From: Richard Sandiford <richard.sandiford@arm.com>
To: Richard Biener <richard.guenther@gmail.com>
Cc: GCC Patches <gcc-patches@gcc.gnu.org>
Subject: Re: [36/46] Add a pattern_stmt_p field to stmt_vec_info
Date: Wed, 25 Jul 2018 11:09:00 -0000 [thread overview]
Message-ID: <87in53k1vm.fsf@arm.com> (raw)
In-Reply-To: <CAFiYyc23rzC4WGaf_SDVBnWQi_VRj1pe=30B9W5n1-2k9PqFOg@mail.gmail.com> (Richard Biener's message of "Wed, 25 Jul 2018 12:15:25 +0200")
Richard Biener <richard.guenther@gmail.com> writes:
> On Tue, Jul 24, 2018 at 12:07 PM Richard Sandiford
> <richard.sandiford@arm.com> wrote:
>>
>> This patch adds a pattern_stmt_p field to stmt_vec_info, so that it's
>> possible to tell whether the statement is a pattern statement without
>> referring to other statements. The new field goes in what was
>> previously a hole in the structure, so the size is the same as before.
>
> Not sure what the advantage is? is_pattern_stmt_p () looks nicer
> than ->is_pattern_p
I can keep the function wrapper if you prefer that. But having a
statement "know" whether it's a pattern stmt makes things like
freeing stmt_vec_infos simpler (see later patches in the series).
It should also be cheaper to test, but that's much more minor.
Thanks,
Richard
>
>>
>> 2018-07-24 Richard Sandiford <richard.sandiford@arm.com>
>>
>> gcc/
>> * tree-vectorizer.h (_stmt_vec_info::pattern_stmt_p): New field.
>> (is_pattern_stmt_p): Delete.
>> * tree-vect-patterns.c (vect_init_pattern_stmt): Set pattern_stmt_p
>> on pattern statements.
>> (vect_split_statement, vect_mark_pattern_stmts): Use the new
>> pattern_stmt_p field instead of is_pattern_stmt_p.
>> * tree-vect-data-refs.c (vect_preserves_scalar_order_p): Likewise.
>> * tree-vect-loop.c (vectorizable_live_operation): Likewise.
>> * tree-vect-slp.c (vect_build_slp_tree_2): Likewise.
>> (vect_find_last_scalar_stmt_in_slp, vect_remove_slp_scalar_calls)
>> (vect_schedule_slp): Likewise.
>> * tree-vect-stmts.c (vect_mark_stmts_to_be_vectorized): Likewise.
>> (vectorizable_call, vectorizable_simd_clone_call, vectorizable_shift)
>> (vectorizable_store, vect_remove_stores): Likewise.
>>
>> Index: gcc/tree-vectorizer.h
>> ===================================================================
>> --- gcc/tree-vectorizer.h 2018-07-24 10:23:56.440544995 +0100
>> +++ gcc/tree-vectorizer.h 2018-07-24 10:24:02.364492386 +0100
>> @@ -791,6 +791,12 @@ struct _stmt_vec_info {
>> /* Stmt is part of some pattern (computation idiom) */
>> bool in_pattern_p;
>>
>> + /* True if the statement was created during pattern recognition as
>> + part of the replacement for RELATED_STMT. This implies that the
>> + statement isn't part of any basic block, although for convenience
>> + its gimple_bb is the same as for RELATED_STMT. */
>> + bool pattern_stmt_p;
>> +
>> /* Is this statement vectorizable or should it be skipped in (partial)
>> vectorization. */
>> bool vectorizable;
>> @@ -1151,16 +1157,6 @@ get_later_stmt (stmt_vec_info stmt1_info
>> return stmt2_info;
>> }
>>
>> -/* Return TRUE if a statement represented by STMT_INFO is a part of a
>> - pattern. */
>> -
>> -static inline bool
>> -is_pattern_stmt_p (stmt_vec_info stmt_info)
>> -{
>> - stmt_vec_info related_stmt_info = STMT_VINFO_RELATED_STMT (stmt_info);
>> - return related_stmt_info && STMT_VINFO_IN_PATTERN_P (related_stmt_info);
>> -}
>> -
>> /* Return true if BB is a loop header. */
>>
>> static inline bool
>> Index: gcc/tree-vect-patterns.c
>> ===================================================================
>> --- gcc/tree-vect-patterns.c 2018-07-24 10:23:59.408518638 +0100
>> +++ gcc/tree-vect-patterns.c 2018-07-24 10:24:02.360492422 +0100
>> @@ -108,6 +108,7 @@ vect_init_pattern_stmt (gimple *pattern_
>> pattern_stmt_info = orig_stmt_info->vinfo->add_stmt (pattern_stmt);
>> gimple_set_bb (pattern_stmt, gimple_bb (orig_stmt_info->stmt));
>>
>> + pattern_stmt_info->pattern_stmt_p = true;
>> STMT_VINFO_RELATED_STMT (pattern_stmt_info) = orig_stmt_info;
>> STMT_VINFO_DEF_TYPE (pattern_stmt_info)
>> = STMT_VINFO_DEF_TYPE (orig_stmt_info);
>> @@ -630,7 +631,7 @@ vect_recog_temp_ssa_var (tree type, gimp
>> vect_split_statement (stmt_vec_info stmt2_info, tree new_rhs,
>> gimple *stmt1, tree vectype)
>> {
>> - if (is_pattern_stmt_p (stmt2_info))
>> + if (stmt2_info->pattern_stmt_p)
>> {
>> /* STMT2_INFO is part of a pattern. Get the statement to which
>> the pattern is attached. */
>> @@ -4726,7 +4727,7 @@ vect_mark_pattern_stmts (stmt_vec_info o
>> gimple *def_seq = STMT_VINFO_PATTERN_DEF_SEQ (orig_stmt_info);
>>
>> gimple *orig_pattern_stmt = NULL;
>> - if (is_pattern_stmt_p (orig_stmt_info))
>> + if (orig_stmt_info->pattern_stmt_p)
>> {
>> /* We're replacing a statement in an existing pattern definition
>> sequence. */
>> Index: gcc/tree-vect-data-refs.c
>> ===================================================================
>> --- gcc/tree-vect-data-refs.c 2018-07-24 10:23:53.204573732 +0100
>> +++ gcc/tree-vect-data-refs.c 2018-07-24 10:24:02.356492457 +0100
>> @@ -212,9 +212,9 @@ vect_preserves_scalar_order_p (stmt_vec_
>> (but could happen later) while reads will happen no later than their
>> current position (but could happen earlier). Reordering is therefore
>> only possible if the first access is a write. */
>> - if (is_pattern_stmt_p (stmtinfo_a))
>> + if (stmtinfo_a->pattern_stmt_p)
>> stmtinfo_a = STMT_VINFO_RELATED_STMT (stmtinfo_a);
>> - if (is_pattern_stmt_p (stmtinfo_b))
>> + if (stmtinfo_b->pattern_stmt_p)
>> stmtinfo_b = STMT_VINFO_RELATED_STMT (stmtinfo_b);
>> stmt_vec_info earlier_stmt_info = get_earlier_stmt (stmtinfo_a,
> stmtinfo_b);
>> return !DR_IS_WRITE (STMT_VINFO_DATA_REF (earlier_stmt_info));
>> Index: gcc/tree-vect-loop.c
>> ===================================================================
>> --- gcc/tree-vect-loop.c 2018-07-24 10:23:56.436545030 +0100
>> +++ gcc/tree-vect-loop.c 2018-07-24 10:24:02.360492422 +0100
>> @@ -7907,7 +7907,7 @@ vectorizable_live_operation (stmt_vec_in
>> }
>>
>> /* If stmt has a related stmt, then use that for getting the lhs. */
>> - gimple *stmt = (is_pattern_stmt_p (stmt_info)
>> + gimple *stmt = (stmt_info->pattern_stmt_p
>> ? STMT_VINFO_RELATED_STMT (stmt_info)->stmt
>> : stmt_info->stmt);
>>
>> Index: gcc/tree-vect-slp.c
>> ===================================================================
>> --- gcc/tree-vect-slp.c 2018-07-24 10:23:53.204573732 +0100
>> +++ gcc/tree-vect-slp.c 2018-07-24 10:24:02.360492422 +0100
>> @@ -376,7 +376,7 @@ vect_get_and_check_slp_defs (vec_info *v
>> /* Check if DEF_STMT_INFO is a part of a pattern in LOOP and get
>> the def stmt from the pattern. Check that all the stmts of the
>> node are in the pattern. */
>> - if (def_stmt_info && is_pattern_stmt_p (def_stmt_info))
>> + if (def_stmt_info && def_stmt_info->pattern_stmt_p)
>> {
>> pattern = true;
>> if (!first && !oprnd_info->first_pattern
>> @@ -1315,7 +1315,7 @@ vect_build_slp_tree_2 (vec_info *vinfo,
>> /* ??? Rejecting patterns this way doesn't work. We'd have to
>> do extra work to cancel the pattern so the uses see the
>> scalar version. */
>> - && !is_pattern_stmt_p (SLP_TREE_SCALAR_STMTS (child)[0]))
>> + && !SLP_TREE_SCALAR_STMTS (child)[0]->pattern_stmt_p)
>> {
>> slp_tree grandchild;
>>
>> @@ -1359,7 +1359,7 @@ vect_build_slp_tree_2 (vec_info *vinfo,
>> /* ??? Rejecting patterns this way doesn't work. We'd have to
>> do extra work to cancel the pattern so the uses see the
>> scalar version. */
>> - && !is_pattern_stmt_p (stmt_info))
>> + && !stmt_info->pattern_stmt_p)
>> {
>> dump_printf_loc (MSG_NOTE, vect_location,
>> "Building vector operands from scalars\n");
>> @@ -1486,7 +1486,7 @@ vect_build_slp_tree_2 (vec_info *vinfo,
>> /* ??? Rejecting patterns this way doesn't work. We'd have
>> to do extra work to cancel the pattern so the uses see the
>> scalar version. */
>> - && !is_pattern_stmt_p (SLP_TREE_SCALAR_STMTS (child)[0]))
>> + && !SLP_TREE_SCALAR_STMTS (child)[0]->pattern_stmt_p)
>> {
>> unsigned int j;
>> slp_tree grandchild;
>> @@ -1848,7 +1848,7 @@ vect_find_last_scalar_stmt_in_slp (slp_t
>>
>> for (int i = 0; SLP_TREE_SCALAR_STMTS (node).iterate (i, &stmt_vinfo); i++)
>> {
>> - if (is_pattern_stmt_p (stmt_vinfo))
>> + if (stmt_vinfo->pattern_stmt_p)
>> stmt_vinfo = STMT_VINFO_RELATED_STMT (stmt_vinfo);
>> last = last ? get_later_stmt (stmt_vinfo, last) : stmt_vinfo;
>> }
>> @@ -4044,8 +4044,7 @@ vect_remove_slp_scalar_calls (slp_tree n
>> gcall *stmt = dyn_cast <gcall *> (stmt_info->stmt);
>> if (!stmt || gimple_bb (stmt) == NULL)
>> continue;
>> - if (is_pattern_stmt_p (stmt_info)
>> - || !PURE_SLP_STMT (stmt_info))
>> + if (stmt_info->pattern_stmt_p || !PURE_SLP_STMT (stmt_info))
>> continue;
>> lhs = gimple_call_lhs (stmt);
>> new_stmt = gimple_build_assign (lhs, build_zero_cst (TREE_TYPE (lhs)));
>> @@ -4106,7 +4105,7 @@ vect_schedule_slp (vec_info *vinfo)
>> if (!STMT_VINFO_DATA_REF (store_info))
>> break;
>>
>> - if (is_pattern_stmt_p (store_info))
>> + if (store_info->pattern_stmt_p)
>> store_info = STMT_VINFO_RELATED_STMT (store_info);
>> /* Free the attached stmt_vec_info and remove the stmt. */
>> gsi = gsi_for_stmt (store_info);
>> Index: gcc/tree-vect-stmts.c
>> ===================================================================
>> --- gcc/tree-vect-stmts.c 2018-07-24 10:23:56.440544995 +0100
>> +++ gcc/tree-vect-stmts.c 2018-07-24 10:24:02.364492386 +0100
>> @@ -731,7 +731,7 @@ vect_mark_stmts_to_be_vectorized (loop_v
>> break;
>> }
>>
>> - if (is_pattern_stmt_p (stmt_vinfo))
>> + if (stmt_vinfo->pattern_stmt_p)
>> {
>> /* Pattern statements are not inserted into the code, so
>> FOR_EACH_PHI_OR_STMT_USE optimizes their operands out, and we
>> @@ -3623,7 +3623,7 @@ vectorizable_call (stmt_vec_info stmt_in
>> if (slp_node)
>> return true;
>>
>> - if (is_pattern_stmt_p (stmt_info))
>> + if (stmt_info->pattern_stmt_p)
>> stmt_info = STMT_VINFO_RELATED_STMT (stmt_info);
>> lhs = gimple_get_lhs (stmt_info->stmt);
>>
>> @@ -4362,7 +4362,7 @@ vectorizable_simd_clone_call (stmt_vec_i
>> if (scalar_dest)
>> {
>> type = TREE_TYPE (scalar_dest);
>> - if (is_pattern_stmt_p (stmt_info))
>> + if (stmt_info->pattern_stmt_p)
>> lhs = gimple_call_lhs (STMT_VINFO_RELATED_STMT (stmt_info)->stmt);
>> else
>> lhs = gimple_call_lhs (stmt);
>> @@ -5552,7 +5552,7 @@ vectorizable_shift (stmt_vec_info stmt_i
>> /* If the shift amount is computed by a pattern stmt we cannot
>> use the scalar amount directly thus give up and use a vector
>> shift. */
>> - if (op1_def_stmt_info && is_pattern_stmt_p (op1_def_stmt_info))
>> + if (op1_def_stmt_info && op1_def_stmt_info->pattern_stmt_p)
>> scalar_shift_arg = false;
>> }
>> else
>> @@ -6286,7 +6286,7 @@ vectorizable_store (stmt_vec_info stmt_i
>> {
>> tree scalar_dest = gimple_assign_lhs (assign);
>> if (TREE_CODE (scalar_dest) == VIEW_CONVERT_EXPR
>> - && is_pattern_stmt_p (stmt_info))
>> + && stmt_info->pattern_stmt_p)
>> scalar_dest = TREE_OPERAND (scalar_dest, 0);
>> if (TREE_CODE (scalar_dest) != ARRAY_REF
>> && TREE_CODE (scalar_dest) != BIT_FIELD_REF
>> @@ -9839,7 +9839,7 @@ vect_remove_stores (stmt_vec_info first_
>> while (next_stmt_info)
>> {
>> stmt_vec_info tmp = DR_GROUP_NEXT_ELEMENT (next_stmt_info);
>> - if (is_pattern_stmt_p (next_stmt_info))
>> + if (next_stmt_info->pattern_stmt_p)
>> next_stmt_info = STMT_VINFO_RELATED_STMT (next_stmt_info);
>> /* Free the attached stmt_vec_info and remove the stmt. */
>> next_si = gsi_for_stmt (next_stmt_info->stmt);
next prev parent reply other threads:[~2018-07-25 11:09 UTC|newest]
Thread overview: 108+ messages / expand[flat|nested] mbox.gz Atom feed top
2018-07-24 9:52 [00/46] Remove vinfo_for_stmt etc Richard Sandiford
2018-07-24 9:52 ` [01/46] Move special cases out of get_initial_def_for_reduction Richard Sandiford
2018-07-25 8:42 ` Richard Biener
2018-07-24 9:53 ` [03/46] Remove unnecessary update of NUM_SLP_USES Richard Sandiford
2018-07-25 8:46 ` Richard Biener
2018-07-24 9:53 ` [02/46] Remove dead vectorizable_reduction code Richard Sandiford
2018-07-25 8:43 ` Richard Biener
2018-07-24 9:54 ` [05/46] Fix make_ssa_name call in vectorizable_reduction Richard Sandiford
2018-07-25 8:47 ` Richard Biener
2018-07-24 9:54 ` [04/46] Factor out the test for a valid reduction input Richard Sandiford
2018-07-25 8:46 ` Richard Biener
2018-07-24 9:55 ` [08/46] Add vec_info::lookup_def Richard Sandiford
2018-07-25 9:12 ` Richard Biener
2018-07-24 9:55 ` [06/46] Add vec_info::add_stmt Richard Sandiford
2018-07-25 9:10 ` Richard Biener
2018-07-24 9:55 ` [07/46] Add vec_info::lookup_stmt Richard Sandiford
2018-07-25 9:11 ` Richard Biener
2018-07-24 9:56 ` [09/46] Add vec_info::lookup_single_use Richard Sandiford
2018-07-25 9:13 ` Richard Biener
2018-07-24 9:57 ` [11/46] Pass back a stmt_vec_info from vect_is_simple_use Richard Sandiford
2018-07-25 9:18 ` Richard Biener
2018-07-24 9:57 ` [10/46] Temporarily make stmt_vec_info a class Richard Sandiford
2018-07-25 9:14 ` Richard Biener
2018-07-24 9:58 ` [12/46] Make vect_finish_stmt_generation return a stmt_vec_info Richard Sandiford
2018-07-25 9:19 ` Richard Biener
2018-07-24 9:58 ` [13/46] Make STMT_VINFO_RELATED_STMT " Richard Sandiford
2018-07-25 9:19 ` Richard Biener
2018-07-24 9:58 ` [14/46] Make STMT_VINFO_VEC_STMT " Richard Sandiford
2018-07-25 9:21 ` Richard Biener
2018-07-25 11:03 ` Richard Sandiford
2018-08-02 0:22 ` H.J. Lu
2018-08-02 9:58 ` Richard Sandiford
2018-07-24 9:59 ` [16/46] Make STMT_VINFO_REDUC_DEF " Richard Sandiford
2018-07-25 9:22 ` Richard Biener
2018-07-24 9:59 ` [17/46] Make LOOP_VINFO_REDUCTIONS an auto_vec<stmt_vec_info> Richard Sandiford
2018-07-25 9:23 ` Richard Biener
2018-07-24 9:59 ` [15/46] Make SLP_TREE_VEC_STMTS a vec<stmt_vec_info> Richard Sandiford
2018-07-25 9:22 ` Richard Biener
2018-07-24 10:00 ` [18/46] Make SLP_TREE_SCALAR_STMTS " Richard Sandiford
2018-07-25 9:27 ` Richard Biener
2018-07-31 15:03 ` Richard Sandiford
2018-07-24 10:01 ` [19/46] Make vect_dr_stmt return a stmt_vec_info Richard Sandiford
2018-07-25 9:28 ` Richard Biener
2018-07-24 10:01 ` [20/46] Make *FIRST_ELEMENT and *NEXT_ELEMENT stmt_vec_infos Richard Sandiford
2018-07-25 9:28 ` Richard Biener
2018-07-24 10:01 ` [21/46] Make grouped_stores and reduction_chains use stmt_vec_infos Richard Sandiford
2018-07-25 9:28 ` Richard Biener
2018-07-24 10:02 ` [22/46] Make DR_GROUP_SAME_DR_STMT a stmt_vec_info Richard Sandiford
2018-07-25 9:29 ` Richard Biener
2018-07-24 10:02 ` [24/46] Make stmt_info_for_cost use " Richard Sandiford
2018-07-25 9:30 ` Richard Biener
2018-07-24 10:02 ` [23/46] Make LOOP_VINFO_MAY_MISALIGN_STMTS use stmt_vec_info Richard Sandiford
2018-07-25 9:29 ` Richard Biener
2018-07-24 10:03 ` [25/46] Make get_earlier/later_stmt take and return stmt_vec_infos Richard Sandiford
2018-07-25 9:31 ` Richard Biener
2018-07-24 10:03 ` [26/46] Make more use of dyn_cast in tree-vect* Richard Sandiford
2018-07-25 9:31 ` Richard Biener
2018-07-24 10:03 ` [27/46] Remove duplicated stmt_vec_info lookups Richard Sandiford
2018-07-25 9:32 ` Richard Biener
2018-07-24 10:04 ` [29/46] Use stmt_vec_info instead of gimple stmts internally (part 2) Richard Sandiford
2018-07-25 10:03 ` Richard Biener
2018-07-24 10:04 ` [28/46] Use stmt_vec_info instead of gimple stmts internally (part 1) Richard Sandiford
2018-07-25 9:33 ` Richard Biener
2018-07-24 10:04 ` [30/46] Use stmt_vec_infos rather than gimple stmts for worklists Richard Sandiford
2018-07-25 10:04 ` Richard Biener
2018-07-24 10:05 ` [32/46] Use stmt_vec_info in function interfaces (part 2) Richard Sandiford
2018-07-25 10:06 ` Richard Biener
2018-07-24 10:05 ` [31/46] Use stmt_vec_info in function interfaces (part 1) Richard Sandiford
2018-07-25 10:05 ` Richard Biener
2018-07-24 10:06 ` [34/46] Alter interface to vect_get_vec_def_for_stmt_copy Richard Sandiford
2018-07-25 10:13 ` Richard Biener
2018-07-24 10:06 ` [33/46] Use stmt_vec_infos instead of vec_info/gimple stmt pairs Richard Sandiford
2018-07-25 10:06 ` Richard Biener
2018-07-24 10:06 ` [35/46] Alter interfaces within vect_pattern_recog Richard Sandiford
2018-07-25 10:14 ` Richard Biener
2018-07-24 10:07 ` [36/46] Add a pattern_stmt_p field to stmt_vec_info Richard Sandiford
2018-07-25 10:15 ` Richard Biener
2018-07-25 11:09 ` Richard Sandiford [this message]
2018-07-25 11:48 ` Richard Biener
2018-07-26 10:29 ` Richard Sandiford
2018-07-26 11:15 ` Richard Biener
2018-07-24 10:07 ` [37/46] Associate alignment information with stmt_vec_infos Richard Sandiford
2018-07-25 10:18 ` Richard Biener
2018-07-26 10:55 ` Richard Sandiford
2018-07-26 11:13 ` Richard Biener
2018-07-24 10:08 ` [39/46] Replace STMT_VINFO_UNALIGNED_DR with the associated statement Richard Sandiford
2018-07-26 11:08 ` [39/46 v2] Change STMT_VINFO_UNALIGNED_DR to a dr_vec_info Richard Sandiford
2018-07-26 11:13 ` Richard Biener
2018-07-24 10:08 ` [38/46] Pass stmt_vec_infos instead of data_references where relevant Richard Sandiford
2018-07-25 10:21 ` Richard Biener
2018-07-25 11:21 ` Richard Sandiford
2018-07-26 11:05 ` Richard Sandiford
2018-07-26 11:13 ` Richard Biener
2018-07-24 10:09 ` [40/46] Add vec_info::lookup_dr Richard Sandiford
2018-07-26 11:10 ` [40/46 v2] " Richard Sandiford
2018-07-26 11:16 ` Richard Biener
2018-07-24 10:09 ` [42/46] Add vec_info::replace_stmt Richard Sandiford
2018-07-31 12:03 ` Richard Biener
2018-07-24 10:09 ` [41/46] Add vec_info::remove_stmt Richard Sandiford
2018-07-31 12:02 ` Richard Biener
2018-07-24 10:10 ` [43/46] Make free_stmt_vec_info take a stmt_vec_info Richard Sandiford
2018-07-31 12:03 ` Richard Biener
2018-07-24 10:10 ` [45/46] Remove vect_stmt_in_region_p Richard Sandiford
2018-07-31 12:06 ` Richard Biener
2018-07-24 10:10 ` [44/46] Remove global vinfo_for_stmt-related routines Richard Sandiford
2018-07-31 12:05 ` Richard Biener
2018-07-24 10:11 ` [46/46] Turn stmt_vec_info back into a typedef Richard Sandiford
2018-07-31 12:07 ` Richard Biener
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=87in53k1vm.fsf@arm.com \
--to=richard.sandiford@arm.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=richard.guenther@gmail.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
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).