[LoongArch] Add codegen support for [X]VF{MSUB/NMADD/NMSUB}.{S/D} instructions (#74819)

This is similar to single and double-precision floating-point
instructions.

Author: wangleiat

llvm/lib/Target/LoongArch/LoongArchLASXInstrInfo.td

@@ -1455,6 +1455,32 @@ def : Pat<(fma v8f32:$xj, v8f32:$xk, v8f32:$xa),
 def : Pat<(fma v4f64:$xj, v4f64:$xk, v4f64:$xa),
           (XVFMADD_D v4f64:$xj, v4f64:$xk, v4f64:$xa)>;
 
+// XVFMSUB_{S/D}
+def : Pat<(fma v8f32:$xj, v8f32:$xk, (fneg v8f32:$xa)),
+          (XVFMSUB_S v8f32:$xj, v8f32:$xk, v8f32:$xa)>;
+def : Pat<(fma v4f64:$xj, v4f64:$xk, (fneg v4f64:$xa)),
+          (XVFMSUB_D v4f64:$xj, v4f64:$xk, v4f64:$xa)>;
+
+// XVFNMADD_{S/D}
+def : Pat<(fneg (fma v8f32:$xj, v8f32:$xk, v8f32:$xa)),
+          (XVFNMADD_S v8f32:$xj, v8f32:$xk, v8f32:$xa)>;
+def : Pat<(fneg (fma v4f64:$xj, v4f64:$xk, v4f64:$xa)),
+          (XVFNMADD_D v4f64:$xj, v4f64:$xk, v4f64:$xa)>;
+def : Pat<(fma_nsz (fneg v8f32:$xj), v8f32:$xk, (fneg v8f32:$xa)),
+          (XVFNMADD_S v8f32:$xj, v8f32:$xk, v8f32:$xa)>;
+def : Pat<(fma_nsz (fneg v4f64:$xj), v4f64:$xk, (fneg v4f64:$xa)),
+          (XVFNMADD_D v4f64:$xj, v4f64:$xk, v4f64:$xa)>;
+
+// XVFNMSUB_{S/D}
+def : Pat<(fneg (fma v8f32:$xj, v8f32:$xk, (fneg v8f32:$xa))),
+          (XVFNMSUB_S v8f32:$xj, v8f32:$xk, v8f32:$xa)>;
+def : Pat<(fneg (fma v4f64:$xj, v4f64:$xk, (fneg v4f64:$xa))),
+          (XVFNMSUB_D v4f64:$xj, v4f64:$xk, v4f64:$xa)>;
+def : Pat<(fma_nsz (fneg v8f32:$xj), v8f32:$xk, v8f32:$xa),
+          (XVFNMSUB_S v8f32:$xj, v8f32:$xk, v8f32:$xa)>;
+def : Pat<(fma_nsz (fneg v4f64:$xj), v4f64:$xk, v4f64:$xa),
+          (XVFNMSUB_D v4f64:$xj, v4f64:$xk, v4f64:$xa)>;
+
 // XVFSQRT_{S/D}
 defm : PatXrF<fsqrt, "XVFSQRT">;
 

llvm/lib/Target/LoongArch/LoongArchLSXInstrInfo.td

@@ -1555,6 +1555,32 @@ def : Pat<(fma v4f32:$vj, v4f32:$vk, v4f32:$va),
 def : Pat<(fma v2f64:$vj, v2f64:$vk, v2f64:$va),
           (VFMADD_D v2f64:$vj, v2f64:$vk, v2f64:$va)>;
 
+// VFMSUB_{S/D}
+def : Pat<(fma v4f32:$vj, v4f32:$vk, (fneg v4f32:$va)),
+          (VFMSUB_S v4f32:$vj, v4f32:$vk, v4f32:$va)>;
+def : Pat<(fma v2f64:$vj, v2f64:$vk, (fneg v2f64:$va)),
+          (VFMSUB_D v2f64:$vj, v2f64:$vk, v2f64:$va)>;
+
+// VFNMADD_{S/D}
+def : Pat<(fneg (fma v4f32:$vj, v4f32:$vk, v4f32:$va)),
+          (VFNMADD_S v4f32:$vj, v4f32:$vk, v4f32:$va)>;
+def : Pat<(fneg (fma v2f64:$vj, v2f64:$vk, v2f64:$va)),
+          (VFNMADD_D v2f64:$vj, v2f64:$vk, v2f64:$va)>;
+def : Pat<(fma_nsz (fneg v4f32:$vj), v4f32:$vk, (fneg v4f32:$va)),
+          (VFNMADD_S v4f32:$vj, v4f32:$vk, v4f32:$va)>;
+def : Pat<(fma_nsz (fneg v2f64:$vj), v2f64:$vk, (fneg v2f64:$va)),
+          (VFNMADD_D v2f64:$vj, v2f64:$vk, v2f64:$va)>;
+
+// VFNMSUB_{S/D}
+def : Pat<(fneg (fma v4f32:$vj, v4f32:$vk, (fneg v4f32:$va))),
+          (VFNMSUB_S v4f32:$vj, v4f32:$vk, v4f32:$va)>;
+def : Pat<(fneg (fma v2f64:$vj, v2f64:$vk, (fneg v2f64:$va))),
+          (VFNMSUB_D v2f64:$vj, v2f64:$vk, v2f64:$va)>;
+def : Pat<(fma_nsz (fneg v4f32:$vj), v4f32:$vk, v4f32:$va),
+          (VFNMSUB_S v4f32:$vj, v4f32:$vk, v4f32:$va)>;
+def : Pat<(fma_nsz (fneg v2f64:$vj), v2f64:$vk, v2f64:$va),
+          (VFNMSUB_D v2f64:$vj, v2f64:$vk, v2f64:$va)>;
+
 // VFSQRT_{S/D}
 defm : PatVrF<fsqrt, "VFSQRT">;