[llvm] [AMDGPU][True16][CodeGen] Add patterns to reduce intermediates (PR #162047)

[Brox Chen via llvm-commits]

================
@@ -3717,6 +3717,23 @@ def : GCNPat <
 } // End foreach Ty = ...
 } // End AddedComplexity = 1
 
+let True16Predicate = UseRealTrue16Insts in {
+def : GCNPat<
+  (i32 (DivergentBinFrag<or>
+    (i32 (zext i16:$src_lo)),
+    (i32 (bitconvert (v2i16 (build_vector (i16 0), (i16 VGPR_32:$src_hi)))))
+  )),
+  (REG_SEQUENCE VGPR_32, $src_lo, lo16, $src_hi, hi16)
+>;
+def : GCNPat<
+  (i32 (DivergentBinFrag<or>
+    (i32 (bitconvert (v2i16 (build_vector (i16 0), (i16 VGPR_32:$src_hi))))),
----------------
broxigarchen wrote:

same as above

https://github.com/llvm/llvm-project/pull/162047