jojo added inline comments.

================
Comment at: lib/Support/TargetParser.cpp:441
@@ +440,3 @@
+  if (Extensions & AArch64::AEK_PROFILE)
+    Features.push_back("+spe");
+
----------------
bsmith wrote:
> For ARM there is a table that defines these extensions and how they map to 
> backend features, it would be good to do this in a similar manner.
There is a similar one for aarch64.But only in the context of switch-case,they 
can be map to backend features by table look-up. eg,getArchExtName().
We need to judge the value of "Extensions" by means of bit operations here.It 
is not possible to use the table.

================
Comment at: lib/Support/TargetParser.cpp:770
@@ +769,3 @@
+      if (A.ID == ARM::AK_ARMV8_2A)
+        Features.push_back("+v8.2a");
+      return A.ID;
----------------
bsmith wrote:
> Why do we need to add these features explicitly, can't we just pass through 
> the correct triple?
Sometimes people do not give a complete and standardized triple, which leads to 
the obtaining of these features by triple impossible.
It maybe safer to add these features by "-march".


Repository:
  rL LLVM

http://reviews.llvm.org/D20089



_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to