fp.isa revision 2649
1// -*- mode:c++ -*- 2 3// Copyright (c) 2003-2005 The Regents of The University of Michigan 4// All rights reserved. 5// 6// Redistribution and use in source and binary forms, with or without 7// modification, are permitted provided that the following conditions are 8// met: redistributions of source code must retain the above copyright 9// notice, this list of conditions and the following disclaimer; 10// redistributions in binary form must reproduce the above copyright 11// notice, this list of conditions and the following disclaimer in the 12// documentation and/or other materials provided with the distribution; 13// neither the name of the copyright holders nor the names of its 14// contributors may be used to endorse or promote products derived from 15// this software without specific prior written permission. 16// 17// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 18// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 19// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 20// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 21// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 22// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 23// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 24// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 25// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 26// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 27// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 28 29//////////////////////////////////////////////////////////////////// 30// 31// Floating-point instructions 32// 33// Note that many FP-type instructions which do not support all the 34// various rounding & trapping modes use the simpler format 35// BasicOperateWithNopCheck. 36// 37 38output exec {{ 39 /// Check "FP enabled" machine status bit. Called when executing any FP 40 /// instruction in full-system mode. 41 /// @retval Full-system mode: NoFault if FP is enabled, FenFault 42 /// if not. Non-full-system mode: always returns NoFault. 43#if FULL_SYSTEM 44 inline Fault checkFpEnableFault(%(CPU_exec_context)s *xc) 45 { 46 Fault fault = NoFault; // dummy... this ipr access should not fault 47 if (!EV5::ICSR_FPE(xc->readMiscRegWithEffect(AlphaISA::IPR_ICSR, fault))) { 48 fault = new FloatEnableFault; 49 } 50 return fault; 51 } 52#else 53 inline Fault checkFpEnableFault(%(CPU_exec_context)s *xc) 54 { 55 return NoFault; 56 } 57#endif 58}}; 59 60output header {{ 61 /** 62 * Base class for general floating-point instructions. Includes 63 * support for various Alpha rounding and trapping modes. Only FP 64 * instructions that require this support are derived from this 65 * class; the rest derive directly from AlphaStaticInst. 66 */ 67 class AlphaFP : public AlphaStaticInst 68 { 69 public: 70 /// Alpha FP rounding modes. 71 enum RoundingMode { 72 Chopped = 0, ///< round toward zero 73 Minus_Infinity = 1, ///< round toward minus infinity 74 Normal = 2, ///< round to nearest (default) 75 Dynamic = 3, ///< use FPCR setting (in instruction) 76 Plus_Infinity = 3 ///< round to plus inifinity (in FPCR) 77 }; 78 79 /// Alpha FP trapping modes. 80 /// For instructions that produce integer results, the 81 /// "Underflow Enable" modes really mean "Overflow Enable", and 82 /// the assembly modifier is V rather than U. 83 enum TrappingMode { 84 /// default: nothing enabled 85 Imprecise = 0, ///< no modifier 86 /// underflow/overflow traps enabled, inexact disabled 87 Underflow_Imprecise = 1, ///< /U or /V 88 Underflow_Precise = 5, ///< /SU or /SV 89 /// underflow/overflow and inexact traps enabled 90 Underflow_Inexact_Precise = 7 ///< /SUI or /SVI 91 }; 92 93 protected: 94 /// Map Alpha rounding mode to C99 constants from <fenv.h>. 95 static const int alphaToC99RoundingMode[]; 96 97 /// Map enum RoundingMode values to disassembly suffixes. 98 static const char *roundingModeSuffix[]; 99 /// Map enum TrappingMode values to FP disassembly suffixes. 100 static const char *fpTrappingModeSuffix[]; 101 /// Map enum TrappingMode values to integer disassembly suffixes. 102 static const char *intTrappingModeSuffix[]; 103 104 /// This instruction's rounding mode. 105 RoundingMode roundingMode; 106 /// This instruction's trapping mode. 107 TrappingMode trappingMode; 108 109 /// Have we warned about this instruction's unsupported 110 /// rounding mode (if applicable)? 111 mutable bool warnedOnRounding; 112 113 /// Have we warned about this instruction's unsupported 114 /// trapping mode (if applicable)? 115 mutable bool warnedOnTrapping; 116 117 /// Constructor 118 AlphaFP(const char *mnem, ExtMachInst _machInst, OpClass __opClass) 119 : AlphaStaticInst(mnem, _machInst, __opClass), 120 roundingMode((enum RoundingMode)FP_ROUNDMODE), 121 trappingMode((enum TrappingMode)FP_TRAPMODE), 122 warnedOnRounding(false), 123 warnedOnTrapping(false) 124 { 125 } 126 127 int getC99RoundingMode(uint64_t fpcr_val) const; 128 129 // This differs from the AlphaStaticInst version only in 130 // printing suffixes for non-default rounding & trapping modes. 131 std::string 132 generateDisassembly(Addr pc, const SymbolTable *symtab) const; 133 }; 134 135}}; 136 137 138output decoder {{ 139 int 140 AlphaFP::getC99RoundingMode(uint64_t fpcr_val) const 141 { 142 if (roundingMode == Dynamic) { 143 return alphaToC99RoundingMode[bits(fpcr_val, 59, 58)]; 144 } 145 else { 146 return alphaToC99RoundingMode[roundingMode]; 147 } 148 } 149 150 std::string 151 AlphaFP::generateDisassembly(Addr pc, const SymbolTable *symtab) const 152 { 153 std::string mnem_str(mnemonic); 154 155#ifndef SS_COMPATIBLE_DISASSEMBLY 156 std::string suffix(""); 157 suffix += ((_destRegIdx[0] >= FP_Base_DepTag) 158 ? fpTrappingModeSuffix[trappingMode] 159 : intTrappingModeSuffix[trappingMode]); 160 suffix += roundingModeSuffix[roundingMode]; 161 162 if (suffix != "") { 163 mnem_str = csprintf("%s/%s", mnemonic, suffix); 164 } 165#endif 166 167 std::stringstream ss; 168 ccprintf(ss, "%-10s ", mnem_str.c_str()); 169 170 // just print the first two source regs... if there's 171 // a third one, it's a read-modify-write dest (Rc), 172 // e.g. for CMOVxx 173 if (_numSrcRegs > 0) { 174 printReg(ss, _srcRegIdx[0]); 175 } 176 if (_numSrcRegs > 1) { 177 ss << ","; 178 printReg(ss, _srcRegIdx[1]); 179 } 180 181 // just print the first dest... if there's a second one, 182 // it's generally implicit 183 if (_numDestRegs > 0) { 184 if (_numSrcRegs > 0) 185 ss << ","; 186 printReg(ss, _destRegIdx[0]); 187 } 188 189 return ss.str(); 190 } 191 192 const int AlphaFP::alphaToC99RoundingMode[] = { 193 FE_TOWARDZERO, // Chopped 194 FE_DOWNWARD, // Minus_Infinity 195 FE_TONEAREST, // Normal 196 FE_UPWARD // Dynamic in inst, Plus_Infinity in FPCR 197 }; 198 199 const char *AlphaFP::roundingModeSuffix[] = { "c", "m", "", "d" }; 200 // mark invalid trapping modes, but don't fail on them, because 201 // you could decode anything on a misspeculated path 202 const char *AlphaFP::fpTrappingModeSuffix[] = 203 { "", "u", "INVTM2", "INVTM3", "INVTM4", "su", "INVTM6", "sui" }; 204 const char *AlphaFP::intTrappingModeSuffix[] = 205 { "", "v", "INVTM2", "INVTM3", "INVTM4", "sv", "INVTM6", "svi" }; 206}}; 207 208// FP instruction class execute method template. Handles non-standard 209// rounding modes. 210def template FloatingPointExecute {{ 211 Fault %(class_name)s::execute(%(CPU_exec_context)s *xc, 212 Trace::InstRecord *traceData) const 213 { 214 if (trappingMode != Imprecise && !warnedOnTrapping) { 215 warn("%s: non-standard trapping mode not supported", 216 generateDisassembly(0, NULL)); 217 warnedOnTrapping = true; 218 } 219 220 Fault fault = NoFault; 221 222 %(fp_enable_check)s; 223 %(op_decl)s; 224 %(op_rd)s; 225#if USE_FENV 226 if (roundingMode == Normal) { 227 %(code)s; 228 } else { 229 fesetround(getC99RoundingMode( 230 xc->readMiscReg(AlphaISA::Fpcr_DepTag))); 231 %(code)s; 232 fesetround(FE_TONEAREST); 233 } 234#else 235 if (roundingMode != Normal && !warnedOnRounding) { 236 warn("%s: non-standard rounding mode not supported", 237 generateDisassembly(0, NULL)); 238 warnedOnRounding = true; 239 } 240 %(code)s; 241#endif 242 243 if (fault == NoFault) { 244 %(op_wb)s; 245 } 246 247 return fault; 248 } 249}}; 250 251// FP instruction class execute method template where no dynamic 252// rounding mode control is needed. Like BasicExecute, but includes 253// check & warning for non-standard trapping mode. 254def template FPFixedRoundingExecute {{ 255 Fault %(class_name)s::execute(%(CPU_exec_context)s *xc, 256 Trace::InstRecord *traceData) const 257 { 258 if (trappingMode != Imprecise && !warnedOnTrapping) { 259 warn("%s: non-standard trapping mode not supported", 260 generateDisassembly(0, NULL)); 261 warnedOnTrapping = true; 262 } 263 264 Fault fault = NoFault; 265 266 %(fp_enable_check)s; 267 %(op_decl)s; 268 %(op_rd)s; 269 %(code)s; 270 271 if (fault == NoFault) { 272 %(op_wb)s; 273 } 274 275 return fault; 276 } 277}}; 278 279def template FloatingPointDecode {{ 280 { 281 AlphaStaticInst *i = new %(class_name)s(machInst); 282 if (FC == 31) { 283 i = makeNop(i); 284 } 285 return i; 286 } 287}}; 288 289// General format for floating-point operate instructions: 290// - Checks trapping and rounding mode flags. Trapping modes 291// currently unimplemented (will fail). 292// - Generates NOP if FC == 31. 293def format FloatingPointOperate(code, *opt_args) {{ 294 iop = InstObjParams(name, Name, 'AlphaFP', CodeBlock(code), opt_args) 295 decode_block = FloatingPointDecode.subst(iop) 296 header_output = BasicDeclare.subst(iop) 297 decoder_output = BasicConstructor.subst(iop) 298 exec_output = FloatingPointExecute.subst(iop) 299}}; 300 301// Special format for cvttq where rounding mode is pre-decoded 302def format FPFixedRounding(code, class_suffix, *opt_args) {{ 303 Name += class_suffix 304 iop = InstObjParams(name, Name, 'AlphaFP', CodeBlock(code), opt_args) 305 decode_block = FloatingPointDecode.subst(iop) 306 header_output = BasicDeclare.subst(iop) 307 decoder_output = BasicConstructor.subst(iop) 308 exec_output = FPFixedRoundingExecute.subst(iop) 309}}; 310 311