diff --git a/tests/datasets/GEMMLowpFusedOffsetOutputDataset.h b/tests/datasets/GEMMLowpFusedOffsetOutputDataset.h
index 7ab068c..b87b2fa 100644
--- a/tests/datasets/GEMMLowpFusedOffsetOutputDataset.h
+++ b/tests/datasets/GEMMLowpFusedOffsetOutputDataset.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -167,10 +167,10 @@
 public:
     SmallGEMMLowpFusedOffsetOutputUint8Dataset()
     {
-        add_config(TensorShape(21U, 13U), TensorShape(1U, 21U), TensorShape(1U, 13U), 0, 0, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN, -100, 2, 13, 10, 210));
-        add_config(TensorShape(52U, 13U), TensorShape(33U, 52U), TensorShape(33U, 13U), 0, 4, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN, 100, 2, 13, 10, 210));
-        add_config(TensorShape(31U, 27U), TensorShape(23U, 31U), TensorShape(23U, 27U), 18, 23, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN, 200, 2, 13, 10, 210));
-        add_config(TensorShape(32U, 72U), TensorShape(16U, 32U), TensorShape(16U, 72U), -9, 1, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN, -100, 2, 13, 10, 210));
+        add_config(TensorShape(21U, 13U), TensorShape(1U, 21U), TensorShape(1U, 13U), 0, 0, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT, -100, 2, 13, 10, 210));
+        add_config(TensorShape(52U, 13U), TensorShape(33U, 52U), TensorShape(33U, 13U), 0, 4, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT, 100, 2, 13, 10, 210));
+        add_config(TensorShape(31U, 27U), TensorShape(23U, 31U), TensorShape(23U, 27U), 18, 23, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT, 200, 2, 13, 10, 210));
+        add_config(TensorShape(32U, 72U), TensorShape(16U, 32U), TensorShape(16U, 72U), -9, 1, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT, -100, 2, 13, 10, 210));
 
         add_config(TensorShape(21U, 1U), TensorShape(43U, 21U), TensorShape(43U, 1U), 0, 0, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT, -2, 254601600, 10, 10, 210));
         add_config(TensorShape(31U, 3U), TensorShape(72U, 31U), TensorShape(72U, 3U), -2, 13, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT, 0, 254601600, 10, 10, 210));
@@ -179,6 +179,38 @@
     }
 };
 
+class SmallGEMMLowpFusedOffsetOutputOutput3DUint8Dataset final : public GEMMLowpFusedOffsetOutputDataset
+{
+public:
+    SmallGEMMLowpFusedOffsetOutputOutput3DUint8Dataset()
+    {
+        add_config(TensorShape(21U, 1421U, 33U), TensorShape(34U, 21U), TensorShape(34U, 7U, 203U, 33U), 0, 0, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT, -100, 2, 13, 10, 210));
+        add_config(TensorShape(31U, 102U, 55U), TensorShape(23U, 31U), TensorShape(23U, 1U, 102U, 55U), 0, 4, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT, 100, 2, 13, 10, 210));
+        add_config(TensorShape(38U, 1200U, 77U), TensorShape(21U, 38U), TensorShape(21U, 4U, 300U, 77U), 18, 23, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT, 200, 2, 13, 10, 210));
+        add_config(TensorShape(32U, 103U, 99U), TensorShape(17U, 32U), TensorShape(17U, 1U, 103U, 99U), -9, 1, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT, -100, 2, 13, 10, 210));
+        add_config(TensorShape(16U, 1600U, 111U), TensorShape(8U, 16U), TensorShape(8U, 8U, 200U, 111U), 0, 0, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT, -2, 254601600, 10, 10,
+                                                                                                                               210));
+        add_config(TensorShape(16U, 1600U, 113U), TensorShape(8U, 16U), TensorShape(8U, 8U, 200U, 113U), -2, 13, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT, 0, 254601600, 10, 10,
+                   210));
+    }
+};
+
+class SmallGEMMLowpFusedOffsetOutputInputOutput3DUint8Dataset final : public GEMMLowpFusedOffsetOutputDataset
+{
+public:
+    SmallGEMMLowpFusedOffsetOutputInputOutput3DUint8Dataset()
+    {
+        add_config(TensorShape(21U, 7U, 203U, 33U), TensorShape(34U, 21U), TensorShape(34U, 7U, 203U, 33U), 0, 0, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT, -100, 2, 13, 10, 210));
+        add_config(TensorShape(31U, 1U, 102U, 55U), TensorShape(23U, 31U), TensorShape(23U, 1U, 102U, 55U), 0, 4, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT, 100, 2, 13, 10, 210));
+        add_config(TensorShape(38U, 4U, 300U, 77U), TensorShape(21U, 38U), TensorShape(21U, 4U, 300U, 77U), 18, 23, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT, 200, 2, 13, 10, 210));
+        add_config(TensorShape(32U, 1U, 103U, 99U), TensorShape(17U, 32U), TensorShape(17U, 1U, 103U, 99U), -9, 1, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT, -100, 2, 13, 10, 210));
+        add_config(TensorShape(16U, 8U, 200U, 111U), TensorShape(8U, 16U), TensorShape(8U, 8U, 200U, 111U), 0, 0, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT, -2, 254601600, 10, 10,
+                   210));
+        add_config(TensorShape(16U, 8U, 200U, 113U), TensorShape(8U, 16U), TensorShape(8U, 8U, 200U, 113U), -2, 13, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT, 0, 254601600, 10, 10,
+                   210));
+    }
+};
+
 class SmallGEMMLowpFusedOffsetOutputInt8Dataset final : public GEMMLowpFusedOffsetOutputDataset
 {
 public:
@@ -214,10 +246,10 @@
 public:
     LargeGEMMLowpFusedOffsetOutputUint8Dataset()
     {
-        add_config(TensorShape(923U, 429U), TensorShape(871U, 923U), TensorShape(871U, 429U), 0, 0, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN, -100, 2, 18, 10, 210));
-        add_config(TensorShape(873U, 513U), TensorShape(784U, 873U), TensorShape(784U, 513U), 0, 4, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN, 100, 2, 18, 10, 210));
-        add_config(TensorShape(1021U, 973U), TensorShape(783U, 1021U), TensorShape(783U, 973U), 5, 13, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN, 200, 2, 18, 10, 210));
-        add_config(TensorShape(941U, 1011U), TensorShape(623U, 941U), TensorShape(623U, 1011U), -9, 1, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN, -100, 2, 18, 10, 210));
+        add_config(TensorShape(923U, 429U), TensorShape(871U, 923U), TensorShape(871U, 429U), 0, 0, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT, -100, 2, 18, 10, 210));
+        add_config(TensorShape(873U, 513U), TensorShape(784U, 873U), TensorShape(784U, 513U), 0, 4, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT, 100, 2, 18, 10, 210));
+        add_config(TensorShape(1021U, 973U), TensorShape(783U, 1021U), TensorShape(783U, 973U), 5, 13, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT, 200, 2, 18, 10, 210));
+        add_config(TensorShape(941U, 1011U), TensorShape(623U, 941U), TensorShape(623U, 1011U), -9, 1, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT, -100, 2, 18, 10, 210));
 
         add_config(TensorShape(923U, 429U), TensorShape(871U, 923U), TensorShape(871U, 429U), 0, 0, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT, -1, 254601600, 15, 10, 210));
         add_config(TensorShape(873U, 513U), TensorShape(784U, 873U), TensorShape(784U, 513U), 0, 4, OutputStageInfo(GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT, 1, 254601600, 15, 10, 210));
diff --git a/tests/validation/CL/GEMMLowp.cpp b/tests/validation/CL/GEMMLowp.cpp
index 5a1971b..1c7446f 100644
--- a/tests/validation/CL/GEMMLowp.cpp
+++ b/tests/validation/CL/GEMMLowp.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2020 Arm Limited.
+ * Copyright (c) 2017-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -68,16 +68,44 @@
 
 TEST_SUITE(FusedOffsetOutput)
 TEST_SUITE(QASYMM8)
-using CLGEMMLowpMatrixMultiplyCoreFusedOffsetOutputUint8Fixture = GEMMLowpMatrixMultiplyCoreFusedOffsetOutputValidationFixture<CLTensor, CLAccessor, CLGEMMLowpMatrixMultiplyCore>;
-FIXTURE_DATA_TEST_CASE(RunSmall, CLGEMMLowpMatrixMultiplyCoreFusedOffsetOutputUint8Fixture, framework::DatasetMode::ALL, combine(datasets::SmallGEMMLowpFusedOffsetOutputUint8Dataset(),
-                       framework::dataset::make("DataType", { DataType::QASYMM8 })))
+using CLGEMMLowpMatrixMultiplyCoreFusedOffsetOutputUint8Fixture = GEMMLowpMatrixMultiplyCoreFusedOffsetOutputGenericValidationFixture<CLTensor, CLAccessor, CLGEMMLowpMatrixMultiplyCore>;
+FIXTURE_DATA_TEST_CASE(RunSmall, CLGEMMLowpMatrixMultiplyCoreFusedOffsetOutputUint8Fixture, framework::DatasetMode::ALL, combine(combine(datasets::SmallGEMMLowpFusedOffsetOutputUint8Dataset(),
+                       framework::dataset::make("DataType", { DataType::QASYMM8 })),
+                       framework::dataset::make("reshape_b_only_on_first_run", { true, false })))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_quant);
 }
 
-FIXTURE_DATA_TEST_CASE(RunLarge, CLGEMMLowpMatrixMultiplyCoreFusedOffsetOutputUint8Fixture, framework::DatasetMode::NIGHTLY, combine(datasets::LargeGEMMLowpFusedOffsetOutputUint8Dataset(),
-                       framework::dataset::make("DataType", { DataType::QASYMM8 })))
+TEST_SUITE(Output3D)
+using CLGEMMLowpMatrixMultiplyCoreFusedOffsetOutputOutput3DUint8Fixture =
+    GEMMLowpMatrixMultiplyCoreFusedOffsetOutputGenericValidationFixture<CLTensor, CLAccessor, CLGEMMLowpMatrixMultiplyCore, false, true>;
+FIXTURE_DATA_TEST_CASE(RunSmall, CLGEMMLowpMatrixMultiplyCoreFusedOffsetOutputOutput3DUint8Fixture, framework::DatasetMode::ALL,
+                       combine(combine(datasets::SmallGEMMLowpFusedOffsetOutputOutput3DUint8Dataset(),
+                                       framework::dataset::make("DataType", { DataType::QASYMM8 })),
+                               framework::dataset::make("reshape_b_only_on_first_run", { true, false })))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_quant);
+}
+TEST_SUITE_END() // Output3D
+
+TEST_SUITE(InputOutput3D)
+using CLGEMMLowpMatrixMultiplyCoreFusedOffsetOutputInputOutput3DUint8Fixture =
+    GEMMLowpMatrixMultiplyCoreFusedOffsetOutputGenericValidationFixture<CLTensor, CLAccessor, CLGEMMLowpMatrixMultiplyCore, true, true>;
+FIXTURE_DATA_TEST_CASE(RunSmall, CLGEMMLowpMatrixMultiplyCoreFusedOffsetOutputInputOutput3DUint8Fixture, framework::DatasetMode::ALL,
+                       combine(combine(datasets::SmallGEMMLowpFusedOffsetOutputInputOutput3DUint8Dataset(),
+                                       framework::dataset::make("DataType", { DataType::QASYMM8 })),
+                               framework::dataset::make("reshape_b_only_on_first_run", { true, false })))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_quant);
+}
+TEST_SUITE_END() // InputOutput3D
+
+FIXTURE_DATA_TEST_CASE(RunLarge, CLGEMMLowpMatrixMultiplyCoreFusedOffsetOutputUint8Fixture, framework::DatasetMode::NIGHTLY, combine(combine(datasets::LargeGEMMLowpFusedOffsetOutputUint8Dataset(),
+                       framework::dataset::make("DataType", { DataType::QASYMM8 })),
+                       framework::dataset::make("reshape_b_only_on_first_run", { true, false })))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_quant);
diff --git a/tests/validation/fixtures/GEMMLowpFixture.h b/tests/validation/fixtures/GEMMLowpFixture.h
index 5cf210b..ab9d35d 100644
--- a/tests/validation/fixtures/GEMMLowpFixture.h
+++ b/tests/validation/fixtures/GEMMLowpFixture.h
@@ -97,7 +97,7 @@
 template <typename TensorType, typename AccessorType, typename FunctionType, bool reinterpret_input_as_3d, bool reinterpret_output_as_3d, typename OutputType, bool is_fused = false>
 TensorType compute_gemmlowp_target(const TensorShape &shape_a, const TensorShape &shape_b, const TensorShape &shape_output, int32_t a_offset, int32_t b_offset,
                                    GEMMLowpOutputStageInfo output_stage = GEMMLowpOutputStageInfo(), DataType data_type_a = DataType::QASYMM8, DataType data_type_b = DataType::QASYMM8,
-                                   QuantizationInfo b_qinfo = QuantizationInfo())
+                                   QuantizationInfo b_qinfo = QuantizationInfo(), bool reshape_b_only_on_first_run = false)
 {
     // Create tensors
     DataType data_type_output = output_stage.type == GEMMLowpOutputStageType::NONE ? DataType::S32 : data_type_a;
@@ -126,7 +126,8 @@
     // Create and configure function
     // The GEMMinfo includes the values of the depth in case of reinterpreted 3d input/output
     FunctionType gemmlowp;
-    gemmlowp.configure(&a, &b, is_fused ? &bias : nullptr, &output, GEMMInfo(false, false, false, (reinterpret_output_as_3d ? shape_output[2] : 0), reinterpret_input_as_3d, false, output_stage));
+    gemmlowp.configure(&a, &b, is_fused ? &bias : nullptr, &output, GEMMInfo(false, false, reshape_b_only_on_first_run, (reinterpret_output_as_3d ? shape_output[2] : 0), reinterpret_input_as_3d, false,
+                                                                             output_stage));
 
     ARM_COMPUTE_ASSERT(a.info()->is_resizable());
     ARM_COMPUTE_ASSERT(b.info()->is_resizable());
@@ -208,11 +209,12 @@
 };
 
 template <typename TensorType, typename AccessorType, typename FunctionType, bool reinterpret_input_as_3d = false, bool reinterpret_output_as_3d = false, typename TI = uint8_t, typename TW = uint8_t>
-class GEMMLowpMatrixMultiplyCoreFusedOffsetOutputValidationFixture : public framework::Fixture
+class GEMMLowpMatrixMultiplyCoreFusedOffsetOutputGenericValidationFixture : public framework::Fixture
 {
 public:
     template <typename...>
-    void setup(TensorShape shape_a, TensorShape shape_b, TensorShape shape_output, int32_t a_offset, int32_t b_offset, GEMMLowpOutputStageInfo output_stage, DataType data_type_b)
+    void setup(TensorShape shape_a, TensorShape shape_b, TensorShape shape_output, int32_t a_offset, int32_t b_offset, GEMMLowpOutputStageInfo output_stage, DataType data_type_b,
+               bool reshape_b_only_on_first_run)
     {
         ARM_COMPUTE_ASSERT(output_stage.type != GEMMLowpOutputStageType::NONE);
         DataType data_type_a = data_type_b == DataType::QASYMM8_SIGNED ? DataType::QASYMM8_SIGNED : DataType::QASYMM8;
@@ -232,21 +234,21 @@
             }
 
             _reference = compute_reference(shape_a, shape_b, shape_output, a_offset, 0, output_stage, data_type_a, data_type_b, QuantizationInfo(scales));
-            _target    = compute_target(shape_a, shape_b, shape_output, a_offset, 0, output_stage, data_type_a, data_type_b, QuantizationInfo(scales));
+            _target    = compute_target(shape_a, shape_b, shape_output, a_offset, 0, output_stage, data_type_a, data_type_b, QuantizationInfo(scales), reshape_b_only_on_first_run);
         }
         else
         {
             _reference = compute_reference(shape_a, shape_b, shape_output, a_offset, b_offset, output_stage, data_type_a, data_type_b, QuantizationInfo());
-            _target    = compute_target(shape_a, shape_b, shape_output, a_offset, b_offset, output_stage, data_type_a, data_type_b, QuantizationInfo());
+            _target    = compute_target(shape_a, shape_b, shape_output, a_offset, b_offset, output_stage, data_type_a, data_type_b, QuantizationInfo(), reshape_b_only_on_first_run);
         }
     }
 
 protected:
     TensorType compute_target(const TensorShape &shape_a, const TensorShape &shape_b, const TensorShape &shape_output, int32_t a_offset, int32_t b_offset, GEMMLowpOutputStageInfo output_stage,
-                              DataType data_type_a, DataType data_type_b, QuantizationInfo b_qinfo)
+                              DataType data_type_a, DataType data_type_b, QuantizationInfo b_qinfo, bool reshape_b_only_on_first_run = false)
     {
         return compute_gemmlowp_target<TensorType, AccessorType, FunctionType, reinterpret_input_as_3d, reinterpret_output_as_3d, qasymm8_t, true>(shape_a, shape_b, shape_output, a_offset, b_offset,
-                output_stage, data_type_a, data_type_b, b_qinfo);
+                output_stage, data_type_a, data_type_b, b_qinfo, reshape_b_only_on_first_run);
     }
 
     SimpleTensor<TI> compute_reference(const TensorShape &shape_a, const TensorShape &shape_b, const TensorShape &shape_output, int32_t a_offset, int32_t b_offset,
@@ -277,6 +279,19 @@
     SimpleTensor<TI> _reference{};
 };
 
+template <typename TensorType, typename AccessorType, typename FunctionType, bool reinterpret_input_as_3d = false, bool reinterpret_output_as_3d = false, typename TI = uint8_t, typename TW = uint8_t>
+class GEMMLowpMatrixMultiplyCoreFusedOffsetOutputValidationFixture : public
+    GEMMLowpMatrixMultiplyCoreFusedOffsetOutputGenericValidationFixture<TensorType, AccessorType, FunctionType, reinterpret_input_as_3d, reinterpret_output_as_3d, TI, TW>
+{
+public:
+    template <typename...>
+    void setup(TensorShape shape_a, TensorShape shape_b, TensorShape shape_output, int32_t a_offset, int32_t b_offset, GEMMLowpOutputStageInfo output_stage, DataType data_type_b)
+    {
+        GEMMLowpMatrixMultiplyCoreFusedOffsetOutputGenericValidationFixture<TensorType, AccessorType, FunctionType, reinterpret_input_as_3d, reinterpret_output_as_3d, TI, TW>::setup(shape_a, shape_b,
+                shape_output, a_offset, b_offset, output_stage, data_type_b, false);
+    }
+};
+
 template <typename TensorType, typename AccessorType, typename FunctionType>
 class GEMMLowpQuantizeDownInt32ToUint8ScaleValidationFixture : public framework::Fixture
 {