Add quantized CL MatMul kernels for Lhs NT/T, Rhs NT

Implement OpenCL kernels for batched Matrix Multiplication for the quantized data types QASYMM8 and QASYMM8_SIGNED.

Quantized MatMul is supported with the following MatMul attributes:
* adj_x = false, adj_y = false
* adj_x = true, adj_y = false

We consider native format kernels only. In other words, no reshaping of the operand matrices is done.

Resolves: COMPMID-5921, COMPMID-5922

Change-Id: I99e0f68054a2bd635c60ec2641acc2e7ff398473
Signed-off-by: Omar Al Khatib <omar.alkhatib@arm.com>
Signed-off-by: Gunes Bayir <gunes.bayir@arm.com>
Signed-off-by: Jakub Sujak <jakub.sujak@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9435
Reviewed-by: SiCong Li <sicong.li@arm.com>
Reviewed-by: Viet-Hoa Do <viet-hoa.do@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>

13 files changed, 1486 insertions, 171 deletions

diff --git a/Android.bp b/Android.bp
index 4bd307447b..32651b539c 100644
--- a/Android.bp
+++ b/Android.bp
@@ -51,6 +51,7 @@ opencl_srcs = [
         "src/core/CL/cl_kernels/common/instance_normalization.cl",
         "src/core/CL/cl_kernels/common/l2_normalize.cl",
         "src/core/CL/cl_kernels/common/mat_mul.cl",
+        "src/core/CL/cl_kernels/common/mat_mul_quantized.cl",
         "src/core/CL/cl_kernels/common/mean_stddev_normalization.cl",
         "src/core/CL/cl_kernels/common/memset.cl",
         "src/core/CL/cl_kernels/common/minmax_layer.cl",
@@ -695,6 +696,7 @@ cc_library_static {
         "src/gpu/cl/kernels/ClIm2ColKernel.cpp",
         "src/gpu/cl/kernels/ClIndirectConv2dAddressPrecalculationKernel.cpp",
         "src/gpu/cl/kernels/ClIndirectConv2dKernel.cpp",
+        "src/gpu/cl/kernels/ClMatMulLowpNativeKernel.cpp",
         "src/gpu/cl/kernels/ClMatMulNativeKernel.cpp",
         "src/gpu/cl/kernels/ClMulKernel.cpp",
         "src/gpu/cl/kernels/ClPermuteKernel.cpp",
diff --git a/SConscript b/SConscript
index 03b94b6bd1..e6ef73cc34 100644
--- a/SConscript
+++ b/SConscript
@@ -390,6 +390,7 @@ if env['opencl'] and env['embed_kernels']:
                        'src/core/CL/cl_kernels/common/instance_normalization.cl',
                        'src/core/CL/cl_kernels/common/l2_normalize.cl',
                        'src/core/CL/cl_kernels/common/mat_mul.cl',
+                       'src/core/CL/cl_kernels/common/mat_mul_quantized.cl',
                        'src/core/CL/cl_kernels/common/mean_stddev_normalization.cl',
                        'src/core/CL/cl_kernels/common/memset.cl',
                        'src/core/CL/cl_kernels/common/minmax_layer.cl',
diff --git a/filelist.json b/filelist.json
index 5418c2bfd0..ed12dca8b3 100644
--- a/filelist.json
+++ b/filelist.json
@@ -512,6 +512,7 @@
     "MatMul": {
       "files": {
         "common": [
+          "src/gpu/cl/kernels/ClMatMulLowpNativeKernel.cpp",
           "src/gpu/cl/kernels/ClMatMulNativeKernel.cpp",
           "src/gpu/cl/operators/ClMatMul.cpp",
           "src/runtime/CL/functions/CLMatMul.cpp"
diff --git a/src/core/CL/cl_kernels/common/mat_mul_quantized.cl b/src/core/CL/cl_kernels/common/mat_mul_quantized.cl
new file mode 100644
index 0000000000..c250b4b988
--- /dev/null
+++ b/src/core/CL/cl_kernels/common/mat_mul_quantized.cl
@@ -0,0 +1,387 @@
+/*
+ * Copyright (c) 2023 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "helpers.h"
+#include "tile_helpers.h"
+
+#if defined(MAT_MUL_NATIVE_QUANTIZED_NT_NT)
+/** This OpenCL kernel performs the batch matrix multiplication (BatchMatMul): LHS non-transposed, RHS non-transposed - buffer only
+ *
+ * @note the "batch" here expresses the number of matrix multiplications to run in parallel. However, it
+ *       should NOT be confused with the batch size of the model. For NHWC the "batch" is the "H" dimension
+ * @note The data type must be passed at compile time using -DDATA_TYPE (e.g. -DDATA_TYPE=uchar)
+ * @note The block's dimensions used for the LHS and RHS matrices (M0, N0 and K0) must be passed at compile time using -DN0, -DM0 and -DK0 (e.g. -DN0=8, -DM0=4, -DK0=4).
+ * @note The number of leftover outputs rows/columns must be passed using -DPARTIAL_STORE_N0 and -DPARTIAL_STORE_M0 (e.g. -DPARTIAL_STORE_N0=2, -DPARTIAL_STORE_M0=3)
+ * @note The dimension K must be passed at compile time using -DK (e.g. -DK=6)
+ * @note The kernel name in uppercase must be passed at compile time (e.g. -DMAT_MUL_NATIVE_QUANTIZED_NT_NT)
+ * @note Only the following configurations of M0, N0 and K0 are currently supported:
+ *  - M0 > 0
+ *  - N0 = 1, 2, 3, 4, 8, 16
+ *  - K0 = 1, 2, 3, 4, 8, 16
+ * @note Values > 8 for M0 are not expected to be efficient
+ *
+ * @param[in]  lhs_ptr                           Pointer to the lhs matrix. Supported data types: QASYMM8_SIGNED/QASYMM8
+ * @param[in]  lhs_stride_y                      Stride of the lhs matrix in Y (2nd) dimension (in bytes)
+ * @param[in]  lhs_stride_z                      Stride of the lhs tensor in Z (3rd) dimension (in bytes)
+ * @param[in]  lhs_w                             The width of the lhs tensor
+ * @param[in]  lhs_h                             The height of the lhs tensor
+ * @param[in]  lhs_n                             Number of the matrices (buffers) in the batch
+ * @param[in]  lhs_offset_first_element_in_bytes The offset of the first element in the lhs matrix
+ * @param[in]  rhs_ptr                           Pointer to the rhs matrix. Supported data types: same as @p lhs_ptr
+ * @param[in]  rhs_stride_y                      Stride of the rhs matrix in Y (2nd) dimension (in bytes)
+ * @param[in]  rhs_stride_z                      Stride of the rhs tensor in Z (3rd) dimension (in bytes)
+ * @param[in]  rhs_w                             The width of the rhs tensor
+ * @param[in]  rhs_h                             The height of the rhs tensor
+ * @param[in]  rhs_n                             Number of the matrices (buffers) in the batch
+ * @param[in]  rhs_offset_first_element_in_bytes The offset of the first element in the rhs matrix
+ * @param[out] dst_ptr                           Pointer to the dst matrix. Supported data types: same as @p lhs_ptr
+ * @param[in]  dst_stride_y                      Stride of the dst matrix in Y (2nd) dimension (in bytes)
+ * @param[in]  dst_stride_z                      Stride of the dst tensor in Z (3rd) dimension (in bytes)
+ * @param[in]  dst_w                             The width of the dst tensor
+ * @param[in]  dst_h                             The height of the dst tensor
+ * @param[in]  dst_n                             Number of the matrices (buffers) in the batch
+ * @param[in]  dst_offset_first_element_in_bytes The offset of the first element in the dst matrix
+ */
+__kernel void mat_mul_native_quantized_nt_nt(
+    TENSOR3D_T(lhs, BUFFER),
+    TENSOR3D_T(rhs, BUFFER),
+    TENSOR3D_T(dst, BUFFER))
+{
+    const uint x = GET_SPATIAL_IDX(0, N0, PARTIAL_STORE_N0);
+    const uint y = GET_SPATIAL_IDX(1, M0, PARTIAL_STORE_M0);
+    const uint z = GET_SPATIAL_IDX(2, 1, 0);
+
+    // Compute LHS/RHS/DST matrix address
+    lhs_offset_first_element_in_bytes += y * lhs_stride_y + z * lhs_stride_z;
+    rhs_offset_first_element_in_bytes += x * sizeof(DATA_TYPE) + z * rhs_stride_z;
+    dst_offset_first_element_in_bytes += x * sizeof(DATA_TYPE) + y * dst_stride_y + z * dst_stride_z;
+
+    // Initialize the accumulators
+    TILE(int, M0, N0, acc);
+    LOOP_UNROLLING(int, i, 0, 1, M0,
+    {
+        acc[i].v = K * ((int)LHS_OFFSET) * ((int)RHS_OFFSET);
+    })
+
+    TILE(int, 1, N0, b_sum);
+    b_sum[0].v = 0;
+
+    TILE(int, 1, M0, a_sum);
+    a_sum[0].v = 0;
+
+    int k;
+    for(k = 0; k <= K - K0; k += K0)
+    {
+        TILE(DATA_TYPE, M0, K0, a);
+        TILE(DATA_TYPE, N0, K0, b);
+
+        LOOP_UNROLLING(int, i, 0, 1, M0,
+        {
+            a[i].v = 0;
+        })
+
+        LOOP_UNROLLING(int, i, 0, 1, N0,
+        {
+            b[i].v = 0;
+        })
+
+        // Load tile from the lhs tensor
+        T_LOAD(DATA_TYPE, M0, K0, BUFFER, lhs, 0, 0, 1, lhs_stride_y, a);
+
+        // Load tile from the rhs tensor in a transposed fashion
+        // in order to use T_MMUL_NT_T macro because only this macro
+        // can utilize dot product instruction for Int8/UInt8 by
+        // directly multiplying the rows of Lhs and Rhs tensors.
+        T_LOAD_TRANSPOSED(DATA_TYPE, K0, N0, BUFFER, rhs, 0, 0, 1, rhs_stride_y, b);
+
+        T_MMUL(DATA_TYPE, DATA_TYPE, int, M0, N0, K0, NT, T, a, b, acc);
+
+        LOOP_UNROLLING(int, i, 0, 1, M0,
+        {
+            LOOP_UNROLLING(int, j, 0, 1, K0,
+            {
+                a_sum[0].s[i] += (int)a[i].s[j];
+            })
+        })
+
+        LOOP_UNROLLING(int, i, 0, 1, K0,
+        {
+            LOOP_UNROLLING(int, j, 0, 1, N0,
+            {
+                b_sum[0].s[j] += (int)b[j].s[i];
+            })
+        })
+
+        lhs_offset_first_element_in_bytes += K0 * sizeof(DATA_TYPE);
+        rhs_offset_first_element_in_bytes += K0 * rhs_stride_y;
+    }
+
+#if((K % K0) != 0)
+    /* Leftover Loop */
+    for(; k < K; ++k)
+    {
+        TILE(DATA_TYPE, M0, 1, a);
+        TILE(DATA_TYPE, N0, 1, b);
+
+        LOOP_UNROLLING(int, i, 0, 1, M0,
+        {
+            a[i].v = 0;
+        })
+
+        LOOP_UNROLLING(int, i, 0, 1, N0,
+        {
+            b[i].v = 0;
+        })
+
+        // Load tile from the lhs tensor
+        T_LOAD(DATA_TYPE, M0, 1, BUFFER, lhs, 0, 0, 1, lhs_stride_y, a);
+
+        // Load tile from the rhs tensor in a transposed fashion.
+        // See the main loop for more explanation
+        T_LOAD_TRANSPOSED(DATA_TYPE, 1, N0, BUFFER, rhs, 0, 0, 1, rhs_stride_y, b);
+
+        T_MMUL(DATA_TYPE, DATA_TYPE, int, M0, N0, 1, NT, T, a, b, acc);
+
+        LOOP_UNROLLING(int, i, 0, 1, M0,
+        {
+            LOOP_UNROLLING(int, j, 0, 1, 1,
+            {
+                a_sum[0].s[i] += (int)a[i].s[j];
+            })
+        })
+
+        LOOP_UNROLLING(int, i, 0, 1, 1,
+        {
+            LOOP_UNROLLING(int, j, 0, 1, N0,
+            {
+                b_sum[0].s[j] += (int)b[j].s[i];
+            })
+        })
+
+        lhs_offset_first_element_in_bytes += 1 * sizeof(DATA_TYPE);
+        rhs_offset_first_element_in_bytes += 1 * rhs_stride_y;
+    }
+#endif // ((K % K0) != 0)
+
+    LOOP_UNROLLING(int, i, 0, 1, M0,
+    {
+        LOOP_UNROLLING(int, j, 0, 1, N0,
+        {
+            acc[i].s[j] += ((int)RHS_OFFSET) * a_sum[0].s[i] + ((int)(LHS_OFFSET)) * b_sum[0].s[j];
+        })
+    })
+
+    const bool x_cond = PARTIAL_STORE_N0 != 0 && get_global_id(0) == 0;
+    const bool y_cond = PARTIAL_STORE_M0 != 0 && get_global_id(1) == 0;
+
+    // Quantize the tile
+    TILE(DATA_TYPE, M0, N0, accq);
+    T_QUANTIZE8_ASYMMETRIC(int, DATA_TYPE, M0, N0, DST_OFFSET, DST_SHIFT, DST_MULTIPLIER, acc, accq);
+
+    TILE(int, M0, 1, indirect_buffer);
+    LOOP_UNROLLING(int, _i, 0, 1, M0,
+    {
+        indirect_buffer[_i].v = min(_i, select(M0 - 1, PARTIAL_STORE_M0 - 1, y_cond));
+    });
+
+    T_STORE_INDIRECT_WIDTH_SELECT(DATA_TYPE, M0, N0, PARTIAL_STORE_N0, BUFFER, dst, 0, dst_stride_y, x_cond, accq, indirect_buffer);
+}
+#endif // defined(MAT_MUL_NATIVE_QUANTIZED_NT_NT)
+
+#if defined(MAT_MUL_NATIVE_QUANTIZED_T_NT)
+/** This OpenCL kernel performs the batch matrix multiplication (BatchMatMul): LHS transposed, RHS non-transposed
+ *
+ * @note the "batch" here expresses the number of matrix multiplications to run in parallel. However, it
+ *       should NOT be confused with the batch size of the model. For NHWC the "batch" is the "H" dimension
+ * @note The data type must be passed at compile time using -DDATA_TYPE (e.g. -DDATA_TYPE=uchar)
+ * @note The block's dimensions used for the LHS and RHS matrices (M0, N0 and K0) must be passed at compile time using -DN0, -DM0 and -DK0 (e.g. -DN0=8, -DM0=4, -DK0=4).
+ * @note The number of leftover outputs rows/columns must be passed using -DPARTIAL_STORE_N0 and -DPARTIAL_STORE_M0 (e.g. -DPARTIAL_STORE_N0=2, -DPARTIAL_STORE_M0=3)
+ * @note The dimension K must be passed at compile time using -DK (e.g. -DK=6)
+ * @note The kernel name in uppercase must be passed at compile time (e.g. -DMAT_MUL_NATIVE_QUANTIZED_T_NT)
+ * @note Only the following configurations of M0, N0 and K0 are currently supported:
+ *  - M0 > 0
+ *  - N0 = 1, 2, 3, 4, 8, 16
+ *  - K0 = 1, 2, 3, 4, 8, 16
+ * @note Values > 8 for M0, N0 and K0 are not expected to be efficient
+ *
+ * @param[in]  lhs_ptr                           Pointer to the lhs matrix. Supported data types: QASYMM8/QASYMM8_SIGNED
+ * @param[in]  lhs_stride_y                      Stride of the lhs matrix in Y (2nd) dimension (in bytes)
+ * @param[in]  lhs_stride_z                      Stride of the lhs tensor in Z (3rd) dimension (in bytes)
+ * @param[in]  lhs_w                             The width of the lhs tensor
+ * @param[in]  lhs_h                             The height of the lhs tensor
+ * @param[in]  lhs_n                             Number of the matrices (buffers) in the batch
+ * @param[in]  lhs_offset_first_element_in_bytes The offset of the first element in the lhs matrix
+ * @param[in]  rhs_ptr                           Pointer to the rhs matrix. Supported data types: same as @p lhs_ptr
+ * @param[in]  rhs_stride_y                      Stride of the rhs matrix in Y (2nd) dimension (in bytes)
+ * @param[in]  rhs_stride_z                      Stride of the rhs tensor in Z (3rd) dimension (in bytes)
+ * @param[in]  rhs_w                             The width of the rhs tensor
+ * @param[in]  rhs_h                             The height of the rhs tensor
+ * @param[in]  rhs_n                             Number of the matrices (buffers) in the batch
+ * @param[in]  rhs_offset_first_element_in_bytes The offset of the first element in the rhs matrix
+ * @param[out] dst_ptr                           Pointer to the dst matrix. Supported data types: same as @p lhs_ptr
+ * @param[in]  dst_stride_y                      Stride of the dst matrix in Y (2nd) dimension (in bytes)
+ * @param[in]  dst_stride_z                      Stride of the dst tensor in Z (3rd) dimension (in bytes)
+ * @param[in]  dst_w                             The width of the dst tensor
+ * @param[in]  dst_h                             The height of the dst tensor
+ * @param[in]  dst_n                             Number of the matrices (buffers) in the batch
+ * @param[in]  dst_offset_first_element_in_bytes The offset of the first element in the dst matrix
+ */
+__kernel void mat_mul_native_quantized_t_nt(
+    TENSOR3D_T(lhs, BUFFER),
+    TENSOR3D_T(rhs, BUFFER),
+    TENSOR3D_T(dst, BUFFER))
+{
+    const uint x = GET_SPATIAL_IDX(0, N0, PARTIAL_STORE_N0);
+    const uint y = GET_SPATIAL_IDX(1, M0, PARTIAL_STORE_M0);
+    const uint z = GET_SPATIAL_IDX(2, 1, 0);
+
+    // Compute LHS/RHS/DST matrix address
+    lhs_offset_first_element_in_bytes += y * sizeof(DATA_TYPE) + z * lhs_stride_z;
+    rhs_offset_first_element_in_bytes += x * sizeof(DATA_TYPE) + z * rhs_stride_z;
+    dst_offset_first_element_in_bytes += x * sizeof(DATA_TYPE) + y * dst_stride_y + z * dst_stride_z;
+
+    // Initialize the accumulators
+    TILE(int, M0, N0, acc);
+    LOOP_UNROLLING(int, i, 0, 1, M0,
+    {
+        acc[i].v = K * ((int)LHS_OFFSET) * ((int)RHS_OFFSET);
+    })
+
+    TILE(int, 1, N0, b_sum);
+    b_sum[0].v = 0;
+
+    TILE(int, 1, M0, a_sum);
+    a_sum[0].v = 0;
+
+    int k;
+    for(k = 0; k <= K - K0; k += K0)
+    {
+        TILE(DATA_TYPE, M0, K0, a);
+        TILE(DATA_TYPE, N0, K0, b);
+
+        LOOP_UNROLLING(int, i, 0, 1, M0,
+        {
+            a[i].v = 0;
+        })
+
+        LOOP_UNROLLING(int, i, 0, 1, N0,
+        {
+            b[i].v = 0;
+        })
+
+        // Load tile from the lhs/rhs tensors in a transposed fashion
+        // see mat_mul_native_quantized_nt_nt main loop for more explanation
+        T_LOAD_TRANSPOSED(DATA_TYPE, K0, M0, BUFFER, lhs, 0, 0, 1, lhs_stride_y, a);
+        T_LOAD_TRANSPOSED(DATA_TYPE, K0, N0, BUFFER, rhs, 0, 0, 1, rhs_stride_y, b);
+
+        T_MMUL(DATA_TYPE, DATA_TYPE, int, M0, N0, K0, NT, T, a, b, acc);
+
+        LOOP_UNROLLING(int, i, 0, 1, K0,
+        {
+            LOOP_UNROLLING(int, j, 0, 1, M0,
+            {
+                a_sum[0].s[j] += (int)a[j].s[i];
+            })
+        })
+
+        LOOP_UNROLLING(int, i, 0, 1, K0,
+        {
+            LOOP_UNROLLING(int, j, 0, 1, N0,
+            {
+                b_sum[0].s[j] += (int)b[j].s[i];
+            })
+        })
+
+        lhs_offset_first_element_in_bytes += K0 * lhs_stride_y;
+        rhs_offset_first_element_in_bytes += K0 * rhs_stride_y;
+    }
+
+#if((K % K0) != 0)
+    /* Leftover Loop */
+    for(; k < K; ++k)
+    {
+        TILE(DATA_TYPE, M0, 1, a);
+        TILE(DATA_TYPE, N0, 1, b);
+
+        LOOP_UNROLLING(int, i, 0, 1, M0,
+        {
+            a[i].v = 0;
+        })
+
+        LOOP_UNROLLING(int, i, 0, 1, N0,
+        {
+            b[i].v = 0;
+        })
+
+        // Load tile from the lhs/rhs tensors in a transposed fashion
+        // see mat_mul_native_quantized_nt_nt main loop for more explanation
+        T_LOAD_TRANSPOSED(DATA_TYPE, 1, M0, BUFFER, lhs, 0, 0, 1, lhs_stride_y, a);
+        T_LOAD_TRANSPOSED(DATA_TYPE, 1, N0, BUFFER, rhs, 0, 0, 1, rhs_stride_y, b);
+
+        T_MMUL(DATA_TYPE, DATA_TYPE, int, M0, N0, 1, NT, T, a, b, acc);
+
+        LOOP_UNROLLING(int, i, 0, 1, 1,
+        {
+            LOOP_UNROLLING(int, j, 0, 1, M0,
+            {
+                a_sum[0].s[j] += (int)a[j].s[i];
+            })
+        })
+
+        LOOP_UNROLLING(int, i, 0, 1, 1,
+        {
+            LOOP_UNROLLING(int, j, 0, 1, N0,
+            {
+                b_sum[0].s[j] += (int)b[j].s[i];
+            })
+        })
+
+        lhs_offset_first_element_in_bytes += 1 * lhs_stride_y;
+        rhs_offset_first_element_in_bytes += 1 * rhs_stride_y;
+    }
+#endif // ((K % K0) != 0)
+
+    LOOP_UNROLLING(int, i, 0, 1, M0,
+    {
+        LOOP_UNROLLING(int, j, 0, 1, N0,
+        {
+            acc[i].s[j] += ((int)(RHS_OFFSET)) * a_sum[0].s[i] + ((int)(LHS_OFFSET)) * b_sum[0].s[j];
+        })
+    })
+
+    const bool x_cond = PARTIAL_STORE_N0 != 0 && get_global_id(0) == 0;
+    const bool y_cond = PARTIAL_STORE_M0 != 0 && get_global_id(1) == 0;
+
+    // Quantize the tile
+    TILE(DATA_TYPE, M0, N0, accq);
+    T_QUANTIZE8_ASYMMETRIC(int, DATA_TYPE, M0, N0, DST_OFFSET, DST_SHIFT, DST_MULTIPLIER, acc, accq);
+
+    TILE(int, M0, 1, indirect_buffer);
+    LOOP_UNROLLING(int, _i, 0, 1, M0,
+    {
+        indirect_buffer[_i].v = min(_i, select(M0 - 1, PARTIAL_STORE_M0 - 1, y_cond));
+    });
+
+    T_STORE_INDIRECT_WIDTH_SELECT(DATA_TYPE, M0, N0, PARTIAL_STORE_N0, BUFFER, dst, 0, dst_stride_y, x_cond, accq, indirect_buffer);
+}
+#endif // defined(MAT_MUL_NATIVE_QUANTIZED_T_NT)
diff --git a/src/core/CL/cl_kernels/tile_helpers.h b/src/core/CL/cl_kernels/tile_helpers.h
index 872f4c0b57..c9b5370dea 100644
--- a/src/core/CL/cl_kernels/tile_helpers.h
+++ b/src/core/CL/cl_kernels/tile_helpers.h
@@ -536,6 +536,100 @@
         })                                                                                                             \
     })
 
+/** Store a VECTOR variable (e.g. int4, int8, char2 etc.) to a specified column in the TILE object
+ *
+ * @param[in]      VECTOR Vector variable to store
+ * @param[in, out] TILE   Tile variable to store to
+ * @param[in]      WIDTH  Width of the vector variable, also height of the tile (e.g. 2 if char2)
+ * @param[in]      COLUMN Column index of the tile
+ */
+#define COPY_VECTOR_TO_TILE_COLUMN(VECTOR, TILE, WIDTH, COLUMN) COPY_VECTOR_TO_TILE_COLUMN_STR(VECTOR, TILE, WIDTH, COLUMN)
+#define COPY_VECTOR_TO_TILE_COLUMN_STR(VECTOR, TILE, WIDTH, COLUMN) COPY_##WIDTH##_VECTOR_TO_TILE_COLUMN(VECTOR, TILE, COLUMN)
+#define COPY_1_VECTOR_TO_TILE_COLUMN(VECTOR, TILE, COLUMN) \
+    ({                                                      \
+        TILE[0].s[COLUMN] = VECTOR;                         \
+    })
+
+#define COPY_2_VECTOR_TO_TILE_COLUMN(VECTOR, TILE, COLUMN) \
+    ({                                                      \
+        TILE[0].s[COLUMN] = VECTOR.s0;                      \
+        TILE[1].s[COLUMN] = VECTOR.s1;                      \
+    })
+
+#define COPY_3_VECTOR_TO_TILE_COLUMN(VECTOR, TILE, COLUMN) \
+    ({                                                      \
+        TILE[0].s[COLUMN] = VECTOR.s0;                      \
+        TILE[1].s[COLUMN] = VECTOR.s1;                      \
+        TILE[2].s[COLUMN] = VECTOR.s2;                      \
+    })
+
+#define COPY_4_VECTOR_TO_TILE_COLUMN(VECTOR, TILE, COLUMN) \
+    ({                                                      \
+        TILE[0].s[COLUMN] = VECTOR.s0;                      \
+        TILE[1].s[COLUMN] = VECTOR.s1;                      \
+        TILE[2].s[COLUMN] = VECTOR.s2;                      \
+        TILE[3].s[COLUMN] = VECTOR.s3;                      \
+    })
+
+#define COPY_8_VECTOR_TO_TILE_COLUMN(VECTOR, TILE, COLUMN) \
+    ({                                                      \
+        TILE[0].s[COLUMN] = VECTOR.s0;                      \
+        TILE[1].s[COLUMN] = VECTOR.s1;                      \
+        TILE[2].s[COLUMN] = VECTOR.s2;                      \
+        TILE[3].s[COLUMN] = VECTOR.s3;                      \
+        TILE[4].s[COLUMN] = VECTOR.s4;                      \
+        TILE[5].s[COLUMN] = VECTOR.s5;                      \
+        TILE[6].s[COLUMN] = VECTOR.s6;                      \
+        TILE[7].s[COLUMN] = VECTOR.s7;                      \
+    })
+
+#define COPY_16_VECTOR_TO_TILE_COLUMN(VECTOR, TILE, COLUMN) \
+    ({                                                      \
+        TILE[0].s[COLUMN] = VECTOR.s0;                      \
+        TILE[1].s[COLUMN] = VECTOR.s1;                      \
+        TILE[2].s[COLUMN] = VECTOR.s2;                      \
+        TILE[3].s[COLUMN] = VECTOR.s3;                      \
+        TILE[4].s[COLUMN] = VECTOR.s4;                      \
+        TILE[5].s[COLUMN] = VECTOR.s5;                      \
+        TILE[6].s[COLUMN] = VECTOR.s6;                      \
+        TILE[7].s[COLUMN] = VECTOR.s7;                      \
+        TILE[8].s[COLUMN] = VECTOR.s8;                      \
+        TILE[9].s[COLUMN] = VECTOR.s9;                      \
+        TILE[10].s[COLUMN] = VECTOR.sA;                     \
+        TILE[11].s[COLUMN] = VECTOR.sB;                     \
+        TILE[12].s[COLUMN] = VECTOR.sC;                     \
+        TILE[13].s[COLUMN] = VECTOR.sD;                     \
+        TILE[14].s[COLUMN] = VECTOR.sE;                     \
+        TILE[15].s[COLUMN] = VECTOR.sF;                     \
+    })
+
+/** Load SRC_HEIGHT x SRC_WIDTH elements from global memory (tensor), and store them in a SRC_WIDTH x SRC_HEIGHT tile
+ *
+ * @param[in]  DATA_TYPE     Data type
+ * @param[in]  SRC_HEIGHT    Number of source rows, or number of columns of the output tile
+ * @param[in]  SRC_WIDTH     Number of source columns, or number of tile rows
+ * @param[in]  TENSOR_TYPE   Type of cl_type used to store the tensor in global memory (BUFFER=cl_buffer, IMAGE=cl_image).
+ *                           In case of cl_image, only WIDTH multiples of 4 are supported (4, 8, 16)
+ * @param[in]  TENSOR        Tensor basename
+ * @param[in]  X             Starting X position
+ * @param[in]  Y             Starting Y position
+ * @param[in]  YI_MULTIPLIER Parameter used to multiply the internal row increment (_i).
+ *                           In common cases should be 1 but it becomes useful when we want to load rows which are multiple of STRIDE_Y.
+ *                           (e.g. loading the weights of convolution layer).
+ *                           In this case the address calculation is performed as: (Y + _i * Y_MULTIPLIER) * STRIDE_Y
+ * @param[in]  STRIDE_Y      Stride Y (in bytes) used to load each row.
+ * @param[out] dst           Output tile
+ */
+#define T_LOAD_TRANSPOSED(DATA_TYPE, SRC_HEIGHT, SRC_WIDTH, TENSOR_TYPE, TENSOR, X, Y, YI_MULTIPLIER, STRIDE_Y, dst)     \
+    ({                                                                                                                   \
+        LOOP_UNROLLING(int, _i, 0, 1, SRC_HEIGHT,                                                                        \
+        {                                                                                                                \
+            VEC_DATA_TYPE(DATA_TYPE, SRC_WIDTH)                                                                          \
+                tmp = V_LOAD(DATA_TYPE, SRC_WIDTH, TENSOR_TYPE, TENSOR, X, ((Y) + _i * (int)(YI_MULTIPLIER)), STRIDE_Y); \
+            COPY_VECTOR_TO_TILE_COLUMN(tmp, dst, SRC_WIDTH, _i);                                                         \
+        })                                                                                                               \
+    })
+
 /** Load a tile from global memory (tensor) using an indirect Y index tile
  *
  * @param[in]  DATA_TYPE   Data type
@@ -1259,6 +1353,11 @@
  * @param[in]      lhs           LHS tile
  * @param[in]      rhs           RHS tile
  * @param[in, out] dst           DST tile
+ *
+ * @note For Int8/UInt8 multiplications, we only have T_MMUL_NT_T because we need
+ *       the multiply the rows of Lhs and Rhs tensors to utilize dot product extension.
+ *       Addition of other versions requires dealing with on the fly transposition of
+ *       these tile elements and therefore is not favored.
  */
 #define T_MMUL(LHS_DATA_TYPE, RHS_DATA_TYPE, DST_DATA_TYPE, M0, N0, K0, LHS_LAYOUT, RHS_LAYOUT, lhs, rhs, dst) T_MMUL_##LHS_LAYOUT##_##RHS_LAYOUT(LHS_DATA_TYPE, RHS_DATA_TYPE, DST_DATA_TYPE, M0, N0, K0, lhs, rhs, dst)
 #define T_MMUL_NT_T(LHS_DATA_TYPE, RHS_DATA_TYPE, DST_DATA_TYPE, M0, N0, K0, lhs, rhs, dst) T_MMUL_NT_T_##LHS_DATA_TYPE##_##RHS_DATA_TYPE##_##DST_DATA_TYPE(LHS_DATA_TYPE, RHS_DATA_TYPE, DST_DATA_TYPE, M0, N0, K0, lhs, rhs, dst)
diff --git a/src/gpu/cl/ClKernelLibrary.cpp b/src/gpu/cl/ClKernelLibrary.cpp
index 44b086f2fc..e657687887 100644
--- a/src/gpu/cl/ClKernelLibrary.cpp
+++ b/src/gpu/cl/ClKernelLibrary.cpp
@@ -323,6 +323,8 @@ const std::map<std::string, std::string> ClKernelLibrary::_kernel_program_map =
     { "mat_mul_native_nt_t", "common/mat_mul.cl" },
     { "mat_mul_native_t_nt", "common/mat_mul.cl" },
     { "mat_mul_native_t_t", "common/mat_mul.cl" },
+    { "mat_mul_native_quantized_nt_nt", "common/mat_mul_quantized.cl" },
+    { "mat_mul_native_quantized_t_nt", "common/mat_mul_quantized.cl" },
     { "max_unpooling_layer_2", "common/unpooling_layer.cl" },
     { "mean_stddev_normalization", "common/mean_stddev_normalization.cl" },
     { "memset", "common/memset.cl" },
@@ -794,6 +796,10 @@ const std::map<std::string, std::string> ClKernelLibrary::_program_source_map =
         "common/mat_mul.cl",
 #include "./cl_kernels/common/mat_mul.clembed"
     },
+    {
+        "common/mat_mul_quantized.cl",
+#include "./cl_kernels/common/mat_mul_quantized.clembed"
+    },
 #ifdef ENABLE_NCHW_KERNELS
     {
         "nchw/batch_to_space.cl",
diff --git a/src/gpu/cl/kernels/ClMatMulLowpNativeKernel.cpp b/src/gpu/cl/kernels/ClMatMulLowpNativeKernel.cpp
new file mode 100644
index 0000000000..d5ecdf7dd2
--- /dev/null
+++ b/src/gpu/cl/kernels/ClMatMulLowpNativeKernel.cpp
@@ -0,0 +1,224 @@
+/*
+ * Copyright (c) 2023 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "src/gpu/cl/kernels/ClMatMulLowpNativeKernel.h"
+
+#include "arm_compute/core/CL/CLHelpers.h"
+#include "arm_compute/core/CL/ICLTensor.h"
+#include "arm_compute/core/ITensorPack.h"
+#include "arm_compute/core/TensorInfo.h"
+#include "arm_compute/core/utils/misc/ShapeCalculator.h"
+#include "arm_compute/core/utils/quantization/AsymmHelpers.h"
+
+#include "src/common/utils/Log.h"
+#include "src/core/helpers/AutoConfiguration.h"
+#include "src/core/helpers/WindowHelpers.h"
+#include "src/gpu/cl/ClCompileContext.h"
+
+#include "support/Cast.h"
+#include "support/StringSupport.h"
+
+namespace arm_compute
+{
+namespace opencl
+{
+namespace kernels
+{
+namespace
+{
+Status validate_matmul_kernel_info(const MatMulKernelInfo &matmul_kernel_info)
+{
+    const bool adj_lhs = matmul_kernel_info.adj_lhs;
+    const bool adj_rhs = matmul_kernel_info.adj_rhs;
+    const int  m0      = matmul_kernel_info.m0;
+    const int  n0      = matmul_kernel_info.n0;
+    const int  k0      = matmul_kernel_info.k0;
+
+    // Validate M0
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG(m0 < 1, "Only positive integers are supported for M0");
+
+    if(adj_lhs)
+    {
+        ARM_COMPUTE_RETURN_ERROR_ON_MSG(((m0 & (m0 - 1)) && (m0 != 3)) || (m0 > 16), "Only 1,2,3,4,8,16 are supported for M0 for Lhs transposed");
+    }
+
+    // Validate N0
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG(n0 < 1, "Only positive integers are supported for N0");
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG(((n0 & (n0 - 1)) && (n0 != 3)) || (n0 > 16), "Only 1,2,3,4,8,16 are supported for N0");
+
+    // Validate K0
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG(k0 < 1, "Only positive integers are supported for K0");
+    if(!adj_lhs || adj_rhs)
+    {
+        ARM_COMPUTE_RETURN_ERROR_ON_MSG(((k0 & (k0 - 1)) && (k0 != 3)) || (k0 > 16), "Only 1,2,3,4,8,16 are supported for K0");
+    }
+
+    return Status{};
+}
+
+Status validate_input_shapes(const TensorShape &lhs_shape, const TensorShape &rhs_shape, const MatMulKernelInfo &matmul_kernel_info)
+{
+    const size_t lhs_k = matmul_kernel_info.adj_lhs ? lhs_shape.y() : lhs_shape.x();
+    const size_t rhs_k = matmul_kernel_info.adj_rhs ? rhs_shape.x() : rhs_shape.y();
+
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG(lhs_k != rhs_k, "K dimension in Lhs and Rhs matrices must match.");
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG(lhs_shape.total_size() == 0, "Lhs tensor can't be empty");
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG(rhs_shape.total_size() == 0, "Rhs tensor can't be empty");
+
+    constexpr size_t batch_dim_start = 2;
+    for(size_t i = batch_dim_start; i < Coordinates::num_max_dimensions; ++i)
+    {
+        ARM_COMPUTE_RETURN_ERROR_ON_MSG(lhs_shape[i] != rhs_shape[i], "Batch dimension broadcasting is not supported");
+    }
+
+    return Status{};
+}
+}
+ClMatMulLowpNativeKernel::ClMatMulLowpNativeKernel()
+{
+    _type = CLKernelType::GEMM;
+}
+Status ClMatMulLowpNativeKernel::validate(const ITensorInfo *lhs, const ITensorInfo *rhs, const ITensorInfo *output, const MatMulKernelInfo &matmul_kernel_info)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(lhs, rhs, output);
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(lhs, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED);
+    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(lhs, rhs);
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_matmul_kernel_info(matmul_kernel_info));
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_input_shapes(lhs->tensor_shape(), rhs->tensor_shape(), matmul_kernel_info));
+
+    if(output->total_size() != 0)
+    {
+        const TensorInfo tensor_info_output = output->clone()->set_tensor_shape(misc::shape_calculator::compute_matmul_shape(lhs->tensor_shape(), rhs->tensor_shape(), matmul_kernel_info));
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(output, &tensor_info_output);
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(lhs, output);
+    }
+
+    return Status{};
+}
+void ClMatMulLowpNativeKernel::configure(const ClCompileContext &compile_context, ITensorInfo *lhs, ITensorInfo *rhs, ITensorInfo *output, const MatMulKernelInfo &matmul_kernel_info)
+{
+    ARM_COMPUTE_ERROR_ON_NULLPTR(lhs, rhs, output, &compile_context, &matmul_kernel_info);
+    ARM_COMPUTE_LOG_PARAMS(lhs, rhs, output, matmul_kernel_info);
+    ARM_COMPUTE_ERROR_THROW_ON(validate(lhs, rhs, output, matmul_kernel_info));
+
+    // output tensor auto initialization if not yet initialized
+    auto_init_if_empty(*output, lhs->clone()->set_tensor_shape(misc::shape_calculator::compute_matmul_shape(lhs->tensor_shape(), rhs->tensor_shape(), matmul_kernel_info)));
+
+    const int  m       = output->dimension(1);
+    const int  n       = output->dimension(0);
+    const int  k       = matmul_kernel_info.adj_lhs ? lhs->tensor_shape().y() : lhs->tensor_shape().x();
+    const bool adj_lhs = matmul_kernel_info.adj_lhs;
+
+    int m0 = adj_lhs ? adjust_vec_size(matmul_kernel_info.m0, m) : std::min(matmul_kernel_info.m0, m);
+    int n0 = adjust_vec_size(matmul_kernel_info.n0, n);
+
+    // Configure kernel window
+    Window win = calculate_max_window(*output, Steps(n0, m0));
+    win        = win.collapse(win, Window::DimZ);
+    IClKernel::configure_internal(win);
+
+    // Calculate partial (store instead of load) M0 and partial N0 for the partial blocks at the end of a row/column if any. This is to avoid padding.
+    const unsigned int partial_store_m0 = m % m0;
+    const unsigned int partial_store_n0 = n % n0;
+
+    CLBuildOptions build_opts;
+    build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(lhs->data_type()));
+    build_opts.add_option("-DM0=" + support::cpp11::to_string(m0));
+    build_opts.add_option("-DN0=" + support::cpp11::to_string(n0));
+    build_opts.add_option("-DK0=" + support::cpp11::to_string(matmul_kernel_info.k0));
+    build_opts.add_option("-DPARTIAL_STORE_M0=" + support::cpp11::to_string(partial_store_m0));
+    build_opts.add_option("-DPARTIAL_STORE_N0=" + support::cpp11::to_string(partial_store_n0));
+    build_opts.add_option("-DK=" + support::cpp11::to_string(k));
+
+    const UniformQuantizationInfo lqinfo = lhs->quantization_info().uniform();
+    const UniformQuantizationInfo rqinfo = rhs->quantization_info().uniform();
+    const UniformQuantizationInfo dqinfo = output->quantization_info().uniform();
+
+    float multiplier        = lqinfo.scale * rqinfo.scale / dqinfo.scale;
+    int   output_multiplier = 0;
+    int   output_shift      = 0;
+    arm_compute::quantization::calculate_quantized_multiplier(multiplier, &output_multiplier, &output_shift);
+
+    build_opts.add_option("-DDST_MULTIPLIER=" + support::cpp11::to_string(output_multiplier));
+    build_opts.add_option("-DDST_SHIFT=" + support::cpp11::to_string(output_shift));
+
+    build_opts.add_option("-DLHS_OFFSET=" + support::cpp11::to_string(-lqinfo.offset)); // Note this is passed as negative to maintain similarity with CLDirectConv2D
+    build_opts.add_option("-DRHS_OFFSET=" + support::cpp11::to_string(-rqinfo.offset)); // Note this is passed as negative to maintain similarity with CLDirectConv2D
+    build_opts.add_option("-DDST_OFFSET=" + support::cpp11::to_string(dqinfo.offset));  // Passed as positive (unlike the above two)
+
+    std::string kernel_name("mat_mul_native_quantized");
+    kernel_name += matmul_kernel_info.adj_lhs ? "_t" : "_nt";
+    kernel_name += matmul_kernel_info.adj_rhs ? "_t" : "_nt";
+
+    // A macro guard to compile ONLY the kernel of interest
+    build_opts.add_option("-D" + upper_string(kernel_name));
+
+    // Create kernel
+    _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
+
+    // Set config_id for enabling LWS tuning
+    const size_t number_of_batches = output->tensor_shape().total_size() / (m * n);
+
+    _config_id = kernel_name;
+    _config_id += "_";
+    _config_id += lower_string(string_from_data_type(lhs->data_type()));
+    _config_id += "_";
+    _config_id += support::cpp11::to_string(m);
+    _config_id += "_";
+    _config_id += support::cpp11::to_string(n);
+    _config_id += "_";
+    _config_id += support::cpp11::to_string(k);
+    _config_id += "_";
+    _config_id += support::cpp11::to_string(number_of_batches);
+    _config_id += "_";
+    _config_id += support::cpp11::to_string(m0);
+    _config_id += "_";
+    _config_id += support::cpp11::to_string(n0);
+    _config_id += "_";
+    _config_id += support::cpp11::to_string(matmul_kernel_info.k0);
+}
+
+void ClMatMulLowpNativeKernel::run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue)
+{
+    ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
+    ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
+
+    const ICLTensor *lhs    = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC_0));
+    const ICLTensor *rhs    = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC_1));
+    ICLTensor       *output = utils::cast::polymorphic_downcast<ICLTensor *>(tensors.get_tensor(TensorType::ACL_DST));
+    ARM_COMPUTE_ERROR_ON_NULLPTR(lhs, rhs, output);
+    ARM_COMPUTE_LOG_PARAMS(lhs, rhs, output);
+
+    unsigned int idx              = 0;
+    Window       window_collapsed = window.collapse(ICLKernel::window(), Window::DimZ);
+
+    add_3d_tensor_nhw_argument(idx, lhs);
+    add_3d_tensor_nhw_argument(idx, rhs);
+    add_3d_tensor_nhw_argument(idx, output);
+
+    enqueue(queue, *this, window_collapsed, lws_hint());
+}
+
+} // namespace kernels
+} // namespace opencl
+} // namespace arm_compute
diff --git a/src/gpu/cl/kernels/ClMatMulLowpNativeKernel.h b/src/gpu/cl/kernels/ClMatMulLowpNativeKernel.h
new file mode 100644
index 0000000000..13a33fbd62
--- /dev/null
+++ b/src/gpu/cl/kernels/ClMatMulLowpNativeKernel.h
@@ -0,0 +1,69 @@
+/*
+ * Copyright (c) 2023 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef ACL_SRC_GPU_CL_KERNELS_CLMATMULLOWPNATIVEKERNEL
+#define ACL_SRC_GPU_CL_KERNELS_CLMATMULLOWPNATIVEKERNEL
+
+#include "src/core/common/Macros.h"
+#include "src/gpu/cl/ClCompileContext.h"
+#include "src/gpu/cl/IClKernel.h"
+
+namespace arm_compute
+{
+// Forward declerations
+struct MatMulKernelInfo;
+namespace opencl
+{
+namespace kernels
+{
+class ClMatMulLowpNativeKernel : public IClKernel
+{
+public:
+    ClMatMulLowpNativeKernel();
+    ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClMatMulLowpNativeKernel);
+    /** Initialise the kernel's input and output.
+     *
+     * @param[in]  compile_context The compile context to be used.
+     * @param[in]  lhs             Input tensor for the LHS matrix. Data type supported: QASYMM8_SIGNED/QASYMM8.
+     *                             Dimensions above 2 are collapsed onto dimension 2 and represent the batch.
+     * @param[in]  rhs             Input tensor for the RHS matrix. Data type supported: same as @p lhs.
+     *                             Dimensions above 2 are collapsed onto dimension 2 and represent the batch.
+     * @param[out] output          Output tensor info. Data type supported: same as @p lhs
+     * @param[in]  matmul_info     Attributes for Batch MatMul Kernel
+     */
+    void configure(const ClCompileContext &compile_context, ITensorInfo *lhs, ITensorInfo *rhs, ITensorInfo *output, const MatMulKernelInfo &matmul_info);
+    /** Static function to check if given info will lead to a valid configuration
+     *
+     * Similar to @ref ClMatMulLowpNativeKernel::configure()
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *lhs, const ITensorInfo *rhs, const ITensorInfo *output, const MatMulKernelInfo &matmul_info);
+
+    // Inherited methods overridden:
+    void run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue) override;
+};
+} // namespace kernels
+} // namespace opencl
+} // namespace arm_compute
+#endif /* ACL_SRC_GPU_CL_KERNELS_CLMATMULLOWPNATIVEKERNEL */
diff --git a/tests/validation/CL/MatMulKernel.cpp b/tests/validation/CL/MatMulKernel.cpp
index 9c19e42d04..ff872aaa0a 100644
--- a/tests/validation/CL/MatMulKernel.cpp
+++ b/tests/validation/CL/MatMulKernel.cpp
@@ -73,7 +73,7 @@ const auto k0_values_nightly_rhs_t         = framework::dataset::make("K0", { 1,
 const auto k0_values_nightly_lhs_t_rhs_nt  = framework::dataset::make("K0", { 1, 2, 3, 4, 5, 6, 7, 8 });
 
 template <typename T>
-using CLMatMulKernelFixture = MatMulKernelValidationFixture<T>;
+using CLMatMulKernelFixture = MatMulKernelValidationFixture<T, ClMatMulNativeKernel>;
 
 TEST_SUITE(CL)
 TEST_SUITE(MatMulKernel)
@@ -95,8 +95,8 @@ TEST_CASE(SupportedBlockSizes, framework::DatasetMode::ALL)
         { MatMulKernelInfo(false, false, 9, 1, 2), true },
         { MatMulKernelInfo(false, false, 3, 16, 3), true },
         { MatMulKernelInfo(false, false, 7, 3, 4), true },
-        { MatMulKernelInfo(false, false, 7, 3, 4, true), false }, // N0 not in {4, 8, 16}
-        { MatMulKernelInfo(false, false, 7, 1, 4, true), false }, // N0 not in {4, 8, 16}
+        { MatMulKernelInfo(false, false, 7, 3, 4, true), false },  // N0 not in {4, 8, 16}
+        { MatMulKernelInfo(false, false, 7, 1, 4, true), false },  // N0 not in {4, 8, 16}
         { MatMulKernelInfo(false, false, 7, 12, 4, true), false }, // N0 not in {4, 8, 16}
         { MatMulKernelInfo(false, false, 7, 4, 4, true), true },
         { MatMulKernelInfo(false, false, 7, 8, 4, true), true },
@@ -166,7 +166,7 @@ TEST_CASE(SupportedBlockSizes, framework::DatasetMode::ALL)
 
         if(!pair.first.export_rhs_to_cl_image || export_to_cl_image_supported)
         {
-           ARM_COMPUTE_EXPECT(bool(status) == pair.second, framework::LogLevel::ERRORS);
+            ARM_COMPUTE_EXPECT(bool(status) == pair.second, framework::LogLevel::ERRORS);
         }
     }
 }
@@ -176,9 +176,9 @@ TEST_CASE(ExportToCLImage, framework::DatasetMode::ALL)
     // We skip this test if the hardware does not support exporting to CL Image
     if(image2d_from_buffer_supported(CLKernelLibrary::get().get_device()))
     {
-        constexpr size_t pixel_size = 4;
-        const size_t max_image_w = pixel_size * CLKernelLibrary::get().get_device().getInfo<CL_DEVICE_IMAGE2D_MAX_WIDTH>();
-        const size_t max_image_h = CLKernelLibrary::get().get_device().getInfo<CL_DEVICE_IMAGE2D_MAX_HEIGHT>();
+        constexpr size_t pixel_size  = 4;
+        const size_t     max_image_w = pixel_size * CLKernelLibrary::get().get_device().getInfo<CL_DEVICE_IMAGE2D_MAX_WIDTH>();
+        const size_t     max_image_h = CLKernelLibrary::get().get_device().getInfo<CL_DEVICE_IMAGE2D_MAX_HEIGHT>();
 
         using ShapeConfigurationTuple = std::tuple<TensorShape, TensorShape, bool, bool, bool>;
         const std::vector<ShapeConfigurationTuple> shape_configurations =
@@ -186,18 +186,18 @@ TEST_CASE(ExportToCLImage, framework::DatasetMode::ALL)
             // lhs_shape, rhs_shape, adj_lhs, adj_rhs, expected
             // Lhs t/Nt, Rhs Nt
             // Transposition of Lhs doesn't add any value to the tests, therefore always assumed false below
-            { TensorShape(5U, 1U), TensorShape(3U, 5U), false, false, false }, // N should be multiple of 4
+            { TensorShape(5U, 1U), TensorShape(3U, 5U), false, false, false },  // N should be multiple of 4
             { TensorShape(5U, 1U), TensorShape(14U, 5U), false, false, false }, // N should be multiple of 4
             { TensorShape(5U, 1U), TensorShape(12U, 5U), false, false, true },
             { TensorShape(5U, 1U), TensorShape(8U, 5U), false, false, true },
             { TensorShape(5U, 1U), TensorShape(4U, 5U), false, false, true },
             { TensorShape(max_image_h + 1, 1U), TensorShape(4U, max_image_h + 1), false, false, false }, // Cannot fit into CL Image memory's height
-            { TensorShape(5U, 1U), TensorShape(max_image_w + 1, 5U), false, false, false }, // Cannot fit into CL Image memory's width
-            { TensorShape(max_image_h, 1U), TensorShape(4U, max_image_h), false, false, true }, // Barely fits into CL Image memory's height
-            { TensorShape(5U, 1U), TensorShape(max_image_w, 5U), false, false, true }, // Barely fits into CL Image memory's width
+            { TensorShape(5U, 1U), TensorShape(max_image_w + 1, 5U), false, false, false },              // Cannot fit into CL Image memory's width
+            { TensorShape(max_image_h, 1U), TensorShape(4U, max_image_h), false, false, true },          // Barely fits into CL Image memory's height
+            { TensorShape(5U, 1U), TensorShape(max_image_w, 5U), false, false, true },                   // Barely fits into CL Image memory's width
 
             // Lhs Nt/T , Rhs T
-            { TensorShape(5U, 1U), TensorShape(5U, 3U), false, true, false }, // K should be multiple of 4
+            { TensorShape(5U, 1U), TensorShape(5U, 3U), false, true, false },  // K should be multiple of 4
             { TensorShape(5U, 1U), TensorShape(5U, 14U), false, true, false }, // K should be multiple of 4
             { TensorShape(4U, 1U), TensorShape(4U, 10U), false, true, true },
             { TensorShape(8U, 1U), TensorShape(8U, 9U), false, true, true },
@@ -216,7 +216,10 @@ TEST_CASE(ExportToCLImage, framework::DatasetMode::ALL)
             const bool adj_rhs = std::get<3>(tuple);
 
             // We choose M0, N0, K0 equal to 4 so that they're always valid for CLImage in any combination
-            const MatMulKernelInfo matmul_kernel_info {adj_lhs, adj_rhs, 4, 4, 4, true /* export_rhs_to_cl_image */};
+            const MatMulKernelInfo matmul_kernel_info
+            {
+                adj_lhs, adj_rhs, 4, 4, 4, true /* export_rhs_to_cl_image */
+            };
 
             TensorInfo output_info;
             Status     status = ClMatMulNativeKernel::validate(&lhs_info, &rhs_info, &output_info, matmul_kernel_info);
@@ -330,60 +333,60 @@ TEST_SUITE(Float)
 TEST_SUITE(FP32)
 TEST_SUITE(Buffer)
 FIXTURE_DATA_TEST_CASE(RunTiny, CLMatMulKernelFixture<float>, framework::DatasetMode::ALL, combine(combine(combine(combine(combine(combine(combine(datasets::TinyMatMulDataset(),
-                                                                                                                       framework::dataset::make("pretransose_A", { false, true })),
-                                                                                                                       framework::dataset::make("pretransose_B", { false, true })),
-                                                                                                                       m0_values_precommit),
-                                                                                                                       n0_values_precommit),
-                                                                                                                       k0_values_precommit),
-                                                                                                                       framework::dataset::make("export_rhs_to_cl_image", { false })),
-                                                                                                               framework::dataset::make("DataType", DataType::F32)))
+                                                                                                                   framework::dataset::make("TransposeA", { false, true })),
+                                                                                                                   framework::dataset::make("TransposeB", { false, true })),
+                                                                                                                   m0_values_precommit),
+                                                                                                                   n0_values_precommit),
+                                                                                                                   k0_values_precommit),
+                                                                                                           framework::dataset::make("ExportRhsToCLImage", { false })),
+                                                                                                   framework::dataset::make("DataType", DataType::F32)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_f32, 0.f, abs_tolerance_f32);
 }
 FIXTURE_DATA_TEST_CASE(RunSmall, CLMatMulKernelFixture<float>, framework::DatasetMode::ALL, combine(combine(combine(combine(combine(combine(combine(datasets::SmallMatMulDataset(),
-                                                                                                                       framework::dataset::make("pretransose_A", { false, true })),
-                                                                                                                       framework::dataset::make("pretransose_B", { false, true })),
-                                                                                                                       m0_values_precommit),
-                                                                                                                       n0_values_precommit),
-                                                                                                                       k0_values_precommit),
-                                                                                                                       framework::dataset::make("export_rhs_to_cl_image", { false })),
-                                                                                                               framework::dataset::make("DataType", DataType::F32)))
+                                                                                                                    framework::dataset::make("TransposeA", { false, true })),
+                                                                                                                    framework::dataset::make("TransposeB", { false, true })),
+                                                                                                                    m0_values_precommit),
+                                                                                                                    n0_values_precommit),
+                                                                                                                    k0_values_precommit),
+                                                                                                            framework::dataset::make("ExportRhsToCLImage", { false })),
+                                                                                                    framework::dataset::make("DataType", DataType::F32)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_f32, 0.f, abs_tolerance_f32);
 }
 FIXTURE_DATA_TEST_CASE(RunLargeNoTranspose, CLMatMulKernelFixture<float>, framework::DatasetMode::NIGHTLY, combine(combine(combine(combine(combine(combine(combine(datasets::LargeMatMulDataset(),
-                                                                                                                   framework::dataset::make("pretransose_A", { false })),
-                                                                                                                   framework::dataset::make("pretransose_B", { false })),
+                                                                                                                   framework::dataset::make("TransposeA", { false })),
+                                                                                                                   framework::dataset::make("TransposeB", { false })),
                                                                                                                    m0_values_nightly_lhs_nt),
                                                                                                                    n0_values_nightly_rhs_nt),
                                                                                                                    k0_values_nightly_lhs_nt_rhs_nt),
-                                                                                                                   framework::dataset::make("export_rhs_to_cl_image", { false })),
+                                                                                                                   framework::dataset::make("ExportRhsToCLImage", { false })),
                                                                                                                    framework::dataset::make("DataType", DataType::F32)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_f32, 0.f, abs_tolerance_f32);
 }
 FIXTURE_DATA_TEST_CASE(RunLargeRhsTransposed, CLMatMulKernelFixture<float>, framework::DatasetMode::NIGHTLY, combine(combine(combine(combine(combine(combine(combine(datasets::LargeMatMulDataset(),
-                                                                                                                     framework::dataset::make("pretransose_A", { false })),
-                                                                                                                     framework::dataset::make("pretransose_B", { true })),
+                                                                                                                     framework::dataset::make("TransposeA", { false })),
+                                                                                                                     framework::dataset::make("TransposeB", { true })),
                                                                                                                      m0_values_nightly_lhs_nt),
                                                                                                                      n0_values_nightly_rhs_t),
                                                                                                                      k0_values_nightly_rhs_t),
-                                                                                                                     framework::dataset::make("export_rhs_to_cl_image", { false })),
+                                                                                                                     framework::dataset::make("ExportRhsToCLImage", { false })),
                                                                                                                      framework::dataset::make("DataType", DataType::F32)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_f32, 0.f, abs_tolerance_f32);
 }
 FIXTURE_DATA_TEST_CASE(RunLargeLhsTransposed, CLMatMulKernelFixture<float>, framework::DatasetMode::NIGHTLY, combine(combine(combine(combine(combine(combine(combine(datasets::LargeMatMulDataset(),
-                                                                                                                     framework::dataset::make("pretransose_A", { true })),
-                                                                                                                     framework::dataset::make("pretransose_B", { false })),
+                                                                                                                     framework::dataset::make("TransposeA", { true })),
+                                                                                                                     framework::dataset::make("TransposeB", { false })),
                                                                                                                      m0_values_nightly_lhs_t),
                                                                                                                      n0_values_nightly_rhs_nt),
                                                                                                                      k0_values_nightly_lhs_t_rhs_nt),
-                                                                                                                     framework::dataset::make("export_rhs_to_cl_image", { false })),
+                                                                                                                     framework::dataset::make("ExportRhsToCLImage", { false })),
                                                                                                                      framework::dataset::make("DataType", DataType::F32)))
 {
     // Validate output
@@ -391,12 +394,12 @@ FIXTURE_DATA_TEST_CASE(RunLargeLhsTransposed, CLMatMulKernelFixture<float>, fram
 }
 FIXTURE_DATA_TEST_CASE(RunLargeLhsTransposedRhsTransposed, CLMatMulKernelFixture<float>, framework::DatasetMode::NIGHTLY,
                        combine(combine(combine(combine(combine(combine(combine(datasets::LargeMatMulDataset(),
-                                                                       framework::dataset::make("pretransose_A", { true })),
-                                                               framework::dataset::make("pretransose_B", { true })),
-                                                       m0_values_nightly_lhs_t),
-                                               n0_values_nightly_rhs_t),
-                                       k0_values_nightly_rhs_t),
-                                       framework::dataset::make("export_rhs_to_cl_image", { false })),
+                                                                               framework::dataset::make("TransposeA", { true })),
+                                                                       framework::dataset::make("TransposeB", { true })),
+                                                               m0_values_nightly_lhs_t),
+                                                       n0_values_nightly_rhs_t),
+                                               k0_values_nightly_rhs_t),
+                                       framework::dataset::make("ExportRhsToCLImage", { false })),
                                framework::dataset::make("DataType", DataType::F32)))
 {
     // Validate output
@@ -405,13 +408,13 @@ FIXTURE_DATA_TEST_CASE(RunLargeLhsTransposedRhsTransposed, CLMatMulKernelFixture
 // Running High Dimensional test is enough for FP32, because we're stressing the number of dimensions, not data type or M0/N0/K0
 // It's a good idea to test for each Lhs/Rhs T/NT combinations because they're different CL kernels
 FIXTURE_DATA_TEST_CASE(RunHighDimensional, CLMatMulKernelFixture<float>, framework::DatasetMode::ALL, combine(combine(combine(combine(combine(combine(combine(datasets::HighDimensionalMatMulDataset(),
-                                                                                                                 framework::dataset::make("pretransose_A", { false, true })),
-                                                                                                                 framework::dataset::make("pretransose_B", { false, true })),
-                                                                                                                 framework::dataset::make("M0", { 2 })),
-                                                                                                                 framework::dataset::make("N0", { 2 })),
-                                                                                                                 framework::dataset::make("K0", { 2 })),
-                                                                                                                 framework::dataset::make("export_rhs_to_cl_image", { false })),
-                                                                                                                 framework::dataset::make("DataType", DataType::F32)))
+                                                                                                                      framework::dataset::make("TransposeA", { false, true })),
+                                                                                                                      framework::dataset::make("TransposeB", { false, true })),
+                                                                                                                      framework::dataset::make("M0", { 2 })),
+                                                                                                                      framework::dataset::make("N0", { 2 })),
+                                                                                                                      framework::dataset::make("K0", { 2 })),
+                                                                                                                      framework::dataset::make("ExportRhsToCLImage", { false })),
+                                                                                                              framework::dataset::make("DataType", DataType::F32)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_f32, 0.f, abs_tolerance_f32);
@@ -419,14 +422,15 @@ FIXTURE_DATA_TEST_CASE(RunHighDimensional, CLMatMulKernelFixture<float>, framewo
 TEST_SUITE_END() // Buffer
 
 TEST_SUITE(ExportRhsToCLImage)
-FIXTURE_DATA_TEST_CASE(RunSmallRhsNotTransposed, CLMatMulKernelFixture<float>, framework::DatasetMode::ALL, combine(combine(combine(combine(combine(combine(combine(datasets::SmallMatMulDatasetRhsExportToCLImageRhsNT(),
-                                                                                                                       framework::dataset::make("pretransose_A", { true, false })),
-                                                                                                                       framework::dataset::make("pretransose_B", { false })),
-                                                                                                                       framework::dataset::make("M0", { 2 })),
-                                                                                                                       framework::dataset::make("N0", { 4, 8, 16 })),
-                                                                                                                       framework::dataset::make("K0", { 2, 4 })),
-                                                                                                                       framework::dataset::make("export_rhs_to_cl_image", { true })),
-                                                                                                               framework::dataset::make("DataType", DataType::F32)))
+FIXTURE_DATA_TEST_CASE(RunSmallRhsNotTransposed, CLMatMulKernelFixture<float>, framework::DatasetMode::ALL,
+                       combine(combine(combine(combine(combine(combine(combine(datasets::SmallMatMulDatasetRhsExportToCLImageRhsNT(),
+                                                                               framework::dataset::make("TransposeA", { true, false })),
+                                                                       framework::dataset::make("TransposeB", { false })),
+                                                               framework::dataset::make("M0", { 2 })),
+                                                       framework::dataset::make("N0", { 4, 8, 16 })),
+                                               framework::dataset::make("K0", { 2, 4 })),
+                                       framework::dataset::make("ExportRhsToCLImage", { true })),
+                               framework::dataset::make("DataType", DataType::F32)))
 {
     // Validate output
     if(_device_supports_export_to_cl_image)
@@ -434,14 +438,15 @@ FIXTURE_DATA_TEST_CASE(RunSmallRhsNotTransposed, CLMatMulKernelFixture<float>, f
         validate(CLAccessor(_target), _reference, tolerance_f32, 0.f, abs_tolerance_f32);
     }
 }
-FIXTURE_DATA_TEST_CASE(RunLargeRhsNotTransposed, CLMatMulKernelFixture<float>, framework::DatasetMode::NIGHTLY, combine(combine(combine(combine(combine(combine(combine(datasets::LargeMatMulDatasetRhsExportToCLImageRhsNT(),
-                                                                                                                       framework::dataset::make("pretransose_A", { true, false })),
-                                                                                                                       framework::dataset::make("pretransose_B", { false })),
-                                                                                                                       framework::dataset::make("M0", { 2 })),              // Choices of M0 does not matter much because it's related to Lhs tensor
-                                                                                                                       framework::dataset::make("N0", { 4, 8, 16 })),
-                                                                                                                       framework::dataset::make("K0", { 1, 2, 3, 4 })),
-                                                                                                                       framework::dataset::make("export_rhs_to_cl_image", { true })),
-                                                                                                               framework::dataset::make("DataType", DataType::F32)))
+FIXTURE_DATA_TEST_CASE(RunLargeRhsNotTransposed, CLMatMulKernelFixture<float>, framework::DatasetMode::NIGHTLY,
+                       combine(combine(combine(combine(combine(combine(combine(datasets::LargeMatMulDatasetRhsExportToCLImageRhsNT(),
+                                                                               framework::dataset::make("TransposeA", { true, false })),
+                                                                       framework::dataset::make("TransposeB", { false })),
+                                                               framework::dataset::make("M0", { 2 })), // Choices of M0 does not matter much because it's related to Lhs tensor
+                                                       framework::dataset::make("N0", { 4, 8, 16 })),
+                                               framework::dataset::make("K0", { 1, 2, 3, 4 })),
+                                       framework::dataset::make("ExportRhsToCLImage", { true })),
+                               framework::dataset::make("DataType", DataType::F32)))
 {
     // Validate output
     if(_device_supports_export_to_cl_image)
@@ -449,14 +454,15 @@ FIXTURE_DATA_TEST_CASE(RunLargeRhsNotTransposed, CLMatMulKernelFixture<float>, f
         validate(CLAccessor(_target), _reference, tolerance_f32, 0.f, abs_tolerance_f32);
     }
 }
-FIXTURE_DATA_TEST_CASE(RunSmallRhsTransposed, CLMatMulKernelFixture<float>, framework::DatasetMode::ALL, combine(combine(combine(combine(combine(combine(combine(datasets::SmallMatMulDatasetRhsExportToCLImageRhsT(),
-                                                                                                                       framework::dataset::make("pretransose_A", { true, false })),
-                                                                                                                       framework::dataset::make("pretransose_B", { true })),
-                                                                                                                       framework::dataset::make("M0", { 2 })),
-                                                                                                                       framework::dataset::make("N0", { 2, 4 })),
-                                                                                                                       framework::dataset::make("K0", { 4, 8, 16 })),
-                                                                                                                       framework::dataset::make("export_rhs_to_cl_image", { true })),
-                                                                                                               framework::dataset::make("DataType", DataType::F32)))
+FIXTURE_DATA_TEST_CASE(RunSmallRhsTransposed, CLMatMulKernelFixture<float>, framework::DatasetMode::ALL,
+                       combine(combine(combine(combine(combine(combine(combine(datasets::SmallMatMulDatasetRhsExportToCLImageRhsT(),
+                                                                               framework::dataset::make("TransposeA", { true, false })),
+                                                                       framework::dataset::make("TransposeB", { true })),
+                                                               framework::dataset::make("M0", { 2 })),
+                                                       framework::dataset::make("N0", { 2, 4 })),
+                                               framework::dataset::make("K0", { 4, 8, 16 })),
+                                       framework::dataset::make("ExportRhsToCLImage", { true })),
+                               framework::dataset::make("DataType", DataType::F32)))
 {
     // Validate output
     if(_device_supports_export_to_cl_image)
@@ -464,14 +470,15 @@ FIXTURE_DATA_TEST_CASE(RunSmallRhsTransposed, CLMatMulKernelFixture<float>, fram
         validate(CLAccessor(_target), _reference, tolerance_f32, 0.f, abs_tolerance_f32);
     }
 }
-FIXTURE_DATA_TEST_CASE(RunLargeRhsTransposed, CLMatMulKernelFixture<float>, framework::DatasetMode::NIGHTLY, combine(combine(combine(combine(combine(combine(combine(datasets::LargeMatMulDatasetRhsExportToCLImageRhsT(),
-                                                                                                                       framework::dataset::make("pretransose_A", { true, false })),
-                                                                                                                       framework::dataset::make("pretransose_B", { true })),
-                                                                                                                       framework::dataset::make("M0", { 2 })),              // Choices of M0 does not matter much because it's related to Lhs tensor
-                                                                                                                       framework::dataset::make("N0", { 1, 2, 3, 4 })),
-                                                                                                                       framework::dataset::make("K0", { 4, 8, 16 })),
-                                                                                                                       framework::dataset::make("export_rhs_to_cl_image", { true })),
-                                                                                                               framework::dataset::make("DataType", DataType::F32)))
+FIXTURE_DATA_TEST_CASE(RunLargeRhsTransposed, CLMatMulKernelFixture<float>, framework::DatasetMode::NIGHTLY,
+                       combine(combine(combine(combine(combine(combine(combine(datasets::LargeMatMulDatasetRhsExportToCLImageRhsT(),
+                                                                               framework::dataset::make("TransposeA", { true, false })),
+                                                                       framework::dataset::make("TransposeB", { true })),
+                                                               framework::dataset::make("M0", { 2 })), // Choices of M0 does not matter much because it's related to Lhs tensor
+                                                       framework::dataset::make("N0", { 1, 2, 3, 4 })),
+                                               framework::dataset::make("K0", { 4, 8, 16 })),
+                                       framework::dataset::make("ExportRhsToCLImage", { true })),
+                               framework::dataset::make("DataType", DataType::F32)))
 {
     // Validate output
     if(_device_supports_export_to_cl_image)
@@ -485,61 +492,62 @@ TEST_SUITE_END() // FP32
 TEST_SUITE(FP16)
 TEST_SUITE(Buffer)
 FIXTURE_DATA_TEST_CASE(RunSmall, CLMatMulKernelFixture<half>, framework::DatasetMode::ALL, combine(combine(combine(combine(combine(combine(combine(datasets::SmallMatMulDataset(),
-                                                                                                                      framework::dataset::make("pretransose_A", { false, true })),
-                                                                                                                      framework::dataset::make("pretransose_B", { false, true })),
-                                                                                                                      m0_values_precommit),
-                                                                                                                      n0_values_precommit),
-                                                                                                                      k0_values_precommit),
-                                                                                                                      framework::dataset::make("export_rhs_to_cl_image", { false })),
-                                                                                                              framework::dataset::make("DataType", DataType::F16)))
+                                                                                                                   framework::dataset::make("TransposeA", { false, true })),
+                                                                                                                   framework::dataset::make("TransposeB", { false, true })),
+                                                                                                                   m0_values_precommit),
+                                                                                                                   n0_values_precommit),
+                                                                                                                   k0_values_precommit),
+                                                                                                           framework::dataset::make("ExportRhsToCLImage", { false })),
+                                                                                                   framework::dataset::make("DataType", DataType::F16)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_f16, 0.f, abs_tolerance_f16);
 }
 FIXTURE_DATA_TEST_CASE(RunLargeNoTranspose, CLMatMulKernelFixture<half>, framework::DatasetMode::NIGHTLY, combine(combine(combine(combine(combine(combine(combine(datasets::LargeMatMulDataset(),
-                                                                                                                  framework::dataset::make("pretransose_A", { false })),
-                                                                                                                  framework::dataset::make("pretransose_B", { false })),
+                                                                                                                  framework::dataset::make("TransposeA", { false })),
+                                                                                                                  framework::dataset::make("TransposeB", { false })),
                                                                                                                   m0_values_nightly_lhs_nt),
                                                                                                                   n0_values_nightly_rhs_nt),
                                                                                                                   k0_values_nightly_lhs_nt_rhs_nt),
-                                                                                                                  framework::dataset::make("export_rhs_to_cl_image", { false })),
+                                                                                                                  framework::dataset::make("ExportRhsToCLImage", { false })),
                                                                                                                   framework::dataset::make("DataType", DataType::F16)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_f16, 0.f, abs_tolerance_f16);
 }
 FIXTURE_DATA_TEST_CASE(RunLargeRhsTransposed, CLMatMulKernelFixture<half>, framework::DatasetMode::NIGHTLY, combine(combine(combine(combine(combine(combine(combine(datasets::LargeMatMulDataset(),
-                                                                                                                    framework::dataset::make("pretransose_A", { false })),
-                                                                                                                    framework::dataset::make("pretransose_B", { true })),
+                                                                                                                    framework::dataset::make("TransposeA", { false })),
+                                                                                                                    framework::dataset::make("TransposeB", { true })),
                                                                                                                     m0_values_nightly_lhs_nt),
                                                                                                                     n0_values_nightly_rhs_t),
                                                                                                                     k0_values_nightly_rhs_t),
-                                                                                                                    framework::dataset::make("export_rhs_to_cl_image", { false })),
+                                                                                                                    framework::dataset::make("ExportRhsToCLImage", { false })),
                                                                                                                     framework::dataset::make("DataType", DataType::F16)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_f16, 0.f, abs_tolerance_f16);
 }
 FIXTURE_DATA_TEST_CASE(RunLargeLhsTransposed, CLMatMulKernelFixture<half>, framework::DatasetMode::NIGHTLY, combine(combine(combine(combine(combine(combine(combine(datasets::LargeMatMulDataset(),
-                                                                                                                    framework::dataset::make("pretransose_A", { true })),
-                                                                                                                    framework::dataset::make("pretransose_B", { false })),
+                                                                                                                    framework::dataset::make("TransposeA", { true })),
+                                                                                                                    framework::dataset::make("TransposeB", { false })),
                                                                                                                     m0_values_nightly_lhs_t),
                                                                                                                     n0_values_nightly_rhs_nt),
                                                                                                                     k0_values_nightly_lhs_t_rhs_nt),
-                                                                                                                    framework::dataset::make("export_rhs_to_cl_image", { false })),
+                                                                                                                    framework::dataset::make("ExportRhsToCLImage", { false })),
                                                                                                                     framework::dataset::make("DataType", DataType::F16)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_f16, 0.f, abs_tolerance_f16);
 }
-FIXTURE_DATA_TEST_CASE(RunLargeLhsTransposedRhsTransposed, CLMatMulKernelFixture<half>, framework::DatasetMode::NIGHTLY, combine(combine(combine(combine(combine(combine(combine(datasets::LargeMatMulDataset(),
-                       framework::dataset::make("pretransose_A", { true })),
-                       framework::dataset::make("pretransose_B", { true })),
-                       m0_values_nightly_lhs_t),
-                       n0_values_nightly_rhs_t),
-                       k0_values_nightly_rhs_t),
-                       framework::dataset::make("export_rhs_to_cl_image", { false })),
-                       framework::dataset::make("DataType", DataType::F16)))
+FIXTURE_DATA_TEST_CASE(RunLargeLhsTransposedRhsTransposed, CLMatMulKernelFixture<half>, framework::DatasetMode::NIGHTLY,
+                       combine(combine(combine(combine(combine(combine(combine(datasets::LargeMatMulDataset(),
+                                                                               framework::dataset::make("TransposeA", { true })),
+                                                                       framework::dataset::make("TransposeB", { true })),
+                                                               m0_values_nightly_lhs_t),
+                                                       n0_values_nightly_rhs_t),
+                                               k0_values_nightly_rhs_t),
+                                       framework::dataset::make("ExportRhsToCLImage", { false })),
+                               framework::dataset::make("DataType", DataType::F16)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_f16, 0.f, abs_tolerance_f16);
@@ -547,14 +555,15 @@ FIXTURE_DATA_TEST_CASE(RunLargeLhsTransposedRhsTransposed, CLMatMulKernelFixture
 TEST_SUITE_END() // Buffer
 
 TEST_SUITE(ExportRhsToCLImage)
-FIXTURE_DATA_TEST_CASE(RunSmallRhsNotTransposed, CLMatMulKernelFixture<half>, framework::DatasetMode::ALL, combine(combine(combine(combine(combine(combine(combine(datasets::SmallMatMulDatasetRhsExportToCLImageRhsNT(),
-                                                                                                                       framework::dataset::make("pretransose_A", { true, false })),
-                                                                                                                       framework::dataset::make("pretransose_B", { false })),
-                                                                                                                       framework::dataset::make("M0", { 2 })),
-                                                                                                                       framework::dataset::make("N0", { 4, 8, 16 })),
-                                                                                                                       framework::dataset::make("K0", { 2, 4 })),
-                                                                                                                       framework::dataset::make("export_rhs_to_cl_image", { true })),
-                                                                                                               framework::dataset::make("DataType", DataType::F16)))
+FIXTURE_DATA_TEST_CASE(RunSmallRhsNotTransposed, CLMatMulKernelFixture<half>, framework::DatasetMode::ALL,
+                       combine(combine(combine(combine(combine(combine(combine(datasets::SmallMatMulDatasetRhsExportToCLImageRhsNT(),
+                                                                               framework::dataset::make("TransposeA", { true, false })),
+                                                                       framework::dataset::make("TransposeB", { false })),
+                                                               framework::dataset::make("M0", { 2 })),
+                                                       framework::dataset::make("N0", { 4, 8, 16 })),
+                                               framework::dataset::make("K0", { 2, 4 })),
+                                       framework::dataset::make("ExportRhsToCLImage", { true })),
+                               framework::dataset::make("DataType", DataType::F16)))
 {
     // Validate output
     if(_device_supports_export_to_cl_image)
@@ -562,14 +571,15 @@ FIXTURE_DATA_TEST_CASE(RunSmallRhsNotTransposed, CLMatMulKernelFixture<half>, fr
         validate(CLAccessor(_target), _reference, tolerance_f16, 0.f, abs_tolerance_f16);
     }
 }
-FIXTURE_DATA_TEST_CASE(RunLargeRhsNotTransposed, CLMatMulKernelFixture<half>, framework::DatasetMode::NIGHTLY, combine(combine(combine(combine(combine(combine(combine(datasets::LargeMatMulDatasetRhsExportToCLImageRhsNT(),
-                                                                                                                       framework::dataset::make("pretransose_A", { true, false })),
-                                                                                                                       framework::dataset::make("pretransose_B", { false })),
-                                                                                                                       framework::dataset::make("M0", { 2 })),              // Choices of M0 does not matter much because it's related to Lhs tensor
-                                                                                                                       framework::dataset::make("N0", { 4, 8, 16 })),
-                                                                                                                       framework::dataset::make("K0", { 1, 2, 3, 4 })),
-                                                                                                                       framework::dataset::make("export_rhs_to_cl_image", { true })),
-                                                                                                               framework::dataset::make("DataType", DataType::F16)))
+FIXTURE_DATA_TEST_CASE(RunLargeRhsNotTransposed, CLMatMulKernelFixture<half>, framework::DatasetMode::NIGHTLY,
+                       combine(combine(combine(combine(combine(combine(combine(datasets::LargeMatMulDatasetRhsExportToCLImageRhsNT(),
+                                                                               framework::dataset::make("TransposeA", { true, false })),
+                                                                       framework::dataset::make("TransposeB", { false })),
+                                                               framework::dataset::make("M0", { 2 })), // Choices of M0 does not matter much because it's related to Lhs tensor
+                                                       framework::dataset::make("N0", { 4, 8, 16 })),
+                                               framework::dataset::make("K0", { 1, 2, 3, 4 })),
+                                       framework::dataset::make("ExportRhsToCLImage", { true })),
+                               framework::dataset::make("DataType", DataType::F16)))
 {
     // Validate output
     if(_device_supports_export_to_cl_image)
@@ -577,14 +587,15 @@ FIXTURE_DATA_TEST_CASE(RunLargeRhsNotTransposed, CLMatMulKernelFixture<half>, fr
         validate(CLAccessor(_target), _reference, tolerance_f16, 0.f, abs_tolerance_f16);
     }
 }
-FIXTURE_DATA_TEST_CASE(RunSmallRhsTransposed, CLMatMulKernelFixture<half>, framework::DatasetMode::ALL, combine(combine(combine(combine(combine(combine(combine(datasets::SmallMatMulDatasetRhsExportToCLImageRhsT(),
-                                                                                                                       framework::dataset::make("pretransose_A", { true, false })),
-                                                                                                                       framework::dataset::make("pretransose_B", { true })),
-                                                                                                                       framework::dataset::make("M0", { 2 })),
-                                                                                                                       framework::dataset::make("N0", { 2, 4 })),
-                                                                                                                       framework::dataset::make("K0", { 4, 8, 16 })),
-                                                                                                                       framework::dataset::make("export_rhs_to_cl_image", { true })),
-                                                                                                               framework::dataset::make("DataType", DataType::F16)))
+FIXTURE_DATA_TEST_CASE(RunSmallRhsTransposed, CLMatMulKernelFixture<half>, framework::DatasetMode::ALL,
+                       combine(combine(combine(combine(combine(combine(combine(datasets::SmallMatMulDatasetRhsExportToCLImageRhsT(),
+                                                                               framework::dataset::make("TransposeA", { true, false })),
+                                                                       framework::dataset::make("TransposeB", { true })),
+                                                               framework::dataset::make("M0", { 2 })),
+                                                       framework::dataset::make("N0", { 2, 4 })),
+                                               framework::dataset::make("K0", { 4, 8, 16 })),
+                                       framework::dataset::make("ExportRhsToCLImage", { true })),
+                               framework::dataset::make("DataType", DataType::F16)))
 {
     // Validate output
     if(_device_supports_export_to_cl_image)
@@ -592,14 +603,15 @@ FIXTURE_DATA_TEST_CASE(RunSmallRhsTransposed, CLMatMulKernelFixture<half>, frame
         validate(CLAccessor(_target), _reference, tolerance_f16, 0.f, abs_tolerance_f16);
     }
 }
-FIXTURE_DATA_TEST_CASE(RunLargeRhsTransposed, CLMatMulKernelFixture<half>, framework::DatasetMode::NIGHTLY, combine(combine(combine(combine(combine(combine(combine(datasets::LargeMatMulDatasetRhsExportToCLImageRhsT(),
-                                                                                                                       framework::dataset::make("pretransose_A", { true, false })),
-                                                                                                                       framework::dataset::make("pretransose_B", { true })),
-                                                                                                                       framework::dataset::make("M0", { 2 })),              // Choices of M0 does not matter much because it's related to Lhs tensor
-                                                                                                                       framework::dataset::make("N0", { 1, 2, 3, 4 })),
-                                                                                                                       framework::dataset::make("K0", { 4, 8, 16 })),
-                                                                                                                       framework::dataset::make("export_rhs_to_cl_image", { true })),
-                                                                                                               framework::dataset::make("DataType", DataType::F16)))
+FIXTURE_DATA_TEST_CASE(RunLargeRhsTransposed, CLMatMulKernelFixture<half>, framework::DatasetMode::NIGHTLY,
+                       combine(combine(combine(combine(combine(combine(combine(datasets::LargeMatMulDatasetRhsExportToCLImageRhsT(),
+                                                                               framework::dataset::make("TransposeA", { true, false })),
+                                                                       framework::dataset::make("TransposeB", { true })),
+                                                               framework::dataset::make("M0", { 2 })), // Choices of M0 does not matter much because it's related to Lhs tensor
+                                                       framework::dataset::make("N0", { 1, 2, 3, 4 })),
+                                               framework::dataset::make("K0", { 4, 8, 16 })),
+                                       framework::dataset::make("ExportRhsToCLImage", { true })),
+                               framework::dataset::make("DataType", DataType::F16)))
 {
     // Validate output
     if(_device_supports_export_to_cl_image)
diff --git a/tests/validation/CL/MatMulLowpNativeKernel.cpp b/tests/validation/CL/MatMulLowpNativeKernel.cpp
new file mode 100644
index 0000000000..5932fa7c21
--- /dev/null
+++ b/tests/validation/CL/MatMulLowpNativeKernel.cpp
@@ -0,0 +1,337 @@
+/*
+ * Copyright (c) 2023 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include "arm_compute/runtime/CL/CLTensor.h"
+
+#include "src/gpu/cl/kernels/ClMatMulLowpNativeKernel.h"
+
+#include "tests/datasets/LargeMatMulDataset.h"
+#include "tests/datasets/SmallMatMulDataset.h"
+#include "tests/framework/Macros.h"
+#include "tests/framework/datasets/Datasets.h"
+#include "tests/validation/Validation.h"
+#include "tests/validation/fixtures/MatMulKernelFixture.h"
+#include "tests/validation/reference/Permute.h"
+
+#include <tuple>
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace
+{
+constexpr AbsoluteTolerance<float> tolerance_quant(1); /**< Tolerance value for comparing reference's output against implementation's output for quantized data types */
+}
+template <typename T>
+using CLMatMulLowpNativeKernelFixture = MatMulKernelValidationFixture<T, ClMatMulLowpNativeKernel>;
+
+/** M0 values to test --precommit*/
+const auto m0_values_precommit = framework::dataset::make("M0", { 1, 3 });
+
+/** N0 values to test --precommit*/
+const auto n0_values_precommit = framework::dataset::make("N0", { 2, 4 });
+
+/** K0 values to test --precommit*/
+const auto k0_values_precommit = framework::dataset::make("K0", { 2, 3 });
+
+/** M0 values to test --nightly*/
+const auto m0_values_nightly_lhs_nt = framework::dataset::make("M0", { 1, 2, 3, 4, 5, 6, 7, 8 });
+const auto m0_values_nightly_lhs_t  = framework::dataset::make("M0", { 1, 2, 3, 4, 8 });
+
+/** N0 values to test --nightly*/
+const auto n0_values_nightly_rhs_nt = framework::dataset::make("N0", { 1, 2, 3, 4, 8, 16 });
+// const auto n0_values_nightly_rhs_t  = framework::dataset::make("N0", { 1, 2, 3, 4, 8 });
+
+/** K0 values to test --nightly*/
+const auto k0_values_nightly_lhs_nt_rhs_nt = framework::dataset::make("K0", { 1, 2, 3, 4, 8, 16 });
+// const auto k0_values_nightly_rhs_t         = framework::dataset::make("K0", { 1, 2, 3, 4, 8 });
+const auto k0_values_nightly_lhs_t_rhs_nt = framework::dataset::make("K0", { 1, 2, 3, 4, 5, 6, 7, 8 });
+
+TEST_SUITE(CL)
+TEST_SUITE(MatMulLowpNativeKernel)
+TEST_SUITE(Validate)
+
+TEST_CASE(SupportedKernelConfigurations, framework::DatasetMode::ALL)
+{
+    using MatMulConfigurationPair = std::pair<MatMulKernelInfo, bool>;
+
+    const std::vector<MatMulConfigurationPair> supported_block_sizes =
+    {
+        // MatMulKernelInfo(adj_lhs, adj_rhs, M0, N0, K0, export_rhs_to_cl_image = false)
+        // Lhs not-transposed, Rhs-not-transposed
+        { MatMulKernelInfo(false, false, 0, 1, 1), false },  // M0 should be > 0
+        { MatMulKernelInfo(false, false, 3, 5, 1), false },  // N0 not in {1, 2, 3, 4, 8, 16}
+        { MatMulKernelInfo(false, false, 3, 6, 1), false },  // N0 not in {1, 2, 3, 4, 8, 16}
+        { MatMulKernelInfo(false, false, 3, 3, 17), false }, // K0 not in {1, 2, 3, 4, 8, 16}
+        { MatMulKernelInfo(false, false, 3, 3, 7), false },  // K0 not in {1, 2, 3, 4, 8, 16}
+        { MatMulKernelInfo(false, false, 9, 1, 2), true },
+        { MatMulKernelInfo(false, false, 3, 16, 3), true },
+        { MatMulKernelInfo(false, false, 7, 3, 4), true },
+        { MatMulKernelInfo(false, false, 7, 3, 4, true), true }, // export to CLImage is unsupported for quantized types
+    };
+
+    // Set big enough shapes so that block sizes are not truncated. Also, set all dimensions equal
+    // so that it doesn't fail for different NT/T configurations. We aim to test the block sizes here,
+    // not the shapes themselves.
+    const TensorInfo lhs_info = TensorInfo(TensorShape(100U, 100U), 1, DataType::QASYMM8_SIGNED);
+    const TensorInfo rhs_info = TensorInfo(TensorShape(100U, 100U), 1, DataType::QASYMM8_SIGNED);
+
+    for(auto &pair : supported_block_sizes)
+    {
+        TensorInfo output_info;
+        Status     status = ClMatMulLowpNativeKernel::validate(&lhs_info, &rhs_info, &output_info, pair.first);
+
+        ARM_COMPUTE_EXPECT(bool(status) == pair.second, framework::LogLevel::ERRORS);
+    }
+}
+
+TEST_CASE(ValidateInputShapes, framework::DatasetMode::ALL)
+{
+    // Configurations are assumed to be Nt/Nt, but will be transposed inside the test to test other configurations
+    using ShapeConfigurationTuple = std::tuple<TensorShape, TensorShape, bool>;
+    const std::vector<ShapeConfigurationTuple> shape_configurations =
+    {
+        { TensorShape(5U, 1U), TensorShape(3U, 5U), true },
+        { TensorShape(10U, 12U), TensorShape(3U, 10U), true },
+        { TensorShape(8U, 4U), TensorShape(2U, 8U), true },
+        { TensorShape(8U, 4U), TensorShape(2U, 5U), false }, // Mismatch in the K dimension
+        { TensorShape(5U, 0U), TensorShape(2U, 5U), false }, // Invalid dimension
+        { TensorShape(5U, 4U, 3U, 4U, 5U, 6U), TensorShape(2U, 5U, 3U, 4U, 5U, 6U), true },
+        { TensorShape(5U, 4U, 3U, 4U, 5U, 1U), TensorShape(2U, 5U, 3U, 4U, 5U, 6U), false }, // no batch broadcasting
+        { TensorShape(5U, 4U, 3U, 4U, 9U, 6U), TensorShape(2U, 5U, 3U, 4U, 5U, 6U), false }, // mismatch in batch dimension
+    };
+
+    for(auto &tuple : shape_configurations)
+    {
+        const bool expected = std::get<2>(tuple);
+
+        for(bool adj_lhs :
+            {
+                false, true
+            })
+        {
+            for(bool adj_rhs :
+                {
+                    false, true
+                })
+            {
+                TensorShape lhs_shape = std::get<0>(tuple);
+                TensorShape rhs_shape = std::get<1>(tuple);
+
+                if(adj_lhs)
+                {
+                    permute(lhs_shape, PermutationVector(1U, 0U));
+                }
+
+                if(adj_rhs)
+                {
+                    permute(rhs_shape, PermutationVector(1U, 0U));
+                }
+
+                const TensorInfo lhs_info = TensorInfo(lhs_shape, 1, DataType::QASYMM8_SIGNED);
+                const TensorInfo rhs_info = TensorInfo(rhs_shape, 1, DataType::QASYMM8_SIGNED);
+                TensorInfo       output_info;
+
+                MatMulKernelInfo matmul_kernel_info{ adj_lhs, adj_rhs, 1, 1, 1, false /* export_rhs_to_cl_image */ };
+
+                Status status = ClMatMulLowpNativeKernel::validate(&lhs_info, &rhs_info, &output_info, matmul_kernel_info);
+                ARM_COMPUTE_EXPECT(bool(status) == expected, framework::LogLevel::ERRORS);
+            }
+        }
+    }
+}
+
+TEST_CASE(ValidateDataTypes, framework::DatasetMode::ALL)
+{
+    using DataTypeConfigurationTuple = std::tuple<DataType, DataType, DataType, bool>;
+    const std::vector<DataTypeConfigurationTuple> data_type_configurations =
+    {
+        { DataType::F32, DataType::F32, DataType::F32, false }, // no floating point types
+        { DataType::F16, DataType::F16, DataType::F16, false }, // no floating point types
+        { DataType::F64, DataType::F64, DataType::F64, false }, // no double precision
+        { DataType::QASYMM8, DataType::QASYMM8, DataType::QASYMM8, true },
+        { DataType::QASYMM8_SIGNED, DataType::QASYMM8_SIGNED, DataType::QASYMM8_SIGNED, true },
+        { DataType::QSYMM8_PER_CHANNEL, DataType::QSYMM8_PER_CHANNEL, DataType::QSYMM8_PER_CHANNEL, false }, // only qasymm8/qasymm8_signed is supported
+        { DataType::QASYMM16, DataType::QASYMM16, DataType::QASYMM16, false },                               // only qasymm8/qasymm8_signed is supported
+        { DataType::QSYMM16, DataType::QSYMM16, DataType::QSYMM16, false },                                  // only qasymm8/qasymm8_signed is supported
+        { DataType::QSYMM8, DataType::QSYMM8, DataType::QSYMM8, false },                                     // only qasymm8/qasymm8_signed is supported
+        { DataType::QASYMM8, DataType::QASYMM8_SIGNED, DataType::QASYMM8, false },                           // no mixed data types
+        { DataType::S64, DataType::S64, DataType::S64, false },                                              // no integral types
+        { DataType::S32, DataType::S32, DataType::S32, false },                                              // no integral types
+        { DataType::S16, DataType::S16, DataType::S16, false },                                              // no integral types
+        { DataType::S8, DataType::S8, DataType::S8, false },                                                 // no integral types
+        { DataType::U64, DataType::U64, DataType::U64, false },                                              // no integral types
+        { DataType::U32, DataType::U32, DataType::U32, false },                                              // no integral types
+        { DataType::U16, DataType::U16, DataType::U16, false },                                              // no integral types
+        { DataType::U8, DataType::U8, DataType::U8, false },                                                 // no integral types
+    };
+
+    // It's enough to test a single shape and block size configuration while checking data types
+    const TensorShape      shape = TensorShape(10U, 10U);
+    const MatMulKernelInfo matmul_kernel_info{ false, false, 1, 1, 1, false };
+    for(auto &tuple : data_type_configurations)
+    {
+        const bool expected = std::get<3>(tuple);
+
+        const TensorInfo lhs_info(shape, 1, std::get<0>(tuple));
+        const TensorInfo rhs_info(shape, 1, std::get<1>(tuple));
+        TensorInfo       output_info(shape, 1, std::get<2>(tuple));
+
+        Status status = ClMatMulLowpNativeKernel::validate(&lhs_info, &rhs_info, &output_info, matmul_kernel_info);
+        ARM_COMPUTE_EXPECT(bool(status) == expected, framework::LogLevel::ERRORS);
+    }
+}
+
+TEST_SUITE_END() // Validate
+
+TEST_SUITE(Quantized)
+TEST_SUITE(QASYMM8_SIGNED)
+FIXTURE_DATA_TEST_CASE(RunTiny, CLMatMulLowpNativeKernelFixture<int8_t>, framework::DatasetMode::ALL, combine(combine(combine(combine(combine(combine(combine(datasets::TinyMatMulDataset(),
+                                                                                                                      framework::dataset::make("TransposeA", { true, false })),
+                                                                                                                      framework::dataset::make("TransposeB", { false })),
+                                                                                                                      m0_values_precommit),
+                                                                                                                      n0_values_precommit),
+                                                                                                                      k0_values_precommit),
+                                                                                                                      framework::dataset::make("ExportRhsToCLImage", { false })),
+                                                                                                              framework::dataset::make("DataType", DataType::QASYMM8_SIGNED)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_quant);
+}
+FIXTURE_DATA_TEST_CASE(RunSmall, CLMatMulLowpNativeKernelFixture<int8_t>, framework::DatasetMode::ALL, combine(combine(combine(combine(combine(combine(combine(datasets::SmallMatMulDataset(),
+                                                                                                                       framework::dataset::make("TransposeA", { true, false })),
+                                                                                                                       framework::dataset::make("TransposeB", { false })),
+                                                                                                                       m0_values_precommit),
+                                                                                                                       n0_values_precommit),
+                                                                                                                       k0_values_precommit),
+                                                                                                                       framework::dataset::make("ExportRhsToCLImage", { false })),
+                                                                                                               framework::dataset::make("DataType", DataType::QASYMM8_SIGNED)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_quant);
+}
+FIXTURE_DATA_TEST_CASE(RunLargeNoTranspose, CLMatMulLowpNativeKernelFixture<int8_t>, framework::DatasetMode::NIGHTLY,
+                       combine(combine(combine(combine(combine(combine(combine(datasets::LargeMatMulDataset(),
+                                                                               framework::dataset::make("TransposeA", { false })),
+                                                                       framework::dataset::make("TransposeB", { false })),
+                                                               m0_values_nightly_lhs_nt),
+                                                       n0_values_nightly_rhs_nt),
+                                               k0_values_nightly_lhs_nt_rhs_nt),
+                                       framework::dataset::make("ExportRhsToCLImage", { false })),
+                               framework::dataset::make("DataType", DataType::QASYMM8_SIGNED)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_quant);
+}
+FIXTURE_DATA_TEST_CASE(RunLargeLhsTransposed, CLMatMulLowpNativeKernelFixture<int8_t>, framework::DatasetMode::NIGHTLY,
+                       combine(combine(combine(combine(combine(combine(combine(datasets::LargeMatMulDataset(),
+                                                                               framework::dataset::make("TransposeA", { true })),
+                                                                       framework::dataset::make("TransposeB", { false })),
+                                                               m0_values_nightly_lhs_t),
+                                                       n0_values_nightly_rhs_nt),
+                                               k0_values_nightly_lhs_t_rhs_nt),
+                                       framework::dataset::make("ExportRhsToCLImage", { false })),
+                               framework::dataset::make("DataType", DataType::QASYMM8_SIGNED)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_quant);
+}
+// Running High Dimensional test is enough for qasymm8_signed, because we're stressing the number of dimensions, not data type or M0/N0/K0
+// It's a good idea to test for each Lhs/Rhs T/NT combinations because they're different CL kernels
+FIXTURE_DATA_TEST_CASE(RunHighDimensional, CLMatMulLowpNativeKernelFixture<int8_t>, framework::DatasetMode::ALL,
+                       combine(combine(combine(combine(combine(combine(combine(datasets::HighDimensionalMatMulDataset(),
+                                                                               framework::dataset::make("TransposeA", { true, false })),
+                                                                       framework::dataset::make("TransposeB", { false })),
+                                                               framework::dataset::make("M0", { 2 })),
+                                                       framework::dataset::make("N0", { 2 })),
+                                               framework::dataset::make("K0", { 2 })),
+                                       framework::dataset::make("ExportRhsToCLImage", { false })),
+                               framework::dataset::make("DataType", DataType::QASYMM8_SIGNED)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_quant);
+}
+TEST_SUITE_END() // QASYMM8_SIGNED
+
+TEST_SUITE(QASYMM8)
+FIXTURE_DATA_TEST_CASE(RunTiny, CLMatMulLowpNativeKernelFixture<uint8_t>, framework::DatasetMode::ALL, combine(combine(combine(combine(combine(combine(combine(datasets::TinyMatMulDataset(),
+                                                                                                                       framework::dataset::make("TransposeA", { true, false })),
+                                                                                                                       framework::dataset::make("TransposeB", { false })),
+                                                                                                                       m0_values_precommit),
+                                                                                                                       n0_values_precommit),
+                                                                                                                       k0_values_precommit),
+                                                                                                                       framework::dataset::make("ExportRhsToCLImage", { false })),
+                                                                                                               framework::dataset::make("DataType", DataType::QASYMM8)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_quant);
+}
+FIXTURE_DATA_TEST_CASE(RunSmall, CLMatMulLowpNativeKernelFixture<uint8_t>, framework::DatasetMode::ALL, combine(combine(combine(combine(combine(combine(combine(datasets::SmallMatMulDataset(),
+                                                                                                                        framework::dataset::make("TransposeA", { true, false })),
+                                                                                                                        framework::dataset::make("TransposeB", { false })),
+                                                                                                                        m0_values_precommit),
+                                                                                                                        n0_values_precommit),
+                                                                                                                        k0_values_precommit),
+                                                                                                                        framework::dataset::make("ExportRhsToCLImage", { false })),
+                                                                                                                framework::dataset::make("DataType", DataType::QASYMM8)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_quant);
+}
+FIXTURE_DATA_TEST_CASE(RunLargeNoTranspose, CLMatMulLowpNativeKernelFixture<uint8_t>, framework::DatasetMode::NIGHTLY,
+                       combine(combine(combine(combine(combine(combine(combine(datasets::LargeMatMulDataset(),
+                                                                               framework::dataset::make("TransposeA", { false })),
+                                                                       framework::dataset::make("TransposeB", { false })),
+                                                               m0_values_nightly_lhs_nt),
+                                                       n0_values_nightly_rhs_nt),
+                                               k0_values_nightly_lhs_nt_rhs_nt),
+                                       framework::dataset::make("ExportRhsToCLImage", { false })),
+                               framework::dataset::make("DataType", DataType::QASYMM8)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_quant);
+}
+FIXTURE_DATA_TEST_CASE(RunLargeLhsTransposed, CLMatMulLowpNativeKernelFixture<uint8_t>, framework::DatasetMode::NIGHTLY,
+                       combine(combine(combine(combine(combine(combine(combine(datasets::LargeMatMulDataset(),
+                                                                               framework::dataset::make("TransposeA", { true })),
+                                                                       framework::dataset::make("TransposeB", { false })),
+                                                               m0_values_nightly_lhs_t),
+                                                       n0_values_nightly_rhs_nt),
+                                               k0_values_nightly_lhs_t_rhs_nt),
+                                       framework::dataset::make("ExportRhsToCLImage", { false })),
+                               framework::dataset::make("DataType", DataType::QASYMM8)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_quant);
+}
+TEST_SUITE_END() // QASYMM8
+TEST_SUITE_END() // Quantized
+TEST_SUITE_END() // MatMulLowpNativeKernel
+TEST_SUITE_END() // CL
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/Helpers.cpp b/tests/validation/Helpers.cpp
index be194dd266..110325c5a0 100644
--- a/tests/validation/Helpers.cpp
+++ b/tests/validation/Helpers.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2022 Arm Limited.
+ * Copyright (c) 2017-2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -22,6 +22,7 @@
  * SOFTWARE.
  */
 #include "tests/validation/Helpers.h"
+#include "tests/framework/Asserts.h"
 
 #include <algorithm>
 #include <cmath>
@@ -373,6 +374,105 @@ void add_padding_y(std::initializer_list<ITensor *> tensors, const DataLayout &d
     }
 }
 
+QuantizationInfo calculate_mat_mul_dst_q_info(const QuantizationInfo &a_q_info, const QuantizationInfo &b_q_info, int m, int n, int k, DataType data_type)
+{
+    ARM_COMPUTE_UNUSED(m, n);
+    QuantizationInfo c_q_info;
+
+    ARM_COMPUTE_ASSERT(data_type == DataType::QASYMM8 || data_type == DataType::QASYMM8_SIGNED);
+
+    const int32_t t_max = static_cast<int32_t>(data_type == DataType::QASYMM8 ? std::numeric_limits<uint8_t>::max() : std::numeric_limits<int8_t>::max());
+    const int32_t t_min = static_cast<int32_t>(data_type == DataType::QASYMM8 ? std::numeric_limits<uint8_t>::min() : std::numeric_limits<int8_t>::min());
+
+    /**  Quantization Setup of matrix multiplication
+     *
+     *  We have a matrix multiplication of the form C = A * B
+     *  where A is (M X K), B is (K x N) and C is therefore (M x N).
+     *
+     *  If we have some distributions statistics of A and B, i.e. mean and variance,
+     *  we can estimate the mean and variance of a single value in C matrix and
+     *  pick good scale and offset values for the output and have non-saturated tests.
+     *
+     *  Each element in the output matrix can be calculated as follows:
+     *      C_ij = sum_k(A_ik * B_kj)
+     *
+     *      All values are float above.
+     *
+     * Note: All possible A_ik, B_kj random variables are assumed mutually independent.
+     *
+     * Terminology:
+     *  E[X]: Mean of the random variable X (sometimes referred as mu_x)
+     *  var(X): Variance of the random variable X (someimes referred as sigma^2_x)
+     *  std(X): sqrt(var(X)), standard deviation of X
+     *
+     * 1) Calculate the mean:
+     *      E[C_ij] = sum_k( E[A_ik] * E[B_kj] ) = K * mean_a * mean_b
+     *
+     *      Since elements of A and B are uniformly distributed random variables, we have
+     *          mean_a = (max_a + min_a) / 2, mean_b = (max_b + min_b ) / 2
+     *          max_a and min_a can be calculated with the scale_a/b and offset_a/b
+     *              by replacing data type minimum and maximums in the equations
+     *
+     * 2) Calculate the variance:
+     *      var(C_ij) = sum_k( var(A_ik * B_kj) )
+     *                = sum_k ( E[A_ik^2 * B_kj^2] - E[A_ik]^2E[B_kj^2] )
+     *                = ...
+     *                = K * (var_a * var_b + var_a * mean^2_b + var_b * mean^2_a)
+     *
+     *      Similarly, due to uniform random variable properties, we have
+     *          var_a = (max_a - min_a)^2 / 12
+     *          var_b = (max_b - min_b)^2 / 12
+     *
+     *
+     * 3) Now, we have an idea of what would an average C_ij will like and how much deviation
+     *    is present around it. The exact distribution of C is not easy to come up with dependent on K.
+     *    But, as K increases, due to Central Limit Theorem, it'll look more like a bell shaped figure,
+     *    approaching normal distribution.
+     *
+     *    This is useful because, in normal distribution, we know that values +- 2 std_deviation around
+     *    the mean constitute 95% of the values. Therefore, setting a plausible range for us:
+     *      C_range = [C_min, C_max] = [mean_c - 2 * std_c, mean_c + 2 * std_c]
+     *
+     * 4)
+     *    If we map this [C_min, C_max] to [0, 255] or [-128, 127] depending on the signedness of the
+     *    data type, we can find a suitable scale and offset for the output. On average, it's expected
+     *    that 5% of the output values will saturate and 95% will remain in the range.
+     *
+     *    The equations to be solved for offset_c and scale_c are:
+     *          C_min = scale_c * (type_min - offset_c)
+     *          C_max = scale_c * (type_max - offset_c)
+     */
+
+    const int32_t a_offset = a_q_info.uniform().offset;
+    const float   a_scale  = a_q_info.uniform().scale;
+    const int32_t b_offset = b_q_info.uniform().offset;
+    const float   b_scale  = b_q_info.uniform().scale;
+
+    // Lhs/A stats
+    const float max_a  = (t_max - a_offset) * a_scale;
+    const float min_a  = (t_min - a_offset) * a_scale;
+    const float mean_a = (max_a + min_a) / 2;
+    const float var_a  = (max_a - min_a) * (max_a - min_a) / 12;
+
+    // Rhs/B stats
+    const float max_b  = (t_max - b_offset) * b_scale;
+    const float min_b  = (t_min - b_offset) * b_scale;
+    const float mean_b = (max_b + min_b) / 2;
+    const float var_b  = (max_b - min_b) * (max_b - min_b) / 12;
+
+    // Output stats
+    const float mean_out = k * mean_a * mean_b;
+    const float var_out  = k * (var_a * var_b + var_a * mean_b * mean_b + var_b * mean_a * mean_a);
+    const float std_out  = sqrt(var_out);
+
+    // Output quantization setup
+    const float   scale_out  = 4 * std_out / 255;
+    const int32_t offset_out = static_cast<int32_t>(t_min - (mean_out - 2.f * std_out) / scale_out);
+
+    c_q_info = QuantizationInfo(scale_out, offset_out);
+    return c_q_info;
+}
+
 template void get_tile(const SimpleTensor<float> &in, SimpleTensor<float> &roi, const Coordinates &coord);
 template void get_tile(const SimpleTensor<half> &in, SimpleTensor<half> &roi, const Coordinates &coord);
 template void get_tile(const SimpleTensor<int> &in, SimpleTensor<int> &roi, const Coordinates &coord);
diff --git a/tests/validation/Helpers.h b/tests/validation/Helpers.h
index 2e48a6b8c6..3449239e45 100644
--- a/tests/validation/Helpers.h
+++ b/tests/validation/Helpers.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2022 Arm Limited.
+ * Copyright (c) 2017-2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -21,8 +21,8 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
-#ifndef ARM_COMPUTE_TEST_VALIDATION_HELPERS_H
-#define ARM_COMPUTE_TEST_VALIDATION_HELPERS_H
+#ifndef ACL_TESTS_VALIDATION_HELPERS
+#define ACL_TESTS_VALIDATION_HELPERS
 
 #include "arm_compute/core/Types.h"
 #include "arm_compute/core/Utils.h"
@@ -250,7 +250,12 @@ void add_padding_x(std::initializer_list<ITensor *> tensors, const DataLayout &d
  * @note This function adds padding to the input tensors only if data_layout == DataLayout::NHWC
  */
 void add_padding_y(std::initializer_list<ITensor *> tensors, const DataLayout &data_layout = DataLayout::NHWC);
+
+/** For MatMulLowp, given the Lhs/Rhs matrix quantization informations and the matrix multiplication dimensions,
+ *  calculate a suitable output quantization for obtaining non-saturated outputs with high probability.
+ */
+QuantizationInfo calculate_mat_mul_dst_q_info(const QuantizationInfo &lhs_q_info, const QuantizationInfo &rhs_q_info, int m, int n, int k, DataType data_type);
 } // namespace validation
 } // namespace test
 } // namespace arm_compute
-#endif /* ARM_COMPUTE_TEST_VALIDATION_HELPERS_H */
+#endif /* ACL_TESTS_VALIDATION_HELPERS */
diff --git a/tests/validation/fixtures/MatMulKernelFixture.h b/tests/validation/fixtures/MatMulKernelFixture.h
index 10e2a0659a..7d0b1a40a9 100644
--- a/tests/validation/fixtures/MatMulKernelFixture.h
+++ b/tests/validation/fixtures/MatMulKernelFixture.h
@@ -25,11 +25,15 @@
 #define ACL_TESTS_VALIDATION_FIXTURES_MATMULKERNELFIXTURE
 
 #include "arm_compute/core/KernelDescriptors.h"
-#include "src/gpu/cl/kernels/ClMatMulNativeKernel.h"
+#include "arm_compute/core/Utils.h"
+#include "arm_compute/core/utils/quantization/AsymmHelpers.h"
+
 #include "tests/CL/CLAccessor.h"
 #include "tests/CL/Helper.h"
 #include "tests/framework/Fixture.h"
+#include "tests/validation/Helpers.h"
 #include "tests/validation/reference/GEMM.h"
+#include "tests/validation/reference/GEMMLowp.h"
 #include "tests/validation/reference/Permute.h"
 #include "tests/validation/reference/ReshapeLayer.h"
 
@@ -43,14 +47,43 @@ namespace validation
 {
 using namespace arm_compute::opencl::kernels;
 
-template <typename T>
+template <typename T, typename KernelType>
 class MatMulKernelValidationFixture : public framework::Fixture
 {
 public:
     template <typename...>
-    void setup(TensorShape shape_a, TensorShape shape_b, TensorShape output_shape, bool pretranspose_a, bool pretranspose_b, const int M0, const int N0, const int K0, bool export_rhs_to_cl_image, DataType data_type)
+    void setup(TensorShape shape_a, TensorShape shape_b, TensorShape output_shape, bool pretranspose_a, bool pretranspose_b, int M0, int N0, int K0, bool export_rhs_to_cl_image, DataType data_type)
     {
         // For brevity, the input shapes are assumed to be not-transposed for both Lhs and Rhs matrices.
+        QuantizationInfo lhs_q_info;
+        QuantizationInfo rhs_q_info;
+        QuantizationInfo dst_q_info;
+
+        if(is_data_type_quantized(data_type))
+        {
+            const int32_t t_max = static_cast<int32_t>(std::numeric_limits<T>::max());
+            const int32_t t_min = static_cast<int32_t>(std::numeric_limits<T>::min());
+
+            std::mt19937                           generator(library->seed());
+            std::uniform_real_distribution<float>  distribution_float(-5.0f, 3.0f);
+            std::uniform_int_distribution<int32_t> distribution_t(t_min, t_max);
+
+            const float scale_lhs = pow(2, distribution_float(generator)); // [2^-5, 2^3]
+            const float scale_rhs = pow(2, distribution_float(generator)); // [2^-5, 2^3]
+
+            const int32_t offset_lhs = distribution_t(generator);
+            const int32_t offset_rhs = distribution_t(generator);
+
+            lhs_q_info = QuantizationInfo(scale_lhs, offset_lhs);
+            rhs_q_info = QuantizationInfo(scale_rhs, offset_rhs);
+
+            const int m = shape_a.y();
+            const int n = shape_b.x();
+            const int k = shape_a.x();
+
+            dst_q_info = calculate_mat_mul_dst_q_info(lhs_q_info, rhs_q_info, m, n, k, data_type);
+        }
+
         if(pretranspose_a)
         {
             permute(shape_a, PermutationVector(1U, 0U));
@@ -65,8 +98,8 @@ public:
 
         if(!export_rhs_to_cl_image || _device_supports_export_to_cl_image)
         {
-            _target    = compute_target(shape_a, shape_b, output_shape, pretranspose_a, pretranspose_b, M0, N0, K0, export_rhs_to_cl_image, data_type);
-            _reference = compute_reference(shape_a, shape_b, output_shape, pretranspose_a, pretranspose_b, data_type);
+            _target    = compute_target(shape_a, shape_b, output_shape, pretranspose_a, pretranspose_b, M0, N0, K0, export_rhs_to_cl_image, data_type, lhs_q_info, rhs_q_info, dst_q_info);
+            _reference = compute_reference(shape_a, shape_b, output_shape, pretranspose_a, pretranspose_b, data_type, lhs_q_info, rhs_q_info, dst_q_info);
         }
     }
 
@@ -93,23 +126,29 @@ protected:
         }
     }
 
+    template <typename U, typename D>
+    void fill_constant(U &&tensor, D value)
+    {
+        library->fill_tensor_value(tensor, value);
+    }
+
     CLTensor compute_target(const TensorShape &shape_a, const TensorShape &shape_b, const TensorShape &output_shape, bool pretranspose_a, bool pretranspose_b, const int M0, const int N0, const int K0,
-                            bool export_rhs_to_cl_image, DataType data_type)
+                            bool export_rhs_to_cl_image, DataType data_type, const QuantizationInfo &lhs_q_info, const QuantizationInfo &rhs_q_info, const QuantizationInfo &dst_q_info)
     {
-        // Create tensors
-        CLTensor a   = create_tensor<CLTensor>(shape_a, data_type, 1);
-        CLTensor b   = create_tensor<CLTensor>(shape_b, data_type, 1);
-        CLTensor dst = create_tensor<CLTensor>(output_shape, data_type, 1);
-
-        CLSynthetizeOperator<ClMatMulNativeKernel> matMul{};
-        MatMulKernelInfo                           matmul_info;
-        matmul_info.adj_lhs = pretranspose_a;
-        matmul_info.adj_rhs = pretranspose_b;
-        matmul_info.m0      = M0;
-        matmul_info.n0      = N0;
-        matmul_info.k0      = K0;
+        CLSynthetizeOperator<KernelType> matMul{};
+        MatMulKernelInfo                 matmul_info;
+        matmul_info.adj_lhs                = pretranspose_a;
+        matmul_info.adj_rhs                = pretranspose_b;
+        matmul_info.m0                     = M0;
+        matmul_info.n0                     = N0;
+        matmul_info.k0                     = K0;
         matmul_info.export_rhs_to_cl_image = export_rhs_to_cl_image;
 
+        // Create tensors
+        CLTensor a   = create_tensor<CLTensor>(shape_a, data_type, 1, lhs_q_info);
+        CLTensor b   = create_tensor<CLTensor>(shape_b, data_type, 1, rhs_q_info);
+        CLTensor dst = create_tensor<CLTensor>(output_shape, data_type, 1, dst_q_info);
+
         matMul.configure(a.info(), b.info(), dst.info(), matmul_info);
         ARM_COMPUTE_ASSERT(a.info()->is_resizable());
         ARM_COMPUTE_ASSERT(b.info()->is_resizable());
@@ -138,18 +177,19 @@ protected:
         return dst;
     }
 
-    SimpleTensor<T> compute_reference(const TensorShape &shape_a, const TensorShape &shape_b, const TensorShape &output_shape, bool pretranspose_a, bool pretranspose_b, DataType data_type)
+    SimpleTensor<T> compute_reference(const TensorShape &shape_a, const TensorShape &shape_b, const TensorShape &output_shape, bool pretranspose_a, bool pretranspose_b, DataType data_type,
+                                      const QuantizationInfo &lhs_q_info, const QuantizationInfo &rhs_q_info, const QuantizationInfo &dst_q_info)
     {
         // We collapse dimensions > 3 onto dimension 3, i.e. 5D+ tensors will look like 4D
         // This is necessary unless we choose to extend gemm reference for 5D+ tensors
-        TensorShape output_shape_collapsed = output_shape.collapsed_from(Window::DimW);
-        TensorShape shape_a_collapsed      = shape_a.collapsed_from(Window::DimW);
-        TensorShape shape_b_collapsed      = shape_b.collapsed_from(Window::DimW);
+        TensorShape output_shape_collapsed = output_shape.collapsed_from(Window::DimZ);
+        TensorShape shape_a_collapsed      = shape_a.collapsed_from(Window::DimZ);
+        TensorShape shape_b_collapsed      = shape_b.collapsed_from(Window::DimZ);
 
         // Create reference
-        SimpleTensor<T> a{ shape_a_collapsed, data_type, 1 };
-        SimpleTensor<T> b{ shape_b_collapsed, data_type, 1 };
-        SimpleTensor<T> c{ output_shape_collapsed, data_type, 1 };
+        SimpleTensor<T> a{ shape_a_collapsed, data_type, 1, lhs_q_info };
+        SimpleTensor<T> b{ shape_b_collapsed, data_type, 1, rhs_q_info };
+        SimpleTensor<T> c{ output_shape_collapsed, data_type, 1, dst_q_info };
 
         // Fill reference
         fill(a, 0);
@@ -185,10 +225,8 @@ protected:
             b_transposed = reference::permute<T>(b, PermutationVector(1U, 0U));
         }
 
-        // Setting beta to 0 will effectively disable C for the
-        // computation of the reference: alpha * A * B + 0 * C
         // Use transposed tensors if boolean enabled else use original tensors
-        SimpleTensor<T> result = reference::gemm<T>((pretranspose_a) ? a_transposed : a, (pretranspose_b) ? b_transposed : b, c, 1.0f, 0.f);
+        SimpleTensor<T> result = gemm_reference<T>((pretranspose_a) ? a_transposed : a, (pretranspose_b) ? b_transposed : b, c);
 
         // We reshape the gemm output back if the tensor is high dimensional
         if(output_shape_collapsed != output_shape)
@@ -199,9 +237,43 @@ protected:
         return result;
     }
 
+    template <typename U = T>
+    typename std::enable_if < std::is_same<U, float>::value || std::is_same<U, half>::value, SimpleTensor<U >>::type gemm_reference(SimpleTensor<U> &a, SimpleTensor<U> &b, SimpleTensor<U> &c)
+    {
+        // Setting beta to 0 will effectively disable C for the
+        // computation of the reference: alpha * A * B + 0 * C
+        return reference::gemm<U>(a, b, c, 1.0f, 0.f);
+    }
+
+    template <typename U = T>
+    typename std::enable_if < std::is_same<U, int8_t>::value || std::is_same<U, uint8_t>::value, SimpleTensor<U >>::type gemm_reference(SimpleTensor<U> &a, SimpleTensor<U> &b, SimpleTensor<U> &c)
+    {
+        const UniformQuantizationInfo aq = a.quantization_info().uniform();
+        const UniformQuantizationInfo bq = b.quantization_info().uniform();
+        const UniformQuantizationInfo cq = c.quantization_info().uniform();
+
+        const SimpleTensor<int32_t> result = reference::gemmlowp_matrix_multiply_core<int32_t, U, U>(a, b, c.shape(), -aq.offset, -bq.offset);
+
+        std::vector<int32_t> gemmlowp_multipliers{ 1 };
+        std::vector<int32_t> gemmlowp_shifts{ 1 };
+        const int            gemmlowp_offset = cq.offset;
+        const float          scale           = aq.scale * bq.scale / cq.scale;
+
+        quantization::calculate_quantized_multiplier(scale, &gemmlowp_multipliers[0], &gemmlowp_shifts[0]);
+        constexpr int32_t gemmlowp_min_bound = std::numeric_limits<int32_t>::min();
+        constexpr int32_t gemmlowp_max_bound = std::numeric_limits<int32_t>::max();
+
+        SimpleTensor<int> bias{ c.shape(), DataType::S32 };
+        fill_constant(bias, static_cast<int32_t>(0));
+
+        const SimpleTensor<U> final_result = reference::gemmlowp_quantize_down_scale_by_fixedpoint<int32_t, U>(result, bias,
+                                                                                                               gemmlowp_multipliers, gemmlowp_shifts, gemmlowp_offset, gemmlowp_min_bound, gemmlowp_max_bound);
+        return final_result;
+    }
+
     CLTensor        _target{};
     SimpleTensor<T> _reference{};
-    bool _device_supports_export_to_cl_image { true };
+    bool            _device_supports_export_to_cl_image{ true };
 };
 
 } // namespace validation