[Intel-gfx] [PATCH] drm/i915: Use SSE4.1 movntqda to accelerate reads from WC memory

[Chris Wilson]

On Sat, Jul 16, 2016 at 01:07:35PM +0100, Chris Wilson wrote:
> +void i915_memcpy_from_wc(void *dst, const void *src, unsigned long len)
> +{
> +	GEM_BUG_ON((unsigned long)dst & 15);
> +	GEM_BUG_ON((unsigned long)src & 15);
> +	GEM_BUG_ON((unsigned long)len & 15);
> +
> +#ifdef CONFIG_AS_MOVNTDQA
> +	if (static_branch_likely(&has_movntqa))
> +		 __movntqda(dst, src, len);
> +	else
> +#endif
> +		memcpy(dst, src, len);

Ok, actually using this in a patch, the automatic fixup to a plain
memcpy from WC is bad behaviour and instead I want to punt the fallback
to the caller (i.e. return false;)
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre