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:
178 src = i915_emit_param4fv(p,
179 &program->Base.Parameters->ParameterValues[source->Index][0].f);
183 i915_program_error(p, "Bad source->File: %d", source->File);
188 GET_SWZ(source->Swizzle, 0),
189 GET_SWZ(source->Swizzle, 1),
190 GET_SWZ(source->Swizzle, 2), GET_SWZ(source->Swizzle, 3));
194 GET_BIT(source->Negate, 0),
195 GET_BIT(source->Negate, 1),
196 GET_BIT(source->Negate, 2),
197 GET_BIT(source->Negate, 3));
204 get_result_vector(struct i915_fragment_program *p,
205 const struct prog_instruction *inst)
207 switch (inst->DstReg.File) {
209 switch (inst->DstReg.Index) {
210 case FRAG_RESULT_COLOR:
211 case FRAG_RESULT_DATA0:
212 return UREG(REG_TYPE_OC, 0);
213 case FRAG_RESULT_DEPTH:
214 p->depth_written = 1;
215 return UREG(REG_TYPE_OD, 0);
217 i915_program_error(p, "Bad inst->DstReg.Index: %d",
221 case PROGRAM_TEMPORARY:
222 return UREG(REG_TYPE_R, inst->DstReg.Index);
224 i915_program_error(p, "Bad inst->DstReg.File: %d", inst->DstReg.File);
230 get_result_flags(const struct prog_instruction *inst)
234 if (inst->SaturateMode == SATURATE_ZERO_ONE)
235 flags |= A0_DEST_SATURATE;
236 if (inst->DstReg.WriteMask & WRITEMASK_X)
237 flags |= A0_DEST_CHANNEL_X;
238 if (inst->DstReg.WriteMask & WRITEMASK_Y)
239 flags |= A0_DEST_CHANNEL_Y;
240 if (inst->DstReg.WriteMask & WRITEMASK_Z)
241 flags |= A0_DEST_CHANNEL_Z;
242 if (inst->DstReg.WriteMask & WRITEMASK_W)
243 flags |= A0_DEST_CHANNEL_W;
249 translate_tex_src_target(struct i915_fragment_program *p, GLubyte bit)
252 case TEXTURE_1D_INDEX:
253 return D0_SAMPLE_TYPE_2D;
254 case TEXTURE_2D_INDEX:
255 return D0_SAMPLE_TYPE_2D;
256 case TEXTURE_RECT_INDEX:
257 return D0_SAMPLE_TYPE_2D;
258 case TEXTURE_3D_INDEX:
259 return D0_SAMPLE_TYPE_VOLUME;
260 case TEXTURE_CUBE_INDEX:
261 return D0_SAMPLE_TYPE_CUBE;
263 i915_program_error(p, "TexSrcBit: %d", bit);
268 #define EMIT_TEX( OP ) \
270 GLuint dim = translate_tex_src_target( p, inst->TexSrcTarget ); \
271 const struct gl_fragment_program *program = &p->FragProg; \
272 GLuint unit = program->Base.SamplerUnits[inst->TexSrcUnit]; \
273 GLuint sampler = i915_emit_decl(p, REG_TYPE_S, \
275 GLuint coord = src_vector( p, &inst->SrcReg[0], program); \
278 i915_emit_texld( p, get_live_regs(p, inst), \
279 get_result_vector( p, inst ), \
280 get_result_flags( inst ), \
286 #define EMIT_ARITH( OP, N ) \
288 i915_emit_arith( p, \
290 get_result_vector( p, inst ), \
291 get_result_flags( inst ), 0, \
292 (N<1)?0:src_vector( p, &inst->SrcReg[0], program), \
293 (N<2)?0:src_vector( p, &inst->SrcReg[1], program), \
294 (N<3)?0:src_vector( p, &inst->SrcReg[2], program)); \
297 #define EMIT_1ARG_ARITH( OP ) EMIT_ARITH( OP, 1 )
298 #define EMIT_2ARG_ARITH( OP ) EMIT_ARITH( OP, 2 )
299 #define EMIT_3ARG_ARITH( OP ) EMIT_ARITH( OP, 3 )
302 * TODO: consider moving this into core
304 static bool calc_live_regs( struct i915_fragment_program *p )
306 const struct gl_fragment_program *program = &p->FragProg;
307 GLuint regsUsed = 0xffff0000;
308 uint8_t live_components[16] = { 0, };
311 for (i = program->Base.NumInstructions - 1; i >= 0; i--) {
312 struct prog_instruction *inst = &program->Base.Instructions[i];
313 int opArgs = _mesa_num_inst_src_regs(inst->Opcode);
316 /* Register is written to: unmark as live for this and preceeding ops */
317 if (inst->DstReg.File == PROGRAM_TEMPORARY) {
318 if (inst->DstReg.Index > 16)
321 live_components[inst->DstReg.Index] &= ~inst->DstReg.WriteMask;
322 if (live_components[inst->DstReg.Index] == 0)
323 regsUsed &= ~(1 << inst->DstReg.Index);
326 for (a = 0; a < opArgs; a++) {
327 /* Register is read from: mark as live for this and preceeding ops */
328 if (inst->SrcReg[a].File == PROGRAM_TEMPORARY) {
331 if (inst->SrcReg[a].Index > 16)
334 regsUsed |= 1 << inst->SrcReg[a].Index;
336 for (c = 0; c < 4; c++) {
337 const unsigned field = GET_SWZ(inst->SrcReg[a].Swizzle, c);
339 if (field <= SWIZZLE_W)
340 live_components[inst->SrcReg[a].Index] |= (1U << field);
345 p->usedRegs[i] = regsUsed;
351 static GLuint get_live_regs( struct i915_fragment_program *p,
352 const struct prog_instruction *inst )
354 const struct gl_fragment_program *program = &p->FragProg;
355 GLuint nr = inst - program->Base.Instructions;
357 return p->usedRegs[nr];
361 /* Possible concerns:
363 * SIN, COS -- could use another taylor step?
364 * LIT -- results seem a little different to sw mesa
365 * LOG -- different to mesa on negative numbers, but this is conformant.
367 * Parse failures -- Mesa doesn't currently give a good indication
368 * internally whether a particular program string parsed or not. This
369 * can lead to confusion -- hopefully we cope with it ok now.
373 upload_program(struct i915_fragment_program *p)
375 const struct gl_fragment_program *program = &p->FragProg;
376 const struct prog_instruction *inst = program->Base.Instructions;
378 if (INTEL_DEBUG & DEBUG_WM)
379 _mesa_print_program(&program->Base);
381 /* Is this a parse-failed program? Ensure a valid program is
382 * loaded, as the flagging of an error isn't sufficient to stop
383 * this being uploaded to hardware.
385 if (inst[0].Opcode == OPCODE_END) {
386 GLuint tmp = i915_get_utemp(p);
389 UREG(REG_TYPE_OC, 0),
390 A0_DEST_CHANNEL_ALL, 0,
391 swizzle(tmp, ONE, ZERO, ONE, ONE), 0, 0);
395 if (program->Base.NumInstructions > I915_MAX_INSN) {
396 i915_program_error(p, "Exceeded max instructions (%d out of %d)",
397 program->Base.NumInstructions, I915_MAX_INSN);
401 /* Not always needed:
403 if (!calc_live_regs(p)) {
404 i915_program_error(p, "Could not allocate registers");
409 GLuint src0, src1, src2, flags;
410 GLuint tmp = 0, dst, consts0 = 0, consts1 = 0;
412 switch (inst->Opcode) {
414 src0 = src_vector(p, &inst->SrcReg[0], program);
417 get_result_vector(p, inst),
418 get_result_flags(inst), 0,
419 src0, negate(src0, 1, 1, 1, 1), 0);
423 EMIT_2ARG_ARITH(A0_ADD);
427 src0 = src_vector(p, &inst->SrcReg[0], program);
428 src1 = src_vector(p, &inst->SrcReg[1], program);
429 src2 = src_vector(p, &inst->SrcReg[2], program);
430 i915_emit_arith(p, A0_CMP, get_result_vector(p, inst), get_result_flags(inst), 0, src0, src2, src1); /* NOTE: order of src2, src1 */
434 src0 = src_vector(p, &inst->SrcReg[0], program);
435 tmp = i915_get_utemp(p);
436 consts0 = i915_emit_const4fv(p, sin_quad_constants[0]);
437 consts1 = i915_emit_const4fv(p, sin_quad_constants[1]);
439 /* Reduce range from repeating about [-pi,pi] to [-1,1] */
442 tmp, A0_DEST_CHANNEL_X, 0,
444 swizzle(consts1, Z, ZERO, ZERO, ZERO), /* 1/(2pi) */
445 swizzle(consts0, W, ZERO, ZERO, ZERO)); /* .75 */
447 i915_emit_arith(p, A0_FRC, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
451 tmp, A0_DEST_CHANNEL_X, 0,
453 swizzle(consts0, X, ZERO, ZERO, ZERO), /* 2 */
454 swizzle(consts0, Y, ZERO, ZERO, ZERO)); /* -1 */
456 /* Compute COS with the same calculation used for SIN, but a
457 * different source range has been mapped to [-1,1] this time.
460 /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
463 tmp, A0_DEST_CHANNEL_Y, 0,
464 swizzle(tmp, ZERO, X, ZERO, ZERO),
465 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
468 /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
471 tmp, A0_DEST_CHANNEL_Y, 0,
472 swizzle(tmp, ZERO, X, ZERO, ZERO),
476 /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
479 tmp, A0_DEST_CHANNEL_X, 0,
481 swizzle(consts1, X, Y, ZERO, ZERO),
484 /* tmp.x now contains a first approximation (y). Now, weight it
485 * against tmp.y**2 to get closer.
489 tmp, A0_DEST_CHANNEL_Y, 0,
490 swizzle(tmp, ZERO, X, ZERO, ZERO),
491 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
494 /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
497 tmp, A0_DEST_CHANNEL_Y, 0,
498 swizzle(tmp, ZERO, X, ZERO, ZERO),
499 swizzle(tmp, ZERO, Y, ZERO, ZERO),
500 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0));
502 /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
505 get_result_vector(p, inst),
506 get_result_flags(inst), 0,
507 swizzle(consts1, W, W, W, W),
508 swizzle(tmp, Y, Y, Y, Y),
509 swizzle(tmp, X, X, X, X));
513 src0 = src_vector(p, &inst->SrcReg[0], program);
514 src1 = src_vector(p, &inst->SrcReg[1], program);
517 get_result_vector(p, inst),
518 get_result_flags(inst), 0,
519 swizzle(src0, X, Y, ZERO, ZERO),
520 swizzle(src1, X, Y, ZERO, ZERO),
525 EMIT_2ARG_ARITH(A0_DP3);
529 EMIT_2ARG_ARITH(A0_DP4);
533 src0 = src_vector(p, &inst->SrcReg[0], program);
534 src1 = src_vector(p, &inst->SrcReg[1], program);
538 get_result_vector(p, inst),
539 get_result_flags(inst), 0,
540 swizzle(src0, X, Y, Z, ONE), src1, 0);
544 src0 = src_vector(p, &inst->SrcReg[0], program);
545 src1 = src_vector(p, &inst->SrcReg[1], program);
547 /* result[0] = 1 * 1;
548 * result[1] = a[1] * b[1];
549 * result[2] = a[2] * 1;
550 * result[3] = 1 * b[3];
554 get_result_vector(p, inst),
555 get_result_flags(inst), 0,
556 swizzle(src0, ONE, Y, Z, ONE),
557 swizzle(src1, ONE, Y, ONE, W), 0);
561 src0 = src_vector(p, &inst->SrcReg[0], program);
565 get_result_vector(p, inst),
566 get_result_flags(inst), 0,
567 swizzle(src0, X, X, X, X), 0, 0);
571 EMIT_1ARG_ARITH(A0_FLR);
575 EMIT_1ARG_ARITH(A0_TRC);
579 EMIT_1ARG_ARITH(A0_FRC);
583 src0 = src_vector(p, &inst->SrcReg[0], program);
584 tmp = i915_get_utemp(p);
586 i915_emit_texld(p, get_live_regs(p, inst),
587 tmp, A0_DEST_CHANNEL_ALL, /* use a dummy dest reg */
588 0, src0, T0_TEXKILL);
592 if (inst->DstReg.CondMask == COND_TR) {
593 tmp = i915_get_utemp(p);
595 /* The KIL instruction discards the fragment if any component of
596 * the source is < 0. Emit an immediate operand of {-1}.xywz.
598 i915_emit_texld(p, get_live_regs(p, inst),
599 tmp, A0_DEST_CHANNEL_ALL,
600 0, /* use a dummy dest reg */
601 negate(swizzle(tmp, ONE, ONE, ONE, ONE),
606 i915_program_error(p, "Unsupported KIL_NV condition code: %d",
607 inst->DstReg.CondMask);
612 src0 = src_vector(p, &inst->SrcReg[0], program);
616 get_result_vector(p, inst),
617 get_result_flags(inst), 0,
618 swizzle(src0, X, X, X, X), 0, 0);
622 src0 = src_vector(p, &inst->SrcReg[0], program);
623 tmp = i915_get_utemp(p);
625 /* tmp = max( a.xyzw, a.00zw )
626 * XXX: Clamp tmp.w to -128..128
628 * tmp.y = tmp.w * tmp.y
630 * result = cmp (a.11-x1, a.1x01, a.1xy1 )
632 i915_emit_arith(p, A0_MAX, tmp, A0_DEST_CHANNEL_ALL, 0,
633 src0, swizzle(src0, ZERO, ZERO, Z, W), 0);
635 i915_emit_arith(p, A0_LOG, tmp, A0_DEST_CHANNEL_Y, 0,
636 swizzle(tmp, Y, Y, Y, Y), 0, 0);
638 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_Y, 0,
639 swizzle(tmp, ZERO, Y, ZERO, ZERO),
640 swizzle(tmp, ZERO, W, ZERO, ZERO), 0);
642 i915_emit_arith(p, A0_EXP, tmp, A0_DEST_CHANNEL_Y, 0,
643 swizzle(tmp, Y, Y, Y, Y), 0, 0);
645 i915_emit_arith(p, A0_CMP,
646 get_result_vector(p, inst),
647 get_result_flags(inst), 0,
648 negate(swizzle(tmp, ONE, ONE, X, ONE), 0, 0, 1, 0),
649 swizzle(tmp, ONE, X, ZERO, ONE),
650 swizzle(tmp, ONE, X, Y, ONE));
655 src0 = src_vector(p, &inst->SrcReg[0], program);
656 src1 = src_vector(p, &inst->SrcReg[1], program);
657 src2 = src_vector(p, &inst->SrcReg[2], program);
658 flags = get_result_flags(inst);
659 tmp = i915_get_utemp(p);
666 * result = (-c)*a + tmp
668 i915_emit_arith(p, A0_MAD, tmp,
669 flags & A0_DEST_CHANNEL_ALL, 0, src1, src0, src2);
671 i915_emit_arith(p, A0_MAD,
672 get_result_vector(p, inst),
673 flags, 0, negate(src2, 1, 1, 1, 1), src0, tmp);
677 EMIT_3ARG_ARITH(A0_MAD);
681 EMIT_2ARG_ARITH(A0_MAX);
685 src0 = src_vector(p, &inst->SrcReg[0], program);
686 src1 = src_vector(p, &inst->SrcReg[1], program);
687 tmp = i915_get_utemp(p);
688 flags = get_result_flags(inst);
692 tmp, flags & A0_DEST_CHANNEL_ALL, 0,
693 negate(src0, 1, 1, 1, 1),
694 negate(src1, 1, 1, 1, 1), 0);
698 get_result_vector(p, inst),
699 flags, 0, negate(tmp, 1, 1, 1, 1), 0, 0);
703 EMIT_1ARG_ARITH(A0_MOV);
707 EMIT_2ARG_ARITH(A0_MUL);
711 src0 = src_vector(p, &inst->SrcReg[0], program);
712 src1 = src_vector(p, &inst->SrcReg[1], program);
713 tmp = i915_get_utemp(p);
714 flags = get_result_flags(inst);
716 /* XXX: masking on intermediate values, here and elsewhere.
720 tmp, A0_DEST_CHANNEL_X, 0,
721 swizzle(src0, X, X, X, X), 0, 0);
723 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_X, 0, tmp, src1, 0);
728 get_result_vector(p, inst),
729 flags, 0, swizzle(tmp, X, X, X, X), 0, 0);
734 src0 = src_vector(p, &inst->SrcReg[0], program);
738 get_result_vector(p, inst),
739 get_result_flags(inst), 0,
740 swizzle(src0, X, X, X, X), 0, 0);
745 src0 = src_vector(p, &inst->SrcReg[0], program);
749 get_result_vector(p, inst),
750 get_result_flags(inst), 0,
751 swizzle(src0, X, X, X, X), 0, 0);
755 src0 = src_vector(p, &inst->SrcReg[0], program);
756 tmp = i915_get_utemp(p);
759 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
760 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
761 * t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x
762 * scs.x = DP4 t1, sin_constants
763 * t1 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1
764 * scs.y = DP4 t1, cos_constants
768 tmp, A0_DEST_CHANNEL_XY, 0,
769 swizzle(src0, X, X, ONE, ONE),
770 swizzle(src0, X, ONE, ONE, ONE), 0);
774 tmp, A0_DEST_CHANNEL_ALL, 0,
775 swizzle(tmp, X, Y, X, Y),
776 swizzle(tmp, X, X, ONE, ONE), 0);
778 if (inst->DstReg.WriteMask & WRITEMASK_Y) {
781 if (inst->DstReg.WriteMask & WRITEMASK_X)
782 tmp1 = i915_get_utemp(p);
788 tmp1, A0_DEST_CHANNEL_ALL, 0,
789 swizzle(tmp, X, Y, Y, W),
790 swizzle(tmp, X, Z, ONE, ONE), 0);
794 get_result_vector(p, inst),
795 A0_DEST_CHANNEL_Y, 0,
796 swizzle(tmp1, W, Z, Y, X),
797 i915_emit_const4fv(p, sin_constants), 0);
800 if (inst->DstReg.WriteMask & WRITEMASK_X) {
803 tmp, A0_DEST_CHANNEL_XYZ, 0,
804 swizzle(tmp, X, X, Z, ONE),
805 swizzle(tmp, Z, ONE, ONE, ONE), 0);
809 get_result_vector(p, inst),
810 A0_DEST_CHANNEL_X, 0,
811 swizzle(tmp, ONE, Z, Y, X),
812 i915_emit_const4fv(p, cos_constants), 0);
817 tmp = i915_get_utemp(p);
818 flags = get_result_flags(inst);
819 dst = get_result_vector(p, inst);
821 /* tmp = src1 >= src2 */
826 src_vector(p, &inst->SrcReg[0], program),
827 src_vector(p, &inst->SrcReg[1], program),
829 /* dst = src1 <= src2 */
834 negate(src_vector(p, &inst->SrcReg[0], program),
836 negate(src_vector(p, &inst->SrcReg[1], program),
839 /* dst = tmp && dst */
850 src0 = src_vector(p, &inst->SrcReg[0], program);
851 tmp = i915_get_utemp(p);
852 consts0 = i915_emit_const4fv(p, sin_quad_constants[0]);
853 consts1 = i915_emit_const4fv(p, sin_quad_constants[1]);
855 /* Reduce range from repeating about [-pi,pi] to [-1,1] */
858 tmp, A0_DEST_CHANNEL_X, 0,
860 swizzle(consts1, Z, ZERO, ZERO, ZERO), /* 1/(2pi) */
861 swizzle(consts0, Z, ZERO, ZERO, ZERO)); /* .5 */
863 i915_emit_arith(p, A0_FRC, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
867 tmp, A0_DEST_CHANNEL_X, 0,
869 swizzle(consts0, X, ZERO, ZERO, ZERO), /* 2 */
870 swizzle(consts0, Y, ZERO, ZERO, ZERO)); /* -1 */
872 /* Compute sin using a quadratic and quartic. It gives continuity
873 * that repeating the Taylor series lacks every 2*pi, and has
876 * The idea was described at:
877 * http://www.devmaster.net/forums/showthread.php?t=5784
880 /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
883 tmp, A0_DEST_CHANNEL_Y, 0,
884 swizzle(tmp, ZERO, X, ZERO, ZERO),
885 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
888 /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
891 tmp, A0_DEST_CHANNEL_Y, 0,
892 swizzle(tmp, ZERO, X, ZERO, ZERO),
896 /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
899 tmp, A0_DEST_CHANNEL_X, 0,
901 swizzle(consts1, X, Y, ZERO, ZERO),
904 /* tmp.x now contains a first approximation (y). Now, weight it
905 * against tmp.y**2 to get closer.
909 tmp, A0_DEST_CHANNEL_Y, 0,
910 swizzle(tmp, ZERO, X, ZERO, ZERO),
911 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
914 /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
917 tmp, A0_DEST_CHANNEL_Y, 0,
918 swizzle(tmp, ZERO, X, ZERO, ZERO),
919 swizzle(tmp, ZERO, Y, ZERO, ZERO),
920 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0));
922 /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
925 get_result_vector(p, inst),
926 get_result_flags(inst), 0,
927 swizzle(consts1, W, W, W, W),
928 swizzle(tmp, Y, Y, Y, Y),
929 swizzle(tmp, X, X, X, X));
934 EMIT_2ARG_ARITH(A0_SGE);
940 get_result_vector( p, inst ),
941 get_result_flags( inst ), 0,
942 negate(src_vector( p, &inst->SrcReg[0], program),
944 negate(src_vector( p, &inst->SrcReg[1], program),
952 get_result_vector( p, inst ),
953 get_result_flags( inst ), 0,
954 negate(src_vector( p, &inst->SrcReg[0], program),
956 negate(src_vector( p, &inst->SrcReg[1], program),
962 EMIT_2ARG_ARITH(A0_SLT);
966 tmp = i915_get_utemp(p);
967 flags = get_result_flags(inst);
968 dst = get_result_vector(p, inst);
970 /* tmp = src1 < src2 */
975 src_vector(p, &inst->SrcReg[0], program),
976 src_vector(p, &inst->SrcReg[1], program),
978 /* dst = src1 > src2 */
983 negate(src_vector(p, &inst->SrcReg[0], program),
985 negate(src_vector(p, &inst->SrcReg[1], program),
988 /* dst = tmp || dst */
992 flags | A0_DEST_SATURATE, 0,
999 dst = get_result_vector(p, inst);
1000 flags = get_result_flags(inst);
1001 src0 = src_vector(p, &inst->SrcReg[0], program);
1002 tmp = i915_get_utemp(p);
1004 /* tmp = (src < 0.0) */
1010 swizzle(src0, ZERO, ZERO, ZERO, ZERO),
1013 /* dst = (0.0 < src) */
1018 swizzle(src0, ZERO, ZERO, ZERO, ZERO),
1022 /* dst = (src > 0.0) - (src < 0.0) */
1028 negate(tmp, 1, 1, 1, 1),
1034 src0 = src_vector(p, &inst->SrcReg[0], program);
1035 src1 = src_vector(p, &inst->SrcReg[1], program);
1039 get_result_vector(p, inst),
1040 get_result_flags(inst), 0,
1041 src0, negate(src1, 1, 1, 1, 1), 0);
1045 EMIT_1ARG_ARITH(A0_MOV); /* extended swizzle handled natively */
1053 EMIT_TEX(T0_TEXLDB);
1057 EMIT_TEX(T0_TEXLDP);
1062 * result.x = src0.y * src1.z - src0.z * src1.y;
1063 * result.y = src0.z * src1.x - src0.x * src1.z;
1064 * result.z = src0.x * src1.y - src0.y * src1.x;
1067 src0 = src_vector(p, &inst->SrcReg[0], program);
1068 src1 = src_vector(p, &inst->SrcReg[1], program);
1069 tmp = i915_get_utemp(p);
1073 tmp, A0_DEST_CHANNEL_ALL, 0,
1074 swizzle(src0, Z, X, Y, ONE),
1075 swizzle(src1, Y, Z, X, ONE), 0);
1079 get_result_vector(p, inst),
1080 get_result_flags(inst), 0,
1081 swizzle(src0, Y, Z, X, ONE),
1082 swizzle(src1, Z, X, Y, ONE),
1083 negate(tmp, 1, 1, 1, 0));
1089 case OPCODE_BGNLOOP:
1099 case OPCODE_ENDLOOP:
1104 i915_program_error(p, "Unsupported opcode: %s",
1105 _mesa_opcode_string(inst->Opcode));
1110 /* These opcodes are claimed as GLSL, NV_vp, and ARB_vp in
1111 * prog_instruction.h, but apparently GLSL doesn't ever emit them.
1112 * Instead, it translates to EX2 or LG2.
1116 /* These opcodes are claimed by GLSL in prog_instruction.h, but
1117 * only NV_vp/fp appears to emit them.
1120 i915_program_error(p, "bad opcode: %s",
1121 _mesa_opcode_string(inst->Opcode));
1126 i915_release_utemps(p);
1130 /* Rather than trying to intercept and jiggle depth writes during
1131 * emit, just move the value into its correct position at the end of
1135 fixup_depth_write(struct i915_fragment_program *p)
1137 if (p->depth_written) {
1138 GLuint depth = UREG(REG_TYPE_OD, 0);
1142 depth, A0_DEST_CHANNEL_W, 0,
1143 swizzle(depth, X, Y, Z, Z), 0, 0);
1149 check_wpos(struct i915_fragment_program *p)
1151 GLuint inputs = p->FragProg.Base.InputsRead;
1156 for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
1157 if (inputs & (FRAG_BIT_TEX(i) | FRAG_BIT_VAR(i)))
1159 else if (inputs & FRAG_BIT_WPOS) {
1161 inputs &= ~FRAG_BIT_WPOS;
1165 if (inputs & FRAG_BIT_WPOS) {
1166 i915_program_error(p, "No free texcoord for wpos value");
1172 translate_program(struct i915_fragment_program *p)
1174 struct i915_context *i915 = I915_CONTEXT(p->ctx);
1176 if (INTEL_DEBUG & DEBUG_WM) {
1178 _mesa_print_program(&p->FragProg.Base);
1182 i915_init_program(i915, p);
1185 fixup_depth_write(p);
1186 i915_fini_program(p);
1193 track_params(struct i915_fragment_program *p)
1198 _mesa_load_state_parameters(p->ctx, p->FragProg.Base.Parameters);
1200 for (i = 0; i < p->nr_params; i++) {
1201 GLint reg = p->param[i].reg;
1202 COPY_4V(p->constant[reg], p->param[i].values);
1205 p->params_uptodate = 1;
1206 p->on_hardware = 0; /* overkill */
1211 i915BindProgram(struct gl_context * ctx, GLenum target, struct gl_program *prog)
1213 if (target == GL_FRAGMENT_PROGRAM_ARB) {
1214 struct i915_context *i915 = I915_CONTEXT(ctx);
1215 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1217 if (i915->current_program == p)
1220 if (i915->current_program) {
1221 i915->current_program->on_hardware = 0;
1222 i915->current_program->params_uptodate = 0;
1225 i915->current_program = p;
1227 assert(p->on_hardware == 0);
1228 assert(p->params_uptodate == 0);
1233 static struct gl_program *
1234 i915NewProgram(struct gl_context * ctx, GLenum target, GLuint id)
1237 case GL_VERTEX_PROGRAM_ARB:
1238 return _mesa_init_vertex_program(ctx, CALLOC_STRUCT(gl_vertex_program),
1241 case GL_FRAGMENT_PROGRAM_ARB:{
1242 struct i915_fragment_program *prog =
1243 CALLOC_STRUCT(i915_fragment_program);
1245 i915_init_program(I915_CONTEXT(ctx), prog);
1247 return _mesa_init_fragment_program(ctx, &prog->FragProg,
1257 return _mesa_new_program(ctx, target, id);
1262 i915DeleteProgram(struct gl_context * ctx, struct gl_program *prog)
1264 if (prog->Target == GL_FRAGMENT_PROGRAM_ARB) {
1265 struct i915_context *i915 = I915_CONTEXT(ctx);
1266 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1268 if (i915->current_program == p)
1269 i915->current_program = 0;
1272 _mesa_delete_program(ctx, prog);
1277 i915IsProgramNative(struct gl_context * ctx, GLenum target, struct gl_program *prog)
1279 if (target == GL_FRAGMENT_PROGRAM_ARB) {
1280 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1283 translate_program(p);
1292 i915ProgramStringNotify(struct gl_context * ctx,
1293 GLenum target, struct gl_program *prog)
1295 if (target == GL_FRAGMENT_PROGRAM_ARB) {
1296 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1300 (void) _tnl_program_string(ctx, target, prog);
1302 /* XXX check if program is legal, within limits */
1307 i915_update_program(struct gl_context *ctx)
1309 struct intel_context *intel = intel_context(ctx);
1310 struct i915_context *i915 = i915_context(&intel->ctx);
1311 struct i915_fragment_program *fp =
1312 (struct i915_fragment_program *) ctx->FragmentProgram._Current;
1314 if (i915->current_program != fp) {
1315 if (i915->current_program) {
1316 i915->current_program->on_hardware = 0;
1317 i915->current_program->params_uptodate = 0;
1320 i915->current_program = fp;
1323 if (!fp->translated)
1324 translate_program(fp);
1326 FALLBACK(&i915->intel, I915_FALLBACK_PROGRAM, fp->error);
1330 i915ValidateFragmentProgram(struct i915_context *i915)
1332 struct gl_context *ctx = &i915->intel.ctx;
1333 struct intel_context *intel = intel_context(ctx);
1334 TNLcontext *tnl = TNL_CONTEXT(ctx);
1335 struct vertex_buffer *VB = &tnl->vb;
1337 struct i915_fragment_program *p =
1338 (struct i915_fragment_program *) ctx->FragmentProgram._Current;
1340 const GLuint inputsRead = p->FragProg.Base.InputsRead;
1341 GLuint s4 = i915->state.Ctx[I915_CTXREG_LIS4] & ~S4_VFMT_MASK;
1342 GLuint s2 = S2_TEXCOORD_NONE;
1347 VB->AttribPtr[VERT_ATTRIB_POS] = VB->NdcPtr;
1350 translate_program(p);
1352 intel->vertex_attr_count = 0;
1353 intel->wpos_offset = 0;
1354 intel->coloroffset = 0;
1355 intel->specoffset = 0;
1357 if (inputsRead & FRAG_BITS_TEX_ANY || p->wpos_tex != -1) {
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)) {
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) {
1396 int wpos_size = 4 * sizeof(float);
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(wpos_size));
1403 intel->wpos_offset = offset;
1404 EMIT_PAD(wpos_size);
1408 if (s2 != i915->state.Ctx[I915_CTXREG_LIS2] ||
1409 s4 != i915->state.Ctx[I915_CTXREG_LIS4]) {
1412 I915_STATECHANGE(i915, I915_UPLOAD_CTX);
1414 /* Must do this *after* statechange, so as not to affect
1415 * buffered vertices reliant on the old state:
1417 intel->vertex_size = _tnl_install_attrs(&intel->ctx,
1418 intel->vertex_attrs,
1419 intel->vertex_attr_count,
1420 intel->ViewportMatrix.m, 0);
1422 assert(intel->prim.current_offset == intel->prim.start_offset);
1423 intel->prim.start_offset = (intel->prim.current_offset + intel->vertex_size-1) / intel->vertex_size * intel->vertex_size;
1424 intel->prim.current_offset = intel->prim.start_offset;
1426 intel->vertex_size >>= 2;
1428 i915->state.Ctx[I915_CTXREG_LIS2] = s2;
1429 i915->state.Ctx[I915_CTXREG_LIS4] = s4;
1431 k = intel->vtbl.check_vertex_size(intel, intel->vertex_size);
1435 if (!p->params_uptodate)
1438 if (!p->on_hardware)
1439 i915_upload_program(i915, p);
1441 if (INTEL_DEBUG & DEBUG_WM) {
1443 i915_disassemble_program(i915->state.Program, i915->state.ProgramSize);
1448 i915InitFragProgFuncs(struct dd_function_table *functions)
1450 functions->BindProgram = i915BindProgram;
1451 functions->NewProgram = i915NewProgram;
1452 functions->DeleteProgram = i915DeleteProgram;
1453 functions->IsProgramNative = i915IsProgramNative;
1454 functions->ProgramStringNotify = i915ProgramStringNotify;