cmd/compile: schedule block with predicted branch

[ctk-1998]

Related Issue(s) & Descriptions

Original pr: https://atomgit.com/openeuler/golang/pull/85, authored by Aleksey Markin.

Port the predicted-branch block layout optimization from the Kunpeng fork. Basic blocks are now laid out along the statically predicted hot path: once layout follows a likely successor, it keeps following single-successor blocks so the whole hot path stays fall-through and error handling is pushed into the cold tail.

• ir: add IfStmt.UnLikely, mutually exclusive with Likely
• walk: heuristic prediction of error handling patterns, enabled with -d=blockpredict>=2; "err == nil" is likely, "err != nil" is unlikely, and a body/else returning nil vs non-nil error hints at the hot side
• ssagen: map Likely/UnLikely to BranchLikely/BranchUnlikely
• ssa: extend layoutOrder to follow the trace through single-successor blocks after a predicted edge (-d=blockpredict>=1)
• port the two tests that came with the patch: ssa/layout_test.go (asserts the predicted-path block order) and test/errorlikely_test.go (end-to-end assembly order)
• newly add some benchmarks in test/blockpredict_test.go to quantify the benefit

Tests

Benchmarks on sg2044:

goos: linux
goarch: riscv64
pkg: cmd/compile/internal/test
                            │  default(0) │            blockpredict=1        │         blockpredict=2          │
                            │   sec/op    │   sec/op     vs base              │   sec/op     vs base              │
ErrorCheckChain/errrate-0     192.8n ± 1%   195.6n ± 1%  +1.48% (p=0.002 n=6)   192.8n ± 0%       ~ (p=0.491 n=6)
ErrorCheckChain/errrate-10    891.4n ± 0%   867.5n ± 0%  -2.67% (p=0.002 n=6)   837.4n ± 2%  -6.05% (p=0.002 n=6)
ErrorCheckChain/errrate-50    868.5n ± 1%   832.1n ± 2%  -4.20% (p=0.002 n=6)   818.8n ± 1%  -5.73% (p=0.002 n=6)
ErrorCheckChain/errrate-100   854.2n ± 1%   824.6n ± 0%  -3.47% (p=0.002 n=6)   798.6n ± 1%  -6.51% (p=0.002 n=6)
MiddlewareChain/errrate-0     88.97n ± 0%   93.81n ± 0%  +5.44% (p=0.002 n=6)   89.38n ± 0%  +0.47% (p=0.004 n=6)
MiddlewareChain/errrate-10    1.929µ ± 1%   1.939µ ± 2%       ~ (p=0.121 n=6)   1.941µ ± 1%       ~ (p=0.093 n=6)
MiddlewareChain/errrate-50    9.260µ ± 1%   9.305µ ± 1%       ~ (p=0.180 n=6)   9.282µ ± 2%       ~ (p=0.589 n=6)
MiddlewareChain/errrate-100   18.28µ ± 0%   18.44µ ± 1%  +0.91% (p=0.004 n=6)   18.43µ ± 1%       ~ (p=0.065 n=6)
NoErrorBaseline/rate-0        190.6n ± 0%   192.2n ± 1%  +0.87% (p=0.002 n=6)   191.0n ± 2%       ~ (p=0.320 n=6)
NoErrorBaseline/rate-50       241.7n ± 0%   242.7n ± 0%       ~ (p=0.128 n=6)   243.5n ± 1%  +0.77% (p=0.004 n=6)
geomean                       837.9n        837.4n       -0.06%                 824.8n       -1.56%

Checklist

• Tests were added or are not required
• Documentation was added or is not required