mirror/gcc
1
0
Fork 0
mirror of https://github.com/gcc-mirror/gcc.git synced 2024-11-21 13:40:47 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

tree-optimization/116973 - SLP permute lower heuristic and single-lane SLP

Browse Source 
When forcing single-lane SLP to emulate non-SLP behavior we need to
disable heuristics designed to optimize SLP loads and instead in
all cases resort to an interleaving scheme as requested by forcefully
doing single-lane SLP.

This fixes the remaining fallout for --param vect-force-slp=1 on x86.

	PR tree-optimization/116973
	* tree-vect-slp.cc (vect_lower_load_permutations): Add
	force_single_lane parameter.  Disable heuristic that keeps
	some load-permutations.
	(vect_analyze_slp): Pass force_single_lane to
	vect_lower_load_permutations.
This commit is contained in:
Richard Biener 2024-11-12 10:31:30 +01:00 committed by Richard Biener
parent 1b35b92935
commit 0d4b254b20
1 changed files with 11 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
gcc/tree-vect-slp.cc
View File

@ -4402,7 +4402,8 @@ vllp_cmp (const void *a_, const void *b_)
static void static void
vect_lower_load_permutations (loop_vec_info loop_vinfo, vect_lower_load_permutations (loop_vec_info loop_vinfo,
scalar_stmts_to_slp_tree_map_t *bst_map, scalar_stmts_to_slp_tree_map_t *bst_map,
const array_slice<slp_tree> &loads) const array_slice<slp_tree> &loads,
bool force_single_lane)
{ {
/* We at this point want to lower without a fixed VF or vector /* We at this point want to lower without a fixed VF or vector
size in mind which means we cannot actually compute whether we size in mind which means we cannot actually compute whether we
@ -4494,7 +4495,8 @@ vect_lower_load_permutations (loop_vec_info loop_vinfo,
extracting it from the larger load. extracting it from the larger load.
??? Long-term some of the lowering should move to where ??? Long-term some of the lowering should move to where
the vector types involved are fixed. */ the vector types involved are fixed. */
if (ld_lanes_lanes == 0 if (!force_single_lane
&& ld_lanes_lanes == 0
&& contiguous && contiguous
&& (SLP_TREE_LANES (load) > 1 || loads.size () == 1) && (SLP_TREE_LANES (load) > 1 || loads.size () == 1)
&& pow2p_hwi (SLP_TREE_LANES (load)) && pow2p_hwi (SLP_TREE_LANES (load))
@ -4668,7 +4670,8 @@ vect_lower_load_permutations (loop_vec_info loop_vinfo,
static void static void
vect_lower_load_permutations (loop_vec_info loop_vinfo, vect_lower_load_permutations (loop_vec_info loop_vinfo,
scalar_stmts_to_slp_tree_map_t *bst_map) scalar_stmts_to_slp_tree_map_t *bst_map,
bool force_single_lane)
{ {
/* Gather and sort loads across all instances. */ /* Gather and sort loads across all instances. */
hash_set<slp_tree> visited; hash_set<slp_tree> visited;
@ -4696,14 +4699,16 @@ vect_lower_load_permutations (loop_vec_info loop_vinfo,
if (STMT_VINFO_GROUPED_ACCESS (a0)) if (STMT_VINFO_GROUPED_ACCESS (a0))
vect_lower_load_permutations (loop_vinfo, bst_map, vect_lower_load_permutations (loop_vinfo, bst_map,
make_array_slice (&loads[firsti], make_array_slice (&loads[firsti],
i - firsti)); i - firsti),
force_single_lane);
firsti = i; firsti = i;
} }
if (firsti < loads.length () if (firsti < loads.length ()
&& STMT_VINFO_GROUPED_ACCESS (SLP_TREE_SCALAR_STMTS (loads[firsti])[0])) && STMT_VINFO_GROUPED_ACCESS (SLP_TREE_SCALAR_STMTS (loads[firsti])[0]))
vect_lower_load_permutations (loop_vinfo, bst_map, vect_lower_load_permutations (loop_vinfo, bst_map,
make_array_slice (&loads[firsti], make_array_slice (&loads[firsti],
loads.length () - firsti)); loads.length () - firsti),
force_single_lane);
} }
/* Check if there are stmts in the loop can be vectorized using SLP. Build SLP /* Check if there are stmts in the loop can be vectorized using SLP. Build SLP
@ -5097,7 +5102,7 @@ vect_analyze_slp (vec_info *vinfo, unsigned max_tree_size,
like schemes. */ like schemes. */
if (loop_vec_info loop_vinfo = dyn_cast <loop_vec_info> (vinfo)) if (loop_vec_info loop_vinfo = dyn_cast <loop_vec_info> (vinfo))
{ {
vect_lower_load_permutations (loop_vinfo, bst_map); vect_lower_load_permutations (loop_vinfo, bst_map, force_single_lane);
if (dump_enabled_p ()) if (dump_enabled_p ())
{ {
dump_printf_loc (MSG_NOTE, vect_location, dump_printf_loc (MSG_NOTE, vect_location,