compiler: implement clear builtin for slices

6 files changed, 94 insertions, 20 deletions

diff --git a/compiler/compiler.go b/compiler/compiler.go
index 46c3a0672..6af6debe8 100644
--- a/compiler/compiler.go
+++ b/compiler/compiler.go
@@ -1600,6 +1600,41 @@ func (b *builder) createBuiltin(argTypes []types.Type, argValues []llvm.Value, c
 		cplx = b.CreateInsertValue(cplx, r, 0, "")
 		cplx = b.CreateInsertValue(cplx, i, 1, "")
 		return cplx, nil
+	case "clear":
+		value := argValues[0]
+		switch typ := argTypes[0].Underlying().(type) {
+		case *types.Slice:
+			elementType := b.getLLVMType(typ.Elem())
+			elementSize := b.targetData.TypeAllocSize(elementType)
+			elementAlign := b.targetData.ABITypeAlignment(elementType)
+
+			// The pointer to the data to be cleared.
+			llvmBuf := b.CreateExtractValue(value, 0, "buf")
+			if llvmBuf.Type() != b.i8ptrType { // compatibility with LLVM 14
+				llvmBuf = b.CreateBitCast(llvmBuf, b.i8ptrType, "")
+			}
+
+			// The length (in bytes) to be cleared.
+			llvmLen := b.CreateExtractValue(value, 1, "len")
+			llvmLen = b.CreateMul(llvmLen, llvm.ConstInt(llvmLen.Type(), elementSize, false), "")
+
+			// Do the clear operation using the LLVM memset builtin.
+			// This is also correct for nil slices: in those cases, len will be
+			// 0 which means the memset call is a no-op (according to the LLVM
+			// LangRef).
+			memset := b.getMemsetFunc()
+			call := b.createCall(memset.GlobalValueType(), memset, []llvm.Value{
+				llvmBuf, // dest
+				llvm.ConstInt(b.ctx.Int8Type(), 0, false), // val
+				llvmLen, // len
+				llvm.ConstInt(b.ctx.Int1Type(), 0, false), // isVolatile
+			}, "")
+			call.AddCallSiteAttribute(1, b.ctx.CreateEnumAttribute(llvm.AttributeKindID("align"), uint64(elementAlign)))
+
+			return llvm.Value{}, nil
+		default:
+			return llvm.Value{}, b.makeError(pos, "unsupported type in clear builtin: "+typ.String())
+		}
 	case "copy":
 		dst := argValues[0]
 		src := argValues[1]
diff --git a/compiler/intrinsics.go b/compiler/intrinsics.go
index 5761a438f..af3a57de1 100644
--- a/compiler/intrinsics.go
+++ b/compiler/intrinsics.go
@@ -70,15 +70,7 @@ func (b *builder) createMemoryCopyImpl() {
 // regular libc memset calls if they aren't optimized out in a different way.
 func (b *builder) createMemoryZeroImpl() {
 	b.createFunctionStart(true)
-	fnName := "llvm.memset.p0.i" + strconv.Itoa(b.uintptrType.IntTypeWidth())
-	if llvmutil.Major() < 15 { // compatibility with LLVM 14
-		fnName = "llvm.memset.p0i8.i" + strconv.Itoa(b.uintptrType.IntTypeWidth())
-	}
-	llvmFn := b.mod.NamedFunction(fnName)
-	if llvmFn.IsNil() {
-		fnType := llvm.FunctionType(b.ctx.VoidType(), []llvm.Type{b.i8ptrType, b.ctx.Int8Type(), b.uintptrType, b.ctx.Int1Type()}, false)
-		llvmFn = llvm.AddFunction(b.mod, fnName, fnType)
-	}
+	llvmFn := b.getMemsetFunc()
 	params := []llvm.Value{
 		b.getValue(b.fn.Params[0], getPos(b.fn)),
 		llvm.ConstInt(b.ctx.Int8Type(), 0, false),
@@ -89,6 +81,20 @@ func (b *builder) createMemoryZeroImpl() {
 	b.CreateRetVoid()
 }
 
+// Return the llvm.memset.p0.i8 function declaration.
+func (c *compilerContext) getMemsetFunc() llvm.Value {
+	fnName := "llvm.memset.p0.i" + strconv.Itoa(c.uintptrType.IntTypeWidth())
+	if llvmutil.Major() < 15 { // compatibility with LLVM 14
+		fnName = "llvm.memset.p0i8.i" + strconv.Itoa(c.uintptrType.IntTypeWidth())
+	}
+	llvmFn := c.mod.NamedFunction(fnName)
+	if llvmFn.IsNil() {
+		fnType := llvm.FunctionType(c.ctx.VoidType(), []llvm.Type{c.i8ptrType, c.ctx.Int8Type(), c.uintptrType, c.ctx.Int1Type()}, false)
+		llvmFn = llvm.AddFunction(c.mod, fnName, fnType)
+	}
+	return llvmFn
+}
+
 // createKeepAlive creates the runtime.KeepAlive function. It is implemented
 // using inline assembly.
 func (b *builder) createKeepAliveImpl() {
diff --git a/compiler/testdata/go1.21.go b/compiler/testdata/go1.21.go
index 5541b489d..3d92b69b1 100644
--- a/compiler/testdata/go1.21.go
+++ b/compiler/testdata/go1.21.go
@@ -51,3 +51,11 @@ func maxFloat32(a, b float32) float32 {
 func maxString(a, b string) string {
 	return max(a, b)
 }
+
+func clearSlice(s []int) {
+	clear(s)
+}
+
+func clearZeroSizedSlice(s []struct{}) {
+	clear(s)
+}
diff --git a/compiler/testdata/go1.21.ll b/compiler/testdata/go1.21.ll
index 5d4a70197..73a368383 100644
--- a/compiler/testdata/go1.21.ll
+++ b/compiler/testdata/go1.21.ll
@@ -84,10 +84,10 @@ entry:
   %2 = insertvalue %runtime._string zeroinitializer, ptr %b.data, 0
   %3 = insertvalue %runtime._string %2, i32 %b.len, 1
   %stackalloc = alloca i8, align 1
-  %4 = call i1 @runtime.stringLess(ptr %a.data, i32 %a.len, ptr %b.data, i32 %b.len, ptr undef) #4
+  %4 = call i1 @runtime.stringLess(ptr %a.data, i32 %a.len, ptr %b.data, i32 %b.len, ptr undef) #5
   %5 = select i1 %4, %runtime._string %1, %runtime._string %3
   %6 = extractvalue %runtime._string %5, 0
-  call void @runtime.trackPointer(ptr %6, ptr nonnull %stackalloc, ptr undef) #4
+  call void @runtime.trackPointer(ptr %6, ptr nonnull %stackalloc, ptr undef) #5
   ret %runtime._string %5
 }
 
@@ -123,30 +123,48 @@ entry:
   %2 = insertvalue %runtime._string zeroinitializer, ptr %b.data, 0
   %3 = insertvalue %runtime._string %2, i32 %b.len, 1
   %stackalloc = alloca i8, align 1
-  %4 = call i1 @runtime.stringLess(ptr %b.data, i32 %b.len, ptr %a.data, i32 %a.len, ptr undef) #4
+  %4 = call i1 @runtime.stringLess(ptr %b.data, i32 %b.len, ptr %a.data, i32 %a.len, ptr undef) #5
   %5 = select i1 %4, %runtime._string %1, %runtime._string %3
   %6 = extractvalue %runtime._string %5, 0
-  call void @runtime.trackPointer(ptr %6, ptr nonnull %stackalloc, ptr undef) #4
+  call void @runtime.trackPointer(ptr %6, ptr nonnull %stackalloc, ptr undef) #5
   ret %runtime._string %5
 }
 
+; Function Attrs: nounwind
+define hidden void @main.clearSlice(ptr %s.data, i32 %s.len, i32 %s.cap, ptr %context) unnamed_addr #2 {
+entry:
+  %0 = shl i32 %s.len, 2
+  call void @llvm.memset.p0.i32(ptr align 4 %s.data, i8 0, i32 %0, i1 false)
+  ret void
+}
+
+; Function Attrs: argmemonly nocallback nofree nounwind willreturn writeonly
+declare void @llvm.memset.p0.i32(ptr nocapture writeonly, i8, i32, i1 immarg) #3
+
+; Function Attrs: nounwind
+define hidden void @main.clearZeroSizedSlice(ptr %s.data, i32 %s.len, i32 %s.cap, ptr %context) unnamed_addr #2 {
+entry:
+  ret void
+}
+
 ; Function Attrs: nocallback nofree nosync nounwind readnone speculatable willreturn
-declare i32 @llvm.smin.i32(i32, i32) #3
+declare i32 @llvm.smin.i32(i32, i32) #4
 
 ; Function Attrs: nocallback nofree nosync nounwind readnone speculatable willreturn
-declare i8 @llvm.umin.i8(i8, i8) #3
+declare i8 @llvm.umin.i8(i8, i8) #4
 
 ; Function Attrs: nocallback nofree nosync nounwind readnone speculatable willreturn
-declare i32 @llvm.umin.i32(i32, i32) #3
+declare i32 @llvm.umin.i32(i32, i32) #4
 
 ; Function Attrs: nocallback nofree nosync nounwind readnone speculatable willreturn
-declare i32 @llvm.smax.i32(i32, i32) #3
+declare i32 @llvm.smax.i32(i32, i32) #4
 
 ; Function Attrs: nocallback nofree nosync nounwind readnone speculatable willreturn
-declare i32 @llvm.umax.i32(i32, i32) #3
+declare i32 @llvm.umax.i32(i32, i32) #4
 
 attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+nontrapping-fptoint,+sign-ext" }
 attributes #1 = { "target-features"="+bulk-memory,+nontrapping-fptoint,+sign-ext" }
 attributes #2 = { nounwind "target-features"="+bulk-memory,+nontrapping-fptoint,+sign-ext" }
-attributes #3 = { nocallback nofree nosync nounwind readnone speculatable willreturn }
-attributes #4 = { nounwind }
+attributes #3 = { argmemonly nocallback nofree nounwind willreturn writeonly }
+attributes #4 = { nocallback nofree nosync nounwind readnone speculatable willreturn }
+attributes #5 = { nounwind }
diff --git a/testdata/go1.21.go b/testdata/go1.21.go
index 885e588da..184bb2d8a 100644
--- a/testdata/go1.21.go
+++ b/testdata/go1.21.go
@@ -1,6 +1,7 @@
 package main
 
 func main() {
+	// The new min/max builtins.
 	ia := 1
 	ib := 5
 	ic := -3
@@ -9,4 +10,9 @@ func main() {
 	fc := -3.0
 	println("min/max:", min(ia, ib, ic), max(ia, ib, ic))
 	println("min/max:", min(fa, fb, fc), max(fa, fb, fc))
+
+	// The clear builtin, for slices.
+	s := []int{1, 2, 3, 4, 5}
+	clear(s[:3])
+	println("cleared s[:3]:", s[0], s[1], s[2], s[3], s[4])
 }
diff --git a/testdata/go1.21.txt b/testdata/go1.21.txt
index ad81dcfe9..459631a30 100644
--- a/testdata/go1.21.txt
+++ b/testdata/go1.21.txt
@@ -1,2 +1,3 @@
 min/max: -3 5
 min/max: -3.000000e+000 +5.000000e+000
+cleared s[:3]: 0 0 0 4 5