cmd/compile: pass register parameters to called function

still needs morestack
still needs results
lots of corner cases also not dealt with.

For #40724.

Change-Id: I03abdf1e8363d75c52969560b427e488a48cd37a
Reviewed-on: https://go-review.googlesource.com/c/go/+/293889
Trust: David Chase <[email protected]>
Run-TryBot: David Chase <[email protected]>
TryBot-Result: Go Bot <[email protected]>
Reviewed-by: Cherry Zhang <[email protected]>
Reviewed-by: Jeremy Faller <[email protected]>

Author: dr2chase

src/cmd/compile/internal/ssa/gen/AMD64Ops.go

@@ -771,7 +771,8 @@ func init() {
 			faultOnNilArg0: true,
 		},
 
-		{name: "CALLstatic", argLength: 1, reg: regInfo{clobbers: callerSave}, aux: "CallOff", clobberFlags: true, call: true},                                              // call static function aux.(*obj.LSym).  arg0=mem, auxint=argsize, returns mem
+		// With a register ABI, the actual register info for these instructions (i.e., what is used in regalloc) is augmented with per-call-site bindings of additional arguments to specific registers.
+		{name: "CALLstatic", argLength: -1, reg: regInfo{clobbers: callerSave}, aux: "CallOff", clobberFlags: true, call: true},                                             // call static function aux.(*obj.LSym).  arg0=mem, auxint=argsize, returns mem
 		{name: "CALLclosure", argLength: 3, reg: regInfo{inputs: []regMask{gpsp, buildReg("DX"), 0}, clobbers: callerSave}, aux: "CallOff", clobberFlags: true, call: true}, // call function via closure.  arg0=codeptr, arg1=closure, arg2=mem, auxint=argsize, returns mem
 		{name: "CALLinter", argLength: 2, reg: regInfo{inputs: []regMask{gp}, clobbers: callerSave}, aux: "CallOff", clobberFlags: true, call: true},                        // call fn by pointer.  arg0=codeptr, arg1=mem, auxint=argsize, returns mem
 

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

@@ -96,6 +96,54 @@ type AuxCall struct {
 	abiInfo *abi.ABIParamResultInfo // TODO remove fields above redundant with this information.
 }
 
+// Reg returns the regInfo for a given call, combining the derived in/out register masks
+// with the machine-specific register information in the input i.  (The machine-specific
+// regInfo is much handier at the call site than it is when the AuxCall is being constructed,
+// therefore do this lazily).
+//
+// TODO: there is a Clever Hack that allows pre-generation of a small-ish number of the slices
+// of inputInfo and outputInfo used here, provided that we are willing to reorder the inputs
+// and outputs from calls, so that all integer registers come first, then all floating registers.
+// At this point (active development of register ABI) that is very premature,
+// but if this turns out to be a cost, we could do it.
+func (a *AuxCall) Reg(i *regInfo, c *Config) *regInfo {
+	if a.reg.clobbers != 0 {
+		// Already updated
+		return a.reg
+	}
+	if a.abiInfo.InRegistersUsed()+a.abiInfo.OutRegistersUsed() == 0 {
+		// Shortcut for zero case, also handles old ABI.
+		a.reg = i
+		return a.reg
+	}
+	a.reg.inputs = append(a.reg.inputs, i.inputs...)
+	for _, p := range a.abiInfo.InParams() {
+		for _, r := range p.Registers {
+			m := archRegForAbiReg(r, c)
+			a.reg.inputs = append(a.reg.inputs, inputInfo{idx: len(a.reg.inputs), regs: (1 << m)})
+		}
+	}
+	a.reg.outputs = append(a.reg.outputs, i.outputs...)
+	for _, p := range a.abiInfo.OutParams() {
+		for _, r := range p.Registers {
+			m := archRegForAbiReg(r, c)
+			a.reg.outputs = append(a.reg.outputs, outputInfo{idx: len(a.reg.outputs), regs: (1 << m)})
+		}
+	}
+	a.reg.clobbers = i.clobbers
+	return a.reg
+}
+
+func archRegForAbiReg(r abi.RegIndex, c *Config) uint8 {
+	var m int8
+	if int(r) < len(c.intParamRegs) {
+		m = c.intParamRegs[r]
+	} else {
+		m = c.floatParamRegs[int(r)-len(c.intParamRegs)]
+	}
+	return uint8(m)
+}
+
 // OffsetOfResult returns the SP offset of result which (indexed 0, 1, etc).
 func (a *AuxCall) OffsetOfResult(which int64) int64 {
 	n := int64(a.abiInfo.OutParam(int(which)).Offset())
@@ -217,7 +265,11 @@ func StaticAuxCall(sym *obj.LSym, args []Param, results []Param, paramResultInfo
 	if paramResultInfo == nil {
 		panic(fmt.Errorf("Nil paramResultInfo, sym=%v", sym))
 	}
-	return &AuxCall{Fn: sym, args: args, results: results, abiInfo: paramResultInfo}
+	var reg *regInfo
+	if paramResultInfo.InRegistersUsed()+paramResultInfo.OutRegistersUsed() > 0 {
+		reg = &regInfo{}
+	}
+	return &AuxCall{Fn: sym, args: args, results: results, abiInfo: paramResultInfo, reg: reg}
 }
 
 // InterfaceAuxCall returns an AuxCall for an interface call.

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

@@ -151,6 +151,14 @@ type register uint8
 
 const noRegister register = 255
 
+// For bulk initializing
+var noRegisters [32]register = [32]register{
+	noRegister, noRegister, noRegister, noRegister, noRegister, noRegister, noRegister, noRegister,
+	noRegister, noRegister, noRegister, noRegister, noRegister, noRegister, noRegister, noRegister,
+	noRegister, noRegister, noRegister, noRegister, noRegister, noRegister, noRegister, noRegister,
+	noRegister, noRegister, noRegister, noRegister, noRegister, noRegister, noRegister, noRegister,
+}
+
 // A regMask encodes a set of machine registers.
 // TODO: regMask -> regSet?
 type regMask uint64
@@ -818,9 +826,8 @@ func (s *regAllocState) regspec(v *Value) regInfo {
 		return regInfo{outputs: []outputInfo{{regs: 1 << uint(reg)}}}
 	}
 	if op.IsCall() {
-		// TODO Panic if not okay
 		if ac, ok := v.Aux.(*AuxCall); ok && ac.reg != nil {
-			return *ac.reg
+			return *ac.Reg(&opcodeTable[op].reg, s.f.Config)
 		}
 	}
 	return opcodeTable[op].reg
@@ -1456,7 +1463,8 @@ func (s *regAllocState) regalloc(f *Func) {
 
 			// Pick registers for outputs.
 			{
-				outRegs := [2]register{noRegister, noRegister}
+				outRegs := noRegisters // TODO if this is costly, hoist and clear incrementally below.
+				maxOutIdx := -1
 				var used regMask
 				for _, out := range regspec.outputs {
 					mask := out.regs & s.allocatable &^ used
@@ -1502,6 +1510,9 @@ func (s *regAllocState) regalloc(f *Func) {
 						mask &^= desired.avoid
 					}
 					r := s.allocReg(mask, v)
+					if out.idx > maxOutIdx {
+						maxOutIdx = out.idx
+					}
 					outRegs[out.idx] = r
 					used |= regMask(1) << r
 					s.tmpused |= regMask(1) << r
@@ -1518,8 +1529,14 @@ func (s *regAllocState) regalloc(f *Func) {
 					s.f.setHome(v, outLocs)
 					// Note that subsequent SelectX instructions will do the assignReg calls.
 				} else if v.Type.IsResults() {
-					// TODO register arguments need to make this work
-					panic("Oops, implement this.")
+					// preallocate outLocs to the right size, which is maxOutIdx+1
+					outLocs := make(LocResults, maxOutIdx+1, maxOutIdx+1)
+					for i := 0; i <= maxOutIdx; i++ {
+						if r := outRegs[i]; r != noRegister {
+							outLocs[i] = &s.registers[r]
+						}
+					}
+					s.f.setHome(v, outLocs)
 				} else {
 					if r := outRegs[0]; r != noRegister {
 						s.assignReg(r, v, v)