1 /**************************************************************************
3 * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **************************************************************************/
32 #include "i915_context.h"
35 #include "pipe/p_shader_tokens.h"
36 #include "util/u_math.h"
37 #include "util/u_memory.h"
38 #include "util/u_string.h"
39 #include "tgsi/tgsi_parse.h"
40 #include "tgsi/tgsi_dump.h"
42 #include "draw/draw_vertex.h"
45 #define M_PI 3.14159265358979323846
49 * Simple pass-through fragment shader to use when we don't have
50 * a real shader (or it fails to compile for some reason).
52 static unsigned passthrough[] =
54 _3DSTATE_PIXEL_SHADER_PROGRAM | ((2*3)-1),
56 /* declare input color:
59 (REG_TYPE_T << D0_TYPE_SHIFT) |
60 (T_DIFFUSE << D0_NR_SHIFT) |
65 /* move to output color:
68 (REG_TYPE_OC << A0_DEST_TYPE_SHIFT) |
70 (REG_TYPE_T << A0_SRC0_TYPE_SHIFT) |
71 (T_DIFFUSE << A0_SRC0_NR_SHIFT)),
72 0x01230000, /* .xyzw */
77 /* 1, -1/3!, 1/5!, -1/7! */
78 static const float scs_sin_constants[4] = { 1.0,
80 1.0f / (5 * 4 * 3 * 2 * 1),
81 -1.0f / (7 * 6 * 5 * 4 * 3 * 2 * 1)
84 /* 1, -1/2!, 1/4!, -1/6! */
85 static const float scs_cos_constants[4] = { 1.0,
87 1.0f / (4 * 3 * 2 * 1),
88 -1.0f / (6 * 5 * 4 * 3 * 2 * 1)
91 /* 2*pi, -(2*pi)^3/3!, (2*pi)^5/5!, -(2*pi)^7/7! */
92 static const float sin_constants[4] = { 2.0 * M_PI,
93 -8.0f * M_PI * M_PI * M_PI / (3 * 2 * 1),
94 32.0f * M_PI * M_PI * M_PI * M_PI * M_PI / (5 * 4 * 3 * 2 * 1),
95 -128.0f * M_PI * M_PI * M_PI * M_PI * M_PI * M_PI * M_PI / (7 * 6 * 5 * 4 * 3 * 2 * 1)
98 /* 1, -(2*pi)^2/2!, (2*pi)^4/4!, -(2*pi)^6/6! */
99 static const float cos_constants[4] = { 1.0,
100 -4.0f * M_PI * M_PI / (2 * 1),
101 16.0f * M_PI * M_PI * M_PI * M_PI / (4 * 3 * 2 * 1),
102 -64.0f * M_PI * M_PI * M_PI * M_PI * M_PI * M_PI / (6 * 5 * 4 * 3 * 2 * 1)
108 * component-wise negation of ureg
111 negate(int reg, int x, int y, int z, int w)
113 /* Another neat thing about the UREG representation */
114 return reg ^ (((x & 1) << UREG_CHANNEL_X_NEGATE_SHIFT) |
115 ((y & 1) << UREG_CHANNEL_Y_NEGATE_SHIFT) |
116 ((z & 1) << UREG_CHANNEL_Z_NEGATE_SHIFT) |
117 ((w & 1) << UREG_CHANNEL_W_NEGATE_SHIFT));
122 * In the event of a translation failure, we'll generate a simple color
123 * pass-through program.
126 i915_use_passthrough_shader(struct i915_fragment_shader *fs)
128 fs->program = (uint *) MALLOC(sizeof(passthrough));
130 memcpy(fs->program, passthrough, sizeof(passthrough));
131 fs->program_len = Elements(passthrough);
133 fs->num_constants = 0;
138 i915_program_error(struct i915_fp_compile *p, const char *msg, ...)
143 debug_printf("i915_program_error: ");
144 va_start( args, msg );
145 util_vsnprintf( buffer, sizeof(buffer), msg, args );
147 debug_printf("%s", buffer);
153 static uint get_mapping(struct i915_fragment_shader* fs, int unit)
156 for (i = 0; i < I915_TEX_UNITS; i++)
158 if (fs->generic_mapping[i] == -1) {
159 fs->generic_mapping[i] = unit;
162 if (fs->generic_mapping[i] == unit)
165 debug_printf("Exceeded max generics\n");
170 * Construct a ureg for the given source register. Will emit
171 * constants, apply swizzling and negation as needed.
174 src_vector(struct i915_fp_compile *p,
175 const struct i915_full_src_register *source,
176 struct i915_fragment_shader* fs)
178 uint index = source->Register.Index;
179 uint src = 0, sem_name, sem_ind;
181 switch (source->Register.File) {
182 case TGSI_FILE_TEMPORARY:
183 if (source->Register.Index >= I915_MAX_TEMPORARY) {
184 i915_program_error(p, "Exceeded max temporary reg");
187 src = UREG(REG_TYPE_R, index);
189 case TGSI_FILE_INPUT:
190 /* XXX: Packing COL1, FOGC into a single attribute works for
191 * texenv programs, but will fail for real fragment programs
192 * that use these attributes and expect them to be a full 4
193 * components wide. Could use a texcoord to pass these
194 * attributes if necessary, but that won't work in the general
197 * We also use a texture coordinate to pass wpos when possible.
200 sem_name = p->shader->info.input_semantic_name[index];
201 sem_ind = p->shader->info.input_semantic_index[index];
204 case TGSI_SEMANTIC_POSITION:
207 int real_tex_unit = get_mapping(fs, I915_SEMANTIC_POS);
208 src = i915_emit_decl(p, REG_TYPE_T, T_TEX0 + real_tex_unit, D0_CHANNEL_ALL);
211 case TGSI_SEMANTIC_COLOR:
213 src = i915_emit_decl(p, REG_TYPE_T, T_DIFFUSE, D0_CHANNEL_ALL);
216 /* secondary color */
217 assert(sem_ind == 1);
218 src = i915_emit_decl(p, REG_TYPE_T, T_SPECULAR, D0_CHANNEL_XYZ);
219 src = swizzle(src, X, Y, Z, ONE);
222 case TGSI_SEMANTIC_FOG:
223 src = i915_emit_decl(p, REG_TYPE_T, T_FOG_W, D0_CHANNEL_W);
224 src = swizzle(src, W, W, W, W);
226 case TGSI_SEMANTIC_GENERIC:
228 int real_tex_unit = get_mapping(fs, sem_ind);
229 src = i915_emit_decl(p, REG_TYPE_T, T_TEX0 + real_tex_unit, D0_CHANNEL_ALL);
232 case TGSI_SEMANTIC_FACE:
234 /* for back/front faces */
235 int real_tex_unit = get_mapping(fs, I915_SEMANTIC_FACE);
236 src = i915_emit_decl(p, REG_TYPE_T, T_TEX0 + real_tex_unit, D0_CHANNEL_X);
240 i915_program_error(p, "Bad source->Index");
245 case TGSI_FILE_IMMEDIATE:
246 assert(index < p->num_immediates);
247 index = p->immediates_map[index];
249 case TGSI_FILE_CONSTANT:
250 src = UREG(REG_TYPE_CONST, index);
254 i915_program_error(p, "Bad source->File");
259 source->Register.SwizzleX,
260 source->Register.SwizzleY,
261 source->Register.SwizzleZ,
262 source->Register.SwizzleW);
264 /* There's both negate-all-components and per-component negation.
265 * Try to handle both here.
268 int n = source->Register.Negate;
269 src = negate(src, n, n, n, n);
274 /* XXX assertions disabled to allow arbfplight.c to run */
275 /* XXX enable these assertions, or fix things */
276 assert(!source->Register.Absolute);
278 if (source->Register.Absolute)
279 debug_printf("Unhandled absolute value\n");
286 * Construct a ureg for a destination register.
289 get_result_vector(struct i915_fp_compile *p,
290 const struct i915_full_dst_register *dest)
292 switch (dest->Register.File) {
293 case TGSI_FILE_OUTPUT:
295 uint sem_name = p->shader->info.output_semantic_name[dest->Register.Index];
297 case TGSI_SEMANTIC_POSITION:
298 return UREG(REG_TYPE_OD, 0);
299 case TGSI_SEMANTIC_COLOR:
300 return UREG(REG_TYPE_OC, 0);
302 i915_program_error(p, "Bad inst->DstReg.Index/semantics");
306 case TGSI_FILE_TEMPORARY:
307 return UREG(REG_TYPE_R, dest->Register.Index);
309 i915_program_error(p, "Bad inst->DstReg.File");
316 * Compute flags for saturation and writemask.
319 get_result_flags(const struct i915_full_instruction *inst)
322 = inst->Dst[0].Register.WriteMask;
325 if (inst->Instruction.Saturate == TGSI_SAT_ZERO_ONE)
326 flags |= A0_DEST_SATURATE;
328 if (writeMask & TGSI_WRITEMASK_X)
329 flags |= A0_DEST_CHANNEL_X;
330 if (writeMask & TGSI_WRITEMASK_Y)
331 flags |= A0_DEST_CHANNEL_Y;
332 if (writeMask & TGSI_WRITEMASK_Z)
333 flags |= A0_DEST_CHANNEL_Z;
334 if (writeMask & TGSI_WRITEMASK_W)
335 flags |= A0_DEST_CHANNEL_W;
342 * Convert TGSI_TEXTURE_x token to DO_SAMPLE_TYPE_x token
345 translate_tex_src_target(struct i915_fp_compile *p, uint tex)
348 case TGSI_TEXTURE_SHADOW1D:
350 case TGSI_TEXTURE_1D:
351 return D0_SAMPLE_TYPE_2D;
353 case TGSI_TEXTURE_SHADOW2D:
355 case TGSI_TEXTURE_2D:
356 return D0_SAMPLE_TYPE_2D;
358 case TGSI_TEXTURE_SHADOWRECT:
360 case TGSI_TEXTURE_RECT:
361 return D0_SAMPLE_TYPE_2D;
363 case TGSI_TEXTURE_3D:
364 return D0_SAMPLE_TYPE_VOLUME;
366 case TGSI_TEXTURE_CUBE:
367 return D0_SAMPLE_TYPE_CUBE;
370 i915_program_error(p, "TexSrc type");
376 * Convert TGSI_TEXTURE_x token to DO_SAMPLE_TYPE_x token
379 texture_num_coords(struct i915_fp_compile *p, uint tex)
382 case TGSI_TEXTURE_SHADOW1D:
383 case TGSI_TEXTURE_1D:
386 case TGSI_TEXTURE_SHADOW2D:
387 case TGSI_TEXTURE_2D:
388 case TGSI_TEXTURE_SHADOWRECT:
389 case TGSI_TEXTURE_RECT:
392 case TGSI_TEXTURE_3D:
393 case TGSI_TEXTURE_CUBE:
397 i915_program_error(p, "Num coords");
404 * Generate texel lookup instruction.
407 emit_tex(struct i915_fp_compile *p,
408 const struct i915_full_instruction *inst,
410 struct i915_fragment_shader* fs)
412 uint texture = inst->Texture.Texture;
413 uint unit = inst->Src[1].Register.Index;
414 uint tex = translate_tex_src_target( p, texture );
415 uint sampler = i915_emit_decl(p, REG_TYPE_S, unit, tex);
416 uint coord = src_vector( p, &inst->Src[0], fs);
419 get_result_vector( p, &inst->Dst[0] ),
420 get_result_flags( inst ),
424 texture_num_coords(p, texture) );
429 * Generate a simple arithmetic instruction
430 * \param opcode the i915 opcode
431 * \param numArgs the number of input/src arguments
434 emit_simple_arith(struct i915_fp_compile *p,
435 const struct i915_full_instruction *inst,
436 uint opcode, uint numArgs,
437 struct i915_fragment_shader* fs)
439 uint arg1, arg2, arg3;
441 assert(numArgs <= 3);
443 arg1 = (numArgs < 1) ? 0 : src_vector( p, &inst->Src[0], fs );
444 arg2 = (numArgs < 2) ? 0 : src_vector( p, &inst->Src[1], fs );
445 arg3 = (numArgs < 3) ? 0 : src_vector( p, &inst->Src[2], fs );
449 get_result_vector( p, &inst->Dst[0]),
450 get_result_flags( inst ), 0,
457 /** As above, but swap the first two src regs */
459 emit_simple_arith_swap2(struct i915_fp_compile *p,
460 const struct i915_full_instruction *inst,
461 uint opcode, uint numArgs,
462 struct i915_fragment_shader* fs)
464 struct i915_full_instruction inst2;
466 assert(numArgs == 2);
468 /* transpose first two registers */
470 inst2.Src[0] = inst->Src[1];
471 inst2.Src[1] = inst->Src[0];
473 emit_simple_arith(p, &inst2, opcode, numArgs, fs);
477 * Translate TGSI instruction to i915 instruction.
481 * DDX, DDY -- return 0
482 * SIN, COS -- could use another taylor step?
483 * LIT -- results seem a little different to sw mesa
484 * LOG -- different to mesa on negative numbers, but this is conformant.
487 i915_translate_instruction(struct i915_fp_compile *p,
488 const struct i915_full_instruction *inst,
489 struct i915_fragment_shader *fs)
492 uint src0, src1, src2, flags;
495 switch (inst->Instruction.Opcode) {
496 case TGSI_OPCODE_ABS:
497 src0 = src_vector(p, &inst->Src[0], fs);
500 get_result_vector(p, &inst->Dst[0]),
501 get_result_flags(inst), 0,
502 src0, negate(src0, 1, 1, 1, 1), 0);
505 case TGSI_OPCODE_ADD:
506 emit_simple_arith(p, inst, A0_ADD, 2, fs);
509 case TGSI_OPCODE_CMP:
510 src0 = src_vector(p, &inst->Src[0], fs);
511 src1 = src_vector(p, &inst->Src[1], fs);
512 src2 = src_vector(p, &inst->Src[2], fs);
513 i915_emit_arith(p, A0_CMP,
514 get_result_vector(p, &inst->Dst[0]),
515 get_result_flags(inst),
516 0, src0, src2, src1); /* NOTE: order of src2, src1 */
519 case TGSI_OPCODE_COS:
520 src0 = src_vector(p, &inst->Src[0], fs);
521 tmp = i915_get_utemp(p);
525 tmp, A0_DEST_CHANNEL_X, 0,
526 src0, i915_emit_const1f(p, 1.0f / (float) (M_PI * 2.0)), 0);
528 i915_emit_arith(p, A0_MOD, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
531 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
532 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, 1
533 * t0 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1
534 * result = DP4 t0, cos_constants
538 tmp, A0_DEST_CHANNEL_XY, 0,
539 swizzle(tmp, X, X, ONE, ONE),
540 swizzle(tmp, X, ONE, ONE, ONE), 0);
544 tmp, A0_DEST_CHANNEL_XYZ, 0,
545 swizzle(tmp, X, Y, X, ONE),
546 swizzle(tmp, X, X, ONE, ONE), 0);
550 tmp, A0_DEST_CHANNEL_XYZ, 0,
551 swizzle(tmp, X, X, Z, ONE),
552 swizzle(tmp, Z, ONE, ONE, ONE), 0);
556 get_result_vector(p, &inst->Dst[0]),
557 get_result_flags(inst), 0,
558 swizzle(tmp, ONE, Z, Y, X),
559 i915_emit_const4fv(p, cos_constants), 0);
562 case TGSI_OPCODE_DDX:
563 case TGSI_OPCODE_DDY:
564 /* XXX We just output 0 here */
565 debug_printf("Punting DDX/DDX\n");
566 src0 = get_result_vector(p, &inst->Dst[0]);
569 get_result_vector(p, &inst->Dst[0]),
570 get_result_flags(inst), 0,
571 swizzle(src0, ZERO, ZERO, ZERO, ZERO), 0, 0);
574 case TGSI_OPCODE_DP2:
575 src0 = src_vector(p, &inst->Src[0], fs);
576 src1 = src_vector(p, &inst->Src[1], fs);
580 get_result_vector(p, &inst->Dst[0]),
581 get_result_flags(inst), 0,
582 swizzle(src0, X, Y, ZERO, ZERO), src1, 0);
585 case TGSI_OPCODE_DP3:
586 emit_simple_arith(p, inst, A0_DP3, 2, fs);
589 case TGSI_OPCODE_DP4:
590 emit_simple_arith(p, inst, A0_DP4, 2, fs);
593 case TGSI_OPCODE_DPH:
594 src0 = src_vector(p, &inst->Src[0], fs);
595 src1 = src_vector(p, &inst->Src[1], fs);
599 get_result_vector(p, &inst->Dst[0]),
600 get_result_flags(inst), 0,
601 swizzle(src0, X, Y, Z, ONE), src1, 0);
604 case TGSI_OPCODE_DST:
605 src0 = src_vector(p, &inst->Src[0], fs);
606 src1 = src_vector(p, &inst->Src[1], fs);
608 /* result[0] = 1 * 1;
609 * result[1] = a[1] * b[1];
610 * result[2] = a[2] * 1;
611 * result[3] = 1 * b[3];
615 get_result_vector(p, &inst->Dst[0]),
616 get_result_flags(inst), 0,
617 swizzle(src0, ONE, Y, Z, ONE),
618 swizzle(src1, ONE, Y, ONE, W), 0);
621 case TGSI_OPCODE_END:
625 case TGSI_OPCODE_EX2:
626 src0 = src_vector(p, &inst->Src[0], fs);
630 get_result_vector(p, &inst->Dst[0]),
631 get_result_flags(inst), 0,
632 swizzle(src0, X, X, X, X), 0, 0);
635 case TGSI_OPCODE_FLR:
636 emit_simple_arith(p, inst, A0_FLR, 1, fs);
639 case TGSI_OPCODE_FRC:
640 emit_simple_arith(p, inst, A0_FRC, 1, fs);
643 case TGSI_OPCODE_KIL:
644 /* kill if src[0].x < 0 || src[0].y < 0 ... */
645 src0 = src_vector(p, &inst->Src[0], fs);
646 tmp = i915_get_utemp(p);
649 tmp, /* dest reg: a dummy reg */
650 A0_DEST_CHANNEL_ALL, /* dest writemask */
653 T0_TEXKILL, /* opcode */
657 case TGSI_OPCODE_KILP:
658 assert(0); /* not tested yet */
661 case TGSI_OPCODE_LG2:
662 src0 = src_vector(p, &inst->Src[0], fs);
666 get_result_vector(p, &inst->Dst[0]),
667 get_result_flags(inst), 0,
668 swizzle(src0, X, X, X, X), 0, 0);
671 case TGSI_OPCODE_LIT:
672 src0 = src_vector(p, &inst->Src[0], fs);
673 tmp = i915_get_utemp(p);
675 /* tmp = max( a.xyzw, a.00zw )
676 * XXX: Clamp tmp.w to -128..128
678 * tmp.y = tmp.w * tmp.y
680 * result = cmp (a.11-x1, a.1x01, a.1xy1 )
682 i915_emit_arith(p, A0_MAX, tmp, A0_DEST_CHANNEL_ALL, 0,
683 src0, swizzle(src0, ZERO, ZERO, Z, W), 0);
685 i915_emit_arith(p, A0_LOG, tmp, A0_DEST_CHANNEL_Y, 0,
686 swizzle(tmp, Y, Y, Y, Y), 0, 0);
688 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_Y, 0,
689 swizzle(tmp, ZERO, Y, ZERO, ZERO),
690 swizzle(tmp, ZERO, W, ZERO, ZERO), 0);
692 i915_emit_arith(p, A0_EXP, tmp, A0_DEST_CHANNEL_Y, 0,
693 swizzle(tmp, Y, Y, Y, Y), 0, 0);
695 i915_emit_arith(p, A0_CMP,
696 get_result_vector(p, &inst->Dst[0]),
697 get_result_flags(inst), 0,
698 negate(swizzle(tmp, ONE, ONE, X, ONE), 0, 0, 1, 0),
699 swizzle(tmp, ONE, X, ZERO, ONE),
700 swizzle(tmp, ONE, X, Y, ONE));
704 case TGSI_OPCODE_LRP:
705 src0 = src_vector(p, &inst->Src[0], fs);
706 src1 = src_vector(p, &inst->Src[1], fs);
707 src2 = src_vector(p, &inst->Src[2], fs);
708 flags = get_result_flags(inst);
709 tmp = i915_get_utemp(p);
716 * result = (-c)*a + tmp
718 i915_emit_arith(p, A0_MAD, tmp,
719 flags & A0_DEST_CHANNEL_ALL, 0, src1, src0, src2);
721 i915_emit_arith(p, A0_MAD,
722 get_result_vector(p, &inst->Dst[0]),
723 flags, 0, negate(src2, 1, 1, 1, 1), src0, tmp);
726 case TGSI_OPCODE_MAD:
727 emit_simple_arith(p, inst, A0_MAD, 3, fs);
730 case TGSI_OPCODE_MAX:
731 emit_simple_arith(p, inst, A0_MAX, 2, fs);
734 case TGSI_OPCODE_MIN:
735 src0 = src_vector(p, &inst->Src[0], fs);
736 src1 = src_vector(p, &inst->Src[1], fs);
737 tmp = i915_get_utemp(p);
738 flags = get_result_flags(inst);
742 tmp, flags & A0_DEST_CHANNEL_ALL, 0,
743 negate(src0, 1, 1, 1, 1),
744 negate(src1, 1, 1, 1, 1), 0);
748 get_result_vector(p, &inst->Dst[0]),
749 flags, 0, negate(tmp, 1, 1, 1, 1), 0, 0);
752 case TGSI_OPCODE_MOV:
753 emit_simple_arith(p, inst, A0_MOV, 1, fs);
756 case TGSI_OPCODE_MUL:
757 emit_simple_arith(p, inst, A0_MUL, 2, fs);
760 case TGSI_OPCODE_NOP:
763 case TGSI_OPCODE_POW:
764 src0 = src_vector(p, &inst->Src[0], fs);
765 src1 = src_vector(p, &inst->Src[1], fs);
766 tmp = i915_get_utemp(p);
767 flags = get_result_flags(inst);
769 /* XXX: masking on intermediate values, here and elsewhere.
773 tmp, A0_DEST_CHANNEL_X, 0,
774 swizzle(src0, X, X, X, X), 0, 0);
776 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_X, 0, tmp, src1, 0);
780 get_result_vector(p, &inst->Dst[0]),
781 flags, 0, swizzle(tmp, X, X, X, X), 0, 0);
784 case TGSI_OPCODE_RET:
788 case TGSI_OPCODE_RCP:
789 src0 = src_vector(p, &inst->Src[0], fs);
793 get_result_vector(p, &inst->Dst[0]),
794 get_result_flags(inst), 0,
795 swizzle(src0, X, X, X, X), 0, 0);
798 case TGSI_OPCODE_RSQ:
799 src0 = src_vector(p, &inst->Src[0], fs);
803 get_result_vector(p, &inst->Dst[0]),
804 get_result_flags(inst), 0,
805 swizzle(src0, X, X, X, X), 0, 0);
808 case TGSI_OPCODE_SCS:
809 src0 = src_vector(p, &inst->Src[0], fs);
810 tmp = i915_get_utemp(p);
813 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
814 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
815 * t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x
816 * scs.x = DP4 t1, scs_sin_constants
817 * t1 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1
818 * scs.y = DP4 t1, scs_cos_constants
822 tmp, A0_DEST_CHANNEL_XY, 0,
823 swizzle(src0, X, X, ONE, ONE),
824 swizzle(src0, X, ONE, ONE, ONE), 0);
828 tmp, A0_DEST_CHANNEL_ALL, 0,
829 swizzle(tmp, X, Y, X, Y),
830 swizzle(tmp, X, X, ONE, ONE), 0);
832 writemask = inst->Dst[0].Register.WriteMask;
834 if (writemask & TGSI_WRITEMASK_Y) {
837 if (writemask & TGSI_WRITEMASK_X)
838 tmp1 = i915_get_utemp(p);
844 tmp1, A0_DEST_CHANNEL_ALL, 0,
845 swizzle(tmp, X, Y, Y, W),
846 swizzle(tmp, X, Z, ONE, ONE), 0);
850 get_result_vector(p, &inst->Dst[0]),
851 A0_DEST_CHANNEL_Y, 0,
852 swizzle(tmp1, W, Z, Y, X),
853 i915_emit_const4fv(p, scs_sin_constants), 0);
856 if (writemask & TGSI_WRITEMASK_X) {
859 tmp, A0_DEST_CHANNEL_XYZ, 0,
860 swizzle(tmp, X, X, Z, ONE),
861 swizzle(tmp, Z, ONE, ONE, ONE), 0);
865 get_result_vector(p, &inst->Dst[0]),
866 A0_DEST_CHANNEL_X, 0,
867 swizzle(tmp, ONE, Z, Y, X),
868 i915_emit_const4fv(p, scs_cos_constants), 0);
872 case TGSI_OPCODE_SEQ:
873 /* if we're both >= and <= then we're == */
874 src0 = src_vector(p, &inst->Src[0], fs);
875 src1 = src_vector(p, &inst->Src[1], fs);
876 tmp = i915_get_utemp(p);
880 tmp, A0_DEST_CHANNEL_ALL, 0,
886 get_result_vector(p, &inst->Dst[0]),
887 A0_DEST_CHANNEL_ALL, 0,
893 get_result_vector(p, &inst->Dst[0]),
894 A0_DEST_CHANNEL_ALL, 0,
895 get_result_vector(p, &inst->Dst[0]),
900 case TGSI_OPCODE_SGE:
901 emit_simple_arith(p, inst, A0_SGE, 2, fs);
904 case TGSI_OPCODE_SIN:
905 src0 = src_vector(p, &inst->Src[0], fs);
906 tmp = i915_get_utemp(p);
910 tmp, A0_DEST_CHANNEL_X, 0,
911 src0, i915_emit_const1f(p, 1.0f / (float) (M_PI * 2.0)), 0);
913 i915_emit_arith(p, A0_MOD, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
916 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
917 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
918 * t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x
919 * result = DP4 t1.wzyx, sin_constants
923 tmp, A0_DEST_CHANNEL_XY, 0,
924 swizzle(tmp, X, X, ONE, ONE),
925 swizzle(tmp, X, ONE, ONE, ONE), 0);
929 tmp, A0_DEST_CHANNEL_ALL, 0,
930 swizzle(tmp, X, Y, X, Y),
931 swizzle(tmp, X, X, ONE, ONE), 0);
935 tmp, A0_DEST_CHANNEL_ALL, 0,
936 swizzle(tmp, X, Y, Y, W),
937 swizzle(tmp, X, Z, ONE, ONE), 0);
941 get_result_vector(p, &inst->Dst[0]),
942 get_result_flags(inst), 0,
943 swizzle(tmp, W, Z, Y, X),
944 i915_emit_const4fv(p, sin_constants), 0);
947 case TGSI_OPCODE_SLE:
948 /* like SGE, but swap reg0, reg1 */
949 emit_simple_arith_swap2(p, inst, A0_SGE, 2, fs);
952 case TGSI_OPCODE_SLT:
953 emit_simple_arith(p, inst, A0_SLT, 2, fs);
956 case TGSI_OPCODE_SGT:
957 /* like SLT, but swap reg0, reg1 */
958 emit_simple_arith_swap2(p, inst, A0_SLT, 2, fs);
961 case TGSI_OPCODE_SNE:
962 /* if we're < or > then we're != */
963 src0 = src_vector(p, &inst->Src[0], fs);
964 src1 = src_vector(p, &inst->Src[1], fs);
965 tmp = i915_get_utemp(p);
970 A0_DEST_CHANNEL_ALL, 0,
976 get_result_vector(p, &inst->Dst[0]),
977 A0_DEST_CHANNEL_ALL, 0,
983 get_result_vector(p, &inst->Dst[0]),
984 A0_DEST_CHANNEL_ALL, 0,
985 get_result_vector(p, &inst->Dst[0]),
989 case TGSI_OPCODE_SSG:
990 /* compute (src>0) - (src<0) */
991 src0 = src_vector(p, &inst->Src[0], fs);
992 tmp = i915_get_utemp(p);
997 A0_DEST_CHANNEL_ALL, 0,
999 swizzle(src0, ZERO, ZERO, ZERO, ZERO), 0);
1003 get_result_vector(p, &inst->Dst[0]),
1004 A0_DEST_CHANNEL_ALL, 0,
1005 swizzle(src0, ZERO, ZERO, ZERO, ZERO),
1010 get_result_vector(p, &inst->Dst[0]),
1011 A0_DEST_CHANNEL_ALL, 0,
1012 get_result_vector(p, &inst->Dst[0]),
1013 negate(tmp, 1, 1, 1, 1), 0);
1016 case TGSI_OPCODE_SUB:
1017 src0 = src_vector(p, &inst->Src[0], fs);
1018 src1 = src_vector(p, &inst->Src[1], fs);
1022 get_result_vector(p, &inst->Dst[0]),
1023 get_result_flags(inst), 0,
1024 src0, negate(src1, 1, 1, 1, 1), 0);
1027 case TGSI_OPCODE_TEX:
1028 emit_tex(p, inst, T0_TEXLD, fs);
1031 case TGSI_OPCODE_TRUNC:
1032 emit_simple_arith(p, inst, A0_TRC, 1, fs);
1035 case TGSI_OPCODE_TXB:
1036 emit_tex(p, inst, T0_TEXLDB, fs);
1039 case TGSI_OPCODE_TXP:
1040 emit_tex(p, inst, T0_TEXLDP, fs);
1043 case TGSI_OPCODE_XPD:
1045 * result.x = src0.y * src1.z - src0.z * src1.y;
1046 * result.y = src0.z * src1.x - src0.x * src1.z;
1047 * result.z = src0.x * src1.y - src0.y * src1.x;
1050 src0 = src_vector(p, &inst->Src[0], fs);
1051 src1 = src_vector(p, &inst->Src[1], fs);
1052 tmp = i915_get_utemp(p);
1056 tmp, A0_DEST_CHANNEL_ALL, 0,
1057 swizzle(src0, Z, X, Y, ONE),
1058 swizzle(src1, Y, Z, X, ONE), 0);
1062 get_result_vector(p, &inst->Dst[0]),
1063 get_result_flags(inst), 0,
1064 swizzle(src0, Y, Z, X, ONE),
1065 swizzle(src1, Z, X, Y, ONE),
1066 negate(tmp, 1, 1, 1, 0));
1070 i915_program_error(p, "bad opcode %d", inst->Instruction.Opcode);
1075 i915_release_utemps(p);
1079 static void i915_translate_token(struct i915_fp_compile *p,
1080 const union i915_full_token* token,
1081 struct i915_fragment_shader *fs)
1083 struct i915_fragment_shader *ifs = p->shader;
1084 switch( token->Token.Type ) {
1085 case TGSI_TOKEN_TYPE_PROPERTY:
1087 * We only support one cbuf, but we still need to ignore the property
1088 * correctly so we don't hit the assert at the end of the switch case.
1090 assert(token->FullProperty.Property.PropertyName ==
1091 TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS);
1094 case TGSI_TOKEN_TYPE_DECLARATION:
1095 if (token->FullDeclaration.Declaration.File
1096 == TGSI_FILE_CONSTANT) {
1098 for (i = token->FullDeclaration.Range.First;
1099 i <= token->FullDeclaration.Range.Last;
1101 assert(ifs->constant_flags[i] == 0x0);
1102 ifs->constant_flags[i] = I915_CONSTFLAG_USER;
1103 ifs->num_constants = MAX2(ifs->num_constants, i + 1);
1106 else if (token->FullDeclaration.Declaration.File
1107 == TGSI_FILE_TEMPORARY) {
1109 for (i = token->FullDeclaration.Range.First;
1110 i <= token->FullDeclaration.Range.Last;
1112 if (i >= I915_MAX_TEMPORARY)
1113 debug_printf("Too many temps (%d)\n",i);
1115 /* XXX just use shader->info->file_mask[TGSI_FILE_TEMPORARY] */
1116 p->temp_flag |= (1 << i); /* mark temp as used */
1121 case TGSI_TOKEN_TYPE_IMMEDIATE:
1123 const struct tgsi_full_immediate *imm
1124 = &token->FullImmediate;
1125 const uint pos = p->num_immediates++;
1127 assert( imm->Immediate.NrTokens <= 4 + 1 );
1128 for (j = 0; j < imm->Immediate.NrTokens - 1; j++) {
1129 p->immediates[pos][j] = imm->u[j].Float;
1134 case TGSI_TOKEN_TYPE_INSTRUCTION:
1135 if (p->first_instruction) {
1136 /* resolve location of immediates */
1138 for (i = 0; i < p->num_immediates; i++) {
1139 /* find constant slot for this immediate */
1140 for (j = 0; j < I915_MAX_CONSTANT; j++) {
1141 if (ifs->constant_flags[j] == 0x0) {
1142 memcpy(ifs->constants[j],
1145 /*printf("immediate %d maps to const %d\n", i, j);*/
1146 ifs->constant_flags[j] = 0xf; /* all four comps used */
1147 p->immediates_map[i] = j;
1148 ifs->num_constants = MAX2(ifs->num_constants, j + 1);
1154 p->first_instruction = FALSE;
1157 i915_translate_instruction(p, &token->FullInstruction, fs);
1167 * Translate TGSI fragment shader into i915 hardware instructions.
1168 * \param p the translation state
1169 * \param tokens the TGSI token array
1172 i915_translate_instructions(struct i915_fp_compile *p,
1173 const struct i915_token_list *tokens,
1174 struct i915_fragment_shader *fs)
1177 for(i = 0; i<tokens->NumTokens; i++) {
1178 i915_translate_token(p, &tokens->Tokens[i], fs);
1183 static struct i915_fp_compile *
1184 i915_init_compile(struct i915_context *i915,
1185 struct i915_fragment_shader *ifs)
1187 struct i915_fp_compile *p = CALLOC_STRUCT(i915_fp_compile);
1192 /* Put new constants at end of const buffer, growing downward.
1193 * The problem is we don't know how many user-defined constants might
1194 * be specified with pipe->set_constant_buffer().
1195 * Should pre-scan the user's program to determine the highest-numbered
1196 * constant referenced.
1198 ifs->num_constants = 0;
1199 memset(ifs->constant_flags, 0, sizeof(ifs->constant_flags));
1201 memset(&p->register_phases, 0, sizeof(p->register_phases));
1203 for (i = 0; i < I915_TEX_UNITS; i++)
1204 ifs->generic_mapping[i] = -1;
1206 p->first_instruction = TRUE;
1208 p->nr_tex_indirect = 1; /* correct? */
1211 p->nr_decl_insn = 0;
1213 p->csr = p->program;
1214 p->decl = p->declarations;
1217 p->temp_flag = ~0x0 << I915_MAX_TEMPORARY;
1218 p->utemp_flag = ~0x7;
1220 /* initialize the first program word */
1221 *(p->decl++) = _3DSTATE_PIXEL_SHADER_PROGRAM;
1227 /* Copy compile results to the fragment program struct and destroy the
1228 * compilation context.
1231 i915_fini_compile(struct i915_context *i915, struct i915_fp_compile *p)
1233 struct i915_fragment_shader *ifs = p->shader;
1234 unsigned long program_size = (unsigned long) (p->csr - p->program);
1235 unsigned long decl_size = (unsigned long) (p->decl - p->declarations);
1237 if (p->nr_tex_indirect > I915_MAX_TEX_INDIRECT)
1238 debug_printf("Exceeded max nr indirect texture lookups\n");
1240 if (p->nr_tex_insn > I915_MAX_TEX_INSN)
1241 i915_program_error(p, "Exceeded max TEX instructions");
1243 if (p->nr_alu_insn > I915_MAX_ALU_INSN)
1244 i915_program_error(p, "Exceeded max ALU instructions");
1246 if (p->nr_decl_insn > I915_MAX_DECL_INSN)
1247 i915_program_error(p, "Exceeded max DECL instructions");
1250 p->NumNativeInstructions = 0;
1251 p->NumNativeAluInstructions = 0;
1252 p->NumNativeTexInstructions = 0;
1253 p->NumNativeTexIndirections = 0;
1255 i915_use_passthrough_shader(ifs);
1258 p->NumNativeInstructions
1259 = p->nr_alu_insn + p->nr_tex_insn + p->nr_decl_insn;
1260 p->NumNativeAluInstructions = p->nr_alu_insn;
1261 p->NumNativeTexInstructions = p->nr_tex_insn;
1262 p->NumNativeTexIndirections = p->nr_tex_indirect;
1264 /* patch in the program length */
1265 p->declarations[0] |= program_size + decl_size - 2;
1267 /* Copy compilation results to fragment program struct:
1269 assert(!ifs->program);
1271 = (uint *) MALLOC((program_size + decl_size) * sizeof(uint));
1273 ifs->program_len = program_size + decl_size;
1275 memcpy(ifs->program,
1277 decl_size * sizeof(uint));
1279 memcpy(ifs->program + decl_size,
1281 program_size * sizeof(uint));
1285 /* Release the compilation struct:
1295 * Rather than trying to intercept and jiggle depth writes during
1296 * emit, just move the value into its correct position at the end of
1300 i915_fixup_depth_write(struct i915_fp_compile *p)
1302 /* XXX assuming pos/depth is always in output[0] */
1303 if (p->shader->info.output_semantic_name[0] == TGSI_SEMANTIC_POSITION) {
1304 const uint depth = UREG(REG_TYPE_OD, 0);
1307 A0_MOV, /* opcode */
1308 depth, /* dest reg */
1309 A0_DEST_CHANNEL_W, /* write mask */
1311 swizzle(depth, X, Y, Z, Z), /* src0 */
1312 0, 0 /* src1, src2 */);
1318 i915_translate_fragment_program( struct i915_context *i915,
1319 struct i915_fragment_shader *fs)
1321 struct i915_fp_compile *p;
1322 const struct tgsi_token *tokens = fs->state.tokens;
1323 struct i915_token_list* i_tokens;
1326 tgsi_dump(tokens, 0);
1329 /* hw doesn't seem to like empty frag programs, even when the depth write
1330 * fixup gets emitted below - may that one is fishy, too? */
1331 if (fs->info.num_instructions == 1) {
1332 i915_use_passthrough_shader(fs);
1337 p = i915_init_compile(i915, fs);
1339 i_tokens = i915_optimize(tokens);
1340 i915_translate_instructions(p, i_tokens, fs);
1341 i915_fixup_depth_write(p);
1343 i915_fini_compile(i915, p);
1344 i915_optimize_free(i_tokens);
1347 i915_disassemble_program(NULL, fs->program, fs->program_len);