[Mesa-dev] [PATCH 3/4] nir: Add a lowering pass for texture projectors.
Jason Ekstrand
jason at jlekstrand.net
Fri Apr 3 11:56:07 PDT 2015
On Wed, Apr 1, 2015 at 3:11 PM, Eric Anholt <eric at anholt.net> wrote:
> Not much hardware wants them these days, and it might give us a chance to
> do CSE or algebraic at the NIR level.
> ---
>
> I wrote this originally for vc4, but I'm not sure I'm going to turn it
> on -- I'm using a non-Newton-Raphson RCP in my TXP handling right now,
> and if I do this pass then I get the N-R step added. If it's not
> necessary, I probably shouldn't spend those instructions on it.
>
> src/glsl/Makefile.sources | 1 +
> src/glsl/nir/nir.h | 1 +
> src/glsl/nir/nir_lower_tex_projector.c | 142 +++++++++++++++++++++++++++++++++
> 3 files changed, 144 insertions(+)
> create mode 100644 src/glsl/nir/nir_lower_tex_projector.c
>
> diff --git a/src/glsl/Makefile.sources b/src/glsl/Makefile.sources
> index b56fa26..ffce706 100644
> --- a/src/glsl/Makefile.sources
> +++ b/src/glsl/Makefile.sources
> @@ -37,6 +37,7 @@ NIR_FILES = \
> nir/nir_lower_phis_to_scalar.c \
> nir/nir_lower_samplers.cpp \
> nir/nir_lower_system_values.c \
> + nir/nir_lower_tex_projector.c \
> nir/nir_lower_to_source_mods.c \
> nir/nir_lower_vars_to_ssa.c \
> nir/nir_lower_var_copies.c \
> diff --git a/src/glsl/nir/nir.h b/src/glsl/nir/nir.h
> index 24deb82..6e2aa97 100644
> --- a/src/glsl/nir/nir.h
> +++ b/src/glsl/nir/nir.h
> @@ -1601,6 +1601,7 @@ void nir_lower_samplers(nir_shader *shader,
> struct gl_program *prog);
>
> void nir_lower_system_values(nir_shader *shader);
> +void nir_lower_tex_projector(nir_shader *shader);
>
> void nir_lower_atomics(nir_shader *shader);
> void nir_lower_to_source_mods(nir_shader *shader);
> diff --git a/src/glsl/nir/nir_lower_tex_projector.c b/src/glsl/nir/nir_lower_tex_projector.c
> new file mode 100644
> index 0000000..6327b23
> --- /dev/null
> +++ b/src/glsl/nir/nir_lower_tex_projector.c
> @@ -0,0 +1,142 @@
> +/*
> + * Copyright © 2015 Broadcom
> + *
> + * Permission is hereby granted, free of charge, to any person obtaining a
> + * copy of this software and associated documentation files (the "Software"),
> + * to deal in the Software without restriction, including without limitation
> + * the rights to use, copy, modify, merge, publish, distribute, sublicense,
> + * and/or sell copies of the Software, and to permit persons to whom the
> + * Software is furnished to do so, subject to the following conditions:
> + *
> + * The above copyright notice and this permission notice (including the next
> + * paragraph) shall be included in all copies or substantial portions of the
> + * Software.
> + *
> + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
> + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
> + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
> + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
> + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
> + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
> + * IN THE SOFTWARE.
> + */
> +
> +/*
> + * This lowering pass converts the coordinate division for texture projection
> + * to be done in ALU instructions instead of asking the texture operation to
> + * do so.
> + */
> +
> +#include "nir.h"
> +#include "nir_builder.h"
> +
> +static nir_ssa_def *
> +channel(nir_builder *b, nir_ssa_def *def, int c)
> +{
> + return nir_swizzle(b, def, (unsigned[4]){c, c, c, c}, 1, false);
> +}
> +
> +static bool
> +nir_lower_tex_projector_block(nir_block *block, void *void_state)
> +{
> + nir_builder *b = void_state;
> +
> + nir_foreach_instr_safe(block, instr) {
> + if (instr->type != nir_instr_type_tex)
> + continue;
> +
> + nir_tex_instr *tex = nir_instr_as_tex(instr);
> + nir_builder_insert_before_instr(b, &tex->instr);
> +
> + /* Find the projector in the srcs list, if present. */
> + int proj_index;
> + for (proj_index = 0; proj_index < tex->num_srcs; proj_index++) {
> + if (tex->src[proj_index].src_type == nir_tex_src_projector)
> + break;
> + }
> + if (proj_index == tex->num_srcs)
> + continue;
> + nir_ssa_def *inv_proj =
> + nir_frcp(b, nir_ssa_for_src(b, tex->src[proj_index].src, 1));
> +
> + /* Walk through the sources projecting the arguments. */
> + for (int i = 0; i < tex->num_srcs; i++) {
> + switch (tex->src[i].src_type) {
> + case nir_tex_src_coord:
> + case nir_tex_src_comparitor:
> + break;
> + default:
> + continue;
> + }
> + nir_ssa_def *unprojected =
> + nir_ssa_for_src(b, tex->src[i].src, nir_tex_instr_src_size(tex, i));
> + nir_ssa_def *projected = nir_fmul(b, unprojected, inv_proj);
> +
> + /* Array indices don't get projected, so make an new vector with the
> + * coordinate's array index untouched.
> + */
> + if (tex->is_array && tex->src[i].src_type == nir_tex_src_coord) {
I don't think this is quite correct. I ran into this issued in my
cubemap pass where some opcodes such as lod_query act on array
textures but don't actually have an array index component. For those,
you want to use the original projected thing instead of leaving the
last component alone.
> + switch (tex->coord_components) {
> + case 4:
> + projected = nir_vec4(b,
> + channel(b, projected, 0),
> + channel(b, projected, 1),
> + channel(b, projected, 2),
> + channel(b, unprojected, 3));
> + break;
> + case 3:
> + projected = nir_vec3(b,
> + channel(b, projected, 0),
> + channel(b, projected, 1),
> + channel(b, unprojected, 2));
> + break;
> + case 2:
> + projected = nir_vec2(b,
> + channel(b, projected, 0),
> + channel(b, unprojected, 1));
> + break;
> + default:
> + unreachable("bad texture coord count for array");
> + break;
> + }
> + }
> +
> + nir_instr_rewrite_src(&tex->instr,
> + &tex->src[i].src,
> + nir_src_for_ssa(projected));
> + }
> +
> + /* Now move the later tex sources down the array so that the projector
> + * disappears.
> + */
> + nir_src dead = {.is_ssa = false, .ssa = NULL};
> + nir_instr_rewrite_src(&tex->instr, &tex->src[proj_index].src, dead);
> + memmove(&tex->src[proj_index],
> + &tex->src[proj_index + 1],
> + (tex->num_srcs - proj_index) * sizeof(*tex->src));
> + tex->num_srcs--;
> + }
> +
> + return true;
> +}
> +
> +static void
> +nir_lower_tex_projector_impl(nir_function_impl *impl)
> +{
> + nir_builder b;
> + nir_builder_init(&b, impl);
> +
> + nir_foreach_block(impl, nir_lower_tex_projector_block, &b);
> +
> + nir_metadata_preserve(impl, nir_metadata_block_index |
> + nir_metadata_dominance);
> +}
> +
> +void
> +nir_lower_tex_projector(nir_shader *shader)
> +{
> + nir_foreach_overload(shader, overload) {
> + if (overload->impl)
> + nir_lower_tex_projector_impl(overload->impl);
> + }
> +}
> --
> 2.1.4
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/mesa-dev
More information about the mesa-dev
mailing list