Skip to content

kvcache framework restructure #1139

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
Adrenaline-S wants to merge 2 commits into
base: main
Choose a base branch
from
Draft

kvcache framework restructure #1139

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
a88a7da
chore: restructure kvcache directory layout (rename only)
Adrenaline-S Jun 25, 2026
8e3adcd
refactor: sync kvcache framework from refactor-kvcache and dsv4_on_de…
Adrenaline-S Jun 26, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 5 additions & 0 deletions BUILD
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -45,6 +45,11 @@ config_setting(
values = {"define": "using_cuda12_x86=true"},
)

config_setting(
name = "using_cuda13_x86",
values = {"define": "using_cuda13_x86=true"},
)

config_setting(
name = "using_rocm",
values = {"define": "using_rocm=true"},
Expand Down
167 changes: 136 additions & 31 deletions rtp_llm/cpp/cache/BUILD
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,13 +1,34 @@
load("//:def.bzl", "copts")
load("@arch_config//:arch_select.bzl", "torch_deps")

cc_library(
name = "cp_slot_mapper",
srcs = ["CPSlotMapper.cc"],
hdrs = ["CPSlotMapper.h"],
copts = copts(),
visibility = ["//visibility:public"],
)

cc_library(
name = "cache_group_type",
hdrs = [
"CacheGroupType.h",
"spec/CacheGroupType.h",
],
copts = copts(),
visibility = ["//visibility:public"],
)

cc_library(
name = "kv_cache_spec_desc_types",
hdrs = [
"spec/KVCacheSpecDescTypes.h",
],
copts = copts(),
visibility = ["//visibility:public"],
deps = [
":cache_group_type",
"//rtp_llm/models_py/bindings/core:types",
],
)

cc_library(
Expand All @@ -19,17 +40,49 @@ cc_library(
visibility = ["//visibility:public"],
)

cc_library(
name = "kv_cache_specs",
hdrs = [
"spec/KVCacheSpec.h",
"spec/KVCacheSpecBase.h",
"spec/KVCacheSpecDesc.h",
"spec/KVCacheSpecDescTypes.h",
"spec/MHAKVCacheSpec.h",
"spec/LinearKVCacheSpec.h",
"spec/MLAKVCacheSpec.h",
"spec/OpaqueKVCacheSpec.h",
"Types.h",
],
copts = copts(),
visibility = ["//visibility:public"],
deps = [
":block_info",
":batch_kv_cache_resource",
":cache_group_type",
":cp_slot_mapper",
"//rtp_llm/cpp/config:config_modules",
"//rtp_llm/cpp/model_utils:model_utils",
"//rtp_llm/models_py/bindings/core:types",
],
)

cc_library(
name = "cache_types",
srcs = [
"Types.cc",
],
hdrs = [
"BufferTypes.h",
"CacheConfig.h",
"CacheGroupType.h",
"KVCacheSpec.h",
"KVCacheSpecBase.h",
"MHAKVCacheSpec.h",
"LinearKVCacheSpec.h",
"MLAKVCacheSpec.h",
"spec/CacheGroupType.h",
"spec/KVCacheSpec.h",
"spec/KVCacheSpecBase.h",
"spec/KVCacheSpecDesc.h",
"spec/KVCacheSpecDescTypes.h",
"spec/MHAKVCacheSpec.h",
"spec/LinearKVCacheSpec.h",
"spec/MLAKVCacheSpec.h",
"spec/OpaqueKVCacheSpec.h",
"Types.h",
"WarmUpResult.h",
],
Expand All @@ -39,8 +92,11 @@ cc_library(
":block_info",
":batch_kv_cache_resource",
":cache_group_type",
":cp_slot_mapper",
"//:rtp_compute_ops",
"//rtp_llm/cpp/config:config_modules",
"//rtp_llm/cpp/engine_base/stream:complete_token_ids",
"//rtp_llm/cpp/model_utils:model_utils",
"//rtp_llm/models_py/bindings/core:types",
] + torch_deps(),
)
Expand All @@ -51,6 +107,7 @@ cc_library(
"BlockCache.cc",
"BlockPool.cc",
"MemoryLayoutStrategy.cc",
"SharedBlockCache.cc",
],
hdrs = [
"BlockCache.h",
Expand All @@ -60,55 +117,100 @@ cc_library(
"BlockRefCounter.h",
"MemoryLayoutStrategy.h",
"MemoryLayoutConfig.h",
"SharedBlockCache.h",
],
copts = copts(),
visibility = ["//visibility:public"],
deps = [
":cache_types",
"//rtp_llm/models_py/bindings/core:exec_ops_hdr",
"//rtp_llm/cpp/disaggregate/cache_store",
"//rtp_llm/cpp/engine_base/stream:complete_token_ids",
"//rtp_llm/cpp/disaggregate/cache_store:cache_store_interface",
"//rtp_llm/cpp/utils:kv_cache_utils",
"//rtp_llm/cpp/utils:lru_cache",
"//rtp_llm/cpp/utils:profiling_scope",
"//rtp_llm/models_py/bindings/core:type_convert",
] + select({
"@//:using_cuda": [
"//rtp_llm/models_py/bindings/cuda:cuda_host_utils",
"@local_config_cuda//cuda:cuda_headers",
"@local_config_cuda//cuda:cudart",
],
"//conditions:default": [],
}),
)

cc_library(
name = "kv_cache_transfer_planner",
srcs = [
"KVCacheTransferPlanner.cc",
],
hdrs = [
"KVCacheTransferPlanner.h",
],
copts = copts(),
visibility = ["//visibility:public"],
deps = [
":cache_group_type",
],
)

cc_library(
name = "kv_cache_group",
srcs = [
"FullKVCacheGroup.cc",
"KVCacheGroup.cc",
"LinearKVCacheGroup.cc",
"group/FullKVCacheGroup.cc",
"group/KVCacheGroup.cc",
"group/LinearKVCacheGroup.cc",
"group/SWAKVCacheGroup.cc",
],
hdrs = [
"FullKVCacheGroup.h",
"KVCacheGroup.h",
"LinearKVCacheGroup.h",
"group/FullKVCacheGroup.h",
"group/KVCacheGroup.h",
"group/LinearKVCacheGroup.h",
"group/SWAKVCacheGroup.h",
],
copts = copts(),
visibility = ["//visibility:public"],
deps = [
":block_pool",
":cache_types",
],
)

cc_library(
name = "kv_cache_allocator_hdr",
hdrs = [
"allocator/KVCacheAllocator.h",
],
copts = copts(),
visibility = ["//visibility:public"],
deps = [
":block_pool",
":cache_types",
"//rtp_llm/cpp/metrics:metrics",
] + torch_deps(),
)

cc_library(
name = "kv_cache_allocator",
srcs = [
"HybridTypeKVCacheAllocator.cc",
"KVCacheAllocator.cc",
"SingleTypeKVCacheAllocator.cc",
"allocator/HybridKVCacheAllocator.cc",
"allocator/HybridPoolKVCacheAllocator.cc",
"allocator/HybridTypeKVCacheAllocator.cc",
"allocator/KVCacheAllocator.cc",
"allocator/SingleTypeKVCacheAllocator.cc",
],
hdrs = [
"HybridTypeKVCacheAllocator.h",
"KVCacheAllocator.h",
"SingleTypeKVCacheAllocator.h",
"allocator/HybridKVCacheAllocator.h",
"allocator/HybridPoolKVCacheAllocator.h",
"allocator/HybridTypeKVCacheAllocator.h",
"allocator/KVCacheAllocator.h",
"allocator/SingleTypeKVCacheAllocator.h",
],
copts = copts(),
visibility = ["//visibility:public"],
deps = [
":kv_cache_allocator_hdr",
":kv_cache_group",
"//rtp_llm/cpp/engine_base/stream:complete_token_ids",
"//rtp_llm/models_py/bindings/core:exec_ops_hdr",
],
)
Expand All @@ -124,25 +226,27 @@ cc_library(
visibility = ["//visibility:public"],
deps = [
"//rtp_llm/cpp/utils:core_utils",
"//rtp_llm/cpp/cache:cache_group_type",
":cache_group_type",
],
)

cc_library(
name = "cache_core",
srcs = [
"CacheConfigCreator.cc",
"HybridConfigCreator.cc",
"config_creator/CacheConfigCreator.cc",
"config_creator/HybridConfigCreator.cc",
"config_creator/HybridPoolConfigCreator.cc",
"KVCacheHashUtil.cc",
"MemoryEvaluationHelper.cc",
"SingleConfigCreator.cc",
"config_creator/MemoryEvaluationHelper.cc",
"config_creator/SingleConfigCreator.cc",
],
hdrs = [
"CacheConfigCreator.h",
"HybridConfigCreator.h",
"config_creator/CacheConfigCreator.h",
"config_creator/HybridConfigCreator.h",
"config_creator/HybridPoolConfigCreator.h",
"KVCacheHashUtil.h",
"MemoryEvaluationHelper.h",
"SingleConfigCreator.h",
"config_creator/MemoryEvaluationHelper.h",
"config_creator/SingleConfigCreator.h",
],
copts = copts(),
visibility = ["//visibility:public"],
Expand All @@ -151,6 +255,7 @@ cc_library(
":cache_types",
":kv_cache_allocator",
"//rtp_llm/cpp/config:model_config",
"//rtp_llm/cpp/engine_base/stream:complete_token_ids",
"//rtp_llm/models_py/bindings/core:exec_ops_hdr",
"//rtp_llm/models_py/bindings/core:type_convert",
"//rtp_llm/cpp/disaggregate/cache_store",
Expand Down
Loading
Loading