GPU: Eliminate CB0 accesses when storage buffer accesses are resolved (#3847)

* Eliminate CB0 accesses

Still some work to do, decouple from hle?

* Forgot the important part somehow

* Fix and improve alignment test

* Address Feedback

* Remove some complexity when checking storage buffer alignment

* Update Ryujinx.Graphics.Shader/Translation/Optimizations/GlobalToStorage.cs

Co-authored-by: gdkchan <gab.dark.100@gmail.com>

Co-authored-by: gdkchan <gab.dark.100@gmail.com>

2 files changed, 35 insertions, 18 deletions

diff --git a/Ryujinx.Graphics.Gpu/Engine/Compute/ComputeClass.cs b/Ryujinx.Graphics.Gpu/Engine/Compute/ComputeClass.cs
index bc291174..cd509471 100644
--- a/Ryujinx.Graphics.Gpu/Engine/Compute/ComputeClass.cs
+++ b/Ryujinx.Graphics.Gpu/Engine/Compute/ComputeClass.cs
@@ -138,7 +138,8 @@ namespace Ryujinx.Graphics.Gpu.Engine.Compute
                 qmd.CtaThreadDimension1,
                 qmd.CtaThreadDimension2,
                 localMemorySize,
-                sharedMemorySize);
+                sharedMemorySize,
+                _channel.BufferManager.HasUnalignedStorageBuffers);
 
             CachedShaderProgram cs = memoryManager.Physical.ShaderCache.GetComputeShader(_channel, poolState, computeState, shaderGpuVa);
 
@@ -150,6 +151,33 @@ namespace Ryujinx.Graphics.Gpu.Engine.Compute
 
             ShaderProgramInfo info = cs.Shaders[0].Info;
 
+            bool hasUnaligned = _channel.BufferManager.HasUnalignedStorageBuffers;
+
+            for (int index = 0; index < info.SBuffers.Count; index++)
+            {
+                BufferDescriptor sb = info.SBuffers[index];
+
+                ulong sbDescAddress = _channel.BufferManager.GetComputeUniformBufferAddress(0);
+
+                int sbDescOffset = 0x310 + sb.Slot * 0x10;
+
+                sbDescAddress += (ulong)sbDescOffset;
+
+                SbDescriptor sbDescriptor = _channel.MemoryManager.Physical.Read<SbDescriptor>(sbDescAddress);
+
+                _channel.BufferManager.SetComputeStorageBuffer(sb.Slot, sbDescriptor.PackAddress(), (uint)sbDescriptor.Size, sb.Flags);
+            }
+
+            if ((_channel.BufferManager.HasUnalignedStorageBuffers) != hasUnaligned)
+            {
+                // Refetch the shader, as assumptions about storage buffer alignment have changed.
+                cs = memoryManager.Physical.ShaderCache.GetComputeShader(_channel, poolState, computeState, shaderGpuVa);
+
+                _context.Renderer.Pipeline.SetProgram(cs.HostProgram);
+
+                info = cs.Shaders[0].Info;
+            }
+
             for (int index = 0; index < info.CBuffers.Count; index++)
             {
                 BufferDescriptor cb = info.CBuffers[index];
@@ -174,21 +202,6 @@ namespace Ryujinx.Graphics.Gpu.Engine.Compute
                 _channel.BufferManager.SetComputeUniformBuffer(cb.Slot, cbDescriptor.PackAddress(), (uint)cbDescriptor.Size);
             }
 
-            for (int index = 0; index < info.SBuffers.Count; index++)
-            {
-                BufferDescriptor sb = info.SBuffers[index];
-
-                ulong sbDescAddress = _channel.BufferManager.GetComputeUniformBufferAddress(0);
-
-                int sbDescOffset = 0x310 + sb.Slot * 0x10;
-
-                sbDescAddress += (ulong)sbDescOffset;
-
-                SbDescriptor sbDescriptor = _channel.MemoryManager.Physical.Read<SbDescriptor>(sbDescAddress);
-
-                _channel.BufferManager.SetComputeStorageBuffer(sb.Slot, sbDescriptor.PackAddress(), (uint)sbDescriptor.Size, sb.Flags);
-            }
-
             _channel.BufferManager.SetComputeStorageBufferBindings(info.SBuffers);
             _channel.BufferManager.SetComputeUniformBufferBindings(info.CBuffers);
 
diff --git a/Ryujinx.Graphics.Gpu/Engine/Threed/StateUpdater.cs b/Ryujinx.Graphics.Gpu/Engine/Threed/StateUpdater.cs
index 3f71172c..d51077dc 100644
--- a/Ryujinx.Graphics.Gpu/Engine/Threed/StateUpdater.cs
+++ b/Ryujinx.Graphics.Gpu/Engine/Threed/StateUpdater.cs
@@ -293,9 +293,12 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
         /// </summary>
         private void CommitBindings()
         {
+            var buffers = _channel.BufferManager;
+            var hasUnaligned = buffers.HasUnalignedStorageBuffers;
+
             UpdateStorageBuffers();
 
-            if (!_channel.TextureManager.CommitGraphicsBindings(_shaderSpecState))
+            if (!_channel.TextureManager.CommitGraphicsBindings(_shaderSpecState) || (buffers.HasUnalignedStorageBuffers != hasUnaligned))
             {
                 // Shader must be reloaded.
                 UpdateShaderState();
@@ -1361,7 +1364,8 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
                 _state.State.AlphaTestFunc,
                 _state.State.AlphaTestRef,
                 ref attributeTypes,
-                _drawState.HasConstantBufferDrawParameters);
+                _drawState.HasConstantBufferDrawParameters,
+                _channel.BufferManager.HasUnalignedStorageBuffers);
         }
 
         /// <summary>