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

2
include/ck_tile/ops/smoothquant/kernel/moe_smoothquant_kernel.hpp
View File

@@ -138,7 +138,7 @@ struct MoeSmoothquant
const index_t i_topk = blockIdx.x;
const index_t i_token = blockIdx.y * Block_M;
const index_t i_token_in_thrd =
__builtin_amdgcn_readfirstlane(threadIdx.x / Problem::BlockShape::ThreadPerBlock_N);
amd_wave_read_first_lane(threadIdx.x / Problem::BlockShape::ThreadPerBlock_N);
const index_t i_expert = reinterpret_cast<const index_t*>(
kargs.p_topk_ids)[(i_token + i_token_in_thrd) * kargs.topk + i_topk];

6
include/ck_tile/ops/smoothquant/pipeline/smoothquant_pipeline_two_pass.hpp
View File

@@ -57,7 +57,7 @@ struct SmoothquantPipelineTwoPass
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));
auto reduce_absmax_func = ReduceOp::AbsMax{};
auto reduce_absmax3_func = [](auto acc_, auto v_0_, auto v_1_) {
@@ -77,7 +77,7 @@ struct SmoothquantPipelineTwoPass
auto absmax = block_reduce2d.template MakeYBlockTile<XTensorType>();
set_tile(absmax, reduce_absmax_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 x = load_tile(x_window);
const auto smscale = load_tile(smscale_window);
@@ -121,7 +121,7 @@ struct SmoothquantPipelineTwoPass
move_tile_window(qy_window, {0, stride_to_right_most_window});
// recompute y and quantize y to qy
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 x = load_tile(x_window);
const auto smscale = load_tile(smscale_window);