Enable quantized data types for CpuElementwiseUnary on Armv7a

- Adding fallback functions neon_qasymm8_signed_elementwise_unary() and
  neon_qasymm8_elementwise_unary()
- They would be called in case target is not aarch64

Resolves: COMPMID-5994
Change-Id: Id0db1e7cb0fe92f1eaef0b3a9ed2bea01b3f2a15
Signed-off-by: Ramy Elgammal <ramy.elgammal@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9416
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Viet-Hoa Do <viet-hoa.do@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>

7 files changed, 34 insertions, 7 deletions

diff --git a/tests/validation/NEON/ElementwiseAbsoluteValue.cpp b/tests/validation/NEON/ElementwiseAbsoluteValue.cpp
index 7f6a6a5bb2..0667ac73f9 100644
--- a/tests/validation/NEON/ElementwiseAbsoluteValue.cpp
+++ b/tests/validation/NEON/ElementwiseAbsoluteValue.cpp
@@ -46,8 +46,13 @@ RelativeTolerance<float> tolerance_fp32(0.000001f);
 #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
 RelativeTolerance<float> tolerance_fp16(0.01f);
 #endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+#if defined(__aarch64__)
 constexpr AbsoluteTolerance<uint8_t> tolerance_qasymm8(0);
 constexpr AbsoluteTolerance<int8_t>  tolerance_qasymm8_signed(0);
+#else  // #if !defined(__aarch64__)
+constexpr AbsoluteTolerance<uint8_t> tolerance_qasymm8(1); // There is difference of 1, because quantizing in reference uses round policy "TO_NEAREST_UP", where the armv7a neon kernel uses "TO_ZERO"
+constexpr AbsoluteTolerance<int8_t>  tolerance_qasymm8_signed(1);
+#endif // #if !defined(__aarch64__)
 } // namespace
 
 TEST_SUITE(NEON)
@@ -136,8 +141,8 @@ FIXTURE_DATA_TEST_CASE(RunSmall, NEAbsLayerQuantizedFixture<int8_t>, framework::
     validate(Accessor(_target), _reference, tolerance_qasymm8_signed);
 }
 TEST_SUITE_END() // QASYMM8_SIGNED
-
 TEST_SUITE_END() // Quantized
+
 TEST_SUITE_END() // AbsLayer
 TEST_SUITE_END() // Neon
 } // namespace validation
diff --git a/tests/validation/NEON/ElementwiseExpLayer.cpp b/tests/validation/NEON/ElementwiseExpLayer.cpp
index e8940c5385..31cd78626f 100644
--- a/tests/validation/NEON/ElementwiseExpLayer.cpp
+++ b/tests/validation/NEON/ElementwiseExpLayer.cpp
@@ -46,8 +46,15 @@ RelativeTolerance<float> tolerance_fp32(0.000001f);
 #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
 RelativeTolerance<float> tolerance_fp16(0.01f);
 #endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+
+#if defined(__aarch64__)
 constexpr AbsoluteTolerance<uint8_t> tolerance_qasymm8(0);
 constexpr AbsoluteTolerance<int8_t>  tolerance_qasymm8_signed(0);
+#else  // #if !defined(__aarch64__)
+constexpr AbsoluteTolerance<uint8_t> tolerance_qasymm8(1); // There is difference of 1, because quantizing in reference uses round policy "TO_NEAREST_UP", where the armv7a neon kernel uses "TO_ZERO"
+constexpr AbsoluteTolerance<int8_t>  tolerance_qasymm8_signed(1);
+#endif // #if !defined(__aarch64__)
+
 } // namespace
 TEST_SUITE(NEON)
 TEST_SUITE(ExpLayer)
@@ -111,7 +118,6 @@ FIXTURE_DATA_TEST_CASE(RunSmall, NEExpLayerQuantizedFixture<int8_t>, framework::
     validate(Accessor(_target), _reference, tolerance_qasymm8_signed);
 }
 TEST_SUITE_END() // QASYMM8_SIGNED
-
 TEST_SUITE_END() // Quantized
 
 TEST_SUITE_END() // ExpLayer
diff --git a/tests/validation/NEON/ElementwiseLog.cpp b/tests/validation/NEON/ElementwiseLog.cpp
index 49a88ced1c..320ece2e73 100644
--- a/tests/validation/NEON/ElementwiseLog.cpp
+++ b/tests/validation/NEON/ElementwiseLog.cpp
@@ -46,8 +46,15 @@ RelativeTolerance<float> tolerance_fp32(0.000001f);
 #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
 RelativeTolerance<float> tolerance_fp16(0.01f);
 #endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+
+#if defined(__aarch64__)
 constexpr AbsoluteTolerance<uint8_t> tolerance_qasymm8(0);
 constexpr AbsoluteTolerance<int8_t>  tolerance_qasymm8_signed(0);
+#else  // #if !defined(__aarch64__)
+constexpr AbsoluteTolerance<uint8_t> tolerance_qasymm8(1); // There is difference of 1, because quantizing in reference uses round policy "TO_NEAREST_UP", where the armv7a neon kernel uses "TO_ZERO"
+constexpr AbsoluteTolerance<int8_t>  tolerance_qasymm8_signed(1);
+#endif // #if !defined(__aarch64__)
+
 } // namespace
 TEST_SUITE(NEON)
 TEST_SUITE(LogLayer)
@@ -118,8 +125,8 @@ FIXTURE_DATA_TEST_CASE(RunSmall, NELogLayerQuantizedFixture<int8_t>, framework::
     validate(Accessor(_target), _reference, tolerance_qasymm8_signed);
 }
 TEST_SUITE_END() // QASYMM8_SIGNED
-
 TEST_SUITE_END() // Quantized
+
 TEST_SUITE_END() // LogLayer
 TEST_SUITE_END() // Neon
 } // namespace validation
diff --git a/tests/validation/NEON/ElementwiseNegation.cpp b/tests/validation/NEON/ElementwiseNegation.cpp
index 038058c70c..5b8ae8fc64 100644
--- a/tests/validation/NEON/ElementwiseNegation.cpp
+++ b/tests/validation/NEON/ElementwiseNegation.cpp
@@ -46,8 +46,13 @@ RelativeTolerance<float> tolerance_fp32(0.000001f);
 #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
 RelativeTolerance<float> tolerance_fp16(0.01f);
 #endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+#if defined(__aarch64__)
 constexpr AbsoluteTolerance<uint8_t> tolerance_qasymm8(0);
 constexpr AbsoluteTolerance<int8_t>  tolerance_qasymm8_signed(0);
+#else  // #if !defined(__aarch64__)
+constexpr AbsoluteTolerance<uint8_t> tolerance_qasymm8(1); // There is difference of 1, because quantizing in reference uses round policy "TO_NEAREST_UP", where the armv7a neon kernel uses "TO_ZERO"
+constexpr AbsoluteTolerance<int8_t>  tolerance_qasymm8_signed(1);
+#endif // #if !defined(__aarch64__)
 } // namespace
 TEST_SUITE(NEON)
 TEST_SUITE(NegLayer)
@@ -142,8 +147,8 @@ FIXTURE_DATA_TEST_CASE(RunSmall, NENegLayerQuantizedFixture<int8_t>, framework::
     validate(Accessor(_target), _reference, tolerance_qasymm8_signed);
 }
 TEST_SUITE_END() // QASYMM8_SIGNED
-
 TEST_SUITE_END() // Quantized
+
 TEST_SUITE_END() // NegLayer
 TEST_SUITE_END() // Neon
 } // namespace validation
diff --git a/tests/validation/NEON/ElementwiseRound.cpp b/tests/validation/NEON/ElementwiseRound.cpp
index a6ff47c830..620618cb0b 100644
--- a/tests/validation/NEON/ElementwiseRound.cpp
+++ b/tests/validation/NEON/ElementwiseRound.cpp
@@ -114,8 +114,8 @@ FIXTURE_DATA_TEST_CASE(RunSmall, NERoundLayerQuantizedFixture<int8_t>, framework
     validate(Accessor(_target), _reference, tolerance_qasymm8_signed);
 }
 TEST_SUITE_END() // QASYMM8_SIGNED
-
 TEST_SUITE_END() // Quantized
+
 TEST_SUITE_END() // RoundLayer
 TEST_SUITE_END() // Neon
 } // namespace validation
diff --git a/tests/validation/NEON/ElementwiseRsqrtLayer.cpp b/tests/validation/NEON/ElementwiseRsqrtLayer.cpp
index 1d291ac6dc..80788c893f 100644
--- a/tests/validation/NEON/ElementwiseRsqrtLayer.cpp
+++ b/tests/validation/NEON/ElementwiseRsqrtLayer.cpp
@@ -46,8 +46,13 @@ RelativeTolerance<float> tolerance_fp32(0.000001f);
 #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
 RelativeTolerance<float> tolerance_fp16(0.01f);
 #endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+#if defined(__aarch64__)
 constexpr AbsoluteTolerance<uint8_t> tolerance_qasymm8(0);
 constexpr AbsoluteTolerance<int8_t>  tolerance_qasymm8_signed(0);
+#else  // #if !defined(__aarch64__)
+constexpr AbsoluteTolerance<uint8_t> tolerance_qasymm8(1); // There is difference of 1, because quantizing in reference uses round policy "TO_NEAREST_UP", where the armv7a neon kernel uses "TO_ZERO"
+constexpr AbsoluteTolerance<int8_t>  tolerance_qasymm8_signed(1);
+#endif // #if !defined(__aarch64__)
 } // namespace
 TEST_SUITE(NEON)
 TEST_SUITE(RsqrtLayer)
@@ -131,8 +136,8 @@ FIXTURE_DATA_TEST_CASE(RunSmall, NERsqrtLayerQuantizedFixture<int8_t>, framework
     validate(Accessor(_target), _reference, tolerance_qasymm8_signed);
 }
 TEST_SUITE_END() // QASYMM8_SIGNED
-
 TEST_SUITE_END() // Quantized
+
 TEST_SUITE_END() // RsqrtLayer
 TEST_SUITE_END() // Neon
 } // namespace validation
diff --git a/tests/validation/NEON/ElementwiseSin.cpp b/tests/validation/NEON/ElementwiseSin.cpp
index 76f4c50b46..9c2d7ae268 100644
--- a/tests/validation/NEON/ElementwiseSin.cpp
+++ b/tests/validation/NEON/ElementwiseSin.cpp
@@ -120,7 +120,6 @@ FIXTURE_DATA_TEST_CASE(RunSmall, NESinLayerQuantizedFixture<int8_t>, framework::
 TEST_SUITE_END() // QASYMM8_SIGNED
 
 TEST_SUITE_END() // Quantized
-
 TEST_SUITE_END() // SinLayer
 TEST_SUITE_END() // Neon
 } // namespace validation