Commit 6464e5dc authored by Cherry Zhang's avatar Cherry Zhang

cmd/compile: do not fold offset into load/store for args on ARM64

Args may be not at 8-byte aligned offset to SP. When the stack
frame is large, folding the offset of args may cause large
unaligned offsets that does not fit in a machine instruction on
ARM64. Therefore disable folding offsets for args.

This has small performance impact (see below). A better fix would
be letting the assembler backend fix up the offset by loading it
into a register if it doesn't fit into an instruction. And the
compiler can simply generate large load/stores with offset. Since
in most of the cases the offset is aligned or the stack frame is
small, it can fit in an instruction and no fixup is needed. But
this is too complicated for Go 1.8.

name                     old time/op    new time/op    delta
BinaryTree17-8              8.30s ± 0%     8.31s ± 0%    ~     (p=0.579 n=10+10)
Fannkuch11-8                6.14s ± 0%     6.18s ± 0%  +0.53%  (p=0.000 n=9+10)
FmtFprintfEmpty-8           117ns ± 0%     117ns ± 0%    ~     (all equal)
FmtFprintfString-8          196ns ± 0%     197ns ± 0%  +0.72%  (p=0.000 n=10+10)
FmtFprintfInt-8             204ns ± 0%     205ns ± 0%  +0.49%  (p=0.000 n=9+10)
FmtFprintfIntInt-8          302ns ± 0%     307ns ± 1%  +1.46%  (p=0.000 n=10+10)
FmtFprintfPrefixedInt-8     329ns ± 2%     326ns ± 0%    ~     (p=0.083 n=10+10)
FmtFprintfFloat-8           540ns ± 0%     542ns ± 0%  +0.46%  (p=0.000 n=8+7)
FmtManyArgs-8              1.20µs ± 1%    1.19µs ± 1%  -1.02%  (p=0.000 n=10+10)
GobDecode-8                17.3ms ± 1%    17.8ms ± 0%  +2.75%  (p=0.000 n=10+7)
GobEncode-8                15.3ms ± 1%    15.4ms ± 0%  +0.57%  (p=0.004 n=9+10)
Gzip-8                      789ms ± 0%     803ms ± 0%  +1.78%  (p=0.000 n=9+10)
Gunzip-8                    128ms ± 0%     130ms ± 0%  +1.73%  (p=0.000 n=10+9)
HTTPClientServer-8          202µs ± 6%     201µs ±10%    ~     (p=0.739 n=10+10)
JSONEncode-8               42.0ms ± 0%    42.1ms ± 0%  +0.19%  (p=0.028 n=10+9)
JSONDecode-8                159ms ± 0%     161ms ± 0%  +1.05%  (p=0.000 n=9+10)
Mandelbrot200-8            10.1ms ± 0%    10.1ms ± 0%  -0.07%  (p=0.000 n=10+9)
GoParse-8                  8.46ms ± 1%    8.61ms ± 1%  +1.77%  (p=0.000 n=10+10)
RegexpMatchEasy0_32-8       227ns ± 1%     226ns ± 0%  -0.35%  (p=0.001 n=10+9)
RegexpMatchEasy0_1K-8      1.63µs ± 0%    1.63µs ± 0%  -0.13%  (p=0.000 n=10+9)
RegexpMatchEasy1_32-8       250ns ± 0%     249ns ± 0%  -0.40%  (p=0.001 n=8+9)
RegexpMatchEasy1_1K-8      2.07µs ± 0%    2.08µs ± 0%  +0.05%  (p=0.027 n=9+9)
RegexpMatchMedium_32-8      350ns ± 0%     350ns ± 0%    ~     (p=0.412 n=9+8)
RegexpMatchMedium_1K-8      104µs ± 0%     104µs ± 0%  +0.31%  (p=0.000 n=10+7)
RegexpMatchHard_32-8       5.82µs ± 0%    5.82µs ± 0%    ~     (p=0.937 n=9+9)
RegexpMatchHard_1K-8        176µs ± 0%     176µs ± 0%  +0.03%  (p=0.000 n=9+8)
Revcomp-8                   1.36s ± 1%     1.37s ± 1%    ~     (p=0.218 n=10+10)
Template-8                  151ms ± 1%     156ms ± 1%  +3.21%  (p=0.000 n=10+10)
TimeParse-8                 737ns ± 0%     758ns ± 2%  +2.74%  (p=0.000 n=10+10)
TimeFormat-8                801ns ± 2%     789ns ± 1%  -1.51%  (p=0.000 n=10+10)
[Geo mean]                  142µs          143µs       +0.50%

Fixes #19137.

Change-Id: Ib8a21ea98c0ffb2d282a586535b213cc163e1b67
Reviewed-on: https://go-review.googlesource.com/37251
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: 's avatarDavid Chase <drchase@google.com>
parent 17405803
......@@ -20,3 +20,16 @@ func f(b [6]byte) T {
_ = x
return T{b: b}
}
// Arg symbol's base address may be not at an aligned offset to
// SP. Folding arg's address into load/store may cause odd offset.
func move(a, b [20]byte) [20]byte {
var x [1000]int // a large stack frame
_ = x
return b // b is not 8-byte aligned to SP
}
func zero() ([20]byte, [20]byte) {
var x [1000]int // a large stack frame
_ = x
return [20]byte{}, [20]byte{} // the second return value is not 8-byte aligned to SP
}
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment