我在理解下面的code(bimpy.v)进行无符号2位乘法运算时遇到问题。
编辑:添加了我的一位朋友的评论:以下修改具有更少逻辑!
o_r <= (i_a[0] ? i_b : 2'b0) + ((i_a[1] ? i_b : 2'b0) << 1);
bimpy.v中的两个信号(w_r和c)的用途是什么?
assign w_r = { ((i_a[1])?i_b:{(BW){1'b0}}), 1'b0 }
^ { 1'b0, ((i_a[0])?i_b:{(BW){1'b0}}) };
assign c = { ((i_a[1])?i_b[(BW-2):0]:{(BW-1){1'b0}}) }
& ((i_a[0])?i_b[(BW-1):1]:{(BW-1){1'b0}});
代码与2位乘2位二进制乘法器gate-level diagram不匹配,如果有误,请纠正我
我还附加了bimpy.v的工作波形,用于一个简单的2x2无符号乘法器。
我还为bimpy.v生成了门级表示图。
////////////////////////////////////////////////////////////////////////////////
//
// Filename: bimpy
//
// Project: A multiply core generator
//
// Purpose: An unsigned 2-bit multiply based upon the fact that LUT's allow
// 6-bits of input, but a 2x2 bit multiply will never carry more
// than one bit. While this multiply is hardware independent, it is
// really motivated by trying to optimize for a specific piece of
// hardware (Xilinx-7 series ...) that has 4-input LUT's with carry
// chains.
//
// Creator: Dan Gisselquist, Ph.D.
// Gisselquist Technology, LLC
//
////////////////////////////////////////////////////////////////////////////////
//
// Copyright (C) 2015,2017-2019, Gisselquist Technology, LLC
//
// This program is free software (firmware): you can redistribute it and/or
// modify it under the terms of the GNU General Public License as published
// by the Free Software Foundation, either version 3 of the License, or (at
// your option) any later version.
//
// This program is distributed in the hope that it will be useful, but WITHOUT
// ANY WARRANTY; without even the implied warranty of MERCHANTIBILITY or
// FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
// for more details.
//
// You should have received a copy of the GNU General Public License along
// with this program. If not, see <http://www.gnu.org/licenses/> for a
// copy.
//
// License: GPL, v3, as defined and found on www.gnu.org,
// http://www.gnu.org/licenses/gpl.html
//
//
////////////////////////////////////////////////////////////////////////////////
module bimpy(i_clk, i_reset, i_ce, i_a, i_b, o_r);
parameter BW=2, LUTB=2;
input i_clk, i_reset, i_ce;
input [(LUTB-1):0] i_a;
input [(BW-1):0] i_b;
output reg [(BW+LUTB-1):0] o_r;
wire [(BW+LUTB-2):0] w_r;
wire [(BW+LUTB-3):1] c;
assign w_r = { ((i_a[1])?i_b:{(BW){1'b0}}), 1'b0 }
^ { 1'b0, ((i_a[0])?i_b:{(BW){1'b0}}) };
assign c = { ((i_a[1])?i_b[(BW-2):0]:{(BW-1){1'b0}}) }
& ((i_a[0])?i_b[(BW-1):1]:{(BW-1){1'b0}});
initial o_r = 0;
always @(posedge i_clk)
if (i_reset)
o_r <= 0;
else if (i_ce)
o_r <= w_r + { c, 2'b0 };
endmodule
答案 0 :(得分:1)
关于MUX的说明
回想一下?
描述了一个多路复用器(MUX),因此声明:
out = sel ? x : y
在门级实现中等效于:
out = (sel & x) | (~sel & y)
(当sel=1
,out <= x
,sel=0
,out <= y
时)
如果y=0
,则MUX减少为x
与sel
之间的AND:out = (sel & x) | (~sel & 0) = sel & x
派生w_r
假设BW=2
和LUTB=2
w_r
是4位信号。让我们分解一下:
w_r = w_rL ^ x_rR
w_rL = { ((i_a[1])?i_b:{(BW){1'b0}}), 1'b0 }
w_rR = { 1'b0, ((i_a[0])?i_b:{(BW){1'b0}}) }
请注意,MUX的两个“ else”值如何都为零,因此将MUX简化为AND,如上面的“ Note”中所述:
w_rL = { BW{i_a[1]} & i_b, 1'b0 } = { A1 & B1, A1 & B0, 0 }
w_rR = { 1'b0, BW{i_a[0]} & i_b } = { 0, A0 & B1, A0 & B0}
为了简化表示,我替换了i_a = {A1, A0}
和i_b = {B1, B0}
。最终,通过按位对它们进行XOR:
w_r[0] = 0 ^ (A0 & B0) = A0 & B0
w_r[1] = (A1 & B0) ^ (A0 & B1)
w_r[2] = (A1 & B1) ^ 0 = A1 & B1
w_r[3] = 0
(隐式设置)
推导c
类似地,对于1位c
信号:
c = cL & cR
cL = i_a[1] ? i_b[(BW-2):0]:{(BW-1){1'b0}} = {A1 & B0}
cR = i_a[0] ? i_b[(BW-1):1]:{(BW-1){1'b0}} = {A0 & B1)
最终:
c = {A1 & B0 & A0 & B1}
派生o_r
如果我们分解o_r
位:
o_r[0] = 0 + w_r[0] = A0 & B0
o_r[1] = 0 + w_r[1] = (A1 & B0) ^ (A0 & B1)
o_r[2] = c + w_r[2] = (A1 & B0 & A0 & B1) + (A1 & B1)
-如果我们将它们相加,则总和是它们的XOR,而进位是它们的AND,即:o_r[2] = (A1 & B0 & A0 & B1) ^ (A1 & B1)
o_r[3] = <carry from o_r[2] addition> = A1 & B0 & A0 & B1 & A1 & B1 = A1 & B0 & A0 & B1
(请记住,与自己进行AND运算等于自己,即x & x = x
)
门级图输出
您的门级图描述了以下等式:
C0 = A0 & B0
(=o_r[0]
)
C1 = (A0 & B1) ^ (A1 & B0)
(=o_r[1]
)
C2 = (A0 & B1 & A1 & B0) ^ (A1 & B1)
(总金额为=o_r[2]
C3 = (A0 & B1 & A1 & B0) & (A1 & B1) = A0 & B1 & A1 & B0
(=o_r[3]
携带)
为什么实施如此奇怪?!
代码注释表明乘法器单元是为特定的FPGA架构构建的,看起来原始编码器的意图是将每个乘法器单元装入该架构的单个LUT中。所以我敢打赌,原来的编码器试图“引导”一个旧的,笨拙的工具,以FPGA有效的方式构建乘法器,而这通常不是门级有效的方式。我认为,这种“手动” RTL级优化对于当今的EDA工具毫无用处(希望如此!)。