Re: [PATCH] PR88497 - Extend reassoc for vector bit_field_ref

on 2019/3/8 下午6:57, Richard Biener wrote: > On Fri, Mar 8, 2019 at 2:40 AM Kewen.Lin wrote: >> >> Hi, >> >> As PR88497 (https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88497), >> when we meet some code pattern like: >> >>V1[0] + V1[1] + ... + V1[k] + V2[0] + ... + V2[k] + ... Vn[k] >>//

[PATCH, i386]: PR 68924, improve *vec_extractv2di_0_sse for 32bit targets

This patch implements additional suggestion from PR 68924. The patch avoids memory for DImode extracts from V2DImode for SSE4.1 targets. 2019-03-10 Uroš Bizjak PR target/68924 * config/i386/sse.md (*vec_extractv2di_0_sse): Add (=r,x) alternative and corresponding splitter.

[PATCH, PR d/87824] Committed fix for failing test gdc.test/runnable/nulltype.d

This patch merges the D front-end implementation with dmd upstream fcc235e8e, fixing the failing nulltype.d test with -m32 on x86_64-linux. Bootstrapped and regression tested on x86_64-linux-gnu with RUNTESTFLAGS="--target_board=unix/\{,-m32\}". Committed to trunk as r269561. -- Iain --- diff

[PATCH] [PATCH/AARCH64] Use neon_move instead of move_logic in some places

From: Andrew Pinski Hi, In some places in the aarch64 backend, neon_logic is used for mov v0.16b, v1.16b mov v0., v1. This patch moves them over to use neon_move instead. For most cores, this does not matter but for OcteonTX2 core it does matter. OK? Bootstrapped and tested on

Re: [patch, fortran] Fix PR 87673, rejects-valid

Am 10.03.19 um 22:12 schrieb Thomas Koenig: You're probably right, I will fix this.  Ugh... OK, so here is the updated test case, without the copyright stuff. OK for trunk now? :-) Regards Thomas ! { dg-do compile } ! PR 87673 - used to cause errors about non-pure functions.

[PATCH] Fix in -std=c++2a mode (PR libstdc++/89641)

Hi! The following patch fixes in make -j32 -k check 'RUNTESTFLAGS=--target_board=unix/-std=gnu++2a' -FAIL: 29_atomics/atomic/69301.cc (test for excess errors) -FAIL: 29_atomics/atomic/requirements/compare_exchange_lowering.cc (test for excess errors) -FAIL: 29_atomics/atomic/70766.cc (test for

[PR fortran/87045, patch] - pointer to array of character

The code in the testcase died due to a run-time bounds-check generated slightly too early, leading to a crash for deferred character length. Moving the character length check after the assignment solves the issue. Regtests cleanly on x86_64-pc-linux-gnu. OK for trunk? Thanks, Harald

Re: [patch, fortran] Fix PR 87673, rejects-valid

Am 10.03.19 um 21:15 schrieb Steve Kargl: On Sun, Mar 10, 2019 at 09:04:18PM +0100, Thomas Koenig wrote: ! { dg-do compile } ! PR 87673 - used to cause errors about non-pure functions. ! A fragment from Richard Townsend's iso_varying_string, as modified by ! Ian Harvey to include F2003

Re: [patch, fortran] Fix PR 87673, rejects-valid

On Sun, Mar 10, 2019 at 09:04:18PM +0100, Thomas Koenig wrote: > ! { dg-do compile } > ! PR 87673 - used to cause errors about non-pure functions. > ! A fragment from Richard Townsend's iso_varying_string, as modified by > ! Ian Harvey to include F2003 language capabilities. For copyright

[patch, fortran] Fix PR 87673, rejects-valid

Hello world, the attached patch fixes a 7/8/9 regression by removing the call to gfc_resovle_expr during matching. It turned out that gfc_resolve_expr depends on some conditions that are not yet met during the matching phase. Fortunately, the bug that this originally fixed (PR 82049) has been

[PATCH/AARCH64] Fix zero_extendsidi2_aarch64 type attribute

Hi, "uxtw x0, w1" is an alias for "mov w0, w1" but currently the back-end marks it as extend type rather than mov_reg. This patch fixes that. For most schedule models, this does not matter; I am adding one where mov (both 32bit and 64bit register mov) can be considered as zero latency in some

[PATCH, PR d/89016] Committed updated fix for ICE in dmd/expression.c

Hi, This patch merges the D front-end implementation with dmd upstream da26db819, backporting a supplementary fix for ICE on importing modules marked as deprecated, but with a non-string message. Bootstrapped and regression tested on x86_64-linux-gnu. Committed to trunk as r269560. -- Iain

Re: [Patch, fortran] PRs 89363 and 89364 - wrinkles with assumed rank

Hi Paul, (ii) PR89364 points out that ubound and shape should be returning -1 for the final dimension and not 0, for assumed rank entities argument associated with assumed size arrays. What a mix of features... Bootstrapped and regtested on FC29/x86_64 - OK for trunk? OK. Thanks for the

Re: [testsuite] Mark gdc.dg/pr89041.d as compile test

On Sun, 10 Mar 2019 at 17:52, Rainer Orth wrote: > > The new gdc.dg/pr89041.d FAILs for the non-default multilib (e.g. on > i386-pc-solaris2.11 and x86_64-pc-linux-gnu): > > +FAIL: gdc.dg/pr89041.d -O0 (test for excess errors) > +UNRESOLVED: gdc.dg/pr89041.d -O0 compilation failed to

[testsuite] Mark gdc.dg/pr89041.d as compile test

The new gdc.dg/pr89041.d FAILs for the non-default multilib (e.g. on i386-pc-solaris2.11 and x86_64-pc-linux-gnu): +FAIL: gdc.dg/pr89041.d -O0 (test for excess errors) +UNRESOLVED: gdc.dg/pr89041.d -O0 compilation failed to produce executable Excess errors: Undefined

[testsuite, i386] Compile gcc.target/i386/indirect-thunk-extern-7.c with -fjump-tables

gcc.target/i386/indirect-thunk-extern-7.c currently FAILs on every x86 target: +FAIL: gcc.target/i386/indirect-thunk-extern-7.c scan-assembler jmp[ \\t]*__x86_indirect_thunk(_nt|)_(r|e)ax on 32 and 64-bit i386-pc-solaris2.11, on x86_64-pc-linux-gnu also +FAIL:

Re: [PATCH] Fix PR89618

> This fixes a missed vectorization because loop_version (and in the end > copy_loop_info) didn't copy IVDEP info (safelen) during if-conversion > versioning. > > Bootstrap & regtest running on x86_64-unknown-linux-gnu. > > Even though this isn't a regression I'd like to fix this for GCC 9, > it

[PATCH PR d/88958] Committed fix for ICE in walk_aliased_vdefs_1

Hi, This patch merges the D front-end implementation with dmd upstream 0fc786f49, fixing PR d/88958. Bootstrapped and regression tested on x86_64-linux-gnu. Committed to trunk as r269557. -- Iain --- diff --git a/gcc/d/dmd/MERGE b/gcc/d/dmd/MERGE index 3f416dbfb7b..98bf8254554 100644 ---

[Patch, fortran] PRs 89363 and 89364 - wrinkles with assumed rank

The attached patch implements fixes for a couple of wrinkles with assumed rank: (i) PR89363 flagged the fact that the rank was not be set for assumed rank entities, associated with unallocated or unassociated arrays. The fix is straightforward, the work being done by

[committed] Add testcase for already fixed PR c++/89648

Hi! The following testcase reduced from webkitgtk started to be rejected with r268850 and is accepted again since r269265. The testcase isn't very similar to the committed one, so I've tested following on x86_64-linux both with trunk and with r269264 and committed to trunk as obvious.

Re: [PATCH] Use -c instead of check for sha512sum

Hi Tommy, On Sat, Mar 09, 2019 at 06:48:07AM -0500, Tommy Nguyen wrote: > Busybox doesn't support "check", which is an alias for "-c" in > coreutils. Tested on Alpine Edge and Fedora 29. > > 2019-03-09 Tommy Nguyen > > PR contrib/82704 > * contrib/download_prerequisites: Use -c

[PATCHv2] Fix not 8-byte aligned ldrd/strd on ARMv5 (PR 89544)

Hi, This patch is an update to the previous patch, which takes into account that the middle-end is not supposed to use the unaligned DI value directly which was passed in an unaligned stack slot due to the AAPCS parameter passing rules. The patch works by changing use_register_for_decl to return