cmd/compile: register abi, morestack work and mole whacking

Morestack works for non-pointer register parameters

Within a function body, pointer-typed parameters are correctly
tracked.

Results still not hooked up.

For #40724.

Change-Id: Icaee0b51d0da54af983662d945d939b756088746
Reviewed-on: https://go-review.googlesource.com/c/go/+/294410
Trust: David Chase <drchase@google.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>

src/cmd/compile/internal/abi/abiutils.go

@@ -5,6 +5,7 @@
 package abi
 
 import (
+	"cmd/compile/internal/ir"
 	"cmd/compile/internal/types"
 	"cmd/internal/src"
 	"fmt"
@@ -337,6 +338,9 @@ func (config *ABIConfig) updateOffset(result *ABIParamResultInfo, f *types.Field
 		if fOffset == types.BOGUS_FUNARG_OFFSET {
 			// Set the Offset the first time. After that, we may recompute it, but it should never change.
 			f.Offset = off
+			if f.Nname != nil {
+				f.Nname.(*ir.Name).SetFrameOffset(off)
+			}
 		} else if fOffset != off {
 			panic(fmt.Errorf("Offset changed from %d to %d", fOffset, off))
 		}

src/cmd/compile/internal/amd64/ssa.go

@@ -980,6 +980,17 @@ func ssaGenValue(s *ssagen.State, v *ssa.Value) {
 		ssagen.AddAux(&p.From, v)
 		p.To.Type = obj.TYPE_REG
 		p.To.Reg = v.Reg()
+	case ssa.OpArgIntReg, ssa.OpArgFloatReg:
+		// The assembler needs to wrap the entry safepoint/stack growth code with spill/unspill
+		// The loop only runs once.
+		for _, ap := range v.Block.Func.RegArgs {
+			// Pass the spill/unspill information along to the assembler, offset by size of return PC pushed on stack.
+			addr := ssagen.SpillSlotAddr(ap.Mem(), x86.REG_SP, v.Block.Func.Config.PtrSize)
+			s.FuncInfo().AddSpill(
+				obj.RegSpill{Reg: ap.Reg(), Addr: addr, Unspill: loadByType(ap.Type()), Spill: storeByType(ap.Type())})
+		}
+		v.Block.Func.RegArgs = nil
+		ssagen.CheckArgReg(v)
 	case ssa.OpAMD64LoweredGetClosurePtr:
 		// Closure pointer is DX.
 		ssagen.CheckLoweredGetClosurePtr(v)

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

@@ -901,10 +901,10 @@ func (state *debugState) buildLocationLists(blockLocs []*BlockDebug) {
 
 			if opcodeTable[v.Op].zeroWidth {
 				if changed {
-					if v.Op == OpArg || v.Op == OpPhi || v.Op.isLoweredGetClosurePtr() {
+					if hasAnyArgOp(v) || v.Op == OpPhi || v.Op.isLoweredGetClosurePtr() {
 						// These ranges begin at true beginning of block, not after first instruction
 						if zeroWidthPending {
-							b.Func.Fatalf("Unexpected op mixed with OpArg/OpPhi/OpLoweredGetClosurePtr at beginning of block %s in %s\n%s", b, b.Func.Name, b.Func)
+							panic(fmt.Errorf("Unexpected op '%s' mixed with OpArg/OpPhi/OpLoweredGetClosurePtr at beginning of block %s in %s\n%s", v.LongString(), b, b.Func.Name, b.Func))
 						}
 						apcChangedSize = len(state.changedVars.contents())
 						continue

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

@@ -112,7 +112,7 @@ func (s *stackAllocState) init(f *Func, spillLive [][]ID) {
 		for _, v := range b.Values {
 			s.values[v.ID].typ = v.Type
 			s.values[v.ID].needSlot = !v.Type.IsMemory() && !v.Type.IsVoid() && !v.Type.IsFlags() && f.getHome(v.ID) == nil && !v.rematerializeable() && !v.OnWasmStack
-			s.values[v.ID].isArg = v.Op == OpArg
+			s.values[v.ID].isArg = hasAnyArgOp(v)
 			if f.pass.debug > stackDebug && s.values[v.ID].needSlot {
 				fmt.Printf("%s needs a stack slot\n", v)
 			}
@@ -151,28 +151,29 @@ func (s *stackAllocState) stackalloc() {
 
 	// Allocate args to their assigned locations.
 	for _, v := range f.Entry.Values {
-		if v.Op != OpArg { // && v.Op != OpArgFReg && v.Op != OpArgIReg  {
+		if !hasAnyArgOp(v) {
 			continue
 		}
 		if v.Aux == nil {
 			f.Fatalf("%s has nil Aux\n", v.LongString())
 		}
-		var loc LocalSlot
-		var name *ir.Name
-		var offset int64
 		if v.Op == OpArg {
-			name = v.Aux.(*ir.Name)
-			offset = v.AuxInt
-		} else {
-			nameOff := v.Aux.(*AuxNameOffset)
-			name = nameOff.Name
-			offset = nameOff.Offset
+			loc := LocalSlot{N: v.Aux.(*ir.Name), Type: v.Type, Off: v.AuxInt}
+			if f.pass.debug > stackDebug {
+				fmt.Printf("stackalloc OpArg %s to %s\n", v, loc)
+			}
+			f.setHome(v, loc)
+			continue
 		}
-		loc = LocalSlot{N: name, Type: v.Type, Off: offset}
+
+		nameOff := v.Aux.(*AuxNameOffset)
+		loc := LocalSlot{N: nameOff.Name, Type: v.Type, Off: nameOff.Offset}
 		if f.pass.debug > stackDebug {
-			fmt.Printf("stackalloc %s to %s\n", v, loc)
+			fmt.Printf("stackalloc Op%s %s to %s\n", v.Op, v, loc)
 		}
-		f.setHome(v, loc)
+		// register args already allocated to registers, but need to know the stack allocation for later
+		reg := f.getHome(v.ID).(*Register)
+		f.RegArgs = append(f.RegArgs, ArgPair{reg: reg, mem: loc})
 	}
 
 	// For each type, we keep track of all the stack slots we
@@ -209,7 +210,7 @@ func (s *stackAllocState) stackalloc() {
 				s.nNotNeed++
 				continue
 			}
-			if v.Op == OpArg {
+			if hasAnyArgOp(v) {
 				s.nArgSlot++
 				continue // already picked
 			}
@@ -396,7 +397,7 @@ func (s *stackAllocState) buildInterferenceGraph() {
 				for _, id := range live.contents() {
 					// Note: args can have different types and still interfere
 					// (with each other or with other values). See issue 23522.
-					if s.values[v.ID].typ.Compare(s.values[id].typ) == types.CMPeq || v.Op == OpArg || s.values[id].isArg {
+					if s.values[v.ID].typ.Compare(s.values[id].typ) == types.CMPeq || hasAnyArgOp(v) || s.values[id].isArg {
 						s.interfere[v.ID] = append(s.interfere[v.ID], id)
 						s.interfere[id] = append(s.interfere[id], v.ID)
 					}
@@ -407,13 +408,15 @@ func (s *stackAllocState) buildInterferenceGraph() {
 					live.add(a.ID)
 				}
 			}
-			if v.Op == OpArg && s.values[v.ID].needSlot {
+			if hasAnyArgOp(v) && s.values[v.ID].needSlot {
 				// OpArg is an input argument which is pre-spilled.
 				// We add back v.ID here because we want this value
 				// to appear live even before this point. Being live
 				// all the way to the start of the entry block prevents other
 				// values from being allocated to the same slot and clobbering
 				// the input value before we have a chance to load it.
+
+				// TODO(register args) this is apparently not wrong for register args -- is it necessary?
 				live.add(v.ID)
 			}
 		}
@@ -430,3 +433,7 @@ func (s *stackAllocState) buildInterferenceGraph() {
 		}
 	}
 }
+
+func hasAnyArgOp(v *Value) bool {
+	return v.Op == OpArg || v.Op == OpArgIntReg || v.Op == OpArgFloatReg
+}

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

@@ -221,7 +221,7 @@ func AbiForFunc(fn *ir.Func) *abi.ABIConfig {
 // Passing a nil function returns ABIInternal.
 func abiForFunc(fn *ir.Func, abi0, abi1 *abi.ABIConfig) *abi.ABIConfig {
 	a := abi1
-	if true || objabi.Regabi_enabled == 0 {
+	if !regabiEnabledForAllCompilation() {
 		a = abi0
 	}
 	if fn != nil && fn.Pragma&ir.RegisterParams != 0 { // TODO(register args) remove after register abi is working
@@ -235,6 +235,11 @@ func abiForFunc(fn *ir.Func, abi0, abi1 *abi.ABIConfig) *abi.ABIConfig {
 	return a
 }
 
+func regabiEnabledForAllCompilation() bool {
+	// TODO compiler does not yet change behavior for GOEXPERIMENT=regabi
+	return false && objabi.Regabi_enabled != 0
+}
+
 // getParam returns the Field of ith param of node n (which is a
 // function/method/interface call), where the receiver of a method call is
 // considered as the 0th parameter. This does not include the receiver of an
@@ -6404,6 +6409,10 @@ type State struct {
 	OnWasmStackSkipped int
 }
 
+func (s *State) FuncInfo() *obj.FuncInfo {
+	return s.pp.CurFunc.LSym.Func()
+}
+
 // Prog appends a new Prog.
 func (s *State) Prog(as obj.As) *obj.Prog {
 	p := s.pp.Prog(as)
@@ -6561,11 +6570,9 @@ func genssa(f *ssa.Func, pp *objw.Progs) {
 				// memory arg needs no code
 			case ssa.OpArg:
 				// input args need no code
-			case ssa.OpArgIntReg, ssa.OpArgFloatReg:
-				CheckArgReg(v)
 			case ssa.OpSP, ssa.OpSB:
 				// nothing to do
-			case ssa.OpSelect0, ssa.OpSelect1, ssa.OpSelectN:
+			case ssa.OpSelect0, ssa.OpSelect1, ssa.OpSelectN, ssa.OpMakeResult:
 				// nothing to do
 			case ssa.OpGetG:
 				// nothing to do when there's a g register,
@@ -7470,6 +7477,39 @@ func deferstruct(stksize int64) *types.Type {
 	return s
 }
 
+// SlotAddr uses LocalSlot information to initialize an obj.Addr
+// The resulting addr is used in a non-standard context -- in the prologue
+// of a function, before the frame has been constructed, so the standard
+// addressing for the parameters will be wrong.
+func SpillSlotAddr(slot *ssa.LocalSlot, baseReg int16, extraOffset int64) obj.Addr {
+	n, off := slot.N, slot.Off
+	if n.Class != ir.PPARAM && n.Class != ir.PPARAMOUT {
+		panic("Only expected to see param and returns here")
+	}
+	return obj.Addr{
+		Name:   obj.NAME_NONE,
+		Type:   obj.TYPE_MEM,
+		Reg:    baseReg,
+		Offset: off + extraOffset + n.FrameOffset(),
+	}
+}
+
+// AddrForParamSlot fills in an Addr appropriately for a Spill,
+// Restore, or VARLIVE.
+func AddrForParamSlot(slot *ssa.LocalSlot, addr *obj.Addr) {
+	// TODO replace this boilerplate in a couple of places.
+	n, off := slot.N, slot.Off
+	addr.Type = obj.TYPE_MEM
+	addr.Sym = n.Linksym()
+	addr.Offset = off
+	if n.Class == ir.PPARAM || n.Class == ir.PPARAMOUT {
+		addr.Name = obj.NAME_PARAM
+		addr.Offset += n.FrameOffset()
+	} else {
+		addr.Name = obj.NAME_AUTO
+	}
+}
+
 var (
 	BoundsCheckFunc [ssa.BoundsKindCount]*obj.LSym
 	ExtendCheckFunc [ssa.BoundsKindCount]*obj.LSym

src/cmd/internal/obj/link.go

@@ -473,6 +473,7 @@ type FuncInfo struct {
 	Autot    map[*LSym]struct{}
 	Pcln     Pcln
 	InlMarks []InlMark
+	spills   []RegSpill
 
 	dwarfInfoSym       *LSym
 	dwarfLocSym        *LSym
@@ -552,6 +553,11 @@ func (fi *FuncInfo) AddInlMark(p *Prog, id int32) {
 	fi.InlMarks = append(fi.InlMarks, InlMark{p: p, id: id})
 }
 
+// AddSpill appends a spill record to the list for FuncInfo fi
+func (fi *FuncInfo) AddSpill(s RegSpill) {
+	fi.spills = append(fi.spills, s)
+}
+
 // Record the type symbol for an auto variable so that the linker
 // an emit DWARF type information for the type.
 func (fi *FuncInfo) RecordAutoType(gotype *LSym) {
@@ -803,12 +809,12 @@ type Auto struct {
 	Gotype  *LSym
 }
 
-// RegArg provides spill/fill information for a register-resident argument
+// RegSpill provides spill/fill information for a register-resident argument
 // to a function.  These need spilling/filling in the safepoint/stackgrowth case.
 // At the time of fill/spill, the offset must be adjusted by the architecture-dependent
 // adjustment to hardware SP that occurs in a call instruction.  E.g., for AMD64,
 // at Offset+8 because the return address was pushed.
-type RegArg struct {
+type RegSpill struct {
 	Addr           Addr
 	Reg            int16
 	Spill, Unspill As
@@ -844,7 +850,6 @@ type Link struct {
 	DebugInfo          func(fn *LSym, info *LSym, curfn interface{}) ([]dwarf.Scope, dwarf.InlCalls) // if non-nil, curfn is a *gc.Node
 	GenAbstractFunc    func(fn *LSym)
 	Errors             int
-	RegArgs            []RegArg
 
 	InParallel    bool // parallel backend phase in effect
 	UseBASEntries bool // use Base Address Selection Entries in location lists and PC ranges
@@ -893,9 +898,11 @@ func (ctxt *Link) Logf(format string, args ...interface{}) {
 	ctxt.Bso.Flush()
 }
 
-func (ctxt *Link) SpillRegisterArgs(last *Prog, pa ProgAlloc) *Prog {
+// SpillRegisterArgs emits the code to spill register args into whatever
+// locations the spill records specify.
+func (fi *FuncInfo) SpillRegisterArgs(last *Prog, pa ProgAlloc) *Prog {
 	// Spill register args.
-	for _, ra := range ctxt.RegArgs {
+	for _, ra := range fi.spills {
 		spill := Appendp(last, pa)
 		spill.As = ra.Spill
 		spill.From.Type = TYPE_REG
@@ -906,9 +913,11 @@ func (ctxt *Link) SpillRegisterArgs(last *Prog, pa ProgAlloc) *Prog {
 	return last
 }
 
-func (ctxt *Link) UnspillRegisterArgs(last *Prog, pa ProgAlloc) *Prog {
+// UnspillRegisterArgs emits the code to restore register args from whatever
+// locations the spill records specify.
+func (fi *FuncInfo) UnspillRegisterArgs(last *Prog, pa ProgAlloc) *Prog {
 	// Unspill any spilled register args
-	for _, ra := range ctxt.RegArgs {
+	for _, ra := range fi.spills {
 		unspill := Appendp(last, pa)
 		unspill.As = ra.Unspill
 		unspill.From = ra.Addr

src/cmd/internal/obj/x86/obj6.go

@@ -135,7 +135,7 @@ func progedit(ctxt *obj.Link, p *obj.Prog, newprog obj.ProgAlloc) {
 			p.To.Index = REG_NONE
 		}
 	} else {
-		// load_g_cx, below, always inserts the 1-instruction sequence. Rewrite it
+		// load_g, below, always inserts the 1-instruction sequence. Rewrite it
 		// as the 2-instruction sequence if necessary.
 		//	MOVQ 0(TLS), BX
 		// becomes
@@ -644,8 +644,12 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) {
 			regg = REGG // use the g register directly in ABIInternal
 		} else {
 			p = obj.Appendp(p, newprog)
-			p = load_g_cx(ctxt, p, newprog) // load g into CX
 			regg = REG_CX
+			if ctxt.Arch.Family == sys.AMD64 {
+				// Using this register means that stacksplit works w/ //go:registerparams even when objabi.Regabi_enabled == 0
+				regg = REGG // == REG_R14
+			}
+			p = load_g(ctxt, p, newprog, regg) // load g into regg
 		}
 	}
 
@@ -963,7 +967,7 @@ func indir_cx(ctxt *obj.Link, a *obj.Addr) {
 // Overwriting p is unusual but it lets use this in both the
 // prologue (caller must call appendp first) and in the epilogue.
 // Returns last new instruction.
-func load_g_cx(ctxt *obj.Link, p *obj.Prog, newprog obj.ProgAlloc) *obj.Prog {
+func load_g(ctxt *obj.Link, p *obj.Prog, newprog obj.ProgAlloc, rg int16) *obj.Prog {
 	p.As = AMOVQ
 	if ctxt.Arch.PtrSize == 4 {
 		p.As = AMOVL
@@ -972,7 +976,7 @@ func load_g_cx(ctxt *obj.Link, p *obj.Prog, newprog obj.ProgAlloc) *obj.Prog {
 	p.From.Reg = REG_TLS
 	p.From.Offset = 0
 	p.To.Type = obj.TYPE_REG
-	p.To.Reg = REG_CX
+	p.To.Reg = rg
 
 	next := p.Link
 	progedit(ctxt, p, newprog)
@@ -1027,9 +1031,14 @@ func stacksplit(ctxt *obj.Link, cursym *obj.LSym, p *obj.Prog, newprog obj.ProgA
 		// unnecessarily. See issue #35470.
 		p = ctxt.StartUnsafePoint(p, newprog)
 	} else if framesize <= objabi.StackBig {
+		tmp := int16(REG_AX) // use AX for 32-bit
+		if ctxt.Arch.Family == sys.AMD64 {
+			// for 64-bit, stay away from register ABI parameter registers, even w/o GOEXPERIMENT=regabi
+			tmp = int16(REG_R13)
+		}
 		// large stack: SP-framesize <= stackguard-StackSmall
-		//	LEAQ -xxx(SP), AX
-		//	CMPQ AX, stackguard
+		//	LEAQ -xxx(SP), tmp
+		//	CMPQ tmp, stackguard
 		p = obj.Appendp(p, newprog)
 
 		p.As = lea
@@ -1037,12 +1046,12 @@ func stacksplit(ctxt *obj.Link, cursym *obj.LSym, p *obj.Prog, newprog obj.ProgA
 		p.From.Reg = REG_SP
 		p.From.Offset = -(int64(framesize) - objabi.StackSmall)
 		p.To.Type = obj.TYPE_REG
-		p.To.Reg = REG_AX
+		p.To.Reg = tmp
 
 		p = obj.Appendp(p, newprog)
 		p.As = cmp
 		p.From.Type = obj.TYPE_REG
-		p.From.Reg = REG_AX
+		p.From.Reg = tmp
 		p.To.Type = obj.TYPE_MEM
 		p.To.Reg = rg
 		p.To.Offset = 2 * int64(ctxt.Arch.PtrSize) // G.stackguard0
@@ -1052,6 +1061,12 @@ func stacksplit(ctxt *obj.Link, cursym *obj.LSym, p *obj.Prog, newprog obj.ProgA
 
 		p = ctxt.StartUnsafePoint(p, newprog) // see the comment above
 	} else {
+		tmp1 := int16(REG_SI)
+		tmp2 := int16(REG_AX)
+		if ctxt.Arch.Family == sys.AMD64 {
+			tmp1 = int16(REG_R13) // register ABI uses REG_SI and REG_AX for parameters.
+			tmp2 = int16(REG_R12)
+		}
 		// Such a large stack we need to protect against wraparound.
 		// If SP is close to zero:
 		//	SP-stackguard+StackGuard <= framesize + (StackGuard-StackSmall)
@@ -1060,12 +1075,12 @@ func stacksplit(ctxt *obj.Link, cursym *obj.LSym, p *obj.Prog, newprog obj.ProgA
 		//
 		// Preemption sets stackguard to StackPreempt, a very large value.
 		// That breaks the math above, so we have to check for that explicitly.
-		//	MOVQ	stackguard, SI
+		//	MOVQ	stackguard, tmp1
 		//	CMPQ	SI, $StackPreempt
 		//	JEQ	label-of-call-to-morestack
-		//	LEAQ	StackGuard(SP), AX
-		//	SUBQ	SI, AX
-		//	CMPQ	AX, $(framesize+(StackGuard-StackSmall))
+		//	LEAQ	StackGuard(SP), tmp2
+		//	SUBQ	tmp1, tmp2
+		//	CMPQ	tmp2, $(framesize+(StackGuard-StackSmall))
 
 		p = obj.Appendp(p, newprog)
 
@@ -1077,14 +1092,14 @@ func stacksplit(ctxt *obj.Link, cursym *obj.LSym, p *obj.Prog, newprog obj.ProgA
 			p.From.Offset = 3 * int64(ctxt.Arch.PtrSize) // G.stackguard1
 		}
 		p.To.Type = obj.TYPE_REG
-		p.To.Reg = REG_SI
+		p.To.Reg = tmp1
 
 		p = ctxt.StartUnsafePoint(p, newprog) // see the comment above
 
 		p = obj.Appendp(p, newprog)
 		p.As = cmp
 		p.From.Type = obj.TYPE_REG
-		p.From.Reg = REG_SI
+		p.From.Reg = tmp1
 		p.To.Type = obj.TYPE_CONST
 		p.To.Offset = objabi.StackPreempt
 		if ctxt.Arch.Family == sys.I386 {
@@ -1102,19 +1117,19 @@ func stacksplit(ctxt *obj.Link, cursym *obj.LSym, p *obj.Prog, newprog obj.ProgA
 		p.From.Reg = REG_SP
 		p.From.Offset = int64(objabi.StackGuard)
 		p.To.Type = obj.TYPE_REG
-		p.To.Reg = REG_AX
+		p.To.Reg = tmp2
 
 		p = obj.Appendp(p, newprog)
 		p.As = sub
 		p.From.Type = obj.TYPE_REG
-		p.From.Reg = REG_SI
+		p.From.Reg = tmp1
 		p.To.Type = obj.TYPE_REG
-		p.To.Reg = REG_AX
+		p.To.Reg = tmp2
 
 		p = obj.Appendp(p, newprog)
 		p.As = cmp
 		p.From.Type = obj.TYPE_REG
-		p.From.Reg = REG_AX
+		p.From.Reg = tmp2
 		p.To.Type = obj.TYPE_CONST
 		p.To.Offset = int64(framesize) + (int64(objabi.StackGuard) - objabi.StackSmall)
 	}
@@ -1139,7 +1154,7 @@ func stacksplit(ctxt *obj.Link, cursym *obj.LSym, p *obj.Prog, newprog obj.ProgA
 
 	pcdata := ctxt.EmitEntryStackMap(cursym, spfix, newprog)
 	spill := ctxt.StartUnsafePoint(pcdata, newprog)
-	pcdata = ctxt.SpillRegisterArgs(spill, newprog)
+	pcdata = cursym.Func().SpillRegisterArgs(spill, newprog)
 
 	call := obj.Appendp(pcdata, newprog)
 	call.Pos = cursym.Func().Text.Pos
@@ -1164,7 +1179,7 @@ func stacksplit(ctxt *obj.Link, cursym *obj.LSym, p *obj.Prog, newprog obj.ProgA
 		progedit(ctxt, callend.Link, newprog)
 	}
 
-	pcdata = ctxt.UnspillRegisterArgs(callend, newprog)
+	pcdata = cursym.Func().UnspillRegisterArgs(callend, newprog)
 	pcdata = ctxt.EndUnsafePoint(pcdata, newprog, -1)
 
 	jmp := obj.Appendp(pcdata, newprog)

test/abi/f_ret_z_not.go

@@ -16,18 +16,18 @@ type NZ struct {
 }
 
 //go:noinline
-func f(x,y int) (Z,NZ,Z) {
+func f(x, y int) (Z, NZ, Z) {
 	var z Z
-	return z,NZ{x,y},z
+	return z, NZ{x, y}, z
 }
 
 //go:noinline
-func g() (Z,NZ,Z) {
-	a,b,c := f(3,4)
-	return c,b,a
+func g() (Z, NZ, Z) {
+	a, b, c := f(3, 4)
+	return c, b, a
 }
 
 func main() {
-	_,b,_ := g()
-	fmt.Println(b.x+b.y)
+	_, b, _ := g()
+	fmt.Println(b.x + b.y)
 }

test/abi/many_int_input.go

@@ -0,0 +1,30 @@
+// run
+
+//go:build !wasm
+// +build !wasm
+
+// Copyright 2021 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package main
+
+import (
+	"fmt"
+)
+
+//go:registerparams
+//go:noinline
+func F(a, b, c, d, e, f, g, h, i, j, k, l, m, n, o, p, q, r, s, t, u, v, w, x, y, z int64) {
+	G(z, y, x, w, v, u, t, s, r, q, p, o, n, m, l, k, j, i, h, g, f, e, d, c, b, a)
+}
+
+//go:registerparams
+//go:noinline
+func G(a, b, c, d, e, f, g, h, i, j, k, l, m, n, o, p, q, r, s, t, u, v, w, x, y, z int64) {
+	fmt.Println(a, b, c, d, e, f, g, h, i, j, k, l, m, n, o, p, q, r, s, t, u, v, w, x, y, z)
+}
+
+func main() {
+	F(1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26)
+}

test/abi/many_int_input.out

@@ -0,0 +1 @@
+26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1

test/abi/regabipragma.go

@@ -1,5 +1,6 @@
 // skip
 // runindir -gcflags=-c=1
+//go:build !windows
 // +build !windows
 
 // Copyright 2021 The Go Authors. All rights reserved.