[dev.typeparams] cmd/compile: unified IR construction

This CL adds a new unified IR construction mode to the frontend.  It's
purely additive, and all files include "UNREVIEWED" at the top, like
how types2 was initially imported. The next CL adds a -d=unified flag
to actually enable unified IR mode.

See below for more details, but some highlights:

1. It adds ~6kloc (excluding enum listings and stringer output), but I
estimate it will allow removing ~14kloc (see CL 324670, including its
commit message);

2. When enabled by default, it passes more tests than -G=3 does (see
CL 325213 and CL 324673);

3. Without requiring any new code, it supports inlining of more code
than the current inliner (see CL 324574; contrast CL 283112 and CL
266203, which added support for inlining function literals and type
switches, respectively);

4. Aside from dictionaries (which I intend to add still), its support
for generics is more complete (e.g., it fully supports local types,
including local generic types within generic functions and
instantiating generic types with local types; see
test/typeparam/nested.go);

5. It supports lazy loading of types and objects for types2 type
checking;

6. It supports re-exporting of types, objects, and inline bodies
without needing to parse them into IR;

7. The new export data format has extensive support for debugging with
"sync" markers, so mistakes during development are easier to catch;

8. When compiling with -d=inlfuncswithclosures=0, it enables "quirks
mode" where it generates output that passes toolstash -cmp.

--

The new unified IR pipeline combines noding, stenciling, inlining, and
import/export into a single, shared code path. Previously, IR trees
went through multiple phases of copying during compilation:

1. "Noding": the syntax AST is copied into the initial IR form. To
support generics, there's now also "irgen", which implements the same
idea, but takes advantage of types2 type-checking results to more
directly construct IR.

2. "Stenciling": generic IR forms are copied into instantiated IR
forms, substituting type parameters as appropriate.

3. "Inlining": the inliner made backup copies of inlinable functions,
and then copied them again when inlining into a call site, with some
modifications (e.g., updating position information, rewriting variable
references, changing "return" statements into "goto").

4. "Importing/exporting": the exporter wrote out the IR as saved by
the inliner, and then the importer read it back as to be used by the
inliner again. Normal functions are imported/exported "desugared",
while generic functions are imported/exported in source form.

These passes are all conceptually the same thing: make a copy of a
function body, maybe with some minor changes/substitutions. However,
they're all completely separate implementations that frequently run
into the same issues because IR has many nuanced corner cases.

For example, inlining currently doesn't support local defined types,
"range" loops, or labeled "for"/"switch" statements, because these
require special handling around Sym references. We've recently
extended the inliner to support new features like inlining type
switches and function literals, and they've had issues. The exporter
only knows how to export from IR form, so when re-exporting inlinable
functions (e.g., methods on imported types that are exposed via
exported APIs), these functions may need to be imported as IR for the
sole purpose of being immediately exported back out again.

By unifying all of these modes of copying into a single code path that
cleanly separates concerns, we eliminate many of these possible
issues. Some recent examples:

1. Issues #45743 and #46472 were issues where type switches were
mishandled by inlining and stenciling, respectively; but neither of
these affected unified IR, because it constructs type switches using
the exact same code as for normal functions.

2. CL 325409 fixes an issue in stenciling with implicit conversion of
values of type-parameter type to variables of interface type, but this
issue did not affect unified IR.

Change-Id: I5a05991fe16d68bb0f712503e034cb9f2d19e296
Reviewed-on: https://go-review.googlesource.com/c/go/+/324573
Trust: Matthew Dempsky <mdempsky@google.com>
Trust: Robert Griesemer <gri@golang.org>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>

src/cmd/compile/internal/noder/linker.go

@@ -0,0 +1,296 @@
+// UNREVIEWED
+
+// Copyright 2021 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package noder
+
+import (
+	"io"
+
+	"cmd/compile/internal/base"
+	"cmd/compile/internal/ir"
+	"cmd/compile/internal/reflectdata"
+	"cmd/compile/internal/types"
+	"cmd/internal/goobj"
+	"cmd/internal/obj"
+)
+
+// This file implements the unified IR linker, which combines the
+// local package's stub data with imported package data to produce a
+// complete export data file. It also rewrites the compiler's
+// extension data sections based on the results of compilation (e.g.,
+// the function inlining cost and linker symbol index assignments).
+//
+// TODO(mdempsky): Using the name "linker" here is confusing, because
+// readers are likely to mistake references to it for cmd/link. But
+// there's a shortage of good names for "something that combines
+// multiple parts into a cohesive whole"... e.g., "assembler" and
+// "compiler" are also already taken.
+
+type linker struct {
+	pw pkgEncoder
+
+	pkgs  map[string]int
+	decls map[*types.Sym]int
+}
+
+func (l *linker) relocAll(pr *pkgReader, relocs []relocEnt) []relocEnt {
+	res := make([]relocEnt, len(relocs))
+	for i, rent := range relocs {
+		rent.idx = l.relocIdx(pr, rent.kind, rent.idx)
+		res[i] = rent
+	}
+	return res
+}
+
+func (l *linker) relocIdx(pr *pkgReader, k reloc, idx int) int {
+	assert(pr != nil)
+
+	absIdx := pr.absIdx(k, idx)
+
+	if newidx := pr.newindex[absIdx]; newidx != 0 {
+		return ^newidx
+	}
+
+	var newidx int
+	switch k {
+	case relocString:
+		newidx = l.relocString(pr, idx)
+	case relocPkg:
+		newidx = l.relocPkg(pr, idx)
+	case relocObj:
+		newidx = l.relocObj(pr, idx)
+
+	default:
+		// Generic relocations.
+		//
+		// TODO(mdempsky): Deduplicate more sections? In fact, I think
+		// every section could be deduplicated. This would also be easier
+		// if we do external relocations.
+
+		w := l.pw.newEncoderRaw(k)
+		l.relocCommon(pr, &w, k, idx)
+		newidx = w.idx
+	}
+
+	pr.newindex[absIdx] = ^newidx
+
+	return newidx
+}
+
+func (l *linker) relocString(pr *pkgReader, idx int) int {
+	return l.pw.stringIdx(pr.stringIdx(idx))
+}
+
+func (l *linker) relocPkg(pr *pkgReader, idx int) int {
+	path := pr.peekPkgPath(idx)
+
+	if newidx, ok := l.pkgs[path]; ok {
+		return newidx
+	}
+
+	r := pr.newDecoder(relocPkg, idx, syncPkgDef)
+	w := l.pw.newEncoder(relocPkg, syncPkgDef)
+	l.pkgs[path] = w.idx
+
+	// TODO(mdempsky): We end up leaving an empty string reference here
+	// from when the package was originally written as "". Probably not
+	// a big deal, but a little annoying. Maybe relocating
+	// cross-references in place is the way to go after all.
+	w.relocs = l.relocAll(pr, r.relocs)
+
+	_ = r.string() // original path
+	w.string(path)
+
+	io.Copy(&w.data, &r.data)
+
+	return w.flush()
+}
+
+func (l *linker) relocObj(pr *pkgReader, idx int) int {
+	path, name, tag, _ := pr.peekObj(idx)
+	sym := types.NewPkg(path, "").Lookup(name)
+
+	if newidx, ok := l.decls[sym]; ok {
+		return newidx
+	}
+
+	if tag == objStub && path != "builtin" && path != "unsafe" {
+		pri, ok := objReader[sym]
+		if !ok {
+			base.Fatalf("missing reader for %q.%v", path, name)
+		}
+		assert(ok)
+
+		pr = pri.pr
+		idx = pri.idx
+
+		path2, name2, tag2, _ := pr.peekObj(idx)
+		sym2 := types.NewPkg(path2, "").Lookup(name2)
+		assert(sym == sym2)
+		assert(tag2 != objStub)
+	}
+
+	w := l.pw.newEncoderRaw(relocObj)
+	bside := l.pw.newEncoderRaw(relocObjExt)
+	assert(bside.idx == w.idx)
+	l.decls[sym] = w.idx
+
+	l.relocCommon(pr, &w, relocObj, idx)
+
+	var obj *ir.Name
+	if path == "" {
+		var ok bool
+		obj, ok = sym.Def.(*ir.Name)
+
+		// Generic types and functions won't have definitions.
+		// For now, just generically copy their extension data.
+		if !ok && base.Flag.G == 0 {
+			base.Fatalf("missing definition for %v", sym)
+		}
+	}
+
+	if obj != nil {
+		bside.sync(syncObject1)
+		switch tag {
+		case objFunc:
+			l.relocFuncExt(&bside, obj)
+		case objType:
+			l.relocTypeExt(&bside, obj)
+		case objVar:
+			l.relocVarExt(&bside, obj)
+		}
+		bside.flush()
+	} else {
+		l.relocCommon(pr, &bside, relocObjExt, idx)
+	}
+
+	return w.idx
+}
+
+func (l *linker) relocCommon(pr *pkgReader, w *encoder, k reloc, idx int) {
+	r := pr.newDecoderRaw(k, idx)
+	w.relocs = l.relocAll(pr, r.relocs)
+	io.Copy(&w.data, &r.data)
+	w.flush()
+}
+
+func (l *linker) pragmaFlag(w *encoder, pragma ir.PragmaFlag) {
+	w.sync(syncPragma)
+	w.int(int(pragma))
+}
+
+func (l *linker) relocFuncExt(w *encoder, name *ir.Name) {
+	w.sync(syncFuncExt)
+
+	l.pragmaFlag(w, name.Func.Pragma)
+	l.linkname(w, name)
+
+	// Relocated extension data.
+	w.bool(true)
+
+	// Record definition ABI so cross-ABI calls can be direct.
+	// This is important for the performance of calling some
+	// common functions implemented in assembly (e.g., bytealg).
+	w.uint64(uint64(name.Func.ABI))
+
+	// Escape analysis.
+	for _, fs := range &types.RecvsParams {
+		for _, f := range fs(name.Type()).FieldSlice() {
+			w.string(f.Note)
+		}
+	}
+
+	if inl := name.Func.Inl; w.bool(inl != nil) {
+		w.len(int(inl.Cost))
+		w.bool(inl.CanDelayResults)
+
+		pri, ok := bodyReader[name.Func]
+		assert(ok)
+		w.sync(syncAddBody)
+		w.reloc(relocBody, l.relocIdx(pri.pr, relocBody, pri.idx))
+	}
+
+	w.sync(syncEOF)
+}
+
+func (l *linker) relocTypeExt(w *encoder, name *ir.Name) {
+	w.sync(syncTypeExt)
+
+	typ := name.Type()
+
+	l.pragmaFlag(w, name.Pragma())
+
+	// For type T, export the index of type descriptor symbols of T and *T.
+	l.lsymIdx(w, "", reflectdata.TypeLinksym(typ))
+	l.lsymIdx(w, "", reflectdata.TypeLinksym(typ.PtrTo()))
+
+	if typ.Kind() != types.TINTER {
+		for _, method := range typ.Methods().Slice() {
+			l.relocFuncExt(w, method.Nname.(*ir.Name))
+		}
+	}
+}
+
+func (l *linker) relocVarExt(w *encoder, name *ir.Name) {
+	w.sync(syncVarExt)
+	l.linkname(w, name)
+}
+
+func (l *linker) linkname(w *encoder, name *ir.Name) {
+	w.sync(syncLinkname)
+
+	linkname := name.Sym().Linkname
+	if !l.lsymIdx(w, linkname, name.Linksym()) {
+		w.string(linkname)
+	}
+}
+
+func (l *linker) lsymIdx(w *encoder, linkname string, lsym *obj.LSym) bool {
+	if lsym.PkgIdx > goobj.PkgIdxSelf || (lsym.PkgIdx == goobj.PkgIdxInvalid && !lsym.Indexed()) || linkname != "" {
+		w.int64(-1)
+		return false
+	}
+
+	// For a defined symbol, export its index.
+	// For re-exporting an imported symbol, pass its index through.
+	w.int64(int64(lsym.SymIdx))
+	return true
+}
+
+// @@@ Helpers
+
+// TODO(mdempsky): These should probably be removed. I think they're a
+// smell that the export data format is not yet quite right.
+
+func (pr *pkgDecoder) peekPkgPath(idx int) string {
+	r := pr.newDecoder(relocPkg, idx, syncPkgDef)
+	path := r.string()
+	if path == "" {
+		path = pr.pkgPath
+	}
+	return path
+}
+
+func (pr *pkgDecoder) peekObj(idx int) (string, string, codeObj, []int) {
+	r := pr.newDecoder(relocObj, idx, syncObject1)
+	r.sync(syncSym)
+	r.sync(syncPkg)
+	path := pr.peekPkgPath(r.reloc(relocPkg))
+	name := r.string()
+	assert(name != "")
+
+	r.sync(syncTypeParamBounds)
+	r.len() // implicits
+	bounds := make([]int, r.len())
+	for i := range bounds {
+		r.sync(syncType)
+		bounds[i] = r.reloc(relocType)
+	}
+
+	tag := codeObj(r.code(syncCodeObj))
+
+	return path, name, tag, bounds
+}