6502/llvm-6502
1
0
Fork 0
mirror of https://github.com/c64scene-ar/llvm-6502.git synced 2025-01-14 16:33:28 +00:00
Code Issues Projects Releases Wiki Activity

Sphixify the GEP FAQ.

Browse Source 
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@158858 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Bill Wendling 2012-06-20 21:54:22 +00:00
parent d67582e276
commit 3950e9e650
3 changed files with 544 additions and 754 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

753
docs/GetElementPtr.html
View File

@ -1,753 +0,0 @@
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
"http://www.w3.org/TR/html4/strict.dtd">
<html>
<head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8">
<title>The Often Misunderstood GEP Instruction</title>
<link rel="stylesheet" href="_static/llvm.css" type="text/css">
<style type="text/css">
TABLE { text-align: left; border: 1px solid black; border-collapse: collapse; margin: 0 0 0 0; }
</style>
</head>
<body>
<h1>
The Often Misunderstood GEP Instruction
</h1>
<ol>
<li><a href="#intro">Introduction</a></li>
<li><a href="#addresses">Address Computation</a>
<ol>
<li><a href="#extra_index">Why is the extra 0 index required?</a></li>
<li><a href="#deref">What is dereferenced by GEP?</a></li>
<li><a href="#firstptr">Why can you index through the first pointer but not
subsequent ones?</a></li>
<li><a href="#lead0">Why don't GEP x,0,0,1 and GEP x,1 alias? </a></li>
<li><a href="#trail0">Why do GEP x,1,0,0 and GEP x,1 alias? </a></li>
<li><a href="#vectors">Can GEP index into vector elements?</a>
<li><a href="#addrspace">What effect do address spaces have on GEPs?</a>
<li><a href="#int">How is GEP different from ptrtoint, arithmetic, and inttoptr?</a></li>
<li><a href="#be">I'm writing a backend for a target which needs custom lowering for GEP. How do I do this?</a>
<li><a href="#vla">How does VLA addressing work with GEPs?</a>
</ol></li>
<li><a href="#rules">Rules</a>
<ol>
<li><a href="#bounds">What happens if an array index is out of bounds?</a>
<li><a href="#negative">Can array indices be negative?</a>
<li><a href="#compare">Can I compare two values computed with GEPs?</a>
<li><a href="#types">Can I do GEP with a different pointer type than the type of the underlying object?</a>
<li><a href="#null">Can I cast an object's address to integer and add it to null?</a>
<li><a href="#ptrdiff">Can I compute the distance between two objects, and add that value to one address to compute the other address?</a>
<li><a href="#tbaa">Can I do type-based alias analysis on LLVM IR?</a>
<li><a href="#overflow">What happens if a GEP computation overflows?</a>
<li><a href="#check">How can I tell if my front-end is following the rules?</a>
</ol></li>
<li><a href="#rationale">Rationale</a>
<ol>
<li><a href="#goals">Why is GEP designed this way?</a></li>
<li><a href="#i32">Why do struct member indices always use i32?</a></li>
<li><a href="#uglygep">What's an uglygep?</a>
</ol></li>
<li><a href="#summary">Summary</a></li>
</ol>
<div class="doc_author">
<p>Written by: <a href="mailto:rspencer@reidspencer.com">Reid Spencer</a>.</p>
</div>
<!-- *********************************************************************** -->
<h2><a name="intro">Introduction</a></h2>
<!-- *********************************************************************** -->
<div>
<p>This document seeks to dispel the mystery and confusion surrounding LLVM's
<a href="LangRef.html#i_getelementptr">GetElementPtr</a> (GEP) instruction.
Questions about the wily GEP instruction are
probably the most frequently occurring questions once a developer gets down to
coding with LLVM. Here we lay out the sources of confusion and show that the
GEP instruction is really quite simple.
</p>
</div>
<!-- *********************************************************************** -->
<h2><a name="addresses">Address Computation</a></h2>
<!-- *********************************************************************** -->
<div>
<p>When people are first confronted with the GEP instruction, they tend to
relate it to known concepts from other programming paradigms, most notably C
array indexing and field selection. GEP closely resembles C array indexing
and field selection, however it's is a little different and this leads to
the following questions.</p>
<!-- *********************************************************************** -->
<h3>
<a name="firstptr">What is the first index of the GEP instruction?</a>
</h3>
<div>
<p>Quick answer: The index stepping through the first operand.</p>
<p>The confusion with the first index usually arises from thinking about
the GetElementPtr instruction as if it was a C index operator. They aren't the
same. For example, when we write, in "C":</p>
<div class="doc_code">
<pre>
AType *Foo;
...
X = &amp;Foo-&gt;F;
</pre>
</div>
<p>it is natural to think that there is only one index, the selection of the
field <tt>F</tt>. However, in this example, <tt>Foo</tt> is a pointer. That
pointer must be indexed explicitly in LLVM. C, on the other hand, indices
through it transparently. To arrive at the same address location as the C
code, you would provide the GEP instruction with two index operands. The
first operand indexes through the pointer; the second operand indexes the
field <tt>F</tt> of the structure, just as if you wrote:</p>
<div class="doc_code">
<pre>
X = &amp;Foo[0].F;
</pre>
</div>
<p>Sometimes this question gets rephrased as:</p>
<blockquote><p><i>Why is it okay to index through the first pointer, but
subsequent pointers won't be dereferenced?</i></p></blockquote>
<p>The answer is simply because memory does not have to be accessed to
perform the computation. The first operand to the GEP instruction must be a
value of a pointer type. The value of the pointer is provided directly to
the GEP instruction as an operand without any need for accessing memory. It
must, therefore be indexed and requires an index operand. Consider this
example:</p>
<div class="doc_code">
<pre>
struct munger_struct {
int f1;
int f2;
};
void munge(struct munger_struct *P) {
P[0].f1 = P[1].f1 + P[2].f2;
}
...
munger_struct Array[3];
...
munge(Array);
</pre>
</div>
<p>In this "C" example, the front end compiler (llvm-gcc) will generate three
GEP instructions for the three indices through "P" in the assignment
statement. The function argument <tt>P</tt> will be the first operand of each
of these GEP instructions. The second operand indexes through that pointer.
The third operand will be the field offset into the
<tt>struct munger_struct</tt> type, for either the <tt>f1</tt> or
<tt>f2</tt> field. So, in LLVM assembly the <tt>munge</tt> function looks
like:</p>
<div class="doc_code">
<pre>
void %munge(%struct.munger_struct* %P) {
entry:
%tmp = getelementptr %struct.munger_struct* %P, i32 1, i32 0
%tmp = load i32* %tmp
%tmp6 = getelementptr %struct.munger_struct* %P, i32 2, i32 1
%tmp7 = load i32* %tmp6
%tmp8 = add i32 %tmp7, %tmp
%tmp9 = getelementptr %struct.munger_struct* %P, i32 0, i32 0
store i32 %tmp8, i32* %tmp9
ret void
}
</pre>
</div>
<p>In each case the first operand is the pointer through which the GEP
instruction starts. The same is true whether the first operand is an
argument, allocated memory, or a global variable. </p>
<p>To make this clear, let's consider a more obtuse example:</p>
<div class="doc_code">
<pre>
%MyVar = uninitialized global i32
...
%idx1 = getelementptr i32* %MyVar, i64 0
%idx2 = getelementptr i32* %MyVar, i64 1
%idx3 = getelementptr i32* %MyVar, i64 2
</pre>
</div>
<p>These GEP instructions are simply making address computations from the
base address of <tt>MyVar</tt>. They compute, as follows (using C syntax):
</p>
<div class="doc_code">
<pre>
idx1 = (char*) &amp;MyVar + 0
idx2 = (char*) &amp;MyVar + 4
idx3 = (char*) &amp;MyVar + 8
</pre>
</div>
<p>Since the type <tt>i32</tt> is known to be four bytes long, the indices
0, 1 and 2 translate into memory offsets of 0, 4, and 8, respectively. No
memory is accessed to make these computations because the address of
<tt>%MyVar</tt> is passed directly to the GEP instructions.</p>
<p>The obtuse part of this example is in the cases of <tt>%idx2</tt> and
<tt>%idx3</tt>. They result in the computation of addresses that point to
memory past the end of the <tt>%MyVar</tt> global, which is only one
<tt>i32</tt> long, not three <tt>i32</tt>s long. While this is legal in LLVM,
it is inadvisable because any load or store with the pointer that results
from these GEP instructions would produce undefined results.</p>
</div>
<!-- *********************************************************************** -->
<h3>
<a name="extra_index">Why is the extra 0 index required?</a>
</h3>
<!-- *********************************************************************** -->
<div>
<p>Quick answer: there are no superfluous indices.</p>
<p>This question arises most often when the GEP instruction is applied to a
global variable which is always a pointer type. For example, consider
this:</p>
<div class="doc_code">
<pre>
%MyStruct = uninitialized global { float*, i32 }
...
%idx = getelementptr { float*, i32 }* %MyStruct, i64 0, i32 1
</pre>
</div>
<p>The GEP above yields an <tt>i32*</tt> by indexing the <tt>i32</tt> typed
field of the structure <tt>%MyStruct</tt>. When people first look at it, they
wonder why the <tt>i64 0</tt> index is needed. However, a closer inspection
of how globals and GEPs work reveals the need. Becoming aware of the following
facts will dispel the confusion:</p>
<ol>
<li>The type of <tt>%MyStruct</tt> is <i>not</i> <tt>{ float*, i32 }</tt>
but rather <tt>{ float*, i32 }*</tt>. That is, <tt>%MyStruct</tt> is a
pointer to a structure containing a pointer to a <tt>float</tt> and an
<tt>i32</tt>.</li>
<li>Point #1 is evidenced by noticing the type of the first operand of
the GEP instruction (<tt>%MyStruct</tt>) which is
<tt>{ float*, i32 }*</tt>.</li>
<li>The first index, <tt>i64 0</tt> is required to step over the global
variable <tt>%MyStruct</tt>. Since the first argument to the GEP
instruction must always be a value of pointer type, the first index
steps through that pointer. A value of 0 means 0 elements offset from that
pointer.</li>
<li>The second index, <tt>i32 1</tt> selects the second field of the
structure (the <tt>i32</tt>). </li>
</ol>
</div>
<!-- *********************************************************************** -->
<h3>
<a name="deref">What is dereferenced by GEP?</a>
</h3>
<div>
<p>Quick answer: nothing.</p>
<p>The GetElementPtr instruction dereferences nothing. That is, it doesn't
access memory in any way. That's what the Load and Store instructions are for.
GEP is only involved in the computation of addresses. For example, consider
this:</p>
<div class="doc_code">
<pre>
%MyVar = uninitialized global { [40 x i32 ]* }
...
%idx = getelementptr { [40 x i32]* }* %MyVar, i64 0, i32 0, i64 0, i64 17
</pre>
</div>
<p>In this example, we have a global variable, <tt>%MyVar</tt> that is a
pointer to a structure containing a pointer to an array of 40 ints. The
GEP instruction seems to be accessing the 18th integer of the structure's
array of ints. However, this is actually an illegal GEP instruction. It
won't compile. The reason is that the pointer in the structure <i>must</i>
be dereferenced in order to index into the array of 40 ints. Since the
GEP instruction never accesses memory, it is illegal.</p>
<p>In order to access the 18th integer in the array, you would need to do the
following:</p>
<div class="doc_code">
<pre>
%idx = getelementptr { [40 x i32]* }* %, i64 0, i32 0
%arr = load [40 x i32]** %idx
%idx = getelementptr [40 x i32]* %arr, i64 0, i64 17
</pre>
</div>
<p>In this case, we have to load the pointer in the structure with a load
instruction before we can index into the array. If the example was changed
to:</p>
<div class="doc_code">
<pre>
%MyVar = uninitialized global { [40 x i32 ] }
...
%idx = getelementptr { [40 x i32] }*, i64 0, i32 0, i64 17
</pre>
</div>
<p>then everything works fine. In this case, the structure does not contain a
pointer and the GEP instruction can index through the global variable,
into the first field of the structure and access the 18th <tt>i32</tt> in the
array there.</p>
</div>
<!-- *********************************************************************** -->
<h3>
<a name="lead0">Why don't GEP x,0,0,1 and GEP x,1 alias?</a>
</h3>
<div>
<p>Quick Answer: They compute different address locations.</p>
<p>If you look at the first indices in these GEP
instructions you find that they are different (0 and 1), therefore the address
computation diverges with that index. Consider this example:</p>
<div class="doc_code">
<pre>
%MyVar = global { [10 x i32 ] }
%idx1 = getelementptr { [10 x i32 ] }* %MyVar, i64 0, i32 0, i64 1
%idx2 = getelementptr { [10 x i32 ] }* %MyVar, i64 1
</pre>
</div>
<p>In this example, <tt>idx1</tt> computes the address of the second integer
in the array that is in the structure in <tt>%MyVar</tt>, that is
<tt>MyVar+4</tt>. The type of <tt>idx1</tt> is <tt>i32*</tt>. However,
<tt>idx2</tt> computes the address of <i>the next</i> structure after
<tt>%MyVar</tt>. The type of <tt>idx2</tt> is <tt>{ [10 x i32] }*</tt> and its
value is equivalent to <tt>MyVar + 40</tt> because it indexes past the ten
4-byte integers in <tt>MyVar</tt>. Obviously, in such a situation, the
pointers don't alias.</p>
</div>
<!-- *********************************************************************** -->
<h3>
<a name="trail0">Why do GEP x,1,0,0 and GEP x,1 alias?</a>
</h3>
<div>
<p>Quick Answer: They compute the same address location.</p>
<p>These two GEP instructions will compute the same address because indexing
through the 0th element does not change the address. However, it does change
the type. Consider this example:</p>
<div class="doc_code">
<pre>
%MyVar = global { [10 x i32 ] }
%idx1 = getelementptr { [10 x i32 ] }* %MyVar, i64 1, i32 0, i64 0
%idx2 = getelementptr { [10 x i32 ] }* %MyVar, i64 1
</pre>
</div>
<p>In this example, the value of <tt>%idx1</tt> is <tt>%MyVar+40</tt> and
its type is <tt>i32*</tt>. The value of <tt>%idx2</tt> is also
<tt>MyVar+40</tt> but its type is <tt>{ [10 x i32] }*</tt>.</p>
</div>
<!-- *********************************************************************** -->
<h3>
<a name="vectors">Can GEP index into vector elements?</a>
</h3>
<div>
<p>This hasn't always been forcefully disallowed, though it's not recommended.
It leads to awkward special cases in the optimizers, and fundamental
inconsistency in the IR. In the future, it will probably be outright
disallowed.</p>
</div>
<!-- *********************************************************************** -->
<h3>
<a name="addrspace">What effect do address spaces have on GEPs?</a>
</h3>
<div>
<p>None, except that the address space qualifier on the first operand pointer
type always matches the address space qualifier on the result type.</p>
</div>
<!-- *********************************************************************** -->
<h3>
<a name="int">
How is GEP different from ptrtoint, arithmetic, and inttoptr?
</a>
</h3>
<div>
<p>It's very similar; there are only subtle differences.</p>
<p>With ptrtoint, you have to pick an integer type. One approach is to pick i64;
this is safe on everything LLVM supports (LLVM internally assumes pointers
are never wider than 64 bits in many places), and the optimizer will actually
narrow the i64 arithmetic down to the actual pointer size on targets which
don't support 64-bit arithmetic in most cases. However, there are some cases
where it doesn't do this. With GEP you can avoid this problem.
<p>Also, GEP carries additional pointer aliasing rules. It's invalid to take a
GEP from one object, address into a different separately allocated
object, and dereference it. IR producers (front-ends) must follow this rule,
and consumers (optimizers, specifically alias analysis) benefit from being
able to rely on it. See the <a href="#rules">Rules</a> section for more
information.</p>
<p>And, GEP is more concise in common cases.</p>
<p>However, for the underlying integer computation implied, there
is no difference.</p>
</div>
<!-- *********************************************************************** -->
<h3>
<a name="be">
I'm writing a backend for a target which needs custom lowering for GEP.
How do I do this?
</a>
</h3>
<div>
<p>You don't. The integer computation implied by a GEP is target-independent.
Typically what you'll need to do is make your backend pattern-match
expressions trees involving ADD, MUL, etc., which are what GEP is lowered
into. This has the advantage of letting your code work correctly in more
cases.</p>
<p>GEP does use target-dependent parameters for the size and layout of data
types, which targets can customize.</p>
<p>If you require support for addressing units which are not 8 bits, you'll
need to fix a lot of code in the backend, with GEP lowering being only a
small piece of the overall picture.</p>
</div>
<!-- *********************************************************************** -->
<h3>
<a name="vla">How does VLA addressing work with GEPs?</a>
</h3>
<div>
<p>GEPs don't natively support VLAs. LLVM's type system is entirely static,
and GEP address computations are guided by an LLVM type.</p>
<p>VLA indices can be implemented as linearized indices. For example, an
expression like X[a][b][c], must be effectively lowered into a form
like X[a*m+b*n+c], so that it appears to the GEP as a single-dimensional
array reference.</p>
<p>This means if you want to write an analysis which understands array
indices and you want to support VLAs, your code will have to be
prepared to reverse-engineer the linearization. One way to solve this
problem is to use the ScalarEvolution library, which always presents
VLA and non-VLA indexing in the same manner.</p>
</div>
</div>
<!-- *********************************************************************** -->
<h2><a name="rules">Rules</a></h2>
<!-- *********************************************************************** -->
<div>
<!-- *********************************************************************** -->
<h3>
<a name="bounds">What happens if an array index is out of bounds?</a>
</h3>
<div>
<p>There are two senses in which an array index can be out of bounds.</p>
<p>First, there's the array type which comes from the (static) type of
the first operand to the GEP. Indices greater than the number of elements
in the corresponding static array type are valid. There is no problem with
out of bounds indices in this sense. Indexing into an array only depends
on the size of the array element, not the number of elements.</p>
<p>A common example of how this is used is arrays where the size is not known.
It's common to use array types with zero length to represent these. The
fact that the static type says there are zero elements is irrelevant; it's
perfectly valid to compute arbitrary element indices, as the computation
only depends on the size of the array element, not the number of
elements. Note that zero-sized arrays are not a special case here.</p>
<p>This sense is unconnected with <tt>inbounds</tt> keyword. The
<tt>inbounds</tt> keyword is designed to describe low-level pointer
arithmetic overflow conditions, rather than high-level array
indexing rules.
<p>Analysis passes which wish to understand array indexing should not
assume that the static array type bounds are respected.</p>
<p>The second sense of being out of bounds is computing an address that's
beyond the actual underlying allocated object.</p>
<p>With the <tt>inbounds</tt> keyword, the result value of the GEP is
undefined if the address is outside the actual underlying allocated
object and not the address one-past-the-end.</p>
<p>Without the <tt>inbounds</tt> keyword, there are no restrictions
on computing out-of-bounds addresses. Obviously, performing a load or
a store requires an address of allocated and sufficiently aligned
memory. But the GEP itself is only concerned with computing addresses.</p>
</div>
<!-- *********************************************************************** -->
<h3>
<a name="negative">Can array indices be negative?</a>
</h3>
<div>
<p>Yes. This is basically a special case of array indices being out
of bounds.</p>
</div>
<!-- *********************************************************************** -->
<h3>
<a name="compare">Can I compare two values computed with GEPs?</a>
</h3>
<div>
<p>Yes. If both addresses are within the same allocated object, or
one-past-the-end, you'll get the comparison result you expect. If either
is outside of it, integer arithmetic wrapping may occur, so the
comparison may not be meaningful.</p>
</div>
<!-- *********************************************************************** -->
<h3>
<a name="types">
Can I do GEP with a different pointer type than the type of
the underlying object?
</a>
</h3>
<div>
<p>Yes. There are no restrictions on bitcasting a pointer value to an arbitrary
pointer type. The types in a GEP serve only to define the parameters for the
underlying integer computation. They need not correspond with the actual
type of the underlying object.</p>
<p>Furthermore, loads and stores don't have to use the same types as the type
of the underlying object. Types in this context serve only to specify
memory size and alignment. Beyond that there are merely a hint to the
optimizer indicating how the value will likely be used.</p>
</div>
<!-- *********************************************************************** -->
<h3>
<a name="null">
Can I cast an object's address to integer and add it to null?
</a>
</h3>
<div>
<p>You can compute an address that way, but if you use GEP to do the add,
you can't use that pointer to actually access the object, unless the
object is managed outside of LLVM.</p>
<p>The underlying integer computation is sufficiently defined; null has a
defined value -- zero -- and you can add whatever value you want to it.</p>
<p>However, it's invalid to access (load from or store to) an LLVM-aware
object with such a pointer. This includes GlobalVariables, Allocas, and
objects pointed to by noalias pointers.</p>
<p>If you really need this functionality, you can do the arithmetic with
explicit integer instructions, and use inttoptr to convert the result to
an address. Most of GEP's special aliasing rules do not apply to pointers
computed from ptrtoint, arithmetic, and inttoptr sequences.</p>
</div>
<!-- *********************************************************************** -->
<h3>
<a name="ptrdiff">
Can I compute the distance between two objects, and add
that value to one address to compute the other address?
</a>
</h3>
<div>
<p>As with arithmetic on null, You can use GEP to compute an address that
way, but you can't use that pointer to actually access the object if you
do, unless the object is managed outside of LLVM.</p>
<p>Also as above, ptrtoint and inttoptr provide an alternative way to do this
which do not have this restriction.</p>
</div>
<!-- *********************************************************************** -->
<h3>
<a name="tbaa">Can I do type-based alias analysis on LLVM IR?</a>
</h3>
<div>
<p>You can't do type-based alias analysis using LLVM's built-in type system,
because LLVM has no restrictions on mixing types in addressing, loads or
stores.</p>
<p>LLVM's type-based alias analysis pass uses metadata to describe a different
type system (such as the C type system), and performs type-based aliasing
on top of that. Further details are in the
<a href="LangRef.html#tbaa">language reference</a>.</p>
</div>
<!-- *********************************************************************** -->
<h3>
<a name="overflow">What happens if a GEP computation overflows?</a>
</h3>
<div>
<p>If the GEP lacks the <tt>inbounds</tt> keyword, the value is the result
from evaluating the implied two's complement integer computation. However,
since there's no guarantee of where an object will be allocated in the
address space, such values have limited meaning.</p>
<p>If the GEP has the <tt>inbounds</tt> keyword, the result value is
undefined (a "<a href="LangRef.html#trapvalues">trap value</a>") if the GEP
overflows (i.e. wraps around the end of the address space).</p>
<p>As such, there are some ramifications of this for inbounds GEPs: scales
implied by array/vector/pointer indices are always known to be "nsw" since
they are signed values that are scaled by the element size. These values
are also allowed to be negative (e.g. "gep i32 *%P, i32 -1") but the
pointer itself is logically treated as an unsigned value. This means that
GEPs have an asymmetric relation between the pointer base (which is treated
as unsigned) and the offset applied to it (which is treated as signed). The
result of the additions within the offset calculation cannot have signed
overflow, but when applied to the base pointer, there can be signed
overflow.
</p>
</div>
<!-- *********************************************************************** -->
<h3>
<a name="check">
How can I tell if my front-end is following the rules?
</a>
</h3>
<div>
<p>There is currently no checker for the getelementptr rules. Currently,
the only way to do this is to manually check each place in your front-end
where GetElementPtr operators are created.</p>
<p>It's not possible to write a checker which could find all rule
violations statically. It would be possible to write a checker which
works by instrumenting the code with dynamic checks though. Alternatively,
it would be possible to write a static checker which catches a subset of
possible problems. However, no such checker exists today.</p>
</div>
</div>
<!-- *********************************************************************** -->
<h2><a name="rationale">Rationale</a></h2>
<!-- *********************************************************************** -->
<div>
<!-- *********************************************************************** -->
<h3>
<a name="goals">Why is GEP designed this way?</a>
</h3>
<div>
<p>The design of GEP has the following goals, in rough unofficial
order of priority:</p>
<ul>
<li>Support C, C-like languages, and languages which can be
conceptually lowered into C (this covers a lot).</li>
<li>Support optimizations such as those that are common in
C compilers. In particular, GEP is a cornerstone of LLVM's
<a href="LangRef.html#pointeraliasing">pointer aliasing model</a>.</li>
<li>Provide a consistent method for computing addresses so that
address computations don't need to be a part of load and
store instructions in the IR.</li>
<li>Support non-C-like languages, to the extent that it doesn't
interfere with other goals.</li>
<li>Minimize target-specific information in the IR.</li>
</ul>
</div>
<!-- *********************************************************************** -->
<h3>
<a name="i32">Why do struct member indices always use i32?</a>
</h3>
<div>
<p>The specific type i32 is probably just a historical artifact, however it's
wide enough for all practical purposes, so there's been no need to change it.
It doesn't necessarily imply i32 address arithmetic; it's just an identifier
which identifies a field in a struct. Requiring that all struct indices be
the same reduces the range of possibilities for cases where two GEPs are
effectively the same but have distinct operand types.</p>
</div>
<!-- *********************************************************************** -->
<h3>
<a name="uglygep">What's an uglygep?</a>
</h3>
<div>
<p>Some LLVM optimizers operate on GEPs by internally lowering them into
more primitive integer expressions, which allows them to be combined
with other integer expressions and/or split into multiple separate
integer expressions. If they've made non-trivial changes, translating
back into LLVM IR can involve reverse-engineering the structure of
the addressing in order to fit it into the static type of the original
first operand. It isn't always possibly to fully reconstruct this
structure; sometimes the underlying addressing doesn't correspond with
the static type at all. In such cases the optimizer instead will emit
a GEP with the base pointer casted to a simple address-unit pointer,
using the name "uglygep". This isn't pretty, but it's just as
valid, and it's sufficient to preserve the pointer aliasing guarantees
that GEP provides.</p>
</div>
</div>
<!-- *********************************************************************** -->
<h2><a name="summary">Summary</a></h2>
<!-- *********************************************************************** -->
<div>
<p>In summary, here's some things to always remember about the GetElementPtr
instruction:</p>
<ol>
<li>The GEP instruction never accesses memory, it only provides pointer
computations.</li>
<li>The first operand to the GEP instruction is always a pointer and it must
be indexed.</li>
<li>There are no superfluous indices for the GEP instruction.</li>
<li>Trailing zero indices are superfluous for pointer aliasing, but not for
the types of the pointers.</li>
<li>Leading zero indices are not superfluous for pointer aliasing nor the
types of the pointers.</li>
</ol>
</div>
<!-- *********************************************************************** -->
<hr>
<address>
<a href="http://jigsaw.w3.org/css-validator/check/referer"><img
src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
<a href="http://validator.w3.org/check/referer"><img
src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
<a href="http://llvm.org/">The LLVM Compiler Infrastructure</a><br>
Last modified: $Date$
</address>
</body>
</html>

538
docs/GetElementPtr.rst Normal file
View File

@ -0,0 +1,538 @@
.. _gep:
=======================================
The Often Misunderstood GEP Instruction
=======================================
.. contents::
:local:
Introduction
============
This document seeks to dispel the mystery and confusion surrounding LLVM's
`GetElementPtr <LangRef.html#i_getelementptr>`_ (GEP) instruction. Questions
about the wily GEP instruction are probably the most frequently occurring
questions once a developer gets down to coding with LLVM. Here we lay out the
sources of confusion and show that the GEP instruction is really quite simple.
Address Computation
===================
When people are first confronted with the GEP instruction, they tend to relate
it to known concepts from other programming paradigms, most notably C array
indexing and field selection. GEP closely resembles C array indexing and field
selection, however it's is a little different and this leads to the following
questions.
What is the first index of the GEP instruction?
-----------------------------------------------
Quick answer: The index stepping through the first operand.
The confusion with the first index usually arises from thinking about the
GetElementPtr instruction as if it was a C index operator. They aren't the
same. For example, when we write, in "C":
.. code-block:: c++
AType *Foo;
...
X = &Foo->F;
it is natural to think that there is only one index, the selection of the field
``F``. However, in this example, ``Foo`` is a pointer. That pointer
must be indexed explicitly in LLVM. C, on the other hand, indices through it
transparently. To arrive at the same address location as the C code, you would
provide the GEP instruction with two index operands. The first operand indexes
through the pointer; the second operand indexes the field ``F`` of the
structure, just as if you wrote:
.. code-block:: c++
X = &Foo[0].F;
Sometimes this question gets rephrased as:
.. _GEP index through first pointer:
*Why is it okay to index through the first pointer, but subsequent pointers
won't be dereferenced?*
The answer is simply because memory does not have to be accessed to perform the
computation. The first operand to the GEP instruction must be a value of a
pointer type. The value of the pointer is provided directly to the GEP
instruction as an operand without any need for accessing memory. It must,
therefore be indexed and requires an index operand. Consider this example:
.. code-block:: c++
struct munger_struct {
int f1;
int f2;
};
void munge(struct munger_struct *P) {
P[0].f1 = P[1].f1 + P[2].f2;
}
...
munger_struct Array[3];
...
munge(Array);
In this "C" example, the front end compiler (llvm-gcc) will generate three GEP
instructions for the three indices through "P" in the assignment statement. The
function argument ``P`` will be the first operand of each of these GEP
instructions. The second operand indexes through that pointer. The third
operand will be the field offset into the ``struct munger_struct`` type, for
either the ``f1`` or ``f2`` field. So, in LLVM assembly the ``munge`` function
looks like:
.. code-block:: llvm
void %munge(%struct.munger_struct* %P) {
entry:
%tmp = getelementptr %struct.munger_struct* %P, i32 1, i32 0
%tmp = load i32* %tmp
%tmp6 = getelementptr %struct.munger_struct* %P, i32 2, i32 1
%tmp7 = load i32* %tmp6
%tmp8 = add i32 %tmp7, %tmp
%tmp9 = getelementptr %struct.munger_struct* %P, i32 0, i32 0
store i32 %tmp8, i32* %tmp9
ret void
}
In each case the first operand is the pointer through which the GEP instruction
starts. The same is true whether the first operand is an argument, allocated
memory, or a global variable.
To make this clear, let's consider a more obtuse example:
.. code-block:: llvm
%MyVar = uninitialized global i32
...
%idx1 = getelementptr i32* %MyVar, i64 0
%idx2 = getelementptr i32* %MyVar, i64 1
%idx3 = getelementptr i32* %MyVar, i64 2
These GEP instructions are simply making address computations from the base
address of ``MyVar``. They compute, as follows (using C syntax):
.. code-block:: c++
idx1 = (char*) &MyVar + 0
idx2 = (char*) &MyVar + 4
idx3 = (char*) &MyVar + 8
Since the type ``i32`` is known to be four bytes long, the indices 0, 1 and 2
translate into memory offsets of 0, 4, and 8, respectively. No memory is
accessed to make these computations because the address of ``%MyVar`` is passed
directly to the GEP instructions.
The obtuse part of this example is in the cases of ``%idx2`` and ``%idx3``. They
result in the computation of addresses that point to memory past the end of the
``%MyVar`` global, which is only one ``i32`` long, not three ``i32``\s long.
While this is legal in LLVM, it is inadvisable because any load or store with
the pointer that results from these GEP instructions would produce undefined
results.
Why is the extra 0 index required?
----------------------------------
Quick answer: there are no superfluous indices.
This question arises most often when the GEP instruction is applied to a global
variable which is always a pointer type. For example, consider this:
.. code-block:: llvm
%MyStruct = uninitialized global { float*, i32 }
...
%idx = getelementptr { float*, i32 }* %MyStruct, i64 0, i32 1
The GEP above yields an ``i32*`` by indexing the ``i32`` typed field of the
structure ``%MyStruct``. When people first look at it, they wonder why the ``i64
0`` index is needed. However, a closer inspection of how globals and GEPs work
reveals the need. Becoming aware of the following facts will dispel the
confusion:
#. The type of ``%MyStruct`` is *not* ``{ float*, i32 }`` but rather ``{ float*,
i32 }*``. That is, ``%MyStruct`` is a pointer to a structure containing a
pointer to a ``float`` and an ``i32``.
#. Point #1 is evidenced by noticing the type of the first operand of the GEP
instruction (``%MyStruct``) which is ``{ float*, i32 }*``.
#. The first index, ``i64 0`` is required to step over the global variable
``%MyStruct``. Since the first argument to the GEP instruction must always
be a value of pointer type, the first index steps through that pointer. A
value of 0 means 0 elements offset from that pointer.
#. The second index, ``i32 1`` selects the second field of the structure (the
``i32``).
What is dereferenced by GEP?
----------------------------
Quick answer: nothing.
The GetElementPtr instruction dereferences nothing. That is, it doesn't access
memory in any way. That's what the Load and Store instructions are for. GEP is
only involved in the computation of addresses. For example, consider this:
.. code-block:: llvm
%MyVar = uninitialized global { [40 x i32 ]* }
...
%idx = getelementptr { [40 x i32]* }* %MyVar, i64 0, i32 0, i64 0, i64 17
In this example, we have a global variable, ``%MyVar`` that is a pointer to a
structure containing a pointer to an array of 40 ints. The GEP instruction seems
to be accessing the 18th integer of the structure's array of ints. However, this
is actually an illegal GEP instruction. It won't compile. The reason is that the
pointer in the structure <i>must</i> be dereferenced in order to index into the
array of 40 ints. Since the GEP instruction never accesses memory, it is
illegal.
In order to access the 18th integer in the array, you would need to do the
following:
.. code-block:: llvm
%idx = getelementptr { [40 x i32]* }* %, i64 0, i32 0
%arr = load [40 x i32]** %idx
%idx = getelementptr [40 x i32]* %arr, i64 0, i64 17
In this case, we have to load the pointer in the structure with a load
instruction before we can index into the array. If the example was changed to:
.. code-block:: llvm
%MyVar = uninitialized global { [40 x i32 ] }
...
%idx = getelementptr { [40 x i32] }*, i64 0, i32 0, i64 17
then everything works fine. In this case, the structure does not contain a
pointer and the GEP instruction can index through the global variable, into the
first field of the structure and access the 18th ``i32`` in the array there.
Why don't GEP x,0,0,1 and GEP x,1 alias?
----------------------------------------
Quick Answer: They compute different address locations.
If you look at the first indices in these GEP instructions you find that they
are different (0 and 1), therefore the address computation diverges with that
index. Consider this example:
.. code-block:: llvm
%MyVar = global { [10 x i32 ] }
%idx1 = getelementptr { [10 x i32 ] }* %MyVar, i64 0, i32 0, i64 1
%idx2 = getelementptr { [10 x i32 ] }* %MyVar, i64 1
In this example, ``idx1`` computes the address of the second integer in the
array that is in the structure in ``%MyVar``, that is ``MyVar+4``. The type of
``idx1`` is ``i32*``. However, ``idx2`` computes the address of *the next*
structure after ``%MyVar``. The type of ``idx2`` is ``{ [10 x i32] }*`` and its
value is equivalent to ``MyVar + 40`` because it indexes past the ten 4-byte
integers in ``MyVar``. Obviously, in such a situation, the pointers don't
alias.
Why do GEP x,1,0,0 and GEP x,1 alias?
-------------------------------------
Quick Answer: They compute the same address location.
These two GEP instructions will compute the same address because indexing
through the 0th element does not change the address. However, it does change the
type. Consider this example:
.. code-block:: llvm
%MyVar = global { [10 x i32 ] }
%idx1 = getelementptr { [10 x i32 ] }* %MyVar, i64 1, i32 0, i64 0
%idx2 = getelementptr { [10 x i32 ] }* %MyVar, i64 1
In this example, the value of ``%idx1`` is ``%MyVar+40`` and its type is
``i32*``. The value of ``%idx2`` is also ``MyVar+40`` but its type is ``{ [10 x
i32] }*``.
Can GEP index into vector elements?
-----------------------------------
This hasn't always been forcefully disallowed, though it's not recommended. It
leads to awkward special cases in the optimizers, and fundamental inconsistency
in the IR. In the future, it will probably be outright disallowed.
What effect do address spaces have on GEPs?
-------------------------------------------
None, except that the address space qualifier on the first operand pointer type
always matches the address space qualifier on the result type.
How is GEP different from ``ptrtoint``, arithmetic, and ``inttoptr``?
---------------------------------------------------------------------
It's very similar; there are only subtle differences.
With ptrtoint, you have to pick an integer type. One approach is to pick i64;
this is safe on everything LLVM supports (LLVM internally assumes pointers are
never wider than 64 bits in many places), and the optimizer will actually narrow
the i64 arithmetic down to the actual pointer size on targets which don't
support 64-bit arithmetic in most cases. However, there are some cases where it
doesn't do this. With GEP you can avoid this problem.
Also, GEP carries additional pointer aliasing rules. It's invalid to take a GEP
from one object, address into a different separately allocated object, and
dereference it. IR producers (front-ends) must follow this rule, and consumers
(optimizers, specifically alias analysis) benefit from being able to rely on
it. See the `Rules`_ section for more information.
And, GEP is more concise in common cases.
However, for the underlying integer computation implied, there is no
difference.
I'm writing a backend for a target which needs custom lowering for GEP. How do I do this?
-----------------------------------------------------------------------------------------
You don't. The integer computation implied by a GEP is target-independent.
Typically what you'll need to do is make your backend pattern-match expressions
trees involving ADD, MUL, etc., which are what GEP is lowered into. This has the
advantage of letting your code work correctly in more cases.
GEP does use target-dependent parameters for the size and layout of data types,
which targets can customize.
If you require support for addressing units which are not 8 bits, you'll need to
fix a lot of code in the backend, with GEP lowering being only a small piece of
the overall picture.
How does VLA addressing work with GEPs?
---------------------------------------
GEPs don't natively support VLAs. LLVM's type system is entirely static, and GEP
address computations are guided by an LLVM type.
VLA indices can be implemented as linearized indices. For example, an expression
like ``X[a][b][c]``, must be effectively lowered into a form like
``X[a*m+b*n+c]``, so that it appears to the GEP as a single-dimensional array
reference.
This means if you want to write an analysis which understands array indices and
you want to support VLAs, your code will have to be prepared to reverse-engineer
the linearization. One way to solve this problem is to use the ScalarEvolution
library, which always presents VLA and non-VLA indexing in the same manner.
.. _Rules:
Rules
=====
What happens if an array index is out of bounds?
------------------------------------------------
There are two senses in which an array index can be out of bounds.
First, there's the array type which comes from the (static) type of the first
operand to the GEP. Indices greater than the number of elements in the
corresponding static array type are valid. There is no problem with out of
bounds indices in this sense. Indexing into an array only depends on the size of
the array element, not the number of elements.
A common example of how this is used is arrays where the size is not known.
It's common to use array types with zero length to represent these. The fact
that the static type says there are zero elements is irrelevant; it's perfectly
valid to compute arbitrary element indices, as the computation only depends on
the size of the array element, not the number of elements. Note that zero-sized
arrays are not a special case here.
This sense is unconnected with ``inbounds`` keyword. The ``inbounds`` keyword is
designed to describe low-level pointer arithmetic overflow conditions, rather
than high-level array indexing rules.
Analysis passes which wish to understand array indexing should not assume that
the static array type bounds are respected.
The second sense of being out of bounds is computing an address that's beyond
the actual underlying allocated object.
With the ``inbounds`` keyword, the result value of the GEP is undefined if the
address is outside the actual underlying allocated object and not the address
one-past-the-end.
Without the ``inbounds`` keyword, there are no restrictions on computing
out-of-bounds addresses. Obviously, performing a load or a store requires an
address of allocated and sufficiently aligned memory. But the GEP itself is only
concerned with computing addresses.
Can array indices be negative?
------------------------------
Yes. This is basically a special case of array indices being out of bounds.
Can I compare two values computed with GEPs?
--------------------------------------------
Yes. If both addresses are within the same allocated object, or
one-past-the-end, you'll get the comparison result you expect. If either is
outside of it, integer arithmetic wrapping may occur, so the comparison may not
be meaningful.
Can I do GEP with a different pointer type than the type of the underlying object?
----------------------------------------------------------------------------------
Yes. There are no restrictions on bitcasting a pointer value to an arbitrary
pointer type. The types in a GEP serve only to define the parameters for the
underlying integer computation. They need not correspond with the actual type of
the underlying object.
Furthermore, loads and stores don't have to use the same types as the type of
the underlying object. Types in this context serve only to specify memory size
and alignment. Beyond that there are merely a hint to the optimizer indicating
how the value will likely be used.
Can I cast an object's address to integer and add it to null?
-------------------------------------------------------------
You can compute an address that way, but if you use GEP to do the add, you can't
use that pointer to actually access the object, unless the object is managed
outside of LLVM.
The underlying integer computation is sufficiently defined; null has a defined
value --- zero --- and you can add whatever value you want to it.
However, it's invalid to access (load from or store to) an LLVM-aware object
with such a pointer. This includes ``GlobalVariables``, ``Allocas``, and objects
pointed to by noalias pointers.
If you really need this functionality, you can do the arithmetic with explicit
integer instructions, and use inttoptr to convert the result to an address. Most
of GEP's special aliasing rules do not apply to pointers computed from ptrtoint,
arithmetic, and inttoptr sequences.
Can I compute the distance between two objects, and add that value to one address to compute the other address?
---------------------------------------------------------------------------------------------------------------
As with arithmetic on null, You can use GEP to compute an address that way, but
you can't use that pointer to actually access the object if you do, unless the
object is managed outside of LLVM.
Also as above, ptrtoint and inttoptr provide an alternative way to do this which
do not have this restriction.
Can I do type-based alias analysis on LLVM IR?
----------------------------------------------
You can't do type-based alias analysis using LLVM's built-in type system,
because LLVM has no restrictions on mixing types in addressing, loads or stores.
LLVM's type-based alias analysis pass uses metadata to describe a different type
system (such as the C type system), and performs type-based aliasing on top of
that. Further details are in the `language reference <LangRef.html#tbaa>`_.
What happens if a GEP computation overflows?
--------------------------------------------
If the GEP lacks the ``inbounds`` keyword, the value is the result from
evaluating the implied two's complement integer computation. However, since
there's no guarantee of where an object will be allocated in the address space,
such values have limited meaning.
If the GEP has the ``inbounds`` keyword, the result value is undefined (a "trap
value") if the GEP overflows (i.e. wraps around the end of the address space).
As such, there are some ramifications of this for inbounds GEPs: scales implied
by array/vector/pointer indices are always known to be "nsw" since they are
signed values that are scaled by the element size. These values are also
allowed to be negative (e.g. "``gep i32 *%P, i32 -1``") but the pointer itself
is logically treated as an unsigned value. This means that GEPs have an
asymmetric relation between the pointer base (which is treated as unsigned) and
the offset applied to it (which is treated as signed). The result of the
additions within the offset calculation cannot have signed overflow, but when
applied to the base pointer, there can be signed overflow.
How can I tell if my front-end is following the rules?
------------------------------------------------------
There is currently no checker for the getelementptr rules. Currently, the only
way to do this is to manually check each place in your front-end where
GetElementPtr operators are created.
It's not possible to write a checker which could find all rule violations
statically. It would be possible to write a checker which works by instrumenting
the code with dynamic checks though. Alternatively, it would be possible to
write a static checker which catches a subset of possible problems. However, no
such checker exists today.
Rationale
=========
Why is GEP designed this way?
-----------------------------
The design of GEP has the following goals, in rough unofficial order of
priority:
* Support C, C-like languages, and languages which can be conceptually lowered
into C (this covers a lot).
* Support optimizations such as those that are common in C compilers. In
particular, GEP is a cornerstone of LLVM's `pointer aliasing
model <LangRef.html#pointeraliasing>`_.
* Provide a consistent method for computing addresses so that address
computations don't need to be a part of load and store instructions in the IR.
* Support non-C-like languages, to the extent that it doesn't interfere with
other goals.
* Minimize target-specific information in the IR.
Why do struct member indices always use ``i32``?
------------------------------------------------
The specific type i32 is probably just a historical artifact, however it's wide
enough for all practical purposes, so there's been no need to change it. It
doesn't necessarily imply i32 address arithmetic; it's just an identifier which
identifies a field in a struct. Requiring that all struct indices be the same
reduces the range of possibilities for cases where two GEPs are effectively the
same but have distinct operand types.
What's an uglygep?
------------------
Some LLVM optimizers operate on GEPs by internally lowering them into more
primitive integer expressions, which allows them to be combined with other
integer expressions and/or split into multiple separate integer expressions. If
they've made non-trivial changes, translating back into LLVM IR can involve
reverse-engineering the structure of the addressing in order to fit it into the
static type of the original first operand. It isn't always possibly to fully
reconstruct this structure; sometimes the underlying addressing doesn't
correspond with the static type at all. In such cases the optimizer instead will
emit a GEP with the base pointer casted to a simple address-unit pointer, using
the name "uglygep". This isn't pretty, but it's just as valid, and it's
sufficient to preserve the pointer aliasing guarantees that GEP provides.
Summary
=======
In summary, here's some things to always remember about the GetElementPtr
instruction:
#. The GEP instruction never accesses memory, it only provides pointer
computations.
#. The first operand to the GEP instruction is always a pointer and it must be
indexed.
#. There are no superfluous indices for the GEP instruction.
#. Trailing zero indices are superfluous for pointer aliasing, but not for the
types of the pointers.
#. Leading zero indices are not superfluous for pointer aliasing nor the types
of the pointers.

7
docs/design_and_overview.rst
View File

@ -3,6 +3,11 @@
LLVM Design & Overview
======================
.. toctree::
:hidden:
GetElementPtr
* `LLVM Language Reference Manual <LangRef.html>`_
Defines the LLVM intermediate representation.
@ -25,7 +30,7 @@ LLVM Design & Overview
More details (quite old now).
* `GetElementPtr FAQ <GetElementPtr.html>`_
* :ref:`gep`
Answers to some very frequent questions about LLVM's most frequently
misunderstood instruction.