Optimize Quantized/Integer Bilinear Scale for Neon™

This patch introduces several performance optimizations regarding the Bilinear Scale operator with REPLICATE Border mode. Changes apply only to NHWC.

This patch
   - Reduces the memory footprint by disabling precomputation of indices and weights when they're not used
   - Rewrites the kernels for QASYMM8/QASYMM8_SIGNED/U8(Uint8)
   - Adds S8(Int8) Bilinear Scale for Border mode REPLICATE
   - Removes Bilinear Scale SVE kernels for Quantized and Integer types and adjust the heuristics to choose the Neon™ implementation
   - Adds new test cases where the input and output of the Bilinear Scale operator have different quantization scale and offset

Resolves: COMPMID-5453, COMPMID-5454

Change-Id: I3d251e76e0c6978fd5a0a1795ec62ab536bec93c
Signed-off-by: Gunes Bayir <gunes.bayir@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8250
Reviewed-by: SiCong Li <sicong.li@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>

3 files changed, 41 insertions, 13 deletions

diff --git a/src/core/NEON/wrapper/intrinsics/cvt.h b/src/core/NEON/wrapper/intrinsics/cvt.h
index e52e3dd0c4..baad1319b2 100644
--- a/src/core/NEON/wrapper/intrinsics/cvt.h
+++ b/src/core/NEON/wrapper/intrinsics/cvt.h
@@ -59,19 +59,35 @@ VCVT_TO_F16_IMPL(float16x4_t, float32x4_t, vcvt, f16, f32)
 #endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
 
 template <typename T>
-inline typename std::enable_if<std::is_same<T, uint8_t>::value, uint32x4_t>::type
+inline typename std::enable_if < std::is_same<T, uint8_t>::value || std::is_same<T, uint32_t>::value, uint32x4_t >::type
 vcvt(const float32x4_t &a)
 {
     return vcvtq_u32_f32(a);
 }
 
 template <typename T>
-inline typename std::enable_if<std::is_same<T, int8_t>::value, int32x4_t>::type
+inline typename std::enable_if < std::is_same<T, int8_t>::value || std::is_same<T, int32_t>::value, int32x4_t >::type
 vcvt(const float32x4_t &a)
 {
     return vcvtq_s32_f32(a);
 }
 
+#ifdef __aarch64__
+template <typename T>
+inline typename std::enable_if<std::is_same<T, uint32_t>::value, uint32x4_t>::type
+vcvta(const float32x4_t &a)
+{
+    return vcvtaq_u32_f32(a);
+}
+
+template <typename T>
+inline typename std::enable_if<std::is_same<T, int32_t>::value, int32x4_t>::type
+vcvta(const float32x4_t &a)
+{
+    return vcvtaq_s32_f32(a);
+}
+#endif //__aarch64__
+
 #if defined(ARM_COMPUTE_ENABLE_BF16)
 /** Convert 2x128-bit floating point vectors into 1x128-bit bfloat16 vector
  *
diff --git a/src/core/utils/ScaleUtils.cpp b/src/core/utils/ScaleUtils.cpp
index 82c6405e89..ee57a8e7a7 100644
--- a/src/core/utils/ScaleUtils.cpp
+++ b/src/core/utils/ScaleUtils.cpp
@@ -40,12 +40,26 @@ float arm_compute::scale_utils::calculate_resize_ratio(size_t input_size, size_t
     return static_cast<float>(in) / static_cast<float>(out);
 }
 
-bool arm_compute::scale_utils::is_precomputation_required(DataLayout data_layout, DataType data_type, InterpolationPolicy policy)
+bool arm_compute::scale_utils::is_precomputation_required(DataLayout data_layout, DataType data_type,
+                                                          InterpolationPolicy policy, BorderMode border_mode)
 {
-    // whether to precompute indices & weights
-    // The Neon™ kernels (which are preferred over SVE when policy is BILINEAR) do not use
-    // precomputed index and weights when data type is FP32/16.
-    // If policy is nearest_neighbor for SVE, then precompute because it's being used
-    // To be revised in COMPMID-5453/5454
-    return data_layout != DataLayout::NHWC || (data_type != DataType::F32 && data_type != DataType::F16) || (CPUInfo::get().get_isa().sve == true && policy == InterpolationPolicy::NEAREST_NEIGHBOR);
+    // Do not calculate precomputed weights and indices if kernel code doesn't use them
+    if(data_layout == DataLayout::NHWC)
+    {
+        switch(data_type)
+        {
+            case DataType::F32:
+            case DataType::F16:
+                return (CPUInfo::get().get_isa().sve == true && policy == InterpolationPolicy::NEAREST_NEIGHBOR);
+            case DataType::U8:
+            case DataType::S8:
+            case DataType::QASYMM8:
+            case DataType::QASYMM8_SIGNED:
+                return (border_mode != BorderMode::REPLICATE) || (policy == InterpolationPolicy::NEAREST_NEIGHBOR);
+            default:
+                return true;
+        }
+    }
+
+    return true;
 }
\ No newline at end of file
diff --git a/src/core/utils/ScaleUtils.h b/src/core/utils/ScaleUtils.h
index c09509253c..1484824a7f 100644
--- a/src/core/utils/ScaleUtils.h
+++ b/src/core/utils/ScaleUtils.h
@@ -26,9 +26,6 @@
 
 #include "arm_compute/core/Types.h"
 
-#include <cstdint>
-#include <cstdlib>
-
 namespace arm_compute
 {
 namespace scale_utils
@@ -59,10 +56,11 @@ inline bool is_align_corners_allowed_sampling_policy(SamplingPolicy sampling_pol
  * @param[in] data_layout Data layout
  * @param[in] data_type   Data type
  * @param[in] policy      Interpolation policy
+ * @param[in] border_mode Border Mode
  *
  * @return True if precomputation is required
  */
-bool is_precomputation_required(DataLayout data_layout, DataType data_type, InterpolationPolicy policy);
+bool is_precomputation_required(DataLayout data_layout, DataType data_type, InterpolationPolicy policy, BorderMode border_mode);
 
 } // namespace scale_utils
 } // namespace arm_compute