[Mesa-dev] [PATCH 1/2] i965: Remove hand-rolled memcpy implementation.

[Ian Romanick]

On 02/22/2015 08:40 PM, Matt Turner wrote:
> ---
> I can't believe this is beneficial, especially since it came from the
> i915tex import.

My recollection is that memcpy does some weird optimizations (copying
from the end first) assuming the memory is mapped in a certain way.  We
don't necessarily map memory that way, and those optimizations hurt
quite a bit.  It's worth testing especially on Atom.

>  src/mesa/drivers/dri/i965/brw_context.h       | 27 ---------------------------
>  src/mesa/drivers/dri/i965/intel_batchbuffer.c |  2 +-
>  2 files changed, 1 insertion(+), 28 deletions(-)
> 
> diff --git a/src/mesa/drivers/dri/i965/brw_context.h b/src/mesa/drivers/dri/i965/brw_context.h
> index 197ed3e..7e86c1b 100644
> --- a/src/mesa/drivers/dri/i965/brw_context.h
> +++ b/src/mesa/drivers/dri/i965/brw_context.h
> @@ -1920,33 +1920,6 @@ gen6_upload_push_constants(struct brw_context *brw,
>                             struct brw_stage_state *stage_state,
>                             enum aub_state_struct_type type);
>  
> -/* ================================================================
> - * From linux kernel i386 header files, copes with odd sizes better
> - * than COPY_DWORDS would:
> - * XXX Put this in src/mesa/main/imports.h ???
> - */
> -#if defined(i386) || defined(__i386__)
> -static inline void * __memcpy(void * to, const void * from, size_t n)
> -{
> -   int d0, d1, d2;
> -   __asm__ __volatile__(
> -      "rep ; movsl\n\t"
> -      "testb $2,%b4\n\t"
> -      "je 1f\n\t"
> -      "movsw\n"
> -      "1:\ttestb $1,%b4\n\t"
> -      "je 2f\n\t"
> -      "movsb\n"
> -      "2:"
> -      : "=&c" (d0), "=&D" (d1), "=&S" (d2)
> -      :"0" (n/4), "q" (n),"1" ((long) to),"2" ((long) from)
> -      : "memory");
> -   return (to);
> -}
> -#else
> -#define __memcpy(a,b,c) memcpy(a,b,c)
> -#endif
> -
>  #ifdef __cplusplus
>  }
>  #endif
> diff --git a/src/mesa/drivers/dri/i965/intel_batchbuffer.c b/src/mesa/drivers/dri/i965/intel_batchbuffer.c
> index 7b96005..45c7493 100644
> --- a/src/mesa/drivers/dri/i965/intel_batchbuffer.c
> +++ b/src/mesa/drivers/dri/i965/intel_batchbuffer.c
> @@ -390,7 +390,7 @@ intel_batchbuffer_data(struct brw_context *brw,
>  {
>     assert((bytes & 3) == 0);
>     intel_batchbuffer_require_space(brw, bytes, ring);
> -   __memcpy(brw->batch.map + brw->batch.used, data, bytes);
> +   memcpy(brw->batch.map + brw->batch.used, data, bytes);
>     brw->batch.used += bytes >> 2;
>  }
>  
>