Remove broken szdiff tests.

tests_lit/llvm2ice_tests/nacl-atomic-cmpxchg-optimization.ll

 ; This tests the optimization of atomic cmpxchg w/ following cmp + branches.
 
 ; RUN: %p2i -i %s --args -O2 --verbose none \
 ; RUN:   | llvm-mc -triple=i686-none-nacl -x86-asm-syntax=intel -filetype=obj \
 ; RUN:   | llvm-objdump -d -symbolize -x86-asm-syntax=intel - \
 ; RUN:   | FileCheck --check-prefix=O2 %s
 ; RUN: %p2i -i %s --args -Om1 --verbose none \
 ; RUN:   | llvm-mc -triple=i686-none-nacl -x86-asm-syntax=intel -filetype=obj \
 ; RUN:   | llvm-objdump -d -symbolize -x86-asm-syntax=intel - \
 ; RUN:   | FileCheck --check-prefix=OM1 %s
 ; RUN: %p2i -i %s --args --verbose none | FileCheck --check-prefix=ERRORS %s
-; TODO(kschimpf) Find out why lc2i is needed.
-; RUN: %lc2i -i %s --insts | %szdiff %s | FileCheck --check-prefix=DUMP %s
 
 declare i32 @llvm.nacl.atomic.cmpxchg.i32(i32*, i32, i32, i32, i32)
 
 
 ; Test that a cmpxchg followed by icmp eq and branch can be optimized to
 ; reuse the flags set by the cmpxchg instruction itself.
 ; This is only expected to work w/ O2, based on lightweight liveness.
 ; (Or if we had other means to detect the only use).
 declare void @use_value(i32)
 
@@ skipping 121 matching lines @@
   ret i32 %r
 }
 ; O2-LABEL: test_atomic_cmpxchg_no_opt2
 ; O2: lock
 ; O2-NEXT: cmpxchg dword ptr [e{{[^a].}}], e{{[^a]}}
 ; O2: mov {{.*}}
 ; O2: cmp
 ; O2: je
 
 ; ERRORS-NOT: ICE translation error
-; DUMP-NOT: SZ