cmd/compile: PGO compiler diagnostics

[prattmic]

Today, the compiler has some limited diagnostics of behavior with PGO using the -d=pgoinline=1 flag. These are really just ad-hoc logging points. We want more robust diagnostics, for reasons @josharian lays out well in #55022 (comment):

The more opaque the toolchain and its inputs are, the harder it is to (a) write safety checks that detect performance problems before they make it to production and (b) diagnose performance issues after they make it to production.

Plus (c) of making it easier for us to investigate compilers issues filed here.

Our current thinking for initial work here is to provide more complete/structured output of all optimization decisions, rather than just "decisions due to PGO", because the latter is ambiguous. e.g., if we decide to do an inline because of PGO and that affects the non-PGO heuristics further up the call tree, does that count as a "PGO" decision.

cc @cherrymui @aclements

[prattmic]

Bumping to 1.22. There is more structural work to be done.

For 1.21, PGO devirtualization does include structured JSON output with -d=pgodebug=3 detailing every single call with its type (direct, indirect interface, etc) and whether it was devirtualized. That can then be used for external analysis (e.g., with a tool like https://go.dev/cl/494717).

[gopherbot]

Change https://go.dev/cl/528400 mentions this issue: cmd/compile: add pgohash for debugging/bisecting PGO optimizations

[prattmic]

Minor change: -m differentiates PGO vs static devirtualization, but doesn't note PGO-enabled inlines (-d=pgodebug=1 is required for that).

[qinlonglong123]

In addition to structured output, we also hope to print some statistics, such as the total number of hotspot function calls in package A is xx, the actual number of inlined function calls is xx, and the hotspot function inlining rate is xx%.

[dr2chase]

I saw this and thought, "can't this be obtained from the compiler's -json output"?
My experiment is to rebuild the compiler, with and without its profile, with json logging enabled, and to extract the inline information from the json files that are generated.

go install -a -gcflags=all=-json=0,`pwd`/nopgo -pgo=off cmd/compile
go install -a -gcflags=all=-json=0,`pwd`/pgo                      cmd/compile
cd pgo
grepall "*" -i "cannotinlinecall" > ../pgo.cic
grepall "*" -i caninlinefunction > ../pgo.cif
cd ../nopgo
grepall "*" -i "cannotinlinecall" > ../nopgo.cic
grepall "*" -i caninlinefunction > ../nopgo.cif
cd ../
wc -l *.cic
   4337 nopgo.cic
   41172 pgo.cic
   45509 total
wc -l *.cif
   10309 nopgo.cif
   11051 pgo.cif
   21360 total

and then if you diff the .cif files, you'll see things like

diff -C 0 nopgo.cif pgo.cif | more
*** nopgo.cif   Tue Jan 21 16:06:35 2025
--- pgo.cif     Tue Jan 21 16:06:16 2025
***************
*** 78 ****
--- 79 ----
+ ./bufio/bufio.json:{"range":{"start":{"line":677,"character":6},"end":{"line":677,"character":6}},"severity":3,"code":"canInlineFunction","source":"go compiler","message":"cost: 206"}
***************
*** 140 ****
--- 142 ----
+ ./cmd%2Fcompile%2Finternal%2Fdwarfgen/dwarf.json:{"range":{"start":{"line":297,"character":6},"end":{"line":297,"character":6}},"severity":3,"code":"canInlineFunction","source":"go compiler","message":"cost: 239"}
***************
*** 145 ****
--- 148 ----
+ ./cmd%2Fcompile%2Finternal%2Fdwarfgen/dwarf.json:{"range":{"start":{"line":346,"character":6},"end":{"line":346,"character":6}},"severity":3,"code":"canInlineFunction","source":"go compiler","message":"cost: 105"}
***************
*** 146 ****
--- 150,151 ----
+ ./cmd%2Fcompile%2Finternal%2Fdwarfgen/dwarf.json:{"range":{"start":{"line":472,"character":6},"end":{"line":472,"character":6}},"severity":3,"code":"canInlineFunction","source":"go compiler","message":"cost: 1986"}
+ ./cmd%2Fcompile%2Finternal%2Fdwarfgen/dwarf.json:{"range":{"start":{"line":497,"character":6},"end":{"line":497,"character":6}},"severity":3,"code":"canInlineFunction","source":"go compiler","message":"cost: 1871"}
***************
*** 148 ****
--- 154,156 ----
+ ./cmd%2Fcompile%2Finternal%2Fdwarfgen/scope.json:{"range":{"start":{"line":18,"character":6},"end":{"line":18,"character":6}},"severity":3,"code":"canInlineFunction","source":"go compiler","message":"cost: 139"}
+ ./cmd%2Fcompile%2Finternal%2Fdwarfgen/scope.json:{"range":{"start":{"line":22,"character":6},"end":{"line":22,"character":6}},"severity":3,"code":"canInlineFunction","source":"go compiler","message":"cost: 93"}
+ ./cmd%2Fcompile%2Finternal%2Fdwarfgen/scope.json:{"range":{"start":{"line":23,"character":31},"end":{"line":23,"character":31}},"severity":3,"code":"canInlineFunction","source":"go compiler","message":"cost: 147"}
***************
*** 149 ****
--- 158,160 ----
+ ./cmd%2Fcompile%2Finternal%2Fdwarfgen/scope.json:{"range":{"start":{"line":23,"character":31},"end":{"line":23,"character":31}},"severity":3,"code":"canInlineFunction","source":"go compiler","message":"cost: 65"}
+ ./cmd%2Fcompile%2Finternal%2Fdwarfgen/scope.json:{"range":{"start":{"line":32,"character":6},"end":{"line":32,"character":6}},"severity":3,"code":"canInlineFunction","source":"go compiler","message":"cost: 540"}
+ ./cmd%2Fcompile%2Finternal%2Fdwarfgen/scope.json:{"range":{"start":{"line":68,"character":6},"end":{"line":68,"character":6}},"severity":3,"code":"canInlineFunction","source":"go compiler","message":"cost: 344"}
***************
*** 155 ****
--- 167,168 ----
+ ./cmd%2Fcompile%2Finternal%2Fdwarfgen/dwinl.json:{"range":{"start":{"line":29,"character":6},"end":{"line":29,"character":6}},"severity":3,"code":"canInlineFunction","source":"go compiler","message":"cost: 1545"}
+ ./cmd%2Fcompile%2Finternal%2Fdwarfgen/dwinl.json:{"range":{"start":{"line":227,"character":6},"end":{"line":227,"character":6}},"severity":3,"code":"canInlineFunction","source":"go compiler","message":"cost: 340"}

There will be more complicated changes in places, because sometimes inlining can reduce a function's cost.

note: "grepall" is

#!/bin/bash

if [ $# -lt 2 ] ; then
  echo grepall "<find-pattern>" "[grep-options]" "<grep-pattern>"
  exit 1
fi

findpat=$1
shift
echo find . -name "$findpat" -exec egrep "$@" /dev/null  "{}" ";"
find . \( -name "$findpat" -and -type f \) -exec egrep "$@" /dev/null "{}" ";"