[WebAssembly] extadd_pairwise for PartialReduce (#157669)

Avoid using extends, and adding the high and low half and use
extadd_pairwise instead.

Author: sparker-arm

llvm/lib/Target/WebAssembly/WebAssemblyISD.def

@@ -27,6 +27,7 @@ HANDLE_NODETYPE(WrapperREL)
 HANDLE_NODETYPE(BR_IF)
 HANDLE_NODETYPE(BR_TABLE)
 HANDLE_NODETYPE(DOT)
+HANDLE_NODETYPE(EXT_ADD_PAIRWISE_U)
 HANDLE_NODETYPE(SHUFFLE)
 HANDLE_NODETYPE(SWIZZLE)
 HANDLE_NODETYPE(VEC_SHL)

llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.cpp

@@ -2183,13 +2183,10 @@ SDValue performLowerPartialReduction(SDNode *N, SelectionDAG &DAG) {
     SDValue MulLow = DAG.getNode(ISD::MUL, DL, MVT::v8i16, LowLHS, LowRHS);
     SDValue MulHigh = DAG.getNode(ISD::MUL, DL, MVT::v8i16, HighLHS, HighRHS);
 
-    SDValue LowLow = DAG.getNode(LowOpc, DL, MVT::v4i32, MulLow);
-    SDValue LowHigh = DAG.getNode(LowOpc, DL, MVT::v4i32, MulHigh);
-    SDValue HighLow = DAG.getNode(HighOpc, DL, MVT::v4i32, MulLow);
-    SDValue HighHigh = DAG.getNode(HighOpc, DL, MVT::v4i32, MulHigh);
-
-    SDValue AddLow = DAG.getNode(ISD::ADD, DL, MVT::v4i32, LowLow, HighLow);
-    SDValue AddHigh = DAG.getNode(ISD::ADD, DL, MVT::v4i32, LowHigh, HighHigh);
+    SDValue AddLow =
+        DAG.getNode(WebAssemblyISD::EXT_ADD_PAIRWISE_U, DL, MVT::v4i32, MulLow);
+    SDValue AddHigh = DAG.getNode(WebAssemblyISD::EXT_ADD_PAIRWISE_U, DL,
+                                  MVT::v4i32, MulHigh);
     SDValue Add = DAG.getNode(ISD::ADD, DL, MVT::v4i32, AddLow, AddHigh);
     return DAG.getNode(ISD::ADD, DL, MVT::v4i32, N->getOperand(1), Add);
   }

llvm/lib/Target/WebAssembly/WebAssemblyInstrSIMD.td

@@ -1453,15 +1453,22 @@ if !ne(t1, t2) then
 def : Pat<(t1.vt (bitconvert (t2.vt V128:$v))), (t1.vt V128:$v)>;
 
 // Extended pairwise addition
+def extadd_pairwise_u : SDNode<"WebAssemblyISD::EXT_ADD_PAIRWISE_U", extend_t>;
+
 defm "" : SIMDConvert<I16x8, I8x16, int_wasm_extadd_pairwise_signed,
                       "extadd_pairwise_i8x16_s", 0x7c>;
-defm "" : SIMDConvert<I16x8, I8x16, int_wasm_extadd_pairwise_unsigned,
+defm "" : SIMDConvert<I16x8, I8x16, extadd_pairwise_u,
                       "extadd_pairwise_i8x16_u", 0x7d>;
 defm "" : SIMDConvert<I32x4, I16x8, int_wasm_extadd_pairwise_signed,
                       "extadd_pairwise_i16x8_s", 0x7e>;
-defm "" : SIMDConvert<I32x4, I16x8, int_wasm_extadd_pairwise_unsigned,
+defm "" : SIMDConvert<I32x4, I16x8, extadd_pairwise_u,
                       "extadd_pairwise_i16x8_u", 0x7f>;
 
+def : Pat<(v4i32 (int_wasm_extadd_pairwise_unsigned (v8i16 V128:$in))),
+          (extadd_pairwise_u_I32x4 V128:$in)>;
+def : Pat<(v8i16 (int_wasm_extadd_pairwise_unsigned (v16i8 V128:$in))),
+          (extadd_pairwise_u_I16x8 V128:$in)>;
+
 // f64x2 <-> f32x4 conversions
 def demote_t : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisVec<1>]>;
 def demote_zero : SDNode<"WebAssemblyISD::DEMOTE_ZERO", demote_t>;

llvm/test/CodeGen/WebAssembly/int-mac-reduction-loops.ll

@@ -191,13 +191,9 @@ define hidden i32 @i32_mac_u8(ptr nocapture noundef readonly %a, ptr nocapture n
 ; MAX-BANDWIDTH: v128.load
 ; MAX-BANDWIDTH: v128.load
 ; MAX-BANDWIDTH: i16x8.extmul_low_i8x16_u
-; MAX-BANDWIDTH: i32x4.extend_low_i16x8_u
-; MAX-BANDWIDTH: i32x4.extend_high_i16x8_u
-; MAX-BANDWIDTH: i32x4.add
+; MAX-BANDWIDTH: i32x4.extadd_pairwise_i16x8_u
 ; MAX-BANDWIDTH: i16x8.extmul_high_i8x16_u
-; MAX-BANDWIDTH: i32x4.extend_low_i16x8_u
-; MAX-BANDWIDTH: i32x4.extend_high_i16x8_u
-; MAX-BANDWIDTH: i32x4.add
+; MAX-BANDWIDTH: i32x4.extadd_pairwise_i16x8_u
 ; MAX-BANDWIDTH: i32x4.add
 ; MAX-BANDWIDTH: i32x4.add