[all-commits] [llvm/llvm-project] dac020: [HIP] Perform implicit pointer cast when compiling...
Juan Manuel Martinez Caamaño via All-commits
all-commits at lists.llvm.org
Tue Oct 28 06:24:00 PDT 2025
Branch: refs/heads/users/jmmartinez/fix/load_lds_typesignature/2
Home: https://github.com/llvm/llvm-project
Commit: dac020c6c2cf10ab3963ae8c236851d869f6ae1d
https://github.com/llvm/llvm-project/commit/dac020c6c2cf10ab3963ae8c236851d869f6ae1d
Author: Juan Manuel Martinez Caamaño <jmartinezcaamao at gmail.com>
Date: 2025-10-28 (Tue, 28 Oct 2025)
Changed paths:
M clang/lib/Sema/SemaExpr.cpp
M clang/test/SemaHIP/amdgpu-gfx950-load-to-lds.hip
Log Message:
-----------
[HIP] Perform implicit pointer cast when compiling device code, not when -fcuda-is-device
When compiling HIP device code, we add implicit casts for the pointer
arguments being passed to builtin calls.
When compiling for the host, apply the same casts for __device__ or __kernel__ functions,
since the device side of the source should still pass type checks.
This patch changes the condition depending on -fcuda-is-device to depend
on if the builtin's caller is marked as __device__ or __kernel__.
Commit: d70fd596bcc6e177fc7770b6c16d03f5f13fae97
https://github.com/llvm/llvm-project/commit/d70fd596bcc6e177fc7770b6c16d03f5f13fae97
Author: Juan Manuel Martinez Caamaño <jmartinezcaamao at gmail.com>
Date: 2025-10-28 (Tue, 28 Oct 2025)
Changed paths:
M clang/test/SemaHIP/amdgpu-gfx950-load-to-lds.hip
Log Message:
-----------
[NFC][HIP] Add __builtin_*_load_lds type check test cases
This tests show how typechecking is performed for
__builtin_amdgcn_load_to_lds, but not for
__builtin_amdgcn_raw_ptr_buffer_load_lds,
__builtin_amdgcn_struct_ptr_buffer_load_lds and
__builtin_amdgcn_global_load_lds since they are declared with the 't'
attribute.
Compare: https://github.com/llvm/llvm-project/compare/dac020c6c2cf%5E...d70fd596bcc6
To unsubscribe from these emails, change your notification settings at https://github.com/llvm/llvm-project/settings/notifications
More information about the All-commits
mailing list