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)
79 static GLint get_texcoord_mapping(struct i915_fragment_program *p,
84 for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
85 if (p->texcoord_mapping[i] == texcoord)
90 return p->ctx->Const.MaxTextureCoordUnits - 1;
94 * Retrieve a ureg for the given source register. Will emit
95 * constants, apply swizzling and negation as needed.
98 src_vector(struct i915_fragment_program *p,
99 const struct prog_src_register *source,
100 const struct gl_fragment_program *program)
105 switch (source->File) {
109 case PROGRAM_TEMPORARY:
110 if (source->Index >= I915_MAX_TEMPORARY) {
111 i915_program_error(p, "Exceeded max temporary reg: %d/%d",
112 source->Index, I915_MAX_TEMPORARY);
115 src = UREG(REG_TYPE_R, source->Index);
118 switch (source->Index) {
119 case VARYING_SLOT_POS:
120 src = i915_emit_decl(p, REG_TYPE_T, p->wpos_tex, D0_CHANNEL_ALL);
122 case VARYING_SLOT_COL0:
123 src = i915_emit_decl(p, REG_TYPE_T, T_DIFFUSE, D0_CHANNEL_ALL);
125 case VARYING_SLOT_COL1:
126 src = i915_emit_decl(p, REG_TYPE_T, T_SPECULAR, D0_CHANNEL_XYZ);
127 src = swizzle(src, X, Y, Z, ONE);
129 case VARYING_SLOT_FOGC:
130 src = i915_emit_decl(p, REG_TYPE_T, T_FOG_W, D0_CHANNEL_W);
131 src = swizzle(src, W, ZERO, ZERO, ONE);
133 case VARYING_SLOT_TEX0:
134 case VARYING_SLOT_TEX1:
135 case VARYING_SLOT_TEX2:
136 case VARYING_SLOT_TEX3:
137 case VARYING_SLOT_TEX4:
138 case VARYING_SLOT_TEX5:
139 case VARYING_SLOT_TEX6:
140 case VARYING_SLOT_TEX7:
141 unit = get_texcoord_mapping(p, (source->Index -
142 VARYING_SLOT_TEX0) | TEXCOORD_TEX);
143 src = i915_emit_decl(p, REG_TYPE_T,
148 case VARYING_SLOT_VAR0:
149 case VARYING_SLOT_VAR0 + 1:
150 case VARYING_SLOT_VAR0 + 2:
151 case VARYING_SLOT_VAR0 + 3:
152 case VARYING_SLOT_VAR0 + 4:
153 case VARYING_SLOT_VAR0 + 5:
154 case VARYING_SLOT_VAR0 + 6:
155 case VARYING_SLOT_VAR0 + 7:
156 unit = get_texcoord_mapping(p, (source->Index -
157 VARYING_SLOT_VAR0) | TEXCOORD_VAR);
158 src = i915_emit_decl(p, REG_TYPE_T,
164 i915_program_error(p, "Bad source->Index: %d", source->Index);
170 switch (source->Index) {
171 case FRAG_RESULT_COLOR:
172 case FRAG_RESULT_DATA0:
173 src = UREG(REG_TYPE_OC, 0);
175 case FRAG_RESULT_DEPTH:
176 src = UREG(REG_TYPE_OD, 0);
179 i915_program_error(p, "Bad source->Index: %d", source->Index);
184 /* Various paramters and env values. All emitted to
185 * hardware as program constants.
187 case PROGRAM_CONSTANT:
188 case PROGRAM_STATE_VAR:
189 case PROGRAM_UNIFORM:
190 src = i915_emit_param4fv(p,
191 &program->Base.Parameters->ParameterValues[source->Index][0].f);
195 i915_program_error(p, "Bad source->File: %d", source->File);
200 GET_SWZ(source->Swizzle, 0),
201 GET_SWZ(source->Swizzle, 1),
202 GET_SWZ(source->Swizzle, 2), GET_SWZ(source->Swizzle, 3));
206 GET_BIT(source->Negate, 0),
207 GET_BIT(source->Negate, 1),
208 GET_BIT(source->Negate, 2),
209 GET_BIT(source->Negate, 3));
216 get_result_vector(struct i915_fragment_program *p,
217 const struct prog_instruction *inst)
219 switch (inst->DstReg.File) {
221 switch (inst->DstReg.Index) {
222 case FRAG_RESULT_COLOR:
223 case FRAG_RESULT_DATA0:
224 return UREG(REG_TYPE_OC, 0);
225 case FRAG_RESULT_DEPTH:
226 p->depth_written = 1;
227 return UREG(REG_TYPE_OD, 0);
229 i915_program_error(p, "Bad inst->DstReg.Index: %d",
233 case PROGRAM_TEMPORARY:
234 return UREG(REG_TYPE_R, inst->DstReg.Index);
236 i915_program_error(p, "Bad inst->DstReg.File: %d", inst->DstReg.File);
242 get_result_flags(const struct prog_instruction *inst)
246 if (inst->SaturateMode == SATURATE_ZERO_ONE)
247 flags |= A0_DEST_SATURATE;
248 if (inst->DstReg.WriteMask & WRITEMASK_X)
249 flags |= A0_DEST_CHANNEL_X;
250 if (inst->DstReg.WriteMask & WRITEMASK_Y)
251 flags |= A0_DEST_CHANNEL_Y;
252 if (inst->DstReg.WriteMask & WRITEMASK_Z)
253 flags |= A0_DEST_CHANNEL_Z;
254 if (inst->DstReg.WriteMask & WRITEMASK_W)
255 flags |= A0_DEST_CHANNEL_W;
261 translate_tex_src_target(struct i915_fragment_program *p, GLubyte bit)
264 case TEXTURE_1D_INDEX:
265 return D0_SAMPLE_TYPE_2D;
266 case TEXTURE_2D_INDEX:
267 case TEXTURE_EXTERNAL_INDEX:
268 return D0_SAMPLE_TYPE_2D;
269 case TEXTURE_RECT_INDEX:
270 return D0_SAMPLE_TYPE_2D;
271 case TEXTURE_3D_INDEX:
272 return D0_SAMPLE_TYPE_VOLUME;
273 case TEXTURE_CUBE_INDEX:
274 return D0_SAMPLE_TYPE_CUBE;
276 i915_program_error(p, "TexSrcBit: %d", bit);
281 #define EMIT_TEX( OP ) \
283 GLuint dim = translate_tex_src_target( p, inst->TexSrcTarget ); \
284 const struct gl_fragment_program *program = &p->FragProg; \
285 GLuint unit = program->Base.SamplerUnits[inst->TexSrcUnit]; \
286 GLuint sampler = i915_emit_decl(p, REG_TYPE_S, \
288 GLuint coord = src_vector( p, &inst->SrcReg[0], program); \
291 i915_emit_texld( p, get_live_regs(p, inst), \
292 get_result_vector( p, inst ), \
293 get_result_flags( inst ), \
299 #define EMIT_ARITH( OP, N ) \
301 i915_emit_arith( p, \
303 get_result_vector( p, inst ), \
304 get_result_flags( inst ), 0, \
305 (N<1)?0:src_vector( p, &inst->SrcReg[0], program), \
306 (N<2)?0:src_vector( p, &inst->SrcReg[1], program), \
307 (N<3)?0:src_vector( p, &inst->SrcReg[2], program)); \
310 #define EMIT_1ARG_ARITH( OP ) EMIT_ARITH( OP, 1 )
311 #define EMIT_2ARG_ARITH( OP ) EMIT_ARITH( OP, 2 )
312 #define EMIT_3ARG_ARITH( OP ) EMIT_ARITH( OP, 3 )
315 * TODO: consider moving this into core
317 static bool calc_live_regs( struct i915_fragment_program *p )
319 const struct gl_fragment_program *program = &p->FragProg;
320 GLuint regsUsed = ~((1 << I915_MAX_TEMPORARY) - 1);
321 uint8_t live_components[I915_MAX_TEMPORARY] = { 0, };
324 for (i = program->Base.NumInstructions - 1; i >= 0; i--) {
325 struct prog_instruction *inst = &program->Base.Instructions[i];
326 int opArgs = _mesa_num_inst_src_regs(inst->Opcode);
329 /* Register is written to: unmark as live for this and preceeding ops */
330 if (inst->DstReg.File == PROGRAM_TEMPORARY) {
331 if (inst->DstReg.Index >= I915_MAX_TEMPORARY)
334 live_components[inst->DstReg.Index] &= ~inst->DstReg.WriteMask;
335 if (live_components[inst->DstReg.Index] == 0)
336 regsUsed &= ~(1 << inst->DstReg.Index);
339 for (a = 0; a < opArgs; a++) {
340 /* Register is read from: mark as live for this and preceeding ops */
341 if (inst->SrcReg[a].File == PROGRAM_TEMPORARY) {
344 if (inst->SrcReg[a].Index >= I915_MAX_TEMPORARY)
347 regsUsed |= 1 << inst->SrcReg[a].Index;
349 for (c = 0; c < 4; c++) {
350 const unsigned field = GET_SWZ(inst->SrcReg[a].Swizzle, c);
352 if (field <= SWIZZLE_W)
353 live_components[inst->SrcReg[a].Index] |= (1U << field);
358 p->usedRegs[i] = regsUsed;
364 static GLuint get_live_regs( struct i915_fragment_program *p,
365 const struct prog_instruction *inst )
367 const struct gl_fragment_program *program = &p->FragProg;
368 GLuint nr = inst - program->Base.Instructions;
370 return p->usedRegs[nr];
374 /* Possible concerns:
376 * SIN, COS -- could use another taylor step?
377 * LIT -- results seem a little different to sw mesa
378 * LOG -- different to mesa on negative numbers, but this is conformant.
380 * Parse failures -- Mesa doesn't currently give a good indication
381 * internally whether a particular program string parsed or not. This
382 * can lead to confusion -- hopefully we cope with it ok now.
386 upload_program(struct i915_fragment_program *p)
388 const struct gl_fragment_program *program = &p->FragProg;
389 const struct prog_instruction *inst = program->Base.Instructions;
391 if (INTEL_DEBUG & DEBUG_WM)
392 _mesa_print_program(&program->Base);
394 /* Is this a parse-failed program? Ensure a valid program is
395 * loaded, as the flagging of an error isn't sufficient to stop
396 * this being uploaded to hardware.
398 if (inst[0].Opcode == OPCODE_END) {
399 GLuint tmp = i915_get_utemp(p);
402 UREG(REG_TYPE_OC, 0),
403 A0_DEST_CHANNEL_ALL, 0,
404 swizzle(tmp, ONE, ZERO, ONE, ONE), 0, 0);
408 if (program->Base.NumInstructions > I915_MAX_INSN) {
409 i915_program_error(p, "Exceeded max instructions (%d out of %d)",
410 program->Base.NumInstructions, I915_MAX_INSN);
414 /* Not always needed:
416 if (!calc_live_regs(p)) {
417 i915_program_error(p, "Could not allocate registers");
422 GLuint src0, src1, src2, flags;
423 GLuint tmp = 0, dst, consts0 = 0, consts1 = 0;
425 switch (inst->Opcode) {
427 src0 = src_vector(p, &inst->SrcReg[0], program);
430 get_result_vector(p, inst),
431 get_result_flags(inst), 0,
432 src0, negate(src0, 1, 1, 1, 1), 0);
436 EMIT_2ARG_ARITH(A0_ADD);
440 src0 = src_vector(p, &inst->SrcReg[0], program);
441 src1 = src_vector(p, &inst->SrcReg[1], program);
442 src2 = src_vector(p, &inst->SrcReg[2], program);
443 i915_emit_arith(p, A0_CMP, get_result_vector(p, inst), get_result_flags(inst), 0, src0, src2, src1); /* NOTE: order of src2, src1 */
447 src0 = src_vector(p, &inst->SrcReg[0], program);
448 tmp = i915_get_utemp(p);
449 consts0 = i915_emit_const4fv(p, sin_quad_constants[0]);
450 consts1 = i915_emit_const4fv(p, sin_quad_constants[1]);
452 /* Reduce range from repeating about [-pi,pi] to [-1,1] */
455 tmp, A0_DEST_CHANNEL_X, 0,
457 swizzle(consts1, Z, ZERO, ZERO, ZERO), /* 1/(2pi) */
458 swizzle(consts0, W, ZERO, ZERO, ZERO)); /* .75 */
460 i915_emit_arith(p, A0_FRC, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
464 tmp, A0_DEST_CHANNEL_X, 0,
466 swizzle(consts0, X, ZERO, ZERO, ZERO), /* 2 */
467 swizzle(consts0, Y, ZERO, ZERO, ZERO)); /* -1 */
469 /* Compute COS with the same calculation used for SIN, but a
470 * different source range has been mapped to [-1,1] this time.
473 /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
476 tmp, A0_DEST_CHANNEL_Y, 0,
477 swizzle(tmp, ZERO, X, ZERO, ZERO),
478 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
481 /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
484 tmp, A0_DEST_CHANNEL_Y, 0,
485 swizzle(tmp, ZERO, X, ZERO, ZERO),
489 /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
492 tmp, A0_DEST_CHANNEL_X, 0,
494 swizzle(consts1, X, Y, ZERO, ZERO),
497 /* tmp.x now contains a first approximation (y). Now, weight it
498 * against tmp.y**2 to get closer.
502 tmp, A0_DEST_CHANNEL_Y, 0,
503 swizzle(tmp, ZERO, X, ZERO, ZERO),
504 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
507 /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
510 tmp, A0_DEST_CHANNEL_Y, 0,
511 swizzle(tmp, ZERO, X, ZERO, ZERO),
512 swizzle(tmp, ZERO, Y, ZERO, ZERO),
513 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0));
515 /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
518 get_result_vector(p, inst),
519 get_result_flags(inst), 0,
520 swizzle(consts1, W, W, W, W),
521 swizzle(tmp, Y, Y, Y, Y),
522 swizzle(tmp, X, X, X, X));
526 src0 = src_vector(p, &inst->SrcReg[0], program);
527 src1 = src_vector(p, &inst->SrcReg[1], program);
530 get_result_vector(p, inst),
531 get_result_flags(inst), 0,
532 swizzle(src0, X, Y, ZERO, ZERO),
533 swizzle(src1, X, Y, ZERO, ZERO),
538 EMIT_2ARG_ARITH(A0_DP3);
542 EMIT_2ARG_ARITH(A0_DP4);
546 src0 = src_vector(p, &inst->SrcReg[0], program);
547 src1 = src_vector(p, &inst->SrcReg[1], program);
551 get_result_vector(p, inst),
552 get_result_flags(inst), 0,
553 swizzle(src0, X, Y, Z, ONE), src1, 0);
557 src0 = src_vector(p, &inst->SrcReg[0], program);
558 src1 = src_vector(p, &inst->SrcReg[1], program);
560 /* result[0] = 1 * 1;
561 * result[1] = a[1] * b[1];
562 * result[2] = a[2] * 1;
563 * result[3] = 1 * b[3];
567 get_result_vector(p, inst),
568 get_result_flags(inst), 0,
569 swizzle(src0, ONE, Y, Z, ONE),
570 swizzle(src1, ONE, Y, ONE, W), 0);
574 src0 = src_vector(p, &inst->SrcReg[0], program);
578 get_result_vector(p, inst),
579 get_result_flags(inst), 0,
580 swizzle(src0, X, X, X, X), 0, 0);
584 EMIT_1ARG_ARITH(A0_FLR);
588 EMIT_1ARG_ARITH(A0_TRC);
592 EMIT_1ARG_ARITH(A0_FRC);
596 src0 = src_vector(p, &inst->SrcReg[0], program);
597 tmp = i915_get_utemp(p);
599 i915_emit_texld(p, get_live_regs(p, inst),
600 tmp, A0_DEST_CHANNEL_ALL, /* use a dummy dest reg */
601 0, src0, T0_TEXKILL);
605 if (inst->DstReg.CondMask == COND_TR) {
606 tmp = i915_get_utemp(p);
608 /* The KIL instruction discards the fragment if any component of
609 * the source is < 0. Emit an immediate operand of {-1}.xywz.
611 i915_emit_texld(p, get_live_regs(p, inst),
612 tmp, A0_DEST_CHANNEL_ALL,
613 0, /* use a dummy dest reg */
614 negate(swizzle(tmp, ONE, ONE, ONE, ONE),
619 i915_program_error(p, "Unsupported KIL_NV condition code: %d",
620 inst->DstReg.CondMask);
625 src0 = src_vector(p, &inst->SrcReg[0], program);
629 get_result_vector(p, inst),
630 get_result_flags(inst), 0,
631 swizzle(src0, X, X, X, X), 0, 0);
635 src0 = src_vector(p, &inst->SrcReg[0], program);
636 tmp = i915_get_utemp(p);
638 /* tmp = max( a.xyzw, a.00zw )
639 * XXX: Clamp tmp.w to -128..128
641 * tmp.y = tmp.w * tmp.y
643 * result = cmp (a.11-x1, a.1x01, a.1xy1 )
645 i915_emit_arith(p, A0_MAX, tmp, A0_DEST_CHANNEL_ALL, 0,
646 src0, swizzle(src0, ZERO, ZERO, Z, W), 0);
648 i915_emit_arith(p, A0_LOG, tmp, A0_DEST_CHANNEL_Y, 0,
649 swizzle(tmp, Y, Y, Y, Y), 0, 0);
651 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_Y, 0,
652 swizzle(tmp, ZERO, Y, ZERO, ZERO),
653 swizzle(tmp, ZERO, W, ZERO, ZERO), 0);
655 i915_emit_arith(p, A0_EXP, tmp, A0_DEST_CHANNEL_Y, 0,
656 swizzle(tmp, Y, Y, Y, Y), 0, 0);
658 i915_emit_arith(p, A0_CMP,
659 get_result_vector(p, inst),
660 get_result_flags(inst), 0,
661 negate(swizzle(tmp, ONE, ONE, X, ONE), 0, 0, 1, 0),
662 swizzle(tmp, ONE, X, ZERO, ONE),
663 swizzle(tmp, ONE, X, Y, ONE));
668 src0 = src_vector(p, &inst->SrcReg[0], program);
669 src1 = src_vector(p, &inst->SrcReg[1], program);
670 src2 = src_vector(p, &inst->SrcReg[2], program);
671 flags = get_result_flags(inst);
672 tmp = i915_get_utemp(p);
679 * result = (-c)*a + tmp
681 i915_emit_arith(p, A0_MAD, tmp,
682 flags & A0_DEST_CHANNEL_ALL, 0, src1, src0, src2);
684 i915_emit_arith(p, A0_MAD,
685 get_result_vector(p, inst),
686 flags, 0, negate(src2, 1, 1, 1, 1), src0, tmp);
690 EMIT_3ARG_ARITH(A0_MAD);
694 EMIT_2ARG_ARITH(A0_MAX);
698 src0 = src_vector(p, &inst->SrcReg[0], program);
699 src1 = src_vector(p, &inst->SrcReg[1], program);
700 tmp = i915_get_utemp(p);
701 flags = get_result_flags(inst);
705 tmp, flags & A0_DEST_CHANNEL_ALL, 0,
706 negate(src0, 1, 1, 1, 1),
707 negate(src1, 1, 1, 1, 1), 0);
711 get_result_vector(p, inst),
712 flags, 0, negate(tmp, 1, 1, 1, 1), 0, 0);
716 EMIT_1ARG_ARITH(A0_MOV);
720 EMIT_2ARG_ARITH(A0_MUL);
724 src0 = src_vector(p, &inst->SrcReg[0], program);
725 src1 = src_vector(p, &inst->SrcReg[1], program);
726 tmp = i915_get_utemp(p);
727 flags = get_result_flags(inst);
729 /* XXX: masking on intermediate values, here and elsewhere.
733 tmp, A0_DEST_CHANNEL_X, 0,
734 swizzle(src0, X, X, X, X), 0, 0);
736 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_X, 0, tmp, src1, 0);
741 get_result_vector(p, inst),
742 flags, 0, swizzle(tmp, X, X, X, X), 0, 0);
747 src0 = src_vector(p, &inst->SrcReg[0], program);
751 get_result_vector(p, inst),
752 get_result_flags(inst), 0,
753 swizzle(src0, X, X, X, X), 0, 0);
758 src0 = src_vector(p, &inst->SrcReg[0], program);
762 get_result_vector(p, inst),
763 get_result_flags(inst), 0,
764 swizzle(src0, X, X, X, X), 0, 0);
768 src0 = src_vector(p, &inst->SrcReg[0], program);
769 tmp = i915_get_utemp(p);
772 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
773 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
774 * t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x
775 * scs.x = DP4 t1, sin_constants
776 * t1 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1
777 * scs.y = DP4 t1, cos_constants
781 tmp, A0_DEST_CHANNEL_XY, 0,
782 swizzle(src0, X, X, ONE, ONE),
783 swizzle(src0, X, ONE, ONE, ONE), 0);
787 tmp, A0_DEST_CHANNEL_ALL, 0,
788 swizzle(tmp, X, Y, X, Y),
789 swizzle(tmp, X, X, ONE, ONE), 0);
791 if (inst->DstReg.WriteMask & WRITEMASK_Y) {
794 if (inst->DstReg.WriteMask & WRITEMASK_X)
795 tmp1 = i915_get_utemp(p);
801 tmp1, A0_DEST_CHANNEL_ALL, 0,
802 swizzle(tmp, X, Y, Y, W),
803 swizzle(tmp, X, Z, ONE, ONE), 0);
807 get_result_vector(p, inst),
808 A0_DEST_CHANNEL_Y, 0,
809 swizzle(tmp1, W, Z, Y, X),
810 i915_emit_const4fv(p, sin_constants), 0);
813 if (inst->DstReg.WriteMask & WRITEMASK_X) {
816 tmp, A0_DEST_CHANNEL_XYZ, 0,
817 swizzle(tmp, X, X, Z, ONE),
818 swizzle(tmp, Z, ONE, ONE, ONE), 0);
822 get_result_vector(p, inst),
823 A0_DEST_CHANNEL_X, 0,
824 swizzle(tmp, ONE, Z, Y, X),
825 i915_emit_const4fv(p, cos_constants), 0);
830 tmp = i915_get_utemp(p);
831 flags = get_result_flags(inst);
832 dst = get_result_vector(p, inst);
834 /* If both operands are uniforms or constants, we get 5 instructions
837 * U[1] = MOV CONST[1]
838 * U[0].xyz = SGE CONST[0].xxxx, U[1]
839 * U[1] = MOV CONST[1].-x-y-z-w
840 * R[0].xyz = SGE CONST[0].-x-x-x-x, U[1]
841 * R[0].xyz = MUL R[0], U[0]
843 * This code is stupid. Instead of having the individual calls to
844 * i915_emit_arith generate the moves to utemps, do it in the caller.
845 * This results in code like:
847 * U[1] = MOV CONST[1]
848 * U[0].xyz = SGE CONST[0].xxxx, U[1]
849 * R[0].xyz = SGE CONST[0].-x-x-x-x, U[1].-x-y-z-w
850 * R[0].xyz = MUL R[0], U[0]
852 src0 = src_vector(p, &inst->SrcReg[0], program);
853 src1 = src_vector(p, &inst->SrcReg[1], program);
855 if (GET_UREG_TYPE(src0) == REG_TYPE_CONST
856 && GET_UREG_TYPE(src1) == REG_TYPE_CONST) {
857 unsigned tmp = i915_get_utemp(p);
859 i915_emit_arith(p, A0_MOV, tmp, A0_DEST_CHANNEL_ALL, 0,
865 /* tmp = src1 >= src2 */
873 /* dst = src1 <= src2 */
878 negate(src0, 1, 1, 1, 1),
879 negate(src1, 1, 1, 1, 1),
881 /* dst = tmp && dst */
892 src0 = src_vector(p, &inst->SrcReg[0], program);
893 tmp = i915_get_utemp(p);
894 consts0 = i915_emit_const4fv(p, sin_quad_constants[0]);
895 consts1 = i915_emit_const4fv(p, sin_quad_constants[1]);
897 /* Reduce range from repeating about [-pi,pi] to [-1,1] */
900 tmp, A0_DEST_CHANNEL_X, 0,
902 swizzle(consts1, Z, ZERO, ZERO, ZERO), /* 1/(2pi) */
903 swizzle(consts0, Z, ZERO, ZERO, ZERO)); /* .5 */
905 i915_emit_arith(p, A0_FRC, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
909 tmp, A0_DEST_CHANNEL_X, 0,
911 swizzle(consts0, X, ZERO, ZERO, ZERO), /* 2 */
912 swizzle(consts0, Y, ZERO, ZERO, ZERO)); /* -1 */
914 /* Compute sin using a quadratic and quartic. It gives continuity
915 * that repeating the Taylor series lacks every 2*pi, and has
918 * The idea was described at:
919 * http://www.devmaster.net/forums/showthread.php?t=5784
922 /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
925 tmp, A0_DEST_CHANNEL_Y, 0,
926 swizzle(tmp, ZERO, X, ZERO, ZERO),
927 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
930 /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
933 tmp, A0_DEST_CHANNEL_Y, 0,
934 swizzle(tmp, ZERO, X, ZERO, ZERO),
938 /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
941 tmp, A0_DEST_CHANNEL_X, 0,
943 swizzle(consts1, X, Y, ZERO, ZERO),
946 /* tmp.x now contains a first approximation (y). Now, weight it
947 * against tmp.y**2 to get closer.
951 tmp, A0_DEST_CHANNEL_Y, 0,
952 swizzle(tmp, ZERO, X, ZERO, ZERO),
953 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
956 /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
959 tmp, A0_DEST_CHANNEL_Y, 0,
960 swizzle(tmp, ZERO, X, ZERO, ZERO),
961 swizzle(tmp, ZERO, Y, ZERO, ZERO),
962 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0));
964 /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
967 get_result_vector(p, inst),
968 get_result_flags(inst), 0,
969 swizzle(consts1, W, W, W, W),
970 swizzle(tmp, Y, Y, Y, Y),
971 swizzle(tmp, X, X, X, X));
976 EMIT_2ARG_ARITH(A0_SGE);
982 get_result_vector( p, inst ),
983 get_result_flags( inst ), 0,
984 negate(src_vector( p, &inst->SrcReg[0], program),
986 negate(src_vector( p, &inst->SrcReg[1], program),
994 get_result_vector( p, inst ),
995 get_result_flags( inst ), 0,
996 negate(src_vector( p, &inst->SrcReg[0], program),
998 negate(src_vector( p, &inst->SrcReg[1], program),
1004 EMIT_2ARG_ARITH(A0_SLT);
1008 tmp = i915_get_utemp(p);
1009 flags = get_result_flags(inst);
1010 dst = get_result_vector(p, inst);
1012 /* If both operands are uniforms or constants, we get 5 instructions
1015 * U[1] = MOV CONST[1]
1016 * U[0].xyz = SLT CONST[0].xxxx, U[1]
1017 * U[1] = MOV CONST[1].-x-y-z-w
1018 * R[0].xyz = SLT CONST[0].-x-x-x-x, U[1]
1019 * R[0].xyz = MUL R[0], U[0]
1021 * This code is stupid. Instead of having the individual calls to
1022 * i915_emit_arith generate the moves to utemps, do it in the caller.
1023 * This results in code like:
1025 * U[1] = MOV CONST[1]
1026 * U[0].xyz = SLT CONST[0].xxxx, U[1]
1027 * R[0].xyz = SLT CONST[0].-x-x-x-x, U[1].-x-y-z-w
1028 * R[0].xyz = MUL R[0], U[0]
1030 src0 = src_vector(p, &inst->SrcReg[0], program);
1031 src1 = src_vector(p, &inst->SrcReg[1], program);
1033 if (GET_UREG_TYPE(src0) == REG_TYPE_CONST
1034 && GET_UREG_TYPE(src1) == REG_TYPE_CONST) {
1035 unsigned tmp = i915_get_utemp(p);
1037 i915_emit_arith(p, A0_MOV, tmp, A0_DEST_CHANNEL_ALL, 0,
1043 /* tmp = src1 < src2 */
1051 /* dst = src1 > src2 */
1056 negate(src0, 1, 1, 1, 1),
1057 negate(src1, 1, 1, 1, 1),
1059 /* dst = tmp || dst */
1063 flags | A0_DEST_SATURATE, 0,
1070 dst = get_result_vector(p, inst);
1071 flags = get_result_flags(inst);
1072 src0 = src_vector(p, &inst->SrcReg[0], program);
1073 tmp = i915_get_utemp(p);
1075 /* tmp = (src < 0.0) */
1081 swizzle(src0, ZERO, ZERO, ZERO, ZERO),
1084 /* dst = (0.0 < src) */
1089 swizzle(src0, ZERO, ZERO, ZERO, ZERO),
1093 /* dst = (src > 0.0) - (src < 0.0) */
1099 negate(tmp, 1, 1, 1, 1),
1105 src0 = src_vector(p, &inst->SrcReg[0], program);
1106 src1 = src_vector(p, &inst->SrcReg[1], program);
1110 get_result_vector(p, inst),
1111 get_result_flags(inst), 0,
1112 src0, negate(src1, 1, 1, 1, 1), 0);
1116 EMIT_1ARG_ARITH(A0_MOV); /* extended swizzle handled natively */
1124 EMIT_TEX(T0_TEXLDB);
1128 EMIT_TEX(T0_TEXLDP);
1133 * result.x = src0.y * src1.z - src0.z * src1.y;
1134 * result.y = src0.z * src1.x - src0.x * src1.z;
1135 * result.z = src0.x * src1.y - src0.y * src1.x;
1138 src0 = src_vector(p, &inst->SrcReg[0], program);
1139 src1 = src_vector(p, &inst->SrcReg[1], program);
1140 tmp = i915_get_utemp(p);
1144 tmp, A0_DEST_CHANNEL_ALL, 0,
1145 swizzle(src0, Z, X, Y, ONE),
1146 swizzle(src1, Y, Z, X, ONE), 0);
1150 get_result_vector(p, inst),
1151 get_result_flags(inst), 0,
1152 swizzle(src0, Y, Z, X, ONE),
1153 swizzle(src1, Z, X, Y, ONE),
1154 negate(tmp, 1, 1, 1, 0));
1160 case OPCODE_BGNLOOP:
1169 case OPCODE_ENDLOOP:
1174 i915_program_error(p, "Unsupported opcode: %s",
1175 _mesa_opcode_string(inst->Opcode));
1180 /* These opcodes are claimed as GLSL, NV_vp, and ARB_vp in
1181 * prog_instruction.h, but apparently GLSL doesn't ever emit them.
1182 * Instead, it translates to EX2 or LG2.
1186 /* These opcodes are claimed by GLSL in prog_instruction.h, but
1187 * only NV_vp/fp appears to emit them.
1190 i915_program_error(p, "bad opcode: %s",
1191 _mesa_opcode_string(inst->Opcode));
1196 i915_release_utemps(p);
1200 /* Rather than trying to intercept and jiggle depth writes during
1201 * emit, just move the value into its correct position at the end of
1205 fixup_depth_write(struct i915_fragment_program *p)
1207 if (p->depth_written) {
1208 GLuint depth = UREG(REG_TYPE_OD, 0);
1212 depth, A0_DEST_CHANNEL_W, 0,
1213 swizzle(depth, X, Y, Z, Z), 0, 0);
1218 check_texcoord_mapping(struct i915_fragment_program *p)
1220 GLbitfield64 inputs = p->FragProg.Base.InputsRead;
1224 for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
1225 if (inputs & VARYING_BIT_TEX(i)) {
1226 if (unit >= p->ctx->Const.MaxTextureCoordUnits) {
1230 p->texcoord_mapping[unit++] = i | TEXCOORD_TEX;
1232 if (inputs & VARYING_BIT_VAR(i)) {
1233 if (unit >= p->ctx->Const.MaxTextureCoordUnits) {
1237 p->texcoord_mapping[unit++] = i | TEXCOORD_VAR;
1241 if (unit > p->ctx->Const.MaxTextureCoordUnits)
1242 i915_program_error(p, "Too many texcoord units");
1246 check_wpos(struct i915_fragment_program *p)
1248 GLbitfield64 inputs = p->FragProg.Base.InputsRead;
1254 if ((inputs & VARYING_BIT_POS) == 0)
1257 for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
1258 unit += !!(inputs & VARYING_BIT_TEX(i));
1259 unit += !!(inputs & VARYING_BIT_VAR(i));
1262 if (unit < p->ctx->Const.MaxTextureCoordUnits)
1265 i915_program_error(p, "No free texcoord for wpos value");
1270 translate_program(struct i915_fragment_program *p)
1272 struct i915_context *i915 = I915_CONTEXT(p->ctx);
1274 if (INTEL_DEBUG & DEBUG_WM) {
1276 _mesa_print_program(&p->FragProg.Base);
1280 i915_init_program(i915, p);
1281 check_texcoord_mapping(p);
1284 fixup_depth_write(p);
1285 i915_fini_program(p);
1292 track_params(struct i915_fragment_program *p)
1297 _mesa_load_state_parameters(p->ctx, p->FragProg.Base.Parameters);
1299 for (i = 0; i < p->nr_params; i++) {
1300 GLint reg = p->param[i].reg;
1301 COPY_4V(p->constant[reg], p->param[i].values);
1304 p->params_uptodate = 1;
1305 p->on_hardware = 0; /* overkill */
1310 i915BindProgram(struct gl_context * ctx, GLenum target, struct gl_program *prog)
1312 if (target == GL_FRAGMENT_PROGRAM_ARB) {
1313 struct i915_context *i915 = I915_CONTEXT(ctx);
1314 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1316 if (i915->current_program == p)
1319 if (i915->current_program) {
1320 i915->current_program->on_hardware = 0;
1321 i915->current_program->params_uptodate = 0;
1324 i915->current_program = p;
1326 assert(p->on_hardware == 0);
1327 assert(p->params_uptodate == 0);
1332 static struct gl_program *
1333 i915NewProgram(struct gl_context * ctx, GLenum target, GLuint id)
1336 case GL_VERTEX_PROGRAM_ARB:
1337 return _mesa_init_vertex_program(ctx, CALLOC_STRUCT(gl_vertex_program),
1340 case GL_FRAGMENT_PROGRAM_ARB:{
1341 struct i915_fragment_program *prog =
1342 CALLOC_STRUCT(i915_fragment_program);
1344 i915_init_program(I915_CONTEXT(ctx), prog);
1346 return _mesa_init_fragment_program(ctx, &prog->FragProg,
1356 return _mesa_new_program(ctx, target, id);
1361 i915DeleteProgram(struct gl_context * ctx, struct gl_program *prog)
1363 if (prog->Target == GL_FRAGMENT_PROGRAM_ARB) {
1364 struct i915_context *i915 = I915_CONTEXT(ctx);
1365 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1367 if (i915->current_program == p)
1368 i915->current_program = 0;
1371 _mesa_delete_program(ctx, prog);
1376 i915IsProgramNative(struct gl_context * ctx, GLenum target, struct gl_program *prog)
1378 if (target == GL_FRAGMENT_PROGRAM_ARB) {
1379 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1382 translate_program(p);
1391 i915ProgramStringNotify(struct gl_context * ctx,
1392 GLenum target, struct gl_program *prog)
1394 if (target == GL_FRAGMENT_PROGRAM_ARB) {
1395 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1399 (void) _tnl_program_string(ctx, target, prog);
1401 /* XXX check if program is legal, within limits */
1406 i915SamplerUniformChange(struct gl_context *ctx,
1407 GLenum target, struct gl_program *prog)
1409 i915ProgramStringNotify(ctx, target, prog);
1413 i915_update_program(struct gl_context *ctx)
1415 struct intel_context *intel = intel_context(ctx);
1416 struct i915_context *i915 = i915_context(&intel->ctx);
1417 struct i915_fragment_program *fp =
1418 (struct i915_fragment_program *) ctx->FragmentProgram._Current;
1420 if (i915->current_program != fp) {
1421 if (i915->current_program) {
1422 i915->current_program->on_hardware = 0;
1423 i915->current_program->params_uptodate = 0;
1426 i915->current_program = fp;
1429 if (!fp->translated)
1430 translate_program(fp);
1432 FALLBACK(&i915->intel, I915_FALLBACK_PROGRAM, fp->error);
1436 i915ValidateFragmentProgram(struct i915_context *i915)
1438 struct gl_context *ctx = &i915->intel.ctx;
1439 struct intel_context *intel = intel_context(ctx);
1440 TNLcontext *tnl = TNL_CONTEXT(ctx);
1441 struct vertex_buffer *VB = &tnl->vb;
1443 struct i915_fragment_program *p =
1444 (struct i915_fragment_program *) ctx->FragmentProgram._Current;
1446 const GLbitfield64 inputsRead = p->FragProg.Base.InputsRead;
1447 GLuint s4 = i915->state.Ctx[I915_CTXREG_LIS4] & ~S4_VFMT_MASK;
1448 GLuint s2 = S2_TEXCOORD_NONE;
1453 VB->AttribPtr[VERT_ATTRIB_POS] = VB->NdcPtr;
1456 translate_program(p);
1458 intel->vertex_attr_count = 0;
1459 intel->wpos_offset = 0;
1460 intel->coloroffset = 0;
1461 intel->specoffset = 0;
1463 if (inputsRead & VARYING_BITS_TEX_ANY || p->wpos_tex != -1) {
1464 EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_4F_VIEWPORT, S4_VFMT_XYZW, 16);
1467 EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_3F_VIEWPORT, S4_VFMT_XYZ, 12);
1470 /* Handle gl_PointSize builtin var here */
1471 if (ctx->Point._Attenuated || ctx->VertexProgram.PointSizeEnabled)
1472 EMIT_ATTR(_TNL_ATTRIB_POINTSIZE, EMIT_1F, S4_VFMT_POINT_WIDTH, 4);
1474 if (inputsRead & VARYING_BIT_COL0) {
1475 intel->coloroffset = offset / 4;
1476 EMIT_ATTR(_TNL_ATTRIB_COLOR0, EMIT_4UB_4F_BGRA, S4_VFMT_COLOR, 4);
1479 if (inputsRead & VARYING_BIT_COL1) {
1480 intel->specoffset = offset / 4;
1481 EMIT_ATTR(_TNL_ATTRIB_COLOR1, EMIT_4UB_4F_BGRA, S4_VFMT_SPEC_FOG, 4);
1484 if ((inputsRead & VARYING_BIT_FOGC)) {
1485 EMIT_ATTR(_TNL_ATTRIB_FOG, EMIT_1F, S4_VFMT_FOG_PARAM, 4);
1488 for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
1489 if (inputsRead & VARYING_BIT_TEX(i)) {
1490 int unit = get_texcoord_mapping(p, i | TEXCOORD_TEX);
1491 int sz = VB->AttribPtr[_TNL_ATTRIB_TEX0 + i]->size;
1493 s2 &= ~S2_TEXCOORD_FMT(unit, S2_TEXCOORD_FMT0_MASK);
1494 s2 |= S2_TEXCOORD_FMT(unit, SZ_TO_HW(sz));
1496 EMIT_ATTR(_TNL_ATTRIB_TEX0 + i, EMIT_SZ(sz), 0, sz * 4);
1498 if (inputsRead & VARYING_BIT_VAR(i)) {
1499 int unit = get_texcoord_mapping(p, i | TEXCOORD_VAR);
1500 int sz = VB->AttribPtr[_TNL_ATTRIB_GENERIC0 + i]->size;
1502 s2 &= ~S2_TEXCOORD_FMT(unit, S2_TEXCOORD_FMT0_MASK);
1503 s2 |= S2_TEXCOORD_FMT(unit, SZ_TO_HW(sz));
1505 EMIT_ATTR(_TNL_ATTRIB_GENERIC0 + i, EMIT_SZ(sz), 0, sz * 4);
1507 if (i == p->wpos_tex) {
1508 int wpos_size = 4 * sizeof(float);
1509 /* If WPOS is required, duplicate the XYZ position data in an
1510 * unused texture coordinate:
1512 s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1513 s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(wpos_size));
1515 intel->wpos_offset = offset;
1516 EMIT_PAD(wpos_size);
1520 if (s2 != i915->state.Ctx[I915_CTXREG_LIS2] ||
1521 s4 != i915->state.Ctx[I915_CTXREG_LIS4]) {
1524 I915_STATECHANGE(i915, I915_UPLOAD_CTX);
1526 /* Must do this *after* statechange, so as not to affect
1527 * buffered vertices reliant on the old state:
1529 intel->vertex_size = _tnl_install_attrs(&intel->ctx,
1530 intel->vertex_attrs,
1531 intel->vertex_attr_count,
1532 intel->ViewportMatrix.m, 0);
1534 assert(intel->prim.current_offset == intel->prim.start_offset);
1535 intel->prim.start_offset = (intel->prim.current_offset + intel->vertex_size-1) / intel->vertex_size * intel->vertex_size;
1536 intel->prim.current_offset = intel->prim.start_offset;
1538 intel->vertex_size >>= 2;
1540 i915->state.Ctx[I915_CTXREG_LIS2] = s2;
1541 i915->state.Ctx[I915_CTXREG_LIS4] = s4;
1543 k = intel->vtbl.check_vertex_size(intel, intel->vertex_size);
1547 if (!p->params_uptodate)
1550 if (!p->on_hardware)
1551 i915_upload_program(i915, p);
1553 if (INTEL_DEBUG & DEBUG_WM) {
1555 i915_disassemble_program(i915->state.Program, i915->state.ProgramSize);
1560 i915InitFragProgFuncs(struct dd_function_table *functions)
1562 functions->BindProgram = i915BindProgram;
1563 functions->NewProgram = i915NewProgram;
1564 functions->DeleteProgram = i915DeleteProgram;
1565 functions->IsProgramNative = i915IsProgramNative;
1566 functions->ProgramStringNotify = i915ProgramStringNotify;
1567 functions->SamplerUniformChange = i915SamplerUniformChange;