[ghstack-poisoned]

In order to support effect tokens with invoke subgraph in Inductor, I reimplemented how we handle with_effects, which now has a similar behavior as control_deps.

Previously the behavior of with_effects is that we turn each with_effects(op, ..) call into an `ir.EffectfulKernel`, and the `op` is treated as a FallbackKernel. However in the case of `invoke_subgraph`, we want it to be an `ir.InvokeSubgraph` to trigger the invoke_subgraph inductor lowering logic. So now instead of turning with_effects into an `ir.EffectfulKernel`, I just add this dependency information to `V.graph.additional_star_deps`, similar to `V.graph.additional_buffer_deps` which is used by `control_deps`. 




cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov coconutruben

[ghstack-poisoned]

In order to support effect tokens with invoke subgraph in Inductor, I reimplemented how we handle with_effects, which now has a similar behavior as control_deps.

Previously the behavior of with_effects is that we turn each with_effects(op, ..) call into an `ir.EffectfulKernel`, and the `op` is treated as a FallbackKernel. However in the case of `invoke_subgraph`, we want it to be an `ir.InvokeSubgraph` to trigger the invoke_subgraph inductor lowering logic. So now instead of turning with_effects into an `ir.EffectfulKernel`, I just add this dependency information to `V.graph.additional_star_deps`, similar to `V.graph.additional_buffer_deps` which is used by `control_deps`. 




cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov coconutruben

[ghstack-poisoned]

In order to support effect tokens with invoke subgraph in Inductor, I reimplemented how we handle with_effects, which now has a similar behavior as control_deps.

Previously the behavior of with_effects is that we turn each with_effects(op, ..) call into an `ir.EffectfulKernel`, and the `op` is treated as a FallbackKernel. However in the case of `invoke_subgraph`, we want it to be an `ir.InvokeSubgraph` to trigger the invoke_subgraph inductor lowering logic. So now instead of turning with_effects into an `ir.EffectfulKernel`, I just add this dependency information to `V.graph.additional_star_deps`, similar to `V.graph.additional_buffer_deps` which is used by `control_deps`. 




cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov coconutruben

[ghstack-poisoned]

In order to support effect tokens with invoke subgraph in Inductor, I reimplemented how we handle with_effects, which now has a similar behavior as control_deps.

Previously the behavior of with_effects is that we turn each with_effects(op, ..) call into an `ir.EffectfulKernel`, and the `op` is treated as a FallbackKernel. However in the case of `invoke_subgraph`, we want it to be an `ir.InvokeSubgraph` to trigger the invoke_subgraph inductor lowering logic. So now instead of turning with_effects into an `ir.EffectfulKernel`, I just add this dependency information to `V.graph.additional_star_deps`, similar to `V.graph.additional_buffer_deps` which is used by `control_deps`. 




cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov coconutruben

[ghstack-poisoned]

In order to support effect tokens with invoke subgraph in Inductor, I reimplemented how we handle with_effects, which now has a similar behavior as control_deps.

Previously the behavior of with_effects is that we turn each with_effects(op, ..) call into an `ir.EffectfulKernel`, and the `op` is treated as a FallbackKernel. However in the case of `invoke_subgraph`, we want it to be an `ir.InvokeSubgraph` to trigger the invoke_subgraph inductor lowering logic. So now instead of turning with_effects into an `ir.EffectfulKernel`, I just add this dependency information to `V.graph.additional_star_deps`, similar to `V.graph.additional_buffer_deps` which is used by `control_deps`. 




cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov coconutruben

[ghstack-poisoned]

In order to support effect tokens with invoke subgraph in Inductor, I reimplemented how we handle with_effects, which now has a similar behavior as control_deps.

Previously the behavior of with_effects is that we turn each with_effects(op, ..) call into an `ir.EffectfulKernel`, and the `op` is treated as a FallbackKernel. However in the case of `invoke_subgraph`, we want it to be an `ir.InvokeSubgraph` to trigger the invoke_subgraph inductor lowering logic. So now instead of turning with_effects into an `ir.EffectfulKernel`, I just add this dependency information to `V.graph.additional_star_deps`, similar to `V.graph.additional_buffer_deps` which is used by `control_deps`. 




cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov coconutruben

Differential Revision: [D87668982](https://our.internmc.facebook.com/intern/diff/D87668982)

[ghstack-poisoned]

In order to support effect tokens with invoke subgraph in Inductor, I reimplemented how we handle with_effects, which now has a similar behavior as control_deps.

Previously the behavior of with_effects is that we turn each with_effects(op, ..) call into an `ir.EffectfulKernel`, and the `op` is treated as a FallbackKernel. However in the case of `invoke_subgraph`, we want it to be an `ir.InvokeSubgraph` to trigger the invoke_subgraph inductor lowering logic. So now instead of turning with_effects into an `ir.EffectfulKernel`, I just add this dependency information to `V.graph.additional_star_deps`, similar to `V.graph.additional_buffer_deps` which is used by `control_deps`. 




cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov coconutruben

Differential Revision: [D87668982](https://our.internmc.facebook.com/intern/diff/D87668982)

[ghstack-poisoned]