1 /**************************************************************************
3 * Copyright 2007 VMware, Inc.
5 * Copyright 2009 VMware, Inc. All Rights Reserved.
6 * Copyright © 2010-2011 Intel Corporation
8 * Permission is hereby granted, free of charge, to any person obtaining a
9 * copy of this software and associated documentation files (the
10 * "Software"), to deal in the Software without restriction, including
11 * without limitation the rights to use, copy, modify, merge, publish,
12 * distribute, sub license, and/or sell copies of the Software, and to
13 * permit persons to whom the Software is furnished to do so, subject to
14 * the following conditions:
16 * The above copyright notice and this permission notice (including the
17 * next paragraph) shall be included in all copies or substantial portions
20 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
21 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
22 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
23 * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
24 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
25 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
26 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
28 **************************************************************************/
30 #include "main/glheader.h"
31 #include "main/context.h"
32 #include "main/imports.h"
33 #include "main/macros.h"
34 #include "main/samplerobj.h"
35 #include "main/shaderobj.h"
36 #include "main/texenvprogram.h"
37 #include "main/texobj.h"
38 #include "main/uniforms.h"
39 #include "compiler/glsl/ir_builder.h"
40 #include "compiler/glsl/ir_optimization.h"
41 #include "compiler/glsl/glsl_parser_extras.h"
42 #include "compiler/glsl/glsl_symbol_table.h"
43 #include "compiler/glsl_types.h"
44 #include "program/ir_to_mesa.h"
45 #include "program/program.h"
46 #include "program/programopt.h"
47 #include "program/prog_cache.h"
48 #include "program/prog_instruction.h"
49 #include "program/prog_parameter.h"
50 #include "program/prog_print.h"
51 #include "program/prog_statevars.h"
52 #include "util/bitscan.h"
54 using namespace ir_builder
;
57 * Note on texture units:
59 * The number of texture units supported by fixed-function fragment
60 * processing is MAX_TEXTURE_COORD_UNITS, not MAX_TEXTURE_IMAGE_UNITS.
61 * That's because there's a one-to-one correspondence between texture
62 * coordinates and samplers in fixed-function processing.
64 * Since fixed-function vertex processing is limited to MAX_TEXTURE_COORD_UNITS
65 * sets of texcoords, so is fixed-function fragment processing.
67 * We can safely use ctx->Const.MaxTextureUnits for loop bounds.
72 texenv_doing_secondary_color(struct gl_context
*ctx
)
74 if (ctx
->Light
.Enabled
&&
75 (ctx
->Light
.Model
.ColorControl
== GL_SEPARATE_SPECULAR_COLOR
))
78 if (ctx
->Fog
.ColorSumEnabled
)
86 __extension__ GLubyte Source
:4; /**< SRC_x */
87 __extension__ GLubyte Operand
:3; /**< OPR_x */
89 GLubyte Source
; /**< SRC_x */
90 GLubyte Operand
; /**< OPR_x */
95 GLuint nr_enabled_units
:4;
96 GLuint separate_specular
:1;
97 GLuint fog_mode
:2; /**< FOG_x */
98 GLuint inputs_available
:12;
99 GLuint num_draw_buffers
:4;
101 /* NOTE: This array of structs must be last! (see "keySize" below) */
104 GLuint source_index
:4; /**< TEXTURE_x_INDEX */
106 GLuint ScaleShiftRGB
:2;
107 GLuint ScaleShiftA
:2;
109 GLuint NumArgsRGB
:3; /**< up to MAX_COMBINER_TERMS */
110 GLuint ModeRGB
:5; /**< MODE_x */
112 GLuint NumArgsA
:3; /**< up to MAX_COMBINER_TERMS */
113 GLuint ModeA
:5; /**< MODE_x */
115 struct mode_opt OptRGB
[MAX_COMBINER_TERMS
];
116 struct mode_opt OptA
[MAX_COMBINER_TERMS
];
117 } unit
[MAX_TEXTURE_COORD_UNITS
];
125 static GLuint
translate_fog_mode( GLenum mode
)
128 case GL_LINEAR
: return FOG_LINEAR
;
129 case GL_EXP
: return FOG_EXP
;
130 case GL_EXP2
: return FOG_EXP2
;
131 default: return FOG_NONE
;
135 #define OPR_SRC_COLOR 0
136 #define OPR_ONE_MINUS_SRC_COLOR 1
137 #define OPR_SRC_ALPHA 2
138 #define OPR_ONE_MINUS_SRC_ALPHA 3
141 #define OPR_UNKNOWN 7
143 static GLuint
translate_operand( GLenum operand
)
146 case GL_SRC_COLOR
: return OPR_SRC_COLOR
;
147 case GL_ONE_MINUS_SRC_COLOR
: return OPR_ONE_MINUS_SRC_COLOR
;
148 case GL_SRC_ALPHA
: return OPR_SRC_ALPHA
;
149 case GL_ONE_MINUS_SRC_ALPHA
: return OPR_ONE_MINUS_SRC_ALPHA
;
150 case GL_ZERO
: return OPR_ZERO
;
151 case GL_ONE
: return OPR_ONE
;
158 #define SRC_TEXTURE 0
159 #define SRC_TEXTURE0 1
160 #define SRC_TEXTURE1 2
161 #define SRC_TEXTURE2 3
162 #define SRC_TEXTURE3 4
163 #define SRC_TEXTURE4 5
164 #define SRC_TEXTURE5 6
165 #define SRC_TEXTURE6 7
166 #define SRC_TEXTURE7 8
167 #define SRC_CONSTANT 9
168 #define SRC_PRIMARY_COLOR 10
169 #define SRC_PREVIOUS 11
171 #define SRC_UNKNOWN 15
173 static GLuint
translate_source( GLenum src
)
176 case GL_TEXTURE
: return SRC_TEXTURE
;
184 case GL_TEXTURE7
: return SRC_TEXTURE0
+ (src
- GL_TEXTURE0
);
185 case GL_CONSTANT
: return SRC_CONSTANT
;
186 case GL_PRIMARY_COLOR
: return SRC_PRIMARY_COLOR
;
187 case GL_PREVIOUS
: return SRC_PREVIOUS
;
196 #define MODE_REPLACE 0 /* r = a0 */
197 #define MODE_MODULATE 1 /* r = a0 * a1 */
198 #define MODE_ADD 2 /* r = a0 + a1 */
199 #define MODE_ADD_SIGNED 3 /* r = a0 + a1 - 0.5 */
200 #define MODE_INTERPOLATE 4 /* r = a0 * a2 + a1 * (1 - a2) */
201 #define MODE_SUBTRACT 5 /* r = a0 - a1 */
202 #define MODE_DOT3_RGB 6 /* r = a0 . a1 */
203 #define MODE_DOT3_RGB_EXT 7 /* r = a0 . a1 */
204 #define MODE_DOT3_RGBA 8 /* r = a0 . a1 */
205 #define MODE_DOT3_RGBA_EXT 9 /* r = a0 . a1 */
206 #define MODE_MODULATE_ADD_ATI 10 /* r = a0 * a2 + a1 */
207 #define MODE_MODULATE_SIGNED_ADD_ATI 11 /* r = a0 * a2 + a1 - 0.5 */
208 #define MODE_MODULATE_SUBTRACT_ATI 12 /* r = a0 * a2 - a1 */
209 #define MODE_ADD_PRODUCTS 13 /* r = a0 * a1 + a2 * a3 */
210 #define MODE_ADD_PRODUCTS_SIGNED 14 /* r = a0 * a1 + a2 * a3 - 0.5 */
211 #define MODE_UNKNOWN 16
214 * Translate GL combiner state into a MODE_x value
216 static GLuint
translate_mode( GLenum envMode
, GLenum mode
)
219 case GL_REPLACE
: return MODE_REPLACE
;
220 case GL_MODULATE
: return MODE_MODULATE
;
222 if (envMode
== GL_COMBINE4_NV
)
223 return MODE_ADD_PRODUCTS
;
227 if (envMode
== GL_COMBINE4_NV
)
228 return MODE_ADD_PRODUCTS_SIGNED
;
230 return MODE_ADD_SIGNED
;
231 case GL_INTERPOLATE
: return MODE_INTERPOLATE
;
232 case GL_SUBTRACT
: return MODE_SUBTRACT
;
233 case GL_DOT3_RGB
: return MODE_DOT3_RGB
;
234 case GL_DOT3_RGB_EXT
: return MODE_DOT3_RGB_EXT
;
235 case GL_DOT3_RGBA
: return MODE_DOT3_RGBA
;
236 case GL_DOT3_RGBA_EXT
: return MODE_DOT3_RGBA_EXT
;
237 case GL_MODULATE_ADD_ATI
: return MODE_MODULATE_ADD_ATI
;
238 case GL_MODULATE_SIGNED_ADD_ATI
: return MODE_MODULATE_SIGNED_ADD_ATI
;
239 case GL_MODULATE_SUBTRACT_ATI
: return MODE_MODULATE_SUBTRACT_ATI
;
248 * Do we need to clamp the results of the given texture env/combine mode?
249 * If the inputs to the mode are in [0,1] we don't always have to clamp
253 need_saturate( GLuint mode
)
258 case MODE_INTERPOLATE
:
261 case MODE_ADD_SIGNED
:
264 case MODE_DOT3_RGB_EXT
:
266 case MODE_DOT3_RGBA_EXT
:
267 case MODE_MODULATE_ADD_ATI
:
268 case MODE_MODULATE_SIGNED_ADD_ATI
:
269 case MODE_MODULATE_SUBTRACT_ATI
:
270 case MODE_ADD_PRODUCTS
:
271 case MODE_ADD_PRODUCTS_SIGNED
:
279 #define VERT_BIT_TEX_ANY (0xff << VERT_ATTRIB_TEX0)
282 * Identify all possible varying inputs. The fragment program will
283 * never reference non-varying inputs, but will track them via state
286 * This function figures out all the inputs that the fragment program
287 * has access to and filters input bitmask.
289 static GLbitfield
filter_fp_input_mask( GLbitfield fp_inputs
,
290 struct gl_context
*ctx
)
292 if (ctx
->VertexProgram
._Overriden
) {
293 /* Somebody's messing with the vertex program and we don't have
294 * a clue what's happening. Assume that it could be producing
295 * all possible outputs.
300 if (ctx
->RenderMode
== GL_FEEDBACK
) {
301 /* _NEW_RENDERMODE */
302 return fp_inputs
& (VARYING_BIT_COL0
| VARYING_BIT_TEX0
);
306 const GLboolean vertexShader
=
307 ctx
->_Shader
->CurrentProgram
[MESA_SHADER_VERTEX
] != NULL
;
308 const GLboolean vertexProgram
= ctx
->VertexProgram
._Enabled
;
310 if (!(vertexProgram
|| vertexShader
)) {
311 /* Fixed function vertex logic */
312 GLbitfield possible_inputs
= 0;
314 /* _NEW_VARYING_VP_INPUTS */
315 GLbitfield64 varying_inputs
= ctx
->varying_vp_inputs
;
317 /* These get generated in the setup routine regardless of the
321 if (ctx
->Point
.PointSprite
) {
322 /* All texture varyings are possible to use */
323 possible_inputs
= VARYING_BITS_TEX_ANY
;
326 /* _NEW_TEXTURE_STATE */
327 const GLbitfield possible_tex_inputs
=
328 ctx
->Texture
._TexGenEnabled
|
329 ctx
->Texture
._TexMatEnabled
|
330 ((varying_inputs
& VERT_BIT_TEX_ANY
) >> VERT_ATTRIB_TEX0
);
332 possible_inputs
= (possible_tex_inputs
<< VARYING_SLOT_TEX0
);
335 /* First look at what values may be computed by the generated
339 if (ctx
->Light
.Enabled
) {
340 possible_inputs
|= VARYING_BIT_COL0
;
342 if (texenv_doing_secondary_color(ctx
))
343 possible_inputs
|= VARYING_BIT_COL1
;
346 /* Then look at what might be varying as a result of enabled
349 if (varying_inputs
& VERT_BIT_COLOR0
)
350 possible_inputs
|= VARYING_BIT_COL0
;
351 if (varying_inputs
& VERT_BIT_COLOR1
)
352 possible_inputs
|= VARYING_BIT_COL1
;
354 return fp_inputs
& possible_inputs
;
357 /* calculate from vp->outputs */
358 struct gl_program
*vprog
;
360 /* Choose GLSL vertex shader over ARB vertex program. Need this
361 * since vertex shader state validation comes after fragment state
362 * validation (see additional comments in state.c).
365 vprog
= ctx
->_Shader
->CurrentProgram
[MESA_SHADER_VERTEX
];
367 vprog
= ctx
->VertexProgram
.Current
;
369 GLbitfield possible_inputs
= vprog
->info
.outputs_written
;
371 /* These get generated in the setup routine regardless of the
375 if (ctx
->Point
.PointSprite
) {
376 /* All texture varyings are possible to use */
377 possible_inputs
|= VARYING_BITS_TEX_ANY
;
380 return fp_inputs
& possible_inputs
;
385 * Examine current texture environment state and generate a unique
386 * key to identify it.
388 static GLuint
make_state_key( struct gl_context
*ctx
, struct state_key
*key
)
391 GLbitfield inputs_referenced
= VARYING_BIT_COL0
;
395 memset(key
, 0, sizeof(*key
));
397 /* _NEW_TEXTURE_OBJECT */
398 mask
= ctx
->Texture
._EnabledCoordUnits
;
401 i
= u_bit_scan(&mask
);
402 const struct gl_texture_unit
*texUnit
= &ctx
->Texture
.Unit
[i
];
403 const struct gl_texture_object
*texObj
= texUnit
->_Current
;
404 const struct gl_tex_env_combine_state
*comb
= texUnit
->_CurrentCombine
;
405 const struct gl_sampler_object
*samp
;
411 samp
= _mesa_get_samplerobj(ctx
, i
);
412 format
= _mesa_texture_base_format(texObj
);
414 key
->unit
[i
].enabled
= 1;
415 inputs_referenced
|= VARYING_BIT_TEX(i
);
417 key
->unit
[i
].source_index
= texObj
->TargetIndex
;
419 key
->unit
[i
].shadow
=
420 ((samp
->CompareMode
== GL_COMPARE_R_TO_TEXTURE
) &&
421 ((format
== GL_DEPTH_COMPONENT
) ||
422 (format
== GL_DEPTH_STENCIL_EXT
)));
424 key
->unit
[i
].NumArgsRGB
= comb
->_NumArgsRGB
;
425 key
->unit
[i
].NumArgsA
= comb
->_NumArgsA
;
427 key
->unit
[i
].ModeRGB
=
428 translate_mode(texUnit
->EnvMode
, comb
->ModeRGB
);
430 translate_mode(texUnit
->EnvMode
, comb
->ModeA
);
432 key
->unit
[i
].ScaleShiftRGB
= comb
->ScaleShiftRGB
;
433 key
->unit
[i
].ScaleShiftA
= comb
->ScaleShiftA
;
435 for (j
= 0; j
< MAX_COMBINER_TERMS
; j
++) {
436 key
->unit
[i
].OptRGB
[j
].Operand
= translate_operand(comb
->OperandRGB
[j
]);
437 key
->unit
[i
].OptA
[j
].Operand
= translate_operand(comb
->OperandA
[j
]);
438 key
->unit
[i
].OptRGB
[j
].Source
= translate_source(comb
->SourceRGB
[j
]);
439 key
->unit
[i
].OptA
[j
].Source
= translate_source(comb
->SourceA
[j
]);
443 key
->nr_enabled_units
= i
+ 1;
445 /* _NEW_LIGHT | _NEW_FOG */
446 if (texenv_doing_secondary_color(ctx
)) {
447 key
->separate_specular
= 1;
448 inputs_referenced
|= VARYING_BIT_COL1
;
452 if (ctx
->Fog
.Enabled
)
453 key
->fog_mode
= translate_fog_mode(ctx
->Fog
.Mode
);
456 key
->num_draw_buffers
= ctx
->DrawBuffer
->_NumColorDrawBuffers
;
459 if (ctx
->Color
.AlphaEnabled
&& key
->num_draw_buffers
== 0) {
460 /* if alpha test is enabled we need to emit at least one color */
461 key
->num_draw_buffers
= 1;
464 key
->inputs_available
= filter_fp_input_mask(inputs_referenced
, ctx
);
466 /* compute size of state key, ignoring unused texture units */
467 keySize
= sizeof(*key
) - sizeof(key
->unit
)
468 + key
->nr_enabled_units
* sizeof(key
->unit
[0]);
474 /** State used to build the fragment program:
476 class texenv_fragment_program
: public ir_factory
{
478 struct gl_shader_program
*shader_program
;
479 struct gl_shader
*shader
;
480 exec_list
*top_instructions
;
481 struct state_key
*state
;
483 ir_variable
*src_texture
[MAX_TEXTURE_COORD_UNITS
];
484 /* Reg containing each texture unit's sampled texture color,
488 /* Texcoord override from bumpmapping. */
489 ir_variable
*texcoord_tex
[MAX_TEXTURE_COORD_UNITS
];
491 /* Reg containing texcoord for a texture unit,
492 * needed for bump mapping, else undef.
495 ir_rvalue
*src_previous
; /**< Reg containing color from previous
496 * stage. May need to be decl'd.
501 get_current_attrib(texenv_fragment_program
*p
, GLuint attrib
)
503 ir_variable
*current
;
506 current
= p
->shader
->symbols
->get_variable("gl_CurrentAttribFragMESA");
508 current
->data
.max_array_access
= MAX2(current
->data
.max_array_access
, (int)attrib
);
509 val
= new(p
->mem_ctx
) ir_dereference_variable(current
);
510 ir_rvalue
*index
= new(p
->mem_ctx
) ir_constant(attrib
);
511 return new(p
->mem_ctx
) ir_dereference_array(val
, index
);
515 get_gl_Color(texenv_fragment_program
*p
)
517 if (p
->state
->inputs_available
& VARYING_BIT_COL0
) {
518 ir_variable
*var
= p
->shader
->symbols
->get_variable("gl_Color");
520 return new(p
->mem_ctx
) ir_dereference_variable(var
);
522 return get_current_attrib(p
, VERT_ATTRIB_COLOR0
);
527 get_source(texenv_fragment_program
*p
,
528 GLuint src
, GLuint unit
)
531 ir_dereference
*deref
;
535 return new(p
->mem_ctx
) ir_dereference_variable(p
->src_texture
[unit
]);
545 return new(p
->mem_ctx
)
546 ir_dereference_variable(p
->src_texture
[src
- SRC_TEXTURE0
]);
549 var
= p
->shader
->symbols
->get_variable("gl_TextureEnvColor");
551 deref
= new(p
->mem_ctx
) ir_dereference_variable(var
);
552 var
->data
.max_array_access
= MAX2(var
->data
.max_array_access
, (int)unit
);
553 return new(p
->mem_ctx
) ir_dereference_array(deref
,
554 new(p
->mem_ctx
) ir_constant(unit
));
556 case SRC_PRIMARY_COLOR
:
557 var
= p
->shader
->symbols
->get_variable("gl_Color");
559 return new(p
->mem_ctx
) ir_dereference_variable(var
);
562 return new(p
->mem_ctx
) ir_constant(0.0f
);
565 if (!p
->src_previous
) {
566 return get_gl_Color(p
);
568 return p
->src_previous
->clone(p
->mem_ctx
, NULL
);
578 emit_combine_source(texenv_fragment_program
*p
,
585 src
= get_source(p
, source
, unit
);
588 case OPR_ONE_MINUS_SRC_COLOR
:
589 return sub(new(p
->mem_ctx
) ir_constant(1.0f
), src
);
592 return src
->type
->is_scalar() ? src
: swizzle_w(src
);
594 case OPR_ONE_MINUS_SRC_ALPHA
: {
595 ir_rvalue
*const scalar
= src
->type
->is_scalar() ? src
: swizzle_w(src
);
597 return sub(new(p
->mem_ctx
) ir_constant(1.0f
), scalar
);
601 return new(p
->mem_ctx
) ir_constant(0.0f
);
603 return new(p
->mem_ctx
) ir_constant(1.0f
);
613 * Check if the RGB and Alpha sources and operands match for the given
614 * texture unit's combinder state. When the RGB and A sources and
615 * operands match, we can emit fewer instructions.
617 static GLboolean
args_match( const struct state_key
*key
, GLuint unit
)
619 GLuint i
, numArgs
= key
->unit
[unit
].NumArgsRGB
;
621 for (i
= 0; i
< numArgs
; i
++) {
622 if (key
->unit
[unit
].OptA
[i
].Source
!= key
->unit
[unit
].OptRGB
[i
].Source
)
625 switch (key
->unit
[unit
].OptA
[i
].Operand
) {
627 switch (key
->unit
[unit
].OptRGB
[i
].Operand
) {
635 case OPR_ONE_MINUS_SRC_ALPHA
:
636 switch (key
->unit
[unit
].OptRGB
[i
].Operand
) {
637 case OPR_ONE_MINUS_SRC_COLOR
:
638 case OPR_ONE_MINUS_SRC_ALPHA
:
645 return GL_FALSE
; /* impossible */
653 smear(ir_rvalue
*val
)
655 if (!val
->type
->is_scalar())
658 return swizzle_xxxx(val
);
662 emit_combine(texenv_fragment_program
*p
,
666 const struct mode_opt
*opt
)
668 ir_rvalue
*src
[MAX_COMBINER_TERMS
];
669 ir_rvalue
*tmp0
, *tmp1
;
672 assert(nr
<= MAX_COMBINER_TERMS
);
674 for (i
= 0; i
< nr
; i
++)
675 src
[i
] = emit_combine_source( p
, unit
, opt
[i
].Source
, opt
[i
].Operand
);
682 return mul(src
[0], src
[1]);
685 return add(src
[0], src
[1]);
687 case MODE_ADD_SIGNED
:
688 return add(add(src
[0], src
[1]), new(p
->mem_ctx
) ir_constant(-0.5f
));
690 case MODE_INTERPOLATE
:
691 /* Arg0 * (Arg2) + Arg1 * (1-Arg2) */
692 tmp0
= mul(src
[0], src
[2]);
693 tmp1
= mul(src
[1], sub(new(p
->mem_ctx
) ir_constant(1.0f
),
694 src
[2]->clone(p
->mem_ctx
, NULL
)));
695 return add(tmp0
, tmp1
);
698 return sub(src
[0], src
[1]);
701 case MODE_DOT3_RGBA_EXT
:
702 case MODE_DOT3_RGB_EXT
:
703 case MODE_DOT3_RGB
: {
704 tmp0
= mul(src
[0], new(p
->mem_ctx
) ir_constant(2.0f
));
705 tmp0
= add(tmp0
, new(p
->mem_ctx
) ir_constant(-1.0f
));
707 tmp1
= mul(src
[1], new(p
->mem_ctx
) ir_constant(2.0f
));
708 tmp1
= add(tmp1
, new(p
->mem_ctx
) ir_constant(-1.0f
));
710 return dot(swizzle_xyz(smear(tmp0
)), swizzle_xyz(smear(tmp1
)));
712 case MODE_MODULATE_ADD_ATI
:
713 return add(mul(src
[0], src
[2]), src
[1]);
715 case MODE_MODULATE_SIGNED_ADD_ATI
:
716 return add(add(mul(src
[0], src
[2]), src
[1]),
717 new(p
->mem_ctx
) ir_constant(-0.5f
));
719 case MODE_MODULATE_SUBTRACT_ATI
:
720 return sub(mul(src
[0], src
[2]), src
[1]);
722 case MODE_ADD_PRODUCTS
:
723 return add(mul(src
[0], src
[1]), mul(src
[2], src
[3]));
725 case MODE_ADD_PRODUCTS_SIGNED
:
726 return add(add(mul(src
[0], src
[1]), mul(src
[2], src
[3])),
727 new(p
->mem_ctx
) ir_constant(-0.5f
));
735 * Generate instructions for one texture unit's env/combiner mode.
738 emit_texenv(texenv_fragment_program
*p
, GLuint unit
)
740 const struct state_key
*key
= p
->state
;
741 GLboolean rgb_saturate
, alpha_saturate
;
742 GLuint rgb_shift
, alpha_shift
;
744 if (!key
->unit
[unit
].enabled
) {
745 return get_source(p
, SRC_PREVIOUS
, 0);
748 switch (key
->unit
[unit
].ModeRGB
) {
749 case MODE_DOT3_RGB_EXT
:
750 alpha_shift
= key
->unit
[unit
].ScaleShiftA
;
753 case MODE_DOT3_RGBA_EXT
:
758 rgb_shift
= key
->unit
[unit
].ScaleShiftRGB
;
759 alpha_shift
= key
->unit
[unit
].ScaleShiftA
;
763 /* If we'll do rgb/alpha shifting don't saturate in emit_combine().
764 * We don't want to clamp twice.
767 rgb_saturate
= GL_FALSE
; /* saturate after rgb shift */
768 else if (need_saturate(key
->unit
[unit
].ModeRGB
))
769 rgb_saturate
= GL_TRUE
;
771 rgb_saturate
= GL_FALSE
;
774 alpha_saturate
= GL_FALSE
; /* saturate after alpha shift */
775 else if (need_saturate(key
->unit
[unit
].ModeA
))
776 alpha_saturate
= GL_TRUE
;
778 alpha_saturate
= GL_FALSE
;
780 ir_variable
*temp_var
= p
->make_temp(glsl_type::vec4_type
, "texenv_combine");
781 ir_dereference
*deref
;
784 /* Emit the RGB and A combine ops
786 if (key
->unit
[unit
].ModeRGB
== key
->unit
[unit
].ModeA
&&
787 args_match(key
, unit
)) {
788 val
= emit_combine(p
, unit
,
789 key
->unit
[unit
].NumArgsRGB
,
790 key
->unit
[unit
].ModeRGB
,
791 key
->unit
[unit
].OptRGB
);
796 p
->emit(assign(temp_var
, val
));
798 else if (key
->unit
[unit
].ModeRGB
== MODE_DOT3_RGBA_EXT
||
799 key
->unit
[unit
].ModeRGB
== MODE_DOT3_RGBA
) {
800 ir_rvalue
*val
= emit_combine(p
, unit
,
801 key
->unit
[unit
].NumArgsRGB
,
802 key
->unit
[unit
].ModeRGB
,
803 key
->unit
[unit
].OptRGB
);
807 p
->emit(assign(temp_var
, val
));
810 /* Need to do something to stop from re-emitting identical
811 * argument calculations here:
813 val
= emit_combine(p
, unit
,
814 key
->unit
[unit
].NumArgsRGB
,
815 key
->unit
[unit
].ModeRGB
,
816 key
->unit
[unit
].OptRGB
);
817 val
= swizzle_xyz(smear(val
));
820 p
->emit(assign(temp_var
, val
, WRITEMASK_XYZ
));
822 val
= emit_combine(p
, unit
,
823 key
->unit
[unit
].NumArgsA
,
824 key
->unit
[unit
].ModeA
,
825 key
->unit
[unit
].OptA
);
826 val
= swizzle_w(smear(val
));
829 p
->emit(assign(temp_var
, val
, WRITEMASK_W
));
832 deref
= new(p
->mem_ctx
) ir_dereference_variable(temp_var
);
834 /* Deal with the final shift:
836 if (alpha_shift
|| rgb_shift
) {
839 if (rgb_shift
== alpha_shift
) {
840 shift
= new(p
->mem_ctx
) ir_constant((float)(1 << rgb_shift
));
843 ir_constant_data const_data
;
845 const_data
.f
[0] = float(1 << rgb_shift
);
846 const_data
.f
[1] = float(1 << rgb_shift
);
847 const_data
.f
[2] = float(1 << rgb_shift
);
848 const_data
.f
[3] = float(1 << alpha_shift
);
850 shift
= new(p
->mem_ctx
) ir_constant(glsl_type::vec4_type
,
854 return saturate(mul(deref
, shift
));
862 * Generate instruction for getting a texture source term.
864 static void load_texture( texenv_fragment_program
*p
, GLuint unit
)
866 ir_dereference
*deref
;
868 if (p
->src_texture
[unit
])
871 const GLuint texTarget
= p
->state
->unit
[unit
].source_index
;
874 if (!(p
->state
->inputs_available
& (VARYING_BIT_TEX0
<< unit
))) {
875 texcoord
= get_current_attrib(p
, VERT_ATTRIB_TEX0
+ unit
);
876 } else if (p
->texcoord_tex
[unit
]) {
877 texcoord
= new(p
->mem_ctx
) ir_dereference_variable(p
->texcoord_tex
[unit
]);
879 ir_variable
*tc_array
= p
->shader
->symbols
->get_variable("gl_TexCoord");
881 texcoord
= new(p
->mem_ctx
) ir_dereference_variable(tc_array
);
882 ir_rvalue
*index
= new(p
->mem_ctx
) ir_constant(unit
);
883 texcoord
= new(p
->mem_ctx
) ir_dereference_array(texcoord
, index
);
884 tc_array
->data
.max_array_access
= MAX2(tc_array
->data
.max_array_access
, (int)unit
);
887 if (!p
->state
->unit
[unit
].enabled
) {
888 p
->src_texture
[unit
] = p
->make_temp(glsl_type::vec4_type
,
890 p
->emit(p
->src_texture
[unit
]);
892 p
->emit(assign(p
->src_texture
[unit
], new(p
->mem_ctx
) ir_constant(0.0f
)));
896 const glsl_type
*sampler_type
= NULL
;
900 case TEXTURE_1D_INDEX
:
901 if (p
->state
->unit
[unit
].shadow
)
902 sampler_type
= glsl_type::sampler1DShadow_type
;
904 sampler_type
= glsl_type::sampler1D_type
;
907 case TEXTURE_1D_ARRAY_INDEX
:
908 if (p
->state
->unit
[unit
].shadow
)
909 sampler_type
= glsl_type::sampler1DArrayShadow_type
;
911 sampler_type
= glsl_type::sampler1DArray_type
;
914 case TEXTURE_2D_INDEX
:
915 if (p
->state
->unit
[unit
].shadow
)
916 sampler_type
= glsl_type::sampler2DShadow_type
;
918 sampler_type
= glsl_type::sampler2D_type
;
921 case TEXTURE_2D_ARRAY_INDEX
:
922 if (p
->state
->unit
[unit
].shadow
)
923 sampler_type
= glsl_type::sampler2DArrayShadow_type
;
925 sampler_type
= glsl_type::sampler2DArray_type
;
928 case TEXTURE_RECT_INDEX
:
929 if (p
->state
->unit
[unit
].shadow
)
930 sampler_type
= glsl_type::sampler2DRectShadow_type
;
932 sampler_type
= glsl_type::sampler2DRect_type
;
935 case TEXTURE_3D_INDEX
:
936 assert(!p
->state
->unit
[unit
].shadow
);
937 sampler_type
= glsl_type::sampler3D_type
;
940 case TEXTURE_CUBE_INDEX
:
941 if (p
->state
->unit
[unit
].shadow
)
942 sampler_type
= glsl_type::samplerCubeShadow_type
;
944 sampler_type
= glsl_type::samplerCube_type
;
947 case TEXTURE_EXTERNAL_INDEX
:
948 assert(!p
->state
->unit
[unit
].shadow
);
949 sampler_type
= glsl_type::samplerExternalOES_type
;
954 p
->src_texture
[unit
] = p
->make_temp(glsl_type::vec4_type
,
957 ir_texture
*tex
= new(p
->mem_ctx
) ir_texture(ir_tex
);
960 char *sampler_name
= ralloc_asprintf(p
->mem_ctx
, "sampler_%d", unit
);
961 ir_variable
*sampler
= new(p
->mem_ctx
) ir_variable(sampler_type
,
964 p
->top_instructions
->push_head(sampler
);
966 /* Set the texture unit for this sampler in the same way that
967 * layout(binding=X) would.
969 sampler
->data
.explicit_binding
= true;
970 sampler
->data
.binding
= unit
;
972 deref
= new(p
->mem_ctx
) ir_dereference_variable(sampler
);
973 tex
->set_sampler(deref
, glsl_type::vec4_type
);
975 tex
->coordinate
= new(p
->mem_ctx
) ir_swizzle(texcoord
, 0, 1, 2, 3, coords
);
977 if (p
->state
->unit
[unit
].shadow
) {
978 texcoord
= texcoord
->clone(p
->mem_ctx
, NULL
);
979 tex
->shadow_comparator
= new(p
->mem_ctx
) ir_swizzle(texcoord
,
985 texcoord
= texcoord
->clone(p
->mem_ctx
, NULL
);
986 tex
->projector
= swizzle_w(texcoord
);
988 p
->emit(assign(p
->src_texture
[unit
], tex
));
992 load_texenv_source(texenv_fragment_program
*p
,
993 GLuint src
, GLuint unit
)
997 load_texture(p
, unit
);
1008 load_texture(p
, src
- SRC_TEXTURE0
);
1012 /* not a texture src - do nothing */
1019 * Generate instructions for loading all texture source terms.
1022 load_texunit_sources( texenv_fragment_program
*p
, GLuint unit
)
1024 const struct state_key
*key
= p
->state
;
1027 for (i
= 0; i
< key
->unit
[unit
].NumArgsRGB
; i
++) {
1028 load_texenv_source( p
, key
->unit
[unit
].OptRGB
[i
].Source
, unit
);
1031 for (i
= 0; i
< key
->unit
[unit
].NumArgsA
; i
++) {
1032 load_texenv_source( p
, key
->unit
[unit
].OptA
[i
].Source
, unit
);
1039 * Applies the fog calculations.
1041 * This is basically like the ARB_fragment_prorgam fog options. Note
1042 * that ffvertex_prog.c produces fogcoord for us when
1043 * GL_FOG_COORDINATE_EXT is set to GL_FRAGMENT_DEPTH_EXT.
1046 emit_fog_instructions(texenv_fragment_program
*p
,
1047 ir_rvalue
*fragcolor
)
1049 struct state_key
*key
= p
->state
;
1050 ir_rvalue
*f
, *temp
;
1051 ir_variable
*params
, *oparams
;
1052 ir_variable
*fogcoord
;
1054 /* Temporary storage for the whole fog result. Fog calculations
1055 * only affect rgb so we're hanging on to the .a value of fragcolor
1058 ir_variable
*fog_result
= p
->make_temp(glsl_type::vec4_type
, "fog_result");
1059 p
->emit(assign(fog_result
, fragcolor
));
1061 fragcolor
= swizzle_xyz(fog_result
);
1063 oparams
= p
->shader
->symbols
->get_variable("gl_FogParamsOptimizedMESA");
1065 fogcoord
= p
->shader
->symbols
->get_variable("gl_FogFragCoord");
1067 params
= p
->shader
->symbols
->get_variable("gl_Fog");
1069 f
= new(p
->mem_ctx
) ir_dereference_variable(fogcoord
);
1071 ir_variable
*f_var
= p
->make_temp(glsl_type::float_type
, "fog_factor");
1073 switch (key
->fog_mode
) {
1075 /* f = (end - z) / (end - start)
1077 * gl_MesaFogParamsOptimized gives us (-1 / (end - start)) and
1078 * (end / (end - start)) so we can generate a single MAD.
1080 f
= add(mul(f
, swizzle_x(oparams
)), swizzle_y(oparams
));
1083 /* f = e^(-(density * fogcoord))
1085 * gl_MesaFogParamsOptimized gives us density/ln(2) so we can
1086 * use EXP2 which is generally the native instruction without
1087 * having to do any further math on the fog density uniform.
1089 f
= mul(f
, swizzle_z(oparams
));
1090 f
= new(p
->mem_ctx
) ir_expression(ir_unop_neg
, f
);
1091 f
= new(p
->mem_ctx
) ir_expression(ir_unop_exp2
, f
);
1094 /* f = e^(-(density * fogcoord)^2)
1096 * gl_MesaFogParamsOptimized gives us density/sqrt(ln(2)) so we
1097 * can do this like FOG_EXP but with a squaring after the
1098 * multiply by density.
1100 ir_variable
*temp_var
= p
->make_temp(glsl_type::float_type
, "fog_temp");
1101 p
->emit(assign(temp_var
, mul(f
, swizzle_w(oparams
))));
1103 f
= mul(temp_var
, temp_var
);
1104 f
= new(p
->mem_ctx
) ir_expression(ir_unop_neg
, f
);
1105 f
= new(p
->mem_ctx
) ir_expression(ir_unop_exp2
, f
);
1109 p
->emit(assign(f_var
, saturate(f
)));
1111 f
= sub(new(p
->mem_ctx
) ir_constant(1.0f
), f_var
);
1112 temp
= new(p
->mem_ctx
) ir_dereference_variable(params
);
1113 temp
= new(p
->mem_ctx
) ir_dereference_record(temp
, "color");
1114 temp
= mul(swizzle_xyz(temp
), f
);
1116 p
->emit(assign(fog_result
, add(temp
, mul(fragcolor
, f_var
)), WRITEMASK_XYZ
));
1118 return new(p
->mem_ctx
) ir_dereference_variable(fog_result
);
1122 emit_instructions(texenv_fragment_program
*p
)
1124 struct state_key
*key
= p
->state
;
1127 if (key
->nr_enabled_units
) {
1128 /* First pass - to support texture_env_crossbar, first identify
1129 * all referenced texture sources and emit texld instructions
1132 for (unit
= 0; unit
< key
->nr_enabled_units
; unit
++)
1133 if (key
->unit
[unit
].enabled
) {
1134 load_texunit_sources(p
, unit
);
1137 /* Second pass - emit combine instructions to build final color:
1139 for (unit
= 0; unit
< key
->nr_enabled_units
; unit
++) {
1140 if (key
->unit
[unit
].enabled
) {
1141 p
->src_previous
= emit_texenv(p
, unit
);
1146 ir_rvalue
*cf
= get_source(p
, SRC_PREVIOUS
, 0);
1148 if (key
->separate_specular
) {
1149 ir_variable
*spec_result
= p
->make_temp(glsl_type::vec4_type
,
1151 p
->emit(assign(spec_result
, cf
));
1153 ir_rvalue
*secondary
;
1154 if (p
->state
->inputs_available
& VARYING_BIT_COL1
) {
1156 p
->shader
->symbols
->get_variable("gl_SecondaryColor");
1158 secondary
= swizzle_xyz(var
);
1160 secondary
= swizzle_xyz(get_current_attrib(p
, VERT_ATTRIB_COLOR1
));
1163 p
->emit(assign(spec_result
, add(swizzle_xyz(spec_result
), secondary
),
1166 cf
= new(p
->mem_ctx
) ir_dereference_variable(spec_result
);
1169 if (key
->fog_mode
) {
1170 cf
= emit_fog_instructions(p
, cf
);
1173 ir_variable
*frag_color
= p
->shader
->symbols
->get_variable("gl_FragColor");
1175 p
->emit(assign(frag_color
, cf
));
1179 * Generate a new fragment program which implements the context's
1180 * current texture env/combine mode.
1182 static struct gl_shader_program
*
1183 create_new_program(struct gl_context
*ctx
, struct state_key
*key
)
1185 texenv_fragment_program p
;
1187 _mesa_glsl_parse_state
*state
;
1189 p
.mem_ctx
= ralloc_context(NULL
);
1190 p
.shader
= _mesa_new_shader(0, MESA_SHADER_FRAGMENT
);
1192 p
.shader
->SourceChecksum
= 0xf18ed; /* fixed */
1194 p
.shader
->ir
= new(p
.shader
) exec_list
;
1195 state
= new(p
.shader
) _mesa_glsl_parse_state(ctx
, MESA_SHADER_FRAGMENT
,
1197 p
.shader
->symbols
= state
->symbols
;
1198 p
.top_instructions
= p
.shader
->ir
;
1199 p
.instructions
= p
.shader
->ir
;
1201 p
.shader_program
= _mesa_new_shader_program(0);
1203 /* Tell the linker to ignore the fact that we're building a
1204 * separate shader, in case we're in a GLES2 context that would
1205 * normally reject that. The real problem is that we're building a
1206 * fixed function program in a GLES2 context at all, but that's a
1207 * big mess to clean up.
1209 p
.shader_program
->SeparateShader
= GL_TRUE
;
1211 /* The legacy GLSL shadow functions follow the depth texture
1212 * mode and return vec4. The GLSL 1.30 shadow functions return float and
1213 * ignore the depth texture mode. That's a shader and state dependency
1214 * that's difficult to deal with. st/mesa uses a simple but not
1215 * completely correct solution: if the shader declares GLSL >= 1.30 and
1216 * the depth texture mode is GL_ALPHA (000X), it sets the XXXX swizzle
1217 * instead. Thus, the GLSL 1.30 shadow function will get the result in .x
1218 * and legacy shadow functions will get it in .w as expected.
1219 * For the fixed-function fragment shader, use 120 to get correct behavior
1222 state
->language_version
= 120;
1224 state
->es_shader
= false;
1225 if (_mesa_is_gles(ctx
) && ctx
->Extensions
.OES_EGL_image_external
)
1226 state
->OES_EGL_image_external_enable
= true;
1227 _mesa_glsl_initialize_types(state
);
1228 _mesa_glsl_initialize_variables(p
.instructions
, state
);
1230 for (unit
= 0; unit
< ctx
->Const
.MaxTextureUnits
; unit
++) {
1231 p
.src_texture
[unit
] = NULL
;
1232 p
.texcoord_tex
[unit
] = NULL
;
1235 p
.src_previous
= NULL
;
1237 ir_function
*main_f
= new(p
.mem_ctx
) ir_function("main");
1239 state
->symbols
->add_function(main_f
);
1241 ir_function_signature
*main_sig
=
1242 new(p
.mem_ctx
) ir_function_signature(glsl_type::void_type
);
1243 main_sig
->is_defined
= true;
1244 main_f
->add_signature(main_sig
);
1246 p
.instructions
= &main_sig
->body
;
1247 if (key
->num_draw_buffers
)
1248 emit_instructions(&p
);
1250 validate_ir_tree(p
.shader
->ir
);
1252 const struct gl_shader_compiler_options
*options
=
1253 &ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_FRAGMENT
];
1255 /* Conservative approach: Don't optimize here, the linker does it too. */
1256 if (!ctx
->Const
.GLSLOptimizeConservatively
) {
1257 while (do_common_optimization(p
.shader
->ir
, false, false, options
,
1258 ctx
->Const
.NativeIntegers
))
1262 reparent_ir(p
.shader
->ir
, p
.shader
->ir
);
1264 p
.shader
->CompileStatus
= compile_success
;
1265 p
.shader
->Version
= state
->language_version
;
1266 p
.shader_program
->Shaders
=
1267 (gl_shader
**)malloc(sizeof(*p
.shader_program
->Shaders
));
1268 p
.shader_program
->Shaders
[0] = p
.shader
;
1269 p
.shader_program
->NumShaders
= 1;
1271 _mesa_glsl_link_shader(ctx
, p
.shader_program
);
1273 if (!p
.shader_program
->data
->LinkStatus
)
1274 _mesa_problem(ctx
, "Failed to link fixed function fragment shader: %s\n",
1275 p
.shader_program
->data
->InfoLog
);
1277 ralloc_free(p
.mem_ctx
);
1278 return p
.shader_program
;
1284 * Return a fragment program which implements the current
1285 * fixed-function texture, fog and color-sum operations.
1287 struct gl_shader_program
*
1288 _mesa_get_fixed_func_fragment_program(struct gl_context
*ctx
)
1290 struct gl_shader_program
*shader_program
;
1291 struct state_key key
;
1294 keySize
= make_state_key(ctx
, &key
);
1296 shader_program
= (struct gl_shader_program
*)
1297 _mesa_search_program_cache(ctx
->FragmentProgram
.Cache
,
1300 if (!shader_program
) {
1301 shader_program
= create_new_program(ctx
, &key
);
1303 _mesa_shader_cache_insert(ctx
, ctx
->FragmentProgram
.Cache
,
1304 &key
, keySize
, shader_program
);
1307 return shader_program
;