On Thu, Jun 28, 2018 at 07:40:19AM +0100, Frediano Ziglio wrote:
> Make code faster on platforms not supporting unaligned access
> by default.
How does this work exactly? Rather than silencing the warning with
SPICE_UNALIGNED_CAST, and then potentially dereferencing a non-aligned
32 bit value, and getting a trap from the kernel/cpu, we instead give a
lot of hints to the compiler to let it know the value is not going to be
aligned, and we expect it's going to emit the right code?
Christophe
>
> Signed-off-by: Frediano Ziglio <fziglio@xxxxxxxxxx>
> ---
> common/canvas_base.c | 10 +++++++++-
> 1 file changed, 9 insertions(+), 1 deletion(-)
>
> diff --git a/common/canvas_base.c b/common/canvas_base.c
> index 6bf6e5d..36546b3 100644
> --- a/common/canvas_base.c
> +++ b/common/canvas_base.c
> @@ -45,6 +45,14 @@
> #include "mem.h"
> #include "macros.h"
>
> +#include <spice/start-packed.h>
> +typedef struct SPICE_ATTR_PACKED {
> + uint32_t v;
> +} uint32_unaligned_t;
> +#include <spice/end-packed.h>
> +
> +#define read_uint32_be(ptr) ntohl(((uint32_unaligned_t *)(ptr))->v)
> +
> #define ROUND(_x) ((int)floor((_x) + 0.5))
>
> static inline int fix_to_int(SPICE_FIXED28_4 fixed)
> @@ -572,7 +580,7 @@ static pixman_image_t *canvas_get_lz4(CanvasBase *canvas, SpiceImage *image)
>
> do {
> // Read next compressed block
> - enc_size = ntohl(*SPICE_UNALIGNED_CAST(uint32_t *, data));
> + enc_size = read_uint32_be(data);
> data += 4;
> dec_size = LZ4_decompress_safe_continue(stream, (const char *) data,
> (char *) dest, enc_size, available);
> --
> 2.17.1
>
> _______________________________________________
> Spice-devel mailing list
> Spice-devel@xxxxxxxxxxxxxxxxxxxxx
> https://lists.freedesktop.org/mailman/listinfo/spice-devel
Attachment:
signature.asc
Description: PGP signature
_______________________________________________ Spice-devel mailing list Spice-devel@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/spice-devel