[llvm] [RISCV][llvm] Preliminary P extension codegen support (PR #162668)

[Craig Topper via llvm-commits]

================
@@ -1455,3 +1455,127 @@ let Predicates = [HasStdExtP, IsRV32] in {
   def PMAXU_DW     : RVPPairBinaryExchanged_rr<0b1111, 0b01, "pmaxu.dw">;
   def PMAXU_DB     : RVPPairBinaryExchanged_rr<0b1111, 0b10, "pmaxu.db">;
 } // Predicates = [HasStdExtP, IsRV32]
+
+let Predicates = [HasStdExtP, IsRV64] in {
+  // Basic arithmetic patterns for v4i16 (16-bit elements in 64-bit GPR)
+  def: Pat<(v4i16 (add v4i16:$rs1, v4i16:$rs2)), (!cast<Instruction>("PADD_H") GPR:$rs1, GPR:$rs2)>;
+  def: Pat<(v4i16 (sub v4i16:$rs1, v4i16:$rs2)), (!cast<Instruction>("PSUB_H") GPR:$rs1, GPR:$rs2)>;
+  
+  // Saturating add/sub patterns for v4i16
+  def: Pat<(v4i16 (saddsat v4i16:$rs1, v4i16:$rs2)), (!cast<Instruction>("PSADD_H") GPR:$rs1, GPR:$rs2)>;
+  def: Pat<(v4i16 (uaddsat v4i16:$rs1, v4i16:$rs2)), (!cast<Instruction>("PSADDU_H") GPR:$rs1, GPR:$rs2)>;
+  def: Pat<(v4i16 (ssubsat v4i16:$rs1, v4i16:$rs2)), (!cast<Instruction>("PSSUB_H") GPR:$rs1, GPR:$rs2)>;
+  def: Pat<(v4i16 (usubsat v4i16:$rs1, v4i16:$rs2)), (!cast<Instruction>("PSSUBU_H") GPR:$rs1, GPR:$rs2)>;
+  
+  // Averaging patterns for v4i16
+  def: Pat<(v4i16 (avgfloors v4i16:$rs1, v4i16:$rs2)), (!cast<Instruction>("PAADD_H") GPR:$rs1, GPR:$rs2)>;
+  def: Pat<(v4i16 (avgflooru v4i16:$rs1, v4i16:$rs2)), (!cast<Instruction>("PAADDU_H") GPR:$rs1, GPR:$rs2)>;
+  
+  // Averaging subtraction patterns for v4i16  
+  // PASUB_H: signed (a - b) >> 1
----------------
topperc wrote:

This is not the definition of averaging subtract.

It's (trunc ((sext(a) - sext(b)) >> 1)). You need to have at least 17 bits for the subtract to avoid overflow.

https://github.com/llvm/llvm-project/pull/162668