Merge pull request #10781 from tellowkrinkle/UberVertexLoader

VideoCommon: Add dynamic vertex loader for ubershaders to reduce pipeline count

36 files changed, 558 insertions, 163 deletions

diff --git a/Source/Core/Core/State.cpp b/Source/Core/Core/State.cpp
index a6467f2b2c..f97e166f19 100644
--- a/Source/Core/Core/State.cpp
+++ b/Source/Core/Core/State.cpp
@@ -74,7 +74,7 @@ static std::recursive_mutex g_save_thread_mutex;
 static std::thread g_save_thread;
 
 // Don't forget to increase this after doing changes on the savestate system
-constexpr u32 STATE_VERSION = 148;  // Last changed in PR 10768
+constexpr u32 STATE_VERSION = 149;  // Last changed in PR 10781
 
 // Maps savestate versions to Dolphin versions.
 // Versions after 42 don't need to be added to this list,
diff --git a/Source/Core/VideoBackends/D3D/D3DMain.cpp b/Source/Core/VideoBackends/D3D/D3DMain.cpp
index b8d7fb5871..755f0cf590 100644
--- a/Source/Core/VideoBackends/D3D/D3DMain.cpp
+++ b/Source/Core/VideoBackends/D3D/D3DMain.cpp
@@ -112,6 +112,7 @@ void VideoBackend::FillBackendInfo()
   g_Config.backend_info.bSupportsLogicOp = D3D::SupportsLogicOp(g_Config.iAdapter);
   g_Config.backend_info.bSupportsSettingObjectNames = true;
   g_Config.backend_info.bSupportsPartialMultisampleResolve = true;
+  g_Config.backend_info.bSupportsDynamicVertexLoader = false;
 
   g_Config.backend_info.Adapters = D3DCommon::GetAdapterNames();
   g_Config.backend_info.AAModes = D3D::GetAAModes(g_Config.iAdapter);
diff --git a/Source/Core/VideoBackends/D3D12/D3D12Renderer.cpp b/Source/Core/VideoBackends/D3D12/D3D12Renderer.cpp
index b1cb3378ed..80872086a9 100644
--- a/Source/Core/VideoBackends/D3D12/D3D12Renderer.cpp
+++ b/Source/Core/VideoBackends/D3D12/D3D12Renderer.cpp
@@ -180,7 +180,8 @@ void Renderer::SetPipeline(const AbstractPipeline* pipeline)
       m_state.root_signature = dx_pipeline->GetRootSignature();
       m_dirty_bits |= DirtyState_RootSignature | DirtyState_PS_CBV | DirtyState_VS_CBV |
                       DirtyState_GS_CBV | DirtyState_SRV_Descriptor |
-                      DirtyState_Sampler_Descriptor | DirtyState_UAV_Descriptor;
+                      DirtyState_Sampler_Descriptor | DirtyState_UAV_Descriptor |
+                      DirtyState_VS_SRV_Descriptor;
     }
     if (dx_pipeline->UseIntegerRTV() != m_state.using_integer_rtv)
     {
@@ -362,6 +363,11 @@ void Renderer::DrawIndexed(u32 base_index, u32 num_indices, u32 base_vertex)
   if (!ApplyState())
     return;
 
+  // DX12 is great and doesn't include the base vertex in SV_VertexID
+  if (static_cast<const DXPipeline*>(m_current_pipeline)->GetUsage() ==
+      AbstractPipelineUsage::GXUber)
+    g_dx_context->GetCommandList()->SetGraphicsRoot32BitConstant(
+        ROOT_PARAMETER_BASE_VERTEX_CONSTANT, base_vertex, 0);
   g_dx_context->GetCommandList()->DrawIndexedInstanced(num_indices, 1, base_index, base_vertex, 0);
 }
 
@@ -494,18 +500,22 @@ void Renderer::SetPixelShaderUAV(D3D12_CPU_DESCRIPTOR_HANDLE handle)
   m_dirty_bits |= DirtyState_PS_UAV;
 }
 
-void Renderer::SetVertexBuffer(D3D12_GPU_VIRTUAL_ADDRESS address, u32 stride, u32 size)
+void Renderer::SetVertexBuffer(D3D12_GPU_VIRTUAL_ADDRESS address, D3D12_CPU_DESCRIPTOR_HANDLE srv,
+                               u32 stride, u32 size)
 {
-  if (m_state.vertex_buffer.BufferLocation == address &&
-      m_state.vertex_buffer.StrideInBytes == stride && m_state.vertex_buffer.SizeInBytes == size)
+  if (m_state.vertex_buffer.BufferLocation != address ||
+      m_state.vertex_buffer.StrideInBytes != stride || m_state.vertex_buffer.SizeInBytes != size)
   {
-    return;
+    m_state.vertex_buffer.BufferLocation = address;
+    m_state.vertex_buffer.StrideInBytes = stride;
+    m_state.vertex_buffer.SizeInBytes = size;
+    m_dirty_bits |= DirtyState_VertexBuffer;
+  }
+  if (m_state.vs_srv.ptr != srv.ptr)
+  {
+    m_state.vs_srv = srv;
+    m_dirty_bits |= DirtyState_VS_SRV;
   }
-
-  m_state.vertex_buffer.BufferLocation = address;
-  m_state.vertex_buffer.StrideInBytes = stride;
-  m_state.vertex_buffer.SizeInBytes = size;
-  m_dirty_bits |= DirtyState_VertexBuffer;
 }
 
 void Renderer::SetIndexBuffer(D3D12_GPU_VIRTUAL_ADDRESS address, u32 size, DXGI_FORMAT format)
@@ -535,15 +545,17 @@ bool Renderer::ApplyState()
   // Clear bits before actually changing state. Some state (e.g. cbuffers) can't be set
   // if utility pipelines are bound.
   const u32 dirty_bits = m_dirty_bits;
-  m_dirty_bits &= ~(
-      DirtyState_Framebuffer | DirtyState_Pipeline | DirtyState_Viewport | DirtyState_ScissorRect |
-      DirtyState_PS_UAV | DirtyState_PS_CBV | DirtyState_VS_CBV | DirtyState_GS_CBV |
-      DirtyState_SRV_Descriptor | DirtyState_Sampler_Descriptor | DirtyState_UAV_Descriptor |
-      DirtyState_VertexBuffer | DirtyState_IndexBuffer | DirtyState_PrimitiveTopology);
+  m_dirty_bits &=
+      ~(DirtyState_Framebuffer | DirtyState_Pipeline | DirtyState_Viewport |
+        DirtyState_ScissorRect | DirtyState_PS_UAV | DirtyState_PS_CBV | DirtyState_VS_CBV |
+        DirtyState_GS_CBV | DirtyState_SRV_Descriptor | DirtyState_Sampler_Descriptor |
+        DirtyState_UAV_Descriptor | DirtyState_VertexBuffer | DirtyState_IndexBuffer |
+        DirtyState_PrimitiveTopology | DirtyState_VS_SRV_Descriptor);
 
   auto* const cmdlist = g_dx_context->GetCommandList();
+  auto* const pipeline = static_cast<const DXPipeline*>(m_current_pipeline);
   if (dirty_bits & DirtyState_Pipeline)
-    cmdlist->SetPipelineState(static_cast<const DXPipeline*>(m_current_pipeline)->GetPipeline());
+    cmdlist->SetPipelineState(pipeline->GetPipeline());
 
   if (dirty_bits & DirtyState_Framebuffer)
     BindFramebuffer(static_cast<DXFramebuffer*>(m_current_framebuffer));
@@ -572,7 +584,7 @@ bool Renderer::ApplyState()
                                             m_state.sampler_descriptor_base);
   }
 
-  if (static_cast<const DXPipeline*>(m_current_pipeline)->GetUsage() == AbstractPipelineUsage::GX)
+  if (pipeline->GetUsage() != AbstractPipelineUsage::Utility)
   {
     if (dirty_bits & DirtyState_VS_CBV)
     {
@@ -589,6 +601,13 @@ bool Renderer::ApplyState()
       }
     }
 
+    if (dirty_bits & DirtyState_VS_SRV_Descriptor &&
+        pipeline->GetUsage() == AbstractPipelineUsage::GXUber)
+    {
+      cmdlist->SetGraphicsRootDescriptorTable(ROOT_PARAMETER_VS_SRV,
+                                              m_state.vertex_srv_descriptor_base);
+    }
+
     if (dirty_bits & DirtyState_GS_CBV)
     {
       cmdlist->SetGraphicsRootConstantBufferView(ROOT_PARAMETER_GS_CBV,
@@ -641,7 +660,9 @@ void Renderer::UpdateDescriptorTables()
   const bool sampler_update_failed =
       (m_dirty_bits & DirtyState_Samplers) && !UpdateSamplerDescriptorTable();
   const bool uav_update_failed = (m_dirty_bits & DirtyState_PS_UAV) && !UpdateUAVDescriptorTable();
-  if (texture_update_failed || sampler_update_failed || uav_update_failed)
+  const bool srv_update_failed =
+      (m_dirty_bits & DirtyState_VS_SRV) && !UpdateVSSRVDescriptorTable();
+  if (texture_update_failed || sampler_update_failed || uav_update_failed || srv_update_failed)
   {
     WARN_LOG_FMT(VIDEO, "Executing command list while waiting for temporary {}",
                  texture_update_failed ? "descriptors" : "samplers");
@@ -651,6 +672,7 @@ void Renderer::UpdateDescriptorTables()
     UpdateSRVDescriptorTable();
     UpdateSamplerDescriptorTable();
     UpdateUAVDescriptorTable();
+    UpdateVSSRVDescriptorTable();
   }
 }
 
@@ -700,6 +722,26 @@ bool Renderer::UpdateUAVDescriptorTable()
   return true;
 }
 
+bool Renderer::UpdateVSSRVDescriptorTable()
+{
+  if (!g_ActiveConfig.backend_info.bSupportsDynamicVertexLoader ||
+      static_cast<const DXPipeline*>(m_current_pipeline)->GetUsage() !=
+          AbstractPipelineUsage::GXUber)
+  {
+    return true;
+  }
+
+  DescriptorHandle handle;
+  if (!g_dx_context->GetDescriptorAllocator()->Allocate(1, &handle))
+    return false;
+
+  g_dx_context->GetDevice()->CopyDescriptorsSimple(1, handle.cpu_handle, m_state.vs_srv,
+                                                   D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV);
+  m_state.vertex_srv_descriptor_base = handle.gpu_handle;
+  m_dirty_bits = (m_dirty_bits & ~DirtyState_VS_SRV) | DirtyState_VS_SRV_Descriptor;
+  return true;
+}
+
 bool Renderer::UpdateComputeUAVDescriptorTable()
 {
   DescriptorHandle handle;
diff --git a/Source/Core/VideoBackends/D3D12/D3D12Renderer.h b/Source/Core/VideoBackends/D3D12/D3D12Renderer.h
index 9624199aa4..19188a5acf 100644
--- a/Source/Core/VideoBackends/D3D12/D3D12Renderer.h
+++ b/Source/Core/VideoBackends/D3D12/D3D12Renderer.h
@@ -88,7 +88,8 @@ public:
   void SetPixelShaderUAV(D3D12_CPU_DESCRIPTOR_HANDLE handle);
 
   // Graphics vertex/index buffer binding.
-  void SetVertexBuffer(D3D12_GPU_VIRTUAL_ADDRESS address, u32 stride, u32 size);
+  void SetVertexBuffer(D3D12_GPU_VIRTUAL_ADDRESS address, D3D12_CPU_DESCRIPTOR_HANDLE srv,
+                       u32 stride, u32 size);
   void SetIndexBuffer(D3D12_GPU_VIRTUAL_ADDRESS address, u32 size, DXGI_FORMAT format);
 
   // Binds all dirty state
@@ -126,6 +127,8 @@ private:
     DirtyState_RootSignature = (1 << 17),
     DirtyState_ComputeRootSignature = (1 << 18),
     DirtyState_DescriptorHeaps = (1 << 19),
+    DirtyState_VS_SRV = (1 << 20),
+    DirtyState_VS_SRV_Descriptor = (1 << 21),
 
     DirtyState_All =
         DirtyState_Framebuffer | DirtyState_Pipeline | DirtyState_Textures | DirtyState_Samplers |
@@ -133,7 +136,8 @@ private:
         DirtyState_PS_UAV | DirtyState_PS_CBV | DirtyState_VS_CBV | DirtyState_GS_CBV |
         DirtyState_SRV_Descriptor | DirtyState_Sampler_Descriptor | DirtyState_UAV_Descriptor |
         DirtyState_VertexBuffer | DirtyState_IndexBuffer | DirtyState_PrimitiveTopology |
-        DirtyState_RootSignature | DirtyState_ComputeRootSignature | DirtyState_DescriptorHeaps
+        DirtyState_RootSignature | DirtyState_ComputeRootSignature | DirtyState_DescriptorHeaps |
+        DirtyState_VS_SRV | DirtyState_VS_SRV_Descriptor
   };
 
   void CheckForSwapChainChanges();
@@ -144,6 +148,7 @@ private:
   void UpdateDescriptorTables();
   bool UpdateSRVDescriptorTable();
   bool UpdateUAVDescriptorTable();
+  bool UpdateVSSRVDescriptorTable();
   bool UpdateComputeUAVDescriptorTable();
   bool UpdateSamplerDescriptorTable();
 
@@ -157,11 +162,13 @@ private:
     DXShader* compute_shader = nullptr;
     std::array<D3D12_GPU_VIRTUAL_ADDRESS, 3> constant_buffers = {};
     std::array<D3D12_CPU_DESCRIPTOR_HANDLE, MAX_TEXTURES> textures = {};
+    D3D12_CPU_DESCRIPTOR_HANDLE vs_srv = {};
     D3D12_CPU_DESCRIPTOR_HANDLE ps_uav = {};
     SamplerStateSet samplers = {};
     const DXTexture* compute_image_texture = nullptr;
     D3D12_VIEWPORT viewport = {};
     D3D12_RECT scissor = {};
+    D3D12_GPU_DESCRIPTOR_HANDLE vertex_srv_descriptor_base = {};
     D3D12_GPU_DESCRIPTOR_HANDLE srv_descriptor_base = {};
     D3D12_GPU_DESCRIPTOR_HANDLE sampler_descriptor_base = {};
     D3D12_GPU_DESCRIPTOR_HANDLE uav_descriptor_base = {};
diff --git a/Source/Core/VideoBackends/D3D12/D3D12VertexManager.cpp b/Source/Core/VideoBackends/D3D12/D3D12VertexManager.cpp
index b85e4e459c..5a9efc0d6c 100644
--- a/Source/Core/VideoBackends/D3D12/D3D12VertexManager.cpp
+++ b/Source/Core/VideoBackends/D3D12/D3D12VertexManager.cpp
@@ -64,6 +64,18 @@ bool VertexManager::Initialize()
                                                         &srv_desc, dh.cpu_handle);
   }
 
+  if (!g_dx_context->GetDescriptorHeapManager().Allocate(&m_vertex_srv))
+  {
+    PanicAlertFmt("Failed to allocate descriptor for vertex srv");
+    return false;
+  }
+
+  D3D12_SHADER_RESOURCE_VIEW_DESC srv_desc = {DXGI_FORMAT_R32_UINT, D3D12_SRV_DIMENSION_BUFFER,
+                                              D3D12_DEFAULT_SHADER_4_COMPONENT_MAPPING};
+  srv_desc.Buffer.NumElements = m_vertex_stream_buffer.GetSize() / sizeof(u32);
+  g_dx_context->GetDevice()->CreateShaderResourceView(m_vertex_stream_buffer.GetBuffer(), &srv_desc,
+                                                      m_vertex_srv.cpu_handle);
+
   UploadAllConstants();
   return true;
 }
@@ -115,7 +127,8 @@ void VertexManager::CommitBuffer(u32 num_vertices, u32 vertex_stride, u32 num_in
   ADDSTAT(g_stats.this_frame.bytes_vertex_streamed, static_cast<int>(vertex_data_size));
   ADDSTAT(g_stats.this_frame.bytes_index_streamed, static_cast<int>(index_data_size));
 
-  Renderer::GetInstance()->SetVertexBuffer(m_vertex_stream_buffer.GetGPUPointer(), vertex_stride,
+  Renderer::GetInstance()->SetVertexBuffer(m_vertex_stream_buffer.GetGPUPointer(),
+                                           m_vertex_srv.cpu_handle, vertex_stride,
                                            m_vertex_stream_buffer.GetSize());
   Renderer::GetInstance()->SetIndexBuffer(m_index_stream_buffer.GetGPUPointer(),
                                           m_index_stream_buffer.GetSize(), DXGI_FORMAT_R16_UINT);
diff --git a/Source/Core/VideoBackends/D3D12/D3D12VertexManager.h b/Source/Core/VideoBackends/D3D12/D3D12VertexManager.h
index 6bcff92bb7..2dd2cdfee9 100644
--- a/Source/Core/VideoBackends/D3D12/D3D12VertexManager.h
+++ b/Source/Core/VideoBackends/D3D12/D3D12VertexManager.h
@@ -46,6 +46,7 @@ protected:
   StreamBuffer m_uniform_stream_buffer;
   StreamBuffer m_texel_stream_buffer;
   std::array<DescriptorHandle, NUM_TEXEL_BUFFER_FORMATS> m_texel_buffer_views = {};
+  DescriptorHandle m_vertex_srv = {};
 };
 
 }  // namespace DX12
diff --git a/Source/Core/VideoBackends/D3D12/DX12Context.cpp b/Source/Core/VideoBackends/D3D12/DX12Context.cpp
index e3ff8a7417..5cf0e414c4 100644
--- a/Source/Core/VideoBackends/D3D12/DX12Context.cpp
+++ b/Source/Core/VideoBackends/D3D12/DX12Context.cpp
@@ -261,6 +261,16 @@ bool DXContext::CreateDescriptorHeaps()
   return true;
 }
 
+static void SetRootParamConstant(D3D12_ROOT_PARAMETER* rp, u32 shader_reg, u32 num_values,
+                                 D3D12_SHADER_VISIBILITY visibility)
+{
+  rp->ParameterType = D3D12_ROOT_PARAMETER_TYPE_32BIT_CONSTANTS;
+  rp->Constants.Num32BitValues = num_values;
+  rp->Constants.ShaderRegister = shader_reg;
+  rp->Constants.RegisterSpace = 0;
+  rp->ShaderVisibility = visibility;
+}
+
 static void SetRootParamCBV(D3D12_ROOT_PARAMETER* rp, u32 shader_reg,
                             D3D12_SHADER_VISIBILITY visibility)
 {
@@ -345,6 +355,11 @@ bool DXContext::CreateGXRootSignature()
   param_count++;
   SetRootParamCBV(&params[param_count], 0, D3D12_SHADER_VISIBILITY_GEOMETRY);
   param_count++;
+  SetRootParamTable(&params[param_count], &ranges[param_count], D3D12_DESCRIPTOR_RANGE_TYPE_SRV, 3,
+                    1, D3D12_SHADER_VISIBILITY_VERTEX);
+  param_count++;
+  SetRootParamConstant(&params[param_count], 2, 1, D3D12_SHADER_VISIBILITY_VERTEX);
+  param_count++;
 
   // Since these must be contiguous, pixel lighting goes to bbox if not enabled.
   if (g_ActiveConfig.bBBoxEnable)
diff --git a/Source/Core/VideoBackends/D3D12/DX12Context.h b/Source/Core/VideoBackends/D3D12/DX12Context.h
index d1c5b38798..ad9ef9fe74 100644
--- a/Source/Core/VideoBackends/D3D12/DX12Context.h
+++ b/Source/Core/VideoBackends/D3D12/DX12Context.h
@@ -27,6 +27,8 @@ enum ROOT_PARAMETER
   ROOT_PARAMETER_VS_CBV,
   ROOT_PARAMETER_VS_CBV2,
   ROOT_PARAMETER_GS_CBV,
+  ROOT_PARAMETER_VS_SRV,
+  ROOT_PARAMETER_BASE_VERTEX_CONSTANT,
   ROOT_PARAMETER_PS_UAV_OR_CBV2,
   ROOT_PARAMETER_PS_CBV2,  // ROOT_PARAMETER_PS_UAV_OR_CBV2 if bbox is not enabled
   NUM_ROOT_PARAMETERS
diff --git a/Source/Core/VideoBackends/D3D12/DX12Pipeline.cpp b/Source/Core/VideoBackends/D3D12/DX12Pipeline.cpp
index 52b73e6911..bca820a2c3 100644
--- a/Source/Core/VideoBackends/D3D12/DX12Pipeline.cpp
+++ b/Source/Core/VideoBackends/D3D12/DX12Pipeline.cpp
@@ -165,6 +165,7 @@ std::unique_ptr<DXPipeline> DXPipeline::Create(const AbstractPipelineConfig& con
   switch (config.usage)
   {
   case AbstractPipelineUsage::GX:
+  case AbstractPipelineUsage::GXUber:
     desc.pRootSignature = g_dx_context->GetGXRootSignature();
     break;
   case AbstractPipelineUsage::Utility:
diff --git a/Source/Core/VideoBackends/D3D12/VideoBackend.cpp b/Source/Core/VideoBackends/D3D12/VideoBackend.cpp
index b0611a8690..b901b994e0 100644
--- a/Source/Core/VideoBackends/D3D12/VideoBackend.cpp
+++ b/Source/Core/VideoBackends/D3D12/VideoBackend.cpp
@@ -87,6 +87,7 @@ void VideoBackend::FillBackendInfo()
   g_Config.backend_info.bSupportsLodBiasInSampler = true;
   g_Config.backend_info.bSupportsSettingObjectNames = true;
   g_Config.backend_info.bSupportsPartialMultisampleResolve = true;
+  g_Config.backend_info.bSupportsDynamicVertexLoader = true;
 
   // We can only check texture support once we have a device.
   if (g_dx_context)
diff --git a/Source/Core/VideoBackends/Metal/MTLObjectCache.mm b/Source/Core/VideoBackends/Metal/MTLObjectCache.mm
index 3181e1fdee..48a8ad7dae 100644
--- a/Source/Core/VideoBackends/Metal/MTLObjectCache.mm
+++ b/Source/Core/VideoBackends/Metal/MTLObjectCache.mm
@@ -220,7 +220,7 @@ static MTLCullMode Convert(CullMode cull)
   switch (cull)
   {
   case CullMode::None:
-  case CullMode::All:  // Handled by disabling rasterization
+  case CullMode::All:  // Handled by VertexLoaderManager::RunVertices
     return MTLCullModeNone;
   case CullMode::Front:
     return MTLCullModeFront;
@@ -289,7 +289,8 @@ public:
       }
     };
     template <size_t N>
-    static void CopyAll(std::array<VertexAttribute, N>& output, const AttributeFormat (&input)[N])
+    static void CopyAll(std::array<VertexAttribute, N>& output,
+                        const std::array<AttributeFormat, N>& input)
     {
       for (size_t i = 0; i < N; ++i)
         output[i] = VertexAttribute(input[i]);
@@ -327,13 +328,14 @@ public:
         blend.subtractAlpha  = cfg.blending_state.subtractAlpha.Value();
         // clang-format on
       }
-      // Throw extras in bits we don't otherwise use
-      if (cfg.rasterization_state.cullmode == CullMode::All)
-        blend.hex |= 1 << 29;
-      if (cfg.rasterization_state.primitive == PrimitiveType::Points)
-        blend.hex |= 1 << 30;
-      else if (cfg.rasterization_state.primitive == PrimitiveType::Lines)
-        blend.hex |= 1 << 31;
+
+      if (cfg.usage != AbstractPipelineUsage::GXUber)
+      {
+        if (cfg.rasterization_state.primitive == PrimitiveType::Points)
+          is_points = true;
+        else if (cfg.rasterization_state.primitive == PrimitiveType::Lines)
+          is_lines = true;
+      }
     }
     PipelineID() { memset(this, 0, sizeof(*this)); }
     PipelineID(const PipelineID& other) { memcpy(this, &other, sizeof(*this)); }
@@ -359,7 +361,13 @@ public:
     VertexAttribute v_posmtx;
     const Shader* vertex_shader;
     const Shader* fragment_shader;
-    BlendingState blend;
+    union
+    {
+      BlendingState blend;
+      // Throw extras in bits we don't otherwise use
+      BitField<30, 1, bool, u32> is_points;
+      BitField<31, 1, bool, u32> is_lines;
+    };
     FramebufferState framebuffer;
   };
 
@@ -377,24 +385,17 @@ public:
       auto desc = MRCTransfer([MTLRenderPipelineDescriptor new]);
       [desc setVertexFunction:static_cast<const Shader*>(config.vertex_shader)->GetShader()];
       [desc setFragmentFunction:static_cast<const Shader*>(config.pixel_shader)->GetShader()];
-      if (config.usage == AbstractPipelineUsage::GX)
-      {
-        if ([[[desc vertexFunction] label] containsString:@"Uber"])
-          [desc
-              setLabel:[NSString stringWithFormat:@"GX Uber Pipeline %d", m_pipeline_counter[0]++]];
-        else
-          [desc setLabel:[NSString stringWithFormat:@"GX Pipeline %d", m_pipeline_counter[1]++]];
-      }
+      if (config.usage == AbstractPipelineUsage::GXUber)
+        [desc setLabel:[NSString stringWithFormat:@"GX Uber Pipeline %d", m_pipeline_counter[0]++]];
+      else if (config.usage == AbstractPipelineUsage::GX)
+        [desc setLabel:[NSString stringWithFormat:@"GX Pipeline %d", m_pipeline_counter[1]++]];
       else
-      {
         [desc setLabel:[NSString stringWithFormat:@"Utility Pipeline %d", m_pipeline_counter[2]++]];
-      }
       if (config.vertex_format)
         [desc setVertexDescriptor:static_cast<const VertexFormat*>(config.vertex_format)->Get()];
       RasterizationState rs = config.rasterization_state;
-      [desc setInputPrimitiveTopology:GetClass(rs.primitive)];
-      if (rs.cullmode == CullMode::All)
-        [desc setRasterizationEnabled:NO];
+      if (config.usage != AbstractPipelineUsage::GXUber)
+        [desc setInputPrimitiveTopology:GetClass(rs.primitive)];
       MTLRenderPipelineColorAttachmentDescriptor* color0 =
           [[desc colorAttachments] objectAtIndexedSubscript:0];
       BlendingState bs = config.blending_state;
diff --git a/Source/Core/VideoBackends/Metal/MTLStateTracker.mm b/Source/Core/VideoBackends/Metal/MTLStateTracker.mm
index 3ab6224f83..0004c81089 100644
--- a/Source/Core/VideoBackends/Metal/MTLStateTracker.mm
+++ b/Source/Core/VideoBackends/Metal/MTLStateTracker.mm
@@ -608,7 +608,7 @@ void Metal::StateTracker::PrepareRender()
     BeginRenderPass(MTLLoadActionLoad);
   id<MTLRenderCommandEncoder> enc = m_current_render_encoder;
   const Pipeline* pipe = m_state.render_pipeline;
-  bool is_gx = pipe->Usage() == AbstractPipelineUsage::GX;
+  bool is_gx = pipe->Usage() != AbstractPipelineUsage::Utility;
   NSString* label = is_gx ? LABEL_GX : LABEL_UTIL;
   if (m_flags.should_apply_label && m_current.label != label)
   {
diff --git a/Source/Core/VideoBackends/Metal/MTLUtil.mm b/Source/Core/VideoBackends/Metal/MTLUtil.mm
index 14c83b5c17..30eb305de3 100644
--- a/Source/Core/VideoBackends/Metal/MTLUtil.mm
+++ b/Source/Core/VideoBackends/Metal/MTLUtil.mm
@@ -74,6 +74,7 @@ void Metal::Util::PopulateBackendInfo(VideoConfig* config)
   config->backend_info.bSupportsSettingObjectNames = true;
   // Metal requires multisample resolve to be done on a render pass
   config->backend_info.bSupportsPartialMultisampleResolve = false;
+  config->backend_info.bSupportsDynamicVertexLoader = true;
 }
 
 void Metal::Util::PopulateBackendInfoAdapters(VideoConfig* config,
@@ -426,6 +427,7 @@ std::optional<std::string> Metal::Util::TranslateShaderToMSL(ShaderStage stage,
   static const spirv_cross::MSLResourceBinding resource_bindings[] = {
       MakeResourceBinding(spv::ExecutionModelVertex,    0, 0, 1, 0, 0), // vs/ubo
       MakeResourceBinding(spv::ExecutionModelVertex,    0, 1, 1, 0, 0), // vs/ubo
+      MakeResourceBinding(spv::ExecutionModelVertex,    2, 1, 0, 0, 0), // vs/ssbo
       MakeResourceBinding(spv::ExecutionModelFragment,  0, 0, 0, 0, 0), // vs/ubo
       MakeResourceBinding(spv::ExecutionModelFragment,  0, 1, 1, 0, 0), // vs/ubo
       MakeResourceBinding(spv::ExecutionModelFragment,  1, 0, 0, 0, 0), // ps/samp0
diff --git a/Source/Core/VideoBackends/Metal/MTLVertexFormat.mm b/Source/Core/VideoBackends/Metal/MTLVertexFormat.mm
index 6453c5189a..d9407402e2 100644
--- a/Source/Core/VideoBackends/Metal/MTLVertexFormat.mm
+++ b/Source/Core/VideoBackends/Metal/MTLVertexFormat.mm
@@ -125,7 +125,7 @@ static void SetAttribute(MTLVertexDescriptor* desc, u32 attribute, const Attribu
 
 template <size_t N>
 static void SetAttributes(MTLVertexDescriptor* desc, u32 attribute,
-                          const AttributeFormat (&format)[N])
+                          const std::array<AttributeFormat, N>& format)
 {
   for (size_t i = 0; i < N; ++i)
     SetAttribute(desc, attribute + i, format[i]);
diff --git a/Source/Core/VideoBackends/Null/NullBackend.cpp b/Source/Core/VideoBackends/Null/NullBackend.cpp
index b68c9cfd94..7cc8919ad0 100644
--- a/Source/Core/VideoBackends/Null/NullBackend.cpp
+++ b/Source/Core/VideoBackends/Null/NullBackend.cpp
@@ -60,6 +60,7 @@ void VideoBackend::InitBackendInfo()
   g_Config.backend_info.bSupportsLodBiasInSampler = false;
   g_Config.backend_info.bSupportsSettingObjectNames = false;
   g_Config.backend_info.bSupportsPartialMultisampleResolve = true;
+  g_Config.backend_info.bSupportsDynamicVertexLoader = false;
 
   // aamodes: We only support 1 sample, so no MSAA
   g_Config.backend_info.Adapters.clear();
diff --git a/Source/Core/VideoBackends/OGL/OGLMain.cpp b/Source/Core/VideoBackends/OGL/OGLMain.cpp
index 24a65f746d..f6a84240e0 100644
--- a/Source/Core/VideoBackends/OGL/OGLMain.cpp
+++ b/Source/Core/VideoBackends/OGL/OGLMain.cpp
@@ -94,6 +94,8 @@ void VideoBackend::InitBackendInfo()
   g_Config.backend_info.bSupportsPipelineCacheData = false;
   g_Config.backend_info.bSupportsLodBiasInSampler = true;
   g_Config.backend_info.bSupportsPartialMultisampleResolve = true;
+  // Unneccessary since OGL doesn't use pipelines
+  g_Config.backend_info.bSupportsDynamicVertexLoader = false;
 
   // TODO: There is a bug here, if texel buffers or SSBOs/atomics are not supported the graphics
   // options will show the option when it is not supported. The only way around this would be
diff --git a/Source/Core/VideoBackends/Software/SWmain.cpp b/Source/Core/VideoBackends/Software/SWmain.cpp
index 80dc4603f2..5076106089 100644
--- a/Source/Core/VideoBackends/Software/SWmain.cpp
+++ b/Source/Core/VideoBackends/Software/SWmain.cpp
@@ -89,6 +89,7 @@ void VideoSoftware::InitBackendInfo()
   g_Config.backend_info.bSupportsLodBiasInSampler = false;
   g_Config.backend_info.bSupportsSettingObjectNames = false;
   g_Config.backend_info.bSupportsPartialMultisampleResolve = true;
+  g_Config.backend_info.bSupportsDynamicVertexLoader = false;
 
   // aamodes
   g_Config.backend_info.AAModes = {1};
diff --git a/Source/Core/VideoBackends/Vulkan/Constants.h b/Source/Core/VideoBackends/Vulkan/Constants.h
index c1218f9a67..52d59b4f45 100644
--- a/Source/Core/VideoBackends/Vulkan/Constants.h
+++ b/Source/Core/VideoBackends/Vulkan/Constants.h
@@ -39,6 +39,8 @@ enum DESCRIPTOR_SET_LAYOUT
 //       - Per-stage UBO (VS/GS/PS, VS constants accessible from PS) [set=0, binding=0-2]
 //       - 8 combined image samplers (accessible from PS) [set=1, binding=0-7]
 //       - 1 SSBO accessible from PS if supported [set=2, binding=0]
+//   - Uber
+//       - Like standard, plus 1 SSBO accessible from VS if supported [set=2, binding=1]
 //   - Utility
 //       - 1 combined UBO, accessible from VS/GS/PS [set=0, binding=0]
 //       - 8 combined image samplers (accessible from PS) [set=1, binding=0-7]
@@ -55,6 +57,7 @@ enum DESCRIPTOR_SET_LAYOUT
 enum PIPELINE_LAYOUT
 {
   PIPELINE_LAYOUT_STANDARD,
+  PIPELINE_LAYOUT_UBER,
   PIPELINE_LAYOUT_UTILITY,
   PIPELINE_LAYOUT_COMPUTE,
   NUM_PIPELINE_LAYOUTS
diff --git a/Source/Core/VideoBackends/Vulkan/ObjectCache.cpp b/Source/Core/VideoBackends/Vulkan/ObjectCache.cpp
index fd296457e4..353cb6da3f 100644
--- a/Source/Core/VideoBackends/Vulkan/ObjectCache.cpp
+++ b/Source/Core/VideoBackends/Vulkan/ObjectCache.cpp
@@ -123,8 +123,10 @@ bool ObjectCache::CreateDescriptorSetLayouts()
        VK_SHADER_STAGE_FRAGMENT_BIT},
   }};
 
-  static const std::array<VkDescriptorSetLayoutBinding, 1> standard_ssbo_bindings{{
+  // The dynamic veretex loader's vertex buffer must be last here, for similar reasons
+  static const std::array<VkDescriptorSetLayoutBinding, 2> standard_ssbo_bindings{{
       {0, VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, 1, VK_SHADER_STAGE_FRAGMENT_BIT},
+      {1, VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, 1, VK_SHADER_STAGE_VERTEX_BIT},
   }};
 
   static const std::array<VkDescriptorSetLayoutBinding, 1> utility_ubo_bindings{{
@@ -173,6 +175,10 @@ bool ObjectCache::CreateDescriptorSetLayouts()
   if (!g_ActiveConfig.backend_info.bSupportsGeometryShaders)
     create_infos[DESCRIPTOR_SET_LAYOUT_STANDARD_UNIFORM_BUFFERS].bindingCount--;
 
+  // Remove the dynamic vertex loader's buffer if it'll never be needed
+  if (!g_ActiveConfig.backend_info.bSupportsDynamicVertexLoader)
+    create_infos[DESCRIPTOR_SET_LAYOUT_STANDARD_SHADER_STORAGE_BUFFERS].bindingCount--;
+
   for (size_t i = 0; i < create_infos.size(); i++)
   {
     VkResult res = vkCreateDescriptorSetLayout(g_vulkan_context->GetDevice(), &create_infos[i],
@@ -206,6 +212,11 @@ bool ObjectCache::CreatePipelineLayouts()
       m_descriptor_set_layouts[DESCRIPTOR_SET_LAYOUT_STANDARD_SAMPLERS],
       m_descriptor_set_layouts[DESCRIPTOR_SET_LAYOUT_STANDARD_SHADER_STORAGE_BUFFERS],
   };
+  const std::array<VkDescriptorSetLayout, 3> uber_sets{
+      m_descriptor_set_layouts[DESCRIPTOR_SET_LAYOUT_STANDARD_UNIFORM_BUFFERS],
+      m_descriptor_set_layouts[DESCRIPTOR_SET_LAYOUT_STANDARD_SAMPLERS],
+      m_descriptor_set_layouts[DESCRIPTOR_SET_LAYOUT_STANDARD_SHADER_STORAGE_BUFFERS],
+  };
   const std::array<VkDescriptorSetLayout, 2> utility_sets{
       m_descriptor_set_layouts[DESCRIPTOR_SET_LAYOUT_UTILITY_UNIFORM_BUFFER],
       m_descriptor_set_layouts[DESCRIPTOR_SET_LAYOUT_UTILITY_SAMPLERS],
@@ -220,6 +231,10 @@ bool ObjectCache::CreatePipelineLayouts()
       {VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO, nullptr, 0,
        static_cast<u32>(standard_sets.size()), standard_sets.data(), 0, nullptr},
 
+      // Uber
+      {VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO, nullptr, 0,
+       static_cast<u32>(uber_sets.size()), uber_sets.data(), 0, nullptr},
+
       // Utility
       {VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO, nullptr, 0,
        static_cast<u32>(utility_sets.size()), utility_sets.data(), 0, nullptr},
@@ -232,6 +247,10 @@ bool ObjectCache::CreatePipelineLayouts()
   // If bounding box is unsupported, don't bother with the SSBO descriptor set.
   if (!g_ActiveConfig.backend_info.bSupportsBBox)
     pipeline_layout_info[PIPELINE_LAYOUT_STANDARD].setLayoutCount--;
+  // If neither SSBO-using feature is supported, skip in ubershaders too
+  if (!g_ActiveConfig.backend_info.bSupportsBBox &&
+      !g_ActiveConfig.backend_info.bSupportsDynamicVertexLoader)
+    pipeline_layout_info[PIPELINE_LAYOUT_UBER].setLayoutCount--;
 
   for (size_t i = 0; i < pipeline_layout_info.size(); i++)
   {
diff --git a/Source/Core/VideoBackends/Vulkan/StateTracker.cpp b/Source/Core/VideoBackends/Vulkan/StateTracker.cpp
index db59b16398..686bbce973 100644
--- a/Source/Core/VideoBackends/Vulkan/StateTracker.cpp
+++ b/Source/Core/VideoBackends/Vulkan/StateTracker.cpp
@@ -77,14 +77,23 @@ bool StateTracker::Initialize()
   return true;
 }
 
-void StateTracker::SetVertexBuffer(VkBuffer buffer, VkDeviceSize offset)
+void StateTracker::SetVertexBuffer(VkBuffer buffer, VkDeviceSize offset, u32 size)
 {
-  if (m_vertex_buffer == buffer && m_vertex_buffer_offset == offset)
-    return;
-
-  m_vertex_buffer = buffer;
-  m_vertex_buffer_offset = offset;
-  m_dirty_flags |= DIRTY_FLAG_VERTEX_BUFFER;
+  if (m_vertex_buffer != buffer || m_vertex_buffer_offset != offset)
+  {
+    m_vertex_buffer = buffer;
+    m_vertex_buffer_offset = offset;
+    m_dirty_flags |= DIRTY_FLAG_VERTEX_BUFFER;
+  }
+  if (m_bindings.gx_uber_vertex_ssbo.buffer != buffer ||
+      m_bindings.gx_uber_vertex_ssbo.offset != offset ||
+      m_bindings.gx_uber_vertex_ssbo.range != size)
+  {
+    m_bindings.gx_uber_vertex_ssbo.buffer = buffer;
+    m_bindings.gx_uber_vertex_ssbo.offset = offset;
+    m_bindings.gx_uber_vertex_ssbo.range = size;
+    m_dirty_flags |= DIRTY_FLAG_GX_SSBO;
+  }
 }
 
 void StateTracker::SetIndexBuffer(VkBuffer buffer, VkDeviceSize offset, VkIndexType type)
@@ -366,8 +375,13 @@ bool StateTracker::Bind()
 
   // Re-bind parts of the pipeline
   const VkCommandBuffer command_buffer = g_command_buffer_mgr->GetCurrentCommandBuffer();
-  if (m_dirty_flags & DIRTY_FLAG_VERTEX_BUFFER)
+  const bool needs_vertex_buffer = !g_ActiveConfig.backend_info.bSupportsDynamicVertexLoader ||
+                                   m_pipeline->GetUsage() != AbstractPipelineUsage::GXUber;
+  if (needs_vertex_buffer && (m_dirty_flags & DIRTY_FLAG_VERTEX_BUFFER))
+  {
     vkCmdBindVertexBuffers(command_buffer, 0, 1, &m_vertex_buffer, &m_vertex_buffer_offset);
+    m_dirty_flags &= ~DIRTY_FLAG_VERTEX_BUFFER;
+  }
 
   if (m_dirty_flags & DIRTY_FLAG_INDEX_BUFFER)
     vkCmdBindIndexBuffer(command_buffer, m_index_buffer, m_index_buffer_offset, m_index_type);
@@ -381,8 +395,8 @@ bool StateTracker::Bind()
   if (m_dirty_flags & DIRTY_FLAG_SCISSOR)
     vkCmdSetScissor(command_buffer, 0, 1, &m_scissor);
 
-  m_dirty_flags &= ~(DIRTY_FLAG_VERTEX_BUFFER | DIRTY_FLAG_INDEX_BUFFER | DIRTY_FLAG_PIPELINE |
-                     DIRTY_FLAG_VIEWPORT | DIRTY_FLAG_SCISSOR);
+  m_dirty_flags &=
+      ~(DIRTY_FLAG_INDEX_BUFFER | DIRTY_FLAG_PIPELINE | DIRTY_FLAG_VIEWPORT | DIRTY_FLAG_SCISSOR);
   return true;
 }
 
@@ -452,7 +466,7 @@ void StateTracker::EndClearRenderPass()
 
 bool StateTracker::UpdateDescriptorSet()
 {
-  if (m_pipeline->GetUsage() == AbstractPipelineUsage::GX)
+  if (m_pipeline->GetUsage() != AbstractPipelineUsage::Utility)
     return UpdateGXDescriptorSet();
   else
     return UpdateUtilityDescriptorSet();
@@ -462,7 +476,7 @@ bool StateTracker::UpdateGXDescriptorSet()
 {
   const size_t MAX_DESCRIPTOR_WRITES = NUM_UBO_DESCRIPTOR_SET_BINDINGS +  // UBO
                                        1 +                                // Samplers
-                                       1;                                 // SSBO
+                                       2;                                 // SSBO
   std::array<VkWriteDescriptorSet, MAX_DESCRIPTOR_WRITES> writes;
   u32 num_writes = 0;
 
@@ -516,7 +530,12 @@ bool StateTracker::UpdateGXDescriptorSet()
     m_dirty_flags = (m_dirty_flags & ~DIRTY_FLAG_GX_SAMPLERS) | DIRTY_FLAG_DESCRIPTOR_SETS;
   }
 
-  if (g_ActiveConfig.backend_info.bSupportsBBox &&
+  const bool needs_bbox_ssbo = g_ActiveConfig.backend_info.bSupportsBBox;
+  const bool needs_vertex_ssbo = g_ActiveConfig.backend_info.bSupportsDynamicVertexLoader &&
+                                 m_pipeline->GetUsage() == AbstractPipelineUsage::GXUber;
+  const bool needs_ssbo = needs_bbox_ssbo || needs_vertex_ssbo;
+
+  if (needs_ssbo &&
       (m_dirty_flags & DIRTY_FLAG_GX_SSBO || m_gx_descriptor_sets[2] == VK_NULL_HANDLE))
   {
     m_gx_descriptor_sets[2] =
@@ -528,6 +547,21 @@ bool StateTracker::UpdateGXDescriptorSet()
     writes[num_writes++] = {
         VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET, nullptr, m_gx_descriptor_sets[2], 0,      0, 1,
         VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,      nullptr, &m_bindings.ssbo,        nullptr};
+
+    if (g_ActiveConfig.backend_info.bSupportsDynamicVertexLoader)
+    {
+      writes[num_writes++] = {VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
+                              nullptr,
+                              m_gx_descriptor_sets[2],
+                              1,
+                              0,
+                              1,
+                              VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+                              nullptr,
+                              &m_bindings.gx_uber_vertex_ssbo,
+                              nullptr};
+    }
+
     m_dirty_flags = (m_dirty_flags & ~DIRTY_FLAG_GX_SSBO) | DIRTY_FLAG_DESCRIPTOR_SETS;
   }
 
@@ -538,9 +572,7 @@ bool StateTracker::UpdateGXDescriptorSet()
   {
     vkCmdBindDescriptorSets(g_command_buffer_mgr->GetCurrentCommandBuffer(),
                             VK_PIPELINE_BIND_POINT_GRAPHICS, m_pipeline->GetVkPipelineLayout(), 0,
-                            g_ActiveConfig.backend_info.bSupportsBBox ?
-                                NUM_GX_DESCRIPTOR_SETS :
-                                (NUM_GX_DESCRIPTOR_SETS - 1),
+                            needs_ssbo ? NUM_GX_DESCRIPTOR_SETS : (NUM_GX_DESCRIPTOR_SETS - 1),
                             m_gx_descriptor_sets.data(),
                             g_ActiveConfig.backend_info.bSupportsGeometryShaders ?
                                 NUM_UBO_DESCRIPTOR_SET_BINDINGS :
diff --git a/Source/Core/VideoBackends/Vulkan/StateTracker.h b/Source/Core/VideoBackends/Vulkan/StateTracker.h
index 7184a2ae4c..bb7311d932 100644
--- a/Source/Core/VideoBackends/Vulkan/StateTracker.h
+++ b/Source/Core/VideoBackends/Vulkan/StateTracker.h
@@ -32,7 +32,7 @@ public:
 
   VKFramebuffer* GetFramebuffer() const { return m_framebuffer; }
   const VKPipeline* GetPipeline() const { return m_pipeline; }
-  void SetVertexBuffer(VkBuffer buffer, VkDeviceSize offset);
+  void SetVertexBuffer(VkBuffer buffer, VkDeviceSize offset, u32 size);
   void SetIndexBuffer(VkBuffer buffer, VkDeviceSize offset, VkIndexType type);
   void SetFramebuffer(VKFramebuffer* framebuffer);
   void SetPipeline(const VKPipeline* pipeline);
@@ -145,6 +145,7 @@ private:
     std::array<VkDescriptorImageInfo, NUM_PIXEL_SHADER_SAMPLERS> samplers;
     std::array<VkBufferView, NUM_COMPUTE_TEXEL_BUFFERS> texel_buffers;
     VkDescriptorBufferInfo ssbo;
+    VkDescriptorBufferInfo gx_uber_vertex_ssbo;
     VkDescriptorImageInfo image_texture;
   } m_bindings = {};
   std::array<VkDescriptorSet, NUM_GX_DESCRIPTOR_SETS> m_gx_descriptor_sets = {};
diff --git a/Source/Core/VideoBackends/Vulkan/VKPipeline.cpp b/Source/Core/VideoBackends/Vulkan/VKPipeline.cpp
index f21fb9bf5e..564f93d802 100644
--- a/Source/Core/VideoBackends/Vulkan/VKPipeline.cpp
+++ b/Source/Core/VideoBackends/Vulkan/VKPipeline.cpp
@@ -251,6 +251,9 @@ std::unique_ptr<VKPipeline> VKPipeline::Create(const AbstractPipelineConfig& con
   case AbstractPipelineUsage::GX:
     pipeline_layout = g_object_cache->GetPipelineLayout(PIPELINE_LAYOUT_STANDARD);
     break;
+  case AbstractPipelineUsage::GXUber:
+    pipeline_layout = g_object_cache->GetPipelineLayout(PIPELINE_LAYOUT_UBER);
+    break;
   case AbstractPipelineUsage::Utility:
     pipeline_layout = g_object_cache->GetPipelineLayout(PIPELINE_LAYOUT_UTILITY);
     break;
diff --git a/Source/Core/VideoBackends/Vulkan/VKVertexManager.cpp b/Source/Core/VideoBackends/Vulkan/VKVertexManager.cpp
index 1c47ea0b8e..146eb2921e 100644
--- a/Source/Core/VideoBackends/Vulkan/VKVertexManager.cpp
+++ b/Source/Core/VideoBackends/Vulkan/VKVertexManager.cpp
@@ -62,7 +62,8 @@ bool VertexManager::Initialize()
     return false;
 
   m_vertex_stream_buffer =
-      StreamBuffer::Create(VK_BUFFER_USAGE_VERTEX_BUFFER_BIT, VERTEX_STREAM_BUFFER_SIZE);
+      StreamBuffer::Create(VK_BUFFER_USAGE_VERTEX_BUFFER_BIT | VK_BUFFER_USAGE_STORAGE_BUFFER_BIT,
+                           VERTEX_STREAM_BUFFER_SIZE);
   m_index_stream_buffer =
       StreamBuffer::Create(VK_BUFFER_USAGE_INDEX_BUFFER_BIT, INDEX_STREAM_BUFFER_SIZE);
   m_uniform_stream_buffer =
@@ -186,7 +187,8 @@ void VertexManager::CommitBuffer(u32 num_vertices, u32 vertex_stride, u32 num_in
   ADDSTAT(g_stats.this_frame.bytes_vertex_streamed, static_cast<int>(vertex_data_size));
   ADDSTAT(g_stats.this_frame.bytes_index_streamed, static_cast<int>(index_data_size));
 
-  StateTracker::GetInstance()->SetVertexBuffer(m_vertex_stream_buffer->GetBuffer(), 0);
+  StateTracker::GetInstance()->SetVertexBuffer(m_vertex_stream_buffer->GetBuffer(), 0,
+                                               VERTEX_STREAM_BUFFER_SIZE);
   StateTracker::GetInstance()->SetIndexBuffer(m_index_stream_buffer->GetBuffer(), 0,
                                               VK_INDEX_TYPE_UINT16);
 }
diff --git a/Source/Core/VideoBackends/Vulkan/VulkanContext.cpp b/Source/Core/VideoBackends/Vulkan/VulkanContext.cpp
index cad326c83a..5b970cc9b1 100644
--- a/Source/Core/VideoBackends/Vulkan/VulkanContext.cpp
+++ b/Source/Core/VideoBackends/Vulkan/VulkanContext.cpp
@@ -295,6 +295,7 @@ void VulkanContext::PopulateBackendInfo(VideoConfig* config)
   config->backend_info.bSupportsLodBiasInSampler = false;          // Dependent on OS.
   config->backend_info.bSupportsSettingObjectNames = false;        // Dependent on features.
   config->backend_info.bSupportsPartialMultisampleResolve = true;  // Assumed support.
+  config->backend_info.bSupportsDynamicVertexLoader = true;        // Assumed support.
 }
 
 void VulkanContext::PopulateBackendInfoAdapters(VideoConfig* config, const GPUList& gpu_list)
diff --git a/Source/Core/VideoCommon/AbstractPipeline.h b/Source/Core/VideoCommon/AbstractPipeline.h
index 0be46eaf5a..60381d0cd9 100644
--- a/Source/Core/VideoCommon/AbstractPipeline.h
+++ b/Source/Core/VideoCommon/AbstractPipeline.h
@@ -20,6 +20,8 @@ class NativeVertexFormat;
 //       - Per-stage UBO (VS/GS/PS, VS constants accessible from PS)
 //       - 8 combined image samplers (accessible from PS)
 //       - 1 SSBO, accessible from PS if bounding box is enabled
+//   - GX Uber
+//       - Same as GX, plus one VS SSBO for vertices if dynamic vertex loading is enabled
 //   - Utility
 //       - Single UBO, accessible from all stages [set=0, binding=1]
 //       - 8 combined image samplers (accessible from PS) [set=1, binding=0-7]
@@ -32,6 +34,7 @@ class NativeVertexFormat;
 enum class AbstractPipelineUsage
 {
   GX,
+  GXUber,
   Utility
 };
 
diff --git a/Source/Core/VideoCommon/ConstantManager.h b/Source/Core/VideoCommon/ConstantManager.h
index 9fd4c060cd..c3427ba37f 100644
--- a/Source/Core/VideoCommon/ConstantManager.h
+++ b/Source/Core/VideoCommon/ConstantManager.h
@@ -93,6 +93,13 @@ struct VertexShaderConstants
 
   float4 cached_tangent;
   float4 cached_binormal;
+  // For UberShader vertex loader
+  u32 vertex_stride;
+  std::array<u32, 3> vertex_offset_normals;
+  u32 vertex_offset_position;
+  u32 vertex_offset_posmtx;
+  std::array<u32, 2> vertex_offset_colors;
+  std::array<u32, 8> vertex_offset_texcoords;
 };
 
 struct GeometryShaderConstants
diff --git a/Source/Core/VideoCommon/NativeVertexFormat.h b/Source/Core/VideoCommon/NativeVertexFormat.h
index 55e1a40178..89f3a06db5 100644
--- a/Source/Core/VideoCommon/NativeVertexFormat.h
+++ b/Source/Core/VideoCommon/NativeVertexFormat.h
@@ -58,9 +58,9 @@ struct PortableVertexDeclaration
   int stride;
 
   AttributeFormat position;
-  AttributeFormat normals[3];
-  AttributeFormat colors[2];
-  AttributeFormat texcoords[8];
+  std::array<AttributeFormat, 3> normals;
+  std::array<AttributeFormat, 2> colors;
+  std::array<AttributeFormat, 8> texcoords;
   AttributeFormat posmtx;
 
   inline bool operator<(const PortableVertexDeclaration& b) const
diff --git a/Source/Core/VideoCommon/ShaderCache.cpp b/Source/Core/VideoCommon/ShaderCache.cpp
index 6049fd29a7..f281024f22 100644
--- a/Source/Core/VideoCommon/ShaderCache.cpp
+++ b/Source/Core/VideoCommon/ShaderCache.cpp
@@ -588,10 +588,10 @@ AbstractPipelineConfig ShaderCache::GetGXPipelineConfig(
     const NativeVertexFormat* vertex_format, const AbstractShader* vertex_shader,
     const AbstractShader* geometry_shader, const AbstractShader* pixel_shader,
     const RasterizationState& rasterization_state, const DepthState& depth_state,
-    const BlendingState& blending_state)
+    const BlendingState& blending_state, AbstractPipelineUsage usage)
 {
   AbstractPipelineConfig config = {};
-  config.usage = AbstractPipelineUsage::GX;
+  config.usage = usage;
   config.vertex_format = vertex_format;
   config.vertex_shader = vertex_shader;
   config.geometry_shader = geometry_shader;
@@ -735,7 +735,7 @@ ShaderCache::GetGXPipelineConfig(const GXPipelineUid& config_in)
   }
 
   return GetGXPipelineConfig(config.vertex_format, vs, gs, ps, config.rasterization_state,
-                             config.depth_state, config.blending_state);
+                             config.depth_state, config.blending_state, AbstractPipelineUsage::GX);
 }
 
 /// Edits the UID based on driver bugs and other special configurations
@@ -743,6 +743,8 @@ static GXUberPipelineUid ApplyDriverBugs(const GXUberPipelineUid& in)
 {
   GXUberPipelineUid out;
   memcpy(&out, &in, sizeof(out));  // Copy padding
+  if (g_ActiveConfig.backend_info.bSupportsDynamicVertexLoader)
+    out.vertex_format = nullptr;
   if (g_ActiveConfig.backend_info.bSupportsFramebufferFetch)
   {
     // Always blend in shader
@@ -798,7 +800,8 @@ ShaderCache::GetGXPipelineConfig(const GXUberPipelineUid& config_in)
   }
 
   return GetGXPipelineConfig(config.vertex_format, vs, gs, ps, config.rasterization_state,
-                             config.depth_state, config.blending_state);
+                             config.depth_state, config.blending_state,
+                             AbstractPipelineUsage::GXUber);
 }
 
 const AbstractPipeline* ShaderCache::InsertGXPipeline(const GXPipelineUid& config,
@@ -1233,32 +1236,32 @@ void ShaderCache::QueueUberShaderPipelines()
   dummy_vertex_decl.stride = sizeof(float) * 4;
   NativeVertexFormat* dummy_vertex_format =
       VertexLoaderManager::GetUberVertexFormat(dummy_vertex_decl);
-  auto QueueDummyPipeline = [&](const UberShader::VertexShaderUid& vs_uid,
-                                const GeometryShaderUid& gs_uid,
-                                const UberShader::PixelShaderUid& ps_uid) {
-    GXUberPipelineUid config;
-    config.vertex_format = dummy_vertex_format;
-    config.vs_uid = vs_uid;
-    config.gs_uid = gs_uid;
-    config.ps_uid = ps_uid;
-    config.rasterization_state = RenderState::GetCullBackFaceRasterizationState(
-        static_cast<PrimitiveType>(gs_uid.GetUidData()->primitive_type));
-    config.depth_state = RenderState::GetNoDepthTestingDepthState();
-    config.blending_state = RenderState::GetNoBlendingBlendState();
-    if (ps_uid.GetUidData()->uint_output)
-    {
-      // uint_output is only ever enabled when logic ops are enabled.
-      config.blending_state.logicopenable = true;
-      config.blending_state.logicmode = LogicOp::And;
-    }
+  auto QueueDummyPipeline =
+      [&](const UberShader::VertexShaderUid& vs_uid, const GeometryShaderUid& gs_uid,
+          const UberShader::PixelShaderUid& ps_uid, const BlendingState& blend) {
+        GXUberPipelineUid config;
+        config.vertex_format = dummy_vertex_format;
+        config.vs_uid = vs_uid;
+        config.gs_uid = gs_uid;
+        config.ps_uid = ps_uid;
+        config.rasterization_state = RenderState::GetCullBackFaceRasterizationState(
+            static_cast<PrimitiveType>(gs_uid.GetUidData()->primitive_type));
+        config.depth_state = RenderState::GetNoDepthTestingDepthState();
+        config.blending_state = blend;
+        if (ps_uid.GetUidData()->uint_output)
+        {
+          // uint_output is only ever enabled when logic ops are enabled.
+          config.blending_state.logicopenable = true;
+          config.blending_state.logicmode = LogicOp::And;
+        }
 
-    auto iter = m_gx_uber_pipeline_cache.find(config);
-    if (iter != m_gx_uber_pipeline_cache.end())
-      return;
+        auto iter = m_gx_uber_pipeline_cache.find(config);
+        if (iter != m_gx_uber_pipeline_cache.end())
+          return;
 
-    auto& entry = m_gx_uber_pipeline_cache[config];
-    entry.second = false;
-  };
+        auto& entry = m_gx_uber_pipeline_cache[config];
+        entry.second = false;
+      };
 
   // Populate the pipeline configs with empty entries, these will be compiled afterwards.
   UberShader::EnumerateVertexShaderUids([&](const UberShader::VertexShaderUid& vuid) {
@@ -1275,7 +1278,45 @@ void ShaderCache::QueueUberShaderPipelines()
         {
           return;
         }
-        QueueDummyPipeline(vuid, guid, cleared_puid);
+        BlendingState blend = RenderState::GetNoBlendingBlendState();
+        QueueDummyPipeline(vuid, guid, cleared_puid, blend);
+        if (g_ActiveConfig.backend_info.bSupportsDynamicVertexLoader)
+        {
+          // Not all GPUs need all the pipeline state compiled into shaders, so they tend to key
+          // compiled shaders based on some subset of the pipeline state.
+          // Some test results:
+          //   (GPUs tested: AMD Radeon Pro 5600M, Nvidia GT 750M, Intel UHD 630,
+          //    Intel Iris Pro 5200, Apple M1)
+          // MacOS Metal:
+          //  - AMD, Nvidia, Intel GPUs: Shaders are keyed on vertex layout and whether or not
+          //    dual source blend is enabled.  That's it.
+          //  - Apple GPUs: Shaders are keyed on vertex layout and all blending settings.  We use
+          //    framebuffer fetch here, so the only blending settings used by ubershaders are the
+          //    alphaupdate and colorupdate ones.  Also keyed on primitive type, but Metal supports
+          //    setting it to "unknown" and we do for ubershaders (but MoltenVK won't).
+          // Windows Vulkan:
+          //  - AMD, Nvidia: Definitely keyed on dual source blend, but the others seem more random
+          //    Changing a setting on one shader will require a recompile, but changing the same
+          //    setting on another won't.  Compiling a copy with alphaupdate off, colorupdate off,
+          //    and one with DSB on seems to get pretty good coverage though.
+          // Windows D3D12:
+          //  - AMD: Keyed on dual source blend and vertex layout
+          //  - Nvidia Kepler: No recompiles for changes to vertex layout or blend
+          blend.alphaupdate = false;
+          QueueDummyPipeline(vuid, guid, cleared_puid, blend);
+          blend.alphaupdate = true;
+          blend.colorupdate = false;
+          QueueDummyPipeline(vuid, guid, cleared_puid, blend);
+          blend.colorupdate = true;
+          if (!cleared_puid.GetUidData()->no_dual_src && !cleared_puid.GetUidData()->uint_output)
+          {
+            blend.blendenable = true;
+            blend.usedualsrc = true;
+            blend.srcfactor = SrcBlendFactor::SrcAlpha;
+            blend.dstfactor = DstBlendFactor::InvSrcAlpha;
+            QueueDummyPipeline(vuid, guid, cleared_puid, blend);
+          }
+        }
       });
     });
   });
diff --git a/Source/Core/VideoCommon/ShaderCache.h b/Source/Core/VideoCommon/ShaderCache.h
index ee59fe9ae7..2721cd7319 100644
--- a/Source/Core/VideoCommon/ShaderCache.h
+++ b/Source/Core/VideoCommon/ShaderCache.h
@@ -151,7 +151,7 @@ private:
   GetGXPipelineConfig(const NativeVertexFormat* vertex_format, const AbstractShader* vertex_shader,
                       const AbstractShader* geometry_shader, const AbstractShader* pixel_shader,
                       const RasterizationState& rasterization_state, const DepthState& depth_state,
-                      const BlendingState& blending_state);
+                      const BlendingState& blending_state, AbstractPipelineUsage usage);
   std::optional<AbstractPipelineConfig> GetGXPipelineConfig(const GXPipelineUid& uid);
   std::optional<AbstractPipelineConfig> GetGXPipelineConfig(const GXUberPipelineUid& uid);
   const AbstractPipeline* InsertGXPipeline(const GXPipelineUid& config,
diff --git a/Source/Core/VideoCommon/ShaderGenCommon.cpp b/Source/Core/VideoCommon/ShaderGenCommon.cpp
index b9965421cc..7b407f9aaf 100644
--- a/Source/Core/VideoCommon/ShaderGenCommon.cpp
+++ b/Source/Core/VideoCommon/ShaderGenCommon.cpp
@@ -43,6 +43,7 @@ ShaderHostConfig ShaderHostConfig::GetCurrent()
   bits.manual_texture_sampling_custom_texture_sizes =
       g_ActiveConfig.ManualTextureSamplingWithHiResTextures();
   bits.backend_sampler_lod_bias = g_ActiveConfig.backend_info.bSupportsLodBiasInSampler;
+  bits.backend_dynamic_vertex_loader = g_ActiveConfig.backend_info.bSupportsDynamicVertexLoader;
   return bits;
 }
 
diff --git a/Source/Core/VideoCommon/ShaderGenCommon.h b/Source/Core/VideoCommon/ShaderGenCommon.h
index 1cbff2bfcf..73fa68af03 100644
--- a/Source/Core/VideoCommon/ShaderGenCommon.h
+++ b/Source/Core/VideoCommon/ShaderGenCommon.h
@@ -177,6 +177,7 @@ union ShaderHostConfig
   BitField<24, 1, bool, u32> manual_texture_sampling;
   BitField<25, 1, bool, u32> manual_texture_sampling_custom_texture_sizes;
   BitField<26, 1, bool, u32> backend_sampler_lod_bias;
+  BitField<27, 1, bool, u32> backend_dynamic_vertex_loader;
 
   static ShaderHostConfig GetCurrent();
 };
@@ -302,6 +303,15 @@ static const char s_shader_uniforms[] = "\tuint    components;\n"
                                         "\tuint4   xfmem_pack1[8];\n"
                                         "\tfloat4 " I_CACHED_TANGENT ";\n"
                                         "\tfloat4 " I_CACHED_BINORMAL ";\n"
+                                        "\tuint vertex_stride;\n"
+                                        "\tuint vertex_offset_rawnormal;\n"
+                                        "\tuint vertex_offset_rawtangent;\n"
+                                        "\tuint vertex_offset_rawbinormal;\n"
+                                        "\tuint vertex_offset_rawpos;\n"
+                                        "\tuint vertex_offset_posmtx;\n"
+                                        "\tuint vertex_offset_rawcolor0;\n"
+                                        "\tuint vertex_offset_rawcolor1;\n"
+                                        "\tuint4 vertex_offset_rawtex[2];\n"  // std140 is pain
                                         "\t#define xfmem_texMtxInfo(i) (xfmem_pack1[(i)].x)\n"
                                         "\t#define xfmem_postMtxInfo(i) (xfmem_pack1[(i)].y)\n"
                                         "\t#define xfmem_color(i) (xfmem_pack1[(i)].z)\n"
diff --git a/Source/Core/VideoCommon/UberShaderVertex.cpp b/Source/Core/VideoCommon/UberShaderVertex.cpp
index 67dd84bb7f..04f2b27a14 100644
--- a/Source/Core/VideoCommon/UberShaderVertex.cpp
+++ b/Source/Core/VideoCommon/UberShaderVertex.cpp
@@ -22,7 +22,11 @@ VertexShaderUid GetVertexShaderUid()
   return out;
 }
 
-static void GenVertexShaderTexGens(APIType api_type, u32 num_texgen, ShaderCode& out);
+static void GenVertexShaderTexGens(APIType api_type, const ShaderHostConfig& host_config,
+                                   u32 num_texgen, ShaderCode& out);
+static void LoadVertexAttribute(ShaderCode& code, const ShaderHostConfig& host_config, u32 indent,
+                                std::string_view name, std::string_view shader_type,
+                                std::string_view stored_type, std::string_view offset_name = {});
 
 ShaderCode GenVertexShader(APIType api_type, const ShaderHostConfig& host_config,
                            const vertex_ubershader_uid_data* uid_data)
@@ -50,15 +54,99 @@ ShaderCode GenVertexShader(APIType api_type, const ShaderHostConfig& host_config
   WriteBitfieldExtractHeader(out, api_type, host_config);
   WriteLightingFunction(out);
 
-  out.Write("ATTRIBUTE_LOCATION({}) in float4 rawpos;\n", SHADER_POSITION_ATTRIB);
-  out.Write("ATTRIBUTE_LOCATION({}) in uint4 posmtx;\n", SHADER_POSMTX_ATTRIB);
-  out.Write("ATTRIBUTE_LOCATION({}) in float3 rawnormal;\n", SHADER_NORMAL_ATTRIB);
-  out.Write("ATTRIBUTE_LOCATION({}) in float3 rawtangent;\n", SHADER_TANGENT_ATTRIB);
-  out.Write("ATTRIBUTE_LOCATION({}) in float3 rawbinormal;\n", SHADER_BINORMAL_ATTRIB);
-  out.Write("ATTRIBUTE_LOCATION({}) in float4 rawcolor0;\n", SHADER_COLOR0_ATTRIB);
-  out.Write("ATTRIBUTE_LOCATION({}) in float4 rawcolor1;\n", SHADER_COLOR1_ATTRIB);
-  for (int i = 0; i < 8; ++i)
-    out.Write("ATTRIBUTE_LOCATION({}) in float3 rawtex{};\n", SHADER_TEXTURE0_ATTRIB + i, i);
+  if (host_config.backend_dynamic_vertex_loader)
+  {
+    out.Write(R"(
+SSBO_BINDING(1) readonly restrict buffer Vertices {{
+  uint vertex_buffer[];
+}};
+)");
+    if (api_type == APIType::D3D)
+    {
+      // Write a function to get an offset into vertex_buffer corresponding to this vertex.
+      // This must be done differently for D3D compared to OpenGL/Vulkan/Metal, as on OpenGL, etc.,
+      // gl_VertexID starts counting at the base vertex specified in glDrawElementsBaseVertex,
+      // while on D3D, SV_VertexID (which spirv-cross translates gl_VertexID into) starts counting
+      // at 0 regardless of the BaseVertexLocation value passed to DrawIndexed.  In both cases,
+      // offset 0 of vertex_buffer corresponds to index 0 with basevertex set to 0, so we have to
+      // manually apply the basevertex offset for D3D
+      // D3D12 uses a root constant for this uniform, since it changes with every draw.
+      // D3D11 doesn't currently support dynamic vertex loader, and we'll have to figure something
+      // out for it if we want to support it in the future.
+      out.Write("UBO_BINDING(std140, 3) uniform DX_Constants {{\n"
+                "  uint base_vertex;\n"
+                "}};\n\n"
+                "uint GetVertexBaseOffset() {{\n"
+                "  return (gl_VertexID + base_vertex) * vertex_stride;\n"
+                "}}\n");
+    }
+    else
+    {
+      out.Write("uint GetVertexBaseOffset() {{\n"
+                "  return gl_VertexID * vertex_stride;\n"
+                "}}\n");
+    }
+
+    out.Write(R"(
+uint4 load_input_uint4_ubyte4(uint vtx_offset, uint attr_offset) {{
+  uint value = vertex_buffer[vtx_offset + attr_offset];
+  return uint4(value & 0xff, (value >> 8) & 0xff, (value >> 16) & 0xff, value >> 24);
+}}
+
+float4 load_input_float4_ubyte4(uint vtx_offset, uint attr_offset) {{
+  return float4(load_input_uint4_ubyte4(vtx_offset, attr_offset)) / 255.0f;
+}}
+
+float3 load_input_float3_float3(uint vtx_offset, uint attr_offset) {{
+  uint offset = vtx_offset + attr_offset;
+  return float3(uintBitsToFloat(vertex_buffer[offset + 0]),
+                uintBitsToFloat(vertex_buffer[offset + 1]),
+                uintBitsToFloat(vertex_buffer[offset + 2]));
+}}
+
+float4 load_input_float4_rawpos(uint vtx_offset, uint attr_offset) {{
+  uint components = attr_offset >> 16;
+  uint offset = vtx_offset + (attr_offset & 0xffff);
+  if (components < 3)
+    return float4(uintBitsToFloat(vertex_buffer[offset + 0]),
+                  uintBitsToFloat(vertex_buffer[offset + 1]),
+                  0.0f, 1.0f);
+  else
+    return float4(uintBitsToFloat(vertex_buffer[offset + 0]),
+                  uintBitsToFloat(vertex_buffer[offset + 1]),
+                  uintBitsToFloat(vertex_buffer[offset + 2]),
+                  1.0f);
+}}
+
+float3 load_input_float3_rawtex(uint vtx_offset, uint attr_offset) {{
+  uint components = attr_offset >> 16;
+  uint offset = vtx_offset + (attr_offset & 0xffff);
+  if (components < 2)
+    return float3(uintBitsToFloat(vertex_buffer[offset + 0]), 0.0f, 0.0f);
+  else if (components < 3)
+    return float3(uintBitsToFloat(vertex_buffer[offset + 0]),
+                  uintBitsToFloat(vertex_buffer[offset + 1]),
+                  0.0f);
+  else
+    return float3(uintBitsToFloat(vertex_buffer[offset + 0]),
+                  uintBitsToFloat(vertex_buffer[offset + 1]),
+                  uintBitsToFloat(vertex_buffer[offset + 2]));
+}}
+
+)");
+  }
+  else
+  {
+    out.Write("ATTRIBUTE_LOCATION({}) in float4 rawpos;\n", SHADER_POSITION_ATTRIB);
+    out.Write("ATTRIBUTE_LOCATION({}) in uint4 posmtx;\n", SHADER_POSMTX_ATTRIB);
+    out.Write("ATTRIBUTE_LOCATION({}) in float3 rawnormal;\n", SHADER_NORMAL_ATTRIB);
+    out.Write("ATTRIBUTE_LOCATION({}) in float3 rawtangent;\n", SHADER_TANGENT_ATTRIB);
+    out.Write("ATTRIBUTE_LOCATION({}) in float3 rawbinormal;\n", SHADER_BINORMAL_ATTRIB);
+    out.Write("ATTRIBUTE_LOCATION({}) in float4 rawcolor0;\n", SHADER_COLOR0_ATTRIB);
+    out.Write("ATTRIBUTE_LOCATION({}) in float4 rawcolor1;\n", SHADER_COLOR1_ATTRIB);
+    for (int i = 0; i < 8; ++i)
+      out.Write("ATTRIBUTE_LOCATION({}) in float3 rawtex{};\n", SHADER_TEXTURE0_ATTRIB + i, i);
+  }
 
   if (host_config.backend_geometry_shaders)
   {
@@ -99,7 +187,12 @@ ShaderCode GenVertexShader(APIType api_type, const ShaderHostConfig& host_config
 
   out.Write("VS_OUTPUT o;\n"
             "\n");
-
+  if (host_config.backend_dynamic_vertex_loader)
+  {
+    out.Write("uint vertex_base_offset = GetVertexBaseOffset();\n");
+  }
+  // rawpos is always needed
+  LoadVertexAttribute(out, host_config, 0, "rawpos", "float4", "rawpos");
   // Transforms
   out.Write("// Position matrix\n"
             "float4 P0;\n"
@@ -113,6 +206,7 @@ ShaderCode GenVertexShader(APIType api_type, const ShaderHostConfig& host_config
             "\n"
             "if ((components & {}u) != 0u) {{ // VB_HAS_POSMTXIDX\n",
             VB_HAS_POSMTXIDX);
+  LoadVertexAttribute(out, host_config, 2, "posmtx", "uint4", "ubyte4");
   out.Write("  // Vertex format has a per-vertex matrix\n"
             "  int posidx = int(posmtx.r);\n"
             "  P0 = " I_TRANSFORMMATRICES "[posidx];\n"
@@ -144,27 +238,40 @@ ShaderCode GenVertexShader(APIType api_type, const ShaderHostConfig& host_config
             "// by lighting calculations and needs to be unit length), the same transform matrix\n"
             "// can do double duty, scaling for emboss mapping, and not scaling for lighting.\n"
             "float3 _normal = float3(0.0, 0.0, 0.0);\n"
-            "if ((components & {}u) != 0u) // VB_HAS_NORMAL\n",
+            "if ((components & {}u) != 0u) // VB_HAS_NORMAL\n"
+            "{{\n",
             VB_HAS_NORMAL);
+  LoadVertexAttribute(out, host_config, 2, "rawnormal", "float3", "float3");
   out.Write("  _normal = normalize(float3(dot(N0, rawnormal), dot(N1, rawnormal), dot(N2, "
             "rawnormal)));\n"
+            "}}\n"
             "\n"
             "float3 _tangent = float3(0.0, 0.0, 0.0);\n"
-            "if ((components & {}u) != 0u) // VB_HAS_TANGENT\n",
+            "if ((components & {}u) != 0u) // VB_HAS_TANGENT\n"
+            "{{\n",
             VB_HAS_TANGENT);
+  LoadVertexAttribute(out, host_config, 2, "rawtangent", "float3", "float3");
   out.Write("  _tangent = float3(dot(N0, rawtangent), dot(N1, rawtangent), dot(N2, rawtangent));\n"
+            "}}\n"
             "else\n"
+            "{{\n"
             "  _tangent = float3(dot(N0, " I_CACHED_TANGENT ".xyz), dot(N1, " I_CACHED_TANGENT
             ".xyz), dot(N2, " I_CACHED_TANGENT ".xyz));\n"
+            "}}\n"
             "\n"
             "float3 _binormal = float3(0.0, 0.0, 0.0);\n"
-            "if ((components & {}u) != 0u) // VB_HAS_BINORMAL\n",
+            "if ((components & {}u) != 0u) // VB_HAS_BINORMAL\n"
+            "{{\n",
             VB_HAS_BINORMAL);
+  LoadVertexAttribute(out, host_config, 2, "rawbinormal", "float3", "float3");
   out.Write("  _binormal = float3(dot(N0, rawbinormal), dot(N1, rawbinormal), dot(N2, "
             "rawbinormal));\n"
+            "}}\n"
             "else\n"
+            "{{\n"
             "  _binormal = float3(dot(N0, " I_CACHED_BINORMAL ".xyz), dot(N1, " I_CACHED_BINORMAL
             ".xyz), dot(N2, " I_CACHED_BINORMAL ".xyz));\n"
+            "}}\n"
             "\n");
 
   // Hardware Lighting
@@ -178,34 +285,40 @@ ShaderCode GenVertexShader(APIType api_type, const ShaderHostConfig& host_config
             "bool use_color_1 = ((components & {0}u) == {0}u); // VB_HAS_COL0 | VB_HAS_COL1\n",
             VB_HAS_COL0 | VB_HAS_COL1);
 
-  out.Write("for (uint color = 0u; color < {}u; color++) {{\n", NUM_XF_COLOR_CHANNELS);
-  out.Write("  if ((color == 0u || use_color_1) && (components & ({}u << color)) != 0u) {{\n",
+  out.Write("if ((components & {0}u) == {0}u) // VB_HAS_COL0 | VB_HAS_COL1\n"
+            "{{\n",
+            VB_HAS_COL0 | VB_HAS_COL1);
+  LoadVertexAttribute(out, host_config, 2, "rawcolor0", "float4", "ubyte4");
+  LoadVertexAttribute(out, host_config, 2, "rawcolor1", "float4", "ubyte4");
+  out.Write("  vertex_color_0 = rawcolor0;\n"
+            "  vertex_color_1 = rawcolor1;\n"
+            "}}\n"
+            "else if ((components & {}u) != 0u) // VB_HAS_COL0\n"
+            "{{\n",
             VB_HAS_COL0);
-  out.Write("    // Use color0 for channel 0, and color1 for channel 1 if both colors 0 and 1 are "
-            "present.\n"
-            "    if (color == 0u)\n"
-            "      vertex_color_0 = rawcolor0;\n"
-            "    else\n"
-            "      vertex_color_1 = rawcolor1;\n"
-            "  }} else if (color == 0u && (components & {}u) != 0u) {{\n",
+  LoadVertexAttribute(out, host_config, 2, "rawcolor0", "float4", "ubyte4");
+  out.Write("  vertex_color_0 = rawcolor0;\n"
+            "  vertex_color_1 = rawcolor0;\n"
+            "}}\n"
+            "else if ((components & {}u) != 0u) // VB_HAS_COL1\n"
+            "{{\n",
             VB_HAS_COL1);
-  out.Write("    // Use color1 for channel 0 if color0 is not present.\n"
-            "    vertex_color_0 = rawcolor1;\n"
-            "  }} else {{\n"
-            "    if (color == 0u)\n"
-            "      vertex_color_0 = missing_color_value;\n"
-            "    else\n"
-            "      vertex_color_1 = missing_color_value;\n"
-            "  }}\n"
+  LoadVertexAttribute(out, host_config, 2, "rawcolor1", "float4", "ubyte4");
+  out.Write("  vertex_color_0 = rawcolor1;\n"
+            "  vertex_color_1 = rawcolor1;\n"
             "}}\n"
-            "\n");
+            "else\n"
+            "{{\n"
+            "  vertex_color_0 = missing_color_value;\n"
+            "  vertex_color_1 = missing_color_value;\n"
+            "}}\n");
 
   WriteVertexLighting(out, api_type, "pos.xyz", "_normal", "vertex_color_0", "vertex_color_1",
                       "o.colors_0", "o.colors_1");
 
   // Texture Coordinates
   if (num_texgen > 0)
-    GenVertexShaderTexGens(api_type, num_texgen, out);
+    GenVertexShaderTexGens(api_type, host_config, num_texgen, out);
 
   if (per_pixel_lighting)
   {
@@ -352,7 +465,8 @@ ShaderCode GenVertexShader(APIType api_type, const ShaderHostConfig& host_config
   return out;
 }
 
-static void GenVertexShaderTexGens(APIType api_type, u32 num_texgen, ShaderCode& out)
+static void GenVertexShaderTexGens(APIType api_type, const ShaderHostConfig& host_config,
+                                   u32 num_texgen, ShaderCode& out)
 {
   // The HLSL compiler complains that the output texture coordinates are uninitialized when trying
   // to dynamically index them.
@@ -377,27 +491,40 @@ static void GenVertexShaderTexGens(APIType api_type, u32 num_texgen, ShaderCode&
   out.Write("    coord.xyz = rawpos.xyz;\n");
   out.Write("    break;\n\n");
   out.Write("  case {:s}:\n", SourceRow::Normal);
-  out.Write("    coord.xyz = ((components & {}u /* VB_HAS_NORMAL */) != 0u) ? rawnormal.xyz : "
-            "coord.xyz;",
+  out.Write("    if ((components & {}u) != 0u) // VB_HAS_NORMAL\n"
+            "    {{\n",
             VB_HAS_NORMAL);
-  out.Write("    break;\n\n");
+  LoadVertexAttribute(out, host_config, 6, "rawnormal", "float3", "float3");
+  out.Write("      coord.xyz = rawnormal.xyz;\n"
+            "    }}\n"
+            "    break;\n\n");
   out.Write("  case {:s}:\n", SourceRow::BinormalT);
-  out.Write("    coord.xyz = ((components & {}u /* VB_HAS_TANGENT */) != 0u) ? rawtangent.xyz : "
-            "coord.xyz;",
+  out.Write("    if ((components & {}u) != 0u) // VB_HAS_TANGENT\n"
+            "    {{\n",
             VB_HAS_TANGENT);
-  out.Write("    break;\n\n");
+  LoadVertexAttribute(out, host_config, 6, "rawtangent", "float3", "float3");
+  out.Write("      coord.xyz = rawtangent.xyz;\n"
+            "    }}\n"
+            "    break;\n\n");
   out.Write("  case {:s}:\n", SourceRow::BinormalB);
-  out.Write("    coord.xyz = ((components & {}u /* VB_HAS_BINORMAL */) != 0u) ? rawbinormal.xyz : "
-            "coord.xyz;",
+  out.Write("    if ((components & {}u) != 0u) // VB_HAS_BINORMAL\n"
+            "    {{\n",
             VB_HAS_BINORMAL);
-  out.Write("    break;\n\n");
+  LoadVertexAttribute(out, host_config, 6, "rawbinormal", "float3", "float3");
+  out.Write("      coord.xyz = rawbinormal.xyz;\n"
+            "    }}\n"
+            "    break;\n\n");
   for (u32 i = 0; i < 8; i++)
   {
     out.Write("  case {:s}:\n", static_cast<SourceRow>(static_cast<u32>(SourceRow::Tex0) + i));
-    out.Write(
-        "    coord = ((components & {}u /* VB_HAS_UV{} */) != 0u) ? float4(rawtex{}.x, rawtex{}.y, "
-        "1.0, 1.0) : coord;\n",
-        VB_HAS_UV0 << i, i, i, i);
+    out.Write("    if ((components & {}u) != 0u) // VB_HAS_UV{}\n"
+              "    {{\n",
+              VB_HAS_UV0 << i, i);
+    LoadVertexAttribute(out, host_config, 6, fmt::format("rawtex{}", i), "float3", "rawtex",
+                        fmt::format("rawtex[{}][{}]", i / 4, i % 4));
+    out.Write("      coord = float4(rawtex{}.x, rawtex{}.y, 1.0f, 1.0f);\n"
+              "    }}\n",
+              i, i);
     out.Write("    break;\n\n");
   }
   out.Write("  }}\n"
@@ -447,14 +574,24 @@ static void GenVertexShaderTexGens(APIType api_type, u32 num_texgen, ShaderCode&
             "    {{\n");
   out.Write("      if ((components & ({}u /* VB_HAS_TEXMTXIDX0 */ << texgen)) != 0u) {{\n",
             VB_HAS_TEXMTXIDX0);
-  out.Write("        // This is messy, due to dynamic indexing of the input texture coordinates.\n"
-            "        // Hopefully the compiler will unroll this whole loop anyway and the switch.\n"
-            "        int tmp = 0;\n"
-            "        switch (texgen) {{\n");
-  for (u32 i = 0; i < num_texgen; i++)
-    out.Write("        case {}u: tmp = int(rawtex{}.z); break;\n", i, i);
-  out.Write("        }}\n"
-            "\n");
+  if (host_config.backend_dynamic_vertex_loader)
+  {
+    out.Write("        int tmp = int(load_input_float3_rawtex(vertex_base_offset, "
+              "vertex_offset_rawtex[texgen / 4][texgen % 4]).z);\n"
+              "\n");
+  }
+  else
+  {
+    out.Write(
+        "        // This is messy, due to dynamic indexing of the input texture coordinates.\n"
+        "        // Hopefully the compiler will unroll this whole loop anyway and the switch.\n"
+        "        int tmp = 0;\n"
+        "        switch (texgen) {{\n");
+    for (u32 i = 0; i < num_texgen; i++)
+      out.Write("        case {}u: tmp = int(rawtex{}.z); break;\n", i, i);
+    out.Write("        }}\n"
+              "\n");
+  }
   out.Write("        if ({} == {:s}) {{\n", BitfieldExtract<&TexMtxInfo::projection>("texMtxInfo"),
             TexSize::STQ);
   out.Write("          output_tex.xyz = float3(dot(coord, " I_TRANSFORMMATRICES "[tmp]),\n"
@@ -514,6 +651,19 @@ static void GenVertexShaderTexGens(APIType api_type, u32 num_texgen, ShaderCode&
             "}}\n");
 }
 
+static void LoadVertexAttribute(ShaderCode& code, const ShaderHostConfig& host_config, u32 indent,
+                                std::string_view name, std::string_view shader_type,
+                                std::string_view stored_type, std::string_view offset_name)
+{
+  if (host_config.backend_dynamic_vertex_loader)
+  {
+    code.Write("{:{}}{} {} = load_input_{}_{}(vertex_base_offset, vertex_offset_{});\n", "", indent,
+               shader_type, name, shader_type, stored_type,
+               offset_name.empty() ? name : offset_name);
+  }
+  // else inputs are always available
+}
+
 void EnumerateVertexShaderUids(const std::function<void(const VertexShaderUid&)>& callback)
 {
   VertexShaderUid uid;
diff --git a/Source/Core/VideoCommon/VertexLoaderManager.cpp b/Source/Core/VideoCommon/VertexLoaderManager.cpp
index 607de66bee..bb843a6120 100644
--- a/Source/Core/VideoCommon/VertexLoaderManager.cpp
+++ b/Source/Core/VideoCommon/VertexLoaderManager.cpp
@@ -353,7 +353,8 @@ int RunVertices(int vtx_attr_group, OpcodeDecoder::Primitive primitive, int coun
     }
     s_current_vtx_fmt = loader->m_native_vertex_format;
     g_current_components = loader->m_native_components;
-    VertexShaderManager::SetVertexFormat(loader->m_native_components);
+    VertexShaderManager::SetVertexFormat(loader->m_native_components,
+                                         loader->m_native_vertex_format->GetVertexDeclaration());
 
     // if cull mode is CULL_ALL, tell VertexManager to skip triangles and quads.
     // They still need to go through vertex loading, because we need to calculate a zfreeze refrence
diff --git a/Source/Core/VideoCommon/VertexShaderManager.cpp b/Source/Core/VideoCommon/VertexShaderManager.cpp
index ce43235977..b74c820271 100644
--- a/Source/Core/VideoCommon/VertexShaderManager.cpp
+++ b/Source/Core/VideoCommon/VertexShaderManager.cpp
@@ -606,13 +606,42 @@ void VertexShaderManager::SetMaterialColorChanged(int index)
   nMaterialsChanged[index] = true;
 }
 
-void VertexShaderManager::SetVertexFormat(u32 components)
+static void UpdateValue(bool* dirty, u32* old_value, u32 new_value)
 {
-  if (components != constants.components)
-  {
-    constants.components = components;
-    dirty = true;
-  }
+  if (*old_value == new_value)
+    return;
+  *old_value = new_value;
+  *dirty = true;
+}
+
+static void UpdateOffset(bool* dirty, bool include_components, u32* old_value,
+                         const AttributeFormat& attribute)
+{
+  if (!attribute.enable)
+    return;
+  u32 new_value = attribute.offset / 4;  // GPU uses uint offsets
+  if (include_components)
+    new_value |= attribute.components << 16;
+  UpdateValue(dirty, old_value, new_value);
+}
+
+template <size_t N>
+static void UpdateOffsets(bool* dirty, bool include_components, std::array<u32, N>* old_value,
+                          const std::array<AttributeFormat, N>& attribute)
+{
+  for (size_t i = 0; i < N; i++)
+    UpdateOffset(dirty, include_components, &(*old_value)[i], attribute[i]);
+}
+
+void VertexShaderManager::SetVertexFormat(u32 components, const PortableVertexDeclaration& format)
+{
+  UpdateValue(&dirty, &constants.components, components);
+  UpdateValue(&dirty, &constants.vertex_stride, format.stride / 4);
+  UpdateOffset(&dirty, true, &constants.vertex_offset_position, format.position);
+  UpdateOffset(&dirty, false, &constants.vertex_offset_posmtx, format.posmtx);
+  UpdateOffsets(&dirty, true, &constants.vertex_offset_texcoords, format.texcoords);
+  UpdateOffsets(&dirty, false, &constants.vertex_offset_colors, format.colors);
+  UpdateOffsets(&dirty, false, &constants.vertex_offset_normals, format.normals);
 }
 
 void VertexShaderManager::SetTexMatrixInfoChanged(int index)
diff --git a/Source/Core/VideoCommon/VertexShaderManager.h b/Source/Core/VideoCommon/VertexShaderManager.h
index 3bddf28fb0..2a8aa7b596 100644
--- a/Source/Core/VideoCommon/VertexShaderManager.h
+++ b/Source/Core/VideoCommon/VertexShaderManager.h
@@ -10,6 +10,7 @@
 #include "VideoCommon/ConstantManager.h"
 
 class PointerWrap;
+struct PortableVertexDeclaration;
 
 // The non-API dependent parts.
 class VertexShaderManager
@@ -29,7 +30,7 @@ public:
   static void SetProjectionChanged();
   static void SetMaterialColorChanged(int index);
 
-  static void SetVertexFormat(u32 components);
+  static void SetVertexFormat(u32 components, const PortableVertexDeclaration& format);
   static void SetTexMatrixInfoChanged(int index);
   static void SetLightingConfigChanged();
 
diff --git a/Source/Core/VideoCommon/VideoConfig.h b/Source/Core/VideoCommon/VideoConfig.h
index 22817d1f61..105840bafa 100644
--- a/Source/Core/VideoCommon/VideoConfig.h
+++ b/Source/Core/VideoCommon/VideoConfig.h
@@ -232,6 +232,7 @@ struct VideoConfig final
     bool bSupportsLodBiasInSampler = false;
     bool bSupportsSettingObjectNames = false;
     bool bSupportsPartialMultisampleResolve = false;
+    bool bSupportsDynamicVertexLoader = false;
   } backend_info;
 
   // Utility