furnace/extern/fftw/dft/simd/codlist.mk

80 lines
3.5 KiB
Makefile
Raw Normal View History

# This file contains a standard list of DFT SIMD codelets. It is
# included by common/Makefile to generate the C files with the actual
# codelets in them. It is included by {sse,sse2,...}/Makefile to
# generate and compile stub files that include common/*.c
# You can customize FFTW for special needs, e.g. to handle certain
# sizes more efficiently, by adding new codelets to the lists of those
# included by default. If you change the list of codelets, any new
# ones you added will be automatically generated when you run the
# bootstrap script (see "Generating your own code" in the FFTW
# manual).
###########################################################################
# n1fv_<n> is a hard-coded FFTW_FORWARD FFT of size <n>, using SIMD
N1F = n1fv_2.c n1fv_3.c n1fv_4.c n1fv_5.c n1fv_6.c n1fv_7.c n1fv_8.c \
n1fv_9.c n1fv_10.c n1fv_11.c n1fv_12.c n1fv_13.c n1fv_14.c n1fv_15.c \
n1fv_16.c n1fv_32.c n1fv_64.c n1fv_128.c n1fv_20.c n1fv_25.c
# as above, with restricted input vector stride
N2F = n2fv_2.c n2fv_4.c n2fv_6.c n2fv_8.c n2fv_10.c n2fv_12.c \
n2fv_14.c n2fv_16.c n2fv_32.c n2fv_64.c n2fv_20.c
# as above, but FFTW_BACKWARD
N1B = n1bv_2.c n1bv_3.c n1bv_4.c n1bv_5.c n1bv_6.c n1bv_7.c n1bv_8.c \
n1bv_9.c n1bv_10.c n1bv_11.c n1bv_12.c n1bv_13.c n1bv_14.c n1bv_15.c \
n1bv_16.c n1bv_32.c n1bv_64.c n1bv_128.c n1bv_20.c n1bv_25.c
N2B = n2bv_2.c n2bv_4.c n2bv_6.c n2bv_8.c n2bv_10.c n2bv_12.c \
n2bv_14.c n2bv_16.c n2bv_32.c n2bv_64.c n2bv_20.c
# split-complex codelets
N2S = n2sv_4.c n2sv_8.c n2sv_16.c n2sv_32.c n2sv_64.c
###########################################################################
# t1fv_<r> is a "twiddle" FFT of size <r>, implementing a radix-r DIT step
# for an FFTW_FORWARD transform, using SIMD
T1F = t1fv_2.c t1fv_3.c t1fv_4.c t1fv_5.c t1fv_6.c t1fv_7.c t1fv_8.c \
t1fv_9.c t1fv_10.c t1fv_12.c t1fv_15.c t1fv_16.c t1fv_32.c t1fv_64.c \
t1fv_20.c t1fv_25.c
# same as t1fv_*, but with different twiddle storage scheme
T2F = t2fv_2.c t2fv_4.c t2fv_8.c t2fv_16.c t2fv_32.c t2fv_64.c \
t2fv_5.c t2fv_10.c t2fv_20.c t2fv_25.c
T3F = t3fv_4.c t3fv_8.c t3fv_16.c t3fv_32.c t3fv_5.c t3fv_10.c \
t3fv_20.c t3fv_25.c
T1FU = t1fuv_2.c t1fuv_3.c t1fuv_4.c t1fuv_5.c t1fuv_6.c t1fuv_7.c \
t1fuv_8.c t1fuv_9.c t1fuv_10.c
# as above, but FFTW_BACKWARD
T1B = t1bv_2.c t1bv_3.c t1bv_4.c t1bv_5.c t1bv_6.c t1bv_7.c t1bv_8.c \
t1bv_9.c t1bv_10.c t1bv_12.c t1bv_15.c t1bv_16.c t1bv_32.c t1bv_64.c \
t1bv_20.c t1bv_25.c
# same as t1bv_*, but with different twiddle storage scheme
T2B = t2bv_2.c t2bv_4.c t2bv_8.c t2bv_16.c t2bv_32.c t2bv_64.c \
t2bv_5.c t2bv_10.c t2bv_20.c t2bv_25.c
T3B = t3bv_4.c t3bv_8.c t3bv_16.c t3bv_32.c t3bv_5.c t3bv_10.c \
t3bv_20.c t3bv_25.c
T1BU = t1buv_2.c t1buv_3.c t1buv_4.c t1buv_5.c t1buv_6.c t1buv_7.c \
t1buv_8.c t1buv_9.c t1buv_10.c
# split-complex codelets
T1S = t1sv_2.c t1sv_4.c t1sv_8.c t1sv_16.c t1sv_32.c
T2S = t2sv_4.c t2sv_8.c t2sv_16.c t2sv_32.c
###########################################################################
# q1fv_<r> is <r> twiddle FFTW_FORWARD FFTs of size <r> (DIF step),
# where the output is transposed, using SIMD. This is used for
# in-place transposes in sizes that are divisible by <r>^2. These
# codelets have size ~ <r>^2, so you should probably not use <r>
# bigger than 8 or so.
Q1F = q1fv_2.c q1fv_4.c q1fv_5.c q1fv_8.c
# as above, but FFTW_BACKWARD
Q1B = q1bv_2.c q1bv_4.c q1bv_5.c q1bv_8.c
###########################################################################
SIMD_CODELETS = $(N1F) $(N1B) $(N2F) $(N2B) $(N2S) $(T1FU) $(T1F) \
$(T2F) $(T3F) $(T1BU) $(T1B) $(T2B) $(T3B) $(T1S) $(T2S) $(Q1F) $(Q1B)