cmd/compile/internal/gc: refactor ACALL Prog creation

This abstracts creation of ACALL Progs into package gc. The main
benefit of this today is we can refactor away a lot of common
boilerplate code.

Later, once liveness analysis happens on the SSA graph, this will also
provide an easy insertion point for emitting the PCDATA Progs
immediately before call instructions.

Passes toolstash-check -all.

Change-Id: Ia15108ace97201cd84314f1ca916dfeb4f09d61c
Reviewed-on: https://go-review.googlesource.com/38081
Reviewed-by: Keith Randall <khr@golang.org>

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

@@ -22,6 +22,7 @@ func Init() {
 	gc.Thearch.MAXWIDTH = 1 << 50
 
 	gc.Thearch.Defframe = defframe
+	gc.Thearch.Ginsnop = ginsnop
 	gc.Thearch.Proginfo = proginfo
 
 	gc.Thearch.SSAMarkMoves = ssaMarkMoves

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

@@ -743,39 +743,8 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 			q.To.Type = obj.TYPE_REG
 			q.To.Reg = r
 		}
-	case ssa.OpAMD64CALLstatic:
-		if v.Aux.(*obj.LSym) == gc.Deferreturn {
-			// Deferred calls will appear to be returning to
-			// the CALL deferreturn(SB) that we are about to emit.
-			// However, the stack trace code will show the line
-			// of the instruction byte before the return PC.
-			// To avoid that being an unrelated instruction,
-			// insert an actual hardware NOP that will have the right line number.
-			// This is different from obj.ANOP, which is a virtual no-op
-			// that doesn't make it into the instruction stream.
-			ginsnop()
-		}
-		p := gc.Prog(obj.ACALL)
-		p.To.Type = obj.TYPE_MEM
-		p.To.Name = obj.NAME_EXTERN
-		p.To.Sym = v.Aux.(*obj.LSym)
-		if gc.Maxarg < v.AuxInt {
-			gc.Maxarg = v.AuxInt
-		}
-	case ssa.OpAMD64CALLclosure:
-		p := gc.Prog(obj.ACALL)
-		p.To.Type = obj.TYPE_REG
-		p.To.Reg = v.Args[0].Reg()
-		if gc.Maxarg < v.AuxInt {
-			gc.Maxarg = v.AuxInt
-		}
-	case ssa.OpAMD64CALLinter:
-		p := gc.Prog(obj.ACALL)
-		p.To.Type = obj.TYPE_REG
-		p.To.Reg = v.Args[0].Reg()
-		if gc.Maxarg < v.AuxInt {
-			gc.Maxarg = v.AuxInt
-		}
+	case ssa.OpAMD64CALLstatic, ssa.OpAMD64CALLclosure, ssa.OpAMD64CALLinter:
+		s.Call(v)
 	case ssa.OpAMD64NEGQ, ssa.OpAMD64NEGL,
 		ssa.OpAMD64BSWAPQ, ssa.OpAMD64BSWAPL,
 		ssa.OpAMD64NOTQ, ssa.OpAMD64NOTL:

src/cmd/compile/internal/arm/galign.go

@@ -16,6 +16,7 @@ func Init() {
 	gc.Thearch.MAXWIDTH = (1 << 32) - 1
 
 	gc.Thearch.Defframe = defframe
+	gc.Thearch.Ginsnop = ginsnop
 	gc.Thearch.Proginfo = proginfo
 
 	gc.Thearch.SSAMarkMoves = func(s *gc.SSAGenState, b *ssa.Block) {}

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

@@ -625,41 +625,8 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 		p.From.Offset = v.AuxInt
 		p.To.Type = obj.TYPE_REG
 		p.To.Reg = v.Reg()
-	case ssa.OpARMCALLstatic:
-		if v.Aux.(*obj.LSym) == gc.Deferreturn {
-			// Deferred calls will appear to be returning to
-			// the CALL deferreturn(SB) that we are about to emit.
-			// However, the stack trace code will show the line
-			// of the instruction byte before the return PC.
-			// To avoid that being an unrelated instruction,
-			// insert an actual hardware NOP that will have the right line number.
-			// This is different from obj.ANOP, which is a virtual no-op
-			// that doesn't make it into the instruction stream.
-			ginsnop()
-		}
-		p := gc.Prog(obj.ACALL)
-		p.To.Type = obj.TYPE_MEM
-		p.To.Name = obj.NAME_EXTERN
-		p.To.Sym = v.Aux.(*obj.LSym)
-		if gc.Maxarg < v.AuxInt {
-			gc.Maxarg = v.AuxInt
-		}
-	case ssa.OpARMCALLclosure:
-		p := gc.Prog(obj.ACALL)
-		p.To.Type = obj.TYPE_MEM
-		p.To.Offset = 0
-		p.To.Reg = v.Args[0].Reg()
-		if gc.Maxarg < v.AuxInt {
-			gc.Maxarg = v.AuxInt
-		}
-	case ssa.OpARMCALLinter:
-		p := gc.Prog(obj.ACALL)
-		p.To.Type = obj.TYPE_MEM
-		p.To.Offset = 0
-		p.To.Reg = v.Args[0].Reg()
-		if gc.Maxarg < v.AuxInt {
-			gc.Maxarg = v.AuxInt
-		}
+	case ssa.OpARMCALLstatic, ssa.OpARMCALLclosure, ssa.OpARMCALLinter:
+		s.Call(v)
 	case ssa.OpARMDUFFZERO:
 		p := gc.Prog(obj.ADUFFZERO)
 		p.To.Type = obj.TYPE_MEM

src/cmd/compile/internal/arm64/galign.go

@@ -16,6 +16,7 @@ func Init() {
 	gc.Thearch.MAXWIDTH = 1 << 50
 
 	gc.Thearch.Defframe = defframe
+	gc.Thearch.Ginsnop = ginsnop
 	gc.Thearch.Proginfo = proginfo
 
 	gc.Thearch.SSAMarkMoves = func(s *gc.SSAGenState, b *ssa.Block) {}

src/cmd/compile/internal/arm64/ssa.go

@@ -622,41 +622,8 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 		p4 := gc.Prog(arm64.ABLE)
 		p4.To.Type = obj.TYPE_BRANCH
 		gc.Patch(p4, p)
-	case ssa.OpARM64CALLstatic:
-		if v.Aux.(*obj.LSym) == gc.Deferreturn {
-			// Deferred calls will appear to be returning to
-			// the CALL deferreturn(SB) that we are about to emit.
-			// However, the stack trace code will show the line
-			// of the instruction byte before the return PC.
-			// To avoid that being an unrelated instruction,
-			// insert an actual hardware NOP that will have the right line number.
-			// This is different from obj.ANOP, which is a virtual no-op
-			// that doesn't make it into the instruction stream.
-			ginsnop()
-		}
-		p := gc.Prog(obj.ACALL)
-		p.To.Type = obj.TYPE_MEM
-		p.To.Name = obj.NAME_EXTERN
-		p.To.Sym = v.Aux.(*obj.LSym)
-		if gc.Maxarg < v.AuxInt {
-			gc.Maxarg = v.AuxInt
-		}
-	case ssa.OpARM64CALLclosure:
-		p := gc.Prog(obj.ACALL)
-		p.To.Type = obj.TYPE_MEM
-		p.To.Offset = 0
-		p.To.Reg = v.Args[0].Reg()
-		if gc.Maxarg < v.AuxInt {
-			gc.Maxarg = v.AuxInt
-		}
-	case ssa.OpARM64CALLinter:
-		p := gc.Prog(obj.ACALL)
-		p.To.Type = obj.TYPE_MEM
-		p.To.Offset = 0
-		p.To.Reg = v.Args[0].Reg()
-		if gc.Maxarg < v.AuxInt {
-			gc.Maxarg = v.AuxInt
-		}
+	case ssa.OpARM64CALLstatic, ssa.OpARM64CALLclosure, ssa.OpARM64CALLinter:
+		s.Call(v)
 	case ssa.OpARM64LoweredNilCheck:
 		// Issue a load which will fault if arg is nil.
 		p := gc.Prog(arm64.AMOVB)

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

@@ -365,10 +365,11 @@ type Arch struct {
 
 	REGSP    int
 	MAXWIDTH int64
+	Use387   bool // should 386 backend use 387 FP instructions instead of sse2.
 
 	Defframe func(*obj.Prog)
+	Ginsnop  func()
 	Proginfo func(*obj.Prog) ProgInfo
-	Use387   bool // should 8g use 387 FP instructions instead of sse2.
 
 	// SSAMarkMoves marks any MOVXconst ops that need to avoid clobbering flags.
 	SSAMarkMoves func(*SSAGenState, *ssa.Block)

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

@@ -4756,6 +4756,42 @@ func (s *SSAGenState) AddrScratch(a *obj.Addr) {
 	a.Offset = s.ScratchFpMem.Xoffset
 }
 
+func (s *SSAGenState) Call(v *ssa.Value) *obj.Prog {
+	if sym, _ := v.Aux.(*obj.LSym); sym == Deferreturn {
+		// Deferred calls will appear to be returning to
+		// the CALL deferreturn(SB) that we are about to emit.
+		// However, the stack trace code will show the line
+		// of the instruction byte before the return PC.
+		// To avoid that being an unrelated instruction,
+		// insert an actual hardware NOP that will have the right line number.
+		// This is different from obj.ANOP, which is a virtual no-op
+		// that doesn't make it into the instruction stream.
+		Thearch.Ginsnop()
+	}
+
+	p := Prog(obj.ACALL)
+	if sym, ok := v.Aux.(*obj.LSym); ok {
+		p.To.Type = obj.TYPE_MEM
+		p.To.Name = obj.NAME_EXTERN
+		p.To.Sym = sym
+	} else {
+		// TODO(mdempsky): Can these differences be eliminated?
+		switch Thearch.LinkArch.Family {
+		case sys.AMD64, sys.I386, sys.PPC64, sys.S390X:
+			p.To.Type = obj.TYPE_REG
+		case sys.ARM, sys.ARM64, sys.MIPS, sys.MIPS64:
+			p.To.Type = obj.TYPE_MEM
+		default:
+			Fatalf("unknown indirect call family")
+		}
+		p.To.Reg = v.Args[0].Reg()
+	}
+	if Maxarg < v.AuxInt {
+		Maxarg = v.AuxInt
+	}
+	return p
+}
+
 // fieldIdx finds the index of the field referred to by the ODOT node n.
 func fieldIdx(n *Node) int {
 	t := n.Left.Type

src/cmd/compile/internal/mips/galign.go

@@ -19,6 +19,7 @@ func Init() {
 	gc.Thearch.REGSP = mips.REGSP
 	gc.Thearch.MAXWIDTH = (1 << 31) - 1
 	gc.Thearch.Defframe = defframe
+	gc.Thearch.Ginsnop = ginsnop
 	gc.Thearch.Proginfo = proginfo
 	gc.Thearch.SSAMarkMoves = func(s *gc.SSAGenState, b *ssa.Block) {}
 	gc.Thearch.SSAGenValue = ssaGenValue

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

@@ -477,41 +477,8 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 		p6.Reg = mips.REG_R1
 		p6.To.Type = obj.TYPE_BRANCH
 		gc.Patch(p6, p2)
-	case ssa.OpMIPSCALLstatic:
-		if v.Aux.(*obj.LSym) == gc.Deferreturn {
-			// Deferred calls will appear to be returning to
-			// the CALL deferreturn(SB) that we are about to emit.
-			// However, the stack trace code will show the line
-			// of the instruction byte before the return PC.
-			// To avoid that being an unrelated instruction,
-			// insert an actual hardware NOP that will have the right line number.
-			// This is different from obj.ANOP, which is a virtual no-op
-			// that doesn't make it into the instruction stream.
-			ginsnop()
-		}
-		p := gc.Prog(obj.ACALL)
-		p.To.Type = obj.TYPE_MEM
-		p.To.Name = obj.NAME_EXTERN
-		p.To.Sym = v.Aux.(*obj.LSym)
-		if gc.Maxarg < v.AuxInt {
-			gc.Maxarg = v.AuxInt
-		}
-	case ssa.OpMIPSCALLclosure:
-		p := gc.Prog(obj.ACALL)
-		p.To.Type = obj.TYPE_MEM
-		p.To.Offset = 0
-		p.To.Reg = v.Args[0].Reg()
-		if gc.Maxarg < v.AuxInt {
-			gc.Maxarg = v.AuxInt
-		}
-	case ssa.OpMIPSCALLinter:
-		p := gc.Prog(obj.ACALL)
-		p.To.Type = obj.TYPE_MEM
-		p.To.Offset = 0
-		p.To.Reg = v.Args[0].Reg()
-		if gc.Maxarg < v.AuxInt {
-			gc.Maxarg = v.AuxInt
-		}
+	case ssa.OpMIPSCALLstatic, ssa.OpMIPSCALLclosure, ssa.OpMIPSCALLinter:
+		s.Call(v)
 	case ssa.OpMIPSLoweredAtomicLoad:
 		gc.Prog(mips.ASYNC)
 

src/cmd/compile/internal/mips64/galign.go

@@ -20,6 +20,7 @@ func Init() {
 	gc.Thearch.MAXWIDTH = 1 << 50
 
 	gc.Thearch.Defframe = defframe
+	gc.Thearch.Ginsnop = ginsnop
 	gc.Thearch.Proginfo = proginfo
 
 	gc.Thearch.SSAMarkMoves = func(s *gc.SSAGenState, b *ssa.Block) {}

src/cmd/compile/internal/mips64/ssa.go

@@ -480,41 +480,8 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 		p6.Reg = mips.REG_R1
 		p6.To.Type = obj.TYPE_BRANCH
 		gc.Patch(p6, p2)
-	case ssa.OpMIPS64CALLstatic:
-		if v.Aux.(*obj.LSym) == gc.Deferreturn {
-			// Deferred calls will appear to be returning to
-			// the CALL deferreturn(SB) that we are about to emit.
-			// However, the stack trace code will show the line
-			// of the instruction byte before the return PC.
-			// To avoid that being an unrelated instruction,
-			// insert an actual hardware NOP that will have the right line number.
-			// This is different from obj.ANOP, which is a virtual no-op
-			// that doesn't make it into the instruction stream.
-			ginsnop()
-		}
-		p := gc.Prog(obj.ACALL)
-		p.To.Type = obj.TYPE_MEM
-		p.To.Name = obj.NAME_EXTERN
-		p.To.Sym = v.Aux.(*obj.LSym)
-		if gc.Maxarg < v.AuxInt {
-			gc.Maxarg = v.AuxInt
-		}
-	case ssa.OpMIPS64CALLclosure:
-		p := gc.Prog(obj.ACALL)
-		p.To.Type = obj.TYPE_MEM
-		p.To.Offset = 0
-		p.To.Reg = v.Args[0].Reg()
-		if gc.Maxarg < v.AuxInt {
-			gc.Maxarg = v.AuxInt
-		}
-	case ssa.OpMIPS64CALLinter:
-		p := gc.Prog(obj.ACALL)
-		p.To.Type = obj.TYPE_MEM
-		p.To.Offset = 0
-		p.To.Reg = v.Args[0].Reg()
-		if gc.Maxarg < v.AuxInt {
-			gc.Maxarg = v.AuxInt
-		}
+	case ssa.OpMIPS64CALLstatic, ssa.OpMIPS64CALLclosure, ssa.OpMIPS64CALLinter:
+		s.Call(v)
 	case ssa.OpMIPS64LoweredNilCheck:
 		// Issue a load which will fault if arg is nil.
 		p := gc.Prog(mips.AMOVB)

src/cmd/compile/internal/ppc64/galign.go

@@ -19,6 +19,7 @@ func Init() {
 	gc.Thearch.MAXWIDTH = 1 << 50
 
 	gc.Thearch.Defframe = defframe
+	gc.Thearch.Ginsnop = ginsnop2
 	gc.Thearch.Proginfo = proginfo
 
 	gc.Thearch.SSAMarkMoves = ssaMarkMoves

src/cmd/compile/internal/ppc64/ggen.go

@@ -98,3 +98,29 @@ func ginsnop() {
 	p.To.Type = obj.TYPE_REG
 	p.To.Reg = ppc64.REG_R0
 }
+
+func ginsnop2() {
+	// PPC64 is unusual because TWO nops are required
+	// (see gc/cgen.go, gc/plive.go -- copy of comment below)
+	//
+	// On ppc64, when compiling Go into position
+	// independent code on ppc64le we insert an
+	// instruction to reload the TOC pointer from the
+	// stack as well. See the long comment near
+	// jmpdefer in runtime/asm_ppc64.s for why.
+	// If the MOVD is not needed, insert a hardware NOP
+	// so that the same number of instructions are used
+	// on ppc64 in both shared and non-shared modes.
+
+	ginsnop()
+	if gc.Ctxt.Flag_shared {
+		p := gc.Prog(ppc64.AMOVD)
+		p.From.Type = obj.TYPE_MEM
+		p.From.Offset = 24
+		p.From.Reg = ppc64.REGSP
+		p.To.Type = obj.TYPE_REG
+		p.To.Reg = ppc64.REG_R2
+	} else {
+		ginsnop()
+	}
+}