[PATCH] [InstCombine][X86] Improved folding of calls to Intrinsic::x86_sse4a_insertqi.

Quentin Colombet qcolombet at apple.com
Tue Dec 9 13:18:05 PST 2014


Hi Andrea,

LGTM.

Thanks,
-Quentin

================
Comment at: lib/Transforms/InstCombine/InstCombineCalls.cpp:779
@@ +778,3 @@
+        unsigned Index = CIStart->getZExtValue();
+        // A value of zero in the field length is defined as length of 64.
+        unsigned Length = CIWidth->equalsInt(0) ? 64 : CIWidth->getZExtValue();
----------------
Maybe add: per AMD64 manual, or something like that.

http://reviews.llvm.org/D6583






More information about the llvm-commits mailing list