[PATCH] Propagate -fno-unwind-tables to EHABI

Logan Chien tzuhsiang.chien at gmail.com
Thu Mar 20 08:13:46 PDT 2014


  BTW, I found that I was wrong in the previous comment.  If we wish to keep the test cases unchanged, then we should emit unwinding information for the functions with neither `uwtable` nor `nounwind` attributes.  (We have this already in the diff#2.)

  But this will raise another problem: we are not going further with this patch.  We still can't control the availability of unwind table with an function attribute.  Maybe further consideration is required?  I have just read your mail in llvm-dev mailing list; however I am too busy to reply it now.  I will reply it as soon as possible.


================
Comment at: lib/CodeGen/AsmPrinter/ARMException.cpp:91
@@ -80,3 +90,3 @@
   ARMTargetStreamer &ATS = getTargetStreamer();
-  if (!Asm->MF->getFunction()->needsUnwindTableEntry())
+  if (Asm->MF->getFunction()->doesNotThrow())
     ATS.emitCantUnwind();
----------------
It seems that this line should not be changed?  Otherwise, the `uwtable+nounwind` case will emit cantunwind.


http://llvm-reviews.chandlerc.com/D3079



More information about the llvm-commits mailing list