From af18bce87cc7ee1ffc68f91abefa241ab209539e Mon Sep 17 00:00:00 2001 From: Cherry Zhang Date: Sun, 20 Sep 2020 23:29:20 -0400 Subject: cmd/link: consider interface conversions only in reachable code The linker prunes methods that are not directly reachable if the receiver type is never converted to interface. Currently, this "never" is too strong: it is invalidated even if the interface conversion is in an unreachable function. This CL improves it by only considering interface conversions in reachable code. To do that, we introduce a marker relocation R_USEIFACE, which marks the target symbol as UsedInIface if the source symbol is reached. binary size before after cmd/compile 18897528 18887400 cmd/go 13607372 13470652 Change-Id: I66c6b69eeff9ae02d84d2e6f2bc7f1b29dd53910 Reviewed-on: https://go-review.googlesource.com/c/go/+/256797 Trust: Cherry Zhang Reviewed-by: Jeremy Faller Reviewed-by: Than McIntosh --- src/cmd/internal/obj/s390x/asmz.go | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) (limited to 'src/cmd/internal/obj/s390x') diff --git a/src/cmd/internal/obj/s390x/asmz.go b/src/cmd/internal/obj/s390x/asmz.go index 68f01f1c5d..cb3a2c3196 100644 --- a/src/cmd/internal/obj/s390x/asmz.go +++ b/src/cmd/internal/obj/s390x/asmz.go @@ -461,6 +461,7 @@ func spanz(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { buffer := make([]byte, 0) changed := true loop := 0 + nrelocs0 := len(c.cursym.R) for changed { if loop > 100 { c.ctxt.Diag("stuck in spanz loop") @@ -468,7 +469,10 @@ func spanz(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { } changed = false buffer = buffer[:0] - c.cursym.R = make([]obj.Reloc, 0) + for i := range c.cursym.R[nrelocs0:] { + c.cursym.R[nrelocs0+i] = obj.Reloc{} + } + c.cursym.R = c.cursym.R[:nrelocs0] // preserve marker relocations generated by the compiler for p := c.cursym.Func.Text; p != nil; p = p.Link { pc := int64(len(buffer)) if pc != p.Pc { -- cgit v1.3 From 912262b806a432a29302e0cee45e4f42ef7038a2 Mon Sep 17 00:00:00 2001 From: Russ Cox Date: Sun, 19 Jul 2020 00:30:12 -0400 Subject: cmd/internal/obj: move LSym.Func into LSym.Extra This creates space for a different kind of extension field in LSym without making the struct any larger. (There are many LSym, so we care about keeping the struct small.) Change-Id: Ib16edb9e15f54c2a7351c8b875e19684058711e5 Reviewed-on: https://go-review.googlesource.com/c/go/+/243943 Trust: Russ Cox Run-TryBot: Russ Cox TryBot-Result: Go Bot Reviewed-by: Cherry Zhang --- src/cmd/asm/internal/asm/asm.go | 2 +- src/cmd/compile/internal/gc/dwinl.go | 4 +- src/cmd/compile/internal/gc/gsubr.go | 2 +- src/cmd/compile/internal/gc/obj.go | 9 ++-- src/cmd/compile/internal/gc/pgen.go | 18 ++++---- src/cmd/compile/internal/gc/plive.go | 9 ++-- src/cmd/compile/internal/gc/scope.go | 4 +- src/cmd/compile/internal/gc/ssa.go | 10 ++-- src/cmd/internal/obj/arm/asm5.go | 8 ++-- src/cmd/internal/obj/arm/obj5.go | 34 +++++++------- src/cmd/internal/obj/arm64/asm7.go | 12 ++--- src/cmd/internal/obj/arm64/obj7.go | 36 +++++++-------- src/cmd/internal/obj/dwarf.go | 31 +++++++------ src/cmd/internal/obj/ld.go | 2 +- src/cmd/internal/obj/link.go | 23 ++++++++- src/cmd/internal/obj/mips/asm0.go | 10 ++-- src/cmd/internal/obj/mips/obj0.go | 42 ++++++++--------- src/cmd/internal/obj/objfile.go | 90 ++++++++++++++++++------------------ src/cmd/internal/obj/pass.go | 6 +-- src/cmd/internal/obj/pcln.go | 28 +++++------ src/cmd/internal/obj/plist.go | 14 +++--- src/cmd/internal/obj/ppc64/asm9.go | 10 ++-- src/cmd/internal/obj/ppc64/obj9.go | 28 +++++------ src/cmd/internal/obj/riscv/obj.go | 38 +++++++-------- src/cmd/internal/obj/s390x/asmz.go | 6 +-- src/cmd/internal/obj/s390x/objz.go | 28 +++++------ src/cmd/internal/obj/sym.go | 6 ++- src/cmd/internal/obj/wasm/wasmobj.go | 34 +++++++------- src/cmd/internal/obj/x86/asm6.go | 8 ++-- src/cmd/internal/obj/x86/obj6.go | 26 +++++------ 30 files changed, 304 insertions(+), 274 deletions(-) (limited to 'src/cmd/internal/obj/s390x') diff --git a/src/cmd/asm/internal/asm/asm.go b/src/cmd/asm/internal/asm/asm.go index 7878d74549..b9efa454ed 100644 --- a/src/cmd/asm/internal/asm/asm.go +++ b/src/cmd/asm/internal/asm/asm.go @@ -181,7 +181,7 @@ func (p *Parser) asmText(operands [][]lex.Token) { // Argsize set below. }, } - nameAddr.Sym.Func.Text = prog + nameAddr.Sym.Func().Text = prog prog.To.Val = int32(argSize) p.append(prog, "", true) } diff --git a/src/cmd/compile/internal/gc/dwinl.go b/src/cmd/compile/internal/gc/dwinl.go index 27e2cbcd98..5120fa1166 100644 --- a/src/cmd/compile/internal/gc/dwinl.go +++ b/src/cmd/compile/internal/gc/dwinl.go @@ -34,7 +34,7 @@ func assembleInlines(fnsym *obj.LSym, dwVars []*dwarf.Var) dwarf.InlCalls { // Walk progs to build up the InlCalls data structure var prevpos src.XPos - for p := fnsym.Func.Text; p != nil; p = p.Link { + for p := fnsym.Func().Text; p != nil; p = p.Link { if p.Pos == prevpos { continue } @@ -150,7 +150,7 @@ func assembleInlines(fnsym *obj.LSym, dwVars []*dwarf.Var) dwarf.InlCalls { start := int64(-1) curii := -1 var prevp *obj.Prog - for p := fnsym.Func.Text; p != nil; prevp, p = p, p.Link { + for p := fnsym.Func().Text; p != nil; prevp, p = p, p.Link { if prevp != nil && p.Pos == prevp.Pos { continue } diff --git a/src/cmd/compile/internal/gc/gsubr.go b/src/cmd/compile/internal/gc/gsubr.go index 480d411f49..14c217ff3b 100644 --- a/src/cmd/compile/internal/gc/gsubr.go +++ b/src/cmd/compile/internal/gc/gsubr.go @@ -199,7 +199,7 @@ func (pp *Progs) settext(fn *Node) { ptxt := pp.Prog(obj.ATEXT) pp.Text = ptxt - fn.Func.lsym.Func.Text = ptxt + fn.Func.lsym.Func().Text = ptxt ptxt.From.Type = obj.TYPE_MEM ptxt.From.Name = obj.NAME_EXTERN ptxt.From.Sym = fn.Func.lsym diff --git a/src/cmd/compile/internal/gc/obj.go b/src/cmd/compile/internal/gc/obj.go index d7f4a94041..f6557e2d15 100644 --- a/src/cmd/compile/internal/gc/obj.go +++ b/src/cmd/compile/internal/gc/obj.go @@ -305,20 +305,21 @@ func dumpglobls() { // global symbols can't be declared during parallel compilation. func addGCLocals() { for _, s := range Ctxt.Text { - if s.Func == nil { + fn := s.Func() + if fn == nil { continue } - for _, gcsym := range []*obj.LSym{s.Func.GCArgs, s.Func.GCLocals, s.Func.GCRegs} { + for _, gcsym := range []*obj.LSym{fn.GCArgs, fn.GCLocals, fn.GCRegs} { if gcsym != nil && !gcsym.OnList() { ggloblsym(gcsym, int32(len(gcsym.P)), obj.RODATA|obj.DUPOK) } } - if x := s.Func.StackObjects; x != nil { + if x := fn.StackObjects; x != nil { attr := int16(obj.RODATA) ggloblsym(x, int32(len(x.P)), attr) x.Set(obj.AttrStatic, true) } - if x := s.Func.OpenCodedDeferInfo; x != nil { + if x := fn.OpenCodedDeferInfo; x != nil { ggloblsym(x, int32(len(x.P)), obj.RODATA|obj.DUPOK) } } diff --git a/src/cmd/compile/internal/gc/pgen.go b/src/cmd/compile/internal/gc/pgen.go index 52b1ed351d..353f4b08c9 100644 --- a/src/cmd/compile/internal/gc/pgen.go +++ b/src/cmd/compile/internal/gc/pgen.go @@ -266,8 +266,8 @@ func compile(fn *Node) { dtypesym(n.Type) // Also make sure we allocate a linker symbol // for the stack object data, for the same reason. - if fn.Func.lsym.Func.StackObjects == nil { - fn.Func.lsym.Func.StackObjects = Ctxt.Lookup(fn.Func.lsym.Name + ".stkobj") + if fn.Func.lsym.Func().StackObjects == nil { + fn.Func.lsym.Func().StackObjects = Ctxt.Lookup(fn.Func.lsym.Name + ".stkobj") } } } @@ -415,7 +415,7 @@ func debuginfo(fnsym *obj.LSym, infosym *obj.LSym, curfn interface{}) ([]dwarf.S case PAUTO: if !n.Name.Used() { // Text == nil -> generating abstract function - if fnsym.Func.Text != nil { + if fnsym.Func().Text != nil { Fatalf("debuginfo unused node (AllocFrame should truncate fn.Func.Dcl)") } continue @@ -425,7 +425,7 @@ func debuginfo(fnsym *obj.LSym, infosym *obj.LSym, curfn interface{}) ([]dwarf.S continue } apdecls = append(apdecls, n) - fnsym.Func.RecordAutoType(ngotype(n).Linksym()) + fnsym.Func().RecordAutoType(ngotype(n).Linksym()) } decls, dwarfVars := createDwarfVars(fnsym, fn.Func, apdecls) @@ -435,7 +435,7 @@ func debuginfo(fnsym *obj.LSym, infosym *obj.LSym, curfn interface{}) ([]dwarf.S // the function symbol to insure that the type included in DWARF // processing during linking. typesyms := []*obj.LSym{} - for t, _ := range fnsym.Func.Autot { + for t, _ := range fnsym.Func().Autot { typesyms = append(typesyms, t) } sort.Sort(obj.BySymName(typesyms)) @@ -444,7 +444,7 @@ func debuginfo(fnsym *obj.LSym, infosym *obj.LSym, curfn interface{}) ([]dwarf.S r.Sym = sym r.Type = objabi.R_USETYPE } - fnsym.Func.Autot = nil + fnsym.Func().Autot = nil var varScopes []ScopeID for _, decl := range decls { @@ -522,7 +522,7 @@ func createSimpleVar(fnsym *obj.LSym, n *Node) *dwarf.Var { } typename := dwarf.InfoPrefix + typesymname(n.Type) - delete(fnsym.Func.Autot, ngotype(n).Linksym()) + delete(fnsym.Func().Autot, ngotype(n).Linksym()) inlIndex := 0 if genDwarfInline > 1 { if n.Name.InlFormal() || n.Name.InlLocal() { @@ -667,7 +667,7 @@ func createDwarfVars(fnsym *obj.LSym, fn *Func, apDecls []*Node) ([]*Node, []*dw ChildIndex: -1, }) // Record go type of to insure that it gets emitted by the linker. - fnsym.Func.RecordAutoType(ngotype(n).Linksym()) + fnsym.Func().RecordAutoType(ngotype(n).Linksym()) } return decls, vars @@ -731,7 +731,7 @@ func createComplexVar(fnsym *obj.LSym, fn *Func, varID ssa.VarID) *dwarf.Var { } gotype := ngotype(n).Linksym() - delete(fnsym.Func.Autot, gotype) + delete(fnsym.Func().Autot, gotype) typename := dwarf.InfoPrefix + gotype.Name[len("type."):] inlIndex := 0 if genDwarfInline > 1 { diff --git a/src/cmd/compile/internal/gc/plive.go b/src/cmd/compile/internal/gc/plive.go index a9ea37701e..b471accb65 100644 --- a/src/cmd/compile/internal/gc/plive.go +++ b/src/cmd/compile/internal/gc/plive.go @@ -1552,26 +1552,27 @@ func liveness(e *ssafn, f *ssa.Func, pp *Progs) LivenessMap { // Emit the live pointer map data structures ls := e.curfn.Func.lsym - ls.Func.GCArgs, ls.Func.GCLocals, ls.Func.GCRegs = lv.emit() + fninfo := ls.Func() + fninfo.GCArgs, fninfo.GCLocals, fninfo.GCRegs = lv.emit() p := pp.Prog(obj.AFUNCDATA) Addrconst(&p.From, objabi.FUNCDATA_ArgsPointerMaps) p.To.Type = obj.TYPE_MEM p.To.Name = obj.NAME_EXTERN - p.To.Sym = ls.Func.GCArgs + p.To.Sym = fninfo.GCArgs p = pp.Prog(obj.AFUNCDATA) Addrconst(&p.From, objabi.FUNCDATA_LocalsPointerMaps) p.To.Type = obj.TYPE_MEM p.To.Name = obj.NAME_EXTERN - p.To.Sym = ls.Func.GCLocals + p.To.Sym = fninfo.GCLocals if !go115ReduceLiveness { p = pp.Prog(obj.AFUNCDATA) Addrconst(&p.From, objabi.FUNCDATA_RegPointerMaps) p.To.Type = obj.TYPE_MEM p.To.Name = obj.NAME_EXTERN - p.To.Sym = ls.Func.GCRegs + p.To.Sym = fninfo.GCRegs } return lv.livenessMap diff --git a/src/cmd/compile/internal/gc/scope.go b/src/cmd/compile/internal/gc/scope.go index d7239d5693..e66b859e10 100644 --- a/src/cmd/compile/internal/gc/scope.go +++ b/src/cmd/compile/internal/gc/scope.go @@ -62,9 +62,9 @@ func scopePCs(fnsym *obj.LSym, marks []Mark, dwarfScopes []dwarf.Scope) { if len(marks) == 0 { return } - p0 := fnsym.Func.Text + p0 := fnsym.Func().Text scope := findScope(marks, p0.Pos) - for p := fnsym.Func.Text; p != nil; p = p.Link { + for p := p0; p != nil; p = p.Link { if p.Pos == p0.Pos { continue } diff --git a/src/cmd/compile/internal/gc/ssa.go b/src/cmd/compile/internal/gc/ssa.go index 3d5fa4cd0a..d8f627c213 100644 --- a/src/cmd/compile/internal/gc/ssa.go +++ b/src/cmd/compile/internal/gc/ssa.go @@ -240,7 +240,7 @@ func dvarint(x *obj.LSym, off int, v int64) int { // - Offset of where argument should be placed in the args frame when making call func (s *state) emitOpenDeferInfo() { x := Ctxt.Lookup(s.curfn.Func.lsym.Name + ".opendefer") - s.curfn.Func.lsym.Func.OpenCodedDeferInfo = x + s.curfn.Func.lsym.Func().OpenCodedDeferInfo = x off := 0 // Compute maxargsize (max size of arguments for all defers) @@ -6108,7 +6108,7 @@ func emitStackObjects(e *ssafn, pp *Progs) { // Populate the stack object data. // Format must match runtime/stack.go:stackObjectRecord. - x := e.curfn.Func.lsym.Func.StackObjects + x := e.curfn.Func.lsym.Func().StackObjects off := 0 off = duintptr(x, off, uint64(len(vars))) for _, v := range vars { @@ -6145,7 +6145,7 @@ func genssa(f *ssa.Func, pp *Progs) { s.livenessMap = liveness(e, f, pp) emitStackObjects(e, pp) - openDeferInfo := e.curfn.Func.lsym.Func.OpenCodedDeferInfo + openDeferInfo := e.curfn.Func.lsym.Func().OpenCodedDeferInfo if openDeferInfo != nil { // This function uses open-coded defers -- write out the funcdata // info that we computed at the end of genssa. @@ -6350,7 +6350,7 @@ func genssa(f *ssa.Func, pp *Progs) { // some of the inline marks. // Use this instruction instead. p.Pos = p.Pos.WithIsStmt() // promote position to a statement - pp.curfn.Func.lsym.Func.AddInlMark(p, inlMarks[m]) + pp.curfn.Func.lsym.Func().AddInlMark(p, inlMarks[m]) // Make the inline mark a real nop, so it doesn't generate any code. m.As = obj.ANOP m.Pos = src.NoXPos @@ -6362,7 +6362,7 @@ func genssa(f *ssa.Func, pp *Progs) { // Any unmatched inline marks now need to be added to the inlining tree (and will generate a nop instruction). for _, p := range inlMarkList { if p.As != obj.ANOP { - pp.curfn.Func.lsym.Func.AddInlMark(p, inlMarks[p]) + pp.curfn.Func.lsym.Func().AddInlMark(p, inlMarks[p]) } } } diff --git a/src/cmd/internal/obj/arm/asm5.go b/src/cmd/internal/obj/arm/asm5.go index 269a4223d5..ebb98b4859 100644 --- a/src/cmd/internal/obj/arm/asm5.go +++ b/src/cmd/internal/obj/arm/asm5.go @@ -390,7 +390,7 @@ func span5(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { var p *obj.Prog var op *obj.Prog - p = cursym.Func.Text + p = cursym.Func().Text if p == nil || p.Link == nil { // handle external functions and ELF section symbols return } @@ -482,8 +482,8 @@ func span5(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { bflag = 0 pc = 0 times++ - c.cursym.Func.Text.Pc = 0 // force re-layout the code. - for p = c.cursym.Func.Text; p != nil; p = p.Link { + c.cursym.Func().Text.Pc = 0 // force re-layout the code. + for p = c.cursym.Func().Text; p != nil; p = p.Link { o = c.oplook(p) if int64(pc) > p.Pc { p.Pc = int64(pc) @@ -558,7 +558,7 @@ func span5(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { * perhaps we'd be able to parallelize the span loop above. */ - p = c.cursym.Func.Text + p = c.cursym.Func().Text c.autosize = p.To.Offset + 4 c.cursym.Grow(c.cursym.Size) diff --git a/src/cmd/internal/obj/arm/obj5.go b/src/cmd/internal/obj/arm/obj5.go index 4d9187b530..f2bfb9679f 100644 --- a/src/cmd/internal/obj/arm/obj5.go +++ b/src/cmd/internal/obj/arm/obj5.go @@ -249,13 +249,13 @@ const ( func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { autosize := int32(0) - if cursym.Func.Text == nil || cursym.Func.Text.Link == nil { + if cursym.Func().Text == nil || cursym.Func().Text.Link == nil { return } c := ctxt5{ctxt: ctxt, cursym: cursym, newprog: newprog} - p := c.cursym.Func.Text + p := c.cursym.Func().Text autoffset := int32(p.To.Offset) if autoffset == -4 { // Historical way to mark NOFRAME. @@ -271,30 +271,30 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { } } - cursym.Func.Locals = autoffset - cursym.Func.Args = p.To.Val.(int32) + cursym.Func().Locals = autoffset + cursym.Func().Args = p.To.Val.(int32) /* * find leaf subroutines */ - for p := cursym.Func.Text; p != nil; p = p.Link { + for p := cursym.Func().Text; p != nil; p = p.Link { switch p.As { case obj.ATEXT: p.Mark |= LEAF case ADIV, ADIVU, AMOD, AMODU: - cursym.Func.Text.Mark &^= LEAF + cursym.Func().Text.Mark &^= LEAF case ABL, ABX, obj.ADUFFZERO, obj.ADUFFCOPY: - cursym.Func.Text.Mark &^= LEAF + cursym.Func().Text.Mark &^= LEAF } } var q2 *obj.Prog - for p := cursym.Func.Text; p != nil; p = p.Link { + for p := cursym.Func().Text; p != nil; p = p.Link { o := p.As switch o { case obj.ATEXT: @@ -311,20 +311,20 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { autosize += 4 } - if autosize == 0 && cursym.Func.Text.Mark&LEAF == 0 { + if autosize == 0 && cursym.Func().Text.Mark&LEAF == 0 { // A very few functions that do not return to their caller // are not identified as leaves but still have no frame. if ctxt.Debugvlog { ctxt.Logf("save suppressed in: %s\n", cursym.Name) } - cursym.Func.Text.Mark |= LEAF + cursym.Func().Text.Mark |= LEAF } // FP offsets need an updated p.To.Offset. p.To.Offset = int64(autosize) - 4 - if cursym.Func.Text.Mark&LEAF != 0 { + if cursym.Func().Text.Mark&LEAF != 0 { cursym.Set(obj.AttrLeaf, true) if p.From.Sym.NoFrame() { break @@ -347,7 +347,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { p.To.Reg = REGSP p.Spadj = autosize - if cursym.Func.Text.From.Sym.Wrapper() { + if cursym.Func().Text.From.Sym.Wrapper() { // if(g->panic != nil && g->panic->argp == FP) g->panic->argp = bottom-of-frame // // MOVW g_panic(g), R1 @@ -460,7 +460,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { case obj.ARET: nocache(p) - if cursym.Func.Text.Mark&LEAF != 0 { + if cursym.Func().Text.Mark&LEAF != 0 { if autosize == 0 { p.As = AB p.From = obj.Addr{} @@ -508,7 +508,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { } case ADIV, ADIVU, AMOD, AMODU: - if cursym.Func.Text.From.Sym.NoSplit() { + if cursym.Func().Text.From.Sym.NoSplit() { ctxt.Diag("cannot divide in NOSPLIT function") } const debugdivmod = false @@ -720,7 +720,7 @@ func (c *ctxt5) stacksplit(p *obj.Prog, framesize int32) *obj.Prog { end := c.ctxt.EndUnsafePoint(bls, c.newprog, -1) var last *obj.Prog - for last = c.cursym.Func.Text; last.Link != nil; last = last.Link { + for last = c.cursym.Func().Text; last.Link != nil; last = last.Link { } // Now we are at the end of the function, but logically @@ -751,7 +751,7 @@ func (c *ctxt5) stacksplit(p *obj.Prog, framesize int32) *obj.Prog { switch { case c.cursym.CFunc(): morestack = "runtime.morestackc" - case !c.cursym.Func.Text.From.Sym.NeedCtxt(): + case !c.cursym.Func().Text.From.Sym.NeedCtxt(): morestack = "runtime.morestack_noctxt" } call.To.Sym = c.ctxt.Lookup(morestack) @@ -762,7 +762,7 @@ func (c *ctxt5) stacksplit(p *obj.Prog, framesize int32) *obj.Prog { b := obj.Appendp(pcdata, c.newprog) b.As = obj.AJMP b.To.Type = obj.TYPE_BRANCH - b.To.SetTarget(c.cursym.Func.Text.Link) + b.To.SetTarget(c.cursym.Func().Text.Link) b.Spadj = +framesize return end diff --git a/src/cmd/internal/obj/arm64/asm7.go b/src/cmd/internal/obj/arm64/asm7.go index c46066313e..6b9fe27c05 100644 --- a/src/cmd/internal/obj/arm64/asm7.go +++ b/src/cmd/internal/obj/arm64/asm7.go @@ -913,7 +913,7 @@ func span7(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { ctxt.Retpoline = false // don't keep printing } - p := cursym.Func.Text + p := cursym.Func().Text if p == nil || p.Link == nil { // handle external functions and ELF section symbols return } @@ -943,8 +943,8 @@ func span7(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { alignedValue := p.From.Offset m = pcAlignPadLength(pc, alignedValue, ctxt) // Update the current text symbol alignment value. - if int32(alignedValue) > cursym.Func.Align { - cursym.Func.Align = int32(alignedValue) + if int32(alignedValue) > cursym.Func().Align { + cursym.Func().Align = int32(alignedValue) } break case obj.ANOP, obj.AFUNCDATA, obj.APCDATA: @@ -983,7 +983,7 @@ func span7(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { for bflag != 0 { bflag = 0 pc = 0 - for p = c.cursym.Func.Text.Link; p != nil; p = p.Link { + for p = c.cursym.Func().Text.Link; p != nil; p = p.Link { if p.As == ADWORD && (pc&7) != 0 { pc += 4 } @@ -1047,7 +1047,7 @@ func span7(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { psz := int32(0) var i int var out [6]uint32 - for p := c.cursym.Func.Text.Link; p != nil; p = p.Link { + for p := c.cursym.Func().Text.Link; p != nil; p = p.Link { c.pc = p.Pc o = c.oplook(p) @@ -1088,7 +1088,7 @@ func span7(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { // We use REGTMP as a scratch register during call injection, // so instruction sequences that use REGTMP are unsafe to // preempt asynchronously. - obj.MarkUnsafePoints(c.ctxt, c.cursym.Func.Text, c.newprog, c.isUnsafePoint, c.isRestartable) + obj.MarkUnsafePoints(c.ctxt, c.cursym.Func().Text, c.newprog, c.isUnsafePoint, c.isRestartable) } // isUnsafePoint returns whether p is an unsafe point. diff --git a/src/cmd/internal/obj/arm64/obj7.go b/src/cmd/internal/obj/arm64/obj7.go index f1bc2583cb..0baf51973a 100644 --- a/src/cmd/internal/obj/arm64/obj7.go +++ b/src/cmd/internal/obj/arm64/obj7.go @@ -166,7 +166,7 @@ func (c *ctxt7) stacksplit(p *obj.Prog, framesize int32) *obj.Prog { end := c.ctxt.EndUnsafePoint(bls, c.newprog, -1) var last *obj.Prog - for last = c.cursym.Func.Text; last.Link != nil; last = last.Link { + for last = c.cursym.Func().Text; last.Link != nil; last = last.Link { } // Now we are at the end of the function, but logically @@ -209,7 +209,7 @@ func (c *ctxt7) stacksplit(p *obj.Prog, framesize int32) *obj.Prog { switch { case c.cursym.CFunc(): morestack = "runtime.morestackc" - case !c.cursym.Func.Text.From.Sym.NeedCtxt(): + case !c.cursym.Func().Text.From.Sym.NeedCtxt(): morestack = "runtime.morestack_noctxt" } call.To.Sym = c.ctxt.Lookup(morestack) @@ -220,7 +220,7 @@ func (c *ctxt7) stacksplit(p *obj.Prog, framesize int32) *obj.Prog { jmp := obj.Appendp(pcdata, c.newprog) jmp.As = AB jmp.To.Type = obj.TYPE_BRANCH - jmp.To.SetTarget(c.cursym.Func.Text.Link) + jmp.To.SetTarget(c.cursym.Func().Text.Link) jmp.Spadj = +framesize return end @@ -441,13 +441,13 @@ func (c *ctxt7) rewriteToUseGot(p *obj.Prog) { } func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { - if cursym.Func.Text == nil || cursym.Func.Text.Link == nil { + if cursym.Func().Text == nil || cursym.Func().Text.Link == nil { return } c := ctxt7{ctxt: ctxt, newprog: newprog, cursym: cursym} - p := c.cursym.Func.Text + p := c.cursym.Func().Text textstksiz := p.To.Offset if textstksiz == -8 { // Historical way to mark NOFRAME. @@ -463,13 +463,13 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { } } - c.cursym.Func.Args = p.To.Val.(int32) - c.cursym.Func.Locals = int32(textstksiz) + c.cursym.Func().Args = p.To.Val.(int32) + c.cursym.Func().Locals = int32(textstksiz) /* * find leaf subroutines */ - for p := c.cursym.Func.Text; p != nil; p = p.Link { + for p := c.cursym.Func().Text; p != nil; p = p.Link { switch p.As { case obj.ATEXT: p.Mark |= LEAF @@ -477,18 +477,18 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { case ABL, obj.ADUFFZERO, obj.ADUFFCOPY: - c.cursym.Func.Text.Mark &^= LEAF + c.cursym.Func().Text.Mark &^= LEAF } } var q *obj.Prog var q1 *obj.Prog var retjmp *obj.LSym - for p := c.cursym.Func.Text; p != nil; p = p.Link { + for p := c.cursym.Func().Text; p != nil; p = p.Link { o := p.As switch o { case obj.ATEXT: - c.cursym.Func.Text = p + c.cursym.Func().Text = p c.autosize = int32(textstksiz) if p.Mark&LEAF != 0 && c.autosize == 0 { @@ -514,7 +514,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { c.ctxt.Diag("%v: unaligned frame size %d - must be 16 aligned", p, c.autosize-8) } c.autosize += extrasize - c.cursym.Func.Locals += extrasize + c.cursym.Func().Locals += extrasize // low 32 bits for autosize // high 32 bits for extrasize @@ -524,14 +524,14 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { p.To.Offset = 0 } - if c.autosize == 0 && c.cursym.Func.Text.Mark&LEAF == 0 { + if c.autosize == 0 && c.cursym.Func().Text.Mark&LEAF == 0 { if c.ctxt.Debugvlog { - c.ctxt.Logf("save suppressed in: %s\n", c.cursym.Func.Text.From.Sym.Name) + c.ctxt.Logf("save suppressed in: %s\n", c.cursym.Func().Text.From.Sym.Name) } - c.cursym.Func.Text.Mark |= LEAF + c.cursym.Func().Text.Mark |= LEAF } - if cursym.Func.Text.Mark&LEAF != 0 { + if cursym.Func().Text.Mark&LEAF != 0 { cursym.Set(obj.AttrLeaf, true) if p.From.Sym.NoFrame() { break @@ -641,7 +641,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { q1.To.Reg = REGFP } - if c.cursym.Func.Text.From.Sym.Wrapper() { + if c.cursym.Func().Text.From.Sym.Wrapper() { // if(g->panic != nil && g->panic->argp == FP) g->panic->argp = bottom-of-frame // // MOV g_panic(g), R1 @@ -755,7 +755,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { retjmp = p.To.Sym p.To = obj.Addr{} - if c.cursym.Func.Text.Mark&LEAF != 0 { + if c.cursym.Func().Text.Mark&LEAF != 0 { if c.autosize != 0 { p.As = AADD p.From.Type = obj.TYPE_CONST diff --git a/src/cmd/internal/obj/dwarf.go b/src/cmd/internal/obj/dwarf.go index 9abb31b558..328fb03b24 100644 --- a/src/cmd/internal/obj/dwarf.go +++ b/src/cmd/internal/obj/dwarf.go @@ -46,12 +46,12 @@ func (ctxt *Link) generateDebugLinesSymbol(s, lines *LSym) { // we expect at the start of a new sequence. stmt := true line := int64(1) - pc := s.Func.Text.Pc + pc := s.Func().Text.Pc var lastpc int64 // last PC written to line table, not last PC in func name := "" prologue, wrotePrologue := false, false // Walk the progs, generating the DWARF table. - for p := s.Func.Text; p != nil; p = p.Link { + for p := s.Func().Text; p != nil; p = p.Link { prologue = prologue || (p.Pos.Xlogue() == src.PosPrologueEnd) // If we're not at a real instruction, keep looping! if p.Pos.Line() == 0 || (p.Link != nil && p.Link.Pc == p.Pc) { @@ -103,7 +103,7 @@ func (ctxt *Link) generateDebugLinesSymbol(s, lines *LSym) { // text address before the end sequence op. If this isn't done, // GDB will assign a line number of zero the last row in the line // table, which we don't want. - lastlen := uint64(s.Size - (lastpc - s.Func.Text.Pc)) + lastlen := uint64(s.Size - (lastpc - s.Func().Text.Pc)) putpclcdelta(ctxt, dctxt, lines, lastlen, 0) dctxt.AddUint8(lines, 0) // start extended opcode dwarf.Uleb128put(dctxt, lines, 1) @@ -301,26 +301,27 @@ func (ctxt *Link) dwarfSym(s *LSym) (dwarfInfoSym, dwarfLocSym, dwarfRangesSym, if s.Type != objabi.STEXT { ctxt.Diag("dwarfSym of non-TEXT %v", s) } - if s.Func.dwarfInfoSym == nil { - s.Func.dwarfInfoSym = &LSym{ + fn := s.Func() + if fn.dwarfInfoSym == nil { + fn.dwarfInfoSym = &LSym{ Type: objabi.SDWARFFCN, } if ctxt.Flag_locationlists { - s.Func.dwarfLocSym = &LSym{ + fn.dwarfLocSym = &LSym{ Type: objabi.SDWARFLOC, } } - s.Func.dwarfRangesSym = &LSym{ + fn.dwarfRangesSym = &LSym{ Type: objabi.SDWARFRANGE, } - s.Func.dwarfDebugLinesSym = &LSym{ + fn.dwarfDebugLinesSym = &LSym{ Type: objabi.SDWARFLINES, } if s.WasInlined() { - s.Func.dwarfAbsFnSym = ctxt.DwFixups.AbsFuncDwarfSym(s) + fn.dwarfAbsFnSym = ctxt.DwFixups.AbsFuncDwarfSym(s) } } - return s.Func.dwarfInfoSym, s.Func.dwarfLocSym, s.Func.dwarfRangesSym, s.Func.dwarfAbsFnSym, s.Func.dwarfDebugLinesSym + return fn.dwarfInfoSym, fn.dwarfLocSym, fn.dwarfRangesSym, fn.dwarfAbsFnSym, fn.dwarfDebugLinesSym } func (s *LSym) Length(dwarfContext interface{}) int64 { @@ -331,7 +332,7 @@ func (s *LSym) Length(dwarfContext interface{}) int64 { // first instruction (prog) of the specified function. This will // presumably be the file in which the function is defined. func (ctxt *Link) fileSymbol(fn *LSym) *LSym { - p := fn.Func.Text + p := fn.Func().Text if p != nil { f, _ := linkgetlineFromPos(ctxt, p.Pos) fsym := ctxt.Lookup(f) @@ -405,8 +406,8 @@ func (ctxt *Link) DwarfAbstractFunc(curfn interface{}, s *LSym, myimportpath str if absfn.Size != 0 { ctxt.Diag("internal error: DwarfAbstractFunc double process %v", s) } - if s.Func == nil { - s.Func = new(FuncInfo) + if s.Func() == nil { + s.NewFuncInfo() } scopes, _ := ctxt.DebugInfo(s, absfn, curfn) dwctxt := dwCtxt{ctxt} @@ -527,8 +528,8 @@ func (ft *DwarfFixupTable) SetPrecursorFunc(s *LSym, fn interface{}) { // wrapper generation as opposed to the main inlining phase) it's // possible that we didn't cache the abstract function sym for the // text symbol -- do so now if needed. See issue 38068. - if s.Func != nil && s.Func.dwarfAbsFnSym == nil { - s.Func.dwarfAbsFnSym = absfn + if fn := s.Func(); fn != nil && fn.dwarfAbsFnSym == nil { + fn.dwarfAbsFnSym = absfn } ft.precursor[s] = fnState{precursor: fn, absfn: absfn} diff --git a/src/cmd/internal/obj/ld.go b/src/cmd/internal/obj/ld.go index 4ba52c7785..5d6c000dc6 100644 --- a/src/cmd/internal/obj/ld.go +++ b/src/cmd/internal/obj/ld.go @@ -59,7 +59,7 @@ func mkfwd(sym *LSym) { } i := 0 - for p := sym.Func.Text; p != nil && p.Link != nil; p = p.Link { + for p := sym.Func().Text; p != nil && p.Link != nil; p = p.Link { i-- if i < 0 { i = LOG - 1 diff --git a/src/cmd/internal/obj/link.go b/src/cmd/internal/obj/link.go index f14b691802..ae85dbbe4e 100644 --- a/src/cmd/internal/obj/link.go +++ b/src/cmd/internal/obj/link.go @@ -38,6 +38,7 @@ import ( "cmd/internal/src" "cmd/internal/sys" "fmt" + "log" "sync" ) @@ -400,7 +401,7 @@ type LSym struct { P []byte R []Reloc - Func *FuncInfo + Extra *interface{} // *FuncInfo if present Pkg string PkgIdx int32 @@ -433,6 +434,26 @@ type FuncInfo struct { FuncInfoSym *LSym } +// NewFuncInfo allocates and returns a FuncInfo for LSym. +func (s *LSym) NewFuncInfo() *FuncInfo { + if s.Extra != nil { + log.Fatalf("invalid use of LSym - NewFuncInfo with Extra of type %T", *s.Extra) + } + f := new(FuncInfo) + s.Extra = new(interface{}) + *s.Extra = f + return f +} + +// Func returns the *FuncInfo associated with s, or else nil. +func (s *LSym) Func() *FuncInfo { + if s.Extra == nil { + return nil + } + f, _ := (*s.Extra).(*FuncInfo) + return f +} + type InlMark struct { // When unwinding from an instruction in an inlined body, mark // where we should unwind to. diff --git a/src/cmd/internal/obj/mips/asm0.go b/src/cmd/internal/obj/mips/asm0.go index 6107974745..fd29f9fa21 100644 --- a/src/cmd/internal/obj/mips/asm0.go +++ b/src/cmd/internal/obj/mips/asm0.go @@ -410,7 +410,7 @@ func span0(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { ctxt.Retpoline = false // don't keep printing } - p := cursym.Func.Text + p := cursym.Func().Text if p == nil || p.Link == nil { // handle external functions and ELF section symbols return } @@ -455,7 +455,7 @@ func span0(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { for bflag != 0 { bflag = 0 pc = 0 - for p = c.cursym.Func.Text.Link; p != nil; p = p.Link { + for p = c.cursym.Func().Text.Link; p != nil; p = p.Link { p.Pc = pc o = c.oplook(p) @@ -512,7 +512,7 @@ func span0(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { bp := c.cursym.P var i int32 var out [4]uint32 - for p := c.cursym.Func.Text.Link; p != nil; p = p.Link { + for p := c.cursym.Func().Text.Link; p != nil; p = p.Link { c.pc = p.Pc o = c.oplook(p) if int(o.size) > 4*len(out) { @@ -529,7 +529,7 @@ func span0(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { // We use REGTMP as a scratch register during call injection, // so instruction sequences that use REGTMP are unsafe to // preempt asynchronously. - obj.MarkUnsafePoints(c.ctxt, c.cursym.Func.Text, c.newprog, c.isUnsafePoint, c.isRestartable) + obj.MarkUnsafePoints(c.ctxt, c.cursym.Func().Text, c.newprog, c.isUnsafePoint, c.isRestartable) } // isUnsafePoint returns whether p is an unsafe point. @@ -1302,7 +1302,7 @@ func (c *ctxt0) asmout(p *obj.Prog, o *Optab, out []uint32) { } o1 = OP_JMP(c.opirr(p.As), uint32(v)) if p.To.Sym == nil { - p.To.Sym = c.cursym.Func.Text.From.Sym + p.To.Sym = c.cursym.Func().Text.From.Sym p.To.Offset = p.To.Target().Pc } rel := obj.Addrel(c.cursym) diff --git a/src/cmd/internal/obj/mips/obj0.go b/src/cmd/internal/obj/mips/obj0.go index f19facc00c..135a8df3aa 100644 --- a/src/cmd/internal/obj/mips/obj0.go +++ b/src/cmd/internal/obj/mips/obj0.go @@ -133,11 +133,11 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { // a switch for enabling/disabling instruction scheduling nosched := true - if c.cursym.Func.Text == nil || c.cursym.Func.Text.Link == nil { + if c.cursym.Func().Text == nil || c.cursym.Func().Text.Link == nil { return } - p := c.cursym.Func.Text + p := c.cursym.Func().Text textstksiz := p.To.Offset if textstksiz == -ctxt.FixedFrameSize() { // Historical way to mark NOFRAME. @@ -153,8 +153,8 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { } } - c.cursym.Func.Args = p.To.Val.(int32) - c.cursym.Func.Locals = int32(textstksiz) + c.cursym.Func().Args = p.To.Val.(int32) + c.cursym.Func().Locals = int32(textstksiz) /* * find leaf subroutines @@ -162,7 +162,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { * expand BECOME pseudo */ - for p := c.cursym.Func.Text; p != nil; p = p.Link { + for p := c.cursym.Func().Text; p != nil; p = p.Link { switch p.As { /* too hard, just leave alone */ case obj.ATEXT: @@ -203,7 +203,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { AJAL, obj.ADUFFZERO, obj.ADUFFCOPY: - c.cursym.Func.Text.Mark &^= LEAF + c.cursym.Func().Text.Mark &^= LEAF fallthrough case AJMP, @@ -267,7 +267,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { autosize := int32(0) var p1 *obj.Prog var p2 *obj.Prog - for p := c.cursym.Func.Text; p != nil; p = p.Link { + for p := c.cursym.Func().Text; p != nil; p = p.Link { o := p.As switch o { case obj.ATEXT: @@ -288,19 +288,19 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { autosize += 4 } - if autosize == 0 && c.cursym.Func.Text.Mark&LEAF == 0 { - if c.cursym.Func.Text.From.Sym.NoSplit() { + if autosize == 0 && c.cursym.Func().Text.Mark&LEAF == 0 { + if c.cursym.Func().Text.From.Sym.NoSplit() { if ctxt.Debugvlog { ctxt.Logf("save suppressed in: %s\n", c.cursym.Name) } - c.cursym.Func.Text.Mark |= LEAF + c.cursym.Func().Text.Mark |= LEAF } } p.To.Offset = int64(autosize) - ctxt.FixedFrameSize() - if c.cursym.Func.Text.Mark&LEAF != 0 { + if c.cursym.Func().Text.Mark&LEAF != 0 { c.cursym.Set(obj.AttrLeaf, true) if p.From.Sym.NoFrame() { break @@ -344,7 +344,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { q = c.ctxt.EndUnsafePoint(q, c.newprog, -1) } - if c.cursym.Func.Text.From.Sym.Wrapper() && c.cursym.Func.Text.Mark&LEAF == 0 { + if c.cursym.Func().Text.From.Sym.Wrapper() && c.cursym.Func().Text.Mark&LEAF == 0 { // if(g->panic != nil && g->panic->argp == FP) g->panic->argp = bottom-of-frame // // MOV g_panic(g), R1 @@ -438,7 +438,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { p.To.Name = obj.NAME_NONE // clear fields as we may modify p to other instruction p.To.Sym = nil - if c.cursym.Func.Text.Mark&LEAF != 0 { + if c.cursym.Func().Text.Mark&LEAF != 0 { if autosize == 0 { p.As = AJMP p.From = obj.Addr{} @@ -540,7 +540,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { if c.ctxt.Arch.Family == sys.MIPS { // rewrite MOVD into two MOVF in 32-bit mode to avoid unaligned memory access - for p = c.cursym.Func.Text; p != nil; p = p1 { + for p = c.cursym.Func().Text; p != nil; p = p1 { p1 = p.Link if p.As != AMOVD { @@ -580,7 +580,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { if nosched { // if we don't do instruction scheduling, simply add // NOP after each branch instruction. - for p = c.cursym.Func.Text; p != nil; p = p.Link { + for p = c.cursym.Func().Text; p != nil; p = p.Link { if p.Mark&BRANCH != 0 { c.addnop(p) } @@ -589,10 +589,10 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { } // instruction scheduling - q = nil // p - 1 - q1 = c.cursym.Func.Text // top of block - o := 0 // count of instructions - for p = c.cursym.Func.Text; p != nil; p = p1 { + q = nil // p - 1 + q1 = c.cursym.Func().Text // top of block + o := 0 // count of instructions + for p = c.cursym.Func().Text; p != nil; p = p1 { p1 = p.Link o++ if p.Mark&NOSCHED != 0 { @@ -791,7 +791,7 @@ func (c *ctxt0) stacksplit(p *obj.Prog, framesize int32) *obj.Prog { p.To.Type = obj.TYPE_BRANCH if c.cursym.CFunc() { p.To.Sym = c.ctxt.Lookup("runtime.morestackc") - } else if !c.cursym.Func.Text.From.Sym.NeedCtxt() { + } else if !c.cursym.Func().Text.From.Sym.NeedCtxt() { p.To.Sym = c.ctxt.Lookup("runtime.morestack_noctxt") } else { p.To.Sym = c.ctxt.Lookup("runtime.morestack") @@ -805,7 +805,7 @@ func (c *ctxt0) stacksplit(p *obj.Prog, framesize int32) *obj.Prog { p.As = AJMP p.To.Type = obj.TYPE_BRANCH - p.To.SetTarget(c.cursym.Func.Text.Link) + p.To.SetTarget(c.cursym.Func().Text.Link) p.Mark |= BRANCH // placeholder for q1's jump target diff --git a/src/cmd/internal/obj/objfile.go b/src/cmd/internal/obj/objfile.go index fa60c9ad6d..a08de891d3 100644 --- a/src/cmd/internal/obj/objfile.go +++ b/src/cmd/internal/obj/objfile.go @@ -189,8 +189,8 @@ func WriteObjFile(ctxt *Link, b *bio.Writer) { // object file, and the Pcln variables haven't been filled in. As such, we // need to check that Pcsp exists, and assume the other pcln variables exist // as well. Tests like test/fixedbugs/issue22200.go demonstrate this issue. - if s.Func != nil && s.Func.Pcln.Pcsp != nil { - pc := &s.Func.Pcln + if fn := s.Func(); fn != nil && fn.Pcln.Pcsp != nil { + pc := &fn.Pcln w.Bytes(pc.Pcsp.P) w.Bytes(pc.Pcfile.P) w.Bytes(pc.Pcline.P) @@ -303,8 +303,8 @@ func (w *writer) Sym(s *LSym) { name = filepath.ToSlash(name) } var align uint32 - if s.Func != nil { - align = uint32(s.Func.Align) + if fn := s.Func(); fn != nil { + align = uint32(fn.Align) } if s.ContentAddressable() { // We generally assume data symbols are natually aligned, @@ -470,38 +470,38 @@ func (w *writer) Aux(s *LSym) { if s.Gotype != nil { w.aux1(goobj.AuxGotype, s.Gotype) } - if s.Func != nil { - w.aux1(goobj.AuxFuncInfo, s.Func.FuncInfoSym) + if fn := s.Func(); fn != nil { + w.aux1(goobj.AuxFuncInfo, fn.FuncInfoSym) - for _, d := range s.Func.Pcln.Funcdata { + for _, d := range fn.Pcln.Funcdata { w.aux1(goobj.AuxFuncdata, d) } - if s.Func.dwarfInfoSym != nil && s.Func.dwarfInfoSym.Size != 0 { - w.aux1(goobj.AuxDwarfInfo, s.Func.dwarfInfoSym) + if fn.dwarfInfoSym != nil && fn.dwarfInfoSym.Size != 0 { + w.aux1(goobj.AuxDwarfInfo, fn.dwarfInfoSym) } - if s.Func.dwarfLocSym != nil && s.Func.dwarfLocSym.Size != 0 { - w.aux1(goobj.AuxDwarfLoc, s.Func.dwarfLocSym) + if fn.dwarfLocSym != nil && fn.dwarfLocSym.Size != 0 { + w.aux1(goobj.AuxDwarfLoc, fn.dwarfLocSym) } - if s.Func.dwarfRangesSym != nil && s.Func.dwarfRangesSym.Size != 0 { - w.aux1(goobj.AuxDwarfRanges, s.Func.dwarfRangesSym) + if fn.dwarfRangesSym != nil && fn.dwarfRangesSym.Size != 0 { + w.aux1(goobj.AuxDwarfRanges, fn.dwarfRangesSym) } - if s.Func.dwarfDebugLinesSym != nil && s.Func.dwarfDebugLinesSym.Size != 0 { - w.aux1(goobj.AuxDwarfLines, s.Func.dwarfDebugLinesSym) + if fn.dwarfDebugLinesSym != nil && fn.dwarfDebugLinesSym.Size != 0 { + w.aux1(goobj.AuxDwarfLines, fn.dwarfDebugLinesSym) } - if s.Func.Pcln.Pcsp != nil && s.Func.Pcln.Pcsp.Size != 0 { - w.aux1(goobj.AuxPcsp, s.Func.Pcln.Pcsp) + if fn.Pcln.Pcsp != nil && fn.Pcln.Pcsp.Size != 0 { + w.aux1(goobj.AuxPcsp, fn.Pcln.Pcsp) } - if s.Func.Pcln.Pcfile != nil && s.Func.Pcln.Pcfile.Size != 0 { - w.aux1(goobj.AuxPcfile, s.Func.Pcln.Pcfile) + if fn.Pcln.Pcfile != nil && fn.Pcln.Pcfile.Size != 0 { + w.aux1(goobj.AuxPcfile, fn.Pcln.Pcfile) } - if s.Func.Pcln.Pcline != nil && s.Func.Pcln.Pcline.Size != 0 { - w.aux1(goobj.AuxPcline, s.Func.Pcln.Pcline) + if fn.Pcln.Pcline != nil && fn.Pcln.Pcline.Size != 0 { + w.aux1(goobj.AuxPcline, fn.Pcln.Pcline) } - if s.Func.Pcln.Pcinline != nil && s.Func.Pcln.Pcinline.Size != 0 { - w.aux1(goobj.AuxPcinline, s.Func.Pcln.Pcinline) + if fn.Pcln.Pcinline != nil && fn.Pcln.Pcinline.Size != 0 { + w.aux1(goobj.AuxPcinline, fn.Pcln.Pcinline) } - for _, pcSym := range s.Func.Pcln.Pcdata { + for _, pcSym := range fn.Pcln.Pcdata { w.aux1(goobj.AuxPcdata, pcSym) } @@ -571,34 +571,34 @@ func nAuxSym(s *LSym) int { if s.Gotype != nil { n++ } - if s.Func != nil { + if fn := s.Func(); fn != nil { // FuncInfo is an aux symbol, each Funcdata is an aux symbol - n += 1 + len(s.Func.Pcln.Funcdata) - if s.Func.dwarfInfoSym != nil && s.Func.dwarfInfoSym.Size != 0 { + n += 1 + len(fn.Pcln.Funcdata) + if fn.dwarfInfoSym != nil && fn.dwarfInfoSym.Size != 0 { n++ } - if s.Func.dwarfLocSym != nil && s.Func.dwarfLocSym.Size != 0 { + if fn.dwarfLocSym != nil && fn.dwarfLocSym.Size != 0 { n++ } - if s.Func.dwarfRangesSym != nil && s.Func.dwarfRangesSym.Size != 0 { + if fn.dwarfRangesSym != nil && fn.dwarfRangesSym.Size != 0 { n++ } - if s.Func.dwarfDebugLinesSym != nil && s.Func.dwarfDebugLinesSym.Size != 0 { + if fn.dwarfDebugLinesSym != nil && fn.dwarfDebugLinesSym.Size != 0 { n++ } - if s.Func.Pcln.Pcsp != nil && s.Func.Pcln.Pcsp.Size != 0 { + if fn.Pcln.Pcsp != nil && fn.Pcln.Pcsp.Size != 0 { n++ } - if s.Func.Pcln.Pcfile != nil && s.Func.Pcln.Pcfile.Size != 0 { + if fn.Pcln.Pcfile != nil && fn.Pcln.Pcfile.Size != 0 { n++ } - if s.Func.Pcln.Pcline != nil && s.Func.Pcln.Pcline.Size != 0 { + if fn.Pcln.Pcline != nil && fn.Pcln.Pcline.Size != 0 { n++ } - if s.Func.Pcln.Pcinline != nil && s.Func.Pcln.Pcinline.Size != 0 { + if fn.Pcln.Pcinline != nil && fn.Pcln.Pcinline.Size != 0 { n++ } - n += len(s.Func.Pcln.Pcdata) + n += len(fn.Pcln.Pcdata) } return n } @@ -620,15 +620,16 @@ func genFuncInfoSyms(ctxt *Link) { var b bytes.Buffer symidx := int32(len(ctxt.defs)) for _, s := range ctxt.Text { - if s.Func == nil { + fn := s.Func() + if fn == nil { continue } o := goobj.FuncInfo{ - Args: uint32(s.Func.Args), - Locals: uint32(s.Func.Locals), - FuncID: objabi.FuncID(s.Func.FuncID), + Args: uint32(fn.Args), + Locals: uint32(fn.Locals), + FuncID: objabi.FuncID(fn.FuncID), } - pc := &s.Func.Pcln + pc := &fn.Pcln o.Pcsp = makeSymRef(preparePcSym(pc.Pcsp)) o.Pcfile = makeSymRef(preparePcSym(pc.Pcfile)) o.Pcline = makeSymRef(preparePcSym(pc.Pcline)) @@ -670,10 +671,10 @@ func genFuncInfoSyms(ctxt *Link) { isym.Set(AttrIndexed, true) symidx++ infosyms = append(infosyms, isym) - s.Func.FuncInfoSym = isym + fn.FuncInfoSym = isym b.Reset() - dwsyms := []*LSym{s.Func.dwarfRangesSym, s.Func.dwarfLocSym, s.Func.dwarfDebugLinesSym, s.Func.dwarfInfoSym} + dwsyms := []*LSym{fn.dwarfRangesSym, fn.dwarfLocSym, fn.dwarfDebugLinesSym, fn.dwarfInfoSym} for _, s := range dwsyms { if s == nil || s.Size == 0 { continue @@ -744,14 +745,15 @@ func (ctxt *Link) writeSymDebugNamed(s *LSym, name string) { } fmt.Fprintf(ctxt.Bso, "size=%d", s.Size) if s.Type == objabi.STEXT { - fmt.Fprintf(ctxt.Bso, " args=%#x locals=%#x funcid=%#x", uint64(s.Func.Args), uint64(s.Func.Locals), uint64(s.Func.FuncID)) + fn := s.Func() + fmt.Fprintf(ctxt.Bso, " args=%#x locals=%#x funcid=%#x", uint64(fn.Args), uint64(fn.Locals), uint64(fn.FuncID)) if s.Leaf() { fmt.Fprintf(ctxt.Bso, " leaf") } } fmt.Fprintf(ctxt.Bso, "\n") if s.Type == objabi.STEXT { - for p := s.Func.Text; p != nil; p = p.Link { + for p := s.Func().Text; p != nil; p = p.Link { fmt.Fprintf(ctxt.Bso, "\t%#04x ", uint(int(p.Pc))) if ctxt.Debugasm > 1 { io.WriteString(ctxt.Bso, p.String()) diff --git a/src/cmd/internal/obj/pass.go b/src/cmd/internal/obj/pass.go index 09d520b4e9..01657dd4f6 100644 --- a/src/cmd/internal/obj/pass.go +++ b/src/cmd/internal/obj/pass.go @@ -118,7 +118,7 @@ func checkaddr(ctxt *Link, p *Prog, a *Addr) { } func linkpatch(ctxt *Link, sym *LSym, newprog ProgAlloc) { - for p := sym.Func.Text; p != nil; p = p.Link { + for p := sym.Func().Text; p != nil; p = p.Link { checkaddr(ctxt, p, &p.From) if p.GetFrom3() != nil { checkaddr(ctxt, p, p.GetFrom3()) @@ -138,7 +138,7 @@ func linkpatch(ctxt *Link, sym *LSym, newprog ProgAlloc) { if p.To.Sym != nil { continue } - q := sym.Func.Text + q := sym.Func().Text for q != nil && p.To.Offset != q.Pc { if q.Forwd != nil && p.To.Offset >= q.Forwd.Pc { q = q.Forwd @@ -164,7 +164,7 @@ func linkpatch(ctxt *Link, sym *LSym, newprog ProgAlloc) { } // Collapse series of jumps to jumps. - for p := sym.Func.Text; p != nil; p = p.Link { + for p := sym.Func().Text; p != nil; p = p.Link { if p.To.Target() == nil { continue } diff --git a/src/cmd/internal/obj/pcln.go b/src/cmd/internal/obj/pcln.go index ce0d3714c0..67c4f9a62b 100644 --- a/src/cmd/internal/obj/pcln.go +++ b/src/cmd/internal/obj/pcln.go @@ -35,20 +35,21 @@ func funcpctab(ctxt *Link, func_ *LSym, desc string, valfunc func(*Link, *LSym, val := int32(-1) oldval := val - if func_.Func.Text == nil { + fn := func_.Func() + if fn.Text == nil { // Return the emtpy symbol we've built so far. return sym } - pc := func_.Func.Text.Pc + pc := fn.Text.Pc if dbg { - ctxt.Logf("%6x %6d %v\n", uint64(pc), val, func_.Func.Text) + ctxt.Logf("%6x %6d %v\n", uint64(pc), val, fn.Text) } buf := make([]byte, binary.MaxVarintLen32) started := false - for p := func_.Func.Text; p != nil; p = p.Link { + for p := fn.Text; p != nil; p = p.Link { // Update val. If it's not changing, keep going. val = valfunc(ctxt, func_, val, p, 0, arg) @@ -107,7 +108,7 @@ func funcpctab(ctxt *Link, func_ *LSym, desc string, valfunc func(*Link, *LSym, if started { if dbg { - ctxt.Logf("%6x done\n", uint64(func_.Func.Text.Pc+func_.Size)) + ctxt.Logf("%6x done\n", uint64(fn.Text.Pc+func_.Size)) } v := (func_.Size - pc) / int64(ctxt.Arch.MinLC) if v < 0 { @@ -257,12 +258,12 @@ func pctopcdata(ctxt *Link, sym *LSym, oldval int32, p *Prog, phase int32, arg i } func linkpcln(ctxt *Link, cursym *LSym) { - pcln := &cursym.Func.Pcln + pcln := &cursym.Func().Pcln pcln.UsedFiles = make(map[goobj.CUFileIndex]struct{}) npcdata := 0 nfuncdata := 0 - for p := cursym.Func.Text; p != nil; p = p.Link { + for p := cursym.Func().Text; p != nil; p = p.Link { // Find the highest ID of any used PCDATA table. This ignores PCDATA table // that consist entirely of "-1", since that's the assumed default value. // From.Offset is table ID @@ -288,11 +289,12 @@ func linkpcln(ctxt *Link, cursym *LSym) { // Check that all the Progs used as inline markers are still reachable. // See issue #40473. - inlMarkProgs := make(map[*Prog]struct{}, len(cursym.Func.InlMarks)) - for _, inlMark := range cursym.Func.InlMarks { + fn := cursym.Func() + inlMarkProgs := make(map[*Prog]struct{}, len(fn.InlMarks)) + for _, inlMark := range fn.InlMarks { inlMarkProgs[inlMark.p] = struct{}{} } - for p := cursym.Func.Text; p != nil; p = p.Link { + for p := fn.Text; p != nil; p = p.Link { if _, ok := inlMarkProgs[p]; ok { delete(inlMarkProgs, p) } @@ -303,7 +305,7 @@ func linkpcln(ctxt *Link, cursym *LSym) { pcinlineState := new(pcinlineState) pcln.Pcinline = funcpctab(ctxt, cursym, "pctoinline", pcinlineState.pctoinline, nil) - for _, inlMark := range cursym.Func.InlMarks { + for _, inlMark := range fn.InlMarks { pcinlineState.setParentPC(ctxt, int(inlMark.id), int32(inlMark.p.Pc)) } pcln.InlTree = pcinlineState.localTree @@ -316,7 +318,7 @@ func linkpcln(ctxt *Link, cursym *LSym) { // tabulate which pc and func data we have. havepc := make([]uint32, (npcdata+31)/32) havefunc := make([]uint32, (nfuncdata+31)/32) - for p := cursym.Func.Text; p != nil; p = p.Link { + for p := fn.Text; p != nil; p = p.Link { if p.As == AFUNCDATA { if (havefunc[p.From.Offset/32]>>uint64(p.From.Offset%32))&1 != 0 { ctxt.Diag("multiple definitions for FUNCDATA $%d", p.From.Offset) @@ -344,7 +346,7 @@ func linkpcln(ctxt *Link, cursym *LSym) { // funcdata if nfuncdata > 0 { - for p := cursym.Func.Text; p != nil; p = p.Link { + for p := fn.Text; p != nil; p = p.Link { if p.As != AFUNCDATA { continue } diff --git a/src/cmd/internal/obj/plist.go b/src/cmd/internal/obj/plist.go index 6e33f29959..eb54c67f6a 100644 --- a/src/cmd/internal/obj/plist.go +++ b/src/cmd/internal/obj/plist.go @@ -81,7 +81,7 @@ func Flushplist(ctxt *Link, plist *Plist, newprog ProgAlloc, myimportpath string continue } found := false - for p := s.Func.Text; p != nil; p = p.Link { + for p := s.Func().Text; p != nil; p = p.Link { if p.As == AFUNCDATA && p.From.Type == TYPE_CONST && p.From.Offset == objabi.FUNCDATA_ArgsPointerMaps { found = true break @@ -89,7 +89,7 @@ func Flushplist(ctxt *Link, plist *Plist, newprog ProgAlloc, myimportpath string } if !found { - p := Appendp(s.Func.Text, newprog) + p := Appendp(s.Func().Text, newprog) p.As = AFUNCDATA p.From.Type = TYPE_CONST p.From.Offset = objabi.FUNCDATA_ArgsPointerMaps @@ -120,15 +120,15 @@ func (ctxt *Link) InitTextSym(s *LSym, flag int) { // func _() { } return } - if s.Func != nil { + if s.Func() != nil { ctxt.Diag("InitTextSym double init for %s", s.Name) } - s.Func = new(FuncInfo) + s.NewFuncInfo() if s.OnList() { ctxt.Diag("symbol %s listed multiple times", s.Name) } name := strings.Replace(s.Name, "\"\"", ctxt.Pkgpath, -1) - s.Func.FuncID = objabi.GetFuncID(name, flag&WRAPPER != 0) + s.Func().FuncID = objabi.GetFuncID(name, flag&WRAPPER != 0) s.Set(AttrOnList, true) s.Set(AttrDuplicateOK, flag&DUPOK != 0) s.Set(AttrNoSplit, flag&NOSPLIT != 0) @@ -185,7 +185,7 @@ func (ctxt *Link) EmitEntryLiveness(s *LSym, p *Prog, newprog ProgAlloc) *Prog { // Similar to EmitEntryLiveness, but just emit stack map. func (ctxt *Link) EmitEntryStackMap(s *LSym, p *Prog, newprog ProgAlloc) *Prog { pcdata := Appendp(p, newprog) - pcdata.Pos = s.Func.Text.Pos + pcdata.Pos = s.Func().Text.Pos pcdata.As = APCDATA pcdata.From.Type = TYPE_CONST pcdata.From.Offset = objabi.PCDATA_StackMapIndex @@ -198,7 +198,7 @@ func (ctxt *Link) EmitEntryStackMap(s *LSym, p *Prog, newprog ProgAlloc) *Prog { // Similar to EmitEntryLiveness, but just emit register map. func (ctxt *Link) EmitEntryRegMap(s *LSym, p *Prog, newprog ProgAlloc) *Prog { pcdata := Appendp(p, newprog) - pcdata.Pos = s.Func.Text.Pos + pcdata.Pos = s.Func().Text.Pos pcdata.As = APCDATA pcdata.From.Type = TYPE_CONST pcdata.From.Offset = objabi.PCDATA_RegMapIndex diff --git a/src/cmd/internal/obj/ppc64/asm9.go b/src/cmd/internal/obj/ppc64/asm9.go index c2e8e9e9d0..dcabb3cd6a 100644 --- a/src/cmd/internal/obj/ppc64/asm9.go +++ b/src/cmd/internal/obj/ppc64/asm9.go @@ -663,8 +663,8 @@ func addpad(pc, a int64, ctxt *obj.Link, cursym *obj.LSym) int { // the function alignment is not changed which might // result in 16 byte alignment but that is still fine. // TODO: alignment on AIX - if ctxt.Headtype != objabi.Haix && cursym.Func.Align < 32 { - cursym.Func.Align = 32 + if ctxt.Headtype != objabi.Haix && cursym.Func().Align < 32 { + cursym.Func().Align = 32 } default: ctxt.Diag("Unexpected alignment: %d for PCALIGN directive\n", a) @@ -673,7 +673,7 @@ func addpad(pc, a int64, ctxt *obj.Link, cursym *obj.LSym) int { } func span9(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { - p := cursym.Func.Text + p := cursym.Func().Text if p == nil || p.Link == nil { // handle external functions and ELF section symbols return } @@ -722,7 +722,7 @@ func span9(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { for bflag != 0 { bflag = 0 pc = 0 - for p = c.cursym.Func.Text.Link; p != nil; p = p.Link { + for p = c.cursym.Func().Text.Link; p != nil; p = p.Link { p.Pc = pc o = c.oplook(p) @@ -784,7 +784,7 @@ func span9(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { bp := c.cursym.P var i int32 var out [6]uint32 - for p := c.cursym.Func.Text.Link; p != nil; p = p.Link { + for p := c.cursym.Func().Text.Link; p != nil; p = p.Link { c.pc = p.Pc o = c.oplook(p) if int(o.size) > 4*len(out) { diff --git a/src/cmd/internal/obj/ppc64/obj9.go b/src/cmd/internal/obj/ppc64/obj9.go index c012762a18..3ab19de602 100644 --- a/src/cmd/internal/obj/ppc64/obj9.go +++ b/src/cmd/internal/obj/ppc64/obj9.go @@ -402,13 +402,13 @@ func (c *ctxt9) rewriteToUseGot(p *obj.Prog) { func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { // TODO(minux): add morestack short-cuts with small fixed frame-size. - if cursym.Func.Text == nil || cursym.Func.Text.Link == nil { + if cursym.Func().Text == nil || cursym.Func().Text.Link == nil { return } c := ctxt9{ctxt: ctxt, cursym: cursym, newprog: newprog} - p := c.cursym.Func.Text + p := c.cursym.Func().Text textstksiz := p.To.Offset if textstksiz == -8 { // Compatibility hack. @@ -424,8 +424,8 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { } } - c.cursym.Func.Args = p.To.Val.(int32) - c.cursym.Func.Locals = int32(textstksiz) + c.cursym.Func().Args = p.To.Val.(int32) + c.cursym.Func().Locals = int32(textstksiz) /* * find leaf subroutines @@ -435,7 +435,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { var q *obj.Prog var q1 *obj.Prog - for p := c.cursym.Func.Text; p != nil; p = p.Link { + for p := c.cursym.Func().Text; p != nil; p = p.Link { switch p.As { /* too hard, just leave alone */ case obj.ATEXT: @@ -541,7 +541,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { ABCL, obj.ADUFFZERO, obj.ADUFFCOPY: - c.cursym.Func.Text.Mark &^= LEAF + c.cursym.Func().Text.Mark &^= LEAF fallthrough case ABC, @@ -598,7 +598,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { autosize := int32(0) var p1 *obj.Prog var p2 *obj.Prog - for p := c.cursym.Func.Text; p != nil; p = p.Link { + for p := c.cursym.Func().Text; p != nil; p = p.Link { o := p.As switch o { case obj.ATEXT: @@ -664,7 +664,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { rel.Type = objabi.R_ADDRPOWER_PCREL } - if !c.cursym.Func.Text.From.Sym.NoSplit() { + if !c.cursym.Func().Text.From.Sym.NoSplit() { q = c.stacksplit(q, autosize) // emit split check } @@ -732,14 +732,14 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { q = c.ctxt.EndUnsafePoint(q, c.newprog, -1) } - } else if c.cursym.Func.Text.Mark&LEAF == 0 { + } else if c.cursym.Func().Text.Mark&LEAF == 0 { // A very few functions that do not return to their caller // (e.g. gogo) are not identified as leaves but still have // no frame. - c.cursym.Func.Text.Mark |= LEAF + c.cursym.Func().Text.Mark |= LEAF } - if c.cursym.Func.Text.Mark&LEAF != 0 { + if c.cursym.Func().Text.Mark&LEAF != 0 { c.cursym.Set(obj.AttrLeaf, true) break } @@ -755,7 +755,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { q.To.Offset = 24 } - if c.cursym.Func.Text.From.Sym.Wrapper() { + if c.cursym.Func().Text.From.Sym.Wrapper() { // if(g->panic != nil && g->panic->argp == FP) g->panic->argp = bottom-of-frame // // MOVD g_panic(g), R3 @@ -853,7 +853,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { retTarget := p.To.Sym - if c.cursym.Func.Text.Mark&LEAF != 0 { + if c.cursym.Func().Text.Mark&LEAF != 0 { if autosize == 0 || c.cursym.Name == "runtime.racecallbackthunk" { p.As = ABR p.From = obj.Addr{} @@ -1161,7 +1161,7 @@ func (c *ctxt9) stacksplit(p *obj.Prog, framesize int32) *obj.Prog { var morestacksym *obj.LSym if c.cursym.CFunc() { morestacksym = c.ctxt.Lookup("runtime.morestackc") - } else if !c.cursym.Func.Text.From.Sym.NeedCtxt() { + } else if !c.cursym.Func().Text.From.Sym.NeedCtxt() { morestacksym = c.ctxt.Lookup("runtime.morestack_noctxt") } else { morestacksym = c.ctxt.Lookup("runtime.morestack") diff --git a/src/cmd/internal/obj/riscv/obj.go b/src/cmd/internal/obj/riscv/obj.go index 841b30d85c..045c2250b5 100644 --- a/src/cmd/internal/obj/riscv/obj.go +++ b/src/cmd/internal/obj/riscv/obj.go @@ -427,7 +427,7 @@ func InvertBranch(as obj.As) obj.As { // instruction. Must be called after progedit. func containsCall(sym *obj.LSym) bool { // CALLs are CALL or JAL(R) with link register LR. - for p := sym.Func.Text; p != nil; p = p.Link { + for p := sym.Func().Text; p != nil; p = p.Link { switch p.As { case obj.ACALL: return true @@ -499,12 +499,12 @@ func stackOffset(a *obj.Addr, stacksize int64) { // concrete, real RISC-V instructions or directive pseudo-ops like TEXT, // PCDATA, and FUNCDATA. func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { - if cursym.Func.Text == nil || cursym.Func.Text.Link == nil { + if cursym.Func().Text == nil || cursym.Func().Text.Link == nil { return } // Generate the prologue. - text := cursym.Func.Text + text := cursym.Func().Text if text.As != obj.ATEXT { ctxt.Diag("preprocess: found symbol that does not start with TEXT directive") return @@ -538,12 +538,12 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { stacksize += ctxt.FixedFrameSize() } - cursym.Func.Args = text.To.Val.(int32) - cursym.Func.Locals = int32(stacksize) + cursym.Func().Args = text.To.Val.(int32) + cursym.Func().Locals = int32(stacksize) prologue := text - if !cursym.Func.Text.From.Sym.NoSplit() { + if !cursym.Func().Text.From.Sym.NoSplit() { prologue = stacksplit(ctxt, prologue, cursym, newprog, stacksize) // emit split check } @@ -567,7 +567,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { prologue = ctxt.EndUnsafePoint(prologue, newprog, -1) } - if cursym.Func.Text.From.Sym.Wrapper() { + if cursym.Func().Text.From.Sym.Wrapper() { // if(g->panic != nil && g->panic->argp == FP) g->panic->argp = bottom-of-frame // // MOV g_panic(g), X11 @@ -647,13 +647,13 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { } // Update stack-based offsets. - for p := cursym.Func.Text; p != nil; p = p.Link { + for p := cursym.Func().Text; p != nil; p = p.Link { stackOffset(&p.From, stacksize) stackOffset(&p.To, stacksize) } // Additional instruction rewriting. - for p := cursym.Func.Text; p != nil; p = p.Link { + for p := cursym.Func().Text; p != nil; p = p.Link { switch p.As { case obj.AGETCALLERPC: if cursym.Leaf() { @@ -733,7 +733,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { // Rewrite MOV pseudo-instructions. This cannot be done in // progedit, as SP offsets need to be applied before we split // up some of the Addrs. - for p := cursym.Func.Text; p != nil; p = p.Link { + for p := cursym.Func().Text; p != nil; p = p.Link { switch p.As { case AMOV, AMOVB, AMOVH, AMOVW, AMOVBU, AMOVHU, AMOVWU, AMOVF, AMOVD: rewriteMOV(ctxt, newprog, p) @@ -741,7 +741,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { } // Split immediates larger than 12-bits. - for p := cursym.Func.Text; p != nil; p = p.Link { + for p := cursym.Func().Text; p != nil; p = p.Link { switch p.As { // $imm, REG, TO case AADDI, AANDI, AORI, AXORI: @@ -858,9 +858,9 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { // a fixed point will be reached). No attempt to handle functions > 2GiB. for { rescan := false - setPCs(cursym.Func.Text, 0) + setPCs(cursym.Func().Text, 0) - for p := cursym.Func.Text; p != nil; p = p.Link { + for p := cursym.Func().Text; p != nil; p = p.Link { switch p.As { case ABEQ, ABEQZ, ABGE, ABGEU, ABGEZ, ABGT, ABGTU, ABGTZ, ABLE, ABLEU, ABLEZ, ABLT, ABLTU, ABLTZ, ABNE, ABNEZ: if p.To.Type != obj.TYPE_BRANCH { @@ -917,7 +917,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { // Now that there are no long branches, resolve branch and jump targets. // At this point, instruction rewriting which changes the number of // instructions will break everything--don't do it! - for p := cursym.Func.Text; p != nil; p = p.Link { + for p := cursym.Func().Text; p != nil; p = p.Link { switch p.As { case ABEQ, ABEQZ, ABGE, ABGEU, ABGEZ, ABGT, ABGTU, ABGTZ, ABLE, ABLEU, ABLEZ, ABLT, ABLTU, ABLTZ, ABNE, ABNEZ, AJAL: switch p.To.Type { @@ -940,7 +940,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { } // Validate all instructions - this provides nice error messages. - for p := cursym.Func.Text; p != nil; p = p.Link { + for p := cursym.Func().Text; p != nil; p = p.Link { for _, ins := range instructionsForProg(p) { ins.validate(ctxt) } @@ -1068,7 +1068,7 @@ func stacksplit(ctxt *obj.Link, p *obj.Prog, cursym *obj.LSym, newprog obj.ProgA p.To.Type = obj.TYPE_BRANCH if cursym.CFunc() { p.To.Sym = ctxt.Lookup("runtime.morestackc") - } else if !cursym.Func.Text.From.Sym.NeedCtxt() { + } else if !cursym.Func().Text.From.Sym.NeedCtxt() { p.To.Sym = ctxt.Lookup("runtime.morestack_noctxt") } else { p.To.Sym = ctxt.Lookup("runtime.morestack") @@ -1083,7 +1083,7 @@ func stacksplit(ctxt *obj.Link, p *obj.Prog, cursym *obj.LSym, newprog obj.ProgA p.As = AJAL p.To = obj.Addr{Type: obj.TYPE_BRANCH} p.From = obj.Addr{Type: obj.TYPE_REG, Reg: REG_ZERO} - p.To.SetTarget(cursym.Func.Text.Link) + p.To.SetTarget(cursym.Func().Text.Link) // placeholder for to_done's jump target p = obj.Appendp(p, newprog) @@ -1926,7 +1926,7 @@ func assemble(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { } var symcode []uint32 - for p := cursym.Func.Text; p != nil; p = p.Link { + for p := cursym.Func().Text; p != nil; p = p.Link { switch p.As { case AJALR: if p.To.Sym != nil { @@ -1981,7 +1981,7 @@ func assemble(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { ctxt.Arch.ByteOrder.PutUint32(p, symcode[i]) } - obj.MarkUnsafePoints(ctxt, cursym.Func.Text, newprog, isUnsafePoint, nil) + obj.MarkUnsafePoints(ctxt, cursym.Func().Text, newprog, isUnsafePoint, nil) } func isUnsafePoint(p *obj.Prog) bool { diff --git a/src/cmd/internal/obj/s390x/asmz.go b/src/cmd/internal/obj/s390x/asmz.go index cb3a2c3196..da14dd3c41 100644 --- a/src/cmd/internal/obj/s390x/asmz.go +++ b/src/cmd/internal/obj/s390x/asmz.go @@ -447,7 +447,7 @@ func spanz(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { ctxt.Retpoline = false // don't keep printing } - p := cursym.Func.Text + p := cursym.Func().Text if p == nil || p.Link == nil { // handle external functions and ELF section symbols return } @@ -473,7 +473,7 @@ func spanz(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { c.cursym.R[nrelocs0+i] = obj.Reloc{} } c.cursym.R = c.cursym.R[:nrelocs0] // preserve marker relocations generated by the compiler - for p := c.cursym.Func.Text; p != nil; p = p.Link { + for p := c.cursym.Func().Text; p != nil; p = p.Link { pc := int64(len(buffer)) if pc != p.Pc { changed = true @@ -504,7 +504,7 @@ func spanz(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { // We use REGTMP as a scratch register during call injection, // so instruction sequences that use REGTMP are unsafe to // preempt asynchronously. - obj.MarkUnsafePoints(c.ctxt, c.cursym.Func.Text, c.newprog, c.isUnsafePoint, nil) + obj.MarkUnsafePoints(c.ctxt, c.cursym.Func().Text, c.newprog, c.isUnsafePoint, nil) } // Return whether p is an unsafe point. diff --git a/src/cmd/internal/obj/s390x/objz.go b/src/cmd/internal/obj/s390x/objz.go index 625bb0f7b4..3af5425b36 100644 --- a/src/cmd/internal/obj/s390x/objz.go +++ b/src/cmd/internal/obj/s390x/objz.go @@ -205,13 +205,13 @@ func (c *ctxtz) rewriteToUseGot(p *obj.Prog) { func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { // TODO(minux): add morestack short-cuts with small fixed frame-size. - if cursym.Func.Text == nil || cursym.Func.Text.Link == nil { + if cursym.Func().Text == nil || cursym.Func().Text.Link == nil { return } c := ctxtz{ctxt: ctxt, cursym: cursym, newprog: newprog} - p := c.cursym.Func.Text + p := c.cursym.Func().Text textstksiz := p.To.Offset if textstksiz == -8 { // Compatibility hack. @@ -227,8 +227,8 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { } } - c.cursym.Func.Args = p.To.Val.(int32) - c.cursym.Func.Locals = int32(textstksiz) + c.cursym.Func().Args = p.To.Val.(int32) + c.cursym.Func().Locals = int32(textstksiz) /* * find leaf subroutines @@ -237,7 +237,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { */ var q *obj.Prog - for p := c.cursym.Func.Text; p != nil; p = p.Link { + for p := c.cursym.Func().Text; p != nil; p = p.Link { switch p.As { case obj.ATEXT: q = p @@ -245,7 +245,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { case ABL, ABCL: q = p - c.cursym.Func.Text.Mark &^= LEAF + c.cursym.Func().Text.Mark &^= LEAF fallthrough case ABC, @@ -294,7 +294,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { var pPre *obj.Prog var pPreempt *obj.Prog wasSplit := false - for p := c.cursym.Func.Text; p != nil; p = p.Link { + for p := c.cursym.Func().Text; p != nil; p = p.Link { pLast = p switch p.As { case obj.ATEXT: @@ -356,19 +356,19 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { q.Spadj = autosize q = c.ctxt.EndUnsafePoint(q, c.newprog, -1) - } else if c.cursym.Func.Text.Mark&LEAF == 0 { + } else if c.cursym.Func().Text.Mark&LEAF == 0 { // A very few functions that do not return to their caller // (e.g. gogo) are not identified as leaves but still have // no frame. - c.cursym.Func.Text.Mark |= LEAF + c.cursym.Func().Text.Mark |= LEAF } - if c.cursym.Func.Text.Mark&LEAF != 0 { + if c.cursym.Func().Text.Mark&LEAF != 0 { c.cursym.Set(obj.AttrLeaf, true) break } - if c.cursym.Func.Text.From.Sym.Wrapper() { + if c.cursym.Func().Text.From.Sym.Wrapper() { // if(g->panic != nil && g->panic->argp == FP) g->panic->argp = bottom-of-frame // // MOVD g_panic(g), R3 @@ -461,7 +461,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { case obj.ARET: retTarget := p.To.Sym - if c.cursym.Func.Text.Mark&LEAF != 0 { + if c.cursym.Func().Text.Mark&LEAF != 0 { if autosize == 0 { p.As = ABR p.From = obj.Addr{} @@ -696,7 +696,7 @@ func (c *ctxtz) stacksplitPost(p *obj.Prog, pPre *obj.Prog, pPreempt *obj.Prog, p.To.Type = obj.TYPE_BRANCH if c.cursym.CFunc() { p.To.Sym = c.ctxt.Lookup("runtime.morestackc") - } else if !c.cursym.Func.Text.From.Sym.NeedCtxt() { + } else if !c.cursym.Func().Text.From.Sym.NeedCtxt() { p.To.Sym = c.ctxt.Lookup("runtime.morestack_noctxt") } else { p.To.Sym = c.ctxt.Lookup("runtime.morestack") @@ -709,7 +709,7 @@ func (c *ctxtz) stacksplitPost(p *obj.Prog, pPre *obj.Prog, pPreempt *obj.Prog, p.As = ABR p.To.Type = obj.TYPE_BRANCH - p.To.SetTarget(c.cursym.Func.Text.Link) + p.To.SetTarget(c.cursym.Func().Text.Link) return p } diff --git a/src/cmd/internal/obj/sym.go b/src/cmd/internal/obj/sym.go index e5d7b2cbfd..0182773f8e 100644 --- a/src/cmd/internal/obj/sym.go +++ b/src/cmd/internal/obj/sym.go @@ -358,7 +358,8 @@ func (ctxt *Link) traverseSyms(flag traverseFlag, fn func(*LSym)) { } func (ctxt *Link) traverseFuncAux(flag traverseFlag, fsym *LSym, fn func(parent *LSym, aux *LSym)) { - pc := &fsym.Func.Pcln + fninfo := fsym.Func() + pc := &fninfo.Pcln if flag&traverseAux == 0 { // NB: should it become necessary to walk aux sym reloc references // without walking the aux syms themselves, this can be changed. @@ -389,7 +390,8 @@ func (ctxt *Link) traverseFuncAux(flag traverseFlag, fsym *LSym, fn func(parent fn(fsym, filesym) } } - dwsyms := []*LSym{fsym.Func.dwarfRangesSym, fsym.Func.dwarfLocSym, fsym.Func.dwarfDebugLinesSym, fsym.Func.dwarfInfoSym} + + dwsyms := []*LSym{fninfo.dwarfRangesSym, fninfo.dwarfLocSym, fninfo.dwarfDebugLinesSym, fninfo.dwarfInfoSym} for _, dws := range dwsyms { if dws == nil || dws.Size == 0 { continue diff --git a/src/cmd/internal/obj/wasm/wasmobj.go b/src/cmd/internal/obj/wasm/wasmobj.go index a9e093a8ad..f7f66a1255 100644 --- a/src/cmd/internal/obj/wasm/wasmobj.go +++ b/src/cmd/internal/obj/wasm/wasmobj.go @@ -182,14 +182,14 @@ func preprocess(ctxt *obj.Link, s *obj.LSym, newprog obj.ProgAlloc) { return p } - framesize := s.Func.Text.To.Offset + framesize := s.Func().Text.To.Offset if framesize < 0 { panic("bad framesize") } - s.Func.Args = s.Func.Text.To.Val.(int32) - s.Func.Locals = int32(framesize) + s.Func().Args = s.Func().Text.To.Val.(int32) + s.Func().Locals = int32(framesize) - if s.Func.Text.From.Sym.Wrapper() { + if s.Func().Text.From.Sym.Wrapper() { // if g._panic != nil && g._panic.argp == FP { // g._panic.argp = bottom-of-frame // } @@ -222,7 +222,7 @@ func preprocess(ctxt *obj.Link, s *obj.LSym, newprog obj.ProgAlloc) { Offset: 0, // panic.argp } - p := s.Func.Text + p := s.Func().Text p = appendp(p, AMOVD, gpanic, regAddr(REG_R0)) p = appendp(p, AGet, regAddr(REG_R0)) @@ -245,7 +245,7 @@ func preprocess(ctxt *obj.Link, s *obj.LSym, newprog obj.ProgAlloc) { } if framesize > 0 { - p := s.Func.Text + p := s.Func().Text p = appendp(p, AGet, regAddr(REG_SP)) p = appendp(p, AI32Const, constAddr(framesize)) p = appendp(p, AI32Sub) @@ -260,8 +260,8 @@ func preprocess(ctxt *obj.Link, s *obj.LSym, newprog obj.ProgAlloc) { pc := int64(0) // pc is only incremented when necessary, this avoids bloat of the BrTable instruction var tableIdxs []uint64 tablePC := int64(0) - base := ctxt.PosTable.Pos(s.Func.Text.Pos).Base() - for p := s.Func.Text; p != nil; p = p.Link { + base := ctxt.PosTable.Pos(s.Func().Text.Pos).Base() + for p := s.Func().Text; p != nil; p = p.Link { prevBase := base base = ctxt.PosTable.Pos(p.Pos).Base() switch p.As { @@ -313,8 +313,8 @@ func preprocess(ctxt *obj.Link, s *obj.LSym, newprog obj.ProgAlloc) { tableIdxs = append(tableIdxs, uint64(numResumePoints)) s.Size = pc + 1 - if !s.Func.Text.From.Sym.NoSplit() { - p := s.Func.Text + if !s.Func().Text.From.Sym.NoSplit() { + p := s.Func().Text if framesize <= objabi.StackSmall { // small stack: SP <= stackguard @@ -352,7 +352,7 @@ func preprocess(ctxt *obj.Link, s *obj.LSym, newprog obj.ProgAlloc) { p = appendp(p, AIf) p = appendp(p, obj.ACALL, constAddr(0)) - if s.Func.Text.From.Sym.NeedCtxt() { + if s.Func().Text.From.Sym.NeedCtxt() { p.To = obj.Addr{Type: obj.TYPE_MEM, Name: obj.NAME_EXTERN, Sym: morestack} } else { p.To = obj.Addr{Type: obj.TYPE_MEM, Name: obj.NAME_EXTERN, Sym: morestackNoCtxt} @@ -365,7 +365,7 @@ func preprocess(ctxt *obj.Link, s *obj.LSym, newprog obj.ProgAlloc) { var entryPointLoopBranches []*obj.Prog var unwindExitBranches []*obj.Prog currentDepth := 0 - for p := s.Func.Text; p != nil; p = p.Link { + for p := s.Func().Text; p != nil; p = p.Link { switch p.As { case ABlock, ALoop, AIf: currentDepth++ @@ -562,7 +562,7 @@ func preprocess(ctxt *obj.Link, s *obj.LSym, newprog obj.ProgAlloc) { } } - for p := s.Func.Text; p != nil; p = p.Link { + for p := s.Func().Text; p != nil; p = p.Link { switch p.From.Name { case obj.NAME_AUTO: p.From.Offset += int64(framesize) @@ -712,7 +712,7 @@ func preprocess(ctxt *obj.Link, s *obj.LSym, newprog obj.ProgAlloc) { } { - p := s.Func.Text + p := s.Func().Text if len(unwindExitBranches) > 0 { p = appendp(p, ABlock) // unwindExit, used to return 1 when unwinding the stack for _, b := range unwindExitBranches { @@ -749,7 +749,7 @@ func preprocess(ctxt *obj.Link, s *obj.LSym, newprog obj.ProgAlloc) { currentDepth = 0 blockDepths := make(map[*obj.Prog]int) - for p := s.Func.Text; p != nil; p = p.Link { + for p := s.Func().Text; p != nil; p = p.Link { switch p.As { case ABlock, ALoop, AIf: currentDepth++ @@ -850,7 +850,7 @@ func assemble(ctxt *obj.Link, s *obj.LSym, newprog obj.ProgAlloc) { hasLocalSP = true var regUsed [MAXREG - MINREG]bool - for p := s.Func.Text; p != nil; p = p.Link { + for p := s.Func().Text; p != nil; p = p.Link { if p.From.Reg != 0 { regUsed[p.From.Reg-MINREG] = true } @@ -896,7 +896,7 @@ func assemble(ctxt *obj.Link, s *obj.LSym, newprog obj.ProgAlloc) { updateLocalSP(w) } - for p := s.Func.Text; p != nil; p = p.Link { + for p := s.Func().Text; p != nil; p = p.Link { switch p.As { case AGet: if p.From.Type != obj.TYPE_REG { diff --git a/src/cmd/internal/obj/x86/asm6.go b/src/cmd/internal/obj/x86/asm6.go index 4940c79eaa..c412f4945d 100644 --- a/src/cmd/internal/obj/x86/asm6.go +++ b/src/cmd/internal/obj/x86/asm6.go @@ -2050,7 +2050,7 @@ func span6(ctxt *obj.Link, s *obj.LSym, newprog obj.ProgAlloc) { ctxt.Diag("x86 tables not initialized, call x86.instinit first") } - for p := s.Func.Text; p != nil; p = p.Link { + for p := s.Func().Text; p != nil; p = p.Link { if p.To.Type == obj.TYPE_BRANCH && p.To.Target() == nil { p.To.SetTarget(p) } @@ -2085,7 +2085,7 @@ func span6(ctxt *obj.Link, s *obj.LSym, newprog obj.ProgAlloc) { } var count int64 // rough count of number of instructions - for p := s.Func.Text; p != nil; p = p.Link { + for p := s.Func().Text; p != nil; p = p.Link { count++ p.Back = branchShort // use short branches first time through if q := p.To.Target(); q != nil && (q.Back&branchShort != 0) { @@ -2113,7 +2113,7 @@ func span6(ctxt *obj.Link, s *obj.LSym, newprog obj.ProgAlloc) { c = 0 var pPrev *obj.Prog nops = nops[:0] - for p := s.Func.Text; p != nil; p = p.Link { + for p := s.Func().Text; p != nil; p = p.Link { c0 := c c = pjc.padJump(ctxt, s, p, c) @@ -2227,7 +2227,7 @@ func span6(ctxt *obj.Link, s *obj.LSym, newprog obj.ProgAlloc) { // the first instruction.) return p.From.Index == REG_TLS } - obj.MarkUnsafePoints(ctxt, s.Func.Text, newprog, useTLS, nil) + obj.MarkUnsafePoints(ctxt, s.Func().Text, newprog, useTLS, nil) } } diff --git a/src/cmd/internal/obj/x86/obj6.go b/src/cmd/internal/obj/x86/obj6.go index 18a6afcd77..e11fa13f65 100644 --- a/src/cmd/internal/obj/x86/obj6.go +++ b/src/cmd/internal/obj/x86/obj6.go @@ -563,11 +563,11 @@ func rewriteToPcrel(ctxt *obj.Link, p *obj.Prog, newprog obj.ProgAlloc) { } func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { - if cursym.Func.Text == nil || cursym.Func.Text.Link == nil { + if cursym.Func().Text == nil || cursym.Func().Text.Link == nil { return } - p := cursym.Func.Text + p := cursym.Func().Text autoffset := int32(p.To.Offset) if autoffset < 0 { autoffset = 0 @@ -602,12 +602,12 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { } textarg := int64(p.To.Val.(int32)) - cursym.Func.Args = int32(textarg) - cursym.Func.Locals = int32(p.To.Offset) + cursym.Func().Args = int32(textarg) + cursym.Func().Locals = int32(p.To.Offset) // TODO(rsc): Remove. - if ctxt.Arch.Family == sys.I386 && cursym.Func.Locals < 0 { - cursym.Func.Locals = 0 + if ctxt.Arch.Family == sys.I386 && cursym.Func().Locals < 0 { + cursym.Func().Locals = 0 } // TODO(rsc): Remove 'ctxt.Arch.Family == sys.AMD64 &&'. @@ -642,7 +642,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { p = load_g_cx(ctxt, p, newprog) // load g into CX } - if !cursym.Func.Text.From.Sym.NoSplit() { + if !cursym.Func().Text.From.Sym.NoSplit() { p = stacksplit(ctxt, cursym, p, newprog, autoffset, int32(textarg)) // emit split check } @@ -690,7 +690,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { p.To.Reg = REG_BP } - if cursym.Func.Text.From.Sym.Wrapper() { + if cursym.Func().Text.From.Sym.Wrapper() { // if g._panic != nil && g._panic.argp == FP { // g._panic.argp = bottom-of-frame // } @@ -808,7 +808,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { } var deltasp int32 - for p = cursym.Func.Text; p != nil; p = p.Link { + for p = cursym.Func().Text; p != nil; p = p.Link { pcsize := ctxt.Arch.RegSize switch p.From.Name { case obj.NAME_AUTO: @@ -1103,7 +1103,7 @@ func stacksplit(ctxt *obj.Link, cursym *obj.LSym, p *obj.Prog, newprog obj.ProgA end := ctxt.EndUnsafePoint(jls, newprog, -1) var last *obj.Prog - for last = cursym.Func.Text; last.Link != nil; last = last.Link { + for last = cursym.Func().Text; last.Link != nil; last = last.Link { } // Now we are at the end of the function, but logically @@ -1117,7 +1117,7 @@ func stacksplit(ctxt *obj.Link, cursym *obj.LSym, p *obj.Prog, newprog obj.ProgA pcdata = ctxt.StartUnsafePoint(pcdata, newprog) call := obj.Appendp(pcdata, newprog) - call.Pos = cursym.Func.Text.Pos + call.Pos = cursym.Func().Text.Pos call.As = obj.ACALL call.To.Type = obj.TYPE_BRANCH call.To.Name = obj.NAME_EXTERN @@ -1125,7 +1125,7 @@ func stacksplit(ctxt *obj.Link, cursym *obj.LSym, p *obj.Prog, newprog obj.ProgA switch { case cursym.CFunc(): morestack = "runtime.morestackc" - case !cursym.Func.Text.From.Sym.NeedCtxt(): + case !cursym.Func().Text.From.Sym.NeedCtxt(): morestack = "runtime.morestack_noctxt" } call.To.Sym = ctxt.Lookup(morestack) @@ -1144,7 +1144,7 @@ func stacksplit(ctxt *obj.Link, cursym *obj.LSym, p *obj.Prog, newprog obj.ProgA jmp := obj.Appendp(pcdata, newprog) jmp.As = obj.AJMP jmp.To.Type = obj.TYPE_BRANCH - jmp.To.SetTarget(cursym.Func.Text.Link) + jmp.To.SetTarget(cursym.Func().Text.Link) jmp.Spadj = +framesize jls.To.SetTarget(call) -- cgit v1.3 From 751c37fd629e0ebb18c39a219fab2b224d016b70 Mon Sep 17 00:00:00 2001 From: Cherry Zhang Date: Wed, 21 Oct 2020 21:29:43 -0400 Subject: cmd/internal/obj/{arm,s390x}: make return jump print nicer When a function with non-zero frame size makes a return jump (RET target), it assembles to, conceptually, MOV (SP), LR ADD $framesize, SP JMP target We did not clear some fields in the first instruction's Prog.To, causing it printed like (on ARM) MOVW.P 4(R13), (R14)(R14)(REG) Clear the fields to make it print nicer. Change-Id: I180901aeea41f1ff287d7c6034a6d69005927744 Reviewed-on: https://go-review.googlesource.com/c/go/+/264343 Trust: Cherry Zhang Reviewed-by: Joel Sing --- src/cmd/internal/obj/arm/obj5.go | 2 ++ src/cmd/internal/obj/s390x/objz.go | 6 ++++-- 2 files changed, 6 insertions(+), 2 deletions(-) (limited to 'src/cmd/internal/obj/s390x') diff --git a/src/cmd/internal/obj/arm/obj5.go b/src/cmd/internal/obj/arm/obj5.go index f2bfb9679f..29d3a5867d 100644 --- a/src/cmd/internal/obj/arm/obj5.go +++ b/src/cmd/internal/obj/arm/obj5.go @@ -487,6 +487,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { // If there are instructions following // this ARET, they come from a branch // with the same stackframe, so no spadj. + if p.To.Sym != nil { // retjmp p.To.Reg = REGLINK q2 = obj.Appendp(p, newprog) @@ -494,6 +495,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { q2.To.Type = obj.TYPE_BRANCH q2.To.Sym = p.To.Sym p.To.Sym = nil + p.To.Name = obj.NAME_NONE p = q2 } diff --git a/src/cmd/internal/obj/s390x/objz.go b/src/cmd/internal/obj/s390x/objz.go index 3af5425b36..970cf827d6 100644 --- a/src/cmd/internal/obj/s390x/objz.go +++ b/src/cmd/internal/obj/s390x/objz.go @@ -497,8 +497,10 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) { p.From.Type = obj.TYPE_MEM p.From.Reg = REGSP p.From.Offset = 0 - p.To.Type = obj.TYPE_REG - p.To.Reg = REG_LR + p.To = obj.Addr{ + Type: obj.TYPE_REG, + Reg: REG_LR, + } q = p -- cgit v1.3 From 15131caeaa83e57fd8bdf87dde2801443f5602db Mon Sep 17 00:00:00 2001 From: fanzha02 Date: Fri, 8 May 2020 10:51:29 +0800 Subject: cmd/internal/obj/arm64: add CASx/CASPx instructions This patch adds support for CASx and CASPx atomic instructions. go syntax gnu syntax CASD Rs, (Rn|RSP), Rt => cas Xs, Xt, (Xn|SP) CASALW Rs, (Rn|RSP), Rt => casal Ws, Wt, (Xn|SP) CASPD (Rs, Rs+1), (Rn|RSP), (Rt, Rt+1) => casp Xs, Xs+1, Xt, Xt+1, (Xn|SP) CASPW (Rs, Rs+1), (Rn|RSP), (Rt, Rt+1) => casp Ws, Ws+1, Wt, Wt+1, (Xn|SP) This patch changes the type of prog.RestArgs from "[]Addr" to "[]struct{Addr, Pos}", Pos is a enum, indicating the position of the operand. This patch also adds test cases. Change-Id: Ib971cfda7890b7aa895d17bab22dea326c7fcaa4 Reviewed-on: https://go-review.googlesource.com/c/go/+/233277 Trust: fannie zhang Reviewed-by: Cherry Zhang --- src/cmd/asm/internal/arch/arm64.go | 13 ++- src/cmd/asm/internal/asm/asm.go | 18 +++- src/cmd/asm/internal/asm/testdata/arm64.s | 18 ++++ src/cmd/asm/internal/asm/testdata/arm64error.s | 4 + src/cmd/compile/internal/s390x/ssa.go | 12 +-- src/cmd/internal/obj/arm64/a.out.go | 14 +++ src/cmd/internal/obj/arm64/anames.go | 14 +++ src/cmd/internal/obj/arm64/asm7.go | 132 +++++++++++++++++++------ src/cmd/internal/obj/link.go | 84 +++++++++++----- src/cmd/internal/obj/riscv/obj.go | 8 +- src/cmd/internal/obj/s390x/asmz.go | 2 +- src/cmd/internal/obj/util.go | 15 ++- src/cmd/internal/obj/x86/asm6.go | 6 +- 13 files changed, 266 insertions(+), 74 deletions(-) (limited to 'src/cmd/internal/obj/s390x') diff --git a/src/cmd/asm/internal/arch/arm64.go b/src/cmd/asm/internal/arch/arm64.go index e643889aef..e557630ca6 100644 --- a/src/cmd/asm/internal/arch/arm64.go +++ b/src/cmd/asm/internal/arch/arm64.go @@ -75,7 +75,7 @@ func IsARM64STLXR(op obj.As) bool { arm64.ASTXP, arm64.ASTXPW, arm64.ASTLXP, arm64.ASTLXPW: return true } - // atomic instructions + // LDADDx/SWPx/CASx atomic instructions if arm64.IsAtomicInstruction(op) { return true } @@ -93,6 +93,17 @@ func IsARM64TBL(op obj.As) bool { return false } +// IsARM64CASP reports whether the op (as defined by an arm64.A* +// constant) is one of the CASP-like instructions, and its 2nd +// destination is a register pair that require special handling. +func IsARM64CASP(op obj.As) bool { + switch op { + case arm64.ACASPD, arm64.ACASPW: + return true + } + return false +} + // ARM64Suffix handles the special suffix for the ARM64. // It returns a boolean to indicate success; failure means // cond was unrecognized. diff --git a/src/cmd/asm/internal/asm/asm.go b/src/cmd/asm/internal/asm/asm.go index b9efa454ed..c4032759bb 100644 --- a/src/cmd/asm/internal/asm/asm.go +++ b/src/cmd/asm/internal/asm/asm.go @@ -637,6 +637,18 @@ func (p *Parser) asmInstruction(op obj.As, cond string, a []obj.Addr) { prog.From = a[0] prog.SetFrom3(a[1]) prog.To = a[2] + case arch.IsARM64CASP(op): + prog.From = a[0] + prog.To = a[1] + // both 1st operand and 3rd operand are (Rs, Rs+1) register pair. + // And the register pair must be contiguous. + if (a[0].Type != obj.TYPE_REGREG) || (a[2].Type != obj.TYPE_REGREG) { + p.errorf("invalid addressing modes for 1st or 3rd operand to %s instruction, must be register pair", op) + return + } + // For ARM64 CASP-like instructions, its 2nd destination operand is register pair(Rt, Rt+1) that can + // not fit into prog.RegTo2, so save it to the prog.RestArgs. + prog.SetTo2(a[2]) default: prog.From = a[0] prog.Reg = p.getRegister(prog, op, &a[1]) @@ -725,7 +737,7 @@ func (p *Parser) asmInstruction(op obj.As, cond string, a []obj.Addr) { } if p.arch.Family == sys.AMD64 { prog.From = a[0] - prog.RestArgs = []obj.Addr{a[1], a[2]} + prog.SetRestArgs([]obj.Addr{a[1], a[2]}) prog.To = a[3] break } @@ -808,13 +820,13 @@ func (p *Parser) asmInstruction(op obj.As, cond string, a []obj.Addr) { } if p.arch.Family == sys.AMD64 { prog.From = a[0] - prog.RestArgs = []obj.Addr{a[1], a[2], a[3]} + prog.SetRestArgs([]obj.Addr{a[1], a[2], a[3]}) prog.To = a[4] break } if p.arch.Family == sys.S390X { prog.From = a[0] - prog.RestArgs = []obj.Addr{a[1], a[2], a[3]} + prog.SetRestArgs([]obj.Addr{a[1], a[2], a[3]}) prog.To = a[4] break } diff --git a/src/cmd/asm/internal/asm/testdata/arm64.s b/src/cmd/asm/internal/asm/testdata/arm64.s index 7943990e16..5547cf634c 100644 --- a/src/cmd/asm/internal/asm/testdata/arm64.s +++ b/src/cmd/asm/internal/asm/testdata/arm64.s @@ -777,6 +777,24 @@ again: LDORLH R5, (RSP), R7 // e7336578 LDORLB R5, (R6), R7 // c7306538 LDORLB R5, (RSP), R7 // e7336538 + CASD R1, (R2), ZR // 5f7ca1c8 + CASW R1, (RSP), ZR // ff7fa188 + CASB ZR, (R5), R3 // a37cbf08 + CASH R3, (RSP), ZR // ff7fa348 + CASW R5, (R7), R6 // e67ca588 + CASLD ZR, (RSP), R8 // e8ffbfc8 + CASLW R9, (R10), ZR // 5ffda988 + CASAD R7, (R11), R15 // 6f7de7c8 + CASAW R10, (RSP), R19 // f37fea88 + CASALD R5, (R6), R7 // c7fce5c8 + CASALD R5, (RSP), R7 // e7ffe5c8 + CASALW R5, (R6), R7 // c7fce588 + CASALW R5, (RSP), R7 // e7ffe588 + CASALH ZR, (R5), R8 // a8fcff48 + CASALB R8, (R9), ZR // 3ffde808 + CASPD (R30, ZR), (RSP), (R8, R9) // e87f3e48 + CASPW (R6, R7), (R8), (R4, R5) // 047d2608 + CASPD (R2, R3), (R2), (R8, R9) // 487c2248 // RET RET diff --git a/src/cmd/asm/internal/asm/testdata/arm64error.s b/src/cmd/asm/internal/asm/testdata/arm64error.s index c3a617066a..99e4d62d25 100644 --- a/src/cmd/asm/internal/asm/testdata/arm64error.s +++ b/src/cmd/asm/internal/asm/testdata/arm64error.s @@ -357,4 +357,8 @@ TEXT errors(SB),$0 VUADDW2 V9.B8, V12.S4, V14.S4 // ERROR "operand mismatch" VSLI $64, V7.D2, V8.D2 // ERROR "shift out of range" VUSRA $0, V7.D2, V8.D2 // ERROR "shift out of range" + CASPD (R3, R4), (R2), (R8, R9) // ERROR "source register pair must start from even register" + CASPD (R2, R3), (R2), (R9, R10) // ERROR "destination register pair must start from even register" + CASPD (R2, R4), (R2), (R8, R9) // ERROR "source register pair must be contiguous" + CASPD (R2, R3), (R2), (R8, R10) // ERROR "destination register pair must be contiguous" RET diff --git a/src/cmd/compile/internal/s390x/ssa.go b/src/cmd/compile/internal/s390x/ssa.go index e23b31f385..84b9f491e4 100644 --- a/src/cmd/compile/internal/s390x/ssa.go +++ b/src/cmd/compile/internal/s390x/ssa.go @@ -182,11 +182,11 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) { i := v.Aux.(s390x.RotateParams) p := s.Prog(v.Op.Asm()) p.From = obj.Addr{Type: obj.TYPE_CONST, Offset: int64(i.Start)} - p.RestArgs = []obj.Addr{ + p.SetRestArgs([]obj.Addr{ {Type: obj.TYPE_CONST, Offset: int64(i.End)}, {Type: obj.TYPE_CONST, Offset: int64(i.Amount)}, {Type: obj.TYPE_REG, Reg: r2}, - } + }) p.To = obj.Addr{Type: obj.TYPE_REG, Reg: r1} case ssa.OpS390XADD, ssa.OpS390XADDW, ssa.OpS390XSUB, ssa.OpS390XSUBW, @@ -913,7 +913,7 @@ func ssaGenBlock(s *gc.SSAGenState, b, next *ssa.Block) { p.From.Type = obj.TYPE_CONST p.From.Offset = int64(s390x.NotEqual & s390x.NotUnordered) // unordered is not possible p.Reg = s390x.REG_R3 - p.RestArgs = []obj.Addr{{Type: obj.TYPE_CONST, Offset: 0}} + p.SetFrom3(obj.Addr{Type: obj.TYPE_CONST, Offset: 0}) if b.Succs[0].Block() != next { s.Br(s390x.ABR, b.Succs[0].Block()) } @@ -956,17 +956,17 @@ func ssaGenBlock(s *gc.SSAGenState, b, next *ssa.Block) { p.From.Type = obj.TYPE_CONST p.From.Offset = int64(mask & s390x.NotUnordered) // unordered is not possible p.Reg = b.Controls[0].Reg() - p.RestArgs = []obj.Addr{{Type: obj.TYPE_REG, Reg: b.Controls[1].Reg()}} + p.SetFrom3(obj.Addr{Type: obj.TYPE_REG, Reg: b.Controls[1].Reg()}) case ssa.BlockS390XCGIJ, ssa.BlockS390XCIJ: p.From.Type = obj.TYPE_CONST p.From.Offset = int64(mask & s390x.NotUnordered) // unordered is not possible p.Reg = b.Controls[0].Reg() - p.RestArgs = []obj.Addr{{Type: obj.TYPE_CONST, Offset: int64(int8(b.AuxInt))}} + p.SetFrom3(obj.Addr{Type: obj.TYPE_CONST, Offset: int64(int8(b.AuxInt))}) case ssa.BlockS390XCLGIJ, ssa.BlockS390XCLIJ: p.From.Type = obj.TYPE_CONST p.From.Offset = int64(mask & s390x.NotUnordered) // unordered is not possible p.Reg = b.Controls[0].Reg() - p.RestArgs = []obj.Addr{{Type: obj.TYPE_CONST, Offset: int64(uint8(b.AuxInt))}} + p.SetFrom3(obj.Addr{Type: obj.TYPE_CONST, Offset: int64(uint8(b.AuxInt))}) default: b.Fatalf("branch not implemented: %s", b.LongString()) } diff --git a/src/cmd/internal/obj/arm64/a.out.go b/src/cmd/internal/obj/arm64/a.out.go index 98504353e2..5844b71ca7 100644 --- a/src/cmd/internal/obj/arm64/a.out.go +++ b/src/cmd/internal/obj/arm64/a.out.go @@ -843,6 +843,20 @@ const ( ASWPLW ASWPLH ASWPLB + ACASD + ACASW + ACASH + ACASB + ACASAD + ACASAW + ACASLD + ACASLW + ACASALD + ACASALW + ACASALH + ACASALB + ACASPD + ACASPW ABEQ ABNE ABCS diff --git a/src/cmd/internal/obj/arm64/anames.go b/src/cmd/internal/obj/arm64/anames.go index 126eefd032..fb216f9a94 100644 --- a/src/cmd/internal/obj/arm64/anames.go +++ b/src/cmd/internal/obj/arm64/anames.go @@ -337,6 +337,20 @@ var Anames = []string{ "SWPLW", "SWPLH", "SWPLB", + "CASD", + "CASW", + "CASH", + "CASB", + "CASAD", + "CASAW", + "CASLD", + "CASLW", + "CASALD", + "CASALW", + "CASALH", + "CASALB", + "CASPD", + "CASPW", "BEQ", "BNE", "BCS", diff --git a/src/cmd/internal/obj/arm64/asm7.go b/src/cmd/internal/obj/arm64/asm7.go index 8cbf5e719f..4fc62d5c7f 100644 --- a/src/cmd/internal/obj/arm64/asm7.go +++ b/src/cmd/internal/obj/arm64/asm7.go @@ -80,12 +80,17 @@ type Optab struct { } func IsAtomicInstruction(as obj.As) bool { - _, ok := atomicInstructions[as] - return ok + if _, ok := atomicLDADD[as]; ok { + return true + } + if _, ok := atomicSWP[as]; ok { + return true + } + return false } // known field values of an instruction. -var atomicInstructions = map[obj.As]uint32{ +var atomicLDADD = map[obj.As]uint32{ ALDADDAD: 3<<30 | 0x1c5<<21 | 0x00<<10, ALDADDAW: 2<<30 | 0x1c5<<21 | 0x00<<10, ALDADDAH: 1<<30 | 0x1c5<<21 | 0x00<<10, @@ -150,22 +155,41 @@ var atomicInstructions = map[obj.As]uint32{ ALDORLW: 2<<30 | 0x1c3<<21 | 0x0c<<10, ALDORLH: 1<<30 | 0x1c3<<21 | 0x0c<<10, ALDORLB: 0<<30 | 0x1c3<<21 | 0x0c<<10, - ASWPAD: 3<<30 | 0x1c5<<21 | 0x20<<10, - ASWPAW: 2<<30 | 0x1c5<<21 | 0x20<<10, - ASWPAH: 1<<30 | 0x1c5<<21 | 0x20<<10, - ASWPAB: 0<<30 | 0x1c5<<21 | 0x20<<10, - ASWPALD: 3<<30 | 0x1c7<<21 | 0x20<<10, - ASWPALW: 2<<30 | 0x1c7<<21 | 0x20<<10, - ASWPALH: 1<<30 | 0x1c7<<21 | 0x20<<10, - ASWPALB: 0<<30 | 0x1c7<<21 | 0x20<<10, - ASWPD: 3<<30 | 0x1c1<<21 | 0x20<<10, - ASWPW: 2<<30 | 0x1c1<<21 | 0x20<<10, - ASWPH: 1<<30 | 0x1c1<<21 | 0x20<<10, - ASWPB: 0<<30 | 0x1c1<<21 | 0x20<<10, - ASWPLD: 3<<30 | 0x1c3<<21 | 0x20<<10, - ASWPLW: 2<<30 | 0x1c3<<21 | 0x20<<10, - ASWPLH: 1<<30 | 0x1c3<<21 | 0x20<<10, - ASWPLB: 0<<30 | 0x1c3<<21 | 0x20<<10, +} + +var atomicSWP = map[obj.As]uint32{ + ASWPAD: 3<<30 | 0x1c5<<21 | 0x20<<10, + ASWPAW: 2<<30 | 0x1c5<<21 | 0x20<<10, + ASWPAH: 1<<30 | 0x1c5<<21 | 0x20<<10, + ASWPAB: 0<<30 | 0x1c5<<21 | 0x20<<10, + ASWPALD: 3<<30 | 0x1c7<<21 | 0x20<<10, + ASWPALW: 2<<30 | 0x1c7<<21 | 0x20<<10, + ASWPALH: 1<<30 | 0x1c7<<21 | 0x20<<10, + ASWPALB: 0<<30 | 0x1c7<<21 | 0x20<<10, + ASWPD: 3<<30 | 0x1c1<<21 | 0x20<<10, + ASWPW: 2<<30 | 0x1c1<<21 | 0x20<<10, + ASWPH: 1<<30 | 0x1c1<<21 | 0x20<<10, + ASWPB: 0<<30 | 0x1c1<<21 | 0x20<<10, + ASWPLD: 3<<30 | 0x1c3<<21 | 0x20<<10, + ASWPLW: 2<<30 | 0x1c3<<21 | 0x20<<10, + ASWPLH: 1<<30 | 0x1c3<<21 | 0x20<<10, + ASWPLB: 0<<30 | 0x1c3<<21 | 0x20<<10, + ACASD: 3<<30 | 0x45<<21 | 0x1f<<10, + ACASW: 2<<30 | 0x45<<21 | 0x1f<<10, + ACASH: 1<<30 | 0x45<<21 | 0x1f<<10, + ACASB: 0<<30 | 0x45<<21 | 0x1f<<10, + ACASAD: 3<<30 | 0x47<<21 | 0x1f<<10, + ACASAW: 2<<30 | 0x47<<21 | 0x1f<<10, + ACASLD: 3<<30 | 0x45<<21 | 0x3f<<10, + ACASLW: 2<<30 | 0x45<<21 | 0x3f<<10, + ACASALD: 3<<30 | 0x47<<21 | 0x3f<<10, + ACASALW: 2<<30 | 0x47<<21 | 0x3f<<10, + ACASALH: 1<<30 | 0x47<<21 | 0x3f<<10, + ACASALB: 0<<30 | 0x47<<21 | 0x3f<<10, +} +var atomicCASP = map[obj.As]uint32{ + ACASPD: 1<<30 | 0x41<<21 | 0x1f<<10, + ACASPW: 0<<30 | 0x41<<21 | 0x1f<<10, } var oprange [ALAST & obj.AMask][]Optab @@ -794,8 +818,10 @@ var optab = []Optab{ {ASTPW, C_PAIR, C_NONE, C_NONE, C_LOREG, 77, 12, 0, LTO, C_XPOST}, {ASTPW, C_PAIR, C_NONE, C_NONE, C_ADDR, 87, 12, 0, 0, 0}, - {ASWPD, C_REG, C_NONE, C_NONE, C_ZOREG, 47, 4, 0, 0, 0}, // RegTo2=C_REG - {ASWPD, C_REG, C_NONE, C_NONE, C_ZAUTO, 47, 4, REGSP, 0, 0}, // RegTo2=C_REG + {ASWPD, C_REG, C_NONE, C_NONE, C_ZOREG, 47, 4, 0, 0, 0}, // RegTo2=C_REG + {ASWPD, C_REG, C_NONE, C_NONE, C_ZAUTO, 47, 4, REGSP, 0, 0}, // RegTo2=C_REG + {ACASPD, C_PAIR, C_NONE, C_NONE, C_ZOREG, 106, 4, 0, 0, 0}, // RegTo2=C_REGREG + {ACASPD, C_PAIR, C_NONE, C_NONE, C_ZAUTO, 106, 4, REGSP, 0, 0}, // RegTo2=C_REGREG {ALDAR, C_ZOREG, C_NONE, C_NONE, C_REG, 58, 4, 0, 0, 0}, {ALDXR, C_ZOREG, C_NONE, C_NONE, C_REG, 58, 4, 0, 0, 0}, {ALDAXR, C_ZOREG, C_NONE, C_NONE, C_REG, 58, 4, 0, 0, 0}, @@ -2011,7 +2037,7 @@ func (c *ctxt7) oplook(p *obj.Prog) *Optab { a1-- a3 := C_NONE + 1 - if p.GetFrom3() != nil { + if p.GetFrom3() != nil && p.RestArgs[0].Pos == 0 { a3 = int(p.GetFrom3().Class) if a3 == 0 { a3 = c.aclass(p.GetFrom3()) + 1 @@ -2496,10 +2522,18 @@ func buildop(ctxt *obj.Link) { oprangeset(AMOVZW, t) case ASWPD: - for i := range atomicInstructions { + for i := range atomicLDADD { + oprangeset(i, t) + } + for i := range atomicSWP { + if i == ASWPD { + continue + } oprangeset(i, t) } + case ACASPD: + oprangeset(ACASPW, t) case ABEQ: oprangeset(ABNE, t) oprangeset(ABCS, t) @@ -3994,17 +4028,27 @@ func (c *ctxt7) asmout(p *obj.Prog, o *Optab, out []uint32) { o1 |= uint32(p.From.Reg&31) << 5 o1 |= uint32(p.To.Reg & 31) - case 47: /* SWPx/LDADDx/LDANDx/LDEORx/LDORx Rs, (Rb), Rt */ + case 47: // SWPx/LDADDx/LDANDx/LDEORx/LDORx/CASx Rs, (Rb), Rt rs := p.From.Reg rt := p.RegTo2 rb := p.To.Reg - fields := atomicInstructions[p.As] - // rt can't be sp. rt can't be r31 when field A is 0, A bit is the 23rd bit. - if rt == REG_RSP || (rt == REGZERO && (fields&(1<<23) == 0)) { + // rt can't be sp. + if rt == REG_RSP { c.ctxt.Diag("illegal destination register: %v\n", p) } - o1 |= fields | uint32(rs&31)<<16 | uint32(rb&31)<<5 | uint32(rt&31) + if enc, ok := atomicLDADD[p.As]; ok { + // for LDADDx-like instructions, rt can't be r31 when field.enc A is 0, A bit is the 23rd bit. + if (rt == REGZERO) && (enc&(1<<23) == 0) { + c.ctxt.Diag("illegal destination register: %v\n", p) + } + o1 |= enc + } else if enc, ok := atomicSWP[p.As]; ok { + o1 |= enc + } else { + c.ctxt.Diag("invalid atomic instructions: %v\n", p) + } + o1 |= uint32(rs&31)<<16 | uint32(rb&31)<<5 | uint32(rt&31) case 48: /* ADD $C_ADDCON2, Rm, Rd */ // NOTE: this case does not use REGTMP. If it ever does, @@ -5351,6 +5395,38 @@ func (c *ctxt7) asmout(p *obj.Prog, o *Optab, out []uint32) { rt := int((p.To.Reg) & 31) r := int((p.Reg) & 31) o1 |= ((Q & 1) << 30) | ((size & 3) << 22) | (uint32(rf&31) << 16) | (uint32(r&31) << 5) | uint32(rt&31) + + case 106: // CASPx (Rs, Rs+1), (Rb), (Rt, Rt+1) + rs := p.From.Reg + rt := p.GetTo2().Reg + rb := p.To.Reg + rs1 := int16(p.From.Offset) + rt1 := int16(p.GetTo2().Offset) + + enc, ok := atomicCASP[p.As] + if !ok { + c.ctxt.Diag("invalid CASP-like atomic instructions: %v\n", p) + } + // for CASPx-like instructions, Rs<0> != 1 && Rt<0> != 1 + switch { + case rs&1 != 0: + c.ctxt.Diag("source register pair must start from even register: %v\n", p) + break + case rt&1 != 0: + c.ctxt.Diag("destination register pair must start from even register: %v\n", p) + break + case rs != rs1-1: + c.ctxt.Diag("source register pair must be contiguous: %v\n", p) + break + case rt != rt1-1: + c.ctxt.Diag("destination register pair must be contiguous: %v\n", p) + break + } + // rt can't be sp. + if rt == REG_RSP { + c.ctxt.Diag("illegal destination register: %v\n", p) + } + o1 |= enc | uint32(rs&31)<<16 | uint32(rb&31)<<5 | uint32(rt&31) } out[0] = o1 out[1] = o2 diff --git a/src/cmd/internal/obj/link.go b/src/cmd/internal/obj/link.go index 2037beca72..b578b6a09a 100644 --- a/src/cmd/internal/obj/link.go +++ b/src/cmd/internal/obj/link.go @@ -283,28 +283,41 @@ func (a *Addr) SetTarget(t *Prog) { // The other fields not yet mentioned are for use by the back ends and should // be left zeroed by creators of Prog lists. type Prog struct { - Ctxt *Link // linker context - Link *Prog // next Prog in linked list - From Addr // first source operand - RestArgs []Addr // can pack any operands that not fit into {Prog.From, Prog.To} - To Addr // destination operand (second is RegTo2 below) - Pool *Prog // constant pool entry, for arm,arm64 back ends - Forwd *Prog // for x86 back end - Rel *Prog // for x86, arm back ends - Pc int64 // for back ends or assembler: virtual or actual program counter, depending on phase - Pos src.XPos // source position of this instruction - Spadj int32 // effect of instruction on stack pointer (increment or decrement amount) - As As // assembler opcode - Reg int16 // 2nd source operand - RegTo2 int16 // 2nd destination operand - Mark uint16 // bitmask of arch-specific items - Optab uint16 // arch-specific opcode index - Scond uint8 // bits that describe instruction suffixes (e.g. ARM conditions) - Back uint8 // for x86 back end: backwards branch state - Ft uint8 // for x86 back end: type index of Prog.From - Tt uint8 // for x86 back end: type index of Prog.To - Isize uint8 // for x86 back end: size of the instruction in bytes -} + Ctxt *Link // linker context + Link *Prog // next Prog in linked list + From Addr // first source operand + RestArgs []AddrPos // can pack any operands that not fit into {Prog.From, Prog.To} + To Addr // destination operand (second is RegTo2 below) + Pool *Prog // constant pool entry, for arm,arm64 back ends + Forwd *Prog // for x86 back end + Rel *Prog // for x86, arm back ends + Pc int64 // for back ends or assembler: virtual or actual program counter, depending on phase + Pos src.XPos // source position of this instruction + Spadj int32 // effect of instruction on stack pointer (increment or decrement amount) + As As // assembler opcode + Reg int16 // 2nd source operand + RegTo2 int16 // 2nd destination operand + Mark uint16 // bitmask of arch-specific items + Optab uint16 // arch-specific opcode index + Scond uint8 // bits that describe instruction suffixes (e.g. ARM conditions) + Back uint8 // for x86 back end: backwards branch state + Ft uint8 // for x86 back end: type index of Prog.From + Tt uint8 // for x86 back end: type index of Prog.To + Isize uint8 // for x86 back end: size of the instruction in bytes +} + +// Pos indicates whether the oprand is the source or the destination. +type AddrPos struct { + Addr + Pos OperandPos +} + +type OperandPos int8 + +const ( + Source OperandPos = iota + Destination +) // From3Type returns p.GetFrom3().Type, or TYPE_NONE when // p.GetFrom3() returns nil. @@ -330,15 +343,36 @@ func (p *Prog) GetFrom3() *Addr { if p.RestArgs == nil { return nil } - return &p.RestArgs[0] + return &p.RestArgs[0].Addr } -// SetFrom3 assigns []Addr{a} to p.RestArgs. +// SetFrom3 assigns []Args{{a, 0}} to p.RestArgs. // In pair with Prog.GetFrom3 it can help in emulation of Prog.From3. // // Deprecated: for the same reasons as Prog.GetFrom3. func (p *Prog) SetFrom3(a Addr) { - p.RestArgs = []Addr{a} + p.RestArgs = []AddrPos{{a, Source}} +} + +// SetTo2 assings []Args{{a, 1}} to p.RestArgs when the second destination +// operand does not fit into prog.RegTo2. +func (p *Prog) SetTo2(a Addr) { + p.RestArgs = []AddrPos{{a, Destination}} +} + +// GetTo2 returns the second destination operand. +func (p *Prog) GetTo2() *Addr { + if p.RestArgs == nil { + return nil + } + return &p.RestArgs[0].Addr +} + +// SetRestArgs assigns more than one source operands to p.RestArgs. +func (p *Prog) SetRestArgs(args []Addr) { + for i := range args { + p.RestArgs = append(p.RestArgs, AddrPos{args[i], Source}) + } } // An As denotes an assembler opcode. diff --git a/src/cmd/internal/obj/riscv/obj.go b/src/cmd/internal/obj/riscv/obj.go index 5301e44002..0cffa54fa6 100644 --- a/src/cmd/internal/obj/riscv/obj.go +++ b/src/cmd/internal/obj/riscv/obj.go @@ -48,7 +48,7 @@ func jalrToSym(ctxt *obj.Link, p *obj.Prog, newprog obj.ProgAlloc, lr int16) *ob p.As = AAUIPC p.Mark |= NEED_PCREL_ITYPE_RELOC - p.RestArgs = []obj.Addr{obj.Addr{Type: obj.TYPE_CONST, Offset: to.Offset, Sym: to.Sym}} + p.SetFrom3(obj.Addr{Type: obj.TYPE_CONST, Offset: to.Offset, Sym: to.Sym}) p.From = obj.Addr{Type: obj.TYPE_CONST, Offset: 0} p.Reg = 0 p.To = obj.Addr{Type: obj.TYPE_REG, Reg: REG_TMP} @@ -234,7 +234,7 @@ func rewriteMOV(ctxt *obj.Link, newprog obj.ProgAlloc, p *obj.Prog) { p.As = AAUIPC p.Mark |= NEED_PCREL_ITYPE_RELOC - p.RestArgs = []obj.Addr{obj.Addr{Type: obj.TYPE_CONST, Offset: p.From.Offset, Sym: p.From.Sym}} + p.SetFrom3(obj.Addr{Type: obj.TYPE_CONST, Offset: p.From.Offset, Sym: p.From.Sym}) p.From = obj.Addr{Type: obj.TYPE_CONST, Offset: 0} p.Reg = 0 p.To = obj.Addr{Type: obj.TYPE_REG, Reg: to.Reg} @@ -275,7 +275,7 @@ func rewriteMOV(ctxt *obj.Link, newprog obj.ProgAlloc, p *obj.Prog) { p.As = AAUIPC p.Mark |= NEED_PCREL_STYPE_RELOC - p.RestArgs = []obj.Addr{obj.Addr{Type: obj.TYPE_CONST, Offset: p.To.Offset, Sym: p.To.Sym}} + p.SetFrom3(obj.Addr{Type: obj.TYPE_CONST, Offset: p.To.Offset, Sym: p.To.Sym}) p.From = obj.Addr{Type: obj.TYPE_CONST, Offset: 0} p.Reg = 0 p.To = obj.Addr{Type: obj.TYPE_REG, Reg: REG_TMP} @@ -340,7 +340,7 @@ func rewriteMOV(ctxt *obj.Link, newprog obj.ProgAlloc, p *obj.Prog) { p.As = AAUIPC p.Mark |= NEED_PCREL_ITYPE_RELOC - p.RestArgs = []obj.Addr{obj.Addr{Type: obj.TYPE_CONST, Offset: p.From.Offset, Sym: p.From.Sym}} + p.SetFrom3(obj.Addr{Type: obj.TYPE_CONST, Offset: p.From.Offset, Sym: p.From.Sym}) p.From = obj.Addr{Type: obj.TYPE_CONST, Offset: 0} p.Reg = 0 p.To = to diff --git a/src/cmd/internal/obj/s390x/asmz.go b/src/cmd/internal/obj/s390x/asmz.go index da14dd3c41..f0f9d5cefc 100644 --- a/src/cmd/internal/obj/s390x/asmz.go +++ b/src/cmd/internal/obj/s390x/asmz.go @@ -718,7 +718,7 @@ func (c *ctxtz) oplook(p *obj.Prog) *Optab { p.From.Class = int8(c.aclass(&p.From) + 1) p.To.Class = int8(c.aclass(&p.To) + 1) for i := range p.RestArgs { - p.RestArgs[i].Class = int8(c.aclass(&p.RestArgs[i]) + 1) + p.RestArgs[i].Addr.Class = int8(c.aclass(&p.RestArgs[i].Addr) + 1) } // Mirrors the argument list in Optab. diff --git a/src/cmd/internal/obj/util.go b/src/cmd/internal/obj/util.go index 21e28807a6..b9bacb7a22 100644 --- a/src/cmd/internal/obj/util.go +++ b/src/cmd/internal/obj/util.go @@ -175,9 +175,11 @@ func (p *Prog) WriteInstructionString(w io.Writer) { sep = ", " } for i := range p.RestArgs { - io.WriteString(w, sep) - WriteDconv(w, p, &p.RestArgs[i]) - sep = ", " + if p.RestArgs[i].Pos == Source { + io.WriteString(w, sep) + WriteDconv(w, p, &p.RestArgs[i].Addr) + sep = ", " + } } if p.As == ATEXT { @@ -198,6 +200,13 @@ func (p *Prog) WriteInstructionString(w io.Writer) { if p.RegTo2 != REG_NONE { fmt.Fprintf(w, "%s%v", sep, Rconv(int(p.RegTo2))) } + for i := range p.RestArgs { + if p.RestArgs[i].Pos == Destination { + io.WriteString(w, sep) + WriteDconv(w, p, &p.RestArgs[i].Addr) + sep = ", " + } + } } func (ctxt *Link) NewProg() *Prog { diff --git a/src/cmd/internal/obj/x86/asm6.go b/src/cmd/internal/obj/x86/asm6.go index c412f4945d..94aed44871 100644 --- a/src/cmd/internal/obj/x86/asm6.go +++ b/src/cmd/internal/obj/x86/asm6.go @@ -4270,7 +4270,7 @@ func (ab *AsmBuf) doasm(ctxt *obj.Link, cursym *obj.LSym, p *obj.Prog) { args = append(args, ft) } for i := range p.RestArgs { - args = append(args, oclass(ctxt, p, &p.RestArgs[i])*Ymax) + args = append(args, oclass(ctxt, p, &p.RestArgs[i].Addr)*Ymax) } if tt != Ynone*Ymax { args = append(args, tt) @@ -5438,10 +5438,10 @@ func (ab *AsmBuf) asmins(ctxt *obj.Link, cursym *obj.LSym, p *obj.Prog) { // unpackOps4 extracts 4 operands from p. func unpackOps4(p *obj.Prog) (arg0, arg1, arg2, dst *obj.Addr) { - return &p.From, &p.RestArgs[0], &p.RestArgs[1], &p.To + return &p.From, &p.RestArgs[0].Addr, &p.RestArgs[1].Addr, &p.To } // unpackOps5 extracts 5 operands from p. func unpackOps5(p *obj.Prog) (arg0, arg1, arg2, arg3, dst *obj.Addr) { - return &p.From, &p.RestArgs[0], &p.RestArgs[1], &p.RestArgs[2], &p.To + return &p.From, &p.RestArgs[0].Addr, &p.RestArgs[1].Addr, &p.RestArgs[2].Addr, &p.To } -- cgit v1.3 From 854e892ce17e2555c59fce5b92f64bc505ba5d8c Mon Sep 17 00:00:00 2001 From: Michael Munday Date: Mon, 11 May 2020 09:44:48 -0700 Subject: cmd/compile: optimize shift pairs and masks on s390x MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Optimize combinations of left and right shifts by a constant value into a 'rotate then insert selected bits [into zero]' instruction. Use the same instruction for contiguous masks since it has some benefits over 'and immediate' (not restricted to 32-bits, does not overwrite source register). To keep the complexity of this change under control I've only implemented 64 bit operations for now. There are a lot more optimizations that can be done with this instruction family. However, since their function overlaps with other instructions we need to be somewhat careful not to break existing optimization rules by creating optimization dead ends. This is particularly true of the load/store merging rules which contain lots of zero extensions and shifts. This CL does interfere with the store merging rules when an operand is shifted left before it is stored: binary.BigEndian.PutUint64(b, x << 1) This is unfortunate but it's not critical and somewhat complex so I plan to fix that in a follow up CL. file before after Δ % addr2line 4117446 4117282 -164 -0.004% api 4945184 4942752 -2432 -0.049% asm 4998079 4991891 -6188 -0.124% buildid 2685158 2684074 -1084 -0.040% cgo 4553732 4553394 -338 -0.007% compile 19294446 19245070 -49376 -0.256% cover 4897105 4891319 -5786 -0.118% dist 3544389 3542785 -1604 -0.045% doc 3926795 3927617 +822 +0.021% fix 3302958 3293868 -9090 -0.275% link 6546274 6543456 -2818 -0.043% nm 4102021 4100825 -1196 -0.029% objdump 4542431 4548483 +6052 +0.133% pack 2482465 2416389 -66076 -2.662% pprof 13366541 13363915 -2626 -0.020% test2json 2829007 2761515 -67492 -2.386% trace 10216164 10219684 +3520 +0.034% vet 6773956 6773572 -384 -0.006% total 107124151 106917891 -206260 -0.193% Change-Id: I7591cce41e06867ba10a745daae9333513062746 Reviewed-on: https://go-review.googlesource.com/c/go/+/233317 Run-TryBot: Michael Munday TryBot-Result: Go Bot Reviewed-by: Keith Randall Trust: Michael Munday --- src/cmd/compile/internal/s390x/ssa.go | 14 +- src/cmd/compile/internal/ssa/gen/S390X.rules | 162 ++++- src/cmd/compile/internal/ssa/gen/S390XOps.go | 20 +- src/cmd/compile/internal/ssa/opGen.go | 25 +- src/cmd/compile/internal/ssa/rewriteS390X.go | 844 +++++++++++++++++++-------- src/cmd/internal/obj/s390x/rotate.go | 82 ++- src/cmd/internal/obj/s390x/rotate_test.go | 122 ++++ test/codegen/bitfield.go | 18 +- test/codegen/bits.go | 12 + test/codegen/mathbits.go | 2 +- test/codegen/rotate.go | 6 +- test/codegen/shift.go | 33 +- 12 files changed, 1007 insertions(+), 333 deletions(-) create mode 100644 src/cmd/internal/obj/s390x/rotate_test.go (limited to 'src/cmd/internal/obj/s390x') diff --git a/src/cmd/compile/internal/s390x/ssa.go b/src/cmd/compile/internal/s390x/ssa.go index 84b9f491e4..8037357131 100644 --- a/src/cmd/compile/internal/s390x/ssa.go +++ b/src/cmd/compile/internal/s390x/ssa.go @@ -188,6 +188,18 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) { {Type: obj.TYPE_REG, Reg: r2}, }) p.To = obj.Addr{Type: obj.TYPE_REG, Reg: r1} + case ssa.OpS390XRISBGZ: + r1 := v.Reg() + r2 := v.Args[0].Reg() + i := v.Aux.(s390x.RotateParams) + p := s.Prog(v.Op.Asm()) + p.From = obj.Addr{Type: obj.TYPE_CONST, Offset: int64(i.Start)} + p.SetRestArgs([]obj.Addr{ + {Type: obj.TYPE_CONST, Offset: int64(i.End)}, + {Type: obj.TYPE_CONST, Offset: int64(i.Amount)}, + {Type: obj.TYPE_REG, Reg: r2}, + }) + p.To = obj.Addr{Type: obj.TYPE_REG, Reg: r1} case ssa.OpS390XADD, ssa.OpS390XADDW, ssa.OpS390XSUB, ssa.OpS390XSUBW, ssa.OpS390XAND, ssa.OpS390XANDW, @@ -360,7 +372,7 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) { case ssa.OpS390XSLDconst, ssa.OpS390XSLWconst, ssa.OpS390XSRDconst, ssa.OpS390XSRWconst, ssa.OpS390XSRADconst, ssa.OpS390XSRAWconst, - ssa.OpS390XRLLGconst, ssa.OpS390XRLLconst: + ssa.OpS390XRLLconst: p := s.Prog(v.Op.Asm()) p.From.Type = obj.TYPE_CONST p.From.Offset = v.AuxInt diff --git a/src/cmd/compile/internal/ssa/gen/S390X.rules b/src/cmd/compile/internal/ssa/gen/S390X.rules index 1b56361c00..39949edbc2 100644 --- a/src/cmd/compile/internal/ssa/gen/S390X.rules +++ b/src/cmd/compile/internal/ssa/gen/S390X.rules @@ -643,8 +643,18 @@ // equivalent to the leftmost 32 bits being set. // TODO(mundaym): modify the assembler to accept 64-bit values // and use isU32Bit(^c). -(AND x (MOVDconst [c])) && is32Bit(c) && c < 0 => (ANDconst [c] x) -(AND x (MOVDconst [c])) && is32Bit(c) && c >= 0 => (MOVWZreg (ANDWconst [int32(c)] x)) +(AND x (MOVDconst [c])) + && s390x.NewRotateParams(0, 63, 0).OutMerge(uint64(c)) != nil + => (RISBGZ x {*s390x.NewRotateParams(0, 63, 0).OutMerge(uint64(c))}) +(AND x (MOVDconst [c])) + && is32Bit(c) + && c < 0 + => (ANDconst [c] x) +(AND x (MOVDconst [c])) + && is32Bit(c) + && c >= 0 + => (MOVWZreg (ANDWconst [int32(c)] x)) + (ANDW x (MOVDconst [c])) => (ANDWconst [int32(c)] x) ((AND|ANDW)const [c] ((AND|ANDW)const [d] x)) => ((AND|ANDW)const [c&d] x) @@ -653,14 +663,20 @@ ((OR|XOR)W x (MOVDconst [c])) => ((OR|XOR)Wconst [int32(c)] x) // Constant shifts. -(S(LD|RD|RAD|LW|RW|RAW) x (MOVDconst [c])) - => (S(LD|RD|RAD|LW|RW|RAW)const x [int8(c&63)]) +(S(LD|RD|RAD) x (MOVDconst [c])) => (S(LD|RD|RAD)const x [int8(c&63)]) +(S(LW|RW|RAW) x (MOVDconst [c])) && c&32 == 0 => (S(LW|RW|RAW)const x [int8(c&31)]) +(S(LW|RW) _ (MOVDconst [c])) && c&32 != 0 => (MOVDconst [0]) +(SRAW x (MOVDconst [c])) && c&32 != 0 => (SRAWconst x [31]) // Shifts only use the rightmost 6 bits of the shift value. +(S(LD|RD|RAD|LW|RW|RAW) x (RISBGZ y {r})) + && r.Amount == 0 + && r.OutMask()&63 == 63 + => (S(LD|RD|RAD|LW|RW|RAW) x y) (S(LD|RD|RAD|LW|RW|RAW) x (AND (MOVDconst [c]) y)) - => (S(LD|RD|RAD|LW|RW|RAW) x (ANDWconst [int32(c&63)] y)) + => (S(LD|RD|RAD|LW|RW|RAW) x (ANDWconst [int32(c&63)] y)) (S(LD|RD|RAD|LW|RW|RAW) x (ANDWconst [c] y)) && c&63 == 63 - => (S(LD|RD|RAD|LW|RW|RAW) x y) + => (S(LD|RD|RAD|LW|RW|RAW) x y) (SLD x (MOV(W|H|B|WZ|HZ|BZ)reg y)) => (SLD x y) (SRD x (MOV(W|H|B|WZ|HZ|BZ)reg y)) => (SRD x y) (SRAD x (MOV(W|H|B|WZ|HZ|BZ)reg y)) => (SRAD x y) @@ -668,17 +684,13 @@ (SRW x (MOV(W|H|B|WZ|HZ|BZ)reg y)) => (SRW x y) (SRAW x (MOV(W|H|B|WZ|HZ|BZ)reg y)) => (SRAW x y) -// Constant rotate generation -(RLL x (MOVDconst [c])) => (RLLconst x [int8(c&31)]) -(RLLG x (MOVDconst [c])) => (RLLGconst x [int8(c&63)]) - -(ADD (SLDconst x [c]) (SRDconst x [d])) && d == 64-c => (RLLGconst [c] x) -( OR (SLDconst x [c]) (SRDconst x [d])) && d == 64-c => (RLLGconst [c] x) -(XOR (SLDconst x [c]) (SRDconst x [d])) && d == 64-c => (RLLGconst [c] x) +// Match rotate by constant. +(RLLG x (MOVDconst [c])) => (RISBGZ x {s390x.NewRotateParams(0, 63, int8(c&63))}) +(RLL x (MOVDconst [c])) => (RLLconst x [int8(c&31)]) -(ADDW (SLWconst x [c]) (SRWconst x [d])) && d == 32-c => (RLLconst [c] x) -( ORW (SLWconst x [c]) (SRWconst x [d])) && d == 32-c => (RLLconst [c] x) -(XORW (SLWconst x [c]) (SRWconst x [d])) && d == 32-c => (RLLconst [c] x) +// Match rotate by constant pattern. +((ADD|OR|XOR) (SLDconst x [c]) (SRDconst x [64-c])) => (RISBGZ x {s390x.NewRotateParams(0, 63, c)}) +((ADD|OR|XOR)W (SLWconst x [c]) (SRWconst x [32-c])) => (RLLconst x [c]) // Signed 64-bit comparison with immediate. (CMP x (MOVDconst [c])) && is32Bit(c) => (CMPconst x [int32(c)]) @@ -692,15 +704,97 @@ (CMP(W|WU) x (MOVDconst [c])) => (CMP(W|WU)const x [int32(c)]) (CMP(W|WU) (MOVDconst [c]) x) => (InvertFlags (CMP(W|WU)const x [int32(c)])) +// Match (x >> c) << d to 'rotate then insert selected bits [into zero]'. +(SLDconst (SRDconst x [c]) [d]) => (RISBGZ x {s390x.NewRotateParams(max8(0, c-d), 63-d, (d-c)&63)}) + +// Match (x << c) >> d to 'rotate then insert selected bits [into zero]'. +(SRDconst (SLDconst x [c]) [d]) => (RISBGZ x {s390x.NewRotateParams(d, min8(63, 63-c+d), (c-d)&63)}) + +// Absorb input zero extension into 'rotate then insert selected bits [into zero]'. +(RISBGZ (MOVWZreg x) {r}) && r.InMerge(0xffffffff) != nil => (RISBGZ x {*r.InMerge(0xffffffff)}) +(RISBGZ (MOVHZreg x) {r}) && r.InMerge(0x0000ffff) != nil => (RISBGZ x {*r.InMerge(0x0000ffff)}) +(RISBGZ (MOVBZreg x) {r}) && r.InMerge(0x000000ff) != nil => (RISBGZ x {*r.InMerge(0x000000ff)}) + +// Absorb 'rotate then insert selected bits [into zero]' into zero extension. +(MOVWZreg (RISBGZ x {r})) && r.OutMerge(0xffffffff) != nil => (RISBGZ x {*r.OutMerge(0xffffffff)}) +(MOVHZreg (RISBGZ x {r})) && r.OutMerge(0x0000ffff) != nil => (RISBGZ x {*r.OutMerge(0x0000ffff)}) +(MOVBZreg (RISBGZ x {r})) && r.OutMerge(0x000000ff) != nil => (RISBGZ x {*r.OutMerge(0x000000ff)}) + +// Absorb shift into 'rotate then insert selected bits [into zero]'. +// +// Any unsigned shift can be represented as a rotate and mask operation: +// +// x << c => RotateLeft64(x, c) & (^uint64(0) << c) +// x >> c => RotateLeft64(x, -c) & (^uint64(0) >> c) +// +// Therefore when a shift is used as the input to a rotate then insert +// selected bits instruction we can merge the two together. We just have +// to be careful that the resultant mask is representable (non-zero and +// contiguous). For example, assuming that x is variable and c, y and m +// are constants, a shift followed by a rotate then insert selected bits +// could be represented as: +// +// RotateLeft64(RotateLeft64(x, c) & (^uint64(0) << c), y) & m +// +// We can split the rotation by y into two, one rotate for x and one for +// the mask: +// +// RotateLeft64(RotateLeft64(x, c), y) & (RotateLeft64(^uint64(0) << c, y)) & m +// +// The rotations of x by c followed by y can then be combined: +// +// RotateLeft64(x, c+y) & (RotateLeft64(^uint64(0) << c, y)) & m +// ^^^^^^^^^^^^^^^^^^^^ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ +// rotate mask +// +// To perform this optimization we therefore just need to check that it +// is valid to merge the shift mask (^(uint64(0)< (RISBGZ x {(*r.InMerge(^uint64(0)<>c) != nil => (RISBGZ x {(*r.InMerge(^uint64(0)>>c)).RotateLeft(-c)}) + +// Absorb 'rotate then insert selected bits [into zero]' into left shift. +(SLDconst (RISBGZ x {r}) [c]) + && s390x.NewRotateParams(0, 63-c, c).InMerge(r.OutMask()) != nil + => (RISBGZ x {(*s390x.NewRotateParams(0, 63-c, c).InMerge(r.OutMask())).RotateLeft(r.Amount)}) + +// Absorb 'rotate then insert selected bits [into zero]' into right shift. +(SRDconst (RISBGZ x {r}) [c]) + && s390x.NewRotateParams(c, 63, -c&63).InMerge(r.OutMask()) != nil + => (RISBGZ x {(*s390x.NewRotateParams(c, 63, -c&63).InMerge(r.OutMask())).RotateLeft(r.Amount)}) + +// Merge 'rotate then insert selected bits [into zero]' instructions together. +(RISBGZ (RISBGZ x {y}) {z}) + && z.InMerge(y.OutMask()) != nil + => (RISBGZ x {(*z.InMerge(y.OutMask())).RotateLeft(y.Amount)}) + +// Convert RISBGZ into 64-bit shift (helps CSE). +(RISBGZ x {r}) && r.End == 63 && r.Start == -r.Amount&63 => (SRDconst x [-r.Amount&63]) +(RISBGZ x {r}) && r.Start == 0 && r.End == 63-r.Amount => (SLDconst x [r.Amount]) + +// Optimize single bit isolation when it is known to be equivalent to +// the most significant bit due to mask produced by arithmetic shift. +// Simply isolate the most significant bit itself and place it in the +// correct position. +// +// Example: (int64(x) >> 63) & 0x8 -> RISBGZ $60, $60, $4, Rsrc, Rdst +(RISBGZ (SRADconst x [c]) {r}) + && r.Start == r.End // single bit selected + && (r.Start+r.Amount)&63 <= c // equivalent to most significant bit of x + => (RISBGZ x {s390x.NewRotateParams(r.Start, r.Start, -r.Start&63)}) + // Canonicalize the order of arguments to comparisons - helps with CSE. ((CMP|CMPW|CMPU|CMPWU) x y) && x.ID > y.ID => (InvertFlags ((CMP|CMPW|CMPU|CMPWU) y x)) -// Using MOV{W,H,B}Zreg instead of AND is cheaper. -(AND x (MOVDconst [0xFF])) => (MOVBZreg x) -(AND x (MOVDconst [0xFFFF])) => (MOVHZreg x) -(AND x (MOVDconst [0xFFFFFFFF])) => (MOVWZreg x) -(ANDWconst [0xFF] x) => (MOVBZreg x) -(ANDWconst [0xFFFF] x) => (MOVHZreg x) +// Use sign/zero extend instead of RISBGZ. +(RISBGZ x {r}) && r == s390x.NewRotateParams(56, 63, 0) => (MOVBZreg x) +(RISBGZ x {r}) && r == s390x.NewRotateParams(48, 63, 0) => (MOVHZreg x) +(RISBGZ x {r}) && r == s390x.NewRotateParams(32, 63, 0) => (MOVWZreg x) + +// Use sign/zero extend instead of ANDW. +(ANDWconst [0x00ff] x) => (MOVBZreg x) +(ANDWconst [0xffff] x) => (MOVHZreg x) // Strength reduce multiplication to the sum (or difference) of two powers of two. // @@ -773,21 +867,22 @@ // detect attempts to set/clear the sign bit // may need to be reworked when NIHH/OIHH are added -(SRDconst [1] (SLDconst [1] (LGDR x))) => (LGDR (LPDFR x)) -(LDGR (SRDconst [1] (SLDconst [1] x))) => (LPDFR (LDGR x)) -(AND (MOVDconst [^(-1<<63)]) (LGDR x)) => (LGDR (LPDFR x)) -(LDGR (AND (MOVDconst [^(-1<<63)]) x)) => (LPDFR (LDGR x)) -(OR (MOVDconst [-1<<63]) (LGDR x)) => (LGDR (LNDFR x)) -(LDGR (OR (MOVDconst [-1<<63]) x)) => (LNDFR (LDGR x)) +(RISBGZ (LGDR x) {r}) && r == s390x.NewRotateParams(1, 63, 0) => (LGDR (LPDFR x)) +(LDGR (RISBGZ x {r})) && r == s390x.NewRotateParams(1, 63, 0) => (LPDFR (LDGR x)) +(OR (MOVDconst [-1<<63]) (LGDR x)) => (LGDR (LNDFR x)) +(LDGR (OR (MOVDconst [-1<<63]) x)) => (LNDFR (LDGR x)) // detect attempts to set the sign bit with load (LDGR x:(ORload [off] {sym} (MOVDconst [-1<<63]) ptr mem)) && x.Uses == 1 && clobber(x) => @x.Block (LNDFR (LDGR (MOVDload [off] {sym} ptr mem))) // detect copysign -(OR (SLDconst [63] (SRDconst [63] (LGDR x))) (LGDR (LPDFR y))) => (LGDR (CPSDR y x)) -(OR (SLDconst [63] (SRDconst [63] (LGDR x))) (MOVDconst [c])) && c & -1<<63 == 0 => (LGDR (CPSDR (FMOVDconst [math.Float64frombits(uint64(c))]) x)) -(OR (AND (MOVDconst [-1<<63]) (LGDR x)) (LGDR (LPDFR y))) => (LGDR (CPSDR y x)) -(OR (AND (MOVDconst [-1<<63]) (LGDR x)) (MOVDconst [c])) && c & -1<<63 == 0 => (LGDR (CPSDR (FMOVDconst [math.Float64frombits(uint64(c))]) x)) +(OR (RISBGZ (LGDR x) {r}) (LGDR (LPDFR y))) + && r == s390x.NewRotateParams(0, 0, 0) + => (LGDR (CPSDR y x)) +(OR (RISBGZ (LGDR x) {r}) (MOVDconst [c])) + && c >= 0 + && r == s390x.NewRotateParams(0, 0, 0) + => (LGDR (CPSDR (FMOVDconst [math.Float64frombits(uint64(c))]) x)) (CPSDR y (FMOVDconst [c])) && !math.Signbit(c) => (LPDFR y) (CPSDR y (FMOVDconst [c])) && math.Signbit(c) => (LNDFR y) @@ -966,6 +1061,9 @@ (CMPWconst (ANDWconst _ [m]) [n]) && int32(m) >= 0 && int32(m) < int32(n) => (FlagLT) (CMPWUconst (ANDWconst _ [m]) [n]) && uint32(m) < uint32(n) => (FlagLT) +(CMPconst (RISBGZ x {r}) [c]) && c > 0 && r.OutMask() < uint64(c) => (FlagLT) +(CMPUconst (RISBGZ x {r}) [c]) && r.OutMask() < uint64(uint32(c)) => (FlagLT) + // Constant compare-and-branch with immediate. (CGIJ {c} (MOVDconst [x]) [y] yes no) && c&s390x.Equal != 0 && int64(x) == int64(y) => (First yes no) (CGIJ {c} (MOVDconst [x]) [y] yes no) && c&s390x.Less != 0 && int64(x) < int64(y) => (First yes no) diff --git a/src/cmd/compile/internal/ssa/gen/S390XOps.go b/src/cmd/compile/internal/ssa/gen/S390XOps.go index 728cfb5508..f0cf2f2f6e 100644 --- a/src/cmd/compile/internal/ssa/gen/S390XOps.go +++ b/src/cmd/compile/internal/ssa/gen/S390XOps.go @@ -331,25 +331,26 @@ func init() { {name: "LTEBR", argLength: 1, reg: fp1flags, asm: "LTEBR", typ: "Flags"}, // arg0 compare to 0, f32 {name: "SLD", argLength: 2, reg: sh21, asm: "SLD"}, // arg0 << arg1, shift amount is mod 64 - {name: "SLW", argLength: 2, reg: sh21, asm: "SLW"}, // arg0 << arg1, shift amount is mod 32 + {name: "SLW", argLength: 2, reg: sh21, asm: "SLW"}, // arg0 << arg1, shift amount is mod 64 {name: "SLDconst", argLength: 1, reg: gp11, asm: "SLD", aux: "Int8"}, // arg0 << auxint, shift amount 0-63 {name: "SLWconst", argLength: 1, reg: gp11, asm: "SLW", aux: "Int8"}, // arg0 << auxint, shift amount 0-31 {name: "SRD", argLength: 2, reg: sh21, asm: "SRD"}, // unsigned arg0 >> arg1, shift amount is mod 64 - {name: "SRW", argLength: 2, reg: sh21, asm: "SRW"}, // unsigned uint32(arg0) >> arg1, shift amount is mod 32 + {name: "SRW", argLength: 2, reg: sh21, asm: "SRW"}, // unsigned uint32(arg0) >> arg1, shift amount is mod 64 {name: "SRDconst", argLength: 1, reg: gp11, asm: "SRD", aux: "Int8"}, // unsigned arg0 >> auxint, shift amount 0-63 {name: "SRWconst", argLength: 1, reg: gp11, asm: "SRW", aux: "Int8"}, // unsigned uint32(arg0) >> auxint, shift amount 0-31 // Arithmetic shifts clobber flags. {name: "SRAD", argLength: 2, reg: sh21, asm: "SRAD", clobberFlags: true}, // signed arg0 >> arg1, shift amount is mod 64 - {name: "SRAW", argLength: 2, reg: sh21, asm: "SRAW", clobberFlags: true}, // signed int32(arg0) >> arg1, shift amount is mod 32 + {name: "SRAW", argLength: 2, reg: sh21, asm: "SRAW", clobberFlags: true}, // signed int32(arg0) >> arg1, shift amount is mod 64 {name: "SRADconst", argLength: 1, reg: gp11, asm: "SRAD", aux: "Int8", clobberFlags: true}, // signed arg0 >> auxint, shift amount 0-63 {name: "SRAWconst", argLength: 1, reg: gp11, asm: "SRAW", aux: "Int8", clobberFlags: true}, // signed int32(arg0) >> auxint, shift amount 0-31 - {name: "RLLG", argLength: 2, reg: sh21, asm: "RLLG"}, // arg0 rotate left arg1, rotate amount 0-63 - {name: "RLL", argLength: 2, reg: sh21, asm: "RLL"}, // arg0 rotate left arg1, rotate amount 0-31 - {name: "RLLGconst", argLength: 1, reg: gp11, asm: "RLLG", aux: "Int8"}, // arg0 rotate left auxint, rotate amount 0-63 - {name: "RLLconst", argLength: 1, reg: gp11, asm: "RLL", aux: "Int8"}, // arg0 rotate left auxint, rotate amount 0-31 + // Rotate instructions. + // Note: no RLLGconst - use RISBGZ instead. + {name: "RLLG", argLength: 2, reg: sh21, asm: "RLLG"}, // arg0 rotate left arg1, rotate amount 0-63 + {name: "RLL", argLength: 2, reg: sh21, asm: "RLL"}, // arg0 rotate left arg1, rotate amount 0-31 + {name: "RLLconst", argLength: 1, reg: gp11, asm: "RLL", aux: "Int8"}, // arg0 rotate left auxint, rotate amount 0-31 // Rotate then (and|or|xor|insert) selected bits instructions. // @@ -371,6 +372,7 @@ func init() { // +-------------+-------+-----+--------+-----------------------+-----------------------+-----------------------+ // {name: "RXSBG", argLength: 2, reg: gp21, asm: "RXSBG", resultInArg0: true, aux: "S390XRotateParams", clobberFlags: true}, // rotate then xor selected bits + {name: "RISBGZ", argLength: 1, reg: gp11, asm: "RISBGZ", aux: "S390XRotateParams", clobberFlags: true}, // rotate then insert selected bits [into zero] // unary ops {name: "NEG", argLength: 1, reg: gp11, asm: "NEG", clobberFlags: true}, // -arg0 @@ -547,9 +549,9 @@ func init() { // Atomic bitwise operations. // Note: 'floor' operations round the pointer down to the nearest word boundary // which reflects how they are used in the runtime. - {name: "LAN", argLength: 3, reg: gpstore, asm: "LAN", typ: "Mem", clobberFlags: true, hasSideEffects: true}, // *arg0 &= arg1. arg2 = mem. + {name: "LAN", argLength: 3, reg: gpstore, asm: "LAN", typ: "Mem", clobberFlags: true, hasSideEffects: true}, // *arg0 &= arg1. arg2 = mem. {name: "LANfloor", argLength: 3, reg: gpstorelab, asm: "LAN", typ: "Mem", clobberFlags: true, hasSideEffects: true}, // *(floor(arg0, 4)) &= arg1. arg2 = mem. - {name: "LAO", argLength: 3, reg: gpstore, asm: "LAO", typ: "Mem", clobberFlags: true, hasSideEffects: true}, // *arg0 |= arg1. arg2 = mem. + {name: "LAO", argLength: 3, reg: gpstore, asm: "LAO", typ: "Mem", clobberFlags: true, hasSideEffects: true}, // *arg0 |= arg1. arg2 = mem. {name: "LAOfloor", argLength: 3, reg: gpstorelab, asm: "LAO", typ: "Mem", clobberFlags: true, hasSideEffects: true}, // *(floor(arg0, 4)) |= arg1. arg2 = mem. // Compare and swap. diff --git a/src/cmd/compile/internal/ssa/opGen.go b/src/cmd/compile/internal/ssa/opGen.go index c0b663cd8f..eceef1d91a 100644 --- a/src/cmd/compile/internal/ssa/opGen.go +++ b/src/cmd/compile/internal/ssa/opGen.go @@ -2285,9 +2285,9 @@ const ( OpS390XSRAWconst OpS390XRLLG OpS390XRLL - OpS390XRLLGconst OpS390XRLLconst OpS390XRXSBG + OpS390XRISBGZ OpS390XNEG OpS390XNEGW OpS390XNOT @@ -30740,10 +30740,10 @@ var opcodeTable = [...]opInfo{ }, }, { - name: "RLLGconst", + name: "RLLconst", auxType: auxInt8, argLen: 1, - asm: s390x.ARLLG, + asm: s390x.ARLL, reg: regInfo{ inputs: []inputInfo{ {0, 23551}, // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R11 R12 R14 @@ -30754,13 +30754,16 @@ var opcodeTable = [...]opInfo{ }, }, { - name: "RLLconst", - auxType: auxInt8, - argLen: 1, - asm: s390x.ARLL, + name: "RXSBG", + auxType: auxS390XRotateParams, + argLen: 2, + resultInArg0: true, + clobberFlags: true, + asm: s390x.ARXSBG, reg: regInfo{ inputs: []inputInfo{ {0, 23551}, // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R11 R12 R14 + {1, 23551}, // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R11 R12 R14 }, outputs: []outputInfo{ {0, 23551}, // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R11 R12 R14 @@ -30768,16 +30771,14 @@ var opcodeTable = [...]opInfo{ }, }, { - name: "RXSBG", + name: "RISBGZ", auxType: auxS390XRotateParams, - argLen: 2, - resultInArg0: true, + argLen: 1, clobberFlags: true, - asm: s390x.ARXSBG, + asm: s390x.ARISBGZ, reg: regInfo{ inputs: []inputInfo{ {0, 23551}, // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R11 R12 R14 - {1, 23551}, // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R11 R12 R14 }, outputs: []outputInfo{ {0, 23551}, // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R11 R12 R14 diff --git a/src/cmd/compile/internal/ssa/rewriteS390X.go b/src/cmd/compile/internal/ssa/rewriteS390X.go index 8c3c61d584..d66113d111 100644 --- a/src/cmd/compile/internal/ssa/rewriteS390X.go +++ b/src/cmd/compile/internal/ssa/rewriteS390X.go @@ -699,6 +699,8 @@ func rewriteValueS390X(v *Value) bool { return rewriteValueS390X_OpS390XORconst(v) case OpS390XORload: return rewriteValueS390X_OpS390XORload(v) + case OpS390XRISBGZ: + return rewriteValueS390X_OpS390XRISBGZ(v) case OpS390XRLL: return rewriteValueS390X_OpS390XRLL(v) case OpS390XRLLG: @@ -5272,9 +5274,8 @@ func rewriteValueS390X_OpS390XADD(v *Value) bool { } break } - // match: (ADD (SLDconst x [c]) (SRDconst x [d])) - // cond: d == 64-c - // result: (RLLGconst [c] x) + // match: (ADD (SLDconst x [c]) (SRDconst x [64-c])) + // result: (RISBGZ x {s390x.NewRotateParams(0, 63, c)}) for { for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 { if v_0.Op != OpS390XSLDconst { @@ -5282,15 +5283,11 @@ func rewriteValueS390X_OpS390XADD(v *Value) bool { } c := auxIntToInt8(v_0.AuxInt) x := v_0.Args[0] - if v_1.Op != OpS390XSRDconst { + if v_1.Op != OpS390XSRDconst || auxIntToInt8(v_1.AuxInt) != 64-c || x != v_1.Args[0] { continue } - d := auxIntToInt8(v_1.AuxInt) - if x != v_1.Args[0] || !(d == 64-c) { - continue - } - v.reset(OpS390XRLLGconst) - v.AuxInt = int8ToAuxInt(c) + v.reset(OpS390XRISBGZ) + v.Aux = s390xRotateParamsToAux(s390x.NewRotateParams(0, 63, c)) v.AddArg(x) return true } @@ -5470,9 +5467,8 @@ func rewriteValueS390X_OpS390XADDW(v *Value) bool { } break } - // match: (ADDW (SLWconst x [c]) (SRWconst x [d])) - // cond: d == 32-c - // result: (RLLconst [c] x) + // match: (ADDW (SLWconst x [c]) (SRWconst x [32-c])) + // result: (RLLconst x [c]) for { for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 { if v_0.Op != OpS390XSLWconst { @@ -5480,11 +5476,7 @@ func rewriteValueS390X_OpS390XADDW(v *Value) bool { } c := auxIntToInt8(v_0.AuxInt) x := v_0.Args[0] - if v_1.Op != OpS390XSRWconst { - continue - } - d := auxIntToInt8(v_1.AuxInt) - if x != v_1.Args[0] || !(d == 32-c) { + if v_1.Op != OpS390XSRWconst || auxIntToInt8(v_1.AuxInt) != 32-c || x != v_1.Args[0] { continue } v.reset(OpS390XRLLconst) @@ -5844,8 +5836,8 @@ func rewriteValueS390X_OpS390XAND(v *Value) bool { b := v.Block typ := &b.Func.Config.Types // match: (AND x (MOVDconst [c])) - // cond: is32Bit(c) && c < 0 - // result: (ANDconst [c] x) + // cond: s390x.NewRotateParams(0, 63, 0).OutMerge(uint64(c)) != nil + // result: (RISBGZ x {*s390x.NewRotateParams(0, 63, 0).OutMerge(uint64(c))}) for { for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 { x := v_0 @@ -5853,19 +5845,19 @@ func rewriteValueS390X_OpS390XAND(v *Value) bool { continue } c := auxIntToInt64(v_1.AuxInt) - if !(is32Bit(c) && c < 0) { + if !(s390x.NewRotateParams(0, 63, 0).OutMerge(uint64(c)) != nil) { continue } - v.reset(OpS390XANDconst) - v.AuxInt = int64ToAuxInt(c) + v.reset(OpS390XRISBGZ) + v.Aux = s390xRotateParamsToAux(*s390x.NewRotateParams(0, 63, 0).OutMerge(uint64(c))) v.AddArg(x) return true } break } // match: (AND x (MOVDconst [c])) - // cond: is32Bit(c) && c >= 0 - // result: (MOVWZreg (ANDWconst [int32(c)] x)) + // cond: is32Bit(c) && c < 0 + // result: (ANDconst [c] x) for { for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 { x := v_0 @@ -5873,72 +5865,32 @@ func rewriteValueS390X_OpS390XAND(v *Value) bool { continue } c := auxIntToInt64(v_1.AuxInt) - if !(is32Bit(c) && c >= 0) { - continue - } - v.reset(OpS390XMOVWZreg) - v0 := b.NewValue0(v.Pos, OpS390XANDWconst, typ.UInt32) - v0.AuxInt = int32ToAuxInt(int32(c)) - v0.AddArg(x) - v.AddArg(v0) - return true - } - break - } - // match: (AND x (MOVDconst [0xFF])) - // result: (MOVBZreg x) - for { - for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 { - x := v_0 - if v_1.Op != OpS390XMOVDconst || auxIntToInt64(v_1.AuxInt) != 0xFF { + if !(is32Bit(c) && c < 0) { continue } - v.reset(OpS390XMOVBZreg) + v.reset(OpS390XANDconst) + v.AuxInt = int64ToAuxInt(c) v.AddArg(x) return true } break } - // match: (AND x (MOVDconst [0xFFFF])) - // result: (MOVHZreg x) + // match: (AND x (MOVDconst [c])) + // cond: is32Bit(c) && c >= 0 + // result: (MOVWZreg (ANDWconst [int32(c)] x)) for { for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 { x := v_0 - if v_1.Op != OpS390XMOVDconst || auxIntToInt64(v_1.AuxInt) != 0xFFFF { + if v_1.Op != OpS390XMOVDconst { continue } - v.reset(OpS390XMOVHZreg) - v.AddArg(x) - return true - } - break - } - // match: (AND x (MOVDconst [0xFFFFFFFF])) - // result: (MOVWZreg x) - for { - for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 { - x := v_0 - if v_1.Op != OpS390XMOVDconst || auxIntToInt64(v_1.AuxInt) != 0xFFFFFFFF { + c := auxIntToInt64(v_1.AuxInt) + if !(is32Bit(c) && c >= 0) { continue } v.reset(OpS390XMOVWZreg) - v.AddArg(x) - return true - } - break - } - // match: (AND (MOVDconst [^(-1<<63)]) (LGDR x)) - // result: (LGDR (LPDFR x)) - for { - for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 { - if v_0.Op != OpS390XMOVDconst || auxIntToInt64(v_0.AuxInt) != ^(-1<<63) || v_1.Op != OpS390XLGDR { - continue - } - t := v_1.Type - x := v_1.Args[0] - v.reset(OpS390XLGDR) - v.Type = t - v0 := b.NewValue0(v.Pos, OpS390XLPDFR, x.Type) + v0 := b.NewValue0(v.Pos, OpS390XANDWconst, typ.UInt32) + v0.AuxInt = int32ToAuxInt(int32(c)) v0.AddArg(x) v.AddArg(v0) return true @@ -6103,10 +6055,10 @@ func rewriteValueS390X_OpS390XANDWconst(v *Value) bool { v.AddArg(x) return true } - // match: (ANDWconst [0xFF] x) + // match: (ANDWconst [0x00ff] x) // result: (MOVBZreg x) for { - if auxIntToInt32(v.AuxInt) != 0xFF { + if auxIntToInt32(v.AuxInt) != 0x00ff { break } x := v_0 @@ -6114,10 +6066,10 @@ func rewriteValueS390X_OpS390XANDWconst(v *Value) bool { v.AddArg(x) return true } - // match: (ANDWconst [0xFFFF] x) + // match: (ANDWconst [0xffff] x) // result: (MOVHZreg x) for { - if auxIntToInt32(v.AuxInt) != 0xFFFF { + if auxIntToInt32(v.AuxInt) != 0xffff { break } x := v_0 @@ -6515,6 +6467,21 @@ func rewriteValueS390X_OpS390XCMPUconst(v *Value) bool { v.reset(OpS390XFlagLT) return true } + // match: (CMPUconst (RISBGZ x {r}) [c]) + // cond: r.OutMask() < uint64(uint32(c)) + // result: (FlagLT) + for { + c := auxIntToInt32(v.AuxInt) + if v_0.Op != OpS390XRISBGZ { + break + } + r := auxToS390xRotateParams(v_0.Aux) + if !(r.OutMask() < uint64(uint32(c))) { + break + } + v.reset(OpS390XFlagLT) + return true + } // match: (CMPUconst (MOVWZreg x) [c]) // result: (CMPWUconst x [c]) for { @@ -7152,6 +7119,21 @@ func rewriteValueS390X_OpS390XCMPconst(v *Value) bool { v.reset(OpS390XFlagGT) return true } + // match: (CMPconst (RISBGZ x {r}) [c]) + // cond: c > 0 && r.OutMask() < uint64(c) + // result: (FlagLT) + for { + c := auxIntToInt32(v.AuxInt) + if v_0.Op != OpS390XRISBGZ { + break + } + r := auxToS390xRotateParams(v_0.Aux) + if !(c > 0 && r.OutMask() < uint64(c)) { + break + } + v.reset(OpS390XFlagLT) + return true + } // match: (CMPconst (MOVWreg x) [c]) // result: (CMPWconst x [c]) for { @@ -7684,47 +7666,25 @@ func rewriteValueS390X_OpS390XFNEGS(v *Value) bool { func rewriteValueS390X_OpS390XLDGR(v *Value) bool { v_0 := v.Args[0] b := v.Block - // match: (LDGR (SRDconst [1] (SLDconst [1] x))) + // match: (LDGR (RISBGZ x {r})) + // cond: r == s390x.NewRotateParams(1, 63, 0) // result: (LPDFR (LDGR x)) for { t := v.Type - if v_0.Op != OpS390XSRDconst || auxIntToInt8(v_0.AuxInt) != 1 { + if v_0.Op != OpS390XRISBGZ { break } - v_0_0 := v_0.Args[0] - if v_0_0.Op != OpS390XSLDconst || auxIntToInt8(v_0_0.AuxInt) != 1 { + r := auxToS390xRotateParams(v_0.Aux) + x := v_0.Args[0] + if !(r == s390x.NewRotateParams(1, 63, 0)) { break } - x := v_0_0.Args[0] v.reset(OpS390XLPDFR) v0 := b.NewValue0(v.Pos, OpS390XLDGR, t) v0.AddArg(x) v.AddArg(v0) return true } - // match: (LDGR (AND (MOVDconst [^(-1<<63)]) x)) - // result: (LPDFR (LDGR x)) - for { - t := v.Type - if v_0.Op != OpS390XAND { - break - } - _ = v_0.Args[1] - v_0_0 := v_0.Args[0] - v_0_1 := v_0.Args[1] - for _i0 := 0; _i0 <= 1; _i0, v_0_0, v_0_1 = _i0+1, v_0_1, v_0_0 { - if v_0_0.Op != OpS390XMOVDconst || auxIntToInt64(v_0_0.AuxInt) != ^(-1<<63) { - continue - } - x := v_0_1 - v.reset(OpS390XLPDFR) - v0 := b.NewValue0(v.Pos, OpS390XLDGR, t) - v0.AddArg(x) - v.AddArg(v0) - return true - } - break - } // match: (LDGR (OR (MOVDconst [-1<<63]) x)) // result: (LNDFR (LDGR x)) for { @@ -8309,6 +8269,23 @@ func rewriteValueS390X_OpS390XMOVBZreg(v *Value) bool { v.copyOf(x) return true } + // match: (MOVBZreg (RISBGZ x {r})) + // cond: r.OutMerge(0x000000ff) != nil + // result: (RISBGZ x {*r.OutMerge(0x000000ff)}) + for { + if v_0.Op != OpS390XRISBGZ { + break + } + r := auxToS390xRotateParams(v_0.Aux) + x := v_0.Args[0] + if !(r.OutMerge(0x000000ff) != nil) { + break + } + v.reset(OpS390XRISBGZ) + v.Aux = s390xRotateParamsToAux(*r.OutMerge(0x000000ff)) + v.AddArg(x) + return true + } // match: (MOVBZreg (ANDWconst [m] x)) // result: (MOVWZreg (ANDWconst [int32( uint8(m))] x)) for { @@ -9697,6 +9674,23 @@ func rewriteValueS390X_OpS390XMOVHZreg(v *Value) bool { v.AuxInt = int64ToAuxInt(int64(uint16(c))) return true } + // match: (MOVHZreg (RISBGZ x {r})) + // cond: r.OutMerge(0x0000ffff) != nil + // result: (RISBGZ x {*r.OutMerge(0x0000ffff)}) + for { + if v_0.Op != OpS390XRISBGZ { + break + } + r := auxToS390xRotateParams(v_0.Aux) + x := v_0.Args[0] + if !(r.OutMerge(0x0000ffff) != nil) { + break + } + v.reset(OpS390XRISBGZ) + v.Aux = s390xRotateParamsToAux(*r.OutMerge(0x0000ffff)) + v.AddArg(x) + return true + } // match: (MOVHZreg (ANDWconst [m] x)) // result: (MOVWZreg (ANDWconst [int32(uint16(m))] x)) for { @@ -10547,6 +10541,23 @@ func rewriteValueS390X_OpS390XMOVWZreg(v *Value) bool { v.AuxInt = int64ToAuxInt(int64(uint32(c))) return true } + // match: (MOVWZreg (RISBGZ x {r})) + // cond: r.OutMerge(0xffffffff) != nil + // result: (RISBGZ x {*r.OutMerge(0xffffffff)}) + for { + if v_0.Op != OpS390XRISBGZ { + break + } + r := auxToS390xRotateParams(v_0.Aux) + x := v_0.Args[0] + if !(r.OutMerge(0xffffffff) != nil) { + break + } + v.reset(OpS390XRISBGZ) + v.Aux = s390xRotateParamsToAux(*r.OutMerge(0xffffffff)) + v.AddArg(x) + return true + } return false } func rewriteValueS390X_OpS390XMOVWload(v *Value) bool { @@ -11622,9 +11633,8 @@ func rewriteValueS390X_OpS390XOR(v *Value) bool { } break } - // match: ( OR (SLDconst x [c]) (SRDconst x [d])) - // cond: d == 64-c - // result: (RLLGconst [c] x) + // match: (OR (SLDconst x [c]) (SRDconst x [64-c])) + // result: (RISBGZ x {s390x.NewRotateParams(0, 63, c)}) for { for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 { if v_0.Op != OpS390XSLDconst { @@ -11632,15 +11642,11 @@ func rewriteValueS390X_OpS390XOR(v *Value) bool { } c := auxIntToInt8(v_0.AuxInt) x := v_0.Args[0] - if v_1.Op != OpS390XSRDconst { - continue - } - d := auxIntToInt8(v_1.AuxInt) - if x != v_1.Args[0] || !(d == 64-c) { + if v_1.Op != OpS390XSRDconst || auxIntToInt8(v_1.AuxInt) != 64-c || x != v_1.Args[0] { continue } - v.reset(OpS390XRLLGconst) - v.AuxInt = int8ToAuxInt(c) + v.reset(OpS390XRISBGZ) + v.Aux = s390xRotateParamsToAux(s390x.NewRotateParams(0, 63, c)) v.AddArg(x) return true } @@ -11664,22 +11670,20 @@ func rewriteValueS390X_OpS390XOR(v *Value) bool { } break } - // match: (OR (SLDconst [63] (SRDconst [63] (LGDR x))) (LGDR (LPDFR y))) + // match: (OR (RISBGZ (LGDR x) {r}) (LGDR (LPDFR y))) + // cond: r == s390x.NewRotateParams(0, 0, 0) // result: (LGDR (CPSDR y x)) for { for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 { - if v_0.Op != OpS390XSLDconst || auxIntToInt8(v_0.AuxInt) != 63 { + if v_0.Op != OpS390XRISBGZ { continue } + r := auxToS390xRotateParams(v_0.Aux) v_0_0 := v_0.Args[0] - if v_0_0.Op != OpS390XSRDconst || auxIntToInt8(v_0_0.AuxInt) != 63 { - continue - } - v_0_0_0 := v_0_0.Args[0] - if v_0_0_0.Op != OpS390XLGDR { + if v_0_0.Op != OpS390XLGDR { continue } - x := v_0_0_0.Args[0] + x := v_0_0.Args[0] if v_1.Op != OpS390XLGDR { continue } @@ -11689,6 +11693,9 @@ func rewriteValueS390X_OpS390XOR(v *Value) bool { } t := v_1_0.Type y := v_1_0.Args[0] + if !(r == s390x.NewRotateParams(0, 0, 0)) { + continue + } v.reset(OpS390XLGDR) v0 := b.NewValue0(v.Pos, OpS390XCPSDR, t) v0.AddArg2(y, x) @@ -11697,28 +11704,25 @@ func rewriteValueS390X_OpS390XOR(v *Value) bool { } break } - // match: (OR (SLDconst [63] (SRDconst [63] (LGDR x))) (MOVDconst [c])) - // cond: c & -1<<63 == 0 + // match: (OR (RISBGZ (LGDR x) {r}) (MOVDconst [c])) + // cond: c >= 0 && r == s390x.NewRotateParams(0, 0, 0) // result: (LGDR (CPSDR (FMOVDconst [math.Float64frombits(uint64(c))]) x)) for { for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 { - if v_0.Op != OpS390XSLDconst || auxIntToInt8(v_0.AuxInt) != 63 { + if v_0.Op != OpS390XRISBGZ { continue } + r := auxToS390xRotateParams(v_0.Aux) v_0_0 := v_0.Args[0] - if v_0_0.Op != OpS390XSRDconst || auxIntToInt8(v_0_0.AuxInt) != 63 { + if v_0_0.Op != OpS390XLGDR { continue } - v_0_0_0 := v_0_0.Args[0] - if v_0_0_0.Op != OpS390XLGDR { - continue - } - x := v_0_0_0.Args[0] + x := v_0_0.Args[0] if v_1.Op != OpS390XMOVDconst { continue } c := auxIntToInt64(v_1.AuxInt) - if !(c&-1<<63 == 0) { + if !(c >= 0 && r == s390x.NewRotateParams(0, 0, 0)) { continue } v.reset(OpS390XLGDR) @@ -11731,73 +11735,6 @@ func rewriteValueS390X_OpS390XOR(v *Value) bool { } break } - // match: (OR (AND (MOVDconst [-1<<63]) (LGDR x)) (LGDR (LPDFR y))) - // result: (LGDR (CPSDR y x)) - for { - for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 { - if v_0.Op != OpS390XAND { - continue - } - _ = v_0.Args[1] - v_0_0 := v_0.Args[0] - v_0_1 := v_0.Args[1] - for _i1 := 0; _i1 <= 1; _i1, v_0_0, v_0_1 = _i1+1, v_0_1, v_0_0 { - if v_0_0.Op != OpS390XMOVDconst || auxIntToInt64(v_0_0.AuxInt) != -1<<63 || v_0_1.Op != OpS390XLGDR { - continue - } - x := v_0_1.Args[0] - if v_1.Op != OpS390XLGDR { - continue - } - v_1_0 := v_1.Args[0] - if v_1_0.Op != OpS390XLPDFR { - continue - } - t := v_1_0.Type - y := v_1_0.Args[0] - v.reset(OpS390XLGDR) - v0 := b.NewValue0(v.Pos, OpS390XCPSDR, t) - v0.AddArg2(y, x) - v.AddArg(v0) - return true - } - } - break - } - // match: (OR (AND (MOVDconst [-1<<63]) (LGDR x)) (MOVDconst [c])) - // cond: c & -1<<63 == 0 - // result: (LGDR (CPSDR (FMOVDconst [math.Float64frombits(uint64(c))]) x)) - for { - for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 { - if v_0.Op != OpS390XAND { - continue - } - _ = v_0.Args[1] - v_0_0 := v_0.Args[0] - v_0_1 := v_0.Args[1] - for _i1 := 0; _i1 <= 1; _i1, v_0_0, v_0_1 = _i1+1, v_0_1, v_0_0 { - if v_0_0.Op != OpS390XMOVDconst || auxIntToInt64(v_0_0.AuxInt) != -1<<63 || v_0_1.Op != OpS390XLGDR { - continue - } - x := v_0_1.Args[0] - if v_1.Op != OpS390XMOVDconst { - continue - } - c := auxIntToInt64(v_1.AuxInt) - if !(c&-1<<63 == 0) { - continue - } - v.reset(OpS390XLGDR) - v0 := b.NewValue0(v.Pos, OpS390XCPSDR, x.Type) - v1 := b.NewValue0(v.Pos, OpS390XFMOVDconst, x.Type) - v1.AuxInt = float64ToAuxInt(math.Float64frombits(uint64(c))) - v0.AddArg2(v1, x) - v.AddArg(v0) - return true - } - } - break - } // match: (OR (MOVDconst [c]) (MOVDconst [d])) // result: (MOVDconst [c|d]) for { @@ -12394,9 +12331,8 @@ func rewriteValueS390X_OpS390XORW(v *Value) bool { } break } - // match: ( ORW (SLWconst x [c]) (SRWconst x [d])) - // cond: d == 32-c - // result: (RLLconst [c] x) + // match: (ORW (SLWconst x [c]) (SRWconst x [32-c])) + // result: (RLLconst x [c]) for { for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 { if v_0.Op != OpS390XSLWconst { @@ -12404,11 +12340,7 @@ func rewriteValueS390X_OpS390XORW(v *Value) bool { } c := auxIntToInt8(v_0.AuxInt) x := v_0.Args[0] - if v_1.Op != OpS390XSRWconst { - continue - } - d := auxIntToInt8(v_1.AuxInt) - if x != v_1.Args[0] || !(d == 32-c) { + if v_1.Op != OpS390XSRWconst || auxIntToInt8(v_1.AuxInt) != 32-c || x != v_1.Args[0] { continue } v.reset(OpS390XRLLconst) @@ -12980,6 +12912,221 @@ func rewriteValueS390X_OpS390XORload(v *Value) bool { } return false } +func rewriteValueS390X_OpS390XRISBGZ(v *Value) bool { + v_0 := v.Args[0] + b := v.Block + // match: (RISBGZ (MOVWZreg x) {r}) + // cond: r.InMerge(0xffffffff) != nil + // result: (RISBGZ x {*r.InMerge(0xffffffff)}) + for { + r := auxToS390xRotateParams(v.Aux) + if v_0.Op != OpS390XMOVWZreg { + break + } + x := v_0.Args[0] + if !(r.InMerge(0xffffffff) != nil) { + break + } + v.reset(OpS390XRISBGZ) + v.Aux = s390xRotateParamsToAux(*r.InMerge(0xffffffff)) + v.AddArg(x) + return true + } + // match: (RISBGZ (MOVHZreg x) {r}) + // cond: r.InMerge(0x0000ffff) != nil + // result: (RISBGZ x {*r.InMerge(0x0000ffff)}) + for { + r := auxToS390xRotateParams(v.Aux) + if v_0.Op != OpS390XMOVHZreg { + break + } + x := v_0.Args[0] + if !(r.InMerge(0x0000ffff) != nil) { + break + } + v.reset(OpS390XRISBGZ) + v.Aux = s390xRotateParamsToAux(*r.InMerge(0x0000ffff)) + v.AddArg(x) + return true + } + // match: (RISBGZ (MOVBZreg x) {r}) + // cond: r.InMerge(0x000000ff) != nil + // result: (RISBGZ x {*r.InMerge(0x000000ff)}) + for { + r := auxToS390xRotateParams(v.Aux) + if v_0.Op != OpS390XMOVBZreg { + break + } + x := v_0.Args[0] + if !(r.InMerge(0x000000ff) != nil) { + break + } + v.reset(OpS390XRISBGZ) + v.Aux = s390xRotateParamsToAux(*r.InMerge(0x000000ff)) + v.AddArg(x) + return true + } + // match: (RISBGZ (SLDconst x [c]) {r}) + // cond: r.InMerge(^uint64(0)<>c) != nil + // result: (RISBGZ x {(*r.InMerge(^uint64(0)>>c)).RotateLeft(-c)}) + for { + r := auxToS390xRotateParams(v.Aux) + if v_0.Op != OpS390XSRDconst { + break + } + c := auxIntToInt8(v_0.AuxInt) + x := v_0.Args[0] + if !(r.InMerge(^uint64(0)>>c) != nil) { + break + } + v.reset(OpS390XRISBGZ) + v.Aux = s390xRotateParamsToAux((*r.InMerge(^uint64(0) >> c)).RotateLeft(-c)) + v.AddArg(x) + return true + } + // match: (RISBGZ (RISBGZ x {y}) {z}) + // cond: z.InMerge(y.OutMask()) != nil + // result: (RISBGZ x {(*z.InMerge(y.OutMask())).RotateLeft(y.Amount)}) + for { + z := auxToS390xRotateParams(v.Aux) + if v_0.Op != OpS390XRISBGZ { + break + } + y := auxToS390xRotateParams(v_0.Aux) + x := v_0.Args[0] + if !(z.InMerge(y.OutMask()) != nil) { + break + } + v.reset(OpS390XRISBGZ) + v.Aux = s390xRotateParamsToAux((*z.InMerge(y.OutMask())).RotateLeft(y.Amount)) + v.AddArg(x) + return true + } + // match: (RISBGZ x {r}) + // cond: r.End == 63 && r.Start == -r.Amount&63 + // result: (SRDconst x [-r.Amount&63]) + for { + r := auxToS390xRotateParams(v.Aux) + x := v_0 + if !(r.End == 63 && r.Start == -r.Amount&63) { + break + } + v.reset(OpS390XSRDconst) + v.AuxInt = int8ToAuxInt(-r.Amount & 63) + v.AddArg(x) + return true + } + // match: (RISBGZ x {r}) + // cond: r.Start == 0 && r.End == 63-r.Amount + // result: (SLDconst x [r.Amount]) + for { + r := auxToS390xRotateParams(v.Aux) + x := v_0 + if !(r.Start == 0 && r.End == 63-r.Amount) { + break + } + v.reset(OpS390XSLDconst) + v.AuxInt = int8ToAuxInt(r.Amount) + v.AddArg(x) + return true + } + // match: (RISBGZ (SRADconst x [c]) {r}) + // cond: r.Start == r.End && (r.Start+r.Amount)&63 <= c + // result: (RISBGZ x {s390x.NewRotateParams(r.Start, r.Start, -r.Start&63)}) + for { + r := auxToS390xRotateParams(v.Aux) + if v_0.Op != OpS390XSRADconst { + break + } + c := auxIntToInt8(v_0.AuxInt) + x := v_0.Args[0] + if !(r.Start == r.End && (r.Start+r.Amount)&63 <= c) { + break + } + v.reset(OpS390XRISBGZ) + v.Aux = s390xRotateParamsToAux(s390x.NewRotateParams(r.Start, r.Start, -r.Start&63)) + v.AddArg(x) + return true + } + // match: (RISBGZ x {r}) + // cond: r == s390x.NewRotateParams(56, 63, 0) + // result: (MOVBZreg x) + for { + r := auxToS390xRotateParams(v.Aux) + x := v_0 + if !(r == s390x.NewRotateParams(56, 63, 0)) { + break + } + v.reset(OpS390XMOVBZreg) + v.AddArg(x) + return true + } + // match: (RISBGZ x {r}) + // cond: r == s390x.NewRotateParams(48, 63, 0) + // result: (MOVHZreg x) + for { + r := auxToS390xRotateParams(v.Aux) + x := v_0 + if !(r == s390x.NewRotateParams(48, 63, 0)) { + break + } + v.reset(OpS390XMOVHZreg) + v.AddArg(x) + return true + } + // match: (RISBGZ x {r}) + // cond: r == s390x.NewRotateParams(32, 63, 0) + // result: (MOVWZreg x) + for { + r := auxToS390xRotateParams(v.Aux) + x := v_0 + if !(r == s390x.NewRotateParams(32, 63, 0)) { + break + } + v.reset(OpS390XMOVWZreg) + v.AddArg(x) + return true + } + // match: (RISBGZ (LGDR x) {r}) + // cond: r == s390x.NewRotateParams(1, 63, 0) + // result: (LGDR (LPDFR x)) + for { + r := auxToS390xRotateParams(v.Aux) + if v_0.Op != OpS390XLGDR { + break + } + t := v_0.Type + x := v_0.Args[0] + if !(r == s390x.NewRotateParams(1, 63, 0)) { + break + } + v.reset(OpS390XLGDR) + v.Type = t + v0 := b.NewValue0(v.Pos, OpS390XLPDFR, x.Type) + v0.AddArg(x) + v.AddArg(v0) + return true + } + return false +} func rewriteValueS390X_OpS390XRLL(v *Value) bool { v_1 := v.Args[1] v_0 := v.Args[0] @@ -13002,15 +13149,15 @@ func rewriteValueS390X_OpS390XRLLG(v *Value) bool { v_1 := v.Args[1] v_0 := v.Args[0] // match: (RLLG x (MOVDconst [c])) - // result: (RLLGconst x [int8(c&63)]) + // result: (RISBGZ x {s390x.NewRotateParams(0, 63, int8(c&63))}) for { x := v_0 if v_1.Op != OpS390XMOVDconst { break } c := auxIntToInt64(v_1.AuxInt) - v.reset(OpS390XRLLGconst) - v.AuxInt = int8ToAuxInt(int8(c & 63)) + v.reset(OpS390XRISBGZ) + v.Aux = s390xRotateParamsToAux(s390x.NewRotateParams(0, 63, int8(c&63))) v.AddArg(x) return true } @@ -13034,6 +13181,23 @@ func rewriteValueS390X_OpS390XSLD(v *Value) bool { v.AddArg(x) return true } + // match: (SLD x (RISBGZ y {r})) + // cond: r.Amount == 0 && r.OutMask()&63 == 63 + // result: (SLD x y) + for { + x := v_0 + if v_1.Op != OpS390XRISBGZ { + break + } + r := auxToS390xRotateParams(v_1.Aux) + y := v_1.Args[0] + if !(r.Amount == 0 && r.OutMask()&63 == 63) { + break + } + v.reset(OpS390XSLD) + v.AddArg2(x, y) + return true + } // match: (SLD x (AND (MOVDconst [c]) y)) // result: (SLD x (ANDWconst [int32(c&63)] y)) for { @@ -13152,6 +13316,38 @@ func rewriteValueS390X_OpS390XSLD(v *Value) bool { } func rewriteValueS390X_OpS390XSLDconst(v *Value) bool { v_0 := v.Args[0] + // match: (SLDconst (SRDconst x [c]) [d]) + // result: (RISBGZ x {s390x.NewRotateParams(max8(0, c-d), 63-d, (d-c)&63)}) + for { + d := auxIntToInt8(v.AuxInt) + if v_0.Op != OpS390XSRDconst { + break + } + c := auxIntToInt8(v_0.AuxInt) + x := v_0.Args[0] + v.reset(OpS390XRISBGZ) + v.Aux = s390xRotateParamsToAux(s390x.NewRotateParams(max8(0, c-d), 63-d, (d-c)&63)) + v.AddArg(x) + return true + } + // match: (SLDconst (RISBGZ x {r}) [c]) + // cond: s390x.NewRotateParams(0, 63-c, c).InMerge(r.OutMask()) != nil + // result: (RISBGZ x {(*s390x.NewRotateParams(0, 63-c, c).InMerge(r.OutMask())).RotateLeft(r.Amount)}) + for { + c := auxIntToInt8(v.AuxInt) + if v_0.Op != OpS390XRISBGZ { + break + } + r := auxToS390xRotateParams(v_0.Aux) + x := v_0.Args[0] + if !(s390x.NewRotateParams(0, 63-c, c).InMerge(r.OutMask()) != nil) { + break + } + v.reset(OpS390XRISBGZ) + v.Aux = s390xRotateParamsToAux((*s390x.NewRotateParams(0, 63-c, c).InMerge(r.OutMask())).RotateLeft(r.Amount)) + v.AddArg(x) + return true + } // match: (SLDconst x [0]) // result: x for { @@ -13170,18 +13366,54 @@ func rewriteValueS390X_OpS390XSLW(v *Value) bool { b := v.Block typ := &b.Func.Config.Types // match: (SLW x (MOVDconst [c])) - // result: (SLWconst x [int8(c&63)]) + // cond: c&32 == 0 + // result: (SLWconst x [int8(c&31)]) for { x := v_0 if v_1.Op != OpS390XMOVDconst { break } c := auxIntToInt64(v_1.AuxInt) + if !(c&32 == 0) { + break + } v.reset(OpS390XSLWconst) - v.AuxInt = int8ToAuxInt(int8(c & 63)) + v.AuxInt = int8ToAuxInt(int8(c & 31)) v.AddArg(x) return true } + // match: (SLW _ (MOVDconst [c])) + // cond: c&32 != 0 + // result: (MOVDconst [0]) + for { + if v_1.Op != OpS390XMOVDconst { + break + } + c := auxIntToInt64(v_1.AuxInt) + if !(c&32 != 0) { + break + } + v.reset(OpS390XMOVDconst) + v.AuxInt = int64ToAuxInt(0) + return true + } + // match: (SLW x (RISBGZ y {r})) + // cond: r.Amount == 0 && r.OutMask()&63 == 63 + // result: (SLW x y) + for { + x := v_0 + if v_1.Op != OpS390XRISBGZ { + break + } + r := auxToS390xRotateParams(v_1.Aux) + y := v_1.Args[0] + if !(r.Amount == 0 && r.OutMask()&63 == 63) { + break + } + v.reset(OpS390XSLW) + v.AddArg2(x, y) + return true + } // match: (SLW x (AND (MOVDconst [c]) y)) // result: (SLW x (ANDWconst [int32(c&63)] y)) for { @@ -13330,6 +13562,23 @@ func rewriteValueS390X_OpS390XSRAD(v *Value) bool { v.AddArg(x) return true } + // match: (SRAD x (RISBGZ y {r})) + // cond: r.Amount == 0 && r.OutMask()&63 == 63 + // result: (SRAD x y) + for { + x := v_0 + if v_1.Op != OpS390XRISBGZ { + break + } + r := auxToS390xRotateParams(v_1.Aux) + y := v_1.Args[0] + if !(r.Amount == 0 && r.OutMask()&63 == 63) { + break + } + v.reset(OpS390XSRAD) + v.AddArg2(x, y) + return true + } // match: (SRAD x (AND (MOVDconst [c]) y)) // result: (SRAD x (ANDWconst [int32(c&63)] y)) for { @@ -13478,18 +13727,56 @@ func rewriteValueS390X_OpS390XSRAW(v *Value) bool { b := v.Block typ := &b.Func.Config.Types // match: (SRAW x (MOVDconst [c])) - // result: (SRAWconst x [int8(c&63)]) + // cond: c&32 == 0 + // result: (SRAWconst x [int8(c&31)]) for { x := v_0 if v_1.Op != OpS390XMOVDconst { break } c := auxIntToInt64(v_1.AuxInt) + if !(c&32 == 0) { + break + } v.reset(OpS390XSRAWconst) - v.AuxInt = int8ToAuxInt(int8(c & 63)) + v.AuxInt = int8ToAuxInt(int8(c & 31)) v.AddArg(x) return true } + // match: (SRAW x (MOVDconst [c])) + // cond: c&32 != 0 + // result: (SRAWconst x [31]) + for { + x := v_0 + if v_1.Op != OpS390XMOVDconst { + break + } + c := auxIntToInt64(v_1.AuxInt) + if !(c&32 != 0) { + break + } + v.reset(OpS390XSRAWconst) + v.AuxInt = int8ToAuxInt(31) + v.AddArg(x) + return true + } + // match: (SRAW x (RISBGZ y {r})) + // cond: r.Amount == 0 && r.OutMask()&63 == 63 + // result: (SRAW x y) + for { + x := v_0 + if v_1.Op != OpS390XRISBGZ { + break + } + r := auxToS390xRotateParams(v_1.Aux) + y := v_1.Args[0] + if !(r.Amount == 0 && r.OutMask()&63 == 63) { + break + } + v.reset(OpS390XSRAW) + v.AddArg2(x, y) + return true + } // match: (SRAW x (AND (MOVDconst [c]) y)) // result: (SRAW x (ANDWconst [int32(c&63)] y)) for { @@ -13650,6 +13937,23 @@ func rewriteValueS390X_OpS390XSRD(v *Value) bool { v.AddArg(x) return true } + // match: (SRD x (RISBGZ y {r})) + // cond: r.Amount == 0 && r.OutMask()&63 == 63 + // result: (SRD x y) + for { + x := v_0 + if v_1.Op != OpS390XRISBGZ { + break + } + r := auxToS390xRotateParams(v_1.Aux) + y := v_1.Args[0] + if !(r.Amount == 0 && r.OutMask()&63 == 63) { + break + } + v.reset(OpS390XSRD) + v.AddArg2(x, y) + return true + } // match: (SRD x (AND (MOVDconst [c]) y)) // result: (SRD x (ANDWconst [int32(c&63)] y)) for { @@ -13768,24 +14072,36 @@ func rewriteValueS390X_OpS390XSRD(v *Value) bool { } func rewriteValueS390X_OpS390XSRDconst(v *Value) bool { v_0 := v.Args[0] - b := v.Block - // match: (SRDconst [1] (SLDconst [1] (LGDR x))) - // result: (LGDR (LPDFR x)) + // match: (SRDconst (SLDconst x [c]) [d]) + // result: (RISBGZ x {s390x.NewRotateParams(d, min8(63, 63-c+d), (c-d)&63)}) for { - if auxIntToInt8(v.AuxInt) != 1 || v_0.Op != OpS390XSLDconst || auxIntToInt8(v_0.AuxInt) != 1 { + d := auxIntToInt8(v.AuxInt) + if v_0.Op != OpS390XSLDconst { break } - v_0_0 := v_0.Args[0] - if v_0_0.Op != OpS390XLGDR { + c := auxIntToInt8(v_0.AuxInt) + x := v_0.Args[0] + v.reset(OpS390XRISBGZ) + v.Aux = s390xRotateParamsToAux(s390x.NewRotateParams(d, min8(63, 63-c+d), (c-d)&63)) + v.AddArg(x) + return true + } + // match: (SRDconst (RISBGZ x {r}) [c]) + // cond: s390x.NewRotateParams(c, 63, -c&63).InMerge(r.OutMask()) != nil + // result: (RISBGZ x {(*s390x.NewRotateParams(c, 63, -c&63).InMerge(r.OutMask())).RotateLeft(r.Amount)}) + for { + c := auxIntToInt8(v.AuxInt) + if v_0.Op != OpS390XRISBGZ { break } - t := v_0_0.Type - x := v_0_0.Args[0] - v.reset(OpS390XLGDR) - v.Type = t - v0 := b.NewValue0(v.Pos, OpS390XLPDFR, x.Type) - v0.AddArg(x) - v.AddArg(v0) + r := auxToS390xRotateParams(v_0.Aux) + x := v_0.Args[0] + if !(s390x.NewRotateParams(c, 63, -c&63).InMerge(r.OutMask()) != nil) { + break + } + v.reset(OpS390XRISBGZ) + v.Aux = s390xRotateParamsToAux((*s390x.NewRotateParams(c, 63, -c&63).InMerge(r.OutMask())).RotateLeft(r.Amount)) + v.AddArg(x) return true } // match: (SRDconst x [0]) @@ -13806,18 +14122,54 @@ func rewriteValueS390X_OpS390XSRW(v *Value) bool { b := v.Block typ := &b.Func.Config.Types // match: (SRW x (MOVDconst [c])) - // result: (SRWconst x [int8(c&63)]) + // cond: c&32 == 0 + // result: (SRWconst x [int8(c&31)]) for { x := v_0 if v_1.Op != OpS390XMOVDconst { break } c := auxIntToInt64(v_1.AuxInt) + if !(c&32 == 0) { + break + } v.reset(OpS390XSRWconst) - v.AuxInt = int8ToAuxInt(int8(c & 63)) + v.AuxInt = int8ToAuxInt(int8(c & 31)) v.AddArg(x) return true } + // match: (SRW _ (MOVDconst [c])) + // cond: c&32 != 0 + // result: (MOVDconst [0]) + for { + if v_1.Op != OpS390XMOVDconst { + break + } + c := auxIntToInt64(v_1.AuxInt) + if !(c&32 != 0) { + break + } + v.reset(OpS390XMOVDconst) + v.AuxInt = int64ToAuxInt(0) + return true + } + // match: (SRW x (RISBGZ y {r})) + // cond: r.Amount == 0 && r.OutMask()&63 == 63 + // result: (SRW x y) + for { + x := v_0 + if v_1.Op != OpS390XRISBGZ { + break + } + r := auxToS390xRotateParams(v_1.Aux) + y := v_1.Args[0] + if !(r.Amount == 0 && r.OutMask()&63 == 63) { + break + } + v.reset(OpS390XSRW) + v.AddArg2(x, y) + return true + } // match: (SRW x (AND (MOVDconst [c]) y)) // result: (SRW x (ANDWconst [int32(c&63)] y)) for { @@ -14564,9 +14916,8 @@ func rewriteValueS390X_OpS390XXOR(v *Value) bool { } break } - // match: (XOR (SLDconst x [c]) (SRDconst x [d])) - // cond: d == 64-c - // result: (RLLGconst [c] x) + // match: (XOR (SLDconst x [c]) (SRDconst x [64-c])) + // result: (RISBGZ x {s390x.NewRotateParams(0, 63, c)}) for { for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 { if v_0.Op != OpS390XSLDconst { @@ -14574,15 +14925,11 @@ func rewriteValueS390X_OpS390XXOR(v *Value) bool { } c := auxIntToInt8(v_0.AuxInt) x := v_0.Args[0] - if v_1.Op != OpS390XSRDconst { - continue - } - d := auxIntToInt8(v_1.AuxInt) - if x != v_1.Args[0] || !(d == 64-c) { + if v_1.Op != OpS390XSRDconst || auxIntToInt8(v_1.AuxInt) != 64-c || x != v_1.Args[0] { continue } - v.reset(OpS390XRLLGconst) - v.AuxInt = int8ToAuxInt(c) + v.reset(OpS390XRISBGZ) + v.Aux = s390xRotateParamsToAux(s390x.NewRotateParams(0, 63, c)) v.AddArg(x) return true } @@ -14665,9 +15012,8 @@ func rewriteValueS390X_OpS390XXORW(v *Value) bool { } break } - // match: (XORW (SLWconst x [c]) (SRWconst x [d])) - // cond: d == 32-c - // result: (RLLconst [c] x) + // match: (XORW (SLWconst x [c]) (SRWconst x [32-c])) + // result: (RLLconst x [c]) for { for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 { if v_0.Op != OpS390XSLWconst { @@ -14675,11 +15021,7 @@ func rewriteValueS390X_OpS390XXORW(v *Value) bool { } c := auxIntToInt8(v_0.AuxInt) x := v_0.Args[0] - if v_1.Op != OpS390XSRWconst { - continue - } - d := auxIntToInt8(v_1.AuxInt) - if x != v_1.Args[0] || !(d == 32-c) { + if v_1.Op != OpS390XSRWconst || auxIntToInt8(v_1.AuxInt) != 32-c || x != v_1.Args[0] { continue } v.reset(OpS390XRLLconst) diff --git a/src/cmd/internal/obj/s390x/rotate.go b/src/cmd/internal/obj/s390x/rotate.go index fd2d5482db..7dbc45e648 100644 --- a/src/cmd/internal/obj/s390x/rotate.go +++ b/src/cmd/internal/obj/s390x/rotate.go @@ -4,6 +4,10 @@ package s390x +import ( + "math/bits" +) + // RotateParams represents the immediates required for a "rotate // then ... selected bits instruction". // @@ -24,12 +28,18 @@ package s390x // input left by. Note that this rotation is performed // before the masked region is used. type RotateParams struct { - Start uint8 // big-endian start bit index [0..63] - End uint8 // big-endian end bit index [0..63] - Amount uint8 // amount to rotate left + Start int8 // big-endian start bit index [0..63] + End int8 // big-endian end bit index [0..63] + Amount int8 // amount to rotate left } -func NewRotateParams(start, end, amount int64) RotateParams { +// NewRotateParams creates a set of parameters representing a +// rotation left by the amount provided and a selection of the bits +// between the provided start and end indexes (inclusive). +// +// The start and end indexes and the rotation amount must all +// be in the range 0-63 inclusive or this function will panic. +func NewRotateParams(start, end, amount int8) RotateParams { if start&^63 != 0 { panic("start out of bounds") } @@ -40,8 +50,66 @@ func NewRotateParams(start, end, amount int64) RotateParams { panic("amount out of bounds") } return RotateParams{ - Start: uint8(start), - End: uint8(end), - Amount: uint8(amount), + Start: start, + End: end, + Amount: amount, } } + +// RotateLeft generates a new set of parameters with the rotation amount +// increased by the given value. The selected bits are left unchanged. +func (r RotateParams) RotateLeft(amount int8) RotateParams { + r.Amount += amount + r.Amount &= 63 + return r +} + +// OutMask provides a mask representing the selected bits. +func (r RotateParams) OutMask() uint64 { + // Note: z must be unsigned for bootstrap compiler + z := uint8(63-r.End+r.Start) & 63 // number of zero bits in mask + return bits.RotateLeft64(^uint64(0)<> 1, outMask: ^uint64(0) >> 1}, + {start: 0, end: 62, amount: 0, inMask: ^uint64(1), outMask: ^uint64(1)}, + {start: 1, end: 62, amount: 0, inMask: ^uint64(3) >> 1, outMask: ^uint64(3) >> 1}, + + // end before start, no rotation + {start: 63, end: 0, amount: 0, inMask: 1<<63 | 1, outMask: 1<<63 | 1}, + {start: 62, end: 0, amount: 0, inMask: 1<<63 | 3, outMask: 1<<63 | 3}, + {start: 63, end: 1, amount: 0, inMask: 3<<62 | 1, outMask: 3<<62 | 1}, + {start: 62, end: 1, amount: 0, inMask: 3<<62 | 3, outMask: 3<<62 | 3}, + + // rotation + {start: 32, end: 63, amount: 32, inMask: 0xffffffff00000000, outMask: 0x00000000ffffffff}, + {start: 48, end: 15, amount: 16, inMask: 0xffffffff00000000, outMask: 0xffff00000000ffff}, + {start: 0, end: 7, amount: -8 & 63, inMask: 0xff, outMask: 0xff << 56}, + } + for i, test := range tests { + r := NewRotateParams(test.start, test.end, test.amount) + if m := r.OutMask(); m != test.outMask { + t.Errorf("out mask %v: want %#x, got %#x", i, test.outMask, m) + } + if m := r.InMask(); m != test.inMask { + t.Errorf("in mask %v: want %#x, got %#x", i, test.inMask, m) + } + } +} + +func TestRotateParamsMerge(t *testing.T) { + tests := []struct { + // inputs + src RotateParams + mask uint64 + + // results + in *RotateParams + out *RotateParams + }{ + { + src: RotateParams{Start: 48, End: 15, Amount: 16}, + mask: 0xffffffffffffffff, + in: &RotateParams{Start: 48, End: 15, Amount: 16}, + out: &RotateParams{Start: 48, End: 15, Amount: 16}, + }, + { + src: RotateParams{Start: 16, End: 47, Amount: 0}, + mask: 0x00000000ffffffff, + in: &RotateParams{Start: 32, End: 47, Amount: 0}, + out: &RotateParams{Start: 32, End: 47, Amount: 0}, + }, + { + src: RotateParams{Start: 16, End: 47, Amount: 0}, + mask: 0xffff00000000ffff, + in: nil, + out: nil, + }, + { + src: RotateParams{Start: 0, End: 63, Amount: 0}, + mask: 0xf7f0000000000000, + in: nil, + out: nil, + }, + { + src: RotateParams{Start: 0, End: 63, Amount: 1}, + mask: 0x000000000000ff00, + in: &RotateParams{Start: 47, End: 54, Amount: 1}, + out: &RotateParams{Start: 48, End: 55, Amount: 1}, + }, + { + src: RotateParams{Start: 32, End: 63, Amount: 32}, + mask: 0xffff00000000ffff, + in: &RotateParams{Start: 32, End: 47, Amount: 32}, + out: &RotateParams{Start: 48, End: 63, Amount: 32}, + }, + { + src: RotateParams{Start: 0, End: 31, Amount: 32}, + mask: 0x8000000000000000, + in: nil, + out: &RotateParams{Start: 0, End: 0, Amount: 32}, + }, + { + src: RotateParams{Start: 0, End: 31, Amount: 32}, + mask: 0x0000000080000000, + in: &RotateParams{Start: 0, End: 0, Amount: 32}, + out: nil, + }, + } + + eq := func(x, y *RotateParams) bool { + if x == nil && y == nil { + return true + } + if x == nil || y == nil { + return false + } + return *x == *y + } + + for _, test := range tests { + if r := test.src.InMerge(test.mask); !eq(r, test.in) { + t.Errorf("%v merged with %#x (input): want %v, got %v", test.src, test.mask, test.in, r) + } + if r := test.src.OutMerge(test.mask); !eq(r, test.out) { + t.Errorf("%v merged with %#x (output): want %v, got %v", test.src, test.mask, test.out, r) + } + } +} diff --git a/test/codegen/bitfield.go b/test/codegen/bitfield.go index 08788f1447..7abc1c2783 100644 --- a/test/codegen/bitfield.go +++ b/test/codegen/bitfield.go @@ -127,11 +127,13 @@ func sbfx6(x int32) int32 { // ubfiz func ubfiz1(x uint64) uint64 { // arm64:"UBFIZ\t[$]3, R[0-9]+, [$]12",-"LSL",-"AND" + // s390x:"RISBGZ\t[$]49, [$]60, [$]3,",-"SLD",-"AND" return (x & 0xfff) << 3 } func ubfiz2(x uint64) uint64 { // arm64:"UBFIZ\t[$]4, R[0-9]+, [$]12",-"LSL",-"AND" + // s390x:"RISBGZ\t[$]48, [$]59, [$]4,",-"SLD",-"AND" return (x << 4) & 0xfff0 } @@ -149,6 +151,7 @@ func ubfiz5(x uint8) uint64 { func ubfiz6(x uint64) uint64 { // arm64:"UBFIZ\t[$]1, R[0-9]+, [$]60",-"LSL",-"LSR" + // s390x:"RISBGZ\t[$]3, [$]62, [$]1, ",-"SLD",-"SRD" return (x << 4) >> 3 } @@ -159,6 +162,7 @@ func ubfiz7(x uint32) uint32 { func ubfiz8(x uint64) uint64 { // arm64:"UBFIZ\t[$]1, R[0-9]+, [$]20",-"LSL",-"LSR" + // s390x:"RISBGZ\t[$]43, [$]62, [$]1, ",-"SLD",-"SRD",-"AND" return ((x & 0xfffff) << 4) >> 3 } @@ -169,17 +173,20 @@ func ubfiz9(x uint64) uint64 { func ubfiz10(x uint64) uint64 { // arm64:"UBFIZ\t[$]7, R[0-9]+, [$]12",-"LSL",-"LSR",-"AND" + // s390x:"RISBGZ\t[$]45, [$]56, [$]7, ",-"SLD",-"SRD",-"AND" return ((x << 5) & (0xfff << 5)) << 2 } // ubfx func ubfx1(x uint64) uint64 { // arm64:"UBFX\t[$]25, R[0-9]+, [$]10",-"LSR",-"AND" + // s390x:"RISBGZ\t[$]54, [$]63, [$]39, ",-"SRD",-"AND" return (x >> 25) & 1023 } func ubfx2(x uint64) uint64 { // arm64:"UBFX\t[$]4, R[0-9]+, [$]8",-"LSR",-"AND" + // s390x:"RISBGZ\t[$]56, [$]63, [$]60, ",-"SRD",-"AND" return (x & 0x0ff0) >> 4 } @@ -196,30 +203,37 @@ func ubfx5(x uint8) uint64 { } func ubfx6(x uint64) uint64 { - return (x << 1) >> 2 // arm64:"UBFX\t[$]1, R[0-9]+, [$]62",-"LSL",-"LSR" + // arm64:"UBFX\t[$]1, R[0-9]+, [$]62",-"LSL",-"LSR" + // s390x:"RISBGZ\t[$]2, [$]63, [$]63,",-"SLD",-"SRD" + return (x << 1) >> 2 } func ubfx7(x uint32) uint32 { - return (x << 1) >> 2 // arm64:"UBFX\t[$]1, R[0-9]+, [$]30",-"LSL",-"LSR" + // arm64:"UBFX\t[$]1, R[0-9]+, [$]30",-"LSL",-"LSR" + return (x << 1) >> 2 } func ubfx8(x uint64) uint64 { // arm64:"UBFX\t[$]1, R[0-9]+, [$]12",-"LSL",-"LSR",-"AND" + // s390x:"RISBGZ\t[$]52, [$]63, [$]63,",-"SLD",-"SRD",-"AND" return ((x << 1) >> 2) & 0xfff } func ubfx9(x uint64) uint64 { // arm64:"UBFX\t[$]4, R[0-9]+, [$]11",-"LSL",-"LSR",-"AND" + // s390x:"RISBGZ\t[$]53, [$]63, [$]60, ",-"SLD",-"SRD",-"AND" return ((x >> 3) & 0xfff) >> 1 } func ubfx10(x uint64) uint64 { // arm64:"UBFX\t[$]5, R[0-9]+, [$]56",-"LSL",-"LSR" + // s390x:"RISBGZ\t[$]8, [$]63, [$]59, ",-"SLD",-"SRD" return ((x >> 2) << 5) >> 8 } func ubfx11(x uint64) uint64 { // arm64:"UBFX\t[$]1, R[0-9]+, [$]19",-"LSL",-"LSR" + // s390x:"RISBGZ\t[$]45, [$]63, [$]63, ",-"SLD",-"SRD",-"AND" return ((x & 0xfffff) << 3) >> 4 } diff --git a/test/codegen/bits.go b/test/codegen/bits.go index 398dd84e9e..56e0f3474e 100644 --- a/test/codegen/bits.go +++ b/test/codegen/bits.go @@ -340,3 +340,15 @@ func bitSetTest(x int) bool { // amd64:"CMPQ\tAX, [$]9" return x&9 == 9 } + +// mask contiguous one bits +func cont1Mask64U(x uint64) uint64 { + // s390x:"RISBGZ\t[$]16, [$]47, [$]0," + return x&0x0000ffffffff0000 +} + +// mask contiguous zero bits +func cont0Mask64U(x uint64) uint64 { + // s390x:"RISBGZ\t[$]48, [$]15, [$]0," + return x&0xffff00000000ffff +} diff --git a/test/codegen/mathbits.go b/test/codegen/mathbits.go index 4c35f26997..fff6639546 100644 --- a/test/codegen/mathbits.go +++ b/test/codegen/mathbits.go @@ -213,7 +213,7 @@ func RotateLeft64(n uint64) uint64 { // arm64:"ROR" // ppc64:"ROTL" // ppc64le:"ROTL" - // s390x:"RLLG" + // s390x:"RISBGZ\t[$]0, [$]63, [$]37, " // wasm:"I64Rotl" return bits.RotateLeft64(n, 37) } diff --git a/test/codegen/rotate.go b/test/codegen/rotate.go index 0c8b030970..e0bcd0abbc 100644 --- a/test/codegen/rotate.go +++ b/test/codegen/rotate.go @@ -17,21 +17,21 @@ func rot64(x uint64) uint64 { // amd64:"ROLQ\t[$]7" // arm64:"ROR\t[$]57" - // s390x:"RLLG\t[$]7" + // s390x:"RISBGZ\t[$]0, [$]63, [$]7, " // ppc64:"ROTL\t[$]7" // ppc64le:"ROTL\t[$]7" a += x<<7 | x>>57 // amd64:"ROLQ\t[$]8" // arm64:"ROR\t[$]56" - // s390x:"RLLG\t[$]8" + // s390x:"RISBGZ\t[$]0, [$]63, [$]8, " // ppc64:"ROTL\t[$]8" // ppc64le:"ROTL\t[$]8" a += x<<8 + x>>56 // amd64:"ROLQ\t[$]9" // arm64:"ROR\t[$]55" - // s390x:"RLLG\t[$]9" + // s390x:"RISBGZ\t[$]0, [$]63, [$]9, " // ppc64:"ROTL\t[$]9" // ppc64le:"ROTL\t[$]9" a += x<<9 ^ x>>55 diff --git a/test/codegen/shift.go b/test/codegen/shift.go index a45f27c9cf..d19a1984c1 100644 --- a/test/codegen/shift.go +++ b/test/codegen/shift.go @@ -11,84 +11,84 @@ package codegen // ------------------ // func lshMask64x64(v int64, s uint64) int64 { - // s390x:-".*AND",-".*MOVDGE" + // s390x:-"RISBGZ",-"AND",-"LOCGR" // ppc64le:"ANDCC",-"ORN",-"ISEL" // ppc64:"ANDCC",-"ORN",-"ISEL" return v << (s & 63) } func rshMask64Ux64(v uint64, s uint64) uint64 { - // s390x:-".*AND",-".*MOVDGE" + // s390x:-"RISBGZ",-"AND",-"LOCGR" // ppc64le:"ANDCC",-"ORN",-"ISEL" // ppc64:"ANDCC",-"ORN",-"ISEL" return v >> (s & 63) } func rshMask64x64(v int64, s uint64) int64 { - // s390x:-".*AND",-".*MOVDGE" + // s390x:-"RISBGZ",-"AND",-"LOCGR" // ppc64le:"ANDCC",-ORN",-"ISEL" // ppc64:"ANDCC",-"ORN",-"ISEL" return v >> (s & 63) } func lshMask32x64(v int32, s uint64) int32 { - // s390x:-".*AND",-".*MOVDGE" + // s390x:-"RISBGZ",-"AND",-"LOCGR" // ppc64le:"ISEL",-"ORN" // ppc64:"ISEL",-"ORN" return v << (s & 63) } func rshMask32Ux64(v uint32, s uint64) uint32 { - // s390x:-".*AND",-".*MOVDGE" + // s390x:-"RISBGZ",-"AND",-"LOCGR" // ppc64le:"ISEL",-"ORN" // ppc64:"ISEL",-"ORN" return v >> (s & 63) } func rshMask32x64(v int32, s uint64) int32 { - // s390x:-".*AND",-".*MOVDGE" + // s390x:-"RISBGZ",-"AND",-"LOCGR" // ppc64le:"ISEL",-"ORN" // ppc64:"ISEL",-"ORN" return v >> (s & 63) } func lshMask64x32(v int64, s uint32) int64 { - // s390x:-".*AND",-".*MOVDGE" + // s390x:-"RISBGZ",-"AND",-"LOCGR" // ppc64le:"ANDCC",-"ORN" // ppc64:"ANDCC",-"ORN" return v << (s & 63) } func rshMask64Ux32(v uint64, s uint32) uint64 { - // s390x:-".*AND",-".*MOVDGE" + // s390x:-"RISBGZ",-"AND",-"LOCGR" // ppc64le:"ANDCC",-"ORN" // ppc64:"ANDCC",-"ORN" return v >> (s & 63) } func rshMask64x32(v int64, s uint32) int64 { - // s390x:-".*AND",-".*MOVDGE" + // s390x:-"RISBGZ",-"AND",-"LOCGR" // ppc64le:"ANDCC",-"ORN",-"ISEL" // ppc64:"ANDCC",-"ORN",-"ISEL" return v >> (s & 63) } func lshMask64x32Ext(v int64, s int32) int64 { - // s390x:-".*AND",-".*MOVDGE" + // s390x:-"RISBGZ",-"AND",-"LOCGR" // ppc64le:"ANDCC",-"ORN",-"ISEL" // ppc64:"ANDCC",-"ORN",-"ISEL" return v << uint(s&63) } func rshMask64Ux32Ext(v uint64, s int32) uint64 { - // s390x:-".*AND",-".*MOVDGE" + // s390x:-"RISBGZ",-"AND",-"LOCGR" // ppc64le:"ANDCC",-"ORN",-"ISEL" // ppc64:"ANDCC",-"ORN",-"ISEL" return v >> uint(s&63) } func rshMask64x32Ext(v int64, s int32) int64 { - // s390x:-".*AND",-".*MOVDGE" + // s390x:-"RISBGZ",-"AND",-"LOCGR" // ppc64le:"ANDCC",-"ORN",-"ISEL" // ppc64:"ANDCC",-"ORN",-"ISEL" return v >> uint(s&63) @@ -128,7 +128,8 @@ func lshSignedMasked(v8 int8, v16 int16, v32 int32, v64 int64, x int) { func rshGuarded64(v int64, s uint) int64 { if s < 64 { - // s390x:-".*AND",-".*MOVDGE" wasm:-"Select",-".*LtU" + // s390x:-"RISBGZ",-"AND",-"LOCGR" + // wasm:-"Select",-".*LtU" return v >> s } panic("shift too large") @@ -136,7 +137,8 @@ func rshGuarded64(v int64, s uint) int64 { func rshGuarded64U(v uint64, s uint) uint64 { if s < 64 { - // s390x:-".*AND",-".*MOVDGE" wasm:-"Select",-".*LtU" + // s390x:-"RISBGZ",-"AND",-"LOCGR" + // wasm:-"Select",-".*LtU" return v >> s } panic("shift too large") @@ -144,7 +146,8 @@ func rshGuarded64U(v uint64, s uint) uint64 { func lshGuarded64(v int64, s uint) int64 { if s < 64 { - // s390x:-".*AND",-".*MOVDGE" wasm:-"Select",-".*LtU" + // s390x:-"RISBGZ",-"AND",-"LOCGR" + // wasm:-"Select",-".*LtU" return v << s } panic("shift too large") -- cgit v1.3 From 189931296f6b56090d9d7f49b7936b817189d87d Mon Sep 17 00:00:00 2001 From: Michael Munday Date: Tue, 10 Nov 2020 02:31:28 -0800 Subject: cmd/internal/obj/s390x: fix SYNC instruction encoding SYNC is supposed to correspond to 'fast-BCR-serialization' which is encoded as 'bcr 14,0'. In CL 197178 I accidentally modified the encoding to 'bcr 7,0' which is a no-op. This CL reverses that change. Fixes #42479. Change-Id: I9918d93d720f5e12acc3014cde20d2d32cc87ee5 Reviewed-on: https://go-review.googlesource.com/c/go/+/268797 Run-TryBot: Michael Munday Trust: Michael Munday TryBot-Result: Go Bot Reviewed-by: Cherry Zhang --- src/cmd/asm/internal/asm/testdata/s390x.s | 2 ++ src/cmd/internal/obj/s390x/asmz.go | 2 +- 2 files changed, 3 insertions(+), 1 deletion(-) (limited to 'src/cmd/internal/obj/s390x') diff --git a/src/cmd/asm/internal/asm/testdata/s390x.s b/src/cmd/asm/internal/asm/testdata/s390x.s index 03b84cfa62..7c5d26be33 100644 --- a/src/cmd/asm/internal/asm/testdata/s390x.s +++ b/src/cmd/asm/internal/asm/testdata/s390x.s @@ -412,6 +412,8 @@ TEXT main·foo(SB),DUPOK|NOSPLIT,$16-0 // TEXT main.foo(SB), DUPOK|NOSPLIT, $16- UNDEF // 00000000 NOPH // 0700 + SYNC // 07e0 + // vector add and sub instructions VAB V3, V4, V4 // e743400000f3 VAH V3, V4, V4 // e743400010f3 diff --git a/src/cmd/internal/obj/s390x/asmz.go b/src/cmd/internal/obj/s390x/asmz.go index f0f9d5cefc..06921085c9 100644 --- a/src/cmd/internal/obj/s390x/asmz.go +++ b/src/cmd/internal/obj/s390x/asmz.go @@ -3700,7 +3700,7 @@ func (c *ctxtz) asmout(p *obj.Prog, asm *[]byte) { } case 80: // sync - zRR(op_BCR, uint32(NotEqual), 0, asm) + zRR(op_BCR, 14, 0, asm) // fast-BCR-serialization case 81: // float to fixed and fixed to float moves (no conversion) switch p.As { -- cgit v1.3