COMPMID-1736: Fixed out-of-bound write in CLIm2Col

The issue was related to CLIm2Col when the number of input channels was less than
the number of elements processed by each thread.
The bug has been fixed in the validate_and_configure_window() function setting the correct number of elements accessed
in the output tensor.

Also fixed an issue GEMM3D when we have a single output channel

Change-Id: I094292d0c7662599c4a4c3916ec5f5821df5faef

3 files changed, 9 insertions, 7 deletions

diff --git a/src/runtime/NEON/functions/NEGEMM.cpp b/src/runtime/NEON/functions/NEGEMM.cpp
index 321ecf85d8..e8bf6732b2 100644
--- a/src/runtime/NEON/functions/NEGEMM.cpp
+++ b/src/runtime/NEON/functions/NEGEMM.cpp
@@ -140,7 +140,7 @@ Status NEGEMM::validate(const ITensorInfo *a, const ITensorInfo *b, const ITenso
 
     if(c != nullptr)
     {
-        ARM_COMPUTE_RETURN_ERROR_ON(gemm_info.depth_output_gemm3d() != 1);
+        ARM_COMPUTE_RETURN_ERROR_ON(gemm_info.depth_output_gemm3d() != 0);
         ARM_COMPUTE_RETURN_ERROR_ON(gemm_info.reinterpret_input_as_3d());
         ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(a, c);
         ARM_COMPUTE_RETURN_ERROR_ON_MSG(a->dimension(1) != c->dimension(1), "The C matrix must have the same number of rows as the matrix A");
@@ -150,7 +150,7 @@ Status NEGEMM::validate(const ITensorInfo *a, const ITensorInfo *b, const ITenso
     if(output->total_size() != 0)
     {
         ARM_COMPUTE_RETURN_ERROR_ON(b->dimension(0) != output->dimension(0));
-        if(gemm_info.depth_output_gemm3d() != 1)
+        if(gemm_info.depth_output_gemm3d() != 0)
         {
             if(gemm_info.reinterpret_input_as_3d())
             {
@@ -174,7 +174,7 @@ Status NEGEMM::validate(const ITensorInfo *a, const ITensorInfo *b, const ITenso
     if(!run_optimised)
     {
         ARM_COMPUTE_RETURN_ERROR_ON_MSG(gemm_info.reinterpret_input_as_3d(), "NEGEMM cannot reinterpret the input tensor as 3D");
-        ARM_COMPUTE_RETURN_ERROR_ON_MSG(gemm_info.depth_output_gemm3d() != 1, "NEGEMM cannot reinterpret the output tensor as 3D");
+        ARM_COMPUTE_RETURN_ERROR_ON_MSG(gemm_info.depth_output_gemm3d() != 0, "NEGEMM cannot reinterpret the output tensor as 3D");
 
         // Check if the first input tensor is a vector.
         const bool run_vector_matrix_multiplication = a->dimension(1) < 2;
diff --git a/src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp b/src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp
index fc65469488..02ae1715da 100644
--- a/src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp
+++ b/src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp
@@ -276,7 +276,9 @@ void NEGEMMConvolutionLayer::configure(const ITensor *input, const ITensor *weig
     }
 
     // Configure GEMM
-    configure_mm(gemm_input_to_use, &_weights_reshaped, gemm_output_to_use, _skip_col2im ? conv_h : 1);
+    // In case we need to skip col2im, GEMM3D (gemm_3d_depth != 0) must be called in order to avoid reshaping the output matrix
+    const unsigned int gemm_3d_depth = _skip_col2im ? conv_h : 0;
+    configure_mm(gemm_input_to_use, &_weights_reshaped, gemm_output_to_use, gemm_3d_depth);
 
     if(!_skip_im2col)
     {
@@ -477,7 +479,7 @@ Status NEGEMMConvolutionLayer::validate(const ITensorInfo *input, const ITensorI
         gemm_output_to_use = &info_gemm;
     }
 
-    ARM_COMPUTE_RETURN_ON_ERROR(validate_mm(gemm_input_to_use, weights_to_use, gemm_output_to_use, skip_col2im ? conv_h : 1, skip_im2col));
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_mm(gemm_input_to_use, weights_to_use, gemm_output_to_use, skip_col2im ? conv_h : 0, skip_im2col));
 
     if(is_quantized)
     {
@@ -518,7 +520,7 @@ Status NEGEMMConvolutionLayer::validate(const ITensorInfo *input, const ITensorI
         }
 
         // Validate output stage for quantized case
-        NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint::validate(gemm_output_to_use, biases, gemm_output_staged_to_use, min, max, skip_reshape ? conv_h : 1);
+        NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint::validate(gemm_output_to_use, biases, gemm_output_staged_to_use, min, max, skip_reshape ? conv_h : 0);
     }
 
     // Validate Col2Im/ReshapeLayer
diff --git a/src/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.cpp b/src/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.cpp
index 80f5ab0c93..16ee3d07fd 100644
--- a/src/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.cpp
+++ b/src/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.cpp
@@ -198,7 +198,7 @@ Status NEGEMMLowpMatrixMultiplyCore::validate(const ITensorInfo *a, const ITenso
     ARM_COMPUTE_RETURN_ERROR_ON_MSG(gemm_info.is_a_reshaped(), "Matrix A already reshaped is not supported");
     ARM_COMPUTE_RETURN_ERROR_ON_MSG(gemm_info.is_b_reshaped(), "Matrix B already reshaped is not supported");
     ARM_COMPUTE_RETURN_ERROR_ON_MSG(gemm_info.reinterpret_input_as_3d(), "NEGEMMLowpMatrixMultiplyCore cannot reinterpret the input tensor as 3D");
-    ARM_COMPUTE_RETURN_ERROR_ON_MSG(gemm_info.depth_output_gemm3d() != 1, "NEGEMMLowpMatrixMultiplyCore cannot reinterpret the output tensor as 3D");
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG(gemm_info.depth_output_gemm3d() != 0, "NEGEMMLowpMatrixMultiplyCore cannot reinterpret the output tensor as 3D");
 
     int32_t a_offset                         = a->quantization_info().offset;
     int32_t b_offset                         = b->quantization_info().offset;