Re: [PATCH v6 3/3] PR80791 Consider doloop cmp use in ivopts
on 2019/9/12 下午4:14, Richard Biener wrote: > On Wed, 11 Sep 2019, Kewen.Lin wrote: > >> Hi, >> >> Sorry for the late update. I've updated the words of target hooks part. >> >> Could someone help to review it? Thanks in advance! >> >> By the way, as previous emails in this thread, Bin has approved the IVOPTs >> part, while Segher has approved the rs6000 part. > > The target hooks part is OK. I guess we'll have to extend it eventually > in case other targets want to make use of it. > Thanks Richard! Committed by r275713. Yes, it's enough when doloop IV costs zero or infinite for generic/address use, but if one target wants some other values, we may have to take it as one common cost shared for all generic/address uses. It's like IV candidate cost but not the same since it's only needed when doloop IV is used for generic/address uses, I guess it requires some changes in candidate set cost calculation. I chose to keep it simple at the first place, but radar on for any other target adoptions. Thanks, Kewen > Thanks, > Richard. > >> >> Thanks, >> Kewen >> >> - >> >> gcc/ChangeLog >> >> 2019-09-11 Kewen Lin >> >> PR middle-end/80791 >> * config/rs6000/rs6000.c (TARGET_HAVE_COUNT_REG_DECR_P): New macro. >> (TARGET_DOLOOP_COST_FOR_GENERIC): Likewise. >> (TARGET_DOLOOP_COST_FOR_ADDRESS): Likewise. >> * target.def (have_count_reg_decr_p): New hook. >> (doloop_cost_for_generic): Likewise. >> (doloop_cost_for_address): Likewise. >> * doc/tm.texi.in (TARGET_HAVE_COUNT_REG_DECR_P): Likewise. >> (TARGET_DOLOOP_COST_FOR_GENERIC): Likewise. >> (TARGET_DOLOOP_COST_FOR_ADDRESS): Likewise. >> * doc/tm.texi: Regenerate. >> * tree-ssa-loop-ivopts.c (comp_cost::operator+=): Consider infinite cost >> addend. >> (record_group): Init doloop_p. >> (add_candidate_1): Add optional argument doloop, change the handlings >> accordingly. >> (add_candidate): Likewise. >> (generic_predict_doloop_p): Update attribute. >> (force_expr_to_var_cost): Add costing for expressions COND_EXPR/LT_EXPR/ >> LE_EXPR/GT_EXPR/GE_EXPR/EQ_EXPR/NE_EXPR/UNORDERED_EXPR/ORDERED_EXPR/ >> UNLT_EXPR/UNLE_EXPR/UNGT_EXPR/UNGE_EXPR/UNEQ_EXPR/LTGT_EXPR/MAX_EXPR/ >> MIN_EXPR. >> (get_computation_cost): Update for doloop IV cand extra cost. >> (determine_group_iv_cost_cond): Update for doloop IV cand. >> (determine_iv_cost): Likewise. >> (ivopts_estimate_reg_pressure): Likewise. >> (may_eliminate_iv): Update handlings for doloop IV cand. >> (add_iv_candidate_for_doloop): New function. >> (find_iv_candidates): Call function add_iv_candidate_for_doloop. >> (iv_ca_set_no_cp): Update for doloop IV cand. >> (iv_ca_set_cp): Likewise. >> (iv_ca_dump): Dump register cost. >> (find_doloop_use): New function. >> (analyze_and_mark_doloop_use): Likewise. >> (tree_ssa_iv_optimize_loop): Call function analyze_and_mark_doloop_use. >> >> gcc/testsuite/ChangeLog >> >> 2019-09-11 Kewen Lin >> >> PR middle-end/80791 >> * gcc.dg/tree-ssa/ivopts-3.c: Adjust for doloop change. >> * gcc.dg/tree-ssa/ivopts-lt.c: Likewise. >> * gcc.dg/tree-ssa/pr32044.c: Likewise. >> >> >> on 2019/8/23 下午6:18, Segher Boessenkool wrote: >>> Hi! >>> >>> On Fri, Aug 23, 2019 at 05:43:32PM +0800, Bin.Cheng wrote: On Fri, Aug 23, 2019 at 4:27 PM Kewen.Lin wrote: Not sure if non-ivopts parts are already approved? If so, the patch is okay with above issues addressed. >>> >>> The rs6000 part is fine. The target.def entries need some spell check >>> and copy-editing, but are obvious and trivial otherwise, and/or you can >>> approve it as ivopts maintainer. >>> Thanks very much for your time! >>> >>> And thank you as well Bin :-) >>> >>> >>> Segher >>> >> >
[patch,committed][og9][openacc-gcc-9-branch] Fix segfault with plugin-gcn under libgomp
Hi all,
exactly the same as for plugin-hsa on the trunk:
https://gcc.gnu.org/ml/gcc-patches/2019-09/msg00718.html
Only a different file (which is only in this branch).
Tobias
commit 392044a8db285d9aea0a280983ce7c5014a4e99c
Author: Tobias Burnus
Date: Thu Sep 12 18:07:53 2019 +0200
libgomp plugin-gcn - init string
libgomp/
2019-09-13 Tobias Burnus
* plugin/plugin-gcn.c (hsa_warn, hsa_fatal, hsa_error): Ensure
string is initialized.
diff --git a/libgomp/ChangeLog.openacc b/libgomp/ChangeLog.openacc
index 355e406d4e3..14ed4e0ec2c 100644
--- a/libgomp/ChangeLog.openacc
+++ b/libgomp/ChangeLog.openacc
@@ -1,3 +1,8 @@
+2019-09-13 Tobias Burnus
+
+ * plugin/plugin-gcn.c (hsa_warn, hsa_fatal, hsa_error): Ensure
+ string is initialized.
+
2019-09-10 Julian Brown
* plugin/plugin-gcn.c (GOMP_hsa_kernel_dispatch): Remove
diff --git a/libgomp/plugin/plugin-gcn.c b/libgomp/plugin/plugin-gcn.c
index f7e3554f297..b8ec96391f7 100644
--- a/libgomp/plugin/plugin-gcn.c
+++ b/libgomp/plugin/plugin-gcn.c
@@ -489,7 +489,7 @@ hsa_warn (const char *str, hsa_status_t status)
if (!debug)
return;
- const char *hsa_error_msg;
+ const char *hsa_error_msg = "[unknown]";
hsa_fns.hsa_status_string_fn (status, &hsa_error_msg);
fprintf (stderr, "GCN warning: %s\nRuntime message: %s\n", str,
@@ -502,7 +502,7 @@ hsa_warn (const char *str, hsa_status_t status)
static void
hsa_fatal (const char *str, hsa_status_t status)
{
- const char *hsa_error_msg;
+ const char *hsa_error_msg = "[unknown]";
hsa_fns.hsa_status_string_fn (status, &hsa_error_msg);
GOMP_PLUGIN_fatal ("GCN fatal error: %s\nRuntime message: %s\n", str,
hsa_error_msg);
@@ -514,7 +514,7 @@ hsa_fatal (const char *str, hsa_status_t status)
static bool
hsa_error (const char *str, hsa_status_t status)
{
- const char *hsa_error_msg;
+ const char *hsa_error_msg = "[unknown]";
hsa_fns.hsa_status_string_fn (status, &hsa_error_msg);
GOMP_PLUGIN_error ("GCN fatal error: %s\nRuntime message: %s\n", str,
hsa_error_msg);
[patch, fortran] Improve error messages for mismatched arguments
Hello world,
the attached patch improves the rather hard to read error
messages for argument mismatches. With this patch, this reads
argument_checking_21.f90:7:11:
6 | call foo(1.0) ! { dg-warning "Rank mismatch" }
| 2
7 | call foo(b) ! { dg-warning "Rank mismatch" }
| 1
Fehler: Rank mismatch between actual argument at (1) and actual argument
at (2) (scalar and rank-2)
which I think is fairly clear. It also makes sure that warnings are
always emitted by -fallow-argument-mismatch by removing
-Wargument-mismatch. Finally, for people who do not want to have too
many warnings cluttering up their logs, a type mismatch is only
shown once if it is a warning.
While I was going on about fixing warnings, I also fixed PR 91557 with
the bit in trans-expr.c. This part is trivial, I will backport it
to the other affected branches.
After this, I think we can close PR 91556. Regression-tested. OK for
trunk?
2019-09-13 Thomas Koenig
PR fortran/91557
PR fortran/91556
* frontend-passes.c (check_externals_procedure): Reformat argument
list. Use gfc_compare_actual_formal instead of gfc_procedure_use.
* gfortran.h (gfc_symbol): Add flag error.
* interface.c (gfc_compare_interfaces): Reformat.
(argument_rank_mismatch): Add where_formal argument. If it is
present, note that the error is between different calls.
(compare_parameter): Change warnings that previously dependended
on -Wargument-mismatch to unconditional. Issue an error / warning
on type mismatch only once. Pass where_formal to
argument_rank_mismatch for artificial variables.
(compare_actual_formal): Change warnings that previously
dependeded on -Wargument-mismatch to unconditional.
(gfc_check_typebound_override): Likewise.
(gfc_get_formal_from_actual_arglist): Set declared_at for
artificial symbol.
* invoke.texi: Extend description of -fallow-argument-mismatch.
Delete -Wargument-mismatch.
* lang.opt: Change -Wargument-mismatch to do-nothing option.
* resolve.c (resolve_structure_cons): Change warnings that
previously depended on -Wargument-mismatch to unconditional.
* trans-decl.c (generate_local_decl): Do not warn if the symbol is
artificial.
2019-09-13 Thomas Koenig
PR fortran/91557
PR fortran/91556
* gfortran.dg/argument_checking_20.f90: New test.
* gfortran.dg/argument_checking_21.f90: New test.
* gfortran.dg/argument_checking_22.f90: New test.
* gfortran.dg/argument_checking_23.f90: New test.
* gfortran.dg/warn_unused_dummy_argument_5.f90: New test.
* gfortran.dg/bessel_3.f90: Add pattern for type mismatch.
* gfortran.dg/g77/20010519-1.f: Adjust dg-warning messages to new
handling.
* gfortran.dg/pr24823.f: Likewise.
* gfortran.dg/pr39937.f: Likewise.
Index: fortran/frontend-passes.c
===
--- fortran/frontend-passes.c (Revision 275713)
+++ fortran/frontend-passes.c (Arbeitskopie)
@@ -5373,7 +5373,8 @@ gfc_code_walker (gfc_code **c, walk_code_fn_t code
/* Common tests for argument checking for both functions and subroutines. */
static int
-check_externals_procedure (gfc_symbol *sym, locus *loc, gfc_actual_arglist *actual)
+check_externals_procedure (gfc_symbol *sym, locus *loc,
+ gfc_actual_arglist *actual)
{
gfc_gsymbol *gsym;
gfc_symbol *def_sym = NULL;
@@ -5396,7 +5397,7 @@ static int
if (def_sym)
{
- gfc_procedure_use (def_sym, &actual, loc);
+ gfc_compare_actual_formal (&actual, def_sym->formal, 0, 0, 0, loc);
return 0;
}
Index: fortran/gfortran.h
===
--- fortran/gfortran.h (Revision 275713)
+++ fortran/gfortran.h (Arbeitskopie)
@@ -1610,6 +1610,9 @@ typedef struct gfc_symbol
/* Set if this is a module function or subroutine with the
abreviated declaration in a submodule. */
unsigned abr_modproc_decl:1;
+ /* Set if a previous error or warning has occurred and no other
+ should be reported. */
+ unsigned error:1;
int refs;
struct gfc_namespace *ns; /* namespace containing this symbol */
Index: fortran/interface.c
===
--- fortran/interface.c (Revision 275713)
+++ fortran/interface.c (Arbeitskopie)
@@ -1807,9 +1807,9 @@ gfc_compare_interfaces (gfc_symbol *s1, gfc_symbol
if (!compare_rank (f2->sym, f1->sym))
{
if (errmsg != NULL)
- snprintf (errmsg, err_len, "Rank mismatch in argument '%s' "
- "(%i/%i)", f1->sym->name, symbol_rank (f1->sym),
- symbol_rank (f2->sym));
+ snprintf (errmsg, err_len, "Rank mismatch in argument "
+ "'%s' (%i/%i)", f1->sym->name,
+ symbol_rank (f1->sym), symbol_rank (f2->sym)
[PR target/85401] initialize the move cost table before using it
This seems to be the way the rest of ira-color.c does it. I hope it's OK. It does fix the segfault. 2019-09-10 Maya Rashish PR target/85401 * ira-color.c: (allocno_copy_cost_saving) Call ira_init_register_move_cost_if_necessary diff --git a/gcc/ira-color.c b/gcc/ira-color.c index 99236994d64..5d721102e19 100644 --- a/gcc/ira-color.c +++ b/gcc/ira-color.c @@ -2828,6 +2828,7 @@ allocno_copy_cost_saving (ira_allocno_t allocno, int hard_regno) } else gcc_unreachable (); + ira_init_register_move_cost_if_necessary(allocno_mode); cost += cp->freq * ira_register_move_cost[allocno_mode][rclass][rclass]; } return cost;
Re: [patch, fortran] Improve error messages for mismatched arguments
On Sat, Sep 14, 2019 at 02:27:15PM +0200, Thomas Koenig wrote:
>
> the attached patch improves the rather hard to read error
> messages for argument mismatches. With this patch, this reads
>
> argument_checking_21.f90:7:11:
>
> 6 | call foo(1.0) ! { dg-warning "Rank mismatch" }
>| 2
> 7 | call foo(b) ! { dg-warning "Rank mismatch" }
>| 1
> Fehler: Rank mismatch between actual argument at (1) and actual argument
> at (2) (scalar and rank-2)
>
> which I think is fairly clear. It also makes sure that warnings are
> always emitted by -fallow-argument-mismatch by removing
> -Wargument-mismatch. Finally, for people who do not want to have too
> many warnings cluttering up their logs, a type mismatch is only
> shown once if it is a warning.
>
> While I was going on about fixing warnings, I also fixed PR 91557 with
> the bit in trans-expr.c. This part is trivial, I will backport it
> to the other affected branches.
>
> After this, I think we can close PR 91556. Regression-tested. OK for
> trunk?
>
Looks good to me. Thanks for working of this issue.
--
Steve
20170425 https://www.youtube.com/watch?v=VWUpyCsUKR4
20161221 https://www.youtube.com/watch?v=IbCHE-hONow
Re: [PATCH V2] Setup predicate for switch default case in IPA (PR ipa/91089)
> It is somewhat hard to write a testcase to show role of range info only with
> this patch. If another patch "Generalized predicate/condition for parameter
> reference in IPA (PR ipa/91088)" is accepted, it will become easy and I will
> update this testcase to show that.
>
> And this new version also resolves a problem that we might generate very
> complicated predicate for basic block in convergence point reached from
> all switch cases.
>
>switch (index)
> / | \
> case1 case2 ... default
> \ | /
>convergence point
>
> For switch/if statement, current algorithm synthesizes predicate of
> convergence point by OR-combining predicates of all its cases/branches, which
> might give us a very complicated one. Actually, we can get simplified
> predicate
> by using the fact that predicate for a basic block must also hold true for
> its post
> dominators. To be specific, convergence point should include (at least equal
> to)
> predicate of the switch/if statement.
>
> Honza & Martin,
>
> Would please review this new patch when you have time? Thanks.
>
> Feng
>
> -
> diff --git a/gcc/ipa-fnsummary.c b/gcc/ipa-fnsummary.c
> index 278bf606661..dc5752fc005 100644
> --- a/gcc/ipa-fnsummary.c
> +++ b/gcc/ipa-fnsummary.c
> @@ -1198,7 +1198,8 @@ set_cond_stmt_execution_predicate (struct
> ipa_func_body_info *fbi,
> /* invert_tree_comparison will return ERROR_MARK on FP
>comparsions that are not EQ/NE instead of returning proper
>unordered one. Be sure it is not confused with NON_CONSTANT. */
> - if (this_code != ERROR_MARK)
> + if (this_code != ERROR_MARK
> + && !dominated_by_p (CDI_POST_DOMINATORS, bb, e->dest))
> {
> predicate p
> = add_condition (summary, index, size, &aggpos, this_code,
So this change is handling the diamond conditional you describe above?
This is bit of hack since it leaves the edge predicate unnecesarily
conservative though I see it saves some conditions to be inserted and
makes things to go smoother.
Please add a comment that explain this and reffers to the other places
where we do this (in the switch handling below).
> @@ -1269,13 +1270,31 @@ set_switch_stmt_execution_predicate (struct
> ipa_func_body_info *fbi,
>if (!unmodified_parm_or_parm_agg_item (fbi, last, op, &index, &size,
> &aggpos))
> return;
>
> + auto_vec > ranges;
> + tree type = TREE_TYPE (op);
> + tree one_cst = build_one_cst (type);
> + int bound_limit = PARAM_VALUE (PARAM_IPA_MAX_SWITCH_PREDICATE_BOUNDS);
> + int bound_count = 0;
> + value_range_base vr;
> +
> + get_range_info (op, vr);
> +
>FOR_EACH_EDGE (e, ei, bb->succs)
> {
>e->aux = edge_predicate_pool.allocate ();
>*(predicate *) e->aux = false;
> }
> +
> + e = gimple_switch_edge (cfun, last, 0);
> + /* Set BOUND_COUNT to maximum count to bypass computing predicate for
> + default case if its target basic block is in convergence point of all
> + switch cases, which can be determined by checking whether it
> + post-dominates the switch statement. */
> + if (dominated_by_p (CDI_POST_DOMINATORS, bb, e->dest))
> +bound_count = INT_MAX;
> +
>n = gimple_switch_num_labels (last);
> - for (case_idx = 0; case_idx < n; ++case_idx)
> + for (case_idx = 1; case_idx < n; ++case_idx)
> {
>tree cl = gimple_switch_label (last, case_idx);
>tree min, max;
> @@ -1285,10 +1304,10 @@ set_switch_stmt_execution_predicate (struct
> ipa_func_body_info *fbi,
>min = CASE_LOW (cl);
>max = CASE_HIGH (cl);
>
> - /* For default we might want to construct predicate that none
> - of cases is met, but it is bit hard to do not having negations
> - of conditionals handy. */
> - if (!min && !max)
> + /* The case's target basic block is in convergence point of all switch
> + cases, its predicate should be at least as that of the switch
> + statement. */
> + if (dominated_by_p (CDI_POST_DOMINATORS, bb, e->dest))
> p = true;
>else if (!max)
> p = add_condition (summary, index, size, &aggpos, EQ_EXPR,
> @@ -1304,7 +1323,111 @@ set_switch_stmt_execution_predicate (struct
> ipa_func_body_info *fbi,
> }
>*(class predicate *) e->aux
> = p.or_with (summary->conds, *(class predicate *) e->aux);
> +
> + /* If there are too many disjoint case ranges, predicate for default
> + case might become too complicated. So add a limit here. */
> + if (bound_count > bound_limit)
> + continue;
> +
> + bool new_range = true;
> +
> + if (!ranges.is_empty ())
> + {
> + tree last_max_plus_one
> + = int_const_binop (PLUS_EXPR, ranges.last ().second, one_cst);
> +
> + /* Merge case ranges if they are continuous. */
> + if (tree_int_cst_equal (min, last_max_plus_one))
> + new_range = false;
> +
Re: [PATCH V3] Generalized predicate/condition for parameter reference in IPA (PR ipa/91088)
> +/* Analyze EXPR if it represents a series of simple operations performed on
> + a function parameter and return true if so. FBI, STMT, EXPR, INDEX_P and
> + AGGPOS have the same meaning like in unmodified_parm_or_parm_agg_item.
> + Type of the parameter or load from an aggregate via the parameter is
> + stored in *TYPE_P. Operations on the parameter are recorded to
> + PARAM_OPS_P if it is not NULL. */
> +
> +static bool
> +decompose_param_expr (struct ipa_func_body_info *fbi,
> + gimple *stmt, tree expr,
> + int *index_p, tree *type_p,
> + struct agg_position_info *aggpos,
> + expr_eval_ops *param_ops_p = NULL)
> +{
> + int op_limit = PARAM_VALUE (PARAM_IPA_MAX_PARAM_EXPR_OPS);
> + int op_count = 0;
> +
> + if (param_ops_p)
> +*param_ops_p = NULL;
> +
> + while (true)
> +{
> + expr_eval_op eval_op;
> + poly_int64 size;
> +
> + if (unmodified_parm_or_parm_agg_item (fbi, stmt, expr, index_p, &size,
> + aggpos))
> + {
> + tree type = TREE_TYPE (expr);
> +
> + /* Stop if found bit-field whose size does not equal any natural
> + integral type. */
> + if (maybe_ne (tree_to_poly_int64 (TYPE_SIZE (type)), size))
> + break;
Why is this needed?
> +
> + *type_p = type;
> + return true;
> + }
> +
> + if (TREE_CODE (expr) != SSA_NAME || SSA_NAME_IS_DEFAULT_DEF (expr))
> + break;
> +
> + if (!is_gimple_assign (stmt = SSA_NAME_DEF_STMT (expr)))
> + break;
> +
> + switch (gimple_assign_rhs_class (stmt))
> + {
> + case GIMPLE_SINGLE_RHS:
> + expr = gimple_assign_rhs1 (stmt);
> + continue;
> +
> + case GIMPLE_UNARY_RHS:
> + expr = gimple_assign_rhs1 (stmt);
> +
> + eval_op.val_is_rhs = false;
I find val_is_rhs to be confusing, lhs/rhs is usually used for the
gimple assignments. Here everything is rhs, but it depends whether
the val is operand0 or operand1. So I guess we could do val_is_op1.
> @@ -1183,10 +1301,8 @@ set_cond_stmt_execution_predicate (struct
> ipa_func_body_info *fbi,
>if (!is_gimple_ip_invariant (gimple_cond_rhs (last)))
> return;
>op = gimple_cond_lhs (last);
> - /* TODO: handle conditionals like
> - var = op0 < 4;
> - if (var != 0). */
> - if (unmodified_parm_or_parm_agg_item (fbi, last, op, &index, &size,
> &aggpos))
> +
> + if (decompose_param_expr (fbi, last, op, &index, &type, &aggpos,
> ¶m_ops))
> {
>code = gimple_cond_code (last);
>inverted_code = invert_tree_comparison (code, HONOR_NANS (op));
> @@ -1201,13 +1317,13 @@ set_cond_stmt_execution_predicate (struct
> ipa_func_body_info *fbi,
> if (this_code != ERROR_MARK)
> {
> predicate p
> - = add_condition (summary, index, size, &aggpos, this_code,
> - unshare_expr_without_location
> - (gimple_cond_rhs (last)));
> + = add_condition (summary, index, type, &aggpos, this_code,
> + gimple_cond_rhs (last), param_ops);
An why unsharing is no longer needed here?
It is importnat to avoid anything which reffers to function local blocks
to get into the global LTO stream.
> +/* Check whether two set of operations have same effects. */
> +static bool
> +expr_eval_ops_equal_p (expr_eval_ops ops1, expr_eval_ops ops2)
> +{
> + if (ops1)
> +{
> + if (!ops2 || ops1->length () != ops2->length ())
> + return false;
> +
> + for (unsigned i = 0; i < ops1->length (); i++)
> + {
> + expr_eval_op &op1 = (*ops1)[i];
> + expr_eval_op &op2 = (*ops2)[i];
> +
> + if (op1.code != op2.code
> + || op1.val_is_rhs != op2.val_is_rhs
> + || !vrp_operand_equal_p (op1.val, op2.val)
Why you need vrp_operand_equal_p instead of operand_equal_p here?
Overall the patch looks very reasonable. We may end up with bit more
general expressions (i.e. supporting arithmetics involving more than one
operand).
I see you also fixed a lot of typos in comments, please go head and
commit them separately as obvious.
Thank for all the work!
Honza
[testsuite, obvious] Remove explicit "dg-do run" from g++.dg/vect/pr87914.cc
I found another instance of PR testsuite/83889 in the g++ testsuite.
I've committed the attached patch to fix it in the same way as all the
others.
-Sandra
2019-09-14 Sandra Loosemore
PR testsuite/83889
gcc/testsuite/
* g++.dg/vect/pr87914.cc: Remove explicit dg-do run.
Index: gcc/testsuite/g++.dg/vect/pr87914.cc
===
--- gcc/testsuite/g++.dg/vect/pr87914.cc (revision 275717)
+++ gcc/testsuite/g++.dg/vect/pr87914.cc (working copy)
@@ -1,4 +1,3 @@
-// { dg-do run }
// { dg-additional-options "-fopenmp-simd" }
// { dg-additional-options "-mavx2" { target { avx2_runtime } } }
Re: [PATCH, AArch64, v3 0/6] LSE atomics out-of-line
On 9/5/19 10:35 AM, Wilco Dijkstra wrote:
> Agreed. I've got a couple of general comments:
>
> * The option name -matomic-ool sounds too abbreviated. I think eg.
> -moutline-atomics is more descriptive and user friendlier.
Changed.
> * Similarly the exported __aa64_have_atomics variable could be named
> __aarch64_have_lse_atomics so it's clear that it is about LSE atomics.
Changed.
> +@item -matomic-ool
> +@itemx -mno-atomic-ool
> +Enable or disable calls to out-of-line helpers to implement atomic
> operations.
> +These helpers will, at runtime, determine if ARMv8.1-Atomics instructions
> +should be used; if not, they will use the load/store-exclusive instructions
> +that are present in the base ARMv8.0 ISA.
> +
> +This option is only applicable when compiling for the base ARMv8.0
> +instruction set. If using a later revision, e.g. @option{-march=armv8.1-a}
> +or @option{-march=armv8-a+lse}, the ARMv8.1-Atomics instructions will be
> +used directly.
>
> So what is the behaviour when you explicitly select a specific CPU?
Selecting a specific cpu selects the specific architecture that the cpu
supports, does it not? Thus the architecture example above still applies.
Unless I don't understand what distinction that you're making?
> +/* Branch to LABEL if LSE is enabled.
> + The branch should be easily predicted, in that it will, after
> constructors,
> + always branch the same way. The expectation is that systems that
> implement
> + ARMv8.1-Atomics are "beefier" than those that omit the extension.
> + By arranging for the fall-through path to use load-store-exclusive insns,
> + we aid the branch predictor of the smallest cpus. */
>
> I'd say that by the time GCC10 is released and used in distros, systems
> without
> LSE atomics would be practically non-existent. So we should favour LSE atomics
> by default.
I suppose. Does it not continue to be true that an a53 is more impacted by the
branch prediction than an a76?
r~
Re: [PATCH][ARM] Enable code hoisting with -Os (PR80155)
On Wed, 11 Sep 2019 at 11:50, Wilco Dijkstra wrote:
>
> While code hoisting generally improves codesize, it can affect performance
> negatively. Benchmarking shows it doesn't help SPEC and negatively affects
> embedded benchmarks, so only enable code hoisting with -Os on Arm.
>
> Bootstrap OK, OK for commit?
Hi Wilco,
My only concern with the patch is that the issue isn't specific to
code-hoisting.
For this particular case (reproducible with pr77445-2.c), disabling
jump threading
doesn't cause the register spill with hoisting enabled.
Likewise disabling forwprop3 and forwprop4 prevents the spill.
The last time I tried, setting setting
max-jump-thread-duplication-stmts to 20 and
fsm-scale-path-stmts to 3, not only removed the spill but also
resulted in 9 more hoistings,
but regressed some other test on jump threading.
So I was wondering whether we should look into fine-tuning relevant params,
instead of disabling code hoisting entirely (we will end up regressing
some test cases either way) ?
Thanks,
Prathamesh
>
> ChangeLog:
> 2019-09-11 Wilco Dijkstra
>
>
> PR tree-optimization/80155
> * common/config/arm/arm-common.c (arm_option_optimization_table):
> Enable -fcode-hoisting with -Os.
>
> --
> diff --git a/gcc/common/config/arm/arm-common.c
> b/gcc/common/config/arm/arm-common.c
> index
> 41a920f6dc96833e778faa8dbcc19beac483734c..b0d5fb300bf01acc1fb6f4631635f8a1bfe6441c
> 100644
> --- a/gcc/common/config/arm/arm-common.c
> +++ b/gcc/common/config/arm/arm-common.c
> @@ -39,6 +39,8 @@ static const struct default_options
> arm_option_optimization_table[] =
> /* Enable section anchors by default at -O1 or higher. */
> { OPT_LEVELS_1_PLUS, OPT_fsection_anchors, NULL, 1 },
> { OPT_LEVELS_1_PLUS, OPT_fsched_pressure, NULL, 1 },
> +/* Enable code hoisting only with -Os. */
> +{ OPT_LEVELS_SIZE, OPT_fcode_hoisting, NULL, 1 },
> { OPT_LEVELS_NONE, 0, NULL, 0 }
>};
>
>
