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 "program/prog_instruction.h"
33 #include "program/prog_parameter.h"
34 #include "program/program.h"
35 #include "program/programopt.h"
36 #include "program/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
);
147 switch (source
->Index
) {
148 case FRAG_RESULT_COLOR
:
149 src
= UREG(REG_TYPE_OC
, 0);
151 case FRAG_RESULT_DEPTH
:
152 src
= UREG(REG_TYPE_OD
, 0);
155 i915_program_error(p
, "Bad source->Index: %d", source
->Index
);
160 /* Various paramters and env values. All emitted to
161 * hardware as program constants.
163 case PROGRAM_LOCAL_PARAM
:
164 src
= i915_emit_param4fv(p
, program
->Base
.LocalParams
[source
->Index
]);
167 case PROGRAM_ENV_PARAM
:
169 i915_emit_param4fv(p
,
170 p
->ctx
->FragmentProgram
.Parameters
[source
->
174 case PROGRAM_CONSTANT
:
175 case PROGRAM_STATE_VAR
:
176 case PROGRAM_NAMED_PARAM
:
177 case PROGRAM_UNIFORM
:
179 i915_emit_param4fv(p
,
180 program
->Base
.Parameters
->ParameterValues
[source
->
185 i915_program_error(p
, "Bad source->File: %d", source
->File
);
190 GET_SWZ(source
->Swizzle
, 0),
191 GET_SWZ(source
->Swizzle
, 1),
192 GET_SWZ(source
->Swizzle
, 2), GET_SWZ(source
->Swizzle
, 3));
196 GET_BIT(source
->Negate
, 0),
197 GET_BIT(source
->Negate
, 1),
198 GET_BIT(source
->Negate
, 2),
199 GET_BIT(source
->Negate
, 3));
206 get_result_vector(struct i915_fragment_program
*p
,
207 const struct prog_instruction
*inst
)
209 switch (inst
->DstReg
.File
) {
211 switch (inst
->DstReg
.Index
) {
212 case FRAG_RESULT_COLOR
:
213 return UREG(REG_TYPE_OC
, 0);
214 case FRAG_RESULT_DEPTH
:
215 p
->depth_written
= 1;
216 return UREG(REG_TYPE_OD
, 0);
218 i915_program_error(p
, "Bad inst->DstReg.Index: %d",
222 case PROGRAM_TEMPORARY
:
223 return UREG(REG_TYPE_R
, inst
->DstReg
.Index
);
225 i915_program_error(p
, "Bad inst->DstReg.File: %d", inst
->DstReg
.File
);
231 get_result_flags(const struct prog_instruction
*inst
)
235 if (inst
->SaturateMode
== SATURATE_ZERO_ONE
)
236 flags
|= A0_DEST_SATURATE
;
237 if (inst
->DstReg
.WriteMask
& WRITEMASK_X
)
238 flags
|= A0_DEST_CHANNEL_X
;
239 if (inst
->DstReg
.WriteMask
& WRITEMASK_Y
)
240 flags
|= A0_DEST_CHANNEL_Y
;
241 if (inst
->DstReg
.WriteMask
& WRITEMASK_Z
)
242 flags
|= A0_DEST_CHANNEL_Z
;
243 if (inst
->DstReg
.WriteMask
& WRITEMASK_W
)
244 flags
|= A0_DEST_CHANNEL_W
;
250 translate_tex_src_target(struct i915_fragment_program
*p
, GLubyte bit
)
253 case TEXTURE_1D_INDEX
:
254 return D0_SAMPLE_TYPE_2D
;
255 case TEXTURE_2D_INDEX
:
256 return D0_SAMPLE_TYPE_2D
;
257 case TEXTURE_RECT_INDEX
:
258 return D0_SAMPLE_TYPE_2D
;
259 case TEXTURE_3D_INDEX
:
260 return D0_SAMPLE_TYPE_VOLUME
;
261 case TEXTURE_CUBE_INDEX
:
262 return D0_SAMPLE_TYPE_CUBE
;
264 i915_program_error(p
, "TexSrcBit: %d", bit
);
269 #define EMIT_TEX( OP ) \
271 GLuint dim = translate_tex_src_target( p, inst->TexSrcTarget ); \
272 const struct gl_fragment_program *program = p->ctx->FragmentProgram._Current; \
273 GLuint unit = program->Base.SamplerUnits[inst->TexSrcUnit]; \
274 GLuint sampler = i915_emit_decl(p, REG_TYPE_S, \
276 GLuint coord = src_vector( p, &inst->SrcReg[0], program); \
279 i915_emit_texld( p, get_live_regs(p, inst), \
280 get_result_vector( p, inst ), \
281 get_result_flags( inst ), \
287 #define EMIT_ARITH( OP, N ) \
289 i915_emit_arith( p, \
291 get_result_vector( p, inst ), \
292 get_result_flags( inst ), 0, \
293 (N<1)?0:src_vector( p, &inst->SrcReg[0], program), \
294 (N<2)?0:src_vector( p, &inst->SrcReg[1], program), \
295 (N<3)?0:src_vector( p, &inst->SrcReg[2], program)); \
298 #define EMIT_1ARG_ARITH( OP ) EMIT_ARITH( OP, 1 )
299 #define EMIT_2ARG_ARITH( OP ) EMIT_ARITH( OP, 2 )
300 #define EMIT_3ARG_ARITH( OP ) EMIT_ARITH( OP, 3 )
303 * TODO: consider moving this into core
305 static void calc_live_regs( struct i915_fragment_program
*p
)
307 const struct gl_fragment_program
*program
= p
->ctx
->FragmentProgram
._Current
;
308 GLuint regsUsed
= 0xffff0000;
311 for (i
= program
->Base
.NumInstructions
- 1; i
>= 0; i
--) {
312 struct prog_instruction
*inst
= &program
->Base
.Instructions
[i
];
313 int opArgs
= _mesa_num_inst_src_regs(inst
->Opcode
);
316 /* Register is written to: unmark as live for this and preceeding ops */
317 if (inst
->DstReg
.File
== PROGRAM_TEMPORARY
)
318 regsUsed
&= ~(1 << inst
->DstReg
.Index
);
320 for (a
= 0; a
< opArgs
; a
++) {
321 /* Register is read from: mark as live for this and preceeding ops */
322 if (inst
->SrcReg
[a
].File
== PROGRAM_TEMPORARY
)
323 regsUsed
|= 1 << inst
->SrcReg
[a
].Index
;
326 p
->usedRegs
[i
] = regsUsed
;
330 static GLuint
get_live_regs( struct i915_fragment_program
*p
,
331 const struct prog_instruction
*inst
)
333 const struct gl_fragment_program
*program
= p
->ctx
->FragmentProgram
._Current
;
334 GLuint nr
= inst
- program
->Base
.Instructions
;
336 return p
->usedRegs
[nr
];
340 /* Possible concerns:
342 * SIN, COS -- could use another taylor step?
343 * LIT -- results seem a little different to sw mesa
344 * LOG -- different to mesa on negative numbers, but this is conformant.
346 * Parse failures -- Mesa doesn't currently give a good indication
347 * internally whether a particular program string parsed or not. This
348 * can lead to confusion -- hopefully we cope with it ok now.
352 upload_program(struct i915_fragment_program
*p
)
354 const struct gl_fragment_program
*program
=
355 p
->ctx
->FragmentProgram
._Current
;
356 const struct prog_instruction
*inst
= program
->Base
.Instructions
;
358 if (INTEL_DEBUG
& DEBUG_WM
)
359 _mesa_print_program(&program
->Base
);
361 /* Is this a parse-failed program? Ensure a valid program is
362 * loaded, as the flagging of an error isn't sufficient to stop
363 * this being uploaded to hardware.
365 if (inst
[0].Opcode
== OPCODE_END
) {
366 GLuint tmp
= i915_get_utemp(p
);
369 UREG(REG_TYPE_OC
, 0),
370 A0_DEST_CHANNEL_ALL
, 0,
371 swizzle(tmp
, ONE
, ZERO
, ONE
, ONE
), 0, 0);
375 if (program
->Base
.NumInstructions
> I915_MAX_INSN
) {
376 i915_program_error(p
, "Exceeded max instructions (%d out of %d)",
377 program
->Base
.NumInstructions
, I915_MAX_INSN
);
381 /* Not always needed:
386 GLuint src0
, src1
, src2
, flags
;
387 GLuint tmp
= 0, dst
, consts0
= 0, consts1
= 0;
389 switch (inst
->Opcode
) {
391 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
394 get_result_vector(p
, inst
),
395 get_result_flags(inst
), 0,
396 src0
, negate(src0
, 1, 1, 1, 1), 0);
400 EMIT_2ARG_ARITH(A0_ADD
);
404 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
405 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
406 src2
= src_vector(p
, &inst
->SrcReg
[2], program
);
407 i915_emit_arith(p
, A0_CMP
, get_result_vector(p
, inst
), get_result_flags(inst
), 0, src0
, src2
, src1
); /* NOTE: order of src2, src1 */
411 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
412 tmp
= i915_get_utemp(p
);
413 consts0
= i915_emit_const4fv(p
, sin_quad_constants
[0]);
414 consts1
= i915_emit_const4fv(p
, sin_quad_constants
[1]);
416 /* Reduce range from repeating about [-pi,pi] to [-1,1] */
419 tmp
, A0_DEST_CHANNEL_X
, 0,
421 swizzle(consts1
, Z
, ZERO
, ZERO
, ZERO
), /* 1/(2pi) */
422 swizzle(consts0
, W
, ZERO
, ZERO
, ZERO
)); /* .75 */
424 i915_emit_arith(p
, A0_FRC
, tmp
, A0_DEST_CHANNEL_X
, 0, tmp
, 0, 0);
428 tmp
, A0_DEST_CHANNEL_X
, 0,
430 swizzle(consts0
, X
, ZERO
, ZERO
, ZERO
), /* 2 */
431 swizzle(consts0
, Y
, ZERO
, ZERO
, ZERO
)); /* -1 */
433 /* Compute COS with the same calculation used for SIN, but a
434 * different source range has been mapped to [-1,1] this time.
437 /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
440 tmp
, A0_DEST_CHANNEL_Y
, 0,
441 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
442 negate(swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
), 0, 1, 0, 0),
445 /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
448 tmp
, A0_DEST_CHANNEL_Y
, 0,
449 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
453 /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
456 tmp
, A0_DEST_CHANNEL_X
, 0,
458 swizzle(consts1
, X
, Y
, ZERO
, ZERO
),
461 /* tmp.x now contains a first approximation (y). Now, weight it
462 * against tmp.y**2 to get closer.
466 tmp
, A0_DEST_CHANNEL_Y
, 0,
467 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
468 negate(swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
), 0, 1, 0, 0),
471 /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
474 tmp
, A0_DEST_CHANNEL_Y
, 0,
475 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
476 swizzle(tmp
, ZERO
, Y
, ZERO
, ZERO
),
477 negate(swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
), 0, 1, 0, 0));
479 /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
482 get_result_vector(p
, inst
),
483 get_result_flags(inst
), 0,
484 swizzle(consts1
, W
, W
, W
, W
),
485 swizzle(tmp
, Y
, Y
, Y
, Y
),
486 swizzle(tmp
, X
, X
, X
, X
));
490 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
491 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
494 get_result_vector(p
, inst
),
495 get_result_flags(inst
), 0,
496 swizzle(src0
, X
, Y
, ZERO
, ZERO
),
497 swizzle(src1
, X
, Y
, ZERO
, ZERO
),
502 EMIT_2ARG_ARITH(A0_DP3
);
506 EMIT_2ARG_ARITH(A0_DP4
);
510 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
511 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
515 get_result_vector(p
, inst
),
516 get_result_flags(inst
), 0,
517 swizzle(src0
, X
, Y
, Z
, ONE
), src1
, 0);
521 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
522 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
524 /* result[0] = 1 * 1;
525 * result[1] = a[1] * b[1];
526 * result[2] = a[2] * 1;
527 * result[3] = 1 * b[3];
531 get_result_vector(p
, inst
),
532 get_result_flags(inst
), 0,
533 swizzle(src0
, ONE
, Y
, Z
, ONE
),
534 swizzle(src1
, ONE
, Y
, ONE
, W
), 0);
538 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
542 get_result_vector(p
, inst
),
543 get_result_flags(inst
), 0,
544 swizzle(src0
, X
, X
, X
, X
), 0, 0);
548 EMIT_1ARG_ARITH(A0_FLR
);
552 EMIT_1ARG_ARITH(A0_TRC
);
556 EMIT_1ARG_ARITH(A0_FRC
);
560 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
561 tmp
= i915_get_utemp(p
);
563 i915_emit_texld(p
, get_live_regs(p
, inst
),
564 tmp
, A0_DEST_CHANNEL_ALL
, /* use a dummy dest reg */
565 0, src0
, T0_TEXKILL
);
569 if (inst
->DstReg
.CondMask
== COND_TR
) {
570 tmp
= i915_get_utemp(p
);
572 i915_emit_texld(p
, get_live_regs(p
, inst
),
573 tmp
, A0_DEST_CHANNEL_ALL
,
574 0, /* use a dummy dest reg */
575 swizzle(tmp
, ONE
, ONE
, ONE
, ONE
), /* always */
579 i915_program_error(p
, "Unsupported KIL_NV condition code: %d",
580 inst
->DstReg
.CondMask
);
585 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
589 get_result_vector(p
, inst
),
590 get_result_flags(inst
), 0,
591 swizzle(src0
, X
, X
, X
, X
), 0, 0);
595 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
596 tmp
= i915_get_utemp(p
);
598 /* tmp = max( a.xyzw, a.00zw )
599 * XXX: Clamp tmp.w to -128..128
601 * tmp.y = tmp.w * tmp.y
603 * result = cmp (a.11-x1, a.1x01, a.1xy1 )
605 i915_emit_arith(p
, A0_MAX
, tmp
, A0_DEST_CHANNEL_ALL
, 0,
606 src0
, swizzle(src0
, ZERO
, ZERO
, Z
, W
), 0);
608 i915_emit_arith(p
, A0_LOG
, tmp
, A0_DEST_CHANNEL_Y
, 0,
609 swizzle(tmp
, Y
, Y
, Y
, Y
), 0, 0);
611 i915_emit_arith(p
, A0_MUL
, tmp
, A0_DEST_CHANNEL_Y
, 0,
612 swizzle(tmp
, ZERO
, Y
, ZERO
, ZERO
),
613 swizzle(tmp
, ZERO
, W
, ZERO
, ZERO
), 0);
615 i915_emit_arith(p
, A0_EXP
, tmp
, A0_DEST_CHANNEL_Y
, 0,
616 swizzle(tmp
, Y
, Y
, Y
, Y
), 0, 0);
618 i915_emit_arith(p
, A0_CMP
,
619 get_result_vector(p
, inst
),
620 get_result_flags(inst
), 0,
621 negate(swizzle(tmp
, ONE
, ONE
, X
, ONE
), 0, 0, 1, 0),
622 swizzle(tmp
, ONE
, X
, ZERO
, ONE
),
623 swizzle(tmp
, ONE
, X
, Y
, ONE
));
628 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
629 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
630 src2
= src_vector(p
, &inst
->SrcReg
[2], program
);
631 flags
= get_result_flags(inst
);
632 tmp
= i915_get_utemp(p
);
639 * result = (-c)*a + tmp
641 i915_emit_arith(p
, A0_MAD
, tmp
,
642 flags
& A0_DEST_CHANNEL_ALL
, 0, src1
, src0
, src2
);
644 i915_emit_arith(p
, A0_MAD
,
645 get_result_vector(p
, inst
),
646 flags
, 0, negate(src2
, 1, 1, 1, 1), src0
, tmp
);
650 EMIT_3ARG_ARITH(A0_MAD
);
654 EMIT_2ARG_ARITH(A0_MAX
);
658 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
659 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
660 tmp
= i915_get_utemp(p
);
661 flags
= get_result_flags(inst
);
665 tmp
, flags
& A0_DEST_CHANNEL_ALL
, 0,
666 negate(src0
, 1, 1, 1, 1),
667 negate(src1
, 1, 1, 1, 1), 0);
671 get_result_vector(p
, inst
),
672 flags
, 0, negate(tmp
, 1, 1, 1, 1), 0, 0);
676 EMIT_1ARG_ARITH(A0_MOV
);
680 EMIT_2ARG_ARITH(A0_MUL
);
684 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
685 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
686 tmp
= i915_get_utemp(p
);
687 flags
= get_result_flags(inst
);
689 /* XXX: masking on intermediate values, here and elsewhere.
693 tmp
, A0_DEST_CHANNEL_X
, 0,
694 swizzle(src0
, X
, X
, X
, X
), 0, 0);
696 i915_emit_arith(p
, A0_MUL
, tmp
, A0_DEST_CHANNEL_X
, 0, tmp
, src1
, 0);
701 get_result_vector(p
, inst
),
702 flags
, 0, swizzle(tmp
, X
, X
, X
, X
), 0, 0);
707 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
711 get_result_vector(p
, inst
),
712 get_result_flags(inst
), 0,
713 swizzle(src0
, X
, X
, X
, X
), 0, 0);
718 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
722 get_result_vector(p
, inst
),
723 get_result_flags(inst
), 0,
724 swizzle(src0
, X
, X
, X
, X
), 0, 0);
728 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
729 tmp
= i915_get_utemp(p
);
732 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
733 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
734 * t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x
735 * scs.x = DP4 t1, sin_constants
736 * t1 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1
737 * scs.y = DP4 t1, cos_constants
741 tmp
, A0_DEST_CHANNEL_XY
, 0,
742 swizzle(src0
, X
, X
, ONE
, ONE
),
743 swizzle(src0
, X
, ONE
, ONE
, ONE
), 0);
747 tmp
, A0_DEST_CHANNEL_ALL
, 0,
748 swizzle(tmp
, X
, Y
, X
, Y
),
749 swizzle(tmp
, X
, X
, ONE
, ONE
), 0);
751 if (inst
->DstReg
.WriteMask
& WRITEMASK_Y
) {
754 if (inst
->DstReg
.WriteMask
& WRITEMASK_X
)
755 tmp1
= i915_get_utemp(p
);
761 tmp1
, A0_DEST_CHANNEL_ALL
, 0,
762 swizzle(tmp
, X
, Y
, Y
, W
),
763 swizzle(tmp
, X
, Z
, ONE
, ONE
), 0);
767 get_result_vector(p
, inst
),
768 A0_DEST_CHANNEL_Y
, 0,
769 swizzle(tmp1
, W
, Z
, Y
, X
),
770 i915_emit_const4fv(p
, sin_constants
), 0);
773 if (inst
->DstReg
.WriteMask
& WRITEMASK_X
) {
776 tmp
, A0_DEST_CHANNEL_XYZ
, 0,
777 swizzle(tmp
, X
, X
, Z
, ONE
),
778 swizzle(tmp
, Z
, ONE
, ONE
, ONE
), 0);
782 get_result_vector(p
, inst
),
783 A0_DEST_CHANNEL_X
, 0,
784 swizzle(tmp
, ONE
, Z
, Y
, X
),
785 i915_emit_const4fv(p
, cos_constants
), 0);
790 tmp
= i915_get_utemp(p
);
791 flags
= get_result_flags(inst
);
792 dst
= get_result_vector(p
, inst
);
794 /* dst = src1 >= src2 */
799 src_vector(p
, &inst
->SrcReg
[0], program
),
800 src_vector(p
, &inst
->SrcReg
[1], program
),
802 /* tmp = src1 <= src2 */
807 negate(src_vector(p
, &inst
->SrcReg
[0], program
),
809 negate(src_vector(p
, &inst
->SrcReg
[1], program
),
812 /* dst = tmp && dst */
823 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
824 tmp
= i915_get_utemp(p
);
825 consts0
= i915_emit_const4fv(p
, sin_quad_constants
[0]);
826 consts1
= i915_emit_const4fv(p
, sin_quad_constants
[1]);
828 /* Reduce range from repeating about [-pi,pi] to [-1,1] */
831 tmp
, A0_DEST_CHANNEL_X
, 0,
833 swizzle(consts1
, Z
, ZERO
, ZERO
, ZERO
), /* 1/(2pi) */
834 swizzle(consts0
, Z
, ZERO
, ZERO
, ZERO
)); /* .5 */
836 i915_emit_arith(p
, A0_FRC
, tmp
, A0_DEST_CHANNEL_X
, 0, tmp
, 0, 0);
840 tmp
, A0_DEST_CHANNEL_X
, 0,
842 swizzle(consts0
, X
, ZERO
, ZERO
, ZERO
), /* 2 */
843 swizzle(consts0
, Y
, ZERO
, ZERO
, ZERO
)); /* -1 */
845 /* Compute sin using a quadratic and quartic. It gives continuity
846 * that repeating the Taylor series lacks every 2*pi, and has
849 * The idea was described at:
850 * http://www.devmaster.net/forums/showthread.php?t=5784
853 /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
856 tmp
, A0_DEST_CHANNEL_Y
, 0,
857 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
858 negate(swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
), 0, 1, 0, 0),
861 /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
864 tmp
, A0_DEST_CHANNEL_Y
, 0,
865 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
869 /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
872 tmp
, A0_DEST_CHANNEL_X
, 0,
874 swizzle(consts1
, X
, Y
, ZERO
, ZERO
),
877 /* tmp.x now contains a first approximation (y). Now, weight it
878 * against tmp.y**2 to get closer.
882 tmp
, A0_DEST_CHANNEL_Y
, 0,
883 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
884 negate(swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
), 0, 1, 0, 0),
887 /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
890 tmp
, A0_DEST_CHANNEL_Y
, 0,
891 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
892 swizzle(tmp
, ZERO
, Y
, ZERO
, ZERO
),
893 negate(swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
), 0, 1, 0, 0));
895 /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
898 get_result_vector(p
, inst
),
899 get_result_flags(inst
), 0,
900 swizzle(consts1
, W
, W
, W
, W
),
901 swizzle(tmp
, Y
, Y
, Y
, Y
),
902 swizzle(tmp
, X
, X
, X
, X
));
907 EMIT_2ARG_ARITH(A0_SGE
);
913 get_result_vector( p
, inst
),
914 get_result_flags( inst
), 0,
915 negate(src_vector( p
, &inst
->SrcReg
[0], program
),
917 negate(src_vector( p
, &inst
->SrcReg
[1], program
),
925 get_result_vector( p
, inst
),
926 get_result_flags( inst
), 0,
927 negate(src_vector( p
, &inst
->SrcReg
[0], program
),
929 negate(src_vector( p
, &inst
->SrcReg
[1], program
),
935 EMIT_2ARG_ARITH(A0_SLT
);
939 tmp
= i915_get_utemp(p
);
940 flags
= get_result_flags(inst
);
941 dst
= get_result_vector(p
, inst
);
943 /* dst = src1 < src2 */
948 src_vector(p
, &inst
->SrcReg
[0], program
),
949 src_vector(p
, &inst
->SrcReg
[1], program
),
951 /* tmp = src1 > src2 */
956 negate(src_vector(p
, &inst
->SrcReg
[0], program
),
958 negate(src_vector(p
, &inst
->SrcReg
[1], program
),
961 /* dst = tmp || dst */
965 flags
| A0_DEST_SATURATE
, 0,
972 dst
= get_result_vector(p
, inst
);
973 flags
= get_result_flags(inst
);
974 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
975 tmp
= i915_get_utemp(p
);
977 /* tmp = (src < 0.0) */
983 swizzle(src0
, ZERO
, ZERO
, ZERO
, ZERO
),
986 /* dst = (0.0 < src) */
991 swizzle(src0
, ZERO
, ZERO
, ZERO
, ZERO
),
995 /* dst = (src > 0.0) - (src < 0.0) */
1001 negate(tmp
, 1, 1, 1, 1),
1007 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
1008 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
1012 get_result_vector(p
, inst
),
1013 get_result_flags(inst
), 0,
1014 src0
, negate(src1
, 1, 1, 1, 1), 0);
1018 EMIT_1ARG_ARITH(A0_MOV
); /* extended swizzle handled natively */
1026 EMIT_TEX(T0_TEXLDB
);
1030 EMIT_TEX(T0_TEXLDP
);
1035 * result.x = src0.y * src1.z - src0.z * src1.y;
1036 * result.y = src0.z * src1.x - src0.x * src1.z;
1037 * result.z = src0.x * src1.y - src0.y * src1.x;
1040 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
1041 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
1042 tmp
= i915_get_utemp(p
);
1046 tmp
, A0_DEST_CHANNEL_ALL
, 0,
1047 swizzle(src0
, Z
, X
, Y
, ONE
),
1048 swizzle(src1
, Y
, Z
, X
, ONE
), 0);
1052 get_result_vector(p
, inst
),
1053 get_result_flags(inst
), 0,
1054 swizzle(src0
, Y
, Z
, X
, ONE
),
1055 swizzle(src1
, Z
, X
, Y
, ONE
),
1056 negate(tmp
, 1, 1, 1, 0));
1062 case OPCODE_BGNLOOP
:
1072 case OPCODE_ENDLOOP
:
1077 i915_program_error(p
, "Unsupported opcode: %s",
1078 _mesa_opcode_string(inst
->Opcode
));
1083 /* These opcodes are claimed as GLSL, NV_vp, and ARB_vp in
1084 * prog_instruction.h, but apparently GLSL doesn't ever emit them.
1085 * Instead, it translates to EX2 or LG2.
1089 /* These opcodes are claimed by GLSL in prog_instruction.h, but
1090 * only NV_vp/fp appears to emit them.
1093 i915_program_error(p
, "bad opcode: %s",
1094 _mesa_opcode_string(inst
->Opcode
));
1099 i915_release_utemps(p
);
1103 /* Rather than trying to intercept and jiggle depth writes during
1104 * emit, just move the value into its correct position at the end of
1108 fixup_depth_write(struct i915_fragment_program
*p
)
1110 if (p
->depth_written
) {
1111 GLuint depth
= UREG(REG_TYPE_OD
, 0);
1115 depth
, A0_DEST_CHANNEL_W
, 0,
1116 swizzle(depth
, X
, Y
, Z
, Z
), 0, 0);
1122 check_wpos(struct i915_fragment_program
*p
)
1124 GLuint inputs
= p
->FragProg
.Base
.InputsRead
;
1129 for (i
= 0; i
< p
->ctx
->Const
.MaxTextureCoordUnits
; i
++) {
1130 if (inputs
& (FRAG_BIT_TEX(i
) | FRAG_BIT_VAR(i
)))
1132 else if (inputs
& FRAG_BIT_WPOS
) {
1134 inputs
&= ~FRAG_BIT_WPOS
;
1138 if (inputs
& FRAG_BIT_WPOS
) {
1139 i915_program_error(p
, "No free texcoord for wpos value");
1145 translate_program(struct i915_fragment_program
*p
)
1147 struct i915_context
*i915
= I915_CONTEXT(p
->ctx
);
1149 if (INTEL_DEBUG
& DEBUG_WM
) {
1151 _mesa_print_program(&p
->ctx
->FragmentProgram
._Current
->Base
);
1155 i915_init_program(i915
, p
);
1158 fixup_depth_write(p
);
1159 i915_fini_program(p
);
1161 if (INTEL_DEBUG
& DEBUG_WM
) {
1163 i915_disassemble_program(i915
->state
.Program
, i915
->state
.ProgramSize
);
1171 track_params(struct i915_fragment_program
*p
)
1176 _mesa_load_state_parameters(p
->ctx
, p
->FragProg
.Base
.Parameters
);
1178 for (i
= 0; i
< p
->nr_params
; i
++) {
1179 GLint reg
= p
->param
[i
].reg
;
1180 COPY_4V(p
->constant
[reg
], p
->param
[i
].values
);
1183 p
->params_uptodate
= 1;
1184 p
->on_hardware
= 0; /* overkill */
1189 i915BindProgram(struct gl_context
* ctx
, GLenum target
, struct gl_program
*prog
)
1191 if (target
== GL_FRAGMENT_PROGRAM_ARB
) {
1192 struct i915_context
*i915
= I915_CONTEXT(ctx
);
1193 struct i915_fragment_program
*p
= (struct i915_fragment_program
*) prog
;
1195 if (i915
->current_program
== p
)
1198 if (i915
->current_program
) {
1199 i915
->current_program
->on_hardware
= 0;
1200 i915
->current_program
->params_uptodate
= 0;
1203 i915
->current_program
= p
;
1205 assert(p
->on_hardware
== 0);
1206 assert(p
->params_uptodate
== 0);
1211 static struct gl_program
*
1212 i915NewProgram(struct gl_context
* ctx
, GLenum target
, GLuint id
)
1215 case GL_VERTEX_PROGRAM_ARB
:
1216 return _mesa_init_vertex_program(ctx
, CALLOC_STRUCT(gl_vertex_program
),
1219 case GL_FRAGMENT_PROGRAM_ARB
:{
1220 struct i915_fragment_program
*prog
=
1221 CALLOC_STRUCT(i915_fragment_program
);
1223 i915_init_program(I915_CONTEXT(ctx
), prog
);
1225 return _mesa_init_fragment_program(ctx
, &prog
->FragProg
,
1235 return _mesa_new_program(ctx
, target
, id
);
1240 i915DeleteProgram(struct gl_context
* ctx
, struct gl_program
*prog
)
1242 if (prog
->Target
== GL_FRAGMENT_PROGRAM_ARB
) {
1243 struct i915_context
*i915
= I915_CONTEXT(ctx
);
1244 struct i915_fragment_program
*p
= (struct i915_fragment_program
*) prog
;
1246 if (i915
->current_program
== p
)
1247 i915
->current_program
= 0;
1250 _mesa_delete_program(ctx
, prog
);
1255 i915IsProgramNative(struct gl_context
* ctx
, GLenum target
, struct gl_program
*prog
)
1257 if (target
== GL_FRAGMENT_PROGRAM_ARB
) {
1258 struct i915_fragment_program
*p
= (struct i915_fragment_program
*) prog
;
1261 translate_program(p
);
1270 i915ProgramStringNotify(struct gl_context
* ctx
,
1271 GLenum target
, struct gl_program
*prog
)
1273 if (target
== GL_FRAGMENT_PROGRAM_ARB
) {
1274 struct i915_fragment_program
*p
= (struct i915_fragment_program
*) prog
;
1277 /* Hack: make sure fog is correctly enabled according to this
1278 * fragment program's fog options.
1280 if (p
->FragProg
.FogOption
) {
1281 /* add extra instructions to do fog, then turn off FogOption field */
1282 _mesa_append_fog_code(ctx
, &p
->FragProg
);
1283 p
->FragProg
.FogOption
= GL_NONE
;
1287 (void) _tnl_program_string(ctx
, target
, prog
);
1289 /* XXX check if program is legal, within limits */
1294 i915_update_program(struct gl_context
*ctx
)
1296 struct intel_context
*intel
= intel_context(ctx
);
1297 struct i915_context
*i915
= i915_context(&intel
->ctx
);
1298 struct i915_fragment_program
*fp
=
1299 (struct i915_fragment_program
*) ctx
->FragmentProgram
._Current
;
1301 if (i915
->current_program
!= fp
) {
1302 if (i915
->current_program
) {
1303 i915
->current_program
->on_hardware
= 0;
1304 i915
->current_program
->params_uptodate
= 0;
1307 i915
->current_program
= fp
;
1310 if (!fp
->translated
)
1311 translate_program(fp
);
1313 FALLBACK(&i915
->intel
, I915_FALLBACK_PROGRAM
, fp
->error
);
1317 i915ValidateFragmentProgram(struct i915_context
*i915
)
1319 struct gl_context
*ctx
= &i915
->intel
.ctx
;
1320 struct intel_context
*intel
= intel_context(ctx
);
1321 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
1322 struct vertex_buffer
*VB
= &tnl
->vb
;
1324 struct i915_fragment_program
*p
=
1325 (struct i915_fragment_program
*) ctx
->FragmentProgram
._Current
;
1327 const GLuint inputsRead
= p
->FragProg
.Base
.InputsRead
;
1328 GLuint s4
= i915
->state
.Ctx
[I915_CTXREG_LIS4
] & ~S4_VFMT_MASK
;
1329 GLuint s2
= S2_TEXCOORD_NONE
;
1334 VB
->AttribPtr
[VERT_ATTRIB_POS
] = VB
->NdcPtr
;
1337 translate_program(p
);
1339 intel
->vertex_attr_count
= 0;
1340 intel
->wpos_offset
= 0;
1341 intel
->wpos_size
= 0;
1342 intel
->coloroffset
= 0;
1343 intel
->specoffset
= 0;
1345 if (inputsRead
& FRAG_BITS_TEX_ANY
) {
1346 EMIT_ATTR(_TNL_ATTRIB_POS
, EMIT_4F_VIEWPORT
, S4_VFMT_XYZW
, 16);
1349 EMIT_ATTR(_TNL_ATTRIB_POS
, EMIT_3F_VIEWPORT
, S4_VFMT_XYZ
, 12);
1352 if (inputsRead
& FRAG_BIT_COL0
) {
1353 intel
->coloroffset
= offset
/ 4;
1354 EMIT_ATTR(_TNL_ATTRIB_COLOR0
, EMIT_4UB_4F_BGRA
, S4_VFMT_COLOR
, 4);
1357 if (inputsRead
& FRAG_BIT_COL1
) {
1358 intel
->specoffset
= offset
/ 4;
1359 EMIT_ATTR(_TNL_ATTRIB_COLOR1
, EMIT_4UB_4F_BGRA
, S4_VFMT_SPEC_FOG
, 4);
1362 if ((inputsRead
& FRAG_BIT_FOGC
) || i915
->vertex_fog
!= I915_FOG_NONE
) {
1363 EMIT_ATTR(_TNL_ATTRIB_FOG
, EMIT_1F
, S4_VFMT_FOG_PARAM
, 4);
1366 for (i
= 0; i
< p
->ctx
->Const
.MaxTextureCoordUnits
; i
++) {
1367 if (inputsRead
& FRAG_BIT_TEX(i
)) {
1368 int sz
= VB
->AttribPtr
[_TNL_ATTRIB_TEX0
+ i
]->size
;
1370 s2
&= ~S2_TEXCOORD_FMT(i
, S2_TEXCOORD_FMT0_MASK
);
1371 s2
|= S2_TEXCOORD_FMT(i
, SZ_TO_HW(sz
));
1373 EMIT_ATTR(_TNL_ATTRIB_TEX0
+ i
, EMIT_SZ(sz
), 0, sz
* 4);
1375 else if (inputsRead
& FRAG_BIT_VAR(i
)) {
1376 int sz
= VB
->AttribPtr
[_TNL_ATTRIB_GENERIC0
+ i
]->size
;
1378 s2
&= ~S2_TEXCOORD_FMT(i
, S2_TEXCOORD_FMT0_MASK
);
1379 s2
|= S2_TEXCOORD_FMT(i
, SZ_TO_HW(sz
));
1381 EMIT_ATTR(_TNL_ATTRIB_GENERIC0
+ i
, EMIT_SZ(sz
), 0, sz
* 4);
1383 else if (i
== p
->wpos_tex
) {
1385 /* If WPOS is required, duplicate the XYZ position data in an
1386 * unused texture coordinate:
1388 s2
&= ~S2_TEXCOORD_FMT(i
, S2_TEXCOORD_FMT0_MASK
);
1389 s2
|= S2_TEXCOORD_FMT(i
, SZ_TO_HW(3));
1391 intel
->wpos_offset
= offset
;
1392 intel
->wpos_size
= 3 * sizeof(GLuint
);
1394 EMIT_PAD(intel
->wpos_size
);
1398 if (s2
!= i915
->state
.Ctx
[I915_CTXREG_LIS2
] ||
1399 s4
!= i915
->state
.Ctx
[I915_CTXREG_LIS4
]) {
1402 I915_STATECHANGE(i915
, I915_UPLOAD_CTX
);
1404 /* Must do this *after* statechange, so as not to affect
1405 * buffered vertices reliant on the old state:
1407 intel
->vertex_size
= _tnl_install_attrs(&intel
->ctx
,
1408 intel
->vertex_attrs
,
1409 intel
->vertex_attr_count
,
1410 intel
->ViewportMatrix
.m
, 0);
1412 intel
->vertex_size
>>= 2;
1414 i915
->state
.Ctx
[I915_CTXREG_LIS2
] = s2
;
1415 i915
->state
.Ctx
[I915_CTXREG_LIS4
] = s4
;
1417 k
= intel
->vtbl
.check_vertex_size(intel
, intel
->vertex_size
);
1421 if (!p
->params_uptodate
)
1424 if (!p
->on_hardware
)
1425 i915_upload_program(i915
, p
);
1429 i915InitFragProgFuncs(struct dd_function_table
*functions
)
1431 functions
->BindProgram
= i915BindProgram
;
1432 functions
->NewProgram
= i915NewProgram
;
1433 functions
->DeleteProgram
= i915DeleteProgram
;
1434 functions
->IsProgramNative
= i915IsProgramNative
;
1435 functions
->ProgramStringNotify
= i915ProgramStringNotify
;