zldsp::fft(develop)

zldsp::fft(develop) aims at FFT implementation and analysis.

If you want to use zldsp::fft, please refer to zldsp_fft.

Usage

Please make sure Clang (AppleClang 16+ or LLVM/Clang 17+), cmake (minimum 3.20) are installed and configured on your OS.

git clone https://github.com/ZL-Audio/zldsp_fft_develop
cd zldsp_fft_develop
git submodule update --init

You may need to edit the building commands in benchmark/build_config.py.

Accuracy Benchmark

CFFT Accuracy Benchmark

Run fftw3, kfr, zldsp, pffft CFFT from order n0 to n1.

python3 benchmark/accuracy_mid_cfft.py <n0> <n1> <--avx2> <--double>

Example

python3 benchmark/accuracy_mid_cfft.py 5 10
Running accuracy CFFT benchmark from order 5 to 10 (10 reps each)...
Order      kfr                fftw3              zldsp              pffft             
--------------------------------------------------------------------------------------
5          2.39176700e-14     1.17213200e-13     3.95274100e-14     1.63440100e-13     
6          1.64665400e-13     2.15260600e-13     1.08373000e-13     3.01948200e-13     
7          5.78072000e-13     5.63297200e-13     4.81077900e-13     8.61865300e-13     
8          1.05322900e-12     1.27926900e-12     8.51770200e-13     2.17439600e-12     
9          3.05149800e-12     2.91082200e-12     2.65517300e-12     5.28153300e-12     
10         6.62576900e-12     6.73661400e-12     6.31371100e-12     1.16500700e-11

CFFT Consistent Benchmark

Run zldsp different CFFT forward/backward APIs from order n0 to n1.

python3 benchmark/accuracy_cfft.py <n0> <n1> <--avx2> <--double>

Example

python3 benchmark/accuracy_cfft.py 5 10
Running accuracy CFFT benchmark for zldsp from order 5 to 10...
Order      Fwd Max MSE        Bwd Max MSE        Id Max MSE        
------------------------------------------------------------------
5          0.00000000e+00     0.00000000e+00     4.70630500e-15    
6          0.00000000e+00     0.00000000e+00     7.68742700e-15    
7          0.00000000e+00     0.00000000e+00     1.11504800e-14    
8          0.00000000e+00     0.00000000e+00     1.19794100e-14    
9          0.00000000e+00     0.00000000e+00     1.32861400e-14    
10         0.00000000e+00     0.00000000e+00     1.51891300e-14

RFFT Accuracy Benchmark

Run fftw3, kfr, zldsp, pffft RFFT from order n0 to n1.

python3 benchmark/accuracy_mid_rfft.py <n0> <n1> <--avx2> <--double>

Example

python3 benchmark/accuracy_mid_rfft.py 5 10
Running accuracy RFFT benchmark from order 5 to 10 (10 reps each)...
Order      kfr                fftw3              zldsp              pffft             
--------------------------------------------------------------------------------------
5          7.26960800e-14     3.77433200e-14     3.38191300e-14     3.27714500e-14     
6          1.63731400e-13     1.01401400e-13     9.75312900e-14     1.05184300e-13     
7          3.56730700e-13     2.27370500e-13     2.42351800e-13     2.67729800e-13     
8          1.07237800e-12     5.58322300e-13     8.09634900e-13     5.90561800e-13     
9          1.91429100e-12     1.27614100e-12     1.43604000e-12     1.51917200e-12     
10         5.22122600e-12     2.91316900e-12     3.52626300e-12     3.38997800e-12

RFFT Consistent Benchmark

Run zldsp different RFFT forward/backward APIs from order n0 to n1.

python3 benchmark/accuracy_rfft.py <n0> <n1> <--avx2> <--double>

Example

python3 benchmark/accuracy_rfft.py 5 10
Running accuracy RFFT benchmark for zldsp from order 5 to 10...
Order      Fwd Max MSE        Bwd Max MSE        Id Max MSE        
------------------------------------------------------------------
5          0.00000000e+00     0.00000000e+00     5.30825400e-15    
6          0.00000000e+00     0.00000000e+00     7.23466400e-15    
7          0.00000000e+00     0.00000000e+00     5.84959600e-15    
8          0.00000000e+00     0.00000000e+00     8.47672600e-15    
9          0.00000000e+00     0.00000000e+00     8.37233100e-15    
10         0.00000000e+00     0.00000000e+00     9.01347400e-15

Throughput Benchmark

CFFT Throughput Benchmark

Run <algorithm> CFFT throughput benchmark from order n0 to n1.

python3 benchmark/throughput_cfft.py <n0> <n1> <algorithm> <--avx2> <--double>

Example

python3 benchmark/throughput_cfft.py 5 10 zldsp
Running CFFT throughput benchmark for zldsp from order 5 to 10...
Order      Time (us)       Throughput (MFLOPS) 
---------------------------------------------
5          0.0189          42283.9468          
6          0.0281          68408.6276          
7          0.0635          70521.8659          
8          0.1303          78574.4614          
9          0.2998          76850.4027          
10         0.6359          80514.4208

RFFT Throughput Benchmark

Run <algorithm> RFFT throughput benchmark from order n0 to n1.

python3 benchmark/throughput_rfft.py <n0> <n1> <algorithm> <--avx2> <--double>

Example

python3 benchmark/throughput_rfft.py 5 10 zldsp
Running RFFT throughput benchmark for zldsp from order 5 to 10...
Order      Time (us)       Throughput (MFLOPS) 
---------------------------------------------
5          0.0111          36143.1067          
6          0.0221          43344.2067          
7          0.0406          55150.3424          
8          0.0840          60944.2159          
9          0.1705          67558.5149          
10         0.3770          67910.7675

Algorithms

• zldsp

External libraries:

• fftw3 / fftw3_estimate
• kfr
• pffft
• ipp
• vdsp
• armpl / armpl_estimate

ArmPL benchmarks require an Arm Performance Libraries installation. Set ARMPL_DIR to its installation directory before running the benchmark scripts. armpl uses FFTW_MEASURE; armpl_estimate uses FFTW_ESTIMATE.

Use zldsp::fft in another project

Static Dispatch

target_compile_definitions(my_static_target PRIVATE HWY_COMPILE_ONLY_STATIC)

Use the compiler's architecture option to select an SSE, AVX2, or NEON static target.

SIMD Target	GCC/Clang	MSVC
SSE2	-march=x86-64	no flag required
SSE4	-march=x86-64-v2 -maes -mpclmul	not supported
AVX2	-march=x86-64-v3 -maes -mpclmul	/arch:AVX2
NEON	-march=armv8-a+simd	/arch:armv8.0

See static_dispatch_caller for CFFT and RFFT static dispatch examples with AoS and SoA layouts.

Caller-owned Dynamic Dispatch

For example, an x86 application can enable only SSE2 and AVX2:

target_compile_definitions(my_dynamic_target PRIVATE "HWY_DISABLED_TARGETS=~(HWY_SSE2|HWY_AVX2)")

Compile this target for the oldest supported baseline (for example, -march=x86-64 for SSE2).

See wrapper, its interface, and dynamic_dispatch_caller for CFFT and RFFT dynamic dispatch examples with AoS and SoA layouts.

License

zldsp_fft is licensed under Apache-2.0 license, as found in the LICENSE.md file.

All external libraries (submodules) are not covered by this license. All trademarks, product names, and company names are the property of their respective owners and are used for identification purposes only. Please refer to the individual licenses within each submodule:

• FFTW3: fftw3_impl/fftw3 (GPL-2.0 license)
• KFR: kfr_impl/kfr (GPL-2.0 license)
• pffft: pffft_impl/pffft (BSD-like license)
• Google benchmark: google/benchmark (Apache-2.0 license)
• Google highway: google/highway (Apache-2.0 license)