gem5/arch/alpha/isa/fp.isa
Kevin Lim e7825aab59 Changes to support automatic renaming of the shadow registers at decode time. This requires using an ExtMachInst (uint64_t) instead of the normal MachInst; the ExtMachInst is packed with extra decode context information. In the case of Alpha, the PAL mode is included.
The shadow registers are folded into the normal integer registers to ease renaming indexing.

Include the removed Opcdec class of instructions for faulting when a pal mode only instruction is decoded in non-pal mode.

arch/alpha/ev5.cc:
    Changes to automatically map the shadow registers if the instruction is in PAL mode.
arch/alpha/isa/branch.isa:
arch/alpha/isa/decoder.isa:
arch/alpha/isa/fp.isa:
arch/alpha/isa/int.isa:
arch/alpha/isa/mem.isa:
arch/alpha/isa/pal.isa:
arch/alpha/isa/unimp.isa:
    Changes for automatically using the shadow registers.  Now instructions must decode based on an ExtMachInst, which is a MachInst with any decode context information concatenated onto the higher order bits.
arch/alpha/isa/main.isa:
    Changes for automatically using the shadow registers.  Now instructions must decode based on an ExtMachInst, which is a MachInst with any decode context information concatenated onto the higher order bits.

    The decoder (for Alpha) uses the 32nd bit in order to determine if the machine is in PAL mode.  If it is, then it refers to the reg_redir table to determine the true index of the register it is using.

    Also include the opcdec instruction definition.
arch/alpha/isa_traits.hh:
    Define ExtMachInst type that is used by the static inst in order to decode the instruction, given the context of being in pal mode or not.

    Redefine the number of Int registers, splitting it into NumIntArchRegs (32) and NumIntRegs (32 + 8 shadow registers).

    Change the dependence tags to reflect the integer registers include the 8 shadow registers.

    Define function to make an ExtMachInst.  Currently it is somewhat specific to Alpha; in the future it must be decided to make this more generic and possibly slower, or leave it specific to each architecture and ifdef it within the CPU.
arch/isa_parser.py:
    Have static insts decode on the ExtMachInst.
base/remote_gdb.cc:
    Support the automatic remapping of shadow registers.  Remote GDB must now look at the PC being read in order to tell if it should use the normal register indices or the shadow register indices.
cpu/o3/regfile.hh:
    Comment out the pal registers; they are now a part of the integer registers.
cpu/simple/cpu.cc:
    Create an ExtMachInst to decode on, based on the normal MachInst and the PC of the instructoin.
cpu/static_inst.hh:
    Change from MachInst to ExtMachInst to support shadow register renaming.

--HG--
extra : convert_revision : 1d23eabf735e297068e1917445a6348e9f8c88d5
2006-03-03 15:28:25 -05:00

302 lines
10 KiB
C++

// -*- mode:c++ -*-
// Copyright (c) 2003-2005 The Regents of The University of Michigan
// All rights reserved.
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions are
// met: redistributions of source code must retain the above copyright
// notice, this list of conditions and the following disclaimer;
// redistributions in binary form must reproduce the above copyright
// notice, this list of conditions and the following disclaimer in the
// documentation and/or other materials provided with the distribution;
// neither the name of the copyright holders nor the names of its
// contributors may be used to endorse or promote products derived from
// this software without specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
output exec {{
/// Check "FP enabled" machine status bit. Called when executing any FP
/// instruction in full-system mode.
/// @retval Full-system mode: NoFault if FP is enabled, FenFault
/// if not. Non-full-system mode: always returns NoFault.
#if FULL_SYSTEM
inline Fault checkFpEnableFault(%(CPU_exec_context)s *xc)
{
Fault fault = NoFault; // dummy... this ipr access should not fault
if (!EV5::ICSR_FPE(xc->readMiscRegWithEffect(AlphaISA::IPR_ICSR, fault))) {
fault = new FloatEnableFault;
}
return fault;
}
#else
inline Fault checkFpEnableFault(%(CPU_exec_context)s *xc)
{
return NoFault;
}
#endif
}};
output header {{
/**
* Base class for general floating-point instructions. Includes
* support for various Alpha rounding and trapping modes. Only FP
* instructions that require this support are derived from this
* class; the rest derive directly from AlphaStaticInst.
*/
class AlphaFP : public AlphaStaticInst
{
public:
/// Alpha FP rounding modes.
enum RoundingMode {
Chopped = 0, ///< round toward zero
Minus_Infinity = 1, ///< round toward minus infinity
Normal = 2, ///< round to nearest (default)
Dynamic = 3, ///< use FPCR setting (in instruction)
Plus_Infinity = 3 ///< round to plus inifinity (in FPCR)
};
/// Alpha FP trapping modes.
/// For instructions that produce integer results, the
/// "Underflow Enable" modes really mean "Overflow Enable", and
/// the assembly modifier is V rather than U.
enum TrappingMode {
/// default: nothing enabled
Imprecise = 0, ///< no modifier
/// underflow/overflow traps enabled, inexact disabled
Underflow_Imprecise = 1, ///< /U or /V
Underflow_Precise = 5, ///< /SU or /SV
/// underflow/overflow and inexact traps enabled
Underflow_Inexact_Precise = 7 ///< /SUI or /SVI
};
protected:
/// Map Alpha rounding mode to C99 constants from <fenv.h>.
static const int alphaToC99RoundingMode[];
/// Map enum RoundingMode values to disassembly suffixes.
static const char *roundingModeSuffix[];
/// Map enum TrappingMode values to FP disassembly suffixes.
static const char *fpTrappingModeSuffix[];
/// Map enum TrappingMode values to integer disassembly suffixes.
static const char *intTrappingModeSuffix[];
/// This instruction's rounding mode.
RoundingMode roundingMode;
/// This instruction's trapping mode.
TrappingMode trappingMode;
/// Have we warned about this instruction's unsupported
/// rounding mode (if applicable)?
mutable bool warnedOnRounding;
/// Have we warned about this instruction's unsupported
/// trapping mode (if applicable)?
mutable bool warnedOnTrapping;
/// Constructor
AlphaFP(const char *mnem, ExtMachInst _machInst, OpClass __opClass)
: AlphaStaticInst(mnem, _machInst, __opClass),
roundingMode((enum RoundingMode)FP_ROUNDMODE),
trappingMode((enum TrappingMode)FP_TRAPMODE),
warnedOnRounding(false),
warnedOnTrapping(false)
{
}
int getC99RoundingMode(uint64_t fpcr_val) const;
// This differs from the AlphaStaticInst version only in
// printing suffixes for non-default rounding & trapping modes.
std::string
generateDisassembly(Addr pc, const SymbolTable *symtab) const;
};
}};
output decoder {{
int
AlphaFP::getC99RoundingMode(uint64_t fpcr_val) const
{
if (roundingMode == Dynamic) {
return alphaToC99RoundingMode[bits(fpcr_val, 59, 58)];
}
else {
return alphaToC99RoundingMode[roundingMode];
}
}
std::string
AlphaFP::generateDisassembly(Addr pc, const SymbolTable *symtab) const
{
std::string mnem_str(mnemonic);
#ifndef SS_COMPATIBLE_DISASSEMBLY
std::string suffix("");
suffix += ((_destRegIdx[0] >= FP_Base_DepTag)
? fpTrappingModeSuffix[trappingMode]
: intTrappingModeSuffix[trappingMode]);
suffix += roundingModeSuffix[roundingMode];
if (suffix != "") {
mnem_str = csprintf("%s/%s", mnemonic, suffix);
}
#endif
std::stringstream ss;
ccprintf(ss, "%-10s ", mnem_str.c_str());
// just print the first two source regs... if there's
// a third one, it's a read-modify-write dest (Rc),
// e.g. for CMOVxx
if (_numSrcRegs > 0) {
printReg(ss, _srcRegIdx[0]);
}
if (_numSrcRegs > 1) {
ss << ",";
printReg(ss, _srcRegIdx[1]);
}
// just print the first dest... if there's a second one,
// it's generally implicit
if (_numDestRegs > 0) {
if (_numSrcRegs > 0)
ss << ",";
printReg(ss, _destRegIdx[0]);
}
return ss.str();
}
const int AlphaFP::alphaToC99RoundingMode[] = {
FE_TOWARDZERO, // Chopped
FE_DOWNWARD, // Minus_Infinity
FE_TONEAREST, // Normal
FE_UPWARD // Dynamic in inst, Plus_Infinity in FPCR
};
const char *AlphaFP::roundingModeSuffix[] = { "c", "m", "", "d" };
// mark invalid trapping modes, but don't fail on them, because
// you could decode anything on a misspeculated path
const char *AlphaFP::fpTrappingModeSuffix[] =
{ "", "u", "INVTM2", "INVTM3", "INVTM4", "su", "INVTM6", "sui" };
const char *AlphaFP::intTrappingModeSuffix[] =
{ "", "v", "INVTM2", "INVTM3", "INVTM4", "sv", "INVTM6", "svi" };
}};
// FP instruction class execute method template. Handles non-standard
// rounding modes.
def template FloatingPointExecute {{
Fault %(class_name)s::execute(%(CPU_exec_context)s *xc,
Trace::InstRecord *traceData) const
{
if (trappingMode != Imprecise && !warnedOnTrapping) {
warn("%s: non-standard trapping mode not supported",
generateDisassembly(0, NULL));
warnedOnTrapping = true;
}
Fault fault = NoFault;
%(fp_enable_check)s;
%(op_decl)s;
%(op_rd)s;
#if USE_FENV
if (roundingMode == Normal) {
%(code)s;
} else {
fesetround(getC99RoundingMode(
xc->readMiscReg(AlphaISA::Fpcr_DepTag)));
%(code)s;
fesetround(FE_TONEAREST);
}
#else
if (roundingMode != Normal && !warnedOnRounding) {
warn("%s: non-standard rounding mode not supported",
generateDisassembly(0, NULL));
warnedOnRounding = true;
}
%(code)s;
#endif
if (fault == NoFault) {
%(op_wb)s;
}
return fault;
}
}};
// FP instruction class execute method template where no dynamic
// rounding mode control is needed. Like BasicExecute, but includes
// check & warning for non-standard trapping mode.
def template FPFixedRoundingExecute {{
Fault %(class_name)s::execute(%(CPU_exec_context)s *xc,
Trace::InstRecord *traceData) const
{
if (trappingMode != Imprecise && !warnedOnTrapping) {
warn("%s: non-standard trapping mode not supported",
generateDisassembly(0, NULL));
warnedOnTrapping = true;
}
Fault fault = NoFault;
%(fp_enable_check)s;
%(op_decl)s;
%(op_rd)s;
%(code)s;
if (fault == NoFault) {
%(op_wb)s;
}
return fault;
}
}};
def template FloatingPointDecode {{
{
AlphaStaticInst *i = new %(class_name)s(machInst);
if (FC == 31) {
i = makeNop(i);
}
return i;
}
}};
// General format for floating-point operate instructions:
// - Checks trapping and rounding mode flags. Trapping modes
// currently unimplemented (will fail).
// - Generates NOP if FC == 31.
def format FloatingPointOperate(code, *opt_args) {{
iop = InstObjParams(name, Name, 'AlphaFP', CodeBlock(code), opt_args)
decode_block = FloatingPointDecode.subst(iop)
header_output = BasicDeclare.subst(iop)
decoder_output = BasicConstructor.subst(iop)
exec_output = FloatingPointExecute.subst(iop)
}};
// Special format for cvttq where rounding mode is pre-decoded
def format FPFixedRounding(code, class_suffix, *opt_args) {{
Name += class_suffix
iop = InstObjParams(name, Name, 'AlphaFP', CodeBlock(code), opt_args)
decode_block = FloatingPointDecode.subst(iop)
header_output = BasicDeclare.subst(iop)
decoder_output = BasicConstructor.subst(iop)
exec_output = FPFixedRoundingExecute.subst(iop)
}};