Subzero: Fixes for Hello World and bisection debugging.

tests_lit/llvm2ice_tests/nacl-other-intrinsics.ll

 ; This tests the NaCl intrinsics not related to atomic operations.
 
-; RUN: %llvm2ice -O2 --verbose none %s | FileCheck %s
-; RUN: %llvm2ice -Om1 --verbose none %s | FileCheck %s
+; RUN: %llvm2ice -O2 --verbose none -sandbox %s | FileCheck %s
+; RUN: %llvm2ice -Om1 --verbose none -sandbox %s | FileCheck %s
 
 ; Do another run w/ O2 and a different check-prefix (otherwise O2 and Om1
 ; share the same "CHECK" prefix). This separate run helps check that
 ; some code is optimized out.
-; RUN: %llvm2ice -O2 --verbose none %s | FileCheck %s --check-prefix=CHECKO2REM
+; RUN: %llvm2ice -O2 --verbose none -sandbox %s \
+; RUN:     | FileCheck %s --check-prefix=CHECKO2REM
+
+; Do O2 runs without -sandbox to make sure llvm.nacl.read.tp gets
+; lowered to __nacl_read_tp instead of gs:[0x0].
+; RUN: %llvm2ice -O2 --verbose none %s \
+; RUN:     | FileCheck --check-prefix=CHECKO2UNSANDBOXED %s
+; RUN: %llvm2ice -O2 --verbose none %s \
+; RUN:     | FileCheck --check-prefix=CHECKO2UNSANDBOXEDREM %s
 
 ; RUN: %llvm2ice -O2 --verbose none %s \
 ; RUN:     | llvm-mc -triple=i686-none-nacl -x86-asm-syntax=intel -filetype=obj
 ; RUN: %llvm2ice -Om1 --verbose none %s \
 ; RUN:     | llvm-mc -triple=i686-none-nacl -x86-asm-syntax=intel -filetype=obj
 ; RUN: %llvm2ice --verbose none %s | FileCheck --check-prefix=ERRORS %s
 ; RUN: %llvm2iceinsts %s | %szdiff %s | FileCheck --check-prefix=DUMP %s
 ; RUN: %llvm2iceinsts --pnacl %s | %szdiff %s \
 ; RUN:                           | FileCheck --check-prefix=DUMP %s
 
@@ skipping 21 matching lines @@
 define i32 @test_nacl_read_tp() {
 entry:
   %ptr = call i8* @llvm.nacl.read.tp()
   %__1 = ptrtoint i8* %ptr to i32
   ret i32 %__1
 }
 ; CHECK-LABEL: test_nacl_read_tp
 ; CHECK: mov e{{.*}}, dword ptr gs:[0]
 ; CHECKO2REM-LABEL: test_nacl_read_tp
 ; CHECKO2REM: mov e{{.*}}, dword ptr gs:[0]
+; CHECKO2UNSANDBOXED-LABEL: test_nacl_read_tp
+; CHECKO2UNSANDBOXED: call __nacl_read_tp
+; CHECKO2UNSANDBOXEDREM-LABEL: test_nacl_read_tp
+; CHECKO2UNSANDBOXEDREM: call __nacl_read_tp
 
 define i32 @test_nacl_read_tp_more_addressing() {
 entry:
   %ptr = call i8* @llvm.nacl.read.tp()
   %__1 = ptrtoint i8* %ptr to i32
   %x = add i32 %__1, %__1
   %__3 = inttoptr i32 %x to i32*
   %v = load i32* %__3, align 1
   %ptr2 = call i8* @llvm.nacl.read.tp()
   %__6 = ptrtoint i8* %ptr2 to i32
   %y = add i32 %__6, 4
   %__8 = inttoptr i32 %y to i32*
   store i32 %v, i32* %__8, align 1
   ret i32 %v
 }
 ; CHECK-LABEL: test_nacl_read_tp_more_addressing
 ; CHECK: mov e{{.*}}, dword ptr gs:[0]
 ; CHECK: mov e{{.*}}, dword ptr gs:[0]
 ; CHECKO2REM-LABEL: test_nacl_read_tp_more_addressing
 ; CHECKO2REM: mov e{{.*}}, dword ptr gs:[0]
 ; CHECKO2REM: mov e{{.*}}, dword ptr gs:[0]
+; CHECKO2UNSANDBOXED-LABEL: test_nacl_read_tp_more_addressing
+; CHECKO2UNSANDBOXED: call __nacl_read_tp
+; CHECKO2UNSANDBOXED: call __nacl_read_tp
+; CHECKO2UNSANDBOXEDREM-LABEL: test_nacl_read_tp_more_addressing
+; CHECKO2UNSANDBOXEDREM: call __nacl_read_tp
+; CHECKO2UNSANDBOXEDREM: call __nacl_read_tp
 
 define i32 @test_nacl_read_tp_dead(i32 %a) {
 entry:
   %ptr = call i8* @llvm.nacl.read.tp()
   ; Not actually using the result of nacl read tp call.
   ; In O2 mode this should be DCE'ed.
   ret i32 %a
 }
 ; Consider nacl.read.tp side-effect free, so it can be eliminated.
 ; CHECKO2REM-LABEL: test_nacl_read_tp_dead
 ; CHECKO2REM-NOT: mov e{{.*}}, dword ptr gs:[0]
+; CHECKO2UNSANDBOXEDREM-LABEL: test_nacl_read_tp_dead
+; CHECKO2UNSANDBOXEDREM-NOT: call __nacl_read_tp
 
 define void @test_memcpy(i32 %iptr_dst, i32 %iptr_src, i32 %len) {
 entry:
   %dst = inttoptr i32 %iptr_dst to i8*
   %src = inttoptr i32 %iptr_src to i8*
   call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dst, i8* %src,
                                        i32 %len, i32 1, i1 0)
   ret void
 }
 ; CHECK-LABEL: test_memcpy
 ; CHECK: call memcpy
+; CHECKO2REM-LABEL: test_memcpy
+; CHECKO2UNSANDBOXEDREM-LABEL: test_memcpy
 
 ; TODO(jvoung) -- if we want to be clever, we can do this and the memmove,
 ; memset without a function call.
 define void @test_memcpy_const_len_align(i32 %iptr_dst, i32 %iptr_src) {
 entry:
   %dst = inttoptr i32 %iptr_dst to i8*
   %src = inttoptr i32 %iptr_src to i8*
   call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dst, i8* %src,
                                        i32 8, i32 1, i1 0)
   ret void
@@ skipping 352 matching lines @@
 }
 ; CHECK-LABEL: test_stacksave_multiple
 ; At least 3 copies of esp, but probably more from having to do the allocas.
 ; CHECK: mov {{.*}}, esp
 ; CHECK: mov {{.*}}, esp
 ; CHECK: mov {{.*}}, esp
 ; CHECK: mov esp, {{.*}}
 
 ; ERRORS-NOT: ICE translation error
 ; DUMP-NOT: SZ