[PATCH] TargetParser prototype - NFC (ARM only)

John Brawn john.brawn at arm.com
Fri May 1 10:21:32 PDT 2015

A general question: how far are you planning to go with this, in particular are you going to sort out cpu <-> arch mapping (which is duplicated in several places)?

I ask because I'm currently thinking about adjusting the ARM -mcpu and -march handling in clang to be more like AArch64, in particular making -march=garbage not mean -mcpu=arm7tdmi (and instead give an error), and it's be nice to not collide with what you're doing.

Comment at: include/llvm/Support/ARMTargetParser.h:90-91
@@ +89,4 @@
+class ARMTargetParser {
+  static StringRef getFPUSynonim(StringRef FPU);
+  static StringRef getArchSynonim(StringRef Arch);

Comment at: lib/Support/ARMTargetParser.cpp:157-158
@@ +156,4 @@
+    .Case("vfp4-d16", "vfpv4-d16")
+    .Cases("fp4-sp-d16", "fpv4-sp-d16", "vfpv4-d16")
+    .Cases("fp5-sp-d16", "fpv5-sp-d16", "fpv5-d16")
+    .Case("fp5-dp-d16", "fpv5-d16")
The -sp-d16 FP variants aren't synonyms for the -d16 variants. This preserves the current behaviour, where clang -S -mfpu=fpv5-sp-d16 gets you .fpu fpv5-d16 (which doesn't seem right to me), but it could at least do with a comment.



More information about the cfe-commits mailing list