1 /**************************************************************************
3 * Copyright 2003 VMware, Inc.
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 VMWARE 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)
75 /* texcoord_mapping[unit] = index | TEXCOORD_{TEX,VAR} */
76 #define TEXCOORD_TEX (0<<7)
77 #define TEXCOORD_VAR (1<<7)
80 get_texcoord_mapping(struct i915_fragment_program
*p
, uint8_t texcoord
)
82 for (unsigned i
= 0; i
< p
->ctx
->Const
.MaxTextureCoordUnits
; i
++) {
83 if (p
->texcoord_mapping
[i
] == texcoord
)
88 return p
->ctx
->Const
.MaxTextureCoordUnits
- 1;
92 * Retrieve a ureg for the given source register. Will emit
93 * constants, apply swizzling and negation as needed.
96 src_vector(struct i915_fragment_program
*p
,
97 const struct prog_src_register
*source
,
98 const struct gl_fragment_program
*program
)
103 switch (source
->File
) {
107 case PROGRAM_TEMPORARY
:
108 if (source
->Index
>= I915_MAX_TEMPORARY
) {
109 i915_program_error(p
, "Exceeded max temporary reg: %d/%d",
110 source
->Index
, I915_MAX_TEMPORARY
);
113 src
= UREG(REG_TYPE_R
, source
->Index
);
116 switch (source
->Index
) {
117 case VARYING_SLOT_POS
:
118 src
= i915_emit_decl(p
, REG_TYPE_T
, p
->wpos_tex
, D0_CHANNEL_ALL
);
120 case VARYING_SLOT_COL0
:
121 src
= i915_emit_decl(p
, REG_TYPE_T
, T_DIFFUSE
, D0_CHANNEL_ALL
);
123 case VARYING_SLOT_COL1
:
124 src
= i915_emit_decl(p
, REG_TYPE_T
, T_SPECULAR
, D0_CHANNEL_XYZ
);
125 src
= swizzle(src
, X
, Y
, Z
, ONE
);
127 case VARYING_SLOT_FOGC
:
128 src
= i915_emit_decl(p
, REG_TYPE_T
, T_FOG_W
, D0_CHANNEL_W
);
129 src
= swizzle(src
, W
, ZERO
, ZERO
, ONE
);
131 case VARYING_SLOT_TEX0
:
132 case VARYING_SLOT_TEX1
:
133 case VARYING_SLOT_TEX2
:
134 case VARYING_SLOT_TEX3
:
135 case VARYING_SLOT_TEX4
:
136 case VARYING_SLOT_TEX5
:
137 case VARYING_SLOT_TEX6
:
138 case VARYING_SLOT_TEX7
:
139 unit
= get_texcoord_mapping(p
, (source
->Index
-
140 VARYING_SLOT_TEX0
) | TEXCOORD_TEX
);
141 src
= i915_emit_decl(p
, REG_TYPE_T
,
146 case VARYING_SLOT_VAR0
:
147 case VARYING_SLOT_VAR0
+ 1:
148 case VARYING_SLOT_VAR0
+ 2:
149 case VARYING_SLOT_VAR0
+ 3:
150 case VARYING_SLOT_VAR0
+ 4:
151 case VARYING_SLOT_VAR0
+ 5:
152 case VARYING_SLOT_VAR0
+ 6:
153 case VARYING_SLOT_VAR0
+ 7:
154 unit
= get_texcoord_mapping(p
, (source
->Index
-
155 VARYING_SLOT_VAR0
) | TEXCOORD_VAR
);
156 src
= i915_emit_decl(p
, REG_TYPE_T
,
162 i915_program_error(p
, "Bad source->Index: %d", source
->Index
);
168 switch (source
->Index
) {
169 case FRAG_RESULT_COLOR
:
170 case FRAG_RESULT_DATA0
:
171 src
= UREG(REG_TYPE_OC
, 0);
173 case FRAG_RESULT_DEPTH
:
174 src
= UREG(REG_TYPE_OD
, 0);
177 i915_program_error(p
, "Bad source->Index: %d", source
->Index
);
182 /* Various paramters and env values. All emitted to
183 * hardware as program constants.
185 case PROGRAM_CONSTANT
:
186 case PROGRAM_STATE_VAR
:
187 case PROGRAM_UNIFORM
:
188 src
= i915_emit_param4fv(p
,
189 &program
->Base
.Parameters
->ParameterValues
[source
->Index
][0].f
);
193 i915_program_error(p
, "Bad source->File: %d", source
->File
);
198 GET_SWZ(source
->Swizzle
, 0),
199 GET_SWZ(source
->Swizzle
, 1),
200 GET_SWZ(source
->Swizzle
, 2), GET_SWZ(source
->Swizzle
, 3));
204 GET_BIT(source
->Negate
, 0),
205 GET_BIT(source
->Negate
, 1),
206 GET_BIT(source
->Negate
, 2),
207 GET_BIT(source
->Negate
, 3));
214 get_result_vector(struct i915_fragment_program
*p
,
215 const struct prog_instruction
*inst
)
217 switch (inst
->DstReg
.File
) {
219 switch (inst
->DstReg
.Index
) {
220 case FRAG_RESULT_COLOR
:
221 case FRAG_RESULT_DATA0
:
222 return UREG(REG_TYPE_OC
, 0);
223 case FRAG_RESULT_DEPTH
:
224 p
->depth_written
= 1;
225 return UREG(REG_TYPE_OD
, 0);
227 i915_program_error(p
, "Bad inst->DstReg.Index: %d",
231 case PROGRAM_TEMPORARY
:
232 return UREG(REG_TYPE_R
, inst
->DstReg
.Index
);
234 i915_program_error(p
, "Bad inst->DstReg.File: %d", inst
->DstReg
.File
);
240 get_result_flags(const struct prog_instruction
*inst
)
245 flags
|= A0_DEST_SATURATE
;
246 if (inst
->DstReg
.WriteMask
& WRITEMASK_X
)
247 flags
|= A0_DEST_CHANNEL_X
;
248 if (inst
->DstReg
.WriteMask
& WRITEMASK_Y
)
249 flags
|= A0_DEST_CHANNEL_Y
;
250 if (inst
->DstReg
.WriteMask
& WRITEMASK_Z
)
251 flags
|= A0_DEST_CHANNEL_Z
;
252 if (inst
->DstReg
.WriteMask
& WRITEMASK_W
)
253 flags
|= A0_DEST_CHANNEL_W
;
259 translate_tex_src_target(struct i915_fragment_program
*p
, GLubyte bit
)
262 case TEXTURE_1D_INDEX
:
263 return D0_SAMPLE_TYPE_2D
;
264 case TEXTURE_2D_INDEX
:
265 return D0_SAMPLE_TYPE_2D
;
266 case TEXTURE_RECT_INDEX
:
267 return D0_SAMPLE_TYPE_2D
;
268 case TEXTURE_3D_INDEX
:
269 return D0_SAMPLE_TYPE_VOLUME
;
270 case TEXTURE_CUBE_INDEX
:
271 return D0_SAMPLE_TYPE_CUBE
;
273 i915_program_error(p
, "TexSrcBit: %d", bit
);
278 #define EMIT_TEX( OP ) \
280 GLuint dim = translate_tex_src_target( p, inst->TexSrcTarget ); \
281 const struct gl_fragment_program *program = &p->FragProg; \
282 GLuint unit = program->Base.SamplerUnits[inst->TexSrcUnit]; \
283 GLuint sampler = i915_emit_decl(p, REG_TYPE_S, \
285 GLuint coord = src_vector( p, &inst->SrcReg[0], program); \
288 i915_emit_texld( p, get_live_regs(p, inst), \
289 get_result_vector( p, inst ), \
290 get_result_flags( inst ), \
296 #define EMIT_ARITH( OP, N ) \
298 i915_emit_arith( p, \
300 get_result_vector( p, inst ), \
301 get_result_flags( inst ), 0, \
302 (N<1)?0:src_vector( p, &inst->SrcReg[0], program), \
303 (N<2)?0:src_vector( p, &inst->SrcReg[1], program), \
304 (N<3)?0:src_vector( p, &inst->SrcReg[2], program)); \
307 #define EMIT_1ARG_ARITH( OP ) EMIT_ARITH( OP, 1 )
308 #define EMIT_2ARG_ARITH( OP ) EMIT_ARITH( OP, 2 )
309 #define EMIT_3ARG_ARITH( OP ) EMIT_ARITH( OP, 3 )
312 * TODO: consider moving this into core
314 static bool calc_live_regs( struct i915_fragment_program
*p
)
316 const struct gl_fragment_program
*program
= &p
->FragProg
;
317 GLuint regsUsed
= ~((1 << I915_MAX_TEMPORARY
) - 1);
318 uint8_t live_components
[I915_MAX_TEMPORARY
] = { 0, };
321 for (i
= program
->Base
.NumInstructions
- 1; i
>= 0; i
--) {
322 struct prog_instruction
*inst
= &program
->Base
.Instructions
[i
];
323 int opArgs
= _mesa_num_inst_src_regs(inst
->Opcode
);
326 /* Register is written to: unmark as live for this and preceeding ops */
327 if (inst
->DstReg
.File
== PROGRAM_TEMPORARY
) {
328 if (inst
->DstReg
.Index
>= I915_MAX_TEMPORARY
)
331 live_components
[inst
->DstReg
.Index
] &= ~inst
->DstReg
.WriteMask
;
332 if (live_components
[inst
->DstReg
.Index
] == 0)
333 regsUsed
&= ~(1 << inst
->DstReg
.Index
);
336 for (a
= 0; a
< opArgs
; a
++) {
337 /* Register is read from: mark as live for this and preceeding ops */
338 if (inst
->SrcReg
[a
].File
== PROGRAM_TEMPORARY
) {
341 if (inst
->SrcReg
[a
].Index
>= I915_MAX_TEMPORARY
)
344 regsUsed
|= 1 << inst
->SrcReg
[a
].Index
;
346 for (c
= 0; c
< 4; c
++) {
347 const unsigned field
= GET_SWZ(inst
->SrcReg
[a
].Swizzle
, c
);
349 if (field
<= SWIZZLE_W
)
350 live_components
[inst
->SrcReg
[a
].Index
] |= (1U << field
);
355 p
->usedRegs
[i
] = regsUsed
;
361 static GLuint
get_live_regs( struct i915_fragment_program
*p
,
362 const struct prog_instruction
*inst
)
364 const struct gl_fragment_program
*program
= &p
->FragProg
;
365 GLuint nr
= inst
- program
->Base
.Instructions
;
367 return p
->usedRegs
[nr
];
371 /* Possible concerns:
373 * SIN, COS -- could use another taylor step?
374 * LIT -- results seem a little different to sw mesa
375 * LOG -- different to mesa on negative numbers, but this is conformant.
377 * Parse failures -- Mesa doesn't currently give a good indication
378 * internally whether a particular program string parsed or not. This
379 * can lead to confusion -- hopefully we cope with it ok now.
383 upload_program(struct i915_fragment_program
*p
)
385 const struct gl_fragment_program
*program
= &p
->FragProg
;
386 const struct prog_instruction
*inst
= program
->Base
.Instructions
;
388 if (INTEL_DEBUG
& DEBUG_WM
)
389 _mesa_print_program(&program
->Base
);
391 /* Is this a parse-failed program? Ensure a valid program is
392 * loaded, as the flagging of an error isn't sufficient to stop
393 * this being uploaded to hardware.
395 if (inst
[0].Opcode
== OPCODE_END
) {
396 GLuint tmp
= i915_get_utemp(p
);
399 UREG(REG_TYPE_OC
, 0),
400 A0_DEST_CHANNEL_ALL
, 0,
401 swizzle(tmp
, ONE
, ZERO
, ONE
, ONE
), 0, 0);
405 if (program
->Base
.NumInstructions
> I915_MAX_INSN
) {
406 i915_program_error(p
, "Exceeded max instructions (%d out of %d)",
407 program
->Base
.NumInstructions
, I915_MAX_INSN
);
411 /* Not always needed:
413 if (!calc_live_regs(p
)) {
414 i915_program_error(p
, "Could not allocate registers");
419 GLuint src0
, src1
, src2
, flags
;
420 GLuint tmp
= 0, dst
, consts0
= 0, consts1
= 0;
422 switch (inst
->Opcode
) {
424 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
427 get_result_vector(p
, inst
),
428 get_result_flags(inst
), 0,
429 src0
, negate(src0
, 1, 1, 1, 1), 0);
433 EMIT_2ARG_ARITH(A0_ADD
);
437 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
438 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
439 src2
= src_vector(p
, &inst
->SrcReg
[2], program
);
440 i915_emit_arith(p
, A0_CMP
, get_result_vector(p
, inst
), get_result_flags(inst
), 0, src0
, src2
, src1
); /* NOTE: order of src2, src1 */
444 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
445 tmp
= i915_get_utemp(p
);
446 consts0
= i915_emit_const4fv(p
, sin_quad_constants
[0]);
447 consts1
= i915_emit_const4fv(p
, sin_quad_constants
[1]);
449 /* Reduce range from repeating about [-pi,pi] to [-1,1] */
452 tmp
, A0_DEST_CHANNEL_X
, 0,
454 swizzle(consts1
, Z
, ZERO
, ZERO
, ZERO
), /* 1/(2pi) */
455 swizzle(consts0
, W
, ZERO
, ZERO
, ZERO
)); /* .75 */
457 i915_emit_arith(p
, A0_FRC
, tmp
, A0_DEST_CHANNEL_X
, 0, tmp
, 0, 0);
461 tmp
, A0_DEST_CHANNEL_X
, 0,
463 swizzle(consts0
, X
, ZERO
, ZERO
, ZERO
), /* 2 */
464 swizzle(consts0
, Y
, ZERO
, ZERO
, ZERO
)); /* -1 */
466 /* Compute COS with the same calculation used for SIN, but a
467 * different source range has been mapped to [-1,1] this time.
470 /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
473 tmp
, A0_DEST_CHANNEL_Y
, 0,
474 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
475 negate(swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
), 0, 1, 0, 0),
478 /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
481 tmp
, A0_DEST_CHANNEL_Y
, 0,
482 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
486 /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
489 tmp
, A0_DEST_CHANNEL_X
, 0,
491 swizzle(consts1
, X
, Y
, ZERO
, ZERO
),
494 /* tmp.x now contains a first approximation (y). Now, weight it
495 * against tmp.y**2 to get closer.
499 tmp
, A0_DEST_CHANNEL_Y
, 0,
500 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
501 negate(swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
), 0, 1, 0, 0),
504 /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
507 tmp
, A0_DEST_CHANNEL_Y
, 0,
508 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
509 swizzle(tmp
, ZERO
, Y
, ZERO
, ZERO
),
510 negate(swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
), 0, 1, 0, 0));
512 /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
515 get_result_vector(p
, inst
),
516 get_result_flags(inst
), 0,
517 swizzle(consts1
, W
, W
, W
, W
),
518 swizzle(tmp
, Y
, Y
, Y
, Y
),
519 swizzle(tmp
, X
, X
, X
, X
));
523 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
524 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
527 get_result_vector(p
, inst
),
528 get_result_flags(inst
), 0,
529 swizzle(src0
, X
, Y
, ZERO
, ZERO
),
530 swizzle(src1
, X
, Y
, ZERO
, ZERO
),
535 EMIT_2ARG_ARITH(A0_DP3
);
539 EMIT_2ARG_ARITH(A0_DP4
);
543 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
544 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
548 get_result_vector(p
, inst
),
549 get_result_flags(inst
), 0,
550 swizzle(src0
, X
, Y
, Z
, ONE
), src1
, 0);
554 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
555 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
557 /* result[0] = 1 * 1;
558 * result[1] = a[1] * b[1];
559 * result[2] = a[2] * 1;
560 * result[3] = 1 * b[3];
564 get_result_vector(p
, inst
),
565 get_result_flags(inst
), 0,
566 swizzle(src0
, ONE
, Y
, Z
, ONE
),
567 swizzle(src1
, ONE
, Y
, ONE
, W
), 0);
571 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
575 get_result_vector(p
, inst
),
576 get_result_flags(inst
), 0,
577 swizzle(src0
, X
, X
, X
, X
), 0, 0);
581 EMIT_1ARG_ARITH(A0_FLR
);
585 EMIT_1ARG_ARITH(A0_TRC
);
589 EMIT_1ARG_ARITH(A0_FRC
);
593 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
594 tmp
= i915_get_utemp(p
);
596 i915_emit_texld(p
, get_live_regs(p
, inst
),
597 tmp
, A0_DEST_CHANNEL_ALL
, /* use a dummy dest reg */
598 0, src0
, T0_TEXKILL
);
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 EMIT_2ARG_ARITH(A0_MIN
);
679 EMIT_1ARG_ARITH(A0_MOV
);
683 EMIT_2ARG_ARITH(A0_MUL
);
687 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
688 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
689 tmp
= i915_get_utemp(p
);
690 flags
= get_result_flags(inst
);
692 /* XXX: masking on intermediate values, here and elsewhere.
696 tmp
, A0_DEST_CHANNEL_X
, 0,
697 swizzle(src0
, X
, X
, X
, X
), 0, 0);
699 i915_emit_arith(p
, A0_MUL
, tmp
, A0_DEST_CHANNEL_X
, 0, tmp
, src1
, 0);
704 get_result_vector(p
, inst
),
705 flags
, 0, swizzle(tmp
, X
, X
, X
, X
), 0, 0);
710 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
714 get_result_vector(p
, inst
),
715 get_result_flags(inst
), 0,
716 swizzle(src0
, X
, X
, X
, X
), 0, 0);
721 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
725 get_result_vector(p
, inst
),
726 get_result_flags(inst
), 0,
727 swizzle(src0
, X
, X
, X
, X
), 0, 0);
731 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
732 tmp
= i915_get_utemp(p
);
735 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
736 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
737 * t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x
738 * scs.x = DP4 t1, sin_constants
739 * t1 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1
740 * scs.y = DP4 t1, cos_constants
744 tmp
, A0_DEST_CHANNEL_XY
, 0,
745 swizzle(src0
, X
, X
, ONE
, ONE
),
746 swizzle(src0
, X
, ONE
, ONE
, ONE
), 0);
750 tmp
, A0_DEST_CHANNEL_ALL
, 0,
751 swizzle(tmp
, X
, Y
, X
, Y
),
752 swizzle(tmp
, X
, X
, ONE
, ONE
), 0);
754 if (inst
->DstReg
.WriteMask
& WRITEMASK_Y
) {
757 if (inst
->DstReg
.WriteMask
& WRITEMASK_X
)
758 tmp1
= i915_get_utemp(p
);
764 tmp1
, A0_DEST_CHANNEL_ALL
, 0,
765 swizzle(tmp
, X
, Y
, Y
, W
),
766 swizzle(tmp
, X
, Z
, ONE
, ONE
), 0);
770 get_result_vector(p
, inst
),
771 A0_DEST_CHANNEL_Y
, 0,
772 swizzle(tmp1
, W
, Z
, Y
, X
),
773 i915_emit_const4fv(p
, sin_constants
), 0);
776 if (inst
->DstReg
.WriteMask
& WRITEMASK_X
) {
779 tmp
, A0_DEST_CHANNEL_XYZ
, 0,
780 swizzle(tmp
, X
, X
, Z
, ONE
),
781 swizzle(tmp
, Z
, ONE
, ONE
, ONE
), 0);
785 get_result_vector(p
, inst
),
786 A0_DEST_CHANNEL_X
, 0,
787 swizzle(tmp
, ONE
, Z
, Y
, X
),
788 i915_emit_const4fv(p
, cos_constants
), 0);
793 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
794 tmp
= i915_get_utemp(p
);
795 consts0
= i915_emit_const4fv(p
, sin_quad_constants
[0]);
796 consts1
= i915_emit_const4fv(p
, sin_quad_constants
[1]);
798 /* Reduce range from repeating about [-pi,pi] to [-1,1] */
801 tmp
, A0_DEST_CHANNEL_X
, 0,
803 swizzle(consts1
, Z
, ZERO
, ZERO
, ZERO
), /* 1/(2pi) */
804 swizzle(consts0
, Z
, ZERO
, ZERO
, ZERO
)); /* .5 */
806 i915_emit_arith(p
, A0_FRC
, tmp
, A0_DEST_CHANNEL_X
, 0, tmp
, 0, 0);
810 tmp
, A0_DEST_CHANNEL_X
, 0,
812 swizzle(consts0
, X
, ZERO
, ZERO
, ZERO
), /* 2 */
813 swizzle(consts0
, Y
, ZERO
, ZERO
, ZERO
)); /* -1 */
815 /* Compute sin using a quadratic and quartic. It gives continuity
816 * that repeating the Taylor series lacks every 2*pi, and has
819 * The idea was described at:
820 * http://www.devmaster.net/forums/showthread.php?t=5784
823 /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
826 tmp
, A0_DEST_CHANNEL_Y
, 0,
827 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
828 negate(swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
), 0, 1, 0, 0),
831 /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
834 tmp
, A0_DEST_CHANNEL_Y
, 0,
835 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
839 /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
842 tmp
, A0_DEST_CHANNEL_X
, 0,
844 swizzle(consts1
, X
, Y
, ZERO
, ZERO
),
847 /* tmp.x now contains a first approximation (y). Now, weight it
848 * against tmp.y**2 to get closer.
852 tmp
, A0_DEST_CHANNEL_Y
, 0,
853 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
854 negate(swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
), 0, 1, 0, 0),
857 /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
860 tmp
, A0_DEST_CHANNEL_Y
, 0,
861 swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
),
862 swizzle(tmp
, ZERO
, Y
, ZERO
, ZERO
),
863 negate(swizzle(tmp
, ZERO
, X
, ZERO
, ZERO
), 0, 1, 0, 0));
865 /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
868 get_result_vector(p
, inst
),
869 get_result_flags(inst
), 0,
870 swizzle(consts1
, W
, W
, W
, W
),
871 swizzle(tmp
, Y
, Y
, Y
, Y
),
872 swizzle(tmp
, X
, X
, X
, X
));
877 EMIT_2ARG_ARITH(A0_SGE
);
881 EMIT_2ARG_ARITH(A0_SLT
);
885 dst
= get_result_vector(p
, inst
);
886 flags
= get_result_flags(inst
);
887 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
888 tmp
= i915_get_utemp(p
);
890 /* tmp = (src < 0.0) */
896 swizzle(src0
, ZERO
, ZERO
, ZERO
, ZERO
),
899 /* dst = (0.0 < src) */
904 swizzle(src0
, ZERO
, ZERO
, ZERO
, ZERO
),
908 /* dst = (src > 0.0) - (src < 0.0) */
914 negate(tmp
, 1, 1, 1, 1),
920 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
921 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
925 get_result_vector(p
, inst
),
926 get_result_flags(inst
), 0,
927 src0
, negate(src1
, 1, 1, 1, 1), 0);
931 EMIT_1ARG_ARITH(A0_MOV
); /* extended swizzle handled natively */
948 * result.x = src0.y * src1.z - src0.z * src1.y;
949 * result.y = src0.z * src1.x - src0.x * src1.z;
950 * result.z = src0.x * src1.y - src0.y * src1.x;
953 src0
= src_vector(p
, &inst
->SrcReg
[0], program
);
954 src1
= src_vector(p
, &inst
->SrcReg
[1], program
);
955 tmp
= i915_get_utemp(p
);
959 tmp
, A0_DEST_CHANNEL_ALL
, 0,
960 swizzle(src0
, Z
, X
, Y
, ONE
),
961 swizzle(src1
, Y
, Z
, X
, ONE
), 0);
965 get_result_vector(p
, inst
),
966 get_result_flags(inst
), 0,
967 swizzle(src0
, Y
, Z
, X
, ONE
),
968 swizzle(src1
, Z
, X
, Y
, ONE
),
969 negate(tmp
, 1, 1, 1, 0));
989 i915_program_error(p
, "Unsupported opcode: %s",
990 _mesa_opcode_string(inst
->Opcode
));
995 /* These opcodes are claimed as GLSL, NV_vp, and ARB_vp in
996 * prog_instruction.h, but apparently GLSL doesn't ever emit them.
997 * Instead, it translates to EX2 or LG2.
1001 /* These opcodes are claimed by GLSL in prog_instruction.h, but
1002 * only NV_vp/fp appears to emit them.
1005 i915_program_error(p
, "bad opcode: %s",
1006 _mesa_opcode_string(inst
->Opcode
));
1011 i915_release_utemps(p
);
1015 /* Rather than trying to intercept and jiggle depth writes during
1016 * emit, just move the value into its correct position at the end of
1020 fixup_depth_write(struct i915_fragment_program
*p
)
1022 if (p
->depth_written
) {
1023 GLuint depth
= UREG(REG_TYPE_OD
, 0);
1027 depth
, A0_DEST_CHANNEL_W
, 0,
1028 swizzle(depth
, X
, Y
, Z
, Z
), 0, 0);
1033 check_texcoord_mapping(struct i915_fragment_program
*p
)
1035 GLbitfield64 inputs
= p
->FragProg
.Base
.InputsRead
;
1038 for (unsigned i
= 0; i
< p
->ctx
->Const
.MaxTextureCoordUnits
; i
++) {
1039 if (inputs
& VARYING_BIT_TEX(i
)) {
1040 if (unit
>= p
->ctx
->Const
.MaxTextureCoordUnits
) {
1044 p
->texcoord_mapping
[unit
++] = i
| TEXCOORD_TEX
;
1046 if (inputs
& VARYING_BIT_VAR(i
)) {
1047 if (unit
>= p
->ctx
->Const
.MaxTextureCoordUnits
) {
1051 p
->texcoord_mapping
[unit
++] = i
| TEXCOORD_VAR
;
1055 if (unit
> p
->ctx
->Const
.MaxTextureCoordUnits
)
1056 i915_program_error(p
, "Too many texcoord units");
1060 check_wpos(struct i915_fragment_program
*p
)
1062 GLbitfield64 inputs
= p
->FragProg
.Base
.InputsRead
;
1068 if ((inputs
& VARYING_BIT_POS
) == 0)
1071 for (i
= 0; i
< p
->ctx
->Const
.MaxTextureCoordUnits
; i
++) {
1072 unit
+= !!(inputs
& VARYING_BIT_TEX(i
));
1073 unit
+= !!(inputs
& VARYING_BIT_VAR(i
));
1076 if (unit
< p
->ctx
->Const
.MaxTextureCoordUnits
)
1079 i915_program_error(p
, "No free texcoord for wpos value");
1084 translate_program(struct i915_fragment_program
*p
)
1086 struct i915_context
*i915
= I915_CONTEXT(p
->ctx
);
1088 if (INTEL_DEBUG
& DEBUG_WM
) {
1090 _mesa_print_program(&p
->FragProg
.Base
);
1094 i915_init_program(i915
, p
);
1095 check_texcoord_mapping(p
);
1098 fixup_depth_write(p
);
1099 i915_fini_program(p
);
1106 track_params(struct i915_fragment_program
*p
)
1111 _mesa_load_state_parameters(p
->ctx
, p
->FragProg
.Base
.Parameters
);
1113 for (i
= 0; i
< p
->nr_params
; i
++) {
1114 GLint reg
= p
->param
[i
].reg
;
1115 COPY_4V(p
->constant
[reg
], p
->param
[i
].values
);
1118 p
->params_uptodate
= 1;
1119 p
->on_hardware
= 0; /* overkill */
1124 i915BindProgram(struct gl_context
* ctx
, GLenum target
, struct gl_program
*prog
)
1126 if (target
== GL_FRAGMENT_PROGRAM_ARB
) {
1127 struct i915_context
*i915
= I915_CONTEXT(ctx
);
1128 struct i915_fragment_program
*p
= (struct i915_fragment_program
*) prog
;
1130 if (i915
->current_program
== p
)
1133 if (i915
->current_program
) {
1134 i915
->current_program
->on_hardware
= 0;
1135 i915
->current_program
->params_uptodate
= 0;
1138 i915
->current_program
= p
;
1140 assert(p
->on_hardware
== 0);
1141 assert(p
->params_uptodate
== 0);
1146 static struct gl_program
*
1147 i915NewProgram(struct gl_context
* ctx
, GLenum target
, GLuint id
)
1150 case GL_VERTEX_PROGRAM_ARB
: {
1151 struct gl_program
*prog
= CALLOC_STRUCT(gl_program
);
1152 return _mesa_init_gl_program(prog
, target
, id
);
1155 case GL_FRAGMENT_PROGRAM_ARB
:{
1156 struct i915_fragment_program
*prog
=
1157 CALLOC_STRUCT(i915_fragment_program
);
1159 i915_init_program(I915_CONTEXT(ctx
), prog
);
1161 return _mesa_init_gl_program(&prog
->FragProg
.Base
, target
, id
);
1170 return _mesa_new_program(ctx
, target
, id
);
1175 i915DeleteProgram(struct gl_context
* ctx
, struct gl_program
*prog
)
1177 if (prog
->Target
== GL_FRAGMENT_PROGRAM_ARB
) {
1178 struct i915_context
*i915
= I915_CONTEXT(ctx
);
1179 struct i915_fragment_program
*p
= (struct i915_fragment_program
*) prog
;
1181 if (i915
->current_program
== p
)
1182 i915
->current_program
= 0;
1185 _mesa_delete_program(ctx
, prog
);
1190 i915IsProgramNative(struct gl_context
* ctx
, GLenum target
, struct gl_program
*prog
)
1192 if (target
== GL_FRAGMENT_PROGRAM_ARB
) {
1193 struct i915_fragment_program
*p
= (struct i915_fragment_program
*) prog
;
1196 translate_program(p
);
1205 i915ProgramStringNotify(struct gl_context
* ctx
,
1206 GLenum target
, struct gl_program
*prog
)
1208 if (target
== GL_FRAGMENT_PROGRAM_ARB
) {
1209 struct i915_fragment_program
*p
= (struct i915_fragment_program
*) prog
;
1213 (void) _tnl_program_string(ctx
, target
, prog
);
1215 /* XXX check if program is legal, within limits */
1220 i915SamplerUniformChange(struct gl_context
*ctx
,
1221 GLenum target
, struct gl_program
*prog
)
1223 i915ProgramStringNotify(ctx
, target
, prog
);
1227 i915_update_program(struct gl_context
*ctx
)
1229 struct intel_context
*intel
= intel_context(ctx
);
1230 struct i915_context
*i915
= i915_context(&intel
->ctx
);
1231 struct i915_fragment_program
*fp
=
1232 (struct i915_fragment_program
*) ctx
->FragmentProgram
._Current
;
1234 if (i915
->current_program
!= fp
) {
1235 if (i915
->current_program
) {
1236 i915
->current_program
->on_hardware
= 0;
1237 i915
->current_program
->params_uptodate
= 0;
1240 i915
->current_program
= fp
;
1243 if (!fp
->translated
)
1244 translate_program(fp
);
1246 FALLBACK(&i915
->intel
, I915_FALLBACK_PROGRAM
, fp
->error
);
1250 i915ValidateFragmentProgram(struct i915_context
*i915
)
1252 struct gl_context
*ctx
= &i915
->intel
.ctx
;
1253 struct intel_context
*intel
= intel_context(ctx
);
1254 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
1255 struct vertex_buffer
*VB
= &tnl
->vb
;
1257 struct i915_fragment_program
*p
=
1258 (struct i915_fragment_program
*) ctx
->FragmentProgram
._Current
;
1260 const GLbitfield64 inputsRead
= p
->FragProg
.Base
.InputsRead
;
1261 GLuint s4
= i915
->state
.Ctx
[I915_CTXREG_LIS4
] & ~S4_VFMT_MASK
;
1262 GLuint s2
= S2_TEXCOORD_NONE
;
1267 VB
->AttribPtr
[VERT_ATTRIB_POS
] = VB
->NdcPtr
;
1270 translate_program(p
);
1272 intel
->vertex_attr_count
= 0;
1273 intel
->wpos_offset
= 0;
1274 intel
->coloroffset
= 0;
1275 intel
->specoffset
= 0;
1277 if (inputsRead
& VARYING_BITS_TEX_ANY
|| p
->wpos_tex
!= -1) {
1278 EMIT_ATTR(_TNL_ATTRIB_POS
, EMIT_4F_VIEWPORT
, S4_VFMT_XYZW
, 16);
1281 EMIT_ATTR(_TNL_ATTRIB_POS
, EMIT_3F_VIEWPORT
, S4_VFMT_XYZ
, 12);
1284 /* Handle gl_PointSize builtin var here */
1285 if (ctx
->Point
._Attenuated
|| ctx
->VertexProgram
.PointSizeEnabled
)
1286 EMIT_ATTR(_TNL_ATTRIB_POINTSIZE
, EMIT_1F
, S4_VFMT_POINT_WIDTH
, 4);
1288 if (inputsRead
& VARYING_BIT_COL0
) {
1289 intel
->coloroffset
= offset
/ 4;
1290 EMIT_ATTR(_TNL_ATTRIB_COLOR0
, EMIT_4UB_4F_BGRA
, S4_VFMT_COLOR
, 4);
1293 if (inputsRead
& VARYING_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
& VARYING_BIT_FOGC
)) {
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
& VARYING_BIT_TEX(i
)) {
1304 int unit
= get_texcoord_mapping(p
, i
| TEXCOORD_TEX
);
1305 int sz
= VB
->AttribPtr
[_TNL_ATTRIB_TEX0
+ i
]->size
;
1307 s2
&= ~S2_TEXCOORD_FMT(unit
, S2_TEXCOORD_FMT0_MASK
);
1308 s2
|= S2_TEXCOORD_FMT(unit
, SZ_TO_HW(sz
));
1310 EMIT_ATTR(_TNL_ATTRIB_TEX0
+ i
, EMIT_SZ(sz
), 0, sz
* 4);
1312 if (inputsRead
& VARYING_BIT_VAR(i
)) {
1313 int unit
= get_texcoord_mapping(p
, i
| TEXCOORD_VAR
);
1314 int sz
= VB
->AttribPtr
[_TNL_ATTRIB_GENERIC0
+ i
]->size
;
1316 s2
&= ~S2_TEXCOORD_FMT(unit
, S2_TEXCOORD_FMT0_MASK
);
1317 s2
|= S2_TEXCOORD_FMT(unit
, SZ_TO_HW(sz
));
1319 EMIT_ATTR(_TNL_ATTRIB_GENERIC0
+ i
, EMIT_SZ(sz
), 0, sz
* 4);
1321 if (i
== p
->wpos_tex
) {
1322 int wpos_size
= 4 * sizeof(float);
1323 /* If WPOS is required, duplicate the XYZ position data in an
1324 * unused texture coordinate:
1326 s2
&= ~S2_TEXCOORD_FMT(i
, S2_TEXCOORD_FMT0_MASK
);
1327 s2
|= S2_TEXCOORD_FMT(i
, SZ_TO_HW(wpos_size
));
1329 intel
->wpos_offset
= offset
;
1330 EMIT_PAD(wpos_size
);
1334 if (s2
!= i915
->state
.Ctx
[I915_CTXREG_LIS2
] ||
1335 s4
!= i915
->state
.Ctx
[I915_CTXREG_LIS4
]) {
1336 I915_STATECHANGE(i915
, I915_UPLOAD_CTX
);
1338 /* Must do this *after* statechange, so as not to affect
1339 * buffered vertices reliant on the old state:
1341 intel
->vertex_size
= _tnl_install_attrs(&intel
->ctx
,
1342 intel
->vertex_attrs
,
1343 intel
->vertex_attr_count
,
1344 intel
->ViewportMatrix
.m
, 0);
1346 assert(intel
->prim
.current_offset
== intel
->prim
.start_offset
);
1347 intel
->prim
.start_offset
= (intel
->prim
.current_offset
+ intel
->vertex_size
-1) / intel
->vertex_size
* intel
->vertex_size
;
1348 intel
->prim
.current_offset
= intel
->prim
.start_offset
;
1350 intel
->vertex_size
>>= 2;
1352 i915
->state
.Ctx
[I915_CTXREG_LIS2
] = s2
;
1353 i915
->state
.Ctx
[I915_CTXREG_LIS4
] = s4
;
1355 assert(intel
->vtbl
.check_vertex_size(intel
, intel
->vertex_size
));
1358 if (!p
->params_uptodate
)
1361 if (!p
->on_hardware
)
1362 i915_upload_program(i915
, p
);
1364 if (INTEL_DEBUG
& DEBUG_WM
) {
1366 i915_disassemble_program(i915
->state
.Program
, i915
->state
.ProgramSize
);
1371 i915InitFragProgFuncs(struct dd_function_table
*functions
)
1373 functions
->BindProgram
= i915BindProgram
;
1374 functions
->NewProgram
= i915NewProgram
;
1375 functions
->DeleteProgram
= i915DeleteProgram
;
1376 functions
->IsProgramNative
= i915IsProgramNative
;
1377 functions
->ProgramStringNotify
= i915ProgramStringNotify
;
1378 functions
->SamplerUniformChange
= i915SamplerUniformChange
;