341 swizzle
= inst
->Predicate
.SwizzleX
;
1344 swizzle
= inst
->Predicate
.SwizzleY
;
1347 swizzle
= inst
->Predicate
.SwizzleZ
;
1350 swizzle
= inst
->Predicate
.SwizzleW
;
1357 assert(inst
->Predicate
.Index
== 0);
1359 pred
= &mach
->Predicates
[inst
->Predicate
.Index
].xyzw
[swizzle
];
1361 if (inst
->Predicate
.Negate
) {
1362 for (i
= 0; i
< QUAD_SIZE
; i
++) {
1364 execmask
&= ~(1 << i
);
1368 for (i
= 0; i
< QUAD_SIZE
; i
++) {
1370 execmask
&= ~(1 << i
);
1376 switch (inst
->Instruction
.Saturate
) {
1378 for (i
= 0; i
< QUAD_SIZE
; i
++)
1379 if (execmask
& (1 << i
))
1380 dst
->i
[i
] = chan
->i
[i
];
1383 case TGSI_SAT_ZERO_ONE
:
1384 for (i
= 0; i
< QUAD_SIZE
; i
++)
1385 if (execmask
& (1 << i
)) {
1386 if (chan
->f
[i
] < 0.0f
)
1388 else if (chan
->f
[i
] > 1.0f
)
1391 dst
->i
[i
] = chan
->i
[i
];
1395 case TGSI_SAT_MINUS_PLUS_ONE
:
1396 for (i
= 0; i
< QUAD_SIZE
; i
++)
1397 if (execmask
& (1 << i
)) {
1398 if (chan
->f
[i
] < -1.0f
)
1400 else if (chan
->f
[i
] > 1.0f
)
1403 dst
->i
[i
] = chan
->i
[i
];
1412 #define FETCH(VAL,INDEX,CHAN)\
1413 fetch_source(mach, VAL, &inst->Src[INDEX], CHAN, TGSI_EXEC_DATA_FLOAT)
1415 #define STORE(VAL,INDEX,CHAN)\
1416 store_dest(mach, VAL, &inst->Dst[INDEX], inst, CHAN, TGSI_EXEC_DATA_FLOAT)
1420 * Execute ARB-style KIL which is predicated by a src register.
1421 * Kill fragment if any of the four values is less than zero.
1424 exec_kil(struct tgsi_exec_machine
*mach
,
1425 const struct tgsi_full_instruction
*inst
)
1429 uint kilmask
= 0; /* bit 0 = pixel 0, bit 1 = pixel 1, etc */
1430 union tgsi_exec_channel r
[1];
1432 /* This mask stores component bits that were already tested. */
1435 for (chan_index
= 0; chan_index
< 4; chan_index
++)
1440 /* unswizzle channel */
1441 swizzle
= tgsi_util_get_full_src_register_swizzle (
1445 /* check if the component has not been already tested */
1446 if (uniquemask
& (1 << swizzle
))
1448 uniquemask
|= 1 << swizzle
;
1450 FETCH(&r
[0], 0, chan_index
);
1451 for (i
= 0; i
< 4; i
++)
1452 if (r
[0].f
[i
] < 0.0f
)
1456 mach
->Temps
[TEMP_KILMASK_I
].xyzw
[TEMP_KILMASK_C
].u
[0] |= kilmask
;
1460 * Execute NVIDIA-style KIL which is predicated by a condition code.
1461 * Kill fragment if the condition code is TRUE.
1464 exec_kilp(struct tgsi_exec_machine
*mach
,
1465 const struct tgsi_full_instruction
*inst
)
1467 uint kilmask
; /* bit 0 = pixel 0, bit 1 = pixel 1, etc */
1469 /* "unconditional" kil */
1470 kilmask
= mach
->ExecMask
;
1471 mach
->Temps
[TEMP_KILMASK_I
].xyzw
[TEMP_KILMASK_C
].u
[0] |= kilmask
;
1475 emit_vertex(struct tgsi_exec_machine
*mach
)
1477 /* FIXME: check for exec mask correctly
1479 for (i = 0; i < QUAD_SIZE; ++i) {
1480 if ((mach->ExecMask & (1 << i)))
1482 if (mach
->ExecMask
) {
1483 mach
->Temps
[TEMP_OUTPUT_I
].xyzw
[TEMP_OUTPUT_C
].u
[0] += mach
->NumOutputs
;
1484 mach
->Primitives
[mach
->Temps
[TEMP_PRIMITIVE_I
].xyzw
[TEMP_PRIMITIVE_C
].u
[0]]++;
1489 emit_primitive(struct tgsi_exec_machine
*mach
)
1491 unsigned *prim_count
= &mach
->Temps
[TEMP_PRIMITIVE_I
].xyzw
[TEMP_PRIMITIVE_C
].u
[0];
1492 /* FIXME: check for exec mask correctly
1494 for (i = 0; i < QUAD_SIZE; ++i) {
1495 if ((mach->ExecMask & (1 << i)))
1497 if (mach
->ExecMask
) {
1499 debug_assert((*prim_count
* mach
->NumOutputs
) < mach
->MaxGeometryShaderOutputs
);
1500 mach
->Primitives
[*prim_count
] = 0;
1505 * Fetch a four texture samples using STR texture coordinates.
1508 fetch_texel( struct tgsi_sampler
*sampler
,
1509 const union tgsi_exec_channel
*s
,
1510 const union tgsi_exec_channel
*t
,
1511 const union tgsi_exec_channel
*p
,
1512 const union tgsi_exec_channel
*c0
,
1513 enum tgsi_sampler_control control
,
1514 union tgsi_exec_channel
*r
,
1515 union tgsi_exec_channel
*g
,
1516 union tgsi_exec_channel
*b
,
1517 union tgsi_exec_channel
*a
)
1520 float rgba
[NUM_CHANNELS
][QUAD_SIZE
];
1522 sampler
->get_samples(sampler
, s
->f
, t
->f
, p
->f
, c0
->f
, control
, rgba
);
1524 for (j
= 0; j
< 4; j
++) {
1525 r
->f
[j
] = rgba
[0][j
];
1526 g
->f
[j
] = rgba
[1][j
];
1527 b
->f
[j
] = rgba
[2][j
];
1528 a
->f
[j
] = rgba
[3][j
];
1533 #define TEX_MODIFIER_NONE 0
1534 #define TEX_MODIFIER_PROJECTED 1
1535 #define TEX_MODIFIER_LOD_BIAS 2
1536 #define TEX_MODIFIER_EXPLICIT_LOD 3
1540 exec_tex(struct tgsi_exec_machine
*mach
,
1541 const struct tgsi_full_instruction
*inst
,
1544 const uint unit
= inst
->Src
[1].Register
.Index
;
1545 union tgsi_exec_channel r
[4];
1546 const union tgsi_exec_channel
*lod
= &ZeroVec
;
1547 enum tgsi_sampler_control control
;
1550 if (modifier
!= TEX_MODIFIER_NONE
) {
1551 FETCH(&r
[3], 0, CHAN_W
);
1552 if (modifier
!= TEX_MODIFIER_PROJECTED
) {
1557 if (modifier
== TEX_MODIFIER_EXPLICIT_LOD
) {
1558 control
= tgsi_sampler_lod_explicit
;
1560 control
= tgsi_sampler_lod_bias
;
1563 switch (inst
->Texture
.Texture
) {
1564 case TGSI_TEXTURE_1D
:
1565 case TGSI_TEXTURE_SHADOW1D
:
1566 FETCH(&r
[0], 0, CHAN_X
);
1568 if (modifier
== TEX_MODIFIER_PROJECTED
) {
1569 micro_div(&r
[0], &r
[0], &r
[3]);
1572 fetch_texel(mach
->Samplers
[unit
],
1573 &r
[0], &ZeroVec
, &ZeroVec
, lod
, /* S, T, P, LOD */
1575 &r
[0], &r
[1], &r
[2], &r
[3]); /* R, G, B, A */
1578 case TGSI_TEXTURE_2D
:
1579 case TGSI_TEXTURE_RECT
:
1580 case TGSI_TEXTURE_SHADOW2D
:
1581 case TGSI_TEXTURE_SHADOWRECT
:
1582 FETCH(&r
[0], 0, CHAN_X
);
1583 FETCH(&r
[1], 0, CHAN_Y
);
1584 FETCH(&r
[2], 0, CHAN_Z
);
1586 if (modifier
== TEX_MODIFIER_PROJECTED
) {
1587 micro_div(&r
[0], &r
[0], &r
[3]);
1588 micro_div(&r
[1], &r
[1], &r
[3]);
1589 micro_div(&r
[2], &r
[2], &r
[3]);
1592 fetch_texel(mach
->Samplers
[unit
],
1593 &r
[0], &r
[1], &r
[2], lod
, /* S, T, P, LOD */
1595 &r
[0], &r
[1], &r
[2], &r
[3]); /* outputs */
1598 case TGSI_TEXTURE_3D
:
1599 case TGSI_TEXTURE_CUBE
:
1600 FETCH(&r
[0], 0, CHAN_X
);
1601 FETCH(&r
[1], 0, CHAN_Y
);
1602 FETCH(&r
[2], 0, CHAN_Z
);
1604 if (modifier
== TEX_MODIFIER_PROJECTED
) {
1605 micro_div(&r
[0], &r
[0], &r
[3]);
1606 micro_div(&r
[1], &r
[1], &r
[3]);
1607 micro_div(&r
[2], &r
[2], &r
[3]);
1610 fetch_texel(mach
->Samplers
[unit
],
1611 &r
[0], &r
[1], &r
[2], lod
,
1613 &r
[0], &r
[1], &r
[2], &r
[3]);
1620 FOR_EACH_ENABLED_CHANNEL(*inst
, chan_index
) {
1621 STORE(&r
[chan_index
], 0, chan_index
);
1626 exec_txd(struct tgsi_exec_machine
*mach
,
1627 const struct tgsi_full_instruction
*inst
)
1629 const uint unit
= inst
->Src
[3].Register
.Index
;
1630 union tgsi_exec_channel r
[4];
1634 * XXX: This is fake TXD -- the derivatives are not taken into account, yet.
1637 switch (inst
->Texture
.Texture
) {
1638 case TGSI_TEXTURE_1D
:
1639 case TGSI_TEXTURE_SHADOW1D
:
1641 FETCH(&r
[0], 0, CHAN_X
);
1643 fetch_texel(mach
->Samplers
[unit
],
1644 &r
[0], &ZeroVec
, &ZeroVec
, &ZeroVec
, /* S, T, P, BIAS */
1645 tgsi_sampler_lod_bias
,
1646 &r
[0], &r
[1], &r
[2], &r
[3]); /* R, G, B, A */
1649 case TGSI_TEXTURE_2D
:
1650 case TGSI_TEXTURE_RECT
:
1651 case TGSI_TEXTURE_SHADOW2D
:
1652 case TGSI_TEXTURE_SHADOWRECT
:
1654 FETCH(&r
[0], 0, CHAN_X
);
1655 FETCH(&r
[1], 0, CHAN_Y
);
1656 FETCH(&r
[2], 0, CHAN_Z
);
1658 fetch_texel(mach
->Samplers
[unit
],
1659 &r
[0], &r
[1], &r
[2], &ZeroVec
, /* inputs */
1660 tgsi_sampler_lod_bias
,
1661 &r
[0], &r
[1], &r
[2], &r
[3]); /* outputs */
1664 case TGSI_TEXTURE_3D
:
1665 case TGSI_TEXTURE_CUBE
:
1667 FETCH(&r
[0], 0, CHAN_X
);
1668 FETCH(&r
[1], 0, CHAN_Y
);
1669 FETCH(&r
[2], 0, CHAN_Z
);
1671 fetch_texel(mach
->Samplers
[unit
],
1672 &r
[0], &r
[1], &r
[2], &ZeroVec
,
1673 tgsi_sampler_lod_bias
,
1674 &r
[0], &r
[1], &r
[2], &r
[3]);
1681 FOR_EACH_ENABLED_CHANNEL(*inst
, chan_index
) {
1682 STORE(&r
[chan_index
], 0, chan_index
);
1688 * Evaluate a constant-valued coefficient at the position of the
1693 struct tgsi_exec_machine
*mach
,
1699 for( i
= 0; i
< QUAD_SIZE
; i
++ ) {
1700 mach
->Inputs
[attrib
].xyzw
[chan
].f
[i
] = mach
->InterpCoefs
[attrib
].a0
[chan
];
1705 * Evaluate a linear-valued coefficient at the position of the
1710 struct tgsi_exec_machine
*mach
,
1714 const float x
= mach
->QuadPos
.xyzw
[0].f
[0];
1715 const float y
= mach
->QuadPos
.xyzw
[1].f
[0];
1716 const float dadx
= mach
->InterpCoefs
[attrib
].dadx
[chan
];
1717 const float dady
= mach
->InterpCoefs
[attrib
].dady
[chan
];
1718 const float a0
= mach
->InterpCoefs
[attrib
].a0
[chan
] + dadx
* x
+ dady
* y
;
1719 mach
->Inputs
[attrib
].xyzw
[chan
].f
[0] = a0
;
1720 mach
->Inputs
[attrib
].xyzw
[chan
].f
[1] = a0
+ dadx
;
1721 mach
->Inputs
[attrib
].xyzw
[chan
].f
[2] = a0
+ dady
;
1722 mach
->Inputs
[attrib
].xyzw
[chan
].f
[3] = a0
+ dadx
+ dady
;
1726 * Evaluate a perspective-valued coefficient at the position of the
1730 eval_perspective_coef(
1731 struct tgsi_exec_machine
*mach
,
1735 const float x
= mach
->QuadPos
.xyzw
[0].f
[0];
1736 const float y
= mach
->QuadPos
.xyzw
[1].f
[0];
1737 const float dadx
= mach
->InterpCoefs
[attrib
].dadx
[chan
];
1738 const float dady
= mach
->InterpCoefs
[attrib
].dady
[chan
];
1739 const float a0
= mach
->InterpCoefs
[attrib
].a0
[chan
] + dadx
* x
+ dady
* y
;
1740 const float *w
= mach
->QuadPos
.xyzw
[3].f
;
1741 /* divide by W here */
1742 mach
->Inputs
[attrib
].xyzw
[chan
].f
[0] = a0
/ w
[0];
1743 mach
->Inputs
[attrib
].xyzw
[chan
].f
[1] = (a0
+ dadx
) / w
[1];
1744 mach
->Inputs
[attrib
].xyzw
[chan
].f
[2] = (a0
+ dady
) / w
[2];
1745 mach
->Inputs
[attrib
].xyzw
[chan
].f
[3] = (a0
+ dadx
+ dady
) / w
[3];
1749 typedef void (* eval_coef_func
)(
1750 struct tgsi_exec_machine
*mach
,
1755 exec_declaration(struct tgsi_exec_machine
*mach
,
1756 const struct tgsi_full_declaration
*decl
)
1758 if (mach
->Processor
== TGSI_PROCESSOR_FRAGMENT
) {
1759 if (decl
->Declaration
.File
== TGSI_FILE_INPUT
||
1760 decl
->Declaration
.File
== TGSI_FILE_SYSTEM_VALUE
) {
1761 uint first
, last
, mask
;
1763 first
= decl
->Range
.First
;
1764 last
= decl
->Range
.Last
;
1765 mask
= decl
->Declaration
.UsageMask
;
1767 if (decl
->Semantic
.Name
== TGSI_SEMANTIC_POSITION
) {
1768 assert(decl
->Semantic
.Index
== 0);
1769 assert(first
== last
);
1770 assert(mask
== TGSI_WRITEMASK_XYZW
);
1772 mach
->Inputs
[first
] = mach
->QuadPos
;
1773 } else if (decl
->Semantic
.Name
== TGSI_SEMANTIC_FACE
) {
1776 assert(decl
->Semantic
.Index
== 0);
1777 assert(first
== last
);
1779 for (i
= 0; i
< QUAD_SIZE
; i
++) {
1780 mach
->Inputs
[first
].xyzw
[0].f
[i
] = mach
->Face
;
1783 eval_coef_func eval
;
1786 switch (decl
->Declaration
.Interpolate
) {
1787 case TGSI_INTERPOLATE_CONSTANT
:
1788 eval
= eval_constant_coef
;
1791 case TGSI_INTERPOLATE_LINEAR
:
1792 eval
= eval_linear_coef
;
1795 case TGSI_INTERPOLATE_PERSPECTIVE
:
1796 eval
= eval_perspective_coef
;
1804 for (j
= 0; j
< NUM_CHANNELS
; j
++) {
1805 if (mask
& (1 << j
)) {
1806 for (i
= first
; i
<= last
; i
++) {
1816 typedef void (* micro_op
)(union tgsi_exec_channel
*dst
,
1817 const union tgsi_exec_channel
*src
);
1820 exec_scalar_unary(struct tgsi_exec_machine
*mach
,
1821 const struct tgsi_full_instruction
*inst
,
1823 enum tgsi_exec_datatype dst_datatype
,
1824 enum tgsi_exec_datatype src_datatype
)
1827 union tgsi_exec_channel src
;
1828 union tgsi_exec_channel dst
;
1830 fetch_source(mach
, &src
, &inst
->Src
[0], CHAN_X
, src_datatype
);
1832 for (chan
= 0; chan
< NUM_CHANNELS
; chan
++) {
1833 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
1834 store_dest(mach
, &dst
, &inst
->Dst
[0], inst
, chan
, dst_datatype
);
1840 exec_vector_unary(struct tgsi_exec_machine
*mach
,
1841 const struct tgsi_full_instruction
*inst
,
1843 enum tgsi_exec_datatype dst_datatype
,
1844 enum tgsi_exec_datatype src_datatype
)
1847 struct tgsi_exec_vector dst
;
1849 for (chan
= 0; chan
< NUM_CHANNELS
; chan
++) {
1850 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
1851 union tgsi_exec_channel src
;
1853 fetch_source(mach
, &src
, &inst
->Src
[0], chan
, src_datatype
);
1854 op(&dst
.xyzw
[chan
], &src
);
1857 for (chan
= 0; chan
< NUM_CHANNELS
; chan
++) {
1858 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
1859 store_dest(mach
, &dst
.xyzw
[chan
], &inst
->Dst
[0], inst
, chan
, dst_datatype
);
1865 exec_vector_binary(struct tgsi_exec_machine
*mach
,
1866 const struct tgsi_full_instruction
*inst
,
1868 enum tgsi_exec_datatype dst_datatype
,
1869 enum tgsi_exec_datatype src_datatype
)
1872 struct tgsi_exec_vector dst
;
1874 for (chan
= 0; chan
< NUM_CHANNELS
; chan
++) {
1875 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
1876 union tgsi_exec_channel src
[2];
1878 fetch_source(mach
, &src
[0], &inst
->Src
[0], chan
, src_datatype
);
1879 fetch_source(mach
, &src
[1], &inst
->Src
[1], chan
, src_datatype
);
1880 op(&dst
.xyzw
[chan
], src
);
1883 for (chan
= 0; chan
< NUM_CHANNELS
; chan
++) {
1884 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
1885 store_dest(mach
, &dst
.xyzw
[chan
], &inst
->Dst
[0], inst
, chan
, dst_datatype
);
1891 exec_vector_trinary(struct tgsi_exec_machine
*mach
,
1892 const struct tgsi_full_instruction
*inst
,
1894 enum tgsi_exec_datatype dst_datatype
,
1895 enum tgsi_exec_datatype src_datatype
)
1898 struct tgsi_exec_vector dst
;
1900 for (chan
= 0; chan
< NUM_CHANNELS
; chan
++) {
1901 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
1902 union tgsi_exec_channel src
[3];
1904 fetch_source(mach
, &src
[0], &inst
->Src
[0], chan
, src_datatype
);
1905 fetch_source(mach
, &src
[1], &inst
->Src
[1], chan
, src_datatype
);
1906 fetch_source(mach
, &src
[2], &inst
->Src
[2], chan
, src_datatype
);
1907 op(&dst
.xyzw
[chan
], src
);
1910 for (chan
= 0; chan
< NUM_CHANNELS
; chan
++) {
1911 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
1912 store_dest(mach
, &dst
.xyzw
[chan
], &inst
->Dst
[0], inst
, chan
, dst_datatype
);
1918 exec_dp3(struct tgsi_exec_machine
*mach
,
1919 const struct tgsi_full_instruction
*inst
)
1922 union tgsi_exec_channel arg
[3];
1924 fetch_source(mach
, &arg
[0], &inst
->Src
[0], CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
1925 fetch_source(mach
, &arg
[1], &inst
->Src
[1], CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
1926 micro_mul(&arg
[2], &arg
[0], &arg
[1]);
1928 for (chan
= CHAN_Y
; chan
<= CHAN_Z
; chan
++) {
1929 fetch_source(mach
, &arg
[0], &inst
->Src
[0], chan
, TGSI_EXEC_DATA_FLOAT
);
1930 fetch_source(mach
, &arg
[1], &inst
->Src
[1], chan
, TGSI_EXEC_DATA_FLOAT
);
1931 micro_mad(&arg
[2], arg
);
1934 for (chan
= 0; chan
< NUM_CHANNELS
; chan
++) {
1935 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
1936 store_dest(mach
, &arg
[2], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
1942 exec_dp4(struct tgsi_exec_machine
*mach
,
1943 const struct tgsi_full_instruction
*inst
)
1946 union tgsi_exec_channel arg
[3];
1948 fetch_source(mach
, &arg
[0], &inst
->Src
[0], CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
1949 fetch_source(mach
, &arg
[1], &inst
->Src
[1], CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
1950 micro_mul(&arg
[2], &arg
[0], &arg
[1]);
1952 for (chan
= CHAN_Y
; chan
<= CHAN_W
; chan
++) {
1953 fetch_source(mach
, &arg
[0], &inst
->Src
[0], chan
, TGSI_EXEC_DATA_FLOAT
);
1954 fetch_source(mach
, &arg
[1], &inst
->Src
[1], chan
, TGSI_EXEC_DATA_FLOAT
);
1955 micro_mad(&arg
[2], arg
);
1958 for (chan
= 0; chan
< NUM_CHANNELS
; chan
++) {
1959 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
1960 store_dest(mach
, &arg
[2], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
1966 exec_dp2a(struct tgsi_exec_machine
*mach
,
1967 const struct tgsi_full_instruction
*inst
)
1970 union tgsi_exec_channel arg
[3];
1972 fetch_source(mach
, &arg
[0], &inst
->Src
[0], CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
1973 fetch_source(mach
, &arg
[1], &inst
->Src
[1], CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
1974 micro_mul(&arg
[2], &arg
[0], &arg
[1]);
1976 fetch_source(mach
, &arg
[0], &inst
->Src
[0], CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
1977 fetch_source(mach
, &arg
[1], &inst
->Src
[1], CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
1978 micro_mad(&arg
[0], arg
);
1980 fetch_source(mach
, &arg
[1], &inst
->Src
[2], CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
1981 micro_add(&arg
[0], &arg
[0], &arg
[1]);
1983 for (chan
= 0; chan
< NUM_CHANNELS
; chan
++) {
1984 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
1985 store_dest(mach
, &arg
[0], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
1991 exec_dph(struct tgsi_exec_machine
*mach
,
1992 const struct tgsi_full_instruction
*inst
)
1995 union tgsi_exec_channel arg
[3];
1997 fetch_source(mach
, &arg
[0], &inst
->Src
[0], CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
1998 fetch_source(mach
, &arg
[1], &inst
->Src
[1], CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
1999 micro_mul(&arg
[2], &arg
[0], &arg
[1]);
2001 fetch_source(mach
, &arg
[0], &inst
->Src
[0], CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2002 fetch_source(mach
, &arg
[1], &inst
->Src
[1], CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2003 micro_mad(&arg
[2], arg
);
2005 fetch_source(mach
, &arg
[0], &inst
->Src
[0], CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
2006 fetch_source(mach
, &arg
[1], &inst
->Src
[1], CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
2007 micro_mad(&arg
[0], arg
);
2009 fetch_source(mach
, &arg
[1], &inst
->Src
[1], CHAN_W
, TGSI_EXEC_DATA_FLOAT
);
2010 micro_add(&arg
[0], &arg
[0], &arg
[1]);
2012 for (chan
= 0; chan
< NUM_CHANNELS
; chan
++) {
2013 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2014 store_dest(mach
, &arg
[0], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2020 exec_dp2(struct tgsi_exec_machine
*mach
,
2021 const struct tgsi_full_instruction
*inst
)
2024 union tgsi_exec_channel arg
[3];
2026 fetch_source(mach
, &arg
[0], &inst
->Src
[0], CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2027 fetch_source(mach
, &arg
[1], &inst
->Src
[1], CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2028 micro_mul(&arg
[2], &arg
[0], &arg
[1]);
2030 fetch_source(mach
, &arg
[0], &inst
->Src
[0], CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2031 fetch_source(mach
, &arg
[1], &inst
->Src
[1], CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2032 micro_mad(&arg
[2], arg
);
2034 for (chan
= 0; chan
< NUM_CHANNELS
; chan
++) {
2035 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2036 store_dest(mach
, &arg
[2], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2042 exec_break(struct tgsi_exec_machine
*mach
)
2044 if (mach
->BreakType
== TGSI_EXEC_BREAK_INSIDE_LOOP
) {
2045 /* turn off loop channels for each enabled exec channel */
2046 mach
->LoopMask
&= ~mach
->ExecMask
;
2047 /* Todo: if mach->LoopMask == 0, jump to end of loop */
2048 UPDATE_EXEC_MASK(mach
);
2050 assert(mach
->BreakType
== TGSI_EXEC_BREAK_INSIDE_SWITCH
);
2052 mach
->Switch
.mask
= 0x0;
2054 UPDATE_EXEC_MASK(mach
);
2059 exec_switch(struct tgsi_exec_machine
*mach
,
2060 const struct tgsi_full_instruction
*inst
)
2062 assert(mach
->SwitchStackTop
< TGSI_EXEC_MAX_SWITCH_NESTING
);
2063 assert(mach
->BreakStackTop
< TGSI_EXEC_MAX_BREAK_STACK
);
2065 mach
->SwitchStack
[mach
->SwitchStackTop
++] = mach
->Switch
;
2066 fetch_source(mach
, &mach
->Switch
.selector
, &inst
->Src
[0], CHAN_X
, TGSI_EXEC_DATA_UINT
);
2067 mach
->Switch
.mask
= 0x0;
2068 mach
->Switch
.defaultMask
= 0x0;
2070 mach
->BreakStack
[mach
->BreakStackTop
++] = mach
->BreakType
;
2071 mach
->BreakType
= TGSI_EXEC_BREAK_INSIDE_SWITCH
;
2073 UPDATE_EXEC_MASK(mach
);
2077 exec_case(struct tgsi_exec_machine
*mach
,
2078 const struct tgsi_full_instruction
*inst
)
2080 uint prevMask
= mach
->SwitchStack
[mach
->SwitchStackTop
- 1].mask
;
2081 union tgsi_exec_channel src
;
2084 fetch_source(mach
, &src
, &inst
->Src
[0], CHAN_X
, TGSI_EXEC_DATA_UINT
);
2086 if (mach
->Switch
.selector
.u
[0] == src
.u
[0]) {
2089 if (mach
->Switch
.selector
.u
[1] == src
.u
[1]) {
2092 if (mach
->Switch
.selector
.u
[2] == src
.u
[2]) {
2095 if (mach
->Switch
.selector
.u
[3] == src
.u
[3]) {
2099 mach
->Switch
.defaultMask
|= mask
;
2101 mach
->Switch
.mask
|= mask
& prevMask
;
2103 UPDATE_EXEC_MASK(mach
);
2107 exec_default(struct tgsi_exec_machine
*mach
)
2109 uint prevMask
= mach
->SwitchStack
[mach
->SwitchStackTop
- 1].mask
;
2111 mach
->Switch
.mask
|= ~mach
->Switch
.defaultMask
& prevMask
;
2113 UPDATE_EXEC_MASK(mach
);
2117 exec_endswitch(struct tgsi_exec_machine
*mach
)
2119 mach
->Switch
= mach
->SwitchStack
[--mach
->SwitchStackTop
];
2120 mach
->BreakType
= mach
->BreakStack
[--mach
->BreakStackTop
];
2122 UPDATE_EXEC_MASK(mach
);
2126 micro_i2f(union tgsi_exec_channel
*dst
,
2127 const union tgsi_exec_channel
*src
)
2129 dst
->f
[0] = (float)src
->i
[0];
2130 dst
->f
[1] = (float)src
->i
[1];
2131 dst
->f
[2] = (float)src
->i
[2];
2132 dst
->f
[3] = (float)src
->i
[3];
2136 micro_not(union tgsi_exec_channel
*dst
,
2137 const union tgsi_exec_channel
*src
)
2139 dst
->u
[0] = ~src
->u
[0];
2140 dst
->u
[1] = ~src
->u
[1];
2141 dst
->u
[2] = ~src
->u
[2];
2142 dst
->u
[3] = ~src
->u
[3];
2146 micro_shl(union tgsi_exec_channel
*dst
,
2147 const union tgsi_exec_channel
*src
)
2149 dst
->u
[0] = src
[0].u
[0] << src
[1].u
[0];
2150 dst
->u
[1] = src
[0].u
[1] << src
[1].u
[1];
2151 dst
->u
[2] = src
[0].u
[2] << src
[1].u
[2];
2152 dst
->u
[3] = src
[0].u
[3] << src
[1].u
[3];
2156 micro_and(union tgsi_exec_channel
*dst
,
2157 const union tgsi_exec_channel
*src
)
2159 dst
->u
[0] = src
[0].u
[0] & src
[1].u
[0];
2160 dst
->u
[1] = src
[0].u
[1] & src
[1].u
[1];
2161 dst
->u
[2] = src
[0].u
[2] & src
[1].u
[2];
2162 dst
->u
[3] = src
[0].u
[3] & src
[1].u
[3];
2166 micro_or(union tgsi_exec_channel
*dst
,
2167 const union tgsi_exec_channel
*src
)
2169 dst
->u
[0] = src
[0].u
[0] | src
[1].u
[0];
2170 dst
->u
[1] = src
[0].u
[1] | src
[1].u
[1];
2171 dst
->u
[2] = src
[0].u
[2] | src
[1].u
[2];
2172 dst
->u
[3] = src
[0].u
[3] | src
[1].u
[3];
2176 micro_xor(union tgsi_exec_channel
*dst
,
2177 const union tgsi_exec_channel
*src
)
2179 dst
->u
[0] = src
[0].u
[0] ^ src
[1].u
[0];
2180 dst
->u
[1] = src
[0].u
[1] ^ src
[1].u
[1];
2181 dst
->u
[2] = src
[0].u
[2] ^ src
[1].u
[2];
2182 dst
->u
[3] = src
[0].u
[3] ^ src
[1].u
[3];
2186 micro_f2i(union tgsi_exec_channel
*dst
,
2187 const union tgsi_exec_channel
*src
)
2189 dst
->i
[0] = (int)src
->f
[0];
2190 dst
->i
[1] = (int)src
->f
[1];
2191 dst
->i
[2] = (int)src
->f
[2];
2192 dst
->i
[3] = (int)src
->f
[3];
2196 micro_idiv(union tgsi_exec_channel
*dst
,
2197 const union tgsi_exec_channel
*src
)
2199 dst
->i
[0] = src
[0].i
[0] / src
[1].i
[0];
2200 dst
->i
[1] = src
[0].i
[1] / src
[1].i
[1];
2201 dst
->i
[2] = src
[0].i
[2] / src
[1].i
[2];
2202 dst
->i
[3] = src
[0].i
[3] / src
[1].i
[3];
2206 micro_imax(union tgsi_exec_channel
*dst
,
2207 const union tgsi_exec_channel
*src
)
2209 dst
->i
[0] = src
[0].i
[0] > src
[1].i
[0] ? src
[0].i
[0] : src
[1].i
[0];
2210 dst
->i
[1] = src
[0].i
[1] > src
[1].i
[1] ? src
[0].i
[1] : src
[1].i
[1];
2211 dst
->i
[2] = src
[0].i
[2] > src
[1].i
[2] ? src
[0].i
[2] : src
[1].i
[2];
2212 dst
->i
[3] = src
[0].i
[3] > src
[1].i
[3] ? src
[0].i
[3] : src
[1].i
[3];
2216 micro_imin(union tgsi_exec_channel
*dst
,
2217 const union tgsi_exec_channel
*src
)
2219 dst
->i
[0] = src
[0].i
[0] < src
[1].i
[0] ? src
[0].i
[0] : src
[1].i
[0];
2220 dst
->i
[1] = src
[0].i
[1] < src
[1].i
[1] ? src
[0].i
[1] : src
[1].i
[1];
2221 dst
->i
[2] = src
[0].i
[2] < src
[1].i
[2] ? src
[0].i
[2] : src
[1].i
[2];
2222 dst
->i
[3] = src
[0].i
[3] < src
[1].i
[3] ? src
[0].i
[3] : src
[1].i
[3];
2226 micro_isge(union tgsi_exec_channel
*dst
,
2227 const union tgsi_exec_channel
*src
)
2229 dst
->i
[0] = src
[0].i
[0] >= src
[1].i
[0] ? -1 : 0;
2230 dst
->i
[1] = src
[0].i
[1] >= src
[1].i
[1] ? -1 : 0;
2231 dst
->i
[2] = src
[0].i
[2] >= src
[1].i
[2] ? -1 : 0;
2232 dst
->i
[3] = src
[0].i
[3] >= src
[1].i
[3] ? -1 : 0;
2236 micro_ishr(union tgsi_exec_channel
*dst
,
2237 const union tgsi_exec_channel
*src
)
2239 dst
->i
[0] = src
[0].i
[0] >> src
[1].i
[0];
2240 dst
->i
[1] = src
[0].i
[1] >> src
[1].i
[1];
2241 dst
->i
[2] = src
[0].i
[2] >> src
[1].i
[2];
2242 dst
->i
[3] = src
[0].i
[3] >> src
[1].i
[3];
2246 micro_islt(union tgsi_exec_channel
*dst
,
2247 const union tgsi_exec_channel
*src
)
2249 dst
->i
[0] = src
[0].i
[0] < src
[1].i
[0] ? -1 : 0;
2250 dst
->i
[1] = src
[0].i
[1] < src
[1].i
[1] ? -1 : 0;
2251 dst
->i
[2] = src
[0].i
[2] < src
[1].i
[2] ? -1 : 0;
2252 dst
->i
[3] = src
[0].i
[3] < src
[1].i
[3] ? -1 : 0;
2256 micro_f2u(union tgsi_exec_channel
*dst
,
2257 const union tgsi_exec_channel
*src
)
2259 dst
->u
[0] = (uint
)src
->f
[0];
2260 dst
->u
[1] = (uint
)src
->f
[1];
2261 dst
->u
[2] = (uint
)src
->f
[2];
2262 dst
->u
[3] = (uint
)src
->f
[3];
2266 micro_u2f(union tgsi_exec_channel
*dst
,
2267 const union tgsi_exec_channel
*src
)
2269 dst
->f
[0] = (float)src
->u
[0];
2270 dst
->f
[1] = (float)src
->u
[1];
2271 dst
->f
[2] = (float)src
->u
[2];
2272 dst
->f
[3] = (float)src
->u
[3];
2276 micro_uadd(union tgsi_exec_channel
*dst
,
2277 const union tgsi_exec_channel
*src
)
2279 dst
->u
[0] = src
[0].u
[0] + src
[1].u
[0];
2280 dst
->u
[1] = src
[0].u
[1] + src
[1].u
[1];
2281 dst
->u
[2] = src
[0].u
[2] + src
[1].u
[2];
2282 dst
->u
[3] = src
[0].u
[3] + src
[1].u
[3];
2286 micro_udiv(union tgsi_exec_channel
*dst
,
2287 const union tgsi_exec_channel
*src
)
2289 dst
->u
[0] = src
[0].u
[0] / src
[1].u
[0];
2290 dst
->u
[1] = src
[0].u
[1] / src
[1].u
[1];
2291 dst
->u
[2] = src
[0].u
[2] / src
[1].u
[2];
2292 dst
->u
[3] = src
[0].u
[3] / src
[1].u
[3];
2296 micro_umad(union tgsi_exec_channel
*dst
,
2297 const union tgsi_exec_channel
*src
)
2299 dst
->u
[0] = src
[0].u
[0] * src
[1].u
[0] + src
[2].u
[0];
2300 dst
->u
[1] = src
[0].u
[1] * src
[1].u
[1] + src
[2].u
[1];
2301 dst
->u
[2] = src
[0].u
[2] * src
[1].u
[2] + src
[2].u
[2];
2302 dst
->u
[3] = src
[0].u
[3] * src
[1].u
[3] + src
[2].u
[3];
2306 micro_umax(union tgsi_exec_channel
*dst
,
2307 const union tgsi_exec_channel
*src
)
2309 dst
->u
[0] = src
[0].u
[0] > src
[1].u
[0] ? src
[0].u
[0] : src
[1].u
[0];
2310 dst
->u
[1] = src
[0].u
[1] > src
[1].u
[1] ? src
[0].u
[1] : src
[1].u
[1];
2311 dst
->u
[2] = src
[0].u
[2] > src
[1].u
[2] ? src
[0].u
[2] : src
[1].u
[2];
2312 dst
->u
[3] = src
[0].u
[3] > src
[1].u
[3] ? src
[0].u
[3] : src
[1].u
[3];
2316 micro_umin(union tgsi_exec_channel
*dst
,
2317 const union tgsi_exec_channel
*src
)
2319 dst
->u
[0] = src
[0].u
[0] < src
[1].u
[0] ? src
[0].u
[0] : src
[1].u
[0];
2320 dst
->u
[1] = src
[0].u
[1] < src
[1].u
[1] ? src
[0].u
[1] : src
[1].u
[1];
2321 dst
->u
[2] = src
[0].u
[2] < src
[1].u
[2] ? src
[0].u
[2] : src
[1].u
[2];
2322 dst
->u
[3] = src
[0].u
[3] < src
[1].u
[3] ? src
[0].u
[3] : src
[1].u
[3];
2326 micro_umod(union tgsi_exec_channel
*dst
,
2327 const union tgsi_exec_channel
*src
)
2329 dst
->u
[0] = src
[0].u
[0] % src
[1].u
[0];
2330 dst
->u
[1] = src
[0].u
[1] % src
[1].u
[1];
2331 dst
->u
[2] = src
[0].u
[2] % src
[1].u
[2];
2332 dst
->u
[3] = src
[0].u
[3] % src
[1].u
[3];
2336 micro_umul(union tgsi_exec_channel
*dst
,
2337 const union tgsi_exec_channel
*src
)
2339 dst
->u
[0] = src
[0].u
[0] * src
[1].u
[0];
2340 dst
->u
[1] = src
[0].u
[1] * src
[1].u
[1];
2341 dst
->u
[2] = src
[0].u
[2] * src
[1].u
[2];
2342 dst
->u
[3] = src
[0].u
[3] * src
[1].u
[3];
2346 micro_useq(union tgsi_exec_channel
*dst
,
2347 const union tgsi_exec_channel
*src
)
2349 dst
->u
[0] = src
[0].u
[0] == src
[1].u
[0] ? ~0 : 0;
2350 dst
->u
[1] = src
[0].u
[1] == src
[1].u
[1] ? ~0 : 0;
2351 dst
->u
[2] = src
[0].u
[2] == src
[1].u
[2] ? ~0 : 0;
2352 dst
->u
[3] = src
[0].u
[3] == src
[1].u
[3] ? ~0 : 0;
2356 micro_usge(union tgsi_exec_channel
*dst
,
2357 const union tgsi_exec_channel
*src
)
2359 dst
->u
[0] = src
[0].u
[0] >= src
[1].u
[0] ? ~0 : 0;
2360 dst
->u
[1] = src
[0].u
[1] >= src
[1].u
[1] ? ~0 : 0;
2361 dst
->u
[2] = src
[0].u
[2] >= src
[1].u
[2] ? ~0 : 0;
2362 dst
->u
[3] = src
[0].u
[3] >= src
[1].u
[3] ? ~0 : 0;
2366 micro_ushr(union tgsi_exec_channel
*dst
,
2367 const union tgsi_exec_channel
*src
)
2369 dst
->u
[0] = src
[0].u
[0] >> src
[1].u
[0];
2370 dst
->u
[1] = src
[0].u
[1] >> src
[1].u
[1];
2371 dst
->u
[2] = src
[0].u
[2] >> src
[1].u
[2];
2372 dst
->u
[3] = src
[0].u
[3] >> src
[1].u
[3];
2376 micro_uslt(union tgsi_exec_channel
*dst
,
2377 const union tgsi_exec_channel
*src
)
2379 dst
->u
[0] = src
[0].u
[0] < src
[1].u
[0] ? ~0 : 0;
2380 dst
->u
[1] = src
[0].u
[1] < src
[1].u
[1] ? ~0 : 0;
2381 dst
->u
[2] = src
[0].u
[2] < src
[1].u
[2] ? ~0 : 0;
2382 dst
->u
[3] = src
[0].u
[3] < src
[1].u
[3] ? ~0 : 0;
2386 micro_usne(union tgsi_exec_channel
*dst
,
2387 const union tgsi_exec_channel
*src
)
2389 dst
->u
[0] = src
[0].u
[0] != src
[1].u
[0] ? ~0 : 0;
2390 dst
->u
[1] = src
[0].u
[1] != src
[1].u
[1] ? ~0 : 0;
2391 dst
->u
[2] = src
[0].u
[2] != src
[1].u
[2] ? ~0 : 0;
2392 dst
->u
[3] = src
[0].u
[3] != src
[1].u
[3] ? ~0 : 0;
2397 struct tgsi_exec_machine
*mach
,
2398 const struct tgsi_full_instruction
*inst
,
2402 union tgsi_exec_channel r
[10];
2403 union tgsi_exec_channel d
[8];
2407 switch (inst
->Instruction
.Opcode
) {
2408 case TGSI_OPCODE_ARL
:
2409 exec_vector_unary(mach
, inst
, micro_arl
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_FLOAT
);
2412 case TGSI_OPCODE_MOV
:
2413 exec_vector_unary(mach
, inst
, micro_mov
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_FLOAT
);
2416 case TGSI_OPCODE_LIT
:
2417 if (IS_CHANNEL_ENABLED( *inst
, CHAN_Y
) || IS_CHANNEL_ENABLED( *inst
, CHAN_Z
)) {
2418 FETCH( &r
[0], 0, CHAN_X
);
2419 if (IS_CHANNEL_ENABLED( *inst
, CHAN_Y
)) {
2420 micro_max(&d
[CHAN_Y
], &r
[0], &mach
->Temps
[TEMP_0_I
].xyzw
[TEMP_0_C
]);
2423 if (IS_CHANNEL_ENABLED( *inst
, CHAN_Z
)) {
2424 FETCH( &r
[1], 0, CHAN_Y
);
2425 micro_max( &r
[1], &r
[1], &mach
->Temps
[TEMP_0_I
].xyzw
[TEMP_0_C
] );
2427 FETCH( &r
[2], 0, CHAN_W
);
2428 micro_min( &r
[2], &r
[2], &mach
->Temps
[TEMP_128_I
].xyzw
[TEMP_128_C
] );
2429 micro_max( &r
[2], &r
[2], &mach
->Temps
[TEMP_M128_I
].xyzw
[TEMP_M128_C
] );
2430 micro_pow( &r
[1], &r
[1], &r
[2] );
2431 micro_lt(&d
[CHAN_Z
], &mach
->Temps
[TEMP_0_I
].xyzw
[TEMP_0_C
], &r
[0], &r
[1], &mach
->Temps
[TEMP_0_I
].xyzw
[TEMP_0_C
]);
2434 if (IS_CHANNEL_ENABLED(*inst
, CHAN_Y
)) {
2435 STORE(&d
[CHAN_Y
], 0, CHAN_Y
);
2437 if (IS_CHANNEL_ENABLED(*inst
, CHAN_Z
)) {
2438 STORE(&d
[CHAN_Z
], 0, CHAN_Z
);
2441 if (IS_CHANNEL_ENABLED( *inst
, CHAN_X
)) {
2442 STORE( &mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], 0, CHAN_X
);
2444 if (IS_CHANNEL_ENABLED( *inst
, CHAN_W
)) {
2445 STORE( &mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], 0, CHAN_W
);
2449 case TGSI_OPCODE_RCP
:
2450 exec_scalar_unary(mach
, inst
, micro_rcp
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2453 case TGSI_OPCODE_RSQ
:
2454 exec_scalar_unary(mach
, inst
, micro_rsq
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2457 case TGSI_OPCODE_EXP
:
2458 FETCH( &r
[0], 0, CHAN_X
);
2459 micro_flr( &r
[1], &r
[0] ); /* r1 = floor(r0) */
2460 if (IS_CHANNEL_ENABLED( *inst
, CHAN_X
)) {
2461 micro_exp2( &r
[2], &r
[1] ); /* r2 = 2 ^ r1 */
2462 STORE( &r
[2], 0, CHAN_X
); /* store r2 */
2464 if (IS_CHANNEL_ENABLED( *inst
, CHAN_Y
)) {
2465 micro_sub( &r
[2], &r
[0], &r
[1] ); /* r2 = r0 - r1 */
2466 STORE( &r
[2], 0, CHAN_Y
); /* store r2 */
2468 if (IS_CHANNEL_ENABLED( *inst
, CHAN_Z
)) {
2469 micro_exp2( &r
[2], &r
[0] ); /* r2 = 2 ^ r0 */
2470 STORE( &r
[2], 0, CHAN_Z
); /* store r2 */
2472 if (IS_CHANNEL_ENABLED( *inst
, CHAN_W
)) {
2473 STORE( &mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], 0, CHAN_W
);
2477 case TGSI_OPCODE_LOG
:
2478 FETCH( &r
[0], 0, CHAN_X
);
2479 micro_abs( &r
[2], &r
[0] ); /* r2 = abs(r0) */
2480 micro_lg2( &r
[1], &r
[2] ); /* r1 = lg2(r2) */
2481 micro_flr( &r
[0], &r
[1] ); /* r0 = floor(r1) */
2482 if (IS_CHANNEL_ENABLED( *inst
, CHAN_X
)) {
2483 STORE( &r
[0], 0, CHAN_X
);
2485 if (IS_CHANNEL_ENABLED( *inst
, CHAN_Y
)) {
2486 micro_exp2( &r
[0], &r
[0] ); /* r0 = 2 ^ r0 */
2487 micro_div( &r
[0], &r
[2], &r
[0] ); /* r0 = r2 / r0 */
2488 STORE( &r
[0], 0, CHAN_Y
);
2490 if (IS_CHANNEL_ENABLED( *inst
, CHAN_Z
)) {
2491 STORE( &r
[1], 0, CHAN_Z
);
2493 if (IS_CHANNEL_ENABLED( *inst
, CHAN_W
)) {
2494 STORE( &mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], 0, CHAN_W
);
2498 case TGSI_OPCODE_MUL
:
2499 FOR_EACH_ENABLED_CHANNEL(*inst
, chan_index
) {
2500 FETCH(&r
[0], 0, chan_index
);
2501 FETCH(&r
[1], 1, chan_index
);
2502 micro_mul(&d
[chan_index
], &r
[0], &r
[1]);
2504 FOR_EACH_ENABLED_CHANNEL(*inst
, chan_index
) {
2505 STORE(&d
[chan_index
], 0, chan_index
);
2509 case TGSI_OPCODE_ADD
:
2510 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2511 FETCH( &r
[0], 0, chan_index
);
2512 FETCH( &r
[1], 1, chan_index
);
2513 micro_add(&d
[chan_index
], &r
[0], &r
[1]);
2515 FOR_EACH_ENABLED_CHANNEL(*inst
, chan_index
) {
2516 STORE(&d
[chan_index
], 0, chan_index
);
2520 case TGSI_OPCODE_DP3
:
2521 exec_dp3(mach
, inst
);
2524 case TGSI_OPCODE_DP4
:
2525 exec_dp4(mach
, inst
);
2528 case TGSI_OPCODE_DST
:
2529 if (IS_CHANNEL_ENABLED( *inst
, CHAN_Y
)) {
2530 FETCH( &r
[0], 0, CHAN_Y
);
2531 FETCH( &r
[1], 1, CHAN_Y
);
2532 micro_mul(&d
[CHAN_Y
], &r
[0], &r
[1]);
2534 if (IS_CHANNEL_ENABLED( *inst
, CHAN_Z
)) {
2535 FETCH(&d
[CHAN_Z
], 0, CHAN_Z
);
2537 if (IS_CHANNEL_ENABLED( *inst
, CHAN_W
)) {
2538 FETCH(&d
[CHAN_W
], 1, CHAN_W
);
2541 if (IS_CHANNEL_ENABLED(*inst
, CHAN_X
)) {
2542 STORE(&mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], 0, CHAN_X
);
2544 if (IS_CHANNEL_ENABLED(*inst
, CHAN_Y
)) {
2545 STORE(&d
[CHAN_Y
], 0, CHAN_Y
);
2547 if (IS_CHANNEL_ENABLED(*inst
, CHAN_Z
)) {
2548 STORE(&d
[CHAN_Z
], 0, CHAN_Z
);
2550 if (IS_CHANNEL_ENABLED(*inst
, CHAN_W
)) {
2551 STORE(&d
[CHAN_W
], 0, CHAN_W
);
2555 case TGSI_OPCODE_MIN
:
2556 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2557 FETCH(&r
[0], 0, chan_index
);
2558 FETCH(&r
[1], 1, chan_index
);
2560 /* XXX use micro_min()?? */
2561 micro_lt(&d
[chan_index
], &r
[0], &r
[1], &r
[0], &r
[1]);
2563 FOR_EACH_ENABLED_CHANNEL(*inst
, chan_index
) {
2564 STORE(&d
[chan_index
], 0, chan_index
);
2568 case TGSI_OPCODE_MAX
:
2569 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2570 FETCH(&r
[0], 0, chan_index
);
2571 FETCH(&r
[1], 1, chan_index
);
2573 /* XXX use micro_max()?? */
2574 micro_lt(&d
[chan_index
], &r
[0], &r
[1], &r
[1], &r
[0] );
2576 FOR_EACH_ENABLED_CHANNEL(*inst
, chan_index
) {
2577 STORE(&d
[chan_index
], 0, chan_index
);
2581 case TGSI_OPCODE_SLT
:
2582 exec_vector_binary(mach
, inst
, micro_slt
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2585 case TGSI_OPCODE_SGE
:
2586 exec_vector_binary(mach
, inst
, micro_sge
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2589 case TGSI_OPCODE_MAD
:
2590 exec_vector_trinary(mach
, inst
, micro_mad
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2593 case TGSI_OPCODE_SUB
:
2594 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2595 FETCH(&r
[0], 0, chan_index
);
2596 FETCH(&r
[1], 1, chan_index
);
2597 micro_sub(&d
[chan_index
], &r
[0], &r
[1]);
2599 FOR_EACH_ENABLED_CHANNEL(*inst
, chan_index
) {
2600 STORE(&d
[chan_index
], 0, chan_index
);
2604 case TGSI_OPCODE_LRP
:
2605 exec_vector_trinary(mach
, inst
, micro_lrp
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2608 case TGSI_OPCODE_CND
:
2609 FOR_EACH_ENABLED_CHANNEL(*inst
, chan_index
) {
2610 FETCH(&r
[0], 0, chan_index
);
2611 FETCH(&r
[1], 1, chan_index
);
2612 FETCH(&r
[2], 2, chan_index
);
2613 micro_lt(&d
[chan_index
], &mach
->Temps
[TEMP_HALF_I
].xyzw
[TEMP_HALF_C
], &r
[2], &r
[0], &r
[1]);
2615 FOR_EACH_ENABLED_CHANNEL(*inst
, chan_index
) {
2616 STORE(&d
[chan_index
], 0, chan_index
);
2620 case TGSI_OPCODE_DP2A
:
2621 exec_dp2a(mach
, inst
);
2624 case TGSI_OPCODE_FRC
:
2625 exec_vector_unary(mach
, inst
, micro_frc
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2628 case TGSI_OPCODE_CLAMP
:
2629 FOR_EACH_ENABLED_CHANNEL(*inst
, chan_index
) {
2630 FETCH(&r
[0], 0, chan_index
);
2631 FETCH(&r
[1], 1, chan_index
);
2632 micro_max(&r
[0], &r
[0], &r
[1]);
2633 FETCH(&r
[1], 2, chan_index
);
2634 micro_min(&d
[chan_index
], &r
[0], &r
[1]);
2636 FOR_EACH_ENABLED_CHANNEL(*inst
, chan_index
) {
2637 STORE(&d
[chan_index
], 0, chan_index
);
2641 case TGSI_OPCODE_FLR
:
2642 exec_vector_unary(mach
, inst
, micro_flr
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2645 case TGSI_OPCODE_ROUND
:
2646 exec_vector_unary(mach
, inst
, micro_rnd
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2649 case TGSI_OPCODE_EX2
:
2650 exec_scalar_unary(mach
, inst
, micro_exp2
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2653 case TGSI_OPCODE_LG2
:
2654 exec_scalar_unary(mach
, inst
, micro_lg2
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2657 case TGSI_OPCODE_POW
:
2658 FETCH(&r
[0], 0, CHAN_X
);
2659 FETCH(&r
[1], 1, CHAN_X
);
2661 micro_pow( &r
[0], &r
[0], &r
[1] );
2663 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2664 STORE( &r
[0], 0, chan_index
);
2668 case TGSI_OPCODE_XPD
:
2669 FETCH(&r
[0], 0, CHAN_Y
);
2670 FETCH(&r
[1], 1, CHAN_Z
);
2672 micro_mul( &r
[2], &r
[0], &r
[1] );
2674 FETCH(&r
[3], 0, CHAN_Z
);
2675 FETCH(&r
[4], 1, CHAN_Y
);
2677 micro_mul( &r
[5], &r
[3], &r
[4] );
2678 micro_sub(&d
[CHAN_X
], &r
[2], &r
[5]);
2680 FETCH(&r
[2], 1, CHAN_X
);
2682 micro_mul( &r
[3], &r
[3], &r
[2] );
2684 FETCH(&r
[5], 0, CHAN_X
);
2686 micro_mul( &r
[1], &r
[1], &r
[5] );
2687 micro_sub(&d
[CHAN_Y
], &r
[3], &r
[1]);
2689 micro_mul( &r
[5], &r
[5], &r
[4] );
2690 micro_mul( &r
[0], &r
[0], &r
[2] );
2691 micro_sub(&d
[CHAN_Z
], &r
[5], &r
[0]);
2693 if (IS_CHANNEL_ENABLED(*inst
, CHAN_X
)) {
2694 STORE(&d
[CHAN_X
], 0, CHAN_X
);
2696 if (IS_CHANNEL_ENABLED(*inst
, CHAN_Y
)) {
2697 STORE(&d
[CHAN_Y
], 0, CHAN_Y
);
2699 if (IS_CHANNEL_ENABLED(*inst
, CHAN_Z
)) {
2700 STORE(&d
[CHAN_Z
], 0, CHAN_Z
);
2702 if (IS_CHANNEL_ENABLED( *inst
, CHAN_W
)) {
2703 STORE( &mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], 0, CHAN_W
);
2707 case TGSI_OPCODE_ABS
:
2708 exec_vector_unary(mach
, inst
, micro_abs
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2711 case TGSI_OPCODE_RCC
:
2712 FETCH(&r
[0], 0, CHAN_X
);
2713 micro_div(&r
[0], &mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], &r
[0]);
2714 micro_float_clamp(&r
[0], &r
[0]);
2715 FOR_EACH_ENABLED_CHANNEL(*inst
, chan_index
) {
2716 STORE(&r
[0], 0, chan_index
);
2720 case TGSI_OPCODE_DPH
:
2721 exec_dph(mach
, inst
);
2724 case TGSI_OPCODE_COS
:
2725 exec_scalar_unary(mach
, inst
, micro_cos
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2728 case TGSI_OPCODE_DDX
:
2729 exec_vector_unary(mach
, inst
, micro_ddx
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2732 case TGSI_OPCODE_DDY
:
2733 exec_vector_unary(mach
, inst
, micro_ddy
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2736 case TGSI_OPCODE_KILP
:
2737 exec_kilp (mach
, inst
);
2740 case TGSI_OPCODE_KIL
:
2741 exec_kil (mach
, inst
);
2744 case TGSI_OPCODE_PK2H
:
2748 case TGSI_OPCODE_PK2US
:
2752 case TGSI_OPCODE_PK4B
:
2756 case TGSI_OPCODE_PK4UB
:
2760 case TGSI_OPCODE_RFL
:
2761 if (IS_CHANNEL_ENABLED(*inst
, CHAN_X
) ||
2762 IS_CHANNEL_ENABLED(*inst
, CHAN_Y
) ||
2763 IS_CHANNEL_ENABLED(*inst
, CHAN_Z
)) {
2764 /* r0 = dp3(src0, src0) */
2765 FETCH(&r
[2], 0, CHAN_X
);
2766 micro_mul(&r
[0], &r
[2], &r
[2]);
2767 FETCH(&r
[4], 0, CHAN_Y
);
2768 micro_mul(&r
[8], &r
[4], &r
[4]);
2769 micro_add(&r
[0], &r
[0], &r
[8]);
2770 FETCH(&r
[6], 0, CHAN_Z
);
2771 micro_mul(&r
[8], &r
[6], &r
[6]);
2772 micro_add(&r
[0], &r
[0], &r
[8]);
2774 /* r1 = dp3(src0, src1) */
2775 FETCH(&r
[3], 1, CHAN_X
);
2776 micro_mul(&r
[1], &r
[2], &r
[3]);
2777 FETCH(&r
[5], 1, CHAN_Y
);
2778 micro_mul(&r
[8], &r
[4], &r
[5]);
2779 micro_add(&r
[1], &r
[1], &r
[8]);
2780 FETCH(&r
[7], 1, CHAN_Z
);
2781 micro_mul(&r
[8], &r
[6], &r
[7]);
2782 micro_add(&r
[1], &r
[1], &r
[8]);
2784 /* r1 = 2 * r1 / r0 */
2785 micro_add(&r
[1], &r
[1], &r
[1]);
2786 micro_div(&r
[1], &r
[1], &r
[0]);
2788 if (IS_CHANNEL_ENABLED(*inst
, CHAN_X
)) {
2789 micro_mul(&r
[2], &r
[2], &r
[1]);
2790 micro_sub(&r
[2], &r
[2], &r
[3]);
2791 STORE(&r
[2], 0, CHAN_X
);
2793 if (IS_CHANNEL_ENABLED(*inst
, CHAN_Y
)) {
2794 micro_mul(&r
[4], &r
[4], &r
[1]);
2795 micro_sub(&r
[4], &r
[4], &r
[5]);
2796 STORE(&r
[4], 0, CHAN_Y
);
2798 if (IS_CHANNEL_ENABLED(*inst
, CHAN_Z
)) {
2799 micro_mul(&r
[6], &r
[6], &r
[1]);
2800 micro_sub(&r
[6], &r
[6], &r
[7]);
2801 STORE(&r
[6], 0, CHAN_Z
);
2804 if (IS_CHANNEL_ENABLED(*inst
, CHAN_W
)) {
2805 STORE(&mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], 0, CHAN_W
);
2809 case TGSI_OPCODE_SEQ
:
2810 exec_vector_binary(mach
, inst
, micro_seq
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2813 case TGSI_OPCODE_SFL
:
2814 FOR_EACH_ENABLED_CHANNEL(*inst
, chan_index
) {
2815 STORE(&mach
->Temps
[TEMP_0_I
].xyzw
[TEMP_0_C
], 0, chan_index
);
2819 case TGSI_OPCODE_SGT
:
2820 exec_vector_binary(mach
, inst
, micro_sgt
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2823 case TGSI_OPCODE_SIN
:
2824 exec_scalar_unary(mach
, inst
, micro_sin
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2827 case TGSI_OPCODE_SLE
:
2828 exec_vector_binary(mach
, inst
, micro_sle
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2831 case TGSI_OPCODE_SNE
:
2832 exec_vector_binary(mach
, inst
, micro_sne
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2835 case TGSI_OPCODE_STR
:
2836 FOR_EACH_ENABLED_CHANNEL(*inst
, chan_index
) {
2837 STORE(&mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], 0, chan_index
);
2841 case TGSI_OPCODE_TEX
:
2842 /* simple texture lookup */
2843 /* src[0] = texcoord */
2844 /* src[1] = sampler unit */
2845 exec_tex(mach
, inst
, TEX_MODIFIER_NONE
);
2848 case TGSI_OPCODE_TXB
:
2849 /* Texture lookup with lod bias */
2850 /* src[0] = texcoord (src[0].w = LOD bias) */
2851 /* src[1] = sampler unit */
2852 exec_tex(mach
, inst
, TEX_MODIFIER_LOD_BIAS
);
2855 case TGSI_OPCODE_TXD
:
2856 /* Texture lookup with explict partial derivatives */
2857 /* src[0] = texcoord */
2858 /* src[1] = d[strq]/dx */
2859 /* src[2] = d[strq]/dy */
2860 /* src[3] = sampler unit */
2861 exec_txd(mach
, inst
);
2864 case TGSI_OPCODE_TXL
:
2865 /* Texture lookup with explit LOD */
2866 /* src[0] = texcoord (src[0].w = LOD) */
2867 /* src[1] = sampler unit */
2868 exec_tex(mach
, inst
, TEX_MODIFIER_EXPLICIT_LOD
);
2871 case TGSI_OPCODE_TXP
:
2872 /* Texture lookup with projection */
2873 /* src[0] = texcoord (src[0].w = projection) */
2874 /* src[1] = sampler unit */
2875 exec_tex(mach
, inst
, TEX_MODIFIER_PROJECTED
);
2878 case TGSI_OPCODE_UP2H
:
2882 case TGSI_OPCODE_UP2US
:
2886 case TGSI_OPCODE_UP4B
:
2890 case TGSI_OPCODE_UP4UB
:
2894 case TGSI_OPCODE_X2D
:
2895 FETCH(&r
[0], 1, CHAN_X
);
2896 FETCH(&r
[1], 1, CHAN_Y
);
2897 if (IS_CHANNEL_ENABLED(*inst
, CHAN_X
) ||
2898 IS_CHANNEL_ENABLED(*inst
, CHAN_Z
)) {
2899 FETCH(&r
[2], 2, CHAN_X
);
2900 micro_mul(&r
[2], &r
[2], &r
[0]);
2901 FETCH(&r
[3], 2, CHAN_Y
);
2902 micro_mul(&r
[3], &r
[3], &r
[1]);
2903 micro_add(&r
[2], &r
[2], &r
[3]);
2904 FETCH(&r
[3], 0, CHAN_X
);
2905 micro_add(&d
[CHAN_X
], &r
[2], &r
[3]);
2908 if (IS_CHANNEL_ENABLED(*inst
, CHAN_Y
) ||
2909 IS_CHANNEL_ENABLED(*inst
, CHAN_W
)) {
2910 FETCH(&r
[2], 2, CHAN_Z
);
2911 micro_mul(&r
[2], &r
[2], &r
[0]);
2912 FETCH(&r
[3], 2, CHAN_W
);
2913 micro_mul(&r
[3], &r
[3], &r
[1]);
2914 micro_add(&r
[2], &r
[2], &r
[3]);
2915 FETCH(&r
[3], 0, CHAN_Y
);
2916 micro_add(&d
[CHAN_Y
], &r
[2], &r
[3]);
2919 if (IS_CHANNEL_ENABLED(*inst
, CHAN_X
)) {
2920 STORE(&d
[CHAN_X
], 0, CHAN_X
);
2922 if (IS_CHANNEL_ENABLED(*inst
, CHAN_Y
)) {
2923 STORE(&d
[CHAN_Y
], 0, CHAN_Y
);
2925 if (IS_CHANNEL_ENABLED(*inst
, CHAN_Z
)) {
2926 STORE(&d
[CHAN_X
], 0, CHAN_Z
);
2928 if (IS_CHANNEL_ENABLED(*inst
, CHAN_W
)) {
2929 STORE(&d
[CHAN_Y
], 0, CHAN_W
);
2933 case TGSI_OPCODE_ARA
:
2937 case TGSI_OPCODE_ARR
:
2938 exec_vector_unary(mach
, inst
, micro_arr
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_FLOAT
);
2941 case TGSI_OPCODE_BRA
:
2945 case TGSI_OPCODE_CAL
:
2946 /* skip the call if no execution channels are enabled */
2947 if (mach
->ExecMask
) {
2950 /* First, record the depths of the execution stacks.
2951 * This is important for deeply nested/looped return statements.
2952 * We have to unwind the stacks by the correct amount. For a
2953 * real code generator, we could determine the number of entries
2954 * to pop off each stack with simple static analysis and avoid
2955 * implementing this data structure at run time.
2957 mach
->CallStack
[mach
->CallStackTop
].CondStackTop
= mach
->CondStackTop
;
2958 mach
->CallStack
[mach
->CallStackTop
].LoopStackTop
= mach
->LoopStackTop
;
2959 mach
->CallStack
[mach
->CallStackTop
].ContStackTop
= mach
->ContStackTop
;
2960 mach
->CallStack
[mach
->CallStackTop
].SwitchStackTop
= mach
->SwitchStackTop
;
2961 mach
->CallStack
[mach
->CallStackTop
].BreakStackTop
= mach
->BreakStackTop
;
2962 /* note that PC was already incremented above */
2963 mach
->CallStack
[mach
->CallStackTop
].ReturnAddr
= *pc
;
2965 mach
->CallStackTop
++;
2967 /* Second, push the Cond, Loop, Cont, Func stacks */
2968 assert(mach
->CondStackTop
< TGSI_EXEC_MAX_COND_NESTING
);
2969 assert(mach
->LoopStackTop
< TGSI_EXEC_MAX_LOOP_NESTING
);
2970 assert(mach
->ContStackTop
< TGSI_EXEC_MAX_LOOP_NESTING
);
2971 assert(mach
->SwitchStackTop
< TGSI_EXEC_MAX_SWITCH_NESTING
);
2972 assert(mach
->BreakStackTop
< TGSI_EXEC_MAX_BREAK_STACK
);
2973 assert(mach
->FuncStackTop
< TGSI_EXEC_MAX_CALL_NESTING
);
2975 mach
->CondStack
[mach
->CondStackTop
++] = mach
->CondMask
;
2976 mach
->LoopStack
[mach
->LoopStackTop
++] = mach
->LoopMask
;
2977 mach
->ContStack
[mach
->ContStackTop
++] = mach
->ContMask
;
2978 mach
->SwitchStack
[mach
->SwitchStackTop
++] = mach
->Switch
;
2979 mach
->BreakStack
[mach
->BreakStackTop
++] = mach
->BreakType
;
2980 mach
->FuncStack
[mach
->FuncStackTop
++] = mach
->FuncMask
;
2982 /* Finally, jump to the subroutine */
2983 *pc
= inst
->Label
.Label
;
2987 case TGSI_OPCODE_RET
:
2988 mach
->FuncMask
&= ~mach
->ExecMask
;
2989 UPDATE_EXEC_MASK(mach
);
2991 if (mach
->FuncMask
== 0x0) {
2992 /* really return now (otherwise, keep executing */
2994 if (mach
->CallStackTop
== 0) {
2995 /* returning from main() */
3000 assert(mach
->CallStackTop
> 0);
3001 mach
->CallStackTop
--;
3003 mach
->CondStackTop
= mach
->CallStack
[mach
->CallStackTop
].CondStackTop
;
3004 mach
->CondMask
= mach
->CondStack
[mach
->CondStackTop
];
3006 mach
->LoopStackTop
= mach
->CallStack
[mach
->CallStackTop
].LoopStackTop
;
3007 mach
->LoopMask
= mach
->LoopStack
[mach
->LoopStackTop
];
3009 mach
->ContStackTop
= mach
->CallStack
[mach
->CallStackTop
].ContStackTop
;
3010 mach
->ContMask
= mach
->ContStack
[mach
->ContStackTop
];
3012 mach
->SwitchStackTop
= mach
->CallStack
[mach
->CallStackTop
].SwitchStackTop
;
3013 mach
->Switch
= mach
->SwitchStack
[mach
->SwitchStackTop
];
3015 mach
->BreakStackTop
= mach
->CallStack
[mach
->CallStackTop
].BreakStackTop
;
3016 mach
->BreakType
= mach
->BreakStack
[mach
->BreakStackTop
];
3018 assert(mach
->FuncStackTop
> 0);
3019 mach
->FuncMask
= mach
->FuncStack
[--mach
->FuncStackTop
];
3021 *pc
= mach
->CallStack
[mach
->CallStackTop
].ReturnAddr
;
3023 UPDATE_EXEC_MASK(mach
);
3027 case TGSI_OPCODE_SSG
:
3028 exec_vector_unary(mach
, inst
, micro_sgn
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3031 case TGSI_OPCODE_CMP
:
3032 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
3033 FETCH(&r
[0], 0, chan_index
);
3034 FETCH(&r
[1], 1, chan_index
);
3035 FETCH(&r
[2], 2, chan_index
);
3036 micro_lt(&d
[chan_index
], &r
[0], &mach
->Temps
[TEMP_0_I
].xyzw
[TEMP_0_C
], &r
[1], &r
[2]);
3038 FOR_EACH_ENABLED_CHANNEL(*inst
, chan_index
) {
3039 STORE(&d
[chan_index
], 0, chan_index
);
3043 case TGSI_OPCODE_SCS
:
3044 if( IS_CHANNEL_ENABLED( *inst
, CHAN_X
) || IS_CHANNEL_ENABLED( *inst
, CHAN_Y
) ) {
3045 FETCH( &r
[0], 0, CHAN_X
);
3046 if (IS_CHANNEL_ENABLED(*inst
, CHAN_X
)) {
3047 micro_cos(&r
[1], &r
[0]);
3048 STORE(&r
[1], 0, CHAN_X
);
3050 if (IS_CHANNEL_ENABLED(*inst
, CHAN_Y
)) {
3051 micro_sin(&r
[1], &r
[0]);
3052 STORE(&r
[1], 0, CHAN_Y
);
3055 if( IS_CHANNEL_ENABLED( *inst
, CHAN_Z
) ) {
3056 STORE( &mach
->Temps
[TEMP_0_I
].xyzw
[TEMP_0_C
], 0, CHAN_Z
);
3058 if( IS_CHANNEL_ENABLED( *inst
, CHAN_W
) ) {
3059 STORE( &mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], 0, CHAN_W
);
3063 case TGSI_OPCODE_NRM
:
3064 /* 3-component vector normalize */
3065 if(IS_CHANNEL_ENABLED(*inst
, CHAN_X
) ||
3066 IS_CHANNEL_ENABLED(*inst
, CHAN_Y
) ||
3067 IS_CHANNEL_ENABLED(*inst
, CHAN_Z
)) {
3068 /* r3 = sqrt(dp3(src0, src0)) */
3069 FETCH(&r
[0], 0, CHAN_X
);
3070 micro_mul(&r
[3], &r
[0], &r
[0]);
3071 FETCH(&r
[1], 0, CHAN_Y
);
3072 micro_mul(&r
[4], &r
[1], &r
[1]);
3073 micro_add(&r
[3], &r
[3], &r
[4]);
3074 FETCH(&r
[2], 0, CHAN_Z
);
3075 micro_mul(&r
[4], &r
[2], &r
[2]);
3076 micro_add(&r
[3], &r
[3], &r
[4]);
3077 micro_sqrt(&r
[3], &r
[3]);
3079 if (IS_CHANNEL_ENABLED(*inst
, CHAN_X
)) {
3080 micro_div(&r
[0], &r
[0], &r
[3]);
3081 STORE(&r
[0], 0, CHAN_X
);
3083 if (IS_CHANNEL_ENABLED(*inst
, CHAN_Y
)) {
3084 micro_div(&r
[1], &r
[1], &r
[3]);
3085 STORE(&r
[1], 0, CHAN_Y
);
3087 if (IS_CHANNEL_ENABLED(*inst
, CHAN_Z
)) {
3088 micro_div(&r
[2], &r
[2], &r
[3]);
3089 STORE(&r
[2], 0, CHAN_Z
);
3092 if (IS_CHANNEL_ENABLED(*inst
, CHAN_W
)) {
3093 STORE(&mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], 0, CHAN_W
);
3097 case TGSI_OPCODE_NRM4
:
3098 /* 4-component vector normalize */
3100 union tgsi_exec_channel tmp
, dot
;
3102 /* tmp = dp4(src0, src0): */
3103 FETCH( &r
[0], 0, CHAN_X
);
3104 micro_mul( &tmp
, &r
[0], &r
[0] );
3106 FETCH( &r
[1], 0, CHAN_Y
);
3107 micro_mul( &dot
, &r
[1], &r
[1] );
3108 micro_add( &tmp
, &tmp
, &dot
);
3110 FETCH( &r
[2], 0, CHAN_Z
);
3111 micro_mul( &dot
, &r
[2], &r
[2] );
3112 micro_add( &tmp
, &tmp
, &dot
);
3114 FETCH( &r
[3], 0, CHAN_W
);
3115 micro_mul( &dot
, &r
[3], &r
[3] );
3116 micro_add( &tmp
, &tmp
, &dot
);
3118 /* tmp = 1 / sqrt(tmp) */
3119 micro_sqrt( &tmp
, &tmp
);
3120 micro_div( &tmp
, &mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], &tmp
);
3122 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
3123 /* chan = chan * tmp */
3124 micro_mul( &r
[chan_index
], &tmp
, &r
[chan_index
] );
3125 STORE( &r
[chan_index
], 0, chan_index
);
3130 case TGSI_OPCODE_DIV
:
3134 case TGSI_OPCODE_DP2
:
3135 exec_dp2(mach
, inst
);
3138 case TGSI_OPCODE_IF
:
3140 assert(mach
->CondStackTop
< TGSI_EXEC_MAX_COND_NESTING
);
3141 mach
->CondStack
[mach
->CondStackTop
++] = mach
->CondMask
;
3142 FETCH( &r
[0], 0, CHAN_X
);
3143 /* update CondMask */
3145 mach
->CondMask
&= ~0x1;
3148 mach
->CondMask
&= ~0x2;
3151 mach
->CondMask
&= ~0x4;
3154 mach
->CondMask
&= ~0x8;
3156 UPDATE_EXEC_MASK(mach
);
3157 /* Todo: If CondMask==0, jump to ELSE */
3160 case TGSI_OPCODE_ELSE
:
3161 /* invert CondMask wrt previous mask */
3164 assert(mach
->CondStackTop
> 0);
3165 prevMask
= mach
->CondStack
[mach
->CondStackTop
- 1];
3166 mach
->CondMask
= ~mach
->CondMask
& prevMask
;
3167 UPDATE_EXEC_MASK(mach
);
3168 /* Todo: If CondMask==0, jump to ENDIF */
3172 case TGSI_OPCODE_ENDIF
:
3174 assert(mach
->CondStackTop
> 0);
3175 mach
->CondMask
= mach
->CondStack
[--mach
->CondStackTop
];
3176 UPDATE_EXEC_MASK(mach
);
3179 case TGSI_OPCODE_END
:
3180 /* halt execution */
3184 case TGSI_OPCODE_REP
:
3188 case TGSI_OPCODE_ENDREP
:
3192 case TGSI_OPCODE_PUSHA
:
3196 case TGSI_OPCODE_POPA
:
3200 case TGSI_OPCODE_CEIL
:
3201 exec_vector_unary(mach
, inst
, micro_ceil
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3204 case TGSI_OPCODE_I2F
:
3205 exec_vector_unary(mach
, inst
, micro_i2f
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_INT
);
3208 case TGSI_OPCODE_NOT
:
3209 exec_vector_unary(mach
, inst
, micro_not
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
3212 case TGSI_OPCODE_TRUNC
:
3213 exec_vector_unary(mach
, inst
, micro_trunc
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3216 case TGSI_OPCODE_SHL
:
3217 exec_vector_binary(mach
, inst
, micro_shl
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
3220 case TGSI_OPCODE_AND
:
3221 exec_vector_binary(mach
, inst
, micro_and
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
3224 case TGSI_OPCODE_OR
:
3225 exec_vector_binary(mach
, inst
, micro_or
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
3228 case TGSI_OPCODE_MOD
:
3232 case TGSI_OPCODE_XOR
:
3233 exec_vector_binary(mach
, inst
, micro_xor
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
3236 case TGSI_OPCODE_SAD
:
3240 case TGSI_OPCODE_TXF
:
3244 case TGSI_OPCODE_TXQ
:
3248 case TGSI_OPCODE_EMIT
:
3252 case TGSI_OPCODE_ENDPRIM
:
3253 emit_primitive(mach
);
3256 case TGSI_OPCODE_BGNFOR
:
3257 assert(mach
->LoopCounterStackTop
< TGSI_EXEC_MAX_LOOP_NESTING
);
3258 for (chan_index
= 0; chan_index
< 3; chan_index
++) {
3259 FETCH( &mach
->LoopCounterStack
[mach
->LoopCounterStackTop
].xyzw
[chan_index
], 0, chan_index
);
3261 ++mach
->LoopCounterStackTop
;
3262 STORE(&mach
->LoopCounterStack
[mach
->LoopCounterStackTop
- 1].xyzw
[CHAN_X
], 0, CHAN_X
);
3263 /* update LoopMask */
3264 if (mach
->LoopCounterStack
[mach
->LoopCounterStackTop
- 1].xyzw
[CHAN_Y
].f
[0] <= 0.0f
) {
3265 mach
->LoopMask
&= ~0x1;
3267 if (mach
->LoopCounterStack
[mach
->LoopCounterStackTop
- 1].xyzw
[CHAN_Y
].f
[1] <= 0.0f
) {
3268 mach
->LoopMask
&= ~0x2;
3270 if (mach
->LoopCounterStack
[mach
->LoopCounterStackTop
- 1].xyzw
[CHAN_Y
].f
[2] <= 0.0f
) {
3271 mach
->LoopMask
&= ~0x4;
3273 if (mach
->LoopCounterStack
[mach
->LoopCounterStackTop
- 1].xyzw
[CHAN_Y
].f
[3] <= 0.0f
) {
3274 mach
->LoopMask
&= ~0x8;
3276 /* TODO: if mach->LoopMask == 0, jump to end of loop */
3277 UPDATE_EXEC_MASK(mach
);
3278 /* fall-through (for now) */
3279 case TGSI_OPCODE_BGNLOOP
:
3280 /* push LoopMask and ContMasks */
3281 assert(mach
->LoopStackTop
< TGSI_EXEC_MAX_LOOP_NESTING
);
3282 assert(mach
->ContStackTop
< TGSI_EXEC_MAX_LOOP_NESTING
);
3283 assert(mach
->LoopLabelStackTop
< TGSI_EXEC_MAX_LOOP_NESTING
);
3284 assert(mach
->BreakStackTop
< TGSI_EXEC_MAX_BREAK_STACK
);
3286 mach
->LoopStack
[mach
->LoopStackTop
++] = mach
->LoopMask
;
3287 mach
->ContStack
[mach
->ContStackTop
++] = mach
->ContMask
;
3288 mach
->LoopLabelStack
[mach
->LoopLabelStackTop
++] = *pc
- 1;
3289 mach
->BreakStack
[mach
->BreakStackTop
++] = mach
->BreakType
;
3290 mach
->BreakType
= TGSI_EXEC_BREAK_INSIDE_LOOP
;
3293 case TGSI_OPCODE_ENDFOR
:
3294 assert(mach
->LoopCounterStackTop
> 0);
3295 micro_sub(&mach
->LoopCounterStack
[mach
->LoopCounterStackTop
- 1].xyzw
[CHAN_Y
],
3296 &mach
->LoopCounterStack
[mach
->LoopCounterStackTop
- 1].xyzw
[CHAN_Y
],
3297 &mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
]);
3298 /* update LoopMask */
3299 if (mach
->LoopCounterStack
[mach
->LoopCounterStackTop
- 1].xyzw
[CHAN_Y
].f
[0] <= 0.0f
) {
3300 mach
->LoopMask
&= ~0x1;
3302 if (mach
->LoopCounterStack
[mach
->LoopCounterStackTop
- 1].xyzw
[CHAN_Y
].f
[1] <= 0.0f
) {
3303 mach
->LoopMask
&= ~0x2;
3305 if (mach
->LoopCounterStack
[mach
->LoopCounterStackTop
- 1].xyzw
[CHAN_Y
].f
[2] <= 0.0f
) {
3306 mach
->LoopMask
&= ~0x4;
3308 if (mach
->LoopCounterStack
[mach
->LoopCounterStackTop
- 1].xyzw
[CHAN_Y
].f
[3] <= 0.0f
) {
3309 mach
->LoopMask
&= ~0x8;
3311 micro_add(&mach
->LoopCounterStack
[mach
->LoopCounterStackTop
- 1].xyzw
[CHAN_X
],
3312 &mach
->LoopCounterStack
[mach
->LoopCounterStackTop
- 1].xyzw
[CHAN_X
],
3313 &mach
->LoopCounterStack
[mach
->LoopCounterStackTop
- 1].xyzw
[CHAN_Z
]);
3314 assert(mach
->LoopLabelStackTop
> 0);
3315 inst
= mach
->Instructions
+ mach
->LoopLabelStack
[mach
->LoopLabelStackTop
- 1];
3316 STORE(&mach
->LoopCounterStack
[mach
->LoopCounterStackTop
].xyzw
[CHAN_X
], 0, CHAN_X
);
3317 /* Restore ContMask, but don't pop */
3318 assert(mach
->ContStackTop
> 0);
3319 mach
->ContMask
= mach
->ContStack
[mach
->ContStackTop
- 1];
3320 UPDATE_EXEC_MASK(mach
);
3321 if (mach
->ExecMask
) {
3322 /* repeat loop: jump to instruction just past BGNLOOP */
3323 assert(mach
->LoopLabelStackTop
> 0);
3324 *pc
= mach
->LoopLabelStack
[mach
->LoopLabelStackTop
- 1] + 1;
3327 /* exit loop: pop LoopMask */
3328 assert(mach
->LoopStackTop
> 0);
3329 mach
->LoopMask
= mach
->LoopStack
[--mach
->LoopStackTop
];
3331 assert(mach
->ContStackTop
> 0);
3332 mach
->ContMask
= mach
->ContStack
[--mach
->ContStackTop
];
3333 assert(mach
->LoopLabelStackTop
> 0);
3334 --mach
->LoopLabelStackTop
;
3335 assert(mach
->LoopCounterStackTop
> 0);
3336 --mach
->LoopCounterStackTop
;
3338 mach
->BreakType
= mach
->BreakStack
[--mach
->BreakStackTop
];
3340 UPDATE_EXEC_MASK(mach
);
3343 case TGSI_OPCODE_ENDLOOP
:
3344 /* Restore ContMask, but don't pop */
3345 assert(mach
->ContStackTop
> 0);
3346 mach
->ContMask
= mach
->ContStack
[mach
->ContStackTop
- 1];
3347 UPDATE_EXEC_MASK(mach
);
3348 if (mach
->ExecMask
) {
3349 /* repeat loop: jump to instruction just past BGNLOOP */
3350 assert(mach
->LoopLabelStackTop
> 0);
3351 *pc
= mach
->LoopLabelStack
[mach
->LoopLabelStackTop
- 1] + 1;
3354 /* exit loop: pop LoopMask */
3355 assert(mach
->LoopStackTop
> 0);
3356 mach
->LoopMask
= mach
->LoopStack
[--mach
->LoopStackTop
];
3358 assert(mach
->ContStackTop
> 0);
3359 mach
->ContMask
= mach
->ContStack
[--mach
->ContStackTop
];
3360 assert(mach
->LoopLabelStackTop
> 0);
3361 --mach
->LoopLabelStackTop
;
3363 mach
->BreakType
= mach
->BreakStack
[--mach
->BreakStackTop
];
3365 UPDATE_EXEC_MASK(mach
);
3368 case TGSI_OPCODE_BRK
:
3372 case TGSI_OPCODE_CONT
:
3373 /* turn off cont channels for each enabled exec channel */
3374 mach
->ContMask
&= ~mach
->ExecMask
;
3375 /* Todo: if mach->LoopMask == 0, jump to end of loop */
3376 UPDATE_EXEC_MASK(mach
);
3379 case TGSI_OPCODE_BGNSUB
:
3383 case TGSI_OPCODE_ENDSUB
:
3385 * XXX: This really should be a no-op. We should never reach this opcode.
3388 assert(mach
->CallStackTop
> 0);
3389 mach
->CallStackTop
--;
3391 mach
->CondStackTop
= mach
->CallStack
[mach
->CallStackTop
].CondStackTop
;
3392 mach
->CondMask
= mach
->CondStack
[mach
->CondStackTop
];
3394 mach
->LoopStackTop
= mach
->CallStack
[mach
->CallStackTop
].LoopStackTop
;
3395 mach
->LoopMask
= mach
->LoopStack
[mach
->LoopStackTop
];
3397 mach
->ContStackTop
= mach
->CallStack
[mach
->CallStackTop
].ContStackTop
;
3398 mach
->ContMask
= mach
->ContStack
[mach
->ContStackTop
];
3400 mach
->SwitchStackTop
= mach
->CallStack
[mach
->CallStackTop
].SwitchStackTop
;
3401 mach
->Switch
= mach
->SwitchStack
[mach
->SwitchStackTop
];
3403 mach
->BreakStackTop
= mach
->CallStack
[mach
->CallStackTop
].BreakStackTop
;
3404 mach
->BreakType
= mach
->BreakStack
[mach
->BreakStackTop
];
3406 assert(mach
->FuncStackTop
> 0);
3407 mach
->FuncMask
= mach
->FuncStack
[--mach
->FuncStackTop
];
3409 *pc
= mach
->CallStack
[mach
->CallStackTop
].ReturnAddr
;
3411 UPDATE_EXEC_MASK(mach
);
3414 case TGSI_OPCODE_NOP
:
3417 case TGSI_OPCODE_BREAKC
:
3418 FETCH(&r
[0], 0, CHAN_X
);
3419 /* update CondMask */
3420 if (r
[0].u
[0] && (mach
->ExecMask
& 0x1)) {
3421 mach
->LoopMask
&= ~0x1;
3423 if (r
[0].u
[1] && (mach
->ExecMask
& 0x2)) {
3424 mach
->LoopMask
&= ~0x2;
3426 if (r
[0].u
[2] && (mach
->ExecMask
& 0x4)) {
3427 mach
->LoopMask
&= ~0x4;
3429 if (r
[0].u
[3] && (mach
->ExecMask
& 0x8)) {
3430 mach
->LoopMask
&= ~0x8;
3432 /* Todo: if mach->LoopMask == 0, jump to end of loop */
3433 UPDATE_EXEC_MASK(mach
);
3436 case TGSI_OPCODE_F2I
:
3437 exec_vector_unary(mach
, inst
, micro_f2i
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_FLOAT
);
3440 case TGSI_OPCODE_IDIV
:
3441 exec_vector_binary(mach
, inst
, micro_idiv
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
3444 case TGSI_OPCODE_IMAX
:
3445 exec_vector_binary(mach
, inst
, micro_imax
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
3448 case TGSI_OPCODE_IMIN
:
3449 exec_vector_binary(mach
, inst
, micro_imin
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
3452 case TGSI_OPCODE_INEG
:
3453 exec_vector_unary(mach
, inst
, micro_ineg
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
3456 case TGSI_OPCODE_ISGE
:
3457 exec_vector_binary(mach
, inst
, micro_isge
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
3460 case TGSI_OPCODE_ISHR
:
3461 exec_vector_binary(mach
, inst
, micro_ishr
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
3464 case TGSI_OPCODE_ISLT
:
3465 exec_vector_binary(mach
, inst
, micro_islt
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
3468 case TGSI_OPCODE_F2U
:
3469 exec_vector_unary(mach
, inst
, micro_f2u
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_FLOAT
);
3472 case TGSI_OPCODE_U2F
:
3473 exec_vector_unary(mach
, inst
, micro_u2f
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_UINT
);
3476 case TGSI_OPCODE_UADD
:
3477 exec_vector_binary(mach
, inst
, micro_uadd
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
3480 case TGSI_OPCODE_UDIV
:
3481 exec_vector_binary(mach
, inst
, micro_udiv
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
3484 case TGSI_OPCODE_UMAD
:
3485 exec_vector_trinary(mach
, inst
, micro_umad
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
3488 case TGSI_OPCODE_UMAX
:
3489 exec_vector_binary(mach
, inst
, micro_umax
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
3492 case TGSI_OPCODE_UMIN
:
3493 exec_vector_binary(mach
, inst
, micro_umin
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
3496 case TGSI_OPCODE_UMOD
:
3497 exec_vector_binary(mach
, inst
, micro_umod
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
3500 case TGSI_OPCODE_UMUL
:
3501 exec_vector_binary(mach
, inst
, micro_umul
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
3504 case TGSI_OPCODE_USEQ
:
3505 exec_vector_binary(mach
, inst
, micro_useq
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
3508 case TGSI_OPCODE_USGE
:
3509 exec_vector_binary(mach
, inst
, micro_usge
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
3512 case TGSI_OPCODE_USHR
:
3513 exec_vector_binary(mach
, inst
, micro_ushr
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
3516 case TGSI_OPCODE_USLT
:
3517 exec_vector_binary(mach
, inst
, micro_uslt
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
3520 case TGSI_OPCODE_USNE
:
3521 exec_vector_binary(mach
, inst
, micro_usne
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
3524 case TGSI_OPCODE_SWITCH
:
3525 exec_switch(mach
, inst
);
3528 case TGSI_OPCODE_CASE
:
3529 exec_case(mach
, inst
);
3532 case TGSI_OPCODE_DEFAULT
:
3536 case TGSI_OPCODE_ENDSWITCH
:
3537 exec_endswitch(mach
);
3546 #define DEBUG_EXECUTION 0
3550 * Run TGSI interpreter.
3551 * \return bitmask of "alive" quad components
3554 tgsi_exec_machine_run( struct tgsi_exec_machine
*mach
)
3559 mach
->CondMask
= 0xf;
3560 mach
->LoopMask
= 0xf;
3561 mach
->ContMask
= 0xf;
3562 mach
->FuncMask
= 0xf;
3563 mach
->ExecMask
= 0xf;
3565 mach
->Switch
.mask
= 0xf;
3567 assert(mach
->CondStackTop
== 0);
3568 assert(mach
->LoopStackTop
== 0);
3569 assert(mach
->ContStackTop
== 0);
3570 assert(mach
->SwitchStackTop
== 0);
3571 assert(mach
->BreakStackTop
== 0);
3572 assert(mach
->CallStackTop
== 0);
3574 mach
->Temps
[TEMP_KILMASK_I
].xyzw
[TEMP_KILMASK_C
].u
[0] = 0;
3575 mach
->Temps
[TEMP_OUTPUT_I
].xyzw
[TEMP_OUTPUT_C
].u
[0] = 0;
3577 if( mach
->Processor
== TGSI_PROCESSOR_GEOMETRY
) {
3578 mach
->Temps
[TEMP_PRIMITIVE_I
].xyzw
[TEMP_PRIMITIVE_C
].u
[0] = 0;
3579 mach
->Primitives
[0] = 0;
3582 for (i
= 0; i
< QUAD_SIZE
; i
++) {
3583 mach
->Temps
[TEMP_CC_I
].xyzw
[TEMP_CC_C
].u
[i
] =
3584 (TGSI_EXEC_CC_EQ
<< TGSI_EXEC_CC_X_SHIFT
) |
3585 (TGSI_EXEC_CC_EQ
<< TGSI_EXEC_CC_Y_SHIFT
) |
3586 (TGSI_EXEC_CC_EQ
<< TGSI_EXEC_CC_Z_SHIFT
) |
3587 (TGSI_EXEC_CC_EQ
<< TGSI_EXEC_CC_W_SHIFT
);
3590 /* execute declarations (interpolants) */
3591 for (i
= 0; i
< mach
->NumDeclarations
; i
++) {
3592 exec_declaration( mach
, mach
->Declarations
+i
);
3597 struct tgsi_exec_vector temps
[TGSI_EXEC_NUM_TEMPS
+ TGSI_EXEC_NUM_TEMP_EXTRAS
];
3598 struct tgsi_exec_vector outputs
[PIPE_MAX_ATTRIBS
];
3601 memcpy(temps
, mach
->Temps
, sizeof(temps
));
3602 memcpy(outputs
, mach
->Outputs
, sizeof(outputs
));
3605 /* execute instructions, until pc is set to -1 */
3611 tgsi_dump_instruction(&mach
->Instructions
[pc
], inst
++);
3614 assert(pc
< (int) mach
->NumInstructions
);
3615 exec_instruction(mach
, mach
->Instructions
+ pc
, &pc
);
3618 for (i
= 0; i
< TGSI_EXEC_NUM_TEMPS
+ TGSI_EXEC_NUM_TEMP_EXTRAS
; i
++) {
3619 if (memcmp(&temps
[i
], &mach
->Temps
[i
], sizeof(temps
[i
]))) {
3622 memcpy(&temps
[i
], &mach
->Temps
[i
], sizeof(temps
[i
]));
3623 debug_printf("TEMP[%2u] = ", i
);
3624 for (j
= 0; j
< 4; j
++) {
3628 debug_printf("(%6f %u, %6f %u, %6f %u, %6f %u)\n",
3629 temps
[i
].xyzw
[0].f
[j
], temps
[i
].xyzw
[0].u
[j
],
3630 temps
[i
].xyzw
[1].f
[j
], temps
[i
].xyzw
[1].u
[j
],
3631 temps
[i
].xyzw
[2].f
[j
], temps
[i
].xyzw
[2].u
[j
],
3632 temps
[i
].xyzw
[3].f
[j
], temps
[i
].xyzw
[3].u
[j
]);
3636 for (i
= 0; i
< PIPE_MAX_ATTRIBS
; i
++) {
3637 if (memcmp(&outputs
[i
], &mach
->Outputs
[i
], sizeof(outputs
[i
]))) {
3640 memcpy(&outputs
[i
], &mach
->Outputs
[i
], sizeof(outputs
[i
]));
3641 debug_printf("OUT[%2u] = ", i
);
3642 for (j
= 0; j
< 4; j
++) {
3646 debug_printf("(%6f %u, %6f %u, %6f %u, %6f %u)\n",
3647 outputs
[i
].xyzw
[0].f
[j
], outputs
[i
].xyzw
[0].u
[j
],
3648 outputs
[i
].xyzw
[1].f
[j
], outputs
[i
].xyzw
[1].u
[j
],
3649 outputs
[i
].xyzw
[2].f
[j
], outputs
[i
].xyzw
[2].u
[j
],
3650 outputs
[i
].xyzw
[3].f
[j
], outputs
[i
].xyzw
[3].u
[j
]);
3659 /* we scale from floats in [0,1] to Zbuffer ints in sp_quad_depth_test.c */
3660 if (mach
->Processor
== TGSI_PROCESSOR_FRAGMENT
) {
3662 * Scale back depth component.
3664 for (i
= 0; i
< 4; i
++)
3665 mach
->Outputs
[0].xyzw
[2].f
[i
] *= ctx
->DrawBuffer
->_DepthMaxF
;
3669 assert(mach
->CondStackTop
== 0);
3670 assert(mach
->LoopStackTop
== 0);
3671 assert(mach
->ContStackTop
== 0);
3672 assert(mach
->SwitchStackTop
== 0);
3673 assert(mach
->BreakStackTop
== 0);
3674 assert(mach
->CallStackTop
== 0);
3676 return ~mach
->Temps
[TEMP_KILMASK_I
].xyzw
[TEMP_KILMASK_C
].u
[0];