![]() ![]() ![]() |
// ************************************************************************** // // // // eses eses // // eses eses // // eses eseses esesese eses Embedded Systems Group // // ese ese ese ese ese // // ese eseseses eseseses ese Department of Computer Science // // eses eses ese eses // // eses eseses eseseses eses University of Kaiserslautern // // eses eses // // // // ************************************************************************** // // The following module implements a multiplication of B-complement numbers. // // The computation is done by summing up the usual partial products by the use// // of carry-save adders for the intermediate steps and a carry-ripple adder // // for the final step. // // The depth of the circuit is seen by observing that the evaluation can // // proceed row-wise for i=0..M-2, and columnwise for the final row. Thus, the // // depth of the circuit is O(M+N) and therefore not optimal. // // ** Note: We are only interested in even values of B ** // // ************************************************************************** // macro B = 4; // base of the radix numbers macro M = 5; // number of digits used macro N = 3; // number of digits used macro alpha(x) = (x<(B/2) ? +x : +x-B); macro gamma(y) = (y<0 ? y+B : y); macro dval(x,i,k) = (i==k-1 ? alpha(x[i]) : +x[i]); macro intval(x,k) = sum(i=0..k-1) (dval(x,i,k) * exp(B,i)); module IntMulCSACRA([M]nat{B} ?x,[N]nat{B} ?y,[M+N]nat{B} p) { event [M][N]nat{B} pp; // digits of partial products event [M][N-1]int{B} cp; // carries for summation in CSA array event [N-1]int{2} cr; // carries for summation in final CRA assert(B==2*(B/2)); // check that B is even // construct carry-save adder array // this part has depth O(M) since each row can be evaluated in O(1) for(i=0..M-1) { for(j=0..N-2) { let(xin = sat{B}((i==M-1?alpha(x[i]):+x[i]))) let(pin = (i==0 ? 0 : pp[i-1][j])) let(cin = (i==0 ? 0 : cp[i-1][j])) let(pout = (j==0 ? p[i] : pp[i][j-1])) let(cout = cp[i][j]) FullMulInt(xin,y[j],pin,cin,cout,pout); } // most significant digits of pp and y let(xin = sat{B}((i==M-1?alpha(x[i]):+x[i]))) let(pin = (i==0 ? 0 : pp[i-1][N-1])) let(pout = pp[i][N-2]) let(cout = pp[i][N-1]) FullMulBC(xin,y[N-1],pin,0,cout,pout); } // final addition is done by carry-ripple addition; in this final row, // the numbers pp[M-1][N-1..0] and cp[M-1][N-2..0] are added; // this part has depth O(N) due to the carry propagation for(j=0..N-2) let(cin = (j==0 ? 0 : cr[j-1])) FullAddInt(pp[M-1][j],cp[M-1][j],cin,cr[j],p[M+j]); p[M+N-1] = gamma(pp[M-1][N-1] + cr[N-2]); // check the result assert(intval(p,M+N) == intval(x,M) * intval(y,N)); } drivenby MinInt_MinInt { x[M-1] = B/2; for(i=0..M-2) x[i] = 0; y[N-1] = B/2; for(i=0..N-2) y[i] = 0; } drivenby MinInt_MaxInt { x[M-1] = B/2; for(i=0..M-2) x[i] = 0; y[N-1] = (B/2)-1; for(i=0..N-2) y[i] = B-1; } drivenby MaxInt_MinInt { x[M-1] = (B/2)-1; for(i=0..M-2) x[i] = B-1; y[N-1] = B/2; for(i=0..N-2) y[i] = 0; } drivenby MaxInt_MaxInt { x[M-1] = (B/2)-1; for(i=0..M-2) x[i] = B-1; y[N-1] = (B/2)-1; for(i=0..N-2) y[i] = B-1; } drivenby Test01 { for(i=0..M-1) x[i] = (-B/2-(i+1)) % B; for(i=0..N-1) y[i] = (-B/2-(i+1)) % B; } drivenby Test02 { for(i=0..M-1) x[i] = (-B/2-(i+1)) % B; for(i=0..N-1) y[i] = (-B-1-i) % B; } drivenby Test03 { for(i=0..M-1) x[i] = (-B-1-i) % B; for(i=0..N-1) y[i] = (-B/2-(i+1)) % B; } drivenby Test04 { for(i=0..M-1) x[i] = (-B-1-i) % B; for(i=0..N-1) y[i] = (-B-1-i) % B; }

|