Fix bug when atomic load is fused with an arith op (and not in the entry BB)

tests_lit/llvm2ice_tests/nacl-atomic-intrinsics.ll

Index: tests_lit/llvm2ice_tests/nacl-atomic-intrinsics.ll
diff --git a/tests_lit/llvm2ice_tests/nacl-atomic-intrinsics.ll b/tests_lit/llvm2ice_tests/nacl-atomic-intrinsics.ll
index a852e87161845ff6352d584295731f9d89e59cf6..8ea72d9bbc7ec5499231fb3c4eb258f064743614 100644
--- a/tests_lit/llvm2ice_tests/nacl-atomic-intrinsics.ll
+++ b/tests_lit/llvm2ice_tests/nacl-atomic-intrinsics.ll
@@ -2,7 +2,7 @@
 ; size allowed.
 
 ; RUN: %llvm2ice -O2 --verbose none %s | FileCheck %s
-; RUN: %llvm2ice -O2 --verbose none %s | FileCheck %s --check-prefix=CHECKO2REM
+; RUN: %llvm2ice -O2 --verbose none %s | FileCheck %s --check-prefix=CHECKO2
 ; RUN: %llvm2ice -Om1 --verbose none %s | FileCheck %s
 ; RUN: %llvm2ice -O2 --verbose none %s \
 ; RUN:               | llvm-mc -arch=x86 -x86-asm-syntax=intel -filetype=obj
@@ -88,6 +88,9 @@ entry:
 
 define i32 @test_atomic_load_32_with_arith(i32 %iptr) {
 entry:
+  br label %next
+
+next:
   %ptr = inttoptr i32 %iptr to i32*
   %r = call i32 @llvm.nacl.atomic.load.i32(i32* %ptr, i32 6)
   %r2 = add i32 %r, 32
@@ -96,6 +99,11 @@ entry:
 ; CHECK-LABEL: test_atomic_load_32_with_arith
 ; CHECK: mov {{.*}}, dword
 ; The next instruction may be a separate load or folded into an add.
+;
+; In O2 mode, we know that the load and add are going to be fused.
+; CHECKO2-LABEL: test_atomic_load_32_with_arith
+; CHECKO2: mov {{.*}}, dword
+; CHECKO2: add {{.*}}, dword
 
 define i32 @test_atomic_load_32_ignored(i32 %iptr) {
 entry:
@@ -106,6 +114,9 @@ entry:
 ; CHECK-LABEL: test_atomic_load_32_ignored
 ; CHECK: mov {{.*}}, dword
 ; CHECK: mov {{.*}}, dword
+; CHECKO2-LABEL: test_atomic_load_32_ignored
+; CHECKO2: mov {{.*}}, dword
+; CHECKO2: mov {{.*}}, dword
 
 define i64 @test_atomic_load_64_ignored(i32 %iptr) {
 entry:
@@ -879,9 +890,9 @@ entry:
 ; CHECK-LABEL: test_atomic_is_lock_free_ignored
 ; CHECK: mov {{.*}}, 0
 ; This can get optimized out, because it's side-effect-free.
-; CHECKO2REM-LABEL: test_atomic_is_lock_free_ignored
-; CHECKO2REM-NOT: mov {{.*}}, 1
-; CHECKO2REM: mov {{.*}}, 0
+; CHECKO2-LABEL: test_atomic_is_lock_free_ignored
+; CHECKO2-NOT: mov {{.*}}, 1
+; CHECKO2: mov {{.*}}, 0
 
 ; TODO(jvoung): at some point we can take advantage of the
 ; fact that nacl.atomic.is.lock.free will resolve to a constant