[WIP] ssa: add memcpyopt pass to optimize load/store to memcpy

- Add TargetMachine and DataLayout methods to Program
- Set data layout and target triple on Module creation
- Run memcpyopt pass after IR generation (skip runtime packages)

This optimization converts consecutive load/store operations on aggregate
types to more efficient memcpy/memmove calls.

Fixes goplus#1605

Author: luoliwoshang

internal/build/build.go

@@ -54,6 +54,7 @@ import (
 	"github.com/goplus/llgo/ssa/abi"
 	xenv "github.com/goplus/llgo/xtool/env"
 	"github.com/goplus/llgo/xtool/env/llvm"
+	gollvm "github.com/goplus/llvm"
 
 	llruntime "github.com/goplus/llgo/runtime"
 	llssa "github.com/goplus/llgo/ssa"
@@ -1158,6 +1159,13 @@ func buildPkg(ctx *context, aPkg *aPackage, verbose bool) error {
 
 	ctx.cTransformer.TransformModule(ret.Path(), ret.Module())
 
+	// Run LLVM optimization passes (memcpyopt converts load/store to memcpy)
+	pbo := gollvm.NewPassBuilderOptions()
+	defer pbo.Dispose()
+	if err := ret.Module().RunPasses("memcpyopt", ctx.prog.TargetMachine(), pbo); err != nil {
+		return fmt.Errorf("run LLVM passes failed for %v: %v", pkgPath, err)
+	}
+
 	printCmds := ctx.shouldPrintCommands(verbose)
 	cgoLLFiles, cgoLdflags, err := buildCgo(ctx, aPkg, aPkg.Package.Syntax, externs, printCmds)
 	if err != nil {

ssa/package.go

@@ -119,7 +119,7 @@ type aProgram struct {
 
 	target *Target
 	td     llvm.TargetData
-	// tm  llvm.TargetMachine
+	tm     llvm.TargetMachine
 	named   map[string]Type
 	fnnamed map[string]int
 
@@ -247,7 +247,7 @@ func NewProgram(target *Target) Program {
 		}
 	}
 	ctx := llvm.NewContext()
-	td := target.targetData() // TODO(xsw): target config
+	td, tm := target.targetDataAndMachine()
 	/*
 		arch := target.GOARCH
 		if arch == "" {
@@ -261,7 +261,7 @@ func NewProgram(target *Target) Program {
 	is32Bits := (td.PointerSize() == 4 || is32Bits(target.GOARCH))
 	return &aProgram{
 		ctx: ctx, gocvt: newGoTypes(),
-		target: target, td: td, is32Bits: is32Bits,
+		target: target, td: td, tm: tm, is32Bits: is32Bits,
 		ptrSize: td.PointerSize(), named: make(map[string]Type), fnnamed: make(map[string]int),
 		linkname: make(map[string]string), abiSymbol: make(map[string]Type),
 	}
@@ -275,6 +275,14 @@ func (p Program) TargetData() llvm.TargetData {
 	return p.td
 }
 
+func (p Program) TargetMachine() llvm.TargetMachine {
+	return p.tm
+}
+
+func (p Program) DataLayout() string {
+	return p.td.String()
+}
+
 func (p Program) SetPatch(patchType func(types.Type) types.Type) {
 	p.patchType = patchType
 }
@@ -408,11 +416,9 @@ func (p Program) tyComplex128() llvm.Type {
 // NewPackage creates a new package.
 func (p Program) NewPackage(name, pkgPath string) Package {
 	mod := p.ctx.NewModule(pkgPath)
-	// TODO(lijie): enable target output will check module override, but can't
-	// pass the snapshot test, so disable it for now
-	// if p.target.GOARCH != runtime.GOARCH && p.target.GOOS != runtime.GOOS {
-	// 	mod.SetTarget(p.target.Spec().Triple)
-	// }
+	// Set data layout and target triple for LLVM optimization passes
+	mod.SetDataLayout(p.DataLayout())
+	mod.SetTarget(p.target.Spec().Triple)
 
 	// TODO(xsw): Finalize may cause panic, so comment it.
 	// mod.Finalize()

ssa/target.go

@@ -62,7 +62,7 @@ func (t *Target) CtxRegister() CtxRegister {
 	return CtxRegister{Name: info.Name, Constraint: info.Constraint}
 }
 
-func (p *Target) targetData() llvm.TargetData {
+func (p *Target) targetDataAndMachine() (llvm.TargetData, llvm.TargetMachine) {
 	spec := p.Spec()
 	if spec.Triple == "" {
 		spec.Triple = llvm.DefaultTargetTriple()
@@ -72,30 +72,9 @@ func (p *Target) targetData() llvm.TargetData {
 		panic(err)
 	}
 	machine := t.CreateTargetMachine(spec.Triple, spec.CPU, spec.Features, llvm.CodeGenLevelDefault, llvm.RelocDefault, llvm.CodeModelDefault)
-	return machine.CreateTargetData()
+	return machine.CreateTargetData(), machine
 }
 
-/*
-func (p *Program) targetMachine() llvm.TargetMachine {
-	if p.tm.C == nil {
-		spec := p.target.toSpec()
-		target, err := llvm.GetTargetFromTriple(spec.triple)
-		if err != nil {
-			panic(err)
-		}
-		p.tm = target.CreateTargetMachine(
-			spec.triple,
-			spec.cpu,
-			spec.features,
-			llvm.CodeGenLevelDefault,
-			llvm.RelocDefault,
-			llvm.CodeModelDefault,
-		)
-	}
-	return p.tm
-}
-*/
-
 type TargetSpec struct {
 	Triple   string
 	CPU      string