[NVPTX] Add lowering of i128 numbers as struct fields

[denzp]

Summary

Addition to D34555 - override VTs computation with ComputePTXValueVTs for struct fields.

[tra]

lib/Target/NVPTX/NVPTXISelLowering.cpp

186–188 ↗

(On Diff #176188)

for(auto const *EI : STy->elements()) ?

[denzp]

@tra thanks for the review! If it looks good now, could you please commit it?

[denzp]

We still need to know the current element index, so there is an outer "counter". Anyway, the code looks more readable now!

[tra]

@denzp: FYI. Apparently we still can't handle [1 x i128]

https://bugs.llvm.org/show_bug.cgi?id=49877

llvm/trunk/lib/Target/NVPTX/NVPTXISelLowering.cpp

if (Ty->isIntegerTy(128)) {
     if (Offsets) {
       Offsets->push_back(StartingOffset + 0);
       Offsets->push_back(StartingOffset + 8);
     }
 
     return;
   }
 
+  // Given a struct type, recursively traverse the elements with custom ComputePTXValueVTs.
+  if (StructType *STy = dyn_cast<StructType>(Ty)) {
+    auto const *SL = DL.getStructLayout(STy);
+    auto ElementNum = 0;
+    for(auto *EI : STy->elements()) {
+      ComputePTXValueVTs(TLI, DL, EI, ValueVTs, Offsets,
+                         StartingOffset + SL->getElementOffset(ElementNum));
+      ++ElementNum;
+    }
+    return;
+  }
+
   ComputeValueVTs(TLI, DL, Ty, TempVTs, &TempOffsets, StartingOffset);
   for (unsigned i = 0, e = TempVTs.size(); i != e; ++i) {
     EVT VT = TempVTs[i];
     uint64_t Off = TempOffsets[i];
     // Split vectors into individual elements, except for v2f16, which
     // we will pass as a single scalar.
     if (VT.isVector()) {
       unsigned NumElts = VT.getVectorNumElements();

llvm/trunk/test/CodeGen/NVPTX/i128-struct.ll

+; RUN: llc < %s -O0 -march=nvptx64 -mcpu=sm_20 | FileCheck %s
+
+; CHECK-LABEL: .visible .func (.param .align 16 .b8 func_retval0[32]) foo(
+define { i128, i128 } @foo(i64 %a, i32 %b) {
+  %1 = sext i64 %a to i128
+  %2 = sext i32 %b to i128
+  %3 = insertvalue { i128, i128 } undef, i128 %1, 0
+  %4 = insertvalue { i128, i128 } %3, i128 %2, 1
+
+  ; CHECK: st.param.v2.b64 [func_retval0+0],  {%[[REG1:rd[0-9]+]], %[[REG2:rd[0-9]+]]};
+  ; CHECK: st.param.v2.b64 [func_retval0+16], {%[[REG3:rd[0-9]+]], %[[REG4:rd[0-9]+]]};
+  ret { i128, i128 } %4
+}