2 * Copyright 2011 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.h"
24 #include "codegen/nv50_ir_target.h"
28 const uint8_t Target::operationSrcNr
[OP_LAST
+ 1] =
31 0, 0, 0, 0, // UNION, SPLIT, MERGE, CONSTRAINT
32 1, 1, 2, // MOV, LOAD, STORE
33 2, 2, 2, 2, 2, 3, 3, 3, // ADD, SUB, MUL, DIV, MOD, MAD, FMA, SAD
34 1, 1, 1, // ABS, NEG, NOT
35 2, 2, 2, 2, 2, // AND, OR, XOR, SHL, SHR
36 2, 2, 1, // MAX, MIN, SAT
37 1, 1, 1, 1, // CEIL, FLOOR, TRUNC, CVT
38 3, 3, 3, 2, 3, 3, // SET_AND,OR,XOR, SET, SELP, SLCT
39 1, 1, 1, 1, 1, 1, // RCP, RSQ, LG2, SIN, COS, EX2
40 1, 1, 1, 1, 1, 2, // EXP, LOG, PRESIN, PREEX2, SQRT, POW
41 0, 0, 0, 0, 0, // BRA, CALL, RET, CONT, BREAK,
42 0, 0, 0, // PRERET,CONT,BREAK
43 0, 0, 0, 0, 0, 0, // BRKPT, JOINAT, JOIN, DISCARD, EXIT, MEMBAR
44 1, 1, 2, 1, 2, // VFETCH, PFETCH, EXPORT, LINTERP, PINTERP
45 1, 1, // EMIT, RESTART
46 1, 1, 1, // TEX, TXB, TXL,
47 1, 1, 1, 1, 1, 2, // TXF, TXQ, TXD, TXG, TEXCSAA, TEXPREP
48 1, 1, 2, 2, 2, 2, 2, // SULDB, SULDP, SUSTB, SUSTP, SUREDB, SUREDP, SULEA
49 3, 3, 3, 3, // SUBFM, SUCLAMP, SUEAU, MADSP
52 1, 2, 2, 0, 0, // RDSV, WRSV, QUADOP, QUADON, QUADPOP
53 2, 3, 2, 3, // POPCNT, INSBF, EXTBF, PERMT
55 2, 2, 2, 2, 3, 2, // VADD, VAVG, VMIN, VMAX, VSAD, VSET,
56 2, 2, 2, 1, // VSHR, VSHL, VSEL, CCTL
60 const OpClass
Target::operationClass
[OP_LAST
+ 1] =
62 // NOP; PHI; UNION, SPLIT, MERGE, CONSTRAINT
65 OPCLASS_PSEUDO
, OPCLASS_PSEUDO
, OPCLASS_PSEUDO
, OPCLASS_PSEUDO
,
70 // ADD, SUB, MUL; DIV, MOD; MAD, FMA, SAD
71 OPCLASS_ARITH
, OPCLASS_ARITH
, OPCLASS_ARITH
,
72 OPCLASS_ARITH
, OPCLASS_ARITH
,
73 OPCLASS_ARITH
, OPCLASS_ARITH
, OPCLASS_ARITH
,
74 // ABS, NEG; NOT, AND, OR, XOR; SHL, SHR
75 OPCLASS_CONVERT
, OPCLASS_CONVERT
,
76 OPCLASS_LOGIC
, OPCLASS_LOGIC
, OPCLASS_LOGIC
, OPCLASS_LOGIC
,
77 OPCLASS_SHIFT
, OPCLASS_SHIFT
,
79 OPCLASS_COMPARE
, OPCLASS_COMPARE
,
80 // SAT, CEIL, FLOOR, TRUNC; CVT
81 OPCLASS_CONVERT
, OPCLASS_CONVERT
, OPCLASS_CONVERT
, OPCLASS_CONVERT
,
83 // SET(AND,OR,XOR); SELP, SLCT
84 OPCLASS_COMPARE
, OPCLASS_COMPARE
, OPCLASS_COMPARE
, OPCLASS_COMPARE
,
85 OPCLASS_COMPARE
, OPCLASS_COMPARE
,
86 // RCP, RSQ, LG2, SIN, COS; EX2, EXP, LOG, PRESIN, PREEX2; SQRT, POW
87 OPCLASS_SFU
, OPCLASS_SFU
, OPCLASS_SFU
, OPCLASS_SFU
, OPCLASS_SFU
,
88 OPCLASS_SFU
, OPCLASS_SFU
, OPCLASS_SFU
, OPCLASS_SFU
, OPCLASS_SFU
,
89 OPCLASS_SFU
, OPCLASS_SFU
,
90 // BRA, CALL, RET; CONT, BREAK, PRE(RET,CONT,BREAK); BRKPT, JOINAT, JOIN
91 OPCLASS_FLOW
, OPCLASS_FLOW
, OPCLASS_FLOW
,
92 OPCLASS_FLOW
, OPCLASS_FLOW
, OPCLASS_FLOW
, OPCLASS_FLOW
, OPCLASS_FLOW
,
93 OPCLASS_FLOW
, OPCLASS_FLOW
, OPCLASS_FLOW
,
95 OPCLASS_FLOW
, OPCLASS_FLOW
,
98 // VFETCH, PFETCH, EXPORT
99 OPCLASS_LOAD
, OPCLASS_OTHER
, OPCLASS_STORE
,
101 OPCLASS_SFU
, OPCLASS_SFU
,
103 OPCLASS_CONTROL
, OPCLASS_CONTROL
,
104 // TEX, TXB, TXL, TXF; TXQ, TXD, TXG, TEXCSAA; TEXPREP
105 OPCLASS_TEXTURE
, OPCLASS_TEXTURE
, OPCLASS_TEXTURE
, OPCLASS_TEXTURE
,
106 OPCLASS_TEXTURE
, OPCLASS_TEXTURE
, OPCLASS_TEXTURE
, OPCLASS_TEXTURE
,
108 // SULDB, SULDP, SUSTB, SUSTP; SUREDB, SUREDP, SULEA
109 OPCLASS_SURFACE
, OPCLASS_SURFACE
, OPCLASS_ATOMIC
, OPCLASS_SURFACE
,
110 OPCLASS_SURFACE
, OPCLASS_SURFACE
, OPCLASS_SURFACE
,
111 // SUBFM, SUCLAMP, SUEAU, MADSP
112 OPCLASS_OTHER
, OPCLASS_OTHER
, OPCLASS_OTHER
, OPCLASS_ARITH
,
115 // DFDX, DFDY, RDSV, WRSV; QUADOP, QUADON, QUADPOP
116 OPCLASS_OTHER
, OPCLASS_OTHER
, OPCLASS_OTHER
, OPCLASS_OTHER
,
117 OPCLASS_OTHER
, OPCLASS_CONTROL
, OPCLASS_CONTROL
,
118 // POPCNT, INSBF, EXTBF, PERMT
119 OPCLASS_BITFIELD
, OPCLASS_BITFIELD
, OPCLASS_BITFIELD
, OPCLASS_BITFIELD
,
121 OPCLASS_ATOMIC
, OPCLASS_CONTROL
,
122 // VADD, VAVG, VMIN, VMAX
123 OPCLASS_VECTOR
, OPCLASS_VECTOR
, OPCLASS_VECTOR
, OPCLASS_VECTOR
,
124 // VSAD, VSET, VSHR, VSHL
125 OPCLASS_VECTOR
, OPCLASS_VECTOR
, OPCLASS_VECTOR
, OPCLASS_VECTOR
,
127 OPCLASS_VECTOR
, OPCLASS_CONTROL
,
128 OPCLASS_PSEUDO
// LAST
132 extern Target
*getTargetNVC0(unsigned int chipset
);
133 extern Target
*getTargetNV50(unsigned int chipset
);
135 Target
*Target::create(unsigned int chipset
)
137 switch (chipset
& ~0xf) {
143 return getTargetNVC0(chipset
);
148 return getTargetNV50(chipset
);
150 ERROR("unsupported target: NV%x\n", chipset
);
155 void Target::destroy(Target
*targ
)
160 CodeEmitter::CodeEmitter(const Target
*target
) : targ(target
)
165 CodeEmitter::setCodeLocation(void *ptr
, uint32_t size
)
167 code
= reinterpret_cast<uint32_t *>(ptr
);
169 codeSizeLimit
= size
;
173 CodeEmitter::printBinary() const
175 uint32_t *bin
= code
- codeSize
/ 4;
176 INFO("program binary (%u bytes)", codeSize
);
177 for (unsigned int pos
= 0; pos
< codeSize
/ 4; ++pos
) {
180 INFO("%08x ", bin
[pos
]);
185 static inline uint32_t sizeToBundlesNVE4(uint32_t size
)
187 return (size
+ 55) / 56;
191 CodeEmitter::prepareEmission(Program
*prog
)
193 for (ArrayList::Iterator fi
= prog
->allFuncs
.iterator();
194 !fi
.end(); fi
.next()) {
195 Function
*func
= reinterpret_cast<Function
*>(fi
.get());
196 func
->binPos
= prog
->binSize
;
197 prepareEmission(func
);
199 // adjust sizes & positions for schedulding info:
200 if (prog
->getTarget()->hasSWSched
) {
201 uint32_t adjPos
= func
->binPos
;
202 BasicBlock
*bb
= NULL
;
203 for (int i
= 0; i
< func
->bbCount
; ++i
) {
204 bb
= func
->bbArray
[i
];
205 int32_t adjSize
= bb
->binSize
;
207 adjSize
-= 64 - adjPos
% 64;
211 adjSize
= bb
->binSize
+ sizeToBundlesNVE4(adjSize
) * 8;
213 bb
->binSize
= adjSize
;
217 func
->binSize
= adjPos
- func
->binPos
;
220 prog
->binSize
+= func
->binSize
;
225 CodeEmitter::prepareEmission(Function
*func
)
228 func
->bbArray
= new BasicBlock
* [func
->cfg
.getSize()];
230 BasicBlock::get(func
->cfg
.getRoot())->binPos
= func
->binPos
;
232 for (IteratorRef it
= func
->cfg
.iteratorCFG(); !it
->end(); it
->next())
233 prepareEmission(BasicBlock::get(*it
));
237 CodeEmitter::prepareEmission(BasicBlock
*bb
)
239 Instruction
*i
, *next
;
240 Function
*func
= bb
->getFunction();
244 for (j
= func
->bbCount
- 1; j
>= 0 && !func
->bbArray
[j
]->binSize
; --j
);
246 for (; j
>= 0; --j
) {
247 BasicBlock
*in
= func
->bbArray
[j
];
248 Instruction
*exit
= in
->getExit();
250 if (exit
&& exit
->op
== OP_BRA
&& exit
->asFlow()->target
.bb
== bb
) {
254 for (++j
; j
< func
->bbCount
; ++j
)
255 func
->bbArray
[j
]->binPos
-= 8;
259 bb
->binPos
= in
->binPos
+ in
->binSize
;
260 if (in
->binSize
) // no more no-op branches to bb
263 func
->bbArray
[func
->bbCount
++] = bb
;
268 // determine encoding size, try to group short instructions
270 for (i
= bb
->getEntry(); i
; i
= next
) {
273 if (i
->op
== OP_MEMBAR
&& !targ
->isOpSupported(OP_MEMBAR
, TYPE_NONE
)) {
278 i
->encSize
= getMinEncodingSize(i
);
279 if (next
&& i
->encSize
< 8)
282 if ((nShort
& 1) && next
&& getMinEncodingSize(next
) == 4) {
283 if (i
->isCommutationLegal(i
->next
)) {
284 bb
->permuteAdjacent(i
, next
);
290 if (i
->isCommutationLegal(i
->prev
) && next
->next
) {
291 bb
->permuteAdjacent(i
->prev
, i
);
298 i
->prev
->encSize
= 8;
305 i
->prev
->encSize
= 8;
310 bb
->binSize
+= i
->encSize
;
313 if (bb
->getExit()->encSize
== 4) {
315 bb
->getExit()->encSize
= 8;
318 if ((bb
->getExit()->prev
->encSize
== 4) && !(nShort
& 1)) {
320 bb
->getExit()->prev
->encSize
= 8;
323 assert(!bb
->getEntry() || (bb
->getExit() && bb
->getExit()->encSize
== 8));
325 func
->binSize
+= bb
->binSize
;
329 Program::emitSymbolTable(struct nv50_ir_prog_info
*info
)
331 unsigned int n
= 0, nMax
= allFuncs
.getSize();
334 (struct nv50_ir_prog_symbol
*)MALLOC(nMax
* sizeof(*info
->bin
.syms
));
336 for (ArrayList::Iterator fi
= allFuncs
.iterator();
339 Function
*f
= (Function
*)fi
.get();
342 info
->bin
.syms
[n
].label
= f
->getLabel();
343 info
->bin
.syms
[n
].offset
= f
->binPos
;
346 info
->bin
.numSyms
= n
;
350 Program::emitBinary(struct nv50_ir_prog_info
*info
)
352 CodeEmitter
*emit
= target
->getCodeEmitter(progType
);
354 emit
->prepareEmission(this);
356 if (dbgFlags
& NV50_IR_DEBUG_BASIC
)
363 code
= reinterpret_cast<uint32_t *>(MALLOC(binSize
));
366 emit
->setCodeLocation(code
, binSize
);
368 for (ArrayList::Iterator fi
= allFuncs
.iterator(); !fi
.end(); fi
.next()) {
369 Function
*fn
= reinterpret_cast<Function
*>(fi
.get());
371 assert(emit
->getCodeSize() == fn
->binPos
);
373 for (int b
= 0; b
< fn
->bbCount
; ++b
)
374 for (Instruction
*i
= fn
->bbArray
[b
]->getEntry(); i
; i
= i
->next
)
375 emit
->emitInstruction(i
);
377 info
->bin
.relocData
= emit
->getRelocInfo();
379 emitSymbolTable(info
);
381 // the nvc0 driver will print the binary iself together with the header
382 if ((dbgFlags
& NV50_IR_DEBUG_BASIC
) && getTarget()->getChipset() < 0xc0)
389 #define RELOC_ALLOC_INCREMENT 8
392 CodeEmitter::addReloc(RelocEntry::Type ty
, int w
, uint32_t data
, uint32_t m
,
395 unsigned int n
= relocInfo
? relocInfo
->count
: 0;
397 if (!(n
% RELOC_ALLOC_INCREMENT
)) {
398 size_t size
= sizeof(RelocInfo
) + n
* sizeof(RelocEntry
);
399 relocInfo
= reinterpret_cast<RelocInfo
*>(
400 REALLOC(relocInfo
, n
? size
: 0,
401 size
+ RELOC_ALLOC_INCREMENT
* sizeof(RelocEntry
)));
405 memset(relocInfo
, 0, sizeof(RelocInfo
));
409 relocInfo
->entry
[n
].data
= data
;
410 relocInfo
->entry
[n
].mask
= m
;
411 relocInfo
->entry
[n
].offset
= codeSize
+ w
* 4;
412 relocInfo
->entry
[n
].bitPos
= s
;
413 relocInfo
->entry
[n
].type
= ty
;
419 RelocEntry::apply(uint32_t *binary
, const RelocInfo
*info
) const
424 case TYPE_CODE
: value
= info
->codePos
; break;
425 case TYPE_BUILTIN
: value
= info
->libPos
; break;
426 case TYPE_DATA
: value
= info
->dataPos
; break;
432 value
= (bitPos
< 0) ? (value
>> -bitPos
) : (value
<< bitPos
);
434 binary
[offset
/ 4] &= ~mask
;
435 binary
[offset
/ 4] |= value
& mask
;
438 } // namespace nv50_ir
441 #include "codegen/nv50_ir_driver.h"
446 nv50_ir_relocate_code(void *relocData
, uint32_t *code
,
451 nv50_ir::RelocInfo
*info
= reinterpret_cast<nv50_ir::RelocInfo
*>(relocData
);
453 info
->codePos
= codePos
;
454 info
->libPos
= libPos
;
455 info
->dataPos
= dataPos
;
457 for (unsigned int i
= 0; i
< info
->count
; ++i
)
458 info
->entry
[i
].apply(code
, info
);
462 nv50_ir_get_target_library(uint32_t chipset
,
463 const uint32_t **code
, uint32_t *size
)
465 nv50_ir::Target
*targ
= nv50_ir::Target::create(chipset
);
466 targ
->getBuiltinCode(code
, size
);
467 nv50_ir::Target::destroy(targ
);