COMPMID-3271: Add support for QASYMM8_SIGNED in CPPTopKVKernel/CPPTopKV

Signed-off-by: Sheri Zhang <sheri.zhang@arm.com>
Change-Id: Ic34616fc3480ca85cc582e4e3db031d631ed5861
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/2887
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>

diff --git a/arm_compute/core/CPP/kernels/CPPTopKVKernel.h b/arm_compute/core/CPP/kernels/CPPTopKVKernel.h
index fc62b42ab87d83f23c63e63c0b508bcb7c34ff0b..4b9bfdd3c9cce65c2620a8c97ef99661c4100288 100644 (file)
--- a/arm_compute/core/CPP/kernels/CPPTopKVKernel.h
+++ b/arm_compute/core/CPP/kernels/CPPTopKVKernel.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019 ARM Limited.
+ * Copyright (c) 2019-2020 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -53,7 +53,7 @@ public:
 
     /** Set the input and output of the kernel.
      *
-     * @param[in]  predictions A batch_size x classes tensor. Data types supported: F16/S32/F32/QASYMM8
+     * @param[in]  predictions A batch_size x classes tensor. Data types supported: F16/S32/F32/QASYMM8/QASYMM8_SIGNED
      * @param[in]  targets     A batch_size 1D tensor of class ids. Data types supported: S32
      * @param[out] output      Computed precision at @p k as a bool 1D tensor. Data types supported: U8
      * @param[in]  k           Number of top elements to look at for computing precision.
@@ -62,7 +62,7 @@ public:
 
     /** Static function to check if given info will lead to a valid configuration of @ref CPPTopKVKernel
      *
-     * @param[in] predictions A batch_size x classes tensor info. Data types supported: F16/S32/F32/QASYMM8
+     * @param[in] predictions A batch_size x classes tensor info. Data types supported: F16/S32/F32/QASYMM8/QASYMM8_SIGNED
      * @param[in] targets     A batch_size 1D tensor info of class ids. Data types supported: S32
      * @param[in] output      Computed precision at @p k as a bool 1D tensor info. Data types supported: U8
      * @param[in] k           Number of top elements to look at for computing precision.

diff --git a/arm_compute/runtime/CPP/functions/CPPTopKV.h b/arm_compute/runtime/CPP/functions/CPPTopKV.h
index d41b7d300f4615fbfdf5af00d93ab9a4f472b378..c94e277312b6ec651469544dc26f1fcc6988af01 100644 (file)
--- a/arm_compute/runtime/CPP/functions/CPPTopKV.h
+++ b/arm_compute/runtime/CPP/functions/CPPTopKV.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019 ARM Limited.
+ * Copyright (c) 2019-2020 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -38,7 +38,7 @@ class CPPTopKV : public ICPPSimpleFunction
 public:
     /** Set the input and output of the kernel.
      *
-     * @param[in]  predictions A batch_size x classes tensor. Data types supported: F16/S32/F32/QASYMM8
+     * @param[in]  predictions A batch_size x classes tensor. Data types supported: F16/S32/F32/QASYMM8/QASYMM8_SIGNED
      * @param[in]  targets     A batch_size 1D tensor of class ids. Data types supported: S32
      * @param[out] output      Computed precision at @p k as a bool 1D tensor. Data types supported: U8
      * @param[in]  k           Number of top elements to look at for computing precision.
@@ -47,7 +47,7 @@ public:
 
     /** Static function to check if given info will lead to a valid configuration of @ref CPPTopKVKernel
      *
-     * @param[in] predictions A batch_size x classes tensor info. Data types supported: F16/S32/F32/QASYMM8
+     * @param[in] predictions A batch_size x classes tensor info. Data types supported: F16/S32/F32/QASYMM8/QASYMM8_SIGNED
      * @param[in] targets     A batch_size 1D tensor info of class ids. Data types supported: S32
      * @param[in] output      Computed precision at @p k as a bool 1D tensor info. Data types supported: U8
      * @param[in] k           Number of top elements to look at for computing precision.

diff --git a/src/core/CPP/kernels/CPPTopKVKernel.cpp b/src/core/CPP/kernels/CPPTopKVKernel.cpp
index 533543a98856ecbea5900444317e14c663ebca69..7f284d4e1e90fe593b5d0572b8c64aaa4b64c348 100644 (file)
--- a/src/core/CPP/kernels/CPPTopKVKernel.cpp
+++ b/src/core/CPP/kernels/CPPTopKVKernel.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019 ARM Limited.
+ * Copyright (c) 2019-2020 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -54,7 +54,7 @@ inline bool greater_than(T a, T b)
 Status validate_arguments(const ITensorInfo *predictions, const ITensorInfo *targets, ITensorInfo *output, const unsigned int k)
 {
     ARM_COMPUTE_UNUSED(k);
-    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(predictions, 1, DataType::QASYMM8, DataType::S32, DataType::F16, DataType::F32);
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(predictions, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED, DataType::S32, DataType::F16, DataType::F32);
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(targets, 1, DataType::U32);
 
     ARM_COMPUTE_RETURN_ERROR_ON(predictions->num_dimensions() > 2);
@@ -145,6 +145,9 @@ void CPPTopKVKernel::run(const Window &window, const ThreadInfo &info)
         case DataType::QASYMM8:
             run_topkv<uint8_t>();
             break;
+        case DataType::QASYMM8_SIGNED:
+            run_topkv<int8_t>();
+            break;
         default:
             ARM_COMPUTE_ERROR("Not supported");
     }

diff --git a/tests/validation/CPP/TopKV.cpp b/tests/validation/CPP/TopKV.cpp
index 02178192aa377139eed7bcde6016f52e48369bbc..e528c622b0c50c2fd36a138a1c0bd379993fcf95 100644 (file)
--- a/tests/validation/CPP/TopKV.cpp
+++ b/tests/validation/CPP/TopKV.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019 ARM Limited.
+ * Copyright (c) 2019-2020 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -134,7 +134,7 @@ TEST_CASE(Float, framework::DatasetMode::ALL)
     validate(Accessor(output), expected_output);
 }
 
-TEST_CASE(Quantized, framework::DatasetMode::ALL)
+TEST_CASE(QASYMM8, framework::DatasetMode::ALL)
 {
     const unsigned int k = 5;
 
@@ -187,6 +187,59 @@ TEST_CASE(Quantized, framework::DatasetMode::ALL)
     validate(Accessor(output), expected_output);
 }
 
+TEST_CASE(QASYMM8_SIGNED, framework::DatasetMode::ALL)
+{
+    const unsigned int k = 5;
+
+    Tensor predictions = create_tensor<Tensor>(TensorShape(10, 20), DataType::QASYMM8_SIGNED, 1, QuantizationInfo());
+    Tensor targets     = create_tensor<Tensor>(TensorShape(20), DataType::U32);
+
+    predictions.allocator()->allocate();
+    targets.allocator()->allocate();
+
+    // Fill the tensors with random pre-generated values
+    fill_tensor(Accessor(predictions), std::vector<int8_t>
+    {
+        123, -34, 69, 118, 20, -45, 99, -98, 127, 117,  //-34
+        -99, 1, -128, 90, 60, 25, 102, 76, 24, -110,    //25
+        99, 119, 49, 43, -40, 60, 43, 84, 29, 67,       //84
+        33, 109, 74, 90, 90, 44, 98, 90, 35, 123,       //74
+        62, 118, 24, -32, 34, 21, 114, 113, 124, 20,    //124
+        74, 98, 48, 107, 127, 125, 6, -98, 127, 59,     //98
+        75, 83, 75, -118, 56, 101, 85, 97, 49, 127,     //75
+        72, -20, 40, 14, 28, -30, 109, 43, 127, -31,    //-20
+        78, 121, 109, 66, 29, 90, 70, 21, 38, 48,       //109
+        18, 10, 115, 124, 17, 123, 51, 54, 15, 17,      //17
+        66, 46, -66, 125, 104, 90, 123, 113, -54, -126, //125
+        58, -85, 74, 39, 115, 39, 111, -27, 44, 51,     //51
+        71, 122, -34, -123, 94, 113, 125, 111, 38, 64,  //94
+        -17, 40, 42, 68, 96, 68, 101, 40, 79, 71,       //40
+        89, 88, 54, 82, 127, 12, 112, 52, 125, 22,      //22
+        -128, 56, 82, 31, 98, 94, 102, 105, 127, 123,   //123
+        112, 50, 61, 41, 39, 63, -77, 92, 26, 70,       //39
+        2, 90, 31, 99, -34, 114, 112, 126, 127, 87,     //90
+        125, 63, 56, 123, 50, -77, 97, -93, 1, 29,      //56
+        100, -35, 116, 64, 36, 92, 56, 82, -22, -118    //36
+    });
+
+    fill_tensor(Accessor(targets), std::vector<int> { 1, 5, 7, 2, 8, 1, 2, 1, 2, 4, 3, 9, 4, 1, 9, 9, 4, 1, 2, 4 });
+
+    // Determine the output through the CPP kernel
+    Tensor   output;
+    CPPTopKV topkv;
+    topkv.configure(&predictions, &targets, &output, k);
+
+    output.allocator()->allocate();
+
+    // Run the kernel
+    topkv.run();
+
+    // Validate against the expected values
+    SimpleTensor<int8_t> expected_output(TensorShape(20), DataType::U8);
+    fill_tensor(expected_output, std::vector<int8_t> { 0, 1, 1, 0, 1, 1, 0, 0, 1, 0, 1, 1, 1, 0, 0, 1, 0, 0, 1, 0 });
+    validate(Accessor(output), expected_output);
+}
+
 TEST_SUITE_END() // TopKV
 TEST_SUITE_END() // CPP
 } // namespace validation