<div dir="ltr"><div>LGTM. I would still rather do the allocation in live_variables not on the stack, but I can clean that up later. </div>Reviewed-by: Jason Ekstrand <<a href="mailto:jason.ekstrand@intel.com">jason.ekstrand@intel.com</a>> </div><div class="gmail_extra"> <div class="gmail_quote">On Tue, Mar 3, 2015 at 5:19 AM, Jose Fonseca <<a href="mailto:jfonseca@vmware.com" target="_blank">jfonseca@vmware.com</a>> wrote: <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">v2: - Single statement, by using memset return value as suggested by Ian Romanick. - No internal declaration, as suggested by Jason Ekstrand. - Move macros to a header. --- src/glsl/nir/nir_from_ssa.c | 23 ++++++--------- src/glsl/nir/nir_live_variables.c | 4 +-- src/glsl/nir/nir_lower_vars_to_ssa.c | 12 +++----- src/glsl/nir/nir_vla.h | 54 ++++++++++++++++++++++++++++++++++++ 4 files changed, 69 insertions(+), 24 deletions(-) create mode 100644 src/glsl/nir/nir_vla.h diff --git a/src/glsl/nir/nir_from_ssa.c b/src/glsl/nir/nir_from_ssa.c index 66339f3..c3090fb 100644 --- a/src/glsl/nir/nir_from_ssa.c +++ b/src/glsl/nir/nir_from_ssa.c @@ -26,7 +26,7 @@ */ #include "nir.h" -#include "c99_alloca.h" +#include "nir_vla.h" /* * This file implements an out-of-SSA pass as described in "Revisiting @@ -182,7 +182,7 @@ merge_merge_sets(merge_set *a, merge_set *b) static bool merge_sets_interfere(merge_set *a, merge_set *b) { - merge_node **dom = alloca((a->size + b->size) * sizeof *dom); + NIR_VLA(merge_node *, dom, a->size + b->size); int dom_idx = -1; struct exec_node *an = exec_list_get_head(&a->nodes); @@ -674,21 +674,16 @@ resolve_parallel_copy(nir_parallel_copy_instr *pcopy, } /* The register/source corresponding to the given index */ - nir_src *values = alloca(num_copies * 2 * sizeof *values); - memset(values, 0, num_copies * 2 * sizeof *values); + NIR_VLA_ZERO(nir_src, values, num_copies * 2); - /* The current location of a given piece of data */ - int *loc = alloca(num_copies * 2 * sizeof *loc); + /* The current location of a given piece of data. We will use -1 for "null" */ + NIR_VLA_FILL(int, loc, num_copies * 2, -1); - /* The piece of data that the given piece of data is to be copied from */ - int *pred = alloca(num_copies * 2 * sizeof *pred); - - /* Initialize loc and pred. We will use -1 for "null" */ - memset(loc, -1, num_copies * 2 * sizeof *loc); - memset(pred, -1, num_copies * 2 * sizeof *pred); + /* The piece of data that the given piece of data is to be copied from. We will use -1 for "null" */ + NIR_VLA_FILL(int, pred, num_copies * 2, -1); /* The destinations we have yet to properly fill */ - int *to_do = alloca(num_copies * 2 * sizeof *to_do); + NIR_VLA(int, to_do, num_copies * 2); int to_do_idx = -1; /* Now we set everything up: @@ -738,7 +733,7 @@ resolve_parallel_copy(nir_parallel_copy_instr *pcopy, } /* Currently empty destinations we can go ahead and fill */ - int *ready = alloca(num_copies * 2 * sizeof *ready); + NIR_VLA(int, ready, num_copies * 2); int ready_idx = -1; /* Mark the ones that are ready for copying. We know an index is a diff --git a/src/glsl/nir/nir_live_variables.c b/src/glsl/nir/nir_live_variables.c index b57ca3a..1c96dcf 100644 --- a/src/glsl/nir/nir_live_variables.c +++ b/src/glsl/nir/nir_live_variables.c @@ -26,7 +26,7 @@ #include "nir.h" #include "nir_worklist.h" -#include "c99_alloca.h" +#include "nir_vla.h" /* * Basic liveness analysis. This works only in SSA form. @@ -131,7 +131,7 @@ static bool propagate_across_edge(nir_block *pred, nir_block *succ, struct live_variables_state *state) { - BITSET_WORD *live = alloca(state->bitset_words * sizeof *live); + NIR_VLA(BITSET_WORD, live, state->bitset_words); memcpy(live, succ->live_in, state->bitset_words * sizeof *live); nir_foreach_instr(succ, instr) { diff --git a/src/glsl/nir/nir_lower_vars_to_ssa.c b/src/glsl/nir/nir_lower_vars_to_ssa.c index f54d1b7..9e9a418 100644 --- a/src/glsl/nir/nir_lower_vars_to_ssa.c +++ b/src/glsl/nir/nir_lower_vars_to_ssa.c @@ -26,8 +26,7 @@ */ #include "nir.h" - -#include "c99_alloca.h" +#include "nir_vla.h" struct deref_node { @@ -902,8 +901,8 @@ rename_variables_block(nir_block *block, struct lower_variables_state *state) static void insert_phi_nodes(struct lower_variables_state *state) { - unsigned *work = alloca(state->impl->num_blocks * sizeof *work); - unsigned *has_already = alloca(state->impl->num_blocks * sizeof *has_already); + NIR_VLA_ZERO(unsigned, work, state->impl->num_blocks); + NIR_VLA_ZERO(unsigned, has_already, state->impl->num_blocks); /* * Since the work flags already prevent us from inserting a node that has @@ -913,10 +912,7 @@ insert_phi_nodes(struct lower_variables_state *state) * function. So all we need to handle W is an array and a pointer to the * next element to be inserted and the next element to be removed. */ - nir_block **W = alloca(state->impl->num_blocks * sizeof *W); - - memset(work, 0, state->impl->num_blocks * sizeof *work); - memset(has_already, 0, state->impl->num_blocks * sizeof *has_already); + NIR_VLA(nir_block *, W, state->impl->num_blocks); unsigned w_start, w_end; unsigned iter_count = 0; diff --git a/src/glsl/nir/nir_vla.h b/src/glsl/nir/nir_vla.h new file mode 100644 index 0000000..7537833 --- /dev/null +++ b/src/glsl/nir/nir_vla.h @@ -0,0 +1,54 @@ +/************************************************************************** + * + * Copyright 2015 VMware, Inc. + * All Rights Reserved. + * + * 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, sub license, 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 NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS 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. + * + **************************************************************************/ + +#pragma once + + +#include "c99_alloca.h" + + +/* Declare a variable length array, with no initialization */ +#define NIR_VLA(_type, _name, _length) \ + _type *_name = alloca((_length) * sizeof *_name) + + +/* Declare a variable length array, and initialize it with the given byte. + * + * _length is evaluated twice, so expressions with side-effects must be + * avoided. + */ +#define NIR_VLA_FILL(_type, _name, _length, _byte) \ + _type *_name = memset(alloca((_length) * sizeof *_name), _byte, (_length) * sizeof *_name) + + +/* Declare a variable length array, and zero it. + * + * Just like NIR_VLA_FILL, _length is evaluated twice, so expressions with + * side-effects must be avoided. + */ +#define NIR_VLA_ZERO(_type, _name, _length) \ + NIR_VLA_FILL(_type, _name, _length, 0) -- 2.1.0 </blockquote></div> </div>