Repository navigation
249 lines (241 loc) · 11.6 KB
/
Copy pathci.yml
File metadata and controls
249 lines (241 loc) · 11.6 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
name: ci
on:
push:
branches: [main]
pull_request:
branches: [main]
permissions:
contents: read
# Split-CI coverage policy (mirroring the go-asmgen / go-fft orgs).
#
# A single `go test -coverpkg` run cannot reach 100% statement coverage over a
# SIMD-dispatched package: each arch compiles exactly one sumSIMD wrapper — the
# amd64 SSE2 wrapper (reduce_amd64.go -> sum_amd64.s), the arm64 NEON wrapper
# (reduce_arm64.go -> sum_arm64.s), the ppc64le VSX wrapper (reduce_ppc64le.go
# -> sum_ppc64le.s), the loong64 LASX wrapper (reduce_loong64.go ->
# sum_loong64.s), the riscv64 RVV wrapper (reduce_riscv64.go -> sum_riscv64.s),
# the s390x vector wrapper (reduce_s390x.go -> sum_s390x.s), or the
# scalar-alias wrapper (reduce_generic.go, on the 32-bit targets). No
# single build sees
# more than one, so no single run covers them all.
#
# The honest split:
# * The pure-Go scalar CORE plus the multicore kernels (parallel.go) are held
# to 100% statement coverage by the `test` gate. It runs on the amd64 hosts
# (ubuntu/windows) AND the arm64 host (macos); on each, exactly one sumSIMD
# wrapper compiles and IS exercised by TestSumSIMD, so the gate legitimately
# reaches 100% on every OS — nothing reachable is excluded.
# * The generated Plan 9 assembly carries no Go statements, so it cannot be
# statement-covered. It is validated by the per-arch execution jobs
# (arch-native / arch-qemu): each regenerates the committed .s with
# go-asmgen and fails if it is stale, vets it (asmdecl cross-checks every
# name+offset(FP)), builds it (cmd/asm encodes the mnemonics), and RUNS the
# suite — TestSumSIMD asserts the SIMD sum matches the scalar oracle to a
# tight relative tolerance (the lane-parallel grouping is a valid reordering,
# exactly the trade-off numpy's pairwise summation makes; bit-identity is
# not claimed for a reduction).
#
# So every line is either statement-covered by the gate or executed by a
# per-arch job; the gate is never lowered and there is no coverage-gaming knob.
# amd64/arm64/ppc64le/loong64 ship SIMD kernels (ppc64le's and loong64's were
# also run on real POWER8/POWER9 and a Loongson 3C5000L); riscv64 ships RVV
# kernels (used when the CPU reports V) and s390x vector-facility kernels (used
# when AT_HWCAP reports it). Under qemu the riscv64 lane runs with V, the
# loong64 lane on la464 (LSX+LASX) and the s390x lane on max (VX+VXE), and each
# FAILS if its CPU does not report the extension (NDARRAY_REQUIRE_RVV / _LASX /
# _VXE): until 2026-10-26 they ran qemu's default CPUs, and the riscv64 and
# loong64 lanes only ever tested the scalar fallbacks.
jobs:
test:
name: test (${{ matrix.os }})
runs-on: ${{ matrix.os }}
strategy:
fail-fast: false
matrix:
os: [ubuntu-latest, macos-latest, windows-latest]
steps:
- uses: actions/checkout@v7
- uses: actions/setup-go@v7
with:
go-version: 1.27.3
- run: go version
- run: go vet ./...
- run: go build ./...
- name: Test + 100% coverage gate
shell: bash
run: |
COVERPKG=$(go list ./... | paste -sd, -)
go test -race -coverpkg="$COVERPKG" -coverprofile=cover.out ./...
go tool cover -func=cover.out | tail -n 40
# Count the uncovered blocks: `go tool cover -func` rounds to one
# decimal, and a gate on its total passed at 99.98% with two
# statements never run.
go run github.com/go-fleettools/fleettools/cmd/covblocks@v0.27.0 cover.out
# The refbench inventory (docs/perf.md): the allocation count of every case
# equals docs/perf.json, docs/perf-table.md is what perf.json generates, and
# every public function has a case or an entry in docs/perf-exempt.md. No
# Python, no timing: timing runs by hand at a 1-minute load below 8,
# cd bench && GOMAXPROCS=1 go test -run TestPerf -cpu 1 -perf.run .
# GOMAXPROCS=1 is in the environment because go-fft/fft reads it once at init.
perf:
name: perf (refbench CI mode)
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v7
- uses: actions/setup-go@v7
with:
go-version: 1.27.3
- name: Allocation counts, table, public-function inventory
working-directory: bench
env:
GOMAXPROCS: "1"
run: go test -count=1 -run TestPerf -cpu 1 .
# Native per-arch execution. amd64/arm64 have GitHub-hosted runners, so the
# generated assembly runs natively. Each regenerates the committed .s via
# go-asmgen and fails if it is stale, then vets/builds/runs the suite —
# TestSumSIMD asserts SIMD sum == scalar oracle (to tolerance).
arch-native:
name: arch ${{ matrix.label }}
runs-on: ${{ matrix.os }}
strategy:
fail-fast: false
matrix:
include:
- { os: ubuntu-latest, label: amd64, regen: amd64 }
- { os: ubuntu-24.04-arm, label: arm64, regen: arm64 }
steps:
- uses: actions/checkout@v7
- uses: actions/setup-go@v7
with:
go-version: 1.27.3
- run: go env GOARCH
- name: Regenerate committed assembly and check it is current
run: |
( cd internal/kernels/asmgen/${{ matrix.regen }} && GOFLAGS=-mod=mod go run gen.go )
mv internal/kernels/asmgen/${{ matrix.regen }}/sum_${{ matrix.regen }}.s \
internal/kernels/sum_${{ matrix.regen }}.s
git diff --exit-code -- internal/kernels/sum_${{ matrix.regen }}.s || {
echo "::error::generated ${{ matrix.regen }} assembly is stale; run go generate and commit"; exit 1; }
- name: Vet (asmdecl) + build + run (SIMD == scalar)
run: |
go vet ./...
go build ./...
go test ./...
# qemu-user execution for the targets without a hosted runner, plus 32-bit
# arm. It runs their SIMD kernels where qemu emulates the extension: RVV on
# riscv64, LSX/LASX on loong64, VSX on ppc64le, and the vector facility with
# its z14 enhancements on s390x. It is also the endianness and word-size
# guard.
arch-qemu:
name: arch ${{ matrix.goarch }} (qemu)
runs-on: ubuntu-latest
strategy:
fail-fast: false
matrix:
include:
# V enabled: qemu's default riscv64 CPU has none, and the RVV kernels
# went untested here until 2026-10-26. NDARRAY_REQUIRE_RVV makes a
# CPU without V fail the lane (TestRVVWhenRequired).
- { goarch: riscv64, qemu: qemu-riscv64-static, cpu: "rv64,v=true,vlen=256", require: NDARRAY_REQUIRE_RVV }
# la464 has LSX and LASX; NDARRAY_REQUIRE_LASX makes a CPU that does
# not report LASX fail the lane (TestLASXWhenRequired).
- { goarch: loong64, qemu: qemu-loongarch64-static, cpu: "la464", require: NDARRAY_REQUIRE_LASX }
- { goarch: ppc64le, qemu: qemu-ppc64le-static }
# max includes z14's vector enhancements (VXE), which the float32
# kernels need; NDARRAY_REQUIRE_VXE fails the lane without them.
- { goarch: s390x, qemu: qemu-s390x-static, cpu: "max", require: NDARRAY_REQUIRE_VXE }
# 32-bit: int is 32 bits, so shape/size overflow paths differ.
- { goarch: arm, qemu: qemu-arm-static }
steps:
- uses: actions/checkout@v7
- uses: actions/setup-go@v7
with:
go-version: 1.27.3
# ⛔ `qemu-user-static` is a VIRTUAL package on Ubuntu 26.04, which
# `ubuntu-latest` became on 2026-10-19, and apt refuses to choose between its
# two providers:
#
# E: Package 'qemu-user-static' has no installation candidate
#
# ⚠ Installing a provider is NOT the fix: `qemu-user-binfmt` registers binfmt
# handlers, it does not ship the `qemu-<arch>-static` binaries this lane
# invokes. `qemu-user` ships the dynamic `qemu-<arch>`, which is all that
# `go test -exec` needs, so both are installed and BOTH NAMES are accepted
# below. go-tex/engine#645 has the full account.
- name: install qemu user-mode emulators
timeout-minutes: 10
run: sudo apt-get update && sudo apt-get install -y qemu-user qemu-user-binfmt
- name: Regenerate committed assembly and check it is current
if: matrix.goarch == 'ppc64le' || matrix.goarch == 'loong64' || matrix.goarch == 'riscv64' || matrix.goarch == 's390x'
run: |
( cd internal/kernels/asmgen/${{ matrix.goarch }} && go run . )
mv internal/kernels/asmgen/${{ matrix.goarch }}/sum_${{ matrix.goarch }}.s internal/kernels/sum_${{ matrix.goarch }}.s
git diff --exit-code -- internal/kernels/sum_${{ matrix.goarch }}.s || {
echo "::error::generated ${{ matrix.goarch }} assembly is stale; run go generate and commit"; exit 1; }
- name: Vet + test under qemu (${{ matrix.goarch }})
run: |
base="${{ matrix.qemu }}"
q=""
for cand in "$base" "${base%-static}"; do
command -v "$cand" >/dev/null 2>&1 && q="$cand" && break
done
# ⛔ A missing emulator FAILS this lane. A lane that reports success for
# an architecture it never ran is the one outcome this matrix exists to
# rule out.
if [ -z "$q" ]; then
echo "::error::no qemu user emulator for ${{ matrix.goarch }} (looked for $base and ${base%-static})"
exit 1
fi
# Printed on purpose: without it, "ran under emulation" and "skipped" are
# indistinguishable from a green tick.
if [ -n "${{ matrix.cpu }}" ]; then
export QEMU_CPU="${{ matrix.cpu }}"
export ${{ matrix.require }}=1
fi
echo "running ${{ matrix.goarch }} under $q (QEMU_CPU=${QEMU_CPU:-default})"
GOARCH=${{ matrix.goarch }} go test -exec="$q" ./...
# 386 runs natively on the amd64 runner (the kernel executes 32-bit x86
# binaries), not under qemu-i386, whose emulation loses x87/SSE state across
# preemption. It is the second 32-bit word-size lane, beside arm.
arch-386:
name: arch 386
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v7
- uses: actions/setup-go@v7
with:
go-version: 1.27.3
- name: Vet + test (GOARCH=386)
run: |
GOARCH=386 go vet ./...
GOARCH=386 go test ./...
# Every GOOS/GOARCH pair Go supports must at least compile the library's
# tests: a test that does not compile on a target is a test
# that never runs there, and nothing else would notice (a 1<<62 literal
# once broke every 32-bit target this way). Execution is the lanes above.
cross-compile:
name: compile tests on every target
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v7
- uses: actions/setup-go@v7
with:
go-version: 1.27.3
- name: go test -c (go vet on android/ios) for each go tool dist list pair
run: |
fail=0
for t in $(go tool dist list); do
os=${t%/*}; arch=${t#*/}
# android/ios refuse to build without a cgo linker, even for vet;
# with cgo nominally on and CC=false, vet type-checks the package
# and its tests (there is no import "C" to compile).
case $os in android|ios) mode=vet ;; *) mode=test ;; esac
for p in $(go list ./...); do
if [ $mode = test ]; then
out=$(CGO_ENABLED=0 GOOS=$os GOARCH=$arch go test -c -o /dev/null "$p" 2>&1) || { echo "::error::$t $p: $out"; fail=1; }
else
out=$(CGO_ENABLED=1 CC=false GOOS=$os GOARCH=$arch go vet "$p" 2>&1) || { echo "::error::$t $p: $out"; fail=1; }
fi
done
done
exit $fail