2 Copyright (C) Intel Corp. 2006. All Rights Reserved.
3 Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
4 develop this 3D driver.
6 Permission is hereby granted, free of charge, to any person obtaining
7 a 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, sublicense, 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
16 portions of the Software.
18 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
19 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
21 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
22 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
23 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
24 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **********************************************************************/
29 * Keith Whitwell <keith@tungstengraphics.com>
33 #include "brw_context.h"
34 #include "brw_defines.h"
37 #include "glsl/ralloc.h"
39 /* Returns the corresponding conditional mod for swapping src0 and
43 brw_swap_cmod(uint32_t cmod
)
46 case BRW_CONDITIONAL_Z
:
47 case BRW_CONDITIONAL_NZ
:
49 case BRW_CONDITIONAL_G
:
50 return BRW_CONDITIONAL_LE
;
51 case BRW_CONDITIONAL_GE
:
52 return BRW_CONDITIONAL_L
;
53 case BRW_CONDITIONAL_L
:
54 return BRW_CONDITIONAL_GE
;
55 case BRW_CONDITIONAL_LE
:
56 return BRW_CONDITIONAL_G
;
63 /* How does predicate control work when execution_size != 8? Do I
64 * need to test/set for 0xffff when execution_size is 16?
66 void brw_set_predicate_control_flag_value( struct brw_compile
*p
, GLuint value
)
68 p
->current
->header
.predicate_control
= BRW_PREDICATE_NONE
;
71 if (value
!= p
->flag_value
) {
72 brw_push_insn_state(p
);
73 brw_MOV(p
, brw_flag_reg(), brw_imm_uw(value
));
74 p
->flag_value
= value
;
75 brw_pop_insn_state(p
);
78 p
->current
->header
.predicate_control
= BRW_PREDICATE_NORMAL
;
82 void brw_set_predicate_control( struct brw_compile
*p
, GLuint pc
)
84 p
->current
->header
.predicate_control
= pc
;
87 void brw_set_predicate_inverse(struct brw_compile
*p
, bool predicate_inverse
)
89 p
->current
->header
.predicate_inverse
= predicate_inverse
;
92 void brw_set_conditionalmod( struct brw_compile
*p
, GLuint conditional
)
94 p
->current
->header
.destreg__conditionalmod
= conditional
;
97 void brw_set_access_mode( struct brw_compile
*p
, GLuint access_mode
)
99 p
->current
->header
.access_mode
= access_mode
;
103 brw_set_compression_control(struct brw_compile
*p
,
104 enum brw_compression compression_control
)
106 p
->compressed
= (compression_control
== BRW_COMPRESSION_COMPRESSED
);
108 if (p
->brw
->intel
.gen
>= 6) {
109 /* Since we don't use the 32-wide support in gen6, we translate
110 * the pre-gen6 compression control here.
112 switch (compression_control
) {
113 case BRW_COMPRESSION_NONE
:
114 /* This is the "use the first set of bits of dmask/vmask/arf
115 * according to execsize" option.
117 p
->current
->header
.compression_control
= GEN6_COMPRESSION_1Q
;
119 case BRW_COMPRESSION_2NDHALF
:
120 /* For 8-wide, this is "use the second set of 8 bits." */
121 p
->current
->header
.compression_control
= GEN6_COMPRESSION_2Q
;
123 case BRW_COMPRESSION_COMPRESSED
:
124 /* For 16-wide instruction compression, use the first set of 16 bits
125 * since we don't do 32-wide dispatch.
127 p
->current
->header
.compression_control
= GEN6_COMPRESSION_1H
;
130 assert(!"not reached");
131 p
->current
->header
.compression_control
= GEN6_COMPRESSION_1H
;
135 p
->current
->header
.compression_control
= compression_control
;
139 void brw_set_mask_control( struct brw_compile
*p
, GLuint value
)
141 p
->current
->header
.mask_control
= value
;
144 void brw_set_saturate( struct brw_compile
*p
, GLuint value
)
146 p
->current
->header
.saturate
= value
;
149 void brw_set_acc_write_control(struct brw_compile
*p
, GLuint value
)
151 if (p
->brw
->intel
.gen
>= 6)
152 p
->current
->header
.acc_wr_control
= value
;
155 void brw_push_insn_state( struct brw_compile
*p
)
157 assert(p
->current
!= &p
->stack
[BRW_EU_MAX_INSN_STACK
-1]);
158 memcpy(p
->current
+1, p
->current
, sizeof(struct brw_instruction
));
159 p
->compressed_stack
[p
->current
- p
->stack
] = p
->compressed
;
163 void brw_pop_insn_state( struct brw_compile
*p
)
165 assert(p
->current
!= p
->stack
);
167 p
->compressed
= p
->compressed_stack
[p
->current
- p
->stack
];
171 /***********************************************************************
174 brw_init_compile(struct brw_context
*brw
, struct brw_compile
*p
, void *mem_ctx
)
178 * Set the initial instruction store array size to 1024, if found that
179 * isn't enough, then it will double the store size at brw_next_insn()
180 * until out of memory.
182 p
->store_size
= 1024;
183 p
->store
= rzalloc_array(mem_ctx
, struct brw_instruction
, p
->store_size
);
185 p
->current
= p
->stack
;
186 p
->compressed
= false;
187 memset(p
->current
, 0, sizeof(p
->current
[0]));
189 p
->mem_ctx
= mem_ctx
;
193 brw_set_mask_control(p
, BRW_MASK_ENABLE
); /* what does this do? */
194 brw_set_saturate(p
, 0);
195 brw_set_compression_control(p
, BRW_COMPRESSION_NONE
);
196 brw_set_predicate_control_flag_value(p
, 0xff);
198 /* Set up control flow stack */
199 p
->if_stack_depth
= 0;
200 p
->if_stack_array_size
= 16;
201 p
->if_stack
= rzalloc_array(mem_ctx
, int, p
->if_stack_array_size
);
203 p
->loop_stack_depth
= 0;
204 p
->loop_stack_array_size
= 16;
205 p
->loop_stack
= rzalloc_array(mem_ctx
, int, p
->loop_stack_array_size
);
206 p
->if_depth_in_loop
= rzalloc_array(mem_ctx
, int, p
->loop_stack_array_size
);
210 const GLuint
*brw_get_program( struct brw_compile
*p
,
215 for (i
= 0; i
< 8; i
++)
218 *sz
= p
->nr_insn
* sizeof(struct brw_instruction
);
219 return (const GLuint
*)p
->store
;
225 * Subroutine calls require special attention.
226 * Mesa instructions may be expanded into multiple hardware instructions
227 * so the prog_instruction::BranchTarget field can't be used as an index
228 * into the hardware instructions.
230 * The BranchTarget field isn't needed, however. Mesa's GLSL compiler
231 * emits CAL and BGNSUB instructions with labels that can be used to map
232 * subroutine calls to actual subroutine code blocks.
234 * The structures and function here implement patching of CAL instructions
235 * so they jump to the right subroutine code...
240 * For each OPCODE_BGNSUB we create one of these.
242 struct brw_glsl_label
244 const char *name
; /**< the label string */
245 GLuint position
; /**< the position of the brw instruction for this label */
246 struct brw_glsl_label
*next
; /**< next in linked list */
251 * For each OPCODE_CAL we create one of these.
255 GLuint call_inst_pos
; /**< location of the CAL instruction */
256 const char *sub_name
; /**< name of subroutine to call */
257 struct brw_glsl_call
*next
; /**< next in linked list */
262 * Called for each OPCODE_BGNSUB.
265 brw_save_label(struct brw_compile
*c
, const char *name
, GLuint position
)
267 struct brw_glsl_label
*label
= CALLOC_STRUCT(brw_glsl_label
);
269 label
->position
= position
;
270 label
->next
= c
->first_label
;
271 c
->first_label
= label
;
276 * Called for each OPCODE_CAL.
279 brw_save_call(struct brw_compile
*c
, const char *name
, GLuint call_pos
)
281 struct brw_glsl_call
*call
= CALLOC_STRUCT(brw_glsl_call
);
282 call
->call_inst_pos
= call_pos
;
283 call
->sub_name
= name
;
284 call
->next
= c
->first_call
;
285 c
->first_call
= call
;
290 * Lookup a label, return label's position/offset.
293 brw_lookup_label(struct brw_compile
*c
, const char *name
)
295 const struct brw_glsl_label
*label
;
296 for (label
= c
->first_label
; label
; label
= label
->next
) {
297 if (strcmp(name
, label
->name
) == 0) {
298 return label
->position
;
301 abort(); /* should never happen */
307 * When we're done generating code, this function is called to resolve
311 brw_resolve_cals(struct brw_compile
*c
)
313 const struct brw_glsl_call
*call
;
315 for (call
= c
->first_call
; call
; call
= call
->next
) {
316 const GLuint sub_loc
= brw_lookup_label(c
, call
->sub_name
);
317 struct brw_instruction
*brw_call_inst
= &c
->store
[call
->call_inst_pos
];
318 struct brw_instruction
*brw_sub_inst
= &c
->store
[sub_loc
];
319 GLint offset
= brw_sub_inst
- brw_call_inst
;
321 /* patch brw_inst1 to point to brw_inst2 */
322 brw_set_src1(c
, brw_call_inst
, brw_imm_d(offset
* 16));
325 /* free linked list of calls */
327 struct brw_glsl_call
*call
, *next
;
328 for (call
= c
->first_call
; call
; call
= next
) {
332 c
->first_call
= NULL
;
335 /* free linked list of labels */
337 struct brw_glsl_label
*label
, *next
;
338 for (label
= c
->first_label
; label
; label
= next
) {
342 c
->first_label
= NULL
;