2 * Mesa 3-D graphics library
5 * Copyright (C) 1999-2006 Brian Paul 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 "Software"),
9 * to deal in the Software without restriction, including without limitation
10 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
11 * and/or sell copies of the Software, and to permit persons to whom the
12 * Software is furnished to do so, subject to the following conditions:
14 * The above copyright notice and this permission notice shall be included
15 * in all copies or substantial portions of the Software.
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
18 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
21 * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
22 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 * Regarding GL_NV_fragment_program:
28 * Portions of this software may use or implement intellectual
29 * property owned and licensed by NVIDIA Corporation. NVIDIA disclaims
30 * any and all warranties with respect to such intellectual property,
31 * including any use thereof or modifications thereto.
37 #include "program_instruction.h"
40 #include "s_nvfragprog.h"
44 /* if 1, print some debugging info */
49 * Virtual machine state used during execution of a fragment programs.
53 GLfloat Temporaries
[MAX_NV_FRAGMENT_PROGRAM_TEMPS
][4];
54 GLfloat Inputs
[MAX_NV_FRAGMENT_PROGRAM_INPUTS
][4];
55 GLfloat Outputs
[MAX_NV_FRAGMENT_PROGRAM_OUTPUTS
][4];
60 #if FEATURE_MESA_program_debug
61 static struct fp_machine
*CurrentMachine
= NULL
;
64 * For GL_MESA_program_debug.
65 * Return current value (4*GLfloat) of a fragment program register.
66 * Called via ctx->Driver.GetFragmentProgramRegister().
69 _swrast_get_program_register(GLcontext
*ctx
, enum register_file file
,
70 GLuint index
, GLfloat val
[4])
75 COPY_4V(val
, CurrentMachine
->Inputs
[index
]);
78 COPY_4V(val
, CurrentMachine
->Outputs
[index
]);
80 case PROGRAM_TEMPORARY
:
81 COPY_4V(val
, CurrentMachine
->Temporaries
[index
]);
85 "bad register file in _swrast_get_program_register");
89 #endif /* FEATURE_MESA_program_debug */
96 fetch_texel( GLcontext
*ctx
, const GLfloat texcoord
[4], GLfloat lambda
,
97 GLuint unit
, GLfloat color
[4] )
100 SWcontext
*swrast
= SWRAST_CONTEXT(ctx
);
102 /* XXX use a float-valued TextureSample routine here!!! */
103 swrast
->TextureSample
[unit
](ctx
, ctx
->Texture
.Unit
[unit
]._Current
,
104 1, (const GLfloat (*)[4]) texcoord
,
106 color
[0] = CHAN_TO_FLOAT(rgba
[0]);
107 color
[1] = CHAN_TO_FLOAT(rgba
[1]);
108 color
[2] = CHAN_TO_FLOAT(rgba
[2]);
109 color
[3] = CHAN_TO_FLOAT(rgba
[3]);
114 * Fetch a texel with the given partial derivatives to compute a level
115 * of detail in the mipmap.
118 fetch_texel_deriv( GLcontext
*ctx
, const GLfloat texcoord
[4],
119 const GLfloat texdx
[4], const GLfloat texdy
[4],
120 GLuint unit
, GLfloat color
[4] )
122 SWcontext
*swrast
= SWRAST_CONTEXT(ctx
);
123 const struct gl_texture_object
*texObj
= ctx
->Texture
.Unit
[unit
]._Current
;
124 const struct gl_texture_image
*texImg
= texObj
->Image
[0][texObj
->BaseLevel
];
125 const GLfloat texW
= (GLfloat
) texImg
->WidthScale
;
126 const GLfloat texH
= (GLfloat
) texImg
->HeightScale
;
129 GLfloat lambda
= _swrast_compute_lambda(texdx
[0], texdy
[0], /* ds/dx, ds/dy */
130 texdx
[1], texdy
[1], /* dt/dx, dt/dy */
131 texdx
[3], texdy
[2], /* dq/dx, dq/dy */
133 texcoord
[0], texcoord
[1], texcoord
[3],
136 swrast
->TextureSample
[unit
](ctx
, ctx
->Texture
.Unit
[unit
]._Current
,
137 1, (const GLfloat (*)[4]) texcoord
,
139 color
[0] = CHAN_TO_FLOAT(rgba
[0]);
140 color
[1] = CHAN_TO_FLOAT(rgba
[1]);
141 color
[2] = CHAN_TO_FLOAT(rgba
[2]);
142 color
[3] = CHAN_TO_FLOAT(rgba
[3]);
147 * Return a pointer to the 4-element float vector specified by the given
150 static INLINE
const GLfloat
*
151 get_register_pointer( GLcontext
*ctx
,
152 const struct prog_src_register
*source
,
153 const struct fp_machine
*machine
,
154 const struct gl_fragment_program
*program
)
157 switch (source
->File
) {
158 case PROGRAM_TEMPORARY
:
159 ASSERT(source
->Index
< MAX_NV_FRAGMENT_PROGRAM_TEMPS
);
160 src
= machine
->Temporaries
[source
->Index
];
163 ASSERT(source
->Index
< MAX_NV_FRAGMENT_PROGRAM_INPUTS
);
164 src
= machine
->Inputs
[source
->Index
];
167 /* This is only for PRINT */
168 ASSERT(source
->Index
< MAX_NV_FRAGMENT_PROGRAM_OUTPUTS
);
169 src
= machine
->Outputs
[source
->Index
];
171 case PROGRAM_LOCAL_PARAM
:
172 ASSERT(source
->Index
< MAX_PROGRAM_LOCAL_PARAMS
);
173 src
= program
->Base
.LocalParams
[source
->Index
];
175 case PROGRAM_ENV_PARAM
:
176 ASSERT(source
->Index
< MAX_NV_FRAGMENT_PROGRAM_PARAMS
);
177 src
= ctx
->FragmentProgram
.Parameters
[source
->Index
];
179 case PROGRAM_STATE_VAR
:
181 case PROGRAM_NAMED_PARAM
:
182 ASSERT(source
->Index
< (GLint
) program
->Base
.Parameters
->NumParameters
);
183 src
= program
->Base
.Parameters
->ParameterValues
[source
->Index
];
186 _mesa_problem(ctx
, "Invalid input register file %d in fetch_vector4", source
->File
);
194 * Fetch a 4-element float vector from the given source register.
195 * Apply swizzling and negating as needed.
198 fetch_vector4( GLcontext
*ctx
,
199 const struct prog_src_register
*source
,
200 const struct fp_machine
*machine
,
201 const struct gl_fragment_program
*program
,
204 const GLfloat
*src
= get_register_pointer(ctx
, source
, machine
, program
);
207 result
[0] = src
[GET_SWZ(source
->Swizzle
, 0)];
208 result
[1] = src
[GET_SWZ(source
->Swizzle
, 1)];
209 result
[2] = src
[GET_SWZ(source
->Swizzle
, 2)];
210 result
[3] = src
[GET_SWZ(source
->Swizzle
, 3)];
212 if (source
->NegateBase
) {
213 result
[0] = -result
[0];
214 result
[1] = -result
[1];
215 result
[2] = -result
[2];
216 result
[3] = -result
[3];
219 result
[0] = FABSF(result
[0]);
220 result
[1] = FABSF(result
[1]);
221 result
[2] = FABSF(result
[2]);
222 result
[3] = FABSF(result
[3]);
224 if (source
->NegateAbs
) {
225 result
[0] = -result
[0];
226 result
[1] = -result
[1];
227 result
[2] = -result
[2];
228 result
[3] = -result
[3];
234 * Fetch the derivative with respect to X for the given register.
235 * \return GL_TRUE if it was easily computed or GL_FALSE if we
236 * need to execute another instance of the program (ugh)!
239 fetch_vector4_deriv( GLcontext
*ctx
,
240 const struct prog_src_register
*source
,
242 char xOrY
, GLint column
, GLfloat result
[4] )
246 ASSERT(xOrY
== 'X' || xOrY
== 'Y');
248 switch (source
->Index
) {
249 case FRAG_ATTRIB_WPOS
:
253 src
[2] = span
->dzdx
/ ctx
->DrawBuffer
->_DepthMaxF
;
259 src
[2] = span
->dzdy
/ ctx
->DrawBuffer
->_DepthMaxF
;
263 case FRAG_ATTRIB_COL0
:
265 src
[0] = span
->drdx
* (1.0F
/ CHAN_MAXF
);
266 src
[1] = span
->dgdx
* (1.0F
/ CHAN_MAXF
);
267 src
[2] = span
->dbdx
* (1.0F
/ CHAN_MAXF
);
268 src
[3] = span
->dadx
* (1.0F
/ CHAN_MAXF
);
271 src
[0] = span
->drdy
* (1.0F
/ CHAN_MAXF
);
272 src
[1] = span
->dgdy
* (1.0F
/ CHAN_MAXF
);
273 src
[2] = span
->dbdy
* (1.0F
/ CHAN_MAXF
);
274 src
[3] = span
->dady
* (1.0F
/ CHAN_MAXF
);
277 case FRAG_ATTRIB_COL1
:
279 src
[0] = span
->dsrdx
* (1.0F
/ CHAN_MAXF
);
280 src
[1] = span
->dsgdx
* (1.0F
/ CHAN_MAXF
);
281 src
[2] = span
->dsbdx
* (1.0F
/ CHAN_MAXF
);
282 src
[3] = 0.0; /* XXX need this */
285 src
[0] = span
->dsrdy
* (1.0F
/ CHAN_MAXF
);
286 src
[1] = span
->dsgdy
* (1.0F
/ CHAN_MAXF
);
287 src
[2] = span
->dsbdy
* (1.0F
/ CHAN_MAXF
);
288 src
[3] = 0.0; /* XXX need this */
291 case FRAG_ATTRIB_FOGC
:
293 src
[0] = span
->dfogdx
;
299 src
[0] = span
->dfogdy
;
305 case FRAG_ATTRIB_TEX0
:
306 case FRAG_ATTRIB_TEX1
:
307 case FRAG_ATTRIB_TEX2
:
308 case FRAG_ATTRIB_TEX3
:
309 case FRAG_ATTRIB_TEX4
:
310 case FRAG_ATTRIB_TEX5
:
311 case FRAG_ATTRIB_TEX6
:
312 case FRAG_ATTRIB_TEX7
:
314 const GLuint u
= source
->Index
- FRAG_ATTRIB_TEX0
;
315 /* this is a little tricky - I think I've got it right */
316 const GLfloat invQ
= 1.0f
/ (span
->tex
[u
][3]
317 + span
->texStepX
[u
][3] * column
);
318 src
[0] = span
->texStepX
[u
][0] * invQ
;
319 src
[1] = span
->texStepX
[u
][1] * invQ
;
320 src
[2] = span
->texStepX
[u
][2] * invQ
;
321 src
[3] = span
->texStepX
[u
][3] * invQ
;
324 const GLuint u
= source
->Index
- FRAG_ATTRIB_TEX0
;
325 /* Tricky, as above, but in Y direction */
326 const GLfloat invQ
= 1.0f
/ (span
->tex
[u
][3] + span
->texStepY
[u
][3]);
327 src
[0] = span
->texStepY
[u
][0] * invQ
;
328 src
[1] = span
->texStepY
[u
][1] * invQ
;
329 src
[2] = span
->texStepY
[u
][2] * invQ
;
330 src
[3] = span
->texStepY
[u
][3] * invQ
;
337 result
[0] = src
[GET_SWZ(source
->Swizzle
, 0)];
338 result
[1] = src
[GET_SWZ(source
->Swizzle
, 1)];
339 result
[2] = src
[GET_SWZ(source
->Swizzle
, 2)];
340 result
[3] = src
[GET_SWZ(source
->Swizzle
, 3)];
342 if (source
->NegateBase
) {
343 result
[0] = -result
[0];
344 result
[1] = -result
[1];
345 result
[2] = -result
[2];
346 result
[3] = -result
[3];
349 result
[0] = FABSF(result
[0]);
350 result
[1] = FABSF(result
[1]);
351 result
[2] = FABSF(result
[2]);
352 result
[3] = FABSF(result
[3]);
354 if (source
->NegateAbs
) {
355 result
[0] = -result
[0];
356 result
[1] = -result
[1];
357 result
[2] = -result
[2];
358 result
[3] = -result
[3];
365 * As above, but only return result[0] element.
368 fetch_vector1( GLcontext
*ctx
,
369 const struct prog_src_register
*source
,
370 const struct fp_machine
*machine
,
371 const struct gl_fragment_program
*program
,
374 const GLfloat
*src
= get_register_pointer(ctx
, source
, machine
, program
);
377 result
[0] = src
[GET_SWZ(source
->Swizzle
, 0)];
379 if (source
->NegateBase
) {
380 result
[0] = -result
[0];
383 result
[0] = FABSF(result
[0]);
385 if (source
->NegateAbs
) {
386 result
[0] = -result
[0];
392 * Test value against zero and return GT, LT, EQ or UN if NaN.
395 generate_cc( float value
)
398 return COND_UN
; /* NaN */
408 * Test if the ccMaskRule is satisfied by the given condition code.
409 * Used to mask destination writes according to the current condition codee.
411 static INLINE GLboolean
412 test_cc(GLuint condCode
, GLuint ccMaskRule
)
414 switch (ccMaskRule
) {
415 case COND_EQ
: return (condCode
== COND_EQ
);
416 case COND_NE
: return (condCode
!= COND_EQ
);
417 case COND_LT
: return (condCode
== COND_LT
);
418 case COND_GE
: return (condCode
== COND_GT
|| condCode
== COND_EQ
);
419 case COND_LE
: return (condCode
== COND_LT
|| condCode
== COND_EQ
);
420 case COND_GT
: return (condCode
== COND_GT
);
421 case COND_TR
: return GL_TRUE
;
422 case COND_FL
: return GL_FALSE
;
423 default: return GL_TRUE
;
429 * Store 4 floats into a register. Observe the instructions saturate and
430 * set-condition-code flags.
433 store_vector4( const struct prog_instruction
*inst
,
434 struct fp_machine
*machine
,
435 const GLfloat value
[4] )
437 const struct prog_dst_register
*dest
= &(inst
->DstReg
);
438 const GLboolean clamp
= inst
->SaturateMode
== SATURATE_ZERO_ONE
;
439 const GLboolean updateCC
= inst
->CondUpdate
;
442 GLfloat clampedValue
[4];
443 GLboolean condWriteMask
[4];
444 GLuint writeMask
= dest
->WriteMask
;
446 switch (dest
->File
) {
448 dstReg
= machine
->Outputs
[dest
->Index
];
450 case PROGRAM_TEMPORARY
:
451 dstReg
= machine
->Temporaries
[dest
->Index
];
453 case PROGRAM_WRITE_ONLY
:
457 _mesa_problem(NULL
, "bad register file in store_vector4(fp)");
462 if (value
[0] > 1.0e10
||
463 IS_INF_OR_NAN(value
[0]) ||
464 IS_INF_OR_NAN(value
[1]) ||
465 IS_INF_OR_NAN(value
[2]) ||
466 IS_INF_OR_NAN(value
[3]) )
467 printf("store %g %g %g %g\n", value
[0], value
[1], value
[2], value
[3]);
471 clampedValue
[0] = CLAMP(value
[0], 0.0F
, 1.0F
);
472 clampedValue
[1] = CLAMP(value
[1], 0.0F
, 1.0F
);
473 clampedValue
[2] = CLAMP(value
[2], 0.0F
, 1.0F
);
474 clampedValue
[3] = CLAMP(value
[3], 0.0F
, 1.0F
);
475 value
= clampedValue
;
478 if (dest
->CondMask
!= COND_TR
) {
479 condWriteMask
[0] = GET_BIT(writeMask
, 0)
480 && test_cc(machine
->CondCodes
[GET_SWZ(dest
->CondSwizzle
, 0)], dest
->CondMask
);
481 condWriteMask
[1] = GET_BIT(writeMask
, 1)
482 && test_cc(machine
->CondCodes
[GET_SWZ(dest
->CondSwizzle
, 1)], dest
->CondMask
);
483 condWriteMask
[2] = GET_BIT(writeMask
, 2)
484 && test_cc(machine
->CondCodes
[GET_SWZ(dest
->CondSwizzle
, 2)], dest
->CondMask
);
485 condWriteMask
[3] = GET_BIT(writeMask
, 3)
486 && test_cc(machine
->CondCodes
[GET_SWZ(dest
->CondSwizzle
, 3)], dest
->CondMask
);
488 writeMask
= ((condWriteMask
[0] << 0) |
489 (condWriteMask
[1] << 1) |
490 (condWriteMask
[2] << 2) |
491 (condWriteMask
[3] << 3));
494 if (GET_BIT(writeMask
, 0)) {
495 dstReg
[0] = value
[0];
497 machine
->CondCodes
[0] = generate_cc(value
[0]);
499 if (GET_BIT(writeMask
, 1)) {
500 dstReg
[1] = value
[1];
502 machine
->CondCodes
[1] = generate_cc(value
[1]);
504 if (GET_BIT(writeMask
, 2)) {
505 dstReg
[2] = value
[2];
507 machine
->CondCodes
[2] = generate_cc(value
[2]);
509 if (GET_BIT(writeMask
, 3)) {
510 dstReg
[3] = value
[3];
512 machine
->CondCodes
[3] = generate_cc(value
[3]);
518 * Initialize a new machine state instance from an existing one, adding
519 * the partial derivatives onto the input registers.
520 * Used to implement DDX and DDY instructions in non-trivial cases.
523 init_machine_deriv( GLcontext
*ctx
,
524 const struct fp_machine
*machine
,
525 const struct gl_fragment_program
*program
,
526 const SWspan
*span
, char xOrY
,
527 struct fp_machine
*dMachine
)
531 ASSERT(xOrY
== 'X' || xOrY
== 'Y');
533 /* copy existing machine */
534 _mesa_memcpy(dMachine
, machine
, sizeof(struct fp_machine
));
536 if (program
->Base
.Target
== GL_FRAGMENT_PROGRAM_NV
) {
537 /* Clear temporary registers (undefined for ARB_f_p) */
538 _mesa_bzero( (void*) machine
->Temporaries
,
539 MAX_NV_FRAGMENT_PROGRAM_TEMPS
* 4 * sizeof(GLfloat
));
542 /* Add derivatives */
543 if (program
->Base
.InputsRead
& (1 << FRAG_ATTRIB_WPOS
)) {
544 GLfloat
*wpos
= (GLfloat
*) machine
->Inputs
[FRAG_ATTRIB_WPOS
];
548 wpos
[2] += span
->dzdx
;
549 wpos
[3] += span
->dwdx
;
554 wpos
[2] += span
->dzdy
;
555 wpos
[3] += span
->dwdy
;
558 if (program
->Base
.InputsRead
& (1 << FRAG_ATTRIB_COL0
)) {
559 GLfloat
*col0
= (GLfloat
*) machine
->Inputs
[FRAG_ATTRIB_COL0
];
561 col0
[0] += span
->drdx
* (1.0F
/ CHAN_MAXF
);
562 col0
[1] += span
->dgdx
* (1.0F
/ CHAN_MAXF
);
563 col0
[2] += span
->dbdx
* (1.0F
/ CHAN_MAXF
);
564 col0
[3] += span
->dadx
* (1.0F
/ CHAN_MAXF
);
567 col0
[0] += span
->drdy
* (1.0F
/ CHAN_MAXF
);
568 col0
[1] += span
->dgdy
* (1.0F
/ CHAN_MAXF
);
569 col0
[2] += span
->dbdy
* (1.0F
/ CHAN_MAXF
);
570 col0
[3] += span
->dady
* (1.0F
/ CHAN_MAXF
);
573 if (program
->Base
.InputsRead
& (1 << FRAG_ATTRIB_COL1
)) {
574 GLfloat
*col1
= (GLfloat
*) machine
->Inputs
[FRAG_ATTRIB_COL1
];
576 col1
[0] += span
->dsrdx
* (1.0F
/ CHAN_MAXF
);
577 col1
[1] += span
->dsgdx
* (1.0F
/ CHAN_MAXF
);
578 col1
[2] += span
->dsbdx
* (1.0F
/ CHAN_MAXF
);
579 col1
[3] += 0.0; /*XXX fix */
582 col1
[0] += span
->dsrdy
* (1.0F
/ CHAN_MAXF
);
583 col1
[1] += span
->dsgdy
* (1.0F
/ CHAN_MAXF
);
584 col1
[2] += span
->dsbdy
* (1.0F
/ CHAN_MAXF
);
585 col1
[3] += 0.0; /*XXX fix */
588 if (program
->Base
.InputsRead
& (1 << FRAG_ATTRIB_FOGC
)) {
589 GLfloat
*fogc
= (GLfloat
*) machine
->Inputs
[FRAG_ATTRIB_FOGC
];
591 fogc
[0] += span
->dfogdx
;
594 fogc
[0] += span
->dfogdy
;
597 for (u
= 0; u
< ctx
->Const
.MaxTextureCoordUnits
; u
++) {
598 if (program
->Base
.InputsRead
& (1 << (FRAG_ATTRIB_TEX0
+ u
))) {
599 GLfloat
*tex
= (GLfloat
*) machine
->Inputs
[FRAG_ATTRIB_TEX0
+ u
];
600 /* XXX perspective-correct interpolation */
602 tex
[0] += span
->texStepX
[u
][0];
603 tex
[1] += span
->texStepX
[u
][1];
604 tex
[2] += span
->texStepX
[u
][2];
605 tex
[3] += span
->texStepX
[u
][3];
608 tex
[0] += span
->texStepY
[u
][0];
609 tex
[1] += span
->texStepY
[u
][1];
610 tex
[2] += span
->texStepY
[u
][2];
611 tex
[3] += span
->texStepY
[u
][3];
616 /* init condition codes */
617 dMachine
->CondCodes
[0] = COND_EQ
;
618 dMachine
->CondCodes
[1] = COND_EQ
;
619 dMachine
->CondCodes
[2] = COND_EQ
;
620 dMachine
->CondCodes
[3] = COND_EQ
;
625 * Execute the given vertex program.
626 * NOTE: we do everything in single-precision floating point; we don't
627 * currently observe the single/half/fixed-precision qualifiers.
628 * \param ctx - rendering context
629 * \param program - the fragment program to execute
630 * \param machine - machine state (register file)
631 * \param maxInst - max number of instructions to execute
632 * \return GL_TRUE if program completed or GL_FALSE if program executed KIL.
635 execute_program( GLcontext
*ctx
,
636 const struct gl_fragment_program
*program
, GLuint maxInst
,
637 struct fp_machine
*machine
, const SWspan
*span
,
643 printf("execute fragment program --------------------\n");
646 for (pc
= 0; pc
< maxInst
; pc
++) {
647 const struct prog_instruction
*inst
= program
->Base
.Instructions
+ pc
;
649 if (ctx
->FragmentProgram
.CallbackEnabled
&&
650 ctx
->FragmentProgram
.Callback
) {
651 ctx
->FragmentProgram
.CurrentPosition
= inst
->StringPos
;
652 ctx
->FragmentProgram
.Callback(program
->Base
.Target
,
653 ctx
->FragmentProgram
.CallbackData
);
656 switch (inst
->Opcode
) {
659 GLfloat a
[4], result
[4];
660 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
661 result
[0] = FABSF(a
[0]);
662 result
[1] = FABSF(a
[1]);
663 result
[2] = FABSF(a
[2]);
664 result
[3] = FABSF(a
[3]);
665 store_vector4( inst
, machine
, result
);
670 GLfloat a
[4], b
[4], result
[4];
671 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
672 fetch_vector4( ctx
, &inst
->SrcReg
[1], machine
, program
, b
);
673 result
[0] = a
[0] + b
[0];
674 result
[1] = a
[1] + b
[1];
675 result
[2] = a
[2] + b
[2];
676 result
[3] = a
[3] + b
[3];
677 store_vector4( inst
, machine
, result
);
682 GLfloat a
[4], b
[4], c
[4], result
[4];
683 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
684 fetch_vector4( ctx
, &inst
->SrcReg
[1], machine
, program
, b
);
685 fetch_vector4( ctx
, &inst
->SrcReg
[2], machine
, program
, c
);
686 result
[0] = a
[0] < 0.0F
? b
[0] : c
[0];
687 result
[1] = a
[1] < 0.0F
? b
[1] : c
[1];
688 result
[2] = a
[2] < 0.0F
? b
[2] : c
[2];
689 result
[3] = a
[3] < 0.0F
? b
[3] : c
[3];
690 store_vector4( inst
, machine
, result
);
695 GLfloat a
[4], result
[4];
696 fetch_vector1( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
697 result
[0] = result
[1] = result
[2] = result
[3] = (GLfloat
)_mesa_cos(a
[0]);
698 store_vector4( inst
, machine
, result
);
701 case OPCODE_DDX
: /* Partial derivative with respect to X */
703 GLfloat a
[4], aNext
[4], result
[4];
704 struct fp_machine dMachine
;
705 if (!fetch_vector4_deriv(ctx
, &inst
->SrcReg
[0], span
, 'X',
707 /* This is tricky. Make a copy of the current machine state,
708 * increment the input registers by the dx or dy partial
709 * derivatives, then re-execute the program up to the
710 * preceeding instruction, then fetch the source register.
711 * Finally, find the difference in the register values for
712 * the original and derivative runs.
714 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
715 init_machine_deriv(ctx
, machine
, program
, span
,
717 execute_program(ctx
, program
, pc
, &dMachine
, span
, column
);
718 fetch_vector4( ctx
, &inst
->SrcReg
[0], &dMachine
, program
, aNext
);
719 result
[0] = aNext
[0] - a
[0];
720 result
[1] = aNext
[1] - a
[1];
721 result
[2] = aNext
[2] - a
[2];
722 result
[3] = aNext
[3] - a
[3];
724 store_vector4( inst
, machine
, result
);
727 case OPCODE_DDY
: /* Partial derivative with respect to Y */
729 GLfloat a
[4], aNext
[4], result
[4];
730 struct fp_machine dMachine
;
731 if (!fetch_vector4_deriv(ctx
, &inst
->SrcReg
[0], span
, 'Y',
733 init_machine_deriv(ctx
, machine
, program
, span
,
735 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
736 execute_program(ctx
, program
, pc
, &dMachine
, span
, column
);
737 fetch_vector4( ctx
, &inst
->SrcReg
[0], &dMachine
, program
, aNext
);
738 result
[0] = aNext
[0] - a
[0];
739 result
[1] = aNext
[1] - a
[1];
740 result
[2] = aNext
[2] - a
[2];
741 result
[3] = aNext
[3] - a
[3];
743 store_vector4( inst
, machine
, result
);
748 GLfloat a
[4], b
[4], result
[4];
749 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
750 fetch_vector4( ctx
, &inst
->SrcReg
[1], machine
, program
, b
);
751 result
[0] = result
[1] = result
[2] = result
[3] =
752 a
[0] * b
[0] + a
[1] * b
[1] + a
[2] * b
[2];
753 store_vector4( inst
, machine
, result
);
755 printf("DP3 %g = (%g %g %g) . (%g %g %g)\n",
756 result
[0], a
[0], a
[1], a
[2], b
[0], b
[1], b
[2]);
762 GLfloat a
[4], b
[4], result
[4];
763 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
764 fetch_vector4( ctx
, &inst
->SrcReg
[1], machine
, program
, b
);
765 result
[0] = result
[1] = result
[2] = result
[3] =
766 a
[0] * b
[0] + a
[1] * b
[1] + a
[2] * b
[2] + a
[3] * b
[3];
767 store_vector4( inst
, machine
, result
);
769 printf("DP4 %g = (%g, %g %g %g) . (%g, %g %g %g)\n",
770 result
[0], a
[0], a
[1], a
[2], a
[3], b
[0], b
[1], b
[2], b
[3]);
776 GLfloat a
[4], b
[4], result
[4];
777 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
778 fetch_vector4( ctx
, &inst
->SrcReg
[1], machine
, program
, b
);
779 result
[0] = result
[1] = result
[2] = result
[3] =
780 a
[0] * b
[0] + a
[1] * b
[1] + a
[2] * b
[2] + b
[3];
781 store_vector4( inst
, machine
, result
);
784 case OPCODE_DST
: /* Distance vector */
786 GLfloat a
[4], b
[4], result
[4];
787 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
788 fetch_vector4( ctx
, &inst
->SrcReg
[1], machine
, program
, b
);
790 result
[1] = a
[1] * b
[1];
793 store_vector4( inst
, machine
, result
);
796 case OPCODE_EX2
: /* Exponential base 2 */
798 GLfloat a
[4], result
[4];
799 fetch_vector1( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
800 result
[0] = result
[1] = result
[2] = result
[3] =
801 (GLfloat
) _mesa_pow(2.0, a
[0]);
802 store_vector4( inst
, machine
, result
);
807 GLfloat a
[4], result
[4];
808 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
809 result
[0] = FLOORF(a
[0]);
810 result
[1] = FLOORF(a
[1]);
811 result
[2] = FLOORF(a
[2]);
812 result
[3] = FLOORF(a
[3]);
813 store_vector4( inst
, machine
, result
);
818 GLfloat a
[4], result
[4];
819 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
820 result
[0] = a
[0] - FLOORF(a
[0]);
821 result
[1] = a
[1] - FLOORF(a
[1]);
822 result
[2] = a
[2] - FLOORF(a
[2]);
823 result
[3] = a
[3] - FLOORF(a
[3]);
824 store_vector4( inst
, machine
, result
);
827 case OPCODE_KIL_NV
: /* NV_f_p only */
829 const GLuint swizzle
= inst
->DstReg
.CondSwizzle
;
830 const GLuint condMask
= inst
->DstReg
.CondMask
;
831 if (test_cc(machine
->CondCodes
[GET_SWZ(swizzle
, 0)], condMask
) ||
832 test_cc(machine
->CondCodes
[GET_SWZ(swizzle
, 1)], condMask
) ||
833 test_cc(machine
->CondCodes
[GET_SWZ(swizzle
, 2)], condMask
) ||
834 test_cc(machine
->CondCodes
[GET_SWZ(swizzle
, 3)], condMask
)) {
839 case OPCODE_KIL
: /* ARB_f_p only */
842 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
843 if (a
[0] < 0.0F
|| a
[1] < 0.0F
|| a
[2] < 0.0F
|| a
[3] < 0.0F
) {
848 case OPCODE_LG2
: /* log base 2 */
850 GLfloat a
[4], result
[4];
851 fetch_vector1( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
852 result
[0] = result
[1] = result
[2] = result
[3]
854 store_vector4( inst
, machine
, result
);
859 const GLfloat epsilon
= 1.0F
/ 256.0F
; /* from NV VP spec */
860 GLfloat a
[4], result
[4];
861 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
862 a
[0] = MAX2(a
[0], 0.0F
);
863 a
[1] = MAX2(a
[1], 0.0F
);
864 /* XXX ARB version clamps a[3], NV version doesn't */
865 a
[3] = CLAMP(a
[3], -(128.0F
- epsilon
), (128.0F
- epsilon
));
868 /* XXX we could probably just use pow() here */
870 if (a
[1] == 0.0 && a
[3] == 0.0)
873 result
[2] = EXPF(a
[3] * LOGF(a
[1]));
879 store_vector4( inst
, machine
, result
);
884 GLfloat a
[4], b
[4], c
[4], result
[4];
885 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
886 fetch_vector4( ctx
, &inst
->SrcReg
[1], machine
, program
, b
);
887 fetch_vector4( ctx
, &inst
->SrcReg
[2], machine
, program
, c
);
888 result
[0] = a
[0] * b
[0] + (1.0F
- a
[0]) * c
[0];
889 result
[1] = a
[1] * b
[1] + (1.0F
- a
[1]) * c
[1];
890 result
[2] = a
[2] * b
[2] + (1.0F
- a
[2]) * c
[2];
891 result
[3] = a
[3] * b
[3] + (1.0F
- a
[3]) * c
[3];
892 store_vector4( inst
, machine
, result
);
897 GLfloat a
[4], b
[4], c
[4], result
[4];
898 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
899 fetch_vector4( ctx
, &inst
->SrcReg
[1], machine
, program
, b
);
900 fetch_vector4( ctx
, &inst
->SrcReg
[2], machine
, program
, c
);
901 result
[0] = a
[0] * b
[0] + c
[0];
902 result
[1] = a
[1] * b
[1] + c
[1];
903 result
[2] = a
[2] * b
[2] + c
[2];
904 result
[3] = a
[3] * b
[3] + c
[3];
905 store_vector4( inst
, machine
, result
);
910 GLfloat a
[4], b
[4], result
[4];
911 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
912 fetch_vector4( ctx
, &inst
->SrcReg
[1], machine
, program
, b
);
913 result
[0] = MAX2(a
[0], b
[0]);
914 result
[1] = MAX2(a
[1], b
[1]);
915 result
[2] = MAX2(a
[2], b
[2]);
916 result
[3] = MAX2(a
[3], b
[3]);
917 store_vector4( inst
, machine
, result
);
919 printf("MAX (%g %g %g %g) = (%g %g %g %g), (%g %g %g %g)\n",
920 result
[0], result
[1], result
[2], result
[3],
921 a
[0], a
[1], a
[2], a
[3],
922 b
[0], b
[1], b
[2], b
[3]);
928 GLfloat a
[4], b
[4], result
[4];
929 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
930 fetch_vector4( ctx
, &inst
->SrcReg
[1], machine
, program
, b
);
931 result
[0] = MIN2(a
[0], b
[0]);
932 result
[1] = MIN2(a
[1], b
[1]);
933 result
[2] = MIN2(a
[2], b
[2]);
934 result
[3] = MIN2(a
[3], b
[3]);
935 store_vector4( inst
, machine
, result
);
941 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, result
);
942 store_vector4( inst
, machine
, result
);
944 printf("MOV (%g %g %g %g)\n",
945 result
[0], result
[1], result
[2], result
[3]);
951 GLfloat a
[4], b
[4], result
[4];
952 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
953 fetch_vector4( ctx
, &inst
->SrcReg
[1], machine
, program
, b
);
954 result
[0] = a
[0] * b
[0];
955 result
[1] = a
[1] * b
[1];
956 result
[2] = a
[2] * b
[2];
957 result
[3] = a
[3] * b
[3];
958 store_vector4( inst
, machine
, result
);
960 printf("MUL (%g %g %g %g) = (%g %g %g %g) * (%g %g %g %g)\n",
961 result
[0], result
[1], result
[2], result
[3],
962 a
[0], a
[1], a
[2], a
[3],
963 b
[0], b
[1], b
[2], b
[3]);
967 case OPCODE_PK2H
: /* pack two 16-bit floats in one 32-bit float */
969 GLfloat a
[4], result
[4];
971 GLuint
*rawResult
= (GLuint
*) result
;
973 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
974 hx
= _mesa_float_to_half(a
[0]);
975 hy
= _mesa_float_to_half(a
[1]);
976 twoHalves
= hx
| (hy
<< 16);
977 rawResult
[0] = rawResult
[1] = rawResult
[2] = rawResult
[3]
979 store_vector4( inst
, machine
, result
);
982 case OPCODE_PK2US
: /* pack two GLushorts into one 32-bit float */
984 GLfloat a
[4], result
[4];
985 GLuint usx
, usy
, *rawResult
= (GLuint
*) result
;
986 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
987 a
[0] = CLAMP(a
[0], 0.0F
, 1.0F
);
988 a
[1] = CLAMP(a
[1], 0.0F
, 1.0F
);
989 usx
= IROUND(a
[0] * 65535.0F
);
990 usy
= IROUND(a
[1] * 65535.0F
);
991 rawResult
[0] = rawResult
[1] = rawResult
[2] = rawResult
[3]
993 store_vector4( inst
, machine
, result
);
996 case OPCODE_PK4B
: /* pack four GLbytes into one 32-bit float */
998 GLfloat a
[4], result
[4];
999 GLuint ubx
, uby
, ubz
, ubw
, *rawResult
= (GLuint
*) result
;
1000 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
1001 a
[0] = CLAMP(a
[0], -128.0F
/ 127.0F
, 1.0F
);
1002 a
[1] = CLAMP(a
[1], -128.0F
/ 127.0F
, 1.0F
);
1003 a
[2] = CLAMP(a
[2], -128.0F
/ 127.0F
, 1.0F
);
1004 a
[3] = CLAMP(a
[3], -128.0F
/ 127.0F
, 1.0F
);
1005 ubx
= IROUND(127.0F
* a
[0] + 128.0F
);
1006 uby
= IROUND(127.0F
* a
[1] + 128.0F
);
1007 ubz
= IROUND(127.0F
* a
[2] + 128.0F
);
1008 ubw
= IROUND(127.0F
* a
[3] + 128.0F
);
1009 rawResult
[0] = rawResult
[1] = rawResult
[2] = rawResult
[3]
1010 = ubx
| (uby
<< 8) | (ubz
<< 16) | (ubw
<< 24);
1011 store_vector4( inst
, machine
, result
);
1014 case OPCODE_PK4UB
: /* pack four GLubytes into one 32-bit float */
1016 GLfloat a
[4], result
[4];
1017 GLuint ubx
, uby
, ubz
, ubw
, *rawResult
= (GLuint
*) result
;
1018 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
1019 a
[0] = CLAMP(a
[0], 0.0F
, 1.0F
);
1020 a
[1] = CLAMP(a
[1], 0.0F
, 1.0F
);
1021 a
[2] = CLAMP(a
[2], 0.0F
, 1.0F
);
1022 a
[3] = CLAMP(a
[3], 0.0F
, 1.0F
);
1023 ubx
= IROUND(255.0F
* a
[0]);
1024 uby
= IROUND(255.0F
* a
[1]);
1025 ubz
= IROUND(255.0F
* a
[2]);
1026 ubw
= IROUND(255.0F
* a
[3]);
1027 rawResult
[0] = rawResult
[1] = rawResult
[2] = rawResult
[3]
1028 = ubx
| (uby
<< 8) | (ubz
<< 16) | (ubw
<< 24);
1029 store_vector4( inst
, machine
, result
);
1034 GLfloat a
[4], b
[4], result
[4];
1035 fetch_vector1( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
1036 fetch_vector1( ctx
, &inst
->SrcReg
[1], machine
, program
, b
);
1037 result
[0] = result
[1] = result
[2] = result
[3]
1038 = (GLfloat
)_mesa_pow(a
[0], b
[0]);
1039 store_vector4( inst
, machine
, result
);
1044 GLfloat a
[4], result
[4];
1045 fetch_vector1( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
1049 else if (IS_INF_OR_NAN(a
[0]))
1050 printf("RCP(inf)\n");
1052 result
[0] = result
[1] = result
[2] = result
[3]
1054 store_vector4( inst
, machine
, result
);
1059 GLfloat axis
[4], dir
[4], result
[4], tmp
[4];
1060 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, axis
);
1061 fetch_vector4( ctx
, &inst
->SrcReg
[1], machine
, program
, dir
);
1062 tmp
[3] = axis
[0] * axis
[0]
1064 + axis
[2] * axis
[2];
1065 tmp
[0] = (2.0F
* (axis
[0] * dir
[0] +
1067 axis
[2] * dir
[2])) / tmp
[3];
1068 result
[0] = tmp
[0] * axis
[0] - dir
[0];
1069 result
[1] = tmp
[0] * axis
[1] - dir
[1];
1070 result
[2] = tmp
[0] * axis
[2] - dir
[2];
1071 /* result[3] is never written! XXX enforce in parser! */
1072 store_vector4( inst
, machine
, result
);
1075 case OPCODE_RSQ
: /* 1 / sqrt() */
1077 GLfloat a
[4], result
[4];
1078 fetch_vector1( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
1080 result
[0] = result
[1] = result
[2] = result
[3] = INV_SQRTF(a
[0]);
1081 store_vector4( inst
, machine
, result
);
1083 printf("RSQ %g = 1/sqrt(|%g|)\n", result
[0], a
[0]);
1087 case OPCODE_SCS
: /* sine and cos */
1089 GLfloat a
[4], result
[4];
1090 fetch_vector1( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
1091 result
[0] = (GLfloat
)_mesa_cos(a
[0]);
1092 result
[1] = (GLfloat
)_mesa_sin(a
[0]);
1093 result
[2] = 0.0; /* undefined! */
1094 result
[3] = 0.0; /* undefined! */
1095 store_vector4( inst
, machine
, result
);
1098 case OPCODE_SEQ
: /* set on equal */
1100 GLfloat a
[4], b
[4], result
[4];
1101 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
1102 fetch_vector4( ctx
, &inst
->SrcReg
[1], machine
, program
, b
);
1103 result
[0] = (a
[0] == b
[0]) ? 1.0F
: 0.0F
;
1104 result
[1] = (a
[1] == b
[1]) ? 1.0F
: 0.0F
;
1105 result
[2] = (a
[2] == b
[2]) ? 1.0F
: 0.0F
;
1106 result
[3] = (a
[3] == b
[3]) ? 1.0F
: 0.0F
;
1107 store_vector4( inst
, machine
, result
);
1110 case OPCODE_SFL
: /* set false, operands ignored */
1112 static const GLfloat result
[4] = { 0.0F
, 0.0F
, 0.0F
, 0.0F
};
1113 store_vector4( inst
, machine
, result
);
1116 case OPCODE_SGE
: /* set on greater or equal */
1118 GLfloat a
[4], b
[4], result
[4];
1119 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
1120 fetch_vector4( ctx
, &inst
->SrcReg
[1], machine
, program
, b
);
1121 result
[0] = (a
[0] >= b
[0]) ? 1.0F
: 0.0F
;
1122 result
[1] = (a
[1] >= b
[1]) ? 1.0F
: 0.0F
;
1123 result
[2] = (a
[2] >= b
[2]) ? 1.0F
: 0.0F
;
1124 result
[3] = (a
[3] >= b
[3]) ? 1.0F
: 0.0F
;
1125 store_vector4( inst
, machine
, result
);
1128 case OPCODE_SGT
: /* set on greater */
1130 GLfloat a
[4], b
[4], result
[4];
1131 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
1132 fetch_vector4( ctx
, &inst
->SrcReg
[1], machine
, program
, b
);
1133 result
[0] = (a
[0] > b
[0]) ? 1.0F
: 0.0F
;
1134 result
[1] = (a
[1] > b
[1]) ? 1.0F
: 0.0F
;
1135 result
[2] = (a
[2] > b
[2]) ? 1.0F
: 0.0F
;
1136 result
[3] = (a
[3] > b
[3]) ? 1.0F
: 0.0F
;
1137 store_vector4( inst
, machine
, result
);
1142 GLfloat a
[4], result
[4];
1143 fetch_vector1( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
1144 result
[0] = result
[1] = result
[2] =
1145 result
[3] = (GLfloat
)_mesa_sin(a
[0]);
1146 store_vector4( inst
, machine
, result
);
1149 case OPCODE_SLE
: /* set on less or equal */
1151 GLfloat a
[4], b
[4], result
[4];
1152 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
1153 fetch_vector4( ctx
, &inst
->SrcReg
[1], machine
, program
, b
);
1154 result
[0] = (a
[0] <= b
[0]) ? 1.0F
: 0.0F
;
1155 result
[1] = (a
[1] <= b
[1]) ? 1.0F
: 0.0F
;
1156 result
[2] = (a
[2] <= b
[2]) ? 1.0F
: 0.0F
;
1157 result
[3] = (a
[3] <= b
[3]) ? 1.0F
: 0.0F
;
1158 store_vector4( inst
, machine
, result
);
1161 case OPCODE_SLT
: /* set on less */
1163 GLfloat a
[4], b
[4], result
[4];
1164 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
1165 fetch_vector4( ctx
, &inst
->SrcReg
[1], machine
, program
, b
);
1166 result
[0] = (a
[0] < b
[0]) ? 1.0F
: 0.0F
;
1167 result
[1] = (a
[1] < b
[1]) ? 1.0F
: 0.0F
;
1168 result
[2] = (a
[2] < b
[2]) ? 1.0F
: 0.0F
;
1169 result
[3] = (a
[3] < b
[3]) ? 1.0F
: 0.0F
;
1170 store_vector4( inst
, machine
, result
);
1173 case OPCODE_SNE
: /* set on not equal */
1175 GLfloat a
[4], b
[4], result
[4];
1176 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
1177 fetch_vector4( ctx
, &inst
->SrcReg
[1], machine
, program
, b
);
1178 result
[0] = (a
[0] != b
[0]) ? 1.0F
: 0.0F
;
1179 result
[1] = (a
[1] != b
[1]) ? 1.0F
: 0.0F
;
1180 result
[2] = (a
[2] != b
[2]) ? 1.0F
: 0.0F
;
1181 result
[3] = (a
[3] != b
[3]) ? 1.0F
: 0.0F
;
1182 store_vector4( inst
, machine
, result
);
1185 case OPCODE_STR
: /* set true, operands ignored */
1187 static const GLfloat result
[4] = { 1.0F
, 1.0F
, 1.0F
, 1.0F
};
1188 store_vector4( inst
, machine
, result
);
1193 GLfloat a
[4], b
[4], result
[4];
1194 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
1195 fetch_vector4( ctx
, &inst
->SrcReg
[1], machine
, program
, b
);
1196 result
[0] = a
[0] - b
[0];
1197 result
[1] = a
[1] - b
[1];
1198 result
[2] = a
[2] - b
[2];
1199 result
[3] = a
[3] - b
[3];
1200 store_vector4( inst
, machine
, result
);
1205 const struct prog_src_register
*source
= &inst
->SrcReg
[0];
1206 const GLfloat
*src
= get_register_pointer(ctx
, source
,
1211 /* do extended swizzling here */
1212 for (i
= 0; i
< 4; i
++) {
1213 if (GET_SWZ(source
->Swizzle
, i
) == SWIZZLE_ZERO
)
1215 else if (GET_SWZ(source
->Swizzle
, i
) == SWIZZLE_ONE
)
1218 result
[i
] = src
[GET_SWZ(source
->Swizzle
, i
)];
1220 if (source
->NegateBase
& (1 << i
))
1221 result
[i
] = -result
[i
];
1223 store_vector4( inst
, machine
, result
);
1226 case OPCODE_TEX
: /* Both ARB and NV frag prog */
1229 GLfloat texcoord
[4], color
[4];
1230 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, texcoord
);
1231 /* Note: we pass 0 for LOD. The ARB extension requires it
1232 * while the NV extension says it's implementation dependant.
1234 /* KW: Previously lambda was passed as zero, but I
1235 * believe this is incorrect, the spec seems to
1236 * indicate rather that lambda should not be
1237 * changed/biased, unlike TXB where texcoord[3] is
1238 * added to the lambda calculations. The lambda should
1239 * still be calculated normally for TEX & TXP though,
1240 * not set to zero. Otherwise it's very difficult to
1241 * implement normal GL semantics through the fragment
1244 fetch_texel( ctx
, texcoord
,
1245 span
->array
->lambda
[inst
->TexSrcUnit
][column
],
1246 inst
->TexSrcUnit
, color
);
1249 printf("color[3] = %f\n", color
[3]);
1251 store_vector4( inst
, machine
, color
);
1254 case OPCODE_TXB
: /* GL_ARB_fragment_program only */
1255 /* Texel lookup with LOD bias */
1257 GLfloat texcoord
[4], color
[4], bias
, lambda
;
1259 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, texcoord
);
1260 /* texcoord[3] is the bias to add to lambda */
1261 bias
= ctx
->Texture
.Unit
[inst
->TexSrcUnit
].LodBias
1262 + ctx
->Texture
.Unit
[inst
->TexSrcUnit
]._Current
->LodBias
1264 lambda
= span
->array
->lambda
[inst
->TexSrcUnit
][column
] + bias
;
1265 fetch_texel( ctx
, texcoord
, lambda
,
1266 inst
->TexSrcUnit
, color
);
1267 store_vector4( inst
, machine
, color
);
1270 case OPCODE_TXD
: /* GL_NV_fragment_program only */
1271 /* Texture lookup w/ partial derivatives for LOD */
1273 GLfloat texcoord
[4], dtdx
[4], dtdy
[4], color
[4];
1274 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, texcoord
);
1275 fetch_vector4( ctx
, &inst
->SrcReg
[1], machine
, program
, dtdx
);
1276 fetch_vector4( ctx
, &inst
->SrcReg
[2], machine
, program
, dtdy
);
1277 fetch_texel_deriv( ctx
, texcoord
, dtdx
, dtdy
, inst
->TexSrcUnit
,
1279 store_vector4( inst
, machine
, color
);
1282 case OPCODE_TXP
: /* GL_ARB_fragment_program only */
1283 /* Texture lookup w/ projective divide */
1285 GLfloat texcoord
[4], color
[4];
1286 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, texcoord
);
1287 /* Not so sure about this test - if texcoord[3] is
1288 * zero, we'd probably be fine except for an ASSERT in
1289 * IROUND_POS() which gets triggered by the inf values created.
1291 if (texcoord
[3] != 0.0) {
1292 texcoord
[0] /= texcoord
[3];
1293 texcoord
[1] /= texcoord
[3];
1294 texcoord
[2] /= texcoord
[3];
1296 /* KW: Previously lambda was passed as zero, but I
1297 * believe this is incorrect, the spec seems to
1298 * indicate rather that lambda should not be
1299 * changed/biased, unlike TXB where texcoord[3] is
1300 * added to the lambda calculations. The lambda should
1301 * still be calculated normally for TEX & TXP though,
1304 fetch_texel( ctx
, texcoord
,
1305 span
->array
->lambda
[inst
->TexSrcUnit
][column
],
1306 inst
->TexSrcUnit
, color
);
1307 store_vector4( inst
, machine
, color
);
1310 case OPCODE_TXP_NV
: /* GL_NV_fragment_program only */
1311 /* Texture lookup w/ projective divide */
1313 GLfloat texcoord
[4], color
[4];
1314 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, texcoord
);
1315 if (inst
->TexSrcTarget
!= TEXTURE_CUBE_INDEX
&&
1316 texcoord
[3] != 0.0) {
1317 texcoord
[0] /= texcoord
[3];
1318 texcoord
[1] /= texcoord
[3];
1319 texcoord
[2] /= texcoord
[3];
1321 fetch_texel( ctx
, texcoord
,
1322 span
->array
->lambda
[inst
->TexSrcUnit
][column
],
1323 inst
->TexSrcUnit
, color
);
1324 store_vector4( inst
, machine
, color
);
1327 case OPCODE_UP2H
: /* unpack two 16-bit floats */
1329 GLfloat a
[4], result
[4];
1330 const GLuint
*rawBits
= (const GLuint
*) a
;
1332 fetch_vector1( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
1333 hx
= rawBits
[0] & 0xffff;
1334 hy
= rawBits
[0] >> 16;
1335 result
[0] = result
[2] = _mesa_half_to_float(hx
);
1336 result
[1] = result
[3] = _mesa_half_to_float(hy
);
1337 store_vector4( inst
, machine
, result
);
1340 case OPCODE_UP2US
: /* unpack two GLushorts */
1342 GLfloat a
[4], result
[4];
1343 const GLuint
*rawBits
= (const GLuint
*) a
;
1345 fetch_vector1( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
1346 usx
= rawBits
[0] & 0xffff;
1347 usy
= rawBits
[0] >> 16;
1348 result
[0] = result
[2] = usx
* (1.0f
/ 65535.0f
);
1349 result
[1] = result
[3] = usy
* (1.0f
/ 65535.0f
);
1350 store_vector4( inst
, machine
, result
);
1353 case OPCODE_UP4B
: /* unpack four GLbytes */
1355 GLfloat a
[4], result
[4];
1356 const GLuint
*rawBits
= (const GLuint
*) a
;
1357 fetch_vector1( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
1358 result
[0] = (((rawBits
[0] >> 0) & 0xff) - 128) / 127.0F
;
1359 result
[1] = (((rawBits
[0] >> 8) & 0xff) - 128) / 127.0F
;
1360 result
[2] = (((rawBits
[0] >> 16) & 0xff) - 128) / 127.0F
;
1361 result
[3] = (((rawBits
[0] >> 24) & 0xff) - 128) / 127.0F
;
1362 store_vector4( inst
, machine
, result
);
1365 case OPCODE_UP4UB
: /* unpack four GLubytes */
1367 GLfloat a
[4], result
[4];
1368 const GLuint
*rawBits
= (const GLuint
*) a
;
1369 fetch_vector1( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
1370 result
[0] = ((rawBits
[0] >> 0) & 0xff) / 255.0F
;
1371 result
[1] = ((rawBits
[0] >> 8) & 0xff) / 255.0F
;
1372 result
[2] = ((rawBits
[0] >> 16) & 0xff) / 255.0F
;
1373 result
[3] = ((rawBits
[0] >> 24) & 0xff) / 255.0F
;
1374 store_vector4( inst
, machine
, result
);
1377 case OPCODE_XPD
: /* cross product */
1379 GLfloat a
[4], b
[4], result
[4];
1380 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
1381 fetch_vector4( ctx
, &inst
->SrcReg
[1], machine
, program
, b
);
1382 result
[0] = a
[1] * b
[2] - a
[2] * b
[1];
1383 result
[1] = a
[2] * b
[0] - a
[0] * b
[2];
1384 result
[2] = a
[0] * b
[1] - a
[1] * b
[0];
1386 store_vector4( inst
, machine
, result
);
1389 case OPCODE_X2D
: /* 2-D matrix transform */
1391 GLfloat a
[4], b
[4], c
[4], result
[4];
1392 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
1393 fetch_vector4( ctx
, &inst
->SrcReg
[1], machine
, program
, b
);
1394 fetch_vector4( ctx
, &inst
->SrcReg
[2], machine
, program
, c
);
1395 result
[0] = a
[0] + b
[0] * c
[0] + b
[1] * c
[1];
1396 result
[1] = a
[1] + b
[0] * c
[2] + b
[1] * c
[3];
1397 result
[2] = a
[2] + b
[0] * c
[0] + b
[1] * c
[1];
1398 result
[3] = a
[3] + b
[0] * c
[2] + b
[1] * c
[3];
1399 store_vector4( inst
, machine
, result
);
1404 if (inst
->SrcReg
[0].File
!= -1) {
1406 fetch_vector4( ctx
, &inst
->SrcReg
[0], machine
, program
, a
);
1407 _mesa_printf("%s%g, %g, %g, %g\n", (const char *) inst
->Data
,
1408 a
[0], a
[1], a
[2], a
[3]);
1411 _mesa_printf("%s\n", (const char *) inst
->Data
);
1418 _mesa_problem(ctx
, "Bad opcode %d in _mesa_exec_fragment_program",
1420 return GL_TRUE
; /* return value doesn't matter */
1428 * Initialize the virtual fragment program machine state prior to running
1429 * fragment program on a fragment. This involves initializing the input
1430 * registers, condition codes, etc.
1431 * \param machine the virtual machine state to init
1432 * \param program the fragment program we're about to run
1433 * \param span the span of pixels we'll operate on
1434 * \param col which element (column) of the span we'll operate on
1437 init_machine( GLcontext
*ctx
, struct fp_machine
*machine
,
1438 const struct gl_fragment_program
*program
,
1439 const SWspan
*span
, GLuint col
)
1441 GLuint inputsRead
= program
->Base
.InputsRead
;
1444 if (ctx
->FragmentProgram
.CallbackEnabled
)
1447 if (program
->Base
.Target
== GL_FRAGMENT_PROGRAM_NV
) {
1448 /* Clear temporary registers (undefined for ARB_f_p) */
1449 _mesa_bzero(machine
->Temporaries
,
1450 MAX_NV_FRAGMENT_PROGRAM_TEMPS
* 4 * sizeof(GLfloat
));
1453 /* Load input registers */
1454 if (inputsRead
& (1 << FRAG_ATTRIB_WPOS
)) {
1455 GLfloat
*wpos
= machine
->Inputs
[FRAG_ATTRIB_WPOS
];
1456 ASSERT(span
->arrayMask
& SPAN_Z
);
1457 if (span
->arrayMask
& SPAN_XY
) {
1458 wpos
[0] = (GLfloat
) span
->array
->x
[col
];
1459 wpos
[1] = (GLfloat
) span
->array
->y
[col
];
1462 wpos
[0] = (GLfloat
) span
->x
+ col
;
1463 wpos
[1] = (GLfloat
) span
->y
;
1465 wpos
[2] = (GLfloat
) span
->array
->z
[col
] / ctx
->DrawBuffer
->_DepthMaxF
;
1466 wpos
[3] = span
->w
+ col
* span
->dwdx
;
1468 if (inputsRead
& (1 << FRAG_ATTRIB_COL0
)) {
1469 GLfloat
*col0
= machine
->Inputs
[FRAG_ATTRIB_COL0
];
1470 ASSERT(span
->arrayMask
& SPAN_RGBA
);
1471 if (span
->array
->ChanType
== GL_UNSIGNED_BYTE
) {
1472 GLubyte (*rgba
)[4] = span
->array
->color
.sz1
.rgba
;
1473 col0
[0] = UBYTE_TO_FLOAT(rgba
[col
][RCOMP
]);
1474 col0
[1] = UBYTE_TO_FLOAT(rgba
[col
][GCOMP
]);
1475 col0
[2] = UBYTE_TO_FLOAT(rgba
[col
][BCOMP
]);
1476 col0
[3] = UBYTE_TO_FLOAT(rgba
[col
][ACOMP
]);
1478 else if (span
->array
->ChanType
== GL_UNSIGNED_SHORT
) {
1479 GLushort (*rgba
)[4] = span
->array
->color
.sz2
.rgba
;
1480 col0
[0] = USHORT_TO_FLOAT(rgba
[col
][RCOMP
]);
1481 col0
[1] = USHORT_TO_FLOAT(rgba
[col
][GCOMP
]);
1482 col0
[2] = USHORT_TO_FLOAT(rgba
[col
][BCOMP
]);
1483 col0
[3] = USHORT_TO_FLOAT(rgba
[col
][ACOMP
]);
1486 GLfloat (*rgba
)[4] = span
->array
->color
.sz4
.rgba
;
1487 COPY_4V(col0
, rgba
[col
]);
1490 if (inputsRead
& (1 << FRAG_ATTRIB_COL1
)) {
1491 GLfloat
*col1
= machine
->Inputs
[FRAG_ATTRIB_COL1
];
1492 if (span
->array
->ChanType
== GL_UNSIGNED_BYTE
) {
1493 GLubyte (*rgba
)[4] = span
->array
->color
.sz1
.spec
;
1494 col1
[0] = UBYTE_TO_FLOAT(rgba
[col
][RCOMP
]);
1495 col1
[1] = UBYTE_TO_FLOAT(rgba
[col
][GCOMP
]);
1496 col1
[2] = UBYTE_TO_FLOAT(rgba
[col
][BCOMP
]);
1497 col1
[3] = UBYTE_TO_FLOAT(rgba
[col
][ACOMP
]);
1499 else if (span
->array
->ChanType
== GL_UNSIGNED_SHORT
) {
1500 GLushort (*rgba
)[4] = span
->array
->color
.sz2
.spec
;
1501 col1
[0] = USHORT_TO_FLOAT(rgba
[col
][RCOMP
]);
1502 col1
[1] = USHORT_TO_FLOAT(rgba
[col
][GCOMP
]);
1503 col1
[2] = USHORT_TO_FLOAT(rgba
[col
][BCOMP
]);
1504 col1
[3] = USHORT_TO_FLOAT(rgba
[col
][ACOMP
]);
1507 GLfloat (*rgba
)[4] = span
->array
->color
.sz4
.spec
;
1508 COPY_4V(col1
, rgba
[col
]);
1511 if (inputsRead
& (1 << FRAG_ATTRIB_FOGC
)) {
1512 GLfloat
*fogc
= machine
->Inputs
[FRAG_ATTRIB_FOGC
];
1513 ASSERT(span
->arrayMask
& SPAN_FOG
);
1514 fogc
[0] = span
->array
->fog
[col
];
1519 for (u
= 0; u
< ctx
->Const
.MaxTextureCoordUnits
; u
++) {
1520 if (inputsRead
& (1 << (FRAG_ATTRIB_TEX0
+ u
))) {
1521 GLfloat
*tex
= machine
->Inputs
[FRAG_ATTRIB_TEX0
+ u
];
1522 /*ASSERT(ctx->Texture._EnabledCoordUnits & (1 << u));*/
1523 COPY_4V(tex
, span
->array
->texcoords
[u
][col
]);
1524 /*ASSERT(tex[0] != 0 || tex[1] != 0 || tex[2] != 0);*/
1528 /* init condition codes */
1529 machine
->CondCodes
[0] = COND_EQ
;
1530 machine
->CondCodes
[1] = COND_EQ
;
1531 machine
->CondCodes
[2] = COND_EQ
;
1532 machine
->CondCodes
[3] = COND_EQ
;
1537 * Run fragment program on the pixels in span from 'start' to 'end' - 1.
1540 run_program(GLcontext
*ctx
, SWspan
*span
, GLuint start
, GLuint end
)
1542 const struct gl_fragment_program
*program
= ctx
->FragmentProgram
._Current
;
1543 struct fp_machine machine
;
1546 CurrentMachine
= &machine
;
1548 for (i
= start
; i
< end
; i
++) {
1549 if (span
->array
->mask
[i
]) {
1550 init_machine(ctx
, &machine
, program
, span
, i
);
1552 if (!execute_program(ctx
, program
, ~0, &machine
, span
, i
)) {
1553 span
->array
->mask
[i
] = GL_FALSE
; /* killed fragment */
1554 span
->writeAll
= GL_FALSE
;
1557 /* Store output registers */
1559 const GLfloat
*colOut
= machine
.Outputs
[FRAG_RESULT_COLR
];
1560 if (span
->array
->ChanType
== GL_UNSIGNED_BYTE
) {
1561 GLubyte (*rgba
)[4] = span
->array
->color
.sz1
.rgba
;
1562 UNCLAMPED_FLOAT_TO_UBYTE(rgba
[i
][RCOMP
], colOut
[0]);
1563 UNCLAMPED_FLOAT_TO_UBYTE(rgba
[i
][GCOMP
], colOut
[1]);
1564 UNCLAMPED_FLOAT_TO_UBYTE(rgba
[i
][BCOMP
], colOut
[2]);
1565 UNCLAMPED_FLOAT_TO_UBYTE(rgba
[i
][ACOMP
], colOut
[3]);
1567 else if (span
->array
->ChanType
== GL_UNSIGNED_BYTE
) {
1568 GLushort (*rgba
)[4] = span
->array
->color
.sz2
.rgba
;
1569 UNCLAMPED_FLOAT_TO_USHORT(rgba
[i
][RCOMP
], colOut
[0]);
1570 UNCLAMPED_FLOAT_TO_USHORT(rgba
[i
][GCOMP
], colOut
[1]);
1571 UNCLAMPED_FLOAT_TO_USHORT(rgba
[i
][BCOMP
], colOut
[2]);
1572 UNCLAMPED_FLOAT_TO_USHORT(rgba
[i
][ACOMP
], colOut
[3]);
1575 GLfloat (*rgba
)[4] = span
->array
->color
.sz4
.rgba
;
1576 COPY_4V(rgba
[i
], colOut
);
1580 if (program
->Base
.OutputsWritten
& (1 << FRAG_RESULT_DEPR
)) {
1581 const GLfloat depth
= machine
.Outputs
[FRAG_RESULT_DEPR
][2];
1583 span
->array
->z
[i
] = 0;
1584 else if (depth
>= 1.0)
1585 span
->array
->z
[i
] = ctx
->DrawBuffer
->_DepthMax
;
1587 span
->array
->z
[i
] = IROUND(depth
* ctx
->DrawBuffer
->_DepthMaxF
);
1592 CurrentMachine
= NULL
;
1597 * Execute the current fragment program for all the fragments
1598 * in the given span.
1601 _swrast_exec_fragment_program( GLcontext
*ctx
, SWspan
*span
)
1603 const struct gl_fragment_program
*program
= ctx
->FragmentProgram
._Current
;
1605 ctx
->_CurrentProgram
= GL_FRAGMENT_PROGRAM_ARB
; /* or NV, doesn't matter */
1607 if (program
->Base
.Parameters
) {
1608 _mesa_load_state_parameters(ctx
, program
->Base
.Parameters
);
1611 run_program(ctx
, span
, 0, span
->end
);
1613 if (program
->Base
.OutputsWritten
& (1 << FRAG_RESULT_DEPR
)) {
1614 span
->interpMask
&= ~SPAN_Z
;
1615 span
->arrayMask
|= SPAN_Z
;
1618 ctx
->_CurrentProgram
= 0;