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->FragProg; \
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
->FragProg
;
308 GLuint regsUsed
= 0xffff0000;
309 uint8_t live_components
[16] = { 0, };
312 for (i
= program
->Base
.NumInstructions
- 1; i
>= 0; i
--) {
313 struct prog_instruction
*inst
= &program
->Base
.Instructions
[i
];
314 int opArgs
= _mesa_num_inst_src_regs(inst
->Opcode
);
317 /* Register is written to: unmark as live for this and preceeding ops */
318 if (inst
->DstReg
.File
== PROGRAM_TEMPORARY
) {
319 live_components
[inst
->DstReg
.Index
] &= ~inst
->DstReg
.WriteMask
;
320 if (live_components
[inst
->DstReg
.Index
] == 0)
321 regsUsed
&= ~(1 << inst
->DstReg
.Index
);
324 for (a
= 0; a
< opArgs
; a
++) {
325 /* Register is read from: mark as live for this and preceeding ops */
326 if (inst
->SrcReg
[a
].File
== PROGRAM_TEMPORARY
) {
329 regsUsed
|= 1 << inst
->SrcReg
[a
].Index
;
331 for (c
= 0; c
< 4; c
++) {
332 const unsigned field
= GET_SWZ(inst
->SrcReg
[a
].Swizzle
, c
);
334 if (field
<= SWIZZLE_W
)
335 live_components
[inst
->SrcReg
[a
].Index
] |= (1U << field
);
340 p
->usedRegs
[i
] = regsUsed
;
344 static GLuint
get_live_regs( struct i915_fragment_program
*p
,
345 const struct prog_instruction
*inst
)
347 const struct gl_fragment_program
*program
= &p
->FragProg
;
348 GLuint nr
= inst
- program
->Base
.Instructions
;
350 return p
->usedRegs
[nr
];
354 /* Possible concerns:
356 * SIN, COS -- could use another taylor step?
357 * LIT -- results seem a little different to sw mesa
358 * LOG -- different to mesa on negative numbers, but this is conformant.
360 * Parse failures -- Mesa doesn't currently give a good indication
361 * internally whether a particular program string parsed or not. This
362 * can lead to confusion -- hopefully we cope with it ok now.
366 upload_program(struct i915_fragment_program
*p
)
368 const struct gl_fragment_program
*program
= &p
->FragProg
;
369 const struct prog_instruction
*inst
= program
->Base
.Instructions
;
371 if (INTEL_DEBUG
& DEBUG_WM
)
372 _mesa_print_program(&program
->Base
);
374 /* Is this a parse-failed program? Ensure a valid program is
375 * loaded, as the flagging of an error isn't sufficient to stop
376 * this being uploaded to hardware.
378 if (inst
[0].Opcode
== OPCODE_END
) {
379 GLuint tmp
= i915_get_utemp(p
);
382 UREG(REG_TYPE_OC
, 0),
383 A0_DEST_CHANNEL_ALL
, 0,
384 swizzle(tmp
, ONE
, ZERO
, ONE
, ONE
), 0, 0);
388 if (program
->Base
.NumInstructions
> I915_MAX_INSN
) {
389 i915_program_error(p
, "Exceeded max instructions (%d out of %d)",
390 program
->Base
.NumInstructions
, I915_MAX_INSN
);
394 /* Not always needed:
399 GLuint src0
, src1
, src2
, flags
;
400 GLuint tmp
= 0, dst
, consts0
= 0, consts1
= 0;
402 switch (inst
->Opcode
) {
404 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
407 get_result_vector(p
, inst
),
408 get_result_flags(inst
), 0,
409 src0
, negate(src0
, 1, 1, 1, 1), 0);
413 EMIT_2ARG_ARITH(A0_ADD
);
417 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
418 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
419 src2
= src_vector(p
, &inst
->SrcReg
[2], program
);
420 i915_emit_arith(p
, A0_CMP
, get_result_vector(p
, inst
), get_result_flags(inst
), 0, src0
, src2
, src1
); /* NOTE: order of src2, src1 */
424 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
425 tmp
= i915_get_utemp(p
);
426 consts0
= i915_emit_const4fv(p
, sin_quad_constants
[0]);
427 consts1
= i915_emit_const4fv(p
, sin_quad_constants
[1]);
429 /* Reduce range from repeating about [-pi,pi] to [-1,1] */
432 tmp
, A0_DEST_CHANNEL_X
, 0,
434 swizzle(consts1
, Z
, ZERO
, ZERO
, ZERO
), /* 1/(2pi) */
435 swizzle(consts0
, W
, ZERO
, ZERO
, ZERO
)); /* .75 */
437 i915_emit_arith(p
, A0_FRC
, tmp
, A0_DEST_CHANNEL_X
, 0, tmp
, 0, 0);
441 tmp
, A0_DEST_CHANNEL_X
, 0,
443 swizzle(consts0
, X
, ZERO
, ZERO
, ZERO
), /* 2 */
444 swizzle(consts0
, Y
, ZERO
, ZERO
, ZERO
)); /* -1 */
446 /* Compute COS with the same calculation used for SIN, but a
447 * different source range has been mapped to [-1,1] this time.
450 /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
453 tmp
, A0_DEST_CHANNEL_Y
, 0,
454 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
455 negate(swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
), 0, 1, 0, 0),
458 /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
461 tmp
, A0_DEST_CHANNEL_Y
, 0,
462 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
466 /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
469 tmp
, A0_DEST_CHANNEL_X
, 0,
471 swizzle(consts1
, X
, Y
, ZERO
, ZERO
),
474 /* tmp.x now contains a first approximation (y). Now, weight it
475 * against tmp.y**2 to get closer.
479 tmp
, A0_DEST_CHANNEL_Y
, 0,
480 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
481 negate(swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
), 0, 1, 0, 0),
484 /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
487 tmp
, A0_DEST_CHANNEL_Y
, 0,
488 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
489 swizzle(tmp
, ZERO
, Y
, ZERO
, ZERO
),
490 negate(swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
), 0, 1, 0, 0));
492 /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
495 get_result_vector(p
, inst
),
496 get_result_flags(inst
), 0,
497 swizzle(consts1
, W
, W
, W
, W
),
498 swizzle(tmp
, Y
, Y
, Y
, Y
),
499 swizzle(tmp
, X
, X
, X
, X
));
503 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
504 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
507 get_result_vector(p
, inst
),
508 get_result_flags(inst
), 0,
509 swizzle(src0
, X
, Y
, ZERO
, ZERO
),
510 swizzle(src1
, X
, Y
, ZERO
, ZERO
),
515 EMIT_2ARG_ARITH(A0_DP3
);
519 EMIT_2ARG_ARITH(A0_DP4
);
523 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
524 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
528 get_result_vector(p
, inst
),
529 get_result_flags(inst
), 0,
530 swizzle(src0
, X
, Y
, Z
, ONE
), src1
, 0);
534 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
535 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
537 /* result[0] = 1 * 1;
538 * result[1] = a[1] * b[1];
539 * result[2] = a[2] * 1;
540 * result[3] = 1 * b[3];
544 get_result_vector(p
, inst
),
545 get_result_flags(inst
), 0,
546 swizzle(src0
, ONE
, Y
, Z
, ONE
),
547 swizzle(src1
, ONE
, Y
, ONE
, W
), 0);
551 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
555 get_result_vector(p
, inst
),
556 get_result_flags(inst
), 0,
557 swizzle(src0
, X
, X
, X
, X
), 0, 0);
561 EMIT_1ARG_ARITH(A0_FLR
);
565 EMIT_1ARG_ARITH(A0_TRC
);
569 EMIT_1ARG_ARITH(A0_FRC
);
573 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
574 tmp
= i915_get_utemp(p
);
576 i915_emit_texld(p
, get_live_regs(p
, inst
),
577 tmp
, A0_DEST_CHANNEL_ALL
, /* use a dummy dest reg */
578 0, src0
, T0_TEXKILL
);
582 if (inst
->DstReg
.CondMask
== COND_TR
) {
583 tmp
= i915_get_utemp(p
);
585 /* The KIL instruction discards the fragment if any component of
586 * the source is < 0. Emit an immediate operand of {-1}.xywz.
588 i915_emit_texld(p
, get_live_regs(p
, inst
),
589 tmp
, A0_DEST_CHANNEL_ALL
,
590 0, /* use a dummy dest reg */
591 negate(swizzle(tmp
, ONE
, ONE
, ONE
, ONE
),
596 i915_program_error(p
, "Unsupported KIL_NV condition code: %d",
597 inst
->DstReg
.CondMask
);
602 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
606 get_result_vector(p
, inst
),
607 get_result_flags(inst
), 0,
608 swizzle(src0
, X
, X
, X
, X
), 0, 0);
612 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
613 tmp
= i915_get_utemp(p
);
615 /* tmp = max( a.xyzw, a.00zw )
616 * XXX: Clamp tmp.w to -128..128
618 * tmp.y = tmp.w * tmp.y
620 * result = cmp (a.11-x1, a.1x01, a.1xy1 )
622 i915_emit_arith(p
, A0_MAX
, tmp
, A0_DEST_CHANNEL_ALL
, 0,
623 src0
, swizzle(src0
, ZERO
, ZERO
, Z
, W
), 0);
625 i915_emit_arith(p
, A0_LOG
, tmp
, A0_DEST_CHANNEL_Y
, 0,
626 swizzle(tmp
, Y
, Y
, Y
, Y
), 0, 0);
628 i915_emit_arith(p
, A0_MUL
, tmp
, A0_DEST_CHANNEL_Y
, 0,
629 swizzle(tmp
, ZERO
, Y
, ZERO
, ZERO
),
630 swizzle(tmp
, ZERO
, W
, ZERO
, ZERO
), 0);
632 i915_emit_arith(p
, A0_EXP
, tmp
, A0_DEST_CHANNEL_Y
, 0,
633 swizzle(tmp
, Y
, Y
, Y
, Y
), 0, 0);
635 i915_emit_arith(p
, A0_CMP
,
636 get_result_vector(p
, inst
),
637 get_result_flags(inst
), 0,
638 negate(swizzle(tmp
, ONE
, ONE
, X
, ONE
), 0, 0, 1, 0),
639 swizzle(tmp
, ONE
, X
, ZERO
, ONE
),
640 swizzle(tmp
, ONE
, X
, Y
, ONE
));
645 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
646 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
647 src2
= src_vector(p
, &inst
->SrcReg
[2], program
);
648 flags
= get_result_flags(inst
);
649 tmp
= i915_get_utemp(p
);
656 * result = (-c)*a + tmp
658 i915_emit_arith(p
, A0_MAD
, tmp
,
659 flags
& A0_DEST_CHANNEL_ALL
, 0, src1
, src0
, src2
);
661 i915_emit_arith(p
, A0_MAD
,
662 get_result_vector(p
, inst
),
663 flags
, 0, negate(src2
, 1, 1, 1, 1), src0
, tmp
);
667 EMIT_3ARG_ARITH(A0_MAD
);
671 EMIT_2ARG_ARITH(A0_MAX
);
675 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
676 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
677 tmp
= i915_get_utemp(p
);
678 flags
= get_result_flags(inst
);
682 tmp
, flags
& A0_DEST_CHANNEL_ALL
, 0,
683 negate(src0
, 1, 1, 1, 1),
684 negate(src1
, 1, 1, 1, 1), 0);
688 get_result_vector(p
, inst
),
689 flags
, 0, negate(tmp
, 1, 1, 1, 1), 0, 0);
693 EMIT_1ARG_ARITH(A0_MOV
);
697 EMIT_2ARG_ARITH(A0_MUL
);
701 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
702 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
703 tmp
= i915_get_utemp(p
);
704 flags
= get_result_flags(inst
);
706 /* XXX: masking on intermediate values, here and elsewhere.
710 tmp
, A0_DEST_CHANNEL_X
, 0,
711 swizzle(src0
, X
, X
, X
, X
), 0, 0);
713 i915_emit_arith(p
, A0_MUL
, tmp
, A0_DEST_CHANNEL_X
, 0, tmp
, src1
, 0);
718 get_result_vector(p
, inst
),
719 flags
, 0, swizzle(tmp
, X
, X
, X
, X
), 0, 0);
724 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
728 get_result_vector(p
, inst
),
729 get_result_flags(inst
), 0,
730 swizzle(src0
, X
, X
, X
, X
), 0, 0);
735 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
739 get_result_vector(p
, inst
),
740 get_result_flags(inst
), 0,
741 swizzle(src0
, X
, X
, X
, X
), 0, 0);
745 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
746 tmp
= i915_get_utemp(p
);
749 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
750 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
751 * t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x
752 * scs.x = DP4 t1, sin_constants
753 * t1 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1
754 * scs.y = DP4 t1, cos_constants
758 tmp
, A0_DEST_CHANNEL_XY
, 0,
759 swizzle(src0
, X
, X
, ONE
, ONE
),
760 swizzle(src0
, X
, ONE
, ONE
, ONE
), 0);
764 tmp
, A0_DEST_CHANNEL_ALL
, 0,
765 swizzle(tmp
, X
, Y
, X
, Y
),
766 swizzle(tmp
, X
, X
, ONE
, ONE
), 0);
768 if (inst
->DstReg
.WriteMask
& WRITEMASK_Y
) {
771 if (inst
->DstReg
.WriteMask
& WRITEMASK_X
)
772 tmp1
= i915_get_utemp(p
);
778 tmp1
, A0_DEST_CHANNEL_ALL
, 0,
779 swizzle(tmp
, X
, Y
, Y
, W
),
780 swizzle(tmp
, X
, Z
, ONE
, ONE
), 0);
784 get_result_vector(p
, inst
),
785 A0_DEST_CHANNEL_Y
, 0,
786 swizzle(tmp1
, W
, Z
, Y
, X
),
787 i915_emit_const4fv(p
, sin_constants
), 0);
790 if (inst
->DstReg
.WriteMask
& WRITEMASK_X
) {
793 tmp
, A0_DEST_CHANNEL_XYZ
, 0,
794 swizzle(tmp
, X
, X
, Z
, ONE
),
795 swizzle(tmp
, Z
, ONE
, ONE
, ONE
), 0);
799 get_result_vector(p
, inst
),
800 A0_DEST_CHANNEL_X
, 0,
801 swizzle(tmp
, ONE
, Z
, Y
, X
),
802 i915_emit_const4fv(p
, cos_constants
), 0);
807 tmp
= i915_get_utemp(p
);
808 flags
= get_result_flags(inst
);
809 dst
= get_result_vector(p
, inst
);
811 /* tmp = src1 >= src2 */
816 src_vector(p
, &inst
->SrcReg
[0], program
),
817 src_vector(p
, &inst
->SrcReg
[1], program
),
819 /* dst = src1 <= src2 */
824 negate(src_vector(p
, &inst
->SrcReg
[0], program
),
826 negate(src_vector(p
, &inst
->SrcReg
[1], program
),
829 /* dst = tmp && dst */
840 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
841 tmp
= i915_get_utemp(p
);
842 consts0
= i915_emit_const4fv(p
, sin_quad_constants
[0]);
843 consts1
= i915_emit_const4fv(p
, sin_quad_constants
[1]);
845 /* Reduce range from repeating about [-pi,pi] to [-1,1] */
848 tmp
, A0_DEST_CHANNEL_X
, 0,
850 swizzle(consts1
, Z
, ZERO
, ZERO
, ZERO
), /* 1/(2pi) */
851 swizzle(consts0
, Z
, ZERO
, ZERO
, ZERO
)); /* .5 */
853 i915_emit_arith(p
, A0_FRC
, tmp
, A0_DEST_CHANNEL_X
, 0, tmp
, 0, 0);
857 tmp
, A0_DEST_CHANNEL_X
, 0,
859 swizzle(consts0
, X
, ZERO
, ZERO
, ZERO
), /* 2 */
860 swizzle(consts0
, Y
, ZERO
, ZERO
, ZERO
)); /* -1 */
862 /* Compute sin using a quadratic and quartic. It gives continuity
863 * that repeating the Taylor series lacks every 2*pi, and has
866 * The idea was described at:
867 * http://www.devmaster.net/forums/showthread.php?t=5784
870 /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
873 tmp
, A0_DEST_CHANNEL_Y
, 0,
874 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
875 negate(swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
), 0, 1, 0, 0),
878 /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
881 tmp
, A0_DEST_CHANNEL_Y
, 0,
882 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
886 /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
889 tmp
, A0_DEST_CHANNEL_X
, 0,
891 swizzle(consts1
, X
, Y
, ZERO
, ZERO
),
894 /* tmp.x now contains a first approximation (y). Now, weight it
895 * against tmp.y**2 to get closer.
899 tmp
, A0_DEST_CHANNEL_Y
, 0,
900 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
901 negate(swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
), 0, 1, 0, 0),
904 /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
907 tmp
, A0_DEST_CHANNEL_Y
, 0,
908 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
909 swizzle(tmp
, ZERO
, Y
, ZERO
, ZERO
),
910 negate(swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
), 0, 1, 0, 0));
912 /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
915 get_result_vector(p
, inst
),
916 get_result_flags(inst
), 0,
917 swizzle(consts1
, W
, W
, W
, W
),
918 swizzle(tmp
, Y
, Y
, Y
, Y
),
919 swizzle(tmp
, X
, X
, X
, X
));
924 EMIT_2ARG_ARITH(A0_SGE
);
930 get_result_vector( p
, inst
),
931 get_result_flags( inst
), 0,
932 negate(src_vector( p
, &inst
->SrcReg
[0], program
),
934 negate(src_vector( p
, &inst
->SrcReg
[1], program
),
942 get_result_vector( p
, inst
),
943 get_result_flags( inst
), 0,
944 negate(src_vector( p
, &inst
->SrcReg
[0], program
),
946 negate(src_vector( p
, &inst
->SrcReg
[1], program
),
952 EMIT_2ARG_ARITH(A0_SLT
);
956 tmp
= i915_get_utemp(p
);
957 flags
= get_result_flags(inst
);
958 dst
= get_result_vector(p
, inst
);
960 /* tmp = src1 < src2 */
965 src_vector(p
, &inst
->SrcReg
[0], program
),
966 src_vector(p
, &inst
->SrcReg
[1], program
),
968 /* dst = src1 > src2 */
973 negate(src_vector(p
, &inst
->SrcReg
[0], program
),
975 negate(src_vector(p
, &inst
->SrcReg
[1], program
),
978 /* dst = tmp || dst */
982 flags
| A0_DEST_SATURATE
, 0,
989 dst
= get_result_vector(p
, inst
);
990 flags
= get_result_flags(inst
);
991 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
992 tmp
= i915_get_utemp(p
);
994 /* tmp = (src < 0.0) */
1000 swizzle(src0
, ZERO
, ZERO
, ZERO
, ZERO
),
1003 /* dst = (0.0 < src) */
1008 swizzle(src0
, ZERO
, ZERO
, ZERO
, ZERO
),
1012 /* dst = (src > 0.0) - (src < 0.0) */
1018 negate(tmp
, 1, 1, 1, 1),
1024 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
1025 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
1029 get_result_vector(p
, inst
),
1030 get_result_flags(inst
), 0,
1031 src0
, negate(src1
, 1, 1, 1, 1), 0);
1035 EMIT_1ARG_ARITH(A0_MOV
); /* extended swizzle handled natively */
1043 EMIT_TEX(T0_TEXLDB
);
1047 EMIT_TEX(T0_TEXLDP
);
1052 * result.x = src0.y * src1.z - src0.z * src1.y;
1053 * result.y = src0.z * src1.x - src0.x * src1.z;
1054 * result.z = src0.x * src1.y - src0.y * src1.x;
1057 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
1058 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
1059 tmp
= i915_get_utemp(p
);
1063 tmp
, A0_DEST_CHANNEL_ALL
, 0,
1064 swizzle(src0
, Z
, X
, Y
, ONE
),
1065 swizzle(src1
, Y
, Z
, X
, ONE
), 0);
1069 get_result_vector(p
, inst
),
1070 get_result_flags(inst
), 0,
1071 swizzle(src0
, Y
, Z
, X
, ONE
),
1072 swizzle(src1
, Z
, X
, Y
, ONE
),
1073 negate(tmp
, 1, 1, 1, 0));
1079 case OPCODE_BGNLOOP
:
1089 case OPCODE_ENDLOOP
:
1094 i915_program_error(p
, "Unsupported opcode: %s",
1095 _mesa_opcode_string(inst
->Opcode
));
1100 /* These opcodes are claimed as GLSL, NV_vp, and ARB_vp in
1101 * prog_instruction.h, but apparently GLSL doesn't ever emit them.
1102 * Instead, it translates to EX2 or LG2.
1106 /* These opcodes are claimed by GLSL in prog_instruction.h, but
1107 * only NV_vp/fp appears to emit them.
1110 i915_program_error(p
, "bad opcode: %s",
1111 _mesa_opcode_string(inst
->Opcode
));
1116 i915_release_utemps(p
);
1120 /* Rather than trying to intercept and jiggle depth writes during
1121 * emit, just move the value into its correct position at the end of
1125 fixup_depth_write(struct i915_fragment_program
*p
)
1127 if (p
->depth_written
) {
1128 GLuint depth
= UREG(REG_TYPE_OD
, 0);
1132 depth
, A0_DEST_CHANNEL_W
, 0,
1133 swizzle(depth
, X
, Y
, Z
, Z
), 0, 0);
1139 check_wpos(struct i915_fragment_program
*p
)
1141 GLuint inputs
= p
->FragProg
.Base
.InputsRead
;
1146 for (i
= 0; i
< p
->ctx
->Const
.MaxTextureCoordUnits
; i
++) {
1147 if (inputs
& (FRAG_BIT_TEX(i
) | FRAG_BIT_VAR(i
)))
1149 else if (inputs
& FRAG_BIT_WPOS
) {
1151 inputs
&= ~FRAG_BIT_WPOS
;
1155 if (inputs
& FRAG_BIT_WPOS
) {
1156 i915_program_error(p
, "No free texcoord for wpos value");
1162 translate_program(struct i915_fragment_program
*p
)
1164 struct i915_context
*i915
= I915_CONTEXT(p
->ctx
);
1166 if (INTEL_DEBUG
& DEBUG_WM
) {
1168 _mesa_print_program(&p
->FragProg
.Base
);
1172 i915_init_program(i915
, p
);
1175 fixup_depth_write(p
);
1176 i915_fini_program(p
);
1183 track_params(struct i915_fragment_program
*p
)
1188 _mesa_load_state_parameters(p
->ctx
, p
->FragProg
.Base
.Parameters
);
1190 for (i
= 0; i
< p
->nr_params
; i
++) {
1191 GLint reg
= p
->param
[i
].reg
;
1192 COPY_4V(p
->constant
[reg
], p
->param
[i
].values
);
1195 p
->params_uptodate
= 1;
1196 p
->on_hardware
= 0; /* overkill */
1201 i915BindProgram(struct gl_context
* ctx
, GLenum target
, struct gl_program
*prog
)
1203 if (target
== GL_FRAGMENT_PROGRAM_ARB
) {
1204 struct i915_context
*i915
= I915_CONTEXT(ctx
);
1205 struct i915_fragment_program
*p
= (struct i915_fragment_program
*) prog
;
1207 if (i915
->current_program
== p
)
1210 if (i915
->current_program
) {
1211 i915
->current_program
->on_hardware
= 0;
1212 i915
->current_program
->params_uptodate
= 0;
1215 i915
->current_program
= p
;
1217 assert(p
->on_hardware
== 0);
1218 assert(p
->params_uptodate
== 0);
1223 static struct gl_program
*
1224 i915NewProgram(struct gl_context
* ctx
, GLenum target
, GLuint id
)
1227 case GL_VERTEX_PROGRAM_ARB
:
1228 return _mesa_init_vertex_program(ctx
, CALLOC_STRUCT(gl_vertex_program
),
1231 case GL_FRAGMENT_PROGRAM_ARB
:{
1232 struct i915_fragment_program
*prog
=
1233 CALLOC_STRUCT(i915_fragment_program
);
1235 i915_init_program(I915_CONTEXT(ctx
), prog
);
1237 return _mesa_init_fragment_program(ctx
, &prog
->FragProg
,
1247 return _mesa_new_program(ctx
, target
, id
);
1252 i915DeleteProgram(struct gl_context
* ctx
, struct gl_program
*prog
)
1254 if (prog
->Target
== GL_FRAGMENT_PROGRAM_ARB
) {
1255 struct i915_context
*i915
= I915_CONTEXT(ctx
);
1256 struct i915_fragment_program
*p
= (struct i915_fragment_program
*) prog
;
1258 if (i915
->current_program
== p
)
1259 i915
->current_program
= 0;
1262 _mesa_delete_program(ctx
, prog
);
1267 i915IsProgramNative(struct gl_context
* ctx
, GLenum target
, struct gl_program
*prog
)
1269 if (target
== GL_FRAGMENT_PROGRAM_ARB
) {
1270 struct i915_fragment_program
*p
= (struct i915_fragment_program
*) prog
;
1273 translate_program(p
);
1282 i915ProgramStringNotify(struct gl_context
* ctx
,
1283 GLenum target
, struct gl_program
*prog
)
1285 if (target
== GL_FRAGMENT_PROGRAM_ARB
) {
1286 struct i915_fragment_program
*p
= (struct i915_fragment_program
*) prog
;
1289 /* Hack: make sure fog is correctly enabled according to this
1290 * fragment program's fog options.
1292 if (p
->FragProg
.FogOption
) {
1293 /* add extra instructions to do fog, then turn off FogOption field */
1294 _mesa_append_fog_code(ctx
, &p
->FragProg
, GL_TRUE
);
1295 p
->FragProg
.FogOption
= GL_NONE
;
1299 (void) _tnl_program_string(ctx
, target
, prog
);
1301 /* XXX check if program is legal, within limits */
1306 i915_update_program(struct gl_context
*ctx
)
1308 struct intel_context
*intel
= intel_context(ctx
);
1309 struct i915_context
*i915
= i915_context(&intel
->ctx
);
1310 struct i915_fragment_program
*fp
=
1311 (struct i915_fragment_program
*) ctx
->FragmentProgram
._Current
;
1313 if (i915
->current_program
!= fp
) {
1314 if (i915
->current_program
) {
1315 i915
->current_program
->on_hardware
= 0;
1316 i915
->current_program
->params_uptodate
= 0;
1319 i915
->current_program
= fp
;
1322 if (!fp
->translated
)
1323 translate_program(fp
);
1325 FALLBACK(&i915
->intel
, I915_FALLBACK_PROGRAM
, fp
->error
);
1329 i915ValidateFragmentProgram(struct i915_context
*i915
)
1331 struct gl_context
*ctx
= &i915
->intel
.ctx
;
1332 struct intel_context
*intel
= intel_context(ctx
);
1333 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
1334 struct vertex_buffer
*VB
= &tnl
->vb
;
1336 struct i915_fragment_program
*p
=
1337 (struct i915_fragment_program
*) ctx
->FragmentProgram
._Current
;
1339 const GLuint inputsRead
= p
->FragProg
.Base
.InputsRead
;
1340 GLuint s4
= i915
->state
.Ctx
[I915_CTXREG_LIS4
] & ~S4_VFMT_MASK
;
1341 GLuint s2
= S2_TEXCOORD_NONE
;
1346 VB
->AttribPtr
[VERT_ATTRIB_POS
] = VB
->NdcPtr
;
1349 translate_program(p
);
1351 intel
->vertex_attr_count
= 0;
1352 intel
->wpos_offset
= 0;
1353 intel
->wpos_size
= 0;
1354 intel
->coloroffset
= 0;
1355 intel
->specoffset
= 0;
1357 if (inputsRead
& FRAG_BITS_TEX_ANY
) {
1358 EMIT_ATTR(_TNL_ATTRIB_POS
, EMIT_4F_VIEWPORT
, S4_VFMT_XYZW
, 16);
1361 EMIT_ATTR(_TNL_ATTRIB_POS
, EMIT_3F_VIEWPORT
, S4_VFMT_XYZ
, 12);
1364 if (inputsRead
& FRAG_BIT_COL0
) {
1365 intel
->coloroffset
= offset
/ 4;
1366 EMIT_ATTR(_TNL_ATTRIB_COLOR0
, EMIT_4UB_4F_BGRA
, S4_VFMT_COLOR
, 4);
1369 if (inputsRead
& FRAG_BIT_COL1
) {
1370 intel
->specoffset
= offset
/ 4;
1371 EMIT_ATTR(_TNL_ATTRIB_COLOR1
, EMIT_4UB_4F_BGRA
, S4_VFMT_SPEC_FOG
, 4);
1374 if ((inputsRead
& FRAG_BIT_FOGC
) || i915
->vertex_fog
!= I915_FOG_NONE
) {
1375 EMIT_ATTR(_TNL_ATTRIB_FOG
, EMIT_1F
, S4_VFMT_FOG_PARAM
, 4);
1378 for (i
= 0; i
< p
->ctx
->Const
.MaxTextureCoordUnits
; i
++) {
1379 if (inputsRead
& FRAG_BIT_TEX(i
)) {
1380 int sz
= VB
->AttribPtr
[_TNL_ATTRIB_TEX0
+ i
]->size
;
1382 s2
&= ~S2_TEXCOORD_FMT(i
, S2_TEXCOORD_FMT0_MASK
);
1383 s2
|= S2_TEXCOORD_FMT(i
, SZ_TO_HW(sz
));
1385 EMIT_ATTR(_TNL_ATTRIB_TEX0
+ i
, EMIT_SZ(sz
), 0, sz
* 4);
1387 else if (inputsRead
& FRAG_BIT_VAR(i
)) {
1388 int sz
= VB
->AttribPtr
[_TNL_ATTRIB_GENERIC0
+ i
]->size
;
1390 s2
&= ~S2_TEXCOORD_FMT(i
, S2_TEXCOORD_FMT0_MASK
);
1391 s2
|= S2_TEXCOORD_FMT(i
, SZ_TO_HW(sz
));
1393 EMIT_ATTR(_TNL_ATTRIB_GENERIC0
+ i
, EMIT_SZ(sz
), 0, sz
* 4);
1395 else if (i
== p
->wpos_tex
) {
1397 /* If WPOS is required, duplicate the XYZ position data in an
1398 * unused texture coordinate:
1400 s2
&= ~S2_TEXCOORD_FMT(i
, S2_TEXCOORD_FMT0_MASK
);
1401 s2
|= S2_TEXCOORD_FMT(i
, SZ_TO_HW(3));
1403 intel
->wpos_offset
= offset
;
1404 intel
->wpos_size
= 3 * sizeof(GLuint
);
1406 EMIT_PAD(intel
->wpos_size
);
1410 if (s2
!= i915
->state
.Ctx
[I915_CTXREG_LIS2
] ||
1411 s4
!= i915
->state
.Ctx
[I915_CTXREG_LIS4
]) {
1414 I915_STATECHANGE(i915
, I915_UPLOAD_CTX
);
1416 /* Must do this *after* statechange, so as not to affect
1417 * buffered vertices reliant on the old state:
1419 intel
->vertex_size
= _tnl_install_attrs(&intel
->ctx
,
1420 intel
->vertex_attrs
,
1421 intel
->vertex_attr_count
,
1422 intel
->ViewportMatrix
.m
, 0);
1424 assert(intel
->prim
.current_offset
== intel
->prim
.start_offset
);
1425 intel
->prim
.start_offset
= (intel
->prim
.current_offset
+ intel
->vertex_size
-1) / intel
->vertex_size
* intel
->vertex_size
;
1426 intel
->prim
.current_offset
= intel
->prim
.start_offset
;
1428 intel
->vertex_size
>>= 2;
1430 i915
->state
.Ctx
[I915_CTXREG_LIS2
] = s2
;
1431 i915
->state
.Ctx
[I915_CTXREG_LIS4
] = s4
;
1433 k
= intel
->vtbl
.check_vertex_size(intel
, intel
->vertex_size
);
1437 if (!p
->params_uptodate
)
1440 if (!p
->on_hardware
)
1441 i915_upload_program(i915
, p
);
1443 if (INTEL_DEBUG
& DEBUG_WM
) {
1445 i915_disassemble_program(i915
->state
.Program
, i915
->state
.ProgramSize
);
1450 i915InitFragProgFuncs(struct dd_function_table
*functions
)
1452 functions
->BindProgram
= i915BindProgram
;
1453 functions
->NewProgram
= i915NewProgram
;
1454 functions
->DeleteProgram
= i915DeleteProgram
;
1455 functions
->IsProgramNative
= i915IsProgramNative
;
1456 functions
->ProgramStringNotify
= i915ProgramStringNotify
;