ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-03 05:01:25 +00:00
Code Issues Packages Projects Releases Wiki Activity

[Ck tile] support rmsnorm and related fusion (#1605)

Browse Source 
* Add reduce2d new api

* Prevent user use cross warp reduction

* Fix bug of std caculation

* Add rmsnorm2d

* Add rmsnorm small example

* Remove static assert to prevent compile fail

* Add script to test performance and correctness

* Add missing cmake change

* refine naming

* refine example of rmsnorm

* Fix bug of rmsnorm

* Refine naming

* Fix cmake

* clang format

* Refine pipeline name

* Add add_rmsnorm2d_rdquant kernel

* Add reduce op

* host verification

* Fix bug of one pass pipeline

* Refine tile size

* Add two pass pipeline

* Rename two pass to three pass

* Fix bug of kSaveX == false

* Add instance library

* Add test script

* Fix bug of x verification

* Add save_x to trait

* Add README

* Move reduce2d into reduce folder

* Fix bug of welford when number of m warp > 1

* remove reduncant comment

* 1. move 06_rmsnorm2d to 10_rmsnorm2d
2. move 07_add_rmsnorm2d_rdquant to 11_add_rmsnorm2d_rdquant

* clang format and add missing header

* Add host validation of add + layernorm2d + rsquant

* Revert "Add host validation of add + layernorm2d + rsquant"

This reverts commit 936cb45797.

* Remove deprecated flag
This commit is contained in:
rocking
2024-10-30 15:22:56 +08:00
committed by GitHub
parent 8632221814
commit 3d60953477
90 changed files with 4674 additions and 128 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
include/ck_tile/ops/welford/block/block_welford.hpp
View File

@@ -276,8 +276,8 @@ struct BlockWelfordCrossWarpSync
fp32x4_t all_scratch[thread_buf_size * num_reduce_warps];
static_for<0, thread_buf_size, 1>{}([&](auto i_0) {
static_for<0, num_reduce_warps, 1>{}([&](auto i_1) {
all_scratch[i_0 * num_warps + i_1] =
smem_ptr[i_0 * num_reduce_warps + local_smem_os + i_1];
all_scratch[i_0 * num_reduce_warps + i_1] =
smem_ptr[i_0 * num_warps + local_smem_os + i_1];
});
});
block_sync_lds(); // TODO: we don't need sync here
@@ -286,7 +286,7 @@ struct BlockWelfordCrossWarpSync
static_for<0, thread_buf_size, 1>{}([&](auto i_0) {
// TODO: use descriptor for this
auto v_local = all_scratch[i_0 * num_warps];
auto v_local = all_scratch[i_0 * num_reduce_warps];
auto v_local_mean = bit_cast<DataType>(v_local[0]);
auto v_local_var = bit_cast<DataType>(v_local[1]);
auto v_local_count = bit_cast<int>(v_local[2]);
@@ -294,7 +294,7 @@ struct BlockWelfordCrossWarpSync
// further reduce mean/var
static_for<0, num_reduce_warps - 1, 1>{}([&](auto i_1_n1) {
constexpr auto i_1 = number<i_1_n1 + 1>{};
const fp32x4_t v_remote = all_scratch[i_0 * num_warps + i_1];
const fp32x4_t v_remote = all_scratch[i_0 * num_reduce_warps + i_1];
const auto v_remote_mean = bit_cast<DataType>(v_remote[0]);
const auto v_remote_var = bit_cast<DataType>(v_remote[1]);
const auto v_remote_count = bit_cast<int>(v_remote[2]);