From 311661e4041e182e9bd6d8a6bd3d168901bb5c87 Mon Sep 17 00:00:00 2001
From: Philip Rebohle <philip.rebohle@tu-dortmund.de>
Date: Sat, 9 Feb 2019 22:21:57 +0100
Subject: [PATCH] [dxvk] Use vkCmdUpdateBuffer to clear tiny buffers

While this might consume a few more CPU cycles, UpdateBuffer may
be cheaper on the GPU for very small buffers, so we should use
it instead.

Also seems to fix rendering issues in Far Cry Primal for unknown reasons.
---
 src/dxvk/dxvk_context.cpp | 29 +++++++++++++++++++++--------
 1 file changed, 21 insertions(+), 8 deletions(-)

diff --git a/src/dxvk/dxvk_context.cpp b/src/dxvk/dxvk_context.cpp
index 8b10e889..2dd68fad 100644
--- a/src/dxvk/dxvk_context.cpp
+++ b/src/dxvk/dxvk_context.cpp
@@ -260,19 +260,32 @@ namespace dxvk {
           uint32_t              value) {
     this->spillRenderPass();
     
-    if (length == buffer->info().size)
-      length = align(length, 4);
-    
+    length = align(length, sizeof(uint32_t));
     auto slice = buffer->getSliceHandle(offset, length);
 
     if (m_barriers.isBufferDirty(slice, DxvkAccess::Write))
       m_barriers.recordCommands(m_cmd);
     
-    m_cmd->cmdFillBuffer(
-      slice.handle,
-      slice.offset,
-      slice.length,
-      value);
+    constexpr VkDeviceSize updateThreshold = 256;
+
+    if (length <= updateThreshold * sizeof(uint32_t)) {
+      std::array<uint32_t, updateThreshold> data;
+
+      for (uint32_t i = 0; i < length / sizeof(uint32_t); i++)
+        data[i] = value;
+      
+      m_cmd->cmdUpdateBuffer(
+        slice.handle,
+        slice.offset,
+        slice.length,
+        data.data());
+    } else {
+      m_cmd->cmdFillBuffer(
+        slice.handle,
+        slice.offset,
+        slice.length,
+        value);
+    }
     
     m_barriers.accessBuffer(slice,
       VK_PIPELINE_STAGE_TRANSFER_BIT,