Merge "Merge branch 'dev.ssa' into mergebranch"

8 files changed, 97 insertions, 54 deletions

diff --git a/src/cmd/internal/obj/link.go b/src/cmd/internal/obj/link.go
index e89c1412e2..4ae819178d 100644
--- a/src/cmd/internal/obj/link.go
+++ b/src/cmd/internal/obj/link.go
@@ -214,14 +214,14 @@ type Prog struct {
 	Spadj  int32
 	As     int16
 	Reg    int16
-	RegTo2 int16 // 2nd register output operand
-	Mark   uint16
+	RegTo2 int16  // 2nd register output operand
+	Mark   uint16 // bitmask of arch-specific items
 	Optab  uint16
 	Scond  uint8
 	Back   uint8
 	Ft     uint8
 	Tt     uint8
-	Isize  uint8
+	Isize  uint8 // size of the instruction in bytes (x86 only)
 	Mode   int8
 
 	Info ProgInfo
diff --git a/src/cmd/internal/obj/obj.go b/src/cmd/internal/obj/obj.go
index 343c93a6ee..f38078fca8 100644
--- a/src/cmd/internal/obj/obj.go
+++ b/src/cmd/internal/obj/obj.go
@@ -25,12 +25,13 @@ import (
 //	  together, so that given (only) calls Push(10, "x.go", 1) and Pop(15),
 //	  virtual line 12 corresponds to x.go line 3.
 type LineHist struct {
-	Top            *LineStack  // current top of stack
-	Ranges         []LineRange // ranges for lookup
-	Dir            string      // directory to qualify relative paths
-	TrimPathPrefix string      // remove leading TrimPath from recorded file names
-	GOROOT         string      // current GOROOT
-	GOROOT_FINAL   string      // target GOROOT
+	Top               *LineStack  // current top of stack
+	Ranges            []LineRange // ranges for lookup
+	Dir               string      // directory to qualify relative paths
+	TrimPathPrefix    string      // remove leading TrimPath from recorded file names
+	PrintFilenameOnly bool        // ignore path when pretty-printing a line; internal use only
+	GOROOT            string      // current GOROOT
+	GOROOT_FINAL      string      // target GOROOT
 }
 
 // A LineStack is an entry in the recorded line history.
@@ -221,20 +222,28 @@ func (h *LineHist) LineString(lineno int) string {
 		return "<unknown line number>"
 	}
 
-	text := fmt.Sprintf("%s:%d", stk.File, stk.fileLineAt(lineno))
+	filename := stk.File
+	if h.PrintFilenameOnly {
+		filename = filepath.Base(filename)
+	}
+	text := fmt.Sprintf("%s:%d", filename, stk.fileLineAt(lineno))
 	if stk.Directive && stk.Parent != nil {
 		stk = stk.Parent
-		text += fmt.Sprintf("[%s:%d]", stk.File, stk.fileLineAt(lineno))
+		filename = stk.File
+		if h.PrintFilenameOnly {
+			filename = filepath.Base(filename)
+		}
+		text += fmt.Sprintf("[%s:%d]", filename, stk.fileLineAt(lineno))
 	}
 	const showFullStack = false // was used by old C compilers
 	if showFullStack {
 		for stk.Parent != nil {
 			lineno = stk.Lineno - 1
 			stk = stk.Parent
-			text += fmt.Sprintf(" %s:%d", stk.File, stk.fileLineAt(lineno))
+			text += fmt.Sprintf(" %s:%d", filename, stk.fileLineAt(lineno))
 			if stk.Directive && stk.Parent != nil {
 				stk = stk.Parent
-				text += fmt.Sprintf("[%s:%d]", stk.File, stk.fileLineAt(lineno))
+				text += fmt.Sprintf("[%s:%d]", filename, stk.fileLineAt(lineno))
 			}
 		}
 	}
diff --git a/src/cmd/internal/obj/pass.go b/src/cmd/internal/obj/pass.go
index ae38c4130a..f10fc60413 100644
--- a/src/cmd/internal/obj/pass.go
+++ b/src/cmd/internal/obj/pass.go
@@ -204,7 +204,6 @@ func linkpatch(ctxt *Link, sym *LSym) {
 
 	if ctxt.Flag_optimize {
 		for p := sym.Text; p != nil; p = p.Link {
-			p.Mark = 0 /* initialization for follow */
 			if p.Pcond != nil {
 				p.Pcond = brloop(ctxt, p.Pcond)
 				if p.Pcond != nil {
diff --git a/src/cmd/internal/obj/util.go b/src/cmd/internal/obj/util.go
index 8ecf800502..b6e644372f 100644
--- a/src/cmd/internal/obj/util.go
+++ b/src/cmd/internal/obj/util.go
@@ -406,6 +406,9 @@ func Dconv(p *Prog, a *Addr) string {
 		if a.Index != REG_NONE {
 			str += fmt.Sprintf("(%v*%d)", Rconv(int(a.Index)), int(a.Scale))
 		}
+		if p.As == ATYPE && a.Gotype != nil {
+			str += fmt.Sprintf("%s", a.Gotype.Name)
+		}
 
 	case TYPE_CONST:
 		if a.Reg != 0 {
diff --git a/src/cmd/internal/obj/x86/a.out.go b/src/cmd/internal/obj/x86/a.out.go
index dacf612bc2..64bd865e42 100644
--- a/src/cmd/internal/obj/x86/a.out.go
+++ b/src/cmd/internal/obj/x86/a.out.go
@@ -34,6 +34,12 @@ import "cmd/internal/obj"
 
 //go:generate go run ../stringer.go -i $GOFILE -o anames.go -p x86
 
+const (
+	/* mark flags */
+	DONE          = 1 << iota
+	PRESERVEFLAGS // not allowed to clobber flags
+)
+
 /*
  *	amd64
  */
@@ -114,23 +120,23 @@ const (
 	AINTO
 	AIRETL
 	AIRETW
-	AJCC
-	AJCS
+	AJCC // >= unsigned
+	AJCS // < unsigned
 	AJCXZL
-	AJEQ
-	AJGE
-	AJGT
-	AJHI
-	AJLE
-	AJLS
-	AJLT
-	AJMI
-	AJNE
-	AJOC
-	AJOS
-	AJPC
-	AJPL
-	AJPS
+	AJEQ // == (zero)
+	AJGE // >= signed
+	AJGT // > signed
+	AJHI // > unsigned
+	AJLE // <= signed
+	AJLS // <= unsigned
+	AJLT // < signed
+	AJMI // sign bit set (negative)
+	AJNE // != (nonzero)
+	AJOC // overflow clear
+	AJOS // overflow set
+	AJPC // parity clear
+	AJPL // sign bit clear (positive)
+	AJPS // parity set
 	ALAHF
 	ALARL
 	ALARW
diff --git a/src/cmd/internal/obj/x86/asm6.go b/src/cmd/internal/obj/x86/asm6.go
index 115350637f..9ab6615e43 100644
--- a/src/cmd/internal/obj/x86/asm6.go
+++ b/src/cmd/internal/obj/x86/asm6.go
@@ -1895,7 +1895,7 @@ func span6(ctxt *obj.Link, s *obj.LSym) {
 
 			// process forward jumps to p
 			for q = p.Rel; q != nil; q = q.Forwd {
-				v = int32(p.Pc - (q.Pc + int64(q.Mark)))
+				v = int32(p.Pc - (q.Pc + int64(q.Isize)))
 				if q.Back&2 != 0 { // short
 					if v > 127 {
 						loop++
@@ -1908,7 +1908,7 @@ func span6(ctxt *obj.Link, s *obj.LSym) {
 						s.P[q.Pc+1] = byte(v)
 					}
 				} else {
-					bp = s.P[q.Pc+int64(q.Mark)-4:]
+					bp = s.P[q.Pc+int64(q.Isize)-4:]
 					bp[0] = byte(v)
 					bp = bp[1:]
 					bp[0] = byte(v >> 8)
@@ -1931,7 +1931,6 @@ func span6(ctxt *obj.Link, s *obj.LSym) {
 
 			obj.Symgrow(ctxt, s, p.Pc+int64(m))
 			copy(s.P[p.Pc:][:m], ctxt.And[:m])
-			p.Mark = uint16(m)
 			c += int32(m)
 		}
 
@@ -2272,8 +2271,28 @@ func oclass(ctxt *obj.Link, p *obj.Prog, a *obj.Addr) int {
 		return Yxxx
 
 	case obj.TYPE_MEM:
-		if a.Name != obj.NAME_NONE {
-			if ctxt.Asmode == 64 && (a.Reg != REG_NONE || a.Index != REG_NONE || a.Scale != 0) {
+		if a.Index == REG_SP {
+			// Can't use SP as the index register
+			return Yxxx
+		}
+		if ctxt.Asmode == 64 {
+			switch a.Name {
+			case obj.NAME_EXTERN, obj.NAME_STATIC, obj.NAME_GOTREF:
+				// Global variables can't use index registers and their
+				// base register is %rip (%rip is encoded as REG_NONE).
+				if a.Reg != REG_NONE || a.Index != REG_NONE || a.Scale != 0 {
+					return Yxxx
+				}
+			case obj.NAME_AUTO, obj.NAME_PARAM:
+				// These names must have a base of SP.  The old compiler
+				// uses 0 for the base register.  SSA uses REG_SP.
+				if a.Reg != REG_SP && a.Reg != 0 {
+					return Yxxx
+				}
+			case obj.NAME_NONE:
+				// everything is ok
+			default:
+				// unknown name
 				return Yxxx
 			}
 		}
@@ -2321,6 +2340,10 @@ func oclass(ctxt *obj.Link, p *obj.Prog, a *obj.Addr) int {
 			v = int64(int32(v))
 		}
 		if v == 0 {
+			if p.Mark&PRESERVEFLAGS != 0 {
+				// If PRESERVEFLAGS is set, avoid MOV $0, AX turning into XOR AX, AX.
+				return Yu7
+			}
 			return Yi0
 		}
 		if v == 1 {
diff --git a/src/cmd/internal/obj/x86/obj6.go b/src/cmd/internal/obj/x86/obj6.go
index 6332b7e4d9..3f8426ae38 100644
--- a/src/cmd/internal/obj/x86/obj6.go
+++ b/src/cmd/internal/obj/x86/obj6.go
@@ -38,7 +38,7 @@ import (
 	"math"
 )
 
-func canuse1insntls(ctxt *obj.Link) bool {
+func CanUse1InsnTLS(ctxt *obj.Link) bool {
 	if isAndroid {
 		// For android, we use a disgusting hack that assumes
 		// the thread-local storage slot for g is allocated
@@ -130,7 +130,7 @@ func progedit(ctxt *obj.Link, p *obj.Prog) {
 	// rewriting the instructions more comprehensively, and it only does because
 	// we only support a single TLS variable (g).
 
-	if canuse1insntls(ctxt) {
+	if CanUse1InsnTLS(ctxt) {
 		// Reduce 2-instruction sequence to 1-instruction sequence.
 		// Sequences like
 		//	MOVQ TLS, BX
@@ -231,7 +231,8 @@ func progedit(ctxt *obj.Link, p *obj.Prog) {
 	// Convert AMOVSS $(0), Xx to AXORPS Xx, Xx
 	case AMOVSS:
 		if p.From.Type == obj.TYPE_FCONST {
-			if p.From.Val.(float64) == 0 {
+			//  f == 0 can't be used here due to -0, so use Float64bits
+			if f := p.From.Val.(float64); math.Float64bits(f) == 0 {
 				if p.To.Type == obj.TYPE_REG && REG_X0 <= p.To.Reg && p.To.Reg <= REG_X15 {
 					p.As = AXORPS
 					p.From = p.To
@@ -271,7 +272,8 @@ func progedit(ctxt *obj.Link, p *obj.Prog) {
 	case AMOVSD:
 		// Convert AMOVSD $(0), Xx to AXORPS Xx, Xx
 		if p.From.Type == obj.TYPE_FCONST {
-			if p.From.Val.(float64) == 0 {
+			//  f == 0 can't be used here due to -0, so use Float64bits
+			if f := p.From.Val.(float64); math.Float64bits(f) == 0 {
 				if p.To.Type == obj.TYPE_REG && REG_X0 <= p.To.Reg && p.To.Reg <= REG_X15 {
 					p.As = AXORPS
 					p.From = p.To
@@ -375,7 +377,7 @@ func rewriteToUseGot(ctxt *obj.Link, p *obj.Prog) {
 	}
 	if p.From.Type == obj.TYPE_ADDR && p.From.Name == obj.NAME_EXTERN && !p.From.Sym.Local {
 		// $MOV $sym, Rx becomes $MOV sym@GOT, Rx
-		// $MOV $sym+<off>, Rx becomes $MOV sym@GOT, Rx; $ADD <off>, Rx
+		// $MOV $sym+<off>, Rx becomes $MOV sym@GOT, Rx; $LEA <off>(Rx), Rx
 		// On 386 only, more complicated things like PUSHL $sym become $MOV sym@GOT, CX; PUSHL CX
 		cmplxdest := false
 		pAs := p.As
@@ -397,8 +399,9 @@ func rewriteToUseGot(ctxt *obj.Link, p *obj.Prog) {
 		q := p
 		if p.From.Offset != 0 {
 			q = obj.Appendp(ctxt, p)
-			q.As = add
-			q.From.Type = obj.TYPE_CONST
+			q.As = lea
+			q.From.Type = obj.TYPE_MEM
+			q.From.Reg = p.To.Reg
 			q.From.Offset = p.From.Offset
 			q.To = p.To
 			p.From.Offset = 0
@@ -1212,16 +1215,16 @@ loop:
 		q = p.Pcond
 		if q != nil && q.As != obj.ATEXT {
 			/* mark instruction as done and continue layout at target of jump */
-			p.Mark = 1
+			p.Mark |= DONE
 
 			p = q
-			if p.Mark == 0 {
+			if p.Mark&DONE == 0 {
 				goto loop
 			}
 		}
 	}
 
-	if p.Mark != 0 {
+	if p.Mark&DONE != 0 {
 		/*
 		 * p goes here, but already used it elsewhere.
 		 * copy up to 4 instructions or else branch to other copy.
@@ -1244,7 +1247,7 @@ loop:
 			if nofollow(a) || pushpop(a) {
 				break // NOTE(rsc): arm does goto copy
 			}
-			if q.Pcond == nil || q.Pcond.Mark != 0 {
+			if q.Pcond == nil || q.Pcond.Mark&DONE != 0 {
 				continue
 			}
 			if a == obj.ACALL || a == ALOOP {
@@ -1258,10 +1261,10 @@ loop:
 
 				q = obj.Copyp(ctxt, p)
 				p = p.Link
-				q.Mark = 1
+				q.Mark |= DONE
 				(*last).Link = q
 				*last = q
-				if int(q.As) != a || q.Pcond == nil || q.Pcond.Mark != 0 {
+				if int(q.As) != a || q.Pcond == nil || q.Pcond.Mark&DONE != 0 {
 					continue
 				}
 
@@ -1271,7 +1274,7 @@ loop:
 				q.Link = p
 				xfol(ctxt, q.Link, last)
 				p = q.Link
-				if p.Mark != 0 {
+				if p.Mark&DONE != 0 {
 					return
 				}
 				goto loop
@@ -1288,7 +1291,7 @@ loop:
 	}
 
 	/* emit p */
-	p.Mark = 1
+	p.Mark |= DONE
 
 	(*last).Link = p
 	*last = p
@@ -1326,7 +1329,7 @@ loop:
 			}
 		} else {
 			q = p.Link
-			if q.Mark != 0 {
+			if q.Mark&DONE != 0 {
 				if a != ALOOP {
 					p.As = relinv(int16(a))
 					p.Link = p.Pcond
@@ -1336,7 +1339,7 @@ loop:
 		}
 
 		xfol(ctxt, p.Link, last)
-		if p.Pcond.Mark != 0 {
+		if p.Pcond.Mark&DONE != 0 {
 			return
 		}
 		p = p.Pcond
diff --git a/src/cmd/internal/obj/x86/obj6_test.go b/src/cmd/internal/obj/x86/obj6_test.go
index 5fa1d3bfcf..a5c80cea3b 100644
--- a/src/cmd/internal/obj/x86/obj6_test.go
+++ b/src/cmd/internal/obj/x86/obj6_test.go
@@ -20,9 +20,9 @@ const testdata = `
 MOVQ AX, AX -> MOVQ AX, AX
 
 LEAQ name(SB), AX -> MOVQ name@GOT(SB), AX
-LEAQ name+10(SB), AX -> MOVQ name@GOT(SB), AX; ADDQ $10, AX
+LEAQ name+10(SB), AX -> MOVQ name@GOT(SB), AX; LEAQ 10(AX), AX
 MOVQ $name(SB), AX -> MOVQ name@GOT(SB), AX
-MOVQ $name+10(SB), AX -> MOVQ name@GOT(SB), AX; ADDQ $10, AX
+MOVQ $name+10(SB), AX -> MOVQ name@GOT(SB), AX; LEAQ 10(AX), AX
 
 MOVQ name(SB), AX -> NOP; MOVQ name@GOT(SB), R15; MOVQ (R15), AX
 MOVQ name+10(SB), AX -> NOP; MOVQ name@GOT(SB), R15; MOVQ 10(R15), AX