cmd, runtime: remove PC_F & PC_B globals on Wasm

Following the previous CL, this removes more global variables on
Wasm.

PC_B is used mostly for intra-function jumps, and for a function
telling its callee where to start or resume. This usage can be
served by a parameter. The top level loop (wasm_pc_f_loop) uses
PC_B for resuming a function. This value is either set by gogo,
or loaded from the Go stack at function return. Instead of
loading PC_B at each function return, we could make gogo stores
PC_B at the same stack location, and let the top level loop do
the load. This way, we don't need to use global PC_B to
communicate with the top level loop, and we can replace global
PC_B with a parameter.

PC_F is similar. It is even more so in that the only reader is
the top level loop. Let the top level loop read it from the stack,
and we can get rid of PC_F entirely.

PC_F and PC_B are used less entensively as SP, so this CL has
smaller performance gain.

Running on Chrome 74.0.3729.108 on Linux/AMD64,

name                   old time/op    new time/op    delta
BinaryTree17              16.6s ± 0%     16.2s ± 1%   -2.59%  (p=0.016 n=4+5)
Fannkuch11                11.1s ± 1%     10.8s ± 0%   -2.65%  (p=0.008 n=5+5)
FmtFprintfEmpty           231ns ± 1%     217ns ± 0%   -6.06%  (p=0.008 n=5+5)
FmtFprintfString          407ns ± 3%     375ns ± 2%   -7.81%  (p=0.008 n=5+5)
FmtFprintfInt             466ns ± 2%     430ns ± 0%   -7.79%  (p=0.016 n=5+4)
FmtFprintfIntInt          719ns ± 2%     673ns ± 2%   -6.37%  (p=0.008 n=5+5)
FmtFprintfPrefixedInt     706ns ± 1%     676ns ± 3%   -4.31%  (p=0.008 n=5+5)
FmtFprintfFloat          1.01µs ± 1%    0.97µs ± 1%   -4.30%  (p=0.008 n=5+5)
FmtManyArgs              2.67µs ± 1%    2.51µs ± 1%   -5.95%  (p=0.008 n=5+5)
GobDecode                30.7ms ± 9%    31.3ms ±34%     ~     (p=0.222 n=5+5)
GobEncode                24.2ms ±23%    20.2ms ± 0%  -16.36%  (p=0.016 n=5+4)
Gzip                      852ms ± 0%     823ms ± 0%   -3.38%  (p=0.016 n=4+5)
Gunzip                    160ms ± 1%     151ms ± 1%   -5.37%  (p=0.008 n=5+5)
JSONEncode               35.7ms ± 1%    34.3ms ± 1%   -3.81%  (p=0.008 n=5+5)
JSONDecode                247ms ± 8%     254ms ± 7%     ~     (p=0.548 n=5+5)
Mandelbrot200            5.39ms ± 0%    5.41ms ± 0%   +0.42%  (p=0.008 n=5+5)
GoParse                  18.5ms ± 1%    18.3ms ± 2%     ~     (p=0.343 n=4+4)
RegexpMatchEasy0_32       424ns ± 2%     397ns ± 0%   -6.23%  (p=0.008 n=5+5)
RegexpMatchEasy0_1K      2.88µs ± 0%    2.86µs ± 1%     ~     (p=0.079 n=5+5)
RegexpMatchEasy1_32       395ns ± 2%     370ns ± 1%   -6.23%  (p=0.008 n=5+5)
RegexpMatchEasy1_1K      3.26µs ± 0%    3.19µs ± 1%   -2.06%  (p=0.008 n=5+5)
RegexpMatchMedium_32      564ns ± 1%     532ns ± 0%   -5.71%  (p=0.008 n=5+5)
RegexpMatchMedium_1K      146µs ± 2%     140µs ± 1%   -4.62%  (p=0.008 n=5+5)
RegexpMatchHard_32       8.47µs ± 1%    7.91µs ± 1%   -6.65%  (p=0.008 n=5+5)
RegexpMatchHard_1K        253µs ± 1%     236µs ± 2%   -6.66%  (p=0.008 n=5+5)
Revcomp                   1.78s ± 4%     1.76s ± 5%     ~     (p=1.000 n=5+5)
Template                  292ms ±29%     269ms ± 5%     ~     (p=0.690 n=5+5)
TimeParse                1.61µs ± 4%    1.54µs ± 1%   -4.42%  (p=0.008 n=5+5)
TimeFormat               1.66µs ± 3%    1.58µs ± 1%   -5.22%  (p=0.008 n=5+5)
[Geo mean]                232µs          221µs        -4.54%

name                   old speed      new speed      delta
GobDecode              25.0MB/s ± 8%  25.1MB/s ±27%     ~     (p=0.222 n=5+5)
GobEncode              32.8MB/s ±21%  38.0MB/s ± 0%  +15.84%  (p=0.016 n=5+4)
Gzip                   22.8MB/s ± 0%  23.6MB/s ± 0%   +3.49%  (p=0.016 n=4+5)
Gunzip                  121MB/s ± 1%   128MB/s ± 1%   +5.68%  (p=0.008 n=5+5)
JSONEncode             54.4MB/s ± 1%  56.5MB/s ± 1%   +3.97%  (p=0.008 n=5+5)
JSONDecode             7.88MB/s ± 8%  7.65MB/s ± 8%     ~     (p=0.548 n=5+5)
GoParse                3.07MB/s ± 8%  3.00MB/s ±22%     ~     (p=0.579 n=5+5)
RegexpMatchEasy0_32    75.6MB/s ± 2%  80.5MB/s ± 0%   +6.58%  (p=0.008 n=5+5)
RegexpMatchEasy0_1K     356MB/s ± 0%   358MB/s ± 1%     ~     (p=0.095 n=5+5)
RegexpMatchEasy1_32    81.1MB/s ± 2%  86.5MB/s ± 1%   +6.69%  (p=0.008 n=5+5)
RegexpMatchEasy1_1K     314MB/s ± 0%   320MB/s ± 0%   +2.10%  (p=0.008 n=5+5)
RegexpMatchMedium_32   1.77MB/s ± 1%  1.88MB/s ± 0%   +6.09%  (p=0.016 n=5+4)
RegexpMatchMedium_1K   6.99MB/s ± 2%  7.33MB/s ± 1%   +4.83%  (p=0.008 n=5+5)
RegexpMatchHard_32     3.78MB/s ± 1%  4.04MB/s ± 1%   +7.04%  (p=0.008 n=5+5)
RegexpMatchHard_1K     4.04MB/s ± 1%  4.33MB/s ± 2%   +7.17%  (p=0.008 n=5+5)
Revcomp                 143MB/s ± 4%   145MB/s ± 5%     ~     (p=1.000 n=5+5)
Template               6.77MB/s ±24%  7.22MB/s ± 5%     ~     (p=0.690 n=5+5)
[Geo mean]             24.4MB/s       25.4MB/s        +4.18%

Change-Id: Ib80716e62992aec28b2c4a96af280c278f83aa49
Reviewed-on: https://go-review.googlesource.com/c/go/+/173980
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Richard Musiol <neelance@gmail.com>
This commit is contained in:
Cherry Zhang 2019-04-25 17:25:54 -04:00
parent 7ed7669c0d
commit 496b8dbbfc
5 changed files with 145 additions and 87 deletions

View file

@ -102,10 +102,11 @@ func asmb2(ctxt *ld.Link) {
}
types := []*wasmFuncType{
// For normal Go functions the return value is
// For normal Go functions, the single parameter is PC_B,
// the return value is
// 0 if the function returned normally or
// 1 if the stack needs to be unwound.
{Results: []byte{I32}},
{Params: []byte{I32}, Results: []byte{I32}},
}
// collect host imports (functions that get imported from the WebAssembly host, usually JavaScript)
@ -320,16 +321,14 @@ func writeGlobalSec(ctxt *ld.Link) {
sizeOffset := writeSecHeader(ctxt, sectionGlobal)
globalRegs := []byte{
I32, // 0: PC_F
I32, // 1: PC_B
I32, // 2: SP
I64, // 3: CTXT
I64, // 4: g
I64, // 5: RET0
I64, // 6: RET1
I64, // 7: RET2
I64, // 8: RET3
I32, // 9: PAUSE
I32, // 0: SP
I64, // 1: CTXT
I64, // 2: g
I64, // 3: RET0
I64, // 4: RET1
I64, // 5: RET2
I64, // 6: RET3
I32, // 7: PAUSE
}
writeUleb128(ctxt.Out, uint64(len(globalRegs))) // number of globals