Skip to content

ROCm Compile Time Reduction #966

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
wants to merge 2 commits into
base: develop
Choose a base branch
from
Open

ROCm Compile Time Reduction #966

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
9d8a30c
ROCm Compile Time Reduction
Spaarsh Oct 5, 2025
de9af17
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot] Oct 18, 2025
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
113 changes: 92 additions & 21 deletions CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -101,6 +101,45 @@ set(USE_ACCEL
CACHE STRING "Build with acceleration support (default: none)")
set_property(CACHE USE_ACCEL PROPERTY STRINGS "" opencl cuda hip)

# Add new multi-architecture option
option(MULTI_ARCH "Build for multiple GPU architectures" OFF)

# Multi-architecture support (only when enabled)
if (MULTI_ARCH AND MULTI_GPU_BUILD)
set(WITH_GPU_LIST
"P100"
CACHE STRING
"List of GPU architectures to build for (semicolon-separated)")

# Define supported architectures for the property
set(SUPPORTED_CUDA_ARCHITECTURES
K20X
K40
K80
P100
V100
A100
H100)
set(SUPPORTED_HIP_ARCHITECTURES Mi50 Mi100 Mi250 Mi300)

set_property(
CACHE WITH_GPU_LIST PROPERTY STRINGS ${SUPPORTED_CUDA_ARCHITECTURES}
${SUPPORTED_HIP_ARCHITECTURES})

# Parse the list
string(REPLACE ";" " " WITH_GPU_LIST_STR "${WITH_GPU_LIST}")
list(LENGTH WITH_GPU_LIST GPU_COUNT)

if (GPU_COUNT GREATER 1)
set(MULTI_GPU_BUILD ON)
message(STATUS "Multi-GPU build enabled for: ${WITH_GPU_LIST_STR}")
else ()
set(MULTI_GPU_BUILD OFF)
list(GET WITH_GPU_LIST 0 WITH_GPU)
message(STATUS "Single GPU build for: ${WITH_GPU}")
endif ()
endif ()

set(SUPPORTED_CUDA_ARCHITECTURES
K20X
K40
Expand All @@ -110,16 +149,24 @@ set(SUPPORTED_CUDA_ARCHITECTURES
A100
H100)
set(SUPPORTED_HIP_ARCHITECTURES Mi50 Mi100 Mi250 Mi300)
set(WITH_GPU
$<IF:$<STREQUAL:${USE_ACCEL},"opencl">,"","P100">
CACHE
STRING
"Select GPU arch. and embed parameters (default: CUDA/HIP=P100, OPENCL=all)"
)
set(WITH_GPU_PARAMS "${WITH_GPU}")
set_property(CACHE WITH_GPU PROPERTY STRINGS ${SUPPORTED_CUDA_ARCHITECTURES}
${SUPPORTED_HIP_ARCHITECTURES})

if (NOT MULTI_ARCH)
set(WITH_GPU
$<IF:$<STREQUAL:${USE_ACCEL},"opencl">,"","P100">
CACHE
STRING
"Select GPU arch. and embed parameters (default: CUDA/HIP=P100, OPENCL=all)"
)
set(WITH_GPU_PARAMS "${WITH_GPU}")
set_property(CACHE WITH_GPU PROPERTY STRINGS ${SUPPORTED_CUDA_ARCHITECTURES}
${SUPPORTED_HIP_ARCHITECTURES})
else ()
# For multi-arch builds, set WITH_GPU_PARAMS to first architecture for
# compatibility
list(GET WITH_GPU_LIST 0 WITH_GPU_PARAMS)
list(GET WITH_GPU_LIST 0 WITH_GPU) # Set WITH_GPU for compatibility with
# existing code
endif ()
option(WITH_CUDA_PROFILING "Enable profiling within CUDA" OFF)
option(WITH_HIP_PROFILING "Enable profiling within HIP" OFF)

Expand Down Expand Up @@ -298,12 +345,42 @@ if (USE_ACCEL MATCHES "hip")
endif ()
enable_language(HIP)

# Make sure the GPU required is supported
list(FIND SUPPORTED_HIP_ARCHITECTURES ${WITH_GPU} GPU_SUPPORTED)
if (GPU_SUPPORTED EQUAL -1)
message(
FATAL_ERROR "GPU architecture requested (${WITH_GPU}) is not supported. "
"Please choose from: ${SUPPORTED_HIP_ARCHITECTURES}")
if (MULTI_ARCH AND MULTI_GPU_BUILD)
# Validate all GPU architectures in the list
foreach (GPU_ARCH IN LISTS WITH_GPU_LIST)
list(FIND SUPPORTED_HIP_ARCHITECTURES ${GPU_ARCH} GPU_SUPPORTED)
if (GPU_SUPPORTED EQUAL -1)
message(
FATAL_ERROR
"GPU architecture requested (${GPU_ARCH}) is not supported. "
"Please choose from: ${SUPPORTED_HIP_ARCHITECTURES}")
endif ()
endforeach ()

set(ACC_ARCH_NUMBERS "")
foreach (GPU_ARCH IN LISTS WITH_GPU_LIST)
list(APPEND ACC_ARCH_NUMBERS ${GPU_ARCH_NUMBER_${GPU_ARCH}})
endforeach ()

message(STATUS "Multi-GPU HIP build for architectures: ${WITH_GPU_LIST}")
message(STATUS "HIP architecture numbers: ${ACC_ARCH_NUMBERS}")
message(STATUS "Kernel parameters will be generated for: ${WITH_GPU_LIST}")

else ()
# Make sure the GPU required is supported
list(FIND SUPPORTED_HIP_ARCHITECTURES ${WITH_GPU} GPU_SUPPORTED)
if (GPU_SUPPORTED EQUAL -1)
message(
FATAL_ERROR
"GPU architecture requested (${WITH_GPU}) is not supported. "
"Please choose from: ${SUPPORTED_HIP_ARCHITECTURES}")
endif ()

set(ACC_ARCH_NUMBER ${GPU_ARCH_NUMBER_${WITH_GPU}})
message(STATUS "GPU target architecture: " ${WITH_GPU})
message(STATUS "Kernel parameters: " ${WITH_GPU_PARAMS})
message(STATUS "GPU architecture number: " ${ACC_ARCH_NUMBER})
message(STATUS "GPU profiling enabled: " ${WITH_HIP_PROFILING})
endif ()

# ROCm is typically installed in /opt/rocm; otherwise let the user set
Expand All @@ -329,12 +406,6 @@ if (USE_ACCEL MATCHES "hip")
message(FATAL_ERROR "HIP version >= 4.4.0 is required.")
endif ()

set(ACC_ARCH_NUMBER ${GPU_ARCH_NUMBER_${WITH_GPU}})
message(STATUS "GPU target architecture: " ${WITH_GPU})
message(STATUS "Kernel parameters: " ${WITH_GPU_PARAMS})
message(STATUS "GPU architecture number: " ${ACC_ARCH_NUMBER})
message(STATUS "GPU profiling enabled: " ${WITH_HIP_PROFILING})

# =================================== BLAS on GPU backend
find_package(hipblas CONFIG REQUIRED HINTS ${ROCM_PATH})
endif ()
Expand Down
Loading