[llvm] [LoongArch] Optimize codegen for ISD::ROTL (PR #100344)

[Craig Topper via llvm-commits]

topperc wrote:

Since LoongArch copied some code from RISC-V, I think you can do this with these 2 tablegen changes.

```
diff --git a/llvm/lib/Target/LoongArch/LoongArchInstrInfo.td b/llvm/lib/Target/LoongArch/LoongArchInstrInfo.td
index ec0d071453c3..030e3500d528 100644
--- a/llvm/lib/Target/LoongArch/LoongArchInstrInfo.td
+++ b/llvm/lib/Target/LoongArch/LoongArchInstrInfo.td
@@ -1119,7 +1119,7 @@ def : PatGprGpr<urem, MOD_WU>;
 def : PatGprGpr<mul, MUL_W>;
 def : PatGprGpr<mulhs, MULH_W>;
 def : PatGprGpr<mulhu, MULH_WU>;
-def : PatGprGpr<rotr, ROTR_W>;
+def : PatGprGpr<shiftop<rotr>, ROTR_W>;
 def : PatGprImm<rotr, ROTRI_W, uimm5>;
 
 foreach Idx = 1...3 in {
@@ -1143,7 +1143,7 @@ def : PatGprGpr_32<srem, MOD_W>;
 def : PatGprGpr<urem, MOD_DU>;
 def : PatGprGpr<loongarch_mod_wu, MOD_WU>;
 def : PatGprGpr<rotr, ROTR_D>;
-def : PatGprGpr<loongarch_rotr_w, ROTR_W>;
+def : PatGprGpr<shiftopw<loongarch_rotr_w>, ROTR_W>;
 def : PatGprImm<rotr, ROTRI_D, uimm6>;
 def : PatGprImm_32<rotr, ROTRI_W, uimm5>;
 def : PatGprImm<loongarch_rotr_w, ROTRI_W, uimm5>;
```

https://github.com/llvm/llvm-project/pull/100344