[clang] [llvm] [Clang][RISCV] Add packed widening multiply intrinsics (PR #217534)

[Craig Topper via llvm-commits]

================
@@ -2791,6 +2789,26 @@ let append Predicates = [IsRV64] in {
   def : Pat<(v2i32 (mul GPR:$rs1, GPR:$rs2)),
             (PACK (MUL_W00 GPR:$rs1, GPR:$rs2), (MUL_W11 GPR:$rs1, GPR:$rs2))>;
 
+  // Packed widening multiply patterns.
+  def : Pat<(v4i16 (riscv_pwmul (v8i8 GPR:$rs1), (v8i8 GPR:$rs2))),
+            (PMUL_H_B01 (ZIP8P GPR:$rs1, GPR:$rs2),
----------------
topperc wrote:

The ZIP8P should be emitted with RISCVISD::PZIP from lowering. And we should have new ISD nodes for PMUL_H_B01/PMUL_W_H01/PMULSU_H_B00/PMULSU_W_H00.

https://github.com/llvm/llvm-project/pull/217534