Changelog
V1.2.0
Note
This 1.2.0 version will no longer support old gcc 10 verison, and it now only support Nuclei Toolchain 2023.10(gcc13 and clang17) or later. The major changes that can be felt are as follows:
The prefix of toolchain has changed from
riscv-nuclei-elf-
toriscv64-unknown-elf-
The
-march
option has changed a lot, for example:b
extension changed to_zba_zbb_zbc_zbs
extension,p
extension changed to_xxldsp
or_xxldspn1x
or_xxldspn2x
or_xxldspn3x
extensions which means stardard DSP extension, Nuclei N1, N2, N3 DSP extensionsv
extension changed tov
or_zve32f
or_zve64f
extensions according to the riscv cpu isa used
These extensions need be combined in a certain order to get a correct arch name to match the prebuilt library name, please be cautious
The name of libraries changed due to
-march
in gcc13 updated, for example, the library namedlibnmsis_dsp_rv32imacb.a
is now namedlibnmsis_dsp_rv32imac_zba_zbb_zbc_zbs.a
sinceb
extension changed to_zba_zbb_zbc_zbs
This is release 1.2.0
version of NMSIS-DSP library.
Defined
NUCLEI_DSP_DEFAULT
,NUCLEI_DSP_N1
,NUCLEI_DSP_N2
,NUCLEI_DSP_N3
inriscv_math_types.h
according to gcc options. This means that if compile with--march=rv32imafc_xxldspn1x
, theNUCLEI_DSP_N1
will defined, if compile with--march=rv32imafc_xxldspn2x
, theNUCLEI_DSP_N1
andNUCLEI_DSP_N2
will defined, and so onOptimize some functions with DSP N1/N2/N3 (such as FilteringFunctions, TransformFunctions, ComplexMathFunctions)
RVV intrinsic APIs is update to v0.12.0
Add f16 support(include f16 rvv extension support)
Fix the use of expd80 instruction(Nuclei default dsp instruction)
Fix some testcases bugs(such as MatrixFunctions, TransformFunctions)
V1.1.1
This is release 1.1.1
version of NMSIS-DSP library.
Sync changes from CMSIS DSP commit 1d9e38a, version after v1.14.4
Optimized more for RVP/RVV
Add support for RV32 Vector
Some bugfix(riscv_mat_inverse_f32.c rvv fix, riscv_offset_q15.c p fix, riscv_fir_q15.c rvv fix etc.)
V1.1.0
This is release 1.1.0
version of NMSIS-DSP library.
Sync changes from CMSIS 5.9.0 release
Optimized more for RVP/RVV
Add experimental support for RV32 Vector
V1.0.3
This is release 1.0.3
version of NMSIS-DSP library.
Update build system for NMSIS-DSP library
Rename
RISCV_VECTOR
toRISCV_MATH_VECTOR
in header file and source codeUsing new python script to generate NMSIS-DSP library
Fix riscv_float_to_q31 function for rv64imafcv target
Change vfredsum to vfredusum when using vector intrinsic function due to vector spec 1.0
Support Nuclei RISC-V GCC 10.2
V1.0.2
This is release 1.0.2
version of NMSIS-DSP library.
Sync up to CMSIS DSP library 1.9.0
Adding initial support for RISC-V vector extension support
Caution:
riscv_math.h
is separated into several header files. ExtraPrivateInclude
folder is included as header folder.
V1.0.1
This is release V1.0.1
version of NMSIS-DSP library.
Both Nuclei RISC-V 32 and 64 bit cores are supported now.
Libraries are optimized for RISC-V 32 and 64 bit DSP instructions.
The NN examples are now using Nuclei SDK as running environment.
V1.0.0
This is the first version of NMSIS-DSP library.
We adapt the CMSIS-DSP v1.6.0 library to use RISCV DSP instructions, all the API names now are renamed from arm_xxx
to riscv_xxx
.