2 * Mesa 3-D graphics library
4 * Copyright (C) 1999-2007 Brian Paul All Rights Reserved.
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the "Software"),
8 * to deal in the Software without restriction, including without limitation
9 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10 * and/or sell copies of the Software, and to permit persons to whom the
11 * Software is furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included
14 * in all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
17 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
20 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
21 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
22 * OTHER DEALINGS IN THE SOFTWARE.
27 * Vertex and fragment program support functions.
32 #include "main/glheader.h"
33 #include "main/context.h"
34 #include "main/hash.h"
35 #include "main/macros.h"
37 #include "prog_cache.h"
38 #include "prog_parameter.h"
39 #include "prog_instruction.h"
43 * A pointer to this dummy program is put into the hash table when
44 * glGenPrograms is called.
46 struct gl_program _mesa_DummyProgram
;
50 * Init context's vertex/fragment program state
53 _mesa_init_program(struct gl_context
*ctx
)
56 * If this assertion fails, we need to increase the field
57 * size for register indexes (see INST_INDEX_BITS).
59 ASSERT(ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxUniformComponents
/ 4
60 <= (1 << INST_INDEX_BITS
));
61 ASSERT(ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxUniformComponents
/ 4
62 <= (1 << INST_INDEX_BITS
));
64 ASSERT(ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTemps
<= (1 << INST_INDEX_BITS
));
65 ASSERT(ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxLocalParams
<= (1 << INST_INDEX_BITS
));
66 ASSERT(ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTemps
<= (1 << INST_INDEX_BITS
));
67 ASSERT(ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxLocalParams
<= (1 << INST_INDEX_BITS
));
69 ASSERT(ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxUniformComponents
<= 4 * MAX_UNIFORMS
);
70 ASSERT(ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxUniformComponents
<= 4 * MAX_UNIFORMS
);
72 ASSERT(ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAddressOffset
<= (1 << INST_INDEX_BITS
));
73 ASSERT(ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxAddressOffset
<= (1 << INST_INDEX_BITS
));
75 /* If this fails, increase prog_instruction::TexSrcUnit size */
76 STATIC_ASSERT(MAX_TEXTURE_UNITS
<= (1 << 5));
78 /* If this fails, increase prog_instruction::TexSrcTarget size */
79 STATIC_ASSERT(NUM_TEXTURE_TARGETS
<= (1 << 4));
81 ctx
->Program
.ErrorPos
= -1;
82 ctx
->Program
.ErrorString
= _mesa_strdup("");
84 ctx
->VertexProgram
.Enabled
= GL_FALSE
;
85 ctx
->VertexProgram
.PointSizeEnabled
=
86 (ctx
->API
== API_OPENGLES2
) ? GL_TRUE
: GL_FALSE
;
87 ctx
->VertexProgram
.TwoSideEnabled
= GL_FALSE
;
88 _mesa_reference_vertprog(ctx
, &ctx
->VertexProgram
.Current
,
89 ctx
->Shared
->DefaultVertexProgram
);
90 assert(ctx
->VertexProgram
.Current
);
91 ctx
->VertexProgram
.Cache
= _mesa_new_program_cache();
93 ctx
->FragmentProgram
.Enabled
= GL_FALSE
;
94 _mesa_reference_fragprog(ctx
, &ctx
->FragmentProgram
.Current
,
95 ctx
->Shared
->DefaultFragmentProgram
);
96 assert(ctx
->FragmentProgram
.Current
);
97 ctx
->FragmentProgram
.Cache
= _mesa_new_program_cache();
99 ctx
->GeometryProgram
.Enabled
= GL_FALSE
;
100 /* right now by default we don't have a geometry program */
101 _mesa_reference_geomprog(ctx
, &ctx
->GeometryProgram
.Current
,
103 ctx
->GeometryProgram
.Cache
= _mesa_new_program_cache();
105 /* XXX probably move this stuff */
106 ctx
->ATIFragmentShader
.Enabled
= GL_FALSE
;
107 ctx
->ATIFragmentShader
.Current
= ctx
->Shared
->DefaultFragmentShader
;
108 assert(ctx
->ATIFragmentShader
.Current
);
109 ctx
->ATIFragmentShader
.Current
->RefCount
++;
114 * Free a context's vertex/fragment program state
117 _mesa_free_program_data(struct gl_context
*ctx
)
119 _mesa_reference_vertprog(ctx
, &ctx
->VertexProgram
.Current
, NULL
);
120 _mesa_delete_program_cache(ctx
, ctx
->VertexProgram
.Cache
);
121 _mesa_reference_fragprog(ctx
, &ctx
->FragmentProgram
.Current
, NULL
);
122 _mesa_delete_shader_cache(ctx
, ctx
->FragmentProgram
.Cache
);
123 _mesa_reference_geomprog(ctx
, &ctx
->GeometryProgram
.Current
, NULL
);
124 _mesa_delete_program_cache(ctx
, ctx
->GeometryProgram
.Cache
);
126 /* XXX probably move this stuff */
127 if (ctx
->ATIFragmentShader
.Current
) {
128 ctx
->ATIFragmentShader
.Current
->RefCount
--;
129 if (ctx
->ATIFragmentShader
.Current
->RefCount
<= 0) {
130 free(ctx
->ATIFragmentShader
.Current
);
134 free((void *) ctx
->Program
.ErrorString
);
139 * Update the default program objects in the given context to reference those
140 * specified in the shared state and release those referencing the old
144 _mesa_update_default_objects_program(struct gl_context
*ctx
)
146 _mesa_reference_vertprog(ctx
, &ctx
->VertexProgram
.Current
,
147 ctx
->Shared
->DefaultVertexProgram
);
148 assert(ctx
->VertexProgram
.Current
);
150 _mesa_reference_fragprog(ctx
, &ctx
->FragmentProgram
.Current
,
151 ctx
->Shared
->DefaultFragmentProgram
);
152 assert(ctx
->FragmentProgram
.Current
);
154 _mesa_reference_geomprog(ctx
, &ctx
->GeometryProgram
.Current
,
155 ctx
->Shared
->DefaultGeometryProgram
);
157 /* XXX probably move this stuff */
158 if (ctx
->ATIFragmentShader
.Current
) {
159 ctx
->ATIFragmentShader
.Current
->RefCount
--;
160 if (ctx
->ATIFragmentShader
.Current
->RefCount
<= 0) {
161 free(ctx
->ATIFragmentShader
.Current
);
164 ctx
->ATIFragmentShader
.Current
= (struct ati_fragment_shader
*) ctx
->Shared
->DefaultFragmentShader
;
165 assert(ctx
->ATIFragmentShader
.Current
);
166 ctx
->ATIFragmentShader
.Current
->RefCount
++;
171 * Set the vertex/fragment program error state (position and error string).
172 * This is generally called from within the parsers.
175 _mesa_set_program_error(struct gl_context
*ctx
, GLint pos
, const char *string
)
177 ctx
->Program
.ErrorPos
= pos
;
178 free((void *) ctx
->Program
.ErrorString
);
181 ctx
->Program
.ErrorString
= _mesa_strdup(string
);
186 * Find the line number and column for 'pos' within 'string'.
187 * Return a copy of the line which contains 'pos'. Free the line with
189 * \param string the program string
190 * \param pos the position within the string
191 * \param line returns the line number corresponding to 'pos'.
192 * \param col returns the column number corresponding to 'pos'.
193 * \return copy of the line containing 'pos'.
196 _mesa_find_line_column(const GLubyte
*string
, const GLubyte
*pos
,
197 GLint
*line
, GLint
*col
)
199 const GLubyte
*lineStart
= string
;
200 const GLubyte
*p
= string
;
207 if (*p
== (GLubyte
) '\n') {
214 *col
= (pos
- lineStart
) + 1;
216 /* return copy of this line */
217 while (*p
!= 0 && *p
!= '\n')
221 memcpy(s
, lineStart
, len
);
229 * Initialize a new vertex/fragment program object.
231 static struct gl_program
*
232 _mesa_init_program_struct( struct gl_context
*ctx
, struct gl_program
*prog
,
233 GLenum target
, GLuint id
)
238 memset(prog
, 0, sizeof(*prog
));
240 prog
->Target
= target
;
242 prog
->Format
= GL_PROGRAM_FORMAT_ASCII_ARB
;
244 /* default mapping from samplers to texture units */
245 for (i
= 0; i
< MAX_SAMPLERS
; i
++)
246 prog
->SamplerUnits
[i
] = i
;
254 * Initialize a new fragment program object.
257 _mesa_init_fragment_program( struct gl_context
*ctx
, struct gl_fragment_program
*prog
,
258 GLenum target
, GLuint id
)
261 return _mesa_init_program_struct( ctx
, &prog
->Base
, target
, id
);
268 * Initialize a new vertex program object.
271 _mesa_init_vertex_program( struct gl_context
*ctx
, struct gl_vertex_program
*prog
,
272 GLenum target
, GLuint id
)
275 return _mesa_init_program_struct( ctx
, &prog
->Base
, target
, id
);
282 * Initialize a new compute program object.
285 _mesa_init_compute_program(struct gl_context
*ctx
,
286 struct gl_compute_program
*prog
, GLenum target
,
290 return _mesa_init_program_struct( ctx
, &prog
->Base
, target
, id
);
297 * Initialize a new geometry program object.
300 _mesa_init_geometry_program( struct gl_context
*ctx
, struct gl_geometry_program
*prog
,
301 GLenum target
, GLuint id
)
304 return _mesa_init_program_struct( ctx
, &prog
->Base
, target
, id
);
311 * Allocate and initialize a new fragment/vertex program object but
312 * don't put it into the program hash table. Called via
313 * ctx->Driver.NewProgram. May be overridden (ie. replaced) by a
314 * device driver function to implement OO deriviation with additional
315 * types not understood by this function.
318 * \param id program id/number
319 * \param target program target/type
320 * \return pointer to new program object
323 _mesa_new_program(struct gl_context
*ctx
, GLenum target
, GLuint id
)
325 struct gl_program
*prog
;
327 case GL_VERTEX_PROGRAM_ARB
: /* == GL_VERTEX_PROGRAM_NV */
328 prog
= _mesa_init_vertex_program(ctx
, CALLOC_STRUCT(gl_vertex_program
),
331 case GL_FRAGMENT_PROGRAM_NV
:
332 case GL_FRAGMENT_PROGRAM_ARB
:
333 prog
=_mesa_init_fragment_program(ctx
,
334 CALLOC_STRUCT(gl_fragment_program
),
337 case MESA_GEOMETRY_PROGRAM
:
338 prog
= _mesa_init_geometry_program(ctx
,
339 CALLOC_STRUCT(gl_geometry_program
),
342 case GL_COMPUTE_PROGRAM_NV
:
343 prog
= _mesa_init_compute_program(ctx
,
344 CALLOC_STRUCT(gl_compute_program
),
348 _mesa_problem(ctx
, "bad target in _mesa_new_program");
356 * Delete a program and remove it from the hash table, ignoring the
358 * Called via ctx->Driver.DeleteProgram. May be wrapped (OO deriviation)
359 * by a device driver function.
362 _mesa_delete_program(struct gl_context
*ctx
, struct gl_program
*prog
)
366 ASSERT(prog
->RefCount
==0);
368 if (prog
== &_mesa_DummyProgram
)
372 free(prog
->LocalParams
);
374 if (prog
->Instructions
) {
375 _mesa_free_instructions(prog
->Instructions
, prog
->NumInstructions
);
377 if (prog
->Parameters
) {
378 _mesa_free_parameter_list(prog
->Parameters
);
386 * Return the gl_program object for a given ID.
387 * Basically just a wrapper for _mesa_HashLookup() to avoid a lot of
391 _mesa_lookup_program(struct gl_context
*ctx
, GLuint id
)
394 return (struct gl_program
*) _mesa_HashLookup(ctx
->Shared
->Programs
, id
);
401 * Reference counting for vertex/fragment programs
402 * This is normally only called from the _mesa_reference_program() macro
403 * when there's a real pointer change.
406 _mesa_reference_program_(struct gl_context
*ctx
,
407 struct gl_program
**ptr
,
408 struct gl_program
*prog
)
414 if ((*ptr
)->Target
== GL_VERTEX_PROGRAM_ARB
)
415 ASSERT(prog
->Target
== GL_VERTEX_PROGRAM_ARB
);
416 else if ((*ptr
)->Target
== GL_FRAGMENT_PROGRAM_ARB
)
417 ASSERT(prog
->Target
== GL_FRAGMENT_PROGRAM_ARB
||
418 prog
->Target
== GL_FRAGMENT_PROGRAM_NV
);
419 else if ((*ptr
)->Target
== MESA_GEOMETRY_PROGRAM
)
420 ASSERT(prog
->Target
== MESA_GEOMETRY_PROGRAM
);
425 GLboolean deleteFlag
;
427 /*_glthread_LOCK_MUTEX((*ptr)->Mutex);*/
429 printf("Program %p ID=%u Target=%s Refcount-- to %d\n",
431 ((*ptr
)->Target
== GL_VERTEX_PROGRAM_ARB
? "VP" :
432 ((*ptr
)->Target
== MESA_GEOMETRY_PROGRAM
? "GP" : "FP")),
433 (*ptr
)->RefCount
- 1);
435 ASSERT((*ptr
)->RefCount
> 0);
438 deleteFlag
= ((*ptr
)->RefCount
== 0);
439 /*_glthread_UNLOCK_MUTEX((*ptr)->Mutex);*/
443 ctx
->Driver
.DeleteProgram(ctx
, *ptr
);
451 /*_glthread_LOCK_MUTEX(prog->Mutex);*/
454 printf("Program %p ID=%u Target=%s Refcount++ to %d\n",
456 (prog
->Target
== GL_VERTEX_PROGRAM_ARB
? "VP" :
457 (prog
->Target
== MESA_GEOMETRY_PROGRAM
? "GP" : "FP")),
460 /*_glthread_UNLOCK_MUTEX(prog->Mutex);*/
468 * Return a copy of a program.
469 * XXX Problem here if the program object is actually OO-derivation
470 * made by a device driver.
473 _mesa_clone_program(struct gl_context
*ctx
, const struct gl_program
*prog
)
475 struct gl_program
*clone
;
477 clone
= ctx
->Driver
.NewProgram(ctx
, prog
->Target
, prog
->Id
);
481 assert(clone
->Target
== prog
->Target
);
482 assert(clone
->RefCount
== 1);
484 clone
->String
= (GLubyte
*) _mesa_strdup((char *) prog
->String
);
485 clone
->Format
= prog
->Format
;
486 clone
->Instructions
= _mesa_alloc_instructions(prog
->NumInstructions
);
487 if (!clone
->Instructions
) {
488 _mesa_reference_program(ctx
, &clone
, NULL
);
491 _mesa_copy_instructions(clone
->Instructions
, prog
->Instructions
,
492 prog
->NumInstructions
);
493 clone
->InputsRead
= prog
->InputsRead
;
494 clone
->OutputsWritten
= prog
->OutputsWritten
;
495 clone
->SamplersUsed
= prog
->SamplersUsed
;
496 clone
->ShadowSamplers
= prog
->ShadowSamplers
;
497 memcpy(clone
->TexturesUsed
, prog
->TexturesUsed
, sizeof(prog
->TexturesUsed
));
499 if (prog
->Parameters
)
500 clone
->Parameters
= _mesa_clone_parameter_list(prog
->Parameters
);
501 if (prog
->LocalParams
) {
502 clone
->LocalParams
= malloc(MAX_PROGRAM_LOCAL_PARAMS
*
504 if (!clone
->LocalParams
) {
505 _mesa_reference_program(ctx
, &clone
, NULL
);
508 memcpy(clone
->LocalParams
, prog
->LocalParams
,
509 MAX_PROGRAM_LOCAL_PARAMS
* sizeof(float[4]));
511 clone
->IndirectRegisterFiles
= prog
->IndirectRegisterFiles
;
512 clone
->NumInstructions
= prog
->NumInstructions
;
513 clone
->NumTemporaries
= prog
->NumTemporaries
;
514 clone
->NumParameters
= prog
->NumParameters
;
515 clone
->NumAttributes
= prog
->NumAttributes
;
516 clone
->NumAddressRegs
= prog
->NumAddressRegs
;
517 clone
->NumNativeInstructions
= prog
->NumNativeInstructions
;
518 clone
->NumNativeTemporaries
= prog
->NumNativeTemporaries
;
519 clone
->NumNativeParameters
= prog
->NumNativeParameters
;
520 clone
->NumNativeAttributes
= prog
->NumNativeAttributes
;
521 clone
->NumNativeAddressRegs
= prog
->NumNativeAddressRegs
;
522 clone
->NumAluInstructions
= prog
->NumAluInstructions
;
523 clone
->NumTexInstructions
= prog
->NumTexInstructions
;
524 clone
->NumTexIndirections
= prog
->NumTexIndirections
;
525 clone
->NumNativeAluInstructions
= prog
->NumNativeAluInstructions
;
526 clone
->NumNativeTexInstructions
= prog
->NumNativeTexInstructions
;
527 clone
->NumNativeTexIndirections
= prog
->NumNativeTexIndirections
;
529 switch (prog
->Target
) {
530 case GL_VERTEX_PROGRAM_ARB
:
532 const struct gl_vertex_program
*vp
= gl_vertex_program_const(prog
);
533 struct gl_vertex_program
*vpc
= gl_vertex_program(clone
);
534 vpc
->IsPositionInvariant
= vp
->IsPositionInvariant
;
537 case GL_FRAGMENT_PROGRAM_ARB
:
539 const struct gl_fragment_program
*fp
= gl_fragment_program_const(prog
);
540 struct gl_fragment_program
*fpc
= gl_fragment_program(clone
);
541 fpc
->UsesKill
= fp
->UsesKill
;
542 fpc
->UsesDFdy
= fp
->UsesDFdy
;
543 fpc
->OriginUpperLeft
= fp
->OriginUpperLeft
;
544 fpc
->PixelCenterInteger
= fp
->PixelCenterInteger
;
547 case MESA_GEOMETRY_PROGRAM
:
549 const struct gl_geometry_program
*gp
= gl_geometry_program_const(prog
);
550 struct gl_geometry_program
*gpc
= gl_geometry_program(clone
);
551 gpc
->VerticesOut
= gp
->VerticesOut
;
552 gpc
->InputType
= gp
->InputType
;
553 gpc
->OutputType
= gp
->OutputType
;
557 _mesa_problem(NULL
, "Unexpected target in _mesa_clone_program");
565 * Insert 'count' NOP instructions at 'start' in the given program.
566 * Adjust branch targets accordingly.
569 _mesa_insert_instructions(struct gl_program
*prog
, GLuint start
, GLuint count
)
571 const GLuint origLen
= prog
->NumInstructions
;
572 const GLuint newLen
= origLen
+ count
;
573 struct prog_instruction
*newInst
;
576 /* adjust branches */
577 for (i
= 0; i
< prog
->NumInstructions
; i
++) {
578 struct prog_instruction
*inst
= prog
->Instructions
+ i
;
579 if (inst
->BranchTarget
> 0) {
580 if ((GLuint
)inst
->BranchTarget
>= start
) {
581 inst
->BranchTarget
+= count
;
586 /* Alloc storage for new instructions */
587 newInst
= _mesa_alloc_instructions(newLen
);
592 /* Copy 'start' instructions into new instruction buffer */
593 _mesa_copy_instructions(newInst
, prog
->Instructions
, start
);
595 /* init the new instructions */
596 _mesa_init_instructions(newInst
+ start
, count
);
598 /* Copy the remaining/tail instructions to new inst buffer */
599 _mesa_copy_instructions(newInst
+ start
+ count
,
600 prog
->Instructions
+ start
,
603 /* free old instructions */
604 _mesa_free_instructions(prog
->Instructions
, origLen
);
606 /* install new instructions */
607 prog
->Instructions
= newInst
;
608 prog
->NumInstructions
= newLen
;
614 * Delete 'count' instructions at 'start' in the given program.
615 * Adjust branch targets accordingly.
618 _mesa_delete_instructions(struct gl_program
*prog
, GLuint start
, GLuint count
)
620 const GLuint origLen
= prog
->NumInstructions
;
621 const GLuint newLen
= origLen
- count
;
622 struct prog_instruction
*newInst
;
625 /* adjust branches */
626 for (i
= 0; i
< prog
->NumInstructions
; i
++) {
627 struct prog_instruction
*inst
= prog
->Instructions
+ i
;
628 if (inst
->BranchTarget
> 0) {
629 if (inst
->BranchTarget
> (GLint
) start
) {
630 inst
->BranchTarget
-= count
;
635 /* Alloc storage for new instructions */
636 newInst
= _mesa_alloc_instructions(newLen
);
641 /* Copy 'start' instructions into new instruction buffer */
642 _mesa_copy_instructions(newInst
, prog
->Instructions
, start
);
644 /* Copy the remaining/tail instructions to new inst buffer */
645 _mesa_copy_instructions(newInst
+ start
,
646 prog
->Instructions
+ start
+ count
,
649 /* free old instructions */
650 _mesa_free_instructions(prog
->Instructions
, origLen
);
652 /* install new instructions */
653 prog
->Instructions
= newInst
;
654 prog
->NumInstructions
= newLen
;
661 * Search instructions for registers that match (oldFile, oldIndex),
662 * replacing them with (newFile, newIndex).
665 replace_registers(struct prog_instruction
*inst
, GLuint numInst
,
666 GLuint oldFile
, GLuint oldIndex
,
667 GLuint newFile
, GLuint newIndex
)
670 for (i
= 0; i
< numInst
; i
++) {
672 for (j
= 0; j
< _mesa_num_inst_src_regs(inst
[i
].Opcode
); j
++) {
673 if (inst
[i
].SrcReg
[j
].File
== oldFile
&&
674 inst
[i
].SrcReg
[j
].Index
== oldIndex
) {
675 inst
[i
].SrcReg
[j
].File
= newFile
;
676 inst
[i
].SrcReg
[j
].Index
= newIndex
;
680 if (inst
[i
].DstReg
.File
== oldFile
&& inst
[i
].DstReg
.Index
== oldIndex
) {
681 inst
[i
].DstReg
.File
= newFile
;
682 inst
[i
].DstReg
.Index
= newIndex
;
689 * Search instructions for references to program parameters. When found,
690 * increment the parameter index by 'offset'.
691 * Used when combining programs.
694 adjust_param_indexes(struct prog_instruction
*inst
, GLuint numInst
,
698 for (i
= 0; i
< numInst
; i
++) {
699 for (j
= 0; j
< _mesa_num_inst_src_regs(inst
[i
].Opcode
); j
++) {
700 GLuint f
= inst
[i
].SrcReg
[j
].File
;
701 if (f
== PROGRAM_CONSTANT
||
702 f
== PROGRAM_UNIFORM
||
703 f
== PROGRAM_STATE_VAR
) {
704 inst
[i
].SrcReg
[j
].Index
+= offset
;
712 * Combine two programs into one. Fix instructions so the outputs of
713 * the first program go to the inputs of the second program.
716 _mesa_combine_programs(struct gl_context
*ctx
,
717 const struct gl_program
*progA
,
718 const struct gl_program
*progB
)
720 struct prog_instruction
*newInst
;
721 struct gl_program
*newProg
;
722 const GLuint lenA
= progA
->NumInstructions
- 1; /* omit END instr */
723 const GLuint lenB
= progB
->NumInstructions
;
724 const GLuint numParamsA
= _mesa_num_parameters(progA
->Parameters
);
725 const GLuint newLength
= lenA
+ lenB
;
726 GLboolean usedTemps
[MAX_PROGRAM_TEMPS
];
727 GLuint firstTemp
= 0;
728 GLbitfield64 inputsB
;
731 ASSERT(progA
->Target
== progB
->Target
);
733 newInst
= _mesa_alloc_instructions(newLength
);
737 _mesa_copy_instructions(newInst
, progA
->Instructions
, lenA
);
738 _mesa_copy_instructions(newInst
+ lenA
, progB
->Instructions
, lenB
);
740 /* adjust branch / instruction addresses for B's instructions */
741 for (i
= 0; i
< lenB
; i
++) {
742 newInst
[lenA
+ i
].BranchTarget
+= lenA
;
745 newProg
= ctx
->Driver
.NewProgram(ctx
, progA
->Target
, 0);
746 newProg
->Instructions
= newInst
;
747 newProg
->NumInstructions
= newLength
;
749 /* find used temp regs (we may need new temps below) */
750 _mesa_find_used_registers(newProg
, PROGRAM_TEMPORARY
,
751 usedTemps
, MAX_PROGRAM_TEMPS
);
753 if (newProg
->Target
== GL_FRAGMENT_PROGRAM_ARB
) {
754 const struct gl_fragment_program
*fprogA
, *fprogB
;
755 struct gl_fragment_program
*newFprog
;
756 GLbitfield64 progB_inputsRead
= progB
->InputsRead
;
757 GLint progB_colorFile
, progB_colorIndex
;
759 fprogA
= gl_fragment_program_const(progA
);
760 fprogB
= gl_fragment_program_const(progB
);
761 newFprog
= gl_fragment_program(newProg
);
763 newFprog
->UsesKill
= fprogA
->UsesKill
|| fprogB
->UsesKill
;
764 newFprog
->UsesDFdy
= fprogA
->UsesDFdy
|| fprogB
->UsesDFdy
;
766 /* We'll do a search and replace for instances
767 * of progB_colorFile/progB_colorIndex below...
769 progB_colorFile
= PROGRAM_INPUT
;
770 progB_colorIndex
= VARYING_SLOT_COL0
;
773 * The fragment program may get color from a state var rather than
774 * a fragment input (vertex output) if it's constant.
775 * See the texenvprogram.c code.
776 * So, search the program's parameter list now to see if the program
777 * gets color from a state var instead of a conventional fragment
780 for (i
= 0; i
< progB
->Parameters
->NumParameters
; i
++) {
781 struct gl_program_parameter
*p
= &progB
->Parameters
->Parameters
[i
];
782 if (p
->Type
== PROGRAM_STATE_VAR
&&
783 p
->StateIndexes
[0] == STATE_INTERNAL
&&
784 p
->StateIndexes
[1] == STATE_CURRENT_ATTRIB
&&
785 (int) p
->StateIndexes
[2] == (int) VERT_ATTRIB_COLOR0
) {
786 progB_inputsRead
|= VARYING_BIT_COL0
;
787 progB_colorFile
= PROGRAM_STATE_VAR
;
788 progB_colorIndex
= i
;
793 /* Connect color outputs of fprogA to color inputs of fprogB, via a
794 * new temporary register.
796 if ((progA
->OutputsWritten
& BITFIELD64_BIT(FRAG_RESULT_COLOR
)) &&
797 (progB_inputsRead
& VARYING_BIT_COL0
)) {
798 GLint tempReg
= _mesa_find_free_register(usedTemps
, MAX_PROGRAM_TEMPS
,
801 _mesa_problem(ctx
, "No free temp regs found in "
802 "_mesa_combine_programs(), using 31");
805 firstTemp
= tempReg
+ 1;
807 /* replace writes to result.color[0] with tempReg */
808 replace_registers(newInst
, lenA
,
809 PROGRAM_OUTPUT
, FRAG_RESULT_COLOR
,
810 PROGRAM_TEMPORARY
, tempReg
);
811 /* replace reads from the input color with tempReg */
812 replace_registers(newInst
+ lenA
, lenB
,
813 progB_colorFile
, progB_colorIndex
, /* search for */
814 PROGRAM_TEMPORARY
, tempReg
/* replace with */ );
817 /* compute combined program's InputsRead */
818 inputsB
= progB_inputsRead
;
819 if (progA
->OutputsWritten
& BITFIELD64_BIT(FRAG_RESULT_COLOR
)) {
820 inputsB
&= ~(1 << VARYING_SLOT_COL0
);
822 newProg
->InputsRead
= progA
->InputsRead
| inputsB
;
823 newProg
->OutputsWritten
= progB
->OutputsWritten
;
824 newProg
->SamplersUsed
= progA
->SamplersUsed
| progB
->SamplersUsed
;
828 assert(0); /* XXX todo */
832 * Merge parameters (uniforms, constants, etc)
834 newProg
->Parameters
= _mesa_combine_parameter_lists(progA
->Parameters
,
837 adjust_param_indexes(newInst
+ lenA
, lenB
, numParamsA
);
845 * Populate the 'used' array with flags indicating which registers (TEMPs,
846 * INPUTs, OUTPUTs, etc, are used by the given program.
847 * \param file type of register to scan for
848 * \param used returns true/false flags for in use / free
849 * \param usedSize size of the 'used' array
852 _mesa_find_used_registers(const struct gl_program
*prog
,
853 gl_register_file file
,
854 GLboolean used
[], GLuint usedSize
)
858 memset(used
, 0, usedSize
);
860 for (i
= 0; i
< prog
->NumInstructions
; i
++) {
861 const struct prog_instruction
*inst
= prog
->Instructions
+ i
;
862 const GLuint n
= _mesa_num_inst_src_regs(inst
->Opcode
);
864 if (inst
->DstReg
.File
== file
) {
865 ASSERT(inst
->DstReg
.Index
< usedSize
);
866 if(inst
->DstReg
.Index
< usedSize
)
867 used
[inst
->DstReg
.Index
] = GL_TRUE
;
870 for (j
= 0; j
< n
; j
++) {
871 if (inst
->SrcReg
[j
].File
== file
) {
872 ASSERT(inst
->SrcReg
[j
].Index
< (GLint
) usedSize
);
873 if (inst
->SrcReg
[j
].Index
< (GLint
) usedSize
)
874 used
[inst
->SrcReg
[j
].Index
] = GL_TRUE
;
882 * Scan the given 'used' register flag array for the first entry
883 * that's >= firstReg.
884 * \param used vector of flags indicating registers in use (as returned
885 * by _mesa_find_used_registers())
886 * \param usedSize size of the 'used' array
887 * \param firstReg first register to start searching at
888 * \return index of unused register, or -1 if none.
891 _mesa_find_free_register(const GLboolean used
[],
892 GLuint usedSize
, GLuint firstReg
)
896 assert(firstReg
< usedSize
);
898 for (i
= firstReg
; i
< usedSize
; i
++)
908 * Check if the given register index is valid (doesn't exceed implementation-
910 * \return GL_TRUE if OK, GL_FALSE if bad index
913 _mesa_valid_register_index(const struct gl_context
*ctx
,
914 gl_shader_stage shaderType
,
915 gl_register_file file
, GLint index
)
917 const struct gl_program_constants
*c
;
919 assert(0 <= shaderType
&& shaderType
< MESA_SHADER_STAGES
);
920 c
= &ctx
->Const
.Program
[shaderType
];
923 case PROGRAM_UNDEFINED
:
924 return GL_TRUE
; /* XXX or maybe false? */
926 case PROGRAM_TEMPORARY
:
927 return index
>= 0 && index
< (GLint
) c
->MaxTemps
;
929 case PROGRAM_UNIFORM
:
930 case PROGRAM_STATE_VAR
:
931 /* aka constant buffer */
932 return index
>= 0 && index
< (GLint
) c
->MaxUniformComponents
/ 4;
934 case PROGRAM_CONSTANT
:
935 /* constant buffer w/ possible relative negative addressing */
936 return (index
> (int) c
->MaxUniformComponents
/ -4 &&
937 index
< (int) c
->MaxUniformComponents
/ 4);
943 switch (shaderType
) {
944 case MESA_SHADER_VERTEX
:
945 return index
< VERT_ATTRIB_GENERIC0
+ (GLint
) c
->MaxAttribs
;
946 case MESA_SHADER_FRAGMENT
:
947 return index
< VARYING_SLOT_VAR0
+ (GLint
) ctx
->Const
.MaxVarying
;
948 case MESA_SHADER_GEOMETRY
:
949 return index
< VARYING_SLOT_VAR0
+ (GLint
) ctx
->Const
.MaxVarying
;
958 switch (shaderType
) {
959 case MESA_SHADER_VERTEX
:
960 return index
< VARYING_SLOT_VAR0
+ (GLint
) ctx
->Const
.MaxVarying
;
961 case MESA_SHADER_FRAGMENT
:
962 return index
< FRAG_RESULT_DATA0
+ (GLint
) ctx
->Const
.MaxDrawBuffers
;
963 case MESA_SHADER_GEOMETRY
:
964 return index
< VARYING_SLOT_VAR0
+ (GLint
) ctx
->Const
.MaxVarying
;
969 case PROGRAM_ADDRESS
:
970 return index
>= 0 && index
< (GLint
) c
->MaxAddressRegs
;
974 "unexpected register file in _mesa_valid_register_index()");
982 * "Post-process" a GPU program. This is intended to be used for debugging.
983 * Example actions include no-op'ing instructions or changing instruction
987 _mesa_postprocess_program(struct gl_context
*ctx
, struct gl_program
*prog
)
989 static const GLfloat white
[4] = { 0.5, 0.5, 0.5, 0.5 };
992 GLint whiteIndex
= _mesa_add_unnamed_constant(prog
->Parameters
,
993 (gl_constant_value
*) white
,
998 for (i
= 0; i
< prog
->NumInstructions
; i
++) {
999 struct prog_instruction
*inst
= prog
->Instructions
+ i
;
1000 const GLuint n
= _mesa_num_inst_src_regs(inst
->Opcode
);
1004 if (_mesa_is_tex_instruction(inst
->Opcode
)) {
1006 /* replace TEX/TXP/TXB with MOV */
1007 inst
->Opcode
= OPCODE_MOV
;
1008 inst
->DstReg
.WriteMask
= WRITEMASK_XYZW
;
1009 inst
->SrcReg
[0].Swizzle
= SWIZZLE_XYZW
;
1010 inst
->SrcReg
[0].Negate
= NEGATE_NONE
;
1014 /* disable shadow texture mode */
1015 inst
->TexShadow
= 0;
1019 if (inst
->Opcode
== OPCODE_TXP
) {
1021 inst
->Opcode
= OPCODE_MOV
;
1022 inst
->DstReg
.WriteMask
= WRITEMASK_XYZW
;
1023 inst
->SrcReg
[0].File
= PROGRAM_CONSTANT
;
1024 inst
->SrcReg
[0].Index
= whiteIndex
;
1025 inst
->SrcReg
[0].Swizzle
= SWIZZLE_XYZW
;
1026 inst
->SrcReg
[0].Negate
= NEGATE_NONE
;
1029 inst
->TexShadow
= 0;
1032 inst
->Opcode
= OPCODE_TEX
;
1033 inst
->TexShadow
= 0;
1040 /* Gets the minimum number of shader invocations per fragment.
1041 * This function is useful to determine if we need to do per
1042 * sample shading or per fragment shading.
1045 _mesa_get_min_invocations_per_fragment(struct gl_context
*ctx
,
1046 const struct gl_fragment_program
*prog
,
1047 bool ignore_sample_qualifier
)
1049 /* From ARB_sample_shading specification:
1050 * "Using gl_SampleID in a fragment shader causes the entire shader
1051 * to be evaluated per-sample."
1053 * "Using gl_SamplePosition in a fragment shader causes the entire
1054 * shader to be evaluated per-sample."
1056 * "If MULTISAMPLE or SAMPLE_SHADING_ARB is disabled, sample shading
1059 if (ctx
->Multisample
.Enabled
) {
1060 /* The ARB_gpu_shader5 specification says:
1062 * "Use of the "sample" qualifier on a fragment shader input
1063 * forces per-sample shading"
1065 if (prog
->IsSample
&& !ignore_sample_qualifier
)
1066 return MAX2(ctx
->DrawBuffer
->Visual
.samples
, 1);
1068 if (prog
->Base
.SystemValuesRead
& (SYSTEM_BIT_SAMPLE_ID
|
1069 SYSTEM_BIT_SAMPLE_POS
))
1070 return MAX2(ctx
->DrawBuffer
->Visual
.samples
, 1);
1071 else if (ctx
->Multisample
.SampleShading
)
1072 return MAX2(ceil(ctx
->Multisample
.MinSampleShadingValue
*
1073 ctx
->DrawBuffer
->Visual
.samples
), 1);