[PATCH] CodeGen target hook for emitting intrinsic or libcall for pow*

[John McCall]

On Jul 10, 2013, at 12:30 PM, Eli Bendersky <eliben at google.com> wrote:
> On Wed, Jul 10, 2013 at 11:53 AM, John McCall <rjmccall at apple.com> wrote:
> On Jul 10, 2013, at 11:39 AM, Eli Bendersky <eliben at google.com> wrote:
>> On Wed, Jul 10, 2013 at 11:26 AM, John McCall <rjmccall at apple.com> wrote:
>> On Jul 10, 2013, at 11:18 AM, Eli Bendersky <eliben at google.com> wrote:
>>> On Wed, Jul 10, 2013 at 3:03 AM, John McCall <rjmccall at apple.com> wrote:
>>> On Jun 27, 2013, at 3:57 PM, Eli Bendersky <eliben at google.com> wrote:
>>> > Without fmath-errno, Clang currently generates calls to @llvm.pow.* intrinsics when it sees pow*(). This may not be suitable for all targets (for example PNaCl), so the attached patch adds a target hook that CodeGen queries. The target can state its preference for having or not having the intrinsic generated. Non-PNaCl behavior remains unchanged; PNaCl-specific test added.
>>> 
>>> Isn't a more straightforward and less invasive approach to just invoke the compiler with -fno-builtin=pow or something along those lines?
>>> 
>>> We don't control the flags users supply to Clang though. 
>> 
>> Then handle it in the driver.
>> 
>> Would that not be more intrusive, then? Or perhaps I'm misunderstanding what you're proposing.
> 
> It would be isolated to target-specific logic in target-specific parts of the toolchain instead of inventing a new option that's basically "do what PNaCl wants" and then propagating it through the entire compiler.
> 
> Clang currently supports le32 as a target. So it's not an issue of a single "driver wrapper". We purposefully keep PNaCl as open as possible, and other developers and toolchain writers may use this target to generate PNaCl bitcode or something similar with Clang.

I'm not asking you to wrap the driver.  I'm asking you to change clang's driver logic when targeting *-*-nacl to use the existing logic to suppress builtin treatment of functions.

> There's no new option added and no additional plumbing done in this change. Target hooks are used in a number of places in Clang codegen already to customize parts of the process. A new hook was added for this specific purpose, and the only PNaCl-specific part of the code is in PNaClTargetCodeGenInfo. This is not unlike already existing target hooks for architectures, OSes, etc.

Those hooks are used for mandatory ABI things, though, not to adjust our interpretation of builtin functions.  The builtin model is already complicated enough.

John.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20130710/6bc2b59f/attachment.html>