2 * Copyright (C) 2005 Ben Skeggs.
4 * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
5 * Adaptation and modification for ATI/AMD Radeon R500 GPU chipsets.
9 * Permission is hereby granted, free of charge, to any person obtaining
10 * a copy of this software and associated documentation files (the
11 * "Software"), to deal in the Software without restriction, including
12 * without limitation the rights to use, copy, modify, merge, publish,
13 * distribute, sublicense, and/or sell copies of the Software, and to
14 * permit persons to whom the Software is furnished to do so, subject to
15 * the following conditions:
17 * The above copyright notice and this permission notice (including the
18 * next paragraph) shall be included in all copies or substantial
19 * portions of the Software.
21 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
22 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
23 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
24 * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
25 * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
26 * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
27 * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
34 * \author Ben Skeggs <darktama@iinet.net.au>
36 * \author Jerome Glisse <j.glisse@gmail.com>
38 * \author Corbin Simpson <MostAwesomeDude@gmail.com>
42 #include "r500_fragprog.h"
44 #include "../r300_reg.h"
46 #include "radeon_program_pair.h"
48 #define MAX_BRANCH_DEPTH_FULL 32
49 #define MAX_BRANCH_DEPTH_PARTIAL 4
52 struct r500_fragment_program_code *code = &c->code->code.r500
54 #define error(fmt, args...) do { \
55 rc_error(&c->Base, "%s::%s(): " fmt "\n", \
56 __FILE__, __FUNCTION__, ##args); \
66 struct r500_loop_info
{
80 struct radeon_compiler
* C
;
81 struct r500_fragment_program_code
* Code
;
83 struct branch_info
* Branches
;
84 unsigned int CurrentBranchDepth
;
85 unsigned int BranchesReserved
;
87 struct r500_loop_info
* Loops
;
88 unsigned int CurrentLoopDepth
;
89 unsigned int LoopsReserved
;
91 unsigned int MaxBranchDepth
;
95 static unsigned int translate_rgb_op(struct r300_fragment_program_compiler
*c
, rc_opcode opcode
)
98 case RC_OPCODE_CMP
: return R500_ALU_RGBA_OP_CMP
;
99 case RC_OPCODE_DDX
: return R500_ALU_RGBA_OP_MDH
;
100 case RC_OPCODE_DDY
: return R500_ALU_RGBA_OP_MDV
;
101 case RC_OPCODE_DP3
: return R500_ALU_RGBA_OP_DP3
;
102 case RC_OPCODE_DP4
: return R500_ALU_RGBA_OP_DP4
;
103 case RC_OPCODE_FRC
: return R500_ALU_RGBA_OP_FRC
;
105 error("translate_rgb_op: unknown opcode %s\n", rc_get_opcode_info(opcode
)->Name
);
109 case RC_OPCODE_MAD
: return R500_ALU_RGBA_OP_MAD
;
110 case RC_OPCODE_MAX
: return R500_ALU_RGBA_OP_MAX
;
111 case RC_OPCODE_MIN
: return R500_ALU_RGBA_OP_MIN
;
112 case RC_OPCODE_REPL_ALPHA
: return R500_ALU_RGBA_OP_SOP
;
116 static unsigned int translate_alpha_op(struct r300_fragment_program_compiler
*c
, rc_opcode opcode
)
119 case RC_OPCODE_CMP
: return R500_ALPHA_OP_CMP
;
120 case RC_OPCODE_COS
: return R500_ALPHA_OP_COS
;
121 case RC_OPCODE_DDX
: return R500_ALPHA_OP_MDH
;
122 case RC_OPCODE_DDY
: return R500_ALPHA_OP_MDV
;
123 case RC_OPCODE_DP3
: return R500_ALPHA_OP_DP
;
124 case RC_OPCODE_DP4
: return R500_ALPHA_OP_DP
;
125 case RC_OPCODE_EX2
: return R500_ALPHA_OP_EX2
;
126 case RC_OPCODE_FRC
: return R500_ALPHA_OP_FRC
;
127 case RC_OPCODE_LG2
: return R500_ALPHA_OP_LN2
;
129 error("translate_alpha_op: unknown opcode %s\n", rc_get_opcode_info(opcode
)->Name
);
133 case RC_OPCODE_MAD
: return R500_ALPHA_OP_MAD
;
134 case RC_OPCODE_MAX
: return R500_ALPHA_OP_MAX
;
135 case RC_OPCODE_MIN
: return R500_ALPHA_OP_MIN
;
136 case RC_OPCODE_RCP
: return R500_ALPHA_OP_RCP
;
137 case RC_OPCODE_RSQ
: return R500_ALPHA_OP_RSQ
;
138 case RC_OPCODE_SIN
: return R500_ALPHA_OP_SIN
;
142 static unsigned int fix_hw_swizzle(unsigned int swz
)
145 case RC_SWIZZLE_ZERO
:
146 case RC_SWIZZLE_UNUSED
:
149 case RC_SWIZZLE_HALF
:
160 static unsigned int translate_arg_rgb(struct rc_pair_instruction
*inst
, int arg
)
162 unsigned int t
= inst
->RGB
.Arg
[arg
].Source
;
164 t
|= inst
->RGB
.Arg
[arg
].Negate
<< 11;
165 t
|= inst
->RGB
.Arg
[arg
].Abs
<< 12;
167 for(comp
= 0; comp
< 3; ++comp
)
168 t
|= fix_hw_swizzle(GET_SWZ(inst
->RGB
.Arg
[arg
].Swizzle
, comp
)) << (3*comp
+ 2);
173 static unsigned int translate_arg_alpha(struct rc_pair_instruction
*inst
, int i
)
175 unsigned int t
= inst
->Alpha
.Arg
[i
].Source
;
176 t
|= fix_hw_swizzle(inst
->Alpha
.Arg
[i
].Swizzle
) << 2;
177 t
|= inst
->Alpha
.Arg
[i
].Negate
<< 5;
178 t
|= inst
->Alpha
.Arg
[i
].Abs
<< 6;
182 static uint32_t translate_alu_result_op(struct r300_fragment_program_compiler
* c
, rc_compare_func func
)
185 case RC_COMPARE_FUNC_EQUAL
: return R500_INST_ALU_RESULT_OP_EQ
;
186 case RC_COMPARE_FUNC_LESS
: return R500_INST_ALU_RESULT_OP_LT
;
187 case RC_COMPARE_FUNC_GEQUAL
: return R500_INST_ALU_RESULT_OP_GE
;
188 case RC_COMPARE_FUNC_NOTEQUAL
: return R500_INST_ALU_RESULT_OP_NE
;
190 rc_error(&c
->Base
, "%s: unsupported compare func %i\n", __FUNCTION__
, func
);
195 static void use_temporary(struct r500_fragment_program_code
* code
, unsigned int index
)
197 if (index
> code
->max_temp_idx
)
198 code
->max_temp_idx
= index
;
201 static unsigned int use_source(struct r500_fragment_program_code
* code
, struct rc_pair_instruction_source src
)
203 if (src
.File
== RC_FILE_CONSTANT
) {
204 return src
.Index
| 0x100;
205 } else if (src
.File
== RC_FILE_TEMPORARY
) {
206 use_temporary(code
, src
.Index
);
214 * NOP the specified instruction if it is not a texture lookup.
216 static void alu_nop(struct r300_fragment_program_compiler
*c
, int ip
)
220 if ((code
->inst
[ip
].inst0
& 0x3) != R500_INST_TYPE_TEX
) {
221 code
->inst
[ip
].inst0
|= R500_INST_NOP
;
226 * Emit a paired ALU instruction.
228 static void emit_paired(struct r300_fragment_program_compiler
*c
, struct rc_pair_instruction
*inst
)
233 if (code
->inst_end
>= c
->Base
.max_alu_insts
-1) {
234 error("emit_alu: Too many instructions");
238 ip
= ++code
->inst_end
;
240 /* Quirk: MDH/MDV (DDX/DDY) need a NOP on previous non-TEX instructions. */
241 if (inst
->RGB
.Opcode
== RC_OPCODE_DDX
|| inst
->Alpha
.Opcode
== RC_OPCODE_DDX
||
242 inst
->RGB
.Opcode
== RC_OPCODE_DDY
|| inst
->Alpha
.Opcode
== RC_OPCODE_DDY
) {
248 code
->inst
[ip
].inst5
= translate_rgb_op(c
, inst
->RGB
.Opcode
);
249 code
->inst
[ip
].inst4
= translate_alpha_op(c
, inst
->Alpha
.Opcode
);
251 if (inst
->RGB
.OutputWriteMask
|| inst
->Alpha
.OutputWriteMask
|| inst
->Alpha
.DepthWriteMask
) {
252 code
->inst
[ip
].inst0
= R500_INST_TYPE_OUT
;
253 if (inst
->WriteALUResult
) {
254 error("Cannot write output and ALU result at the same time");
258 code
->inst
[ip
].inst0
= R500_INST_TYPE_ALU
;
260 code
->inst
[ip
].inst0
|= R500_INST_TEX_SEM_WAIT
;
262 code
->inst
[ip
].inst0
|= (inst
->RGB
.WriteMask
<< 11) | (inst
->Alpha
.WriteMask
<< 14);
263 code
->inst
[ip
].inst0
|= (inst
->RGB
.OutputWriteMask
<< 15) | (inst
->Alpha
.OutputWriteMask
<< 18);
265 code
->inst
[ip
].inst0
|= R500_INST_NOP
;
267 if (inst
->Alpha
.DepthWriteMask
) {
268 code
->inst
[ip
].inst4
|= R500_ALPHA_W_OMASK
;
269 c
->code
->writes_depth
= 1;
272 code
->inst
[ip
].inst4
|= R500_ALPHA_ADDRD(inst
->Alpha
.DestIndex
);
273 code
->inst
[ip
].inst5
|= R500_ALU_RGBA_ADDRD(inst
->RGB
.DestIndex
);
274 use_temporary(code
, inst
->Alpha
.DestIndex
);
275 use_temporary(code
, inst
->RGB
.DestIndex
);
277 if (inst
->RGB
.Saturate
)
278 code
->inst
[ip
].inst0
|= R500_INST_RGB_CLAMP
;
279 if (inst
->Alpha
.Saturate
)
280 code
->inst
[ip
].inst0
|= R500_INST_ALPHA_CLAMP
;
282 /* Set the presubtract operation. */
283 switch(inst
->RGB
.Src
[RC_PAIR_PRESUB_SRC
].Index
) {
285 code
->inst
[ip
].inst1
|= R500_RGB_SRCP_OP_1_MINUS_2RGB0
;
288 code
->inst
[ip
].inst1
|= R500_RGB_SRCP_OP_RGB1_MINUS_RGB0
;
291 code
->inst
[ip
].inst1
|= R500_RGB_SRCP_OP_RGB1_PLUS_RGB0
;
294 code
->inst
[ip
].inst1
|= R500_RGB_SRCP_OP_1_MINUS_RGB0
;
299 switch(inst
->Alpha
.Src
[RC_PAIR_PRESUB_SRC
].Index
) {
301 code
->inst
[ip
].inst2
|= R500_ALPHA_SRCP_OP_1_MINUS_2A0
;
304 code
->inst
[ip
].inst2
|= R500_ALPHA_SRCP_OP_A1_MINUS_A0
;
307 code
->inst
[ip
].inst2
|= R500_ALPHA_SRCP_OP_A1_PLUS_A0
;
310 code
->inst
[ip
].inst2
|= R500_ALPHA_SRCP_OP_1_MINUS_A0
;
316 code
->inst
[ip
].inst1
|= R500_RGB_ADDR0(use_source(code
, inst
->RGB
.Src
[0]));
317 code
->inst
[ip
].inst1
|= R500_RGB_ADDR1(use_source(code
, inst
->RGB
.Src
[1]));
318 code
->inst
[ip
].inst1
|= R500_RGB_ADDR2(use_source(code
, inst
->RGB
.Src
[2]));
320 code
->inst
[ip
].inst2
|= R500_ALPHA_ADDR0(use_source(code
, inst
->Alpha
.Src
[0]));
321 code
->inst
[ip
].inst2
|= R500_ALPHA_ADDR1(use_source(code
, inst
->Alpha
.Src
[1]));
322 code
->inst
[ip
].inst2
|= R500_ALPHA_ADDR2(use_source(code
, inst
->Alpha
.Src
[2]));
324 code
->inst
[ip
].inst3
|= translate_arg_rgb(inst
, 0) << R500_ALU_RGB_SEL_A_SHIFT
;
325 code
->inst
[ip
].inst3
|= translate_arg_rgb(inst
, 1) << R500_ALU_RGB_SEL_B_SHIFT
;
326 code
->inst
[ip
].inst5
|= translate_arg_rgb(inst
, 2) << R500_ALU_RGBA_SEL_C_SHIFT
;
328 code
->inst
[ip
].inst4
|= translate_arg_alpha(inst
, 0) << R500_ALPHA_SEL_A_SHIFT
;
329 code
->inst
[ip
].inst4
|= translate_arg_alpha(inst
, 1) << R500_ALPHA_SEL_B_SHIFT
;
330 code
->inst
[ip
].inst5
|= translate_arg_alpha(inst
, 2) << R500_ALU_RGBA_ALPHA_SEL_C_SHIFT
;
332 code
->inst
[ip
].inst3
|= R500_ALU_RGB_TARGET(inst
->RGB
.Target
);
333 code
->inst
[ip
].inst4
|= R500_ALPHA_TARGET(inst
->Alpha
.Target
);
335 if (inst
->WriteALUResult
) {
336 code
->inst
[ip
].inst3
|= R500_ALU_RGB_WMASK
;
338 if (inst
->WriteALUResult
== RC_ALURESULT_X
)
339 code
->inst
[ip
].inst0
|= R500_INST_ALU_RESULT_SEL_RED
;
341 code
->inst
[ip
].inst0
|= R500_INST_ALU_RESULT_SEL_ALPHA
;
343 code
->inst
[ip
].inst0
|= translate_alu_result_op(c
, inst
->ALUResultCompare
);
347 static unsigned int translate_strq_swizzle(unsigned int swizzle
)
349 unsigned int swiz
= 0;
351 for (i
= 0; i
< 4; i
++)
352 swiz
|= (GET_SWZ(swizzle
, i
) & 0x3) << i
*2;
357 * Emit a single TEX instruction
359 static int emit_tex(struct r300_fragment_program_compiler
*c
, struct rc_sub_instruction
*inst
)
364 if (code
->inst_end
>= c
->Base
.max_alu_insts
-1) {
365 error("emit_tex: Too many instructions");
369 ip
= ++code
->inst_end
;
371 code
->inst
[ip
].inst0
= R500_INST_TYPE_TEX
372 | (inst
->DstReg
.WriteMask
<< 11)
373 | R500_INST_TEX_SEM_WAIT
;
374 code
->inst
[ip
].inst1
= R500_TEX_ID(inst
->TexSrcUnit
)
375 | R500_TEX_SEM_ACQUIRE
| R500_TEX_IGNORE_UNCOVERED
;
377 if (inst
->TexSrcTarget
== RC_TEXTURE_RECT
)
378 code
->inst
[ip
].inst1
|= R500_TEX_UNSCALED
;
380 switch (inst
->Opcode
) {
382 code
->inst
[ip
].inst1
|= R500_TEX_INST_TEXKILL
;
385 code
->inst
[ip
].inst1
|= R500_TEX_INST_LD
;
388 code
->inst
[ip
].inst1
|= R500_TEX_INST_LODBIAS
;
391 code
->inst
[ip
].inst1
|= R500_TEX_INST_PROJ
;
394 error("emit_tex can't handle opcode %s\n", rc_get_opcode_info(inst
->Opcode
)->Name
);
397 use_temporary(code
, inst
->SrcReg
[0].Index
);
398 if (inst
->Opcode
!= RC_OPCODE_KIL
)
399 use_temporary(code
, inst
->DstReg
.Index
);
401 code
->inst
[ip
].inst2
= R500_TEX_SRC_ADDR(inst
->SrcReg
[0].Index
)
402 | (translate_strq_swizzle(inst
->SrcReg
[0].Swizzle
) << 8)
403 | R500_TEX_DST_ADDR(inst
->DstReg
.Index
)
404 | R500_TEX_DST_R_SWIZ_R
| R500_TEX_DST_G_SWIZ_G
405 | R500_TEX_DST_B_SWIZ_B
| R500_TEX_DST_A_SWIZ_A
;
410 static void emit_flowcontrol(struct emit_state
* s
, struct rc_instruction
* inst
)
414 if (s
->Code
->inst_end
>= s
->C
->max_alu_insts
-1) {
415 rc_error(s
->C
, "emit_tex: Too many instructions");
419 newip
= ++s
->Code
->inst_end
;
421 /* Currently all loops use the same integer constant to intialize
422 * the loop variables. */
423 if(!s
->Code
->int_constants
[0]) {
424 s
->Code
->int_constants
[0] = R500_FC_INT_CONST_KR(0xff);
425 s
->Code
->int_constant_count
= 1;
427 s
->Code
->inst
[newip
].inst0
= R500_INST_TYPE_FC
| R500_INST_ALU_WAIT
;
429 switch(inst
->U
.I
.Opcode
){
430 struct branch_info
* branch
;
431 struct r500_loop_info
* loop
;
432 case RC_OPCODE_BGNLOOP
:
433 memory_pool_array_reserve(&s
->C
->Pool
, struct r500_loop_info
,
434 s
->Loops
, s
->CurrentLoopDepth
, s
->LoopsReserved
, 1);
436 loop
= &s
->Loops
[s
->CurrentLoopDepth
++];
437 memset(loop
, 0, sizeof(struct r500_loop_info
));
438 loop
->BranchDepth
= s
->CurrentBranchDepth
;
439 loop
->BgnLoop
= newip
;
441 s
->Code
->inst
[newip
].inst2
= R500_FC_OP_LOOP
442 | R500_FC_JUMP_FUNC(0x00)
443 | R500_FC_IGNORE_UNCOVERED
447 loop
= &s
->Loops
[s
->CurrentLoopDepth
- 1];
448 memory_pool_array_reserve(&s
->C
->Pool
, int, loop
->Brks
,
449 loop
->BrkCount
, loop
->BrkReserved
, 1);
451 loop
->Brks
[loop
->BrkCount
++] = newip
;
452 s
->Code
->inst
[newip
].inst2
= R500_FC_OP_BREAKLOOP
453 | R500_FC_JUMP_FUNC(0xff)
456 s
->CurrentBranchDepth
- loop
->BranchDepth
)
457 | R500_FC_IGNORE_UNCOVERED
462 loop
= &s
->Loops
[s
->CurrentLoopDepth
- 1];
463 memory_pool_array_reserve(&s
->C
->Pool
, int, loop
->Conts
,
464 loop
->ContCount
, loop
->ContReserved
, 1);
465 loop
->Conts
[loop
->ContCount
++] = newip
;
466 s
->Code
->inst
[newip
].inst2
= R500_FC_OP_CONTINUE
467 | R500_FC_JUMP_FUNC(0xff)
470 s
->CurrentBranchDepth
- loop
->BranchDepth
)
471 | R500_FC_IGNORE_UNCOVERED
475 case RC_OPCODE_ENDLOOP
:
477 loop
= &s
->Loops
[s
->CurrentLoopDepth
- 1];
479 s
->Code
->inst
[newip
].inst2
= R500_FC_OP_ENDLOOP
480 | R500_FC_JUMP_FUNC(0xff)
482 | R500_FC_IGNORE_UNCOVERED
484 /* The constant integer at index 0 is used by all loops. */
485 s
->Code
->inst
[newip
].inst3
= R500_FC_INT_ADDR(0)
486 | R500_FC_JUMP_ADDR(loop
->BgnLoop
+ 1)
489 /* Set jump address and int constant for BGNLOOP */
490 s
->Code
->inst
[loop
->BgnLoop
].inst3
= R500_FC_INT_ADDR(0)
491 | R500_FC_JUMP_ADDR(newip
)
494 /* Set jump address for the BRK instructions. */
495 while(loop
->BrkCount
--) {
496 s
->Code
->inst
[loop
->Brks
[loop
->BrkCount
]].inst3
=
497 R500_FC_JUMP_ADDR(newip
+ 1);
500 /* Set jump address for CONT instructions. */
501 while(loop
->ContCount
--) {
502 s
->Code
->inst
[loop
->Conts
[loop
->ContCount
]].inst3
=
503 R500_FC_JUMP_ADDR(newip
);
505 s
->CurrentLoopDepth
--;
509 if ( s
->CurrentBranchDepth
>= MAX_BRANCH_DEPTH_FULL
) {
510 rc_error(s
->C
, "Branch depth exceeds hardware limit");
513 memory_pool_array_reserve(&s
->C
->Pool
, struct branch_info
,
514 s
->Branches
, s
->CurrentBranchDepth
, s
->BranchesReserved
, 1);
516 branch
= &s
->Branches
[s
->CurrentBranchDepth
++];
521 if (s
->CurrentBranchDepth
> s
->MaxBranchDepth
)
522 s
->MaxBranchDepth
= s
->CurrentBranchDepth
;
524 /* actual instruction is filled in at ENDIF time */
528 if (!s
->CurrentBranchDepth
) {
529 rc_error(s
->C
, "%s: got ELSE outside a branch", __FUNCTION__
);
533 branch
= &s
->Branches
[s
->CurrentBranchDepth
- 1];
534 branch
->Else
= newip
;
536 /* actual instruction is filled in at ENDIF time */
539 case RC_OPCODE_ENDIF
:
540 if (!s
->CurrentBranchDepth
) {
541 rc_error(s
->C
, "%s: got ELSE outside a branch", __FUNCTION__
);
545 branch
= &s
->Branches
[s
->CurrentBranchDepth
- 1];
546 branch
->Endif
= newip
;
548 s
->Code
->inst
[branch
->Endif
].inst2
= R500_FC_OP_JUMP
549 | R500_FC_A_OP_NONE
/* no address stack */
550 | R500_FC_JUMP_ANY
/* docs says set this, but I don't understand why */
551 | R500_FC_B_OP0_DECR
/* decrement branch counter if stay */
552 | R500_FC_B_OP1_NONE
/* no branch counter if stay */
553 | R500_FC_B_POP_CNT(1)
555 s
->Code
->inst
[branch
->Endif
].inst3
= R500_FC_JUMP_ADDR(branch
->Endif
+ 1);
556 s
->Code
->inst
[branch
->If
].inst2
= R500_FC_OP_JUMP
557 | R500_FC_A_OP_NONE
/* no address stack */
558 | R500_FC_JUMP_FUNC(0x0f) /* jump if ALU result is false */
559 | R500_FC_B_OP0_INCR
/* increment branch counter if stay */
560 | R500_FC_IGNORE_UNCOVERED
563 if (branch
->Else
>= 0) {
564 /* increment branch counter also if jump */
565 s
->Code
->inst
[branch
->If
].inst2
|= R500_FC_B_OP1_INCR
;
566 s
->Code
->inst
[branch
->If
].inst3
= R500_FC_JUMP_ADDR(branch
->Else
+ 1);
568 s
->Code
->inst
[branch
->Else
].inst2
= R500_FC_OP_JUMP
569 | R500_FC_A_OP_NONE
/* no address stack */
570 | R500_FC_B_ELSE
/* all active pixels want to jump */
571 | R500_FC_B_OP0_NONE
/* no counter op if stay */
572 | R500_FC_B_OP1_DECR
/* decrement branch counter if jump */
573 | R500_FC_B_POP_CNT(1)
575 s
->Code
->inst
[branch
->Else
].inst3
= R500_FC_JUMP_ADDR(branch
->Endif
+ 1);
577 /* don't touch branch counter on jump */
578 s
->Code
->inst
[branch
->If
].inst2
|= R500_FC_B_OP1_NONE
;
579 s
->Code
->inst
[branch
->If
].inst3
= R500_FC_JUMP_ADDR(branch
->Endif
+ 1);
583 s
->CurrentBranchDepth
--;
586 rc_error(s
->C
, "%s: unknown opcode %s\n", __FUNCTION__
, rc_get_opcode_info(inst
->U
.I
.Opcode
)->Name
);
590 void r500BuildFragmentProgramHwCode(struct radeon_compiler
*c
, void *user
)
592 struct r300_fragment_program_compiler
*compiler
= (struct r300_fragment_program_compiler
*)c
;
594 struct r500_fragment_program_code
*code
= &compiler
->code
->code
.r500
;
596 memset(&s
, 0, sizeof(s
));
597 s
.C
= &compiler
->Base
;
600 memset(code
, 0, sizeof(*code
));
601 code
->max_temp_idx
= 1;
604 for(struct rc_instruction
* inst
= compiler
->Base
.Program
.Instructions
.Next
;
605 inst
!= &compiler
->Base
.Program
.Instructions
&& !compiler
->Base
.Error
;
607 if (inst
->Type
== RC_INSTRUCTION_NORMAL
) {
608 const struct rc_opcode_info
* opcode
= rc_get_opcode_info(inst
->U
.I
.Opcode
);
610 if (opcode
->IsFlowControl
) {
611 emit_flowcontrol(&s
, inst
);
612 } else if (inst
->U
.I
.Opcode
== RC_OPCODE_BEGIN_TEX
) {
615 emit_tex(compiler
, &inst
->U
.I
);
618 emit_paired(compiler
, &inst
->U
.P
);
622 if (code
->max_temp_idx
>= compiler
->Base
.max_temp_regs
)
623 rc_error(&compiler
->Base
, "Too many hardware temporaries used");
625 if (compiler
->Base
.Error
)
628 if (code
->inst_end
== -1 ||
629 (code
->inst
[code
->inst_end
].inst0
& R500_INST_TYPE_MASK
) != R500_INST_TYPE_OUT
) {
632 /* This may happen when dead-code elimination is disabled or
633 * when most of the fragment program logic is leading to a KIL */
634 if (code
->inst_end
>= compiler
->Base
.max_alu_insts
-1) {
635 rc_error(&compiler
->Base
, "Introducing fake OUT: Too many instructions");
639 ip
= ++code
->inst_end
;
640 code
->inst
[ip
].inst0
= R500_INST_TYPE_OUT
| R500_INST_TEX_SEM_WAIT
;
643 /* Enable full flow control mode if we are using loops or have if
644 * statements nested at least four deep. */
645 if (s
.MaxBranchDepth
>= 4 || s
.LoopsReserved
> 0) {
646 if (code
->max_temp_idx
< 1)
647 code
->max_temp_idx
= 1;
649 code
->us_fc_ctrl
|= R500_FC_FULL_FC_EN
;