diff --git a/libs/vkd3d/command.c b/libs/vkd3d/command.c
index f27bf314..774ea97d 100644
--- a/libs/vkd3d/command.c
+++ b/libs/vkd3d/command.c
@@ -2265,6 +2265,9 @@ static VkImageLayout dsv_plane_optimal_mask_to_layout(uint32_t plane_optimal_mas
         VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL,
     };
 
+    if (plane_optimal_mask & VKD3D_DEPTH_STENCIL_PLANE_GENERAL)
+        return VK_IMAGE_LAYOUT_GENERAL;
+
     if (image_aspects != (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT))
     {
         /* If aspects is only DEPTH or only STENCIL, we should use the OPTIMAL or READ_ONLY layout.
@@ -2327,8 +2330,9 @@ static uint32_t d3d12_command_list_notify_decay_dsv_resource(struct d3d12_comman
     uint32_t decay_aspects;
     size_t i, n;
 
-    /* No point in adding these since they are always deduced to be optimal. */
-    if (resource->desc.Flags & D3D12_RESOURCE_FLAG_DENY_SHADER_RESOURCE)
+    /* No point in adding these since they are always deduced to be optimal or general. */
+    if ((resource->desc.Flags & D3D12_RESOURCE_FLAG_DENY_SHADER_RESOURCE) ||
+            resource->common_layout == VK_IMAGE_LAYOUT_GENERAL)
         return 0;
 
     for (i = 0, n = list->dsv_resource_tracking_count; i < n; i++)
@@ -2353,6 +2357,8 @@ static uint32_t d3d12_command_list_promote_dsv_resource(struct d3d12_command_lis
     /* No point in adding these since they are always deduced to be optimal. */
     if (resource->desc.Flags & D3D12_RESOURCE_FLAG_DENY_SHADER_RESOURCE)
         return VKD3D_DEPTH_PLANE_OPTIMAL | VKD3D_STENCIL_PLANE_OPTIMAL;
+    else if (resource->common_layout == VK_IMAGE_LAYOUT_GENERAL)
+        return VKD3D_DEPTH_STENCIL_PLANE_GENERAL;
 
     /* For single aspect images, mirror the optimal mask in the unused aspect. This avoids some
      * extra checks elsewhere (particularly graphics pipeline setup and compat render passes)
@@ -2382,6 +2388,9 @@ static uint32_t d3d12_command_list_promote_dsv_resource(struct d3d12_command_lis
 static uint32_t d3d12_command_list_notify_dsv_writes(struct d3d12_command_list *list,
         struct d3d12_resource *resource, const struct vkd3d_view *view, uint32_t plane_write_mask)
 {
+    if (plane_write_mask & VKD3D_DEPTH_STENCIL_PLANE_GENERAL)
+        return VKD3D_DEPTH_STENCIL_PLANE_GENERAL;
+
     assert(!(plane_write_mask & ~(VKD3D_DEPTH_PLANE_OPTIMAL | VKD3D_STENCIL_PLANE_OPTIMAL)));
 
     /* If we cover the entire resource, we can promote it to our target layout. */
@@ -2502,6 +2511,12 @@ static VkImageLayout d3d12_command_list_get_depth_stencil_resource_layout(const
             *plane_optimal_mask = VKD3D_DEPTH_PLANE_OPTIMAL | VKD3D_STENCIL_PLANE_OPTIMAL;
         return VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;
     }
+    else if (resource->common_layout == VK_IMAGE_LAYOUT_GENERAL)
+    {
+        if (plane_optimal_mask)
+            *plane_optimal_mask = VKD3D_DEPTH_STENCIL_PLANE_GENERAL;
+        return VK_IMAGE_LAYOUT_GENERAL;
+    }
 
     for (i = 0, n = list->dsv_resource_tracking_count; i < n; i++)
     {
@@ -2521,16 +2536,30 @@ static VkImageLayout d3d12_command_list_get_depth_stencil_resource_layout(const
 
 static VkImageLayout vk_separate_depth_layout(VkImageLayout combined_layout)
 {
-    return (combined_layout == VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL ||
-            combined_layout == VK_IMAGE_LAYOUT_DEPTH_ATTACHMENT_STENCIL_READ_ONLY_OPTIMAL) ?
-            VK_IMAGE_LAYOUT_DEPTH_ATTACHMENT_OPTIMAL : VK_IMAGE_LAYOUT_DEPTH_READ_ONLY_OPTIMAL;
+    if (combined_layout == VK_IMAGE_LAYOUT_GENERAL)
+    {
+        return combined_layout;
+    }
+    else
+    {
+        return (combined_layout == VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL ||
+                combined_layout == VK_IMAGE_LAYOUT_DEPTH_ATTACHMENT_STENCIL_READ_ONLY_OPTIMAL) ?
+                VK_IMAGE_LAYOUT_DEPTH_ATTACHMENT_OPTIMAL : VK_IMAGE_LAYOUT_DEPTH_READ_ONLY_OPTIMAL;
+    }
 }
 
 static VkImageLayout vk_separate_stencil_layout(VkImageLayout combined_layout)
 {
-    return (combined_layout == VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL ||
-            combined_layout == VK_IMAGE_LAYOUT_DEPTH_READ_ONLY_STENCIL_ATTACHMENT_OPTIMAL) ?
-            VK_IMAGE_LAYOUT_STENCIL_ATTACHMENT_OPTIMAL : VK_IMAGE_LAYOUT_STENCIL_READ_ONLY_OPTIMAL;
+    if (combined_layout == VK_IMAGE_LAYOUT_GENERAL)
+    {
+        return combined_layout;
+    }
+    else
+    {
+        return (combined_layout == VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL ||
+                combined_layout == VK_IMAGE_LAYOUT_DEPTH_READ_ONLY_STENCIL_ATTACHMENT_OPTIMAL) ?
+                VK_IMAGE_LAYOUT_STENCIL_ATTACHMENT_OPTIMAL : VK_IMAGE_LAYOUT_STENCIL_READ_ONLY_OPTIMAL;
+    }
 }
 
 static void d3d12_command_list_clear_attachment_pass(struct d3d12_command_list *list, struct d3d12_resource *resource,
diff --git a/libs/vkd3d/resource.c b/libs/vkd3d/resource.c
index 72fbb39c..249e2702 100644
--- a/libs/vkd3d/resource.c
+++ b/libs/vkd3d/resource.c
@@ -299,7 +299,31 @@ static bool vkd3d_is_linear_tiling_supported(const struct d3d12_device *device,
     return supported;
 }
 
-static VkImageLayout vk_common_image_layout_from_d3d12_desc(const D3D12_RESOURCE_DESC1 *desc)
+static bool d3d12_device_prefers_general_depth_stencil(const struct d3d12_device *device)
+{
+    if (device->vk_info.KHR_driver_properties)
+    {
+        if (device->device_info.driver_properties.driverID == VK_DRIVER_ID_NVIDIA_PROPRIETARY)
+        {
+            /* NVIDIA doesn't really care about layouts for the most part. */
+            return true;
+        }
+        else if (device->device_info.driver_properties.driverID == VK_DRIVER_ID_MESA_RADV)
+        {
+            /* RADV can use TC-compat HTILE without too much issues on Polaris and later.
+             * Use GENERAL for these GPUs.
+             * Pre-Polaris we run into issues where even read-only depth requires decompress
+             * so using GENERAL shouldn't really make things worse, it's going to run pretty bad
+             * either way. */
+            return true;
+        }
+    }
+
+    return false;
+}
+
+static VkImageLayout vk_common_image_layout_from_d3d12_desc(const struct d3d12_device *device,
+        const D3D12_RESOURCE_DESC1 *desc)
 {
     /* We need aggressive decay and promotion into anything. */
     if (desc->Flags & D3D12_RESOURCE_FLAG_ALLOW_SIMULTANEOUS_ACCESS)
@@ -307,6 +331,20 @@ static VkImageLayout vk_common_image_layout_from_d3d12_desc(const D3D12_RESOURCE
     if (desc->Layout == D3D12_TEXTURE_LAYOUT_ROW_MAJOR)
         return VK_IMAGE_LAYOUT_GENERAL;
 
+    /* This is counter-intuitive, but using GENERAL layout for depth-stencils works around
+     * having to perform DSV plane tracking all the time, since we don't necessarily know at recording time
+     * if a DSV image is OPTIMAL or READ_ONLY.
+     * This saves us many redundant barriers while rendering, especially since games tend
+     * to split their rendering across many command lists in parallel.
+     * On several implementations, GENERAL is a perfectly fine layout to use,
+     * on others it is a disaster since compression is disabled :') */
+    if (((desc->Flags & (D3D12_RESOURCE_FLAG_ALLOW_DEPTH_STENCIL | D3D12_RESOURCE_FLAG_DENY_SHADER_RESOURCE)) ==
+            D3D12_RESOURCE_FLAG_ALLOW_DEPTH_STENCIL) &&
+            d3d12_device_prefers_general_depth_stencil(device))
+    {
+        return VK_IMAGE_LAYOUT_GENERAL;
+    }
+
     /* DENY_SHADER_RESOURCE only allowed with ALLOW_DEPTH_STENCIL */
     if (desc->Flags & D3D12_RESOURCE_FLAG_DENY_SHADER_RESOURCE)
         return VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;
@@ -708,7 +746,7 @@ static HRESULT vkd3d_create_image(struct d3d12_device *device,
             resource->common_layout = VK_IMAGE_LAYOUT_GENERAL;
         }
         else
-            resource->common_layout = vk_common_image_layout_from_d3d12_desc(desc);
+            resource->common_layout = vk_common_image_layout_from_d3d12_desc(device, desc);
 
         if (desc->Flags & D3D12_RESOURCE_FLAG_ALLOW_SIMULTANEOUS_ACCESS)
             resource->flags |= VKD3D_RESOURCE_SIMULTANEOUS_ACCESS;
@@ -2952,7 +2990,7 @@ VKD3D_EXPORT HRESULT vkd3d_create_image_resource(ID3D12Device *device,
     object->flags = create_info->flags;
     object->flags |= VKD3D_RESOURCE_EXTERNAL;
     object->initial_layout_transition = 1;
-    object->common_layout = vk_common_image_layout_from_d3d12_desc(&object->desc);
+    object->common_layout = vk_common_image_layout_from_d3d12_desc(d3d12_device, &object->desc);
 
     memset(&object->sparse, 0, sizeof(object->sparse));
 
diff --git a/libs/vkd3d/vkd3d_private.h b/libs/vkd3d/vkd3d_private.h
index bc9a2c1a..412c3611 100644
--- a/libs/vkd3d/vkd3d_private.h
+++ b/libs/vkd3d/vkd3d_private.h
@@ -1445,11 +1445,11 @@ struct vkd3d_shader_debug_ring_spec_info
     VkSpecializationInfo spec_info;
 };
 
-/* One render pass for each plane optimal mask. */
 enum vkd3d_plane_optimal_flag
 {
     VKD3D_DEPTH_PLANE_OPTIMAL = (1 << 0),
     VKD3D_STENCIL_PLANE_OPTIMAL = (1 << 1),
+    VKD3D_DEPTH_STENCIL_PLANE_GENERAL = (1 << 2),
 };
 
 struct d3d12_graphics_pipeline_state