cmd/compile: exponential behavior for deeply nested structs

[randall77]

Go version

tip

Output of go env in your module/workspace:

GO111MODULE=""
GOARCH="arm64"
GOBIN=""
GOCACHE="/Users/khr/Library/Caches/go-build"
GOENV="/Users/khr/Library/Application Support/go/env"
GOEXE=""
GOEXPERIMENT=""
GOFLAGS=""
GOHOSTARCH="arm64"
GOHOSTOS="darwin"
GOINSECURE=""
GOMODCACHE="/Users/khr/gopath/pkg/mod"
GONOPROXY=""
GONOSUMDB=""
GOOS="darwin"
GOPATH="/Users/khr/gopath"
GOPRIVATE=""
GOPROXY="https://proxy.golang.org,direct"
GOROOT="/Users/khr/go1.20.6"
GOSUMDB="sum.golang.org"
GOTMPDIR=""
GOTOOLDIR="/Users/khr/go1.20.6/pkg/tool/darwin_arm64"
GOVCS=""
GOVERSION="go1.20.6"
GCCGO="gccgo"
AR="ar"
CC="clang"
CXX="clang++"
CGO_ENABLED="1"
GOMOD="/dev/null"
GOWORK=""
CGO_CFLAGS="-O2 -g"
CGO_CPPFLAGS=""
CGO_CXXFLAGS="-O2 -g"
CGO_FFLAGS="-O2 -g"
CGO_LDFLAGS="-O2 -g"
PKG_CONFIG="pkg-config"
GOGCCFLAGS="-fPIC -arch arm64 -pthread -fno-caret-diagnostics -Qunused-arguments -fmessage-length=0 -fdebug-prefix-map=/var/folders/z9/dty110711l9cr9w3ktv1_2380000gn/T/go-build2801687852=/tmp/go-build -gno-record-gcc-switches -fno-common"

What did you do?

test.go:

package a

type S0 struct{ a, b S1 }
type S1 struct{ a, b S2 }
type S2 struct{ a, b S3 }
type S3 struct{ a, b S4 }
type S4 struct{ a, b S5 }
type S5 struct{ a, b S6 }
type S6 struct{ a, b S7 }
type S7 struct{ a, b S8 }
type S8 struct{ a, b S9 }
type S9 struct{ a, b S10 }
type S10 struct{ a, b S11 }
type S11 struct{ a, b S12 }
type S12 struct{ a, b S13 }
type S13 struct{ a, b S14 }
type S14 struct{ a, b S15 }
type S15 struct{ a, b S16 }
type S16 struct{ a, b S17 }
type S17 struct{ a, b S18 }
type S18 struct{ a, b S19 }
type S19 struct{ a, b S20 }
type S20 struct{ a, b S21 }
type S21 struct{ a, b S22 }
type S22 struct{ a, b S23 }
type S23 struct{ a, b int }

Compile with

go tool compile -cpuprofile=foo.prof ~/gowork/test.go

It takes a surprisingly long time. The pprof output points to cmd/compile/internal/types/alg.go:AlgType as the culprit. It is walking the exponential-sized object to determine its algorithm class. It should probably memoize the computation somehow so it isn't exponential.

Add more lines to the example to make it even more slower.

Found during discussion in #65495.

What did you see happen?

      flat  flat%   sum%        cum   cum%
    2330ms 28.14% 28.14%     7470ms 90.22%  cmd/compile/internal/types.AlgType
    2040ms 24.64% 52.78%     2070ms 25.00%  cmd/compile/internal/types.(*Type).Fields
    1010ms 12.20% 64.98%     3850ms 46.50%  cmd/compile/internal/types.IsPaddedField
     640ms  7.73% 72.71%      790ms  9.54%  cmd/compile/internal/types.PtrDataSize

What did you expect to see?

Fast compilation.

[gopherbot]

Change https://go.dev/cl/561875 mentions this issue: cmd/compile: caching result of AlgType

[gopherbot]

Change https://go.dev/cl/561936 mentions this issue: cmd/compile: caching result of AlgType, EqCanPanic and PtrDataSize

[randall77]

@mdempsky

[randall77]

Looking at this a bit more, the exponential behavior is in both cmd/compile/internal/types.AlgType and cmd/compile/internal/types.PtrDataSize.
However, at tip, there are two additional offenders, cmd/compile/internal/types2.(*comparer).identical and cmd/compile/internal/types2.(*Checker).validType0. Those do not appear in profiles of the test case in earlier versions of Go than tip (unlike the first two). Bisect points to https://go-review.googlesource.com/c/go/+/567976 @griesemer

[gopherbot]

Change https://go.dev/cl/571543 mentions this issue: cmd/compile: compute ptrBytes during CalcSize instead of on demand

[gopherbot]

Change https://go.dev/cl/571542 mentions this issue: cmd/compile: compute type eq/hash algorithm in CalcSize instead of on demand

[griesemer]

I will look into the validType issue again. The previously existing code was wrong, even though it helped in this case.

[griesemer]

Optimization for esoteric cases. Moving to 1.25.