One is the MAC instruction which can be used in the entire SuperH family.
The MAC instruction transfers data, multiplies that data and executes accumulation with the MAC register in 1 instruction.
The instructions and data use a common bus, so it requires 2 or 3 clock cycles.
However, the SH3-DSP DSP function transfers 2 data items from memory to the DSP data register in 1 clock cycle, at multiplying between DSP data registers in 1 clock cycle and accumulating in 1 clock cycle.
It takes 3 clock cycles to complete multiplication and accumulation for the initial data, but data operations are pipelined so subsequent multiplication and accumulation results are obtained in 1 clock cycle. This is the same performance at the general DSP.
The instruction that uses the DSP function specifies 2 data transfers and multiplication operations, and subtraction and addition circuit operations.
For that reason the DSP function instruction is extended to 32 bits.