[Lldb-commits] [PATCH] D91293: [lldb] [Process/FreeBSDRemote] Modernize and simplify YMM logic

Pavel Labath via Phabricator via lldb-commits lldb-commits at lists.llvm.org
Fri Nov 13 02:11:19 PST 2020

labath accepted this revision.
labath added inline comments.

Comment at: lldb/source/Plugins/Process/FreeBSDRemote/NativeRegisterContextFreeBSD_x86_64.cpp:615
+    return false;
mgorny wrote:
> labath wrote:
> > When does this return false? When the ymm regs are not available?
> Yes. Note that this change brings us closer to being able to disable whole regsets — I just need to figure out how to do it ;-).
Ok, got it.

Comment at: lldb/source/Plugins/Process/FreeBSDRemote/NativeRegisterContextFreeBSD_x86_64.h:35-38
+struct YMMSplitPtr {
+  void *xmm;
+  void *ymm_hi;
Declare this inside the class (next to the function returning it), since it's really a private thing.

Comment at: lldb/source/Plugins/Process/FreeBSDRemote/NativeRegisterContextFreeBSD_x86_64.h:83
   size_t GetDBROffset() const;
+  bool GetYMMSplitReg(uint32_t reg, void **xmm, void **ymm_hi);
labath wrote:
> mgorny wrote:
> > labath wrote:
> > > `void *&`
> > Are you sure about this? I feel like the whole `&` deal is quite confusing (read: shouldn't have happened). You don't know whether the method modifies the variables passed to it unless you look at the method prototype.
> well.. there certainly are code styles which prohibit non-const reference arguments. However, llvm is not one of them. Also lldb has a issue with compulsive null checks, so I think it's important to communicate the fact the two arguments cannot be null, which reference arguments do.
> Another possibility would be to just make this a real return value (pair<void *, void*> ?). The downside there is that it's not as self-documenting. Pick your poison...
Ok, that feels a bit over-engineered, but works too...



More information about the lldb-commits mailing list