[PATCH] D17308: AMDGPU: Change how alignment is computed for argument lowering

[Matt Arsenault via llvm-commits]

arsenm added inline comments.

================
Comment at: lib/Target/AMDGPU/SIISelLowering.cpp:565-566
@@ +564,4 @@
+
+  SDValue Load = DAG.getLoad(
+    ISD::UNINDEXED, ExtTy,
+    LoadVT, SL, Chain, Ptr, PtrOffset, PtrInfo, MemVT,
----------------
You should be able to use one of the simpler getLoad operands that doesn't include the ISD::UNINDEXED

================
Comment at: lib/Target/AMDGPU/SIISelLowering.cpp:1512
@@ -1491,1 +1511,3 @@
+
+  // FIXME: Get alignment from somewhere. Base pointer should be 256 aligned.
   SDValue Param = LowerParameter(DAG, MVT::i32, MVT::i32, SL,
----------------
This should be the same MinAlign(Subtarget->getKernargSegmentPtrAlignment(), Offset)

================
Comment at: lib/Target/AMDGPU/SIISelLowering.cpp:1600-1601
@@ -1577,4 +1599,4 @@
 
     return LowerParameter(DAG, VT, VT, DL, DAG.getEntryNode(),
-                          SI::KernelInputOffsets::NGROUPS_X, false);
+                          SI::KernelInputOffsets::NGROUPS_X, 4, false);
   case Intrinsic::r600_read_ngroups_y:
----------------
Should be the same MinAlign again


Repository:
  rL LLVM

http://reviews.llvm.org/D17308