1 /**************************************************************************
3 * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **************************************************************************/
28 #include "main/glheader.h"
29 #include "main/macros.h"
30 #include "main/enums.h"
32 #include "shader/prog_instruction.h"
33 #include "shader/prog_parameter.h"
34 #include "shader/program.h"
35 #include "shader/programopt.h"
36 #include "shader/prog_print.h"
39 #include "tnl/t_context.h"
41 #include "intel_batchbuffer.h"
44 #include "i915_context.h"
45 #include "i915_program.h"
47 static const GLfloat sin_quad_constants
[2][4] = {
62 static const GLfloat sin_constants
[4] = { 1.0,
64 1.0 / (5 * 4 * 3 * 2 * 1),
65 -1.0 / (7 * 6 * 5 * 4 * 3 * 2 * 1)
68 /* 1, -1/2!, 1/4!, -1/6! */
69 static const GLfloat cos_constants
[4] = { 1.0,
71 1.0 / (4 * 3 * 2 * 1),
72 -1.0 / (6 * 5 * 4 * 3 * 2 * 1)
76 * Retrieve a ureg for the given source register. Will emit
77 * constants, apply swizzling and negation as needed.
80 src_vector(struct i915_fragment_program
*p
,
81 const struct prog_src_register
*source
,
82 const struct gl_fragment_program
*program
)
86 switch (source
->File
) {
90 case PROGRAM_TEMPORARY
:
91 if (source
->Index
>= I915_MAX_TEMPORARY
) {
92 i915_program_error(p
, "Exceeded max temporary reg");
95 src
= UREG(REG_TYPE_R
, source
->Index
);
98 switch (source
->Index
) {
99 case FRAG_ATTRIB_WPOS
:
100 src
= i915_emit_decl(p
, REG_TYPE_T
, p
->wpos_tex
, D0_CHANNEL_ALL
);
102 case FRAG_ATTRIB_COL0
:
103 src
= i915_emit_decl(p
, REG_TYPE_T
, T_DIFFUSE
, D0_CHANNEL_ALL
);
105 case FRAG_ATTRIB_COL1
:
106 src
= i915_emit_decl(p
, REG_TYPE_T
, T_SPECULAR
, D0_CHANNEL_XYZ
);
107 src
= swizzle(src
, X
, Y
, Z
, ONE
);
109 case FRAG_ATTRIB_FOGC
:
110 src
= i915_emit_decl(p
, REG_TYPE_T
, T_FOG_W
, D0_CHANNEL_W
);
111 src
= swizzle(src
, W
, ZERO
, ZERO
, ONE
);
113 case FRAG_ATTRIB_TEX0
:
114 case FRAG_ATTRIB_TEX1
:
115 case FRAG_ATTRIB_TEX2
:
116 case FRAG_ATTRIB_TEX3
:
117 case FRAG_ATTRIB_TEX4
:
118 case FRAG_ATTRIB_TEX5
:
119 case FRAG_ATTRIB_TEX6
:
120 case FRAG_ATTRIB_TEX7
:
121 src
= i915_emit_decl(p
, REG_TYPE_T
,
122 T_TEX0
+ (source
->Index
- FRAG_ATTRIB_TEX0
),
127 i915_program_error(p
, "Bad source->Index");
132 /* Various paramters and env values. All emitted to
133 * hardware as program constants.
135 case PROGRAM_LOCAL_PARAM
:
136 src
= i915_emit_param4fv(p
, program
->Base
.LocalParams
[source
->Index
]);
139 case PROGRAM_ENV_PARAM
:
141 i915_emit_param4fv(p
,
142 p
->ctx
->FragmentProgram
.Parameters
[source
->
146 case PROGRAM_CONSTANT
:
147 case PROGRAM_STATE_VAR
:
148 case PROGRAM_NAMED_PARAM
:
150 i915_emit_param4fv(p
,
151 program
->Base
.Parameters
->ParameterValues
[source
->
156 i915_program_error(p
, "Bad source->File");
161 GET_SWZ(source
->Swizzle
, 0),
162 GET_SWZ(source
->Swizzle
, 1),
163 GET_SWZ(source
->Swizzle
, 2), GET_SWZ(source
->Swizzle
, 3));
167 GET_BIT(source
->Negate
, 0),
168 GET_BIT(source
->Negate
, 1),
169 GET_BIT(source
->Negate
, 2),
170 GET_BIT(source
->Negate
, 3));
177 get_result_vector(struct i915_fragment_program
*p
,
178 const struct prog_instruction
*inst
)
180 switch (inst
->DstReg
.File
) {
182 switch (inst
->DstReg
.Index
) {
183 case FRAG_RESULT_COLOR
:
184 return UREG(REG_TYPE_OC
, 0);
185 case FRAG_RESULT_DEPTH
:
186 p
->depth_written
= 1;
187 return UREG(REG_TYPE_OD
, 0);
189 i915_program_error(p
, "Bad inst->DstReg.Index");
192 case PROGRAM_TEMPORARY
:
193 return UREG(REG_TYPE_R
, inst
->DstReg
.Index
);
195 i915_program_error(p
, "Bad inst->DstReg.File");
201 get_result_flags(const struct prog_instruction
*inst
)
205 if (inst
->SaturateMode
== SATURATE_ZERO_ONE
)
206 flags
|= A0_DEST_SATURATE
;
207 if (inst
->DstReg
.WriteMask
& WRITEMASK_X
)
208 flags
|= A0_DEST_CHANNEL_X
;
209 if (inst
->DstReg
.WriteMask
& WRITEMASK_Y
)
210 flags
|= A0_DEST_CHANNEL_Y
;
211 if (inst
->DstReg
.WriteMask
& WRITEMASK_Z
)
212 flags
|= A0_DEST_CHANNEL_Z
;
213 if (inst
->DstReg
.WriteMask
& WRITEMASK_W
)
214 flags
|= A0_DEST_CHANNEL_W
;
220 translate_tex_src_target(struct i915_fragment_program
*p
, GLubyte bit
)
223 case TEXTURE_1D_INDEX
:
224 return D0_SAMPLE_TYPE_2D
;
225 case TEXTURE_2D_INDEX
:
226 return D0_SAMPLE_TYPE_2D
;
227 case TEXTURE_RECT_INDEX
:
228 return D0_SAMPLE_TYPE_2D
;
229 case TEXTURE_3D_INDEX
:
230 return D0_SAMPLE_TYPE_VOLUME
;
231 case TEXTURE_CUBE_INDEX
:
232 return D0_SAMPLE_TYPE_CUBE
;
234 i915_program_error(p
, "TexSrcBit");
239 #define EMIT_TEX( OP ) \
241 GLuint dim = translate_tex_src_target( p, inst->TexSrcTarget ); \
242 GLuint sampler = i915_emit_decl(p, REG_TYPE_S, \
243 inst->TexSrcUnit, dim); \
244 GLuint coord = src_vector( p, &inst->SrcReg[0], program); \
247 i915_emit_texld( p, get_live_regs(p, inst), \
248 get_result_vector( p, inst ), \
249 get_result_flags( inst ), \
255 #define EMIT_ARITH( OP, N ) \
257 i915_emit_arith( p, \
259 get_result_vector( p, inst ), \
260 get_result_flags( inst ), 0, \
261 (N<1)?0:src_vector( p, &inst->SrcReg[0], program), \
262 (N<2)?0:src_vector( p, &inst->SrcReg[1], program), \
263 (N<3)?0:src_vector( p, &inst->SrcReg[2], program)); \
266 #define EMIT_1ARG_ARITH( OP ) EMIT_ARITH( OP, 1 )
267 #define EMIT_2ARG_ARITH( OP ) EMIT_ARITH( OP, 2 )
268 #define EMIT_3ARG_ARITH( OP ) EMIT_ARITH( OP, 3 )
271 * TODO: consider moving this into core
273 static void calc_live_regs( struct i915_fragment_program
*p
)
275 const struct gl_fragment_program
*program
= p
->ctx
->FragmentProgram
._Current
;
276 GLuint regsUsed
= 0xffff0000;
279 for (i
= program
->Base
.NumInstructions
- 1; i
>= 0; i
--) {
280 struct prog_instruction
*inst
= &program
->Base
.Instructions
[i
];
281 int opArgs
= _mesa_num_inst_src_regs(inst
->Opcode
);
284 /* Register is written to: unmark as live for this and preceeding ops */
285 if (inst
->DstReg
.File
== PROGRAM_TEMPORARY
)
286 regsUsed
&= ~(1 << inst
->DstReg
.Index
);
288 for (a
= 0; a
< opArgs
; a
++) {
289 /* Register is read from: mark as live for this and preceeding ops */
290 if (inst
->SrcReg
[a
].File
== PROGRAM_TEMPORARY
)
291 regsUsed
|= 1 << inst
->SrcReg
[a
].Index
;
294 p
->usedRegs
[i
] = regsUsed
;
298 static GLuint
get_live_regs( struct i915_fragment_program
*p
,
299 const struct prog_instruction
*inst
)
301 const struct gl_fragment_program
*program
= p
->ctx
->FragmentProgram
._Current
;
302 GLuint nr
= inst
- program
->Base
.Instructions
;
304 return p
->usedRegs
[nr
];
308 /* Possible concerns:
310 * SIN, COS -- could use another taylor step?
311 * LIT -- results seem a little different to sw mesa
312 * LOG -- different to mesa on negative numbers, but this is conformant.
314 * Parse failures -- Mesa doesn't currently give a good indication
315 * internally whether a particular program string parsed or not. This
316 * can lead to confusion -- hopefully we cope with it ok now.
320 upload_program(struct i915_fragment_program
*p
)
322 const struct gl_fragment_program
*program
=
323 p
->ctx
->FragmentProgram
._Current
;
324 const struct prog_instruction
*inst
= program
->Base
.Instructions
;
326 if (INTEL_DEBUG
& DEBUG_WM
)
327 _mesa_print_program(&program
->Base
);
329 /* Is this a parse-failed program? Ensure a valid program is
330 * loaded, as the flagging of an error isn't sufficient to stop
331 * this being uploaded to hardware.
333 if (inst
[0].Opcode
== OPCODE_END
) {
334 GLuint tmp
= i915_get_utemp(p
);
337 UREG(REG_TYPE_OC
, 0),
338 A0_DEST_CHANNEL_ALL
, 0,
339 swizzle(tmp
, ONE
, ZERO
, ONE
, ONE
), 0, 0);
343 if (program
->Base
.NumInstructions
> I915_MAX_INSN
) {
344 i915_program_error( p
, "Exceeded max instructions" );
348 /* Not always needed:
353 GLuint src0
, src1
, src2
, flags
;
354 GLuint tmp
= 0, consts0
= 0, consts1
= 0;
356 switch (inst
->Opcode
) {
358 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
361 get_result_vector(p
, inst
),
362 get_result_flags(inst
), 0,
363 src0
, negate(src0
, 1, 1, 1, 1), 0);
367 EMIT_2ARG_ARITH(A0_ADD
);
371 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
372 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
373 src2
= src_vector(p
, &inst
->SrcReg
[2], program
);
374 i915_emit_arith(p
, A0_CMP
, get_result_vector(p
, inst
), get_result_flags(inst
), 0, src0
, src2
, src1
); /* NOTE: order of src2, src1 */
378 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
379 tmp
= i915_get_utemp(p
);
380 consts0
= i915_emit_const4fv(p
, sin_quad_constants
[0]);
381 consts1
= i915_emit_const4fv(p
, sin_quad_constants
[1]);
383 /* Reduce range from repeating about [-pi,pi] to [-1,1] */
386 tmp
, A0_DEST_CHANNEL_X
, 0,
388 swizzle(consts1
, Z
, ZERO
, ZERO
, ZERO
), /* 1/(2pi) */
389 swizzle(consts0
, W
, ZERO
, ZERO
, ZERO
)); /* .75 */
391 i915_emit_arith(p
, A0_FRC
, tmp
, A0_DEST_CHANNEL_X
, 0, tmp
, 0, 0);
395 tmp
, A0_DEST_CHANNEL_X
, 0,
397 swizzle(consts0
, X
, ZERO
, ZERO
, ZERO
), /* 2 */
398 swizzle(consts0
, Y
, ZERO
, ZERO
, ZERO
)); /* -1 */
400 /* Compute COS with the same calculation used for SIN, but a
401 * different source range has been mapped to [-1,1] this time.
404 /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
407 tmp
, A0_DEST_CHANNEL_Y
, 0,
408 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
409 negate(swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
), 0, 1, 0, 0),
412 /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
415 tmp
, A0_DEST_CHANNEL_Y
, 0,
416 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
420 /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
423 tmp
, A0_DEST_CHANNEL_X
, 0,
425 swizzle(consts1
, X
, Y
, ZERO
, ZERO
),
428 /* tmp.x now contains a first approximation (y). Now, weight it
429 * against tmp.y**2 to get closer.
433 tmp
, A0_DEST_CHANNEL_Y
, 0,
434 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
435 negate(swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
), 0, 1, 0, 0),
438 /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
441 tmp
, A0_DEST_CHANNEL_Y
, 0,
442 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
443 swizzle(tmp
, ZERO
, Y
, ZERO
, ZERO
),
444 negate(swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
), 0, 1, 0, 0));
446 /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
449 get_result_vector(p
, inst
),
450 get_result_flags(inst
), 0,
451 swizzle(consts1
, W
, W
, W
, W
),
452 swizzle(tmp
, Y
, Y
, Y
, Y
),
453 swizzle(tmp
, X
, X
, X
, X
));
457 EMIT_2ARG_ARITH(A0_DP3
);
461 EMIT_2ARG_ARITH(A0_DP4
);
465 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
466 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
470 get_result_vector(p
, inst
),
471 get_result_flags(inst
), 0,
472 swizzle(src0
, X
, Y
, Z
, ONE
), src1
, 0);
476 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
477 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
479 /* result[0] = 1 * 1;
480 * result[1] = a[1] * b[1];
481 * result[2] = a[2] * 1;
482 * result[3] = 1 * b[3];
486 get_result_vector(p
, inst
),
487 get_result_flags(inst
), 0,
488 swizzle(src0
, ONE
, Y
, Z
, ONE
),
489 swizzle(src1
, ONE
, Y
, ONE
, W
), 0);
493 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
497 get_result_vector(p
, inst
),
498 get_result_flags(inst
), 0,
499 swizzle(src0
, X
, X
, X
, X
), 0, 0);
503 EMIT_1ARG_ARITH(A0_FLR
);
507 EMIT_1ARG_ARITH(A0_FRC
);
511 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
512 tmp
= i915_get_utemp(p
);
514 i915_emit_texld(p
, get_live_regs(p
, inst
),
515 tmp
, A0_DEST_CHANNEL_ALL
, /* use a dummy dest reg */
516 0, src0
, T0_TEXKILL
);
520 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
524 get_result_vector(p
, inst
),
525 get_result_flags(inst
), 0,
526 swizzle(src0
, X
, X
, X
, X
), 0, 0);
530 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
531 tmp
= i915_get_utemp(p
);
533 /* tmp = max( a.xyzw, a.00zw )
534 * XXX: Clamp tmp.w to -128..128
536 * tmp.y = tmp.w * tmp.y
538 * result = cmp (a.11-x1, a.1x01, a.1xy1 )
540 i915_emit_arith(p
, A0_MAX
, tmp
, A0_DEST_CHANNEL_ALL
, 0,
541 src0
, swizzle(src0
, ZERO
, ZERO
, Z
, W
), 0);
543 i915_emit_arith(p
, A0_LOG
, tmp
, A0_DEST_CHANNEL_Y
, 0,
544 swizzle(tmp
, Y
, Y
, Y
, Y
), 0, 0);
546 i915_emit_arith(p
, A0_MUL
, tmp
, A0_DEST_CHANNEL_Y
, 0,
547 swizzle(tmp
, ZERO
, Y
, ZERO
, ZERO
),
548 swizzle(tmp
, ZERO
, W
, ZERO
, ZERO
), 0);
550 i915_emit_arith(p
, A0_EXP
, tmp
, A0_DEST_CHANNEL_Y
, 0,
551 swizzle(tmp
, Y
, Y
, Y
, Y
), 0, 0);
553 i915_emit_arith(p
, A0_CMP
,
554 get_result_vector(p
, inst
),
555 get_result_flags(inst
), 0,
556 negate(swizzle(tmp
, ONE
, ONE
, X
, ONE
), 0, 0, 1, 0),
557 swizzle(tmp
, ONE
, X
, ZERO
, ONE
),
558 swizzle(tmp
, ONE
, X
, Y
, ONE
));
563 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
564 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
565 src2
= src_vector(p
, &inst
->SrcReg
[2], program
);
566 flags
= get_result_flags(inst
);
567 tmp
= i915_get_utemp(p
);
574 * result = (-c)*a + tmp
576 i915_emit_arith(p
, A0_MAD
, tmp
,
577 flags
& A0_DEST_CHANNEL_ALL
, 0, src1
, src0
, src2
);
579 i915_emit_arith(p
, A0_MAD
,
580 get_result_vector(p
, inst
),
581 flags
, 0, negate(src2
, 1, 1, 1, 1), src0
, tmp
);
585 EMIT_3ARG_ARITH(A0_MAD
);
589 EMIT_2ARG_ARITH(A0_MAX
);
593 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
594 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
595 tmp
= i915_get_utemp(p
);
596 flags
= get_result_flags(inst
);
600 tmp
, flags
& A0_DEST_CHANNEL_ALL
, 0,
601 negate(src0
, 1, 1, 1, 1),
602 negate(src1
, 1, 1, 1, 1), 0);
606 get_result_vector(p
, inst
),
607 flags
, 0, negate(tmp
, 1, 1, 1, 1), 0, 0);
611 EMIT_1ARG_ARITH(A0_MOV
);
615 EMIT_2ARG_ARITH(A0_MUL
);
619 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
620 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
621 tmp
= i915_get_utemp(p
);
622 flags
= get_result_flags(inst
);
624 /* XXX: masking on intermediate values, here and elsewhere.
628 tmp
, A0_DEST_CHANNEL_X
, 0,
629 swizzle(src0
, X
, X
, X
, X
), 0, 0);
631 i915_emit_arith(p
, A0_MUL
, tmp
, A0_DEST_CHANNEL_X
, 0, tmp
, src1
, 0);
636 get_result_vector(p
, inst
),
637 flags
, 0, swizzle(tmp
, X
, X
, X
, X
), 0, 0);
642 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
646 get_result_vector(p
, inst
),
647 get_result_flags(inst
), 0,
648 swizzle(src0
, X
, X
, X
, X
), 0, 0);
653 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
657 get_result_vector(p
, inst
),
658 get_result_flags(inst
), 0,
659 swizzle(src0
, X
, X
, X
, X
), 0, 0);
663 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
664 tmp
= i915_get_utemp(p
);
667 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
668 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
669 * t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x
670 * scs.x = DP4 t1, sin_constants
671 * t1 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1
672 * scs.y = DP4 t1, cos_constants
676 tmp
, A0_DEST_CHANNEL_XY
, 0,
677 swizzle(src0
, X
, X
, ONE
, ONE
),
678 swizzle(src0
, X
, ONE
, ONE
, ONE
), 0);
682 tmp
, A0_DEST_CHANNEL_ALL
, 0,
683 swizzle(tmp
, X
, Y
, X
, Y
),
684 swizzle(tmp
, X
, X
, ONE
, ONE
), 0);
686 if (inst
->DstReg
.WriteMask
& WRITEMASK_Y
) {
689 if (inst
->DstReg
.WriteMask
& WRITEMASK_X
)
690 tmp1
= i915_get_utemp(p
);
696 tmp1
, A0_DEST_CHANNEL_ALL
, 0,
697 swizzle(tmp
, X
, Y
, Y
, W
),
698 swizzle(tmp
, X
, Z
, ONE
, ONE
), 0);
702 get_result_vector(p
, inst
),
703 A0_DEST_CHANNEL_Y
, 0,
704 swizzle(tmp1
, W
, Z
, Y
, X
),
705 i915_emit_const4fv(p
, sin_constants
), 0);
708 if (inst
->DstReg
.WriteMask
& WRITEMASK_X
) {
711 tmp
, A0_DEST_CHANNEL_XYZ
, 0,
712 swizzle(tmp
, X
, X
, Z
, ONE
),
713 swizzle(tmp
, Z
, ONE
, ONE
, ONE
), 0);
717 get_result_vector(p
, inst
),
718 A0_DEST_CHANNEL_X
, 0,
719 swizzle(tmp
, ONE
, Z
, Y
, X
),
720 i915_emit_const4fv(p
, cos_constants
), 0);
725 EMIT_2ARG_ARITH(A0_SGE
);
729 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
730 tmp
= i915_get_utemp(p
);
731 consts0
= i915_emit_const4fv(p
, sin_quad_constants
[0]);
732 consts1
= i915_emit_const4fv(p
, sin_quad_constants
[1]);
734 /* Reduce range from repeating about [-pi,pi] to [-1,1] */
737 tmp
, A0_DEST_CHANNEL_X
, 0,
739 swizzle(consts1
, Z
, ZERO
, ZERO
, ZERO
), /* 1/(2pi) */
740 swizzle(consts0
, Z
, ZERO
, ZERO
, ZERO
)); /* .5 */
742 i915_emit_arith(p
, A0_FRC
, tmp
, A0_DEST_CHANNEL_X
, 0, tmp
, 0, 0);
746 tmp
, A0_DEST_CHANNEL_X
, 0,
748 swizzle(consts0
, X
, ZERO
, ZERO
, ZERO
), /* 2 */
749 swizzle(consts0
, Y
, ZERO
, ZERO
, ZERO
)); /* -1 */
751 /* Compute sin using a quadratic and quartic. It gives continuity
752 * that repeating the Taylor series lacks every 2*pi, and has
755 * The idea was described at:
756 * http://www.devmaster.net/forums/showthread.php?t=5784
759 /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
762 tmp
, A0_DEST_CHANNEL_Y
, 0,
763 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
764 negate(swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
), 0, 1, 0, 0),
767 /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
770 tmp
, A0_DEST_CHANNEL_Y
, 0,
771 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
775 /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
778 tmp
, A0_DEST_CHANNEL_X
, 0,
780 swizzle(consts1
, X
, Y
, ZERO
, ZERO
),
783 /* tmp.x now contains a first approximation (y). Now, weight it
784 * against tmp.y**2 to get closer.
788 tmp
, A0_DEST_CHANNEL_Y
, 0,
789 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
790 negate(swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
), 0, 1, 0, 0),
793 /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
796 tmp
, A0_DEST_CHANNEL_Y
, 0,
797 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
798 swizzle(tmp
, ZERO
, Y
, ZERO
, ZERO
),
799 negate(swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
), 0, 1, 0, 0));
801 /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
804 get_result_vector(p
, inst
),
805 get_result_flags(inst
), 0,
806 swizzle(consts1
, W
, W
, W
, W
),
807 swizzle(tmp
, Y
, Y
, Y
, Y
),
808 swizzle(tmp
, X
, X
, X
, X
));
813 EMIT_2ARG_ARITH(A0_SLT
);
817 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
818 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
822 get_result_vector(p
, inst
),
823 get_result_flags(inst
), 0,
824 src0
, negate(src1
, 1, 1, 1, 1), 0);
828 EMIT_1ARG_ARITH(A0_MOV
); /* extended swizzle handled natively */
845 * result.x = src0.y * src1.z - src0.z * src1.y;
846 * result.y = src0.z * src1.x - src0.x * src1.z;
847 * result.z = src0.x * src1.y - src0.y * src1.x;
850 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
851 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
852 tmp
= i915_get_utemp(p
);
856 tmp
, A0_DEST_CHANNEL_ALL
, 0,
857 swizzle(src0
, Z
, X
, Y
, ONE
),
858 swizzle(src1
, Y
, Z
, X
, ONE
), 0);
862 get_result_vector(p
, inst
),
863 get_result_flags(inst
), 0,
864 swizzle(src0
, Y
, Z
, X
, ONE
),
865 swizzle(src1
, Z
, X
, Y
, ONE
),
866 negate(tmp
, 1, 1, 1, 0));
873 i915_program_error(p
, "bad opcode");
878 i915_release_utemps(p
);
882 /* Rather than trying to intercept and jiggle depth writes during
883 * emit, just move the value into its correct position at the end of
887 fixup_depth_write(struct i915_fragment_program
*p
)
889 if (p
->depth_written
) {
890 GLuint depth
= UREG(REG_TYPE_OD
, 0);
894 depth
, A0_DEST_CHANNEL_W
, 0,
895 swizzle(depth
, X
, Y
, Z
, Z
), 0, 0);
901 check_wpos(struct i915_fragment_program
*p
)
903 GLuint inputs
= p
->FragProg
.Base
.InputsRead
;
908 for (i
= 0; i
< p
->ctx
->Const
.MaxTextureCoordUnits
; i
++) {
909 if (inputs
& FRAG_BIT_TEX(i
))
911 else if (inputs
& FRAG_BIT_WPOS
) {
913 inputs
&= ~FRAG_BIT_WPOS
;
917 if (inputs
& FRAG_BIT_WPOS
) {
918 i915_program_error(p
, "No free texcoord for wpos value");
924 translate_program(struct i915_fragment_program
*p
)
926 struct i915_context
*i915
= I915_CONTEXT(p
->ctx
);
928 i915_init_program(i915
, p
);
931 fixup_depth_write(p
);
932 i915_fini_program(p
);
939 track_params(struct i915_fragment_program
*p
)
944 _mesa_load_state_parameters(p
->ctx
, p
->FragProg
.Base
.Parameters
);
946 for (i
= 0; i
< p
->nr_params
; i
++) {
947 GLint reg
= p
->param
[i
].reg
;
948 COPY_4V(p
->constant
[reg
], p
->param
[i
].values
);
951 p
->params_uptodate
= 1;
952 p
->on_hardware
= 0; /* overkill */
957 i915BindProgram(GLcontext
* ctx
, GLenum target
, struct gl_program
*prog
)
959 if (target
== GL_FRAGMENT_PROGRAM_ARB
) {
960 struct i915_context
*i915
= I915_CONTEXT(ctx
);
961 struct i915_fragment_program
*p
= (struct i915_fragment_program
*) prog
;
963 if (i915
->current_program
== p
)
966 if (i915
->current_program
) {
967 i915
->current_program
->on_hardware
= 0;
968 i915
->current_program
->params_uptodate
= 0;
971 i915
->current_program
= p
;
973 assert(p
->on_hardware
== 0);
974 assert(p
->params_uptodate
== 0);
979 static struct gl_program
*
980 i915NewProgram(GLcontext
* ctx
, GLenum target
, GLuint id
)
983 case GL_VERTEX_PROGRAM_ARB
:
984 return _mesa_init_vertex_program(ctx
, CALLOC_STRUCT(gl_vertex_program
),
987 case GL_FRAGMENT_PROGRAM_ARB
:{
988 struct i915_fragment_program
*prog
=
989 CALLOC_STRUCT(i915_fragment_program
);
991 i915_init_program(I915_CONTEXT(ctx
), prog
);
993 return _mesa_init_fragment_program(ctx
, &prog
->FragProg
,
1003 return _mesa_new_program(ctx
, target
, id
);
1008 i915DeleteProgram(GLcontext
* ctx
, struct gl_program
*prog
)
1010 if (prog
->Target
== GL_FRAGMENT_PROGRAM_ARB
) {
1011 struct i915_context
*i915
= I915_CONTEXT(ctx
);
1012 struct i915_fragment_program
*p
= (struct i915_fragment_program
*) prog
;
1014 if (i915
->current_program
== p
)
1015 i915
->current_program
= 0;
1018 _mesa_delete_program(ctx
, prog
);
1023 i915IsProgramNative(GLcontext
* ctx
, GLenum target
, struct gl_program
*prog
)
1025 if (target
== GL_FRAGMENT_PROGRAM_ARB
) {
1026 struct i915_fragment_program
*p
= (struct i915_fragment_program
*) prog
;
1029 translate_program(p
);
1038 i915ProgramStringNotify(GLcontext
* ctx
,
1039 GLenum target
, struct gl_program
*prog
)
1041 if (target
== GL_FRAGMENT_PROGRAM_ARB
) {
1042 struct i915_fragment_program
*p
= (struct i915_fragment_program
*) prog
;
1045 /* Hack: make sure fog is correctly enabled according to this
1046 * fragment program's fog options.
1048 if (p
->FragProg
.FogOption
) {
1049 /* add extra instructions to do fog, then turn off FogOption field */
1050 _mesa_append_fog_code(ctx
, &p
->FragProg
);
1051 p
->FragProg
.FogOption
= GL_NONE
;
1055 _tnl_program_string(ctx
, target
, prog
);
1060 i915ValidateFragmentProgram(struct i915_context
*i915
)
1062 GLcontext
*ctx
= &i915
->intel
.ctx
;
1063 struct intel_context
*intel
= intel_context(ctx
);
1064 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
1065 struct vertex_buffer
*VB
= &tnl
->vb
;
1067 struct i915_fragment_program
*p
=
1068 (struct i915_fragment_program
*) ctx
->FragmentProgram
._Current
;
1070 const GLuint inputsRead
= p
->FragProg
.Base
.InputsRead
;
1071 GLuint s4
= i915
->state
.Ctx
[I915_CTXREG_LIS4
] & ~S4_VFMT_MASK
;
1072 GLuint s2
= S2_TEXCOORD_NONE
;
1075 if (i915
->current_program
!= p
) {
1076 if (i915
->current_program
) {
1077 i915
->current_program
->on_hardware
= 0;
1078 i915
->current_program
->params_uptodate
= 0;
1081 i915
->current_program
= p
;
1087 VB
->AttribPtr
[VERT_ATTRIB_POS
] = VB
->NdcPtr
;
1090 translate_program(p
);
1092 intel
->vertex_attr_count
= 0;
1093 intel
->wpos_offset
= 0;
1094 intel
->wpos_size
= 0;
1095 intel
->coloroffset
= 0;
1096 intel
->specoffset
= 0;
1098 if (inputsRead
& FRAG_BITS_TEX_ANY
) {
1099 EMIT_ATTR(_TNL_ATTRIB_POS
, EMIT_4F_VIEWPORT
, S4_VFMT_XYZW
, 16);
1102 EMIT_ATTR(_TNL_ATTRIB_POS
, EMIT_3F_VIEWPORT
, S4_VFMT_XYZ
, 12);
1105 if (inputsRead
& FRAG_BIT_COL0
) {
1106 intel
->coloroffset
= offset
/ 4;
1107 EMIT_ATTR(_TNL_ATTRIB_COLOR0
, EMIT_4UB_4F_BGRA
, S4_VFMT_COLOR
, 4);
1110 if (inputsRead
& FRAG_BIT_COL1
) {
1111 intel
->specoffset
= offset
/ 4;
1112 EMIT_ATTR(_TNL_ATTRIB_COLOR1
, EMIT_4UB_4F_BGRA
, S4_VFMT_SPEC_FOG
, 4);
1115 if ((inputsRead
& FRAG_BIT_FOGC
) || i915
->vertex_fog
!= I915_FOG_NONE
) {
1116 EMIT_ATTR(_TNL_ATTRIB_FOG
, EMIT_1F
, S4_VFMT_FOG_PARAM
, 4);
1119 for (i
= 0; i
< p
->ctx
->Const
.MaxTextureCoordUnits
; i
++) {
1120 if (inputsRead
& FRAG_BIT_TEX(i
)) {
1121 int sz
= VB
->TexCoordPtr
[i
]->size
;
1123 s2
&= ~S2_TEXCOORD_FMT(i
, S2_TEXCOORD_FMT0_MASK
);
1124 s2
|= S2_TEXCOORD_FMT(i
, SZ_TO_HW(sz
));
1126 EMIT_ATTR(_TNL_ATTRIB_TEX0
+ i
, EMIT_SZ(sz
), 0, sz
* 4);
1128 else if (i
== p
->wpos_tex
) {
1130 /* If WPOS is required, duplicate the XYZ position data in an
1131 * unused texture coordinate:
1133 s2
&= ~S2_TEXCOORD_FMT(i
, S2_TEXCOORD_FMT0_MASK
);
1134 s2
|= S2_TEXCOORD_FMT(i
, SZ_TO_HW(3));
1136 intel
->wpos_offset
= offset
;
1137 intel
->wpos_size
= 3 * sizeof(GLuint
);
1139 EMIT_PAD(intel
->wpos_size
);
1143 if (s2
!= i915
->state
.Ctx
[I915_CTXREG_LIS2
] ||
1144 s4
!= i915
->state
.Ctx
[I915_CTXREG_LIS4
]) {
1147 I915_STATECHANGE(i915
, I915_UPLOAD_CTX
);
1149 /* Must do this *after* statechange, so as not to affect
1150 * buffered vertices reliant on the old state:
1152 intel
->vertex_size
= _tnl_install_attrs(&intel
->ctx
,
1153 intel
->vertex_attrs
,
1154 intel
->vertex_attr_count
,
1155 intel
->ViewportMatrix
.m
, 0);
1157 intel
->vertex_size
>>= 2;
1159 i915
->state
.Ctx
[I915_CTXREG_LIS2
] = s2
;
1160 i915
->state
.Ctx
[I915_CTXREG_LIS4
] = s4
;
1162 k
= intel
->vtbl
.check_vertex_size(intel
, intel
->vertex_size
);
1166 if (!p
->params_uptodate
)
1169 if (!p
->on_hardware
)
1170 i915_upload_program(i915
, p
);
1174 i915InitFragProgFuncs(struct dd_function_table
*functions
)
1176 functions
->BindProgram
= i915BindProgram
;
1177 functions
->NewProgram
= i915NewProgram
;
1178 functions
->DeleteProgram
= i915DeleteProgram
;
1179 functions
->IsProgramNative
= i915IsProgramNative
;
1180 functions
->ProgramStringNotify
= i915ProgramStringNotify
;