arsenm wrote:
> Interesting, the important bit is that this is done before the LTO link so we
> know which functions to pull in. Unfortunately I don't think there's a good
> location that actually states which math calls are supported, but I could
> probably make a script that goes through eac
Artem-B wrote:
Given that the prevalent compilation for CUDA has no standard library
whatsoever, preserving libcalls may break some existing users that may be
relying on library call lowering to an intrinsic that *is* implemented by the
back-end.
Perhaps this "no library call to intrinsic con
https://github.com/jhuber6 updated
https://github.com/llvm/llvm-project/pull/98209
>From 605e9e78c1cba3b1947a538c566ffedbb9525be0 Mon Sep 17 00:00:00 2001
From: Joseph Huber
Date: Wed, 10 Jul 2024 09:39:44 -0500
Subject: [PATCH 1/2] [Clang] Correctly enable the f16 type for offloading
Summary:
jhuber6 wrote:
> This is going to break the library build. We use the __builtin functions to
> access the intrinsic in the cases where the llvm intrinsic lowering provides
> the implementation of the function. In a more sensible world, the library
> would not provide the implementations of fun
https://github.com/arsenm edited https://github.com/llvm/llvm-project/pull/98209
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
https://github.com/arsenm commented:
This is going to break the library build. We use the __builtin functions to
access the intrinsic in the cases where the llvm intrinsic lowering provides
the implementation of the function. In a more sensible world, the library would
not provide the implemen
@@ -0,0 +1,51 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
UTC_ARGS: --version 5
+// RUN: %clang_cc1 -triple amdgcn-amd-amdhsa %s -emit-llvm -o - | FileCheck %s
--check-prefix AMDGPU
+// RUN: %clang_cc1 -triple nvptx64-nvidia-cuda %s -emit-ll
https://github.com/jhuber6 created
https://github.com/llvm/llvm-project/pull/98209
Summary:
Currently, the GPU gets its math by using wrapper headers that eagerly
replace libcalls with calls to the vendor's math library. e.g.
```
// __clang_cuda_math.h
[[gnu::always_inline]] double sin(double __
llvmbot wrote:
@llvm/pr-subscribers-clang
@llvm/pr-subscribers-clang-codegen
Author: Joseph Huber (jhuber6)
Changes
Summary:
Currently, the GPU gets its math by using wrapper headers that eagerly
replace libcalls with calls to the vendor's math library. e.g.
```
// __clang_cuda_math.h
[[gn
