[Mesa-dev] [PATCH 2/2] i965/vec4: Add ability to reswizzle arbitrary swizzles.

[Kenneth Graunke]

On Sunday, August 31, 2014 11:34:11 AM Matt Turner wrote:
> Before commit 04895f5c we would only reswizzle dot product instructions
> (since they wrote the same value into all channels, and we didn't have
> to think about anything else). That commit extended reswizzling to cases
> when the swizzle was single valued -- i.e., writing the same result into
> all channels.
> 
> But allowing reswizzling of arbitrary things is actually really easy and
> is even less code. (Why didn't we do this in the first place?!)
> 
> total instructions in shared programs: 4266079 -> 4261000 (-0.12%)
> instructions in affected programs:     351933 -> 346854 (-1.44%)
> ---
>  src/mesa/drivers/dri/i965/brw_vec4.cpp | 74 +++++++++-------------------------
>  1 file changed, 18 insertions(+), 56 deletions(-)
> 
> diff --git a/src/mesa/drivers/dri/i965/brw_vec4.cpp b/src/mesa/drivers/dri/i965/brw_vec4.cpp
> index cbea437..536a4b6 100644
> --- a/src/mesa/drivers/dri/i965/brw_vec4.cpp
> +++ b/src/mesa/drivers/dri/i965/brw_vec4.cpp
> @@ -943,30 +943,10 @@ vec4_instruction::can_reswizzle(int dst_writemask,
>     if (dst.writemask & ~swizzle_mask)
>        return false;
>  
> -   switch (opcode) {
> -   default:
> -      if (!brw_is_single_value_swizzle(swizzle)) {
> -         /* Check if there happens to be no reswizzling required. */
> -         for (int c = 0; c < 4; c++) {
> -            int bit = 1 << BRW_GET_SWZ(swizzle, c);
> -            /* Skip components of the swizzle not used by the dst. */
> -            if (!(dst_writemask & (1 << c)))
> -               continue;
> +   if (is_tex())

Presuambly the problem here is that you can't reswizzle the destination of messages...so it seems like you'd want is_tex() || is_send_from_grf()...

Or better yet:

   if (this->mlen > 0)
      return false;

With that change, this patch is:
Reviewed-by: Kenneth Graunke <kenneth at whitecape.org>

> +      return false;
>  
> -            /* We don't do the reswizzling yet, so just sanity check that we
> -             * don't have to.
> -             */
> -            if (bit != (1 << c))
> -               return false;
> -         }
> -         return true;
> -      }
> -      /* fallthrough */
> -   case BRW_OPCODE_DP4:
> -   case BRW_OPCODE_DP3:
> -   case BRW_OPCODE_DP2:
> -      return true;
> -   }
> +   return true;
>  }
>  
>  /**
> @@ -982,22 +962,9 @@ vec4_instruction::reswizzle(int dst_writemask, int swizzle)
>     int new_writemask = 0;
>     int new_swizzle[4] = { 0 };
>  
> -   switch (opcode) {
> -   default:
> -      if (!brw_is_single_value_swizzle(swizzle)) {
> -         for (int c = 0; c < 4; c++) {
> -            /* Skip components of the swizzle not used by the dst. */
> -            if (!(dst_writemask & (1 << c)))
> -               continue;
> -
> -            /* We don't do the reswizzling yet, so just sanity check that we
> -             * don't have to.
> -             */
> -            assert((1 << BRW_GET_SWZ(swizzle, c)) == (1 << c));
> -         }
> -         break;
> -      }
> -
> +   /* Dot product instructions write a single result into all channels. */
> +   if (opcode != BRW_OPCODE_DP4 && opcode != BRW_OPCODE_DPH &&
> +       opcode != BRW_OPCODE_DP3 && opcode != BRW_OPCODE_DP2) {
>        for (int i = 0; i < 3; i++) {
>           if (src[i].file == BAD_FILE || src[i].file == IMM)
>              continue;
> @@ -1009,25 +976,20 @@ vec4_instruction::reswizzle(int dst_writemask, int swizzle)
>           src[i].swizzle = BRW_SWIZZLE4(new_swizzle[0], new_swizzle[1],
>                                         new_swizzle[2], new_swizzle[3]);
>        }
> +   }
>  
> -      /* fallthrough */
> -   case BRW_OPCODE_DP4:
> -   case BRW_OPCODE_DP3:
> -   case BRW_OPCODE_DP2:
> -      for (int c = 0; c < 4; c++) {
> -         int bit = 1 << BRW_GET_SWZ(swizzle, c);
> -         /* Skip components of the swizzle not used by the dst. */
> -         if (!(dst_writemask & (1 << c)))
> -            continue;
> -         /* If we were populating this component, then populate the
> -          * corresponding channel of the new dst.
> -          */
> -         if (dst.writemask & bit)
> -            new_writemask |= (1 << c);
> -      }
> -      dst.writemask = new_writemask;
> -      break;
> +   for (int c = 0; c < 4; c++) {
> +      int bit = 1 << BRW_GET_SWZ(swizzle, c);
> +      /* Skip components of the swizzle not used by the dst. */
> +      if (!(dst_writemask & (1 << c)))
> +         continue;
> +      /* If we were populating this component, then populate the
> +       * corresponding channel of the new dst.
> +       */
> +      if (dst.writemask & bit)
> +         new_writemask |= (1 << c);
>     }
> +   dst.writemask = new_writemask;
>  }
>  
>  /*
> 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: This is a digitally signed message part.
URL: <http://lists.freedesktop.org/archives/mesa-dev/attachments/20140905/ef42e118/attachment-0001.sig>