x87

The x87 is the floating-point unit (FPU) and instruction set used in Intel's x86 architecture. It originated with the 8087 coprocessor, released in 1987 to provide hardware support for floating-point calculations alongside the 8086/8088. In later generations, the FPU was integrated onto the same die as the processor (starting with the 80486 family), and the x87 instructions remained the primary means of performing floating-point operations in many software systems for years.

The x87 FPU uses a stack-based register model with eight 80-bit registers, designated ST0 through ST7. Operations typically operate on the top of the stack, and instructions may push, pop, or exchange stack elements. By default, the FPU uses 80-bit extended precision, with options to convert to 32-bit single or 64-bit double precision. The instruction set includes arithmetic operations (FADD, FSUB, FMUL, FDIV), comparisons (FCOM, FCOMI), data transfers, and transcendental functions, as well as control flow and data conversion instructions. The FPU maintains a control word, status word, and tag word that govern precision, rounding, exception masks, and stack state.

With the rise of SIMD floating-point extensions in the late 1990s, Intel introduced SSE and later SSE2, which provide separate 128-bit XMM registers for vector operations. These technologies offered higher throughput for most workloads, and compilers and libraries gradually shifted toward SSE-based code, though the x87 unit remained present for backward compatibility and certain extended-precision or complex-number routines.

Today, x87 remains backward compatible on modern CPUs, but it is largely superseded for new code by SSE and related technologies. Context switching and operating-system saving of the FPU state continue to be important considerations in multitasking environments.