22 Pull requests merged by 15 people

• Update 35 Equality logic yaml sizes.

  #1378 merged Nov 23, 2024

• update 38 Equality logic yaml sizes

  #1372 merged Nov 23, 2024

• gfx942 38cu F8BS NN TN NT grid tune

  #1345 merged Nov 22, 2024

• gfx942 38cu HHS BBS NN TN NT grid tune

  #1331 merged Nov 22, 2024

• Fix invalid stream-k test case, make dynamic grid the default

  #1359 merged Nov 22, 2024

• update gfx942 xf32 freesize

  #1375 merged Nov 22, 2024

• Add gfx942 xf32 NN/NT/TN Equality yamls for 1105 xf32

  #1376 merged Nov 22, 2024

• Fix: incorrect required workspace size for singleKernel GSU

  #1371 merged Nov 22, 2024

• gridbased search for batched gemm

  #1362 merged Nov 22, 2024

• Add profiling to TensileCreateLibrary

  #1329 merged Nov 22, 2024

• Set Python_ROOT virtual.env

  #1344 merged Nov 21, 2024

• Fixing and adding test for DepthU=48

  #1303 merged Nov 20, 2024

• [Hotfix] Disable setOccupancyLimit for gfx120X

  #1368 merged Nov 20, 2024

• Remove alias for MirrorDims in logic yaml

  #1361 merged Nov 20, 2024

• GFX942 equality tuning for F8HS and F8B8HS for TN,NT,NN

  #1302 merged Nov 19, 2024

• Add setOccupancyLimit

  #1364 merged Nov 19, 2024

• Remove Min/Max/TotalVgprNumber in Common.py

  #1355 merged Nov 19, 2024

• gfx12 - change to use byte_sel modifier for v_cvt_f32_fp8 and v_cvt_f…

  #1172 merged Nov 19, 2024

• Add profile logging and standardize scaleA and scaleB datatypes

  #1275 merged Nov 18, 2024

• Fix F8/BF8 failed cases for GWVW=8 and Beta != 0

  #1333 merged Nov 18, 2024

• adding bpl64 support to addLdsLoad (for Bias and scaleAlphaVector)

  #1336 merged Nov 18, 2024

• Add sgpr occupancy

  #1349 merged Nov 18, 2024

14 Pull requests opened by 13 people

• Modify to check if alpha is in host memory.

  #1356 opened Nov 18, 2024

• Refactoy the pack scheduling for scheduleIterAlg = 3.

  #1358 opened Nov 18, 2024

• Fix F32 FMAC Perf Bugs for gfx11/12

  #1360 opened Nov 19, 2024

• Bump rocm-docs-core from 1.8.3 to 1.8.5 in /docs/sphinx

  #1363 opened Nov 19, 2024

• Library Logic Format Simplification

  #1365 opened Nov 19, 2024

• Remove PackageLibrary option

  #1367 opened Nov 20, 2024

• [Sparse] fix sparse kernel generation failure

  #1369 opened Nov 20, 2024

• Update 12 Equality logic yamls.

  #1370 opened Nov 20, 2024

• Avoid divide by 0 when calculating predicted performance with streamk

  #1373 opened Nov 21, 2024

• Code object compression via bundling

  #1374 opened Nov 22, 2024

• [Experimental] hipBLASLt tensor swizzling integration

  #1377 opened Nov 22, 2024

• Fp8 tuning upstream

  #1380 opened Nov 22, 2024

• Find python

  #1381 opened Nov 22, 2024

• Logic fix to exclude streamk by default

  #1382 opened Nov 22, 2024

14 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• Adding HostLibraryTests back to hipBLASLt

  #1147 commented on Nov 20, 2024 • 10 new comments

• Add support for fallback from compute type f16 to f32

  #1263 commented on Nov 19, 2024 • 8 new comments

• feature: DTV with Swizzling (tensorA)

  #1246 commented on Nov 22, 2024 • 1 new comment

• Enable variant builds via device ID and cu count

  #1222 commented on Nov 18, 2024 • 0 new comments

• dot2 fp16 mac kernel for gfx942

  #1258 commented on Nov 23, 2024 • 0 new comments

• Tune Aquavanjaram942X F8F8S Equality TN 1 GEMM size

  #1270 commented on Nov 20, 2024 • 0 new comments

• Static build

  #1283 commented on Nov 23, 2024 • 0 new comments

• Check arguments in yaml file, abort if not recognized.

  #1294 commented on Nov 19, 2024 • 0 new comments

• Tune Aldebaran BF16 NN TN NT GEMM sizes

  #1323 commented on Nov 21, 2024 • 0 new comments

• Add --experimental flag to TensileCreateLibrary

  #1328 commented on Nov 22, 2024 • 0 new comments

• Tune Aquavanjaram 942 20CU HHS NN and TN GEMM sizes tuning in equality library

  #1330 commented on Nov 21, 2024 • 0 new comments

• [CQE only] gfx12 - change to use byte_sel modifier for v_cvt_f32_fp8 and v_cvt_f…

  #1339 commented on Nov 18, 2024 • 0 new comments

• Add initial optional stream-k libraries

  #1347 commented on Nov 21, 2024 • 0 new comments

• [OPT] Optimize tail loop

  #1353 commented on Nov 22, 2024 • 0 new comments