diff --git a/doc/services/dsp/index.rst b/doc/services/dsp/index.rst
index f37aefb4f6..38d4ee32d6 100644
--- a/doc/services/dsp/index.rst
+++ b/doc/services/dsp/index.rst
@@ -14,7 +14,7 @@ optimized. The status of the various architectures can be found below:
 ============ =============
 Architecture Status
 ============ =============
-ARC          Unoptimized
+ARC          Optimized
 ARM          Optimized
 ARM64        Optimized
 MIPS         Unoptimized
@@ -46,11 +46,13 @@ Optimizing for your architecture
 
 If your architecture is showing as ``Unoptimized``, it's possible to add a new
 zDSP backend to better support it. To do that, a new Kconfig option should be
-added to `subsys/dsp/Kconfig`_ along with the required dependencies and the
+added to :file:`subsys/dsp/Kconfig` along with the required dependencies and the
 ``default`` set for ``DSP_BACKEND`` Kconfig choice.
 
 Next, the implementation should be added at ``subsys/dsp/<backend>/`` and
-linked in at `subsys/dsp/CMakeLists.txt`_.
+linked in at :file:`subsys/dsp/CMakeLists.txt`. To add architecture-specific attributes,
+its corresponding Kconfig option should be added to :file:`subsys/dsp/Kconfig` and use
+them to update ``DSP_DATA`` and ``DSP_STATIC_DATA`` in :file:`include/zephyr/dsp/dsp.h`.
 
 API Reference
 *************
@@ -59,3 +61,4 @@ API Reference
 
 .. _subsys/dsp/Kconfig: https://github.com/zephyrproject-rtos/zephyr/blob/main/subsys/dsp/Kconfig
 .. _subsys/dsp/CMakeLists.txt: https://github.com/zephyrproject-rtos/zephyr/blob/main/subsys/dsp/CMakeLists.txt
+.. _include/zephyr/dsp/dsp.h: https://github.com/zephyrproject-rtos/zephyr/blob/main/include/zephyr/dsp/dsp.h
diff --git a/include/zephyr/dsp/dsp.h b/include/zephyr/dsp/dsp.h
index 9c720962c0..d475c68a00 100644
--- a/include/zephyr/dsp/dsp.h
+++ b/include/zephyr/dsp/dsp.h
@@ -17,9 +17,17 @@
 #define DSP_FUNC_SCOPE
 #endif
 
+#ifdef CONFIG_DSP_BACKEND_HAS_AGU
+#define DSP_DATA __agu
+#else
 #define DSP_DATA
+#endif
 
+#ifdef CONFIG_DSP_BACKEND_HAS_XDATA_SECTION
+#define DSP_STATIC_DATA DSP_DATA __attribute__((section(".Xdata")))
+#else
 #define DSP_STATIC_DATA DSP_DATA
+#endif
 
 /**
  * @brief DSP Interface
diff --git a/soc/arc/snps_nsim/CMakeLists.txt b/soc/arc/snps_nsim/CMakeLists.txt
index 2a103f3dce..103cb7463c 100644
--- a/soc/arc/snps_nsim/CMakeLists.txt
+++ b/soc/arc/snps_nsim/CMakeLists.txt
@@ -28,6 +28,8 @@ else()
 			       -Xshift_assist -Xfpus_div -Xfpu_mac -Xfpuda -Xfpus_mpy_slow
 			       -Xfpus_div_slow -Xbitstream -Xtimer0 -Xtimer1)
 
+  zephyr_ld_option_ifdef(CONFIG_SOC_NSIM_EM11D -Hlib=em9d_nrg_fpusp -Hdsplib)
+
   if(CONFIG_SOC_NSIM_EM11D)
     set_property(GLOBAL PROPERTY z_arc_dsp_options -Xxy -Xagu_large -Hfxapi -Xdsp2
               -Xdsp_accshift=full -Xdsp_divsqrt=radix2 -Xdsp_complex -Xdsp_itu
diff --git a/subsys/dsp/CMakeLists.txt b/subsys/dsp/CMakeLists.txt
index 8a07484fec..2908c03a36 100644
--- a/subsys/dsp/CMakeLists.txt
+++ b/subsys/dsp/CMakeLists.txt
@@ -2,3 +2,4 @@
 # SPDX-License-Identifier: Apache-2.0
 
 add_subdirectory_ifdef(CONFIG_DSP_BACKEND_CMSIS cmsis)
+add_subdirectory_ifdef(CONFIG_DSP_BACKEND_ARCMWDT arcmwdt)
diff --git a/subsys/dsp/Kconfig b/subsys/dsp/Kconfig
index d4e203c526..ac101be9fb 100644
--- a/subsys/dsp/Kconfig
+++ b/subsys/dsp/Kconfig
@@ -13,9 +13,16 @@ if DSP
 config DSP_BACKEND_HAS_STATIC
 	bool
 
+config DSP_BACKEND_HAS_AGU
+	bool
+
+config DSP_BACKEND_HAS_XDATA_SECTION
+	bool
+
 choice DSP_BACKEND
 	prompt "DSP library backend selection"
 	default DSP_BACKEND_CMSIS if CMSIS_DSP
+	default DSP_BACKEND_ARCMWDT if ARC && "$(ZEPHYR_TOOLCHAIN_VARIANT)" = "arcmwdt"
 	default DSP_BACKEND_CUSTOM
 
 config DSP_BACKEND_CMSIS
@@ -32,6 +39,17 @@ config DSP_BACKEND_CUSTOM
 	  Rely on the application to provide a custom DSP backend. The implementation should be
 	  added to the 'zdsp' build target by the application or one of its modules.
 
+config DSP_BACKEND_ARCMWDT
+	bool "Use the mwdt library as the math backend"
+	depends on ARCMWDT_LIBC
+	depends on CMSIS_DSP
+	select DSP_BACKEND_HAS_STATIC
+	select DSP_BACKEND_HAS_AGU
+	select DSP_BACKEND_HAS_XDATA_SECTION
+	help
+	  Implement the various zephyr DSP functions using the MWDT-DSP library. This feature
+	  requires the MetaWare toolchain and CMSIS module to be selected.
+
 endchoice
 
 endif # DSP
diff --git a/subsys/dsp/arcmwdt/CMakeLists.txt b/subsys/dsp/arcmwdt/CMakeLists.txt
new file mode 100644
index 0000000000..1996f407ac
--- /dev/null
+++ b/subsys/dsp/arcmwdt/CMakeLists.txt
@@ -0,0 +1,8 @@
+# Copyright (c) 2022 Synopsys
+# SPDX-License-Identifier: Apache-2.0
+
+zephyr_include_directories(public)
+
+zephyr_include_directories_ifdef(CONFIG_DSP_BACKEND_ARCMWDT
+  ${ARCMWDT_TOOLCHAIN_PATH}/MetaWare/arc/lib/src/dsp/include/
+)
diff --git a/subsys/dsp/arcmwdt/public/zdsp_backend.h b/subsys/dsp/arcmwdt/public/zdsp_backend.h
new file mode 100644
index 0000000000..1a6870819c
--- /dev/null
+++ b/subsys/dsp/arcmwdt/public/zdsp_backend.h
@@ -0,0 +1,293 @@
+/*
+ * Copyright (c) 2022 Synopsys
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+#ifndef SUBSYS_MATH_ARC_BACKEND_PUBLIC_ZDSP_BACKEND_DSP_H_
+#define SUBSYS_MATH_ARC_BACKEND_PUBLIC_ZDSP_BACKEND_DSP_H_
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/* This include MUST be done before arm_math.h so we can let the arch specific
+ * logic set up the right #define values for arm_math.h
+ */
+#include <zephyr/kernel.h>
+
+#include <arm_math.h>
+#include "dsplib.h"
+
+static inline void zdsp_mult_q7(const DSP_DATA q7_t *src_a, const DSP_DATA q7_t *src_b,
+				DSP_DATA q7_t *dst, uint32_t block_size)
+{
+	dsp_mult_q7(src_a, src_b, dst, block_size);
+}
+static inline void zdsp_mult_q15(const DSP_DATA q15_t *src_a, const DSP_DATA q15_t *src_b,
+				DSP_DATA q15_t *dst, uint32_t block_size)
+{
+	dsp_mult_q15(src_a, src_b, dst, block_size);
+}
+static inline void zdsp_mult_q31(const DSP_DATA q31_t *src_a, const DSP_DATA q31_t *src_b,
+				DSP_DATA q31_t *dst, uint32_t block_size)
+{
+	dsp_mult_q31(src_a, src_b, dst, block_size);
+}
+static inline void zdsp_mult_f32(const DSP_DATA float32_t *src_a, const DSP_DATA float32_t *src_b,
+				DSP_DATA float32_t *dst, uint32_t block_size)
+{
+	dsp_mult_f32(src_a, src_b, dst, block_size);
+}
+
+static inline void zdsp_add_q7(const DSP_DATA q7_t *src_a, const DSP_DATA q7_t *src_b,
+				DSP_DATA q7_t *dst, uint32_t block_size)
+{
+	dsp_add_q7(src_a, src_b, dst, block_size);
+}
+static inline void zdsp_add_q15(const DSP_DATA q15_t *src_a, const DSP_DATA q15_t *src_b,
+				DSP_DATA q15_t *dst, uint32_t block_size)
+{
+	dsp_add_q15(src_a, src_b, dst, block_size);
+}
+static inline void zdsp_add_q31(const DSP_DATA q31_t *src_a, const DSP_DATA q31_t *src_b,
+				DSP_DATA q31_t *dst, uint32_t block_size)
+{
+	dsp_add_q31(src_a, src_b, dst, block_size);
+}
+static inline void zdsp_add_f32(const DSP_DATA float32_t *src_a, const DSP_DATA float32_t *src_b,
+				DSP_DATA float32_t *dst, uint32_t block_size)
+{
+	dsp_add_f32(src_a, src_b, dst, block_size);
+}
+
+static inline void zdsp_sub_q7(const DSP_DATA q7_t *src_a, const DSP_DATA q7_t *src_b,
+				DSP_DATA q7_t *dst, uint32_t block_size)
+{
+	dsp_sub_q7(src_a, src_b, dst, block_size);
+}
+static inline void zdsp_sub_q15(const DSP_DATA q15_t *src_a, const DSP_DATA q15_t *src_b,
+				DSP_DATA q15_t *dst, uint32_t block_size)
+{
+	dsp_sub_q15(src_a, src_b, dst, block_size);
+}
+static inline void zdsp_sub_q31(const DSP_DATA q31_t *src_a, const DSP_DATA q31_t *src_b,
+				DSP_DATA q31_t *dst, uint32_t block_size)
+{
+	dsp_sub_q31(src_a, src_b, dst, block_size);
+}
+static inline void zdsp_sub_f32(const DSP_DATA float32_t *src_a, const DSP_DATA float32_t *src_b,
+				DSP_DATA float32_t *dst, uint32_t block_size)
+{
+	dsp_sub_f32(src_a, src_b, dst, block_size);
+}
+
+static inline void zdsp_scale_q7(const DSP_DATA q7_t *src, q7_t scale_fract, int8_t shift,
+				DSP_DATA q7_t *dst, uint32_t block_size)
+{
+	dsp_scale_q7(src, scale_fract, shift, dst, block_size);
+}
+static inline void zdsp_scale_q15(const DSP_DATA q15_t *src, q15_t scale_fract, int8_t shift,
+				DSP_DATA q15_t *dst, uint32_t block_size)
+{
+	dsp_scale_q15(src, scale_fract, shift, dst, block_size);
+}
+static inline void zdsp_scale_q31(const DSP_DATA q31_t *src, q31_t scale_fract, int8_t shift,
+				DSP_DATA q31_t *dst, uint32_t block_size)
+{
+	dsp_scale_q31(src, scale_fract, shift, dst, block_size);
+}
+
+static inline void zdsp_scale_f32(const DSP_DATA float32_t *src, float32_t scale,
+				DSP_DATA float32_t *dst, uint32_t block_size)
+{
+	dsp_scale_f32(src, scale, dst, block_size);
+}
+
+static inline void zdsp_abs_q7(const DSP_DATA q7_t *src, DSP_DATA q7_t *dst, uint32_t block_size)
+{
+	dsp_abs_q7(src, dst, block_size);
+}
+static inline void zdsp_abs_q15(const DSP_DATA q15_t *src, DSP_DATA q15_t *dst, uint32_t block_size)
+{
+	dsp_abs_q15(src, dst, block_size);
+}
+static inline void zdsp_abs_q31(const DSP_DATA q31_t *src, DSP_DATA q31_t *dst, uint32_t block_size)
+{
+	dsp_abs_q31(src, dst, block_size);
+}
+static inline void zdsp_abs_f32(const DSP_DATA float32_t *src, DSP_DATA float32_t *dst,
+				uint32_t block_size)
+{
+	dsp_abs_f32(src, dst, block_size);
+}
+
+static inline void zdsp_negate_q7(const DSP_DATA q7_t *src, DSP_DATA q7_t *dst, uint32_t block_size)
+{
+	dsp_negate_q7(src, dst, block_size);
+}
+static inline void zdsp_negate_q15(const DSP_DATA q15_t *src, DSP_DATA q15_t *dst,
+				uint32_t block_size)
+{
+	dsp_negate_q15(src, dst, block_size);
+}
+static inline void zdsp_negate_q31(const DSP_DATA q31_t *src, DSP_DATA q31_t *dst,
+				uint32_t block_size)
+{
+	dsp_negate_q31(src, dst, block_size);
+}
+static inline void zdsp_negate_f32(const DSP_DATA float32_t *src, DSP_DATA float32_t *dst,
+				uint32_t block_size)
+{
+	dsp_negate_f32(src, dst, block_size);
+}
+
+static inline void zdsp_dot_prod_q7(const DSP_DATA q7_t *src_a, const DSP_DATA q7_t *src_b,
+				uint32_t block_size, DSP_DATA q31_t *dst)
+{
+	dsp_dot_prod_q7(src_a, src_b, block_size, dst);
+}
+static inline void zdsp_dot_prod_q15(const DSP_DATA q15_t *src_a, const DSP_DATA q15_t *src_b,
+				uint32_t block_size, DSP_DATA q63_t *dst)
+{
+	dsp_dot_prod_q15(src_a, src_b, block_size, dst);
+}
+static inline void zdsp_dot_prod_q31(const DSP_DATA q31_t *src_a, const DSP_DATA q31_t *src_b,
+				uint32_t block_size, DSP_DATA q63_t *dst)
+{
+	dsp_dot_prod_q31(src_a, src_b, block_size, dst);
+}
+static inline void zdsp_dot_prod_f32(const DSP_DATA float32_t *src_a,
+				const DSP_DATA float32_t *src_b, uint32_t block_size,
+				DSP_DATA float32_t *dst)
+{
+	dsp_dot_prod_f32(src_a, src_b, block_size, dst);
+}
+
+static inline void zdsp_shift_q7(const DSP_DATA q7_t *src, int8_t shift_bits, DSP_DATA q7_t *dst,
+				uint32_t block_size)
+{
+	dsp_shift_q7(src, shift_bits, dst, block_size);
+}
+static inline void zdsp_shift_q15(const DSP_DATA q15_t *src, int8_t shift_bits, DSP_DATA q15_t *dst,
+				uint32_t block_size)
+{
+	dsp_shift_q15(src, shift_bits, dst, block_size);
+}
+static inline void zdsp_shift_q31(const DSP_DATA q31_t *src, int8_t shift_bits, DSP_DATA q31_t *dst,
+				uint32_t block_size)
+{
+	dsp_shift_q31(src, shift_bits, dst, block_size);
+}
+
+static inline void zdsp_offset_q7(const DSP_DATA q7_t *src, q7_t offset, DSP_DATA q7_t *dst,
+				uint32_t block_size)
+{
+	dsp_offset_q7(src, offset, dst, block_size);
+}
+static inline void zdsp_offset_q15(const DSP_DATA q15_t *src, q15_t offset, DSP_DATA q15_t *dst,
+				uint32_t block_size)
+{
+	dsp_offset_q15(src, offset, dst, block_size);
+}
+static inline void zdsp_offset_q31(const DSP_DATA q31_t *src, q31_t offset, DSP_DATA q31_t *dst,
+				uint32_t block_size)
+{
+	dsp_offset_q31(src, offset, dst, block_size);
+}
+static inline void zdsp_offset_f32(const DSP_DATA float32_t *src, float32_t offset,
+				DSP_DATA float32_t *dst, uint32_t block_size)
+{
+	dsp_offset_f32(src, offset, dst, block_size);
+}
+
+static inline void zdsp_clip_q7(const DSP_DATA q7_t *src, DSP_DATA q7_t *dst, q7_t low, q7_t high,
+				uint32_t num_samples)
+{
+	arm_clip_q7(src, dst, low, high, num_samples);
+}
+static inline void zdsp_clip_q15(const DSP_DATA q15_t *src, DSP_DATA q15_t *dst, q15_t low,
+				q15_t high, uint32_t num_samples)
+{
+	arm_clip_q15(src, dst, low, high, num_samples);
+}
+static inline void zdsp_clip_q31(const DSP_DATA q31_t *src, DSP_DATA q31_t *dst, q31_t low,
+				q31_t high, uint32_t num_samples)
+{
+	arm_clip_q31(src, dst, low, high, num_samples);
+}
+static inline void zdsp_clip_f32(const DSP_DATA float32_t *src, DSP_DATA float32_t *dst,
+				float32_t low, float32_t high, uint32_t num_samples)
+{
+	arm_clip_f32(src, dst, low, high, num_samples);
+}
+
+static inline void zdsp_and_u8(const DSP_DATA uint8_t *src_a, const DSP_DATA uint8_t *src_b,
+				DSP_DATA uint8_t *dst, uint32_t block_size)
+{
+	arm_and_u8(src_a, src_b, dst, block_size);
+}
+static inline void zdsp_and_u16(const DSP_DATA uint16_t *src_a, const DSP_DATA uint16_t *src_b,
+				DSP_DATA uint16_t *dst, uint32_t block_size)
+{
+	arm_and_u16(src_a, src_b, dst, block_size);
+}
+static inline void zdsp_and_u32(const DSP_DATA uint32_t *src_a, const DSP_DATA uint32_t *src_b,
+				DSP_DATA uint32_t *dst, uint32_t block_size)
+{
+	arm_and_u32(src_a, src_b, dst, block_size);
+}
+
+static inline void zdsp_or_u8(const DSP_DATA uint8_t *src_a, const DSP_DATA uint8_t *src_b,
+				DSP_DATA uint8_t *dst, uint32_t block_size)
+{
+	arm_or_u8(src_a, src_b, dst, block_size);
+}
+static inline void zdsp_or_u16(const DSP_DATA uint16_t *src_a, const DSP_DATA uint16_t *src_b,
+				DSP_DATA uint16_t *dst, uint32_t block_size)
+{
+	arm_or_u16(src_a, src_b, dst, block_size);
+}
+static inline void zdsp_or_u32(const DSP_DATA uint32_t *src_a, const DSP_DATA uint32_t *src_b,
+				DSP_DATA uint32_t *dst, uint32_t block_size)
+{
+	arm_or_u32(src_a, src_b, dst, block_size);
+}
+
+static inline void zdsp_xor_u8(const DSP_DATA uint8_t *src_a, const DSP_DATA uint8_t *src_b,
+				DSP_DATA uint8_t *dst, uint32_t block_size)
+{
+	arm_xor_u8(src_a, src_b, dst, block_size);
+}
+static inline void zdsp_xor_u16(const DSP_DATA uint16_t *src_a, const DSP_DATA uint16_t *src_b,
+				DSP_DATA uint16_t *dst, uint32_t block_size)
+{
+	arm_xor_u16(src_a, src_b, dst, block_size);
+}
+static inline void zdsp_xor_u32(const DSP_DATA uint32_t *src_a, const DSP_DATA uint32_t *src_b,
+				DSP_DATA uint32_t *dst, uint32_t block_size)
+{
+	arm_xor_u32(src_a, src_b, dst, block_size);
+}
+
+static inline void zdsp_not_u8(const DSP_DATA uint8_t *src, DSP_DATA uint8_t *dst,
+				uint32_t block_size)
+{
+	arm_not_u8(src, dst, block_size);
+}
+static inline void zdsp_not_u16(const DSP_DATA uint16_t *src, DSP_DATA uint16_t *dst,
+				uint32_t block_size)
+{
+	arm_not_u16(src, dst, block_size);
+}
+static inline void zdsp_not_u32(const DSP_DATA uint32_t *src, DSP_DATA uint32_t *dst,
+				uint32_t block_size)
+{
+	arm_not_u32(src, dst, block_size);
+}
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* SUBSYS_MATH_ARC_BACKEND_PUBLIC_ZDSP_BACKEND_DSP_H_ */
diff --git a/tests/subsys/dsp/basicmath/CMakeLists.txt b/tests/subsys/dsp/basicmath/CMakeLists.txt
index 22ec493cdf..15512830bb 100644
--- a/tests/subsys/dsp/basicmath/CMakeLists.txt
+++ b/tests/subsys/dsp/basicmath/CMakeLists.txt
@@ -14,3 +14,8 @@ target_sources(app PRIVATE
 target_sources_ifdef(CONFIG_FP16 app PRIVATE src/f16.c)
 
 target_include_directories(app PRIVATE ${ZEPHYR_BASE}/tests/lib/cmsis_dsp)
+
+if(COMPILER STREQUAL arcmwdt)
+  get_property(Z_ARC_DSP_OPTIONS GLOBAL PROPERTY z_arc_dsp_options)
+  target_compile_options(app PRIVATE ${Z_ARC_DSP_OPTIONS})
+endif()
diff --git a/tests/subsys/dsp/basicmath/prj_arc.conf b/tests/subsys/dsp/basicmath/prj_arc.conf
new file mode 100644
index 0000000000..c84bec1f3d
--- /dev/null
+++ b/tests/subsys/dsp/basicmath/prj_arc.conf
@@ -0,0 +1,8 @@
+CONFIG_ZTEST=y
+CONFIG_ZTEST_NEW_API=y
+CONFIG_ARCMWDT_LIBC=y
+CONFIG_CPLUSPLUS=y
+CONFIG_DSP=y
+CONFIG_CMSIS_DSP=y
+CONFIG_CMSIS_DSP_BASICMATH=y
+CONFIG_DSP_BACKEND_ARCMWDT=y
diff --git a/tests/subsys/dsp/basicmath/testcase.yaml b/tests/subsys/dsp/basicmath/testcase.yaml
index b0cd202d3f..a31b4ffe68 100644
--- a/tests/subsys/dsp/basicmath/testcase.yaml
+++ b/tests/subsys/dsp/basicmath/testcase.yaml
@@ -19,3 +19,8 @@ tests:
       - CONFIG_FPU=y
     min_flash: 128
     min_ram: 64
+  zdsp.basicmath.arcmwdt:
+    filter: CONFIG_ISA_ARCV2
+    toolchain_allow: arcmwdt
+    platform_allow: nsim_em11d
+    extra_args: CONF_FILE=prj_arc.conf