Subversion Repositories Open64

[/] [branches/] [x86_open64-4.2.4/] - Rev 3931

Rev

Go to most recent revision

Filtering Options

Clear current filter

Rev Log message Author Age Path
3241 Update version number and release notes for x86 Open64 4.2.4 release. dcoakley 1057d 17h /branches/x86_open64-4.2.4/
3238 Update version for second release candidate. dcoakley 1062d 15h /branches/x86_open64-4.2.4/
3237 Update release documents with draft versions. dcoakley 1062d 17h /branches/x86_open64-4.2.4/
3236 fix the issue with formal preg V16C8 lowering. dcoakley 1062d 20h /branches/x86_open64-4.2.4/
3235 handling test/cmp(8/16) compare op in cflow optimization, compare constant
source need sign extend.
dcoakley 1062d 20h /branches/x86_open64-4.2.4/
3234 Fix bug 15820: out of memory for application boost-1-38-0/wave

The problem is running out of memory in cfg transformation when
adding paths for conditional constant.

In function find_conditional_const, it could possibly process the
same coderep (and add its paths) for the number of BB in cfg.
The fix is to add a cr_visited bvector to record CRs that are visited.
dcoakley 1062d 20h /branches/x86_open64-4.2.4/
3233 Fix to bug bug 15773 (bug 559 in the main Open64 bugzilla data base):
__ompc_can_fork undefined when building on systems with newer versions
of GCC.

Newer versions of GCC need a function prototype for __ompc_can_fork(),
when the definition of __ompc_can_fork() contains the inline attribute.
dcoakley 1062d 20h /branches/x86_open64-4.2.4/
3224 Add release notes to install directory. dcoakley 1076d 08h /branches/x86_open64-4.2.4/
3223 Prepare to build x86 Open64 4.2.4 release candidate. dcoakley 1076d 16h /branches/x86_open64-4.2.4/
3222 Create x86 Open64 4.2.4 release branch. dcoakley 1076d 17h /branches/x86_open64-4.2.4/
3220 Update copyright notices. dcoakley 1077d 17h /branches/open64-booster/
3219 Fix bug 15816: root app: segmentation fault in ebo

The last_op in an empty block is NULL. This case should be considered
seperately.
dcoakley 1077d 17h /branches/open64-booster/
3218 This change corrects an Assert failure from lra.cxx casued by
LRA missing opportunities to use free regsiters.

The fix is to allocate a register to a live range if the live
range's last use is before the register's (live range) definition
and if the live range does not overlap any upward exposed use of
the register in the current basis block.
dcoakley 1077d 17h /branches/open64-booster/
3217 Fix bug 15358 (again): wrong register used for inline asm instruction.

A previous bug fix worked for -O2. A similar problem was found while
trying to compile ffmpeg at -O3: an input to inline assembly with the
type of "unsigned char" was being promoted to register %ecx instead of
the expected %cl.

A side effect of scalar_rename is that the type of the renamed variable
is promoted to 32 bits. The fix is to exclude inputs to inline assembly
from renaming when the type would be promoted.
dcoakley 1077d 17h /branches/open64-booster/
3216 Fix the issue with V16C8 EQ op. dcoakley 1077d 17h /branches/open64-booster/
3215 Fixes for bugs #15650 (cpuv6 bzip2 training data set problem - cross iteration
copy removal optimization was incorrectly removing stores), #15792 (polyhedron
induct problem - CG copy propagation was overly aggressive with operands that
were not involved with swapped subtract operands).
dcoakley 1077d 17h /branches/open64-booster/
3214 Bug 15798 - Regression: t262 group of Plumhall libc++(LVS) fails for
-O2 option from SVN 940
The bug is caused by some function which has complex <long double> type
as its parameter not be inlined, and the reason not being inlined is
the Stid_Opcode[] in ipa_inline.cxx is not initialized properly for
MTYPE_C10.
dcoakley 1077d 17h /branches/open64-booster/
3213 Fix the defect of memory large block management. The original implementation is tricky. At a very small possibility, it would illegally try to free something in memory block, rather than memory large block. dcoakley 1077d 17h /branches/open64-booster/
3212 Fix a free-of-unallocated-memory bug.

As stated in one of the comments, only the first node in the sequence
of blocks created by Update_Map is potentially an allocated pointer.
dcoakley 1077d 17h /branches/open64-booster/
3211 Fix Bug 15552: The EP benchmark fails with segment fault when used with
openmp option.

According to OMP specification, "variables with automatic storage
duration that are declared in a scope inside the construct are private".
Thus, WN_PRAGMA_LOCAL pragma is generated for SCLASS_AUTO ST instead of
for anonymous ST.
dcoakley 1077d 17h /branches/open64-booster/

1 2 Next >

Show All