[llvm] [AMDGPU][MC][True16] Support VOP2 instructions with true16 format (PR #115233)
Joe Nash via llvm-commits
llvm-commits at lists.llvm.org
Fri Nov 8 11:53:44 PST 2024
================
@@ -467,6 +479,42 @@ class VOP_MAC <ValueType vt0, ValueType vt1=vt0> : VOPProfile <[vt0, vt1, vt1, v
}
def VOP_MAC_F16 : VOP_MAC <f16>;
+def VOP_MAC_F16_t16 : VOP_MAC <f16> {
----------------
Sisyph wrote:
I don't see a way to reduce it much. These lets are pretty much standard for a true16 instruction. But we have the true16 lets, and all the VOP_MAC lets to support. We could inherit from VOPProfile_TRUE16 instead of VOP_MAC, but then we'd have all the same lets here as in VOP_MAC. One upside is when the default is switched to true16, we can delete 6 lines related to VOP3DPP.
https://github.com/llvm/llvm-project/pull/115233
More information about the llvm-commits
mailing list