5a: add SQRTF and SQRTD
5l: add ASQRTF and ASQRTD
Use ARMv7 VFP VSQRT instruction to speed up math.Sqrt
R=rsc, dave, m
CC=golang-dev
https://golang.org/cl/4551082
I started looking at this code because the nm in GNU
binutils was ignoring the first symbol in the .symtab
section. Apparently, the System V ABI reserves the
first entry and requires all fields inside to be set
to zero.
The list of changes is as follows:
· reserve the first symbol entry (as noted above)
· fix the section indices for .data and .bss symbols
· factor out common code for Elf32 and Elf64
· remove the special case for elfsymo in [568]l/asm.c:/^asmb
· add the "etext" symbol in 6l
· add static symbols
R=rsc
CC=golang-dev
https://golang.org/cl/4524075
The ld time was dominated by symbol table processing, so
* increase hash table size
* emit fewer symbols in gc (just 1 per string, 1 per type)
* add read-only lookup to avoid creating spurious symbols
* add linked list to speed whole-table traversals
Breaks dwarf generator (no idea why), so disable dwarf.
Reduces time for 6l to link godoc by 25%.
R=ken2
CC=golang-dev
https://golang.org/cl/4383047
Much of the bulk of Go binaries is the symbol tables,
which give a name to every C string, Go string,
and reflection type symbol. These names are not worth
much other than seeing what's where in a binary.
This CL deletes all those names from the symbol table,
instead aggregating the symbols into contiguous blocks
and giving them the names "string.*", "go.string.*", and "type.*".
Before:
$ 6nm $(which godoc.old) | sort | grep ' string\.' | tail -10
59eda4 D string."aa87ca22be8b05378eb1c71...
59ee08 D string."b3312fa7e23ee7e4988e056...
59ee6c D string."func(*token.FileSet, st...
59eed0 D string."func(io.Writer, []uint8...
59ef34 D string."func(*tls.Config, *tls....
59ef98 D string."func(*bool, **template....
59effc D string."method(p *printer.print...
59f060 D string."method(S *scanner.Scann...
59f12c D string."func(*struct { begin in...
59f194 D string."method(ka *tls.ecdheRSA...
$
After:
$ 6nm $(which godoc) | sort | grep ' string\.' | tail -10
5e6a30 D string.*
$
Those names in the "Before" are truncated for the CL.
In the real binary they are the complete string, up to
a certain length, or else a unique identifier.
The same applies to the type and go.string symbols.
Removing the names cuts godoc by more than half:
-rwxr-xr-x 1 rsc rsc 9153405 2011-03-07 23:19 godoc.old
-rwxr-xr-x 1 rsc rsc 4290071 2011-03-07 23:19 godoc
For what it's worth, only 80% of what's left gets loaded
into memory; the other 20% is dwarf debugging information
only ever accessed by gdb:
-rwxr-xr-x 1 rsc rsc 3397787 2011-03-07 23:19 godoc.nodwarf
R=r, cw
CC=golang-dev
https://golang.org/cl/4245072
A reference to the address of weak.foo resolves at link time
to the address of the symbol foo if foo would end up in the
binary anyway, or to zero if foo would not be in the binary.
For example:
int xxx = 1;
int yyy = 2;
int weak·xxx;
int weak·yyy;
void main·main(void) {
runtime·printf("%p %p %p\n", &xxx, &weak·xxx, &weak·yyy);
}
prints the same non-nil address twice, then 0 (because yyy is not
referenced so it was dropped from the binary).
This will be used by the reflection tables.
R=iant
CC=golang-dev
https://golang.org/cl/4223044
Fix problems found.
On amd64, various library routines had bigger
stack frames than expected, because large function
calls had been added.
runtime.assertI2T: nosplit stack overflow
120 assumed on entry to runtime.assertI2T
8 after runtime.assertI2T uses 112
0 on entry to runtime.newTypeAssertionError
-8 on entry to runtime.morestack01
runtime.assertE2E: nosplit stack overflow
120 assumed on entry to runtime.assertE2E
16 after runtime.assertE2E uses 104
8 on entry to runtime.panic
0 on entry to runtime.morestack16
-8 after runtime.morestack16 uses 8
runtime.assertE2T: nosplit stack overflow
120 assumed on entry to runtime.assertE2T
16 after runtime.assertE2T uses 104
8 on entry to runtime.panic
0 on entry to runtime.morestack16
-8 after runtime.morestack16 uses 8
runtime.newselect: nosplit stack overflow
120 assumed on entry to runtime.newselect
56 after runtime.newselect uses 64
48 on entry to runtime.printf
8 after runtime.printf uses 40
0 on entry to vprintf
-8 on entry to runtime.morestack16
runtime.selectdefault: nosplit stack overflow
120 assumed on entry to runtime.selectdefault
56 after runtime.selectdefault uses 64
48 on entry to runtime.printf
8 after runtime.printf uses 40
0 on entry to vprintf
-8 on entry to runtime.morestack16
runtime.selectgo: nosplit stack overflow
120 assumed on entry to runtime.selectgo
0 after runtime.selectgo uses 120
-8 on entry to runtime.gosched
On arm, 5c was tagging functions NOSPLIT that should
not have been, like the recursive function printpanics:
printpanics: nosplit stack overflow
124 assumed on entry to printpanics
112 after printpanics uses 12
108 on entry to printpanics
96 after printpanics uses 12
92 on entry to printpanics
80 after printpanics uses 12
76 on entry to printpanics
64 after printpanics uses 12
60 on entry to printpanics
48 after printpanics uses 12
44 on entry to printpanics
32 after printpanics uses 12
28 on entry to printpanics
16 after printpanics uses 12
12 on entry to printpanics
0 after printpanics uses 12
-4 on entry to printpanics
R=r, r2
CC=golang-dev
https://golang.org/cl/4188061
Shame on me: I fixed the same bug in 6l in 8691fcc6a66e
(https://golang.org/cl/2609041) and neglected
to look at 5l and 8l to see if they were affected.
On the positive side, the check I added in that CL is the
one that detected this bug.
Fixes#1457.
R=ken2
CC=golang-dev
https://golang.org/cl/3981052
The fault was lucky: when it wasn't faulting it was silently
copying a word from some other block and later putting
that same word back. If some other goroutine had changed
that word of memory in the interim, too bad.
The ARM code was inconsistent about whether the
"argument frame" included the saved LR. Including it made
some things more regular but mostly just caused confusion
in the places where the regularity broke. Now the rule
reflects reality: argp is always a pointer to arguments,
never a saved link register.
Renamed struct fields to make meaning clearer.
Running ARM in QEMU, package time's gotest:
* before: 27/58 failed
* after: 0/50
R=r, r2
CC=golang-dev
https://golang.org/cl/3993041
The recent linker changes broke NaCl support
a month ago, and there are no known users of it.
The NaCl code can always be recovered from the
repository history.
R=adg, r
CC=golang-dev
https://golang.org/cl/3671042
Sub-symbols are laid out inside a larger symbol
but can be addressed directly.
Use to make Mach-O pointer array not a special case.
Will use later to describe ELF sections.
Glimpses of the beginning of ELF loading.
R=ken2
CC=golang-dev
https://golang.org/cl/2623043
Load the entire archive file instead.
Reduces I/O by avoiding additional passes
through libraries to resolve symbols.
Go packages always need all the files anyway
(most often, all 1 of them).
R=ken2
CC=golang-dev
https://golang.org/cl/2613042
That is, move the pc/ln table and the symbol table
into the read-only data segment. This eliminates
the need for a special load command to map the
symbol table into memory, which makes the
information available on systems that couldn't handle
the magic load to 0x99000000, like NaCl and ARM QEMU
and Linux without config_highmem=y. It also
eliminates an #ifdef and some clumsy code to
find the symbol table on Windows.
The bad news is that the binary appears to be bigger
than it used to be. This is not actually the case, though:
the same amount of data is being mapped into memory
as before, and the tables are still read-only, so they're
still shared across multiple instances of the binary as
they were before. The difference is just that the tables
aren't squirreled away in some section that "size" doesn't
know to look at.
This is a checkpoint.
It probably breaks Windows and breaks NaCl more
than it used to be broken, but those will be fixed.
The logic involving -s needs to be revisited too.
Fixes#871.
R=ken2
CC=golang-dev
https://golang.org/cl/2587041
Because the SB is only good for 8k and Go programs
tend to have much more data than that, SB doesn't
save very much. A fmt.Printf-based hello world program
has 360 kB text segment. Removing SB makes the text
500 bytes (0.14%) longer.
R=ken2, r2, ken3
CC=golang-dev
https://golang.org/cl/2487042
Also change the span-dependent jump algorithm
to use fewer iterations:
* resolve forward jumps at their targets (comefrom list)
* mark jumps as small or big and only do small->big
* record whether a jump failed to be encodable
These changes mean that a function with only small
jumps can be laid out in a single iteration, and the
vast majority of functions take just two iterations.
I was seeing a maximum of 5 iterations before; the
max now is 3 and there are fewer that get even that far.
R=ken2
CC=golang-dev
https://golang.org/cl/2537041
Using explicit relocations internally, we can
represent the data for a particular symbol as
an initialized block of memory instead of a
linked list of ADATA instructions. The real
goal here is to be able to hand off some of the
relocations to the dynamic linker when interacting
with system libraries, but a pleasant side effect is
that the memory image is much more compact
than the ADATA list, so the linkers use less memory.
R=ken2
CC=golang-dev
https://golang.org/cl/2512041
