1/**********************************************************
2 * Copyright 2008-2009 VMware, Inc.  All rights reserved.
3 *
4 * Permission is hereby granted, free of charge, to any person
5 * obtaining a copy of this software and associated documentation
6 * files (the "Software"), to deal in the Software without
7 * restriction, including without limitation the rights to use, copy,
8 * modify, merge, publish, distribute, sublicense, and/or sell copies
9 * of the Software, and to permit persons to whom the Software is
10 * furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice shall be
13 * included in all copies or substantial portions of the Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
16 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
17 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
18 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
19 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
20 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
21 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22 * SOFTWARE.
23 *
24 **********************************************************/
25
26
27#include "pipe/p_shader_tokens.h"
28#include "tgsi/tgsi_dump.h"
29#include "tgsi/tgsi_parse.h"
30#include "util/u_memory.h"
31#include "util/u_math.h"
32#include "util/u_pstipple.h"
33
34#include "svga_tgsi_emit.h"
35#include "svga_context.h"
36
37
38static boolean emit_vs_postamble( struct svga_shader_emitter *emit );
39static boolean emit_ps_postamble( struct svga_shader_emitter *emit );
40
41
42static SVGA3dShaderOpCodeType
43translate_opcode(enum tgsi_opcode opcode)
44{
45   switch (opcode) {
46   case TGSI_OPCODE_ADD:        return SVGA3DOP_ADD;
47   case TGSI_OPCODE_DP3:        return SVGA3DOP_DP3;
48   case TGSI_OPCODE_DP4:        return SVGA3DOP_DP4;
49   case TGSI_OPCODE_FRC:        return SVGA3DOP_FRC;
50   case TGSI_OPCODE_MAD:        return SVGA3DOP_MAD;
51   case TGSI_OPCODE_MAX:        return SVGA3DOP_MAX;
52   case TGSI_OPCODE_MIN:        return SVGA3DOP_MIN;
53   case TGSI_OPCODE_MOV:        return SVGA3DOP_MOV;
54   case TGSI_OPCODE_MUL:        return SVGA3DOP_MUL;
55   case TGSI_OPCODE_NOP:        return SVGA3DOP_NOP;
56   default:
57      assert(!"svga: unexpected opcode in translate_opcode()");
58      return SVGA3DOP_LAST_INST;
59   }
60}
61
62
63static SVGA3dShaderRegType
64translate_file(enum tgsi_file_type file)
65{
66   switch (file) {
67   case TGSI_FILE_TEMPORARY: return SVGA3DREG_TEMP;
68   case TGSI_FILE_INPUT:     return SVGA3DREG_INPUT;
69   case TGSI_FILE_OUTPUT:    return SVGA3DREG_OUTPUT; /* VS3.0+ only */
70   case TGSI_FILE_IMMEDIATE: return SVGA3DREG_CONST;
71   case TGSI_FILE_CONSTANT:  return SVGA3DREG_CONST;
72   case TGSI_FILE_SAMPLER:   return SVGA3DREG_SAMPLER;
73   case TGSI_FILE_ADDRESS:   return SVGA3DREG_ADDR;
74   default:
75      assert(!"svga: unexpected register file in translate_file()");
76      return SVGA3DREG_TEMP;
77   }
78}
79
80
81/**
82 * Translate a TGSI destination register to an SVGA3DShaderDestToken.
83 * \param insn  the TGSI instruction
84 * \param idx  which TGSI dest register to translate (usually (always?) zero)
85 */
86static SVGA3dShaderDestToken
87translate_dst_register( struct svga_shader_emitter *emit,
88                        const struct tgsi_full_instruction *insn,
89                        unsigned idx )
90{
91   const struct tgsi_full_dst_register *reg = &insn->Dst[idx];
92   SVGA3dShaderDestToken dest;
93
94   switch (reg->Register.File) {
95   case TGSI_FILE_OUTPUT:
96      /* Output registers encode semantic information in their name.
97       * Need to lookup a table built at decl time:
98       */
99      dest = emit->output_map[reg->Register.Index];
100      emit->num_output_writes++;
101      break;
102
103   default:
104      {
105         unsigned index = reg->Register.Index;
106         assert(index < SVGA3D_TEMPREG_MAX);
107         index = MIN2(index, SVGA3D_TEMPREG_MAX - 1);
108         dest = dst_register(translate_file(reg->Register.File), index);
109      }
110      break;
111   }
112
113   if (reg->Register.Indirect) {
114      debug_warning("Indirect indexing of dest registers is not supported!\n");
115   }
116
117   dest.mask = reg->Register.WriteMask;
118   assert(dest.mask);
119
120   if (insn->Instruction.Saturate)
121      dest.dstMod = SVGA3DDSTMOD_SATURATE;
122
123   return dest;
124}
125
126
127/**
128 * Apply a swizzle to a src_register, returning a new src_register
129 * Ex: swizzle(SRC.ZZYY, SWIZZLE_Z, SWIZZLE_W, SWIZZLE_X, SWIZZLE_Y)
130 * would return SRC.YYZZ
131 */
132static struct src_register
133swizzle(struct src_register src,
134        unsigned x, unsigned y, unsigned z, unsigned w)
135{
136   assert(x < 4);
137   assert(y < 4);
138   assert(z < 4);
139   assert(w < 4);
140   x = (src.base.swizzle >> (x * 2)) & 0x3;
141   y = (src.base.swizzle >> (y * 2)) & 0x3;
142   z = (src.base.swizzle >> (z * 2)) & 0x3;
143   w = (src.base.swizzle >> (w * 2)) & 0x3;
144
145   src.base.swizzle = TRANSLATE_SWIZZLE(x, y, z, w);
146
147   return src;
148}
149
150
151/**
152 * Apply a "scalar" swizzle to a src_register returning a new
153 * src_register where all the swizzle terms are the same.
154 * Ex: scalar(SRC.WZYX, SWIZZLE_Y) would return SRC.ZZZZ
155 */
156static struct src_register
157scalar(struct src_register src, unsigned comp)
158{
159   assert(comp < 4);
160   return swizzle( src, comp, comp, comp, comp );
161}
162
163
164static boolean
165svga_arl_needs_adjustment( const struct svga_shader_emitter *emit )
166{
167   unsigned i;
168
169   for (i = 0; i < emit->num_arl_consts; ++i) {
170      if (emit->arl_consts[i].arl_num == emit->current_arl)
171         return TRUE;
172   }
173   return FALSE;
174}
175
176
177static int
178svga_arl_adjustment( const struct svga_shader_emitter *emit )
179{
180   unsigned i;
181
182   for (i = 0; i < emit->num_arl_consts; ++i) {
183      if (emit->arl_consts[i].arl_num == emit->current_arl)
184         return emit->arl_consts[i].number;
185   }
186   return 0;
187}
188
189
190/**
191 * Translate a TGSI src register to a src_register.
192 */
193static struct src_register
194translate_src_register( const struct svga_shader_emitter *emit,
195                        const struct tgsi_full_src_register *reg )
196{
197   struct src_register src;
198
199   switch (reg->Register.File) {
200   case TGSI_FILE_INPUT:
201      /* Input registers are referred to by their semantic name rather
202       * than by index.  Use the mapping build up from the decls:
203       */
204      src = emit->input_map[reg->Register.Index];
205      break;
206
207   case TGSI_FILE_IMMEDIATE:
208      /* Immediates are appended after TGSI constants in the D3D
209       * constant buffer.
210       */
211      src = src_register( translate_file( reg->Register.File ),
212                          reg->Register.Index + emit->imm_start );
213      break;
214
215   default:
216      src = src_register( translate_file( reg->Register.File ),
217                          reg->Register.Index );
218      break;
219   }
220
221   /* Indirect addressing.
222    */
223   if (reg->Register.Indirect) {
224      if (emit->unit == PIPE_SHADER_FRAGMENT) {
225         /* Pixel shaders have only loop registers for relative
226          * addressing into inputs. Ignore the redundant address
227          * register, the contents of aL should be in sync with it.
228          */
229         if (reg->Register.File == TGSI_FILE_INPUT) {
230            src.base.relAddr = 1;
231            src.indirect = src_token(SVGA3DREG_LOOP, 0);
232         }
233      }
234      else {
235         /* Constant buffers only.
236          */
237         if (reg->Register.File == TGSI_FILE_CONSTANT) {
238            /* we shift the offset towards the minimum */
239            if (svga_arl_needs_adjustment( emit )) {
240               src.base.num -= svga_arl_adjustment( emit );
241            }
242            src.base.relAddr = 1;
243
244            /* Not really sure what should go in the second token:
245             */
246            src.indirect = src_token( SVGA3DREG_ADDR,
247                                      reg->Indirect.Index );
248
249            src.indirect.swizzle = SWIZZLE_XXXX;
250         }
251      }
252   }
253
254   src = swizzle( src,
255                  reg->Register.SwizzleX,
256                  reg->Register.SwizzleY,
257                  reg->Register.SwizzleZ,
258                  reg->Register.SwizzleW );
259
260   /* src.mod isn't a bitfield, unfortunately */
261   if (reg->Register.Absolute) {
262      if (reg->Register.Negate)
263         src.base.srcMod = SVGA3DSRCMOD_ABSNEG;
264      else
265         src.base.srcMod = SVGA3DSRCMOD_ABS;
266   }
267   else {
268      if (reg->Register.Negate)
269         src.base.srcMod = SVGA3DSRCMOD_NEG;
270      else
271         src.base.srcMod = SVGA3DSRCMOD_NONE;
272   }
273
274   return src;
275}
276
277
278/*
279 * Get a temporary register.
280 * Note: if we exceed the temporary register limit we just use
281 * register SVGA3D_TEMPREG_MAX - 1.
282 */
283static SVGA3dShaderDestToken
284get_temp( struct svga_shader_emitter *emit )
285{
286   int i = emit->nr_hw_temp + emit->internal_temp_count++;
287   if (i >= SVGA3D_TEMPREG_MAX) {
288      debug_warn_once("svga: Too many temporary registers used in shader\n");
289      i = SVGA3D_TEMPREG_MAX - 1;
290   }
291   return dst_register( SVGA3DREG_TEMP, i );
292}
293
294
295/**
296 * Release a single temp.  Currently only effective if it was the last
297 * allocated temp, otherwise release will be delayed until the next
298 * call to reset_temp_regs().
299 */
300static void
301release_temp( struct svga_shader_emitter *emit,
302              SVGA3dShaderDestToken temp )
303{
304   if (temp.num == emit->internal_temp_count - 1)
305      emit->internal_temp_count--;
306}
307
308
309/**
310 * Release all temps.
311 */
312static void
313reset_temp_regs(struct svga_shader_emitter *emit)
314{
315   emit->internal_temp_count = 0;
316}
317
318
319/** Emit bytecode for a src_register */
320static boolean
321emit_src(struct svga_shader_emitter *emit, const struct src_register src)
322{
323   if (src.base.relAddr) {
324      assert(src.base.reserved0);
325      assert(src.indirect.reserved0);
326      return (svga_shader_emit_dword( emit, src.base.value ) &&
327              svga_shader_emit_dword( emit, src.indirect.value ));
328   }
329   else {
330      assert(src.base.reserved0);
331      return svga_shader_emit_dword( emit, src.base.value );
332   }
333}
334
335
336/** Emit bytecode for a dst_register */
337static boolean
338emit_dst(struct svga_shader_emitter *emit, SVGA3dShaderDestToken dest)
339{
340   assert(dest.reserved0);
341   assert(dest.mask);
342   return svga_shader_emit_dword( emit, dest.value );
343}
344
345
346/** Emit bytecode for a 1-operand instruction */
347static boolean
348emit_op1(struct svga_shader_emitter *emit,
349         SVGA3dShaderInstToken inst,
350         SVGA3dShaderDestToken dest,
351         struct src_register src0)
352{
353   return (emit_instruction(emit, inst) &&
354           emit_dst(emit, dest) &&
355           emit_src(emit, src0));
356}
357
358
359/** Emit bytecode for a 2-operand instruction */
360static boolean
361emit_op2(struct svga_shader_emitter *emit,
362         SVGA3dShaderInstToken inst,
363         SVGA3dShaderDestToken dest,
364         struct src_register src0,
365         struct src_register src1)
366{
367   return (emit_instruction(emit, inst) &&
368           emit_dst(emit, dest) &&
369           emit_src(emit, src0) &&
370           emit_src(emit, src1));
371}
372
373
374/** Emit bytecode for a 3-operand instruction */
375static boolean
376emit_op3(struct svga_shader_emitter *emit,
377         SVGA3dShaderInstToken inst,
378         SVGA3dShaderDestToken dest,
379         struct src_register src0,
380         struct src_register src1,
381         struct src_register src2)
382{
383   return (emit_instruction(emit, inst) &&
384           emit_dst(emit, dest) &&
385           emit_src(emit, src0) &&
386           emit_src(emit, src1) &&
387           emit_src(emit, src2));
388}
389
390
391/** Emit bytecode for a 4-operand instruction */
392static boolean
393emit_op4(struct svga_shader_emitter *emit,
394         SVGA3dShaderInstToken inst,
395         SVGA3dShaderDestToken dest,
396         struct src_register src0,
397         struct src_register src1,
398         struct src_register src2,
399         struct src_register src3)
400{
401   return (emit_instruction(emit, inst) &&
402           emit_dst(emit, dest) &&
403           emit_src(emit, src0) &&
404           emit_src(emit, src1) &&
405           emit_src(emit, src2) &&
406           emit_src(emit, src3));
407}
408
409
410/**
411 * Apply the absolute value modifier to the given src_register, returning
412 * a new src_register.
413 */
414static struct src_register
415absolute(struct src_register src)
416{
417   src.base.srcMod = SVGA3DSRCMOD_ABS;
418   return src;
419}
420
421
422/**
423 * Apply the negation modifier to the given src_register, returning
424 * a new src_register.
425 */
426static struct src_register
427negate(struct src_register src)
428{
429   switch (src.base.srcMod) {
430   case SVGA3DSRCMOD_ABS:
431      src.base.srcMod = SVGA3DSRCMOD_ABSNEG;
432      break;
433   case SVGA3DSRCMOD_ABSNEG:
434      src.base.srcMod = SVGA3DSRCMOD_ABS;
435      break;
436   case SVGA3DSRCMOD_NEG:
437      src.base.srcMod = SVGA3DSRCMOD_NONE;
438      break;
439   case SVGA3DSRCMOD_NONE:
440      src.base.srcMod = SVGA3DSRCMOD_NEG;
441      break;
442   }
443   return src;
444}
445
446
447
448/* Replace the src with the temporary specified in the dst, but copying
449 * only the necessary channels, and preserving the original swizzle (which is
450 * important given that several opcodes have constraints in the allowed
451 * swizzles).
452 */
453static boolean
454emit_repl(struct svga_shader_emitter *emit,
455          SVGA3dShaderDestToken dst,
456          struct src_register *src0)
457{
458   unsigned src0_swizzle;
459   unsigned chan;
460
461   assert(SVGA3dShaderGetRegType(dst.value) == SVGA3DREG_TEMP);
462
463   src0_swizzle = src0->base.swizzle;
464
465   dst.mask = 0;
466   for (chan = 0; chan < 4; ++chan) {
467      unsigned swizzle = (src0_swizzle >> (chan *2)) & 0x3;
468      dst.mask |= 1 << swizzle;
469   }
470   assert(dst.mask);
471
472   src0->base.swizzle = SVGA3DSWIZZLE_NONE;
473
474   if (!emit_op1( emit, inst_token( SVGA3DOP_MOV ), dst, *src0 ))
475      return FALSE;
476
477   *src0 = src( dst );
478   src0->base.swizzle = src0_swizzle;
479
480   return TRUE;
481}
482
483
484/**
485 * Submit/emit an instruction with zero operands.
486 */
487static boolean
488submit_op0(struct svga_shader_emitter *emit,
489           SVGA3dShaderInstToken inst,
490           SVGA3dShaderDestToken dest)
491{
492   return (emit_instruction( emit, inst ) &&
493           emit_dst( emit, dest ));
494}
495
496
497/**
498 * Submit/emit an instruction with one operand.
499 */
500static boolean
501submit_op1(struct svga_shader_emitter *emit,
502           SVGA3dShaderInstToken inst,
503           SVGA3dShaderDestToken dest,
504           struct src_register src0)
505{
506   return emit_op1( emit, inst, dest, src0 );
507}
508
509
510/**
511 * Submit/emit an instruction with two operands.
512 *
513 * SVGA shaders may not refer to >1 constant register in a single
514 * instruction.  This function checks for that usage and inserts a
515 * move to temporary if detected.
516 *
517 * The same applies to input registers -- at most a single input
518 * register may be read by any instruction.
519 */
520static boolean
521submit_op2(struct svga_shader_emitter *emit,
522           SVGA3dShaderInstToken inst,
523           SVGA3dShaderDestToken dest,
524           struct src_register src0,
525           struct src_register src1)
526{
527   SVGA3dShaderDestToken temp;
528   SVGA3dShaderRegType type0, type1;
529   boolean need_temp = FALSE;
530
531   temp.value = 0;
532   type0 = SVGA3dShaderGetRegType( src0.base.value );
533   type1 = SVGA3dShaderGetRegType( src1.base.value );
534
535   if (type0 == SVGA3DREG_CONST &&
536       type1 == SVGA3DREG_CONST &&
537       src0.base.num != src1.base.num)
538      need_temp = TRUE;
539
540   if (type0 == SVGA3DREG_INPUT &&
541       type1 == SVGA3DREG_INPUT &&
542       src0.base.num != src1.base.num)
543      need_temp = TRUE;
544
545   if (need_temp) {
546      temp = get_temp( emit );
547
548      if (!emit_repl( emit, temp, &src0 ))
549         return FALSE;
550   }
551
552   if (!emit_op2( emit, inst, dest, src0, src1 ))
553      return FALSE;
554
555   if (need_temp)
556      release_temp( emit, temp );
557
558   return TRUE;
559}
560
561
562/**
563 * Submit/emit an instruction with three operands.
564 *
565 * SVGA shaders may not refer to >1 constant register in a single
566 * instruction.  This function checks for that usage and inserts a
567 * move to temporary if detected.
568 */
569static boolean
570submit_op3(struct svga_shader_emitter *emit,
571           SVGA3dShaderInstToken inst,
572           SVGA3dShaderDestToken dest,
573           struct src_register src0,
574           struct src_register src1,
575           struct src_register src2)
576{
577   SVGA3dShaderDestToken temp0;
578   SVGA3dShaderDestToken temp1;
579   boolean need_temp0 = FALSE;
580   boolean need_temp1 = FALSE;
581   SVGA3dShaderRegType type0, type1, type2;
582
583   temp0.value = 0;
584   temp1.value = 0;
585   type0 = SVGA3dShaderGetRegType( src0.base.value );
586   type1 = SVGA3dShaderGetRegType( src1.base.value );
587   type2 = SVGA3dShaderGetRegType( src2.base.value );
588
589   if (inst.op != SVGA3DOP_SINCOS) {
590      if (type0 == SVGA3DREG_CONST &&
591          ((type1 == SVGA3DREG_CONST && src0.base.num != src1.base.num) ||
592           (type2 == SVGA3DREG_CONST && src0.base.num != src2.base.num)))
593         need_temp0 = TRUE;
594
595      if (type1 == SVGA3DREG_CONST &&
596          (type2 == SVGA3DREG_CONST && src1.base.num != src2.base.num))
597         need_temp1 = TRUE;
598   }
599
600   if (type0 == SVGA3DREG_INPUT &&
601       ((type1 == SVGA3DREG_INPUT && src0.base.num != src1.base.num) ||
602        (type2 == SVGA3DREG_INPUT && src0.base.num != src2.base.num)))
603      need_temp0 = TRUE;
604
605   if (type1 == SVGA3DREG_INPUT &&
606       (type2 == SVGA3DREG_INPUT && src1.base.num != src2.base.num))
607      need_temp1 = TRUE;
608
609   if (need_temp0) {
610      temp0 = get_temp( emit );
611
612      if (!emit_repl( emit, temp0, &src0 ))
613         return FALSE;
614   }
615
616   if (need_temp1) {
617      temp1 = get_temp( emit );
618
619      if (!emit_repl( emit, temp1, &src1 ))
620         return FALSE;
621   }
622
623   if (!emit_op3( emit, inst, dest, src0, src1, src2 ))
624      return FALSE;
625
626   if (need_temp1)
627      release_temp( emit, temp1 );
628   if (need_temp0)
629      release_temp( emit, temp0 );
630   return TRUE;
631}
632
633
634/**
635 * Submit/emit an instruction with four operands.
636 *
637 * SVGA shaders may not refer to >1 constant register in a single
638 * instruction.  This function checks for that usage and inserts a
639 * move to temporary if detected.
640 */
641static boolean
642submit_op4(struct svga_shader_emitter *emit,
643           SVGA3dShaderInstToken inst,
644           SVGA3dShaderDestToken dest,
645           struct src_register src0,
646           struct src_register src1,
647           struct src_register src2,
648           struct src_register src3)
649{
650   SVGA3dShaderDestToken temp0;
651   SVGA3dShaderDestToken temp3;
652   boolean need_temp0 = FALSE;
653   boolean need_temp3 = FALSE;
654   SVGA3dShaderRegType type0, type1, type2, type3;
655
656   temp0.value = 0;
657   temp3.value = 0;
658   type0 = SVGA3dShaderGetRegType( src0.base.value );
659   type1 = SVGA3dShaderGetRegType( src1.base.value );
660   type2 = SVGA3dShaderGetRegType( src2.base.value );
661   type3 = SVGA3dShaderGetRegType( src2.base.value );
662
663   /* Make life a little easier - this is only used by the TXD
664    * instruction which is guaranteed not to have a constant/input reg
665    * in one slot at least:
666    */
667   assert(type1 == SVGA3DREG_SAMPLER);
668   (void) type1;
669
670   if (type0 == SVGA3DREG_CONST &&
671       ((type3 == SVGA3DREG_CONST && src0.base.num != src3.base.num) ||
672        (type2 == SVGA3DREG_CONST && src0.base.num != src2.base.num)))
673      need_temp0 = TRUE;
674
675   if (type3 == SVGA3DREG_CONST &&
676       (type2 == SVGA3DREG_CONST && src3.base.num != src2.base.num))
677      need_temp3 = TRUE;
678
679   if (type0 == SVGA3DREG_INPUT &&
680       ((type3 == SVGA3DREG_INPUT && src0.base.num != src3.base.num) ||
681        (type2 == SVGA3DREG_INPUT && src0.base.num != src2.base.num)))
682      need_temp0 = TRUE;
683
684   if (type3 == SVGA3DREG_INPUT &&
685       (type2 == SVGA3DREG_INPUT && src3.base.num != src2.base.num))
686      need_temp3 = TRUE;
687
688   if (need_temp0) {
689      temp0 = get_temp( emit );
690
691      if (!emit_repl( emit, temp0, &src0 ))
692         return FALSE;
693   }
694
695   if (need_temp3) {
696      temp3 = get_temp( emit );
697
698      if (!emit_repl( emit, temp3, &src3 ))
699         return FALSE;
700   }
701
702   if (!emit_op4( emit, inst, dest, src0, src1, src2, src3 ))
703      return FALSE;
704
705   if (need_temp3)
706      release_temp( emit, temp3 );
707   if (need_temp0)
708      release_temp( emit, temp0 );
709   return TRUE;
710}
711
712
713/**
714 * Do the src and dest registers refer to the same register?
715 */
716static boolean
717alias_src_dst(struct src_register src,
718              SVGA3dShaderDestToken dst)
719{
720   if (src.base.num != dst.num)
721      return FALSE;
722
723   if (SVGA3dShaderGetRegType(dst.value) !=
724       SVGA3dShaderGetRegType(src.base.value))
725      return FALSE;
726
727   return TRUE;
728}
729
730
731/**
732 * Helper for emitting SVGA immediate values using the SVGA3DOP_DEF[I]
733 * instructions.
734 */
735static boolean
736emit_def_const(struct svga_shader_emitter *emit,
737               SVGA3dShaderConstType type,
738               unsigned idx, float a, float b, float c, float d)
739{
740   SVGA3DOpDefArgs def;
741   SVGA3dShaderInstToken opcode;
742
743   switch (type) {
744   case SVGA3D_CONST_TYPE_FLOAT:
745      opcode = inst_token( SVGA3DOP_DEF );
746      def.dst = dst_register( SVGA3DREG_CONST, idx );
747      def.constValues[0] = a;
748      def.constValues[1] = b;
749      def.constValues[2] = c;
750      def.constValues[3] = d;
751      break;
752   case SVGA3D_CONST_TYPE_INT:
753      opcode = inst_token( SVGA3DOP_DEFI );
754      def.dst = dst_register( SVGA3DREG_CONSTINT, idx );
755      def.constIValues[0] = (int)a;
756      def.constIValues[1] = (int)b;
757      def.constIValues[2] = (int)c;
758      def.constIValues[3] = (int)d;
759      break;
760   default:
761      assert(0);
762      opcode = inst_token( SVGA3DOP_NOP );
763      break;
764   }
765
766   if (!emit_instruction(emit, opcode) ||
767       !svga_shader_emit_dwords( emit, def.values, ARRAY_SIZE(def.values)))
768      return FALSE;
769
770   return TRUE;
771}
772
773
774static boolean
775create_loop_const( struct svga_shader_emitter *emit )
776{
777   unsigned idx = emit->nr_hw_int_const++;
778
779   if (!emit_def_const( emit, SVGA3D_CONST_TYPE_INT, idx,
780                        255, /* iteration count */
781                        0, /* initial value */
782                        1, /* step size */
783                        0 /* not used, must be 0 */))
784      return FALSE;
785
786   emit->loop_const_idx = idx;
787   emit->created_loop_const = TRUE;
788
789   return TRUE;
790}
791
792static boolean
793create_arl_consts( struct svga_shader_emitter *emit )
794{
795   int i;
796
797   for (i = 0; i < emit->num_arl_consts; i += 4) {
798      int j;
799      unsigned idx = emit->nr_hw_float_const++;
800      float vals[4];
801      for (j = 0; j < 4 && (j + i) < emit->num_arl_consts; ++j) {
802         vals[j] = (float) emit->arl_consts[i + j].number;
803         emit->arl_consts[i + j].idx = idx;
804         switch (j) {
805         case 0:
806            emit->arl_consts[i + 0].swizzle = TGSI_SWIZZLE_X;
807            break;
808         case 1:
809            emit->arl_consts[i + 0].swizzle = TGSI_SWIZZLE_Y;
810            break;
811         case 2:
812            emit->arl_consts[i + 0].swizzle = TGSI_SWIZZLE_Z;
813            break;
814         case 3:
815            emit->arl_consts[i + 0].swizzle = TGSI_SWIZZLE_W;
816            break;
817         }
818      }
819      while (j < 4)
820         vals[j++] = 0;
821
822      if (!emit_def_const( emit, SVGA3D_CONST_TYPE_FLOAT, idx,
823                           vals[0], vals[1],
824                           vals[2], vals[3]))
825         return FALSE;
826   }
827
828   return TRUE;
829}
830
831
832/**
833 * Return the register which holds the pixel shaders front/back-
834 * facing value.
835 */
836static struct src_register
837get_vface( struct svga_shader_emitter *emit )
838{
839   assert(emit->emitted_vface);
840   return src_register(SVGA3DREG_MISCTYPE, SVGA3DMISCREG_FACE);
841}
842
843
844/**
845 * Create/emit a "common" constant with values {0, 0.5, -1, 1}.
846 * We can swizzle this to produce other useful constants such as
847 * {0, 0, 0, 0}, {1, 1, 1, 1}, etc.
848 */
849static boolean
850create_common_immediate( struct svga_shader_emitter *emit )
851{
852   unsigned idx = emit->nr_hw_float_const++;
853
854   /* Emit the constant (0, 0.5, -1, 1) and use swizzling to generate
855    * other useful vectors.
856    */
857   if (!emit_def_const( emit, SVGA3D_CONST_TYPE_FLOAT,
858                        idx, 0.0f, 0.5f, -1.0f, 1.0f ))
859      return FALSE;
860   emit->common_immediate_idx[0] = idx;
861   idx++;
862
863   /* Emit constant {2, 0, 0, 0} (only the 2 is used for now) */
864   if (emit->key.vs.adjust_attrib_range) {
865      if (!emit_def_const( emit, SVGA3D_CONST_TYPE_FLOAT,
866                           idx, 2.0f, 0.0f, 0.0f, 0.0f ))
867         return FALSE;
868      emit->common_immediate_idx[1] = idx;
869   }
870   else {
871      emit->common_immediate_idx[1] = -1;
872   }
873
874   emit->created_common_immediate = TRUE;
875
876   return TRUE;
877}
878
879
880/**
881 * Return swizzle/position for the given value in the "common" immediate.
882 */
883static inline unsigned
884common_immediate_swizzle(float value)
885{
886   if (value == 0.0f)
887      return TGSI_SWIZZLE_X;
888   else if (value == 0.5f)
889      return TGSI_SWIZZLE_Y;
890   else if (value == -1.0f)
891      return TGSI_SWIZZLE_Z;
892   else if (value == 1.0f)
893      return TGSI_SWIZZLE_W;
894   else {
895      assert(!"illegal value in common_immediate_swizzle");
896      return TGSI_SWIZZLE_X;
897   }
898}
899
900
901/**
902 * Returns an immediate reg where all the terms are either 0, 1, 2 or 0.5
903 */
904static struct src_register
905get_immediate(struct svga_shader_emitter *emit,
906              float x, float y, float z, float w)
907{
908   unsigned sx = common_immediate_swizzle(x);
909   unsigned sy = common_immediate_swizzle(y);
910   unsigned sz = common_immediate_swizzle(z);
911   unsigned sw = common_immediate_swizzle(w);
912   assert(emit->created_common_immediate);
913   assert(emit->common_immediate_idx[0] >= 0);
914   return swizzle(src_register(SVGA3DREG_CONST, emit->common_immediate_idx[0]),
915                  sx, sy, sz, sw);
916}
917
918
919/**
920 * returns {0, 0, 0, 0} immediate
921 */
922static struct src_register
923get_zero_immediate( struct svga_shader_emitter *emit )
924{
925   assert(emit->created_common_immediate);
926   assert(emit->common_immediate_idx[0] >= 0);
927   return swizzle(src_register( SVGA3DREG_CONST,
928                                emit->common_immediate_idx[0]),
929                  0, 0, 0, 0);
930}
931
932
933/**
934 * returns {1, 1, 1, 1} immediate
935 */
936static struct src_register
937get_one_immediate( struct svga_shader_emitter *emit )
938{
939   assert(emit->created_common_immediate);
940   assert(emit->common_immediate_idx[0] >= 0);
941   return swizzle(src_register( SVGA3DREG_CONST,
942                                emit->common_immediate_idx[0]),
943                  3, 3, 3, 3);
944}
945
946
947/**
948 * returns {0.5, 0.5, 0.5, 0.5} immediate
949 */
950static struct src_register
951get_half_immediate( struct svga_shader_emitter *emit )
952{
953   assert(emit->created_common_immediate);
954   assert(emit->common_immediate_idx[0] >= 0);
955   return swizzle(src_register(SVGA3DREG_CONST, emit->common_immediate_idx[0]),
956                  1, 1, 1, 1);
957}
958
959
960/**
961 * returns {2, 2, 2, 2} immediate
962 */
963static struct src_register
964get_two_immediate( struct svga_shader_emitter *emit )
965{
966   /* Note we use the second common immediate here */
967   assert(emit->created_common_immediate);
968   assert(emit->common_immediate_idx[1] >= 0);
969   return swizzle(src_register( SVGA3DREG_CONST,
970                                emit->common_immediate_idx[1]),
971                  0, 0, 0, 0);
972}
973
974
975/**
976 * returns the loop const
977 */
978static struct src_register
979get_loop_const( struct svga_shader_emitter *emit )
980{
981   assert(emit->created_loop_const);
982   assert(emit->loop_const_idx >= 0);
983   return src_register( SVGA3DREG_CONSTINT,
984                        emit->loop_const_idx );
985}
986
987
988static struct src_register
989get_fake_arl_const( struct svga_shader_emitter *emit )
990{
991   struct src_register reg;
992   int idx = 0, swizzle = 0, i;
993
994   for (i = 0; i < emit->num_arl_consts; ++ i) {
995      if (emit->arl_consts[i].arl_num == emit->current_arl) {
996         idx = emit->arl_consts[i].idx;
997         swizzle = emit->arl_consts[i].swizzle;
998      }
999   }
1000
1001   reg = src_register( SVGA3DREG_CONST, idx );
1002   return scalar(reg, swizzle);
1003}
1004
1005
1006/**
1007 * Return a register which holds the width and height of the texture
1008 * currently bound to the given sampler.
1009 */
1010static struct src_register
1011get_tex_dimensions( struct svga_shader_emitter *emit, int sampler_num )
1012{
1013   int idx;
1014   struct src_register reg;
1015
1016   /* the width/height indexes start right after constants */
1017   idx = emit->key.tex[sampler_num].width_height_idx +
1018         emit->info.file_max[TGSI_FILE_CONSTANT] + 1;
1019
1020   reg = src_register( SVGA3DREG_CONST, idx );
1021   return reg;
1022}
1023
1024
1025static boolean
1026emit_fake_arl(struct svga_shader_emitter *emit,
1027              const struct tgsi_full_instruction *insn)
1028{
1029   const struct src_register src0 =
1030      translate_src_register(emit, &insn->Src[0] );
1031   struct src_register src1 = get_fake_arl_const( emit );
1032   SVGA3dShaderDestToken dst = translate_dst_register( emit, insn, 0 );
1033   SVGA3dShaderDestToken tmp = get_temp( emit );
1034
1035   if (!submit_op1(emit, inst_token( SVGA3DOP_MOV ), tmp, src0))
1036      return FALSE;
1037
1038   if (!submit_op2( emit, inst_token( SVGA3DOP_ADD ), tmp, src( tmp ),
1039                    src1))
1040      return FALSE;
1041
1042   /* replicate the original swizzle */
1043   src1 = src(tmp);
1044   src1.base.swizzle = src0.base.swizzle;
1045
1046   return submit_op1( emit, inst_token( SVGA3DOP_MOVA ),
1047                      dst, src1 );
1048}
1049
1050
1051static boolean
1052emit_if(struct svga_shader_emitter *emit,
1053        const struct tgsi_full_instruction *insn)
1054{
1055   struct src_register src0 =
1056      translate_src_register(emit, &insn->Src[0]);
1057   struct src_register zero = get_zero_immediate(emit);
1058   SVGA3dShaderInstToken if_token = inst_token( SVGA3DOP_IFC );
1059
1060   if_token.control = SVGA3DOPCOMPC_NE;
1061
1062   if (SVGA3dShaderGetRegType(src0.base.value) == SVGA3DREG_CONST) {
1063      /*
1064       * Max different constant registers readable per IFC instruction is 1.
1065       */
1066      SVGA3dShaderDestToken tmp = get_temp( emit );
1067
1068      if (!submit_op1(emit, inst_token( SVGA3DOP_MOV ), tmp, src0))
1069         return FALSE;
1070
1071      src0 = scalar(src( tmp ), TGSI_SWIZZLE_X);
1072   }
1073
1074   emit->dynamic_branching_level++;
1075
1076   return (emit_instruction( emit, if_token ) &&
1077           emit_src( emit, src0 ) &&
1078           emit_src( emit, zero ) );
1079}
1080
1081
1082static boolean
1083emit_else(struct svga_shader_emitter *emit,
1084          const struct tgsi_full_instruction *insn)
1085{
1086   return emit_instruction(emit, inst_token(SVGA3DOP_ELSE));
1087}
1088
1089
1090static boolean
1091emit_endif(struct svga_shader_emitter *emit,
1092           const struct tgsi_full_instruction *insn)
1093{
1094   emit->dynamic_branching_level--;
1095
1096   return emit_instruction(emit, inst_token(SVGA3DOP_ENDIF));
1097}
1098
1099
1100/**
1101 * Translate the following TGSI FLR instruction.
1102 *    FLR  DST, SRC
1103 * To the following SVGA3D instruction sequence.
1104 *    FRC  TMP, SRC
1105 *    SUB  DST, SRC, TMP
1106 */
1107static boolean
1108emit_floor(struct svga_shader_emitter *emit,
1109           const struct tgsi_full_instruction *insn )
1110{
1111   SVGA3dShaderDestToken dst = translate_dst_register( emit, insn, 0 );
1112   const struct src_register src0 =
1113      translate_src_register(emit, &insn->Src[0] );
1114   SVGA3dShaderDestToken temp = get_temp( emit );
1115
1116   /* FRC  TMP, SRC */
1117   if (!submit_op1( emit, inst_token( SVGA3DOP_FRC ), temp, src0 ))
1118      return FALSE;
1119
1120   /* SUB  DST, SRC, TMP */
1121   if (!submit_op2( emit, inst_token( SVGA3DOP_ADD ), dst, src0,
1122                    negate( src( temp ) ) ))
1123      return FALSE;
1124
1125   return TRUE;
1126}
1127
1128
1129/**
1130 * Translate the following TGSI CEIL instruction.
1131 *    CEIL  DST, SRC
1132 * To the following SVGA3D instruction sequence.
1133 *    FRC  TMP, -SRC
1134 *    ADD  DST, SRC, TMP
1135 */
1136static boolean
1137emit_ceil(struct svga_shader_emitter *emit,
1138          const struct tgsi_full_instruction *insn)
1139{
1140   SVGA3dShaderDestToken dst = translate_dst_register(emit, insn, 0);
1141   const struct src_register src0 =
1142      translate_src_register(emit, &insn->Src[0]);
1143   SVGA3dShaderDestToken temp = get_temp(emit);
1144
1145   /* FRC  TMP, -SRC */
1146   if (!submit_op1(emit, inst_token(SVGA3DOP_FRC), temp, negate(src0)))
1147      return FALSE;
1148
1149   /* ADD DST, SRC, TMP */
1150   if (!submit_op2(emit, inst_token(SVGA3DOP_ADD), dst, src0, src(temp)))
1151      return FALSE;
1152
1153   return TRUE;
1154}
1155
1156
1157/**
1158 * Translate the following TGSI DIV instruction.
1159 *    DIV  DST.xy, SRC0, SRC1
1160 * To the following SVGA3D instruction sequence.
1161 *    RCP  TMP.x, SRC1.xxxx
1162 *    RCP  TMP.y, SRC1.yyyy
1163 *    MUL  DST.xy, SRC0, TMP
1164 */
1165static boolean
1166emit_div(struct svga_shader_emitter *emit,
1167         const struct tgsi_full_instruction *insn )
1168{
1169   SVGA3dShaderDestToken dst = translate_dst_register( emit, insn, 0 );
1170   const struct src_register src0 =
1171      translate_src_register(emit, &insn->Src[0] );
1172   const struct src_register src1 =
1173      translate_src_register(emit, &insn->Src[1] );
1174   SVGA3dShaderDestToken temp = get_temp( emit );
1175   unsigned i;
1176
1177   /* For each enabled element, perform a RCP instruction.  Note that
1178    * RCP is scalar in SVGA3D:
1179    */
1180   for (i = 0; i < 4; i++) {
1181      unsigned channel = 1 << i;
1182      if (dst.mask & channel) {
1183         /* RCP  TMP.?, SRC1.???? */
1184         if (!submit_op1( emit, inst_token( SVGA3DOP_RCP ),
1185                          writemask(temp, channel),
1186                          scalar(src1, i) ))
1187            return FALSE;
1188      }
1189   }
1190
1191   /* Vector mul:
1192    * MUL  DST, SRC0, TMP
1193    */
1194   if (!submit_op2( emit, inst_token( SVGA3DOP_MUL ), dst, src0,
1195                    src( temp ) ))
1196      return FALSE;
1197
1198   return TRUE;
1199}
1200
1201
1202/**
1203 * Translate the following TGSI DP2 instruction.
1204 *    DP2  DST, SRC1, SRC2
1205 * To the following SVGA3D instruction sequence.
1206 *    MUL  TMP, SRC1, SRC2
1207 *    ADD  DST, TMP.xxxx, TMP.yyyy
1208 */
1209static boolean
1210emit_dp2(struct svga_shader_emitter *emit,
1211         const struct tgsi_full_instruction *insn )
1212{
1213   SVGA3dShaderDestToken dst = translate_dst_register( emit, insn, 0 );
1214   const struct src_register src0 =
1215      translate_src_register(emit, &insn->Src[0]);
1216   const struct src_register src1 =
1217      translate_src_register(emit, &insn->Src[1]);
1218   SVGA3dShaderDestToken temp = get_temp( emit );
1219   struct src_register temp_src0, temp_src1;
1220
1221   /* MUL  TMP, SRC1, SRC2 */
1222   if (!submit_op2( emit, inst_token( SVGA3DOP_MUL ), temp, src0, src1 ))
1223      return FALSE;
1224
1225   temp_src0 = scalar(src( temp ), TGSI_SWIZZLE_X);
1226   temp_src1 = scalar(src( temp ), TGSI_SWIZZLE_Y);
1227
1228   /* ADD  DST, TMP.xxxx, TMP.yyyy */
1229   if (!submit_op2( emit, inst_token( SVGA3DOP_ADD ), dst,
1230                    temp_src0, temp_src1 ))
1231      return FALSE;
1232
1233   return TRUE;
1234}
1235
1236
1237/**
1238 * Sine / Cosine helper function.
1239 */
1240static boolean
1241do_emit_sincos(struct svga_shader_emitter *emit,
1242               SVGA3dShaderDestToken dst,
1243               struct src_register src0)
1244{
1245   src0 = scalar(src0, TGSI_SWIZZLE_X);
1246   return submit_op1(emit, inst_token(SVGA3DOP_SINCOS), dst, src0);
1247}
1248
1249
1250/**
1251 * Translate TGSI SIN instruction into:
1252 * SCS TMP SRC
1253 * MOV DST TMP.yyyy
1254 */
1255static boolean
1256emit_sin(struct svga_shader_emitter *emit,
1257         const struct tgsi_full_instruction *insn )
1258{
1259   SVGA3dShaderDestToken dst = translate_dst_register( emit, insn, 0 );
1260   struct src_register src0 =
1261      translate_src_register(emit, &insn->Src[0] );
1262   SVGA3dShaderDestToken temp = get_temp( emit );
1263
1264   /* SCS TMP SRC */
1265   if (!do_emit_sincos(emit, writemask(temp, TGSI_WRITEMASK_Y), src0))
1266      return FALSE;
1267
1268   src0 = scalar(src( temp ), TGSI_SWIZZLE_Y);
1269
1270   /* MOV DST TMP.yyyy */
1271   if (!submit_op1( emit, inst_token( SVGA3DOP_MOV ), dst, src0 ))
1272      return FALSE;
1273
1274   return TRUE;
1275}
1276
1277
1278/*
1279 * Translate TGSI COS instruction into:
1280 * SCS TMP SRC
1281 * MOV DST TMP.xxxx
1282 */
1283static boolean
1284emit_cos(struct svga_shader_emitter *emit,
1285         const struct tgsi_full_instruction *insn)
1286{
1287   SVGA3dShaderDestToken dst = translate_dst_register( emit, insn, 0 );
1288   struct src_register src0 =
1289      translate_src_register(emit, &insn->Src[0] );
1290   SVGA3dShaderDestToken temp = get_temp( emit );
1291
1292   /* SCS TMP SRC */
1293   if (!do_emit_sincos( emit, writemask(temp, TGSI_WRITEMASK_X), src0 ))
1294      return FALSE;
1295
1296   src0 = scalar(src( temp ), TGSI_SWIZZLE_X);
1297
1298   /* MOV DST TMP.xxxx */
1299   if (!submit_op1( emit, inst_token( SVGA3DOP_MOV ), dst, src0 ))
1300      return FALSE;
1301
1302   return TRUE;
1303}
1304
1305
1306/**
1307 * Translate/emit TGSI SSG (Set Sign: -1, 0, +1) instruction.
1308 */
1309static boolean
1310emit_ssg(struct svga_shader_emitter *emit,
1311         const struct tgsi_full_instruction *insn)
1312{
1313   SVGA3dShaderDestToken dst = translate_dst_register( emit, insn, 0 );
1314   struct src_register src0 =
1315      translate_src_register(emit, &insn->Src[0] );
1316   SVGA3dShaderDestToken temp0 = get_temp( emit );
1317   SVGA3dShaderDestToken temp1 = get_temp( emit );
1318   struct src_register zero, one;
1319
1320   if (emit->unit == PIPE_SHADER_VERTEX) {
1321      /* SGN  DST, SRC0, TMP0, TMP1 */
1322      return submit_op3( emit, inst_token( SVGA3DOP_SGN ), dst, src0,
1323                         src( temp0 ), src( temp1 ) );
1324   }
1325
1326   one = get_one_immediate(emit);
1327   zero = get_zero_immediate(emit);
1328
1329   /* CMP  TMP0, SRC0, one, zero */
1330   if (!submit_op3( emit, inst_token( SVGA3DOP_CMP ),
1331                    writemask( temp0, dst.mask ), src0, one, zero ))
1332      return FALSE;
1333
1334   /* CMP  TMP1, negate(SRC0), negate(one), zero */
1335   if (!submit_op3( emit, inst_token( SVGA3DOP_CMP ),
1336                    writemask( temp1, dst.mask ), negate( src0 ), negate( one ),
1337                    zero ))
1338      return FALSE;
1339
1340   /* ADD  DST, TMP0, TMP1 */
1341   return submit_op2( emit, inst_token( SVGA3DOP_ADD ), dst, src( temp0 ),
1342                      src( temp1 ) );
1343}
1344
1345
1346/**
1347 * Translate/emit the conditional discard instruction (discard if
1348 * any of X,Y,Z,W are negative).
1349 */
1350static boolean
1351emit_cond_discard(struct svga_shader_emitter *emit,
1352                  const struct tgsi_full_instruction *insn)
1353{
1354   const struct tgsi_full_src_register *reg = &insn->Src[0];
1355   struct src_register src0, srcIn;
1356   const boolean special = (reg->Register.Absolute ||
1357                            reg->Register.Negate ||
1358                            reg->Register.Indirect ||
1359                            reg->Register.SwizzleX != 0 ||
1360                            reg->Register.SwizzleY != 1 ||
1361                            reg->Register.SwizzleZ != 2 ||
1362                            reg->Register.File != TGSI_FILE_TEMPORARY);
1363   SVGA3dShaderDestToken temp;
1364
1365   src0 = srcIn = translate_src_register( emit, reg );
1366
1367   if (special) {
1368      /* need a temp reg */
1369      temp = get_temp( emit );
1370   }
1371
1372   if (special) {
1373      /* move the source into a temp register */
1374      submit_op1(emit, inst_token(SVGA3DOP_MOV), temp, src0);
1375
1376      src0 = src( temp );
1377   }
1378
1379   /* Do the discard by checking if any of the XYZW components are < 0.
1380    * Note that ps_2_0 and later take XYZW in consideration, while ps_1_x
1381    * only used XYZ.  The MSDN documentation about this is incorrect.
1382    */
1383   if (!submit_op0( emit, inst_token( SVGA3DOP_TEXKILL ), dst(src0) ))
1384      return FALSE;
1385
1386   return TRUE;
1387}
1388
1389
1390/**
1391 * Translate/emit the unconditional discard instruction (usually found inside
1392 * an IF/ELSE/ENDIF block).
1393 */
1394static boolean
1395emit_discard(struct svga_shader_emitter *emit,
1396             const struct tgsi_full_instruction *insn)
1397{
1398   SVGA3dShaderDestToken temp;
1399   struct src_register one = get_one_immediate(emit);
1400   SVGA3dShaderInstToken inst = inst_token( SVGA3DOP_TEXKILL );
1401
1402   /* texkill doesn't allow negation on the operand so lets move
1403    * negation of {1} to a temp register */
1404   temp = get_temp( emit );
1405   if (!submit_op1( emit, inst_token( SVGA3DOP_MOV ), temp,
1406                    negate( one ) ))
1407      return FALSE;
1408
1409   return submit_op0( emit, inst, temp );
1410}
1411
1412
1413/**
1414 * Test if r1 and r2 are the same register.
1415 */
1416static boolean
1417same_register(struct src_register r1, struct src_register r2)
1418{
1419   return (r1.base.num == r2.base.num &&
1420           r1.base.type_upper == r2.base.type_upper &&
1421           r1.base.type_lower == r2.base.type_lower);
1422}
1423
1424
1425
1426/**
1427 * Implement conditionals by initializing destination reg to 'fail',
1428 * then set predicate reg with UFOP_SETP, then move 'pass' to dest
1429 * based on predicate reg.
1430 *
1431 * SETP src0, cmp, src1  -- do this first to avoid aliasing problems.
1432 * MOV dst, fail
1433 * MOV dst, pass, p0
1434 */
1435static boolean
1436emit_conditional(struct svga_shader_emitter *emit,
1437                 enum pipe_compare_func compare_func,
1438                 SVGA3dShaderDestToken dst,
1439                 struct src_register src0,
1440                 struct src_register src1,
1441                 struct src_register pass,
1442                 struct src_register fail)
1443{
1444   SVGA3dShaderDestToken pred_reg = dst_register( SVGA3DREG_PREDICATE, 0 );
1445   SVGA3dShaderInstToken setp_token;
1446
1447   switch (compare_func) {
1448   case PIPE_FUNC_NEVER:
1449      return submit_op1( emit, inst_token( SVGA3DOP_MOV ),
1450                         dst, fail );
1451      break;
1452   case PIPE_FUNC_LESS:
1453      setp_token = inst_token_setp(SVGA3DOPCOMP_LT);
1454      break;
1455   case PIPE_FUNC_EQUAL:
1456      setp_token = inst_token_setp(SVGA3DOPCOMP_EQ);
1457      break;
1458   case PIPE_FUNC_LEQUAL:
1459      setp_token = inst_token_setp(SVGA3DOPCOMP_LE);
1460      break;
1461   case PIPE_FUNC_GREATER:
1462      setp_token = inst_token_setp(SVGA3DOPCOMP_GT);
1463      break;
1464   case PIPE_FUNC_NOTEQUAL:
1465      setp_token = inst_token_setp(SVGA3DOPCOMPC_NE);
1466      break;
1467   case PIPE_FUNC_GEQUAL:
1468      setp_token = inst_token_setp(SVGA3DOPCOMP_GE);
1469      break;
1470   case PIPE_FUNC_ALWAYS:
1471      return submit_op1( emit, inst_token( SVGA3DOP_MOV ),
1472                         dst, pass );
1473      break;
1474   }
1475
1476   if (same_register(src(dst), pass)) {
1477      /* We'll get bad results if the dst and pass registers are the same
1478       * so use a temp register containing pass.
1479       */
1480      SVGA3dShaderDestToken temp = get_temp(emit);
1481      if (!submit_op1(emit, inst_token(SVGA3DOP_MOV), temp, pass))
1482         return FALSE;
1483      pass = src(temp);
1484   }
1485
1486   /* SETP src0, COMPOP, src1 */
1487   if (!submit_op2( emit, setp_token, pred_reg,
1488                    src0, src1 ))
1489      return FALSE;
1490
1491   /* MOV dst, fail */
1492   if (!submit_op1(emit, inst_token(SVGA3DOP_MOV), dst, fail))
1493      return FALSE;
1494
1495   /* MOV dst, pass (predicated)
1496    *
1497    * Note that the predicate reg (and possible modifiers) is passed
1498    * as the first source argument.
1499    */
1500   if (!submit_op2(emit,
1501                   inst_token_predicated(SVGA3DOP_MOV), dst,
1502                   src(pred_reg), pass))
1503      return FALSE;
1504
1505   return TRUE;
1506}
1507
1508
1509/**
1510 * Helper for emiting 'selection' commands.  Basically:
1511 * if (src0 OP src1)
1512 *    dst = 1.0;
1513 * else
1514 *    dst = 0.0;
1515 */
1516static boolean
1517emit_select(struct svga_shader_emitter *emit,
1518            enum pipe_compare_func compare_func,
1519            SVGA3dShaderDestToken dst,
1520            struct src_register src0,
1521            struct src_register src1 )
1522{
1523   /* There are some SVGA instructions which implement some selects
1524    * directly, but they are only available in the vertex shader.
1525    */
1526   if (emit->unit == PIPE_SHADER_VERTEX) {
1527      switch (compare_func) {
1528      case PIPE_FUNC_GEQUAL:
1529         return submit_op2( emit, inst_token( SVGA3DOP_SGE ), dst, src0, src1 );
1530      case PIPE_FUNC_LEQUAL:
1531         return submit_op2( emit, inst_token( SVGA3DOP_SGE ), dst, src1, src0 );
1532      case PIPE_FUNC_GREATER:
1533         return submit_op2( emit, inst_token( SVGA3DOP_SLT ), dst, src1, src0 );
1534      case PIPE_FUNC_LESS:
1535         return submit_op2( emit, inst_token( SVGA3DOP_SLT ), dst, src0, src1 );
1536      default:
1537         break;
1538      }
1539   }
1540
1541   /* Otherwise, need to use the setp approach:
1542    */
1543   {
1544      struct src_register one, zero;
1545      /* zero immediate is 0,0,0,1 */
1546      zero = get_zero_immediate(emit);
1547      one = get_one_immediate(emit);
1548
1549      return emit_conditional(emit, compare_func, dst, src0, src1, one, zero);
1550   }
1551}
1552
1553
1554/**
1555 * Translate/emit a TGSI SEQ, SNE, SLT, SGE, etc. instruction.
1556 */
1557static boolean
1558emit_select_op(struct svga_shader_emitter *emit,
1559               unsigned compare,
1560               const struct tgsi_full_instruction *insn)
1561{
1562   SVGA3dShaderDestToken dst = translate_dst_register( emit, insn, 0 );
1563   struct src_register src0 = translate_src_register(
1564      emit, &insn->Src[0] );
1565   struct src_register src1 = translate_src_register(
1566      emit, &insn->Src[1] );
1567
1568   return emit_select( emit, compare, dst, src0, src1 );
1569}
1570
1571
1572/**
1573 * Translate TGSI CMP instruction.  Component-wise:
1574 * dst = (src0 < 0.0) ? src1 : src2
1575 */
1576static boolean
1577emit_cmp(struct svga_shader_emitter *emit,
1578         const struct tgsi_full_instruction *insn)
1579{
1580   SVGA3dShaderDestToken dst = translate_dst_register( emit, insn, 0 );
1581   const struct src_register src0 =
1582      translate_src_register(emit, &insn->Src[0] );
1583   const struct src_register src1 =
1584      translate_src_register(emit, &insn->Src[1] );
1585   const struct src_register src2 =
1586      translate_src_register(emit, &insn->Src[2] );
1587
1588   if (emit->unit == PIPE_SHADER_VERTEX) {
1589      struct src_register zero = get_zero_immediate(emit);
1590      /* We used to simulate CMP with SLT+LRP.  But that didn't work when
1591       * src1 or src2 was Inf/NaN.  In particular, GLSL sqrt(0) failed
1592       * because it involves a CMP to handle the 0 case.
1593       * Use a conditional expression instead.
1594       */
1595      return emit_conditional(emit, PIPE_FUNC_LESS, dst,
1596                              src0, zero, src1, src2);
1597   }
1598   else {
1599      assert(emit->unit == PIPE_SHADER_FRAGMENT);
1600
1601      /* CMP  DST, SRC0, SRC2, SRC1 */
1602      return submit_op3( emit, inst_token( SVGA3DOP_CMP ), dst,
1603                         src0, src2, src1);
1604   }
1605}
1606
1607
1608/**
1609 * Translate/emit 2-operand (coord, sampler) texture instructions.
1610 */
1611static boolean
1612emit_tex2(struct svga_shader_emitter *emit,
1613          const struct tgsi_full_instruction *insn,
1614          SVGA3dShaderDestToken dst)
1615{
1616   SVGA3dShaderInstToken inst;
1617   struct src_register texcoord;
1618   struct src_register sampler;
1619   SVGA3dShaderDestToken tmp;
1620
1621   inst.value = 0;
1622
1623   switch (insn->Instruction.Opcode) {
1624   case TGSI_OPCODE_TEX:
1625      inst.op = SVGA3DOP_TEX;
1626      break;
1627   case TGSI_OPCODE_TXP:
1628      inst.op = SVGA3DOP_TEX;
1629      inst.control = SVGA3DOPCONT_PROJECT;
1630      break;
1631   case TGSI_OPCODE_TXB:
1632      inst.op = SVGA3DOP_TEX;
1633      inst.control = SVGA3DOPCONT_BIAS;
1634      break;
1635   case TGSI_OPCODE_TXL:
1636      inst.op = SVGA3DOP_TEXLDL;
1637      break;
1638   default:
1639      assert(0);
1640      return FALSE;
1641   }
1642
1643   texcoord = translate_src_register( emit, &insn->Src[0] );
1644   sampler = translate_src_register( emit, &insn->Src[1] );
1645
1646   if (emit->key.tex[sampler.base.num].unnormalized ||
1647       emit->dynamic_branching_level > 0)
1648      tmp = get_temp( emit );
1649
1650   /* Can't do mipmapping inside dynamic branch constructs.  Force LOD
1651    * zero in that case.
1652    */
1653   if (emit->dynamic_branching_level > 0 &&
1654       inst.op == SVGA3DOP_TEX &&
1655       SVGA3dShaderGetRegType(texcoord.base.value) == SVGA3DREG_TEMP) {
1656      struct src_register zero = get_zero_immediate(emit);
1657
1658      /* MOV  tmp, texcoord */
1659      if (!submit_op1( emit,
1660                       inst_token( SVGA3DOP_MOV ),
1661                       tmp,
1662                       texcoord ))
1663         return FALSE;
1664
1665      /* MOV  tmp.w, zero */
1666      if (!submit_op1( emit,
1667                       inst_token( SVGA3DOP_MOV ),
1668                       writemask( tmp, TGSI_WRITEMASK_W ),
1669                       zero ))
1670         return FALSE;
1671
1672      texcoord = src( tmp );
1673      inst.op = SVGA3DOP_TEXLDL;
1674   }
1675
1676   /* Explicit normalization of texcoords:
1677    */
1678   if (emit->key.tex[sampler.base.num].unnormalized) {
1679      struct src_register wh = get_tex_dimensions( emit, sampler.base.num );
1680
1681      /* MUL  tmp, SRC0, WH */
1682      if (!submit_op2( emit, inst_token( SVGA3DOP_MUL ),
1683                       tmp, texcoord, wh ))
1684         return FALSE;
1685
1686      texcoord = src( tmp );
1687   }
1688
1689   return submit_op2( emit, inst, dst, texcoord, sampler );
1690}
1691
1692
1693/**
1694 * Translate/emit 4-operand (coord, ddx, ddy, sampler) texture instructions.
1695 */
1696static boolean
1697emit_tex4(struct svga_shader_emitter *emit,
1698          const struct tgsi_full_instruction *insn,
1699          SVGA3dShaderDestToken dst )
1700{
1701   SVGA3dShaderInstToken inst;
1702   struct src_register texcoord;
1703   struct src_register ddx;
1704   struct src_register ddy;
1705   struct src_register sampler;
1706
1707   texcoord = translate_src_register( emit, &insn->Src[0] );
1708   ddx      = translate_src_register( emit, &insn->Src[1] );
1709   ddy      = translate_src_register( emit, &insn->Src[2] );
1710   sampler  = translate_src_register( emit, &insn->Src[3] );
1711
1712   inst.value = 0;
1713
1714   switch (insn->Instruction.Opcode) {
1715   case TGSI_OPCODE_TXD:
1716      inst.op = SVGA3DOP_TEXLDD; /* 4 args! */
1717      break;
1718   default:
1719      assert(0);
1720      return FALSE;
1721   }
1722
1723   return submit_op4( emit, inst, dst, texcoord, sampler, ddx, ddy );
1724}
1725
1726
1727/**
1728 * Emit texture swizzle code.  We do this here since SVGA samplers don't
1729 * directly support swizzles.
1730 */
1731static boolean
1732emit_tex_swizzle(struct svga_shader_emitter *emit,
1733                 SVGA3dShaderDestToken dst,
1734                 struct src_register src,
1735                 unsigned swizzle_x,
1736                 unsigned swizzle_y,
1737                 unsigned swizzle_z,
1738                 unsigned swizzle_w)
1739{
1740   const unsigned swizzleIn[4] = {swizzle_x, swizzle_y, swizzle_z, swizzle_w};
1741   unsigned srcSwizzle[4];
1742   unsigned srcWritemask = 0x0, zeroWritemask = 0x0, oneWritemask = 0x0;
1743   unsigned i;
1744
1745   /* build writemasks and srcSwizzle terms */
1746   for (i = 0; i < 4; i++) {
1747      if (swizzleIn[i] == PIPE_SWIZZLE_0) {
1748         srcSwizzle[i] = TGSI_SWIZZLE_X + i;
1749         zeroWritemask |= (1 << i);
1750      }
1751      else if (swizzleIn[i] == PIPE_SWIZZLE_1) {
1752         srcSwizzle[i] = TGSI_SWIZZLE_X + i;
1753         oneWritemask |= (1 << i);
1754      }
1755      else {
1756         srcSwizzle[i] = swizzleIn[i];
1757         srcWritemask |= (1 << i);
1758      }
1759   }
1760
1761   /* write x/y/z/w comps */
1762   if (dst.mask & srcWritemask) {
1763      if (!submit_op1(emit,
1764                      inst_token(SVGA3DOP_MOV),
1765                      writemask(dst, srcWritemask),
1766                      swizzle(src,
1767                              srcSwizzle[0],
1768                              srcSwizzle[1],
1769                              srcSwizzle[2],
1770                              srcSwizzle[3])))
1771         return FALSE;
1772   }
1773
1774   /* write 0 comps */
1775   if (dst.mask & zeroWritemask) {
1776      if (!submit_op1(emit,
1777                      inst_token(SVGA3DOP_MOV),
1778                      writemask(dst, zeroWritemask),
1779                      get_zero_immediate(emit)))
1780         return FALSE;
1781   }
1782
1783   /* write 1 comps */
1784   if (dst.mask & oneWritemask) {
1785      if (!submit_op1(emit,
1786                      inst_token(SVGA3DOP_MOV),
1787                      writemask(dst, oneWritemask),
1788                      get_one_immediate(emit)))
1789         return FALSE;
1790   }
1791
1792   return TRUE;
1793}
1794
1795
1796/**
1797 * Translate/emit a TGSI texture sample instruction.
1798 */
1799static boolean
1800emit_tex(struct svga_shader_emitter *emit,
1801         const struct tgsi_full_instruction *insn)
1802{
1803   SVGA3dShaderDestToken dst =
1804      translate_dst_register( emit, insn, 0 );
1805   struct src_register src0 =
1806      translate_src_register( emit, &insn->Src[0] );
1807   struct src_register src1 =
1808      translate_src_register( emit, &insn->Src[1] );
1809
1810   SVGA3dShaderDestToken tex_result;
1811   const unsigned unit = src1.base.num;
1812
1813   /* check for shadow samplers */
1814   boolean compare = (emit->key.tex[unit].compare_mode ==
1815                      PIPE_TEX_COMPARE_R_TO_TEXTURE);
1816
1817   /* texture swizzle */
1818   boolean swizzle = (emit->key.tex[unit].swizzle_r != PIPE_SWIZZLE_X ||
1819                      emit->key.tex[unit].swizzle_g != PIPE_SWIZZLE_Y ||
1820                      emit->key.tex[unit].swizzle_b != PIPE_SWIZZLE_Z ||
1821                      emit->key.tex[unit].swizzle_a != PIPE_SWIZZLE_W);
1822
1823   boolean saturate = insn->Instruction.Saturate;
1824
1825   /* If doing compare processing or tex swizzle or saturation, we need to put
1826    * the fetched color into a temporary so it can be used as a source later on.
1827    */
1828   if (compare || swizzle || saturate) {
1829      tex_result = get_temp( emit );
1830   }
1831   else {
1832      tex_result = dst;
1833   }
1834
1835   switch(insn->Instruction.Opcode) {
1836   case TGSI_OPCODE_TEX:
1837   case TGSI_OPCODE_TXB:
1838   case TGSI_OPCODE_TXP:
1839   case TGSI_OPCODE_TXL:
1840      if (!emit_tex2( emit, insn, tex_result ))
1841         return FALSE;
1842      break;
1843   case TGSI_OPCODE_TXD:
1844      if (!emit_tex4( emit, insn, tex_result ))
1845         return FALSE;
1846      break;
1847   default:
1848      assert(0);
1849   }
1850
1851   if (compare) {
1852      SVGA3dShaderDestToken dst2;
1853
1854      if (swizzle || saturate)
1855         dst2 = tex_result;
1856      else
1857         dst2 = dst;
1858
1859      if (dst.mask & TGSI_WRITEMASK_XYZ) {
1860         SVGA3dShaderDestToken src0_zdivw = get_temp( emit );
1861         /* When sampling a depth texture, the result of the comparison is in
1862          * the Y component.
1863          */
1864         struct src_register tex_src_x = scalar(src(tex_result), TGSI_SWIZZLE_Y);
1865         struct src_register r_coord;
1866
1867         if (insn->Instruction.Opcode == TGSI_OPCODE_TXP) {
1868            /* Divide texcoord R by Q */
1869            if (!submit_op1( emit, inst_token( SVGA3DOP_RCP ),
1870                             writemask(src0_zdivw, TGSI_WRITEMASK_X),
1871                             scalar(src0, TGSI_SWIZZLE_W) ))
1872               return FALSE;
1873
1874            if (!submit_op2( emit, inst_token( SVGA3DOP_MUL ),
1875                             writemask(src0_zdivw, TGSI_WRITEMASK_X),
1876                             scalar(src0, TGSI_SWIZZLE_Z),
1877                             scalar(src(src0_zdivw), TGSI_SWIZZLE_X) ))
1878               return FALSE;
1879
1880            r_coord = scalar(src(src0_zdivw), TGSI_SWIZZLE_X);
1881         }
1882         else {
1883            r_coord = scalar(src0, TGSI_SWIZZLE_Z);
1884         }
1885
1886         /* Compare texture sample value against R component of texcoord */
1887         if (!emit_select(emit,
1888                          emit->key.tex[unit].compare_func,
1889                          writemask( dst2, TGSI_WRITEMASK_XYZ ),
1890                          r_coord,
1891                          tex_src_x))
1892            return FALSE;
1893      }
1894
1895      if (dst.mask & TGSI_WRITEMASK_W) {
1896         struct src_register one = get_one_immediate(emit);
1897
1898        if (!submit_op1( emit, inst_token( SVGA3DOP_MOV ),
1899                         writemask( dst2, TGSI_WRITEMASK_W ),
1900                         one ))
1901           return FALSE;
1902      }
1903   }
1904
1905   if (saturate && !swizzle) {
1906      /* MOV_SAT real_dst, dst */
1907      if (!submit_op1( emit, inst_token( SVGA3DOP_MOV ), dst, src(tex_result) ))
1908         return FALSE;
1909   }
1910   else if (swizzle) {
1911      /* swizzle from tex_result to dst (handles saturation too, if any) */
1912      emit_tex_swizzle(emit,
1913                       dst, src(tex_result),
1914                       emit->key.tex[unit].swizzle_r,
1915                       emit->key.tex[unit].swizzle_g,
1916                       emit->key.tex[unit].swizzle_b,
1917                       emit->key.tex[unit].swizzle_a);
1918   }
1919
1920   return TRUE;
1921}
1922
1923
1924static boolean
1925emit_bgnloop(struct svga_shader_emitter *emit,
1926             const struct tgsi_full_instruction *insn)
1927{
1928   SVGA3dShaderInstToken inst = inst_token( SVGA3DOP_LOOP );
1929   struct src_register loop_reg = src_register( SVGA3DREG_LOOP, 0 );
1930   struct src_register const_int = get_loop_const( emit );
1931
1932   emit->dynamic_branching_level++;
1933
1934   return (emit_instruction( emit, inst ) &&
1935           emit_src( emit, loop_reg ) &&
1936           emit_src( emit, const_int ) );
1937}
1938
1939
1940static boolean
1941emit_endloop(struct svga_shader_emitter *emit,
1942             const struct tgsi_full_instruction *insn)
1943{
1944   SVGA3dShaderInstToken inst = inst_token( SVGA3DOP_ENDLOOP );
1945
1946   emit->dynamic_branching_level--;
1947
1948   return emit_instruction( emit, inst );
1949}
1950
1951
1952/**
1953 * Translate/emit TGSI BREAK (out of loop) instruction.
1954 */
1955static boolean
1956emit_brk(struct svga_shader_emitter *emit,
1957         const struct tgsi_full_instruction *insn)
1958{
1959   SVGA3dShaderInstToken inst = inst_token( SVGA3DOP_BREAK );
1960   return emit_instruction( emit, inst );
1961}
1962
1963
1964/**
1965 * Emit simple instruction which operates on one scalar value (not
1966 * a vector).  Ex: LG2, RCP, RSQ.
1967 */
1968static boolean
1969emit_scalar_op1(struct svga_shader_emitter *emit,
1970                SVGA3dShaderOpCodeType opcode,
1971                const struct tgsi_full_instruction *insn)
1972{
1973   SVGA3dShaderInstToken inst;
1974   SVGA3dShaderDestToken dst;
1975   struct src_register src;
1976
1977   inst = inst_token( opcode );
1978   dst = translate_dst_register( emit, insn, 0 );
1979   src = translate_src_register( emit, &insn->Src[0] );
1980   src = scalar( src, TGSI_SWIZZLE_X );
1981
1982   return submit_op1( emit, inst, dst, src );
1983}
1984
1985
1986/**
1987 * Translate/emit a simple instruction (one which has no special-case
1988 * code) such as ADD, MUL, MIN, MAX.
1989 */
1990static boolean
1991emit_simple_instruction(struct svga_shader_emitter *emit,
1992                        SVGA3dShaderOpCodeType opcode,
1993                        const struct tgsi_full_instruction *insn)
1994{
1995   const struct tgsi_full_src_register *src = insn->Src;
1996   SVGA3dShaderInstToken inst;
1997   SVGA3dShaderDestToken dst;
1998
1999   inst = inst_token( opcode );
2000   dst = translate_dst_register( emit, insn, 0 );
2001
2002   switch (insn->Instruction.NumSrcRegs) {
2003   case 0:
2004      return submit_op0( emit, inst, dst );
2005   case 1:
2006      return submit_op1( emit, inst, dst,
2007                         translate_src_register( emit, &src[0] ));
2008   case 2:
2009      return submit_op2( emit, inst, dst,
2010                         translate_src_register( emit, &src[0] ),
2011                         translate_src_register( emit, &src[1] ) );
2012   case 3:
2013      return submit_op3( emit, inst, dst,
2014                         translate_src_register( emit, &src[0] ),
2015                         translate_src_register( emit, &src[1] ),
2016                         translate_src_register( emit, &src[2] ) );
2017   default:
2018      assert(0);
2019      return FALSE;
2020   }
2021}
2022
2023
2024/**
2025 * TGSI_OPCODE_MOVE is only special-cased here to detect the
2026 * svga_fragment_shader::constant_color_output case.
2027 */
2028static boolean
2029emit_mov(struct svga_shader_emitter *emit,
2030         const struct tgsi_full_instruction *insn)
2031{
2032   const struct tgsi_full_src_register *src = &insn->Src[0];
2033   const struct tgsi_full_dst_register *dst = &insn->Dst[0];
2034
2035   if (emit->unit == PIPE_SHADER_FRAGMENT &&
2036       dst->Register.File == TGSI_FILE_OUTPUT &&
2037       dst->Register.Index == 0 &&
2038       src->Register.File == TGSI_FILE_CONSTANT &&
2039       !src->Register.Indirect) {
2040      emit->constant_color_output = TRUE;
2041   }
2042
2043   return emit_simple_instruction(emit, SVGA3DOP_MOV, insn);
2044}
2045
2046
2047/**
2048 * Translate TGSI SQRT instruction
2049 * if src1 == 0
2050 *    mov dst, src1
2051 * else
2052 *    rsq temp, src1
2053 *    rcp dst, temp
2054 * endif
2055 */
2056static boolean
2057emit_sqrt(struct svga_shader_emitter *emit,
2058         const struct tgsi_full_instruction *insn)
2059{
2060   const struct src_register src1 = translate_src_register(emit, &insn->Src[0]);
2061   const struct src_register zero = get_zero_immediate(emit);
2062   SVGA3dShaderDestToken dst = translate_dst_register(emit, insn, 0);
2063   SVGA3dShaderDestToken temp = get_temp(emit);
2064   SVGA3dShaderInstToken if_token = inst_token(SVGA3DOP_IFC);
2065   boolean ret = TRUE;
2066
2067   if_token.control = SVGA3DOPCOMP_EQ;
2068
2069   if (!(emit_instruction(emit, if_token) &&
2070         emit_src(emit, src1) &&
2071         emit_src(emit, zero))) {
2072      ret = FALSE;
2073      goto cleanup;
2074   }
2075
2076   if (!submit_op1(emit,
2077              inst_token(SVGA3DOP_MOV),
2078              dst, src1)) {
2079      ret = FALSE;
2080      goto cleanup;
2081   }
2082
2083   if (!emit_instruction(emit, inst_token(SVGA3DOP_ELSE))) {
2084      ret = FALSE;
2085      goto cleanup;
2086   }
2087
2088   if (!submit_op1(emit,
2089              inst_token(SVGA3DOP_RSQ),
2090              temp, src1)) {
2091      ret = FALSE;
2092      goto cleanup;
2093   }
2094
2095   if (!submit_op1(emit,
2096              inst_token(SVGA3DOP_RCP),
2097              dst, src(temp))) {
2098      ret = FALSE;
2099      goto cleanup;
2100   }
2101
2102   if (!emit_instruction(emit, inst_token(SVGA3DOP_ENDIF))) {
2103      ret = FALSE;
2104      goto cleanup;
2105   }
2106
2107cleanup:
2108   release_temp(emit, temp);
2109
2110   return ret;
2111}
2112
2113
2114/**
2115 * Translate/emit TGSI DDX, DDY instructions.
2116 */
2117static boolean
2118emit_deriv(struct svga_shader_emitter *emit,
2119           const struct tgsi_full_instruction *insn )
2120{
2121   if (emit->dynamic_branching_level > 0 &&
2122       insn->Src[0].Register.File == TGSI_FILE_TEMPORARY)
2123   {
2124      SVGA3dShaderDestToken dst =
2125         translate_dst_register( emit, insn, 0 );
2126
2127      /* Deriv opcodes not valid inside dynamic branching, workaround
2128       * by zeroing out the destination.
2129       */
2130      if (!submit_op1(emit,
2131                      inst_token( SVGA3DOP_MOV ),
2132                      dst,
2133                      get_zero_immediate(emit)))
2134         return FALSE;
2135
2136      return TRUE;
2137   }
2138   else {
2139      SVGA3dShaderOpCodeType opcode;
2140      const struct tgsi_full_src_register *reg = &insn->Src[0];
2141      SVGA3dShaderInstToken inst;
2142      SVGA3dShaderDestToken dst;
2143      struct src_register src0;
2144
2145      switch (insn->Instruction.Opcode) {
2146      case TGSI_OPCODE_DDX:
2147         opcode = SVGA3DOP_DSX;
2148         break;
2149      case TGSI_OPCODE_DDY:
2150         opcode = SVGA3DOP_DSY;
2151         break;
2152      default:
2153         return FALSE;
2154      }
2155
2156      inst = inst_token( opcode );
2157      dst = translate_dst_register( emit, insn, 0 );
2158      src0 = translate_src_register( emit, reg );
2159
2160      /* We cannot use negate or abs on source to dsx/dsy instruction.
2161       */
2162      if (reg->Register.Absolute ||
2163          reg->Register.Negate) {
2164         SVGA3dShaderDestToken temp = get_temp( emit );
2165
2166         if (!emit_repl( emit, temp, &src0 ))
2167            return FALSE;
2168      }
2169
2170      return submit_op1( emit, inst, dst, src0 );
2171   }
2172}
2173
2174
2175/**
2176 * Translate/emit ARL (Address Register Load) instruction.  Used to
2177 * move a value into the special 'address' register.  Used to implement
2178 * indirect/variable indexing into arrays.
2179 */
2180static boolean
2181emit_arl(struct svga_shader_emitter *emit,
2182         const struct tgsi_full_instruction *insn)
2183{
2184   ++emit->current_arl;
2185   if (emit->unit == PIPE_SHADER_FRAGMENT) {
2186      /* MOVA not present in pixel shader instruction set.
2187       * Ignore this instruction altogether since it is
2188       * only used for loop counters -- and for that
2189       * we reference aL directly.
2190       */
2191      return TRUE;
2192   }
2193   if (svga_arl_needs_adjustment( emit )) {
2194      return emit_fake_arl( emit, insn );
2195   } else {
2196      /* no need to adjust, just emit straight arl */
2197      return emit_simple_instruction(emit, SVGA3DOP_MOVA, insn);
2198   }
2199}
2200
2201
2202static boolean
2203emit_pow(struct svga_shader_emitter *emit,
2204         const struct tgsi_full_instruction *insn)
2205{
2206   SVGA3dShaderDestToken dst = translate_dst_register( emit, insn, 0 );
2207   struct src_register src0 = translate_src_register(
2208      emit, &insn->Src[0] );
2209   struct src_register src1 = translate_src_register(
2210      emit, &insn->Src[1] );
2211   boolean need_tmp = FALSE;
2212
2213   /* POW can only output to a temporary */
2214   if (insn->Dst[0].Register.File != TGSI_FILE_TEMPORARY)
2215      need_tmp = TRUE;
2216
2217   /* POW src1 must not be the same register as dst */
2218   if (alias_src_dst( src1, dst ))
2219      need_tmp = TRUE;
2220
2221   /* it's a scalar op */
2222   src0 = scalar( src0, TGSI_SWIZZLE_X );
2223   src1 = scalar( src1, TGSI_SWIZZLE_X );
2224
2225   if (need_tmp) {
2226      SVGA3dShaderDestToken tmp =
2227         writemask(get_temp( emit ), TGSI_WRITEMASK_X );
2228
2229      if (!submit_op2(emit, inst_token( SVGA3DOP_POW ), tmp, src0, src1))
2230         return FALSE;
2231
2232      return submit_op1(emit, inst_token( SVGA3DOP_MOV ),
2233                        dst, scalar(src(tmp), 0) );
2234   }
2235   else {
2236      return submit_op2(emit, inst_token( SVGA3DOP_POW ), dst, src0, src1);
2237   }
2238}
2239
2240
2241/**
2242 * Emit a LRP (linear interpolation) instruction.
2243 */
2244static boolean
2245submit_lrp(struct svga_shader_emitter *emit,
2246           SVGA3dShaderDestToken dst,
2247           struct src_register src0,
2248           struct src_register src1,
2249           struct src_register src2)
2250{
2251   SVGA3dShaderDestToken tmp;
2252   boolean need_dst_tmp = FALSE;
2253
2254   /* The dst reg must be a temporary, and not be the same as src0 or src2 */
2255   if (SVGA3dShaderGetRegType(dst.value) != SVGA3DREG_TEMP ||
2256       alias_src_dst(src0, dst) ||
2257       alias_src_dst(src2, dst))
2258      need_dst_tmp = TRUE;
2259
2260   if (need_dst_tmp) {
2261      tmp = get_temp( emit );
2262      tmp.mask = dst.mask;
2263   }
2264   else {
2265      tmp = dst;
2266   }
2267
2268   if (!submit_op3(emit, inst_token( SVGA3DOP_LRP ), tmp, src0, src1, src2))
2269      return FALSE;
2270
2271   if (need_dst_tmp) {
2272      if (!submit_op1(emit, inst_token( SVGA3DOP_MOV ), dst, src( tmp )))
2273         return FALSE;
2274   }
2275
2276   return TRUE;
2277}
2278
2279
2280/**
2281 * Translate/emit LRP (Linear Interpolation) instruction.
2282 */
2283static boolean
2284emit_lrp(struct svga_shader_emitter *emit,
2285         const struct tgsi_full_instruction *insn)
2286{
2287   SVGA3dShaderDestToken dst = translate_dst_register( emit, insn, 0 );
2288   const struct src_register src0 = translate_src_register(
2289      emit, &insn->Src[0] );
2290   const struct src_register src1 = translate_src_register(
2291      emit, &insn->Src[1] );
2292   const struct src_register src2 = translate_src_register(
2293      emit, &insn->Src[2] );
2294
2295   return submit_lrp(emit, dst, src0, src1, src2);
2296}
2297
2298/**
2299 * Translate/emit DST (Distance function) instruction.
2300 */
2301static boolean
2302emit_dst_insn(struct svga_shader_emitter *emit,
2303              const struct tgsi_full_instruction *insn)
2304{
2305   if (emit->unit == PIPE_SHADER_VERTEX) {
2306      /* SVGA/DX9 has a DST instruction, but only for vertex shaders:
2307       */
2308      return emit_simple_instruction(emit, SVGA3DOP_DST, insn);
2309   }
2310   else {
2311      /* result[0] = 1    * 1;
2312       * result[1] = a[1] * b[1];
2313       * result[2] = a[2] * 1;
2314       * result[3] = 1    * b[3];
2315       */
2316      SVGA3dShaderDestToken dst = translate_dst_register( emit, insn, 0 );
2317      SVGA3dShaderDestToken tmp;
2318      const struct src_register src0 = translate_src_register(
2319         emit, &insn->Src[0] );
2320      const struct src_register src1 = translate_src_register(
2321         emit, &insn->Src[1] );
2322      boolean need_tmp = FALSE;
2323
2324      if (SVGA3dShaderGetRegType(dst.value) != SVGA3DREG_TEMP ||
2325          alias_src_dst(src0, dst) ||
2326          alias_src_dst(src1, dst))
2327         need_tmp = TRUE;
2328
2329      if (need_tmp) {
2330         tmp = get_temp( emit );
2331      }
2332      else {
2333         tmp = dst;
2334      }
2335
2336      /* tmp.xw = 1.0
2337       */
2338      if (tmp.mask & TGSI_WRITEMASK_XW) {
2339         if (!submit_op1( emit, inst_token( SVGA3DOP_MOV ),
2340                          writemask(tmp, TGSI_WRITEMASK_XW ),
2341                          get_one_immediate(emit)))
2342            return FALSE;
2343      }
2344
2345      /* tmp.yz = src0
2346       */
2347      if (tmp.mask & TGSI_WRITEMASK_YZ) {
2348         if (!submit_op1( emit, inst_token( SVGA3DOP_MOV ),
2349                          writemask(tmp, TGSI_WRITEMASK_YZ ),
2350                          src0))
2351            return FALSE;
2352      }
2353
2354      /* tmp.yw = tmp * src1
2355       */
2356      if (tmp.mask & TGSI_WRITEMASK_YW) {
2357         if (!submit_op2( emit, inst_token( SVGA3DOP_MUL ),
2358                          writemask(tmp, TGSI_WRITEMASK_YW ),
2359                          src(tmp),
2360                          src1))
2361            return FALSE;
2362      }
2363
2364      /* dst = tmp
2365       */
2366      if (need_tmp) {
2367         if (!submit_op1( emit, inst_token( SVGA3DOP_MOV ),
2368                          dst,
2369                          src(tmp)))
2370            return FALSE;
2371      }
2372   }
2373
2374   return TRUE;
2375}
2376
2377
2378static boolean
2379emit_exp(struct svga_shader_emitter *emit,
2380         const struct tgsi_full_instruction *insn)
2381{
2382   SVGA3dShaderDestToken dst = translate_dst_register( emit, insn, 0 );
2383   struct src_register src0 =
2384      translate_src_register( emit, &insn->Src[0] );
2385   SVGA3dShaderDestToken fraction;
2386
2387   if (dst.mask & TGSI_WRITEMASK_Y)
2388      fraction = dst;
2389   else if (dst.mask & TGSI_WRITEMASK_X)
2390      fraction = get_temp( emit );
2391   else
2392      fraction.value = 0;
2393
2394   /* If y is being written, fill it with src0 - floor(src0).
2395    */
2396   if (dst.mask & TGSI_WRITEMASK_XY) {
2397      if (!submit_op1( emit, inst_token( SVGA3DOP_FRC ),
2398                       writemask( fraction, TGSI_WRITEMASK_Y ),
2399                       src0 ))
2400         return FALSE;
2401   }
2402
2403   /* If x is being written, fill it with 2 ^ floor(src0).
2404    */
2405   if (dst.mask & TGSI_WRITEMASK_X) {
2406      if (!submit_op2( emit, inst_token( SVGA3DOP_ADD ),
2407                       writemask( dst, TGSI_WRITEMASK_X ),
2408                       src0,
2409                       scalar( negate( src( fraction ) ), TGSI_SWIZZLE_Y ) ) )
2410         return FALSE;
2411
2412      if (!submit_op1( emit, inst_token( SVGA3DOP_EXP ),
2413                       writemask( dst, TGSI_WRITEMASK_X ),
2414                       scalar( src( dst ), TGSI_SWIZZLE_X ) ) )
2415         return FALSE;
2416
2417      if (!(dst.mask & TGSI_WRITEMASK_Y))
2418         release_temp( emit, fraction );
2419   }
2420
2421   /* If z is being written, fill it with 2 ^ src0 (partial precision).
2422    */
2423   if (dst.mask & TGSI_WRITEMASK_Z) {
2424      if (!submit_op1( emit, inst_token( SVGA3DOP_EXPP ),
2425                       writemask( dst, TGSI_WRITEMASK_Z ),
2426                       src0 ) )
2427         return FALSE;
2428   }
2429
2430   /* If w is being written, fill it with one.
2431    */
2432   if (dst.mask & TGSI_WRITEMASK_W) {
2433      if (!submit_op1( emit, inst_token( SVGA3DOP_MOV ),
2434                       writemask(dst, TGSI_WRITEMASK_W),
2435                       get_one_immediate(emit)))
2436         return FALSE;
2437   }
2438
2439   return TRUE;
2440}
2441
2442
2443/**
2444 * Translate/emit LIT (Lighting helper) instruction.
2445 */
2446static boolean
2447emit_lit(struct svga_shader_emitter *emit,
2448         const struct tgsi_full_instruction *insn)
2449{
2450   if (emit->unit == PIPE_SHADER_VERTEX) {
2451      /* SVGA/DX9 has a LIT instruction, but only for vertex shaders:
2452       */
2453      return emit_simple_instruction(emit, SVGA3DOP_LIT, insn);
2454   }
2455   else {
2456      /* D3D vs. GL semantics can be fairly easily accommodated by
2457       * variations on this sequence.
2458       *
2459       * GL:
2460       *   tmp.y = src.x
2461       *   tmp.z = pow(src.y,src.w)
2462       *   p0 = src0.xxxx > 0
2463       *   result = zero.wxxw
2464       *   (p0) result.yz = tmp
2465       *
2466       * D3D:
2467       *   tmp.y = src.x
2468       *   tmp.z = pow(src.y,src.w)
2469       *   p0 = src0.xxyy > 0
2470       *   result = zero.wxxw
2471       *   (p0) result.yz = tmp
2472       *
2473       * Will implement the GL version for now.
2474       */
2475      SVGA3dShaderDestToken dst = translate_dst_register( emit, insn, 0 );
2476      SVGA3dShaderDestToken tmp = get_temp( emit );
2477      const struct src_register src0 = translate_src_register(
2478         emit, &insn->Src[0] );
2479
2480      /* tmp = pow(src.y, src.w)
2481       */
2482      if (dst.mask & TGSI_WRITEMASK_Z) {
2483         if (!submit_op2(emit, inst_token( SVGA3DOP_POW ),
2484                         tmp,
2485                         scalar(src0, 1),
2486                         scalar(src0, 3)))
2487            return FALSE;
2488      }
2489
2490      /* tmp.y = src.x
2491       */
2492      if (dst.mask & TGSI_WRITEMASK_Y) {
2493         if (!submit_op1( emit, inst_token( SVGA3DOP_MOV ),
2494                          writemask(tmp, TGSI_WRITEMASK_Y ),
2495                          scalar(src0, 0)))
2496            return FALSE;
2497      }
2498
2499      /* Can't quite do this with emit conditional due to the extra
2500       * writemask on the predicated mov:
2501       */
2502      {
2503         SVGA3dShaderDestToken pred_reg = dst_register( SVGA3DREG_PREDICATE, 0 );
2504         struct src_register predsrc;
2505
2506         /* D3D vs GL semantics:
2507          */
2508         if (0)
2509            predsrc = swizzle(src0, 0, 0, 1, 1); /* D3D */
2510         else
2511            predsrc = swizzle(src0, 0, 0, 0, 0); /* GL */
2512
2513         /* SETP src0.xxyy, GT, {0}.x */
2514         if (!submit_op2( emit,
2515                          inst_token_setp(SVGA3DOPCOMP_GT),
2516                          pred_reg,
2517                          predsrc,
2518                          get_zero_immediate(emit)))
2519            return FALSE;
2520
2521         /* MOV dst, fail */
2522         if (!submit_op1( emit, inst_token( SVGA3DOP_MOV ), dst,
2523                          get_immediate(emit, 1.0f, 0.0f, 0.0f, 1.0f)))
2524             return FALSE;
2525
2526         /* MOV dst.yz, tmp (predicated)
2527          *
2528          * Note that the predicate reg (and possible modifiers) is passed
2529          * as the first source argument.
2530          */
2531         if (dst.mask & TGSI_WRITEMASK_YZ) {
2532            if (!submit_op2( emit,
2533                             inst_token_predicated(SVGA3DOP_MOV),
2534                             writemask(dst, TGSI_WRITEMASK_YZ),
2535                             src( pred_reg ), src( tmp ) ))
2536               return FALSE;
2537         }
2538      }
2539   }
2540
2541   return TRUE;
2542}
2543
2544
2545static boolean
2546emit_ex2(struct svga_shader_emitter *emit,
2547         const struct tgsi_full_instruction *insn)
2548{
2549   SVGA3dShaderInstToken inst;
2550   SVGA3dShaderDestToken dst;
2551   struct src_register src0;
2552
2553   inst = inst_token( SVGA3DOP_EXP );
2554   dst = translate_dst_register( emit, insn, 0 );
2555   src0 = translate_src_register( emit, &insn->Src[0] );
2556   src0 = scalar( src0, TGSI_SWIZZLE_X );
2557
2558   if (dst.mask != TGSI_WRITEMASK_XYZW) {
2559      SVGA3dShaderDestToken tmp = get_temp( emit );
2560
2561      if (!submit_op1( emit, inst, tmp, src0 ))
2562         return FALSE;
2563
2564      return submit_op1( emit, inst_token( SVGA3DOP_MOV ),
2565                         dst,
2566                         scalar( src( tmp ), TGSI_SWIZZLE_X ) );
2567   }
2568
2569   return submit_op1( emit, inst, dst, src0 );
2570}
2571
2572
2573static boolean
2574emit_log(struct svga_shader_emitter *emit,
2575         const struct tgsi_full_instruction *insn)
2576{
2577   SVGA3dShaderDestToken dst = translate_dst_register( emit, insn, 0 );
2578   struct src_register src0 =
2579      translate_src_register( emit, &insn->Src[0] );
2580   SVGA3dShaderDestToken abs_tmp;
2581   struct src_register abs_src0;
2582   SVGA3dShaderDestToken log2_abs;
2583
2584   abs_tmp.value = 0;
2585
2586   if (dst.mask & TGSI_WRITEMASK_Z)
2587      log2_abs = dst;
2588   else if (dst.mask & TGSI_WRITEMASK_XY)
2589      log2_abs = get_temp( emit );
2590   else
2591      log2_abs.value = 0;
2592
2593   /* If z is being written, fill it with log2( abs( src0 ) ).
2594    */
2595   if (dst.mask & TGSI_WRITEMASK_XYZ) {
2596      if (!src0.base.srcMod || src0.base.srcMod == SVGA3DSRCMOD_ABS)
2597         abs_src0 = src0;
2598      else {
2599         abs_tmp = get_temp( emit );
2600
2601         if (!submit_op1( emit, inst_token( SVGA3DOP_MOV ),
2602                          abs_tmp,
2603                          src0 ) )
2604            return FALSE;
2605
2606         abs_src0 = src( abs_tmp );
2607      }
2608
2609      abs_src0 = absolute( scalar( abs_src0, TGSI_SWIZZLE_X ) );
2610
2611      if (!submit_op1( emit, inst_token( SVGA3DOP_LOG ),
2612                       writemask( log2_abs, TGSI_WRITEMASK_Z ),
2613                       abs_src0 ) )
2614         return FALSE;
2615   }
2616
2617   if (dst.mask & TGSI_WRITEMASK_XY) {
2618      SVGA3dShaderDestToken floor_log2;
2619
2620      if (dst.mask & TGSI_WRITEMASK_X)
2621         floor_log2 = dst;
2622      else
2623         floor_log2 = get_temp( emit );
2624
2625      /* If x is being written, fill it with floor( log2( abs( src0 ) ) ).
2626       */
2627      if (!submit_op1( emit, inst_token( SVGA3DOP_FRC ),
2628                       writemask( floor_log2, TGSI_WRITEMASK_X ),
2629                       scalar( src( log2_abs ), TGSI_SWIZZLE_Z ) ) )
2630         return FALSE;
2631
2632      if (!submit_op2( emit, inst_token( SVGA3DOP_ADD ),
2633                       writemask( floor_log2, TGSI_WRITEMASK_X ),
2634                       scalar( src( log2_abs ), TGSI_SWIZZLE_Z ),
2635                       negate( src( floor_log2 ) ) ) )
2636         return FALSE;
2637
2638      /* If y is being written, fill it with
2639       * abs ( src0 ) / ( 2 ^ floor( log2( abs( src0 ) ) ) ).
2640       */
2641      if (dst.mask & TGSI_WRITEMASK_Y) {
2642         if (!submit_op1( emit, inst_token( SVGA3DOP_EXP ),
2643                          writemask( dst, TGSI_WRITEMASK_Y ),
2644                          negate( scalar( src( floor_log2 ),
2645                                          TGSI_SWIZZLE_X ) ) ) )
2646            return FALSE;
2647
2648         if (!submit_op2( emit, inst_token( SVGA3DOP_MUL ),
2649                          writemask( dst, TGSI_WRITEMASK_Y ),
2650                          src( dst ),
2651                          abs_src0 ) )
2652            return FALSE;
2653      }
2654
2655      if (!(dst.mask & TGSI_WRITEMASK_X))
2656         release_temp( emit, floor_log2 );
2657
2658      if (!(dst.mask & TGSI_WRITEMASK_Z))
2659         release_temp( emit, log2_abs );
2660   }
2661
2662   if (dst.mask & TGSI_WRITEMASK_XYZ && src0.base.srcMod &&
2663       src0.base.srcMod != SVGA3DSRCMOD_ABS)
2664      release_temp( emit, abs_tmp );
2665
2666   /* If w is being written, fill it with one.
2667    */
2668   if (dst.mask & TGSI_WRITEMASK_W) {
2669      if (!submit_op1( emit, inst_token( SVGA3DOP_MOV ),
2670                       writemask(dst, TGSI_WRITEMASK_W),
2671                       get_one_immediate(emit)))
2672         return FALSE;
2673   }
2674
2675   return TRUE;
2676}
2677
2678
2679/**
2680 * Translate TGSI TRUNC or ROUND instruction.
2681 * We need to truncate toward zero. Ex: trunc(-1.9) = -1
2682 * Different approaches are needed for VS versus PS.
2683 */
2684static boolean
2685emit_trunc_round(struct svga_shader_emitter *emit,
2686                 const struct tgsi_full_instruction *insn,
2687                 boolean round)
2688{
2689   SVGA3dShaderDestToken dst = translate_dst_register(emit, insn, 0);
2690   const struct src_register src0 =
2691      translate_src_register(emit, &insn->Src[0] );
2692   SVGA3dShaderDestToken t1 = get_temp(emit);
2693
2694   if (round) {
2695      SVGA3dShaderDestToken t0 = get_temp(emit);
2696      struct src_register half = get_half_immediate(emit);
2697
2698      /* t0 = abs(src0) + 0.5 */
2699      if (!submit_op2(emit, inst_token(SVGA3DOP_ADD), t0,
2700                      absolute(src0), half))
2701         return FALSE;
2702
2703      /* t1 = fract(t0) */
2704      if (!submit_op1(emit, inst_token(SVGA3DOP_FRC), t1, src(t0)))
2705         return FALSE;
2706
2707      /* t1 = t0 - t1 */
2708      if (!submit_op2(emit, inst_token(SVGA3DOP_ADD), t1, src(t0),
2709                      negate(src(t1))))
2710         return FALSE;
2711   }
2712   else {
2713      /* trunc */
2714
2715      /* t1 = fract(abs(src0)) */
2716      if (!submit_op1(emit, inst_token(SVGA3DOP_FRC), t1, absolute(src0)))
2717         return FALSE;
2718
2719      /* t1 = abs(src0) - t1 */
2720      if (!submit_op2(emit, inst_token(SVGA3DOP_ADD), t1, absolute(src0),
2721                      negate(src(t1))))
2722         return FALSE;
2723   }
2724
2725   /*
2726    * Now we need to multiply t1 by the sign of the original value.
2727   */
2728   if (emit->unit == PIPE_SHADER_VERTEX) {
2729      /* For VS: use SGN instruction */
2730      /* Need two extra/dummy registers: */
2731      SVGA3dShaderDestToken t2 = get_temp(emit), t3 = get_temp(emit),
2732         t4 = get_temp(emit);
2733
2734      /* t2 = sign(src0) */
2735      if (!submit_op3(emit, inst_token(SVGA3DOP_SGN), t2, src0,
2736                      src(t3), src(t4)))
2737         return FALSE;
2738
2739      /* dst = t1 * t2 */
2740      if (!submit_op2(emit, inst_token(SVGA3DOP_MUL), dst, src(t1), src(t2)))
2741         return FALSE;
2742   }
2743   else {
2744      /* For FS: Use CMP instruction */
2745      return submit_op3(emit, inst_token( SVGA3DOP_CMP ), dst,
2746                        src0, src(t1), negate(src(t1)));
2747   }
2748
2749   return TRUE;
2750}
2751
2752
2753/**
2754 * Translate/emit "begin subroutine" instruction/marker/label.
2755 */
2756static boolean
2757emit_bgnsub(struct svga_shader_emitter *emit,
2758            unsigned position,
2759            const struct tgsi_full_instruction *insn)
2760{
2761   unsigned i;
2762
2763   /* Note that we've finished the main function and are now emitting
2764    * subroutines.  This affects how we terminate the generated
2765    * shader.
2766    */
2767   emit->in_main_func = FALSE;
2768
2769   for (i = 0; i < emit->nr_labels; i++) {
2770      if (emit->label[i] == position) {
2771         return (emit_instruction( emit, inst_token( SVGA3DOP_RET ) ) &&
2772                 emit_instruction( emit, inst_token( SVGA3DOP_LABEL ) ) &&
2773                 emit_src( emit, src_register( SVGA3DREG_LABEL, i )));
2774      }
2775   }
2776
2777   assert(0);
2778   return TRUE;
2779}
2780
2781
2782/**
2783 * Translate/emit subroutine call instruction.
2784 */
2785static boolean
2786emit_call(struct svga_shader_emitter *emit,
2787          const struct tgsi_full_instruction *insn)
2788{
2789   unsigned position = insn->Label.Label;
2790   unsigned i;
2791
2792   for (i = 0; i < emit->nr_labels; i++) {
2793      if (emit->label[i] == position)
2794         break;
2795   }
2796
2797   if (emit->nr_labels == ARRAY_SIZE(emit->label))
2798      return FALSE;
2799
2800   if (i == emit->nr_labels) {
2801      emit->label[i] = position;
2802      emit->nr_labels++;
2803   }
2804
2805   return (emit_instruction( emit, inst_token( SVGA3DOP_CALL ) ) &&
2806           emit_src( emit, src_register( SVGA3DREG_LABEL, i )));
2807}
2808
2809
2810/**
2811 * Called at the end of the shader.  Actually, emit special "fix-up"
2812 * code for the vertex/fragment shader.
2813 */
2814static boolean
2815emit_end(struct svga_shader_emitter *emit)
2816{
2817   if (emit->unit == PIPE_SHADER_VERTEX) {
2818      return emit_vs_postamble( emit );
2819   }
2820   else {
2821      return emit_ps_postamble( emit );
2822   }
2823}
2824
2825
2826/**
2827 * Translate any TGSI instruction to SVGA.
2828 */
2829static boolean
2830svga_emit_instruction(struct svga_shader_emitter *emit,
2831                      unsigned position,
2832                      const struct tgsi_full_instruction *insn)
2833{
2834   switch (insn->Instruction.Opcode) {
2835
2836   case TGSI_OPCODE_ARL:
2837      return emit_arl( emit, insn );
2838
2839   case TGSI_OPCODE_TEX:
2840   case TGSI_OPCODE_TXB:
2841   case TGSI_OPCODE_TXP:
2842   case TGSI_OPCODE_TXL:
2843   case TGSI_OPCODE_TXD:
2844      return emit_tex( emit, insn );
2845
2846   case TGSI_OPCODE_DDX:
2847   case TGSI_OPCODE_DDY:
2848      return emit_deriv( emit, insn );
2849
2850   case TGSI_OPCODE_BGNSUB:
2851      return emit_bgnsub( emit, position, insn );
2852
2853   case TGSI_OPCODE_ENDSUB:
2854      return TRUE;
2855
2856   case TGSI_OPCODE_CAL:
2857      return emit_call( emit, insn );
2858
2859   case TGSI_OPCODE_FLR:
2860      return emit_floor( emit, insn );
2861
2862   case TGSI_OPCODE_TRUNC:
2863      return emit_trunc_round( emit, insn, FALSE );
2864
2865   case TGSI_OPCODE_ROUND:
2866      return emit_trunc_round( emit, insn, TRUE );
2867
2868   case TGSI_OPCODE_CEIL:
2869      return emit_ceil( emit, insn );
2870
2871   case TGSI_OPCODE_CMP:
2872      return emit_cmp( emit, insn );
2873
2874   case TGSI_OPCODE_DIV:
2875      return emit_div( emit, insn );
2876
2877   case TGSI_OPCODE_DP2:
2878      return emit_dp2( emit, insn );
2879
2880   case TGSI_OPCODE_COS:
2881      return emit_cos( emit, insn );
2882
2883   case TGSI_OPCODE_SIN:
2884      return emit_sin( emit, insn );
2885
2886   case TGSI_OPCODE_END:
2887      /* TGSI always finishes the main func with an END */
2888      return emit_end( emit );
2889
2890   case TGSI_OPCODE_KILL_IF:
2891      return emit_cond_discard( emit, insn );
2892
2893      /* Selection opcodes.  The underlying language is fairly
2894       * non-orthogonal about these.
2895       */
2896   case TGSI_OPCODE_SEQ:
2897      return emit_select_op( emit, PIPE_FUNC_EQUAL, insn );
2898
2899   case TGSI_OPCODE_SNE:
2900      return emit_select_op( emit, PIPE_FUNC_NOTEQUAL, insn );
2901
2902   case TGSI_OPCODE_SGT:
2903      return emit_select_op( emit, PIPE_FUNC_GREATER, insn );
2904
2905   case TGSI_OPCODE_SGE:
2906      return emit_select_op( emit, PIPE_FUNC_GEQUAL, insn );
2907
2908   case TGSI_OPCODE_SLT:
2909      return emit_select_op( emit, PIPE_FUNC_LESS, insn );
2910
2911   case TGSI_OPCODE_SLE:
2912      return emit_select_op( emit, PIPE_FUNC_LEQUAL, insn );
2913
2914   case TGSI_OPCODE_POW:
2915      return emit_pow( emit, insn );
2916
2917   case TGSI_OPCODE_EX2:
2918      return emit_ex2( emit, insn );
2919
2920   case TGSI_OPCODE_EXP:
2921      return emit_exp( emit, insn );
2922
2923   case TGSI_OPCODE_LOG:
2924      return emit_log( emit, insn );
2925
2926   case TGSI_OPCODE_LG2:
2927      return emit_scalar_op1( emit, SVGA3DOP_LOG, insn );
2928
2929   case TGSI_OPCODE_RSQ:
2930      return emit_scalar_op1( emit, SVGA3DOP_RSQ, insn );
2931
2932   case TGSI_OPCODE_RCP:
2933      return emit_scalar_op1( emit, SVGA3DOP_RCP, insn );
2934
2935   case TGSI_OPCODE_CONT:
2936      /* not expected (we return PIPE_SHADER_CAP_CONT_SUPPORTED = 0) */
2937      return FALSE;
2938
2939   case TGSI_OPCODE_RET:
2940      /* This is a noop -- we tell mesa that we can't support RET
2941       * within a function (early return), so this will always be
2942       * followed by an ENDSUB.
2943       */
2944      return TRUE;
2945
2946      /* These aren't actually used by any of the frontends we care
2947       * about:
2948       */
2949   case TGSI_OPCODE_AND:
2950   case TGSI_OPCODE_OR:
2951   case TGSI_OPCODE_I2F:
2952   case TGSI_OPCODE_NOT:
2953   case TGSI_OPCODE_SHL:
2954   case TGSI_OPCODE_ISHR:
2955   case TGSI_OPCODE_XOR:
2956      return FALSE;
2957
2958   case TGSI_OPCODE_IF:
2959      return emit_if( emit, insn );
2960   case TGSI_OPCODE_ELSE:
2961      return emit_else( emit, insn );
2962   case TGSI_OPCODE_ENDIF:
2963      return emit_endif( emit, insn );
2964
2965   case TGSI_OPCODE_BGNLOOP:
2966      return emit_bgnloop( emit, insn );
2967   case TGSI_OPCODE_ENDLOOP:
2968      return emit_endloop( emit, insn );
2969   case TGSI_OPCODE_BRK:
2970      return emit_brk( emit, insn );
2971
2972   case TGSI_OPCODE_KILL:
2973      return emit_discard( emit, insn );
2974
2975   case TGSI_OPCODE_DST:
2976      return emit_dst_insn( emit, insn );
2977
2978   case TGSI_OPCODE_LIT:
2979      return emit_lit( emit, insn );
2980
2981   case TGSI_OPCODE_LRP:
2982      return emit_lrp( emit, insn );
2983
2984   case TGSI_OPCODE_SSG:
2985      return emit_ssg( emit, insn );
2986
2987   case TGSI_OPCODE_MOV:
2988      return emit_mov( emit, insn );
2989
2990   case TGSI_OPCODE_SQRT:
2991      return emit_sqrt( emit, insn );
2992
2993   default:
2994      {
2995         SVGA3dShaderOpCodeType opcode =
2996            translate_opcode(insn->Instruction.Opcode);
2997
2998         if (opcode == SVGA3DOP_LAST_INST)
2999            return FALSE;
3000
3001         if (!emit_simple_instruction( emit, opcode, insn ))
3002            return FALSE;
3003      }
3004   }
3005
3006   return TRUE;
3007}
3008
3009
3010/**
3011 * Translate/emit a TGSI IMMEDIATE declaration.
3012 * An immediate vector is a constant that's hard-coded into the shader.
3013 */
3014static boolean
3015svga_emit_immediate(struct svga_shader_emitter *emit,
3016                    const struct tgsi_full_immediate *imm)
3017{
3018   static const float id[4] = {0,0,0,1};
3019   float value[4];
3020   unsigned i;
3021
3022   assert(1 <= imm->Immediate.NrTokens && imm->Immediate.NrTokens <= 5);
3023   for (i = 0; i < 4 && i < imm->Immediate.NrTokens - 1; i++) {
3024      float f = imm->u[i].Float;
3025      value[i] = util_is_inf_or_nan(f) ? 0.0f : f;
3026   }
3027
3028   /* If the immediate has less than four values, fill in the remaining
3029    * positions from id={0,0,0,1}.
3030    */
3031   for ( ; i < 4; i++ )
3032      value[i] = id[i];
3033
3034   return emit_def_const( emit, SVGA3D_CONST_TYPE_FLOAT,
3035                          emit->imm_start + emit->internal_imm_count++,
3036                          value[0], value[1], value[2], value[3]);
3037}
3038
3039
3040static boolean
3041make_immediate(struct svga_shader_emitter *emit,
3042               float a, float b, float c, float d,
3043               struct src_register *out )
3044{
3045   unsigned idx = emit->nr_hw_float_const++;
3046
3047   if (!emit_def_const( emit, SVGA3D_CONST_TYPE_FLOAT,
3048                        idx, a, b, c, d ))
3049      return FALSE;
3050
3051   *out = src_register( SVGA3DREG_CONST, idx );
3052
3053   return TRUE;
3054}
3055
3056
3057/**
3058 * Emit special VS instructions at top of shader.
3059 */
3060static boolean
3061emit_vs_preamble(struct svga_shader_emitter *emit)
3062{
3063   if (!emit->key.vs.need_prescale) {
3064      if (!make_immediate( emit, 0, 0, .5, .5,
3065                           &emit->imm_0055))
3066         return FALSE;
3067   }
3068
3069   return TRUE;
3070}
3071
3072
3073/**
3074 * Emit special PS instructions at top of shader.
3075 */
3076static boolean
3077emit_ps_preamble(struct svga_shader_emitter *emit)
3078{
3079   if (emit->ps_reads_pos && emit->info.reads_z) {
3080      /*
3081       * Assemble the position from various bits of inputs. Depth and W are
3082       * passed in a texcoord this is due to D3D's vPos not hold Z or W.
3083       * Also fixup the perspective interpolation.
3084       *
3085       * temp_pos.xy = vPos.xy
3086       * temp_pos.w = rcp(texcoord1.w);
3087       * temp_pos.z = texcoord1.z * temp_pos.w;
3088       */
3089      if (!submit_op1( emit,
3090                       inst_token(SVGA3DOP_MOV),
3091                       writemask( emit->ps_temp_pos, TGSI_WRITEMASK_XY ),
3092                       emit->ps_true_pos ))
3093         return FALSE;
3094
3095      if (!submit_op1( emit,
3096                       inst_token(SVGA3DOP_RCP),
3097                       writemask( emit->ps_temp_pos, TGSI_WRITEMASK_W ),
3098                       scalar( emit->ps_depth_pos, TGSI_SWIZZLE_W ) ))
3099         return FALSE;
3100
3101      if (!submit_op2( emit,
3102                       inst_token(SVGA3DOP_MUL),
3103                       writemask( emit->ps_temp_pos, TGSI_WRITEMASK_Z ),
3104                       scalar( emit->ps_depth_pos, TGSI_SWIZZLE_Z ),
3105                       scalar( src(emit->ps_temp_pos), TGSI_SWIZZLE_W ) ))
3106         return FALSE;
3107   }
3108
3109   return TRUE;
3110}
3111
3112
3113/**
3114 * Emit special PS instructions at end of shader.
3115 */
3116static boolean
3117emit_ps_postamble(struct svga_shader_emitter *emit)
3118{
3119   unsigned i;
3120
3121   /* PS oDepth is incredibly fragile and it's very hard to catch the
3122    * types of usage that break it during shader emit.  Easier just to
3123    * redirect the main program to a temporary and then only touch
3124    * oDepth with a hand-crafted MOV below.
3125    */
3126   if (SVGA3dShaderGetRegType(emit->true_pos.value) != 0) {
3127      if (!submit_op1( emit,
3128                       inst_token(SVGA3DOP_MOV),
3129                       emit->true_pos,
3130                       scalar(src(emit->temp_pos), TGSI_SWIZZLE_Z) ))
3131         return FALSE;
3132   }
3133
3134   for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) {
3135      if (SVGA3dShaderGetRegType(emit->true_color_output[i].value) != 0) {
3136         /* Potentially override output colors with white for XOR
3137          * logicop workaround.
3138          */
3139         if (emit->unit == PIPE_SHADER_FRAGMENT &&
3140             emit->key.fs.white_fragments) {
3141            struct src_register one = get_one_immediate(emit);
3142
3143            if (!submit_op1( emit,
3144                             inst_token(SVGA3DOP_MOV),
3145                             emit->true_color_output[i],
3146                             one ))
3147               return FALSE;
3148         }
3149         else if (emit->unit == PIPE_SHADER_FRAGMENT &&
3150                  i < emit->key.fs.write_color0_to_n_cbufs) {
3151            /* Write temp color output [0] to true output [i] */
3152            if (!submit_op1(emit, inst_token(SVGA3DOP_MOV),
3153                            emit->true_color_output[i],
3154                            src(emit->temp_color_output[0]))) {
3155               return FALSE;
3156            }
3157         }
3158         else {
3159            if (!submit_op1( emit,
3160                             inst_token(SVGA3DOP_MOV),
3161                             emit->true_color_output[i],
3162                             src(emit->temp_color_output[i]) ))
3163               return FALSE;
3164         }
3165      }
3166   }
3167
3168   return TRUE;
3169}
3170
3171
3172/**
3173 * Emit special VS instructions at end of shader.
3174 */
3175static boolean
3176emit_vs_postamble(struct svga_shader_emitter *emit)
3177{
3178   /* PSIZ output is incredibly fragile and it's very hard to catch
3179    * the types of usage that break it during shader emit.  Easier
3180    * just to redirect the main program to a temporary and then only
3181    * touch PSIZ with a hand-crafted MOV below.
3182    */
3183   if (SVGA3dShaderGetRegType(emit->true_psiz.value) != 0) {
3184      if (!submit_op1( emit,
3185                       inst_token(SVGA3DOP_MOV),
3186                       emit->true_psiz,
3187                       scalar(src(emit->temp_psiz), TGSI_SWIZZLE_X) ))
3188         return FALSE;
3189   }
3190
3191   /* Need to perform various manipulations on vertex position to cope
3192    * with the different GL and D3D clip spaces.
3193    */
3194   if (emit->key.vs.need_prescale) {
3195      SVGA3dShaderDestToken temp_pos = emit->temp_pos;
3196      SVGA3dShaderDestToken depth = emit->depth_pos;
3197      SVGA3dShaderDestToken pos = emit->true_pos;
3198      unsigned offset = emit->info.file_max[TGSI_FILE_CONSTANT] + 1;
3199      struct src_register prescale_scale = src_register( SVGA3DREG_CONST,
3200                                                         offset + 0 );
3201      struct src_register prescale_trans = src_register( SVGA3DREG_CONST,
3202                                                         offset + 1 );
3203
3204      if (!submit_op1( emit,
3205                       inst_token(SVGA3DOP_MOV),
3206                       writemask(depth, TGSI_WRITEMASK_W),
3207                       scalar(src(temp_pos), TGSI_SWIZZLE_W) ))
3208         return FALSE;
3209
3210      /* MUL temp_pos.xyz,    temp_pos,      prescale.scale
3211       * MAD result.position, temp_pos.wwww, prescale.trans, temp_pos
3212       *   --> Note that prescale.trans.w == 0
3213       */
3214      if (!submit_op2( emit,
3215                       inst_token(SVGA3DOP_MUL),
3216                       writemask(temp_pos, TGSI_WRITEMASK_XYZ),
3217                       src(temp_pos),
3218                       prescale_scale ))
3219         return FALSE;
3220
3221      if (!submit_op3( emit,
3222                       inst_token(SVGA3DOP_MAD),
3223                       pos,
3224                       swizzle(src(temp_pos), 3, 3, 3, 3),
3225                       prescale_trans,
3226                       src(temp_pos)))
3227         return FALSE;
3228
3229      /* Also write to depth value */
3230      if (!submit_op3( emit,
3231                       inst_token(SVGA3DOP_MAD),
3232                       writemask(depth, TGSI_WRITEMASK_Z),
3233                       swizzle(src(temp_pos), 3, 3, 3, 3),
3234                       prescale_trans,
3235                       src(temp_pos) ))
3236         return FALSE;
3237   }
3238   else {
3239      SVGA3dShaderDestToken temp_pos = emit->temp_pos;
3240      SVGA3dShaderDestToken depth = emit->depth_pos;
3241      SVGA3dShaderDestToken pos = emit->true_pos;
3242      struct src_register imm_0055 = emit->imm_0055;
3243
3244      /* Adjust GL clipping coordinate space to hardware (D3D-style):
3245       *
3246       * DP4 temp_pos.z, {0,0,.5,.5}, temp_pos
3247       * MOV result.position, temp_pos
3248       */
3249      if (!submit_op2( emit,
3250                       inst_token(SVGA3DOP_DP4),
3251                       writemask(temp_pos, TGSI_WRITEMASK_Z),
3252                       imm_0055,
3253                       src(temp_pos) ))
3254         return FALSE;
3255
3256      if (!submit_op1( emit,
3257                       inst_token(SVGA3DOP_MOV),
3258                       pos,
3259                       src(temp_pos) ))
3260         return FALSE;
3261
3262      /* Move the manipulated depth into the extra texcoord reg */
3263      if (!submit_op1( emit,
3264                       inst_token(SVGA3DOP_MOV),
3265                       writemask(depth, TGSI_WRITEMASK_ZW),
3266                       src(temp_pos) ))
3267         return FALSE;
3268   }
3269
3270   return TRUE;
3271}
3272
3273
3274/**
3275 * For the pixel shader: emit the code which chooses the front
3276 * or back face color depending on triangle orientation.
3277 * This happens at the top of the fragment shader.
3278 *
3279 *  0: IF VFACE :4
3280 *  1:   COLOR = FrontColor;
3281 *  2: ELSE
3282 *  3:   COLOR = BackColor;
3283 *  4: ENDIF
3284 */
3285static boolean
3286emit_light_twoside(struct svga_shader_emitter *emit)
3287{
3288   struct src_register vface, zero;
3289   struct src_register front[2];
3290   struct src_register back[2];
3291   SVGA3dShaderDestToken color[2];
3292   int count = emit->internal_color_count;
3293   unsigned i;
3294   SVGA3dShaderInstToken if_token;
3295
3296   if (count == 0)
3297      return TRUE;
3298
3299   vface = get_vface( emit );
3300   zero = get_zero_immediate(emit);
3301
3302   /* Can't use get_temp() to allocate the color reg as such
3303    * temporaries will be reclaimed after each instruction by the call
3304    * to reset_temp_regs().
3305    */
3306   for (i = 0; i < count; i++) {
3307      color[i] = dst_register( SVGA3DREG_TEMP, emit->nr_hw_temp++ );
3308      front[i] = emit->input_map[emit->internal_color_idx[i]];
3309
3310      /* Back is always the next input:
3311       */
3312      back[i] = front[i];
3313      back[i].base.num = front[i].base.num + 1;
3314
3315      /* Reassign the input_map to the actual front-face color:
3316       */
3317      emit->input_map[emit->internal_color_idx[i]] = src(color[i]);
3318   }
3319
3320   if_token = inst_token( SVGA3DOP_IFC );
3321
3322   if (emit->key.fs.front_ccw)
3323      if_token.control = SVGA3DOPCOMP_LT;
3324   else
3325      if_token.control = SVGA3DOPCOMP_GT;
3326
3327   if (!(emit_instruction( emit, if_token ) &&
3328         emit_src( emit, vface ) &&
3329         emit_src( emit, zero ) ))
3330      return FALSE;
3331
3332   for (i = 0; i < count; i++) {
3333      if (!submit_op1( emit, inst_token( SVGA3DOP_MOV ), color[i], front[i] ))
3334         return FALSE;
3335   }
3336
3337   if (!(emit_instruction( emit, inst_token( SVGA3DOP_ELSE))))
3338      return FALSE;
3339
3340   for (i = 0; i < count; i++) {
3341      if (!submit_op1( emit, inst_token( SVGA3DOP_MOV ), color[i], back[i] ))
3342         return FALSE;
3343   }
3344
3345   if (!emit_instruction( emit, inst_token( SVGA3DOP_ENDIF ) ))
3346      return FALSE;
3347
3348   return TRUE;
3349}
3350
3351
3352/**
3353 * Emit special setup code for the front/back face register in the FS.
3354 *  0: SETP_GT TEMP, VFACE, 0
3355 *  where TEMP is a fake frontface register
3356 */
3357static boolean
3358emit_frontface(struct svga_shader_emitter *emit)
3359{
3360   struct src_register vface;
3361   SVGA3dShaderDestToken temp;
3362   struct src_register pass, fail;
3363
3364   vface = get_vface( emit );
3365
3366   /* Can't use get_temp() to allocate the fake frontface reg as such
3367    * temporaries will be reclaimed after each instruction by the call
3368    * to reset_temp_regs().
3369    */
3370   temp = dst_register( SVGA3DREG_TEMP,
3371                        emit->nr_hw_temp++ );
3372
3373   if (emit->key.fs.front_ccw) {
3374      pass = get_zero_immediate(emit);
3375      fail = get_one_immediate(emit);
3376   } else {
3377      pass = get_one_immediate(emit);
3378      fail = get_zero_immediate(emit);
3379   }
3380
3381   if (!emit_conditional(emit, PIPE_FUNC_GREATER,
3382                         temp, vface, get_zero_immediate(emit),
3383                         pass, fail))
3384      return FALSE;
3385
3386   /* Reassign the input_map to the actual front-face color:
3387    */
3388   emit->input_map[emit->internal_frontface_idx] = src(temp);
3389
3390   return TRUE;
3391}
3392
3393
3394/**
3395 * Emit code to invert the T component of the incoming texture coordinate.
3396 * This is used for drawing point sprites when
3397 * pipe_rasterizer_state::sprite_coord_mode == PIPE_SPRITE_COORD_LOWER_LEFT.
3398 */
3399static boolean
3400emit_inverted_texcoords(struct svga_shader_emitter *emit)
3401{
3402   unsigned inverted_texcoords = emit->inverted_texcoords;
3403
3404   while (inverted_texcoords) {
3405      const unsigned unit = ffs(inverted_texcoords) - 1;
3406
3407      assert(emit->inverted_texcoords & (1 << unit));
3408
3409      assert(unit < ARRAY_SIZE(emit->ps_true_texcoord));
3410
3411      assert(unit < ARRAY_SIZE(emit->ps_inverted_texcoord_input));
3412
3413      assert(emit->ps_inverted_texcoord_input[unit]
3414             < ARRAY_SIZE(emit->input_map));
3415
3416      /* inverted = coord * (1, -1, 1, 1) + (0, 1, 0, 0) */
3417      if (!submit_op3(emit,
3418                      inst_token(SVGA3DOP_MAD),
3419                      dst(emit->ps_inverted_texcoord[unit]),
3420                      emit->ps_true_texcoord[unit],
3421                      get_immediate(emit, 1.0f, -1.0f, 1.0f, 1.0f),
3422                      get_immediate(emit, 0.0f, 1.0f, 0.0f, 0.0f)))
3423         return FALSE;
3424
3425      /* Reassign the input_map entry to the new texcoord register */
3426      emit->input_map[emit->ps_inverted_texcoord_input[unit]] =
3427         emit->ps_inverted_texcoord[unit];
3428
3429      inverted_texcoords &= ~(1 << unit);
3430   }
3431
3432   return TRUE;
3433}
3434
3435
3436/**
3437 * Emit code to adjust vertex shader inputs/attributes:
3438 * - Change range from [0,1] to [-1,1] (for normalized byte/short attribs).
3439 * - Set attrib W component = 1.
3440 */
3441static boolean
3442emit_adjusted_vertex_attribs(struct svga_shader_emitter *emit)
3443{
3444   unsigned adjust_mask = (emit->key.vs.adjust_attrib_range |
3445                           emit->key.vs.adjust_attrib_w_1);
3446
3447   while (adjust_mask) {
3448      /* Adjust vertex attrib range and/or set W component = 1 */
3449      const unsigned index = u_bit_scan(&adjust_mask);
3450      struct src_register tmp;
3451
3452      /* allocate a temp reg */
3453      tmp = src_register(SVGA3DREG_TEMP, emit->nr_hw_temp);
3454      emit->nr_hw_temp++;
3455
3456      if (emit->key.vs.adjust_attrib_range & (1 << index)) {
3457         /* The vertex input/attribute is supposed to be a signed value in
3458          * the range [-1,1] but we actually fetched/converted it to the
3459          * range [0,1].  This most likely happens when the app specifies a
3460          * signed byte attribute but we interpreted it as unsigned bytes.
3461          * See also svga_translate_vertex_format().
3462          *
3463          * Here, we emit some extra instructions to adjust
3464          * the attribute values from [0,1] to [-1,1].
3465          *
3466          * The adjustment we implement is:
3467          *   new_attrib = attrib * 2.0;
3468          *   if (attrib >= 0.5)
3469          *      new_attrib = new_attrib - 2.0;
3470          * This isn't exactly right (it's off by a bit or so) but close enough.
3471          */
3472         SVGA3dShaderDestToken pred_reg = dst_register(SVGA3DREG_PREDICATE, 0);
3473
3474         /* tmp = attrib * 2.0 */
3475         if (!submit_op2(emit,
3476                         inst_token(SVGA3DOP_MUL),
3477                         dst(tmp),
3478                         emit->input_map[index],
3479                         get_two_immediate(emit)))
3480            return FALSE;
3481
3482         /* pred = (attrib >= 0.5) */
3483         if (!submit_op2(emit,
3484                         inst_token_setp(SVGA3DOPCOMP_GE),
3485                         pred_reg,
3486                         emit->input_map[index],  /* vert attrib */
3487                         get_half_immediate(emit)))  /* 0.5 */
3488            return FALSE;
3489
3490         /* sub(pred) tmp, tmp, 2.0 */
3491         if (!submit_op3(emit,
3492                         inst_token_predicated(SVGA3DOP_SUB),
3493                         dst(tmp),
3494                         src(pred_reg),
3495                         tmp,
3496                         get_two_immediate(emit)))
3497            return FALSE;
3498      }
3499      else {
3500         /* just copy the vertex input attrib to the temp register */
3501         if (!submit_op1(emit,
3502                         inst_token(SVGA3DOP_MOV),
3503                         dst(tmp),
3504                         emit->input_map[index]))
3505            return FALSE;
3506      }
3507
3508      if (emit->key.vs.adjust_attrib_w_1 & (1 << index)) {
3509         /* move 1 into W position of tmp */
3510         if (!submit_op1(emit,
3511                         inst_token(SVGA3DOP_MOV),
3512                         writemask(dst(tmp), TGSI_WRITEMASK_W),
3513                         get_one_immediate(emit)))
3514            return FALSE;
3515      }
3516
3517      /* Reassign the input_map entry to the new tmp register */
3518      emit->input_map[index] = tmp;
3519   }
3520
3521   return TRUE;
3522}
3523
3524
3525/**
3526 * Determine if we need to create the "common" immediate value which is
3527 * used for generating useful vector constants such as {0,0,0,0} and
3528 * {1,1,1,1}.
3529 * We could just do this all the time except that we want to conserve
3530 * registers whenever possible.
3531 */
3532static boolean
3533needs_to_create_common_immediate(const struct svga_shader_emitter *emit)
3534{
3535   unsigned i;
3536
3537   if (emit->unit == PIPE_SHADER_FRAGMENT) {
3538      if (emit->key.fs.light_twoside)
3539         return TRUE;
3540
3541      if (emit->key.fs.white_fragments)
3542         return TRUE;
3543
3544      if (emit->emit_frontface)
3545         return TRUE;
3546
3547      if (emit->info.opcode_count[TGSI_OPCODE_DST] >= 1 ||
3548          emit->info.opcode_count[TGSI_OPCODE_SSG] >= 1 ||
3549          emit->info.opcode_count[TGSI_OPCODE_LIT] >= 1)
3550         return TRUE;
3551
3552      if (emit->inverted_texcoords)
3553         return TRUE;
3554
3555      /* look for any PIPE_SWIZZLE_0/ONE terms */
3556      for (i = 0; i < emit->key.num_textures; i++) {
3557         if (emit->key.tex[i].swizzle_r > PIPE_SWIZZLE_W ||
3558             emit->key.tex[i].swizzle_g > PIPE_SWIZZLE_W ||
3559             emit->key.tex[i].swizzle_b > PIPE_SWIZZLE_W ||
3560             emit->key.tex[i].swizzle_a > PIPE_SWIZZLE_W)
3561            return TRUE;
3562      }
3563
3564      for (i = 0; i < emit->key.num_textures; i++) {
3565         if (emit->key.tex[i].compare_mode
3566             == PIPE_TEX_COMPARE_R_TO_TEXTURE)
3567            return TRUE;
3568      }
3569   }
3570   else if (emit->unit == PIPE_SHADER_VERTEX) {
3571      if (emit->info.opcode_count[TGSI_OPCODE_CMP] >= 1)
3572         return TRUE;
3573      if (emit->key.vs.adjust_attrib_range ||
3574          emit->key.vs.adjust_attrib_w_1)
3575         return TRUE;
3576   }
3577
3578   if (emit->info.opcode_count[TGSI_OPCODE_IF] >= 1 ||
3579       emit->info.opcode_count[TGSI_OPCODE_BGNLOOP] >= 1 ||
3580       emit->info.opcode_count[TGSI_OPCODE_DDX] >= 1 ||
3581       emit->info.opcode_count[TGSI_OPCODE_DDY] >= 1 ||
3582       emit->info.opcode_count[TGSI_OPCODE_ROUND] >= 1 ||
3583       emit->info.opcode_count[TGSI_OPCODE_SGE] >= 1 ||
3584       emit->info.opcode_count[TGSI_OPCODE_SGT] >= 1 ||
3585       emit->info.opcode_count[TGSI_OPCODE_SLE] >= 1 ||
3586       emit->info.opcode_count[TGSI_OPCODE_SLT] >= 1 ||
3587       emit->info.opcode_count[TGSI_OPCODE_SNE] >= 1 ||
3588       emit->info.opcode_count[TGSI_OPCODE_SEQ] >= 1 ||
3589       emit->info.opcode_count[TGSI_OPCODE_EXP] >= 1 ||
3590       emit->info.opcode_count[TGSI_OPCODE_LOG] >= 1 ||
3591       emit->info.opcode_count[TGSI_OPCODE_KILL] >= 1 ||
3592       emit->info.opcode_count[TGSI_OPCODE_SQRT] >= 1)
3593      return TRUE;
3594
3595   return FALSE;
3596}
3597
3598
3599/**
3600 * Do we need to create a looping constant?
3601 */
3602static boolean
3603needs_to_create_loop_const(const struct svga_shader_emitter *emit)
3604{
3605   return (emit->info.opcode_count[TGSI_OPCODE_BGNLOOP] >= 1);
3606}
3607
3608
3609static boolean
3610needs_to_create_arl_consts(const struct svga_shader_emitter *emit)
3611{
3612   return (emit->num_arl_consts > 0);
3613}
3614
3615
3616static boolean
3617pre_parse_add_indirect( struct svga_shader_emitter *emit,
3618                        int num, int current_arl)
3619{
3620   unsigned i;
3621   assert(num < 0);
3622
3623   for (i = 0; i < emit->num_arl_consts; ++i) {
3624      if (emit->arl_consts[i].arl_num == current_arl)
3625         break;
3626   }
3627   /* new entry */
3628   if (emit->num_arl_consts == i) {
3629      ++emit->num_arl_consts;
3630   }
3631   emit->arl_consts[i].number = (emit->arl_consts[i].number > num) ?
3632                                num :
3633                                emit->arl_consts[i].number;
3634   emit->arl_consts[i].arl_num = current_arl;
3635   return TRUE;
3636}
3637
3638
3639static boolean
3640pre_parse_instruction( struct svga_shader_emitter *emit,
3641                       const struct tgsi_full_instruction *insn,
3642                       int current_arl)
3643{
3644   if (insn->Src[0].Register.Indirect &&
3645       insn->Src[0].Indirect.File == TGSI_FILE_ADDRESS) {
3646      const struct tgsi_full_src_register *reg = &insn->Src[0];
3647      if (reg->Register.Index < 0) {
3648         pre_parse_add_indirect(emit, reg->Register.Index, current_arl);
3649      }
3650   }
3651
3652   if (insn->Src[1].Register.Indirect &&
3653       insn->Src[1].Indirect.File == TGSI_FILE_ADDRESS) {
3654      const struct tgsi_full_src_register *reg = &insn->Src[1];
3655      if (reg->Register.Index < 0) {
3656         pre_parse_add_indirect(emit, reg->Register.Index, current_arl);
3657      }
3658   }
3659
3660   if (insn->Src[2].Register.Indirect &&
3661       insn->Src[2].Indirect.File == TGSI_FILE_ADDRESS) {
3662      const struct tgsi_full_src_register *reg = &insn->Src[2];
3663      if (reg->Register.Index < 0) {
3664         pre_parse_add_indirect(emit, reg->Register.Index, current_arl);
3665      }
3666   }
3667
3668   return TRUE;
3669}
3670
3671
3672static boolean
3673pre_parse_tokens( struct svga_shader_emitter *emit,
3674                  const struct tgsi_token *tokens )
3675{
3676   struct tgsi_parse_context parse;
3677   int current_arl = 0;
3678
3679   tgsi_parse_init( &parse, tokens );
3680
3681   while (!tgsi_parse_end_of_tokens( &parse )) {
3682      tgsi_parse_token( &parse );
3683      switch (parse.FullToken.Token.Type) {
3684      case TGSI_TOKEN_TYPE_IMMEDIATE:
3685      case TGSI_TOKEN_TYPE_DECLARATION:
3686         break;
3687      case TGSI_TOKEN_TYPE_INSTRUCTION:
3688         if (parse.FullToken.FullInstruction.Instruction.Opcode ==
3689             TGSI_OPCODE_ARL) {
3690            ++current_arl;
3691         }
3692         if (!pre_parse_instruction( emit, &parse.FullToken.FullInstruction,
3693                                     current_arl ))
3694            return FALSE;
3695         break;
3696      default:
3697         break;
3698      }
3699
3700   }
3701   return TRUE;
3702}
3703
3704
3705static boolean
3706svga_shader_emit_helpers(struct svga_shader_emitter *emit)
3707{
3708   if (needs_to_create_common_immediate( emit )) {
3709      create_common_immediate( emit );
3710   }
3711   if (needs_to_create_loop_const( emit )) {
3712      create_loop_const( emit );
3713   }
3714   if (needs_to_create_arl_consts( emit )) {
3715      create_arl_consts( emit );
3716   }
3717
3718   if (emit->unit == PIPE_SHADER_FRAGMENT) {
3719      if (!svga_shader_emit_samplers_decl( emit ))
3720         return FALSE;
3721
3722      if (!emit_ps_preamble( emit ))
3723         return FALSE;
3724
3725      if (emit->key.fs.light_twoside) {
3726         if (!emit_light_twoside( emit ))
3727            return FALSE;
3728      }
3729      if (emit->emit_frontface) {
3730         if (!emit_frontface( emit ))
3731            return FALSE;
3732      }
3733      if (emit->inverted_texcoords) {
3734         if (!emit_inverted_texcoords( emit ))
3735            return FALSE;
3736      }
3737   }
3738   else {
3739      assert(emit->unit == PIPE_SHADER_VERTEX);
3740      if (emit->key.vs.adjust_attrib_range) {
3741         if (!emit_adjusted_vertex_attribs(emit) ||
3742             emit->key.vs.adjust_attrib_w_1) {
3743            return FALSE;
3744         }
3745      }
3746   }
3747
3748   return TRUE;
3749}
3750
3751
3752/**
3753 * This is the main entrypoint into the TGSI instruction translater.
3754 * Translate TGSI shader tokens into an SVGA shader.
3755 */
3756boolean
3757svga_shader_emit_instructions(struct svga_shader_emitter *emit,
3758                              const struct tgsi_token *tokens)
3759{
3760   struct tgsi_parse_context parse;
3761   const struct tgsi_token *new_tokens = NULL;
3762   boolean ret = TRUE;
3763   boolean helpers_emitted = FALSE;
3764   unsigned line_nr = 0;
3765
3766   if (emit->unit == PIPE_SHADER_FRAGMENT && emit->key.fs.pstipple) {
3767      unsigned unit;
3768
3769      new_tokens = util_pstipple_create_fragment_shader(tokens, &unit, 0,
3770                                                        TGSI_FILE_INPUT);
3771
3772      if (new_tokens) {
3773         /* Setup texture state for stipple */
3774         emit->sampler_target[unit] = TGSI_TEXTURE_2D;
3775         emit->key.tex[unit].swizzle_r = TGSI_SWIZZLE_X;
3776         emit->key.tex[unit].swizzle_g = TGSI_SWIZZLE_Y;
3777         emit->key.tex[unit].swizzle_b = TGSI_SWIZZLE_Z;
3778         emit->key.tex[unit].swizzle_a = TGSI_SWIZZLE_W;
3779
3780         emit->pstipple_sampler_unit = unit;
3781
3782         tokens = new_tokens;
3783      }
3784   }
3785
3786   tgsi_parse_init( &parse, tokens );
3787   emit->internal_imm_count = 0;
3788
3789   if (emit->unit == PIPE_SHADER_VERTEX) {
3790      ret = emit_vs_preamble( emit );
3791      if (!ret)
3792         goto done;
3793   }
3794
3795   pre_parse_tokens(emit, tokens);
3796
3797   while (!tgsi_parse_end_of_tokens( &parse )) {
3798      tgsi_parse_token( &parse );
3799
3800      switch (parse.FullToken.Token.Type) {
3801      case TGSI_TOKEN_TYPE_IMMEDIATE:
3802         ret = svga_emit_immediate( emit, &parse.FullToken.FullImmediate );
3803         if (!ret)
3804            goto done;
3805         break;
3806
3807      case TGSI_TOKEN_TYPE_DECLARATION:
3808         ret = svga_translate_decl_sm30( emit, &parse.FullToken.FullDeclaration );
3809         if (!ret)
3810            goto done;
3811         break;
3812
3813      case TGSI_TOKEN_TYPE_INSTRUCTION:
3814         if (!helpers_emitted) {
3815            if (!svga_shader_emit_helpers( emit ))
3816               goto done;
3817            helpers_emitted = TRUE;
3818         }
3819         ret = svga_emit_instruction( emit,
3820                                      line_nr++,
3821                                      &parse.FullToken.FullInstruction );
3822         if (!ret)
3823            goto done;
3824         break;
3825      default:
3826         break;
3827      }
3828
3829      reset_temp_regs( emit );
3830   }
3831
3832   /* Need to terminate the current subroutine.  Note that the
3833    * hardware doesn't tolerate shaders without sub-routines
3834    * terminating with RET+END.
3835    */
3836   if (!emit->in_main_func) {
3837      ret = emit_instruction( emit, inst_token( SVGA3DOP_RET ) );
3838      if (!ret)
3839         goto done;
3840   }
3841
3842   assert(emit->dynamic_branching_level == 0);
3843
3844   /* Need to terminate the whole shader:
3845    */
3846   ret = emit_instruction( emit, inst_token( SVGA3DOP_END ) );
3847   if (!ret)
3848      goto done;
3849
3850done:
3851   tgsi_parse_free( &parse );
3852   if (new_tokens) {
3853      tgsi_free_tokens(new_tokens);
3854   }
3855
3856   return ret;
3857}
3858