Google Git
Sign in
gem5 / public / gem5 / f0a53b8024836b3916d78453f37f9068781232a8 / . / src / arch / arm / isa / insts / sve.isa
blob: c46a34da4ee7c13aa6ff776f2d1c8dd9b1e4311b [file] [log] [blame]
// Copyright (c) 2017-2019 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
// @file Definition of SVE instructions.
output header {{
// Decodes unary, constructive, predicated (merging) SVE instructions,
// handling signed and unsigned variants.
template <template <typename T> class BaseS,
template <typename T> class BaseU>
StaticInstPtr
decodeSveUnaryPred(unsigned size, unsigned u, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op1, IntRegIndex gp)
{
switch (size) {
case 0:
if (u) {
return new BaseU<uint8_t>(machInst, dest, op1, gp);
} else {
return new BaseS<int8_t>(machInst, dest, op1, gp);
}
case 1:
if (u) {
return new BaseU<uint16_t>(machInst, dest, op1, gp);
} else {
return new BaseS<int16_t>(machInst, dest, op1, gp);
}
case 2:
if (u) {
return new BaseU<uint32_t>(machInst, dest, op1, gp);
} else {
return new BaseS<int32_t>(machInst, dest, op1, gp);
}
case 3:
if (u) {
return new BaseU<uint64_t>(machInst, dest, op1, gp);
} else {
return new BaseS<int64_t>(machInst, dest, op1, gp);
}
default:
return new Unknown64(machInst);
}
}
// Decodes SVE widening reductions.
// handling signed and unsigned variants.
template <template <typename T1, typename T2> class BaseS,
template <typename T1, typename T2> class BaseU>
StaticInstPtr
decodeSveWideningReduc(unsigned size, unsigned u, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op1, IntRegIndex gp)
{
switch (size) {
case 0:
if (u) {
return new BaseU<uint8_t, uint64_t>(machInst, dest, op1, gp);
} else {
return new BaseS<int8_t, int64_t>(machInst, dest, op1, gp);
}
case 1:
if (u) {
return new BaseU<uint16_t, uint64_t>(machInst, dest, op1, gp);
} else {
return new BaseS<int16_t, int64_t>(machInst, dest, op1, gp);
}
case 2:
if (u) {
return new BaseU<uint32_t, uint64_t>(machInst, dest, op1, gp);
} else {
return new BaseS<int32_t, int64_t>(machInst, dest, op1, gp);
}
case 3:
assert(u);
return new BaseU<uint64_t, uint64_t>(machInst, dest, op1, gp);
default:
return new Unknown64(machInst);
}
}
// Decodes unary, constructive, predicated (merging) SVE instructions,
// handling signed variants only.
template <template <typename T> class Base>
StaticInstPtr
decodeSveUnaryPredS(unsigned size, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op1, IntRegIndex gp)
{
switch (size) {
case 0:
return new Base<int8_t>(machInst, dest, op1, gp);
case 1:
return new Base<int16_t>(machInst, dest, op1, gp);
case 2:
return new Base<int32_t>(machInst, dest, op1, gp);
case 3:
return new Base<int64_t>(machInst, dest, op1, gp);
default:
return new Unknown64(machInst);
}
}
// Decodes unary, constructive, predicated (merging) SVE instructions,
// handling unsigned variants only.
template <template <typename T> class Base>
StaticInstPtr
decodeSveUnaryPredU(unsigned size, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op1, IntRegIndex gp)
{
switch (size) {
case 0:
return new Base<uint8_t>(machInst, dest, op1, gp);
case 1:
return new Base<uint16_t>(machInst, dest, op1, gp);
case 2:
return new Base<uint32_t>(machInst, dest, op1, gp);
case 3:
return new Base<uint64_t>(machInst, dest, op1, gp);
default:
return new Unknown64(machInst);
}
}
// Decodes unary, constructive, predicated (merging) SVE instructions,
// handling signed and unsigned variants, for small element sizes (8- to
// 32-bit).
template <template <typename T> class BaseS,
template <typename T> class BaseU>
StaticInstPtr
decodeSveUnaryPredSmall(unsigned size, unsigned u, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op1, IntRegIndex gp)
{
switch (size) {
case 0:
if (u) {
return new BaseU<uint8_t>(machInst, dest, op1, gp);
} else {
return new BaseS<int8_t>(machInst, dest, op1, gp);
}
case 1:
if (u) {
return new BaseU<uint16_t>(machInst, dest, op1, gp);
} else {
return new BaseS<int16_t>(machInst, dest, op1, gp);
}
case 2:
if (u) {
return new BaseU<uint32_t>(machInst, dest, op1, gp);
} else {
return new BaseS<int32_t>(machInst, dest, op1, gp);
}
default:
return new Unknown64(machInst);
}
}
// Decodes unary, constructive, predicated (merging) SVE instructions,
// handling floating point variants only.
template <template <typename T> class Base>
StaticInstPtr
decodeSveUnaryPredF(unsigned size, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op1, IntRegIndex gp)
{
switch (size) {
case 1:
return new Base<uint16_t>(machInst, dest, op1, gp);
case 2:
return new Base<uint32_t>(machInst, dest, op1, gp);
case 3:
return new Base<uint64_t>(machInst, dest, op1, gp);
default:
return new Unknown64(machInst);
}
}
// Decodes unary, constructive, unpredicated SVE instructions, handling
// unsigned variants only.
template <template <typename T> class Base>
StaticInstPtr
decodeSveUnaryUnpredU(unsigned size, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op1)
{
switch (size) {
case 0:
return new Base<uint8_t>(machInst, dest, op1);
case 1:
return new Base<uint16_t>(machInst, dest, op1);
case 2:
return new Base<uint32_t>(machInst, dest, op1);
case 3:
return new Base<uint64_t>(machInst, dest, op1);
default:
return new Unknown64(machInst);
}
}
// Decodes unary, constructive, unpredicated SVE instructions, handling
// floating-point variants only.
template <template <typename T> class Base>
StaticInstPtr
decodeSveUnaryUnpredF(unsigned size, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op1)
{
switch (size) {
case 1:
return new Base<uint16_t>(machInst, dest, op1);
case 2:
return new Base<uint32_t>(machInst, dest, op1);
case 3:
return new Base<uint64_t>(machInst, dest, op1);
default:
return new Unknown64(machInst);
}
}
// Decodes binary, destructive, predicated (merging) SVE instructions,
// handling signed and unsigned variants.
template <template <typename T> class BaseS,
template <typename T> class BaseU>
StaticInstPtr
decodeSveBinDestrPred(unsigned size, unsigned u, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op2, IntRegIndex gp)
{
switch (size) {
case 0:
if (u) {
return new BaseU<uint8_t>(machInst, dest, op2, gp);
} else {
return new BaseS<int8_t>(machInst, dest, op2, gp);
}
case 1:
if (u) {
return new BaseU<uint16_t>(machInst, dest, op2, gp);
} else {
return new BaseS<int16_t>(machInst, dest, op2, gp);
}
case 2:
if (u) {
return new BaseU<uint32_t>(machInst, dest, op2, gp);
} else {
return new BaseS<int32_t>(machInst, dest, op2, gp);
}
case 3:
if (u) {
return new BaseU<uint64_t>(machInst, dest, op2, gp);
} else {
return new BaseS<int64_t>(machInst, dest, op2, gp);
}
default:
return new Unknown64(machInst);
}
}
// Decodes binary with immediate operand, constructive, unpredicated
// SVE instructions, handling signed variants only.
template <template <typename T> class Base>
StaticInstPtr
decodeSveBinImmUnpredS(unsigned size, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op1, unsigned immediate)
{
switch (size) {
case 0:
return new Base<int8_t>(machInst, dest, op1, immediate);
case 1:
return new Base<int16_t>(machInst, dest, op1, immediate);
case 2:
return new Base<int32_t>(machInst, dest, op1, immediate);
case 3:
return new Base<int64_t>(machInst, dest, op1, immediate);
default:
return new Unknown64(machInst);
}
}
// Decodes binary with immediate operand, constructive, unpredicated
// SVE instructions, handling unsigned variants only.
template <template <typename T> class Base>
StaticInstPtr
decodeSveBinImmUnpredU(unsigned size, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op1, unsigned immediate)
{
switch (size) {
case 0:
return new Base<uint8_t>(machInst, dest, op1, immediate);
case 1:
return new Base<uint16_t>(machInst, dest, op1, immediate);
case 2:
return new Base<uint32_t>(machInst, dest, op1, immediate);
case 3:
return new Base<uint64_t>(machInst, dest, op1, immediate);
default:
return new Unknown64(machInst);
}
}
// Decodes binary with immediate operand, destructive, predicated (merging)
// SVE instructions, handling unsigned variants only.
template <template <typename T> class Base>
StaticInstPtr
decodeSveBinImmPredU(unsigned size, ExtMachInst machInst, IntRegIndex dest,
unsigned immediate, IntRegIndex gp)
{
switch (size) {
case 0:
return new Base<uint8_t>(machInst, dest, immediate, gp);
case 1:
return new Base<uint16_t>(machInst, dest, immediate, gp);
case 2:
return new Base<uint32_t>(machInst, dest, immediate, gp);
case 3:
return new Base<uint64_t>(machInst, dest, immediate, gp);
default:
return new Unknown64(machInst);
}
}
// Decodes binary with immediate operand, destructive, predicated (merging)
// SVE instructions, handling signed variants only.
template <template <typename T> class Base>
StaticInstPtr
decodeSveBinImmPredS(unsigned size, ExtMachInst machInst, IntRegIndex dest,
unsigned immediate, IntRegIndex gp)
{
switch (size) {
case 0:
return new Base<int8_t>(machInst, dest, immediate, gp);
case 1:
return new Base<int16_t>(machInst, dest, immediate, gp);
case 2:
return new Base<int32_t>(machInst, dest, immediate, gp);
case 3:
return new Base<int64_t>(machInst, dest, immediate, gp);
default:
return new Unknown64(machInst);
}
}
// Decodes binary with immediate operand, destructive, predicated (merging)
// SVE instructions, handling floating-point variants only.
template <template <typename T> class Base>
StaticInstPtr
decodeSveBinImmPredF(unsigned size, ExtMachInst machInst, IntRegIndex dest,
uint64_t immediate, IntRegIndex gp)
{
switch (size) {
case 1:
return new Base<uint16_t>(machInst, dest, immediate, gp);
case 2:
return new Base<uint32_t>(machInst, dest, immediate, gp);
case 3:
return new Base<uint64_t>(machInst, dest, immediate, gp);
default:
return new Unknown64(machInst);
}
}
// Decodes unary/binary with wide immediate operand, destructive,
// unpredicated SVE instructions, handling unsigned variants only.
template <template <typename T> class Base>
StaticInstPtr
decodeSveWideImmUnpredU(unsigned size, ExtMachInst machInst,
IntRegIndex dest, uint64_t immediate)
{
switch (size) {
case 0:
return new Base<uint8_t>(machInst, dest, immediate);
case 1:
return new Base<uint16_t>(machInst, dest, immediate);
case 2:
return new Base<uint32_t>(machInst, dest, immediate);
case 3:
return new Base<uint64_t>(machInst, dest, immediate);
default:
return new Unknown64(machInst);
}
}
// Decodes unary/binary with wide immediate operand, destructive,
// unpredicated SVE instructions, handling signed variants only.
template <template <typename T> class Base>
StaticInstPtr
decodeSveWideImmUnpredS(unsigned size, ExtMachInst machInst,
IntRegIndex dest, uint64_t immediate)
{
switch (size) {
case 0:
return new Base<int8_t>(machInst, dest, immediate);
case 1:
return new Base<int16_t>(machInst, dest, immediate);
case 2:
return new Base<int32_t>(machInst, dest, immediate);
case 3:
return new Base<int64_t>(machInst, dest, immediate);
default:
return new Unknown64(machInst);
}
}
// Decodes unary/binary with wide immediate operand, destructive,
// unpredicated SVE instructions, handling floating-point variants only.
template <template <typename T> class Base>
StaticInstPtr
decodeSveWideImmUnpredF(unsigned size, ExtMachInst machInst,
IntRegIndex dest, uint64_t immediate)
{
switch (size) {
case 1:
return new Base<uint16_t>(machInst, dest, immediate);
case 2:
return new Base<uint32_t>(machInst, dest, immediate);
case 3:
return new Base<uint64_t>(machInst, dest, immediate);
default:
return new Unknown64(machInst);
}
}
// Decodes unary/binary with wide immediate operand, destructive,
// predicated SVE instructions, handling unsigned variants only.
template <template <typename T> class Base>
StaticInstPtr
decodeSveWideImmPredU(unsigned size, ExtMachInst machInst,
IntRegIndex dest, uint64_t immediate, IntRegIndex gp,
bool isMerging = true)
{
switch (size) {
case 0:
return new Base<uint8_t>(machInst, dest, immediate, gp,
isMerging);
case 1:
return new Base<uint16_t>(machInst, dest, immediate, gp,
isMerging);
case 2:
return new Base<uint32_t>(machInst, dest, immediate, gp,
isMerging);
case 3:
return new Base<uint64_t>(machInst, dest, immediate, gp,
isMerging);
default:
return new Unknown64(machInst);
}
}
// Decodes unary/binary with wide immediate operand, destructive,
// predicated SVE instructions, handling floating-point variants only.
template <template <typename T> class Base>
StaticInstPtr
decodeSveWideImmPredF(unsigned size, ExtMachInst machInst,
IntRegIndex dest, uint64_t immediate, IntRegIndex gp)
{
switch (size) {
case 1:
return new Base<uint16_t>(machInst, dest, immediate, gp);
case 2:
return new Base<uint32_t>(machInst, dest, immediate, gp);
case 3:
return new Base<uint64_t>(machInst, dest, immediate, gp);
default:
return new Unknown64(machInst);
}
}
// Decodes binary, destructive, predicated (merging) SVE instructions,
// handling unsigned variants only.
template <template <typename T> class Base>
StaticInstPtr
decodeSveBinDestrPredU(unsigned size, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op2, IntRegIndex gp)
{
switch (size) {
case 0:
return new Base<uint8_t>(machInst, dest, op2, gp);
case 1:
return new Base<uint16_t>(machInst, dest, op2, gp);
case 2:
return new Base<uint32_t>(machInst, dest, op2, gp);
case 3:
return new Base<uint64_t>(machInst, dest, op2, gp);
default:
return new Unknown64(machInst);
}
}
// Decodes binary, destructive, predicated (merging) SVE instructions,
// handling signed variants only.
template <template <typename T> class Base>
StaticInstPtr
decodeSveBinDestrPredS(unsigned size, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op2, IntRegIndex gp)
{
switch (size) {
case 0:
return new Base<int8_t>(machInst, dest, op2, gp);
case 1:
return new Base<int16_t>(machInst, dest, op2, gp);
case 2:
return new Base<int32_t>(machInst, dest, op2, gp);
case 3:
return new Base<int64_t>(machInst, dest, op2, gp);
default:
return new Unknown64(machInst);
}
}
// Decodes binary, destructive, predicated (merging) SVE instructions,
// handling floating-point variants only.
template <template <typename T> class Base>
StaticInstPtr
decodeSveBinDestrPredF(unsigned size, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op2, IntRegIndex gp)
{
switch (size) {
case 1:
return new Base<uint16_t>(machInst, dest, op2, gp);
case 2:
return new Base<uint32_t>(machInst, dest, op2, gp);
case 3:
return new Base<uint64_t>(machInst, dest, op2, gp);
default:
return new Unknown64(machInst);
}
}
// Decodes binary, constructive, predicated SVE instructions, handling
// unsigned variants only.
template <template <typename T> class Base>
StaticInstPtr
decodeSveBinConstrPredU(unsigned size, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op1, IntRegIndex op2,
IntRegIndex gp, SvePredType predType)
{
switch (size) {
case 0:
return new Base<uint8_t>(machInst, dest, op1, op2, gp, predType);
case 1:
return new Base<uint16_t>(machInst, dest, op1, op2, gp, predType);
case 2:
return new Base<uint32_t>(machInst, dest, op1, op2, gp, predType);
case 3:
return new Base<uint64_t>(machInst, dest, op1, op2, gp, predType);
default:
return new Unknown64(machInst);
}
}
// Decodes binary, constructive, unpredicated SVE instructions.
template <template <typename T> class Base>
StaticInstPtr
decodeSveBinUnpred(unsigned size, unsigned u, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op1, IntRegIndex op2)
{
switch (size) {
case 0:
if (u) {
return new Base<uint8_t>(machInst, dest, op1, op2);
} else {
return new Base<int8_t>(machInst, dest, op1, op2);
}
case 1:
if (u) {
return new Base<uint16_t>(machInst, dest, op1, op2);
} else {
return new Base<int16_t>(machInst, dest, op1, op2);
}
case 2:
if (u) {
return new Base<uint32_t>(machInst, dest, op1, op2);
} else {
return new Base<int32_t>(machInst, dest, op1, op2);
}
case 3:
if (u) {
return new Base<uint64_t>(machInst, dest, op1, op2);
} else {
return new Base<int64_t>(machInst, dest, op1, op2);
}
default:
return new Unknown64(machInst);
}
}
// Decodes binary, constructive, unpredicated SVE instructions.
// Unsigned instructions only.
template <template <typename T> class Base>
StaticInstPtr
decodeSveBinUnpredU(unsigned size, ExtMachInst machInst, IntRegIndex dest,
IntRegIndex op1, IntRegIndex op2)
{
switch (size) {
case 0:
return new Base<uint8_t>(machInst, dest, op1, op2);
case 1:
return new Base<uint16_t>(machInst, dest, op1, op2);
case 2:
return new Base<uint32_t>(machInst, dest, op1, op2);
case 3:
return new Base<uint64_t>(machInst, dest, op1, op2);
default:
return new Unknown64(machInst);
}
}
// Decodes binary, constructive, unpredicated SVE instructions.
// Signed instructions only.
template <template <typename T> class Base>
StaticInstPtr
decodeSveBinUnpredS(unsigned size, ExtMachInst machInst, IntRegIndex dest,
IntRegIndex op1, IntRegIndex op2)
{
switch (size) {
case 0:
return new Base<int8_t>(machInst, dest, op1, op2);
case 1:
return new Base<int16_t>(machInst, dest, op1, op2);
case 2:
return new Base<int32_t>(machInst, dest, op1, op2);
case 3:
return new Base<int64_t>(machInst, dest, op1, op2);
default:
return new Unknown64(machInst);
}
}
// Decodes binary, costructive, unpredicated SVE instructions, handling
// floating-point variants only.
template <template <typename T> class Base>
StaticInstPtr
decodeSveBinUnpredF(unsigned size, ExtMachInst machInst, IntRegIndex dest,
IntRegIndex op1, IntRegIndex op2)
{
switch (size) {
case 1:
return new Base<uint16_t>(machInst, dest, op1, op2);
case 2:
return new Base<uint32_t>(machInst, dest, op1, op2);
case 3:
return new Base<uint64_t>(machInst, dest, op1, op2);
default:
return new Unknown64(machInst);
}
}
// Decodes SVE compare instructions - binary, predicated (zeroing),
// generating a predicate - handling floating-point variants only.
template <template <typename T> class Base>
StaticInstPtr
decodeSveCmpF(unsigned size, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op1, IntRegIndex op2,
IntRegIndex gp)
{
switch (size) {
case 1:
return new Base<uint16_t>(machInst, dest, op1, op2, gp);
case 2:
return new Base<uint32_t>(machInst, dest, op1, op2, gp);
case 3:
return new Base<uint64_t>(machInst, dest, op1, op2, gp);
default:
return new Unknown64(machInst);
}
}
// Decodes SVE compare-with-immediate instructions - binary, predicated
// (zeroing), generating a predicate - handling floating-point variants
// only.
template <template <typename T> class Base>
StaticInstPtr
decodeSveCmpImmF(unsigned size, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op1, uint64_t imm,
IntRegIndex gp)
{
switch (size) {
case 1:
return new Base<uint16_t>(machInst, dest, op1, imm, gp);
case 2:
return new Base<uint32_t>(machInst, dest, op1, imm, gp);
case 3:
return new Base<uint64_t>(machInst, dest, op1, imm, gp);
default:
return new Unknown64(machInst);
}
}
// Decodes ternary, destructive, predicated (merging) SVE instructions.
template <template <typename T> class Base>
StaticInstPtr
decodeSveTerPred(unsigned size, unsigned u, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op1, IntRegIndex op2,
IntRegIndex gp)
{
switch (size) {
case 0:
if (u) {
return new Base<uint8_t>(machInst, dest, op1, op2, gp);
} else {
return new Base<int8_t>(machInst, dest, op1, op2, gp);
}
case 1:
if (u) {
return new Base<uint16_t>(machInst, dest, op1, op2, gp);
} else {
return new Base<int16_t>(machInst, dest, op1, op2, gp);
}
case 2:
if (u) {
return new Base<uint32_t>(machInst, dest, op1, op2, gp);
} else {
return new Base<int32_t>(machInst, dest, op1, op2, gp);
}
case 3:
if (u) {
return new Base<uint64_t>(machInst, dest, op1, op2, gp);
} else {
return new Base<int64_t>(machInst, dest, op1, op2, gp);
}
default:
return new Unknown64(machInst);
}
}
// Decodes ternary, destructive, predicated (merging) SVE instructions,
// handling wide signed variants only. XXX: zeroing for CMP instructions.
template <template <typename T> class Base>
StaticInstPtr
decodeSveTerPredWS(unsigned size, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op1, IntRegIndex op2,
IntRegIndex gp)
{
switch (size) {
case 0:
return new Base<int8_t>(machInst, dest, op1, op2, gp);
case 1:
return new Base<int16_t>(machInst, dest, op1, op2, gp);
case 2:
return new Base<int32_t>(machInst, dest, op1, op2, gp);
default:
return new Unknown64(machInst);
}
}
// Decodes ternary, destructive, predicated (merging) SVE instructions,
// handling wide unsigned variants only. XXX: zeroing for CMP instructions.
template <template <typename T> class Base>
StaticInstPtr
decodeSveTerPredWU(unsigned size, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op1, IntRegIndex op2,
IntRegIndex gp)
{
switch (size) {
case 0:
return new Base<uint8_t>(machInst, dest, op1, op2, gp);
case 1:
return new Base<uint16_t>(machInst, dest, op1, op2, gp);
case 2:
return new Base<uint32_t>(machInst, dest, op1, op2, gp);
default:
return new Unknown64(machInst);
}
}
// Decodes ternary, destructive, predicated (merging) SVE instructions,
// handling signed variants only. XXX: zeroing for CMP instructions.
template <template <typename T> class Base>
StaticInstPtr
decodeSveTerPredS(unsigned size, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op1, IntRegIndex op2,
IntRegIndex gp)
{
switch (size) {
case 0:
return new Base<int8_t>(machInst, dest, op1, op2, gp);
case 1:
return new Base<int16_t>(machInst, dest, op1, op2, gp);
case 2:
return new Base<int32_t>(machInst, dest, op1, op2, gp);
case 3:
return new Base<int64_t>(machInst, dest, op1, op2, gp);
default:
return new Unknown64(machInst);
}
}
// Decodes ternary, destructive, predicated (merging) SVE instructions,
// handling unsigned variants only. XXX: zeroing for CMP instructions.
template <template <typename T> class Base>
StaticInstPtr
decodeSveTerPredU(unsigned size, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op1, IntRegIndex op2,
IntRegIndex gp)
{
switch (size) {
case 0:
return new Base<uint8_t>(machInst, dest, op1, op2, gp);
case 1:
return new Base<uint16_t>(machInst, dest, op1, op2, gp);
case 2:
return new Base<uint32_t>(machInst, dest, op1, op2, gp);
case 3:
return new Base<uint64_t>(machInst, dest, op1, op2, gp);
default:
return new Unknown64(machInst);
}
}
// Decodes SVE signed unary extension instructions (8-bit source element
// size)
template <template <typename TS, typename TD> class Base>
StaticInstPtr
decodeSveUnaryExtendFromBPredS(unsigned dsize, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op1,
IntRegIndex gp)
{
switch (dsize) {
case 1:
return new Base<int8_t, int16_t>(machInst, dest, op1, gp);
case 2:
return new Base<int8_t, int32_t>(machInst, dest, op1, gp);
case 3:
return new Base<int8_t, int64_t>(machInst, dest, op1, gp);
}
return new Unknown64(machInst);
}
// Decodes SVE unsigned unary extension instructions (8-bit source element
// size)
template <template <typename TS, typename TD> class Base>
StaticInstPtr
decodeSveUnaryExtendFromBPredU(unsigned dsize, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op1,
IntRegIndex gp)
{
switch (dsize) {
case 1:
return new Base<uint8_t, uint16_t>(machInst, dest, op1, gp);
case 2:
return new Base<uint8_t, uint32_t>(machInst, dest, op1, gp);
case 3:
return new Base<uint8_t, uint64_t>(machInst, dest, op1, gp);
}
return new Unknown64(machInst);
}
// Decodes SVE signed unary extension instructions (16-bit source element
// size)
template <template <typename TS, typename TD> class Base>
StaticInstPtr
decodeSveUnaryExtendFromHPredS(unsigned dsize, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op1,
IntRegIndex gp)
{
switch (dsize) {
case 2:
return new Base<int16_t, int32_t>(machInst, dest, op1, gp);
case 3:
return new Base<int16_t, int64_t>(machInst, dest, op1, gp);
}
return new Unknown64(machInst);
}
// Decodes SVE unsigned unary extension instructions (16-bit source element
// size)
template <template <typename TS, typename TD> class Base>
StaticInstPtr
decodeSveUnaryExtendFromHPredU(unsigned dsize, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op1,
IntRegIndex gp)
{
switch (dsize) {
case 2:
return new Base<uint16_t, uint32_t>(machInst, dest, op1, gp);
case 3:
return new Base<uint16_t, uint64_t>(machInst, dest, op1, gp);
}
return new Unknown64(machInst);
}
// Decodes ternary, destructive, predicated (merging) SVE instructions,
// handling floating-point variants only.
template <template <typename T> class Base>
StaticInstPtr
decodeSveTerPredF(unsigned size, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op1, IntRegIndex op2,
IntRegIndex gp)
{
switch (size) {
case 1:
return new Base<uint16_t>(machInst, dest, op1, op2, gp);
case 2:
return new Base<uint32_t>(machInst, dest, op1, op2, gp);
case 3:
return new Base<uint64_t>(machInst, dest, op1, op2, gp);
default:
return new Unknown64(machInst);
}
}
// Decodes ternary with immediate operand, destructive, unpredicated SVE
// instructions handling floating-point variants only.
template <template <typename T> class Base>
StaticInstPtr
decodeSveTerImmUnpredF(unsigned size, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op2, uint8_t imm)
{
switch (size) {
case 1:
return new Base<uint16_t>(machInst, dest, op2, imm);
case 2:
return new Base<uint32_t>(machInst, dest, op2, imm);
case 3:
return new Base<uint64_t>(machInst, dest, op2, imm);
default:
return new Unknown64(machInst);
}
}
// Decodes SVE PTRUE(S) instructions.
template <template <typename T> class Base>
StaticInstPtr
decodeSvePtrue(unsigned size, ExtMachInst machInst,
IntRegIndex dest, uint8_t imm)
{
switch (size) {
case 0:
return new Base<uint8_t>(machInst, dest, imm);
case 1:
return new Base<uint16_t>(machInst, dest, imm);
case 2:
return new Base<uint32_t>(machInst, dest, imm);
case 3:
return new Base<uint64_t>(machInst, dest, imm);
default:
return new Unknown64(machInst);
}
}
// Decodes SVE predicate count instructions, scalar signed variant only
template <template <typename T> class Base>
StaticInstPtr
decodeSvePredCountS(unsigned size, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op1)
{
switch (size) {
case 0:
return new Base<int8_t>(machInst, dest, op1);
case 1:
return new Base<int16_t>(machInst, dest, op1);
case 2:
return new Base<int32_t>(machInst, dest, op1);
case 3:
return new Base<int64_t>(machInst, dest, op1);
default:
return new Unknown64(machInst);
}
}
// Decodes SVE predicate count instructions, scalar unsigned variant only
template <template <typename T> class Base>
StaticInstPtr
decodeSvePredCountU(unsigned size, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op1)
{
switch (size) {
case 0:
return new Base<uint8_t>(machInst, dest, op1);
case 1:
return new Base<uint16_t>(machInst, dest, op1);
case 2:
return new Base<uint32_t>(machInst, dest, op1);
case 3:
return new Base<uint64_t>(machInst, dest, op1);
default:
return new Unknown64(machInst);
}
}
// Decodes SVE predicate count instructions, vector signed variant only
template <template <typename T> class Base>
StaticInstPtr
decodeSvePredCountVS(unsigned size, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op1)
{
switch (size) {
case 1:
return new Base<int16_t>(machInst, dest, op1);
case 2:
return new Base<int32_t>(machInst, dest, op1);
case 3:
return new Base<int64_t>(machInst, dest, op1);
default:
return new Unknown64(machInst);
}
}
// Decodes SVE predicate count instructions, vector unsigned variant only
template <template <typename T> class Base>
StaticInstPtr
decodeSvePredCountVU(unsigned size, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op1)
{
switch (size) {
case 1:
return new Base<uint16_t>(machInst, dest, op1);
case 2:
return new Base<uint32_t>(machInst, dest, op1);
case 3:
return new Base<uint64_t>(machInst, dest, op1);
default:
return new Unknown64(machInst);
}
}
// Decodes ternary with immediate operand, predicated SVE
// instructions handling unsigned variants only.
template <template <typename T> class Base>
StaticInstPtr
decodeSveTerImmPredU(unsigned size, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op1, int64_t imm, IntRegIndex gp)
{
switch (size) {
case 0:
return new Base<uint8_t>(machInst, dest, op1, imm, gp);
case 1:
return new Base<uint16_t>(machInst, dest, op1, imm, gp);
case 2:
return new Base<uint32_t>(machInst, dest, op1, imm, gp);
case 3:
return new Base<uint64_t>(machInst, dest, op1, imm, gp);
default:
return new Unknown64(machInst);
}
}
// Decodes ternary with immediate operand, predicated SVE
// instructions handling signed variants only.
template <template <typename T> class Base>
StaticInstPtr
decodeSveTerImmPredS(unsigned size, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op1, int64_t imm, IntRegIndex gp)
{
switch (size) {
case 0:
return new Base<int8_t>(machInst, dest, op1, imm, gp);
case 1:
return new Base<int16_t>(machInst, dest, op1, imm, gp);
case 2:
return new Base<int32_t>(machInst, dest, op1, imm, gp);
case 3:
return new Base<int64_t>(machInst, dest, op1, imm, gp);
default:
return new Unknown64(machInst);
}
}
// Decodes integer element count SVE instructions, handling
// signed variants only.
template <template <typename T> class Base>
StaticInstPtr
decodeSveElemIntCountS(unsigned size, ExtMachInst machInst,
IntRegIndex dest, uint8_t pattern, uint8_t imm4)
{
switch (size) {
case 0:
return new Base<int8_t>(machInst, dest, pattern, imm4);
case 1:
return new Base<int16_t>(machInst, dest, pattern, imm4);
case 2:
return new Base<int32_t>(machInst, dest, pattern, imm4);
case 3:
return new Base<int64_t>(machInst, dest, pattern, imm4);
default:
return new Unknown64(machInst);
}
}
// Decodes integer element count SVE instructions, handling
// unsigned variants only.
template <template <typename T> class Base>
StaticInstPtr
decodeSveElemIntCountU(unsigned size, ExtMachInst machInst,
IntRegIndex dest, uint8_t pattern, uint8_t imm4)
{
switch (size) {
case 0:
return new Base<uint8_t>(machInst, dest, pattern, imm4);
case 1:
return new Base<uint16_t>(machInst, dest, pattern, imm4);
case 2:
return new Base<uint32_t>(machInst, dest, pattern, imm4);
case 3:
return new Base<uint64_t>(machInst, dest, pattern, imm4);
default:
return new Unknown64(machInst);
}
}
// Decodes integer element count SVE instructions, handling
// signed variants from 16 to 64 bits only.
template <template <typename T> class Base>
StaticInstPtr
decodeSveElemIntCountLS(unsigned size, ExtMachInst machInst,
IntRegIndex dest, uint8_t pattern, uint8_t imm4)
{
switch (size) {
case 1:
return new Base<int16_t>(machInst, dest, pattern, imm4);
case 2:
return new Base<int32_t>(machInst, dest, pattern, imm4);
case 3:
return new Base<int64_t>(machInst, dest, pattern, imm4);
default:
return new Unknown64(machInst);
}
}
// Decodes integer element count SVE instructions, handling
// unsigned variants from 16 to 64 bits only.
template <template <typename T> class Base>
StaticInstPtr
decodeSveElemIntCountLU(unsigned size, ExtMachInst machInst,
IntRegIndex dest, uint8_t pattern, uint8_t imm4)
{
switch (size) {
case 1:
return new Base<uint16_t>(machInst, dest, pattern, imm4);
case 2:
return new Base<uint32_t>(machInst, dest, pattern, imm4);
case 3:
return new Base<uint64_t>(machInst, dest, pattern, imm4);
default:
return new Unknown64(machInst);
}
}
// Decodes SVE unpack instructions. Handling signed variants.
template <template <typename T1, typename T2> class Base>
StaticInstPtr
decodeSveUnpackS(unsigned size, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op1)
{
switch (size) {
case 1:
return new Base<int8_t, int16_t>(machInst, dest, op1);
case 2:
return new Base<int16_t, int32_t>(machInst, dest, op1);
case 3:
return new Base<int32_t, int64_t>(machInst, dest, op1);
default:
return new Unknown64(machInst);
}
}
// Decodes SVE unpack instructions. Handling unsigned variants.
template <template <typename T1, typename T2> class Base>
StaticInstPtr
decodeSveUnpackU(unsigned size, ExtMachInst machInst,
IntRegIndex dest, IntRegIndex op1)
{
switch (size) {
case 1:
return new Base<uint8_t, uint16_t>(machInst, dest, op1);
case 2:
return new Base<uint16_t, uint32_t>(machInst, dest, op1);
case 3:
return new Base<uint32_t, uint64_t>(machInst, dest, op1);
default:
return new Unknown64(machInst);
}
}
}};
let {{
header_output = ''
exec_output = ''
decoders = { 'Generic': {} }
class PredType:
NONE = 0
MERGE = 1
ZERO = 2
SELECT = 3
class CvtDir:
Narrow = 0
Widen = 1
class IndexFormat(object):
ImmImm = 'II'
ImmReg = 'IR'
RegImm = 'RI'
RegReg = 'RR'
class SrcRegType(object):
Vector = 0
Scalar = 1
SimdFpScalar = 2
Predicate = 3
class DstRegType(object):
Vector = 0
Scalar = 1
SimdFpScalar = 2
Predicate = 3
class DestType(object):
Scalar = 'false'
Vector = 'true'
class SrcSize(object):
Src32bit = 'true'
Src64bit = 'false'
class Break(object):
Before = 0
After = 1
class Unpack(object):
High = 0
Low = 1
# Generates definitions for SVE ADR instructions
def sveAdrInst(name, Name, opClass, types, op):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());
for (unsigned i = 0; i < eCount; i++) {
const Element& srcElem1 = AA64FpOp1_x[i];
Element srcElem2 = AA64FpOp2_x[i];
Element destElem = 0;
%(op)s
AA64FpDest_x[i] = destElem;
}''' % {'op': op}
iop = InstObjParams(name, 'Sve' + Name, 'SveAdrOp',
{'code': code, 'op_class': opClass}, [])
header_output += SveAdrOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in types:
substDict = {'targs' : type,
'class_name' : 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generates definition for SVE while predicate generation instructions
def sveWhileInst(name, Name, opClass, types, op,
srcSize = SrcSize.Src64bit):
global header_output, exec_output, decoders
extraPrologCode = '''
auto& destPred = PDest;'''
if 'int32_t' in types:
srcType = 'int64_t' if srcSize == SrcSize.Src64bit else 'int32_t'
else:
srcType = 'uint64_t' if srcSize == SrcSize.Src64bit else 'uint32_t'
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());
%(stype)s srcElem1 = static_cast<%(stype)s>(XOp1);
%(stype)s srcElem2 = static_cast<%(stype)s>(XOp2);
bool cond, first = false, none = true, last = true;
destPred.reset();
for (unsigned i = 0; i < eCount; i++) {
%(op)s;
last = last && cond;
none = none && !cond;
first = first || (i == 0 && cond);
PDest_x[i] = last;
srcElem1++;
}
CondCodesNZ = (first << 1) | none;
CondCodesC = !last;
CondCodesV = false;
'''%{'op': op, 'stype': srcType}
iop = InstObjParams(name, 'Sve' + Name, 'SveWhileOp',
{'code': code, 'op_class': opClass, 'srcIs32b': srcSize}, [])
iop.snippets['code'] = extraPrologCode + iop.snippets['code']
header_output += SveWhileOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in types:
substDict = {'targs' : type, 'class_name' : 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict);
# Generate definition for SVE compare & terminate instructions
def sveCompTermInst(name, Name, opClass, types, op):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
bool destElem;
Element srcElem1 = static_cast<Element>(XOp1);
Element srcElem2 = static_cast<Element>(XOp2);
%(op)s;
if (destElem) {
CondCodesNZ = CondCodesNZ | 0x2;
CondCodesV = 0;
} else {
CondCodesNZ = CondCodesNZ & ~0x2;
CondCodesV = !CondCodesC;
}
''' % {'op': op}
iop = InstObjParams(name, 'Sve' + Name, 'SveCompTermOp',
{'code': code, 'op_class': opClass}, [])
header_output += SveCompTermOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in types:
substDict = {'targs' : type, 'class_name': 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict);
# Generates definition for SVE predicate count instructions
def svePredCountInst(name, Name, opClass, types, op,
destType=DestType.Vector,
srcSize=SrcSize.Src64bit):
global header_output, exec_output, decoders
assert not (destType == DestType.Vector and
srcSize != SrcSize.Src64bit)
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());
int count = 0;
for (unsigned i = 0; i < eCount; i++) {
if (GpOp_x[i]) {
count++;
}
}'''
if destType == DestType.Vector:
code += '''
for (unsigned i = 0; i < eCount; i++) {
Element destElem = 0;
const Element& srcElem = AA64FpDestMerge_x[i];
%(op)s
AA64FpDest_x[i] = destElem;
}''' % {'op': op}
else:
code += '''
%(op)s''' % {'op': op}
iop = InstObjParams(name, 'Sve' + Name, 'SvePredCountOp',
{'code': code, 'op_class': opClass, 'srcIs32b': srcSize,
'destIsVec': destType}, [])
header_output += SvePredCountOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in types:
substDict = {'targs' : type, 'class_name' : 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict);
# Generates definition for SVE predicate count instructions (predicated)
def svePredCountPredInst(name, Name, opClass, types):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());
int count = 0;
for (unsigned i = 0; i < eCount; i++) {
if (POp1_x[i] && GpOp_x[i]) {
count++;
}
}
XDest = count;
'''
iop = InstObjParams(name, 'Sve' + Name, 'SvePredCountPredOp',
{'code': code, 'op_class': opClass}, [])
header_output += SvePredCountPredOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in types:
substDict = {'targs' : type, 'class_name' : 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generates definition for SVE Index generation instructions
def sveIndex(fmt):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());'''
if fmt == IndexFormat.ImmReg or fmt == IndexFormat.ImmImm:
code += '''
const Element& srcElem1 = imm1;'''
if fmt == IndexFormat.RegImm or fmt == IndexFormat.RegReg:
code += '''
const Element& srcElem1 = XOp1;'''
if fmt == IndexFormat.RegImm or fmt == IndexFormat.ImmImm:
code += '''
const Element& srcElem2 = imm2;'''
if fmt == IndexFormat.ImmReg or fmt == IndexFormat.RegReg:
code += '''
const Element& srcElem2 = XOp2;'''
code +='''
for (unsigned i = 0; i < eCount; i++) {
AA64FpDest_x[i] = srcElem1 + i * srcElem2;
}'''
iop = InstObjParams('index', 'SveIndex'+fmt, 'SveIndex'+fmt+'Op',
{'code': code, 'op_class': 'SimdAluOp'})
if fmt == IndexFormat.ImmImm:
header_output += SveIndexIIOpDeclare.subst(iop)
elif fmt == IndexFormat.ImmReg:
header_output += SveIndexIROpDeclare.subst(iop)
elif fmt == IndexFormat.RegImm:
header_output += SveIndexRIOpDeclare.subst(iop)
elif fmt == IndexFormat.RegReg:
header_output += SveIndexRROpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in ['int8_t', 'int16_t', 'int32_t', 'int64_t']:
substDict = {'targs': type, 'class_name': 'SveIndex'+fmt}
exec_output += SveOpExecDeclare.subst(substDict)
# Generates definitions for widening unary SVE instructions
# (always constructive)
def sveWidenUnaryInst(name, Name, opClass, types, op,
predType=PredType.NONE, decoder='Generic'):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<DElement>(
xc->tcBase());
for (unsigned i = 0; i < eCount; i++) {
SElement srcElem1 = AA64FpOp1_xd[i];
DElement destElem = 0;'''
if predType != PredType.NONE:
code += '''
if (GpOp_xd[i]) {
%(op)s
} else {
destElem = %(dest_elem)s;
}''' % {'op': op,
'dest_elem': 'AA64FpDestMerge_xd[i]'
if predType == PredType.MERGE
else '0'}
else:
code += '''
%(op)s''' % {'op': op}
code += '''
AA64FpDest_xd[i] = destElem;
}'''
iop = InstObjParams(name, 'Sve' + Name,
'SveUnaryPredOp' if predType != PredType.NONE
else 'SveUnaryUnpredOp',
{'code': code, 'op_class': opClass}, [])
if predType != PredType.NONE:
header_output += SveWideningUnaryPredOpDeclare.subst(iop)
else:
header_output += SveWideningUnaryUnpredOpDeclare.subst(iop)
exec_output += SveWideningOpExecute.subst(iop)
for type in types:
substDict = {'targs' : type,
'class_name' : 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generates definitions for unary SVE instructions (always constructive)
def sveUnaryInst(name, Name, opClass, types, op, predType=PredType.NONE,
srcRegType=SrcRegType.Vector, decoder='Generic'):
global header_output, exec_output, decoders
op1 = ('AA64FpOp1_x[i]' if srcRegType == SrcRegType.Vector
else 'XOp1' if srcRegType == SrcRegType.Scalar
else 'AA64FpOp1_x[0]')
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());
for (unsigned i = 0; i < eCount; i++) {
Element srcElem1 = %s;
Element destElem = 0;''' % op1
if predType != PredType.NONE:
code += '''
if (GpOp_x[i]) {
%(op)s
} else {
destElem = %(dest_elem)s;
}''' % {'op': op,
'dest_elem': 'AA64FpDestMerge_x[i]'
if predType == PredType.MERGE
else '0'}
else:
code += '''
%(op)s''' % {'op': op}
code += '''
AA64FpDest_x[i] = destElem;
}'''
iop = InstObjParams(name, 'Sve' + Name,
'SveUnaryPredOp' if predType != PredType.NONE
else 'SveUnaryUnpredOp',
{'code': code, 'op_class': opClass}, [])
if predType != PredType.NONE:
header_output += SveUnaryPredOpDeclare.subst(iop)
else:
header_output += SveUnaryUnpredOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in types:
substDict = {'targs' : type,
'class_name' : 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generates definitions for SVE floating-point conversions (always
# unary, constructive, merging
def sveCvtInst(name, Name, opClass, types, op, direction=CvtDir.Narrow,
decoder='Generic'):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<%(bigElemType)s>(
xc->tcBase());
for (unsigned i = 0; i < eCount; i++) {
SElement srcElem1 = AA64FpOp1_x%(bigElemSuffix)s[i] &
mask(sizeof(SElement) * 8);
DElement destElem = 0;
if (GpOp_x%(bigElemSuffix)s[i]) {
%(op)s
AA64FpDest_x%(bigElemSuffix)s[i] = destElem;
} else {
AA64FpDest_x%(bigElemSuffix)s[i] =
AA64FpDestMerge_x%(bigElemSuffix)s[i];
}
}
''' % {'op': op,
'bigElemType': 'SElement' if direction == CvtDir.Narrow
else 'DElement',
'bigElemSuffix': 's' if direction == CvtDir.Narrow else 'd'}
iop = InstObjParams(name, 'Sve' + Name, 'SveUnaryPredOp',
{'code': code, 'op_class': opClass}, [])
header_output += SveWideningUnaryPredOpDeclare.subst(iop)
exec_output += SveWideningOpExecute.subst(iop)
for type in types:
substDict = {'targs' : type,
'class_name' : 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generates definitions for associative SVE reductions
def sveAssocReducInst(name, Name, opClass, types, op, identity,
decoder='Generic'):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());
ArmISA::VecRegContainer tmpVecC;
auto auxOp1 = tmpVecC.as<Element>();
for (unsigned i = 0; i < eCount; ++i) {
auxOp1[i] = AA64FpOp1_x[i];
}
Element destElem = %(identity)s;
for (unsigned i = 0; i < eCount; i++) {
AA64FpDest_x[i] = 0; // zero upper part
if (GpOp_x[i]) {
const Element& srcElem1 = auxOp1[i];
%(op)s
}
}
AA64FpDest_x[0] = destElem;
''' % {'op': op, 'identity': identity}
iop = InstObjParams(name, 'Sve' + Name, 'SveReducOp',
{'code': code, 'op_class': opClass}, [])
header_output += SveReducOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in types:
substDict = {'targs' : type,
'class_name' : 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generates definitions for widening associative SVE reductions
def sveWideningAssocReducInst(name, Name, opClass, types, op, identity,
decoder='Generic'):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<SElement>(
xc->tcBase());
unsigned eWideCount = ArmStaticInst::getCurSveVecLen<DElement>(
xc->tcBase());
DElement destElem = %(identity)s;
for (unsigned i = 0; i < eCount; i++) {
if (GpOp_xs[i]) {
DElement srcElem1 = AA64FpOp1_xs[i];
%(op)s
}
}
AA64FpDest_xd[0] = destElem;
for (int i = 1; i < eWideCount; i++) {
AA64FpDest_xd[i] = 0;
}
''' % {'op': op, 'identity': identity}
iop = InstObjParams(name, 'Sve' + Name, 'SveReducOp',
{'code': code, 'op_class': opClass}, [])
header_output += SveWideningReducOpDeclare.subst(iop)
exec_output += SveWideningOpExecute.subst(iop)
for type in types:
substDict = {'targs' : type,
'class_name' : 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generates definitions for non-associative SVE reductions
def sveNonAssocReducInst(name, Name, opClass, types, op, identity,
decoder='Generic'):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());
ArmISA::VecRegContainer tmpVecC;
auto tmpVec = tmpVecC.as<Element>();
int ePow2Count = 1;
while (ePow2Count < eCount) {
ePow2Count *= 2;
}
for (unsigned i = 0; i < ePow2Count; i++) {
if (i < eCount && GpOp_x[i]) {
tmpVec[i] = AA64FpOp1_x[i];
} else {
tmpVec[i] = %(identity)s;
}
}
unsigned n = ePow2Count;
while (n > 1) {
unsigned max = n;
n = 0;
for (unsigned i = 0; i < max; i += 2) {
Element srcElem1 = tmpVec[i];
Element srcElem2 = tmpVec[i + 1];
Element destElem = 0;
%(op)s
tmpVec[n] = destElem;
n++;
}
}
AA64FpDest_x[0] = tmpVec[0];
for (unsigned i = 1; i < eCount; i++) {
AA64FpDest_x[i] = 0; // zero upper part
}
''' % {'op': op, 'identity': identity}
iop = InstObjParams(name, 'Sve' + Name, 'SveReducOp',
{'code': code, 'op_class': opClass}, [])
header_output += SveReducOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in types:
substDict = {'targs' : type,
'class_name' : 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generates definitions for binary SVE instructions with immediate operand
def sveBinImmInst(name, Name, opClass, types, op, predType=PredType.NONE,
decoder='Generic'):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());
for (unsigned i = 0; i < eCount; i++) {'''
if predType != PredType.NONE:
code += '''
const Element& srcElem1 = %s;''' % (
'AA64FpDestMerge_x[i]' if predType == PredType.MERGE else '0')
else:
code += '''
const Element& srcElem1 = AA64FpOp1_x[i];'''
code += '''
Element srcElem2 = imm;
Element destElem = 0;'''
if predType != PredType.NONE:
code += '''
if (GpOp_x[i]) {
%(op)s
} else {
destElem = %(dest_elem)s;
}''' % {'op': op,
'dest_elem': 'AA64FpDestMerge_x[i]'
if predType == PredType.MERGE else '0'}
else:
code += '''
%(op)s''' % {'op': op}
code += '''
AA64FpDest_x[i] = destElem;
}'''
iop = InstObjParams(name, 'Sve' + Name,
'SveBinImmPredOp' if predType != PredType.NONE
else 'SveBinImmUnpredConstrOp',
{'code': code, 'op_class': opClass}, [])
if predType != PredType.NONE:
header_output += SveBinImmPredOpDeclare.subst(iop)
else:
header_output += SveBinImmUnpredOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in types:
substDict = {'targs' : type,
'class_name' : 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generates definitions for unary and binary SVE instructions with wide
# immediate operand
def sveWideImmInst(name, Name, opClass, types, op, predType=PredType.NONE,
isUnary=False, decoder='Generic'):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());
for (unsigned i = 0; i < eCount; i++) {'''
# TODO: handle unsigned-to-signed conversion properly...
if isUnary:
code += '''
Element srcElem1 = imm;'''
else:
code += '''
const Element& srcElem1 = AA64FpDestMerge_x[i];
Element srcElem2 = imm;'''
code += '''
Element destElem = 0;'''
if predType != PredType.NONE:
code += '''
if (GpOp_x[i]) {
%(op)s
} else {
destElem = %(dest_elem)s;
}''' % {'op': op,
'dest_elem': 'AA64FpDestMerge_x[i]'
if predType == PredType.MERGE else '0'}
else:
code += '''
%(op)s''' % {'op': op}
code += '''
AA64FpDest_x[i] = destElem;
}'''
iop = InstObjParams(name, 'Sve' + Name,
'Sve%sWideImm%sOp' % (
'Unary' if isUnary else 'Bin',
'Unpred' if predType == PredType.NONE else 'Pred'),
{'code': code, 'op_class': opClass}, [])
if predType == PredType.NONE:
header_output += SveWideImmUnpredOpDeclare.subst(iop)
else:
header_output += SveWideImmPredOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in types:
substDict = {'targs' : type,
'class_name' : 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generates definitions for shift SVE instructions with wide elements
def sveShiftByWideElemsInst(name, Name, opClass, types, op,
predType=PredType.NONE, decoder='Generic'):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());
ArmISA::VecRegContainer tmpVecC;
auto auxOp2 = tmpVecC.as<Element>();
for (unsigned i = 0; i < eCount; i++) {
auxOp2[i] = AA64FpOp2_ud[i];
}
for (unsigned i = 0; i < eCount; i++) {'''
if predType != PredType.NONE:
code += '''
const Element& srcElem1 = AA64FpDestMerge_x[i];'''
else:
code += '''
const Element& srcElem1 = AA64FpOp1_x[i];'''
code += '''
const auto& srcElem2 = auxOp2[
(i * sizeof(Element) * 8) / 64];
Element destElem = 0;'''
if predType != PredType.NONE:
code += '''
if (GpOp_x[i]) {
%(op)s
} else {
destElem = %(dest_elem)s;
}''' % {'op': op,
'dest_elem': 'AA64FpDestMerge_x[i]'
if predType == PredType.MERGE else '0'}
else:
code += '''
%(op)s''' % {'op': op}
code += '''
AA64FpDest_x[i] = destElem;
}'''
iop = InstObjParams(name, 'Sve' + Name,
'SveBinDestrPredOp' if predType != PredType.NONE
else 'SveBinUnpredOp',
{'code': code, 'op_class': opClass}, [])
if predType != PredType.NONE:
header_output += SveBinDestrPredOpDeclare.subst(iop)
else:
header_output += SveBinUnpredOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in types:
substDict = {'targs' : type,
'class_name' : 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generates definitions for binary indexed SVE instructions
# (always unpredicated)
def sveBinIdxInst(name, Name, opClass, types, op, decoder='Generic'):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());
// Number of elements in a 128 bit segment
constexpr unsigned ePerSegment = 128 / sizeof(Element);
'''
code += '''
for (unsigned i = 0; i < eCount; i++) {
const auto segmentBase = i - i % ePerSegment;
const auto segmentIdx = segmentBase + index;
const Element& srcElem1 = AA64FpOp1_x[i];
const Element& srcElem2 = AA64FpOp2_x[segmentIdx];
Element destElem = 0;
'''
code += '''
%(op)s
AA64FpDest_x[i] = destElem;
}
''' % {'op': op}
baseClass = 'SveBinIdxUnpredOp'
iop = InstObjParams(name, 'Sve' + Name, baseClass,
{'code': code, 'op_class': opClass}, [])
header_output += SveBinIdxUnpredOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in types:
substDict = {'targs' : type,
'class_name' : 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generates definitions for binary SVE instructions
def sveBinInst(name, Name, opClass, types, op, predType=PredType.NONE,
isDestructive=False, customIterCode=None,
decoder='Generic'):
assert not (predType in (PredType.NONE, PredType.SELECT) and
isDestructive)
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());'''
if customIterCode is None:
code += '''
for (unsigned i = 0; i < eCount; i++) {'''
if predType == PredType.MERGE:
code += '''
const Element& srcElem1 = AA64FpDestMerge_x[i];'''
else:
code += '''
const Element& srcElem1 = AA64FpOp1_x[i];'''
code += '''
const Element& srcElem2 = AA64FpOp2_x[i];
Element destElem = 0;'''
if predType != PredType.NONE:
code += '''
if (GpOp_x[i]) {
%(op)s
} else {
destElem = %(dest_elem)s;
}''' % {'op': op,
'dest_elem':
'AA64FpDestMerge_x[i]' if predType == PredType.MERGE
else '0' if predType == PredType.ZERO
else 'srcElem2'}
else:
code += '''
%(op)s''' % {'op': op}
code += '''
AA64FpDest_x[i] = destElem;
}'''
else:
code += customIterCode
if predType == PredType.NONE:
baseClass = 'SveBinUnpredOp'
elif isDestructive:
baseClass = 'SveBinDestrPredOp'
else:
baseClass = 'SveBinConstrPredOp'
iop = InstObjParams(name, 'Sve' + Name, baseClass,
{'code': code, 'op_class': opClass}, [])
if predType == PredType.NONE:
header_output += SveBinUnpredOpDeclare.subst(iop)
elif isDestructive:
header_output += SveBinDestrPredOpDeclare.subst(iop)
else:
header_output += SveBinConstrPredOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in types:
substDict = {'targs' : type,
'class_name' : 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generates definitions for predicate logical instructions
def svePredLogicalInst(name, Name, opClass, types, op,
predType=PredType.ZERO, isFlagSetting=False,
decoder='Generic'):
global header_output, exec_output, decoders
assert predType in (PredType.ZERO, PredType.SELECT)
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());
ArmISA::VecPredRegContainer tmpPredC;
auto auxGpOp = tmpPredC.as<Element>();
for (unsigned i = 0; i < eCount; i++) {
auxGpOp[i] = GpOp_x[i];
}
for (unsigned i = 0; i < eCount; i++) {
bool srcElem1 = POp1_x[i];
bool srcElem2 = POp2_x[i];
bool destElem = false;
if (auxGpOp[i]) {
%(op)s
} else {
destElem = %(dest_elem)s;
}
PDest_x[i] = destElem;
}''' % {'op': op,
'dest_elem': 'false' if predType == PredType.ZERO
else 'srcElem2'}
extraPrologCode = ''
if isFlagSetting:
code += '''
CondCodesNZ = (destPred.firstActive(auxGpOp, eCount) << 1) |
destPred.noneActive(auxGpOp, eCount);
CondCodesC = !destPred.lastActive(auxGpOp, eCount);
CondCodesV = 0;'''
extraPrologCode += '''
auto& destPred = PDest;'''
iop = InstObjParams(name, 'Sve' + Name, 'SvePredLogicalOp',
{'code': code, 'op_class': opClass}, [])
iop.snippets['code'] = extraPrologCode + iop.snippets['code']
header_output += SvePredLogicalOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in types:
substDict = {'targs' : type,
'class_name' : 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generates definitions for predicate permute instructions
def svePredBinPermInst(name, Name, opClass, types, iterCode,
decoder='Generic'):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());'''
code += iterCode
iop = InstObjParams(name, 'Sve' + Name, 'SvePredBinPermOp',
{'code': code, 'op_class': opClass}, [])
header_output += SveBinUnpredOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in types:
substDict = {'targs' : type,
'class_name' : 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generates definitions for SVE compare instructions
# NOTE: compares are all predicated zeroing
def sveCmpInst(name, Name, opClass, types, op, isImm=False,
decoder='Generic'):
global header_output, exec_output, decoders
extraPrologCode = '''
auto& destPred = PDest;'''
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());
ArmISA::VecPredRegContainer tmpPredC;
auto tmpPred = tmpPredC.as<Element>();
for (unsigned i = 0; i < eCount; ++i)
tmpPred[i] = GpOp_x[i];
destPred.reset();
for (unsigned i = 0; i < eCount; i++) {
const Element& srcElem1 = AA64FpOp1_x[i];
%(src_elem_2_ty)s srcElem2 __attribute__((unused)) =
%(src_elem_2)s;
bool destElem = false;
if (tmpPred[i]) {
%(op)s
} else {
destElem = false;
}
PDest_x[i] = destElem;
}''' % {'op': op,
'src_elem_2_ty': 'Element' if isImm else 'const Element&',
'src_elem_2': 'imm' if isImm else 'AA64FpOp2_x[i]'}
iop = InstObjParams(name, 'Sve' + Name,
'SveCmpImmOp' if isImm else 'SveCmpOp',
{'code': code, 'op_class': opClass}, [])
iop.snippets['code'] = extraPrologCode + iop.snippets['code']
if isImm:
header_output += SveCmpImmOpDeclare.subst(iop)
else:
header_output += SveCmpOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in types:
substDict = {'targs' : type,
'class_name' : 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generates definitions for ternary SVE intructions (always predicated -
# merging)
def sveTerInst(name, Name, opClass, types, op, decoder='Generic'):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());
for (unsigned i = 0; i < eCount; i++) {
const Element& srcElem1 = AA64FpOp1_x[i];
const Element& srcElem2 = AA64FpOp2_x[i];
Element destElem = AA64FpDestMerge_x[i];
if (GpOp_x[i]) {
%(op)s
}
AA64FpDest_x[i] = destElem;
}''' % {'op': op}
iop = InstObjParams(name, 'Sve' + Name, 'SveTerPredOp',
{'code': code, 'op_class': opClass}, [])
header_output += SveTerPredOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in types:
substDict = {'targs' : type,
'class_name' : 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generates definitions for ternary SVE instructions with indexed operand
def sveTerIdxInst(name, Name, opClass, types, op, decoder='Generic'):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());
// Number of elements in a 128 bit segment
constexpr unsigned ePerSegment = 128 / sizeof(Element);
for (unsigned i = 0; i < eCount; i++) {
const auto segmentBase = i - i % ePerSegment;
const auto segmentIdx = segmentBase + index;
const Element& srcElem1 = AA64FpOp1_x[i];
const Element& srcElem2 = AA64FpOp2_x[segmentIdx];
Element destElem = AA64FpDestMerge_x[i];
'''
code += '''
%(op)s
AA64FpDest_x[i] = destElem;
}''' % {'op': op}
iop = InstObjParams(name, 'Sve' + Name, 'SveBinIdxUnpredOp',
{'code': code, 'op_class': opClass}, [])
header_output += SveBinIdxUnpredOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in types:
substDict = {'targs' : type,
'class_name' : 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generates definitions for ternary SVE intructions with immediate operand
# (always unpredicated)
def sveTerImmInst(name, Name, opClass, types, op, decoder='Generic'):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());
for (unsigned i = 0; i < eCount; i++) {
const Element& srcElem2 = AA64FpOp2_x[i];
Element srcElem3 = imm;
Element destElem = AA64FpDestMerge_x[i];
%(op)s
AA64FpDest_x[i] = destElem;
}''' % {'op': op}
iop = InstObjParams(name, 'Sve' + Name, 'SveTerImmUnpredOp',
{'code': code, 'op_class': opClass}, [])
header_output += SveTerImmUnpredOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in types:
substDict = {'targs' : type,
'class_name' : 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generates definitions for PTRUE and PTRUES instructions.
def svePtrueInst(name, Name, opClass, types, isFlagSetting=False,
decoder='Generic'):
global header_output, exec_output, decoders
extraPrologCode = '''
auto& destPred = PDest;'''
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());
unsigned predCount = sveDecodePredCount(imm, eCount);
destPred.reset();
for (unsigned i = 0; i < eCount; i++) {
PDest_x[i] = (i < predCount);
}'''
if isFlagSetting:
code += '''
CondCodesNZ = (destPred.firstActive(destPred, eCount) << 1) |
destPred.noneActive(destPred, eCount);
CondCodesC = !destPred.lastActive(destPred, eCount);
CondCodesV = 0;'''
iop = InstObjParams(name, 'Sve' + Name, 'SvePtrueOp',
{'code': code, 'op_class': opClass}, [])
iop.snippets['code'] = extraPrologCode + iop.snippets['code']
header_output += SvePtrueOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in types:
substDict = {'targs' : type,
'class_name' : 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generate definitions for integer CMP<cc> instructions
def sveIntCmpInst(name, Name, opClass, types, op, wideop = False,
decoder = 'Generic'):
global header_output, exec_output, decoders
signed = 'int8_t' in types
srcType = 'Element'
op2Suffix = 'x'
if wideop:
srcType = 'int64_t' if signed else 'uint64_t'
op2Suffix = 'sd' if signed else 'ud'
extraPrologCode = '''
auto& destPred = PDest;'''
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());
ArmISA::VecPredRegContainer tmpPredC;
auto tmpPred = tmpPredC.as<Element>();
for (unsigned i = 0; i < eCount; ++i)
tmpPred[i] = GpOp_x[i];
destPred.reset();
for (unsigned i = 0; i < eCount; ++i) {
%(srcType)s srcElem1 = (%(srcType)s) AA64FpOp1_x[i];
%(srcType)s srcElem2 = AA64FpOp2_%(op2Suffix)s[%(op2Index)s];
bool destElem = false;
if (tmpPred[i]) {
%(op)s
}
PDest_x[i] = destElem;
}
CondCodesNZ = (destPred.firstActive(tmpPred, eCount) << 1) |
destPred.noneActive(tmpPred, eCount);
CondCodesC = !destPred.lastActive(tmpPred, eCount);
CondCodesV = 0;''' % {
'op': op,
'srcType': srcType,
'op2Suffix': op2Suffix,
'op2Index': '(i * sizeof(Element)) / 8' if wideop else 'i'
}
iop = InstObjParams(name, 'Sve' + Name, 'SveIntCmpOp',
{
'code': code,
'op_class': opClass,
'op2IsWide': 'true' if wideop else 'false',
}, [])
iop.snippets['code'] = extraPrologCode + iop.snippets['code']
header_output += SveIntCmpOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in types:
substDict = {'targs' : type,
'class_name' : 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generate definitions for integer CMP<cc> instructions (with immediate)
def sveIntCmpImmInst(name, Name, opClass, types, op, decoder = 'Generic'):
global header_output, exec_output, decoders
extraPrologCode = '''
auto& destPred = PDest;'''
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());
ArmISA::VecPredRegContainer tmpPredC;
auto tmpPred = tmpPredC.as<Element>();
for (unsigned i = 0; i < eCount; ++i)
tmpPred[i] = GpOp_x[i];
destPred.reset();
for (unsigned i = 0; i < eCount; ++i) {
Element srcElem1 = AA64FpOp1_x[i];
Element srcElem2 = static_cast<Element>(imm);
bool destElem = false;
if (tmpPred[i]) {
%(op)s
}
PDest_x[i] = destElem;
}
CondCodesNZ = (destPred.firstActive(tmpPred, eCount) << 1) |
destPred.noneActive(tmpPred, eCount);
CondCodesC = !destPred.lastActive(tmpPred, eCount);
CondCodesV = 0;'''%{'op': op}
iop = InstObjParams(name, 'Sve' + Name, 'SveIntCmpImmOp',
{'code': code, 'op_class': opClass,}, [])
iop.snippets['code'] = extraPrologCode + iop.snippets['code']
header_output += SveIntCmpImmOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in types:
substDict = {'targs' : type,
'class_name' : 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generate definitions for SVE element count instructions
def sveElemCountInst(name, Name, opClass, types, op,
destType = DestType.Scalar, dstIs32b = False,
dstAcc = True, decoder = 'Generic'):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());
unsigned count = sveDecodePredCount(pattern, eCount);
'''
if destType == DestType.Vector:
code += '''
for (unsigned i = 0; i < eCount; ++i) {
Element srcElem1 = AA64FpDestMerge_x[i];
Element destElem = 0;
%(op)s
AA64FpDest_x[i] = destElem;
}'''%{'op': op}
else:
if 'uint16_t' in types:
if dstIs32b:
dstType = 'uint32_t'
else:
dstType = 'uint64_t'
else:
if dstIs32b:
dstType = 'int32_t'
else:
dstType = 'int64_t'
if dstAcc:
code += '''
%(dstType)s srcElem1 = XDest;
'''%{'dstType': dstType}
code += '''
%(dstType)s destElem = 0;
%(op)s;
XDest = destElem;
'''%{'op': op, 'dstType': dstType}
iop = InstObjParams(name, 'Sve' + Name, 'SveElemCountOp',
{'code': code, 'op_class': opClass, 'dstIsVec': destType,
'dstIs32b': 'true' if dstIs32b else 'false'}, [])
header_output += SveElemCountOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in types:
substDict = {'targs' : type,
'class_name' : 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict);
def svePartBrkInst(name, Name, opClass, isFlagSetting, predType, whenBrk,
decoder = 'Generic'):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<uint8_t>(
xc->tcBase());
bool dobreak = false;
ArmISA::VecPredRegContainer tmpPredC;
auto auxGpOp = tmpPredC.as<uint8_t>();
for (unsigned i = 0; i < eCount; ++i) {
auxGpOp[i] = GpOp_ub[i];
}
for (unsigned i = 0; i < eCount; ++i) {
bool element = POp1_ub[i] == 1;
if (auxGpOp[i]) {'''
breakCode = '''
dobreak = dobreak || element;'''
if whenBrk == Break.Before:
code += breakCode
code += '''
PDest_ub[i] = !dobreak;'''
if whenBrk == Break.After:
code += breakCode
code += '''
}'''
if predType == PredType.ZERO:
code += ''' else {
PDest_ub[i] = 0;
}'''
elif predType == PredType.MERGE:
code += ''' else {
PDest_ub[i] = PDestMerge_ub[i];
}'''
code += '''
}'''
extraPrologCode = ''
if isFlagSetting:
code += '''
CondCodesNZ = (destPred.firstActive(auxGpOp, eCount) << 1) |
destPred.noneActive(auxGpOp, eCount);
CondCodesC = !destPred.lastActive(auxGpOp, eCount);
CondCodesV = 0;'''
extraPrologCode += '''
auto& destPred = PDest;'''
iop = InstObjParams(name, 'Sve' + Name, 'SvePartBrkOp',
{'code': code, 'op_class': opClass,
'isMerging': 'true' if predType == PredType.MERGE
else 'false'}, [])
iop.snippets['code'] = extraPrologCode + iop.snippets['code']
header_output += SvePartBrkOpDeclare.subst(iop)
exec_output += SveNonTemplatedOpExecute.subst(iop)
def svePartBrkPropPrevInst(name, Name, opClass, isFlagSetting, whenBrk,
decoder = 'Generic'):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<uint8_t>(
xc->tcBase());
bool last = POp1_ub.lastActive(GpOp_ub, eCount);
ArmISA::VecPredRegContainer tmpPredC;
auto auxGpOp = tmpPredC.as<uint8_t>();
for (unsigned i = 0; i < eCount; ++i) {
auxGpOp[i] = GpOp_ub[i];
}
for (unsigned i = 0; i < eCount; ++i) {
if (auxGpOp[i]) {'''
breakCode = '''
last = last && (POp2_ub[i] == 0);'''
if whenBrk == Break.Before:
code += breakCode
code += '''
PDest_ub[i] = last;'''
if whenBrk == Break.After:
code += breakCode
code += '''
} else {
PDest_ub[i] = 0;
}
}'''
extraPrologCode = ''
if isFlagSetting:
code += '''
CondCodesNZ = (destPred.firstActive(auxGpOp, eCount) << 1) |
destPred.noneActive(auxGpOp, eCount);
CondCodesC = !destPred.lastActive(auxGpOp, eCount);
CondCodesV = 0;'''
extraPrologCode += '''
auto& destPred = PDest;'''
iop = InstObjParams(name, 'Sve' + Name, 'SvePartBrkPropOp',
{'code': code, 'op_class': opClass}, [])
iop.snippets['code'] = extraPrologCode + iop.snippets['code']
header_output += SvePartBrkPropOpDeclare.subst(iop)
exec_output += SveNonTemplatedOpExecute.subst(iop)
def svePartBrkPropNextInst(name, Name, opClass, isFlagSetting,
decoder = 'Generic'):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<uint8_t>(
xc->tcBase());
bool last = POp1_ub.lastActive(GpOp_ub, eCount);
for (unsigned i = 0; i < eCount; i++) {
if (!last) {
PDest_ub[i] = 0;
} else {
PDest_ub[i] = PDestMerge_ub[i];
}
}'''
extraPrologCode = ''
if isFlagSetting:
code += '''
VecPredRegT<uint8_t, MaxSveVecLenInBytes, false, false>::Container c;
VecPredRegT<uint8_t, MaxSveVecLenInBytes, false, false> predOnes(c);
for (unsigned i = 0; i < eCount; i++) {
predOnes[i] = 1;
}
CondCodesNZ = (destPred.firstActive(predOnes, eCount) << 1) |
destPred.noneActive(predOnes, eCount);
CondCodesC = !destPred.lastActive(predOnes, eCount);
CondCodesV = 0;'''
extraPrologCode += '''
auto& destPred = PDest;'''
iop = InstObjParams(name, 'Sve' + Name, 'SvePartBrkPropOp',
{'code': code, 'op_class': opClass}, [])
iop.snippets['code'] = extraPrologCode + iop.snippets['code']
header_output += SvePartBrkPropOpDeclare.subst(iop)
exec_output += SveNonTemplatedOpExecute.subst(iop)
# Generate definitions for scalar select instructions
def sveSelectInst(name, Name, opClass, types, op, isCond,
destType = DstRegType.Scalar, decoder = 'Generic'):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());
int last;
for (last = eCount - 1; last >= 0; --last) {
if (GpOp_x[last]) {
break;
}
}
'''
if isCond:
code += '''
if (last >= 0) {'''
code += '''
Element destElem;
%(op)s'''%{'op': op}
if destType == DstRegType.Vector:
code += '''
for (unsigned i = 0; i < eCount; ++i)
AA64FpDest_x[i] = destElem;'''
elif destType == DstRegType.Scalar:
code += '''
XDest = destElem;'''
elif destType == DstRegType.SimdFpScalar:
code += '''
AA64FpDest_x[0] = destElem;'''
if isCond:
code += '''
}'''
if destType == DstRegType.Scalar:
code += ''' else {
XDest = (Element) XDest;
}'''
elif destType == DstRegType.Vector:
code += ''' else {
for (unsigned i = 0; i < eCount; ++i)
AA64FpDest_x[i] = AA64FpDestMerge_x[i];
}'''
elif destType == DstRegType.SimdFpScalar:
code += ''' else {
AA64FpDest_x[0] = AA64FpDestMerge_x[0];
}'''
iop = InstObjParams(name, 'Sve' + Name, 'SveSelectOp',
{'code': code, 'op_class': opClass,
'isCond': 'true' if isCond else 'false',
'isScalar': 'true'
if destType == DstRegType.Scalar else 'false',
'isSimdFp': 'true'
if destType == DstRegType.SimdFpScalar
else 'false'},
[])
header_output += SveSelectOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in types:
substDict = {'targs' : type,
'class_name' : 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generate definitions for PNEXT (find next active predicate)
# instructions
def svePNextInst(name, Name, opClass, types, decoder = 'Generic'):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());
ArmISA::VecPredRegContainer tmpPredC;
auto auxGpOp = tmpPredC.as<Element>();
for (unsigned i = 0; i < eCount; ++i) {
auxGpOp[i] = GpOp_x[i];
}
int last;
for (last = eCount - 1; last >= 0; --last) {
if (POp1_x[last]) {
break;
}
}
int next = last + 1;
while (next < eCount && GpOp_x[next] == 0) {
next++;
}
destPred.reset();
if (next < eCount) {
PDest_x[next] = 1;
}
CondCodesNZ = (destPred.firstActive(auxGpOp, eCount) << 1) |
destPred.noneActive(auxGpOp, eCount);
CondCodesC = !destPred.lastActive(auxGpOp, eCount);
CondCodesV = 0;'''
extraPrologCode = '''
auto& destPred = PDest;'''
iop = InstObjParams(name, 'Sve' + Name, 'SveUnaryPredPredOp',
{'code': code, 'op_class': opClass}, [])
iop.snippets['code'] = extraPrologCode + iop.snippets['code']
header_output += SveUnaryPredOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in types:
substDict = {'targs' : type,
'class_name' : 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generate definitions for PFIRST (set first active predicate)
# instructions
def svePFirstInst(name, Name, opClass, decoder = 'Generic'):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());
ArmISA::VecPredRegContainer tmpPredC;
auto auxGpOp = tmpPredC.as<Element>();
for (unsigned i = 0; i < eCount; ++i)
auxGpOp[i] = GpOp_x[i];
int first = -1;
for (int i = 0; i < eCount; ++i) {
if (auxGpOp[i] && first == -1) {
first = i;
}
}
for (int i = 0; i < eCount; ++i) {
PDest_x[i] = PDestMerge_x[i];
}
if (first >= 0) {
PDest_x[first] = 1;
}
CondCodesNZ = (destPred.firstActive(auxGpOp, eCount) << 1) |
destPred.noneActive(auxGpOp, eCount);
CondCodesC = !destPred.lastActive(auxGpOp, eCount);
CondCodesV = 0;'''
extraPrologCode = '''
auto& destPred = PDest;'''
iop = InstObjParams(name, 'Sve' + Name, 'SveUnaryPredPredOp',
{'code': code, 'op_class': opClass}, [])
iop.snippets['code'] = extraPrologCode + iop.snippets['code']
header_output += SveUnaryPredOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
substDict = {'targs' : 'uint8_t',
'class_name' : 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generate definitions for SVE TBL instructions
def sveTblInst(name, Name, opClass, decoder = 'Generic'):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());
for (int i = 0; i < eCount; ++i) {
Element idx = AA64FpOp2_x[i];
Element val;
if (idx < eCount) {
val = AA64FpOp1_x[idx];
} else {
val = 0;
}
AA64FpDest_x[i] = val;
}'''
iop = InstObjParams(name, 'Sve' + Name, 'SveTblOp',
{'code': code, 'op_class': opClass}, [])
header_output += SveBinUnpredOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in unsignedTypes:
substDict = {'targs' : type,
'class_name' : 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generate definitions for SVE Unpack instructions
def sveUnpackInst(name, Name, opClass, sdtypes, unpackHalf,
regType, decoder = 'Generic'):
global header_output, exec_output, decoders
extraPrologCode = '''
auto& destPred = PDest;'''
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<DElement>(
xc->tcBase());'''
if unpackHalf == Unpack.Low:
if regType == SrcRegType.Predicate:
code += '''
ArmISA::VecPredRegContainer tmpPredC;
auto auxPOp1 = tmpPredC.as<SElement>();
for (int i = 0; i < eCount; ++i) {
auxPOp1[i] = POp1_xs[i];
}'''
else:
code += '''
ArmISA::VecRegContainer tmpVecC;
auto auxOp1 = tmpVecC.as<SElement>();
for (int i = 0; i < eCount; ++i) {
auxOp1[i] = AA64FpOp1_xs[i];
}'''
code += '''
for (int i = 0; i < eCount; ++i) {'''
if regType == SrcRegType.Predicate:
if unpackHalf == Unpack.High:
code +='''
const SElement& srcElem1 = POp1_xs[i + eCount];'''
else:
code +='''
const SElement& srcElem1 = auxPOp1[i];'''
code += '''
destPred.set_raw(i, 0);
PDest_xd[i] = srcElem1;'''
else:
if unpackHalf == Unpack.High:
code +='''
const SElement& srcElem1 = AA64FpOp1_xs[i + eCount];'''
else:
code +='''
const SElement& srcElem1 = auxOp1[i];'''
code += '''
AA64FpDest_xd[i] = static_cast<DElement>(srcElem1);'''
code += '''
}
'''
iop = InstObjParams(name, 'Sve' + Name, 'SveUnpackOp',
{'code': code, 'op_class': opClass}, [])
if regType == SrcRegType.Predicate:
iop.snippets['code'] = extraPrologCode + iop.snippets['code']
header_output += SveUnpackOpDeclare.subst(iop)
exec_output += SveWideningOpExecute.subst(iop)
for srcType, dstType in sdtypes:
substDict = {'targs': srcType + ', ' + dstType,
'class_name': 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generate definition for SVE predicate test instructions
def svePredTestInst(name, Name, opClass, decoder = 'Generic'):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<uint8_t>(
xc->tcBase());
CondCodesNZ = (POp1_ub.firstActive(GpOp_ub, eCount) << 1) |
POp1_ub.noneActive(GpOp_ub, eCount);
CondCodesC = !POp1_ub.lastActive(GpOp_ub, eCount);
CondCodesV = 0;'''
iop = InstObjParams(name, 'Sve' + Name, 'SvePredTestOp',
{'code': code, 'op_class': opClass}, [])
header_output += SvePredicateTestOpDeclare.subst(iop)
exec_output += SveNonTemplatedOpExecute.subst(iop)
# Generate definition for SVE predicate compact operations
def sveCompactInst(name, Name, opClass, types, decoder = 'Generic'):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());
ArmISA::VecRegContainer tmpVecC;
auto auxOp1 = tmpVecC.as<Element>();
for (unsigned i = 0; i < eCount; ++i) {
auxOp1[i] = AA64FpOp1_x[i];
}
unsigned x = 0;
for (unsigned i = 0; i < eCount; ++i) {
AA64FpDest_x[i] = 0;
if (GpOp_x[i]) {
AA64FpDest_x[x] = auxOp1[i];
x++;
}
}'''
iop = InstObjParams(name, 'Sve' + Name, 'SveUnaryPredOp',
{'code': code, 'op_class': opClass}, [])
header_output += SveUnaryPredOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in types:
substDict = {'targs': type, 'class_name': 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generate definition for unary SVE predicate instructions with implicit
# source operand (PFALSE, RDFFR(S))
def svePredUnaryWImplicitSrcInst(name, Name, opClass, op,
predType=PredType.NONE, isFlagSetting=False, decoder='Generic'):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + op
if isFlagSetting:
code += '''
CondCodesNZ = (destPred.firstActive(GpOp, eCount) << 1) |
destPred.noneActive(GpOp, eCount);
CondCodesC = !destPred.lastActive(GpOp, eCount);
CondCodesV = 0;'''
extraPrologCode = '''
auto& destPred M5_VAR_USED = PDest;'''
baseClass = ('SvePredUnaryWImplicitSrcOp' if predType == PredType.NONE
else 'SvePredUnaryWImplicitSrcPredOp')
iop = InstObjParams(name, 'Sve' + Name, baseClass,
{'code': code, 'op_class': opClass}, [])
iop.snippets['code'] = extraPrologCode + iop.snippets['code']
if predType == PredType.NONE:
header_output += SvePredUnaryOpWImplicitSrcDeclare.subst(iop)
else:
header_output += SvePredUnaryPredOpWImplicitSrcDeclare.subst(iop)
exec_output += SveNonTemplatedOpExecute.subst(iop)
# Generate definition for SVE instructions writing to the FFR (SETFFR,
# WRFFR)
def svePredWriteFfrInst(name, Name, opClass, op, isSetFfr,
decoder='Generic'):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + op
extraPrologCode = '''
auto& destPred M5_VAR_USED = Ffr;'''
baseClass = ('SveWImplicitSrcDstOp' if isSetFfr
else 'SvePredUnaryWImplicitDstOp')
iop = InstObjParams(name, 'Sve' + Name, baseClass,
{'code': code, 'op_class': opClass}, [])
iop.snippets['code'] = extraPrologCode + iop.snippets['code']
if isSetFfr:
header_output += SveOpWImplicitSrcDstDeclare.subst(iop)
else:
header_output += SvePredUnaryOpWImplicitDstDeclare.subst(iop)
exec_output += SveNonTemplatedOpExecute.subst(iop)
# Generate definition for SVE Ext instruction
def sveExtInst(name, Name, opClass, decoder = 'Generic'):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());
ArmISA::VecRegContainer tmpVecC;
auto auxOp1 = tmpVecC.as<Element>();
for (unsigned i = 0; i < eCount; ++i) {
auxOp1[i] = AA64FpOp1_x[i];
}
uint64_t pos = imm;
if (pos >= eCount)
pos = 0;
for (int i = 0; i < eCount; ++i, ++pos)
{
if (pos < eCount)
AA64FpDest_x[i] = AA64FpDestMerge_x[pos];
else
AA64FpDest_x[i] = auxOp1[pos-eCount];
}
'''
iop = InstObjParams(name, 'Sve' + Name, 'SveBinImmUnpredDestrOp',
{'code': code, 'op_class': opClass}, [])
header_output += SveBinImmUnpredOpDeclare.subst(iop);
exec_output += SveOpExecute.subst(iop)
substDict = {'targs': 'uint8_t', 'class_name': 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generate definition for SVE Slice instruction
def sveSpliceInst(name, Name, opClass, types, decoder = 'Generic'):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());
ArmISA::VecRegContainer tmpVecC;
auto auxDest = tmpVecC.as<Element>();
int firstelem = -1, lastelem = -2;
for (int i = 0; i < eCount; ++i) {
if (GpOp_x[i]) {
lastelem = i;
if (firstelem < 0)
firstelem = i;
}
}
int x = 0;
for (int i = firstelem; i <= lastelem; ++i, ++x) {
auxDest[x] = AA64FpDestMerge_x[i];
}
int remaining = eCount - x;
for (int i = 0; i < remaining; ++i, ++x) {
auxDest[x] = AA64FpOp2_x[i];
}
for (int i = 0; i < eCount; ++i) {
AA64FpDest_x[i] = auxDest[i];
}
'''
iop = InstObjParams(name, 'Sve' + Name, 'SveBinDestrPredOp',
{'code': code, 'op_class': opClass}, [])
header_output += SveBinDestrPredOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in types:
substDict = {'targs': type, 'class_name': 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generate definition for SVE DUP (index) instruction
def sveDupIndexInst(name, Name, opClass, types, decoder = 'Generic'):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());
Element srcElem1 = 0;
if (imm < eCount) {
srcElem1 = AA64FpOp1_x[imm];
}
for (int i = 0; i < eCount; ++i) {
AA64FpDest_x[i] = srcElem1;
}'''
iop = InstObjParams(name, 'Sve' + Name, 'SveBinImmIdxUnpredOp',
{'code': code, 'op_class': opClass}, [])
header_output += SveBinImmUnpredOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in types:
substDict = {'targs': type, 'class_name': 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generate definition for SVE reverse elements instructions
def sveReverseElementsInst(name, Name, opClass, types,
srcType = SrcRegType.Vector, decoder = 'Generic'):
assert srcType in (SrcRegType.Vector, SrcRegType.Predicate)
global header_output, exec_output, decoders
extraPrologCode = '''
auto& destPred = PDest;'''
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());'''
if srcType == SrcRegType.Predicate:
code += '''
ArmISA::VecPredRegContainer tmpPredC;
auto auxPOp1 = tmpPredC.as<Element>();
for (unsigned i = 0; i < eCount; ++i) {
uint8_t v = POp1_x.get_raw(i);
auxPOp1.set_raw(i, v);
}
PDest_x[0] = 0;'''
else:
code += '''
ArmISA::VecRegContainer tmpRegC;
auto auxOp1 = tmpRegC.as<Element>();
for (unsigned i = 0; i < eCount; ++i) {
auxOp1[i] = AA64FpOp1_x[i];
}'''
code += '''
for (int i = 0; i < eCount; ++i) {'''
if srcType == SrcRegType.Vector:
code += '''
AA64FpDest_x[i] = auxOp1[eCount - i - 1];'''
else:
code += '''
destPred.set_raw(i, auxPOp1.get_raw(eCount - i - 1));'''
code += '''
}'''
iop = InstObjParams(name, 'Sve' + Name, 'SveUnaryUnpredOp',
{'code': code, 'op_class': opClass}, [])
if srcType == SrcRegType.Predicate:
iop.snippets['code'] = extraPrologCode + iop.snippets['code']
header_output += SveUnaryUnpredOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in types:
substDict = {'targs': type, 'class_name': 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generate definition for shift & insert instructions
def sveShiftAndInsertInst(name, Name, opClass, types,
srcType = SrcRegType.Scalar, decoder = 'Generic'):
assert srcType in (SrcRegType.SimdFpScalar, SrcRegType.Scalar)
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());'''
if srcType == SrcRegType.Scalar:
code += '''
auto& srcElem1 = XOp1;'''
elif srcType == SrcRegType.SimdFpScalar:
code += '''
auto& srcElem1 = AA64FpOp1_x[0];'''
code += '''
for (int i = eCount - 1; i > 0; --i) {
AA64FpDest_x[i] = AA64FpDestMerge_x[i-1];
}
AA64FpDest_x[0] = srcElem1;'''
iop = InstObjParams(name, 'Sve' + Name, 'SveUnarySca2VecUnpredOp',
{'code': code, 'op_class': opClass,
'isSimdFp': 'true' if srcType == SrcRegType.SimdFpScalar
else 'false'}, [])
header_output += SveShiftAndInsertOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in types:
substDict = {'targs': type, 'class_name': 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generate definition for DOT instructions
def sveDotInst(name, Name, opClass, types, isIndexed = True):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());
for (int i = 0; i < eCount; ++i) {'''
if isIndexed:
code += '''
int segbase = i - i % (16 / sizeof(Element));
int s = segbase + imm;'''
code += '''
DElement res = AA64FpDest_xd[i];
DElement srcElem1, srcElem2;
for (int j = 0; j <= 3; ++j) {
srcElem1 = static_cast<DElement>(AA64FpOp1_xs[4 * i + j]);'''
if isIndexed:
code += '''
srcElem2 = static_cast<DElement>(AA64FpOp2_xs[4 * s + j]);'''
else:
code += '''
srcElem2 = static_cast<DElement>(AA64FpOp2_xs[4 * i + j]);'''
code += '''
res += srcElem1 * srcElem2;
}
AA64FpDestMerge_xd[i] = res;
}'''
iop = InstObjParams(name, 'Sve' + Name,
'SveDotProdIdxOp' if isIndexed else
'SveDotProdOp',
{'code': code, 'op_class': opClass}, [])
if isIndexed:
header_output += SveWideningTerImmOpDeclare.subst(iop)
else:
header_output += SveWideningTerOpDeclare.subst(iop)
exec_output += SveWideningOpExecute.subst(iop)
for type in types:
substDict = {'targs': type, 'class_name': 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generate definition for ordered reduction
def sveOrderedReduction(name, Name, opClass, types, op,
decoder = 'Generic'):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());'''
code += '''
Element destElem = AA64FpDestMerge_x[0];
for (int i = 0; i < eCount; ++i) {
if (GpOp_x[i]) {
Element srcElem1 = AA64FpOp1_x[i];
%(op)s
}
}
for (int i = 1; i < eCount; ++i) {
AA64FpDest_x[i] = 0;
}
AA64FpDest_x[0] = destElem;'''%{'op': op}
iop = InstObjParams(name, 'Sve' + Name, 'SveOrdReducOp',
{'code': code, 'op_class': opClass}, [])
header_output += SveReducOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in types:
substDict = {'targs' : type,
'class_name' : 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generate definitions for complex addition instructions
def sveComplexAddInst(name, Name, opClass, types,
decoder = 'Generic'):
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());'''
code += '''
bool sub_i = (rot == 1);
bool sub_r = (rot == 3);
for (int i = 0; i < eCount / 2; ++i) {
Element acc_r = AA64FpOp1_x[2 * i];
Element acc_i = AA64FpOp1_x[2 * i + 1];
Element elt2_r = AA64FpOp2_x[2 * i];
Element elt2_i = AA64FpOp2_x[2 * i + 1];
FPSCR fpscr;
if (GpOp_x[2 * i]) {
if (sub_i) {
elt2_i = fplibNeg<Element>(elt2_i);
}
fpscr = (FPSCR) FpscrExc;
acc_r = fplibAdd<Element>(acc_r, elt2_i, fpscr);
FpscrExc = fpscr;
}
if (GpOp_x[2 * i + 1]) {
if (sub_r) {
elt2_r = fplibNeg<Element>(elt2_r);
}
fpscr = (FPSCR) FpscrExc;
acc_i = fplibAdd<Element>(acc_i, elt2_r, fpscr);
FpscrExc = fpscr;
}
AA64FpDest_x[2 * i] = acc_r;
AA64FpDest_x[2 * i + 1] = acc_i;
}
'''
iop = InstObjParams(name, 'Sve' + Name, 'SveComplexOp',
{'code': code, 'op_class': opClass}, [])
header_output += SveComplexOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in types:
substDict = {'targs' : type,
'class_name' : 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
# Generate definitions for complex multiply and accumulate instructions
def sveComplexMulAddInst(name, Name, opClass, types,
predType=PredType.NONE, decoder='Generic'):
assert predType in (PredType.NONE, PredType.MERGE)
global header_output, exec_output, decoders
code = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<Element>(
xc->tcBase());'''
code += '''
uint32_t sel_a = rot & 0x1;
uint32_t sel_b = sel_a ? 0 : 1;
bool neg_i = (rot & 0x2) == 1;
bool neg_r = (rot & 0x1) != (rot & 0x2);'''
if predType == PredType.NONE:
code += '''
uint32_t eltspersegment = 16 / (2 * sizeof(Element));'''
code += '''
for (int i = 0; i < eCount / 2; ++i) {'''
if predType == PredType.NONE:
code += '''
uint32_t segmentbase = i - (i % eltspersegment);
uint32_t s = segmentbase + imm;'''
else:
code += '''
uint32_t s = i;'''
code += '''
Element addend_r = AA64FpDestMerge_x[2 * i];
Element addend_i = AA64FpDestMerge_x[2 * i + 1];
Element elt1_a = AA64FpOp1_x[2 * i + sel_a];
Element elt2_a = AA64FpOp2_x[2 * s + sel_a];
Element elt2_b = AA64FpOp2_x[2 * s + sel_b];
FPSCR fpscr;
'''
if predType != PredType.NONE:
code += '''
if (GpOp_x[2 * i]) {'''
code += '''
if (neg_r) {
elt2_a = fplibNeg<Element>(elt2_a);
}
fpscr = (FPSCR) FpscrExc;
addend_r = fplibMulAdd<Element>(addend_r, elt1_a, elt2_a, fpscr);
FpscrExc = fpscr;'''
if predType != PredType.NONE:
code += '''
}'''
if predType != PredType.NONE:
code += '''
if (GpOp_x[2 * i + 1]) {'''
code += '''
if (neg_i) {
elt2_b = fplibNeg<Element>(elt2_b);
}
fpscr = (FPSCR) FpscrExc;
addend_i = fplibMulAdd<Element>(addend_i, elt1_a, elt2_b, fpscr);
FpscrExc = fpscr;'''
if predType != PredType.NONE:
code += '''
}'''
code += '''
AA64FpDest_x[2 * i] = addend_r;
AA64FpDest_x[2 * i + 1] = addend_i;
}'''
iop = InstObjParams(name, 'Sve' + Name,
'SveComplexIdxOp' if predType == PredType.NONE
else 'SveComplexOp',
{'code': code, 'op_class': opClass}, [])
if predType == PredType.NONE:
header_output += SveComplexIndexOpDeclare.subst(iop)
else:
header_output += SveComplexOpDeclare.subst(iop)
exec_output += SveOpExecute.subst(iop)
for type in types:
substDict = {'targs' : type,
'class_name' : 'Sve' + Name}
exec_output += SveOpExecDeclare.subst(substDict)
fpTypes = ('uint16_t', 'uint32_t', 'uint64_t')
signedTypes = ('int8_t', 'int16_t', 'int32_t', 'int64_t')
unsignedTypes = ('uint8_t', 'uint16_t', 'uint32_t', 'uint64_t')
smallSignedTypes = ('int8_t', 'int16_t', 'int32_t')
bigSignedTypes = ('int16_t', 'int32_t', 'int64_t')
smallUnsignedTypes = ('uint8_t', 'uint16_t', 'uint32_t')
bigUnsignedTypes = ('uint16_t', 'uint32_t', 'uint64_t')
unsignedWideSDTypes = (('uint8_t', 'uint16_t'),
('uint16_t', 'uint32_t'), ('uint32_t', 'uint64_t'))
signedWideSDTypes = (('int8_t', 'int16_t'),
('int16_t', 'int32_t'), ('int32_t', 'int64_t'))
# ABS
absCode = 'destElem = (Element) std::abs(srcElem1);'
sveUnaryInst('abs', 'Abs', 'SimdAluOp', signedTypes, absCode,
PredType.MERGE)
# ADD (immediate)
sveWideImmInst('add', 'AddImm', 'SimdAddOp', unsignedTypes, addCode, False)
# ADD (vectors, predicated)
addCode = 'destElem = srcElem1 + srcElem2;'
sveBinInst('add', 'AddPred', 'SimdAddOp', unsignedTypes, addCode,
PredType.MERGE, True)
# ADD (vectors, unpredicated)
addCode = 'destElem = srcElem1 + srcElem2;'
sveBinInst('add', 'AddUnpred', 'SimdAddOp', unsignedTypes, addCode)
# ADDPL
addvlCode = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<uint%d_t>(
xc->tcBase());
XDest = XOp1 + eCount * (int64_t) imm;
'''
buildXImmDataInst('addpl', addvlCode % 64, buildCc=False)
# ADDVL
buildXImmDataInst('addvl', addvlCode % 8, buildCc=False)
# ADR
adrCode = '''
if (offsetFormat == SveAdrOffsetUnpackedSigned) {
srcElem2 = sext<32>(srcElem2 & mask(32));
} else if (offsetFormat == SveAdrOffsetUnpackedUnsigned) {
srcElem2 = srcElem2 & mask(32);
}
destElem = srcElem1 + srcElem2 * mult;
'''
sveAdrInst('adr', 'Adr', 'SimdAddOp', ('uint32_t', 'uint64_t'), adrCode)
# AND (immediate)
andCode = 'destElem = srcElem1 & srcElem2;'
sveWideImmInst('and', 'AndImm', 'SimdAluOp', ('uint64_t',), andCode)
# AND (vectors, predicated)
sveBinInst('and', 'AndPred', 'SimdAluOp', unsignedTypes, andCode,
PredType.MERGE, True)
# AND (vectors, unpredicated)
andCode = 'destElem = srcElem1 & srcElem2;'
sveBinInst('and', 'AndUnpred', 'SimdAluOp', ('uint64_t',), andCode)
# AND, ANDS (predicates)
svePredLogicalInst('and', 'PredAnd', 'SimdPredAluOp', ('uint8_t',),
andCode)
svePredLogicalInst('ands', 'PredAnds', 'SimdPredAluOp', ('uint8_t',),
andCode, isFlagSetting=True)
# ANDV
andvCode = 'destElem &= srcElem1;'
sveAssocReducInst('andv', 'Andv', 'SimdReduceAluOp', unsignedTypes,
andvCode, 'std::numeric_limits<Element>::max()')
# ASR (immediate, predicated)
asrCode = '''
int sign_bit = bits(srcElem1, sizeof(Element) * 8 - 1);
if (srcElem2 == 0) {
destElem = srcElem1;
} else if (srcElem2 >= sizeof(Element) * 8) {
destElem = sign_bit ? std::numeric_limits<Element>::max() : 0;
} else {
destElem = srcElem1 >> srcElem2;
if (sign_bit) {
destElem |= ~mask(sizeof(Element) * 8 - srcElem2);
}
}
'''
sveBinImmInst('asr', 'AsrImmPred', 'SimdAluOp', unsignedTypes, asrCode,
PredType.MERGE)
# ASR (immediate, unpredicated)
sveBinImmInst('asr', 'AsrImmUnpred', 'SimdAluOp', unsignedTypes, asrCode)
# ASR (vectors)
sveBinInst('asr', 'AsrPred', 'SimdAluOp', unsignedTypes, asrCode,
PredType.MERGE, True)
# ASR (wide elements, predicated)
sveShiftByWideElemsInst('asr', 'AsrWidePred', 'SimdAluOp', unsignedTypes,
asrCode, PredType.MERGE)
# ASR (wide elements, unpredicated)
sveShiftByWideElemsInst('asr', 'AsrWideUnpred', 'SimdAluOp', unsignedTypes,
asrCode)
# ASRD
asrdCode = '''
Element element1 = srcElem1;
Element shift = srcElem2;
if (srcElem1 < 0) {
Element tmp = ((1L << shift) - 1L);
if (tmp == -1L) {
element1 = 0;
} else {
element1 = element1 + tmp;
}
}
destElem = (element1 >> shift);
'''
sveBinImmInst('asrd', 'Asrd', 'SimdAluOp', signedTypes, asrdCode,
PredType.MERGE)
# ASRR
asrrCode = '''
int sign_bit = bits(srcElem2, sizeof(Element) * 8 - 1);
if (srcElem1 == 0) {
destElem = srcElem2;
} else if (srcElem1 >= sizeof(Element) * 8) {
destElem = sign_bit ? std::numeric_limits<Element>::max() : 0;
} else {
destElem = srcElem2 >> srcElem1;
if (sign_bit) {
destElem |= ~mask(sizeof(Element) * 8 - srcElem1);
}
}
'''
sveBinInst('asrr', 'Asrr', 'SimdAluOp', unsignedTypes, asrrCode,
PredType.MERGE, True)
# BIC (vectors, predicated)
bicCode = 'destElem = srcElem1 & ~srcElem2;'
sveBinInst('bic', 'BicPred', 'SimdAluOp', unsignedTypes, bicCode,
PredType.MERGE, True)
# BIC (vectors, unpredicated)
sveBinInst('bic', 'BicUnpred', 'SimdAluOp', unsignedTypes, bicCode)
# BIC, BICS (predicates)
bicCode = 'destElem = srcElem1 && !srcElem2;'
svePredLogicalInst('bic', 'PredBic', 'SimdPredAluOp', ('uint8_t',),
bicCode)
svePredLogicalInst('bics', 'PredBics', 'SimdPredAluOp', ('uint8_t',),
bicCode, isFlagSetting=True)
# BRKA (merging)
svePartBrkInst('brka', 'Brkam', 'SimdPredAluOp', isFlagSetting = False,
predType = PredType.MERGE, whenBrk = Break.After)
# BRKA (zeroing)
svePartBrkInst('brka', 'Brkaz', 'SimdPredAluOp', isFlagSetting = False,
predType = PredType.ZERO, whenBrk = Break.After)
# BRKAS
svePartBrkInst('brkas', 'Brkas', 'SimdPredAluOp', isFlagSetting = True,
predType = PredType.ZERO, whenBrk = Break.After)
# BRKB (merging)
svePartBrkInst('brkb', 'Brkbm', 'SimdPredAluOp', isFlagSetting = False,
predType = PredType.MERGE, whenBrk = Break.Before)
# BRKB (zeroging)
svePartBrkInst('brkb', 'Brkbz', 'SimdPredAluOp', isFlagSetting = False,
predType = PredType.ZERO, whenBrk = Break.Before)
# BRKBS
svePartBrkInst('brkbs', 'Brkbs', 'SimdPredAluOp', isFlagSetting = True,
predType = PredType.ZERO, whenBrk = Break.Before)
# BRKN
svePartBrkPropNextInst('brkn', 'Brkn', 'SimdPredAluOp',
isFlagSetting = False)
# BRKNS
svePartBrkPropNextInst('brkns', 'Brkns', 'SimdPredAluOp',
isFlagSetting = True)
# BRKPA
svePartBrkPropPrevInst('brkpa', 'Brkpa', 'SimdPredAluOp',
isFlagSetting = False, whenBrk = Break.After)
# BRKPAS
svePartBrkPropPrevInst('brkpas', 'Brkpas', 'SimdPredAluOp',
isFlagSetting = True, whenBrk = Break.After)
# BRKPB
svePartBrkPropPrevInst('brkpb', 'Brkpb', 'SimdPredAluOp',
isFlagSetting = False, whenBrk = Break.Before)
# BRKPBS
svePartBrkPropPrevInst('brkpbs', 'Brkpbs', 'SimdPredAluOp',
isFlagSetting = True, whenBrk = Break.Before)
# CLASTA (scalar)
clastaCode = '''
last++;
if (last >= eCount)
last = 0;
destElem = AA64FpOp1_x[last];'''
sveSelectInst('clasta', 'Clasta', 'SimdAluOp', unsignedTypes, clastaCode,
isCond = True, destType = DstRegType.Scalar)
# CLASTA (SIMD&FP scalar)
sveSelectInst('clasta', 'Clastaf', 'SimdAluOp', unsignedTypes, clastaCode,
isCond = True, destType = DstRegType.SimdFpScalar)
# CLASTA (vector)
sveSelectInst('clasta', 'Clastav', 'SimdAluOp', unsignedTypes, clastaCode,
isCond = True, destType = DstRegType.Vector)
# CLASTB (scalar)
clastbCode = '''
destElem = AA64FpOp1_x[last];'''
sveSelectInst('clastb', 'Clastb', 'SimdAluOp', unsignedTypes, clastbCode,
isCond = True, destType = DstRegType.Scalar)
# CLASTB (SIMD&FP scalar)
sveSelectInst('clastb', 'Clastbf', 'SimdAluOp', unsignedTypes, clastbCode,
isCond = True, destType = DstRegType.SimdFpScalar)
# CLASTB (vectors)
sveSelectInst('clastb', 'Clastbv', 'SimdAluOp', unsignedTypes, clastbCode,
isCond = True, destType = DstRegType.Vector)
# CLS
clsCode = '''
destElem = 0;
Element val = srcElem1;
if (val < 0) {
val <<= 1;
while (val < 0) {
destElem++;
val <<= 1;
}
} else {
val <<= 1;
while (val >= 0 && destElem < sizeof(Element) * 8 - 1) {
destElem++;
val <<= 1;
}
}
'''
sveUnaryInst('cls', 'Cls', 'SimdAluOp', signedTypes, clsCode,
PredType.MERGE)
# CLZ
clzCode = '''
destElem = 0;
Element val = srcElem1;
while (val >= 0 && destElem < sizeof(Element) * 8) {
destElem++;
val <<= 1;
}
'''
sveUnaryInst('clz', 'Clz', 'SimdAluOp', signedTypes, clzCode,
PredType.MERGE)
# CMPEQ (immediate)
cmpeqCode = '''
destElem = (srcElem1 == srcElem2);
'''
sveIntCmpImmInst('cmpeq', 'Cmpeqi', 'SimdCmpOp', unsignedTypes, cmpeqCode)
# CMPEQ (vectors)
sveIntCmpInst('cmpeq', 'Cmpeq', 'SimdCmpOp', unsignedTypes, cmpeqCode)
# CMPEQ (wide elements)
sveIntCmpInst('cmpeq', 'Cmpeqw', 'SimdCmpOp', smallUnsignedTypes,
cmpeqCode, True)
# CMPGE (immediate)
cmpgeCode = '''
destElem = (srcElem1 >= srcElem2);
'''
sveIntCmpImmInst('cmpge', 'Cmpgei', 'SimdCmpOp', signedTypes, cmpgeCode)
# CMPGE (vectors)
sveIntCmpInst('cmpge', 'Cmpge', 'SimdCmpOp', signedTypes, cmpgeCode)
# CMPGE (wide elements)
sveIntCmpInst('cmpge', 'Cmpgew', 'SimdCmpOp', smallSignedTypes,
cmpgeCode, True)
# CMPGT (immediate)
cmpgtCode = '''
destElem = (srcElem1 > srcElem2);
'''
sveIntCmpImmInst('cmpge', 'Cmpgti', 'SimdCmpOp', signedTypes, cmpgtCode)
# CMPGT (vectors)
sveIntCmpInst('cmpge', 'Cmpgt', 'SimdCmpOp', signedTypes, cmpgtCode)
# CMPGT (wide elements)
sveIntCmpInst('cmpge', 'Cmpgtw', 'SimdCmpOp', smallSignedTypes,
cmpgtCode, True)
# CMPHI (immediate)
sveIntCmpImmInst('cmphi', 'Cmphii', 'SimdCmpOp', unsignedTypes, cmpgtCode)
# CMPHI (vectors)
sveIntCmpInst('cmphi', 'Cmphi', 'SimdCmpOp', unsignedTypes, cmpgtCode)
# CMPHI (wide elements)
sveIntCmpInst('cmphi', 'Cmphiw', 'SimdCmpOp', smallUnsignedTypes,
cmpgtCode, True)
# CMPHS (immediate)
sveIntCmpImmInst('cmphs', 'Cmphsi', 'SimdCmpOp', unsignedTypes, cmpgeCode)
# CMPHS (vectors)
sveIntCmpInst('cmphs', 'Cmphs', 'SimdCmpOp', unsignedTypes, cmpgeCode)
# CMPHS (wide elements)
sveIntCmpInst('cmphs', 'Cmphsw', 'SimdCmpOp', smallUnsignedTypes,
cmpgeCode, True)
# CMPLE (immediate)
cmpleCode = '''
destElem = (srcElem1 <= srcElem2);
'''
sveIntCmpImmInst('cmple', 'Cmplei', 'SimdCmpOp', signedTypes, cmpleCode)
# CMPLE (wide elements)
sveIntCmpInst('cmple', 'Cmplew', 'SimdCmpOp', smallSignedTypes,
cmpleCode, True)
# CMPLO (immediate)
cmpltCode = '''
destElem = (srcElem1 < srcElem2);
'''
sveIntCmpImmInst('cmplo', 'Cmploi', 'SimdCmpOp', unsignedTypes, cmpltCode)
# CMPLO (wide elements)
sveIntCmpInst('cmplo', 'Cmplow', 'SimdCmpOp', smallUnsignedTypes,
cmpltCode, True)
# CMPLS (immediate)
sveIntCmpImmInst('cmpls', 'Cmplsi', 'SimdCmpOp', unsignedTypes, cmpleCode)
# CMPLS (wide elements)
sveIntCmpInst('cmpls', 'Cmplsw', 'SimdCmpOp', smallUnsignedTypes,
cmpleCode, True)
# CMPLT (immediate)
sveIntCmpImmInst('cmplt', 'Cmplti', 'SimdCmpOp', signedTypes, cmpltCode)
# CMPLT (wide elements)
sveIntCmpInst('cmplt', 'Cmpltw', 'SimdCmpOp', smallSignedTypes,
cmpltCode, True)
# CMPNE (immediate)
cmpneCode = '''
destElem = (srcElem1 != srcElem2);
'''
sveIntCmpImmInst('cmpeq', 'Cmpnei', 'SimdCmpOp', unsignedTypes, cmpneCode)
# CMPNE (vectors)
sveIntCmpInst('cmpeq', 'Cmpne', 'SimdCmpOp', unsignedTypes, cmpneCode)
# CMPNE (wide elements)
sveIntCmpInst('cmpeq', 'Cmpnew', 'SimdCmpOp', smallUnsignedTypes,
cmpneCode, True)
# CNOT
cnotCode = '''
destElem = srcElem1?0:1;
'''
sveUnaryInst('cnot', 'Cnot', 'SimdAluOp', unsignedTypes, cnotCode,
PredType.MERGE)
# CNT
cntCode = '''
destElem = 0;
Element val = srcElem1;
while (val) {
destElem += val & 0x1;
val >>= 1;
}
'''
sveUnaryInst('cnt', 'Cnt', 'SimdAluOp', unsignedTypes, cntCode,
PredType.MERGE)
# CNTB, CNTD, CNTH, CNTW
cntxCode = '''
destElem = (count * imm);
'''
sveElemCountInst('cnt', 'Cntx', 'SimdAluOp', unsignedTypes, cntxCode,
destType = DestType.Scalar, dstIs32b = False, dstAcc = False)
# COMPACT
sveCompactInst('compact', 'Compact', 'SimdPredAluOp',
('uint32_t', 'uint64_t'))
# CPY (immediate)
dupCode = 'destElem = srcElem1;'
sveWideImmInst('cpy', 'CpyImmMerge', 'SimdAluOp', unsignedTypes, dupCode,
predType=PredType.MERGE, isUnary=True)
sveWideImmInst('cpy', 'CpyImmZero', 'SimdAluOp', unsignedTypes, dupCode,
predType=PredType.ZERO, isUnary=True)
# CPY (scalar)
sveUnaryInst('cpy', 'CpyScalar', 'SimdAluOp', unsignedTypes, dupCode,
PredType.MERGE, srcRegType=SrcRegType.Scalar)
# CPY (SIMD&FP scalar)
sveUnaryInst('cpy', 'CpySimdFpScalar', 'SimdAluOp', unsignedTypes, dupCode,
PredType.MERGE, srcRegType=SrcRegType.SimdFpScalar)
# CNTP
svePredCountPredInst('cntp', 'Cntp', 'SimdAluOp', unsignedTypes)
# CTERMEQ
cteqCode = '''
destElem = srcElem1 == srcElem2;
'''
sveCompTermInst('ctermeq', 'Ctermeq', 'IntAluOp',
['uint32_t', 'uint64_t'], cteqCode)
# CTERMNE
ctneCode = '''
destElem = srcElem1 != srcElem2;
'''
sveCompTermInst('ctermne', 'Ctermne', 'IntAluOp',
['uint32_t', 'uint64_t'], ctneCode)
# DECB, DECH, DECW, DECD (scalar)
decxCode = '''
destElem = srcElem1 - (count * imm);
'''
sveElemCountInst('dec', 'Dec', 'SimdAluOp', unsignedTypes, decxCode,
destType = DestType.Scalar, dstIs32b = False)
# DECH, DECW, DECD (vector)
sveElemCountInst('dec', 'Decv', 'SimdAluOp', bigUnsignedTypes, decxCode,
destType = DestType.Vector, dstIs32b = False)
# DECP (scalar)
decpCode = '''
XDest = XDest - count;
'''
svePredCountInst('decp', 'Decp', 'SimdAluOp', unsignedTypes, decpCode,
DestType.Scalar, SrcSize.Src64bit)
# DECP (vector)
decpvCode = '''
destElem = srcElem - count;
'''
svePredCountInst('decp', 'Decpv', 'SimdAluOp', unsignedTypes, decpvCode,
DestType.Vector)
# DUP (immediate)
sveWideImmInst('dup', 'DupImm', 'SimdAluOp', unsignedTypes, dupCode,
isUnary=True)
# DUP (indexed)
sveDupIndexInst('mov', 'DupIdx', 'SimdAluOp',
list(unsignedTypes) + ['__uint128_t'])
# DUP (scalar)
sveUnaryInst('dup', 'DupScalar', 'SimdAluOp', unsignedTypes, dupCode,
PredType.NONE, srcRegType=SrcRegType.Scalar)
# DUPM
sveWideImmInst('dupm', 'Dupm', 'SimdAluOp', unsignedTypes, dupCode,
isUnary=True)
# EOR (immediate)
eorCode = 'destElem = srcElem1 ^ srcElem2;'
sveWideImmInst('eor', 'EorImm', 'SimdAluOp', ('uint64_t',), eorCode)
# EOR (vectors, predicated)
sveBinInst('eor', 'EorPred', 'SimdAluOp', unsignedTypes, eorCode,
PredType.MERGE, True)
# EOR (vectors, unpredicated)
eorCode = 'destElem = srcElem1 ^ srcElem2;'
sveBinInst('eor', 'EorUnpred', 'SimdAluOp', ('uint64_t',), eorCode)
# EOR, EORS (predicates)
svePredLogicalInst('eor', 'PredEor', 'SimdPredAluOp', ('uint8_t',),
eorCode)
svePredLogicalInst('eors', 'PredEors', 'SimdPredAluOp', ('uint8_t',),
eorCode, isFlagSetting=True)
# EORV
eorvCode = 'destElem ^= srcElem1;'
sveAssocReducInst('eorv', 'Eorv', 'SimdReduceAluOp', unsignedTypes,
eorvCode, '0')
# EXT
sveExtInst('ext', 'Ext', 'SimdAluOp')
# FABD
fpOp = '''
FPSCR fpscr = (FPSCR) FpscrExc;
destElem = %s;
FpscrExc = fpscr;
'''
fabdCode = fpOp % 'fplibAbs<Element>(fplibSub(srcElem1, srcElem2, fpscr))'
sveBinInst('fabd', 'Fabd', 'SimdFloatAddOp', floatTypes, fabdCode,
PredType.MERGE, True)
# FABS
fabsCode = 'destElem = fplibAbs<Element>(srcElem1);'
sveUnaryInst('fabs', 'Fabs', 'SimdFloatAluOp', fpTypes, fabsCode,
PredType.MERGE)
# FACGE
fpCmpAbsOp = fpOp % ('fplibCompare%s<Element>(fplibAbs<Element>(srcElem1),'
' fplibAbs<Element>(srcElem2), fpscr)')
facgeCode = fpCmpAbsOp % 'GE'
sveCmpInst('facge', 'Facge', 'SimdFloatCmpOp', fpTypes, facgeCode)
# FACGT
facgtCode = fpCmpAbsOp % 'GT'
sveCmpInst('facgt', 'Facgt', 'SimdFloatCmpOp', fpTypes, facgtCode)
# FADD (immediate)
fpBinOp = fpOp % 'fplib%s<Element>(srcElem1, srcElem2, fpscr)'
faddCode = fpBinOp % 'Add'
sveBinImmInst('fadd', 'FaddImm', 'SimdFloatAddOp', floatTypes, faddCode,
PredType.MERGE)
# FADD (vectors, predicated)
sveBinInst('fadd', 'FaddPred', 'SimdFloatAddOp', floatTypes, faddCode,
PredType.MERGE, True)
# FADD (vectors, unpredicated)
sveBinInst('fadd', 'FaddUnpred', 'SimdFloatAddOp', floatTypes, faddCode)
# FADDA
fpAddaOp = '''
FPSCR fpscr = (FPSCR) FpscrExc;
destElem = fplibAdd<Element>(destElem, srcElem1, fpscr);
FpscrExc = FpscrExc | fpscr;
'''
sveOrderedReduction('fadda', 'Fadda', 'SimdFloatReduceAddOp', floatTypes,
fpAddaOp)
# FADDV
fpReduceOp = '''
FPSCR fpscr = (FPSCR) FpscrExc;
destElem = fplib%s<Element>(srcElem1, srcElem2, fpscr);
FpscrExc = FpscrExc | fpscr;
'''
faddvCode = fpReduceOp % 'Add'
sveNonAssocReducInst('faddv', 'Faddv', 'SimdFloatReduceAddOp', floatTypes,
faddvCode, '0')
# FCADD
sveComplexAddInst('fcadd','Fcadd', 'SimdFloatAddOp', fpTypes)
# FCMEQ (vectors)
fpCmpOp = fpOp % ('fplibCompare%s<Element>(srcElem1, srcElem2, fpscr)')
fcmeqCode = fpCmpOp % 'EQ'
sveCmpInst('fcmeq', 'Fcmeq', 'SimdFloatCmpOp', fpTypes, fcmeqCode)
# FCMEQ (zero)
fpCmpZeroOp = fpOp % 'fplibCompare%s<Element>(srcElem1, 0, fpscr)'
fcmeqZeroCode = fpCmpZeroOp % 'EQ'
sveCmpInst('fcmeq', 'FcmeqZero', 'SimdFloatCmpOp', fpTypes, fcmeqZeroCode,
True)
# FCMGE (vectors)
fcmgeCode = fpCmpOp % 'GE'
sveCmpInst('fcmge', 'Fcmge', 'SimdFloatCmpOp', fpTypes, fcmgeCode)
# FCMGE (zero)
fcmgeZeroCode = fpCmpZeroOp % 'GE'
sveCmpInst('fcmge', 'FcmgeZero', 'SimdFloatCmpOp', fpTypes, fcmgeZeroCode,
True)
# FCMGT (vectors)
fcmgtCode = fpCmpOp % 'GT'
sveCmpInst('fcmgt', 'Fcmgt', 'SimdFloatCmpOp', fpTypes, fcmgtCode)
# FCMGT (zero)
fcmgtZeroCode = fpCmpZeroOp % 'GT'
sveCmpInst('fcmgt', 'FcmgtZero', 'SimdFloatCmpOp', fpTypes, fcmgtZeroCode,
True)
# FCMLE (zero)
fpCmpRevZeroOp = fpOp % ('fplibCompare%s<Element>(0, srcElem1, fpscr)')
fcmleZeroCode = fpCmpRevZeroOp % 'GE'
sveCmpInst('fcmle', 'FcmleZero', 'SimdFloatCmpOp', fpTypes, fcmleZeroCode,
True)
# FCMLT (zero)
fcmltZeroCode = fpCmpRevZeroOp % 'GT'
sveCmpInst('fcmlt', 'FcmltZero', 'SimdFloatCmpOp', fpTypes, fcmltZeroCode,
True)
# FCMNE (vectors)
fcmneCode = fpOp % ('!fplibCompareEQ<Element>(srcElem1, srcElem2, fpscr)')
sveCmpInst('fcmne', 'Fcmne', 'SimdFloatCmpOp', fpTypes, fcmneCode)
# FCMNE (zero)
fcmneZeroCode = fpOp % ('!fplibCompareEQ<Element>(srcElem1, 0, fpscr)')
sveCmpInst('fcmne', 'FcmneZero', 'SimdFloatCmpOp', fpTypes, fcmneZeroCode,
True)
# FCMUO (vectors)
fcmuoCode = fpCmpOp % 'UN'
sveCmpInst('fcmuo', 'Fcmuo', 'SimdFloatCmpOp', fpTypes, fcmuoCode)
# FCMLA (indexed)
sveComplexMulAddInst('fcmla', 'Fcmlai', 'SimdFloatMultAccOp',
fpTypes[1:], predType = PredType.NONE)
# FCMLA (vectors)
sveComplexMulAddInst('fcmla', 'Fcmlav', 'SimdFloatMultAccOp',
fpTypes, predType = PredType.MERGE)
# FCPY
sveWideImmInst('fcpy', 'Fcpy', 'SimdAluOp', unsignedTypes, dupCode,
predType=PredType.MERGE, isUnary=True)
# FCVT
fcvtCode = fpOp % ('fplibConvert<SElement, DElement>('
'srcElem1, FPCRRounding(fpscr), fpscr)')
sveCvtInst('fcvt', 'FcvtNarrow', 'SimdCvtOp',
('uint32_t, uint16_t',
'uint64_t, uint16_t',
'uint64_t, uint32_t'),
fcvtCode, CvtDir.Narrow)
sveCvtInst('fcvt', 'FcvtWiden', 'SimdCvtOp',
('uint16_t, uint32_t',
'uint16_t, uint64_t',
'uint32_t, uint64_t'),
fcvtCode, CvtDir.Widen)
# FCVTZS
fcvtIntCode = fpOp % ('fplibFPToFixed<SElement, DElement>('
'srcElem1, %s, %s, %s, fpscr)')
fcvtzsCode = fcvtIntCode % ('0', 'false', 'FPRounding_ZERO')
sveCvtInst('fcvtzs', 'FcvtzsNarrow', 'SimdCvtOp',
('uint16_t, uint16_t',
'uint32_t, uint32_t',
'uint64_t, uint32_t',
'uint64_t, uint64_t'),
fcvtzsCode, CvtDir.Narrow)
sveCvtInst('fcvtzs', 'FcvtzsWiden', 'SimdCvtOp',
('uint16_t, uint32_t',
'uint16_t, uint64_t',
'uint32_t, uint64_t'),
fcvtzsCode, CvtDir.Widen)
# FCVTZU
fcvtzuCode = fcvtIntCode % ('0', 'true', 'FPRounding_ZERO')
sveCvtInst('fcvtzu', 'FcvtzuNarrow', 'SimdCvtOp',
('uint16_t, uint16_t',
'uint32_t, uint32_t',
'uint64_t, uint32_t',
'uint64_t, uint64_t'),
fcvtzuCode, CvtDir.Narrow)
sveCvtInst('fcvtzu', 'FcvtzuWiden', 'SimdCvtOp',
('uint16_t, uint32_t',
'uint16_t, uint64_t',
'uint32_t, uint64_t'),
fcvtzuCode, CvtDir.Widen)
# FDIV
fdivCode = fpBinOp % 'Div'
sveBinInst('fdiv', 'Fdiv', 'SimdFloatDivOp', floatTypes, fdivCode,
PredType.MERGE, True)
# FDIVR
fpBinRevOp = fpOp % 'fplib%s<Element>(srcElem2, srcElem1, fpscr)'
fdivrCode = fpBinRevOp % 'Div'
sveBinInst('fdivr', 'Fdivr', 'SimdFloatDivOp', floatTypes, fdivrCode,
PredType.MERGE, True)
# FDUP
sveWideImmInst('fdup', 'Fdup', 'SimdFloatAluOp', floatTypes, dupCode,
isUnary=True)
# FEXPA
fexpaCode = 'destElem = fplibExpA<Element>(srcElem1);'
sveUnaryInst('fexpa', 'Fexpa', 'SimdFloatAluOp', fpTypes, fexpaCode)
# FMAD
fmadCode = fpOp % ('fplibMulAdd<Element>('
'srcElem1, destElem, srcElem2, fpscr)')
sveTerInst('fmad', 'Fmad', 'SimdFloatMultAccOp', floatTypes, fmadCode,
PredType.MERGE)
# FMAX (immediate)
fmaxCode = fpBinOp % 'Max'
sveBinImmInst('fmax', 'FmaxImm', 'SimdFloatCmpOp', floatTypes, fmaxCode,
PredType.MERGE)
# FMAX (vectors)
sveBinInst('fmax', 'Fmax', 'SimdFloatCmpOp', floatTypes, fmaxCode,
PredType.MERGE, True)
# FMAXNM (immediate)
fmaxnmCode = fpBinOp % 'MaxNum'
sveBinImmInst('fmaxnm', 'FmaxnmImm', 'SimdFloatCmpOp', floatTypes,
fmaxnmCode, PredType.MERGE)
# FMAXNM (vectors)
sveBinInst('fmaxnm', 'Fmaxnm', 'SimdFloatCmpOp', floatTypes, fmaxnmCode,
PredType.MERGE, True)
# FMAXNMV
fmaxnmvCode = fpReduceOp % 'MaxNum'
sveNonAssocReducInst('fmaxnmv', 'Fmaxnmv', 'SimdFloatReduceCmpOp',
floatTypes, fmaxnmvCode, 'fplibDefaultNaN<Element>()')
# FMAXV
fmaxvCode = fpReduceOp % 'Max'
sveNonAssocReducInst('fmaxv', 'Fmaxv', 'SimdFloatReduceCmpOp', floatTypes,
fmaxvCode, 'fplibInfinity<Element>(1)')
# FMIN (immediate)
fminCode = fpBinOp % 'Min'
sveBinImmInst('fmin', 'FminImm', 'SimdFloatCmpOp', floatTypes, fminCode,
PredType.MERGE)
# FMIN (vectors)
sveBinInst('fmin', 'Fmin', 'SimdFloatCmpOp', floatTypes, fminCode,
PredType.MERGE, True)
# FMINNM (immediate)
fminnmCode = fpBinOp % 'MinNum'
sveBinImmInst('fminnm', 'FminnmImm', 'SimdFloatCmpOp', floatTypes,
fminnmCode, PredType.MERGE)
# FMINNM (vectors)
sveBinInst('fminnm', 'Fminnm', 'SimdFloatCmpOp', floatTypes, fminnmCode,
PredType.MERGE, True)
# FMINNMV
fminnmvCode = fpReduceOp % 'MinNum'
sveNonAssocReducInst('fminnmv', 'Fminnmv', 'SimdFloatReduceCmpOp',
floatTypes, fminnmvCode, 'fplibDefaultNaN<Element>()')
# FMINV
fminvCode = fpReduceOp % 'Min'
sveNonAssocReducInst('fminv', 'Fminv', 'SimdFloatReduceCmpOp', floatTypes,
fminvCode, 'fplibInfinity<Element>(0)')
fmlaCode = fpOp % ('fplibMulAdd<Element>('
'destElem, srcElem1, srcElem2, fpscr)')
# FMLA (indexed)
sveTerIdxInst('fmla', 'FmlaIdx', 'SimdFloatMultAccOp', floatTypes,
fmlaCode, PredType.MERGE)
# FMLA (vectors)
sveTerInst('fmla', 'Fmla', 'SimdFloatMultAccOp', floatTypes, fmlaCode,
PredType.MERGE)
fmlsCode = fpOp % ('fplibMulAdd<Element>(destElem, '
'fplibNeg<Element>(srcElem1), srcElem2, fpscr)')
# FMLS (indexed)
sveTerIdxInst('fmls', 'FmlsIdx', 'SimdFloatMultAccOp', floatTypes,
fmlsCode, PredType.MERGE)
# FMLS (vectors)
sveTerInst('fmls', 'Fmls', 'SimdFloatMultAccOp', floatTypes, fmlsCode,
PredType.MERGE)
# FMSB
fmsbCode = fpOp % ('fplibMulAdd<Element>(srcElem1, '
'fplibNeg<Element>(destElem), srcElem2, fpscr)')
sveTerInst('fmsb', 'Fmsb', 'SimdFloatMultAccOp', floatTypes, fmsbCode,
PredType.MERGE)
# FMUL (immediate)
fpBinOp = fpOp % 'fplib%s<Element>(srcElem1, srcElem2, fpscr)'
fmulCode = fpBinOp % 'Mul'
sveBinImmInst('fmul', 'FmulImm', 'SimdFloatMultOp', floatTypes, fmulCode,
PredType.MERGE)
# TODO: FMUL (indexed)
# FMUL (vectors, predicated)
fmulCode = fpBinOp % 'Mul'
sveBinInst('fmul', 'FmulPred', 'SimdFloatMultOp', floatTypes, fmulCode,
PredType.MERGE, True)
# FMUL (vectors, unpredicated)
sveBinInst('fmul', 'FmulUnpred', 'SimdFloatMultOp', floatTypes, fmulCode)
# FMUL (indexed)
sveBinIdxInst('fmul', 'FmulIdx', 'SimdFloatMultOp', floatTypes, fmulCode)
# FMULX
fmulxCode = fpBinOp % 'MulX'
sveBinInst('fmulx', 'Fmulx', 'SimdFloatMultOp', floatTypes, fmulxCode,
PredType.MERGE, True)
# FNEG
fnegCode = 'destElem = fplibNeg<Element>(srcElem1);'
sveUnaryInst('fneg', 'Fneg', 'SimdFloatAluOp', fpTypes, fnegCode,
PredType.MERGE)
# FNMAD
fnmadCode = fpOp % ('fplibMulAdd<Element>('
'fplibNeg<Element>(srcElem1), '
'fplibNeg<Element>(destElem), srcElem2, fpscr)')
sveTerInst('fnmad', 'Fnmad', 'SimdFloatMultAccOp', floatTypes, fnmadCode,
PredType.MERGE)
# FNMLA
fnmlaCode = fpOp % ('fplibMulAdd<Element>('
'fplibNeg<Element>(destElem), '
'fplibNeg<Element>(srcElem1), srcElem2, fpscr)')
sveTerInst('fnmla', 'Fnmla', 'SimdFloatMultAccOp', floatTypes, fnmlaCode,
PredType.MERGE)
# FNMLS
fnmlsCode = fpOp % ('fplibMulAdd<Element>('
'fplibNeg<Element>(destElem), srcElem1, srcElem2, '
'fpscr)')
sveTerInst('fnmls', 'Fnmls', 'SimdFloatMultAccOp', floatTypes, fnmlsCode,
PredType.MERGE)
# FNMSB
fnmsbCode = fpOp % ('fplibMulAdd<Element>('
'fplibNeg<Element>(srcElem1), destElem, srcElem2, '
'fpscr)')
sveTerInst('fnmsb', 'Fnmsb', 'SimdFloatMultAccOp', floatTypes, fnmsbCode,
PredType.MERGE)
# FRECPE
frecpeCode = fpOp % 'fplibRecipEstimate<Element>(srcElem1, fpscr)'
sveUnaryInst('frecpe', 'Frecpe', 'SimdFloatMultAccOp', floatTypes,
frecpeCode)
# FRECPS
frecpsCode = fpBinOp % 'RecipStepFused'
sveBinInst('frecps', 'Frecps', 'SimdFloatMultAccOp', floatTypes,
frecpsCode)
# FRECPX
frecpxCode = fpOp % "fplibRecpX<Element>(srcElem1, fpscr)"
sveUnaryInst('frecpx', 'Frecpx', 'SimdFloatMultAccOp', floatTypes,
frecpxCode, PredType.MERGE)
# FRINTA
frintCode = fpOp % 'fplibRoundInt<Element>(srcElem1, %s, %s, fpscr)'
frintaCode = frintCode % ('FPRounding_TIEAWAY', 'false')
sveUnaryInst('frinta', 'Frinta', 'SimdCvtOp', floatTypes, frintaCode,
PredType.MERGE)
# FRINTI
frintiCode = frintCode % ('FPCRRounding(fpscr)', 'false')
sveUnaryInst('frinti', 'Frinti', 'SimdCvtOp', floatTypes, frintiCode,
PredType.MERGE)
# FRINTM
frintmCode = frintCode % ('FPRounding_NEGINF', 'false')
sveUnaryInst('frintm', 'Frintm', 'SimdCvtOp', floatTypes, frintmCode,
PredType.MERGE)
# FRINTN
frintnCode = frintCode % ('FPRounding_TIEEVEN', 'false')
sveUnaryInst('frintn', 'Frintn', 'SimdCvtOp', floatTypes, frintnCode,
PredType.MERGE)
# FRINTP
frintpCode = frintCode % ('FPRounding_POSINF', 'false')
sveUnaryInst('frintp', 'Frintp', 'SimdCvtOp', floatTypes, frintpCode,
PredType.MERGE)
# FRINTX
frintxCode = frintCode % ('FPCRRounding(fpscr)', 'true')
sveUnaryInst('frintx', 'Frintx', 'SimdCvtOp', floatTypes, frintxCode,
PredType.MERGE)
# FRINTZ
frintzCode = frintCode % ('FPRounding_ZERO', 'false')
sveUnaryInst('frintz', 'Frintz', 'SimdCvtOp', floatTypes, frintzCode,
PredType.MERGE)
# FRSQRTE
frsqrteCode = fpOp % 'fplibRSqrtEstimate<Element>(srcElem1, fpscr)'
sveUnaryInst('frsqrte', 'Frsqrte', 'SimdFloatSqrtOp', floatTypes,
frsqrteCode)
# FRSQRTS
frsqrtsCode = fpBinOp % 'RSqrtStepFused'
sveBinInst('frsqrts', 'Frsqrts', 'SimdFloatMiscOp', floatTypes,
frsqrtsCode)
# FSCALE
fscaleCode = fpBinOp % 'Scale'
sveBinInst('fscale', 'Fscale', 'SimdFloatMiscOp', floatTypes, fscaleCode,
PredType.MERGE, True)
# FSQRT
fsqrtCode = fpOp % "fplibSqrt<Element>(srcElem1, fpscr)"
sveUnaryInst('fsqrt', 'Fsqrt', 'SimdFloatSqrtOp', floatTypes, fsqrtCode,
PredType.MERGE)
# FSUB (immediate)
fsubCode = fpBinOp % 'Sub'
sveBinImmInst('fsub', 'FsubImm', 'SimdFloatAddOp', floatTypes, fsubCode,
PredType.MERGE)
# FSUB (vectors, predicated)
sveBinInst('fsub', 'FsubPred', 'SimdFloatAddOp', floatTypes, fsubCode,
PredType.MERGE, True)
# FSUB (vectors, unpredicated)
sveBinInst('fsub', 'FsubUnpred', 'SimdFloatAddOp', floatTypes, fsubCode)
# FSUBR (immediate)
fsubrCode = fpBinRevOp % 'Sub'
sveBinImmInst('fsubr', 'FsubrImm', 'SimdFloatAddOp', floatTypes, fsubrCode,
PredType.MERGE)
# FSUBR (vectors)
sveBinInst('fsubr', 'Fsubr', 'SimdFloatAddOp', floatTypes, fsubrCode,
PredType.MERGE, True)
# FTMAD
ftmadCode = fpOp % ('fplibTrigMulAdd<Element>('
'srcElem3, destElem, srcElem2, fpscr)')
sveTerImmInst('ftmad', 'Ftmad', 'SimdFloatMultAccOp', floatTypes,
ftmadCode)
# FTSMUL
ftsmulCode = fpBinOp % 'TrigSMul'
sveBinInst('ftsmul', 'Ftsmul', 'SimdFloatMiscOp', floatTypes, ftsmulCode)
# FTSSEL
ftsselCode = fpBinOp % 'TrigSSel'
sveBinInst('ftssel', 'Ftssel', 'SimdFloatMultOp', floatTypes, ftsselCode)
# INCB, INCH, INCW, INCD (scalar)
incxCode = '''
destElem = srcElem1 + (count * imm);
'''
sveElemCountInst('inc', 'Inc', 'SimdAluOp', unsignedTypes, incxCode,
destType = DestType.Scalar, dstIs32b = False)
# INCH, INCW, INCD (vector)
sveElemCountInst('inc', 'Incv', 'SimdAluOp', bigUnsignedTypes, incxCode,
destType = DestType.Vector, dstIs32b = False)
# INCP (scalar)
incpCode = '''
XDest = XDest + count;
'''
svePredCountInst('incp', 'Incp', 'SimdAluOp', unsignedTypes, incpCode,
DestType.Scalar, SrcSize.Src64bit)
# INCP (vector)
incpvCode = '''
destElem = srcElem + count;
'''
svePredCountInst('incp', 'Incpv', 'SimdAluOp', unsignedTypes, incpvCode,
DestType.Vector)
# INDEX (immediate, scalar)
sveIndex(IndexFormat.ImmReg)
# INDEX (immediates)
sveIndex(IndexFormat.ImmImm)
# INDEX (scalar, immediate)
sveIndex(IndexFormat.RegImm)
# INDEX (scalars)
sveIndex(IndexFormat.RegReg)
# INSR (scalar)
sveShiftAndInsertInst('insr', 'Insr', 'SimdAluOp', unsignedTypes,
srcType = SrcRegType.Scalar)
# INSR (SIMD&FP scalar)
sveShiftAndInsertInst('insr', 'Insrf', 'SimdAluOp', unsignedTypes,
srcType = SrcRegType.SimdFpScalar)
# LASTA (scalar)
lastaCode = '''
last++;
if (last >= eCount) {
last = 0;
}
destElem = AA64FpOp1_x[last];'''
sveSelectInst('lasta', 'Lasta', 'SimdAluOp', unsignedTypes, lastaCode,
isCond = False)
# LASTA (SIMD&FP scalar)
sveSelectInst('lasta', 'Lastaf', 'SimdAluOp', unsignedTypes, lastaCode,
isCond = False, destType = DstRegType.SimdFpScalar)
# LASTB (scalar)
lastbCode = '''
if (last < 0) {
last = eCount - 1;
}
destElem = AA64FpOp1_x[last];'''
sveSelectInst('lastb', 'Lastb', 'SimdAluOp', unsignedTypes, lastbCode,
isCond = False)
# LASTB (SIMD&FP scalar)
sveSelectInst('lastb', 'Lastbf', 'SimdAluOp', unsignedTypes, lastbCode,
isCond = False, destType = DstRegType.SimdFpScalar)
# LSL (immediate, predicated)
lslCode = '''
if (srcElem2 == 0) {
destElem = srcElem1;
} else if (srcElem2 >= sizeof(Element) * 8) {
destElem = 0;
} else {
destElem = srcElem1 << srcElem2;
}
'''
sveBinImmInst('lsl', 'LslImmPred', 'SimdAluOp', unsignedTypes, lslCode,
PredType.MERGE)
# LSL (immediate, unpredicated)
sveBinImmInst('lsl', 'LslImmUnpred', 'SimdAluOp', unsignedTypes, lslCode)
# LSL (vectors)
sveBinInst('lsl', 'LslPred', 'SimdAluOp', unsignedTypes, lslCode,
PredType.MERGE, True)
# LSL (wide elements, predicated)
sveShiftByWideElemsInst('lsl', 'LslWidePred', 'SimdAluOp', unsignedTypes,
lslCode, PredType.MERGE)
# LSL (wide elements, unpredicated)
sveShiftByWideElemsInst('lsl', 'LslWideUnpred', 'SimdAluOp', unsignedTypes,
lslCode)
# LSLR
lslrCode = '''
if (srcElem1 == 0) {
destElem = srcElem2;
} else if (srcElem1 >= sizeof(Element) * 8) {
destElem = 0;
} else {
destElem = srcElem2 << srcElem1;
}
'''
sveBinInst('lslr', 'Lslr', 'SimdAluOp', unsignedTypes, lslrCode,
PredType.MERGE, True)
# LSR (immediate, predicated)
lsrCode = '''
if (srcElem2 >= sizeof(Element) * 8) {
destElem = 0;
} else {
destElem = srcElem1 >> srcElem2;
}
'''
sveBinImmInst('lsr', 'LsrImmPred', 'SimdAluOp', unsignedTypes, lsrCode,
PredType.MERGE)
# LSR (immediate, unpredicated)
sveBinImmInst('lsr', 'LsrImmUnpred', 'SimdAluOp', unsignedTypes, lsrCode)
# LSR (vectors)
sveBinInst('lsr', 'LsrPred', 'SimdAluOp', unsignedTypes, lsrCode,
PredType.MERGE, True)
# LSR (wide elements, predicated)
sveShiftByWideElemsInst('lsr', 'LsrWidePred', 'SimdAluOp', unsignedTypes,
lsrCode, PredType.MERGE)
# LSR (wide elements, unpredicated)
sveShiftByWideElemsInst('lsr', 'LsrWideUnpred', 'SimdAluOp', unsignedTypes,
lsrCode)
# LSRR
lsrrCode = '''
if (srcElem1 >= sizeof(Element) * 8) {
destElem = 0;
} else {
destElem = srcElem2 >> srcElem1;
}
'''
sveBinInst('lsrr', 'Lsrr', 'SimdAluOp', unsignedTypes, lsrrCode,
PredType.MERGE, True)
# MAD
madCode = 'destElem = srcElem1 + destElem * srcElem2;'
sveTerInst('mad', 'Mad', 'SimdMultAccOp', signedTypes, madCode)
# MLA
mlaCode = 'destElem += srcElem1 * srcElem2;'
sveTerInst('mla', 'Mla', 'SimdMultAccOp', signedTypes, mlaCode)
# MLS
mlsCode = 'destElem -= srcElem1 * srcElem2;'
sveTerInst('mls', 'Mls', 'SimdMultAccOp', signedTypes, mlsCode)
# MOVPRFX (predicated)
movCode = 'destElem = srcElem1;'
sveUnaryInst('movprfx', 'MovprfxPredM', 'SimdMiscOp', unsignedTypes,
movCode, PredType.MERGE)
sveUnaryInst('movprfx', 'MovprfxPredZ', 'SimdMiscOp', unsignedTypes,
movCode, PredType.ZERO)
# MOVPRFX (unpredicated)
sveUnaryInst('movprfx', 'MovprfxUnpred', 'SimdMiscOp', ('uint64_t',),
movCode)
# MSB
msbCode = 'destElem = srcElem1 - destElem * srcElem2;'
sveTerInst('msb', 'Msb', 'SimdMultAccOp', signedTypes, msbCode)
# MUL (immediate)
mulCode = 'destElem = srcElem1 * srcElem2;'
sveWideImmInst('mul', 'MulImm', 'SimdMultOp', unsignedTypes, mulCode)
# MUL (vectors)
sveBinInst('mul', 'Mul', 'SimdMultOp', unsignedTypes, mulCode,
PredType.MERGE, True)
# NAND, NANDS
nandCode = 'destElem = !(srcElem1 & srcElem2);';
svePredLogicalInst('nand', 'PredNand', 'SimdPredAluOp', ('uint8_t',),
nandCode)
svePredLogicalInst('nands', 'PredNands', 'SimdPredAluOp', ('uint8_t',),
nandCode, isFlagSetting=True)
# NEG
negCode = 'destElem = -srcElem1;'
sveUnaryInst('neg', 'Neg', 'SimdAluOp', signedTypes, negCode,
PredType.MERGE)
# NOR, NORS
norCode = 'destElem = !(srcElem1 | srcElem2);';
svePredLogicalInst('nor', 'PredNor', 'SimdPredAluOp', ('uint8_t',),
norCode)
svePredLogicalInst('nors', 'PredNors', 'SimdPredAluOp', ('uint8_t',),
norCode, isFlagSetting=True)
# NOT (vector)
notCode = 'destElem = ~srcElem1;'
sveUnaryInst('not', 'Not', 'SimdAluOp', unsignedTypes, notCode,
PredType.MERGE)
# ORN, ORNS (predicates)
ornCode = 'destElem = srcElem1 | !srcElem2;';
svePredLogicalInst('orn', 'PredOrn', 'SimdPredAluOp', ('uint8_t',),
ornCode)
svePredLogicalInst('orns', 'PredOrns', 'SimdPredAluOp', ('uint8_t',),
ornCode, isFlagSetting=True)
# ORR (immediate)
orCode = 'destElem = srcElem1 | srcElem2;'
sveWideImmInst('orr', 'OrrImm', 'SimdAluOp', ('uint64_t',), orCode)
# ORR (vectors, predicated)
sveBinInst('orr', 'OrrPred', 'SimdAluOp', unsignedTypes, orCode,
PredType.MERGE, True)
# ORR (vectors, unpredicated)
orCode = 'destElem = srcElem1 | srcElem2;'
sveBinInst('orr', 'OrrUnpred', 'SimdAluOp', ('uint64_t',), orCode)
# ORR, ORRS (predicates)
svePredLogicalInst('orr', 'PredOrr', 'SimdPredAluOp', ('uint8_t',), orCode)
svePredLogicalInst('orrs', 'PredOrrs', 'SimdPredAluOp', ('uint8_t',),
orCode, isFlagSetting=True)
# ORV
orvCode = 'destElem |= srcElem1;'
sveAssocReducInst('orv', 'Orv', 'SimdReduceAluOp', unsignedTypes,
orvCode, '0')
# PFALSE
pfalseCode = '''
PDest_ub[0] = 0;
destPred.reset();
'''
svePredUnaryWImplicitSrcInst('pfalse', 'Pfalse', 'SimdPredAluOp',
pfalseCode)
# PFIRST
svePFirstInst('pfirst', 'Pfirst', 'SimdPredAluOp')
# PNEXT
svePNextInst('pnext', 'Pnext', 'SimdPredAluOp', unsignedTypes)
# PTEST
svePredTestInst('ptest', 'Ptest', 'SimdPredAluOp')
# PTRUE
svePtrueInst('ptrue', 'Ptrue', 'SimdPredAluOp', unsignedTypes, False)
# PTRUES
svePtrueInst('ptrues', 'Ptrues', 'SimdPredAluOp', unsignedTypes, True)
# PUNPKHI
sveUnpackInst('punpkhi', 'Punpkhi', 'SimdPredAluOp', unsignedWideSDTypes,
unpackHalf = Unpack.High, regType = SrcRegType.Predicate)
# PUNPKLO
sveUnpackInst('punpklo', 'Punpklo', 'SimdPredAluOp', unsignedWideSDTypes,
unpackHalf = Unpack.Low, regType = SrcRegType.Predicate)
# RBIT
rbitCode = '''
destElem = reverseBits(srcElem1);'''
sveUnaryInst('rbit', 'Rbit', 'SimdAluOp', unsignedTypes, rbitCode,
predType=PredType.MERGE, srcRegType=SrcRegType.Vector)
# RDFFR (unpredicated)
rdffrUnpredCode = '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<uint8_t>(
xc->tcBase());
for (unsigned i = 0; i < eCount; i++) {
PDest_ub[i] = Ffr_ub[i];
}'''
svePredUnaryWImplicitSrcInst('rdffr', 'RdffrUnpred', 'SimdPredAluOp',
rdffrUnpredCode)
# RDFFR, RDFFRS (predicated)
rdffrPredCode = '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<uint8_t>(
xc->tcBase());
for (unsigned i = 0; i < eCount; i++) {
if (GpOp_ub[i]) {
PDest_ub[i] = Ffr_ub[i];
} else {
PDest_ub[i] = false;
}
}'''
svePredUnaryWImplicitSrcInst('rdffr', 'RdffrPred', 'SimdPredAluOp',
rdffrPredCode, PredType.ZERO, False)
svePredUnaryWImplicitSrcInst('rdffrs', 'RdffrsPred', 'SimdPredAluOp',
rdffrPredCode, PredType.ZERO, True)
# RDVL
rdvlCode = sveEnabledCheckCode + '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<uint8_t>(
xc->tcBase());
XDest = eCount * (int64_t) imm;
'''
rdvlIop = InstObjParams('rdvl', 'SveRdvl', 'RegImmOp', rdvlCode, [])
header_output += RegImmOpDeclare.subst(rdvlIop)
decoder_output += RegImmOpConstructor.subst(rdvlIop)
exec_output += BasicExecute.subst(rdvlIop)
# REV (predicate)
sveReverseElementsInst('rev', 'Revp', 'SimdPredAluOp', unsignedTypes,
srcType = SrcRegType.Predicate)
# REV (vector)
sveReverseElementsInst('rev', 'Revv', 'SimdAluOp', unsignedTypes,
srcType = SrcRegType.Vector)
# REVB
revCode = '''
%(revtype)s* srcPtr = reinterpret_cast<%(revtype)s*>(&srcElem1);
%(revtype)s* dstPtr = reinterpret_cast<%(revtype)s*>(&destElem);
uint8_t subelements = sizeof(Element) / sizeof(%(revtype)s);
for(int i = 0; i < subelements; ++i) {
dstPtr[subelements - i - 1] = srcPtr[i];
}'''
sveUnaryInst('revb', 'Revb', 'SimdAluOp',
['uint16_t', 'uint32_t', 'uint64_t'],
revCode % {'revtype' : 'uint8_t'}, predType=PredType.MERGE,
srcRegType=SrcRegType.Vector, decoder='Generic')
# REVH
sveUnaryInst('revh', 'Revh', 'SimdAluOp', ['uint32_t', 'uint64_t'],
revCode % {'revtype' : 'uint16_t'}, predType=PredType.MERGE,
srcRegType=SrcRegType.Vector, decoder='Generic')
# REVW
sveUnaryInst('revw', 'Revw', 'SimdAluOp', ['uint64_t'],
revCode % {'revtype' : 'uint32_t'}, predType=PredType.MERGE,
srcRegType=SrcRegType.Vector, decoder='Generic')
# SABD
abdCode = '''
destElem = (srcElem1 > srcElem2) ? (srcElem1 - srcElem2) :
(srcElem2 - srcElem1);
'''
sveBinInst('sabd', 'Sabd', 'SimdAddOp', signedTypes, abdCode,
PredType.MERGE, True)
# SADDV
addvCode = 'destElem += srcElem1;'
sveWideningAssocReducInst('saddv', 'Saddv', 'SimdReduceAddOp',
['int8_t, int64_t', 'int16_t, int64_t', 'int32_t, int64_t'],
addvCode, '0')
# SCVTF
scvtfCode = fpOp % ('fplibFixedToFP<DElement>('
'sext<sizeof(SElement) * 8>(srcElem1), 0,'
' false, FPCRRounding(fpscr), fpscr)')
sveCvtInst('scvtf', 'ScvtfNarrow', 'SimdCvtOp',
('uint16_t, uint16_t',
'uint32_t, uint16_t',
'uint64_t, uint16_t',
'uint32_t, uint32_t',
'uint64_t, uint32_t',
'uint64_t, uint64_t'),
scvtfCode, CvtDir.Narrow)
sveCvtInst('scvtf', 'ScvtfWiden', 'SimdCvtOp', ('uint32_t, uint64_t',),
scvtfCode, CvtDir.Widen)
# SDIV
sdivCode = '''
constexpr Element ELEM_MIN = std::numeric_limits<Element>::min();
destElem = (srcElem2 == 0) ? 0 :
(srcElem2 == -1 && srcElem1 == ELEM_MIN) ? ELEM_MIN :
(srcElem1 / srcElem2);
'''
sveBinInst('sdiv', 'Sdiv', 'SimdDivOp', signedTypes, sdivCode,
PredType.MERGE, True)
# SDIVR
sdivrCode = '''
constexpr Element ELEM_MIN = std::numeric_limits<Element>::min();
destElem = (srcElem1 == 0) ? 0 :
(srcElem1 == -1 && srcElem2 == ELEM_MIN) ? ELEM_MIN :
(srcElem2 / srcElem1);
'''
sveBinInst('sdivr', 'Sdivr', 'SimdDivOp', signedTypes, sdivrCode,
PredType.MERGE, True)
# SDOT (indexed)
sveDotInst('sdot', 'Sdoti', 'SimdAluOp', ['int8_t, int32_t',
'int16_t, int64_t'], isIndexed = True)
# SDOT (vectors)
sveDotInst('sdot', 'Sdotv', 'SimdAluOp', ['int8_t, int32_t',
'int16_t, int64_t'], isIndexed = False)
# SEL (predicates)
selCode = 'destElem = srcElem1;'
svePredLogicalInst('sel', 'PredSel', 'SimdPredAluOp', ('uint8_t',),
selCode, PredType.SELECT)
# SEL (vectors)
sveBinInst('sel', 'Sel', 'SimdAluOp', unsignedTypes, selCode,
PredType.SELECT, False)
# SETFFR
setffrCode = '''
Ffr_ub[0] = true;
destPred.set();'''
svePredWriteFfrInst('setffr', 'Setffr', 'SimdPredAluOp', setffrCode, True)
# SMAX (immediate)
maxCode = 'destElem = (srcElem1 > srcElem2) ? srcElem1 : srcElem2;'
sveWideImmInst('smax', 'SmaxImm', 'SimdCmpOp', signedTypes, maxCode)
# SMAX (vectors)
sveBinInst('smax', 'Smax', 'SimdCmpOp', signedTypes, maxCode,
PredType.MERGE, True)
# SMAXV
maxvCode = '''
if (srcElem1 > destElem)
destElem = srcElem1;
'''
sveAssocReducInst('smaxv', 'Smaxv', 'SimdReduceCmpOp', signedTypes,
maxvCode, 'std::numeric_limits<Element>::min()')
# SMIN (immediate)
minCode = 'destElem = (srcElem1 < srcElem2) ? srcElem1 : srcElem2;'
sveWideImmInst('smin', 'SminImm', 'SimdCmpOp', signedTypes, minCode)
# SMIN (vectors)
sveBinInst('smin', 'Smin', 'SimdCmpOp', signedTypes, minCode,
PredType.MERGE, True)
# SMINV
minvCode = '''
if (srcElem1 < destElem)
destElem = srcElem1;
'''
sveAssocReducInst('sminv', 'Sminv', 'SimdReduceCmpOp', signedTypes,
minvCode, 'std::numeric_limits<Element>::max()')
# SMULH
exec_output += '''
template <class T>
T do_mulh(T srcElem1, T srcElem2)
{
return ((int64_t)srcElem1 * (int64_t)srcElem2) >> sizeof(T) * 8;
}
int64_t do_mulh(int64_t srcElem1, int64_t srcElem2)
{
uint64_t x = (uint64_t) llabs(srcElem1);
uint64_t y = (uint64_t) llabs(srcElem2);
uint64_t a = x >> 32;
uint64_t b = x & 0xFFFFFFFF;
uint64_t c = y >> 32;
uint64_t d = y & 0xFFFFFFFF;
uint64_t hi = a * c;
uint64_t lo = b * d;
hi += (a * d) >> 32;
uint64_t tmp = lo;
lo += ((a * d) & 0xFFFFFFFF) << 32;
if (lo < tmp)
hi++;
hi += (b * c) >> 32;
tmp = lo;
lo += ((b * c) & 0xFFFFFFFF) << 32;
if (lo < tmp)
hi++;
uint64_t destElem = hi;
if ((srcElem1 < 0) ^ (srcElem2 < 0)) {
uint64_t tmp = lo = ~lo;
destElem = ~hi;
if (++lo < tmp)
destElem++;
}
return destElem;
}
uint64_t do_mulh(uint64_t srcElem1, uint64_t srcElem2)
{
uint64_t x = srcElem1;
uint64_t y = srcElem2;
uint64_t a = x >> 32;
uint64_t b = x & 0xFFFFFFFF;
uint64_t c = y >> 32;
uint64_t d = y & 0xFFFFFFFF;
uint64_t hi = a * c;
uint64_t lo = b * d;
hi += (a * d) >> 32;
uint64_t tmp = lo;
lo += ((a * d) & 0xFFFFFFFF) << 32;
if (lo < tmp)
hi++;
hi += (b * c) >> 32;
tmp = lo;
lo += ((b * c) & 0xFFFFFFFF) << 32;
if (lo < tmp)
hi++;
return hi;
}'''
mulhCode = '''
destElem = do_mulh(srcElem1, srcElem2);'''
sveBinInst('smulh', 'Smulh', 'SimdMultOp', signedTypes, mulhCode,
PredType.MERGE, True)
# SPLICE
sveSpliceInst('splice', 'Splice', 'SimdAluOp', unsignedTypes)
# SQADD (immediate)
sqaddCode = '''
destElem = srcElem1 + srcElem2;
bool negDest = (destElem < 0);
bool negSrc1 = (srcElem1 < 0);
bool negSrc2 = (srcElem2 < 0);
if ((negDest != negSrc1) && (negSrc1 == negSrc2)) {
destElem = static_cast<Element>(
(Element)1 << (sizeof(Element) * 8 - 1)
);
if (negDest)
destElem -= 1;
}
'''
sveWideImmInst('sqadd', 'SqaddImm', 'SimdAddOp', signedTypes, sqaddCode)
# SQADD (vectors)
sveBinInst('sqadd', 'Sqadd', 'SimdAddOp', signedTypes, sqaddCode)
# SQDECB, SQDECH, SQDECW, SQDECD (scalar, 32-bit)
sqdecCode = '''
destElem = srcElem1 - (count * imm);
bool negDest = (destElem < 0);
bool negSrc = (srcElem1 < 0);
bool posCount = ((count * imm) >= 0);
if ((negDest != negSrc) && (negSrc == posCount)) {
destElem = static_cast<%(dstType)s>(
(%(dstType)s)1 << (sizeof(%(dstType)s) * 8 - 1)
);
if (negDest)
destElem -= 1;
}
'''
sveElemCountInst('sqdec', 'Sqdec32', 'SimdAluOp', signedTypes,
sqdecCode%{'dstType':'int32_t'}, destType = DestType.Scalar,
dstIs32b = True)
# SQDECB, SQDECH, SQDECW, SQDECD (scalar, 64-bit)
sveElemCountInst('sqdec', 'Sqdec', 'SimdAluOp', signedTypes,
sqdecCode%{'dstType':'int64_t'}, destType = DestType.Scalar,
dstIs32b = False)
# SQDECH, SQDECW, SQDECD (vector)
sveElemCountInst('sqdec', 'Sqdecv', 'SimdAluOp', bigSignedTypes,
sqdecCode%{'dstType':'Element'}, destType = DestType.Vector,
dstIs32b = False)
# SQDECP (scalar, 32-bit)
sqdecpCode = '''
destElem = srcElem - count;
bool negDest = (destElem < 0);
bool negSrc = (srcElem < 0);
bool posCount = (count >= 0);
if ((negDest != negSrc) && (negSrc == posCount)) {
destElem = std::numeric_limits<%s>::min();
if (negDest)
destElem -= 1;
}
'''
sqdecp32Code = '''
int32_t srcElem = WDest;
int32_t destElem;''' + (sqdecpCode % 'int32_t') + '''
if (destElem < 0) {
XDest = static_cast<uint32_t>(destElem) | ~mask(32);
} else {
XDest = destElem;
}
'''
svePredCountInst('sqdecp', 'Sqdecp32', 'SimdAluOp', signedTypes,
sqdecp32Code, DestType.Scalar, SrcSize.Src32bit)
# SQDECP (scalar, 64-bit)
sqdecp64Code = '''
int64_t srcElem = XDest;
int64_t destElem;''' + (sqdecpCode % 'int64_t') + '''
XDest = destElem;
'''
svePredCountInst('sqdecp', 'Sqdecp64', 'SimdAluOp', signedTypes,
sqdecp64Code, DestType.Scalar, SrcSize.Src64bit)
# SQDECP (vector)
svePredCountInst('sqdecp', 'Sqdecpv', 'SimdAluOp', signedTypes,
sqdecpCode % 'Element', DestType.Vector)
# SQINCB, SQINCH, SQINCW, SQINCD (scalar, 32-bit)
sqincCode = '''
destElem = srcElem1 + (count * imm);
bool negDest = (destElem < 0);
bool negSrc = (srcElem1 < 0);
bool negCount = ((count * imm) < 0);
if ((negDest != negSrc) && (negSrc == negCount)) {
destElem = static_cast<%(dstType)s>(
(%(dstType)s)1 << (sizeof(%(dstType)s) * 8 - 1)
);
if (negDest)
destElem -= 1;
}
'''
sveElemCountInst('sqinc', 'Sqinc32', 'SimdAluOp', signedTypes,
sqincCode%{'dstType':'int32_t'}, destType = DestType.Scalar,
dstIs32b = True)
# SQINCB, SQINCH, SQINCW, SQINCD (scalar, 64-bit)
sveElemCountInst('sqinc', 'Sqinc', 'SimdAluOp', signedTypes,
sqincCode%{'dstType':'int64_t'}, destType = DestType.Scalar,
dstIs32b = False)
# SQINCH, SQINCW, SQINCD (vector)
sveElemCountInst('sqinc', 'Sqincv', 'SimdAluOp', bigSignedTypes,
sqincCode%{'dstType':'Element'}, destType = DestType.Vector,
dstIs32b = False)
# SQINCP (scalar, 32-bit)
sqincpCode = '''
destElem = srcElem + count;
bool negDest = (destElem < 0);
bool negSrc = (srcElem < 0);
bool negCount = (count < 0);
if ((negDest != negSrc) && (negSrc == negCount)) {
destElem = std::numeric_limits<%s>::min();
if (negDest)
destElem -= 1;
}
'''
sqincp32Code = '''
int32_t srcElem = WDest;
int32_t destElem;''' + (sqincpCode % 'int32_t') + '''
if (destElem < 0) {
XDest = static_cast<uint32_t>(destElem) | ~mask(32);
} else {
XDest = destElem;
}
'''
svePredCountInst('sqincp', 'Sqincp32', 'SimdAluOp', signedTypes,
sqincp32Code, DestType.Scalar, SrcSize.Src32bit)
# SQINCP (scalar, 64-bit)
sqincp64Code = '''
int64_t srcElem = XDest;
int64_t destElem;''' + (sqincpCode % 'int64_t') + '''
XDest = destElem;
'''
svePredCountInst('sqincp', 'Sqincp64', 'SimdAluOp', signedTypes,
sqincp64Code, DestType.Scalar, SrcSize.Src64bit)
# SQINCP (vector)
svePredCountInst('sqincp', 'Sqincpv', 'SimdAluOp', signedTypes,
sqincpCode % 'Element', DestType.Vector)
# SQSUB (immediate)
sqsubCode = '''
destElem = srcElem1 - srcElem2;
bool negDest = (destElem < 0);
bool negSrc1 = (srcElem1 < 0);
bool posSrc2 = (srcElem2 >= 0);
if ((negDest != negSrc1) && (negSrc1 == posSrc2)) {
destElem = static_cast<Element>(
(Element)1 << (sizeof(Element) * 8 - 1)
);
if (negDest)
destElem -= 1;
}
'''
sveWideImmInst('sqsub', 'SqsubImm', 'SimdAddOp', signedTypes, sqsubCode)
# SQSUB (vectors)
sveBinInst('sqsub', 'Sqsub', 'SimdAddOp', signedTypes, sqsubCode)
# SUB (immediate)
subCode = 'destElem = srcElem1 - srcElem2;'
sveWideImmInst('sub', 'SubImm', 'SimdAddOp', unsignedTypes, subCode)
# SUB (vectors, predicated)
sveBinInst('sub', 'SubPred', 'SimdAddOp', unsignedTypes, subCode,
PredType.MERGE, True)
# SUB (vectors, unpredicated)
subCode = 'destElem = srcElem1 - srcElem2;'
sveBinInst('sub', 'SubUnpred', 'SimdAddOp', unsignedTypes, subCode)
# SUBR (immediate)
subrCode = 'destElem = srcElem2 - srcElem1;'
sveWideImmInst('subr', 'SubrImm', 'SimdAddOp', unsignedTypes, subrCode)
# SUBR (vectors)
sveBinInst('subr', 'Subr', 'SimdAddOp', unsignedTypes, subrCode,
PredType.MERGE, True)
# SUNPKHI
sveUnpackInst('sunpkhi', 'Sunpkhi', 'SimdAluOp', signedWideSDTypes,
unpackHalf = Unpack.High, regType = SrcRegType.Vector)
# SUNPKLO
sveUnpackInst('sunpklo', 'Sunpklo', 'SimdAluOp', signedWideSDTypes,
unpackHalf = Unpack.Low, regType = SrcRegType.Vector)
# SXTB
sxtCode = 'destElem = sext<8 * sizeof(SElement)>(srcElem1);'
sveWidenUnaryInst('sxtb', 'Sxtb', 'SimdAluOp',
['uint8_t, uint16_t', 'uint8_t, uint32_t', 'uint8_t, uint64_t'],
sxtCode, PredType.MERGE)
# SXTH
sveWidenUnaryInst('sxth', 'Sxth', 'SimdAluOp',
['uint16_t, uint32_t', 'uint16_t, uint64_t'],
sxtCode, PredType.MERGE)
# SXTW
sveWidenUnaryInst('sxtw', 'Sxtw', 'SimdAluOp',
['uint32_t, uint64_t'],
sxtCode, PredType.MERGE)
# TBL
sveTblInst('tbl', 'Tbl', 'SimdAluOp')
# TRN1, TRN2 (predicates)
trnPredIterCode = '''
constexpr unsigned sz = sizeof(Element);
int s;
int part = %d;
ArmISA::VecPredRegContainer tmpPredC;
auto auxPDest = tmpPredC.as<uint8_t>();
for (unsigned i = 0; i < eCount / 2; i++) {
s = 2 * i + part;
for (unsigned j = 0; j < sz; j++) {
auxPDest[(2 * i) * sz + j] = POp1_pb[s * sz + j];
auxPDest[(2 * i + 1) * sz + j] = POp2_pb[s * sz + j];
}
}
for (unsigned i = 0; i < eCount * sz; i++) {
PDest_pb[i] = auxPDest[i];
}
'''
svePredBinPermInst('trn1', 'Trn1Pred', 'SimdPredAluOp', unsignedTypes,
trnPredIterCode % 0)
svePredBinPermInst('trn2', 'Trn2Pred', 'SimdPredAluOp', unsignedTypes,
trnPredIterCode % 1)
# TRN1, TRN2 (vectors)
trnIterCode = '''
int s;
int part = %d;
ArmISA::VecRegContainer tmpVecC;
auto auxDest = tmpVecC.as<Element>();
for (unsigned i = 0; i < eCount / 2; i++) {
s = 2 * i + part;
auxDest[2 * i] = AA64FpOp1_x[s];
auxDest[2 * i + 1] = AA64FpOp2_x[s];
}
for (unsigned i = 0; i < eCount; i++) {
AA64FpDest_x[i] = auxDest[i];
}
'''
sveBinInst('trn1', 'Trn1', 'SimdAluOp', unsignedTypes, '',
customIterCode=trnIterCode % 0)
sveBinInst('trn2', 'Trn2', 'SimdAluOp', unsignedTypes, '',
customIterCode=trnIterCode % 1)
# UABD
sveBinInst('uabd', 'Uabd', 'SimdAddOp', unsignedTypes, abdCode,
PredType.MERGE, True)
# UADDV
sveWideningAssocReducInst('uaddv', 'Uaddv', 'SimdReduceAddOp',
['uint8_t, uint64_t', 'uint16_t, uint64_t', 'uint32_t, uint64_t',
'uint64_t, uint64_t'],
addvCode, '0')
# UCVTF
ucvtfCode = fpOp % ('fplibFixedToFP<DElement>(srcElem1, 0, true,'
' FPCRRounding(fpscr), fpscr)')
sveCvtInst('ucvtf', 'UcvtfNarrow', 'SimdCvtOp',
('uint16_t, uint16_t',
'uint32_t, uint16_t',
'uint64_t, uint16_t',
'uint32_t, uint32_t',
'uint64_t, uint32_t',
'uint64_t, uint64_t'),
ucvtfCode, CvtDir.Narrow)
sveCvtInst('ucvtf', 'UcvtfWiden', 'SimdCvtOp', ('uint32_t, uint64_t',),
ucvtfCode, CvtDir.Widen)
# UDIV
udivCode = 'destElem = (srcElem2 == 0) ? 0 : (srcElem1 / srcElem2);'
sveBinInst('udiv', 'Udiv', 'SimdDivOp', unsignedTypes, udivCode,
PredType.MERGE, True)
# UDIVR
udivrCode = 'destElem = (srcElem1 == 0) ? 0 : (srcElem2 / srcElem1);'
sveBinInst('udivr', 'Udivr', 'SimdDivOp', unsignedTypes, udivrCode,
PredType.MERGE, True)
# UDOT (indexed)
sveDotInst('udot', 'Udoti', 'SimdAluOp', ['uint8_t, uint32_t',
'uint16_t, uint64_t'], isIndexed = True)
# UDOT (vectors)
sveDotInst('udot', 'Udotv', 'SimdAluOp', ['uint8_t, uint32_t',
'uint16_t, uint64_t'], isIndexed = False)
# UMAX (immediate)
sveWideImmInst('umax', 'UmaxImm', 'SimdCmpOp', unsignedTypes, maxCode)
# UMAX (vectors)
sveBinInst('umax', 'Umax', 'SimdCmpOp', unsignedTypes, maxCode,
PredType.MERGE, True)
# UMAXV
sveAssocReducInst('umaxv', 'Umaxv', 'SimdReduceCmpOp', unsignedTypes,
maxvCode, 'std::numeric_limits<Element>::min()')
# UMIN (immediate)
sveWideImmInst('umin', 'UminImm', 'SimdCmpOp', unsignedTypes, minCode)
# UMIN (vectors)
sveBinInst('umin', 'Umin', 'SimdCmpOp', unsignedTypes, minCode,
PredType.MERGE, True)
# UMINV
sveAssocReducInst('uminv', 'Uminv', 'SimdReduceCmpOp', unsignedTypes,
minvCode, 'std::numeric_limits<Element>::max()')
# UMULH
sveBinInst('umulh', 'Umulh', 'SimdMultOp', unsignedTypes, mulhCode,
PredType.MERGE, True)
# UQADD (immediate)
uqaddCode = '''
destElem = srcElem1 + srcElem2;
if (destElem < srcElem1 || destElem < srcElem2) {
destElem = (Element)(-1);
}
'''
sveWideImmInst('uqadd', 'UqaddImm', 'SimdAddOp', unsignedTypes, uqaddCode)
# UQADD (vectors)
sveBinInst('uqadd', 'Uqadd', 'SimdAddOp', unsignedTypes, uqaddCode)
# UQDECB, UQDECH, UQDECW, UQDECD (scalar, 32-bit)
uqdecCode = '''
destElem = srcElem1 - (imm * count);
if (destElem > srcElem1) {
destElem = 0;
}
'''
sveElemCountInst('uqdec', 'Uqdec32', 'SimdAluOp', unsignedTypes,
uqdecCode, destType = DestType.Scalar, dstIs32b = True)
# UQDECB, UQDECH, UQDECW, UQDECD (scalar, 64-bit)
sveElemCountInst('uqdec', 'Uqdec', 'SimdAluOp', unsignedTypes,
uqdecCode, destType = DestType.Scalar, dstIs32b = False)
# UQDECH, UQDECW, UQDECD (vector)
sveElemCountInst('uqdec', 'Uqdecv', 'SimdAluOp', bigUnsignedTypes,
uqdecCode, destType = DestType.Vector, dstIs32b = False)
# UQDECP (scalar, 32-bit)
uqdecpCode = '''
destElem = srcElem - count;
if (destElem > srcElem) {
destElem = 0;
}
'''
uqdecp32Code = '''
uint32_t srcElem = WDest;
uint32_t destElem;''' + uqdecpCode + '''
WDest = destElem;
'''
svePredCountInst('uqdecp', 'Uqdecp32', 'SimdAluOp', unsignedTypes,
uqdecp32Code, DestType.Scalar, SrcSize.Src32bit)
# UQDECP (scalar, 64-bit)
uqdecp64Code = '''
uint64_t srcElem = XDest;
uint64_t destElem;''' + uqdecpCode + '''
XDest = destElem;
'''
svePredCountInst('uqdecp', 'Uqdecp64', 'SimdAluOp', unsignedTypes,
uqdecp64Code, DestType.Scalar, SrcSize.Src64bit)
# UQDECP (vector)
svePredCountInst('uqdecp', 'Uqdecpv', 'SimdAluOp', unsignedTypes,
uqdecpCode, DestType.Vector)
# UQDECB, UQDECH, UQDECW, UQDECD (scalar, 32-bit)
uqincCode = '''
destElem = srcElem1 + (imm * count);
if (destElem < srcElem1 || destElem < (imm * count)) {
destElem = static_cast<%(destType)s>(-1);
}
'''
sveElemCountInst('uqinc', 'Uqinc32', 'SimdAluOp', unsignedTypes,
uqincCode%{'destType': 'uint32_t'}, destType = DestType.Scalar,
dstIs32b = True)
# UQDECB, UQDECH, UQDECW, UQDECD (scalar, 64-bit)
sveElemCountInst('uqinc', 'Uqinc', 'SimdAluOp', unsignedTypes,
uqincCode%{'destType': 'uint64_t'}, destType = DestType.Scalar,
dstIs32b = False)
# UQDECH, UQDECW, UQDECD (vector)
sveElemCountInst('uqinc', 'Uqincv', 'SimdAluOp', bigUnsignedTypes,
uqincCode%{'destType': 'Element'}, destType = DestType.Vector,
dstIs32b = False)
# UQINCP (scalar, 32-bit)
uqincpCode = '''
destElem = srcElem + count;
if (destElem < srcElem || destElem < count) {
destElem = std::numeric_limits<%s>::max();
}
'''
uqincp32Code = '''
uint32_t srcElem = WDest;
uint32_t destElem;''' + (uqincpCode % 'uint32_t') + '''
XDest = destElem;
'''
svePredCountInst('uqincp', 'Uqincp32', 'SimdAluOp', unsignedTypes,
uqincp32Code, DestType.Scalar, SrcSize.Src32bit)
# UQINCP (scalar, 64-bit)
uqincp64Code = '''
uint64_t srcElem = XDest;
uint64_t destElem;''' + (uqincpCode % 'uint64_t') + '''
XDest = destElem;
'''
svePredCountInst('uqincp', 'Uqincp64', 'SimdAluOp', unsignedTypes,
uqincp64Code, DestType.Scalar, SrcSize.Src64bit)
# UQINCP (vector)
svePredCountInst('uqincp', 'Uqincpv', 'SimdAluOp', unsignedTypes,
uqincpCode % 'Element', DestType.Vector)
# UQSUB (immediate)
uqsubCode = '''
destElem = srcElem1 - srcElem2;
if (destElem > srcElem1) {
destElem = 0;
}
'''
sveWideImmInst('uqsub', 'UqsubImm', 'SimdAddOp', unsignedTypes, uqsubCode)
# UQSUB (vectors)
sveBinInst('uqsub', 'Uqsub', 'SimdAddOp', unsignedTypes, uqsubCode)
# UUNPKHI
sveUnpackInst('uunpkhi', 'Uunpkhi', 'SimdAluOp', unsignedWideSDTypes,
unpackHalf = Unpack.High, regType = SrcRegType.Vector)
# UUNPKLO
sveUnpackInst('uunpklo', 'Uunpklo', 'SimdAluOp', unsignedWideSDTypes,
unpackHalf = Unpack.Low, regType = SrcRegType.Vector)
# UXTB
uxtCode = 'destElem = srcElem1;'
sveWidenUnaryInst('uxtb', 'Uxtb', 'SimdAluOp',
['uint8_t, uint16_t', 'uint8_t, uint32_t', 'uint8_t, uint64_t'],
uxtCode, PredType.MERGE)
# UXTH
sveWidenUnaryInst('uxth', 'Uxth', 'SimdAluOp',
['uint16_t, uint32_t', 'uint16_t, uint64_t'],
uxtCode, PredType.MERGE)
# UXTW
sveWidenUnaryInst('uxtw', 'Uxtw', 'SimdAluOp',
['uint32_t, uint64_t'],
uxtCode, PredType.MERGE)
# UZP1, UZP2 (predicates)
uzpPredIterCode = '''
constexpr unsigned sz = sizeof(Element);
int s;
int part = %d;
ArmISA::VecPredRegContainer tmpPredC;
auto auxPDest = tmpPredC.as<uint8_t>();
for (unsigned i = 0; i < eCount; i++) {
s = 2 * i + part;
for (unsigned j = 0; j < sz; j++) {
if (s < eCount) {
auxPDest[i * sz + j] = POp1_pb[s * sz + j];
} else {
auxPDest[i * sz + j] = POp2_pb[(s - eCount) * sz + j];
}
}
}
for (unsigned i = 0; i < eCount * sz; i++) {
PDest_pb[i] = auxPDest[i];
}
'''
svePredBinPermInst('uzp1', 'Uzp1Pred', 'SimdPredAluOp', unsignedTypes,
uzpPredIterCode % 0)
svePredBinPermInst('uzp2', 'Uzp2Pred', 'SimdPredAluOp', unsignedTypes,
uzpPredIterCode % 1)
# UZP1, UZP2 (vectors)
uzpIterCode = '''
int s;
int part = %d;
ArmISA::VecRegContainer tmpVecC;
auto auxDest = tmpVecC.as<Element>();
for (unsigned i = 0; i < eCount; i++) {
s = 2 * i + part;
if (s < eCount) {
auxDest[i] = AA64FpOp1_x[s];
} else {
auxDest[i] = AA64FpOp2_x[s - eCount];
}
}
for (unsigned i = 0; i < eCount; i++) {
AA64FpDest_x[i] = auxDest[i];
}
'''
sveBinInst('uzp1', 'Uzp1', 'SimdAluOp', unsignedTypes, '',
customIterCode=uzpIterCode % 0)
sveBinInst('uzp2', 'Uzp2', 'SimdAluOp', unsignedTypes, '',
customIterCode=uzpIterCode % 1)
# WHILELE (32-bit)
whileLECode = '''
cond = srcElem1 <= srcElem2;
'''
sveWhileInst('whilele', 'Whilele32', 'SimdCmpOp', signedTypes, whileLECode,
SrcSize.Src32bit)
# WHILELE (64-bit)
sveWhileInst('whilele', 'Whilele64', 'SimdCmpOp', signedTypes, whileLECode,
SrcSize.Src64bit)
# WHILELO (32-bit)
whileLTCode = '''
cond = srcElem1 < srcElem2;
'''
sveWhileInst('whilelo', 'Whilelo32', 'SimdCmpOp', unsignedTypes,
whileLTCode, SrcSize.Src32bit)
# WHILELO (64-bit)
sveWhileInst('whilelo', 'Whilelo64', 'SimdCmpOp', unsignedTypes,
whileLTCode, SrcSize.Src64bit)
# WHILELS (32-bit)
sveWhileInst('whilels', 'Whilels32', 'SimdCmpOp', unsignedTypes,
whileLECode, SrcSize.Src32bit)
# WHILELS (64-bit)
sveWhileInst('whilels', 'Whilels64', 'SimdCmpOp', unsignedTypes,
whileLECode, SrcSize.Src64bit)
# WHILELT (32-bit)
sveWhileInst('whilelt', 'Whilelt32', 'SimdCmpOp', signedTypes,
whileLTCode, SrcSize.Src32bit)
# WHILELT (64-bit)
sveWhileInst('whilelt', 'Whilelt64', 'SimdCmpOp', signedTypes,
whileLTCode, SrcSize.Src64bit)
# WRFFR
wrffrCode = '''
unsigned eCount = ArmStaticInst::getCurSveVecLen<uint8_t>(
xc->tcBase());
for (unsigned i = 0; i < eCount; i++) {
Ffr_ub[i] = POp1_ub[i];
}'''
svePredWriteFfrInst('wrffr', 'Wrffr', 'SimdPredAluOp', wrffrCode, False)
# ZIP1, ZIP2 (predicates)
zipPredIterCode = '''
constexpr unsigned sz = sizeof(Element);
int s;
int part = %d;
ArmISA::VecPredRegContainer tmpPredC;
auto auxPDest = tmpPredC.as<uint8_t>();
for (unsigned i = 0; i < eCount / 2; i++) {
s = i + (part * (eCount / 2));
for (unsigned j = 0; j < sz; j++) {
auxPDest[(2 * i) * sz + j] = POp1_pb[s * sz + j];
auxPDest[(2 * i + 1) * sz + j] = POp2_pb[s * sz + j];
}
}
for (unsigned i = 0; i < eCount * sz; i++) {
PDest_pb[i] = auxPDest[i];
}
'''
svePredBinPermInst('zip1', 'Zip1Pred', 'SimdPredAluOp', unsignedTypes,
zipPredIterCode % 0)
svePredBinPermInst('zip2', 'Zip2Pred', 'SimdPredAluOp', unsignedTypes,
zipPredIterCode % 1)
# ZIP1, ZIP2 (vectors)
zipIterCode = '''
int s;
int part = %d;
ArmISA::VecRegContainer tmpVecC;
auto auxDest = tmpVecC.as<Element>();
for (unsigned i = 0; i < eCount / 2; i++) {
s = i + (part * (eCount / 2));
auxDest[2 * i] = AA64FpOp1_x[s];
auxDest[2 * i + 1] = AA64FpOp2_x[s];
}
for (unsigned i = 0; i < eCount; i++) {
AA64FpDest_x[i] = auxDest[i];
}
'''
sveBinInst('zip1', 'Zip1', 'SimdAluOp', unsignedTypes, '',
customIterCode=zipIterCode % 0)
sveBinInst('zip2', 'Zip2', 'SimdAluOp', unsignedTypes, '',
customIterCode=zipIterCode % 1)
}};