| // Copyright (c) 2009 The Regents of The University of Michigan |
| // Copyright (c) 2015 Advanced Micro Devices, Inc. |
| // |
| // All rights reserved. |
| // |
| // Redistribution and use in source and binary forms, with or without |
| // modification, are permitted provided that the following conditions are |
| // met: redistributions of source code must retain the above copyright |
| // notice, this list of conditions and the following disclaimer; |
| // redistributions in binary form must reproduce the above copyright |
| // notice, this list of conditions and the following disclaimer in the |
| // documentation and/or other materials provided with the distribution; |
| // neither the name of the copyright holders nor the names of its |
| // contributors may be used to endorse or promote products derived from |
| // this software without specific prior written permission. |
| // |
| // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
| // "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
| // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR |
| // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT |
| // OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, |
| // SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT |
| // LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, |
| // DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY |
| // THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
| // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE |
| // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
| // |
| // Authors: Gabe Black |
| |
| def template MediaOpExecute {{ |
| Fault %(class_name)s::execute(ExecContext *xc, |
| Trace::InstRecord *traceData) const |
| { |
| Fault fault = NoFault; |
| |
| %(op_decl)s; |
| %(op_rd)s; |
| |
| %(code)s; |
| |
| //Write the resulting state to the execution context |
| if(fault == NoFault) |
| { |
| %(op_wb)s; |
| } |
| return fault; |
| } |
| }}; |
| |
| def template MediaOpRegDeclare {{ |
| class %(class_name)s : public %(base_class)s |
| { |
| public: |
| %(class_name)s(ExtMachInst _machInst, |
| const char * instMnem, uint64_t setFlags, |
| InstRegIndex _src1, InstRegIndex _src2, InstRegIndex _dest, |
| uint8_t _srcSize, uint8_t _destSize, uint16_t _ext); |
| |
| Fault execute(ExecContext *, Trace::InstRecord *) const; |
| }; |
| }}; |
| |
| def template MediaOpImmDeclare {{ |
| |
| class %(class_name)s : public %(base_class)s |
| { |
| public: |
| %(class_name)s(ExtMachInst _machInst, |
| const char * instMnem, uint64_t setFlags, |
| InstRegIndex _src1, uint16_t _imm8, InstRegIndex _dest, |
| uint8_t _srcSize, uint8_t _destSize, uint16_t _ext); |
| |
| Fault execute(ExecContext *, Trace::InstRecord *) const; |
| }; |
| }}; |
| |
| def template MediaOpRegConstructor {{ |
| %(class_name)s::%(class_name)s( |
| ExtMachInst machInst, const char * instMnem, uint64_t setFlags, |
| InstRegIndex _src1, InstRegIndex _src2, InstRegIndex _dest, |
| uint8_t _srcSize, uint8_t _destSize, uint16_t _ext) : |
| %(base_class)s(machInst, "%(mnemonic)s", instMnem, setFlags, |
| _src1, _src2, _dest, _srcSize, _destSize, _ext, |
| %(op_class)s) |
| { |
| %(constructor)s; |
| } |
| }}; |
| |
| def template MediaOpImmConstructor {{ |
| %(class_name)s::%(class_name)s( |
| ExtMachInst machInst, const char * instMnem, uint64_t setFlags, |
| InstRegIndex _src1, uint16_t _imm8, InstRegIndex _dest, |
| uint8_t _srcSize, uint8_t _destSize, uint16_t _ext) : |
| %(base_class)s(machInst, "%(mnemonic)s", instMnem, setFlags, |
| _src1, _imm8, _dest, _srcSize, _destSize, _ext, |
| %(op_class)s) |
| { |
| %(constructor)s; |
| } |
| }}; |
| |
| let {{ |
| # Make these empty strings so that concatenating onto |
| # them will always work. |
| header_output = "" |
| decoder_output = "" |
| exec_output = "" |
| |
| immTemplates = ( |
| MediaOpImmDeclare, |
| MediaOpImmConstructor, |
| MediaOpExecute) |
| |
| regTemplates = ( |
| MediaOpRegDeclare, |
| MediaOpRegConstructor, |
| MediaOpExecute) |
| |
| class MediaOpMeta(type): |
| def buildCppClasses(self, name, Name, suffix, code): |
| |
| # Globals to stick the output in |
| global header_output |
| global decoder_output |
| global exec_output |
| |
| # If op2 is used anywhere, make register and immediate versions |
| # of this code. |
| matcher = re.compile(r"(?<!\w)(?P<prefix>s?)op2(?P<typeQual>_[^\W_]+)?") |
| match = matcher.search(code) |
| if match: |
| typeQual = "" |
| if match.group("typeQual"): |
| typeQual = match.group("typeQual") |
| src2_name = "%sFpSrcReg2%s" % (match.group("prefix"), typeQual) |
| self.buildCppClasses(name, Name, suffix, |
| matcher.sub(src2_name, code)) |
| self.buildCppClasses(name + "i", Name, suffix + "Imm", |
| matcher.sub("imm8", code)) |
| return |
| |
| base = "X86ISA::MediaOp" |
| |
| # If imm8 shows up in the code, use the immediate templates, if |
| # not, hopefully the register ones will be correct. |
| matcher = re.compile("(?<!\w)imm8(?!\w)") |
| if matcher.search(code): |
| base += "Imm" |
| templates = immTemplates |
| else: |
| base += "Reg" |
| templates = regTemplates |
| |
| # Get everything ready for the substitution |
| opt_args = [] |
| if self.op_class: |
| opt_args.append(self.op_class) |
| iop = InstObjParams(name, Name + suffix, base, {"code" : code}, |
| opt_args) |
| |
| # Generate the actual code (finally!) |
| header_output += templates[0].subst(iop) |
| decoder_output += templates[1].subst(iop) |
| exec_output += templates[2].subst(iop) |
| |
| |
| def __new__(mcls, Name, bases, dict): |
| abstract = False |
| name = Name.lower() |
| if "abstract" in dict: |
| abstract = dict['abstract'] |
| del dict['abstract'] |
| if not "op_class" in dict: |
| dict["op_class"] = None |
| |
| cls = super(MediaOpMeta, mcls).__new__(mcls, Name, bases, dict) |
| if not abstract: |
| cls.className = Name |
| cls.base_mnemonic = name |
| code = cls.code |
| |
| # Set up the C++ classes |
| mcls.buildCppClasses(cls, name, Name, "", code) |
| |
| # Hook into the microassembler dict |
| global microopClasses |
| microopClasses[name] = cls |
| |
| # If op2 is used anywhere, make register and immediate versions |
| # of this code. |
| matcher = re.compile(r"op2(?P<typeQual>_[^\W_]+)?") |
| if matcher.search(code): |
| microopClasses[name + 'i'] = cls |
| return cls |
| |
| |
| class MediaOp(X86Microop): |
| __metaclass__ = MediaOpMeta |
| # This class itself doesn't act as a microop |
| abstract = True |
| |
| def __init__(self, dest, src1, op2, |
| size = None, destSize = None, srcSize = None, ext = None): |
| self.dest = dest |
| self.src1 = src1 |
| self.op2 = op2 |
| if size is not None: |
| self.srcSize = size |
| self.destSize = size |
| if srcSize is not None: |
| self.srcSize = srcSize |
| if destSize is not None: |
| self.destSize = destSize |
| if self.srcSize is None: |
| raise Exception, "Source size not set." |
| if self.destSize is None: |
| raise Exception, "Dest size not set." |
| if ext is None: |
| self.ext = 0 |
| else: |
| self.ext = ext |
| |
| def getAllocator(self, microFlags): |
| className = self.className |
| if self.mnemonic == self.base_mnemonic + 'i': |
| className += "Imm" |
| allocator = '''new %(class_name)s(machInst, macrocodeBlock, |
| %(flags)s, %(src1)s, %(op2)s, %(dest)s, |
| %(srcSize)s, %(destSize)s, %(ext)s)''' % { |
| "class_name" : className, |
| "flags" : self.microFlagsText(microFlags), |
| "src1" : self.src1, "op2" : self.op2, |
| "dest" : self.dest, |
| "srcSize" : self.srcSize, |
| "destSize" : self.destSize, |
| "ext" : self.ext} |
| return allocator |
| |
| class Mov2int(MediaOp): |
| def __init__(self, dest, src1, src2 = 0, \ |
| size = None, destSize = None, srcSize = None, ext = None): |
| super(Mov2int, self).__init__(dest, src1,\ |
| src2, size, destSize, srcSize, ext) |
| op_class = 'SimdMiscOp' |
| code = ''' |
| int items = sizeof(FloatRegBits) / srcSize; |
| int offset = imm8; |
| if (bits(src1, 0) && (ext & 0x1)) |
| offset -= items; |
| if (offset >= 0 && offset < items) { |
| uint64_t fpSrcReg1 = |
| bits(FpSrcReg1_uqw, |
| (offset + 1) * srcSize * 8 - 1, |
| (offset + 0) * srcSize * 8); |
| DestReg = merge(0, fpSrcReg1, destSize); |
| } else { |
| DestReg = DestReg; |
| } |
| ''' |
| |
| class Mov2fp(MediaOp): |
| def __init__(self, dest, src1, src2 = 0, \ |
| size = None, destSize = None, srcSize = None, ext = None): |
| super(Mov2fp, self).__init__(dest, src1,\ |
| src2, size, destSize, srcSize, ext) |
| op_class = 'SimdMiscOp' |
| code = ''' |
| int items = sizeof(FloatRegBits) / destSize; |
| int offset = imm8; |
| if (bits(dest, 0) && (ext & 0x1)) |
| offset -= items; |
| if (offset >= 0 && offset < items) { |
| uint64_t srcReg1 = pick(SrcReg1, 0, srcSize); |
| FpDestReg_uqw = |
| insertBits(FpDestReg_uqw, |
| (offset + 1) * destSize * 8 - 1, |
| (offset + 0) * destSize * 8, srcReg1); |
| } else { |
| FpDestReg_uqw = FpDestReg_uqw; |
| } |
| ''' |
| |
| class Movsign(MediaOp): |
| def __init__(self, dest, src, \ |
| size = None, destSize = None, srcSize = None, ext = None): |
| super(Movsign, self).__init__(dest, src,\ |
| "InstRegIndex(0)", size, destSize, srcSize, ext) |
| op_class = 'SimdMiscOp' |
| code = ''' |
| int items = sizeof(FloatRegBits) / srcSize; |
| uint64_t result = 0; |
| int offset = (ext & 0x1) ? items : 0; |
| for (int i = 0; i < items; i++) { |
| uint64_t picked = |
| bits(FpSrcReg1_uqw, (i + 1) * 8 * srcSize - 1); |
| result = insertBits(result, i + offset, i + offset, picked); |
| } |
| DestReg = DestReg | result; |
| ''' |
| |
| class Maskmov(MediaOp): |
| op_class = 'SimdMiscOp' |
| code = ''' |
| assert(srcSize == destSize); |
| int size = srcSize; |
| int sizeBits = size * 8; |
| int items = numItems(size); |
| uint64_t result = FpDestReg_uqw; |
| |
| for (int i = 0; i < items; i++) { |
| int hiIndex = (i + 1) * sizeBits - 1; |
| int loIndex = (i + 0) * sizeBits; |
| uint64_t arg1Bits = bits(FpSrcReg1_uqw, hiIndex, loIndex); |
| if (bits(FpSrcReg2_uqw, hiIndex)) |
| result = insertBits(result, hiIndex, loIndex, arg1Bits); |
| } |
| FpDestReg_uqw = result; |
| ''' |
| |
| class shuffle(MediaOp): |
| op_class = 'SimdMiscOp' |
| code = ''' |
| assert(srcSize == destSize); |
| int size = srcSize; |
| int sizeBits = size * 8; |
| int items = sizeof(FloatRegBits) / size; |
| int options; |
| int optionBits; |
| if (size == 8) { |
| options = 2; |
| optionBits = 1; |
| } else { |
| options = 4; |
| optionBits = 2; |
| } |
| |
| uint64_t result = 0; |
| uint8_t sel = ext; |
| |
| for (int i = 0; i < items; i++) { |
| uint64_t resBits; |
| uint8_t lsel = sel & mask(optionBits); |
| if (lsel * size >= sizeof(FloatRegBits)) { |
| lsel -= options / 2; |
| resBits = bits(FpSrcReg2_uqw, |
| (lsel + 1) * sizeBits - 1, |
| (lsel + 0) * sizeBits); |
| } else { |
| resBits = bits(FpSrcReg1_uqw, |
| (lsel + 1) * sizeBits - 1, |
| (lsel + 0) * sizeBits); |
| } |
| |
| sel >>= optionBits; |
| |
| int hiIndex = (i + 1) * sizeBits - 1; |
| int loIndex = (i + 0) * sizeBits; |
| result = insertBits(result, hiIndex, loIndex, resBits); |
| } |
| FpDestReg_uqw = result; |
| ''' |
| |
| class Unpack(MediaOp): |
| op_class = 'SimdMiscOp' |
| code = ''' |
| assert(srcSize == destSize); |
| int size = destSize; |
| int items = (sizeof(FloatRegBits) / size) / 2; |
| int offset = ext ? items : 0; |
| uint64_t result = 0; |
| for (int i = 0; i < items; i++) { |
| uint64_t pickedLow = |
| bits(FpSrcReg1_uqw, (i + offset + 1) * 8 * size - 1, |
| (i + offset) * 8 * size); |
| result = insertBits(result, |
| (2 * i + 1) * 8 * size - 1, |
| (2 * i + 0) * 8 * size, |
| pickedLow); |
| uint64_t pickedHigh = |
| bits(FpSrcReg2_uqw, (i + offset + 1) * 8 * size - 1, |
| (i + offset) * 8 * size); |
| result = insertBits(result, |
| (2 * i + 2) * 8 * size - 1, |
| (2 * i + 1) * 8 * size, |
| pickedHigh); |
| } |
| FpDestReg_uqw = result; |
| ''' |
| |
| class Pack(MediaOp): |
| op_class = 'SimdMiscOp' |
| code = ''' |
| assert(srcSize == destSize * 2); |
| int items = (sizeof(FloatRegBits) / destSize); |
| int destBits = destSize * 8; |
| int srcBits = srcSize * 8; |
| uint64_t result = 0; |
| int i; |
| for (i = 0; i < items / 2; i++) { |
| uint64_t picked = |
| bits(FpSrcReg1_uqw, (i + 1) * srcBits - 1, |
| (i + 0) * srcBits); |
| unsigned signBit = bits(picked, srcBits - 1); |
| uint64_t overflow = bits(picked, srcBits - 1, destBits - 1); |
| |
| // Handle saturation. |
| if (signBit) { |
| if (overflow != mask(destBits - srcBits + 1)) { |
| if (signedOp()) |
| picked = (ULL(1) << (destBits - 1)); |
| else |
| picked = 0; |
| } |
| } else { |
| if (overflow != 0) { |
| if (signedOp()) |
| picked = mask(destBits - 1); |
| else |
| picked = mask(destBits); |
| } |
| } |
| result = insertBits(result, |
| (i + 1) * destBits - 1, |
| (i + 0) * destBits, |
| picked); |
| } |
| for (;i < items; i++) { |
| uint64_t picked = |
| bits(FpSrcReg2_uqw, (i - items + 1) * srcBits - 1, |
| (i - items + 0) * srcBits); |
| unsigned signBit = bits(picked, srcBits - 1); |
| uint64_t overflow = bits(picked, srcBits - 1, destBits - 1); |
| |
| // Handle saturation. |
| if (signBit) { |
| if (overflow != mask(destBits - srcBits + 1)) { |
| if (signedOp()) |
| picked = (ULL(1) << (destBits - 1)); |
| else |
| picked = 0; |
| } |
| } else { |
| if (overflow != 0) { |
| if (signedOp()) |
| picked = mask(destBits - 1); |
| else |
| picked = mask(destBits); |
| } |
| } |
| result = insertBits(result, |
| (i + 1) * destBits - 1, |
| (i + 0) * destBits, |
| picked); |
| } |
| FpDestReg_uqw = result; |
| ''' |
| |
| class Mxor(MediaOp): |
| def __init__(self, dest, src1, src2): |
| super(Mxor, self).__init__(dest, src1, src2, 1) |
| op_class = 'SimdAluOp' |
| code = ''' |
| FpDestReg_uqw = FpSrcReg1_uqw ^ FpSrcReg2_uqw; |
| ''' |
| |
| class Mor(MediaOp): |
| def __init__(self, dest, src1, src2): |
| super(Mor, self).__init__(dest, src1, src2, 1) |
| op_class = 'SimdAluOp' |
| code = ''' |
| FpDestReg_uqw = FpSrcReg1_uqw | FpSrcReg2_uqw; |
| ''' |
| |
| class Mand(MediaOp): |
| def __init__(self, dest, src1, src2): |
| super(Mand, self).__init__(dest, src1, src2, 1) |
| op_class = 'SimdAluOp' |
| code = ''' |
| FpDestReg_uqw = FpSrcReg1_uqw & FpSrcReg2_uqw; |
| ''' |
| |
| class Mandn(MediaOp): |
| def __init__(self, dest, src1, src2): |
| super(Mandn, self).__init__(dest, src1, src2, 1) |
| op_class = 'SimdAluOp' |
| code = ''' |
| FpDestReg_uqw = ~FpSrcReg1_uqw & FpSrcReg2_uqw; |
| ''' |
| |
| class Mminf(MediaOp): |
| op_class = 'SimdFloatCmpOp' |
| code = ''' |
| union floatInt |
| { |
| float f; |
| uint32_t i; |
| }; |
| union doubleInt |
| { |
| double d; |
| uint64_t i; |
| }; |
| |
| assert(srcSize == destSize); |
| int size = srcSize; |
| int sizeBits = size * 8; |
| assert(srcSize == 4 || srcSize == 8); |
| int items = numItems(size); |
| uint64_t result = FpDestReg_uqw; |
| |
| for (int i = 0; i < items; i++) { |
| double arg1, arg2; |
| int hiIndex = (i + 1) * sizeBits - 1; |
| int loIndex = (i + 0) * sizeBits; |
| uint64_t arg1Bits = bits(FpSrcReg1_uqw, hiIndex, loIndex); |
| uint64_t arg2Bits = bits(FpSrcReg2_uqw, hiIndex, loIndex); |
| |
| if (size == 4) { |
| floatInt fi; |
| fi.i = arg1Bits; |
| arg1 = fi.f; |
| fi.i = arg2Bits; |
| arg2 = fi.f; |
| } else { |
| doubleInt di; |
| di.i = arg1Bits; |
| arg1 = di.d; |
| di.i = arg2Bits; |
| arg2 = di.d; |
| } |
| |
| if (arg1 < arg2) { |
| result = insertBits(result, hiIndex, loIndex, arg1Bits); |
| } else { |
| result = insertBits(result, hiIndex, loIndex, arg2Bits); |
| } |
| } |
| FpDestReg_uqw = result; |
| ''' |
| |
| class Mmaxf(MediaOp): |
| op_class = 'SimdFloatCmpOp' |
| code = ''' |
| union floatInt |
| { |
| float f; |
| uint32_t i; |
| }; |
| union doubleInt |
| { |
| double d; |
| uint64_t i; |
| }; |
| |
| assert(srcSize == destSize); |
| int size = srcSize; |
| int sizeBits = size * 8; |
| assert(srcSize == 4 || srcSize == 8); |
| int items = numItems(size); |
| uint64_t result = FpDestReg_uqw; |
| |
| for (int i = 0; i < items; i++) { |
| double arg1, arg2; |
| int hiIndex = (i + 1) * sizeBits - 1; |
| int loIndex = (i + 0) * sizeBits; |
| uint64_t arg1Bits = bits(FpSrcReg1_uqw, hiIndex, loIndex); |
| uint64_t arg2Bits = bits(FpSrcReg2_uqw, hiIndex, loIndex); |
| |
| if (size == 4) { |
| floatInt fi; |
| fi.i = arg1Bits; |
| arg1 = fi.f; |
| fi.i = arg2Bits; |
| arg2 = fi.f; |
| } else { |
| doubleInt di; |
| di.i = arg1Bits; |
| arg1 = di.d; |
| di.i = arg2Bits; |
| arg2 = di.d; |
| } |
| |
| if (arg1 > arg2) { |
| result = insertBits(result, hiIndex, loIndex, arg1Bits); |
| } else { |
| result = insertBits(result, hiIndex, loIndex, arg2Bits); |
| } |
| } |
| FpDestReg_uqw = result; |
| ''' |
| |
| class Mmini(MediaOp): |
| op_class = 'SimdCmpOp' |
| code = ''' |
| |
| assert(srcSize == destSize); |
| int size = srcSize; |
| int sizeBits = size * 8; |
| int items = numItems(size); |
| uint64_t result = FpDestReg_uqw; |
| |
| for (int i = 0; i < items; i++) { |
| int hiIndex = (i + 1) * sizeBits - 1; |
| int loIndex = (i + 0) * sizeBits; |
| uint64_t arg1Bits = bits(FpSrcReg1_uqw, hiIndex, loIndex); |
| int64_t arg1 = arg1Bits | |
| (0 - (arg1Bits & (ULL(1) << (sizeBits - 1)))); |
| uint64_t arg2Bits = bits(FpSrcReg2_uqw, hiIndex, loIndex); |
| int64_t arg2 = arg2Bits | |
| (0 - (arg2Bits & (ULL(1) << (sizeBits - 1)))); |
| uint64_t resBits; |
| |
| if (signedOp()) { |
| if (arg1 < arg2) { |
| resBits = arg1Bits; |
| } else { |
| resBits = arg2Bits; |
| } |
| } else { |
| if (arg1Bits < arg2Bits) { |
| resBits = arg1Bits; |
| } else { |
| resBits = arg2Bits; |
| } |
| } |
| result = insertBits(result, hiIndex, loIndex, resBits); |
| } |
| FpDestReg_uqw = result; |
| ''' |
| |
| class Mmaxi(MediaOp): |
| op_class = 'SimdCmpOp' |
| code = ''' |
| |
| assert(srcSize == destSize); |
| int size = srcSize; |
| int sizeBits = size * 8; |
| int items = numItems(size); |
| uint64_t result = FpDestReg_uqw; |
| |
| for (int i = 0; i < items; i++) { |
| int hiIndex = (i + 1) * sizeBits - 1; |
| int loIndex = (i + 0) * sizeBits; |
| uint64_t arg1Bits = bits(FpSrcReg1_uqw, hiIndex, loIndex); |
| int64_t arg1 = arg1Bits | |
| (0 - (arg1Bits & (ULL(1) << (sizeBits - 1)))); |
| uint64_t arg2Bits = bits(FpSrcReg2_uqw, hiIndex, loIndex); |
| int64_t arg2 = arg2Bits | |
| (0 - (arg2Bits & (ULL(1) << (sizeBits - 1)))); |
| uint64_t resBits; |
| |
| if (signedOp()) { |
| if (arg1 > arg2) { |
| resBits = arg1Bits; |
| } else { |
| resBits = arg2Bits; |
| } |
| } else { |
| if (arg1Bits > arg2Bits) { |
| resBits = arg1Bits; |
| } else { |
| resBits = arg2Bits; |
| } |
| } |
| result = insertBits(result, hiIndex, loIndex, resBits); |
| } |
| FpDestReg_uqw = result; |
| ''' |
| |
| class Msqrt(MediaOp): |
| op_class = 'SimdFloatSqrtOp' |
| def __init__(self, dest, src, \ |
| size = None, destSize = None, srcSize = None, ext = None): |
| super(Msqrt, self).__init__(dest, src,\ |
| "InstRegIndex(0)", size, destSize, srcSize, ext) |
| code = ''' |
| union floatInt |
| { |
| float f; |
| uint32_t i; |
| }; |
| union doubleInt |
| { |
| double d; |
| uint64_t i; |
| }; |
| |
| assert(srcSize == destSize); |
| int size = srcSize; |
| int sizeBits = size * 8; |
| assert(srcSize == 4 || srcSize == 8); |
| int items = numItems(size); |
| uint64_t result = FpDestReg_uqw; |
| |
| for (int i = 0; i < items; i++) { |
| int hiIndex = (i + 1) * sizeBits - 1; |
| int loIndex = (i + 0) * sizeBits; |
| uint64_t argBits = bits(FpSrcReg1_uqw, hiIndex, loIndex); |
| |
| if (size == 4) { |
| floatInt fi; |
| fi.i = argBits; |
| fi.f = sqrt(fi.f); |
| argBits = fi.i; |
| } else { |
| doubleInt di; |
| di.i = argBits; |
| di.d = sqrt(di.d); |
| argBits = di.i; |
| } |
| result = insertBits(result, hiIndex, loIndex, argBits); |
| } |
| FpDestReg_uqw = result; |
| ''' |
| |
| # compute approximate reciprocal --- single-precision only |
| class Mrcp(MediaOp): |
| def __init__(self, dest, src, \ |
| size = None, destSize = None, srcSize = None, ext = None): |
| super(Mrcp, self).__init__(dest, src,\ |
| "InstRegIndex(0)", size, destSize, srcSize, ext) |
| op_class = 'SimdFloatAluOp' |
| code = ''' |
| union floatInt |
| { |
| float f; |
| uint32_t i; |
| }; |
| |
| assert(srcSize == 4); // ISA defines single-precision only |
| assert(srcSize == destSize); |
| const int size = 4; |
| const int sizeBits = size * 8; |
| int items = numItems(size); |
| uint64_t result = FpDestReg_uqw; |
| |
| for (int i = 0; i < items; i++) { |
| int hiIndex = (i + 1) * sizeBits - 1; |
| int loIndex = (i + 0) * sizeBits; |
| uint64_t argBits = bits(FpSrcReg1_uqw, hiIndex, loIndex); |
| |
| floatInt fi; |
| fi.i = argBits; |
| // This is more accuracy than HW provides, but oh well |
| fi.f = 1.0 / fi.f; |
| argBits = fi.i; |
| result = insertBits(result, hiIndex, loIndex, argBits); |
| } |
| FpDestReg_uqw = result; |
| ''' |
| |
| class Maddf(MediaOp): |
| op_class = 'SimdFloatAddOp' |
| code = ''' |
| union floatInt |
| { |
| float f; |
| uint32_t i; |
| }; |
| union doubleInt |
| { |
| double d; |
| uint64_t i; |
| }; |
| |
| assert(srcSize == destSize); |
| int size = srcSize; |
| int sizeBits = size * 8; |
| assert(srcSize == 4 || srcSize == 8); |
| int items = numItems(size); |
| uint64_t result = FpDestReg_uqw; |
| |
| for (int i = 0; i < items; i++) { |
| int hiIndex = (i + 1) * sizeBits - 1; |
| int loIndex = (i + 0) * sizeBits; |
| uint64_t arg1Bits = bits(FpSrcReg1_uqw, hiIndex, loIndex); |
| uint64_t arg2Bits = bits(FpSrcReg2_uqw, hiIndex, loIndex); |
| uint64_t resBits; |
| |
| if (size == 4) { |
| floatInt arg1, arg2, res; |
| arg1.i = arg1Bits; |
| arg2.i = arg2Bits; |
| res.f = arg1.f + arg2.f; |
| resBits = res.i; |
| } else { |
| doubleInt arg1, arg2, res; |
| arg1.i = arg1Bits; |
| arg2.i = arg2Bits; |
| res.d = arg1.d + arg2.d; |
| resBits = res.i; |
| } |
| |
| result = insertBits(result, hiIndex, loIndex, resBits); |
| } |
| FpDestReg_uqw = result; |
| ''' |
| |
| class Msubf(MediaOp): |
| op_class = 'SimdFloatAddOp' |
| code = ''' |
| union floatInt |
| { |
| float f; |
| uint32_t i; |
| }; |
| union doubleInt |
| { |
| double d; |
| uint64_t i; |
| }; |
| |
| assert(srcSize == destSize); |
| int size = srcSize; |
| int sizeBits = size * 8; |
| assert(srcSize == 4 || srcSize == 8); |
| int items = numItems(size); |
| uint64_t result = FpDestReg_uqw; |
| |
| for (int i = 0; i < items; i++) { |
| int hiIndex = (i + 1) * sizeBits - 1; |
| int loIndex = (i + 0) * sizeBits; |
| uint64_t arg1Bits = bits(FpSrcReg1_uqw, hiIndex, loIndex); |
| uint64_t arg2Bits = bits(FpSrcReg2_uqw, hiIndex, loIndex); |
| uint64_t resBits; |
| |
| if (size == 4) { |
| floatInt arg1, arg2, res; |
| arg1.i = arg1Bits; |
| arg2.i = arg2Bits; |
| res.f = arg1.f - arg2.f; |
| resBits = res.i; |
| } else { |
| doubleInt arg1, arg2, res; |
| arg1.i = arg1Bits; |
| arg2.i = arg2Bits; |
| res.d = arg1.d - arg2.d; |
| resBits = res.i; |
| } |
| |
| result = insertBits(result, hiIndex, loIndex, resBits); |
| } |
| FpDestReg_uqw = result; |
| ''' |
| |
| class Mmulf(MediaOp): |
| op_class = 'SimdFloatMultOp' |
| code = ''' |
| union floatInt |
| { |
| float f; |
| uint32_t i; |
| }; |
| union doubleInt |
| { |
| double d; |
| uint64_t i; |
| }; |
| |
| assert(srcSize == destSize); |
| int size = srcSize; |
| int sizeBits = size * 8; |
| assert(srcSize == 4 || srcSize == 8); |
| int items = numItems(size); |
| uint64_t result = FpDestReg_uqw; |
| |
| for (int i = 0; i < items; i++) { |
| int hiIndex = (i + 1) * sizeBits - 1; |
| int loIndex = (i + 0) * sizeBits; |
| uint64_t arg1Bits = bits(FpSrcReg1_uqw, hiIndex, loIndex); |
| uint64_t arg2Bits = bits(FpSrcReg2_uqw, hiIndex, loIndex); |
| uint64_t resBits; |
| |
| if (size == 4) { |
| floatInt arg1, arg2, res; |
| arg1.i = arg1Bits; |
| arg2.i = arg2Bits; |
| res.f = arg1.f * arg2.f; |
| resBits = res.i; |
| } else { |
| doubleInt arg1, arg2, res; |
| arg1.i = arg1Bits; |
| arg2.i = arg2Bits; |
| res.d = arg1.d * arg2.d; |
| resBits = res.i; |
| } |
| |
| result = insertBits(result, hiIndex, loIndex, resBits); |
| } |
| FpDestReg_uqw = result; |
| ''' |
| |
| class Mdivf(MediaOp): |
| op_class = 'SimdFloatDivOp' |
| code = ''' |
| union floatInt |
| { |
| float f; |
| uint32_t i; |
| }; |
| union doubleInt |
| { |
| double d; |
| uint64_t i; |
| }; |
| |
| assert(srcSize == destSize); |
| int size = srcSize; |
| int sizeBits = size * 8; |
| assert(srcSize == 4 || srcSize == 8); |
| int items = numItems(size); |
| uint64_t result = FpDestReg_uqw; |
| |
| for (int i = 0; i < items; i++) { |
| int hiIndex = (i + 1) * sizeBits - 1; |
| int loIndex = (i + 0) * sizeBits; |
| uint64_t arg1Bits = bits(FpSrcReg1_uqw, hiIndex, loIndex); |
| uint64_t arg2Bits = bits(FpSrcReg2_uqw, hiIndex, loIndex); |
| uint64_t resBits; |
| |
| if (size == 4) { |
| floatInt arg1, arg2, res; |
| arg1.i = arg1Bits; |
| arg2.i = arg2Bits; |
| res.f = arg1.f / arg2.f; |
| resBits = res.i; |
| } else { |
| doubleInt arg1, arg2, res; |
| arg1.i = arg1Bits; |
| arg2.i = arg2Bits; |
| res.d = arg1.d / arg2.d; |
| resBits = res.i; |
| } |
| |
| result = insertBits(result, hiIndex, loIndex, resBits); |
| } |
| FpDestReg_uqw = result; |
| ''' |
| |
| class Maddi(MediaOp): |
| op_class = 'SimdAddOp' |
| code = ''' |
| assert(srcSize == destSize); |
| int size = srcSize; |
| int sizeBits = size * 8; |
| int items = numItems(size); |
| uint64_t result = FpDestReg_uqw; |
| |
| for (int i = 0; i < items; i++) { |
| int hiIndex = (i + 1) * sizeBits - 1; |
| int loIndex = (i + 0) * sizeBits; |
| uint64_t arg1Bits = bits(FpSrcReg1_uqw, hiIndex, loIndex); |
| uint64_t arg2Bits = bits(FpSrcReg2_uqw, hiIndex, loIndex); |
| uint64_t resBits = arg1Bits + arg2Bits; |
| |
| if (ext & 0x2) { |
| if (signedOp()) { |
| int arg1Sign = bits(arg1Bits, sizeBits - 1); |
| int arg2Sign = bits(arg2Bits, sizeBits - 1); |
| int resSign = bits(resBits, sizeBits - 1); |
| if ((arg1Sign == arg2Sign) && (arg1Sign != resSign)) { |
| if (resSign == 0) |
| resBits = (ULL(1) << (sizeBits - 1)); |
| else |
| resBits = mask(sizeBits - 1); |
| } |
| } else { |
| if (findCarry(sizeBits, resBits, arg1Bits, arg2Bits)) |
| resBits = mask(sizeBits); |
| } |
| } |
| |
| result = insertBits(result, hiIndex, loIndex, resBits); |
| } |
| FpDestReg_uqw = result; |
| ''' |
| |
| class Msubi(MediaOp): |
| op_class = 'SimdAddOp' |
| code = ''' |
| assert(srcSize == destSize); |
| int size = srcSize; |
| int sizeBits = size * 8; |
| int items = numItems(size); |
| uint64_t result = FpDestReg_uqw; |
| |
| for (int i = 0; i < items; i++) { |
| int hiIndex = (i + 1) * sizeBits - 1; |
| int loIndex = (i + 0) * sizeBits; |
| uint64_t arg1Bits = bits(FpSrcReg1_uqw, hiIndex, loIndex); |
| uint64_t arg2Bits = bits(FpSrcReg2_uqw, hiIndex, loIndex); |
| uint64_t resBits = arg1Bits - arg2Bits; |
| |
| if (ext & 0x2) { |
| if (signedOp()) { |
| int arg1Sign = bits(arg1Bits, sizeBits - 1); |
| int arg2Sign = !bits(arg2Bits, sizeBits - 1); |
| int resSign = bits(resBits, sizeBits - 1); |
| if ((arg1Sign == arg2Sign) && (arg1Sign != resSign)) { |
| if (resSign == 0) |
| resBits = (ULL(1) << (sizeBits - 1)); |
| else |
| resBits = mask(sizeBits - 1); |
| } |
| } else { |
| if (arg2Bits > arg1Bits) { |
| resBits = 0; |
| } else if (!findCarry(sizeBits, resBits, |
| arg1Bits, ~arg2Bits)) { |
| resBits = mask(sizeBits); |
| } |
| } |
| } |
| |
| result = insertBits(result, hiIndex, loIndex, resBits); |
| } |
| FpDestReg_uqw = result; |
| ''' |
| |
| class Mmuli(MediaOp): |
| op_class = 'SimdMultOp' |
| code = ''' |
| int srcBits = srcSize * 8; |
| int destBits = destSize * 8; |
| assert(destBits <= 64); |
| assert(destSize >= srcSize); |
| int items = numItems(destSize); |
| uint64_t result = FpDestReg_uqw; |
| |
| for (int i = 0; i < items; i++) { |
| int offset = 0; |
| if (ext & 16) { |
| if (ext & 32) |
| offset = i * (destBits - srcBits); |
| else |
| offset = i * (destBits - srcBits) + srcBits; |
| } |
| int srcHiIndex = (i + 1) * srcBits - 1 + offset; |
| int srcLoIndex = (i + 0) * srcBits + offset; |
| uint64_t arg1Bits = bits(FpSrcReg1_uqw, srcHiIndex, srcLoIndex); |
| uint64_t arg2Bits = bits(FpSrcReg2_uqw, srcHiIndex, srcLoIndex); |
| uint64_t resBits; |
| |
| if (signedOp()) { |
| int64_t arg1 = arg1Bits | |
| (0 - (arg1Bits & (ULL(1) << (srcBits - 1)))); |
| int64_t arg2 = arg2Bits | |
| (0 - (arg2Bits & (ULL(1) << (srcBits - 1)))); |
| resBits = (uint64_t)(arg1 * arg2); |
| } else { |
| resBits = arg1Bits * arg2Bits; |
| } |
| |
| if (ext & 0x4) |
| resBits += (ULL(1) << (destBits - 1)); |
| |
| if (multHi()) |
| resBits >>= destBits; |
| |
| int destHiIndex = (i + 1) * destBits - 1; |
| int destLoIndex = (i + 0) * destBits; |
| result = insertBits(result, destHiIndex, destLoIndex, resBits); |
| } |
| FpDestReg_uqw = result; |
| ''' |
| |
| class Mavg(MediaOp): |
| op_class = 'SimdAddOp' |
| code = ''' |
| assert(srcSize == destSize); |
| int size = srcSize; |
| int sizeBits = size * 8; |
| int items = numItems(size); |
| uint64_t result = FpDestReg_uqw; |
| |
| for (int i = 0; i < items; i++) { |
| int hiIndex = (i + 1) * sizeBits - 1; |
| int loIndex = (i + 0) * sizeBits; |
| uint64_t arg1Bits = bits(FpSrcReg1_uqw, hiIndex, loIndex); |
| uint64_t arg2Bits = bits(FpSrcReg2_uqw, hiIndex, loIndex); |
| uint64_t resBits = (arg1Bits + arg2Bits + 1) / 2; |
| |
| result = insertBits(result, hiIndex, loIndex, resBits); |
| } |
| FpDestReg_uqw = result; |
| ''' |
| |
| class Msad(MediaOp): |
| op_class = 'SimdAddOp' |
| code = ''' |
| int srcBits = srcSize * 8; |
| int items = sizeof(FloatRegBits) / srcSize; |
| |
| uint64_t sum = 0; |
| for (int i = 0; i < items; i++) { |
| int hiIndex = (i + 1) * srcBits - 1; |
| int loIndex = (i + 0) * srcBits; |
| uint64_t arg1Bits = bits(FpSrcReg1_uqw, hiIndex, loIndex); |
| uint64_t arg2Bits = bits(FpSrcReg2_uqw, hiIndex, loIndex); |
| int64_t resBits = arg1Bits - arg2Bits; |
| if (resBits < 0) |
| resBits = -resBits; |
| sum += resBits; |
| } |
| FpDestReg_uqw = sum & mask(destSize * 8); |
| ''' |
| |
| class Msrl(MediaOp): |
| op_class = 'SimdShiftOp' |
| code = ''' |
| |
| assert(srcSize == destSize); |
| int size = srcSize; |
| int sizeBits = size * 8; |
| int items = numItems(size); |
| uint64_t shiftAmt = op2_uqw; |
| uint64_t result = FpDestReg_uqw; |
| |
| for (int i = 0; i < items; i++) { |
| int hiIndex = (i + 1) * sizeBits - 1; |
| int loIndex = (i + 0) * sizeBits; |
| uint64_t arg1Bits = bits(FpSrcReg1_uqw, hiIndex, loIndex); |
| uint64_t resBits; |
| if (shiftAmt >= sizeBits) { |
| resBits = 0; |
| } else { |
| resBits = (arg1Bits >> shiftAmt) & |
| mask(sizeBits - shiftAmt); |
| } |
| |
| result = insertBits(result, hiIndex, loIndex, resBits); |
| } |
| FpDestReg_uqw = result; |
| ''' |
| |
| class Msra(MediaOp): |
| op_class = 'SimdShiftOp' |
| code = ''' |
| |
| assert(srcSize == destSize); |
| int size = srcSize; |
| int sizeBits = size * 8; |
| int items = numItems(size); |
| uint64_t shiftAmt = op2_uqw; |
| uint64_t result = FpDestReg_uqw; |
| |
| for (int i = 0; i < items; i++) { |
| int hiIndex = (i + 1) * sizeBits - 1; |
| int loIndex = (i + 0) * sizeBits; |
| uint64_t arg1Bits = bits(FpSrcReg1_uqw, hiIndex, loIndex); |
| uint64_t resBits; |
| if (shiftAmt >= sizeBits) { |
| if (bits(arg1Bits, sizeBits - 1)) |
| resBits = mask(sizeBits); |
| else |
| resBits = 0; |
| } else { |
| resBits = (arg1Bits >> shiftAmt); |
| resBits = resBits | |
| (0 - (resBits & (ULL(1) << (sizeBits - 1 - shiftAmt)))); |
| } |
| |
| result = insertBits(result, hiIndex, loIndex, resBits); |
| } |
| FpDestReg_uqw = result; |
| ''' |
| |
| class Msll(MediaOp): |
| op_class = 'SimdShiftOp' |
| code = ''' |
| |
| assert(srcSize == destSize); |
| int size = srcSize; |
| int sizeBits = size * 8; |
| int items = numItems(size); |
| uint64_t shiftAmt = op2_uqw; |
| uint64_t result = FpDestReg_uqw; |
| |
| for (int i = 0; i < items; i++) { |
| int hiIndex = (i + 1) * sizeBits - 1; |
| int loIndex = (i + 0) * sizeBits; |
| uint64_t arg1Bits = bits(FpSrcReg1_uqw, hiIndex, loIndex); |
| uint64_t resBits; |
| if (shiftAmt >= sizeBits) { |
| resBits = 0; |
| } else { |
| resBits = (arg1Bits << shiftAmt); |
| } |
| |
| result = insertBits(result, hiIndex, loIndex, resBits); |
| } |
| FpDestReg_uqw = result; |
| ''' |
| |
| class Cvtf2i(MediaOp): |
| def __init__(self, dest, src, \ |
| size = None, destSize = None, srcSize = None, ext = None): |
| super(Cvtf2i, self).__init__(dest, src,\ |
| "InstRegIndex(0)", size, destSize, srcSize, ext) |
| op_class = 'SimdFloatCvtOp' |
| code = ''' |
| union floatInt |
| { |
| float f; |
| uint32_t i; |
| }; |
| union doubleInt |
| { |
| double d; |
| uint64_t i; |
| }; |
| |
| assert(destSize == 4 || destSize == 8); |
| assert(srcSize == 4 || srcSize == 8); |
| int srcSizeBits = srcSize * 8; |
| int destSizeBits = destSize * 8; |
| int items; |
| int srcStart = 0; |
| int destStart = 0; |
| if (srcSize == 2 * destSize) { |
| items = numItems(srcSize); |
| if (ext & 0x2) |
| destStart = destSizeBits * items; |
| } else if (destSize == 2 * srcSize) { |
| items = numItems(destSize); |
| if (ext & 0x2) |
| srcStart = srcSizeBits * items; |
| } else { |
| items = numItems(destSize); |
| } |
| uint64_t result = FpDestReg_uqw; |
| |
| for (int i = 0; i < items; i++) { |
| int srcHiIndex = srcStart + (i + 1) * srcSizeBits - 1; |
| int srcLoIndex = srcStart + (i + 0) * srcSizeBits; |
| uint64_t argBits = bits(FpSrcReg1_uqw, srcHiIndex, srcLoIndex); |
| double arg; |
| |
| if (srcSize == 4) { |
| floatInt fi; |
| fi.i = argBits; |
| arg = fi.f; |
| } else { |
| doubleInt di; |
| di.i = argBits; |
| arg = di.d; |
| } |
| |
| if (ext & 0x4) { |
| if (arg >= 0) |
| arg += 0.5; |
| else |
| arg -= 0.5; |
| } |
| |
| if (destSize == 4) { |
| int32_t i_arg = (int32_t)arg; |
| argBits = *((uint32_t*)&i_arg); |
| } else { |
| int64_t i_arg = (int64_t)arg; |
| argBits = *((uint64_t*)&i_arg); |
| } |
| int destHiIndex = destStart + (i + 1) * destSizeBits - 1; |
| int destLoIndex = destStart + (i + 0) * destSizeBits; |
| result = insertBits(result, destHiIndex, destLoIndex, argBits); |
| } |
| FpDestReg_uqw = result; |
| ''' |
| |
| class Cvti2f(MediaOp): |
| def __init__(self, dest, src, \ |
| size = None, destSize = None, srcSize = None, ext = None): |
| super(Cvti2f, self).__init__(dest, src,\ |
| "InstRegIndex(0)", size, destSize, srcSize, ext) |
| op_class = 'SimdFloatCvtOp' |
| code = ''' |
| union floatInt |
| { |
| float f; |
| uint32_t i; |
| }; |
| union doubleInt |
| { |
| double d; |
| uint64_t i; |
| }; |
| |
| assert(destSize == 4 || destSize == 8); |
| assert(srcSize == 4 || srcSize == 8); |
| int srcSizeBits = srcSize * 8; |
| int destSizeBits = destSize * 8; |
| int items; |
| int srcStart = 0; |
| int destStart = 0; |
| if (srcSize == 2 * destSize) { |
| items = numItems(srcSize); |
| if (ext & 0x2) |
| destStart = destSizeBits * items; |
| } else if (destSize == 2 * srcSize) { |
| items = numItems(destSize); |
| if (ext & 0x2) |
| srcStart = srcSizeBits * items; |
| } else { |
| items = numItems(destSize); |
| } |
| uint64_t result = FpDestReg_uqw; |
| |
| for (int i = 0; i < items; i++) { |
| int srcHiIndex = srcStart + (i + 1) * srcSizeBits - 1; |
| int srcLoIndex = srcStart + (i + 0) * srcSizeBits; |
| uint64_t argBits = bits(FpSrcReg1_uqw, srcHiIndex, srcLoIndex); |
| |
| int64_t sArg = argBits | |
| (0 - (argBits & (ULL(1) << (srcSizeBits - 1)))); |
| double arg = sArg; |
| |
| if (destSize == 4) { |
| floatInt fi; |
| fi.f = arg; |
| argBits = fi.i; |
| } else { |
| doubleInt di; |
| di.d = arg; |
| argBits = di.i; |
| } |
| int destHiIndex = destStart + (i + 1) * destSizeBits - 1; |
| int destLoIndex = destStart + (i + 0) * destSizeBits; |
| result = insertBits(result, destHiIndex, destLoIndex, argBits); |
| } |
| FpDestReg_uqw = result; |
| ''' |
| |
| class Cvtf2f(MediaOp): |
| def __init__(self, dest, src, \ |
| size = None, destSize = None, srcSize = None, ext = None): |
| super(Cvtf2f, self).__init__(dest, src,\ |
| "InstRegIndex(0)", size, destSize, srcSize, ext) |
| op_class = 'SimdFloatCvtOp' |
| code = ''' |
| union floatInt |
| { |
| float f; |
| uint32_t i; |
| }; |
| union doubleInt |
| { |
| double d; |
| uint64_t i; |
| }; |
| |
| assert(destSize == 4 || destSize == 8); |
| assert(srcSize == 4 || srcSize == 8); |
| int srcSizeBits = srcSize * 8; |
| int destSizeBits = destSize * 8; |
| int items; |
| int srcStart = 0; |
| int destStart = 0; |
| if (srcSize == 2 * destSize) { |
| items = numItems(srcSize); |
| if (ext & 0x2) |
| destStart = destSizeBits * items; |
| } else if (destSize == 2 * srcSize) { |
| items = numItems(destSize); |
| if (ext & 0x2) |
| srcStart = srcSizeBits * items; |
| } else { |
| items = numItems(destSize); |
| } |
| uint64_t result = FpDestReg_uqw; |
| |
| for (int i = 0; i < items; i++) { |
| int srcHiIndex = srcStart + (i + 1) * srcSizeBits - 1; |
| int srcLoIndex = srcStart + (i + 0) * srcSizeBits; |
| uint64_t argBits = bits(FpSrcReg1_uqw, srcHiIndex, srcLoIndex); |
| double arg; |
| |
| if (srcSize == 4) { |
| floatInt fi; |
| fi.i = argBits; |
| arg = fi.f; |
| } else { |
| doubleInt di; |
| di.i = argBits; |
| arg = di.d; |
| } |
| if (destSize == 4) { |
| floatInt fi; |
| fi.f = arg; |
| argBits = fi.i; |
| } else { |
| doubleInt di; |
| di.d = arg; |
| argBits = di.i; |
| } |
| int destHiIndex = destStart + (i + 1) * destSizeBits - 1; |
| int destLoIndex = destStart + (i + 0) * destSizeBits; |
| result = insertBits(result, destHiIndex, destLoIndex, argBits); |
| } |
| FpDestReg_uqw = result; |
| ''' |
| |
| class Mcmpi2r(MediaOp): |
| op_class = 'SimdCvtOp' |
| code = ''' |
| union floatInt |
| { |
| float f; |
| uint32_t i; |
| }; |
| union doubleInt |
| { |
| double d; |
| uint64_t i; |
| }; |
| |
| assert(srcSize == destSize); |
| int size = srcSize; |
| int sizeBits = size * 8; |
| int items = numItems(size); |
| uint64_t result = FpDestReg_uqw; |
| |
| for (int i = 0; i < items; i++) { |
| int hiIndex = (i + 1) * sizeBits - 1; |
| int loIndex = (i + 0) * sizeBits; |
| uint64_t arg1Bits = bits(FpSrcReg1_uqw, hiIndex, loIndex); |
| int64_t arg1 = arg1Bits | |
| (0 - (arg1Bits & (ULL(1) << (sizeBits - 1)))); |
| uint64_t arg2Bits = bits(FpSrcReg2_uqw, hiIndex, loIndex); |
| int64_t arg2 = arg2Bits | |
| (0 - (arg2Bits & (ULL(1) << (sizeBits - 1)))); |
| |
| uint64_t resBits = 0; |
| if (((ext & 0x2) == 0 && arg1 == arg2) || |
| ((ext & 0x2) == 0x2 && arg1 > arg2)) |
| resBits = mask(sizeBits); |
| |
| result = insertBits(result, hiIndex, loIndex, resBits); |
| } |
| FpDestReg_uqw = result; |
| ''' |
| |
| class Mcmpf2r(MediaOp): |
| op_class = 'SimdFloatCvtOp' |
| code = ''' |
| union floatInt |
| { |
| float f; |
| uint32_t i; |
| }; |
| union doubleInt |
| { |
| double d; |
| uint64_t i; |
| }; |
| |
| assert(srcSize == destSize); |
| int size = srcSize; |
| int sizeBits = size * 8; |
| int items = numItems(size); |
| uint64_t result = FpDestReg_uqw; |
| |
| for (int i = 0; i < items; i++) { |
| int hiIndex = (i + 1) * sizeBits - 1; |
| int loIndex = (i + 0) * sizeBits; |
| uint64_t arg1Bits = bits(FpSrcReg1_uqw, hiIndex, loIndex); |
| uint64_t arg2Bits = bits(FpSrcReg2_uqw, hiIndex, loIndex); |
| double arg1, arg2; |
| |
| if (size == 4) { |
| floatInt fi; |
| fi.i = arg1Bits; |
| arg1 = fi.f; |
| fi.i = arg2Bits; |
| arg2 = fi.f; |
| } else { |
| doubleInt di; |
| di.i = arg1Bits; |
| arg1 = di.d; |
| di.i = arg2Bits; |
| arg2 = di.d; |
| } |
| |
| uint64_t resBits = 0; |
| bool nanop = std::isnan(arg1) || std::isnan(arg2); |
| switch (ext & mask(3)) { |
| case 0: |
| if (arg1 == arg2 && !nanop) |
| resBits = mask(sizeBits); |
| break; |
| case 1: |
| if (arg1 < arg2 && !nanop) |
| resBits = mask(sizeBits); |
| break; |
| case 2: |
| if (arg1 <= arg2 && !nanop) |
| resBits = mask(sizeBits); |
| break; |
| case 3: |
| if (nanop) |
| resBits = mask(sizeBits); |
| break; |
| case 4: |
| if (arg1 != arg2 || nanop) |
| resBits = mask(sizeBits); |
| break; |
| case 5: |
| if (!(arg1 < arg2) || nanop) |
| resBits = mask(sizeBits); |
| break; |
| case 6: |
| if (!(arg1 <= arg2) || nanop) |
| resBits = mask(sizeBits); |
| break; |
| case 7: |
| if (!nanop) |
| resBits = mask(sizeBits); |
| break; |
| }; |
| |
| result = insertBits(result, hiIndex, loIndex, resBits); |
| } |
| FpDestReg_uqw = result; |
| ''' |
| |
| class Mcmpf2rf(MediaOp): |
| def __init__(self, src1, src2,\ |
| size = None, destSize = None, srcSize = None, ext = None): |
| super(Mcmpf2rf, self).__init__("InstRegIndex(0)", src1,\ |
| src2, size, destSize, srcSize, ext) |
| op_class = 'SimdFloatCvtOp' |
| code = ''' |
| union floatInt |
| { |
| float f; |
| uint32_t i; |
| }; |
| union doubleInt |
| { |
| double d; |
| uint64_t i; |
| }; |
| |
| assert(srcSize == destSize); |
| assert(srcSize == 4 || srcSize == 8); |
| int size = srcSize; |
| int sizeBits = size * 8; |
| |
| double arg1, arg2; |
| uint64_t arg1Bits = bits(FpSrcReg1_uqw, sizeBits - 1, 0); |
| uint64_t arg2Bits = bits(FpSrcReg2_uqw, sizeBits - 1, 0); |
| if (size == 4) { |
| floatInt fi; |
| fi.i = arg1Bits; |
| arg1 = fi.f; |
| fi.i = arg2Bits; |
| arg2 = fi.f; |
| } else { |
| doubleInt di; |
| di.i = arg1Bits; |
| arg1 = di.d; |
| di.i = arg2Bits; |
| arg2 = di.d; |
| } |
| |
| // ZF PF CF |
| // Unordered 1 1 1 |
| // Greater than 0 0 0 |
| // Less than 0 0 1 |
| // Equal 1 0 0 |
| // OF = SF = AF = 0 |
| ccFlagBits = ccFlagBits & ~(SFBit | AFBit | ZFBit | PFBit); |
| cfofBits = cfofBits & ~(OFBit | CFBit); |
| |
| if (std::isnan(arg1) || std::isnan(arg2)) { |
| ccFlagBits = ccFlagBits | (ZFBit | PFBit); |
| cfofBits = cfofBits | CFBit; |
| } |
| else if(arg1 < arg2) |
| cfofBits = cfofBits | CFBit; |
| else if(arg1 == arg2) |
| ccFlagBits = ccFlagBits | ZFBit; |
| ''' |
| |
| class Emms(MediaOp): |
| op_class = 'FloatMiscOp' |
| def __init__(self): |
| super(Emms, self).__init__('InstRegIndex(MISCREG_FTW)', |
| 'InstRegIndex(0)', 'InstRegIndex(0)', 2) |
| code = 'FTW = 0xFFFF;' |
| }}; |
