[clang] [llvm] [RISCV] Support ZVqdot Codegen and C intrinsics (PR #154915)

[Brandon Wu via cfe-commits]

================
@@ -52,10 +52,71 @@ let Predicates = [HasStdExtZvqdotq], mayLoad = 0, mayStore = 0,
   defm PseudoVQDOT : VPseudoVQDOT_VV_VX;
   defm PseudoVQDOTU : VPseudoVQDOT_VV_VX;
   defm PseudoVQDOTSU : VPseudoVQDOT_VV_VX;
+  // VQDOTUS does not have a VV variant
+  foreach m = MxListVF4 in {
+    defm "PseudoVQDOTUS_VX" : VPseudoTernaryWithPolicy<m.vrclass, m.vrclass, GPR, m>;
+  }
 }
 
 defvar AllE32Vectors = [VI32MF2, VI32M1, VI32M2, VI32M4, VI32M8];
 defm : VPatBinaryVL_VV_VX<riscv_vqdot_vl, "PseudoVQDOT", AllE32Vectors>;
 defm : VPatBinaryVL_VV_VX<riscv_vqdotu_vl, "PseudoVQDOTU", AllE32Vectors>;
 defm : VPatBinaryVL_VV_VX<riscv_vqdotsu_vl, "PseudoVQDOTSU", AllE32Vectors>;
 
+// These VPat definitions are for vqdot because they have a different operand
+// order with other ternary instructions (i.e. vop.vx vd, vs2, rs1)
+multiclass VPatTernaryV_VX_AAXASwapped<string intrinsic, string instruction,
----------------
4vtomat wrote:

you are right, the name AABX should be the proper one lol

https://github.com/llvm/llvm-project/pull/154915