FBGEMM

FBGEMM copied to clipboard

Deepcopy to avoid errors from inplace operators

5

Differential Revision: D38716708

jiecaoyu

Add Mx2, Mx4, 2xN, and 4xN avx512 transpose

6

Add Mx2, Mx4, 2xN, and 4xN specific transposes on avx512 to improve the transpose performance of shapes of Mx2, Mx4, 2xN, and 4xN. * When the shape is Mx2 or...

CaoE

Clang 14 Build Failure

18

Building FBGEMM with clang 14 fails. The errors look like this: `In file included from /home/juanpaez/FBGEMM/src/EmbeddingSpMDM.cc:11: In file included from /home/juanpaez/FBGEMM/third_party/asmjit/src/asmjit/asmjit.h:27: In file included from /home/juanpaez/FBGEMM/third_party/asmjit/src/asmjit/./core.h:2009: In file included from...

juanpaez22

Change the hardcoded 128 items per warp in embeddingBag to a variable and optimize for ROCm

9

Make @weihanmines's PR https://github.com/ROCmSoftwarePlatform/FBGEMM/pull/13 upstreamable. @sryap, would you please review the PR and consider converting it to a draft? Thank you.

liligwu

Adding warmup option to device benchmark

2

Summary: In the current code, we generate a different input batch for each iteration. For training benchmarks, we use global batch size of >= 64K. So, it becomes expensive (in...

mustafaozdal

FBGEMM_gpu support for CUDA 11.5

10

We are trying to compile FBGEMM_gpu from source and are running into some errors during the build process that we suspect are related to an unsupported CUDA version. Please let...

chantat

error: comparison of integer expressions of different signedness

2

In the QuantUtilsTest.cc on line 641 I am getting the error "error: comparison of integer expressions of different signedness: ‘std::vector::size_type’ {aka ‘long unsigned int’} and ‘int’ [-Werror=sign-compare] [build] 641 |...

dkbattle

Conanfile.py??

1

Does anyone have a conanfile.py created for this dependency?

dkbattle

Add prototype for optimized dense2jagged kernel for 1 jagged dim

2

Sharing optimized prototype kernel for 2D dense_to_jagged operations using vectorized operations. CC: @mjanderson09

chirayuG-nvidia

Enable uvm using p2p (UVM on a pair GPU)

4

Summary: Add the APIs for using UVM where the preferred location is on GPU device instead of on CPU device. Differential Revision: D36657705

jianyuh