[igt-dev] [PATCH i-g-t 1/8] lib/igt_draw: Refactor get_tiling calls

[Imre Deak]

On Tue, Feb 11, 2020 at 11:10:13AM -0800, Dixit, Ashutosh wrote:
> On Mon, 10 Feb 2020 03:37:54 -0800, Imre Deak wrote:
> >
> > On Sat, Feb 08, 2020 at 02:36:38PM -0800, Dixit, Ashutosh wrote:
> > > On Fri, 07 Feb 2020 11:15:17 -0800, Imre Deak wrote:
> > > >  void igt_draw_rect(int fd, drm_intel_bufmgr *bufmgr, drm_intel_context *context,
> > > >		   uint32_t buf_handle, uint32_t buf_size, uint32_t buf_stride,
> > > > -		   enum igt_draw_method method, int rect_x, int rect_y,
> > > > -		   int rect_w, int rect_h, uint32_t color, int bpp)
> > > > +		   uint32_t tiling, enum igt_draw_method method,
> > > > +		   int rect_x, int rect_y, int rect_w, int rect_h,
> > > > +		   uint32_t color, int bpp)
> > > >  {
> > > > +	uint32_t buf_tiling, swizzle;
> > > > +
> > > >	struct cmd_data cmd_data = {
> > > >		.bufmgr = bufmgr,
> > > >		.context = context,
> > > > @@ -667,24 +660,32 @@ void igt_draw_rect(int fd, drm_intel_bufmgr *bufmgr, drm_intel_context *context,
> > > >		.h = rect_h,
> > > >	};
> > > >
> > > > +	swizzle = I915_BIT_6_SWIZZLE_NONE;
> > > > +	if (tiling != I915_TILING_NONE && gem_available_fences(fd)) {
> > > > +		gem_get_tiling(fd, buf_handle, &buf_tiling, &swizzle);
> > > > +		igt_assert(tiling == buf_tiling);
> > > > +	}
> > >
> > > Probably a nit, but looks a little strange to call gem_get_tiling() to get
> > > the swizzle and then doing an assert.
> >
> > gem_get_tiling() is the way to get the swizzling for a buffer. The
> > assert only makes sure that the caller's and kernel's idea of the tiling
> > matches. While the callers will pick the tiling (when creating the
> > buffer), the swizzling is fixed based on this tiling and platform.
> 
> Let me ask a different way, do we even need the new tiling argument to
> igt_draw_rect()? Can't we just do:
> 
>         if (gem_available_fences(fd))
>                 gem_get_tiling(fd, buf_handle, &buf_tiling, &swizzle);
> 
> Basically if we have the handle we should just be able to query the tiling.

We do need to handle different tilings even if
!gem_available_fences(fd), so we need the tiling param. Note that HW
detiling support via GGTT != tiling support by display,gt engines.

> 
> > > Instead, since igt_draw_rect() has only two callers how about moving
> > > the gem_get_tiling() call into the callers and passing both tiling and
> > > swizzle into igt_draw_rect()?
> >
> > Why?
> >
> > >
> > > > diff --git a/tests/kms_frontbuffer_tracking.c b/tests/kms_frontbuffer_tracking.c
> > > > index 2c765c34..9c2c3a5d 100644
> > > > --- a/tests/kms_frontbuffer_tracking.c
> > > > +++ b/tests/kms_frontbuffer_tracking.c
> > > > @@ -291,6 +291,7 @@ struct {
> > > >	int height;
> > > >	uint32_t color;
> > > >	int bpp;
> > > > +	uint32_t tiling;
> > >
> > > Should not need this if we follow the suggestion above.
> >
> > This contains the tiling of the buffer as now we won't have a way to
> > retrieve the same from the kernel.
> 
> Again, won't need these other changes if the tiling arg to igt_draw_rect()
> can be eliminated.