From 2408b881a2fce6b46af2723d6db2e9fc3838fe3b Mon Sep 17 00:00:00 2001
From: Joshua Ashton <joshua@froggi.es>
Date: Tue, 18 Feb 2020 18:28:06 +0000
Subject: [PATCH] [dxso] Use SSBOs for vertex constant buffers if we'd be
 outside of uniform range

Impacts #1375
---
 src/d3d9/d3d9_device.cpp   |  9 +++++----
 src/d3d9/d3d9_device.h     |  2 +-
 src/dxso/dxso_compiler.cpp | 14 ++++++++++++--
 src/dxso/dxso_options.cpp  |  8 +++++++-
 src/dxso/dxso_options.h    |  6 +++++-
 5 files changed, 30 insertions(+), 9 deletions(-)

diff --git a/src/d3d9/d3d9_device.cpp b/src/d3d9/d3d9_device.cpp
index 78599b91b..21279e322 100644
--- a/src/d3d9/d3d9_device.cpp
+++ b/src/d3d9/d3d9_device.cpp
@@ -49,7 +49,6 @@ namespace dxvk {
     , m_multithread    ( BehaviorFlags & D3DCREATE_MULTITHREADED )
     , m_shaderModules  ( new D3D9ShaderModuleSet )
     , m_d3d9Options    ( dxvkDevice, pParent->GetInstance()->config() )
-    , m_dxsoOptions    ( m_dxvkDevice, m_d3d9Options )
     , m_isSWVP         ( (BehaviorFlags & D3DCREATE_SOFTWARE_VERTEXPROCESSING) ? TRUE : FALSE ) {
     // If we can SWVP, then we use an extended constant set
     // as SWVP has many more slots available than HWVP. 
@@ -73,11 +72,13 @@ namespace dxvk {
       ctx->setLogicOpState(loState);
     });
 
-    CreateConstantBuffers();
-
     if (!(BehaviorFlags & D3DCREATE_FPU_PRESERVE))
       SetupFPU();
 
+    m_dxsoOptions = DxsoOptions(this, m_d3d9Options);
+
+    CreateConstantBuffers();
+
     m_availableMemory = DetermineInitialTextureMemory();
   }
 
@@ -4468,7 +4469,7 @@ namespace dxvk {
 
   void D3D9DeviceEx::CreateConstantBuffers() {
     m_consts[DxsoProgramTypes::VertexShader].buffer =
-      CreateConstantBuffer(false,
+      CreateConstantBuffer(m_dxsoOptions.vertexConstantBufferAsSSBO,
                            m_vsLayout.totalSize(),
                            DxsoProgramType::VertexShader,
                            DxsoConstantBuffers::VSConstantBuffer);
diff --git a/src/d3d9/d3d9_device.h b/src/d3d9/d3d9_device.h
index c79bbcca7..3b0d3e7c5 100644
--- a/src/d3d9/d3d9_device.h
+++ b/src/d3d9/d3d9_device.h
@@ -992,7 +992,7 @@ namespace dxvk {
     D3D9UPBufferSlice               m_upBuffer;
 
     const D3D9Options               m_d3d9Options;
-    const DxsoOptions               m_dxsoOptions;
+    DxsoOptions                     m_dxsoOptions;
 
     BOOL                            m_isSWVP;
 
diff --git a/src/dxso/dxso_compiler.cpp b/src/dxso/dxso_compiler.cpp
index d887b9819..134de1c3c 100644
--- a/src/dxso/dxso_compiler.cpp
+++ b/src/dxso/dxso_compiler.cpp
@@ -274,6 +274,9 @@ namespace dxvk {
 
 
   void DxsoCompiler::emitDclConstantBuffer() {
+    const bool asSsbo = m_moduleInfo.options.vertexConstantBufferAsSSBO &&
+      m_programInfo.type() == DxsoProgramType::VertexShader;
+
     std::array<uint32_t, 3> members = {
       // float f[256 or 224 or 8192]
       m_module.defArrayTypeUnique(
@@ -309,7 +312,9 @@ namespace dxvk {
     const uint32_t structType =
       m_module.defStructType(swvp ? 3 : 2, members.data());
 
-    m_module.decorateBlock(structType);
+    m_module.decorate(structType, asSsbo
+      ? spv::DecorationBufferBlock
+      : spv::DecorationBlock);
 
     m_module.memberDecorateOffset(structType, 0, m_layout->floatOffset());
     m_module.memberDecorateOffset(structType, 1, m_layout->intOffset());
@@ -337,9 +342,14 @@ namespace dxvk {
     m_module.decorateDescriptorSet(m_cBuffer, 0);
     m_module.decorateBinding(m_cBuffer, bindingId);
 
+    if (asSsbo)
+      m_module.decorate(m_cBuffer, spv::DecorationNonWritable);
+
     DxvkResourceSlot resource;
     resource.slot   = bindingId;
-    resource.type   = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
+    resource.type   = asSsbo
+      ? VK_DESCRIPTOR_TYPE_STORAGE_BUFFER
+      : VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
     resource.view   = VK_IMAGE_VIEW_TYPE_MAX_ENUM;
     resource.access = VK_ACCESS_UNIFORM_READ_BIT;
     m_resourceSlots.push_back(resource);
diff --git a/src/dxso/dxso_options.cpp b/src/dxso/dxso_options.cpp
index ef402b0c5..41d5fb4c5 100644
--- a/src/dxso/dxso_options.cpp
+++ b/src/dxso/dxso_options.cpp
@@ -1,10 +1,14 @@
 #include "dxso_options.h"
 
+#include "../d3d9/d3d9_device.h"
+
 namespace dxvk {
 
   DxsoOptions::DxsoOptions() {}
 
-  DxsoOptions::DxsoOptions(const Rc<DxvkDevice>& device, const D3D9Options& options) {
+  DxsoOptions::DxsoOptions(D3D9DeviceEx* pDevice, const D3D9Options& options) {
+    const Rc<DxvkDevice> device = pDevice->GetDXVKDevice();
+
     const Rc<DxvkAdapter> adapter = device->adapter();
 
     const DxvkDeviceFeatures& devFeatures = device->features();
@@ -39,6 +43,8 @@ namespace dxvk {
     invariantPosition    = options.invariantPosition;
 
     forceSamplerTypeSpecConstants = options.forceSamplerTypeSpecConstants;
+
+    vertexConstantBufferAsSSBO = pDevice->GetVertexConstantLayout().totalSize() > devInfo.core.properties.limits.maxUniformBufferRange;
   }
 
 }
\ No newline at end of file
diff --git a/src/dxso/dxso_options.h b/src/dxso/dxso_options.h
index 5a8e5c56a..efc877193 100644
--- a/src/dxso/dxso_options.h
+++ b/src/dxso/dxso_options.h
@@ -5,11 +5,12 @@
 
 namespace dxvk {
 
+  class D3D9DeviceEx;
   struct D3D9Options;
 
   struct DxsoOptions {
     DxsoOptions();
-    DxsoOptions(const Rc<DxvkDevice>& device, const D3D9Options& options);
+    DxsoOptions(D3D9DeviceEx* pDevice, const D3D9Options& options);
 
     /// Use a SPIR-V extension to implement D3D-style discards
     bool useDemoteToHelperInvocation = false;
@@ -41,6 +42,9 @@ namespace dxvk {
     /// Always use a spec constant to determine sampler type (instead of just in PS 1.x)
     /// Works around a game bug in Halo CE where it gives cube textures to 2d/volume samplers
     bool forceSamplerTypeSpecConstants;
+
+    /// Should the VS constant buffer be an SSBO (swvp on NV)
+    bool vertexConstantBufferAsSSBO;
   };
 
 }
\ No newline at end of file