annotate fft/fftw/fftw-3.3.4/dft/scalar/codelets/Makefile.am @ 40:223f770b5341 kissfft-double tip

Try a double-precision kissfft
author Chris Cannam
date Wed, 07 Sep 2016 10:40:32 +0100
parents 26056e866c29
children
rev   line source
Chris@19 1 # This Makefile.am specifies a set of codelets, efficient transforms
Chris@19 2 # of small sizes, that are used as building blocks (kernels) by FFTW
Chris@19 3 # to build up large transforms, as well as the options for generating
Chris@19 4 # and compiling them.
Chris@19 5
Chris@19 6 # You can customize FFTW for special needs, e.g. to handle certain
Chris@19 7 # sizes more efficiently, by adding new codelets to the lists of those
Chris@19 8 # included by default. If you change the list of codelets, any new
Chris@19 9 # ones you added will be automatically generated when you run the
Chris@19 10 # bootstrap script (see "Generating your own code" in the FFTW
Chris@19 11 # manual).
Chris@19 12
Chris@19 13 ###########################################################################
Chris@19 14 AM_CPPFLAGS = -I$(top_srcdir)/kernel -I$(top_srcdir)/dft \
Chris@19 15 -I$(top_srcdir)/dft/scalar
Chris@19 16 noinst_LTLIBRARIES = libdft_scalar_codelets.la
Chris@19 17
Chris@19 18 ###########################################################################
Chris@19 19 # n1_<n> is a hard-coded FFT of size <n> (base cases of FFT recursion)
Chris@19 20 N1 = n1_2.c n1_3.c n1_4.c n1_5.c n1_6.c n1_7.c n1_8.c n1_9.c n1_10.c \
Chris@19 21 n1_11.c n1_12.c n1_13.c n1_14.c n1_15.c n1_16.c n1_32.c n1_64.c \
Chris@19 22 n1_20.c n1_25.c # n1_30.c n1_40.c n1_50.c
Chris@19 23
Chris@19 24 ###########################################################################
Chris@19 25 # t1_<r> is a "twiddle" FFT of size <r>, implementing a radix-r DIT step
Chris@19 26 T1 = t1_2.c t1_3.c t1_4.c t1_5.c t1_6.c t1_7.c t1_8.c t1_9.c \
Chris@19 27 t1_10.c t1_12.c t1_15.c t1_16.c t1_32.c t1_64.c \
Chris@19 28 t1_20.c t1_25.c # t1_30.c t1_40.c t1_50.c
Chris@19 29
Chris@19 30 # t2_<r> is also a twiddle FFT, but instead of using a complete lookup table
Chris@19 31 # of trig. functions, it partially generates the trig. values on the fly
Chris@19 32 # (this is faster for large sizes).
Chris@19 33 T2 = t2_4.c t2_8.c t2_16.c t2_32.c t2_64.c \
Chris@19 34 t2_5.c t2_10.c t2_20.c t2_25.c
Chris@19 35
Chris@19 36 ###########################################################################
Chris@19 37 # The F (DIF) codelets are used for a kind of in-place transform algorithm,
Chris@19 38 # but the planner seems to never (or hardly ever) use them on the machines
Chris@19 39 # we have access to, preferring the Q codelets and the use of buffers
Chris@19 40 # for sub-transforms. So, we comment them out, at least for now.
Chris@19 41
Chris@19 42 # f1_<r> is a "twiddle" FFT of size <r>, implementing a radix-r DIF step
Chris@19 43 F1 = # f1_2.c f1_3.c f1_4.c f1_5.c f1_6.c f1_7.c f1_8.c f1_9.c f1_10.c f1_12.c f1_15.c f1_16.c f1_32.c f1_64.c
Chris@19 44
Chris@19 45 # like f1, but partially generates its trig. table on the fly
Chris@19 46 F2 = # f2_4.c f2_8.c f2_16.c f2_32.c f2_64.c
Chris@19 47
Chris@19 48 ###########################################################################
Chris@19 49 # q1_<r> is <r> twiddle FFTs of size <r> (DIF step), where the output is
Chris@19 50 # transposed. This is used for in-place transposes in sizes that are
Chris@19 51 # divisible by <r>^2. These codelets have size ~ <r>^2, so you should
Chris@19 52 # probably not use <r> bigger than 8 or so.
Chris@19 53 Q1 = q1_2.c q1_4.c q1_8.c q1_3.c q1_5.c q1_6.c
Chris@19 54
Chris@19 55 ###########################################################################
Chris@19 56 ALL_CODELETS = $(N1) $(T1) $(T2) $(F1) $(F2) $(Q1)
Chris@19 57 BUILT_SOURCES= $(ALL_CODELETS) $(CODLIST)
Chris@19 58
Chris@19 59 libdft_scalar_codelets_la_SOURCES = $(BUILT_SOURCES)
Chris@19 60
Chris@19 61 SOLVTAB_NAME = X(solvtab_dft_standard)
Chris@19 62 XRENAME=X
Chris@19 63
Chris@19 64 # special rules for regenerating codelets.
Chris@19 65 include $(top_srcdir)/support/Makefile.codelets
Chris@19 66
Chris@19 67 if MAINTAINER_MODE
Chris@19 68 FLAGS_N1=$(DFT_FLAGS_COMMON)
Chris@19 69 FLAGS_T1=$(DFT_FLAGS_COMMON)
Chris@19 70 FLAGS_T2=$(DFT_FLAGS_COMMON) -twiddle-log3 -precompute-twiddles
Chris@19 71 FLAGS_F1=$(DFT_FLAGS_COMMON)
Chris@19 72 FLAGS_F2=$(DFT_FLAGS_COMMON) -twiddle-log3 -precompute-twiddles
Chris@19 73 FLAGS_Q1=$(DFT_FLAGS_COMMON) -reload-twiddle
Chris@19 74 FLAGS_Q2=$(DFT_FLAGS_COMMON) -twiddle-log3 -precompute-twiddles
Chris@19 75
Chris@19 76 n1_%.c: $(CODELET_DEPS) $(GEN_NOTW)
Chris@19 77 ($(PRELUDE_COMMANDS_DFT); $(TWOVERS) $(GEN_NOTW) $(FLAGS_N1) -n $* -name n1_$* -include "n.h") | $(ADD_DATE) | $(INDENT) >$@
Chris@19 78
Chris@19 79 t1_%.c: $(CODELET_DEPS) $(GEN_TWIDDLE)
Chris@19 80 ($(PRELUDE_COMMANDS_DFT); $(TWOVERS) $(GEN_TWIDDLE) $(FLAGS_T1) -n $* -name t1_$* -include "t.h") | $(ADD_DATE) | $(INDENT) >$@
Chris@19 81
Chris@19 82 t2_%.c: $(CODELET_DEPS) $(GEN_TWIDDLE)
Chris@19 83 ($(PRELUDE_COMMANDS_DFT); $(TWOVERS) $(GEN_TWIDDLE) $(FLAGS_T2) -n $* -name t2_$* -include "t.h") | $(ADD_DATE) | $(INDENT) >$@
Chris@19 84
Chris@19 85 f1_%.c: $(CODELET_DEPS) $(GEN_TWIDDLE)
Chris@19 86 ($(PRELUDE_COMMANDS_DFT); $(TWOVERS) $(GEN_TWIDDLE) $(FLAGS_F1) -dif -n $* -name f1_$* -include "f.h") | $(ADD_DATE) | $(INDENT) >$@
Chris@19 87
Chris@19 88 f2_%.c: $(CODELET_DEPS) $(GEN_TWIDDLE)
Chris@19 89 ($(PRELUDE_COMMANDS_DFT); $(TWOVERS) $(GEN_TWIDDLE) $(FLAGS_F2) -dif -n $* -name f2_$* -include "f.h") | $(ADD_DATE) | $(INDENT) >$@
Chris@19 90
Chris@19 91 q1_%.c: $(CODELET_DEPS) $(GEN_TWIDSQ)
Chris@19 92 ($(PRELUDE_COMMANDS_DFT); $(TWOVERS) $(GEN_TWIDSQ) $(FLAGS_Q1) -dif -n $* -name q1_$* -include "q.h") | $(ADD_DATE) | $(INDENT) >$@
Chris@19 93
Chris@19 94 q2_%.c: $(CODELET_DEPS) $(GEN_TWIDSQ)
Chris@19 95 ($(PRELUDE_COMMANDS_DFT); $(TWOVERS) $(GEN_TWIDSQ) $(FLAGS_Q2) -dif -n $* -name q2_$* -include "q.h") | $(ADD_DATE) | $(INDENT) >$@
Chris@19 96
Chris@19 97 endif # MAINTAINER_MODE