On Tue, Apr 10, 2018 at 7:37 AM, Daniel Schürmann <
daniel.schuerm...@campus.tu-berlin.de> wrote:
> ---
> src/compiler/nir/nir_lower_subgroups.c | 5 ++---
> src/compiler/nir/nir_opcodes.py | 12 ++++++------
> 2 files changed, 8 insertions(+), 9 deletions(-)
>
> diff --git a/src/compiler/nir/nir_lower_subgroups.c
> b/src/compiler/nir/nir_lower_subgroups.c
> index 0d3c83b795..9dc7be7947 100644
> --- a/src/compiler/nir/nir_lower_subgroups.c
> +++ b/src/compiler/nir/nir_lower_subgroups.c
> @@ -357,9 +357,8 @@ lower_subgroups_intrin(nir_builder *b,
> nir_intrinsic_instr *intrin,
> switch (intrin->intrinsic) {
> case nir_intrinsic_ballot_bitfield_extract:
> assert(intrin->src[1].is_ssa);
> - return nir_i2b(b, nir_iand(b, nir_ushr(b, int_val,
> - intrin->src[1].ssa),
> - nir_imm_int(b, 1)));
> + return nir_i2b(b, nir_iand(b, nir_ushr(b, int_val,
> intrin->src[1].ssa),
> + nir_imm_intN_t(b, 1, options->ballot_bit_size)));
>
Oops. This didn't automatically scale as I intended. :-/ The indentation
seems a bit weird but there's not much you can do. This hunk (as it's own
patch) looks good.
> case nir_intrinsic_ballot_bit_count_reduce:
> return nir_bit_count(b, int_val);
> case nir_intrinsic_ballot_find_lsb:
> diff --git a/src/compiler/nir/nir_opcodes.py b/src/compiler/nir/nir_
> opcodes.py
> index a762fdd220..89a6c6becc 100644
> --- a/src/compiler/nir/nir_opcodes.py
> +++ b/src/compiler/nir/nir_opcodes.py
> @@ -308,17 +308,17 @@ dst = 0;
> for (unsigned bit = 0; bit < 32; bit++)
> dst |= ((src0 >> bit) & 1) << (31 - bit);
> """)
> -unop("bit_count", tuint32, """
> +unop_convert("bit_count", tuint32, tuint, """
> dst = 0;
> -for (unsigned bit = 0; bit < 32; bit++) {
> +for (unsigned bit = 0; bit < bit_size; bit++) {
> if ((src0 >> bit) & 1)
> dst++;
> }
> """)
>
> -unop_convert("ufind_msb", tint32, tuint32, """
> +unop_convert("ufind_msb", tint32, tuint, """
> dst = -1;
> -for (int bit = 31; bit >= 0; bit--) {
> +for (int bit = bit_size - 1; bit >= 0; bit--) {
> if ((src0 >> bit) & 1) {
> dst = bit;
> break;
> @@ -340,9 +340,9 @@ for (int bit = 31; bit >= 0; bit--) {
> }
> """)
>
> -unop("find_lsb", tint32, """
> +unop_convert("find_lsb", tint32, tint, """
> dst = -1;
> -for (unsigned bit = 0; bit < 32; bit++) {
> +for (unsigned bit = 0; bit < bit_size; bit++) {
>
These three changes should probably be their own patch. They look fine to
me though.
> if ((src0 >> bit) & 1) {
> dst = bit;
> break;
> --
> 2.14.1
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
>
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
