[Intel-gfx] [RFC 2/6] drm/i915: Add tiled framebuffer modifiers

[Daniel Vetter]

On Mon, Feb 02, 2015 at 04:42:32PM +0000, Tvrtko Ursulin wrote:
> 
> On 02/02/2015 04:32 PM, Rob Clark wrote:
> >On Mon, Feb 2, 2015 at 4:41 AM, Daniel Vetter <daniel at ffwll.ch> wrote:
> >>On Fri, Jan 30, 2015 at 05:36:54PM +0000, Tvrtko Ursulin wrote:
> >>>From: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
> >>>
> >>>To be used from the new addfb2 extension.
> >>>
> >>>Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
> >>>---
> >>>  include/uapi/drm/i915_drm.h | 13 +++++++++++++
> >>>  1 file changed, 13 insertions(+)
> >>>
> >>>diff --git a/include/uapi/drm/i915_drm.h b/include/uapi/drm/i915_drm.h
> >>>index 6eed16b..a7327fd 100644
> >>>--- a/include/uapi/drm/i915_drm.h
> >>>+++ b/include/uapi/drm/i915_drm.h
> >>>@@ -28,6 +28,7 @@
> >>>  #define _UAPI_I915_DRM_H_
> >>>
> >>>  #include <drm/drm.h>
> >>>+#include <uapi/drm/drm_fourcc.h>
> >>>
> >>>  /* Please note that modifications to all structs defined here are
> >>>   * subject to backwards-compatibility constraints.
> >>>@@ -1101,4 +1102,16 @@ struct drm_i915_gem_context_param {
> >>>       __u64 value;
> >>>  };
> >>>
> >>>+/** @{
> >>>+ * Intel framebuffer modifiers
> >>>+ *
> >>>+ * Tiling modes supported by the display hardware
> >>>+ * to be passed in via the DRM addfb2 ioctl.
> >>>+ */
> >>>+/** None */
> >>>+#define I915_FORMAT_MOD_NONE fourcc_mod_code(INTEL, 0x00000000000000L)
> >>>+/** X tiling */
> >>>+#define I915_FORMAT_MOD_X_TILED      fourcc_mod_code(INTEL, 0x00000000000001L)
> >>
> >>One thing I wonder here is whether we should have a modifier for each
> >>physical layout (tiling modes do change slightly between hw) or whether we
> >>should just continue to assume that this is Intel-specific and add a
> >>disclaimer that the precise layout depends upon the actual intel box
> >>you're running on?
> >
> >I'd kind of lean towards different modifiers per physical layout..
> >that seems more useful for cases where nvidia/amd support some of the
> >formats for buffer sharing..
> 
> Hm.. we've got physical layout, alignment restrictions, geometry
> restrictions, what are the odds this will be shareable or compatible, and
> how will the token names even looks when one puts all of this into them?

On top of that there's a _lot_ of different physical layouts for just X
tiling. At least if you look at more than just modern platforms. And often
userspace doesn't even know which precise variant it is.

I think if we eventually have a match with some other vendor format (the
one with nvidia wasn't intentionally, it only works if you have swizzling
enabled, not without swizzling) then we could do some aliasing: Define a
new vendor neutral code which then all drivers supporting it would remap
to the correct internal/vendor-specific representation.

Of course integrated gpus are special, with plug-in pci devices you really
have to spec the full thing.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch