void genCodeForIndexAddr(GenTreeIndexAddr* tree);
void genCodeForIndir(GenTreeIndir* tree);
void genCodeForNegNot(GenTree* tree);
+ void genCodeForBswap(GenTree* tree);
void genCodeForLclVar(GenTreeLclVar* tree);
void genCodeForLclFld(GenTreeLclFld* tree);
void genCodeForStoreLclFld(GenTreeLclFld* tree);
genProduceReg(tree);
}
+//------------------------------------------------------------------------
+// genCodeForBswap: Produce code for a GT_BSWAP / GT_BSWAP16 node.
+//
+// Arguments:
+// tree - the node
+//
+void CodeGen::genCodeForBswap(GenTree* tree)
+{
+ // TODO: If we're swapping immediately after a read from memory or immediately before
+ // a write to memory, use the MOVBE instruction instead of the BSWAP instruction if
+ // the platform supports it.
+
+ assert(tree->OperIs(GT_BSWAP, GT_BSWAP16));
+
+ regNumber targetReg = tree->gtRegNum;
+ var_types targetType = tree->TypeGet();
+
+ GenTree* operand = tree->gtGetOp1();
+ assert(operand->isUsedFromReg());
+ regNumber operandReg = genConsumeReg(operand);
+
+ if (operandReg != targetReg)
+ {
+ inst_RV_RV(INS_mov, targetReg, operandReg, targetType);
+ }
+
+ if (tree->OperIs(GT_BSWAP))
+ {
+ // 32-bit and 64-bit byte swaps use "bswap reg"
+ inst_RV(INS_bswap, targetReg, targetType);
+ }
+ else
+ {
+ // 16-bit byte swaps use "ror reg.16, 8"
+ inst_RV_IV(INS_ror_N, targetReg, 8 /* val */, emitAttr::EA_2BYTE);
+ }
+
+ genProduceReg(tree);
+}
+
// Generate code to get the high N bits of a N*N=2N bit multiplication result
void CodeGen::genCodeForMulHi(GenTreeOp* treeNode)
{
genCodeForNegNot(treeNode);
break;
+ case GT_BSWAP:
+ case GT_BSWAP16:
+ genCodeForBswap(treeNode);
+ break;
+
case GT_DIV:
if (varTypeIsFloating(treeNode->TypeGet()))
{
// Standard unary operators
case GT_NOT:
case GT_NEG:
+ case GT_BSWAP:
+ case GT_BSWAP16:
case GT_COPY:
case GT_RELOAD:
case GT_ARR_LENGTH:
case GT_STORE_LCL_FLD:
case GT_NOT:
case GT_NEG:
+ case GT_BSWAP:
+ case GT_BSWAP16:
case GT_COPY:
case GT_RELOAD:
case GT_ARR_LENGTH:
dst += emitOutputByte(dst, code);
break;
+ case INS_bswap:
+ {
+ assert(size >= EA_4BYTE && size <= EA_PTRSIZE); // 16-bit BSWAP is undefined
+
+ // The Intel instruction set reference for BSWAP states that extended registers
+ // should be enabled via REX.R, but per Vol. 2A, Sec. 2.2.1.2 (see also Figure 2-7),
+ // REX.B should instead be used if the register is encoded in the opcode byte itself.
+ // Therefore the default logic of insEncodeReg012 is correct for this case.
+
+ code = insCodeRR(ins);
+
+ if (TakesRexWPrefix(ins, size))
+ {
+ code = AddRexWPrefix(ins, code);
+ }
+
+ // Register...
+ unsigned regcode = insEncodeReg012(ins, reg, size, &code);
+
+ // Output the REX prefix
+ dst += emitOutputRexOrVexPrefixIfNeeded(ins, dst, code);
+
+ dst += emitOutputWord(dst, code | (regcode << 8));
+ break;
+ }
+
case INS_seto:
case INS_setno:
case INS_setb:
case GT_NOP:
case GT_RETURN:
case GT_RETFILT:
+ case GT_BSWAP:
+ case GT_BSWAP16:
if (def == this->AsUnOp()->gtOp1)
{
*use = &this->AsUnOp()->gtOp1;
case GT_NULLCHECK:
case GT_PUTARG_REG:
case GT_PUTARG_STK:
+ case GT_BSWAP:
+ case GT_BSWAP16:
#if FEATURE_ARG_SPLIT
case GT_PUTARG_SPLIT:
#endif // FEATURE_ARG_SPLIT
i1 = -i1;
break;
+ case GT_BSWAP:
+ i1 = ((i1 >> 24) & 0xFF) | ((i1 >> 8) & 0xFF00) | ((i1 << 8) & 0xFF0000) |
+ ((i1 << 24) & 0xFF000000);
+ break;
+
+ case GT_BSWAP16:
+ i1 = ((i1 >> 8) & 0xFF) | ((i1 << 8) & 0xFF00);
+ break;
+
case GT_CAST:
// assert (genActualType(tree->CastToType()) == tree->gtType);
switch (tree->CastToType())
lval1 = -lval1;
break;
+ case GT_BSWAP:
+ lval1 = ((lval1 >> 56) & 0xFF) | ((lval1 >> 40) & 0xFF00) | ((lval1 >> 24) & 0xFF0000) |
+ ((lval1 >> 8) & 0xFF000000) | ((lval1 << 8) & 0xFF00000000) |
+ ((lval1 << 24) & 0xFF0000000000) | ((lval1 << 40) & 0xFF000000000000) |
+ ((lval1 << 56) & 0xFF00000000000000);
+ break;
+
case GT_CAST:
assert(genActualType(tree->CastToType()) == tree->gtType);
switch (tree->CastToType())
GTNODE(RUNTIMELOOKUP , GenTreeRuntimeLookup, 0,GTK_UNOP|GTK_EXOP) // Runtime handle lookup
+GTNODE(BSWAP , GenTreeOp ,0,GTK_UNOP) // Byte swap (32-bit or 64-bit)
+GTNODE(BSWAP16 , GenTreeOp ,0,GTK_UNOP) // Byte swap (16-bit)
+
//-----------------------------------------------------------------------------
// Binary operators (2 operands):
//-----------------------------------------------------------------------------
break;
}
+ case NI_System_Buffers_Binary_BinaryPrimitives_ReverseEndianness:
+ {
+ assert(sig->numArgs == 1);
+
+ // We expect the return type of the ReverseEndianness routine to match the type of the
+ // one and only argument to the method. We use a special instruction for 16-bit
+ // BSWAPs since on x86 processors this is implemented as ROR <16-bit reg>, 8. Additionally,
+ // we only emit 64-bit BSWAP instructions on 64-bit archs; if we're asked to perform a
+ // 64-bit byte swap on a 32-bit arch, we'll fall to the default case in the switch block below.
+
+ switch (sig->retType)
+ {
+ case CorInfoType::CORINFO_TYPE_SHORT:
+ case CorInfoType::CORINFO_TYPE_USHORT:
+ retNode = gtNewOperNode(GT_BSWAP16, callType, impPopStack().val);
+ break;
+
+ case CorInfoType::CORINFO_TYPE_INT:
+ case CorInfoType::CORINFO_TYPE_UINT:
+#ifdef _TARGET_64BIT_
+ case CorInfoType::CORINFO_TYPE_LONG:
+ case CorInfoType::CORINFO_TYPE_ULONG:
+#endif // _TARGET_64BIT_
+ retNode = gtNewOperNode(GT_BSWAP, callType, impPopStack().val);
+ break;
+
+ default:
+ // This default case gets hit on 32-bit archs when a call to a 64-bit overload
+ // of ReverseEndianness is encountered. In that case we'll let JIT treat this as a standard
+ // method call, where the implementation decomposes the operation into two 32-bit
+ // bswap routines. If the input to the 64-bit function is a constant, then we rely
+ // on inlining + constant folding of 32-bit bswaps to effectively constant fold
+ // the 64-bit call site.
+ break;
+ }
+
+ break;
+ }
+
default:
break;
}
result = NI_Math_Round;
}
}
+#if defined(_TARGET_XARCH_) // We currently only support BSWAP on x86
+ else if (strcmp(namespaceName, "System.Buffers.Binary") == 0)
+ {
+ if ((strcmp(className, "BinaryPrimitives") == 0) && (strcmp(methodName, "ReverseEndianness") == 0))
+ {
+ result = NI_System_Buffers_Binary_BinaryPrimitives_ReverseEndianness;
+ }
+ }
+#endif // !defined(_TARGET_XARCH_)
else if (strcmp(namespaceName, "System.Collections.Generic") == 0)
{
if ((strcmp(className, "EqualityComparer`1") == 0) && (strcmp(methodName, "get_Default") == 0))
INST5(dec, "dec", IUM_RW, 0x0008FE, BAD_CODE, BAD_CODE, BAD_CODE, 0x000048, INS_FLAGS_WritesFlags)
INST5(dec_l, "dec", IUM_RW, 0x0008FE, BAD_CODE, BAD_CODE, BAD_CODE, 0x00C8FE, INS_FLAGS_WritesFlags)
+// Multi-byte opcodes without modrm are represented in mixed endian fashion.
+// See comment around quarter way through this file for more information.
+INST5(bswap, "bswap", IUM_RW, 0x0F00C8, BAD_CODE, BAD_CODE, BAD_CODE, 0x00C80F, INS_FLAGS_None)
+
// id nm um mr mi rm a4 flags
INST4(add, "add", IUM_RW, 0x000000, 0x000080, 0x000002, 0x000004, INS_FLAGS_WritesFlags)
INST4(or, "or", IUM_RW, 0x000008, 0x000880, 0x00000A, 0x00000C, INS_FLAGS_WritesFlags)
enum NamedIntrinsic : unsigned short
{
- NI_Illegal = 0,
- NI_System_Enum_HasFlag = 1,
- NI_MathF_Round = 2,
- NI_Math_Round = 3,
- NI_System_Collections_Generic_EqualityComparer_get_Default = 4,
+ NI_Illegal = 0,
+ NI_System_Enum_HasFlag = 1,
+ NI_MathF_Round = 2,
+ NI_Math_Round = 3,
+ NI_System_Collections_Generic_EqualityComparer_get_Default = 4,
+ NI_System_Buffers_Binary_BinaryPrimitives_ReverseEndianness = 5,
#ifdef FEATURE_HW_INTRINSICS
NI_HW_INTRINSIC_START,
#if defined(_TARGET_XARCH_)
case GT_NEG:
case GT_NOT:
+ case GT_BSWAP:
+ case GT_BSWAP16:
case GT_CAST:
return true; // CSE these Unary Operators
--- /dev/null
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+// See the LICENSE file in the project root for more information.
+//
+
+using System;
+using System.Buffers.Binary;
+using Internal;
+
+namespace BinaryPrimitivesReverseEndianness
+{
+ class Program
+ {
+ public const int Pass = 100;
+ public const int Fail = 0;
+
+ private const ushort ConstantUInt16Input = 0x9876;
+ private const ushort ConstantUInt16Expected = 0x7698;
+
+ private const uint ConstantUInt32Input = 0x98765432;
+ private const uint ConstantUInt32Expected = 0x32547698;
+
+ private const ulong ConstantUInt64Input = 0xfedcba9876543210;
+ private const ulong ConstantUInt64Expected = 0x1032547698badcfe;
+
+
+ static int Main(string[] args)
+ {
+ /*
+ * CONST VALUE TESTS
+ */
+
+ ushort swappedUInt16 = BinaryPrimitives.ReverseEndianness(ConstantUInt16Input);
+ if (swappedUInt16 != ConstantUInt16Expected)
+ {
+ Console.WriteLine($"BinaryPrimitives.ReverseEndianness(const UInt16) failed.");
+ Console.WriteLine($"Input: 0x{ConstantUInt16Input:X4}");
+ Console.WriteLine($"Output: 0x{swappedUInt16:X4}");
+ Console.WriteLine($"Expected: 0x{ConstantUInt16Expected:X4}");
+ }
+
+ uint swappedUInt32 = BinaryPrimitives.ReverseEndianness(ConstantUInt32Input);
+ if (swappedUInt32 != ConstantUInt32Expected)
+ {
+ Console.WriteLine($"BinaryPrimitives.ReverseEndianness(const UInt32) failed.");
+ Console.WriteLine($"Input: 0x{ConstantUInt32Input:X8}");
+ Console.WriteLine($"Output: 0x{swappedUInt32:X8}");
+ Console.WriteLine($"Expected: 0x{ConstantUInt32Expected:X8}");
+ }
+
+ ulong swappedUInt64 = BinaryPrimitives.ReverseEndianness(ConstantUInt64Input);
+ if (swappedUInt64 != ConstantUInt64Expected)
+ {
+ Console.WriteLine($"BinaryPrimitives.ReverseEndianness(const UInt32) failed.");
+ Console.WriteLine($"Input: 0x{ConstantUInt64Input:X16}");
+ Console.WriteLine($"Output: 0x{swappedUInt64:X16}");
+ Console.WriteLine($"Expected: 0x{ConstantUInt64Expected:X16}");
+ }
+
+ /*
+ * NON-CONST VALUE TESTS
+ */
+
+ ushort nonConstUInt16Input = (ushort)DateTime.UtcNow.Ticks;
+ ushort nonConstUInt16Output = BinaryPrimitives.ReverseEndianness(nonConstUInt16Input);
+ ushort nonConstUInt16Expected = ByteSwapUInt16_Control(nonConstUInt16Input);
+ if (nonConstUInt16Output != nonConstUInt16Expected)
+ {
+ Console.WriteLine($"BinaryPrimitives.ReverseEndianness(non-const UInt16) failed.");
+ Console.WriteLine($"Input: 0x{nonConstUInt16Input:X4}");
+ Console.WriteLine($"Output: 0x{nonConstUInt16Output:X4}");
+ Console.WriteLine($"Expected: 0x{nonConstUInt16Expected:X4}");
+ }
+
+ uint nonConstUInt32Input = (uint)DateTime.UtcNow.Ticks;
+ uint nonConstUInt32Output = BinaryPrimitives.ReverseEndianness(nonConstUInt32Input);
+ uint nonConstUInt32Expected = ByteSwapUInt32_Control(nonConstUInt32Input);
+ if (nonConstUInt32Output != nonConstUInt32Expected)
+ {
+ Console.WriteLine($"BinaryPrimitives.ReverseEndianness(non-const UInt32) failed.");
+ Console.WriteLine($"Input: 0x{nonConstUInt32Input:X8}");
+ Console.WriteLine($"Output: 0x{nonConstUInt32Output:X8}");
+ Console.WriteLine($"Expected: 0x{nonConstUInt32Expected:X8}");
+ }
+
+ ulong nonConstUInt64Input = (ulong)DateTime.UtcNow.Ticks;
+ ulong nonConstUInt64Output = BinaryPrimitives.ReverseEndianness(nonConstUInt64Input);
+ ulong nonConstUInt64Expected = ByteSwapUInt64_Control(nonConstUInt64Input);
+ if (nonConstUInt64Output != nonConstUInt64Expected)
+ {
+ Console.WriteLine($"BinaryPrimitives.ReverseEndianness(non-const UInt64) failed.");
+ Console.WriteLine($"Input: 0x{nonConstUInt64Input:X16}");
+ Console.WriteLine($"Output: 0x{nonConstUInt64Output:X16}");
+ Console.WriteLine($"Expected: 0x{nonConstUInt64Expected:X16}");
+ }
+
+ return Pass;
+ }
+
+ private static ushort ByteSwapUInt16_Control(ushort value)
+ {
+ return (ushort)ByteSwapUnsigned_General(value, sizeof(ushort));
+ }
+
+ private static uint ByteSwapUInt32_Control(uint value)
+ {
+ return (uint)ByteSwapUnsigned_General(value, sizeof(uint));
+ }
+
+ private static ulong ByteSwapUInt64_Control(ulong value)
+ {
+ return (ulong)ByteSwapUnsigned_General(value, sizeof(ulong));
+ }
+
+ private static ulong ByteSwapUnsigned_General(ulong value, int width)
+ {
+ // A naive byte swap routine that works on integers of any arbitrary width.
+ // Width is specified in bytes.
+
+ ulong retVal = 0;
+ do
+ {
+ retVal = retVal << 8 | (byte)value;
+ value >>= 8;
+ } while (--width > 0);
+
+ if (value != 0)
+ {
+ // All bits of value should've been shifted out at this point.
+ throw new Exception("Unexpected data width specified - error in test program?");
+ }
+
+ return retVal;
+ }
+ }
+}
--- /dev/null
+<?xml version="1.0" encoding="utf-8"?>
+<Project ToolsVersion="12.0" DefaultTargets="Build" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
+ <Import Project="$([MSBuild]::GetDirectoryNameOfFileAbove($(MSBuildThisFileDirectory), dir.props))\dir.props" />
+ <PropertyGroup>
+ <Configuration Condition=" '$(Configuration)' == '' ">Debug</Configuration>
+ <Platform Condition=" '$(Platform)' == '' ">AnyCPU</Platform>
+ <SchemaVersion>2.0</SchemaVersion>
+ <ProjectGuid>{95DFC527-4DC1-495E-97D7-E94EE1F7140D}</ProjectGuid>
+ <OutputType>Exe</OutputType>
+ <ProjectTypeGuids>{786C830F-07A1-408B-BD7F-6EE04809D6DB};{FAE04EC0-301F-11D3-BF4B-00C04F79EFBC}</ProjectTypeGuids>
+ <SolutionDir Condition="$(SolutionDir) == '' Or $(SolutionDir) == '*Undefined*'">..\..\</SolutionDir>
+ <ReferenceSystemPrivateCoreLib>true</ReferenceSystemPrivateCoreLib>
+ </PropertyGroup>
+ <!-- Default configurations to help VS understand the configurations -->
+ <PropertyGroup Condition=" '$(Configuration)|$(Platform)' == 'Debug|AnyCPU' "></PropertyGroup>
+ <PropertyGroup Condition=" '$(Configuration)|$(Platform)' == 'Release|AnyCPU' " />
+ <ItemGroup>
+ <CodeAnalysisDependentAssemblyPaths Condition=" '$(VS100COMNTOOLS)' != '' " Include="$(VS100COMNTOOLS)..\IDE\PrivateAssemblies">
+ <Visible>False</Visible>
+ </CodeAnalysisDependentAssemblyPaths>
+ </ItemGroup>
+ <PropertyGroup>
+ <AllowUnsafeBlocks>true</AllowUnsafeBlocks>
+ <DebugType>None</DebugType>
+ <Optimize></Optimize>
+ </PropertyGroup>
+ <ItemGroup>
+ <Service Include="{82A7F48D-3B50-4B1E-B82E-3ADA8210C358}" />
+ </ItemGroup>
+ <ItemGroup>
+ <Compile Include="BinaryPrimitivesReverseEndianness.cs" />
+ </ItemGroup>
+ <Import Project="$([MSBuild]::GetDirectoryNameOfFileAbove($(MSBuildThisFileDirectory), dir.targets))\dir.targets" />
+ <PropertyGroup Condition=" '$(MsBuildProjectDirOverride)' != '' "></PropertyGroup>
+</Project>
--- /dev/null
+<?xml version="1.0" encoding="utf-8"?>
+<Project ToolsVersion="12.0" DefaultTargets="Build" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
+ <Import Project="$([MSBuild]::GetDirectoryNameOfFileAbove($(MSBuildThisFileDirectory), dir.props))\dir.props" />
+ <PropertyGroup>
+ <Configuration Condition=" '$(Configuration)' == '' ">Debug</Configuration>
+ <Platform Condition=" '$(Platform)' == '' ">AnyCPU</Platform>
+ <SchemaVersion>2.0</SchemaVersion>
+ <ProjectGuid>{95DFC527-4DC1-495E-97D7-E94EE1F7140D}</ProjectGuid>
+ <OutputType>Exe</OutputType>
+ <ProjectTypeGuids>{786C830F-07A1-408B-BD7F-6EE04809D6DB};{FAE04EC0-301F-11D3-BF4B-00C04F79EFBC}</ProjectTypeGuids>
+ <SolutionDir Condition="$(SolutionDir) == '' Or $(SolutionDir) == '*Undefined*'">..\..\</SolutionDir>
+ <ReferenceSystemPrivateCoreLib>true</ReferenceSystemPrivateCoreLib>
+ </PropertyGroup>
+ <!-- Default configurations to help VS understand the configurations -->
+ <PropertyGroup Condition=" '$(Configuration)|$(Platform)' == 'Debug|AnyCPU' "></PropertyGroup>
+ <PropertyGroup Condition=" '$(Configuration)|$(Platform)' == 'Release|AnyCPU' " />
+ <ItemGroup>
+ <CodeAnalysisDependentAssemblyPaths Condition=" '$(VS100COMNTOOLS)' != '' " Include="$(VS100COMNTOOLS)..\IDE\PrivateAssemblies">
+ <Visible>False</Visible>
+ </CodeAnalysisDependentAssemblyPaths>
+ </ItemGroup>
+ <PropertyGroup>
+ <AllowUnsafeBlocks>true</AllowUnsafeBlocks>
+ <DebugType>None</DebugType>
+ <Optimize>True</Optimize>
+ </PropertyGroup>
+ <ItemGroup>
+ <Service Include="{82A7F48D-3B50-4B1E-B82E-3ADA8210C358}" />
+ </ItemGroup>
+ <ItemGroup>
+ <Compile Include="BinaryPrimitivesReverseEndianness.cs" />
+ </ItemGroup>
+ <Import Project="$([MSBuild]::GetDirectoryNameOfFileAbove($(MSBuildThisFileDirectory), dir.targets))\dir.targets" />
+ <PropertyGroup Condition=" '$(MsBuildProjectDirOverride)' != '' "></PropertyGroup>
+</Project>
/// <summary>
/// Reverses a primitive value - performs an endianness swap
/// </summary>
+ [Intrinsic]
[MethodImpl(MethodImplOptions.AggressiveInlining)]
- public static short ReverseEndianness(short value)
- {
- return (short)((value & 0x00FF) << 8 | (value & 0xFF00) >> 8);
- }
+ public static short ReverseEndianness(short value) => (short)ReverseEndianness((ushort)value);
/// <summary>
/// Reverses a primitive value - performs an endianness swap
/// </summary>
+ [Intrinsic]
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public static int ReverseEndianness(int value) => (int)ReverseEndianness((uint)value);
/// <summary>
/// Reverses a primitive value - performs an endianness swap
/// </summary>
+ [Intrinsic]
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public static long ReverseEndianness(long value) => (long)ReverseEndianness((ulong)value);
/// rather than having to skip byte fields.
/// </summary>
[MethodImpl(MethodImplOptions.AggressiveInlining)]
- public static byte ReverseEndianness(byte value)
+ public static byte ReverseEndianness(byte value)
{
return value;
}
/// Reverses a primitive value - performs an endianness swap
/// </summary>
[CLSCompliant(false)]
+ [Intrinsic]
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public static ushort ReverseEndianness(ushort value)
{
/// Reverses a primitive value - performs an endianness swap
/// </summary>
[CLSCompliant(false)]
+ [Intrinsic]
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public static uint ReverseEndianness(uint value)
{
/// Reverses a primitive value - performs an endianness swap
/// </summary>
[CLSCompliant(false)]
+ [Intrinsic]
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public static ulong ReverseEndianness(ulong value)
{