[Intel-gfx] [PATCH] igt/gem_tiled_wc: Exercise wc mmaps with swizzling

[Chris Wilson]

On Tue, Nov 04, 2014 at 10:43:23AM +0100, Daniel Vetter wrote:
> > +			switch (swizzle) {
> > +			case I915_BIT_6_SWIZZLE_NONE:
> > +				swizzled_offset = j;
> > +				swizzle_str = "none";
> > +				break;
> > +			case I915_BIT_6_SWIZZLE_9:
> > +				swizzled_offset = j ^
> > +					swizzle_bit(9, j);
> > +				swizzle_str = "bit9";
> > +				break;
> > +			case I915_BIT_6_SWIZZLE_9_10:
> > +				swizzled_offset = j ^
> > +					swizzle_bit(9, j) ^
> > +					swizzle_bit(10, j);
> > +				swizzle_str = "bit9^10";
> > +				break;
> > +			case I915_BIT_6_SWIZZLE_9_11:
> > +				swizzled_offset = j ^
> > +					swizzle_bit(9, j) ^
> > +					swizzle_bit(11, j);
> > +				swizzle_str = "bit9^11";
> > +				break;
> > +			case I915_BIT_6_SWIZZLE_9_10_11:
> > +				swizzled_offset = j ^
> > +					swizzle_bit(9, j) ^
> > +					swizzle_bit(10, j) ^
> > +					swizzle_bit(11, j);
> > +				swizzle_str = "bit9^10^11";
> > +				break;
> 
> I think an explicit list of all the bit17 swizzle modes we know about with
> an igt_skip("bit17 swizzling") should be here. Otherwise the test will
> fail. Otherwise lgtm.

bit17 swizzling will trigger an igt_require() earlier on in get_tiling.
But yeah, we should skip an unknown swizzle rather than fail.

For Thomas,

/** @file gem_tiled_wc.c
 *
 * This is a test of write-combining mmap's behavior on tiled objects
 * with respect to the reported swizzling value.
 *
 * The goal is to exercise the complications that arise when using a linear
 * view of a tiled object that is subject to hardware swizzling. This is
 * useful to check that we are presenting the correct view of the object
 * to userspace, and that userspace has to respect the swizzle.
 */
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre