Code

ROCm Port#1087

Code

Merged

SlyEcho merged 105 commits intoggerganov:masterggerganov/llama.cpp:masterfrom SlyEcho:hipblasSlyEcho/llama.cpp:hipblasCopy head branch name to clipboardAug 25, 2023

+335-59

Commits

Commits on Apr 20, 2023

• use hipblas based on cublas

  SlyEcho

  committed
  0fd8363

  Copy full SHA for 0fd8363

• Update Makefile for the Cuda kernels

  SlyEcho

  committed
  54a63c1

  Copy full SHA for 54a63c1

• Build file changes

  Show description for 0e005f7
  SlyEcho

  committed
  0e005f7

  Copy full SHA for 0e005f7

Commits on Apr 21, 2023

• add rpath

  SlyEcho

  committed
  d3e1984

  Copy full SHA for d3e1984

Commits on Apr 22, 2023

• More build file changes

  SlyEcho

  committed
  3677235

  Copy full SHA for 3677235

Commits on Apr 23, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  db7a012

  Copy full SHA for db7a012

• add rpath

  SlyEcho

  committed
  3a004b2

  Copy full SHA for 3a004b2

Commits on Apr 25, 2023

• change default GPU arch to match CMake

  SlyEcho

  committed
  608aa33

  Copy full SHA for 608aa33

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  d571d16

  Copy full SHA for d571d16

Commits on Apr 26, 2023

• Merge branch 'ggerganov:master' into hipblas

  SlyEcho

  authored
  ef51e9e

  Copy full SHA for ef51e9e

Commits on Apr 27, 2023

• only .cu file needs to be complied as device

  SlyEcho

  committed
  ecc0565

  Copy full SHA for ecc0565

Commits on Apr 28, 2023

• add more cuda defines

  Show description for a1caa48
  SlyEcho

  committed
  a1caa48

  Copy full SHA for a1caa48

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  3b4a531

  Copy full SHA for 3b4a531

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  2ab9d11

  Copy full SHA for 2ab9d11

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  d194586

  Copy full SHA for d194586

Commits on Apr 29, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  d8ea75e

  Copy full SHA for d8ea75e

Commits on Apr 30, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  c73def1

  Copy full SHA for c73def1

Commits on May 1, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  fcbc262

  Copy full SHA for fcbc262

Commits on May 3, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  b67cc50

  Copy full SHA for b67cc50

Commits on May 4, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  d83cfba

  Copy full SHA for d83cfba

• Move all HIP stuff to ggml-cuda.cu

  SlyEcho

  committed
  04c0d48

  Copy full SHA for 04c0d48

Commits on May 5, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  1107194

  Copy full SHA for 1107194

Commits on May 6, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  289073a

  Copy full SHA for 289073a

Commits on May 7, 2023

• Revert to default copy

  SlyEcho

  committed
  baeb482

  Copy full SHA for baeb482

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  0aefa6a

  Copy full SHA for 0aefa6a

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  a3296d5

  Copy full SHA for a3296d5

• occupanct function

  SlyEcho

  committed
  070cbcc

  Copy full SHA for 070cbcc

Commits on May 11, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  127f68e

  Copy full SHA for 127f68e

Commits on May 12, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  605560d

  Copy full SHA for 605560d

• fix makefile

  SlyEcho

  committed
  0fe6384

  Copy full SHA for 0fe6384

Commits on May 13, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  2956630

  Copy full SHA for 2956630

Commits on May 14, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  8bab456

  Copy full SHA for 8bab456

Commits on May 16, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  a0b2d5f

  Copy full SHA for a0b2d5f

Commits on May 20, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  c66115b

  Copy full SHA for c66115b

• Forwardcompat

  SlyEcho

  committed
  b19fefe

  Copy full SHA for b19fefe

• update warp size

  SlyEcho

  committed
  600ace3

  Copy full SHA for 600ace3

Commits on May 24, 2023

• Merge branch 'origin/master' into hipblas

  SlyEcho

  committed
  f80ce7a

  Copy full SHA for f80ce7a

Commits on May 25, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  174bf6a

  Copy full SHA for 174bf6a

• Add missing parameters

  SlyEcho

  committed
  a593a4f

  Copy full SHA for a593a4f

• and makefile

  SlyEcho

  committed
  30d921a

  Copy full SHA for 30d921a

• add configurable vars

  SlyEcho

  committed
  4c8b3fb

  Copy full SHA for 4c8b3fb

Commits on May 27, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  a4648c1

  Copy full SHA for a4648c1

• Add more defs

  Show description for 9fdaa1d
  SlyEcho

  committed
  9fdaa1d

  Copy full SHA for 9fdaa1d

Commits on Jun 6, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  33091a9

  Copy full SHA for 33091a9

• warp size fixes

  SlyEcho

  committed
  5d6eb72

  Copy full SHA for 5d6eb72

• Revert "warp size fixes"

  Show description for 1ba4ce4
  SlyEcho

  committed
  1ba4ce4

  Copy full SHA for 1ba4ce4

• fix makefile.

  SlyEcho

  committed
  fa5b3d7

  Copy full SHA for fa5b3d7

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  4362e80

  Copy full SHA for 4362e80

Commits on Jun 8, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  85f902d

  Copy full SHA for 85f902d

Commits on Jun 14, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  a836529

  Copy full SHA for a836529

• add cudaMemset

  SlyEcho

  committed
  61df8e9

  Copy full SHA for 61df8e9

Commits on Jun 17, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  6f7c156

  Copy full SHA for 6f7c156

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  67e229b

  Copy full SHA for 67e229b

Commits on Jun 19, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  5dd2fbe

  Copy full SHA for 5dd2fbe

Commits on Jun 22, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  df7346c

  Copy full SHA for df7346c

Commits on Jun 25, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  35a6031

  Copy full SHA for 35a6031

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  c1e5c83

  Copy full SHA for c1e5c83

Commits on Jun 27, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  c8ae945

  Copy full SHA for c8ae945

Commits on Jun 28, 2023

• headers fix; add kquants_iter for hipblas and add gfx803 (#1)

  Show description for bb16eff
  YellowRoseCx

  authored
  bb16eff

  Copy full SHA for bb16eff

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  04419f1

  Copy full SHA for 04419f1

Commits on Jul 2, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  15db19a

  Copy full SHA for 15db19a

• ROCm fixes

  SlyEcho

  committed
  c3e3733

  Copy full SHA for c3e3733

Commits on Jul 4, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  7735c5a

  Copy full SHA for 7735c5a

Commits on Jul 9, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  80e4e54

  Copy full SHA for 80e4e54

Commits on Jul 11, 2023

• Expand arch list and make it overrideable

  SlyEcho

  committed
  e610466

  Copy full SHA for e610466

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  8c2c497

  Copy full SHA for 8c2c497

• Add new config option

  SlyEcho

  committed
  afcb8fe

  Copy full SHA for afcb8fe

Commits on Jul 13, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  cd36b18

  Copy full SHA for cd36b18

• Update build flags.

  Show description for 2ec4466
  

  SlyEcho

  and

  YellowRoseCx

  committed
  2ec4466

  Copy full SHA for 2ec4466

Commits on Jul 17, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  3db70b5

  Copy full SHA for 3db70b5

Commits on Jul 24, 2023

• Fix multi GPU on multiple amd architectures with rocblas_initialize() (#5)

  Show description for 1f6294d
  YellowRoseCx

  authored
  1f6294d

  Copy full SHA for 1f6294d

• Add rocblas to build files

  SlyEcho

  committed
  8e8054a

  Copy full SHA for 8e8054a

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  cde52d6

  Copy full SHA for cde52d6

Commits on Jul 29, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  d2ade63

  Copy full SHA for d2ade63

• rocblas init stuff

  SlyEcho

  committed
  f8e3fc6

  Copy full SHA for f8e3fc6

• add hipBLAS to README

  Show description for 4336231
  

  SlyEcho

  and

  ardfork

  committed
  4336231

  Copy full SHA for 4336231

Commits on Jul 31, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  c1664a0

  Copy full SHA for c1664a0

• new build arg LLAMA_CUDA_MMQ_Y

  SlyEcho

  committed
  c1cb70d

  Copy full SHA for c1cb70d

• fix half2 decomposition

  

  ardfork

  authored and

  SlyEcho

  committed
  d91456a

  Copy full SHA for d91456a

Commits on Aug 8, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  ab62128

  Copy full SHA for ab62128

• Add intrinsics polyfills for AMD

  Show description for 4024f91
  

  4 peoplecommitted
  4024f91

  Copy full SHA for 4024f91

Commits on Aug 9, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  610ba4c

  Copy full SHA for 610ba4c

• hipLDFLAG Path change Unix to multisystem in Makefile

  Show description for 8f8ab6c
  YellowRoseCx

  authored
  8f8ab6c

  Copy full SHA for 8f8ab6c

Commits on Aug 10, 2023

• Fix merge

  Show description for 29a59b5
  

  3 peoplecommitted
  29a59b5

  Copy full SHA for 29a59b5

• AMD assembly optimized __dp4a

  Show description for f41920e
  

  Engininja2

  authored and

  SlyEcho

  committed
  f41920e

  Copy full SHA for f41920e

• ws fix

  SlyEcho

  committed
  42e055d

  Copy full SHA for 42e055d

Commits on Aug 11, 2023

• Undo mess

  Show description for e6b6ae5
  

  SlyEcho

  and

  ardfork

  committed
  e6b6ae5

  Copy full SHA for e6b6ae5

• New __dp4a assembly

  Show description for c299c4a
  

  Engininja2

  authored and

  SlyEcho

  committed
  c299c4a

  Copy full SHA for c299c4a

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  b815e97

  Copy full SHA for b815e97

• Allow overriding CC_TURING

  SlyEcho

  committed
  4e58a05

  Copy full SHA for 4e58a05

Commits on Aug 12, 2023

• gfx1100 support

  Show description for 6415610
  

  4 peoplecommitted
  6415610

  Copy full SHA for 6415610

Commits on Aug 14, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  70e2f7c

  Copy full SHA for 70e2f7c

Commits on Aug 16, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  68e79cc

  Copy full SHA for 68e79cc

• reenable LLAMA_CUDA_FORCE_DMMV

  SlyEcho

  committed
  3de6a9a

  Copy full SHA for 3de6a9a

• makefile rewrite

  SlyEcho

  committed
  bbbc0ce

  Copy full SHA for bbbc0ce

• probably lld is not required

  SlyEcho

  committed
  c88c2a9

  Copy full SHA for c88c2a9

Commits on Aug 21, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  423db74

  Copy full SHA for 423db74

Commits on Aug 22, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  391dd9a

  Copy full SHA for 391dd9a

• use "ROCm" instead of "CUDA"

  SlyEcho

  committed
  5d3e7b2

  Copy full SHA for 5d3e7b2

Commits on Aug 24, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  7b84217

  Copy full SHA for 7b84217

• ignore all build dirs

  SlyEcho

  committed
  058f905

  Copy full SHA for 058f905

• Add Dockerfiles

  SlyEcho

  committed
  a60231f

  Copy full SHA for a60231f

• fix llama-bench

  SlyEcho

  committed
  81ecaa4

  Copy full SHA for 81ecaa4

• fix -nommq help for non CUDA/HIP

  SlyEcho

  committed
  238335f

  Copy full SHA for 238335f

Commits on Aug 25, 2023

• Merge 'origin/master' into hipblas

  SlyEcho

  committed
  9035cfc

  Copy full SHA for 9035cfc