ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-16 19:09:59 +00:00
Code Issues Packages Projects Releases Wiki Activity

hotfix fix sorting int64 (#2025)

Browse Source 
* fix sorting int64

* clang format

* fix example issue

* update WA issue #

---------

Co-authored-by: coderfeli <coderfeli@163.com>
Co-authored-by: carlushuang <carlus.huang@amd.com>

[ROCm/composable_kernel commit: a82f338fb9]
This commit is contained in:
felix
2025-03-28 11:31:52 +08:00
committed by GitHub
parent 895ba2b497
commit 20ffa0f474
4 changed files with 33 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
example/ck_tile/13_moe_sorting/moe_sorting.cpp
View File

@@ -74,7 +74,7 @@ bool test_moe_sorting(ck_tile::ArgParser args)
int topk = args.get_int("k");
int seed = args.get_int("seed");
int unit_size = args.get_int("unit");
int moe_buf_size = args.get_int("moe_buf_size");
int64_t moe_buf_size = static_cast<int64_t>(args.get_uint64("moe_buf_size"));
int kname = args.get_int("kname");
int warmup = args.get_int("warmup");
int repeat = args.get_int("repeat");
@@ -175,7 +175,7 @@ bool test_moe_sorting(ck_tile::ArgParser args)
unit_size,
num_experts,
topk,
static_cast<ck_tile::index_t>(moe_buf_size * sizeof(float))};
static_cast<ck_tile::long_index_t>(moe_buf_size * sizeof(float))};
ck_tile::stream_config sc{nullptr,
true,

29
example/ck_tile/15_fused_moe/instances/fused_moe_api.cpp
View File

@@ -19,20 +19,21 @@ float fused_moe(fused_moe_traits t, fused_moe_args a, const ck_tile::stream_conf
auto t0 = fused_moesorting_trait{"int32", "fp32", t.local_expert_masking};
auto a0 = fused_moesorting_args{
a.topk_ids_ptr, // const void* p_topk_ids;
a.topk_weight_ptr, // const void* p_weights;
a.local_expert_mask_ptr, // const void* p_local_expert_mask;
a.sorted_token_ids_ptr, // void* p_sorted_token_ids;
a.sorted_weight_ptr, // void* p_sorted_weights;
a.sorted_expert_ids_ptr, // void* p_sorted_expert_ids;
a.num_sorted_tiles_ptr, // void* p_total_tokens_post_pad;
a.o_ptr, // void* p_moe_buf;
a.ws_ptr, // void* p_ws;
a.num_tokens, // index_t tokens;
a.block_m, // index_t unit_size;
a.num_experts, // index_t num_experts;
a.topk, // index_t topk;
a.num_tokens * a.stride_token * o_data_bytes // index_t moe_buf_bytes;
a.topk_ids_ptr, // const void* p_topk_ids;
a.topk_weight_ptr, // const void* p_weights;
a.local_expert_mask_ptr, // const void* p_local_expert_mask;
a.sorted_token_ids_ptr, // void* p_sorted_token_ids;
a.sorted_weight_ptr, // void* p_sorted_weights;
a.sorted_expert_ids_ptr, // void* p_sorted_expert_ids;
a.num_sorted_tiles_ptr, // void* p_total_tokens_post_pad;
a.o_ptr, // void* p_moe_buf;
a.ws_ptr, // void* p_ws;
a.num_tokens, // index_t tokens;
a.block_m, // index_t unit_size;
a.num_experts, // index_t num_experts;
a.topk, // index_t topk;
static_cast<ck_tile::long_index_t>(a.num_tokens) * a.stride_token *
o_data_bytes // index_t moe_buf_bytes;
};
auto t1 = fused_moegemm_traits{t.prec_i,