TensorComprehensions

TensorComprehensions copied to clipboard

daniel-s-ingram

Convolution Backward Pass with kernel_size > 1

5

Hi, I'm currently trying to implement a custom convolution operation (foward and backward passes). I am starting from the code provided [here](https://facebookresearch.github.io/TensorComprehensions/_sources/framework/pytorch_integration/autograd_with_tc.rst.txt) I am getting an error with the backward...

pclucas14

Update Conda package?

1

First, let me say I love, love, love this project. It makes creating fast kernels a real breeze, unlike anything else I've seen. I am running into a bug (I...

larspars

use templated isl types

2

Templated isl types keep track of the number and nature of tuples in the spaces of an isl object as well as (optionally) their internal structure. This allows the compiler...

skimo-openhub

Autotuning improvements

1

This PR isolates a small subset from @ttheodor's #453. I took the commits that look independent, fix clear bugs or implement simple new behavior and do not touch autotuner-inl.h (which...

nicolasvasilache

Support -mpcu flag and autodetect to trigger vectorization on proper vector lengths

1

This PR passes proper llvm::TargetMachine information in llvm_jit and codegen_llvm by introducing a proper TargetMachine at the LLVMJit level and avoids introducing adhoc objects. The TargetMachine is constructed either from...

nicolasvasilache

Cuda Performance Metrics/Profiling

14

Add CUPTI-based profiling functionality in `CudaRTCFun`. There are several performance metrics (listed [here](http://docs.nvidia.com/cuda/cupti/r_main.html#metrics-reference). Each metric requires measuring (possibly multiple types of) hardware events. Since not all events can be measured...

thetheodor

[Build] issues finding Cuda, incomplete config

15

Hi, I want to build TC from source on a cluster (https://www.macs.hw.ac.uk/~hv15/robotarium/about) and run some of the benchmarks, here's the information regarding my working environment: - OS: Scientific Linux release...

keightyfive

Register promotion: limit the number of elements promoted per thread

1

cuda::MappedScop: introduce maxPrivateElements mapping option This mapping option controls the maximum number of elements per thread that are promoted into the private memory (hopefully, registers, but we cannot guarantee this...

ftynse

Allow computed expressions on the left-hand-side

11

abadams