On September 26, 2020 10:20:20 AM GMT+02:00, Jan Hubicka wrote:
>Hi,
>this patchs finishes the parameter tracking by implementing the
>iterative
>dataflow in propagation stage. This is necessary since we now can
>propagate how the pointers are passed around recursive calls (as done
>in
>a testcase
On Fri, 25 Sep 2020, Jason Merrill wrote:
> On 9/25/20 2:30 AM, Richard Biener wrote:
> > On Thu, 24 Sep 2020, Jason Merrill wrote:
> >
> >> On 9/24/20 3:43 AM, Richard Biener wrote:
> >>> On Wed, 23 Sep 2020, Jason Merrill wrote:
> >>>
> >&g
ef_may_alias_errno (ref)));
> - }
> - /* Freeing memory kills the pointed-to memory. More importantly
> -the call has to serve as a barrier for moving loads and stores
> -across it. */
> - case BUILT_IN_FREE:
> - case BUILT_IN_VA_END:
> - {
> - tree ptr = gimple_call_arg (call, 0);
> - return ptr_deref_may_alias_ref_p_1 (ptr, ref);
> - }
> - /* Realloc serves both as allocation point and deallocation point. */
> - case BUILT_IN_REALLOC:
> - {
> - tree ptr = gimple_call_arg (call, 0);
> - /* Unix98 specifies that errno is set on allocation failure. */
> - return ((flag_errno_math
> - && targetm.ref_may_alias_errno (ref))
> - || ptr_deref_may_alias_ref_p_1 (ptr, ref));
> - }
> - case BUILT_IN_GAMMA_R:
> - case BUILT_IN_GAMMAF_R:
> - case BUILT_IN_GAMMAL_R:
> - case BUILT_IN_LGAMMA_R:
> - case BUILT_IN_LGAMMAF_R:
> - case BUILT_IN_LGAMMAL_R:
> - {
> - tree out = gimple_call_arg (call, 1);
> - if (ptr_deref_may_alias_ref_p_1 (out, ref))
> - return true;
> - if (flag_errno_math)
> - break;
> - return false;
> - }
> - case BUILT_IN_FREXP:
> - case BUILT_IN_FREXPF:
> - case BUILT_IN_FREXPL:
> - case BUILT_IN_MODF:
> - case BUILT_IN_MODFF:
> - case BUILT_IN_MODFL:
> - {
> - tree out = gimple_call_arg (call, 1);
> - return ptr_deref_may_alias_ref_p_1 (out, ref);
> - }
> - case BUILT_IN_REMQUO:
> - case BUILT_IN_REMQUOF:
> - case BUILT_IN_REMQUOL:
> - {
> - tree out = gimple_call_arg (call, 2);
> - if (ptr_deref_may_alias_ref_p_1 (out, ref))
> - return true;
> - if (flag_errno_math)
> - break;
> - return false;
> - }
> - case BUILT_IN_SINCOS:
> - case BUILT_IN_SINCOSF:
> - case BUILT_IN_SINCOSL:
> - {
> - tree sin = gimple_call_arg (call, 1);
> - tree cos = gimple_call_arg (call, 2);
> - return (ptr_deref_may_alias_ref_p_1 (sin, ref)
> - || ptr_deref_may_alias_ref_p_1 (cos, ref));
> - }
> - /* __sync_* builtins and some OpenMP builtins act as threading
> -barriers. */
> -#undef DEF_SYNC_BUILTIN
> -#define DEF_SYNC_BUILTIN(ENUM, NAME, TYPE, ATTRS) case ENUM:
> -#include "sync-builtins.def"
> -#undef DEF_SYNC_BUILTIN
> - case BUILT_IN_GOMP_ATOMIC_START:
> - case BUILT_IN_GOMP_ATOMIC_END:
> - case BUILT_IN_GOMP_BARRIER:
> - case BUILT_IN_GOMP_BARRIER_CANCEL:
> - case BUILT_IN_GOMP_TASKWAIT:
> - case BUILT_IN_GOMP_TASKGROUP_END:
> - case BUILT_IN_GOMP_CRITICAL_START:
> - case BUILT_IN_GOMP_CRITICAL_END:
> - case BUILT_IN_GOMP_CRITICAL_NAME_START:
> - case BUILT_IN_GOMP_CRITICAL_NAME_END:
> - case BUILT_IN_GOMP_LOOP_END:
> - case BUILT_IN_GOMP_LOOP_END_CANCEL:
> - case BUILT_IN_GOMP_ORDERED_START:
> - case BUILT_IN_GOMP_ORDERED_END:
> - case BUILT_IN_GOMP_SECTIONS_END:
> - case BUILT_IN_GOMP_SECTIONS_END_CANCEL:
> - case BUILT_IN_GOMP_SINGLE_COPY_START:
> - case BUILT_IN_GOMP_SINGLE_COPY_END:
> - return true;
> - default:
> - /* Fallthru to general call handling. */;
> - }
> + }
> +}
>
>/* Check if base is a global static variable that is not written
> by the function. */
>
--
Richard Biener
SUSE Software Solutions Germany GmbH, Maxfeldstrasse 5, 90409 Nuernberg,
Germany; GF: Felix Imend
):
> > > - CASE_BUILT_IN_TM_STORE (4):
> > > - CASE_BUILT_IN_TM_STORE (8):
> > > - CASE_BUILT_IN_TM_STORE (FLOAT):
> > > - CASE_BUILT_IN_TM_STORE (DOUBLE):
> > > - CASE_BUILT_IN_TM_STORE (LDOUBLE):
> > > - CASE_BUILT_IN_TM_STORE (M64):
> &g
On Mon, 28 Sep 2020, Jan Hubicka wrote:
> >
> > Hmm, no - it expects the gimple_call_use/clobber_set to include
> > actions of the called function itself (const functions none here)
> > but for passed-by-value the loads may be in the call itself
> > which is where it uses gimple_call_arg_flags to
On Mon, 28 Sep 2020, Jan Hubicka wrote:
> > On Mon, 28 Sep 2020, Jan Hubicka wrote:
> >
> > > >
> > > > Hmm, no - it expects the gimple_call_use/clobber_set to include
> > > > actions of the called function itself (const functions none here)
> > > > but for passed-by-value the loads may be in th
On Fri, 25 Sep 2020, Tamar Christina wrote:
> Hi All,
>
> This patch series adds support for SLP vectorization of complex instructions
> [1].
>
> These instructions exist only in their vector forms and require you to
> recognize
> two statements in parallel. Complex operations usually require
gt; * tree-vect-slp.c (_slp_tree::_slp_tree, _slp_tree::~_slp_tree,
> vect_free_slp_tree, vect_build_slp_tree, vect_print_slp_tree,
> slp_copy_subtree, vect_attempt_slp_rearrange_stmts): Use it.
>
>
--
Richard Biener
SUSE Software Solutions Germany GmbH, Maxfeldstrasse 5, 90409 Nuernberg,
Germany; GF: Felix Imend
> Thanks,
> Tamar
>
> gcc/ChangeLog:
>
> * Makefile.in (tree-vect-slp-patterns.o): New.
> * doc/passes.texi: Update documentation.
> * tree-vect-slp.c (vect_match_slp_patterns_2, vect_match_slp_patterns):
> New.
> (vect_analyze_slp_
MT simply gets replaced). I
wonder if it would be easier to record the SLP pattern stmt
only in SLP_TREE_REPRESENTATIVE but _not_ in SLP_TREE_SCALAR_STMTS
(just leave those alone)?
Richard.
> Thanks,
> Tamar
>
> gcc/ChangeLog:
>
> * tree-vect-loop.c (vect_dissolve_sl
seem to share
the initial mixed plus/minus, then we have the multiplication
on one or both operand cases.
Richard.
> Thanks,
> Tamar
>
> gcc/ChangeLog:
>
> * tree-vect-slp-patterns.c (complex_operation_t,class ComplexPattern):
> New.
>
>
--
Richard Biener
SUSE Software Solutions Germany GmbH, Maxfeldstrasse 5, 90409 Nuernberg,
Germany; GF: Felix Imend
On Mon, 28 Sep 2020, Tamar Christina wrote:
> Hi Richi,
>
> > -Original Message-
> > From: rguent...@c653.arch.suse.de On
> > Behalf Of Richard Biener
> > Sent: Monday, September 28, 2020 2:22 PM
> > To: Tamar Christina
> > Cc: gcc-patches@gcc
On Mon, 28 Sep 2020, Tamar Christina wrote:
>
>
> > -Original Message-
> > From: Gcc-patches On Behalf Of Tamar
> > Christina
> > Sent: Monday, September 28, 2020 3:56 PM
> > To: Richard Biener
> > Cc: nd ; gcc-patches@gcc.gnu.org; o...@ucw
On Tue, 29 Sep 2020, Richard Sandiford wrote:
> Richard Biener writes:
> >> > > @@ -2192,6 +2378,17 @@ vect_analyze_slp_instance (vec_info *vinfo,
> >> > > &tree_size, bst_map);
> >> > >if (node != NULL)
gt; +OPTAB_D (cadd270_optab, "cadd270$a3")
> > OPTAB_D (cos_optab, "cos$a2")
> > OPTAB_D (cosh_optab, "cosh$a2")
> > OPTAB_D (exp10_optab, "exp10$a2")
> > diff --git a/gcc/tree-vect-slp-patterns.c b/gcc/tree-vect-slp-patterns.c
> > index
> > 6453a5b1b6464dba833adc2c2a194db5e712bb79..b2b0ac62e9a69145470f41d2bac736dd970be735
> > 100644
> > --- a/gcc/tree-vect-slp-patterns.c
> > +++ b/gcc/tree-vect-slp-patterns.c
> > @@ -663,12 +663,94 @@ graceful_exit:
> > }
> > };
> >
> > +class ComplexAddPattern : public ComplexPattern
>
> Another nitpick, sorry, but type names should be lower case rather than
> CamelCase.
>
> Thanks,
> Richard
>
--
Richard Biener
SUSE Software Solutions Germany GmbH, Maxfeldstrasse 5, 90409 Nuernberg,
Germany; GF: Felix Imend
This fixes a typo causing a NULL dereference.
Bootstrapped and tested on x86_64-unknown-linux-gnu, pushed.
2020-09-29 Richard Biener
PR tree-optimization/97238
* tree-ssa-reassoc.c (ovce_extract_ops): Fix typo.
* gcc.dg/pr97238.c: New testcase.
---
gcc/testsuite
This moves optimizing permutes of SLP reductions to vect_optimize_slp,
eliding the global slp_loads array.
Bootstrapped and tested on x86_64-unknown-linux-gnu, pushed.
2020-09-29 Richard Biener
* tree-vect-slp.c (vect_analyze_slp): Move SLP reduction
re-arrangement and SLP
The following moves an ad-hoc attempt at discovering the SLP node
for a stmt to the place where we can find it in lock-step when
we find the stmt itself.
Bootstrapped / tested on x86_64-unknown-linux-gnu, pushed.
2020-09-29 Richard Biener
PR tree-optimization/97241
* tree
On September 29, 2020 4:17:30 PM GMT+02:00, Jan Hubicka wrote:
>Hi,
>this patch fixes accidental \000 in fnspec strings for internal fns.
>OK?
OK.
Richard.
>Honza
>
> * internal-fn.c (DEF_INTERNAL_FN): Fix call of build_string.
>diff --git a/gcc/internal-fn.c b/gcc/internal-fn.c
>index
On September 29, 2020 4:20:42 PM GMT+02:00, Jan Hubicka wrote:
>Hi,
>this patch is not needed but makes it possible to sanity check that
>fnspec match function signature. It turns out that there are quite few
>mistakes in that in trans-decl and one mistake here.
>Transfer_derived has additional pa
node, 4, dt_parm_type, pvoid_type_node,
> > > integer_type_node, gfc_charlen_type_node);
> > >
> > >iocall[IOCALL_X_ARRAY_WRITE] =
> > >gfc_build_library_function_decl_with_spec (
> > >- get_identifier (PREFIX("transfer_array_write")), ".wr&
vector constructors.
>
> gcc/testsuite/ChangeLog:
>
> 2020-09-29 Joel Hutton
>
> PR target/96837
> * gcc.dg/vect/bb-slp-49.c: New test.
--
Richard Biener
SUSE Software Solutions Germany GmbH, Maxfeldstrasse 5, 90409 Nuernberg,
Germany; GF: Felix Imend
*/
> > +/* { dg-require-effective-target fenv_exceptions } */
> > +/* { dg-options "-lm -fno-builtin" } */
>
> That -fno-builtin looks very strange... Comment what it is for?
>
> > +#define FAIL(v, e) printf("ERROR, __builtin_fegetround() returned %d," \
> > + " not the expecected value %d\n", v, e);
>
> (Typo, "expected")
>
> The rs6000 part is okay for trunk (with those modifications), after the
> generic parts is approved. Thanks!
>
>
> Segher
>
--
Richard Biener
SUSE Software Solutions Germany GmbH, Maxfeldstrasse 5, 90409 Nuernberg,
Germany; GF: Felix Imend
This amends SLP reduction testcases that currently trigger
vect_attempt_slp_rearrange_stmts eliding load permutations to
verify this is actually happening.
tested on x86_64-unknown-linux-gnu, pushed
2020-09-30 Richard Biener
* gcc.dg/vect/pr37027.c: Amend.
* gcc.dg/vect
On Wed, 30 Sep 2020, Segher Boessenkool wrote:
> On Wed, Sep 30, 2020 at 09:02:34AM +0200, Richard Biener wrote:
> > On Tue, 29 Sep 2020, Segher Boessenkool wrote:
> > > I don't see much about optabs in the docs either. Add some text to
> > > optabs.def itse
-gnu, pushed to trunk.
2020-10-01 Richard Biener
* tree-vect-patterns.c (vect_recog_bool_pattern): Also handle
VIEW_CONVERT_EXPR.
* g++.dg/vect/pr97255.cc: New testcase.
---
gcc/testsuite/g++.dg/vect/pr97255.cc | 44
gcc/tree-vect-patterns.c
On Wed, 30 Sep 2020, Jason Merrill wrote:
> On 9/28/20 3:09 PM, Jason Merrill wrote:
> > On 9/28/20 3:56 AM, Richard Biener wrote:
> >> On Fri, 25 Sep 2020, Jason Merrill wrote:
> >>
> >>> On 9/25/20 2:30 AM, Richard Biener wrote:
> >>>> On Th
This avoids using VMAT_CONTIGUOUS with single-element interleaving
when using V1mode vectors. Instead keep VMAT_ELEMENTWISE but
continue to avoid load-lanes and gathers.
Bootstrapped on x86_64-unknown-linux-gnu, testing in progress.
Richard.
2020-10-01 Richard Biener
PR tree
ol
> + arg_direct_p (unsigned int i)
> + {
> +unsigned int idx = arg_idx (i);
> +gcc_checking_assert (arg_specified_p (i));
> +return str[idx] == 'R' || str[idx] == 'W';
> + }
> +
> + /* True if argument is used. */
> + bool
> + arg_use
e fnspec = lookup_attribute ("fn spec",
> - TYPE_ATTRIBUTES (TREE_TYPE (fun->decl)));
> - if (fnspec)
> + tree fnspec_tree
> + = lookup_attribute ("fn spec",
> + TYPE_ATTRIBUTES (TREE_TYPE (fun->decl)));
> + if (fnspec_tree)
> {
> - fnspec = TREE_VALUE (TREE_VALUE (fnspec));
> - unsigned i = 1;
> + attr_fnspec fnspec (TREE_VALUE (TREE_VALUE (fnspec_tree)));
> + unsigned i = 0;
>for (tree arg = DECL_ARGUMENTS (cfun->decl);
> arg; arg = DECL_CHAIN (arg), ++i)
> {
> - if (i >= (unsigned) TREE_STRING_LENGTH (fnspec))
> - break;
> - if (TREE_STRING_POINTER (fnspec)[i] == 'R'
> - || TREE_STRING_POINTER (fnspec)[i] == 'r')
> + if (!fnspec.arg_specified_p (i))
> +break;
> + if (fnspec.arg_readonly_p (i))
> {
> tree name = ssa_default_def (fun, arg);
> if (name)
> diff --git a/gcc/tree-ssa-alias.c b/gcc/tree-ssa-alias.c
> index fe390d4ffbe..2b2d53ebb88 100644
> --- a/gcc/tree-ssa-alias.c
> +++ b/gcc/tree-ssa-alias.c
> @@ -40,6 +40,7 @@ along with GCC; see the file COPYING3. If not see
> #include "varasm.h"
> #include "ipa-modref-tree.h"
> #include "ipa-modref.h"
> +#include "attr-fnspec.h"
>
> /* Broad overview of how alias analysis on gimple works:
>
> @@ -3984,3 +3985,42 @@ walk_aliased_vdefs (ao_ref *ref, tree vdef,
>return ret;
> }
>
> +/* Verify validity of the fnspec string.
> + See attr-fnspec.h for details. */
> +
> +void
> +attr_fnspec::verify ()
> +{
> + /* FIXME: Fortran trans-decl.c contains multiple wrong fnspec strings.
> + re-enable verification after these are fixed. */
> + return;
> + bool err = false;
> +
> + /* Check return value specifier. */
> + if (len < return_desc_size)
> +err = true;
> + else if (str[0] < '1' || str[0] > '4')
> +&& str[0] != '.' && str[0] != 'm')
> +err = true;
> +
> + /* Now check all parameters. */
> + for (unsigned int i = 0; arg_specified_p (i); i++)
> +{
> + unsigned int idx = arg_idx (i);
> + switch (str[idx])
> + {
> + case 'x':
> + case 'X':
> + case 'r':
> + case 'R':
> + case 'w':
> + case 'W':
> + case '.':
> + break;
> + default:
> + err = true;
> + }
> +}
> + if (err)
> +internal_error ("invalid fn spec attribute %s", str);
> +}
>
--
Richard Biener
SUSE Software Solutions Germany GmbH, Maxfeldstrasse 5, 90409 Nuernberg,
Germany; GF: Felix Imend
;
> +poly_int64 streamer_read_poly_int64 (class lto_input_block *);
> gcov_type streamer_read_gcov_count (class lto_input_block *);
> wide_int streamer_read_wide_int (class lto_input_block *);
> widest_int streamer_read_widest_int (class lto_input_block *);
>
--
Richard Biener
SUSE Software Solutions Germany GmbH, Maxfeldstrasse 5, 90409 Nuernberg,
Germany; GF: Felix Imend
On Thu, 1 Oct 2020, Jason Merrill wrote:
> On 10/1/20 5:26 AM, Richard Biener wrote:
> > On Wed, 30 Sep 2020, Jason Merrill wrote:
> >
> >> On 9/28/20 3:09 PM, Jason Merrill wrote:
> >>> On 9/28/20 3:56 AM, Richard Biener wrote:
> >>>> On Fri,
& poly_int_tree_p (size, &size_hwi)
> + && coeffs_in_range_p (size_hwi, 0, HOST_WIDE_INT_MAX / BITS_PER_UNIT))
> +{
> + size_hwi = size_hwi * BITS_PER_UNIT;
> + ao_ref_init_from_ptr_and_range (ref, ptr, true, 0, size_hwi, size_hwi);
> +}
> + else
> +ao_ref_init_from_ptr_and_range (ref, ptr, false, 0, -1, -1);
> +}
> +
> /* S1 and S2 are TYPE_SIZE or DECL_SIZE. Compare them:
> Return -1 if S1 < S2
> Return 1 if S1 > S2
>
--
Richard Biener
SUSE Software Solutions Germany GmbH, Maxfeldstrasse 5, 90409 Nuernberg,
Germany; GF: Felix Imend
On Fri, 2 Oct 2020, Jan Hubicka wrote:
> Hi,
> as discussed this patch makes return value and arg specifiers to be 2
> characters long and updates (I hope) all fnspec strings.
> I also enabled part of the verification (just accepting the fortran bug
> with 'R' and 'W' in return value specifiers) t
ux-gnu, testing in progress.
Richard.
2020-10-02 Richard Biener
* tree-vect-data-refs.c (vect_slp_analyze_instance_dependence):
Use SLP_TREE_REPRESENTATIVE.
* tree-vectorizer.h (_slp_tree::vertex): New member used
for graphds interfacing.
*
5,8 +2505,8 @@ modref_may_conflict (const gimple *stmt,
> }
>
> /* TBAA checks did not disambiguate, try to use base pointer, for
> - that we however need to have ref->ref. */
> - if (ref_node->every_access || !ref->ref)
> + tha
On October 3, 2020 10:41:26 AM GMT+02:00, Jakub Jelinek
wrote:
>On Fri, Oct 02, 2020 at 04:21:12PM +0200, Stefan Schulze Frielinghaus
>via Gcc-patches wrote:
>> > > Sure, no problem at all. In that case I stop to investigate
>further and
>> > > wait for you.
>> >
>> > Here is a patch that imple
On October 5, 2020 9:08:41 AM GMT+02:00, Jakub Jelinek wrote:
>On Sun, Oct 04, 2020 at 09:16:00PM +0200, Jakub Jelinek via Gcc-patches
>wrote:
>> On Sun, Oct 04, 2020 at 09:13:29AM +0200, Andreas Schwab wrote:
>> > This breaks ia64:
>> >
>> > In file included from ./tm.h:23,
>> >
is necessary.
> Without going through this folding we are unable to disambiguate base
> pointers: MEM_REF containing ADDR_EXPR of DECL does not go through the
> usual decl compares.
>
> Honza
> >
> > >
> > > > I'm not sure that automatically openi
On Fri, 2 Oct 2020, Richard Sandiford wrote:
> Richard Biener writes:
> > This introduces a permute optimization phase for SLP which is
> > intended to cover the existing permute eliding for SLP reductions
> > plus handling commonizing the easy cases.
> >
> > It
On Mon, 5 Oct 2020, Tom de Vries wrote:
> [ was: Re: [PATCH][omp, ftracer] Don't duplicate blocks in SIMT region ]
>
> On 10/5/20 9:05 AM, Tom de Vries wrote:
> > Ack, updated the patch accordingly, and split it up in two bits, one
> > that does refactoring, and one that adds the actual caching:
On Mon, 5 Oct 2020, Tom de Vries wrote:
> [ was: Re: [PATCH][omp, ftracer] Don't duplicate blocks in SIMT region ]
>
> On 10/5/20 9:05 AM, Tom de Vries wrote:
> > Ack, updated the patch accordingly, and split it up in two bits, one
> > that does refactoring, and one that adds the actual caching:
0200
> +++ gcc/testsuite/gcc.target/i386/pr97282.c 2020-10-05 15:28:51.864499619
> +0200
> @@ -0,0 +1,25 @@
> +/* PR rtl-optimization/97282 */
> +/* { dg-do compile } */
> +/* { dg-options "-O2" } */
> +/* { dg-final { scan-assembler "call\[^\n\r]*__udivmod\[dt]i4" } } */
> +
> +#ifdef __SIZEOF_INT128__
> +typedef __uint128_t T;
> +#else
> +typedef unsigned long long T;
> +#endif
> +
> +unsigned long
> +foo (T x)
> +{
> + if (x == 0)
> +return 0;
> +
> + unsigned long ret = 0;
> + while (x > 0)
> +{
> + ret = ret + x % 10;
> + x = x / 10;
> +}
> + return ret;
> +}
>
> Jakub
>
>
--
Richard Biener
SUSE Software Solutions Germany GmbH, Maxfeldstrasse 5, 90409 Nuernberg,
Germany; GF: Felix Imend
or (i = 0; i < n_target_char; i++) {
> + if (j == 0 && k == 0) {
> + print " unsigned HOST_WIDE_INT mask = 0;";
> + }
> print " if (opts_set->x_" var_target_char[i] ") mask |=
> HOST_WIDE_INT_1U << " j ";";
> j++;
> if (j == 64) {
> @@ -676,6 +693,9 @@ for (i = 0; i < n_target_char; i++) {
> }
>
> for (i = 0; i < n_target_string; i++) {
> + if (j == 0 && k == 0) {
> + print " unsigned HOST_WIDE_INT mask = 0;";
> + }
> print " if (opts_set->x_" var_target_string[i] ") mask |=
> HOST_WIDE_INT_1U << " j ";";
> j++;
> if (j == 64) {
> @@ -732,7 +752,9 @@ for (i = 0; i < n_target_string; i++) {
> }
>
> print "";
> -print " unsigned HOST_WIDE_INT mask;";
> +if (has_target_explicit_mask) {
> + print " unsigned HOST_WIDE_INT mask;";
> +}
>
> j = 64;
> k = 0;
>
>
> Jakub
>
>
--
Richard Biener
SUSE Software Solutions Germany GmbH, Maxfeldstrasse 5, 90409 Nuernberg,
Germany; GF: Felix Imend
On Tue, 6 Oct 2020, Richard Biener wrote:
> On Fri, 2 Oct 2020, Richard Sandiford wrote:
>
> > Richard Biener writes:
> > > This introduces a permute optimization phase for SLP which is
> > > intended to cover the existing permute eliding for SLP reductions
>
n_bbs, copied_bbs,
>true))
> {
> - fprintf (dump_file, "Duplication failed.\n");
> + if (dump_file && (dump_flags & TDF_DETAILS))
> + fprintf (dump_file, "Duplication failed.\n");
> continue;
> }
>copied.safe_push (std::make_pair (entry, loop));
>
--
Richard Biener
SUSE Software Solutions Germany GmbH, Maxfeldstrasse 5, 90409 Nuernberg,
Germany; GF: Felix Imend
ritten with that same value. Such re-writing is safe in a single
> +threaded context but may be unsafe in a multi-threaded context. Note that on
> +some processors, if-conversions may be required in order to enable
> +vectorization.
>
> Enabled at level @option{-Ofast}.
>
>
--
Richard Biener
SUSE Software Solutions Germany GmbH, Maxfeldstrasse 5, 90409 Nuernberg,
Germany; GF: Felix Imend
This excludes BOOLEAN_TYPE which are signed or have TYPE_PRECISION != 1
from being handled as VECTOR_BOOLEAN_TYPE_P. This fixes a bootstrap
ICE with LTO and PGO in the Ada frontend.
Bootstrapped on x86_64-unknown-linux-gnu, testing in progress.
Richard.
2020-10-07 Richard Biener
PR
This improves the heuristics finding a sink location for loads that does
not cross any store.
Bootstrapped and tested on x86_64-unknown-linux-gnu, pushed.
2020-10-07 Richard Biener
PR tree-optimization/97307
* tree-ssa-sink.c (statement_sink_location): Change heuristic
ref2.ref_alias_set = ref_set;
> + ref2.base_alias_set = base_set;
> + if (refs_may_alias_p_1 (&ref2, ref, tbaa_p))
> + return true;
> + }
> + else if (ptr_deref_may_alias_ref_p_1 (arg, ref))
> return true;
> }
> num_tests++;
>
--
Richard Biener
SUSE Software Solutions Germany GmbH, Maxfeldstrasse 5, 90409 Nuernberg,
Germany; GF: Felix Imend
(not) vectorized loops instead of
relying on fragile counts.
Bootstrapped / tested on x86_64-unknown-linux-gnu, pushed.
Richard.
2020-10-08 Richard Biener
* tree-vectorizer.c (try_vectorize_loop_1): Do not dump
"basic block vectorized".
(pass_slp_vectorize::exe
This fixes bad placement of sunk loads.
Bootstrapped / tested on x86_64-unknown-linux-gnu, pushed.
2020-10-08 Richard Biener
PR tree-optimization/97330
* tree-ssa-sink.c (statement_sink_location): Avoid skipping
PHIs when they dominate the insert location
as well.
Final re-bootstrap/regtest running on x86_64-unknown-linux-gnu.
Richard.
2020-10-08 Martin Liska
Richard Biener
* tree-vectorizer.h (_bb_vec_info::const_iterator): Remove.
(_bb_vec_info::const_reverse_iterator): Likewise.
(_bb_vec_info::region_st
25.c 2020-10-08
> 14:27:22.518708841 +0200
> @@ -0,0 +1,15 @@
> +/* PR tree-optimization/97325 */
> +
> +unsigned long long
> +foo (unsigned long long c)
> +{
> + return c ? __builtin_ffs (-(unsigned short) c) : 0;
> +}
> +
> +int
> +main ()
> +{
> + if (foo (2) != 2)
> +__builtin_abort ();
> + return 0;
> +}
>
>
> Jakub
>
>
--
Richard Biener
SUSE Software Solutions Germany GmbH, Maxfeldstrasse 5, 90409 Nuernberg,
Germany; GF: Felix Imend
tail_calls (basic_block bb, struct tailcall **ret)
>if (TREE_CODE (var) != PARM_DECL
> && auto_var_in_fn_p (var, cfun->decl)
> && may_be_aliased (var)
> - && (ref_maybe_used_by_stmt_p (call, var)
> - || call_may_clobber_ref_p (call, var)))
> + && (ref_maybe_used_by_stmt_p (call, var, false)
> + || call_may_clobber_ref_p (call, var, false)))
> {
> if (!VAR_P (var))
> {
>
--
Richard Biener
SUSE Software Solutions Germany GmbH, Maxfeldstrasse 5, 90409 Nuernberg,
Germany; GF: Felix Imend
This fixes a vector CTOR insertion issue when we try to insert after
a PHI node.
Bootstrap / regtest running on x86_64-unknown-linux-gnu.
2020-10-09 Richard Biener
* tree-vect-slp.c (vect_create_constant_vectors): Properly insert
after PHIs.
* gcc.dg/vect/bb-slp-phis
This fixes leaks discovered checking whether I introduced new ones
with the last vectorizer changes.
Bootstrap / regtest running on x86_64-unknown-linux-gnu. Parts
are also applicable for branches.
2020-10-09 Richard Biener
* cgraphunit.c (expand_all_functions): Free
On Thu, 8 Oct 2020, Thomas Schwinge wrote:
> Hi Richard!
>
> On 2020-10-08T13:34:02+0200, Richard Biener wrote:
> > It might be interesting to work on adding sth like
> > dg-warning to look for -fopt-info-{optimized,missing} so
> > we could directly annotate (not)
Just use edge insertion which will appropriately handle the situation
from botan.
Bootstrapped / tested on x86_64-unknown-linux-gnu, applied.
2020-10-09 Richard Biener
PR tree-optimization/97347
* tree-vect-slp.c (vect_create_constant_vectors): Use
edge insertion when
On Thu, 8 Oct 2020, sunil.k.pandey wrote:
> On Linux/x86_64,
>
> 532e882f8872b1b4437e3a0fa8c61d2af2d999d4 is the first bad commit
> commit 532e882f8872b1b4437e3a0fa8c61d2af2d999d4
> Author: Richard Biener
> Date: Thu Oct 8 11:53:51 2020 +0200
>
> adjust BB vec
following patch makes the
case work where the first stmt has a good operand by not fatally
failing the SLP build for the operand but communicating upwards
how to commutate.
Bootstrapped / tested on x86_64-unknown-linux-gnu, pushed.
2020-10-09 Richard Biener
PR tree-optimization/97334
The following adds a effective target to properly allow
the gcc.dg/vect/pr65947-3.c expected vectorization to be adjusted
when run with, say, -march=cascadelake.
Tested on x86_64-unknown-linux-gnu, pushed.
2020-10-09 Richard Biener
gcc/
* doc/sourcebuild.texi (vect_masked_load
te_bb_no_insn_iter_p (bb))
> +return false;
> +
> + for (gimple_stmt_iterator gsi = gsi_start_bb (CONST_CAST_BB (bb));
> + !gsi_end_p (gsi); gsi_next (&gsi))
> +if (!can_duplicate_insn_p (gsi_stmt (gsi)))
> + return false;
> +
> return true;
> }
&g
This avoids abnormals in another place.
Bootstrap / regtest in progress on x86_64-unknown-linux-gnu.
2020-10-12 Richard Biener
PR tree-optimization/97357
* tree-ssa-loop-split.c (ssa_semi_invariant_p): Abnormal
SSA names are not semi invariant.
* gcc.dg
This appropriately makes matches all true after successful SLP discovery
to reliably succeed splitting. We were picking up an eventual all
false built-up from scalars state in some cases.
Bootstrap / regtest in progress on x86_64-unknown-linux-gnu.
2020-10-12 Richard Biener
* tree
profitable to vectorize (we can at least use
a vector load from the constant pool).
Bootstrapped & tested on x86_64-unknown-linux-gnu, applied.
Richard.
2019-10-23 Richard Biener
* tree-vect-slp.c (vect_build_slp_tree_2): Do not build
op from scalars in case there's a
On Tue, Oct 22, 2019 at 2:47 PM Martin Liška wrote:
>
> Hi.
>
> We should not call to_gcov_type on a count that is uninitialized.
> That's the case for a THUNK cgraph_node that we inline.
>
> Patch can bootstrap on x86_64-linux-gnu and survives regression tests.
>
> Ready to be installed?
OK.
>
On Wed, Oct 23, 2019 at 10:33 AM Eric Botcazou wrote:
>
> This is a regression present on mainline, 9 and 8 branches, but the underlying
> issue is probably latent on the 7 branch. compare_values in tree-vrp.c can
> rely on undefined overflow to compare symbolic expressions with constants so
> we
On Wed, Oct 23, 2019 at 5:36 AM Feng Xue OS wrote:
>
> Michael,
>
> > I've only noticed a couple typos, and one minor remark.
> Typos corrected.
>
> > I just wonder why you duplicated these three loops instead of integrating
> > the real body into the existing LI_FROM_INNERMOST loop. I would have
More vectorizable_shift fallout - I've removed another disparity
of SLP vs. non-SLP plus allowed type but not mode changing cases
for SLP to go through (which originally caused the reported ICEs).
Bootstrapped on x86_64-unknown-linux-gnu, testing in progress.
Richard.
2019-10-23 Ri
nal splitting I guess).
Don't we eventually want to recurse on that?
Otherwise the patch looks reasonable. Sorry for the many bits above and the
late real review from me...
Thanks,
Richard.
> Feng
>
>
> From: Richard Biener
> S
On Wed, Oct 23, 2019 at 1:00 PM Richard Sandiford
wrote:
>
> This patch is the first of a series that tries to remove two
> assumptions:
>
> (1) that all vectors involved in vectorisation must be the same size
>
> (2) that there is only one vector mode for a given element mode and
> number of
On Wed, Oct 23, 2019 at 1:51 PM Richard Sandiford
wrote:
>
> Richard Biener writes:
> > On Wed, Oct 23, 2019 at 1:00 PM Richard Sandiford
> > wrote:
> >>
> >> This patch is the first of a series that tries to remove two
> >> assumptions:
> >&g
On Mon, Oct 21, 2019 at 10:04 AM Kugan Vivekanandarajah
wrote:
>
> Hi Richard,
>
> Thanks for the pointers.
>
>
>
> On Fri, 11 Oct 2019 at 22:33, Richard Biener
> wrote:
> >
> > On Fri, Oct 11, 2019 at 6:15 AM Kugan Vivekanandarajah
> > wrote:
ces of the above
thus I refrained from adding another testcase.
Bootstraped and tested on x86_64-unknown-linux-gnu, applied.
Richard.
2019-10-23 Richard Biener
PR tree-optimization/65930
* tree-vect-loop.c (check_reduction_path): Allow conversions
th
This helps saving some IVs (though I guess the situation where it
matches in practice is scarce).
Bootstrapped and tested on x86_64-unknown-linux-gnu, applied to trunk.
Richard.
2019-10-24 Richard Biener
* tree-vect-slp.c (vect_analyze_slp): When reduction group
SLP
The no longer xfailed testcases lacked appropriate target restriction.
The following should fix the sparc fallout.
Applied to trunk.
Richard.
2019-10-24 Richard Biener
PR tree-optimization/65930
* gcc.dg/vect/vect-reduc-2char-big-array.c: Adjust again.
* gcc.dg
Committed.
Richard.
2019-10-24 Richard Biener
PR tree-optimization/65930
* gcc.dg/vect/vect-reduc-2short.c: Fix typo.
Index: gcc/testsuite/gcc.dg/vect/vect-reduc-2short.c
===
--- gcc/testsuite/gcc.dg/vect/vect
Bootstrapped and tested on x86_64-unknown-linux-gnu, applied.
Richard.
2019-10-24 Richard Biener
PR tree-optimization/92203
* treee-ssa-sccvn.c (eliminate_dom_walker::eliminate_stmt):
Skip eliminating conversion stmts inserted by insertion.
* gcc.dg/torture
Bootstrapped and tested on x86_64-unknown-linux-gnu, applied.
Richard.
2019-10-24 Richard Biener
PR tree-optimization/92205
* tree-vect-loop.c (vectorizable_reduction): Restrict
search for alternate vectype_in to lane-reducing patterns
we support
nc" "$as_ac_var"
> @@ -11448,6 +11460,61 @@ fi
> done
>
>
> +for ac_func in mallinfo
> +do
> + ac_tr_decl=`$as_echo "HAVE_DECL_$ac_func" | $as_tr_cpp`
> +{ $as_echo "$as_me:${as_lineno-$LINENO}: checking whether $ac_func is
> declared" >&5
> +$as_echo_n "checking whether $ac_func is declared... " >&6; }
> +if eval \${gcc_cv_have_decl_$ac_func+:} false; then :
> + $as_echo_n "(cached) " >&6
> +else
> + cat confdefs.h - <<_ACEOF >conftest.$ac_ext
> +/* end confdefs.h. */
> +#undef $ac_tr_decl
> +#define $ac_tr_decl 1
> +
> +#include "ansidecl.h"
> +#include "system.h"
> +#ifdef HAVE_MALLOC_H
> +#include
> +#endif
> +
> +
> +int
> +main ()
> +{
> +#ifndef $ac_func
> +char *(*pfn) = (char *(*)) $ac_func ;
> +#endif
> + ;
> + return 0;
> +}
> +_ACEOF
> +if ac_fn_cxx_try_compile "$LINENO"; then :
> + eval "gcc_cv_have_decl_$ac_func=yes"
> +else
> + eval "gcc_cv_have_decl_$ac_func=no"
> +fi
> +rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
> +fi
> +
> +if eval "test \"`echo '$gcc_cv_have_decl_'$ac_func`\" = yes"; then
> + { $as_echo "$as_me:${as_lineno-$LINENO}: result: yes" >&5
> +$as_echo "yes" >&6; } ; cat >>confdefs.h <<_ACEOF
> +#define $ac_tr_decl 1
> +_ACEOF
> +
> +else
> + { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
> +$as_echo "no" >&6; } ; cat >>confdefs.h <<_ACEOF
> +#define $ac_tr_decl 0
> +_ACEOF
> +
> +fi
> +
> +done
> +
> +
> cat confdefs.h - <<_ACEOF >conftest.$ac_ext
> /* end confdefs.h. */
>
> @@ -18851,7 +18918,7 @@ else
>lt_dlunknown=0; lt_dlno_uscore=1; lt_dlneed_uscore=2
>lt_status=$lt_dlunknown
>cat > conftest.$ac_ext <<_LT_EOF
> -#line 18854 "configure"
> +#line 18921 "configure"
> #include "confdefs.h"
>
> #if HAVE_DLFCN_H
> @@ -18957,7 +19024,7 @@ else
>lt_dlunknown=0; lt_dlno_uscore=1; lt_dlneed_uscore=2
>lt_status=$lt_dlunknown
>cat > conftest.$ac_ext <<_LT_EOF
> -#line 18960 "configure"
> +#line 19027 "configure"
> #include "confdefs.h"
>
> #if HAVE_DLFCN_H
>
--
Richard Biener
SUSE Software Solutions Germany GmbH, Maxfeldstrasse 5, 90409 Nuernberg,
Germany; GF: Felix Imendörffer; HRB 36809 (AG Nuernberg)
-gnu, applied.
Richard.
2019-10-24 Richard Biener
* tree-vect-slp.c (vect_get_and_check_slp_defs): For reduction
chains try harder with operand swapping and instead of
putting a shifted chain into the reduction operands put
a repetition of the final
On Wed, Oct 23, 2019 at 2:12 PM Richard Sandiford
wrote:
>
> Richard Biener writes:
> > On Wed, Oct 23, 2019 at 1:51 PM Richard Sandiford
> > wrote:
> >>
> >> Richard Biener writes:
> >> > On Wed, Oct 23, 2019 at 1:00 PM Richard Sandiford
> &
On Fri, Oct 25, 2019 at 1:52 AM Indu Bhagat wrote:
>
>
>
> On 10/11/2019 04:41 AM, Jakub Jelinek wrote:
> > On Fri, Oct 11, 2019 at 01:23:12PM +0200, Richard Biener wrote:
> >>> (coreutils-0.22)
> >>>.debug_info(D1) | .debug_abbrev(D2) | .debug_st
On Wed, Oct 23, 2019 at 11:45 PM Prathamesh Kulkarni
wrote:
>
> Hi,
> The attached patch tries to fix PR92163 by calling
> gimple_purge_dead_eh_edges from ifcvt_local_dce if we need eh cleanup.
> Does it look OK ?
Hmm. I think it shows an issue with the return value of remove_stmt_form_eh_lp
whi
...
Seeing Richard using STMT_VINFO_REDUC_IDX more I'm committing this now.
Bootstrapped and tested on x86_64-unknown-linux-gnu, applied.
Richard.
2019-10-25 Richard Biener
* tree-vect-loop.c (vectorizable_reduction): Verify
STMT_VINFO_REDUC_IDX on the to be vectorized
On Thu, Oct 24, 2019 at 9:29 AM Richard Sandiford
wrote:
>
> Now that vectorizable_operation vectorises most loop stmts involved
> in a reduction, it needs to be aware of reductions in fully-masked loops.
> The LOOP_VINFO_CAN_FULLY_MASK_P parts of vectorizable_reduction now only
> apply to cases t
On Thu, Oct 24, 2019 at 9:45 AM Richard Sandiford
wrote:
>
> Bernhard Reutner-Fischer writes:
> > On 23 October 2019 13:16:19 CEST, Richard Sandiford
> > wrote:
> >
> >>+++ gcc/config/gcn/gcn.c 2019-10-23 12:13:54.091122156 +0100
> >>@@ -3786,8 +3786,7 @@ gcn_expand_builtin (tree exp, rtx
On Wed, Oct 23, 2019 at 1:10 PM Richard Sandiford
wrote:
>
> Callers of vect_halve_mask_nunits and vect_double_mask_nunits
> already know what mode the resulting vector type should have,
> so we might as well create the vector type directly with that mode,
> just like build_vector_type_for_mode le
On Wed, Oct 23, 2019 at 1:13 PM Richard Sandiford
wrote:
>
> build_same_sized_truth_vector_type was confusingly named, since for
> SVE and AVX512 the returned vector isn't the same byte size (although
> it does have the same number of elements). What it really returns
> is the "truth" vector type
-gnu.
Richard.
2019-10-25 Richard Biener
PR tree-optimization/9
* tree-vect-slp.c (_slp_oprnd_info::first_pattern): Remove.
(_slp_oprnd_info::second_pattern): Likewise.
(_slp_oprnd_info::any_pattern): New.
(vect_create_oprnd_info): Ad
When I remove the swapping of operands in the IL I didn't actually
relax swapping restrictions.
Bootstrapped on x86_64-unknown-linux-gnu, testing in progress.
Richard.
2019-10-25 Richard Biener
* tree-vect-slp.c (vect_get_and_check_slp_defs): Only fail
swapping
parts).
Bootstrap and regtest running on x86_64-unknown-linux-gnu.
Richard.
2019-10-25 Richard Biener
* tree-vect-loop.c (vect_create_epilog_for_reduction): Use
STMT_VINFO_REDUC_IDX from the actual stmt.
(vect_transform_reduction): Likewise.
(vectorizable_redu
On October 26, 2019 6:35:43 AM GMT+02:00, Alexandre Oliva
wrote:
>This was first submitted many years ago
>https://gcc.gnu.org/ml/gcc-patches/2010-10/msg02468.html
>
>The command line option -fcallgraph-info is added and makes the
>compiler generate another output file (xxx.ci) for each compilati
powerpc/pr78604.c
> +++ b/gcc/testsuite/gcc.target/powerpc/pr78604.c
> @@ -1,7 +1,7 @@
> /* { dg-do compile { target { powerpc*-*-* } } } */
> /* { dg-skip-if "" { powerpc*-*-darwin* } } */
> /* { dg-require-effective-target powerpc_p8vector_ok } */
> -/* { dg-opti
On Sun, 27 Oct 2019, Alexandre Oliva wrote:
> On Oct 26, 2019, Alexandre Oliva wrote:
>
> > E.g., the reason we gather expanded calls rather than just use
> > cgraph_edges is that the latter would dump several "calls" that are
> > builtins expanded internally by the compiler, and would NOT dump
On Sat, 26 Oct 2019, Richard Sandiford wrote:
> Richard Biener writes:
> > We have to check each operand for being in a pattern, not just the
> > first when avoiding build from scalars (we could possibly handle
> > the special case of some of them being the pattern stmt roo
On Sun, Oct 27, 2019 at 6:32 PM Jeff Law wrote:
>
> On 9/29/19 1:51 PM, Martin Sebor wrote:
> > -Wstringop-overflow detects a subset of past-the-end read and write
> > accesses by built-in functions such as memcpy and strcpy. It relies
> > on the functions' effects the knowledge of which is hardw
On Mon, 28 Oct 2019, Jiufu Guo wrote:
> Richard Biener writes:
>
> > On Fri, 25 Oct 2019, Jiufu Guo wrote:
> >
> >> Hi,
> >>
> >> In PR88760, there are a few disscussion about improve or tune unroller for
> >> targets. And we would agree to
Bootstrap and regtest running on x86_64-unknown-linux-gnu.
Richard.
2019-10-28 Richard Biener
PR tree-optimization/92252
* tree-vect-slp.c (vect_get_and_check_slp_defs): Adjust
STMT_VINFO_REDUC_IDX when swapping operands.
* gcc.dg/torture/pr92252.c: New
701 - 800 of 26740 matches
Mail list logo
