DeepSeek/DeepEP
1
0
Fork 0
mirror of https://github.com/deepseek-ai/DeepEP synced 2025-06-26 18:28:11 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Add fast-debugging flags

Browse Source
This commit is contained in:
Chenggang Zhao
2025-06-23 11:47:51 +08:00
parent 1c277c303e
commit 6a708ec14c
4 changed files with 9 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
csrc/CMakeLists.txt
View File

@@ -6,6 +6,7 @@ set(CMAKE_VERBOSE_MAKEFILE ON)
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -O3 -fPIC") set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -O3 -fPIC")
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -O3 -fPIC") set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -O3 -fPIC")
set(CUDA_SEPARABLE_COMPILATION ON) set(CUDA_SEPARABLE_COMPILATION ON)
list(APPEND CUDA_NVCC_FLAGS "-DENABLE_FAST_DEBUG")
list(APPEND CUDA_NVCC_FLAGS "-O3") list(APPEND CUDA_NVCC_FLAGS "-O3")
list(APPEND CUDA_NVCC_FLAGS "--ptxas-options=--verbose,--register-usage-level=10,--warn-on-local-memory-usage") list(APPEND CUDA_NVCC_FLAGS "--ptxas-options=--verbose,--register-usage-level=10,--warn-on-local-memory-usage")

8
csrc/kernels/configs.cuh
View File

@@ -7,9 +7,15 @@
#define NUM_BUFFER_ALIGNMENT_BYTES 128 #define NUM_BUFFER_ALIGNMENT_BYTES 128
#define FINISHED_SUM_TAG 1024 #define FINISHED_SUM_TAG 1024
#define NUM_WAIT_NANOSECONDS 500
#ifndef ENABLE_FAST_DEBUG
#define NUM_CPU_TIMEOUT_SECS 100 #define NUM_CPU_TIMEOUT_SECS 100
#define NUM_TIMEOUT_CYCLES 200000000000ull // 200G cycles ~= 100s #define NUM_TIMEOUT_CYCLES 200000000000ull // 200G cycles ~= 100s
#define NUM_WAIT_NANOSECONDS 500 #else
#define NUM_CPU_TIMEOUT_SECS 10
#define NUM_TIMEOUT_CYCLES 20000000000ull // 20G cycles ~= 10s
#endif
#define LOW_LATENCY_SEND_PHASE 1 #define LOW_LATENCY_SEND_PHASE 1
#define LOW_LATENCY_RECV_PHASE 2 #define LOW_LATENCY_RECV_PHASE 2

4
csrc/kernels/launch.cuh
View File

@@ -58,12 +58,9 @@ cfg.dynamicSmemBytes = smem_size;
#define SWITCH_RDMA_RANKS(case_macro) \ #define SWITCH_RDMA_RANKS(case_macro) \
switch (num_ranks / NUM_MAX_NVL_PEERS) { \ switch (num_ranks / NUM_MAX_NVL_PEERS) { \
case 2: case_macro(2); \ case 2: case_macro(2); \
case 3: case_macro(3); \
case 4: case_macro(4); \ case 4: case_macro(4); \
case 8: case_macro(8); \ case 8: case_macro(8); \
case 16: case_macro(16); \ case 16: case_macro(16); \
case 18: case_macro(18); \
case 20: case_macro(20); \
default: EP_HOST_ASSERT(false and "Unsupported RDMA ranks"); \ default: EP_HOST_ASSERT(false and "Unsupported RDMA ranks"); \
} while (false) } while (false)
@@ -78,7 +75,6 @@ cfg.dynamicSmemBytes = smem_size;
#define SWITCH_TYPES(case_macro) \ #define SWITCH_TYPES(case_macro) \
switch (type) { \ switch (type) { \
case CUDA_R_16BF: case_macro(nv_bfloat16); \ case CUDA_R_16BF: case_macro(nv_bfloat16); \
case CUDA_R_32F: case_macro(float); \
default: EP_HOST_ASSERT(false && "Unsupported type"); \ default: EP_HOST_ASSERT(false && "Unsupported type"); \
} while (false) } while (false)

2
tests/test_internode.py
View File

@@ -220,7 +220,7 @@ def test_main(num_sms: int, local_rank: int, num_local_ranks: int, num_ranks: in
def test_loop(local_rank: int, num_local_ranks: int): def test_loop(local_rank: int, num_local_ranks: int):
num_nodes = int(os.getenv('WORLD_SIZE', 1)) num_nodes = int(os.getenv('WORLD_SIZE', 1))
rank, num_ranks, group = init_dist(local_rank, num_local_ranks) rank, num_ranks, group = init_dist(local_rank, num_local_ranks)
test_ll_compatibility = True test_ll_compatibility = os.getenv('EP_TEST_LL_COMPATIBILITY', False)
if test_ll_compatibility: if test_ll_compatibility:
ll_num_tokens, ll_hidden, ll_num_experts, ll_num_topk = 16, 5120, 256, 9 ll_num_tokens, ll_hidden, ll_num_experts, ll_num_topk = 16, 5120, 256, 9