[WebAssembly] Expand BUILD_PAIR nodes

[aheejin]

LGTM with nits

llvm/test/CodeGen/WebAssembly/build-pair.ll

1 ↗

(On Diff #183676)

Maybe rename this file to simd-build-pair.ll or something to be clear this is about SIMD?

4 ↗

(On Diff #183676)

No need for -wasm

[aheejin]

llvm/test/CodeGen/WebAssembly/build-pair.ll

6 ↗

(On Diff #183676)

Adding DAG printed by -debug or some explanation on which part of the code becomes build_pair might help reading.

llvm/trunk/lib/Target/WebAssembly/WebAssemblyISelLowering.cpp

for (auto T : {MVT::v16i8, MVT::v8i16, MVT::v4i32, MVT::v2i64, MVT::v4f32,
           setTruncStoreAction(T, MemT, Expand);
           for (auto Ext : {ISD::EXTLOAD, ISD::ZEXTLOAD, ISD::SEXTLOAD})
             setLoadExtAction(Ext, T, MemT, Expand);
         }
       }
     }
   }
 
+  // Don't do anything clever with build_pairs
+  setOperationAction(ISD::BUILD_PAIR, MVT::i64, Expand);
+
   // Trap lowers to wasm unreachable
   setOperationAction(ISD::TRAP, MVT::Other, Legal);
 
   // Exception handling intrinsics
   setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::Other, Custom);
   setOperationAction(ISD::INTRINSIC_VOID, MVT::Other, Custom);
 
   setMaxAtomicSizeInBitsSupported(64);

llvm/trunk/test/CodeGen/WebAssembly/simd-build-pair.ll

+; RUN: llc < %s -asm-verbose=false -verify-machineinstrs -disable-wasm-fallthrough-return-opt -wasm-disable-explicit-locals -wasm-keep-registers -mattr=simd128 | FileCheck %s --check-prefixes CHECK
+
+target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
+target triple = "wasm32-unknown-unknown"
+
+; Test that BUILD_PAIR dag nodes are correctly lowered.
+; This code produces a selection DAG like the following:
+
+;    t0: ch = EntryToken
+;  t3: v4i32,ch = load<(load 16 from `<4 x i32>* undef`)> t0, undef:i32, undef:i32
+;        t30: i32 = extract_vector_elt t3, Constant:i32<2>
+;        t28: i32 = extract_vector_elt t3, Constant:i32<3>
+;      t24: i64 = build_pair t30, t28
+;    t8: ch = store<(store 8 into `i64* undef`, align 1)> t3:1, t24, undef:i32, undef:i32
+;  t9: ch = WebAssemblyISD::RETURN t8
+
+; CHECK:      i32x4.extract_lane
+; CHECK-NEXT: i64.extend_i32_u
+; CHECK-NEXT: i32x4.extract_lane
+; CHECK-NEXT: i64.extend_i32_u
+; CHECK-NEXT: i64.const {{.*}} 32
+; CHECK-NEXT: i64.shl
+; CHECK-NEXT: i64.or
+; CHECK-NEXT: i64.store
+define void @build_pair_i32s() {
+entry:
+  %0 = load <4 x i32>, <4 x i32>* undef, align 16
+  %shuffle.i184 = shufflevector <4 x i32> %0, <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
+  %bc357 = bitcast <4 x i32> %shuffle.i184 to <2 x i64>
+  %1 = extractelement <2 x i64> %bc357, i32 0
+  store i64 %1, i64* undef, align 1
+  ret void
+}