*******************************************************************************
 Copyright 2023 Arm Limited and affiliates.
 SPDX-License-Identifier: Apache-2.0

 Licensed under the Apache License, Version 2.0 (the "License");
 you may not use this file except in compliance with the License.
 You may obtain a copy of the License at

     http://www.apache.org/licenses/LICENSE-2.0

 Unless required by applicable law or agreed to in writing, software
 distributed under the License is distributed on an "AS IS" BASIS,
 WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License.
 *******************************************************************************
diff --git a/src/cpu/aarch64/acl_convolution_utils.cpp b/src/cpu/aarch64/acl_convolution_utils.cpp
index 6b57374643..85e45ace9d 100644
--- a/src/cpu/aarch64/acl_convolution_utils.cpp
+++ b/src/cpu/aarch64/acl_convolution_utils.cpp
@@ -48,11 +48,14 @@ status_t acl_init_conf(acl_conv_conf_t &acp, memory_desc_t &src_md,
     if (!is_fwd) return status::unimplemented;
 
     const int ndims = src_d.ndims();
+    const bool is_depthwise = wei_d.ndims() == 5 && wei_d.dims()[1] == 1
+            && wei_d.dims()[2] == 1;
 
-    ACL_CHECK_SUPPORT(ndims != 4, " only supports 2 spatial dimensions");
+    ACL_CHECK_SUPPORT(
+            ndims != 4 && !is_depthwise, " only supports 2 spatial dimensions");
 
     const int with_groups = wei_d.ndims() == src_d.ndims() + 1;
-    ACL_CHECK_SUPPORT(with_groups, " does not support groups");
+    ACL_CHECK_SUPPORT(with_groups && !is_depthwise, " does not support groups");
 
     ACL_CHECK_SUPPORT(src_d.data_type() != data_type::f32
                     || wei_d.data_type() != data_type::f32
@@ -108,7 +111,8 @@ status_t acl_init_conf(acl_conv_conf_t &acp, memory_desc_t &src_md,
 
     acp.with_bias = cd.bias_desc.format_kind != format_kind::undef;
 
-    if (wei_d.format_kind() != format_kind::any) return status::unimplemented;
+    if (wei_d.format_kind() != format_kind::any && !is_depthwise)
+        return status::unimplemented;
 
     auto src_tag = memory_desc_matches_one_of_tag(
             src_md, format_tag::nhwc, format_tag::nchw);
@@ -138,8 +142,12 @@ status_t acl_init_conf(acl_conv_conf_t &acp, memory_desc_t &src_md,
             || src_tag != dst_tag)
         return status::unimplemented;
 
-    // Set weights to initially be the same as src
-    CHECK(memory_desc_init_by_tag(weights_md, src_tag));
+    if (is_depthwise) {
+        CHECK(memory_desc_init_by_tag(weights_md, format_tag::hwigo));
+    } else {
+        // Set weights to initially be the same as src
+        CHECK(memory_desc_init_by_tag(weights_md, src_tag));
+    }
 
     // Bias is just 1D, set to be the obvious format
     if (acp.with_bias && bias_md.format_kind == format_kind::any)
@@ -166,6 +174,11 @@ status_t acl_init_conf(acl_conv_conf_t &acp, memory_desc_t &src_md,
             1,
             acl_data_type,
             acl_layout);
+    if(is_depthwise) {
+       // We need to set that values are not constant so that we
+       // we can update them in-place in ACL
+      acp.wei_tensor_info.set_are_values_constant(false);
+    }
 
     acp.dst_tensor_info = arm_compute::TensorInfo(
             is_nhwc ? arm_compute::TensorShape(oc, ow, oh, mb) :
@@ -185,6 +198,11 @@ status_t acl_init_conf(acl_conv_conf_t &acp, memory_desc_t &src_md,
     // Are we allowed to cast down to bf16 or not?
     acp.fast_math
             = one_of(attr.fpmath_mode_, fpmath_mode::bf16, fpmath_mode::any);
+    if (is_depthwise) {
+        // There is no support for fixed format kernels for depthwise convolution
+        // in ACL so we are going to use weight format that we set up earlier
+        return status::success;
+    }
 
     // WeightFormat::ANY tells ACL we can handle any format
     acp.weights_info = arm_compute::WeightsInfo(
@@ -252,6 +270,7 @@ status_t init_conf_gemm(acl_conv_conf_t &acp, memory_desc_t &src_md,
         memory_desc_t &weights_md, memory_desc_t &dst_md,
         memory_desc_t &bias_md, const convolution_desc_t &cd,
         const primitive_attr_t &attr) {
+    if (weights_md.ndims != 4) return status::unimplemented;
 
     // General Compute Library checks, memory tags are also set there
     CHECK(acl_init_conf(acp, src_md, weights_md, dst_md, bias_md, cd, attr));
@@ -277,6 +296,7 @@ status_t init_conf_indirect_gemm(acl_conv_conf_t &acp, memory_desc_t &src_md,
         memory_desc_t &weights_md, memory_desc_t &dst_md,
         memory_desc_t &bias_md, const convolution_desc_t &cd,
         const primitive_attr_t &attr) {
+    if (weights_md.ndims != 4) return status::unimplemented;
 
     // Indirect is slower for small convolution kernels
     if (weights_md.dims[2] == 1 && weights_md.dims[3] == 1)
@@ -314,6 +334,22 @@ status_t init_conf_indirect_gemm(acl_conv_conf_t &acp, memory_desc_t &src_md,
     return status::success;
 }
 
+status_t init_conf_depthwise(acl_conv_conf_t &acp, memory_desc_t &src_md,
+        memory_desc_t &weights_md, memory_desc_t &dst_md,
+        memory_desc_t &bias_md, const convolution_desc_t &cd,
+        const primitive_attr_t &attr) {
+    if (weights_md.ndims != 5) return status::unimplemented;
+
+    CHECK(acl_init_conf(acp, src_md, weights_md, dst_md, bias_md, cd, attr));
+
+    ACL_CHECK_VALID(arm_compute::NEDepthwiseConvolutionLayer::validate(
+            &acp.src_tensor_info, &acp.wei_tensor_info,
+            acp.with_bias ? &acp.bia_tensor_info : nullptr,
+            &acp.dst_tensor_info, acp.padstride_info));
+
+    return status::success;
+}
+
 } // namespace acl_convolution_utils
 
 } // namespace aarch64
diff --git a/src/cpu/aarch64/acl_convolution_utils.hpp b/src/cpu/aarch64/acl_convolution_utils.hpp
index e3d40a5e75..1ded5826c4 100644
--- a/src/cpu/aarch64/acl_convolution_utils.hpp
+++ b/src/cpu/aarch64/acl_convolution_utils.hpp
@@ -66,6 +66,11 @@ status_t init_conf_indirect_gemm(acl_conv_conf_t &acp, memory_desc_t &src_md,
         memory_desc_t &bias_md, const convolution_desc_t &cd,
         const primitive_attr_t &attr);
 
+status_t init_conf_depthwise(acl_conv_conf_t &acp, memory_desc_t &src_md,
+        memory_desc_t &weights_md, memory_desc_t &dst_md,
+        memory_desc_t &bias_md, const convolution_desc_t &cd,
+        const primitive_attr_t &attr);
+
 } // namespace acl_convolution_utils
 
 template <typename conv_obj_t, typename conv_pd_t, typename src_data_t,
diff --git a/src/cpu/aarch64/acl_depthwise_convolution.cpp b/src/cpu/aarch64/acl_depthwise_convolution.cpp
new file mode 100644
index 0000000000..70ae6bceea
--- /dev/null
+++ b/src/cpu/aarch64/acl_depthwise_convolution.cpp
@@ -0,0 +1,42 @@
+/*******************************************************************************
+* Copyright 2023 Arm Ltd. and affiliates
+*
+* Licensed under the Apache License, Version 2.0 (the "License");
+* you may not use this file except in compliance with the License.
+* You may obtain a copy of the License at
+*
+*     http://www.apache.org/licenses/LICENSE-2.0
+*
+* Unless required by applicable law or agreed to in writing, software
+* distributed under the License is distributed on an "AS IS" BASIS,
+* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+* See the License for the specific language governing permissions and
+* limitations under the License.
+*******************************************************************************/
+
+#include "cpu/aarch64/acl_depthwise_convolution.hpp"
+
+namespace dnnl {
+namespace impl {
+namespace cpu {
+namespace aarch64 {
+
+status_t acl_depthwise_convolution_fwd_t::execute_forward(
+        const exec_ctx_t &ctx) const {
+    std::lock_guard<std::mutex> _lock {this->mtx};
+
+    auto *acl_resource
+            = ctx.get_resource_mapper()
+                      ->get<acl_depthwise_convolution_resource_t>(this);
+    acl_obj_t<arm_compute::NEDepthwiseConvolutionLayer> &acl_depthwise_obj
+            = acl_resource->get_acl_obj();
+
+    return execute_forward_conv_acl<
+            acl_obj_t<arm_compute::NEDepthwiseConvolutionLayer>, pd_t, data_t>(
+            ctx, acl_depthwise_obj, pd());
+}
+
+} // namespace aarch64
+} // namespace cpu
+} // namespace impl
+} // namespace dnnl
diff --git a/src/cpu/aarch64/acl_depthwise_convolution.hpp b/src/cpu/aarch64/acl_depthwise_convolution.hpp
new file mode 100644
index 0000000000..3e3d02cf41
--- /dev/null
+++ b/src/cpu/aarch64/acl_depthwise_convolution.hpp
@@ -0,0 +1,141 @@
+/*******************************************************************************
+* Copyright 2023 Arm Ltd. and affiliates
+*
+* Licensed under the Apache License, Version 2.0 (the "License");
+* you may not use this file except in compliance with the License.
+* You may obtain a copy of the License at
+*
+*     http://www.apache.org/licenses/LICENSE-2.0
+*
+* Unless required by applicable law or agreed to in writing, software
+* distributed under the License is distributed on an "AS IS" BASIS,
+* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+* See the License for the specific language governing permissions and
+* limitations under the License.
+*******************************************************************************/
+
+#ifndef CPU_AARCH64_ACL_DEPTHWISE_CONVOLUTION_HPP
+#define CPU_AARCH64_ACL_DEPTHWISE_CONVOLUTION_HPP
+
+#include "cpu/aarch64/acl_convolution_utils.hpp"
+#include "cpu/cpu_convolution_pd.hpp"
+
+namespace dnnl {
+namespace impl {
+namespace cpu {
+namespace aarch64 {
+
+struct acl_depthwise_convolution_resource_t : public resource_t {
+    acl_depthwise_convolution_resource_t()
+        : acl_obj_(utils::make_unique<
+                acl_obj_t<arm_compute::NEDepthwiseConvolutionLayer>>()) {}
+
+    status_t configure(const acl_conv_conf_t &acp) {
+        if (!acl_obj_) return status::out_of_memory;
+
+        acl_obj_->src_tensor.allocator()->init(acp.src_tensor_info);
+        acl_obj_->wei_tensor.allocator()->init(acp.wei_tensor_info);
+        acl_obj_->dst_tensor.allocator()->init(acp.dst_tensor_info);
+        acl_obj_->bia_tensor.allocator()->init(acp.bia_tensor_info);
+
+        // clang-format off
+        acl_obj_->conv.configure(
+            &acl_obj_->src_tensor,
+            &acl_obj_->wei_tensor,
+            acp.with_bias ? &acl_obj_->bia_tensor : nullptr,
+            &acl_obj_->dst_tensor,
+            acp.padstride_info,
+            1, // depth multiplier default value
+            acp.act_info);
+
+        // clang-format on
+        return status::success;
+    }
+
+    acl_obj_t<arm_compute::NEDepthwiseConvolutionLayer> &get_acl_obj() const {
+        return *acl_obj_;
+    }
+
+    DNNL_DISALLOW_COPY_AND_ASSIGN(acl_depthwise_convolution_resource_t);
+
+private:
+    std::unique_ptr<acl_obj_t<arm_compute::NEDepthwiseConvolutionLayer>>
+            acl_obj_;
+};
+
+struct acl_depthwise_convolution_fwd_t : public primitive_t {
+
+    struct pd_t : public cpu_convolution_fwd_pd_t {
+        pd_t(const convolution_desc_t *adesc, const primitive_attr_t *attr,
+                const typename pd_t::base_class *hint_fwd_pd)
+            : cpu_convolution_fwd_pd_t(adesc, attr, hint_fwd_pd), acp_() {}
+
+        DECLARE_COMMON_PD_T("depthwise_convolution:acl",
+                acl_depthwise_convolution_fwd_t, USE_GLOBAL_SCRATCHPAD);
+
+        status_t init(engine_t *engine) {
+            using namespace data_type;
+
+            const bool is_fp16_ok = expect_data_types(f16, f16, f16, f16, undef)
+                    && attr()->has_default_values(
+                            primitive_attr_t::skip_mask_t::post_ops, f16);
+            const bool is_fp32_ok = expect_data_types(f32, f32, f32, f32, undef)
+                    && attr()->has_default_values(
+                            primitive_attr_t::skip_mask_t::post_ops, f32);
+            bool ok = is_fwd()
+                    && set_default_alg_kind(alg_kind::convolution_direct)
+                    && utils::one_of(true, is_fp16_ok, is_fp32_ok)
+                    && !has_zero_dim_memory();
+            if (!ok) return status::unimplemented;
+
+            CHECK(acl_convolution_utils::init_conf_depthwise(acp_, src_md_,
+                    weights_md_, dst_md_, bias_md_, *desc(), *attr()));
+
+            CHECK(post_ops.init(
+                    engine, attr_.post_ops_, dst_md_, acp_.act_info));
+            acp_.use_dst_acc = post_ops.has_sum();
+
+            return status::success;
+        }
+
+        acl_conv_conf_t acp_;
+
+        acl_post_ops_t post_ops;
+    };
+
+    acl_depthwise_convolution_fwd_t(const pd_t *apd) : primitive_t(apd) {}
+
+    status_t create_resource(
+            engine_t *engine, resource_mapper_t &mapper) const override {
+        if (mapper.has_resource(this)) return status::success;
+
+        auto r = utils::make_unique<acl_depthwise_convolution_resource_t>();
+        if (!r) return status::out_of_memory;
+
+        CHECK(r->configure(pd()->acp_));
+        mapper.add(this, std::move(r));
+
+        CHECK(pd()->post_ops.create_resource(engine, mapper));
+
+        return status::success;
+    }
+
+    typedef typename prec_traits<data_type::f32>::type data_t;
+
+    status_t execute(const exec_ctx_t &ctx) const override {
+        return execute_forward(ctx);
+    }
+
+private:
+    mutable std::mutex mtx;
+    status_t execute_forward(const exec_ctx_t &ctx) const;
+
+    const pd_t *pd() const { return (const pd_t *)primitive_t::pd().get(); }
+};
+
+} // namespace aarch64
+} // namespace cpu
+} // namespace impl
+} // namespace dnnl
+
+#endif // CPU_AARCH64_ACL_DEPTHWISE_CONVOLUTION_HPP
diff --git a/src/cpu/cpu_convolution_list.cpp b/src/cpu/cpu_convolution_list.cpp
index 094c73aa36..80385432d8 100644
--- a/src/cpu/cpu_convolution_list.cpp
+++ b/src/cpu/cpu_convolution_list.cpp
@@ -63,6 +63,7 @@ using namespace dnnl::impl::cpu::x64;
 #include "cpu/aarch64/jit_sve_512_x8s8s32x_convolution.hpp"
 #include "cpu/aarch64/jit_uni_dw_convolution.hpp"
 #if DNNL_AARCH64 && DNNL_AARCH64_USE_ACL
+#include "cpu/aarch64/acl_depthwise_convolution.hpp"
 #include "cpu/aarch64/acl_gemm_convolution.hpp"
 #include "cpu/aarch64/acl_indirect_gemm_convolution.hpp"
 #endif
@@ -102,6 +103,7 @@ const std::map<pk_dt_impl_key_t, std::vector<impl_list_item_t>> &impl_list_map()
             CPU_INSTANCE_AARCH64(jit_sve_512_dw_convolution_fwd_t)
             CPU_INSTANCE_AARCH64(jit_sve_512_1x1_convolution_fwd_f32_t)
             CPU_INSTANCE_AARCH64(jit_sve_512_convolution_fwd_t<f32>)
+            CPU_INSTANCE_AARCH64_ACL(acl_depthwise_convolution_fwd_t)
             CPU_INSTANCE_AARCH64_ACL(acl_indirect_gemm_convolution_fwd_t)
             CPU_INSTANCE_AARCH64_ACL(acl_gemm_convolution_fwd_t<f32>)
             CPU_INSTANCE(gemm_convolution_fwd_t)