Bit-Slice Design: Controllers and ALUs

The ALU and Basic Arithmetic

Last Edit November 1, 1996; May 1, 1999 ; July 15, 2001

Sample Operations

Some operations will be examined fro the two-RALU slices, the Am2901 and the Am2903. [AMD also did an Am29203 that was released after this book was published.]

Increment

It is desired to increment a register and at the same time output its original value:

R'_A <-- R_A + 1; R_A --> out

Assume that the register is B₁₅ and that it has been previously loaded with a value. If it has not been previously loaded with a value, the contents are unknown and undetermined.

Am2901 Version

The Am2901 microcontrols are derived by examining each subfield table. The A and B addresses will both be F₁₆. The only way to output from a scratchpad is to select A with the output MUX. This leaves the B,0 pair as the source control.; therefore I₂I₁I₀ is set at 3₈. The ALU function is add or increment; therefore I₅I₄I₃ is set to 0₈, and CN is set to 1₈.

The result of the ADD is to be passed around from the ALU output F to the RAM shift register and then to the scratchpad memory, to R15. This is the destination control F--> B, A --> Y; therefore I₈I₇I₆ is set to 28₈. The complete microword for the Am2901 is:

The data flow is given in Figure 6-3.

Figure 6-3 Am2901: Increment a register and output its original value

Am2903 Version

The same problem approach is used with the Am2903. The only way to output directly from the scratchpad is via the DB_out path, which requires OE'_n be held LOW. The ALU will add A + 0; therefore I₄I₃I₂I₁ will be set to 6₁₆, C_in set to 1, and I₀ set to 0.

The A port input MUX select E'_A is set LOW. The A and B addresses are set to F₁₆, assuming that R₁₅ is in use. Since Y will be used to input back into the scratchpad, OE_Y = LOW and WE' will be driven by WRITE'/MSS'. The complete microword for the Am2903 is:

The data flow is given in Figure 6-4.

Figure 6-4 Am2903: Increment a register and output its original value

Byte Swap

A 16-bit byte swap exchanges bits B₁₅-B₈, with B₇-B₀ in order and visa versa.

Assume a 16-bit ALU (4 slices) and that the register is question is R₁₅.

Am2901 Version

The Am2901 microcode solution can be constructed by taking advantage of the fact that a shift up is equivalent to multiplying by 2. If the result of an add of R₁₅ + R₁₅ is shifted (prior to storing the result) into R₁₅, the effect is a 2-bit rotate, assuming that

1. the high order input, output of the RAM shift is connected to the low order input/output of the RAM
2. the carryout C_n+4 is connected to the carry-in C_in.

Since we are selecting source operands as A, B, adding, and shifting, 2F --> B; the microword is:

The microinstruction must be repeated four times to cause an 8-bit rotate, for a total timing of four microcycles of approximately 165 ns worst-case minimum (dependent on the actual hardware). [Remember that this was the late 1970s and early 1980s - times change.]

Am2903 Version

The Am2903 solution is approached in the same way. The S shift is connected for rotate as was the RAM shift of the Am2901. The carry out is input to the carry in, and an ADD is performed. The complete microword is:

Am2901 Hardware Version

The byte-swap operation can be performed faster using additional hardware (such as Am25LS240/244) tristate buffers (select non-inverting). The Am2901 Y outputs are gated out to the buffers, which in turn input to the appropriate D_A inputs of another Am2901. The operation performed by the Am2901 in this case is to output from the scratchpad, Y -> A, and to pass the D_A input through the ALU and back into the scratchpad memory, F -> B. The additional hardware allows the byte swap to be done in one microcycle. The microword is:

Am2903 Hardware Version

There are two ways to perform a hardware-assisted byte swap using the Am2903 and the Am25LS240/244 buffers.

One approach is to use the D_B as the output to the buffers and to use D_A as the input point. This is the relatively slower approach, since the D_A input passes the data through the ALU and back to the scratchpad.

The other approach uses the Y port as the data input, which provides a datapath directly into the scratchpad. This method avoids the propagation delay of the ALU, shift register, tristate buffer, and input MUX. The microwords are:

The basic difference is the OE_Y control.