[PATCH] D61634: [clang/llvm] Allow efficient implementation of libc's memory functions in C/C++

[Guillaume Chatelet via Phabricator via llvm-commits]

gchatelet added a comment.

In D61634#1586047 <https://reviews.llvm.org/D61634#1586047>, @tejohnson wrote:

> Checking in to see where we are on this issue. I haven't had any time to work on 4 but hopefully can start on that soon. But it needs the first part done to be effective.


Thx for the heads up @tejohnson.
The patch is missing a bit of documentation but it shouldn't be too far from complete:

- it adds a clang function attribute, e.g. `__attribute__((no_builtin("memcpy")))`
- instructs clang to compose individual IR attributes from the clang attribute above, e.g. `no-builtin-memcpy`, `no-builtin-memset`, `no-builtin-sqrt`...
- adds a specific builtin to clang for the memcpy case `__builtin_memcpy_inline`
- adds an LLVM IR intrinsic `int_memcpy_inline`
- adds an LLVM Instruction `MemCpyInlineInst`
- instructs LLVM to forward the `no-builtin-memcpy` IR attribute from the function declaration to the actual memcpy calls inside the function's body (same for `memset` and `memmove`)
- adds code to turn the `MemCpyInlineInst` into code, using `DAG.getMemcpy` with `always_inline` set.

Left to do:

- finish implementing memset / memmove
- check attributes and reject the one that are not implemented
- add documentation

There's too many things going on in this patch and it may worth splitting it.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D61634/new/

https://reviews.llvm.org/D61634