From 17e981f360be4d3102a05a1f26811590913d16bc Mon Sep 17 00:00:00 2001
From: Philip Rebohle <philip.rebohle@tu-dortmund.de>
Date: Wed, 21 Mar 2018 04:56:33 +0100
Subject: [PATCH] [dxvk] Increase UAV slot count to 64 for the graphics
 pipeline

D3D11 raised this limit from 8 UAVs in the fragment shader to
64 UAVs in all graphics stages combined.
---
 src/d3d11/d3d11_context.cpp |  9 ++++-----
 src/d3d11/d3d11_context.h   |  3 +--
 src/dxbc/dxbc_util.cpp      | 15 +++++++--------
 src/dxvk/dxvk_limits.h      |  2 +-
 4 files changed, 13 insertions(+), 16 deletions(-)

diff --git a/src/d3d11/d3d11_context.cpp b/src/d3d11/d3d11_context.cpp
index 522a6206d..918b47ecc 100644
--- a/src/d3d11/d3d11_context.cpp
+++ b/src/d3d11/d3d11_context.cpp
@@ -2682,8 +2682,8 @@ namespace dxvk {
     RestoreShaderResources(DxbcProgramType::PixelShader,    m_state.ps.shaderResources);
     RestoreShaderResources(DxbcProgramType::ComputeShader,  m_state.cs.shaderResources);
     
-    RestoreUnorderedAccessViews(DxbcProgramType::PixelShader,   m_state.ps.unorderedAccessViews, D3D11_SIMULTANEOUS_RENDER_TARGET_COUNT);
-    RestoreUnorderedAccessViews(DxbcProgramType::ComputeShader, m_state.cs.unorderedAccessViews, D3D11_1_UAV_SLOT_COUNT);
+    RestoreUnorderedAccessViews(DxbcProgramType::PixelShader,   m_state.ps.unorderedAccessViews);
+    RestoreUnorderedAccessViews(DxbcProgramType::ComputeShader, m_state.cs.unorderedAccessViews);
   }
   
   
@@ -2722,15 +2722,14 @@ namespace dxvk {
   
   void D3D11DeviceContext::RestoreUnorderedAccessViews(
           DxbcProgramType                   Stage,
-          D3D11UnorderedAccessBindings&     Bindings,
-          UINT                              SlotCount) {
+          D3D11UnorderedAccessBindings&     Bindings) {
     const uint32_t uavSlotId = computeResourceSlotId(
       Stage, DxbcBindingType::UnorderedAccessView, 0);
     
     const uint32_t ctrSlotId = computeResourceSlotId(
       Stage, DxbcBindingType::UavCounter, 0);
     
-    for (uint32_t i = 0; i < SlotCount; i++) {
+    for (uint32_t i = 0; i < Bindings.size(); i++) {
       BindUnorderedAccessView(
         uavSlotId + i, ctrSlotId + i,
         Bindings[i].ptr());
diff --git a/src/d3d11/d3d11_context.h b/src/d3d11/d3d11_context.h
index 7f6d79b2a..6b7bef9e8 100644
--- a/src/d3d11/d3d11_context.h
+++ b/src/d3d11/d3d11_context.h
@@ -767,8 +767,7 @@ namespace dxvk {
     
     void RestoreUnorderedAccessViews(
             DxbcProgramType                   Stage,
-            D3D11UnorderedAccessBindings&     Bindings,
-            UINT                              SlotCount);
+            D3D11UnorderedAccessBindings&     Bindings);
     
     DxvkDataSlice AllocUpdateBufferSlice(size_t Size);
     
diff --git a/src/dxbc/dxbc_util.cpp b/src/dxbc/dxbc_util.cpp
index 1c42e0bfc..6197992c1 100644
--- a/src/dxbc/dxbc_util.cpp
+++ b/src/dxbc/dxbc_util.cpp
@@ -6,13 +6,14 @@ namespace dxvk {
           DxbcProgramType shaderStage,
           DxbcBindingType bindingType,
           uint32_t        bindingIndex) {
+    // First resource slot index for per-stage resources
+    const uint32_t stageOffset = 132 + 158 * static_cast<uint32_t>(shaderStage);
+    
     if (shaderStage == DxbcProgramType::ComputeShader) {
       //   0 -  13: Constant buffers
       //  14 -  29: Samplers
       //  30 - 157: Shader resources
-      // 158 - 221: Uniform access views
-      const uint32_t stageOffset = 20 + 158 * 5;
-      
+      // 158 - 221: Unordered access views
       switch (bindingType) {
         case DxbcBindingType::ConstantBuffer:     return bindingIndex + stageOffset +  0;
         case DxbcBindingType::ImageSampler:       return bindingIndex + stageOffset +  14;
@@ -23,15 +24,13 @@ namespace dxvk {
       }
     } else {
       // Global resource slots
-      //   0 -   7: Unordered access views
-      //   8 -  15: UAV counter buffers
-      //  16 -  19: Stream output buffers
+      //   0 -   3: Stream output buffers
+      //   4 -  67: Unordered access views
+      //  68 - 131: UAV counter buffers
       // Per-stage resource slots:
       //   0 -  13: Constant buffers
       //  14 -  29: Samplers
       //  30 - 157: Shader resources
-      const uint32_t stageOffset = 20 + 158 * static_cast<uint32_t>(shaderStage);
-      
       switch (bindingType) {
         case DxbcBindingType::UnorderedAccessView:return bindingIndex + 0;
         case DxbcBindingType::UavCounter:         return bindingIndex + 8;
diff --git a/src/dxvk/dxvk_limits.h b/src/dxvk/dxvk_limits.h
index 0ea5143dc..7c77e2d03 100644
--- a/src/dxvk/dxvk_limits.h
+++ b/src/dxvk/dxvk_limits.h
@@ -10,7 +10,7 @@ namespace dxvk {
     MaxNumVertexBindings        =   32,
     MaxNumOutputStreams         =    4,
     MaxNumViewports             =   16,
-    MaxNumResourceSlots         = 1096,
+    MaxNumResourceSlots         = 1208,
     MaxNumActiveBindings        =  128,
     MaxNumQueuedCommandBuffers  =    8,
     MaxNumQueryCountPerPool     =  128,