2 * Copyright 2012 Christoph Bumiller
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20 * OTHER DEALINGS IN THE SOFTWARE.
23 #include "codegen/nv50_ir_target_nvc0.h"
25 // CodeEmitter for GK110 encoding of the Fermi/Kepler ISA.
29 class CodeEmitterGK110
: public CodeEmitter
32 CodeEmitterGK110(const TargetNVC0
*);
34 virtual bool emitInstruction(Instruction
*);
35 virtual uint32_t getMinEncodingSize(const Instruction
*) const;
36 virtual void prepareEmission(Function
*);
38 inline void setProgramType(Program::Type pType
) { progType
= pType
; }
41 const TargetNVC0
*targNVC0
;
43 Program::Type progType
;
45 const bool writeIssueDelays
;
48 void emitForm_21(const Instruction
*, uint32_t opc2
, uint32_t opc1
);
49 void emitForm_C(const Instruction
*, uint32_t opc
, uint8_t ctg
);
50 void emitForm_L(const Instruction
*, uint32_t opc
, uint8_t ctg
, Modifier
, int sCount
= 3);
52 void emitPredicate(const Instruction
*);
54 void setCAddress14(const ValueRef
&);
55 void setShortImmediate(const Instruction
*, const int s
);
56 void setImmediate32(const Instruction
*, const int s
, Modifier
);
57 void setSUConst16(const Instruction
*, const int s
);
59 void modNegAbsF32_3b(const Instruction
*, const int s
);
61 void emitCondCode(CondCode cc
, int pos
, uint8_t mask
);
62 void emitInterpMode(const Instruction
*);
63 void emitLoadStoreType(DataType ty
, const int pos
);
64 void emitCachingMode(CacheMode c
, const int pos
);
65 void emitSUGType(DataType
, const int pos
);
66 void emitSUCachingMode(CacheMode c
);
68 inline uint8_t getSRegEncoding(const ValueRef
&);
70 void emitRoundMode(RoundMode
, const int pos
, const int rintPos
);
71 void emitRoundModeF(RoundMode
, const int pos
);
72 void emitRoundModeI(RoundMode
, const int pos
);
74 void emitNegAbs12(const Instruction
*);
76 void emitNOP(const Instruction
*);
78 void emitLOAD(const Instruction
*);
79 void emitSTORE(const Instruction
*);
80 void emitMOV(const Instruction
*);
81 void emitATOM(const Instruction
*);
82 void emitCCTL(const Instruction
*);
84 void emitINTERP(const Instruction
*);
85 void emitAFETCH(const Instruction
*);
86 void emitPFETCH(const Instruction
*);
87 void emitVFETCH(const Instruction
*);
88 void emitEXPORT(const Instruction
*);
89 void emitOUT(const Instruction
*);
91 void emitUADD(const Instruction
*);
92 void emitFADD(const Instruction
*);
93 void emitDADD(const Instruction
*);
94 void emitIMUL(const Instruction
*);
95 void emitFMUL(const Instruction
*);
96 void emitDMUL(const Instruction
*);
97 void emitIMAD(const Instruction
*);
98 void emitISAD(const Instruction
*);
99 void emitSHLADD(const Instruction
*);
100 void emitFMAD(const Instruction
*);
101 void emitDMAD(const Instruction
*);
102 void emitMADSP(const Instruction
*i
);
104 void emitNOT(const Instruction
*);
105 void emitLogicOp(const Instruction
*, uint8_t subOp
);
106 void emitPOPC(const Instruction
*);
107 void emitINSBF(const Instruction
*);
108 void emitEXTBF(const Instruction
*);
109 void emitBFIND(const Instruction
*);
110 void emitPERMT(const Instruction
*);
111 void emitShift(const Instruction
*);
112 void emitShift64(const Instruction
*);
114 void emitSFnOp(const Instruction
*, uint8_t subOp
);
116 void emitCVT(const Instruction
*);
117 void emitMINMAX(const Instruction
*);
118 void emitPreOp(const Instruction
*);
120 void emitSET(const CmpInstruction
*);
121 void emitSLCT(const CmpInstruction
*);
122 void emitSELP(const Instruction
*);
124 void emitTEXBAR(const Instruction
*);
125 void emitTEX(const TexInstruction
*);
126 void emitTEXCSAA(const TexInstruction
*);
127 void emitTXQ(const TexInstruction
*);
129 void emitQUADOP(const Instruction
*, uint8_t qOp
, uint8_t laneMask
);
131 void emitPIXLD(const Instruction
*);
133 void emitBAR(const Instruction
*);
134 void emitMEMBAR(const Instruction
*);
136 void emitFlow(const Instruction
*);
138 void emitSHFL(const Instruction
*);
140 void emitVOTE(const Instruction
*);
142 void emitSULDGB(const TexInstruction
*);
143 void emitSUSTGx(const TexInstruction
*);
144 void emitSUCLAMPMode(uint16_t);
145 void emitSUCalc(Instruction
*);
147 void emitVSHL(const Instruction
*);
148 void emitVectorSubOp(const Instruction
*);
150 inline void defId(const ValueDef
&, const int pos
);
151 inline void srcId(const ValueRef
&, const int pos
);
152 inline void srcId(const ValueRef
*, const int pos
);
153 inline void srcId(const Instruction
*, int s
, const int pos
);
155 inline void srcAddr32(const ValueRef
&, const int pos
); // address / 4
157 inline bool isLIMM(const ValueRef
&, DataType ty
, bool mod
= false);
160 #define GK110_GPR_ZERO 255
163 if (i->src(s).mod.neg()) code[(0x##b) / 32] |= 1 << ((0x##b) % 32)
165 if (i->src(s).mod.abs()) code[(0x##b) / 32] |= 1 << ((0x##b) % 32)
167 #define NOT_(b, s) if (i->src(s).mod & Modifier(NV50_IR_MOD_NOT)) \
168 code[(0x##b) / 32] |= 1 << ((0x##b) % 32)
170 #define FTZ_(b) if (i->ftz) code[(0x##b) / 32] |= 1 << ((0x##b) % 32)
171 #define DNZ_(b) if (i->dnz) code[(0x##b) / 32] |= 1 << ((0x##b) % 32)
173 #define SAT_(b) if (i->saturate) code[(0x##b) / 32] |= 1 << ((0x##b) % 32)
175 #define RND_(b, t) emitRoundMode##t(i->rnd, 0x##b)
177 #define SDATA(a) ((a).rep()->reg.data)
178 #define DDATA(a) ((a).rep()->reg.data)
180 void CodeEmitterGK110::srcId(const ValueRef
& src
, const int pos
)
182 code
[pos
/ 32] |= (src
.get() ? SDATA(src
).id
: GK110_GPR_ZERO
) << (pos
% 32);
185 void CodeEmitterGK110::srcId(const ValueRef
*src
, const int pos
)
187 code
[pos
/ 32] |= (src
? SDATA(*src
).id
: GK110_GPR_ZERO
) << (pos
% 32);
190 void CodeEmitterGK110::srcId(const Instruction
*insn
, int s
, int pos
)
192 int r
= insn
->srcExists(s
) ? SDATA(insn
->src(s
)).id
: GK110_GPR_ZERO
;
193 code
[pos
/ 32] |= r
<< (pos
% 32);
196 void CodeEmitterGK110::srcAddr32(const ValueRef
& src
, const int pos
)
198 code
[pos
/ 32] |= (SDATA(src
).offset
>> 2) << (pos
% 32);
201 void CodeEmitterGK110::defId(const ValueDef
& def
, const int pos
)
203 code
[pos
/ 32] |= (def
.get() && def
.getFile() != FILE_FLAGS
? DDATA(def
).id
: GK110_GPR_ZERO
) << (pos
% 32);
206 bool CodeEmitterGK110::isLIMM(const ValueRef
& ref
, DataType ty
, bool mod
)
208 const ImmediateValue
*imm
= ref
.get()->asImm();
211 return imm
&& imm
->reg
.data
.u32
& 0xfff;
213 return imm
&& (imm
->reg
.data
.s32
> 0x7ffff ||
214 imm
->reg
.data
.s32
< -0x80000);
218 CodeEmitterGK110::emitRoundMode(RoundMode rnd
, const int pos
, const int rintPos
)
224 case ROUND_MI
: rint
= true; /* fall through */ case ROUND_M
: n
= 1; break;
225 case ROUND_PI
: rint
= true; /* fall through */ case ROUND_P
: n
= 2; break;
226 case ROUND_ZI
: rint
= true; /* fall through */ case ROUND_Z
: n
= 3; break;
228 rint
= rnd
== ROUND_NI
;
230 assert(rnd
== ROUND_N
|| rnd
== ROUND_NI
);
233 code
[pos
/ 32] |= n
<< (pos
% 32);
234 if (rint
&& rintPos
>= 0)
235 code
[rintPos
/ 32] |= 1 << (rintPos
% 32);
239 CodeEmitterGK110::emitRoundModeF(RoundMode rnd
, const int pos
)
244 case ROUND_M
: n
= 1; break;
245 case ROUND_P
: n
= 2; break;
246 case ROUND_Z
: n
= 3; break;
249 assert(rnd
== ROUND_N
);
252 code
[pos
/ 32] |= n
<< (pos
% 32);
256 CodeEmitterGK110::emitRoundModeI(RoundMode rnd
, const int pos
)
261 case ROUND_MI
: n
= 1; break;
262 case ROUND_PI
: n
= 2; break;
263 case ROUND_ZI
: n
= 3; break;
266 assert(rnd
== ROUND_NI
);
269 code
[pos
/ 32] |= n
<< (pos
% 32);
272 void CodeEmitterGK110::emitCondCode(CondCode cc
, int pos
, uint8_t mask
)
277 case CC_FL
: n
= 0x00; break;
278 case CC_LT
: n
= 0x01; break;
279 case CC_EQ
: n
= 0x02; break;
280 case CC_LE
: n
= 0x03; break;
281 case CC_GT
: n
= 0x04; break;
282 case CC_NE
: n
= 0x05; break;
283 case CC_GE
: n
= 0x06; break;
284 case CC_LTU
: n
= 0x09; break;
285 case CC_EQU
: n
= 0x0a; break;
286 case CC_LEU
: n
= 0x0b; break;
287 case CC_GTU
: n
= 0x0c; break;
288 case CC_NEU
: n
= 0x0d; break;
289 case CC_GEU
: n
= 0x0e; break;
290 case CC_TR
: n
= 0x0f; break;
291 case CC_NO
: n
= 0x10; break;
292 case CC_NC
: n
= 0x11; break;
293 case CC_NS
: n
= 0x12; break;
294 case CC_NA
: n
= 0x13; break;
295 case CC_A
: n
= 0x14; break;
296 case CC_S
: n
= 0x15; break;
297 case CC_C
: n
= 0x16; break;
298 case CC_O
: n
= 0x17; break;
301 assert(!"invalid condition code");
304 code
[pos
/ 32] |= (n
& mask
) << (pos
% 32);
308 CodeEmitterGK110::emitPredicate(const Instruction
*i
)
310 if (i
->predSrc
>= 0) {
311 srcId(i
->src(i
->predSrc
), 18);
312 if (i
->cc
== CC_NOT_P
)
313 code
[0] |= 8 << 18; // negate
314 assert(i
->getPredicate()->reg
.file
== FILE_PREDICATE
);
321 CodeEmitterGK110::setCAddress14(const ValueRef
& src
)
323 const Storage
& res
= src
.get()->asSym()->reg
;
324 const int32_t addr
= res
.data
.offset
/ 4;
326 code
[0] |= (addr
& 0x01ff) << 23;
327 code
[1] |= (addr
& 0x3e00) >> 9;
328 code
[1] |= res
.fileIndex
<< 5;
332 CodeEmitterGK110::setShortImmediate(const Instruction
*i
, const int s
)
334 const uint32_t u32
= i
->getSrc(s
)->asImm()->reg
.data
.u32
;
335 const uint64_t u64
= i
->getSrc(s
)->asImm()->reg
.data
.u64
;
337 if (i
->sType
== TYPE_F32
) {
338 assert(!(u32
& 0x00000fff));
339 code
[0] |= ((u32
& 0x001ff000) >> 12) << 23;
340 code
[1] |= ((u32
& 0x7fe00000) >> 21);
341 code
[1] |= ((u32
& 0x80000000) >> 4);
343 if (i
->sType
== TYPE_F64
) {
344 assert(!(u64
& 0x00000fffffffffffULL
));
345 code
[0] |= ((u64
& 0x001ff00000000000ULL
) >> 44) << 23;
346 code
[1] |= ((u64
& 0x7fe0000000000000ULL
) >> 53);
347 code
[1] |= ((u64
& 0x8000000000000000ULL
) >> 36);
349 assert((u32
& 0xfff80000) == 0 || (u32
& 0xfff80000) == 0xfff80000);
350 code
[0] |= (u32
& 0x001ff) << 23;
351 code
[1] |= (u32
& 0x7fe00) >> 9;
352 code
[1] |= (u32
& 0x80000) << 8;
357 CodeEmitterGK110::setImmediate32(const Instruction
*i
, const int s
,
360 uint32_t u32
= i
->getSrc(s
)->asImm()->reg
.data
.u32
;
363 ImmediateValue
imm(i
->getSrc(s
)->asImm(), i
->sType
);
365 u32
= imm
.reg
.data
.u32
;
368 code
[0] |= u32
<< 23;
373 CodeEmitterGK110::emitForm_L(const Instruction
*i
, uint32_t opc
, uint8_t ctg
,
374 Modifier mod
, int sCount
)
383 for (int s
= 0; s
< sCount
&& i
->srcExists(s
); ++s
) {
384 switch (i
->src(s
).getFile()) {
386 srcId(i
->src(s
), s
? 42 : 10);
389 setImmediate32(i
, s
, mod
);
399 CodeEmitterGK110::emitForm_C(const Instruction
*i
, uint32_t opc
, uint8_t ctg
)
408 switch (i
->src(0).getFile()) {
409 case FILE_MEMORY_CONST
:
410 code
[1] |= 0x4 << 28;
411 setCAddress14(i
->src(0));
414 code
[1] |= 0xc << 28;
415 srcId(i
->src(0), 23);
423 // 0x2 for GPR, c[] and 0x1 for short immediate
425 CodeEmitterGK110::emitForm_21(const Instruction
*i
, uint32_t opc2
,
428 const bool imm
= i
->srcExists(1) && i
->src(1).getFile() == FILE_IMMEDIATE
;
431 if (i
->srcExists(2) && i
->src(2).getFile() == FILE_MEMORY_CONST
)
436 code
[1] = opc1
<< 20;
439 code
[1] = (0xc << 28) | (opc2
<< 20);
446 for (int s
= 0; s
< 3 && i
->srcExists(s
); ++s
) {
447 switch (i
->src(s
).getFile()) {
448 case FILE_MEMORY_CONST
:
449 code
[1] &= (s
== 2) ? ~(0x4 << 28) : ~(0x8 << 28);
450 setCAddress14(i
->src(s
));
453 setShortImmediate(i
, s
);
456 srcId(i
->src(s
), s
? ((s
== 2) ? 42 : s1
) : 10);
459 if (i
->op
== OP_SELP
) {
460 assert(s
== 2 && i
->src(s
).getFile() == FILE_PREDICATE
);
461 srcId(i
->src(s
), 42);
463 // ignore here, can be predicate or flags, but must not be address
471 assert(imm
|| (code
[1] & (0xc << 28)));
475 CodeEmitterGK110::modNegAbsF32_3b(const Instruction
*i
, const int s
)
477 if (i
->src(s
).mod
.abs()) code
[1] &= ~(1 << 27);
478 if (i
->src(s
).mod
.neg()) code
[1] ^= (1 << 27);
482 CodeEmitterGK110::emitNOP(const Instruction
*i
)
484 code
[0] = 0x00003c02;
485 code
[1] = 0x85800000;
490 code
[0] = 0x001c3c02;
494 CodeEmitterGK110::emitFMAD(const Instruction
*i
)
496 bool neg1
= (i
->src(0).mod
^ i
->src(1).mod
).neg();
498 if (isLIMM(i
->src(1), TYPE_F32
)) {
499 assert(i
->getDef(0)->reg
.data
.id
== i
->getSrc(2)->reg
.data
.id
);
501 // last source is dst, so force 2 sources
502 emitForm_L(i
, 0x600, 0x0, 0, 2);
504 if (i
->flagsDef
>= 0)
514 emitForm_21(i
, 0x0c0, 0x940);
534 CodeEmitterGK110::emitDMAD(const Instruction
*i
)
536 assert(!i
->saturate
);
539 emitForm_21(i
, 0x1b8, 0xb38);
544 bool neg1
= (i
->src(0).mod
^ i
->src(1).mod
).neg();
556 CodeEmitterGK110::emitMADSP(const Instruction
*i
)
558 emitForm_21(i
, 0x140, 0xa40);
560 if (i
->subOp
== NV50_IR_SUBOP_MADSP_SD
) {
561 code
[1] |= 0x00c00000;
563 code
[1] |= (i
->subOp
& 0x00f) << 19; // imadp1
564 code
[1] |= (i
->subOp
& 0x0f0) << 20; // imadp2
565 code
[1] |= (i
->subOp
& 0x100) << 11; // imadp3
566 code
[1] |= (i
->subOp
& 0x200) << 15; // imadp3
567 code
[1] |= (i
->subOp
& 0xc00) << 12; // imadp3
570 if (i
->flagsDef
>= 0)
575 CodeEmitterGK110::emitFMUL(const Instruction
*i
)
577 bool neg
= (i
->src(0).mod
^ i
->src(1).mod
).neg();
579 assert(i
->postFactor
>= -3 && i
->postFactor
<= 3);
581 if (isLIMM(i
->src(1), TYPE_F32
)) {
582 emitForm_L(i
, 0x200, 0x2, Modifier(0));
590 assert(i
->postFactor
== 0);
592 emitForm_21(i
, 0x234, 0xc34);
593 code
[1] |= ((i
->postFactor
> 0) ?
594 (7 - i
->postFactor
) : (0 - i
->postFactor
)) << 12;
612 CodeEmitterGK110::emitDMUL(const Instruction
*i
)
614 bool neg
= (i
->src(0).mod
^ i
->src(1).mod
).neg();
616 assert(!i
->postFactor
);
617 assert(!i
->saturate
);
621 emitForm_21(i
, 0x240, 0xc40);
635 CodeEmitterGK110::emitIMUL(const Instruction
*i
)
637 assert(!i
->src(0).mod
.neg() && !i
->src(1).mod
.neg());
638 assert(!i
->src(0).mod
.abs() && !i
->src(1).mod
.abs());
640 if (isLIMM(i
->src(1), TYPE_S32
)) {
641 emitForm_L(i
, 0x280, 2, Modifier(0));
643 if (i
->subOp
== NV50_IR_SUBOP_MUL_HIGH
)
645 if (i
->sType
== TYPE_S32
)
648 emitForm_21(i
, 0x21c, 0xc1c);
650 if (i
->subOp
== NV50_IR_SUBOP_MUL_HIGH
)
652 if (i
->sType
== TYPE_S32
)
658 CodeEmitterGK110::emitFADD(const Instruction
*i
)
660 if (isLIMM(i
->src(1), TYPE_F32
)) {
661 assert(i
->rnd
== ROUND_N
);
662 assert(!i
->saturate
);
664 Modifier mod
= i
->src(1).mod
^
665 Modifier(i
->op
== OP_SUB
? NV50_IR_MOD_NEG
: 0);
667 emitForm_L(i
, 0x400, 0, mod
);
673 emitForm_21(i
, 0x22c, 0xc2c);
682 modNegAbsF32_3b(i
, 1);
683 if (i
->op
== OP_SUB
) code
[1] ^= 1 << 27;
687 if (i
->op
== OP_SUB
) code
[1] ^= 1 << 16;
693 CodeEmitterGK110::emitDADD(const Instruction
*i
)
695 assert(!i
->saturate
);
698 emitForm_21(i
, 0x238, 0xc38);
703 modNegAbsF32_3b(i
, 1);
704 if (i
->op
== OP_SUB
) code
[1] ^= 1 << 27;
708 if (i
->op
== OP_SUB
) code
[1] ^= 1 << 16;
713 CodeEmitterGK110::emitUADD(const Instruction
*i
)
715 uint8_t addOp
= (i
->src(0).mod
.neg() << 1) | i
->src(1).mod
.neg();
720 assert(!i
->src(0).mod
.abs() && !i
->src(1).mod
.abs());
722 if (isLIMM(i
->src(1), TYPE_S32
)) {
723 emitForm_L(i
, 0x400, 1, Modifier((addOp
& 1) ? NV50_IR_MOD_NEG
: 0));
728 assert(i
->flagsDef
< 0);
729 assert(i
->flagsSrc
< 0);
733 emitForm_21(i
, 0x208, 0xc08);
735 assert(addOp
!= 3); // would be add-plus-one
737 code
[1] |= addOp
<< 19;
739 if (i
->flagsDef
>= 0)
740 code
[1] |= 1 << 18; // write carry
741 if (i
->flagsSrc
>= 0)
742 code
[1] |= 1 << 14; // add carry
749 CodeEmitterGK110::emitIMAD(const Instruction
*i
)
752 i
->src(2).mod
.neg() | ((i
->src(0).mod
.neg() ^ i
->src(1).mod
.neg()) << 1);
754 emitForm_21(i
, 0x100, 0xa00);
757 code
[1] |= addOp
<< 26;
759 if (i
->sType
== TYPE_S32
)
760 code
[1] |= (1 << 19) | (1 << 24);
762 if (i
->subOp
== NV50_IR_SUBOP_MUL_HIGH
)
765 if (i
->flagsDef
>= 0) code
[1] |= 1 << 18;
766 if (i
->flagsSrc
>= 0) code
[1] |= 1 << 20;
772 CodeEmitterGK110::emitISAD(const Instruction
*i
)
774 assert(i
->dType
== TYPE_S32
|| i
->dType
== TYPE_U32
);
776 emitForm_21(i
, 0x1f4, 0xb74);
778 if (i
->dType
== TYPE_S32
)
783 CodeEmitterGK110::emitSHLADD(const Instruction
*i
)
785 uint8_t addOp
= (i
->src(0).mod
.neg() << 1) | i
->src(2).mod
.neg();
786 const ImmediateValue
*imm
= i
->src(1).get()->asImm();
789 if (i
->src(2).getFile() == FILE_IMMEDIATE
) {
791 code
[1] = 0xc0c << 20;
794 code
[1] = 0x20c << 20;
796 code
[1] |= addOp
<< 19;
801 srcId(i
->src(0), 10);
803 if (i
->flagsDef
>= 0)
806 assert(!(imm
->reg
.data
.u32
& 0xffffffe0));
807 code
[1] |= imm
->reg
.data
.u32
<< 10;
809 switch (i
->src(2).getFile()) {
811 assert(code
[0] & 0x2);
812 code
[1] |= 0xc << 28;
813 srcId(i
->src(2), 23);
815 case FILE_MEMORY_CONST
:
816 assert(code
[0] & 0x2);
817 code
[1] |= 0x4 << 28;
818 setCAddress14(i
->src(2));
821 assert(code
[0] & 0x1);
822 setShortImmediate(i
, 2);
825 assert(!"bad src2 file");
831 CodeEmitterGK110::emitNOT(const Instruction
*i
)
833 code
[0] = 0x0003fc02; // logop(mov2) dst, 0, not src
834 code
[1] = 0x22003800;
840 switch (i
->src(0).getFile()) {
842 code
[1] |= 0xc << 28;
843 srcId(i
->src(0), 23);
845 case FILE_MEMORY_CONST
:
846 code
[1] |= 0x4 << 28;
847 setCAddress14(i
->src(0));
856 CodeEmitterGK110::emitLogicOp(const Instruction
*i
, uint8_t subOp
)
858 if (i
->def(0).getFile() == FILE_PREDICATE
) {
859 code
[0] = 0x00000002 | (subOp
<< 27);
860 code
[1] = 0x84800000;
865 srcId(i
->src(0), 14);
866 if (i
->src(0).mod
== Modifier(NV50_IR_MOD_NOT
)) code
[0] |= 1 << 17;
867 srcId(i
->src(1), 32);
868 if (i
->src(1).mod
== Modifier(NV50_IR_MOD_NOT
)) code
[1] |= 1 << 3;
870 if (i
->defExists(1)) {
876 if (i
->predSrc
!= 2 && i
->srcExists(2)) {
877 code
[1] |= subOp
<< 16;
878 srcId(i
->src(2), 42);
879 if (i
->src(2).mod
== Modifier(NV50_IR_MOD_NOT
)) code
[1] |= 1 << 13;
884 if (isLIMM(i
->src(1), TYPE_S32
)) {
885 emitForm_L(i
, 0x200, 0, i
->src(1).mod
);
886 code
[1] |= subOp
<< 24;
889 emitForm_21(i
, 0x220, 0xc20);
890 code
[1] |= subOp
<< 12;
897 CodeEmitterGK110::emitPOPC(const Instruction
*i
)
899 assert(!isLIMM(i
->src(1), TYPE_S32
, true));
901 emitForm_21(i
, 0x204, 0xc04);
904 if (!(code
[0] & 0x1))
909 CodeEmitterGK110::emitINSBF(const Instruction
*i
)
911 emitForm_21(i
, 0x1f8, 0xb78);
915 CodeEmitterGK110::emitEXTBF(const Instruction
*i
)
917 emitForm_21(i
, 0x600, 0xc00);
919 if (i
->dType
== TYPE_S32
)
921 if (i
->subOp
== NV50_IR_SUBOP_EXTBF_REV
)
926 CodeEmitterGK110::emitBFIND(const Instruction
*i
)
928 emitForm_C(i
, 0x218, 0x2);
930 if (i
->dType
== TYPE_S32
)
932 if (i
->src(0).mod
== Modifier(NV50_IR_MOD_NOT
))
934 if (i
->subOp
== NV50_IR_SUBOP_BFIND_SAMT
)
939 CodeEmitterGK110::emitPERMT(const Instruction
*i
)
941 emitForm_21(i
, 0x1e0, 0xb60);
943 code
[1] |= i
->subOp
<< 19;
947 CodeEmitterGK110::emitShift(const Instruction
*i
)
949 if (i
->op
== OP_SHR
) {
950 emitForm_21(i
, 0x214, 0xc14);
951 if (isSignedType(i
->dType
))
954 emitForm_21(i
, 0x224, 0xc24);
957 if (i
->subOp
== NV50_IR_SUBOP_SHIFT_WRAP
)
962 CodeEmitterGK110::emitShift64(const Instruction
*i
)
964 if (i
->op
== OP_SHR
) {
965 emitForm_21(i
, 0x27c, 0xc7c);
966 if (isSignedType(i
->sType
))
968 if (i
->subOp
& NV50_IR_SUBOP_SHIFT_HIGH
)
971 emitForm_21(i
, 0xdfc, 0xf7c);
975 if (i
->subOp
& NV50_IR_SUBOP_SHIFT_WRAP
)
980 CodeEmitterGK110::emitPreOp(const Instruction
*i
)
982 emitForm_C(i
, 0x248, 0x2);
984 if (i
->op
== OP_PREEX2
)
992 CodeEmitterGK110::emitSFnOp(const Instruction
*i
, uint8_t subOp
)
994 code
[0] = 0x00000002 | (subOp
<< 23);
995 code
[1] = 0x84000000;
1000 srcId(i
->src(0), 10);
1008 CodeEmitterGK110::emitMINMAX(const Instruction
*i
)
1032 emitForm_21(i
, op2
, op1
);
1034 if (i
->dType
== TYPE_S32
)
1036 code
[1] |= (i
->op
== OP_MIN
) ? 0x1c00 : 0x3c00; // [!]pt
1037 code
[1] |= i
->subOp
<< 14;
1038 if (i
->flagsDef
>= 0)
1039 code
[1] |= i
->subOp
<< 18;
1044 if (code
[0] & 0x1) {
1045 modNegAbsF32_3b(i
, 1);
1053 CodeEmitterGK110::emitCVT(const Instruction
*i
)
1055 const bool f2f
= isFloatType(i
->dType
) && isFloatType(i
->sType
);
1056 const bool f2i
= !isFloatType(i
->dType
) && isFloatType(i
->sType
);
1057 const bool i2f
= isFloatType(i
->dType
) && !isFloatType(i
->sType
);
1059 bool sat
= i
->saturate
;
1060 bool abs
= i
->src(0).mod
.abs();
1061 bool neg
= i
->src(0).mod
.neg();
1063 RoundMode rnd
= i
->rnd
;
1066 case OP_CEIL
: rnd
= f2f
? ROUND_PI
: ROUND_P
; break;
1067 case OP_FLOOR
: rnd
= f2f
? ROUND_MI
: ROUND_M
; break;
1068 case OP_TRUNC
: rnd
= f2f
? ROUND_ZI
: ROUND_Z
; break;
1069 case OP_SAT
: sat
= true; break;
1070 case OP_NEG
: neg
= !neg
; break;
1071 case OP_ABS
: abs
= true; neg
= false; break;
1078 if (i
->op
== OP_NEG
&& i
->dType
== TYPE_U32
)
1086 if (f2f
) op
= 0x254;
1087 else if (f2i
) op
= 0x258;
1088 else if (i2f
) op
= 0x25c;
1091 emitForm_C(i
, op
, 0x2);
1094 if (neg
) code
[1] |= 1 << 16;
1095 if (abs
) code
[1] |= 1 << 20;
1096 if (sat
) code
[1] |= 1 << 21;
1098 emitRoundMode(rnd
, 32 + 10, f2f
? (32 + 13) : -1);
1100 code
[0] |= typeSizeofLog2(dType
) << 10;
1101 code
[0] |= typeSizeofLog2(i
->sType
) << 12;
1102 code
[1] |= i
->subOp
<< 12;
1104 if (isSignedIntType(dType
))
1106 if (isSignedIntType(i
->sType
))
1111 CodeEmitterGK110::emitSET(const CmpInstruction
*i
)
1115 if (i
->def(0).getFile() == FILE_PREDICATE
) {
1117 case TYPE_F32
: op2
= 0x1d8; op1
= 0xb58; break;
1118 case TYPE_F64
: op2
= 0x1c0; op1
= 0xb40; break;
1124 emitForm_21(i
, op2
, op1
);
1128 if (!(code
[0] & 0x1)) {
1132 modNegAbsF32_3b(i
, 1);
1136 // normal DST field is negated predicate result
1137 code
[0] = (code
[0] & ~0xfc) | ((code
[0] << 3) & 0xe0);
1138 if (i
->defExists(1))
1139 defId(i
->def(1), 2);
1144 case TYPE_F32
: op2
= 0x000; op1
= 0x800; break;
1145 case TYPE_F64
: op2
= 0x080; op1
= 0x900; break;
1151 emitForm_21(i
, op2
, op1
);
1155 if (!(code
[0] & 0x1)) {
1159 modNegAbsF32_3b(i
, 1);
1163 if (i
->dType
== TYPE_F32
) {
1164 if (isFloatType(i
->sType
))
1170 if (i
->sType
== TYPE_S32
)
1173 if (i
->op
!= OP_SET
) {
1175 case OP_SET_AND
: code
[1] |= 0x0 << 16; break;
1176 case OP_SET_OR
: code
[1] |= 0x1 << 16; break;
1177 case OP_SET_XOR
: code
[1] |= 0x2 << 16; break;
1182 srcId(i
->src(2), 0x2a);
1184 code
[1] |= 0x7 << 10;
1186 if (i
->flagsSrc
>= 0)
1188 emitCondCode(i
->setCond
,
1189 isFloatType(i
->sType
) ? 0x33 : 0x34,
1190 isFloatType(i
->sType
) ? 0xf : 0x7);
1194 CodeEmitterGK110::emitSLCT(const CmpInstruction
*i
)
1196 CondCode cc
= i
->setCond
;
1197 if (i
->src(2).mod
.neg())
1198 cc
= reverseCondCode(cc
);
1200 if (i
->dType
== TYPE_F32
) {
1201 emitForm_21(i
, 0x1d0, 0xb50);
1203 emitCondCode(cc
, 0x33, 0xf);
1205 emitForm_21(i
, 0x1a0, 0xb20);
1206 emitCondCode(cc
, 0x34, 0x7);
1207 if (i
->dType
== TYPE_S32
)
1213 selpFlip(const FixupEntry
*entry
, uint32_t *code
, const FixupData
& data
)
1215 int loc
= entry
->loc
;
1216 if (data
.force_persample_interp
)
1217 code
[loc
+ 1] |= 1 << 13;
1219 code
[loc
+ 1] &= ~(1 << 13);
1222 void CodeEmitterGK110::emitSELP(const Instruction
*i
)
1224 emitForm_21(i
, 0x250, 0x050);
1226 if (i
->src(2).mod
& Modifier(NV50_IR_MOD_NOT
))
1229 if (i
->subOp
== 1) {
1230 addInterp(0, 0, selpFlip
);
1234 void CodeEmitterGK110::emitTEXBAR(const Instruction
*i
)
1236 code
[0] = 0x0000003e | (i
->subOp
<< 23);
1237 code
[1] = 0x77000000;
1242 void CodeEmitterGK110::emitTEXCSAA(const TexInstruction
*i
)
1244 code
[0] = 0x00000002;
1245 code
[1] = 0x76c00000;
1247 code
[1] |= i
->tex
.r
<< 9;
1248 // code[1] |= i->tex.s << (9 + 8);
1250 if (i
->tex
.liveOnly
)
1251 code
[0] |= 0x80000000;
1253 defId(i
->def(0), 2);
1254 srcId(i
->src(0), 10);
1258 isNextIndependentTex(const TexInstruction
*i
)
1260 if (!i
->next
|| !isTextureOp(i
->next
->op
))
1262 if (i
->getDef(0)->interfers(i
->next
->getSrc(0)))
1264 return !i
->next
->srcExists(1) || !i
->getDef(0)->interfers(i
->next
->getSrc(1));
1268 CodeEmitterGK110::emitTEX(const TexInstruction
*i
)
1270 const bool ind
= i
->tex
.rIndirectSrc
>= 0;
1273 code
[0] = 0x00000002;
1276 code
[1] = 0x7e000000;
1279 code
[1] = 0x7e800000;
1282 code
[1] = 0x78000000;
1285 code
[1] = 0x7dc00000;
1288 code
[1] = 0x7d800000;
1294 code
[0] = 0x00000002;
1295 code
[1] = 0x76000000;
1296 code
[1] |= i
->tex
.r
<< 9;
1299 code
[0] = 0x00000002;
1300 code
[1] = 0x76800000;
1301 code
[1] |= i
->tex
.r
<< 9;
1304 code
[0] = 0x00000002;
1305 code
[1] = 0x70000000;
1306 code
[1] |= i
->tex
.r
<< 13;
1309 code
[0] = 0x00000001;
1310 code
[1] = 0x70000000;
1311 code
[1] |= i
->tex
.r
<< 15;
1314 code
[0] = 0x00000001;
1315 code
[1] = 0x60000000;
1316 code
[1] |= i
->tex
.r
<< 15;
1321 code
[1] |= isNextIndependentTex(i
) ? 0x1 : 0x2; // t : p mode
1323 if (i
->tex
.liveOnly
)
1324 code
[0] |= 0x80000000;
1328 case OP_TXB
: code
[1] |= 0x2000; break;
1329 case OP_TXL
: code
[1] |= 0x3000; break;
1333 case OP_TXLQ
: break;
1335 assert(!"invalid texture op");
1339 if (i
->op
== OP_TXF
) {
1340 if (!i
->tex
.levelZero
)
1343 if (i
->tex
.levelZero
) {
1347 if (i
->op
!= OP_TXD
&& i
->tex
.derivAll
)
1352 code
[1] |= i
->tex
.mask
<< 2;
1354 const int src1
= (i
->predSrc
== 1) ? 2 : 1; // if predSrc == 1, !srcExists(2)
1356 defId(i
->def(0), 2);
1357 srcId(i
->src(0), 10);
1360 if (i
->op
== OP_TXG
) code
[1] |= i
->tex
.gatherComp
<< 13;
1363 code
[1] |= (i
->tex
.target
.isCube() ? 3 : (i
->tex
.target
.getDim() - 1)) << 7;
1364 if (i
->tex
.target
.isArray())
1366 if (i
->tex
.target
.isShadow())
1368 if (i
->tex
.target
== TEX_TARGET_2D_MS
||
1369 i
->tex
.target
== TEX_TARGET_2D_MS_ARRAY
)
1372 if (i
->srcExists(src1
) && i
->src(src1
).getFile() == FILE_IMMEDIATE
) {
1376 if (i
->tex
.useOffsets
== 1) {
1378 case OP_TXF
: code
[1] |= 0x200; break;
1379 case OP_TXD
: code
[1] |= 0x00400000; break;
1380 default: code
[1] |= 0x800; break;
1383 if (i
->tex
.useOffsets
== 4)
1388 CodeEmitterGK110::emitTXQ(const TexInstruction
*i
)
1390 code
[0] = 0x00000002;
1391 code
[1] = 0x75400001;
1393 switch (i
->tex
.query
) {
1394 case TXQ_DIMS
: code
[0] |= 0x01 << 25; break;
1395 case TXQ_TYPE
: code
[0] |= 0x02 << 25; break;
1396 case TXQ_SAMPLE_POSITION
: code
[0] |= 0x05 << 25; break;
1397 case TXQ_FILTER
: code
[0] |= 0x10 << 25; break;
1398 case TXQ_LOD
: code
[0] |= 0x12 << 25; break;
1399 case TXQ_BORDER_COLOUR
: code
[0] |= 0x16 << 25; break;
1401 assert(!"invalid texture query");
1405 code
[1] |= i
->tex
.mask
<< 2;
1406 code
[1] |= i
->tex
.r
<< 9;
1407 if (/*i->tex.sIndirectSrc >= 0 || */i
->tex
.rIndirectSrc
>= 0)
1408 code
[1] |= 0x08000000;
1410 defId(i
->def(0), 2);
1411 srcId(i
->src(0), 10);
1417 CodeEmitterGK110::emitQUADOP(const Instruction
*i
, uint8_t qOp
, uint8_t laneMask
)
1419 code
[0] = 0x00000002 | ((qOp
& 1) << 31);
1420 code
[1] = 0x7fc00200 | (qOp
>> 1) | (laneMask
<< 12); // dall
1422 defId(i
->def(0), 2);
1423 srcId(i
->src(0), 10);
1424 srcId((i
->srcExists(1) && i
->predSrc
!= 1) ? i
->src(1) : i
->src(0), 23);
1430 CodeEmitterGK110::emitPIXLD(const Instruction
*i
)
1432 emitForm_L(i
, 0x7f4, 2, Modifier(0));
1433 code
[1] |= i
->subOp
<< 2;
1434 code
[1] |= 0x00070000;
1438 CodeEmitterGK110::emitBAR(const Instruction
*i
)
1440 code
[0] = 0x00000002;
1441 code
[1] = 0x85400000;
1444 case NV50_IR_SUBOP_BAR_ARRIVE
: code
[1] |= 0x08; break;
1445 case NV50_IR_SUBOP_BAR_RED_AND
: code
[1] |= 0x50; break;
1446 case NV50_IR_SUBOP_BAR_RED_OR
: code
[1] |= 0x90; break;
1447 case NV50_IR_SUBOP_BAR_RED_POPC
: code
[1] |= 0x10; break;
1449 assert(i
->subOp
== NV50_IR_SUBOP_BAR_SYNC
);
1456 if (i
->src(0).getFile() == FILE_GPR
) {
1457 srcId(i
->src(0), 10);
1459 ImmediateValue
*imm
= i
->getSrc(0)->asImm();
1461 code
[0] |= imm
->reg
.data
.u32
<< 10;
1466 if (i
->src(1).getFile() == FILE_GPR
) {
1467 srcId(i
->src(1), 23);
1469 ImmediateValue
*imm
= i
->getSrc(0)->asImm();
1471 assert(imm
->reg
.data
.u32
<= 0xfff);
1472 code
[0] |= imm
->reg
.data
.u32
<< 23;
1473 code
[1] |= imm
->reg
.data
.u32
>> 9;
1477 if (i
->srcExists(2) && (i
->predSrc
!= 2)) {
1478 srcId(i
->src(2), 32 + 10);
1479 if (i
->src(2).mod
== Modifier(NV50_IR_MOD_NOT
))
1486 void CodeEmitterGK110::emitMEMBAR(const Instruction
*i
)
1488 code
[0] = 0x00000002 | NV50_IR_SUBOP_MEMBAR_SCOPE(i
->subOp
) << 8;
1489 code
[1] = 0x7cc00000;
1495 CodeEmitterGK110::emitFlow(const Instruction
*i
)
1497 const FlowInstruction
*f
= i
->asFlow();
1499 unsigned mask
; // bit 0: predicate, bit 1: target
1501 code
[0] = 0x00000000;
1505 code
[1] = f
->absolute
? 0x10800000 : 0x12000000;
1506 if (i
->srcExists(0) && i
->src(0).getFile() == FILE_MEMORY_CONST
)
1511 code
[1] = f
->absolute
? 0x11000000 : 0x13000000;
1512 if (i
->srcExists(0) && i
->src(0).getFile() == FILE_MEMORY_CONST
)
1517 case OP_EXIT
: code
[1] = 0x18000000; mask
= 1; break;
1518 case OP_RET
: code
[1] = 0x19000000; mask
= 1; break;
1519 case OP_DISCARD
: code
[1] = 0x19800000; mask
= 1; break;
1520 case OP_BREAK
: code
[1] = 0x1a000000; mask
= 1; break;
1521 case OP_CONT
: code
[1] = 0x1a800000; mask
= 1; break;
1523 case OP_JOINAT
: code
[1] = 0x14800000; mask
= 2; break;
1524 case OP_PREBREAK
: code
[1] = 0x15000000; mask
= 2; break;
1525 case OP_PRECONT
: code
[1] = 0x15800000; mask
= 2; break;
1526 case OP_PRERET
: code
[1] = 0x13800000; mask
= 2; break;
1528 case OP_QUADON
: code
[1] = 0x1b800000; mask
= 0; break;
1529 case OP_QUADPOP
: code
[1] = 0x1c000000; mask
= 0; break;
1530 case OP_BRKPT
: code
[1] = 0x00000000; mask
= 0; break;
1532 assert(!"invalid flow operation");
1538 if (i
->flagsSrc
< 0)
1550 if (f
->op
== OP_CALL
) {
1552 assert(f
->absolute
);
1553 uint32_t pcAbs
= targNVC0
->getBuiltinOffset(f
->target
.builtin
);
1554 addReloc(RelocEntry::TYPE_BUILTIN
, 0, pcAbs
, 0xff800000, 23);
1555 addReloc(RelocEntry::TYPE_BUILTIN
, 1, pcAbs
, 0x007fffff, -9);
1557 assert(!f
->absolute
);
1558 int32_t pcRel
= f
->target
.fn
->binPos
- (codeSize
+ 8);
1559 code
[0] |= (pcRel
& 0x1ff) << 23;
1560 code
[1] |= (pcRel
>> 9) & 0x7fff;
1564 int32_t pcRel
= f
->target
.bb
->binPos
- (codeSize
+ 8);
1565 if (writeIssueDelays
&& !(f
->target
.bb
->binPos
& 0x3f))
1567 // currently we don't want absolute branches
1568 assert(!f
->absolute
);
1569 code
[0] |= (pcRel
& 0x1ff) << 23;
1570 code
[1] |= (pcRel
>> 9) & 0x7fff;
1575 CodeEmitterGK110::emitSHFL(const Instruction
*i
)
1577 const ImmediateValue
*imm
;
1579 code
[0] = 0x00000002;
1580 code
[1] = 0x78800000 | (i
->subOp
<< 1);
1584 defId(i
->def(0), 2);
1585 srcId(i
->src(0), 10);
1587 switch (i
->src(1).getFile()) {
1589 srcId(i
->src(1), 23);
1591 case FILE_IMMEDIATE
:
1592 imm
= i
->getSrc(1)->asImm();
1593 assert(imm
&& imm
->reg
.data
.u32
< 0x20);
1594 code
[0] |= imm
->reg
.data
.u32
<< 23;
1598 assert(!"invalid src1 file");
1602 switch (i
->src(2).getFile()) {
1604 srcId(i
->src(2), 42);
1606 case FILE_IMMEDIATE
:
1607 imm
= i
->getSrc(2)->asImm();
1608 assert(imm
&& imm
->reg
.data
.u32
< 0x2000);
1609 code
[1] |= imm
->reg
.data
.u32
<< 5;
1613 assert(!"invalid src2 file");
1617 if (!i
->defExists(1))
1620 assert(i
->def(1).getFile() == FILE_PREDICATE
);
1621 defId(i
->def(1), 51);
1626 CodeEmitterGK110::emitVOTE(const Instruction
*i
)
1628 const ImmediateValue
*imm
;
1631 code
[0] = 0x00000002;
1632 code
[1] = 0x86c00000 | (i
->subOp
<< 19);
1637 for (int d
= 0; i
->defExists(d
); d
++) {
1638 if (i
->def(d
).getFile() == FILE_PREDICATE
) {
1641 defId(i
->def(d
), 48);
1642 } else if (i
->def(d
).getFile() == FILE_GPR
) {
1645 defId(i
->def(d
), 2);
1647 assert(!"Unhandled def");
1651 code
[0] |= 255 << 2;
1655 switch (i
->src(0).getFile()) {
1656 case FILE_PREDICATE
:
1657 if (i
->src(0).mod
== Modifier(NV50_IR_MOD_NOT
))
1659 srcId(i
->src(0), 42);
1661 case FILE_IMMEDIATE
:
1662 imm
= i
->getSrc(0)->asImm();
1664 u32
= imm
->reg
.data
.u32
;
1665 assert(u32
== 0 || u32
== 1);
1666 code
[1] |= (u32
== 1 ? 0x7 : 0xf) << 10;
1669 assert(!"Unhandled src");
1675 CodeEmitterGK110::emitSUGType(DataType ty
, const int pos
)
1680 case TYPE_S32
: n
= 1; break;
1681 case TYPE_U8
: n
= 2; break;
1682 case TYPE_S8
: n
= 3; break;
1684 assert(ty
== TYPE_U32
);
1687 code
[pos
/ 32] |= n
<< (pos
% 32);
1691 CodeEmitterGK110::emitSUCachingMode(CacheMode c
)
1711 assert(!"invalid caching mode");
1714 code
[0] |= (n
& 1) << 31;
1715 code
[1] |= (n
& 2) >> 1;
1719 CodeEmitterGK110::setSUConst16(const Instruction
*i
, const int s
)
1721 const uint32_t offset
= i
->getSrc(s
)->reg
.data
.offset
;
1723 assert(offset
== (offset
& 0xfffc));
1725 code
[0] |= offset
<< 21;
1726 code
[1] |= offset
>> 11;
1727 code
[1] |= i
->getSrc(s
)->reg
.fileIndex
<< 5;
1731 CodeEmitterGK110::emitSULDGB(const TexInstruction
*i
)
1733 code
[0] = 0x00000002;
1734 code
[1] = 0x30000000 | (i
->subOp
<< 14);
1736 if (i
->src(1).getFile() == FILE_MEMORY_CONST
) {
1737 emitLoadStoreType(i
->dType
, 0x38);
1738 emitCachingMode(i
->cache
, 0x36);
1743 assert(i
->src(1).getFile() == FILE_GPR
);
1744 code
[1] |= 0x49800000;
1746 emitLoadStoreType(i
->dType
, 0x21);
1747 emitSUCachingMode(i
->cache
);
1749 srcId(i
->src(1), 23);
1752 emitSUGType(i
->sType
, 0x34);
1755 defId(i
->def(0), 2); // destination
1756 srcId(i
->src(0), 10); // address
1758 // surface predicate
1759 if (!i
->srcExists(2) || (i
->predSrc
== 2)) {
1760 code
[1] |= 0x7 << 10;
1762 if (i
->src(2).mod
== Modifier(NV50_IR_MOD_NOT
))
1764 srcId(i
->src(2), 32 + 10);
1769 CodeEmitterGK110::emitSUSTGx(const TexInstruction
*i
)
1771 assert(i
->op
== OP_SUSTP
);
1773 code
[0] = 0x00000002;
1774 code
[1] = 0x38000000;
1776 if (i
->src(1).getFile() == FILE_MEMORY_CONST
) {
1777 code
[0] |= i
->subOp
<< 2;
1779 if (i
->op
== OP_SUSTP
)
1780 code
[0] |= i
->tex
.mask
<< 4;
1782 emitSUGType(i
->sType
, 0x8);
1783 emitCachingMode(i
->cache
, 0x36);
1788 assert(i
->src(1).getFile() == FILE_GPR
);
1790 code
[0] |= i
->subOp
<< 23;
1791 code
[1] |= 0x41c00000;
1793 if (i
->op
== OP_SUSTP
)
1794 code
[0] |= i
->tex
.mask
<< 25;
1796 emitSUGType(i
->sType
, 0x1d);
1797 emitSUCachingMode(i
->cache
);
1799 srcId(i
->src(1), 2);
1803 srcId(i
->src(0), 10); // address
1804 srcId(i
->src(3), 42); // values
1806 // surface predicate
1807 if (!i
->srcExists(2) || (i
->predSrc
== 2)) {
1808 code
[1] |= 0x7 << 18;
1810 if (i
->src(2).mod
== Modifier(NV50_IR_MOD_NOT
))
1812 srcId(i
->src(2), 32 + 18);
1817 CodeEmitterGK110::emitSUCLAMPMode(uint16_t subOp
)
1820 switch (subOp
& ~NV50_IR_SUBOP_SUCLAMP_2D
) {
1821 case NV50_IR_SUBOP_SUCLAMP_SD(0, 1): m
= 0; break;
1822 case NV50_IR_SUBOP_SUCLAMP_SD(1, 1): m
= 1; break;
1823 case NV50_IR_SUBOP_SUCLAMP_SD(2, 1): m
= 2; break;
1824 case NV50_IR_SUBOP_SUCLAMP_SD(3, 1): m
= 3; break;
1825 case NV50_IR_SUBOP_SUCLAMP_SD(4, 1): m
= 4; break;
1826 case NV50_IR_SUBOP_SUCLAMP_PL(0, 1): m
= 5; break;
1827 case NV50_IR_SUBOP_SUCLAMP_PL(1, 1): m
= 6; break;
1828 case NV50_IR_SUBOP_SUCLAMP_PL(2, 1): m
= 7; break;
1829 case NV50_IR_SUBOP_SUCLAMP_PL(3, 1): m
= 8; break;
1830 case NV50_IR_SUBOP_SUCLAMP_PL(4, 1): m
= 9; break;
1831 case NV50_IR_SUBOP_SUCLAMP_BL(0, 1): m
= 10; break;
1832 case NV50_IR_SUBOP_SUCLAMP_BL(1, 1): m
= 11; break;
1833 case NV50_IR_SUBOP_SUCLAMP_BL(2, 1): m
= 12; break;
1834 case NV50_IR_SUBOP_SUCLAMP_BL(3, 1): m
= 13; break;
1835 case NV50_IR_SUBOP_SUCLAMP_BL(4, 1): m
= 14; break;
1840 if (subOp
& NV50_IR_SUBOP_SUCLAMP_2D
)
1845 CodeEmitterGK110::emitSUCalc(Instruction
*i
)
1847 ImmediateValue
*imm
= NULL
;
1848 uint64_t opc1
, opc2
;
1850 if (i
->srcExists(2)) {
1851 imm
= i
->getSrc(2)->asImm();
1853 i
->setSrc(2, NULL
); // special case, make emitForm_21 not assert
1857 case OP_SUCLAMP
: opc1
= 0xb00; opc2
= 0x580; break;
1858 case OP_SUBFM
: opc1
= 0xb68; opc2
= 0x1e8; break;
1859 case OP_SUEAU
: opc1
= 0xb6c; opc2
= 0x1ec; break;
1864 emitForm_21(i
, opc2
, opc1
);
1866 if (i
->op
== OP_SUCLAMP
) {
1867 if (i
->dType
== TYPE_S32
)
1869 emitSUCLAMPMode(i
->subOp
);
1872 if (i
->op
== OP_SUBFM
&& i
->subOp
== NV50_IR_SUBOP_SUBFM_3D
)
1875 if (i
->op
!= OP_SUEAU
) {
1876 const uint8_t pos
= i
->op
== OP_SUBFM
? 19 : 16;
1877 if (i
->def(0).getFile() == FILE_PREDICATE
) { // p, #
1878 code
[0] |= 255 << 2;
1879 code
[1] |= i
->getDef(1)->reg
.data
.id
<< pos
;
1881 if (i
->defExists(1)) { // r, p
1882 assert(i
->def(1).getFile() == FILE_PREDICATE
);
1883 code
[1] |= i
->getDef(1)->reg
.data
.id
<< pos
;
1885 code
[1] |= 7 << pos
;
1890 assert(i
->op
== OP_SUCLAMP
);
1892 code
[1] |= (imm
->reg
.data
.u32
& 0x3f) << 10; // sint6
1898 CodeEmitterGK110::emitVectorSubOp(const Instruction
*i
)
1900 switch (NV50_IR_SUBOP_Vn(i
->subOp
)) {
1902 code
[1] |= (i
->subOp
& 0x000f) << 7; // vsrc1
1903 code
[1] |= (i
->subOp
& 0x00e0) >> 6; // vsrc2
1904 code
[1] |= (i
->subOp
& 0x0100) << 13; // vsrc2
1905 code
[1] |= (i
->subOp
& 0x3c00) << 12; // vdst
1914 CodeEmitterGK110::emitVSHL(const Instruction
*i
)
1916 code
[0] = 0x00000002;
1917 code
[1] = 0xb8000000;
1919 assert(NV50_IR_SUBOP_Vn(i
->subOp
) == 0);
1921 if (isSignedType(i
->dType
)) code
[1] |= 1 << 25;
1922 if (isSignedType(i
->sType
)) code
[1] |= 1 << 19;
1927 defId(i
->def(0), 2);
1928 srcId(i
->src(0), 10);
1930 if (i
->getSrc(1)->reg
.file
== FILE_IMMEDIATE
) {
1931 ImmediateValue
*imm
= i
->getSrc(1)->asImm();
1933 code
[0] |= (imm
->reg
.data
.u32
& 0x01ff) << 23;
1934 code
[1] |= (imm
->reg
.data
.u32
& 0xfe00) >> 9;
1936 assert(i
->getSrc(1)->reg
.file
== FILE_GPR
);
1938 srcId(i
->src(1), 23);
1940 srcId(i
->src(2), 42);
1944 if (i
->flagsDef
>= 0)
1949 CodeEmitterGK110::emitAFETCH(const Instruction
*i
)
1951 uint32_t offset
= i
->src(0).get()->reg
.data
.offset
& 0x7ff;
1953 code
[0] = 0x00000002 | (offset
<< 23);
1954 code
[1] = 0x7d000000 | (offset
>> 9);
1956 if (i
->getSrc(0)->reg
.file
== FILE_SHADER_OUTPUT
)
1961 defId(i
->def(0), 2);
1962 srcId(i
->src(0).getIndirect(0), 10);
1966 CodeEmitterGK110::emitPFETCH(const Instruction
*i
)
1968 uint32_t prim
= i
->src(0).get()->reg
.data
.u32
;
1970 code
[0] = 0x00000002 | ((prim
& 0xff) << 23);
1971 code
[1] = 0x7f800000;
1975 const int src1
= (i
->predSrc
== 1) ? 2 : 1; // if predSrc == 1, !srcExists(2)
1977 defId(i
->def(0), 2);
1982 CodeEmitterGK110::emitVFETCH(const Instruction
*i
)
1984 unsigned int size
= typeSizeof(i
->dType
);
1985 uint32_t offset
= i
->src(0).get()->reg
.data
.offset
;
1987 code
[0] = 0x00000002 | (offset
<< 23);
1988 code
[1] = 0x7ec00000 | (offset
>> 9);
1989 code
[1] |= (size
/ 4 - 1) << 18;
1993 if (i
->getSrc(0)->reg
.file
== FILE_SHADER_OUTPUT
)
1994 code
[1] |= 0x8; // yes, TCPs can read from *outputs* of other threads
1998 defId(i
->def(0), 2);
1999 srcId(i
->src(0).getIndirect(0), 10);
2000 srcId(i
->src(0).getIndirect(1), 32 + 10); // vertex address
2004 CodeEmitterGK110::emitEXPORT(const Instruction
*i
)
2006 unsigned int size
= typeSizeof(i
->dType
);
2007 uint32_t offset
= i
->src(0).get()->reg
.data
.offset
;
2009 code
[0] = 0x00000002 | (offset
<< 23);
2010 code
[1] = 0x7f000000 | (offset
>> 9);
2011 code
[1] |= (size
/ 4 - 1) << 18;
2018 assert(i
->src(1).getFile() == FILE_GPR
);
2020 srcId(i
->src(0).getIndirect(0), 10);
2021 srcId(i
->src(0).getIndirect(1), 32 + 10); // vertex base address
2022 srcId(i
->src(1), 2);
2026 CodeEmitterGK110::emitOUT(const Instruction
*i
)
2028 assert(i
->src(0).getFile() == FILE_GPR
);
2030 emitForm_21(i
, 0x1f0, 0xb70);
2032 if (i
->op
== OP_EMIT
)
2034 if (i
->op
== OP_RESTART
|| i
->subOp
== NV50_IR_SUBOP_EMIT_RESTART
)
2039 CodeEmitterGK110::emitInterpMode(const Instruction
*i
)
2041 code
[1] |= (i
->ipa
& 0x3) << 21; // TODO: INTERP_SAMPLEID
2042 code
[1] |= (i
->ipa
& 0xc) << (19 - 2);
2046 interpApply(const FixupEntry
*entry
, uint32_t *code
, const FixupData
& data
)
2048 int ipa
= entry
->ipa
;
2049 int reg
= entry
->reg
;
2050 int loc
= entry
->loc
;
2052 if (data
.flatshade
&&
2053 (ipa
& NV50_IR_INTERP_MODE_MASK
) == NV50_IR_INTERP_SC
) {
2054 ipa
= NV50_IR_INTERP_FLAT
;
2056 } else if (data
.force_persample_interp
&&
2057 (ipa
& NV50_IR_INTERP_SAMPLE_MASK
) == NV50_IR_INTERP_DEFAULT
&&
2058 (ipa
& NV50_IR_INTERP_MODE_MASK
) != NV50_IR_INTERP_FLAT
) {
2059 ipa
|= NV50_IR_INTERP_CENTROID
;
2061 code
[loc
+ 1] &= ~(0xf << 19);
2062 code
[loc
+ 1] |= (ipa
& 0x3) << 21;
2063 code
[loc
+ 1] |= (ipa
& 0xc) << (19 - 2);
2064 code
[loc
+ 0] &= ~(0xff << 23);
2065 code
[loc
+ 0] |= reg
<< 23;
2069 CodeEmitterGK110::emitINTERP(const Instruction
*i
)
2071 const uint32_t base
= i
->getSrc(0)->reg
.data
.offset
;
2073 code
[0] = 0x00000002 | (base
<< 31);
2074 code
[1] = 0x74800000 | (base
>> 1);
2079 if (i
->op
== OP_PINTERP
) {
2080 srcId(i
->src(1), 23);
2081 addInterp(i
->ipa
, SDATA(i
->src(1)).id
, interpApply
);
2083 code
[0] |= 0xff << 23;
2084 addInterp(i
->ipa
, 0xff, interpApply
);
2087 srcId(i
->src(0).getIndirect(0), 10);
2091 defId(i
->def(0), 2);
2093 if (i
->getSampleMode() == NV50_IR_INTERP_OFFSET
)
2094 srcId(i
->src(i
->op
== OP_PINTERP
? 2 : 1), 32 + 10);
2096 code
[1] |= 0xff << 10;
2100 CodeEmitterGK110::emitLoadStoreType(DataType ty
, const int pos
)
2132 assert(!"invalid ld/st type");
2135 code
[pos
/ 32] |= n
<< (pos
% 32);
2139 CodeEmitterGK110::emitCachingMode(CacheMode c
, const int pos
)
2160 assert(!"invalid caching mode");
2163 code
[pos
/ 32] |= n
<< (pos
% 32);
2167 CodeEmitterGK110::emitSTORE(const Instruction
*i
)
2169 int32_t offset
= SDATA(i
->src(0)).offset
;
2171 switch (i
->src(0).getFile()) {
2172 case FILE_MEMORY_GLOBAL
: code
[1] = 0xe0000000; code
[0] = 0x00000000; break;
2173 case FILE_MEMORY_LOCAL
: code
[1] = 0x7a800000; code
[0] = 0x00000002; break;
2174 case FILE_MEMORY_SHARED
:
2175 code
[0] = 0x00000002;
2176 if (i
->subOp
== NV50_IR_SUBOP_STORE_UNLOCKED
)
2177 code
[1] = 0x78400000;
2179 code
[1] = 0x7ac00000;
2182 assert(!"invalid memory file");
2186 if (code
[0] & 0x2) {
2188 emitLoadStoreType(i
->dType
, 0x33);
2189 if (i
->src(0).getFile() == FILE_MEMORY_LOCAL
)
2190 emitCachingMode(i
->cache
, 0x2f);
2192 emitLoadStoreType(i
->dType
, 0x38);
2193 emitCachingMode(i
->cache
, 0x3b);
2195 code
[0] |= offset
<< 23;
2196 code
[1] |= offset
>> 9;
2198 // Unlocked store on shared memory can fail.
2199 if (i
->src(0).getFile() == FILE_MEMORY_SHARED
&&
2200 i
->subOp
== NV50_IR_SUBOP_STORE_UNLOCKED
) {
2201 assert(i
->defExists(0));
2202 defId(i
->def(0), 32 + 16);
2207 srcId(i
->src(1), 2);
2208 srcId(i
->src(0).getIndirect(0), 10);
2209 if (i
->src(0).getFile() == FILE_MEMORY_GLOBAL
&&
2210 i
->src(0).isIndirect(0) &&
2211 i
->getIndirect(0, 0)->reg
.size
== 8)
2216 CodeEmitterGK110::emitLOAD(const Instruction
*i
)
2218 int32_t offset
= SDATA(i
->src(0)).offset
;
2220 switch (i
->src(0).getFile()) {
2221 case FILE_MEMORY_GLOBAL
: code
[1] = 0xc0000000; code
[0] = 0x00000000; break;
2222 case FILE_MEMORY_LOCAL
: code
[1] = 0x7a000000; code
[0] = 0x00000002; break;
2223 case FILE_MEMORY_SHARED
:
2224 code
[0] = 0x00000002;
2225 if (i
->subOp
== NV50_IR_SUBOP_LOAD_LOCKED
)
2226 code
[1] = 0x77400000;
2228 code
[1] = 0x7a400000;
2230 case FILE_MEMORY_CONST
:
2231 if (!i
->src(0).isIndirect(0) && typeSizeof(i
->dType
) == 4) {
2236 code
[0] = 0x00000002;
2237 code
[1] = 0x7c800000 | (i
->src(0).get()->reg
.fileIndex
<< 7);
2238 code
[1] |= i
->subOp
<< 15;
2241 assert(!"invalid memory file");
2245 if (code
[0] & 0x2) {
2247 emitLoadStoreType(i
->dType
, 0x33);
2248 if (i
->src(0).getFile() == FILE_MEMORY_LOCAL
)
2249 emitCachingMode(i
->cache
, 0x2f);
2251 emitLoadStoreType(i
->dType
, 0x38);
2252 emitCachingMode(i
->cache
, 0x3b);
2254 code
[0] |= offset
<< 23;
2255 code
[1] |= offset
>> 9;
2257 // Locked store on shared memory can fail.
2259 if (i
->src(0).getFile() == FILE_MEMORY_SHARED
&&
2260 i
->subOp
== NV50_IR_SUBOP_LOAD_LOCKED
) {
2261 if (i
->def(0).getFile() == FILE_PREDICATE
) { // p, #
2264 } else if (i
->defExists(1)) { // r, p
2267 assert(!"Expected predicate dest for load locked");
2274 defId(i
->def(r
), 2);
2276 code
[0] |= 255 << 2;
2279 defId(i
->def(p
), 32 + 16);
2281 if (i
->getIndirect(0, 0)) {
2282 srcId(i
->src(0).getIndirect(0), 10);
2283 if (i
->getIndirect(0, 0)->reg
.size
== 8)
2286 code
[0] |= 255 << 10;
2291 CodeEmitterGK110::getSRegEncoding(const ValueRef
& ref
)
2293 switch (SDATA(ref
).sv
.sv
) {
2294 case SV_LANEID
: return 0x00;
2295 case SV_PHYSID
: return 0x03;
2296 case SV_VERTEX_COUNT
: return 0x10;
2297 case SV_INVOCATION_ID
: return 0x11;
2298 case SV_YDIR
: return 0x12;
2299 case SV_THREAD_KILL
: return 0x13;
2300 case SV_COMBINED_TID
: return 0x20;
2301 case SV_TID
: return 0x21 + SDATA(ref
).sv
.index
;
2302 case SV_CTAID
: return 0x25 + SDATA(ref
).sv
.index
;
2303 case SV_NTID
: return 0x29 + SDATA(ref
).sv
.index
;
2304 case SV_GRIDID
: return 0x2c;
2305 case SV_NCTAID
: return 0x2d + SDATA(ref
).sv
.index
;
2306 case SV_LBASE
: return 0x34;
2307 case SV_SBASE
: return 0x30;
2308 case SV_LANEMASK_EQ
: return 0x38;
2309 case SV_LANEMASK_LT
: return 0x39;
2310 case SV_LANEMASK_LE
: return 0x3a;
2311 case SV_LANEMASK_GT
: return 0x3b;
2312 case SV_LANEMASK_GE
: return 0x3c;
2313 case SV_CLOCK
: return 0x50 + SDATA(ref
).sv
.index
;
2315 assert(!"no sreg for system value");
2321 CodeEmitterGK110::emitMOV(const Instruction
*i
)
2323 if (i
->def(0).getFile() == FILE_PREDICATE
) {
2324 if (i
->src(0).getFile() == FILE_GPR
) {
2325 // Use ISETP.NE.AND dst, PT, src, RZ, PT
2326 code
[0] = 0x00000002;
2327 code
[1] = 0xdb500000;
2329 code
[0] |= 0x7 << 2;
2330 code
[0] |= 0xff << 23;
2331 code
[1] |= 0x7 << 10;
2332 srcId(i
->src(0), 10);
2334 if (i
->src(0).getFile() == FILE_PREDICATE
) {
2335 // Use PSETP.AND.AND dst, PT, src, PT, PT
2336 code
[0] = 0x00000002;
2337 code
[1] = 0x84800000;
2339 code
[0] |= 0x7 << 2;
2340 code
[1] |= 0x7 << 0;
2341 code
[1] |= 0x7 << 10;
2343 srcId(i
->src(0), 14);
2345 assert(!"Unexpected source for predicate destination");
2349 defId(i
->def(0), 5);
2351 if (i
->src(0).getFile() == FILE_SYSTEM_VALUE
) {
2352 code
[0] = 0x00000002 | (getSRegEncoding(i
->src(0)) << 23);
2353 code
[1] = 0x86400000;
2355 defId(i
->def(0), 2);
2357 if (i
->src(0).getFile() == FILE_IMMEDIATE
) {
2358 code
[0] = 0x00000002 | (i
->lanes
<< 14);
2359 code
[1] = 0x74000000;
2361 defId(i
->def(0), 2);
2362 setImmediate32(i
, 0, Modifier(0));
2364 if (i
->src(0).getFile() == FILE_PREDICATE
) {
2365 code
[0] = 0x00000002;
2366 code
[1] = 0x84401c07;
2368 defId(i
->def(0), 2);
2369 srcId(i
->src(0), 14);
2371 emitForm_C(i
, 0x24c, 2);
2372 code
[1] |= i
->lanes
<< 10;
2377 uses64bitAddress(const Instruction
*ldst
)
2379 return ldst
->src(0).getFile() == FILE_MEMORY_GLOBAL
&&
2380 ldst
->src(0).isIndirect(0) &&
2381 ldst
->getIndirect(0, 0)->reg
.size
== 8;
2385 CodeEmitterGK110::emitATOM(const Instruction
*i
)
2387 const bool hasDst
= i
->defExists(0);
2388 const bool exch
= i
->subOp
== NV50_IR_SUBOP_ATOM_EXCH
;
2390 code
[0] = 0x00000002;
2391 if (i
->subOp
== NV50_IR_SUBOP_ATOM_CAS
)
2392 code
[1] = 0x77800000;
2394 code
[1] = 0x68000000;
2397 case NV50_IR_SUBOP_ATOM_CAS
: break;
2398 case NV50_IR_SUBOP_ATOM_EXCH
: code
[1] |= 0x04000000; break;
2399 default: code
[1] |= i
->subOp
<< 23; break;
2403 case TYPE_U32
: break;
2404 case TYPE_S32
: code
[1] |= 0x00100000; break;
2405 case TYPE_U64
: code
[1] |= 0x00200000; break;
2406 case TYPE_F32
: code
[1] |= 0x00300000; break;
2407 case TYPE_B128
: code
[1] |= 0x00400000; break; /* TODO: U128 */
2408 case TYPE_S64
: code
[1] |= 0x00500000; break;
2409 default: assert(!"unsupported type"); break;
2414 /* TODO: cas: check that src regs line up */
2415 /* TODO: cas: flip bits if $r255 is used */
2416 srcId(i
->src(1), 23);
2419 defId(i
->def(0), 2);
2422 code
[0] |= 255 << 2;
2425 if (hasDst
|| !exch
) {
2426 const int32_t offset
= SDATA(i
->src(0)).offset
;
2427 assert(offset
< 0x80000 && offset
>= -0x80000);
2428 code
[0] |= (offset
& 1) << 31;
2429 code
[1] |= (offset
& 0xffffe) >> 1;
2431 srcAddr32(i
->src(0), 31);
2434 if (i
->getIndirect(0, 0)) {
2435 srcId(i
->getIndirect(0, 0), 10);
2436 if (i
->getIndirect(0, 0)->reg
.size
== 8)
2439 code
[0] |= 255 << 10;
2444 CodeEmitterGK110::emitCCTL(const Instruction
*i
)
2446 int32_t offset
= SDATA(i
->src(0)).offset
;
2448 code
[0] = 0x00000002 | (i
->subOp
<< 2);
2450 if (i
->src(0).getFile() == FILE_MEMORY_GLOBAL
) {
2451 code
[1] = 0x7b000000;
2453 code
[1] = 0x7c000000;
2456 code
[0] |= offset
<< 23;
2457 code
[1] |= offset
>> 9;
2459 if (uses64bitAddress(i
))
2461 srcId(i
->src(0).getIndirect(0), 10);
2467 CodeEmitterGK110::emitInstruction(Instruction
*insn
)
2469 const unsigned int size
= (writeIssueDelays
&& !(codeSize
& 0x3f)) ? 16 : 8;
2471 if (insn
->encSize
!= 8) {
2472 ERROR("skipping unencodable instruction: ");
2476 if (codeSize
+ size
> codeSizeLimit
) {
2477 ERROR("code emitter output buffer too small\n");
2481 if (writeIssueDelays
) {
2482 int id
= (codeSize
& 0x3f) / 8 - 1;
2485 code
[0] = 0x00000000; // cf issue delay "instruction"
2486 code
[1] = 0x08000000;
2490 uint32_t *data
= code
- (id
* 2 + 2);
2493 case 0: data
[0] |= insn
->sched
<< 2; break;
2494 case 1: data
[0] |= insn
->sched
<< 10; break;
2495 case 2: data
[0] |= insn
->sched
<< 18; break;
2496 case 3: data
[0] |= insn
->sched
<< 26; data
[1] |= insn
->sched
>> 6; break;
2497 case 4: data
[1] |= insn
->sched
<< 2; break;
2498 case 5: data
[1] |= insn
->sched
<< 10; break;
2499 case 6: data
[1] |= insn
->sched
<< 18; break;
2506 // assert that instructions with multiple defs don't corrupt registers
2507 for (int d
= 0; insn
->defExists(d
); ++d
)
2508 assert(insn
->asTex() || insn
->def(d
).rep()->reg
.data
.id
>= 0);
2545 if (insn
->dType
== TYPE_F64
)
2547 else if (isFloatType(insn
->dType
))
2553 if (insn
->dType
== TYPE_F64
)
2555 else if (isFloatType(insn
->dType
))
2562 if (insn
->dType
== TYPE_F64
)
2564 else if (isFloatType(insn
->dType
))
2582 emitLogicOp(insn
, 0);
2585 emitLogicOp(insn
, 1);
2588 emitLogicOp(insn
, 2);
2592 if (typeSizeof(insn
->sType
) == 8)
2601 emitSET(insn
->asCmp());
2607 emitSLCT(insn
->asCmp());
2622 if (insn
->def(0).getFile() == FILE_PREDICATE
||
2623 insn
->src(0).getFile() == FILE_PREDICATE
)
2629 emitSFnOp(insn
, 5 + 2 * insn
->subOp
);
2632 emitSFnOp(insn
, 4 + 2 * insn
->subOp
);
2657 emitTEX(insn
->asTex());
2660 emitTXQ(insn
->asTex());
2685 emitQUADOP(insn
, insn
->subOp
, insn
->lanes
);
2688 emitQUADOP(insn
, insn
->src(0).mod
.neg() ? 0x66 : 0x99, 0x4);
2691 emitQUADOP(insn
, insn
->src(0).mod
.neg() ? 0x5a : 0xa5, 0x5);
2731 emitSULDGB(insn
->asTex());
2735 emitSUSTGx(insn
->asTex());
2748 ERROR("operation should have been eliminated");
2754 ERROR("operation should have been lowered\n");
2757 ERROR("unknown op: %u\n", insn
->op
);
2770 CodeEmitterGK110::getMinEncodingSize(const Instruction
*i
) const
2772 // No more short instruction encodings.
2777 CodeEmitterGK110::prepareEmission(Function
*func
)
2779 const Target
*targ
= func
->getProgram()->getTarget();
2781 CodeEmitter::prepareEmission(func
);
2783 if (targ
->hasSWSched
)
2784 calculateSchedDataNVC0(targ
, func
);
2787 CodeEmitterGK110::CodeEmitterGK110(const TargetNVC0
*target
)
2788 : CodeEmitter(target
),
2790 writeIssueDelays(target
->hasSWSched
)
2793 codeSize
= codeSizeLimit
= 0;
2798 TargetNVC0::createCodeEmitterGK110(Program::Type type
)
2800 CodeEmitterGK110
*emit
= new CodeEmitterGK110(this);
2801 emit
->setProgramType(type
);
2805 } // namespace nv50_ir