Ryujinx-uplift/ChocolArm64/Instructions/InstEmitSimdMemory.cs
gdkchan 5001f78b1d Optimize address translation and write tracking on the MMU (#571)
* Implement faster address translation and write tracking on the MMU

* Rename MemoryAlloc to MemoryManagement, and other nits

* Support multi-level page tables

* Fix typo

* Reword comment a bit

* Support scalar vector loads/stores on the memory fast path, and minor fixes

* Add missing cast

* Alignment

* Fix VirtualFree function signature

* Change MemoryProtection enum to uint aswell for consistency
2019-02-24 18:24:35 +11:00

182 lines
4.9 KiB
C#

using ChocolArm64.Decoders;
using ChocolArm64.State;
using ChocolArm64.Translation;
using System;
using System.Reflection.Emit;
using static ChocolArm64.Instructions.InstEmitMemoryHelper;
using static ChocolArm64.Instructions.InstEmitSimdHelper;
namespace ChocolArm64.Instructions
{
static partial class InstEmit
{
public static void Ld__Vms(ILEmitterCtx context)
{
EmitSimdMemMs(context, isLoad: true);
}
public static void Ld__Vss(ILEmitterCtx context)
{
EmitSimdMemSs(context, isLoad: true);
}
public static void St__Vms(ILEmitterCtx context)
{
EmitSimdMemMs(context, isLoad: false);
}
public static void St__Vss(ILEmitterCtx context)
{
EmitSimdMemSs(context, isLoad: false);
}
private static void EmitSimdMemMs(ILEmitterCtx context, bool isLoad)
{
OpCodeSimdMemMs64 op = (OpCodeSimdMemMs64)context.CurrOp;
int offset = 0;
for (int rep = 0; rep < op.Reps; rep++)
for (int elem = 0; elem < op.Elems; elem++)
for (int sElem = 0; sElem < op.SElems; sElem++)
{
int rtt = (op.Rt + rep + sElem) & 0x1f;
if (isLoad)
{
context.EmitLdint(op.Rn);
context.EmitLdc_I8(offset);
context.Emit(OpCodes.Add);
EmitReadZxCall(context, op.Size);
EmitVectorInsert(context, rtt, elem, op.Size);
if (op.RegisterSize == RegisterSize.Simd64 && elem == op.Elems - 1)
{
EmitVectorZeroUpper(context, rtt);
}
}
else
{
context.EmitLdint(op.Rn);
context.EmitLdc_I8(offset);
context.Emit(OpCodes.Add);
EmitVectorExtractZx(context, rtt, elem, op.Size);
EmitWriteCall(context, op.Size);
}
offset += 1 << op.Size;
}
if (op.WBack)
{
EmitSimdMemWBack(context, offset);
}
}
private static void EmitSimdMemSs(ILEmitterCtx context, bool isLoad)
{
OpCodeSimdMemSs64 op = (OpCodeSimdMemSs64)context.CurrOp;
int offset = 0;
void EmitMemAddress()
{
context.EmitLdint(op.Rn);
context.EmitLdc_I8(offset);
context.Emit(OpCodes.Add);
}
if (op.Replicate)
{
//Only loads uses the replicate mode.
if (!isLoad)
{
throw new InvalidOperationException();
}
int bytes = op.GetBitsCount() >> 3;
int elems = bytes >> op.Size;
for (int sElem = 0; sElem < op.SElems; sElem++)
{
int rt = (op.Rt + sElem) & 0x1f;
for (int index = 0; index < elems; index++)
{
EmitMemAddress();
EmitReadZxCall(context, op.Size);
EmitVectorInsert(context, rt, index, op.Size);
}
if (op.RegisterSize == RegisterSize.Simd64)
{
EmitVectorZeroUpper(context, rt);
}
offset += 1 << op.Size;
}
}
else
{
for (int sElem = 0; sElem < op.SElems; sElem++)
{
int rt = (op.Rt + sElem) & 0x1f;
if (isLoad)
{
EmitMemAddress();
EmitReadZxCall(context, op.Size);
EmitVectorInsert(context, rt, op.Index, op.Size);
}
else
{
EmitMemAddress();
EmitVectorExtractZx(context, rt, op.Index, op.Size);
EmitWriteCall(context, op.Size);
}
offset += 1 << op.Size;
}
}
if (op.WBack)
{
EmitSimdMemWBack(context, offset);
}
}
private static void EmitSimdMemWBack(ILEmitterCtx context, int offset)
{
OpCodeMemReg64 op = (OpCodeMemReg64)context.CurrOp;
context.EmitLdint(op.Rn);
if (op.Rm != RegisterAlias.Zr)
{
context.EmitLdint(op.Rm);
}
else
{
context.EmitLdc_I8(offset);
}
context.Emit(OpCodes.Add);
context.EmitStint(op.Rn);
}
}
}