| @@ -159,6 +159,17 @@ Changes to the AArch64 Backend |
|
|
| 159 |
159 |
Changes to the AMDGPU Backend |
| 160 |
160 |
----------------------------- |
| 161 |
161 |
|
|
162 |
+* Initial support for gfx950 |
|
163 |
+ |
|
164 |
+* Improved ``llvm.memcpy``, ``llvm.memmove`` and ``llvm.memset`` lowering |
|
165 |
+ |
|
166 |
+* Fixed expansion of 64-bit flat address space ``atomicrmw`` and |
|
167 |
+``cmpxchg`` operations which may access private |
|
168 |
+ memory. `noalias.addrspace` metadat may be used to avoid the |
|
169 |
+ expansion if the target address is known to not be on the stack. |
|
170 |
+ |
|
171 |
+* Fix compile failures when emitting unreachable functions. |
|
172 |
+ |
| 162 |
173 |
* Removed `llvm.amdgcn.flat.atomic.fadd` and |
| 163 |
174 |
`llvm.amdgcn.global.atomic.fadd` intrinsics. Users should use the |
| 164 |
175 |
{ref}`atomicrmw <i_atomicrmw>` instruction with `fadd` and |