[Mesa-dev] [PATCH 27/42] i965/blorp: wrap LRP

[Pohjolainen, Topi]

On Mon, Jan 20, 2014 at 07:47:45PM -0800, Paul Berry wrote:
>    On 20 December 2013 06:38, Topi Pohjolainen <topi.pohjolainen at intel.com>
>    wrote:
> 
>      The split of the emission of the two halfs into single emission
>      call prapares for fs_generator support that already does similar
>      thing. No regressions seen on IVB (unit tests and piglit quick).
> 
>      Signed-off-by: Topi Pohjolainen <topi.pohjolainen at intel.com>
>      ---
>       src/mesa/drivers/dri/i965/brw_blorp_blit.cpp  | 26
>      ++++++++++----------------
>       src/mesa/drivers/dri/i965/brw_blorp_blit_eu.h | 13 +++++++++++++
>       2 files changed, 23 insertions(+), 16 deletions(-)
> 
>      diff --git a/src/mesa/drivers/dri/i965/brw_blorp_blit.cpp
>      b/src/mesa/drivers/dri/i965/brw_blorp_blit.cpp
>      index 1b7310b..b95104e 100644
>      --- a/src/mesa/drivers/dri/i965/brw_blorp_blit.cpp
>      +++ b/src/mesa/drivers/dri/i965/brw_blorp_blit.cpp
>      @@ -1679,29 +1679,23 @@
>      brw_blorp_blit_program::manual_blend_bilinear(unsigned num_samples)
>          }
> 
>       #define SAMPLE(x, y) offset(texture_data[x], y)
>      -   brw_set_access_mode(&func, BRW_ALIGN_16);
>      -   brw_set_compression_control(&func, BRW_COMPRESSION_NONE);
>          for (int index = 3; index > 0; ) {
>             /* Since we're doing SIMD16, 4 color channels fits in to 8
>      registers.
>              * Counter value of 8 in 'for' loop below is used to interpolate
>      all
>              * the color components.
>              */
>      -      for (int k = 0; k < 8; ++k)
>      -         brw_LRP(&func,
>      -                 vec8(SAMPLE(index - 1, k)),
>      -                 offset(x_frac, k & 1),
>      -                 SAMPLE(index, k),
>      -                 SAMPLE(index - 1, k));
>      +      for (int k = 0; k < 8; k += 2)
>      +         emit_lrp(vec8(SAMPLE(index - 1, k)),
>      +                  offset(x_frac, k & 1),
> 
>    This line can just be "x_frac" now, because k is always even, so (k & 1)
>    is always zero.

Nice catch, I probably should have paid more attention to what I was doing also.

>     
> 
>      +                  SAMPLE(index, k),
>      +                  SAMPLE(index - 1, k));
>             index -= 2;
>          }
>      -   for (int k = 0; k < 8; ++k)
>      -      brw_LRP(&func,
>      -              vec8(SAMPLE(0, k)),
>      -              offset(y_frac, k & 1),
>      -              vec8(SAMPLE(2, k)),
>      -              vec8(SAMPLE(0, k)));
>      -   brw_set_compression_control(&func, BRW_COMPRESSION_COMPRESSED);
>      -   brw_set_access_mode(&func, BRW_ALIGN_1);
>      +   for (int k = 0; k < 8; k += 2)
>      +      emit_lrp(vec8(SAMPLE(0, k)),
>      +               offset(y_frac, k & 1),
> 
>    Same comment applies here.
>     
> 
>      +               vec8(SAMPLE(2, k)),
>      +               vec8(SAMPLE(0, k)));
>       #undef SAMPLE
>       }
> 
>    With those two things fixed, this patch is:
> 
>    Reviewed-by: Paul Berry <stereotype441 at gmail.com>