Undef and Poison round table follow-up & a plan

Project Infrastructure LLVM Dev List Archives
1

Hello all,

Thank everyone who participated in the (impromptu) round table discussion on Tuesday.
For those who are interested, I share the summary of the discussion.
Also, I share a short-term plan regarding this issue and relevant patches.

Fixing Miscompilations using Freeze

2

Hello all,

Thank everyone who participated in the (impromptu) round table discussion on Tuesday.
For those who are interested, I share the summary of the discussion.
Also, I share a short-term plan regarding this issue and relevant patches.

Fixing Miscompilations using Freeze

To reduce the cost of fixing miscompilations using freeze instruction, we need to
optimize freeze away whenever possible.
Using the no-undef/poison assumption from the source language (C/C++ in
this context) can play a significant role.
To make use the assumptions, here are short-term goals:

1. Preserve no-undef/poison assumption of function arguments from C/C++ when valid.

There is an ongoing relevant patch (that is written by others):
https://reviews.llvm.org/D81678

2. Preserve no-undef/poison assumption of lvalue reads in C/C++ when valid.

Reading an indeterminate value from an lvalue that does not have char or
std::byte type is UB [1].
Since reading an lvalue is lowered to load in IR, we suggest attaching a new
!noundef metadata to such loads.
The IR-side change is here: https://reviews.llvm.org/D89050
The clang-side change is going to be made after D81678 is reviewed, because it is likely
that this patch will have a lot of changes in clang tests.

Replacing Undef with Poison

Since undef is known to be the source of many optimizations due to its complexity,
we’d like to suggest gradually moving towards using poison only.
To make it, (1) poison constant should be introduced into LLVM IR first, and (2)
transformations that introduce undef should be updated to introduce poison instead.

For the step (2), we need an experimental result showing that it does not cause
performance degradation. This relies on better support for freeze (the
no-undef/poison analysis patches).

1. Introduce a new poison constant into IR: https://reviews.llvm.org/D71126

Note that poison constant can be used as a true placeholder value as well.
Undef cannot be used in general because it is less undefined than poison.

2. Update transformations that introduce undef to introduce poison instead

(1) There are transformations that introduce undef as a placeholder (e.g. phi operand
from an unreachable block).
For these, poison can be used instead.

(2) The value of an uninitialized object (automatic or dynamic).

They are indeterminate values in C/C++, so okay to use poison instead.
A tricky case is a bitfield access, and we have two possible solutions:

  • i. Introduce a very-packed struct type
<C>
struct {
int a:2, b:6;
} s;

v = s.a;

=>

<IR>

s = alloca

tmp = load **{{i2, i6}}*** s ; load as a very packed struct type
v = extractvalue tmp, 0
  • Pros: Can be used to precisely lower C/C++'s struct typed function argument into IR
    (currently clang coerces a struct into int if small enough; I’ll explain about this detail if anyone requests)
  • Cons: Since optimizations aren’t aware of the new type, they should be updated
  • ii. Use load-freeze
<C>
struct {
int a:2, b:6;
} s;

v = s.a;

=>

<IR>
s = alloca

// Poison bits are frozen and returned
tmp = **load freeze** i8* s
v = tmp & 3
  • Pros: The change is simpler
  • Cons: Store forwarding isn’t free; needs insertion of freeze
    (store x, p; v = load freeze p => store x, p; v = freeze x)

(3) The third case is the value of struct/union padding.
Padding is filled with unspecified value in C, so it is too undefined to use poison.
We can fill it with defined bits nondeterministically chosen at allocation time (freeze poison).

<C>
struct {
char a; // 3 bytes padding
int b;
} s;

v = s.b;

=>

<IR>
s = alloca {i8, i32} // alloca initializes bytes in a type-dependent manner
// s[0], s[4~7]: poison
// s[1~3]: let's fill these bytes with nondet. bits

s2 = gep (bitcast s to i8*), 4
v = load i32 s2

Thanks,
Juneyoung

[1]
C11 6.2.6.1.5: If the stored value of an object has such a representation and is read by an lvalue expression that does not have character type, the behavior is undefined.
(Similarly, C17 6.2.6.1.5)

It is important to note that this applies to trap representations and not to unspecified values. A structure or union never has a trap representation.

C++14 8.5.12: If an indeterminate value is produced by an evaluation, the behavior is undefined except in the following cases: If an indeterminate value of unsigned narrow character type …
(Similarly, C++17 11.6.12 , C++11 4.1.1)

While loading undef for the unsigned character type case merely produces undef, for C++, operations such as sign-extend or zero-extend on an undef i8 is also undefined behaviour.

3

It is important to note that this applies to trap representations and not to unspecified values. A structure or union never has a trap representation.
Yes, nondeterministic bits would work for padding of struct/union, as described in (3) The third case is the value of struct/union padding.
For the members of struct/union, it is allowed to have trap representation, so poison can be used.

Juneyoung

4

It is important to note that this applies to trap representations and not to unspecified values. A structure or union never has a trap representation.
Yes, nondeterministic bits would work for padding of struct/union, as described in (3) The third case is the value of struct/union padding.
For the members of struct/union, it is allowed to have trap representation, so poison can be used.

At what point are the members considered poison? For copying/passing/returning a struct or union, there is no UB even if some members are uninitialized.

5

It is UB when a poison is passed to certain operations that raise UB on poison, such as division by poison/dereferencing poison pointer/branching on poison condition/etc.
Otherwise, poison is simply propagated, but it does not raise UB

Copying poison bytes is okay:

// Members are initialized to poison at object creation.
p = alloca {i8, i32} // p[0], p[4~7] are poison
q = alloca {i8, i32} // we want to copy p to q
v = load i8* p[0] // v is poison
store i8 v, i8* q[0] // poison is simply copied; no UB happened

Similarly, passing/returning poison is allowed as well.

Juneyoung

6

It is UB when a poison is passed to certain operations that raise UB on poison, such as division by poison/dereferencing poison pointer/branching on poison condition/etc.

Got it. Thanks.

Otherwise, poison is simply propagated, but it does not raise UB

Copying poison bytes is okay:

// Members are initialized to poison at object creation.
p = alloca {i8, i32} // p[0], p[4~7] are poison

p[0] is an i8, so it shouldn’t be poison?

7

// Members are initialized to poison at object creation.
p = alloca {i8, i32} // p[0], p[4~7] are poison
p[0] is an i8, so it shouldn’t be poison?

My interpretation of standard is that reading uninitialized char can also yield trap representation.
If uninitialized, char variable has indeterminate value, and C/C++ does not seem to forbid reading trap representation from it.
C++14 explicitly has an example that shows it is indeterminate value at 3.3.2.1 :


The point of declaration for a name is immediately after its complete declarator (Clause 8) and before its initializer (if any), except as noted below. [Example:
unsigned char x = 12;
{ unsigned char x = x; }
Here the second x is initialized with its own (indeterminate) value. —end example]

It seems there was a phrase saying that reading indeterminate value as an unsigned char should yield unspecified value in the C++14 draft in the past, but it is removed: http://www.open-std.org/jtc1/sc22/wg21/docs/cwg_defects.html#1787
The removed phrase did not exist in C++11, so I believe it is fine to use poison for uninitialized char types.

Juneyoung

8

// Members are initialized to poison at object creation.
p = alloca {i8, i32} // p[0], p[4~7] are poison
p[0] is an i8, so it shouldn’t be poison?

My interpretation of standard is that reading uninitialized char can also yield trap representation.
If uninitialized, char variable has indeterminate value, and C/C++ does not seem to forbid reading trap representation from it.

Okay, it’s just not immediately undefined behaviour. The C model has more issues because of the problem with how “trap representation” is defined (which precludes trap representations for unsigned char, two’s complement signed char, etc.). This interpretation is further stressed because C only explicitly ascribes undefined behaviour to trap representations on loads and stores.

C++14 explicitly has an example that shows it is indeterminate value at 3.3.2.1 :


The point of declaration for a name is immediately after its complete declarator (Clause 8) and before its initializer (if any), except as noted below. [Example:
unsigned char x = 12;
{ unsigned char x = x; }
Here the second x is initialized with its own (indeterminate) value. —end example]

It seems there was a phrase saying that reading indeterminate value as an unsigned char should yield unspecified value in the C++14 draft in the past, but it is removed: http://www.open-std.org/jtc1/sc22/wg21/docs/cwg_defects.html#1787
The removed phrase did not exist in C++11, so I believe it is fine to use poison for uninitialized char types.

Thanks for raising this. So, it can be that there are:

  • values that induce broader undefined behaviour (with certain operations),

  • values that may change on their own, and

  • values that become defined to an unspecified value (not of the first category) at some point in time.

The unspecified values put into padding in C appear to be in the last category. Is that how LLVM treats them?

9

Okay, it’s just not immediately undefined behaviour. The C model has more issues because of the problem with how “trap representation” is defined (which precludes trap representations for unsigned char, two’s complement signed char, etc.).

This interpretation is further stressed because C only explicitly ascribes undefined behaviour to trap representations on loads and stores.

In this case, freeze(poison) can be used to represent an uninitialized value, because using freeze(poison) never raises UB.
However, I couldn’t find relevant statements from C standard about these two. Could you elaborate a bit more please?

Thanks for raising this. So, it can be that there are:

  • values that induce broader undefined behaviour (with certain operations),
  • values that may change on their own, and
  • values that become defined to an unspecified value (not of the first category) at some point in time.

The unspecified values put into padding in C appear to be in the last category. Is that how LLVM treats them?

With the current version of LLVM, there is slightly a gap, and with our suggestion this becomes correct.
https://godbolt.org/z/nxhnTK

Reading padding is optimized to undef, but according to the previous link (http://www.open-std.org/jtc1/sc22/wg21/docs/cwg_defects.html#1787), this isn’t valid because unspecified value is a ‘stable’ value.
In our suggestion, well-defined bits are stored into padding at object creation, so it becomes okay.

Juneyoung

10

Okay, it’s just not immediately undefined behaviour. The C model has more issues because of the problem with how “trap representation” is defined (which precludes trap representations for unsigned char, two’s complement signed char, etc.).

This interpretation is further stressed because C only explicitly ascribes undefined behaviour to trap representations on loads and stores.

In this case, freeze(poison) can be used to represent an uninitialized value, because using freeze(poison) never raises UB.
However, I couldn’t find relevant statements from C standard about these two. Could you elaborate a bit more please?

The definition of trap representation in C17 subclause 6.2.6.1 paragraph 5 refers to object representations that do not represent a value of the type. No such object representations can exist for an unsigned char because all of the bits in the object representation of that type are value bits and every combination of value bits represent a value of the type (see C17 subclause 6.2.6.2 paragraph 1). It seems C17 still allows trap representations for two’s complement signed char, but that is only if SCHAR_MIN == -SCHAR_MAX.

11

The definition of trap representation in C17 subclause 6.2.6.1 paragraph 5 refers to object representations that do not represent a value of the type. No such object representations can exist for an unsigned char because all of the bits in the object representation of that type are value bits and every combination of value bits represent a value of the type (see C17 subclause 6.2.6.2 paragraph 1). It seems C17 still allows trap representations for two’s complement signed char, but that is only if SCHAR_MIN == -SCHAR_MAX.

I see, thank you for the info.

Juneyoung