Subversion Repositories dblclockfft

#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <sys/stat.h>
#include <string.h>
#include <math.h>
#include <ctype.h>
#include <assert.h>
 
#define	COREDIR	"fft-core"
 
const char	cpyleft[] = 
"///////////////////////////////////////////////////////////////////////////\n"
"//\n"
"// Copyright (C) 2015, Gisselquist Technology, LLC\n"
"//\n"
"// This program is free software (firmware): you can redistribute it and/or\n"
"// modify it under the terms of  the GNU General Public License as published\n"
"// by the Free Software Foundation, either version 3 of the License, or (at\n"
"// your option) any later version.\n"
"//\n"
"// This program is distributed in the hope that it will be useful, but WITHOUT\n"
"// ANY WARRANTY; without even the implied warranty of MERCHANTIBILITY or\n"
"// FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License\n"
"// for more details.\n"
"//\n"
"// You should have received a copy of the GNU General Public License along\n"
"// with this program.  (It's in the $(ROOT)/doc directory, run make with no\n"
"// target there if the PDF file isn\'t present.)  If not, see\n"
"// <http://www.gnu.org/licenses/> for a copy.\n"
"//\n"
"// License:	GPL, v3, as defined and found on www.gnu.org,\n"
"//		http://www.gnu.org/licenses/gpl.html\n"
"//\n"
"//\n"
"///////////////////////////////////////////////////////////////////////////\n";
const char	prjname[] = "A Doubletime Pipelined FFT\n";
const char	creator[] =	"// Creator:	Dan Gisselquist, Ph.D.\n"
				"//		Gisselquist Tecnology, LLC\n";
 
int	lgval(int vl) {
	int	lg;
 
	for(lg=1; (1<<lg) < vl; lg++)
		;
	return lg;
}
 
int	nextlg(int vl) {
	int	r;
 
	for(r=1; r<vl; r<<=1)
		;
	return r;
}
 
int	lgdelay(int nbits, int xtra) {
	int	cbits = nbits + xtra;
	int	delay = nbits + 2;
	if (nbits+1<cbits)
		delay = nbits+4;
	else
		delay = cbits+3;
	return lgval(delay);
}
 
void	build_quarters(const char *fname) {
	FILE	*fp = fopen(fname, "w");
	if (NULL == fp) {
		fprintf(stderr, "Could not open \'%s\' for writing\n", fname);
		perror("O/S Err was:");
		return;
	}
 
	fprintf(fp,
"///////////////////////////////////////////////////////////////////////////\n"
"//\n"
"// Filename: 	qtrstage.v\n"
"//		\n"
"// Project:	%s\n"
"//\n"
"// Purpose:	This file encapsulates the 4 point stage of a decimation in\n"
"//		frequency FFT.  This particular implementation is optimized\n"
"//		so that all of the multiplies are accomplished by additions\n"
"//		and multiplexers only.\n"
"//\n"
"//\n%s"
"//\n",
		prjname, creator);
	fprintf(fp, "%s", cpyleft);
 
	fprintf(fp,
"module\tqtrstage(i_clk, i_rst, i_ce, i_sync, i_data, o_data, o_sync);\n"
	"\tparameter	IWIDTH=16, OWIDTH=IWIDTH+1;\n"
	"\t// Parameters specific to the core that should be changed when this\n"
	"\t// core is built ... Note that the minimum LGSPAN is 2.  Smaller \n"
	"\t// spans must use the fftdoubles stage.\n"
	"\tparameter\tLGWIDTH=8, ODD=0, INVERSE=0,SHIFT=0;\n"
	"\tinput\t				i_clk, i_rst, i_ce, i_sync;\n"
	"\tinput\t	[(2*IWIDTH-1):0]	i_data;\n"
	"\toutput\treg	[(2*OWIDTH-1):0]	o_data;\n"
	"\toutput\treg				o_sync;\n"
	"\t\n"
	"\treg\t	wait_for_sync;\n"
	"\treg\t[2:0]	pipeline;\n"
"\n"
	"\treg\t[(IWIDTH):0]	sum_r, sum_i, diff_r, diff_i;\n"
	"\twire\t[(IWIDTH):0]	n_diff_i;\n"
	"\tassign n_diff_i = -diff_i;\n"
"\n"
	"\treg\t[(2*OWIDTH-1):0]	ob_a;\n"
	"\twire\t[(2*OWIDTH-1):0]	ob_b;\n"
	"\treg\t[(OWIDTH-1):0]		ob_b_r, ob_b_i;\n"
	"\tassign	ob_b = { ob_b_r, ob_b_i };\n"
"\n"
	"\treg\t[(LGWIDTH-1):0]		iaddr;\n"
	"\treg\t[(2*IWIDTH-1):0]	imem;\n"
"\n"
	"\twire\tsigned\t[(IWIDTH-1):0]\timem_r, imem_i;\n"
	"\tassign\timem_r = imem[(2*IWIDTH-1):(IWIDTH)];\n"
	"\tassign\timem_i = imem[(IWIDTH-1):0];\n"
"\n"
	"\twire\tsigned\t[(IWIDTH-1):0]\ti_data_r, i_data_i;\n"
	"\tassign\ti_data_r = i_data[(2*IWIDTH-1):(IWIDTH)];\n"
	"\tassign\ti_data_i = i_data[(IWIDTH-1):0];\n"
"\n"
	"\treg	[(2*OWIDTH-1):0]	omem;\n"
"\n"
	"\twire	[(IWIDTH-1):0]	rnd;\n"
	"\tassign rnd = ((IWIDTH+1-OWIDTH-SHIFT)!=0) ? { {(IWIDTH-1){1'b0}}, (OWIDTH<IWIDTH+1)? 1'b1:1'b0 } : {{(IWIDTH){1'b0}}};\n"
"\n"
	"\talways @(posedge i_clk)\n"
		"\t\tif (i_rst)\n"
		"\t\tbegin\n"
			"\t\t\twait_for_sync <= 1'b1;\n"
			"\t\t\tiaddr <= 0;\n"
			"\t\t\tpipeline <= 3'b000;\n"
		"\t\tend\n"
		"\t\telse if ((i_ce)&&((~wait_for_sync)||(i_sync)))\n"
		"\t\tbegin\n"
			"\t\t\t// Always\n"
			"\t\t\timem <= i_data;\n"
			"\t\t\tiaddr <= iaddr + 1;\n"
			"\t\t\twait_for_sync <= 1'b0;\n"
"\n"
			"\t\t\t// In sequence, clock = 0\n"
			"\t\t\tif (iaddr[0])\n"
			"\t\t\tbegin\n"
				"\t\t\t\tsum_r  <= imem_r + i_data_r + rnd;\n"
				"\t\t\t\tsum_i  <= imem_i + i_data_i + rnd;\n"
				"\t\t\t\tdiff_r <= imem_r - i_data_r + rnd;\n"
				"\t\t\t\tdiff_i <= imem_i - i_data_i + rnd;\n"
"\n"
			"\t\t\t\tpipeline[2:0] <= { pipeline[1:0], 1'b1 };\n"
			"\t\t\tend else\n"
			"\t\t\t\tpipeline[2:0] <= { pipeline[1:0], 1'b0 };\n"
"\n"
			"\t\t\t// In sequence, clock = 1\n"
			"\t\t\tif (pipeline[1])\n"
			"\t\t\tbegin\n"
			"\t\t\t\tob_a <= { sum_r[(IWIDTH):(IWIDTH+1-OWIDTH)],\n"
						"\t\t\t\t\t\tsum_i[(IWIDTH):(IWIDTH+1-OWIDTH)] };\n"
				"\t\t\t\t// on Even, W = e^{-j2pi 1/4 0} = 1\n"
				"\t\t\t\tif (~ODD)\n"
				"\t\t\t\tbegin\n"
"\t\t\t\t\tob_b_r <= diff_r[(IWIDTH-SHIFT):(IWIDTH+1-OWIDTH-SHIFT)];\n"
"\t\t\t\t\tob_b_i <= diff_i[(IWIDTH-SHIFT):(IWIDTH+1-OWIDTH-SHIFT)];\n"
				"\t\t\t\tend else if (~INVERSE) begin\n"
"\t\t\t\t\t// on Odd, W = e^{-j2pi 1/4} = -j\n"
"\t\t\t\t\tob_b_r <=   diff_i[(IWIDTH-SHIFT):(IWIDTH+1-OWIDTH-SHIFT)];\n"
"\t\t\t\t\tob_b_i <=   diff_r[(IWIDTH-SHIFT):(IWIDTH+1-OWIDTH-SHIFT)];\n"
				"\t\t\t\tend else begin\n"
"\t\t\t\t\t// on Odd, W = e^{j2pi 1/4} = j\n"
"\t\t\t\t\tob_b_r <= n_diff_i[(IWIDTH-SHIFT):(IWIDTH+1-OWIDTH-SHIFT)];\n"
"\t\t\t\t\tob_b_i <=   diff_r[(IWIDTH-SHIFT):(IWIDTH+1-OWIDTH-SHIFT)];\n"
				"\t\t\t\tend\n"
				"\t\t\t\t// (wire) ob_b <= { ob_b_r, ob_b_i };\n"
			"\t\t\tend\n"
			"\t\t\t// In sequence, clock = 2\n"
			"\t\t\tif (pipeline[2])\n"
			"\t\t\tbegin\n"
				"\t\t\t\tomem <= ob_b;\n"
				"\t\t\t\to_data <= ob_a;\n"
			"\t\t\tend else\n"
				"\t\t\t\to_data <= omem;\n"
			"\t\t\t// Don\'t forget in the sync check that we are running\n"
			"\t\t\t// at two clocks per sample.  Thus we need to\n"
			"\t\t\t// produce a sync every 2^(LGWIDTH-1) clocks.\n"
			"\t\t\to_sync <= &(~iaddr[(LGWIDTH-2):3]) && (iaddr[2:0] == 3'b100);\n"
		"\t\tend\n"
"endmodule\n");
}
 
void	build_dblstage(const char *fname) {
	FILE	*fp = fopen(fname, "w");
	if (NULL == fp) {
		fprintf(stderr, "Could not open \'%s\' for writing\n", fname);
		perror("O/S Err was:");
		return;
	}
 
	fprintf(fp,
"///////////////////////////////////////////////////////////////////////////\n"
"//\n"
"// Filename: 	dblstage.v\n"
"//\n"
"// Project:	%s\n"
"//\n"
"// Purpose:	This is part of an FPGA implementation that will process\n"
"//		the final stage of a decimate-in-frequency FFT, running\n"
"//		through the data at two samples per clock.  If you notice\n"
"//		from the derivation of an FFT, the only time both even and\n"
"//		odd samples are used at the same time is in this stage.\n"
"//		Therefore, other than this stage and these twiddles, all of\n"
"//		the other stages can run two stages at a time at one sample\n"
"//		per clock.\n"
"//\n"
"//		In this implementation, the output is valid one clock after\n"
"//		the input is valid.  The output also accumulates one bit\n"
"//		above and beyond the number of bits in the input.\n"
"//		\n"
"//		i_clk	A system clock\n"
"//		i_rst	A synchronous reset\n"
"//		i_ce	Circuit enable--nothing happens unless this line is high\n"
"//		i_sync	A synchronization signal, high once per FFT at the start\n"
"//		i_left	The first (even) complex sample input.  The higher order\n"
"//			bits contain the real portion, low order bits the\n"
"//			imaginary portion, all in two\'s complement.\n"
"//		i_right	The next (odd) complex sample input, same format as\n"
"//			i_left.\n"
"//		o_left	The first (even) complex output.\n"
"//		o_right	The next (odd) complex output.\n"
"//		o_sync	Output synchronization signal.\n"
"//\n%s"
"//\n", prjname, creator);
 
	fprintf(fp, "%s", cpyleft);
	fprintf(fp, 
"module	dblstage(i_clk, i_rst, i_ce, i_sync, i_left, i_right, o_left, o_right, o_sync);\n"
	"\tparameter\tIWIDTH=16,OWIDTH=IWIDTH+1, SHIFT=0;\n"
	"\tinput\t\ti_clk, i_rst, i_ce, i_sync;\n"
	"\tinput\t\t[(2*IWIDTH-1):0]\ti_left, i_right;\n"
	"\toutput\twire\t[(2*OWIDTH-1):0]\to_left, o_right;\n"
	"\toutput\treg\t\t\to_sync;\n"
"\n"
	"\twire\tsigned\t[(IWIDTH-1):0]\ti_in_0r, i_in_0i, i_in_1r, i_in_1i;\n"
	"\tassign\ti_in_0r = i_left[(2*IWIDTH-1):(IWIDTH)]; \n"
	"\tassign\ti_in_0i = i_left[(IWIDTH-1):0]; \n"
	"\tassign\ti_in_1r = i_right[(2*IWIDTH-1):(IWIDTH)]; \n"
	"\tassign\ti_in_1i = i_right[(IWIDTH-1):0]; \n"
	"\twire\t[(OWIDTH-1):0]\t\to_out_0r, o_out_0i,\n"
				"\t\t\t\t\to_out_1r, o_out_1i;\n"
"\n"
	"\t// Don't forget that we accumulate a bit by adding two values\n"
	"\t// together. Therefore our intermediate value must have one more\n"
	"\t// bit than the two originals.\n"
	"\treg\t[IWIDTH:0]\tout_0r, out_0i, out_1r, out_1i;\n"
"\n"
	"\treg\twait_for_sync;\n"
"\n"
	"\talways @(posedge i_clk)\n"
		"\t\tif (i_rst)\n"
			"\t\t\twait_for_sync <= 1'b1;\n"
		"\t\telse if ((i_ce)&&((~wait_for_sync)||(i_sync)))\n"
		"\t\tbegin\n"
			"\t\t\twait_for_sync <= 1'b0;\n"
			"\t\t\t//\n"
			"\t\t\tout_0r <= i_in_0r + i_in_1r;\n"
			"\t\t\tout_0i <= i_in_0i + i_in_1i;\n"
			"\t\t\t//\n"
			"\t\t\tout_1r <= i_in_0r - i_in_1r;\n"
			"\t\t\tout_1i <= i_in_0i - i_in_1i;\n"
			"\t\t\t//\n"
			"\t\t\to_sync <= i_sync;\n"
		"\t\tend\n"
"\n"
	"\t// Now, if the master control program doesn't want to keep all of\n"
	"\t// our bits, we can shift down to OWIDTH bits here.\n"
	"\tassign\to_out_0r = out_0r[(IWIDTH-SHIFT):(IWIDTH+1-OWIDTH-SHIFT)];\n"
	"\tassign\to_out_0i = out_0i[(IWIDTH-SHIFT):(IWIDTH+1-OWIDTH-SHIFT)];\n"
	"\tassign\to_out_1r = out_1r[(IWIDTH-SHIFT):(IWIDTH+1-OWIDTH-SHIFT)];\n"
	"\tassign\to_out_1i = out_1i[(IWIDTH-SHIFT):(IWIDTH+1-OWIDTH-SHIFT)];\n"
"\n"
	"\tassign\to_left  = { o_out_0r, o_out_0i };\n"
	"\tassign\to_right = { o_out_1r, o_out_1i };\n"
"\n"
"endmodule\n");
	fclose(fp);
}
 
void	build_multiply(const char *fname) {
	FILE	*fp = fopen(fname, "w");
	if (NULL == fp) {
		fprintf(stderr, "Could not open \'%s\' for writing\n", fname);
		perror("O/S Err was:");
		return;
	}
 
	fprintf(fp,
"///////////////////////////////////////////////////////////////////////////\n"
"//\n"
"// Filename: 	shiftaddmpy.v\n"
"//\n"
"// Project:	%s\n"
"//\n"
"// Purpose:	A portable shift and add multiply.\n"
"//\n"
"//		While both Xilinx and Altera will offer single clock \n"
"//		multiplies, this simple approach will multiply two numbers\n"
"//		on any architecture.  The result maintains the full width\n"
"//		of the multiply, there are no extra stuff bits, no rounding,\n"
"//		no shifted bits, etc.\n"
"//\n"
"//		Further, for those applications that can support it, this\n"
"//		multiply is pipelined and will produce one answer per clock.\n"
"//\n"
"//		For minimal processing delay, make the first parameter\n"
"//		the one with the least bits, so that AWIDTH <= BWIDTH.\n"
"//\n"
"//		The processing delay in this multiply is (AWIDTH+1) cycles.\n"
"//		That is, if the data is present on the input at clock t=0,\n"
"//		the result will be present on the output at time t=AWIDTH+1;\n"
"//\n"
"//\n%s"
"//\n", prjname, creator);
 
	fprintf(fp, "%s", cpyleft);
	fprintf(fp, 
"module	shiftaddmpy(i_clk, i_ce, i_a, i_b, o_r);\n"
	"\tparameter\tAWIDTH=16,BWIDTH=AWIDTH;\n"
	"\tinput\t\t\t\t\ti_clk, i_ce;\n"
	"\tinput\t\t[(AWIDTH-1):0]\t\ti_a;\n"
	"\tinput\t\t[(BWIDTH-1):0]\t\ti_b;\n"
	"\toutput\treg\t[(AWIDTH+BWIDTH-1):0]\to_r;\n"
"\n"
	"\treg\t[(AWIDTH-1):0]\tu_a;\n"
	"\treg\t[(BWIDTH-1):0]\tu_b;\n"
	"\treg\t\t\tsgn;\n"
"\n"
	"\treg\t[(AWIDTH-2):0]\t\tr_a[0:(AWIDTH-1)];\n"
	"\treg\t[(AWIDTH+BWIDTH-2):0]\tr_b[0:(AWIDTH-1)];\n"
	"\treg\t\t\t\tr_s[0:(AWIDTH-1)];\n"
	"\treg\t[(AWIDTH+BWIDTH-1):0]\tacc[0:(AWIDTH-1)];\n"
	"\tgenvar k;\n"
"\n"
	"\t// If we were forced to stay within two\'s complement arithmetic,\n"
	"\t// taking the absolute value here would require an additional bit.\n"
	"\t// However, because our results are now unsigned, we can stay\n"
	"\t// within the number of bits given (for now).\n"
	"\talways @(posedge i_clk)\n"
		"\t\tif (i_ce)\n"
		"\t\tbegin\n"
			"\t\t\tu_a <= (i_a[AWIDTH-1])?(-i_a):(i_a);\n"
			"\t\t\tu_b <= (i_b[BWIDTH-1])?(-i_b):(i_b);\n"
			"\t\t\tsgn <= i_a[AWIDTH-1] ^ i_b[BWIDTH-1];\n"
		"\t\tend\n"
"\n"
	"\talways @(posedge i_clk)\n"
		"\t\tif (i_ce)\n"
		"\t\tbegin\n"
			"\t\t\tacc[0] <= (u_a[0]) ? { {(AWIDTH){1'b0}}, u_b }\n"
			"\t\t\t\t\t: {(AWIDTH+BWIDTH){1'b0}};\n"
			"\t\t\tr_a[0] <= { u_a[(AWIDTH-1):1] };\n"
			"\t\t\tr_b[0] <= { {(AWIDTH-1){1'b0}}, u_b };\n"
			"\t\t\tr_s[0] <= sgn; // The final sign, needs to be preserved\n"
		"\t\tend\n"
"\n"
	"\tgenerate\n"
	"\talways @(posedge i_clk)\n"
	"\tif (i_ce)\n"
	"\tbegin\n"
		"\t\tfor(k=0; k<AWIDTH-1; k++)\n"
		"\t\tbegin\n"
			"\t\t\tacc[k+1] <= acc[k] + ((r_a[k][0]) ? {r_b[k],1'b0}:0);\n"
			"\t\t\tr_a[k+1] <= { 1'b0, r_a[k][(AWIDTH-2):1] };\n"
			"\t\t\tr_b[k+1] <= { r_b[k][(AWIDTH+BWIDTH-3):0], 1'b0};\n"
			"\t\t\tr_s[k+1] <= r_s[k];\n"
		"\t\tend\n"
	"\tend\n"
	"\tendgenerate\n"
"\n"
	"\talways @(posedge i_clk)\n"
		"\t\tif (i_ce)\n"
			"\t\t\to_r <= (r_s[AWIDTH-1]) ? (-acc[AWIDTH-1]) : acc[AWIDTH-1];\n"
"\n"
"endmodule\n");
 
	fclose(fp);
}
 
void	build_dblreverse(const char *fname) {
	FILE	*fp = fopen(fname, "w");
	if (NULL == fp) {
		fprintf(stderr, "Could not open \'%s\' for writing\n", fname);
		perror("O/S Err was:");
		return;
	}
 
	fprintf(fp,
"///////////////////////////////////////////////////////////////////////////\n"
"//\n"
"// Filename: 	dblreverse.v\n"
"//\n"
"// Project:	%s\n"
"//\n"
"// Purpose:	This module bitreverses a pipelined FFT input.  Operation is\n"
"//		expected as follows:\n"
"//\n"
"//		i_clk	A running clock at whatever system speed is offered.\n"
"//		i_rst	A synchronous reset signal, that resets all internals\n"
"//		i_ce	If this is one, one input is consumed and an output\n"
"//			is produced.\n"
"//		i_in_0, i_in_1\n"
"//			Two inputs to be consumed, each of width WIDTH.\n"
"//		o_out_0, o_out_1\n"
"//			Two of the bitreversed outputs, also of the same\n"
"//			width, WIDTH.  Of course, there is a delay from the\n"
"//			first input to the first output.  For this purpose,\n"
"//			o_sync is present.\n"
"//		o_sync	This will be a 1'b1 for the first value in any block.\n"
"//			Following a reset, this will only become 1'b1 once\n"
"//			the data has been loaded and is now valid.  After that,\n"
"//			all outputs will be valid.\n"
"//\n%s"
"//\n", prjname, creator);
	fprintf(fp, "%s", cpyleft);
	fprintf(fp,
"\n\n"
"//\n"
"// How do we do bit reversing at two smples per clock?  Can we separate out\n"
"// our work into eight memory banks, writing two banks at once and reading\n"
"// another two banks in the same clock?\n"
"//\n"
"//	mem[00xxx0] = s_0[n]\n"
"//	mem[00xxx1] = s_1[n]\n"
"//	o_0[n] = mem[10xxx0]\n"
"//	o_1[n] = mem[11xxx0]\n"
"//	...\n"
"//	mem[01xxx0] = s_0[m]\n"
"//	mem[01xxx1] = s_1[m]\n"
"//	o_0[m] = mem[10xxx1]\n"
"//	o_1[m] = mem[11xxx1]\n"
"//	...\n"
"//	mem[10xxx0] = s_0[n]\n"
"//	mem[10xxx1] = s_1[n]\n"
"//	o_0[n] = mem[00xxx0]\n"
"//	o_1[n] = mem[01xxx0]\n"
"//	...\n"
"//	mem[11xxx0] = s_0[m]\n"
"//	mem[11xxx1] = s_1[m]\n"
"//	o_0[m] = mem[00xxx1]\n"
"//	o_1[m] = mem[01xxx1]\n"
"//	...\n"
"//\n"
"//	The answer is that, yes we can but: we need to use four memory banks\n"
"//	to do it properly.  These four banks are defined by the two bits\n"
"//	that determine the top and bottom of the correct address.  Larger\n"
"//	FFT\'s would require more memories.\n"
"//\n"
"//\n");
	fprintf(fp, 
"module	dblreverse(i_clk, i_rst, i_ce, i_in_0, i_in_1,\n"
	"\t\to_out_0, o_out_1, o_sync);\n"
	"\tparameter\t\t\tLGSIZE=4, WIDTH=24;\n"
	"\tinput\t\t\t\ti_clk, i_rst, i_ce;\n"
	"\tinput\t\t[(2*WIDTH-1):0]\ti_in_0, i_in_1;\n"
	"\toutput\treg\t[(2*WIDTH-1):0]\to_out_0, o_out_1;\n"
	"\toutput\treg\t\t\to_sync;\n"
"\n"
	"\treg\tin_reset;\n"
	"\treg\t[(LGSIZE):0]\tiaddr;\n"
	"\treg\t[(2*WIDTH-1):0]\tmem_0e [0:((1<<(LGSIZE-1))-1)];\n"
	"\treg\t[(2*WIDTH-1):0]\tmem_0o [0:((1<<(LGSIZE-1))-1)];\n"
	"\treg\t[(2*WIDTH-1):0]\tmem_1e [0:((1<<(LGSIZE-1))-1)];\n"
	"\treg\t[(2*WIDTH-1):0]\tmem_1o [0:((1<<(LGSIZE-1))-1)];\n"
"\n"
	"\twire\t[(2*LGSIZE-1):0]	braddr;\n"
	"\tgenvar\tk;\n"
	"\tgenerate for(k=0; k<LGSIZE; k++)\n"
		"\t\tassign braddr[k] = iaddr[LGSIZE-1-k];\n"
	"\tendgenerate\n"
"\n"
	"\talways @(posedge i_clk)\n"
		"\t\tif (i_rst)\n"
		"\t\tbegin\n"
			"\t\t\tiaddr <= 0;\n"
			"\t\t\tin_reset <= 1'b1;\n"
		"\t\tend else if (i_ce)\n"
		"\t\tbegin\n"
			"\t\t\tif (iaddr[(LGSIZE-1)])\n"
			"\t\t\tbegin\n"
				"\t\t\t\tmem_1e[{iaddr[LGSIZE],iaddr[(LGSIZE-2):1]}] <= i_in_0;\n"
				"\t\t\t\tmem_1o[{iaddr[LGSIZE],iaddr[(LGSIZE-2):1]}] <= i_in_1;\n"
			"\t\t\tend else begin\n"
				"\t\t\t\tmem_0e[{iaddr[LGSIZE],iaddr[(LGSIZE-2):1]}] <= i_in_0;\n"
				"\t\t\t\tmem_0o[{iaddr[LGSIZE],iaddr[(LGSIZE-2):1]}] <= i_in_1;\n"
			"\t\t\tend\n"
			"\t\t\tiaddr <= iaddr + 2;\n"
			"\t\t\tif (&iaddr[(LGSIZE-1):1])\n"
				"\t\t\t\tin_reset <= 1'b0;\n"
			"\t\t\tif (in_reset)\n"
			"\t\t\tbegin\n"
				"\t\t\t\to_out_0 <= {(2*WIDTH){1'b0}};\n"
				"\t\t\t\to_out_1 <= {(2*WIDTH){1'b0}};\n"
				"\t\t\t\to_sync <= 1'b0;\n"
			"\t\t\tend else\n"
			"\t\t\tbegin\n"
				"\t\t\t\tif (braddr[0])\n"
				"\t\t\t\tbegin\n"
"\t\t\t\t\to_out_0 <= mem_0o[{~iaddr[LGSIZE],braddr[(LGSIZE-2):1]}];\n"
"\t\t\t\t\to_out_1 <= mem_1o[{~iaddr[LGSIZE],braddr[(LGSIZE-2):1]}];\n"
				"\t\t\t\tend else begin\n"
"\t\t\t\t\to_out_0 <= mem_0e[{~iaddr[LGSIZE],braddr[(LGSIZE-2):1]}];\n"
"\t\t\t\t\to_out_1 <= mem_1e[{~iaddr[LGSIZE],braddr[(LGSIZE-2):1]}];\n"
				"\t\t\t\tend\n"
				"\t\t\t\to_sync <= ~(|iaddr[(LGSIZE-1):0]);\n"
			"\t\t\tend\n"
		"\t\tend\n"
"\n"
"endmodule;\n");
 
	fclose(fp);
}
 
void	build_butterfly(const char *fname) {
	FILE	*fp = fopen(fname, "w");
	if (NULL == fp) {
		fprintf(stderr, "Could not open \'%s\' for writing\n", fname);
		perror("O/S Err was:");
		return;
	}
 
	fprintf(fp,
"///////////////////////////////////////////////////////////////////////////\n"
"//\n"
"// Filename:	butterfly.v\n"
"//\n"
"// Project:	%s\n"
"//\n"
"// Purpose:	This routine caculates a butterfly for a decimation\n"
"//		in frequency version of an FFT.  Specifically, given\n"
"//		complex Left and Right values together with a \n"
"//		coefficient, the output of this routine is given\n"
"//		by:\n"
"//\n"
"//		L' = L + R\n"
"//		R' = (L - R)*C\n"
"//\n"
"//		The rest of the junk below handles timing (mostly),\n"
"//		to make certain that L' and R' reach the output at\n"
"//		the same clock.  Further, just to make certain\n"
"//		that is the case, an 'aux' input exists.  This\n"
"//		aux value will come out of this routine synchronized\n"
"//		to the values it came in with.  (i.e., both L', R',\n"
"//		and aux all have the same delay.)  Hence, a caller\n"
"//		of this routine may set aux on the first input with\n"
"//		valid data, and then wait to see aux set on the output\n"
"//		to know when to find the first output with valid data.\n"
"//\n"
"//		All bits are preserved until the very last clock,\n"
"//		where any more bits than OWIDTH will be quietly\n"
"//		discarded.\n"
"//\n"
"//		This design features no overflow checking.\n"
"// \n"
"// Notes:\n"
"//		CORDIC:\n"
"//		Much as we would like, we can't use a cordic here.\n"
"//		The goal is to accomplish an FFT, as defined, and a\n"
"//		CORDIC places a scale factor onto the data.  Removing\n"
"//		the scale factor would cost a two multiplies, which\n"
"//		is precisely what we are trying to avoid.\n"
"//\n"
"//\n"
"//		3-MULTIPLIES:\n"
"//		It should also be possible to do this with three \n"
"//		multiplies and an extra two addition cycles.  \n"
"//\n"
"//		We want\n"
"//			R+I = (a + jb) * (c + jd)\n"
"//			R+I = (ac-bd) + j(ad+bc)\n"
"//		We multiply\n"
"//			P1 = ac\n"
"//			P2 = bd\n"
"//			P3 = (a+b)(c+d)\n"
"//		Then \n"
"//			R+I=(P1-P2)+j(P3-P2-P1)\n"
"//\n"
"//		WIDTHS:\n"
"//		On multiplying an X width number by an\n"
"//		Y width number, X>Y, the result should be (X+Y)\n"
"//		bits, right?\n"
"//		-2^(X-1) <= a <= 2^(X-1) - 1\n"
"//		-2^(Y-1) <= b <= 2^(Y-1) - 1\n"
"//		(2^(Y-1)-1)*(-2^(X-1)) <= ab <= 2^(X-1)2^(Y-1)\n"
"//		-2^(X+Y-2)+2^(X-1) <= ab <= 2^(X+Y-2) <= 2^(X+Y-1) - 1\n"
"//		-2^(X+Y-1) <= ab <= 2^(X+Y-1)-1\n"
"//		YUP!  But just barely.  Do this and you'll really want\n"
"//		to drop a bit, although you will risk overflow in so\n"
"//		doing.\n"
"//\n%s"
"//\n", prjname, creator);
	fprintf(fp, "%s", cpyleft);
 
	fprintf(fp,
"module\tbutterfly(i_clk, i_rst, i_ce, i_coef, i_left, i_right, i_aux,\n"
		"\t\to_left, o_right, o_aux);\n"
	"\t// Public changeable parameters ...\n"
	"\tparameter IWIDTH=16,CWIDTH=IWIDTH+4,OWIDTH=IWIDTH+1;\n"
	"\t// Parameters specific to the core that should not be changed.\n"
	"\tparameter	MPYDELAY=5'd20, // (IWIDTH+1 < CWIDTH)?(IWIDTH+4):(CWIDTH+3),\n"
			"\t\t\tSHIFT=0, ROUND=0;\n"
	"\t// The LGDELAY should be the base two log of the MPYDELAY.  If\n"
	"\t// this value is fractional, then round up to the nearest\n"
	"\t// integer: LGDELAY=ceil(log(MPYDELAY)/log(2));\n"
	"\tparameter\tLGDELAY=5;\n"
	"\tinput\t\ti_clk, i_rst, i_ce;\n"
	"\tinput\t\t[(2*CWIDTH-1):0] i_coef;\n"
	"\tinput\t\t[(2*IWIDTH-1):0] i_left, i_right;\n"
	"\tinput\t\ti_aux;\n"
	"\toutput\twire	[(2*OWIDTH-1):0] o_left, o_right;\n"
	"\toutput\twire	o_aux;\n"
"\n"
	"\twire\t[(OWIDTH-1):0]	o_left_r, o_left_i, o_right_r, o_right_i;\n"
"\n"
	"\treg\t[(2*IWIDTH-1):0]\tr_left, r_right;\n"
	"\treg\t\t\t\tr_aux, r_aux_2;\n"
	"\treg\t[(2*CWIDTH-1):0]\tr_coef, r_coef_2;\n"
	"\twire\tsigned\t[(CWIDTH-1):0]\tr_coef_r, r_coef_i;\n"
	"\tassign\tr_coef_r  = r_coef_2[ (2*CWIDTH-1):(CWIDTH)];\n"
	"\tassign\tr_coef_i  = r_coef_2[ (  CWIDTH-1):0];\n"
	"\twire\tsigned\t[(IWIDTH-1):0]\tr_left_r, r_left_i, r_right_r, r_right_i;\n"
	"\tassign\tr_left_r  = r_left[ (2*IWIDTH-1):(IWIDTH)];\n"
	"\tassign\tr_left_i  = r_left[ (IWIDTH-1):0];\n"
	"\tassign\tr_right_r = r_right[(2*IWIDTH-1):(IWIDTH)];\n"
	"\tassign\tr_right_i = r_right[(IWIDTH-1):0];\n"
"\n"
	"\treg\tsigned\t[(IWIDTH):0]\tr_sum_r, r_sum_i, r_dif_r, r_dif_i;\n"
"\n"
	"\treg	[(LGDELAY-1):0]	fifo_addr;\n"
	"\twire	[(LGDELAY-1):0]	fifo_read_addr;\n"
	"\tassign\tfifo_read_addr = fifo_addr - MPYDELAY;\n"
	"\treg	[(2*IWIDTH+2):0]	fifo_left [ 0:((1<<LGDELAY)-1)];\n"
	"\treg\t\t\t\tovalid;\n"
"\n");
	fprintf(fp,
	"\t// Set up the input to the multiply\n"
	"\talways @(posedge i_clk)\n"
		"\t\tif (i_ce)\n"
		"\t\tbegin\n"
			"\t\t\t// One clock just latches the inputs\n"
			"\t\t\tr_left <= i_left;	// No change in # of bits\n"
			"\t\t\tr_right <= i_right;\n"
			"\t\t\tr_aux <= i_aux;\n"
			"\t\t\tr_coef  <= i_coef;\n"
			"\t\t\t// Next clock adds/subtracts\n"
			"\t\t\tr_sum_r <= r_left_r + r_right_r; // Now IWIDTH+1 bits\n"
			"\t\t\tr_sum_i <= r_left_i + r_right_i;\n"
			"\t\t\tr_dif_r <= r_left_r - r_right_r;\n"
			"\t\t\tr_dif_i <= r_left_i - r_right_i;\n"
			"\t\t\t// Other inputs are simply delayed on second clock\n"
			"\t\t\tr_aux_2 <= r_aux;\n"
			"\t\t\tr_coef_2<= r_coef;\n"
	"\t\tend\n"
"\n");
	fprintf(fp,
	"\t// Don\'t forget to record the even side, since it doesn\'t need\n"
	"\t// to be multiplied, but yet we still need the results in sync\n"
	"\t// with the answer when it is ready.\n"
	"\talways @(posedge i_clk)\n"
		"\t\tif (i_rst)\n"
		"\t\tbegin\n"
			"\t\t\tfifo_addr <= 0;\n"
			"\t\t\tovalid <= 1'b0;\n"
		"\t\tend else if (i_ce)\n"
		"\t\tbegin\n"
			"\t\t\t// Need to delay the sum side--nothing else happens\n"
			"\t\t\t// to it, but it needs to stay synchronized with the\n"
			"\t\t\t// right side.\n"
			"\t\t\tfifo_left[fifo_addr] <= { r_aux_2, r_sum_r, r_sum_i };\n"
			"\t\t\tfifo_addr <= fifo_addr + 1;\n"
			"\t\t\tovalid <= (ovalid) || (fifo_addr > MPYDELAY+1);\n"
		"\t\tend\n"
"\n"
	"\twire\tsigned\t[(CWIDTH-1):0]	ir_coef_r, ir_coef_i;\n"
	"\tassign\tir_coef_r = r_coef_2[(2*CWIDTH-1):CWIDTH];\n"
	"\tassign\tir_coef_i = r_coef_2[(CWIDTH-1):0];\n"
	"\twire\tsigned\t[((IWIDTH+2)+(CWIDTH+1)-1):0]\tp_one, p_two, p_three;\n"
"\n"
"\n");
	fprintf(fp,
	"\t// Multiply output is always a width of the sum of the widths of\n"
	"\t// the two inputs.  ALWAYS.  This is independent of the number of\n"
	"\t// bits in p_one, p_two, or p_three.  These values needed to \n"
	"\t// accumulate a bit (or two) each.  However, this approach to a\n"
	"\t// three multiply complex multiply cannot increase the total\n"
	"\t// number of bits in our final output.  We\'ll take care of\n"
	"\t// dropping back down to the proper width, OWIDTH, in our routine\n"
	"\t// below.\n"
"\n"
"\n");
	fprintf(fp,
	"\t// We accomplish here \"Karatsuba\" multiplication.  That is,\n"
	"\t// by doing three multiplies we accomplish the work of four.\n"
	"\t// Let\'s prove to ourselves that this works ... We wish to\n"
	"\t// multiply: (a+jb) * (c+jd), where a+jb is given by\n"
	"\t//\ta + jb = r_dif_r + j r_dif_i, and\n"
	"\t//\tc + jd = ir_coef_r + j ir_coef_i.\n"
	"\t// We do this by calculating the intermediate products P1, P2,\n"
	"\t// and P3 as\n"
	"\t//\tP1 = ac\n"
	"\t//\tP2 = bd\n"
	"\t//\tP3 = (a + b) * (c + d)\n"
	"\t// and then complete our final answer with\n"
	"\t//\tac - bd = P1 - P2 (this checks)\n"
	"\t//\tad + bc = P3 - P2 - P1\n"
	"\t//\t        = (ac + bc + ad + bd) - bd - ac\n"
	"\t//\t        = bc + ad (this checks)\n"
"\n"
"\n");
	fprintf(fp,
	"\t// This should really be based upon an IF, such as in\n"
	"\t// if (IWIDTH < CWIDTH) then ...\n"
	"\t// However, this is the only (other) way I know to do it.\n"
	"\tgenerate\n"
	"\tif (CWIDTH < IWIDTH+1)\n"
	"\tbegin\n"
		"\t\t// We need to pad these first two multiplies by an extra\n"
		"\t\t// bit just to keep them aligned with the third,\n"
		"\t\t// simpler, multiply.\n"
		"\t\tshiftaddmpy #(CWIDTH+1,IWIDTH+2) p1(i_clk, i_ce,\n"
				"\t\t\t\t{ir_coef_r[CWIDTH-1],ir_coef_r},\n"
				"\t\t\t\t{r_dif_r[IWIDTH],r_dif_r}, p_one);\n"
		"\t\tshiftaddmpy #(CWIDTH+1,IWIDTH+2) p2(i_clk, i_ce,\n"
				"\t\t\t\t{ir_coef_i[CWIDTH-1],ir_coef_i},\n"
				"\t\t\t\t{r_dif_i[IWIDTH],r_dif_i}, p_two);\n"
		"\t\tshiftaddmpy #(CWIDTH+1,IWIDTH+2) p3(i_clk, i_ce,\n"
			"\t\t\t\tir_coef_i+ir_coef_r,\n"
			"\t\t\t\tr_dif_r + r_dif_i,\n"
			"\t\t\t\tp_three);\n"
	"\tend else begin\n"
		"\t\tshiftaddmpy #(IWIDTH+2,CWIDTH+1) p1a(i_clk, i_ce,\n"
				"\t\t\t\t{r_dif_r[IWIDTH],r_dif_r},\n"
				"\t\t\t\t{ir_coef_r[CWIDTH-1],ir_coef_r}, p_one);\n"
		"\t\tshiftaddmpy #(IWIDTH+2,CWIDTH+1) p2a(i_clk, i_ce,\n"
				"\t\t\t\t{r_dif_i[IWIDTH], r_dif_i},\n"
				"\t\t\t\t{ir_coef_i[CWIDTH-1],ir_coef_i}, p_two);\n"
		"\t\tshiftaddmpy #(IWIDTH+2,CWIDTH+1) p3a(i_clk, i_ce,\n"
				"\t\t\t\tr_dif_r+r_dif_i,\n"
				"\t\t\t\tir_coef_i+ir_coef_r,\n"
				"\t\t\t\tp_three);\n"
	"\tend\n"
	"\tendgenerate\n"
"\n");
	fprintf(fp,
	"\t// These values are held in memory and delayed during the\n"
	"\t// multiply.  Here, we recover them.  During the multiply,\n"
	"\t// values were multiplied by 2^(CWIDTH-2)*exp{-j*2*pi*...},\n"
	"\t// therefore, the left_x values need to be right shifted by\n"
	"\t// CWIDTH-2 as well.  The additional bits come from a sign\n"
	"\t// extension.\n"
	"\twire	aux;\n"
	"\twire\tsigned\t[(IWIDTH+CWIDTH):0]	fifo_i, fifo_r;\n"
	"\treg\t\t[(2*IWIDTH+2):0]	fifo_read;\n"
	"\tassign\tfifo_r = { {2{fifo_read[2*(IWIDTH+1)-1]}}, fifo_read[(2*(IWIDTH+1)-1):(IWIDTH+1)], {(CWIDTH-2){1'b0}} };\n"
	"\tassign\tfifo_i = { {2{fifo_read[(IWIDTH+1)-1]}}, fifo_read[((IWIDTH+1)-1):0], {(CWIDTH-2){1'b0}} };\n"
	"\tassign\taux = fifo_read[2*IWIDTH+2];\n"
"\n"
"\n"
	"\treg\tsigned\t[(CWIDTH+IWIDTH+3-1):0]	b_left_r, b_left_i,\n"
			"\t\t\t\t\t\tb_right_r, b_right_i;\n"
	"\treg\tsigned\t[(CWIDTH+IWIDTH+3-1):0]	mpy_r, mpy_i;\n"
	"\treg\tsigned\t[(CWIDTH+IWIDTH+3-1):0]	rnd;\n"
	"\tgenerate\n"
	"\tif ((~ROUND)||(CWIDTH+IWIDTH-OWIDTH-SHIFT<1))\n"
		"\t\tassign rnd = ({(CWIDTH+IWIDTH+3){1'b0}});\n"
	"\telse\n"
		"\t\tassign rnd = ({ {(OWIDTH+3+SHIFT){1'b0}},1'b1,\n"
		"\t\t\t\t{(CWIDTH+IWIDTH-OWIDTH-SHIFT-1){1'b0}} });\n"
	"\tendgenerate\n"
"\n");
	fprintf(fp,
	"\talways @(posedge i_clk)\n"
		"\t\tif (i_ce)\n"
		"\t\tbegin\n"
			"\t\t\t// First clock, recover all values\n"
			"\t\t\tfifo_read <= fifo_left[fifo_read_addr];\n"
			"\t\t\t// These values are IWIDTH+CWIDTH+3 bits wide\n"
			"\t\t\t// although they only need to be (IWIDTH+1)\n"
			"\t\t\t// + (CWIDTH) bits wide.  (We\'ve got two\n"
			"\t\t\t// extra bits we need to get rid of.)\n"
			"\t\t\tmpy_r <= p_one - p_two;\n"
			"\t\t\tmpy_i <= p_three - p_one - p_two;\n"
"\n"
			"\t\t\t// Second clock, round and latch for final clock\n"
			"\t\t\tb_right_r <= mpy_r + rnd;\n"
			"\t\t\tb_right_i <= mpy_i + rnd;\n"
			"\t\t\tb_left_r <= { {2{fifo_r[(IWIDTH+CWIDTH)]}},fifo_r } + rnd;\n"
			"\t\t\tb_left_i <= { {2{fifo_i[(IWIDTH+CWIDTH)]}},fifo_i } + rnd;\n"
			"\t\t\to_aux <= aux & ovalid;\n"
		"\t\tend\n"
"\n");
	fprintf(fp,
	"\t// Final clock--clock and remove unnecessary bits.\n"
	"\t// We have (IWIDTH+CWIDTH+3) bits here, we need to drop down to\n"
	"\t// OWIDTH, and SHIFT by SHIFT bits in the process.  The trick is\n"
	"\t// that we don\'t need (IWIDTH+CWIDTH+3) bits.  We\'ve accumulated\n"
	"\t// them, but the actual values will never fill all these bits.\n"
	"\t// In particular, we only need:\n"
	"\t//\t IWIDTH bits for the input\n"
	"\t//\t     +1 bit for the add/subtract\n"
	"\t//\t+CWIDTH bits for the coefficient multiply\n"
	"\t//\t     +1 bit for the add/subtract in the complex multiply\n"
	"\t//\t ------\n"
	"\t//\t (IWIDTH+CWIDTH+2) bits at full precision.\n"
	"\t//\n"
	"\t// However, the coefficient multiply multiplied by a maximum value\n"
	"\t// of 2^(CWIDTH-2).  Thus, we only have\n"
	"\t//\t   IWIDTH bits for the input\n"
	"\t//\t       +1 bit for the add/subtract\n"
	"\t//\t+CWIDTH-2 bits for the coefficient multiply\n"
	"\t//\t       +1 (optional) bit for the add/subtract in the cpx mpy.\n"
	"\t//\t -------- ... multiply.  (This last bit may be shifted out.)\n"
	"\t//\t (IWIDTH+CWIDTH) valid output bits. \n"
	"\t// Now, if the user wants to keep any extras of these (via OWIDTH),\n"
	"\t// or if he wishes to arbitrarily shift some of these off (via\n"
	"\t// SHIFT) we accomplish that here.\n"
	"\tassign o_left_r  = b_left_r[ (CWIDTH+IWIDTH-1-SHIFT-1):(CWIDTH+IWIDTH-OWIDTH-SHIFT-1)];\n"
	"\tassign o_left_i  = b_left_i[ (CWIDTH+IWIDTH-1-SHIFT-1):(CWIDTH+IWIDTH-OWIDTH-SHIFT-1)];\n"
	"\tassign o_right_r = b_right_r[(CWIDTH+IWIDTH-1-SHIFT-1):(CWIDTH+IWIDTH-OWIDTH-SHIFT-1)];\n"
	"\tassign o_right_i = b_right_i[(CWIDTH+IWIDTH-1-SHIFT-1):(CWIDTH+IWIDTH-OWIDTH-SHIFT-1)];\n"
"\n"
	"\t// As a final step, we pack our outputs into two packed two\'s\n"
	"\t// complement numbers per output word, so that each output word\n"
	"\t// has (2*OWIDTH) bits in it, with the top half being the real\n"
	"\t// portion and the bottom half being the imaginary portion.\n"
	"\tassign	o_left = { o_left_r, o_left_i };\n"
	"\tassign	o_right= { o_right_r,o_right_i};\n"
"\n"
"endmodule\n");
	fclose(fp);
}
 
void	build_stage(const char *fname, int stage, bool odd, int nbits, bool inv, int xtra) {
	FILE	*fstage = fopen(fname, "w");
	int	cbits = nbits + xtra;
 
	if ((cbits * 2) >= sizeof(long long)*8) {
		fprintf(stderr, "ERROR: CMEM Coefficient precision requested overflows long long data type.\n");
		exit(-1);
	}
 
	if (fstage == NULL) {
		fprintf(stderr, "ERROR: Could not open %s for writing!\n", fname);
		perror("O/S Err was:");
		fprintf(stderr, "Attempting to continue, but this file will be missing.\n");
		return;
	}
 
	fprintf(fstage,
"////////////////////////////////////////////////////////////////////////////\n"
"//\n"
"// Filename: 	%sfftstage_%c%d.v\n"
"//\n"
"// Project:	%s\n"
"//\n"
"// Purpose:	This file is (almost) a Verilog source file.  It is meant to\n"
"//		be used by a FFT core compiler to generate FFTs which may be\n"
"//		used as part of an FFT core.  Specifically, this file \n"
"//		encapsulates the options of an FFT-stage.  For any 2^N length\n"
"//		FFT, there shall be (N-1) of these stages.  \n"
"//\n%s"
"//\n",
		(inv)?"i":"", (odd)?'o':'e', stage*2, prjname, creator);
	fprintf(fstage, "%s", cpyleft);
	fprintf(fstage, "module\t%sfftstage_%c%d(i_clk, i_rst, i_ce, i_sync, i_data, o_data, o_sync);\n",
		(inv)?"i":"", (odd)?'o':'e', stage*2);
	// These parameter values are useless at this point--they are to be
	// replaced by the parameter values in the calling program.  Only
	// problem is, the CWIDTH needs to match exactly!
	fprintf(fstage, "\tparameter\tIWIDTH=%d,CWIDTH=%d,OWIDTH=%d;\n",
		nbits, cbits, nbits+1);
	fprintf(fstage,
"\t// Parameters specific to the core that should be changed when this\n"
"\t// core is built ... Note that the minimum LGSPAN (the base two log\n"
"\t// of the span, or the base two log of the current FFT size) is 3.\n"
"\t// Smaller spans (i.e. the span of 2) must use the dblstage module.\n"
"\tparameter\tLGWIDTH=11, LGSPAN=9, LGBDLY=5, BFLYSHIFT=0;\n");
	fprintf(fstage, 
"\tinput					i_clk, i_rst, i_ce, i_sync;\n"
"\tinput		[(2*IWIDTH-1):0]	i_data;\n"
"\toutput	reg	[(2*OWIDTH-1):0]	o_data;\n"
"\toutput	reg				o_sync;\n"
"\n"
"\treg	wait_for_sync;\n"
"\treg	[(2*IWIDTH-1):0]	ib_a, ib_b;\n"
"\treg	[(2*CWIDTH-1):0]	ib_c;\n"
"\treg	ib_sync, b_ce;\n"
"\n"
"\treg	b_started;\n"
"\twire	ob_sync;\n"
"\twire	[(2*OWIDTH-1):0]	ob_a, ob_b;\n");
	fprintf(fstage, 
"\n"
"\t// %scmem is defined as an array of real and complex values,\n"
"\t// where the top CWIDTH bits are the real value and the bottom\n"
"\t// CWIDTH bits are the imaginary value.\n"
"\t//\n"
"\t// cmem[i] = { (2^(CWIDTH-2)) * cos(2*pi*i/(2^LGWIDTH)),\n"
"\t//		(2^(CWIDTH-2)) * sin(2*pi*i/(2^LGWIDTH)) };\n"
"\t//\n"
"\treg	[(2*CWIDTH-1):0]	%scmem [0:((1<<LGSPAN)-1)];\n"
"\tinitial\t$readmemh(\"%scmem_%c%d.hex\",%scmem);\n\n",
		(inv)?"i":"", (inv)?"i":"",
		(inv)?"i":"", (odd)?'o':'e',stage<<1,
		(inv)?"i":"");
	{
		FILE	*cmem;
		char	memfile[128], *ptr;
 
		strncpy(memfile, fname, 125);
		if ((NULL != (ptr = strrchr(memfile, '/')))&&(ptr>memfile)) {
			ptr++;
			sprintf(ptr, "%scmem_%c%d.hex", (inv)?"i":"", (odd)?'o':'e', stage*2);
		} else {
			sprintf(memfile, "%s/%scmem_%c%d.hex",
				COREDIR, (inv)?"i":"",
				(odd)?'o':'e', stage*2);
		}
		// strcpy(&memfile[strlen(memfile)-2], ".hex");
		cmem = fopen(memfile, "w");
		// fprintf(cmem, "// CBITS = %d, inv = %s\n", cbits, (inv)?"true":"false");
		for(int i=0; i<stage/2; i++) {
			int k = 2*i+odd;
			double	W = ((inv)?1:-1)*2.0*M_PI*k/(double)stage;
			double	c, s;
			long long ic, is, vl;
 
			c = cos(W); s = sin(W);
			ic = (long long)((double)((1ll<<(cbits-2)) * c + 0.5));
			is = (long long)((double)((1ll<<(cbits-2)) * s + 0.5));
			vl = (ic & (~(-1ll << (cbits))));
			vl <<= (cbits);
			vl |= (is & (~(-1ll << (cbits))));
			fprintf(cmem, "%0*llx\n", ((cbits*2+3)/4), vl);
			/*
			fprintf(cmem, "%0*llx\t\t// %f+j%f -> %llx +j%llx\n",
				((cbits*2+3)/4), vl, c, s,
				ic & (~(-1ll<<(((cbits+3)/4)*4))),
				is & (~(-1ll<<(((cbits+3)/4)*4))));
			*/
		} fclose(cmem);
	}
 
	fprintf(fstage,
"\treg	[(LGWIDTH-2):0]		iaddr;\n"
"\treg	[(2*IWIDTH-1):0]	imem	[0:((1<<LGSPAN)-1)];\n"
"\n"
"\treg	[(LGSPAN-1):0]		oB;\n"
"\treg	[(2*OWIDTH-1):0]	omem	[0:((1<<LGSPAN)-1)];\n"
"\n"
"\talways @(posedge i_clk)\n"
	"\t\tif (i_rst)\n"
	"\t\tbegin\n"
		"\t\t\twait_for_sync <= 1'b1;\n"
		"\t\t\tiaddr <= 0;\n"
		"\t\t\toB <= 0;\n"
		"\t\t\tb_ce <= 1'b0;\n"
	"\t\tend\n"
	"\t\telse if ((i_ce)&&((~wait_for_sync)||(i_sync)))\n"
	"\t\tbegin\n"
		"\t\t\t//\n"
		"\t\t\t// First step: Record what we\'re not ready to use yet\n"
		"\t\t\t//\n"
		"\t\t\timem[iaddr[(LGSPAN-1):0]] <= i_data;\n"
		"\t\t\tiaddr <= iaddr + 1;\n"
		"\t\t\twait_for_sync <= 1'b0;\n"
"\n"
		"\t\t\t//\n"
		"\t\t\t// Now, we have all the inputs, so let\'s feed the\n"
		"\t\t\t// butterfly\n"
		"\t\t\t//\n"
		"\t\t\tif (iaddr[LGSPAN])\n"
		"\t\t\tbegin\n"
			"\t\t\t\t// One input from memory, ...\n"
			"\t\t\t\tib_a <= imem[iaddr[(LGSPAN-1):0]];\n"
			"\t\t\t\t// One input clocked in from the top\n"
			"\t\t\t\tib_b <= i_data;\n"
			"\t\t\t\t// Set the sync to true on the very first\n"
			"\t\t\t\t// valid input in, and hence on the very\n"
			"\t\t\t\t// first valid data out per FFT.\n"
			"\t\t\t\tib_sync <= (iaddr==(1<<(LGSPAN)));\n"
			"\t\t\t\tib_c <= %scmem[iaddr[(LGSPAN-1):0]];\n"
			"\t\t\t\tb_ce <= 1'b1;\n"
		"\t\t\tend else\n"
			"\t\t\t\tb_ce <= 1'b0;\n"
"\n"
		"\t\t\t//\n"
		"\t\t\t// Next step: recover the outputs from the butterfly\n"
		"\t\t\t//\n"
		"\t\t\tif ((ob_sync||b_started)&&(b_ce))\n"
		"\t\t\tbegin // A butterfly output is available\n"
			"\t\t\t\tb_started <= 1'b1;\n"
			"\t\t\t\tomem[oB] <= ob_b;\n"
			"\t\t\t\toB <= oB+1;\n"
"\n"
			"\t\t\t\to_sync <= (ob_sync);\n"
			"\t\t\t\to_data <= ob_a;\n"
		"\t\t\tend else if (b_started)\n"
		"\t\t\tbegin // and keep outputting once you start--at a rate\n"
		"\t\t\t// of one guaranteed output per clock that has i_ce set.\n"
			"\t\t\t\to_data <= omem[oB];\n"
			"\t\t\t\toB <= oB + 1;\n"
			"\t\t\t\to_sync <= 1'b0;\n"
		"\t\t\tend else\n"
			"\t\t\t\to_sync <= 1'b0;\n"
	"\t\tend\n"
"\n", (inv)?"i":"");
	fprintf(fstage,
"\tbutterfly #(.IWIDTH(IWIDTH),.CWIDTH(CWIDTH),.OWIDTH(OWIDTH),\n"
"\t\t\t.MPYDELAY(%d\'d%d),.LGDELAY(LGBDLY),.SHIFT(BFLYSHIFT))\n"
"\t\tbfly(i_clk, i_rst, (b_ce&i_ce), ib_c,\n"
"\t\t\tib_a, ib_b, ib_sync, ob_a, ob_b, ob_sync);\n"
"endmodule;\n",
	lgdelay(nbits, xtra), (1<xtra)?(nbits+4):(nbits+xtra+3));
}
 
void	usage(void) {
	fprintf(stderr,
"USAGE:\tfftgen [-f <size>] [-d dir] [-c cbits] [-n nbits] [-m mxbits] [-s01]\n"
// "\tfftgen -i\n"
"\t-c <cbits>\tCauses all internal complex coefficients to be\n"
"\t\tlonger than the corresponding data bits, to help avoid\n"
"\t\tcoefficient truncation errors.\n"
"\t-d <dir>\tPlaces all of the generated verilog files into <dir>.\n"
"\t-f <size>\tSets the size of the FFT as the number of complex\n"
"\t\tsamples input to the transform.\n"
"\t-n <nbits>\tSets the number of bits in the twos complement input\n"
"\t\tto the FFT routine.\n"
"\t-m <mxbits>\tSets the maximum bit width that the FFT should ever\n"
"\t\tproduce.  Internal values greater than this value will be\n"
"\t\ttruncated to this value.\n"
"\t-s\tSkip the final bit reversal stage.  This is useful in\n"
"\t\talgorithms that need to apply a filter without needing to do\n"
"\t\tbin shifting, as these algorithms can, with this option, just\n"
"\t\tmultiply by a bit reversed correlation sequence and then\n"
"\t\tinverse FFT the (still bit reversed) result.\n"
"\t-S\tInclude the final bit reversal stage (default).\n"
"\t-0\tA forward FFT (default), meaning that the coefficients are\n"
"\t\tgiven by e^{-j 2 pi k/N n }.\n"
"\t-1\tAn inverse FFT, meaning that the coefficients are\n"
"\t\tgiven by e^{ j 2 pi k/N n }.\n");
}
 
// Features still needed:
//	Interactivity.
//	Some number of maximum bits, beyond which we won't accumulate any more.
//	Obviously, the build_stage above.
//	Copying the files of interest into the fft-core directory, from
//		whatever directory this file is run out of.
int main(int argc, char **argv) {
	int	fftsize = -1, lgsize = -1;
	int	nbitsin = 16, xtracbits = 4;
	int	nbitsout, maxbitsout = -1;
	bool	bitreverse = true, inverse=false, interactive = false,
		verbose_flag = false;
	FILE	*vmain;
	char	fname[128], coredir[1024] = "fft-core";
 
 
	if (argc <= 1)
		usage();
 
	for(int argn=1; argn<argc; argn++) {
		if ('-' == argv[argn][0]) {
			for(int j=1; (argv[argn][j])&&(j<100); j++) {
				switch(argv[argn][j]) {
					case '0':
						inverse = false;
						break;
					case '1':
						inverse = true;
						break;
					case 'c':
						if (argn+1 >= argc) {
							printf("No extra number of coefficient bits given\n");
							usage(); exit(-1);
						}
						xtracbits = atoi(argv[++argn]);
						j+= 200;
						break;
					case 'd':
						if (argn+1 >= argc) {
							printf("No extra number of coefficient bits given\n");
							usage(); exit(-1);
						}
						strcpy(coredir, argv[++argn]);
						j += 200;
						break;
					case 'f':
						if (argn+1 >= argc) {
							printf("No FFT Size given\n");
							usage(); exit(-1);
						}
						fftsize = atoi(argv[++argn]);
						{ int sln = strlen(argv[argn]);
						if (!isdigit(argv[argn][sln-1])){
							switch(argv[argn][sln-1]) {
							case 'k': case 'K':
								fftsize <<= 10;
								break;
							case 'm': case 'M':
								fftsize <<= 20;
								break;
							case 'g': case 'G':
								fftsize <<= 30;
								break;
							default:
								printf("Unknown FFT size, %s\n", argv[argn]);
								exit(-1);
							}
						}}
						j += 200;
						break;
					case 'h':
						usage();
						exit(0);
						break;
					case 'i':
						interactive = true;
						break;
					case 'm':
						if (argn+1 >= argc) {
							printf("No maximum output bit value given\n");
							exit(-1);
						}
						maxbitsout = atoi(argv[++argn]);
						j += 200;
						break;
					case 'n':
						if (argn+1 >= argc) {
							printf("No input bit size given\n");
							exit(-1);
						}
						nbitsin = atoi(argv[++argn]);
						j += 200;
						break;
					case 'S':
						bitreverse = true;
						break;
					case 's':
						bitreverse = false;
						break;
					case 'v':
						verbose_flag = true;
						break;
					default: 
						printf("Unknown argument, -%c\n", argv[argn][j]);
						usage();
						exit(-1);
				}
			}
		} else {
			printf("Unrecognized argument, %s\n", argv[argn]);
			usage();
			exit(-1);
		}
	}
 
	if ((lgsize < 0)&&(fftsize > 1)) {
		for(lgsize=1; (1<<lgsize) < fftsize; lgsize++)
			;
	}
 
	if ((fftsize <= 0)||(nbitsin < 1)||(nbitsin>48)) {
		printf("INVALID PARAMETERS!!!!\n");
		exit(-1);
	}
 
 
	if (nextlg(fftsize) != fftsize) {
		fprintf(stderr, "ERR: FFTSize (%d) *must* be a power of two\n",
				fftsize);
		exit(-1);
	} else if (fftsize < 2) {
		fprintf(stderr, "ERR: Minimum FFTSize is 2, not %d\n",
				fftsize);
		if (fftsize == 1) {
			fprintf(stderr, "You do realize that a 1 point FFT makes very little sense\n");
			fprintf(stderr, "in an FFT operation that handles two samples per clock?\n");
			fprintf(stderr, "If you really need to do an FFT of this size, the output\n");
			fprintf(stderr, "can be connected straight to the input.\n");
		} else {
			fprintf(stderr, "Indeed, a size of %d doesn\'t make much sense to me at all.\n", fftsize);
			fprintf(stderr, "Is such an operation even defined?\n");
		}
		exit(-1);
	}
 
	// Calculate how many output bits we'll have, and what the log
	// based two size of our FFT is.
	{
		int	tmp_size = fftsize;
 
		// The first stage always accumulates one bit, regardless
		// of whether you need to or not.
		nbitsout = nbitsin + 1;
		tmp_size >>= 1;
 
		while(tmp_size > 4) {
			nbitsout += 1;
			tmp_size >>= 2;
		}
 
		if (tmp_size > 1)
			nbitsout ++;
 
		if (fftsize <= 2)
			bitreverse = false;
	} if ((maxbitsout > 0)&&(nbitsout > maxbitsout))
		nbitsout = maxbitsout;
 
 
	{
		struct stat	sbuf;
		if (lstat(coredir, &sbuf)==0) {
			if (!S_ISDIR(sbuf.st_mode)) {
				fprintf(stderr, "\'%s\' already exists, and is not a directory!\n", coredir);
				fprintf(stderr, "I will stop now, lest I overwrite something you care about.\n");
				fprintf(stderr, "To try again, please remove this file.\n");
				exit(-1);
			}
		} else	
			mkdir(coredir, 0755);
		if (access(coredir, X_OK|W_OK) != 0) {
			fprintf(stderr, "I have no access to the directory \'%s\'.\n", coredir);
			exit(-1);
		}
	}
 
	sprintf(fname, "%s/%sfftmain.v", coredir, (inverse)?"i":"");
	vmain = fopen(fname, "w");
	if (NULL == vmain) {
		fprintf(stderr, "Could not open \'%s\' for writing\n", fname);
		perror("Err from O/S:");
		exit(-1);
	}
 
	fprintf(vmain, "/////////////////////////////////////////////////////////////////////////////\n");
	fprintf(vmain, "//\n");
	fprintf(vmain, "// Filename: 	%sfftmain.v\n", (inverse)?"i":"");
	fprintf(vmain, "//\n");
	fprintf(vmain, "// Project:	%s\n", prjname);
	fprintf(vmain, "//\n");
	fprintf(vmain, "// Purpose:	This is the main module in the Doubletime FPGA FFT project.\n");
	fprintf(vmain, "//		As such, all other modules are subordinate to this one.\n");
	fprintf(vmain, "//		(I have been reading too much legalese this week ...)\n");
	fprintf(vmain, "//		This module accomplish a fixed size Complex FFT on %d data\n", fftsize);
	fprintf(vmain, "//		points.  The FFT is fully pipelined, and accepts as inputs\n");
	fprintf(vmain, "//		two complex two\'s complement samples per clock.\n");
	fprintf(vmain, "//\n");
	fprintf(vmain, "// Parameters:\n");
	fprintf(vmain, "//	i_clk\tThe clock.  All operations are synchronous with this clock.\n");
	fprintf(vmain, "//\ti_rst\tSynchronous reset, active high.  Setting this line will\n");
	fprintf(vmain, "//\t\t\tforce the reset of all of the internals to this routine.\n");
	fprintf(vmain, "//\t\t\tFurther, following a reset, the o_sync line will go\n");
	fprintf(vmain, "//\t\t\thigh the same time the first output sample is valid.\n");
	fprintf(vmain, "//	i_ce\tA clock enable line.  If this line is set, this module\n");
	fprintf(vmain, "//\t\t\twill accept two complex values as inputs, and produce\n");
	fprintf(vmain, "//\t\t\ttwo (possibly empty) complex values as outputs.\n");
	fprintf(vmain, "//\t\ti_left\tThe first of two complex input samples.  This value\n");
	fprintf(vmain, "//\t\t\tis split into two two\'s complement numbers, of \n");
	fprintf(vmain, "//\t\t\t%d bits each, with the real portion in the high\n", nbitsin);
	fprintf(vmain, "//\t\t\torder bits, and the imaginary portion taking the\n");
	fprintf(vmain, "//\t\t\tbottom %d bits.\n", nbitsin);
	fprintf(vmain, "//\t\ti_right\tThis is the same thing as i_left, only this is the\n");
	fprintf(vmain, "//\t\t\tsecond of two such samples.  Hence, i_left would\n");
	fprintf(vmain, "//\t\t\tcontain input sample zero, i_right would contain\n");
	fprintf(vmain, "//\t\t\tsample one.  On the next clock i_left would contain\n");
	fprintf(vmain, "//\t\t\tinput sample two, i_right number three and so forth.\n");
	fprintf(vmain, "//\t\to_left\tThe first of two output samples, of the same\n");
	fprintf(vmain, "//\t\t\tformat as i_left, only having %d bits for each of\n", nbitsout);
	fprintf(vmain, "//\t\t\tthe real and imaginary components, leading to %d\n", nbitsout*2);
	fprintf(vmain, "//\t\t\tbits total.\n");
	fprintf(vmain, "//\t\to_right\tThe second of two output samples produced each clock.\n");
	fprintf(vmain, "//\t\t\tThis has the same format as o_left.\n");
	fprintf(vmain, "//\t\to_sync\tA one bit output indicating the first valid sample\n");
	fprintf(vmain, "//\t\t\tproduced by this FFT following a reset.  Ever after,\n");
	fprintf(vmain, "//\t\t\tthis will indicate the first sample of an FFT frame.\n");
	fprintf(vmain, "//\n");
	fprintf(vmain, "%s", creator);
	fprintf(vmain, "//\n");
	fprintf(vmain, "%s", cpyleft);
 
 
	fprintf(vmain, "//\n");
	fprintf(vmain, "//\n");
	fprintf(vmain, "module %sfftmain(i_clk, i_rst, i_ce,\n", (inverse)?"i":"");
	fprintf(vmain, "\t\ti_left, i_right,\n");
	fprintf(vmain, "\t\to_left, o_right, o_sync);\n");
	fprintf(vmain, "\tparameter\tIWIDTH=%d, OWIDTH=%d, LGWIDTH=%d;\n", nbitsin, nbitsout, lgsize);
	assert(lgsize > 0);
	fprintf(vmain, "\tinput\t\ti_clk, i_rst, i_ce;\n");
	fprintf(vmain, "\tinput\t\t[(2*IWIDTH-1):0]\ti_left, i_right;\n");
	fprintf(vmain, "\toutput\treg\t[(2*OWIDTH-1):0]\to_left, o_right;\n");
	fprintf(vmain, "\toutput\treg\t\t\to_sync;\n");
	fprintf(vmain, "\n\n");
 
	fprintf(vmain, "\t// Outputs of the FFT, ready for bit reversal.\n");
	fprintf(vmain, "\twire\t[(2*OWIDTH-1):0]\tbr_left, br_right;\n"); 
	fprintf(vmain, "\n\n");
 
	int	tmp_size = fftsize, lgtmp = lgsize;
	if (fftsize == 2) {
		if (bitreverse) {
			fprintf(vmain, "\treg\tbr_start;\n");
			fprintf(vmain, "\talways @(posedge i_clk)\n");
			fprintf(vmain, "\t\tif (i_rst)\n");
			fprintf(vmain, "\t\t\tbr_start <= 1'b0;\n");
			fprintf(vmain, "\t\telse if (i_ce)\n");
			fprintf(vmain, "\t\t\tbr_start <= 1'b1;\n");
		}
		fprintf(vmain, "\n\n");
		fprintf(vmain, "\tdblstage\t#(IWIDTH)\tstage_2(i_clk, i_rst, i_ce,\n");
		fprintf(vmain, "\t\t\t(~i_rst), i_left, i_right, br_left, br_right);\n");
		fprintf(vmain, "\n\n");
	} else {
		int	nbits = nbitsin, dropbit=0;
		// Always do a first stage
		fprintf(vmain, "\n\n");
		fprintf(vmain, "\twire\t\tw_s%d, w_os%d;\n", fftsize, fftsize);
		fprintf(vmain, "\twire\t[(2*IWIDTH+1):0]\tw_e%d, w_o%d;\n", fftsize, fftsize);
		fprintf(vmain, "\t%sfftstage_e%d\t#(IWIDTH,IWIDTH+%d,IWIDTH+1,%d,%d,%d,0)\tstage_e%d(i_clk, i_rst, i_ce,\n",
			(inverse)?"i":"", fftsize,
			xtracbits,
			lgsize, lgtmp-2, lgdelay(nbits,xtracbits),
			fftsize);
		fprintf(vmain, "\t\t\t(~i_rst), i_left, w_e%d, w_s%d);\n", fftsize, fftsize);
		fprintf(vmain, "\t%sfftstage_o%d\t#(IWIDTH,IWIDTH+%d,IWIDTH+1,%d,%d,%d,0)\tstage_o%d(i_clk, i_rst, i_ce,\n",
			(inverse)?"i":"", fftsize,
			xtracbits,
			lgsize, lgtmp-2, lgdelay(nbits,xtracbits),
			fftsize);
		fprintf(vmain, "\t\t\t(~i_rst), i_left, w_o%d, w_os%d);\n", fftsize, fftsize);
		fprintf(vmain, "\n\n");
 
		sprintf(fname, "%s/%sfftstage_e%d.v", coredir, (inverse)?"i":"", fftsize);
		build_stage(fname, fftsize/2, 0, nbits, inverse, xtracbits);	// Even stage
		sprintf(fname, "%s/%sfftstage_o%d.v", coredir, (inverse)?"i":"", fftsize);
		build_stage(fname, fftsize/2, 1, nbits, inverse, xtracbits);	// Odd  stage
 
		nbits += 1;	// New number of input bits
		tmp_size >>= 1; lgtmp--;
		dropbit = 0;
		fprintf(vmain, "\n\n");
		while(tmp_size >= 8) {
			int	obits = nbits+((dropbit)?0:1);
 
			if ((maxbitsout > 0)&&(obits > maxbitsout))
				obits = maxbitsout;
 
			fprintf(vmain, "\twire\t\tw_s%d, w_os%d;\n", tmp_size, tmp_size);
			fprintf(vmain, "\twire\t[%d:0]\tw_e%d, w_o%d;\n", 2*obits-1, tmp_size, tmp_size);
			fprintf(vmain, "\t%sfftstage_e%d\t#(%d,%d,%d,%d,%d,%d,%d)\tstage_e%d(i_clk, i_rst, i_ce,\n",
				(inverse)?"i":"", tmp_size,
				nbits, nbits+xtracbits, obits,
				lgsize, lgtmp-2, lgdelay(nbits,xtracbits), dropbit,
				tmp_size);
			fprintf(vmain, "\t\t\t\t\t\tw_s%d, w_e%d, w_e%d, w_s%d);\n", tmp_size<<1, tmp_size<<1, tmp_size, tmp_size);
			fprintf(vmain, "\t%sfftstage_o%d\t#(%d,%d,%d,%d,%d,%d,%d)\tstage_o%d(i_clk, i_rst, i_ce,\n",
				(inverse)?"i":"", tmp_size,
				nbits, nbits+xtracbits, obits,
				lgsize, lgtmp-2, lgdelay(nbits,xtracbits), dropbit,
				tmp_size);
			fprintf(vmain, "\t\t\t\t\t\tw_s%d, w_o%d, w_o%d, w_os%d);\n", tmp_size<<1, tmp_size<<1, tmp_size, tmp_size);
			fprintf(vmain, "\n\n");
 
			sprintf(fname, "%s/%sfftstage_e%d.v", coredir, (inverse)?"i":"", tmp_size);
			build_stage(fname, tmp_size/2, 0, nbits, inverse, xtracbits);	// Even stage
			sprintf(fname, "%s/%sfftstage_o%d.v", coredir, (inverse)?"i":"", tmp_size);
			build_stage(fname, tmp_size/2, 1, nbits, inverse, xtracbits);	// Odd  stage
 
 
			dropbit ^= 1;
			nbits = obits;
			tmp_size >>= 1; lgtmp--;
		}
 
		if (tmp_size == 4) {
			int	obits = nbits+((dropbit)?0:1);
 
			if ((maxbitsout > 0)&&(obits > maxbitsout))
				obits = maxbitsout;
 
			fprintf(vmain, "\twire\t\tw_s4, w_os4;\n");
			fprintf(vmain, "\twire\t[%d:0]\tw_e4, w_o4;\n", 2*obits-1);
			fprintf(vmain, "\tqtrstage\t#(%d,%d,%d,0,%d,%d)\tstage_e4(i_clk, i_rst, i_ce,\n",
				nbits, obits, lgsize, (inverse)?1:0, dropbit);
			fprintf(vmain, "\t\t\t\t\t\tw_s8, w_e8, w_e4, w_s4);\n");
			fprintf(vmain, "\tqtrstage\t#(%d,%d,%d,1,%d,%d)\tstage_o4(i_clk, i_rst, i_ce,\n",
				nbits, obits, lgsize, (inverse)?1:0, dropbit);
			fprintf(vmain, "\t\t\t\t\t\tw_s8, w_o8, w_o4, w_os4);\n");
			dropbit ^= 1;
			nbits = obits;
			tmp_size >>= 1; lgtmp--;
		}
 
		{
			int obits = nbits+((dropbit)?0:1);
			if (obits > nbitsout)
				obits = nbitsout;
			if ((maxbitsout>0)&&(obits > maxbitsout))
				obits = maxbitsout;
			fprintf(vmain, "\twire\t\tw_s2;\n");
			fprintf(vmain, "\twire\t[%d:0]\tw_e2, w_o2;\n", 2*obits-1);
			fprintf(vmain, "\tdblstage\t#(%d,%d,%d)\tstage_2(i_clk, i_rst, i_ce,\n", nbits, obits,dropbit);
			fprintf(vmain, "\t\t\t\t\tw_s4, w_e4, w_o4, w_e2, w_o2, w_s2);\n");
 
			fprintf(vmain, "\n\n");
			nbits = obits;
		}
 
		fprintf(vmain, "\t// Prepare for a (potential) bit-reverse stage.\n");
		fprintf(vmain, "\tassign\tbr_left  = w_e2;\n");
		fprintf(vmain, "\tassign\tbr_right = w_o2;\n");
		fprintf(vmain, "\n");
		if (bitreverse) {
			fprintf(vmain, "\twire\tbr_start;\n");
			fprintf(vmain, "\treg\tr_br_started;\n");
			fprintf(vmain, "\t// A delay of one clock here is perfect, as it matches the delay in\n");
			fprintf(vmain, "\t// our dblstage.\n");
			fprintf(vmain, "\talways @(posedge i_clk)\n");
			fprintf(vmain, "\t\tif (i_rst)\n");
			fprintf(vmain, "\t\t\tr_br_started <= 1'b0;\n");
			fprintf(vmain, "\t\telse\n");
			fprintf(vmain, "\t\t\tr_br_started <= r_br_started || w_s4;\n");
			fprintf(vmain, "\tassign\tbr_start = r_br_started;\n");
		}
	}
 
	fprintf(vmain, "\n");
	fprintf(vmain, "\t// Now for the bit-reversal stage.\n");
	fprintf(vmain, "\twire\tbr_sync;\n");
	fprintf(vmain, "\twire\t[(2*OWIDTH-1):0]\tbr_o_left, br_o_right;\n");
	if (bitreverse) {
		fprintf(vmain, "\tdblreverse\t#(%d,%d)\trevstage(i_clk, i_rst,\n", lgsize, nbitsout);
		fprintf(vmain, "\t\t\t(i_ce & br_start), br_left, br_right,\n");
		fprintf(vmain, "\t\t\tbr_o_left, br_o_right, br_sync);\n");
	} else {
		fprintf(vmain, "\tassign\tbr_o_left  = br_left;\n");
		fprintf(vmain, "\tassign\tbr_o_right = br_right;\n");
		fprintf(vmain, "\tassign\tbr_sync    = w_s2;\n");
	}
 
	fprintf(vmain, "\n\n");
	fprintf(vmain, "\t// Last clock: Register our outputs, we\'re done.\n");
	fprintf(vmain, "\talways @(posedge i_clk)\n");
	fprintf(vmain, "\t\tbegin\n");
	fprintf(vmain, "\t\t\to_left  <= br_o_left;\n");
	fprintf(vmain, "\t\t\to_right <= br_o_right;\n");
	fprintf(vmain, "\t\t\to_sync  <= br_sync;\n");
	fprintf(vmain, "\t\tend\n");
	fprintf(vmain, "\n\n");
	fprintf(vmain, "endmodule\n");
	fclose(vmain);
 
	sprintf(fname, "%s/butterfly.v", coredir);
	build_butterfly(fname);
 
	sprintf(fname, "%s/shiftaddmpy.v", coredir);
	build_multiply(fname);
 
	sprintf(fname, "%s/qtrstage.v", coredir);
	build_quarters(fname);
 
	sprintf(fname, "%s/dblstage.v", coredir);
	build_dblstage(fname);
 
	if (bitreverse) {
		sprintf(fname, "%s/dblreverse.v", coredir);
		build_dblreverse(fname);
	}
}