runtime: add minimal MemProfile allocation records

[cpunion]

Summary

• Track LLGO runtime allocations in fixed MemProfile size-class buckets and expose runtime.MemProfile with the Go public MemProfileRecord/Stack0 shape.
• Add stable test/go/memprofile coverage for the MemProfile sizing loop and tiny allocation bucket.
• Remove the now-fixed finprofiled.go GOROOT xfails only; heapsampling.go remains xfailed because stack-attributed heap sampling is still missing.

Tests

• go test ./test/go/memprofile -run TestRuntimeMemProfileReportsTinyAllocations -count=1
• (cd runtime && go test ./internal/runtime ./internal/lib/runtime -count=1)
• go build -tags=dev -o /tmp/llgo-profile ./cmd/llgo
• /tmp/llgo-profile test -run TestRuntimeMemProfileReportsTinyAllocations -count=1 ./test/go/memprofile
• go test ./test/goroot -run TestGoRootRunCases -count=1 -timeout=10m -args -goroot "$(go env GOROOT)" -dirs . -case '^finprofiled\\.go$' -xfail test/goroot/no-xfail.yaml -run-timeout=3m -build-timeout=5m
• go test ./test/goroot -run TestGoRootRunCases -count=1 -timeout=10m -args -goroot "$(go env GOROOT)" -dirs . -case '^finprofiled\\.go$' -run-timeout=3m -build-timeout=5m

Notes

• Full GOROOT CI is slow/disabled, so this PR uses targeted local GOROOT runs plus normal PR CI from the fork branch.
• Manual GOROOT workflow_dispatch was not run. If it cannot resolve fork refs, keep relying on normal PR CI rather than pushing a branch to xgo-dev/llgo.
• Bounded heapsampling.go check with xfail disabled still fails: panic: main.allocInterleaved1:0 want objects in [45000: 55000], got [0 0 0], which is the remaining stack-attributed heap sampling gap.

[cpunion]

Follow-up for head e68aeb7.

Root cause:

• The MemProfile allocation counter used a uint64 atomic add in runtime/internal/runtime.recordMemProfileAlloc. On ESP32-C3/riscv32 baremetal this lowered to __atomic_fetch_add_8, which is not provided by the target runtime, so build-cache, Targets, and LLGo embedded/demo jobs failed at link time.
• runtime/pprof still had an empty pprof_memProfileInternal stub, so the heap-profile path was not wired to the new MemProfile data.

Fix:

• Split MemProfile counters by build mode: non-baremetal keeps atomic uint64 counters; baremetal uses a plain uintptr counter to avoid 64-bit atomic helper symbols on riscv32.
• Added Go-version-aware runtime/pprof memProfileInternal adapters: Go 1.23+ uses the internal/profilerecord slice-stack layout, older Go keeps the runtime.MemProfileRecord layout.
• Added test/go coverage for runtime/pprof heap profile allocation totals in addition to direct runtime.MemProfile coverage.

Local tests:

• go test ./test/go/memprofile -count=1 -v
• go test ./test/go -count=1
• (cd runtime && go test ./internal/runtime ./internal/lib/runtime -count=1)
• go build -tags=dev -o /tmp/llgo-profile ./cmd/llgo
• LLGO_ROOT=/Users/lijie/source/goplus/llgo-wt-goroot-profile /tmp/llgo-profile test -run 'TestRuntime(MemProfile|PprofHeapProfile)ReportsTinyAllocations' -count=1 ./test/go/memprofile
• LLGO_ROOT=/Users/lijie/source/goplus/llgo-wt-goroot-profile /tmp/llgo-profile build -target=esp32c3 -o /tmp/llgo-buildcache.elf -v . (from test/buildcache)
• LLGO_ROOT=/Users/lijie/source/goplus/llgo-wt-goroot-profile /tmp/llgo-profile build -target=esp32c3 -o /tmp/llgo-target-defer.elf -v ./defer (from _demo/embed/targetsbuild)
• go test ./cl ./ssa -count=1
• go test ./test/goroot -run TestGoRootRunCases -count=1 -timeout=10m -args -goroot "$(go env GOROOT)" -dirs . -case '^finprofiled.go$' -xfail test/goroot/no-xfail.yaml -run-timeout=3m -build-timeout=5m
• git diff --check

CI status as of 2026-05-22 18:34 UTC:

• No failed checks on the new head.
• 28 checks have passed, including build-cache ubuntu, Targets ubuntu, Go ubuntu test with embedded emulator env, LLGo ubuntu 1.21/1.24/1.26, coverage ubuntu, install docs ubuntu, fmt, and linux release artifacts.
• 27 checks are still queued, all macOS jobs. I will keep watching and iterate if any macOS failure appears.

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!