[Mesa-dev] [PATCH] mesa/st: only convert AND(a, NOT(b)) into MAD when not using native integers

[Marek Olšák]

Reviewed-by: Marek Olšák <marek.olsak at amd.com>

Marek

On Sun, Aug 3, 2014 at 7:33 AM, Ilia Mirkin <imirkin at alum.mit.edu> wrote:
> Native integers imply a somewhat different handling of booleans. Instead
> of being 1.0/0.0 floats, they are 0 (true) / -1 (false) integers. As such
> the original optimization no longer applies.
>
> Reported-by: Glenn Kennard <glenn.kennard at gmail.com>
> Signed-off-by: Ilia Mirkin <imirkin at alum.mit.edu>
> ---
>
> This fixes builtin-gl-sample-mask on nvc0, and I presume radeonsi as well. The
> code being generated was a
>
>  31:   UMAD TEMP[16], TEMP[15].xxxx, -TEMP[6].xxxx, TEMP[15].xxxx
>
> instead of
>
>  27:   NOT TEMP[14].x, TEMP[6].xxxx
> ...
>  32:   AND TEMP[17].x, TEMP[14].xxxx, TEMP[16].xxxx
>
> Which is not the same when native integers are being used.
>
>  src/mesa/state_tracker/st_glsl_to_tgsi.cpp | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/src/mesa/state_tracker/st_glsl_to_tgsi.cpp b/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
> index 27e4615..b818139 100644
> --- a/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
> +++ b/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
> @@ -1354,7 +1354,7 @@ glsl_to_tgsi_visitor::visit(ir_expression *ir)
>
>     /* Quick peephole: Emit OPCODE_MAD(-a, -b, a) instead of AND(a, NOT(b))
>      */
> -   if (ir->operation == ir_binop_logic_and) {
> +   if (!native_integers && ir->operation == ir_binop_logic_and) {
>        if (try_emit_mad_for_and_not(ir, 1))
>          return;
>        if (try_emit_mad_for_and_not(ir, 0))
> --
> 1.8.5.5
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/mesa-dev