Build native Windows packages for both RDNA4 targets
Add explicit gfx1200/gfx1201 builds, runtime selection and per-target code objects. Guard staging, packaging and installation against mixed architectures; retain gfx1201 compatibility and strict K/M arithmetic. Adapt realdody's target configuration and MSVC binary16/BOOL portability fixes from ae530b0. Do not import its forced gfx12 layout on RDNA3. Add exhaustive binary16 and C++/PowerShell target contract regressions. Validation: 20 CTest gates on RX 9070 XT; four software-only gfx1200 gates; MSVC portability compile checks; exact 4K K async and M control outputs; isolated package install/restore and mixed-target rejection tests. No physical gfx1200 validation is claimed. Co-authored-by: realdody <[email protected]>
This commit is contained in:
1 parent
8f7058d2ba
commit
98de8e3a3e
51 files changed
+636
-127
No files matched your search
+51
-16
@@ -13,25 +13,36 @@ if(NOT EXISTS "${D4R_HIP_ROOT}/include/hip/hip_runtime_api.h")
|
||||
message(FATAL_ERROR "Set D4R_HIP_ROOT to a Windows HIP SDK (headers and bin/clang++.exe required)")
|
||||
endif()
|
||||
find_program(D4R_HIP_CLANG NAMES clang++.exe PATHS "${D4R_HIP_ROOT}/bin" "${D4R_HIP_ROOT}/lib/llvm/bin" NO_DEFAULT_PATH REQUIRED)
|
||||
set(probe_object "${CMAKE_CURRENT_BINARY_DIR}/probe_gfx1201.hsaco")
|
||||
# Architecture parameterization adapted from realdody/d4r ae530b0.
|
||||
# Keep explicit subtargets: no gfx11 substitution or untested future gfx12 ISA.
|
||||
set(D4R_GPU_ARCH "gfx1201" CACHE STRING "Native Windows package GPU target")
|
||||
set_property(CACHE D4R_GPU_ARCH PROPERTY STRINGS gfx1201 gfx1200)
|
||||
set(d4r_supported_targets gfx1200 gfx1201)
|
||||
if(NOT D4R_GPU_ARCH IN_LIST d4r_supported_targets)
|
||||
message(FATAL_ERROR "D4R_GPU_ARCH must be gfx1200 or gfx1201")
|
||||
endif()
|
||||
add_compile_definitions(D4R_TARGET_ARCH="${D4R_GPU_ARCH}")
|
||||
configure_file(tools/windows/gpu-target.json.in gpu-target.json @ONLY)
|
||||
install(FILES "${CMAKE_CURRENT_BINARY_DIR}/gpu-target.json" DESTINATION .)
|
||||
set(probe_object "${CMAKE_CURRENT_BINARY_DIR}/probe_${D4R_GPU_ARCH}.hsaco")
|
||||
add_custom_command(OUTPUT "${probe_object}"
|
||||
COMMAND "${D4R_HIP_CLANG}" -x hip --offload-arch=gfx1201
|
||||
COMMAND "${D4R_HIP_CLANG}" -x hip --offload-arch=${D4R_GPU_ARCH}
|
||||
--offload-device-only --no-gpu-bundle-output -mno-wavefrontsize64
|
||||
-nogpuinc -nogpulib -O2 "${CMAKE_CURRENT_SOURCE_DIR}/tools/windows/probe_kernel.hip"
|
||||
-o "${probe_object}"
|
||||
DEPENDS tools/windows/probe_kernel.hip VERBATIM)
|
||||
add_custom_target(d4r_probe_code_object ALL DEPENDS "${probe_object}")
|
||||
set(pixel_object "${CMAKE_CURRENT_BINARY_DIR}/pixel_convert_gfx1201.hsaco")
|
||||
set(pixel_object "${CMAKE_CURRENT_BINARY_DIR}/pixel_convert_${D4R_GPU_ARCH}.hsaco")
|
||||
add_custom_command(OUTPUT "${pixel_object}"
|
||||
COMMAND "${D4R_HIP_CLANG}" -x hip -std=c++17 --offload-arch=gfx1201
|
||||
COMMAND "${D4R_HIP_CLANG}" -x hip -std=c++17 --offload-arch=${D4R_GPU_ARCH}
|
||||
--offload-device-only --no-gpu-bundle-output -mno-wavefrontsize64
|
||||
-nogpuinc -nogpulib -O2 "${CMAKE_CURRENT_SOURCE_DIR}/tools/windows/pixel_convert_gfx1201.hip"
|
||||
-o "${pixel_object}"
|
||||
DEPENDS tools/windows/pixel_convert_gfx1201.hip tools/windows/pixel_format.h kernels/common/hip_device_minimal.h VERBATIM)
|
||||
add_custom_target(d4r_pixel_code_object ALL DEPENDS "${pixel_object}")
|
||||
set(wmma_object "${CMAKE_CURRENT_BINARY_DIR}/wmma_gfx1201.hsaco")
|
||||
set(wmma_object "${CMAKE_CURRENT_BINARY_DIR}/wmma_${D4R_GPU_ARCH}.hsaco")
|
||||
add_custom_command(OUTPUT "${wmma_object}"
|
||||
COMMAND "${D4R_HIP_CLANG}" -x hip --offload-arch=gfx1201
|
||||
COMMAND "${D4R_HIP_CLANG}" -x hip --offload-arch=${D4R_GPU_ARCH}
|
||||
--offload-device-only --no-gpu-bundle-output -mno-wavefrontsize64
|
||||
-nogpuinc -nogpulib -O2 "${CMAKE_CURRENT_SOURCE_DIR}/tools/windows/wmma_gfx1201.hip"
|
||||
-o "${wmma_object}"
|
||||
@@ -49,7 +60,7 @@ foreach(family k m)
|
||||
file(GLOB native_headers CONFIGURE_DEPENDS "${CMAKE_CURRENT_SOURCE_DIR}/kernels/${family}/*.h")
|
||||
foreach(source IN LISTS native_sources)
|
||||
get_filename_component(kernel_name "${source}" NAME_WE)
|
||||
set(native_object "${CMAKE_CURRENT_BINARY_DIR}/${kernel_name}_gfx1201.hsaco")
|
||||
set(native_object "${CMAKE_CURRENT_BINARY_DIR}/${kernel_name}_${D4R_GPU_ARCH}.hsaco")
|
||||
set_property(DIRECTORY APPEND PROPERTY CMAKE_CONFIGURE_DEPENDS "${source}")
|
||||
file(READ "${source}" kernel_source)
|
||||
string(REGEX MATCH "// d4r-build-flags: ([^\r\n]+)" flags_line "${kernel_source}")
|
||||
@@ -65,7 +76,7 @@ foreach(family k m)
|
||||
list(APPEND native_flags ${source_flags})
|
||||
endif()
|
||||
add_custom_command(OUTPUT "${native_object}"
|
||||
COMMAND "${D4R_HIP_CLANG}" -x hip -std=c++17 --offload-arch=gfx1201
|
||||
COMMAND "${D4R_HIP_CLANG}" -x hip -std=c++17 --offload-arch=${D4R_GPU_ARCH}
|
||||
--offload-device-only --no-gpu-bundle-output -mno-wavefrontsize64
|
||||
-nogpuinc -nogpulib -O3 -DD4R_DEVICE_ONLY_MINIMAL ${native_flags}
|
||||
"${source}" -o "${native_object}"
|
||||
@@ -75,7 +86,9 @@ foreach(family k m)
|
||||
install(FILES "${native_object}" DESTINATION "experimental/${family}")
|
||||
endforeach()
|
||||
endforeach()
|
||||
add_custom_target(d4r_native_gfx1201_compile ALL DEPENDS ${native_objects})
|
||||
add_custom_target(d4r_native_compile ALL DEPENDS ${native_objects})
|
||||
# Retain the old manual build target for existing scripts.
|
||||
add_custom_target(d4r_native_gfx1201_compile DEPENDS d4r_native_compile)
|
||||
foreach(probe hip_gfx1201_probe gfx12_wmma_probe k_module_probe native_replay_probe cuda_driver_probe cuda_mma_probe cuda_image_probe d3d12_command_probe d3d12_pixel_probe d3d12_hip_interop_probe hip_stream_lifecycle_probe ngx_cuda_init_probe d3d12_dlss_probe)
|
||||
add_executable(d4r_${probe} tools/windows/${probe}.cpp)
|
||||
target_compile_definitions(d4r_${probe} PRIVATE WIN32_LEAN_AND_MEAN NOMINMAX __HIP_PLATFORM_AMD__)
|
||||
@@ -92,6 +105,16 @@ foreach(probe hip_gfx1201_probe gfx12_wmma_probe k_module_probe native_replay_pr
|
||||
endif()
|
||||
install(TARGETS d4r_${probe} RUNTIME DESTINATION bin)
|
||||
endforeach()
|
||||
foreach(probe binary16_probe gpu_target_probe)
|
||||
add_executable(d4r_${probe} tools/windows/${probe}.cpp)
|
||||
if(MINGW)
|
||||
target_link_options(d4r_${probe} PRIVATE -static)
|
||||
elseif(MSVC)
|
||||
set_property(TARGET d4r_${probe} PROPERTY MSVC_RUNTIME_LIBRARY "MultiThreaded$<$<CONFIG:Debug>:Debug>")
|
||||
endif()
|
||||
install(TARGETS d4r_${probe} RUNTIME DESTINATION bin)
|
||||
endforeach()
|
||||
add_dependencies(d4r_gpu_target_probe d4r_probe_code_object d4r_pixel_code_object)
|
||||
set(D4R_MINHOOK_ROOT "${CMAKE_CURRENT_SOURCE_DIR}/external/MinHook" CACHE PATH "MinHook 1.3.4 source checkout")
|
||||
if(NOT EXISTS "${D4R_MINHOOK_ROOT}/include/MinHook.h")
|
||||
message(FATAL_ERROR "Clone MinHook v1.3.4 into external/MinHook or set D4R_MINHOOK_ROOT")
|
||||
@@ -171,11 +194,13 @@ if(EXISTS "${D4R_AGILITY_ROOT}/build/native/bin/x64/d3d12SDKLayers.dll")
|
||||
endif()
|
||||
add_dependencies(d4r_hip_gfx1201_probe d4r_probe_code_object)
|
||||
add_dependencies(d4r_gfx12_wmma_probe d4r_wmma_code_object)
|
||||
add_dependencies(d4r_k_module_probe d4r_native_gfx1201_compile)
|
||||
add_dependencies(d4r_k_module_probe d4r_native_compile)
|
||||
install(FILES "${probe_object}" DESTINATION bin)
|
||||
install(FILES "${pixel_object}" DESTINATION bin)
|
||||
install(FILES "${wmma_object}" DESTINATION bin)
|
||||
install(FILES scripts/windows/test-windows-rdna4.ps1 DESTINATION .)
|
||||
install(FILES scripts/windows/gpu-target.ps1 DESTINATION .)
|
||||
install(FILES scripts/windows/test-gpu-target-contract.ps1 DESTINATION .)
|
||||
install(FILES scripts/windows/test-k-replay.ps1 DESTINATION .)
|
||||
install(FILES scripts/windows/test-k-captures.ps1 DESTINATION .)
|
||||
install(FILES scripts/windows/test-native-overrides.ps1 DESTINATION .)
|
||||
@@ -192,13 +217,23 @@ install(FILES kernels/tools/pwin_model.py kernels/tools/pwin_check.py kernels/to
|
||||
kernels/tools/extract_dlss_ptx.py kernels/tools/kernel_manifest.py DESTINATION reference)
|
||||
install(FILES kernels/tools/model_enc3.py kernels/tools/swin_model.py kernels/tools/swin_check.py DESTINATION reference)
|
||||
install(FILES docs/windows-rdna4-port.md DESTINATION docs)
|
||||
install(FILES docs/windows-gpu-support.md DESTINATION docs)
|
||||
include(CTest)
|
||||
add_test(NAME binary16_rounding COMMAND d4r_binary16_probe)
|
||||
set_tests_properties(binary16_rounding PROPERTIES TIMEOUT 30 LABELS "software;numerics")
|
||||
add_test(NAME gpu_target_contract COMMAND d4r_gpu_target_probe "${probe_object}" "${pixel_object}")
|
||||
set_tests_properties(gpu_target_contract PROPERTIES TIMEOUT 30 LABELS "software;architecture")
|
||||
find_program(D4R_POWERSHELL NAMES pwsh.exe powershell.exe REQUIRED)
|
||||
add_test(NAME gpu_target_script_contract COMMAND "${D4R_POWERSHELL}" -NoProfile -File
|
||||
"${CMAKE_CURRENT_SOURCE_DIR}/scripts/windows/test-gpu-target-contract.ps1"
|
||||
-PackageRoot "${CMAKE_CURRENT_BINARY_DIR}" -CodeObject "${probe_object}")
|
||||
set_tests_properties(gpu_target_script_contract PROPERTIES TIMEOUT 30 LABELS "software;architecture")
|
||||
add_test(NAME d3d12_gpu_pixel_formats COMMAND d4r_d3d12_pixel_probe
|
||||
--hip-root "${D4R_HIP_ROOT}" --module "${pixel_object}" --iterations 2)
|
||||
set_tests_properties(d3d12_gpu_pixel_formats PROPERTIES TIMEOUT 120 LABELS "hardware;gfx1201;formats")
|
||||
set_tests_properties(d3d12_gpu_pixel_formats PROPERTIES TIMEOUT 120 LABELS "hardware;${D4R_GPU_ARCH};formats")
|
||||
add_test(NAME d3d12_gpu_enhanced_formats COMMAND d4r_d3d12_pixel_probe
|
||||
--hip-root "${D4R_HIP_ROOT}" --module "${pixel_object}" --interop-mode enhanced --iterations 2)
|
||||
set_tests_properties(d3d12_gpu_enhanced_formats PROPERTIES TIMEOUT 120 LABELS "hardware;gfx1201;formats;enhanced_barriers")
|
||||
set_tests_properties(d3d12_gpu_enhanced_formats PROPERTIES TIMEOUT 120 LABELS "hardware;${D4R_GPU_ARCH};formats;enhanced_barriers")
|
||||
add_test(NAME d3d12_command_backend COMMAND d4r_d3d12_command_probe
|
||||
--hip-root "${D4R_HIP_ROOT}" --module "$<TARGET_FILE:d4r_nvngx_windows>" --interop-mode indirect --iterations 32)
|
||||
set_tests_properties(d3d12_command_backend PROPERTIES TIMEOUT 120 LABELS "hardware;d3d12;command_lists;lifetime")
|
||||
@@ -211,16 +246,16 @@ set_tests_properties(d3d12_command_root_reset_warp PROPERTIES TIMEOUT 120 LABELS
|
||||
ENVIRONMENT "D4R_COMMAND_PROBE_WARP=1;D4R_COMMAND_PROBE_NO_HIP=1")
|
||||
add_test(NAME hip_gfx1201 COMMAND d4r_hip_gfx1201_probe
|
||||
--hip-root "${D4R_HIP_ROOT}" --module "${probe_object}")
|
||||
set_tests_properties(hip_gfx1201 PROPERTIES TIMEOUT 90 LABELS "hardware;gfx1201")
|
||||
set_tests_properties(hip_gfx1201 PROPERTIES TIMEOUT 90 LABELS "hardware;${D4R_GPU_ARCH}")
|
||||
add_test(NAME gfx12_wmma COMMAND d4r_gfx12_wmma_probe
|
||||
--hip-root "${D4R_HIP_ROOT}" --module "${wmma_object}" --iterations 16)
|
||||
set_tests_properties(gfx12_wmma PROPERTIES TIMEOUT 90 LABELS "hardware;gfx1201;wmma")
|
||||
set_tests_properties(gfx12_wmma PROPERTIES TIMEOUT 90 LABELS "hardware;${D4R_GPU_ARCH};wmma")
|
||||
foreach(layer enc1 enc2)
|
||||
add_test(NAME k_${layer}_execution COMMAND d4r_k_module_probe
|
||||
--hip-root "${D4R_HIP_ROOT}"
|
||||
--module "${CMAKE_CURRENT_BINARY_DIR}/dltss_pwin_${layer}_layer_gfx1201.hsaco"
|
||||
--module "${CMAKE_CURRENT_BINARY_DIR}/dltss_pwin_${layer}_layer_${D4R_GPU_ARCH}.hsaco"
|
||||
--kernel-name ${layer})
|
||||
set_tests_properties(k_${layer}_execution PROPERTIES TIMEOUT 90 LABELS "hardware;gfx1201;k_execution")
|
||||
set_tests_properties(k_${layer}_execution PROPERTIES TIMEOUT 90 LABELS "hardware;${D4R_GPU_ARCH};k_execution")
|
||||
endforeach()
|
||||
add_test(NAME d3d12_hip_interop COMMAND d4r_d3d12_hip_interop_probe
|
||||
--hip-root "${D4R_HIP_ROOT}" --module "${probe_object}" --iterations 32)
|
||||
|
||||
Reference in new issue
Block a user