On Tue, Dec 16, 2014 at 5:53 PM, Jason Ekstrand <ja...@jlekstrand.net> wrote:
> Originally, this field was intended for determining if the given
> instruction acted per-component or if it had mismatching source and
> destination sizes that would have to be interpreted specially. However, we
> can easily derive this from output_size == 0, so it's not really that
> useful. Also, the values we were setting in nir_opcodes.h for this field
> were completely bogus and it was never used.
> ---
> src/glsl/nir/nir.h | 33 +++++++++++++++------------------
> src/glsl/nir/nir_opcodes.c | 3 +--
> src/glsl/nir/nir_opcodes.h | 34 +++++++++++++++++-----------------
> 3 files changed, 33 insertions(+), 37 deletions(-)
>
> diff --git a/src/glsl/nir/nir.h b/src/glsl/nir/nir.h
> index a1a133c..ba04c29 100644
> --- a/src/glsl/nir/nir.h
> +++ b/src/glsl/nir/nir.h
> @@ -533,7 +533,7 @@ typedef struct {
> unsigned write_mask : 4; /* ignored if dest.is_ssa is true */
> } nir_alu_dest;
>
> -#define OPCODE(name, num_inputs, per_component, output_size, output_type, \
> +#define OPCODE(name, num_inputs, output_size, output_type, \
> input_sizes, input_types, algebraic_props) \
> nir_op_##name,
>
> @@ -565,24 +565,21 @@ typedef struct {
> unsigned num_inputs;
>
> /**
> - * If true, the opcode acts in the standard, per-component manner; the
> - * operation is performed on each component (except the ones that are
> masked
> - * out) with the input being taken from the input swizzle for that
> component.
> + * The number of components in the output
> *
> - * If false, the size of the output and inputs are explicitly given;
> swizzle
> - * and writemask are still in effect, but if the output component is
> masked
> - * out, then the input component may still be in use.
> + * If non-zero, this is the size of the output and input sizes are
> + * explicitly given; swizzle and writemask are still in effect, but if
> + * the output component is masked out, then the input component may
> + * still be in use.
> *
> - * The size of some of the inputs may be given (i.e. non-zero) even though
> - * per_component is false; in that case, each component of the input acts
> - * per-component, while the rest of the inputs and the output are normal.
> - * For example, for conditional select the condition is per-component but
> - * everything else is normal.
> - */
> - bool per_component;
> -
> - /**
> - * If per_component is false, the number of components in the output.
> + * If zero, the opcode acts in the standard, per-component manner; the
> + * operation is performed on each component (except the ones that are
> + * masked out) with the input being taken from the input swizzle for
> + * that component.
> + *
> + * The size of some of the inputs may be given (i.e. non-zero) even
> + * though output_size is zero; in that case, the inputs with a zero
> + * size act per-component, while the inputs with non-zero size don't.
> */
> unsigned output_size;
>
> @@ -594,7 +591,7 @@ typedef struct {
> nir_alu_type output_type;
>
> /**
> - * If per_component is false, the number of components in each input.
> + * The number of components in each input
> */
> unsigned input_sizes[4];
>
> diff --git a/src/glsl/nir/nir_opcodes.c b/src/glsl/nir/nir_opcodes.c
> index 3a86641..1e66c55 100644
> --- a/src/glsl/nir/nir_opcodes.c
> +++ b/src/glsl/nir/nir_opcodes.c
> @@ -27,12 +27,11 @@
>
> #include "nir.h"
>
> -#define OPCODE(_name, _num_inputs, _per_component, _output_size,
> _output_type, \
> +#define OPCODE(_name, _num_inputs, _output_size, _output_type, \
> _input_sizes, _input_types, _algebraic_props) \
> { \
> .name = #_name, \
> .num_inputs = _num_inputs, \
> - .per_component = _per_component, \
> .output_size = _output_size, \
> .output_type = _output_type, \
> .input_sizes = _input_sizes, \
> diff --git a/src/glsl/nir/nir_opcodes.h b/src/glsl/nir/nir_opcodes.h
> index 6525fe6..60198b0 100644
> --- a/src/glsl/nir/nir_opcodes.h
> +++ b/src/glsl/nir/nir_opcodes.h
> @@ -29,7 +29,7 @@
> * This header file defines all the available opcodes in one place. It
> expands
> * to a list of macros of the form:
> *
> - * OPCODE(name, num_inputs, per_component, output_size, output_type,
> + * OPCODE(name, num_inputs, output_size, output_type,
> * input_sizes, input_types, algebraic_properties)
> *
> * Which should correspond one-to-one with the nir_op_info structure. It is
> @@ -40,11 +40,11 @@
>
> #define ARR(...) { __VA_ARGS__ }
>
> -#define UNOP(name, type) OPCODE(name, 1, false, 0, type, ARR(0), ARR(type),
> 0)
> +#define UNOP(name, type) OPCODE(name, 1, 0, type, ARR(0), ARR(type), 0)
> #define UNOP_CONVERT(name, in_type, out_type) \
> - OPCODE(name, 1, false, 0, out_type, ARR(0), ARR(in_type), 0)
> + OPCODE(name, 1, 0, out_type, ARR(0), ARR(in_type), 0)
> #define UNOP_HORIZ(name, output_size, output_type, input_size, input_type) \
> - OPCODE(name, 1, true, output_size, output_type, ARR(input_size), \
> + OPCODE(name, 1, output_size, output_type, ARR(input_size), \
> ARR(input_type), 0)
>
> #define UNOP_REDUCE(name, output_size, output_type, input_type) \
> @@ -175,21 +175,21 @@ UNOP_HORIZ(fnoise4_3, 4, nir_type_float, 3,
> nir_type_float)
> UNOP_HORIZ(fnoise4_4, 4, nir_type_float, 4, nir_type_float)
>
> #define BINOP(name, type, alg_props) \
> - OPCODE(name, 2, true, 0, type, ARR(0, 0), ARR(type, type), alg_props)
> + OPCODE(name, 2, 0, type, ARR(0, 0), ARR(type, type), alg_props)
> #define BINOP_CONVERT(name, out_type, in_type, alg_props) \
> - OPCODE(name, 2, true, 0, out_type, ARR(0, 0), ARR(in_type, in_type),
> alg_props)
> + OPCODE(name, 2, 0, out_type, ARR(0, 0), ARR(in_type, in_type), alg_props)
> #define BINOP_COMPARE(name, type, alg_props) \
> - OPCODE(name, 2, true, 0, nir_type_bool, ARR(0, 0), ARR(type, type),
> alg_props)
> + OPCODE(name, 2, 0, nir_type_bool, ARR(0, 0), ARR(type, type), alg_props)
> #define BINOP_HORIZ(name, output_size, output_type, src1_size, src1_type, \
> src2_size, src2_type) \
> - OPCODE(name, 2, true, output_size, output_type, ARR(src1_size,
> src2_size), \
> + OPCODE(name, 2, output_size, output_type, ARR(src1_size, src2_size), \
> ARR(src1_type, src2_type), 0)
> #define BINOP_REDUCE(name, output_size, output_type, src_type) \
> - OPCODE(name##2, 2, false, output_size, output_type, \
> + OPCODE(name##2, 2, output_size, output_type, \
> ARR(2, 2), ARR(src_type, src_type), nir_op_is_commutative) \
> - OPCODE(name##3, 2, false, output_size, output_type, \
> + OPCODE(name##3, 2, output_size, output_type, \
> ARR(3, 3), ARR(src_type, src_type), nir_op_is_commutative) \
> - OPCODE(name##4, 2, false, output_size, output_type, \
> + OPCODE(name##4, 2, output_size, output_type, \
> ARR(4, 4), ARR(src_type, src_type), nir_op_is_commutative)
>
> BINOP(fadd, nir_type_float, nir_op_is_commutative | nir_op_is_associative)
> @@ -314,9 +314,9 @@ BINOP(ldexp, nir_type_unsigned, 0)
> BINOP_HORIZ(vec2, 2, nir_type_unsigned, 1, nir_type_unsigned, 1,
> nir_type_unsigned)
>
> #define TRIOP(name, type) \
> - OPCODE(name, 3, true, 0, type, ARR(0, 0, 0), ARR(type, type, type), 0)
> + OPCODE(name, 3, 0, type, ARR(0, 0, 0), ARR(type, type, type), 0)
> #define TRIOP_HORIZ(name, output_size, src1_size, src2_size, src3_size) \
> - OPCODE(name, 3, false, output_size, nir_type_unsigned, \
> + OPCODE(name, 3, output_size, nir_type_unsigned, \
> ARR(src1_size, src2_size, src3_size), \
> ARR(nir_type_unsigned, nir_type_unsigned, nir_type_unsigned), 0)
>
> @@ -334,13 +334,13 @@ TRIOP(flrp, nir_type_float)
> */
>
> TRIOP(fcsel, nir_type_float)
> -OPCODE(bcsel, 3, true, 0, nir_type_unsigned, ARR(0, 0, 0),
> +OPCODE(bcsel, 3, 0, nir_type_unsigned, ARR(0, 0, 0),
> ARR(nir_type_bool, nir_type_unsigned, nir_type_unsigned), 0)
>
> TRIOP(bfi, nir_type_unsigned)
>
> TRIOP(ubitfield_extract, nir_type_unsigned)
> -OPCODE(ibitfield_extract, 3, true, 0, nir_type_int, ARR(0, 0, 0),
> +OPCODE(ibitfield_extract, 3, 0, nir_type_int, ARR(0, 0, 0),
> ARR(nir_type_int, nir_type_unsigned, nir_type_unsigned), 0)
>
> /**
> @@ -349,12 +349,12 @@ OPCODE(ibitfield_extract, 3, true, 0, nir_type_int,
> ARR(0, 0, 0),
> TRIOP_HORIZ(vec3, 3, 1, 1, 1)
>
> #define QUADOP(name) \
> - OPCODE(name, 4, true, 0, nir_type_unsigned, ARR(0, 0, 0, 0), \
> + OPCODE(name, 4, 0, nir_type_unsigned, ARR(0, 0, 0, 0), \
> ARR(nir_type_unsigned, nir_type_unsigned, nir_type_unsigned,
> nir_type_unsigned), \
> 0)
> #define QUADOP_HORIZ(name, output_size, src1_size, src2_size, src3_size, \
> src4_size) \
> - OPCODE(name, 4, false, output_size, nir_type_unsigned, \
> + OPCODE(name, 4, output_size, nir_type_unsigned, \
> ARR(src1_size, src2_size, src3_size, src4_size), \
> ARR(nir_type_unsigned, nir_type_unsigned, nir_type_unsigned,
> nir_type_unsigned), \
> 0)
> --
> 2.2.0
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev@lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/mesa-dev
Yeah, per_component definitely shouldn't have been there in the first place...
Reviewed-by: Connor Abbott <cwabbo...@gmail.com>
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/mesa-dev
