Fix/agc zero dim and storage (#828)

* fix(videoout): promote guest images to storage usage

* fix(agc): treat zero-dimension indirect compute dispatches as valid no-ops

src/SharpEmu.Core/Memory/PhysicalVirtualMemory.cs

@@ -1,6 +1,7 @@
 // Copyright (C) 2026 SharpEmu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
+using System.Collections.Concurrent;
 using System.Runtime.InteropServices;
 using SharpEmu.Core.Loader;
 using SharpEmu.HLE;
@@ -18,7 +19,7 @@ public sealed unsafe class PhysicalVirtualMemory : IVirtualMemory, IGuestMemoryA
     private readonly object _allocationSearchHintGate = new();
     private readonly List<MemoryRegion> _regions = new();
     private readonly Dictionary<(ulong DesiredAddress, ulong Alignment, bool Executable), ulong> _allocationSearchHints = new();
-    private readonly Dictionary<ulong, ProgramHeaderFlags> _pageProtections = new();
+    private readonly ConcurrentDictionary<ulong, ProgramHeaderFlags> _pageProtections = new();
     private bool _disposed;
 
     [ThreadStatic]

src/SharpEmu.Libs/Agc/AgcExports.cs

@@ -12174,15 +12174,18 @@ public static partial class AgcExports
 
         if (dispatchEndX == 0 || dispatchEndY == 0 || dispatchEndZ == 0)
         {
-            // Indirect dispatches read their dimensions from a guest buffer a
+            // For indirect dispatches (both absolute and base), zero dimensions are a valid outcome
-            // prior GPU dispatch fills. Zero here means that producer has not run
+            // of GPU culling passes (0 workgroups). VulkanVideoPresenter handles groupCount = 0 as a clean no-op.
-            // yet — signal the caller to suspend on the dims buffer and retry,
+            if (opcode == ItDispatchIndirect || dispatchSource is "absolute-indirect" or "base-indirect")
-            // rather than dropping the work (which black-screens GPU-driven games
-            // like Astro Bot). Direct dispatches carry dims inline, so a zero is
-            // genuinely malformed and still rejected.
-            if (opcode == ItDispatchIndirect)
             {
-                indirectDimsRetryAddress = dimensionsAddress;
+                var waveCount = (initiator & (1u << 15)) != 0 ? 32u : 64u;
+                dispatch = new ComputeDispatch(
+                    0, 0, 0,
+                    0, 0, 0,
+                    waveCount,
+                    IsIndirect: true,
+                    0, 0, 0);
+                return true;
             }
 
             return RejectComputeDispatch(

src/SharpEmu.Libs/VideoOut/VulkanVideoPresenter.cs

@@ -13828,16 +13828,10 @@ internal static unsafe class VulkanVideoPresenter
                     existing.LogicalDepth == depth &&
                     existing.Type == type &&
                     existing.MipLevels == mipLevels &&
+                    (!requiresStorage || existing.SupportsStorageUsage) &&
                     (exactFormatMatch ||
-                     (IsAliasableGuestImageFormat(existing.Format, format) &&
+                    IsAliasableGuestImageFormat(existing.Format, format)))
-                      (!requiresStorage || existing.SupportsStorageUsage))))
                 {
-                    if (requiresStorage && !existing.SupportsStorageUsage)
-                    {
-                        throw new InvalidOperationException(
-                            $"Guest image 0x{target.Address:X16} was created without storage usage.");
-                    }
-
                     existing.IsCpuBacked = false;
                     existing.CpuContentFingerprint = 0;
                     if (existing.RenderPass.Handle == 0 &&
@@ -13870,14 +13864,9 @@ internal static unsafe class VulkanVideoPresenter
                 if (existing.Width == target.Width &&
                     existing.Height == target.Height &&
                     existing.MipLevels == mipLevels &&
+                    (!requiresStorage || existing.SupportsStorageUsage) &&
                     IsCompatibleViewFormat(existing.Format, format))
                 {
-                    if (requiresStorage && !existing.SupportsStorageUsage)
-                    {
-                        throw new InvalidOperationException(
-                            $"Guest image 0x{target.Address:X16} was created without storage usage.");
-                    }
-
                     if (_traceGuestImageEvents)
                     {
                         Console.Error.WriteLine(
@@ -13952,10 +13941,11 @@ internal static unsafe class VulkanVideoPresenter
             {
                 if (requiresStorage && !retained.SupportsStorageUsage)
                 {
-                    throw new InvalidOperationException(
+                    // Do not reuse retained image if it lacks required storage usage
-                        $"Retained guest image 0x{target.Address:X16} was created without storage usage.");
+                    DestroyGuestImage(retained);
                 }
-
+                else
+                {
                     retained.IsCpuBacked = false;
                     retained.CpuContentFingerprint = 0;
                     _guestImages.Add(target.Address, retained);
@@ -13997,6 +13987,7 @@ internal static unsafe class VulkanVideoPresenter
 
                     return retained;
                 }
+            }
 
             var imageInfo = new ImageCreateInfo
             {

src/SharpEmu.ShaderCompiler.Vulkan/Gen5SpirvTranslator.cs

@@ -2366,17 +2366,16 @@ public static partial class Gen5SpirvTranslator
                         return;
                     }
 
+                    // GLOBAL_STORE/LOAD_DWORD(x2/x3/x4) are dword-aligned by the GCN ISA, so read/write dwords directly instead of the per-byte loop.
                     for (uint index = 0; index < control.DwordCount; index++)
                     {
-                        var address = index == 0
+                        var indexedDwordAddress = index == 0
-                            ? byteAddress
+                            ? dwordAddress
-                            : IAdd(byteAddress, UInt(index * sizeof(uint)));
+                            : IAdd(dwordAddress, UInt(index));
-                        StoreBufferBytes(
+                        StoreBufferWord(
                             bindingIndex,
-                            address,
+                            indexedDwordAddress,
-                            LoadV(control.VectorData + index),
+                            LoadV(control.VectorData + index));
-                            sizeof(uint),
-                            0);
                     }
                 });
                 return true;
@@ -2404,12 +2403,12 @@ public static partial class Gen5SpirvTranslator
 
             for (uint index = 0; index < control.DwordCount; index++)
             {
-                var address = index == 0
+                var indexedDwordAddress = index == 0
-                    ? byteAddress
+                    ? dwordAddress
-                    : IAdd(byteAddress, UInt(index * sizeof(uint)));
+                    : IAdd(dwordAddress, UInt(index));
                 StoreV(
                     control.VectorData + index,
-                    LoadUnalignedBufferWord(bindingIndex, address));
+                    LoadBufferWord(bindingIndex, indexedDwordAddress));
             }
 
             return true;
@@ -2510,17 +2509,16 @@ public static partial class Gen5SpirvTranslator
                         return;
                     }
 
+                    // BUFFER_STORE/LOAD_DWORD(x2/x3/x4) are dword-aligned by the GCN ISA, same as the GLOBAL case above — no per-byte reassembly needed.
                     for (uint index = 0; index < control.DwordCount; index++)
                     {
-                        var address = index == 0
+                        var indexedDwordAddress = index == 0
-                            ? byteAddress
+                            ? dwordAddress
-                            : IAdd(byteAddress, UInt(index * sizeof(uint)));
+                            : IAdd(dwordAddress, UInt(index));
-                        StoreBufferBytes(
+                        StoreBufferWord(
                             bindingIndex,
-                            address,
+                            indexedDwordAddress,
-                            LoadV(control.VectorData + index),
+                            LoadV(control.VectorData + index));
-                            sizeof(uint),
-                            0);
                     }
                 });
 
@@ -2576,12 +2574,12 @@ public static partial class Gen5SpirvTranslator
 
             for (uint index = 0; index < control.DwordCount; index++)
             {
-                var address = index == 0
+                var indexedDwordAddress = index == 0
-                    ? byteAddress
+                    ? dwordAddress
-                    : IAdd(byteAddress, UInt(index * sizeof(uint)));
+                    : IAdd(dwordAddress, UInt(index));
                 StoreV(
                     control.VectorData + index,
-                    LoadUnalignedBufferWord(bindingIndex, address));
+                    LoadBufferWord(bindingIndex, indexedDwordAddress));
             }
 
             return true;