compiler: insert basic blocks at an appropriate location

For example, this commit moves the 'throw' branch of an assertion (nil
check, slice index check, etc) to the end of the function while
inserting the "continue" branch right after the insert location. This
makes the resulting IR easier to follow.

For some reason, this also reduces code size a bit on average. The
TinyGo smoke tests saw a reduction of 0.22%, mainly from WebAssembly.
The drivers repo saw little average change in code size (-0.01%).

This commit also adds a few compiler tests for the defer keyword.

13 files changed, 272 insertions, 89 deletions

diff --git a/compiler/asserts.go b/compiler/asserts.go
index acd605fab..8d9efdde7 100644
--- a/compiler/asserts.go
+++ b/compiler/asserts.go
@@ -240,8 +240,10 @@ func (b *builder) createRuntimeAssert(assert llvm.Value, blockPrefix, assertFunc
 		}
 	}
 
+	// Put the fault block at the end of the function and the next block at the
+	// current insert position.
 	faultBlock := b.ctx.AddBasicBlock(b.llvmFn, blockPrefix+".throw")
-	nextBlock := b.ctx.AddBasicBlock(b.llvmFn, blockPrefix+".next")
+	nextBlock := b.insertBasicBlock(blockPrefix + ".next")
 	b.blockExits[b.currentBlock] = nextBlock // adjust outgoing block for phi nodes
 
 	// Now branch to the out-of-bounds or the regular block.
diff --git a/compiler/compiler_test.go b/compiler/compiler_test.go
index 9b0c6efff..96797ea70 100644
--- a/compiler/compiler_test.go
+++ b/compiler/compiler_test.go
@@ -49,6 +49,7 @@ func TestCompiler(t *testing.T) {
 		{"float.go", "", ""},
 		{"interface.go", "", ""},
 		{"func.go", "", ""},
+		{"defer.go", "cortex-m-qemu", ""},
 		{"pragma.go", "", ""},
 		{"goroutine.go", "wasm", "asyncify"},
 		{"goroutine.go", "cortex-m-qemu", "tasks"},
diff --git a/compiler/defer.go b/compiler/defer.go
index 1207ce649..235ffaab1 100644
--- a/compiler/defer.go
+++ b/compiler/defer.go
@@ -15,6 +15,7 @@ package compiler
 
 import (
 	"go/types"
+	"strconv"
 
 	"github.com/tinygo-org/tinygo/compiler/llvmutil"
 	"golang.org/x/tools/go/ssa"
@@ -248,11 +249,11 @@ func (b *builder) createRunDefers() {
 	//         }
 	//     }
 
-	// Create loop.
-	loophead := b.ctx.AddBasicBlock(b.llvmFn, "rundefers.loophead")
-	loop := b.ctx.AddBasicBlock(b.llvmFn, "rundefers.loop")
-	unreachable := b.ctx.AddBasicBlock(b.llvmFn, "rundefers.default")
-	end := b.ctx.AddBasicBlock(b.llvmFn, "rundefers.end")
+	// Create loop, in the order: loophead, loop, callback0, callback1, ..., unreachable, end.
+	end := b.insertBasicBlock("rundefers.end")
+	unreachable := b.ctx.InsertBasicBlock(end, "rundefers.default")
+	loop := b.ctx.InsertBasicBlock(unreachable, "rundefers.loop")
+	loophead := b.ctx.InsertBasicBlock(loop, "rundefers.loophead")
 	b.CreateBr(loophead)
 
 	// Create loop head:
@@ -284,7 +285,7 @@ func (b *builder) createRunDefers() {
 		// Create switch case, for example:
 		//     case 0:
 		//         // run first deferred call
-		block := b.ctx.AddBasicBlock(b.llvmFn, "rundefers.callback")
+		block := b.insertBasicBlock("rundefers.callback" + strconv.Itoa(i))
 		sw.AddCase(llvm.ConstInt(b.uintptrType, uint64(i), false), block)
 		b.SetInsertPointAtEnd(block)
 		switch callback := callback.(type) {
diff --git a/compiler/interface.go b/compiler/interface.go
index e98285a2a..03a9e150b 100644
--- a/compiler/interface.go
+++ b/compiler/interface.go
@@ -389,8 +389,8 @@ func (b *builder) createTypeAssert(expr *ssa.TypeAssert) llvm.Value {
 	// value.
 
 	prevBlock := b.GetInsertBlock()
-	okBlock := b.ctx.AddBasicBlock(b.llvmFn, "typeassert.ok")
-	nextBlock := b.ctx.AddBasicBlock(b.llvmFn, "typeassert.next")
+	okBlock := b.insertBasicBlock("typeassert.ok")
+	nextBlock := b.insertBasicBlock("typeassert.next")
 	b.blockExits[b.currentBlock] = nextBlock // adjust outgoing block for phi nodes
 	b.CreateCondBr(commaOk, okBlock, nextBlock)
 
diff --git a/compiler/llvm.go b/compiler/llvm.go
index 81bd0b1f9..2e786f945 100644
--- a/compiler/llvm.go
+++ b/compiler/llvm.go
@@ -23,6 +23,23 @@ func (b *builder) createTemporaryAlloca(t llvm.Type, name string) (alloca, bitca
 	return llvmutil.CreateTemporaryAlloca(b.Builder, b.mod, t, name)
 }
 
+// insertBasicBlock inserts a new basic block after the current basic block.
+// This is useful when inserting new basic blocks while converting a
+// *ssa.BasicBlock to a llvm.BasicBlock and the LLVM basic block needs some
+// extra blocks.
+// It does not update b.blockExits, this must be done by the caller.
+func (b *builder) insertBasicBlock(name string) llvm.BasicBlock {
+	currentBB := b.Builder.GetInsertBlock()
+	nextBB := llvm.NextBasicBlock(currentBB)
+	if nextBB.IsNil() {
+		// Last basic block in the function, so add one to the end.
+		return b.ctx.AddBasicBlock(b.llvmFn, name)
+	}
+	// Insert a basic block before the next basic block - that is, at the
+	// current insert location.
+	return b.ctx.InsertBasicBlock(nextBB, name)
+}
+
 // emitLifetimeEnd signals the end of an (alloca) lifetime by calling the
 // llvm.lifetime.end intrinsic. It is commonly used together with
 // createTemporaryAlloca.
diff --git a/compiler/testdata/basic.ll b/compiler/testdata/basic.ll
index 549205a7a..9f82ddb4a 100644
--- a/compiler/testdata/basic.ll
+++ b/compiler/testdata/basic.ll
@@ -36,10 +36,6 @@ entry:
   %0 = icmp eq i32 %y, 0
   br i1 %0, label %divbyzero.throw, label %divbyzero.next
 
-divbyzero.throw:                                  ; preds = %entry
-  call void @runtime.divideByZeroPanic(i8* undef) #0
-  unreachable
-
 divbyzero.next:                                   ; preds = %entry
   %1 = icmp eq i32 %y, -1
   %2 = icmp eq i32 %x, -2147483648
@@ -47,6 +43,10 @@ divbyzero.next:                                   ; preds = %entry
   %4 = select i1 %3, i32 1, i32 %y
   %5 = sdiv i32 %x, %4
   ret i32 %5
+
+divbyzero.throw:                                  ; preds = %entry
+  call void @runtime.divideByZeroPanic(i8* undef) #0
+  unreachable
 }
 
 declare void @runtime.divideByZeroPanic(i8*)
@@ -57,13 +57,13 @@ entry:
   %0 = icmp eq i32 %y, 0
   br i1 %0, label %divbyzero.throw, label %divbyzero.next
 
-divbyzero.throw:                                  ; preds = %entry
-  call void @runtime.divideByZeroPanic(i8* undef) #0
-  unreachable
-
 divbyzero.next:                                   ; preds = %entry
   %1 = udiv i32 %x, %y
   ret i32 %1
+
+divbyzero.throw:                                  ; preds = %entry
+  call void @runtime.divideByZeroPanic(i8* undef) #0
+  unreachable
 }
 
 ; Function Attrs: nounwind
@@ -72,10 +72,6 @@ entry:
   %0 = icmp eq i32 %y, 0
   br i1 %0, label %divbyzero.throw, label %divbyzero.next
 
-divbyzero.throw:                                  ; preds = %entry
-  call void @runtime.divideByZeroPanic(i8* undef) #0
-  unreachable
-
 divbyzero.next:                                   ; preds = %entry
   %1 = icmp eq i32 %y, -1
   %2 = icmp eq i32 %x, -2147483648
@@ -83,6 +79,10 @@ divbyzero.next:                                   ; preds = %entry
   %4 = select i1 %3, i32 1, i32 %y
   %5 = srem i32 %x, %4
   ret i32 %5
+
+divbyzero.throw:                                  ; preds = %entry
+  call void @runtime.divideByZeroPanic(i8* undef) #0
+  unreachable
 }
 
 ; Function Attrs: nounwind
@@ -91,13 +91,13 @@ entry:
   %0 = icmp eq i32 %y, 0
   br i1 %0, label %divbyzero.throw, label %divbyzero.next
 
-divbyzero.throw:                                  ; preds = %entry
-  call void @runtime.divideByZeroPanic(i8* undef) #0
-  unreachable
-
 divbyzero.next:                                   ; preds = %entry
   %1 = urem i32 %x, %y
   ret i32 %1
+
+divbyzero.throw:                                  ; preds = %entry
+  call void @runtime.divideByZeroPanic(i8* undef) #0
+  unreachable
 }
 
 ; Function Attrs: nounwind
diff --git a/compiler/testdata/defer-cortex-m-qemu.ll b/compiler/testdata/defer-cortex-m-qemu.ll
new file mode 100644
index 000000000..ddb830da1
--- /dev/null
+++ b/compiler/testdata/defer-cortex-m-qemu.ll
@@ -0,0 +1,142 @@
+; ModuleID = 'defer.go'
+source_filename = "defer.go"
+target datalayout = "e-m:e-p:32:32-Fi8-i64:64-v128:64:128-a:0:32-n32-S64"
+target triple = "thumbv7m-unknown-unknown-eabi"
+
+%runtime._defer = type { i32, %runtime._defer* }
+
+declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*)
+
+; Function Attrs: nounwind
+define hidden void @main.init(i8* %context) unnamed_addr #0 {
+entry:
+  ret void
+}
+
+declare void @main.external(i8*)
+
+; Function Attrs: nounwind
+define hidden void @main.deferSimple(i8* %context) unnamed_addr #0 {
+entry:
+  %defer.alloca = alloca { i32, %runtime._defer* }, align 4
+  %deferPtr = alloca %runtime._defer*, align 4
+  store %runtime._defer* null, %runtime._defer** %deferPtr, align 4
+  %defer.alloca.repack = getelementptr inbounds { i32, %runtime._defer* }, { i32, %runtime._defer* }* %defer.alloca, i32 0, i32 0
+  store i32 0, i32* %defer.alloca.repack, align 4
+  %defer.alloca.repack1 = getelementptr inbounds { i32, %runtime._defer* }, { i32, %runtime._defer* }* %defer.alloca, i32 0, i32 1
+  store %runtime._defer* null, %runtime._defer** %defer.alloca.repack1, align 4
+  %0 = bitcast %runtime._defer** %deferPtr to { i32, %runtime._defer* }**
+  store { i32, %runtime._defer* }* %defer.alloca, { i32, %runtime._defer* }** %0, align 4
+  call void @main.external(i8* undef) #0
+  br label %rundefers.loophead
+
+rundefers.loophead:                               ; preds = %rundefers.callback0, %entry
+  %1 = load %runtime._defer*, %runtime._defer** %deferPtr, align 4
+  %stackIsNil = icmp eq %runtime._defer* %1, null
+  br i1 %stackIsNil, label %rundefers.end, label %rundefers.loop
+
+rundefers.loop:                                   ; preds = %rundefers.loophead
+  %stack.next.gep = getelementptr inbounds %runtime._defer, %runtime._defer* %1, i32 0, i32 1
+  %stack.next = load %runtime._defer*, %runtime._defer** %stack.next.gep, align 4
+  store %runtime._defer* %stack.next, %runtime._defer** %deferPtr, align 4
+  %callback.gep = getelementptr inbounds %runtime._defer, %runtime._defer* %1, i32 0, i32 0
+  %callback = load i32, i32* %callback.gep, align 4
+  switch i32 %callback, label %rundefers.default [
+    i32 0, label %rundefers.callback0
+  ]
+
+rundefers.callback0:                              ; preds = %rundefers.loop
+  call void @"main.deferSimple$1"(i8* undef)
+  br label %rundefers.loophead
+
+rundefers.default:                                ; preds = %rundefers.loop
+  unreachable
+
+rundefers.end:                                    ; preds = %rundefers.loophead
+  ret void
+
+recover:                                          ; No predecessors!
+  ret void
+}
+
+; Function Attrs: nounwind
+define hidden void @"main.deferSimple$1"(i8* %context) unnamed_addr #0 {
+entry:
+  call void @runtime.printint32(i32 3, i8* undef) #0
+  ret void
+}
+
+declare void @runtime.printint32(i32, i8*)
+
+; Function Attrs: nounwind
+define hidden void @main.deferMultiple(i8* %context) unnamed_addr #0 {
+entry:
+  %defer.alloca2 = alloca { i32, %runtime._defer* }, align 4
+  %defer.alloca = alloca { i32, %runtime._defer* }, align 4
+  %deferPtr = alloca %runtime._defer*, align 4
+  store %runtime._defer* null, %runtime._defer** %deferPtr, align 4
+  %defer.alloca.repack = getelementptr inbounds { i32, %runtime._defer* }, { i32, %runtime._defer* }* %defer.alloca, i32 0, i32 0
+  store i32 0, i32* %defer.alloca.repack, align 4
+  %defer.alloca.repack5 = getelementptr inbounds { i32, %runtime._defer* }, { i32, %runtime._defer* }* %defer.alloca, i32 0, i32 1
+  store %runtime._defer* null, %runtime._defer** %defer.alloca.repack5, align 4
+  %0 = bitcast %runtime._defer** %deferPtr to { i32, %runtime._defer* }**
+  store { i32, %runtime._defer* }* %defer.alloca, { i32, %runtime._defer* }** %0, align 4
+  %defer.alloca2.repack = getelementptr inbounds { i32, %runtime._defer* }, { i32, %runtime._defer* }* %defer.alloca2, i32 0, i32 0
+  store i32 1, i32* %defer.alloca2.repack, align 4
+  %defer.alloca2.repack6 = getelementptr inbounds { i32, %runtime._defer* }, { i32, %runtime._defer* }* %defer.alloca2, i32 0, i32 1
+  %1 = bitcast %runtime._defer** %defer.alloca2.repack6 to { i32, %runtime._defer* }**
+  store { i32, %runtime._defer* }* %defer.alloca, { i32, %runtime._defer* }** %1, align 4
+  %2 = bitcast %runtime._defer** %deferPtr to { i32, %runtime._defer* }**
+  store { i32, %runtime._defer* }* %defer.alloca2, { i32, %runtime._defer* }** %2, align 4
+  call void @main.external(i8* undef) #0
+  br label %rundefers.loophead
+
+rundefers.loophead:                               ; preds = %rundefers.callback1, %rundefers.callback0, %entry
+  %3 = load %runtime._defer*, %runtime._defer** %deferPtr, align 4
+  %stackIsNil = icmp eq %runtime._defer* %3, null
+  br i1 %stackIsNil, label %rundefers.end, label %rundefers.loop
+
+rundefers.loop:                                   ; preds = %rundefers.loophead
+  %stack.next.gep = getelementptr inbounds %runtime._defer, %runtime._defer* %3, i32 0, i32 1
+  %stack.next = load %runtime._defer*, %runtime._defer** %stack.next.gep, align 4
+  store %runtime._defer* %stack.next, %runtime._defer** %deferPtr, align 4
+  %callback.gep = getelementptr inbounds %runtime._defer, %runtime._defer* %3, i32 0, i32 0
+  %callback = load i32, i32* %callback.gep, align 4
+  switch i32 %callback, label %rundefers.default [
+    i32 0, label %rundefers.callback0
+    i32 1, label %rundefers.callback1
+  ]
+
+rundefers.callback0:                              ; preds = %rundefers.loop
+  call void @"main.deferMultiple$1"(i8* undef)
+  br label %rundefers.loophead
+
+rundefers.callback1:                              ; preds = %rundefers.loop
+  call void @"main.deferMultiple$2"(i8* undef)
+  br label %rundefers.loophead
+
+rundefers.default:                                ; preds = %rundefers.loop
+  unreachable
+
+rundefers.end:                                    ; preds = %rundefers.loophead
+  ret void
+
+recover:                                          ; No predecessors!
+  ret void
+}
+
+; Function Attrs: nounwind
+define hidden void @"main.deferMultiple$1"(i8* %context) unnamed_addr #0 {
+entry:
+  call void @runtime.printint32(i32 3, i8* undef) #0
+  ret void
+}
+
+; Function Attrs: nounwind
+define hidden void @"main.deferMultiple$2"(i8* %context) unnamed_addr #0 {
+entry:
+  call void @runtime.printint32(i32 5, i8* undef) #0
+  ret void
+}
+
+attributes #0 = { nounwind }
diff --git a/compiler/testdata/defer.go b/compiler/testdata/defer.go
new file mode 100644
index 000000000..ae334a656
--- /dev/null
+++ b/compiler/testdata/defer.go
@@ -0,0 +1,20 @@
+package main
+
+func external()
+
+func deferSimple() {
+	defer func() {
+		print(3)
+	}()
+	external()
+}
+
+func deferMultiple() {
+	defer func() {
+		print(3)
+	}()
+	defer func() {
+		print(5)
+	}()
+	external()
+}
diff --git a/compiler/testdata/func.ll b/compiler/testdata/func.ll
index b497ee7c5..405eddfdb 100644
--- a/compiler/testdata/func.ll
+++ b/compiler/testdata/func.ll
@@ -19,14 +19,14 @@ entry:
   %0 = icmp eq void ()* %callback.funcptr, null
   br i1 %0, label %fpcall.throw, label %fpcall.next
 
-fpcall.throw:                                     ; preds = %entry
-  call void @runtime.nilPanic(i8* undef) #0
-  unreachable
-
 fpcall.next:                                      ; preds = %entry
   %1 = bitcast void ()* %callback.funcptr to void (i32, i8*)*
   call void %1(i32 3, i8* %callback.context) #0
   ret void
+
+fpcall.throw:                                     ; preds = %entry
+  call void @runtime.nilPanic(i8* undef) #0
+  unreachable
 }
 
 declare void @runtime.nilPanic(i8*)
diff --git a/compiler/testdata/go1.17.ll b/compiler/testdata/go1.17.ll
index dc83e36e4..56ff52fb1 100644
--- a/compiler/testdata/go1.17.ll
+++ b/compiler/testdata/go1.17.ll
@@ -36,13 +36,13 @@ entry:
   %0 = icmp ult i32 %s.len, 4
   br i1 %0, label %slicetoarray.throw, label %slicetoarray.next
 
-slicetoarray.throw:                               ; preds = %entry
-  call void @runtime.sliceToArrayPointerPanic(i8* undef) #0
-  unreachable
-
 slicetoarray.next:                                ; preds = %entry
   %1 = bitcast i32* %s.data to [4 x i32]*
   ret [4 x i32]* %1
+
+slicetoarray.throw:                               ; preds = %entry
+  call void @runtime.sliceToArrayPointerPanic(i8* undef) #0
+  unreachable
 }
 
 declare void @runtime.sliceToArrayPointerPanic(i8*)
@@ -54,12 +54,12 @@ entry:
   call void @runtime.trackPointer(i8* nonnull %makeslice, i8* undef) #0
   br i1 false, label %slicetoarray.throw, label %slicetoarray.next
 
-slicetoarray.throw:                               ; preds = %entry
-  unreachable
-
 slicetoarray.next:                                ; preds = %entry
   %0 = bitcast i8* %makeslice to [4 x i32]*
   ret [4 x i32]* %0
+
+slicetoarray.throw:                               ; preds = %entry
+  unreachable
 }
 
 ; Function Attrs: nounwind
@@ -72,10 +72,6 @@ entry:
   %4 = or i1 %3, %0
   br i1 %4, label %unsafe.Slice.throw, label %unsafe.Slice.next
 
-unsafe.Slice.throw:                               ; preds = %entry
-  call void @runtime.unsafeSlicePanic(i8* undef) #0
-  unreachable
-
 unsafe.Slice.next:                                ; preds = %entry
   %5 = insertvalue { i32*, i32, i32 } undef, i32* %ptr, 0
   %6 = insertvalue { i32*, i32, i32 } %5, i32 %len, 1
@@ -83,6 +79,10 @@ unsafe.Slice.next:                                ; preds = %entry
   %8 = bitcast i32* %ptr to i8*
   call void @runtime.trackPointer(i8* %8, i8* undef) #0
   ret { i32*, i32, i32 } %7
+
+unsafe.Slice.throw:                               ; preds = %entry
+  call void @runtime.unsafeSlicePanic(i8* undef) #0
+  unreachable
 }
 
 declare void @runtime.unsafeSlicePanic(i8*)
@@ -95,10 +95,6 @@ entry:
   %2 = and i1 %0, %1
   br i1 %2, label %unsafe.Slice.throw, label %unsafe.Slice.next
 
-unsafe.Slice.throw:                               ; preds = %entry
-  call void @runtime.unsafeSlicePanic(i8* undef) #0
-  unreachable
-
 unsafe.Slice.next:                                ; preds = %entry
   %3 = zext i16 %len to i32
   %4 = insertvalue { i8*, i32, i32 } undef, i8* %ptr, 0
@@ -106,6 +102,10 @@ unsafe.Slice.next:                                ; preds = %entry
   %6 = insertvalue { i8*, i32, i32 } %5, i32 %3, 2
   call void @runtime.trackPointer(i8* %ptr, i8* undef) #0
   ret { i8*, i32, i32 } %6
+
+unsafe.Slice.throw:                               ; preds = %entry
+  call void @runtime.unsafeSlicePanic(i8* undef) #0
+  unreachable
 }
 
 ; Function Attrs: nounwind
@@ -118,10 +118,6 @@ entry:
   %4 = or i1 %3, %0
   br i1 %4, label %unsafe.Slice.throw, label %unsafe.Slice.next
 
-unsafe.Slice.throw:                               ; preds = %entry
-  call void @runtime.unsafeSlicePanic(i8* undef) #0
-  unreachable
-
 unsafe.Slice.next:                                ; preds = %entry
   %5 = trunc i64 %len to i32
   %6 = insertvalue { i32*, i32, i32 } undef, i32* %ptr, 0
@@ -130,6 +126,10 @@ unsafe.Slice.next:                                ; preds = %entry
   %9 = bitcast i32* %ptr to i8*
   call void @runtime.trackPointer(i8* %9, i8* undef) #0
   ret { i32*, i32, i32 } %8
+
+unsafe.Slice.throw:                               ; preds = %entry
+  call void @runtime.unsafeSlicePanic(i8* undef) #0
+  unreachable
 }
 
 ; Function Attrs: nounwind
@@ -142,10 +142,6 @@ entry:
   %4 = or i1 %3, %0
   br i1 %4, label %unsafe.Slice.throw, label %unsafe.Slice.next
 
-unsafe.Slice.throw:                               ; preds = %entry
-  call void @runtime.unsafeSlicePanic(i8* undef) #0
-  unreachable
-
 unsafe.Slice.next:                                ; preds = %entry
   %5 = trunc i64 %len to i32
   %6 = insertvalue { i32*, i32, i32 } undef, i32* %ptr, 0
@@ -154,6 +150,10 @@ unsafe.Slice.next:                                ; preds = %entry
   %9 = bitcast i32* %ptr to i8*
   call void @runtime.trackPointer(i8* %9, i8* undef) #0
   ret { i32*, i32, i32 } %8
+
+unsafe.Slice.throw:                               ; preds = %entry
+  call void @runtime.unsafeSlicePanic(i8* undef) #0
+  unreachable
 }
 
 attributes #0 = { nounwind }
diff --git a/compiler/testdata/interface.ll b/compiler/testdata/interface.ll
index 609b91fe1..986854c28 100644
--- a/compiler/testdata/interface.ll
+++ b/compiler/testdata/interface.ll
@@ -70,11 +70,11 @@ entry:
   %typecode = call i1 @runtime.typeAssert(i32 %itf.typecode, i8* nonnull @"reflect/types.typeid:basic:int", i8* undef) #0
   br i1 %typecode, label %typeassert.ok, label %typeassert.next
 
-typeassert.ok:                                    ; preds = %entry
-  br label %typeassert.next
-
 typeassert.next:                                  ; preds = %typeassert.ok, %entry
   ret i1 %typecode
+
+typeassert.ok:                                    ; preds = %entry
+  br label %typeassert.next
 }
 
 declare i1 @runtime.typeAssert(i32, i8* dereferenceable_or_null(1), i8*)
@@ -85,11 +85,11 @@ entry:
   %0 = call i1 @"interface:{Error:func:{}{basic:string}}.$typeassert"(i32 %itf.typecode) #0
   br i1 %0, label %typeassert.ok, label %typeassert.next
 
-typeassert.ok:                                    ; preds = %entry
-  br label %typeassert.next
-
 typeassert.next:                                  ; preds = %typeassert.ok, %entry
   ret i1 %0
+
+typeassert.ok:                                    ; preds = %entry
+  br label %typeassert.next
 }
 
 ; Function Attrs: nounwind
@@ -98,11 +98,11 @@ entry:
   %0 = call i1 @"interface:{String:func:{}{basic:string}}.$typeassert"(i32 %itf.typecode) #0
   br i1 %0, label %typeassert.ok, label %typeassert.next
 
-typeassert.ok:                                    ; preds = %entry
-  br label %typeassert.next
-
 typeassert.next:                                  ; preds = %typeassert.ok, %entry
   ret i1 %0
+
+typeassert.ok:                                    ; preds = %entry
+  br label %typeassert.next
 }
 
 ; Function Attrs: nounwind
diff --git a/compiler/testdata/slice.ll b/compiler/testdata/slice.ll
index eb5fcd6ca..f5947c68f 100644
--- a/compiler/testdata/slice.ll
+++ b/compiler/testdata/slice.ll
@@ -31,14 +31,14 @@ entry:
   %.not = icmp ult i32 %index, %ints.len
   br i1 %.not, label %lookup.next, label %lookup.throw
 
-lookup.throw:                                     ; preds = %entry
-  call void @runtime.lookupPanic(i8* undef) #0
-  unreachable
-
 lookup.next:                                      ; preds = %entry
   %0 = getelementptr inbounds i32, i32* %ints.data, i32 %index
   %1 = load i32, i32* %0, align 4
   ret i32 %1
+
+lookup.throw:                                     ; preds = %entry
+  call void @runtime.lookupPanic(i8* undef) #0
+  unreachable
 }
 
 declare void @runtime.lookupPanic(i8*)
@@ -105,10 +105,6 @@ entry:
   %slice.maxcap = icmp slt i32 %len, 0
   br i1 %slice.maxcap, label %slice.throw, label %slice.next
 
-slice.throw:                                      ; preds = %entry
-  call void @runtime.slicePanic(i8* undef) #0
-  unreachable
-
 slice.next:                                       ; preds = %entry
   %makeslice.buf = call i8* @runtime.alloc(i32 %len, i8* nonnull inttoptr (i32 3 to i8*), i8* undef) #0
   %0 = insertvalue { i8*, i32, i32 } undef, i8* %makeslice.buf, 0
@@ -116,6 +112,10 @@ slice.next:                                       ; preds = %entry
   %2 = insertvalue { i8*, i32, i32 } %1, i32 %len, 2
   call void @runtime.trackPointer(i8* nonnull %makeslice.buf, i8* undef) #0
   ret { i8*, i32, i32 } %2
+
+slice.throw:                                      ; preds = %entry
+  call void @runtime.slicePanic(i8* undef) #0
+  unreachable
 }
 
 declare void @runtime.slicePanic(i8*)
@@ -126,10 +126,6 @@ entry:
   %slice.maxcap = icmp slt i32 %len, 0
   br i1 %slice.maxcap, label %slice.throw, label %slice.next
 
-slice.throw:                                      ; preds = %entry
-  call void @runtime.slicePanic(i8* undef) #0
-  unreachable
-
 slice.next:                                       ; preds = %entry
   %makeslice.cap = shl i32 %len, 1
   %makeslice.buf = call i8* @runtime.alloc(i32 %makeslice.cap, i8* nonnull inttoptr (i32 3 to i8*), i8* undef) #0
@@ -139,6 +135,10 @@ slice.next:                                       ; preds = %entry
   %2 = insertvalue { i16*, i32, i32 } %1, i32 %len, 2
   call void @runtime.trackPointer(i8* nonnull %makeslice.buf, i8* undef) #0
   ret { i16*, i32, i32 } %2
+
+slice.throw:                                      ; preds = %entry
+  call void @runtime.slicePanic(i8* undef) #0
+  unreachable
 }
 
 ; Function Attrs: nounwind
@@ -147,10 +147,6 @@ entry:
   %slice.maxcap = icmp ugt i32 %len, 1431655765
   br i1 %slice.maxcap, label %slice.throw, label %slice.next
 
-slice.throw:                                      ; preds = %entry
-  call void @runtime.slicePanic(i8* undef) #0
-  unreachable
-
 slice.next:                                       ; preds = %entry
   %makeslice.cap = mul i32 %len, 3
   %makeslice.buf = call i8* @runtime.alloc(i32 %makeslice.cap, i8* nonnull inttoptr (i32 3 to i8*), i8* undef) #0
@@ -160,6 +156,10 @@ slice.next:                                       ; preds = %entry
   %2 = insertvalue { [3 x i8]*, i32, i32 } %1, i32 %len, 2
   call void @runtime.trackPointer(i8* nonnull %makeslice.buf, i8* undef) #0
   ret { [3 x i8]*, i32, i32 } %2
+
+slice.throw:                                      ; preds = %entry
+  call void @runtime.slicePanic(i8* undef) #0
+  unreachable
 }
 
 ; Function Attrs: nounwind
@@ -168,10 +168,6 @@ entry:
   %slice.maxcap = icmp ugt i32 %len, 1073741823
   br i1 %slice.maxcap, label %slice.throw, label %slice.next
 
-slice.throw:                                      ; preds = %entry
-  call void @runtime.slicePanic(i8* undef) #0
-  unreachable
-
 slice.next:                                       ; preds = %entry
   %makeslice.cap = shl i32 %len, 2
   %makeslice.buf = call i8* @runtime.alloc(i32 %makeslice.cap, i8* nonnull inttoptr (i32 3 to i8*), i8* undef) #0
@@ -181,6 +177,10 @@ slice.next:                                       ; preds = %entry
   %2 = insertvalue { i32*, i32, i32 } %1, i32 %len, 2
   call void @runtime.trackPointer(i8* nonnull %makeslice.buf, i8* undef) #0
   ret { i32*, i32, i32 } %2
+
+slice.throw:                                      ; preds = %entry
+  call void @runtime.slicePanic(i8* undef) #0
+  unreachable
 }
 
 attributes #0 = { nounwind }
diff --git a/compiler/testdata/string.ll b/compiler/testdata/string.ll
index 99d81bab9..6d25a35fb 100644
--- a/compiler/testdata/string.ll
+++ b/compiler/testdata/string.ll
@@ -41,14 +41,14 @@ entry:
   %.not = icmp ult i32 %index, %s.len
   br i1 %.not, label %lookup.next, label %lookup.throw
 
-lookup.throw:                                     ; preds = %entry
-  call void @runtime.lookupPanic(i8* undef) #0
-  unreachable
-
 lookup.next:                                      ; preds = %entry
   %0 = getelementptr inbounds i8, i8* %s.data, i32 %index
   %1 = load i8, i8* %0, align 1
   ret i8 %1
+
+lookup.throw:                                     ; preds = %entry
+  call void @runtime.lookupPanic(i8* undef) #0
+  unreachable
 }
 
 declare void @runtime.lookupPanic(i8*)
@@ -86,14 +86,14 @@ entry:
   %.not = icmp ult i32 %0, %s.len
   br i1 %.not, label %lookup.next, label %lookup.throw
 
-lookup.throw:                                     ; preds = %entry
-  call void @runtime.lookupPanic(i8* undef) #0
-  unreachable
-
 lookup.next:                                      ; preds = %entry
   %1 = getelementptr inbounds i8, i8* %s.data, i32 %0
   %2 = load i8, i8* %1, align 1
   ret i8 %2
+
+lookup.throw:                                     ; preds = %entry
+  call void @runtime.lookupPanic(i8* undef) #0
+  unreachable
 }
 
 attributes #0 = { nounwind }