Add zeroext/signext attributes to C ABI functions where appropriate #31725
Conversation
We need to supply sext/zext attributes to LLVM to ensure that arguments are extended to the appropriate width in the correct way. This commit adds the necessary infrastructure for handling integer widening and makes the x86-64 code produce the correct information.
Most platforms extend integers less than 32 bits, though not all.
|
r? @arielb1 (rust_highfive has picked a reviewer for you, use r? to override) |
| @@ -0,0 +1,4 @@ | |||
| // ignore-license | |||
| int foo(char x) { | |||
There was a problem hiding this comment.
I think char is signed already. At least, clang specifies that they're signext.
There was a problem hiding this comment.
char is a distinct type from signed char and unsigned char and its signedness is implementation defined. And according to e.g. http://stackoverflow.com/questions/3728045/any-compiler-which-takes-char-as-unsigned char is in fact unsigned on ARM.
|
r? @dotdash |
| @@ -0,0 +1,10 @@ | |||
| -include ../tools.mk | |||
|
|
|||
| all: $(call NATIVE_STATICLIB,cfoo) | |||
There was a problem hiding this comment.
This might be most easily added actually as part of src/rt/rust_test_helpers.c and then a standard run-pass test can be added. Helps avoid run-make at least!
|
☔ The latest upstream changes (presumably #31742) made this pull request unmergeable. Please resolve the merge conflicts. |
|
Updated version of this is #32732 |
The C ABI on most platforms expects integers to be extended up to the appropriate size. This requires annotation in LLVM to indicate whether or not to do sign extension or zero extension. Foreign code may rely on the extension behaviour and for signed integers the value passed may be incorrect.
Fixes #30841 and #31315