[Mesa-dev] [PATCH 2/4] intel/compiler: Don't propagate cmod into integer multiplies

[Matt Turner]

On Wed, Oct 4, 2017 at 4:58 PM, Jason Ekstrand <jason at jlekstrand.net> wrote:
> No shader-db change on Sky Lake.
>
> Cc: mesa-stable at lists.freedesktop.org
> ---
>  src/intel/compiler/brw_fs_cmod_propagation.cpp   | 17 +++++++++++++++++
>  src/intel/compiler/brw_vec4_cmod_propagation.cpp | 17 +++++++++++++++++
>  2 files changed, 34 insertions(+)
>
> diff --git a/src/intel/compiler/brw_fs_cmod_propagation.cpp b/src/intel/compiler/brw_fs_cmod_propagation.cpp
> index db63e94..e8f1069 100644
> --- a/src/intel/compiler/brw_fs_cmod_propagation.cpp
> +++ b/src/intel/compiler/brw_fs_cmod_propagation.cpp
> @@ -150,6 +150,23 @@ opt_cmod_propagation_local(const gen_device_info *devinfo, bblock_t *block)
>              if (scan_inst->saturate)
>                 break;
>
> +            /* From the Sky Lake PRM, Vol 2a, "Multiply":
> +             *
> +             *    "When multiplying integer data types, if one of the sources
> +             *    is a DW, the resulting full precision data is stored in
> +             *    the accumulator. However, if the destination data type is
> +             *    either W or DW, the low bits of the result are written to
> +             *    the destination register and the remaining high bits are
> +             *    discarded. This results in undefined Overflow and Sign
> +             *    flags. Therefore, conditional modifiers and saturation
> +             *    (.sat) cannot be used in this case.

Please indent the lines in the block quote one space more than the "

Patches 1-2 are

Reviewed-by: Matt Turner <mattst88 at gmail.com>