projects / platform / upstream / llvm.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: 5798be8) | patch
[polly][GPGPU] Fixup related to overloading exponent type in llvm.powi
authorBjorn Pettersson <bjorn.a.pettersson@ericsson.com>
Fri, 18 Jun 2021 06:58:44 +0000 (08:58 +0200)
committerBjorn Pettersson <bjorn.a.pettersson@ericsson.com>
Fri, 18 Jun 2021 06:59:06 +0000 (08:59 +0200)
commit6aac2773d80b42bcf9da2414bbd410aa9f10095e
tree20c795a6bf4012fcaa88432f28bc6644b2c06934tree | snapshot
parent5798be84580be233e4cf34c08ceec8f79e80502ecommit | diff
[polly][GPGPU] Fixup related to overloading exponent type in llvm.powi

Commit 4c7f820b2b206bb changed the llvm.powi intrinsic to support
different 'int' sizes for the exponent. That happened to break
the IntrinsicToLibdeviceFunc mapping in PPCGCodeGeneration, which
obviously should have been updated as part of commit 4c7f820b2b
(https://reviews.llvm.org/D99439).

The shortcoming was found by buildbots that use
   -DPOLLY_ENABLE_GPGPU_CODEGEN=ON

This patch should fixup the problem.
polly/lib/CodeGen/PPCGCodeGeneration.cpp diff | blob | history
polly/test/GPGPU/libdevice-functions-copied-into-kernel.ll diff | blob | history
Domain: System / Toolchain;