<html lang="en"> <head> <title>Insn Canonicalizations - GNU Compiler Collection (GCC) Internals</title> <meta http-equiv="Content-Type" content="text/html"> <meta name="description" content="GNU Compiler Collection (GCC) Internals"> <meta name="generator" content="makeinfo 4.13"> <link title="Top" rel="start" href="index.html#Top"> <link rel="up" href="Machine-Desc.html#Machine-Desc" title="Machine Desc"> <link rel="prev" href="Looping-Patterns.html#Looping-Patterns" title="Looping Patterns"> <link rel="next" href="Expander-Definitions.html#Expander-Definitions" title="Expander Definitions"> <link href="http://www.gnu.org/software/texinfo/" rel="generator-home" title="Texinfo Homepage"> <!-- Copyright (C) 1988-2015 Free Software Foundation, Inc. Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.3 or any later version published by the Free Software Foundation; with the Invariant Sections being ``Funding Free Software'', the Front-Cover Texts being (a) (see below), and with the Back-Cover Texts being (b) (see below). A copy of the license is included in the section entitled ``GNU Free Documentation License''. (a) The FSF's Front-Cover Text is: A GNU Manual (b) The FSF's Back-Cover Text is: You have freedom to copy and modify this GNU Manual, like GNU software. Copies published by the Free Software Foundation raise funds for GNU development.--> <meta http-equiv="Content-Style-Type" content="text/css"> <style type="text/css"><!-- pre.display { font-family:inherit } pre.format { font-family:inherit } pre.smalldisplay { font-family:inherit; font-size:smaller } pre.smallformat { font-family:inherit; font-size:smaller } pre.smallexample { font-size:smaller } pre.smalllisp { font-size:smaller } span.sc { font-variant:small-caps } span.roman { font-family:serif; font-weight:normal; } span.sansserif { font-family:sans-serif; font-weight:normal; } --></style> </head> <body> <div class="node"> <a name="Insn-Canonicalizations"></a> <p> Next: <a rel="next" accesskey="n" href="Expander-Definitions.html#Expander-Definitions">Expander Definitions</a>, Previous: <a rel="previous" accesskey="p" href="Looping-Patterns.html#Looping-Patterns">Looping Patterns</a>, Up: <a rel="up" accesskey="u" href="Machine-Desc.html#Machine-Desc">Machine Desc</a> <hr> </div> <h3 class="section">16.14 Canonicalization of Instructions</h3> <p><a name="index-canonicalization-of-instructions-3679"></a><a name="index-insn-canonicalization-3680"></a> There are often cases where multiple RTL expressions could represent an operation performed by a single machine instruction. This situation is most commonly encountered with logical, branch, and multiply-accumulate instructions. In such cases, the compiler attempts to convert these multiple RTL expressions into a single canonical form to reduce the number of insn patterns required. <p>In addition to algebraic simplifications, following canonicalizations are performed: <ul> <li>For commutative and comparison operators, a constant is always made the second operand. If a machine only supports a constant as the second operand, only patterns that match a constant in the second operand need be supplied. <li>For associative operators, a sequence of operators will always chain to the left; for instance, only the left operand of an integer <code>plus</code> can itself be a <code>plus</code>. <code>and</code>, <code>ior</code>, <code>xor</code>, <code>plus</code>, <code>mult</code>, <code>smin</code>, <code>smax</code>, <code>umin</code>, and <code>umax</code> are associative when applied to integers, and sometimes to floating-point. <li><a name="index-g_t_0040code_007bneg_007d_002c-canonicalization-of-3681"></a><a name="index-g_t_0040code_007bnot_007d_002c-canonicalization-of-3682"></a><a name="index-g_t_0040code_007bmult_007d_002c-canonicalization-of-3683"></a><a name="index-g_t_0040code_007bplus_007d_002c-canonicalization-of-3684"></a><a name="index-g_t_0040code_007bminus_007d_002c-canonicalization-of-3685"></a>For these operators, if only one operand is a <code>neg</code>, <code>not</code>, <code>mult</code>, <code>plus</code>, or <code>minus</code> expression, it will be the first operand. <li>In combinations of <code>neg</code>, <code>mult</code>, <code>plus</code>, and <code>minus</code>, the <code>neg</code> operations (if any) will be moved inside the operations as far as possible. For instance, <code>(neg (mult A B))</code> is canonicalized as <code>(mult (neg A) B)</code>, but <code>(plus (mult (neg B) C) A)</code> is canonicalized as <code>(minus A (mult B C))</code>. <p><a name="index-g_t_0040code_007bcompare_007d_002c-canonicalization-of-3686"></a><li>For the <code>compare</code> operator, a constant is always the second operand if the first argument is a condition code register or <code>(cc0)</code>. <li>An operand of <code>neg</code>, <code>not</code>, <code>mult</code>, <code>plus</code>, or <code>minus</code> is made the first operand under the same conditions as above. <li><code>(ltu (plus </code><var>a</var> <var>b</var><code>) </code><var>b</var><code>)</code> is converted to <code>(ltu (plus </code><var>a</var> <var>b</var><code>) </code><var>a</var><code>)</code>. Likewise with <code>geu</code> instead of <code>ltu</code>. <li><code>(minus </code><var>x</var><code> (const_int </code><var>n</var><code>))</code> is converted to <code>(plus </code><var>x</var><code> (const_int </code><var>-n</var><code>))</code>. <li>Within address computations (i.e., inside <code>mem</code>), a left shift is converted into the appropriate multiplication by a power of two. <p><a name="index-g_t_0040code_007bior_007d_002c-canonicalization-of-3687"></a><a name="index-g_t_0040code_007band_007d_002c-canonicalization-of-3688"></a><a name="index-De-Morgan_0027s-law-3689"></a><li>De Morgan's Law is used to move bitwise negation inside a bitwise logical-and or logical-or operation. If this results in only one operand being a <code>not</code> expression, it will be the first one. <p>A machine that has an instruction that performs a bitwise logical-and of one operand with the bitwise negation of the other should specify the pattern for that instruction as <pre class="smallexample"> (define_insn "" [(set (match_operand:<var>m</var> 0 ...) (and:<var>m</var> (not:<var>m</var> (match_operand:<var>m</var> 1 ...)) (match_operand:<var>m</var> 2 ...)))] "..." "...") </pre> <p class="noindent">Similarly, a pattern for a “NAND” instruction should be written <pre class="smallexample"> (define_insn "" [(set (match_operand:<var>m</var> 0 ...) (ior:<var>m</var> (not:<var>m</var> (match_operand:<var>m</var> 1 ...)) (not:<var>m</var> (match_operand:<var>m</var> 2 ...))))] "..." "...") </pre> <p>In both cases, it is not necessary to include patterns for the many logically equivalent RTL expressions. <p><a name="index-g_t_0040code_007bxor_007d_002c-canonicalization-of-3690"></a><li>The only possible RTL expressions involving both bitwise exclusive-or and bitwise negation are <code>(xor:</code><var>m</var> <var>x</var> <var>y</var><code>)</code> and <code>(not:</code><var>m</var><code> (xor:</code><var>m</var> <var>x</var> <var>y</var><code>))</code>. <li>The sum of three items, one of which is a constant, will only appear in the form <pre class="smallexample"> (plus:<var>m</var> (plus:<var>m</var> <var>x</var> <var>y</var>) <var>constant</var>) </pre> <p><a name="index-g_t_0040code_007bzero_005fextract_007d_002c-canonicalization-of-3691"></a><a name="index-g_t_0040code_007bsign_005fextract_007d_002c-canonicalization-of-3692"></a><li>Equality comparisons of a group of bits (usually a single bit) with zero will be written using <code>zero_extract</code> rather than the equivalent <code>and</code> or <code>sign_extract</code> operations. <p><a name="index-g_t_0040code_007bmult_007d_002c-canonicalization-of-3693"></a><li><code>(sign_extend:</code><var>m1</var><code> (mult:</code><var>m2</var><code> (sign_extend:</code><var>m2</var> <var>x</var><code>) (sign_extend:</code><var>m2</var> <var>y</var><code>)))</code> is converted to <code>(mult:</code><var>m1</var><code> (sign_extend:</code><var>m1</var> <var>x</var><code>) (sign_extend:</code><var>m1</var> <var>y</var><code>))</code>, and likewise for <code>zero_extend</code>. <li><code>(sign_extend:</code><var>m1</var><code> (mult:</code><var>m2</var><code> (ashiftrt:</code><var>m2</var> <var>x</var> <var>s</var><code>) (sign_extend:</code><var>m2</var> <var>y</var><code>)))</code> is converted to <code>(mult:</code><var>m1</var><code> (sign_extend:</code><var>m1</var><code> (ashiftrt:</code><var>m2</var> <var>x</var> <var>s</var><code>)) (sign_extend:</code><var>m1</var> <var>y</var><code>))</code>, and likewise for patterns using <code>zero_extend</code> and <code>lshiftrt</code>. If the second operand of <code>mult</code> is also a shift, then that is extended also. This transformation is only applied when it can be proven that the original operation had sufficient precision to prevent overflow. </ul> <p>Further canonicalization rules are defined in the function <code>commutative_operand_precedence</code> in <samp><span class="file">gcc/rtlanal.c</span></samp>. </body></html>