The adder at the heart of Intel's 8087 floating-point chip
12 hours ago
- Intel released the 8087 floating-point coprocessor in 1980, featuring a 69-bit adder that accelerated math operations up to 100 times, including transcendental functions.
- The adder uses a Manchester carry chain combined with a carry-skip technique to speed up addition by breaking it into 4-bit blocks.
- Key concepts include Generate, Propagate, and Delete signals that allow parallel computation of carry values, with NMOS pass transistors and precharging for performance.
- The adder takes 69-bit inputs (including rounding bits) and produces a 70-bit output, requiring two clock cycles per addition due to block count.
- Surrounding circuitry like the Skip Shifter and Sum Shifter enables efficient multiplication, division, and square root operations in hardware.
- The adder's design balances complexity and performance, being simpler than later adders like the Kogge-Stone, but crucial for the 8087's arithmetic capabilities.