[Mesa-dev] [RFC 2/5] nir: Add unordered comparisons and ordered fne

[Bas Nieuwenhuizen]

On Thu, Nov 22, 2018 at 7:47 PM Jason Ekstrand <jason at jlekstrand.net> wrote:
>
> ---
>  src/compiler/nir/nir.h                |  8 ++++++++
>  src/compiler/nir/nir_loop_analyze.c   | 12 ++++++++----
>  src/compiler/nir/nir_opcodes.py       |  4 ++++
>  src/compiler/nir/nir_opt_algebraic.py |  5 +++++
>  4 files changed, 25 insertions(+), 4 deletions(-)
>
> diff --git a/src/compiler/nir/nir.h b/src/compiler/nir/nir.h
> index 4271f237235..20ff9a87297 100644
> --- a/src/compiler/nir/nir.h
> +++ b/src/compiler/nir/nir.h
> @@ -1561,6 +1561,10 @@ nir_alu_instr_is_comparison(const nir_alu_instr *instr)
>     case nir_op_flt:
>     case nir_op_fge:
>     case nir_op_feq:
> +   case nir_op_fne:
> +   case nir_op_fltu:
> +   case nir_op_fgeu:
> +   case nir_op_fequ:
>     case nir_op_fneu:
>     case nir_op_ilt:
>     case nir_op_ult:
> @@ -2132,6 +2136,10 @@ typedef struct nir_shader_compiler_options {
>
>     bool lower_ldexp;
>
> +   bool lower_fltu;
> +   bool lower_fgeu;
> +   bool lower_fne_to_fequ;
> +
>     bool lower_pack_half_2x16;
>     bool lower_pack_unorm_2x16;
>     bool lower_pack_snorm_2x16;
> diff --git a/src/compiler/nir/nir_loop_analyze.c b/src/compiler/nir/nir_loop_analyze.c
> index d73314a8a44..d6cba541a10 100644
> --- a/src/compiler/nir/nir_loop_analyze.c
> +++ b/src/compiler/nir/nir_loop_analyze.c
> @@ -376,6 +376,10 @@ get_iteration(nir_op cond_op, nir_const_value *initial, nir_const_value *step,
>     case nir_op_fge:
>     case nir_op_flt:
>     case nir_op_feq:
> +   case nir_op_fne:
> +   case nir_op_fgeu:
> +   case nir_op_fltu:
> +   case nir_op_fequ:
>     case nir_op_fneu: {
>        float initial_val = initial->f32[0];
>        float span = limit->f32[0] - initial_val;
> @@ -547,10 +551,10 @@ find_trip_count(loop_info_state *state)
>        bool limit_rhs = true;
>
>        switch (alu->op) {
> -      case nir_op_fge:      case nir_op_ige:      case nir_op_uge:
> -      case nir_op_flt:      case nir_op_ilt:      case nir_op_ult:
> -      case nir_op_feq:      case nir_op_ieq:
> -      case nir_op_fneu:     case nir_op_ine:
> +      case nir_op_fgeu: case nir_op_fge:  case nir_op_ige:  case nir_op_uge:
> +      case nir_op_fltu: case nir_op_flt:  case nir_op_ilt:  case nir_op_ult:
> +      case nir_op_fequ: case nir_op_feq:  case nir_op_ieq:
> +      case nir_op_fneu: case nir_op_fne:  case nir_op_ine:
>
>           /* We assume that the limit is the "right" operand */
>           basic_ind = get_loop_var(alu->src[0].src.ssa, state);
> diff --git a/src/compiler/nir/nir_opcodes.py b/src/compiler/nir/nir_opcodes.py
> index 2375309aca6..032168bae49 100644
> --- a/src/compiler/nir/nir_opcodes.py
> +++ b/src/compiler/nir/nir_opcodes.py
> @@ -491,6 +491,10 @@ binop("frem", tfloat, "", "src0 - src1 * truncf(src0 / src1)")
>  binop_compare("flt", tfloat, "", "src0 < src1")
>  binop_compare("fge", tfloat, "", "src0 >= src1")
>  binop_compare("feq", tfloat, commutative, "src0 == src1")
> +binop_compare("fne", tfloat, commutative, "!isnan(src0) && !isnan(src1) && src0 == src1")
> +binop_compare("fltu", tfloat, "", "!(src0 >= src1)")
> +binop_compare("fgeu", tfloat, "", "!(src0 < src1)")
> +binop_compare("fequ", tfloat, commutative, "isnan(src0) || isnan(src1) || src0 == src1")

Please add some comment around this that the ones without suffix are
ordered and the one with u suffix are unordered.

Otherwise

Reviewed-by: Bas Nieuwenhuizen <bas at basnieuwenhuizen.nl

for 1-3, 5
>  binop_compare("fneu", tfloat, commutative, "src0 != src1")
>  binop_compare("ilt", tint, "", "src0 < src1")
>  binop_compare("ige", tint, "", "src0 >= src1")
> diff --git a/src/compiler/nir/nir_opt_algebraic.py b/src/compiler/nir/nir_opt_algebraic.py
> index 131d2721934..64327708a66 100644
> --- a/src/compiler/nir/nir_opt_algebraic.py
> +++ b/src/compiler/nir/nir_opt_algebraic.py
> @@ -164,6 +164,11 @@ optimizations = [
>     (('inot', ('ieq', a, b)), ('ine', a, b)),
>     (('inot', ('ine', a, b)), ('ieq', a, b)),
>
> +   # Comparison lowering
> +   (('fltu', a, b), ('inot', ('fge', a, b)), 'options->lower_fltu'),
> +   (('fgeu', a, b), ('inot', ('flt', a, b)), 'options->lower_fgeu'),
> +   (('fne', a, b), ('inot', ('fequ', a, b)), 'options->lower_fne_to_fequ'),
> +
>     # 0.0 >= b2f(a)
>     # b2f(a) <= 0.0
>     # b2f(a) == 0.0 because b2f(a) can only be 0 or 1
> --
> 2.19.1
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev