Hi!
Since the x86 backend enabled V2SImode vectorization (with
TARGET_MMX_WITH_SSE), slp vectorization can kick in and emit
movq(%rdi), %xmm1
pshufd $225, %xmm1, %xmm0
movq%xmm0, (%rdi)
instead of
rolq$32, (%rdi)
we used to emit (or emit when slp vector
Hi!
As mentioned in the PR, we have 5 split passes (+ splitting during final).
split1 is before RA and is unconditional,
split2 is after RA and is gated on optimize > 0,
split3 is before sched2 and is gated on
defined(INSN_SCHEDULING) && optimize > 0 && flag_schedule_insns_after_reload
split4 is
On Sat, Feb 13, 2021 at 9:47 AM Jakub Jelinek wrote:
>
> Hi!
>
> Since the x86 backend enabled V2SImode vectorization (with
> TARGET_MMX_WITH_SSE), slp vectorization can kick in and emit
> movq(%rdi), %xmm1
> pshufd $225, %xmm1, %xmm0
> movq%xmm0, (%rdi)
> instead
Hi!
The (mod @0 (convert?@3 (power_of_two_cand@1 @2))) simplification
uses tree_nop_conversion_p (type, TREE_TYPE (@3)) condition, but I believe
it doesn't check what it was meant to check. On convert?@3
TREE_TYPE (@3) is not the type of what it has been converted from, but
what it has been conve
On Sat, Feb 13, 2021 at 09:59:57AM +0100, Uros Bizjak wrote:
> BTW: I wonder if this optimization triggers frequently, the code in
> the PR looks quite useful.
On my next bootstrap/regtest I can gather some statistics.
Jakub
Some subtargets don't provide the canonical function names as
the symbol name in C libraries, and libcalls will only work if
the builtins are patched to emit the correct library name.
For example, on NetBSD, cabsl has the symbol name __c99_cabsl,
and the patching is done via netbsd_patch_builtin.
Some subtargets don't provide the canonical function names as
the symbol name in C libraries, and libcalls will only work if
the builtins are patched to emit the correct library name.
For example, on NetBSD, cabsl has the symbol name __c99_cabsl,
and the patching is done via netbsd_patch_builtin.
I hope that writing the detailed commit message will encourage someone
with better knowledge of GCC internals to point out a better place for
this logic. I can follow through with any suggestions :)
On Sat, Feb 13, 2021 at 12:20:30PM +, Maya Rashish wrote:
> Some subtargets don't provide the c
On February 13, 2021 9:58:58 AM GMT+01:00, Jakub Jelinek
wrote:
>Hi!
>
>As mentioned in the PR, we have 5 split passes (+ splitting during
>final).
>split1 is before RA and is unconditional,
>split2 is after RA and is gated on optimize > 0,
>split3 is before sched2 and is gated on
>defined(INSN_S
On Sat, Feb 13, 2021 at 02:54:38PM +0100, Richard Biener wrote:
> Ok. But if required splitting is an IL property maybe we can see sth like
> RTL_split_insns, clear it from passes like selsched and gate split on the
> property and not being present?
I guess better would be to track it after reloa
On February 13, 2021 4:07:03 PM GMT+01:00, Jakub Jelinek
wrote:
>On Sat, Feb 13, 2021 at 02:54:38PM +0100, Richard Biener wrote:
>> Ok. But if required splitting is an IL property maybe we can see sth
>like
>> RTL_split_insns, clear it from passes like selsched and gate split on
>the
>> property
Hi, answers below. Note, do you want my resubmitted patch to be with
commit --amend, and therefore relative to "real" commits in the history,
or do you want me to chain it onto the last submission?
I have already sent to assign at gnu for the form.
I will update the commit message to call it s3 n
*PIN*
On 09.02.21 12:52, Tobias Burnus wrote:
Hi all, hi Thomas,
On 02.02.21 18:15, Tobias Burnus wrote:
I think I will do a combination: If 'identical' is true, I think I
cannot remove it. If it is false, it can be identical or nonoverlapping
– which makes sense.
Well, it turned out that co
We already have a check for riscv_shorten_memrefs in riscv_address_cost.
This adds the same check to riscv_rtx_costs. Making this work also
requires a change to riscv_compressed_lw_address_p to work before reload
by checking the offset and assuming any pseudo reg is OK. Testing shows
that this co
From: Levy Hsu
This expands sub-word loads as a zero/sign extended load, followed by
a subreg. This helps eliminate unnecessary zero/sign extend insns after
the load, particularly for volatiles, but also in some other cases.
Testing shows that it gives consistent code size decreases.
Tested wit
The problem which accessing dim= for an expression is that
when the argument is an array, dim= and expr->rank are relative
to the resulting array – but the array-ref also contains the
DIMEN_ELEMENT which does not count for dim=/rank.
OK for the trunk? (Reported against GCC 11, but I wonder whethe
The test case in PR 99074 invokes dynamic_cast with the this pointer
in a non-static member function called on a null pointer. The call
is, of course, undefined and other different circumstances would be
diagnosed by -Wnonnull. Unfortunately, in the test case, the null
pointer is the result of
Move custom macros to acinclude.m4 so we can autogenerate aclocal.m4
with aclocal. This matches every other project in the tree.
libiberty/ChangeLog:
* Makefile.in (ACLOCAL, ACLOCAL_AMFLAGS, $(srcdir)/aclocal.m4): Define.
(configure_deps): Rename to ...
(aclocal_deps): ..
On Fri, 12 Feb 2021, Patrick Palka wrote:
> Here, the problem ultimately seems to be that tsubst_copy_and_build,
> when called with empty args as we do during non-dependent expression
> folding, doesn't touch BASELINKs at all: it delegates to tsubst_copy
> which then immediately exits early due to
19 matches
Mail list logo
