[Pixman] [PATCH] Add forgotten _mm_empty() calls in the SSE2 fetchers.

Siarhei Siamashka siarhei.siamashka at gmail.com
Thu Feb 17 14:10:34 PST 2011


On Wednesday 16 February 2011 14:13:20 Søren Sandmann wrote:
> From: Søren Sandmann Pedersen <ssp at redhat.com>
> 
> ---
>  pixman/pixman-sse2.c |    6 ++++++
>  1 files changed, 6 insertions(+), 0 deletions(-)
> 
> diff --git a/pixman/pixman-sse2.c b/pixman/pixman-sse2.c
> index 2e135e2..29d2ea8 100644
> --- a/pixman/pixman-sse2.c
> +++ b/pixman/pixman-sse2.c
> @@ -6200,6 +6200,8 @@ sse2_fetch_x8r8g8b8 (pixman_iter_t *iter, const
> uint32_t *mask) w--;
>      }
> 
> +    _mm_empty();
> +
>      return iter->buffer;
>  }
> 
> @@ -6246,6 +6248,8 @@ sse2_fetch_r5g6b5 (pixman_iter_t *iter, const
> uint32_t *mask) w--;
>      }
> 
> +    _mm_empty();
> +
>      return iter->buffer;
>  }
> 
> @@ -6292,6 +6296,8 @@ sse2_fetch_a8 (pixman_iter_t *iter, const uint32_t
> *mask) w--;
>      }
> 
> +    _mm_empty();
> +
>      return iter->buffer;
>  }

This looks ok.

I just wonder whether you have investigated why these missing _mm_empty() calls
are not actually causing pixman tests failure? Is it because fetch operation is
typically followed by combiner which has _mm_empty()? But SRC combiner should
not be using MMX/SSE2. Or no MMX instructions are actually used by these
fetchers (only SSE2)? Or the tests are lacking something?

-- 
Best regards,
Siarhei Siamashka


More information about the Pixman mailing list