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
diff --git a/src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp b/src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp
index fc65469..02ae171 100644
--- a/src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp
+++ b/src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp
@@ -276,7 +276,9 @@
}
// 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 @@
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 @@
}
// 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