Stream: git-wasmtime

Topic: wasmtime / issue #6013 cranelift: Add narrower and wider ...


view this post on Zulip Wasmtime GitHub notifications bot (Mar 14 2023 at 01:04):

elliottt commented on issue #6013:

Could you also diff the generated target/debug/build/cranelift-codegen-*/out/opcodes.rs before and after this change? I'm particularly interested in making sure that OPCODE_CONSTRAINTS isn't different in any interesting way except for the five instructions you've updated to use the new derived types. There are some nice comments in the generated code for the computed value of use_typevar_operand, for example.

OPCODE_CONSTRAINTS only differs in that it's now 80 elements long. There are some new constraints that uses Narrower and Wider, with two previously completely free constraints being replaced with a constrained use of the controlling variable.

view this post on Zulip Wasmtime GitHub notifications bot (Mar 14 2023 at 01:07):

elliottt edited a comment on issue #6013:

Could you also diff the generated target/debug/build/cranelift-codegen-*/out/opcodes.rs before and after this change? I'm particularly interested in making sure that OPCODE_CONSTRAINTS isn't different in any interesting way except for the five instructions you've updated to use the new derived types. There are some nice comments in the generated code for the computed value of use_typevar_operand, for example.

OPCODE_CONSTRAINTS only differs in that it's now 80 elements long. There are some new constraints that use Narrower and Wider, with two previously completely free constraints being replaced with a constrained use of the controlling variable.

view this post on Zulip Wasmtime GitHub notifications bot (Mar 14 2023 at 16:06):

elliottt commented on issue #6013:

I realized that the changes to the is_ctrl_typevar_candidate were only necessary when we were allowing an instructions return type to include a constrained use of the control type. As we needed to avoid that to ensure that instructions always had a well-defined return type, those changes could be reverted as well.

view this post on Zulip Wasmtime GitHub notifications bot (Mar 14 2023 at 16:19):

elliottt edited a comment on issue #6013:

I realized that the changes to the is_ctrl_typevar_candidate were only necessary when we were allowing an instruction's return type to include a constrained use of the control type. As we needed to avoid that to ensure that instructions always had a well-defined return type, those changes could be reverted as well.


Last updated: Dec 23 2024 at 12:05 UTC