Mirrors/Ryujinx
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Implement GPU syncpoints (#980)

Browse source 
* Implement GPU syncpoints

This adds support for GPU syncpoints on the GPU backend & nvservices.

Everything that was implemented here is based on my researches,
hardware testing of the GM20B and reversing of nvservices (8.1.0).

Thanks to @fincs for the informations about some behaviours of the pusher
and for the initial informations about syncpoints.

* syncpoint: address gdkchan's comments

* Add some missing logic to handle SubmitGpfifo correctly

* Handle the NV event API correctly

* evnt => hostEvent

* Finish addressing gdkchan's comments

* nvservices: write the output buffer even when an error is returned

* dma pusher: Implemnet prefetch barrier

lso fix when the commands should be prefetch.

* Partially fix prefetch barrier

* Add a missing syncpoint check in QueryEvent of NvHostSyncPt

* Address Ac_K's comments and fix GetSyncpoint for ChannelResourcePolicy == Channel

* fix SyncptWait & SyncptWaitEx cmds logic

* Address ripinperi's comments

* Address gdkchan's comments

* Move user event management to the control channel

* Fix mm implementation, nvdec works again

* Address ripinperi's comments

* Address gdkchan's comments

* Implement nvhost-ctrl close accurately + make nvservices dispose channels when stopping the emulator

* Fix typo in MultiMediaOperationType
This commit is contained in:
Thog 2020-04-19 03:25:57 +02:00 • committed by GitHub
parent 4960ab85f8
commit 644de99e86
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
37 changed files with 1576 additions and 386 deletions
Download patch file Download diff file Expand all files Collapse all files

171
Ryujinx.HLE/HOS/Services/Nv/NvDrvServices/NvHostChannel/NvHostChannelDeviceFile.cs
View file

@ -1,8 +1,9 @@
using Ryujinx.Common.Logging;
using Ryujinx.Graphics.Gpu;
using Ryujinx.Graphics.Gpu.Memory;
using Ryujinx.HLE.HOS.Services.Nv.Types;
using Ryujinx.HLE.HOS.Services.Nv.NvDrvServices.NvHostAsGpu;
using Ryujinx.HLE.HOS.Services.Nv.NvDrvServices.NvHostChannel.Types;
using Ryujinx.HLE.HOS.Services.Nv.NvDrvServices.NvHostCtrl;
using Ryujinx.HLE.HOS.Services.Nv.NvDrvServices.NvMap;
using System;
using System.Runtime.CompilerServices;
@ -12,21 +13,42 @@ namespace Ryujinx.HLE.HOS.Services.Nv.NvDrvServices.NvHostChannel
{
class NvHostChannelDeviceFile : NvDeviceFile
{
private const uint MaxModuleSyncpoint = 16;
private uint _timeout;
private uint _submitTimeout;
private uint _timeslice;
private GpuContext _gpu;
private Switch _device;
private ARMeilleure.Memory.MemoryManager _memory;
public enum ResourcePolicy
{
Device,
Channel
}
protected static uint[] DeviceSyncpoints = new uint[MaxModuleSyncpoint];
protected uint[] ChannelSyncpoints;
protected static ResourcePolicy ChannelResourcePolicy = ResourcePolicy.Device;
private NvFence _channelSyncpoint;
public NvHostChannelDeviceFile(ServiceCtx context) : base(context)
{
_gpu = context.Device.Gpu;
_device = context.Device;
_memory = context.Memory;
_timeout = 3000;
_submitTimeout = 0;
_timeslice = 0;
ChannelSyncpoints = new uint[MaxModuleSyncpoint];
_channelSyncpoint.Id = _device.System.HostSyncpoint.AllocateSyncpoint(false);
_channelSyncpoint.UpdateValue(_device.System.HostSyncpoint);
}
public override NvInternalResult Ioctl(NvIoctl command, Span<byte> arguments)
@ -132,9 +154,24 @@ namespace Ryujinx.HLE.HOS.Services.Nv.NvDrvServices.NvHostChannel
private NvInternalResult GetSyncpoint(ref GetParameterArguments arguments)
{
arguments.Value = 0;
if (arguments.Parameter >= MaxModuleSyncpoint)
{
return NvInternalResult.InvalidInput;
}
Logger.PrintStub(LogClass.ServiceNv);
if (ChannelResourcePolicy == ResourcePolicy.Device)
{
arguments.Value = GetSyncpointDevice(_device.System.HostSyncpoint, arguments.Parameter, false);
}
else
{
arguments.Value = GetSyncpointChannel(arguments.Parameter, false);
}
if (arguments.Value == 0)
{
return NvInternalResult.TryAgain;
}
return NvInternalResult.Success;
}
@ -293,6 +330,10 @@ namespace Ryujinx.HLE.HOS.Services.Nv.NvDrvServices.NvHostChannel
private NvInternalResult AllocGpfifoEx(ref AllocGpfifoExArguments arguments)
{
_channelSyncpoint.UpdateValue(_device.System.HostSyncpoint);
arguments.Fence = _channelSyncpoint;
Logger.PrintStub(LogClass.ServiceNv);
return NvInternalResult.Success;
@ -300,6 +341,10 @@ namespace Ryujinx.HLE.HOS.Services.Nv.NvDrvServices.NvHostChannel
private NvInternalResult AllocGpfifoEx2(ref AllocGpfifoExArguments arguments)
{
_channelSyncpoint.UpdateValue(_device.System.HostSyncpoint);
arguments.Fence = _channelSyncpoint;
Logger.PrintStub(LogClass.ServiceNv);
return NvInternalResult.Success;
@ -330,17 +375,125 @@ namespace Ryujinx.HLE.HOS.Services.Nv.NvDrvServices.NvHostChannel
protected NvInternalResult SubmitGpfifo(ref SubmitGpfifoArguments header, Span<ulong> entries)
{
foreach (ulong entry in entries)
if (header.Flags.HasFlag(SubmitGpfifoFlags.FenceWait) && header.Flags.HasFlag(SubmitGpfifoFlags.IncrementWithValue))
{
_gpu.DmaPusher.Push(entry);
return NvInternalResult.InvalidInput;
}
header.Fence.Id = 0;
header.Fence.Value = 0;
if (header.Flags.HasFlag(SubmitGpfifoFlags.FenceWait) && !_device.System.HostSyncpoint.IsSyncpointExpired(header.Fence.Id, header.Fence.Value))
{
_device.Gpu.DmaPusher.PushHostCommandBuffer(CreateWaitCommandBuffer(header.Fence));
}
_device.Gpu.DmaPusher.PushEntries(entries);
header.Fence.Id = _channelSyncpoint.Id;
if (header.Flags.HasFlag(SubmitGpfifoFlags.FenceIncrement) || header.Flags.HasFlag(SubmitGpfifoFlags.IncrementWithValue))
{
uint incrementCount = header.Flags.HasFlag(SubmitGpfifoFlags.FenceIncrement) ? 2u : 0u;
if (header.Flags.HasFlag(SubmitGpfifoFlags.IncrementWithValue))
{
incrementCount += header.Fence.Value;
}
header.Fence.Value = _device.System.HostSyncpoint.IncrementSyncpointMaxExt(header.Fence.Id, (int)incrementCount);
}
else
{
header.Fence.Value = _device.System.HostSyncpoint.ReadSyncpointMaxValue(header.Fence.Id);
}
if (header.Flags.HasFlag(SubmitGpfifoFlags.FenceIncrement))
{
_device.Gpu.DmaPusher.PushHostCommandBuffer(CreateIncrementCommandBuffer(ref header.Fence, header.Flags));
}
header.Flags = SubmitGpfifoFlags.None;
_device.Gpu.DmaPusher.SignalNewEntries();
return NvInternalResult.Success;
}
public uint GetSyncpointChannel(uint index, bool isClientManaged)
{
if (ChannelSyncpoints[index] != 0)
{
return ChannelSyncpoints[index];
}
ChannelSyncpoints[index] = _device.System.HostSyncpoint.AllocateSyncpoint(isClientManaged);
return ChannelSyncpoints[index];
}
public static uint GetSyncpointDevice(NvHostSyncpt syncpointManager, uint index, bool isClientManaged)
{
if (DeviceSyncpoints[index] != 0)
{
return DeviceSyncpoints[index];
}
DeviceSyncpoints[index] = syncpointManager.AllocateSyncpoint(isClientManaged);
return DeviceSyncpoints[index];
}
private static int[] CreateWaitCommandBuffer(NvFence fence)
{
int[] commandBuffer = new int[4];
// SyncpointValue = fence.Value;
commandBuffer[0] = 0x2001001C;
commandBuffer[1] = (int)fence.Value;
// SyncpointAction(fence.id, increment: false, switch_en: true);
commandBuffer[2] = 0x2001001D;
commandBuffer[3] = (((int)fence.Id << 8) | (0 << 0) | (1 << 4));
return commandBuffer;
}
private int[] CreateIncrementCommandBuffer(ref NvFence fence, SubmitGpfifoFlags flags)
{
bool hasWfi = !flags.HasFlag(SubmitGpfifoFlags.SuppressWfi);
int[] commandBuffer;
int offset = 0;
if (hasWfi)
{
commandBuffer = new int[8];
// WaitForInterrupt(handle)
commandBuffer[offset++] = 0x2001001E;
commandBuffer[offset++] = 0x0;
}
else
{
commandBuffer = new int[6];
}
// SyncpointValue = 0x0;
commandBuffer[offset++] = 0x2001001C;
commandBuffer[offset++] = 0x0;
// Increment the syncpoint 2 times. (mitigate a hardware bug)
// SyncpointAction(fence.id, increment: true, switch_en: false);
commandBuffer[offset++] = 0x2001001D;
commandBuffer[offset++] = (((int)fence.Id << 8) | (1 << 0) | (0 << 4));
// SyncpointAction(fence.id, increment: true, switch_en: false);
commandBuffer[offset++] = 0x2001001D;
commandBuffer[offset++] = (((int)fence.Id << 8) | (1 << 0) | (0 << 4));
return commandBuffer;
}
public override void Close() { }
}
}