[dev.ssa] cmd/compile: more fixes for 386 shared libraries

Use the destination register for materializing the pc
for GOT references also. See https://go-review.googlesource.com/c/25442/
The SSA backend assumes CX does not get clobbered for these instructions.

Mark duffzero as clobbering CX. The linker needs to clobber CX
to materialize the address to call. (This affects the non-shared-library
duffzero also, but hopefully forbidding one register across duffzero
won't be a big deal.)

Hopefully this is all the cases where the linker is clobbering CX
under the hood and SSA assumes it isn't.

Change-Id: I080c938170193df57cd5ce1f2a956b68a34cc886
Reviewed-on: https://go-review.googlesource.com/26611
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Michael Hudson-Doyle <michael.hudson@canonical.com>

src/cmd/compile/internal/ssa/gen/386.rules

@@ -669,7 +669,8 @@
 
 // We need to fold LEAQ into the MOVx ops so that the live variable analysis knows
 // what variables are being read/written by the ops.
-// Note: we turn off this merging for operations on globals when building position-independent code.
+// Note: we turn off this merging for operations on globals when building
+// position-independent code (when Flag_shared is set).
 // PIC needs a spare register to load the PC into. For loads from globals into integer registers we use
 // the target register, but for other loads and all stores, we need a free register. Having the LEAL be
 // a separate instruction gives us that register.

src/cmd/compile/internal/ssa/gen/386Ops.go

@@ -362,7 +362,8 @@ func init() {
 			argLength: 3,
 			reg: regInfo{
 				inputs:   []regMask{buildReg("DI"), buildReg("AX")},
-				clobbers: buildReg("DI"),
+				clobbers: buildReg("DI CX"),
+				// Note: CX is only clobbered when dynamic linking.
 			},
 		},
 

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

@@ -3511,7 +3511,7 @@ var opcodeTable = [...]opInfo{
 				{0, 128}, // DI
 				{1, 1},   // AX
 			},
-			clobbers: 128, // DI
+			clobbers: 130, // CX DI
 		},
 	},
 	{

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

@@ -334,6 +334,16 @@ func rewriteToUseGot(ctxt *obj.Link, p *obj.Prog) {
 		lea = ALEAL
 		mov = AMOVL
 		reg = REG_CX
+		if p.To.Type == obj.TYPE_REG && p.To.Reg != p.From.Reg && p.To.Reg != p.From.Index {
+			switch p.As {
+			case ALEAL, AMOVL, AMOVWLZX, AMOVBLZX, AMOVWLSX, AMOVBLSX:
+				// Special case: clobber the destination register with
+				// the PC so we don't have to clobber CX.
+				// The SSA backend depends on CX not being clobbered across these instructions.
+				// See cmd/compile/internal/ssa/gen/386.rules (search for Flag_shared).
+				reg = p.To.Reg
+			}
+		}
 	}
 
 	if p.As == obj.ADUFFCOPY || p.As == obj.ADUFFZERO {
@@ -392,7 +402,7 @@ func rewriteToUseGot(ctxt *obj.Link, p *obj.Prog) {
 			dest = p.To
 			p.As = mov
 			p.To.Type = obj.TYPE_REG
-			p.To.Reg = REG_CX
+			p.To.Reg = reg
 			p.To.Sym = nil
 			p.To.Name = obj.NAME_NONE
 		}
@@ -413,7 +423,7 @@ func rewriteToUseGot(ctxt *obj.Link, p *obj.Prog) {
 			q.As = pAs
 			q.To = dest
 			q.From.Type = obj.TYPE_REG
-			q.From.Reg = REG_CX
+			q.From.Reg = reg
 		}
 	}
 	if p.From3 != nil && p.From3.Name == obj.NAME_EXTERN {
@@ -544,14 +554,11 @@ func rewriteToPcrel(ctxt *obj.Link, p *obj.Prog) {
 		return
 	}
 	var dst int16 = REG_CX
-	if isName(&p.From) && p.To.Type == obj.TYPE_REG {
+	if p.To.Type == obj.TYPE_REG && p.To.Reg != p.From.Reg && p.To.Reg != p.From.Index {
 		switch p.As {
 		case ALEAL, AMOVL, AMOVWLZX, AMOVBLZX, AMOVWLSX, AMOVBLSX:
 			dst = p.To.Reg
-			// Special case: clobber the destination register with
-			// the PC so we don't have to clobber CX.
-			// The SSA backend depends on CX not being clobbered across these instructions.
-			// See cmd/compile/internal/ssa/gen/386.rules (search for Flag_shared).
+			// Why?  See the comment near the top of rewriteToUseGot above.
 		}
 	}
 	q := obj.Appendp(ctxt, p)