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
);
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
);
58 void modNegAbsF32_3b(const Instruction
*, const int s
);
60 void emitCondCode(CondCode cc
, int pos
, uint8_t mask
);
61 void emitInterpMode(const Instruction
*);
62 void emitLoadStoreType(DataType ty
, const int pos
);
63 void emitCachingMode(CacheMode c
, const int pos
);
65 inline uint8_t getSRegEncoding(const ValueRef
&);
67 void emitRoundMode(RoundMode
, const int pos
, const int rintPos
);
68 void emitRoundModeF(RoundMode
, const int pos
);
69 void emitRoundModeI(RoundMode
, const int pos
);
71 void emitNegAbs12(const Instruction
*);
73 void emitNOP(const Instruction
*);
75 void emitLOAD(const Instruction
*);
76 void emitSTORE(const Instruction
*);
77 void emitMOV(const Instruction
*);
78 void emitMEMBAR(const Instruction
*);
80 void emitINTERP(const Instruction
*);
81 void emitAFETCH(const Instruction
*);
82 void emitPFETCH(const Instruction
*);
83 void emitVFETCH(const Instruction
*);
84 void emitEXPORT(const Instruction
*);
85 void emitOUT(const Instruction
*);
87 void emitUADD(const Instruction
*);
88 void emitFADD(const Instruction
*);
89 void emitDADD(const Instruction
*);
90 void emitIMUL(const Instruction
*);
91 void emitFMUL(const Instruction
*);
92 void emitDMUL(const Instruction
*);
93 void emitIMAD(const Instruction
*);
94 void emitISAD(const Instruction
*);
95 void emitFMAD(const Instruction
*);
96 void emitDMAD(const Instruction
*);
98 void emitNOT(const Instruction
*);
99 void emitLogicOp(const Instruction
*, uint8_t subOp
);
100 void emitPOPC(const Instruction
*);
101 void emitINSBF(const Instruction
*);
102 void emitEXTBF(const Instruction
*);
103 void emitBFIND(const Instruction
*);
104 void emitShift(const Instruction
*);
106 void emitSFnOp(const Instruction
*, uint8_t subOp
);
108 void emitCVT(const Instruction
*);
109 void emitMINMAX(const Instruction
*);
110 void emitPreOp(const Instruction
*);
112 void emitSET(const CmpInstruction
*);
113 void emitSLCT(const CmpInstruction
*);
114 void emitSELP(const Instruction
*);
116 void emitTEXBAR(const Instruction
*);
117 void emitTEX(const TexInstruction
*);
118 void emitTEXCSAA(const TexInstruction
*);
119 void emitTXQ(const TexInstruction
*);
121 void emitQUADOP(const Instruction
*, uint8_t qOp
, uint8_t laneMask
);
123 void emitPIXLD(const Instruction
*);
125 void emitBAR(const Instruction
*);
127 void emitFlow(const Instruction
*);
129 inline void defId(const ValueDef
&, const int pos
);
130 inline void srcId(const ValueRef
&, const int pos
);
131 inline void srcId(const ValueRef
*, const int pos
);
132 inline void srcId(const Instruction
*, int s
, const int pos
);
134 inline void srcAddr32(const ValueRef
&, const int pos
); // address / 4
136 inline bool isLIMM(const ValueRef
&, DataType ty
, bool mod
= false);
139 #define GK110_GPR_ZERO 255
142 if (i->src(s).mod.neg()) code[(0x##b) / 32] |= 1 << ((0x##b) % 32)
144 if (i->src(s).mod.abs()) code[(0x##b) / 32] |= 1 << ((0x##b) % 32)
146 #define NOT_(b, s) if (i->src(s).mod & Modifier(NV50_IR_MOD_NOT)) \
147 code[(0x##b) / 32] |= 1 << ((0x##b) % 32)
149 #define FTZ_(b) if (i->ftz) code[(0x##b) / 32] |= 1 << ((0x##b) % 32)
150 #define DNZ_(b) if (i->dnz) code[(0x##b) / 32] |= 1 << ((0x##b) % 32)
152 #define SAT_(b) if (i->saturate) code[(0x##b) / 32] |= 1 << ((0x##b) % 32)
154 #define RND_(b, t) emitRoundMode##t(i->rnd, 0x##b)
156 #define SDATA(a) ((a).rep()->reg.data)
157 #define DDATA(a) ((a).rep()->reg.data)
159 void CodeEmitterGK110::srcId(const ValueRef
& src
, const int pos
)
161 code
[pos
/ 32] |= (src
.get() ? SDATA(src
).id
: GK110_GPR_ZERO
) << (pos
% 32);
164 void CodeEmitterGK110::srcId(const ValueRef
*src
, const int pos
)
166 code
[pos
/ 32] |= (src
? SDATA(*src
).id
: GK110_GPR_ZERO
) << (pos
% 32);
169 void CodeEmitterGK110::srcId(const Instruction
*insn
, int s
, int pos
)
171 int r
= insn
->srcExists(s
) ? SDATA(insn
->src(s
)).id
: GK110_GPR_ZERO
;
172 code
[pos
/ 32] |= r
<< (pos
% 32);
175 void CodeEmitterGK110::srcAddr32(const ValueRef
& src
, const int pos
)
177 code
[pos
/ 32] |= (SDATA(src
).offset
>> 2) << (pos
% 32);
180 void CodeEmitterGK110::defId(const ValueDef
& def
, const int pos
)
182 code
[pos
/ 32] |= (def
.get() ? DDATA(def
).id
: GK110_GPR_ZERO
) << (pos
% 32);
185 bool CodeEmitterGK110::isLIMM(const ValueRef
& ref
, DataType ty
, bool mod
)
187 const ImmediateValue
*imm
= ref
.get()->asImm();
189 return imm
&& (imm
->reg
.data
.u32
& ((ty
== TYPE_F32
) ? 0xfff : 0xfff00000));
193 CodeEmitterGK110::emitRoundMode(RoundMode rnd
, const int pos
, const int rintPos
)
199 case ROUND_MI
: rint
= true; /* fall through */ case ROUND_M
: n
= 1; break;
200 case ROUND_PI
: rint
= true; /* fall through */ case ROUND_P
: n
= 2; break;
201 case ROUND_ZI
: rint
= true; /* fall through */ case ROUND_Z
: n
= 3; break;
203 rint
= rnd
== ROUND_NI
;
205 assert(rnd
== ROUND_N
|| rnd
== ROUND_NI
);
208 code
[pos
/ 32] |= n
<< (pos
% 32);
209 if (rint
&& rintPos
>= 0)
210 code
[rintPos
/ 32] |= 1 << (rintPos
% 32);
214 CodeEmitterGK110::emitRoundModeF(RoundMode rnd
, const int pos
)
219 case ROUND_M
: n
= 1; break;
220 case ROUND_P
: n
= 2; break;
221 case ROUND_Z
: n
= 3; break;
224 assert(rnd
== ROUND_N
);
227 code
[pos
/ 32] |= n
<< (pos
% 32);
231 CodeEmitterGK110::emitRoundModeI(RoundMode rnd
, const int pos
)
236 case ROUND_MI
: n
= 1; break;
237 case ROUND_PI
: n
= 2; break;
238 case ROUND_ZI
: n
= 3; break;
241 assert(rnd
== ROUND_NI
);
244 code
[pos
/ 32] |= n
<< (pos
% 32);
247 void CodeEmitterGK110::emitCondCode(CondCode cc
, int pos
, uint8_t mask
)
252 case CC_FL
: n
= 0x00; break;
253 case CC_LT
: n
= 0x01; break;
254 case CC_EQ
: n
= 0x02; break;
255 case CC_LE
: n
= 0x03; break;
256 case CC_GT
: n
= 0x04; break;
257 case CC_NE
: n
= 0x05; break;
258 case CC_GE
: n
= 0x06; break;
259 case CC_LTU
: n
= 0x09; break;
260 case CC_EQU
: n
= 0x0a; break;
261 case CC_LEU
: n
= 0x0b; break;
262 case CC_GTU
: n
= 0x0c; break;
263 case CC_NEU
: n
= 0x0d; break;
264 case CC_GEU
: n
= 0x0e; break;
265 case CC_TR
: n
= 0x0f; break;
266 case CC_NO
: n
= 0x10; break;
267 case CC_NC
: n
= 0x11; break;
268 case CC_NS
: n
= 0x12; break;
269 case CC_NA
: n
= 0x13; break;
270 case CC_A
: n
= 0x14; break;
271 case CC_S
: n
= 0x15; break;
272 case CC_C
: n
= 0x16; break;
273 case CC_O
: n
= 0x17; break;
276 assert(!"invalid condition code");
279 code
[pos
/ 32] |= (n
& mask
) << (pos
% 32);
283 CodeEmitterGK110::emitPredicate(const Instruction
*i
)
285 if (i
->predSrc
>= 0) {
286 srcId(i
->src(i
->predSrc
), 18);
287 if (i
->cc
== CC_NOT_P
)
288 code
[0] |= 8 << 18; // negate
289 assert(i
->getPredicate()->reg
.file
== FILE_PREDICATE
);
296 CodeEmitterGK110::setCAddress14(const ValueRef
& src
)
298 const Storage
& res
= src
.get()->asSym()->reg
;
299 const int32_t addr
= res
.data
.offset
/ 4;
301 code
[0] |= (addr
& 0x01ff) << 23;
302 code
[1] |= (addr
& 0x3e00) >> 9;
303 code
[1] |= res
.fileIndex
<< 5;
307 CodeEmitterGK110::setShortImmediate(const Instruction
*i
, const int s
)
309 const uint32_t u32
= i
->getSrc(s
)->asImm()->reg
.data
.u32
;
310 const uint64_t u64
= i
->getSrc(s
)->asImm()->reg
.data
.u64
;
312 if (i
->sType
== TYPE_F32
) {
313 assert(!(u32
& 0x00000fff));
314 code
[0] |= ((u32
& 0x001ff000) >> 12) << 23;
315 code
[1] |= ((u32
& 0x7fe00000) >> 21);
316 code
[1] |= ((u32
& 0x80000000) >> 4);
318 if (i
->sType
== TYPE_F64
) {
319 assert(!(u64
& 0x00000fffffffffffULL
));
320 code
[0] |= ((u64
& 0x001ff00000000000ULL
) >> 44) << 23;
321 code
[1] |= ((u64
& 0x7fe0000000000000ULL
) >> 53);
322 code
[1] |= ((u64
& 0x8000000000000000ULL
) >> 36);
324 assert((u32
& 0xfff00000) == 0 || (u32
& 0xfff00000) == 0xfff00000);
325 code
[0] |= (u32
& 0x001ff) << 23;
326 code
[1] |= (u32
& 0x7fe00) >> 9;
327 code
[1] |= (u32
& 0x80000) << 8;
332 CodeEmitterGK110::setImmediate32(const Instruction
*i
, const int s
,
335 uint32_t u32
= i
->getSrc(s
)->asImm()->reg
.data
.u32
;
338 ImmediateValue
imm(i
->getSrc(s
)->asImm(), i
->sType
);
340 u32
= imm
.reg
.data
.u32
;
343 code
[0] |= u32
<< 23;
348 CodeEmitterGK110::emitForm_L(const Instruction
*i
, uint32_t opc
, uint8_t ctg
,
358 for (int s
= 0; s
< 3 && i
->srcExists(s
); ++s
) {
359 switch (i
->src(s
).getFile()) {
361 srcId(i
->src(s
), s
? 42 : 10);
364 setImmediate32(i
, s
, mod
);
374 CodeEmitterGK110::emitForm_C(const Instruction
*i
, uint32_t opc
, uint8_t ctg
)
383 switch (i
->src(0).getFile()) {
384 case FILE_MEMORY_CONST
:
385 code
[1] |= 0x4 << 28;
386 setCAddress14(i
->src(0));
389 code
[1] |= 0xc << 28;
390 srcId(i
->src(0), 23);
398 // 0x2 for GPR, c[] and 0x1 for short immediate
400 CodeEmitterGK110::emitForm_21(const Instruction
*i
, uint32_t opc2
,
403 const bool imm
= i
->srcExists(1) && i
->src(1).getFile() == FILE_IMMEDIATE
;
406 if (i
->srcExists(2) && i
->src(2).getFile() == FILE_MEMORY_CONST
)
411 code
[1] = opc1
<< 20;
414 code
[1] = (0xc << 28) | (opc2
<< 20);
421 for (int s
= 0; s
< 3 && i
->srcExists(s
); ++s
) {
422 switch (i
->src(s
).getFile()) {
423 case FILE_MEMORY_CONST
:
424 code
[1] &= (s
== 2) ? ~(0x4 << 28) : ~(0x8 << 28);
425 setCAddress14(i
->src(s
));
428 setShortImmediate(i
, s
);
431 srcId(i
->src(s
), s
? ((s
== 2) ? 42 : s1
) : 10);
434 // ignore here, can be predicate or flags, but must not be address
442 assert(imm
|| (code
[1] & (0xc << 28)));
446 CodeEmitterGK110::modNegAbsF32_3b(const Instruction
*i
, const int s
)
448 if (i
->src(s
).mod
.abs()) code
[1] &= ~(1 << 27);
449 if (i
->src(s
).mod
.neg()) code
[1] ^= (1 << 27);
453 CodeEmitterGK110::emitNOP(const Instruction
*i
)
455 code
[0] = 0x00003c02;
456 code
[1] = 0x85800000;
461 code
[0] = 0x001c3c02;
465 CodeEmitterGK110::emitFMAD(const Instruction
*i
)
467 assert(!isLIMM(i
->src(1), TYPE_F32
));
469 emitForm_21(i
, 0x0c0, 0x940);
477 bool neg1
= (i
->src(0).mod
^ i
->src(1).mod
).neg();
489 CodeEmitterGK110::emitDMAD(const Instruction
*i
)
491 assert(!i
->saturate
);
494 emitForm_21(i
, 0x1b8, 0xb38);
499 bool neg1
= (i
->src(0).mod
^ i
->src(1).mod
).neg();
511 CodeEmitterGK110::emitFMUL(const Instruction
*i
)
513 bool neg
= (i
->src(0).mod
^ i
->src(1).mod
).neg();
515 assert(i
->postFactor
>= -3 && i
->postFactor
<= 3);
517 if (isLIMM(i
->src(1), TYPE_F32
)) {
518 emitForm_L(i
, 0x200, 0x2, Modifier(0));
526 assert(i
->postFactor
== 0);
528 emitForm_21(i
, 0x234, 0xc34);
529 code
[1] |= ((i
->postFactor
> 0) ?
530 (7 - i
->postFactor
) : (0 - i
->postFactor
)) << 12;
548 CodeEmitterGK110::emitDMUL(const Instruction
*i
)
550 bool neg
= (i
->src(0).mod
^ i
->src(1).mod
).neg();
552 assert(!i
->postFactor
);
553 assert(!i
->saturate
);
557 emitForm_21(i
, 0x240, 0xc40);
571 CodeEmitterGK110::emitIMUL(const Instruction
*i
)
573 assert(!i
->src(0).mod
.neg() && !i
->src(1).mod
.neg());
574 assert(!i
->src(0).mod
.abs() && !i
->src(1).mod
.abs());
576 if (isLIMM(i
->src(1), TYPE_S32
)) {
577 emitForm_L(i
, 0x280, 2, Modifier(0));
579 if (i
->subOp
== NV50_IR_SUBOP_MUL_HIGH
)
581 if (i
->sType
== TYPE_S32
)
584 emitForm_21(i
, 0x21c, 0xc1c);
586 if (i
->subOp
== NV50_IR_SUBOP_MUL_HIGH
)
588 if (i
->sType
== TYPE_S32
)
594 CodeEmitterGK110::emitFADD(const Instruction
*i
)
596 if (isLIMM(i
->src(1), TYPE_F32
)) {
597 assert(i
->rnd
== ROUND_N
);
598 assert(!i
->saturate
);
600 Modifier mod
= i
->src(1).mod
^
601 Modifier(i
->op
== OP_SUB
? NV50_IR_MOD_NEG
: 0);
603 emitForm_L(i
, 0x400, 0, mod
);
609 emitForm_21(i
, 0x22c, 0xc2c);
618 modNegAbsF32_3b(i
, 1);
619 if (i
->op
== OP_SUB
) code
[1] ^= 1 << 27;
623 if (i
->op
== OP_SUB
) code
[1] ^= 1 << 16;
629 CodeEmitterGK110::emitDADD(const Instruction
*i
)
631 assert(!i
->saturate
);
634 emitForm_21(i
, 0x238, 0xc38);
639 modNegAbsF32_3b(i
, 1);
640 if (i
->op
== OP_SUB
) code
[1] ^= 1 << 27;
644 if (i
->op
== OP_SUB
) code
[1] ^= 1 << 16;
649 CodeEmitterGK110::emitUADD(const Instruction
*i
)
651 uint8_t addOp
= (i
->src(0).mod
.neg() << 1) | i
->src(1).mod
.neg();
656 assert(!i
->src(0).mod
.abs() && !i
->src(1).mod
.abs());
658 if (isLIMM(i
->src(1), TYPE_S32
)) {
659 emitForm_L(i
, 0x400, 1, Modifier((addOp
& 1) ? NV50_IR_MOD_NEG
: 0));
664 assert(!i
->defExists(1));
665 assert(i
->flagsSrc
< 0);
669 emitForm_21(i
, 0x208, 0xc08);
671 assert(addOp
!= 3); // would be add-plus-one
673 code
[1] |= addOp
<< 19;
676 code
[1] |= 1 << 18; // write carry
677 if (i
->flagsSrc
>= 0)
678 code
[1] |= 1 << 14; // add carry
686 CodeEmitterGK110::emitIMAD(const Instruction
*i
)
689 (i
->src(2).mod
.neg() << 1) | (i
->src(0).mod
.neg() ^ i
->src(1).mod
.neg());
691 emitForm_21(i
, 0x100, 0xa00);
694 code
[1] |= addOp
<< 26;
696 if (i
->sType
== TYPE_S32
)
697 code
[1] |= (1 << 19) | (1 << 24);
703 if (i
->subOp
== NV50_IR_SUBOP_MUL_HIGH
)
710 CodeEmitterGK110::emitISAD(const Instruction
*i
)
712 assert(i
->dType
== TYPE_S32
|| i
->dType
== TYPE_U32
);
714 emitForm_21(i
, 0x1f4, 0xb74);
716 if (i
->dType
== TYPE_S32
)
721 CodeEmitterGK110::emitNOT(const Instruction
*i
)
723 code
[0] = 0x0003fc02; // logop(mov2) dst, 0, not src
724 code
[1] = 0x22003800;
730 switch (i
->src(0).getFile()) {
732 code
[1] |= 0xc << 28;
733 srcId(i
->src(0), 23);
735 case FILE_MEMORY_CONST
:
736 code
[1] |= 0x4 << 28;
737 setCAddress14(i
->src(1));
746 CodeEmitterGK110::emitLogicOp(const Instruction
*i
, uint8_t subOp
)
748 if (isLIMM(i
->src(1), TYPE_S32
)) {
749 emitForm_L(i
, 0x200, 0, i
->src(1).mod
);
750 code
[1] |= subOp
<< 24;
753 emitForm_21(i
, 0x220, 0xc20);
754 code
[1] |= subOp
<< 12;
761 CodeEmitterGK110::emitPOPC(const Instruction
*i
)
763 assert(!isLIMM(i
->src(1), TYPE_S32
, true));
765 emitForm_21(i
, 0x204, 0xc04);
768 if (!(code
[0] & 0x1))
773 CodeEmitterGK110::emitINSBF(const Instruction
*i
)
775 emitForm_21(i
, 0x1f8, 0xb78);
779 CodeEmitterGK110::emitEXTBF(const Instruction
*i
)
781 emitForm_21(i
, 0x600, 0xc00);
783 if (i
->dType
== TYPE_S32
)
785 if (i
->subOp
== NV50_IR_SUBOP_EXTBF_REV
)
790 CodeEmitterGK110::emitBFIND(const Instruction
*i
)
792 emitForm_C(i
, 0x218, 0x2);
794 if (i
->dType
== TYPE_S32
)
796 if (i
->src(0).mod
== Modifier(NV50_IR_MOD_NOT
))
798 if (i
->subOp
== NV50_IR_SUBOP_BFIND_SAMT
)
803 CodeEmitterGK110::emitShift(const Instruction
*i
)
805 if (i
->op
== OP_SHR
) {
806 emitForm_21(i
, 0x214, 0xc14);
807 if (isSignedType(i
->dType
))
810 emitForm_21(i
, 0x224, 0xc24);
813 if (i
->subOp
== NV50_IR_SUBOP_SHIFT_WRAP
)
818 CodeEmitterGK110::emitPreOp(const Instruction
*i
)
820 emitForm_C(i
, 0x248, 0x2);
822 if (i
->op
== OP_PREEX2
)
830 CodeEmitterGK110::emitSFnOp(const Instruction
*i
, uint8_t subOp
)
832 code
[0] = 0x00000002 | (subOp
<< 23);
833 code
[1] = 0x84000000;
838 srcId(i
->src(0), 10);
846 CodeEmitterGK110::emitMINMAX(const Instruction
*i
)
870 emitForm_21(i
, op2
, op1
);
872 if (i
->dType
== TYPE_S32
)
874 code
[1] |= (i
->op
== OP_MIN
) ? 0x1c00 : 0x3c00; // [!]pt
880 modNegAbsF32_3b(i
, 1);
888 CodeEmitterGK110::emitCVT(const Instruction
*i
)
890 const bool f2f
= isFloatType(i
->dType
) && isFloatType(i
->sType
);
891 const bool f2i
= !isFloatType(i
->dType
) && isFloatType(i
->sType
);
892 const bool i2f
= isFloatType(i
->dType
) && !isFloatType(i
->sType
);
894 bool sat
= i
->saturate
;
895 bool abs
= i
->src(0).mod
.abs();
896 bool neg
= i
->src(0).mod
.neg();
898 RoundMode rnd
= i
->rnd
;
901 case OP_CEIL
: rnd
= f2f
? ROUND_PI
: ROUND_P
; break;
902 case OP_FLOOR
: rnd
= f2f
? ROUND_MI
: ROUND_M
; break;
903 case OP_TRUNC
: rnd
= f2f
? ROUND_ZI
: ROUND_Z
; break;
904 case OP_SAT
: sat
= true; break;
905 case OP_NEG
: neg
= !neg
; break;
906 case OP_ABS
: abs
= true; neg
= false; break;
913 if (i
->op
== OP_NEG
&& i
->dType
== TYPE_U32
)
922 else if (f2i
) op
= 0x258;
923 else if (i2f
) op
= 0x25c;
926 emitForm_C(i
, op
, 0x2);
929 if (neg
) code
[1] |= 1 << 16;
930 if (abs
) code
[1] |= 1 << 20;
931 if (sat
) code
[1] |= 1 << 21;
933 emitRoundMode(rnd
, 32 + 10, f2f
? (32 + 13) : -1);
935 code
[0] |= typeSizeofLog2(dType
) << 10;
936 code
[0] |= typeSizeofLog2(i
->sType
) << 12;
937 code
[1] |= i
->subOp
<< 12;
939 if (isSignedIntType(dType
))
941 if (isSignedIntType(i
->sType
))
946 CodeEmitterGK110::emitSET(const CmpInstruction
*i
)
950 if (i
->def(0).getFile() == FILE_PREDICATE
) {
952 case TYPE_F32
: op2
= 0x1d8; op1
= 0xb58; break;
953 case TYPE_F64
: op2
= 0x1c0; op1
= 0xb40; break;
959 emitForm_21(i
, op2
, op1
);
963 if (!(code
[0] & 0x1)) {
967 modNegAbsF32_3b(i
, 1);
971 // normal DST field is negated predicate result
972 code
[0] = (code
[0] & ~0xfc) | ((code
[0] << 3) & 0xe0);
979 case TYPE_F32
: op2
= 0x000; op1
= 0x800; break;
980 case TYPE_F64
: op2
= 0x080; op1
= 0x900; break;
986 emitForm_21(i
, op2
, op1
);
990 if (!(code
[0] & 0x1)) {
994 modNegAbsF32_3b(i
, 1);
998 if (i
->dType
== TYPE_F32
) {
999 if (isFloatType(i
->sType
))
1005 if (i
->sType
== TYPE_S32
)
1008 if (i
->op
!= OP_SET
) {
1010 case OP_SET_AND
: code
[1] |= 0x0 << 16; break;
1011 case OP_SET_OR
: code
[1] |= 0x1 << 16; break;
1012 case OP_SET_XOR
: code
[1] |= 0x2 << 16; break;
1017 srcId(i
->src(2), 0x2a);
1019 code
[1] |= 0x7 << 10;
1021 emitCondCode(i
->setCond
,
1022 isFloatType(i
->sType
) ? 0x33 : 0x34,
1023 isFloatType(i
->sType
) ? 0xf : 0x7);
1027 CodeEmitterGK110::emitSLCT(const CmpInstruction
*i
)
1029 CondCode cc
= i
->setCond
;
1030 if (i
->src(2).mod
.neg())
1031 cc
= reverseCondCode(cc
);
1033 if (i
->dType
== TYPE_F32
) {
1034 emitForm_21(i
, 0x1d0, 0xb50);
1036 emitCondCode(cc
, 0x33, 0xf);
1038 emitForm_21(i
, 0x1a0, 0xb20);
1039 emitCondCode(cc
, 0x34, 0x7);
1043 void CodeEmitterGK110::emitSELP(const Instruction
*i
)
1045 emitForm_21(i
, 0x250, 0x050);
1047 if ((i
->cc
== CC_NOT_P
) ^ (bool)(i
->src(2).mod
& Modifier(NV50_IR_MOD_NOT
)))
1051 void CodeEmitterGK110::emitTEXBAR(const Instruction
*i
)
1053 code
[0] = 0x0000003e | (i
->subOp
<< 23);
1054 code
[1] = 0x77000000;
1059 void CodeEmitterGK110::emitTEXCSAA(const TexInstruction
*i
)
1061 code
[0] = 0x00000002;
1062 code
[1] = 0x76c00000;
1064 code
[1] |= i
->tex
.r
<< 9;
1065 // code[1] |= i->tex.s << (9 + 8);
1067 if (i
->tex
.liveOnly
)
1068 code
[0] |= 0x80000000;
1070 defId(i
->def(0), 2);
1071 srcId(i
->src(0), 10);
1075 isNextIndependentTex(const TexInstruction
*i
)
1077 if (!i
->next
|| !isTextureOp(i
->next
->op
))
1079 if (i
->getDef(0)->interfers(i
->next
->getSrc(0)))
1081 return !i
->next
->srcExists(1) || !i
->getDef(0)->interfers(i
->next
->getSrc(1));
1085 CodeEmitterGK110::emitTEX(const TexInstruction
*i
)
1087 const bool ind
= i
->tex
.rIndirectSrc
>= 0;
1090 code
[0] = 0x00000002;
1093 code
[1] = 0x7e000000;
1096 code
[1] = 0x7e800000;
1099 code
[1] = 0x78000000;
1102 code
[1] = 0x7dc00000;
1105 code
[1] = 0x7d800000;
1111 code
[0] = 0x00000002;
1112 code
[1] = 0x76000000;
1113 code
[1] |= i
->tex
.r
<< 9;
1116 code
[0] = 0x00000002;
1117 code
[1] = 0x76800000;
1118 code
[1] |= i
->tex
.r
<< 9;
1121 code
[0] = 0x00000002;
1122 code
[1] = 0x70000000;
1123 code
[1] |= i
->tex
.r
<< 13;
1126 code
[0] = 0x00000001;
1127 code
[1] = 0x70000000;
1128 code
[1] |= i
->tex
.r
<< 15;
1131 code
[0] = 0x00000001;
1132 code
[1] = 0x60000000;
1133 code
[1] |= i
->tex
.r
<< 15;
1138 code
[1] |= isNextIndependentTex(i
) ? 0x1 : 0x2; // t : p mode
1140 if (i
->tex
.liveOnly
)
1141 code
[0] |= 0x80000000;
1145 case OP_TXB
: code
[1] |= 0x2000; break;
1146 case OP_TXL
: code
[1] |= 0x3000; break;
1150 case OP_TXLQ
: break;
1152 assert(!"invalid texture op");
1156 if (i
->op
== OP_TXF
) {
1157 if (!i
->tex
.levelZero
)
1160 if (i
->tex
.levelZero
) {
1164 if (i
->op
!= OP_TXD
&& i
->tex
.derivAll
)
1169 code
[1] |= i
->tex
.mask
<< 2;
1171 const int src1
= (i
->predSrc
== 1) ? 2 : 1; // if predSrc == 1, !srcExists(2)
1173 defId(i
->def(0), 2);
1174 srcId(i
->src(0), 10);
1177 if (i
->op
== OP_TXG
) code
[1] |= i
->tex
.gatherComp
<< 13;
1180 code
[1] |= (i
->tex
.target
.isCube() ? 3 : (i
->tex
.target
.getDim() - 1)) << 7;
1181 if (i
->tex
.target
.isArray())
1183 if (i
->tex
.target
.isShadow())
1185 if (i
->tex
.target
== TEX_TARGET_2D_MS
||
1186 i
->tex
.target
== TEX_TARGET_2D_MS_ARRAY
)
1189 if (i
->srcExists(src1
) && i
->src(src1
).getFile() == FILE_IMMEDIATE
) {
1193 if (i
->tex
.useOffsets
== 1) {
1195 case OP_TXF
: code
[1] |= 0x200; break;
1196 case OP_TXD
: code
[1] |= 0x00400000; break;
1197 default: code
[1] |= 0x800; break;
1200 if (i
->tex
.useOffsets
== 4)
1205 CodeEmitterGK110::emitTXQ(const TexInstruction
*i
)
1207 code
[0] = 0x00000002;
1208 code
[1] = 0x75400001;
1210 switch (i
->tex
.query
) {
1211 case TXQ_DIMS
: code
[0] |= 0x01 << 25; break;
1212 case TXQ_TYPE
: code
[0] |= 0x02 << 25; break;
1213 case TXQ_SAMPLE_POSITION
: code
[0] |= 0x05 << 25; break;
1214 case TXQ_FILTER
: code
[0] |= 0x10 << 25; break;
1215 case TXQ_LOD
: code
[0] |= 0x12 << 25; break;
1216 case TXQ_BORDER_COLOUR
: code
[0] |= 0x16 << 25; break;
1218 assert(!"invalid texture query");
1222 code
[1] |= i
->tex
.mask
<< 2;
1223 code
[1] |= i
->tex
.r
<< 9;
1224 if (/*i->tex.sIndirectSrc >= 0 || */i
->tex
.rIndirectSrc
>= 0)
1225 code
[1] |= 0x08000000;
1227 defId(i
->def(0), 2);
1228 srcId(i
->src(0), 10);
1234 CodeEmitterGK110::emitQUADOP(const Instruction
*i
, uint8_t qOp
, uint8_t laneMask
)
1236 code
[0] = 0x00000002 | ((qOp
& 1) << 31);
1237 code
[1] = 0x7fc00000 | (qOp
>> 1) | (laneMask
<< 12);
1239 defId(i
->def(0), 2);
1240 srcId(i
->src(0), 10);
1241 srcId(i
->srcExists(1) ? i
->src(1) : i
->src(0), 23);
1243 if (i
->op
== OP_QUADOP
&& progType
!= Program::TYPE_FRAGMENT
)
1244 code
[1] |= 1 << 9; // dall
1250 CodeEmitterGK110::emitPIXLD(const Instruction
*i
)
1252 emitForm_L(i
, 0x7f4, 2, Modifier(0));
1253 code
[1] |= i
->subOp
<< 2;
1254 code
[1] |= 0x00070000;
1258 CodeEmitterGK110::emitBAR(const Instruction
*i
)
1265 CodeEmitterGK110::emitFlow(const Instruction
*i
)
1267 const FlowInstruction
*f
= i
->asFlow();
1269 unsigned mask
; // bit 0: predicate, bit 1: target
1271 code
[0] = 0x00000000;
1275 code
[1] = f
->absolute
? 0x10800000 : 0x12000000;
1276 if (i
->srcExists(0) && i
->src(0).getFile() == FILE_MEMORY_CONST
)
1281 code
[1] = f
->absolute
? 0x11000000 : 0x13000000;
1282 if (i
->srcExists(0) && i
->src(0).getFile() == FILE_MEMORY_CONST
)
1287 case OP_EXIT
: code
[1] = 0x18000000; mask
= 1; break;
1288 case OP_RET
: code
[1] = 0x19000000; mask
= 1; break;
1289 case OP_DISCARD
: code
[1] = 0x19800000; mask
= 1; break;
1290 case OP_BREAK
: code
[1] = 0x1a000000; mask
= 1; break;
1291 case OP_CONT
: code
[1] = 0x1a800000; mask
= 1; break;
1293 case OP_JOINAT
: code
[1] = 0x14800000; mask
= 2; break;
1294 case OP_PREBREAK
: code
[1] = 0x15000000; mask
= 2; break;
1295 case OP_PRECONT
: code
[1] = 0x15800000; mask
= 2; break;
1296 case OP_PRERET
: code
[1] = 0x13800000; mask
= 2; break;
1298 case OP_QUADON
: code
[1] = 0x1b800000; mask
= 0; break;
1299 case OP_QUADPOP
: code
[1] = 0x1c000000; mask
= 0; break;
1300 case OP_BRKPT
: code
[1] = 0x00000000; mask
= 0; break;
1302 assert(!"invalid flow operation");
1308 if (i
->flagsSrc
< 0)
1320 if (f
->op
== OP_CALL
) {
1322 assert(f
->absolute
);
1323 uint32_t pcAbs
= targNVC0
->getBuiltinOffset(f
->target
.builtin
);
1324 addReloc(RelocEntry::TYPE_BUILTIN
, 0, pcAbs
, 0xff800000, 23);
1325 addReloc(RelocEntry::TYPE_BUILTIN
, 1, pcAbs
, 0x007fffff, -9);
1327 assert(!f
->absolute
);
1328 int32_t pcRel
= f
->target
.fn
->binPos
- (codeSize
+ 8);
1329 code
[0] |= (pcRel
& 0x1ff) << 23;
1330 code
[1] |= (pcRel
>> 9) & 0x7fff;
1334 int32_t pcRel
= f
->target
.bb
->binPos
- (codeSize
+ 8);
1335 if (writeIssueDelays
&& !(f
->target
.bb
->binPos
& 0x3f))
1337 // currently we don't want absolute branches
1338 assert(!f
->absolute
);
1339 code
[0] |= (pcRel
& 0x1ff) << 23;
1340 code
[1] |= (pcRel
>> 9) & 0x7fff;
1345 CodeEmitterGK110::emitAFETCH(const Instruction
*i
)
1347 uint32_t offset
= i
->src(0).get()->reg
.data
.offset
& 0x7ff;
1349 code
[0] = 0x00000002 | (offset
<< 23);
1350 code
[1] = 0x7d000000 | (offset
>> 9);
1352 if (i
->getSrc(0)->reg
.file
== FILE_SHADER_OUTPUT
)
1357 defId(i
->def(0), 2);
1358 srcId(i
->src(0).getIndirect(0), 10);
1362 CodeEmitterGK110::emitPFETCH(const Instruction
*i
)
1364 uint32_t prim
= i
->src(0).get()->reg
.data
.u32
;
1366 code
[0] = 0x00000002 | ((prim
& 0xff) << 23);
1367 code
[1] = 0x7f800000;
1371 const int src1
= (i
->predSrc
== 1) ? 2 : 1; // if predSrc == 1, !srcExists(2)
1373 defId(i
->def(0), 2);
1378 CodeEmitterGK110::emitVFETCH(const Instruction
*i
)
1380 unsigned int size
= typeSizeof(i
->dType
);
1381 uint32_t offset
= i
->src(0).get()->reg
.data
.offset
;
1383 code
[0] = 0x00000002 | (offset
<< 23);
1384 code
[1] = 0x7ec00000 | (offset
>> 9);
1385 code
[1] |= (size
/ 4 - 1) << 18;
1389 if (i
->getSrc(0)->reg
.file
== FILE_SHADER_OUTPUT
)
1390 code
[1] |= 0x8; // yes, TCPs can read from *outputs* of other threads
1394 defId(i
->def(0), 2);
1395 srcId(i
->src(0).getIndirect(0), 10);
1396 srcId(i
->src(0).getIndirect(1), 32 + 10); // vertex address
1400 CodeEmitterGK110::emitEXPORT(const Instruction
*i
)
1402 unsigned int size
= typeSizeof(i
->dType
);
1403 uint32_t offset
= i
->src(0).get()->reg
.data
.offset
;
1405 code
[0] = 0x00000002 | (offset
<< 23);
1406 code
[1] = 0x7f000000 | (offset
>> 9);
1407 code
[1] |= (size
/ 4 - 1) << 18;
1414 assert(i
->src(1).getFile() == FILE_GPR
);
1416 srcId(i
->src(0).getIndirect(0), 10);
1417 srcId(i
->src(0).getIndirect(1), 32 + 10); // vertex base address
1418 srcId(i
->src(1), 2);
1422 CodeEmitterGK110::emitOUT(const Instruction
*i
)
1424 assert(i
->src(0).getFile() == FILE_GPR
);
1426 emitForm_21(i
, 0x1f0, 0xb70);
1428 if (i
->op
== OP_EMIT
)
1430 if (i
->op
== OP_RESTART
|| i
->subOp
== NV50_IR_SUBOP_EMIT_RESTART
)
1435 CodeEmitterGK110::emitInterpMode(const Instruction
*i
)
1437 code
[1] |= (i
->ipa
& 0x3) << 21; // TODO: INTERP_SAMPLEID
1438 code
[1] |= (i
->ipa
& 0xc) << (19 - 2);
1442 interpApply(const InterpEntry
*entry
, uint32_t *code
,
1443 bool force_persample_interp
, bool flatshade
)
1445 int ipa
= entry
->ipa
;
1446 int reg
= entry
->reg
;
1447 int loc
= entry
->loc
;
1450 (ipa
& NV50_IR_INTERP_MODE_MASK
) == NV50_IR_INTERP_SC
) {
1451 ipa
= NV50_IR_INTERP_FLAT
;
1453 } else if (force_persample_interp
&&
1454 (ipa
& NV50_IR_INTERP_SAMPLE_MASK
) == NV50_IR_INTERP_DEFAULT
&&
1455 (ipa
& NV50_IR_INTERP_MODE_MASK
) != NV50_IR_INTERP_FLAT
) {
1456 ipa
|= NV50_IR_INTERP_CENTROID
;
1458 code
[loc
+ 1] &= ~(0xf << 19);
1459 code
[loc
+ 1] |= (ipa
& 0x3) << 21;
1460 code
[loc
+ 1] |= (ipa
& 0xc) << (19 - 2);
1461 code
[loc
+ 0] &= ~(0xff << 23);
1462 code
[loc
+ 0] |= reg
<< 23;
1466 CodeEmitterGK110::emitINTERP(const Instruction
*i
)
1468 const uint32_t base
= i
->getSrc(0)->reg
.data
.offset
;
1470 code
[0] = 0x00000002 | (base
<< 31);
1471 code
[1] = 0x74800000 | (base
>> 1);
1476 if (i
->op
== OP_PINTERP
) {
1477 srcId(i
->src(1), 23);
1478 addInterp(i
->ipa
, SDATA(i
->src(1)).id
, interpApply
);
1480 code
[0] |= 0xff << 23;
1481 addInterp(i
->ipa
, 0xff, interpApply
);
1484 srcId(i
->src(0).getIndirect(0), 10);
1488 defId(i
->def(0), 2);
1490 if (i
->getSampleMode() == NV50_IR_INTERP_OFFSET
)
1491 srcId(i
->src(i
->op
== OP_PINTERP
? 2 : 1), 32 + 10);
1493 code
[1] |= 0xff << 10;
1497 CodeEmitterGK110::emitLoadStoreType(DataType ty
, const int pos
)
1529 assert(!"invalid ld/st type");
1532 code
[pos
/ 32] |= n
<< (pos
% 32);
1536 CodeEmitterGK110::emitCachingMode(CacheMode c
, const int pos
)
1557 assert(!"invalid caching mode");
1560 code
[pos
/ 32] |= n
<< (pos
% 32);
1564 CodeEmitterGK110::emitSTORE(const Instruction
*i
)
1566 int32_t offset
= SDATA(i
->src(0)).offset
;
1568 switch (i
->src(0).getFile()) {
1569 case FILE_MEMORY_GLOBAL
: code
[1] = 0xe0000000; code
[0] = 0x00000000; break;
1570 case FILE_MEMORY_LOCAL
: code
[1] = 0x7a800000; code
[0] = 0x00000002; break;
1571 case FILE_MEMORY_SHARED
: code
[1] = 0x7ac00000; code
[0] = 0x00000002; break;
1573 assert(!"invalid memory file");
1577 if (i
->src(0).getFile() != FILE_MEMORY_GLOBAL
)
1580 if (code
[0] & 0x2) {
1581 emitLoadStoreType(i
->dType
, 0x33);
1582 if (i
->src(0).getFile() == FILE_MEMORY_LOCAL
)
1583 emitCachingMode(i
->cache
, 0x2f);
1585 emitLoadStoreType(i
->dType
, 0x38);
1586 emitCachingMode(i
->cache
, 0x3b);
1588 code
[0] |= offset
<< 23;
1589 code
[1] |= offset
>> 9;
1593 srcId(i
->src(1), 2);
1594 srcId(i
->src(0).getIndirect(0), 10);
1598 CodeEmitterGK110::emitLOAD(const Instruction
*i
)
1600 int32_t offset
= SDATA(i
->src(0)).offset
;
1602 switch (i
->src(0).getFile()) {
1603 case FILE_MEMORY_GLOBAL
: code
[1] = 0xc0000000; code
[0] = 0x00000000; break;
1604 case FILE_MEMORY_LOCAL
: code
[1] = 0x7a000000; code
[0] = 0x00000002; break;
1605 case FILE_MEMORY_SHARED
: code
[1] = 0x7ac00000; code
[0] = 0x00000002; break;
1606 case FILE_MEMORY_CONST
:
1607 if (!i
->src(0).isIndirect(0) && typeSizeof(i
->dType
) == 4) {
1612 code
[0] = 0x00000002;
1613 code
[1] = 0x7c800000 | (i
->src(0).get()->reg
.fileIndex
<< 7);
1614 code
[1] |= i
->subOp
<< 15;
1617 assert(!"invalid memory file");
1621 if (code
[0] & 0x2) {
1623 emitLoadStoreType(i
->dType
, 0x33);
1624 if (i
->src(0).getFile() == FILE_MEMORY_LOCAL
)
1625 emitCachingMode(i
->cache
, 0x2f);
1627 emitLoadStoreType(i
->dType
, 0x38);
1628 emitCachingMode(i
->cache
, 0x3b);
1630 code
[0] |= offset
<< 23;
1631 code
[1] |= offset
>> 9;
1635 defId(i
->def(0), 2);
1636 srcId(i
->src(0).getIndirect(0), 10);
1640 CodeEmitterGK110::getSRegEncoding(const ValueRef
& ref
)
1642 switch (SDATA(ref
).sv
.sv
) {
1643 case SV_LANEID
: return 0x00;
1644 case SV_PHYSID
: return 0x03;
1645 case SV_VERTEX_COUNT
: return 0x10;
1646 case SV_INVOCATION_ID
: return 0x11;
1647 case SV_YDIR
: return 0x12;
1648 case SV_THREAD_KILL
: return 0x13;
1649 case SV_TID
: return 0x21 + SDATA(ref
).sv
.index
;
1650 case SV_CTAID
: return 0x25 + SDATA(ref
).sv
.index
;
1651 case SV_NTID
: return 0x29 + SDATA(ref
).sv
.index
;
1652 case SV_GRIDID
: return 0x2c;
1653 case SV_NCTAID
: return 0x2d + SDATA(ref
).sv
.index
;
1654 case SV_LBASE
: return 0x34;
1655 case SV_SBASE
: return 0x30;
1656 case SV_CLOCK
: return 0x50 + SDATA(ref
).sv
.index
;
1658 assert(!"no sreg for system value");
1664 CodeEmitterGK110::emitMOV(const Instruction
*i
)
1666 if (i
->src(0).getFile() == FILE_SYSTEM_VALUE
) {
1667 code
[0] = 0x00000002 | (getSRegEncoding(i
->src(0)) << 23);
1668 code
[1] = 0x86400000;
1670 defId(i
->def(0), 2);
1672 if (i
->src(0).getFile() == FILE_IMMEDIATE
) {
1673 code
[0] = 0x00000002 | (i
->lanes
<< 14);
1674 code
[1] = 0x74000000;
1676 defId(i
->def(0), 2);
1677 setImmediate32(i
, 0, Modifier(0));
1679 if (i
->src(0).getFile() == FILE_PREDICATE
) {
1680 code
[0] = 0x00000002;
1681 code
[1] = 0x84401c07;
1683 defId(i
->def(0), 2);
1684 srcId(i
->src(0), 14);
1686 emitForm_C(i
, 0x24c, 2);
1687 code
[1] |= i
->lanes
<< 10;
1691 void CodeEmitterGK110::emitMEMBAR(const Instruction
*i
)
1693 code
[0] = 0x00000002 | NV50_IR_SUBOP_MEMBAR_SCOPE(i
->subOp
) << 8;
1694 code
[1] = 0x7cc00000;
1700 CodeEmitterGK110::emitInstruction(Instruction
*insn
)
1702 const unsigned int size
= (writeIssueDelays
&& !(codeSize
& 0x3f)) ? 16 : 8;
1704 if (insn
->encSize
!= 8) {
1705 ERROR("skipping unencodable instruction: ");
1709 if (codeSize
+ size
> codeSizeLimit
) {
1710 ERROR("code emitter output buffer too small\n");
1714 if (writeIssueDelays
) {
1715 int id
= (codeSize
& 0x3f) / 8 - 1;
1718 code
[0] = 0x00000000; // cf issue delay "instruction"
1719 code
[1] = 0x08000000;
1723 uint32_t *data
= code
- (id
* 2 + 2);
1726 case 0: data
[0] |= insn
->sched
<< 2; break;
1727 case 1: data
[0] |= insn
->sched
<< 10; break;
1728 case 2: data
[0] |= insn
->sched
<< 18; break;
1729 case 3: data
[0] |= insn
->sched
<< 26; data
[1] |= insn
->sched
>> 6; break;
1730 case 4: data
[1] |= insn
->sched
<< 2; break;
1731 case 5: data
[1] |= insn
->sched
<< 10; break;
1732 case 6: data
[1] |= insn
->sched
<< 18; break;
1739 // assert that instructions with multiple defs don't corrupt registers
1740 for (int d
= 0; insn
->defExists(d
); ++d
)
1741 assert(insn
->asTex() || insn
->def(d
).rep()->reg
.data
.id
>= 0);
1778 if (insn
->dType
== TYPE_F64
)
1780 else if (isFloatType(insn
->dType
))
1786 if (insn
->dType
== TYPE_F64
)
1788 else if (isFloatType(insn
->dType
))
1795 if (insn
->dType
== TYPE_F64
)
1797 else if (isFloatType(insn
->dType
))
1809 emitLogicOp(insn
, 0);
1812 emitLogicOp(insn
, 1);
1815 emitLogicOp(insn
, 2);
1825 emitSET(insn
->asCmp());
1831 emitSLCT(insn
->asCmp());
1847 emitSFnOp(insn
, 5 + 2 * insn
->subOp
);
1850 emitSFnOp(insn
, 4 + 2 * insn
->subOp
);
1875 emitTEX(insn
->asTex());
1878 emitTXQ(insn
->asTex());
1903 emitQUADOP(insn
, insn
->subOp
, insn
->lanes
);
1906 emitQUADOP(insn
, insn
->src(0).mod
.neg() ? 0x66 : 0x99, 0x4);
1909 emitQUADOP(insn
, insn
->src(0).mod
.neg() ? 0x5a : 0xa5, 0x5);
1936 ERROR("operation should have been eliminated");
1942 ERROR("operation should have been lowered\n");
1945 ERROR("unknown op: %u\n", insn
->op
);
1958 CodeEmitterGK110::getMinEncodingSize(const Instruction
*i
) const
1960 // No more short instruction encodings.
1965 CodeEmitterGK110::prepareEmission(Function
*func
)
1967 const Target
*targ
= func
->getProgram()->getTarget();
1969 CodeEmitter::prepareEmission(func
);
1971 if (targ
->hasSWSched
)
1972 calculateSchedDataNVC0(targ
, func
);
1975 CodeEmitterGK110::CodeEmitterGK110(const TargetNVC0
*target
)
1976 : CodeEmitter(target
),
1978 writeIssueDelays(target
->hasSWSched
)
1981 codeSize
= codeSizeLimit
= 0;
1986 TargetNVC0::createCodeEmitterGK110(Program::Type type
)
1988 CodeEmitterGK110
*emit
= new CodeEmitterGK110(this);
1989 emit
->setProgramType(type
);
1993 } // namespace nv50_ir