Renesas HD6417641 Network Card User Manual


  Open as PDF
of 1036
 
Section 3 DSP Operation
Rev. 4.00 Sep. 14, 2005 Page 108 of 982
REJ09B0023-0400
Table 3.5 Variation of Fixed-Point Multiply Operation
Mnemonic Function Source 1 Source 2 Destination
PMULS Signed multiplication Se Sf Dg
Table 3.6 Correspondence between Operands and Registers
Register Se Sf Dg
A0   Yes
A1 Yes Yes Yes
M0 Yes
M1 Yes
X0 Yes Yes
X1 Yes
Y0 Yes Yes
Y1 Yes
Note: The multiply operations basically generate 32-bit operation results. So when a register
providing the guard-bit parts are specified as a destination operand, the guard-bit parts will
copy bit 31 of the operation result.
The multiply operation of the DSP unit side is not integer but fixed-point arithmetic. So, the upper
words of each multiplier and multiplicand are input into a MAC unit as shown in figure 3.8. In the
SH's standard multiply operations, the lower words of both source operands are input into a MAC
unit. The operation result is also different from the SH's case. The SH's multiply operation result is
aligned to the LSB of the destination, but the fixed-point multiply operation result is aligned to the
MSB, so that the LSB of the fixed-point multiply operation result is always 0.
This fixed-point multiply operation is executed in one cycle Multiply operation is always
unconditional, but does not affect any condition code bits, DC, N, Z, V and GT, in DSR.
Overflow Protection: The S bit in SR is effective for this multiply operation in the DSP unit. See
section 3.1.8, Overflow Protection, for details.
If the S bit is 0, overflow occurs only when H'8000*H'8000 ((-1.0)*(-1.0)) operation is
executed as signed fixed-point multiply. The result is H'00 8000 0000 but it does not mean
(+1.0). If the S bit is 1, overflow is prevented and the result is H'00 7FFF FFFF.