WIP: Add HIP backend support

CMakeLists.txt

@@ -5,8 +5,26 @@ if(MSVC)
   set(CMAKE_CONFIGURATION_TYPES "Debug;Release" CACHE STRING "Available configuration types to select")
 endif()
 
-option(GPU "GPU" ON)
-if(GPU)
+option(GPU "Build with CUDA GPU support" ON)
+option(HIP "Build with HIP GPU support (AMD ROCm or NVIDIA via HIP)" OFF)
+
+if(GPU AND HIP)
+  message(FATAL_ERROR "GPU (CUDA) and HIP options are mutually exclusive. Choose one backend.")
+endif()
+
+if(HIP)
+  if(NOT DEFINED CMAKE_HIP_ARCHITECTURES)
+    set(CMAKE_HIP_ARCHITECTURES native CACHE STRING "Targeted HIP architectures")
+  endif()
+  project(cuda-battery
+    HOMEPAGE_URL "https://github.com/lattice-land/cuda-battery"
+    LANGUAGES HIP CXX)
+  # find_package(hip) provides hip::host which handles HIP runtime linking on both
+  # AMD (amdhip64) and NVIDIA (wraps cudart).  CUDA::cuda_driver is needed on NVIDIA
+  # because hipDeviceGetAttribute routes through the CUDA driver API (libcuda.so).
+  find_package(hip REQUIRED)
+  find_package(CUDAToolkit QUIET)
+elseif(GPU)
   if(NOT DEFINED CMAKE_CUDA_ARCHITECTURES)
     set(CMAKE_CUDA_ARCHITECTURES native CACHE STRING "Targeted CUDA architectures")
   endif()
@@ -41,12 +59,28 @@ target_compile_options(cuda_battery INTERFACE
     "$<$<CXX_COMPILER_ID:GNU,Clang>:-frounding-math>"
     $<$<AND:$<COMPILE_LANGUAGE:CXX>,$<CXX_COMPILER_ID:MSVC>>:/fp:strict>
     "$<$<AND:$<COMPILE_LANGUAGE:CUDA>,$<CUDA_COMPILER_ID:NVIDIA>,$<CXX_COMPILER_ID:MSVC>>:SHELL:--compiler-options /fp:strict>"
+    # HIP compile options (hipcc on AMD ROCm or NVIDIA via HIP_PLATFORM=nvidia)
+    "$<$<COMPILE_LANGUAGE:HIP>:--expt-relaxed-constexpr>"
+    "$<$<AND:$<COMPILE_LANGUAGE:HIP>,$<CONFIG:Debug>>:-ggdb>"
 )
 
 if(REDUCE_PTX_SIZE)
   target_compile_definitions(cuda_battery INTERFACE "$<$<COMPILE_LANGUAGE:CUDA>:REDUCE_PTX_SIZE>")
 endif()
 
+# When building with HIP=ON, link hip::host so both AMD and NVIDIA HIP builds get
+# the correct HIP runtime libraries through a single CMake target.
+if(HIP)
+  target_link_libraries(cuda_battery INTERFACE hip::host)
+endif()
+
+# When building with HIP=ON (either AMD or NVIDIA via HIP), expose BATTERY_HIP_BUILD
+# so that utility.hpp makes HIPE/HIPEX available even when nvcc is the compiler
+# (hip-nvidia-* presets: __CUDACC__ is defined but HIP headers are in scope).
+if(HIP)
+  target_compile_definitions(cuda_battery INTERFACE BATTERY_HIP_BUILD)
+endif()
+
 if(CUDA_BATTERY_BUILD_TESTS)
 
 # Google Test dependency
@@ -88,6 +122,26 @@ if(GPU)
   endforeach()
 endif()
 
+# III. HIP GPU Tests (ending with "_hip.cpp")
+if(HIP)
+  # HIP runtime on NVIDIA routes device-attribute queries through the CUDA Driver
+  # API (cuDeviceGetAttribute in libcuda.so). CMake's implicit HIP link only adds
+  # libcudart_static; link CUDA::cuda_driver explicitly so hipDeviceGetAttribute works.
+  find_package(CUDAToolkit REQUIRED)
+
+  file(GLOB hip_test_files tests/*_hip.cpp)
+  foreach(file ${hip_test_files})
+    cmake_path(GET file STEM test_name)
+    set_source_files_properties(${file} PROPERTIES LANGUAGE HIP)
+    add_executable(${test_name} ${file})
+    target_link_libraries(${test_name} cuda_battery)
+    if(CUDAToolkit_FOUND)
+      target_link_libraries(${test_name} CUDA::cuda_driver)
+    endif()
+    add_test(NAME ${test_name} COMMAND $<TARGET_FILE:${test_name}>)
+  endforeach()
+endif()
+
 endif()
 
 # Documentation

CMakePresets.json

@@ -107,6 +107,50 @@
       "displayName": "CPU Release configuration",
       "description": "Build the project with a CPU compiler such as GCC.",
       "inherits": ["default-cpu","release"]
+    },
+    {
+      "name": "default-hip",
+      "hidden": true,
+      "inherits": "default",
+      "cacheVariables": {
+        "GPU": { "type": "BOOL", "value": "OFF" },
+        "HIP": { "type": "BOOL", "value": "ON" }
+      }
+    },
+    {
+      "name": "default-hip-nvidia",
+      "hidden": true,
+      "inherits": "default-hip",
+      "environment": {
+        "HIP_PLATFORM": "nvidia"
+      },
+      "cacheVariables": {
+        "CMAKE_HIP_ARCHITECTURES": "native"
+      }
+    },
+    {
+      "name": "hip-debug",
+      "displayName": "HIP Debug (AMD ROCm)",
+      "description": "Build using hipcc targeting AMD GPUs.",
+      "inherits": ["default-hip", "debug", "verbose"]
+    },
+    {
+      "name": "hip-release",
+      "displayName": "HIP Release (AMD ROCm)",
+      "description": "Build using hipcc targeting AMD GPUs.",
+      "inherits": ["default-hip", "release"]
+    },
+    {
+      "name": "hip-nvidia-debug",
+      "displayName": "HIP Debug - NVIDIA backend [temporary]",
+      "description": "Build using hipcc with HIP_PLATFORM=nvidia. Used to develop and validate HIP support on NVIDIA hardware before testing on AMD. Remove once AMD testing is established.",
+      "inherits": ["default-hip-nvidia", "debug", "verbose"]
+    },
+    {
+      "name": "hip-nvidia-release",
+      "displayName": "HIP Release - NVIDIA backend [temporary]",
+      "description": "See hip-nvidia-debug. Remove once AMD testing is established.",
+      "inherits": ["default-hip-nvidia", "release"]
     }
   ],
   "buildPresets": [
@@ -125,6 +169,22 @@
     {
       "name": "cpu-release",
       "configurePreset": "cpu-release"
+    },
+    {
+      "name": "hip-debug",
+      "configurePreset": "hip-debug"
+    },
+    {
+      "name": "hip-release",
+      "configurePreset": "hip-release"
+    },
+    {
+      "name": "hip-nvidia-debug",
+      "configurePreset": "hip-nvidia-debug"
+    },
+    {
+      "name": "hip-nvidia-release",
+      "configurePreset": "hip-nvidia-release"
     }
   ],
   "testPresets": [
@@ -143,6 +203,22 @@
     {
       "name": "cpu-release",
       "configurePreset": "cpu-release"
+    },
+    {
+      "name": "hip-debug",
+      "configurePreset": "hip-debug"
+    },
+    {
+      "name": "hip-release",
+      "configurePreset": "hip-release"
+    },
+    {
+      "name": "hip-nvidia-debug",
+      "configurePreset": "hip-nvidia-debug"
+    },
+    {
+      "name": "hip-nvidia-release",
+      "configurePreset": "hip-nvidia-release"
     }
   ],
   "workflowPresets": [
@@ -200,18 +276,41 @@
     {
       "name": "gpu-release",
       "steps": [
-        {
-          "type": "configure",
-          "name": "gpu-release"
-        },
-        {
-          "type": "build",
-          "name": "gpu-release"
-        },
-        {
-          "type": "test",
-          "name": "gpu-release"
-        }
+        { "type": "configure", "name": "gpu-release" },
+        { "type": "build",     "name": "gpu-release" },
+        { "type": "test",      "name": "gpu-release" }
+      ]
+    },
+    {
+      "name": "hip-debug",
+      "steps": [
+        { "type": "configure", "name": "hip-debug" },
+        { "type": "build",     "name": "hip-debug" },
+        { "type": "test",      "name": "hip-debug" }
+      ]
+    },
+    {
+      "name": "hip-release",
+      "steps": [
+        { "type": "configure", "name": "hip-release" },
+        { "type": "build",     "name": "hip-release" },
+        { "type": "test",      "name": "hip-release" }
+      ]
+    },
+    {
+      "name": "hip-nvidia-debug",
+      "steps": [
+        { "type": "configure", "name": "hip-nvidia-debug" },
+        { "type": "build",     "name": "hip-nvidia-debug" },
+        { "type": "test",      "name": "hip-nvidia-debug" }
+      ]
+    },
+    {
+      "name": "hip-nvidia-release",
+      "steps": [
+        { "type": "configure", "name": "hip-nvidia-release" },
+        { "type": "build",     "name": "hip-nvidia-release" },
+        { "type": "test",      "name": "hip-nvidia-release" }
       ]
     }
   ]

README.md

@@ -1,12 +1,25 @@
-# Standard Library for CUDA Programming
+# Standard Library for CUDA/HIP Programming
 
 [![Build Status](https://travis-ci.com/lattice-land/cuda-battery.svg?branch=main)](https://travis-ci.com/lattice-land/cuda-battery)
 
-This library provides data structures to ease programming in CUDA (version 12 or higher).
+This library provides data structures to ease GPU programming in CUDA (version 12 or higher) and HIP (ROCm 6+).
 For a tutorial and further information, please read [this manual](https://lattice-land.github.io/1-cuda-battery.html).
 
+## Supported Backends
+
+| Backend | Compiler | CMake option | Active macro |
+|---|---|---|---|
+| CPU only | GCC / Clang / MSVC | _(default)_ | — |
+| CUDA (NVIDIA) | nvcc 12+ | `GPU=ON` | `BATTERY_CUDA_BACKEND` |
+| HIP (AMD ROCm) | amdclang++ / hipcc, ROCm 6+ | `HIP=ON GPU=OFF` | `BATTERY_HIP_BACKEND` |
+
+The three backends are mutually exclusive at compile time.
+
 ## Example
 
+The example below works unchanged on both CUDA and HIP — just compile with the appropriate backend option.
+For HIP, replace `CUDAEX` with `HIPEX` and `cudaDeviceSynchronize` with `hipDeviceSynchronize`.
+
 Quick example on how to transfer a `std::vector` on CPU to a `battery::vector` on GPU (notice you don't need to do any manual memory allocation or deallocation):
 
 ```cpp
@@ -60,6 +73,7 @@ int main(int argc, char** argv) {
 | Containers | [`vector`](https://lattice-land.github.io/cuda-battery/vector_8hpp.html) ([`std`](https://en.cppreference.com/w/cpp/container/vector)) | [`string`](https://lattice-land.github.io/cuda-battery/string_8hpp.html) ([`std`](https://en.cppreference.com/w/cpp/string/basic_string)) | [`dynamic_bitset`](https://lattice-land.github.io/cuda-battery/dynamic__bitset_8hpp.html) | |
 | | [`tuple`](https://en.cppreference.com/w/cpp/utility/tuple) | [`variant`](https://lattice-land.github.io/cuda-battery/variant_8hpp.html) ([`std`](https://en.cppreference.com/w/cpp/utility/variant)) | [`bitset`](https://lattice-land.github.io/cuda-battery/bitset_8hpp.html) ([`std`](https://en.cppreference.com/w/cpp/utility/bitset)) | |
 | Utility | [`CUDA`](https://lattice-land.github.io/cuda-battery/utility_8hpp.html#a7b01e29f669d6beed251f1b2a547ca93) | [`INLINE`](https://lattice-land.github.io/cuda-battery/utility_8hpp.html#a2eb6f9e0395b47b8d5e3eeae4fe0c116) | [`CUDAE`](https://lattice-land.github.io/cuda-battery/utility_8hpp.html#a289596c1db721f82251de6902f9699db) | [`CUDAEX`](https://lattice-land.github.io/cuda-battery/utility_8hpp.html#af35c92d967acfadd086658422f631100) |
+| | | | [`HIPE`](https://lattice-land.github.io/cuda-battery/utility_8hpp.html) _(HIP)_ | [`HIPEX`](https://lattice-land.github.io/cuda-battery/utility_8hpp.html) _(HIP)_ |
 | | [`limits`](https://lattice-land.github.io/cuda-battery/structbattery_1_1limits.html) | [`ru_cast`](https://lattice-land.github.io/cuda-battery/namespacebattery.html#a7fdea425c76eab201a009ea09b8cbac0) | [`rd_cast`](https://lattice-land.github.io/cuda-battery/namespacebattery.html#aa2296c962277e71780bccf1ba9708f59) | |
 | | [`popcount`](https://lattice-land.github.io/cuda-battery/namespacebattery.html#a2821ae67e8ea81b375f3fd6d70909fef) ([`std`](https://en.cppreference.com/w/cpp/numeric/popcount)) | [`countl_zero`](https://lattice-land.github.io/cuda-battery/namespacebattery.html#aa18a34122dc3e8b7e96c4a54eeffa9aa) ([`std`](https://en.cppreference.com/w/cpp/numeric/countl_zero)) | [`countl_one`](https://lattice-land.github.io/cuda-battery/namespacebattery.html#ae252a50e577d7b092eb368b7e0289772) ([`std`](https://en.cppreference.com/w/cpp/numeric/countl_one)) | [`countr_zero`](https://lattice-land.github.io/cuda-battery/namespacebattery.html#a7338f90fab224e49c3716c5eace58bee) ([`std`](https://en.cppreference.com/w/cpp/numeric/countr_zero)) |
 | | [`countr_one`](https://lattice-land.github.io/cuda-battery/namespacebattery.html#a974d0a682d546e1185ae7dca29c272d6) ([`std`](https://en.cppreference.com/w/cpp/numeric/countr_one)) | [`signum`](https://lattice-land.github.io/cuda-battery/namespacebattery.html#a31b3f5ee3799a73d29c153ebd222cdea) | [`ipow`](https://lattice-land.github.io/cuda-battery/namespacebattery.html#a93472d80842253624e2436eef7b900b6) | |