From c894d481147ead3d26fa1c85bf3ccc94c5b6de14 Mon Sep 17 00:00:00 2001
From: "@ding-jing12" <@ding-jing12>
Date: Wed, 27 Nov 2024 15:51:05 +0800
Subject: [PATCH] =?UTF-8?q?=E6=B7=BB=E5=8A=A0precisonCompare=E5=AF=B9?=
 =?UTF-8?q?=E5=A4=96=E6=8E=A5=E5=8F=A3?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 tf_adapter/kernels/aicpu/npu_cpu_ops.cc             |  8 ++++++++
 tf_adapter/ops/aicpu/npu_cpu_ops.cc                 |  8 ++++++++
 tf_adapter/python/npu_bridge/npu_cpu/npu_cpu_ops.py | 11 +++++++++++
 3 files changed, 27 insertions(+)

diff --git a/tf_adapter/kernels/aicpu/npu_cpu_ops.cc b/tf_adapter/kernels/aicpu/npu_cpu_ops.cc
index ff9dc6d2f..39b2a3218 100644
--- a/tf_adapter/kernels/aicpu/npu_cpu_ops.cc
+++ b/tf_adapter/kernels/aicpu/npu_cpu_ops.cc
@@ -221,6 +221,13 @@ class ScatterElementsV2Op : public OpKernel {
   }
 };
 
+class PrecisionCompareOp : public OpKernel {
+ public:
+  explicit PrecisionCompareOp(OpKernelConstruction *context) : OpKernel(context) {}
+  ~PrecisionCompareOp() override {}
+  void Compute(OpKernelContext *context) override { ADP_LOG(INFO) << "PrecisionCompareOp Compute"; }
+};
+
 REGISTER_KERNEL_BUILDER(Name("ScatterElementsV2").Device(DEVICE_CPU), ScatterElementsV2Op);
 REGISTER_KERNEL_BUILDER(Name("EmbeddingRankId").Device(DEVICE_CPU), EmbeddingRankIdOpKernel);
 REGISTER_KERNEL_BUILDER(Name("EmbeddingLocalIndex").Device(DEVICE_CPU), EmbeddingLocalIndexOpKernel);
@@ -242,6 +249,7 @@ REGISTER_KERNEL_BUILDER(Name("NonZeroWithValueShape").Device(DEVICE_CPU), NonZer
 REGISTER_KERNEL_BUILDER(Name("NonZeroWithValueShapeV2").Device(DEVICE_CPU), NonZeroWithValueShapeV2Op);
 REGISTER_KERNEL_BUILDER(Name("WarpAffineV2").Device(DEVICE_CPU), WarpAffineV2Op);
 REGISTER_KERNEL_BUILDER(Name("ResizeV2").Device(DEVICE_CPU), ResizeV2Op);
+REGISTER_KERNEL_BUILDER(Name("PrecisionCompare").Device(DEVICE_CPU), PrecisionCompareOp);
 
 class DecodeImageV3Op : public OpKernel {
 public:
diff --git a/tf_adapter/ops/aicpu/npu_cpu_ops.cc b/tf_adapter/ops/aicpu/npu_cpu_ops.cc
index 40857b449..048ea0c04 100644
--- a/tf_adapter/ops/aicpu/npu_cpu_ops.cc
+++ b/tf_adapter/ops/aicpu/npu_cpu_ops.cc
@@ -627,4 +627,12 @@ REGISTER_OP("ResizeV2")
     .Output("image: dtype")
     .Attr("expand_animations: bool = true")
     .SetShapeFn(DecodeImageV3ShapeFn);
+
+    REGISTER_OP("PrecisionCompare")
+        .Input("x1: T")
+        .Input("x2: T")
+        .Output("y: uint32")
+        .Attr("detect_type: int = 0")
+        .Attr("T: {float32, float16, bfloat16}")
+        .SetShapeFn(tensorflow::shape_inference::ScalarShape);
 }  // namespace tensorflow
diff --git a/tf_adapter/python/npu_bridge/npu_cpu/npu_cpu_ops.py b/tf_adapter/python/npu_bridge/npu_cpu/npu_cpu_ops.py
index b247e96a9..c49ef9fc5 100644
--- a/tf_adapter/python/npu_bridge/npu_cpu/npu_cpu_ops.py
+++ b/tf_adapter/python/npu_bridge/npu_cpu/npu_cpu_ops.py
@@ -465,3 +465,14 @@ def host_feature_mapping_import(path):
     """ host feature mapping export. """
     result = gen_npu_cpu_ops.FeatureMappingImport(path=path)
     return result
+
+## 提供硬件精度检测功能
+#  @param golden 支持float/float16/bfloat16类型
+#  @param realdata 支持float/float16/bfloat16类型
+#  @param detect_type int32 类型, 取值范围：0/1/2
+#  @return uint32
+def precision_compare(golden, realdata, detect_type):
+    """ precision compare. """
+    result = gen_npu_cpu_ops.PrecisionCompare(
+        x1=golden, x2=realdata, detect_type=detect_type)
+    return result
-- 
Gitee