On 10/14/2018 03:11 PM, Matt Turner wrote:
> Unfortunately some int64 lowerings generate more int64 operations, so we
> need to call this function a few times. Also call
> nir_lower_alu_to_scalar() beforehand to make more int64 operations
> available for lowering.
> ---
> src/intel/compiler/brw_nir.c | 10 +++++++---
> 1 file changed, 7 insertions(+), 3 deletions(-)
>
> diff --git a/src/intel/compiler/brw_nir.c b/src/intel/compiler/brw_nir.c
> index f61baee230a..066724c58a6 100644
> --- a/src/intel/compiler/brw_nir.c
> +++ b/src/intel/compiler/brw_nir.c
> @@ -670,12 +670,16 @@ brw_preprocess_nir(const struct brw_compiler *compiler,
> nir_shader *nir)
> */
> OPT(nir_opt_algebraic);
>
> + if (is_scalar) {
> + OPT(nir_lower_alu_to_scalar);
> + }
> +
> /* Lower int64 instructions before nir_optimize so that loop unrolling
> * sees their actual cost.
> */
> - nir_lower_int64(nir, nir_lower_imul64 |
> - nir_lower_isign64 |
> - nir_lower_divmod64);
> + while (nir_lower_int64(nir, nir_lower_imul64 |
> + nir_lower_isign64 |
> + nir_lower_divmod64));
I don't know that we have a specific coding standard about this, but I
always like
while (foo)
/* empty */ ;
just to make it obvious.
>
> nir = brw_nir_optimize(nir, compiler, is_scalar, true);
>
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
