Added support for negative multiply add

Author: badumbatish

llvm/include/llvm/Target/TargetSelectionDAG.td

@@ -1147,6 +1147,9 @@ def fadd_contract : PatFrag<(ops node:$a, node:$b), (fadd node:$a, node:$b),[{
   return N->getFlags().hasAllowContract();
 }]>;
 
+def fsub_contract : PatFrag<(ops node:$a, node:$b), (fsub node:$a, node:$b),[{
+  return N->getFlags().hasAllowContract();
+}]>;
 
 def not  : PatFrag<(ops node:$in), (xor node:$in, -1)>;
 def vnot : PatFrag<(ops node:$in), (xor node:$in, immAllOnesV)>;

llvm/lib/Target/WebAssembly/WebAssemblyInstrSIMD.td

@@ -1540,6 +1540,8 @@ multiclass SIMDMADD<Vec vec, bits<32> simdopA, bits<32> simdopS, list<Predicate>
   def : Pat<(fadd_contract (vec.vt V128:$a), (fmul_contract (vec.vt V128:$b), (vec.vt V128:$c))),
              (!cast<Instruction>("MADD_"#vec) V128:$a, V128:$b, V128:$c)>, Requires<[HasRelaxedSIMD]>;
 
+  def : Pat<(fsub_contract (vec.vt V128:$a), (fmul_contract (vec.vt V128:$b), (vec.vt V128:$c))),
+             (!cast<Instruction>("NMADD_"#vec) V128:$a, V128:$b, V128:$c)>, Requires<[HasRelaxedSIMD]>;
 }
 
 defm "" : SIMDMADD<F32x4, 0x105, 0x106, [HasRelaxedSIMD]>;

llvm/test/CodeGen/WebAssembly/simd-relaxed-fnma.ll

@@ -27,9 +27,8 @@ define <4 x float> @fsub_fmul_contract_4xf32(<4 x float> %a, <4 x float> %b, <4
 ; RELAXED-LABEL: fsub_fmul_contract_4xf32:
 ; RELAXED:         .functype fsub_fmul_contract_4xf32 (v128, v128, v128) -> (v128)
 ; RELAXED-NEXT:  # %bb.0:
-; RELAXED-NEXT:    f32x4.mul $push0=, $1, $0
-; RELAXED-NEXT:    f32x4.sub $push1=, $2, $pop0
-; RELAXED-NEXT:    return $pop1
+; RELAXED-NEXT:    f32x4.relaxed_nmadd $push0=, $2, $1, $0
+; RELAXED-NEXT:    return $pop0
 ;
 ; STRICT-LABEL: fsub_fmul_contract_4xf32:
 ; STRICT:         .functype fsub_fmul_contract_4xf32 (v128, v128, v128) -> (v128)
@@ -47,9 +46,8 @@ define <8 x half> @fsub_fmul_contract_8xf16(<8 x half> %a, <8 x half> %b, <8 x h
 ; RELAXED-LABEL: fsub_fmul_contract_8xf16:
 ; RELAXED:         .functype fsub_fmul_contract_8xf16 (v128, v128, v128) -> (v128)
 ; RELAXED-NEXT:  # %bb.0:
-; RELAXED-NEXT:    f16x8.mul $push0=, $1, $0
-; RELAXED-NEXT:    f16x8.sub $push1=, $2, $pop0
-; RELAXED-NEXT:    return $pop1
+; RELAXED-NEXT:    f16x8.relaxed_nmadd $push0=, $2, $1, $0
+; RELAXED-NEXT:    return $pop0
 ;
 ; STRICT-LABEL: fsub_fmul_contract_8xf16:
 ; STRICT:         .functype fsub_fmul_contract_8xf16 (v128, v128, v128) -> (v128)
@@ -86,12 +84,10 @@ define <8 x float> @fsub_fmul_contract_8xf32(<8 x float> %a, <8 x float> %b, <8
 ; RELAXED-LABEL: fsub_fmul_contract_8xf32:
 ; RELAXED:         .functype fsub_fmul_contract_8xf32 (i32, v128, v128, v128, v128, v128, v128) -> ()
 ; RELAXED-NEXT:  # %bb.0:
-; RELAXED-NEXT:    f32x4.mul $push0=, $4, $2
-; RELAXED-NEXT:    f32x4.sub $push1=, $6, $pop0
-; RELAXED-NEXT:    v128.store 16($0), $pop1
-; RELAXED-NEXT:    f32x4.mul $push2=, $3, $1
-; RELAXED-NEXT:    f32x4.sub $push3=, $5, $pop2
-; RELAXED-NEXT:    v128.store 0($0), $pop3
+; RELAXED-NEXT:    f32x4.relaxed_nmadd $push0=, $6, $4, $2
+; RELAXED-NEXT:    v128.store 16($0), $pop0
+; RELAXED-NEXT:    f32x4.relaxed_nmadd $push1=, $5, $3, $1
+; RELAXED-NEXT:    v128.store 0($0), $pop1
 ; RELAXED-NEXT:    return
 ;
 ; STRICT-LABEL: fsub_fmul_contract_8xf32:
@@ -114,9 +110,8 @@ define <2 x double> @fsub_fmul_contract_2xf64(<2 x double> %a, <2 x double> %b,
 ; RELAXED-LABEL: fsub_fmul_contract_2xf64:
 ; RELAXED:         .functype fsub_fmul_contract_2xf64 (v128, v128, v128) -> (v128)
 ; RELAXED-NEXT:  # %bb.0:
-; RELAXED-NEXT:    f64x2.mul $push0=, $1, $0
-; RELAXED-NEXT:    f64x2.sub $push1=, $2, $pop0
-; RELAXED-NEXT:    return $pop1
+; RELAXED-NEXT:    f64x2.relaxed_nmadd $push0=, $2, $1, $0
+; RELAXED-NEXT:    return $pop0
 ;
 ; STRICT-LABEL: fsub_fmul_contract_2xf64:
 ; STRICT:         .functype fsub_fmul_contract_2xf64 (v128, v128, v128) -> (v128)