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: %d/%d",
93 source
->Index
, I915_MAX_TEMPORARY
);
96 src
= UREG(REG_TYPE_R
, source
->Index
);
99 switch (source
->Index
) {
100 case FRAG_ATTRIB_WPOS
:
101 src
= i915_emit_decl(p
, REG_TYPE_T
, p
->wpos_tex
, D0_CHANNEL_ALL
);
103 case FRAG_ATTRIB_COL0
:
104 src
= i915_emit_decl(p
, REG_TYPE_T
, T_DIFFUSE
, D0_CHANNEL_ALL
);
106 case FRAG_ATTRIB_COL1
:
107 src
= i915_emit_decl(p
, REG_TYPE_T
, T_SPECULAR
, D0_CHANNEL_XYZ
);
108 src
= swizzle(src
, X
, Y
, Z
, ONE
);
110 case FRAG_ATTRIB_FOGC
:
111 src
= i915_emit_decl(p
, REG_TYPE_T
, T_FOG_W
, D0_CHANNEL_W
);
112 src
= swizzle(src
, W
, ZERO
, ZERO
, ONE
);
114 case FRAG_ATTRIB_TEX0
:
115 case FRAG_ATTRIB_TEX1
:
116 case FRAG_ATTRIB_TEX2
:
117 case FRAG_ATTRIB_TEX3
:
118 case FRAG_ATTRIB_TEX4
:
119 case FRAG_ATTRIB_TEX5
:
120 case FRAG_ATTRIB_TEX6
:
121 case FRAG_ATTRIB_TEX7
:
122 src
= i915_emit_decl(p
, REG_TYPE_T
,
123 T_TEX0
+ (source
->Index
- FRAG_ATTRIB_TEX0
),
127 case FRAG_ATTRIB_VAR0
:
128 case FRAG_ATTRIB_VAR0
+ 1:
129 case FRAG_ATTRIB_VAR0
+ 2:
130 case FRAG_ATTRIB_VAR0
+ 3:
131 case FRAG_ATTRIB_VAR0
+ 4:
132 case FRAG_ATTRIB_VAR0
+ 5:
133 case FRAG_ATTRIB_VAR0
+ 6:
134 case FRAG_ATTRIB_VAR0
+ 7:
135 src
= i915_emit_decl(p
, REG_TYPE_T
,
136 T_TEX0
+ (source
->Index
- FRAG_ATTRIB_VAR0
),
141 i915_program_error(p
, "Bad source->Index: %d", source
->Index
);
146 /* Various paramters and env values. All emitted to
147 * hardware as program constants.
149 case PROGRAM_LOCAL_PARAM
:
150 src
= i915_emit_param4fv(p
, program
->Base
.LocalParams
[source
->Index
]);
153 case PROGRAM_ENV_PARAM
:
155 i915_emit_param4fv(p
,
156 p
->ctx
->FragmentProgram
.Parameters
[source
->
160 case PROGRAM_CONSTANT
:
161 case PROGRAM_STATE_VAR
:
162 case PROGRAM_NAMED_PARAM
:
163 case PROGRAM_UNIFORM
:
165 i915_emit_param4fv(p
,
166 program
->Base
.Parameters
->ParameterValues
[source
->
171 i915_program_error(p
, "Bad source->File: %d", source
->File
);
176 GET_SWZ(source
->Swizzle
, 0),
177 GET_SWZ(source
->Swizzle
, 1),
178 GET_SWZ(source
->Swizzle
, 2), GET_SWZ(source
->Swizzle
, 3));
182 GET_BIT(source
->Negate
, 0),
183 GET_BIT(source
->Negate
, 1),
184 GET_BIT(source
->Negate
, 2),
185 GET_BIT(source
->Negate
, 3));
192 get_result_vector(struct i915_fragment_program
*p
,
193 const struct prog_instruction
*inst
)
195 switch (inst
->DstReg
.File
) {
197 switch (inst
->DstReg
.Index
) {
198 case FRAG_RESULT_COLOR
:
199 return UREG(REG_TYPE_OC
, 0);
200 case FRAG_RESULT_DEPTH
:
201 p
->depth_written
= 1;
202 return UREG(REG_TYPE_OD
, 0);
204 i915_program_error(p
, "Bad inst->DstReg.Index: %d",
208 case PROGRAM_TEMPORARY
:
209 return UREG(REG_TYPE_R
, inst
->DstReg
.Index
);
211 i915_program_error(p
, "Bad inst->DstReg.File: %d", inst
->DstReg
.File
);
217 get_result_flags(const struct prog_instruction
*inst
)
221 if (inst
->SaturateMode
== SATURATE_ZERO_ONE
)
222 flags
|= A0_DEST_SATURATE
;
223 if (inst
->DstReg
.WriteMask
& WRITEMASK_X
)
224 flags
|= A0_DEST_CHANNEL_X
;
225 if (inst
->DstReg
.WriteMask
& WRITEMASK_Y
)
226 flags
|= A0_DEST_CHANNEL_Y
;
227 if (inst
->DstReg
.WriteMask
& WRITEMASK_Z
)
228 flags
|= A0_DEST_CHANNEL_Z
;
229 if (inst
->DstReg
.WriteMask
& WRITEMASK_W
)
230 flags
|= A0_DEST_CHANNEL_W
;
236 translate_tex_src_target(struct i915_fragment_program
*p
, GLubyte bit
)
239 case TEXTURE_1D_INDEX
:
240 return D0_SAMPLE_TYPE_2D
;
241 case TEXTURE_2D_INDEX
:
242 return D0_SAMPLE_TYPE_2D
;
243 case TEXTURE_RECT_INDEX
:
244 return D0_SAMPLE_TYPE_2D
;
245 case TEXTURE_3D_INDEX
:
246 return D0_SAMPLE_TYPE_VOLUME
;
247 case TEXTURE_CUBE_INDEX
:
248 return D0_SAMPLE_TYPE_CUBE
;
250 i915_program_error(p
, "TexSrcBit: %d", bit
);
255 #define EMIT_TEX( OP ) \
257 GLuint dim = translate_tex_src_target( p, inst->TexSrcTarget ); \
258 GLuint sampler = i915_emit_decl(p, REG_TYPE_S, \
259 inst->TexSrcUnit, dim); \
260 GLuint coord = src_vector( p, &inst->SrcReg[0], program); \
263 i915_emit_texld( p, get_live_regs(p, inst), \
264 get_result_vector( p, inst ), \
265 get_result_flags( inst ), \
271 #define EMIT_ARITH( OP, N ) \
273 i915_emit_arith( p, \
275 get_result_vector( p, inst ), \
276 get_result_flags( inst ), 0, \
277 (N<1)?0:src_vector( p, &inst->SrcReg[0], program), \
278 (N<2)?0:src_vector( p, &inst->SrcReg[1], program), \
279 (N<3)?0:src_vector( p, &inst->SrcReg[2], program)); \
282 #define EMIT_1ARG_ARITH( OP ) EMIT_ARITH( OP, 1 )
283 #define EMIT_2ARG_ARITH( OP ) EMIT_ARITH( OP, 2 )
284 #define EMIT_3ARG_ARITH( OP ) EMIT_ARITH( OP, 3 )
287 * TODO: consider moving this into core
289 static void calc_live_regs( struct i915_fragment_program
*p
)
291 const struct gl_fragment_program
*program
= p
->ctx
->FragmentProgram
._Current
;
292 GLuint regsUsed
= 0xffff0000;
295 for (i
= program
->Base
.NumInstructions
- 1; i
>= 0; i
--) {
296 struct prog_instruction
*inst
= &program
->Base
.Instructions
[i
];
297 int opArgs
= _mesa_num_inst_src_regs(inst
->Opcode
);
300 /* Register is written to: unmark as live for this and preceeding ops */
301 if (inst
->DstReg
.File
== PROGRAM_TEMPORARY
)
302 regsUsed
&= ~(1 << inst
->DstReg
.Index
);
304 for (a
= 0; a
< opArgs
; a
++) {
305 /* Register is read from: mark as live for this and preceeding ops */
306 if (inst
->SrcReg
[a
].File
== PROGRAM_TEMPORARY
)
307 regsUsed
|= 1 << inst
->SrcReg
[a
].Index
;
310 p
->usedRegs
[i
] = regsUsed
;
314 static GLuint
get_live_regs( struct i915_fragment_program
*p
,
315 const struct prog_instruction
*inst
)
317 const struct gl_fragment_program
*program
= p
->ctx
->FragmentProgram
._Current
;
318 GLuint nr
= inst
- program
->Base
.Instructions
;
320 return p
->usedRegs
[nr
];
324 /* Possible concerns:
326 * SIN, COS -- could use another taylor step?
327 * LIT -- results seem a little different to sw mesa
328 * LOG -- different to mesa on negative numbers, but this is conformant.
330 * Parse failures -- Mesa doesn't currently give a good indication
331 * internally whether a particular program string parsed or not. This
332 * can lead to confusion -- hopefully we cope with it ok now.
336 upload_program(struct i915_fragment_program
*p
)
338 const struct gl_fragment_program
*program
=
339 p
->ctx
->FragmentProgram
._Current
;
340 const struct prog_instruction
*inst
= program
->Base
.Instructions
;
342 if (INTEL_DEBUG
& DEBUG_WM
)
343 _mesa_print_program(&program
->Base
);
345 /* Is this a parse-failed program? Ensure a valid program is
346 * loaded, as the flagging of an error isn't sufficient to stop
347 * this being uploaded to hardware.
349 if (inst
[0].Opcode
== OPCODE_END
) {
350 GLuint tmp
= i915_get_utemp(p
);
353 UREG(REG_TYPE_OC
, 0),
354 A0_DEST_CHANNEL_ALL
, 0,
355 swizzle(tmp
, ONE
, ZERO
, ONE
, ONE
), 0, 0);
359 if (program
->Base
.NumInstructions
> I915_MAX_INSN
) {
360 i915_program_error( p
, "Exceeded max instructions" );
364 /* Not always needed:
369 GLuint src0
, src1
, src2
, flags
;
370 GLuint tmp
= 0, dst
, consts0
= 0, consts1
= 0;
372 switch (inst
->Opcode
) {
374 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
377 get_result_vector(p
, inst
),
378 get_result_flags(inst
), 0,
379 src0
, negate(src0
, 1, 1, 1, 1), 0);
383 EMIT_2ARG_ARITH(A0_ADD
);
387 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
388 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
389 src2
= src_vector(p
, &inst
->SrcReg
[2], program
);
390 i915_emit_arith(p
, A0_CMP
, get_result_vector(p
, inst
), get_result_flags(inst
), 0, src0
, src2
, src1
); /* NOTE: order of src2, src1 */
394 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
395 tmp
= i915_get_utemp(p
);
396 consts0
= i915_emit_const4fv(p
, sin_quad_constants
[0]);
397 consts1
= i915_emit_const4fv(p
, sin_quad_constants
[1]);
399 /* Reduce range from repeating about [-pi,pi] to [-1,1] */
402 tmp
, A0_DEST_CHANNEL_X
, 0,
404 swizzle(consts1
, Z
, ZERO
, ZERO
, ZERO
), /* 1/(2pi) */
405 swizzle(consts0
, W
, ZERO
, ZERO
, ZERO
)); /* .75 */
407 i915_emit_arith(p
, A0_FRC
, tmp
, A0_DEST_CHANNEL_X
, 0, tmp
, 0, 0);
411 tmp
, A0_DEST_CHANNEL_X
, 0,
413 swizzle(consts0
, X
, ZERO
, ZERO
, ZERO
), /* 2 */
414 swizzle(consts0
, Y
, ZERO
, ZERO
, ZERO
)); /* -1 */
416 /* Compute COS with the same calculation used for SIN, but a
417 * different source range has been mapped to [-1,1] this time.
420 /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
423 tmp
, A0_DEST_CHANNEL_Y
, 0,
424 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
425 negate(swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
), 0, 1, 0, 0),
428 /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
431 tmp
, A0_DEST_CHANNEL_Y
, 0,
432 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
436 /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
439 tmp
, A0_DEST_CHANNEL_X
, 0,
441 swizzle(consts1
, X
, Y
, ZERO
, ZERO
),
444 /* tmp.x now contains a first approximation (y). Now, weight it
445 * against tmp.y**2 to get closer.
449 tmp
, A0_DEST_CHANNEL_Y
, 0,
450 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
451 negate(swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
), 0, 1, 0, 0),
454 /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
457 tmp
, A0_DEST_CHANNEL_Y
, 0,
458 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
459 swizzle(tmp
, ZERO
, Y
, ZERO
, ZERO
),
460 negate(swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
), 0, 1, 0, 0));
462 /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
465 get_result_vector(p
, inst
),
466 get_result_flags(inst
), 0,
467 swizzle(consts1
, W
, W
, W
, W
),
468 swizzle(tmp
, Y
, Y
, Y
, Y
),
469 swizzle(tmp
, X
, X
, X
, X
));
473 EMIT_2ARG_ARITH(A0_DP3
);
477 EMIT_2ARG_ARITH(A0_DP4
);
481 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
482 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
486 get_result_vector(p
, inst
),
487 get_result_flags(inst
), 0,
488 swizzle(src0
, X
, Y
, Z
, ONE
), src1
, 0);
492 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
493 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
495 /* result[0] = 1 * 1;
496 * result[1] = a[1] * b[1];
497 * result[2] = a[2] * 1;
498 * result[3] = 1 * b[3];
502 get_result_vector(p
, inst
),
503 get_result_flags(inst
), 0,
504 swizzle(src0
, ONE
, Y
, Z
, ONE
),
505 swizzle(src1
, ONE
, Y
, ONE
, W
), 0);
509 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
513 get_result_vector(p
, inst
),
514 get_result_flags(inst
), 0,
515 swizzle(src0
, X
, X
, X
, X
), 0, 0);
519 EMIT_1ARG_ARITH(A0_FLR
);
523 EMIT_1ARG_ARITH(A0_TRC
);
527 EMIT_1ARG_ARITH(A0_FRC
);
531 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
532 tmp
= i915_get_utemp(p
);
534 i915_emit_texld(p
, get_live_regs(p
, inst
),
535 tmp
, A0_DEST_CHANNEL_ALL
, /* use a dummy dest reg */
536 0, src0
, T0_TEXKILL
);
540 if (inst
->DstReg
.CondMask
== COND_TR
) {
541 tmp
= i915_get_utemp(p
);
543 i915_emit_texld(p
, get_live_regs(p
, inst
),
544 tmp
, A0_DEST_CHANNEL_ALL
,
545 0, /* use a dummy dest reg */
546 swizzle(tmp
, ONE
, ONE
, ONE
, ONE
), /* always */
550 i915_program_error(p
, "Unsupported KIL_NV condition code: %d",
551 inst
->DstReg
.CondMask
);
556 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
560 get_result_vector(p
, inst
),
561 get_result_flags(inst
), 0,
562 swizzle(src0
, X
, X
, X
, X
), 0, 0);
566 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
567 tmp
= i915_get_utemp(p
);
569 /* tmp = max( a.xyzw, a.00zw )
570 * XXX: Clamp tmp.w to -128..128
572 * tmp.y = tmp.w * tmp.y
574 * result = cmp (a.11-x1, a.1x01, a.1xy1 )
576 i915_emit_arith(p
, A0_MAX
, tmp
, A0_DEST_CHANNEL_ALL
, 0,
577 src0
, swizzle(src0
, ZERO
, ZERO
, Z
, W
), 0);
579 i915_emit_arith(p
, A0_LOG
, tmp
, A0_DEST_CHANNEL_Y
, 0,
580 swizzle(tmp
, Y
, Y
, Y
, Y
), 0, 0);
582 i915_emit_arith(p
, A0_MUL
, tmp
, A0_DEST_CHANNEL_Y
, 0,
583 swizzle(tmp
, ZERO
, Y
, ZERO
, ZERO
),
584 swizzle(tmp
, ZERO
, W
, ZERO
, ZERO
), 0);
586 i915_emit_arith(p
, A0_EXP
, tmp
, A0_DEST_CHANNEL_Y
, 0,
587 swizzle(tmp
, Y
, Y
, Y
, Y
), 0, 0);
589 i915_emit_arith(p
, A0_CMP
,
590 get_result_vector(p
, inst
),
591 get_result_flags(inst
), 0,
592 negate(swizzle(tmp
, ONE
, ONE
, X
, ONE
), 0, 0, 1, 0),
593 swizzle(tmp
, ONE
, X
, ZERO
, ONE
),
594 swizzle(tmp
, ONE
, X
, Y
, ONE
));
599 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
600 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
601 src2
= src_vector(p
, &inst
->SrcReg
[2], program
);
602 flags
= get_result_flags(inst
);
603 tmp
= i915_get_utemp(p
);
610 * result = (-c)*a + tmp
612 i915_emit_arith(p
, A0_MAD
, tmp
,
613 flags
& A0_DEST_CHANNEL_ALL
, 0, src1
, src0
, src2
);
615 i915_emit_arith(p
, A0_MAD
,
616 get_result_vector(p
, inst
),
617 flags
, 0, negate(src2
, 1, 1, 1, 1), src0
, tmp
);
621 EMIT_3ARG_ARITH(A0_MAD
);
625 EMIT_2ARG_ARITH(A0_MAX
);
629 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
630 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
631 tmp
= i915_get_utemp(p
);
632 flags
= get_result_flags(inst
);
636 tmp
, flags
& A0_DEST_CHANNEL_ALL
, 0,
637 negate(src0
, 1, 1, 1, 1),
638 negate(src1
, 1, 1, 1, 1), 0);
642 get_result_vector(p
, inst
),
643 flags
, 0, negate(tmp
, 1, 1, 1, 1), 0, 0);
647 EMIT_1ARG_ARITH(A0_MOV
);
651 EMIT_2ARG_ARITH(A0_MUL
);
658 /* Don't implement noise because we just don't have the instructions
659 * to spare. We aren't the first vendor to do so.
661 i915_program_error(p
, "Stubbed-out noise functions");
664 get_result_vector(p
, inst
),
665 get_result_flags(inst
), 0,
666 swizzle(tmp
, ZERO
, ZERO
, ZERO
, ZERO
), 0, 0);
669 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
670 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
671 tmp
= i915_get_utemp(p
);
672 flags
= get_result_flags(inst
);
674 /* XXX: masking on intermediate values, here and elsewhere.
678 tmp
, A0_DEST_CHANNEL_X
, 0,
679 swizzle(src0
, X
, X
, X
, X
), 0, 0);
681 i915_emit_arith(p
, A0_MUL
, tmp
, A0_DEST_CHANNEL_X
, 0, tmp
, src1
, 0);
686 get_result_vector(p
, inst
),
687 flags
, 0, swizzle(tmp
, X
, X
, X
, X
), 0, 0);
692 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
696 get_result_vector(p
, inst
),
697 get_result_flags(inst
), 0,
698 swizzle(src0
, X
, X
, X
, X
), 0, 0);
703 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
707 get_result_vector(p
, inst
),
708 get_result_flags(inst
), 0,
709 swizzle(src0
, X
, X
, X
, X
), 0, 0);
713 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
714 tmp
= i915_get_utemp(p
);
717 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
718 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
719 * t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x
720 * scs.x = DP4 t1, sin_constants
721 * t1 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1
722 * scs.y = DP4 t1, cos_constants
726 tmp
, A0_DEST_CHANNEL_XY
, 0,
727 swizzle(src0
, X
, X
, ONE
, ONE
),
728 swizzle(src0
, X
, ONE
, ONE
, ONE
), 0);
732 tmp
, A0_DEST_CHANNEL_ALL
, 0,
733 swizzle(tmp
, X
, Y
, X
, Y
),
734 swizzle(tmp
, X
, X
, ONE
, ONE
), 0);
736 if (inst
->DstReg
.WriteMask
& WRITEMASK_Y
) {
739 if (inst
->DstReg
.WriteMask
& WRITEMASK_X
)
740 tmp1
= i915_get_utemp(p
);
746 tmp1
, A0_DEST_CHANNEL_ALL
, 0,
747 swizzle(tmp
, X
, Y
, Y
, W
),
748 swizzle(tmp
, X
, Z
, ONE
, ONE
), 0);
752 get_result_vector(p
, inst
),
753 A0_DEST_CHANNEL_Y
, 0,
754 swizzle(tmp1
, W
, Z
, Y
, X
),
755 i915_emit_const4fv(p
, sin_constants
), 0);
758 if (inst
->DstReg
.WriteMask
& WRITEMASK_X
) {
761 tmp
, A0_DEST_CHANNEL_XYZ
, 0,
762 swizzle(tmp
, X
, X
, Z
, ONE
),
763 swizzle(tmp
, Z
, ONE
, ONE
, ONE
), 0);
767 get_result_vector(p
, inst
),
768 A0_DEST_CHANNEL_X
, 0,
769 swizzle(tmp
, ONE
, Z
, Y
, X
),
770 i915_emit_const4fv(p
, cos_constants
), 0);
775 tmp
= i915_get_utemp(p
);
776 flags
= get_result_flags(inst
);
777 dst
= get_result_vector(p
, inst
);
779 /* dst = src1 >= src2 */
784 src_vector(p
, &inst
->SrcReg
[0], program
),
785 src_vector(p
, &inst
->SrcReg
[1], program
),
787 /* tmp = src1 <= src2 */
792 negate(src_vector(p
, &inst
->SrcReg
[0], program
),
794 negate(src_vector(p
, &inst
->SrcReg
[1], program
),
797 /* dst = tmp && dst */
808 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
809 tmp
= i915_get_utemp(p
);
810 consts0
= i915_emit_const4fv(p
, sin_quad_constants
[0]);
811 consts1
= i915_emit_const4fv(p
, sin_quad_constants
[1]);
813 /* Reduce range from repeating about [-pi,pi] to [-1,1] */
816 tmp
, A0_DEST_CHANNEL_X
, 0,
818 swizzle(consts1
, Z
, ZERO
, ZERO
, ZERO
), /* 1/(2pi) */
819 swizzle(consts0
, Z
, ZERO
, ZERO
, ZERO
)); /* .5 */
821 i915_emit_arith(p
, A0_FRC
, tmp
, A0_DEST_CHANNEL_X
, 0, tmp
, 0, 0);
825 tmp
, A0_DEST_CHANNEL_X
, 0,
827 swizzle(consts0
, X
, ZERO
, ZERO
, ZERO
), /* 2 */
828 swizzle(consts0
, Y
, ZERO
, ZERO
, ZERO
)); /* -1 */
830 /* Compute sin using a quadratic and quartic. It gives continuity
831 * that repeating the Taylor series lacks every 2*pi, and has
834 * The idea was described at:
835 * http://www.devmaster.net/forums/showthread.php?t=5784
838 /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
841 tmp
, A0_DEST_CHANNEL_Y
, 0,
842 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
843 negate(swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
), 0, 1, 0, 0),
846 /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
849 tmp
, A0_DEST_CHANNEL_Y
, 0,
850 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
854 /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
857 tmp
, A0_DEST_CHANNEL_X
, 0,
859 swizzle(consts1
, X
, Y
, ZERO
, ZERO
),
862 /* tmp.x now contains a first approximation (y). Now, weight it
863 * against tmp.y**2 to get closer.
867 tmp
, A0_DEST_CHANNEL_Y
, 0,
868 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
869 negate(swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
), 0, 1, 0, 0),
872 /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
875 tmp
, A0_DEST_CHANNEL_Y
, 0,
876 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
877 swizzle(tmp
, ZERO
, Y
, ZERO
, ZERO
),
878 negate(swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
), 0, 1, 0, 0));
880 /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
883 get_result_vector(p
, inst
),
884 get_result_flags(inst
), 0,
885 swizzle(consts1
, W
, W
, W
, W
),
886 swizzle(tmp
, Y
, Y
, Y
, Y
),
887 swizzle(tmp
, X
, X
, X
, X
));
892 EMIT_2ARG_ARITH(A0_SGE
);
898 get_result_vector( p
, inst
),
899 get_result_flags( inst
), 0,
900 negate(src_vector( p
, &inst
->SrcReg
[0], program
),
902 negate(src_vector( p
, &inst
->SrcReg
[1], program
),
910 get_result_vector( p
, inst
),
911 get_result_flags( inst
), 0,
912 negate(src_vector( p
, &inst
->SrcReg
[0], program
),
914 negate(src_vector( p
, &inst
->SrcReg
[1], program
),
920 EMIT_2ARG_ARITH(A0_SLT
);
924 tmp
= i915_get_utemp(p
);
925 flags
= get_result_flags(inst
);
926 dst
= get_result_vector(p
, inst
);
928 /* dst = src1 < src2 */
933 src_vector(p
, &inst
->SrcReg
[0], program
),
934 src_vector(p
, &inst
->SrcReg
[1], program
),
936 /* tmp = src1 > src2 */
941 negate(src_vector(p
, &inst
->SrcReg
[0], program
),
943 negate(src_vector(p
, &inst
->SrcReg
[1], program
),
946 /* dst = tmp || dst */
950 flags
| A0_DEST_SATURATE
, 0,
957 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
958 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
962 get_result_vector(p
, inst
),
963 get_result_flags(inst
), 0,
964 src0
, negate(src1
, 1, 1, 1, 1), 0);
968 EMIT_1ARG_ARITH(A0_MOV
); /* extended swizzle handled natively */
985 * result.x = src0.y * src1.z - src0.z * src1.y;
986 * result.y = src0.z * src1.x - src0.x * src1.z;
987 * result.z = src0.x * src1.y - src0.y * src1.x;
990 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
991 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
992 tmp
= i915_get_utemp(p
);
996 tmp
, A0_DEST_CHANNEL_ALL
, 0,
997 swizzle(src0
, Z
, X
, Y
, ONE
),
998 swizzle(src1
, Y
, Z
, X
, ONE
), 0);
1002 get_result_vector(p
, inst
),
1003 get_result_flags(inst
), 0,
1004 swizzle(src0
, Y
, Z
, X
, ONE
),
1005 swizzle(src1
, Z
, X
, Y
, ONE
),
1006 negate(tmp
, 1, 1, 1, 0));
1012 case OPCODE_BGNLOOP
:
1022 case OPCODE_ENDLOOP
:
1027 i915_program_error(p
, "Unsupported opcode: %s",
1028 _mesa_opcode_string(inst
->Opcode
));
1033 /* These opcodes are claimed as GLSL, NV_vp, and ARB_vp in
1034 * prog_instruction.h, but apparently GLSL doesn't ever emit them.
1035 * Instead, it translates to EX2 or LG2.
1039 /* These opcodes are claimed by GLSL in prog_instruction.h, but
1040 * only NV_vp/fp appears to emit them.
1043 i915_program_error(p
, "bad opcode: %s",
1044 _mesa_opcode_string(inst
->Opcode
));
1049 i915_release_utemps(p
);
1053 /* Rather than trying to intercept and jiggle depth writes during
1054 * emit, just move the value into its correct position at the end of
1058 fixup_depth_write(struct i915_fragment_program
*p
)
1060 if (p
->depth_written
) {
1061 GLuint depth
= UREG(REG_TYPE_OD
, 0);
1065 depth
, A0_DEST_CHANNEL_W
, 0,
1066 swizzle(depth
, X
, Y
, Z
, Z
), 0, 0);
1072 check_wpos(struct i915_fragment_program
*p
)
1074 GLuint inputs
= p
->FragProg
.Base
.InputsRead
;
1079 for (i
= 0; i
< p
->ctx
->Const
.MaxTextureCoordUnits
; i
++) {
1080 if (inputs
& (FRAG_BIT_TEX(i
) | FRAG_BIT_VAR(i
)))
1082 else if (inputs
& FRAG_BIT_WPOS
) {
1084 inputs
&= ~FRAG_BIT_WPOS
;
1088 if (inputs
& FRAG_BIT_WPOS
) {
1089 i915_program_error(p
, "No free texcoord for wpos value");
1095 translate_program(struct i915_fragment_program
*p
)
1097 struct i915_context
*i915
= I915_CONTEXT(p
->ctx
);
1099 i915_init_program(i915
, p
);
1102 fixup_depth_write(p
);
1103 i915_fini_program(p
);
1110 track_params(struct i915_fragment_program
*p
)
1115 _mesa_load_state_parameters(p
->ctx
, p
->FragProg
.Base
.Parameters
);
1117 for (i
= 0; i
< p
->nr_params
; i
++) {
1118 GLint reg
= p
->param
[i
].reg
;
1119 COPY_4V(p
->constant
[reg
], p
->param
[i
].values
);
1122 p
->params_uptodate
= 1;
1123 p
->on_hardware
= 0; /* overkill */
1128 i915BindProgram(GLcontext
* ctx
, GLenum target
, struct gl_program
*prog
)
1130 if (target
== GL_FRAGMENT_PROGRAM_ARB
) {
1131 struct i915_context
*i915
= I915_CONTEXT(ctx
);
1132 struct i915_fragment_program
*p
= (struct i915_fragment_program
*) prog
;
1134 if (i915
->current_program
== p
)
1137 if (i915
->current_program
) {
1138 i915
->current_program
->on_hardware
= 0;
1139 i915
->current_program
->params_uptodate
= 0;
1142 i915
->current_program
= p
;
1144 assert(p
->on_hardware
== 0);
1145 assert(p
->params_uptodate
== 0);
1150 static struct gl_program
*
1151 i915NewProgram(GLcontext
* ctx
, GLenum target
, GLuint id
)
1154 case GL_VERTEX_PROGRAM_ARB
:
1155 return _mesa_init_vertex_program(ctx
, CALLOC_STRUCT(gl_vertex_program
),
1158 case GL_FRAGMENT_PROGRAM_ARB
:{
1159 struct i915_fragment_program
*prog
=
1160 CALLOC_STRUCT(i915_fragment_program
);
1162 i915_init_program(I915_CONTEXT(ctx
), prog
);
1164 return _mesa_init_fragment_program(ctx
, &prog
->FragProg
,
1174 return _mesa_new_program(ctx
, target
, id
);
1179 i915DeleteProgram(GLcontext
* ctx
, struct gl_program
*prog
)
1181 if (prog
->Target
== GL_FRAGMENT_PROGRAM_ARB
) {
1182 struct i915_context
*i915
= I915_CONTEXT(ctx
);
1183 struct i915_fragment_program
*p
= (struct i915_fragment_program
*) prog
;
1185 if (i915
->current_program
== p
)
1186 i915
->current_program
= 0;
1189 _mesa_delete_program(ctx
, prog
);
1194 i915IsProgramNative(GLcontext
* ctx
, GLenum target
, struct gl_program
*prog
)
1196 if (target
== GL_FRAGMENT_PROGRAM_ARB
) {
1197 struct i915_fragment_program
*p
= (struct i915_fragment_program
*) prog
;
1200 translate_program(p
);
1209 i915ProgramStringNotify(GLcontext
* ctx
,
1210 GLenum target
, struct gl_program
*prog
)
1212 if (target
== GL_FRAGMENT_PROGRAM_ARB
) {
1213 struct i915_fragment_program
*p
= (struct i915_fragment_program
*) prog
;
1216 /* Hack: make sure fog is correctly enabled according to this
1217 * fragment program's fog options.
1219 if (p
->FragProg
.FogOption
) {
1220 /* add extra instructions to do fog, then turn off FogOption field */
1221 _mesa_append_fog_code(ctx
, &p
->FragProg
);
1222 p
->FragProg
.FogOption
= GL_NONE
;
1226 _tnl_program_string(ctx
, target
, prog
);
1230 i915_update_program(GLcontext
*ctx
)
1232 struct intel_context
*intel
= intel_context(ctx
);
1233 struct i915_context
*i915
= i915_context(&intel
->ctx
);
1234 struct i915_fragment_program
*fp
=
1235 (struct i915_fragment_program
*) ctx
->FragmentProgram
._Current
;
1237 if (i915
->current_program
!= fp
) {
1238 if (i915
->current_program
) {
1239 i915
->current_program
->on_hardware
= 0;
1240 i915
->current_program
->params_uptodate
= 0;
1243 i915
->current_program
= fp
;
1246 if (!fp
->translated
)
1247 translate_program(fp
);
1249 FALLBACK(&i915
->intel
, I915_FALLBACK_PROGRAM
, fp
->error
);
1253 i915ValidateFragmentProgram(struct i915_context
*i915
)
1255 GLcontext
*ctx
= &i915
->intel
.ctx
;
1256 struct intel_context
*intel
= intel_context(ctx
);
1257 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
1258 struct vertex_buffer
*VB
= &tnl
->vb
;
1260 struct i915_fragment_program
*p
=
1261 (struct i915_fragment_program
*) ctx
->FragmentProgram
._Current
;
1263 const GLuint inputsRead
= p
->FragProg
.Base
.InputsRead
;
1264 GLuint s4
= i915
->state
.Ctx
[I915_CTXREG_LIS4
] & ~S4_VFMT_MASK
;
1265 GLuint s2
= S2_TEXCOORD_NONE
;
1270 VB
->AttribPtr
[VERT_ATTRIB_POS
] = VB
->NdcPtr
;
1273 translate_program(p
);
1275 intel
->vertex_attr_count
= 0;
1276 intel
->wpos_offset
= 0;
1277 intel
->wpos_size
= 0;
1278 intel
->coloroffset
= 0;
1279 intel
->specoffset
= 0;
1281 if (inputsRead
& FRAG_BITS_TEX_ANY
) {
1282 EMIT_ATTR(_TNL_ATTRIB_POS
, EMIT_4F_VIEWPORT
, S4_VFMT_XYZW
, 16);
1285 EMIT_ATTR(_TNL_ATTRIB_POS
, EMIT_3F_VIEWPORT
, S4_VFMT_XYZ
, 12);
1288 if (inputsRead
& FRAG_BIT_COL0
) {
1289 intel
->coloroffset
= offset
/ 4;
1290 EMIT_ATTR(_TNL_ATTRIB_COLOR0
, EMIT_4UB_4F_BGRA
, S4_VFMT_COLOR
, 4);
1293 if (inputsRead
& FRAG_BIT_COL1
) {
1294 intel
->specoffset
= offset
/ 4;
1295 EMIT_ATTR(_TNL_ATTRIB_COLOR1
, EMIT_4UB_4F_BGRA
, S4_VFMT_SPEC_FOG
, 4);
1298 if ((inputsRead
& FRAG_BIT_FOGC
) || i915
->vertex_fog
!= I915_FOG_NONE
) {
1299 EMIT_ATTR(_TNL_ATTRIB_FOG
, EMIT_1F
, S4_VFMT_FOG_PARAM
, 4);
1302 for (i
= 0; i
< p
->ctx
->Const
.MaxTextureCoordUnits
; i
++) {
1303 if (inputsRead
& FRAG_BIT_TEX(i
)) {
1304 int sz
= VB
->AttribPtr
[_TNL_ATTRIB_TEX0
+ i
]->size
;
1306 s2
&= ~S2_TEXCOORD_FMT(i
, S2_TEXCOORD_FMT0_MASK
);
1307 s2
|= S2_TEXCOORD_FMT(i
, SZ_TO_HW(sz
));
1309 EMIT_ATTR(_TNL_ATTRIB_TEX0
+ i
, EMIT_SZ(sz
), 0, sz
* 4);
1311 else if (inputsRead
& FRAG_BIT_VAR(i
)) {
1312 int sz
= VB
->AttribPtr
[_TNL_ATTRIB_GENERIC0
+ i
]->size
;
1314 s2
&= ~S2_TEXCOORD_FMT(i
, S2_TEXCOORD_FMT0_MASK
);
1315 s2
|= S2_TEXCOORD_FMT(i
, SZ_TO_HW(sz
));
1317 EMIT_ATTR(_TNL_ATTRIB_GENERIC0
+ i
, EMIT_SZ(sz
), 0, sz
* 4);
1319 else if (i
== p
->wpos_tex
) {
1321 /* If WPOS is required, duplicate the XYZ position data in an
1322 * unused texture coordinate:
1324 s2
&= ~S2_TEXCOORD_FMT(i
, S2_TEXCOORD_FMT0_MASK
);
1325 s2
|= S2_TEXCOORD_FMT(i
, SZ_TO_HW(3));
1327 intel
->wpos_offset
= offset
;
1328 intel
->wpos_size
= 3 * sizeof(GLuint
);
1330 EMIT_PAD(intel
->wpos_size
);
1334 if (s2
!= i915
->state
.Ctx
[I915_CTXREG_LIS2
] ||
1335 s4
!= i915
->state
.Ctx
[I915_CTXREG_LIS4
]) {
1338 I915_STATECHANGE(i915
, I915_UPLOAD_CTX
);
1340 /* Must do this *after* statechange, so as not to affect
1341 * buffered vertices reliant on the old state:
1343 intel
->vertex_size
= _tnl_install_attrs(&intel
->ctx
,
1344 intel
->vertex_attrs
,
1345 intel
->vertex_attr_count
,
1346 intel
->ViewportMatrix
.m
, 0);
1348 intel
->vertex_size
>>= 2;
1350 i915
->state
.Ctx
[I915_CTXREG_LIS2
] = s2
;
1351 i915
->state
.Ctx
[I915_CTXREG_LIS4
] = s4
;
1353 k
= intel
->vtbl
.check_vertex_size(intel
, intel
->vertex_size
);
1357 if (!p
->params_uptodate
)
1360 if (!p
->on_hardware
)
1361 i915_upload_program(i915
, p
);
1365 i915InitFragProgFuncs(struct dd_function_table
*functions
)
1367 functions
->BindProgram
= i915BindProgram
;
1368 functions
->NewProgram
= i915NewProgram
;
1369 functions
->DeleteProgram
= i915DeleteProgram
;
1370 functions
->IsProgramNative
= i915IsProgramNative
;
1371 functions
->ProgramStringNotify
= i915ProgramStringNotify
;