[LLGA] copy the quantizer_ when updating LlgaTensorDesc from a logical_tensor (#136)

* [LLGA] set quantizer after query from compilation

* [LLGA] copy the quantizer_ when updating LlgaTensorDesc from a logical_tensor

* fix clang format

Author: chunyuan-w

torch_ipex/csrc/LlgaTensorImpl.h

@@ -102,6 +102,10 @@ struct LlgaTensorDesc {
     return ret;
   }
 
+  LlgaTensorDesc update_desc(const desc &t) const {
+    return LlgaTensorDesc(t).set_quantizer(quantizer_);
+  }
+
   QuantizerPtr get_quantizer() {
     return quantizer_;
   }

torch_ipex/csrc/jit/codegen/onednn/kernel.cpp

@@ -172,7 +172,8 @@ compiled_partition LlgaKernel::compile(const partition &partition) {
   // we need to query them out from compilation and update outputSpecs
   for (size_t i = 0; i < nOutputs_; i++) {
     auto tid = outputSpecs_[i].tid();
-    outputSpecs_[i] = compilation.query_logical_tensor(tid);
+    outputSpecs_[i] =
+        outputSpecs_[i].update_desc(compilation.query_logical_tensor(tid));
   }
 
   // Build static mapping from output id to input offset