On Mon, 2016-01-25 at 15:18 -0800, Matt Turner wrote:
> The uint versions zero extend while the int versions sign extend.
> ---
> src/glsl/nir/nir.h | 3 +++
> src/glsl/nir/nir_opcodes.py | 9 +++++++++
> src/glsl/nir/nir_opt_algebraic.py | 16 ++++++++++++++++
> 3 files changed, 28 insertions(+)
>
> diff --git a/src/glsl/nir/nir.h b/src/glsl/nir/nir.h
> index 1113030..7b39cbb 100644
> --- a/src/glsl/nir/nir.h
> +++ b/src/glsl/nir/nir.h
> @@ -1471,6 +1471,9 @@ typedef struct nir_shader_compiler_options {
> bool lower_pack_half_2x16;
> bool lower_unpack_half_2x16;
>
> + bool lower_extract_byte;
> + bool lower_extract_word;
> +
> /**
> * Does the driver support real 32-bit integers? (Otherwise, integers
> * are simulated by floats.)
> diff --git a/src/glsl/nir/nir_opcodes.py b/src/glsl/nir/nir_opcodes.py
> index a8bbe1a..be3cd17 100644
> --- a/src/glsl/nir/nir_opcodes.py
> +++ b/src/glsl/nir/nir_opcodes.py
> @@ -536,6 +536,15 @@ dst.x = src0.x;
> dst.y = src1.x;
> """)
>
> +# Byte extraction
> +binop("extract_ubyte", tuint, "", "(uint8_t)(src0 >> (src1 * 8))")
> +binop("extract_ibyte", tint, "", "(int8_t)(src0 >> (src1 * 8))")
> +
> +# Word extraction
> +binop("extract_uword", tuint, "", "(uint16_t)(src0 >> (src1 * 16))")
> +binop("extract_iword", tint, "", "(int16_t)(src0 >> (src1 * 16))")
> +
> +
> def triop(name, ty, const_expr):
> opcode(name, 0, ty, [0, 0, 0], [ty, ty, ty], "", const_expr)
> def triop_horiz(name, output_size, src1_size, src2_size, src3_size,
> const_expr):
> diff --git a/src/glsl/nir/nir_opt_algebraic.py
> b/src/glsl/nir/nir_opt_algebraic.py
> index 7745b76..b761b54 100644
> --- a/src/glsl/nir/nir_opt_algebraic.py
> +++ b/src/glsl/nir/nir_opt_algebraic.py
> @@ -242,6 +242,22 @@ optimizations = [
> ('bcsel', ('ult', 31, 'bits'), 'value',
> ('ubfe', 'value', 'offset', 'bits')),
> 'options->lower_bitfield_extract'),
> +
> + (('extract_ibyte', a, b),
> + ('ishr', ('ishl', a, ('imul', ('isub', 3, b), 8)), 8),
> + 'options->lower_extract_byte'),
> +
Is this correct? If I get this right, extract_ibyte(0x00ff0000, 2)
should return 0xff, but this seems to do:
(N << 8) >> 8 = N = 0x00ff0000
It seems like you would like to shr by 24 instead of 8.
With this fixed:
Reviewed-by: Iago Toral Quiroga <ito...@igalia.com>
> + (('extract_ubyte', a, b),
> + ('iand', ('ushr', a, ('imul', b, 8)), 0xff),
> + 'options->lower_extract_byte'),
> +
> + (('extract_iword', a, b),
> + ('ishr', ('ishl', a, ('imul', ('isub', 1, b), 16)), 16),
> + 'options->lower_extract_word'),
> +
> + (('extract_uword', a, b),
> + ('iand', ('ushr', a, ('imul', b, 16)), 0xffff),
> + 'options->lower_extract_word'),
> ]
>
> # Add optimizations to handle the case where the result of a ternary is
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/mesa-dev
