zdsp: add ARC DSPLIB backend for zdsp

Introduce ARC DSPLIB backend zdsp library for ARC target.
Add agu and restrict attributes to map with ARC DSPLIB

Signed-off-by: Siyuan Cheng <siyuanc@synopsys.com>
This commit is contained in:
Siyuan Cheng 2023-01-16 15:51:45 +08:00 committed by Carles Cufí
parent a0db069978
commit b475e1fcbf
10 changed files with 354 additions and 3 deletions

View file

@ -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

View file

@ -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

View file

@ -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

View file

@ -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)

View file

@ -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

View file

@ -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/
)

View file

@ -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_ */

View file

@ -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()

View file

@ -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

View file

@ -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