ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-13 01:36:06 +00:00
Code Issues Packages Projects Releases Wiki Activity

ck: add tf32 in DTYPES to control instances build(#3317)

Browse Source
This commit is contained in:
yinglu
2025-12-08 16:24:20 +08:00
committed by GitHub
parent 86a84ae611
commit 8fec8054b2
24 changed files with 177 additions and 140 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
library/src/tensor_operation_instance/gpu/CMakeLists.txt
View File

@@ -13,6 +13,8 @@ function(add_instance_library INSTANCE_NAME)
set(type1 "_f16")
elseif(type MATCHES "fp32")
set(type1 "_f32")
elseif(type MATCHES "tf32")
set(type1 "_tf32")
elseif(type MATCHES "fp8")
set(type1 "_f8")
elseif(type MATCHES "bf16")
@@ -27,8 +29,8 @@ function(add_instance_library INSTANCE_NAME)
#if filename matches any selected type, exit type loop and do no exclude the file from the list
set(test 0)
break()
elseif((source_name MATCHES "fp8" OR source_name MATCHES "fp32" OR source_name MATCHES "fp64" OR source_name MATCHES "bf16" OR source_name MATCHES "int8" OR source_name MATCHES "fp16" OR
source_name MATCHES "_f8" OR source_name MATCHES "_f32" OR source_name MATCHES "_f64" OR source_name MATCHES "_i8" OR source_name MATCHES "_f16" OR source_name MATCHES "_b16") AND
elseif((source_name MATCHES "fp8" OR source_name MATCHES "fp32" OR source_name MATCHES "tf32" OR source_name MATCHES "fp64" OR source_name MATCHES "bf16" OR source_name MATCHES "int8" OR source_name MATCHES "fp16" OR
source_name MATCHES "_f8" OR source_name MATCHES "_f32" OR source_name MATCHES "_tf32" OR source_name MATCHES "_f64" OR source_name MATCHES "_i8" OR source_name MATCHES "_f16" OR source_name MATCHES "_b16") AND
NOT (source_name MATCHES type OR source_name MATCHES type1))
#if filename contains a type which doesn't match any selected type, mark it for removal
set(test 1)
@@ -102,9 +104,11 @@ function(add_instance_library INSTANCE_NAME)
list(REMOVE_ITEM ARGN "${source}")
endif()
# Only build tf32 instances for gfx942 & gfx950
if(NOT (INST_TARGETS MATCHES "gfx942|gfx950") AND source_name MATCHES "_tf32_")
message(DEBUG "removing tf32 instance ${source} ")
list(REMOVE_ITEM ARGN "${source}")
if(source_name MATCHES "_tf32_")
if(NOT ((INST_TARGETS MATCHES "gfx942|gfx950") AND CK_ENABLE_TF32))
message(DEBUG "removing tf32 instance ${source} ")
list(REMOVE_ITEM ARGN "${source}")
endif()
endif()
endforeach()
@@ -223,6 +227,10 @@ FOREACH(subdir_path ${dir_list})
message(DEBUG "fp32 instance found!")
set(add_inst 1)
endif()
if(("${cmake_instance}" MATCHES "_tf32" OR "${cmake_instance}" MATCHES "_tf32") AND DTYPES MATCHES "tf32")
message(DEBUG "tf32 instance found!")
set(add_inst 1)
endif()
if(("${cmake_instance}" MATCHES "_fp64" OR "${cmake_instance}" MATCHES "_f64") AND DTYPES MATCHES "fp64")
message(DEBUG "fp64 instance found!")
set(add_inst 1)
@@ -237,6 +245,7 @@ FOREACH(subdir_path ${dir_list})
"${cmake_instance}" MATCHES "_f16" OR
"${cmake_instance}" MATCHES "_fp32" OR
"${cmake_instance}" MATCHES "_f32" OR
"${cmake_instance}" MATCHES "_tf32" OR
"${cmake_instance}" MATCHES "_fp64" OR
"${cmake_instance}" MATCHES "_f64" OR
"${cmake_instance}" MATCHES "_bf16" OR
@@ -330,7 +339,7 @@ FOREACH(subdir_path ${dir_list})
list(APPEND CK_DEVICE_OTHER_INSTANCES $<TARGET_OBJECTS:device_${target_dir}_instance>)
endif()
message(DEBUG "add_instance_directory ${subdir_path}")
endif()
endif()
else()
message(DEBUG "skip_instance_directory ${subdir_path}")
endif()