[d3d11] Move immediate constant buffers to non-mapped VRAM

This has the advantage that icbs can be defragmented and don't
take up space in HVV chunks that would potentially go unused.

src/d3d11/d3d11_device.h

@@ -429,6 +429,13 @@ namespace dxvk {
       m_initializer->NotifyContextFlush();
     }
     
+    void InitShaderIcb(
+            D3D11CommonShader*          pShader,
+            size_t                      IcbSize,
+      const void*                       pIcbData) {
+      return m_initializer->InitShaderIcb(pShader, IcbSize, pIcbData);
+    }
+
     VkPipelineStageFlags GetEnabledShaderStages() const {
       return m_dxvkDevice->getShaderPipelineStages();
     }

src/d3d11/d3d11_initializer.cpp

@@ -77,6 +77,33 @@ namespace dxvk {
   }
 
 
+  void D3D11Initializer::InitShaderIcb(
+          D3D11CommonShader*          pShader,
+          size_t                      IcbSize,
+    const void*                       pIcbData) {
+    std::lock_guard<dxvk::mutex> lock(m_mutex);
+    m_transferCommands += 1;
+
+    auto icbSlice = pShader->GetIcb();
+    auto srcSlice = m_stagingBuffer.alloc(icbSlice.length());
+
+    std::memcpy(srcSlice.mapPtr(0), pIcbData, IcbSize);
+
+    if (IcbSize < icbSlice.length())
+      std::memset(srcSlice.mapPtr(IcbSize), 0, icbSlice.length() - IcbSize);
+
+    EmitCs([
+      cIcbSlice = std::move(icbSlice),
+      cSrcSlice = std::move(srcSlice)
+    ] (DxvkContext* ctx) {
+      ctx->copyBuffer(cIcbSlice.buffer(), cIcbSlice.offset(),
+        cSrcSlice.buffer(), cSrcSlice.offset(), cIcbSlice.length());
+    });
+
+    ThrottleAllocationLocked();
+  }
+
+
   void D3D11Initializer::InitDeviceLocalBuffer(
           D3D11Buffer*                pBuffer,
     const D3D11_SUBRESOURCE_DATA*     pInitialData) {

src/d3d11/d3d11_initializer.h

@@ -3,7 +3,9 @@
 #include "../dxvk/dxvk_staging.h"
 
 #include "d3d11_buffer.h"
+#include "d3d11_shader.h"
 #include "d3d11_texture.h"
+#include "d3d11_view_uav.h"
 
 namespace dxvk {
 
@@ -57,6 +59,11 @@ namespace dxvk {
     void InitUavCounter(
             D3D11UnorderedAccessView*   pUav);
     
+    void InitShaderIcb(
+            D3D11CommonShader*          pShader,
+            size_t                      IcbSize,
+      const void*                       pIcbData);
+
   private:
 
     dxvk::mutex       m_mutex;

src/d3d11/d3d11_shader.cpp

@@ -64,19 +64,17 @@ namespace dxvk {
     if (icb.size) {
       DxvkBufferCreateInfo info = { };
       info.size   = align(icb.size, 256u);
-      info.usage  = VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT;
+      info.usage  = VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT
+                  | VK_BUFFER_USAGE_TRANSFER_DST_BIT;
       info.stages = util::pipelineStages(m_shader->info().stage);
-      info.access = VK_ACCESS_UNIFORM_READ_BIT;
+      info.access = VK_ACCESS_UNIFORM_READ_BIT
+                  | VK_ACCESS_TRANSFER_WRITE_BIT;
       info.debugName = "Icb";
       
-      VkMemoryPropertyFlags memFlags
+      m_buffer = pDevice->GetDXVKDevice()->createBuffer(info, VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT);
-        = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT
-        | VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
-        | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT;
 
-      m_buffer = pDevice->GetDXVKDevice()->createBuffer(info, memFlags);
+      // Upload immediate constant buffer to VRAM
-      std::memcpy(m_buffer->mapPtr(0), icb.data, icb.size);
+      pDevice->InitShaderIcb(this, icb.size, icb.data);
-      std::memset(m_buffer->mapPtr(icb.size), 0, info.size - icb.size);
     }
 
     pDevice->GetDXVKDevice()->registerShader(m_shader);