[Intel-gfx] [PATCH i-g-t] kms_flip_tiling: Test flips between same tiling

[Chris Wilson]

On Thu, Apr 23, 2015 at 11:08:35AM +0100, Tvrtko Ursulin wrote:
> +/*
> + * Test that tiled page flips work.

Worth saying that we don't always expect linear->tiled flips to work, so
the mo is to first set the CRTC with a tiled fb, then flip to the
second.

> + */
> +static void
> +test_tiled_flip(data_t *data, igt_output_t *output, uint64_t tiling)
> +{
> +	drmModeModeInfo *mode;
> +	igt_plane_t *primary;
> +	igt_pipe_crc_t *pipe_crc;
> +	igt_crc_t reference_crc, crc;
> +	struct igt_fb fb[2];
> +	int fb_id[2], pipe, ret, width, height;
> +
> +	pipe = output->config.pipe;
> +	pipe_crc = igt_pipe_crc_new(pipe, INTEL_PIPE_CRC_SOURCE_AUTO);
> +	igt_output_set_pipe(output, pipe);
> +
> +	mode = igt_output_get_mode(output);
> +	primary = igt_output_get_plane(output, 0);
> +
> +	width = mode->hdisplay;
> +	height = mode->vdisplay;
> +
> +	/* create two identical fbs */
> +	fb_id[0] = igt_create_fb(data->drm_fd, width, height,
> +				 DRM_FORMAT_XRGB8888, tiling, &fb[0]);
> +	igt_assert(fb_id[0]);
> +
> +	fb_id[1] = igt_create_fb(data->drm_fd, width, height,
> +				 DRM_FORMAT_XRGB8888, tiling, &fb[1]);
> +	igt_assert(fb_id[1]);
> +
> +	fill_fb(&fb[0], data, mode);
> +	fill_fb(&fb[1], data, mode);

Pretty please can we have two different patterns?

> +	/* set the crtc and generate a reference crc */
> +	igt_plane_set_fb(primary, &fb[0]);
> +	igt_display_commit(&data->display);
> +	igt_pipe_crc_collect_crc(pipe_crc, &reference_crc);
> +
> +	/* flip to the other fb */
> +	ret = drmModePageFlip(data->drm_fd, output->config.crtc->crtc_id,
> +			      fb_id[1], 0, NULL);
> +	igt_assert_eq(ret, 0);
> +
> +	igt_wait_for_vblank(data->drm_fd, pipe);

I'd be more comfortable waiting for the flip-completion event (and then
a vblank for the CRC to be computed? idk)

> +	/* get a crc and compare with the reference */
> +	igt_pipe_crc_collect_crc(pipe_crc, &crc);
> +	igt_assert_crc_equal(&reference_crc, &crc);

With 2 patterns you can even test the second flip back again!

So for detecting the current breakage, we should just a
test_tiled_change that if PageFlip returns 0, the crc matches the
reference? And iterate over the modes to find one that causes an
underrun?
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre