You may need to delete and recompile some .o files! This changes the
alignment of 8-byte values in C structs to match what Apple's gcc
does. See Apple's "32-bit PowerPC Function Calling Conventions" at
https://developer.apple.com
/library/content/documentation/DeveloperTools/Conceptual/LowLevelABI
/100-32-bit_PowerPC_Function_Calling_Conventions/32bitPowerPC.html
In the instruction list, put /* kills xer */ for sraw, srawi, subfic;
and correct the (now unused) "addi." and "lfdu".
Change MACHOPT_F from -m3 to -m2. This changes the code for 15 * i
from
slwi r3,r4,4
subfic r5,r4,0
add r3,r3,r5
to
mulli r3,r4,15
If the sequence "slwi subfic addi" takes 3 cycles and 12 bytes, and
mulli takes 3 cycles and 4 bytes, then mulli is better.
The new test rck_e.e segfaults on PowerPC unless I make some changes.
The inline code for _rck_ was wrong because it didn't allow the trap
handler to return. _sig_ forgot to push the old trap handler.
Move plat/linuxppc/libsys/trap.s to mach/powerpc/libem/trp.s and
rewrite it with simplified/extended mnemonics. Remove .trap alias for
.trp procedure. Add a missing `mtspr lr, r0` so we can return from
the trap handler. Call write() and _exit() so trp.s works with both
linuxppc and osxppc. Before, Mac OS X was wrongly using the trap.s
for Linux.
In powerpc/libem, simplify .aar4; teach .csa and .csb to raise the
trap if the default target is zero.
C programs don't need these changes. You may relink your C programs
with the changed .csa and .csb, but C code doesn't raise the trap.
Modula-2 code can raise traps, so you may want to relink your Modula-2
programs with the changed libem, but you might keep your old .o files
from Modula-2. You may need to recompile your Pascal programs (delete
old .o files from Pascal) because the Pascal compiler might use _rck_.
Telling osx386 and osxppc to build and run their own tools, not to
reuse the tools from linux386 and linuxppc. This wastes time to build
identical tools, but it removes some bogus dependencies. OS X tools
had wrongly depended on Linux descr files and aelflod; now they don't.
Discussion in https://github.com/davidgiven/ack/pull/23
No change to linuxppc and qemuppc. They continue to run ego without
any descr file.
I copied m68020.descr to powerpc.descr and changed some numbers. My
numbers are guesses; I know little about PowerPC cycle counts, and
almost nothing about ego. This powerpc.descr causes most of the
example programs to shrink in size (without descr -> with descr):
65429 -> 57237 hilo_b.osxppc -8192
36516 -> 32420 hilo_c.osxppc -4096
55782 -> 51686 hilo_mod.osxppc -4096
20096 -> 20096 hilo_p.osxppc 0
8813 -> 8813 mandelbrot_c.osxppc 0
93355 -> 89259 paranoia_c.osxppc -4096
92751 -> 84559 startrek_c.osxppc -8192
(Each file has 2 Mach segments, then a symbol table. Each segment
takes a multiple of 4096 bytes. When the code shrinks, we lose a
multiple of 4096 bytes.)
I used "ack -mosxppc -O6 -c.so" to examine the assembly code for
hilo.mod and mandelbrot.c, both without and with descr. This reveals
optimizations made only with descr, from 2 ego phases: SP (stack
pollution) and RA (register allocation). In hilo.mod, SP deletes some
instructions that remove items from the stack. These items get
removed when the function returns. In both hilo.mod and mandelbrot.c,
RA moves some values into local variables, so ncg can make them into
register variables. This shrinks code size, probably because register
variables get preserved across function calls. More values stay in
registers, and ncg emits shorter code.
I believe that the ego descr file uses (time,space) tuples but the ncg
table uses (space,time) tuples. This is confusing. Perhaps I am
wrong, and some or all tuples are backwards. My time values are the
cycle counts in latency from the MPC7450 Reference Manual (but not
including complications like "store serialization").
In powerpc.descr, I give the cost for saving and restoring registers
as if I was using chains of stw and lwz instructions. Actually ncg
uses single stmw and lmw instructions with at least 2 instructions.
The (time,space) for stmw and lmw would be much less than the
(time,space) for chains of stw and lwz. But this ignores the pipeline
of the MPC7450. The chains of stw and lwz may run faster than stmw
and lmw in the pipeline, because the throughput may be better than the
latency. By using the wrong values for (time,space), I'm trying to
tell ego that stmw and lmw are not better than chains of stw and lwz.
David Given made top for PowerPC. Copy the asopt phase (running top)
from linuxppc to osxppc.
Remove CC_ALIGN=-Vr to become compatible with Apple's gcc. Apple uses
left adjustment for bitfields; the first bitfield is on the left side
(the big end), not the right side.
Remove unused variables C_LIB and OLD_C_LIB; the file libc-ansi.a
doesn't exist.
Change MACHOPT_F from -m10 to -m3. This means to use no more than 3
adds and shifts to optimize a multiply by a constant. I pick -m3
because -m4 can use too many instructions. At -m4, the compiler
rewrites
n * 14
as
s = n << 1
(s << 3) + (0 - s)
This means (n * 16 - n * 2), but even at ack -O6, the compiler doesn't
rewrite (a + (0 - b)) as (a - b). The compiler emits 5 instructions:
2 of rlinmw for 2 left shifts, then addi to load 0 in a register, subf
to subtract from that 0, then add. These 5 instructions cost 5 cycles
on the MPC7450, using the cycle counts from mach/powerpc/ncg/table.
At -m3, (n * 14) becomes 2 instructions: addi to load 14 in a register
and mullw to multiply. This also costs 5 cycles (because mullw costs
4 cycles), but uses less space.
Also add fstat() and lstat(). I don't #define the constants for
st_mode or d_type, but I provide enough to get the block size of a
file and to list the names in a directory. Some fields of struct stat
get truncated, see XXX in plat/osx/include/sys/stat.h.
In struct dirent, the inode field might be d_ino or d_fileno. I
picked d_ino because Apple's sys/dirent.h uses d_ino (but Apple's
manual pages use d_fileno).
The system call puts the time in a pair of registers, not in the
timeval structure. Add code to move the time to the structure, so
programs see the correct time, not garbage. This fixes our example
programs that use the time as a random seed.
This preserves the name and value of every symbol. The type and other
info of a symbol might be lost. In gdb, one can now "disas main" or
"disas '.ret'" to disassemble functions by name.
Most symbols are in sections, so I also teach cvmach to emit the Mach
section headers. The entry point in plat/osx*/descr moves down to
make room for the section headers and LC_SYMTAB.
I fix some bugs in calculations of cvmach. They were wrong if ROM had
a greater alignment than TEXT, or if DATA did not start on a page
boundary. I introduce machseg[] to simplify the mess of variables in
main(). I declare most functions as static. Also, cvmach becomes the
first program to #include <object.h>.
Before this commit, the headers in plat/osx/include got installed
twice into PLATIND/osx386/include and PLATIND/osxppc/include. This
commit installs them once into PLATIND/osx/include and changes both
descr files to find them.
Several rules in lang/ depend on plat/osx386/include+headers or
plat/osxppc/include+headers. They each become a simplerule that
depends on plat/osx/include+headers.
These produce Mach-o executables for Mac OS X on Intel or PowerPC
processors. Our code generator for PowerPC (mach/powerpc) still has
bugs. Some examples seem to run, but startrek crashes. Our code
generator for Intel (mach/i386) is better.
There is a problem with job control. If you run paranoia or startrek,
then suspend the job (^Z) and resume it ('fg' in bash), then read(2)
might fail with EINTR.
The larger files in this commit are
- plat/osx/cvmach/cvmach.c
- plat/osx/libsys/brk.c
- plat/osx386/libsys/sigaction.s
- plat/osxppc/libsys/sigaction.s
