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, 13 insertions, 10 deletions

diff --git a/arm_compute/core/Types.h b/arm_compute/core/Types.h
index cb1a212797..0240916da8 100644
--- a/arm_compute/core/Types.h
+++ b/arm_compute/core/Types.h
@@ -1192,7 +1192,7 @@ class GEMMReshapeInfo final
 public:
     /** Default constructor */
     GEMMReshapeInfo()
-        : _m(1), _n(1), _k(1), _mult_transpose1xW_width(1), _mult_interleave4x4_height(1), _depth_output_gemm3d(1), _reinterpret_input_as_3d(false)
+        : _m(1), _n(1), _k(1), _mult_transpose1xW_width(1), _mult_interleave4x4_height(1), _depth_output_gemm3d(0), _reinterpret_input_as_3d(false)
     {
     }
     /** Constructor
@@ -1202,11 +1202,12 @@ public:
      * @param[in] k                         Number of matrix A columns or matrix B rows
      * @param[in] mult_transpose1xW_width   (Optional) Multiplication factor for the width of the 1xW transposed block
      * @param[in] mult_interleave4x4_height (Optional) Multiplication factor for the height of the 4x4 interleaved block
-     * @param[in] depth_output_gemm3d       (Optional) Depth (third dimension) of the output tensor to be used with the GEMM3D kernel
+     * @param[in] depth_output_gemm3d       (Optional) Depth (third dimension) of the output tensor to be used with the GEMM3D kernel.
+     *                                      If 0 the output will not be reinterpreted as 3D. Default 0
      * @param[in] reinterpret_input_as_3d   (Optional) Reinterpret the input as 3D tensor. (i.e. this flag should be set to true when GEMM is used
      *                                                 to perform 1x1 convolutions with the NHWC data layout)
      */
-    GEMMReshapeInfo(int m, int n, int k, int mult_transpose1xW_width = 1, int mult_interleave4x4_height = 1, int depth_output_gemm3d = 1, bool reinterpret_input_as_3d = false)
+    GEMMReshapeInfo(int m, int n, int k, int mult_transpose1xW_width = 1, int mult_interleave4x4_height = 1, int depth_output_gemm3d = 0, bool reinterpret_input_as_3d = false)
         : _m(m), _n(n), _k(k), _mult_transpose1xW_width(mult_transpose1xW_width), _mult_interleave4x4_height(mult_interleave4x4_height), _depth_output_gemm3d(depth_output_gemm3d),
           _reinterpret_input_as_3d(reinterpret_input_as_3d)
     {
@@ -1311,7 +1312,7 @@ class GEMMInfo
 public:
     /** Default constructor */
     GEMMInfo()
-        : _is_a_reshaped(false), _is_b_reshaped(false), _reshape_b_only_on_first_run(false), _depth_output_gemm3d(1), _reinterpret_input_as_3d(false), _retain_internal_weights(false), _gemmlowp_output_stage()
+        : _is_a_reshaped(false), _is_b_reshaped(false), _reshape_b_only_on_first_run(false), _depth_output_gemm3d(0), _reinterpret_input_as_3d(false), _retain_internal_weights(false), _gemmlowp_output_stage()
     {
     }
     /** Constructor
@@ -1320,13 +1321,14 @@ public:
      * @param[in] is_b_reshaped               True if the matrix B has been reshaped
      * @param[in] reshape_b_only_on_first_run Reshape matrix B only for the first run
      * @param[in] depth_output_gemm3d         (Optional) Depth (third dimension) of the output tensor to be used with the GEMM3D kernel
+     *                                        If 0 the output will not be reinterpreted as 3D. Default 0
      * @param[in] reinterpret_input_as_3d     (Optional) Reinterpret the input as 3D tensor. (i.e. this flag should be set to true when GEMM is used
      *                                        to perform 1x1 convolutions with the NHWC data layout)
      * @param[in] retain_internal_weights     (Optional) Retain the weights tensor from previous run
      * @param[in] gemmlowp_output_stage       (Optional) GEMMLowp Output stage info
      *
      */
-    GEMMInfo(bool is_a_reshaped, bool is_b_reshaped, bool reshape_b_only_on_first_run, int depth_output_gemm3d = 1, bool reinterpret_input_as_3d = false, bool retain_internal_weights = false,
+    GEMMInfo(bool is_a_reshaped, bool is_b_reshaped, bool reshape_b_only_on_first_run, int depth_output_gemm3d = 0, bool reinterpret_input_as_3d = false, bool retain_internal_weights = false,
              GEMMLowpOutputStageInfo gemmlowp_output_stage = GEMMLowpOutputStageInfo())
         : _is_a_reshaped(is_a_reshaped), _is_b_reshaped(is_b_reshaped), _reshape_b_only_on_first_run(reshape_b_only_on_first_run), _depth_output_gemm3d(depth_output_gemm3d),
           _reinterpret_input_as_3d(reinterpret_input_as_3d), _retain_internal_weights(retain_internal_weights), _gemmlowp_output_stage(gemmlowp_output_stage)
diff --git a/arm_compute/core/utils/misc/ShapeCalculator.h b/arm_compute/core/utils/misc/ShapeCalculator.h
index 56f65d0ba8..1f532ca31e 100644
--- a/arm_compute/core/utils/misc/ShapeCalculator.h
+++ b/arm_compute/core/utils/misc/ShapeCalculator.h
@@ -532,13 +532,14 @@ inline TensorShape compute_mm_shape(const ITensorInfo &input0, const ITensorInfo
     ARM_COMPUTE_ERROR_ON_MSG(is_interleaved_transposed && reshape_info.reinterpret_input_as_3d(), "The first input tensor cannot be reinterpreted as 3D if is_interleaved_transposed is true");
 
     const bool reinterpret_input_as_3d  = reshape_info.reinterpret_input_as_3d();
-    const bool reinterpret_output_as_3d = reshape_info.depth_output_gemm3d() != 1;
+    const bool reinterpret_output_as_3d = reshape_info.depth_output_gemm3d() != 0;
+    const int  depth_output_gemm3d      = reinterpret_output_as_3d ? reshape_info.depth_output_gemm3d() : 1;
     const int  m                        = reshape_info.reinterpret_input_as_3d() ? input0.dimension(1) * input0.dimension(2) : input0.dimension(1);
 
     // If the output of GEMM has to be reinterpreted as 3D, the number of input0 rows (M) is obtained collapsing the second and third
     // dimension of the output tensor
     const int dim0 = is_interleaved_transposed ? reshape_info.n() : input1.dimension(0);
-    const int dim1 = is_interleaved_transposed ? reshape_info.m() / reshape_info.depth_output_gemm3d() : m / reshape_info.depth_output_gemm3d();
+    const int dim1 = is_interleaved_transposed ? reshape_info.m() / depth_output_gemm3d : m / depth_output_gemm3d;
     const int dim2 = reinterpret_input_as_3d ? input0.tensor_shape()[3] : input0.tensor_shape()[2];
     const int dim3 = reinterpret_input_as_3d ? 1 : input0.tensor_shape()[3];
 
@@ -546,7 +547,7 @@ inline TensorShape compute_mm_shape(const ITensorInfo &input0, const ITensorInfo
 
     output_shape.set(0, dim0);
     output_shape.set(1, dim1);
-    output_shape.set(2, reinterpret_output_as_3d ? reshape_info.depth_output_gemm3d() : dim2);
+    output_shape.set(2, reinterpret_output_as_3d ? depth_output_gemm3d : dim2);
     output_shape.set(3, reinterpret_output_as_3d ? dim2 : dim3);
     output_shape.set(4, reinterpret_output_as_3d ? dim3 : 1);
 
diff --git a/arm_compute/core/utils/quantization/AsymmHelpers.h b/arm_compute/core/utils/quantization/AsymmHelpers.h
index 6fd1d80010..d9f20cee2b 100644
--- a/arm_compute/core/utils/quantization/AsymmHelpers.h
+++ b/arm_compute/core/utils/quantization/AsymmHelpers.h
@@ -38,7 +38,7 @@ namespace quantization
  *
  * @return a status
  */
-arm_compute::Status calculate_quantized_multiplier_less_than_one(double multiplier, int *quant_multiplier, int *right_shift);
+arm_compute::Status calculate_quantized_multiplier_less_than_one(float multiplier, int *quant_multiplier, int *right_shift);
 /** Calculate quantized representation of multiplier having value greater than one.
  *
  * @param[in]  multiplier           Real multiplier.
@@ -47,7 +47,7 @@ arm_compute::Status calculate_quantized_multiplier_less_than_one(double multipli
  *
  * @return a status
  */
-arm_compute::Status calculate_quantized_multiplier_greater_than_one(double multiplier, int *quantized_multiplier, int *left_shift);
+arm_compute::Status calculate_quantized_multiplier_greater_than_one(float multiplier, int *quantized_multiplier, int *left_shift);
 } // namespace quantization
 } // namespace arm_compute
 #endif /* __ARM_COMPUTE_IO_FILE_HANDLER_H__ */