Reid Spencer | e00906f | 2006-08-10 20:15:58 +0000 | [diff] [blame^] | 1 | <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN" |
| 2 | "http://www.w3.org/TR/html4/strict.dtd"> |
| 3 | <html> |
| 4 | <head> |
| 5 | <meta http-equiv="Content-Type" content="text/html; charset=utf-8"> |
| 6 | <title>The Often Misunderstood GEP Instruction</title> |
| 7 | <link rel="stylesheet" href="llvm.css" type="text/css"> |
| 8 | </head> |
| 9 | <body> |
| 10 | |
| 11 | <div class="doc_title"> |
| 12 | The Often Misunderstood GEP Instruction |
| 13 | </div> |
| 14 | |
| 15 | <ol> |
| 16 | <li><a href="#intro">Introduction</a></li> |
| 17 | <li><a href="#questions">The Questions</a> |
| 18 | <ol> |
| 19 | <li><a href="#extra_index">Why is the extra 0 index required?</a></li> |
| 20 | <li><a href="#deref">What is dereferenced by GEP?</a></li> |
| 21 | <li><a href="#firstptr">Why can you index through the first pointer but not |
| 22 | subsequent ones?</a></li> |
| 23 | <li><a href="#lead0">Why don't GEP x,0,0,1 and GEP x,1 alias? </a></li> |
| 24 | <li><a href="#trail0">Why do GEP x,1,0,0 and GEP x,1 alias? </a></li> |
| 25 | </ol></li> |
| 26 | <li><a href="#summary">Summary</a></li> |
| 27 | </ol> |
| 28 | |
| 29 | <div class="doc_author"> |
| 30 | <p>Written by: <a href="mailto:rspencer@reidspencer.com">Reid Spencer</a>.</p> |
| 31 | </div> |
| 32 | |
| 33 | |
| 34 | <!-- *********************************************************************** --> |
| 35 | <div class="doc_section"><a name="intro"><b>Introduction</b></a></div> |
| 36 | <!-- *********************************************************************** --> |
| 37 | <div class="doc_text"> |
| 38 | <p>This document seeks to dispel the mystery and confusion surrounding LLVM's |
| 39 | GetElementPtr (GEP) instruction. Questions about the wiley GEP instruction are |
| 40 | probably the most frequently occuring questions once a developer gets down to |
| 41 | coding with LLVM. Here we lay out the sources of confusion and show that the |
| 42 | GEP instruction is really quite simple. |
| 43 | </p> |
| 44 | </div> |
| 45 | |
| 46 | <!-- *********************************************************************** --> |
| 47 | <div class="doc_section"><a name="questions"><b>The Questions</b></a></div> |
| 48 | <!-- *********************************************************************** --> |
| 49 | <div class="doc_text"> |
| 50 | <p>When people are first confronted with the GEP instruction, they tend to |
| 51 | relate it to known concepts from other programming paradigms, most notably C |
| 52 | array indexing and field selection. However, GEP is a little different and |
| 53 | this leads to the following questions, all of which are answered in the |
| 54 | following sections.</p> |
| 55 | <ol> |
| 56 | <li><a href="extra_index">Why is the extra 0 index required?</a></li> |
| 57 | <li><a href="deref">What is dereferenced by GEP?</a></li> |
| 58 | <li><a href="firstptr">Why can you index through the first pointer but not |
| 59 | subsequent ones?</a></li> |
| 60 | <li><a href="lead0">Why don't GEP x,0,0,1 and GEP x,1 alias? </a></li> |
| 61 | <li><a href="trail0">Why do GEP x,1,0,0 and GEP x,1 alias? </a></li> |
| 62 | </ol> |
| 63 | </div> |
| 64 | |
| 65 | <!-- *********************************************************************** --> |
| 66 | <div class="doc_subsection"> |
| 67 | <a name="extra_index"><b>Why is the extra 0 index required?</b></a> |
| 68 | </div> |
| 69 | <!-- *********************************************************************** --> |
| 70 | <div class="doc_text"> |
| 71 | <p>Quick answer: there are no superfluous indices.</p> |
| 72 | <p>This question arises most often when the GEP instruction is applied to a |
| 73 | global variable which is always a pointer type. For example, consider |
| 74 | this:</p><pre> |
| 75 | %MyStruct = uninitialized global { float*, int } |
| 76 | ... |
| 77 | %idx = getelementptr { float*, int }* %MyStruct, long 0, ubyte 1</pre> |
| 78 | <p>The GEP above yields an <tt>int*</tt> by indexing the <tt>int</tt> typed |
| 79 | field of the structure <tt>%MyStruct</tt>. When people first look at it, they |
| 80 | wonder why the <tt>long 0</tt> index is needed. However, a closer inspection |
| 81 | of how globals and GEPs work reveals the need. Becoming aware of the following |
| 82 | facts will dispell the confusion:</p> |
| 83 | <ol> |
| 84 | <li>The type of <tt>%MyStruct</tt> is <i>not</i> <tt>{ float*, int }</tt> |
| 85 | but rather <tt>{ float*, int }*</tt>. That is, <tt>%MyStruct</tt> is a |
| 86 | pointer to a structure containing a pointer to a <tt>float</tt> and an |
| 87 | <tt>int</tt>.</li> |
| 88 | <li>Point #1 is evidenced by noticing the type of the first operand of |
| 89 | the GEP instruction (<tt>%MyStruct</tt>) which is |
| 90 | <tt>{ float*, int }*</tt>.</li> |
| 91 | <li>The first index, <tt>long 0</tt> is required to dereference the |
| 92 | pointer associated with <tt>%MyStruct</tt>.</li> |
| 93 | <li>The second index, <tt>ubyte 1</tt> selects the second field of the |
| 94 | structure (the <tt>int</tt>). </li> |
| 95 | </ol> |
| 96 | </div> |
| 97 | |
| 98 | <!-- *********************************************************************** --> |
| 99 | <div class="doc_subsection"> |
| 100 | <a name="deref"><b>What is dereferenced by GEP?</b></a> |
| 101 | </div> |
| 102 | <div class="doc_text"> |
| 103 | <p>Quick answer: nothing.</p> |
| 104 | <p>The GetElementPtr instruction dereferences nothing. That is, it doesn't |
| 105 | access memory in any way. That's what the Load instruction is for. GEP is |
| 106 | only involved in the computation of addresses. For example, consider this:</p> |
| 107 | <pre> |
| 108 | %MyVar = uninitialized global { [40 x int ]* } |
| 109 | ... |
| 110 | %idx = getelementptr { [40 x int]* }* %MyVar, long 0, ubyte 0, long 0, long 17</pre> |
| 111 | <p>In this example, we have a global variable, <tt>%MyVar</tt> that is a |
| 112 | pointer to a structure containing a pointer to an array of 40 ints. The |
| 113 | GEP instruction seems to be accessing the 18th integer of of the structure's |
| 114 | array of ints. However, this is actually an illegal GEP instruction. It |
| 115 | won't compile. The reason is that the pointer in the structure <i>must</i> |
| 116 | be dereferenced in order to index into the array of 40 ints. Since the |
| 117 | GEP instruction never accesses memory, it is illegal.</p> |
| 118 | <p>In order to access the 18th integer in the array, you would need to do the |
| 119 | following:</p> |
| 120 | <pre> |
| 121 | %idx = getelementptr { [40 x int]* }* %, long 0, ubyte 0 |
| 122 | %arr = load [40 x int]** %idx |
| 123 | %idx = getelementptr [40 x int]* %arr, long 0, long 17</pre> |
| 124 | <p>In this case, we have to load the pointer in the structure with a load |
| 125 | instruction before we can index into the array. If the example was changed |
| 126 | to:</p> |
| 127 | <pre> |
| 128 | %MyVar = uninitialized global { [40 x int ] } |
| 129 | ... |
| 130 | %idx = getelementptr { [40 x int] }*, long 0, ubyte 0, long 17</pre> |
| 131 | <p>then everything works fine. In this case, the structure does not contain a |
| 132 | pointer and the GEP instruction can index through the global variable pointer, |
| 133 | into the first field of the structure and access the 18th <tt>int</tt> in the |
| 134 | array there.</p> |
| 135 | </div> |
| 136 | |
| 137 | <!-- *********************************************************************** --> |
| 138 | <div class="doc_subsection"> |
| 139 | <a name="firstptr"><b>Why can you index through the first pointer?</b></a> |
| 140 | </div> |
| 141 | <div class="doc_text"> |
| 142 | <p>Quick answer: Because its already present.</p> |
| 143 | <p>Having understood the <a href="#deref">previous question</a>, a new |
| 144 | question then arises:</p> |
| 145 | <blockquote><i>Why is it okay to index through the first pointer, but |
| 146 | subsequent pointers won't be dereferenced?</i></blockquote> |
| 147 | <p>The answer is simply because |
| 148 | memory does not have to be accessed to perform the computation. The first |
| 149 | operand to the GEP instruction must be a value of a pointer type. The value |
| 150 | of the pointer is provided directly to the GEP instruction without any need |
| 151 | for accessing memory. It must, therefore be indexed like any other operand. |
| 152 | Consider this example:</p> |
| 153 | <pre> |
| 154 | %MyVar = unintialized global int |
| 155 | ... |
| 156 | %idx1 = getelementptr int* %MyVar, long 0 |
| 157 | %idx2 = getelementptr int* %MyVar, long 1 |
| 158 | %idx3 = getelementptr int* %MyVar, long 2</pre> |
| 159 | <p>These GEP instructions are simply making address computations from the |
| 160 | base address of <tt>MyVar</tt>. They compute, as follows (using C syntax):</p> |
| 161 | <ul> |
| 162 | <li> idx1 = &MyVar + 0</li> |
| 163 | <li> idx2 = &MyVar + 4</li> |
| 164 | <li> idx3 = &MyVar = 8</li> |
| 165 | </ul> |
| 166 | <p>Since the type <tt>int</tt> is known to be four bytes long, the indices |
| 167 | 0, 1 and 2 translate into memory offsets of 0, 4, and 8, respectively. No |
| 168 | memory is accessed to make these computations because the address of |
| 169 | <tt>%MyVar</tt> is passed directly to the GEP instructions.</p> |
| 170 | <p>Note that the cases of <tt>%idx2</tt> and <tt>%idx3</tt> are a bit silly. |
| 171 | They are computing addresses of something of unknown type (and thus |
| 172 | potentially breaking type safety) because <tt>%MyVar</tt> is only one |
| 173 | integer long.</p> |
| 174 | </div> |
| 175 | |
| 176 | <!-- *********************************************************************** --> |
| 177 | <div class="doc_subsection"> |
| 178 | <a name="lead0"><b>Why don't GEP x,0,0,1 and GEP x,1 alias?</b></a> |
| 179 | </div> |
| 180 | <div class="doc_text"> |
| 181 | <p>Quick Answer: They compute different address locations.</p> |
| 182 | <p>If you look at the first indices in these GEP |
| 183 | instructions you find that they are different (0 and 1), therefore the address |
| 184 | computation diverges with that index. Consider this example:</p> |
| 185 | <pre> |
| 186 | %MyVar = global { [10 x int ] } |
| 187 | %idx1 = getlementptr { [10 x int ] }* %MyVar, long 0, byte 0, long 1 |
| 188 | %idx2 = getlementptr { [10 x int ] }* %MyVar, long 1</pre> |
| 189 | <p>In this example, <tt>idx1</tt> computes the address of the second integer |
| 190 | in the array that is in the structure in %MyVar, that is <tt>MyVar+4</tt>. The |
| 191 | type of <tt>idx1</tt> is <tt>int*</tt>. However, <tt>idx2</tt> computes the |
| 192 | address of <i>the next</i> structure after <tt>%MyVar</tt>. The type of |
| 193 | <tt>idx2</tt> is <tt>{ [10 x int] }*</tt> and its value is equivalent |
| 194 | to <tt>MyVar + 40</tt> because it indexes past the ten 4-byte integers |
| 195 | in <tt>MyVar</tt>. Obviously, in such a situation, the pointers don't |
| 196 | alias.</p> |
| 197 | </div> |
| 198 | |
| 199 | <!-- *********************************************************************** --> |
| 200 | <div class="doc_subsection"> |
| 201 | <a name="lead0"><b>Why do GEP x,1,0,0 and GEP x,1 alias?</b></a> |
| 202 | </div> |
| 203 | <div class="doc_text"> |
| 204 | <p>Quick Answer: They compute the same address location.</p> |
| 205 | <p>These two GEP instructions will compute the same address because indexing |
| 206 | through the 0th element does not change the address. However, it does change |
| 207 | the type. Consider this example:</p> |
| 208 | <pre> |
| 209 | %MyVar = global { [10 x int ] } |
| 210 | %idx1 = getlementptr { [10 x int ] }* %MyVar, long 1, byte 0, long 0 |
| 211 | %idx2 = getlementptr { [10 x int ] }* %MyVar, long 1</pre> |
| 212 | <p>In this example, the value of <tt>%idx1</tt> is <tt>%MyVar+40</tt> and |
| 213 | its type is <tt>int*</tt>. The value of <tt>%idx2</tt> is also |
| 214 | <tt>MyVar+40</tt> but its type is <tt>{ [10 x int] }*</tt>.</p> |
| 215 | </div> |
| 216 | |
| 217 | <!-- *********************************************************************** --> |
| 218 | <div class="doc_section"><a name="summary"><b>Summary</b></a></div> |
| 219 | <!-- *********************************************************************** --> |
| 220 | |
| 221 | <div class="doc_text"> |
| 222 | <p>In summary, here's some things to always remember about the GetElementPtr |
| 223 | instruction:</p> |
| 224 | <ol> |
| 225 | <li>The GEP instruction never accesses memory, it only provides pointer |
| 226 | computations.</li> |
| 227 | <li>The first operand to the GEP instruction is always a pointer and it must |
| 228 | be indexed.</li> |
| 229 | <li>There are no superfluous indices for the GEP instruction.</li> |
| 230 | <li>Trailing zero indices are superfluous for pointer aliasing, but not for |
| 231 | the types of the pointers.</li> |
| 232 | <li>Leading zero indices are not superfluous for pointer aliasing nor the |
| 233 | types of the pointers.</li> |
| 234 | </ol> |
| 235 | </div> |
| 236 | |
| 237 | <!-- *********************************************************************** --> |
| 238 | |
| 239 | <hr> |
| 240 | <address> |
| 241 | <a href="http://jigsaw.w3.org/css-validator/check/referer"><img |
| 242 | src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a> |
| 243 | <a href="http://validator.w3.org/check/referer"><img |
| 244 | src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a> |
| 245 | <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br/> |
| 246 | Last modified: $Date$ |
| 247 | </address> |
| 248 | </body> |
| 249 | </html> |