This includes the following information in the per-function summary:
outK = paramJ encoded in outK bits for paramJ
outK = *paramJ encoded in outK bits for paramJ
heap = paramJ EscHeap
heap = *paramJ EscContentEscapes
Note that (currently) if the address of a parameter is taken and
returned, necessarily a heap allocation occurred to contain that
reference, and the heap can never refer to stack, therefore the
parameter and everything downstream from it escapes to the heap.
The per-function summary information now has a tuneable number of bits
(2 is probably noticeably better than 1, 3 is likely overkill, but it
is now easy to check and the -m debugging output includes information
that allows you to figure out if more would be better.)
A new test was added to check pointer flow through struct-typed and
*struct-typed parameters and returns; some of these are sensitive to
the number of summary bits, and ought to yield better results with a
more competent escape analysis algorithm. Another new test checks
(some) correctness with array parameters, results, and operations.
The old analysis inferred a piece of plan9 runtime was non-escaping by
counteracting overconservative analysis with buggy analysis; with the
bug fixed, the result was too conservative (and it's not easy to fix
in this framework) so the source code was tweaked to get the desired
result. A test was added against the discovered bug.
The escape analysis was further improved splitting the "level" into
3 parts, one tracking the conventional "level" and the other two
computing the highest-level-suffix-from-copy, which is used to
generally model the cancelling effect of indirection applied to
address-of.
With the improved escape analysis enabled, it was necessary to
modify one of the runtime tests because it now attempts to allocate
too much on the (small, fixed-size) G0 (system) stack and this
failed the test.
Compiling src/std after touching src/runtime/*.go with -m logging
turned on shows 420 fewer heap allocation sites (10538 vs 10968).
Profiling allocations in src/html/template with
for i in {1..5} ;
do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go;
go tool pprof -alloc_objects -text mastx.${i}.prof ;
done
showed a 15% reduction in allocations performed by the compiler.
Update #3753
Update #4720Fixes#10466
Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432
Reviewed-on: https://go-review.googlesource.com/8202
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
Val is used to hold constant values.
Reg was the odd duck out.
Generated using eg.
No functional changes. Passes toolstash -cmp.
Change-Id: Ic1de769a1f92bb02e09a4428d998b716f307e2f6
Reviewed-on: https://go-review.googlesource.com/8912
Reviewed-by: Russ Cox <rsc@golang.org>
Convert Embedded, Method, and Colas to bools.
I believe that this is the last of the Node fields
that can be trivially converted to bools.
No functional changes. Passes toolstash -cmp.
Change-Id: I81962ee47866596341fc60d24d6959c20cd7fc1c
Reviewed-on: https://go-review.googlesource.com/8440
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
I first prototyped this change in Sept 2011, and I discarded it
because it made no difference in the obvious benchmark loop.
It still makes no difference in the obvious benchmark loop,
but in a less obvious one, doing some extra computation
around the calls to Sqrt, not making the call does have a
significant effect.
benchmark old ns/op new ns/op delta
BenchmarkSqrt 4.56 4.57 +0.22%
BenchmarkSqrtIndirect 4.56 4.56 +0.00%
BenchmarkSqrtGo 69.4 69.4 +0.00%
BenchmarkSqrtPrime 4417 3647 -17.43%
This is a warmup for using hardware expansions for some
calls to 1-line assembly routines in the runtime (for example getg).
Change-Id: Ie66be23f8c09d0f7dc4ddd7ca8a93cfce28f55a4
Reviewed-on: https://go-review.googlesource.com/8356
Reviewed-by: Rob Pike <r@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
This is a follow-up to CL 7360.
It was generated with eg and gofmt -r.
The only manual changes are the unembedding in syntax.go
and backporting changes from y.go to go.y.
Passes toolstash -cmp.
Change-Id: I3d6d06ecb659809a4bc8592395d5b9a18967218e
Reviewed-on: https://go-review.googlesource.com/8053
Reviewed-by: Russ Cox <rsc@golang.org>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Nodes dominate gc's memory usage, but many fields are only used
for a subset of kinds of nodes. This change pulls out fields
used only for func-like Nodes. This reduces the size of the
Node struct on a 64-bit machine from 504 bytes to 416 bytes (-17%).
Compiling the runtime, 1.5% of nodes have a non-nil Func.
In html/template, 2.7% of nodes have a non-nil Func.
This change introduces an extra alloc and associated GC overhead
when Func is non-nil. However, when Func is nil, as it almost
always is, it spares the garbage collector scanning some Node fields.
Empirically, this change appears to be roughly neutral with regard to GC.
To keep the diff readable, this CL uses an embedded Func field.
A subsequent CL will unembed the field.
Passes toolstash -cmp.
Change-Id: Ide86aa954b097fb8e6154f0811d3691497477004
Reviewed-on: https://go-review.googlesource.com/7360
Reviewed-by: Russ Cox <rsc@golang.org>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
This CL moves the bulk of the code that has been copy-and-pasted
since the initial 386 port back into a shared place, cutting 5 copies to 1.
The motivation here is not cleanup per se but instead to reduce the
cost of introducing changes in shared concepts like regalloc or general
expression evaluation. For example, a change after this one will
implement x.(*T) without a call into the runtime. This CL makes that
followup work 5x easier.
The single copy still has more special cases for architecture details
than I'd like, but having them called out explicitly like this at least
opens the door to generalizing the conditions and smoothing out
the distinctions in the future.
This is a LARGE CL. I started by trying to pull in one function at a time
in a sequence of CLs and it became clear that everything was so
interrelated that it had to be moved as a whole. Apologies for the size.
It is not clear how many more releases this code will matter for;
eventually it will be replaced by Keith's SSA work. But as noted above,
the deduplication was necessary to reduce the cost of working on
the current code while we have it.
Passes tests on amd64, 386, arm, and ppc64le.
Can build arm64 binaries but not tested there.
Being able to build binaries means it is probably very close.
Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861
Reviewed-on: https://go-review.googlesource.com/7853
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Change-Id: I3096a7497955bc475739739ee23be387e9162867
Reviewed-on: https://go-review.googlesource.com/7210
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
More cleanups to gc.Node
- make Node.Local a boolean
- make Type.Local a boolean
- reduce the size of Node.Esc to a uint8
Reducing the size of Node.Esc shaves ~45mb off the RSS compiling cmd/internal/gc on amd64
before:
Maximum resident set size (kbytes): 659496
after:
Maximum resident set size (kbytes): 612196
- declare gc.Funcdepth as int32
- declare Node.Funcdepth as int32
In both cases, these were previously machine specific int types. This doesn't result in
any memory saving at the moment due to struct padding.
Change-Id: Iabef8da15e962fe8b79d7fd3d402fb26ce7ec31c
Reviewed-on: https://go-review.googlesource.com/7261
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Dave Cheney <dave@cheney.net>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Convert Node.Isddd to a boolean and simplify usage.
- Node.Isddd converted to bool
- Type.Isddd converted to bool
- mkinlcall converted to take isddd as a bool
- typecheckaste converted to take isddd as a bool
- ascompatte converted to take isddd as a bool
Change-Id: I52586145619c44182bb0c2c5d80a0a3fe3e50a07
Reviewed-on: https://go-review.googlesource.com/7172
Reviewed-by: Keith Randall <khr@golang.org>
More Node cleanups, these ones touch go.y.
- convert Node.Implicit to bool
- convert Node.Used to bool
Change-Id: I85c7ff9e66cee7122b560adedc995166c874f2f2
Reviewed-on: https://go-review.googlesource.com/7124
Reviewed-by: Keith Randall <khr@golang.org>
- make paramoutheap return a bool
- convert Node.Assigned to a bool
- convert Node.Captured to a bool
- convert Node.Byval to a bool
- convert Node.Dupok to a bool
- convert Node.Wrapper to a bool
- convert Node.Reslice to a bool
Change-Id: I5b57c019f936c31d53db4db14459fb2b0aa72305
Reviewed-on: https://go-review.googlesource.com/7030
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Dave Cheney <dave@cheney.net>
Node.Addrtaken is treated as a bool, so make it a bool.
I'll start to batch these changes if they are simple.
Change-Id: I02a3d1131efc4e12b78b83372c1b50f8b160c194
Reviewed-on: https://go-review.googlesource.com/6911
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Node.Hasbreak was treated like a bool, so declare it as bool.
Change-Id: Ied238356dce4da896834bd1412cc21ea56d35e1d
Reviewed-on: https://go-review.googlesource.com/6807
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Convert Node.Readonly to a bool.
Change-Id: Ide9f6f657f498d70d7b9544a38046325d7c82dc8
Reviewed-on: https://go-review.googlesource.com/6809
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Node.Builtin was occasionally set to 1, but never read.
Change-Id: Ia8a76bccc79b0f211296d50bd419860b13077ba5
Reviewed-on: https://go-review.googlesource.com/6808
Reviewed-by: Ian Lance Taylor <iant@golang.org>
It appears that c2go dropped comments inside struct { ... } and enum { ... }.
Restore them.
Identified missing comments by checking for comments present
in the C code but not the Go code, made a list, and then reapplied
with some mechanical help.
Missing comment finder: http://play.golang.org/p/g6qNUAo1Y0
Change-Id: I323ab45c7ef9d51e28eab3b699eb14bee1eef66b
Reviewed-on: https://go-review.googlesource.com/6899
Reviewed-by: Rob Pike <r@golang.org>
Minor comments added. More to come.
Change-Id: I97511db54d59e1009ef934da38f306a2dc83a6e9
Reviewed-on: https://go-review.googlesource.com/6898
Reviewed-by: Rob Pike <r@golang.org>
