徵求最佳化的 CFLAGS

Post by **scsi** » Mon Jan 19, 2004 11:46 am

我家裡有一台amd 2200+的機器
使用 palatis 的 CFLAGS 覺得很滿意

CFLAGS="-O3 -mcpu=athlon-xp -march=athlon-xp -mmmx -m3dnow -msse -mfpmath=sse -pipe -finline-functions -falign-jumps=4 -falign-loops=4 -falign-functions=64 -funroll-loops -fforce-addr -fomit-frame-pointer -frerun-cse-after-loop -frerun-loop-opt -fexpensive-optimizations"

但是我公司裡有一台pentinm 4 1.6G 的機器，用以下的設定

Code: Select all

CFLAGS="-O3 -mcpu=pentium4 -march=pentium4 -mfpmath=sse -msse2 -mmmx -pipe -finline-functions -falign-jumps=4 -falign-loops=4 -falign-functions=64 -funroll-loops -fforce-addr -fomit-frame-pointer -frerun-cse-after-loop -frerun-loop-opt -fexpensive-optimizations -fschedule-insns2 -falign-functions=4"

發現跑的速度好像慢了點，最慘的是

Code: Select all

xine-lib-1_rc2
xine-lib-1_rc3
xine-lib-1_rc3-r1

都無法正確的compile起來
會出現如以下的的錯誤

Code: Select all

source='cfft.c' object='cfft.lo' libtool=yes \
depfile='.deps/cfft.Plo' tmpdepfile='.deps/cfft.TPlo' \
depmode=gcc3 /bin/sh ../../depcomp \
/bin/sh ../../libtool-nofpic --mode=compile gcc -DHAVE_CONFIG_H -I. -I. -I../.. -I../.. -I../../include -I../../include -I../../src -I../../src/xine-engine -I../../src/xine-engine -I../../src/xine-utils  -I../../src/input -I../../src/input  -O3 -mcpu=pentium4 -march=pentium4 -mfpmath=sse -msse2 -mmmx -pipe -finline-functions -falign-jumps=4 -falign-loops=4 -falign-functions=64 -funroll-loops -fforce-addr -fomit-frame-pointer -frerun-cse-after-loop -frerun-loop-opt -fexpensive-optimizations -fschedule-insns2 -falign-functions=4  -Wall -DNDEBUG -D_REENTRANT -D_FILE_OFFSET_BITS=64 -DXINE_COMPILE  -Wpointer-arith -Wnested-externs -Wcast-align -Wchar-subscripts -Wmissing-declarations -Wmissing-prototypes -O3 -mcpu=pentium4 -march=pentium4 -mfpmath=sse -msse2 -mmmx -pipe -finline-functions -falign-jumps=4 -falign-loops=4 -falign-functions=64 -funroll-loops -fforce-addr -fomit-frame-pointer -frerun-cse-after-loop -frerun-loop-opt -fexpensive-optimizations -fschedule-insns2 -falign-functions=4 -mno-sse2 -mno-sse2 -mno-sse2 -c -o cfft.lo `test -f 'cfft.c' || echo './'`cfft.c
cfft.c: In function `passf2':
cfft.c:560: unable to find a register to spill in class `SSE_REGS'
cfft.c:560: this is the insn:
(insn 663 662 664 (set (subreg:SF (reg/v:DI 29 rmm0 [156]) 0)
        (minus:SF (reg:SF 21 exmm0 [301])
            (mem:SF (reg:SI 0 eax [294]) [4 S4 A32]))) 547 {*fop_sf_1_sse} (insn_list 660 (nil))
    (expr_list:REG_DEAD (reg:SF 21 exmm0 [301])
        (nil)))
cfft.c:560: confused by earlier errors, bailing out
make[4]: *** [cfft.lo] Error 1
make[4]: Leaving directory `/var/tmp/portage/xine-lib-1_rc3-r1/work/xine-lib-1-rc3a/src/libfaad'
make[3]: *** [all-recursive] Error 1

感覺似乎是組語的部份出錯
想問問大家，我這樣的設定有沒有問題，或是有沒有一樣是pentiume 4機器的網友能把你們的CFLAGS的設定給我參考嗎？謝謝

Post by **【Palatis】** » Mon Jan 19, 2004 12:45 pm

我改設定了.

Desktop 是這樣 (AthlonXP Barton 2600+ @ 1920Mhz)

Code: Select all

CFLAGS="-Os -mcpu=athlon-xp -march=athlon-xp -mmmx -m3dnow -msse -mfpmath=sse -pipe -finline-functions -falign-jumps=4 -falign-loops=4 -falign-functions=64 -funroll-loops -fforce-addr -fomit-frame-pointer -frerun-cse-after-loop -frerun-loop-opt -fexpensive-optimizations -fPIC"

這是 nb1 的 CFLAGS (pentium-mmx 300Mhz)

Code: Select all

CFLAGS="-Os -mcpu=pentium-mmx -march=pentium-mmx -mmmx -finline-functions -fforce-addr -funroll-loops -frerun-cse-after-loop -frerun-loop-opt -fomit-frame-pointer -fexpensive-optimizations -pipe -fPIC"

然後這是我 nb2 的 CFLAGS (pentium3-m 1Ghz)

Code: Select all

CFLAGS="-Os -mcpu=pentium3 -march=pentium3 -mfpmath=sse,387 -mmmx -fforce-addr -fomit-frame-pointer -funroll-loops -frerun-cse-after-loop -frerun-loop-opt -falign-functions=4 -pipe"

這是 PDA 的設定 = = (COMPAQ iPAQ H3630, StrongARM 1110 @ 206Mhz)

Code: Select all

CFLAGS="-mcpu=strongarm1110 -Os -pipe"

對了, nb2 那台 kernel 是 gentoo-dev-sources (2.6.1-gentoo-r1)

pentium4... 不知道, 手上沒有 p4 電腦, 所以沒有研究他的 CFLAGS.
說不定得去 gentoo 官方 forum 找找.

nightmare · Post by **nightmare** » Mon Jan 19, 2004 6:46 pm

我的 CFLAGS 是 "-march=pentium3 -O3 -pipe -mmmx -msse -mfpmath=sse"，xine-lib 的問題小弟也有碰到，經人指點後關掉與 mmx、sse 有關的 CFLAGS 後就可以成功編譯了，您可以試試....

另外，gcc 3.2 以下用 pentium4 參數好像會產生錯誤的 sse2 code.....

Post by **scsi** » Tue Jan 20, 2004 11:11 am

我目前參考大家的，然後再改一點

Code: Select all

CFLAGS="-O3 -mcpu=pentium4 -march=pentium4 -mfpmath=sse,387 -msse2 -mmmx -fforce-addr -fomit-frame-pointer -funroll-loops -frerun-cse-after-loop -frerun-loop-opt -falign-functions=4 -pipe"

目前xine-lib有compile成功:D
謝謝大家

wishy-washy · Post by **wishy-washy** » Thu Jan 22, 2004 4:22 am

FYI
http://forums.gentoo.org/viewtopic.php? ... ght=cflags
and
http://www.coyotegulch.com/acovea/

And here is my CFLAGS

Code: Select all

 CFLAGS="-O3 -march=pentium3 -mcpu=pentium4 -pipe -fomit-frame-pointer -fprefetch-loop-arrays -funroll-loops"

visa · Post by **visa** » Thu Feb 05, 2004 4:23 pm

æœ‰PIIçš„å—Ž
å°

zqyou · Post by **zqyou** » Tue Mar 09, 2004 8:45 pm

Mobile Pentium II 400 MHz , (Acer TravelMate 518TX)

CFLAGS="-O3 -march=pentium2 -mmmx -fforce-addr -funroll-loops -frerun-loop-opt -fomit-frame-pointer -pipe -fPIC"

visa · Post by **visa** » Thu Mar 18, 2004 11:21 pm

æžœç„¶å¿«å¤šäº† THX

Si · Post by Si » Sun Mar 21, 2004 7:05 pm

天啊！這麼多時間加了那麼多 flag。那些 flag 我都還沒看有什麼用處......
會比用

Code: Select all

CFLAGS="-march=pentium3 -O3 -pipe -mfpmath=sse -msse -mmmx"

這麼點快很多嗎？