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_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->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 case TEXTURE_EXTERNAL_INDEX:
266 return D0_SAMPLE_TYPE_2D;
267 case TEXTURE_RECT_INDEX:
268 return D0_SAMPLE_TYPE_2D;
269 case TEXTURE_3D_INDEX:
270 return D0_SAMPLE_TYPE_VOLUME;
271 case TEXTURE_CUBE_INDEX:
272 return D0_SAMPLE_TYPE_CUBE;
274 i915_program_error(p, "TexSrcBit: %d", bit);
279 #define EMIT_TEX( OP ) \
281 GLuint dim = translate_tex_src_target( p, inst->TexSrcTarget ); \
282 const struct gl_program *program = &p->FragProg; \
283 GLuint unit = program->SamplerUnits[inst->TexSrcUnit]; \
284 GLuint sampler = i915_emit_decl(p, REG_TYPE_S, \
286 GLuint coord = src_vector( p, &inst->SrcReg[0], program); \
289 i915_emit_texld( p, get_live_regs(p, inst), \
290 get_result_vector( p, inst ), \
291 get_result_flags( inst ), \
297 #define EMIT_ARITH( OP, N ) \
299 i915_emit_arith( p, \
301 get_result_vector( p, inst ), \
302 get_result_flags( inst ), 0, \
303 (N<1)?0:src_vector( p, &inst->SrcReg[0], program), \
304 (N<2)?0:src_vector( p, &inst->SrcReg[1], program), \
305 (N<3)?0:src_vector( p, &inst->SrcReg[2], program)); \
308 #define EMIT_1ARG_ARITH( OP ) EMIT_ARITH( OP, 1 )
309 #define EMIT_2ARG_ARITH( OP ) EMIT_ARITH( OP, 2 )
310 #define EMIT_3ARG_ARITH( OP ) EMIT_ARITH( OP, 3 )
313 * TODO: consider moving this into core
315 static bool calc_live_regs( struct i915_fragment_program *p )
317 const struct gl_program *program = &p->FragProg;
318 GLuint regsUsed = ~((1 << I915_MAX_TEMPORARY) - 1);
319 uint8_t live_components[I915_MAX_TEMPORARY] = { 0, };
322 for (i = program->arb.NumInstructions - 1; i >= 0; i--) {
323 struct prog_instruction *inst = &program->arb.Instructions[i];
324 int opArgs = _mesa_num_inst_src_regs(inst->Opcode);
327 /* Register is written to: unmark as live for this and preceeding ops */
328 if (inst->DstReg.File == PROGRAM_TEMPORARY) {
329 if (inst->DstReg.Index >= I915_MAX_TEMPORARY)
332 live_components[inst->DstReg.Index] &= ~inst->DstReg.WriteMask;
333 if (live_components[inst->DstReg.Index] == 0)
334 regsUsed &= ~(1 << inst->DstReg.Index);
337 for (a = 0; a < opArgs; a++) {
338 /* Register is read from: mark as live for this and preceeding ops */
339 if (inst->SrcReg[a].File == PROGRAM_TEMPORARY) {
342 if (inst->SrcReg[a].Index >= I915_MAX_TEMPORARY)
345 regsUsed |= 1 << inst->SrcReg[a].Index;
347 for (c = 0; c < 4; c++) {
348 const unsigned field = GET_SWZ(inst->SrcReg[a].Swizzle, c);
350 if (field <= SWIZZLE_W)
351 live_components[inst->SrcReg[a].Index] |= (1U << field);
356 p->usedRegs[i] = regsUsed;
362 static GLuint get_live_regs( struct i915_fragment_program *p,
363 const struct prog_instruction *inst )
365 const struct gl_program *program = &p->FragProg;
366 GLuint nr = inst - program->arb.Instructions;
368 return p->usedRegs[nr];
372 /* Possible concerns:
374 * SIN, COS -- could use another taylor step?
375 * LIT -- results seem a little different to sw mesa
376 * LOG -- different to mesa on negative numbers, but this is conformant.
378 * Parse failures -- Mesa doesn't currently give a good indication
379 * internally whether a particular program string parsed or not. This
380 * can lead to confusion -- hopefully we cope with it ok now.
384 upload_program(struct i915_fragment_program *p)
386 const struct gl_program *program = &p->FragProg;
387 const struct prog_instruction *inst = program->arb.Instructions;
389 if (INTEL_DEBUG & DEBUG_WM)
390 _mesa_print_program(program);
392 /* Is this a parse-failed program? Ensure a valid program is
393 * loaded, as the flagging of an error isn't sufficient to stop
394 * this being uploaded to hardware.
396 if (inst[0].Opcode == OPCODE_END) {
397 GLuint tmp = i915_get_utemp(p);
400 UREG(REG_TYPE_OC, 0),
401 A0_DEST_CHANNEL_ALL, 0,
402 swizzle(tmp, ONE, ZERO, ONE, ONE), 0, 0);
406 if (program->arb.NumInstructions > I915_MAX_INSN) {
407 i915_program_error(p, "Exceeded max instructions (%d out of %d)",
408 program->arb.NumInstructions, I915_MAX_INSN);
412 /* Not always needed:
414 if (!calc_live_regs(p)) {
415 i915_program_error(p, "Could not allocate registers");
420 GLuint src0, src1, src2, flags;
421 GLuint tmp = 0, dst, consts0 = 0, consts1 = 0;
423 switch (inst->Opcode) {
425 src0 = src_vector(p, &inst->SrcReg[0], program);
428 get_result_vector(p, inst),
429 get_result_flags(inst), 0,
430 src0, negate(src0, 1, 1, 1, 1), 0);
434 EMIT_2ARG_ARITH(A0_ADD);
438 src0 = src_vector(p, &inst->SrcReg[0], program);
439 src1 = src_vector(p, &inst->SrcReg[1], program);
440 src2 = src_vector(p, &inst->SrcReg[2], program);
441 i915_emit_arith(p, A0_CMP, get_result_vector(p, inst), get_result_flags(inst), 0, src0, src2, src1); /* NOTE: order of src2, src1 */
445 src0 = src_vector(p, &inst->SrcReg[0], program);
446 tmp = i915_get_utemp(p);
447 consts0 = i915_emit_const4fv(p, sin_quad_constants[0]);
448 consts1 = i915_emit_const4fv(p, sin_quad_constants[1]);
450 /* Reduce range from repeating about [-pi,pi] to [-1,1] */
453 tmp, A0_DEST_CHANNEL_X, 0,
455 swizzle(consts1, Z, ZERO, ZERO, ZERO), /* 1/(2pi) */
456 swizzle(consts0, W, ZERO, ZERO, ZERO)); /* .75 */
458 i915_emit_arith(p, A0_FRC, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
462 tmp, A0_DEST_CHANNEL_X, 0,
464 swizzle(consts0, X, ZERO, ZERO, ZERO), /* 2 */
465 swizzle(consts0, Y, ZERO, ZERO, ZERO)); /* -1 */
467 /* Compute COS with the same calculation used for SIN, but a
468 * different source range has been mapped to [-1,1] this time.
471 /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
474 tmp, A0_DEST_CHANNEL_Y, 0,
475 swizzle(tmp, ZERO, X, ZERO, ZERO),
476 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
479 /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
482 tmp, A0_DEST_CHANNEL_Y, 0,
483 swizzle(tmp, ZERO, X, ZERO, ZERO),
487 /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
490 tmp, A0_DEST_CHANNEL_X, 0,
492 swizzle(consts1, X, Y, ZERO, ZERO),
495 /* tmp.x now contains a first approximation (y). Now, weight it
496 * against tmp.y**2 to get closer.
500 tmp, A0_DEST_CHANNEL_Y, 0,
501 swizzle(tmp, ZERO, X, ZERO, ZERO),
502 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
505 /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
508 tmp, A0_DEST_CHANNEL_Y, 0,
509 swizzle(tmp, ZERO, X, ZERO, ZERO),
510 swizzle(tmp, ZERO, Y, ZERO, ZERO),
511 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0));
513 /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
516 get_result_vector(p, inst),
517 get_result_flags(inst), 0,
518 swizzle(consts1, W, W, W, W),
519 swizzle(tmp, Y, Y, Y, Y),
520 swizzle(tmp, X, X, X, X));
524 src0 = src_vector(p, &inst->SrcReg[0], program);
525 src1 = src_vector(p, &inst->SrcReg[1], program);
528 get_result_vector(p, inst),
529 get_result_flags(inst), 0,
530 swizzle(src0, X, Y, ZERO, ZERO),
531 swizzle(src1, X, Y, ZERO, ZERO),
536 EMIT_2ARG_ARITH(A0_DP3);
540 EMIT_2ARG_ARITH(A0_DP4);
544 src0 = src_vector(p, &inst->SrcReg[0], program);
545 src1 = src_vector(p, &inst->SrcReg[1], program);
549 get_result_vector(p, inst),
550 get_result_flags(inst), 0,
551 swizzle(src0, X, Y, Z, ONE), src1, 0);
555 src0 = src_vector(p, &inst->SrcReg[0], program);
556 src1 = src_vector(p, &inst->SrcReg[1], program);
558 /* result[0] = 1 * 1;
559 * result[1] = a[1] * b[1];
560 * result[2] = a[2] * 1;
561 * result[3] = 1 * b[3];
565 get_result_vector(p, inst),
566 get_result_flags(inst), 0,
567 swizzle(src0, ONE, Y, Z, ONE),
568 swizzle(src1, ONE, Y, ONE, W), 0);
572 src0 = src_vector(p, &inst->SrcReg[0], program);
576 get_result_vector(p, inst),
577 get_result_flags(inst), 0,
578 swizzle(src0, X, X, X, X), 0, 0);
582 EMIT_1ARG_ARITH(A0_FLR);
586 EMIT_1ARG_ARITH(A0_TRC);
590 EMIT_1ARG_ARITH(A0_FRC);
594 src0 = src_vector(p, &inst->SrcReg[0], program);
595 tmp = i915_get_utemp(p);
597 i915_emit_texld(p, get_live_regs(p, inst),
598 tmp, A0_DEST_CHANNEL_ALL, /* use a dummy dest reg */
599 0, src0, T0_TEXKILL);
603 src0 = src_vector(p, &inst->SrcReg[0], program);
607 get_result_vector(p, inst),
608 get_result_flags(inst), 0,
609 swizzle(src0, X, X, X, X), 0, 0);
613 src0 = src_vector(p, &inst->SrcReg[0], program);
614 tmp = i915_get_utemp(p);
616 /* tmp = max( a.xyzw, a.00zw )
617 * XXX: Clamp tmp.w to -128..128
619 * tmp.y = tmp.w * tmp.y
621 * result = cmp (a.11-x1, a.1x01, a.1xy1 )
623 i915_emit_arith(p, A0_MAX, tmp, A0_DEST_CHANNEL_ALL, 0,
624 src0, swizzle(src0, ZERO, ZERO, Z, W), 0);
626 i915_emit_arith(p, A0_LOG, tmp, A0_DEST_CHANNEL_Y, 0,
627 swizzle(tmp, Y, Y, Y, Y), 0, 0);
629 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_Y, 0,
630 swizzle(tmp, ZERO, Y, ZERO, ZERO),
631 swizzle(tmp, ZERO, W, ZERO, ZERO), 0);
633 i915_emit_arith(p, A0_EXP, tmp, A0_DEST_CHANNEL_Y, 0,
634 swizzle(tmp, Y, Y, Y, Y), 0, 0);
636 i915_emit_arith(p, A0_CMP,
637 get_result_vector(p, inst),
638 get_result_flags(inst), 0,
639 negate(swizzle(tmp, ONE, ONE, X, ONE), 0, 0, 1, 0),
640 swizzle(tmp, ONE, X, ZERO, ONE),
641 swizzle(tmp, ONE, X, Y, ONE));
646 src0 = src_vector(p, &inst->SrcReg[0], program);
647 src1 = src_vector(p, &inst->SrcReg[1], program);
648 src2 = src_vector(p, &inst->SrcReg[2], program);
649 flags = get_result_flags(inst);
650 tmp = i915_get_utemp(p);
657 * result = (-c)*a + tmp
659 i915_emit_arith(p, A0_MAD, tmp,
660 flags & A0_DEST_CHANNEL_ALL, 0, src1, src0, src2);
662 i915_emit_arith(p, A0_MAD,
663 get_result_vector(p, inst),
664 flags, 0, negate(src2, 1, 1, 1, 1), src0, tmp);
668 EMIT_3ARG_ARITH(A0_MAD);
672 EMIT_2ARG_ARITH(A0_MAX);
676 EMIT_2ARG_ARITH(A0_MIN);
680 EMIT_1ARG_ARITH(A0_MOV);
684 EMIT_2ARG_ARITH(A0_MUL);
688 src0 = src_vector(p, &inst->SrcReg[0], program);
689 src1 = src_vector(p, &inst->SrcReg[1], program);
690 tmp = i915_get_utemp(p);
691 flags = get_result_flags(inst);
693 /* XXX: masking on intermediate values, here and elsewhere.
697 tmp, A0_DEST_CHANNEL_X, 0,
698 swizzle(src0, X, X, X, X), 0, 0);
700 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_X, 0, tmp, src1, 0);
705 get_result_vector(p, inst),
706 flags, 0, swizzle(tmp, X, X, X, X), 0, 0);
711 src0 = src_vector(p, &inst->SrcReg[0], program);
715 get_result_vector(p, inst),
716 get_result_flags(inst), 0,
717 swizzle(src0, 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);
732 src0 = src_vector(p, &inst->SrcReg[0], program);
733 tmp = i915_get_utemp(p);
736 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
737 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
738 * t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x
739 * scs.x = DP4 t1, sin_constants
740 * t1 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1
741 * scs.y = DP4 t1, cos_constants
745 tmp, A0_DEST_CHANNEL_XY, 0,
746 swizzle(src0, X, X, ONE, ONE),
747 swizzle(src0, X, ONE, ONE, ONE), 0);
751 tmp, A0_DEST_CHANNEL_ALL, 0,
752 swizzle(tmp, X, Y, X, Y),
753 swizzle(tmp, X, X, ONE, ONE), 0);
755 if (inst->DstReg.WriteMask & WRITEMASK_Y) {
758 if (inst->DstReg.WriteMask & WRITEMASK_X)
759 tmp1 = i915_get_utemp(p);
765 tmp1, A0_DEST_CHANNEL_ALL, 0,
766 swizzle(tmp, X, Y, Y, W),
767 swizzle(tmp, X, Z, ONE, ONE), 0);
771 get_result_vector(p, inst),
772 A0_DEST_CHANNEL_Y, 0,
773 swizzle(tmp1, W, Z, Y, X),
774 i915_emit_const4fv(p, sin_constants), 0);
777 if (inst->DstReg.WriteMask & WRITEMASK_X) {
780 tmp, A0_DEST_CHANNEL_XYZ, 0,
781 swizzle(tmp, X, X, Z, ONE),
782 swizzle(tmp, Z, ONE, ONE, ONE), 0);
786 get_result_vector(p, inst),
787 A0_DEST_CHANNEL_X, 0,
788 swizzle(tmp, ONE, Z, Y, X),
789 i915_emit_const4fv(p, cos_constants), 0);
794 src0 = src_vector(p, &inst->SrcReg[0], program);
795 tmp = i915_get_utemp(p);
796 consts0 = i915_emit_const4fv(p, sin_quad_constants[0]);
797 consts1 = i915_emit_const4fv(p, sin_quad_constants[1]);
799 /* Reduce range from repeating about [-pi,pi] to [-1,1] */
802 tmp, A0_DEST_CHANNEL_X, 0,
804 swizzle(consts1, Z, ZERO, ZERO, ZERO), /* 1/(2pi) */
805 swizzle(consts0, Z, ZERO, ZERO, ZERO)); /* .5 */
807 i915_emit_arith(p, A0_FRC, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
811 tmp, A0_DEST_CHANNEL_X, 0,
813 swizzle(consts0, X, ZERO, ZERO, ZERO), /* 2 */
814 swizzle(consts0, Y, ZERO, ZERO, ZERO)); /* -1 */
816 /* Compute sin using a quadratic and quartic. It gives continuity
817 * that repeating the Taylor series lacks every 2*pi, and has
820 * The idea was described at:
821 * http://www.devmaster.net/forums/showthread.php?t=5784
824 /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
827 tmp, A0_DEST_CHANNEL_Y, 0,
828 swizzle(tmp, ZERO, X, ZERO, ZERO),
829 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
832 /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
835 tmp, A0_DEST_CHANNEL_Y, 0,
836 swizzle(tmp, ZERO, X, ZERO, ZERO),
840 /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
843 tmp, A0_DEST_CHANNEL_X, 0,
845 swizzle(consts1, X, Y, ZERO, ZERO),
848 /* tmp.x now contains a first approximation (y). Now, weight it
849 * against tmp.y**2 to get closer.
853 tmp, A0_DEST_CHANNEL_Y, 0,
854 swizzle(tmp, ZERO, X, ZERO, ZERO),
855 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
858 /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
861 tmp, A0_DEST_CHANNEL_Y, 0,
862 swizzle(tmp, ZERO, X, ZERO, ZERO),
863 swizzle(tmp, ZERO, Y, ZERO, ZERO),
864 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0));
866 /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
869 get_result_vector(p, inst),
870 get_result_flags(inst), 0,
871 swizzle(consts1, W, W, W, W),
872 swizzle(tmp, Y, Y, Y, Y),
873 swizzle(tmp, X, X, X, X));
878 EMIT_2ARG_ARITH(A0_SGE);
882 EMIT_2ARG_ARITH(A0_SLT);
886 dst = get_result_vector(p, inst);
887 flags = get_result_flags(inst);
888 src0 = src_vector(p, &inst->SrcReg[0], program);
889 tmp = i915_get_utemp(p);
891 /* tmp = (src < 0.0) */
897 swizzle(src0, ZERO, ZERO, ZERO, ZERO),
900 /* dst = (0.0 < src) */
905 swizzle(src0, ZERO, ZERO, ZERO, ZERO),
909 /* dst = (src > 0.0) - (src < 0.0) */
915 negate(tmp, 1, 1, 1, 1),
921 src0 = src_vector(p, &inst->SrcReg[0], program);
922 src1 = src_vector(p, &inst->SrcReg[1], program);
926 get_result_vector(p, inst),
927 get_result_flags(inst), 0,
928 src0, negate(src1, 1, 1, 1, 1), 0);
932 EMIT_1ARG_ARITH(A0_MOV); /* extended swizzle handled natively */
949 * result.x = src0.y * src1.z - src0.z * src1.y;
950 * result.y = src0.z * src1.x - src0.x * src1.z;
951 * result.z = src0.x * src1.y - src0.y * src1.x;
954 src0 = src_vector(p, &inst->SrcReg[0], program);
955 src1 = src_vector(p, &inst->SrcReg[1], program);
956 tmp = i915_get_utemp(p);
960 tmp, A0_DEST_CHANNEL_ALL, 0,
961 swizzle(src0, Z, X, Y, ONE),
962 swizzle(src1, Y, Z, X, ONE), 0);
966 get_result_vector(p, inst),
967 get_result_flags(inst), 0,
968 swizzle(src0, Y, Z, X, ONE),
969 swizzle(src1, Z, X, Y, ONE),
970 negate(tmp, 1, 1, 1, 0));
990 i915_program_error(p, "Unsupported opcode: %s",
991 _mesa_opcode_string(inst->Opcode));
996 /* These opcodes are claimed as GLSL, NV_vp, and ARB_vp in
997 * prog_instruction.h, but apparently GLSL doesn't ever emit them.
998 * Instead, it translates to EX2 or LG2.
1002 /* These opcodes are claimed by GLSL in prog_instruction.h, but
1003 * only NV_vp/fp appears to emit them.
1006 i915_program_error(p, "bad opcode: %s",
1007 _mesa_opcode_string(inst->Opcode));
1012 i915_release_utemps(p);
1016 /* Rather than trying to intercept and jiggle depth writes during
1017 * emit, just move the value into its correct position at the end of
1021 fixup_depth_write(struct i915_fragment_program *p)
1023 if (p->depth_written) {
1024 GLuint depth = UREG(REG_TYPE_OD, 0);
1028 depth, A0_DEST_CHANNEL_W, 0,
1029 swizzle(depth, X, Y, Z, Z), 0, 0);
1034 check_texcoord_mapping(struct i915_fragment_program *p)
1036 GLbitfield64 inputs = p->FragProg.info.inputs_read;
1039 for (unsigned i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
1040 if (inputs & VARYING_BIT_TEX(i)) {
1041 if (unit >= p->ctx->Const.MaxTextureCoordUnits) {
1045 p->texcoord_mapping[unit++] = i | TEXCOORD_TEX;
1047 if (inputs & VARYING_BIT_VAR(i)) {
1048 if (unit >= p->ctx->Const.MaxTextureCoordUnits) {
1052 p->texcoord_mapping[unit++] = i | TEXCOORD_VAR;
1056 if (unit > p->ctx->Const.MaxTextureCoordUnits)
1057 i915_program_error(p, "Too many texcoord units");
1061 check_wpos(struct i915_fragment_program *p)
1063 GLbitfield64 inputs = p->FragProg.info.inputs_read;
1069 if ((inputs & VARYING_BIT_POS) == 0)
1072 for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
1073 unit += !!(inputs & VARYING_BIT_TEX(i));
1074 unit += !!(inputs & VARYING_BIT_VAR(i));
1077 if (unit < p->ctx->Const.MaxTextureCoordUnits)
1080 i915_program_error(p, "No free texcoord for wpos value");
1085 translate_program(struct i915_fragment_program *p)
1087 struct i915_context *i915 = I915_CONTEXT(p->ctx);
1089 if (INTEL_DEBUG & DEBUG_WM) {
1091 _mesa_print_program(&p->FragProg);
1095 i915_init_program(i915, p);
1096 check_texcoord_mapping(p);
1099 fixup_depth_write(p);
1100 i915_fini_program(p);
1107 track_params(struct i915_fragment_program *p)
1112 _mesa_load_state_parameters(p->ctx, p->FragProg.Parameters);
1114 for (i = 0; i < p->nr_params; i++) {
1115 GLint reg = p->param[i].reg;
1116 COPY_4V(p->constant[reg], p->param[i].values);
1119 p->params_uptodate = 1;
1120 p->on_hardware = 0; /* overkill */
1125 i915BindProgram(struct gl_context * ctx, GLenum target, struct gl_program *prog)
1127 if (target == GL_FRAGMENT_PROGRAM_ARB) {
1128 struct i915_context *i915 = I915_CONTEXT(ctx);
1129 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1131 if (i915->current_program == p)
1134 if (i915->current_program) {
1135 i915->current_program->on_hardware = 0;
1136 i915->current_program->params_uptodate = 0;
1139 i915->current_program = p;
1141 assert(p->on_hardware == 0);
1142 assert(p->params_uptodate == 0);
1147 static struct gl_program *
1148 i915NewProgram(struct gl_context * ctx, GLenum target, GLuint id,
1152 case GL_VERTEX_PROGRAM_ARB: {
1153 struct gl_program *prog = rzalloc(NULL, struct gl_program);
1154 return _mesa_init_gl_program(prog, target, id, is_arb_asm);
1157 case GL_FRAGMENT_PROGRAM_ARB:{
1158 struct i915_fragment_program *prog =
1159 rzalloc(NULL, struct i915_fragment_program);
1161 i915_init_program(I915_CONTEXT(ctx), prog);
1163 return _mesa_init_gl_program(&prog->FragProg, target, id,
1173 return _mesa_new_program(ctx, target, id, is_arb_asm);
1178 i915DeleteProgram(struct gl_context * ctx, struct gl_program *prog)
1180 if (prog->Target == GL_FRAGMENT_PROGRAM_ARB) {
1181 struct i915_context *i915 = I915_CONTEXT(ctx);
1182 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1184 if (i915->current_program == p)
1185 i915->current_program = 0;
1188 _mesa_delete_program(ctx, prog);
1193 i915IsProgramNative(struct gl_context * ctx, GLenum target, struct gl_program *prog)
1195 if (target == GL_FRAGMENT_PROGRAM_ARB) {
1196 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1199 translate_program(p);
1208 i915ProgramStringNotify(struct gl_context * ctx,
1209 GLenum target, struct gl_program *prog)
1211 if (target == GL_FRAGMENT_PROGRAM_ARB) {
1212 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1216 (void) _tnl_program_string(ctx, target, prog);
1218 /* XXX check if program is legal, within limits */
1223 i915SamplerUniformChange(struct gl_context *ctx,
1224 GLenum target, struct gl_program *prog)
1226 i915ProgramStringNotify(ctx, target, prog);
1230 i915_update_program(struct gl_context *ctx)
1232 struct intel_context *intel = intel_context(ctx);
1233 struct i915_context *i915 = i915_context(&intel->ctx);
1234 struct i915_fragment_program *fp =
1235 (struct i915_fragment_program *) ctx->FragmentProgram._Current;
1237 if (i915->current_program != fp) {
1238 if (i915->current_program) {
1239 i915->current_program->on_hardware = 0;
1240 i915->current_program->params_uptodate = 0;
1243 i915->current_program = fp;
1246 if (!fp->translated)
1247 translate_program(fp);
1249 FALLBACK(&i915->intel, I915_FALLBACK_PROGRAM, fp->error);
1253 i915ValidateFragmentProgram(struct i915_context *i915)
1255 struct gl_context *ctx = &i915->intel.ctx;
1256 struct intel_context *intel = intel_context(ctx);
1257 TNLcontext *tnl = TNL_CONTEXT(ctx);
1258 struct vertex_buffer *VB = &tnl->vb;
1260 struct i915_fragment_program *p =
1261 (struct i915_fragment_program *) ctx->FragmentProgram._Current;
1263 const GLbitfield64 inputsRead = p->FragProg.info.inputs_read;
1264 GLuint s4 = i915->state.Ctx[I915_CTXREG_LIS4] & ~S4_VFMT_MASK;
1265 GLuint s2 = S2_TEXCOORD_NONE;
1270 VB->AttribPtr[VERT_ATTRIB_POS] = VB->NdcPtr;
1273 translate_program(p);
1275 intel->vertex_attr_count = 0;
1276 intel->wpos_offset = 0;
1277 intel->coloroffset = 0;
1278 intel->specoffset = 0;
1280 if (inputsRead & VARYING_BITS_TEX_ANY || p->wpos_tex != -1) {
1281 EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_4F_VIEWPORT, S4_VFMT_XYZW, 16);
1284 EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_3F_VIEWPORT, S4_VFMT_XYZ, 12);
1287 /* Handle gl_PointSize builtin var here */
1288 if (ctx->Point._Attenuated || ctx->VertexProgram.PointSizeEnabled)
1289 EMIT_ATTR(_TNL_ATTRIB_POINTSIZE, EMIT_1F, S4_VFMT_POINT_WIDTH, 4);
1291 if (inputsRead & VARYING_BIT_COL0) {
1292 intel->coloroffset = offset / 4;
1293 EMIT_ATTR(_TNL_ATTRIB_COLOR0, EMIT_4UB_4F_BGRA, S4_VFMT_COLOR, 4);
1296 if (inputsRead & VARYING_BIT_COL1) {
1297 intel->specoffset = offset / 4;
1298 EMIT_ATTR(_TNL_ATTRIB_COLOR1, EMIT_4UB_4F_BGRA, S4_VFMT_SPEC_FOG, 4);
1301 if ((inputsRead & VARYING_BIT_FOGC)) {
1302 EMIT_ATTR(_TNL_ATTRIB_FOG, EMIT_1F, S4_VFMT_FOG_PARAM, 4);
1305 for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
1306 if (inputsRead & VARYING_BIT_TEX(i)) {
1307 int unit = get_texcoord_mapping(p, i | TEXCOORD_TEX);
1308 int sz = VB->AttribPtr[_TNL_ATTRIB_TEX0 + i]->size;
1310 s2 &= ~S2_TEXCOORD_FMT(unit, S2_TEXCOORD_FMT0_MASK);
1311 s2 |= S2_TEXCOORD_FMT(unit, SZ_TO_HW(sz));
1313 EMIT_ATTR(_TNL_ATTRIB_TEX0 + i, EMIT_SZ(sz), 0, sz * 4);
1315 if (inputsRead & VARYING_BIT_VAR(i)) {
1316 int unit = get_texcoord_mapping(p, i | TEXCOORD_VAR);
1317 int sz = VB->AttribPtr[_TNL_ATTRIB_GENERIC0 + i]->size;
1319 s2 &= ~S2_TEXCOORD_FMT(unit, S2_TEXCOORD_FMT0_MASK);
1320 s2 |= S2_TEXCOORD_FMT(unit, SZ_TO_HW(sz));
1322 EMIT_ATTR(_TNL_ATTRIB_GENERIC0 + i, EMIT_SZ(sz), 0, sz * 4);
1324 if (i == p->wpos_tex) {
1325 int wpos_size = 4 * sizeof(float);
1326 /* If WPOS is required, duplicate the XYZ position data in an
1327 * unused texture coordinate:
1329 s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1330 s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(wpos_size));
1332 intel->wpos_offset = offset;
1333 EMIT_PAD(wpos_size);
1337 if (s2 != i915->state.Ctx[I915_CTXREG_LIS2] ||
1338 s4 != i915->state.Ctx[I915_CTXREG_LIS4]) {
1339 I915_STATECHANGE(i915, I915_UPLOAD_CTX);
1341 /* Must do this *after* statechange, so as not to affect
1342 * buffered vertices reliant on the old state:
1344 intel->vertex_size = _tnl_install_attrs(&intel->ctx,
1345 intel->vertex_attrs,
1346 intel->vertex_attr_count,
1347 intel->ViewportMatrix.m, 0);
1349 assert(intel->prim.current_offset == intel->prim.start_offset);
1350 intel->prim.start_offset = (intel->prim.current_offset + intel->vertex_size-1) / intel->vertex_size * intel->vertex_size;
1351 intel->prim.current_offset = intel->prim.start_offset;
1353 intel->vertex_size >>= 2;
1355 i915->state.Ctx[I915_CTXREG_LIS2] = s2;
1356 i915->state.Ctx[I915_CTXREG_LIS4] = s4;
1358 assert(intel->vtbl.check_vertex_size(intel, intel->vertex_size));
1361 if (!p->params_uptodate)
1364 if (!p->on_hardware)
1365 i915_upload_program(i915, p);
1367 if (INTEL_DEBUG & DEBUG_WM) {
1369 i915_disassemble_program(i915->state.Program, i915->state.ProgramSize);
1374 i915InitFragProgFuncs(struct dd_function_table *functions)
1376 functions->BindProgram = i915BindProgram;
1377 functions->NewProgram = i915NewProgram;
1378 functions->DeleteProgram = i915DeleteProgram;
1379 functions->IsProgramNative = i915IsProgramNative;
1380 functions->ProgramStringNotify = i915ProgramStringNotify;
1381 functions->SamplerUniformChange = i915SamplerUniformChange;