1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-10-25 22:12:57 +02:00
llvm-mirror/lib/Target/PowerPC/InstPrinter/PPCInstPrinter.cpp
Hal Finkel 883c64377d Add CR-bit tracking to the PowerPC backend for i1 values
This change enables tracking i1 values in the PowerPC backend using the
condition register bits. These bits can be treated on PowerPC as separate
registers; individual bit operations (and, or, xor, etc.) are supported.
Tracking booleans in CR bits has several advantages:

 - Reduction in register pressure (because we no longer need GPRs to store
   boolean values).

 - Logical operations on booleans can be handled more efficiently; we used to
   have to move all results from comparisons into GPRs, perform promoted
   logical operations in GPRs, and then move the result back into condition
   register bits to be used by conditional branches. This can be very
   inefficient, because the throughput of these CR <-> GPR moves have high
   latency and low throughput (especially when other associated instructions
   are accounted for).

 - On the POWER7 and similar cores, we can increase total throughput by using
   the CR bits. CR bit operations have a dedicated functional unit.

Most of this is more-or-less mechanical: Adjustments were needed in the
calling-convention code, support was added for spilling/restoring individual
condition-register bits, and conditional branch instruction definitions taking
specific CR bits were added (plus patterns and code for generating bit-level
operations).

This is enabled by default when running at -O2 and higher. For -O0 and -O1,
where the ability to debug is more important, this feature is disabled by
default. Individual CR bits do not have assigned DWARF register numbers,
and storing values in CR bits makes them invisible to the debugger.

It is critical, however, that we don't move i1 values that have been promoted
to larger values (such as those passed as function arguments) into bit
registers only to quickly turn around and move the values back into GPRs (such
as happens when values are returned by functions). A pair of target-specific
DAG combines are added to remove the trunc/extends in:
  trunc(binary-ops(binary-ops(zext(x), zext(y)), ...)
and:
  zext(binary-ops(binary-ops(trunc(x), trunc(y)), ...)
In short, we only want to use CR bits where some of the i1 values come from
comparisons or are used by conditional branches or selects. To put it another
way, if we can do the entire i1 computation in GPRs, then we probably should
(on the POWER7, the GPR-operation throughput is higher, and for all cores, the
CR <-> GPR moves are expensive).

POWER7 test-suite performance results (from 10 runs in each configuration):

SingleSource/Benchmarks/Misc/mandel-2: 35% speedup
MultiSource/Benchmarks/Prolangs-C++/city/city: 21% speedup
MultiSource/Benchmarks/MiBench/automotive-susan: 23% speedup
SingleSource/Benchmarks/CoyoteBench/huffbench: 13% speedup
SingleSource/Benchmarks/Misc-C++/Large/sphereflake: 13% speedup
SingleSource/Benchmarks/Misc-C++/mandel-text: 10% speedup

SingleSource/Benchmarks/Misc-C++-EH/spirit: 10% slowdown
MultiSource/Applications/lemon/lemon: 8% slowdown

llvm-svn: 202451
2014-02-28 00:27:01 +00:00

348 lines
10 KiB
C++

//===-- PPCInstPrinter.cpp - Convert PPC MCInst to assembly syntax --------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This class prints an PPC MCInst to a .s file.
//
//===----------------------------------------------------------------------===//
#define DEBUG_TYPE "asm-printer"
#include "PPCInstPrinter.h"
#include "MCTargetDesc/PPCMCTargetDesc.h"
#include "MCTargetDesc/PPCPredicates.h"
#include "llvm/MC/MCExpr.h"
#include "llvm/MC/MCInst.h"
#include "llvm/MC/MCInstrInfo.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Target/TargetOpcodes.h"
using namespace llvm;
// FIXME: Once the integrated assembler supports full register names, tie this
// to the verbose-asm setting.
static cl::opt<bool>
FullRegNames("ppc-asm-full-reg-names", cl::Hidden, cl::init(false),
cl::desc("Use full register names when printing assembly"));
#include "PPCGenAsmWriter.inc"
void PPCInstPrinter::printRegName(raw_ostream &OS, unsigned RegNo) const {
OS << getRegisterName(RegNo);
}
void PPCInstPrinter::printInst(const MCInst *MI, raw_ostream &O,
StringRef Annot) {
// Check for slwi/srwi mnemonics.
if (MI->getOpcode() == PPC::RLWINM) {
unsigned char SH = MI->getOperand(2).getImm();
unsigned char MB = MI->getOperand(3).getImm();
unsigned char ME = MI->getOperand(4).getImm();
bool useSubstituteMnemonic = false;
if (SH <= 31 && MB == 0 && ME == (31-SH)) {
O << "\tslwi "; useSubstituteMnemonic = true;
}
if (SH <= 31 && MB == (32-SH) && ME == 31) {
O << "\tsrwi "; useSubstituteMnemonic = true;
SH = 32-SH;
}
if (useSubstituteMnemonic) {
printOperand(MI, 0, O);
O << ", ";
printOperand(MI, 1, O);
O << ", " << (unsigned int)SH;
printAnnotation(O, Annot);
return;
}
}
if ((MI->getOpcode() == PPC::OR || MI->getOpcode() == PPC::OR8) &&
MI->getOperand(1).getReg() == MI->getOperand(2).getReg()) {
O << "\tmr ";
printOperand(MI, 0, O);
O << ", ";
printOperand(MI, 1, O);
printAnnotation(O, Annot);
return;
}
if (MI->getOpcode() == PPC::RLDICR) {
unsigned char SH = MI->getOperand(2).getImm();
unsigned char ME = MI->getOperand(3).getImm();
// rldicr RA, RS, SH, 63-SH == sldi RA, RS, SH
if (63-SH == ME) {
O << "\tsldi ";
printOperand(MI, 0, O);
O << ", ";
printOperand(MI, 1, O);
O << ", " << (unsigned int)SH;
printAnnotation(O, Annot);
return;
}
}
// For fast-isel, a COPY_TO_REGCLASS may survive this long. This is
// used when converting a 32-bit float to a 64-bit float as part of
// conversion to an integer (see PPCFastISel.cpp:SelectFPToI()),
// as otherwise we have problems with incorrect register classes
// in machine instruction verification. For now, just avoid trying
// to print it as such an instruction has no effect (a 32-bit float
// in a register is already in 64-bit form, just with lower
// precision). FIXME: Is there a better solution?
if (MI->getOpcode() == TargetOpcode::COPY_TO_REGCLASS)
return;
printInstruction(MI, O);
printAnnotation(O, Annot);
}
void PPCInstPrinter::printPredicateOperand(const MCInst *MI, unsigned OpNo,
raw_ostream &O,
const char *Modifier) {
unsigned Code = MI->getOperand(OpNo).getImm();
if (StringRef(Modifier) == "cc") {
switch ((PPC::Predicate)Code) {
case PPC::PRED_LT_MINUS:
case PPC::PRED_LT_PLUS:
case PPC::PRED_LT:
O << "lt";
return;
case PPC::PRED_LE_MINUS:
case PPC::PRED_LE_PLUS:
case PPC::PRED_LE:
O << "le";
return;
case PPC::PRED_EQ_MINUS:
case PPC::PRED_EQ_PLUS:
case PPC::PRED_EQ:
O << "eq";
return;
case PPC::PRED_GE_MINUS:
case PPC::PRED_GE_PLUS:
case PPC::PRED_GE:
O << "ge";
return;
case PPC::PRED_GT_MINUS:
case PPC::PRED_GT_PLUS:
case PPC::PRED_GT:
O << "gt";
return;
case PPC::PRED_NE_MINUS:
case PPC::PRED_NE_PLUS:
case PPC::PRED_NE:
O << "ne";
return;
case PPC::PRED_UN_MINUS:
case PPC::PRED_UN_PLUS:
case PPC::PRED_UN:
O << "un";
return;
case PPC::PRED_NU_MINUS:
case PPC::PRED_NU_PLUS:
case PPC::PRED_NU:
O << "nu";
return;
case PPC::PRED_BIT_SET:
case PPC::PRED_BIT_UNSET:
llvm_unreachable("Invalid use of bit predicate code");
}
llvm_unreachable("Invalid predicate code");
}
if (StringRef(Modifier) == "pm") {
switch ((PPC::Predicate)Code) {
case PPC::PRED_LT:
case PPC::PRED_LE:
case PPC::PRED_EQ:
case PPC::PRED_GE:
case PPC::PRED_GT:
case PPC::PRED_NE:
case PPC::PRED_UN:
case PPC::PRED_NU:
return;
case PPC::PRED_LT_MINUS:
case PPC::PRED_LE_MINUS:
case PPC::PRED_EQ_MINUS:
case PPC::PRED_GE_MINUS:
case PPC::PRED_GT_MINUS:
case PPC::PRED_NE_MINUS:
case PPC::PRED_UN_MINUS:
case PPC::PRED_NU_MINUS:
O << "-";
return;
case PPC::PRED_LT_PLUS:
case PPC::PRED_LE_PLUS:
case PPC::PRED_EQ_PLUS:
case PPC::PRED_GE_PLUS:
case PPC::PRED_GT_PLUS:
case PPC::PRED_NE_PLUS:
case PPC::PRED_UN_PLUS:
case PPC::PRED_NU_PLUS:
O << "+";
return;
case PPC::PRED_BIT_SET:
case PPC::PRED_BIT_UNSET:
llvm_unreachable("Invalid use of bit predicate code");
}
llvm_unreachable("Invalid predicate code");
}
assert(StringRef(Modifier) == "reg" &&
"Need to specify 'cc', 'pm' or 'reg' as predicate op modifier!");
printOperand(MI, OpNo+1, O);
}
void PPCInstPrinter::printS5ImmOperand(const MCInst *MI, unsigned OpNo,
raw_ostream &O) {
int Value = MI->getOperand(OpNo).getImm();
Value = SignExtend32<5>(Value);
O << (int)Value;
}
void PPCInstPrinter::printU5ImmOperand(const MCInst *MI, unsigned OpNo,
raw_ostream &O) {
unsigned int Value = MI->getOperand(OpNo).getImm();
assert(Value <= 31 && "Invalid u5imm argument!");
O << (unsigned int)Value;
}
void PPCInstPrinter::printU6ImmOperand(const MCInst *MI, unsigned OpNo,
raw_ostream &O) {
unsigned int Value = MI->getOperand(OpNo).getImm();
assert(Value <= 63 && "Invalid u6imm argument!");
O << (unsigned int)Value;
}
void PPCInstPrinter::printS16ImmOperand(const MCInst *MI, unsigned OpNo,
raw_ostream &O) {
if (MI->getOperand(OpNo).isImm())
O << (short)MI->getOperand(OpNo).getImm();
else
printOperand(MI, OpNo, O);
}
void PPCInstPrinter::printU16ImmOperand(const MCInst *MI, unsigned OpNo,
raw_ostream &O) {
if (MI->getOperand(OpNo).isImm())
O << (unsigned short)MI->getOperand(OpNo).getImm();
else
printOperand(MI, OpNo, O);
}
void PPCInstPrinter::printBranchOperand(const MCInst *MI, unsigned OpNo,
raw_ostream &O) {
if (!MI->getOperand(OpNo).isImm())
return printOperand(MI, OpNo, O);
// Branches can take an immediate operand. This is used by the branch
// selection pass to print .+8, an eight byte displacement from the PC.
O << ".+";
printAbsBranchOperand(MI, OpNo, O);
}
void PPCInstPrinter::printAbsBranchOperand(const MCInst *MI, unsigned OpNo,
raw_ostream &O) {
if (!MI->getOperand(OpNo).isImm())
return printOperand(MI, OpNo, O);
O << (int)MI->getOperand(OpNo).getImm()*4;
}
void PPCInstPrinter::printcrbitm(const MCInst *MI, unsigned OpNo,
raw_ostream &O) {
unsigned CCReg = MI->getOperand(OpNo).getReg();
unsigned RegNo;
switch (CCReg) {
default: llvm_unreachable("Unknown CR register");
case PPC::CR0: RegNo = 0; break;
case PPC::CR1: RegNo = 1; break;
case PPC::CR2: RegNo = 2; break;
case PPC::CR3: RegNo = 3; break;
case PPC::CR4: RegNo = 4; break;
case PPC::CR5: RegNo = 5; break;
case PPC::CR6: RegNo = 6; break;
case PPC::CR7: RegNo = 7; break;
}
O << (0x80 >> RegNo);
}
void PPCInstPrinter::printMemRegImm(const MCInst *MI, unsigned OpNo,
raw_ostream &O) {
printS16ImmOperand(MI, OpNo, O);
O << '(';
if (MI->getOperand(OpNo+1).getReg() == PPC::R0)
O << "0";
else
printOperand(MI, OpNo+1, O);
O << ')';
}
void PPCInstPrinter::printMemRegReg(const MCInst *MI, unsigned OpNo,
raw_ostream &O) {
// When used as the base register, r0 reads constant zero rather than
// the value contained in the register. For this reason, the darwin
// assembler requires that we print r0 as 0 (no r) when used as the base.
if (MI->getOperand(OpNo).getReg() == PPC::R0)
O << "0";
else
printOperand(MI, OpNo, O);
O << ", ";
printOperand(MI, OpNo+1, O);
}
void PPCInstPrinter::printTLSCall(const MCInst *MI, unsigned OpNo,
raw_ostream &O) {
printBranchOperand(MI, OpNo, O);
O << '(';
printOperand(MI, OpNo+1, O);
O << ')';
}
/// stripRegisterPrefix - This method strips the character prefix from a
/// register name so that only the number is left. Used by for linux asm.
static const char *stripRegisterPrefix(const char *RegName) {
if (FullRegNames)
return RegName;
switch (RegName[0]) {
case 'r':
case 'f':
case 'v': return RegName + 1;
case 'c': if (RegName[1] == 'r') return RegName + 2;
}
return RegName;
}
void PPCInstPrinter::printOperand(const MCInst *MI, unsigned OpNo,
raw_ostream &O) {
const MCOperand &Op = MI->getOperand(OpNo);
if (Op.isReg()) {
const char *RegName = getRegisterName(Op.getReg());
// The linux and AIX assembler does not take register prefixes.
if (!isDarwinSyntax())
RegName = stripRegisterPrefix(RegName);
O << RegName;
return;
}
if (Op.isImm()) {
O << Op.getImm();
return;
}
assert(Op.isExpr() && "unknown operand kind in printOperand");
O << *Op.getExpr();
}