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] Add the API to load SGPR (#2878)

Browse Source 
* Have a workable version for SGPR

* have a workable version for atomic add

* Revert "have a workable version for atomic add"

This reverts commit 792377a590c26cfff9c8f545d9a9e8484a7422eb.

* substitute with the new sgpr read api

* update the CHANGELOG

* have a workable version for atomic add

* Revert "have a workable version for atomic add"

This reverts commit 792377a590c26cfff9c8f545d9a9e8484a7422eb.

* change to static for logic

* have a workable version for atomic add

* Revert "have a workable version for atomic add"

This reverts commit 792377a590c26cfff9c8f545d9a9e8484a7422eb.
This commit is contained in:
Thomas Ning
2025-09-23 01:23:56 -07:00
committed by GitHub
parent b6e8994386
commit 2cbbf5dcb3
40 changed files with 273 additions and 167 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
include/ck_tile/ops/add_rmsnorm2d_rdquant/pipeline/add_rmsnorm2d_rdquant_fwd_pipeline_three_pass.hpp
View File

@@ -92,13 +92,13 @@ struct AddRmsnorm2dRdquantFwdPipelineThreePass
static constexpr index_t Block_N = Problem::BlockShape::Block_N;
index_t num_n_tile_iteration =
__builtin_amdgcn_readfirstlane(integer_divide_ceil(row_size, Block_N));
amd_wave_read_first_lane(integer_divide_ceil(row_size, Block_N));
using XTensorType = decltype(cast_tile<ComputeDataType>(load_tile(a_window)));
auto square_sum = block_reduce2d.template MakeYBlockTile<XTensorType>();
set_tile(square_sum, reduce_square_sum_func.GetIdentityValue<ComputeDataType>());
for(int iN = __builtin_amdgcn_readfirstlane(0); iN < num_n_tile_iteration; ++iN)
for(int iN = amd_wave_read_first_lane(0); iN < num_n_tile_iteration; ++iN)
{
const auto a = load_tile(a_window);
const auto b = load_tile(b_window);
@@ -149,7 +149,7 @@ struct AddRmsnorm2dRdquantFwdPipelineThreePass
if constexpr(kSaveX)
__syncthreads();
for(int iN = __builtin_amdgcn_readfirstlane(0); iN < num_n_tile_iteration; ++iN)
for(int iN = amd_wave_read_first_lane(0); iN < num_n_tile_iteration; ++iN)
{
auto x = [&]() {
if constexpr(kSaveX)
@@ -226,7 +226,7 @@ struct AddRmsnorm2dRdquantFwdPipelineThreePass
}
move_tile_window(gamma_window, {Block_N});
for(int iN = __builtin_amdgcn_readfirstlane(0); iN < num_n_tile_iteration; ++iN)
for(int iN = amd_wave_read_first_lane(0); iN < num_n_tile_iteration; ++iN)
{
auto x = [&]() {
if constexpr(kSaveX)