From f1c2db43dcf35d2cf4715390bd2391c28e42a8c2 Mon Sep 17 00:00:00 2001 From: marha Date: Sun, 22 Feb 2015 14:31:16 +0100 Subject: xwininfo fontconfig libX11 libXdmcp libfontenc libxcb libxcb/xcb-proto mesalib xserver xkeyboard-config mkfontscale git update 22 Feb 2015 xserver commit 3a06faf3fcdb7451125a46181f9152e8e59e9770 libxcb commit e3ec1f74637237ce500dfd0ca59f2e422da4e019 libxcb/xcb-proto commit 4c550465934164aab2449a125f75f4ca07816233 xkeyboard-config commit 26f344c93f8c6141e9233eb68088ba4fd56bc9ef libX11 commit c8e19b393defd53f046ddc2da3a16881221b3c34 libXdmcp commit 9f4cac7656b221ce2a8f97e7bd31e5e23126d001 libfontenc commit de1843aaf76015c9d99416f3122d169fe331b849 mkfontscale commit 87d628f8eec170ec13bb9feefb1ce05aed07d1d6 xwininfo commit 0c49f8f2bd56b1e77721e81030ea948386dcdf4e fontconfig commit d6d5adeb7940c0d0beb86489c2a1c2ce59e5c044 mesa commit 4359954d842caa2a9f8d4b50d70ecc789884b68b --- mesalib/src/glsl/nir/nir_lower_vec_to_movs.c | 155 +++++++++++++++++++++++++++ 1 file changed, 155 insertions(+) create mode 100644 mesalib/src/glsl/nir/nir_lower_vec_to_movs.c (limited to 'mesalib/src/glsl/nir/nir_lower_vec_to_movs.c') diff --git a/mesalib/src/glsl/nir/nir_lower_vec_to_movs.c b/mesalib/src/glsl/nir/nir_lower_vec_to_movs.c new file mode 100644 index 000000000..602853ea6 --- /dev/null +++ b/mesalib/src/glsl/nir/nir_lower_vec_to_movs.c @@ -0,0 +1,155 @@ +/* + * Copyright © 2014 Intel Corporation + * + * 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. + * + * Authors: + * Jason Ekstrand (jason@jlekstrand.net) + * + */ + +#include "nir.h" + +/* + * Implements a simple pass that lowers vecN instructions to a series of + * moves with partial writes. + */ + +static bool +src_matches_dest_reg(nir_dest *dest, nir_src *src) +{ + if (dest->is_ssa || src->is_ssa) + return false; + + return (dest->reg.reg == src->reg.reg && + dest->reg.base_offset == src->reg.base_offset && + !dest->reg.indirect && + !src->reg.indirect); +} + +/** + * For a given starting writemask channel and corresponding source index in + * the vec instruction, insert a MOV to the vec instruction's dest of all the + * writemask channels that get read from the same src reg. + * + * Returns the writemask of our MOV, so the parent loop calling this knows + * which ones have been processed. + */ +static unsigned +insert_mov(nir_alu_instr *vec, unsigned start_channel, + unsigned start_src_idx, void *mem_ctx) +{ + unsigned src_idx = start_src_idx; + assert(src_idx < nir_op_infos[vec->op].num_inputs); + + nir_alu_instr *mov = nir_alu_instr_create(mem_ctx, nir_op_imov); + nir_alu_src_copy(&mov->src[0], &vec->src[src_idx], mem_ctx); + nir_alu_dest_copy(&mov->dest, &vec->dest, mem_ctx); + + mov->dest.write_mask = (1u << start_channel); + mov->src[0].swizzle[start_channel] = vec->src[src_idx].swizzle[0]; + src_idx++; + + for (unsigned i = start_channel + 1; i < 4; i++) { + if (!(vec->dest.write_mask & (1 << i))) + continue; + + if (nir_srcs_equal(vec->src[src_idx].src, vec->src[start_src_idx].src)) { + mov->dest.write_mask |= (1 << i); + mov->src[0].swizzle[i] = vec->src[src_idx].swizzle[0]; + } + src_idx++; + } + + nir_instr_insert_before(&vec->instr, &mov->instr); + + return mov->dest.write_mask; +} + +static bool +lower_vec_to_movs_block(nir_block *block, void *mem_ctx) +{ + nir_foreach_instr_safe(block, instr) { + if (instr->type != nir_instr_type_alu) + continue; + + nir_alu_instr *vec = (nir_alu_instr *)instr; + + switch (vec->op) { + case nir_op_vec2: + case nir_op_vec3: + case nir_op_vec4: + break; + default: + continue; /* The loop */ + } + + /* Since we insert multiple MOVs, we have to be non-SSA. */ + assert(!vec->dest.dest.is_ssa); + + unsigned finished_write_mask = 0; + + /* First, emit a MOV for all the src channels that are in the + * destination reg, in case other values we're populating in the dest + * might overwrite them. + */ + for (unsigned i = 0, src_idx = 0; i < 4; i++) { + if (!(vec->dest.write_mask & (1 << i))) + continue; + + if (src_matches_dest_reg(&vec->dest.dest, &vec->src[src_idx].src)) { + finished_write_mask |= insert_mov(vec, i, src_idx, mem_ctx); + break; + } + src_idx++; + } + + /* Now, emit MOVs for all the other src channels. */ + for (unsigned i = 0, src_idx = 0; i < 4; i++) { + if (!(vec->dest.write_mask & (1 << i))) + continue; + + if (!(finished_write_mask & (1 << i))) + finished_write_mask |= insert_mov(vec, i, src_idx, mem_ctx); + + src_idx++; + } + + nir_instr_remove(&vec->instr); + ralloc_free(vec); + } + + return true; +} + +static void +nir_lower_vec_to_movs_impl(nir_function_impl *impl) +{ + nir_foreach_block(impl, lower_vec_to_movs_block, ralloc_parent(impl)); +} + +void +nir_lower_vec_to_movs(nir_shader *shader) +{ + nir_foreach_overload(shader, overload) { + if (overload->impl) + nir_lower_vec_to_movs_impl(overload->impl); + } +} -- cgit v1.2.3