rocm/composable_kernel issues and pull requests

#1640 - remove gfx940;gfx941 from default target lists

Pull Request - State: closed - Opened by illsilin 3 months ago

#1639 - Prevent instantiation of undefined FP8 operators.

Pull Request - State: closed - Opened by andriy-ca 3 months ago

#1638 - [Issue]: FMHA change of drop_seed_offset to std::variant is breaking builds

Issue - State: closed - Opened by iratebadger 3 months ago - 3 comments
Labels: Under Investigation

#1637 - [Ck tile] layernorm2d fwd optimize

Pull Request - State: closed - Opened by coderfeli 3 months ago - 1 comment

#1636 - Update ck_a8w8

Pull Request - State: open - Opened by aska-0096 3 months ago

#1635 - [generate.py] Override blob list if it already exists

Pull Request - State: closed - Opened by jmmartinez 3 months ago

#1634 - [CK_TILE] fused-moe first version

Pull Request - State: closed - Opened by carlushuang 3 months ago - 2 comments

#1633 - Make sure cmake can handle the xnack+/xnack- targets.

Pull Request - State: closed - Opened by illsilin 3 months ago

#1632 - A prototype of TF32 gemm

Pull Request - State: open - Opened by zjing14 3 months ago

#1631 - Statically Cast Pointer Offset

Pull Request - State: closed - Opened by darren-amd 3 months ago

#1630 - Temporary disable part of dynamic op conv instances

Pull Request - State: closed - Opened by bartekxk 3 months ago

#1629 - [CK_TILE] Allow using default gemm pipeline policy

Pull Request - State: closed - Opened by poyenc 3 months ago - 1 comment

#1628 - [do not review] int4 scale based on jzhang's pre work

Pull Request - State: open - Opened by mtgu0705 3 months ago
Labels: noCI

#1627 - [DO NOT REVIEW] Add int4 dequant with scale only supports based on ZhangJing's PR #1572

Pull Request - State: open - Opened by mtgu0705 3 months ago
Labels: noCI

#1626 - Linsun/convint8 fwd instances

Pull Request - State: closed - Opened by linsun12 3 months ago - 4 comments

#1625 - Linsun/convint8 fwd instances

Pull Request - State: closed - Opened by linsun12 3 months ago

#1624 - Ck tile/moe sorting

Pull Request - State: closed - Opened by coderfeli 3 months ago

#1623 - [CK_TILE] layernorm have more accurate residual

Pull Request - State: closed - Opened by carlushuang 3 months ago - 1 comment

#1622 - [CK_TILE] Add small warp gemm

Pull Request - State: closed - Opened by poyenc 3 months ago - 1 comment

#1621 - Reduce build time.

Pull Request - State: closed - Opened by illsilin 3 months ago

#1620 - [layernorm] hot fix

Pull Request - State: closed - Opened by carlushuang 3 months ago

#1619 - [CK_TILE] Add operator batched_transpose

Pull Request - State: closed - Opened by fangche123 3 months ago - 1 comment

#1618 - Generic threshold calculation after merge fixes

Pull Request - State: closed - Opened by aledudek 3 months ago

#1617 - [Ck_tile] smoothquant

Pull Request - State: closed - Opened by rocking5566 3 months ago

#1616 - [Ck tile] smoothquant

Pull Request - State: closed - Opened by rocking5566 3 months ago

#1615 - Ck tile batched gemm example

Pull Request - State: closed - Opened by aledudek 3 months ago - 1 comment

#1614 - Batched GEMM Multiple D based on Universal GEMM

Pull Request - State: closed - Opened by zjing14 3 months ago - 12 comments

#1613 - fix clang format

Pull Request - State: closed - Opened by illsilin 3 months ago - 1 comment

#1612 - [HOTFIX] fix ci fail

Pull Request - State: closed - Opened by rocking5566 3 months ago - 1 comment

#1611 - CK Tile Batched gemm example

Pull Request - State: closed - Opened by aledudek 3 months ago

#1610 - Remove virtual destructors from unary ops

Pull Request - State: closed - Opened by bartekxk 3 months ago

#1609 - [CK_TILE] add scatter_gather

Pull Request - State: closed - Opened by valarLip 3 months ago

#1608 - [CK_TILE] Add fmha fwd headdim96 support

Pull Request - State: closed - Opened by qianfengz 3 months ago

#1607 - [CK_TILE] add generic_permute

Pull Request - State: closed - Opened by valarLip 3 months ago

#1606 - fix compilation errors for gfx12 with clang20

Pull Request - State: closed - Opened by illsilin 3 months ago

#1605 - [Ck tile] support rmsnorm and related fusion

Pull Request - State: closed - Opened by rocking5566 3 months ago - 1 comment

#1604 - [CK_TILE] layernorm support fused-quant/fused-add

Pull Request - State: closed - Opened by carlushuang 3 months ago

#1603 - [Discussion] Do we have/Where can we find swizzling rules in ck to avoid bank conflict?

Issue - State: closed - Opened by LeiWang1999 3 months ago - 8 comments
Labels: Under Investigation

#1602 - Pipeline matrix b shuffle

Pull Request - State: closed - Opened by ThomasNing 3 months ago - 2 comments

#1601 - Ck tile gemm fixes

Pull Request - State: closed - Opened by jakpiase 3 months ago

#1600 - Polished Grouped GEMM APIs and new BF16 instances

Pull Request - State: closed - Opened by aosewski 3 months ago

#1599 - add rounding converter

Pull Request - State: closed - Opened by coderfeli 3 months ago

#1598 - Hot fix ln precision rounding

Pull Request - State: closed - Opened by dummycoderfe 3 months ago

#1597 - hot_fix epsilon pos

Pull Request - State: closed - Opened by dummycoderfe 3 months ago

#1596 - Update GPU verification

Pull Request - State: closed - Opened by geyyer 4 months ago

#1595 - fix the logic of enabling XDL and WMMA instances

Pull Request - State: closed - Opened by illsilin 4 months ago

#1594 - [POST MERGE PR] Enable grouped conv bwd wei bf16 NHWGC

Pull Request - State: closed - Opened by bartekxk 4 months ago

#1593 - Explicit cast values to half

Pull Request - State: closed - Opened by cjatin 4 months ago

#1592 - topk_softmax

Pull Request - State: closed - Opened by carlushuang 4 months ago - 1 comment

#1591 - add int8 gemm multiply multiply a8w8

Pull Request - State: closed - Opened by valarLip 4 months ago

#1590 - [GEMM] Congruous GEMM optimization

Pull Request - State: open - Opened by aska-0096 4 months ago
Labels: enhancement

#1589 - Enable grouped conv bwd wei bf16 NGCHW

Pull Request - State: closed - Opened by bartekxk 4 months ago

#1588 - [CK_TILE] More fmha splitkv optimizations

Pull Request - State: closed - Opened by poyenc 4 months ago

#1587 - Bump rocm-docs-core from 1.8.2 to 1.8.3 in /docs/sphinx

Pull Request - State: closed - Opened by dependabot[bot] 4 months ago
Labels: documentation, dependencies, ci:docs-only

#1586 - [Development] How to create a temp tile window from a pointer?

Issue - State: closed - Opened by LeiWang1999 4 months ago - 8 comments

#1585 - [PT Inductor] Add parsing grouped conv fwd instances

Pull Request - State: closed - Opened by tenpercent 4 months ago

#1584 - disable bad instance detected on MI308CPX

Pull Request - State: closed - Opened by aska-0096 4 months ago

#1583 - Fix layernorm F16 type in ckProfiler

Pull Request - State: closed - Opened by rocking5566 4 months ago

#1582 - Add --lsr-drop-solution=1 compiler flag.

Pull Request - State: closed - Opened by illsilin 4 months ago

#1581 - [Issue]: Error linking ckProfiler

Issue - State: closed - Opened by RandUser123sa 4 months ago - 5 comments
Labels: Under Investigation

#1580 - [Issue]: Cannot receive the correct result while using DeviceBatchedGemmMultiD_Xdl

Issue - State: closed - Opened by hoangvictor 4 months ago - 6 comments
Labels: Under Investigation

#1579 - Codegen hipRTC compilation

Pull Request - State: closed - Opened by arai713 4 months ago - 1 comment

#1578 - added link to documentation

Pull Request - State: closed - Opened by spolifroni-amd 4 months ago
Labels: documentation, ci:docs-only

#1577 - [CK_TILE] Optimize fmha splitkv & splitkv combine kernels

Pull Request - State: closed - Opened by poyenc 4 months ago

#1576 - Update default stride

Pull Request - State: closed - Opened by geyyer 4 months ago

#1575 - Ck profiler instance support

Pull Request - State: closed - Opened by ThomasNing 4 months ago - 1 comment

#1574 - Rebase the PR #1520 to ROCm repo.

Pull Request - State: open - Opened by illsilin 4 months ago

#1572 - Add a prototype of F16/BF16xINT4 GEMM

Pull Request - State: closed - Opened by zjing14 4 months ago - 4 comments

#1570 - update layernorm

Pull Request - State: closed - Opened by ltqin 4 months ago

#1568 - remove the --rm docker container flags

Pull Request - State: closed - Opened by illsilin 4 months ago

#1567 - [CK_TILE][Mainline] Add fmha fwd compiler issue workaround for ROCm 6.3

Pull Request - State: closed - Opened by poyenc 4 months ago - 1 comment

#1567 - [CK_TILE][Mainline] Add fmha fwd compiler issue workaround for ROCm 6.3

Pull Request - State: closed - Opened by poyenc 4 months ago - 1 comment

#1565 - only build tests and examples if user sets GPU_TARGETS

Pull Request - State: closed - Opened by illsilin 4 months ago

#1562 - Grouped gemm fixes

Pull Request - State: closed - Opened by bartekxk 4 months ago

#1558 - [CK-Tile] Universal gemm memory bound pipeline

Pull Request - State: closed - Opened by aosewski 4 months ago - 1 comment

#1556 - Build codegen as standalone

Pull Request - State: closed - Opened by pfultz2 4 months ago

#1554 - Fixes small memory leak from missing hipEventDestroy

Pull Request - State: closed - Opened by cgmillette 4 months ago

#1553 - [CK_TILE] Fix 'sh' command compatibility of smoke_test_fwd.sh

Pull Request - State: closed - Opened by poyenc 4 months ago

#1551 - [Issue]: amd_wave_read_first_lane is ambiguous

Issue - State: closed - Opened by RichardGe 4 months ago - 4 comments
Labels: Under Investigation

#1546 - Generic threshold calculation

Pull Request - State: closed - Opened by aledudek 4 months ago

#1543 - Introduce gemm_elementwise_gemm

Pull Request - State: open - Opened by mirza-halilcevic 4 months ago

#1542 - Introduce gemm_softmax_gemm to codegen

Pull Request - State: open - Opened by mirza-halilcevic 4 months ago - 1 comment

#1541 - BF16 GEMM Stream-K

Pull Request - State: closed - Opened by ozturkosu 4 months ago - 5 comments

#1540 - Add generating mha static library for gfx90a

Pull Request - State: closed - Opened by BrianHarrisonAMD 4 months ago

#1528 - Add a gpu gemm reference kernel

Pull Request - State: closed - Opened by geyyer 5 months ago - 1 comment
Labels: CI - Pass

#1528 - Add a gpu gemm reference kernel

Pull Request - State: closed - Opened by geyyer 5 months ago - 1 comment
Labels: CI - Pass

#1520 - Enable hipRTC compilation of codegen tests

Pull Request - State: closed - Opened by music-dino 5 months ago - 3 comments

#1514 - [Question] Register data layout for two consecutive GEMMs in flash attention kernel (how is TransposedC implemented)?

Issue - State: closed - Opened by bulffi 5 months ago - 3 comments
Labels: question, Under Investigation

#1471 - [WIP] add more example for permute/scatter-gather/moe

Pull Request - State: open - Opened by carlushuang 6 months ago

#1471 - [WIP] add more example for permute/scatter-gather/moe

Pull Request - State: open - Opened by carlushuang 6 months ago

#1459 - INSTANCES_ONLY=ON quietly override target list and cause issues

Issue - State: closed - Opened by junliume 6 months ago - 3 comments
Labels: urgency_high

#1459 - INSTANCES_ONLY=ON quietly override target list and cause issues

Issue - State: closed - Opened by junliume 6 months ago - 3 comments
Labels: urgency_high

#1434 - WMMA / RDNA3+ kernels for backwards fused attention?

Issue - State: closed - Opened by Googulator 6 months ago - 3 comments
Labels: Under Investigation

#1431 - debug build got error: R_X86_64_REX_GOTPCRELX | R_X86_64_PC32 out of range

Issue - State: closed - Opened by ZJLi2013 6 months ago - 3 comments
Labels: help wanted, Under Investigation

#1426 - Add dynamic elementwise op

Pull Request - State: closed - Opened by bartekxk 6 months ago

#1373 - This macro looks terrible

Issue - State: closed - Opened by atamazov 7 months ago - 1 comment
Labels: Under Investigation

#1334 - cmake issue after #1286: ModuleNotFoundError: No module named 'dataclasses'

Issue - State: closed - Opened by junliume 8 months ago - 5 comments

GitHub / rocm/composable_kernel issues and pull requests