1 /**************************************************************************
3 * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
5 * Copyright 2009 VMware, Inc. All Rights Reserved.
7 * Permission is hereby granted, free of charge, to any person obtaining a
8 * copy of this software and associated documentation files (the
9 * "Software"), to deal in the Software without restriction, including
10 * without limitation the rights to use, copy, modify, merge, publish,
11 * distribute, sub license, and/or sell copies of the Software, and to
12 * permit persons to whom the Software is furnished to do so, subject to
13 * the following conditions:
15 * The above copyright notice and this permission notice (including the
16 * next paragraph) shall be included in all copies or substantial portions
19 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
20 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
21 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
22 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
23 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
24 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
25 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
27 **************************************************************************/
32 #include "shader/program.h"
33 #include "shader/prog_parameter.h"
34 #include "shader/prog_cache.h"
35 #include "shader/prog_instruction.h"
36 #include "shader/prog_print.h"
37 #include "shader/prog_statevars.h"
38 #include "shader/programopt.h"
39 #include "texenvprogram.h"
43 * Note on texture units:
45 * The number of texture units supported by fixed-function fragment
46 * processing is MAX_TEXTURE_COORD_UNITS, not MAX_TEXTURE_IMAGE_UNITS.
47 * That's because there's a one-to-one correspondence between texture
48 * coordinates and samplers in fixed-function processing.
50 * Since fixed-function vertex processing is limited to MAX_TEXTURE_COORD_UNITS
51 * sets of texcoords, so is fixed-function fragment processing.
53 * We can safely use ctx->Const.MaxTextureUnits for loop bounds.
57 struct texenvprog_cache_item
61 struct gl_fragment_program *data;
62 struct texenvprog_cache_item *next;
66 texenv_doing_secondary_color(GLcontext *ctx)
68 if (ctx->Light.Enabled &&
69 (ctx->Light.Model.ColorControl == GL_SEPARATE_SPECULAR_COLOR))
72 if (ctx->Fog.ColorSumEnabled)
79 * Up to nine instructions per tex unit, plus fog, specular color.
81 #define MAX_INSTRUCTIONS ((MAX_TEXTURE_COORD_UNITS * 9) + 12)
83 #define DISASSEM (MESA_VERBOSE & VERBOSE_DISASSEM)
91 GLuint nr_enabled_units:8;
92 GLuint enabled_units:8;
93 GLuint separate_specular:1;
96 GLuint inputs_available:12;
100 GLuint source_index:3; /* one of TEXTURE_1D/2D/3D/CUBE/RECT_INDEX */
102 GLuint ScaleShiftRGB:2;
103 GLuint ScaleShiftA:2;
107 struct mode_opt OptRGB[MAX_COMBINER_TERMS];
111 struct mode_opt OptA[MAX_COMBINER_TERMS];
118 #define FOG_UNKNOWN 3
120 static GLuint translate_fog_mode( GLenum mode )
123 case GL_LINEAR: return FOG_LINEAR;
124 case GL_EXP: return FOG_EXP;
125 case GL_EXP2: return FOG_EXP2;
126 default: return FOG_UNKNOWN;
130 #define OPR_SRC_COLOR 0
131 #define OPR_ONE_MINUS_SRC_COLOR 1
132 #define OPR_SRC_ALPHA 2
133 #define OPR_ONE_MINUS_SRC_ALPHA 3
136 #define OPR_UNKNOWN 7
138 static GLuint translate_operand( GLenum operand )
141 case GL_SRC_COLOR: return OPR_SRC_COLOR;
142 case GL_ONE_MINUS_SRC_COLOR: return OPR_ONE_MINUS_SRC_COLOR;
143 case GL_SRC_ALPHA: return OPR_SRC_ALPHA;
144 case GL_ONE_MINUS_SRC_ALPHA: return OPR_ONE_MINUS_SRC_ALPHA;
145 case GL_ZERO: return OPR_ZERO;
146 case GL_ONE: return OPR_ONE;
153 #define SRC_TEXTURE 0
154 #define SRC_TEXTURE0 1
155 #define SRC_TEXTURE1 2
156 #define SRC_TEXTURE2 3
157 #define SRC_TEXTURE3 4
158 #define SRC_TEXTURE4 5
159 #define SRC_TEXTURE5 6
160 #define SRC_TEXTURE6 7
161 #define SRC_TEXTURE7 8
162 #define SRC_CONSTANT 9
163 #define SRC_PRIMARY_COLOR 10
164 #define SRC_PREVIOUS 11
166 #define SRC_UNKNOWN 15
168 static GLuint translate_source( GLenum src )
171 case GL_TEXTURE: return SRC_TEXTURE;
179 case GL_TEXTURE7: return SRC_TEXTURE0 + (src - GL_TEXTURE0);
180 case GL_CONSTANT: return SRC_CONSTANT;
181 case GL_PRIMARY_COLOR: return SRC_PRIMARY_COLOR;
182 case GL_PREVIOUS: return SRC_PREVIOUS;
191 #define MODE_REPLACE 0 /* r = a0 */
192 #define MODE_MODULATE 1 /* r = a0 * a1 */
193 #define MODE_ADD 2 /* r = a0 + a1 */
194 #define MODE_ADD_SIGNED 3 /* r = a0 + a1 - 0.5 */
195 #define MODE_INTERPOLATE 4 /* r = a0 * a2 + a1 * (1 - a2) */
196 #define MODE_SUBTRACT 5 /* r = a0 - a1 */
197 #define MODE_DOT3_RGB 6 /* r = a0 . a1 */
198 #define MODE_DOT3_RGB_EXT 7 /* r = a0 . a1 */
199 #define MODE_DOT3_RGBA 8 /* r = a0 . a1 */
200 #define MODE_DOT3_RGBA_EXT 9 /* r = a0 . a1 */
201 #define MODE_MODULATE_ADD_ATI 10 /* r = a0 * a2 + a1 */
202 #define MODE_MODULATE_SIGNED_ADD_ATI 11 /* r = a0 * a2 + a1 - 0.5 */
203 #define MODE_MODULATE_SUBTRACT_ATI 12 /* r = a0 * a2 - a1 */
204 #define MODE_ADD_PRODUCTS 13 /* r = a0 * a1 + a2 * a3 */
205 #define MODE_ADD_PRODUCTS_SIGNED 14 /* r = a0 * a1 + a2 * a3 - 0.5 */
206 #define MODE_BUMP_ENVMAP_ATI 15 /* special */
207 #define MODE_UNKNOWN 16
210 * Translate GL combiner state into a MODE_x value
212 static GLuint translate_mode( GLenum envMode, GLenum mode )
215 case GL_REPLACE: return MODE_REPLACE;
216 case GL_MODULATE: return MODE_MODULATE;
218 if (envMode == GL_COMBINE4_NV)
219 return MODE_ADD_PRODUCTS;
223 if (envMode == GL_COMBINE4_NV)
224 return MODE_ADD_PRODUCTS_SIGNED;
226 return MODE_ADD_SIGNED;
227 case GL_INTERPOLATE: return MODE_INTERPOLATE;
228 case GL_SUBTRACT: return MODE_SUBTRACT;
229 case GL_DOT3_RGB: return MODE_DOT3_RGB;
230 case GL_DOT3_RGB_EXT: return MODE_DOT3_RGB_EXT;
231 case GL_DOT3_RGBA: return MODE_DOT3_RGBA;
232 case GL_DOT3_RGBA_EXT: return MODE_DOT3_RGBA_EXT;
233 case GL_MODULATE_ADD_ATI: return MODE_MODULATE_ADD_ATI;
234 case GL_MODULATE_SIGNED_ADD_ATI: return MODE_MODULATE_SIGNED_ADD_ATI;
235 case GL_MODULATE_SUBTRACT_ATI: return MODE_MODULATE_SUBTRACT_ATI;
236 case GL_BUMP_ENVMAP_ATI: return MODE_BUMP_ENVMAP_ATI;
243 #define TEXTURE_UNKNOWN_INDEX 7
244 static GLuint translate_tex_src_bit( GLbitfield bit )
246 /* make sure number of switch cases is correct */
247 assert(NUM_TEXTURE_TARGETS == 7);
249 case TEXTURE_1D_BIT: return TEXTURE_1D_INDEX;
250 case TEXTURE_2D_BIT: return TEXTURE_2D_INDEX;
251 case TEXTURE_RECT_BIT: return TEXTURE_RECT_INDEX;
252 case TEXTURE_3D_BIT: return TEXTURE_3D_INDEX;
253 case TEXTURE_CUBE_BIT: return TEXTURE_CUBE_INDEX;
254 case TEXTURE_1D_ARRAY_BIT: return TEXTURE_1D_ARRAY_INDEX;
255 case TEXTURE_2D_ARRAY_BIT: return TEXTURE_2D_ARRAY_INDEX;
258 return TEXTURE_UNKNOWN_INDEX;
262 #define VERT_BIT_TEX_ANY (0xff << VERT_ATTRIB_TEX0)
263 #define VERT_RESULT_TEX_ANY (0xff << VERT_RESULT_TEX0)
266 * Identify all possible varying inputs. The fragment program will
267 * never reference non-varying inputs, but will track them via state
270 * This function figures out all the inputs that the fragment program
271 * has access to. The bitmask is later reduced to just those which
272 * are actually referenced.
274 static GLbitfield get_fp_input_mask( GLcontext *ctx )
277 const GLboolean vertexShader = (ctx->Shader.CurrentProgram &&
278 ctx->Shader.CurrentProgram->LinkStatus &&
279 ctx->Shader.CurrentProgram->VertexProgram);
280 const GLboolean vertexProgram = ctx->VertexProgram._Enabled;
281 GLbitfield fp_inputs = 0x0;
283 if (ctx->VertexProgram._Overriden) {
284 /* Somebody's messing with the vertex program and we don't have
285 * a clue what's happening. Assume that it could be producing
286 * all possible outputs.
290 else if (ctx->RenderMode == GL_FEEDBACK) {
291 /* _NEW_RENDERMODE */
292 fp_inputs = (FRAG_BIT_COL0 | FRAG_BIT_TEX0);
294 else if (!(vertexProgram || vertexShader) ||
295 !ctx->VertexProgram._Current) {
296 /* Fixed function vertex logic */
298 GLbitfield varying_inputs = ctx->varying_vp_inputs;
300 /* These get generated in the setup routine regardless of the
304 if (ctx->Point.PointSprite)
305 varying_inputs |= FRAG_BITS_TEX_ANY;
307 /* First look at what values may be computed by the generated
311 if (ctx->Light.Enabled) {
312 fp_inputs |= FRAG_BIT_COL0;
314 if (texenv_doing_secondary_color(ctx))
315 fp_inputs |= FRAG_BIT_COL1;
319 fp_inputs |= (ctx->Texture._TexGenEnabled |
320 ctx->Texture._TexMatEnabled) << FRAG_ATTRIB_TEX0;
322 /* Then look at what might be varying as a result of enabled
325 if (varying_inputs & VERT_BIT_COLOR0)
326 fp_inputs |= FRAG_BIT_COL0;
327 if (varying_inputs & VERT_BIT_COLOR1)
328 fp_inputs |= FRAG_BIT_COL1;
330 fp_inputs |= (((varying_inputs & VERT_BIT_TEX_ANY) >> VERT_ATTRIB_TEX0)
331 << FRAG_ATTRIB_TEX0);
335 /* calculate from vp->outputs */
336 struct gl_vertex_program *vprog;
337 GLbitfield vp_outputs;
339 /* Choose GLSL vertex shader over ARB vertex program. Need this
340 * since vertex shader state validation comes after fragment state
341 * validation (see additional comments in state.c).
344 vprog = ctx->Shader.CurrentProgram->VertexProgram;
346 vprog = ctx->VertexProgram.Current;
348 vp_outputs = vprog->Base.OutputsWritten;
350 /* These get generated in the setup routine regardless of the
354 if (ctx->Point.PointSprite)
355 vp_outputs |= FRAG_BITS_TEX_ANY;
357 if (vp_outputs & (1 << VERT_RESULT_COL0))
358 fp_inputs |= FRAG_BIT_COL0;
359 if (vp_outputs & (1 << VERT_RESULT_COL1))
360 fp_inputs |= FRAG_BIT_COL1;
362 fp_inputs |= (((vp_outputs & VERT_RESULT_TEX_ANY) >> VERT_RESULT_TEX0)
363 << FRAG_ATTRIB_TEX0);
371 * Examine current texture environment state and generate a unique
372 * key to identify it.
374 static void make_state_key( GLcontext *ctx, struct state_key *key )
377 GLbitfield inputs_referenced = FRAG_BIT_COL0;
378 GLbitfield inputs_available = get_fp_input_mask( ctx );
380 memset(key, 0, sizeof(*key));
383 for (i = 0; i < ctx->Const.MaxTextureUnits; i++) {
384 const struct gl_texture_unit *texUnit = &ctx->Texture.Unit[i];
387 if (!texUnit->_ReallyEnabled || !texUnit->Enabled)
390 format = texUnit->_Current->Image[0][texUnit->_Current->BaseLevel]->_BaseFormat;
392 key->unit[i].enabled = 1;
393 key->enabled_units |= (1<<i);
394 key->nr_enabled_units = i+1;
395 inputs_referenced |= FRAG_BIT_TEX(i);
397 key->unit[i].source_index =
398 translate_tex_src_bit(texUnit->_ReallyEnabled);
399 key->unit[i].shadow = ((texUnit->_Current->CompareMode == GL_COMPARE_R_TO_TEXTURE) &&
400 ((format == GL_DEPTH_COMPONENT) ||
401 (format == GL_DEPTH_STENCIL_EXT)));
403 key->unit[i].NumArgsRGB = texUnit->_CurrentCombine->_NumArgsRGB;
404 key->unit[i].NumArgsA = texUnit->_CurrentCombine->_NumArgsA;
406 key->unit[i].ModeRGB =
407 translate_mode(texUnit->EnvMode, texUnit->_CurrentCombine->ModeRGB);
409 translate_mode(texUnit->EnvMode, texUnit->_CurrentCombine->ModeA);
411 key->unit[i].ScaleShiftRGB = texUnit->_CurrentCombine->ScaleShiftRGB;
412 key->unit[i].ScaleShiftA = texUnit->_CurrentCombine->ScaleShiftA;
414 for (j = 0; j < MAX_COMBINER_TERMS; j++) {
415 key->unit[i].OptRGB[j].Operand =
416 translate_operand(texUnit->_CurrentCombine->OperandRGB[j]);
417 key->unit[i].OptA[j].Operand =
418 translate_operand(texUnit->_CurrentCombine->OperandA[j]);
419 key->unit[i].OptRGB[j].Source =
420 translate_source(texUnit->_CurrentCombine->SourceRGB[j]);
421 key->unit[i].OptA[j].Source =
422 translate_source(texUnit->_CurrentCombine->SourceA[j]);
425 if (key->unit[i].ModeRGB == MODE_BUMP_ENVMAP_ATI) {
426 /* requires some special translation */
427 key->unit[i].NumArgsRGB = 2;
428 key->unit[i].ScaleShiftRGB = 0;
429 key->unit[i].OptRGB[0].Operand = OPR_SRC_COLOR;
430 key->unit[i].OptRGB[0].Source = SRC_TEXTURE;
431 key->unit[i].OptRGB[1].Operand = OPR_SRC_COLOR;
432 key->unit[i].OptRGB[1].Source = texUnit->BumpTarget - GL_TEXTURE0 + SRC_TEXTURE0;
436 /* _NEW_LIGHT | _NEW_FOG */
437 if (texenv_doing_secondary_color(ctx)) {
438 key->separate_specular = 1;
439 inputs_referenced |= FRAG_BIT_COL1;
443 if (ctx->Fog.Enabled) {
444 key->fog_enabled = 1;
445 key->fog_mode = translate_fog_mode(ctx->Fog.Mode);
446 inputs_referenced |= FRAG_BIT_FOGC; /* maybe */
449 key->inputs_available = (inputs_available & inputs_referenced);
453 * Use uregs to represent registers internally, translate to Mesa's
454 * expected formats on emit.
456 * NOTE: These are passed by value extensively in this file rather
457 * than as usual by pointer reference. If this disturbs you, try
458 * remembering they are just 32bits in size.
460 * GCC is smart enough to deal with these dword-sized structures in
461 * much the same way as if I had defined them as dwords and was using
462 * macros to access and set the fields. This is much nicer and easier
475 static const struct ureg undef = {
486 /** State used to build the fragment program:
488 struct texenv_fragment_program {
489 struct gl_fragment_program *program;
491 struct state_key *state;
493 GLbitfield alu_temps; /**< Track texture indirections, see spec. */
494 GLbitfield temps_output; /**< Track texture indirections, see spec. */
495 GLbitfield temp_in_use; /**< Tracks temporary regs which are in use. */
498 struct ureg src_texture[MAX_TEXTURE_COORD_UNITS];
499 /* Reg containing each texture unit's sampled texture color,
503 struct ureg texcoord_tex[MAX_TEXTURE_COORD_UNITS];
504 /* Reg containing texcoord for a texture unit,
505 * needed for bump mapping, else undef.
508 struct ureg src_previous; /**< Reg containing color from previous
509 * stage. May need to be decl'd.
512 GLuint last_tex_stage; /**< Number of last enabled texture unit */
521 static struct ureg make_ureg(GLuint file, GLuint idx)
529 reg.swz = SWIZZLE_NOOP;
534 static struct ureg swizzle( struct ureg reg, int x, int y, int z, int w )
536 reg.swz = MAKE_SWIZZLE4(GET_SWZ(reg.swz, x),
539 GET_SWZ(reg.swz, w));
544 static struct ureg swizzle1( struct ureg reg, int x )
546 return swizzle(reg, x, x, x, x);
549 static struct ureg negate( struct ureg reg )
555 static GLboolean is_undef( struct ureg reg )
557 return reg.file == PROGRAM_UNDEFINED;
561 static struct ureg get_temp( struct texenv_fragment_program *p )
565 /* First try and reuse temps which have been used already:
567 bit = _mesa_ffs( ~p->temp_in_use & p->alu_temps );
569 /* Then any unused temporary:
572 bit = _mesa_ffs( ~p->temp_in_use );
575 _mesa_problem(NULL, "%s: out of temporaries\n", __FILE__);
579 if ((GLuint) bit > p->program->Base.NumTemporaries)
580 p->program->Base.NumTemporaries = bit;
582 p->temp_in_use |= 1<<(bit-1);
583 return make_ureg(PROGRAM_TEMPORARY, (bit-1));
586 static struct ureg get_tex_temp( struct texenv_fragment_program *p )
590 /* First try to find available temp not previously used (to avoid
591 * starting a new texture indirection). According to the spec, the
592 * ~p->temps_output isn't necessary, but will keep it there for
595 bit = _mesa_ffs( ~p->temp_in_use & ~p->alu_temps & ~p->temps_output );
597 /* Then any unused temporary:
600 bit = _mesa_ffs( ~p->temp_in_use );
603 _mesa_problem(NULL, "%s: out of temporaries\n", __FILE__);
607 if ((GLuint) bit > p->program->Base.NumTemporaries)
608 p->program->Base.NumTemporaries = bit;
610 p->temp_in_use |= 1<<(bit-1);
611 return make_ureg(PROGRAM_TEMPORARY, (bit-1));
615 /** Mark a temp reg as being no longer allocatable. */
616 static void reserve_temp( struct texenv_fragment_program *p, struct ureg r )
618 if (r.file == PROGRAM_TEMPORARY)
619 p->temps_output |= (1 << r.idx);
623 static void release_temps(GLcontext *ctx, struct texenv_fragment_program *p )
625 GLuint max_temp = ctx->Const.FragmentProgram.MaxTemps;
627 /* KW: To support tex_env_crossbar, don't release the registers in
630 if (max_temp >= sizeof(int) * 8)
631 p->temp_in_use = p->temps_output;
633 p->temp_in_use = ~((1<<max_temp)-1) | p->temps_output;
637 static struct ureg register_param5( struct texenv_fragment_program *p,
644 gl_state_index tokens[STATE_LENGTH];
651 idx = _mesa_add_state_reference( p->program->Base.Parameters, tokens );
652 return make_ureg(PROGRAM_STATE_VAR, idx);
656 #define register_param1(p,s0) register_param5(p,s0,0,0,0,0)
657 #define register_param2(p,s0,s1) register_param5(p,s0,s1,0,0,0)
658 #define register_param3(p,s0,s1,s2) register_param5(p,s0,s1,s2,0,0)
659 #define register_param4(p,s0,s1,s2,s3) register_param5(p,s0,s1,s2,s3,0)
661 static GLuint frag_to_vert_attrib( GLuint attrib )
664 case FRAG_ATTRIB_COL0: return VERT_ATTRIB_COLOR0;
665 case FRAG_ATTRIB_COL1: return VERT_ATTRIB_COLOR1;
667 assert(attrib >= FRAG_ATTRIB_TEX0);
668 assert(attrib <= FRAG_ATTRIB_TEX7);
669 return attrib - FRAG_ATTRIB_TEX0 + VERT_ATTRIB_TEX0;
674 static struct ureg register_input( struct texenv_fragment_program *p, GLuint input )
676 if (p->state->inputs_available & (1<<input)) {
677 p->program->Base.InputsRead |= (1 << input);
678 return make_ureg(PROGRAM_INPUT, input);
681 GLuint idx = frag_to_vert_attrib( input );
682 return register_param3( p, STATE_INTERNAL, STATE_CURRENT_ATTRIB, idx );
687 static void emit_arg( struct prog_src_register *reg,
690 reg->File = ureg.file;
691 reg->Index = ureg.idx;
692 reg->Swizzle = ureg.swz;
693 reg->Negate = ureg.negatebase ? NEGATE_XYZW : NEGATE_NONE;
697 static void emit_dst( struct prog_dst_register *dst,
698 struct ureg ureg, GLuint mask )
700 dst->File = ureg.file;
701 dst->Index = ureg.idx;
702 dst->WriteMask = mask;
703 dst->CondMask = COND_TR; /* always pass cond test */
704 dst->CondSwizzle = SWIZZLE_NOOP;
707 static struct prog_instruction *
708 emit_op(struct texenv_fragment_program *p,
717 GLuint nr = p->program->Base.NumInstructions++;
718 struct prog_instruction *inst = &p->program->Base.Instructions[nr];
720 assert(nr < MAX_INSTRUCTIONS);
722 _mesa_init_instructions(inst, 1);
725 emit_arg( &inst->SrcReg[0], src0 );
726 emit_arg( &inst->SrcReg[1], src1 );
727 emit_arg( &inst->SrcReg[2], src2 );
729 inst->SaturateMode = saturate ? SATURATE_ZERO_ONE : SATURATE_OFF;
731 emit_dst( &inst->DstReg, dest, mask );
734 /* Accounting for indirection tracking:
736 if (dest.file == PROGRAM_TEMPORARY)
737 p->temps_output |= 1 << dest.idx;
744 static struct ureg emit_arith( struct texenv_fragment_program *p,
753 emit_op(p, op, dest, mask, saturate, src0, src1, src2);
755 /* Accounting for indirection tracking:
757 if (src0.file == PROGRAM_TEMPORARY)
758 p->alu_temps |= 1 << src0.idx;
760 if (!is_undef(src1) && src1.file == PROGRAM_TEMPORARY)
761 p->alu_temps |= 1 << src1.idx;
763 if (!is_undef(src2) && src2.file == PROGRAM_TEMPORARY)
764 p->alu_temps |= 1 << src2.idx;
766 if (dest.file == PROGRAM_TEMPORARY)
767 p->alu_temps |= 1 << dest.idx;
769 p->program->Base.NumAluInstructions++;
773 static struct ureg emit_texld( struct texenv_fragment_program *p,
782 struct prog_instruction *inst = emit_op( p, op,
784 GL_FALSE, /* don't saturate? */
789 inst->TexSrcTarget = tex_idx;
790 inst->TexSrcUnit = tex_unit;
791 inst->TexShadow = tex_shadow;
793 p->program->Base.NumTexInstructions++;
795 /* Accounting for indirection tracking:
797 reserve_temp(p, dest);
800 /* Is this a texture indirection?
802 if ((coord.file == PROGRAM_TEMPORARY &&
803 (p->temps_output & (1<<coord.idx))) ||
804 (dest.file == PROGRAM_TEMPORARY &&
805 (p->alu_temps & (1<<dest.idx)))) {
806 p->program->Base.NumTexIndirections++;
807 p->temps_output = 1<<coord.idx;
809 assert(0); /* KW: texture env crossbar */
817 static struct ureg register_const4f( struct texenv_fragment_program *p,
830 idx = _mesa_add_unnamed_constant( p->program->Base.Parameters, values, 4,
832 r = make_ureg(PROGRAM_CONSTANT, idx);
837 #define register_scalar_const(p, s0) register_const4f(p, s0, s0, s0, s0)
838 #define register_const1f(p, s0) register_const4f(p, s0, 0, 0, 1)
839 #define register_const2f(p, s0, s1) register_const4f(p, s0, s1, 0, 1)
840 #define register_const3f(p, s0, s1, s2) register_const4f(p, s0, s1, s2, 1)
843 static struct ureg get_one( struct texenv_fragment_program *p )
845 if (is_undef(p->one))
846 p->one = register_scalar_const(p, 1.0);
850 static struct ureg get_half( struct texenv_fragment_program *p )
852 if (is_undef(p->half))
853 p->half = register_scalar_const(p, 0.5);
857 static struct ureg get_zero( struct texenv_fragment_program *p )
859 if (is_undef(p->zero))
860 p->zero = register_scalar_const(p, 0.0);
865 static void program_error( struct texenv_fragment_program *p, const char *msg )
867 _mesa_problem(NULL, msg);
871 static struct ureg get_source( struct texenv_fragment_program *p,
872 GLuint src, GLuint unit )
876 assert(!is_undef(p->src_texture[unit]));
877 return p->src_texture[unit];
887 assert(!is_undef(p->src_texture[src - SRC_TEXTURE0]));
888 return p->src_texture[src - SRC_TEXTURE0];
891 return register_param2(p, STATE_TEXENV_COLOR, unit);
893 case SRC_PRIMARY_COLOR:
894 return register_input(p, FRAG_ATTRIB_COL0);
900 if (is_undef(p->src_previous))
901 return register_input(p, FRAG_ATTRIB_COL0);
903 return p->src_previous;
911 static struct ureg emit_combine_source( struct texenv_fragment_program *p,
917 struct ureg arg, src, one;
919 src = get_source(p, source, unit);
922 case OPR_ONE_MINUS_SRC_COLOR:
924 * Emit tmp = 1.0 - arg.xyzw
928 return emit_arith( p, OPCODE_SUB, arg, mask, 0, one, src, undef);
931 if (mask == WRITEMASK_W)
934 return swizzle1( src, SWIZZLE_W );
935 case OPR_ONE_MINUS_SRC_ALPHA:
937 * Emit tmp = 1.0 - arg.wwww
941 return emit_arith(p, OPCODE_SUB, arg, mask, 0,
942 one, swizzle1(src, SWIZZLE_W), undef);
955 static GLboolean args_match( const struct state_key *key, GLuint unit )
957 GLuint i, nr = key->unit[unit].NumArgsRGB;
959 for (i = 0 ; i < nr ; i++) {
960 if (key->unit[unit].OptA[i].Source != key->unit[unit].OptRGB[i].Source)
963 switch(key->unit[unit].OptA[i].Operand) {
965 switch(key->unit[unit].OptRGB[i].Operand) {
973 case OPR_ONE_MINUS_SRC_ALPHA:
974 switch(key->unit[unit].OptRGB[i].Operand) {
975 case OPR_ONE_MINUS_SRC_COLOR:
976 case OPR_ONE_MINUS_SRC_ALPHA:
983 return GL_FALSE; /* impossible */
990 static struct ureg emit_combine( struct texenv_fragment_program *p,
997 const struct mode_opt *opt)
999 struct ureg src[MAX_COMBINER_TERMS];
1000 struct ureg tmp, half;
1003 assert(nr <= MAX_COMBINER_TERMS);
1005 tmp = undef; /* silence warning (bug 5318) */
1007 for (i = 0; i < nr; i++)
1008 src[i] = emit_combine_source( p, mask, unit, opt[i].Source, opt[i].Operand );
1012 if (mask == WRITEMASK_XYZW && !saturate)
1015 return emit_arith( p, OPCODE_MOV, dest, mask, saturate, src[0], undef, undef );
1017 return emit_arith( p, OPCODE_MUL, dest, mask, saturate,
1018 src[0], src[1], undef );
1020 return emit_arith( p, OPCODE_ADD, dest, mask, saturate,
1021 src[0], src[1], undef );
1022 case MODE_ADD_SIGNED:
1023 /* tmp = arg0 + arg1
1027 tmp = get_temp( p );
1028 emit_arith( p, OPCODE_ADD, tmp, mask, 0, src[0], src[1], undef );
1029 emit_arith( p, OPCODE_SUB, dest, mask, saturate, tmp, half, undef );
1031 case MODE_INTERPOLATE:
1032 /* Arg0 * (Arg2) + Arg1 * (1-Arg2) -- note arguments are reordered:
1034 return emit_arith( p, OPCODE_LRP, dest, mask, saturate, src[2], src[0], src[1] );
1037 return emit_arith( p, OPCODE_SUB, dest, mask, saturate, src[0], src[1], undef );
1039 case MODE_DOT3_RGBA:
1040 case MODE_DOT3_RGBA_EXT:
1041 case MODE_DOT3_RGB_EXT:
1042 case MODE_DOT3_RGB: {
1043 struct ureg tmp0 = get_temp( p );
1044 struct ureg tmp1 = get_temp( p );
1045 struct ureg neg1 = register_scalar_const(p, -1);
1046 struct ureg two = register_scalar_const(p, 2);
1048 /* tmp0 = 2*src0 - 1
1051 * dst = tmp0 dot3 tmp1
1053 emit_arith( p, OPCODE_MAD, tmp0, WRITEMASK_XYZW, 0,
1056 if (_mesa_memcmp(&src[0], &src[1], sizeof(struct ureg)) == 0)
1059 emit_arith( p, OPCODE_MAD, tmp1, WRITEMASK_XYZW, 0,
1061 emit_arith( p, OPCODE_DP3, dest, mask, saturate, tmp0, tmp1, undef);
1064 case MODE_MODULATE_ADD_ATI:
1065 /* Arg0 * Arg2 + Arg1 */
1066 return emit_arith( p, OPCODE_MAD, dest, mask, saturate,
1067 src[0], src[2], src[1] );
1068 case MODE_MODULATE_SIGNED_ADD_ATI: {
1069 /* Arg0 * Arg2 + Arg1 - 0.5 */
1070 struct ureg tmp0 = get_temp(p);
1072 emit_arith( p, OPCODE_MAD, tmp0, mask, 0, src[0], src[2], src[1] );
1073 emit_arith( p, OPCODE_SUB, dest, mask, saturate, tmp0, half, undef );
1076 case MODE_MODULATE_SUBTRACT_ATI:
1077 /* Arg0 * Arg2 - Arg1 */
1078 emit_arith( p, OPCODE_MAD, dest, mask, 0, src[0], src[2], negate(src[1]) );
1080 case MODE_ADD_PRODUCTS:
1081 /* Arg0 * Arg1 + Arg2 * Arg3 */
1083 struct ureg tmp0 = get_temp(p);
1084 emit_arith( p, OPCODE_MUL, tmp0, mask, 0, src[0], src[1], undef );
1085 emit_arith( p, OPCODE_MAD, dest, mask, saturate, src[2], src[3], tmp0 );
1088 case MODE_ADD_PRODUCTS_SIGNED:
1089 /* Arg0 * Arg1 + Arg2 * Arg3 - 0.5 */
1091 struct ureg tmp0 = get_temp(p);
1093 emit_arith( p, OPCODE_MUL, tmp0, mask, 0, src[0], src[1], undef );
1094 emit_arith( p, OPCODE_MAD, tmp0, mask, 0, src[2], src[3], tmp0 );
1095 emit_arith( p, OPCODE_SUB, dest, mask, saturate, tmp0, half, undef );
1098 case MODE_BUMP_ENVMAP_ATI:
1099 /* special - not handled here */
1110 * Generate instructions for one texture unit's env/combiner mode.
1113 emit_texenv(struct texenv_fragment_program *p, GLuint unit)
1115 const struct state_key *key = p->state;
1117 GLuint rgb_shift, alpha_shift;
1118 struct ureg out, dest;
1120 if (!key->unit[unit].enabled) {
1121 return get_source(p, SRC_PREVIOUS, 0);
1123 if (key->unit[unit].ModeRGB == MODE_BUMP_ENVMAP_ATI) {
1124 /* this isn't really a env stage delivering a color and handled elsewhere */
1125 return get_source(p, SRC_PREVIOUS, 0);
1128 switch (key->unit[unit].ModeRGB) {
1129 case MODE_DOT3_RGB_EXT:
1130 alpha_shift = key->unit[unit].ScaleShiftA;
1133 case MODE_DOT3_RGBA_EXT:
1138 rgb_shift = key->unit[unit].ScaleShiftRGB;
1139 alpha_shift = key->unit[unit].ScaleShiftA;
1143 /* If we'll do rgb/alpha shifting don't saturate in emit_combine().
1144 * We don't want to clamp twice.
1146 saturate = !(rgb_shift || alpha_shift);
1148 /* If this is the very last calculation, emit direct to output reg:
1150 if (key->separate_specular ||
1151 unit != p->last_tex_stage ||
1154 dest = get_temp( p );
1156 dest = make_ureg(PROGRAM_OUTPUT, FRAG_RESULT_COLOR);
1158 /* Emit the RGB and A combine ops
1160 if (key->unit[unit].ModeRGB == key->unit[unit].ModeA &&
1161 args_match(key, unit)) {
1162 out = emit_combine( p, dest, WRITEMASK_XYZW, saturate,
1164 key->unit[unit].NumArgsRGB,
1165 key->unit[unit].ModeRGB,
1166 key->unit[unit].OptRGB);
1168 else if (key->unit[unit].ModeRGB == MODE_DOT3_RGBA_EXT ||
1169 key->unit[unit].ModeRGB == MODE_DOT3_RGBA) {
1170 out = emit_combine( p, dest, WRITEMASK_XYZW, saturate,
1172 key->unit[unit].NumArgsRGB,
1173 key->unit[unit].ModeRGB,
1174 key->unit[unit].OptRGB);
1177 /* Need to do something to stop from re-emitting identical
1178 * argument calculations here:
1180 out = emit_combine( p, dest, WRITEMASK_XYZ, saturate,
1182 key->unit[unit].NumArgsRGB,
1183 key->unit[unit].ModeRGB,
1184 key->unit[unit].OptRGB);
1185 out = emit_combine( p, dest, WRITEMASK_W, saturate,
1187 key->unit[unit].NumArgsA,
1188 key->unit[unit].ModeA,
1189 key->unit[unit].OptA);
1192 /* Deal with the final shift:
1194 if (alpha_shift || rgb_shift) {
1197 saturate = GL_TRUE; /* always saturate at this point */
1199 if (rgb_shift == alpha_shift) {
1200 shift = register_scalar_const(p, (GLfloat)(1<<rgb_shift));
1203 shift = register_const4f(p,
1204 (GLfloat)(1<<rgb_shift),
1205 (GLfloat)(1<<rgb_shift),
1206 (GLfloat)(1<<rgb_shift),
1207 (GLfloat)(1<<alpha_shift));
1209 return emit_arith( p, OPCODE_MUL, dest, WRITEMASK_XYZW,
1210 saturate, out, shift, undef );
1218 * Generate instruction for getting a texture source term.
1220 static void load_texture( struct texenv_fragment_program *p, GLuint unit )
1222 if (is_undef(p->src_texture[unit])) {
1223 GLuint texTarget = p->state->unit[unit].source_index;
1224 struct ureg texcoord;
1225 struct ureg tmp = get_tex_temp( p );
1227 if (is_undef(p->texcoord_tex[unit])) {
1228 texcoord = register_input(p, FRAG_ATTRIB_TEX0+unit);
1231 /* might want to reuse this reg for tex output actually */
1232 texcoord = p->texcoord_tex[unit];
1235 if (texTarget == TEXTURE_UNKNOWN_INDEX)
1236 program_error(p, "TexSrcBit");
1238 /* TODO: Use D0_MASK_XY where possible.
1240 if (p->state->unit[unit].enabled) {
1241 GLboolean shadow = GL_FALSE;
1243 if (p->state->unit[unit].shadow) {
1244 p->program->Base.ShadowSamplers |= 1 << unit;
1248 p->src_texture[unit] = emit_texld( p, OPCODE_TXP,
1249 tmp, WRITEMASK_XYZW,
1250 unit, texTarget, shadow,
1253 p->program->Base.SamplersUsed |= (1 << unit);
1254 /* This identity mapping should already be in place
1255 * (see _mesa_init_program_struct()) but let's be safe.
1257 p->program->Base.SamplerUnits[unit] = unit;
1260 p->src_texture[unit] = get_zero(p);
1264 static GLboolean load_texenv_source( struct texenv_fragment_program *p,
1265 GLuint src, GLuint unit )
1269 load_texture(p, unit);
1280 load_texture(p, src - SRC_TEXTURE0);
1284 /* not a texture src - do nothing */
1293 * Generate instructions for loading all texture source terms.
1296 load_texunit_sources( struct texenv_fragment_program *p, int unit )
1298 const struct state_key *key = p->state;
1301 for (i = 0; i < key->unit[unit].NumArgsRGB; i++) {
1302 load_texenv_source( p, key->unit[unit].OptRGB[i].Source, unit );
1305 for (i = 0; i < key->unit[unit].NumArgsA; i++) {
1306 load_texenv_source( p, key->unit[unit].OptA[i].Source, unit );
1313 * Generate instructions for loading bump map textures.
1316 load_texunit_bumpmap( struct texenv_fragment_program *p, int unit )
1318 const struct state_key *key = p->state;
1319 GLuint bumpedUnitNr = key->unit[unit].OptRGB[1].Source - SRC_TEXTURE0;
1320 struct ureg texcDst, bumpMapRes;
1321 struct ureg constdudvcolor = register_const4f(p, 0.0, 0.0, 0.0, 1.0);
1322 struct ureg texcSrc = register_input(p, FRAG_ATTRIB_TEX0 + bumpedUnitNr);
1323 struct ureg rotMat0 = register_param3( p, STATE_INTERNAL, STATE_ROT_MATRIX_0, unit );
1324 struct ureg rotMat1 = register_param3( p, STATE_INTERNAL, STATE_ROT_MATRIX_1, unit );
1326 load_texenv_source( p, unit + SRC_TEXTURE0, unit );
1328 bumpMapRes = get_source(p, key->unit[unit].OptRGB[0].Source, unit);
1329 texcDst = get_tex_temp( p );
1330 p->texcoord_tex[bumpedUnitNr] = texcDst;
1332 /* apply rot matrix and add coords to be available in next phase */
1333 /* dest = (Arg0.xxxx * rotMat0 + Arg1) + (Arg0.yyyy * rotMat1) */
1334 /* note only 2 coords are affected the rest are left unchanged (mul by 0) */
1335 emit_arith( p, OPCODE_MAD, texcDst, WRITEMASK_XYZW, 0,
1336 swizzle1(bumpMapRes, SWIZZLE_X), rotMat0, texcSrc );
1337 emit_arith( p, OPCODE_MAD, texcDst, WRITEMASK_XYZW, 0,
1338 swizzle1(bumpMapRes, SWIZZLE_Y), rotMat1, texcDst );
1340 /* move 0,0,0,1 into bumpmap src if someone (crossbar) is foolish
1341 enough to access this later, should optimize away */
1342 emit_arith( p, OPCODE_MOV, bumpMapRes, WRITEMASK_XYZW, 0, constdudvcolor, undef, undef );
1348 * Generate a new fragment program which implements the context's
1349 * current texture env/combine mode.
1352 create_new_program(GLcontext *ctx, struct state_key *key,
1353 struct gl_fragment_program *program)
1355 struct prog_instruction instBuffer[MAX_INSTRUCTIONS];
1356 struct texenv_fragment_program p;
1358 struct ureg cf, out;
1360 _mesa_memset(&p, 0, sizeof(p));
1363 p.program = program;
1365 /* During code generation, use locally-allocated instruction buffer,
1366 * then alloc dynamic storage below.
1368 p.program->Base.Instructions = instBuffer;
1369 p.program->Base.Target = GL_FRAGMENT_PROGRAM_ARB;
1370 p.program->Base.NumTexIndirections = 1;
1371 p.program->Base.NumTexInstructions = 0;
1372 p.program->Base.NumAluInstructions = 0;
1373 p.program->Base.String = NULL;
1374 p.program->Base.NumInstructions =
1375 p.program->Base.NumTemporaries =
1376 p.program->Base.NumParameters =
1377 p.program->Base.NumAttributes = p.program->Base.NumAddressRegs = 0;
1378 p.program->Base.Parameters = _mesa_new_parameter_list();
1380 p.program->Base.InputsRead = 0;
1381 p.program->Base.OutputsWritten = 1 << FRAG_RESULT_COLOR;
1383 for (unit = 0; unit < ctx->Const.MaxTextureUnits; unit++) {
1384 p.src_texture[unit] = undef;
1385 p.texcoord_tex[unit] = undef;
1388 p.src_previous = undef;
1393 p.last_tex_stage = 0;
1394 release_temps(ctx, &p);
1396 if (key->enabled_units) {
1397 GLboolean needbumpstage = GL_FALSE;
1398 /* Zeroth pass - bump map textures first */
1399 for (unit = 0 ; unit < ctx->Const.MaxTextureUnits ; unit++)
1400 if (key->unit[unit].enabled && key->unit[unit].ModeRGB == MODE_BUMP_ENVMAP_ATI) {
1401 needbumpstage = GL_TRUE;
1402 load_texunit_bumpmap( &p, unit );
1405 p.program->Base.NumTexIndirections++;
1407 /* First pass - to support texture_env_crossbar, first identify
1408 * all referenced texture sources and emit texld instructions
1411 for (unit = 0 ; unit < ctx->Const.MaxTextureUnits ; unit++)
1412 if (key->unit[unit].enabled) {
1413 load_texunit_sources( &p, unit );
1414 p.last_tex_stage = unit;
1417 /* Second pass - emit combine instructions to build final color:
1419 for (unit = 0 ; unit < ctx->Const.MaxTextureUnits; unit++)
1420 if (key->enabled_units & (1<<unit)) {
1421 p.src_previous = emit_texenv( &p, unit );
1422 reserve_temp(&p, p.src_previous); /* don't re-use this temp reg */
1423 release_temps(ctx, &p); /* release all temps */
1427 cf = get_source( &p, SRC_PREVIOUS, 0 );
1428 out = make_ureg( PROGRAM_OUTPUT, FRAG_RESULT_COLOR );
1430 if (key->separate_specular) {
1431 /* Emit specular add.
1433 struct ureg s = register_input(&p, FRAG_ATTRIB_COL1);
1434 emit_arith( &p, OPCODE_ADD, out, WRITEMASK_XYZ, 0, cf, s, undef );
1435 emit_arith( &p, OPCODE_MOV, out, WRITEMASK_W, 0, cf, undef, undef );
1437 else if (_mesa_memcmp(&cf, &out, sizeof(cf)) != 0) {
1438 /* Will wind up in here if no texture enabled or a couple of
1439 * other scenarios (GL_REPLACE for instance).
1441 emit_arith( &p, OPCODE_MOV, out, WRITEMASK_XYZW, 0, cf, undef, undef );
1446 emit_arith( &p, OPCODE_END, undef, WRITEMASK_XYZW, 0, undef, undef, undef);
1448 if (key->fog_enabled) {
1449 /* Pull fog mode from GLcontext, the value in the state key is
1450 * a reduced value and not what is expected in FogOption
1452 p.program->FogOption = ctx->Fog.Mode;
1453 p.program->Base.InputsRead |= FRAG_BIT_FOGC; /* XXX new */
1455 p.program->FogOption = GL_NONE;
1457 if (p.program->Base.NumTexIndirections > ctx->Const.FragmentProgram.MaxTexIndirections)
1458 program_error(&p, "Exceeded max nr indirect texture lookups");
1460 if (p.program->Base.NumTexInstructions > ctx->Const.FragmentProgram.MaxTexInstructions)
1461 program_error(&p, "Exceeded max TEX instructions");
1463 if (p.program->Base.NumAluInstructions > ctx->Const.FragmentProgram.MaxAluInstructions)
1464 program_error(&p, "Exceeded max ALU instructions");
1466 ASSERT(p.program->Base.NumInstructions <= MAX_INSTRUCTIONS);
1468 /* Allocate final instruction array */
1469 p.program->Base.Instructions
1470 = _mesa_alloc_instructions(p.program->Base.NumInstructions);
1471 if (!p.program->Base.Instructions) {
1472 _mesa_error(ctx, GL_OUT_OF_MEMORY,
1473 "generating tex env program");
1476 _mesa_copy_instructions(p.program->Base.Instructions, instBuffer,
1477 p.program->Base.NumInstructions);
1479 if (p.program->FogOption) {
1480 _mesa_append_fog_code(ctx, p.program);
1481 p.program->FogOption = GL_NONE;
1485 /* Notify driver the fragment program has (actually) changed.
1487 if (ctx->Driver.ProgramStringNotify) {
1488 ctx->Driver.ProgramStringNotify( ctx, GL_FRAGMENT_PROGRAM_ARB,
1493 _mesa_print_program(&p.program->Base);
1500 * Return a fragment program which implements the current
1501 * fixed-function texture, fog and color-sum operations.
1503 struct gl_fragment_program *
1504 _mesa_get_fixed_func_fragment_program(GLcontext *ctx)
1506 struct gl_fragment_program *prog;
1507 struct state_key key;
1509 make_state_key(ctx, &key);
1511 prog = (struct gl_fragment_program *)
1512 _mesa_search_program_cache(ctx->FragmentProgram.Cache,
1516 prog = (struct gl_fragment_program *)
1517 ctx->Driver.NewProgram(ctx, GL_FRAGMENT_PROGRAM_ARB, 0);
1519 create_new_program(ctx, &key, prog);
1521 _mesa_program_cache_insert(ctx, ctx->FragmentProgram.Cache,
1522 &key, sizeof(key), &prog->Base);