COMPMID-1413 - Improve the performance of GEMMLowp with 8 bit dot product on OpenCL

COMPMID-1424 - Add dot product support for CLDepthwise QASYMM8 3x3 NHWC non-unit stride

With this patch we are able to improve the performance of MobileNet v1-qasymm8 by 37 %
Tried to use the dot product instruction in CLDepthwise QASYMM8 3x3 NHWC non-unit stride
but I have not seen any benefit (maybe because we have few arithemtic operation and we
do not have more load instructions). However Depthwise convolution has been improved by
30%

Change-Id: Id768a99c2e53a04276707e427af5d0ec93419ada
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/155082
Tested-by: bsgcomp <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>

3 files changed, 9 insertions, 6 deletions

diff --git a/src/runtime/NEON/functions/NEFullyConnectedLayer.cpp b/src/runtime/NEON/functions/NEFullyConnectedLayer.cpp
index 60f6294394..45e21b53d1 100644
--- a/src/runtime/NEON/functions/NEFullyConnectedLayer.cpp
+++ b/src/runtime/NEON/functions/NEFullyConnectedLayer.cpp
@@ -50,6 +50,7 @@ Status validate_mm(const ITensorInfo &input, const ITensorInfo &weights, const I
         // Validate gemmlowp function
         ARM_COMPUTE_RETURN_ON_ERROR(NEGEMMLowpMatrixMultiplyCore::validate(&input.clone()->set_quantization_info(input_quantization_info),
                                                                            &weights.clone()->set_quantization_info(weights_quantization_info),
+                                                                           nullptr,
                                                                            &output));
     }
     else
@@ -93,7 +94,7 @@ void NEFullyConnectedLayer::configure_mm(const ITensor *input, const ITensor *we
         weights->info()->set_quantization_info(QuantizationInfo(weights_quantization_info.scale, -weights_quantization_info.offset));
 
         // Configure gemmlowp function
-        _mm_gemmlowp.configure(input, weights, output);
+        _mm_gemmlowp.configure(input, weights, nullptr, output);
 
         // Revert back QuantizatioInfo as input and weights could be used in other fully connected layers
         input->info()->set_quantization_info(input_quantization_info);
diff --git a/src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp b/src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp
index fb6d4a1847..fc65469488 100644
--- a/src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp
+++ b/src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp
@@ -111,7 +111,7 @@ void NEGEMMConvolutionLayer::configure_mm(const ITensor *input, const ITensor *w
         input->info()->set_quantization_info(QuantizationInfo(input_quantization_info.scale, -input_quantization_info.offset));
         weights->info()->set_quantization_info(QuantizationInfo(weights_quantization_info.scale, -weights_quantization_info.offset));
 
-        _mm_gemmlowp.configure(input, weights, output, GEMMInfo(false, false, true /* Reshape weights only for the first run*/));
+        _mm_gemmlowp.configure(input, weights, nullptr, output, GEMMInfo(false, false, true /* Reshape weights only for the first run*/));
 
         // Revert back QuantizatioInfo as input and weights could be used in other convolution layers
         input->info()->set_quantization_info(input_quantization_info);
@@ -143,7 +143,7 @@ Status NEGEMMConvolutionLayer::validate_mm(const ITensorInfo *input, const ITens
         weights_qa->set_quantization_info(QuantizationInfo(weights_quantization_info.scale, -weights_quantization_info.offset));
 
         // Perform validation step on GEMMLowp
-        return NEGEMMLowpMatrixMultiplyCore::validate(input_qa.get(), weights_qa.get(), output, gemm_info);
+        return NEGEMMLowpMatrixMultiplyCore::validate(input_qa.get(), weights_qa.get(), nullptr, output, gemm_info);
     }
     else
     {
diff --git a/src/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.cpp b/src/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.cpp
index 828011d019..80f5ab0c93 100644
--- a/src/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.cpp
+++ b/src/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.cpp
@@ -47,10 +47,11 @@ NEGEMMLowpMatrixMultiplyCore::NEGEMMLowpMatrixMultiplyCore(std::shared_ptr<IMemo
 {
 }
 
-void NEGEMMLowpMatrixMultiplyCore::configure(const ITensor *a, const ITensor *b, ITensor *output, const GEMMInfo &gemm_info)
+void NEGEMMLowpMatrixMultiplyCore::configure(const ITensor *a, const ITensor *b, const ITensor *c, ITensor *output, const GEMMInfo &gemm_info)
 {
     ARM_COMPUTE_ERROR_ON_NULLPTR(a, b, output);
-    ARM_COMPUTE_ERROR_THROW_ON(NEGEMMLowpMatrixMultiplyCore::validate(a->info(), b->info(), output->info(), gemm_info));
+    ARM_COMPUTE_UNUSED(c);
+    ARM_COMPUTE_ERROR_THROW_ON(NEGEMMLowpMatrixMultiplyCore::validate(a->info(), b->info(), c != nullptr ? c->info() : nullptr, output->info(), gemm_info));
 
     // Clear state
     _mtx_a_reshape_kernel = nullptr;
@@ -181,11 +182,12 @@ void NEGEMMLowpMatrixMultiplyCore::configure(const ITensor *a, const ITensor *b,
     }
 }
 
-Status NEGEMMLowpMatrixMultiplyCore::validate(const ITensorInfo *a, const ITensorInfo *b, const ITensorInfo *output, const GEMMInfo &gemm_info)
+Status NEGEMMLowpMatrixMultiplyCore::validate(const ITensorInfo *a, const ITensorInfo *b, const ITensorInfo *c, const ITensorInfo *output, const GEMMInfo &gemm_info)
 {
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(a, 1, DataType::QASYMM8);
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::S32);
     ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(a, b);
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG(c != nullptr, "Bias addition not supported in NEGEMMLowpMatrixMultiplyCore");
     ARM_COMPUTE_RETURN_ERROR_ON_MSG((a)->dimension(0) != (b)->dimension(1),
                                     "The product AB is defined only if the number of columns in A is equal to the number of rows in B");
     ARM_COMPUTE_RETURN_ERROR_ON_MSG((a)->dimension(1) != (output)->dimension(1),