Stream: git-wasmtime

Topic: wasmtime / issue #5168 Switch to `x64_rbp` to avoid the u...

Wasmtime GitHub notifications bot (Nov 01 2022 at 18:56):

elliottt commented on issue #5168:

@fitzgen there was a test failure due to output differences with get_frame_pointer. This pr introduces an indirection through %rdi on the fp_sp_pc.clif test.

Wasmtime GitHub notifications bot (Nov 01 2022 at 19:01):

jameysharp commented on issue #5168:

What is the MovPReg instruction doing? Is it possible to ensure it doesn't emit any instructions in cases like fp_sp_pc.clif? I hoped it would just impose a register constraint.

Wasmtime GitHub notifications bot (Nov 01 2022 at 19:35):

cfallin commented on issue #5168:

What is the MovPReg instruction doing? Is it possible to ensure it doesn't emit any instructions in cases like fp_sp_pc.clif? I hoped it would just impose a register constraint.

It always emits a move; unfortunately this is required, and the register constraint approach is not possible, because the register (rbp) is not otherwise allocatable. We can't let regalloc decide that the new vreg that was copied from rbp now lives in rbp; the original value in rbp is still live, but that's not tracked by regalloc.

Last updated: Nov 22 2024 at 17:03 UTC