cmd/compile: implement simple register results

at least for ints and strings

includes simple test

For #40724.

Change-Id: Ib8484e5b957b08f961574a67cfd93d3d26551558
Reviewed-on: https://go-review.googlesource.com/c/go/+/295309
Trust: David Chase <[email protected]>
Run-TryBot: David Chase <[email protected]>
Reviewed-by: Cherry Zhang <[email protected]>

Author: dr2chase

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

@@ -101,6 +101,70 @@ func (a *ABIParamAssignment) Offset() int32 {
 	return a.offset
 }
 
+// RegisterTypes returns a slice of the types of the registers
+// corresponding to a slice of parameters.  The returned slice
+// has capacity for one more, likely a memory type.
+func RegisterTypes(apa []ABIParamAssignment) []*types.Type {
+	rcount := 0
+	for _, pa := range apa {
+		rcount += len(pa.Registers)
+	}
+	if rcount == 0 {
+		// Note that this catches top-level struct{} and [0]Foo, which are stack allocated.
+		return make([]*types.Type, 0, 1)
+	}
+	rts := make([]*types.Type, 0, rcount+1)
+	for _, pa := range apa {
+		if len(pa.Registers) == 0 {
+			continue
+		}
+		rts = appendParamRegs(rts, pa.Type)
+	}
+	return rts
+}
+
+func appendParamRegs(rts []*types.Type, t *types.Type) []*types.Type {
+	if t.IsScalar() || t.IsPtrShaped() {
+		if t.IsComplex() {
+			c := types.FloatForComplex(t)
+			return append(rts, c, c)
+		} else {
+			if int(t.Size()) <= types.RegSize {
+				return append(rts, t)
+			}
+			// assume 64bit int on 32-bit machine
+			// TODO endianness? Should high-order (sign bits) word come first?
+			if t.IsSigned() {
+				rts = append(rts, types.Types[types.TINT32])
+			} else {
+				rts = append(rts, types.Types[types.TUINT32])
+			}
+			return append(rts, types.Types[types.TUINT32])
+		}
+	} else {
+		typ := t.Kind()
+		switch typ {
+		case types.TARRAY:
+			for i := int64(0); i < t.Size(); i++ { // 0 gets no registers, plus future-proofing.
+				rts = appendParamRegs(rts, t.Elem())
+			}
+		case types.TSTRUCT:
+			for _, f := range t.FieldSlice() {
+				if f.Type.Size() > 0 { // embedded zero-width types receive no registers
+					rts = appendParamRegs(rts, f.Type)
+				}
+			}
+		case types.TSLICE:
+			return appendParamRegs(rts, synthSlice)
+		case types.TSTRING:
+			return appendParamRegs(rts, synthString)
+		case types.TINTER:
+			return appendParamRegs(rts, synthIface)
+		}
+	}
+	return rts
+}
+
 // SpillOffset returns the offset *within the spill area* for the parameter that "a" describes.
 // Registers will be spilled here; if a memory home is needed (for a pointer method e.g.)
 // then that will be the address.

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

@@ -14,10 +14,10 @@ import (
 )
 
 type selKey struct {
-	from   *Value
-	offset int64
-	size   int64
-	typ    *types.Type
+	from          *Value // what is selected from
+	offsetOrIndex int64  // whatever is appropriate for the selector
+	size          int64
+	typ           *types.Type
 }
 
 type offsetKey struct {
@@ -372,6 +372,7 @@ func (x *expandState) rewriteSelect(leaf *Value, selector *Value, offset int64,
 		// if applied to Op-mumble-call, the Aux tells us which result, regOffset specifies offset within result.  If a register, should rewrite to OpSelectN for new call.
 		// TODO these may be duplicated. Should memoize. Intermediate selectors will go dead, no worries there.
 		call := selector.Args[0]
+		call0 := call
 		aux := call.Aux.(*AuxCall)
 		which := selector.AuxInt
 		if which == aux.NResults() { // mem is after the results.
@@ -398,7 +399,6 @@ func (x *expandState) rewriteSelect(leaf *Value, selector *Value, offset int64,
 			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 call.Op == OpStaticLECall { // TODO this is temporary until all calls are register-able
 					// Create a "mem" for any loads that need to occur.
@@ -413,15 +413,30 @@ func (x *expandState) rewriteSelect(leaf *Value, selector *Value, offset int64,
 						call = mem
 					}
 				}
-				if leaf.Block == call.Block {
-					leaf.reset(OpLoad)
-					leaf.SetArgs2(off, call)
-					leaf.Type = leafType
+				outParam := aux.abiInfo.OutParam(int(which))
+				if len(outParam.Registers) > 0 {
+					reg := int64(outParam.Registers[regOffset])
+					if leaf.Block == call.Block {
+						leaf.reset(OpSelectN)
+						leaf.SetArgs1(call0)
+						leaf.Type = leafType
+						leaf.AuxInt = reg
+					} else {
+						w := call.Block.NewValue1I(leaf.Pos, OpSelectN, leafType, reg, call0)
+						leaf.copyOf(w)
+					}
 				} else {
-					w := call.Block.NewValue2(leaf.Pos, OpLoad, leafType, off, call)
-					leaf.copyOf(w)
-					if x.debug {
-						fmt.Printf("\tnew %s\n", w.LongString())
+					off := x.offsetFrom(x.sp, offset+aux.OffsetOfResult(which), pt)
+					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 x.debug {
+							fmt.Printf("\tnew %s\n", w.LongString())
+						}
 					}
 				}
 				for _, s := range x.namedSelects[selector] {
@@ -812,7 +827,7 @@ func (x *expandState) storeArgOrLoad(pos src.XPos, b *Block, source, mem *Value,
 		s = b.NewValue3A(pos, OpStore, types.TypeMem, t, dst, source, mem)
 	}
 	if x.debug {
-		fmt.Printf("\t\tstoreArg returns %s\n", s.LongString())
+		fmt.Printf("\t\tstoreArg returns %s, storeRc=%s\n", s.LongString(), storeRc.String())
 	}
 	return s
 }
@@ -983,9 +998,11 @@ func expandCalls(f *Func) {
 					mem = x.storeArgOrLoad(v.Pos, b, a, mem, aux.TypeOfResult(i), auxOffset, 0, rc)
 				}
 			}
-			// TODO REGISTER -- keep the Result for block control, splice in contents of AllResults
-			b.SetControl(mem)
-			v.reset(OpInvalid) // otherwise it can have a mem operand which will fail check(), even though it is dead.
+			v.resetArgs()
+			v.AddArgs(allResults...)
+			v.AddArg(mem)
+			v.Type = types.NewResults(append(abi.RegisterTypes(aux.abiInfo.OutParams()), types.TypeMem))
+			b.SetControl(v)
 		}
 	}
 
@@ -1170,7 +1187,7 @@ func expandCalls(f *Func) {
 		case OpArraySelect:
 			offset = size * v.AuxInt
 		case OpSelectN:
-			offset = w.Aux.(*AuxCall).OffsetOfResult(v.AuxInt)
+			offset = v.AuxInt // offset is just a key, really.
 		case OpInt64Hi:
 			offset = x.hiOffset
 		case OpInt64Lo:
@@ -1182,7 +1199,7 @@ func expandCalls(f *Func) {
 		case OpComplexImag:
 			offset = size
 		}
-		sk := selKey{from: w, size: size, offset: offset, typ: typ}
+		sk := selKey{from: w, size: size, offsetOrIndex: offset, typ: typ}
 		dupe := x.commonSelectors[sk]
 		if dupe == nil {
 			x.commonSelectors[sk] = v
@@ -1240,8 +1257,9 @@ func expandCalls(f *Func) {
 				x.rewriteArgToMemOrRegs(v)
 			case OpStaticLECall:
 				v.Op = OpStaticCall
+				rts := abi.RegisterTypes(v.Aux.(*AuxCall).abiInfo.OutParams())
 				// TODO need to insert all the register types.
-				v.Type = types.NewResults([]*types.Type{types.TypeMem})
+				v.Type = types.NewResults(append(rts, types.TypeMem))
 			case OpClosureLECall:
 				v.Op = OpClosureCall
 				v.Type = types.TypeMem

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

@@ -24,7 +24,7 @@ func checkLower(f *Func) {
 			case OpSP, OpSB, OpInitMem, OpArg, OpArgIntReg, OpArgFloatReg, OpPhi, OpVarDef, OpVarKill, OpVarLive, OpKeepAlive, OpSelect0, OpSelect1, OpSelectN, OpConvert, OpInlMark:
 				continue // ok not to lower
 			case OpMakeResult:
-				if len(b.Controls) == 1 && b.Controls[0] == v {
+				if b.Controls[0] == v {
 					continue
 				}
 			case OpGetG:
@@ -34,6 +34,7 @@ func checkLower(f *Func) {
 				}
 			}
 			s := "not lowered: " + v.String() + ", " + v.Op.String() + " " + v.Type.SimpleString()
+
 			for _, a := range v.Args {
 				s += " " + a.Type.SimpleString()
 			}

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

@@ -134,6 +134,24 @@ func (a *AuxCall) Reg(i *regInfo, c *Config) *regInfo {
 	return a.reg
 }
 
+func (a *AuxCall) ResultReg(c *Config) *regInfo {
+	if a.abiInfo.OutRegistersUsed() == 0 {
+		return a.reg
+	}
+	if len(a.reg.inputs) > 0 {
+		return a.reg
+	}
+	k := 0
+	for _, p := range a.abiInfo.OutParams() {
+		for _, r := range p.Registers {
+			m := archRegForAbiReg(r, c)
+			a.reg.inputs = append(a.reg.inputs, inputInfo{idx: k, regs: (1 << m)})
+			k++
+		}
+	}
+	return a.reg
+}
+
 func archRegForAbiReg(r abi.RegIndex, c *Config) uint8 {
 	var m int8
 	if int(r) < len(c.intParamRegs) {
@@ -285,10 +303,13 @@ func ClosureAuxCall(args []Param, results []Param, paramResultInfo *abi.ABIParam
 func (*AuxCall) CanBeAnSSAAux() {}
 
 // OwnAuxCall returns a function's own AuxCall
-
 func OwnAuxCall(fn *obj.LSym, args []Param, results []Param, paramResultInfo *abi.ABIParamResultInfo) *AuxCall {
 	// TODO if this remains identical to ClosureAuxCall above after new ABI is done, should deduplicate.
-	return &AuxCall{Fn: fn, args: args, results: results, abiInfo: paramResultInfo}
+	var reg *regInfo
+	if paramResultInfo.InRegistersUsed()+paramResultInfo.OutRegistersUsed() > 0 {
+		reg = &regInfo{}
+	}
+	return &AuxCall{Fn: fn, args: args, results: results, abiInfo: paramResultInfo, reg: reg}
 }
 
 const (

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

@@ -830,6 +830,9 @@ func (s *regAllocState) regspec(v *Value) regInfo {
 			return *ac.Reg(&opcodeTable[op].reg, s.f.Config)
 		}
 	}
+	if op == OpMakeResult && s.f.OwnAux.reg != nil {
+		return *s.f.OwnAux.ResultReg(s.f.Config)
+	}
 	return opcodeTable[op].reg
 }
 

test/abi/fibish.go

@@ -0,0 +1,33 @@
+// 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"
+
+// Test that register results are correctly returned (and passed)
+
+//go:registerparams
+//go:noinline
+func f(x int) (int, int) {
+
+	if x < 3 {
+		return 0, x
+	}
+
+	a, b := f(x - 2)
+	c, d := f(x - 1)
+	return a + d, b + c
+}
+
+func main() {
+	x := 40
+	a, b := f(x)
+	fmt.Printf("f(%d)=%d,%d\n", x, a, b)
+}

test/abi/fibish.out

@@ -0,0 +1 @@
+f(40)=39088169,126491972