ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-02 12:41:26 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix for Add the API to load SGPR (#2913)

Browse Source 
* Revert "Revert "[CK-Tile] Add the API to load SGPR  (#2878)" (#2904)"

This reverts commit f161b5b738.

* Fix: sgpr minor issue

* cyclic dependency resolved

* clang formatted

* removing unused variable

* clang formatted

---------

Co-authored-by: Illia Silin <98187287+illsilin@users.noreply.github.com>
This commit is contained in:
Khushbu Agarwal
2025-09-25 10:32:42 -07:00
committed by GitHub
parent 64e61b8647
commit b56e5d1d79
41 changed files with 224 additions and 173 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)