[dev.link] cmd/link: remove OutData

OutData was used for a symbol to point to its data in the output
buffer, in order to apply relocations. Now we fold relocation
application to Asmb next to symbol data writing. We can just pass
the output data as a local variable.

Linking cmd/compile,

name         old time/op    new time/op    delta
Asmb_GC        19.0ms ±10%    16.6ms ± 9%  -12.50%  (p=0.032 n=5+5)

name         old alloc/op   new alloc/op   delta
Asmb_GC        3.78MB ± 0%    0.14MB ± 1%  -96.41%  (p=0.008 n=5+5)

name         old live-B     new live-B     delta
Asmb_GC         27.5M ± 0%     23.9M ± 0%  -13.24%  (p=0.008 n=5+5)

Change-Id: Id870a10dce2a0a7447a05029c6d0ab39b47d0a12
Reviewed-on: https://go-review.googlesource.com/c/go/+/244017
Reviewed-by: Jeremy Faller <jeremy@golang.org>

src/cmd/link/internal/ld/asmb.go

@@ -18,7 +18,6 @@ import (
 //  - writing out the architecture specific pieces.
 // This function handles the first part.
 func asmb(ctxt *Link) {
-	ctxt.loader.InitOutData()
 	if ctxt.IsExternal() && !ctxt.StreamExtRelocs() {
 		ctxt.loader.InitExtRelocs()
 	}

src/cmd/link/internal/ld/data.go

@@ -1026,9 +1026,9 @@ func writeBlock(ctxt *Link, out *OutBuf, ldr *loader.Loader, syms []loader.Sym,
 			out.WriteStringPad("", int(val-addr), pad)
 			addr = val
 		}
-		out.WriteSym(ldr, s)
-		st.relocsym(s, ldr.OutData(s))
-		addr += int64(len(ldr.Data(s)))
+		P := out.WriteSym(ldr, s)
+		st.relocsym(s, P)
+		addr += int64(len(P))
 		siz := ldr.SymSize(s)
 		if addr < val+siz {
 			out.WriteStringPad("", int(val+siz-addr), pad)
@@ -2677,8 +2677,8 @@ func compressSyms(ctxt *Link, syms []loader.Sym) []byte {
 		if relocs.Count() != 0 {
 			relocbuf = append(relocbuf[:0], P...)
 			P = relocbuf
+			st.relocsym(s, P)
 		}
-		st.relocsym(s, P)
 		if _, err := z.Write(P); err != nil {
 			log.Fatalf("compression failed: %s", err)
 		}

src/cmd/link/internal/ld/link.go

@@ -159,8 +159,8 @@ func (ctxt *Link) MaxVersion() int {
 // Generator symbols shouldn't grow the symbol size, and might be called in
 // parallel in the future.
 //
-// Generator Symbols have their Data and OutData set to the mmapped area when
-// the generator is called.
+// Generator Symbols have their Data set to the mmapped area when the
+// generator is called.
 type generatorFunc func(*Link, loader.Sym)
 
 // createGeneratorSymbol is a convenience method for creating a generator

src/cmd/link/internal/ld/outbuf.go

@@ -277,23 +277,24 @@ func (out *OutBuf) WriteStringPad(s string, n int, pad []byte) {
 	}
 }
 
-// WriteSym writes the content of a Symbol, then changes the Symbol's content
-// to point to the output buffer that we just wrote, so we can apply further
-// edit to the symbol content.
-// If the output file is not Mmap'd, just writes the content.
-func (out *OutBuf) WriteSym(ldr *loader.Loader, s loader.Sym) {
+// WriteSym writes the content of a Symbol, and returns the output buffer
+// that we just wrote, so we can apply further edit to the symbol content.
+// For generator symbols, it also sets the symbol's Data to the output
+// buffer.
+func (out *OutBuf) WriteSym(ldr *loader.Loader, s loader.Sym) []byte {
 	if !ldr.IsGeneratedSym(s) {
 		P := ldr.Data(s)
 		n := int64(len(P))
 		pos, buf := out.writeLoc(n)
 		copy(buf[pos:], P)
 		out.off += n
-		ldr.SetOutData(s, buf[pos:pos+n])
+		ldr.FreeData(s)
+		return buf[pos : pos+n]
 	} else {
 		n := ldr.SymSize(s)
 		pos, buf := out.writeLoc(n)
 		out.off += n
-		ldr.SetOutData(s, buf[pos:pos+n])
 		ldr.MakeSymbolUpdater(s).SetData(buf[pos : pos+n])
+		return buf[pos : pos+n]
 	}
 }

src/cmd/link/internal/loader/loader.go

@@ -238,7 +238,6 @@ type Loader struct {
 
 	align []uint8 // symbol 2^N alignment, indexed by global index
 
-	outdata   [][]byte     // symbol's data in the output buffer
 	extRelocs [][]ExtReloc // symbol's external relocations
 
 	itablink         map[Sym]struct{} // itablink[j] defined if j is go.itablink.*
@@ -1230,30 +1229,16 @@ func (l *Loader) Data(i Sym) []byte {
 	return r.Data(li)
 }
 
-// Returns the data of the i-th symbol in the output buffer.
-func (l *Loader) OutData(i Sym) []byte {
-	if int(i) < len(l.outdata) && l.outdata[i] != nil {
-		return l.outdata[i]
-	}
-	return l.Data(i)
-}
-
-// SetOutData sets the position of the data of the i-th symbol in the output buffer.
+// FreeData clears the symbol data of an external symbol, allowing the memory
+// to be freed earlier. No-op for non-external symbols.
 // i is global index.
-func (l *Loader) SetOutData(i Sym, data []byte) {
+func (l *Loader) FreeData(i Sym) {
 	if l.IsExternal(i) {
 		pp := l.getPayload(i)
 		if pp != nil {
-			pp.data = data
-			return
+			pp.data = nil
 		}
 	}
-	l.outdata[i] = data
-}
-
-// InitOutData initializes the slice used to store symbol output data.
-func (l *Loader) InitOutData() {
-	l.outdata = make([][]byte, l.extStart)
 }
 
 // SetExtRelocs sets the external relocations of the i-th symbol. i is global index.

src/cmd/link/internal/mips/asm.go

@@ -77,7 +77,7 @@ func machoreloc1(*sys.Arch, *ld.OutBuf, *loader.Loader, loader.Sym, loader.ExtRe
 }
 
 func applyrel(arch *sys.Arch, ldr *loader.Loader, rt objabi.RelocType, off int32, s loader.Sym, val int64, t int64) int64 {
-	o := arch.ByteOrder.Uint32(ldr.OutData(s)[off:])
+	o := uint32(val)
 	switch rt {
 	case objabi.R_ADDRMIPS, objabi.R_ADDRMIPSTLS:
 		return int64(o&0xffff0000 | uint32(t)&0xffff)

src/cmd/link/internal/mips64/asm.go

@@ -129,25 +129,22 @@ func archreloc(target *ld.Target, ldr *loader.Loader, syms *ld.ArchSyms, r loade
 	case objabi.R_ADDRMIPS,
 		objabi.R_ADDRMIPSU:
 		t := ldr.SymValue(rs) + r.Add()
-		o1 := target.Arch.ByteOrder.Uint32(ldr.OutData(s)[r.Off():])
 		if r.Type() == objabi.R_ADDRMIPS {
-			return int64(o1&0xffff0000 | uint32(t)&0xffff), noExtReloc, isOk
+			return int64(val&0xffff0000 | t&0xffff), noExtReloc, isOk
 		}
-		return int64(o1&0xffff0000 | uint32((t+1<<15)>>16)&0xffff), noExtReloc, isOk
+		return int64(val&0xffff0000 | ((t+1<<15)>>16)&0xffff), noExtReloc, isOk
 	case objabi.R_ADDRMIPSTLS:
 		// thread pointer is at 0x7000 offset from the start of TLS data area
 		t := ldr.SymValue(rs) + r.Add() - 0x7000
 		if t < -32768 || t >= 32678 {
 			ldr.Errorf(s, "TLS offset out of range %d", t)
 		}
-		o1 := target.Arch.ByteOrder.Uint32(ldr.OutData(s)[r.Off():])
-		return int64(o1&0xffff0000 | uint32(t)&0xffff), noExtReloc, isOk
+		return int64(val&0xffff0000 | t&0xffff), noExtReloc, isOk
 	case objabi.R_CALLMIPS,
 		objabi.R_JMPMIPS:
 		// Low 26 bits = (S + A) >> 2
 		t := ldr.SymValue(rs) + r.Add()
-		o1 := target.Arch.ByteOrder.Uint32(ldr.OutData(s)[r.Off():])
-		return int64(o1&0xfc000000 | uint32(t>>2)&^0xfc000000), noExtReloc, isOk
+		return int64(val&0xfc000000 | (t>>2)&^0xfc000000), noExtReloc, isOk
 	}
 
 	return val, 0, false