ryujinx/ARMeilleure/CodeGen/X86/CodeGenContext.cs
FICTURE7 a9343c9364
Refactor PtcInfo (#2625)
* Refactor `PtcInfo`

This change reduces the coupling of `PtcInfo` by moving relocation
tracking to the backend. `RelocEntry`s remains as `RelocEntry`s through
out the pipeline until it actually needs to be written to the PTC
streams. Keeping this representation makes inspecting and manipulating
relocations after compilations less painful. This is something I needed
to do to patch relocations to 0 to diff dumps.

Contributes to #1125.

* Turn `Symbol` & `RelocInfo` into readonly structs

* Add documentation to `CompiledFunction`

* Remove `Compiler.Compile<T>`

Remove `Compiler.Compile<T>` and replace it by `Map<T>` of the
`CompiledFunction` returned.
2021-09-14 01:23:37 +02:00

323 lines
11 KiB
C#

using ARMeilleure.CodeGen.Linking;
using ARMeilleure.CodeGen.RegisterAllocators;
using ARMeilleure.Common;
using ARMeilleure.IntermediateRepresentation;
using System;
using System.Collections.Generic;
using System.Diagnostics;
using System.IO;
namespace ARMeilleure.CodeGen.X86
{
class CodeGenContext
{
private const int ReservedBytesForJump = 1;
private readonly Stream _stream;
private readonly bool _relocatable;
public int StreamOffset => (int)_stream.Length;
public AllocationResult AllocResult { get; }
public Assembler Assembler { get; }
public BasicBlock CurrBlock { get; private set; }
public int CallArgsRegionSize { get; }
public int XmmSaveRegionSize { get; }
private readonly long[] _blockOffsets;
private struct Jump
{
public bool IsConditional { get; }
public X86Condition Condition { get; }
public BasicBlock Target { get; }
public long JumpPosition { get; }
public long RelativeOffset { get; set; }
public int InstSize { get; set; }
public Jump(BasicBlock target, long jumpPosition, int instSize = 0)
{
IsConditional = false;
Condition = 0;
Target = target;
JumpPosition = jumpPosition;
RelativeOffset = 0;
InstSize = instSize;
}
public Jump(X86Condition condition, BasicBlock target, long jumpPosition, int instSize = 0)
{
IsConditional = true;
Condition = condition;
Target = target;
JumpPosition = jumpPosition;
RelativeOffset = 0;
InstSize = instSize;
}
}
private readonly List<Jump> _jumps;
private X86Condition _jNearCondition;
private long _jNearPosition;
private int _jNearLength;
public CodeGenContext(Stream stream, AllocationResult allocResult, int maxCallArgs, int blocksCount, bool relocatable)
{
_stream = stream;
_relocatable = relocatable;
_blockOffsets = new long[blocksCount];
_jumps = new List<Jump>();
AllocResult = allocResult;
Assembler = new Assembler(stream, relocatable);
CallArgsRegionSize = GetCallArgsRegionSize(allocResult, maxCallArgs, out int xmmSaveRegionSize);
XmmSaveRegionSize = xmmSaveRegionSize;
}
private static int GetCallArgsRegionSize(AllocationResult allocResult, int maxCallArgs, out int xmmSaveRegionSize)
{
// We need to add 8 bytes to the total size, as the call to this
// function already pushed 8 bytes (the return address).
int intMask = CallingConvention.GetIntCalleeSavedRegisters() & allocResult.IntUsedRegisters;
int vecMask = CallingConvention.GetVecCalleeSavedRegisters() & allocResult.VecUsedRegisters;
xmmSaveRegionSize = BitUtils.CountBits(vecMask) * 16;
int calleeSaveRegionSize = BitUtils.CountBits(intMask) * 8 + xmmSaveRegionSize + 8;
int argsCount = maxCallArgs;
if (argsCount < 0)
{
// When the function has no calls, argsCount is -1.
// In this case, we don't need to allocate the shadow space.
argsCount = 0;
}
else if (argsCount < 4)
{
// The ABI mandates that the space for at least 4 arguments
// is reserved on the stack (this is called shadow space).
argsCount = 4;
}
int frameSize = calleeSaveRegionSize + allocResult.SpillRegionSize;
// TODO: Instead of always multiplying by 16 (the largest possible size of a variable,
// since a V128 has 16 bytes), we should calculate the exact size consumed by the
// arguments passed to the called functions on the stack.
int callArgsAndFrameSize = frameSize + argsCount * 16;
// Ensure that the Stack Pointer will be aligned to 16 bytes.
callArgsAndFrameSize = (callArgsAndFrameSize + 0xf) & ~0xf;
return callArgsAndFrameSize - frameSize;
}
public void EnterBlock(BasicBlock block)
{
_blockOffsets[block.Index] = _stream.Position;
CurrBlock = block;
}
public void JumpTo(BasicBlock target)
{
if (!_relocatable)
{
_jumps.Add(new Jump(target, _stream.Position));
WritePadding(ReservedBytesForJump);
}
else
{
_jumps.Add(new Jump(target, _stream.Position, 5));
WritePadding(5);
}
}
public void JumpTo(X86Condition condition, BasicBlock target)
{
if (!_relocatable)
{
_jumps.Add(new Jump(condition, target, _stream.Position));
WritePadding(ReservedBytesForJump);
}
else
{
_jumps.Add(new Jump(condition, target, _stream.Position, 6));
WritePadding(6);
}
}
public void JumpToNear(X86Condition condition)
{
_jNearCondition = condition;
_jNearPosition = _stream.Position;
_jNearLength = Assembler.GetJccLength(0, _relocatable);
_stream.Seek(_jNearLength, SeekOrigin.Current);
}
public void JumpHere()
{
long currentPosition = _stream.Position;
_stream.Seek(_jNearPosition, SeekOrigin.Begin);
long offset = currentPosition - (_jNearPosition + _jNearLength);
Debug.Assert(_jNearLength == Assembler.GetJccLength(offset, _relocatable), "Relative offset doesn't fit on near jump.");
Assembler.Jcc(_jNearCondition, offset);
_stream.Seek(currentPosition, SeekOrigin.Begin);
}
private void WritePadding(int size)
{
while (size-- > 0)
{
_stream.WriteByte(0);
}
}
public (byte[], RelocInfo) GetCode()
{
// Write jump relative offsets.
bool modified;
do
{
modified = false;
for (int index = 0; index < _jumps.Count; index++)
{
Jump jump = _jumps[index];
long jumpTarget = _blockOffsets[jump.Target.Index];
long offset = jumpTarget - jump.JumpPosition;
if (!_relocatable)
{
if (offset < 0)
{
for (int index2 = index - 1; index2 >= 0; index2--)
{
Jump jump2 = _jumps[index2];
if (jump2.JumpPosition < jumpTarget)
{
break;
}
offset -= jump2.InstSize - ReservedBytesForJump;
}
}
else
{
for (int index2 = index + 1; index2 < _jumps.Count; index2++)
{
Jump jump2 = _jumps[index2];
if (jump2.JumpPosition >= jumpTarget)
{
break;
}
offset += jump2.InstSize - ReservedBytesForJump;
}
offset -= ReservedBytesForJump;
}
if (jump.IsConditional)
{
jump.InstSize = Assembler.GetJccLength(offset);
}
else
{
jump.InstSize = Assembler.GetJmpLength(offset);
}
// The jump is relative to the next instruction, not the current one.
// Since we didn't know the next instruction address when calculating
// the offset (as the size of the current jump instruction was not known),
// we now need to compensate the offset with the jump instruction size.
// It's also worth noting that:
// - This is only needed for backward jumps.
// - The GetJmpLength and GetJccLength also compensates the offset
// internally when computing the jump instruction size.
if (offset < 0)
{
offset -= jump.InstSize;
}
}
else
{
offset -= jump.InstSize;
}
if (jump.RelativeOffset != offset)
{
modified = true;
}
jump.RelativeOffset = offset;
_jumps[index] = jump;
}
}
while (modified);
// Write the code, ignoring the dummy bytes after jumps, into a new stream.
_stream.Seek(0, SeekOrigin.Begin);
using (MemoryStream codeStream = new MemoryStream())
{
Assembler assembler = new Assembler(codeStream, _relocatable);
for (int index = 0; index < _jumps.Count; index++)
{
Jump jump = _jumps[index];
Span<byte> buffer = new byte[jump.JumpPosition - _stream.Position];
_stream.Read(buffer);
_stream.Seek(!_relocatable ? ReservedBytesForJump : jump.InstSize, SeekOrigin.Current);
codeStream.Write(buffer);
if (jump.IsConditional)
{
assembler.Jcc(jump.Condition, jump.RelativeOffset);
}
else
{
assembler.Jmp(jump.RelativeOffset);
}
}
_stream.CopyTo(codeStream);
var code = codeStream.ToArray();
var relocInfo = Assembler.HasRelocs
? new RelocInfo(Assembler.Relocs.ToArray())
: RelocInfo.Empty;
return (code, relocInfo);
}
}
}
}