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/state.h"
37 #include "main/texenvprogram.h"
38 #include "main/texobj.h"
39 #include "main/uniforms.h"
40 #include "compiler/glsl/ir_builder.h"
41 #include "compiler/glsl/ir_optimization.h"
42 #include "compiler/glsl/glsl_parser_extras.h"
43 #include "compiler/glsl/glsl_symbol_table.h"
44 #include "compiler/glsl_types.h"
45 #include "program/ir_to_mesa.h"
46 #include "program/program.h"
47 #include "program/programopt.h"
48 #include "program/prog_cache.h"
49 #include "program/prog_instruction.h"
50 #include "program/prog_parameter.h"
51 #include "program/prog_print.h"
52 #include "program/prog_statevars.h"
53 #include "util/bitscan.h"
55 using namespace ir_builder
;
58 * Note on texture units:
60 * The number of texture units supported by fixed-function fragment
61 * processing is MAX_TEXTURE_COORD_UNITS, not MAX_TEXTURE_IMAGE_UNITS.
62 * That's because there's a one-to-one correspondence between texture
63 * coordinates and samplers in fixed-function processing.
65 * Since fixed-function vertex processing is limited to MAX_TEXTURE_COORD_UNITS
66 * sets of texcoords, so is fixed-function fragment processing.
68 * We can safely use ctx->Const.MaxTextureUnits for loop bounds.
73 texenv_doing_secondary_color(struct gl_context
*ctx
)
75 if (ctx
->Light
.Enabled
&&
76 (ctx
->Light
.Model
.ColorControl
== GL_SEPARATE_SPECULAR_COLOR
))
79 if (ctx
->Fog
.ColorSumEnabled
)
86 GLuint nr_enabled_units
:4;
87 GLuint separate_specular
:1;
88 GLuint fog_mode
:2; /**< FOG_x */
89 GLuint inputs_available
:12;
90 GLuint num_draw_buffers
:4;
92 /* NOTE: This array of structs must be last! (see "keySize" below) */
95 GLuint source_index
:4; /**< TEXTURE_x_INDEX */
99 * These are taken from struct gl_tex_env_combine_packed
104 GLuint ScaleShiftRGB
:2;
105 GLuint ScaleShiftA
:2;
108 struct gl_tex_env_argument ArgsRGB
[MAX_COMBINER_TERMS
];
109 struct gl_tex_env_argument ArgsA
[MAX_COMBINER_TERMS
];
111 } unit
[MAX_TEXTURE_COORD_UNITS
];
116 * Do we need to clamp the results of the given texture env/combine mode?
117 * If the inputs to the mode are in [0,1] we don't always have to clamp
121 need_saturate( GLuint mode
)
124 case TEXENV_MODE_REPLACE
:
125 case TEXENV_MODE_MODULATE
:
126 case TEXENV_MODE_INTERPOLATE
:
128 case TEXENV_MODE_ADD
:
129 case TEXENV_MODE_ADD_SIGNED
:
130 case TEXENV_MODE_SUBTRACT
:
131 case TEXENV_MODE_DOT3_RGB
:
132 case TEXENV_MODE_DOT3_RGB_EXT
:
133 case TEXENV_MODE_DOT3_RGBA
:
134 case TEXENV_MODE_DOT3_RGBA_EXT
:
135 case TEXENV_MODE_MODULATE_ADD_ATI
:
136 case TEXENV_MODE_MODULATE_SIGNED_ADD_ATI
:
137 case TEXENV_MODE_MODULATE_SUBTRACT_ATI
:
138 case TEXENV_MODE_ADD_PRODUCTS_NV
:
139 case TEXENV_MODE_ADD_PRODUCTS_SIGNED_NV
:
147 #define VERT_BIT_TEX_ANY (0xff << VERT_ATTRIB_TEX0)
150 * Identify all possible varying inputs. The fragment program will
151 * never reference non-varying inputs, but will track them via state
154 * This function figures out all the inputs that the fragment program
155 * has access to and filters input bitmask.
157 static GLbitfield
filter_fp_input_mask( GLbitfield fp_inputs
,
158 struct gl_context
*ctx
)
160 if (ctx
->VertexProgram
._Overriden
) {
161 /* Somebody's messing with the vertex program and we don't have
162 * a clue what's happening. Assume that it could be producing
163 * all possible outputs.
168 if (ctx
->RenderMode
== GL_FEEDBACK
) {
169 /* _NEW_RENDERMODE */
170 return fp_inputs
& (VARYING_BIT_COL0
| VARYING_BIT_TEX0
);
174 const GLboolean vertexShader
=
175 ctx
->_Shader
->CurrentProgram
[MESA_SHADER_VERTEX
] != NULL
;
176 const GLboolean vertexProgram
= _mesa_arb_vertex_program_enabled(ctx
);
178 if (!(vertexProgram
|| vertexShader
)) {
179 /* Fixed function vertex logic */
180 GLbitfield possible_inputs
= 0;
182 /* _NEW_VARYING_VP_INPUTS */
183 GLbitfield varying_inputs
= ctx
->varying_vp_inputs
;
185 /* These get generated in the setup routine regardless of the
189 if (ctx
->Point
.PointSprite
) {
190 /* All texture varyings are possible to use */
191 possible_inputs
= VARYING_BITS_TEX_ANY
;
194 /* _NEW_TEXTURE_STATE */
195 const GLbitfield possible_tex_inputs
=
196 ctx
->Texture
._TexGenEnabled
|
197 ctx
->Texture
._TexMatEnabled
|
198 ((varying_inputs
& VERT_BIT_TEX_ANY
) >> VERT_ATTRIB_TEX0
);
200 possible_inputs
= (possible_tex_inputs
<< VARYING_SLOT_TEX0
);
203 /* First look at what values may be computed by the generated
207 if (ctx
->Light
.Enabled
) {
208 possible_inputs
|= VARYING_BIT_COL0
;
210 if (texenv_doing_secondary_color(ctx
))
211 possible_inputs
|= VARYING_BIT_COL1
;
214 /* Then look at what might be varying as a result of enabled
217 if (varying_inputs
& VERT_BIT_COLOR0
)
218 possible_inputs
|= VARYING_BIT_COL0
;
219 if (varying_inputs
& VERT_BIT_COLOR1
)
220 possible_inputs
|= VARYING_BIT_COL1
;
222 return fp_inputs
& possible_inputs
;
225 /* calculate from vp->outputs */
226 struct gl_program
*vprog
;
228 /* Choose GLSL vertex shader over ARB vertex program. Need this
229 * since vertex shader state validation comes after fragment state
230 * validation (see additional comments in state.c).
233 vprog
= ctx
->_Shader
->CurrentProgram
[MESA_SHADER_VERTEX
];
235 vprog
= ctx
->VertexProgram
.Current
;
237 GLbitfield possible_inputs
= vprog
->info
.outputs_written
;
239 /* These get generated in the setup routine regardless of the
243 if (ctx
->Point
.PointSprite
) {
244 /* All texture varyings are possible to use */
245 possible_inputs
|= VARYING_BITS_TEX_ANY
;
248 return fp_inputs
& possible_inputs
;
253 * Examine current texture environment state and generate a unique
254 * key to identify it.
256 static GLuint
make_state_key( struct gl_context
*ctx
, struct state_key
*key
)
258 GLbitfield inputs_referenced
= VARYING_BIT_COL0
;
262 memset(key
, 0, sizeof(*key
));
264 /* _NEW_TEXTURE_OBJECT */
265 mask
= ctx
->Texture
._EnabledCoordUnits
;
268 i
= u_bit_scan(&mask
);
269 const struct gl_texture_unit
*texUnit
= &ctx
->Texture
.Unit
[i
];
270 const struct gl_texture_object
*texObj
= texUnit
->_Current
;
271 const struct gl_tex_env_combine_packed
*comb
= &texUnit
->_CurrentCombinePacked
;
276 key
->unit
[i
].enabled
= 1;
277 inputs_referenced
|= VARYING_BIT_TEX(i
);
279 key
->unit
[i
].source_index
= texObj
->TargetIndex
;
281 const struct gl_sampler_object
*samp
= _mesa_get_samplerobj(ctx
, i
);
282 if (samp
->CompareMode
== GL_COMPARE_R_TO_TEXTURE
) {
283 const GLenum format
= _mesa_texture_base_format(texObj
);
284 key
->unit
[i
].shadow
= (format
== GL_DEPTH_COMPONENT
||
285 format
== GL_DEPTH_STENCIL_EXT
);
288 key
->unit
[i
].ModeRGB
= comb
->ModeRGB
;
289 key
->unit
[i
].ModeA
= comb
->ModeA
;
290 key
->unit
[i
].ScaleShiftRGB
= comb
->ScaleShiftRGB
;
291 key
->unit
[i
].ScaleShiftA
= comb
->ScaleShiftA
;
292 key
->unit
[i
].NumArgsRGB
= comb
->NumArgsRGB
;
293 key
->unit
[i
].NumArgsA
= comb
->NumArgsA
;
295 memcpy(key
->unit
[i
].ArgsRGB
, comb
->ArgsRGB
, sizeof comb
->ArgsRGB
);
296 memcpy(key
->unit
[i
].ArgsA
, comb
->ArgsA
, sizeof comb
->ArgsA
);
299 key
->nr_enabled_units
= i
+ 1;
301 /* _NEW_LIGHT | _NEW_FOG */
302 if (texenv_doing_secondary_color(ctx
)) {
303 key
->separate_specular
= 1;
304 inputs_referenced
|= VARYING_BIT_COL1
;
308 key
->fog_mode
= ctx
->Fog
._PackedEnabledMode
;
311 key
->num_draw_buffers
= ctx
->DrawBuffer
->_NumColorDrawBuffers
;
314 if (ctx
->Color
.AlphaEnabled
&& key
->num_draw_buffers
== 0) {
315 /* if alpha test is enabled we need to emit at least one color */
316 key
->num_draw_buffers
= 1;
319 key
->inputs_available
= filter_fp_input_mask(inputs_referenced
, ctx
);
321 /* compute size of state key, ignoring unused texture units */
322 keySize
= sizeof(*key
) - sizeof(key
->unit
)
323 + key
->nr_enabled_units
* sizeof(key
->unit
[0]);
329 /** State used to build the fragment program:
331 class texenv_fragment_program
: public ir_factory
{
333 struct gl_shader_program
*shader_program
;
334 struct gl_shader
*shader
;
335 exec_list
*top_instructions
;
336 struct state_key
*state
;
338 ir_variable
*src_texture
[MAX_TEXTURE_COORD_UNITS
];
339 /* Reg containing each texture unit's sampled texture color,
343 /* Texcoord override from bumpmapping. */
344 ir_variable
*texcoord_tex
[MAX_TEXTURE_COORD_UNITS
];
346 /* Reg containing texcoord for a texture unit,
347 * needed for bump mapping, else undef.
350 ir_rvalue
*src_previous
; /**< Reg containing color from previous
351 * stage. May need to be decl'd.
356 get_current_attrib(texenv_fragment_program
*p
, GLuint attrib
)
358 ir_variable
*current
;
361 current
= p
->shader
->symbols
->get_variable("gl_CurrentAttribFragMESA");
363 current
->data
.max_array_access
= MAX2(current
->data
.max_array_access
, (int)attrib
);
364 val
= new(p
->mem_ctx
) ir_dereference_variable(current
);
365 ir_rvalue
*index
= new(p
->mem_ctx
) ir_constant(attrib
);
366 return new(p
->mem_ctx
) ir_dereference_array(val
, index
);
370 get_gl_Color(texenv_fragment_program
*p
)
372 if (p
->state
->inputs_available
& VARYING_BIT_COL0
) {
373 ir_variable
*var
= p
->shader
->symbols
->get_variable("gl_Color");
375 return new(p
->mem_ctx
) ir_dereference_variable(var
);
377 return get_current_attrib(p
, VERT_ATTRIB_COLOR0
);
382 get_source(texenv_fragment_program
*p
,
383 GLuint src
, GLuint unit
)
386 ir_dereference
*deref
;
389 case TEXENV_SRC_TEXTURE
:
390 return new(p
->mem_ctx
) ir_dereference_variable(p
->src_texture
[unit
]);
392 case TEXENV_SRC_TEXTURE0
:
393 case TEXENV_SRC_TEXTURE1
:
394 case TEXENV_SRC_TEXTURE2
:
395 case TEXENV_SRC_TEXTURE3
:
396 case TEXENV_SRC_TEXTURE4
:
397 case TEXENV_SRC_TEXTURE5
:
398 case TEXENV_SRC_TEXTURE6
:
399 case TEXENV_SRC_TEXTURE7
:
400 return new(p
->mem_ctx
)
401 ir_dereference_variable(p
->src_texture
[src
- TEXENV_SRC_TEXTURE0
]);
403 case TEXENV_SRC_CONSTANT
:
404 var
= p
->shader
->symbols
->get_variable("gl_TextureEnvColor");
406 deref
= new(p
->mem_ctx
) ir_dereference_variable(var
);
407 var
->data
.max_array_access
= MAX2(var
->data
.max_array_access
, (int)unit
);
408 return new(p
->mem_ctx
) ir_dereference_array(deref
,
409 new(p
->mem_ctx
) ir_constant(unit
));
411 case TEXENV_SRC_PRIMARY_COLOR
:
412 var
= p
->shader
->symbols
->get_variable("gl_Color");
414 return new(p
->mem_ctx
) ir_dereference_variable(var
);
416 case TEXENV_SRC_ZERO
:
417 return new(p
->mem_ctx
) ir_constant(0.0f
);
420 return new(p
->mem_ctx
) ir_constant(1.0f
);
422 case TEXENV_SRC_PREVIOUS
:
423 if (!p
->src_previous
) {
424 return get_gl_Color(p
);
426 return p
->src_previous
->clone(p
->mem_ctx
, NULL
);
436 emit_combine_source(texenv_fragment_program
*p
,
443 src
= get_source(p
, source
, unit
);
446 case TEXENV_OPR_ONE_MINUS_COLOR
:
447 return sub(new(p
->mem_ctx
) ir_constant(1.0f
), src
);
449 case TEXENV_OPR_ALPHA
:
450 return src
->type
->is_scalar() ? src
: swizzle_w(src
);
452 case TEXENV_OPR_ONE_MINUS_ALPHA
: {
453 ir_rvalue
*const scalar
= src
->type
->is_scalar() ? src
: swizzle_w(src
);
455 return sub(new(p
->mem_ctx
) ir_constant(1.0f
), scalar
);
458 case TEXENV_OPR_COLOR
:
468 * Check if the RGB and Alpha sources and operands match for the given
469 * texture unit's combinder state. When the RGB and A sources and
470 * operands match, we can emit fewer instructions.
472 static GLboolean
args_match( const struct state_key
*key
, GLuint unit
)
474 GLuint i
, numArgs
= key
->unit
[unit
].NumArgsRGB
;
476 for (i
= 0; i
< numArgs
; i
++) {
477 if (key
->unit
[unit
].ArgsA
[i
].Source
!= key
->unit
[unit
].ArgsRGB
[i
].Source
)
480 switch (key
->unit
[unit
].ArgsA
[i
].Operand
) {
481 case TEXENV_OPR_ALPHA
:
482 switch (key
->unit
[unit
].ArgsRGB
[i
].Operand
) {
483 case TEXENV_OPR_COLOR
:
484 case TEXENV_OPR_ALPHA
:
490 case TEXENV_OPR_ONE_MINUS_ALPHA
:
491 switch (key
->unit
[unit
].ArgsRGB
[i
].Operand
) {
492 case TEXENV_OPR_ONE_MINUS_COLOR
:
493 case TEXENV_OPR_ONE_MINUS_ALPHA
:
500 return GL_FALSE
; /* impossible */
508 smear(ir_rvalue
*val
)
510 if (!val
->type
->is_scalar())
513 return swizzle_xxxx(val
);
517 emit_combine(texenv_fragment_program
*p
,
521 const struct gl_tex_env_argument
*opt
)
523 ir_rvalue
*src
[MAX_COMBINER_TERMS
];
524 ir_rvalue
*tmp0
, *tmp1
;
527 assert(nr
<= MAX_COMBINER_TERMS
);
529 for (i
= 0; i
< nr
; i
++)
530 src
[i
] = emit_combine_source( p
, unit
, opt
[i
].Source
, opt
[i
].Operand
);
533 case TEXENV_MODE_REPLACE
:
536 case TEXENV_MODE_MODULATE
:
537 return mul(src
[0], src
[1]);
539 case TEXENV_MODE_ADD
:
540 return add(src
[0], src
[1]);
542 case TEXENV_MODE_ADD_SIGNED
:
543 return add(add(src
[0], src
[1]), new(p
->mem_ctx
) ir_constant(-0.5f
));
545 case TEXENV_MODE_INTERPOLATE
:
546 /* Arg0 * (Arg2) + Arg1 * (1-Arg2) */
547 tmp0
= mul(src
[0], src
[2]);
548 tmp1
= mul(src
[1], sub(new(p
->mem_ctx
) ir_constant(1.0f
),
549 src
[2]->clone(p
->mem_ctx
, NULL
)));
550 return add(tmp0
, tmp1
);
552 case TEXENV_MODE_SUBTRACT
:
553 return sub(src
[0], src
[1]);
555 case TEXENV_MODE_DOT3_RGBA
:
556 case TEXENV_MODE_DOT3_RGBA_EXT
:
557 case TEXENV_MODE_DOT3_RGB_EXT
:
558 case TEXENV_MODE_DOT3_RGB
: {
559 tmp0
= mul(src
[0], new(p
->mem_ctx
) ir_constant(2.0f
));
560 tmp0
= add(tmp0
, new(p
->mem_ctx
) ir_constant(-1.0f
));
562 tmp1
= mul(src
[1], new(p
->mem_ctx
) ir_constant(2.0f
));
563 tmp1
= add(tmp1
, new(p
->mem_ctx
) ir_constant(-1.0f
));
565 return dot(swizzle_xyz(smear(tmp0
)), swizzle_xyz(smear(tmp1
)));
567 case TEXENV_MODE_MODULATE_ADD_ATI
:
568 return add(mul(src
[0], src
[2]), src
[1]);
570 case TEXENV_MODE_MODULATE_SIGNED_ADD_ATI
:
571 return add(add(mul(src
[0], src
[2]), src
[1]),
572 new(p
->mem_ctx
) ir_constant(-0.5f
));
574 case TEXENV_MODE_MODULATE_SUBTRACT_ATI
:
575 return sub(mul(src
[0], src
[2]), src
[1]);
577 case TEXENV_MODE_ADD_PRODUCTS_NV
:
578 return add(mul(src
[0], src
[1]), mul(src
[2], src
[3]));
580 case TEXENV_MODE_ADD_PRODUCTS_SIGNED_NV
:
581 return add(add(mul(src
[0], src
[1]), mul(src
[2], src
[3])),
582 new(p
->mem_ctx
) ir_constant(-0.5f
));
590 * Generate instructions for one texture unit's env/combiner mode.
593 emit_texenv(texenv_fragment_program
*p
, GLuint unit
)
595 const struct state_key
*key
= p
->state
;
596 GLboolean rgb_saturate
, alpha_saturate
;
597 GLuint rgb_shift
, alpha_shift
;
599 if (!key
->unit
[unit
].enabled
) {
600 return get_source(p
, TEXENV_SRC_PREVIOUS
, 0);
603 switch (key
->unit
[unit
].ModeRGB
) {
604 case TEXENV_MODE_DOT3_RGB_EXT
:
605 alpha_shift
= key
->unit
[unit
].ScaleShiftA
;
608 case TEXENV_MODE_DOT3_RGBA_EXT
:
613 rgb_shift
= key
->unit
[unit
].ScaleShiftRGB
;
614 alpha_shift
= key
->unit
[unit
].ScaleShiftA
;
618 /* If we'll do rgb/alpha shifting don't saturate in emit_combine().
619 * We don't want to clamp twice.
622 rgb_saturate
= GL_FALSE
; /* saturate after rgb shift */
623 else if (need_saturate(key
->unit
[unit
].ModeRGB
))
624 rgb_saturate
= GL_TRUE
;
626 rgb_saturate
= GL_FALSE
;
629 alpha_saturate
= GL_FALSE
; /* saturate after alpha shift */
630 else if (need_saturate(key
->unit
[unit
].ModeA
))
631 alpha_saturate
= GL_TRUE
;
633 alpha_saturate
= GL_FALSE
;
635 ir_variable
*temp_var
= p
->make_temp(glsl_type::vec4_type
, "texenv_combine");
636 ir_dereference
*deref
;
639 /* Emit the RGB and A combine ops
641 if (key
->unit
[unit
].ModeRGB
== key
->unit
[unit
].ModeA
&&
642 args_match(key
, unit
)) {
643 val
= emit_combine(p
, unit
,
644 key
->unit
[unit
].NumArgsRGB
,
645 key
->unit
[unit
].ModeRGB
,
646 key
->unit
[unit
].ArgsRGB
);
651 p
->emit(assign(temp_var
, val
));
653 else if (key
->unit
[unit
].ModeRGB
== TEXENV_MODE_DOT3_RGBA_EXT
||
654 key
->unit
[unit
].ModeRGB
== TEXENV_MODE_DOT3_RGBA
) {
655 ir_rvalue
*val
= emit_combine(p
, unit
,
656 key
->unit
[unit
].NumArgsRGB
,
657 key
->unit
[unit
].ModeRGB
,
658 key
->unit
[unit
].ArgsRGB
);
662 p
->emit(assign(temp_var
, val
));
665 /* Need to do something to stop from re-emitting identical
666 * argument calculations here:
668 val
= emit_combine(p
, unit
,
669 key
->unit
[unit
].NumArgsRGB
,
670 key
->unit
[unit
].ModeRGB
,
671 key
->unit
[unit
].ArgsRGB
);
672 val
= swizzle_xyz(smear(val
));
675 p
->emit(assign(temp_var
, val
, WRITEMASK_XYZ
));
677 val
= emit_combine(p
, unit
,
678 key
->unit
[unit
].NumArgsA
,
679 key
->unit
[unit
].ModeA
,
680 key
->unit
[unit
].ArgsA
);
681 val
= swizzle_w(smear(val
));
684 p
->emit(assign(temp_var
, val
, WRITEMASK_W
));
687 deref
= new(p
->mem_ctx
) ir_dereference_variable(temp_var
);
689 /* Deal with the final shift:
691 if (alpha_shift
|| rgb_shift
) {
694 if (rgb_shift
== alpha_shift
) {
695 shift
= new(p
->mem_ctx
) ir_constant((float)(1 << rgb_shift
));
698 ir_constant_data const_data
;
700 const_data
.f
[0] = float(1 << rgb_shift
);
701 const_data
.f
[1] = float(1 << rgb_shift
);
702 const_data
.f
[2] = float(1 << rgb_shift
);
703 const_data
.f
[3] = float(1 << alpha_shift
);
705 shift
= new(p
->mem_ctx
) ir_constant(glsl_type::vec4_type
,
709 return saturate(mul(deref
, shift
));
717 * Generate instruction for getting a texture source term.
719 static void load_texture( texenv_fragment_program
*p
, GLuint unit
)
721 ir_dereference
*deref
;
723 if (p
->src_texture
[unit
])
726 const GLuint texTarget
= p
->state
->unit
[unit
].source_index
;
729 if (!(p
->state
->inputs_available
& (VARYING_BIT_TEX0
<< unit
))) {
730 texcoord
= get_current_attrib(p
, VERT_ATTRIB_TEX0
+ unit
);
731 } else if (p
->texcoord_tex
[unit
]) {
732 texcoord
= new(p
->mem_ctx
) ir_dereference_variable(p
->texcoord_tex
[unit
]);
734 ir_variable
*tc_array
= p
->shader
->symbols
->get_variable("gl_TexCoord");
736 texcoord
= new(p
->mem_ctx
) ir_dereference_variable(tc_array
);
737 ir_rvalue
*index
= new(p
->mem_ctx
) ir_constant(unit
);
738 texcoord
= new(p
->mem_ctx
) ir_dereference_array(texcoord
, index
);
739 tc_array
->data
.max_array_access
= MAX2(tc_array
->data
.max_array_access
, (int)unit
);
742 if (!p
->state
->unit
[unit
].enabled
) {
743 p
->src_texture
[unit
] = p
->make_temp(glsl_type::vec4_type
,
745 p
->emit(p
->src_texture
[unit
]);
747 p
->emit(assign(p
->src_texture
[unit
], new(p
->mem_ctx
) ir_constant(0.0f
)));
751 const glsl_type
*sampler_type
= NULL
;
755 case TEXTURE_1D_INDEX
:
756 if (p
->state
->unit
[unit
].shadow
)
757 sampler_type
= glsl_type::sampler1DShadow_type
;
759 sampler_type
= glsl_type::sampler1D_type
;
762 case TEXTURE_1D_ARRAY_INDEX
:
763 if (p
->state
->unit
[unit
].shadow
)
764 sampler_type
= glsl_type::sampler1DArrayShadow_type
;
766 sampler_type
= glsl_type::sampler1DArray_type
;
769 case TEXTURE_2D_INDEX
:
770 if (p
->state
->unit
[unit
].shadow
)
771 sampler_type
= glsl_type::sampler2DShadow_type
;
773 sampler_type
= glsl_type::sampler2D_type
;
776 case TEXTURE_2D_ARRAY_INDEX
:
777 if (p
->state
->unit
[unit
].shadow
)
778 sampler_type
= glsl_type::sampler2DArrayShadow_type
;
780 sampler_type
= glsl_type::sampler2DArray_type
;
783 case TEXTURE_RECT_INDEX
:
784 if (p
->state
->unit
[unit
].shadow
)
785 sampler_type
= glsl_type::sampler2DRectShadow_type
;
787 sampler_type
= glsl_type::sampler2DRect_type
;
790 case TEXTURE_3D_INDEX
:
791 assert(!p
->state
->unit
[unit
].shadow
);
792 sampler_type
= glsl_type::sampler3D_type
;
795 case TEXTURE_CUBE_INDEX
:
796 if (p
->state
->unit
[unit
].shadow
)
797 sampler_type
= glsl_type::samplerCubeShadow_type
;
799 sampler_type
= glsl_type::samplerCube_type
;
802 case TEXTURE_EXTERNAL_INDEX
:
803 assert(!p
->state
->unit
[unit
].shadow
);
804 sampler_type
= glsl_type::samplerExternalOES_type
;
809 p
->src_texture
[unit
] = p
->make_temp(glsl_type::vec4_type
,
812 ir_texture
*tex
= new(p
->mem_ctx
) ir_texture(ir_tex
);
815 char *sampler_name
= ralloc_asprintf(p
->mem_ctx
, "sampler_%d", unit
);
816 ir_variable
*sampler
= new(p
->mem_ctx
) ir_variable(sampler_type
,
819 p
->top_instructions
->push_head(sampler
);
821 /* Set the texture unit for this sampler in the same way that
822 * layout(binding=X) would.
824 sampler
->data
.explicit_binding
= true;
825 sampler
->data
.binding
= unit
;
827 deref
= new(p
->mem_ctx
) ir_dereference_variable(sampler
);
828 tex
->set_sampler(deref
, glsl_type::vec4_type
);
830 tex
->coordinate
= new(p
->mem_ctx
) ir_swizzle(texcoord
, 0, 1, 2, 3, coords
);
832 if (p
->state
->unit
[unit
].shadow
) {
833 texcoord
= texcoord
->clone(p
->mem_ctx
, NULL
);
834 tex
->shadow_comparator
= new(p
->mem_ctx
) ir_swizzle(texcoord
,
840 texcoord
= texcoord
->clone(p
->mem_ctx
, NULL
);
841 tex
->projector
= swizzle_w(texcoord
);
843 p
->emit(assign(p
->src_texture
[unit
], tex
));
847 load_texenv_source(texenv_fragment_program
*p
,
848 GLuint src
, GLuint unit
)
851 case TEXENV_SRC_TEXTURE
:
852 load_texture(p
, unit
);
855 case TEXENV_SRC_TEXTURE0
:
856 case TEXENV_SRC_TEXTURE1
:
857 case TEXENV_SRC_TEXTURE2
:
858 case TEXENV_SRC_TEXTURE3
:
859 case TEXENV_SRC_TEXTURE4
:
860 case TEXENV_SRC_TEXTURE5
:
861 case TEXENV_SRC_TEXTURE6
:
862 case TEXENV_SRC_TEXTURE7
:
863 load_texture(p
, src
- TEXENV_SRC_TEXTURE0
);
867 /* not a texture src - do nothing */
874 * Generate instructions for loading all texture source terms.
877 load_texunit_sources( texenv_fragment_program
*p
, GLuint unit
)
879 const struct state_key
*key
= p
->state
;
882 for (i
= 0; i
< key
->unit
[unit
].NumArgsRGB
; i
++) {
883 load_texenv_source( p
, key
->unit
[unit
].ArgsRGB
[i
].Source
, unit
);
886 for (i
= 0; i
< key
->unit
[unit
].NumArgsA
; i
++) {
887 load_texenv_source( p
, key
->unit
[unit
].ArgsA
[i
].Source
, unit
);
894 * Applies the fog calculations.
896 * This is basically like the ARB_fragment_prorgam fog options. Note
897 * that ffvertex_prog.c produces fogcoord for us when
898 * GL_FOG_COORDINATE_EXT is set to GL_FRAGMENT_DEPTH_EXT.
901 emit_fog_instructions(texenv_fragment_program
*p
,
902 ir_rvalue
*fragcolor
)
904 struct state_key
*key
= p
->state
;
906 ir_variable
*params
, *oparams
;
907 ir_variable
*fogcoord
;
909 /* Temporary storage for the whole fog result. Fog calculations
910 * only affect rgb so we're hanging on to the .a value of fragcolor
913 ir_variable
*fog_result
= p
->make_temp(glsl_type::vec4_type
, "fog_result");
914 p
->emit(assign(fog_result
, fragcolor
));
916 fragcolor
= swizzle_xyz(fog_result
);
918 oparams
= p
->shader
->symbols
->get_variable("gl_FogParamsOptimizedMESA");
920 fogcoord
= p
->shader
->symbols
->get_variable("gl_FogFragCoord");
922 params
= p
->shader
->symbols
->get_variable("gl_Fog");
924 f
= new(p
->mem_ctx
) ir_dereference_variable(fogcoord
);
926 ir_variable
*f_var
= p
->make_temp(glsl_type::float_type
, "fog_factor");
928 switch (key
->fog_mode
) {
930 /* f = (end - z) / (end - start)
932 * gl_MesaFogParamsOptimized gives us (-1 / (end - start)) and
933 * (end / (end - start)) so we can generate a single MAD.
935 f
= add(mul(f
, swizzle_x(oparams
)), swizzle_y(oparams
));
938 /* f = e^(-(density * fogcoord))
940 * gl_MesaFogParamsOptimized gives us density/ln(2) so we can
941 * use EXP2 which is generally the native instruction without
942 * having to do any further math on the fog density uniform.
944 f
= mul(f
, swizzle_z(oparams
));
945 f
= new(p
->mem_ctx
) ir_expression(ir_unop_neg
, f
);
946 f
= new(p
->mem_ctx
) ir_expression(ir_unop_exp2
, f
);
949 /* f = e^(-(density * fogcoord)^2)
951 * gl_MesaFogParamsOptimized gives us density/sqrt(ln(2)) so we
952 * can do this like FOG_EXP but with a squaring after the
953 * multiply by density.
955 ir_variable
*temp_var
= p
->make_temp(glsl_type::float_type
, "fog_temp");
956 p
->emit(assign(temp_var
, mul(f
, swizzle_w(oparams
))));
958 f
= mul(temp_var
, temp_var
);
959 f
= new(p
->mem_ctx
) ir_expression(ir_unop_neg
, f
);
960 f
= new(p
->mem_ctx
) ir_expression(ir_unop_exp2
, f
);
964 p
->emit(assign(f_var
, saturate(f
)));
966 f
= sub(new(p
->mem_ctx
) ir_constant(1.0f
), f_var
);
967 temp
= new(p
->mem_ctx
) ir_dereference_variable(params
);
968 temp
= new(p
->mem_ctx
) ir_dereference_record(temp
, "color");
969 temp
= mul(swizzle_xyz(temp
), f
);
971 p
->emit(assign(fog_result
, add(temp
, mul(fragcolor
, f_var
)), WRITEMASK_XYZ
));
973 return new(p
->mem_ctx
) ir_dereference_variable(fog_result
);
977 emit_instructions(texenv_fragment_program
*p
)
979 struct state_key
*key
= p
->state
;
982 if (key
->nr_enabled_units
) {
983 /* First pass - to support texture_env_crossbar, first identify
984 * all referenced texture sources and emit texld instructions
987 for (unit
= 0; unit
< key
->nr_enabled_units
; unit
++)
988 if (key
->unit
[unit
].enabled
) {
989 load_texunit_sources(p
, unit
);
992 /* Second pass - emit combine instructions to build final color:
994 for (unit
= 0; unit
< key
->nr_enabled_units
; unit
++) {
995 if (key
->unit
[unit
].enabled
) {
996 p
->src_previous
= emit_texenv(p
, unit
);
1001 ir_rvalue
*cf
= get_source(p
, TEXENV_SRC_PREVIOUS
, 0);
1003 if (key
->separate_specular
) {
1004 ir_variable
*spec_result
= p
->make_temp(glsl_type::vec4_type
,
1006 p
->emit(assign(spec_result
, cf
));
1008 ir_rvalue
*secondary
;
1009 if (p
->state
->inputs_available
& VARYING_BIT_COL1
) {
1011 p
->shader
->symbols
->get_variable("gl_SecondaryColor");
1013 secondary
= swizzle_xyz(var
);
1015 secondary
= swizzle_xyz(get_current_attrib(p
, VERT_ATTRIB_COLOR1
));
1018 p
->emit(assign(spec_result
, add(swizzle_xyz(spec_result
), secondary
),
1021 cf
= new(p
->mem_ctx
) ir_dereference_variable(spec_result
);
1024 if (key
->fog_mode
) {
1025 cf
= emit_fog_instructions(p
, cf
);
1028 ir_variable
*frag_color
= p
->shader
->symbols
->get_variable("gl_FragColor");
1030 p
->emit(assign(frag_color
, cf
));
1034 * Generate a new fragment program which implements the context's
1035 * current texture env/combine mode.
1037 static struct gl_shader_program
*
1038 create_new_program(struct gl_context
*ctx
, struct state_key
*key
)
1040 texenv_fragment_program p
;
1042 _mesa_glsl_parse_state
*state
;
1044 p
.mem_ctx
= ralloc_context(NULL
);
1045 p
.shader
= _mesa_new_shader(0, MESA_SHADER_FRAGMENT
);
1047 p
.shader
->SourceChecksum
= 0xf18ed; /* fixed */
1049 p
.shader
->ir
= new(p
.shader
) exec_list
;
1050 state
= new(p
.shader
) _mesa_glsl_parse_state(ctx
, MESA_SHADER_FRAGMENT
,
1052 p
.shader
->symbols
= state
->symbols
;
1053 p
.top_instructions
= p
.shader
->ir
;
1054 p
.instructions
= p
.shader
->ir
;
1056 p
.shader_program
= _mesa_new_shader_program(0);
1058 /* Tell the linker to ignore the fact that we're building a
1059 * separate shader, in case we're in a GLES2 context that would
1060 * normally reject that. The real problem is that we're building a
1061 * fixed function program in a GLES2 context at all, but that's a
1062 * big mess to clean up.
1064 p
.shader_program
->SeparateShader
= GL_TRUE
;
1066 /* The legacy GLSL shadow functions follow the depth texture
1067 * mode and return vec4. The GLSL 1.30 shadow functions return float and
1068 * ignore the depth texture mode. That's a shader and state dependency
1069 * that's difficult to deal with. st/mesa uses a simple but not
1070 * completely correct solution: if the shader declares GLSL >= 1.30 and
1071 * the depth texture mode is GL_ALPHA (000X), it sets the XXXX swizzle
1072 * instead. Thus, the GLSL 1.30 shadow function will get the result in .x
1073 * and legacy shadow functions will get it in .w as expected.
1074 * For the fixed-function fragment shader, use 120 to get correct behavior
1077 state
->language_version
= 120;
1079 state
->es_shader
= false;
1080 if (_mesa_is_gles(ctx
) && ctx
->Extensions
.OES_EGL_image_external
)
1081 state
->OES_EGL_image_external_enable
= true;
1082 _mesa_glsl_initialize_types(state
);
1083 _mesa_glsl_initialize_variables(p
.instructions
, state
);
1085 for (unit
= 0; unit
< ctx
->Const
.MaxTextureUnits
; unit
++) {
1086 p
.src_texture
[unit
] = NULL
;
1087 p
.texcoord_tex
[unit
] = NULL
;
1090 p
.src_previous
= NULL
;
1092 ir_function
*main_f
= new(p
.mem_ctx
) ir_function("main");
1094 state
->symbols
->add_function(main_f
);
1096 ir_function_signature
*main_sig
=
1097 new(p
.mem_ctx
) ir_function_signature(glsl_type::void_type
);
1098 main_sig
->is_defined
= true;
1099 main_f
->add_signature(main_sig
);
1101 p
.instructions
= &main_sig
->body
;
1102 if (key
->num_draw_buffers
)
1103 emit_instructions(&p
);
1105 validate_ir_tree(p
.shader
->ir
);
1107 const struct gl_shader_compiler_options
*options
=
1108 &ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_FRAGMENT
];
1110 /* Conservative approach: Don't optimize here, the linker does it too. */
1111 if (!ctx
->Const
.GLSLOptimizeConservatively
) {
1112 while (do_common_optimization(p
.shader
->ir
, false, false, options
,
1113 ctx
->Const
.NativeIntegers
))
1117 reparent_ir(p
.shader
->ir
, p
.shader
->ir
);
1119 p
.shader
->CompileStatus
= compile_success
;
1120 p
.shader
->Version
= state
->language_version
;
1121 p
.shader_program
->Shaders
=
1122 (gl_shader
**)malloc(sizeof(*p
.shader_program
->Shaders
));
1123 p
.shader_program
->Shaders
[0] = p
.shader
;
1124 p
.shader_program
->NumShaders
= 1;
1126 _mesa_glsl_link_shader(ctx
, p
.shader_program
);
1128 if (!p
.shader_program
->data
->LinkStatus
)
1129 _mesa_problem(ctx
, "Failed to link fixed function fragment shader: %s\n",
1130 p
.shader_program
->data
->InfoLog
);
1132 ralloc_free(p
.mem_ctx
);
1133 return p
.shader_program
;
1139 * Return a fragment program which implements the current
1140 * fixed-function texture, fog and color-sum operations.
1142 struct gl_shader_program
*
1143 _mesa_get_fixed_func_fragment_program(struct gl_context
*ctx
)
1145 struct gl_shader_program
*shader_program
;
1146 struct state_key key
;
1149 keySize
= make_state_key(ctx
, &key
);
1151 shader_program
= (struct gl_shader_program
*)
1152 _mesa_search_program_cache(ctx
->FragmentProgram
.Cache
,
1155 if (!shader_program
) {
1156 shader_program
= create_new_program(ctx
, &key
);
1158 _mesa_shader_cache_insert(ctx
, ctx
->FragmentProgram
.Cache
,
1159 &key
, keySize
, shader_program
);
1162 return shader_program
;