1 <!DOCTYPE HTML PUBLIC
"-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
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 <style type=
"text/css">
9 TABLE
{ text-align: left
; border: 1px solid black
; border-collapse: collapse
; margin: 0 0 0 0; }
14 <div class=
"doc_title">
15 The Often Misunderstood GEP Instruction
19 <li><a href=
"#intro">Introduction
</a></li>
20 <li><a href=
"#questions">The Questions
</a>
22 <li><a href=
"#extra_index">Why is the extra
0 index required?
</a></li>
23 <li><a href=
"#deref">What is dereferenced by GEP?
</a></li>
24 <li><a href=
"#firstptr">Why can you index through the first pointer but not
25 subsequent ones?
</a></li>
26 <li><a href=
"#lead0">Why don't GEP x,
0,
0,
1 and GEP x,
1 alias?
</a></li>
27 <li><a href=
"#trail0">Why do GEP x,
1,
0,
0 and GEP x,
1 alias?
</a></li>
29 <li><a href=
"#summary">Summary
</a></li>
32 <div class=
"doc_author">
33 <p>Written by:
<a href=
"mailto:rspencer@reidspencer.com">Reid Spencer
</a>.
</p>
37 <!-- *********************************************************************** -->
38 <div class=
"doc_section"><a name=
"intro"><b>Introduction
</b></a></div>
39 <!-- *********************************************************************** -->
40 <div class=
"doc_text">
41 <p>This document seeks to dispel the mystery and confusion surrounding LLVM's
42 GetElementPtr (GEP) instruction. Questions about the wiley GEP instruction are
43 probably the most frequently occuring questions once a developer gets down to
44 coding with LLVM. Here we lay out the sources of confusion and show that the
45 GEP instruction is really quite simple.
49 <!-- *********************************************************************** -->
50 <div class=
"doc_section"><a name=
"questions"><b>The Questions
</b></a></div>
51 <!-- *********************************************************************** -->
52 <div class=
"doc_text">
53 <p>When people are first confronted with the GEP instruction, they tend to
54 relate it to known concepts from other programming paradigms, most notably C
55 array indexing and field selection. However, GEP is a little different and
56 this leads to the following questions; all of which are answered in the
57 following sections.
</p>
59 <li><a href=
"#firstptr">What is the first index of the GEP instruction?
</a>
61 <li><a href=
"#extra_index">Why is the extra
0 index required?
</a></li>
62 <li><a href=
"#deref">What is dereferenced by GEP?
</a></li>
63 <li><a href=
"#lead0">Why don't GEP x,
0,
0,
1 and GEP x,
1 alias?
</a></li>
64 <li><a href=
"#trail0">Why do GEP x,
1,
0,
0 and GEP x,
1 alias?
</a></li>
68 <!-- *********************************************************************** -->
69 <div class=
"doc_subsection">
70 <a name=
"firstptr"><b>What is the first index of the GEP instruction?
</b></a>
72 <div class=
"doc_text">
73 <p>Quick answer: The index stepping through the first operand.
</p>
74 <p>The confusion with the first index usually arises from thinking about
75 the GetElementPtr instruction as if it was a C index operator. They aren't the
76 same. For example, when we write, in
"C":
</p>
78 <div class=
"doc_code">
86 <p>it is natural to think that there is only one index, the selection of the
87 field
<tt>F
</tt>. However, in this example,
<tt>Foo
</tt> is a pointer. That
88 pointer must be indexed explicitly in LLVM. C, on the other hand, indexs
89 through it transparently. To arrive at the same address location as the C
90 code, you would provide the GEP instruction with two index operands. The
91 first operand indexes through the pointer; the second operand indexes the
92 field
<tt>F
</tt> of the structure, just as if you wrote:
</p>
94 <div class=
"doc_code">
100 <p>Sometimes this question gets rephrased as:
</p>
101 <blockquote><p><i>Why is it okay to index through the first pointer, but
102 subsequent pointers won't be dereferenced?
</i></p></blockquote>
103 <p>The answer is simply because memory does not have to be accessed to
104 perform the computation. The first operand to the GEP instruction must be a
105 value of a pointer type. The value of the pointer is provided directly to
106 the GEP instruction as an operand without any need for accessing memory. It
107 must, therefore be indexed and requires an index operand. Consider this
110 <div class=
"doc_code">
112 struct munger_struct {
116 void munge(struct munger_struct *P) {
117 P[
0].f1 = P[
1].f1 + P[
2].f2;
120 munger_struct Array[
3];
126 <p>In this
"C" example, the front end compiler (llvm-gcc) will generate three
127 GEP instructions for the three indices through
"P" in the assignment
128 statement. The function argument
<tt>P
</tt> will be the first operand of each
129 of these GEP instructions. The second operand indexes through that pointer.
130 The third operand will be the field offset into the
131 <tt>struct munger_struct
</tt> type, for either the
<tt>f1
</tt> or
132 <tt>f2
</tt> field. So, in LLVM assembly the
<tt>munge
</tt> function looks
135 <div class=
"doc_code">
137 void %munge(%struct.munger_struct* %P) {
139 %tmp = getelementptr %struct.munger_struct* %P, i32
1, i32
0
140 %tmp = load i32* %tmp
141 %tmp6 = getelementptr %struct.munger_struct* %P, i32
2, i32
1
142 %tmp7 = load i32* %tmp6
143 %tmp8 = add i32 %tmp7, %tmp
144 %tmp9 = getelementptr %struct.munger_struct* %P, i32
0, i32
0
145 store i32 %tmp8, i32* %tmp9
151 <p>In each case the first operand is the pointer through which the GEP
152 instruction starts. The same is true whether the first operand is an
153 argument, allocated memory, or a global variable.
</p>
154 <p>To make this clear, let's consider a more obtuse example:
</p>
156 <div class=
"doc_code">
158 %MyVar = unintialized global i32
160 %idx1 = getelementptr i32* %MyVar, i64
0
161 %idx2 = getelementptr i32* %MyVar, i64
1
162 %idx3 = getelementptr i32* %MyVar, i64
2
166 <p>These GEP instructions are simply making address computations from the
167 base address of
<tt>MyVar
</tt>. They compute, as follows (using C syntax):
170 <div class=
"doc_code">
172 idx1 = (char*)
&MyVar +
0
173 idx2 = (char*)
&MyVar +
4
174 idx3 = (char*)
&MyVar +
8
178 <p>Since the type
<tt>i32
</tt> is known to be four bytes long, the indices
179 0,
1 and
2 translate into memory offsets of
0,
4, and
8, respectively. No
180 memory is accessed to make these computations because the address of
181 <tt>%MyVar
</tt> is passed directly to the GEP instructions.
</p>
182 <p>The obtuse part of this example is in the cases of
<tt>%idx2
</tt> and
183 <tt>%idx3
</tt>. They result in the computation of addresses that point to
184 memory past the end of the
<tt>%MyVar
</tt> global, which is only one
185 <tt>i32
</tt> long, not three
<tt>i32
</tt>s long. While this is legal in LLVM,
186 it is inadvisable because any load or store with the pointer that results
187 from these GEP instructions would produce undefined results.
</p>
190 <!-- *********************************************************************** -->
191 <div class=
"doc_subsection">
192 <a name=
"extra_index"><b>Why is the extra
0 index required?
</b></a>
194 <!-- *********************************************************************** -->
195 <div class=
"doc_text">
196 <p>Quick answer: there are no superfluous indices.
</p>
197 <p>This question arises most often when the GEP instruction is applied to a
198 global variable which is always a pointer type. For example, consider
201 <div class=
"doc_code">
203 %MyStruct = uninitialized global { float*, i32 }
205 %idx = getelementptr { float*, i32 }* %MyStruct, i64
0, i32
1
209 <p>The GEP above yields an
<tt>i32*
</tt> by indexing the
<tt>i32
</tt> typed
210 field of the structure
<tt>%MyStruct
</tt>. When people first look at it, they
211 wonder why the
<tt>i64
0</tt> index is needed. However, a closer inspection
212 of how globals and GEPs work reveals the need. Becoming aware of the following
213 facts will dispell the confusion:
</p>
215 <li>The type of
<tt>%MyStruct
</tt> is
<i>not
</i> <tt>{ float*, i32 }
</tt>
216 but rather
<tt>{ float*, i32 }*
</tt>. That is,
<tt>%MyStruct
</tt> is a
217 pointer to a structure containing a pointer to a
<tt>float
</tt> and an
219 <li>Point #
1 is evidenced by noticing the type of the first operand of
220 the GEP instruction (
<tt>%MyStruct
</tt>) which is
221 <tt>{ float*, i32 }*
</tt>.
</li>
222 <li>The first index,
<tt>i64
0</tt> is required to step over the global
223 variable
<tt>%MyStruct
</tt>. Since the first argument to the GEP
224 instruction must always be a value of pointer type, the first index
225 steps through that pointer. A value of
0 means
0 elements offset from that
227 <li>The second index,
<tt>i32
1</tt> selects the second field of the
228 structure (the
<tt>i32
</tt>).
</li>
232 <!-- *********************************************************************** -->
233 <div class=
"doc_subsection">
234 <a name=
"deref"><b>What is dereferenced by GEP?
</b></a>
236 <div class=
"doc_text">
237 <p>Quick answer: nothing.
</p>
238 <p>The GetElementPtr instruction dereferences nothing. That is, it doesn't
239 access memory in any way. That's what the Load and Store instructions are for.
240 GEP is only involved in the computation of addresses. For example, consider
243 <div class=
"doc_code">
245 %MyVar = uninitialized global { [
40 x i32 ]* }
247 %idx = getelementptr { [
40 x i32]* }* %MyVar, i64
0, i32
0, i64
0, i64
17
251 <p>In this example, we have a global variable,
<tt>%MyVar
</tt> that is a
252 pointer to a structure containing a pointer to an array of
40 ints. The
253 GEP instruction seems to be accessing the
18th integer of the structure's
254 array of ints. However, this is actually an illegal GEP instruction. It
255 won't compile. The reason is that the pointer in the structure
<i>must
</i>
256 be dereferenced in order to index into the array of
40 ints. Since the
257 GEP instruction never accesses memory, it is illegal.
</p>
258 <p>In order to access the
18th integer in the array, you would need to do the
261 <div class=
"doc_code">
263 %idx = getelementptr { [
40 x i32]* }* %, i64
0, i32
0
264 %arr = load [
40 x i32]** %idx
265 %idx = getelementptr [
40 x i32]* %arr, i64
0, i64
17
269 <p>In this case, we have to load the pointer in the structure with a load
270 instruction before we can index into the array. If the example was changed
273 <div class=
"doc_code">
275 %MyVar = uninitialized global { [
40 x i32 ] }
277 %idx = getelementptr { [
40 x i32] }*, i64
0, i32
0, i64
17
281 <p>then everything works fine. In this case, the structure does not contain a
282 pointer and the GEP instruction can index through the global variable,
283 into the first field of the structure and access the
18th
<tt>i32
</tt> in the
287 <!-- *********************************************************************** -->
288 <div class=
"doc_subsection">
289 <a name=
"lead0"><b>Why don't GEP x,
0,
0,
1 and GEP x,
1 alias?
</b></a>
291 <div class=
"doc_text">
292 <p>Quick Answer: They compute different address locations.
</p>
293 <p>If you look at the first indices in these GEP
294 instructions you find that they are different (
0 and
1), therefore the address
295 computation diverges with that index. Consider this example:
</p>
297 <div class=
"doc_code">
299 %MyVar = global { [
10 x i32 ] }
300 %idx1 = getlementptr { [
10 x i32 ] }* %MyVar, i64
0, i32
0, i64
1
301 %idx2 = getlementptr { [
10 x i32 ] }* %MyVar, i64
1
305 <p>In this example,
<tt>idx1
</tt> computes the address of the second integer
306 in the array that is in the structure in %MyVar, that is
<tt>MyVar+
4</tt>. The
307 type of
<tt>idx1
</tt> is
<tt>i32*
</tt>. However,
<tt>idx2
</tt> computes the
308 address of
<i>the next
</i> structure after
<tt>%MyVar
</tt>. The type of
309 <tt>idx2
</tt> is
<tt>{ [
10 x i32] }*
</tt> and its value is equivalent
310 to
<tt>MyVar +
40</tt> because it indexes past the ten
4-byte integers
311 in
<tt>MyVar
</tt>. Obviously, in such a situation, the pointers don't
315 <!-- *********************************************************************** -->
316 <div class=
"doc_subsection">
317 <a name=
"trail0"><b>Why do GEP x,
1,
0,
0 and GEP x,
1 alias?
</b></a>
319 <div class=
"doc_text">
320 <p>Quick Answer: They compute the same address location.
</p>
321 <p>These two GEP instructions will compute the same address because indexing
322 through the
0th element does not change the address. However, it does change
323 the type. Consider this example:
</p>
325 <div class=
"doc_code">
327 %MyVar = global { [
10 x i32 ] }
328 %idx1 = getlementptr { [
10 x i32 ] }* %MyVar, i64
1, i32
0, i64
0
329 %idx2 = getlementptr { [
10 x i32 ] }* %MyVar, i64
1
333 <p>In this example, the value of
<tt>%idx1
</tt> is
<tt>%MyVar+
40</tt> and
334 its type is
<tt>i32*
</tt>. The value of
<tt>%idx2
</tt> is also
335 <tt>MyVar+
40</tt> but its type is
<tt>{ [
10 x i32] }*
</tt>.
</p>
338 <!-- *********************************************************************** -->
339 <div class=
"doc_section"><a name=
"summary"><b>Summary
</b></a></div>
340 <!-- *********************************************************************** -->
342 <div class=
"doc_text">
343 <p>In summary, here's some things to always remember about the GetElementPtr
346 <li>The GEP instruction never accesses memory, it only provides pointer
348 <li>The first operand to the GEP instruction is always a pointer and it must
350 <li>There are no superfluous indices for the GEP instruction.
</li>
351 <li>Trailing zero indices are superfluous for pointer aliasing, but not for
352 the types of the pointers.
</li>
353 <li>Leading zero indices are not superfluous for pointer aliasing nor the
354 types of the pointers.
</li>
358 <!-- *********************************************************************** -->
362 <a href=
"http://jigsaw.w3.org/css-validator/check/referer"><img
363 src=
"http://jigsaw.w3.org/css-validator/images/vcss-blue" alt=
"Valid CSS"></a>
364 <a href=
"http://validator.w3.org/check/referer"><img
365 src=
"http://www.w3.org/Icons/valid-html401-blue" alt=
"Valid HTML 4.01"></a>
366 <a href=
"http://llvm.org">The LLVM Compiler Infrastructure
</a><br/>
367 Last modified: $Date$