diff options
| author | David Chase <drchase@google.com> | 2024-08-28 10:34:54 -0400 |
|---|---|---|
| committer | David Chase <drchase@google.com> | 2024-10-22 17:24:49 +0000 |
| commit | a464ffda3e44c87de946cc29f719794a30e9b3f4 (patch) | |
| tree | edc66f977b4e60a3274f1b242446a7e96bb45d58 /src/runtime/debug | |
| parent | 74163c895a7c5d7ebafeaed1f4a0891d218e2704 (diff) | |
| download | go-a464ffda3e44c87de946cc29f719794a30e9b3f4.tar.xz | |
cmd/compile: tweak inlining to favor PPARAM call sites
If a function f being considered for inlining calls
one of its parameters, reduce the normal cost of that
call (57) to 17 to increase the chance that f will
be inlined and (with luck) that parameter will be
revealed as a constant function (which unblocks
escape analysis) or perhaps even be inlined.
The least-change value for that was still effective for
iter_test benchmarks was 32; however tests showed no
particular harm even when reduced as low as 7, and there
have been reports of other performance problems with
rangefunc overheads and so I picked a middling number
in hopes of warding off such reports.
Updates #69015
Change-Id: I2a525c1beffb9f88daa14caa8a622864b023675c
Reviewed-on: https://go-review.googlesource.com/c/go/+/609095
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Keith Randall <khr@google.com>
Reviewed-by: Tim King <taking@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
Diffstat (limited to 'src/runtime/debug')
0 files changed, 0 insertions, 0 deletions
