SHIFT AND ROTATE
Shift and rotate instructions manipulate binary numbers at the binary bit level, as did the AND, OR, Exclusive-OR, and NOT instructions. Shifts and rotates find their most common applications in low-level software used to control I/O devices. The microprocessor contains a complete complement of shift and rotate instructions that are used to shift or rotate any memory data or register.
Shift
Shift instructions position or move numbers to the left or right within a register or memory location. They also perform simple arithmetic such as multiplication by powers of 2+n (left shift) and division by powers of 2–n (right shift). The microprocessor’s instruction set contains four different shift instructions: Two are logical shifts and two are arithmetic shifts. All four shift operations appear in Figure 5–9.
Notice in Figure 5–9 that there are two right shifts and two left shifts. The logical shifts move a 0 into the rightmost bit position for a logical left shift and a 0 into the leftmost bit position for a logical right shift. There are also two arithmetic shifts. The arithmetic shift left and logical left shift are identical. The arithmetic right shift and logical right shift are different because the arithmetic right shift copies the sign-bit through the number, whereas the logical right shift copies a 0 through the number.
Logical shift operations function with unsigned numbers, and arithmetic shifts function with signed numbers. Logical shifts multiply or divide unsigned data, and arithmetic shifts multiply or divide signed data. A shift left always multiplies by 2 for each bit position shifted, and a shift right always divides by 2 for each bit position shifted. Shifting a number two places, to the left or right, multiplies or divides by 4.
Table 5–22 illustrates some addressing modes allowed for the various shift instructions. There are two different forms of shifts that allow any register (except the segment register) or memory location to be shifted. One mode uses an immediate shift count, and the other uses register CL to hold the shift count. Note that CL must hold the shift count. When CL is the shift count, it does not change when the shift instruction executes. Note that the shift count is a modulo-32 count, which means that a shift count of 33 will shift the data one place (33>32 = remainder of 1). The same applies to a 64-bit number, but the shift count is modulo-64.
Example 5–30 shows how to shift the DX register left 14 places in two different ways. The first method uses an immediate shift count of 14. The second method loads 14 into CL and then uses CL as the shift count. Both instructions shift the contents of the DX register logically to the left 14 binary bit positions or places.
Suppose that the contents of AX must be multiplied by 10, as shown in Example 5–31. This can be done in two ways: by the MUL instruction or by shifts and additions. A number is doubled when it shifts left one place. When a number is doubled, and then added to the number times 8, the result is 10 times the number. The number 10 decimal is 1010 in binary. A logic 1 appears in both the 2’s and 8’s positions. If 2 times the number is added to 8 times the number, the result is 10 times the number. Using this technique, a program can be written to multiply by any constant. This technique often executes faster than the multiply instruction found in earlier versions of the Intel microprocessor.
Double-Precision Shifts (80386–Core2 Only). The 80386 and above contain two double precision shifts: SHLD (shift left) and SHRD (shift right). Each instruction contains three operands, instead of the two found with the other shift instructions. Both instructions function with two 16-or 32-bit registers, or with one 16- or 32-bit memory location and a register.
The SHRD AX,BX,12 instruction is an example of the double-precision shift right instruction. This instruction logically shifts AX right by 12 bit positions. The rightmost 12 bits of BX shift into the leftmost 12 bits of AX. The contents of BX remain unchanged by this instruction. The shift count can be an immediate count, as in this example, or it can be found in register CL, as with other shift instructions.
The SHLD EBX,ECX,16 instruction shifts EBX left. The leftmost 16 bits of ECX fill the rightmost 16 bits of EBX after the shift. As before, the contents of ECX, the second operand, remain unchanged. This instruction, as well as SHRD, affects the flag bits.
Rotate
Rotate instructions position binary data by rotating the information in a register or memory location, either from one end to another or through the carry flag. They are often used to shift or position numbers that are wider than 16 bits in the 8086–80286 microprocessors or wider than 32 bits in the 80386 through the Core2. The four available rotate instructions appear in Figure 5–10.
Numbers rotate through a register or memory location, through the C flag (carry), or through a register or memory location only. With either type of rotate instruction, the programmer can select either a left or a right rotate. Addressing modes used with rotate are the same as those used with shifts. A rotate count can be immediate or located in register CL. Table 5–23 lists some of the possible rotate instructions. If CL is used for a rotate count, it does not change. As with shifts, the count in CL is a modulo-32 count for a 32-bit operation and modulo-64 for a 64-bit operation.
Rotate instructions are often used to shift wide numbers to the left or right. The program listed in Example 5–32 shifts the 48-bit number in registers DX, BX, and AX left one binary place. Notice that the least significant 16 bits (AX) shift left first. This moves the leftmost bit of AX into the carry flag bit. Next, the rotate BX instruction rotates carry into BX, and its leftmost bit moves into carry. The last instruction rotates carry into DX, and the shift is complete.
Bit Scan Instructions
Although the bit scan instructions don’t shift or rotate numbers, they do scan through a number searching for a 1-bit. Because this is accomplished within the microprocessor by shifting the number, bit scan instructions are included in this section of the text.
The bit scan instructions BSF (bit scan forward) and BSR (bit scan reverse) are available only in the 80386–Pentium 4 processors. Both forms scan through the source number, searching for the first 1-bit. The BSF instruction scans the number from the leftmost bit toward the right, and BSR scans the number from the rightmost bit toward the left. If a 1-bit is encountered, the zero flag is set and the bit position number of the 1-bit is placed into the destination operand. If no 1-bit is encountered (i.e., the number contains all zeros), the zero flag is cleared. Thus, the result is not-zero if no 1-bit is encountered.
For example, if EAX = 60000000H and the BSF EBX,EAX instruction executes, the number is scanned from the leftmost bit toward the right. The first 1-bit encountered is at bit position 30, which is placed into EBX and the zero flag bit is set. If the same value for EAX is used for the BSR instruction, the EBX register is loaded with 29 and the zero flag bit is set.