AMDGPU: Fix incorrect simm check

Use signed division otherwise all back jumps fail the check
Fixes regression introduced in r269951

Differential Revision: http://reviews.llvm.org/D20380

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@269972 91177308-0d34-0410-b5e6-96231b3b80d8

lib/Target/AMDGPU/MCTargetDesc/AMDGPUAsmBackend.cpp

@@ -96,7 +96,7 @@ void AMDGPUAsmBackend::applyFixup(const MCFixup &Fixup, char *Data,
 
   switch ((unsigned)Fixup.getKind()) {
     case AMDGPU::fixup_si_sopp_br: {
-      int64_t BrImm = (Value - 4) / 4;
+      int64_t BrImm = ((int64_t)Value - 4) / 4;
       if (!isInt<16>(BrImm))
         report_fatal_error("branch size exceeds simm16");