Add llvm.sideeffect to potential infinite loops and recursions

LLVM assumes that a thread will eventually cause side effect. This is
not true in Rust if a loop or recursion does nothing in its body,
causing undefined behavior even in common cases like `loop {}`.
Inserting llvm.sideeffect fixes the undefined behavior.

As a micro-optimization, only insert llvm.sideeffect when jumping back
in blocks or calling a function.

A patch for LLVM is expected to allow empty non-terminate code by
default and fix this issue from LLVM side.

https://github.com/rust-lang/rust/issues/28728

src/test/codegen/alloc-optimisation.rs

@@ -7,7 +7,8 @@
 pub fn alloc_test(data: u32) {
     // CHECK-LABEL: @alloc_test
     // CHECK-NEXT: start:
-    // CHECK-NEXT: ret void
+    // CHECK-NOT: alloc
+    // CHECK: ret void
     let x = Box::new(data);
     drop(x);
 }

src/test/codegen/dealloc-no-unwind.rs

@@ -17,7 +17,7 @@ impl Drop for A {
 pub fn a(a: Box<i32>) {
     // CHECK-LABEL: define void @a
     // CHECK: call void @__rust_dealloc
-    // CHECK-NEXT: call void @foo
+    // CHECK: call void @foo
     let _a = A;
     drop(a);
 }

src/test/codegen/issue-34947-pow-i32.rs

@@ -6,8 +6,11 @@
 #[no_mangle]
 pub fn issue_34947(x: i32) -> i32 {
     // CHECK: mul
-    // CHECK-NEXT: mul
-    // CHECK-NEXT: mul
-    // CHECK-NEXT: ret
+    // CHECK-NOT: br label
+    // CHECK: mul
+    // CHECK-NOT: br label
+    // CHECK: mul
+    // CHECK-NOT: br label
+    // CHECK: ret
     x.pow(5)
 }

src/test/codegen/issue-45222.rs

@@ -1,3 +1,9 @@
+// ignore-test
+
+// FIXME:
+// LLVM can't optimize some loops with a large number of iterations because of
+// @llvm.sideeffect() (see also #59546)
+
 // compile-flags: -O
 // ignore-debug: the debug assertions get in the way
 

src/test/codegen/naked-functions.rs

@@ -1,5 +1,3 @@
-// ignore-tidy-linelength
-
 // compile-flags: -C no-prepopulate-passes
 
 #![crate_type = "lib"]
@@ -53,7 +51,7 @@ pub fn naked_with_args_and_return(a: isize) -> isize {
 #[naked]
 pub fn naked_recursive() {
     // CHECK-NEXT: {{.+}}:
-    // CHECK-NEXT: call void @naked_empty()
+    // CHECK: call void @naked_empty()
 
     // FIXME(#39685) Avoid one block per call.
     // CHECK-NEXT: br label %bb1
@@ -61,19 +59,19 @@ pub fn naked_recursive() {
 
     naked_empty();
 
-    // CHECK-NEXT: %{{[0-9]+}} = call i{{[0-9]+}} @naked_with_return()
+    // CHECK: %{{[0-9]+}} = call i{{[0-9]+}} @naked_with_return()
 
     // FIXME(#39685) Avoid one block per call.
     // CHECK-NEXT: br label %bb2
     // CHECK: bb2:
 
-    // CHECK-NEXT: %{{[0-9]+}} = call i{{[0-9]+}} @naked_with_args_and_return(i{{[0-9]+}} %{{[0-9]+}})
+    // CHECK: %{{[0-9]+}} = call i{{[0-9]+}} @naked_with_args_and_return(i{{[0-9]+}} %{{[0-9]+}})
 
     // FIXME(#39685) Avoid one block per call.
     // CHECK-NEXT: br label %bb3
     // CHECK: bb3:
 
-    // CHECK-NEXT: call void @naked_with_args(i{{[0-9]+}} %{{[0-9]+}})
+    // CHECK: call void @naked_with_args(i{{[0-9]+}} %{{[0-9]+}})
 
     // FIXME(#39685) Avoid one block per call.
     // CHECK-NEXT: br label %bb4

src/test/codegen/non-terminate/infinite-loop-1.rs

@@ -0,0 +1,17 @@
+// compile-flags: -C opt-level=3
+
+#![crate_type = "lib"]
+
+fn infinite_loop() -> u8 {
+    loop {}
+}
+
+// CHECK-LABEL: @test
+#[no_mangle]
+fn test() -> u8 {
+    // CHECK-NOT: unreachable
+    // CHECK: br label %{{.+}}
+    // CHECK-NOT: unreachable
+    let x = infinite_loop();
+    x
+}

src/test/codegen/non-terminate/infinite-loop-2.rs

@@ -0,0 +1,19 @@
+// compile-flags: -C opt-level=3
+
+#![crate_type = "lib"]
+
+fn infinite_loop() -> u8 {
+    let i = 2;
+    while i > 1 {}
+    1
+}
+
+// CHECK-LABEL: @test
+#[no_mangle]
+fn test() -> u8 {
+    // CHECK-NOT: unreachable
+    // CHECK: br label %{{.+}}
+    // CHECK-NOT: unreachable
+    let x = infinite_loop();
+    x
+}

src/test/codegen/non-terminate/infinite-recursion.rs

@@ -0,0 +1,14 @@
+// compile-flags: -C opt-level=3
+
+#![crate_type = "lib"]
+
+#![allow(unconditional_recursion)]
+
+// CHECK-LABEL: @infinite_recursion
+#[no_mangle]
+fn infinite_recursion() -> u8 {
+    // CHECK-NOT: ret i8 undef
+    // CHECK: br label %{{.+}}
+    // CHECK-NOT: ret i8 undef
+    infinite_recursion()
+}

src/test/codegen/repeat-trusted-len.rs

@@ -14,6 +14,10 @@ pub fn helper(_: usize) {
 // CHECK-LABEL: @repeat_take_collect
 #[no_mangle]
 pub fn repeat_take_collect() -> Vec<u8> {
-// CHECK: call void @llvm.memset.p0i8.[[USIZE]](i8* {{(nonnull )?}}align 1 %{{[0-9]+}}, i8 42, [[USIZE]] 100000, i1 false)
+// FIXME: At the time of writing LLVM transforms this loop into a single
+// `store` and then a `memset` with size = 99999. The correct check should be:
+//        call void @llvm.memset.p0i8.[[USIZE]](i8* {{(nonnull )?}}align 1 %{{[a-z0-9.]+}}, i8 42, [[USIZE]] 100000, i1 false)
+
+// CHECK: call void @llvm.memset.p0i8.[[USIZE]](i8* {{(nonnull )?}}align 1 %{{[a-z0-9.]+}}, i8 42, [[USIZE]] 99999, i1 false)
     iter::repeat(42).take(100000).collect()
 }

src/test/codegen/vec-clear.rs

@@ -1,3 +1,9 @@
+// ignore-test
+
+// FIXME:
+// LLVM can't optimize some loops with unknown number of iterations because of
+// @llvm.sideeffect() (see also #59546)
+
 // ignore-debug: the debug assertions get in the way
 // compile-flags: -O
 

src/test/codegen/vec-iter-collect-len.rs

@@ -5,8 +5,6 @@
 
 #[no_mangle]
 pub fn get_len() -> usize {
-    // CHECK-LABEL: @get_len
-    // CHECK-NOT: call
-    // CHECK-NOT: invoke
+    // CHECK-COUNT-1: {{^define}}
     [1, 2, 3].iter().collect::<Vec<_>>().len()
 }

src/test/codegen/vec-optimizes-away.rs

@@ -8,6 +8,6 @@
 pub fn sum_me() -> i32 {
     // CHECK-LABEL: @sum_me
     // CHECK-NEXT: {{^.*:$}}
-    // CHECK-NEXT: ret i32 6
+    // CHECK: ret i32 6
     vec![1, 2, 3].iter().sum::<i32>()
 }