diff --git a/src/core/CL/cl_kernels/depthwise_convolution.cl b/src/core/CL/cl_kernels/depthwise_convolution.cl
index 1c032b1..da22faa 100644
--- a/src/core/CL/cl_kernels/depthwise_convolution.cl
+++ b/src/core/CL/cl_kernels/depthwise_convolution.cl
@@ -1474,6 +1474,19 @@
 
 #define VEC_FLOAT VEC_DATA_TYPE(DATA_TYPE, VEC_SIZE)
 
+#define FILL_ZERO_OUT_OF_BOUND_3(data_type, vec_size, basename, cond)                                                                 \
+    ({                                                                                                                                \
+        basename##0 = select(basename##0, (VEC_DATA_TYPE(data_type, vec_size))0, (SELECT_DATA_TYPE(data_type, vec_size))((cond).s0)); \
+        basename##1 = select(basename##1, (VEC_DATA_TYPE(data_type, vec_size))0, (SELECT_DATA_TYPE(data_type, vec_size))((cond).s1)); \
+        basename##2 = select(basename##2, (VEC_DATA_TYPE(data_type, vec_size))0, (SELECT_DATA_TYPE(data_type, vec_size))((cond).s2)); \
+    })
+
+#define FILL_ZERO_OUT_OF_BOUND_4(data_type, vec_size, basename, cond)                                                                 \
+    ({                                                                                                                                \
+        FILL_ZERO_OUT_OF_BOUND_3(data_type, vec_size, basename, cond);                                                                \
+        basename##3 = select(basename##3, (VEC_DATA_TYPE(data_type, vec_size))0, (SELECT_DATA_TYPE(data_type, vec_size))((cond).s3)); \
+    })
+
 #if defined(CONV_STRIDE_X) && defined(CONV_STRIDE_Y)
 
 /** This function computes the depthwise convolution for NHWC data layout when the stride along the width or height is not 1.
@@ -1485,9 +1498,13 @@
  * @note The convolution pad top must be passed at compile time using -DCONV_PAD_LEFT (e.g. -DCONV_PAD_LEFT=1)
  * @note The convolution stride along the width must be passed at compile time using -DCONV_STRIDE_X (e.g. -DCONV_STRIDE_Y=X)
  * @note The convolution stride along the height must be passed at compile time using -DCONV_STRIDE_Y (e.g. -DCONV_STRIDE_Y=1)
+ * @note The dilation_x and dilation_y must be passed at compile time using -DDILATION_X and -DDILATION_Y: e.g. -DDILATION_X=1, -DDILATION_Y=1
  * @note It is possible to select the activation function to apply using -DACTIVATION_TYPE e.g. -DACTIVATION_TYPE=relu
  * @note A, B variables required by some activation functions are set using -DA_VAL= and -DB_VAL= respectively
  * @note Vector size should be given as a preprocessor argument using -DVEC_SIZE=size
+ * @note The size of the partial store block in x must be passed at compile time using -DPARTIAL_STORE_N0 (e.g. -DPARTIAL_STORE_N0=1)
+ * @note In case of biases, -DHAS_BIAS must to be passed at compile
+ * @note If the output tensor has more than three dimensions, its third dimension must be passed at compile time using -DDST_DEPTH (e.g. -DDST_DEPTH=32)
  *
  * @param[in] src_ptr                               Pointer to the source tensor. Supported data types: F16/F32
  * @param[in] src_stride_x                          Stride of the source tensor in X dimension (in bytes)
@@ -1526,14 +1543,15 @@
 __kernel void depthwise_convolution_3x3_nhwc(
     TENSOR4D_DECLARATION(src),
     TENSOR4D_DECLARATION(dst),
-    TENSOR3D_DECLARATION(weights),
+    TENSOR3D_DECLARATION(weights)
 #if defined(HAS_BIAS)
-    VECTOR_DECLARATION(biases),
+    ,
+    VECTOR_DECLARATION(biases)
 #endif /* defined(HAS_BIAS) */
-    int max_offset)
+)
 {
-    int x = get_global_id(0); // channels
-    int y = get_global_id(1); // spatial coordinate x
+    int x_offset = max((int)(get_global_id(0) * VEC_SIZE - (VEC_SIZE - PARTIAL_STORE_N0) % VEC_SIZE), 0) * sizeof(DATA_TYPE);
+    int y        = get_global_id(1); // spatial coordinate x
 #if defined(DST_DEPTH)
     int z = get_global_id(2) % (int)DST_DEPTH; // spatial coordinate y
     int b = get_global_id(2) / (int)DST_DEPTH; // batch
@@ -1541,90 +1559,89 @@
     int      z               = get_global_id(2); // spatial coordinate y
 #endif                                         // defined(DST_DEPTH)
 
-    Vector weights = CONVERT_TO_VECTOR_STRUCT(weights);
+    __global uchar *weights_addr = weights_ptr + weights_offset_first_element_in_bytes + x_offset;
 
 #if defined(DST_DEPTH)
-    __global uchar *src_addr = src_ptr + src_offset_first_element_in_bytes + x * sizeof(DATA_TYPE) * VEC_SIZE + b * src_stride_w;
+    __global uchar *src_addr = src_ptr + src_offset_first_element_in_bytes + x_offset + b * src_stride_w;
 #else  /* defined(DST_DEPTH) */
-    __global uchar *src_addr = src_ptr + src_offset_first_element_in_bytes + x * sizeof(DATA_TYPE) * VEC_SIZE;
+    __global uchar *src_addr = src_ptr + src_offset_first_element_in_bytes + x_offset;
 #endif /* defined(DST_DEPTH) */
 
-    int  z_coord  = 0;
-    int4 offset   = 0;
-    int4 y_offset = ((int4)(y * CONV_STRIDE_X) + (int4)(0, DILATION_X * 1, DILATION_X * 2, DILATION_X * 3) - CONV_PAD_LEFT) * (int4)src_stride_y;
+    int3 src_coord_y = (int3)(y * CONV_STRIDE_X - CONV_PAD_LEFT) + (int3)(0, DILATION_X, 2 * DILATION_X);
+    int3 src_coord_z = (int3)(z * CONV_STRIDE_Y - CONV_PAD_TOP) + (int3)(0, DILATION_Y, 2 * DILATION_Y);
 
-    // We compute 2x1x1 [C,W,H] elements
-    VEC_FLOAT acc = 0;
+    int3 src_offset_y = clamp(src_coord_y, (int3)0, (int3)(SRC_DIM_1 - 1));
+    int3 src_offset_z = clamp(src_coord_z, (int3)0, (int3)(SRC_DIM_2 - 1));
+
+    // Use these vectors to check whether the unclamped load would have been out of bounds
+    src_coord_y = (src_offset_y != src_coord_y);
+    src_coord_z = (src_offset_z != src_coord_z);
+
+    src_offset_y *= (int3)src_stride_y;
+    src_offset_z *= (int3)src_stride_z;
+
+    // We compute VEC_SIZEx1x1 [C,W,H] elements
+    VEC_FLOAT acc0 = 0;
 
     // Load weights
-    VEC_FLOAT w0 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights.ptr + 0 * weights_stride_y + 0 * weights_stride_z));
-    VEC_FLOAT w1 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights.ptr + 1 * weights_stride_y + 0 * weights_stride_z));
-    VEC_FLOAT w2 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights.ptr + 2 * weights_stride_y + 0 * weights_stride_z));
-    VEC_FLOAT w3 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights.ptr + 0 * weights_stride_y + 1 * weights_stride_z));
-    VEC_FLOAT w4 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights.ptr + 1 * weights_stride_y + 1 * weights_stride_z));
-    VEC_FLOAT w5 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights.ptr + 2 * weights_stride_y + 1 * weights_stride_z));
-    VEC_FLOAT w6 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights.ptr + 0 * weights_stride_y + 2 * weights_stride_z));
-    VEC_FLOAT w7 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights.ptr + 1 * weights_stride_y + 2 * weights_stride_z));
-    VEC_FLOAT w8 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights.ptr + 2 * weights_stride_y + 2 * weights_stride_z));
+    VEC_FLOAT w0 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights_addr + 0 * weights_stride_y + 0 * weights_stride_z));
+    VEC_FLOAT w1 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights_addr + 1 * weights_stride_y + 0 * weights_stride_z));
+    VEC_FLOAT w2 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights_addr + 2 * weights_stride_y + 0 * weights_stride_z));
+    VEC_FLOAT w3 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights_addr + 0 * weights_stride_y + 1 * weights_stride_z));
+    VEC_FLOAT w4 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights_addr + 1 * weights_stride_y + 1 * weights_stride_z));
+    VEC_FLOAT w5 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights_addr + 2 * weights_stride_y + 1 * weights_stride_z));
+    VEC_FLOAT w6 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights_addr + 0 * weights_stride_y + 2 * weights_stride_z));
+    VEC_FLOAT w7 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights_addr + 1 * weights_stride_y + 2 * weights_stride_z));
+    VEC_FLOAT w8 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights_addr + 2 * weights_stride_y + 2 * weights_stride_z));
 
     // Load input values
     // z == 0
-    // Clamp z_coord as for z = 0, it can be negative
-    // z_coord is casted to unsigned int in order to use just a min() operation
-    // A "-1" 32 bit signed variable converted to unsigned gives 4294967295
-    z_coord = z * CONV_STRIDE_Y - (int)CONV_PAD_TOP;
-    z_coord = min((uint)z_coord, (uint)SRC_DIM_2);
-    offset  = y_offset + (int4)(z_coord * src_stride_z);
-    offset  = min(offset, (int4)max_offset);
+    VEC_FLOAT values0 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + src_offset_z.s0 + src_offset_y.s0));
+    VEC_FLOAT values1 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + src_offset_z.s0 + src_offset_y.s1));
+    VEC_FLOAT values2 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + src_offset_z.s0 + src_offset_y.s2));
 
-    VEC_FLOAT values0 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + offset.s0));
-    VEC_FLOAT values1 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + offset.s1));
-    VEC_FLOAT values2 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + offset.s2));
+    FILL_ZERO_OUT_OF_BOUND_3(DATA_TYPE, VEC_SIZE, values, src_coord_y | (int3)src_coord_z.s0);
+
+    acc0 = fma(values0, w0, acc0);
+    acc0 = fma(values1, w1, acc0);
+    acc0 = fma(values2, w2, acc0);
 
     // z == 1
-    // z_coord can be only negative for z = 0 so we do not need to clamp it
-    // Moreover z_coord cannot be out-of-bound for z = 1 so we do not need to clamp the offset
-    z_coord           = z * CONV_STRIDE_Y - (int)CONV_PAD_TOP + DILATION_Y;
-    offset            = y_offset + (int4)(z_coord * src_stride_z);
-    VEC_FLOAT values3 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + offset.s0));
-    VEC_FLOAT values4 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + offset.s1));
-    VEC_FLOAT values5 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + offset.s2));
+    values0 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + src_offset_z.s1 + src_offset_y.s0));
+    values1 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + src_offset_z.s1 + src_offset_y.s1));
+    values2 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + src_offset_z.s1 + src_offset_y.s2));
+
+    FILL_ZERO_OUT_OF_BOUND_3(DATA_TYPE, VEC_SIZE, values, src_coord_y | (int3)src_coord_z.s1);
+
+    acc0 = fma(values0, w3, acc0);
+    acc0 = fma(values1, w4, acc0);
+    acc0 = fma(values2, w5, acc0);
 
     // z == 2
-    // Offset can be out-of-bound so we need to check if it is greater than max_offset
-    z_coord           = z * CONV_STRIDE_Y - (int)CONV_PAD_TOP + DILATION_Y * 2;
-    offset            = y_offset + (int4)(z_coord * src_stride_z);
-    offset            = min(offset, (int4)max_offset);
-    VEC_FLOAT values6 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + offset.s0));
-    VEC_FLOAT values7 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + offset.s1));
-    VEC_FLOAT values8 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + offset.s2));
+    values0 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + src_offset_z.s2 + src_offset_y.s0));
+    values1 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + src_offset_z.s2 + src_offset_y.s1));
+    values2 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + src_offset_z.s2 + src_offset_y.s2));
 
-    acc = fma(values0, w0, acc);
-    acc = fma(values1, w1, acc);
-    acc = fma(values2, w2, acc);
+    FILL_ZERO_OUT_OF_BOUND_3(DATA_TYPE, VEC_SIZE, values, src_coord_y | (int3)src_coord_z.s2);
 
-    acc = fma(values3, w3, acc);
-    acc = fma(values4, w4, acc);
-    acc = fma(values5, w5, acc);
-
-    acc = fma(values6, w6, acc);
-    acc = fma(values7, w7, acc);
-    acc = fma(values8, w8, acc);
+    acc0 = fma(values0, w6, acc0);
+    acc0 = fma(values1, w7, acc0);
+    acc0 = fma(values2, w8, acc0);
 
 #if defined(HAS_BIAS)
-    Vector    biases      = CONVERT_TO_VECTOR_STRUCT(biases);
-    VEC_FLOAT bias_values = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)biases.ptr);
-    acc += bias_values;
+    __global uchar *biases_addr = biases_ptr + biases_offset_first_element_in_bytes + x_offset;
+    VEC_FLOAT bias_values       = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)biases_addr);
+    acc0 += bias_values;
 #endif // defined(HAS_BIAS)
 
 #if defined(DST_DEPTH)
-    __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x * dst_step_x + y * dst_step_y + z * dst_step_z + b * dst_stride_w;
+    __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x_offset + y * dst_step_y + z * dst_step_z + b * dst_stride_w;
 #else  /* defined(DST_DEPTH) */
-    __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x * dst_step_x + y * dst_step_y + z * dst_step_z;
+    __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x_offset + y * dst_step_y + z * dst_step_z;
 #endif /* defined(DST_DEPTH) */
 
-    VSTORE(VEC_SIZE)
-    (ACTIVATION(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, acc, A_VAL, B_VAL), 0, (__global DATA_TYPE *)(dst_addr));
+    acc0 = ACTIVATION(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, acc0, A_VAL, B_VAL);
+    STORE_VECTOR_SELECT(acc, DATA_TYPE, dst_addr, VEC_SIZE, PARTIAL_STORE_N0, PARTIAL_STORE_N0 != 0 && get_global_id(0) == 0)
 }
 #endif // defined(CONV_STRIDE_X) && defined(CONV_STRIDE_Y)
 
@@ -1641,6 +1658,12 @@
  * @note It is possible to select the activation function to apply using -DACTIVATION_TYPE e.g. -DACTIVATION_TYPE=relu
  * @note A, B variables required by some activation functions are set using -DA_VAL= and -DB_VAL= respectively
  * @note Vector size should be given as a preprocessor argument using -DVEC_SIZE=size
+ * @note The size of the partial store block in y must be passed at compile time using -DPARTIAL_STORE_M0 (e.g. -DPARTIAL_STORE_M0=1)
+ * @note The size of the partial store block in x must be passed at compile time using -DPARTIAL_STORE_N0 (e.g. -DPARTIAL_STORE_N0=1)
+ * @note The size of the output's second dimension must be passed at compile time using -DDST_DIM_1 (e.g. -DDST_DIM_1=64)
+ * @note The size of the output's third dimension must be passed at compile time using -DDST_DIM_2 (e.g. -DDST_DIM_2=32)
+ * @note In case of biases, -DHAS_BIAS must to be passed at compile
+ * @note If the output tensor has more than three dimensions, its third dimension must be passed at compile time using -DDST_DEPTH (e.g. -DDST_DEPTH=32)
  *
  * @param[in] src_ptr                               Pointer to the source tensor. Supported data types: F16/F32
  * @param[in] src_stride_x                          Stride of the source tensor in X dimension (in bytes)
@@ -1679,14 +1702,15 @@
 __kernel void depthwise_convolution_3x3_nhwc_stride1(
     TENSOR4D_DECLARATION(src),
     TENSOR4D_DECLARATION(dst),
-    TENSOR3D_DECLARATION(weights),
+    TENSOR3D_DECLARATION(weights)
 #if defined(HAS_BIAS)
-    VECTOR_DECLARATION(biases),
+    ,
+    VECTOR_DECLARATION(biases)
 #endif /* defined(HAS_BIAS) */
-    int max_offset)
+)
 {
-    int x = get_global_id(0); // channels
-    int y = get_global_id(1); // spatial coordinate x
+    int x_offset = max((int)(get_global_id(0) * VEC_SIZE - (VEC_SIZE - PARTIAL_STORE_N0) % VEC_SIZE), 0) * sizeof(DATA_TYPE);
+    int y        = get_global_id(1); // spatial coordinate x
 #if defined(DST_DEPTH)
     int z = get_global_id(2) % (int)DST_DEPTH; // spatial coordinate y
     int b = get_global_id(2) / (int)DST_DEPTH; // batch
@@ -1694,79 +1718,52 @@
     int             z        = get_global_id(2); // spatial coordinate y
 #endif                                         // defined(DST_DEPTH)
 
-    Vector weights = CONVERT_TO_VECTOR_STRUCT(weights);
+    __global uchar *weights_addr = weights_ptr + weights_offset_first_element_in_bytes + x_offset;
 
 #if defined(DST_DEPTH)
-    __global uchar *src_addr = src_ptr + src_offset_first_element_in_bytes + x * sizeof(DATA_TYPE) * VEC_SIZE + b * src_stride_w;
+    __global uchar *src_addr = src_ptr + src_offset_first_element_in_bytes + x_offset + b * src_stride_w;
 #else  /* defined(DST_DEPTH) */
-    __global uchar *src_addr = src_ptr + src_offset_first_element_in_bytes + x * sizeof(DATA_TYPE) * VEC_SIZE;
+    __global uchar *src_addr = src_ptr + src_offset_first_element_in_bytes + x_offset;
 #endif /* defined(DST_DEPTH) */
 
-    int  z_coord  = 0;
-    int4 offset   = 0;
-    int4 y_offset = ((int4)(y * NUM_ROWS_PROCESSED) + (int4)(0, 1, 2, 3) - (int)CONV_PAD_LEFT) * (int4)src_stride_y;
+    int4 src_coord_y = (int4)(y * NUM_ROWS_PROCESSED - CONV_PAD_LEFT) + V_OFFS4(int4);
+    int4 src_coord_z = (int4)(z * NUM_PLANES_PROCESSED - CONV_PAD_TOP) + V_OFFS4(int4);
 
-    // We compute 2x2x2 [C,W,H] elements
+    int4 src_offset_y = clamp(src_coord_y, (int4)0, (int4)(SRC_DIM_1 - 1));
+    int4 src_offset_z = clamp(src_coord_z, (int4)0, (int4)(SRC_DIM_2 - 1));
+
+    // Use these vectors to check whether the unclamped load would have been out of bounds
+    src_coord_y = (src_offset_y != src_coord_y);
+    src_coord_z = (src_offset_z != src_coord_z);
+
+    src_offset_y *= (int4)src_stride_y;
+    src_offset_z *= (int4)src_stride_z;
+
+    // We compute VEC_SIZEx2x2 [C,W,H] elements
     VEC_FLOAT acc0 = 0;
     VEC_FLOAT acc1 = 0;
     VEC_FLOAT acc2 = 0;
     VEC_FLOAT acc3 = 0;
 
     // Load weights
-    VEC_FLOAT w0 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights.ptr + 0 * weights_stride_y + 0 * weights_stride_z));
-    VEC_FLOAT w1 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights.ptr + 1 * weights_stride_y + 0 * weights_stride_z));
-    VEC_FLOAT w2 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights.ptr + 2 * weights_stride_y + 0 * weights_stride_z));
-    VEC_FLOAT w3 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights.ptr + 0 * weights_stride_y + 1 * weights_stride_z));
-    VEC_FLOAT w4 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights.ptr + 1 * weights_stride_y + 1 * weights_stride_z));
-    VEC_FLOAT w5 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights.ptr + 2 * weights_stride_y + 1 * weights_stride_z));
-    VEC_FLOAT w6 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights.ptr + 0 * weights_stride_y + 2 * weights_stride_z));
-    VEC_FLOAT w7 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights.ptr + 1 * weights_stride_y + 2 * weights_stride_z));
-    VEC_FLOAT w8 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights.ptr + 2 * weights_stride_y + 2 * weights_stride_z));
+    VEC_FLOAT w0 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights_addr + 0 * weights_stride_y + 0 * weights_stride_z));
+    VEC_FLOAT w1 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights_addr + 1 * weights_stride_y + 0 * weights_stride_z));
+    VEC_FLOAT w2 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights_addr + 2 * weights_stride_y + 0 * weights_stride_z));
+    VEC_FLOAT w3 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights_addr + 0 * weights_stride_y + 1 * weights_stride_z));
+    VEC_FLOAT w4 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights_addr + 1 * weights_stride_y + 1 * weights_stride_z));
+    VEC_FLOAT w5 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights_addr + 2 * weights_stride_y + 1 * weights_stride_z));
+    VEC_FLOAT w6 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights_addr + 0 * weights_stride_y + 2 * weights_stride_z));
+    VEC_FLOAT w7 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights_addr + 1 * weights_stride_y + 2 * weights_stride_z));
+    VEC_FLOAT w8 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(weights_addr + 2 * weights_stride_y + 2 * weights_stride_z));
 
     // Load input values
     // z == 0
-    // Clamp z_coord as for z = 0, it can be negative
-    // z_coord is casted to unsigned int in order to use just a min() operation
-    // A "-1" 32 bit signed variable converted to unsigned gives 4294967295
-    z_coord = z * (int)NUM_PLANES_PROCESSED - (int)CONV_PAD_TOP;
-    z_coord = min((uint)z_coord, (uint)SRC_DIM_2);
-    offset  = y_offset + (int4)(z_coord * src_stride_z);
-    offset  = min(offset, (int4)max_offset);
+    VEC_FLOAT values0 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + src_offset_z.s0 + src_offset_y.s0));
+    VEC_FLOAT values1 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + src_offset_z.s0 + src_offset_y.s1));
+    VEC_FLOAT values2 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + src_offset_z.s0 + src_offset_y.s2));
+    VEC_FLOAT values3 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + src_offset_z.s0 + src_offset_y.s3));
 
-    VEC_FLOAT values0 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + offset.s0));
-    VEC_FLOAT values1 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + offset.s1));
-    VEC_FLOAT values2 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + offset.s2));
-    VEC_FLOAT values3 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + offset.s3));
-
-    // z == 1
-    // z_coord can be only negative for z = 0 so we do not need to clamp it
-    // Moreover z_coord cannot be out-of-bound for z = 1 so we do not need to clamp the offset
-    z_coord           = z * (int)NUM_PLANES_PROCESSED - (int)CONV_PAD_TOP + 1;
-    offset            = y_offset + (int4)(z_coord * src_stride_z);
-    VEC_FLOAT values4 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + offset.s0));
-    VEC_FLOAT values5 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + offset.s1));
-    VEC_FLOAT values6 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + offset.s2));
-    VEC_FLOAT values7 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + offset.s3));
-
-    // z == 2
-    // After z = 1 we can simply add src_stride_z to offset without updating z_coord
-    // However offset can be out-of-bound so we need to check if it is greater than max_offset
-    offset += (int4)src_stride_z;
-    offset             = min(offset, (int4)max_offset);
-    VEC_FLOAT values8  = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + offset.s0));
-    VEC_FLOAT values9  = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + offset.s1));
-    VEC_FLOAT values10 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + offset.s2));
-    VEC_FLOAT values11 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + offset.s3));
-
-    // z == 3
-    // After z = 1 we can simply add src_stride_z to offset without updating z_coord
-    // However offset can be out-of-bound so we need to check if it is greater than max_offset
-    offset += (int4)src_stride_z;
-    offset             = min(offset, (int4)max_offset);
-    VEC_FLOAT values12 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + offset.s0));
-    VEC_FLOAT values13 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + offset.s1));
-    VEC_FLOAT values14 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + offset.s2));
-    VEC_FLOAT values15 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + offset.s3));
+    FILL_ZERO_OUT_OF_BOUND_4(DATA_TYPE, VEC_SIZE, values, src_coord_y | (int4)src_coord_z.s0);
 
     acc0 = fma(values0, w0, acc0);
     acc0 = fma(values1, w1, acc0);
@@ -1775,45 +1772,69 @@
     acc1 = fma(values2, w1, acc1);
     acc1 = fma(values3, w2, acc1);
 
-    acc0 = fma(values4, w3, acc0);
-    acc0 = fma(values5, w4, acc0);
-    acc0 = fma(values6, w5, acc0);
-    acc1 = fma(values5, w3, acc1);
-    acc1 = fma(values6, w4, acc1);
-    acc1 = fma(values7, w5, acc1);
+    // z == 1
+    values0 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + src_offset_z.s1 + src_offset_y.s0));
+    values1 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + src_offset_z.s1 + src_offset_y.s1));
+    values2 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + src_offset_z.s1 + src_offset_y.s2));
+    values3 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + src_offset_z.s1 + src_offset_y.s3));
 
-    acc0 = fma(values8, w6, acc0);
-    acc0 = fma(values9, w7, acc0);
-    acc0 = fma(values10, w8, acc0);
-    acc1 = fma(values9, w6, acc1);
-    acc1 = fma(values10, w7, acc1);
-    acc1 = fma(values11, w8, acc1);
+    FILL_ZERO_OUT_OF_BOUND_4(DATA_TYPE, VEC_SIZE, values, src_coord_y | (int4)src_coord_z.s1);
 
-    acc2 = fma(values4, w0, acc2);
-    acc2 = fma(values5, w1, acc2);
-    acc2 = fma(values6, w2, acc2);
-    acc3 = fma(values5, w0, acc3);
-    acc3 = fma(values6, w1, acc3);
-    acc3 = fma(values7, w2, acc3);
+    acc0 = fma(values0, w3, acc0);
+    acc0 = fma(values1, w4, acc0);
+    acc0 = fma(values2, w5, acc0);
+    acc1 = fma(values1, w3, acc1);
+    acc1 = fma(values2, w4, acc1);
+    acc1 = fma(values3, w5, acc1);
 
-    acc2 = fma(values8, w3, acc2);
-    acc2 = fma(values9, w4, acc2);
-    acc2 = fma(values10, w5, acc2);
-    acc3 = fma(values9, w3, acc3);
-    acc3 = fma(values10, w4, acc3);
-    acc3 = fma(values11, w5, acc3);
+    acc2 = fma(values0, w0, acc2);
+    acc2 = fma(values1, w1, acc2);
+    acc2 = fma(values2, w2, acc2);
+    acc3 = fma(values1, w0, acc3);
+    acc3 = fma(values2, w1, acc3);
+    acc3 = fma(values3, w2, acc3);
 
-    acc2 = fma(values12, w6, acc2);
-    acc2 = fma(values13, w7, acc2);
-    acc2 = fma(values14, w8, acc2);
-    acc3 = fma(values13, w6, acc3);
-    acc3 = fma(values14, w7, acc3);
-    acc3 = fma(values15, w8, acc3);
+    // z == 2
+    values0 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + src_offset_z.s2 + src_offset_y.s0));
+    values1 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + src_offset_z.s2 + src_offset_y.s1));
+    values2 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + src_offset_z.s2 + src_offset_y.s2));
+    values3 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + src_offset_z.s2 + src_offset_y.s3));
+
+    FILL_ZERO_OUT_OF_BOUND_4(DATA_TYPE, VEC_SIZE, values, src_coord_y | (int4)src_coord_z.s2);
+
+    acc0 = fma(values0, w6, acc0);
+    acc0 = fma(values1, w7, acc0);
+    acc0 = fma(values2, w8, acc0);
+    acc1 = fma(values1, w6, acc1);
+    acc1 = fma(values2, w7, acc1);
+    acc1 = fma(values3, w8, acc1);
+
+    acc2 = fma(values0, w3, acc2);
+    acc2 = fma(values1, w4, acc2);
+    acc2 = fma(values2, w5, acc2);
+    acc3 = fma(values1, w3, acc3);
+    acc3 = fma(values2, w4, acc3);
+    acc3 = fma(values3, w5, acc3);
+
+    // z == 3
+    values0 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + src_offset_z.s3 + src_offset_y.s0));
+    values1 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + src_offset_z.s3 + src_offset_y.s1));
+    values2 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + src_offset_z.s3 + src_offset_y.s2));
+    values3 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + src_offset_z.s3 + src_offset_y.s3));
+
+    FILL_ZERO_OUT_OF_BOUND_4(DATA_TYPE, VEC_SIZE, values, src_coord_y | (int4)src_coord_z.s3);
+
+    acc2 = fma(values0, w6, acc2);
+    acc2 = fma(values1, w7, acc2);
+    acc2 = fma(values2, w8, acc2);
+    acc3 = fma(values1, w6, acc3);
+    acc3 = fma(values2, w7, acc3);
+    acc3 = fma(values3, w8, acc3);
 
 #if defined(HAS_BIAS)
-    Vector biases = CONVERT_TO_VECTOR_STRUCT(biases);
+    __global uchar *biases_addr = biases_ptr + biases_offset_first_element_in_bytes + x_offset;
 
-    VEC_FLOAT bias_values = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)biases.ptr);
+    VEC_FLOAT bias_values = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)biases_addr);
 
     acc0 += bias_values;
     acc1 += bias_values;
@@ -1821,25 +1842,34 @@
     acc3 += bias_values;
 #endif // defined(HAS_BIAS)
 
-#if defined(DST_DEPTH)
-    __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x * dst_step_x + y * dst_step_y + (z * NUM_PLANES_PROCESSED) * dst_step_z + b * dst_stride_w;
-#else  /* defined(DST_DEPTH) */
-    __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x * dst_step_x + y * dst_step_y + (z * NUM_PLANES_PROCESSED) * dst_step_z;
-#endif /* defined(DST_DEPTH) */
+    int2 dst_offset_y = min((int2)(y * NUM_ROWS_PROCESSED) + V_OFFS2(int2), (int2)(DST_DIM_1 - 1)) * (int2)dst_stride_y;
+    int  dst_coord_z  = z * NUM_PLANES_PROCESSED;
 
-    VSTORE(VEC_SIZE)
-    (ACTIVATION(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, acc0, A_VAL, B_VAL), 0, (__global DATA_TYPE *)(dst_addr + 0 * dst_stride_y));
-    VSTORE(VEC_SIZE)
-    (ACTIVATION(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, acc1, A_VAL, B_VAL), 0, (__global DATA_TYPE *)(dst_addr + 1 * dst_stride_y));
+#if defined(DST_DEPTH)
+    __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x_offset + dst_coord_z * dst_stride_z + b * dst_stride_w;
+#else  // defined(DST_DEPTH)
+    __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x_offset + dst_coord_z * dst_stride_z;
+#endif //  defined(DST_DEPTH)
+
+    /* Store vectors in reverse order along the Y. The Y offsets are calculated so that they are forced to be in bound.
+     * If only the first address is in bound, the Y offset of the second address will be brought back and there will be 2 writes in the same location for the same thread.
+     * Since the last vector to be written is always the valid one for that location, it overwrites the wrong values.
+     */
+    values0 = ACTIVATION(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, acc1, A_VAL, B_VAL);
+    STORE_VECTOR_SELECT(values, DATA_TYPE, dst_addr + dst_offset_y.s1, VEC_SIZE, PARTIAL_STORE_N0, PARTIAL_STORE_N0 != 0 && get_global_id(0) == 0)
+
+    values0 = ACTIVATION(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, acc0, A_VAL, B_VAL);
+    STORE_VECTOR_SELECT(values, DATA_TYPE, dst_addr + dst_offset_y.s0, VEC_SIZE, PARTIAL_STORE_N0, PARTIAL_STORE_N0 != 0 && get_global_id(0) == 0)
 
 #if((DST_DIM_2 % NUM_PLANES_PROCESSED) != 0)
-    if((z * NUM_PLANES_PROCESSED + 1) < DST_DIM_2)
+    if((dst_coord_z + 1) < DST_DIM_2)
 #endif // ((DST_DIM_2 % NUM_PLANES_PROCESSED) != 0)
     {
-        VSTORE(VEC_SIZE)
-        (ACTIVATION(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, acc2, A_VAL, B_VAL), 0, (__global DATA_TYPE *)(dst_addr + 0 * dst_stride_y + 1 * dst_stride_z));
-        VSTORE(VEC_SIZE)
-        (ACTIVATION(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, acc3, A_VAL, B_VAL), 0, (__global DATA_TYPE *)(dst_addr + 1 * dst_stride_y + 1 * dst_stride_z));
+        values0 = ACTIVATION(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, acc3, A_VAL, B_VAL);
+        STORE_VECTOR_SELECT(values, DATA_TYPE, dst_addr + dst_stride_z + dst_offset_y.s1, VEC_SIZE, PARTIAL_STORE_N0, PARTIAL_STORE_N0 != 0 && get_global_id(0) == 0)
+
+        values0 = ACTIVATION(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, acc2, A_VAL, B_VAL);
+        STORE_VECTOR_SELECT(values, DATA_TYPE, dst_addr + dst_stride_z + dst_offset_y.s0, VEC_SIZE, PARTIAL_STORE_N0, PARTIAL_STORE_N0 != 0 && get_global_id(0) == 0)
     }
 }
 
diff --git a/src/core/CL/cl_kernels/gemm.cl b/src/core/CL/cl_kernels/gemm.cl
index f29f2c1..653aa55 100644
--- a/src/core/CL/cl_kernels/gemm.cl
+++ b/src/core/CL/cl_kernels/gemm.cl
@@ -1282,7 +1282,7 @@
     const bool cond_x = ((x + 1) * N0 >= N);
 
     // Store output block
-    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, N, cond_y, cond_x);
+    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);
 
 #undef RHS_BLOCK_SIZE
 #undef RHS_OFFSET_X
@@ -1628,7 +1628,7 @@
     const bool cond_x = ((x + 1) * N0 >= N);
 
     // Store output block
-    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, N, cond_y, cond_x);
+    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);
 
 #undef RHS_BLOCK_SIZE
 #undef RHS_OFFSET_X
@@ -2024,7 +2024,7 @@
     const bool cond_x = ((x + 1) * N0 >= N);
 
     // Store output block
-    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, N, cond_y, cond_x);
+    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);
 
 #undef RHS_BLOCK_SIZE
 #undef RHS_OFFSET_X
@@ -2333,7 +2333,7 @@
     const bool cond_x = ((x + 1) * N0 >= N);
 
     // Store output block
-    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, N, cond_y, cond_x);
+    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);
 
 #undef RHS_BLOCK_SIZE
 #undef RHS_OFFSET_X
@@ -2775,9 +2775,9 @@
     // Store output block
 #if defined(MIXED_PRECISION)
     CONVERT_BLOCK(M0, N0, DATA_TYPE, c, c_lp);
-    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c_lp, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, N, cond_y, cond_x);
+    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c_lp, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);
 #else  // defined(MIXED_PRECISION)
-    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, N, cond_y, cond_x);
+    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);
 #endif // defined(MIXED_PRECISION)
 
 #undef LHS_BLOCK_SIZE
@@ -3045,9 +3045,9 @@
     // Store output block
 #if defined(MIXED_PRECISION)
     CONVERT_BLOCK(M0, N0, DATA_TYPE, c, c_lp);
-    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c_lp, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, N, cond_y, cond_x);
+    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c_lp, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);
 #else  // defined(MIXED_PRECISION)
-    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, N, cond_y, cond_x);
+    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);
 #endif // defined(MIXED_PRECISION)
 
 #undef LHS_BLOCK_SIZE
@@ -3539,9 +3539,9 @@
     // Store output block
 #if defined(MIXED_PRECISION)
     CONVERT_BLOCK(M0, N0, DATA_TYPE, c, c_lp);
-    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c_lp, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, N, cond_y, cond_x);
+    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c_lp, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);
 #else  // defined(MIXED_PRECISION)
-    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, N, cond_y, cond_x);
+    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);
 #endif // defined(MIXED_PRECISION)
 
 #undef LHS_BLOCK_SIZE
@@ -3906,9 +3906,9 @@
     // Store output block
 #if defined(MIXED_PRECISION)
     CONVERT_BLOCK(M0, N0, DATA_TYPE, c, c_lp);
-    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c_lp, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, N, cond_y, cond_x);
+    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c_lp, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);
 #else  // defined(MIXED_PRECISION)
-    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, N, cond_y, cond_x);
+    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);
 #endif // defined(MIXED_PRECISION)
 
 #undef LHS_BLOCK_SIZE
@@ -4287,7 +4287,7 @@
     const bool cond_x = ((x + 1) * N0 >= N);
 
     // Store output block
-    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, N, cond_y, cond_x);
+    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);
 
 #undef RHS_BLOCK_SIZE
 #undef RHS_OFFSET_X
diff --git a/src/core/CL/cl_kernels/gemm_helpers.h b/src/core/CL/cl_kernels/gemm_helpers.h
index 1e020e1..2534204 100644
--- a/src/core/CL/cl_kernels/gemm_helpers.h
+++ b/src/core/CL/cl_kernels/gemm_helpers.h
@@ -708,233 +708,6 @@
 #define CALCULATE_Z_OFFSET(M0, DATA_TYPE, Z, Y, HEIGHT_GEMM3D, DEPTH_GEMM3D, CROSS_PLANE_PAD, STRIDE_Y) CALCULATE_Z_OFFSET_STR(M0, DATA_TYPE, Z, Y, HEIGHT_GEMM3D, DEPTH_GEMM3D, CROSS_PLANE_PAD, STRIDE_Y)
 /** @} */ // end of group CALCULATE_Z_OFFSET
 
-/** Store the 0 to (n-1)th rows of the given variables
- * @name STORE_ROW_n
- *
- * @param[in] N0        The width of the passed in vector. Supported: 1, 2, 3, 4, 8, 16
- * @param[in] DATA_TYPE The data type of the vectors
- * @param[in] BASENAME  The basename of the variables
- * @param[in] PTR       The base pointer
- * @param[in] STRIDE_Y  The stride value in y-axis direction
- * @param[in] Z         The offset in z-axis direction
- * @{
- */
-#define STORE_ROW_1(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
-    VSTORE(N0)                                                 \
-    (BASENAME##0, 0, (__global DATA_TYPE *)(PTR + 0 * STRIDE_Y + Z##0));
-
-#define STORE_ROW_2(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
-    STORE_ROW_1(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
-    VSTORE(N0)                                                 \
-    (BASENAME##1, 0, (__global DATA_TYPE *)(PTR + 1 * STRIDE_Y + Z##1));
-
-#define STORE_ROW_3(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
-    STORE_ROW_2(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
-    VSTORE(N0)                                                 \
-    (BASENAME##2, 0, (__global DATA_TYPE *)(PTR + 2 * STRIDE_Y + Z##2));
-
-#define STORE_ROW_4(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
-    STORE_ROW_3(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
-    VSTORE(N0)                                                 \
-    (BASENAME##3, 0, (__global DATA_TYPE *)(PTR + 3 * STRIDE_Y + Z##3));
-
-#define STORE_ROW_5(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
-    STORE_ROW_4(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
-    VSTORE(N0)                                                 \
-    (BASENAME##4, 0, (__global DATA_TYPE *)(PTR + 4 * STRIDE_Y + Z##4));
-
-#define STORE_ROW_6(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
-    STORE_ROW_5(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
-    VSTORE(N0)                                                 \
-    (BASENAME##5, 0, (__global DATA_TYPE *)(PTR + 5 * STRIDE_Y + Z##5));
-
-#define STORE_ROW_7(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
-    STORE_ROW_6(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
-    VSTORE(N0)                                                 \
-    (BASENAME##6, 0, (__global DATA_TYPE *)(PTR + 6 * STRIDE_Y + Z##6));
-
-#define STORE_ROW_8(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
-    STORE_ROW_7(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
-    VSTORE(N0)                                                 \
-    (BASENAME##7, 0, (__global DATA_TYPE *)(PTR + 7 * STRIDE_Y + Z##7));
-
-#define STORE_ROW_9(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
-    STORE_ROW_8(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
-    VSTORE(N0)                                                 \
-    (BASENAME##8, 0, (__global DATA_TYPE *)(PTR + 8 * STRIDE_Y + Z##8));
-
-#define STORE_ROW_10(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
-    STORE_ROW_9(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)      \
-    VSTORE(N0)                                                  \
-    (BASENAME##9, 0, (__global DATA_TYPE *)(PTR + 9 * STRIDE_Y + Z##9));
-
-#define STORE_ROW_11(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
-    STORE_ROW_10(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
-    VSTORE(N0)                                                  \
-    (BASENAME##A, 0, (__global DATA_TYPE *)(PTR + 10 * STRIDE_Y + Z##A));
-
-#define STORE_ROW_12(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
-    STORE_ROW_11(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
-    VSTORE(N0)                                                  \
-    (BASENAME##B, 0, (__global DATA_TYPE *)(PTR + 11 * STRIDE_Y + Z##B));
-
-#define STORE_ROW_13(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
-    STORE_ROW_12(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
-    VSTORE(N0)                                                  \
-    (BASENAME##C, 0, (__global DATA_TYPE *)(PTR + 12 * STRIDE_Y + Z##C));
-
-#define STORE_ROW_14(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
-    STORE_ROW_13(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
-    VSTORE(N0)                                                  \
-    (BASENAME##D, 0, (__global DATA_TYPE *)(PTR + 13 * STRIDE_Y + Z##D));
-
-#define STORE_ROW_15(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
-    STORE_ROW_14(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
-    VSTORE(N0)                                                  \
-    (BASENAME##E, 0, (__global DATA_TYPE *)(PTR + 14 * STRIDE_Y + Z##E));
-
-#define STORE_ROW_16(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
-    STORE_ROW_15(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
-    VSTORE(N0)                                                  \
-    (BASENAME##F, 0, (__global DATA_TYPE *)(PTR + 15 * STRIDE_Y + Z##F));
-/** @} */ // end of groupd STORE_ROW_n
-
-/** Convert and store the 0th to (n-1)th rows of the given variables
- * @name CONVERT_STORE_ROW_n
- *
- * @param[in] N0        The size of the vectors
- * @param[in] DATA_TYPE The data type of the vectors
- * @param[in] BASENAME  The basename of the variables
- * @param[in] PTR       The base pointer
- * @param[in] STRIDE_Y  The stride value in y-axis direction
- * @param[in] Z         The offset in z-axis direction
- * @{
- */
-#define CONVERT_STORE_ROW_1(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
-    VSTORE(N0)                                                         \
-    (CONVERT_SAT((BASENAME##0), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 0 * STRIDE_Y + Z##0));
-
-#define CONVERT_STORE_ROW_2(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
-    CONVERT_STORE_ROW_1(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
-    VSTORE(N0)                                                         \
-    (CONVERT_SAT((BASENAME##1), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 1 * STRIDE_Y + Z##1));
-
-#define CONVERT_STORE_ROW_3(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
-    CONVERT_STORE_ROW_2(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
-    VSTORE(N0)                                                         \
-    (CONVERT_SAT((BASENAME##2), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 2 * STRIDE_Y + Z##2));
-
-#define CONVERT_STORE_ROW_4(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
-    CONVERT_STORE_ROW_3(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
-    VSTORE(N0)                                                         \
-    (CONVERT_SAT((BASENAME##3), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 3 * STRIDE_Y + Z##3));
-
-#define CONVERT_STORE_ROW_5(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
-    CONVERT_STORE_ROW_4(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
-    VSTORE(N0)                                                         \
-    (CONVERT_SAT((BASENAME##4), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 4 * STRIDE_Y + Z##4));
-
-#define CONVERT_STORE_ROW_6(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
-    CONVERT_STORE_ROW_5(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
-    VSTORE(N0)                                                         \
-    (CONVERT_SAT((BASENAME##5), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 5 * STRIDE_Y + Z##5));
-
-#define CONVERT_STORE_ROW_7(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
-    CONVERT_STORE_ROW_6(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
-    VSTORE(N0)                                                         \
-    (CONVERT_SAT((BASENAME##6), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 6 * STRIDE_Y + Z##6));
-
-#define CONVERT_STORE_ROW_8(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
-    CONVERT_STORE_ROW_7(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
-    VSTORE(N0)                                                         \
-    (CONVERT_SAT((BASENAME##7), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 7 * STRIDE_Y + Z##7));
-
-#define CONVERT_STORE_ROW_9(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
-    CONVERT_STORE_ROW_8(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
-    VSTORE(N0)                                                         \
-    (CONVERT_SAT((BASENAME##8), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 8 * STRIDE_Y + Z##8));
-
-#define CONVERT_STORE_ROW_10(N0, DATA, BASENAME, PTR, STRIDE_Y, Z) \
-    CONVERT_STORE_ROW_9(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
-    VSTORE(N0)                                                     \
-    (CONVERT_SAT((BASENAME##9), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 9 * STRIDE_Y + Z##9));
-
-#define CONVERT_STORE_ROW_11(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
-    CONVERT_STORE_ROW_10(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
-    VSTORE(N0)                                                          \
-    (CONVERT_SAT((BASENAME##A), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 10 * STRIDE_Y + Z##A));
-
-#define CONVERT_STORE_ROW_12(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
-    CONVERT_STORE_ROW_11(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
-    VSTORE(N0)                                                          \
-    (CONVERT_SAT((BASENAME##B), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 11 * STRIDE_Y + Z##B));
-
-#define CONVERT_STORE_ROW_13(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
-    CONVERT_STORE_ROW_12(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
-    VSTORE(N0)                                                          \
-    (CONVERT_SAT((BASENAME##C), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 12 * STRIDE_Y + Z##C));
-
-#define CONVERT_STORE_ROW_14(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
-    CONVERT_STORE_ROW_13(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
-    VSTORE(N0)                                                          \
-    (CONVERT_SAT((BASENAME##D), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 13 * STRIDE_Y + Z##D));
-
-#define CONVERT_STORE_ROW_15(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
-    CONVERT_STORE_ROW_14(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
-    VSTORE(N0)                                                          \
-    (CONVERT_SAT((BASENAME##E), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 14 * STRIDE_Y + Z##E));
-
-#define CONVERT_STORE_ROW_16(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
-    CONVERT_STORE_ROW_15(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
-    VSTORE(N0)                                                          \
-    (CONVERT_SAT((BASENAME##F), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 15 * STRIDE_Y + Z##F));
-
-/** @} */ // end of groupd CONVERT_STORE_ROW_n
-
-/** Store a block of the given size M0xN0
- * @name STORE_BLOCK
- *
- * Supported cases are M0=1,2,3,...,16 and N0=2,3,4,8,16.
- * The data to store is expected to have consecutive names for each row.
- * E.g., for M0=3 and basename=c, the expected names are c0, c1 and c2.
- * The Z offset is expected to have consecutive names.
- * E.g., for M0=3 and Z=zin, the expected z offset names are zin0, zin1 and zin2.
- *
- * @param[in] M0        The number of rows to store
- * @param[in] N0        The size of each vector
- * @param[in] DATA_TYPE The data type of the vectors
- * @param[in] BASENAME  The basename of the variables
- * @param[in] PTR       The base pointer
- * @param[in] STRIDE_Y  The stride value in y-axis direction
- * @param[in] Z         The offset in z-axis direction
- * @{
- */
-#define STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) STORE_ROW_##M0(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
-#define STORE_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
-/** @} */ // end of group STORE_BLOCK
-
-/** Convert and store a block of the given size M0xN0
- * @name CONVERT_STORE_BLOCK
- *
- * Supported cases are M0=1,2,3,...,16 and N0=2,3,4,8,16.
- * The data to store is expected to have consecutive names for each row.
- * E.g., for M0=3 and basename=c, the expected names are c0, c1 and c2.
- * The Z offset is expected to have consecutive names.
- * E.g., for M0=3 and Z=zin, the expected z offset names are zin0, zin1 and zin2.
- *
- * @param[in] M0        The number of rows to store
- * @param[in] N0        The size of each vector
- * @param[in] DATA_TYPE The data type of the vectors
- * @param[in] BASENAME  The basename of the variables
- * @param[in] PTR       The base pointer
- * @param[in] STRIDE_Y  The stride value in y-axis direction
- * @param[in] Z         The offset in z-axis direction
- * @{
- */
-#define CONVERT_STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) CONVERT_STORE_ROW_##M0(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
-#define CONVERT_STORE_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) CONVERT_STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
-/** @} */ // end of group CONVERT_STORE_BLOCK
-
 /** Scale the rows in the given variables (BASENAME0 to BASENAMEn-1)
  * @name SCALE_ROW_n
  *
diff --git a/src/core/CL/cl_kernels/gemmlowp.cl b/src/core/CL/cl_kernels/gemmlowp.cl
index da92548..9bdd5a2 100644
--- a/src/core/CL/cl_kernels/gemmlowp.cl
+++ b/src/core/CL/cl_kernels/gemmlowp.cl
@@ -1096,7 +1096,7 @@
     const bool cond_x = ((x + 1) * N0 >= N);
 
     // Store output block
-    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, int, c, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, N, cond_y, cond_x);
+    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, int, c, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);
 }
 #endif // defined(M0) && defined(N0) && defined(K0) && defined(K) && defined(PARTIAL_STORE_M0) && defined(PARTIAL_STORE_N0)
 
diff --git a/src/core/CL/cl_kernels/load_store_utility.h b/src/core/CL/cl_kernels/load_store_utility.h
index cb833d0..84b6d2e 100644
--- a/src/core/CL/cl_kernels/load_store_utility.h
+++ b/src/core/CL/cl_kernels/load_store_utility.h
@@ -21,6 +21,234 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
+
+/** Store the 0 to (n-1)th rows of the given variables
+ * @name STORE_ROW_n
+ *
+ * @param[in] N0        The width of the passed in vector. Supported: 1, 2, 3, 4, 8, 16
+ * @param[in] DATA_TYPE The data type of the vectors
+ * @param[in] BASENAME  The basename of the variables
+ * @param[in] PTR       The base pointer
+ * @param[in] STRIDE_Y  The stride value in y-axis direction
+ * @param[in] Z         The offset in z-axis direction
+ * @{
+ */
+#define STORE_ROW_1(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+    VSTORE(N0)                                                 \
+    (BASENAME##0, 0, (__global DATA_TYPE *)(PTR + 0 * STRIDE_Y + Z##0));
+
+#define STORE_ROW_2(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+    STORE_ROW_1(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
+    VSTORE(N0)                                                 \
+    (BASENAME##1, 0, (__global DATA_TYPE *)(PTR + 1 * STRIDE_Y + Z##1));
+
+#define STORE_ROW_3(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+    STORE_ROW_2(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
+    VSTORE(N0)                                                 \
+    (BASENAME##2, 0, (__global DATA_TYPE *)(PTR + 2 * STRIDE_Y + Z##2));
+
+#define STORE_ROW_4(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+    STORE_ROW_3(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
+    VSTORE(N0)                                                 \
+    (BASENAME##3, 0, (__global DATA_TYPE *)(PTR + 3 * STRIDE_Y + Z##3));
+
+#define STORE_ROW_5(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+    STORE_ROW_4(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
+    VSTORE(N0)                                                 \
+    (BASENAME##4, 0, (__global DATA_TYPE *)(PTR + 4 * STRIDE_Y + Z##4));
+
+#define STORE_ROW_6(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+    STORE_ROW_5(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
+    VSTORE(N0)                                                 \
+    (BASENAME##5, 0, (__global DATA_TYPE *)(PTR + 5 * STRIDE_Y + Z##5));
+
+#define STORE_ROW_7(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+    STORE_ROW_6(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
+    VSTORE(N0)                                                 \
+    (BASENAME##6, 0, (__global DATA_TYPE *)(PTR + 6 * STRIDE_Y + Z##6));
+
+#define STORE_ROW_8(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+    STORE_ROW_7(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
+    VSTORE(N0)                                                 \
+    (BASENAME##7, 0, (__global DATA_TYPE *)(PTR + 7 * STRIDE_Y + Z##7));
+
+#define STORE_ROW_9(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+    STORE_ROW_8(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
+    VSTORE(N0)                                                 \
+    (BASENAME##8, 0, (__global DATA_TYPE *)(PTR + 8 * STRIDE_Y + Z##8));
+
+#define STORE_ROW_10(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+    STORE_ROW_9(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)      \
+    VSTORE(N0)                                                  \
+    (BASENAME##9, 0, (__global DATA_TYPE *)(PTR + 9 * STRIDE_Y + Z##9));
+
+#define STORE_ROW_11(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+    STORE_ROW_10(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
+    VSTORE(N0)                                                  \
+    (BASENAME##A, 0, (__global DATA_TYPE *)(PTR + 10 * STRIDE_Y + Z##A));
+
+#define STORE_ROW_12(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+    STORE_ROW_11(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
+    VSTORE(N0)                                                  \
+    (BASENAME##B, 0, (__global DATA_TYPE *)(PTR + 11 * STRIDE_Y + Z##B));
+
+#define STORE_ROW_13(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+    STORE_ROW_12(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
+    VSTORE(N0)                                                  \
+    (BASENAME##C, 0, (__global DATA_TYPE *)(PTR + 12 * STRIDE_Y + Z##C));
+
+#define STORE_ROW_14(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+    STORE_ROW_13(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
+    VSTORE(N0)                                                  \
+    (BASENAME##D, 0, (__global DATA_TYPE *)(PTR + 13 * STRIDE_Y + Z##D));
+
+#define STORE_ROW_15(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+    STORE_ROW_14(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
+    VSTORE(N0)                                                  \
+    (BASENAME##E, 0, (__global DATA_TYPE *)(PTR + 14 * STRIDE_Y + Z##E));
+
+#define STORE_ROW_16(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+    STORE_ROW_15(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
+    VSTORE(N0)                                                  \
+    (BASENAME##F, 0, (__global DATA_TYPE *)(PTR + 15 * STRIDE_Y + Z##F));
+/** @} */ // end of groupd STORE_ROW_n
+
+/** Convert and store the 0th to (n-1)th rows of the given variables
+ * @name CONVERT_STORE_ROW_n
+ *
+ * @param[in] N0        The size of the vectors
+ * @param[in] DATA_TYPE The data type of the vectors
+ * @param[in] BASENAME  The basename of the variables
+ * @param[in] PTR       The base pointer
+ * @param[in] STRIDE_Y  The stride value in y-axis direction
+ * @param[in] Z         The offset in z-axis direction
+ * @{
+ */
+#define CONVERT_STORE_ROW_1(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+    VSTORE(N0)                                                         \
+    (CONVERT_SAT((BASENAME##0), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 0 * STRIDE_Y + Z##0));
+
+#define CONVERT_STORE_ROW_2(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+    CONVERT_STORE_ROW_1(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
+    VSTORE(N0)                                                         \
+    (CONVERT_SAT((BASENAME##1), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 1 * STRIDE_Y + Z##1));
+
+#define CONVERT_STORE_ROW_3(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+    CONVERT_STORE_ROW_2(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
+    VSTORE(N0)                                                         \
+    (CONVERT_SAT((BASENAME##2), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 2 * STRIDE_Y + Z##2));
+
+#define CONVERT_STORE_ROW_4(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+    CONVERT_STORE_ROW_3(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
+    VSTORE(N0)                                                         \
+    (CONVERT_SAT((BASENAME##3), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 3 * STRIDE_Y + Z##3));
+
+#define CONVERT_STORE_ROW_5(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+    CONVERT_STORE_ROW_4(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
+    VSTORE(N0)                                                         \
+    (CONVERT_SAT((BASENAME##4), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 4 * STRIDE_Y + Z##4));
+
+#define CONVERT_STORE_ROW_6(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+    CONVERT_STORE_ROW_5(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
+    VSTORE(N0)                                                         \
+    (CONVERT_SAT((BASENAME##5), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 5 * STRIDE_Y + Z##5));
+
+#define CONVERT_STORE_ROW_7(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+    CONVERT_STORE_ROW_6(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
+    VSTORE(N0)                                                         \
+    (CONVERT_SAT((BASENAME##6), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 6 * STRIDE_Y + Z##6));
+
+#define CONVERT_STORE_ROW_8(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+    CONVERT_STORE_ROW_7(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
+    VSTORE(N0)                                                         \
+    (CONVERT_SAT((BASENAME##7), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 7 * STRIDE_Y + Z##7));
+
+#define CONVERT_STORE_ROW_9(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+    CONVERT_STORE_ROW_8(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
+    VSTORE(N0)                                                         \
+    (CONVERT_SAT((BASENAME##8), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 8 * STRIDE_Y + Z##8));
+
+#define CONVERT_STORE_ROW_10(N0, DATA, BASENAME, PTR, STRIDE_Y, Z) \
+    CONVERT_STORE_ROW_9(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+    VSTORE(N0)                                                     \
+    (CONVERT_SAT((BASENAME##9), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 9 * STRIDE_Y + Z##9));
+
+#define CONVERT_STORE_ROW_11(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+    CONVERT_STORE_ROW_10(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
+    VSTORE(N0)                                                          \
+    (CONVERT_SAT((BASENAME##A), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 10 * STRIDE_Y + Z##A));
+
+#define CONVERT_STORE_ROW_12(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+    CONVERT_STORE_ROW_11(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
+    VSTORE(N0)                                                          \
+    (CONVERT_SAT((BASENAME##B), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 11 * STRIDE_Y + Z##B));
+
+#define CONVERT_STORE_ROW_13(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+    CONVERT_STORE_ROW_12(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
+    VSTORE(N0)                                                          \
+    (CONVERT_SAT((BASENAME##C), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 12 * STRIDE_Y + Z##C));
+
+#define CONVERT_STORE_ROW_14(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+    CONVERT_STORE_ROW_13(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
+    VSTORE(N0)                                                          \
+    (CONVERT_SAT((BASENAME##D), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 13 * STRIDE_Y + Z##D));
+
+#define CONVERT_STORE_ROW_15(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+    CONVERT_STORE_ROW_14(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
+    VSTORE(N0)                                                          \
+    (CONVERT_SAT((BASENAME##E), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 14 * STRIDE_Y + Z##E));
+
+#define CONVERT_STORE_ROW_16(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+    CONVERT_STORE_ROW_15(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
+    VSTORE(N0)                                                          \
+    (CONVERT_SAT((BASENAME##F), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 15 * STRIDE_Y + Z##F));
+
+/** @} */ // end of groupd CONVERT_STORE_ROW_n
+
+/** Store a block of the given size M0xN0
+ * @name STORE_BLOCK
+ *
+ * Supported cases are M0=1,2,3,...,16 and N0=2,3,4,8,16.
+ * The data to store is expected to have consecutive names for each row.
+ * E.g., for M0=3 and basename=c, the expected names are c0, c1 and c2.
+ * The Z offset is expected to have consecutive names.
+ * E.g., for M0=3 and Z=zin, the expected z offset names are zin0, zin1 and zin2.
+ *
+ * @param[in] M0        The number of rows to store
+ * @param[in] N0        The size of each vector
+ * @param[in] DATA_TYPE The data type of the vectors
+ * @param[in] BASENAME  The basename of the variables
+ * @param[in] PTR       The base pointer
+ * @param[in] STRIDE_Y  The stride value in y-axis direction
+ * @param[in] Z         The offset in z-axis direction
+ * @{
+ */
+#define STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) STORE_ROW_##M0(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
+#define STORE_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
+/** @} */ // end of group STORE_BLOCK
+
+/** Convert and store a block of the given size M0xN0
+ * @name CONVERT_STORE_BLOCK
+ *
+ * Supported cases are M0=1,2,3,...,16 and N0=2,3,4,8,16.
+ * The data to store is expected to have consecutive names for each row.
+ * E.g., for M0=3 and basename=c, the expected names are c0, c1 and c2.
+ * The Z offset is expected to have consecutive names.
+ * E.g., for M0=3 and Z=zin, the expected z offset names are zin0, zin1 and zin2.
+ *
+ * @param[in] M0        The number of rows to store
+ * @param[in] N0        The size of each vector
+ * @param[in] DATA_TYPE The data type of the vectors
+ * @param[in] BASENAME  The basename of the variables
+ * @param[in] PTR       The base pointer
+ * @param[in] STRIDE_Y  The stride value in y-axis direction
+ * @param[in] Z         The offset in z-axis direction
+ * @{
+ */
+#define CONVERT_STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) CONVERT_STORE_ROW_##M0(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
+#define CONVERT_STORE_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) CONVERT_STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
+/** @} */ // end of group CONVERT_STORE_BLOCK
+
 /** Partially store the 0 to (n-1)th rows of the given variables
  * @name STORE_ROW_PARTIAL_n
  * Within each row, store the lower @p STORE_N0 elements of vectors of width @p N0
@@ -157,26 +385,25 @@
  * @param[in] Z                The offset in z-axis direction
  * @param[in] PARTIAL_STORE_M0 The partial size in y, for partial blocks. Supported range: [1, @p M0)
  * @param[in] PARTIAL_STORE_N0 The partial size in x, for partial blocks. Supported range: [1, @p N0)
- * @param[in] N                Total number of columns. Used to detect if current block is at the boundary in x.
  * @param[in] PARTIAL_COND_Y   Condition on the y axis to perform the partial store Y. True to use PARTIAL_STORE_M0 rather than M0.
  * @param[in] PARTIAL_COND_X   Condition on the x axis to perform the partial store X. True to use PARTIAL_STORE_N0 rather than N0.
  */
-#define STORE_BLOCK_PARTIAL_IN_X_AND_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, N, PARTIAL_COND_Y, PARTIAL_COND_X) \
-    if(!(PARTIAL_COND_X) && !(PARTIAL_COND_Y))                                                                                                               \
-    {                                                                                                                                                        \
-        STORE_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                                                              \
-    }                                                                                                                                                        \
-    else if((PARTIAL_COND_Y) && !(PARTIAL_COND_X))                                                                                                           \
-    {                                                                                                                                                        \
-        STORE_BLOCK_PARTIAL(PARTIAL_STORE_M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                                                \
-    }                                                                                                                                                        \
-    else if(!(PARTIAL_COND_Y) && (PARTIAL_COND_X))                                                                                                           \
-    {                                                                                                                                                        \
-        STORE_BLOCK_PARTIAL(M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                                                \
-    }                                                                                                                                                        \
-    else                                                                                                                                                     \
-    {                                                                                                                                                        \
-        STORE_BLOCK_PARTIAL(PARTIAL_STORE_M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                                  \
+#define STORE_BLOCK_PARTIAL_IN_X_AND_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
+    if(!(PARTIAL_COND_X) && !(PARTIAL_COND_Y))                                                                                                            \
+    {                                                                                                                                                     \
+        STORE_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                                                           \
+    }                                                                                                                                                     \
+    else if((PARTIAL_COND_Y) && !(PARTIAL_COND_X))                                                                                                        \
+    {                                                                                                                                                     \
+        STORE_BLOCK_PARTIAL(PARTIAL_STORE_M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                                             \
+    }                                                                                                                                                     \
+    else if(!(PARTIAL_COND_Y) && (PARTIAL_COND_X))                                                                                                        \
+    {                                                                                                                                                     \
+        STORE_BLOCK_PARTIAL(M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                                             \
+    }                                                                                                                                                     \
+    else                                                                                                                                                  \
+    {                                                                                                                                                     \
+        STORE_BLOCK_PARTIAL(PARTIAL_STORE_M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                               \
     }
 /** Store a block that can only be partial in x but not y.
  *
@@ -195,17 +422,16 @@
  * @param[in] STRIDE_Y         The stride value in y-axis direction
  * @param[in] Z                The offset in z-axis direction
  * @param[in] PARTIAL_STORE_N0 The partial size in x, for partial blocks. Supported range: [1, @p N0)
- * @param[in] N                Total number of columns. Used to detect if current block is at the boundary in x.
  * @param[in] PARTIAL_COND_X   Condition on the x axis to perform the partial store X. True to use PARTIAL_STORE_N0 rather than N0.
  */
-#define STORE_BLOCK_PARTIAL_IN_X(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_N0, N, PARTIAL_COND_X) \
-    if(!(PARTIAL_COND_X))                                                                                            \
-    {                                                                                                                \
-        STORE_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                      \
-    }                                                                                                                \
-    else                                                                                                             \
-    {                                                                                                                \
-        STORE_BLOCK_PARTIAL(M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                        \
+#define STORE_BLOCK_PARTIAL_IN_X(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_N0, PARTIAL_COND_X) \
+    if(!(PARTIAL_COND_X))                                                                                         \
+    {                                                                                                             \
+        STORE_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                   \
+    }                                                                                                             \
+    else                                                                                                          \
+    {                                                                                                             \
+        STORE_BLOCK_PARTIAL(M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                     \
     }
 /** Store a block that can only be partial in y but not x.
  *
@@ -286,36 +512,35 @@
  * @param[in] Z                The offset in z-axis direction
  * @param[in] PARTIAL_STORE_M0 The partial size in y, for partial blocks. Supported: [0, @p M0)
  * @param[in] PARTIAL_STORE_N0 The partial size in x, for partial blocks. Supported: [0, @p N0)
- * @param[in] N                Total number of columns. Used to detect if current block is at the boundary in x.
  * @param[in] PARTIAL_COND_Y   Condition on the y axis to perform the partial store Y. True to use PARTIAL_STORE_M0 rather than M0.
  * @param[in] PARTIAL_COND_X   Condition on the x axis to perform the partial store X. True to use PARTIAL_STORE_N0 rather than N0.
  * @{
  */
 #if PARTIAL_STORE_M0 == 0 && PARTIAL_STORE_N0 == 0
 // Case1: No partial blocks in either x or y
-#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, N, PARTIAL_COND_Y, PARTIAL_COND_X) \
+#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
     STORE_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
 
 #elif PARTIAL_STORE_M0 > 0 && PARTIAL_STORE_N0 == 0
 // Case2: Partial blocks in y
-#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, N, PARTIAL_COND_Y, PARTIAL_COND_X) \
+#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
     STORE_BLOCK_PARTIAL_IN_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_COND_Y)
 
 #elif PARTIAL_STORE_M0 == 0 && PARTIAL_STORE_N0 > 0
 // Case3: Partial blocks in x
-#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, N, PARTIAL_COND_Y, PARTIAL_COND_X) \
-    STORE_BLOCK_PARTIAL_IN_X(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_N0, N, PARTIAL_COND_X)
+#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
+    STORE_BLOCK_PARTIAL_IN_X(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_N0, PARTIAL_COND_X)
 
 #else // PARTIAL_STORE_M0 == 0 && PARTIAL_STORE_N0 == 0
 // Case4: Partial blocks in both x and y
-#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, N, PARTIAL_COND_Y, PARTIAL_COND_X) \
-    STORE_BLOCK_PARTIAL_IN_X_AND_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, N, PARTIAL_COND_Y, PARTIAL_COND_X)
+#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
+    STORE_BLOCK_PARTIAL_IN_X_AND_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X)
 
 #endif // PARTIAL_STORE_M0 == 0 && PARTIAL_STORE_N0 == 0
 
 #else // defined(PARTIAL_STORE_M0) && defined(PARTIAL_STORE_N0)
 
-#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, N, PARTIAL_COND_Y, PARTIAL_COND_X) \
+#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
     STORE_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
 
 #endif    // defined(PARTIAL_STORE_M0) && defined(PARTIAL_STORE_N0)
@@ -362,5 +587,5 @@
  * @{
  */
 #define STORE_VECTOR_SELECT(basename, data_type, ptr, vec_size, leftover, cond) \
-    STORE_BLOCK_PARTIAL_IN_X(1, vec_size, data_type, basename, ptr, 0, 0, leftover, 0, cond)
+    STORE_BLOCK_PARTIAL_IN_X(1, vec_size, data_type, basename, ptr, 0, 0, leftover, cond)
 /** @} */ // end of group STORE_VECTOR_SELECT
\ No newline at end of file
diff --git a/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.cpp b/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.cpp
index 5a0d2d0..876ef1e 100644
--- a/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.cpp
+++ b/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.cpp
@@ -124,37 +124,33 @@
                                                         const PadStrideInfo &conv_info, unsigned int depth_multiplier, const Size2D &dilation,
                                                         ITensorInfo *output_multipliers, ITensorInfo *output_shifts)
 {
-    const size_t weights_width  = 3;
-    const size_t weights_height = 3;
+    ARM_COMPUTE_UNUSED(weights);
+    ARM_COMPUTE_UNUSED(depth_multiplier);
 
-    // Get convolved dimensions
-    const TensorShape output_shape = arm_compute::misc::shape_calculator::compute_depthwise_convolution_shape(
-                                         *input, TensorInfo(TensorShape(weights_width, weights_height), 1, weights->data_type()).set_data_layout(DataLayout::NCHW), conv_info, depth_multiplier, dilation);
+    const bool   is_stride_1_dilation_1           = ((conv_info.stride().first == conv_info.stride().second) && (conv_info.stride().first == 1) && dilation.x() == 1 && dilation.y() == 1);
+    unsigned int num_rows_processed_per_iteration = is_stride_1_dilation_1 ? 2 : 1;
 
-    auto_init_if_empty(*output, input->clone()->set_tensor_shape(output_shape).set_quantization_info(output->quantization_info()));
+    Window win{};
+    Status err{};
 
-    const bool is_qasymm              = is_data_type_quantized_asymmetric(input->data_type());
-    const bool is_stride_1_dilation_1 = ((conv_info.stride().first == conv_info.stride().second) && (conv_info.stride().first == 1) && dilation.x() == 1 && dilation.y() == 1);
-
-    const unsigned int num_rows_processed_per_iteration = is_stride_1_dilation_1 ? 2 : 1;
-    const unsigned int num_elems_accessed_per_iteration = is_qasymm ? 4 : (8 / input->element_size());
-    const unsigned int num_rows_read_per_iteration      = num_rows_processed_per_iteration + 2;
-    const unsigned int num_rows_written_per_iteration   = std::ceil(num_rows_processed_per_iteration / static_cast<float>(conv_info.stride().first));
-
-    BorderSize border_size;
-    border_size = BorderSize(conv_info.pad_left(), 0, std::max(std::max(conv_info.pad_right(), conv_info.pad_bottom()), conv_info.pad_top()), 0);
-
-    // Configure kernel window
-    Window win = calculate_max_window(*output, Steps(num_elems_accessed_per_iteration, num_rows_written_per_iteration));
-
-    AccessWindowStatic input_access(input, 0, -border_size.top, ceil_to_multiple(input->dimension(0), num_elems_accessed_per_iteration),
-                                    ceil_to_multiple(input->dimension(1) + border_size.bottom, num_rows_read_per_iteration));
-    AccessWindowRectangle output_access(output, 0, 0, num_elems_accessed_per_iteration, num_rows_written_per_iteration);
-
-    bool window_changed = false;
-
-    if(is_qasymm)
+    if(is_data_type_quantized_asymmetric(input->data_type()))
     {
+        const unsigned int num_elems_accessed_per_iteration = 4;
+        const unsigned int num_rows_read_per_iteration      = num_rows_processed_per_iteration + 2;
+        const unsigned int num_rows_written_per_iteration   = std::ceil(num_rows_processed_per_iteration / static_cast<float>(conv_info.stride().first));
+
+        BorderSize border_size;
+        border_size = BorderSize(conv_info.pad_left(), 0, std::max(std::max(conv_info.pad_right(), conv_info.pad_bottom()), conv_info.pad_top()), 0);
+
+        // Configure kernel window
+        win = calculate_max_window(*output, Steps(num_elems_accessed_per_iteration, num_rows_written_per_iteration));
+
+        AccessWindowStatic input_access(input, 0, -border_size.top, ceil_to_multiple(input->dimension(0), num_elems_accessed_per_iteration),
+                                        ceil_to_multiple(input->dimension(1) + border_size.bottom, num_rows_read_per_iteration));
+        AccessWindowRectangle output_access(output, 0, 0, num_elems_accessed_per_iteration, num_rows_written_per_iteration);
+
+        bool window_changed = false;
+
         if((output_multipliers != nullptr) && (output_shifts != nullptr))
         {
             AccessWindowHorizontal output_multipliers_access(output_multipliers, 0, num_elems_accessed_per_iteration);
@@ -166,27 +162,28 @@
             Status err = ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "output_multipliers and output_shifts must be non-nullptr for quantized input");
             return std::make_pair(err, win);
         }
+
+        if(bias != nullptr)
+        {
+            AccessWindowHorizontal bias_access(bias, 0, num_elems_accessed_per_iteration);
+            window_changed = window_changed || update_window_and_padding(win, bias_access);
+        }
+        output_access.set_valid_region(win, ValidRegion(Coordinates(), output->tensor_shape()));
+
+        err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
     }
     else
     {
-        AccessWindowStatic weights_access(weights, 0, 0, ceil_to_multiple(weights->dimension(0), num_elems_accessed_per_iteration), weights->dimension(1));
-        window_changed = update_window_and_padding(win, input_access, weights_access, output_access);
+        unsigned int num_elems_accessed_per_iteration = adjust_vec_size(4 / input->element_size(), input->dimension(0));
+        win                                           = calculate_max_window(*output, Steps(num_elems_accessed_per_iteration, num_rows_processed_per_iteration));
     }
 
-    if(bias != nullptr)
-    {
-        AccessWindowHorizontal bias_access(bias, 0, num_elems_accessed_per_iteration);
-        window_changed = window_changed || update_window_and_padding(win, bias_access);
-    }
-    output_access.set_valid_region(win, ValidRegion(Coordinates(), output->tensor_shape()));
-
-    Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
     return std::make_pair(err, win);
 }
 } // namespace
 
 CLDepthwiseConvolutionLayer3x3NHWCKernel::CLDepthwiseConvolutionLayer3x3NHWCKernel()
-    : _num_rows_processed_per_iteration(1), _num_planes_processed_per_iteration(1)
+    : _num_planes_processed_per_iteration(1)
 {
 }
 
@@ -211,15 +208,16 @@
                                                   conv_info, depth_multiplier, act_info, dilation,
                                                   (output_multipliers != nullptr) ? output_multipliers->info() : nullptr,
                                                   (output_shifts != nullptr) ? output_shifts->info() : nullptr));
+
+    auto padding_info = get_padding_info({ input, weights, biases, output });
+
     auto win_config = validate_and_configure_window(input->info(), weights->info(), biases != nullptr ? biases->info() : nullptr, output->info(),
                                                     conv_info, depth_multiplier, dilation,
                                                     (output_multipliers != nullptr) ? output_multipliers->info() : nullptr,
                                                     (output_shifts != nullptr) ? output_shifts->info() : nullptr);
-    ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
 
-    const bool is_stride_1            = ((conv_info.stride().first == conv_info.stride().second) && (conv_info.stride().first == 1));
-    const bool is_stride_1_dilation_1 = (is_stride_1 && dilation.x() == 1 && dilation.y() == 1);
-
+    const bool is_stride_1              = ((conv_info.stride().first == conv_info.stride().second) && (conv_info.stride().first == 1));
+    const bool is_stride_1_dilation_1   = (is_stride_1 && dilation.x() == 1 && dilation.y() == 1);
     const bool is_quantized_per_channel = is_data_type_quantized_per_channel(weights->info()->data_type());
     const bool is_dot8_supported        = dot8_supported(CLKernelLibrary::get().get_device()) && !is_quantized_per_channel;
 
@@ -228,31 +226,37 @@
     _weights                            = weights;
     _biases                             = biases;
     _conv_stride_y                      = conv_info.stride().second;
-    _num_rows_processed_per_iteration   = is_stride_1_dilation_1 ? 2 : 1;
     _num_planes_processed_per_iteration = is_stride_1_dilation_1 ? 2 : 1;
     _output_multipliers                 = output_multipliers;
     _output_shifts                      = output_shifts;
     _is_quantized                       = is_data_type_quantized_asymmetric(input->info()->data_type());
 
-    // If QASYMM8 and the 8 bit dot product is available, force _num_planes_processed_per_iteration to 1
-    if(is_dot8_supported && _is_quantized)
+    if(_is_quantized)
     {
-        _num_planes_processed_per_iteration = 1;
+        _border_size = BorderSize(is_stride_1 ? 0 : conv_info.pad_left(), 0, std::max(std::max(conv_info.pad_right(), conv_info.pad_bottom()), conv_info.pad_top()), 0);
+
+        // If QASYMM8 and the 8 bit dot product is available, force _num_planes_processed_per_iteration to 1
+        if(is_dot8_supported)
+        {
+            _num_planes_processed_per_iteration = 1;
+        }
     }
 
-    _border_size = BorderSize(_is_quantized && is_stride_1 ? 0 : conv_info.pad_left(), 0, std::max(std::max(conv_info.pad_right(), conv_info.pad_bottom()), conv_info.pad_top()), 0);
-
-    const unsigned int num_elems_accessed_per_iteration = _is_quantized ? 4 : (8 / input->info()->element_size());
+    unsigned int num_elems_accessed_per_iteration = _is_quantized ? 4 : adjust_vec_size(4 / input->info()->element_size(), input->info()->dimension(0));
+    unsigned int num_rows_processed_per_iteration = is_stride_1_dilation_1 ? 2 : 1;
 
     CLBuildOptions build_opts;
+    build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(_input->info()->data_type()));
     build_opts.add_option("-DACTIVATION_TYPE=" + lower_string(string_from_activation_func(act_info.activation())));
-    build_opts.add_option_if(_biases != nullptr, "-DHAS_BIAS");
     build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(num_elems_accessed_per_iteration));
+    build_opts.add_option("-DSRC_DIM_1=" + support::cpp11::to_string(_input->info()->dimension(1)));
     build_opts.add_option("-DSRC_DIM_2=" + support::cpp11::to_string(_input->info()->dimension(2)));
     build_opts.add_option("-DCONV_PAD_TOP=" + support::cpp11::to_string(conv_info.pad_top()));
     build_opts.add_option("-DCONV_PAD_LEFT=" + support::cpp11::to_string(conv_info.pad_left()));
-    build_opts.add_option("-DDILATION_X=" + support::cpp11::to_string(dilation.x()));
-    build_opts.add_option("-DDILATION_Y=" + support::cpp11::to_string(dilation.y()));
+    build_opts.add_option("-DPARTIAL_STORE_N0=" + support::cpp11::to_string(input->info()->dimension(0) % num_elems_accessed_per_iteration));
+    build_opts.add_option_if(_biases != nullptr, "-DHAS_BIAS");
+    build_opts.add_option_if(_input->info()->tensor_shape().total_size_upper(3) > 1,
+                             "-DDST_DEPTH=" + support::cpp11::to_string(static_cast<int>(std::ceil(_output->info()->dimension(2) / static_cast<float>(_num_planes_processed_per_iteration)))));
 
     if(_is_quantized)
     {
@@ -291,7 +295,6 @@
             build_opts.add_option("-DO1_VAL=" + support::cpp11::to_string(o1));
         }
 
-        build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type()));
         build_opts.add_option("-DWEIGHTS_TYPE=" + get_cl_type_from_data_type(weights->info()->data_type()));
         build_opts.add_option("-DWEIGHTS_PROMOTED_TYPE=" + get_cl_promoted_type_from_data_type(weights->info()->data_type()));
     }
@@ -299,22 +302,23 @@
     {
         build_opts.add_option_if(act_info.enabled(), "-DA_VAL=" + float_to_string_with_full_precision(act_info.a()));
         build_opts.add_option_if(act_info.enabled(), "-DB_VAL=" + float_to_string_with_full_precision(act_info.b()));
-        build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(_input->info()->data_type()));
     }
 
     if(is_stride_1_dilation_1)
     {
-        build_opts.add_option("-DNUM_ROWS_PROCESSED=" + support::cpp11::to_string(_num_rows_processed_per_iteration));
+        build_opts.add_option("-DNUM_ROWS_PROCESSED=" + support::cpp11::to_string(num_rows_processed_per_iteration));
         build_opts.add_option("-DNUM_PLANES_PROCESSED=" + support::cpp11::to_string(_num_planes_processed_per_iteration));
+        build_opts.add_option("-DDST_DIM_1=" + support::cpp11::to_string(_output->info()->dimension(1)));
         build_opts.add_option("-DDST_DIM_2=" + support::cpp11::to_string(_output->info()->dimension(2)));
+        build_opts.add_option("-DPARTIAL_STORE_M0=" + support::cpp11::to_string((input->info()->dimension(1) + conv_info.pad_left() + conv_info.pad_right()) % num_rows_processed_per_iteration));
     }
     else
     {
         build_opts.add_option("-DCONV_STRIDE_X=" + support::cpp11::to_string(conv_info.stride().first));
         build_opts.add_option("-DCONV_STRIDE_Y=" + support::cpp11::to_string(_conv_stride_y));
+        build_opts.add_option("-DDILATION_X=" + support::cpp11::to_string(dilation.x()));
+        build_opts.add_option("-DDILATION_Y=" + support::cpp11::to_string(dilation.y()));
     }
-    build_opts.add_option_if(_input->info()->tensor_shape().total_size_upper(3) > 1,
-                             "-DDST_DEPTH=" + support::cpp11::to_string(static_cast<int>(std::ceil(_output->info()->dimension(2) / static_cast<float>(_num_planes_processed_per_iteration)))));
 
     std::string kernel_name;
     // Create kernel
@@ -331,12 +335,11 @@
         kernel_name += (is_stride_1_dilation_1 ? "_stride1" : "");
     }
 
-    build_opts.add_option_if(input->info()->data_type() == DataType::F16, "-DIS_F16");
-    build_opts.add_option_if(input->info()->data_type() == DataType::F32, "-DIS_F32");
-
     ICLKernel::configure_internal(win_config.second);
     _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
 
+    ARM_COMPUTE_ERROR_ON(!_is_quantized && has_padding_changed(padding_info));
+
     // Set config_id for enabling LWS tuning
     _config_id = kernel_name;
     _config_id += "_";
@@ -364,7 +367,6 @@
                                                               (output_multipliers != nullptr) ? output_multipliers->clone().get() : nullptr,
                                                               (output_shifts != nullptr) ? output_shifts->clone().get() : nullptr)
                                 .first);
-
     return Status{};
 }
 
@@ -373,23 +375,11 @@
     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
     ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(IKernel::window(), window);
 
-    // Collapse window
-    Window       window_collapsed = window.collapse_if_possible(ICLKernel::window(), Window::DimZ);
-    const size_t total_batches    = _input->info()->tensor_shape().total_size_upper(3);
+    const size_t total_batches = _input->info()->tensor_shape().total_size_upper(3);
 
-    Window win = window_collapsed;
+    Window win = window.collapse_if_possible(ICLKernel::window(), Window::DimZ);
     win.set(Window::DimZ, Window::Dimension(0, std::ceil(_output->info()->dimension(2) / static_cast<float>(_num_planes_processed_per_iteration)) * total_batches, 1));
 
-    // Create input window and adjust
-    Window win_in = win;
-    win_in.set_dimension_step(Window::DimY, _num_rows_processed_per_iteration);
-    win_in.set_dimension_step(Window::DimZ, _conv_stride_y);
-
-    ARM_COMPUTE_ERROR_ON((win_in.y().step() < window.y().step()) || (win_in.z().step() < window.z().step()));
-
-    Window slice_in  = win_in.first_slice_window_4D();
-    Window slice_out = win.first_slice_window_4D();
-
     unsigned int idx = 2 * num_arguments_per_4D_tensor() + (_is_quantized ? num_arguments_per_2D_tensor() : num_arguments_per_3D_tensor());
 
     if(_is_quantized)
@@ -409,60 +399,64 @@
         add_1D_tensor_argument(idx, _biases, win_biases);
     }
 
-    // Calculate the max_offset.
-    // max_offset is the offset for the last NOT valid value in the Z dimension (spatial dimension Y for NHWC)
-    //  |******************|
-    //  |     pad_top      |
-    //  |******************|
-    //  |                  |
-    //  |      plane0      |
-    //  |      batch0      |
-    //  |__________________|
-    //  |******************|       Batch 0
-    //  |    pad_bottom    |
-    //  |     pad_top      |
-    //  |******************|
-    //  |                  |
-    //  |      plane1      |
-    //  |      batch0      |
-    //  |__________________|-----> max_offset
-    //  |******************|
-    //  |    pad_bottom    |
-    //  |     pad_top      |
-    //  |******************|
-    //  |                  |
-    //  |      plane0      |
-    //  |      batch1      |
-    //  |__________________|
-    //  |******************|       Batch 1
-    //  |    pad_bottom    |
-    //  |     pad_top      |
-    //  |******************|
-    //  |                  |
-    //  |      plane1      |
-    //  |      batch1      |
-    //  |__________________|
-    //  |     pad_bottom   |
-    //  |******************|
-    const int max_offset = _input->info()->strides_in_bytes().z() * _input->info()->dimension(2) - (_input->info()->padding().bottom + _input->info()->padding().top) *
-                           _input->info()->strides_in_bytes().y();
-    _kernel.setArg(idx, max_offset);
+    if(_is_quantized)
+    {
+        // Calculate the max_offset.
+        // max_offset is the offset for the last NOT valid value in the Z dimension (spatial dimension Y for NHWC)
+        //  |******************|
+        //  |     pad_top      |
+        //  |******************|
+        //  |                  |
+        //  |      plane0      |
+        //  |      batch0      |
+        //  |__________________|
+        //  |******************|       Batch 0
+        //  |    pad_bottom    |
+        //  |     pad_top      |
+        //  |******************|
+        //  |                  |
+        //  |      plane1      |
+        //  |      batch0      |
+        //  |__________________|-----> max_offset
+        //  |******************|
+        //  |    pad_bottom    |
+        //  |     pad_top      |
+        //  |******************|
+        //  |                  |
+        //  |      plane0      |
+        //  |      batch1      |
+        //  |__________________|
+        //  |******************|       Batch 1
+        //  |    pad_bottom    |
+        //  |     pad_top      |
+        //  |******************|
+        //  |                  |
+        //  |      plane1      |
+        //  |      batch1      |
+        //  |__________________|
+        //  |     pad_bottom   |
+        //  |******************|
+        const int max_offset = _input->info()->strides_in_bytes().z() * _input->info()->dimension(2) - (_input->info()->padding().bottom + _input->info()->padding().top) *
+                               _input->info()->strides_in_bytes().y();
+        _kernel.setArg(idx, max_offset);
+    }
 
+    Window slice = win.first_slice_window_4D();
     do
     {
         unsigned int idx = 0;
-        add_4D_tensor_argument(idx, _input, slice_in);
-        add_4D_tensor_argument(idx, _output, slice_out);
+        add_4D_tensor_argument(idx, _input, slice);
+        add_4D_tensor_argument(idx, _output, slice);
         if(_is_quantized)
         {
-            add_2D_tensor_argument(idx, _weights, slice_out);
+            add_2D_tensor_argument(idx, _weights, slice);
         }
         else
         {
-            add_3D_tensor_argument(idx, _weights, slice_out);
+            add_3D_tensor_argument(idx, _weights, slice);
         }
-        enqueue(queue, *this, slice_out, lws_hint());
+        enqueue(queue, *this, slice, lws_hint());
     }
-    while(win.slide_window_slice_4D(slice_out) && win_in.slide_window_slice_4D(slice_in));
+    while(win.slide_window_slice_4D(slice));
 }
 } // namespace arm_compute