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/framebuffer.h"
35 #include "main/hash.h"
36 #include "main/macros.h"
37 #include "main/shaderobj.h"
39 #include "prog_cache.h"
40 #include "prog_parameter.h"
41 #include "prog_instruction.h"
42 #include "util/bitscan.h"
43 #include "util/ralloc.h"
44 #include "util/u_atomic.h"
48 * A pointer to this dummy program is put into the hash table when
49 * glGenPrograms is called.
51 struct gl_program _mesa_DummyProgram
;
55 * Init context's vertex/fragment program state
58 _mesa_init_program(struct gl_context
*ctx
)
61 * If this assertion fails, we need to increase the field
62 * size for register indexes (see INST_INDEX_BITS).
64 assert(ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxUniformComponents
/ 4
65 <= (1 << INST_INDEX_BITS
));
66 assert(ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxUniformComponents
/ 4
67 <= (1 << INST_INDEX_BITS
));
69 assert(ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxTemps
<= (1 << INST_INDEX_BITS
));
70 assert(ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxLocalParams
<= (1 << INST_INDEX_BITS
));
71 assert(ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxTemps
<= (1 << INST_INDEX_BITS
));
72 assert(ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxLocalParams
<= (1 << INST_INDEX_BITS
));
74 assert(ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxUniformComponents
<= 4 * MAX_UNIFORMS
);
75 assert(ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxUniformComponents
<= 4 * MAX_UNIFORMS
);
77 assert(ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAddressOffset
<= (1 << INST_INDEX_BITS
));
78 assert(ctx
->Const
.Program
[MESA_SHADER_FRAGMENT
].MaxAddressOffset
<= (1 << INST_INDEX_BITS
));
80 /* If this fails, increase prog_instruction::TexSrcUnit size */
81 STATIC_ASSERT(MAX_TEXTURE_UNITS
<= (1 << 5));
83 /* If this fails, increase prog_instruction::TexSrcTarget size */
84 STATIC_ASSERT(NUM_TEXTURE_TARGETS
<= (1 << 4));
86 ctx
->Program
.ErrorPos
= -1;
87 ctx
->Program
.ErrorString
= strdup("");
89 ctx
->VertexProgram
.Enabled
= GL_FALSE
;
90 ctx
->VertexProgram
.PointSizeEnabled
=
91 (ctx
->API
== API_OPENGLES2
) ? GL_TRUE
: GL_FALSE
;
92 ctx
->VertexProgram
.TwoSideEnabled
= GL_FALSE
;
93 _mesa_reference_program(ctx
, &ctx
->VertexProgram
.Current
,
94 ctx
->Shared
->DefaultVertexProgram
);
95 assert(ctx
->VertexProgram
.Current
);
96 ctx
->VertexProgram
.Cache
= _mesa_new_program_cache();
98 ctx
->FragmentProgram
.Enabled
= GL_FALSE
;
99 _mesa_reference_program(ctx
, &ctx
->FragmentProgram
.Current
,
100 ctx
->Shared
->DefaultFragmentProgram
);
101 assert(ctx
->FragmentProgram
.Current
);
102 ctx
->FragmentProgram
.Cache
= _mesa_new_program_cache();
103 ctx
->VertexProgram
._VPMode
= VP_MODE_FF
;
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_program(ctx
, &ctx
->VertexProgram
.Current
, NULL
);
120 _mesa_delete_program_cache(ctx
, ctx
->VertexProgram
.Cache
);
121 _mesa_reference_program(ctx
, &ctx
->FragmentProgram
.Current
, NULL
);
122 _mesa_delete_shader_cache(ctx
, ctx
->FragmentProgram
.Cache
);
124 /* XXX probably move this stuff */
125 if (ctx
->ATIFragmentShader
.Current
) {
126 ctx
->ATIFragmentShader
.Current
->RefCount
--;
127 if (ctx
->ATIFragmentShader
.Current
->RefCount
<= 0) {
128 free(ctx
->ATIFragmentShader
.Current
);
132 free((void *) ctx
->Program
.ErrorString
);
137 * Update the default program objects in the given context to reference those
138 * specified in the shared state and release those referencing the old
142 _mesa_update_default_objects_program(struct gl_context
*ctx
)
144 _mesa_reference_program(ctx
, &ctx
->VertexProgram
.Current
,
145 ctx
->Shared
->DefaultVertexProgram
);
146 assert(ctx
->VertexProgram
.Current
);
148 _mesa_reference_program(ctx
, &ctx
->FragmentProgram
.Current
,
149 ctx
->Shared
->DefaultFragmentProgram
);
150 assert(ctx
->FragmentProgram
.Current
);
152 /* XXX probably move this stuff */
153 if (ctx
->ATIFragmentShader
.Current
) {
154 ctx
->ATIFragmentShader
.Current
->RefCount
--;
155 if (ctx
->ATIFragmentShader
.Current
->RefCount
<= 0) {
156 free(ctx
->ATIFragmentShader
.Current
);
159 ctx
->ATIFragmentShader
.Current
= (struct ati_fragment_shader
*) ctx
->Shared
->DefaultFragmentShader
;
160 assert(ctx
->ATIFragmentShader
.Current
);
161 ctx
->ATIFragmentShader
.Current
->RefCount
++;
166 * Set the vertex/fragment program error state (position and error string).
167 * This is generally called from within the parsers.
170 _mesa_set_program_error(struct gl_context
*ctx
, GLint pos
, const char *string
)
172 ctx
->Program
.ErrorPos
= pos
;
173 free((void *) ctx
->Program
.ErrorString
);
176 ctx
->Program
.ErrorString
= strdup(string
);
181 * Initialize a new gl_program object.
184 _mesa_init_gl_program(struct gl_program
*prog
, gl_shader_stage stage
,
185 GLuint id
, bool is_arb_asm
)
190 memset(prog
, 0, sizeof(*prog
));
192 prog
->Target
= _mesa_shader_stage_to_program(stage
);
194 prog
->Format
= GL_PROGRAM_FORMAT_ASCII_ARB
;
195 prog
->info
.stage
= stage
;
196 prog
->is_arb_asm
= is_arb_asm
;
198 /* Uniforms that lack an initializer in the shader code have an initial
199 * value of zero. This includes sampler uniforms.
201 * Page 24 (page 30 of the PDF) of the GLSL 1.20 spec says:
203 * "The link time initial value is either the value of the variable's
204 * initializer, if present, or 0 if no initializer is present. Sampler
205 * types cannot have initializers."
207 * So we only initialise ARB assembly style programs.
210 /* default mapping from samplers to texture units */
211 for (unsigned i
= 0; i
< MAX_SAMPLERS
; i
++)
212 prog
->SamplerUnits
[i
] = i
;
220 * Allocate and initialize a new fragment/vertex program object but
221 * don't put it into the program hash table. Called via
222 * ctx->Driver.NewProgram. May be overridden (ie. replaced) by a
223 * device driver function to implement OO deriviation with additional
224 * types not understood by this function.
227 * \param id program id/number
228 * \param stage shader stage
229 * \return pointer to new program object
232 _mesa_new_program(struct gl_context
*ctx
, gl_shader_stage stage
, GLuint id
,
235 struct gl_program
*prog
= rzalloc(NULL
, struct gl_program
);
237 return _mesa_init_gl_program(prog
, stage
, id
, is_arb_asm
);
242 * Delete a program and remove it from the hash table, ignoring the
244 * Called via ctx->Driver.DeleteProgram. May be wrapped (OO deriviation)
245 * by a device driver function.
248 _mesa_delete_program(struct gl_context
*ctx
, struct gl_program
*prog
)
252 assert(prog
->RefCount
==0);
254 if (prog
== &_mesa_DummyProgram
)
257 if (prog
->Parameters
) {
258 _mesa_free_parameter_list(prog
->Parameters
);
262 ralloc_free(prog
->nir
);
265 if (prog
->sh
.BindlessSamplers
) {
266 ralloc_free(prog
->sh
.BindlessSamplers
);
269 if (prog
->sh
.BindlessImages
) {
270 ralloc_free(prog
->sh
.BindlessImages
);
273 if (prog
->driver_cache_blob
) {
274 ralloc_free(prog
->driver_cache_blob
);
282 * Return the gl_program object for a given ID.
283 * Basically just a wrapper for _mesa_HashLookup() to avoid a lot of
287 _mesa_lookup_program(struct gl_context
*ctx
, GLuint id
)
290 return (struct gl_program
*) _mesa_HashLookup(ctx
->Shared
->Programs
, id
);
297 * Reference counting for vertex/fragment programs
298 * This is normally only called from the _mesa_reference_program() macro
299 * when there's a real pointer change.
302 _mesa_reference_program_(struct gl_context
*ctx
,
303 struct gl_program
**ptr
,
304 struct gl_program
*prog
)
310 if ((*ptr
)->Target
== GL_VERTEX_PROGRAM_ARB
)
311 assert(prog
->Target
== GL_VERTEX_PROGRAM_ARB
);
312 else if ((*ptr
)->Target
== GL_FRAGMENT_PROGRAM_ARB
)
313 assert(prog
->Target
== GL_FRAGMENT_PROGRAM_ARB
||
314 prog
->Target
== GL_FRAGMENT_PROGRAM_NV
);
315 else if ((*ptr
)->Target
== GL_GEOMETRY_PROGRAM_NV
)
316 assert(prog
->Target
== GL_GEOMETRY_PROGRAM_NV
);
321 struct gl_program
*oldProg
= *ptr
;
323 assert(oldProg
->RefCount
> 0);
325 if (p_atomic_dec_zero(&oldProg
->RefCount
)) {
327 _mesa_reference_shader_program_data(ctx
, &oldProg
->sh
.data
, NULL
);
328 ctx
->Driver
.DeleteProgram(ctx
, oldProg
);
336 p_atomic_inc(&prog
->RefCount
);
344 * Insert 'count' NOP instructions at 'start' in the given program.
345 * Adjust branch targets accordingly.
348 _mesa_insert_instructions(struct gl_program
*prog
, GLuint start
, GLuint count
)
350 const GLuint origLen
= prog
->arb
.NumInstructions
;
351 const GLuint newLen
= origLen
+ count
;
352 struct prog_instruction
*newInst
;
355 /* adjust branches */
356 for (i
= 0; i
< prog
->arb
.NumInstructions
; i
++) {
357 struct prog_instruction
*inst
= prog
->arb
.Instructions
+ i
;
358 if (inst
->BranchTarget
> 0) {
359 if ((GLuint
)inst
->BranchTarget
>= start
) {
360 inst
->BranchTarget
+= count
;
365 /* Alloc storage for new instructions */
366 newInst
= rzalloc_array(prog
, struct prog_instruction
, newLen
);
371 /* Copy 'start' instructions into new instruction buffer */
372 _mesa_copy_instructions(newInst
, prog
->arb
.Instructions
, start
);
374 /* init the new instructions */
375 _mesa_init_instructions(newInst
+ start
, count
);
377 /* Copy the remaining/tail instructions to new inst buffer */
378 _mesa_copy_instructions(newInst
+ start
+ count
,
379 prog
->arb
.Instructions
+ start
,
382 /* free old instructions */
383 ralloc_free(prog
->arb
.Instructions
);
385 /* install new instructions */
386 prog
->arb
.Instructions
= newInst
;
387 prog
->arb
.NumInstructions
= newLen
;
393 * Delete 'count' instructions at 'start' in the given program.
394 * Adjust branch targets accordingly.
397 _mesa_delete_instructions(struct gl_program
*prog
, GLuint start
, GLuint count
,
400 const GLuint origLen
= prog
->arb
.NumInstructions
;
401 const GLuint newLen
= origLen
- count
;
402 struct prog_instruction
*newInst
;
405 /* adjust branches */
406 for (i
= 0; i
< prog
->arb
.NumInstructions
; i
++) {
407 struct prog_instruction
*inst
= prog
->arb
.Instructions
+ i
;
408 if (inst
->BranchTarget
> 0) {
409 if (inst
->BranchTarget
> (GLint
) start
) {
410 inst
->BranchTarget
-= count
;
415 /* Alloc storage for new instructions */
416 newInst
= rzalloc_array(mem_ctx
, struct prog_instruction
, newLen
);
421 /* Copy 'start' instructions into new instruction buffer */
422 _mesa_copy_instructions(newInst
, prog
->arb
.Instructions
, start
);
424 /* Copy the remaining/tail instructions to new inst buffer */
425 _mesa_copy_instructions(newInst
+ start
,
426 prog
->arb
.Instructions
+ start
+ count
,
429 /* free old instructions */
430 ralloc_free(prog
->arb
.Instructions
);
432 /* install new instructions */
433 prog
->arb
.Instructions
= newInst
;
434 prog
->arb
.NumInstructions
= newLen
;
441 * Populate the 'used' array with flags indicating which registers (TEMPs,
442 * INPUTs, OUTPUTs, etc, are used by the given program.
443 * \param file type of register to scan for
444 * \param used returns true/false flags for in use / free
445 * \param usedSize size of the 'used' array
448 _mesa_find_used_registers(const struct gl_program
*prog
,
449 gl_register_file file
,
450 GLboolean used
[], GLuint usedSize
)
454 memset(used
, 0, usedSize
);
456 for (i
= 0; i
< prog
->arb
.NumInstructions
; i
++) {
457 const struct prog_instruction
*inst
= prog
->arb
.Instructions
+ i
;
458 const GLuint n
= _mesa_num_inst_src_regs(inst
->Opcode
);
460 if (inst
->DstReg
.File
== file
) {
461 assert(inst
->DstReg
.Index
< usedSize
);
462 if(inst
->DstReg
.Index
< usedSize
)
463 used
[inst
->DstReg
.Index
] = GL_TRUE
;
466 for (j
= 0; j
< n
; j
++) {
467 if (inst
->SrcReg
[j
].File
== file
) {
468 assert(inst
->SrcReg
[j
].Index
< (GLint
) usedSize
);
469 if (inst
->SrcReg
[j
].Index
< (GLint
) usedSize
)
470 used
[inst
->SrcReg
[j
].Index
] = GL_TRUE
;
478 * Scan the given 'used' register flag array for the first entry
479 * that's >= firstReg.
480 * \param used vector of flags indicating registers in use (as returned
481 * by _mesa_find_used_registers())
482 * \param usedSize size of the 'used' array
483 * \param firstReg first register to start searching at
484 * \return index of unused register, or -1 if none.
487 _mesa_find_free_register(const GLboolean used
[],
488 GLuint usedSize
, GLuint firstReg
)
492 assert(firstReg
< usedSize
);
494 for (i
= firstReg
; i
< usedSize
; i
++)
502 /* Gets the minimum number of shader invocations per fragment.
503 * This function is useful to determine if we need to do per
504 * sample shading or per fragment shading.
507 _mesa_get_min_invocations_per_fragment(struct gl_context
*ctx
,
508 const struct gl_program
*prog
)
510 /* From ARB_sample_shading specification:
511 * "Using gl_SampleID in a fragment shader causes the entire shader
512 * to be evaluated per-sample."
514 * "Using gl_SamplePosition in a fragment shader causes the entire
515 * shader to be evaluated per-sample."
517 * "If MULTISAMPLE or SAMPLE_SHADING_ARB is disabled, sample shading
520 if (ctx
->Multisample
.Enabled
) {
521 /* The ARB_gpu_shader5 specification says:
523 * "Use of the "sample" qualifier on a fragment shader input
524 * forces per-sample shading"
526 if (prog
->info
.fs
.uses_sample_qualifier
||
527 (prog
->info
.system_values_read
& (SYSTEM_BIT_SAMPLE_ID
|
528 SYSTEM_BIT_SAMPLE_POS
)))
529 return MAX2(_mesa_geometric_samples(ctx
->DrawBuffer
), 1);
530 else if (ctx
->Multisample
.SampleShading
)
531 return MAX2(ceilf(ctx
->Multisample
.MinSampleShadingValue
*
532 _mesa_geometric_samples(ctx
->DrawBuffer
)), 1);
541 gl_external_samplers(const struct gl_program
*prog
)
543 GLbitfield external_samplers
= 0;
544 GLbitfield mask
= prog
->SamplersUsed
;
547 int idx
= u_bit_scan(&mask
);
548 if (prog
->sh
.SamplerTargets
[idx
] == TEXTURE_EXTERNAL_INDEX
)
549 external_samplers
|= (1 << idx
);
552 return external_samplers
;