mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-05-05 06:01:23 +00:00
[CK TILE ENGINE] GEMM Multi D Restructure (#3121)
* Renaming old code * Adding GEMM code with new Architecture * Partial Progress : Errors * Partial Progress : Working code * Changes to element wise function * Removing Debugging statements * Working GEMM Multi D code * Removing Stale Code * Address Copilot review comments * Address Copilot review comments * Changes to validation file * Changes to common code snippets * Creating common folder * Removing duplicate files * Pointing to right common file * Pointing to right common file * Pointing to right common file * Changing to VERBOSE * Changing CMAKE messages to verbose * Updating Cmake with right layout datatype configs * Working code for GEMM Multi D
This commit is contained in:
Thrupti Raj Lakshmana Gowda
committed by
GitHub
GitHub
parent
04efd282cf
commit
a33d98f8e2
@@ -1,4 +1,4 @@
|
||||
set(GEMM_PRESHUFFLE_DATATYPE "fp16;fp8" CACHE STRING "List of datatypes for GEMM Preshuffle (semicolon-separated)")
|
||||
set(GEMM_PRESHUFFLE_DATATYPE "fp16;fp8;bf16;bf8" CACHE STRING "List of datatypes for GEMM Preshuffle (semicolon-separated)")
|
||||
set(GEMM_PRESHUFFLE_LAYOUT "rcr" CACHE STRING "List of layout for GEMM Preshuffle (semicolon-separated)")
|
||||
set(GEMM_PRESHUFFLE_CONFIG_FILE "" CACHE STRING "Custom config file name (without path, must be in configs/ folder)")
|
||||
option(ENABLE_CCACHE_GEMM_PRESHUFFLE "Enable ccache for GEMM Preshuffle ops compilation" OFF)
|
||||
@@ -122,15 +122,15 @@ function(build_individual_gemm_preshuffle_targets datatype layout)
|
||||
if(DEFINED ENV{GEMM_PRESHUFFLE_CONFIG_FILE} AND NOT "$ENV{GEMM_PRESHUFFLE_CONFIG_FILE}" STREQUAL "")
|
||||
set(config_filename "$ENV{GEMM_PRESHUFFLE_CONFIG_FILE}")
|
||||
set(json_blob "${CMAKE_CURRENT_LIST_DIR}/configs/${config_filename}")
|
||||
message(STATUS " Using config from environment variable: ${config_filename}")
|
||||
message(VERBOSE " Using config from environment variable: ${config_filename}")
|
||||
elseif(NOT "${GEMM_PRESHUFFLE_CONFIG_FILE}" STREQUAL "")
|
||||
# Use CMake variable if set
|
||||
set(json_blob "${CMAKE_CURRENT_LIST_DIR}/configs/${GEMM_PRESHUFFLE_CONFIG_FILE}")
|
||||
message(STATUS " Using custom config: ${GEMM_PRESHUFFLE_CONFIG_FILE}")
|
||||
message(VERBOSE " Using custom config: ${GEMM_PRESHUFFLE_CONFIG_FILE}")
|
||||
else()
|
||||
# Use default config for all layouts
|
||||
set(json_blob "${CMAKE_CURRENT_LIST_DIR}/configs/default_config.json")
|
||||
message(STATUS " Using default config for layout ${layout}")
|
||||
message(VERBOSE " Using default config for layout ${layout}")
|
||||
endif()
|
||||
|
||||
# Check if config file exists
|
||||
@@ -151,18 +151,18 @@ function(build_individual_gemm_preshuffle_targets datatype layout)
|
||||
endif()
|
||||
|
||||
# Generate individual kernel files using parallel version
|
||||
message(STATUS "Generating individual kernels for ${datatype} ${layout} using ${num_workers} workers...")
|
||||
message(STATUS " Working path: ${working_path}")
|
||||
message(STATUS " Config file: ${json_blob}")
|
||||
message(STATUS " Python executable: ${Python3_EXECUTABLE}")
|
||||
message(STATUS " Script path: ${CMAKE_CURRENT_LIST_DIR}/gemm_preshuffle_instance_builder.py")
|
||||
message(VERBOSE "Generating individual kernels for ${datatype} ${layout} using ${num_workers} workers...")
|
||||
message(VERBOSE " Working path: ${working_path}")
|
||||
message(VERBOSE " Config file: ${json_blob}")
|
||||
message(VERBOSE " Python executable: ${Python3_EXECUTABLE}")
|
||||
message(VERBOSE " Script path: ${CMAKE_CURRENT_LIST_DIR}/gemm_preshuffle_instance_builder.py")
|
||||
|
||||
# Create working directory first
|
||||
file(MAKE_DIRECTORY ${working_path})
|
||||
|
||||
# First, just list the kernels (fast operation)
|
||||
message(STATUS " Listing kernel configurations...")
|
||||
message(STATUS " GPU Targets: ${GEMM_PRESHUFFLE_GPU_TARGETS_INDIVIDUAL}")
|
||||
message(VERBOSE " Listing kernel configurations...")
|
||||
message(VERBOSE " GPU Targets: ${GEMM_PRESHUFFLE_GPU_TARGETS_INDIVIDUAL}")
|
||||
execute_process(
|
||||
COMMAND ${Python3_EXECUTABLE} -u ${CMAKE_CURRENT_LIST_DIR}/gemm_preshuffle_instance_builder.py
|
||||
--working_path ${working_path}
|
||||
@@ -185,7 +185,7 @@ function(build_individual_gemm_preshuffle_targets datatype layout)
|
||||
if(EXISTS ${working_path}/gemm_preshuffle_kernel_count.txt)
|
||||
file(READ ${working_path}/gemm_preshuffle_kernel_count.txt kernel_count)
|
||||
string(STRIP "${kernel_count}" kernel_count)
|
||||
message(STATUS " Found ${kernel_count} kernel configurations")
|
||||
message(VERBOSE " Found ${kernel_count} kernel configurations")
|
||||
else()
|
||||
message(FATAL_ERROR "Kernel count file not found")
|
||||
endif()
|
||||
@@ -209,10 +209,10 @@ function(build_individual_gemm_preshuffle_targets datatype layout)
|
||||
endfunction()
|
||||
|
||||
# Main build logic - Only individual builds supported
|
||||
message(STATUS "=== Starting Tile Engine GEMM Preshuffle Configuration ===")
|
||||
message(STATUS "GEMM_PRESHUFFLE_DATATYPE: ${GEMM_PRESHUFFLE_DATATYPE}")
|
||||
message(STATUS "GEMM_PRESHUFFLE_LAYOUT: ${GEMM_PRESHUFFLE_LAYOUT}")
|
||||
message(STATUS "SUPPORTED_GPU_TARGETS: ${SUPPORTED_GPU_TARGETS}")
|
||||
message(VERBOSE "=== Starting Tile Engine GEMM Preshuffle Configuration ===")
|
||||
message(VERBOSE "GEMM_PRESHUFFLE_DATATYPE: ${GEMM_PRESHUFFLE_DATATYPE}")
|
||||
message(VERBOSE "GEMM_PRESHUFFLE_LAYOUT: ${GEMM_PRESHUFFLE_LAYOUT}")
|
||||
message(VERBOSE "SUPPORTED_GPU_TARGETS: ${SUPPORTED_GPU_TARGETS}")
|
||||
|
||||
# Filter GPU targets to only gfx90a, gfx942, and gfx950
|
||||
set(GEMM_PRESHUFFLE_GPU_TARGETS_INDIVIDUAL "")
|
||||
@@ -221,7 +221,7 @@ set(DESIRED_TARGETS "gfx90a;gfx942;gfx950")
|
||||
foreach(target IN LISTS SUPPORTED_GPU_TARGETS)
|
||||
if(target IN_LIST DESIRED_TARGETS)
|
||||
list(APPEND GEMM_PRESHUFFLE_GPU_TARGETS_INDIVIDUAL ${target})
|
||||
message(STATUS " Adding GPU target: ${target}")
|
||||
message(VERBOSE " Adding GPU target: ${target}")
|
||||
endif()
|
||||
endforeach()
|
||||
|
||||
@@ -229,7 +229,7 @@ endforeach()
|
||||
if(NOT GEMM_PRESHUFFLE_GPU_TARGETS_INDIVIDUAL)
|
||||
message(WARNING "Skipping Tile Engine GEMM build: No supported GPU targets (gfx90a, gfx942, gfx950) found in SUPPORTED_GPU_TARGETS: ${SUPPORTED_GPU_TARGETS}")
|
||||
else()
|
||||
message(STATUS "Building individual GEMM Preshuffle targets for GPU targets: ${GEMM_PRESHUFFLE_GPU_TARGETS_INDIVIDUAL}")
|
||||
message(VERBOSE "Building individual GEMM Preshuffle targets for GPU targets: ${GEMM_PRESHUFFLE_GPU_TARGETS_INDIVIDUAL}")
|
||||
|
||||
# Enable parallel compilation optimizations
|
||||
# Set up job pools for better parallel compilation control
|
||||
@@ -244,12 +244,12 @@ else()
|
||||
find_program(CCACHE_PROGRAM ccache)
|
||||
if(CCACHE_PROGRAM)
|
||||
set(CMAKE_CXX_COMPILER_LAUNCHER ${CCACHE_PROGRAM})
|
||||
message(STATUS "Using ccache for faster compilation")
|
||||
message(VERBOSE "Using ccache for faster compilation")
|
||||
else()
|
||||
message(WARNING "ccache requested but not found")
|
||||
endif()
|
||||
else()
|
||||
message(STATUS "ccache disabled for GEMM Preshuffle ops (use -DENABLE_CCACHE_GEMM_PRESHUFFLE=ON to enable)")
|
||||
message(VERBOSE "ccache disabled for GEMM Preshuffle ops (use -DENABLE_CCACHE_GEMM_PRESHUFFLE=ON to enable)")
|
||||
endif()
|
||||
|
||||
# Create master collection targets
|
||||
|
||||
Reference in New Issue
Block a user