Device for forming the exponent in the case of converting a binary number of the fixed point in the sliding point representation

摘要

1,076,207. Digital electric calculating apparatus. SPERRY RAND CORPORATION. Dec. 7, 1964 [Dec. 31, 1963], No. 49645/64. Heading G4A. In an arrangement for normalizing and generating the scale factor of a binary number, the number is shifted one place to the right and compared with the original number, the highest order bit position in which there is a difference providing an indication of the scale factor and hence the number of shifts required to normalize the number. The apparatus described can effect either one of two normalization operations on 36-bit (including sign bit) binary numbers: (1) " normalize floating point " in which the result is an 8-bit scale factor and a 27-bit mantissa, plus sign bit; (2) "normalize non- floating point " in which the result is a 7-bit scale factor and a 35-bit mantissa, plus sign bit. The arrangement has the advantage over known arrangements in which the number to be normalized is shifted, that the time of the normalization operation is independent of the position of the most significant digit in the number. In operation, for a " non-floating point " normalization, the number to be normalized is placed in an A register, Fig. 1 and gated directly to an X register and, with a shift of one place to the right, to a D register. The contents of the A and D registers are then compared, the comparison being effected by employing halfadders in the stages of a parallel adder 1 to compare the individual bits in the various stages of the two registers. The 35 significant bit outputs of the comparison operation are arranged in four 8-bit groups A-D and one 3-bit group E, an arrangement of group selectors A-D, Fig. 3 (not shown) and associated gates 90-93 being arranged to provide an output on one of five leads indicating which of the five groups contains the highest order comparison difference. The four leads from selectors A-E are connected to a group translator 100 (shown in detail in Fig. 5, not shown) which produces outputs on leads 94-96 representing the three highest order digits of the 7-bit scale factor number. The outputs of the group selector circuits also render operative one of five gating circuits in the input to a bit translator 101, thereby causing the comparison digits in the selected group to be translated by the translator 101 (which is shown in more detail in Fig. 5) to provide outputs on leads 97-99 indicating the three lowest order digits of the scale factor number. The scale factor number, which is called a " left shift count " since it indicates the number of single shifts to the left required, is subtracted from the value 72 in a subtracter 103 to provide a " right shift count ", since the shift matrix, top right of Fig. 1, can effect directly only a shift to the right by up to 72 places. The left shift is entered in a register K1, Fig. 1, and the right shift count in a register K3 where its seven bits control the shift matrix, which comprises three gating levels responsive respectively to stages 0, 1; 2, 3; 4, 5, 6 of the register K3. The number to be normalized is now transferred over a bus 41 to the shift matrix where it is shifted by gating by the appropriate number of places to enter the significant digits in the high order places of the D register. The scale factor K2 is then transferred, to the X register. A floating point normalization is generally similar, but a right shift instead of a left shift may be necessary if the highest significant digit is in any of bit positions 27-34. If a left shift is necessary, a positive sign in the number being normalized is effective to complement the scale factor register K2 by subtraction from zero at 5, Fig. 1. The sign bit and scale factor are transmitted to bit positions 35, 27-34, respectively, in the D register to produce the normalized floating point number. If a right shift is necessary in a floating point normalization, the right shift count is supplied to register K3 as before, and its complement is transferred to the scale factor register K1, the sign of the original number in the A register determining whether a further complementation is necessary to produce the correct scale factor.