Uses the same implementation as runtime/internal/atomic.
Reorganize the intrinsic detector to make it more table-driven.
Also works on amd64p32.
Change-Id: I7a5238951d6018d7d5d1bc01f339f6ee9282b2d0
Reviewed-on: https://go-review.googlesource.com/28076
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Atomic swap, add/and/or, compare and swap.
Also works on amd64p32.
Change-Id: Idf2d8f3e1255f71deba759e6e75e293afe4ab2ba
Reviewed-on: https://go-review.googlesource.com/27813
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Add the following optimizations:
- fold constants
- fold address into load/store
- simplify extensions and conditional branches
- remove nil checks
Turn on SSA on MIPS64 by default, and toggle the tests.
Fixes#16359.
Change-Id: I7f1e38c2509e22e42cd024e712990ebbe47176bd
Reviewed-on: https://go-review.googlesource.com/27870
Run-TryBot: Cherry Zhang <cherryyz@google.com>
Reviewed-by: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Inline atomic reads and writes on amd64. There's no reason
to pay the overhead of a call for these.
To keep atomic loads from being reordered, we make them
return a <value,memory> tuple.
Change the meaning of resultInArg0 for tuple-generating ops
to mean the first part of the result tuple, not the second.
This means we can always put the store part of the tuple last,
matching how arguments are laid out. This requires reordering
the outputs of add32carry and sub32carry and their descendents
in various architectures.
benchmark old ns/op new ns/op delta
BenchmarkAtomicLoad64-8 2.09 0.26 -87.56%
BenchmarkAtomicStore64-8 7.54 5.72 -24.14%
TBD (in a different CL): Cas, Or8, ...
Change-Id: I713ea88e7da3026c44ea5bdb56ed094b20bc5207
Reviewed-on: https://go-review.googlesource.com/27641
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Now that we have ops that can return 2 results, have BSF return a result
and flags. We can then get rid of the redundant comparison and use CMOV
instead of CMOVconst ops.
Get rid of a bunch of the ops we don't use. Ctz{8,16}, plus all the Clzs,
and CMOVNEs. I don't think we'll ever use them, and they would be easy
to add back if needed.
Change-Id: I8858a1d017903474ea7e4002fc76a6a86e7bd487
Reviewed-on: https://go-review.googlesource.com/27630
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
This time with the cherry-pick from the proper patch of
the old CL.
Stack size increased.
Corrected NaN-comparison glitches.
Marked g register as clobbered by calls.
Fixed shared libraries.
live_ssa.go still disabled because of differences.
Presumably turning on more optimization will fix
both the stack size and the live_ssa.go glitches.
Enhanced debugging output for shared libs test.
Rebased onto master.
Updates #16010.
Change-Id: I40864faf1ef32c118fb141b7ef8e854498e6b2c4
Reviewed-on: https://go-review.googlesource.com/27159
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
- Use machine instructions for uint64<->float conversions
- Do not enforce alignment on Zero/Move
ARM64 supports unaligned load/stores, but only aligned offset
or small offset can be encoded into instructions.
- Do combined loads
Change-Id: Iffca7dd0f13070b17b784861ce5a30af584680eb
Reviewed-on: https://go-review.googlesource.com/27086
Reviewed-by: David Chase <drchase@google.com>
When T is a scalar, there are no runtime calls
required, which makes this a clear win.
encoding/binary:
WriteInts-8 958ns ± 3% 864ns ± 2% -9.80% (p=0.000 n=15+15)
This also considerably shrinks a core fmt
routine:
Before: "".(*pp).printArg t=1 size=3952 args=0x20 locals=0xf0
After: "".(*pp).printArg t=1 size=2624 args=0x20 locals=0x98
Unfortunately, I find it very hard to get stable
numbers out of the fmt benchmarks due to thermal scaling.
Change-Id: I1278006b030253bf8e48dc7631d18985cdaa143d
Reviewed-on: https://go-review.googlesource.com/26659
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Keith Randall <khr@golang.org>
autopkg == localpkg, so it appears to be a remnant of earlier code.
Change-Id: I65b6c074535e877317cbf9f1f35e94890f0ebf14
Reviewed-on: https://go-review.googlesource.com/26662
Reviewed-by: Keith Randall <khr@golang.org>
Passes ssa_test.
Requires a few new instructions and some scratchpad
memory to move data between G and F registers.
Also fixed comparisons to be correct in case of NaN.
Added missing instructions for run.bash.
Removed some FP registers that are apparently "reserved"
(but that are also apparently also unused except for a
gratuitous multiplication by two when y = x+x would work
just as well).
Currently failing stack splits.
Updates #16010.
Change-Id: I73b161bfff54445d72bd7b813b1479f89fc72602
Reviewed-on: https://go-review.googlesource.com/26813
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Add more ARM64 optimizations:
- use hardware zero register when it is possible.
- use shifted ops.
The assembler supports shifted ops but not documented, nor knows
how to print it. This CL adds them.
- enable fast division.
This was disabled because it makes the old backend generate slower
code. But with SSA it generates faster code.
Turn on SSA by default, also adjust tests.
Change-Id: I7794479954c83bb65008dcb457bc1e21d7496da6
Reviewed-on: https://go-review.googlesource.com/26950
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
Last part of the 386 SSA port.
Modify the x86 backend to simulate SSE registers and
instructions with 387 registers and instructions.
The simulation isn't terribly performant, but it works,
and the old implementation wasn't very performant either.
Leaving to people who care about 387 to optimize if they want.
Turn on SSA backend for 386 by default.
Fixes#16358
Change-Id: I678fb59132620b2c47e993c1c10c4c21135f70c0
Reviewed-on: https://go-review.googlesource.com/25271
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
It's not a new backend, just a PtrSize==4 modification
of the existing AMD64 backend.
Change-Id: Icc63521a5cf4ebb379f7430ef3f070894c09afda
Reviewed-on: https://go-review.googlesource.com/25586
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
GOARCH=386 SSATEST=1 ./all.bash passes
Caveat: still needs changes to test/ files to use *_ssa.go versions. I
won't check those changes in with this CL because the builders will
complain as they don't have SSATEST=1.
Mostly minor fixes.
Implement float <-> uint32 in assembly. It seems the simplest option
for now.
GO386=387 does not work. That's why I can't make SSA the default for
386 yet.
Change-Id: Ic4d4402104d32bcfb1fd612f5bb6539f9acb8ae0
Reviewed-on: https://go-review.googlesource.com/25119
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Make tuple types and their SelectX ops fully generic.
These ops no longer need to be lowered.
Regalloc understands them and their tuple-generating arguments.
We can now have opcodes returning arbitrary pairs of results.
(And it would be easy to move to >2 results if needed.)
Update arm implementation to the new standard.
Implement just enough in 386 port to do 64-bit add.
Change-Id: I370ed5aacce219c82e1954c61d1f63af76c16f79
Reviewed-on: https://go-review.googlesource.com/24976
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
NaCl code runs in sandbox and there are restrictions for its
instruction uses
(https://developer.chrome.com/native-client/reference/sandbox_internals/arm-32-bit-sandbox).
Like the legacy backend, on NaCl,
- don't use R9, which is used as NaCl's "thread pointer".
- don't use Duff's device.
- don't use indexed load/stores.
- the assembler rewrites DIV/MOD to runtime calls, which on NaCl
clobbers R12, so R12 is marked as clobbered for DIV/MOD.
- other restrictions are satisfied by the assembler.
Enable SSA specific tests on nacl/arm, and disable non-SSA ones.
Updates #15365.
Change-Id: I9262693ec6756b89ca29d3ae4e52a96fe5403b02
Reviewed-on: https://go-review.googlesource.com/24859
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
As Josh mentioned in CL 24716, there has been requests for using SSA
for ARM. SSA can still be disabled by setting -ssa=0 for cmd/compile,
or partially enabled with GOSSAFUNC, GOSSAPKG, and GOSSAHASH.
Not enable SSA by default on NaCl, which is not supported yet.
Enable SSA-specific tests on ARM: live_ssa.go and nilptr3_ssa.go;
disable non-SSA tests: live.go, nilptr3.go, and slicepot.go.
Updates #15365.
Change-Id: Ic2ca8d166aeca8517b9d262a55e92f2130683a16
Reviewed-on: https://go-review.googlesource.com/23953
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: David Chase <drchase@google.com>
The comments were mostly duplicated; unify them.
Add a check that the required invariant holds.
Change-Id: I42fe09dcd1fac76d3c4e191f7a58c591c5ce429b
Reviewed-on: https://go-review.googlesource.com/24719
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: David Chase <drchase@google.com>
Encode the size and the alignment into AuxInt of Zero and Move ops.
On AMD64, we simply don't look at the alignment. On ARM and PPC64, we
only generate aligned stores.
Updates #15365.
Change-Id: Ifdcc205c364f67c4516b9adebfe7d50d223b6863
Reviewed-on: https://go-review.googlesource.com/24511
Reviewed-by: David Chase <drchase@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
The frontend may emit node with line number missing. In this case,
use the parent line number. Instead of changing every call site of
pushLine, do it in pushLine itself.
Fixes#16214.
Change-Id: I80390550b56e4d690fc770b01ff725b892ffd6dc
Reviewed-on: https://go-review.googlesource.com/24641
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
We had ~30 one way, and these four new occurrences the other way.
Updates #11626
Change-Id: Ic6403dc4905874916ae292ff739d33482ed8e5bf
Reviewed-on: https://go-review.googlesource.com/24683
Reviewed-by: Rob Pike <r@golang.org>
This adds the initial SSA implementation for PPC64.
Builds golang and all.bash runs correctly. Simple hello.go
builds but does not run.
Change-Id: I7cec211b934cd7a2dd75a6cdfaf9f71867063466
Reviewed-on: https://go-review.googlesource.com/24453
Reviewed-by: David Chase <drchase@google.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
This will be used verbatim in other architectures.
Change-Id: I307891ae597d797fd45f296b6a38ffe9fac6b975
Reviewed-on: https://go-review.googlesource.com/24155
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
This will be needed by other architectures as well.
Put a cleaner encapsulation around it.
Change-Id: I0ac25d600378042b2233301678e9d037e20701d8
Reviewed-on: https://go-review.googlesource.com/24154
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
SSA treats SP as constant throughout a function, so as OffPtr [off] SP.
When the stack moves, spilled OffPtr values become invalid, if they are
not pointer-typed.
(Currently it is fine because of the optimization rules that folds OffPtr
into Load/Store. But it'd better be "optimization", not requirement.)
Updates #15365.
Change-Id: I76cf4008dfdc169e1cb5a55a2605b6678efc915d
Reviewed-on: https://go-review.googlesource.com/23941
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
This matches the behavior of the legacy backend.
Fixes#15975 (if this is the intended behavior)
Change-Id: Id277959069b8b8bf9958fa8f2cbc762c752a1a19
Reviewed-on: https://go-review.googlesource.com/23820
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Make sure auto names don't conflict with function names. Before this CL,
we confused name a.len (the len field of the slice a) with a.len (the function
len declared on a).
Fixes#15961
Change-Id: I14913de697b521fb35db9a1b10ba201f25d552bb
Reviewed-on: https://go-review.googlesource.com/23789
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Machine supports (or the runtime simulates in soft float mode)
(u)int32<->float conversions. The frontend rewrites int64<->float
conversions to call to runtime function.
For int64->float32 conversion, the frontend generates
. . AS u(100) l(10) tc(1)
. . . NAME-main.~r1 u(1) a(true) g(1) l(9) x(8+0) class(PPARAMOUT) f(1) float32
. . . CALLFUNC u(100) l(10) tc(1) float32
. . . . NAME-runtime.int64tofloat64 u(1) a(true) x(0+0) class(PFUNC) tc(1) used(true) FUNC-func(int64) float64
The CALLFUNC node has type float32, whereas runtime.int64tofloat64
returns float64. The legacy backend implicitly makes a float64->float32
conversion. The SSA backend does not do implicit conversion, so we
insert an explicit CONV here.
All cmd/compile/internal/gc/testdata/*_ssa.go tests passed.
Progress on SSA for ARM. Still not complete.
Update #15365.
Change-Id: I30937c8ff977271246b068f48224693776804339
Reviewed-on: https://go-review.googlesource.com/23652
Reviewed-by: Keith Randall <khr@golang.org>
This CL adds support of Div, Mod, Convert, GetClosurePtr and 64-bit indexing
support to SSA backend for ARM.
Add tests for 64-bit indexing to cmd/compile/internal/gc/testdata/string_ssa.go.
Tests cmd/compile/internal/gc/testdata/*_ssa.go passed, except compound_ssa.go
and fp_ssa.go.
Progress on SSA for ARM. Still not complete. Essentially the only unsupported
part is floating point.
Updates #15365.
Change-Id: I269e88b67f641c25e7a813d910c96d356d236bff
Reviewed-on: https://go-review.googlesource.com/23542
Reviewed-by: David Chase <drchase@google.com>
Introduce dec64 rules to (generically) decompose 64-bit integer on
32-bit architectures. 64-bit integer is composed/decomposed with
Int64Make/Hi/Lo ops, as for complex types.
The idea of dealing with Add64 is the following:
(Add64 (Int64Make xh xl) (Int64Make yh yl))
->
(Int64Make
(Add32withcarry xh yh (Select0 (Add32carry xl yl)))
(Select1 (Add32carry xl yl)))
where Add32carry returns a tuple (flags,uint32). Select0 and Select1
read the first and the second component of the tuple, respectively.
The two Add32carry will be CSE'd.
Similarly for multiplication, Mul32uhilo returns a tuple (hi, lo).
Also add support of KeepAlive, to fix build after merge.
Tests addressed_ssa.go, array_ssa.go, break_ssa.go, chan_ssa.go,
cmp_ssa.go, ctl_ssa.go, map_ssa.go, and string_ssa.go in
cmd/compile/internal/gc/testdata passed.
Progress on SSA for ARM. Still not complete.
Updates #15365.
Change-Id: I7867c76785a456312de5d8398a6b3f7ca5a4f7ec
Reviewed-on: https://go-review.googlesource.com/23213
Reviewed-by: Keith Randall <khr@golang.org>
When we do *p = f(), we might need to copy the return value from
f to p with a write barrier. The write barrier itself is a call,
so we need to copy the return value of f to a temporary location
before we call the write barrier function. Otherwise, the call
itself (specifically, marshalling the args to typedmemmove) will
clobber the value we're trying to write.
Fixes#15854
Change-Id: I5703da87634d91a9884e3ec098d7b3af713462e7
Reviewed-on: https://go-review.googlesource.com/23522
Reviewed-by: David Chase <drchase@google.com>
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Also fix argument offset for runtime calls.
Also fix LoadReg/StoreReg by generating instructions by type.
Progress on SSA backend for ARM. Still not complete.
Tests append_ssa.go, assert_ssa.go, loadstore_ssa.go, short_ssa.go, and
deferNoReturn.go in cmd/compile/internal/gc/testdata passed.
Updates #15365.
Change-Id: I0f0a2398cab8bbb461772a55241a16a7da2ecedf
Reviewed-on: https://go-review.googlesource.com/23212
Reviewed-by: David Chase <drchase@google.com>
As in the elimination of PHEAP|PPARAM in CL 23393,
this is something the front end can trivially take care of
and then not bother the back ends with.
It also eliminates some suspect (and only lightly exercised)
code paths in the back ends.
I don't have a smoking gun for this one but it seems
more clearly correct.
Change-Id: I3b3f5e669b3b81d091ff1e2fb13226a6f14c69d5
Reviewed-on: https://go-review.googlesource.com/23431
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
The liveness computation of parameters generally was never
correct, but forcing all parameters to be live throughout the
function covered up that problem. The new SSA back end is
too clever: even though it currently keeps the parameter values live
throughout the function, it may find optimizations that mean
the current values are not written back to the original parameter
stack slots immediately or ever (for example if a parameter is set
to nil, SSA constant propagation may replace all later uses of the
parameter with a constant nil, eliminating the need to write the nil
value back to the stack slot), so the liveness code must now
track the actual operations on the stack slots, exposing these
problems.
One small problem in the handling of arguments is that nodarg
can return ONAME PPARAM nodes with adjusted offsets, so that
there are actually multiple *Node pointers for the same parameter
in the instruction stream. This might be possible to correct, but
not in this CL. For now, we fix this by using n.Orig instead of n
when considering PPARAM and PPARAMOUT nodes.
The major problem in the handling of arguments is general
confusion in the liveness code about the meaning of PPARAM|PHEAP
and PPARAMOUT|PHEAP nodes, especially as contrasted with PAUTO|PHEAP.
The difference between these two is that when a local variable "moves"
to the heap, it's really just allocated there to start with; in contrast,
when an argument moves to the heap, the actual data has to be copied
there from the stack at the beginning of the function, and when a
result "moves" to the heap the value in the heap has to be copied
back to the stack when the function returns
This general confusion is also present in the SSA back end.
The PHEAP bit worked decently when I first introduced it 7 years ago (!)
in 391425ae. The back end did nothing sophisticated, and in particular
there was no analysis at all: no escape analysis, no liveness analysis,
and certainly no SSA back end. But the complications caused in the
various downstream consumers suggest that this should be a detail
kept mainly in the front end.
This CL therefore eliminates both the PHEAP bit and even the idea of
"heap variables" from the back ends.
First, it replaces the PPARAM|PHEAP, PPARAMOUT|PHEAP, and PAUTO|PHEAP
variable classes with the single PAUTOHEAP, a pseudo-class indicating
a variable maintained on the heap and available by indirecting a
local variable kept on the stack (a plain PAUTO).
Second, walkexpr replaces all references to PAUTOHEAP variables
with indirections of the corresponding PAUTO variable.
The back ends and the liveness code now just see plain indirected
variables. This may actually produce better code, but the real goal
here is to eliminate these little-used and somewhat suspect code
paths in the back end analyses.
The OPARAM node type goes away too.
A followup CL will do the same to PPARAMREF. I'm not sure that
the back ends (SSA in particular) are handling those right either,
and with the framework established in this CL that change is trivial
and the result clearly more correct.
Fixes#15747.
Change-Id: I2770b1ce3cbc93981bfc7166be66a9da12013d74
Reviewed-on: https://go-review.googlesource.com/23393
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Fix hardcoded flag register mask in ssa/flagalloc.go by auto-generating
the mask.
Also fix a mistake (in previous CL) about conditional branches.
Progress on SSA backend for ARM. Still not complete. Now "container/ring"
package compiles and tests passed.
Updates #15365.
Change-Id: Id7c8805c30dbb8107baedb485ed0f71f59ed6ea8
Reviewed-on: https://go-review.googlesource.com/23093
Reviewed-by: Keith Randall <khr@golang.org>
Introduce a KeepAlive op which makes sure that its argument is kept
live until the KeepAlive. Use KeepAlive to mark pointer input
arguments as live after each function call and at each return.
We do this change only for pointer arguments. Those are the
critical ones to handle because they might have finalizers.
Doing compound arguments (slices, structs, ...) is more complicated
because we would need to track field liveness individually (we do
that for auto variables now, but inputs requires extra trickery).
Turn off the automatic marking of args as live. That way, when args
are explicitly nulled, plive will know that the original argument is
dead.
The KeepAlive op will be the eventual implementation of
runtime.KeepAlive.
Fixes#15277
Change-Id: I5f223e65d99c9f8342c03fbb1512c4d363e903e5
Reviewed-on: https://go-review.googlesource.com/22365
Reviewed-by: David Chase <drchase@google.com>
Reviewed-by: Russ Cox <rsc@golang.org>
This adds a sparse method for locating nearest ancestors
in a dominator tree, and checks blocks with more than one
predecessor for differences and inserts phi functions where
there are.
Uses reversed post order to cut number of passes, running
it from first def to last use ("last use" for paramout and
mem is end-of-program; last use for a phi input from a
backedge is the source of the back edge)
Includes a cutover from old algorithm to new to avoid paying
large constant factor for small programs. This keeps normal
builds running at about the same time, while not running
over-long on large machine-generated inputs.
Add "phase" flags for ssa/build -- ssa/build/stats prints
number of blocks, values (before and after linking references
and inserting phis, so expansion can be measured), and their
product; the product governs the cutover, where a good value
seems to be somewhere between 1 and 5 million.
Among the files compiled by make.bash, this is the shape of
the tail of the distribution for #blocks, #vars, and their
product:
#blocks #vars product
max 6171 28180 173,898,780
99.9% 1641 6548 10,401,878
99% 463 1909 873,721
95% 152 639 95,235
90% 84 359 30,021
The old algorithm is indeed usually fastest, for 99%ile
values of usually.
The fix to LookupVarOutgoing
( https://go-review.googlesource.com/#/c/22790/ )
deals with some of the same problems addressed by this CL,
but on at least one bug ( #15537 ) this change is still
a significant help.
With this CL:
/tmp/gopath$ rm -rf pkg bin
/tmp/gopath$ time go get -v -gcflags -memprofile=y.mprof \
github.com/gogo/protobuf/test/theproto3/combos/...
...
real 4m35.200s
user 13m16.644s
sys 0m36.712s
and pprof reports 3.4GB allocated in one of the larger profiles
With tip:
/tmp/gopath$ rm -rf pkg bin
/tmp/gopath$ time go get -v -gcflags -memprofile=y.mprof \
github.com/gogo/protobuf/test/theproto3/combos/...
...
real 10m36.569s
user 25m52.286s
sys 4m3.696s
and pprof reports 8.3GB allocated in the same larger profile
With this CL, most of the compilation time on the benchmarked
input is spent in register/stack allocation (cumulative 53%)
and in the sparse lookup algorithm itself (cumulative 20%).
Fixes#15537.
Change-Id: Ia0299dda6a291534d8b08e5f9883216ded677a00
Reviewed-on: https://go-review.googlesource.com/22342
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Progress on SSA backend for ARM. Still not complete. Now "helloworld"
function compiles and runs.
Updates #15365.
Change-Id: I02f66983cefdf07a6aed262fb4af8add464d8e9a
Reviewed-on: https://go-review.googlesource.com/22854
Reviewed-by: Keith Randall <khr@golang.org>
If b has exactly one predecessor, as happens
frequently with static calls, we can make
lookupVarOutgoing generate less garbage.
Instead of generating a value that is just
going to be an OpCopy and then get eliminated,
loop. This can lead to lots of looping.
However, this loop is way cheaper than generating
lots of ssa.Values and then eliminating them.
For a subset of the code in #15537:
Before:
28.31 real 36.17 user 1.68 sys
2282450944 maximum resident set size
After:
9.63 real 11.66 user 0.51 sys
638144512 maximum resident set size
Updates #15537.
Excitingly, it appears that this also helps
regular code:
name old time/op new time/op delta
Template 288ms ± 6% 276ms ± 7% -4.13% (p=0.000 n=21+24)
Unicode 143ms ± 8% 141ms ±10% ~ (p=0.287 n=24+25)
GoTypes 932ms ± 4% 874ms ± 4% -6.20% (p=0.000 n=23+22)
Compiler 4.89s ± 4% 4.58s ± 4% -6.46% (p=0.000 n=22+23)
MakeBash 40.2s ±13% 39.8s ± 9% ~ (p=0.648 n=23+23)
name old user-ns/op new user-ns/op delta
Template 388user-ms ±10% 373user-ms ± 5% -3.80% (p=0.000 n=24+25)
Unicode 203user-ms ± 6% 202user-ms ± 7% ~ (p=0.492 n=22+24)
GoTypes 1.29user-s ± 4% 1.17user-s ± 4% -9.67% (p=0.000 n=25+23)
Compiler 6.86user-s ± 5% 6.28user-s ± 4% -8.49% (p=0.000 n=25+25)
name old alloc/op new alloc/op delta
Template 51.5MB ± 0% 47.6MB ± 0% -7.47% (p=0.000 n=22+25)
Unicode 37.2MB ± 0% 37.1MB ± 0% -0.21% (p=0.000 n=25+25)
GoTypes 166MB ± 0% 138MB ± 0% -16.83% (p=0.000 n=25+25)
Compiler 756MB ± 0% 628MB ± 0% -16.96% (p=0.000 n=25+23)
name old allocs/op new allocs/op delta
Template 450k ± 0% 445k ± 0% -1.02% (p=0.000 n=25+25)
Unicode 356k ± 0% 356k ± 0% ~ (p=0.374 n=24+25)
GoTypes 1.31M ± 0% 1.25M ± 0% -4.18% (p=0.000 n=25+25)
Compiler 5.29M ± 0% 5.02M ± 0% -5.15% (p=0.000 n=25+23)
It also seems to help in other cases in which
phi insertion is a pain point (#14774, #14934).
Change-Id: Ibd05ed7b99d262117ece7bb250dfa8c3d1cc5dd2
Reviewed-on: https://go-review.googlesource.com/22790
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Turn SSAing of variables off when compiling with optimizations off.
This helps keep variable names around that would otherwise be
optimized away.
Fixes#14744
Change-Id: I31db8cf269c068c7c5851808f13e5955a09810ca
Reviewed-on: https://go-review.googlesource.com/22681
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
