Xilinx Virtex-II Pro PPC405, User Manual

Page: 100 / 562

408 www.xilinx.com March 2002 Release
1-800-255-7778 Virtex-II Pro™ Platform FPGA Documentation
Chapter 3: User Programming Model
R
Multiply-Accumulate High-Halfword to Word Instructions
Table 3-41 shows the PPC405 multiply-accumulate high-halfword to word instructions. These
instructions multiply the high halfword of both source operands, r A[0:15] and r B[0:15],
producing a 32-bit product. The product is signed or unsigned, depending on the
instruction. This product is added to the value in the destination register,
r D, producing a
33-bit intermediate result. Generally, r D is loaded with the lower-32 bits of the 33-bit
intermediate result. However, if the instruction performs saturating arithmetic and the
intermediate result overflows,
r D is loaded with the nearest representable value (see
Modulo and Saturating Arithmetic , page 405 ).
For each type of instruction shown in Table 3-41 , the “Operation” column indicates the
multiply-accumulate operation performed. The column also shows, on an instruction-by-
instruction basis, how the XER and CR registers are updated (if at all).
Figure 3-28: Multiply-Accumulate Cross-Halfword to Word Operation
UG011_20_033101
r D
0 31
0 32
r A
0 31 16
r B
0 31 15
×
r D
0 31
+
Intermediate Result
1
Table 3-41: Multiply-Accumulate High-Halfword to Word Instructions
Mnemonic Name Operation
Operand
Syntax
Multiply-Accumulate High-Halfword to Word
Modulo Signed Instructions
r D is added to the signed product ( r A[0:15])
×
( r B[0:15]), producing
a 33-bit result. The low-32 bits of this result are stored in r D.
machhw Multiply Accumulate High Halfword
to Word Modulo Signed
XER and CR0 are
not
updated.
r D, r A, r B
machhw.
Multiply Accumulate High Halfword
to Word Modulo Signed and Record
CR0 is updated to reflect the result.
machhwo Multiply Accumulate High Halfword
to Word Modulo Signed with
Overflow Enabled
XER[OV,SO] are updated to reflect the result.
machhwo. Multiply Accumulate High Halfword
to Word Modulo Signed with
Overflow Enabled and Record
XER[OV,SO] and CR0 are updated to reflect the
result.