Fix regression from RuntimeType.AllocateValueType rewrite by jkoritzinsky · Pull Request #101137 · dotnet/runtime

jkoritzinsky · 2024-04-16T19:30:23Z

Testing out fixes for #101134 and #104777

jkoritzinsky · 2024-04-16T19:30:34Z

/benchmark help

jkoritzinsky · 2024-04-16T21:02:27Z

/benchmark micro aspnet-perf-win runtime --variable filter=System.Reflection.Invoke.StaticMethod

pr-benchmarks · 2024-04-16T21:02:37Z

Benchmark started for micro on aspnet-perf-win with runtime and arguments --variable filter=*System.Reflection.Invoke.StaticMethod*. Logs: link

pr-benchmarks · 2024-04-16T21:34:25Z

micro - aspnet-perf-win

| benchmark | mean (micro.base) | mean (micro.pr) | ratio | allocated (micro.base) | allocated (micro.pr) | ratio || --------------------------------------------------------- | ----------------- | --------------- | ----- | ---------------------- | -------------------- | ----- || StaticMethod4_arrayNotCached_int_string_struct_class | 31.80 ns | 30.08 ns | 0.95 | 104 B | 104 B | 1.00 || StaticMethod4_ByRefParams_int_string_struct_class | 138.6 ns | 137.7 ns | 0.99 | 48 B | 48 B | 1.00 || StaticMethod4_int_string_struct_class | 19.52 ns | 19.80 ns | 1.01 | 0 B | 0 B ||| StaticMethod5_arrayNotCached_int_string_struct_class_bool | 45.05 ns | 43.55 ns | 0.97 | 136 B | 136 B | 1.00 || StaticMethod5_ByRefParams_int_string_struct_class_bool | 197.7 ns | 206.7 ns | 1.05 | 72 B | 72 B | 1.00 |

jkoritzinsky · 2024-04-16T21:56:54Z

Looks like this helps, but not enough.

jkoritzinsky · 2024-04-16T22:34:44Z

/benchmark micro aspnet-perf-win runtime --variable filter=System.Reflection.Invoke.StaticMethod

pr-benchmarks · 2024-04-16T22:34:59Z

Benchmark started for micro on aspnet-perf-win with runtime and arguments --variable filter=*System.Reflection.Invoke.StaticMethod*. Logs: link

pr-benchmarks · 2024-04-16T23:11:55Z

micro - aspnet-perf-win

| benchmark | mean (micro.base) | mean (micro.pr) | ratio | allocated (micro.base) | allocated (micro.pr) | ratio || --------------------------------------------------------- | ----------------- | --------------- | ----- | ---------------------- | -------------------- | ----- || StaticMethod4_arrayNotCached_int_string_struct_class | 29.41 ns | 30.07 ns | 1.02 | 104 B | 104 B | 1.00 || StaticMethod4_ByRefParams_int_string_struct_class | 142.15 ns | 87.21 ns | 0.61 | 48 B | 48 B | 1.00 || StaticMethod4_int_string_struct_class | 20.62 ns | 19.83 ns | 0.96 | 0 B | 0 B ||| StaticMethod5_arrayNotCached_int_string_struct_class_bool | 42.85 ns | 43.83 ns | 1.02 | 136 B | 136 B | 1.00 || StaticMethod5_ByRefParams_int_string_struct_class_bool | 192.8 ns | 120.2 ns | 0.62 | 72 B | 72 B | 1.00 |

dotnet-policy-service · 2024-04-17T17:37:36Z

Tagging subscribers to this area: @mangod9
See info in area-owners.md if you want to be subscribed.

jkotas · 2024-04-19T03:27:18Z

 SetJitHelperFunction(CORINFO_HELP_NEWARR_1_OBJ, JIT_NewArr1OBJ_UP);

 ECall::DynamicallyAssignFCallImpl(GetEEFuncEntryPoint(AllocateStringFastUP), ECall::FastAllocateString);
+ ECall::DynamicallyAssignFCallImpl(GetEEFuncEntryPoint(JIT_BoxFastUP), ECall::NonNullableBox);


This block is Windows specific, so this change only helps on Windows. We need an equivalent fix for non-Windows platforms too.

Yeah I'll probably revert this particular change and make an equivalent one that's more xplat-friendly.

I've inlined the trail allocation logic into JIT_Box and factored out the rest of the logic to be in a separate framed helper instead of using the dynamic FCALL impl logic.
Let me know if this looks good to you.

jkoritzinsky · 2024-06-04T17:25:10Z

/benchmark micro aspnet-perf-win runtime --variable filter=System.Reflection.Invoke.StaticMethod

jkotas

Thanks!

jkoritzinsky · 2024-06-05T01:00:28Z

/ba-g timeouts unrelated

jkoritzinsky · 2024-04-16T19:30:23Z

Testing out fixes for #101134 and #104777

jkoritzinsky · 2024-04-16T19:30:34Z

/benchmark help

jkoritzinsky · 2024-04-16T21:02:27Z

/benchmark micro aspnet-perf-win runtime --variable filter=System.Reflection.Invoke.StaticMethod

pr-benchmarks · 2024-04-16T21:02:37Z

Benchmark started for micro on aspnet-perf-win with runtime and arguments --variable filter=*System.Reflection.Invoke.StaticMethod*. Logs: link

pr-benchmarks · 2024-04-16T21:34:25Z

micro - aspnet-perf-win

| benchmark | mean (micro.base) | mean (micro.pr) | ratio | allocated (micro.base) | allocated (micro.pr) | ratio || --------------------------------------------------------- | ----------------- | --------------- | ----- | ---------------------- | -------------------- | ----- || StaticMethod4_arrayNotCached_int_string_struct_class | 31.80 ns | 30.08 ns | 0.95 | 104 B | 104 B | 1.00 || StaticMethod4_ByRefParams_int_string_struct_class | 138.6 ns | 137.7 ns | 0.99 | 48 B | 48 B | 1.00 || StaticMethod4_int_string_struct_class | 19.52 ns | 19.80 ns | 1.01 | 0 B | 0 B ||| StaticMethod5_arrayNotCached_int_string_struct_class_bool | 45.05 ns | 43.55 ns | 0.97 | 136 B | 136 B | 1.00 || StaticMethod5_ByRefParams_int_string_struct_class_bool | 197.7 ns | 206.7 ns | 1.05 | 72 B | 72 B | 1.00 |

jkoritzinsky · 2024-04-16T21:56:54Z

Looks like this helps, but not enough.

jkoritzinsky · 2024-04-16T22:34:44Z

/benchmark micro aspnet-perf-win runtime --variable filter=System.Reflection.Invoke.StaticMethod

pr-benchmarks · 2024-04-16T22:34:59Z

Benchmark started for micro on aspnet-perf-win with runtime and arguments --variable filter=*System.Reflection.Invoke.StaticMethod*. Logs: link

pr-benchmarks · 2024-04-16T23:11:55Z

micro - aspnet-perf-win

| benchmark | mean (micro.base) | mean (micro.pr) | ratio | allocated (micro.base) | allocated (micro.pr) | ratio || --------------------------------------------------------- | ----------------- | --------------- | ----- | ---------------------- | -------------------- | ----- || StaticMethod4_arrayNotCached_int_string_struct_class | 29.41 ns | 30.07 ns | 1.02 | 104 B | 104 B | 1.00 || StaticMethod4_ByRefParams_int_string_struct_class | 142.15 ns | 87.21 ns | 0.61 | 48 B | 48 B | 1.00 || StaticMethod4_int_string_struct_class | 20.62 ns | 19.83 ns | 0.96 | 0 B | 0 B ||| StaticMethod5_arrayNotCached_int_string_struct_class_bool | 42.85 ns | 43.83 ns | 1.02 | 136 B | 136 B | 1.00 || StaticMethod5_ByRefParams_int_string_struct_class_bool | 192.8 ns | 120.2 ns | 0.62 | 72 B | 72 B | 1.00 |

dotnet-policy-service · 2024-04-17T17:37:36Z

Tagging subscribers to this area: @mangod9
See info in area-owners.md if you want to be subscribed.

jkotas · 2024-04-19T03:27:18Z

 SetJitHelperFunction(CORINFO_HELP_NEWARR_1_OBJ, JIT_NewArr1OBJ_UP);

 ECall::DynamicallyAssignFCallImpl(GetEEFuncEntryPoint(AllocateStringFastUP), ECall::FastAllocateString);
+ ECall::DynamicallyAssignFCallImpl(GetEEFuncEntryPoint(JIT_BoxFastUP), ECall::NonNullableBox);


This block is Windows specific, so this change only helps on Windows. We need an equivalent fix for non-Windows platforms too.

Yeah I'll probably revert this particular change and make an equivalent one that's more xplat-friendly.

I've inlined the trail allocation logic into JIT_Box and factored out the rest of the logic to be in a separate framed helper instead of using the dynamic FCALL impl logic.
Let me know if this looks good to you.

jkoritzinsky · 2024-06-04T17:25:10Z

/benchmark micro aspnet-perf-win runtime --variable filter=System.Reflection.Invoke.StaticMethod

jkotas

Thanks!

jkoritzinsky · 2024-06-05T01:00:28Z

/ba-g timeouts unrelated

jkoritzinsky · 2024-04-16T19:30:23Z

Testing out fixes for #101134 and #104777

jkoritzinsky · 2024-04-16T19:30:34Z

/benchmark help

jkoritzinsky · 2024-04-16T21:02:27Z

/benchmark micro aspnet-perf-win runtime --variable filter=System.Reflection.Invoke.StaticMethod

pr-benchmarks · 2024-04-16T21:02:37Z

Benchmark started for micro on aspnet-perf-win with runtime and arguments --variable filter=*System.Reflection.Invoke.StaticMethod*. Logs: link

pr-benchmarks · 2024-04-16T21:34:25Z

micro - aspnet-perf-win

| benchmark | mean (micro.base) | mean (micro.pr) | ratio | allocated (micro.base) | allocated (micro.pr) | ratio || --------------------------------------------------------- | ----------------- | --------------- | ----- | ---------------------- | -------------------- | ----- || StaticMethod4_arrayNotCached_int_string_struct_class | 31.80 ns | 30.08 ns | 0.95 | 104 B | 104 B | 1.00 || StaticMethod4_ByRefParams_int_string_struct_class | 138.6 ns | 137.7 ns | 0.99 | 48 B | 48 B | 1.00 || StaticMethod4_int_string_struct_class | 19.52 ns | 19.80 ns | 1.01 | 0 B | 0 B ||| StaticMethod5_arrayNotCached_int_string_struct_class_bool | 45.05 ns | 43.55 ns | 0.97 | 136 B | 136 B | 1.00 || StaticMethod5_ByRefParams_int_string_struct_class_bool | 197.7 ns | 206.7 ns | 1.05 | 72 B | 72 B | 1.00 |

jkoritzinsky · 2024-04-16T21:56:54Z

Looks like this helps, but not enough.

jkoritzinsky · 2024-04-16T22:34:44Z

/benchmark micro aspnet-perf-win runtime --variable filter=System.Reflection.Invoke.StaticMethod

pr-benchmarks · 2024-04-16T22:34:59Z

Benchmark started for micro on aspnet-perf-win with runtime and arguments --variable filter=*System.Reflection.Invoke.StaticMethod*. Logs: link

pr-benchmarks · 2024-04-16T23:11:55Z

micro - aspnet-perf-win

| benchmark | mean (micro.base) | mean (micro.pr) | ratio | allocated (micro.base) | allocated (micro.pr) | ratio || --------------------------------------------------------- | ----------------- | --------------- | ----- | ---------------------- | -------------------- | ----- || StaticMethod4_arrayNotCached_int_string_struct_class | 29.41 ns | 30.07 ns | 1.02 | 104 B | 104 B | 1.00 || StaticMethod4_ByRefParams_int_string_struct_class | 142.15 ns | 87.21 ns | 0.61 | 48 B | 48 B | 1.00 || StaticMethod4_int_string_struct_class | 20.62 ns | 19.83 ns | 0.96 | 0 B | 0 B ||| StaticMethod5_arrayNotCached_int_string_struct_class_bool | 42.85 ns | 43.83 ns | 1.02 | 136 B | 136 B | 1.00 || StaticMethod5_ByRefParams_int_string_struct_class_bool | 192.8 ns | 120.2 ns | 0.62 | 72 B | 72 B | 1.00 |

dotnet-policy-service · 2024-04-17T17:37:36Z

Tagging subscribers to this area: @mangod9
See info in area-owners.md if you want to be subscribed.

jkotas · 2024-04-19T03:27:18Z

 SetJitHelperFunction(CORINFO_HELP_NEWARR_1_OBJ, JIT_NewArr1OBJ_UP);

 ECall::DynamicallyAssignFCallImpl(GetEEFuncEntryPoint(AllocateStringFastUP), ECall::FastAllocateString);
+ ECall::DynamicallyAssignFCallImpl(GetEEFuncEntryPoint(JIT_BoxFastUP), ECall::NonNullableBox);


This block is Windows specific, so this change only helps on Windows. We need an equivalent fix for non-Windows platforms too.

Yeah I'll probably revert this particular change and make an equivalent one that's more xplat-friendly.

I've inlined the trail allocation logic into JIT_Box and factored out the rest of the logic to be in a separate framed helper instead of using the dynamic FCALL impl logic.
Let me know if this looks good to you.

jkoritzinsky · 2024-06-04T17:25:10Z

/benchmark micro aspnet-perf-win runtime --variable filter=System.Reflection.Invoke.StaticMethod

jkotas

Thanks!

jkoritzinsky · 2024-06-05T01:00:28Z

/ba-g timeouts unrelated

jkoritzinsky · 2024-04-16T19:30:23Z

Testing out fixes for #101134 and #104777

jkoritzinsky · 2024-04-16T19:30:34Z

/benchmark help

jkoritzinsky · 2024-04-16T21:02:27Z

/benchmark micro aspnet-perf-win runtime --variable filter=System.Reflection.Invoke.StaticMethod

pr-benchmarks · 2024-04-16T21:02:37Z

Benchmark started for micro on aspnet-perf-win with runtime and arguments --variable filter=*System.Reflection.Invoke.StaticMethod*. Logs: link

pr-benchmarks · 2024-04-16T21:34:25Z

micro - aspnet-perf-win

| benchmark | mean (micro.base) | mean (micro.pr) | ratio | allocated (micro.base) | allocated (micro.pr) | ratio || --------------------------------------------------------- | ----------------- | --------------- | ----- | ---------------------- | -------------------- | ----- || StaticMethod4_arrayNotCached_int_string_struct_class | 31.80 ns | 30.08 ns | 0.95 | 104 B | 104 B | 1.00 || StaticMethod4_ByRefParams_int_string_struct_class | 138.6 ns | 137.7 ns | 0.99 | 48 B | 48 B | 1.00 || StaticMethod4_int_string_struct_class | 19.52 ns | 19.80 ns | 1.01 | 0 B | 0 B ||| StaticMethod5_arrayNotCached_int_string_struct_class_bool | 45.05 ns | 43.55 ns | 0.97 | 136 B | 136 B | 1.00 || StaticMethod5_ByRefParams_int_string_struct_class_bool | 197.7 ns | 206.7 ns | 1.05 | 72 B | 72 B | 1.00 |

jkoritzinsky · 2024-04-16T21:56:54Z

Looks like this helps, but not enough.

jkoritzinsky · 2024-04-16T22:34:44Z

/benchmark micro aspnet-perf-win runtime --variable filter=System.Reflection.Invoke.StaticMethod

pr-benchmarks · 2024-04-16T22:34:59Z

Benchmark started for micro on aspnet-perf-win with runtime and arguments --variable filter=*System.Reflection.Invoke.StaticMethod*. Logs: link

pr-benchmarks · 2024-04-16T23:11:55Z

micro - aspnet-perf-win

| benchmark | mean (micro.base) | mean (micro.pr) | ratio | allocated (micro.base) | allocated (micro.pr) | ratio || --------------------------------------------------------- | ----------------- | --------------- | ----- | ---------------------- | -------------------- | ----- || StaticMethod4_arrayNotCached_int_string_struct_class | 29.41 ns | 30.07 ns | 1.02 | 104 B | 104 B | 1.00 || StaticMethod4_ByRefParams_int_string_struct_class | 142.15 ns | 87.21 ns | 0.61 | 48 B | 48 B | 1.00 || StaticMethod4_int_string_struct_class | 20.62 ns | 19.83 ns | 0.96 | 0 B | 0 B ||| StaticMethod5_arrayNotCached_int_string_struct_class_bool | 42.85 ns | 43.83 ns | 1.02 | 136 B | 136 B | 1.00 || StaticMethod5_ByRefParams_int_string_struct_class_bool | 192.8 ns | 120.2 ns | 0.62 | 72 B | 72 B | 1.00 |

dotnet-policy-service · 2024-04-17T17:37:36Z

Tagging subscribers to this area: @mangod9
See info in area-owners.md if you want to be subscribed.

jkotas · 2024-04-19T03:27:18Z

 SetJitHelperFunction(CORINFO_HELP_NEWARR_1_OBJ, JIT_NewArr1OBJ_UP);

 ECall::DynamicallyAssignFCallImpl(GetEEFuncEntryPoint(AllocateStringFastUP), ECall::FastAllocateString);
+ ECall::DynamicallyAssignFCallImpl(GetEEFuncEntryPoint(JIT_BoxFastUP), ECall::NonNullableBox);


This block is Windows specific, so this change only helps on Windows. We need an equivalent fix for non-Windows platforms too.

Yeah I'll probably revert this particular change and make an equivalent one that's more xplat-friendly.

I've inlined the trail allocation logic into JIT_Box and factored out the rest of the logic to be in a separate framed helper instead of using the dynamic FCALL impl logic.
Let me know if this looks good to you.

jkoritzinsky · 2024-06-04T17:25:10Z

/benchmark micro aspnet-perf-win runtime --variable filter=System.Reflection.Invoke.StaticMethod

jkotas

Thanks!

jkoritzinsky · 2024-06-05T01:00:28Z

/ba-g timeouts unrelated

jkoritzinsky · 2024-04-16T19:30:23Z

Testing out fixes for #101134 and #104777

jkoritzinsky · 2024-04-16T19:30:34Z

/benchmark help

jkoritzinsky · 2024-04-16T21:02:27Z

/benchmark micro aspnet-perf-win runtime --variable filter=System.Reflection.Invoke.StaticMethod

pr-benchmarks · 2024-04-16T21:02:37Z

Benchmark started for micro on aspnet-perf-win with runtime and arguments --variable filter=*System.Reflection.Invoke.StaticMethod*. Logs: link

pr-benchmarks · 2024-04-16T21:34:25Z

micro - aspnet-perf-win

| benchmark | mean (micro.base) | mean (micro.pr) | ratio | allocated (micro.base) | allocated (micro.pr) | ratio || --------------------------------------------------------- | ----------------- | --------------- | ----- | ---------------------- | -------------------- | ----- || StaticMethod4_arrayNotCached_int_string_struct_class | 31.80 ns | 30.08 ns | 0.95 | 104 B | 104 B | 1.00 || StaticMethod4_ByRefParams_int_string_struct_class | 138.6 ns | 137.7 ns | 0.99 | 48 B | 48 B | 1.00 || StaticMethod4_int_string_struct_class | 19.52 ns | 19.80 ns | 1.01 | 0 B | 0 B ||| StaticMethod5_arrayNotCached_int_string_struct_class_bool | 45.05 ns | 43.55 ns | 0.97 | 136 B | 136 B | 1.00 || StaticMethod5_ByRefParams_int_string_struct_class_bool | 197.7 ns | 206.7 ns | 1.05 | 72 B | 72 B | 1.00 |

jkoritzinsky · 2024-04-16T21:56:54Z

Looks like this helps, but not enough.

jkoritzinsky · 2024-04-16T22:34:44Z

/benchmark micro aspnet-perf-win runtime --variable filter=System.Reflection.Invoke.StaticMethod

pr-benchmarks · 2024-04-16T22:34:59Z

Benchmark started for micro on aspnet-perf-win with runtime and arguments --variable filter=*System.Reflection.Invoke.StaticMethod*. Logs: link

pr-benchmarks · 2024-04-16T23:11:55Z

micro - aspnet-perf-win

| benchmark | mean (micro.base) | mean (micro.pr) | ratio | allocated (micro.base) | allocated (micro.pr) | ratio || --------------------------------------------------------- | ----------------- | --------------- | ----- | ---------------------- | -------------------- | ----- || StaticMethod4_arrayNotCached_int_string_struct_class | 29.41 ns | 30.07 ns | 1.02 | 104 B | 104 B | 1.00 || StaticMethod4_ByRefParams_int_string_struct_class | 142.15 ns | 87.21 ns | 0.61 | 48 B | 48 B | 1.00 || StaticMethod4_int_string_struct_class | 20.62 ns | 19.83 ns | 0.96 | 0 B | 0 B ||| StaticMethod5_arrayNotCached_int_string_struct_class_bool | 42.85 ns | 43.83 ns | 1.02 | 136 B | 136 B | 1.00 || StaticMethod5_ByRefParams_int_string_struct_class_bool | 192.8 ns | 120.2 ns | 0.62 | 72 B | 72 B | 1.00 |

dotnet-policy-service · 2024-04-17T17:37:36Z

Tagging subscribers to this area: @mangod9
See info in area-owners.md if you want to be subscribed.

jkotas · 2024-04-19T03:27:18Z

 SetJitHelperFunction(CORINFO_HELP_NEWARR_1_OBJ, JIT_NewArr1OBJ_UP);

 ECall::DynamicallyAssignFCallImpl(GetEEFuncEntryPoint(AllocateStringFastUP), ECall::FastAllocateString);
+ ECall::DynamicallyAssignFCallImpl(GetEEFuncEntryPoint(JIT_BoxFastUP), ECall::NonNullableBox);


This block is Windows specific, so this change only helps on Windows. We need an equivalent fix for non-Windows platforms too.

Yeah I'll probably revert this particular change and make an equivalent one that's more xplat-friendly.

I've inlined the trail allocation logic into JIT_Box and factored out the rest of the logic to be in a separate framed helper instead of using the dynamic FCALL impl logic.
Let me know if this looks good to you.

jkoritzinsky · 2024-06-04T17:25:10Z

/benchmark micro aspnet-perf-win runtime --variable filter=System.Reflection.Invoke.StaticMethod

jkotas

Thanks!

jkoritzinsky · 2024-06-05T01:00:28Z

/ba-g timeouts unrelated

jkoritzinsky · 2024-04-16T19:30:23Z

Testing out fixes for #101134 and #104777

jkoritzinsky · 2024-04-16T19:30:34Z

/benchmark help

jkoritzinsky · 2024-04-16T21:02:27Z

/benchmark micro aspnet-perf-win runtime --variable filter=System.Reflection.Invoke.StaticMethod

pr-benchmarks · 2024-04-16T21:02:37Z

Benchmark started for micro on aspnet-perf-win with runtime and arguments --variable filter=*System.Reflection.Invoke.StaticMethod*. Logs: link

pr-benchmarks · 2024-04-16T21:34:25Z

micro - aspnet-perf-win

| benchmark | mean (micro.base) | mean (micro.pr) | ratio | allocated (micro.base) | allocated (micro.pr) | ratio || --------------------------------------------------------- | ----------------- | --------------- | ----- | ---------------------- | -------------------- | ----- || StaticMethod4_arrayNotCached_int_string_struct_class | 31.80 ns | 30.08 ns | 0.95 | 104 B | 104 B | 1.00 || StaticMethod4_ByRefParams_int_string_struct_class | 138.6 ns | 137.7 ns | 0.99 | 48 B | 48 B | 1.00 || StaticMethod4_int_string_struct_class | 19.52 ns | 19.80 ns | 1.01 | 0 B | 0 B ||| StaticMethod5_arrayNotCached_int_string_struct_class_bool | 45.05 ns | 43.55 ns | 0.97 | 136 B | 136 B | 1.00 || StaticMethod5_ByRefParams_int_string_struct_class_bool | 197.7 ns | 206.7 ns | 1.05 | 72 B | 72 B | 1.00 |

jkoritzinsky · 2024-04-16T21:56:54Z

Looks like this helps, but not enough.

jkoritzinsky · 2024-04-16T22:34:44Z

/benchmark micro aspnet-perf-win runtime --variable filter=System.Reflection.Invoke.StaticMethod

pr-benchmarks · 2024-04-16T22:34:59Z

Benchmark started for micro on aspnet-perf-win with runtime and arguments --variable filter=*System.Reflection.Invoke.StaticMethod*. Logs: link

pr-benchmarks · 2024-04-16T23:11:55Z

micro - aspnet-perf-win

| benchmark | mean (micro.base) | mean (micro.pr) | ratio | allocated (micro.base) | allocated (micro.pr) | ratio || --------------------------------------------------------- | ----------------- | --------------- | ----- | ---------------------- | -------------------- | ----- || StaticMethod4_arrayNotCached_int_string_struct_class | 29.41 ns | 30.07 ns | 1.02 | 104 B | 104 B | 1.00 || StaticMethod4_ByRefParams_int_string_struct_class | 142.15 ns | 87.21 ns | 0.61 | 48 B | 48 B | 1.00 || StaticMethod4_int_string_struct_class | 20.62 ns | 19.83 ns | 0.96 | 0 B | 0 B ||| StaticMethod5_arrayNotCached_int_string_struct_class_bool | 42.85 ns | 43.83 ns | 1.02 | 136 B | 136 B | 1.00 || StaticMethod5_ByRefParams_int_string_struct_class_bool | 192.8 ns | 120.2 ns | 0.62 | 72 B | 72 B | 1.00 |

dotnet-policy-service · 2024-04-17T17:37:36Z

Tagging subscribers to this area: @mangod9
See info in area-owners.md if you want to be subscribed.

jkotas · 2024-04-19T03:27:18Z

 SetJitHelperFunction(CORINFO_HELP_NEWARR_1_OBJ, JIT_NewArr1OBJ_UP);

 ECall::DynamicallyAssignFCallImpl(GetEEFuncEntryPoint(AllocateStringFastUP), ECall::FastAllocateString);
+ ECall::DynamicallyAssignFCallImpl(GetEEFuncEntryPoint(JIT_BoxFastUP), ECall::NonNullableBox);


This block is Windows specific, so this change only helps on Windows. We need an equivalent fix for non-Windows platforms too.

Yeah I'll probably revert this particular change and make an equivalent one that's more xplat-friendly.

I've inlined the trail allocation logic into JIT_Box and factored out the rest of the logic to be in a separate framed helper instead of using the dynamic FCALL impl logic.
Let me know if this looks good to you.

jkoritzinsky · 2024-06-04T17:25:10Z

/benchmark micro aspnet-perf-win runtime --variable filter=System.Reflection.Invoke.StaticMethod

jkotas

Thanks!

jkoritzinsky · 2024-06-05T01:00:28Z

/ba-g timeouts unrelated

jkoritzinsky · 2024-04-16T19:30:23Z

Testing out fixes for #101134 and #104777

jkoritzinsky · 2024-04-16T19:30:34Z

/benchmark help

jkoritzinsky · 2024-04-16T21:02:27Z

/benchmark micro aspnet-perf-win runtime --variable filter=System.Reflection.Invoke.StaticMethod

pr-benchmarks · 2024-04-16T21:02:37Z

Benchmark started for micro on aspnet-perf-win with runtime and arguments --variable filter=*System.Reflection.Invoke.StaticMethod*. Logs: link

pr-benchmarks · 2024-04-16T21:34:25Z

micro - aspnet-perf-win

| benchmark | mean (micro.base) | mean (micro.pr) | ratio | allocated (micro.base) | allocated (micro.pr) | ratio || --------------------------------------------------------- | ----------------- | --------------- | ----- | ---------------------- | -------------------- | ----- || StaticMethod4_arrayNotCached_int_string_struct_class | 31.80 ns | 30.08 ns | 0.95 | 104 B | 104 B | 1.00 || StaticMethod4_ByRefParams_int_string_struct_class | 138.6 ns | 137.7 ns | 0.99 | 48 B | 48 B | 1.00 || StaticMethod4_int_string_struct_class | 19.52 ns | 19.80 ns | 1.01 | 0 B | 0 B ||| StaticMethod5_arrayNotCached_int_string_struct_class_bool | 45.05 ns | 43.55 ns | 0.97 | 136 B | 136 B | 1.00 || StaticMethod5_ByRefParams_int_string_struct_class_bool | 197.7 ns | 206.7 ns | 1.05 | 72 B | 72 B | 1.00 |

jkoritzinsky · 2024-04-16T21:56:54Z

Looks like this helps, but not enough.

jkoritzinsky · 2024-04-16T22:34:44Z

/benchmark micro aspnet-perf-win runtime --variable filter=System.Reflection.Invoke.StaticMethod

pr-benchmarks · 2024-04-16T22:34:59Z

Benchmark started for micro on aspnet-perf-win with runtime and arguments --variable filter=*System.Reflection.Invoke.StaticMethod*. Logs: link

pr-benchmarks · 2024-04-16T23:11:55Z

micro - aspnet-perf-win

| benchmark | mean (micro.base) | mean (micro.pr) | ratio | allocated (micro.base) | allocated (micro.pr) | ratio || --------------------------------------------------------- | ----------------- | --------------- | ----- | ---------------------- | -------------------- | ----- || StaticMethod4_arrayNotCached_int_string_struct_class | 29.41 ns | 30.07 ns | 1.02 | 104 B | 104 B | 1.00 || StaticMethod4_ByRefParams_int_string_struct_class | 142.15 ns | 87.21 ns | 0.61 | 48 B | 48 B | 1.00 || StaticMethod4_int_string_struct_class | 20.62 ns | 19.83 ns | 0.96 | 0 B | 0 B ||| StaticMethod5_arrayNotCached_int_string_struct_class_bool | 42.85 ns | 43.83 ns | 1.02 | 136 B | 136 B | 1.00 || StaticMethod5_ByRefParams_int_string_struct_class_bool | 192.8 ns | 120.2 ns | 0.62 | 72 B | 72 B | 1.00 |

dotnet-policy-service · 2024-04-17T17:37:36Z

Tagging subscribers to this area: @mangod9
See info in area-owners.md if you want to be subscribed.

jkotas · 2024-04-19T03:27:18Z

 SetJitHelperFunction(CORINFO_HELP_NEWARR_1_OBJ, JIT_NewArr1OBJ_UP);

 ECall::DynamicallyAssignFCallImpl(GetEEFuncEntryPoint(AllocateStringFastUP), ECall::FastAllocateString);
+ ECall::DynamicallyAssignFCallImpl(GetEEFuncEntryPoint(JIT_BoxFastUP), ECall::NonNullableBox);


This block is Windows specific, so this change only helps on Windows. We need an equivalent fix for non-Windows platforms too.

Yeah I'll probably revert this particular change and make an equivalent one that's more xplat-friendly.

I've inlined the trail allocation logic into JIT_Box and factored out the rest of the logic to be in a separate framed helper instead of using the dynamic FCALL impl logic.
Let me know if this looks good to you.

jkoritzinsky · 2024-06-04T17:25:10Z

/benchmark micro aspnet-perf-win runtime --variable filter=System.Reflection.Invoke.StaticMethod

jkotas

Thanks!

jkoritzinsky · 2024-06-05T01:00:28Z

/ba-g timeouts unrelated

jkoritzinsky · 2024-04-16T19:30:23Z

Testing out fixes for #101134 and #104777

jkoritzinsky · 2024-04-16T19:30:34Z

/benchmark help

jkoritzinsky · 2024-04-16T21:02:27Z

/benchmark micro aspnet-perf-win runtime --variable filter=System.Reflection.Invoke.StaticMethod

pr-benchmarks · 2024-04-16T21:02:37Z

Benchmark started for micro on aspnet-perf-win with runtime and arguments --variable filter=*System.Reflection.Invoke.StaticMethod*. Logs: link

pr-benchmarks · 2024-04-16T21:34:25Z

micro - aspnet-perf-win

| benchmark | mean (micro.base) | mean (micro.pr) | ratio | allocated (micro.base) | allocated (micro.pr) | ratio || --------------------------------------------------------- | ----------------- | --------------- | ----- | ---------------------- | -------------------- | ----- || StaticMethod4_arrayNotCached_int_string_struct_class | 31.80 ns | 30.08 ns | 0.95 | 104 B | 104 B | 1.00 || StaticMethod4_ByRefParams_int_string_struct_class | 138.6 ns | 137.7 ns | 0.99 | 48 B | 48 B | 1.00 || StaticMethod4_int_string_struct_class | 19.52 ns | 19.80 ns | 1.01 | 0 B | 0 B ||| StaticMethod5_arrayNotCached_int_string_struct_class_bool | 45.05 ns | 43.55 ns | 0.97 | 136 B | 136 B | 1.00 || StaticMethod5_ByRefParams_int_string_struct_class_bool | 197.7 ns | 206.7 ns | 1.05 | 72 B | 72 B | 1.00 |

jkoritzinsky · 2024-04-16T21:56:54Z

Looks like this helps, but not enough.

jkoritzinsky · 2024-04-16T22:34:44Z

/benchmark micro aspnet-perf-win runtime --variable filter=System.Reflection.Invoke.StaticMethod

pr-benchmarks · 2024-04-16T22:34:59Z

Benchmark started for micro on aspnet-perf-win with runtime and arguments --variable filter=*System.Reflection.Invoke.StaticMethod*. Logs: link

pr-benchmarks · 2024-04-16T23:11:55Z

micro - aspnet-perf-win

| benchmark | mean (micro.base) | mean (micro.pr) | ratio | allocated (micro.base) | allocated (micro.pr) | ratio || --------------------------------------------------------- | ----------------- | --------------- | ----- | ---------------------- | -------------------- | ----- || StaticMethod4_arrayNotCached_int_string_struct_class | 29.41 ns | 30.07 ns | 1.02 | 104 B | 104 B | 1.00 || StaticMethod4_ByRefParams_int_string_struct_class | 142.15 ns | 87.21 ns | 0.61 | 48 B | 48 B | 1.00 || StaticMethod4_int_string_struct_class | 20.62 ns | 19.83 ns | 0.96 | 0 B | 0 B ||| StaticMethod5_arrayNotCached_int_string_struct_class_bool | 42.85 ns | 43.83 ns | 1.02 | 136 B | 136 B | 1.00 || StaticMethod5_ByRefParams_int_string_struct_class_bool | 192.8 ns | 120.2 ns | 0.62 | 72 B | 72 B | 1.00 |

dotnet-policy-service · 2024-04-17T17:37:36Z

Tagging subscribers to this area: @mangod9
See info in area-owners.md if you want to be subscribed.

jkotas · 2024-04-19T03:27:18Z

 SetJitHelperFunction(CORINFO_HELP_NEWARR_1_OBJ, JIT_NewArr1OBJ_UP);

 ECall::DynamicallyAssignFCallImpl(GetEEFuncEntryPoint(AllocateStringFastUP), ECall::FastAllocateString);
+ ECall::DynamicallyAssignFCallImpl(GetEEFuncEntryPoint(JIT_BoxFastUP), ECall::NonNullableBox);


This block is Windows specific, so this change only helps on Windows. We need an equivalent fix for non-Windows platforms too.

Yeah I'll probably revert this particular change and make an equivalent one that's more xplat-friendly.

I've inlined the trail allocation logic into JIT_Box and factored out the rest of the logic to be in a separate framed helper instead of using the dynamic FCALL impl logic.
Let me know if this looks good to you.

jkoritzinsky · 2024-06-04T17:25:10Z

/benchmark micro aspnet-perf-win runtime --variable filter=System.Reflection.Invoke.StaticMethod

jkotas

Thanks!

jkoritzinsky · 2024-06-05T01:00:28Z

/ba-g timeouts unrelated

ghost added the needs-area-label An area label is needed to ensure this gets routed to the appropriate area owners label Apr 16, 2024

dotnet-policy-serviceBot assigned jkoritzinsky Apr 16, 2024

This comment was marked as outdated.
Sign in to view

This was referenced Apr 16, 2024
System.Text.Json failing some large file tests #59678
Closed
CI error: System.Net.Quic.QuicException: The connection timed out from inactivity #91757
Closed
[wasm] System.Text.Json.Tests running out of memory #98578
Closed

This was referenced Apr 17, 2024
GC\Regressions\v2.0-beta2\452950\452950\452950.cmd failing on Mono minijit Windows x64 #99729
Closed
at GC_Regressions._v2_0_beta2_452950_452950_452950_._v2_0_beta2_452950_452950_452950_cmd() #100174
Closed

teo-tsirpanis added area-VM-coreclr and removed needs-area-label An area label is needed to ensure this gets routed to the appropriate area owners labels Apr 17, 2024

This was referenced Apr 17, 2024
Timeout in System.Text.Json.Tests.Utf8JsonWriterTests.WriteNumbers #101193
Closed
System.IO.Net5Compat.Tests and System.IO.Tests suddenly exiting with error 137 #100558
Closed
Debug assertion failure in System.Net.Http.HttpConnection.SendAsync #100616
Closed

jkotas reviewed Apr 19, 2024
View reviewed changes

jkotas mentioned this pull request Apr 26, 2024
Try to reduce cost of Async #101605
Merged

jkotas reviewed Apr 29, 2024
View reviewed changes

Comment threadsrc/coreclr/vm/jithelpers.cpp Outdated

build-analysisBot mentioned this pull request May 2, 2024
[wasm][AOT][net8.0] emcc : error - received SIGKILL (-9) #89402
Open

jkoritzinsky added 2 commits May 31, 2024 10:23

Avoid stack spill for Mono RuntimeHelpers.Box
d5cdf17

Rewrite RuntimeHelpers.Box to insert into the generic cache.
a32aa66

jkoritzinsky force-pushed the allocate-vt-regression branch from db2d27c to a32aa66Compare May 31, 2024 20:28

jkotas reviewed May 31, 2024
View reviewed changes

Comment threadsrc/coreclr/System.Private.CoreLib/src/System/RuntimeType.BoxCache.cs Outdated

jkotas reviewed May 31, 2024
View reviewed changes

Comment threadsrc/coreclr/System.Private.CoreLib/src/System/RuntimeType.BoxCache.cs

build-analysisBot mentioned this pull request Jun 1, 2024
NativeAOT legs timing out in CI #102239
Closed

jkoritzinsky added 2 commits June 3, 2024 13:55

Remove duplicate checks, fix ordering, add assert.
20415a4

v -> V
b3ebd7b

jkoritzinsky marked this pull request as ready for review June 4, 2024 18:34

jkoritzinsky requested review from lambdageek, marek-safar and thaystg as code owners June 4, 2024 18:34

jkoritzinsky requested a review from jkotas June 4, 2024 18:34

jkotas approved these changes Jun 4, 2024
View reviewed changes

jkoritzinsky merged commit 1753956 into dotnet:mainJun 5, 2024

jkoritzinsky deleted the allocate-vt-regression branch June 5, 2024 01:00

AndyAyersMS mentioned this pull request Jun 11, 2024
[Perf] Linux/x64: 3 Improvements on 6/5/2024 1:00:52 AM dotnet/perf-autofiling-issues#36057
Closed

elinor-fung mentioned this pull request Jul 2, 2024
.NET 9 Arm32 tests are randomly failing from PowerShell hang #104106
Closed

github-actionsBot locked and limited conversation to collaborators Jul 5, 2024