解构 Intel 8087 的正切算法：超越 CORDIC 方法
Intel 8087 于 20 世纪 70 年代后期推出，是一款革命性的协处理器，专为 x86 架构处理浮点运算。在其众多复杂指令中，正切计算因其效率和精度而脱颖而出。虽然 CORDIC（坐标旋转数字计算机）算法常被提及为底层方法，但深入探究后发现，其实现更为精妙且经过优化。本文将逆向分析 8087 的正切算法，揭示其复杂之处，并将其与标准 CORDIC 方法进行比较。
CORDIC 算法：简要概述
在深入探讨 8087 的实现之前，理解 CORDIC 算法至关重要，它是嵌入式和计算机系统中三角函数和双曲函数计算的核心。CORDIC 算法利用迭代方法，仅通过简单的算术运算（加法、减法、乘法和位移）计算这些函数。其简单性使其非常适合计算资源有限的硬件实现。
CORDIC 算法的基本思想是利用一系列旋转来逼近所需角度。对于正切，算法通过预计算的系数迭代调整角度估计值，将其旋转至目标角度。该过程持续进行，直到达到所需精度。
逆向分析 8087 的正切实现
逆向分析 8087 的正切算法揭示了一种引人入胜的 CORDIC 原理与针对硬件能力的优化的结合。原始文章《逆向分析 Intel 8087 的正切算法：远超 CORDIC》的作者，通过仔细反汇编协处理器的微码，揭示了其内部工作原理。
关键观察
-
预计算系数：8087 使用一组预计算系数来加速迭代过程。这些系数存储在查找表中，在计算过程中被访问。这种方法减少了所需的乘法次数，考虑到 8087 的设计限制，这是一个关键的优化。
-
迭代细化：与标准 CORDIC 方法类似，80x87 通过迭代细化正切估计值。然而，迭代次数和每步的精度经过精心调整，以平衡准确性和性能。
-
硬件特定优化：8087 的实现利用了其硬件架构。例如，协处理器执行并行操作的能力以及其高效利用 x86 CPU 的算术逻辑单元（ALU），都促成了算法的速度。
示例代码片段
以下是一个简化的示例，说明 8087 可能如何使用其算法计算角度的正切：
; 假设 X 是以弧度为单位的角
; Y 是 tan(X) 的初始猜测值，设为 0
; Z 是 sin(X) 的初始猜测值，设为 1
START:
; 加载预计算系数
LOAD_COEFFICIENTS
; 对固定步数进行迭代
FOR I = 0 TO N:
; 检查角度是否为正或负
IF X >= 0:
Y = Y + Z * COEFF[I]
Z = Z + Y * COEFF[I]
ELSE:
Y = Y - Z * COEFF[I]
Z = Z - Y * COEFF[I]
ENDIF
X = X - Y
NEXT I
; 将结果存储在 Y (tan(X))
STORE_RESULT
END
这段伪代码说明了迭代过程，其中 COEFF 代表预计算系数。该算法根据 X 的符号调整 Y 和 Z，以确保角度的正确象限。
与标准 CORDIC 的比较
虽然 8087 的正切算法与标准 CORDIC 方法相似，但存在显著差异：
-
系数选择：8087 使用一组针对其特定硬件优化的不同系数。这些系数的选择旨在最小化所需的迭代次数，同时保持高精度。
-
迭代次数：标准 CORDIC 算法通常需要更多迭代才能达到与 8087 实现相同的精度水平。这是由于 8087 的硬件优化和定制的系数选择。
-
性能：8087 的正切计算比标准 CORDIC 实现更快、更高效。这部分是因为协处理器可以并行执行某些操作，并利用其专用硬件。
这为何重要
理解 8087 的正切算法为早期浮点协处理器的设计提供了宝贵的见解。它展示了硬件限制和优化技术如何导致比仅基于理论模型更高效的算法。这些知识在现代计算中尤为重要，因为性能和效率仍然是关键优先事项。
总结
Intel 8087 的正切算法虽然基于 CORDIC 方法，但展示了硬件特定优化的杰出范例。通过利用预计算系数、迭代细化和定制的硬件能力，8087 在三角函数计算中实现了速度和精度。这一见解不仅突出了早期计算机设计的巧妙之处，也为现代开发者在资源受限环境中优化性能提供了宝贵的经验教训。
Unpacking the Intel 8087's Tangent Algorithm: Beyond the CORDIC Method
The Intel 8087, introduced in the late 1970s, was a groundbreaking coprocessor designed to handle floating-point arithmetic for the x86 architecture. Among its many sophisticated instructions, the tangent calculation stood out for its efficiency and precision. While the CORDIC (Coordinate Rotation Digital Computer) algorithm is often cited as the underlying method, a closer look reveals a more nuanced and optimized approach. This exploration delves into the reverse-engineered tangent algorithm of the 8087, uncovering its intricacies and comparing it with the standard CORDIC method.
The CORDIC Algorithm: A Brief Overview
Before diving into the 8087's implementation, it's essential to understand the CORDIC algorithm, a cornerstone in embedded and computer systems for trigonometric and hyperbolic function calculations. The CORDIC algorithm leverages iterative methods to compute these functions using only simple arithmetic operations (addition, subtraction, multiplication, and bit shifts). Its simplicity makes it ideal for hardware implementations with limited computational resources.
The basic idea behind CORDIC is to use a sequence of rotations to approximate the desired angle. For tangent, the algorithm iteratively adjusts an angle estimate by rotating it towards the target angle using precomputed coefficients. The process continues until the desired precision is achieved.
Reverse-Engineering the 8087's Tangent Implementation
Reverse-engineering the 8087's tangent algorithm reveals a fascinating blend of CORDIC principles and optimizations tailored to the hardware's capabilities. The author of the original article, "Reverse-engineering the Intel 8087's tangent algorithm: more than CORDIC," meticulously disassembled the coprocessor's microcode to uncover its inner workings.
Key Observations
-
Precomputed Coefficients: The 8087 uses a set of precomputed coefficients to accelerate the iterative process. These coefficients are stored in a lookup table and accessed during the computation. This approach reduces the number of multiplications required, a critical optimization given the 8087's design constraints.
-
Iterative Refinement: Like the standard CORDIC method, the 80x87 refines the tangent estimate iteratively. However, the number of iterations and the precision of each step are carefully tailored to balance accuracy and performance.
-
Hardware-Specific Optimizations: The 8087's implementation takes advantage of its hardware architecture. For instance, the coprocessor's ability to perform parallel operations and its efficient use of the x86 CPU's arithmetic logic unit (ALU) contribute to the algorithm's speed.
Example Code Snippet
Here's a simplified example of how the 8087 might compute the tangent of an angle using its algorithm:
; Assume X is the angle in radians
; Y is the initial guess for tan(X), set to 0
; Z is the initial guess for sin(X), set to 1
START:
; Load precomputed coefficients
LOAD_COEFFICIENTS
; Iterate for a fixed number of steps
FOR I = 0 TO N:
; Check if the angle is positive or negative
IF X >= 0:
Y = Y + Z * COEFF[I]
Z = Z + Y * COEFF[I]
ELSE:
Y = Y - Z * COEFF[I]
Z = Z - Y * COEFF[I]
ENDIF
X = X - Y
NEXT I
; Store the result in Y (tan(X))
STORE_RESULT
END
This pseudocode illustrates the iterative process, where COEFF represents the precomputed coefficients. The algorithm adjusts Y and Z based on the sign of X to ensure the correct quadrant of the angle.
Comparing with Standard CORDIC
While the 8087's tangent algorithm shares similarities with the standard CORDIC method, there are notable differences:
-
Coefficient Selection: The 8087 uses a different set of coefficients optimized for its specific hardware. These coefficients are chosen to minimize the number of iterations required while maintaining high precision.
-
Iteration Count: The standard CORDIC algorithm typically requires more iterations to achieve the same level of precision as the 8087's implementation. This is due to the 8087's hardware optimizations and the tailored coefficient selection.
-
Performance: The 8087's tangent calculation is faster and more efficient than a standard CORDIC implementation. This is partly because the coprocessor can perform certain operations in parallel and leverages its specialized hardware.
Why This Matters
Understanding the 8087's tangent algorithm provides valuable insights into the design of early floating-point coprocessors. It demonstrates how hardware constraints and optimization techniques can lead to more efficient algorithms than those based on theoretical models alone. This knowledge is particularly relevant in modern computing, where performance and efficiency remain critical priorities.
Takeaway
The Intel 8087's tangent algorithm, while rooted in the CORDIC method, showcases a remarkable example of hardware-specific optimization. By leveraging precomputed coefficients, iterative refinement, and tailored hardware capabilities, the 8087 achieves both speed and precision in trigonometric calculations. This insight not only highlights the ingenuity of early computer design but also offers valuable lessons for modern developers striving to optimize performance in resource-constrained environments.
