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 VARYING_SLOT_POS:
101 src = i915_emit_decl(p, REG_TYPE_T, p->wpos_tex, D0_CHANNEL_ALL);
103 case VARYING_SLOT_COL0:
104 src = i915_emit_decl(p, REG_TYPE_T, T_DIFFUSE, D0_CHANNEL_ALL);
106 case VARYING_SLOT_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 VARYING_SLOT_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 VARYING_SLOT_TEX0:
115 case VARYING_SLOT_TEX1:
116 case VARYING_SLOT_TEX2:
117 case VARYING_SLOT_TEX3:
118 case VARYING_SLOT_TEX4:
119 case VARYING_SLOT_TEX5:
120 case VARYING_SLOT_TEX6:
121 case VARYING_SLOT_TEX7:
122 src = i915_emit_decl(p, REG_TYPE_T,
123 T_TEX0 + (source->Index - VARYING_SLOT_TEX0),
127 case VARYING_SLOT_VAR0:
128 case VARYING_SLOT_VAR0 + 1:
129 case VARYING_SLOT_VAR0 + 2:
130 case VARYING_SLOT_VAR0 + 3:
131 case VARYING_SLOT_VAR0 + 4:
132 case VARYING_SLOT_VAR0 + 5:
133 case VARYING_SLOT_VAR0 + 6:
134 case VARYING_SLOT_VAR0 + 7:
135 src = i915_emit_decl(p, REG_TYPE_T,
136 T_TEX0 + (source->Index - VARYING_SLOT_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_CONSTANT:
164 case PROGRAM_STATE_VAR:
165 case PROGRAM_UNIFORM:
166 src = i915_emit_param4fv(p,
167 &program->Base.Parameters->ParameterValues[source->Index][0].f);
171 i915_program_error(p, "Bad source->File: %d", source->File);
176 GET_SWZ(source->Swizzle, 0),
177 GET_SWZ(source->Swizzle, 1),
178 GET_SWZ(source->Swizzle, 2), GET_SWZ(source->Swizzle, 3));
182 GET_BIT(source->Negate, 0),
183 GET_BIT(source->Negate, 1),
184 GET_BIT(source->Negate, 2),
185 GET_BIT(source->Negate, 3));
192 get_result_vector(struct i915_fragment_program *p,
193 const struct prog_instruction *inst)
195 switch (inst->DstReg.File) {
197 switch (inst->DstReg.Index) {
198 case FRAG_RESULT_COLOR:
199 case FRAG_RESULT_DATA0:
200 return UREG(REG_TYPE_OC, 0);
201 case FRAG_RESULT_DEPTH:
202 p->depth_written = 1;
203 return UREG(REG_TYPE_OD, 0);
205 i915_program_error(p, "Bad inst->DstReg.Index: %d",
209 case PROGRAM_TEMPORARY:
210 return UREG(REG_TYPE_R, inst->DstReg.Index);
212 i915_program_error(p, "Bad inst->DstReg.File: %d", inst->DstReg.File);
218 get_result_flags(const struct prog_instruction *inst)
222 if (inst->SaturateMode == SATURATE_ZERO_ONE)
223 flags |= A0_DEST_SATURATE;
224 if (inst->DstReg.WriteMask & WRITEMASK_X)
225 flags |= A0_DEST_CHANNEL_X;
226 if (inst->DstReg.WriteMask & WRITEMASK_Y)
227 flags |= A0_DEST_CHANNEL_Y;
228 if (inst->DstReg.WriteMask & WRITEMASK_Z)
229 flags |= A0_DEST_CHANNEL_Z;
230 if (inst->DstReg.WriteMask & WRITEMASK_W)
231 flags |= A0_DEST_CHANNEL_W;
237 translate_tex_src_target(struct i915_fragment_program *p, GLubyte bit)
240 case TEXTURE_1D_INDEX:
241 return D0_SAMPLE_TYPE_2D;
242 case TEXTURE_2D_INDEX:
243 return D0_SAMPLE_TYPE_2D;
244 case TEXTURE_RECT_INDEX:
245 return D0_SAMPLE_TYPE_2D;
246 case TEXTURE_3D_INDEX:
247 return D0_SAMPLE_TYPE_VOLUME;
248 case TEXTURE_CUBE_INDEX:
249 return D0_SAMPLE_TYPE_CUBE;
251 i915_program_error(p, "TexSrcBit: %d", bit);
256 #define EMIT_TEX( OP ) \
258 GLuint dim = translate_tex_src_target( p, inst->TexSrcTarget ); \
259 const struct gl_fragment_program *program = &p->FragProg; \
260 GLuint unit = program->Base.SamplerUnits[inst->TexSrcUnit]; \
261 GLuint sampler = i915_emit_decl(p, REG_TYPE_S, \
263 GLuint coord = src_vector( p, &inst->SrcReg[0], program); \
266 i915_emit_texld( p, get_live_regs(p, inst), \
267 get_result_vector( p, inst ), \
268 get_result_flags( inst ), \
274 #define EMIT_ARITH( OP, N ) \
276 i915_emit_arith( p, \
278 get_result_vector( p, inst ), \
279 get_result_flags( inst ), 0, \
280 (N<1)?0:src_vector( p, &inst->SrcReg[0], program), \
281 (N<2)?0:src_vector( p, &inst->SrcReg[1], program), \
282 (N<3)?0:src_vector( p, &inst->SrcReg[2], program)); \
285 #define EMIT_1ARG_ARITH( OP ) EMIT_ARITH( OP, 1 )
286 #define EMIT_2ARG_ARITH( OP ) EMIT_ARITH( OP, 2 )
287 #define EMIT_3ARG_ARITH( OP ) EMIT_ARITH( OP, 3 )
290 * TODO: consider moving this into core
292 static bool calc_live_regs( struct i915_fragment_program *p )
294 const struct gl_fragment_program *program = &p->FragProg;
295 GLuint regsUsed = ~((1 << I915_MAX_TEMPORARY) - 1);
296 uint8_t live_components[I915_MAX_TEMPORARY] = { 0, };
299 for (i = program->Base.NumInstructions - 1; i >= 0; i--) {
300 struct prog_instruction *inst = &program->Base.Instructions[i];
301 int opArgs = _mesa_num_inst_src_regs(inst->Opcode);
304 /* Register is written to: unmark as live for this and preceeding ops */
305 if (inst->DstReg.File == PROGRAM_TEMPORARY) {
306 if (inst->DstReg.Index >= I915_MAX_TEMPORARY)
309 live_components[inst->DstReg.Index] &= ~inst->DstReg.WriteMask;
310 if (live_components[inst->DstReg.Index] == 0)
311 regsUsed &= ~(1 << inst->DstReg.Index);
314 for (a = 0; a < opArgs; a++) {
315 /* Register is read from: mark as live for this and preceeding ops */
316 if (inst->SrcReg[a].File == PROGRAM_TEMPORARY) {
319 if (inst->SrcReg[a].Index >= I915_MAX_TEMPORARY)
322 regsUsed |= 1 << inst->SrcReg[a].Index;
324 for (c = 0; c < 4; c++) {
325 const unsigned field = GET_SWZ(inst->SrcReg[a].Swizzle, c);
327 if (field <= SWIZZLE_W)
328 live_components[inst->SrcReg[a].Index] |= (1U << field);
333 p->usedRegs[i] = regsUsed;
339 static GLuint get_live_regs( struct i915_fragment_program *p,
340 const struct prog_instruction *inst )
342 const struct gl_fragment_program *program = &p->FragProg;
343 GLuint nr = inst - program->Base.Instructions;
345 return p->usedRegs[nr];
349 /* Possible concerns:
351 * SIN, COS -- could use another taylor step?
352 * LIT -- results seem a little different to sw mesa
353 * LOG -- different to mesa on negative numbers, but this is conformant.
355 * Parse failures -- Mesa doesn't currently give a good indication
356 * internally whether a particular program string parsed or not. This
357 * can lead to confusion -- hopefully we cope with it ok now.
361 upload_program(struct i915_fragment_program *p)
363 const struct gl_fragment_program *program = &p->FragProg;
364 const struct prog_instruction *inst = program->Base.Instructions;
366 if (INTEL_DEBUG & DEBUG_WM)
367 _mesa_print_program(&program->Base);
369 /* Is this a parse-failed program? Ensure a valid program is
370 * loaded, as the flagging of an error isn't sufficient to stop
371 * this being uploaded to hardware.
373 if (inst[0].Opcode == OPCODE_END) {
374 GLuint tmp = i915_get_utemp(p);
377 UREG(REG_TYPE_OC, 0),
378 A0_DEST_CHANNEL_ALL, 0,
379 swizzle(tmp, ONE, ZERO, ONE, ONE), 0, 0);
383 if (program->Base.NumInstructions > I915_MAX_INSN) {
384 i915_program_error(p, "Exceeded max instructions (%d out of %d)",
385 program->Base.NumInstructions, I915_MAX_INSN);
389 /* Not always needed:
391 if (!calc_live_regs(p)) {
392 i915_program_error(p, "Could not allocate registers");
397 GLuint src0, src1, src2, flags;
398 GLuint tmp = 0, dst, consts0 = 0, consts1 = 0;
400 switch (inst->Opcode) {
402 src0 = src_vector(p, &inst->SrcReg[0], program);
405 get_result_vector(p, inst),
406 get_result_flags(inst), 0,
407 src0, negate(src0, 1, 1, 1, 1), 0);
411 EMIT_2ARG_ARITH(A0_ADD);
415 src0 = src_vector(p, &inst->SrcReg[0], program);
416 src1 = src_vector(p, &inst->SrcReg[1], program);
417 src2 = src_vector(p, &inst->SrcReg[2], program);
418 i915_emit_arith(p, A0_CMP, get_result_vector(p, inst), get_result_flags(inst), 0, src0, src2, src1); /* NOTE: order of src2, src1 */
422 src0 = src_vector(p, &inst->SrcReg[0], program);
423 tmp = i915_get_utemp(p);
424 consts0 = i915_emit_const4fv(p, sin_quad_constants[0]);
425 consts1 = i915_emit_const4fv(p, sin_quad_constants[1]);
427 /* Reduce range from repeating about [-pi,pi] to [-1,1] */
430 tmp, A0_DEST_CHANNEL_X, 0,
432 swizzle(consts1, Z, ZERO, ZERO, ZERO), /* 1/(2pi) */
433 swizzle(consts0, W, ZERO, ZERO, ZERO)); /* .75 */
435 i915_emit_arith(p, A0_FRC, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
439 tmp, A0_DEST_CHANNEL_X, 0,
441 swizzle(consts0, X, ZERO, ZERO, ZERO), /* 2 */
442 swizzle(consts0, Y, ZERO, ZERO, ZERO)); /* -1 */
444 /* Compute COS with the same calculation used for SIN, but a
445 * different source range has been mapped to [-1,1] this time.
448 /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
451 tmp, A0_DEST_CHANNEL_Y, 0,
452 swizzle(tmp, ZERO, X, ZERO, ZERO),
453 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
456 /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
459 tmp, A0_DEST_CHANNEL_Y, 0,
460 swizzle(tmp, ZERO, X, ZERO, ZERO),
464 /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
467 tmp, A0_DEST_CHANNEL_X, 0,
469 swizzle(consts1, X, Y, ZERO, ZERO),
472 /* tmp.x now contains a first approximation (y). Now, weight it
473 * against tmp.y**2 to get closer.
477 tmp, A0_DEST_CHANNEL_Y, 0,
478 swizzle(tmp, ZERO, X, ZERO, ZERO),
479 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
482 /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
485 tmp, A0_DEST_CHANNEL_Y, 0,
486 swizzle(tmp, ZERO, X, ZERO, ZERO),
487 swizzle(tmp, ZERO, Y, ZERO, ZERO),
488 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0));
490 /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
493 get_result_vector(p, inst),
494 get_result_flags(inst), 0,
495 swizzle(consts1, W, W, W, W),
496 swizzle(tmp, Y, Y, Y, Y),
497 swizzle(tmp, X, X, X, X));
501 src0 = src_vector(p, &inst->SrcReg[0], program);
502 src1 = src_vector(p, &inst->SrcReg[1], program);
505 get_result_vector(p, inst),
506 get_result_flags(inst), 0,
507 swizzle(src0, X, Y, ZERO, ZERO),
508 swizzle(src1, X, Y, ZERO, ZERO),
513 EMIT_2ARG_ARITH(A0_DP3);
517 EMIT_2ARG_ARITH(A0_DP4);
521 src0 = src_vector(p, &inst->SrcReg[0], program);
522 src1 = src_vector(p, &inst->SrcReg[1], program);
526 get_result_vector(p, inst),
527 get_result_flags(inst), 0,
528 swizzle(src0, X, Y, Z, ONE), src1, 0);
532 src0 = src_vector(p, &inst->SrcReg[0], program);
533 src1 = src_vector(p, &inst->SrcReg[1], program);
535 /* result[0] = 1 * 1;
536 * result[1] = a[1] * b[1];
537 * result[2] = a[2] * 1;
538 * result[3] = 1 * b[3];
542 get_result_vector(p, inst),
543 get_result_flags(inst), 0,
544 swizzle(src0, ONE, Y, Z, ONE),
545 swizzle(src1, ONE, Y, ONE, W), 0);
549 src0 = src_vector(p, &inst->SrcReg[0], program);
553 get_result_vector(p, inst),
554 get_result_flags(inst), 0,
555 swizzle(src0, X, X, X, X), 0, 0);
559 EMIT_1ARG_ARITH(A0_FLR);
563 EMIT_1ARG_ARITH(A0_TRC);
567 EMIT_1ARG_ARITH(A0_FRC);
571 src0 = src_vector(p, &inst->SrcReg[0], program);
572 tmp = i915_get_utemp(p);
574 i915_emit_texld(p, get_live_regs(p, inst),
575 tmp, A0_DEST_CHANNEL_ALL, /* use a dummy dest reg */
576 0, src0, T0_TEXKILL);
580 if (inst->DstReg.CondMask == COND_TR) {
581 tmp = i915_get_utemp(p);
583 /* The KIL instruction discards the fragment if any component of
584 * the source is < 0. Emit an immediate operand of {-1}.xywz.
586 i915_emit_texld(p, get_live_regs(p, inst),
587 tmp, A0_DEST_CHANNEL_ALL,
588 0, /* use a dummy dest reg */
589 negate(swizzle(tmp, ONE, ONE, ONE, ONE),
594 i915_program_error(p, "Unsupported KIL_NV condition code: %d",
595 inst->DstReg.CondMask);
600 src0 = src_vector(p, &inst->SrcReg[0], program);
604 get_result_vector(p, inst),
605 get_result_flags(inst), 0,
606 swizzle(src0, X, X, X, X), 0, 0);
610 src0 = src_vector(p, &inst->SrcReg[0], program);
611 tmp = i915_get_utemp(p);
613 /* tmp = max( a.xyzw, a.00zw )
614 * XXX: Clamp tmp.w to -128..128
616 * tmp.y = tmp.w * tmp.y
618 * result = cmp (a.11-x1, a.1x01, a.1xy1 )
620 i915_emit_arith(p, A0_MAX, tmp, A0_DEST_CHANNEL_ALL, 0,
621 src0, swizzle(src0, ZERO, ZERO, Z, W), 0);
623 i915_emit_arith(p, A0_LOG, tmp, A0_DEST_CHANNEL_Y, 0,
624 swizzle(tmp, Y, Y, Y, Y), 0, 0);
626 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_Y, 0,
627 swizzle(tmp, ZERO, Y, ZERO, ZERO),
628 swizzle(tmp, ZERO, W, ZERO, ZERO), 0);
630 i915_emit_arith(p, A0_EXP, tmp, A0_DEST_CHANNEL_Y, 0,
631 swizzle(tmp, Y, Y, Y, Y), 0, 0);
633 i915_emit_arith(p, A0_CMP,
634 get_result_vector(p, inst),
635 get_result_flags(inst), 0,
636 negate(swizzle(tmp, ONE, ONE, X, ONE), 0, 0, 1, 0),
637 swizzle(tmp, ONE, X, ZERO, ONE),
638 swizzle(tmp, ONE, X, Y, ONE));
643 src0 = src_vector(p, &inst->SrcReg[0], program);
644 src1 = src_vector(p, &inst->SrcReg[1], program);
645 src2 = src_vector(p, &inst->SrcReg[2], program);
646 flags = get_result_flags(inst);
647 tmp = i915_get_utemp(p);
654 * result = (-c)*a + tmp
656 i915_emit_arith(p, A0_MAD, tmp,
657 flags & A0_DEST_CHANNEL_ALL, 0, src1, src0, src2);
659 i915_emit_arith(p, A0_MAD,
660 get_result_vector(p, inst),
661 flags, 0, negate(src2, 1, 1, 1, 1), src0, tmp);
665 EMIT_3ARG_ARITH(A0_MAD);
669 EMIT_2ARG_ARITH(A0_MAX);
673 src0 = src_vector(p, &inst->SrcReg[0], program);
674 src1 = src_vector(p, &inst->SrcReg[1], program);
675 tmp = i915_get_utemp(p);
676 flags = get_result_flags(inst);
680 tmp, flags & A0_DEST_CHANNEL_ALL, 0,
681 negate(src0, 1, 1, 1, 1),
682 negate(src1, 1, 1, 1, 1), 0);
686 get_result_vector(p, inst),
687 flags, 0, negate(tmp, 1, 1, 1, 1), 0, 0);
691 EMIT_1ARG_ARITH(A0_MOV);
695 EMIT_2ARG_ARITH(A0_MUL);
699 src0 = src_vector(p, &inst->SrcReg[0], program);
700 src1 = src_vector(p, &inst->SrcReg[1], program);
701 tmp = i915_get_utemp(p);
702 flags = get_result_flags(inst);
704 /* XXX: masking on intermediate values, here and elsewhere.
708 tmp, A0_DEST_CHANNEL_X, 0,
709 swizzle(src0, X, X, X, X), 0, 0);
711 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_X, 0, tmp, src1, 0);
716 get_result_vector(p, inst),
717 flags, 0, swizzle(tmp, X, X, X, X), 0, 0);
722 src0 = src_vector(p, &inst->SrcReg[0], program);
726 get_result_vector(p, inst),
727 get_result_flags(inst), 0,
728 swizzle(src0, X, X, X, X), 0, 0);
733 src0 = src_vector(p, &inst->SrcReg[0], program);
737 get_result_vector(p, inst),
738 get_result_flags(inst), 0,
739 swizzle(src0, X, X, X, X), 0, 0);
743 src0 = src_vector(p, &inst->SrcReg[0], program);
744 tmp = i915_get_utemp(p);
747 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
748 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
749 * t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x
750 * scs.x = DP4 t1, sin_constants
751 * t1 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1
752 * scs.y = DP4 t1, cos_constants
756 tmp, A0_DEST_CHANNEL_XY, 0,
757 swizzle(src0, X, X, ONE, ONE),
758 swizzle(src0, X, ONE, ONE, ONE), 0);
762 tmp, A0_DEST_CHANNEL_ALL, 0,
763 swizzle(tmp, X, Y, X, Y),
764 swizzle(tmp, X, X, ONE, ONE), 0);
766 if (inst->DstReg.WriteMask & WRITEMASK_Y) {
769 if (inst->DstReg.WriteMask & WRITEMASK_X)
770 tmp1 = i915_get_utemp(p);
776 tmp1, A0_DEST_CHANNEL_ALL, 0,
777 swizzle(tmp, X, Y, Y, W),
778 swizzle(tmp, X, Z, ONE, ONE), 0);
782 get_result_vector(p, inst),
783 A0_DEST_CHANNEL_Y, 0,
784 swizzle(tmp1, W, Z, Y, X),
785 i915_emit_const4fv(p, sin_constants), 0);
788 if (inst->DstReg.WriteMask & WRITEMASK_X) {
791 tmp, A0_DEST_CHANNEL_XYZ, 0,
792 swizzle(tmp, X, X, Z, ONE),
793 swizzle(tmp, Z, ONE, ONE, ONE), 0);
797 get_result_vector(p, inst),
798 A0_DEST_CHANNEL_X, 0,
799 swizzle(tmp, ONE, Z, Y, X),
800 i915_emit_const4fv(p, cos_constants), 0);
805 tmp = i915_get_utemp(p);
806 flags = get_result_flags(inst);
807 dst = get_result_vector(p, inst);
809 /* If both operands are uniforms or constants, we get 5 instructions
812 * U[1] = MOV CONST[1]
813 * U[0].xyz = SGE CONST[0].xxxx, U[1]
814 * U[1] = MOV CONST[1].-x-y-z-w
815 * R[0].xyz = SGE CONST[0].-x-x-x-x, U[1]
816 * R[0].xyz = MUL R[0], U[0]
818 * This code is stupid. Instead of having the individual calls to
819 * i915_emit_arith generate the moves to utemps, do it in the caller.
820 * This results in code like:
822 * U[1] = MOV CONST[1]
823 * U[0].xyz = SGE CONST[0].xxxx, U[1]
824 * R[0].xyz = SGE CONST[0].-x-x-x-x, U[1].-x-y-z-w
825 * R[0].xyz = MUL R[0], U[0]
827 src0 = src_vector(p, &inst->SrcReg[0], program);
828 src1 = src_vector(p, &inst->SrcReg[1], program);
830 if (GET_UREG_TYPE(src0) == REG_TYPE_CONST
831 && GET_UREG_TYPE(src1) == REG_TYPE_CONST) {
832 unsigned tmp = i915_get_utemp(p);
834 i915_emit_arith(p, A0_MOV, tmp, A0_DEST_CHANNEL_ALL, 0,
840 /* tmp = src1 >= src2 */
848 /* dst = src1 <= src2 */
853 negate(src0, 1, 1, 1, 1),
854 negate(src1, 1, 1, 1, 1),
856 /* dst = tmp && dst */
867 src0 = src_vector(p, &inst->SrcReg[0], program);
868 tmp = i915_get_utemp(p);
869 consts0 = i915_emit_const4fv(p, sin_quad_constants[0]);
870 consts1 = i915_emit_const4fv(p, sin_quad_constants[1]);
872 /* Reduce range from repeating about [-pi,pi] to [-1,1] */
875 tmp, A0_DEST_CHANNEL_X, 0,
877 swizzle(consts1, Z, ZERO, ZERO, ZERO), /* 1/(2pi) */
878 swizzle(consts0, Z, ZERO, ZERO, ZERO)); /* .5 */
880 i915_emit_arith(p, A0_FRC, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
884 tmp, A0_DEST_CHANNEL_X, 0,
886 swizzle(consts0, X, ZERO, ZERO, ZERO), /* 2 */
887 swizzle(consts0, Y, ZERO, ZERO, ZERO)); /* -1 */
889 /* Compute sin using a quadratic and quartic. It gives continuity
890 * that repeating the Taylor series lacks every 2*pi, and has
893 * The idea was described at:
894 * http://www.devmaster.net/forums/showthread.php?t=5784
897 /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
900 tmp, A0_DEST_CHANNEL_Y, 0,
901 swizzle(tmp, ZERO, X, ZERO, ZERO),
902 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
905 /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
908 tmp, A0_DEST_CHANNEL_Y, 0,
909 swizzle(tmp, ZERO, X, ZERO, ZERO),
913 /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
916 tmp, A0_DEST_CHANNEL_X, 0,
918 swizzle(consts1, X, Y, ZERO, ZERO),
921 /* tmp.x now contains a first approximation (y). Now, weight it
922 * against tmp.y**2 to get closer.
926 tmp, A0_DEST_CHANNEL_Y, 0,
927 swizzle(tmp, ZERO, X, ZERO, ZERO),
928 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
931 /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
934 tmp, A0_DEST_CHANNEL_Y, 0,
935 swizzle(tmp, ZERO, X, ZERO, ZERO),
936 swizzle(tmp, ZERO, Y, ZERO, ZERO),
937 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0));
939 /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
942 get_result_vector(p, inst),
943 get_result_flags(inst), 0,
944 swizzle(consts1, W, W, W, W),
945 swizzle(tmp, Y, Y, Y, Y),
946 swizzle(tmp, X, X, X, X));
951 EMIT_2ARG_ARITH(A0_SGE);
957 get_result_vector( p, inst ),
958 get_result_flags( inst ), 0,
959 negate(src_vector( p, &inst->SrcReg[0], program),
961 negate(src_vector( p, &inst->SrcReg[1], program),
969 get_result_vector( p, inst ),
970 get_result_flags( inst ), 0,
971 negate(src_vector( p, &inst->SrcReg[0], program),
973 negate(src_vector( p, &inst->SrcReg[1], program),
979 EMIT_2ARG_ARITH(A0_SLT);
983 tmp = i915_get_utemp(p);
984 flags = get_result_flags(inst);
985 dst = get_result_vector(p, inst);
987 /* If both operands are uniforms or constants, we get 5 instructions
990 * U[1] = MOV CONST[1]
991 * U[0].xyz = SLT CONST[0].xxxx, U[1]
992 * U[1] = MOV CONST[1].-x-y-z-w
993 * R[0].xyz = SLT CONST[0].-x-x-x-x, U[1]
994 * R[0].xyz = MUL R[0], U[0]
996 * This code is stupid. Instead of having the individual calls to
997 * i915_emit_arith generate the moves to utemps, do it in the caller.
998 * This results in code like:
1000 * U[1] = MOV CONST[1]
1001 * U[0].xyz = SLT CONST[0].xxxx, U[1]
1002 * R[0].xyz = SLT CONST[0].-x-x-x-x, U[1].-x-y-z-w
1003 * R[0].xyz = MUL R[0], U[0]
1005 src0 = src_vector(p, &inst->SrcReg[0], program);
1006 src1 = src_vector(p, &inst->SrcReg[1], program);
1008 if (GET_UREG_TYPE(src0) == REG_TYPE_CONST
1009 && GET_UREG_TYPE(src1) == REG_TYPE_CONST) {
1010 unsigned tmp = i915_get_utemp(p);
1012 i915_emit_arith(p, A0_MOV, tmp, A0_DEST_CHANNEL_ALL, 0,
1018 /* tmp = src1 < src2 */
1026 /* dst = src1 > src2 */
1031 negate(src0, 1, 1, 1, 1),
1032 negate(src1, 1, 1, 1, 1),
1034 /* dst = tmp || dst */
1038 flags | A0_DEST_SATURATE, 0,
1045 dst = get_result_vector(p, inst);
1046 flags = get_result_flags(inst);
1047 src0 = src_vector(p, &inst->SrcReg[0], program);
1048 tmp = i915_get_utemp(p);
1050 /* tmp = (src < 0.0) */
1056 swizzle(src0, ZERO, ZERO, ZERO, ZERO),
1059 /* dst = (0.0 < src) */
1064 swizzle(src0, ZERO, ZERO, ZERO, ZERO),
1068 /* dst = (src > 0.0) - (src < 0.0) */
1074 negate(tmp, 1, 1, 1, 1),
1080 src0 = src_vector(p, &inst->SrcReg[0], program);
1081 src1 = src_vector(p, &inst->SrcReg[1], program);
1085 get_result_vector(p, inst),
1086 get_result_flags(inst), 0,
1087 src0, negate(src1, 1, 1, 1, 1), 0);
1091 EMIT_1ARG_ARITH(A0_MOV); /* extended swizzle handled natively */
1099 EMIT_TEX(T0_TEXLDB);
1103 EMIT_TEX(T0_TEXLDP);
1108 * result.x = src0.y * src1.z - src0.z * src1.y;
1109 * result.y = src0.z * src1.x - src0.x * src1.z;
1110 * result.z = src0.x * src1.y - src0.y * src1.x;
1113 src0 = src_vector(p, &inst->SrcReg[0], program);
1114 src1 = src_vector(p, &inst->SrcReg[1], program);
1115 tmp = i915_get_utemp(p);
1119 tmp, A0_DEST_CHANNEL_ALL, 0,
1120 swizzle(src0, Z, X, Y, ONE),
1121 swizzle(src1, Y, Z, X, ONE), 0);
1125 get_result_vector(p, inst),
1126 get_result_flags(inst), 0,
1127 swizzle(src0, Y, Z, X, ONE),
1128 swizzle(src1, Z, X, Y, ONE),
1129 negate(tmp, 1, 1, 1, 0));
1135 case OPCODE_BGNLOOP:
1144 case OPCODE_ENDLOOP:
1149 i915_program_error(p, "Unsupported opcode: %s",
1150 _mesa_opcode_string(inst->Opcode));
1155 /* These opcodes are claimed as GLSL, NV_vp, and ARB_vp in
1156 * prog_instruction.h, but apparently GLSL doesn't ever emit them.
1157 * Instead, it translates to EX2 or LG2.
1161 /* These opcodes are claimed by GLSL in prog_instruction.h, but
1162 * only NV_vp/fp appears to emit them.
1165 i915_program_error(p, "bad opcode: %s",
1166 _mesa_opcode_string(inst->Opcode));
1171 i915_release_utemps(p);
1175 /* Rather than trying to intercept and jiggle depth writes during
1176 * emit, just move the value into its correct position at the end of
1180 fixup_depth_write(struct i915_fragment_program *p)
1182 if (p->depth_written) {
1183 GLuint depth = UREG(REG_TYPE_OD, 0);
1187 depth, A0_DEST_CHANNEL_W, 0,
1188 swizzle(depth, X, Y, Z, Z), 0, 0);
1194 check_wpos(struct i915_fragment_program *p)
1196 GLbitfield64 inputs = p->FragProg.Base.InputsRead;
1201 for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
1202 if (inputs & (VARYING_BIT_TEX(i) | VARYING_BIT_VAR(i)))
1204 else if (inputs & VARYING_BIT_POS) {
1206 inputs &= ~VARYING_BIT_POS;
1210 if (inputs & VARYING_BIT_POS) {
1211 i915_program_error(p, "No free texcoord for wpos value");
1217 translate_program(struct i915_fragment_program *p)
1219 struct i915_context *i915 = I915_CONTEXT(p->ctx);
1221 if (INTEL_DEBUG & DEBUG_WM) {
1223 _mesa_print_program(&p->FragProg.Base);
1227 i915_init_program(i915, p);
1230 fixup_depth_write(p);
1231 i915_fini_program(p);
1238 track_params(struct i915_fragment_program *p)
1243 _mesa_load_state_parameters(p->ctx, p->FragProg.Base.Parameters);
1245 for (i = 0; i < p->nr_params; i++) {
1246 GLint reg = p->param[i].reg;
1247 COPY_4V(p->constant[reg], p->param[i].values);
1250 p->params_uptodate = 1;
1251 p->on_hardware = 0; /* overkill */
1256 i915BindProgram(struct gl_context * ctx, GLenum target, struct gl_program *prog)
1258 if (target == GL_FRAGMENT_PROGRAM_ARB) {
1259 struct i915_context *i915 = I915_CONTEXT(ctx);
1260 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1262 if (i915->current_program == p)
1265 if (i915->current_program) {
1266 i915->current_program->on_hardware = 0;
1267 i915->current_program->params_uptodate = 0;
1270 i915->current_program = p;
1272 assert(p->on_hardware == 0);
1273 assert(p->params_uptodate == 0);
1278 static struct gl_program *
1279 i915NewProgram(struct gl_context * ctx, GLenum target, GLuint id)
1282 case GL_VERTEX_PROGRAM_ARB:
1283 return _mesa_init_vertex_program(ctx, CALLOC_STRUCT(gl_vertex_program),
1286 case GL_FRAGMENT_PROGRAM_ARB:{
1287 struct i915_fragment_program *prog =
1288 CALLOC_STRUCT(i915_fragment_program);
1290 i915_init_program(I915_CONTEXT(ctx), prog);
1292 return _mesa_init_fragment_program(ctx, &prog->FragProg,
1302 return _mesa_new_program(ctx, target, id);
1307 i915DeleteProgram(struct gl_context * ctx, struct gl_program *prog)
1309 if (prog->Target == GL_FRAGMENT_PROGRAM_ARB) {
1310 struct i915_context *i915 = I915_CONTEXT(ctx);
1311 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1313 if (i915->current_program == p)
1314 i915->current_program = 0;
1317 _mesa_delete_program(ctx, prog);
1322 i915IsProgramNative(struct gl_context * ctx, GLenum target, struct gl_program *prog)
1324 if (target == GL_FRAGMENT_PROGRAM_ARB) {
1325 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1328 translate_program(p);
1337 i915ProgramStringNotify(struct gl_context * ctx,
1338 GLenum target, struct gl_program *prog)
1340 if (target == GL_FRAGMENT_PROGRAM_ARB) {
1341 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1345 (void) _tnl_program_string(ctx, target, prog);
1347 /* XXX check if program is legal, within limits */
1352 i915SamplerUniformChange(struct gl_context *ctx,
1353 GLenum target, struct gl_program *prog)
1355 i915ProgramStringNotify(ctx, target, prog);
1359 i915_update_program(struct gl_context *ctx)
1361 struct intel_context *intel = intel_context(ctx);
1362 struct i915_context *i915 = i915_context(&intel->ctx);
1363 struct i915_fragment_program *fp =
1364 (struct i915_fragment_program *) ctx->FragmentProgram._Current;
1366 if (i915->current_program != fp) {
1367 if (i915->current_program) {
1368 i915->current_program->on_hardware = 0;
1369 i915->current_program->params_uptodate = 0;
1372 i915->current_program = fp;
1375 if (!fp->translated)
1376 translate_program(fp);
1378 FALLBACK(&i915->intel, I915_FALLBACK_PROGRAM, fp->error);
1382 i915ValidateFragmentProgram(struct i915_context *i915)
1384 struct gl_context *ctx = &i915->intel.ctx;
1385 struct intel_context *intel = intel_context(ctx);
1386 TNLcontext *tnl = TNL_CONTEXT(ctx);
1387 struct vertex_buffer *VB = &tnl->vb;
1389 struct i915_fragment_program *p =
1390 (struct i915_fragment_program *) ctx->FragmentProgram._Current;
1392 const GLbitfield64 inputsRead = p->FragProg.Base.InputsRead;
1393 GLuint s4 = i915->state.Ctx[I915_CTXREG_LIS4] & ~S4_VFMT_MASK;
1394 GLuint s2 = S2_TEXCOORD_NONE;
1399 VB->AttribPtr[VERT_ATTRIB_POS] = VB->NdcPtr;
1402 translate_program(p);
1404 intel->vertex_attr_count = 0;
1405 intel->wpos_offset = 0;
1406 intel->coloroffset = 0;
1407 intel->specoffset = 0;
1409 if (inputsRead & VARYING_BITS_TEX_ANY || p->wpos_tex != -1) {
1410 EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_4F_VIEWPORT, S4_VFMT_XYZW, 16);
1413 EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_3F_VIEWPORT, S4_VFMT_XYZ, 12);
1416 /* Handle gl_PointSize builtin var here */
1417 if (ctx->Point._Attenuated || ctx->VertexProgram.PointSizeEnabled)
1418 EMIT_ATTR(_TNL_ATTRIB_POINTSIZE, EMIT_1F, S4_VFMT_POINT_WIDTH, 4);
1420 if (inputsRead & VARYING_BIT_COL0) {
1421 intel->coloroffset = offset / 4;
1422 EMIT_ATTR(_TNL_ATTRIB_COLOR0, EMIT_4UB_4F_BGRA, S4_VFMT_COLOR, 4);
1425 if (inputsRead & VARYING_BIT_COL1) {
1426 intel->specoffset = offset / 4;
1427 EMIT_ATTR(_TNL_ATTRIB_COLOR1, EMIT_4UB_4F_BGRA, S4_VFMT_SPEC_FOG, 4);
1430 if ((inputsRead & VARYING_BIT_FOGC)) {
1431 EMIT_ATTR(_TNL_ATTRIB_FOG, EMIT_1F, S4_VFMT_FOG_PARAM, 4);
1434 for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
1435 if (inputsRead & VARYING_BIT_TEX(i)) {
1436 int sz = VB->AttribPtr[_TNL_ATTRIB_TEX0 + i]->size;
1438 s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1439 s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(sz));
1441 EMIT_ATTR(_TNL_ATTRIB_TEX0 + i, EMIT_SZ(sz), 0, sz * 4);
1443 else if (inputsRead & VARYING_BIT_VAR(i)) {
1444 int sz = VB->AttribPtr[_TNL_ATTRIB_GENERIC0 + i]->size;
1446 s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1447 s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(sz));
1449 EMIT_ATTR(_TNL_ATTRIB_GENERIC0 + i, EMIT_SZ(sz), 0, sz * 4);
1451 else if (i == p->wpos_tex) {
1452 int wpos_size = 4 * sizeof(float);
1453 /* If WPOS is required, duplicate the XYZ position data in an
1454 * unused texture coordinate:
1456 s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1457 s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(wpos_size));
1459 intel->wpos_offset = offset;
1460 EMIT_PAD(wpos_size);
1464 if (s2 != i915->state.Ctx[I915_CTXREG_LIS2] ||
1465 s4 != i915->state.Ctx[I915_CTXREG_LIS4]) {
1468 I915_STATECHANGE(i915, I915_UPLOAD_CTX);
1470 /* Must do this *after* statechange, so as not to affect
1471 * buffered vertices reliant on the old state:
1473 intel->vertex_size = _tnl_install_attrs(&intel->ctx,
1474 intel->vertex_attrs,
1475 intel->vertex_attr_count,
1476 intel->ViewportMatrix.m, 0);
1478 assert(intel->prim.current_offset == intel->prim.start_offset);
1479 intel->prim.start_offset = (intel->prim.current_offset + intel->vertex_size-1) / intel->vertex_size * intel->vertex_size;
1480 intel->prim.current_offset = intel->prim.start_offset;
1482 intel->vertex_size >>= 2;
1484 i915->state.Ctx[I915_CTXREG_LIS2] = s2;
1485 i915->state.Ctx[I915_CTXREG_LIS4] = s4;
1487 k = intel->vtbl.check_vertex_size(intel, intel->vertex_size);
1491 if (!p->params_uptodate)
1494 if (!p->on_hardware)
1495 i915_upload_program(i915, p);
1497 if (INTEL_DEBUG & DEBUG_WM) {
1499 i915_disassemble_program(i915->state.Program, i915->state.ProgramSize);
1504 i915InitFragProgFuncs(struct dd_function_table *functions)
1506 functions->BindProgram = i915BindProgram;
1507 functions->NewProgram = i915NewProgram;
1508 functions->DeleteProgram = i915DeleteProgram;
1509 functions->IsProgramNative = i915IsProgramNative;
1510 functions->ProgramStringNotify = i915ProgramStringNotify;
1511 functions->SamplerUniformChange = i915SamplerUniformChange;