[PATCH] D12089: Implement __aeabi_c{d,f}{cmpeq,cmple,rcmple}.

[Stephen Hines via llvm-commits]

srhines added inline comments.

================
Comment at: lib/builtins/arm/aeabi_cdcmp.S:20
@@ +19,3 @@
+// void __aeabi_cdcmpeq(double a, double b) {
+//   // __aeabi_cdcmpeq_helper takes the most significant words of a, b
+//   if (__aeabi_cdcmpeq(a, b)) { Z = 0, C = 1 }
----------------
Double comment on this line, and the rest of the lines are formatted strangely.

================
Comment at: lib/builtins/arm/aeabi_cdcmp.S:46
@@ +45,3 @@
+//   if (__aeabi_dcmplt(a, b)) {
+//     Z = 1; C = 0;
+//   } else if (__aeabi_dcmpeq(a, b)) {
----------------
These comments don't match the implementation below, so something seems wrong.

================
Comment at: lib/builtins/arm/aeabi_cdcmpeq_helper.c:19
@@ +18,3 @@
+int __aeabi_cdcmpeq_helper(int32_t a, int32_t b) {
+    return isnan_d(a) || isnan_d(b);
+}
----------------
Couldn't this be using __builtin_isnan instead of redefining it?

================
Comment at: lib/builtins/arm/aeabi_cfcmpeq_helper.c:19
@@ +18,3 @@
+int __aeabi_cfcmpeq_helper(int32_t a, int32_t b) {
+    return isnan_f(a) || isnan_f(b);
+}
----------------
I have the same comment about using __builtin_isnan() here too.


Repository:
  rL LLVM

http://reviews.llvm.org/D12089