[PATCH] D18049: AMDGPU i16 implementation

[Tom Stellard via llvm-commits]

tstellarAMD added a comment.

I tried to apply this patch, but I get a lot of failing lit tests in test/CodeGen/AMDGPU.  Do all the tests in this directory pass for you?


================
Comment at: lib/Target/AMDGPU/SIISelLowering.cpp:65
@@ -64,1 +64,3 @@
 
+  // TODO: Subtarget feature for i16
+  if (Subtarget->has16BitInsts())
----------------
Do we still need this comment?

================
Comment at: lib/Target/AMDGPU/SIISelLowering.cpp:1740
@@ -1683,3 +1739,3 @@
   if (ExtType == ISD::NON_EXTLOAD && MemVT.getSizeInBits() < 32) {
-    assert(MemVT == MVT::i1 && "Only i1 non-extloads expected");
+    //assert(MemVT == MVT::i1 && "Only i1 non-extloads expected");
     // FIXME: Copied from PPC
----------------
Does this need to be removed?

================
Comment at: lib/Target/AMDGPU/SIISelLowering.cpp:1748
@@ -1691,1 +1747,3 @@
 
+    EVT mem_vt = (MemVT == MVT::i1) ? MVT::i8 : MVT::i16;
+
----------------
Coding style.  Variable names should start with a captial.


Repository:
  rL LLVM

http://reviews.llvm.org/D18049