public inbox for gcc-cvs@sourceware.org
help / color / mirror / Atom feed
* [gcc r14-2071] GIMPLE_FOLD: Apply LEN_MASK_{LOAD, STORE} into GIMPLE_FOLD
@ 2023-06-25 6:00 Pan Li
0 siblings, 0 replies; only message in thread
From: Pan Li @ 2023-06-25 6:00 UTC (permalink / raw)
To: gcc-cvs
https://gcc.gnu.org/g:ef09afa4767c25e23d5d837ce68a5b7ebd9bad1d
commit r14-2071-gef09afa4767c25e23d5d837ce68a5b7ebd9bad1d
Author: Ju-Zhe Zhong <juzhe.zhong@rivai.ai>
Date: Fri Jun 23 21:48:27 2023 +0800
GIMPLE_FOLD: Apply LEN_MASK_{LOAD,STORE} into GIMPLE_FOLD
Hi, since we are going to have LEN_MASK_{LOAD,STORE} into loopVectorizer.
Currenly,
1. we can fold MASK_{LOAD,STORE} into MEM when mask is all ones.
2. we can fold LEN_{LOAD,STORE} into MEM when (len - bias) is VF.
Now, I think it makes sense that we can support
fold LEN_MASK_{LOAD,STORE} into MEM when both mask = all ones and (len - bias) is VF.
gcc/ChangeLog:
* gimple-fold.cc (arith_overflowed_p): Apply LEN_MASK_{LOAD,STORE}.
(gimple_fold_partial_load_store_mem_ref): Ditto.
(gimple_fold_partial_store): Ditto.
(gimple_fold_call): Ditto.
Diff:
---
gcc/gimple-fold.cc | 23 ++++++++++++++++++-----
1 file changed, 18 insertions(+), 5 deletions(-)
diff --git a/gcc/gimple-fold.cc b/gcc/gimple-fold.cc
index 55e80567708..3d46b76edeb 100644
--- a/gcc/gimple-fold.cc
+++ b/gcc/gimple-fold.cc
@@ -5370,10 +5370,10 @@ arith_overflowed_p (enum tree_code code, const_tree type,
return wi::min_precision (wres, sign) > TYPE_PRECISION (type);
}
-/* If IFN_{MASK,LEN}_LOAD/STORE call CALL is unconditional, return a MEM_REF
- for the memory it references, otherwise return null. VECTYPE is the
- type of the memory vector. MASK_P indicates it's for MASK if true,
- otherwise it's for LEN. */
+/* If IFN_{MASK,LEN,LEN_MASK}_LOAD/STORE call CALL is unconditional,
+ return a MEM_REF for the memory it references, otherwise return null.
+ VECTYPE is the type of the memory vector. MASK_P indicates it's for
+ MASK if true, otherwise it's for LEN. */
static tree
gimple_fold_partial_load_store_mem_ref (gcall *call, tree vectype, bool mask_p)
@@ -5400,6 +5400,16 @@ gimple_fold_partial_load_store_mem_ref (gcall *call, tree vectype, bool mask_p)
if (maybe_ne (wi::to_poly_widest (basic_len) - wi::to_widest (bias),
GET_MODE_SIZE (TYPE_MODE (vectype))))
return NULL_TREE;
+
+ /* For LEN_MASK_{LOAD,STORE}, we should also check whether
+ the mask is all ones mask. */
+ internal_fn ifn = gimple_call_internal_fn (call);
+ if (ifn == IFN_LEN_MASK_LOAD || ifn == IFN_LEN_MASK_STORE)
+ {
+ tree mask = gimple_call_arg (call, internal_fn_mask_index (ifn));
+ if (!integer_all_onesp (mask))
+ return NULL_TREE;
+ }
}
unsigned HOST_WIDE_INT align = tree_to_uhwi (alias_align);
@@ -5438,7 +5448,8 @@ static bool
gimple_fold_partial_store (gimple_stmt_iterator *gsi, gcall *call,
bool mask_p)
{
- tree rhs = gimple_call_arg (call, 3);
+ internal_fn ifn = gimple_call_internal_fn (call);
+ tree rhs = gimple_call_arg (call, internal_fn_stored_value_index (ifn));
if (tree lhs
= gimple_fold_partial_load_store_mem_ref (call, TREE_TYPE (rhs), mask_p))
{
@@ -5676,9 +5687,11 @@ gimple_fold_call (gimple_stmt_iterator *gsi, bool inplace)
changed |= gimple_fold_partial_store (gsi, stmt, true);
break;
case IFN_LEN_LOAD:
+ case IFN_LEN_MASK_LOAD:
changed |= gimple_fold_partial_load (gsi, stmt, false);
break;
case IFN_LEN_STORE:
+ case IFN_LEN_MASK_STORE:
changed |= gimple_fold_partial_store (gsi, stmt, false);
break;
default:
^ permalink raw reply [flat|nested] only message in thread
only message in thread, other threads:[~2023-06-25 6:00 UTC | newest]
Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2023-06-25 6:00 [gcc r14-2071] GIMPLE_FOLD: Apply LEN_MASK_{LOAD, STORE} into GIMPLE_FOLD Pan Li
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).