[dev.typeparams] all: merge dev.regabi (23b0c1f) into dev.typeparams

Merge List:

+ 2021-02-02 23b0c1f76e [dev.regabi] all: merge master (fca94ab) into dev.regabi
+ 2021-02-02 fca94ab3ab spec: improve the example in Type assertions section
+ 2021-02-02 98f8454a73 cmd/link: don't decode type symbol in shared library in deadcode
+ 2021-02-02 1426a571b7 cmd/link: fix off-by-1 error in findShlibSection
+ 2021-02-01 32e789f4fb test: fix incorrectly laid out instructions in issue11656.go
+ 2021-02-01 ca6999e27c [dev.regabi] test: add a test for inlining closures
+ 2021-02-01 0b6cfea634 doc/go1.16: document that on OpenBSD syscalls are now made through libc
+ 2021-02-01 26e29aa15a cmd/link: disable TestPIESize if CGO isn't enabled
+ 2021-02-01 6ac91e460c doc/go1.16: minor markup fixes
+ 2021-01-29 44361140c0 embed: update docs for proposal tweaks
+ 2021-01-29 68058edc39 runtime: document pointer write atomicity for memclrNoHeapPointers
+ 2021-01-28 c8bd8010ff syscall: generate readlen/writelen for openbsd libc
+ 2021-01-28 41bb49b878 cmd/go: revert TestScript/build_trimpath to use ioutil.ReadFile
+ 2021-01-28 725a642c2d runtime: correct syscall10/syscall10X on openbsd/amd64
+ 2021-01-28 4b068cafb5 doc/go1.16: document go/build/constraint package
+ 2021-01-28 376518d77f runtime,syscall: convert syscall on openbsd/arm64 to libc
+ 2021-01-27 aca22bddf2 [dev.regabi] cmd/compile: remove nested functions from expands_calls.go
+ 2021-01-27 667e08ba8c [dev.regabi] cmd/go: Use GOMAXPROCS to limit default build, compile parallelism
+ 2021-01-27 00f2ff5c94 api/go1.16: add go/build/constraint APIs
+ 2021-01-27 35334caf18 crypto/x509: remove leftover CertificateRequest field
+ 2021-01-27 a5a5e2c968 runtime: make sure to remove open-coded defer entries in all cases after a recover
+ 2021-01-27 8cfa01943a runtime: block console ctrlhandler when the signal is handled
+ 2021-01-27 ff9e8364c6 cmd/go: skip issue33139 when the 'cc' script command is unavailable
+ 2021-01-27 cd176b3615 runtime: switch runtime to libc for openbsd/arm64
+ 2021-01-27 6c8fbfbdcf runtime: convert openbsd/arm64 locking to libc
+ 2021-01-27 5cdf0da1bf syscall: clean up mkasm related changes
+ 2021-01-27 210f70e298 doc/go1.16: fix closing brace in .Export format
+ 2021-01-27 0f797f168d math: fix typo in sqrt.go code comment
+ 2021-01-26 9b636feafe [dev.regabi] cmd/compile: missing last patch set for cl286013
+ 2021-01-26 f7dad5eae4 [dev.regabi] cmd/compile: remove leftover code form late call lowering work
+ 2021-01-26 8634a234df runtime,syscall: convert syscall on openbsd/amd64 to libc
+ 2021-01-26 1d5e14632e os: further document limitations around naked file descriptors
+ 2021-01-26 cf263e9f77 os: correct names in CreateTemp and MkdirTemp doc comments
+ 2021-01-26 ce8b318624 net/http/fcgi: remove locking added to prevent a test-only race

Change-Id: Ibd38d559c8a5b0aa32dd0d3a8cdf6876368a3aeb

19 files changed, 711 insertions, 1144 deletions

diff --git a/src/cmd/compile/internal/ssa/compile.go b/src/cmd/compile/internal/ssa/compile.go
index 63994d1778..c267274366 100644
--- a/src/cmd/compile/internal/ssa/compile.go
+++ b/src/cmd/compile/internal/ssa/compile.go
@@ -431,7 +431,6 @@ var passes = [...]pass{
 	{name: "early copyelim", fn: copyelim},
 	{name: "early deadcode", fn: deadcode}, // remove generated dead code to avoid doing pointless work during opt
 	{name: "short circuit", fn: shortcircuit},
-	{name: "decompose args", fn: decomposeArgs, required: !go116lateCallExpansion, disabled: go116lateCallExpansion}, // handled by late call lowering
 	{name: "decompose user", fn: decomposeUser, required: true},
 	{name: "pre-opt deadcode", fn: deadcode},
 	{name: "opt", fn: opt, required: true},               // NB: some generic rules know the name of the opt pass. TODO: split required rules and optimizing rules
diff --git a/src/cmd/compile/internal/ssa/config.go b/src/cmd/compile/internal/ssa/config.go
index 8dc2ee8213..e952c73d9b 100644
--- a/src/cmd/compile/internal/ssa/config.go
+++ b/src/cmd/compile/internal/ssa/config.go
@@ -179,14 +179,6 @@ type Frontend interface {
 	MyImportPath() string
 }
 
-const go116lateCallExpansion = true
-
-// LateCallExpansionEnabledWithin returns true if late call expansion should be tested
-// within compilation of a function/method.
-func LateCallExpansionEnabledWithin(f *Func) bool {
-	return go116lateCallExpansion
-}
-
 // NewConfig returns a new configuration object for the given architecture.
 func NewConfig(arch string, types Types, ctxt *obj.Link, optimize bool) *Config {
 	c := &Config{arch: arch, Types: types}
diff --git a/src/cmd/compile/internal/ssa/decompose.go b/src/cmd/compile/internal/ssa/decompose.go
index bf7f1e826b..ea988e44f6 100644
--- a/src/cmd/compile/internal/ssa/decompose.go
+++ b/src/cmd/compile/internal/ssa/decompose.go
@@ -219,10 +219,6 @@ func decomposeInterfacePhi(v *Value) {
 	v.AddArg(data)
 }
 
-func decomposeArgs(f *Func) {
-	applyRewrite(f, rewriteBlockdecArgs, rewriteValuedecArgs, removeDeadValues)
-}
-
 func decomposeUser(f *Func) {
 	for _, b := range f.Blocks {
 		for _, v := range b.Values {
diff --git a/src/cmd/compile/internal/ssa/expand_calls.go b/src/cmd/compile/internal/ssa/expand_calls.go
index af994d4b5b..579818e4f3 100644
--- a/src/cmd/compile/internal/ssa/expand_calls.go
+++ b/src/cmd/compile/internal/ssa/expand_calls.go
@@ -28,658 +28,666 @@ func isBlockMultiValueExit(b *Block) bool {
 	return (b.Kind == BlockRet || b.Kind == BlockRetJmp) && len(b.Controls) > 0 && b.Controls[0].Op == OpMakeResult
 }
 
-// expandCalls converts LE (Late Expansion) calls that act like they receive value args into a lower-level form
-// that is more oriented to a platform's ABI.  The SelectN operations that extract results are rewritten into
-// more appropriate forms, and any StructMake or ArrayMake inputs are decomposed until non-struct values are
-// reached.  On the callee side, OpArg nodes are not decomposed until this phase is run.
-// TODO results should not be lowered until this phase.
-func expandCalls(f *Func) {
-	// Calls that need lowering have some number of inputs, including a memory input,
-	// and produce a tuple of (value1, value2, ..., mem) where valueK may or may not be SSA-able.
+// removeTrivialWrapperTypes unwraps layers of
+// struct { singleField SomeType } and [1]SomeType
+// until a non-wrapper type is reached.  This is useful
+// for working with assignments to/from interface data
+// fields (either second operand to OpIMake or OpIData)
+// where the wrapping or type conversion can be elided
+// because of type conversions/assertions in source code
+// that do not appear in SSA.
+func removeTrivialWrapperTypes(t *types.Type) *types.Type {
+	for {
+		if t.IsStruct() && t.NumFields() == 1 {
+			t = t.Field(0).Type
+			continue
+		}
+		if t.IsArray() && t.NumElem() == 1 {
+			t = t.Elem()
+			continue
+		}
+		break
+	}
+	return t
+}
 
-	// With the current ABI those inputs need to be converted into stores to memory,
-	// rethreading the call's memory input to the first, and the new call now receiving the last.
+type expandState struct {
+	f            *Func
+	debug        bool
+	canSSAType   func(*types.Type) bool
+	regSize      int64
+	sp           *Value
+	typs         *Types
+	ptrSize      int64
+	hiOffset     int64
+	lowOffset    int64
+	namedSelects map[*Value][]namedVal
+	sdom         SparseTree
+	common       map[selKey]*Value
+	offsets      map[offsetKey]*Value
+}
 
-	// With the current ABI, the outputs need to be converted to loads, which will all use the call's
-	// memory output as their input.
-	if !LateCallExpansionEnabledWithin(f) {
-		return
+// intPairTypes returns the pair of 32-bit int types needed to encode a 64-bit integer type on a target
+// that has no 64-bit integer registers.
+func (x *expandState) intPairTypes(et types.Kind) (tHi, tLo *types.Type) {
+	tHi = x.typs.UInt32
+	if et == types.TINT64 {
+		tHi = x.typs.Int32
 	}
-	debug := f.pass.debug > 0
+	tLo = x.typs.UInt32
+	return
+}
 
-	if debug {
-		fmt.Printf("\nexpandsCalls(%s)\n", f.Name)
+// isAlreadyExpandedAggregateType returns whether a type is an SSA-able "aggregate" (multiple register) type
+// that was expanded in an earlier phase (currently, expand_calls is intended to run after decomposeBuiltin,
+// so this is all aggregate types -- small struct and array, complex, interface, string, slice, and 64-bit
+// integer on 32-bit).
+func (x *expandState) isAlreadyExpandedAggregateType(t *types.Type) bool {
+	if !x.canSSAType(t) {
+		return false
 	}
+	return t.IsStruct() || t.IsArray() || t.IsComplex() || t.IsInterface() || t.IsString() || t.IsSlice() ||
+		t.Size() > x.regSize && t.IsInteger()
+}
 
-	canSSAType := f.fe.CanSSA
-	regSize := f.Config.RegSize
-	sp, _ := f.spSb()
-	typ := &f.Config.Types
-	ptrSize := f.Config.PtrSize
+// offsetFrom creates an offset from a pointer, simplifying chained offsets and offsets from SP
+// TODO should also optimize offsets from SB?
+func (x *expandState) offsetFrom(from *Value, offset int64, pt *types.Type) *Value {
+	if offset == 0 && from.Type == pt { // this is not actually likely
+		return from
+	}
+	// Simplify, canonicalize
+	for from.Op == OpOffPtr {
+		offset += from.AuxInt
+		from = from.Args[0]
+	}
+	if from == x.sp {
+		return x.f.ConstOffPtrSP(pt, offset, x.sp)
+	}
+	key := offsetKey{from, offset, pt}
+	v := x.offsets[key]
+	if v != nil {
+		return v
+	}
+	v = from.Block.NewValue1I(from.Pos.WithNotStmt(), OpOffPtr, pt, offset, from)
+	x.offsets[key] = v
+	return v
+}
 
-	// For 32-bit, need to deal with decomposition of 64-bit integers, which depends on endianness.
-	var hiOffset, lowOffset int64
-	if f.Config.BigEndian {
-		lowOffset = 4
-	} else {
-		hiOffset = 4
+// splitSlots splits one "field" (specified by sfx, offset, and ty) out of the LocalSlots in ls and returns the new LocalSlots this generates.
+func (x *expandState) splitSlots(ls []LocalSlot, sfx string, offset int64, ty *types.Type) []LocalSlot {
+	var locs []LocalSlot
+	for i := range ls {
+		locs = append(locs, x.f.fe.SplitSlot(&ls[i], sfx, offset, ty))
 	}
+	return locs
+}
 
-	namedSelects := make(map[*Value][]namedVal)
+// Calls that need lowering have some number of inputs, including a memory input,
+// and produce a tuple of (value1, value2, ..., mem) where valueK may or may not be SSA-able.
 
-	sdom := f.Sdom()
+// With the current ABI those inputs need to be converted into stores to memory,
+// rethreading the call's memory input to the first, and the new call now receiving the last.
 
-	common := make(map[selKey]*Value)
+// With the current ABI, the outputs need to be converted to loads, which will all use the call's
+// memory output as their input.
 
-	// intPairTypes returns the pair of 32-bit int types needed to encode a 64-bit integer type on a target
-	// that has no 64-bit integer registers.
-	intPairTypes := func(et types.Kind) (tHi, tLo *types.Type) {
-		tHi = typ.UInt32
-		if et == types.TINT64 {
-			tHi = typ.Int32
-		}
-		tLo = typ.UInt32
-		return
+// rewriteSelect recursively walks from leaf selector to a root (OpSelectN, OpLoad, OpArg)
+// through a chain of Struct/Array/builtin Select operations.  If the chain of selectors does not
+// end in an expected root, it does nothing (this can happen depending on compiler phase ordering).
+// The "leaf" provides the type, the root supplies the container, and the leaf-to-root path
+// accumulates the offset.
+// It emits the code necessary to implement the leaf select operation that leads to the root.
+//
+// TODO when registers really arrive, must also decompose anything split across two registers or registers and memory.
+func (x *expandState) rewriteSelect(leaf *Value, selector *Value, offset int64) []LocalSlot {
+	if x.debug {
+		fmt.Printf("rewriteSelect(%s, %s, %d)\n", leaf.LongString(), selector.LongString(), offset)
 	}
-
-	// isAlreadyExpandedAggregateType returns whether a type is an SSA-able "aggregate" (multiple register) type
-	// that was expanded in an earlier phase (currently, expand_calls is intended to run after decomposeBuiltin,
-	// so this is all aggregate types -- small struct and array, complex, interface, string, slice, and 64-bit
-	// integer on 32-bit).
-	isAlreadyExpandedAggregateType := func(t *types.Type) bool {
-		if !canSSAType(t) {
-			return false
+	var locs []LocalSlot
+	leafType := leaf.Type
+	if len(selector.Args) > 0 {
+		w := selector.Args[0]
+		if w.Op == OpCopy {
+			for w.Op == OpCopy {
+				w = w.Args[0]
+			}
+			selector.SetArg(0, w)
 		}
-		return t.IsStruct() || t.IsArray() || t.IsComplex() || t.IsInterface() || t.IsString() || t.IsSlice() ||
-			t.Size() > regSize && t.IsInteger()
 	}
-
-	offsets := make(map[offsetKey]*Value)
-
-	// offsetFrom creates an offset from a pointer, simplifying chained offsets and offsets from SP
-	// TODO should also optimize offsets from SB?
-	offsetFrom := func(from *Value, offset int64, pt *types.Type) *Value {
-		if offset == 0 && from.Type == pt { // this is not actually likely
-			return from
+	switch selector.Op {
+	case OpArg:
+		if !x.isAlreadyExpandedAggregateType(selector.Type) {
+			if leafType == selector.Type { // OpIData leads us here, sometimes.
+				leaf.copyOf(selector)
+			} else {
+				x.f.Fatalf("Unexpected OpArg type, selector=%s, leaf=%s\n", selector.LongString(), leaf.LongString())
+			}
+			if x.debug {
+				fmt.Printf("\tOpArg, break\n")
+			}
+			break
 		}
-		// Simplify, canonicalize
-		for from.Op == OpOffPtr {
-			offset += from.AuxInt
-			from = from.Args[0]
+		switch leaf.Op {
+		case OpIData, OpStructSelect, OpArraySelect:
+			leafType = removeTrivialWrapperTypes(leaf.Type)
 		}
-		if from == sp {
-			return f.ConstOffPtrSP(pt, offset, sp)
+		aux := selector.Aux
+		auxInt := selector.AuxInt + offset
+		if leaf.Block == selector.Block {
+			leaf.reset(OpArg)
+			leaf.Aux = aux
+			leaf.AuxInt = auxInt
+			leaf.Type = leafType
+		} else {
+			w := selector.Block.NewValue0IA(leaf.Pos, OpArg, leafType, auxInt, aux)
+			leaf.copyOf(w)
+			if x.debug {
+				fmt.Printf("\tnew %s\n", w.LongString())
+			}
 		}
-		key := offsetKey{from, offset, pt}
-		v := offsets[key]
-		if v != nil {
-			return v
+		for _, s := range x.namedSelects[selector] {
+			locs = append(locs, x.f.Names[s.locIndex])
 		}
-		v = from.Block.NewValue1I(from.Pos.WithNotStmt(), OpOffPtr, pt, offset, from)
-		offsets[key] = v
-		return v
-	}
 
-	// splitSlots splits one "field" (specified by sfx, offset, and ty) out of the LocalSlots in ls and returns the new LocalSlots this generates.
-	splitSlots := func(ls []LocalSlot, sfx string, offset int64, ty *types.Type) []LocalSlot {
-		var locs []LocalSlot
-		for i := range ls {
-			locs = append(locs, f.fe.SplitSlot(&ls[i], sfx, offset, ty))
+	case OpLoad: // We end up here because of IData of immediate structures.
+		// Failure case:
+		// (note the failure case is very rare; w/o this case, make.bash and run.bash both pass, as well as
+		// the hard cases of building {syscall,math,math/cmplx,math/bits,go/constant} on ppc64le and mips-softfloat).
+		//
+		// GOSSAFUNC='(*dumper).dump' go build -gcflags=-l -tags=math_big_pure_go cmd/compile/internal/gc
+		// cmd/compile/internal/gc/dump.go:136:14: internal compiler error: '(*dumper).dump': not lowered: v827, StructSelect PTR PTR
+		// b2: ← b1
+		// v20 (+142) = StaticLECall <interface {},mem> {AuxCall{reflect.Value.Interface([reflect.Value,0])[interface {},24]}} [40] v8 v1
+		// v21 (142) = SelectN <mem> [1] v20
+		// v22 (142) = SelectN <interface {}> [0] v20
+		// b15: ← b8
+		// v71 (+143) = IData <Nodes> v22 (v[Nodes])
+		// v73 (+146) = StaticLECall <[]*Node,mem> {AuxCall{"".Nodes.Slice([Nodes,0])[[]*Node,8]}} [32] v71 v21
+		//
+		// translates (w/o the "case OpLoad:" above) to:
+		//
+		// b2: ← b1
+		// v20 (+142) = StaticCall <mem> {AuxCall{reflect.Value.Interface([reflect.Value,0])[interface {},24]}} [40] v715
+		// v23 (142) = Load <*uintptr> v19 v20
+		// v823 (142) = IsNonNil <bool> v23
+		// v67 (+143) = Load <*[]*Node> v880 v20
+		// b15: ← b8
+		// v827 (146) = StructSelect <*[]*Node> [0] v67
+		// v846 (146) = Store <mem> {*[]*Node} v769 v827 v20
+		// v73 (+146) = StaticCall <mem> {AuxCall{"".Nodes.Slice([Nodes,0])[[]*Node,8]}} [32] v846
+		// i.e., the struct select is generated and remains in because it is not applied to an actual structure.
+		// The OpLoad was created to load the single field of the IData
+		// This case removes that StructSelect.
+		if leafType != selector.Type {
+			x.f.Fatalf("Unexpected Load as selector, leaf=%s, selector=%s\n", leaf.LongString(), selector.LongString())
 		}
-		return locs
-	}
-
-	// removeTrivialWrapperTypes unwraps layers of
-	// struct { singleField SomeType } and [1]SomeType
-	// until a non-wrapper type is reached.  This is useful
-	// for working with assignments to/from interface data
-	// fields (either second operand to OpIMake or OpIData)
-	// where the wrapping or type conversion can be elided
-	// because of type conversions/assertions in source code
-	// that do not appear in SSA.
-	removeTrivialWrapperTypes := func(t *types.Type) *types.Type {
-		for {
-			if t.IsStruct() && t.NumFields() == 1 {
-				t = t.Field(0).Type
-				continue
-			}
-			if t.IsArray() && t.NumElem() == 1 {
-				t = t.Elem()
-				continue
-			}
-			break
+		leaf.copyOf(selector)
+		for _, s := range x.namedSelects[selector] {
+			locs = append(locs, x.f.Names[s.locIndex])
 		}
-		return t
-	}
-
-	// Calls that need lowering have some number of inputs, including a memory input,
-	// and produce a tuple of (value1, value2, ..., mem) where valueK may or may not be SSA-able.
-
-	// With the current ABI those inputs need to be converted into stores to memory,
-	// rethreading the call's memory input to the first, and the new call now receiving the last.
-
-	// With the current ABI, the outputs need to be converted to loads, which will all use the call's
-	// memory output as their input.
 
-	// rewriteSelect recursively walks from leaf selector to a root (OpSelectN, OpLoad, OpArg)
-	// through a chain of Struct/Array/builtin Select operations.  If the chain of selectors does not
-	// end in an expected root, it does nothing (this can happen depending on compiler phase ordering).
-	// The "leaf" provides the type, the root supplies the container, and the leaf-to-root path
-	// accumulates the offset.
-	// It emits the code necessary to implement the leaf select operation that leads to the root.
-	//
-	// TODO when registers really arrive, must also decompose anything split across two registers or registers and memory.
-	var rewriteSelect func(leaf *Value, selector *Value, offset int64) []LocalSlot
-	rewriteSelect = func(leaf *Value, selector *Value, offset int64) []LocalSlot {
-		if debug {
-			fmt.Printf("rewriteSelect(%s, %s, %d)\n", leaf.LongString(), selector.LongString(), offset)
-		}
-		var locs []LocalSlot
-		leafType := leaf.Type
-		if len(selector.Args) > 0 {
-			w := selector.Args[0]
-			if w.Op == OpCopy {
-				for w.Op == OpCopy {
-					w = w.Args[0]
-				}
-				selector.SetArg(0, w)
-			}
-		}
-		switch selector.Op {
-		case OpArg:
-			if !isAlreadyExpandedAggregateType(selector.Type) {
-				if leafType == selector.Type { // OpIData leads us here, sometimes.
-					leaf.copyOf(selector)
+	case OpSelectN:
+		// TODO these may be duplicated. Should memoize. Intermediate selectors will go dead, no worries there.
+		call := selector.Args[0]
+		aux := call.Aux.(*AuxCall)
+		which := selector.AuxInt
+		if which == aux.NResults() { // mem is after the results.
+			// rewrite v as a Copy of call -- the replacement call will produce a mem.
+			leaf.copyOf(call)
+		} else {
+			leafType := removeTrivialWrapperTypes(leaf.Type)
+			if x.canSSAType(leafType) {
+				pt := types.NewPtr(leafType)
+				off := x.offsetFrom(x.sp, offset+aux.OffsetOfResult(which), pt)
+				// Any selection right out of the arg area/registers has to be same Block as call, use call as mem input.
+				if leaf.Block == call.Block {
+					leaf.reset(OpLoad)
+					leaf.SetArgs2(off, call)
+					leaf.Type = leafType
 				} else {
-					f.Fatalf("Unexpected OpArg type, selector=%s, leaf=%s\n", selector.LongString(), leaf.LongString())
+					w := call.Block.NewValue2(leaf.Pos, OpLoad, leafType, off, call)
+					leaf.copyOf(w)
+					if x.debug {
+						fmt.Printf("\tnew %s\n", w.LongString())
+					}
 				}
-				if debug {
-					fmt.Printf("\tOpArg, break\n")
+				for _, s := range x.namedSelects[selector] {
+					locs = append(locs, x.f.Names[s.locIndex])
 				}
-				break
-			}
-			switch leaf.Op {
-			case OpIData, OpStructSelect, OpArraySelect:
-				leafType = removeTrivialWrapperTypes(leaf.Type)
-			}
-			aux := selector.Aux
-			auxInt := selector.AuxInt + offset
-			if leaf.Block == selector.Block {
-				leaf.reset(OpArg)
-				leaf.Aux = aux
-				leaf.AuxInt = auxInt
-				leaf.Type = leafType
-			} else {
-				w := selector.Block.NewValue0IA(leaf.Pos, OpArg, leafType, auxInt, aux)
-				leaf.copyOf(w)
-				if debug {
-					fmt.Printf("\tnew %s\n", w.LongString())
-				}
-			}
-			for _, s := range namedSelects[selector] {
-				locs = append(locs, f.Names[s.locIndex])
-			}
-
-		case OpLoad: // We end up here because of IData of immediate structures.
-			// Failure case:
-			// (note the failure case is very rare; w/o this case, make.bash and run.bash both pass, as well as
-			// the hard cases of building {syscall,math,math/cmplx,math/bits,go/constant} on ppc64le and mips-softfloat).
-			//
-			// GOSSAFUNC='(*dumper).dump' go build -gcflags=-l -tags=math_big_pure_go cmd/compile/internal/gc
-			// cmd/compile/internal/gc/dump.go:136:14: internal compiler error: '(*dumper).dump': not lowered: v827, StructSelect PTR PTR
-			// b2: ← b1
-			// v20 (+142) = StaticLECall <interface {},mem> {AuxCall{reflect.Value.Interface([reflect.Value,0])[interface {},24]}} [40] v8 v1
-			// v21 (142) = SelectN <mem> [1] v20
-			// v22 (142) = SelectN <interface {}> [0] v20
-			// b15: ← b8
-			// v71 (+143) = IData <Nodes> v22 (v[Nodes])
-			// v73 (+146) = StaticLECall <[]*Node,mem> {AuxCall{"".Nodes.Slice([Nodes,0])[[]*Node,8]}} [32] v71 v21
-			//
-			// translates (w/o the "case OpLoad:" above) to:
-			//
-			// b2: ← b1
-			// v20 (+142) = StaticCall <mem> {AuxCall{reflect.Value.Interface([reflect.Value,0])[interface {},24]}} [40] v715
-			// v23 (142) = Load <*uintptr> v19 v20
-			// v823 (142) = IsNonNil <bool> v23
-			// v67 (+143) = Load <*[]*Node> v880 v20
-			// b15: ← b8
-			// v827 (146) = StructSelect <*[]*Node> [0] v67
-			// v846 (146) = Store <mem> {*[]*Node} v769 v827 v20
-			// v73 (+146) = StaticCall <mem> {AuxCall{"".Nodes.Slice([Nodes,0])[[]*Node,8]}} [32] v846
-			// i.e., the struct select is generated and remains in because it is not applied to an actual structure.
-			// The OpLoad was created to load the single field of the IData
-			// This case removes that StructSelect.
-			if leafType != selector.Type {
-				f.Fatalf("Unexpected Load as selector, leaf=%s, selector=%s\n", leaf.LongString(), selector.LongString())
-			}
-			leaf.copyOf(selector)
-			for _, s := range namedSelects[selector] {
-				locs = append(locs, f.Names[s.locIndex])
-			}
-
-		case OpSelectN:
-			// TODO these may be duplicated. Should memoize. Intermediate selectors will go dead, no worries there.
-			call := selector.Args[0]
-			aux := call.Aux.(*AuxCall)
-			which := selector.AuxInt
-			if which == aux.NResults() { // mem is after the results.
-				// rewrite v as a Copy of call -- the replacement call will produce a mem.
-				leaf.copyOf(call)
 			} else {
-				leafType := removeTrivialWrapperTypes(leaf.Type)
-				if canSSAType(leafType) {
-					pt := types.NewPtr(leafType)
-					off := offsetFrom(sp, offset+aux.OffsetOfResult(which), pt)
-					// Any selection right out of the arg area/registers has to be same Block as call, use call as mem input.
-					if leaf.Block == call.Block {
-						leaf.reset(OpLoad)
-						leaf.SetArgs2(off, call)
-						leaf.Type = leafType
-					} else {
-						w := call.Block.NewValue2(leaf.Pos, OpLoad, leafType, off, call)
-						leaf.copyOf(w)
-						if debug {
-							fmt.Printf("\tnew %s\n", w.LongString())
-						}
-					}
-					for _, s := range namedSelects[selector] {
-						locs = append(locs, f.Names[s.locIndex])
-					}
-				} else {
-					f.Fatalf("Should not have non-SSA-able OpSelectN, selector=%s", selector.LongString())
-				}
+				x.f.Fatalf("Should not have non-SSA-able OpSelectN, selector=%s", selector.LongString())
 			}
+		}
 
-		case OpStructSelect:
-			w := selector.Args[0]
-			var ls []LocalSlot
-			if w.Type.Kind() != types.TSTRUCT { // IData artifact
-				ls = rewriteSelect(leaf, w, offset)
-			} else {
-				ls = rewriteSelect(leaf, w, offset+w.Type.FieldOff(int(selector.AuxInt)))
-				if w.Op != OpIData {
-					for _, l := range ls {
-						locs = append(locs, f.fe.SplitStruct(l, int(selector.AuxInt)))
-					}
+	case OpStructSelect:
+		w := selector.Args[0]
+		var ls []LocalSlot
+		if w.Type.Kind() != types.TSTRUCT { // IData artifact
+			ls = x.rewriteSelect(leaf, w, offset)
+		} else {
+			ls = x.rewriteSelect(leaf, w, offset+w.Type.FieldOff(int(selector.AuxInt)))
+			if w.Op != OpIData {
+				for _, l := range ls {
+					locs = append(locs, x.f.fe.SplitStruct(l, int(selector.AuxInt)))
 				}
 			}
+		}
 
-		case OpArraySelect:
-			w := selector.Args[0]
-			rewriteSelect(leaf, w, offset+selector.Type.Size()*selector.AuxInt)
-
-		case OpInt64Hi:
-			w := selector.Args[0]
-			ls := rewriteSelect(leaf, w, offset+hiOffset)
-			locs = splitSlots(ls, ".hi", hiOffset, leafType)
+	case OpArraySelect:
+		w := selector.Args[0]
+		x.rewriteSelect(leaf, w, offset+selector.Type.Size()*selector.AuxInt)
 
-		case OpInt64Lo:
-			w := selector.Args[0]
-			ls := rewriteSelect(leaf, w, offset+lowOffset)
-			locs = splitSlots(ls, ".lo", lowOffset, leafType)
+	case OpInt64Hi:
+		w := selector.Args[0]
+		ls := x.rewriteSelect(leaf, w, offset+x.hiOffset)
+		locs = x.splitSlots(ls, ".hi", x.hiOffset, leafType)
 
-		case OpStringPtr:
-			ls := rewriteSelect(leaf, selector.Args[0], offset)
-			locs = splitSlots(ls, ".ptr", 0, typ.BytePtr)
+	case OpInt64Lo:
+		w := selector.Args[0]
+		ls := x.rewriteSelect(leaf, w, offset+x.lowOffset)
+		locs = x.splitSlots(ls, ".lo", x.lowOffset, leafType)
 
-		case OpSlicePtr:
-			w := selector.Args[0]
-			ls := rewriteSelect(leaf, w, offset)
-			locs = splitSlots(ls, ".ptr", 0, types.NewPtr(w.Type.Elem()))
+	case OpStringPtr:
+		ls := x.rewriteSelect(leaf, selector.Args[0], offset)
+		locs = x.splitSlots(ls, ".ptr", 0, x.typs.BytePtr)
 
-		case OpITab:
-			w := selector.Args[0]
-			ls := rewriteSelect(leaf, w, offset)
-			sfx := ".itab"
-			if w.Type.IsEmptyInterface() {
-				sfx = ".type"
-			}
-			locs = splitSlots(ls, sfx, 0, typ.Uintptr)
+	case OpSlicePtr:
+		w := selector.Args[0]
+		ls := x.rewriteSelect(leaf, w, offset)
+		locs = x.splitSlots(ls, ".ptr", 0, types.NewPtr(w.Type.Elem()))
 
-		case OpComplexReal:
-			ls := rewriteSelect(leaf, selector.Args[0], offset)
-			locs = splitSlots(ls, ".real", 0, leafType)
+	case OpITab:
+		w := selector.Args[0]
+		ls := x.rewriteSelect(leaf, w, offset)
+		sfx := ".itab"
+		if w.Type.IsEmptyInterface() {
+			sfx = ".type"
+		}
+		locs = x.splitSlots(ls, sfx, 0, x.typs.Uintptr)
 
-		case OpComplexImag:
-			ls := rewriteSelect(leaf, selector.Args[0], offset+leafType.Width) // result is FloatNN, width of result is offset of imaginary part.
-			locs = splitSlots(ls, ".imag", leafType.Width, leafType)
+	case OpComplexReal:
+		ls := x.rewriteSelect(leaf, selector.Args[0], offset)
+		locs = x.splitSlots(ls, ".real", 0, leafType)
 
-		case OpStringLen, OpSliceLen:
-			ls := rewriteSelect(leaf, selector.Args[0], offset+ptrSize)
-			locs = splitSlots(ls, ".len", ptrSize, leafType)
+	case OpComplexImag:
+		ls := x.rewriteSelect(leaf, selector.Args[0], offset+leafType.Width) // result is FloatNN, width of result is offset of imaginary part.
+		locs = x.splitSlots(ls, ".imag", leafType.Width, leafType)
 
-		case OpIData:
-			ls := rewriteSelect(leaf, selector.Args[0], offset+ptrSize)
-			locs = splitSlots(ls, ".data", ptrSize, leafType)
+	case OpStringLen, OpSliceLen:
+		ls := x.rewriteSelect(leaf, selector.Args[0], offset+x.ptrSize)
+		locs = x.splitSlots(ls, ".len", x.ptrSize, leafType)
 
-		case OpSliceCap:
-			ls := rewriteSelect(leaf, selector.Args[0], offset+2*ptrSize)
-			locs = splitSlots(ls, ".cap", 2*ptrSize, leafType)
+	case OpIData:
+		ls := x.rewriteSelect(leaf, selector.Args[0], offset+x.ptrSize)
+		locs = x.splitSlots(ls, ".data", x.ptrSize, leafType)
 
-		case OpCopy: // If it's an intermediate result, recurse
-			locs = rewriteSelect(leaf, selector.Args[0], offset)
-			for _, s := range namedSelects[selector] {
-				// this copy may have had its own name, preserve that, too.
-				locs = append(locs, f.Names[s.locIndex])
-			}
+	case OpSliceCap:
+		ls := x.rewriteSelect(leaf, selector.Args[0], offset+2*x.ptrSize)
+		locs = x.splitSlots(ls, ".cap", 2*x.ptrSize, leafType)
 
-		default:
-			// Ignore dead ends. These can occur if this phase is run before decompose builtin (which is not intended, but allowed).
+	case OpCopy: // If it's an intermediate result, recurse
+		locs = x.rewriteSelect(leaf, selector.Args[0], offset)
+		for _, s := range x.namedSelects[selector] {
+			// this copy may have had its own name, preserve that, too.
+			locs = append(locs, x.f.Names[s.locIndex])
 		}
 
-		return locs
+	default:
+		// Ignore dead ends. These can occur if this phase is run before decompose builtin (which is not intended, but allowed).
 	}
 
-	// storeArgOrLoad converts stores of SSA-able aggregate arguments (passed to a call) into a series of primitive-typed
-	// stores of non-aggregate types.  It recursively walks up a chain of selectors until it reaches a Load or an Arg.
-	// If it does not reach a Load or an Arg, nothing happens; this allows a little freedom in phase ordering.
-	var storeArgOrLoad func(pos src.XPos, b *Block, base, source, mem *Value, t *types.Type, offset int64) *Value
+	return locs
+}
 
-	// decomposeArgOrLoad is a helper for storeArgOrLoad.
-	// It decomposes a Load or an Arg into smaller parts, parameterized by the decomposeOne and decomposeTwo functions
-	// passed to it, and returns the new mem. If the type does not match one of the expected aggregate types, it returns nil instead.
-	decomposeArgOrLoad := func(pos src.XPos, b *Block, base, source, mem *Value, t *types.Type, offset int64,
-		decomposeOne func(pos src.XPos, b *Block, base, source, mem *Value, t1 *types.Type, offArg, offStore int64) *Value,
-		decomposeTwo func(pos src.XPos, b *Block, base, source, mem *Value, t1, t2 *types.Type, offArg, offStore int64) *Value) *Value {
-		u := source.Type
-		switch u.Kind() {
-		case types.TARRAY:
-			elem := u.Elem()
-			for i := int64(0); i < u.NumElem(); i++ {
-				elemOff := i * elem.Size()
-				mem = decomposeOne(pos, b, base, source, mem, elem, source.AuxInt+elemOff, offset+elemOff)
-				pos = pos.WithNotStmt()
-			}
-			return mem
-		case types.TSTRUCT:
-			for i := 0; i < u.NumFields(); i++ {
-				fld := u.Field(i)
-				mem = decomposeOne(pos, b, base, source, mem, fld.Type, source.AuxInt+fld.Offset, offset+fld.Offset)
-				pos = pos.WithNotStmt()
-			}
-			return mem
-		case types.TINT64, types.TUINT64:
-			if t.Width == regSize {
-				break
-			}
-			tHi, tLo := intPairTypes(t.Kind())
-			mem = decomposeOne(pos, b, base, source, mem, tHi, source.AuxInt+hiOffset, offset+hiOffset)
-			pos = pos.WithNotStmt()
-			return decomposeOne(pos, b, base, source, mem, tLo, source.AuxInt+lowOffset, offset+lowOffset)
-		case types.TINTER:
-			return decomposeTwo(pos, b, base, source, mem, typ.Uintptr, typ.BytePtr, source.AuxInt, offset)
-		case types.TSTRING:
-			return decomposeTwo(pos, b, base, source, mem, typ.BytePtr, typ.Int, source.AuxInt, offset)
-		case types.TCOMPLEX64:
-			return decomposeTwo(pos, b, base, source, mem, typ.Float32, typ.Float32, source.AuxInt, offset)
-		case types.TCOMPLEX128:
-			return decomposeTwo(pos, b, base, source, mem, typ.Float64, typ.Float64, source.AuxInt, offset)
-		case types.TSLICE:
-			mem = decomposeTwo(pos, b, base, source, mem, typ.BytePtr, typ.Int, source.AuxInt, offset)
-			return decomposeOne(pos, b, base, source, mem, typ.Int, source.AuxInt+2*ptrSize, offset+2*ptrSize)
-		}
-		return nil
+func (x *expandState) rewriteDereference(b *Block, base, a, mem *Value, offset, size int64, typ *types.Type, pos src.XPos) *Value {
+	source := a.Args[0]
+	dst := x.offsetFrom(base, offset, source.Type)
+	if a.Uses == 1 && a.Block == b {
+		a.reset(OpMove)
+		a.Pos = pos
+		a.Type = types.TypeMem
+		a.Aux = typ
+		a.AuxInt = size
+		a.SetArgs3(dst, source, mem)
+		mem = a
+	} else {
+		mem = b.NewValue3A(pos, OpMove, types.TypeMem, typ, dst, source, mem)
+		mem.AuxInt = size
 	}
+	return mem
+}
 
-	// storeOneArg creates a decomposed (one step) arg that is then stored.
-	// pos and b locate the store instruction, base is the base of the store target, source is the "base" of the value input,
-	// mem is the input mem, t is the type in question, and offArg and offStore are the offsets from the respective bases.
-	storeOneArg := func(pos src.XPos, b *Block, base, source, mem *Value, t *types.Type, offArg, offStore int64) *Value {
-		w := common[selKey{source, offArg, t.Width, t}]
-		if w == nil {
-			w = source.Block.NewValue0IA(source.Pos, OpArg, t, offArg, source.Aux)
-			common[selKey{source, offArg, t.Width, t}] = w
+// decomposeArgOrLoad is a helper for storeArgOrLoad.
+// It decomposes a Load or an Arg into smaller parts, parameterized by the decomposeOne and decomposeTwo functions
+// passed to it, and returns the new mem. If the type does not match one of the expected aggregate types, it returns nil instead.
+func (x *expandState) decomposeArgOrLoad(pos src.XPos, b *Block, base, source, mem *Value, t *types.Type, offset int64,
+	decomposeOne func(x *expandState, pos src.XPos, b *Block, base, source, mem *Value, t1 *types.Type, offArg, offStore int64) *Value,
+	decomposeTwo func(x *expandState, pos src.XPos, b *Block, base, source, mem *Value, t1, t2 *types.Type, offArg, offStore int64) *Value) *Value {
+	u := source.Type
+	switch u.Kind() {
+	case types.TARRAY:
+		elem := u.Elem()
+		for i := int64(0); i < u.NumElem(); i++ {
+			elemOff := i * elem.Size()
+			mem = decomposeOne(x, pos, b, base, source, mem, elem, source.AuxInt+elemOff, offset+elemOff)
+			pos = pos.WithNotStmt()
 		}
-		return storeArgOrLoad(pos, b, base, w, mem, t, offStore)
-	}
-
-	// storeOneLoad creates a decomposed (one step) load that is then stored.
-	storeOneLoad := func(pos src.XPos, b *Block, base, source, mem *Value, t *types.Type, offArg, offStore int64) *Value {
-		from := offsetFrom(source.Args[0], offArg, types.NewPtr(t))
-		w := source.Block.NewValue2(source.Pos, OpLoad, t, from, mem)
-		return storeArgOrLoad(pos, b, base, w, mem, t, offStore)
-	}
-
-	storeTwoArg := func(pos src.XPos, b *Block, base, source, mem *Value, t1, t2 *types.Type, offArg, offStore int64) *Value {
-		mem = storeOneArg(pos, b, base, source, mem, t1, offArg, offStore)
+		return mem
+	case types.TSTRUCT:
+		for i := 0; i < u.NumFields(); i++ {
+			fld := u.Field(i)
+			mem = decomposeOne(x, pos, b, base, source, mem, fld.Type, source.AuxInt+fld.Offset, offset+fld.Offset)
+			pos = pos.WithNotStmt()
+		}
+		return mem
+	case types.TINT64, types.TUINT64:
+		if t.Width == x.regSize {
+			break
+		}
+		tHi, tLo := x.intPairTypes(t.Kind())
+		mem = decomposeOne(x, pos, b, base, source, mem, tHi, source.AuxInt+x.hiOffset, offset+x.hiOffset)
 		pos = pos.WithNotStmt()
-		t1Size := t1.Size()
-		return storeOneArg(pos, b, base, source, mem, t2, offArg+t1Size, offStore+t1Size)
+		return decomposeOne(x, pos, b, base, source, mem, tLo, source.AuxInt+x.lowOffset, offset+x.lowOffset)
+	case types.TINTER:
+		return decomposeTwo(x, pos, b, base, source, mem, x.typs.Uintptr, x.typs.BytePtr, source.AuxInt, offset)
+	case types.TSTRING:
+		return decomposeTwo(x, pos, b, base, source, mem, x.typs.BytePtr, x.typs.Int, source.AuxInt, offset)
+	case types.TCOMPLEX64:
+		return decomposeTwo(x, pos, b, base, source, mem, x.typs.Float32, x.typs.Float32, source.AuxInt, offset)
+	case types.TCOMPLEX128:
+		return decomposeTwo(x, pos, b, base, source, mem, x.typs.Float64, x.typs.Float64, source.AuxInt, offset)
+	case types.TSLICE:
+		mem = decomposeTwo(x, pos, b, base, source, mem, x.typs.BytePtr, x.typs.Int, source.AuxInt, offset)
+		return decomposeOne(x, pos, b, base, source, mem, x.typs.Int, source.AuxInt+2*x.ptrSize, offset+2*x.ptrSize)
 	}
+	return nil
+}
 
-	storeTwoLoad := func(pos src.XPos, b *Block, base, source, mem *Value, t1, t2 *types.Type, offArg, offStore int64) *Value {
-		mem = storeOneLoad(pos, b, base, source, mem, t1, offArg, offStore)
-		pos = pos.WithNotStmt()
-		t1Size := t1.Size()
-		return storeOneLoad(pos, b, base, source, mem, t2, offArg+t1Size, offStore+t1Size)
+// storeOneArg creates a decomposed (one step) arg that is then stored.
+// pos and b locate the store instruction, base is the base of the store target, source is the "base" of the value input,
+// mem is the input mem, t is the type in question, and offArg and offStore are the offsets from the respective bases.
+func storeOneArg(x *expandState, pos src.XPos, b *Block, base, source, mem *Value, t *types.Type, offArg, offStore int64) *Value {
+	w := x.common[selKey{source, offArg, t.Width, t}]
+	if w == nil {
+		w = source.Block.NewValue0IA(source.Pos, OpArg, t, offArg, source.Aux)
+		x.common[selKey{source, offArg, t.Width, t}] = w
 	}
+	return x.storeArgOrLoad(pos, b, base, w, mem, t, offStore)
+}
 
-	storeArgOrLoad = func(pos src.XPos, b *Block, base, source, mem *Value, t *types.Type, offset int64) *Value {
-		if debug {
-			fmt.Printf("\tstoreArgOrLoad(%s;  %s;  %s;  %s; %d)\n", base.LongString(), source.LongString(), mem.String(), t.String(), offset)
-		}
+// storeOneLoad creates a decomposed (one step) load that is then stored.
+func storeOneLoad(x *expandState, pos src.XPos, b *Block, base, source, mem *Value, t *types.Type, offArg, offStore int64) *Value {
+	from := x.offsetFrom(source.Args[0], offArg, types.NewPtr(t))
+	w := source.Block.NewValue2(source.Pos, OpLoad, t, from, mem)
+	return x.storeArgOrLoad(pos, b, base, w, mem, t, offStore)
+}
 
-		switch source.Op {
-		case OpCopy:
-			return storeArgOrLoad(pos, b, base, source.Args[0], mem, t, offset)
+func storeTwoArg(x *expandState, pos src.XPos, b *Block, base, source, mem *Value, t1, t2 *types.Type, offArg, offStore int64) *Value {
+	mem = storeOneArg(x, pos, b, base, source, mem, t1, offArg, offStore)
+	pos = pos.WithNotStmt()
+	t1Size := t1.Size()
+	return storeOneArg(x, pos, b, base, source, mem, t2, offArg+t1Size, offStore+t1Size)
+}
 
-		case OpLoad:
-			ret := decomposeArgOrLoad(pos, b, base, source, mem, t, offset, storeOneLoad, storeTwoLoad)
-			if ret != nil {
-				return ret
-			}
+func storeTwoLoad(x *expandState, pos src.XPos, b *Block, base, source, mem *Value, t1, t2 *types.Type, offArg, offStore int64) *Value {
+	mem = storeOneLoad(x, pos, b, base, source, mem, t1, offArg, offStore)
+	pos = pos.WithNotStmt()
+	t1Size := t1.Size()
+	return storeOneLoad(x, pos, b, base, source, mem, t2, offArg+t1Size, offStore+t1Size)
+}
 
-		case OpArg:
-			ret := decomposeArgOrLoad(pos, b, base, source, mem, t, offset, storeOneArg, storeTwoArg)
-			if ret != nil {
-				return ret
-			}
+// storeArgOrLoad converts stores of SSA-able aggregate arguments (passed to a call) into a series of primitive-typed
+// stores of non-aggregate types.  It recursively walks up a chain of selectors until it reaches a Load or an Arg.
+// If it does not reach a Load or an Arg, nothing happens; this allows a little freedom in phase ordering.
+func (x *expandState) storeArgOrLoad(pos src.XPos, b *Block, base, source, mem *Value, t *types.Type, offset int64) *Value {
+	if x.debug {
+		fmt.Printf("\tstoreArgOrLoad(%s;  %s;  %s;  %s; %d)\n", base.LongString(), source.LongString(), mem.String(), t.String(), offset)
+	}
 
-		case OpArrayMake0, OpStructMake0:
-			return mem
+	switch source.Op {
+	case OpCopy:
+		return x.storeArgOrLoad(pos, b, base, source.Args[0], mem, t, offset)
 
-		case OpStructMake1, OpStructMake2, OpStructMake3, OpStructMake4:
-			for i := 0; i < t.NumFields(); i++ {
-				fld := t.Field(i)
-				mem = storeArgOrLoad(pos, b, base, source.Args[i], mem, fld.Type, offset+fld.Offset)
-				pos = pos.WithNotStmt()
-			}
-			return mem
+	case OpLoad:
+		ret := x.decomposeArgOrLoad(pos, b, base, source, mem, t, offset, storeOneLoad, storeTwoLoad)
+		if ret != nil {
+			return ret
+		}
 
-		case OpArrayMake1:
-			return storeArgOrLoad(pos, b, base, source.Args[0], mem, t.Elem(), offset)
+	case OpArg:
+		ret := x.decomposeArgOrLoad(pos, b, base, source, mem, t, offset, storeOneArg, storeTwoArg)
+		if ret != nil {
+			return ret
+		}
 
-		case OpInt64Make:
-			tHi, tLo := intPairTypes(t.Kind())
-			mem = storeArgOrLoad(pos, b, base, source.Args[0], mem, tHi, offset+hiOffset)
-			pos = pos.WithNotStmt()
-			return storeArgOrLoad(pos, b, base, source.Args[1], mem, tLo, offset+lowOffset)
+	case OpArrayMake0, OpStructMake0:
+		return mem
 
-		case OpComplexMake:
-			tPart := typ.Float32
-			wPart := t.Width / 2
-			if wPart == 8 {
-				tPart = typ.Float64
-			}
-			mem = storeArgOrLoad(pos, b, base, source.Args[0], mem, tPart, offset)
+	case OpStructMake1, OpStructMake2, OpStructMake3, OpStructMake4:
+		for i := 0; i < t.NumFields(); i++ {
+			fld := t.Field(i)
+			mem = x.storeArgOrLoad(pos, b, base, source.Args[i], mem, fld.Type, offset+fld.Offset)
 			pos = pos.WithNotStmt()
-			return storeArgOrLoad(pos, b, base, source.Args[1], mem, tPart, offset+wPart)
+		}
+		return mem
 
-		case OpIMake:
-			mem = storeArgOrLoad(pos, b, base, source.Args[0], mem, typ.Uintptr, offset)
-			pos = pos.WithNotStmt()
-			return storeArgOrLoad(pos, b, base, source.Args[1], mem, typ.BytePtr, offset+ptrSize)
+	case OpArrayMake1:
+		return x.storeArgOrLoad(pos, b, base, source.Args[0], mem, t.Elem(), offset)
 
-		case OpStringMake:
-			mem = storeArgOrLoad(pos, b, base, source.Args[0], mem, typ.BytePtr, offset)
-			pos = pos.WithNotStmt()
-			return storeArgOrLoad(pos, b, base, source.Args[1], mem, typ.Int, offset+ptrSize)
+	case OpInt64Make:
+		tHi, tLo := x.intPairTypes(t.Kind())
+		mem = x.storeArgOrLoad(pos, b, base, source.Args[0], mem, tHi, offset+x.hiOffset)
+		pos = pos.WithNotStmt()
+		return x.storeArgOrLoad(pos, b, base, source.Args[1], mem, tLo, offset+x.lowOffset)
 
-		case OpSliceMake:
-			mem = storeArgOrLoad(pos, b, base, source.Args[0], mem, typ.BytePtr, offset)
-			pos = pos.WithNotStmt()
-			mem = storeArgOrLoad(pos, b, base, source.Args[1], mem, typ.Int, offset+ptrSize)
-			return storeArgOrLoad(pos, b, base, source.Args[2], mem, typ.Int, offset+2*ptrSize)
+	case OpComplexMake:
+		tPart := x.typs.Float32
+		wPart := t.Width / 2
+		if wPart == 8 {
+			tPart = x.typs.Float64
 		}
+		mem = x.storeArgOrLoad(pos, b, base, source.Args[0], mem, tPart, offset)
+		pos = pos.WithNotStmt()
+		return x.storeArgOrLoad(pos, b, base, source.Args[1], mem, tPart, offset+wPart)
 
-		// For nodes that cannot be taken apart -- OpSelectN, other structure selectors.
-		switch t.Kind() {
-		case types.TARRAY:
-			elt := t.Elem()
-			if source.Type != t && t.NumElem() == 1 && elt.Width == t.Width && t.Width == regSize {
-				t = removeTrivialWrapperTypes(t)
-				// it could be a leaf type, but the "leaf" could be complex64 (for example)
-				return storeArgOrLoad(pos, b, base, source, mem, t, offset)
-			}
-			for i := int64(0); i < t.NumElem(); i++ {
-				sel := source.Block.NewValue1I(pos, OpArraySelect, elt, i, source)
-				mem = storeArgOrLoad(pos, b, base, sel, mem, elt, offset+i*elt.Width)
-				pos = pos.WithNotStmt()
-			}
-			return mem
+	case OpIMake:
+		mem = x.storeArgOrLoad(pos, b, base, source.Args[0], mem, x.typs.Uintptr, offset)
+		pos = pos.WithNotStmt()
+		return x.storeArgOrLoad(pos, b, base, source.Args[1], mem, x.typs.BytePtr, offset+x.ptrSize)
 
-		case types.TSTRUCT:
-			if source.Type != t && t.NumFields() == 1 && t.Field(0).Type.Width == t.Width && t.Width == regSize {
-				// This peculiar test deals with accesses to immediate interface data.
-				// It works okay because everything is the same size.
-				// Example code that triggers this can be found in go/constant/value.go, function ToComplex
-				// v119 (+881) = IData <intVal> v6
-				// v121 (+882) = StaticLECall <floatVal,mem> {AuxCall{"".itof([intVal,0])[floatVal,8]}} [16] v119 v1
-				// This corresponds to the generic rewrite rule "(StructSelect [0] (IData x)) => (IData x)"
-				// Guard against "struct{struct{*foo}}"
-				// Other rewriting phases create minor glitches when they transform IData, for instance the
-				// interface-typed Arg "x" of ToFloat in go/constant/value.go
-				//   v6 (858) = Arg <Value> {x} (x[Value], x[Value])
-				// is rewritten by decomposeArgs into
-				//   v141 (858) = Arg <uintptr> {x}
-				//   v139 (858) = Arg <*uint8> {x} [8]
-				// because of a type case clause on line 862 of go/constant/value.go
-				//  	case intVal:
-				//		   return itof(x)
-				// v139 is later stored as an intVal == struct{val *big.Int} which naively requires the fields of
-				// of a *uint8, which does not succeed.
-				t = removeTrivialWrapperTypes(t)
-				// it could be a leaf type, but the "leaf" could be complex64 (for example)
-				return storeArgOrLoad(pos, b, base, source, mem, t, offset)
-			}
+	case OpStringMake:
+		mem = x.storeArgOrLoad(pos, b, base, source.Args[0], mem, x.typs.BytePtr, offset)
+		pos = pos.WithNotStmt()
+		return x.storeArgOrLoad(pos, b, base, source.Args[1], mem, x.typs.Int, offset+x.ptrSize)
 
-			for i := 0; i < t.NumFields(); i++ {
-				fld := t.Field(i)
-				sel := source.Block.NewValue1I(pos, OpStructSelect, fld.Type, int64(i), source)
-				mem = storeArgOrLoad(pos, b, base, sel, mem, fld.Type, offset+fld.Offset)
-				pos = pos.WithNotStmt()
-			}
-			return mem
+	case OpSliceMake:
+		mem = x.storeArgOrLoad(pos, b, base, source.Args[0], mem, x.typs.BytePtr, offset)
+		pos = pos.WithNotStmt()
+		mem = x.storeArgOrLoad(pos, b, base, source.Args[1], mem, x.typs.Int, offset+x.ptrSize)
+		return x.storeArgOrLoad(pos, b, base, source.Args[2], mem, x.typs.Int, offset+2*x.ptrSize)
+	}
 
-		case types.TINT64, types.TUINT64:
-			if t.Width == regSize {
-				break
-			}
-			tHi, tLo := intPairTypes(t.Kind())
-			sel := source.Block.NewValue1(pos, OpInt64Hi, tHi, source)
-			mem = storeArgOrLoad(pos, b, base, sel, mem, tHi, offset+hiOffset)
+	// For nodes that cannot be taken apart -- OpSelectN, other structure selectors.
+	switch t.Kind() {
+	case types.TARRAY:
+		elt := t.Elem()
+		if source.Type != t && t.NumElem() == 1 && elt.Width == t.Width && t.Width == x.regSize {
+			t = removeTrivialWrapperTypes(t)
+			// it could be a leaf type, but the "leaf" could be complex64 (for example)
+			return x.storeArgOrLoad(pos, b, base, source, mem, t, offset)
+		}
+		for i := int64(0); i < t.NumElem(); i++ {
+			sel := source.Block.NewValue1I(pos, OpArraySelect, elt, i, source)
+			mem = x.storeArgOrLoad(pos, b, base, sel, mem, elt, offset+i*elt.Width)
 			pos = pos.WithNotStmt()
-			sel = source.Block.NewValue1(pos, OpInt64Lo, tLo, source)
-			return storeArgOrLoad(pos, b, base, sel, mem, tLo, offset+lowOffset)
+		}
+		return mem
 
-		case types.TINTER:
-			sel := source.Block.NewValue1(pos, OpITab, typ.BytePtr, source)
-			mem = storeArgOrLoad(pos, b, base, sel, mem, typ.BytePtr, offset)
-			pos = pos.WithNotStmt()
-			sel = source.Block.NewValue1(pos, OpIData, typ.BytePtr, source)
-			return storeArgOrLoad(pos, b, base, sel, mem, typ.BytePtr, offset+ptrSize)
+	case types.TSTRUCT:
+		if source.Type != t && t.NumFields() == 1 && t.Field(0).Type.Width == t.Width && t.Width == x.regSize {
+			// This peculiar test deals with accesses to immediate interface data.
+			// It works okay because everything is the same size.
+			// Example code that triggers this can be found in go/constant/value.go, function ToComplex
+			// v119 (+881) = IData <intVal> v6
+			// v121 (+882) = StaticLECall <floatVal,mem> {AuxCall{"".itof([intVal,0])[floatVal,8]}} [16] v119 v1
+			// This corresponds to the generic rewrite rule "(StructSelect [0] (IData x)) => (IData x)"
+			// Guard against "struct{struct{*foo}}"
+			// Other rewriting phases create minor glitches when they transform IData, for instance the
+			// interface-typed Arg "x" of ToFloat in go/constant/value.go
+			//   v6 (858) = Arg <Value> {x} (x[Value], x[Value])
+			// is rewritten by decomposeArgs into
+			//   v141 (858) = Arg <uintptr> {x}
+			//   v139 (858) = Arg <*uint8> {x} [8]
+			// because of a type case clause on line 862 of go/constant/value.go
+			//  	case intVal:
+			//		   return itof(x)
+			// v139 is later stored as an intVal == struct{val *big.Int} which naively requires the fields of
+			// of a *uint8, which does not succeed.
+			t = removeTrivialWrapperTypes(t)
+			// it could be a leaf type, but the "leaf" could be complex64 (for example)
+			return x.storeArgOrLoad(pos, b, base, source, mem, t, offset)
+		}
 
-		case types.TSTRING:
-			sel := source.Block.NewValue1(pos, OpStringPtr, typ.BytePtr, source)
-			mem = storeArgOrLoad(pos, b, base, sel, mem, typ.BytePtr, offset)
+		for i := 0; i < t.NumFields(); i++ {
+			fld := t.Field(i)
+			sel := source.Block.NewValue1I(pos, OpStructSelect, fld.Type, int64(i), source)
+			mem = x.storeArgOrLoad(pos, b, base, sel, mem, fld.Type, offset+fld.Offset)
 			pos = pos.WithNotStmt()
-			sel = source.Block.NewValue1(pos, OpStringLen, typ.Int, source)
-			return storeArgOrLoad(pos, b, base, sel, mem, typ.Int, offset+ptrSize)
+		}
+		return mem
 
-		case types.TSLICE:
-			et := types.NewPtr(t.Elem())
-			sel := source.Block.NewValue1(pos, OpSlicePtr, et, source)
-			mem = storeArgOrLoad(pos, b, base, sel, mem, et, offset)
-			pos = pos.WithNotStmt()
-			sel = source.Block.NewValue1(pos, OpSliceLen, typ.Int, source)
-			mem = storeArgOrLoad(pos, b, base, sel, mem, typ.Int, offset+ptrSize)
-			sel = source.Block.NewValue1(pos, OpSliceCap, typ.Int, source)
-			return storeArgOrLoad(pos, b, base, sel, mem, typ.Int, offset+2*ptrSize)
+	case types.TINT64, types.TUINT64:
+		if t.Width == x.regSize {
+			break
+		}
+		tHi, tLo := x.intPairTypes(t.Kind())
+		sel := source.Block.NewValue1(pos, OpInt64Hi, tHi, source)
+		mem = x.storeArgOrLoad(pos, b, base, sel, mem, tHi, offset+x.hiOffset)
+		pos = pos.WithNotStmt()
+		sel = source.Block.NewValue1(pos, OpInt64Lo, tLo, source)
+		return x.storeArgOrLoad(pos, b, base, sel, mem, tLo, offset+x.lowOffset)
 
-		case types.TCOMPLEX64:
-			sel := source.Block.NewValue1(pos, OpComplexReal, typ.Float32, source)
-			mem = storeArgOrLoad(pos, b, base, sel, mem, typ.Float32, offset)
-			pos = pos.WithNotStmt()
-			sel = source.Block.NewValue1(pos, OpComplexImag, typ.Float32, source)
-			return storeArgOrLoad(pos, b, base, sel, mem, typ.Float32, offset+4)
+	case types.TINTER:
+		sel := source.Block.NewValue1(pos, OpITab, x.typs.BytePtr, source)
+		mem = x.storeArgOrLoad(pos, b, base, sel, mem, x.typs.BytePtr, offset)
+		pos = pos.WithNotStmt()
+		sel = source.Block.NewValue1(pos, OpIData, x.typs.BytePtr, source)
+		return x.storeArgOrLoad(pos, b, base, sel, mem, x.typs.BytePtr, offset+x.ptrSize)
 
-		case types.TCOMPLEX128:
-			sel := source.Block.NewValue1(pos, OpComplexReal, typ.Float64, source)
-			mem = storeArgOrLoad(pos, b, base, sel, mem, typ.Float64, offset)
-			pos = pos.WithNotStmt()
-			sel = source.Block.NewValue1(pos, OpComplexImag, typ.Float64, source)
-			return storeArgOrLoad(pos, b, base, sel, mem, typ.Float64, offset+8)
-		}
+	case types.TSTRING:
+		sel := source.Block.NewValue1(pos, OpStringPtr, x.typs.BytePtr, source)
+		mem = x.storeArgOrLoad(pos, b, base, sel, mem, x.typs.BytePtr, offset)
+		pos = pos.WithNotStmt()
+		sel = source.Block.NewValue1(pos, OpStringLen, x.typs.Int, source)
+		return x.storeArgOrLoad(pos, b, base, sel, mem, x.typs.Int, offset+x.ptrSize)
 
-		dst := offsetFrom(base, offset, types.NewPtr(t))
-		x := b.NewValue3A(pos, OpStore, types.TypeMem, t, dst, source, mem)
-		if debug {
-			fmt.Printf("\t\tstoreArg returns %s\n", x.LongString())
-		}
-		return x
+	case types.TSLICE:
+		et := types.NewPtr(t.Elem())
+		sel := source.Block.NewValue1(pos, OpSlicePtr, et, source)
+		mem = x.storeArgOrLoad(pos, b, base, sel, mem, et, offset)
+		pos = pos.WithNotStmt()
+		sel = source.Block.NewValue1(pos, OpSliceLen, x.typs.Int, source)
+		mem = x.storeArgOrLoad(pos, b, base, sel, mem, x.typs.Int, offset+x.ptrSize)
+		sel = source.Block.NewValue1(pos, OpSliceCap, x.typs.Int, source)
+		return x.storeArgOrLoad(pos, b, base, sel, mem, x.typs.Int, offset+2*x.ptrSize)
+
+	case types.TCOMPLEX64:
+		sel := source.Block.NewValue1(pos, OpComplexReal, x.typs.Float32, source)
+		mem = x.storeArgOrLoad(pos, b, base, sel, mem, x.typs.Float32, offset)
+		pos = pos.WithNotStmt()
+		sel = source.Block.NewValue1(pos, OpComplexImag, x.typs.Float32, source)
+		return x.storeArgOrLoad(pos, b, base, sel, mem, x.typs.Float32, offset+4)
+
+	case types.TCOMPLEX128:
+		sel := source.Block.NewValue1(pos, OpComplexReal, x.typs.Float64, source)
+		mem = x.storeArgOrLoad(pos, b, base, sel, mem, x.typs.Float64, offset)
+		pos = pos.WithNotStmt()
+		sel = source.Block.NewValue1(pos, OpComplexImag, x.typs.Float64, source)
+		return x.storeArgOrLoad(pos, b, base, sel, mem, x.typs.Float64, offset+8)
 	}
 
-	rewriteDereference := func(b *Block, base, a, mem *Value, offset, size int64, typ *types.Type, pos src.XPos) *Value {
-		source := a.Args[0]
-		dst := offsetFrom(base, offset, source.Type)
-		if a.Uses == 1 && a.Block == b {
-			a.reset(OpMove)
-			a.Pos = pos
-			a.Type = types.TypeMem
-			a.Aux = typ
-			a.AuxInt = size
-			a.SetArgs3(dst, source, mem)
-			mem = a
-		} else {
-			mem = b.NewValue3A(pos, OpMove, types.TypeMem, typ, dst, source, mem)
-			mem.AuxInt = size
-		}
-		return mem
+	dst := x.offsetFrom(base, offset, types.NewPtr(t))
+	s := b.NewValue3A(pos, OpStore, types.TypeMem, t, dst, source, mem)
+	if x.debug {
+		fmt.Printf("\t\tstoreArg returns %s\n", s.LongString())
 	}
+	return s
+}
 
-	// rewriteArgs removes all the Args from a call and converts the call args into appropriate
-	// stores (or later, register movement).  Extra args for interface and closure calls are ignored,
-	// but removed.
-	rewriteArgs := func(v *Value, firstArg int) *Value {
-		// Thread the stores on the memory arg
-		aux := v.Aux.(*AuxCall)
-		pos := v.Pos.WithNotStmt()
-		m0 := v.MemoryArg()
-		mem := m0
-		for i, a := range v.Args {
-			if i < firstArg {
-				continue
-			}
-			if a == m0 { // mem is last.
-				break
+// rewriteArgs removes all the Args from a call and converts the call args into appropriate
+// stores (or later, register movement).  Extra args for interface and closure calls are ignored,
+// but removed.
+func (x *expandState) rewriteArgs(v *Value, firstArg int) *Value {
+	// Thread the stores on the memory arg
+	aux := v.Aux.(*AuxCall)
+	pos := v.Pos.WithNotStmt()
+	m0 := v.MemoryArg()
+	mem := m0
+	for i, a := range v.Args {
+		if i < firstArg {
+			continue
+		}
+		if a == m0 { // mem is last.
+			break
+		}
+		auxI := int64(i - firstArg)
+		if a.Op == OpDereference {
+			if a.MemoryArg() != m0 {
+				x.f.Fatalf("Op...LECall and OpDereference have mismatched mem, %s and %s", v.LongString(), a.LongString())
 			}
-			auxI := int64(i - firstArg)
-			if a.Op == OpDereference {
-				if a.MemoryArg() != m0 {
-					f.Fatalf("Op...LECall and OpDereference have mismatched mem, %s and %s", v.LongString(), a.LongString())
-				}
-				// "Dereference" of addressed (probably not-SSA-eligible) value becomes Move
-				// TODO this will be more complicated with registers in the picture.
-				mem = rewriteDereference(v.Block, sp, a, mem, aux.OffsetOfArg(auxI), aux.SizeOfArg(auxI), aux.TypeOfArg(auxI), pos)
-			} else {
-				if debug {
-					fmt.Printf("storeArg %s, %v, %d\n", a.LongString(), aux.TypeOfArg(auxI), aux.OffsetOfArg(auxI))
-				}
-				mem = storeArgOrLoad(pos, v.Block, sp, a, mem, aux.TypeOfArg(auxI), aux.OffsetOfArg(auxI))
+			// "Dereference" of addressed (probably not-SSA-eligible) value becomes Move
+			// TODO this will be more complicated with registers in the picture.
+			mem = x.rewriteDereference(v.Block, x.sp, a, mem, aux.OffsetOfArg(auxI), aux.SizeOfArg(auxI), aux.TypeOfArg(auxI), pos)
+		} else {
+			if x.debug {
+				fmt.Printf("storeArg %s, %v, %d\n", a.LongString(), aux.TypeOfArg(auxI), aux.OffsetOfArg(auxI))
 			}
+			mem = x.storeArgOrLoad(pos, v.Block, x.sp, a, mem, aux.TypeOfArg(auxI), aux.OffsetOfArg(auxI))
 		}
-		v.resetArgs()
-		return mem
+	}
+	v.resetArgs()
+	return mem
+}
+
+// expandCalls converts LE (Late Expansion) calls that act like they receive value args into a lower-level form
+// that is more oriented to a platform's ABI.  The SelectN operations that extract results are rewritten into
+// more appropriate forms, and any StructMake or ArrayMake inputs are decomposed until non-struct values are
+// reached.  On the callee side, OpArg nodes are not decomposed until this phase is run.
+// TODO results should not be lowered until this phase.
+func expandCalls(f *Func) {
+	// Calls that need lowering have some number of inputs, including a memory input,
+	// and produce a tuple of (value1, value2, ..., mem) where valueK may or may not be SSA-able.
+
+	// With the current ABI those inputs need to be converted into stores to memory,
+	// rethreading the call's memory input to the first, and the new call now receiving the last.
+
+	// With the current ABI, the outputs need to be converted to loads, which will all use the call's
+	// memory output as their input.
+	sp, _ := f.spSb()
+	x := &expandState{
+		f:            f,
+		debug:        f.pass.debug > 0,
+		canSSAType:   f.fe.CanSSA,
+		regSize:      f.Config.RegSize,
+		sp:           sp,
+		typs:         &f.Config.Types,
+		ptrSize:      f.Config.PtrSize,
+		namedSelects: make(map[*Value][]namedVal),
+		sdom:         f.Sdom(),
+		common:       make(map[selKey]*Value),
+		offsets:      make(map[offsetKey]*Value),
+	}
+
+	// For 32-bit, need to deal with decomposition of 64-bit integers, which depends on endianness.
+	if f.Config.BigEndian {
+		x.lowOffset = 4
+	} else {
+		x.hiOffset = 4
+	}
+
+	if x.debug {
+		fmt.Printf("\nexpandsCalls(%s)\n", f.Name)
 	}
 
 	// TODO if too slow, whole program iteration can be replaced w/ slices of appropriate values, accumulated in first loop here.
@@ -689,16 +697,16 @@ func expandCalls(f *Func) {
 		for _, v := range b.Values {
 			switch v.Op {
 			case OpStaticLECall:
-				mem := rewriteArgs(v, 0)
+				mem := x.rewriteArgs(v, 0)
 				v.SetArgs1(mem)
 			case OpClosureLECall:
 				code := v.Args[0]
 				context := v.Args[1]
-				mem := rewriteArgs(v, 2)
+				mem := x.rewriteArgs(v, 2)
 				v.SetArgs3(code, context, mem)
 			case OpInterLECall:
 				code := v.Args[0]
-				mem := rewriteArgs(v, 1)
+				mem := x.rewriteArgs(v, 1)
 				v.SetArgs2(code, mem)
 			}
 		}
@@ -715,7 +723,7 @@ func expandCalls(f *Func) {
 					break
 				}
 				auxType := aux.TypeOfResult(i)
-				auxBase := b.NewValue2A(v.Pos, OpLocalAddr, types.NewPtr(auxType), aux.results[i].Name, sp, mem)
+				auxBase := b.NewValue2A(v.Pos, OpLocalAddr, types.NewPtr(auxType), aux.results[i].Name, x.sp, mem)
 				auxOffset := int64(0)
 				auxSize := aux.SizeOfResult(i)
 				if a.Op == OpDereference {
@@ -727,7 +735,7 @@ func expandCalls(f *Func) {
 						}
 						continue
 					}
-					mem = rewriteDereference(v.Block, auxBase, a, mem, auxOffset, auxSize, auxType, pos)
+					mem = x.rewriteDereference(v.Block, auxBase, a, mem, auxOffset, auxSize, auxType, pos)
 				} else {
 					if a.Op == OpLoad && a.Args[0].Op == OpLocalAddr {
 						addr := a.Args[0]
@@ -735,7 +743,7 @@ func expandCalls(f *Func) {
 							continue
 						}
 					}
-					mem = storeArgOrLoad(v.Pos, b, auxBase, a, mem, aux.TypeOfResult(i), auxOffset)
+					mem = x.storeArgOrLoad(v.Pos, b, auxBase, a, mem, aux.TypeOfResult(i), auxOffset)
 				}
 			}
 			b.SetControl(mem)
@@ -745,11 +753,11 @@ func expandCalls(f *Func) {
 
 	for i, name := range f.Names {
 		t := name.Type
-		if isAlreadyExpandedAggregateType(t) {
+		if x.isAlreadyExpandedAggregateType(t) {
 			for j, v := range f.NamedValues[name] {
-				if v.Op == OpSelectN || v.Op == OpArg && isAlreadyExpandedAggregateType(v.Type) {
-					ns := namedSelects[v]
-					namedSelects[v] = append(ns, namedVal{locIndex: i, valIndex: j})
+				if v.Op == OpSelectN || v.Op == OpArg && x.isAlreadyExpandedAggregateType(v.Type) {
+					ns := x.namedSelects[v]
+					x.namedSelects[v] = append(ns, namedVal{locIndex: i, valIndex: j})
 				}
 			}
 		}
@@ -763,22 +771,22 @@ func expandCalls(f *Func) {
 				t := v.Aux.(*types.Type)
 				source := v.Args[1]
 				tSrc := source.Type
-				iAEATt := isAlreadyExpandedAggregateType(t)
+				iAEATt := x.isAlreadyExpandedAggregateType(t)
 
 				if !iAEATt {
 					// guarding against store immediate struct into interface data field -- store type is *uint8
 					// TODO can this happen recursively?
-					iAEATt = isAlreadyExpandedAggregateType(tSrc)
+					iAEATt = x.isAlreadyExpandedAggregateType(tSrc)
 					if iAEATt {
 						t = tSrc
 					}
 				}
 				if iAEATt {
-					if debug {
+					if x.debug {
 						fmt.Printf("Splitting store %s\n", v.LongString())
 					}
 					dst, mem := v.Args[0], v.Args[2]
-					mem = storeArgOrLoad(v.Pos, b, dst, source, mem, t, 0)
+					mem = x.storeArgOrLoad(v.Pos, b, dst, source, mem, t, 0)
 					v.copyOf(mem)
 				}
 			}
@@ -807,7 +815,7 @@ func expandCalls(f *Func) {
 				switch w.Op {
 				case OpStructSelect, OpArraySelect, OpSelectN, OpArg:
 					val2Preds[w] += 1
-					if debug {
+					if x.debug {
 						fmt.Printf("v2p[%s] = %d\n", w.LongString(), val2Preds[w])
 					}
 				}
@@ -816,18 +824,18 @@ func expandCalls(f *Func) {
 			case OpSelectN:
 				if _, ok := val2Preds[v]; !ok {
 					val2Preds[v] = 0
-					if debug {
+					if x.debug {
 						fmt.Printf("v2p[%s] = %d\n", v.LongString(), val2Preds[v])
 					}
 				}
 
 			case OpArg:
-				if !isAlreadyExpandedAggregateType(v.Type) {
+				if !x.isAlreadyExpandedAggregateType(v.Type) {
 					continue
 				}
 				if _, ok := val2Preds[v]; !ok {
 					val2Preds[v] = 0
-					if debug {
+					if x.debug {
 						fmt.Printf("v2p[%s] = %d\n", v.LongString(), val2Preds[v])
 					}
 				}
@@ -838,7 +846,7 @@ func expandCalls(f *Func) {
 				which := v.AuxInt
 				aux := call.Aux.(*AuxCall)
 				pt := v.Type
-				off := offsetFrom(sp, aux.OffsetOfResult(which), pt)
+				off := x.offsetFrom(x.sp, aux.OffsetOfResult(which), pt)
 				v.copyOf(off)
 			}
 		}
@@ -860,7 +868,7 @@ func expandCalls(f *Func) {
 		if bi == bj {
 			return vi.ID < vj.ID
 		}
-		return sdom.domorder(bi) > sdom.domorder(bj) // reverse the order to put dominators last.
+		return x.sdom.domorder(bi) > x.sdom.domorder(bj) // reverse the order to put dominators last.
 	}
 
 	// Accumulate order in allOrdered
@@ -894,7 +902,7 @@ func expandCalls(f *Func) {
 		}
 	}
 
-	common = make(map[selKey]*Value)
+	x.common = make(map[selKey]*Value)
 	// Rewrite duplicate selectors as copies where possible.
 	for i := len(allOrdered) - 1; i >= 0; i-- {
 		v := allOrdered[i]
@@ -926,26 +934,26 @@ func expandCalls(f *Func) {
 		case OpSelectN:
 			offset = w.Aux.(*AuxCall).OffsetOfResult(v.AuxInt)
 		case OpInt64Hi:
-			offset = hiOffset
+			offset = x.hiOffset
 		case OpInt64Lo:
-			offset = lowOffset
+			offset = x.lowOffset
 		case OpStringLen, OpSliceLen, OpIData:
-			offset = ptrSize
+			offset = x.ptrSize
 		case OpSliceCap:
-			offset = 2 * ptrSize
+			offset = 2 * x.ptrSize
 		case OpComplexImag:
 			offset = size
 		}
 		sk := selKey{from: w, size: size, offset: offset, typ: typ}
-		dupe := common[sk]
+		dupe := x.common[sk]
 		if dupe == nil {
-			common[sk] = v
-		} else if sdom.IsAncestorEq(dupe.Block, v.Block) {
+			x.common[sk] = v
+		} else if x.sdom.IsAncestorEq(dupe.Block, v.Block) {
 			v.copyOf(dupe)
 		} else {
 			// Because values are processed in dominator order, the old common[s] will never dominate after a miss is seen.
 			// Installing the new value might match some future values.
-			common[sk] = v
+			x.common[sk] = v
 		}
 	}
 
@@ -954,7 +962,7 @@ func expandCalls(f *Func) {
 
 	// Rewrite selectors.
 	for i, v := range allOrdered {
-		if debug {
+		if x.debug {
 			b := v.Block
 			fmt.Printf("allOrdered[%d] = b%d, %s, uses=%d\n", i, b.ID, v.LongString(), v.Uses)
 		}
@@ -965,13 +973,13 @@ func expandCalls(f *Func) {
 		if v.Op == OpCopy {
 			continue
 		}
-		locs := rewriteSelect(v, v, 0)
+		locs := x.rewriteSelect(v, v, 0)
 		// Install new names.
 		if v.Type.IsMemory() {
 			continue
 		}
 		// Leaf types may have debug locations
-		if !isAlreadyExpandedAggregateType(v.Type) {
+		if !x.isAlreadyExpandedAggregateType(v.Type) {
 			for _, l := range locs {
 				f.NamedValues[l] = append(f.NamedValues[l], v)
 			}
@@ -979,7 +987,7 @@ func expandCalls(f *Func) {
 			continue
 		}
 		// Not-leaf types that had debug locations need to lose them.
-		if ns, ok := namedSelects[v]; ok {
+		if ns, ok := x.namedSelects[v]; ok {
 			toDelete = append(toDelete, ns...)
 		}
 	}
diff --git a/src/cmd/compile/internal/ssa/gen/dec64.rules b/src/cmd/compile/internal/ssa/gen/dec64.rules
index 9297ed8d2e..b0f10d0a0f 100644
--- a/src/cmd/compile/internal/ssa/gen/dec64.rules
+++ b/src/cmd/compile/internal/ssa/gen/dec64.rules
@@ -42,20 +42,20 @@
 		(Store {hi.Type} dst hi mem))
 
 // These are not enabled during decomposeBuiltin if late call expansion, but they are always enabled for softFloat
-(Arg {n} [off]) && is64BitInt(v.Type) && !config.BigEndian && v.Type.IsSigned() && !(go116lateCallExpansion && b.Func.pass.name == "decompose builtin") =>
+(Arg {n} [off]) && is64BitInt(v.Type) && !config.BigEndian && v.Type.IsSigned() && !(b.Func.pass.name == "decompose builtin") =>
   (Int64Make
     (Arg <typ.Int32> {n} [off+4])
     (Arg <typ.UInt32> {n} [off]))
-(Arg {n} [off]) && is64BitInt(v.Type) && !config.BigEndian && !v.Type.IsSigned() && !(go116lateCallExpansion && b.Func.pass.name == "decompose builtin")  =>
+(Arg {n} [off]) && is64BitInt(v.Type) && !config.BigEndian && !v.Type.IsSigned() && !(b.Func.pass.name == "decompose builtin")  =>
   (Int64Make
     (Arg <typ.UInt32> {n} [off+4])
     (Arg <typ.UInt32> {n} [off]))
 
-(Arg {n} [off]) && is64BitInt(v.Type) && config.BigEndian && v.Type.IsSigned() && !(go116lateCallExpansion && b.Func.pass.name == "decompose builtin") =>
+(Arg {n} [off]) && is64BitInt(v.Type) && config.BigEndian && v.Type.IsSigned() && !(b.Func.pass.name == "decompose builtin") =>
   (Int64Make
     (Arg <typ.Int32> {n} [off])
     (Arg <typ.UInt32> {n} [off+4]))
-(Arg {n} [off]) && is64BitInt(v.Type) && config.BigEndian && !v.Type.IsSigned() && !(go116lateCallExpansion && b.Func.pass.name == "decompose builtin") =>
+(Arg {n} [off]) && is64BitInt(v.Type) && config.BigEndian && !v.Type.IsSigned() && !(b.Func.pass.name == "decompose builtin") =>
   (Int64Make
     (Arg <typ.UInt32> {n} [off])
     (Arg <typ.UInt32> {n} [off+4]))
diff --git a/src/cmd/compile/internal/ssa/gen/decArgs.rules b/src/cmd/compile/internal/ssa/gen/decArgs.rules
deleted file mode 100644
index 1c9a0bb23d..0000000000
--- a/src/cmd/compile/internal/ssa/gen/decArgs.rules
+++ /dev/null
@@ -1,58 +0,0 @@
-// Copyright 2018 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.
-
-// Decompose compound argument values
-// Do this early to simplify tracking names for debugging.
-
-(Arg {n} [off]) && v.Type.IsString() =>
-  (StringMake
-    (Arg <typ.BytePtr> {n} [off])
-    (Arg <typ.Int> {n} [off+int32(config.PtrSize)]))
-
-(Arg {n} [off]) && v.Type.IsSlice() =>
-  (SliceMake
-    (Arg <v.Type.Elem().PtrTo()> {n} [off])
-    (Arg <typ.Int> {n} [off+int32(config.PtrSize)])
-    (Arg <typ.Int> {n} [off+2*int32(config.PtrSize)]))
-
-(Arg {n} [off]) && v.Type.IsInterface() =>
-  (IMake
-    (Arg <typ.Uintptr> {n} [off])
-    (Arg <typ.BytePtr> {n} [off+int32(config.PtrSize)]))
-
-(Arg {n} [off]) && v.Type.IsComplex() && v.Type.Size() == 16 =>
-  (ComplexMake
-    (Arg <typ.Float64> {n} [off])
-    (Arg <typ.Float64> {n} [off+8]))
-
-(Arg {n} [off]) && v.Type.IsComplex() && v.Type.Size() == 8 =>
-  (ComplexMake
-    (Arg <typ.Float32> {n} [off])
-    (Arg <typ.Float32> {n} [off+4]))
-
-(Arg <t>) && t.IsStruct() && t.NumFields() == 0 && fe.CanSSA(t) =>
-  (StructMake0)
-(Arg <t> {n} [off]) && t.IsStruct() && t.NumFields() == 1 && fe.CanSSA(t) =>
-  (StructMake1
-    (Arg <t.FieldType(0)> {n} [off+int32(t.FieldOff(0))]))
-(Arg <t> {n} [off]) && t.IsStruct() && t.NumFields() == 2 && fe.CanSSA(t) =>
-  (StructMake2
-    (Arg <t.FieldType(0)> {n} [off+int32(t.FieldOff(0))])
-    (Arg <t.FieldType(1)> {n} [off+int32(t.FieldOff(1))]))
-(Arg <t> {n} [off]) && t.IsStruct() && t.NumFields() == 3 && fe.CanSSA(t) =>
-  (StructMake3
-    (Arg <t.FieldType(0)> {n} [off+int32(t.FieldOff(0))])
-    (Arg <t.FieldType(1)> {n} [off+int32(t.FieldOff(1))])
-    (Arg <t.FieldType(2)> {n} [off+int32(t.FieldOff(2))]))
-(Arg <t> {n} [off]) && t.IsStruct() && t.NumFields() == 4 && fe.CanSSA(t) =>
-  (StructMake4
-    (Arg <t.FieldType(0)> {n} [off+int32(t.FieldOff(0))])
-    (Arg <t.FieldType(1)> {n} [off+int32(t.FieldOff(1))])
-    (Arg <t.FieldType(2)> {n} [off+int32(t.FieldOff(2))])
-    (Arg <t.FieldType(3)> {n} [off+int32(t.FieldOff(3))]))
-
-(Arg <t>) && t.IsArray() && t.NumElem() == 0 =>
-  (ArrayMake0)
-(Arg <t> {n} [off]) && t.IsArray() && t.NumElem() == 1 && fe.CanSSA(t) =>
-  (ArrayMake1 (Arg <t.Elem()> {n} [off]))
diff --git a/src/cmd/compile/internal/ssa/gen/decArgsOps.go b/src/cmd/compile/internal/ssa/gen/decArgsOps.go
deleted file mode 100644
index b73d9d3976..0000000000
--- a/src/cmd/compile/internal/ssa/gen/decArgsOps.go
+++ /dev/null
@@ -1,20 +0,0 @@
-// Copyright 2018 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.
-
-// +build ignore
-
-package main
-
-var decArgsOps = []opData{}
-
-var decArgsBlocks = []blockData{}
-
-func init() {
-	archs = append(archs, arch{
-		name:    "decArgs",
-		ops:     decArgsOps,
-		blocks:  decArgsBlocks,
-		generic: true,
-	})
-}
diff --git a/src/cmd/compile/internal/ssa/rewritedec64.go b/src/cmd/compile/internal/ssa/rewritedec64.go
index c49bc8043e..60b727f45f 100644
--- a/src/cmd/compile/internal/ssa/rewritedec64.go
+++ b/src/cmd/compile/internal/ssa/rewritedec64.go
@@ -184,12 +184,12 @@ func rewriteValuedec64_OpArg(v *Value) bool {
 	config := b.Func.Config
 	typ := &b.Func.Config.Types
 	// match: (Arg {n} [off])
-	// cond: is64BitInt(v.Type) && !config.BigEndian && v.Type.IsSigned() && !(go116lateCallExpansion && b.Func.pass.name == "decompose builtin")
+	// cond: is64BitInt(v.Type) && !config.BigEndian && v.Type.IsSigned() && !(b.Func.pass.name == "decompose builtin")
 	// result: (Int64Make (Arg <typ.Int32> {n} [off+4]) (Arg <typ.UInt32> {n} [off]))
 	for {
 		off := auxIntToInt32(v.AuxInt)
 		n := auxToSym(v.Aux)
-		if !(is64BitInt(v.Type) && !config.BigEndian && v.Type.IsSigned() && !(go116lateCallExpansion && b.Func.pass.name == "decompose builtin")) {
+		if !(is64BitInt(v.Type) && !config.BigEndian && v.Type.IsSigned() && !(b.Func.pass.name == "decompose builtin")) {
 			break
 		}
 		v.reset(OpInt64Make)
@@ -203,12 +203,12 @@ func rewriteValuedec64_OpArg(v *Value) bool {
 		return true
 	}
 	// match: (Arg {n} [off])
-	// cond: is64BitInt(v.Type) && !config.BigEndian && !v.Type.IsSigned() && !(go116lateCallExpansion && b.Func.pass.name == "decompose builtin")
+	// cond: is64BitInt(v.Type) && !config.BigEndian && !v.Type.IsSigned() && !(b.Func.pass.name == "decompose builtin")
 	// result: (Int64Make (Arg <typ.UInt32> {n} [off+4]) (Arg <typ.UInt32> {n} [off]))
 	for {
 		off := auxIntToInt32(v.AuxInt)
 		n := auxToSym(v.Aux)
-		if !(is64BitInt(v.Type) && !config.BigEndian && !v.Type.IsSigned() && !(go116lateCallExpansion && b.Func.pass.name == "decompose builtin")) {
+		if !(is64BitInt(v.Type) && !config.BigEndian && !v.Type.IsSigned() && !(b.Func.pass.name == "decompose builtin")) {
 			break
 		}
 		v.reset(OpInt64Make)
@@ -222,12 +222,12 @@ func rewriteValuedec64_OpArg(v *Value) bool {
 		return true
 	}
 	// match: (Arg {n} [off])
-	// cond: is64BitInt(v.Type) && config.BigEndian && v.Type.IsSigned() && !(go116lateCallExpansion && b.Func.pass.name == "decompose builtin")
+	// cond: is64BitInt(v.Type) && config.BigEndian && v.Type.IsSigned() && !(b.Func.pass.name == "decompose builtin")
 	// result: (Int64Make (Arg <typ.Int32> {n} [off]) (Arg <typ.UInt32> {n} [off+4]))
 	for {
 		off := auxIntToInt32(v.AuxInt)
 		n := auxToSym(v.Aux)
-		if !(is64BitInt(v.Type) && config.BigEndian && v.Type.IsSigned() && !(go116lateCallExpansion && b.Func.pass.name == "decompose builtin")) {
+		if !(is64BitInt(v.Type) && config.BigEndian && v.Type.IsSigned() && !(b.Func.pass.name == "decompose builtin")) {
 			break
 		}
 		v.reset(OpInt64Make)
@@ -241,12 +241,12 @@ func rewriteValuedec64_OpArg(v *Value) bool {
 		return true
 	}
 	// match: (Arg {n} [off])
-	// cond: is64BitInt(v.Type) && config.BigEndian && !v.Type.IsSigned() && !(go116lateCallExpansion && b.Func.pass.name == "decompose builtin")
+	// cond: is64BitInt(v.Type) && config.BigEndian && !v.Type.IsSigned() && !(b.Func.pass.name == "decompose builtin")
 	// result: (Int64Make (Arg <typ.UInt32> {n} [off]) (Arg <typ.UInt32> {n} [off+4]))
 	for {
 		off := auxIntToInt32(v.AuxInt)
 		n := auxToSym(v.Aux)
-		if !(is64BitInt(v.Type) && config.BigEndian && !v.Type.IsSigned() && !(go116lateCallExpansion && b.Func.pass.name == "decompose builtin")) {
+		if !(is64BitInt(v.Type) && config.BigEndian && !v.Type.IsSigned() && !(b.Func.pass.name == "decompose builtin")) {
 			break
 		}
 		v.reset(OpInt64Make)
diff --git a/src/cmd/compile/internal/ssa/rewritedecArgs.go b/src/cmd/compile/internal/ssa/rewritedecArgs.go
deleted file mode 100644
index 23ff417eee..0000000000
--- a/src/cmd/compile/internal/ssa/rewritedecArgs.go
+++ /dev/null
@@ -1,247 +0,0 @@
-// Code generated from gen/decArgs.rules; DO NOT EDIT.
-// generated with: cd gen; go run *.go
-
-package ssa
-
-func rewriteValuedecArgs(v *Value) bool {
-	switch v.Op {
-	case OpArg:
-		return rewriteValuedecArgs_OpArg(v)
-	}
-	return false
-}
-func rewriteValuedecArgs_OpArg(v *Value) bool {
-	b := v.Block
-	config := b.Func.Config
-	fe := b.Func.fe
-	typ := &b.Func.Config.Types
-	// match: (Arg {n} [off])
-	// cond: v.Type.IsString()
-	// result: (StringMake (Arg <typ.BytePtr> {n} [off]) (Arg <typ.Int> {n} [off+int32(config.PtrSize)]))
-	for {
-		off := auxIntToInt32(v.AuxInt)
-		n := auxToSym(v.Aux)
-		if !(v.Type.IsString()) {
-			break
-		}
-		v.reset(OpStringMake)
-		v0 := b.NewValue0(v.Pos, OpArg, typ.BytePtr)
-		v0.AuxInt = int32ToAuxInt(off)
-		v0.Aux = symToAux(n)
-		v1 := b.NewValue0(v.Pos, OpArg, typ.Int)
-		v1.AuxInt = int32ToAuxInt(off + int32(config.PtrSize))
-		v1.Aux = symToAux(n)
-		v.AddArg2(v0, v1)
-		return true
-	}
-	// match: (Arg {n} [off])
-	// cond: v.Type.IsSlice()
-	// result: (SliceMake (Arg <v.Type.Elem().PtrTo()> {n} [off]) (Arg <typ.Int> {n} [off+int32(config.PtrSize)]) (Arg <typ.Int> {n} [off+2*int32(config.PtrSize)]))
-	for {
-		off := auxIntToInt32(v.AuxInt)
-		n := auxToSym(v.Aux)
-		if !(v.Type.IsSlice()) {
-			break
-		}
-		v.reset(OpSliceMake)
-		v0 := b.NewValue0(v.Pos, OpArg, v.Type.Elem().PtrTo())
-		v0.AuxInt = int32ToAuxInt(off)
-		v0.Aux = symToAux(n)
-		v1 := b.NewValue0(v.Pos, OpArg, typ.Int)
-		v1.AuxInt = int32ToAuxInt(off + int32(config.PtrSize))
-		v1.Aux = symToAux(n)
-		v2 := b.NewValue0(v.Pos, OpArg, typ.Int)
-		v2.AuxInt = int32ToAuxInt(off + 2*int32(config.PtrSize))
-		v2.Aux = symToAux(n)
-		v.AddArg3(v0, v1, v2)
-		return true
-	}
-	// match: (Arg {n} [off])
-	// cond: v.Type.IsInterface()
-	// result: (IMake (Arg <typ.Uintptr> {n} [off]) (Arg <typ.BytePtr> {n} [off+int32(config.PtrSize)]))
-	for {
-		off := auxIntToInt32(v.AuxInt)
-		n := auxToSym(v.Aux)
-		if !(v.Type.IsInterface()) {
-			break
-		}
-		v.reset(OpIMake)
-		v0 := b.NewValue0(v.Pos, OpArg, typ.Uintptr)
-		v0.AuxInt = int32ToAuxInt(off)
-		v0.Aux = symToAux(n)
-		v1 := b.NewValue0(v.Pos, OpArg, typ.BytePtr)
-		v1.AuxInt = int32ToAuxInt(off + int32(config.PtrSize))
-		v1.Aux = symToAux(n)
-		v.AddArg2(v0, v1)
-		return true
-	}
-	// match: (Arg {n} [off])
-	// cond: v.Type.IsComplex() && v.Type.Size() == 16
-	// result: (ComplexMake (Arg <typ.Float64> {n} [off]) (Arg <typ.Float64> {n} [off+8]))
-	for {
-		off := auxIntToInt32(v.AuxInt)
-		n := auxToSym(v.Aux)
-		if !(v.Type.IsComplex() && v.Type.Size() == 16) {
-			break
-		}
-		v.reset(OpComplexMake)
-		v0 := b.NewValue0(v.Pos, OpArg, typ.Float64)
-		v0.AuxInt = int32ToAuxInt(off)
-		v0.Aux = symToAux(n)
-		v1 := b.NewValue0(v.Pos, OpArg, typ.Float64)
-		v1.AuxInt = int32ToAuxInt(off + 8)
-		v1.Aux = symToAux(n)
-		v.AddArg2(v0, v1)
-		return true
-	}
-	// match: (Arg {n} [off])
-	// cond: v.Type.IsComplex() && v.Type.Size() == 8
-	// result: (ComplexMake (Arg <typ.Float32> {n} [off]) (Arg <typ.Float32> {n} [off+4]))
-	for {
-		off := auxIntToInt32(v.AuxInt)
-		n := auxToSym(v.Aux)
-		if !(v.Type.IsComplex() && v.Type.Size() == 8) {
-			break
-		}
-		v.reset(OpComplexMake)
-		v0 := b.NewValue0(v.Pos, OpArg, typ.Float32)
-		v0.AuxInt = int32ToAuxInt(off)
-		v0.Aux = symToAux(n)
-		v1 := b.NewValue0(v.Pos, OpArg, typ.Float32)
-		v1.AuxInt = int32ToAuxInt(off + 4)
-		v1.Aux = symToAux(n)
-		v.AddArg2(v0, v1)
-		return true
-	}
-	// match: (Arg <t>)
-	// cond: t.IsStruct() && t.NumFields() == 0 && fe.CanSSA(t)
-	// result: (StructMake0)
-	for {
-		t := v.Type
-		if !(t.IsStruct() && t.NumFields() == 0 && fe.CanSSA(t)) {
-			break
-		}
-		v.reset(OpStructMake0)
-		return true
-	}
-	// match: (Arg <t> {n} [off])
-	// cond: t.IsStruct() && t.NumFields() == 1 && fe.CanSSA(t)
-	// result: (StructMake1 (Arg <t.FieldType(0)> {n} [off+int32(t.FieldOff(0))]))
-	for {
-		t := v.Type
-		off := auxIntToInt32(v.AuxInt)
-		n := auxToSym(v.Aux)
-		if !(t.IsStruct() && t.NumFields() == 1 && fe.CanSSA(t)) {
-			break
-		}
-		v.reset(OpStructMake1)
-		v0 := b.NewValue0(v.Pos, OpArg, t.FieldType(0))
-		v0.AuxInt = int32ToAuxInt(off + int32(t.FieldOff(0)))
-		v0.Aux = symToAux(n)
-		v.AddArg(v0)
-		return true
-	}
-	// match: (Arg <t> {n} [off])
-	// cond: t.IsStruct() && t.NumFields() == 2 && fe.CanSSA(t)
-	// result: (StructMake2 (Arg <t.FieldType(0)> {n} [off+int32(t.FieldOff(0))]) (Arg <t.FieldType(1)> {n} [off+int32(t.FieldOff(1))]))
-	for {
-		t := v.Type
-		off := auxIntToInt32(v.AuxInt)
-		n := auxToSym(v.Aux)
-		if !(t.IsStruct() && t.NumFields() == 2 && fe.CanSSA(t)) {
-			break
-		}
-		v.reset(OpStructMake2)
-		v0 := b.NewValue0(v.Pos, OpArg, t.FieldType(0))
-		v0.AuxInt = int32ToAuxInt(off + int32(t.FieldOff(0)))
-		v0.Aux = symToAux(n)
-		v1 := b.NewValue0(v.Pos, OpArg, t.FieldType(1))
-		v1.AuxInt = int32ToAuxInt(off + int32(t.FieldOff(1)))
-		v1.Aux = symToAux(n)
-		v.AddArg2(v0, v1)
-		return true
-	}
-	// match: (Arg <t> {n} [off])
-	// cond: t.IsStruct() && t.NumFields() == 3 && fe.CanSSA(t)
-	// result: (StructMake3 (Arg <t.FieldType(0)> {n} [off+int32(t.FieldOff(0))]) (Arg <t.FieldType(1)> {n} [off+int32(t.FieldOff(1))]) (Arg <t.FieldType(2)> {n} [off+int32(t.FieldOff(2))]))
-	for {
-		t := v.Type
-		off := auxIntToInt32(v.AuxInt)
-		n := auxToSym(v.Aux)
-		if !(t.IsStruct() && t.NumFields() == 3 && fe.CanSSA(t)) {
-			break
-		}
-		v.reset(OpStructMake3)
-		v0 := b.NewValue0(v.Pos, OpArg, t.FieldType(0))
-		v0.AuxInt = int32ToAuxInt(off + int32(t.FieldOff(0)))
-		v0.Aux = symToAux(n)
-		v1 := b.NewValue0(v.Pos, OpArg, t.FieldType(1))
-		v1.AuxInt = int32ToAuxInt(off + int32(t.FieldOff(1)))
-		v1.Aux = symToAux(n)
-		v2 := b.NewValue0(v.Pos, OpArg, t.FieldType(2))
-		v2.AuxInt = int32ToAuxInt(off + int32(t.FieldOff(2)))
-		v2.Aux = symToAux(n)
-		v.AddArg3(v0, v1, v2)
-		return true
-	}
-	// match: (Arg <t> {n} [off])
-	// cond: t.IsStruct() && t.NumFields() == 4 && fe.CanSSA(t)
-	// result: (StructMake4 (Arg <t.FieldType(0)> {n} [off+int32(t.FieldOff(0))]) (Arg <t.FieldType(1)> {n} [off+int32(t.FieldOff(1))]) (Arg <t.FieldType(2)> {n} [off+int32(t.FieldOff(2))]) (Arg <t.FieldType(3)> {n} [off+int32(t.FieldOff(3))]))
-	for {
-		t := v.Type
-		off := auxIntToInt32(v.AuxInt)
-		n := auxToSym(v.Aux)
-		if !(t.IsStruct() && t.NumFields() == 4 && fe.CanSSA(t)) {
-			break
-		}
-		v.reset(OpStructMake4)
-		v0 := b.NewValue0(v.Pos, OpArg, t.FieldType(0))
-		v0.AuxInt = int32ToAuxInt(off + int32(t.FieldOff(0)))
-		v0.Aux = symToAux(n)
-		v1 := b.NewValue0(v.Pos, OpArg, t.FieldType(1))
-		v1.AuxInt = int32ToAuxInt(off + int32(t.FieldOff(1)))
-		v1.Aux = symToAux(n)
-		v2 := b.NewValue0(v.Pos, OpArg, t.FieldType(2))
-		v2.AuxInt = int32ToAuxInt(off + int32(t.FieldOff(2)))
-		v2.Aux = symToAux(n)
-		v3 := b.NewValue0(v.Pos, OpArg, t.FieldType(3))
-		v3.AuxInt = int32ToAuxInt(off + int32(t.FieldOff(3)))
-		v3.Aux = symToAux(n)
-		v.AddArg4(v0, v1, v2, v3)
-		return true
-	}
-	// match: (Arg <t>)
-	// cond: t.IsArray() && t.NumElem() == 0
-	// result: (ArrayMake0)
-	for {
-		t := v.Type
-		if !(t.IsArray() && t.NumElem() == 0) {
-			break
-		}
-		v.reset(OpArrayMake0)
-		return true
-	}
-	// match: (Arg <t> {n} [off])
-	// cond: t.IsArray() && t.NumElem() == 1 && fe.CanSSA(t)
-	// result: (ArrayMake1 (Arg <t.Elem()> {n} [off]))
-	for {
-		t := v.Type
-		off := auxIntToInt32(v.AuxInt)
-		n := auxToSym(v.Aux)
-		if !(t.IsArray() && t.NumElem() == 1 && fe.CanSSA(t)) {
-			break
-		}
-		v.reset(OpArrayMake1)
-		v0 := b.NewValue0(v.Pos, OpArg, t.Elem())
-		v0.AuxInt = int32ToAuxInt(off)
-		v0.Aux = symToAux(n)
-		v.AddArg(v0)
-		return true
-	}
-	return false
-}
-func rewriteBlockdecArgs(b *Block) bool {
-	switch b.Kind {
-	}
-	return false
-}
diff --git a/src/cmd/compile/internal/ssagen/ssa.go b/src/cmd/compile/internal/ssagen/ssa.go
index e49a9716fe..b042c132d5 100644
--- a/src/cmd/compile/internal/ssagen/ssa.go
+++ b/src/cmd/compile/internal/ssagen/ssa.go
@@ -1803,7 +1803,7 @@ const shareDeferExits = false
 // It returns a BlockRet block that ends the control flow. Its control value
 // will be set to the final memory state.
 func (s *state) exit() *ssa.Block {
-	lateResultLowering := s.f.DebugTest && ssa.LateCallExpansionEnabledWithin(s.f)
+	lateResultLowering := s.f.DebugTest
 	if s.hasdefer {
 		if s.hasOpenDefers {
 			if shareDeferExits && s.lastDeferExit != nil && len(s.openDefers) == s.lastDeferCount {
@@ -4628,7 +4628,6 @@ func (s *state) openDeferExit() {
 	s.lastDeferExit = deferExit
 	s.lastDeferCount = len(s.openDefers)
 	zeroval := s.constInt8(types.Types[types.TUINT8], 0)
-	testLateExpansion := ssa.LateCallExpansionEnabledWithin(s.f)
 	// Test for and run defers in reverse order
 	for i := len(s.openDefers) - 1; i >= 0; i-- {
 		r := s.openDefers[i]
@@ -4670,35 +4669,19 @@ func (s *state) openDeferExit() {
 		if r.rcvr != nil {
 			// rcvr in case of OCALLINTER
 			v := s.load(r.rcvr.Type.Elem(), r.rcvr)
-			addr := s.constOffPtrSP(s.f.Config.Types.UintptrPtr, argStart)
 			ACArgs = append(ACArgs, ssa.Param{Type: types.Types[types.TUINTPTR], Offset: int32(argStart)})
-			if testLateExpansion {
-				callArgs = append(callArgs, v)
-			} else {
-				s.store(types.Types[types.TUINTPTR], addr, v)
-			}
+			callArgs = append(callArgs, v)
 		}
 		for j, argAddrVal := range r.argVals {
 			f := getParam(r.n, j)
-			pt := types.NewPtr(f.Type)
 			ACArgs = append(ACArgs, ssa.Param{Type: f.Type, Offset: int32(argStart + f.Offset)})
-			if testLateExpansion {
-				var a *ssa.Value
-				if !TypeOK(f.Type) {
-					a = s.newValue2(ssa.OpDereference, f.Type, argAddrVal, s.mem())
-				} else {
-					a = s.load(f.Type, argAddrVal)
-				}
-				callArgs = append(callArgs, a)
+			var a *ssa.Value
+			if !TypeOK(f.Type) {
+				a = s.newValue2(ssa.OpDereference, f.Type, argAddrVal, s.mem())
 			} else {
-				addr := s.constOffPtrSP(pt, argStart+f.Offset)
-				if !TypeOK(f.Type) {
-					s.move(f.Type, addr, argAddrVal)
-				} else {
-					argVal := s.load(f.Type, argAddrVal)
-					s.storeType(f.Type, addr, argVal, 0, false)
-				}
+				a = s.load(f.Type, argAddrVal)
 			}
+			callArgs = append(callArgs, a)
 		}
 		var call *ssa.Value
 		if r.closure != nil {
@@ -4706,30 +4689,15 @@ func (s *state) openDeferExit() {
 			s.maybeNilCheckClosure(v, callDefer)
 			codeptr := s.rawLoad(types.Types[types.TUINTPTR], v)
 			aux := ssa.ClosureAuxCall(ACArgs, ACResults)
-			if testLateExpansion {
-				callArgs = append(callArgs, s.mem())
-				call = s.newValue2A(ssa.OpClosureLECall, aux.LateExpansionResultType(), aux, codeptr, v)
-				call.AddArgs(callArgs...)
-			} else {
-				call = s.newValue3A(ssa.OpClosureCall, types.TypeMem, aux, codeptr, v, s.mem())
-			}
+			call = s.newValue2A(ssa.OpClosureLECall, aux.LateExpansionResultType(), aux, codeptr, v)
 		} else {
 			aux := ssa.StaticAuxCall(fn.(*ir.Name).Linksym(), ACArgs, ACResults)
-			if testLateExpansion {
-				callArgs = append(callArgs, s.mem())
-				call = s.newValue0A(ssa.OpStaticLECall, aux.LateExpansionResultType(), aux)
-				call.AddArgs(callArgs...)
-			} else {
-				// Do a static call if the original call was a static function or method
-				call = s.newValue1A(ssa.OpStaticCall, types.TypeMem, aux, s.mem())
-			}
+			call = s.newValue0A(ssa.OpStaticLECall, aux.LateExpansionResultType(), aux)
 		}
+		callArgs = append(callArgs, s.mem())
+		call.AddArgs(callArgs...)
 		call.AuxInt = stksize
-		if testLateExpansion {
-			s.vars[memVar] = s.newValue1I(ssa.OpSelectN, types.TypeMem, int64(len(ACResults)), call)
-		} else {
-			s.vars[memVar] = call
-		}
+		s.vars[memVar] = s.newValue1I(ssa.OpSelectN, types.TypeMem, int64(len(ACResults)), call)
 		// Make sure that the stack slots with pointers are kept live
 		// through the call (which is a pre-emption point). Also, we will
 		// use the first call of the last defer exit to compute liveness
@@ -4782,12 +4750,10 @@ func (s *state) call(n *ir.CallExpr, k callKind, returnResultAddr bool) *ssa.Val
 		}
 	}
 
-	testLateExpansion := false
 	inRegisters := false
 
 	switch n.Op() {
 	case ir.OCALLFUNC:
-		testLateExpansion = k != callDeferStack && ssa.LateCallExpansionEnabledWithin(s.f)
 		if k == callNormal && fn.Op() == ir.ONAME && fn.(*ir.Name).Class == ir.PFUNC {
 			fn := fn.(*ir.Name)
 			callee = fn
@@ -4813,7 +4779,6 @@ func (s *state) call(n *ir.CallExpr, k callKind, returnResultAddr bool) *ssa.Val
 			s.Fatalf("OCALLINTER: n.Left not an ODOTINTER: %v", fn.Op())
 		}
 		fn := fn.(*ir.SelectorExpr)
-		testLateExpansion = k != callDeferStack && ssa.LateCallExpansionEnabledWithin(s.f)
 		var iclosure *ssa.Value
 		iclosure, rcvr = s.getClosureAndRcvr(fn)
 		if k == callNormal {
@@ -4827,7 +4792,6 @@ func (s *state) call(n *ir.CallExpr, k callKind, returnResultAddr bool) *ssa.Val
 
 	var call *ssa.Value
 	if k == callDeferStack {
-		testLateExpansion = ssa.LateCallExpansionEnabledWithin(s.f)
 		// Make a defer struct d on the stack.
 		t := deferstruct(stksize)
 		d := typecheck.TempAt(n.Pos(), s.curfn, t)
@@ -4878,15 +4842,9 @@ func (s *state) call(n *ir.CallExpr, k callKind, returnResultAddr bool) *ssa.Val
 		// Call runtime.deferprocStack with pointer to _defer record.
 		ACArgs = append(ACArgs, ssa.Param{Type: types.Types[types.TUINTPTR], Offset: int32(base.Ctxt.FixedFrameSize())})
 		aux := ssa.StaticAuxCall(ir.Syms.DeferprocStack, ACArgs, ACResults)
-		if testLateExpansion {
-			callArgs = append(callArgs, addr, s.mem())
-			call = s.newValue0A(ssa.OpStaticLECall, aux.LateExpansionResultType(), aux)
-			call.AddArgs(callArgs...)
-		} else {
-			arg0 := s.constOffPtrSP(types.Types[types.TUINTPTR], base.Ctxt.FixedFrameSize())
-			s.store(types.Types[types.TUINTPTR], arg0, addr)
-			call = s.newValue1A(ssa.OpStaticCall, types.TypeMem, aux, s.mem())
-		}
+		callArgs = append(callArgs, addr, s.mem())
+		call = s.newValue0A(ssa.OpStaticLECall, aux.LateExpansionResultType(), aux)
+		call.AddArgs(callArgs...)
 		if stksize < int64(types.PtrSize) {
 			// We need room for both the call to deferprocStack and the call to
 			// the deferred function.
@@ -4903,32 +4861,17 @@ func (s *state) call(n *ir.CallExpr, k callKind, returnResultAddr bool) *ssa.Val
 			// Write argsize and closure (args to newproc/deferproc).
 			argsize := s.constInt32(types.Types[types.TUINT32], int32(stksize))
 			ACArgs = append(ACArgs, ssa.Param{Type: types.Types[types.TUINT32], Offset: int32(argStart)})
-			if testLateExpansion {
-				callArgs = append(callArgs, argsize)
-			} else {
-				addr := s.constOffPtrSP(s.f.Config.Types.UInt32Ptr, argStart)
-				s.store(types.Types[types.TUINT32], addr, argsize)
-			}
+			callArgs = append(callArgs, argsize)
 			ACArgs = append(ACArgs, ssa.Param{Type: types.Types[types.TUINTPTR], Offset: int32(argStart) + int32(types.PtrSize)})
-			if testLateExpansion {
-				callArgs = append(callArgs, closure)
-			} else {
-				addr := s.constOffPtrSP(s.f.Config.Types.UintptrPtr, argStart+int64(types.PtrSize))
-				s.store(types.Types[types.TUINTPTR], addr, closure)
-			}
+			callArgs = append(callArgs, closure)
 			stksize += 2 * int64(types.PtrSize)
 			argStart += 2 * int64(types.PtrSize)
 		}
 
 		// Set receiver (for interface calls).
 		if rcvr != nil {
-			addr := s.constOffPtrSP(s.f.Config.Types.UintptrPtr, argStart)
 			ACArgs = append(ACArgs, ssa.Param{Type: types.Types[types.TUINTPTR], Offset: int32(argStart)})
-			if testLateExpansion {
-				callArgs = append(callArgs, rcvr)
-			} else {
-				s.store(types.Types[types.TUINTPTR], addr, rcvr)
-			}
+			callArgs = append(callArgs, rcvr)
 		}
 
 		// Write args.
@@ -4939,7 +4882,7 @@ func (s *state) call(n *ir.CallExpr, k callKind, returnResultAddr bool) *ssa.Val
 		}
 		for i, n := range args {
 			f := t.Params().Field(i)
-			ACArg, arg := s.putArg(n, f.Type, argStart+f.Offset, testLateExpansion)
+			ACArg, arg := s.putArg(n, f.Type, argStart+f.Offset)
 			ACArgs = append(ACArgs, ACArg)
 			callArgs = append(callArgs, arg)
 		}
@@ -4950,20 +4893,10 @@ func (s *state) call(n *ir.CallExpr, k callKind, returnResultAddr bool) *ssa.Val
 		switch {
 		case k == callDefer:
 			aux := ssa.StaticAuxCall(ir.Syms.Deferproc, ACArgs, ACResults)
-			if testLateExpansion {
-				call = s.newValue0A(ssa.OpStaticLECall, aux.LateExpansionResultType(), aux)
-				call.AddArgs(callArgs...)
-			} else {
-				call = s.newValue1A(ssa.OpStaticCall, types.TypeMem, aux, s.mem())
-			}
+			call = s.newValue0A(ssa.OpStaticLECall, aux.LateExpansionResultType(), aux)
 		case k == callGo:
 			aux := ssa.StaticAuxCall(ir.Syms.Newproc, ACArgs, ACResults)
-			if testLateExpansion {
-				call = s.newValue0A(ssa.OpStaticLECall, aux.LateExpansionResultType(), aux)
-				call.AddArgs(callArgs...)
-			} else {
-				call = s.newValue1A(ssa.OpStaticCall, types.TypeMem, aux, s.mem())
-			}
+			call = s.newValue0A(ssa.OpStaticLECall, aux.LateExpansionResultType(), aux)
 		case closure != nil:
 			// rawLoad because loading the code pointer from a
 			// closure is always safe, but IsSanitizerSafeAddr
@@ -4971,40 +4904,22 @@ func (s *state) call(n *ir.CallExpr, k callKind, returnResultAddr bool) *ssa.Val
 			// critical that we not clobber any arguments already
 			// stored onto the stack.
 			codeptr = s.rawLoad(types.Types[types.TUINTPTR], closure)
-			if testLateExpansion {
-				aux := ssa.ClosureAuxCall(ACArgs, ACResults)
-				call = s.newValue2A(ssa.OpClosureLECall, aux.LateExpansionResultType(), aux, codeptr, closure)
-				call.AddArgs(callArgs...)
-			} else {
-				call = s.newValue3A(ssa.OpClosureCall, types.TypeMem, ssa.ClosureAuxCall(ACArgs, ACResults), codeptr, closure, s.mem())
-			}
+			aux := ssa.ClosureAuxCall(ACArgs, ACResults)
+			call = s.newValue2A(ssa.OpClosureLECall, aux.LateExpansionResultType(), aux, codeptr, closure)
 		case codeptr != nil:
-			if testLateExpansion {
-				aux := ssa.InterfaceAuxCall(ACArgs, ACResults)
-				call = s.newValue1A(ssa.OpInterLECall, aux.LateExpansionResultType(), aux, codeptr)
-				call.AddArgs(callArgs...)
-			} else {
-				call = s.newValue2A(ssa.OpInterCall, types.TypeMem, ssa.InterfaceAuxCall(ACArgs, ACResults), codeptr, s.mem())
-			}
+			aux := ssa.InterfaceAuxCall(ACArgs, ACResults)
+			call = s.newValue1A(ssa.OpInterLECall, aux.LateExpansionResultType(), aux, codeptr)
 		case callee != nil:
-			if testLateExpansion {
-				aux := ssa.StaticAuxCall(callTargetLSym(callee, s.curfn.LSym), ACArgs, ACResults)
-				call = s.newValue0A(ssa.OpStaticLECall, aux.LateExpansionResultType(), aux)
-				call.AddArgs(callArgs...)
-			} else {
-				call = s.newValue1A(ssa.OpStaticCall, types.TypeMem, ssa.StaticAuxCall(callTargetLSym(callee, s.curfn.LSym), ACArgs, ACResults), s.mem())
-			}
+			aux := ssa.StaticAuxCall(callTargetLSym(callee, s.curfn.LSym), ACArgs, ACResults)
+			call = s.newValue0A(ssa.OpStaticLECall, aux.LateExpansionResultType(), aux)
 		default:
 			s.Fatalf("bad call type %v %v", n.Op(), n)
 		}
+		call.AddArgs(callArgs...)
 		call.AuxInt = stksize // Call operations carry the argsize of the callee along with them
 	}
-	if testLateExpansion {
-		s.prevCall = call
-		s.vars[memVar] = s.newValue1I(ssa.OpSelectN, types.TypeMem, int64(len(ACResults)), call)
-	} else {
-		s.vars[memVar] = call
-	}
+	s.prevCall = call
+	s.vars[memVar] = s.newValue1I(ssa.OpSelectN, types.TypeMem, int64(len(ACResults)), call)
 	// Insert OVARLIVE nodes
 	for _, name := range n.KeepAlive {
 		s.stmt(ir.NewUnaryExpr(n.Pos(), ir.OVARLIVE, name))
@@ -5033,16 +4948,10 @@ func (s *state) call(n *ir.CallExpr, k callKind, returnResultAddr bool) *ssa.Val
 	fp := res.Field(0)
 	if returnResultAddr {
 		pt := types.NewPtr(fp.Type)
-		if testLateExpansion {
-			return s.newValue1I(ssa.OpSelectNAddr, pt, 0, call)
-		}
-		return s.constOffPtrSP(pt, fp.Offset+base.Ctxt.FixedFrameSize())
+		return s.newValue1I(ssa.OpSelectNAddr, pt, 0, call)
 	}
 
-	if testLateExpansion {
-		return s.newValue1I(ssa.OpSelectN, fp.Type, 0, call)
-	}
-	return s.load(n.Type(), s.constOffPtrSP(types.NewPtr(fp.Type), fp.Offset+base.Ctxt.FixedFrameSize()))
+	return s.newValue1I(ssa.OpSelectN, fp.Type, 0, call)
 }
 
 // maybeNilCheckClosure checks if a nil check of a closure is needed in some
@@ -5458,7 +5367,6 @@ func (s *state) rtcall(fn *obj.LSym, returns bool, results []*types.Type, args .
 	s.prevCall = nil
 	// Write args to the stack
 	off := base.Ctxt.FixedFrameSize()
-	testLateExpansion := ssa.LateCallExpansionEnabledWithin(s.f)
 	var ACArgs []ssa.Param
 	var ACResults []ssa.Param
 	var callArgs []*ssa.Value
@@ -5468,12 +5376,7 @@ func (s *state) rtcall(fn *obj.LSym, returns bool, results []*types.Type, args .
 		off = types.Rnd(off, t.Alignment())
 		size := t.Size()
 		ACArgs = append(ACArgs, ssa.Param{Type: t, Offset: int32(off)})
-		if testLateExpansion {
-			callArgs = append(callArgs, arg)
-		} else {
-			ptr := s.constOffPtrSP(t.PtrTo(), off)
-			s.store(t, ptr, arg)
-		}
+		callArgs = append(callArgs, arg)
 		off += size
 	}
 	off = types.Rnd(off, int64(types.RegSize))
@@ -5489,15 +5392,10 @@ func (s *state) rtcall(fn *obj.LSym, returns bool, results []*types.Type, args .
 	// Issue call
 	var call *ssa.Value
 	aux := ssa.StaticAuxCall(fn, ACArgs, ACResults)
-	if testLateExpansion {
-		callArgs = append(callArgs, s.mem())
-		call = s.newValue0A(ssa.OpStaticLECall, aux.LateExpansionResultType(), aux)
-		call.AddArgs(callArgs...)
-		s.vars[memVar] = s.newValue1I(ssa.OpSelectN, types.TypeMem, int64(len(ACResults)), call)
-	} else {
-		call = s.newValue1A(ssa.OpStaticCall, types.TypeMem, aux, s.mem())
-		s.vars[memVar] = call
-	}
+	callArgs = append(callArgs, s.mem())
+	call = s.newValue0A(ssa.OpStaticLECall, aux.LateExpansionResultType(), aux)
+	call.AddArgs(callArgs...)
+	s.vars[memVar] = s.newValue1I(ssa.OpSelectN, types.TypeMem, int64(len(ACResults)), call)
 
 	if !returns {
 		// Finish block
@@ -5513,24 +5411,15 @@ func (s *state) rtcall(fn *obj.LSym, returns bool, results []*types.Type, args .
 
 	// Load results
 	res := make([]*ssa.Value, len(results))
-	if testLateExpansion {
-		for i, t := range results {
-			off = types.Rnd(off, t.Alignment())
-			if TypeOK(t) {
-				res[i] = s.newValue1I(ssa.OpSelectN, t, int64(i), call)
-			} else {
-				addr := s.newValue1I(ssa.OpSelectNAddr, types.NewPtr(t), int64(i), call)
-				res[i] = s.rawLoad(t, addr)
-			}
-			off += t.Size()
-		}
-	} else {
-		for i, t := range results {
-			off = types.Rnd(off, t.Alignment())
-			ptr := s.constOffPtrSP(types.NewPtr(t), off)
-			res[i] = s.load(t, ptr)
-			off += t.Size()
+	for i, t := range results {
+		off = types.Rnd(off, t.Alignment())
+		if TypeOK(t) {
+			res[i] = s.newValue1I(ssa.OpSelectN, t, int64(i), call)
+		} else {
+			addr := s.newValue1I(ssa.OpSelectNAddr, types.NewPtr(t), int64(i), call)
+			res[i] = s.rawLoad(t, addr)
 		}
+		off += t.Size()
 	}
 	off = types.Rnd(off, int64(types.PtrSize))
 
@@ -5650,19 +5539,13 @@ func (s *state) storeTypePtrs(t *types.Type, left, right *ssa.Value) {
 	}
 }
 
-// putArg evaluates n for the purpose of passing it as an argument to a function and returns the corresponding Param for the call.
-// If forLateExpandedCall is true, it returns the argument value to pass to the call operation.
-// If forLateExpandedCall is false, then the value is stored at the specified stack offset, and the returned value is nil.
-func (s *state) putArg(n ir.Node, t *types.Type, off int64, forLateExpandedCall bool) (ssa.Param, *ssa.Value) {
+// putArg evaluates n for the purpose of passing it as an argument to a function and returns the corresponding Param and value for the call.
+func (s *state) putArg(n ir.Node, t *types.Type, off int64) (ssa.Param, *ssa.Value) {
 	var a *ssa.Value
-	if forLateExpandedCall {
-		if !TypeOK(t) {
-			a = s.newValue2(ssa.OpDereference, t, s.addr(n), s.mem())
-		} else {
-			a = s.expr(n)
-		}
+	if !TypeOK(t) {
+		a = s.newValue2(ssa.OpDereference, t, s.addr(n), s.mem())
 	} else {
-		s.storeArgWithBase(n, t, s.sp, off)
+		a = s.expr(n)
 	}
 	return ssa.Param{Type: t, Offset: int32(off)}, a
 }
diff --git a/src/cmd/go/alldocs.go b/src/cmd/go/alldocs.go
index 49d390297c..da06e831ae 100644
--- a/src/cmd/go/alldocs.go
+++ b/src/cmd/go/alldocs.go
@@ -111,7 +111,7 @@
 // 	-p n
 // 		the number of programs, such as build commands or
 // 		test binaries, that can be run in parallel.
-// 		The default is the number of CPUs available.
+// 		The default is GOMAXPROCS, normally the number of CPUs available.
 // 	-race
 // 		enable data race detection.
 // 		Supported only on linux/amd64, freebsd/amd64, darwin/amd64, windows/amd64,
diff --git a/src/cmd/go/internal/cfg/cfg.go b/src/cmd/go/internal/cfg/cfg.go
index c48904eacc..322247962f 100644
--- a/src/cmd/go/internal/cfg/cfg.go
+++ b/src/cmd/go/internal/cfg/cfg.go
@@ -28,18 +28,18 @@ var (
 	BuildA                 bool   // -a flag
 	BuildBuildmode         string // -buildmode flag
 	BuildContext           = defaultContext()
-	BuildMod               string             // -mod flag
-	BuildModExplicit       bool               // whether -mod was set explicitly
-	BuildModReason         string             // reason -mod was set, if set by default
-	BuildI                 bool               // -i flag
-	BuildLinkshared        bool               // -linkshared flag
-	BuildMSan              bool               // -msan flag
-	BuildN                 bool               // -n flag
-	BuildO                 string             // -o flag
-	BuildP                 = runtime.NumCPU() // -p flag
-	BuildPkgdir            string             // -pkgdir flag
-	BuildRace              bool               // -race flag
-	BuildToolexec          []string           // -toolexec flag
+	BuildMod               string                  // -mod flag
+	BuildModExplicit       bool                    // whether -mod was set explicitly
+	BuildModReason         string                  // reason -mod was set, if set by default
+	BuildI                 bool                    // -i flag
+	BuildLinkshared        bool                    // -linkshared flag
+	BuildMSan              bool                    // -msan flag
+	BuildN                 bool                    // -n flag
+	BuildO                 string                  // -o flag
+	BuildP                 = runtime.GOMAXPROCS(0) // -p flag
+	BuildPkgdir            string                  // -pkgdir flag
+	BuildRace              bool                    // -race flag
+	BuildToolexec          []string                // -toolexec flag
 	BuildToolchainName     string
 	BuildToolchainCompiler func() string
 	BuildToolchainLinker   func() string
diff --git a/src/cmd/go/internal/work/build.go b/src/cmd/go/internal/work/build.go
index 780d639c5d..0e7af6d33f 100644
--- a/src/cmd/go/internal/work/build.go
+++ b/src/cmd/go/internal/work/build.go
@@ -71,7 +71,7 @@ and test commands:
 	-p n
 		the number of programs, such as build commands or
 		test binaries, that can be run in parallel.
-		The default is the number of CPUs available.
+		The default is GOMAXPROCS, normally the number of CPUs available.
 	-race
 		enable data race detection.
 		Supported only on linux/amd64, freebsd/amd64, darwin/amd64, windows/amd64,
diff --git a/src/cmd/go/internal/work/gc.go b/src/cmd/go/internal/work/gc.go
index 3205fcbffc..2087855b3c 100644
--- a/src/cmd/go/internal/work/gc.go
+++ b/src/cmd/go/internal/work/gc.go
@@ -239,16 +239,19 @@ CheckFlags:
 	//   - it has no successor packages to compile (usually package main)
 	//   - all paths through the build graph pass through it
 	//   - critical path scheduling says it is high priority
-	// and in such a case, set c to runtime.NumCPU.
+	// and in such a case, set c to runtime.GOMAXPROCS(0).
+	// By default this is the same as runtime.NumCPU.
 	// We do this now when p==1.
+	// To limit parallelism, set GOMAXPROCS below numCPU; this may be useful
+	// on a low-memory builder, or if a deterministic build order is required.
+	c := runtime.GOMAXPROCS(0)
 	if cfg.BuildP == 1 {
-		// No process parallelism. Max out c.
-		return runtime.NumCPU()
+		// No process parallelism, do not cap compiler parallelism.
+		return c
 	}
-	// Some process parallelism. Set c to min(4, numcpu).
-	c := 4
-	if ncpu := runtime.NumCPU(); ncpu < c {
-		c = ncpu
+	// Some process parallelism. Set c to min(4, maxprocs).
+	if c > 4 {
+		c = 4
 	}
 	return c
 }
diff --git a/src/cmd/go/testdata/script/build_trimpath.txt b/src/cmd/go/testdata/script/build_trimpath.txt
index e1ea0a48b2..2c3bee8fdc 100644
--- a/src/cmd/go/testdata/script/build_trimpath.txt
+++ b/src/cmd/go/testdata/script/build_trimpath.txt
@@ -121,6 +121,7 @@ package main
 import (
 	"bytes"
 	"fmt"
+	"io/ioutil"
 	"log"
 	"os"
 	"os/exec"
@@ -130,7 +131,7 @@ import (
 
 func main() {
 	exe := os.Args[1]
-	data, err := os.ReadFile(exe)
+	data, err := ioutil.ReadFile(exe)
 	if err != nil {
 		log.Fatal(err)
 	}
diff --git a/src/cmd/go/testdata/script/link_syso_issue33139.txt b/src/cmd/go/testdata/script/link_syso_issue33139.txt
index 26034c9626..8a8cb4aa8c 100644
--- a/src/cmd/go/testdata/script/link_syso_issue33139.txt
+++ b/src/cmd/go/testdata/script/link_syso_issue33139.txt
@@ -2,7 +2,7 @@
 # embedded in a package, that is referenced by a Go assembly function.
 # See issue 33139.
 [!gc] skip
-[!exec:cc] skip
+[!cgo] skip
 
 # External linking is not supported on linux/ppc64.
 # See: https://github.com/golang/go/issues/8912
diff --git a/src/cmd/link/elf_test.go b/src/cmd/link/elf_test.go
index 334f050e88..20754d09f5 100644
--- a/src/cmd/link/elf_test.go
+++ b/src/cmd/link/elf_test.go
@@ -226,6 +226,12 @@ func main() {
 
 func TestPIESize(t *testing.T) {
 	testenv.MustHaveGoBuild(t)
+
+	// We don't want to test -linkmode=external if cgo is not supported.
+	// On some systems -buildmode=pie implies -linkmode=external, so just
+	// always skip the test if cgo is not supported.
+	testenv.MustHaveCGO(t)
+
 	if !sys.BuildModeSupported(runtime.Compiler, "pie", runtime.GOOS, runtime.GOARCH) {
 		t.Skip("-buildmode=pie not supported")
 	}
diff --git a/src/cmd/link/internal/ld/deadcode.go b/src/cmd/link/internal/ld/deadcode.go
index d8813fa936..245076a83a 100644
--- a/src/cmd/link/internal/ld/deadcode.go
+++ b/src/cmd/link/internal/ld/deadcode.go
@@ -165,13 +165,17 @@ func (d *deadcodePass) flood() {
 				// R_USEIFACEMETHOD is a marker relocation that marks an interface
 				// method as used.
 				rs := r.Sym()
-				if d.ldr.SymType(rs) != sym.SDYNIMPORT { // don't decode DYNIMPORT symbol (we'll mark all exported methods anyway)
-					m := d.decodeIfaceMethod(d.ldr, d.ctxt.Arch, rs, r.Add())
-					if d.ctxt.Debugvlog > 1 {
-						d.ctxt.Logf("reached iface method: %v\n", m)
-					}
-					d.ifaceMethod[m] = true
+				if d.ctxt.linkShared && (d.ldr.SymType(rs) == sym.SDYNIMPORT || d.ldr.SymType(rs) == sym.Sxxx) {
+					// Don't decode symbol from shared library (we'll mark all exported methods anyway).
+					// We check for both SDYNIMPORT and Sxxx because name-mangled symbols haven't
+					// been resolved at this point.
+					continue
+				}
+				m := d.decodeIfaceMethod(d.ldr, d.ctxt.Arch, rs, r.Add())
+				if d.ctxt.Debugvlog > 1 {
+					d.ctxt.Logf("reached iface method: %v\n", m)
 				}
+				d.ifaceMethod[m] = true
 				continue
 			}
 			rs := r.Sym()
diff --git a/src/cmd/link/internal/ld/decodesym.go b/src/cmd/link/internal/ld/decodesym.go
index c6e2d8ca7f..fc179fc6e4 100644
--- a/src/cmd/link/internal/ld/decodesym.go
+++ b/src/cmd/link/internal/ld/decodesym.go
@@ -279,7 +279,7 @@ func findShlibSection(ctxt *Link, path string, addr uint64) *elf.Section {
 	for _, shlib := range ctxt.Shlibs {
 		if shlib.Path == path {
 			for _, sect := range shlib.File.Sections[1:] { // skip the NULL section
-				if sect.Addr <= addr && addr <= sect.Addr+sect.Size {
+				if sect.Addr <= addr && addr < sect.Addr+sect.Size {
 					return sect
 				}
 			}