b310570af3
From-SVN: r4505
45 lines
2.3 KiB
Plaintext
45 lines
2.3 KiB
Plaintext
# The svr4 reference port for the i860 contains an alloca.o routine
|
|
# in /usr/ucblib/libucb.a, but we can't just try to get that by
|
|
# setting CLIB to /usr/ucblib/libucb.a because (unfortunately)
|
|
# there are a lot of other routines in libucb.a which are supposed
|
|
# to be the Berkeley versions of library routines normally found in
|
|
# libc.a and many of these Berkeley versions are badly broken. Thus,
|
|
# if we try to link programs with libucb.a before libc.a, those
|
|
# programs tend to crash.
|
|
|
|
# Also, the alloca() routine supplied in early version of svr4 for
|
|
# the i860 is non-ABI compliant. It doesn't keep the stack aligned
|
|
# to a 16-byte boundary as the ABI requires.
|
|
|
|
# More importantly however, even a fully ABI compliant alloca() routine
|
|
# would fail to work correctly with some versions of the native svr4 C
|
|
# compiler currently being distributed for the i860 (as of 1/29/92).
|
|
# The problem is that the native C compiler generates non-ABI-compliant
|
|
# function epilogues which cut back the stack (upon function exit) in
|
|
# an incorrect manner. Specifically, they cut back the stack by adding
|
|
# the nominal *static* frame size (determined statically at compile-time)
|
|
# to the stack pointer rather than setting the stack pointer based upon
|
|
# the current value of the frame pointer (as called for in the i860 ABI).
|
|
# This can cause serious trouble in cases where you repeatedly call a
|
|
# routine which itself calls alloca(). In such cases, the stack will
|
|
# grow continuously until you finally run out of swap space or exceed
|
|
# the system's process size limit. To avoid this problem (which can
|
|
# arise when a stage1 gcc is being used to build a stage2 gcc) you
|
|
# *must* link in the C language version of alloca() which is supplied
|
|
# with gcc to your stage1 version of gcc. The following definition
|
|
# forces that to happen.
|
|
|
|
ALLOCA=alloca.o
|
|
|
|
# We build all stages *without* shared libraries because that may make
|
|
# debugging the compiler easier (until there is a GDB which supports
|
|
# both Dwarf *and* svr4 shared libraries).
|
|
|
|
# Note that the native C compiler for the svr4 reference port on the
|
|
# i860 recognizes a special -gg option. Using that option causes *full*
|
|
# Dwarf debugging information to be generated, whereas using only -g
|
|
# causes only limited Dwarf debugging information to be generated.
|
|
# (This is an undocumented feature of the native svr4 C compiler.)
|
|
|
|
CCLIBFLAGS=-Bstatic -dn -gg
|