repo.or.cz
/
official-gcc.git
/
search
commit
grep
author
committer
pickaxe
?
search:
re
summary
|
log
|
graphiclog1
|
graphiclog2
|
commit
|
commitdiff
|
tree
|
refs
|
edit
|
fork
first
·
prev
·
next
Code scheduling for Cortex-A53 isn't as good as it could be. It turns out
2017-05-05
wilco
Code scheduling
for Cortex-A53 i
s
n't as good as it
.
.
.
commit
|
commitdiff
|
tree
2017-05-04
w
il
c
o
M
a
ny
s
u
p
p
orted cores u
s
e the AUTOPREFETCHER_WEAK set
t
ing
.
.
.
commit
|
commitdiff
|
tree
2017-05-04
w
i
lco
Set j
u
m
p alignment to 4 for
C
ortex
c
ores as
it reduces
.
.
.
commit
|
commitdiff
|
tree
2017-05-04
wilco
With -mcpu=generi
c
the loop ali
g
nment i
s
currently
.
.
.
commit
|
commitdiff
|
tree
2017-05-04
wilco
All cores
w
hich add a
c
pu_addrcost_table
u
se a non
.
.
.
commit
|
commitdiff
|
tree
2017-03-22
wilco
Re
c
ently we've
put a
l
ot of
e
f
fort
i
nto
improv
i
ng ifcvt
.
.
.
commit
|
commitdiff
|
tree
2017-03-22
wilco
M
a
ny supported cores im
p
lem
e
nt fu
s
ion of AES
i
n
structions
.
.
.
commit
|
commitdiff
|
tree
2017-01-18
wilco
SHA1H ins
t
ructions may be sche
d
ul
e
d after
a
SHA1
C
instruction
commit
|
commitdiff
|
tree
2017-01-17
wilco
This patch simplif
i
es th
e
handling of
EH retu
r
n
.
W
e
.
.
.
commit
|
commitdiff
|
tree
2017-01-11
wilco
My previous
change t
o
the Corte
x
-A
5
3
s
c
heduler resulted
.
.
.
commit
|
commitdiff
|
tree
2016-12-16
wilco
The negdi2 pa
t
terns for ARM and T
h
umb-2 are dupli
c
ated
.
.
.
commit
|
commitdiff
|
tree
2016-12-16
wilco
Thumb uses
a specia
l
r
e
gister allocati
o
n order
t
o
i
ncrease
.
.
.
commit
|
commitdiff
|
tree
2016-12-15
wil
c
o
Us
i
ng lea
f
_function_p
i
n
a backend is dang
e
r
o
us a
s
.
.
.
commit
|
commitdiff
|
tree
2016-12-09
wilco
Add the test
this time
.
.
.
commit
|
commitdiff
|
tree
2016-12-08
wil
c
o
This patch f
i
xes an is
s
ue in aarch64_
c
lass
i
f
y
_
address
.
.
.
commit
|
commitdiff
|
tree
2016-12-07
w
i
lco
GCC caches
the
whether a function is a le
a
f in crtl
.
.
.
commit
|
commitdiff
|
tree
2016-12-07
wilco
I
m
pro
v
e TI mode address offs
e
t
s - these may either
.
.
.
commit
|
commitdiff
|
tree
2016-11-16
wil
c
o
Looking at
P
R77308, one o
f
t
he i
s
sues
is that
t
he bswap
.
.
.
commit
|
commitdiff
|
tree
2016-11-14
w
ilco
Th
e
second
p
a
t
ch
updates the
C
ortex-A57 scheduler n
o
w
.
.
.
commit
|
commitdiff
|
tree
2016-11-14
wil
c
o
Curr
e
ntly the SBFM,
UBFM and B
F
M instructio
n
s all use
.
.
.
commit
|
commitdiff
|
tree
2016-11-14
wilco
The
e
xis
t
ing v
e
ctor cost
s
s
t
op
s
ome b
e
ne
f
i
cial
v
ectorization
.
.
.
commit
|
commitdiff
|
tree
2016-11-02
wilco
A
d
d LE/BE
S
HA1H pa
t
terns with a V2S
I
inp
u
t
.
T
his
a
voids
.
.
.
commit
|
commitdiff
|
tree
2016-11-02
w
i
lco
Th
e
a
d
d expan
d
e
r still contains
s
om
e
ex
p
ansion
c
o
d
e
.
.
.
commit
|
commitdiff
|
tree
2016-10-25
w
ilco
With -fpu=n
e
on DI mode shif
t
s are expanded after reload
.
.
.
commit
|
commitdiff
|
tree
2016-10-21
wil
c
o
R
e
a
lly add the test this tim
e
.
.
.
commit
|
commitdiff
|
tree
2016-10-21
wilco
Improve stack adj
u
stment b
y
reu
s
i
ng a t
e
mporary m
o
ve
.
.
.
commit
|
commitdiff
|
tree
2016-10-21
wilco
If the nu
m
ber of integer
c
allee-sa
v
es is odd, any F
P
.
.
.
commit
|
commitdiff
|
tree
2016-10-05
wilco
Move a
l
l existing strchr an
d
strrchr fo
l
ding from
b
u
i
ltins
.
.
.
commit
|
commitdiff
|
tree
2016-09-28
w
i
lco
If strchr
can't
b
e folded in
g
imple-fold, we s
t
ill
.
.
.
commit
|
commitdiff
|
tree
2016-09-28
w
i
lco
Optim
i
ze strchr
(s, 0) to
s
+
strlen
(s
)
.
str
c
hr
.
.
.
commit
|
commitdiff
|
tree
2016-09-07
wil
c
o
Improve aarch6
4
_legit
i
mize_address -
a
v
o
id splitting
.
.
.
commit
|
commitdiff
|
tree
2016-09-01
w
i
lco
This patch add
s
legiti
m
ize
_
addre
s
s_displacement hook
.
.
.
commit
|
commitdiff
|
tree
2016-08-01
wilco
Th
i
s p
a
tch optimizes the prolog and epilog c
o
de to
.
.
.
commit
|
commitdiff
|
tree
2016-07-28
wilco
On
AArch64
th
e
UXTB and
U
XTH i
n
st
r
u
c
t
ions are alia
s
es
.
.
.
commit
|
commitdiff
|
tree
2016-07-28
wilco
This patchset improves zero e
x
tend costs and code generation
.
commit
|
commitdiff
|
tree
2016-07-28
wilco
This pat
c
h improves the re
a
dability of th
e
prolog and
.
.
.
commit
|
commitdiff
|
tree
2016-07-06
w
il
c
o
This patch improves the accu
r
acy of the Cortex-
A
53
.
.
.
commit
|
commitdiff
|
tree
2016-07-06
wilc
o
Fix prot
o
typ
e
in vst1Q_la
n
eu64-
1
.
c
t
o unsigned char*
.
commit
|
commitdiff
|
tree
2016-06-30
wil
c
o
This pat
c
h set
s
t
he br
a
nch
cost to the same m
o
st optimal
.
.
.
commit
|
commitdiff
|
tree
2016-06-29
w
i
lc
o
Increase loop alignment on Cortex
cores to 8 an
d
set
.
.
.
commit
|
commitdiff
|
tree
2016-06-28
wil
c
o
This patch fixes a
b
u
g
in the bswa
p
pas
s
.
In big-
e
ndian
.
.
.
commit
|
commitdiff
|
tree
2016-06-21
wilco
The recently added gcc
.
t
arget/
a
a
r
c
h64/a
d
v
s
imd
-
i
ntrinsi
c
s
.
.
.
commit
|
commitdiff
|
tree
2016-06-21
w
ilco
F
ix tree-ssa/att
r
-hotcold-2
.
c failures now
t
hat the
.
.
.
commit
|
commitdiff
|
tree
2016-06-21
wilco
Due to re
c
e
n
t improvements to the
v
ectoriz
e
r,
t
h
e
n
umb
e
r
.
.
.
commit
|
commitdiff
|
tree
2016-06-20
wil
c
o
This patch clean
s
u
p
the -mp
c
-
r
e
l
ativ
e
-loads opt
i
on
.
.
.
commit
|
commitdiff
|
tree
2016-06-20
wilco
Improv
e
modes_tieable by returni
n
g
true in more cases
.
.
.
commit
|
commitdiff
|
tree
2016-06-20
wi
l
co
T
h
e
Cort
e
x-A57 scheduler is missing
f
csel, so add it
.
commit
|
commitdiff
|
tree
2016-05-27
wilco
Remove aarch64_can
n
o
t_change_mode_class
a
s
the underlying
.
.
.
commit
|
commitdiff
|
tree
2016-05-26
w
i
l
co
GCC expand
s
switch
s
tatements
i
n
a v
e
ry simplis
t
ic
.
.
.
commit
|
commitdiff
|
tree
2016-05-26
w
i
lco
SIM
D
operation
s
like combine prefe
r
to have their
oper
a
nds
.
.
.
commit
|
commitdiff
|
tree
2016-05-16
w
ilco
Update d
o
cumentati
o
n of AArch
6
4
option
s
f
or GCC6
t
o
.
.
.
commit
|
commitdiff
|
tree
2016-05-16
w
ilco
Some pattern
s
ar
e
u
sing '%w2'
for
immediate operands
.
.
.
commit
|
commitdiff
|
tree
2016-05-16
w
i
lc
o
This patch fixes t
h
e attrib
u
t
e
s of
inte
g
er imme
d
i
a
te
.
.
.
commit
|
commitdiff
|
tree
2016-02-17
wi
l
co
A
s part o
f
t
he final clean
u
p of aar
c
h64
_
in
t
ernal_mov_immedia
.
.
.
commit
|
commitdiff
|
tree
2016-02-10
wilco
Enable i
n
struction
f
usion of
d
epe
n
dent AESE; AES
M
C
.
.
.
commit
|
commitdiff
|
tree
2016-02-10
wilco
Enable instr
u
ction fusion of AE
S
instruc
t
ions on ARM
.
.
.
commit
|
commitdiff
|
tree
2016-02-04
wilco
This
patch fixe
s
an expon
e
ntial i
s
sue in ccm
p
.
c
.
Wh
e
n
.
.
.
commit
|
commitdiff
|
tree
2016-02-03
w
i
lco
F
i
x the cc
m
p_1
.
c tes
t
back to use
'0' as regula
r
exp
r
e
s
s
ion
s
.
.
.
commit
|
commitdiff
|
tree
2016-02-02
wilco
I
m
prove TARGET_IRA_CHAN
G
E
_
PSEUDO_AL
L
OCNO_CLASS targ
e
t
.
.
.
commit
|
commitdiff
|
tree
2016-02-02
wilco
Th
i
s
patch a
d
ds su
p
port for the TARGE
T
_IRA_CHANGE_PSEUDO
_
ALL
.
.
.
commit
|
commitdiff
|
tree
2016-01-28
wilco
A rec
e
nt change added UNSPEC to the CCMP pat
t
e
r
ns t
o
.
.
.
commit
|
commitdiff
|
tree
2016-01-28
wilco
Add su
p
port for
v
e
c
tor permute cost since various permutes
.
.
.
commit
|
commitdiff
|
tree
2016-01-28
wil
c
o
Several
instru
c
ti
o
ns disassemble a zero immediate as
.
.
.
commit
|
commitdiff
|
tree
2016-01-27
wil
c
o
2015-01-27 Wi
l
co
D
ijk
s
tra <wdijkstr@arm
.
com>
commit
|
commitdiff
|
tree