i#1729/i#1569 A64 traces: Reintroduce drcachesim trace_entry_t packing.

This now works because XINST_CREATE_store on AArch64 uses OP_stur
instead of OP_str when it detects an "unaligned" offset.

Review-URL: https://codereview.appspot.com/309480043

clients/drcachesim/common/trace_entry.h

@@ -117,11 +117,7 @@ extern const char * const trace_type_names[];
 // - a flush request
 // - a prefetch request
 // - a thread/process
-// FIXME i#1729/i#1569: for offline traces we need to pack this to avoid
-// compiler padding differences.  Packing should also help online performance,
-// and having the two identical is required anyway to avoid complexity in
-// reader_t.  Once the AArch64 instrumentation handles unaligned offsets we'll
-// put back the {START,END}_PACKED_STRUCTURE.
+START_PACKED_STRUCTURE
 struct _trace_entry_t {
     unsigned short type; // 2 bytes: trace_type_t
     // 2 bytes: mem ref size, instr length, or num of instrs for instr bundle
@@ -131,7 +127,7 @@ struct _trace_entry_t {
         // The length of each instr in the instr bundle
         unsigned char length[sizeof(addr_t)];
     };
-};
+} END_PACKED_STRUCTURE;
 typedef struct _trace_entry_t trace_entry_t;
 
 static inline bool

core/arch/aarch64/instr_create.h

@@ -94,7 +94,10 @@
  * \param m   The destination memory opnd.
  * \param r   The source register opnd.
  */
-#define XINST_CREATE_store(dc, m, r) INSTR_CREATE_str((dc), (m), (r))
+#define XINST_CREATE_store(dc, m, r) \
+  (opnd_is_base_disp(m) && \
+   opnd_get_disp(m) % opnd_size_in_bytes(opnd_get_size(m)) != 0 ? \
+   INSTR_CREATE_stur(dc, m, r) : INSTR_CREATE_str(dc, m, r))
 
 /**
  * This platform-independent macro creates an instr_t for a 1-byte
@@ -289,10 +292,14 @@
   instr_create_0dst_0src((dc), OP_nop)
 #define INSTR_CREATE_ret(dc, Rn) \
   instr_create_0dst_1src((dc), OP_ret, (Rn))
-#define INSTR_CREATE_str(dc, mem, Rt) \
-  instr_create_1dst_1src((dc), OP_str, (mem), (Rt))
-#define INSTR_CREATE_strh(dc, mem, Rt) \
-  instr_create_1dst_1src((dc), OP_strh, (mem), (Rt))
+#define INSTR_CREATE_str(dc, mem, rt) \
+  instr_create_1dst_1src(dc, OP_str, mem, rt)
+#define INSTR_CREATE_strh(dc, mem, rt) \
+  instr_create_1dst_1src(dc, OP_strh, mem, rt)
+#define INSTR_CREATE_stur(dc, mem, rt) \
+  instr_create_1dst_1src(dc, OP_stur, mem, rt)
+#define INSTR_CREATE_sturh(dc, mem, rt) \
+  instr_create_1dst_1src(dc, OP_sturh, mem, rt)
 #define INSTR_CREATE_sub(dc, rd, rn, rm_or_imm) \
   INSTR_CREATE_sub_shift(dc, rd, rn, rm_or_imm, OPND_CREATE_LSL(), OPND_CREATE_INT(0))
 #define INSTR_CREATE_sub_extend(dc, rd, rn, rm, ext, exa) \