1 /**************************************************************************
3 * Copyright 2009-2010 VMware, Inc.
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL VMWARE, INC AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **************************************************************************/
29 #include "pipe/p_screen.h"
30 #include "pipe/p_context.h"
31 #include "pipe/p_state.h"
32 #include "tgsi/tgsi_ureg.h"
33 #include "tgsi/tgsi_build.h"
34 #include "tgsi/tgsi_info.h"
35 #include "tgsi/tgsi_dump.h"
36 #include "tgsi/tgsi_sanity.h"
37 #include "util/u_debug.h"
38 #include "util/u_inlines.h"
39 #include "util/u_memory.h"
40 #include "util/u_math.h"
41 #include "util/u_bitmask.h"
43 union tgsi_any_token
{
44 struct tgsi_header header
;
45 struct tgsi_processor processor
;
46 struct tgsi_token token
;
47 struct tgsi_property prop
;
48 struct tgsi_property_data prop_data
;
49 struct tgsi_declaration decl
;
50 struct tgsi_declaration_range decl_range
;
51 struct tgsi_declaration_dimension decl_dim
;
52 struct tgsi_declaration_interp decl_interp
;
53 struct tgsi_declaration_image decl_image
;
54 struct tgsi_declaration_semantic decl_semantic
;
55 struct tgsi_declaration_sampler_view decl_sampler_view
;
56 struct tgsi_declaration_array array
;
57 struct tgsi_immediate imm
;
58 union tgsi_immediate_data imm_data
;
59 struct tgsi_instruction insn
;
60 struct tgsi_instruction_label insn_label
;
61 struct tgsi_instruction_texture insn_texture
;
62 struct tgsi_instruction_memory insn_memory
;
63 struct tgsi_texture_offset insn_texture_offset
;
64 struct tgsi_src_register src
;
65 struct tgsi_ind_register ind
;
66 struct tgsi_dimension dim
;
67 struct tgsi_dst_register dst
;
73 union tgsi_any_token
*tokens
;
79 #define UREG_MAX_INPUT (4 * PIPE_MAX_SHADER_INPUTS)
80 #define UREG_MAX_SYSTEM_VALUE PIPE_MAX_ATTRIBS
81 #define UREG_MAX_OUTPUT (4 * PIPE_MAX_SHADER_OUTPUTS)
82 #define UREG_MAX_CONSTANT_RANGE 32
83 #define UREG_MAX_IMMEDIATE 4096
84 #define UREG_MAX_ADDR 3
85 #define UREG_MAX_ARRAY_TEMPS 256
91 } constant_range
[UREG_MAX_CONSTANT_RANGE
];
92 unsigned nr_constant_ranges
;
101 bool supports_any_inout_decl_range
;
102 int next_shader_processor
;
105 unsigned semantic_name
;
106 unsigned semantic_index
;
108 unsigned char cylindrical_wrap
;
109 unsigned char usage_mask
;
110 unsigned interp_location
;
114 } input
[UREG_MAX_INPUT
];
115 unsigned nr_inputs
, nr_input_regs
;
117 unsigned vs_inputs
[PIPE_MAX_ATTRIBS
/32];
120 unsigned semantic_name
;
121 unsigned semantic_index
;
122 } system_value
[UREG_MAX_SYSTEM_VALUE
];
123 unsigned nr_system_values
;
126 unsigned semantic_name
;
127 unsigned semantic_index
;
129 unsigned usage_mask
; /* = TGSI_WRITEMASK_* */
133 } output
[UREG_MAX_OUTPUT
];
134 unsigned nr_outputs
, nr_output_regs
;
144 } immediate
[UREG_MAX_IMMEDIATE
];
145 unsigned nr_immediates
;
147 struct ureg_src sampler
[PIPE_MAX_SAMPLERS
];
148 unsigned nr_samplers
;
153 unsigned return_type_x
;
154 unsigned return_type_y
;
155 unsigned return_type_z
;
156 unsigned return_type_w
;
157 } sampler_view
[PIPE_MAX_SHADER_SAMPLER_VIEWS
];
158 unsigned nr_sampler_views
;
166 } image
[PIPE_MAX_SHADER_IMAGES
];
172 } buffer
[PIPE_MAX_SHADER_BUFFERS
];
175 struct util_bitmask
*free_temps
;
176 struct util_bitmask
*local_temps
;
177 struct util_bitmask
*decl_temps
;
180 unsigned array_temps
[UREG_MAX_ARRAY_TEMPS
];
181 unsigned nr_array_temps
;
183 struct const_decl const_decls
;
184 struct const_decl const_decls2D
[PIPE_MAX_CONSTANT_BUFFERS
];
186 unsigned properties
[TGSI_PROPERTY_COUNT
];
189 unsigned nr_instructions
;
191 struct ureg_tokens domain
[2];
193 bool use_memory
[TGSI_MEMORY_TYPE_COUNT
];
196 static union tgsi_any_token error_tokens
[32];
198 static void tokens_error( struct ureg_tokens
*tokens
)
200 if (tokens
->tokens
&& tokens
->tokens
!= error_tokens
)
201 FREE(tokens
->tokens
);
203 tokens
->tokens
= error_tokens
;
204 tokens
->size
= ARRAY_SIZE(error_tokens
);
209 static void tokens_expand( struct ureg_tokens
*tokens
,
212 unsigned old_size
= tokens
->size
* sizeof(unsigned);
214 if (tokens
->tokens
== error_tokens
) {
218 while (tokens
->count
+ count
> tokens
->size
) {
219 tokens
->size
= (1 << ++tokens
->order
);
222 tokens
->tokens
= REALLOC(tokens
->tokens
,
224 tokens
->size
* sizeof(unsigned));
225 if (tokens
->tokens
== NULL
) {
226 tokens_error(tokens
);
230 static void set_bad( struct ureg_program
*ureg
)
232 tokens_error(&ureg
->domain
[0]);
237 static union tgsi_any_token
*get_tokens( struct ureg_program
*ureg
,
241 struct ureg_tokens
*tokens
= &ureg
->domain
[domain
];
242 union tgsi_any_token
*result
;
244 if (tokens
->count
+ count
> tokens
->size
)
245 tokens_expand(tokens
, count
);
247 result
= &tokens
->tokens
[tokens
->count
];
248 tokens
->count
+= count
;
253 static union tgsi_any_token
*retrieve_token( struct ureg_program
*ureg
,
257 if (ureg
->domain
[domain
].tokens
== error_tokens
)
258 return &error_tokens
[0];
260 return &ureg
->domain
[domain
].tokens
[nr
];
265 ureg_property(struct ureg_program
*ureg
, unsigned name
, unsigned value
)
267 assert(name
< ARRAY_SIZE(ureg
->properties
));
268 ureg
->properties
[name
] = value
;
272 ureg_DECL_fs_input_cyl_centroid_layout(struct ureg_program
*ureg
,
273 unsigned semantic_name
,
274 unsigned semantic_index
,
275 unsigned interp_mode
,
276 unsigned cylindrical_wrap
,
277 unsigned interp_location
,
285 assert(usage_mask
!= 0);
286 assert(usage_mask
<= TGSI_WRITEMASK_XYZW
);
288 for (i
= 0; i
< ureg
->nr_inputs
; i
++) {
289 if (ureg
->input
[i
].semantic_name
== semantic_name
&&
290 ureg
->input
[i
].semantic_index
== semantic_index
) {
291 assert(ureg
->input
[i
].interp
== interp_mode
);
292 assert(ureg
->input
[i
].cylindrical_wrap
== cylindrical_wrap
);
293 assert(ureg
->input
[i
].interp_location
== interp_location
);
294 if (ureg
->input
[i
].array_id
== array_id
) {
295 ureg
->input
[i
].usage_mask
|= usage_mask
;
298 assert((ureg
->input
[i
].usage_mask
& usage_mask
) == 0);
302 if (ureg
->nr_inputs
< UREG_MAX_INPUT
) {
303 assert(array_size
>= 1);
304 ureg
->input
[i
].semantic_name
= semantic_name
;
305 ureg
->input
[i
].semantic_index
= semantic_index
;
306 ureg
->input
[i
].interp
= interp_mode
;
307 ureg
->input
[i
].cylindrical_wrap
= cylindrical_wrap
;
308 ureg
->input
[i
].interp_location
= interp_location
;
309 ureg
->input
[i
].first
= index
;
310 ureg
->input
[i
].last
= index
+ array_size
- 1;
311 ureg
->input
[i
].array_id
= array_id
;
312 ureg
->input
[i
].usage_mask
= usage_mask
;
313 ureg
->nr_input_regs
= MAX2(ureg
->nr_input_regs
, index
+ array_size
);
320 return ureg_src_array_register(TGSI_FILE_INPUT
, ureg
->input
[i
].first
,
325 ureg_DECL_fs_input_cyl_centroid(struct ureg_program
*ureg
,
326 unsigned semantic_name
,
327 unsigned semantic_index
,
328 unsigned interp_mode
,
329 unsigned cylindrical_wrap
,
330 unsigned interp_location
,
334 return ureg_DECL_fs_input_cyl_centroid_layout(ureg
,
335 semantic_name
, semantic_index
, interp_mode
, cylindrical_wrap
, interp_location
,
336 ureg
->nr_input_regs
, TGSI_WRITEMASK_XYZW
, array_id
, array_size
);
341 ureg_DECL_vs_input( struct ureg_program
*ureg
,
344 assert(ureg
->processor
== PIPE_SHADER_VERTEX
);
345 assert(index
/ 32 < ARRAY_SIZE(ureg
->vs_inputs
));
347 ureg
->vs_inputs
[index
/32] |= 1 << (index
% 32);
348 return ureg_src_register( TGSI_FILE_INPUT
, index
);
353 ureg_DECL_input_layout(struct ureg_program
*ureg
,
354 unsigned semantic_name
,
355 unsigned semantic_index
,
361 return ureg_DECL_fs_input_cyl_centroid_layout(ureg
,
362 semantic_name
, semantic_index
, 0, 0, 0,
363 index
, usage_mask
, array_id
, array_size
);
368 ureg_DECL_input(struct ureg_program
*ureg
,
369 unsigned semantic_name
,
370 unsigned semantic_index
,
374 return ureg_DECL_fs_input_cyl_centroid(ureg
, semantic_name
, semantic_index
,
375 0, 0, 0, array_id
, array_size
);
380 ureg_DECL_system_value(struct ureg_program
*ureg
,
381 unsigned semantic_name
,
382 unsigned semantic_index
)
386 for (i
= 0; i
< ureg
->nr_system_values
; i
++) {
387 if (ureg
->system_value
[i
].semantic_name
== semantic_name
&&
388 ureg
->system_value
[i
].semantic_index
== semantic_index
) {
393 if (ureg
->nr_system_values
< UREG_MAX_SYSTEM_VALUE
) {
394 ureg
->system_value
[ureg
->nr_system_values
].semantic_name
= semantic_name
;
395 ureg
->system_value
[ureg
->nr_system_values
].semantic_index
= semantic_index
;
396 i
= ureg
->nr_system_values
;
397 ureg
->nr_system_values
++;
403 return ureg_src_register(TGSI_FILE_SYSTEM_VALUE
, i
);
408 ureg_DECL_output_layout(struct ureg_program
*ureg
,
409 unsigned semantic_name
,
410 unsigned semantic_index
,
419 assert(usage_mask
!= 0);
420 assert(!(streams
& 0x03) || (usage_mask
& 1));
421 assert(!(streams
& 0x0c) || (usage_mask
& 2));
422 assert(!(streams
& 0x30) || (usage_mask
& 4));
423 assert(!(streams
& 0xc0) || (usage_mask
& 8));
425 for (i
= 0; i
< ureg
->nr_outputs
; i
++) {
426 if (ureg
->output
[i
].semantic_name
== semantic_name
&&
427 ureg
->output
[i
].semantic_index
== semantic_index
) {
428 if (ureg
->output
[i
].array_id
== array_id
) {
429 ureg
->output
[i
].usage_mask
|= usage_mask
;
432 assert((ureg
->output
[i
].usage_mask
& usage_mask
) == 0);
436 if (ureg
->nr_outputs
< UREG_MAX_OUTPUT
) {
437 ureg
->output
[i
].semantic_name
= semantic_name
;
438 ureg
->output
[i
].semantic_index
= semantic_index
;
439 ureg
->output
[i
].usage_mask
= usage_mask
;
440 ureg
->output
[i
].first
= index
;
441 ureg
->output
[i
].last
= index
+ array_size
- 1;
442 ureg
->output
[i
].array_id
= array_id
;
443 ureg
->nr_output_regs
= MAX2(ureg
->nr_output_regs
, index
+ array_size
);
452 ureg
->output
[i
].streams
|= streams
;
454 return ureg_dst_array_register(TGSI_FILE_OUTPUT
, ureg
->output
[i
].first
,
460 ureg_DECL_output_masked(struct ureg_program
*ureg
,
467 return ureg_DECL_output_layout(ureg
, name
, index
, 0,
468 ureg
->nr_output_regs
, usage_mask
, array_id
, array_size
);
473 ureg_DECL_output(struct ureg_program
*ureg
,
477 return ureg_DECL_output_masked(ureg
, name
, index
, TGSI_WRITEMASK_XYZW
,
482 ureg_DECL_output_array(struct ureg_program
*ureg
,
483 unsigned semantic_name
,
484 unsigned semantic_index
,
488 return ureg_DECL_output_masked(ureg
, semantic_name
, semantic_index
,
490 array_id
, array_size
);
494 /* Returns a new constant register. Keep track of which have been
495 * referred to so that we can emit decls later.
497 * Constant operands declared with this function must be addressed
498 * with a two-dimensional index.
500 * There is nothing in this code to bind this constant to any tracked
501 * value or manage any constant_buffer contents -- that's the
502 * resposibility of the calling code.
505 ureg_DECL_constant2D(struct ureg_program
*ureg
,
510 struct const_decl
*decl
= &ureg
->const_decls2D
[index2D
];
512 assert(index2D
< PIPE_MAX_CONSTANT_BUFFERS
);
514 if (decl
->nr_constant_ranges
< UREG_MAX_CONSTANT_RANGE
) {
515 uint i
= decl
->nr_constant_ranges
++;
517 decl
->constant_range
[i
].first
= first
;
518 decl
->constant_range
[i
].last
= last
;
523 /* A one-dimensional, deprecated version of ureg_DECL_constant2D().
525 * Constant operands declared with this function must be addressed
526 * with a one-dimensional index.
529 ureg_DECL_constant(struct ureg_program
*ureg
,
532 struct const_decl
*decl
= &ureg
->const_decls
;
533 unsigned minconst
= index
, maxconst
= index
;
536 /* Inside existing range?
538 for (i
= 0; i
< decl
->nr_constant_ranges
; i
++) {
539 if (decl
->constant_range
[i
].first
<= index
&&
540 decl
->constant_range
[i
].last
>= index
) {
545 /* Extend existing range?
547 for (i
= 0; i
< decl
->nr_constant_ranges
; i
++) {
548 if (decl
->constant_range
[i
].last
== index
- 1) {
549 decl
->constant_range
[i
].last
= index
;
553 if (decl
->constant_range
[i
].first
== index
+ 1) {
554 decl
->constant_range
[i
].first
= index
;
558 minconst
= MIN2(minconst
, decl
->constant_range
[i
].first
);
559 maxconst
= MAX2(maxconst
, decl
->constant_range
[i
].last
);
564 if (decl
->nr_constant_ranges
< UREG_MAX_CONSTANT_RANGE
) {
565 i
= decl
->nr_constant_ranges
++;
566 decl
->constant_range
[i
].first
= index
;
567 decl
->constant_range
[i
].last
= index
;
571 /* Collapse all ranges down to one:
574 decl
->constant_range
[0].first
= minconst
;
575 decl
->constant_range
[0].last
= maxconst
;
576 decl
->nr_constant_ranges
= 1;
579 assert(i
< decl
->nr_constant_ranges
);
580 assert(decl
->constant_range
[i
].first
<= index
);
581 assert(decl
->constant_range
[i
].last
>= index
);
582 return ureg_src_register(TGSI_FILE_CONSTANT
, index
);
585 static struct ureg_dst
alloc_temporary( struct ureg_program
*ureg
,
590 /* Look for a released temporary.
592 for (i
= util_bitmask_get_first_index(ureg
->free_temps
);
593 i
!= UTIL_BITMASK_INVALID_INDEX
;
594 i
= util_bitmask_get_next_index(ureg
->free_temps
, i
+ 1)) {
595 if (util_bitmask_get(ureg
->local_temps
, i
) == local
)
599 /* Or allocate a new one.
601 if (i
== UTIL_BITMASK_INVALID_INDEX
) {
602 i
= ureg
->nr_temps
++;
605 util_bitmask_set(ureg
->local_temps
, i
);
607 /* Start a new declaration when the local flag changes */
608 if (!i
|| util_bitmask_get(ureg
->local_temps
, i
- 1) != local
)
609 util_bitmask_set(ureg
->decl_temps
, i
);
612 util_bitmask_clear(ureg
->free_temps
, i
);
614 return ureg_dst_register( TGSI_FILE_TEMPORARY
, i
);
617 struct ureg_dst
ureg_DECL_temporary( struct ureg_program
*ureg
)
619 return alloc_temporary(ureg
, FALSE
);
622 struct ureg_dst
ureg_DECL_local_temporary( struct ureg_program
*ureg
)
624 return alloc_temporary(ureg
, TRUE
);
627 struct ureg_dst
ureg_DECL_array_temporary( struct ureg_program
*ureg
,
631 unsigned i
= ureg
->nr_temps
;
632 struct ureg_dst dst
= ureg_dst_register( TGSI_FILE_TEMPORARY
, i
);
635 util_bitmask_set(ureg
->local_temps
, i
);
637 /* Always start a new declaration at the start */
638 util_bitmask_set(ureg
->decl_temps
, i
);
640 ureg
->nr_temps
+= size
;
642 /* and also at the end of the array */
643 util_bitmask_set(ureg
->decl_temps
, ureg
->nr_temps
);
645 if (ureg
->nr_array_temps
< UREG_MAX_ARRAY_TEMPS
) {
646 ureg
->array_temps
[ureg
->nr_array_temps
++] = i
;
647 dst
.ArrayID
= ureg
->nr_array_temps
;
653 void ureg_release_temporary( struct ureg_program
*ureg
,
654 struct ureg_dst tmp
)
656 if(tmp
.File
== TGSI_FILE_TEMPORARY
)
657 util_bitmask_set(ureg
->free_temps
, tmp
.Index
);
661 /* Allocate a new address register.
663 struct ureg_dst
ureg_DECL_address( struct ureg_program
*ureg
)
665 if (ureg
->nr_addrs
< UREG_MAX_ADDR
)
666 return ureg_dst_register( TGSI_FILE_ADDRESS
, ureg
->nr_addrs
++ );
669 return ureg_dst_register( TGSI_FILE_ADDRESS
, 0 );
672 /* Allocate a new sampler.
674 struct ureg_src
ureg_DECL_sampler( struct ureg_program
*ureg
,
679 for (i
= 0; i
< ureg
->nr_samplers
; i
++)
680 if (ureg
->sampler
[i
].Index
== nr
)
681 return ureg
->sampler
[i
];
683 if (i
< PIPE_MAX_SAMPLERS
) {
684 ureg
->sampler
[i
] = ureg_src_register( TGSI_FILE_SAMPLER
, nr
);
686 return ureg
->sampler
[i
];
690 return ureg
->sampler
[0];
694 * Allocate a new shader sampler view.
697 ureg_DECL_sampler_view(struct ureg_program
*ureg
,
700 unsigned return_type_x
,
701 unsigned return_type_y
,
702 unsigned return_type_z
,
703 unsigned return_type_w
)
705 struct ureg_src reg
= ureg_src_register(TGSI_FILE_SAMPLER_VIEW
, index
);
708 for (i
= 0; i
< ureg
->nr_sampler_views
; i
++) {
709 if (ureg
->sampler_view
[i
].index
== index
) {
714 if (i
< PIPE_MAX_SHADER_SAMPLER_VIEWS
) {
715 ureg
->sampler_view
[i
].index
= index
;
716 ureg
->sampler_view
[i
].target
= target
;
717 ureg
->sampler_view
[i
].return_type_x
= return_type_x
;
718 ureg
->sampler_view
[i
].return_type_y
= return_type_y
;
719 ureg
->sampler_view
[i
].return_type_z
= return_type_z
;
720 ureg
->sampler_view
[i
].return_type_w
= return_type_w
;
721 ureg
->nr_sampler_views
++;
729 /* Allocate a new image.
732 ureg_DECL_image(struct ureg_program
*ureg
,
739 struct ureg_src reg
= ureg_src_register(TGSI_FILE_IMAGE
, index
);
742 for (i
= 0; i
< ureg
->nr_images
; i
++)
743 if (ureg
->image
[i
].index
== index
)
746 if (i
< PIPE_MAX_SHADER_IMAGES
) {
747 ureg
->image
[i
].index
= index
;
748 ureg
->image
[i
].target
= target
;
749 ureg
->image
[i
].wr
= wr
;
750 ureg
->image
[i
].raw
= raw
;
751 ureg
->image
[i
].format
= format
;
760 /* Allocate a new buffer.
762 struct ureg_src
ureg_DECL_buffer(struct ureg_program
*ureg
, unsigned nr
,
765 struct ureg_src reg
= ureg_src_register(TGSI_FILE_BUFFER
, nr
);
768 for (i
= 0; i
< ureg
->nr_buffers
; i
++)
769 if (ureg
->buffer
[i
].index
== nr
)
772 if (i
< PIPE_MAX_SHADER_BUFFERS
) {
773 ureg
->buffer
[i
].index
= nr
;
774 ureg
->buffer
[i
].atomic
= atomic
;
783 /* Allocate a memory area.
785 struct ureg_src
ureg_DECL_memory(struct ureg_program
*ureg
,
786 unsigned memory_type
)
788 struct ureg_src reg
= ureg_src_register(TGSI_FILE_MEMORY
, memory_type
);
790 ureg
->use_memory
[memory_type
] = true;
795 match_or_expand_immediate64( const unsigned *v
,
802 unsigned nr2
= *pnr2
;
806 for (i
= 0; i
< nr
; i
+= 2) {
807 boolean found
= FALSE
;
809 for (j
= 0; j
< nr2
&& !found
; j
+= 2) {
810 if (v
[i
] == v2
[j
] && v
[i
+ 1] == v2
[j
+ 1]) {
811 *swizzle
|= (j
<< (i
* 2)) | ((j
+ 1) << ((i
+ 1) * 2));
821 v2
[nr2
+ 1] = v
[i
+ 1];
823 *swizzle
|= (nr2
<< (i
* 2)) | ((nr2
+ 1) << ((i
+ 1) * 2));
828 /* Actually expand immediate only when fully succeeded.
835 match_or_expand_immediate( const unsigned *v
,
842 unsigned nr2
= *pnr2
;
845 if (type
== TGSI_IMM_FLOAT64
||
846 type
== TGSI_IMM_UINT64
||
847 type
== TGSI_IMM_INT64
)
848 return match_or_expand_immediate64(v
, type
, nr
, v2
, pnr2
, swizzle
);
852 for (i
= 0; i
< nr
; i
++) {
853 boolean found
= FALSE
;
855 for (j
= 0; j
< nr2
&& !found
; j
++) {
857 *swizzle
|= j
<< (i
* 2);
868 *swizzle
|= nr2
<< (i
* 2);
873 /* Actually expand immediate only when fully succeeded.
880 static struct ureg_src
881 decl_immediate( struct ureg_program
*ureg
,
887 unsigned swizzle
= 0;
889 /* Could do a first pass where we examine all existing immediates
893 for (i
= 0; i
< ureg
->nr_immediates
; i
++) {
894 if (ureg
->immediate
[i
].type
!= type
) {
897 if (match_or_expand_immediate(v
,
900 ureg
->immediate
[i
].value
.u
,
901 &ureg
->immediate
[i
].nr
,
907 if (ureg
->nr_immediates
< UREG_MAX_IMMEDIATE
) {
908 i
= ureg
->nr_immediates
++;
909 ureg
->immediate
[i
].type
= type
;
910 if (match_or_expand_immediate(v
,
913 ureg
->immediate
[i
].value
.u
,
914 &ureg
->immediate
[i
].nr
,
923 /* Make sure that all referenced elements are from this immediate.
924 * Has the effect of making size-one immediates into scalars.
926 if (type
== TGSI_IMM_FLOAT64
||
927 type
== TGSI_IMM_UINT64
||
928 type
== TGSI_IMM_INT64
) {
929 for (j
= nr
; j
< 4; j
+=2) {
930 swizzle
|= (swizzle
& 0xf) << (j
* 2);
933 for (j
= nr
; j
< 4; j
++) {
934 swizzle
|= (swizzle
& 0x3) << (j
* 2);
937 return ureg_swizzle(ureg_src_register(TGSI_FILE_IMMEDIATE
, i
),
938 (swizzle
>> 0) & 0x3,
939 (swizzle
>> 2) & 0x3,
940 (swizzle
>> 4) & 0x3,
941 (swizzle
>> 6) & 0x3);
946 ureg_DECL_immediate( struct ureg_program
*ureg
,
956 for (i
= 0; i
< nr
; i
++) {
960 return decl_immediate(ureg
, fu
.u
, nr
, TGSI_IMM_FLOAT32
);
964 ureg_DECL_immediate_f64( struct ureg_program
*ureg
,
974 assert((nr
/ 2) < 3);
975 for (i
= 0; i
< nr
/ 2; i
++) {
979 return decl_immediate(ureg
, fu
.u
, nr
, TGSI_IMM_FLOAT64
);
983 ureg_DECL_immediate_uint( struct ureg_program
*ureg
,
987 return decl_immediate(ureg
, v
, nr
, TGSI_IMM_UINT32
);
992 ureg_DECL_immediate_block_uint( struct ureg_program
*ureg
,
999 if (ureg
->nr_immediates
+ (nr
+ 3) / 4 > UREG_MAX_IMMEDIATE
) {
1001 return ureg_src_register(TGSI_FILE_IMMEDIATE
, 0);
1004 index
= ureg
->nr_immediates
;
1005 ureg
->nr_immediates
+= (nr
+ 3) / 4;
1007 for (i
= index
; i
< ureg
->nr_immediates
; i
++) {
1008 ureg
->immediate
[i
].type
= TGSI_IMM_UINT32
;
1009 ureg
->immediate
[i
].nr
= nr
> 4 ? 4 : nr
;
1010 memcpy(ureg
->immediate
[i
].value
.u
,
1011 &v
[(i
- index
) * 4],
1012 ureg
->immediate
[i
].nr
* sizeof(uint
));
1016 return ureg_src_register(TGSI_FILE_IMMEDIATE
, index
);
1021 ureg_DECL_immediate_int( struct ureg_program
*ureg
,
1025 return decl_immediate(ureg
, (const unsigned *)v
, nr
, TGSI_IMM_INT32
);
1029 ureg_DECL_immediate_uint64( struct ureg_program
*ureg
,
1039 assert((nr
/ 2) < 3);
1040 for (i
= 0; i
< nr
/ 2; i
++) {
1044 return decl_immediate(ureg
, fu
.u
, nr
, TGSI_IMM_UINT64
);
1048 ureg_DECL_immediate_int64( struct ureg_program
*ureg
,
1058 assert((nr
/ 2) < 3);
1059 for (i
= 0; i
< nr
/ 2; i
++) {
1063 return decl_immediate(ureg
, fu
.u
, nr
, TGSI_IMM_INT64
);
1067 ureg_emit_src( struct ureg_program
*ureg
,
1068 struct ureg_src src
)
1070 unsigned size
= 1 + (src
.Indirect
? 1 : 0) +
1071 (src
.Dimension
? (src
.DimIndirect
? 2 : 1) : 0);
1073 union tgsi_any_token
*out
= get_tokens( ureg
, DOMAIN_INSN
, size
);
1076 assert(src
.File
!= TGSI_FILE_NULL
);
1077 assert(src
.File
< TGSI_FILE_COUNT
);
1080 out
[n
].src
.File
= src
.File
;
1081 out
[n
].src
.SwizzleX
= src
.SwizzleX
;
1082 out
[n
].src
.SwizzleY
= src
.SwizzleY
;
1083 out
[n
].src
.SwizzleZ
= src
.SwizzleZ
;
1084 out
[n
].src
.SwizzleW
= src
.SwizzleW
;
1085 out
[n
].src
.Index
= src
.Index
;
1086 out
[n
].src
.Negate
= src
.Negate
;
1087 out
[0].src
.Absolute
= src
.Absolute
;
1091 out
[0].src
.Indirect
= 1;
1093 out
[n
].ind
.File
= src
.IndirectFile
;
1094 out
[n
].ind
.Swizzle
= src
.IndirectSwizzle
;
1095 out
[n
].ind
.Index
= src
.IndirectIndex
;
1096 if (!ureg
->supports_any_inout_decl_range
&&
1097 (src
.File
== TGSI_FILE_INPUT
|| src
.File
== TGSI_FILE_OUTPUT
))
1098 out
[n
].ind
.ArrayID
= 0;
1100 out
[n
].ind
.ArrayID
= src
.ArrayID
;
1104 if (src
.Dimension
) {
1105 out
[0].src
.Dimension
= 1;
1106 out
[n
].dim
.Dimension
= 0;
1107 out
[n
].dim
.Padding
= 0;
1108 if (src
.DimIndirect
) {
1109 out
[n
].dim
.Indirect
= 1;
1110 out
[n
].dim
.Index
= src
.DimensionIndex
;
1113 out
[n
].ind
.File
= src
.DimIndFile
;
1114 out
[n
].ind
.Swizzle
= src
.DimIndSwizzle
;
1115 out
[n
].ind
.Index
= src
.DimIndIndex
;
1116 if (!ureg
->supports_any_inout_decl_range
&&
1117 (src
.File
== TGSI_FILE_INPUT
|| src
.File
== TGSI_FILE_OUTPUT
))
1118 out
[n
].ind
.ArrayID
= 0;
1120 out
[n
].ind
.ArrayID
= src
.ArrayID
;
1122 out
[n
].dim
.Indirect
= 0;
1123 out
[n
].dim
.Index
= src
.DimensionIndex
;
1133 ureg_emit_dst( struct ureg_program
*ureg
,
1134 struct ureg_dst dst
)
1136 unsigned size
= 1 + (dst
.Indirect
? 1 : 0) +
1137 (dst
.Dimension
? (dst
.DimIndirect
? 2 : 1) : 0);
1139 union tgsi_any_token
*out
= get_tokens( ureg
, DOMAIN_INSN
, size
);
1142 assert(dst
.File
!= TGSI_FILE_NULL
);
1143 assert(dst
.File
!= TGSI_FILE_CONSTANT
);
1144 assert(dst
.File
!= TGSI_FILE_INPUT
);
1145 assert(dst
.File
!= TGSI_FILE_SAMPLER
);
1146 assert(dst
.File
!= TGSI_FILE_SAMPLER_VIEW
);
1147 assert(dst
.File
!= TGSI_FILE_IMMEDIATE
);
1148 assert(dst
.File
< TGSI_FILE_COUNT
);
1151 out
[n
].dst
.File
= dst
.File
;
1152 out
[n
].dst
.WriteMask
= dst
.WriteMask
;
1153 out
[n
].dst
.Indirect
= dst
.Indirect
;
1154 out
[n
].dst
.Index
= dst
.Index
;
1159 out
[n
].ind
.File
= dst
.IndirectFile
;
1160 out
[n
].ind
.Swizzle
= dst
.IndirectSwizzle
;
1161 out
[n
].ind
.Index
= dst
.IndirectIndex
;
1162 if (!ureg
->supports_any_inout_decl_range
&&
1163 (dst
.File
== TGSI_FILE_INPUT
|| dst
.File
== TGSI_FILE_OUTPUT
))
1164 out
[n
].ind
.ArrayID
= 0;
1166 out
[n
].ind
.ArrayID
= dst
.ArrayID
;
1170 if (dst
.Dimension
) {
1171 out
[0].dst
.Dimension
= 1;
1172 out
[n
].dim
.Dimension
= 0;
1173 out
[n
].dim
.Padding
= 0;
1174 if (dst
.DimIndirect
) {
1175 out
[n
].dim
.Indirect
= 1;
1176 out
[n
].dim
.Index
= dst
.DimensionIndex
;
1179 out
[n
].ind
.File
= dst
.DimIndFile
;
1180 out
[n
].ind
.Swizzle
= dst
.DimIndSwizzle
;
1181 out
[n
].ind
.Index
= dst
.DimIndIndex
;
1182 if (!ureg
->supports_any_inout_decl_range
&&
1183 (dst
.File
== TGSI_FILE_INPUT
|| dst
.File
== TGSI_FILE_OUTPUT
))
1184 out
[n
].ind
.ArrayID
= 0;
1186 out
[n
].ind
.ArrayID
= dst
.ArrayID
;
1188 out
[n
].dim
.Indirect
= 0;
1189 out
[n
].dim
.Index
= dst
.DimensionIndex
;
1198 static void validate( unsigned opcode
,
1203 const struct tgsi_opcode_info
*info
= tgsi_get_opcode_info( opcode
);
1206 assert(nr_dst
== info
->num_dst
);
1207 assert(nr_src
== info
->num_src
);
1212 struct ureg_emit_insn_result
1213 ureg_emit_insn(struct ureg_program
*ureg
,
1219 union tgsi_any_token
*out
;
1221 struct ureg_emit_insn_result result
;
1223 validate( opcode
, num_dst
, num_src
);
1225 out
= get_tokens( ureg
, DOMAIN_INSN
, count
);
1226 out
[0].insn
= tgsi_default_instruction();
1227 out
[0].insn
.Opcode
= opcode
;
1228 out
[0].insn
.Saturate
= saturate
;
1229 out
[0].insn
.NumDstRegs
= num_dst
;
1230 out
[0].insn
.NumSrcRegs
= num_src
;
1232 result
.insn_token
= ureg
->domain
[DOMAIN_INSN
].count
- count
;
1233 result
.extended_token
= result
.insn_token
;
1235 ureg
->nr_instructions
++;
1242 * Emit a label token.
1243 * \param label_token returns a token number indicating where the label
1244 * needs to be patched later. Later, this value should be passed to the
1245 * ureg_fixup_label() function.
1248 ureg_emit_label(struct ureg_program
*ureg
,
1249 unsigned extended_token
,
1250 unsigned *label_token
)
1252 union tgsi_any_token
*out
, *insn
;
1257 out
= get_tokens( ureg
, DOMAIN_INSN
, 1 );
1260 insn
= retrieve_token( ureg
, DOMAIN_INSN
, extended_token
);
1261 insn
->insn
.Label
= 1;
1263 *label_token
= ureg
->domain
[DOMAIN_INSN
].count
- 1;
1266 /* Will return a number which can be used in a label to point to the
1267 * next instruction to be emitted.
1270 ureg_get_instruction_number( struct ureg_program
*ureg
)
1272 return ureg
->nr_instructions
;
1275 /* Patch a given label (expressed as a token number) to point to a
1276 * given instruction (expressed as an instruction number).
1279 ureg_fixup_label(struct ureg_program
*ureg
,
1280 unsigned label_token
,
1281 unsigned instruction_number
)
1283 union tgsi_any_token
*out
= retrieve_token( ureg
, DOMAIN_INSN
, label_token
);
1285 out
->insn_label
.Label
= instruction_number
;
1290 ureg_emit_texture(struct ureg_program
*ureg
,
1291 unsigned extended_token
,
1292 unsigned target
, unsigned return_type
, unsigned num_offsets
)
1294 union tgsi_any_token
*out
, *insn
;
1296 out
= get_tokens( ureg
, DOMAIN_INSN
, 1 );
1297 insn
= retrieve_token( ureg
, DOMAIN_INSN
, extended_token
);
1299 insn
->insn
.Texture
= 1;
1302 out
[0].insn_texture
.Texture
= target
;
1303 out
[0].insn_texture
.NumOffsets
= num_offsets
;
1304 out
[0].insn_texture
.ReturnType
= return_type
;
1308 ureg_emit_texture_offset(struct ureg_program
*ureg
,
1309 const struct tgsi_texture_offset
*offset
)
1311 union tgsi_any_token
*out
;
1313 out
= get_tokens( ureg
, DOMAIN_INSN
, 1);
1316 out
[0].insn_texture_offset
= *offset
;
1321 ureg_emit_memory(struct ureg_program
*ureg
,
1322 unsigned extended_token
,
1327 union tgsi_any_token
*out
, *insn
;
1329 out
= get_tokens( ureg
, DOMAIN_INSN
, 1 );
1330 insn
= retrieve_token( ureg
, DOMAIN_INSN
, extended_token
);
1332 insn
->insn
.Memory
= 1;
1335 out
[0].insn_memory
.Qualifier
= qualifier
;
1336 out
[0].insn_memory
.Texture
= texture
;
1337 out
[0].insn_memory
.Format
= format
;
1341 ureg_fixup_insn_size(struct ureg_program
*ureg
,
1344 union tgsi_any_token
*out
= retrieve_token( ureg
, DOMAIN_INSN
, insn
);
1346 assert(out
->insn
.Type
== TGSI_TOKEN_TYPE_INSTRUCTION
);
1347 out
->insn
.NrTokens
= ureg
->domain
[DOMAIN_INSN
].count
- insn
- 1;
1352 ureg_insn(struct ureg_program
*ureg
,
1354 const struct ureg_dst
*dst
,
1356 const struct ureg_src
*src
,
1359 struct ureg_emit_insn_result insn
;
1363 if (nr_dst
&& ureg_dst_is_empty(dst
[0])) {
1367 saturate
= nr_dst
? dst
[0].Saturate
: FALSE
;
1369 insn
= ureg_emit_insn(ureg
,
1375 for (i
= 0; i
< nr_dst
; i
++)
1376 ureg_emit_dst( ureg
, dst
[i
] );
1378 for (i
= 0; i
< nr_src
; i
++)
1379 ureg_emit_src( ureg
, src
[i
] );
1381 ureg_fixup_insn_size( ureg
, insn
.insn_token
);
1385 ureg_tex_insn(struct ureg_program
*ureg
,
1387 const struct ureg_dst
*dst
,
1390 unsigned return_type
,
1391 const struct tgsi_texture_offset
*texoffsets
,
1393 const struct ureg_src
*src
,
1396 struct ureg_emit_insn_result insn
;
1400 if (nr_dst
&& ureg_dst_is_empty(dst
[0])) {
1404 saturate
= nr_dst
? dst
[0].Saturate
: FALSE
;
1406 insn
= ureg_emit_insn(ureg
,
1412 ureg_emit_texture( ureg
, insn
.extended_token
, target
, return_type
,
1415 for (i
= 0; i
< nr_offset
; i
++)
1416 ureg_emit_texture_offset( ureg
, &texoffsets
[i
]);
1418 for (i
= 0; i
< nr_dst
; i
++)
1419 ureg_emit_dst( ureg
, dst
[i
] );
1421 for (i
= 0; i
< nr_src
; i
++)
1422 ureg_emit_src( ureg
, src
[i
] );
1424 ureg_fixup_insn_size( ureg
, insn
.insn_token
);
1429 ureg_memory_insn(struct ureg_program
*ureg
,
1431 const struct ureg_dst
*dst
,
1433 const struct ureg_src
*src
,
1439 struct ureg_emit_insn_result insn
;
1442 insn
= ureg_emit_insn(ureg
,
1448 ureg_emit_memory(ureg
, insn
.extended_token
, qualifier
, texture
, format
);
1450 for (i
= 0; i
< nr_dst
; i
++)
1451 ureg_emit_dst(ureg
, dst
[i
]);
1453 for (i
= 0; i
< nr_src
; i
++)
1454 ureg_emit_src(ureg
, src
[i
]);
1456 ureg_fixup_insn_size(ureg
, insn
.insn_token
);
1461 emit_decl_semantic(struct ureg_program
*ureg
,
1465 unsigned semantic_name
,
1466 unsigned semantic_index
,
1468 unsigned usage_mask
,
1471 union tgsi_any_token
*out
= get_tokens(ureg
, DOMAIN_DECL
, array_id
? 4 : 3);
1474 out
[0].decl
.Type
= TGSI_TOKEN_TYPE_DECLARATION
;
1475 out
[0].decl
.NrTokens
= 3;
1476 out
[0].decl
.File
= file
;
1477 out
[0].decl
.UsageMask
= usage_mask
;
1478 out
[0].decl
.Semantic
= 1;
1479 out
[0].decl
.Array
= array_id
!= 0;
1482 out
[1].decl_range
.First
= first
;
1483 out
[1].decl_range
.Last
= last
;
1486 out
[2].decl_semantic
.Name
= semantic_name
;
1487 out
[2].decl_semantic
.Index
= semantic_index
;
1488 out
[2].decl_semantic
.StreamX
= streams
& 3;
1489 out
[2].decl_semantic
.StreamY
= (streams
>> 2) & 3;
1490 out
[2].decl_semantic
.StreamZ
= (streams
>> 4) & 3;
1491 out
[2].decl_semantic
.StreamW
= (streams
>> 6) & 3;
1495 out
[3].array
.ArrayID
= array_id
;
1501 emit_decl_fs(struct ureg_program
*ureg
,
1505 unsigned semantic_name
,
1506 unsigned semantic_index
,
1507 unsigned interpolate
,
1508 unsigned cylindrical_wrap
,
1509 unsigned interpolate_location
,
1511 unsigned usage_mask
)
1513 union tgsi_any_token
*out
= get_tokens(ureg
, DOMAIN_DECL
,
1517 out
[0].decl
.Type
= TGSI_TOKEN_TYPE_DECLARATION
;
1518 out
[0].decl
.NrTokens
= 4;
1519 out
[0].decl
.File
= file
;
1520 out
[0].decl
.UsageMask
= usage_mask
;
1521 out
[0].decl
.Interpolate
= 1;
1522 out
[0].decl
.Semantic
= 1;
1523 out
[0].decl
.Array
= array_id
!= 0;
1526 out
[1].decl_range
.First
= first
;
1527 out
[1].decl_range
.Last
= last
;
1530 out
[2].decl_interp
.Interpolate
= interpolate
;
1531 out
[2].decl_interp
.CylindricalWrap
= cylindrical_wrap
;
1532 out
[2].decl_interp
.Location
= interpolate_location
;
1535 out
[3].decl_semantic
.Name
= semantic_name
;
1536 out
[3].decl_semantic
.Index
= semantic_index
;
1540 out
[4].array
.ArrayID
= array_id
;
1545 emit_decl_temps( struct ureg_program
*ureg
,
1546 unsigned first
, unsigned last
,
1550 union tgsi_any_token
*out
= get_tokens( ureg
, DOMAIN_DECL
,
1554 out
[0].decl
.Type
= TGSI_TOKEN_TYPE_DECLARATION
;
1555 out
[0].decl
.NrTokens
= 2;
1556 out
[0].decl
.File
= TGSI_FILE_TEMPORARY
;
1557 out
[0].decl
.UsageMask
= TGSI_WRITEMASK_XYZW
;
1558 out
[0].decl
.Local
= local
;
1561 out
[1].decl_range
.First
= first
;
1562 out
[1].decl_range
.Last
= last
;
1565 out
[0].decl
.Array
= 1;
1567 out
[2].array
.ArrayID
= arrayid
;
1571 static void emit_decl_range( struct ureg_program
*ureg
,
1576 union tgsi_any_token
*out
= get_tokens( ureg
, DOMAIN_DECL
, 2 );
1579 out
[0].decl
.Type
= TGSI_TOKEN_TYPE_DECLARATION
;
1580 out
[0].decl
.NrTokens
= 2;
1581 out
[0].decl
.File
= file
;
1582 out
[0].decl
.UsageMask
= TGSI_WRITEMASK_XYZW
;
1583 out
[0].decl
.Semantic
= 0;
1586 out
[1].decl_range
.First
= first
;
1587 out
[1].decl_range
.Last
= first
+ count
- 1;
1591 emit_decl_range2D(struct ureg_program
*ureg
,
1597 union tgsi_any_token
*out
= get_tokens(ureg
, DOMAIN_DECL
, 3);
1600 out
[0].decl
.Type
= TGSI_TOKEN_TYPE_DECLARATION
;
1601 out
[0].decl
.NrTokens
= 3;
1602 out
[0].decl
.File
= file
;
1603 out
[0].decl
.UsageMask
= TGSI_WRITEMASK_XYZW
;
1604 out
[0].decl
.Dimension
= 1;
1607 out
[1].decl_range
.First
= first
;
1608 out
[1].decl_range
.Last
= last
;
1611 out
[2].decl_dim
.Index2D
= index2D
;
1615 emit_decl_sampler_view(struct ureg_program
*ureg
,
1618 unsigned return_type_x
,
1619 unsigned return_type_y
,
1620 unsigned return_type_z
,
1621 unsigned return_type_w
)
1623 union tgsi_any_token
*out
= get_tokens(ureg
, DOMAIN_DECL
, 3);
1626 out
[0].decl
.Type
= TGSI_TOKEN_TYPE_DECLARATION
;
1627 out
[0].decl
.NrTokens
= 3;
1628 out
[0].decl
.File
= TGSI_FILE_SAMPLER_VIEW
;
1629 out
[0].decl
.UsageMask
= TGSI_WRITEMASK_XYZW
;
1632 out
[1].decl_range
.First
= index
;
1633 out
[1].decl_range
.Last
= index
;
1636 out
[2].decl_sampler_view
.Resource
= target
;
1637 out
[2].decl_sampler_view
.ReturnTypeX
= return_type_x
;
1638 out
[2].decl_sampler_view
.ReturnTypeY
= return_type_y
;
1639 out
[2].decl_sampler_view
.ReturnTypeZ
= return_type_z
;
1640 out
[2].decl_sampler_view
.ReturnTypeW
= return_type_w
;
1644 emit_decl_image(struct ureg_program
*ureg
,
1651 union tgsi_any_token
*out
= get_tokens(ureg
, DOMAIN_DECL
, 3);
1654 out
[0].decl
.Type
= TGSI_TOKEN_TYPE_DECLARATION
;
1655 out
[0].decl
.NrTokens
= 3;
1656 out
[0].decl
.File
= TGSI_FILE_IMAGE
;
1657 out
[0].decl
.UsageMask
= TGSI_WRITEMASK_XYZW
;
1660 out
[1].decl_range
.First
= index
;
1661 out
[1].decl_range
.Last
= index
;
1664 out
[2].decl_image
.Resource
= target
;
1665 out
[2].decl_image
.Writable
= wr
;
1666 out
[2].decl_image
.Raw
= raw
;
1667 out
[2].decl_image
.Format
= format
;
1671 emit_decl_buffer(struct ureg_program
*ureg
,
1675 union tgsi_any_token
*out
= get_tokens(ureg
, DOMAIN_DECL
, 2);
1678 out
[0].decl
.Type
= TGSI_TOKEN_TYPE_DECLARATION
;
1679 out
[0].decl
.NrTokens
= 2;
1680 out
[0].decl
.File
= TGSI_FILE_BUFFER
;
1681 out
[0].decl
.UsageMask
= TGSI_WRITEMASK_XYZW
;
1682 out
[0].decl
.Atomic
= atomic
;
1685 out
[1].decl_range
.First
= index
;
1686 out
[1].decl_range
.Last
= index
;
1690 emit_decl_memory(struct ureg_program
*ureg
, unsigned memory_type
)
1692 union tgsi_any_token
*out
= get_tokens(ureg
, DOMAIN_DECL
, 2);
1695 out
[0].decl
.Type
= TGSI_TOKEN_TYPE_DECLARATION
;
1696 out
[0].decl
.NrTokens
= 2;
1697 out
[0].decl
.File
= TGSI_FILE_MEMORY
;
1698 out
[0].decl
.UsageMask
= TGSI_WRITEMASK_XYZW
;
1699 out
[0].decl
.MemType
= memory_type
;
1702 out
[1].decl_range
.First
= memory_type
;
1703 out
[1].decl_range
.Last
= memory_type
;
1707 emit_immediate( struct ureg_program
*ureg
,
1711 union tgsi_any_token
*out
= get_tokens( ureg
, DOMAIN_DECL
, 5 );
1714 out
[0].imm
.Type
= TGSI_TOKEN_TYPE_IMMEDIATE
;
1715 out
[0].imm
.NrTokens
= 5;
1716 out
[0].imm
.DataType
= type
;
1717 out
[0].imm
.Padding
= 0;
1719 out
[1].imm_data
.Uint
= v
[0];
1720 out
[2].imm_data
.Uint
= v
[1];
1721 out
[3].imm_data
.Uint
= v
[2];
1722 out
[4].imm_data
.Uint
= v
[3];
1726 emit_property(struct ureg_program
*ureg
,
1730 union tgsi_any_token
*out
= get_tokens(ureg
, DOMAIN_DECL
, 2);
1733 out
[0].prop
.Type
= TGSI_TOKEN_TYPE_PROPERTY
;
1734 out
[0].prop
.NrTokens
= 2;
1735 out
[0].prop
.PropertyName
= name
;
1737 out
[1].prop_data
.Data
= data
;
1741 static void emit_decls( struct ureg_program
*ureg
)
1745 for (i
= 0; i
< ARRAY_SIZE(ureg
->properties
); i
++)
1746 if (ureg
->properties
[i
] != ~0)
1747 emit_property(ureg
, i
, ureg
->properties
[i
]);
1749 if (ureg
->processor
== PIPE_SHADER_VERTEX
) {
1750 for (i
= 0; i
< PIPE_MAX_ATTRIBS
; i
++) {
1751 if (ureg
->vs_inputs
[i
/32] & (1u << (i
%32))) {
1752 emit_decl_range( ureg
, TGSI_FILE_INPUT
, i
, 1 );
1755 } else if (ureg
->processor
== PIPE_SHADER_FRAGMENT
) {
1756 if (ureg
->supports_any_inout_decl_range
) {
1757 for (i
= 0; i
< ureg
->nr_inputs
; i
++) {
1760 ureg
->input
[i
].first
,
1761 ureg
->input
[i
].last
,
1762 ureg
->input
[i
].semantic_name
,
1763 ureg
->input
[i
].semantic_index
,
1764 ureg
->input
[i
].interp
,
1765 ureg
->input
[i
].cylindrical_wrap
,
1766 ureg
->input
[i
].interp_location
,
1767 ureg
->input
[i
].array_id
,
1768 ureg
->input
[i
].usage_mask
);
1772 for (i
= 0; i
< ureg
->nr_inputs
; i
++) {
1773 for (j
= ureg
->input
[i
].first
; j
<= ureg
->input
[i
].last
; j
++) {
1777 ureg
->input
[i
].semantic_name
,
1778 ureg
->input
[i
].semantic_index
+
1779 (j
- ureg
->input
[i
].first
),
1780 ureg
->input
[i
].interp
,
1781 ureg
->input
[i
].cylindrical_wrap
,
1782 ureg
->input
[i
].interp_location
, 0,
1783 ureg
->input
[i
].usage_mask
);
1788 if (ureg
->supports_any_inout_decl_range
) {
1789 for (i
= 0; i
< ureg
->nr_inputs
; i
++) {
1790 emit_decl_semantic(ureg
,
1792 ureg
->input
[i
].first
,
1793 ureg
->input
[i
].last
,
1794 ureg
->input
[i
].semantic_name
,
1795 ureg
->input
[i
].semantic_index
,
1797 TGSI_WRITEMASK_XYZW
,
1798 ureg
->input
[i
].array_id
);
1802 for (i
= 0; i
< ureg
->nr_inputs
; i
++) {
1803 for (j
= ureg
->input
[i
].first
; j
<= ureg
->input
[i
].last
; j
++) {
1804 emit_decl_semantic(ureg
,
1807 ureg
->input
[i
].semantic_name
,
1808 ureg
->input
[i
].semantic_index
+
1809 (j
- ureg
->input
[i
].first
),
1811 TGSI_WRITEMASK_XYZW
, 0);
1817 for (i
= 0; i
< ureg
->nr_system_values
; i
++) {
1818 emit_decl_semantic(ureg
,
1819 TGSI_FILE_SYSTEM_VALUE
,
1822 ureg
->system_value
[i
].semantic_name
,
1823 ureg
->system_value
[i
].semantic_index
,
1825 TGSI_WRITEMASK_XYZW
, 0);
1828 if (ureg
->supports_any_inout_decl_range
) {
1829 for (i
= 0; i
< ureg
->nr_outputs
; i
++) {
1830 emit_decl_semantic(ureg
,
1832 ureg
->output
[i
].first
,
1833 ureg
->output
[i
].last
,
1834 ureg
->output
[i
].semantic_name
,
1835 ureg
->output
[i
].semantic_index
,
1836 ureg
->output
[i
].streams
,
1837 ureg
->output
[i
].usage_mask
,
1838 ureg
->output
[i
].array_id
);
1842 for (i
= 0; i
< ureg
->nr_outputs
; i
++) {
1843 for (j
= ureg
->output
[i
].first
; j
<= ureg
->output
[i
].last
; j
++) {
1844 emit_decl_semantic(ureg
,
1847 ureg
->output
[i
].semantic_name
,
1848 ureg
->output
[i
].semantic_index
+
1849 (j
- ureg
->output
[i
].first
),
1850 ureg
->output
[i
].streams
,
1851 ureg
->output
[i
].usage_mask
, 0);
1856 for (i
= 0; i
< ureg
->nr_samplers
; i
++) {
1857 emit_decl_range( ureg
,
1859 ureg
->sampler
[i
].Index
, 1 );
1862 for (i
= 0; i
< ureg
->nr_sampler_views
; i
++) {
1863 emit_decl_sampler_view(ureg
,
1864 ureg
->sampler_view
[i
].index
,
1865 ureg
->sampler_view
[i
].target
,
1866 ureg
->sampler_view
[i
].return_type_x
,
1867 ureg
->sampler_view
[i
].return_type_y
,
1868 ureg
->sampler_view
[i
].return_type_z
,
1869 ureg
->sampler_view
[i
].return_type_w
);
1872 for (i
= 0; i
< ureg
->nr_images
; i
++) {
1873 emit_decl_image(ureg
,
1874 ureg
->image
[i
].index
,
1875 ureg
->image
[i
].target
,
1876 ureg
->image
[i
].format
,
1878 ureg
->image
[i
].raw
);
1881 for (i
= 0; i
< ureg
->nr_buffers
; i
++) {
1882 emit_decl_buffer(ureg
, ureg
->buffer
[i
].index
, ureg
->buffer
[i
].atomic
);
1885 for (i
= 0; i
< TGSI_MEMORY_TYPE_COUNT
; i
++) {
1886 if (ureg
->use_memory
[i
])
1887 emit_decl_memory(ureg
, i
);
1890 if (ureg
->const_decls
.nr_constant_ranges
) {
1891 for (i
= 0; i
< ureg
->const_decls
.nr_constant_ranges
; i
++) {
1892 emit_decl_range(ureg
,
1894 ureg
->const_decls
.constant_range
[i
].first
,
1895 ureg
->const_decls
.constant_range
[i
].last
- ureg
->const_decls
.constant_range
[i
].first
+ 1);
1899 for (i
= 0; i
< PIPE_MAX_CONSTANT_BUFFERS
; i
++) {
1900 struct const_decl
*decl
= &ureg
->const_decls2D
[i
];
1902 if (decl
->nr_constant_ranges
) {
1905 for (j
= 0; j
< decl
->nr_constant_ranges
; j
++) {
1906 emit_decl_range2D(ureg
,
1908 decl
->constant_range
[j
].first
,
1909 decl
->constant_range
[j
].last
,
1915 if (ureg
->nr_temps
) {
1917 for (i
= 0; i
< ureg
->nr_temps
;) {
1918 boolean local
= util_bitmask_get(ureg
->local_temps
, i
);
1920 i
= util_bitmask_get_next_index(ureg
->decl_temps
, i
+ 1);
1921 if (i
== UTIL_BITMASK_INVALID_INDEX
)
1924 if (array
< ureg
->nr_array_temps
&& ureg
->array_temps
[array
] == first
)
1925 emit_decl_temps( ureg
, first
, i
- 1, local
, ++array
);
1927 emit_decl_temps( ureg
, first
, i
- 1, local
, 0 );
1931 if (ureg
->nr_addrs
) {
1932 emit_decl_range( ureg
,
1934 0, ureg
->nr_addrs
);
1937 for (i
= 0; i
< ureg
->nr_immediates
; i
++) {
1938 emit_immediate( ureg
,
1939 ureg
->immediate
[i
].value
.u
,
1940 ureg
->immediate
[i
].type
);
1944 /* Append the instruction tokens onto the declarations to build a
1945 * contiguous stream suitable to send to the driver.
1947 static void copy_instructions( struct ureg_program
*ureg
)
1949 unsigned nr_tokens
= ureg
->domain
[DOMAIN_INSN
].count
;
1950 union tgsi_any_token
*out
= get_tokens( ureg
,
1955 ureg
->domain
[DOMAIN_INSN
].tokens
,
1956 nr_tokens
* sizeof out
[0] );
1961 fixup_header_size(struct ureg_program
*ureg
)
1963 union tgsi_any_token
*out
= retrieve_token( ureg
, DOMAIN_DECL
, 0 );
1965 out
->header
.BodySize
= ureg
->domain
[DOMAIN_DECL
].count
- 2;
1970 emit_header( struct ureg_program
*ureg
)
1972 union tgsi_any_token
*out
= get_tokens( ureg
, DOMAIN_DECL
, 2 );
1974 out
[0].header
.HeaderSize
= 2;
1975 out
[0].header
.BodySize
= 0;
1977 out
[1].processor
.Processor
= ureg
->processor
;
1978 out
[1].processor
.Padding
= 0;
1982 const struct tgsi_token
*ureg_finalize( struct ureg_program
*ureg
)
1984 const struct tgsi_token
*tokens
;
1986 switch (ureg
->processor
) {
1987 case PIPE_SHADER_VERTEX
:
1988 case PIPE_SHADER_TESS_EVAL
:
1989 ureg_property(ureg
, TGSI_PROPERTY_NEXT_SHADER
,
1990 ureg
->next_shader_processor
== -1 ?
1991 PIPE_SHADER_FRAGMENT
:
1992 ureg
->next_shader_processor
);
1996 emit_header( ureg
);
1998 copy_instructions( ureg
);
1999 fixup_header_size( ureg
);
2001 if (ureg
->domain
[0].tokens
== error_tokens
||
2002 ureg
->domain
[1].tokens
== error_tokens
) {
2003 debug_printf("%s: error in generated shader\n", __FUNCTION__
);
2008 tokens
= &ureg
->domain
[DOMAIN_DECL
].tokens
[0].token
;
2011 debug_printf("%s: emitted shader %d tokens:\n", __FUNCTION__
,
2012 ureg
->domain
[DOMAIN_DECL
].count
);
2013 tgsi_dump( tokens
, 0 );
2017 if (tokens
&& !tgsi_sanity_check(tokens
)) {
2018 debug_printf("tgsi_ureg.c, sanity check failed on generated tokens:\n");
2019 tgsi_dump(tokens
, 0);
2029 void *ureg_create_shader( struct ureg_program
*ureg
,
2030 struct pipe_context
*pipe
,
2031 const struct pipe_stream_output_info
*so
)
2033 struct pipe_shader_state state
;
2035 pipe_shader_state_from_tgsi(&state
, ureg_finalize(ureg
));
2040 state
.stream_output
= *so
;
2042 switch (ureg
->processor
) {
2043 case PIPE_SHADER_VERTEX
:
2044 return pipe
->create_vs_state(pipe
, &state
);
2045 case PIPE_SHADER_TESS_CTRL
:
2046 return pipe
->create_tcs_state(pipe
, &state
);
2047 case PIPE_SHADER_TESS_EVAL
:
2048 return pipe
->create_tes_state(pipe
, &state
);
2049 case PIPE_SHADER_GEOMETRY
:
2050 return pipe
->create_gs_state(pipe
, &state
);
2051 case PIPE_SHADER_FRAGMENT
:
2052 return pipe
->create_fs_state(pipe
, &state
);
2059 const struct tgsi_token
*ureg_get_tokens( struct ureg_program
*ureg
,
2060 unsigned *nr_tokens
)
2062 const struct tgsi_token
*tokens
;
2064 ureg_finalize(ureg
);
2066 tokens
= &ureg
->domain
[DOMAIN_DECL
].tokens
[0].token
;
2069 *nr_tokens
= ureg
->domain
[DOMAIN_DECL
].count
;
2071 ureg
->domain
[DOMAIN_DECL
].tokens
= 0;
2072 ureg
->domain
[DOMAIN_DECL
].size
= 0;
2073 ureg
->domain
[DOMAIN_DECL
].order
= 0;
2074 ureg
->domain
[DOMAIN_DECL
].count
= 0;
2080 void ureg_free_tokens( const struct tgsi_token
*tokens
)
2082 FREE((struct tgsi_token
*)tokens
);
2086 struct ureg_program
*
2087 ureg_create(unsigned processor
)
2089 return ureg_create_with_screen(processor
, NULL
);
2093 struct ureg_program
*
2094 ureg_create_with_screen(unsigned processor
, struct pipe_screen
*screen
)
2097 struct ureg_program
*ureg
= CALLOC_STRUCT( ureg_program
);
2101 ureg
->processor
= processor
;
2102 ureg
->supports_any_inout_decl_range
=
2104 screen
->get_shader_param(screen
, processor
,
2105 PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE
) != 0;
2106 ureg
->next_shader_processor
= -1;
2108 for (i
= 0; i
< ARRAY_SIZE(ureg
->properties
); i
++)
2109 ureg
->properties
[i
] = ~0;
2111 ureg
->free_temps
= util_bitmask_create();
2112 if (ureg
->free_temps
== NULL
)
2115 ureg
->local_temps
= util_bitmask_create();
2116 if (ureg
->local_temps
== NULL
)
2117 goto no_local_temps
;
2119 ureg
->decl_temps
= util_bitmask_create();
2120 if (ureg
->decl_temps
== NULL
)
2126 util_bitmask_destroy(ureg
->local_temps
);
2128 util_bitmask_destroy(ureg
->free_temps
);
2137 ureg_set_next_shader_processor(struct ureg_program
*ureg
, unsigned processor
)
2139 ureg
->next_shader_processor
= processor
;
2144 ureg_get_nr_outputs( const struct ureg_program
*ureg
)
2148 return ureg
->nr_outputs
;
2152 void ureg_destroy( struct ureg_program
*ureg
)
2156 for (i
= 0; i
< ARRAY_SIZE(ureg
->domain
); i
++) {
2157 if (ureg
->domain
[i
].tokens
&&
2158 ureg
->domain
[i
].tokens
!= error_tokens
)
2159 FREE(ureg
->domain
[i
].tokens
);
2162 util_bitmask_destroy(ureg
->free_temps
);
2163 util_bitmask_destroy(ureg
->local_temps
);
2164 util_bitmask_destroy(ureg
->decl_temps
);