GitHub - ravkum/firFilter: FIR Filter implementation on GPU using OpenCL

ravkum / firFilter Public

Notifications You must be signed in to change notification settings
Fork 0
Star 0

FIR Filter implementation on GPU using OpenCL

Notifications

Name		Name	Last commit message	Last commit date
Latest commit History 12 Commits
.gitignore		.gitignore
README.txt		README.txt
firFilter.cl		firFilter.cl
firFilter.cpp		firFilter.cpp
firFilter.sln		firFilter.sln
firFilter.vcxproj		firFilter.vcxproj
firFilter.vcxproj.filters		firFilter.vcxproj.filters
macros.h		macros.h
utils.cpp		utils.cpp
utils.h		utils.h

Repository files navigation

# firFilter
FIR Filter implementation on GPU using OpenCL

FIR Filter:
  𝑦[𝑖]=∑((𝑥[𝑖−(𝑀/2)+ 𝑗] . 𝑎[𝑗]) 
	  for j ranging from 0 to M-1

  x: Input array and 𝑥[𝑖-(𝑀/2)+𝑗] = 0 if 𝑖−(𝑀/2)+ 𝑗 < 0 or 𝑖-(M/2)+𝑗 > length of 𝑥
  a: Filter array
  M: Filter TAP size
  y: Output array

Note:

The sample assumes input data is commplex array with i and q components being fp32 values. SO every sample is total 8 bytes long.

1) The "master" branch implements a simple FIR filter. 
2) The "filter_pipeline_interleaved" branch implements two FIR filters in a pipeline. The second filter upscales the first filter output by a factor of 2. The upscaling
is done in the kernel to avoid extra global memory read/writes. 


Prerequisite:
1) OpenCL header and lib
2) export AMDAPPSDKROOT=/opt/rocm 
    or to any path where the OpenCL SDK is present.

Build:
1) Linux:
  mkdir build
  cd build
  cmake ../
  make clean all
  
2) Windows:
  Open the .sln fileon Visusl Studio and build. 
  
 
Run:
Usage: firfilter.exe

        [-zeroCopy (0 | 1)] //0 (default) - Device buffer, 1 - zero copy buffer
        [-filtSize (Tap size)]
        [-numElements (220000)]
	[-iter (10) number of times the pipeline should run]
        [-verify (0 | 1]
        [-h (help)]

Example:

                firFilter.exe -numElements 256000 -filtSize 335 -verify 1 -iter 10