[PATCH] wined3d: Force VBOs to be GPU accessible.

[Andrew Wesie]

Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=45279
Signed-off-by: Andrew Wesie 
---
 dlls/wined3d/buffer.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/dlls/wined3d/buffer.c b/dlls/wined3d/buffer.c
index e7b05157cc..247db45bb7 100644
--- a/dlls/wined3d/buffer.c
+++ b/dlls/wined3d/buffer.c
@@ -1399,6 +1399,10 @@ static HRESULT buffer_init(struct wined3d_buffer *buffer, struct wined3d_device
      * (Half-Life 2 and others.) */
     dynamic_buffer_ok = gl_info->supported[APPLE_FLUSH_BUFFER_RANGE] || gl_info->supported[ARB_MAP_BUFFER_RANGE];

+    /* We need vertex buffer objects in OpenGL core profiles. Force GPU access. */
+    if (bind_flags & WINED3D_BIND_VERTEX_BUFFER)
+        access |= WINED3D_RESOURCE_ACCESS_GPU;
+
     if (!gl_info->supported[ARB_VERTEX_BUFFER_OBJECT])
     {
         TRACE("Not creating a BO because GL_ARB_vertex_buffer is not supported.\n");

[Henri Verbeet]

On Mon, 15 Oct 2018 at 08:06, Andrew Wesie  wrote:
> - /* We need vertex buffer objects in OpenGL core profiles. Force GPU access. */
> - if (bind_flags & WINED3D_BIND_VERTEX_BUFFER)
> -
>    access |= WINED3D_RESOURCE_ACCESS_GPU;

I don't think that's correct. If anything, I think we should be
failing resource creation for resources with BIND_VERTEX_BUFFER
without RESOURCE_ACCESS_GPU. We have d3d11 tests to prove that, and
the only reason we currently don't reject these in wined3d is because
we already reject them in d3d11 instead.
> From the logs I've seen for this, one (45486) was using d3d9 with

software vertex processing, in which case we probably should either
just set RESOURCE_ACCESS_GPU on resources in the system memory pool,
or actually implement software vertex processing. The other (45279)
was using ddraw. I think it was using a TnLHalDevice, but in ddraw
often anything goes, except when it doesn't. The log from bug 45987
didn't contain the CreateDevice() flags.