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>

2 files changed, 486 insertions, 137 deletions

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