Cycles: add HIP device support for AMD GPUs

NOTE: this feature is not ready for user testing, and not yet enabled in daily
builds. It is being merged now for easier collaboration on development.

HIP is a heterogenous compute interface allowing C++ code to be executed on
GPUs similar to CUDA. It is intended to bring back AMD GPU rendering support
on Windows and Linux.

https://github.com/ROCm-Developer-Tools/HIP.

As of the time of writing, it should compile and run on Linux with existing
HIP compilers and driver runtimes. Publicly available compilers and drivers
for Windows will come later.

See task T91571 for more details on the current status and work remaining
to be done.

Credits:

Sayak Biswas (AMD)
Arya Rafii (AMD)
Brian Savery (AMD)

Differential Revision: https://developer.blender.org/D12578

1 files changed, 116 insertions, 0 deletions

diff --git a/intern/cycles/kernel/CMakeLists.txt b/intern/cycles/kernel/CMakeLists.txt
index 4196539a9b1..8c2cb2c68de 100644
--- a/intern/cycles/kernel/CMakeLists.txt
+++ b/intern/cycles/kernel/CMakeLists.txt
@@ -35,6 +35,10 @@ set(SRC_DEVICE_CUDA
   device/cuda/kernel.cu
 )
 
+set(SRC_DEVICE_HIP
+  device/hip/kernel.cpp
+)
+
 set(SRC_DEVICE_OPTIX
   device/optix/kernel.cu
   device/optix/kernel_shader_raytrace.cu
@@ -106,6 +110,12 @@ set(SRC_DEVICE_CUDA_HEADERS
   device/cuda/globals.h
 )
 
+set(SRC_DEVICE_HIP_HEADERS
+  device/hip/compat.h
+  device/hip/config.h
+  device/hip/globals.h
+)
+
 set(SRC_DEVICE_OPTIX_HEADERS
   device/optix/compat.h
   device/optix/globals.h
@@ -458,6 +468,104 @@ if(WITH_CYCLES_CUDA_BINARIES)
   cycles_set_solution_folder(cycles_kernel_cuda)
 endif()
 
+####################################################### START
+
+# HIP module
+
+if(WITH_CYCLES_HIP_BINARIES)
+  # 64 bit only
+  set(HIP_BITS 64)
+
+  # HIP version
+  execute_process(COMMAND ${HIP_HIPCC_EXECUTABLE} "--version" OUTPUT_VARIABLE HIPCC_OUT)
+  string(REGEX REPLACE ".*release ([0-9]+)\\.([0-9]+).*" "\\1" HIP_VERSION_MAJOR "${HIPCC_OUT}")
+  string(REGEX REPLACE ".*release ([0-9]+)\\.([0-9]+).*" "\\2" HIP_VERSION_MINOR "${HIPCC_OUT}")
+  set(HIP_VERSION "${HIP_VERSION_MAJOR}${HIP_VERSION_MINOR}")
+
+
+  message(WARNING
+    "HIP version ${HIP_VERSION_MAJOR}.${HIP_VERSION_MINOR} detected")
+
+  # build for each arch
+  set(hip_sources device/hip/kernel.cpp
+    ${SRC_HEADERS}
+    ${SRC_DEVICE_HIP_HEADERS}
+    ${SRC_BVH_HEADERS}
+    ${SRC_SVM_HEADERS}
+    ${SRC_GEOM_HEADERS}
+    ${SRC_INTEGRATOR_HEADERS}
+    ${SRC_CLOSURE_HEADERS}
+    ${SRC_UTIL_HEADERS}
+  )
+  set(hip_fatbins)
+
+  macro(CYCLES_HIP_KERNEL_ADD arch prev_arch name flags sources experimental)
+    if(${arch} MATCHES "compute_.*")
+      set(format "ptx")
+    else()
+      set(format "fatbin")
+    endif()
+    set(hip_file ${name}_${arch}.${format})
+
+    set(kernel_sources ${sources})
+    if(NOT ${prev_arch} STREQUAL "none")
+      if(${prev_arch} MATCHES "compute_.*")
+        set(kernel_sources ${kernel_sources} ${name}_${prev_arch}.ptx)
+      else()
+        set(kernel_sources ${kernel_sources} ${name}_${prev_arch}.fatbin)
+      endif()
+    endif()
+
+    set(hip_kernel_src "/device/hip/${name}.cpp")
+
+    set(hip_flags ${flags}
+      -D CCL_NAMESPACE_BEGIN=
+      -D CCL_NAMESPACE_END=
+      -D HIPCC
+      -m ${HIP_BITS}
+      -I ${CMAKE_CURRENT_SOURCE_DIR}/..
+      -I ${CMAKE_CURRENT_SOURCE_DIR}/device/hip
+      --use_fast_math
+      -o ${CMAKE_CURRENT_BINARY_DIR}/${hip_file})
+
+    if(${experimental})
+      set(hip_flags ${hip_flags} -D __KERNEL_EXPERIMENTAL__)
+      set(name ${name}_experimental)
+    endif()
+
+    if(WITH_CYCLES_DEBUG)
+      set(hip_flags ${hip_flags} -D __KERNEL_DEBUG__)
+    endif()
+
+    if(WITH_NANOVDB)
+      set(hip_flags ${hip_flags}
+        -D WITH_NANOVDB
+        -I "${NANOVDB_INCLUDE_DIR}")
+    endif()
+
+
+  set(prev_arch "none")
+  foreach(arch ${CYCLES_HIP_BINARIES_ARCH})
+      set(hip_hipcc_executable ${HIP_HIPCC_EXECUTABLE})
+      set(hip_toolkit_root_dir ${HIP_TOOLKIT_ROOT_DIR})
+    if(DEFINED hip_hipcc_executable AND DEFINED hip_toolkit_root_dir)
+      # Compile regular kernel
+      CYCLES_HIP_KERNEL_ADD(${arch} ${prev_arch} kernel "" "${hip_sources}" FALSE)
+
+      if(WITH_CYCLES_HIP_BUILD_SERIAL)
+        set(prev_arch ${arch})
+      endif()
+
+      unset(hip_hipcc_executable)
+      unset(hip_toolkit_root_dir)
+    endif()
+  endforeach()
+
+  add_custom_target(cycles_kernel_hip ALL DEPENDS ${hip_fatbins})
+  cycles_set_solution_folder(cycles_kernel_hip)
+endif()
+
+####################################################### END
 # OptiX PTX modules
 
 if(WITH_CYCLES_DEVICE_OPTIX AND WITH_CYCLES_CUDA_BINARIES)
@@ -602,11 +710,13 @@ endif()
 cycles_add_library(cycles_kernel "${LIB}"
   ${SRC_DEVICE_CPU}
   ${SRC_DEVICE_CUDA}
+  ${SRC_DEVICE_HIP}
   ${SRC_DEVICE_OPTIX}
   ${SRC_HEADERS}
   ${SRC_DEVICE_CPU_HEADERS}
   ${SRC_DEVICE_GPU_HEADERS}
   ${SRC_DEVICE_CUDA_HEADERS}
+  ${SRC_DEVICE_HIP_HEADERS}
   ${SRC_DEVICE_OPTIX_HEADERS}
   ${SRC_BVH_HEADERS}
   ${SRC_CLOSURE_HEADERS}
@@ -621,6 +731,7 @@ source_group("geom" FILES ${SRC_GEOM_HEADERS})
 source_group("integrator" FILES ${SRC_INTEGRATOR_HEADERS})
 source_group("kernel" FILES ${SRC_HEADERS})
 source_group("device\\cpu" FILES ${SRC_DEVICE_CPU} ${SRC_DEVICE_CPU_HEADERS})
+source_group("device\\hip" FILES ${SRC_DEVICE_HIP} ${SRC_DEVICE_HIP_HEADERS})
 source_group("device\\gpu" FILES ${SRC_DEVICE_GPU_HEADERS})
 source_group("device\\cuda" FILES ${SRC_DEVICE_CUDA} ${SRC_DEVICE_CUDA_HEADERS})
 source_group("device\\optix" FILES ${SRC_DEVICE_OPTIX} ${SRC_DEVICE_OPTIX_HEADERS})
@@ -632,14 +743,19 @@ endif()
 if(WITH_CYCLES_DEVICE_OPTIX AND WITH_CYCLES_CUDA_BINARIES)
   add_dependencies(cycles_kernel cycles_kernel_optix)
 endif()
+if(WITH_CYCLES_HIP)
+  add_dependencies(cycles_kernel cycles_kernel_hip)
+endif()
 
 # Install kernel source for runtime compilation
 
 delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "${SRC_DEVICE_CUDA}" ${CYCLES_INSTALL_PATH}/source/kernel/device/cuda)
+delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "${SRC_DEVICE_HIP}" ${CYCLES_INSTALL_PATH}/source/kernel/device/hip)
 delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "${SRC_DEVICE_OPTIX}" ${CYCLES_INSTALL_PATH}/source/kernel/device/optix)
 delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "${SRC_HEADERS}" ${CYCLES_INSTALL_PATH}/source/kernel)
 delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "${SRC_DEVICE_GPU_HEADERS}" ${CYCLES_INSTALL_PATH}/source/kernel/device/gpu)
 delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "${SRC_DEVICE_CUDA_HEADERS}" ${CYCLES_INSTALL_PATH}/source/kernel/device/cuda)
+delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "${SRC_DEVICE_HIP_HEADERS}" ${CYCLES_INSTALL_PATH}/source/kernel/device/hip)
 delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "${SRC_DEVICE_OPTIX_HEADERS}" ${CYCLES_INSTALL_PATH}/source/kernel/device/optix)
 delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "${SRC_BVH_HEADERS}" ${CYCLES_INSTALL_PATH}/source/kernel/bvh)
 delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "${SRC_CLOSURE_HEADERS}" ${CYCLES_INSTALL_PATH}/source/kernel/closure)