Christian Sarofeen

Cooperative groups support.

2

Remove whitespace on empty lines. Add support for cooperative groups.

Couple quick changes that may be needed

Work is getting pretty deep, very high level changes this is going after: A lot of work going into separating `IterDomainGraph` from `ComputeAtMap` cleanly converting usage of `BestEffortReplay` to `IterDomainGraph::mapBetween`...

[Repro] View persistent issue push

1

I'm not certain what's going on yet, but... 1) projection of persistent buffer is coming up with a buffer size of 0 (non projected is 3,780). 2) I'm getting an...

Data Swizzle Inlining - Replay consistency

1

### 🐛 Describe the bug I don't understand the following test: https://github.com/csarofeen/pytorch/blob/devel/torch/csrc/jit/codegen/cuda/test/test_gpu_swizzle.cpp#L58-L96 We have straight producer-consumer relationships between: ``` T1_l[ iS2{2}, iS18{( ceilDiv(32, 16) )}, iS20{( ceilDiv(16, 8) )}, iS21{8}...

Segfault with bad schedule

1

### 🐛 Describe the bug I forgot to add an output to a schedule and it resulted in a segfault within computeAt. Uncertain where it came from as I didn't...

Expression Evaluator exposure

1

### 🚀 The feature, motivation and pitch It would be good to make sure we have strong exposure of our expression evaluator. Going to note some interfaces we would likely...

### 🐛 Describe the bug I changed: https://github.com/csarofeen/pytorch/blob/devel/torch/csrc/jit/codegen/cuda/scheduler/pointwise.cpp#L246 from: ``` for (const auto break_point_i : c10::irange(ref_root.size())) { ``` to: ``` for (const auto break_point_i : c10::irange(ref_root.size()-1)) { ``` Because there...

### 🐛 Describe the bug The repro in https://github.com/csarofeen/pytorch/issues/2080 seems to be getting segmented into two kernels because of a trivial reduction pattern being flagged as an issue. Need to...