blob: 2e0ec44952b6cf38ad968b561b40cf002b2966f6 [file] [log] [blame]
// -*- mode: c++ -*-
// Copyright (c) 2012-2013, 2015-2018 ARM Limited
// All rights reserved
//
// The license below extends only to copyright in the software and shall
// not be construed as granting a license to any other intellectual
// property including but not limited to intellectual property relating
// to a hardware implementation of the functionality of the software
// licensed hereunder. You may use the software subject to the license
// terms below provided that you ensure that this notice is replicated
// unmodified and in its entirety in all distributions of the software,
// modified or unmodified, in source code or in binary form.
//
// 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.
//
// Authors: Giacomo Gabrielli
// Mbou Eyole
let {{
header_output = ""
exec_output = ""
decoders = { 'Generic' : {} }
# FP types (FP operations always work with unsigned representations)
floatTypes = ("uint16_t", "uint32_t", "uint64_t")
smallFloatTypes = ("uint32_t",)
zeroSveVecRegUpperPartCode = '''
TheISA::ISA::zeroSveVecRegUpperPart(%s,
ArmStaticInst::getCurSveVecLen<uint64_t>(xc->tcBase()));
'''
def threeEqualRegInstX(name, Name, opClass, types, rCount, op,
readDest=False, pairwise=False, scalar=False,
byElem=False, decoder='Generic'):
assert (not pairwise) or ((not byElem) and (not scalar))
global header_output, exec_output, decoders
eWalkCode = simd64EnabledCheckCode + '''
RegVect srcReg1, destReg;
'''
if byElem:
# 2nd register operand has to be read fully
eWalkCode += '''
FullRegVect srcReg2;
'''
else:
eWalkCode += '''
RegVect srcReg2;
'''
for reg in range(rCount):
eWalkCode += '''
srcReg1.regs[%(reg)d] = htog(AA64FpOp1P%(reg)d_uw);
srcReg2.regs[%(reg)d] = htog(AA64FpOp2P%(reg)d_uw);
''' % { "reg" : reg }
if readDest:
eWalkCode += '''
destReg.regs[%(reg)d] = htog(AA64FpDestP%(reg)d_uw);
''' % { "reg" : reg }
if byElem:
# 2nd operand has to be read fully
for reg in range(rCount, 4):
eWalkCode += '''
srcReg2.regs[%(reg)d] = htog(AA64FpOp2P%(reg)d_uw);
''' % { "reg" : reg }
readDestCode = ''
if readDest:
readDestCode = 'destElem = gtoh(destReg.elements[i]);'
if pairwise:
eWalkCode += '''
for (unsigned i = 0; i < eCount; i++) {
Element srcElem1 = gtoh(2 * i < eCount ?
srcReg1.elements[2 * i] :
srcReg2.elements[2 * i - eCount]);
Element srcElem2 = gtoh(2 * i < eCount ?
srcReg1.elements[2 * i + 1] :
srcReg2.elements[2 * i + 1 - eCount]);
Element destElem;
%(readDest)s
%(op)s
destReg.elements[i] = htog(destElem);
}
''' % { "op" : op, "readDest" : readDestCode }
else:
scalarCheck = '''
if (i != 0) {
destReg.elements[i] = 0;
continue;
}
'''
eWalkCode += '''
for (unsigned i = 0; i < eCount; i++) {
%(scalarCheck)s
Element srcElem1 = gtoh(srcReg1.elements[i]);
Element srcElem2 = gtoh(srcReg2.elements[%(src2Index)s]);
Element destElem;
%(readDest)s
%(op)s
destReg.elements[i] = htog(destElem);
}
''' % { "op" : op, "readDest" : readDestCode,
"scalarCheck" : scalarCheck if scalar else "",
"src2Index" : "imm" if byElem else "i" }
for reg in range(rCount):
eWalkCode += '''
AA64FpDestP%(reg)d_uw = gtoh(destReg.regs[%(reg)d]);
''' % { "reg" : reg }
if rCount < 4: # zero upper half
for reg in range(rCount, 4):
eWalkCode += '''
AA64FpDestP%(reg)d_uw = 0;
''' % { "reg" : reg }
iop = InstObjParams(name, Name,
"DataX2RegImmOp" if byElem else "DataX2RegOp",
{ "code": eWalkCode,
"r_count": rCount,
"op_class": opClass }, [])
iop.snippets["code"] += zeroSveVecRegUpperPartCode % "AA64FpDest"
if byElem:
header_output += NeonX2RegImmOpDeclare.subst(iop)
else:
header_output += NeonX2RegOpDeclare.subst(iop)
exec_output += NeonXEqualRegOpExecute.subst(iop)
for type in types:
substDict = { "targs" : type,
"class_name" : Name }
exec_output += NeonXExecDeclare.subst(substDict)
def threeUnequalRegInstX(name, Name, opClass, types, op,
bigSrc1, bigSrc2, bigDest, readDest, scalar=False,
byElem=False, hi=False):
assert not (scalar and hi)
global header_output, exec_output
src1Cnt = src2Cnt = destCnt = 2
src1Prefix = src2Prefix = destPrefix = ''
if bigSrc1:
src1Cnt = 4
src1Prefix = 'Big'
if bigSrc2:
src2Cnt = 4
src2Prefix = 'Big'
if bigDest:
destCnt = 4
destPrefix = 'Big'
if byElem:
src2Prefix = 'Full'
eWalkCode = simd64EnabledCheckCode + '''
%sRegVect srcReg1;
%sRegVect srcReg2;
%sRegVect destReg;
''' % (src1Prefix, src2Prefix, destPrefix)
srcReg1 = 0
if hi and not bigSrc1: # long/widening operations
srcReg1 = 2
for reg in range(src1Cnt):
eWalkCode += '''
srcReg1.regs[%(reg)d] = htog(AA64FpOp1P%(srcReg1)d_uw);
''' % { "reg" : reg, "srcReg1" : srcReg1 }
srcReg1 += 1
srcReg2 = 0
if (not byElem) and (hi and not bigSrc2): # long/widening operations
srcReg2 = 2
for reg in range(src2Cnt):
eWalkCode += '''
srcReg2.regs[%(reg)d] = htog(AA64FpOp2P%(srcReg2)d_uw);
''' % { "reg" : reg, "srcReg2" : srcReg2 }
srcReg2 += 1
if byElem:
# 2nd operand has to be read fully
for reg in range(src2Cnt, 4):
eWalkCode += '''
srcReg2.regs[%(reg)d] = htog(AA64FpOp2P%(reg)d_uw);
''' % { "reg" : reg }
if readDest:
for reg in range(destCnt):
eWalkCode += '''
destReg.regs[%(reg)d] = htog(AA64FpDestP%(reg)d_uw);
''' % { "reg" : reg }
readDestCode = ''
if readDest:
readDestCode = 'destElem = gtoh(destReg.elements[i]);'
scalarCheck = '''
if (i != 0) {
destReg.elements[i] = 0;
continue;
}
'''
eWalkCode += '''
for (unsigned i = 0; i < eCount; i++) {
%(scalarCheck)s
%(src1Prefix)sElement srcElem1 = gtoh(srcReg1.elements[i]);
%(src1Prefix)sElement srcElem2 = gtoh(srcReg2.elements[%(src2Index)s]);
%(destPrefix)sElement destElem;
%(readDest)s
%(op)s
destReg.elements[i] = htog(destElem);
}
''' % { "op" : op, "readDest" : readDestCode,
"src1Prefix" : src1Prefix, "src2Prefix" : src2Prefix,
"destPrefix" : destPrefix,
"scalarCheck" : scalarCheck if scalar else "",
"src2Index" : "imm" if byElem else "i" }
destReg = 0
if hi and not bigDest:
# narrowing operations
destReg = 2
for reg in range(destCnt):
eWalkCode += '''
AA64FpDestP%(destReg)d_uw = gtoh(destReg.regs[%(reg)d]);
''' % { "reg" : reg, "destReg": destReg }
destReg += 1
if destCnt < 4:
if hi: # Explicitly merge with lower half
for reg in range(0, destCnt):
eWalkCode += '''
AA64FpDestP%(reg)d_uw = AA64FpDestP%(reg)d_uw;''' % { "reg" : reg }
else: # zero upper half
for reg in range(destCnt, 4):
eWalkCode += '''
AA64FpDestP%(reg)d_uw = 0;''' % { "reg" : reg }
iop = InstObjParams(name, Name,
"DataX2RegImmOp" if byElem else "DataX2RegOp",
{ "code": eWalkCode,
"r_count": 2,
"op_class": opClass }, [])
iop.snippets["code"] += zeroSveVecRegUpperPartCode % "AA64FpDest"
if byElem:
header_output += NeonX2RegImmOpDeclare.subst(iop)
else:
header_output += NeonX2RegOpDeclare.subst(iop)
exec_output += NeonXUnequalRegOpExecute.subst(iop)
for type in types:
substDict = { "targs" : type,
"class_name" : Name }
exec_output += NeonXExecDeclare.subst(substDict)
def threeRegNarrowInstX(name, Name, opClass, types, op, readDest=False,
scalar=False, byElem=False, hi=False):
assert not byElem
threeUnequalRegInstX(name, Name, opClass, types, op,
True, True, False, readDest, scalar, byElem, hi)
def threeRegLongInstX(name, Name, opClass, types, op, readDest=False,
scalar=False, byElem=False, hi=False):
threeUnequalRegInstX(name, Name, opClass, types, op,
False, False, True, readDest, scalar, byElem, hi)
def threeRegWideInstX(name, Name, opClass, types, op, readDest=False,
scalar=False, byElem=False, hi=False):
assert not byElem
threeUnequalRegInstX(name, Name, opClass, types, op,
True, False, True, readDest, scalar, byElem, hi)
def twoEqualRegInstX(name, Name, opClass, types, rCount, op,
readDest=False, scalar=False, byElem=False,
hasImm=False, isDup=False):
global header_output, exec_output
assert (not isDup) or byElem
if byElem:
hasImm = True
if isDup:
eWalkCode = simd64EnabledCheckCode + '''
FullRegVect srcReg1;
RegVect destReg;
'''
else:
eWalkCode = simd64EnabledCheckCode + '''
RegVect srcReg1, destReg;
'''
for reg in range(4 if isDup else rCount):
eWalkCode += '''
srcReg1.regs[%(reg)d] = htog(AA64FpOp1P%(reg)d_uw);
''' % { "reg" : reg }
if readDest:
eWalkCode += '''
destReg.regs[%(reg)d] = htog(AA64FpDestP%(reg)d_uw);
''' % { "reg" : reg }
readDestCode = ''
if readDest:
readDestCode = 'destElem = gtoh(destReg.elements[i]);'
scalarCheck = '''
if (i != 0) {
destReg.elements[i] = 0;
continue;
}
'''
eWalkCode += '''
for (unsigned i = 0; i < eCount; i++) {
%(scalarCheck)s
unsigned j = i;
Element srcElem1 = gtoh(srcReg1.elements[%(src1Index)s]);
Element destElem;
%(readDest)s
%(op)s
destReg.elements[j] = htog(destElem);
}
''' % { "op" : op, "readDest" : readDestCode,
"scalarCheck" : scalarCheck if scalar else "",
"src1Index" : "imm" if byElem else "i" }
for reg in range(rCount):
eWalkCode += '''
AA64FpDestP%(reg)d_uw = gtoh(destReg.regs[%(reg)d]);
''' % { "reg" : reg }
if rCount < 4: # zero upper half
for reg in range(rCount, 4):
eWalkCode += '''
AA64FpDestP%(reg)d_uw = 0;
''' % { "reg" : reg }
iop = InstObjParams(name, Name,
"DataX1RegImmOp" if hasImm else "DataX1RegOp",
{ "code": eWalkCode,
"r_count": rCount,
"op_class": opClass }, [])
iop.snippets["code"] += zeroSveVecRegUpperPartCode % "AA64FpDest"
if hasImm:
header_output += NeonX1RegImmOpDeclare.subst(iop)
else:
header_output += NeonX1RegOpDeclare.subst(iop)
exec_output += NeonXEqualRegOpExecute.subst(iop)
for type in types:
substDict = { "targs" : type,
"class_name" : Name }
exec_output += NeonXExecDeclare.subst(substDict)
def twoRegLongInstX(name, Name, opClass, types, op, readDest=False,
hi=False, hasImm=False):
global header_output, exec_output
eWalkCode = simd64EnabledCheckCode + '''
RegVect srcReg1;
BigRegVect destReg;
'''
destReg = 0 if not hi else 2
for reg in range(2):
eWalkCode += '''
srcReg1.regs[%(reg)d] = htog(AA64FpOp1P%(destReg)d_uw);
''' % { "reg" : reg, "destReg": destReg }
destReg += 1
destReg = 0 if not hi else 2
if readDest:
for reg in range(4):
eWalkCode += '''
destReg.regs[%(reg)d] = htog(AA64FpDestP%(reg)d_uw);
''' % { "reg" : reg }
destReg += 1
readDestCode = ''
if readDest:
readDestCode = 'destReg = gtoh(destReg.elements[i]);'
eWalkCode += '''
for (unsigned i = 0; i < eCount; i++) {
Element srcElem1 = gtoh(srcReg1.elements[i]);
BigElement destElem;
%(readDest)s
%(op)s
destReg.elements[i] = htog(destElem);
}
''' % { "op" : op, "readDest" : readDestCode }
for reg in range(4):
eWalkCode += '''
AA64FpDestP%(reg)d_uw = gtoh(destReg.regs[%(reg)d]);
''' % { "reg" : reg }
iop = InstObjParams(name, Name,
"DataX1RegImmOp" if hasImm else "DataX1RegOp",
{ "code": eWalkCode,
"r_count": 2,
"op_class": opClass }, [])
iop.snippets["code"] += zeroSveVecRegUpperPartCode % "AA64FpDest"
if hasImm:
header_output += NeonX1RegImmOpDeclare.subst(iop)
else:
header_output += NeonX1RegOpDeclare.subst(iop)
exec_output += NeonXUnequalRegOpExecute.subst(iop)
for type in types:
substDict = { "targs" : type,
"class_name" : Name }
exec_output += NeonXExecDeclare.subst(substDict)
def twoRegNarrowInstX(name, Name, opClass, types, op, readDest=False,
scalar=False, hi=False, hasImm=False):
global header_output, exec_output
eWalkCode = simd64EnabledCheckCode + '''
BigRegVect srcReg1;
RegVect destReg;
'''
for reg in range(4):
eWalkCode += '''
srcReg1.regs[%(reg)d] = htog(AA64FpOp1P%(reg)d_uw);
''' % { "reg" : reg }
if readDest:
for reg in range(2):
eWalkCode += '''
destReg.regs[%(reg)d] = htog(AA64FpDestP%(reg)d_uw);
''' % { "reg" : reg }
else:
eWalkCode += '''
destReg.elements[0] = 0;
''' % { "reg" : reg }
readDestCode = ''
if readDest:
readDestCode = 'destElem = gtoh(destReg.elements[i]);'
scalarCheck = '''
if (i != 0) {
destReg.elements[i] = 0;
continue;
}
'''
eWalkCode += '''
for (unsigned i = 0; i < eCount; i++) {
%(scalarCheck)s
BigElement srcElem1 = gtoh(srcReg1.elements[i]);
Element destElem;
%(readDest)s
%(op)s
destReg.elements[i] = htog(destElem);
}
''' % { "op" : op, "readDest" : readDestCode,
"scalarCheck" : scalarCheck if scalar else "" }
destReg = 0 if not hi else 2
for reg in range(2):
eWalkCode += '''
AA64FpDestP%(destReg)d_uw = gtoh(destReg.regs[%(reg)d]);
''' % { "reg" : reg, "destReg": destReg }
destReg += 1
if hi:
for reg in range(0, 2): # Explicitly merge with the lower half
eWalkCode += '''
AA64FpDestP%(reg)d_uw = AA64FpDestP%(reg)d_uw;''' % { "reg" : reg }
else:
for reg in range(2, 4): # zero upper half
eWalkCode += '''
AA64FpDestP%(reg)d_uw = 0;
''' % { "reg" : reg }
iop = InstObjParams(name, Name,
"DataX1RegImmOp" if hasImm else "DataX1RegOp",
{ "code": eWalkCode,
"r_count": 2,
"op_class": opClass }, [])
iop.snippets["code"] += zeroSveVecRegUpperPartCode % "AA64FpDest"
if hasImm:
header_output += NeonX1RegImmOpDeclare.subst(iop)
else:
header_output += NeonX1RegOpDeclare.subst(iop)
exec_output += NeonXUnequalRegOpExecute.subst(iop)
for type in types:
substDict = { "targs" : type,
"class_name" : Name }
exec_output += NeonXExecDeclare.subst(substDict)
def threeRegScrambleInstX(name, Name, opClass, types, rCount, op):
global header_output, exec_output
eWalkCode = simd64EnabledCheckCode + '''
RegVect srcReg1, srcReg2, destReg;
'''
for reg in range(rCount):
eWalkCode += '''
srcReg1.regs[%(reg)d] = htog(AA64FpOp1P%(reg)d_uw);
srcReg2.regs[%(reg)d] = htog(AA64FpOp2P%(reg)d_uw);
''' % { "reg" : reg }
eWalkCode += op
for reg in range(rCount):
eWalkCode += '''
AA64FpDestP%(reg)d_uw = gtoh(destReg.regs[%(reg)d]);
''' % { "reg" : reg }
if rCount < 4:
for reg in range(rCount, 4):
eWalkCode += '''
AA64FpDestP%(reg)d_uw = 0;
''' % { "reg" : reg }
iop = InstObjParams(name, Name,
"DataX2RegOp",
{ "code": eWalkCode,
"r_count": rCount,
"op_class": opClass }, [])
iop.snippets["code"] += zeroSveVecRegUpperPartCode % "AA64FpDest"
header_output += NeonX2RegOpDeclare.subst(iop)
exec_output += NeonXEqualRegOpExecute.subst(iop)
for type in types:
substDict = { "targs" : type,
"class_name" : Name }
exec_output += NeonXExecDeclare.subst(substDict)
def insFromVecElemInstX(name, Name, opClass, types, rCount):
global header_output, exec_output
eWalkCode = simd64EnabledCheckCode + '''
FullRegVect srcReg1;
RegVect destReg;
'''
for reg in range(4):
eWalkCode += '''
srcReg1.regs[%(reg)d] = htog(AA64FpOp1P%(reg)d_uw);
''' % { "reg" : reg }
for reg in range(rCount):
eWalkCode += '''
destReg.regs[%(reg)d] = htog(AA64FpDestP%(reg)d_uw);
''' % { "reg" : reg }
eWalkCode += '''
Element srcElem1 = gtoh(srcReg1.elements[imm2]);
Element destElem = srcElem1;
destReg.elements[imm1] = htog(destElem);
'''
for reg in range(rCount):
eWalkCode += '''
AA64FpDestP%(reg)d_uw = gtoh(destReg.regs[%(reg)d]);
''' % { "reg" : reg }
iop = InstObjParams(name, Name,
"DataX1Reg2ImmOp",
{ "code": eWalkCode,
"r_count": rCount,
"op_class": opClass }, [])
iop.snippets["code"] += zeroSveVecRegUpperPartCode % "AA64FpDest"
header_output += NeonX1Reg2ImmOpDeclare.subst(iop)
exec_output += NeonXEqualRegOpExecute.subst(iop)
for type in types:
substDict = { "targs" : type,
"class_name" : Name }
exec_output += NeonXExecDeclare.subst(substDict)
def twoRegPairwiseScInstX(name, Name, opClass, types, rCount, op):
global header_output, exec_output
eWalkCode = simd64EnabledCheckCode + '''
RegVect srcReg1, destReg;
'''
for reg in range(rCount):
eWalkCode += '''
srcReg1.regs[%(reg)d] = htog(AA64FpOp1P%(reg)d_uw);
''' % { "reg" : reg }
eWalkCode += '''
Element srcElem1 = gtoh(srcReg1.elements[0]);
Element srcElem2 = gtoh(srcReg1.elements[1]);
Element destElem;
%(op)s
destReg.elements[0] = htog(destElem);
''' % { "op" : op }
destCnt = rCount / 2
for reg in range(destCnt):
eWalkCode += '''
AA64FpDestP%(reg)d_uw = gtoh(destReg.regs[%(reg)d]);
''' % { "reg" : reg }
for reg in range(destCnt, 4): # zero upper half
eWalkCode += '''
AA64FpDestP%(reg)d_uw = 0;
''' % { "reg" : reg }
iop = InstObjParams(name, Name,
"DataX1RegOp",
{ "code": eWalkCode,
"r_count": rCount,
"op_class": opClass }, [])
iop.snippets["code"] += zeroSveVecRegUpperPartCode % "AA64FpDest"
header_output += NeonX1RegOpDeclare.subst(iop)
exec_output += NeonXEqualRegOpExecute.subst(iop)
for type in types:
substDict = { "targs" : type,
"class_name" : Name }
exec_output += NeonXExecDeclare.subst(substDict)
def twoRegAcrossInstX(name, Name, opClass, types, rCount, op,
doubleDest=False, long=False):
global header_output, exec_output
destPrefix = "Big" if long else ""
eWalkCode = simd64EnabledCheckCode + '''
RegVect srcReg1;
%sRegVect destReg;
''' % destPrefix
for reg in range(rCount):
eWalkCode += '''
srcReg1.regs[%(reg)d] = htog(AA64FpOp1P%(reg)d_uw);
''' % { "reg" : reg }
eWalkCode += '''
destReg.regs[0] = 0;
%(destPrefix)sElement destElem = 0;
for (unsigned i = 0; i < eCount; i++) {
Element srcElem1 = gtoh(srcReg1.elements[i]);
if (i == 0) {
destElem = srcElem1;
} else {
%(op)s
}
}
destReg.elements[0] = htog(destElem);
''' % { "op" : op, "destPrefix" : destPrefix }
destCnt = 2 if doubleDest else 1
for reg in range(destCnt):
eWalkCode += '''
AA64FpDestP%(reg)d_uw = gtoh(destReg.regs[%(reg)d]);
''' % { "reg" : reg }
for reg in range(destCnt, 4): # zero upper half
eWalkCode += '''
AA64FpDestP%(reg)d_uw = 0;
''' % { "reg" : reg }
iop = InstObjParams(name, Name,
"DataX1RegOp",
{ "code": eWalkCode,
"r_count": rCount,
"op_class": opClass }, [])
iop.snippets["code"] += zeroSveVecRegUpperPartCode % "AA64FpDest"
header_output += NeonX1RegOpDeclare.subst(iop)
if long:
exec_output += NeonXUnequalRegOpExecute.subst(iop)
else:
exec_output += NeonXEqualRegOpExecute.subst(iop)
for type in types:
substDict = { "targs" : type,
"class_name" : Name }
exec_output += NeonXExecDeclare.subst(substDict)
def twoRegCondenseInstX(name, Name, opClass, types, rCount, op,
readDest=False):
global header_output, exec_output
eWalkCode = simd64EnabledCheckCode + '''
RegVect srcRegs;
BigRegVect destReg;
'''
for reg in range(rCount):
eWalkCode += '''
srcRegs.regs[%(reg)d] = htog(AA64FpOp1P%(reg)d_uw);
''' % { "reg" : reg }
if readDest:
eWalkCode += '''
destReg.regs[%(reg)d] = htog(AA64FpDestP%(reg)d_uw);
''' % { "reg" : reg }
readDestCode = ''
if readDest:
readDestCode = 'destElem = gtoh(destReg.elements[i]);'
eWalkCode += '''
for (unsigned i = 0; i < eCount / 2; i++) {
Element srcElem1 = gtoh(srcRegs.elements[2 * i]);
Element srcElem2 = gtoh(srcRegs.elements[2 * i + 1]);
BigElement destElem;
%(readDest)s
%(op)s
destReg.elements[i] = htog(destElem);
}
''' % { "op" : op, "readDest" : readDestCode }
for reg in range(rCount):
eWalkCode += '''
AA64FpDestP%(reg)d_uw = gtoh(destReg.regs[%(reg)d]);
''' % { "reg" : reg }
if rCount < 4: # zero upper half
for reg in range(rCount, 4):
eWalkCode += '''
AA64FpDestP%(reg)d_uw = 0;
''' % { "reg" : reg }
iop = InstObjParams(name, Name,
"DataX1RegOp",
{ "code": eWalkCode,
"r_count": rCount,
"op_class": opClass }, [])
iop.snippets["code"] += zeroSveVecRegUpperPartCode % "AA64FpDest"
header_output += NeonX1RegOpDeclare.subst(iop)
exec_output += NeonXUnequalRegOpExecute.subst(iop)
for type in types:
substDict = { "targs" : type,
"class_name" : Name }
exec_output += NeonXExecDeclare.subst(substDict)
def oneRegImmInstX(name, Name, opClass, types, rCount, op, readDest=False):
global header_output, exec_output
eWalkCode = simd64EnabledCheckCode + '''
RegVect destReg;
'''
if readDest:
for reg in range(rCount):
eWalkCode += '''
destReg.regs[%(reg)d] = htog(AA64FpDestP%(reg)d_uw);
''' % { "reg" : reg }
readDestCode = ''
if readDest:
readDestCode = 'destElem = gtoh(destReg.elements[i]);'
eWalkCode += '''
for (unsigned i = 0; i < eCount; i++) {
Element destElem;
%(readDest)s
%(op)s
destReg.elements[i] = htog(destElem);
}
''' % { "op" : op, "readDest" : readDestCode }
for reg in range(rCount):
eWalkCode += '''
AA64FpDestP%(reg)d_uw = gtoh(destReg.regs[%(reg)d]);
''' % { "reg" : reg }
if rCount < 4: # zero upper half
for reg in range(rCount, 4):
eWalkCode += '''
AA64FpDestP%(reg)d_uw = 0;
''' % { "reg" : reg }
iop = InstObjParams(name, Name,
"DataXImmOnlyOp",
{ "code": eWalkCode,
"r_count": rCount,
"op_class": opClass }, [])
iop.snippets["code"] += zeroSveVecRegUpperPartCode % "AA64FpDest"
header_output += NeonX1RegImmOnlyOpDeclare.subst(iop)
exec_output += NeonXEqualRegOpExecute.subst(iop)
for type in types:
substDict = { "targs" : type,
"class_name" : Name }
exec_output += NeonXExecDeclare.subst(substDict)
def dupGprInstX(name, Name, opClass, types, rCount, gprSpec):
global header_output, exec_output
eWalkCode = simd64EnabledCheckCode + '''
RegVect destReg;
for (unsigned i = 0; i < eCount; i++) {
destReg.elements[i] = htog((Element) %sOp1);
}
''' % gprSpec
for reg in range(rCount):
eWalkCode += '''
AA64FpDestP%(reg)d_uw = gtoh(destReg.regs[%(reg)d]);
''' % { "reg" : reg }
if rCount < 4: # zero upper half
for reg in range(rCount, 4):
eWalkCode += '''
AA64FpDestP%(reg)d_uw = 0;
''' % { "reg" : reg }
iop = InstObjParams(name, Name,
"DataX1RegOp",
{ "code": eWalkCode,
"r_count": rCount,
"op_class": opClass }, [])
iop.snippets["code"] += zeroSveVecRegUpperPartCode % "AA64FpDest"
header_output += NeonX1RegOpDeclare.subst(iop)
exec_output += NeonXEqualRegOpExecute.subst(iop)
for type in types:
substDict = { "targs" : type,
"class_name" : Name }
exec_output += NeonXExecDeclare.subst(substDict)
def extInstX(name, Name, opClass, types, rCount, op):
global header_output, exec_output
eWalkCode = simd64EnabledCheckCode + '''
RegVect srcReg1, srcReg2, destReg;
'''
for reg in range(rCount):
eWalkCode += '''
srcReg1.regs[%(reg)d] = htog(AA64FpOp1P%(reg)d_uw);
srcReg2.regs[%(reg)d] = htog(AA64FpOp2P%(reg)d_uw);
''' % { "reg" : reg }
eWalkCode += op
for reg in range(rCount):
eWalkCode += '''
AA64FpDestP%(reg)d_uw = gtoh(destReg.regs[%(reg)d]);
''' % { "reg" : reg }
if rCount < 4: # zero upper half
for reg in range(rCount, 4):
eWalkCode += '''
AA64FpDestP%(reg)d_uw = 0;
''' % { "reg" : reg }
iop = InstObjParams(name, Name,
"DataX2RegImmOp",
{ "code": eWalkCode,
"r_count": rCount,
"op_class": opClass }, [])
iop.snippets["code"] += zeroSveVecRegUpperPartCode % "AA64FpDest"
header_output += NeonX2RegImmOpDeclare.subst(iop)
exec_output += NeonXEqualRegOpExecute.subst(iop)
for type in types:
substDict = { "targs" : type,
"class_name" : Name }
exec_output += NeonXExecDeclare.subst(substDict)
def insFromGprInstX(name, Name, opClass, types, rCount, gprSpec):
global header_output, exec_output
eWalkCode = simd64EnabledCheckCode + '''
RegVect destReg;
'''
for reg in range(rCount):
eWalkCode += '''
destReg.regs[%(reg)d] = htog(AA64FpDestP%(reg)d_uw);
''' % { "reg" : reg }
eWalkCode += '''
destReg.elements[imm] = htog((Element) %sOp1);
''' % gprSpec
for reg in range(rCount):
eWalkCode += '''
AA64FpDestP%(reg)d_uw = gtoh(destReg.regs[%(reg)d]);
''' % { "reg" : reg }
iop = InstObjParams(name, Name,
"DataX1RegImmOp",
{ "code": eWalkCode,
"r_count": rCount,
"op_class": opClass }, [])
iop.snippets["code"] += zeroSveVecRegUpperPartCode % "AA64FpDest"
header_output += NeonX1RegImmOpDeclare.subst(iop)
exec_output += NeonXEqualRegOpExecute.subst(iop)
for type in types:
substDict = { "targs" : type,
"class_name" : Name }
exec_output += NeonXExecDeclare.subst(substDict)
def insToGprInstX(name, Name, opClass, types, rCount, gprSpec,
signExt=False):
global header_output, exec_output
eWalkCode = simd64EnabledCheckCode + '''
FullRegVect srcReg;
'''
for reg in range(4):
eWalkCode += '''
srcReg.regs[%(reg)d] = htog(AA64FpOp1P%(reg)d_uw);
''' % { "reg" : reg }
if signExt:
eWalkCode += '''
%sDest = sext<sizeof(Element) * 8>(srcReg.elements[imm]);
''' % gprSpec
else:
eWalkCode += '''
%sDest = srcReg.elements[imm];
''' % gprSpec
iop = InstObjParams(name, Name,
"DataX1RegImmOp",
{ "code": eWalkCode,
"r_count": rCount,
"op_class": opClass }, [])
header_output += NeonX1RegImmOpDeclare.subst(iop)
exec_output += NeonXEqualRegOpExecute.subst(iop)
for type in types:
substDict = { "targs" : type,
"class_name" : Name }
exec_output += NeonXExecDeclare.subst(substDict)
def tbxTblInstX(name, Name, opClass, types, length, isTbl, rCount):
global header_output, decoder_output, exec_output
code = simd64EnabledCheckCode + '''
union
{
uint8_t bytes[64];
uint32_t regs[16];
} table;
union
{
uint8_t bytes[%(rCount)d * 4];
uint32_t regs[%(rCount)d];
} destReg, srcReg2;
const unsigned length = %(length)d;
const bool isTbl = %(isTbl)s;
''' % { "rCount" : rCount, "length" : length, "isTbl" : isTbl }
for reg in range(rCount):
code += '''
srcReg2.regs[%(reg)d] = htog(AA64FpOp2P%(reg)d_uw);
destReg.regs[%(reg)d] = htog(AA64FpDestP%(reg)d_uw);
''' % { "reg" : reg }
for reg in range(16):
if reg < length * 4:
code += '''
table.regs[%(reg)d] = htog(AA64FpOp1P%(p)dV%(v)dS_uw);
''' % { "reg" : reg, "p" : reg % 4, "v" : reg / 4 }
else:
code += '''
table.regs[%(reg)d] = 0;
''' % { "reg" : reg }
code += '''
for (unsigned i = 0; i < sizeof(destReg); i++) {
uint8_t index = srcReg2.bytes[i];
if (index < 16 * length) {
destReg.bytes[i] = table.bytes[index];
} else {
if (isTbl)
destReg.bytes[i] = 0;
// else destReg.bytes[i] unchanged
}
}
'''
for reg in range(rCount):
code += '''
AA64FpDestP%(reg)d_uw = gtoh(destReg.regs[%(reg)d]);
''' % { "reg" : reg }
if rCount < 4: # zero upper half
for reg in range(rCount, 4):
code += '''
AA64FpDestP%(reg)d_uw = 0;
''' % { "reg" : reg }
iop = InstObjParams(name, Name,
"DataX2RegOp",
{ "code": code,
"r_count": rCount,
"op_class": opClass }, [])
iop.snippets["code"] += zeroSveVecRegUpperPartCode % "AA64FpDest"
header_output += NeonX2RegOpDeclare.subst(iop)
exec_output += NeonXEqualRegOpExecute.subst(iop)
for type in types:
substDict = { "targs" : type,
"class_name" : Name }
exec_output += NeonXExecDeclare.subst(substDict)
# ABS
absCode = '''
if (srcElem1 < 0) {
destElem = -srcElem1;
} else {
destElem = srcElem1;
}
'''
twoEqualRegInstX("abs", "AbsDX", "SimdAluOp", signedTypes, 2, absCode)
twoEqualRegInstX("abs", "AbsQX", "SimdAluOp", signedTypes, 4, absCode)
# ADD
addCode = "destElem = srcElem1 + srcElem2;"
threeEqualRegInstX("add", "AddDX", "SimdAddOp", unsignedTypes, 2, addCode)
threeEqualRegInstX("add", "AddQX", "SimdAddOp", unsignedTypes, 4, addCode)
# ADDHN, ADDHN2
addhnCode = '''
destElem = ((BigElement)srcElem1 + (BigElement)srcElem2) >>
(sizeof(Element) * 8);
'''
threeRegNarrowInstX("addhn", "AddhnX", "SimdAddOp", smallUnsignedTypes,
addhnCode)
threeRegNarrowInstX("addhn2", "Addhn2X", "SimdAddOp", smallUnsignedTypes,
addhnCode, hi=True)
# ADDP (scalar)
twoRegPairwiseScInstX("addp", "AddpScQX", "SimdAddOp", ("uint64_t",), 4,
addCode)
# ADDP (vector)
threeEqualRegInstX("addp", "AddpDX", "SimdAddOp", smallUnsignedTypes, 2,
addCode, pairwise=True)
threeEqualRegInstX("addp", "AddpQX", "SimdAddOp", unsignedTypes, 4,
addCode, pairwise=True)
# ADDV
# Note: SimdAddOp can be a bit optimistic here
addAcrossCode = "destElem += srcElem1;"
twoRegAcrossInstX("addv", "AddvDX", "SimdAddOp", ("uint8_t", "uint16_t"),
2, addAcrossCode)
twoRegAcrossInstX("addv", "AddvQX", "SimdAddOp", smallUnsignedTypes, 4,
addAcrossCode)
# AND
andCode = "destElem = srcElem1 & srcElem2;"
threeEqualRegInstX("and", "AndDX", "SimdAluOp", ("uint64_t",), 2, andCode)
threeEqualRegInstX("and", "AndQX", "SimdAluOp", ("uint64_t",), 4, andCode)
# BIC (immediate)
bicImmCode = "destElem &= ~imm;"
oneRegImmInstX("bic", "BicImmDX", "SimdAluOp", ("uint64_t",), 2,
bicImmCode, True)
oneRegImmInstX("bic", "BicImmQX", "SimdAluOp", ("uint64_t",), 4,
bicImmCode, True)
# BIC (register)
bicCode = "destElem = srcElem1 & ~srcElem2;"
threeEqualRegInstX("bic", "BicDX", "SimdAluOp", ("uint64_t",), 2, bicCode)
threeEqualRegInstX("bic", "BicQX", "SimdAluOp", ("uint64_t",), 4, bicCode)
# BIF
bifCode = "destElem = (destElem & srcElem2) | (srcElem1 & ~srcElem2);"
threeEqualRegInstX("bif", "BifDX", "SimdAluOp", ("uint64_t",), 2, bifCode,
True)
threeEqualRegInstX("bif", "BifQX", "SimdAluOp", ("uint64_t",), 4, bifCode,
True)
# BIT
bitCode = "destElem = (srcElem1 & srcElem2) | (destElem & ~srcElem2);"
threeEqualRegInstX("bit", "BitDX", "SimdAluOp", ("uint64_t",), 2, bitCode,
True)
threeEqualRegInstX("bit", "BitQX", "SimdAluOp", ("uint64_t",), 4, bitCode,
True)
# BSL
bslCode = "destElem = (srcElem1 & destElem) | (srcElem2 & ~destElem);"
threeEqualRegInstX("bsl", "BslDX", "SimdAluOp", ("uint64_t",), 2, bslCode,
True)
threeEqualRegInstX("bsl", "BslQX", "SimdAluOp", ("uint64_t",), 4, bslCode,
True)
# CLS
clsCode = '''
unsigned count = 0;
if (srcElem1 < 0) {
srcElem1 <<= 1;
while (srcElem1 < 0 && count < sizeof(Element) * 8 - 1) {
count++;
srcElem1 <<= 1;
}
} else {
srcElem1 <<= 1;
while (srcElem1 >= 0 && count < sizeof(Element) * 8 - 1) {
count++;
srcElem1 <<= 1;
}
}
destElem = count;
'''
twoEqualRegInstX("cls", "ClsDX", "SimdAluOp", smallSignedTypes, 2, clsCode)
twoEqualRegInstX("cls", "ClsQX", "SimdAluOp", smallSignedTypes, 4, clsCode)
# CLZ
clzCode = '''
unsigned count = 0;
while (srcElem1 >= 0 && count < sizeof(Element) * 8) {
count++;
srcElem1 <<= 1;
}
destElem = count;
'''
twoEqualRegInstX("clz", "ClzDX", "SimdAluOp", smallSignedTypes, 2, clzCode)
twoEqualRegInstX("clz", "ClzQX", "SimdAluOp", smallSignedTypes, 4, clzCode)
# CMEQ (register)
cmeqCode = "destElem = (srcElem1 == srcElem2) ? (Element)(-1) : 0;"
threeEqualRegInstX("cmeq", "CmeqDX", "SimdCmpOp", unsignedTypes, 2,
cmeqCode)
threeEqualRegInstX("cmeq", "CmeqQX", "SimdCmpOp", unsignedTypes, 4,
cmeqCode)
# CMEQ (zero)
cmeqZeroCode = "destElem = (srcElem1 == 0) ? (Element)(-1) : 0;"
twoEqualRegInstX("cmeq", "CmeqZeroDX", "SimdCmpOp", signedTypes, 2,
cmeqZeroCode)
twoEqualRegInstX("cmeq", "CmeqZeroQX", "SimdCmpOp", signedTypes, 4,
cmeqZeroCode)
# CMGE (register)
cmgeCode = "destElem = (srcElem1 >= srcElem2) ? (Element)(-1) : 0;"
threeEqualRegInstX("cmge", "CmgeDX", "SimdCmpOp", signedTypes, 2, cmgeCode)
threeEqualRegInstX("cmge", "CmgeQX", "SimdCmpOp", signedTypes, 4, cmgeCode)
# CMGE (zero)
cmgeZeroCode = "destElem = (srcElem1 >= 0) ? (Element)(-1) : 0;"
twoEqualRegInstX("cmge", "CmgeZeroDX", "SimdCmpOp", signedTypes, 2,
cmgeZeroCode)
twoEqualRegInstX("cmge", "CmgeZeroQX", "SimdCmpOp", signedTypes, 4,
cmgeZeroCode)
# CMGT (register)
cmgtCode = "destElem = (srcElem1 > srcElem2) ? (Element)(-1) : 0;"
threeEqualRegInstX("cmgt", "CmgtDX", "SimdCmpOp", signedTypes, 2, cmgtCode)
threeEqualRegInstX("cmgt", "CmgtQX", "SimdCmpOp", signedTypes, 4, cmgtCode)
# CMGT (zero)
cmgtZeroCode = "destElem = (srcElem1 > 0) ? (Element)(-1) : 0;"
twoEqualRegInstX("cmgt", "CmgtZeroDX", "SimdCmpOp", signedTypes, 2,
cmgtZeroCode)
twoEqualRegInstX("cmgt", "CmgtZeroQX", "SimdCmpOp", signedTypes, 4,
cmgtZeroCode)
# CMHI (register)
threeEqualRegInstX("cmhi", "CmhiDX", "SimdCmpOp", unsignedTypes, 2,
cmgtCode)
threeEqualRegInstX("cmhi", "CmhiQX", "SimdCmpOp", unsignedTypes, 4,
cmgtCode)
# CMHS (register)
threeEqualRegInstX("cmhs", "CmhsDX", "SimdCmpOp", unsignedTypes, 2,
cmgeCode)
threeEqualRegInstX("cmhs", "CmhsQX", "SimdCmpOp", unsignedTypes, 4,
cmgeCode)
# CMLE (zero)
cmleZeroCode = "destElem = (srcElem1 <= 0) ? (Element)(-1) : 0;"
twoEqualRegInstX("cmle", "CmleZeroDX", "SimdCmpOp", signedTypes, 2,
cmleZeroCode)
twoEqualRegInstX("cmle", "CmleZeroQX", "SimdCmpOp", signedTypes, 4,
cmleZeroCode)
# CMLT (zero)
cmltZeroCode = "destElem = (srcElem1 < 0) ? (Element)(-1) : 0;"
twoEqualRegInstX("cmlt", "CmltZeroDX", "SimdCmpOp", signedTypes, 2,
cmltZeroCode)
twoEqualRegInstX("cmlt", "CmltZeroQX", "SimdCmpOp", signedTypes, 4,
cmltZeroCode)
# CMTST (register)
tstCode = "destElem = (srcElem1 & srcElem2) ? (Element)(-1) : 0;"
threeEqualRegInstX("cmtst", "CmtstDX", "SimdAluOp", unsignedTypes, 2,
tstCode)
threeEqualRegInstX("cmtst", "CmtstQX", "SimdAluOp", unsignedTypes, 4,
tstCode)
# CNT
cntCode = '''
unsigned count = 0;
while (srcElem1 && count < sizeof(Element) * 8) {
count += srcElem1 & 0x1;
srcElem1 >>= 1;
}
destElem = count;
'''
twoEqualRegInstX("cnt", "CntDX", "SimdAluOp", ("uint8_t",), 2, cntCode)
twoEqualRegInstX("cnt", "CntQX", "SimdAluOp", ("uint8_t",), 4, cntCode)
# DUP (element)
dupCode = "destElem = srcElem1;"
twoEqualRegInstX("dup", "DupElemDX", "SimdMiscOp", smallUnsignedTypes, 2,
dupCode, isDup=True, byElem=True)
twoEqualRegInstX("dup", "DupElemQX", "SimdMiscOp", unsignedTypes, 4,
dupCode, isDup=True, byElem=True)
twoEqualRegInstX("dup", "DupElemScX", "SimdMiscOp", unsignedTypes, 4,
dupCode, isDup=True, byElem=True, scalar=True)
# DUP (general register)
dupGprInstX("dup", "DupGprWDX", "SimdMiscOp", smallUnsignedTypes, 2, 'W')
dupGprInstX("dup", "DupGprWQX", "SimdMiscOp", smallUnsignedTypes, 4, 'W')
dupGprInstX("dup", "DupGprXQX", "SimdMiscOp", ("uint64_t",), 4, 'X')
# EOR
eorCode = "destElem = srcElem1 ^ srcElem2;"
threeEqualRegInstX("eor", "EorDX", "SimdAluOp", ("uint64_t",), 2, eorCode)
threeEqualRegInstX("eor", "EorQX", "SimdAluOp", ("uint64_t",), 4, eorCode)
# EXT
extCode = '''
for (unsigned i = 0; i < eCount; i++) {
unsigned index = i + imm;
if (index < eCount) {
destReg.elements[i] = srcReg1.elements[index];
} else {
index -= eCount;
if (index >= eCount) {
fault = std::make_shared<UndefinedInstruction>(
machInst, false, mnemonic);
} else {
destReg.elements[i] = srcReg2.elements[index];
}
}
}
'''
extInstX("Ext", "ExtDX", "SimdMiscOp", ("uint8_t",), 2, extCode)
extInstX("Ext", "ExtQX", "SimdMiscOp", ("uint8_t",), 4, extCode)
# FABD
fpOp = '''
FPSCR fpscr = (FPSCR) FpscrExc;
destElem = %s;
FpscrExc = fpscr;
'''
fabdCode = fpOp % "fplibAbs<Element>(fplibSub(srcElem1, srcElem2, fpscr))"
threeEqualRegInstX("fabd", "FabdDX", "SimdFloatAddOp", smallFloatTypes, 2,
fabdCode)
threeEqualRegInstX("fabd", "FabdQX", "SimdFloatAddOp", floatTypes, 4,
fabdCode)
threeEqualRegInstX("fabd", "FabdScX", "SimdFloatAddOp", floatTypes, 4,
fabdCode, scalar=True)
# FABS
fabsCode = fpOp % "fplibAbs<Element>(srcElem1)"
twoEqualRegInstX("Abs", "FabsDX", "SimdFloatAluOp", smallFloatTypes, 2,
fabsCode)
twoEqualRegInstX("Abs", "FabsQX", "SimdFloatAluOp", floatTypes, 4,
fabsCode)
# FACGE
fpCmpAbsOp = fpOp % ("fplibCompare%s<Element>(fplibAbs<Element>(srcElem1),"
" fplibAbs<Element>(srcElem2), fpscr) ? -1 : 0")
facgeCode = fpCmpAbsOp % "GE"
threeEqualRegInstX("facge", "FacgeDX", "SimdFloatCmpOp", smallFloatTypes,
2, facgeCode)
threeEqualRegInstX("facge", "FacgeQX", "SimdFloatCmpOp", floatTypes, 4,
facgeCode)
threeEqualRegInstX("facge", "FacgeScX", "SimdFloatCmpOp", floatTypes, 4,
facgeCode, scalar=True)
# FACGT
facgtCode = fpCmpAbsOp % "GT"
threeEqualRegInstX("facgt", "FacgtDX", "SimdFloatCmpOp", smallFloatTypes,
2, facgtCode)
threeEqualRegInstX("facgt", "FacgtQX", "SimdFloatCmpOp", floatTypes, 4,
facgtCode)
threeEqualRegInstX("facgt", "FacgtScX", "SimdFloatCmpOp", floatTypes, 4,
facgtCode, scalar=True)
# FADD
fpBinOp = fpOp % "fplib%s<Element>(srcElem1, srcElem2, fpscr)"
faddCode = fpBinOp % "Add"
threeEqualRegInstX("fadd", "FaddDX", "SimdFloatAddOp", smallFloatTypes, 2,
faddCode)
threeEqualRegInstX("fadd", "FaddQX", "SimdFloatAddOp", floatTypes, 4,
faddCode)
# FADDP (scalar)
twoRegPairwiseScInstX("faddp", "FaddpScDX", "SimdFloatAddOp",
("uint32_t",), 2, faddCode)
twoRegPairwiseScInstX("faddp", "FaddpScQX", "SimdFloatAddOp",
("uint64_t",), 4, faddCode)
# FADDP (vector)
threeEqualRegInstX("faddp", "FaddpDX", "SimdFloatAddOp", smallFloatTypes,
2, faddCode, pairwise=True)
threeEqualRegInstX("faddp", "FaddpQX", "SimdFloatAddOp", floatTypes, 4,
faddCode, pairwise=True)
# FCMEQ (register)
fpCmpOp = fpOp % ("fplibCompare%s<Element>(srcElem1, srcElem2, fpscr) ?"
" -1 : 0")
fcmeqCode = fpCmpOp % "EQ"
threeEqualRegInstX("fcmeq", "FcmeqDX", "SimdFloatCmpOp", smallFloatTypes,
2, fcmeqCode)
threeEqualRegInstX("fcmeq", "FcmeqQX", "SimdFloatCmpOp", floatTypes, 4,
fcmeqCode)
threeEqualRegInstX("fcmeq", "FcmeqScX", "SimdFloatCmpOp", floatTypes, 4,
fcmeqCode, scalar=True)
# FCMEQ (zero)
fpCmpZeroOp = fpOp % "fplibCompare%s<Element>(srcElem1, 0, fpscr) ? -1 : 0"
fcmeqZeroCode = fpCmpZeroOp % "EQ"
twoEqualRegInstX("fcmeq", "FcmeqZeroDX", "SimdFloatCmpOp", smallFloatTypes,
2, fcmeqZeroCode)
twoEqualRegInstX("fcmeq", "FcmeqZeroQX", "SimdFloatCmpOp", floatTypes, 4,
fcmeqZeroCode)
twoEqualRegInstX("fcmeq", "FcmeqZeroScX", "SimdFloatCmpOp", floatTypes, 4,
fcmeqZeroCode, scalar=True)
# FCMGE (register)
fcmgeCode = fpCmpOp % "GE"
threeEqualRegInstX("fcmge", "FcmgeDX", "SimdFloatCmpOp", smallFloatTypes,
2, fcmgeCode)
threeEqualRegInstX("fcmge", "FcmgeQX", "SimdFloatCmpOp", floatTypes, 4,
fcmgeCode)
threeEqualRegInstX("fcmge", "FcmgeScX", "SimdFloatCmpOp", floatTypes, 4,
fcmgeCode, scalar=True)
# FCMGE (zero)
fcmgeZeroCode = fpCmpZeroOp % "GE"
twoEqualRegInstX("fcmge", "FcmgeZeroDX", "SimdFloatCmpOp", smallFloatTypes,
2, fcmgeZeroCode)
twoEqualRegInstX("fcmge", "FcmgeZeroQX", "SimdFloatCmpOp", floatTypes, 4,
fcmgeZeroCode)
twoEqualRegInstX("fcmge", "FcmgeZeroScX", "SimdFloatCmpOp", floatTypes, 4,
fcmgeZeroCode, scalar=True)
# FCMGT (register)
fcmgtCode = fpCmpOp % "GT"
threeEqualRegInstX("fcmgt", "FcmgtDX", "SimdFloatCmpOp", smallFloatTypes,
2, fcmgtCode)
threeEqualRegInstX("fcmgt", "FcmgtQX", "SimdFloatCmpOp", floatTypes, 4,
fcmgtCode)
threeEqualRegInstX("fcmgt", "FcmgtScX", "SimdFloatCmpOp", floatTypes, 4,
fcmgtCode, scalar=True)
# FCMGT (zero)
fcmgtZeroCode = fpCmpZeroOp % "GT"
twoEqualRegInstX("fcmgt", "FcmgtZeroDX", "SimdFloatCmpOp", smallFloatTypes,
2, fcmgtZeroCode)
twoEqualRegInstX("fcmgt", "FcmgtZeroQX", "SimdFloatCmpOp", floatTypes, 4,
fcmgtZeroCode)
twoEqualRegInstX("fcmgt", "FcmgtZeroScX", "SimdFloatCmpOp", floatTypes, 4,
fcmgtZeroCode, scalar=True)
# FCMLE (zero)
fpCmpRevZeroOp = fpOp % ("fplibCompare%s<Element>(0, srcElem1, fpscr) ?"
" -1 : 0")
fcmleZeroCode = fpCmpRevZeroOp % "GE"
twoEqualRegInstX("fcmle", "FcmleZeroDX", "SimdFloatCmpOp", smallFloatTypes,
2, fcmleZeroCode)
twoEqualRegInstX("fcmle", "FcmleZeroQX", "SimdFloatCmpOp", floatTypes, 4,
fcmleZeroCode)
twoEqualRegInstX("fcmle", "FcmleZeroScX", "SimdFloatCmpOp", floatTypes, 4,
fcmleZeroCode, scalar=True)
# FCMLT (zero)
fcmltZeroCode = fpCmpRevZeroOp % "GT"
twoEqualRegInstX("fcmlt", "FcmltZeroDX", "SimdFloatCmpOp", smallFloatTypes,
2, fcmltZeroCode)
twoEqualRegInstX("fcmlt", "FcmltZeroQX", "SimdFloatCmpOp", floatTypes, 4,
fcmltZeroCode)
twoEqualRegInstX("fcmlt", "FcmltZeroScX", "SimdFloatCmpOp", floatTypes, 4,
fcmltZeroCode, scalar=True)
# FCVTAS
fcvtCode = fpOp % ("fplibFPToFixed<Element, Element>("
"srcElem1, %s, %s, %s, fpscr)")
fcvtasCode = fcvtCode % ("0", "false", "FPRounding_TIEAWAY")
twoEqualRegInstX("fcvtas", "FcvtasDX", "SimdCvtOp", smallFloatTypes, 2,
fcvtasCode)
twoEqualRegInstX("fcvtas", "FcvtasQX", "SimdCvtOp", floatTypes, 4,
fcvtasCode)
twoEqualRegInstX("fcvtas", "FcvtasScX", "SimdCvtOp", floatTypes, 4,
fcvtasCode, scalar=True)
# FCVTAU
fcvtauCode = fcvtCode % ("0", "true", "FPRounding_TIEAWAY")
twoEqualRegInstX("fcvtau", "FcvtauDX", "SimdCvtOp", smallFloatTypes, 2,
fcvtauCode)
twoEqualRegInstX("fcvtau", "FcvtauQX", "SimdCvtOp", floatTypes, 4,
fcvtauCode)
twoEqualRegInstX("fcvtau", "FcvtauScX", "SimdCvtOp", floatTypes, 4,
fcvtauCode, scalar=True)
# FCVTL, FCVTL2
fcvtlCode = fpOp % ("fplibConvert<Element, BigElement>("
"srcElem1, FPCRRounding(fpscr), fpscr)")
twoRegLongInstX("fcvtl", "FcvtlX", "SimdCvtOp", ("uint16_t", "uint32_t"),
fcvtlCode)
twoRegLongInstX("fcvtl", "Fcvtl2X", "SimdCvtOp", ("uint16_t", "uint32_t"),
fcvtlCode, hi=True)
# FCVTMS
fcvtmsCode = fcvtCode % ("0", "false", "FPRounding_NEGINF")
twoEqualRegInstX("fcvtms", "FcvtmsDX", "SimdCvtOp", smallFloatTypes, 2,
fcvtmsCode)
twoEqualRegInstX("fcvtms", "FcvtmsQX", "SimdCvtOp", floatTypes, 4,
fcvtmsCode)
twoEqualRegInstX("fcvtms", "FcvtmsScX", "SimdCvtOp", floatTypes, 4,
fcvtmsCode, scalar=True)
# FCVTMU
fcvtmuCode = fcvtCode % ("0", "true", "FPRounding_NEGINF")
twoEqualRegInstX("fcvtmu", "FcvtmuDX", "SimdCvtOp", smallFloatTypes, 2,
fcvtmuCode)
twoEqualRegInstX("fcvtmu", "FcvtmuQX", "SimdCvtOp", floatTypes, 4,
fcvtmuCode)
twoEqualRegInstX("fcvtmu", "FcvtmuScX", "SimdCvtOp", floatTypes, 4,
fcvtmuCode, scalar=True)
# FCVTN, FCVTN2
fcvtnCode = fpOp % ("fplibConvert<BigElement, Element>("
"srcElem1, FPCRRounding(fpscr), fpscr)")
twoRegNarrowInstX("fcvtn", "FcvtnX", "SimdCvtOp",
("uint16_t", "uint32_t"), fcvtnCode)
twoRegNarrowInstX("fcvtn", "Fcvtn2X", "SimdCvtOp",
("uint16_t", "uint32_t"), fcvtnCode, hi=True)
# FCVTNS
fcvtnsCode = fcvtCode % ("0", "false", "FPRounding_TIEEVEN")
twoEqualRegInstX("fcvtns", "FcvtnsDX", "SimdCvtOp", smallFloatTypes, 2,
fcvtnsCode)
twoEqualRegInstX("fcvtns", "FcvtnsQX", "SimdCvtOp", floatTypes, 4,
fcvtnsCode)
twoEqualRegInstX("fcvtns", "FcvtnsScX", "SimdCvtOp", floatTypes, 4,
fcvtnsCode, scalar=True)
# FCVTNU
fcvtnuCode = fcvtCode % ("0", "true", "FPRounding_TIEEVEN")
twoEqualRegInstX("fcvtnu", "FcvtnuDX", "SimdCvtOp", smallFloatTypes, 2,
fcvtnuCode)
twoEqualRegInstX("fcvtnu", "FcvtnuQX", "SimdCvtOp", floatTypes, 4,
fcvtnuCode)
twoEqualRegInstX("fcvtnu", "FcvtnuScX", "SimdCvtOp", floatTypes, 4,
fcvtnuCode, scalar=True)
# FCVTPS
fcvtpsCode = fcvtCode % ("0", "false", "FPRounding_POSINF")
twoEqualRegInstX("fcvtps", "FcvtpsDX", "SimdCvtOp", smallFloatTypes, 2,
fcvtpsCode)
twoEqualRegInstX("fcvtps", "FcvtpsQX", "SimdCvtOp", floatTypes, 4,
fcvtpsCode)
twoEqualRegInstX("fcvtps", "FcvtpsScX", "SimdCvtOp", floatTypes, 4,
fcvtpsCode, scalar=True)
# FCVTPU
fcvtpuCode = fcvtCode % ("0", "true", "FPRounding_POSINF")
twoEqualRegInstX("fcvtpu", "FcvtpuDX", "SimdCvtOp", smallFloatTypes, 2,
fcvtpuCode)
twoEqualRegInstX("fcvtpu", "FcvtpuQX", "SimdCvtOp", floatTypes, 4,
fcvtpuCode)
twoEqualRegInstX("fcvtpu", "FcvtpuScX", "SimdCvtOp", floatTypes, 4,
fcvtpuCode, scalar=True)
# FCVTXN, FCVTXN2
fcvtxnCode = fpOp % ("fplibConvert<BigElement, Element>("
"srcElem1, FPRounding_ODD, fpscr)")
twoRegNarrowInstX("fcvtxn", "FcvtxnX", "SimdCvtOp", smallFloatTypes,
fcvtxnCode)
twoRegNarrowInstX("fcvtxn", "Fcvtxn2X", "SimdCvtOp", smallFloatTypes,
fcvtxnCode, hi=True)
twoRegNarrowInstX("fcvtxn", "FcvtxnScX", "SimdCvtOp", smallFloatTypes,
fcvtxnCode, scalar=True)
# FCVTZS (fixed-point)
fcvtzsCode = fcvtCode % ("imm", "false", "FPRounding_ZERO")
twoEqualRegInstX("fcvtzs", "FcvtzsFixedDX", "SimdCvtOp", smallFloatTypes,
2, fcvtzsCode, hasImm=True)
twoEqualRegInstX("fcvtzs", "FcvtzsFixedQX", "SimdCvtOp", floatTypes, 4,
fcvtzsCode, hasImm=True)
twoEqualRegInstX("fcvtzs", "FcvtzsFixedScX", "SimdCvtOp", floatTypes, 4,
fcvtzsCode, hasImm=True, scalar=True)
# FCVTZS (integer)
fcvtzsIntCode = fcvtCode % ("0", "false", "FPRounding_ZERO")
twoEqualRegInstX("fcvtzs", "FcvtzsIntDX", "SimdCvtOp", smallFloatTypes,
2, fcvtzsIntCode)
twoEqualRegInstX("fcvtzs", "FcvtzsIntQX", "SimdCvtOp", floatTypes, 4,
fcvtzsIntCode)
twoEqualRegInstX("fcvtzs", "FcvtzsIntScX", "SimdCvtOp", floatTypes, 4,
fcvtzsIntCode, scalar=True)
# FCVTZU (fixed-point)
fcvtzuCode = fcvtCode % ("imm", "true", "FPRounding_ZERO")
twoEqualRegInstX("fcvtzu", "FcvtzuFixedDX", "SimdCvtOp", smallFloatTypes,
2, fcvtzuCode, hasImm=True)
twoEqualRegInstX("fcvtzu", "FcvtzuFixedQX", "SimdCvtOp", floatTypes, 4,
fcvtzuCode, hasImm=True)
twoEqualRegInstX("fcvtzu", "FcvtzuFixedScX", "SimdCvtOp", floatTypes, 4,
fcvtzuCode, hasImm=True, scalar=True)
# FCVTZU (integer)
fcvtzuIntCode = fcvtCode % ("0", "true", "FPRounding_ZERO")
twoEqualRegInstX("fcvtzu", "FcvtzuIntDX", "SimdCvtOp", smallFloatTypes, 2,
fcvtzuIntCode)
twoEqualRegInstX("fcvtzu", "FcvtzuIntQX", "SimdCvtOp", floatTypes, 4,
fcvtzuIntCode)
twoEqualRegInstX("fcvtzu", "FcvtzuIntScX", "SimdCvtOp", floatTypes, 4,
fcvtzuIntCode, scalar=True)
# FDIV
fdivCode = fpBinOp % "Div"
threeEqualRegInstX("fdiv", "FdivDX", "SimdFloatDivOp", smallFloatTypes, 2,
fdivCode)
threeEqualRegInstX("fdiv", "FdivQX", "SimdFloatDivOp", floatTypes, 4,
fdivCode)
# FMAX
fmaxCode = fpBinOp % "Max"
threeEqualRegInstX("fmax", "FmaxDX", "SimdFloatCmpOp", smallFloatTypes, 2,
fmaxCode)
threeEqualRegInstX("fmax", "FmaxQX", "SimdFloatCmpOp", floatTypes, 4,
fmaxCode)
# FMAXNM
fmaxnmCode = fpBinOp % "MaxNum"
threeEqualRegInstX("fmaxnm", "FmaxnmDX", "SimdFloatCmpOp", smallFloatTypes,
2, fmaxnmCode)
threeEqualRegInstX("fmaxnm", "FmaxnmQX", "SimdFloatCmpOp", floatTypes, 4,
fmaxnmCode)
# FMAXNMP (scalar)
twoRegPairwiseScInstX("fmaxnmp", "FmaxnmpScDX", "SimdFloatCmpOp",
("uint32_t",), 2, fmaxnmCode)
twoRegPairwiseScInstX("fmaxnmp", "FmaxnmpScQX", "SimdFloatCmpOp",
("uint64_t",), 4, fmaxnmCode)
# FMAXNMP (vector)
threeEqualRegInstX("fmaxnmp", "FmaxnmpDX", "SimdFloatCmpOp",
smallFloatTypes, 2, fmaxnmCode, pairwise=True)
threeEqualRegInstX("fmaxnmp", "FmaxnmpQX", "SimdFloatCmpOp", floatTypes, 4,
fmaxnmCode, pairwise=True)
# FMAXNMV
# Note: SimdFloatCmpOp can be a bit optimistic here
fpAcrossOp = fpOp % "fplib%s<Element>(destElem, srcElem1, fpscr)"
fmaxnmAcrossCode = fpAcrossOp % "MaxNum"
twoRegAcrossInstX("fmaxnmv", "FmaxnmvQX", "SimdFloatCmpOp", ("uint32_t",),
4, fmaxnmAcrossCode)
# FMAXP (scalar)
twoRegPairwiseScInstX("fmaxp", "FmaxpScDX", "SimdFloatCmpOp",
("uint32_t",), 2, fmaxCode)
twoRegPairwiseScInstX("fmaxp", "FmaxpScQX", "SimdFloatCmpOp",
("uint64_t",), 4, fmaxCode)
# FMAXP (vector)
threeEqualRegInstX("fmaxp", "FmaxpDX", "SimdFloatCmpOp", smallFloatTypes,
2, fmaxCode, pairwise=True)
threeEqualRegInstX("fmaxp", "FmaxpQX", "SimdFloatCmpOp", floatTypes, 4,
fmaxCode, pairwise=True)
# FMAXV
# Note: SimdFloatCmpOp can be a bit optimistic here
fmaxAcrossCode = fpAcrossOp % "Max"
twoRegAcrossInstX("fmaxv", "FmaxvQX", "SimdFloatCmpOp", ("uint32_t",), 4,
fmaxAcrossCode)
# FMIN
fminCode = fpBinOp % "Min"
threeEqualRegInstX("fmin", "FminDX", "SimdFloatCmpOp", smallFloatTypes, 2,
fminCode)
threeEqualRegInstX("fmin", "FminQX", "SimdFloatCmpOp", floatTypes, 4,
fminCode)
# FMINNM
fminnmCode = fpBinOp % "MinNum"
threeEqualRegInstX("fminnm", "FminnmDX", "SimdFloatCmpOp", smallFloatTypes,
2, fminnmCode)
threeEqualRegInstX("fminnm", "FminnmQX", "SimdFloatCmpOp", floatTypes, 4,
fminnmCode)
# FMINNMP (scalar)
twoRegPairwiseScInstX("fminnmp", "FminnmpScDX", "SimdFloatCmpOp",
("uint32_t",), 2, fminnmCode)
twoRegPairwiseScInstX("fminnmp", "FminnmpScQX", "SimdFloatCmpOp",
("uint64_t",), 4, fminnmCode)
# FMINNMP (vector)
threeEqualRegInstX("fminnmp", "FminnmpDX", "SimdFloatCmpOp",
smallFloatTypes, 2, fminnmCode, pairwise=True)
threeEqualRegInstX("fminnmp", "FminnmpQX", "SimdFloatCmpOp", floatTypes, 4,
fminnmCode, pairwise=True)
# FMINNMV
# Note: SimdFloatCmpOp can be a bit optimistic here
fminnmAcrossCode = fpAcrossOp % "MinNum"
twoRegAcrossInstX("fminnmv", "FminnmvQX", "SimdFloatCmpOp", ("uint32_t",),
4, fminnmAcrossCode)
# FMINP (scalar)
twoRegPairwiseScInstX("fminp", "FminpScDX", "SimdFloatCmpOp",
("uint32_t",), 2, fminCode)
twoRegPairwiseScInstX("fminp", "FminpScQX", "SimdFloatCmpOp",
("uint64_t",), 4, fminCode)
# FMINP (vector)
threeEqualRegInstX("fminp", "FminpDX", "SimdFloatCmpOp", smallFloatTypes,
2, fminCode, pairwise=True)
threeEqualRegInstX("fminp", "FminpQX", "SimdFloatCmpOp", floatTypes, 4,
fminCode, pairwise=True)
# FMINV
# Note: SimdFloatCmpOp can be a bit optimistic here
fminAcrossCode = fpAcrossOp % "Min"
twoRegAcrossInstX("fminv", "FminvQX", "SimdFloatCmpOp", ("uint32_t",), 4,
fminAcrossCode)
# FMLA (by element)
fmlaCode = fpOp % ("fplibMulAdd<Element>("
"destElem, srcElem1, srcElem2, fpscr)")
threeEqualRegInstX("fmla", "FmlaElemDX", "SimdFloatMultAccOp",
smallFloatTypes, 2, fmlaCode, True, byElem=True)
threeEqualRegInstX("fmla", "FmlaElemQX", "SimdFloatMultAccOp", floatTypes,
4, fmlaCode, True, byElem=True)
threeEqualRegInstX("fmla", "FmlaElemScX", "SimdFloatMultAccOp", floatTypes,
4, fmlaCode, True, byElem=True, scalar=True)
# FMLA (vector)
threeEqualRegInstX("fmla", "FmlaDX", "SimdFloatMultAccOp", smallFloatTypes,
2, fmlaCode, True)
threeEqualRegInstX("fmla", "FmlaQX", "SimdFloatMultAccOp", floatTypes, 4,
fmlaCode, True)
# FMLS (by element)
fmlsCode = fpOp % ("fplibMulAdd<Element>(destElem,"
" fplibNeg<Element>(srcElem1), srcElem2, fpscr)")
threeEqualRegInstX("fmls", "FmlsElemDX", "SimdFloatMultAccOp",
smallFloatTypes, 2, fmlsCode, True, byElem=True)
threeEqualRegInstX("fmls", "FmlsElemQX", "SimdFloatMultAccOp", floatTypes,
4, fmlsCode, True, byElem=True)
threeEqualRegInstX("fmls", "FmlsElemScX", "SimdFloatMultAccOp", floatTypes,
4, fmlsCode, True, byElem=True, scalar=True)
# FMLS (vector)
threeEqualRegInstX("fmls", "FmlsDX", "SimdFloatMultAccOp", smallFloatTypes,
2, fmlsCode, True)
threeEqualRegInstX("fmls", "FmlsQX", "SimdFloatMultAccOp", floatTypes, 4,
fmlsCode, True)
# FMOV
fmovCode = 'destElem = imm;'
oneRegImmInstX("fmov", "FmovDX", "SimdMiscOp", smallFloatTypes, 2,
fmovCode)
oneRegImmInstX("fmov", "FmovQX", "SimdMiscOp", floatTypes, 4, fmovCode)
# FMUL (by element)
fmulCode = fpBinOp % "Mul"
threeEqualRegInstX("fmul", "FmulElemDX", "SimdFloatMultOp",
smallFloatTypes, 2, fmulCode, byElem=True)
threeEqualRegInstX("fmul", "FmulElemQX", "SimdFloatMultOp", floatTypes, 4,
fmulCode, byElem=True)
threeEqualRegInstX("fmul", "FmulElemScX", "SimdFloatMultOp", floatTypes, 4,
fmulCode, byElem=True, scalar=True)
# FMUL (vector)
threeEqualRegInstX("fmul", "FmulDX", "SimdFloatMultOp", smallFloatTypes, 2,
fmulCode)
threeEqualRegInstX("fmul", "FmulQX", "SimdFloatMultOp", floatTypes, 4,
fmulCode)
# FMULX
fmulxCode = fpBinOp % "MulX"
threeEqualRegInstX("fmulx", "FmulxDX", "SimdFloatMultOp", smallFloatTypes,
2, fmulxCode)
threeEqualRegInstX("fmulx", "FmulxQX", "SimdFloatMultOp", floatTypes, 4,
fmulxCode)
threeEqualRegInstX("fmulx", "FmulxScX", "SimdFloatMultOp", floatTypes, 4,
fmulxCode, scalar=True)
# FMULX (by element)
threeEqualRegInstX("fmulx", "FmulxElemDX", "SimdFloatMultOp",
smallFloatTypes, 2, fmulxCode, byElem=True)
threeEqualRegInstX("fmulx", "FmulxElemQX", "SimdFloatMultOp", floatTypes,
4, fmulxCode, byElem=True)
threeEqualRegInstX("fmulx", "FmulxElemScX", "SimdFloatMultOp", floatTypes,
4, fmulxCode, byElem=True, scalar=True)
# FNEG
fnegCode = fpOp % "fplibNeg<Element>(srcElem1)"
twoEqualRegInstX("Neg", "FnegDX", "SimdFloatAluOp", smallFloatTypes, 2,
fnegCode)
twoEqualRegInstX("Neg", "FnegQX", "SimdFloatAluOp", floatTypes, 4,
fnegCode)
# FRECPE
frecpeCode = fpOp % "fplibRecipEstimate<Element>(srcElem1, fpscr)"
twoEqualRegInstX("frecpe", "FrecpeDX", "SimdFloatMultAccOp",
smallFloatTypes, 2, frecpeCode)
twoEqualRegInstX("frecpe", "FrecpeQX", "SimdFloatMultAccOp", floatTypes, 4,
frecpeCode)
twoEqualRegInstX("frecpe", "FrecpeScX", "SimdFloatMultAccOp", floatTypes,
4, frecpeCode, scalar=True)
# FRECPS
frecpsCode = fpBinOp % "RecipStepFused"
threeEqualRegInstX("frecps", "FrecpsDX", "SimdFloatMultAccOp",
smallFloatTypes, 2, frecpsCode)
threeEqualRegInstX("frecps", "FrecpsQX", "SimdFloatMultAccOp", floatTypes,
4, frecpsCode)
threeEqualRegInstX("frecps", "FrecpsScX", "SimdFloatMultAccOp", floatTypes,
4, frecpsCode, scalar=True)
# FRECPX
frecpxCode = fpOp % "fplibRecpX<Element>(srcElem1, fpscr)"
twoEqualRegInstX("frecpx", "FrecpxX", "SimdFloatMultAccOp", floatTypes, 4,
frecpxCode, scalar=True)
# FRINTA
frintCode = fpOp % "fplibRoundInt<Element>(srcElem1, %s, %s, fpscr)"
frintaCode = frintCode % ("FPRounding_TIEAWAY", "false")
twoEqualRegInstX("frinta", "FrintaDX", "SimdCvtOp", smallFloatTypes, 2,
frintaCode)
twoEqualRegInstX("frinta", "FrintaQX", "SimdCvtOp", floatTypes, 4,
frintaCode)
# FRINTI
frintiCode = frintCode % ("FPCRRounding(fpscr)", "false")
twoEqualRegInstX("frinti", "FrintiDX", "SimdCvtOp", smallFloatTypes, 2,
frintiCode)
twoEqualRegInstX("frinti", "FrintiQX", "SimdCvtOp", floatTypes, 4,
frintiCode)
# FRINTM
frintmCode = frintCode % ("FPRounding_NEGINF", "false")
twoEqualRegInstX("frintm", "FrintmDX", "SimdCvtOp", smallFloatTypes, 2,
frintmCode)
twoEqualRegInstX("frintm", "FrintmQX", "SimdCvtOp", floatTypes, 4,
frintmCode)
# FRINTN
frintnCode = frintCode % ("FPRounding_TIEEVEN", "false")
twoEqualRegInstX("frintn", "FrintnDX", "SimdCvtOp", smallFloatTypes, 2,
frintnCode)
twoEqualRegInstX("frintn", "FrintnQX", "SimdCvtOp", floatTypes, 4,
frintnCode)
# FRINTP
frintpCode = frintCode % ("FPRounding_POSINF", "false")
twoEqualRegInstX("frintp", "FrintpDX", "SimdCvtOp", smallFloatTypes, 2,
frintpCode)
twoEqualRegInstX("frintp", "FrintpQX", "SimdCvtOp", floatTypes, 4,
frintpCode)
# FRINTX
frintxCode = frintCode % ("FPCRRounding(fpscr)", "true")
twoEqualRegInstX("frintx", "FrintxDX", "SimdCvtOp", smallFloatTypes, 2,
frintxCode)
twoEqualRegInstX("frintx", "FrintxQX", "SimdCvtOp", floatTypes, 4,
frintxCode)
# FRINTZ
frintzCode = frintCode % ("FPRounding_ZERO", "false")
twoEqualRegInstX("frintz", "FrintzDX", "SimdCvtOp", smallFloatTypes, 2,
frintzCode)
twoEqualRegInstX("frintz", "FrintzQX", "SimdCvtOp", floatTypes, 4,
frintzCode)
# FRSQRTE
frsqrteCode = fpOp % "fplibRSqrtEstimate<Element>(srcElem1, fpscr)"
twoEqualRegInstX("frsqrte", "FrsqrteDX", "SimdFloatSqrtOp",
smallFloatTypes, 2, frsqrteCode)
twoEqualRegInstX("frsqrte", "FrsqrteQX", "SimdFloatSqrtOp", floatTypes, 4,
frsqrteCode)
twoEqualRegInstX("frsqrte", "FrsqrteScX", "SimdFloatSqrtOp", floatTypes, 4,
frsqrteCode, scalar=True)
# FRSQRTS
frsqrtsCode = fpBinOp % "RSqrtStepFused"
threeEqualRegInstX("frsqrts", "FrsqrtsDX", "SimdFloatMiscOp",
smallFloatTypes, 2, frsqrtsCode)
threeEqualRegInstX("frsqrts", "FrsqrtsQX", "SimdFloatMiscOp", floatTypes,
4, frsqrtsCode)
threeEqualRegInstX("frsqrts", "FrsqrtsScX", "SimdFloatMiscOp", floatTypes,
4, frsqrtsCode, scalar=True)
# FSQRT
fsqrtCode = fpOp % "fplibSqrt<Element>(srcElem1, fpscr)"
twoEqualRegInstX("fsqrt", "FsqrtDX", "SimdFloatSqrtOp", smallFloatTypes, 2,
fsqrtCode)
twoEqualRegInstX("fsqrt", "FsqrtQX", "SimdFloatSqrtOp", floatTypes, 4,
fsqrtCode)
# FSUB
fsubCode = fpBinOp % "Sub"
threeEqualRegInstX("fsub", "FsubDX", "SimdFloatAddOp", smallFloatTypes, 2,
fsubCode)
threeEqualRegInstX("fsub", "FsubQX", "SimdFloatAddOp", floatTypes, 4,
fsubCode)
# INS (element)
insFromVecElemInstX("ins", "InsElemX", "SimdMiscOp", unsignedTypes, 4)
# INS (general register)
insFromGprInstX("ins", "InsGprWX", "SimdMiscOp", smallUnsignedTypes, 4,
'W')
insFromGprInstX("ins", "InsGprXX", "SimdMiscOp", unsignedTypes, 4, 'X')
# MLA (by element)
mlaCode = "destElem += srcElem1 * srcElem2;"
threeEqualRegInstX("mla", "MlaElemDX", "SimdMultAccOp",
("uint16_t", "uint32_t"), 2, mlaCode, True, byElem=True)
threeEqualRegInstX("mla", "MlaElemQX", "SimdMultAccOp",
("uint16_t", "uint32_t"), 4, mlaCode, True, byElem=True)
# MLA (vector)
threeEqualRegInstX("mla", "MlaDX", "SimdMultAccOp", smallUnsignedTypes, 2,
mlaCode, True)
threeEqualRegInstX("mla", "MlaQX", "SimdMultAccOp", smallUnsignedTypes, 4,
mlaCode, True)
# MLS (by element)
mlsCode = "destElem -= srcElem1 * srcElem2;"
threeEqualRegInstX("mls", "MlsElemDX", "SimdMultAccOp",
("uint16_t", "uint32_t"), 2, mlsCode, True, byElem=True)
threeEqualRegInstX("mls", "MlsElemQX", "SimdMultAccOp",
("uint16_t", "uint32_t"), 4, mlsCode, True, byElem=True)
# MLS (vector)
threeEqualRegInstX("mls", "MlsDX", "SimdMultAccOp", smallUnsignedTypes, 2,
mlsCode, True)
threeEqualRegInstX("mls", "MlsQX", "SimdMultAccOp", smallUnsignedTypes, 4,
mlsCode, True)
# MOV (element) -> alias to INS (element)
# MOV (from general) -> alias to INS (general register)
# MOV (scalar) -> alias to DUP (element)
# MOV (to general) -> alias to UMOV
# MOV (vector) -> alias to ORR (register)
# MOVI
movImmCode = "destElem = imm;"
oneRegImmInstX("movi", "MoviDX", "SimdMiscOp", ("uint64_t",), 2,
movImmCode)
oneRegImmInstX("movi", "MoviQX", "SimdMiscOp", ("uint64_t",), 4,
movImmCode)
# MUL (by element)
mulCode = "destElem = srcElem1 * srcElem2;"
threeEqualRegInstX("mul", "MulElemDX", "SimdMultOp",
("uint16_t", "uint32_t"), 2, mulCode, byElem=True)
threeEqualRegInstX("mul", "MulElemQX", "SimdMultOp",
("uint16_t", "uint32_t"), 4, mulCode, byElem=True)
# MUL (vector)
threeEqualRegInstX("mul", "MulDX", "SimdMultOp", smallUnsignedTypes, 2,
mulCode)
threeEqualRegInstX("mul", "MulQX", "SimdMultOp", smallUnsignedTypes, 4,
mulCode)
# MVN
mvnCode = "destElem = ~srcElem1;"
twoEqualRegInstX("mvn", "MvnDX", "SimdAluOp", ("uint64_t",), 2, mvnCode)
twoEqualRegInstX("mvn", "MvnQX", "SimdAluOp", ("uint64_t",), 4, mvnCode)
# MVNI
mvniCode = "destElem = ~imm;"
oneRegImmInstX("mvni", "MvniDX", "SimdAluOp", ("uint64_t",), 2, mvniCode)
oneRegImmInstX("mvni", "MvniQX", "SimdAluOp", ("uint64_t",), 4, mvniCode)
# NEG
negCode = "destElem = -srcElem1;"
twoEqualRegInstX("neg", "NegDX", "SimdAluOp", signedTypes, 2, negCode)
twoEqualRegInstX("neg", "NegQX", "SimdAluOp", signedTypes, 4, negCode)
# NOT -> alias to MVN
# ORN
ornCode = "destElem = srcElem1 | ~srcElem2;"
threeEqualRegInstX("orn", "OrnDX", "SimdAluOp", ("uint64_t",), 2, ornCode)
threeEqualRegInstX("orn", "OrnQX", "SimdAluOp", ("uint64_t",), 4, ornCode)
# ORR (immediate)
orrImmCode = "destElem |= imm;"
oneRegImmInstX("orr", "OrrImmDX", "SimdAluOp", ("uint64_t",), 2,
orrImmCode, True)
oneRegImmInstX("orr", "OrrImmQX", "SimdAluOp", ("uint64_t",), 4,
orrImmCode, True)
# ORR (register)
orrCode = "destElem = srcElem1 | srcElem2;"
threeEqualRegInstX("orr", "OrrDX", "SimdAluOp", ("uint64_t",), 2, orrCode)
threeEqualRegInstX("orr", "OrrQX", "SimdAluOp", ("uint64_t",), 4, orrCode)
# PMUL
pmulCode = '''
destElem = 0;
for (unsigned j = 0; j < sizeof(Element) * 8; j++) {
if (bits(srcElem2, j))
destElem ^= srcElem1 << j;
}
'''
threeEqualRegInstX("pmul", "PmulDX", "SimdMultOp", ("uint8_t",), 2,
pmulCode)
threeEqualRegInstX("pmul", "PmulQX", "SimdMultOp", ("uint8_t",), 4,
pmulCode)
# PMULL, PMULL2
# Note: 64-bit PMULL is not available (Crypto. Extension)
pmullCode = '''
destElem = 0;
for (unsigned j = 0; j < sizeof(Element) * 8; j++) {
if (bits(srcElem2, j))
destElem ^= (BigElement)srcElem1 << j;
}
'''
threeRegLongInstX("pmull", "PmullX", "SimdMultOp", ("uint8_t",), pmullCode)
threeRegLongInstX("pmull", "Pmull2X", "SimdMultOp", ("uint8_t",),
pmullCode, hi=True)
# RADDHN, RADDHN2
raddhnCode = '''
destElem = ((BigElement)srcElem1 + (BigElement)srcElem2 +
((BigElement)1 << (sizeof(Element) * 8 - 1))) >>
(sizeof(Element) * 8);
'''
threeRegNarrowInstX("raddhn", "RaddhnX", "SimdAddOp", smallUnsignedTypes,
raddhnCode)
threeRegNarrowInstX("raddhn2", "Raddhn2X", "SimdAddOp", smallUnsignedTypes,
raddhnCode, hi=True)
# RBIT
rbitCode = '''
destElem = 0;
Element temp = srcElem1;
for (int i = 0; i < 8 * sizeof(Element); i++) {
destElem = destElem | ((temp & 0x1) <<
(8 * sizeof(Element) - 1 - i));
temp >>= 1;
}
'''
twoEqualRegInstX("rbit", "RbitDX", "SimdAluOp", ("uint8_t",), 2, rbitCode)
twoEqualRegInstX("rbit", "RbitQX", "SimdAluOp", ("uint8_t",), 4, rbitCode)
# REV16
rev16Code = '''
destElem = srcElem1;
unsigned groupSize = ((1 << 1) / sizeof(Element));
unsigned reverseMask = (groupSize - 1);
j = i ^ reverseMask;
'''
twoEqualRegInstX("rev16", "Rev16DX", "SimdAluOp", ("uint8_t",), 2,
rev16Code)
twoEqualRegInstX("rev16", "Rev16QX", "SimdAluOp", ("uint8_t",), 4,
rev16Code)
# REV32
rev32Code = '''
destElem = srcElem1;
unsigned groupSize = ((1 << 2) / sizeof(Element));
unsigned reverseMask = (groupSize - 1);
j = i ^ reverseMask;
'''
twoEqualRegInstX("rev32", "Rev32DX", "SimdAluOp", ("uint8_t", "uint16_t"),
2, rev32Code)
twoEqualRegInstX("rev32", "Rev32QX", "SimdAluOp", ("uint8_t", "uint16_t"),
4, rev32Code)
# REV64
rev64Code = '''
destElem = srcElem1;
unsigned groupSize = ((1 << 3) / sizeof(Element));
unsigned reverseMask = (groupSize - 1);
j = i ^ reverseMask;
'''
twoEqualRegInstX("rev64", "Rev64DX", "SimdAluOp", smallUnsignedTypes, 2,
rev64Code)
twoEqualRegInstX("rev64", "Rev64QX", "SimdAluOp", smallUnsignedTypes, 4,
rev64Code)
# RSHRN, RSHRN2
rshrnCode = '''
if (imm > sizeof(srcElem1) * 8) {
destElem = 0;
} else if (imm) {
Element rBit = bits(srcElem1, imm - 1);
destElem = ((srcElem1 >> (imm - 1)) >> 1) + rBit;
} else {
destElem = srcElem1;
}
'''
twoRegNarrowInstX("rshrn", "RshrnX", "SimdShiftOp", smallUnsignedTypes,
rshrnCode, hasImm=True)
twoRegNarrowInstX("rshrn2", "Rshrn2X", "SimdShiftOp", smallUnsignedTypes,
rshrnCode, hasImm=True, hi=True)
# RSUBHN, RSUBHN2
rsubhnCode = '''
destElem = ((BigElement)srcElem1 - (BigElement)srcElem2 +
((BigElement)1 << (sizeof(Element) * 8 - 1))) >>
(sizeof(Element) * 8);
'''
threeRegNarrowInstX("rsubhn", "RsubhnX", "SimdAddOp", smallTypes,
rsubhnCode)
threeRegNarrowInstX("rsubhn2", "Rsubhn2X", "SimdAddOp", smallTypes,
rsubhnCode, hi=True)
# SABA
abaCode = '''
destElem += (srcElem1 > srcElem2) ? (srcElem1 - srcElem2) :
(srcElem2 - srcElem1);
'''
threeEqualRegInstX("saba", "SabaDX", "SimdAddAccOp", smallSignedTypes, 2,
abaCode, True)
threeEqualRegInstX("saba", "SabaQX", "SimdAddAccOp", smallSignedTypes, 4,
abaCode, True)
# SABAL, SABAL2
abalCode = '''
destElem += (srcElem1 > srcElem2) ?
((BigElement)srcElem1 - (BigElement)srcElem2) :
((BigElement)srcElem2 - (BigElement)srcElem1);
'''
threeRegLongInstX("sabal", "SabalX", "SimdAddAccOp", smallSignedTypes,
abalCode, True)
threeRegLongInstX("sabal2", "Sabal2X", "SimdAddAccOp", smallSignedTypes,
abalCode, True, hi=True)
# SABD
abdCode = '''
destElem = (srcElem1 > srcElem2) ? (srcElem1 - srcElem2) :
(srcElem2 - srcElem1);
'''
threeEqualRegInstX("sabd", "SabdDX", "SimdAddOp", smallSignedTypes, 2,
abdCode)
threeEqualRegInstX("sabd", "SabdQX", "SimdAddOp", smallSignedTypes, 4,
abdCode)
# SABDL, SABDL2
abdlCode = '''
destElem = (srcElem1 > srcElem2) ?
((BigElement)srcElem1 - (BigElement)srcElem2) :
((BigElement)srcElem2 - (BigElement)srcElem1);
'''
threeRegLongInstX("sabdl", "SabdlX", "SimdAddAccOp", smallSignedTypes,
abdlCode, True)
threeRegLongInstX("sabdl2", "Sabdl2X", "SimdAddAccOp", smallSignedTypes,
abdlCode, True, hi=True)
# SADALP
adalpCode = "destElem += (BigElement)srcElem1 + (BigElement)srcElem2;"
twoRegCondenseInstX("sadalp", "SadalpDX", "SimdAddOp", smallSignedTypes, 2,
adalpCode, True)
twoRegCondenseInstX("sadalp", "SadalpQX", "SimdAddOp", smallSignedTypes, 4,
adalpCode, True)
# SADDL, SADDL2
addlwCode = "destElem = (BigElement)srcElem1 + (BigElement)srcElem2;"
threeRegLongInstX("saddl", "SaddlX", "SimdAddAccOp", smallSignedTypes,
addlwCode)
threeRegLongInstX("saddl2", "Saddl2X", "SimdAddAccOp", smallSignedTypes,
addlwCode, hi=True)
# SADDLP
twoRegCondenseInstX("saddlp", "SaddlpDX", "SimdAddOp", smallSignedTypes, 2,
addlwCode)
twoRegCondenseInstX("saddlp", "SaddlpQX", "SimdAddOp", smallSignedTypes, 4,
addlwCode)
# SADDLV
# Note: SimdAddOp can be a bit optimistic here
addAcrossLongCode = "destElem += (BigElement)srcElem1;"
twoRegAcrossInstX("saddlv", "SaddlvDX", "SimdAddOp", ("int8_t", "int16_t"),
2, addAcrossLongCode, long=True)
twoRegAcrossInstX("saddlv", "SaddlvQX", "SimdAddOp", ("int8_t", "int16_t"),
4, addAcrossLongCode, long=True)
twoRegAcrossInstX("saddlv", "SaddlvBQX", "SimdAddOp", ("int32_t",), 4,
addAcrossLongCode, doubleDest=True, long=True)
# SADDW, SADDW2
threeRegWideInstX("saddw", "SaddwX", "SimdAddAccOp", smallSignedTypes,
addlwCode)
threeRegWideInstX("saddw2", "Saddw2X", "SimdAddAccOp", smallSignedTypes,
addlwCode, hi=True)
# SCVTF (fixed-point)
scvtfFixedCode = fpOp % ("fplibFixedToFP<Element>((int%d_t) srcElem1, imm,"
" false, FPCRRounding(fpscr), fpscr)")
twoEqualRegInstX("scvtf", "ScvtfFixedDX", "SimdCvtOp", smallFloatTypes, 2,
scvtfFixedCode % 32, hasImm=True)
twoEqualRegInstX("scvtf", "ScvtfFixedSQX", "SimdCvtOp", smallFloatTypes, 4,
scvtfFixedCode % 32, hasImm=True)
twoEqualRegInstX("scvtf", "ScvtfFixedDQX", "SimdCvtOp", ("uint64_t",), 4,
scvtfFixedCode % 64, hasImm=True)
twoEqualRegInstX("scvtf", "ScvtfFixedScSX", "SimdCvtOp", smallFloatTypes,
4, scvtfFixedCode % 32, hasImm=True, scalar=True)
twoEqualRegInstX("scvtf", "ScvtfFixedScDX", "SimdCvtOp", ("uint64_t",), 4,
scvtfFixedCode % 64, hasImm=True, scalar=True)
# SCVTF (integer)
scvtfIntCode = fpOp % ("fplibFixedToFP<Element>((int%d_t) srcElem1, 0,"
" false, FPCRRounding(fpscr), fpscr)")
twoEqualRegInstX("scvtf", "ScvtfIntDX", "SimdCvtOp", smallFloatTypes, 2,
scvtfIntCode % 32)
twoEqualRegInstX("scvtf", "ScvtfIntSQX", "SimdCvtOp", smallFloatTypes, 4,
scvtfIntCode % 32)
twoEqualRegInstX("scvtf", "ScvtfIntDQX", "SimdCvtOp", ("uint64_t",), 4,
scvtfIntCode % 64)
twoEqualRegInstX("scvtf", "ScvtfIntScSX", "SimdCvtOp", smallFloatTypes, 4,
scvtfIntCode % 32, scalar=True)
twoEqualRegInstX("scvtf", "ScvtfIntScDX", "SimdCvtOp", ("uint64_t",), 4,
scvtfIntCode % 64, scalar=True)
# SHADD
haddCode = '''
Element carryBit =
(((unsigned)srcElem1 & 0x1) +
((unsigned)srcElem2 & 0x1)) >> 1;
// Use division instead of a shift to ensure the sign extension works
// right. The compiler will figure out if it can be a shift. Mask the
// inputs so they get truncated correctly.
destElem = (((srcElem1 & ~(Element)1) / 2) +
((srcElem2 & ~(Element)1) / 2)) + carryBit;
'''
threeEqualRegInstX("shadd", "ShaddDX", "SimdAddOp", smallSignedTypes, 2,
haddCode)
threeEqualRegInstX("shadd", "ShaddQX", "SimdAddOp", smallSignedTypes, 4,
haddCode)
# SHL
shlCode = '''
if (imm >= sizeof(Element) * 8)
destElem = (srcElem1 << (sizeof(Element) * 8 - 1)) << 1;
else
destElem = srcElem1 << imm;
'''
twoEqualRegInstX("shl", "ShlDX", "SimdShiftOp", unsignedTypes, 2, shlCode,
hasImm=True)
twoEqualRegInstX("shl", "ShlQX", "SimdShiftOp", unsignedTypes, 4, shlCode,
hasImm=True)
# SHLL, SHLL2
shllCode = "destElem = ((BigElement)srcElem1) << (sizeof(Element) * 8);"
twoRegLongInstX("shll", "ShllX", "SimdShiftOp", smallTypes, shllCode)
twoRegLongInstX("shll", "Shll2X", "SimdShiftOp", smallTypes, shllCode,
hi=True)
# SHRN, SHRN2
shrnCode = '''
if (imm >= sizeof(srcElem1) * 8) {
destElem = 0;
} else {
destElem = srcElem1 >> imm;
}
'''
twoRegNarrowInstX("shrn", "ShrnX", "SimdShiftOp", smallUnsignedTypes,
shrnCode, hasImm=True)
twoRegNarrowInstX("shrn2", "Shrn2X", "SimdShiftOp", smallUnsignedTypes,
shrnCode, hasImm=True, hi=True)
# SHSUB
hsubCode = '''
Element borrowBit =
(((srcElem1 & 0x1) - (srcElem2 & 0x1)) >> 1) & 0x1;
// Use division instead of a shift to ensure the sign extension works
// right. The compiler will figure out if it can be a shift. Mask the
// inputs so they get truncated correctly.
destElem = (((srcElem1 & ~(Element)1) / 2) -
((srcElem2 & ~(Element)1) / 2)) - borrowBit;
'''
threeEqualRegInstX("shsub", "ShsubDX", "SimdAddOp", smallSignedTypes, 2,
hsubCode)
threeEqualRegInstX("shsub", "ShsubQX", "SimdAddOp", smallSignedTypes, 4,
hsubCode)
# SLI
sliCode = '''
if (imm >= sizeof(Element) * 8)
destElem = destElem;
else
destElem = (srcElem1 << imm) | (destElem & mask(imm));
'''
twoEqualRegInstX("sli", "SliDX", "SimdShiftOp", unsignedTypes, 2, sliCode,
True, hasImm=True)
twoEqualRegInstX("sli", "SliQX", "SimdShiftOp", unsignedTypes, 4, sliCode,
True, hasImm=True)
# SMAX
maxCode = "destElem = (srcElem1 > srcElem2) ? srcElem1 : srcElem2;"
threeEqualRegInstX("smax", "SmaxDX", "SimdCmpOp", smallSignedTypes, 2,
maxCode)
threeEqualRegInstX("smax", "SmaxQX", "SimdCmpOp", smallSignedTypes, 4,
maxCode)
# SMAXP
threeEqualRegInstX("smaxp", "SmaxpDX", "SimdCmpOp", smallSignedTypes, 2,
maxCode, pairwise=True)
threeEqualRegInstX("smaxp", "SmaxpQX", "SimdCmpOp", smallSignedTypes, 4,
maxCode, pairwise=True)
# SMAXV
maxAcrossCode = '''
if (i == 0 || srcElem1 > destElem)
destElem = srcElem1;
'''
twoRegAcrossInstX("smaxv", "SmaxvDX", "SimdCmpOp", ("int8_t", "int16_t"),
2, maxAcrossCode)
twoRegAcrossInstX("smaxv", "SmaxvQX", "SimdCmpOp", smallSignedTypes, 4,
maxAcrossCode)
# SMIN
minCode = "destElem = (srcElem1 < srcElem2) ? srcElem1 : srcElem2;"
threeEqualRegInstX("smin", "SminDX", "SimdCmpOp", smallSignedTypes, 2,
minCode)
threeEqualRegInstX("smin", "SminQX", "SimdCmpOp", smallSignedTypes, 4,
minCode)
# SMINP
threeEqualRegInstX("sminp", "SminpDX", "SimdCmpOp", smallSignedTypes, 2,
minCode, pairwise=True)
threeEqualRegInstX("sminp", "SminpQX", "SimdCmpOp", smallSignedTypes, 4,
minCode, pairwise=True)
# SMINV
minAcrossCode = '''
if (i == 0 || srcElem1 < destElem)
destElem = srcElem1;
'''
twoRegAcrossInstX("sminv", "SminvDX", "SimdCmpOp", ("int8_t", "int16_t"),
2, minAcrossCode)
twoRegAcrossInstX("sminv", "SminvQX", "SimdCmpOp", smallSignedTypes, 4,
minAcrossCode)
split('exec')
# SMLAL, SMLAL2 (by element)
mlalCode = "destElem += (BigElement)srcElem1 * (BigElement)srcElem2;"
threeRegLongInstX("smlal", "SmlalElemX", "SimdMultAccOp",
("int16_t", "int32_t"), mlalCode, True, byElem=True)
threeRegLongInstX("smlal", "SmlalElem2X", "SimdMultAccOp",
("int16_t", "int32_t"), mlalCode, True, byElem=True,
hi=True)
# SMLAL, SMLAL2 (vector)
threeRegLongInstX("smlal", "SmlalX", "SimdMultAccOp", smallSignedTypes,
mlalCode, True)
threeRegLongInstX("smlal", "Smlal2X", "SimdMultAccOp", smallSignedTypes,
mlalCode, True, hi=True)
# SMLSL, SMLSL2 (by element)
mlslCode = "destElem -= (BigElement)srcElem1 * (BigElement)srcElem2;"
threeRegLongInstX("smlsl", "SmlslElemX", "SimdMultAccOp", smallSignedTypes,
mlslCode, True, byElem=True)
threeRegLongInstX("smlsl", "SmlslElem2X", "SimdMultAccOp",
smallSignedTypes, mlslCode, True, byElem=True, hi=True)
# SMLSL, SMLSL2 (vector)
threeRegLongInstX("smlsl", "SmlslX", "SimdMultAccOp", smallSignedTypes,
mlslCode, True)
threeRegLongInstX("smlsl", "Smlsl2X", "SimdMultAccOp", smallSignedTypes,
mlslCode, True, hi=True)
# SMOV
insToGprInstX("smov", "SmovWX", "SimdMiscOp", ("int8_t", "int16_t"), 4,
'W', True)
insToGprInstX("smov", "SmovXX", "SimdMiscOp", smallSignedTypes, 4, 'X',
True)
# SMULL, SMULL2 (by element)
mullCode = "destElem = (BigElement)srcElem1 * (BigElement)srcElem2;"
threeRegLongInstX("smull", "SmullElemX", "SimdMultOp", smallSignedTypes,
mullCode, byElem=True)
threeRegLongInstX("smull", "SmullElem2X", "SimdMultOp", smallSignedTypes,
mullCode, byElem=True, hi=True)
# SMULL, SMULL2 (vector)
threeRegLongInstX("smull", "SmullX", "SimdMultOp", smallSignedTypes,
mullCode)
threeRegLongInstX("smull", "Smull2X", "SimdMultOp", smallSignedTypes,
mullCode, hi=True)
# SQABS
sqabsCode = '''
FPSCR fpscr = (FPSCR) FpscrQc;
if (srcElem1 == (Element)(std::numeric_limits<Element>::min())) {
fpscr.qc = 1;
destElem = ~srcElem1;
} else if (srcElem1 < 0) {
destElem = -srcElem1;
} else {
destElem = srcElem1;
}
FpscrQc = fpscr;
'''
twoEqualRegInstX("sqabs", "SqabsDX", "SimdAluOp", smallSignedTypes, 2,
sqabsCode)
twoEqualRegInstX("sqabs", "SqabsQX", "SimdAluOp", signedTypes, 4,
sqabsCode)
twoEqualRegInstX("sqabs", "SqabsScX", "SimdAluOp", signedTypes, 4,
sqabsCode, scalar=True)
# SQADD
sqaddCode = '''
destElem = srcElem1 + srcElem2;
FPSCR fpscr = (FPSCR) FpscrQc;
bool negDest = (destElem < 0);
bool negSrc1 = (srcElem1 < 0);
bool negSrc2 = (srcElem2 < 0);
if ((negDest != negSrc1) && (negSrc1 == negSrc2)) {
destElem = std::numeric_limits<Element>::min();
if (negDest)
destElem -= 1;
fpscr.qc = 1;
}
FpscrQc = fpscr;
'''
threeEqualRegInstX("sqadd", "SqaddDX", "SimdAddOp", smallSignedTypes, 2,
sqaddCode)
threeEqualRegInstX("sqadd", "SqaddQX", "SimdAddOp", signedTypes, 4,
sqaddCode)
threeEqualRegInstX("sqadd", "SqaddScX", "SimdAddOp", signedTypes, 4,
sqaddCode, scalar=True)
# SQDMLAL, SQDMLAL2 (by element)
qdmlalCode = '''
FPSCR fpscr = (FPSCR) FpscrQc;
BigElement midElem = (2 * (int64_t)srcElem1 * (int64_t)srcElem2);
Element maxNeg = std::numeric_limits<Element>::min();
Element halfNeg = maxNeg / 2;
if ((srcElem1 == maxNeg && srcElem2 == maxNeg) ||
(srcElem1 == halfNeg && srcElem2 == maxNeg) ||
(srcElem1 == maxNeg && srcElem2 == halfNeg)) {
midElem = ~((BigElement)maxNeg << (sizeof(Element) * 8));
fpscr.qc = 1;
}
bool negPreDest = ltz(destElem);
destElem += midElem;
bool negDest = ltz(destElem);
bool negMid = ltz(midElem);
if (negPreDest == negMid && negMid != negDest) {
destElem = mask(sizeof(BigElement) * 8 - 1);
if (negPreDest)
destElem = ~destElem;
fpscr.qc = 1;
}
FpscrQc = fpscr;
'''
threeRegLongInstX("sqdmlal", "SqdmlalElemX", "SimdMultAccOp",
("int16_t", "int32_t"), qdmlalCode, True, byElem=True)
threeRegLongInstX("sqdmlal", "SqdmlalElem2X", "SimdMultAccOp",
("int16_t", "int32_t"), qdmlalCode, True, byElem=True,
hi=True)
threeRegLongInstX("sqdmlal", "SqdmlalElemScX", "SimdMultAccOp",
("int16_t", "int32_t"), qdmlalCode, True, byElem=True,
scalar=True)
# SQDMLAL, SQDMLAL2 (vector)
threeRegLongInstX("sqdmlal", "SqdmlalX", "SimdMultAccOp",
("int16_t", "int32_t"), qdmlalCode, True)
threeRegLongInstX("sqdmlal", "Sqdmlal2X", "SimdMultAccOp",
("int16_t", "int32_t"), qdmlalCode, True, hi=True)
threeRegLongInstX("sqdmlal", "SqdmlalScX", "SimdMultAccOp",
("int16_t", "int32_t"), qdmlalCode, True, scalar=True)
# SQDMLSL, SQDMLSL2 (by element)
qdmlslCode = '''
FPSCR fpscr = (FPSCR) FpscrQc;
BigElement midElem = (2 * (int64_t)srcElem1 * (int64_t)srcElem2);
Element maxNeg = std::numeric_limits<Element>::min();
Element halfNeg = maxNeg / 2;
if ((srcElem1 == maxNeg && srcElem2 == maxNeg) ||
(srcElem1 == halfNeg && srcElem2 == maxNeg) ||
(srcElem1 == maxNeg && srcElem2 == halfNeg)) {
midElem = ~((BigElement)maxNeg << (sizeof(Element) * 8));
fpscr.qc = 1;
}
bool negPreDest = ltz(destElem);
destElem -= midElem;
bool negDest = ltz(destElem);
bool posMid = ltz((BigElement)-midElem);
if (negPreDest == posMid && posMid != negDest) {
destElem = mask(sizeof(BigElement) * 8 - 1);
if (negPreDest)
destElem = ~destElem;
fpscr.qc = 1;
}
FpscrQc = fpscr;
'''
threeRegLongInstX("sqdmlsl", "SqdmlslElemX", "SimdMultAccOp",
("int16_t", "int32_t"), qdmlslCode, True, byElem=True)
threeRegLongInstX("sqdmlsl", "SqdmlslElem2X", "SimdMultAccOp",
("int16_t", "int32_t"), qdmlslCode, True, byElem=True,
hi=True)
threeRegLongInstX("sqdmlsl", "SqdmlslElemScX", "SimdMultAccOp",
("int16_t", "int32_t"), qdmlslCode, True, byElem=True,
scalar=True)
# SQDMLSL, SQDMLSL2 (vector)
threeRegLongInstX("sqdmlsl", "SqdmlslX", "SimdMultAccOp",
("int16_t", "int32_t"), qdmlslCode, True)
threeRegLongInstX("sqdmlsl", "Sqdmlsl2X", "SimdMultAccOp",
("int16_t", "int32_t"), qdmlslCode, True, hi=True)
threeRegLongInstX("sqdmlsl", "SqdmlslScX", "SimdMultAccOp",
("int16_t", "int32_t"), qdmlslCode, True, scalar=True)
# SQDMULH (by element)
sqdmulhCode = '''
FPSCR fpscr = (FPSCR) FpscrQc;
destElem = (2 * (int64_t)srcElem1 * (int64_t)srcElem2) >>
(sizeof(Element) * 8);
if (srcElem1 == srcElem2 &&
srcElem1 == (Element)((Element)1 <<
(sizeof(Element) * 8 - 1))) {
destElem = ~srcElem1;
fpscr.qc = 1;
}
FpscrQc = fpscr;
'''
threeEqualRegInstX("sqdmulh", "SqdmulhElemDX", "SimdMultOp",
("int16_t", "int32_t"), 2, sqdmulhCode, byElem=True)
threeEqualRegInstX("sqdmulh", "SqdmulhElemQX", "SimdMultOp",
("int16_t", "int32_t"), 4, sqdmulhCode, byElem=True)
threeEqualRegInstX("sqdmulh", "SqdmulhElemScX", "SimdMultOp",
("int16_t", "int32_t"), 4, sqdmulhCode, byElem=True,
scalar=True)
# SQDMULH (vector)
threeEqualRegInstX("sqdmulh", "SqdmulhDX", "SimdMultOp",
("int16_t", "int32_t"), 2, sqdmulhCode)
threeEqualRegInstX("sqdmulh", "SqdmulhQX", "SimdMultOp",
("int16_t", "int32_t"), 4, sqdmulhCode)
threeEqualRegInstX("sqdmulh", "SqdmulhScX", "SimdMultOp",
("int16_t", "int32_t"), 4, sqdmulhCode, scalar=True)
# SQDMULL, SQDMULL2 (by element)
qdmullCode = '''
FPSCR fpscr = (FPSCR) FpscrQc;
destElem = (2 * (int64_t)srcElem1 * (int64_t)srcElem2);
if (srcElem1 == srcElem2 &&
srcElem1 == (Element)((Element)1 <<
(Element)(sizeof(Element) * 8 - 1))) {
destElem = ~((BigElement)srcElem1 << (sizeof(Element) * 8));
fpscr.qc = 1;
}
FpscrQc = fpscr;
'''
threeRegLongInstX("sqdmull", "SqdmullElemX", "SimdMultOp",
("int16_t", "int32_t"), qdmullCode, True, byElem=True)
threeRegLongInstX("sqdmull", "SqdmullElem2X", "SimdMultOp",
("int16_t", "int32_t"), qdmullCode, True, byElem=True,
hi=True)
threeRegLongInstX("sqdmull", "SqdmullElemScX", "SimdMultOp",
("int16_t", "int32_t"), qdmullCode, True, byElem=True,
scalar=True)
# SQDMULL, SQDMULL2 (vector)
threeRegLongInstX("sqdmull", "SqdmullX", "SimdMultOp",
("int16_t", "int32_t"), qdmullCode, True)
threeRegLongInstX("sqdmull", "Sqdmull2X", "SimdMultOp",
("int16_t", "int32_t"), qdmullCode, True, hi=True)
threeRegLongInstX("sqdmull", "SqdmullScX", "SimdMultOp",
("int16_t", "int32_t"), qdmullCode, True, scalar=True)
# SQNEG
sqnegCode = '''
FPSCR fpscr = (FPSCR) FpscrQc;
if (srcElem1 == (Element)(std::numeric_limits<Element>::min())) {
fpscr.qc = 1;
destElem = ~srcElem1;
} else {
destElem = -srcElem1;
}
FpscrQc = fpscr;
'''
twoEqualRegInstX("sqneg", "SqnegDX", "SimdAluOp", smallSignedTypes, 2,
sqnegCode)
twoEqualRegInstX("sqneg", "SqnegQX", "SimdAluOp", signedTypes, 4,
sqnegCode)
twoEqualRegInstX("sqneg", "SqnegScX", "SimdAluOp", signedTypes, 4,
sqnegCode, scalar=True)
# SQRDMULH (by element)
sqrdmulhCode = '''
FPSCR fpscr = (FPSCR) FpscrQc;
destElem = (2 * (int64_t)srcElem1 * (int64_t)srcElem2 +
((int64_t)1 << (sizeof(Element) * 8 - 1))) >>
(sizeof(Element) * 8);
Element maxNeg = std::numeric_limits<Element>::min();
Element halfNeg = maxNeg / 2;
if ((srcElem1 == maxNeg && srcElem2 == maxNeg) ||
(srcElem1 == halfNeg && srcElem2 == maxNeg) ||
(srcElem1 == maxNeg && srcElem2 == halfNeg)) {
if (destElem < 0) {
destElem = mask(sizeof(Element) * 8 - 1);
} else {
destElem = std::numeric_limits<Element>::min();
}
fpscr.qc = 1;
}
FpscrQc = fpscr;
'''
threeEqualRegInstX("sqrdmulh", "SqrdmulhElemDX", "SimdMultOp",
("int16_t", "int32_t"), 2, sqrdmulhCode, byElem=True)
threeEqualRegInstX("sqrdmulh", "SqrdmulhElemQX", "SimdMultOp",
("int16_t", "int32_t"), 4, sqrdmulhCode, byElem=True)
threeEqualRegInstX("sqrdmulh", "SqrdmulhElemScX", "SimdMultOp",
("int16_t", "int32_t"), 4, sqrdmulhCode, byElem=True,
scalar=True)
# SQRDMULH (vector)
threeEqualRegInstX("sqrdmulh", "SqrdmulhDX", "SimdMultOp",
("int16_t", "int32_t"), 2, sqrdmulhCode)
threeEqualRegInstX("sqrdmulh", "SqrdmulhQX", "SimdMultOp",
("int16_t", "int32_t"), 4, sqrdmulhCode)
threeEqualRegInstX("sqrdmulh", "SqrdmulhScX", "SimdMultOp",
("int16_t", "int32_t"), 4, sqrdmulhCode, scalar=True)
# SQRSHL
sqrshlCode = '''
int16_t shiftAmt = (int8_t)srcElem2;
FPSCR fpscr = (FPSCR) FpscrQc;
if (shiftAmt < 0) {
shiftAmt = -shiftAmt;
Element rBit = 0;
if (shiftAmt <= sizeof(Element) * 8)
rBit = bits(srcElem1, shiftAmt - 1);
if (shiftAmt > sizeof(Element) * 8 && srcElem1 < 0)
rBit = 1;
if (shiftAmt >= sizeof(Element) * 8) {
shiftAmt = sizeof(Element) * 8 - 1;
destElem = 0;
} else {
destElem = (srcElem1 >> shiftAmt);
}
// Make sure the right shift sign extended when it should.
if (srcElem1 < 0 && destElem >= 0) {
destElem |= -((Element)1 << (sizeof(Element) * 8 -
1 - shiftAmt));
}
destElem += rBit;
} else if (shiftAmt > 0) {
bool sat = false;
if (shiftAmt >= sizeof(Element) * 8) {
if (srcElem1 != 0)
sat = true;
else
destElem = 0;
} else {
if (bits((uint64_t) srcElem1, sizeof(Element) * 8 - 1,
sizeof(Element) * 8 - 1 - shiftAmt) !=
((srcElem1 < 0) ? mask(shiftAmt + 1) : 0)) {
sat = true;
} else {
destElem = srcElem1 << shiftAmt;
}
}
if (sat) {
fpscr.qc = 1;
destElem = mask(sizeof(Element) * 8 - 1);
if (srcElem1 < 0)
destElem = ~destElem;
}
} else {
destElem = srcElem1;
}
FpscrQc = fpscr;
'''
threeEqualRegInstX("sqrshl", "SqrshlDX", "SimdCmpOp", smallSignedTypes, 2,
sqrshlCode)
threeEqualRegInstX("sqrshl", "SqrshlQX", "SimdCmpOp", signedTypes, 4,
sqrshlCode)
threeEqualRegInstX("sqrshl", "SqrshlScX", "SimdCmpOp", signedTypes, 4,
sqrshlCode, scalar=True)
# SQRSHRN, SQRSHRN2
sqrshrnCode = '''
FPSCR fpscr = (FPSCR) FpscrQc;
if (imm > sizeof(srcElem1) * 8) {
if (srcElem1 != 0 && srcElem1 != -1)
fpscr.qc = 1;
destElem = 0;
} else if (imm) {
BigElement mid = (srcElem1 >> (imm - 1));
uint64_t rBit = mid & 0x1;
mid >>= 1;
mid |= -(mid & ((BigElement)1 <<
(sizeof(BigElement) * 8 - 1 - imm)));
mid += rBit;
if (mid != (Element)mid) {
destElem = mask(sizeof(Element) * 8 - 1);
if (srcElem1 < 0)
destElem = ~destElem;
fpscr.qc = 1;
} else {
destElem = mid;
}
} else {
if (srcElem1 != (Element)srcElem1) {
destElem = mask(sizeof(Element) * 8 - 1);
if (srcElem1 < 0)
destElem = ~destElem;
fpscr.qc = 1;
} else {
destElem = srcElem1;
}
}
FpscrQc = fpscr;
'''
twoRegNarrowInstX("sqrshrn", "SqrshrnX", "SimdShiftOp", smallSignedTypes,
sqrshrnCode, hasImm=True)
twoRegNarrowInstX("sqrshrn2", "Sqrshrn2X", "SimdShiftOp", smallSignedTypes,
sqrshrnCode, hasImm=True, hi=True)
twoRegNarrowInstX("sqrshrn", "SqrshrnScX", "SimdShiftOp", smallSignedTypes,
sqrshrnCode, hasImm=True, scalar=True)
# SQRSHRUN, SQRSHRUN2
sqrshrunCode = '''
FPSCR fpscr = (FPSCR) FpscrQc;
if (imm > sizeof(srcElem1) * 8) {
if (srcElem1 != 0)
fpscr.qc = 1;
destElem = 0;
} else if (imm) {
BigElement mid = (srcElem1 >> (imm - 1));
uint64_t rBit = mid & 0x1;
mid >>= 1;
mid |= -(mid & ((BigElement)1 <<
(sizeof(BigElement) * 8 - 1 - imm)));
mid += rBit;
if (bits(mid, sizeof(BigElement) * 8 - 1,
sizeof(Element) * 8) != 0) {
if (srcElem1 < 0) {
destElem = 0;
} else {
destElem = mask(sizeof(Element) * 8);
}
fpscr.qc = 1;
} else {
destElem = mid;
}
} else {
if (srcElem1 < 0) {
fpscr.qc = 1;
destElem = 0;
} else {
destElem = srcElem1;
}
}
FpscrQc = fpscr;
'''
twoRegNarrowInstX("sqrshrun", "SqrshrunX", "SimdShiftOp", smallSignedTypes,
sqrshrunCode, hasImm=True)
twoRegNarrowInstX("sqrshrun", "Sqrshrun2X", "SimdShiftOp",
smallSignedTypes, sqrshrunCode, hasImm=True, hi=True)
twoRegNarrowInstX("sqrshrun", "SqrshrunScX", "SimdShiftOp",
smallSignedTypes, sqrshrunCode, hasImm=True, scalar=True)
# SQSHL (immediate)
sqshlImmCode = '''
FPSCR fpscr = (FPSCR) FpscrQc;
if (imm >= sizeof(Element) * 8) {
if (srcElem1 != 0) {
destElem = std::numeric_limits<Element>::min();
if (srcElem1 > 0)
destElem = ~destElem;
fpscr.qc = 1;
} else {
destElem = 0;
}
} else if (imm) {
destElem = (srcElem1 << imm);
uint64_t topBits = bits((uint64_t)srcElem1,
sizeof(Element) * 8 - 1,
sizeof(Element) * 8 - 1 - imm);
if (topBits != 0 && topBits != mask(imm + 1)) {
destElem = std::numeric_limits<Element>::min();
if (srcElem1 > 0)
destElem = ~destElem;
fpscr.qc = 1;
}
} else {
destElem = srcElem1;
}
FpscrQc = fpscr;
'''
twoEqualRegInstX("sqshl", "SqshlImmDX", "SimdAluOp", smallSignedTypes, 2,
sqshlImmCode, hasImm=True)
twoEqualRegInstX("sqshl", "SqshlImmQX", "SimdAluOp", signedTypes, 4,
sqshlImmCode, hasImm=True)
twoEqualRegInstX("sqshl", "SqshlImmScX", "SimdAluOp", signedTypes, 4,
sqshlImmCode, hasImm=True, scalar=True)
# SQSHL (register)
sqshlCode = '''
int16_t shiftAmt = (int8_t)srcElem2;
FPSCR fpscr = (FPSCR) FpscrQc;
if (shiftAmt < 0) {
shiftAmt = -shiftAmt;
if (shiftAmt >= sizeof(Element) * 8) {
shiftAmt = sizeof(Element) * 8 - 1;