forked from bartvdbraak/blender
9815f8a623
The idea is to switch from allocating separate buffers for shader data's structure of arrays to allocating one huge memory block and do some index trickery to make it accessed as SOA. This saves quite reasonable amount of lines of code in device_opencl and also makes it possible to get rid of special declaration of ShaderData structure. As a side effect it also makes it easier to experiment with SOA vs. AOS for split kernel. Works fine here on NVidia GTX580, Intel CPU amd AMD Fiji cards. Reviewers: #cycles, brecht, juicyfruit, dingto Differential Revision: https://developer.blender.org/D1593
349 lines
10 KiB
CMake
349 lines
10 KiB
CMake
remove_extra_strict_flags()
|
|
|
|
set(INC
|
|
.
|
|
../util
|
|
osl
|
|
svm
|
|
)
|
|
|
|
set(INC_SYS
|
|
|
|
)
|
|
|
|
set(SRC
|
|
kernels/cpu/kernel.cpp
|
|
kernels/opencl/kernel.cl
|
|
kernels/opencl/kernel_data_init.cl
|
|
kernels/opencl/kernel_queue_enqueue.cl
|
|
kernels/opencl/kernel_scene_intersect.cl
|
|
kernels/opencl/kernel_lamp_emission.cl
|
|
kernels/opencl/kernel_background_buffer_update.cl
|
|
kernels/opencl/kernel_shader_eval.cl
|
|
kernels/opencl/kernel_holdout_emission_blurring_pathtermination_ao.cl
|
|
kernels/opencl/kernel_direct_lighting.cl
|
|
kernels/opencl/kernel_shadow_blocked.cl
|
|
kernels/opencl/kernel_next_iteration_setup.cl
|
|
kernels/opencl/kernel_sum_all_radiance.cl
|
|
kernels/cuda/kernel.cu
|
|
)
|
|
|
|
set(SRC_HEADERS
|
|
kernel_accumulate.h
|
|
kernel_bake.h
|
|
kernel_camera.h
|
|
kernel_compat_cpu.h
|
|
kernel_compat_cuda.h
|
|
kernel_compat_opencl.h
|
|
kernel_debug.h
|
|
kernel_differential.h
|
|
kernel_emission.h
|
|
kernel_film.h
|
|
kernel_globals.h
|
|
kernel_jitter.h
|
|
kernel_light.h
|
|
kernel_math.h
|
|
kernel_montecarlo.h
|
|
kernel_passes.h
|
|
kernel_path.h
|
|
kernel_path_branched.h
|
|
kernel_path_common.h
|
|
kernel_path_state.h
|
|
kernel_path_surface.h
|
|
kernel_path_volume.h
|
|
kernel_projection.h
|
|
kernel_queues.h
|
|
kernel_random.h
|
|
kernel_shader.h
|
|
kernel_shadow.h
|
|
kernel_subsurface.h
|
|
kernel_textures.h
|
|
kernel_types.h
|
|
kernel_volume.h
|
|
kernel_work_stealing.h
|
|
)
|
|
|
|
set(SRC_KERNELS_CPU_HEADERS
|
|
kernel.h
|
|
kernels/cpu/kernel_cpu.h
|
|
kernels/cpu/kernel_cpu_impl.h
|
|
)
|
|
|
|
set(SRC_CLOSURE_HEADERS
|
|
closure/bsdf.h
|
|
closure/bsdf_ashikhmin_velvet.h
|
|
closure/bsdf_diffuse.h
|
|
closure/bsdf_diffuse_ramp.h
|
|
closure/bsdf_microfacet.h
|
|
closure/bsdf_oren_nayar.h
|
|
closure/bsdf_phong_ramp.h
|
|
closure/bsdf_reflection.h
|
|
closure/bsdf_refraction.h
|
|
closure/bsdf_toon.h
|
|
closure/bsdf_transparent.h
|
|
closure/bsdf_util.h
|
|
closure/bsdf_ashikhmin_shirley.h
|
|
closure/bsdf_hair.h
|
|
closure/bssrdf.h
|
|
closure/emissive.h
|
|
closure/volume.h
|
|
)
|
|
|
|
set(SRC_SVM_HEADERS
|
|
svm/svm.h
|
|
svm/svm_attribute.h
|
|
svm/svm_blackbody.h
|
|
svm/svm_camera.h
|
|
svm/svm_closure.h
|
|
svm/svm_convert.h
|
|
svm/svm_checker.h
|
|
svm/svm_brick.h
|
|
svm/svm_displace.h
|
|
svm/svm_fresnel.h
|
|
svm/svm_wireframe.h
|
|
svm/svm_wavelength.h
|
|
svm/svm_gamma.h
|
|
svm/svm_brightness.h
|
|
svm/svm_geometry.h
|
|
svm/svm_gradient.h
|
|
svm/svm_hsv.h
|
|
svm/svm_image.h
|
|
svm/svm_invert.h
|
|
svm/svm_light_path.h
|
|
svm/svm_magic.h
|
|
svm/svm_mapping.h
|
|
svm/svm_math.h
|
|
svm/svm_math_util.h
|
|
svm/svm_mix.h
|
|
svm/svm_musgrave.h
|
|
svm/svm_noise.h
|
|
svm/svm_noisetex.h
|
|
svm/svm_normal.h
|
|
svm/svm_ramp.h
|
|
svm/svm_sepcomb_hsv.h
|
|
svm/svm_sepcomb_vector.h
|
|
svm/svm_sky.h
|
|
svm/svm_tex_coord.h
|
|
svm/svm_texture.h
|
|
svm/svm_types.h
|
|
svm/svm_value.h
|
|
svm/svm_vector_transform.h
|
|
svm/svm_voronoi.h
|
|
svm/svm_voxel.h
|
|
svm/svm_wave.h
|
|
)
|
|
|
|
set(SRC_GEOM_HEADERS
|
|
geom/geom.h
|
|
geom/geom_attribute.h
|
|
geom/geom_bvh.h
|
|
geom/geom_bvh_shadow.h
|
|
geom/geom_bvh_subsurface.h
|
|
geom/geom_bvh_traversal.h
|
|
geom/geom_bvh_volume.h
|
|
geom/geom_bvh_volume_all.h
|
|
geom/geom_curve.h
|
|
geom/geom_motion_curve.h
|
|
geom/geom_motion_triangle.h
|
|
geom/geom_object.h
|
|
geom/geom_primitive.h
|
|
geom/geom_qbvh.h
|
|
geom/geom_qbvh_shadow.h
|
|
geom/geom_qbvh_subsurface.h
|
|
geom/geom_qbvh_traversal.h
|
|
geom/geom_qbvh_volume.h
|
|
geom/geom_qbvh_volume_all.h
|
|
geom/geom_triangle.h
|
|
geom/geom_triangle_intersect.h
|
|
geom/geom_volume.h
|
|
)
|
|
|
|
set(SRC_UTIL_HEADERS
|
|
../util/util_atomic.h
|
|
../util/util_color.h
|
|
../util/util_half.h
|
|
../util/util_math.h
|
|
../util/util_math_fast.h
|
|
../util/util_transform.h
|
|
../util/util_types.h
|
|
)
|
|
|
|
set(SRC_SPLIT_HEADERS
|
|
split/kernel_background_buffer_update.h
|
|
split/kernel_data_init.h
|
|
split/kernel_direct_lighting.h
|
|
split/kernel_holdout_emission_blurring_pathtermination_ao.h
|
|
split/kernel_lamp_emission.h
|
|
split/kernel_next_iteration_setup.h
|
|
split/kernel_scene_intersect.h
|
|
split/kernel_shader_eval.h
|
|
split/kernel_shadow_blocked.h
|
|
split/kernel_split_common.h
|
|
split/kernel_sum_all_radiance.h
|
|
)
|
|
|
|
# CUDA module
|
|
|
|
if(WITH_CYCLES_CUDA_BINARIES)
|
|
# 32 bit or 64 bit
|
|
if("${CMAKE_SIZEOF_VOID_P}" EQUAL "8")
|
|
set(CUDA_BITS 64)
|
|
else()
|
|
set(CUDA_BITS 32)
|
|
endif()
|
|
|
|
# CUDA version
|
|
execute_process(COMMAND ${CUDA_NVCC_EXECUTABLE} "--version" OUTPUT_VARIABLE NVCC_OUT)
|
|
string(REGEX REPLACE ".*release ([0-9]+)\\.([0-9]+).*" "\\1" CUDA_VERSION_MAJOR ${NVCC_OUT})
|
|
string(REGEX REPLACE ".*release ([0-9]+)\\.([0-9]+).*" "\\2" CUDA_VERSION_MINOR ${NVCC_OUT})
|
|
set(CUDA_VERSION "${CUDA_VERSION_MAJOR}${CUDA_VERSION_MINOR}")
|
|
|
|
# warn for other versions
|
|
if(CUDA_VERSION MATCHES "65")
|
|
else()
|
|
message(WARNING
|
|
"CUDA version ${CUDA_VERSION_MAJOR}.${CUDA_VERSION_MINOR} detected, "
|
|
"build may succeed but only CUDA 6.5 is officially supported")
|
|
endif()
|
|
|
|
# build for each arch
|
|
set(cuda_sources kernels/cuda/kernel.cu ${SRC_HEADERS} ${SRC_SVM_HEADERS} ${SRC_GEOM_HEADERS} ${SRC_CLOSURE_HEADERS} ${SRC_UTIL_HEADERS})
|
|
set(cuda_cubins)
|
|
|
|
macro(CYCLES_CUDA_KERNEL_ADD arch experimental)
|
|
if(${experimental})
|
|
set(cuda_extra_flags "-D__KERNEL_EXPERIMENTAL__")
|
|
set(cuda_cubin kernel_experimental_${arch}.cubin)
|
|
else()
|
|
set(cuda_extra_flags "")
|
|
set(cuda_cubin kernel_${arch}.cubin)
|
|
endif()
|
|
|
|
if(WITH_CYCLES_DEBUG)
|
|
set(cuda_debug_flags "-D__KERNEL_DEBUG__")
|
|
else()
|
|
set(cuda_debug_flags "")
|
|
endif()
|
|
|
|
set(cuda_version_flags "-D__KERNEL_CUDA_VERSION__=${CUDA_VERSION}")
|
|
set(cuda_math_flags "--use_fast_math")
|
|
|
|
add_custom_command(
|
|
OUTPUT ${cuda_cubin}
|
|
COMMAND ${CUDA_NVCC_EXECUTABLE}
|
|
-arch=${arch}
|
|
-m${CUDA_BITS}
|
|
--cubin ${CMAKE_CURRENT_SOURCE_DIR}/kernels/cuda/kernel.cu
|
|
-o ${CMAKE_CURRENT_BINARY_DIR}/${cuda_cubin}
|
|
--ptxas-options="-v"
|
|
${cuda_arch_flags}
|
|
${cuda_version_flags}
|
|
${cuda_math_flags}
|
|
${cuda_extra_flags}
|
|
${cuda_debug_flags}
|
|
-I${CMAKE_CURRENT_SOURCE_DIR}/../util
|
|
-I${CMAKE_CURRENT_SOURCE_DIR}/svm
|
|
-DCCL_NAMESPACE_BEGIN=
|
|
-DCCL_NAMESPACE_END=
|
|
-DNVCC
|
|
|
|
DEPENDS ${cuda_sources})
|
|
|
|
delayed_install("${CMAKE_CURRENT_BINARY_DIR}" "${cuda_cubin}" ${CYCLES_INSTALL_PATH}/lib)
|
|
list(APPEND cuda_cubins ${cuda_cubin})
|
|
|
|
unset(cuda_extra_flags)
|
|
unset(cuda_debug_flags)
|
|
endmacro()
|
|
|
|
foreach(arch ${CYCLES_CUDA_BINARIES_ARCH})
|
|
# Compile regular kernel
|
|
CYCLES_CUDA_KERNEL_ADD(${arch} FALSE)
|
|
endforeach()
|
|
|
|
add_custom_target(cycles_kernel_cuda ALL DEPENDS ${cuda_cubins})
|
|
endif()
|
|
|
|
# OSL module
|
|
|
|
if(WITH_CYCLES_OSL)
|
|
add_subdirectory(osl)
|
|
add_subdirectory(shaders)
|
|
endif()
|
|
|
|
# CPU module
|
|
|
|
include_directories(${INC})
|
|
include_directories(SYSTEM ${INC_SYS})
|
|
|
|
if(CXX_HAS_SSE)
|
|
list(APPEND SRC
|
|
kernels/cpu/kernel_sse2.cpp
|
|
kernels/cpu/kernel_sse3.cpp
|
|
kernels/cpu/kernel_sse41.cpp
|
|
)
|
|
|
|
set_source_files_properties(kernels/cpu/kernel_sse2.cpp PROPERTIES COMPILE_FLAGS "${CYCLES_SSE2_KERNEL_FLAGS}")
|
|
set_source_files_properties(kernels/cpu/kernel_sse3.cpp PROPERTIES COMPILE_FLAGS "${CYCLES_SSE3_KERNEL_FLAGS}")
|
|
set_source_files_properties(kernels/cpu/kernel_sse41.cpp PROPERTIES COMPILE_FLAGS "${CYCLES_SSE41_KERNEL_FLAGS}")
|
|
endif()
|
|
|
|
if(CXX_HAS_AVX)
|
|
list(APPEND SRC
|
|
kernels/cpu/kernel_avx.cpp
|
|
)
|
|
set_source_files_properties(kernels/cpu/kernel_avx.cpp PROPERTIES COMPILE_FLAGS "${CYCLES_AVX_KERNEL_FLAGS}")
|
|
endif()
|
|
|
|
if(CXX_HAS_AVX2)
|
|
list(APPEND SRC
|
|
kernels/cpu/kernel_avx2.cpp
|
|
)
|
|
set_source_files_properties(kernels/cpu/kernel_avx2.cpp PROPERTIES COMPILE_FLAGS "${CYCLES_AVX2_KERNEL_FLAGS}")
|
|
endif()
|
|
|
|
add_library(cycles_kernel
|
|
${SRC}
|
|
${SRC_HEADERS}
|
|
${SRC_KERNELS_CPU_HEADERS}
|
|
${SRC_CLOSURE_HEADERS}
|
|
${SRC_SVM_HEADERS}
|
|
${SRC_GEOM_HEADERS}
|
|
${SRC_SPLIT_HEADERS}
|
|
)
|
|
|
|
if(WITH_CYCLES_CUDA)
|
|
add_dependencies(cycles_kernel cycles_kernel_cuda)
|
|
endif()
|
|
|
|
# OpenCL kernel
|
|
|
|
#set(KERNEL_PREPROCESSED ${CMAKE_CURRENT_BINARY_DIR}/kernel_preprocessed.cl)
|
|
#add_custom_command(
|
|
# OUTPUT ${KERNEL_PREPROCESSED}
|
|
# COMMAND gcc -x c++ -E ${CMAKE_CURRENT_SOURCE_DIR}/kernel.cl -I ${CMAKE_CURRENT_SOURCE_DIR}/../util/ -DCCL_NAMESPACE_BEGIN= -DCCL_NAMESPACE_END= -o ${KERNEL_PREPROCESSED}
|
|
# DEPENDS ${SRC_KERNEL} ${SRC_UTIL_HEADERS})
|
|
#add_custom_target(cycles_kernel_preprocess ALL DEPENDS ${KERNEL_PREPROCESSED})
|
|
#delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "${KERNEL_PREPROCESSED}" ${CYCLES_INSTALL_PATH}/kernel)
|
|
|
|
delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "kernels/opencl/kernel.cl" ${CYCLES_INSTALL_PATH}/kernel/kernels/opencl)
|
|
delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "kernels/opencl/kernel_data_init.cl" ${CYCLES_INSTALL_PATH}/kernel/kernels/opencl)
|
|
delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "kernels/opencl/kernel_queue_enqueue.cl" ${CYCLES_INSTALL_PATH}/kernel/kernels/opencl)
|
|
delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "kernels/opencl/kernel_scene_intersect.cl" ${CYCLES_INSTALL_PATH}/kernel/kernels/opencl)
|
|
delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "kernels/opencl/kernel_lamp_emission.cl" ${CYCLES_INSTALL_PATH}/kernel/kernels/opencl)
|
|
delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "kernels/opencl/kernel_background_buffer_update.cl" ${CYCLES_INSTALL_PATH}/kernel/kernels/opencl)
|
|
delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "kernels/opencl/kernel_shader_eval.cl" ${CYCLES_INSTALL_PATH}/kernel/kernels/opencl)
|
|
delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "kernels/opencl/kernel_holdout_emission_blurring_pathtermination_ao.cl" ${CYCLES_INSTALL_PATH}/kernel/kernels/opencl)
|
|
delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "kernels/opencl/kernel_direct_lighting.cl" ${CYCLES_INSTALL_PATH}/kernel/kernels/opencl)
|
|
delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "kernels/opencl/kernel_shadow_blocked.cl" ${CYCLES_INSTALL_PATH}/kernel/kernels/opencl)
|
|
delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "kernels/opencl/kernel_next_iteration_setup.cl" ${CYCLES_INSTALL_PATH}/kernel/kernels/opencl)
|
|
delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "kernels/opencl/kernel_sum_all_radiance.cl" ${CYCLES_INSTALL_PATH}/kernel/kernels/opencl)
|
|
delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "kernels/cuda/kernel.cu" ${CYCLES_INSTALL_PATH}/kernel/kernels/cuda)
|
|
delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "${SRC_HEADERS}" ${CYCLES_INSTALL_PATH}/kernel)
|
|
delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "${SRC_CLOSURE_HEADERS}" ${CYCLES_INSTALL_PATH}/kernel/closure)
|
|
delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "${SRC_SVM_HEADERS}" ${CYCLES_INSTALL_PATH}/kernel/svm)
|
|
delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "${SRC_GEOM_HEADERS}" ${CYCLES_INSTALL_PATH}/kernel/geom)
|
|
delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "${SRC_UTIL_HEADERS}" ${CYCLES_INSTALL_PATH}/kernel)
|
|
delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "${SRC_SPLIT_HEADERS}" ${CYCLES_INSTALL_PATH}/kernel/split)
|
|
|