build system: use a bit more aggressive gcc/ld optimizations,

mostly related to ELF section ans padding, not code gen.
Resulting reductions in size (only biggest ones are shown)
uClibc.t4 - new, uClibc.t3 - old:

  15673     244      92   16009    3e89 uClibc.t4-stdcfg_so/lib/ld-uClibc-0.9.29.so
  15673     244      92   16009    3e89 uClibc.t4-stdcfg_so/lib/ld-uClibc.so
  15673     244      92   16009    3e89 uClibc.t4-stdcfg_so/lib/ld-uClibc.so.0
  15678     244      92   16014    3e8e uClibc.t3-stdcfg_so/lib/ld-uClibc-0.9.29.so
  15678     244      92   16014    3e8e uClibc.t3-stdcfg_so/lib/ld-uClibc.so
  15678     244      92   16014    3e8e uClibc.t3-stdcfg_so/lib/ld-uClibc.so.0
  39910     200       4   40114    9cb2 uClibc.t4-stdcfg_so/lib/libm-0.9.29.so
  39910     200       4   40114    9cb2 uClibc.t4-stdcfg_so/lib/libm.so
  39910     200       4   40114    9cb2 uClibc.t4-stdcfg_so/lib/libm.so.0
  40179     200       4   40383    9dbf uClibc.t3-stdcfg_so/lib/libm-0.9.29.so
  40179     200       4   40383    9dbf uClibc.t3-stdcfg_so/lib/libm.so
  40179     200       4   40383    9dbf uClibc.t3-stdcfg_so/lib/libm.so.0
 234104    1472    5980  241556   3af94 uClibc.t4-stdcfg_so/lib/libc.so.0
 234104    1472    5980  241556   3af94 uClibc.t4-stdcfg_so/lib/libuClibc-0.9.29.so
 235319    1472    5992  242783   3b45f uClibc.t3-stdcfg_so/lib/libc.so.0
 235319    1472    5992  242783   3b45f uClibc.t3-stdcfg_so/lib/libuClibc-0.9.29.so

diff --git a/Rules.mak b/Rules.mak
index 94285cc..794d2b5 100644 (file)
--- a/Rules.mak
+++ b/Rules.mak
@@ -138,6 +138,7 @@ check_ld=$(shell \
 
 ARFLAGS:=cr
 
+# Flags in OPTIMIZATION are used only for non-debug builds
 OPTIMIZATION:=
 # Use '-Os' optimization if available, else use -O2, allow Config to override
 OPTIMIZATION+=$(call check_gcc,-Os,-O2)
@@ -167,8 +168,53 @@ PIEFLAG_NAME:=-fPIE
 
 # Some nice CPU specific optimizations
 ifeq ($(TARGET_ARCH),i386)
+       OPTIMIZATION+=$(call check_gcc,-fomit-frame-pointer,)
+
+       # NB: this may make SSE insns segfault!
+       # -O1 -march=pentium3, -Os -msse etc are known to be affected.
+       # TODO: conditionally bump to 4
+       # (see http://gcc.gnu.org/bugzilla/show_bug.cgi?id=13685)
        OPTIMIZATION+=$(call check_gcc,-mpreferred-stack-boundary=2,)
-       OPTIMIZATION+=$(call check_gcc,-falign-jumps=0 -falign-loops=0,-malign-jumps=0 -malign-loops=0)
+
+       # Choice of alignment (please document why!)
+       #  -falign-labels: in-line labels
+       #  (reachable by normal code flow, aligning will insert nops
+       #  which will be executed - may even make things slower)
+       #  -falign-jumps: reachable only by a jump
+       # Generic: no alignment at all (smallest code)
+       GCC_FALIGN=$(call check_gcc,-falign-functions=1 -falign-jumps=1 -falign-labels=1 -falign-loops=1,-malign-jumps=1 -malign-loops=1)
+ifeq ($(CONFIG_K7),y)
+       # Align functions to four bytes, use default for labels and loops (why?)
+       GCC_FALIGN=$(call check_gcc,-falign-functions=4 -falign-jumps=1,-malign-functions=4 -falign-jumps=1)
+endif
+ifeq ($(CONFIG_CRUSOE),y)
+       # Use compiler's default for functions, labels and loops (why?)
+       GCC_FALIGN=$(call check_gcc,-falign-functions=0 -falign-jumps=1,-malign-functions=0 -falign-jumps=1)
+endif
+ifeq ($(CONFIG_CYRIXIII),y)
+       # Use compiler's default for functions, labels and loops (why?)
+       GCC_FALIGN=$(call check_gcc,-falign-functions=0 -falign-jumps=1,-malign-functions=0 -falign-jumps=1)
+endif
+       OPTIMIZATION+=$(GCC_FALIGN)
+
+       # Putting each function and data object into its own section
+       # allows for kbytes of less text if users link against static uclibc
+       # using ld --gc-sections.
+       # ld 2.18 can't do that (yet?) for shared libraries, so we itself
+       # do not use --gc-sections at shared lib link time.
+       # However, in combination with sections being sorted by alignment
+       # it does result in much reduced padding:
+       #   text    data     bss     dec     hex
+       # 235319    1472    5992  242783   3b45f old.so
+       # 234104    1472    5980  241556   3af94 new.so
+       # Without -ffunction-sections, all functions will get aligned
+       # to 4 byte boundary by as/ld. This is arguably a bug in as.
+       # It specifies 4 byte align for .text even if not told to do so:
+       # Idx Name          Size      VMA       LMA       File off  Algn
+       #   0 .text         xxxxxxxx  00000000  00000000  xxxxxxxx  2**2 <===!
+       CPU_CFLAGS-y  += $(call check_gcc,-ffunction-sections -fdata-sections,)
+       CPU_LDFLAGS-y += -Wl,--sort-common -Wl,--sort-section -Wl,alignment
+
        CPU_LDFLAGS-y+=-m32
        CPU_CFLAGS-y+=-m32
        CPU_CFLAGS-$(CONFIG_386)+=-march=i386
@@ -181,11 +227,11 @@ ifeq ($(TARGET_ARCH),i386)
        CPU_CFLAGS-$(CONFIG_PENTIUMIII)+=$(call check_gcc,-march=pentium3,-march=i686)
        CPU_CFLAGS-$(CONFIG_PENTIUM4)+=$(call check_gcc,-march=pentium4,-march=i686)
        CPU_CFLAGS-$(CONFIG_K6)+=$(call check_gcc,-march=k6,-march=i586)
-       CPU_CFLAGS-$(CONFIG_K7)+=$(call check_gcc,-march=athlon,-march=i686) $(call check_gcc,-falign-functions=4,-malign-functions=4)
-       CPU_CFLAGS-$(CONFIG_CRUSOE)+=-march=i686 $(call check_gcc,-falign-functions=0,-malign-functions=0)
+       CPU_CFLAGS-$(CONFIG_K7)+=$(call check_gcc,-march=athlon,-march=i686)
+       CPU_CFLAGS-$(CONFIG_CRUSOE)+=-march=i686
        CPU_CFLAGS-$(CONFIG_WINCHIPC6)+=$(call check_gcc,-march=winchip-c6,-march=i586)
        CPU_CFLAGS-$(CONFIG_WINCHIP2)+=$(call check_gcc,-march=winchip2,-march=i586)
-       CPU_CFLAGS-$(CONFIG_CYRIXIII)+=$(call check_gcc,-march=c3,-march=i486) $(call check_gcc,-falign-functions=0,-malign-functions=0)
+       CPU_CFLAGS-$(CONFIG_CYRIXIII)+=$(call check_gcc,-march=c3,-march=i486)
        CPU_CFLAGS-$(CONFIG_NEHEMIAH)+=$(call check_gcc,-march=c3-2,-march=i686)
 endif