Code

Ifu 2024 02 20#56

Code

Merged

liligwu merged 36 commits intomainROCm/FBGEMM:mainfrom IFU-2024-02-20ROCm/FBGEMM:IFU-2024-02-20Copy head branch name to clipboardFeb 20, 2024

+3,835-1,307

Commits

Commits on Feb 1, 2024

• Add int32 overflow unit test for TBE UVM caching (pytorch#2303)

  Show description for b347a43
  

  sryap

  authored and

  facebook-github-bot

  committed
  b347a43

  Copy full SHA for b347a43

• Refactor generate_cache_tbes in unit test (pytorch#2304)

  Show description for 79f4862
  

  sryap

  authored and

  facebook-github-bot

  committed
  79f4862

  Copy full SHA for 79f4862

Commits on Feb 2, 2024

• Re-organize TBE tests, pt 5 (pytorch#2305)

  Show description for c218dc3
  

  q10

  authored and

  facebook-github-bot

  committed
  c218dc3

  Copy full SHA for c218dc3

• Add a new CPU nbit-TBE benchmark that tries to reduce CPU frequency noise (pytorch#2306)

  Show description for aabb4ae
  

  excelle08

  authored and

  facebook-github-bot

  committed
  aabb4ae

  Copy full SHA for aabb4ae

Commits on Feb 4, 2024

• Add early exit to sparse_segment_sum_csr_cuda op (pytorch#2277)

  Show description for dad9720
  

  Mark Eremeev

  authored and

  facebook-github-bot

  committed
  dad9720

  Copy full SHA for dad9720

Commits on Feb 5, 2024

• Flag to disable uvm caching for pt2 export (pytorch#2308)

  Show description for 7889f64
  

  PaulZhang12

  authored and

  facebook-github-bot

  committed
  7889f64

  Copy full SHA for 7889f64

Commits on Feb 6, 2024

• Suppress infinite-recursion warning (pytorch#2307)

  Show description for 86acdbd
  

  q10

  authored and

  facebook-github-bot

  committed
  86acdbd

  Copy full SHA for 86acdbd

• Skip test_backward_dense (pytorch#2311)

  Show description for 9538d62
  

  spcyppt

  authored and

  facebook-github-bot

  committed
  9538d62

  Copy full SHA for 9538d62

• Optimize the cache fetch for forward split, pt. 2B (pytorch#2282)

  Show description for c8cb1a2
  

  q10

  authored and

  facebook-github-bot

  committed
  c8cb1a2

  Copy full SHA for c8cb1a2

Commits on Feb 7, 2024

• Copy from/to cache directly if cache and emb are the same type (pytorch#2309)

  Show description for 6f5fa54
  

  sryap

  authored and

  facebook-github-bot

  committed
  6f5fa54

  Copy full SHA for 6f5fa54

• add dynamic quantize gemm benchmark [step 2: fp16->int8 quantize] (pytorch#2295)

  Show description for 44fc10a
  

  Jiyuan Zhang

  authored and

  facebook-github-bot

  committed
  44fc10a

  Copy full SHA for 44fc10a

• Fix OSS ufmt lint issues (pytorch#2313)

  Show description for c466746
  

  q10

  authored and

  facebook-github-bot

  committed
  c466746

  Copy full SHA for c466746

• Add BF16 support in group_index_select_2d (pytorch#2317)

  Show description for 89a4913
  

  sryap

  authored and

  facebook-github-bot

  committed
  89a4913

  Copy full SHA for 89a4913

• int8 output for seq embeddings (pytorch#2316)

  Show description for af41af1
  

  YazhiGao

  authored and

  facebook-github-bot

  committed
  af41af1

  Copy full SHA for af41af1

• remove d2h sync from keyed jagged index select (pytorch#2315)

  Show description for e80d507
  

  joshuadeng

  authored and

  facebook-github-bot

  committed
  e80d507

  Copy full SHA for e80d507

• add dynamic quantize gemm benchmark [step 1: minmax qparams compute] (pytorch#2297)

  Show description for 3687061
  

  Jiyuan Zhang

  authored and

  facebook-github-bot

  committed
  3687061

  Copy full SHA for 3687061

• Optimize the cache fetch for forward split, pt. 3 (pytorch#2216)

  Show description for 348d439
  

  q10

  authored and

  facebook-github-bot

  committed
  348d439

  Copy full SHA for 348d439

Commits on Feb 8, 2024

• ROCm fix (pytorch#2320)

  Show description for a816f8c
  

  q10

  authored and

  facebook-github-bot

  committed
  a816f8c

  Copy full SHA for a816f8c

• Fix BF16 group_index_select_2d on AMD GPU (pytorch#2321)

  Show description for 86ea895
  

  zoranzhao

  authored and

  facebook-github-bot

  committed
  86ea895

  Copy full SHA for 86ea895

Commits on Feb 9, 2024

• Refactor stochastic rounding state in TBE (pytorch#2312)

  Show description for 8664c84
  

  sryap

  authored and

  facebook-github-bot

  committed
  8664c84

  Copy full SHA for 8664c84

• Enable compilation by clang (pytorch#2323)

  Show description for 533c4bb
  

  q10

  authored and

  facebook-github-bot

  committed
  533c4bb

  Copy full SHA for 533c4bb

• Add unit test for stochastic rounding for UVM caching (pytorch#2324)

  Show description for bc85d73
  

  sryap

  authored and

  facebook-github-bot

  committed
  bc85d73

  Copy full SHA for bc85d73

• Back out "Add BF16 support in group_index_select_2d" (pytorch#2326)

  Show description for eb3c304
  

  sryap

  authored and

  facebook-github-bot

  committed
  eb3c304

  Copy full SHA for eb3c304

Commits on Feb 13, 2024

• PT2 compliant - fbgemm::jagged_dense_elementwise_add_jagged_output (pytorch#2327)

  Show description for 0c43bee
  

  davidberard98

  authored and

  facebook-github-bot

  committed
  0c43bee

  Copy full SHA for 0c43bee

• More flexible test retries (pytorch#2328)

  Show description for 2c519be
  

  q10

  authored and

  facebook-github-bot

  committed
  2c519be

  Copy full SHA for 2c519be

• scale_bias first, 2byte quant padding dequant support

  Show description for 5ca556a
  

  YazhiGao

  authored and

  facebook-github-bot

  committed
  5ca556a

  Copy full SHA for 5ca556a

Commits on Feb 15, 2024

• Enable Clang compilation in OSS for fbgemm_gpu (CPU) (pytorch#2330)

  Show description for d00b5d5
  

  q10

  authored and

  facebook-github-bot

  committed
  d00b5d5

  Copy full SHA for d00b5d5

Commits on Feb 16, 2024

• Remove double type support from fbgemm_cuda_utils.cuh (pytorch#2335)

  Show description for 66a8779
  

  sryap

  authored and

  facebook-github-bot

  committed
  66a8779

  Copy full SHA for 66a8779

Commits on Feb 17, 2024

• Enable -Winfinite-recursion in deeplearning/PACKAGE (pytorch#2329)

  Show description for 89c52b0
  

  r-barnes

  authored and

  facebook-github-bot

  committed
  89c52b0

  Copy full SHA for 89c52b0

Commits on Feb 18, 2024

• Remove semicolon(s) from 3 files inc deeplearning/fbgemm/fb/FbgemmFP32.h (pytorch#2339)

  Show description for 13d6bea
  

  r-barnes

  authored and

  facebook-github-bot

  committed
  13d6bea

  Copy full SHA for 13d6bea

Commits on Feb 19, 2024

• remove symint from tbe_input_combine_with_length_abstract (pytorch#2336)

  Show description for bc4a73e
  

  bradleyhd

  authored and

  facebook-github-bot

  committed
  bc4a73e

  Copy full SHA for bc4a73e

Commits on Feb 20, 2024

• Enable Clang compilation in OSS for fbgemm_gpu (CUDA) (pytorch#2334)

  Show description for 935b075
  

  q10

  authored and

  facebook-github-bot

  committed
  935b075

  Copy full SHA for 935b075

• Improve perf of kernel: _block_bucketize_sparse_features_cuda_kernel1 (pytorch#2331)

  Show description for bbf83f1
  

  AlbertDachiChen

  authored and

  facebook-github-bot

  committed
  bbf83f1

  Copy full SHA for bbf83f1

• Use gpu_library_selector for permute_pooled_embedding_ops_gpu (pytorch#2340)

  Show description for 49cca04
  

  jianyuh

  authored and

  facebook-github-bot

  committed
  49cca04

  Copy full SHA for 49cca04

• Improve logging message for uvm caching (pytorch#2337)

  Show description for b4afb6d
  

  henrylhtsang

  authored and

  facebook-github-bot

  committed
  b4afb6d

  Copy full SHA for b4afb6d

• Merge remote-tracking branch 'upstream/main' into IFU-2024-02-20

  

  root

  committed
  360e3f1

  Copy full SHA for 360e3f1