From 124f2bc5c99e27152ca20916857df2d870cba359 Mon Sep 17 00:00:00 2001
From: Ilya Churaev <ilya.churaev@intel.com>
Date: Tue, 24 Oct 2023 19:19:16 +0400
Subject: [PATCH] Remove NF4 Convert from public API (#20666)

* Remove NF4 Convert from public API

* Fixed build
---
 src/core/{include => dev_api}/openvino/core/type/nf4.hpp  | 0
 src/core/include/openvino/core/type/element_type.hpp      | 1 -
 src/core/include/openvino/op/constant.hpp                 | 7 ++++---
 src/core/reference/include/openvino/reference/convert.hpp | 3 ++-
 src/core/src/op/constant.cpp                              | 7 +++++++
 5 files changed, 13 insertions(+), 5 deletions(-)
 rename src/core/{include => dev_api}/openvino/core/type/nf4.hpp (100%)

diff --git a/src/core/include/openvino/core/type/nf4.hpp b/src/core/dev_api/openvino/core/type/nf4.hpp
similarity index 100%
rename from src/core/include/openvino/core/type/nf4.hpp
rename to src/core/dev_api/openvino/core/type/nf4.hpp
diff --git a/src/core/include/openvino/core/type/element_type.hpp b/src/core/include/openvino/core/type/element_type.hpp
index 1534e9e0cc8..78e200d5035 100644
--- a/src/core/include/openvino/core/type/element_type.hpp
+++ b/src/core/include/openvino/core/type/element_type.hpp
@@ -20,7 +20,6 @@
 #include "openvino/core/rtti.hpp"
 #include "openvino/core/type/bfloat16.hpp"
 #include "openvino/core/type/float16.hpp"
-#include "openvino/core/type/nf4.hpp"
 
 /**
  * @defgroup ov_element_cpp_api Element types
diff --git a/src/core/include/openvino/op/constant.hpp b/src/core/include/openvino/op/constant.hpp
index 100ed2f7e18..6299dde4590 100644
--- a/src/core/include/openvino/op/constant.hpp
+++ b/src/core/include/openvino/op/constant.hpp
@@ -705,15 +705,15 @@ private:
         auto p = get_data_ptr_nc<Type>();
         size_t i = 0;
         for (; i < source.size() / 2; i++) {
-            const auto idx1 = ConvertNF4::quantize(static_cast<float>(source[i * 2]));
-            const auto idx2 = ConvertNF4::quantize(static_cast<float>(source[i * 2 + 1]));
+            const auto idx1 = quantize_nf4(static_cast<float>(source[i * 2]));
+            const auto idx2 = quantize_nf4(static_cast<float>(source[i * 2 + 1]));
             const auto v1 = value_in_range<Type>(idx1) & 0x0F;
             const auto v2 = value_in_range<Type>(idx2) & 0x0F;
             const auto v = (v2 << 4) | v1;
             p[i] = static_cast<StorageDataType>(v);
         }
         if (source.size() % 2) {
-            const auto idx1 = ConvertNF4::quantize(static_cast<float>(source[i * 2]));
+            const auto idx1 = quantize_nf4(static_cast<float>(source[i * 2]));
             const auto v = value_in_range<Type>(idx1) & 0x0F;
             p[i] = static_cast<StorageDataType>(v);
         }
@@ -853,6 +853,7 @@ private:
         }
         return shape_size(m_shape) * m_element_type.size();
     }
+    static uint8_t quantize_nf4(float x);
 
     element::Type m_element_type;
     Shape m_shape{};
diff --git a/src/core/reference/include/openvino/reference/convert.hpp b/src/core/reference/include/openvino/reference/convert.hpp
index bd36b50b033..3924ce69055 100644
--- a/src/core/reference/include/openvino/reference/convert.hpp
+++ b/src/core/reference/include/openvino/reference/convert.hpp
@@ -8,6 +8,7 @@
 
 #include "openvino/core/type/element_type.hpp"
 #include "openvino/core/type/float16.hpp"
+#include "openvino/core/type/nf4.hpp"
 
 namespace ov {
 namespace reference {
@@ -87,7 +88,7 @@ void lp_convert(const TI* arg, TO* out, size_t count, element::Type_t src_type,
         } else if (dst_type == element::i4) {
             detail::set_i4(output, i, detail::get_value<int8_t, TI>(input, i, src_type));
         } else if (src_type == element::nf4) {
-            ConvertNF4::unpack(out, input, i);
+            ov::ConvertNF4::unpack(out, input, i);
         } else {
             out[i] = detail::get_value<TO, TI>(input, i, src_type);
         }
diff --git a/src/core/src/op/constant.cpp b/src/core/src/op/constant.cpp
index 2fe3d024fd9..34e97d73eee 100644
--- a/src/core/src/op/constant.cpp
+++ b/src/core/src/op/constant.cpp
@@ -13,6 +13,9 @@
 #include "ngraph/runtime/aligned_buffer.hpp"
 #include "ngraph/runtime/host_tensor.hpp"
 #include "ngraph/runtime/tensor.hpp"
+#include "openvino/core/type/element_type.hpp"
+#include "openvino/core/type/float16.hpp"
+#include "openvino/core/type/nf4.hpp"
 #include "openvino/runtime/shared_buffer.hpp"
 
 template <typename T>
@@ -606,3 +609,7 @@ bool ov::op::v0::Constant::evaluate_lower(TensorVector& outputs) const {
 bool ov::op::v0::Constant::evaluate_upper(TensorVector& outputs) const {
     return evaluate(outputs, {});
 }
+
+uint8_t ov::op::v0::Constant::quantize_nf4(float x) {
+    return ov::ConvertNF4::quantize(x);
+}