sanchayanmaity/gem5
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

ARM: Cleanup and small fixes to some NEON ops to match the spec.

Browse source 
Only certain bits of the cpacr can be written, some must be equal.
Mult instructions that write the same register should do something sane
This commit is contained in:
William Wang 2011-04-04 11:42:28 -05:00
parent a679cd917a
commit 16fcad3907
5 changed files with 39 additions and 28 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

23
src/arch/arm/isa.cc
View file

@ -268,19 +268,22 @@ ISA::setMiscReg(int misc_reg, const MiscReg &val, ThreadContext *tc)
switch (misc_reg) { switch (misc_reg) {
case MISCREG_CPACR: case MISCREG_CPACR:
{ {
CPACR newCpacr = 0;
CPACR valCpacr = val; const uint32_t ones = (uint32_t)(-1);
newCpacr.cp10 = valCpacr.cp10; CPACR cpacrMask = 0;
newCpacr.cp11 = valCpacr.cp11; // Only cp10, cp11, and ase are implemented, nothing else should
//XXX d32dis isn't implemented. The manual says whether or not // be writable
//it works is implementation defined. cpacrMask.cp10 = ones;
newCpacr.asedis = valCpacr.asedis; cpacrMask.cp11 = ones;
newVal = newCpacr; cpacrMask.asedis = ones;
newVal &= cpacrMask;
DPRINTF(MiscRegs, "Writing misc reg %s: %#x\n",
miscRegName[misc_reg], newVal);
} }
break; break;
case MISCREG_CSSELR: case MISCREG_CSSELR:
warn_once("The csselr register isn't implemented.\n"); warn_once("The csselr register isn't implemented.\n");
break; return;
case MISCREG_FPSCR: case MISCREG_FPSCR:
{ {
const uint32_t ones = (uint32_t)(-1); const uint32_t ones = (uint32_t)(-1);
@ -320,6 +323,8 @@ ISA::setMiscReg(int misc_reg, const MiscReg &val, ThreadContext *tc)
break; break;
case MISCREG_FPEXC: case MISCREG_FPEXC:
{ {
// vfpv3 architecture, section B.6.1 of DDI04068
// bit 29 - valid only if fpexc[31] is 0
const uint32_t fpexcMask = 0x60000000; const uint32_t fpexcMask = 0x60000000;
newVal = (newVal & fpexcMask) | newVal = (newVal & fpexcMask) |
(miscRegs[MISCREG_FPEXC] & ~fpexcMask); (miscRegs[MISCREG_FPEXC] & ~fpexcMask);

8
src/arch/arm/isa/insts/mult.isa
View file

@ -349,8 +349,8 @@ let {{
''') ''')
buildMult4Inst ("smull", '''resTemp = (int64_t)Reg2.sw * buildMult4Inst ("smull", '''resTemp = (int64_t)Reg2.sw *
(int64_t)Reg3.sw; (int64_t)Reg3.sw;
Reg0 = (int32_t)resTemp;
Reg1 = (int32_t)(resTemp >> 32); Reg1 = (int32_t)(resTemp >> 32);
Reg0 = (int32_t)resTemp;
''', "llbit") ''', "llbit")
buildMult3InstUnCc("smulwb", '''Reg0 = resTemp = buildMult3InstUnCc("smulwb", '''Reg0 = resTemp =
(Reg1.sw * (Reg1.sw *
@ -374,16 +374,16 @@ let {{
''') ''')
buildMult4InstUnCc("umaal", '''resTemp = Reg2.ud * Reg3.ud + buildMult4InstUnCc("umaal", '''resTemp = Reg2.ud * Reg3.ud +
Reg0.ud + Reg1.ud; Reg0.ud + Reg1.ud;
Reg0.ud = (uint32_t)resTemp;
Reg1.ud = (uint32_t)(resTemp >> 32); Reg1.ud = (uint32_t)(resTemp >> 32);
Reg0.ud = (uint32_t)resTemp;
''') ''')
buildMult4Inst ("umlal", '''resTemp = Reg2.ud * Reg3.ud + Reg0.ud + buildMult4Inst ("umlal", '''resTemp = Reg2.ud * Reg3.ud + Reg0.ud +
(Reg1.ud << 32); (Reg1.ud << 32);
Reg0.ud = (uint32_t)resTemp;
Reg1.ud = (uint32_t)(resTemp >> 32); Reg1.ud = (uint32_t)(resTemp >> 32);
Reg0.ud = (uint32_t)resTemp;
''', "llbit") ''', "llbit")
buildMult4Inst ("umull", '''resTemp = Reg2.ud * Reg3.ud; buildMult4Inst ("umull", '''resTemp = Reg2.ud * Reg3.ud;
Reg0 = (uint32_t)resTemp;
Reg1 = (uint32_t)(resTemp >> 32); Reg1 = (uint32_t)(resTemp >> 32);
Reg0 = (uint32_t)resTemp;
''', "llbit") ''', "llbit")
}}; }};

28
src/arch/arm/isa/insts/neon.isa
View file

@ -1761,8 +1761,8 @@ let {{
} }
} }
''' '''
threeEqualRegInst("vshl", "VshlD", "SimdAluOp", allTypes, 2, vshlCode) threeEqualRegInst("vshl", "VshlD", "SimdShiftOp", allTypes, 2, vshlCode)
threeEqualRegInst("vshl", "VshlQ", "SimdAluOp", allTypes, 4, vshlCode) threeEqualRegInst("vshl", "VshlQ", "SimdShiftOp", allTypes, 4, vshlCode)
vrshlCode = ''' vrshlCode = '''
int16_t shiftAmt = (int8_t)srcElem2; int16_t shiftAmt = (int8_t)srcElem2;
@ -3204,8 +3204,8 @@ let {{
substDict = { "targs" : type, substDict = { "targs" : type,
"class_name" : Name } "class_name" : Name }
exec_output += NeonExecDeclare.subst(substDict) exec_output += NeonExecDeclare.subst(substDict)
vdupGprInst("vdup", "NVdupDGpr", "SimdAluOp", smallUnsignedTypes, 2) vdupGprInst("vdup", "NVdupDGpr", "SimdMiscOp", smallUnsignedTypes, 2)
vdupGprInst("vdup", "NVdupQGpr", "SimdAluOp", smallUnsignedTypes, 4) vdupGprInst("vdup", "NVdupQGpr", "SimdMiscOp", smallUnsignedTypes, 4)
vmovCode = 'destElem = imm;' vmovCode = 'destElem = imm;'
oneRegImmInst("vmov", "NVmoviD", "SimdMiscOp", ("uint64_t",), 2, vmovCode) oneRegImmInst("vmov", "NVmoviD", "SimdMiscOp", ("uint64_t",), 2, vmovCode)
@ -3309,8 +3309,8 @@ let {{
} }
} }
''' '''
buildVext("vext", "NVextD", "SimdAluOp", ("uint8_t",), 2, vextCode) buildVext("vext", "NVextD", "SimdMiscOp", ("uint8_t",), 2, vextCode)
buildVext("vext", "NVextQ", "SimdAluOp", ("uint8_t",), 4, vextCode) buildVext("vext", "NVextQ", "SimdMiscOp", ("uint8_t",), 4, vextCode)
def buildVtbxl(name, Name, opClass, length, isVtbl): def buildVtbxl(name, Name, opClass, length, isVtbl):
global header_output, decoder_output, exec_output global header_output, decoder_output, exec_output
@ -3366,13 +3366,13 @@ let {{
decoder_output += RegRegRegOpConstructor.subst(iop) decoder_output += RegRegRegOpConstructor.subst(iop)
exec_output += PredOpExecute.subst(iop) exec_output += PredOpExecute.subst(iop)
buildVtbxl("vtbl", "NVtbl1", "SimdAluOp", 1, "true") buildVtbxl("vtbl", "NVtbl1", "SimdMiscOp", 1, "true")
buildVtbxl("vtbl", "NVtbl2", "SimdAluOp", 2, "true") buildVtbxl("vtbl", "NVtbl2", "SimdMiscOp", 2, "true")
buildVtbxl("vtbl", "NVtbl3", "SimdAluOp", 3, "true") buildVtbxl("vtbl", "NVtbl3", "SimdMiscOp", 3, "true")
buildVtbxl("vtbl", "NVtbl4", "SimdAluOp", 4, "true") buildVtbxl("vtbl", "NVtbl4", "SimdMiscOp", 4, "true")
buildVtbxl("vtbx", "NVtbx1", "SimdAluOp", 1, "false") buildVtbxl("vtbx", "NVtbx1", "SimdMiscOp", 1, "false")
buildVtbxl("vtbx", "NVtbx2", "SimdAluOp", 2, "false") buildVtbxl("vtbx", "NVtbx2", "SimdMiscOp", 2, "false")
buildVtbxl("vtbx", "NVtbx3", "SimdAluOp", 3, "false") buildVtbxl("vtbx", "NVtbx3", "SimdMiscOp", 3, "false")
buildVtbxl("vtbx", "NVtbx4", "SimdAluOp", 4, "false") buildVtbxl("vtbx", "NVtbx4", "SimdMiscOp", 4, "false")
}}; }};

1
src/arch/arm/miscregs.hh
View file

@ -310,6 +310,7 @@ namespace ArmISA
Bitfield<23, 22> cp11; Bitfield<23, 22> cp11;
Bitfield<25, 24> cp12; Bitfield<25, 24> cp12;
Bitfield<27, 26> cp13; Bitfield<27, 26> cp13;
Bitfield<29, 28> rsvd;
Bitfield<30> d32dis; Bitfield<30> d32dis;
Bitfield<31> asedis; Bitfield<31> asedis;
EndBitUnion(CPACR) EndBitUnion(CPACR)

5
src/arch/arm/utility.hh
View file

@ -146,7 +146,12 @@ vfpEnabled(CPACR cpacr, CPSR cpsr)
static inline bool static inline bool
vfpEnabled(CPACR cpacr, CPSR cpsr, FPEXC fpexc) vfpEnabled(CPACR cpacr, CPSR cpsr, FPEXC fpexc)
{ {
if ((cpacr.cp11 == 0x3) ||
((cpacr.cp11 == 0x1) && inPrivilegedMode(cpsr)))
return fpexc.en && vfpEnabled(cpacr, cpsr); return fpexc.en && vfpEnabled(cpacr, cpsr);
else
return fpexc.en && vfpEnabled(cpacr, cpsr) &&
(cpacr.cp11 == cpacr.cp10);
} }
static inline bool static inline bool