Skip to content

[None][refactor] Split thop.attention into phased fmha #17558

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

Open
yihwang-nv wants to merge 1 commit into
base: main
Choose a base branch
from
Open

[None][refactor] Split thop.attention into phased fmha #17558

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
41 changes: 41 additions & 0 deletions cpp/tensorrt_llm/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,6 +22,47 @@ set(API_INCLUDE_DIR ${PROJECT_SOURCE_DIR}/include)
include_directories(${CMAKE_CURRENT_SOURCE_DIR}/cutlass_extensions/include
${API_INCLUDE_DIR})

get_filename_component(TRTLLM_REPO_ROOT_DIR "${CMAKE_CURRENT_SOURCE_DIR}/../.."
ABSOLUTE)
set(TRTLLM_FMHA_PARAMS_GENERATOR
"${TRTLLM_REPO_ROOT_DIR}/scripts/generate_fmha_params.py")
# The schema classes live in the modules that own them; the generator parses all
# of them and resolves a nested member by the class its annotation names.
set(TRTLLM_FMHA_PARAMS_SCHEMAS
"${TRTLLM_REPO_ROOT_DIR}/tensorrt_llm/_torch/attention/backends/fmha/interface.py"
"${TRTLLM_REPO_ROOT_DIR}/tensorrt_llm/_torch/attention/backends/interface.py"
"${TRTLLM_REPO_ROOT_DIR}/tensorrt_llm/_torch/attention/backends/sparse/params.py"
)
set(TRTLLM_FMHA_PARAMS_SCHEMA_ARGS "")
foreach(_schema ${TRTLLM_FMHA_PARAMS_SCHEMAS})
list(APPEND TRTLLM_FMHA_PARAMS_SCHEMA_ARGS --module-path "${_schema}")
endforeach()
set(TRTLLM_FMHA_PARAMS_GENERATED_INCLUDE_DIR
"${CMAKE_CURRENT_BINARY_DIR}/generated")
# Generate at configure time rather than as a build step: a configure-only tree,
# which is how the clangd compile database is produced, still needs these
# headers for attentionOp.h to parse. CONFIGURE_DEPENDS re-runs CMake when a
# schema or the generator changes, so editing a schema and building is enough to
# refresh them.
set_property(
DIRECTORY
APPEND
PROPERTY CMAKE_CONFIGURE_DEPENDS ${TRTLLM_FMHA_PARAMS_GENERATOR}
${TRTLLM_FMHA_PARAMS_SCHEMAS})
execute_process(
COMMAND
"${Python3_EXECUTABLE}" "${TRTLLM_FMHA_PARAMS_GENERATOR}"
${TRTLLM_FMHA_PARAMS_SCHEMA_ARGS} --class-name FmhaParams --out-dir
"${TRTLLM_FMHA_PARAMS_GENERATED_INCLUDE_DIR}/tensorrt_llm/thop"
RESULT_VARIABLE TRTLLM_FMHA_PARAMS_RESULT
OUTPUT_VARIABLE TRTLLM_FMHA_PARAMS_OUTPUT
ERROR_VARIABLE TRTLLM_FMHA_PARAMS_OUTPUT)
if(NOT TRTLLM_FMHA_PARAMS_RESULT EQUAL 0)
message(
FATAL_ERROR
"Generating native FmhaParams failed:\n${TRTLLM_FMHA_PARAMS_OUTPUT}")
endif()

set(TARGET_ARCH "unknown")

message(STATUS "CMAKE_SYSTEM_PROCESSOR: ${CMAKE_SYSTEM_PROCESSOR}")
Expand Down
Loading
Loading