cmd/compile,runtime: pass only ptr and len to some runtime calls

Some runtime calls accept a slice, but only use ptr and len.
This change modifies most such routines to accept only ptr and len.

After this change, the only runtime calls that accept an unnecessary
cap arg are concatstrings and slicerunetostring.
Neither is particularly common, and both are complicated to modify.

Negligible compiler performance impact. Shrinks binaries a little.
There are only a few regressions; the one I investigated was
due to register allocation fluctuation.

Passes 'go test -race std cmd', modulo #38265 and #38266.
Wow, does that take a long time to run.

Updates #36890

file      before    after     Δ       %       
compile   19655024  19655152  +128    +0.001% 
cover     5244840   5236648   -8192   -0.156% 
dist      3662376   3658280   -4096   -0.112% 
link      6680056   6675960   -4096   -0.061% 
pprof     14789844  14777556  -12288  -0.083% 
test2json 2824744   2820648   -4096   -0.145% 
trace     11647876  11639684  -8192   -0.070% 
vet       8260472   8256376   -4096   -0.050% 
total     115163736 115118808 -44928  -0.039% 

Change-Id: Idb29fa6a81d6a82bfd3b65740b98cf3275ca0a78
Reviewed-on: https://go-review.googlesource.com/c/go/+/227163
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>

src/cmd/compile/internal/gc/builtin/runtime.go

@@ -69,13 +69,13 @@ func concatstrings(*[32]byte, []string) string
 
 func cmpstring(string, string) int
 func intstring(*[4]byte, int64) string
-func slicebytetostring(*[32]byte, []byte) string
-func slicebytetostringtmp([]byte) string
+func slicebytetostring(buf *[32]byte, ptr *byte, n int) string
+func slicebytetostringtmp(ptr *byte, n int) string
 func slicerunetostring(*[32]byte, []rune) string
 func stringtoslicebyte(*[32]byte, string) []byte
 func stringtoslicerune(*[32]rune, string) []rune
-func slicecopy(to any, fr any, wid uintptr) int
-func slicestringcopy(to any, fr any) int
+func slicecopy(toPtr *any, toLen int, frPtr *any, frLen int, wid uintptr) int
+func slicestringcopy(toPtr *byte, toLen int, fr string) int
 
 func decoderune(string, int) (retv rune, retk int)
 func countrunes(string) int
@@ -162,7 +162,7 @@ var writeBarrier struct {
 // *byte is really *runtime.Type
 func typedmemmove(typ *byte, dst *any, src *any)
 func typedmemclr(typ *byte, dst *any)
-func typedslicecopy(typ *byte, dst any, src any) int
+func typedslicecopy(typ *byte, dstPtr *any, dstLen int, srcPtr *any, srcLen int) int
 
 func selectnbsend(hchan chan<- any, elem *any) bool
 func selectnbrecv(elem *any, hchan <-chan any) bool

src/cmd/compile/internal/gc/ssa.go

@@ -3277,10 +3277,7 @@ func init() {
 				// Compiler frontend optimizations emit OBYTES2STRTMP nodes
 				// for the backend instead of slicebytetostringtmp calls
 				// when not instrumenting.
-				slice := args[0]
-				ptr := s.newValue1(ssa.OpSlicePtr, s.f.Config.Types.BytePtr, slice)
-				len := s.newValue1(ssa.OpSliceLen, types.Types[TINT], slice)
-				return s.newValue2(ssa.OpStringMake, n.Type, ptr, len)
+				return s.newValue2(ssa.OpStringMake, n.Type, args[0], args[1])
 			},
 			all...)
 	}

src/cmd/compile/internal/gc/subr.go

@@ -923,6 +923,21 @@ func (o Op) IsSlice3() bool {
 	return false
 }
 
+// slicePtrLen extracts the pointer and length from a slice.
+// This constructs two nodes referring to n, so n must be a cheapexpr.
+func (n *Node) slicePtrLen() (ptr, len *Node) {
+	var init Nodes
+	c := cheapexpr(n, &init)
+	if c != n || init.Len() != 0 {
+		Fatalf("slicePtrLen not cheap: %v", n)
+	}
+	ptr = nod(OSPTR, n, nil)
+	ptr.Type = n.Type.Elem().PtrTo()
+	len = nod(OLEN, n, nil)
+	len.Type = types.Types[TINT]
+	return ptr, len
+}
+
 // labeledControl returns the control flow Node (for, switch, select)
 // associated with the label n, if any.
 func (n *Node) labeledControl() *Node {

src/cmd/compile/internal/gc/walk.go

@@ -1414,13 +1414,15 @@ opswitch:
 			t := types.NewArray(types.Types[TUINT8], tmpstringbufsize)
 			a = nod(OADDR, temp(t), nil)
 		}
-		fn := "slicebytetostring"
 		if n.Op == ORUNES2STR {
-			fn = "slicerunetostring"
+			// slicerunetostring(*[32]byte, []rune) string
+			n = mkcall("slicerunetostring", n.Type, init, a, n.Left)
+		} else {
+			// slicebytetostring(*[32]byte, ptr *byte, n int) string
+			n.Left = cheapexpr(n.Left, init)
+			ptr, len := n.Left.slicePtrLen()
+			n = mkcall("slicebytetostring", n.Type, init, a, ptr, len)
 		}
-		// slicebytetostring(*[32]byte, []byte) string
-		// slicerunetostring(*[32]byte, []rune) string
-		n = mkcall(fn, n.Type, init, a, n.Left)
 
 	case OBYTES2STRTMP:
 		n.Left = walkexpr(n.Left, init)
@@ -1429,8 +1431,10 @@ opswitch:
 			// to avoid a function call to slicebytetostringtmp.
 			break
 		}
-		// slicebytetostringtmp([]byte) string
-		n = mkcall("slicebytetostringtmp", n.Type, init, n.Left)
+		// slicebytetostringtmp(ptr *byte, n int) string
+		n.Left = cheapexpr(n.Left, init)
+		ptr, len := n.Left.slicePtrLen()
+		n = mkcall("slicebytetostringtmp", n.Type, init, ptr, len)
 
 	case OSTR2BYTES:
 		s := n.Left
@@ -2645,6 +2649,8 @@ func appendslice(n *Node, init *Nodes) *Node {
 
 	l1 := n.List.First()
 	l2 := n.List.Second()
+	l2 = cheapexpr(l2, init)
+	n.List.SetSecond(l2)
 
 	var nodes Nodes
 
@@ -2682,35 +2688,45 @@ func appendslice(n *Node, init *Nodes) *Node {
 	if elemtype.HasHeapPointer() {
 		// copy(s[len(l1):], l2)
 		nptr1 := nod(OSLICE, s, nil)
+		nptr1.Type = s.Type
 		nptr1.SetSliceBounds(nod(OLEN, l1, nil), nil, nil)
+		nptr1 = cheapexpr(nptr1, &nodes)
 
 		nptr2 := l2
 
 		Curfn.Func.setWBPos(n.Pos)
 
-		// instantiate typedslicecopy(typ *type, dst any, src any) int
+		// instantiate typedslicecopy(typ *type, dstPtr *any, dstLen int, srcPtr *any, srcLen int) int
 		fn := syslook("typedslicecopy")
-		fn = substArgTypes(fn, l1.Type, l2.Type)
-		ncopy = mkcall1(fn, types.Types[TINT], &nodes, typename(elemtype), nptr1, nptr2)
+		fn = substArgTypes(fn, l1.Type.Elem(), l2.Type.Elem())
+		ptr1, len1 := nptr1.slicePtrLen()
+		ptr2, len2 := nptr2.slicePtrLen()
+		ncopy = mkcall1(fn, types.Types[TINT], &nodes, typename(elemtype), ptr1, len1, ptr2, len2)
 
 	} else if instrumenting && !compiling_runtime {
 		// rely on runtime to instrument copy.
 		// copy(s[len(l1):], l2)
 		nptr1 := nod(OSLICE, s, nil)
+		nptr1.Type = s.Type
 		nptr1.SetSliceBounds(nod(OLEN, l1, nil), nil, nil)
+		nptr1 = cheapexpr(nptr1, &nodes)
 
 		nptr2 := l2
 
 		if l2.Type.IsString() {
-			// instantiate func slicestringcopy(to any, fr any) int
+			// instantiate func slicestringcopy(toPtr *byte, toLen int, fr string) int
 			fn := syslook("slicestringcopy")
-			fn = substArgTypes(fn, l1.Type, l2.Type)
-			ncopy = mkcall1(fn, types.Types[TINT], &nodes, nptr1, nptr2)
+			ptr, len := nptr1.slicePtrLen()
+			str := nod(OCONVNOP, nptr2, nil)
+			str.Type = types.Types[TSTRING]
+			ncopy = mkcall1(fn, types.Types[TINT], &nodes, ptr, len, str)
 		} else {
 			// instantiate func slicecopy(to any, fr any, wid uintptr) int
 			fn := syslook("slicecopy")
-			fn = substArgTypes(fn, l1.Type, l2.Type)
-			ncopy = mkcall1(fn, types.Types[TINT], &nodes, nptr1, nptr2, nodintconst(elemtype.Width))
+			fn = substArgTypes(fn, l1.Type.Elem(), l2.Type.Elem())
+			ptr1, len1 := nptr1.slicePtrLen()
+			ptr2, len2 := nptr2.slicePtrLen()
+			ncopy = mkcall1(fn, types.Types[TINT], &nodes, ptr1, len1, ptr2, len2, nodintconst(elemtype.Width))
 		}
 
 	} else {
@@ -3009,20 +3025,31 @@ func walkappend(n *Node, init *Nodes, dst *Node) *Node {
 func copyany(n *Node, init *Nodes, runtimecall bool) *Node {
 	if n.Left.Type.Elem().HasHeapPointer() {
 		Curfn.Func.setWBPos(n.Pos)
-		fn := writebarrierfn("typedslicecopy", n.Left.Type, n.Right.Type)
-		return mkcall1(fn, n.Type, init, typename(n.Left.Type.Elem()), n.Left, n.Right)
+		fn := writebarrierfn("typedslicecopy", n.Left.Type.Elem(), n.Right.Type.Elem())
+		n.Left = cheapexpr(n.Left, init)
+		ptrL, lenL := n.Left.slicePtrLen()
+		n.Right = cheapexpr(n.Right, init)
+		ptrR, lenR := n.Right.slicePtrLen()
+		return mkcall1(fn, n.Type, init, typename(n.Left.Type.Elem()), ptrL, lenL, ptrR, lenR)
 	}
 
 	if runtimecall {
 		if n.Right.Type.IsString() {
 			fn := syslook("slicestringcopy")
-			fn = substArgTypes(fn, n.Left.Type, n.Right.Type)
-			return mkcall1(fn, n.Type, init, n.Left, n.Right)
+			n.Left = cheapexpr(n.Left, init)
+			ptr, len := n.Left.slicePtrLen()
+			str := nod(OCONVNOP, n.Right, nil)
+			str.Type = types.Types[TSTRING]
+			return mkcall1(fn, n.Type, init, ptr, len, str)
 		}
 
 		fn := syslook("slicecopy")
-		fn = substArgTypes(fn, n.Left.Type, n.Right.Type)
-		return mkcall1(fn, n.Type, init, n.Left, n.Right, nodintconst(n.Left.Type.Elem().Width))
+		fn = substArgTypes(fn, n.Left.Type.Elem(), n.Right.Type.Elem())
+		n.Left = cheapexpr(n.Left, init)
+		ptrL, lenL := n.Left.slicePtrLen()
+		n.Right = cheapexpr(n.Right, init)
+		ptrR, lenR := n.Right.slicePtrLen()
+		return mkcall1(fn, n.Type, init, ptrL, lenL, ptrR, lenR, nodintconst(n.Left.Type.Elem().Width))
 	}
 
 	n.Left = walkexpr(n.Left, init)

src/runtime/cgocheck.go

@@ -76,23 +76,24 @@ func cgoCheckMemmove(typ *_type, dst, src unsafe.Pointer, off, size uintptr) {
 	cgoCheckTypedBlock(typ, src, off, size)
 }
 
-// cgoCheckSliceCopy is called when copying n elements of a slice from
-// src to dst.  typ is the element type of the slice.
+// cgoCheckSliceCopy is called when copying n elements of a slice.
+// src and dst are pointers to the first element of the slice.
+// typ is the element type of the slice.
 // It throws if the program is copying slice elements that contain Go pointers
 // into non-Go memory.
 //go:nosplit
 //go:nowritebarrier
-func cgoCheckSliceCopy(typ *_type, dst, src slice, n int) {
+func cgoCheckSliceCopy(typ *_type, dst, src unsafe.Pointer, n int) {
 	if typ.ptrdata == 0 {
 		return
 	}
-	if !cgoIsGoPointer(src.array) {
+	if !cgoIsGoPointer(src) {
 		return
 	}
-	if cgoIsGoPointer(dst.array) {
+	if cgoIsGoPointer(dst) {
 		return
 	}
-	p := src.array
+	p := src
 	for i := 0; i < n; i++ {
 		cgoCheckTypedBlock(typ, p, 0, typ.size)
 		p = add(p, typ.size)

src/runtime/mbarrier.go

@@ -231,16 +231,14 @@ func reflectcallmove(typ *_type, dst, src unsafe.Pointer, size uintptr) {
 }
 
 //go:nosplit
-func typedslicecopy(typ *_type, dst, src slice) int {
-	n := dst.len
-	if n > src.len {
-		n = src.len
+func typedslicecopy(typ *_type, dstPtr unsafe.Pointer, dstLen int, srcPtr unsafe.Pointer, srcLen int) int {
+	n := dstLen
+	if n > srcLen {
+		n = srcLen
 	}
 	if n == 0 {
 		return 0
 	}
-	dstp := dst.array
-	srcp := src.array
 
 	// The compiler emits calls to typedslicecopy before
 	// instrumentation runs, so unlike the other copying and
@@ -249,19 +247,19 @@ func typedslicecopy(typ *_type, dst, src slice) int {
 	if raceenabled {
 		callerpc := getcallerpc()
 		pc := funcPC(slicecopy)
-		racewriterangepc(dstp, uintptr(n)*typ.size, callerpc, pc)
-		racereadrangepc(srcp, uintptr(n)*typ.size, callerpc, pc)
+		racewriterangepc(dstPtr, uintptr(n)*typ.size, callerpc, pc)
+		racereadrangepc(srcPtr, uintptr(n)*typ.size, callerpc, pc)
 	}
 	if msanenabled {
-		msanwrite(dstp, uintptr(n)*typ.size)
-		msanread(srcp, uintptr(n)*typ.size)
+		msanwrite(dstPtr, uintptr(n)*typ.size)
+		msanread(srcPtr, uintptr(n)*typ.size)
 	}
 
 	if writeBarrier.cgo {
-		cgoCheckSliceCopy(typ, dst, src, n)
+		cgoCheckSliceCopy(typ, dstPtr, srcPtr, n)
 	}
 
-	if dstp == srcp {
+	if dstPtr == srcPtr {
 		return n
 	}
 
@@ -272,11 +270,11 @@ func typedslicecopy(typ *_type, dst, src slice) int {
 	size := uintptr(n) * typ.size
 	if writeBarrier.needed {
 		pwsize := size - typ.size + typ.ptrdata
-		bulkBarrierPreWrite(uintptr(dstp), uintptr(srcp), pwsize)
+		bulkBarrierPreWrite(uintptr(dstPtr), uintptr(srcPtr), pwsize)
 	}
 	// See typedmemmove for a discussion of the race between the
 	// barrier and memmove.
-	memmove(dstp, srcp, size)
+	memmove(dstPtr, srcPtr, size)
 	return n
 }
 
@@ -306,7 +304,7 @@ func reflect_typedslicecopy(elemType *_type, dst, src slice) int {
 		memmove(dst.array, src.array, size)
 		return n
 	}
-	return typedslicecopy(elemType, dst, src)
+	return typedslicecopy(elemType, dst.array, dst.len, src.array, src.len)
 }
 
 // typedmemclr clears the typed memory at ptr with type typ. The

src/runtime/os_linux.go

@@ -277,13 +277,14 @@ func getHugePageSize() uintptr {
 	if fd < 0 {
 		return 0
 	}
-	n := read(fd, noescape(unsafe.Pointer(&numbuf[0])), int32(len(numbuf)))
+	ptr := noescape(unsafe.Pointer(&numbuf[0]))
+	n := read(fd, ptr, int32(len(numbuf)))
 	closefd(fd)
 	if n <= 0 {
 		return 0
 	}
-	l := n - 1 // remove trailing newline
-	v, ok := atoi(slicebytetostringtmp(numbuf[:l]))
+	n-- // remove trailing newline
+	v, ok := atoi(slicebytetostringtmp((*byte)(ptr), int(n)))
 	if !ok || v < 0 {
 		v = 0
 	}