From a4a5bf5d63054bda6d536edc1f08ead761c9a357 Mon Sep 17 00:00:00 2001
From: Philip Rebohle <philip.rebohle@tu-dortmund.de>
Date: Sat, 1 Mar 2025 14:14:38 +0100
Subject: [PATCH] [dxvk] Add helpers for transfer queue release barriers

---
 src/dxvk/dxvk_context.cpp | 86 +++++++++++++++++++++++++++++++++++++++
 src/dxvk/dxvk_context.h   | 12 ++++++
 2 files changed, 98 insertions(+)

diff --git a/src/dxvk/dxvk_context.cpp b/src/dxvk/dxvk_context.cpp
index ad0fbadc8..d14edddc9 100644
--- a/src/dxvk/dxvk_context.cpp
+++ b/src/dxvk/dxvk_context.cpp
@@ -8087,6 +8087,56 @@ namespace dxvk {
   }
 
 
+  void DxvkContext::accessImageTransfer(
+          DxvkImage&                image,
+    const VkImageSubresourceRange&  subresources,
+          VkImageLayout             srcLayout,
+          VkPipelineStageFlags2     srcStages,
+          VkAccessFlags2            srcAccess) {
+    auto& transferBatch = getBarrierBatch(DxvkCmdBuffer::SdmaBuffer);
+    auto& graphicsBatch = getBarrierBatch(DxvkCmdBuffer::InitBarriers);
+
+    if (srcLayout == VK_IMAGE_LAYOUT_UNDEFINED || srcLayout == VK_IMAGE_LAYOUT_PREINITIALIZED)
+      image.trackInitialization(subresources);
+
+    if (m_device->hasDedicatedTransferQueue()) {
+      VkImageMemoryBarrier2 barrier = { VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2 };
+      barrier.srcStageMask = srcStages;
+      barrier.srcAccessMask = srcAccess;
+      barrier.dstStageMask = srcStages;
+      barrier.dstAccessMask = VK_ACCESS_2_NONE;
+      barrier.oldLayout = srcLayout;
+      barrier.newLayout = image.info().layout;
+      barrier.srcQueueFamilyIndex = m_device->queues().transfer.queueFamily;
+      barrier.dstQueueFamilyIndex = m_device->queues().graphics.queueFamily;
+      barrier.image = image.handle();
+      barrier.subresourceRange = subresources;
+
+      transferBatch.addImageBarrier(barrier);
+
+      barrier.srcAccessMask = VK_ACCESS_2_NONE;
+      barrier.dstStageMask = image.info().stages;
+      barrier.dstAccessMask = image.info().access;
+
+      graphicsBatch.addImageBarrier(barrier);
+    } else {
+      VkImageMemoryBarrier2 barrier = { VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2 };
+      barrier.srcStageMask = srcStages;
+      barrier.srcAccessMask = srcAccess;
+      barrier.dstStageMask = image.info().stages;
+      barrier.dstAccessMask = image.info().access;
+      barrier.oldLayout = srcLayout;
+      barrier.newLayout = image.info().layout;
+      barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
+      barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
+      barrier.image = image.handle();
+      barrier.subresourceRange = subresources;
+
+      transferBatch.addImageBarrier(barrier);
+    }
+  }
+
+
   void DxvkContext::accessBuffer(
           DxvkCmdBuffer             cmdBuffer,
           DxvkBuffer&               buffer,
@@ -8206,6 +8256,42 @@ namespace dxvk {
   }
 
 
+  void DxvkContext::accessBufferTransfer(
+          DxvkBuffer&               buffer,
+          VkPipelineStageFlags2     srcStages,
+          VkAccessFlags2            srcAccess) {
+    auto& transferBatch = getBarrierBatch(DxvkCmdBuffer::SdmaBuffer);
+    auto& graphicsBatch = getBarrierBatch(DxvkCmdBuffer::InitBarriers);
+
+    if (m_device->hasDedicatedTransferQueue()) {
+      // No queue ownership transfer necessary since buffers all
+      // use SHARING_MODE_CONCURRENT, but we need a split barrier.
+      VkMemoryBarrier2 barrier = { VK_STRUCTURE_TYPE_MEMORY_BARRIER_2 };
+      barrier.srcStageMask = srcStages;
+      barrier.srcAccessMask = srcAccess;
+      barrier.dstStageMask = srcStages;
+      barrier.dstAccessMask = VK_ACCESS_2_NONE;
+
+      transferBatch.addMemoryBarrier(barrier);
+
+      barrier.srcStageMask = srcStages;
+      barrier.srcAccessMask = VK_ACCESS_2_NONE;
+      barrier.dstStageMask = buffer.info().stages;
+      barrier.dstAccessMask = buffer.info().access;
+
+      graphicsBatch.addMemoryBarrier(barrier);
+    } else {
+      VkMemoryBarrier2 barrier = { VK_STRUCTURE_TYPE_MEMORY_BARRIER_2 };
+      barrier.srcStageMask = srcStages;
+      barrier.srcAccessMask = srcAccess;
+      barrier.dstStageMask = buffer.info().stages;
+      barrier.dstAccessMask = buffer.info().access;
+
+      transferBatch.addMemoryBarrier(barrier);
+    }
+  }
+
+
   void DxvkContext::accessDrawBuffer(
           VkDeviceSize              offset,
           uint32_t                  count,
diff --git a/src/dxvk/dxvk_context.h b/src/dxvk/dxvk_context.h
index 97805a96b..81cb14bf7 100644
--- a/src/dxvk/dxvk_context.h
+++ b/src/dxvk/dxvk_context.h
@@ -1973,6 +1973,13 @@ namespace dxvk {
             VkAccessFlags2            dstAccess,
             DxvkAccessOp              accessOp);
 
+    void accessImageTransfer(
+            DxvkImage&                image,
+      const VkImageSubresourceRange&  subresources,
+            VkImageLayout             srcLayout,
+            VkPipelineStageFlags2     srcStages,
+            VkAccessFlags2            srcAccess);
+
     void accessBuffer(
             DxvkCmdBuffer             cmdBuffer,
             DxvkBuffer&               buffer,
@@ -2025,6 +2032,11 @@ namespace dxvk {
             VkAccessFlags2            dstAccess,
             DxvkAccessOp              accessOp);
 
+    void accessBufferTransfer(
+            DxvkBuffer&               buffer,
+            VkPipelineStageFlags2     srcStages,
+            VkAccessFlags2            srcAccess);
+
     void accessDrawBuffer(
             VkDeviceSize              offset,
             uint32_t                  count,