1-3-kernel_programming.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\u001b[32m\u001b[1m  Activating\u001b[22m\u001b[39m project at `~/cscs2023`\n",
      "\u001b[36m\u001b[1m[ \u001b[22m\u001b[39m\u001b[36m\u001b[1mInfo: \u001b[22m\u001b[39mPrecompiling BenchmarkTools [6e4b80f9-dd63-53aa-95a3-0cdb28fa8baf]\n"
     ]
    }
   ],
   "source": [
    "using Pkg\n",
    "Pkg.activate(@__DIR__)\n",
    "Pkg.instantiate()\n",
    "\n",
    "using CUDA, BenchmarkTools\n",
    "CUDA.allowscalar(false)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Kernel programming\n",
    "\n",
    "When an array operation is not supported, or you need to perform an operation that you can easily express using existing array abstractions, you might need to write your own kernel. Kernels are **scalar functions that are executed multiple times in parallel**. Each 'thread' runs on one of the many streaming multiprocessors a GPU has, and threads running on a single SM are called a 'block'. Within a SM, some threads are always executed together; these form a 'warp' of 32 threads."
   ]
  },
  {
   "attachments": {
    "image.png": {
     "image/png": "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"
    }
   },
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "![image.png](attachment:image.png)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Key to using a GPU efficiently, is optimizing the communication between these entities:\n",
    "\n",
    "- between blocks: global memory\n",
    "- within a block: shared memory\n",
    "- within a warp: via registers (shuffle)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "But let's start with the basics."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Defining and launching kernels"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "A kernel function is just a regular function in Julia, launched using the `@cuda` macro:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "CUDA.HostKernel for my_kernel()"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "function my_kernel()\n",
    "    return\n",
    "end\n",
    "\n",
    "@cuda my_kernel()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "You can pass any argument to a CUDA kernel, as long as it's an `isbits` type:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "CUDA.HostKernel for my_kernel(Int64)"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "function my_kernel(x)\n",
    "    return\n",
    "end\n",
    "\n",
    "@cuda my_kernel(42)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "ename": "LoadError",
     "evalue": "GPU compilation of MethodInstance for my_kernel(::String) failed\nKernelError: passing and using non-bitstype argument\n\nArgument 2 to your kernel function is of type String, which is not isbits:\n\n",
     "output_type": "error",
     "traceback": [
      "GPU compilation of MethodInstance for my_kernel(::String) failed\nKernelError: passing and using non-bitstype argument\n\nArgument 2 to your kernel function is of type String, which is not isbits:\n\n",
      "",
      "Stacktrace:",
      "  [1] check_invocation(job::GPUCompiler.CompilerJob)",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/validation.jl:92",
      "  [2] macro expansion",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/driver.jl:123 [inlined]",
      "  [3] macro expansion",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/TimerOutputs/RsWnF/src/TimerOutput.jl:253 [inlined]",
      "  [4] codegen(output::Symbol, job::GPUCompiler.CompilerJob; libraries::Bool, toplevel::Bool, optimize::Bool, cleanup::Bool, strip::Bool, validate::Bool, only_entry::Bool, parent_job::Nothing)",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/driver.jl:121",
      "  [5] codegen",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/driver.jl:110 [inlined]",
      "  [6] compile(target::Symbol, job::GPUCompiler.CompilerJob; libraries::Bool, toplevel::Bool, optimize::Bool, cleanup::Bool, strip::Bool, validate::Bool, only_entry::Bool)",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/driver.jl:106",
      "  [7] compile",
      "    @ /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/driver.jl:98 [inlined]",
      "  [8] #1075",
      "    @ /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/src/compiler/compilation.jl:247 [inlined]",
      "  [9] JuliaContext(f::CUDA.var\"#1075#1077\"{GPUCompiler.CompilerJob{GPUCompiler.PTXCompilerTarget, CUDA.CUDACompilerParams}})",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/driver.jl:47",
      " [10] compile(job::GPUCompiler.CompilerJob)",
      "    @ CUDA /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/src/compiler/compilation.jl:246",
      " [11] actual_compilation(cache::Dict{Any, CuFunction}, src::Core.MethodInstance, world::UInt64, cfg::GPUCompiler.CompilerConfig{GPUCompiler.PTXCompilerTarget, CUDA.CUDACompilerParams}, compiler::typeof(CUDA.compile), linker::typeof(CUDA.link))",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/execution.jl:125",
      " [12] cached_compilation(cache::Dict{Any, CuFunction}, src::Core.MethodInstance, cfg::GPUCompiler.CompilerConfig{GPUCompiler.PTXCompilerTarget, CUDA.CUDACompilerParams}, compiler::Function, linker::Function)",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/execution.jl:103",
      " [13] macro expansion",
      "    @ /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/src/compiler/execution.jl:373 [inlined]",
      " [14] macro expansion",
      "    @ ./lock.jl:267 [inlined]",
      " [15] cufunction(f::typeof(my_kernel), tt::Type{Tuple{String}}; kwargs::@Kwargs{})",
      "    @ CUDA /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/src/compiler/execution.jl:368",
      " [16] cufunction(f::typeof(my_kernel), tt::Type{Tuple{String}})",
      "    @ CUDA /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/src/compiler/execution.jl:365",
      " [17] top-level scope",
      "    @ /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/src/compiler/execution.jl:104"
     ]
    }
   ],
   "source": [
    "@cuda my_kernel(\"42\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "For some non-`isbits` arguments, such as `CuArray`, we have automatic conversions in place that make the arguments GPU-compatible:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "false"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "A = CuArray([1,2,3])\n",
    "isbits(A)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "CUDA.HostKernel for my_kernel(CuDeviceVector{Int64, 1})"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "@cuda my_kernel(A)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Launch configuration and indexing\n",
    "\n",
    "Essential to a GPU, is launching multiple threads by setting the `threads` argument to `@cuda` and differentiating the computations by querying the current thread index:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "10-element CuArray{Float32, 1, CUDA.Mem.DeviceBuffer}:\n",
       "  1.0\n",
       "  2.0\n",
       "  3.0\n",
       "  4.0\n",
       "  5.0\n",
       "  6.0\n",
       "  7.0\n",
       "  8.0\n",
       "  9.0\n",
       " 10.0"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "function kernel(A)\n",
    "    A[threadIdx().x] = threadIdx().x\n",
    "    return\n",
    "end\n",
    "\n",
    "A = CUDA.zeros(10)\n",
    "@cuda threads=length(A) kernel(A)\n",
    "A"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "As threads are guaranteed to execute on the same shared multiprocessor, there's a limit on how many threads you can launch:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "1024"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "attribute(device(), CUDA.DEVICE_ATTRIBUTE_MAX_THREADS_PER_BLOCK)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "You can work around this by launching multiple blocks, which can be scheduled on different multiprocessors:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "10-element CuArray{@NamedTuple{thread::Int64, block::Int64}, 1, CUDA.Mem.DeviceBuffer}:\n",
       " (thread = 1, block = 1)\n",
       " (thread = 2, block = 1)\n",
       " (thread = 3, block = 1)\n",
       " (thread = 4, block = 1)\n",
       " (thread = 5, block = 1)\n",
       " (thread = 1, block = 2)\n",
       " (thread = 2, block = 2)\n",
       " (thread = 3, block = 2)\n",
       " (thread = 4, block = 2)\n",
       " (thread = 5, block = 2)"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "function kernel(A)\n",
    "    i = (blockIdx().x-1) * blockDim().x + threadIdx().x\n",
    "    if i <= length(A)\n",
    "        # it's possible blockDim * gridDim > length(A)\n",
    "        A[i] = (; thread=threadIdx().x, block=blockIdx().x)\n",
    "    end\n",
    "    return\n",
    "end\n",
    "\n",
    "A = CuArray{@NamedTuple{thread::Int, block::Int}}(undef, 10)\n",
    "@cuda threads=5 blocks=2 kernel(A)\n",
    "A"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Most of the time, it's best to rely on the CUDA APIs to determine a good launch configuration, using the occupancy API:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "config = launch_configuration(k.fun) = (blocks = 112, threads = 1024)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "10-element CuArray{@NamedTuple{thread::Int64, block::Int64}, 1, CUDA.Mem.DeviceBuffer}:\n",
       " (thread = 1, block = 1)\n",
       " (thread = 2, block = 1)\n",
       " (thread = 3, block = 1)\n",
       " (thread = 4, block = 1)\n",
       " (thread = 5, block = 1)\n",
       " (thread = 6, block = 1)\n",
       " (thread = 7, block = 1)\n",
       " (thread = 8, block = 1)\n",
       " (thread = 9, block = 1)\n",
       " (thread = 10, block = 1)"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "k = @cuda launch=false kernel(A)\n",
    "@show config = launch_configuration(k.fun)\n",
    "threads = min(length(A), config.threads)\n",
    "blocks = cld(length(A), threads)\n",
    "k(A; threads, blocks)\n",
    "A"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Synchronization\n",
    "\n",
    "If threads are working together -- say, they are using the same global memory, or are communicating using shared memory or finer-grained intrinsics -- you may need to have threads wait on each other. Note that this is only possible **within a block**; different blocks generally cannot wait on one another.\n",
    "\n",
    "Let's look at a contrived example:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "true"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# simple reverse kernel without multiple blocks\n",
    "function reverse_kernel(A)\n",
    "    i = threadIdx().x\n",
    "    if i <= length(A)\n",
    "        j = length(A) - i + 1\n",
    "        A[i] = A[j]\n",
    "    end\n",
    "    return\n",
    "end\n",
    "\n",
    "A = CuArray(1:100)\n",
    "@cuda threads=length(A) reverse_kernel(A)\n",
    "Array(A) == 100:-1:1"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Depending on how the threads get scheduled, values may get read after they have been written to by another thread. The solution is to wait for all threads to have read a value before writing to it again:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "true"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# simple reverse kernel without multiple blocks\n",
    "function reverse_kernel(A)\n",
    "    i = threadIdx().x\n",
    "    if i <= length(A)\n",
    "        j = length(A) - i + 1\n",
    "        val = A[j]\n",
    "    end\n",
    "    sync_threads()\n",
    "    if i <= length(A)\n",
    "        A[i] = val\n",
    "    end\n",
    "    return\n",
    "end\n",
    "\n",
    "A = CuArray(1:100)\n",
    "@cuda threads=length(A) reverse_kernel(A)\n",
    "Array(A) == 100:-1:1"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Note how we didn't put `sync_threads()` inside of the branch; All threads need to reach the synchronization point for the kernel to make progress. This makes it dangerous to synchronize from a branch, as the branch cannot be divergent within a block or the kernel would deadlock!"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "When coordinating within the warp, you may need the `sync_warp()` function. A detailed explanation of warp-level programming is out of scope for this notebook, refer to the [NVIDIA developer blog](https://developer.nvidia.com/blog/using-cuda-warp-level-primitives/) for more information."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Shared memory\n",
    "\n",
    "As mentioned before, one of the crucial aspects of GPU programming is efficient communication between threads, where efficient means that we're using the optimal memory hierarchy for the job, instead of always hitting global device memory.\n",
    "\n",
    "For threads that execute in a block, which will be scheduled on a single shared multiprocessor, it's possible to allocate shared memory that can be accessed by all threads.\n",
    "\n",
    "For example, let's update the `reverse` kernel to use shared memory:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "true"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "function reverse_kernel(A::AbstractVector{T}) where T\n",
    "    i = threadIdx().x\n",
    "    j = length(A) - i + 1\n",
    "    buf = CuStaticSharedArray(T, 100)\n",
    "    buf[i] = A[i]\n",
    "    sync_threads()\n",
    "    A[i] = buf[j]\n",
    "    return\n",
    "end\n",
    "\n",
    "A = CuArray(1:100)\n",
    "@cuda threads=length(A) reverse_kernel(A)\n",
    "Array(A) == 100:-1:1"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "With static shared memory, we have to hardcode the size of the shared array. That's often OK, because the size of the thread block is often known (or bounded). In case it isn't, we can use dynamic shared memory, which requires setting the `shmem` keyword argument to `@cuda` to a number of bytes:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "true"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "function reverse_kernel(A::AbstractVector{T}) where T\n",
    "    i = threadIdx().x\n",
    "    j = length(A) - i + 1\n",
    "    buf = CuDynamicSharedArray(T, length(A))\n",
    "    buf[i] = A[i]\n",
    "    sync_threads()\n",
    "    A[i] = buf[j]\n",
    "    return\n",
    "end\n",
    "\n",
    "A = CuArray(1:100)\n",
    "@cuda threads=length(A) shmem=sizeof(A) reverse_kernel(A)\n",
    "Array(A) == 100:-1:1"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Note that we still passed the (dynamic) length of the array to the `CuDynamicSharedArray` constructor, which makes it possible to bounds-check accesses. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Atomic operations\n",
    "\n",
    "When you want to use the same global memory from different threads, you may want to use atomic operations. For example:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "241.49663f0"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "A_sum = CUDA.zeros(1)\n",
    "A = CUDA.rand(512)\n",
    "\n",
    "function kernel(A, A_sum)\n",
    "    i = threadIdx().x\n",
    "    CUDA.@atomic A_sum[] += A[i]\n",
    "    return\n",
    "end\n",
    "@cuda threads=length(A) kernel(A, A_sum)\n",
    "Array(A_sum)[]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "You shouldn't overuse atomics though, as they generally serialize execution and thus are very expensive! But they may be useful for an initial implementation (i.e. before considering more fine-grained communication), or to reduce values from different blocks (because of the difficulty of synchronizing the grid)."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Output\n",
    "\n",
    "To help with implementing a kernel, there's a couple of helpful macros to generate output:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "I'm thread 1\n"
     ]
    }
   ],
   "source": [
    "function kernel()\n",
    "    i = threadIdx().x\n",
    "    @cuprintf \"I'm thread %ld\\n\" Int(i)\n",
    "    return\n",
    "end\n",
    "@cuda kernel();"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "However, `@cuprintf` is a bit cumbersome, so we have `@cuprintln` trying to automatically generate an appropriate formatting string, while even supporting string interpolation:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "I'm thread 1\n"
     ]
    }
   ],
   "source": [
    "function kernel()\n",
    "    i = threadIdx().x\n",
    "    @cuprintln \"I'm thread $i\"\n",
    "    return\n",
    "end\n",
    "@cuda kernel();"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "And for quick debugging, we have a helpful `@cushow` you can surround expressions with:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(threadIdx()).x = 1\n"
     ]
    }
   ],
   "source": [
    "function kernel()\n",
    "    i = @cushow threadIdx().x\n",
    "    return\n",
    "end\n",
    "@cuda kernel();"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Other CUDA features\n",
    "\n",
    "CUDA.jl supports many other device-side CUDA features, like cooperative groups, dynamic parallelism, warp functions, etc. Some high-level Julia features are also supported in device context, such as random number generation and "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Common issues\n",
    "\n",
    "Because some aspects of the Julia language are unsupported, you'll definitely be running into compilation errors when writing GPU device code. With CPU code, Julia being a dynamic language, errors are postponed to run-time, so if you have a typo in your code it will still compile but you will run into an error at run time.\n",
    "\n",
    "With GPU code, it's harder to report errors at run time (for one, they'd be generated by every thread, resulting in a multiplication of errors), so the GPU compiler generally refuses to compile when it encounters certain unsupported code patterns. In other cases, errors are reported, but they are much less informative than on the CPU."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div class=\"alert alert-block alert-info\">\n",
    "    <b>Note</b>: Work is under way to improve the ability for GPU code to call into the CPU for, e.g., dynamic error reporting. That would make it possible to compile unsupported code and have it error at run time just like Julia code on the CPU.\n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's demonstrate a couple of common errors:"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Returning values from kernels\n",
    "\n",
    "Kernel functions cannot return anything; if you do so you'll run into a compilation error:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "ename": "LoadError",
     "evalue": "GPU compilation of MethodInstance for kernel() failed\nKernelError: kernel returns a value of type `Int64`\n\nMake sure your kernel function ends in `return`, `return nothing` or `nothing`.\n",
     "output_type": "error",
     "traceback": [
      "GPU compilation of MethodInstance for kernel() failed\nKernelError: kernel returns a value of type `Int64`\n\nMake sure your kernel function ends in `return`, `return nothing` or `nothing`.\n",
      "",
      "Stacktrace:",
      "  [1] check_method(job::GPUCompiler.CompilerJob)",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/validation.jl:36",
      "  [2] macro expansion",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/driver.jl:122 [inlined]",
      "  [3] macro expansion",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/TimerOutputs/RsWnF/src/TimerOutput.jl:253 [inlined]",
      "  [4] codegen(output::Symbol, job::GPUCompiler.CompilerJob; libraries::Bool, toplevel::Bool, optimize::Bool, cleanup::Bool, strip::Bool, validate::Bool, only_entry::Bool, parent_job::Nothing)",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/driver.jl:121",
      "  [5] codegen",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/driver.jl:110 [inlined]",
      "  [6] compile(target::Symbol, job::GPUCompiler.CompilerJob; libraries::Bool, toplevel::Bool, optimize::Bool, cleanup::Bool, strip::Bool, validate::Bool, only_entry::Bool)",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/driver.jl:106",
      "  [7] compile",
      "    @ /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/driver.jl:98 [inlined]",
      "  [8] #1075",
      "    @ /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/src/compiler/compilation.jl:247 [inlined]",
      "  [9] JuliaContext(f::CUDA.var\"#1075#1077\"{GPUCompiler.CompilerJob{GPUCompiler.PTXCompilerTarget, CUDA.CUDACompilerParams}})",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/driver.jl:47",
      " [10] compile(job::GPUCompiler.CompilerJob)",
      "    @ CUDA /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/src/compiler/compilation.jl:246",
      " [11] actual_compilation(cache::Dict{Any, CuFunction}, src::Core.MethodInstance, world::UInt64, cfg::GPUCompiler.CompilerConfig{GPUCompiler.PTXCompilerTarget, CUDA.CUDACompilerParams}, compiler::typeof(CUDA.compile), linker::typeof(CUDA.link))",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/execution.jl:125",
      " [12] cached_compilation(cache::Dict{Any, CuFunction}, src::Core.MethodInstance, cfg::GPUCompiler.CompilerConfig{GPUCompiler.PTXCompilerTarget, CUDA.CUDACompilerParams}, compiler::Function, linker::Function)",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/execution.jl:103",
      " [13] macro expansion",
      "    @ /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/src/compiler/execution.jl:373 [inlined]",
      " [14] macro expansion",
      "    @ ./lock.jl:267 [inlined]",
      " [15] cufunction(f::typeof(kernel), tt::Type{Tuple{}}; kwargs::@Kwargs{})",
      "    @ CUDA /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/src/compiler/execution.jl:368",
      " [16] cufunction(f::typeof(kernel), tt::Type{Tuple{}})",
      "    @ CUDA /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/src/compiler/execution.jl:365",
      " [17] top-level scope",
      "    @ /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/src/compiler/execution.jl:104"
     ]
    }
   ],
   "source": [
    "kernel() = 42\n",
    "@cuda kernel()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "If you need to return a value from a kernel, use a single-element `CuArray`, or a `Ref`:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "1-element CuArray{Int64, 1, CUDA.Mem.DeviceBuffer}:\n",
       " 42"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "function kernel(A)\n",
    "    A[1] = 42\n",
    "    return\n",
    "end\n",
    "\n",
    "A = CuArray([0])\n",
    "@cuda kernel(A)\n",
    "A"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Base.RefValue{Int64}(42)"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "function kernel(ref)\n",
    "    ref[] = 42\n",
    "    return\n",
    "end\n",
    "\n",
    "ref = Ref(0)\n",
    "@cuda kernel(ref)\n",
    "ref"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Run-time exceptions\n",
    "\n",
    "When a kernel runs into an exception, there's two issues that may make resolving the exception harder than on the CPU:\n",
    "\n",
    "- because kernels execute asynchronously, the exception will only be caught during a subsequent synchronizing API call;\n",
    "- much of the information of the exception will have been lost (notably custom printing, and sometimes even the exception type).\n",
    "\n",
    "For example:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "CUDA.HostKernel for kernel(CuDeviceVector{Int64, 1})"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ERROR: a exception was thrown during kernel execution.\n",
      "       Run Julia on debug level 2 for device stack traces.\n"
     ]
    }
   ],
   "source": [
    "function kernel(a)\n",
    "    if threadIdx().x == 1\n",
    "        a[] += 42.1\n",
    "    end\n",
    "    return\n",
    "end\n",
    "@cuda kernel(CuArray([42]))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Here, execution of the kernel already indicated something went wrong by printing information to the console, but we haven't \"trapped\" the exception yet. That only happepns on a subsequent API call, such as `synchronize()`:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "ename": "LoadError",
     "evalue": "KernelException: exception thrown during kernel execution on device Tesla P100-PCIE-16GB",
     "output_type": "error",
     "traceback": [
      "KernelException: exception thrown during kernel execution on device Tesla P100-PCIE-16GB",
      "",
      "Stacktrace:",
      " [1] check_exceptions()",
      "   @ CUDA /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/src/compiler/exceptions.jl:34",
      " [2] synchronize(stream::CuStream; blocking::Bool, spin::Bool)",
      "   @ CUDA /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/lib/cudadrv/synchronization.jl:194",
      " [3] synchronize",
      "   @ CUDA /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/lib/cudadrv/synchronization.jl:183 [inlined]",
      " [4] synchronize()",
      "   @ CUDA /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/lib/cudadrv/synchronization.jl:183",
      " [5] top-level scope",
      "   @ In[28]:1"
     ]
    }
   ],
   "source": [
    "synchronize()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The best way to debug this, is by starting Julia with `-g2` so that the device-side output will include a stack trace:\n",
    "\n",
    "```\n",
    "$ julia -g2\n",
    "...\n",
    "julia> @cuda kernel(CuArray([42]))\n",
    "ERROR: a exception was thrown during kernel execution.\n",
    "Stacktrace:\n",
    " [1] Int64 at ./float.jl:812\n",
    " [2] convert at ./number.jl:7\n",
    " [3] setindex! at /home/tim/Julia/pkg/CUDA/src/device/array.jl:204\n",
    " [4] setindex! at /home/tim/Julia/pkg/CUDA/src/device/array.jl:217\n",
    " [5] kernel at ./REPL[1]:3\n",
    "```"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Unsupported IR\n",
    "\n",
    "Other errors might result in a failure later during compilation:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "ename": "LoadError",
     "evalue": "InvalidIRError: compiling MethodInstance for kernel(::CuDeviceVector{Int64, 1}) resulted in invalid LLVM IR\n\u001b[31mReason: unsupported call to an unknown function\u001b[39m\u001b[31m (call to julia.new_gc_frame)\u001b[39m\n\u001b[31mReason: unsupported call to an unknown function\u001b[39m\u001b[31m (call to julia.push_gc_frame)\u001b[39m\n\u001b[31mReason: unsupported call to an unknown function\u001b[39m\u001b[31m (call to julia.pop_gc_frame)\u001b[39m\n\u001b[31mReason: unsupported use of an undefined name\u001b[39m\u001b[31m (use of 'threadId')\u001b[39m\nStacktrace:\n [1] \u001b[0m\u001b[1mkernel\u001b[22m\n\u001b[90m   @\u001b[39m \u001b[90m./\u001b[39m\u001b[90m\u001b[4mIn[29]:2\u001b[24m\u001b[39m\n\u001b[31mReason: unsupported call to an unknown function\u001b[39m\u001b[31m (call to julia.get_gc_frame_slot)\u001b[39m\n\u001b[31mReason: unsupported dynamic function invocation\u001b[39m\nStacktrace:\n [1] \u001b[0m\u001b[1mkernel\u001b[22m\n\u001b[90m   @\u001b[39m \u001b[90m./\u001b[39m\u001b[90m\u001b[4mIn[29]:2\u001b[24m\u001b[39m\n\u001b[31mReason: unsupported dynamic function invocation\u001b[39m\u001b[31m (call to getproperty)\u001b[39m\nStacktrace:\n [1] \u001b[0m\u001b[1mkernel\u001b[22m\n\u001b[90m   @\u001b[39m \u001b[90m./\u001b[39m\u001b[90m\u001b[4mIn[29]:2\u001b[24m\u001b[39m\n\u001b[31mReason: unsupported dynamic function invocation\u001b[39m\u001b[31m (call to ==)\u001b[39m\nStacktrace:\n [1] \u001b[0m\u001b[1mkernel\u001b[22m\n\u001b[90m   @\u001b[39m \u001b[90m./\u001b[39m\u001b[90m\u001b[4mIn[29]:2\u001b[24m\u001b[39m\n\u001b[36m\u001b[1mHint\u001b[22m\u001b[39m\u001b[36m: catch this exception as `err` and call `code_typed(err; interactive = true)` to introspect the erronous code with Cthulhu.jl\u001b[39m",
     "output_type": "error",
     "traceback": [
      "InvalidIRError: compiling MethodInstance for kernel(::CuDeviceVector{Int64, 1}) resulted in invalid LLVM IR\n\u001b[31mReason: unsupported call to an unknown function\u001b[39m\u001b[31m (call to julia.new_gc_frame)\u001b[39m\n\u001b[31mReason: unsupported call to an unknown function\u001b[39m\u001b[31m (call to julia.push_gc_frame)\u001b[39m\n\u001b[31mReason: unsupported call to an unknown function\u001b[39m\u001b[31m (call to julia.pop_gc_frame)\u001b[39m\n\u001b[31mReason: unsupported use of an undefined name\u001b[39m\u001b[31m (use of 'threadId')\u001b[39m\nStacktrace:\n [1] \u001b[0m\u001b[1mkernel\u001b[22m\n\u001b[90m   @\u001b[39m \u001b[90m./\u001b[39m\u001b[90m\u001b[4mIn[29]:2\u001b[24m\u001b[39m\n\u001b[31mReason: unsupported call to an unknown function\u001b[39m\u001b[31m (call to julia.get_gc_frame_slot)\u001b[39m\n\u001b[31mReason: unsupported dynamic function invocation\u001b[39m\nStacktrace:\n [1] \u001b[0m\u001b[1mkernel\u001b[22m\n\u001b[90m   @\u001b[39m \u001b[90m./\u001b[39m\u001b[90m\u001b[4mIn[29]:2\u001b[24m\u001b[39m\n\u001b[31mReason: unsupported dynamic function invocation\u001b[39m\u001b[31m (call to getproperty)\u001b[39m\nStacktrace:\n [1] \u001b[0m\u001b[1mkernel\u001b[22m\n\u001b[90m   @\u001b[39m \u001b[90m./\u001b[39m\u001b[90m\u001b[4mIn[29]:2\u001b[24m\u001b[39m\n\u001b[31mReason: unsupported dynamic function invocation\u001b[39m\u001b[31m (call to ==)\u001b[39m\nStacktrace:\n [1] \u001b[0m\u001b[1mkernel\u001b[22m\n\u001b[90m   @\u001b[39m \u001b[90m./\u001b[39m\u001b[90m\u001b[4mIn[29]:2\u001b[24m\u001b[39m\n\u001b[36m\u001b[1mHint\u001b[22m\u001b[39m\u001b[36m: catch this exception as `err` and call `code_typed(err; interactive = true)` to introspect the erronous code with Cthulhu.jl\u001b[39m",
      "",
      "Stacktrace:",
      "  [1] check_ir(job::GPUCompiler.CompilerJob{GPUCompiler.PTXCompilerTarget, CUDA.CUDACompilerParams}, args::LLVM.Module)",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/validation.jl:147",
      "  [2] macro expansion",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/driver.jl:440 [inlined]",
      "  [3] macro expansion",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/TimerOutputs/RsWnF/src/TimerOutput.jl:253 [inlined]",
      "  [4] macro expansion",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/driver.jl:439 [inlined]",
      "  [5] emit_llvm(job::GPUCompiler.CompilerJob; libraries::Bool, toplevel::Bool, optimize::Bool, cleanup::Bool, only_entry::Bool, validate::Bool)",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/utils.jl:92",
      "  [6] emit_llvm",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/utils.jl:86 [inlined]",
      "  [7] codegen(output::Symbol, job::GPUCompiler.CompilerJob; libraries::Bool, toplevel::Bool, optimize::Bool, cleanup::Bool, strip::Bool, validate::Bool, only_entry::Bool, parent_job::Nothing)",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/driver.jl:129",
      "  [8] codegen",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/driver.jl:110 [inlined]",
      "  [9] compile(target::Symbol, job::GPUCompiler.CompilerJob; libraries::Bool, toplevel::Bool, optimize::Bool, cleanup::Bool, strip::Bool, validate::Bool, only_entry::Bool)",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/driver.jl:106",
      " [10] compile",
      "    @ /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/driver.jl:98 [inlined]",
      " [11] #1075",
      "    @ /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/src/compiler/compilation.jl:247 [inlined]",
      " [12] JuliaContext(f::CUDA.var\"#1075#1077\"{GPUCompiler.CompilerJob{GPUCompiler.PTXCompilerTarget, CUDA.CUDACompilerParams}})",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/driver.jl:47",
      " [13] compile(job::GPUCompiler.CompilerJob)",
      "    @ CUDA /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/src/compiler/compilation.jl:246",
      " [14] actual_compilation(cache::Dict{Any, CuFunction}, src::Core.MethodInstance, world::UInt64, cfg::GPUCompiler.CompilerConfig{GPUCompiler.PTXCompilerTarget, CUDA.CUDACompilerParams}, compiler::typeof(CUDA.compile), linker::typeof(CUDA.link))",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/execution.jl:125",
      " [15] cached_compilation(cache::Dict{Any, CuFunction}, src::Core.MethodInstance, cfg::GPUCompiler.CompilerConfig{GPUCompiler.PTXCompilerTarget, CUDA.CUDACompilerParams}, compiler::Function, linker::Function)",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/execution.jl:103",
      " [16] macro expansion",
      "    @ /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/src/compiler/execution.jl:373 [inlined]",
      " [17] macro expansion",
      "    @ ./lock.jl:267 [inlined]",
      " [18] cufunction(f::typeof(kernel), tt::Type{Tuple{CuDeviceVector{Int64, 1}}}; kwargs::@Kwargs{})",
      "    @ CUDA /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/src/compiler/execution.jl:368",
      " [19] cufunction(f::typeof(kernel), tt::Type{Tuple{CuDeviceVector{Int64, 1}}})",
      "    @ CUDA /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/src/compiler/execution.jl:365",
      " [20] top-level scope",
      "    @ /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/src/compiler/execution.jl:104"
     ]
    }
   ],
   "source": [
    "function kernel(a)\n",
    "    if threadId().x == 1\n",
    "        a[] = 42\n",
    "    end\n",
    "    return\n",
    "end\n",
    "@cuda kernel(CuArray([1]))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "CUDA.jl tries its best to figure out why dynamic IR was generated, as can be seen in the error trace here. It's also possible to spot this error using reflection macros (more on those later):"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "PTX CompilerJob of MethodInstance for kernel(::CuDeviceVector{Int64, 1}) for sm_60\n",
      "\n",
      "MethodInstance for kernel(::CuDeviceVector{Int64, 1})\n",
      "  from kernel(\u001b[90ma\u001b[39m)\u001b[90m @\u001b[39m \u001b[90mMain\u001b[39m \u001b[90m\u001b[4mIn[29]:1\u001b[24m\u001b[39m\n",
      "Arguments\n",
      "  #self#\u001b[36m::Core.Const(kernel)\u001b[39m\n",
      "  a\u001b[36m::CuDeviceVector{Int64, 1}\u001b[39m\n",
      "Body\u001b[36m::Nothing\u001b[39m\n",
      "\u001b[90m1 ─\u001b[39m %1 = Main.threadId()\u001b[91m\u001b[1m::Any\u001b[22m\u001b[39m\n",
      "\u001b[90m│  \u001b[39m %2 = Base.getproperty(%1, :x)\u001b[91m\u001b[1m::Any\u001b[22m\u001b[39m\n",
      "\u001b[90m│  \u001b[39m %3 = (%2 == 1)\u001b[91m\u001b[1m::Any\u001b[22m\u001b[39m\n",
      "\u001b[90m└──\u001b[39m      goto #3 if not %3\n",
      "\u001b[90m2 ─\u001b[39m      Base.setindex!(a, 42)\n",
      "\u001b[90m3 ┄\u001b[39m      return nothing\n",
      "\n"
     ]
    },
    {
     "ename": "LoadError",
     "evalue": "InvalidIRError: compiling MethodInstance for kernel(::CuDeviceVector{Int64, 1}) resulted in invalid LLVM IR\n\u001b[31mReason: unsupported call to an unknown function\u001b[39m\u001b[31m (call to julia.new_gc_frame)\u001b[39m\n\u001b[31mReason: unsupported call to an unknown function\u001b[39m\u001b[31m (call to julia.push_gc_frame)\u001b[39m\n\u001b[31mReason: unsupported call to an unknown function\u001b[39m\u001b[31m (call to julia.pop_gc_frame)\u001b[39m\n\u001b[31mReason: unsupported use of an undefined name\u001b[39m\u001b[31m (use of 'threadId')\u001b[39m\nStacktrace:\n [1] \u001b[0m\u001b[1mkernel\u001b[22m\n\u001b[90m   @\u001b[39m \u001b[90m./\u001b[39m\u001b[90m\u001b[4mIn[29]:2\u001b[24m\u001b[39m\n\u001b[31mReason: unsupported call to an unknown function\u001b[39m\u001b[31m (call to julia.get_gc_frame_slot)\u001b[39m\n\u001b[31mReason: unsupported dynamic function invocation\u001b[39m\nStacktrace:\n [1] \u001b[0m\u001b[1mkernel\u001b[22m\n\u001b[90m   @\u001b[39m \u001b[90m./\u001b[39m\u001b[90m\u001b[4mIn[29]:2\u001b[24m\u001b[39m\n\u001b[31mReason: unsupported dynamic function invocation\u001b[39m\u001b[31m (call to getproperty)\u001b[39m\nStacktrace:\n [1] \u001b[0m\u001b[1mkernel\u001b[22m\n\u001b[90m   @\u001b[39m \u001b[90m./\u001b[39m\u001b[90m\u001b[4mIn[29]:2\u001b[24m\u001b[39m\n\u001b[31mReason: unsupported dynamic function invocation\u001b[39m\u001b[31m (call to ==)\u001b[39m\nStacktrace:\n [1] \u001b[0m\u001b[1mkernel\u001b[22m\n\u001b[90m   @\u001b[39m \u001b[90m./\u001b[39m\u001b[90m\u001b[4mIn[29]:2\u001b[24m\u001b[39m\n\u001b[36m\u001b[1mHint\u001b[22m\u001b[39m\u001b[36m: catch this exception as `err` and call `code_typed(err; interactive = true)` to introspect the erronous code with Cthulhu.jl\u001b[39m",
     "output_type": "error",
     "traceback": [
      "InvalidIRError: compiling MethodInstance for kernel(::CuDeviceVector{Int64, 1}) resulted in invalid LLVM IR\n\u001b[31mReason: unsupported call to an unknown function\u001b[39m\u001b[31m (call to julia.new_gc_frame)\u001b[39m\n\u001b[31mReason: unsupported call to an unknown function\u001b[39m\u001b[31m (call to julia.push_gc_frame)\u001b[39m\n\u001b[31mReason: unsupported call to an unknown function\u001b[39m\u001b[31m (call to julia.pop_gc_frame)\u001b[39m\n\u001b[31mReason: unsupported use of an undefined name\u001b[39m\u001b[31m (use of 'threadId')\u001b[39m\nStacktrace:\n [1] \u001b[0m\u001b[1mkernel\u001b[22m\n\u001b[90m   @\u001b[39m \u001b[90m./\u001b[39m\u001b[90m\u001b[4mIn[29]:2\u001b[24m\u001b[39m\n\u001b[31mReason: unsupported call to an unknown function\u001b[39m\u001b[31m (call to julia.get_gc_frame_slot)\u001b[39m\n\u001b[31mReason: unsupported dynamic function invocation\u001b[39m\nStacktrace:\n [1] \u001b[0m\u001b[1mkernel\u001b[22m\n\u001b[90m   @\u001b[39m \u001b[90m./\u001b[39m\u001b[90m\u001b[4mIn[29]:2\u001b[24m\u001b[39m\n\u001b[31mReason: unsupported dynamic function invocation\u001b[39m\u001b[31m (call to getproperty)\u001b[39m\nStacktrace:\n [1] \u001b[0m\u001b[1mkernel\u001b[22m\n\u001b[90m   @\u001b[39m \u001b[90m./\u001b[39m\u001b[90m\u001b[4mIn[29]:2\u001b[24m\u001b[39m\n\u001b[31mReason: unsupported dynamic function invocation\u001b[39m\u001b[31m (call to ==)\u001b[39m\nStacktrace:\n [1] \u001b[0m\u001b[1mkernel\u001b[22m\n\u001b[90m   @\u001b[39m \u001b[90m./\u001b[39m\u001b[90m\u001b[4mIn[29]:2\u001b[24m\u001b[39m\n\u001b[36m\u001b[1mHint\u001b[22m\u001b[39m\u001b[36m: catch this exception as `err` and call `code_typed(err; interactive = true)` to introspect the erronous code with Cthulhu.jl\u001b[39m",
      "",
      "Stacktrace:",
      "  [1] check_ir(job::GPUCompiler.CompilerJob{GPUCompiler.PTXCompilerTarget, CUDA.CUDACompilerParams}, args::LLVM.Module)",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/validation.jl:147",
      "  [2] macro expansion",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/driver.jl:440 [inlined]",
      "  [3] macro expansion",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/TimerOutputs/RsWnF/src/TimerOutput.jl:253 [inlined]",
      "  [4] macro expansion",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/driver.jl:439 [inlined]",
      "  [5] emit_llvm(job::GPUCompiler.CompilerJob; libraries::Bool, toplevel::Bool, optimize::Bool, cleanup::Bool, only_entry::Bool, validate::Bool)",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/utils.jl:92",
      "  [6] emit_llvm",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/utils.jl:86 [inlined]",
      "  [7] codegen(output::Symbol, job::GPUCompiler.CompilerJob; libraries::Bool, toplevel::Bool, optimize::Bool, cleanup::Bool, strip::Bool, validate::Bool, only_entry::Bool, parent_job::Nothing)",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/driver.jl:129",
      "  [8] codegen",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/driver.jl:110 [inlined]",
      "  [9] compile(target::Symbol, job::GPUCompiler.CompilerJob; libraries::Bool, toplevel::Bool, optimize::Bool, cleanup::Bool, strip::Bool, validate::Bool, only_entry::Bool)",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/driver.jl:106",
      " [10] compile",
      "    @ /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/driver.jl:98 [inlined]",
      " [11] #1075",
      "    @ /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/src/compiler/compilation.jl:247 [inlined]",
      " [12] JuliaContext(f::CUDA.var\"#1075#1077\"{GPUCompiler.CompilerJob{GPUCompiler.PTXCompilerTarget, CUDA.CUDACompilerParams}})",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/driver.jl:47",
      " [13] compile(job::GPUCompiler.CompilerJob)",
      "    @ CUDA /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/src/compiler/compilation.jl:246",
      " [14] actual_compilation(cache::Dict{Any, CuFunction}, src::Core.MethodInstance, world::UInt64, cfg::GPUCompiler.CompilerConfig{GPUCompiler.PTXCompilerTarget, CUDA.CUDACompilerParams}, compiler::typeof(CUDA.compile), linker::typeof(CUDA.link))",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/execution.jl:125",
      " [15] cached_compilation(cache::Dict{Any, CuFunction}, src::Core.MethodInstance, cfg::GPUCompiler.CompilerConfig{GPUCompiler.PTXCompilerTarget, CUDA.CUDACompilerParams}, compiler::Function, linker::Function)",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/execution.jl:103",
      " [16] macro expansion",
      "    @ /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/src/compiler/execution.jl:373 [inlined]",
      " [17] macro expansion",
      "    @ ./lock.jl:267 [inlined]",
      " [18] cufunction(f::typeof(kernel), tt::Type{Tuple{CuDeviceVector{Int64, 1}}}; kwargs::@Kwargs{})",
      "    @ CUDA /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/src/compiler/execution.jl:368",
      " [19] cufunction(f::typeof(kernel), tt::Type{Tuple{CuDeviceVector{Int64, 1}}})",
      "    @ CUDA /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/src/compiler/execution.jl:365",
      " [20] macro expansion",
      "    @ /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/src/compiler/execution.jl:104 [inlined]",
      " [21] macro expansion",
      "    @ /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/reflection.jl:206 [inlined]",
      " [22] top-level scope",
      "    @ In[30]:1"
     ]
    }
   ],
   "source": [
    "@device_code_warntype @cuda kernel(CuArray([1]))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In this IR, the `::Any` indicates that type-unstable code will be generated. This is unsupported on the GPU, as it will result in calls to the runtime library."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Passing invalid types\n",
    "\n",
    "Compilation errors are not limited to kernel functions, you can trigger some by using array abstractions in an invalid manner. For example:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "ename": "LoadError",
     "evalue": "GPU compilation of MethodInstance for (::GPUArrays.var\"#broadcast_kernel#38\")(::CUDA.CuKernelContext, ::CuDeviceVector{Int64, 1}, ::Base.Broadcast.Broadcasted{CUDA.CuArrayStyle{1}, Tuple{Base.OneTo{Int64}}, typeof(identity), Tuple{Base.Broadcast.Extruded{Vector{Int64}, Tuple{Bool}, Tuple{Int64}}}}, ::Int64) failed\nKernelError: passing and using non-bitstype argument\n\nArgument 4 to your kernel function is of type Base.Broadcast.Broadcasted{CUDA.CuArrayStyle{1}, Tuple{Base.OneTo{Int64}}, typeof(identity), Tuple{Base.Broadcast.Extruded{Vector{Int64}, Tuple{Bool}, Tuple{Int64}}}}, which is not isbits:\n  .args is of type Tuple{Base.Broadcast.Extruded{Vector{Int64}, Tuple{Bool}, Tuple{Int64}}} which is not isbits.\n    .1 is of type Base.Broadcast.Extruded{Vector{Int64}, Tuple{Bool}, Tuple{Int64}} which is not isbits.\n      .x is of type Vector{Int64} which is not isbits.\n\n",
     "output_type": "error",
     "traceback": [
      "GPU compilation of MethodInstance for (::GPUArrays.var\"#broadcast_kernel#38\")(::CUDA.CuKernelContext, ::CuDeviceVector{Int64, 1}, ::Base.Broadcast.Broadcasted{CUDA.CuArrayStyle{1}, Tuple{Base.OneTo{Int64}}, typeof(identity), Tuple{Base.Broadcast.Extruded{Vector{Int64}, Tuple{Bool}, Tuple{Int64}}}}, ::Int64) failed\nKernelError: passing and using non-bitstype argument\n\nArgument 4 to your kernel function is of type Base.Broadcast.Broadcasted{CUDA.CuArrayStyle{1}, Tuple{Base.OneTo{Int64}}, typeof(identity), Tuple{Base.Broadcast.Extruded{Vector{Int64}, Tuple{Bool}, Tuple{Int64}}}}, which is not isbits:\n  .args is of type Tuple{Base.Broadcast.Extruded{Vector{Int64}, Tuple{Bool}, Tuple{Int64}}} which is not isbits.\n    .1 is of type Base.Broadcast.Extruded{Vector{Int64}, Tuple{Bool}, Tuple{Int64}} which is not isbits.\n      .x is of type Vector{Int64} which is not isbits.\n\n",
      "",
      "Stacktrace:",
      "  [1] check_invocation(job::GPUCompiler.CompilerJob)",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/validation.jl:92",
      "  [2] macro expansion",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/driver.jl:123 [inlined]",
      "  [3] macro expansion",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/TimerOutputs/RsWnF/src/TimerOutput.jl:253 [inlined]",
      "  [4] codegen(output::Symbol, job::GPUCompiler.CompilerJob; libraries::Bool, toplevel::Bool, optimize::Bool, cleanup::Bool, strip::Bool, validate::Bool, only_entry::Bool, parent_job::Nothing)",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/driver.jl:121",
      "  [5] codegen",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/driver.jl:110 [inlined]",
      "  [6] compile(target::Symbol, job::GPUCompiler.CompilerJob; libraries::Bool, toplevel::Bool, optimize::Bool, cleanup::Bool, strip::Bool, validate::Bool, only_entry::Bool)",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/driver.jl:106",
      "  [7] compile",
      "    @ /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/driver.jl:98 [inlined]",
      "  [8] #1075",
      "    @ /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/src/compiler/compilation.jl:247 [inlined]",
      "  [9] JuliaContext(f::CUDA.var\"#1075#1077\"{GPUCompiler.CompilerJob{GPUCompiler.PTXCompilerTarget, CUDA.CUDACompilerParams}})",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/driver.jl:47",
      " [10] compile(job::GPUCompiler.CompilerJob)",
      "    @ CUDA /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/src/compiler/compilation.jl:246",
      " [11] actual_compilation(cache::Dict{Any, CuFunction}, src::Core.MethodInstance, world::UInt64, cfg::GPUCompiler.CompilerConfig{GPUCompiler.PTXCompilerTarget, CUDA.CUDACompilerParams}, compiler::typeof(CUDA.compile), linker::typeof(CUDA.link))",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/execution.jl:125",
      " [12] cached_compilation(cache::Dict{Any, CuFunction}, src::Core.MethodInstance, cfg::GPUCompiler.CompilerConfig{GPUCompiler.PTXCompilerTarget, CUDA.CUDACompilerParams}, compiler::Function, linker::Function)",
      "    @ GPUCompiler /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUCompiler/U36Ed/src/execution.jl:103",
      " [13] macro expansion",
      "    @ /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/src/compiler/execution.jl:373 [inlined]",
      " [14] macro expansion",
      "    @ ./lock.jl:267 [inlined]",
      " [15] cufunction(f::GPUArrays.var\"#broadcast_kernel#38\", tt::Type{Tuple{CUDA.CuKernelContext, CuDeviceVector{Int64, 1}, Base.Broadcast.Broadcasted{CUDA.CuArrayStyle{1}, Tuple{Base.OneTo{Int64}}, typeof(identity), Tuple{Base.Broadcast.Extruded{Vector{Int64}, Tuple{Bool}, Tuple{Int64}}}}, Int64}}; kwargs::@Kwargs{})",
      "    @ CUDA /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/src/compiler/execution.jl:368",
      " [16] cufunction",
      "    @ CUDA /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/src/compiler/execution.jl:365 [inlined]",
      " [17] macro expansion",
      "    @ CUDA /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/src/compiler/execution.jl:104 [inlined]",
      " [18] #launch_heuristic#1120",
      "    @ CUDA /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/src/gpuarrays.jl:17 [inlined]",
      " [19] launch_heuristic",
      "    @ CUDA /scratch/snx3000/julia/tbesard/daint-gpu/packages/CUDA/hfp7r/src/gpuarrays.jl:15 [inlined]",
      " [20] _copyto!",
      "    @ GPUArrays /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUArrays/dAUOE/src/host/broadcast.jl:70 [inlined]",
      " [21] materialize!",
      "    @ GPUArrays /scratch/snx3000/julia/tbesard/daint-gpu/packages/GPUArrays/dAUOE/src/host/broadcast.jl:46 [inlined]",
      " [22] materialize!(dest::CuArray{Int64, 1, CUDA.Mem.DeviceBuffer}, bc::Base.Broadcast.Broadcasted{Base.Broadcast.DefaultArrayStyle{1}, Nothing, typeof(identity), Tuple{Vector{Int64}}})",
      "    @ Base.Broadcast ./broadcast.jl:911",
      " [23] top-level scope",
      "    @ In[31]:1"
     ]
    }
   ],
   "source": [
    "CuArray([1]) .= [1]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Here, compilation fails because GPU code requires all arguments to be `isbits` (remember the difference between `CuArray` and `CuDeviceArray`)."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Exercise: Matrix RMSE\n",
    "\n",
    "Now that we know something about implementing kernels, and debugging issues with them, let's try implementing the RMSE function again:\n",
    "\n",
    "$$\n",
    "    RMSE = \\sqrt{\\frac{1}{N} \\sum_{i=1}^N (A_i - B_i)^2}\n",
    "$$\n",
    "\n",
    "Recall that we implemented it using array operations as such:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "rmse (generic function with 1 method)"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "rmse(A, B) = sqrt(sum((A.-B).^2) / length(A))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's start with an simplified implementation:\n",
    "\n",
    "- only use a single block\n",
    "- use a single-element array to store the result\n",
    "- compute $(A_i - B_i)^2$ on all threads, storing the result in global memory\n",
    "\n",
    "Try to keep the number of API calls and kernel launches to a minimum, i.e., perform as many operations as possible in a single kernel."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.3914029497263703"
      ]
     },
     "execution_count": 37,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "A = rand(512)\n",
    "B = rand(512)\n",
    "rmse(A, B)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [],
   "source": [
    "dA = CuArray(A)\n",
    "dB = CuArray(B)\n",
    "dC = similar(dA, 1);"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "A possible implementation would look as follows:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.3914029497263704"
      ]
     },
     "execution_count": 43,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "function rmse_kernel(C, A, B)\n",
    "    i = threadIdx().x\n",
    "\n",
    "    # initialize the memory\n",
    "    if i == 1\n",
    "        C[] = 0\n",
    "    end\n",
    "    sync_threads()\n",
    "\n",
    "    # process an element on each thread\n",
    "    a = A[i]\n",
    "    b = B[i]\n",
    "    CUDA.@atomic C[] += (a-b)^2\n",
    "    sync_threads()\n",
    "\n",
    "    # finalize the computation\n",
    "    if i == 1\n",
    "        C[1] = sqrt(C[] / length(A))\n",
    "    end\n",
    "    return\n",
    "end\n",
    "\n",
    "@cuda threads=length(dA) rmse_kernel(dC, dA, dB)\n",
    "CUDA.@allowscalar dC[]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "As we've seen before, this kernel will run into an error when it tries to launch too many threads. Instead, we need to use the launch configuration API to make sure we use a valid launch configuration:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.39140294972637024"
      ]
     },
     "execution_count": 44,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "function rmse_kernel(C, A, B)\n",
    "    i = (blockIdx().x-1) * blockDim().x + threadIdx().x\n",
    "\n",
    "    # initialize the memory\n",
    "    if i == 1\n",
    "        C[] = 0\n",
    "    end\n",
    "    sync_threads()\n",
    "\n",
    "    # process an element on each thread\n",
    "    a = A[i]\n",
    "    b = B[i]\n",
    "    CUDA.@atomic C[] += (a-b)^2\n",
    "    sync_threads()\n",
    "\n",
    "    # finalize the computation\n",
    "    if i == 1\n",
    "        C[1] = sqrt(C[] / length(A))\n",
    "    end\n",
    "    return\n",
    "end\n",
    "\n",
    "k = @cuda launch=false rmse_kernel(dC, dA, dB)\n",
    "config = CUDA.launch_configuration(k.fun)\n",
    "threads = min(length(A), config.threads)\n",
    "blocks = cld(length(A), threads)\n",
    "k(dC, dA, dB; threads, blocks)\n",
    "\n",
    "CUDA.@allowscalar dC[]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Things break! Turns out this generalization isn't always valid, here because we performed an initialization and finalization step on thread 1, however, now that we launch multiple blocks that thread does not longer run in lockstep with all other steps. As a result, we need to perform these operations before and after the kernel, from the host:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.39140294972637046"
      ]
     },
     "execution_count": 48,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "function rmse_kernel(C, A, B)\n",
    "    i = (blockIdx().x-1) * blockDim().x + threadIdx().x\n",
    "\n",
    "    if i > length(A)\n",
    "        return\n",
    "    end\n",
    "\n",
    "    # process an element on each thread\n",
    "    a = A[i]\n",
    "    b = B[i]\n",
    "    CUDA.@atomic C[] += (a-b)^2\n",
    "\n",
    "    return\n",
    "end\n",
    "\n",
    "dC .= 0                                     # <-- initialization\n",
    "\n",
    "k = @cuda launch=false rmse_kernel(dC, dA, dB)\n",
    "config = CUDA.launch_configuration(k.fun)\n",
    "threads = min(length(A), config.threads)\n",
    "blocks = cld(length(A), threads)\n",
    "k(dC, dA, dB; threads, blocks)\n",
    "\n",
    "CUDA.@allowscalar sqrt(dC[] / length(A))    # <-- finalization"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "An alternative is to generalize using a so-called grid stride loop, i.e., by having each thread process multiple elements until we exhaust the input. That means converting the following pattern:\n",
    "\n",
    "```julia\n",
    "\n",
    "function kernel(input)\n",
    "    i = (blockIdx().x - 1) * blockDim().x + threadIdx().x\n",
    "\n",
    "    if i <= length(input)\n",
    "        # do something with input[i]\n",
    "    end\n",
    "\n",
    "    return\n",
    "end\n",
    "```\n",
    "\n",
    "to:\n",
    "\n",
    "```julia\n",
    "function kernel(input)\n",
    "    i = (blockIdx().x - 1) * blockDim().x + threadIdx().x\n",
    "    stride = blockDim().x * gridDim().x\n",
    "\n",
    "    while i <= length(input)\n",
    "        # do something with input[i]\n",
    "        i += stride\n",
    "    end\n",
    "\n",
    "    return\n",
    "end\n",
    "```\n",
    "\n",
    "Note that for our simplified kernel that only uses a single block, `i` and `stride` are simply `threadIdx().x` and `blockDim().x`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.39140294972637035"
      ]
     },
     "execution_count": 49,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "function rmse_kernel(C, A, B)\n",
    "    # initialize the memory\n",
    "    if threadIdx().x == 1\n",
    "        C[1] = 0\n",
    "    end\n",
    "    sync_threads()\n",
    "\n",
    "    # grid-stride loop to process all elements\n",
    "    i = threadIdx().x\n",
    "    stride = blockDim().x\n",
    "    while i <= length(A)\n",
    "        a = A[i]\n",
    "        b = B[i]\n",
    "        CUDA.@atomic C[1] += (a-b)^2\n",
    "        i += stride\n",
    "    end\n",
    "    sync_threads()\n",
    "\n",
    "    # finalize the computation\n",
    "    if threadIdx().x == 1\n",
    "        C[1] = sqrt(C[1] / length(A))\n",
    "    end\n",
    "    return\n",
    "end\n",
    "\n",
    "@cuda threads=256 rmse_kernel(dC, dA, dB)\n",
    "CUDA.@allowscalar dC[]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Note how it doesn't matter anymore how many threads we launch, as the kernel adapts automatically. This is a very powerful pattern that generalizes kernel implementations, avoiding assumptions on the launch configuration.\n",
    "\n",
    "**HOWEVER**, it is not all great. Recall the suggested launch configuration:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(blocks = 112, threads = 1024)"
      ]
     },
     "execution_count": 50,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "k = @cuda launch=false rmse_kernel(dC, dA, dB)\n",
    "launch_configuration(k.fun)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The `blocks` in there is a suggestion on how many blocks we should launch to fully utilize our GPU. Meanwhile, we are only launching a single block, severely underutilizing our device! That doesn't mean grid stride loops are useless, as they keep our kernel flexible (dynamically adapting to the actual launch configuration), and making it possible to limit the number of blocks launched (which can sometimes hurt performance). So let's combine both approaches:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.39140294972637035"
      ]
     },
     "execution_count": 51,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "function rmse_kernel(C, A, B)\n",
    "    i = (blockIdx().x-1) * blockDim().x + threadIdx().x\n",
    "    stride = gridDim().x * blockDim().x\n",
    "\n",
    "    # grid-stride loop to process all elements\n",
    "    while i <= length(A)\n",
    "        a = A[i]\n",
    "        b = B[i]\n",
    "        CUDA.@atomic C[] += (a-b)^2\n",
    "\n",
    "        i += stride\n",
    "    end\n",
    "\n",
    "    return\n",
    "end\n",
    "\n",
    "dC .= 0\n",
    "\n",
    "k = @cuda launch=false rmse_kernel(dC, dA, dB)\n",
    "config = CUDA.launch_configuration(k.fun)\n",
    "threads = min(length(A), config.threads)\n",
    "blocks = cld(length(A), threads)\n",
    "blocks = min(blocks, config.blocks)             # <-- avoid oversubscription\n",
    "k(dC, dA, dB; threads, blocks)\n",
    "\n",
    "CUDA.@allowscalar sqrt(dC[] / length(A))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's finish up with a benchmark of different implementations. First, using array operations on the CPU:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "BenchmarkTools.Trial: 246 samples with 1 evaluation.\n",
       " Range \u001b[90m(\u001b[39m\u001b[36m\u001b[1mmin\u001b[22m\u001b[39m … \u001b[35mmax\u001b[39m\u001b[90m):  \u001b[39m\u001b[36m\u001b[1m18.846 ms\u001b[22m\u001b[39m … \u001b[35m24.527 ms\u001b[39m  \u001b[90m┊\u001b[39m GC \u001b[90m(\u001b[39mmin … max\u001b[90m): \u001b[39m0.00% … 21.67%\n",
       " Time  \u001b[90m(\u001b[39m\u001b[34m\u001b[1mmedian\u001b[22m\u001b[39m\u001b[90m):     \u001b[39m\u001b[34m\u001b[1m19.004 ms              \u001b[22m\u001b[39m\u001b[90m┊\u001b[39m GC \u001b[90m(\u001b[39mmedian\u001b[90m):    \u001b[39m0.00%\n",
       " Time  \u001b[90m(\u001b[39m\u001b[32m\u001b[1mmean\u001b[22m\u001b[39m ± \u001b[32mσ\u001b[39m\u001b[90m):   \u001b[39m\u001b[32m\u001b[1m20.319 ms\u001b[22m\u001b[39m ± \u001b[32m 2.322 ms\u001b[39m  \u001b[90m┊\u001b[39m GC \u001b[90m(\u001b[39mmean ± σ\u001b[90m):  \u001b[39m6.38% ±  9.21%\n",
       "\n",
       "  \u001b[39m▆\u001b[39m█\u001b[34m \u001b[39m\u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[32m \u001b[39m\u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m▆\u001b[39m \u001b[39m \u001b[39m \n",
       "  \u001b[39m█\u001b[39m█\u001b[34m█\u001b[39m\u001b[39m▄\u001b[39m▄\u001b[39m▁\u001b[39m▄\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[32m▁\u001b[39m\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▄\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▄\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m█\u001b[39m▅\u001b[39m \u001b[39m▅\n",
       "  18.8 ms\u001b[90m      \u001b[39m\u001b[90mHistogram: \u001b[39m\u001b[90m\u001b[1mlog(\u001b[22m\u001b[39m\u001b[90mfrequency\u001b[39m\u001b[90m\u001b[1m)\u001b[22m\u001b[39m\u001b[90m by time\u001b[39m      24.5 ms \u001b[0m\u001b[1m<\u001b[22m\n",
       "\n",
       " Memory estimate\u001b[90m: \u001b[39m\u001b[33m32.00 MiB\u001b[39m, allocs estimate\u001b[90m: \u001b[39m\u001b[33m4\u001b[39m."
      ]
     },
     "execution_count": 58,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "A = rand(2048, 2048)\n",
    "B = rand(2048, 2048)\n",
    "\n",
    "@benchmark rmse($A, $B)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Then, the straightforward GPU port:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "BenchmarkTools.Trial: 10000 samples with 1 evaluation.\n",
       " Range \u001b[90m(\u001b[39m\u001b[36m\u001b[1mmin\u001b[22m\u001b[39m … \u001b[35mmax\u001b[39m\u001b[90m):  \u001b[39m\u001b[36m\u001b[1m338.241 μs\u001b[22m\u001b[39m … \u001b[35m  5.697 ms\u001b[39m  \u001b[90m┊\u001b[39m GC \u001b[90m(\u001b[39mmin … max\u001b[90m): \u001b[39m0.00% … 14.54%\n",
       " Time  \u001b[90m(\u001b[39m\u001b[34m\u001b[1mmedian\u001b[22m\u001b[39m\u001b[90m):     \u001b[39m\u001b[34m\u001b[1m360.474 μs               \u001b[22m\u001b[39m\u001b[90m┊\u001b[39m GC \u001b[90m(\u001b[39mmedian\u001b[90m):    \u001b[39m0.00%\n",
       " Time  \u001b[90m(\u001b[39m\u001b[32m\u001b[1mmean\u001b[22m\u001b[39m ± \u001b[32mσ\u001b[39m\u001b[90m):   \u001b[39m\u001b[32m\u001b[1m373.237 μs\u001b[22m\u001b[39m ± \u001b[32m209.246 μs\u001b[39m  \u001b[90m┊\u001b[39m GC \u001b[90m(\u001b[39mmean ± σ\u001b[90m):  \u001b[39m0.45% ±  0.74%\n",
       "\n",
       "  \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m▂\u001b[39m▆\u001b[39m▆\u001b[39m▃\u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m▄\u001b[34m▅\u001b[39m\u001b[39m▁\u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[32m \u001b[39m\u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m█\u001b[39m▆\u001b[39m▁\u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \n",
       "  \u001b[39m▁\u001b[39m▁\u001b[39m▂\u001b[39m▂\u001b[39m▅\u001b[39m▆\u001b[39m▇\u001b[39m▆\u001b[39m▆\u001b[39m▃\u001b[39m▂\u001b[39m▃\u001b[39m▄\u001b[39m█\u001b[39m█\u001b[39m█\u001b[39m█\u001b[39m█\u001b[39m▅\u001b[39m▃\u001b[39m▂\u001b[39m▆\u001b[39m█\u001b[34m█\u001b[39m\u001b[39m█\u001b[39m▄\u001b[39m▂\u001b[39m▃\u001b[39m▄\u001b[39m▃\u001b[39m▂\u001b[39m▂\u001b[39m▃\u001b[39m▂\u001b[39m▂\u001b[39m▂\u001b[32m▂\u001b[39m\u001b[39m▂\u001b[39m▂\u001b[39m▃\u001b[39m▃\u001b[39m▂\u001b[39m▃\u001b[39m█\u001b[39m█\u001b[39m█\u001b[39m█\u001b[39m▄\u001b[39m▃\u001b[39m▂\u001b[39m▂\u001b[39m▂\u001b[39m▂\u001b[39m▂\u001b[39m▂\u001b[39m▂\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m \u001b[39m▃\n",
       "  338 μs\u001b[90m           Histogram: frequency by time\u001b[39m          398 μs \u001b[0m\u001b[1m<\u001b[22m\n",
       "\n",
       " Memory estimate\u001b[90m: \u001b[39m\u001b[33m6.23 KiB\u001b[39m, allocs estimate\u001b[90m: \u001b[39m\u001b[33m113\u001b[39m."
      ]
     },
     "execution_count": 59,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dA = CuArray(A)\n",
    "dB = CuArray(B)\n",
    "@benchmark rmse($dA, $dB)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "And finally, our custom kernel:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "BenchmarkTools.Trial: 505 samples with 1 evaluation.\n",
       " Range \u001b[90m(\u001b[39m\u001b[36m\u001b[1mmin\u001b[22m\u001b[39m … \u001b[35mmax\u001b[39m\u001b[90m):  \u001b[39m\u001b[36m\u001b[1m9.883 ms\u001b[22m\u001b[39m … \u001b[35m9.941 ms\u001b[39m  \u001b[90m┊\u001b[39m GC \u001b[90m(\u001b[39mmin … max\u001b[90m): \u001b[39m0.00% … 0.00%\n",
       " Time  \u001b[90m(\u001b[39m\u001b[34m\u001b[1mmedian\u001b[22m\u001b[39m\u001b[90m):     \u001b[39m\u001b[34m\u001b[1m9.901 ms             \u001b[22m\u001b[39m\u001b[90m┊\u001b[39m GC \u001b[90m(\u001b[39mmedian\u001b[90m):    \u001b[39m0.00%\n",
       " Time  \u001b[90m(\u001b[39m\u001b[32m\u001b[1mmean\u001b[22m\u001b[39m ± \u001b[32mσ\u001b[39m\u001b[90m):   \u001b[39m\u001b[32m\u001b[1m9.901 ms\u001b[22m\u001b[39m ± \u001b[32m5.407 μs\u001b[39m  \u001b[90m┊\u001b[39m GC \u001b[90m(\u001b[39mmean ± σ\u001b[90m):  \u001b[39m0.00% ± 0.00%\n",
       "\n",
       "  \u001b[39m \u001b[39m▁\u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m▅\u001b[39m█\u001b[34m▇\u001b[39m\u001b[39m▅\u001b[39m▄\u001b[39m▂\u001b[39m \u001b[39m \u001b[39m \u001b[39m▂\u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \u001b[39m \n",
       "  \u001b[39m▆\u001b[39m█\u001b[39m█\u001b[39m▆\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▆\u001b[39m█\u001b[39m█\u001b[34m█\u001b[39m\u001b[39m█\u001b[39m█\u001b[39m█\u001b[39m█\u001b[39m█\u001b[39m█\u001b[39m█\u001b[39m▆\u001b[39m▅\u001b[39m▄\u001b[39m▅\u001b[39m▁\u001b[39m▁\u001b[39m▁\u001b[39m▄\u001b[39m▁\u001b[39m▁\u001b[39m▄\u001b[39m▄\u001b[39m▄\u001b[39m▁\u001b[39m▁\u001b[39m▄\u001b[39m▅\u001b[39m▅\u001b[39m▁\u001b[39m▄\u001b[39m \u001b[39m▇\n",
       "  9.88 ms\u001b[90m     \u001b[39m\u001b[90mHistogram: \u001b[39m\u001b[90m\u001b[1mlog(\u001b[22m\u001b[39m\u001b[90mfrequency\u001b[39m\u001b[90m\u001b[1m)\u001b[22m\u001b[39m\u001b[90m by time\u001b[39m     9.92 ms \u001b[0m\u001b[1m<\u001b[22m\n",
       "\n",
       " Memory estimate\u001b[90m: \u001b[39m\u001b[33m1.20 KiB\u001b[39m, allocs estimate\u001b[90m: \u001b[39m\u001b[33m23\u001b[39m."
      ]
     },
     "execution_count": 61,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dC = similar(dA, 1)\n",
    "\n",
    "k = @cuda launch=false rmse_kernel(dC, dA, dB)\n",
    "config = CUDA.launch_configuration(k.fun)\n",
    "threads = min(length(A), config.threads)\n",
    "blocks = cld(length(A), threads)\n",
    "blocks = min(blocks, config.blocks)\n",
    "\n",
    "@benchmark CUDA.@sync k(dC, dA, dB; threads, blocks)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Well, that's unexpectedly slow! We'll look into why that is in the next notebook."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## High-level kernel programming\n",
    "\n",
    "Before finishing this notebook, I want to namedrop a couple of packages that aim to simplify kernel programming without resorting to array operations (which generally aren't as flexible)."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Tullio.jl\n",
    "\n",
    "With Tullio, it's easy to write kernels using index notation. This makes it easy to express operations like our RMSE calculation in a single expression which typically will also be compiled to a single kernel:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\u001b[36m\u001b[1m[ \u001b[22m\u001b[39m\u001b[36m\u001b[1mInfo: \u001b[22m\u001b[39mPrecompiling TullioCUDAExt [4e6b9f75-c769-5b50-88dd-0f9543e5de10]\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "10×10 Matrix{Float64}:\n",
       " 1.0  1.0  1.0  1.0  1.0  1.0  1.0  1.0  1.0  1.0\n",
       " 1.0  1.0  1.0  1.0  1.0  1.0  1.0  1.0  1.0  1.0\n",
       " 1.0  1.0  1.0  1.0  1.0  1.0  1.0  1.0  1.0  1.0\n",
       " 1.0  1.0  1.0  1.0  1.0  1.0  1.0  1.0  1.0  1.0\n",
       " 1.0  1.0  1.0  1.0  1.0  1.0  1.0  1.0  1.0  1.0\n",
       " 1.0  1.0  1.0  1.0  1.0  1.0  1.0  1.0  1.0  1.0\n",
       " 1.0  1.0  1.0  1.0  1.0  1.0  1.0  1.0  1.0  1.0\n",
       " 1.0  1.0  1.0  1.0  1.0  1.0  1.0  1.0  1.0  1.0\n",
       " 1.0  1.0  1.0  1.0  1.0  1.0  1.0  1.0  1.0  1.0\n",
       " 1.0  1.0  1.0  1.0  1.0  1.0  1.0  1.0  1.0  1.0"
      ]
     },
     "execution_count": 62,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "using Tullio\n",
    "\n",
    "A = ones(10, 10)\n",
    "\n",
    "# assigning with `:=` creates a new array\n",
    "@tullio C[i,j] := A[i,j]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "10-element Vector{Float64}:\n",
       " 10.0\n",
       " 10.0\n",
       " 10.0\n",
       " 10.0\n",
       " 10.0\n",
       " 10.0\n",
       " 10.0\n",
       " 10.0\n",
       " 10.0\n",
       " 10.0"
      ]
     },
     "execution_count": 63,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# dropping an index will sum across that dimension\n",
    "@tullio C[i] := A[i,j]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0-dimensional Array{Float64, 0}:\n",
       "2.0"
      ]
     },
     "execution_count": 64,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# pipe operators can be used to apply functions 'outside' of the reduction\n",
    "@tullio C[] := A[i,j] |> log10(_)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "With that explained, try to implement the matrix RMSE operation using index notation. Recall the original operation (first on CPU arrays):"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.4084252657854481"
      ]
     },
     "execution_count": 65,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "A = rand(1024, 1024)\n",
    "B = rand(1024, 1024)\n",
    "sqrt(sum((A-B).^2)/length(A))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now with Tullio:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0-dimensional Array{Float64, 0}:\n",
       "0.40842526578544525"
      ]
     },
     "execution_count": 66,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "@tullio C[] := (A[i,j] - B[i,j])^2 |> sqrt(_ / length(A))\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "To use Tullio on the GPU, we simply use CuArrays everywhere:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0-dimensional CuArray{Float64, 0, CUDA.Mem.DeviceBuffer}:\n",
       "0.40804403168851916"
      ]
     },
     "execution_count": 76,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dA = CuArray(A)\n",
    "dB = CuArray(B)\n",
    "@tullio dC[] := (dA[i,j] - dB[i,j])^2 |> sqrt(_ / length(dA))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Tullio is great for quickly creating portable kernels (CPU, different GPU back-ends) for mathematical operations, and it can be seen as a generalization of broadcast. Sadly, for the purpose of our RMSE kernel, it does not efficiently support reduction operations on the GPU."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### KernelAbstractions.jl\n",
    "\n",
    "For a more flexible API, i.e. not restricted to Tullio's index notation, but still retaining Tullio's portability, you can consider the KernelAbstractions.jl framework that Tullio.jl is built on:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 77,
   "metadata": {},
   "outputs": [],
   "source": [
    "using KernelAbstractions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 78,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "ka_rmse_kernel (generic function with 4 methods)"
      ]
     },
     "execution_count": 78,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "@kernel function ka_rmse_kernel(C, A, B)\n",
    "    i = @index(Global)\n",
    "\n",
    "    a = A[i]\n",
    "    b = B[i]\n",
    "    KernelAbstractions.@atomic C[] += (a-b)^2\n",
    "end"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 79,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "rmse(A, B) = 0.40829909701268036\n",
      "typeof(dC) = Vector{Float64}\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "0.40829909701268235"
      ]
     },
     "execution_count": 79,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "A = rand(512, 512)\n",
    "B = rand(512, 512)\n",
    "@show rmse(A, B)\n",
    "\n",
    "backend = CPU()\n",
    "dA = KernelAbstractions.allocate(backend, eltype(A), size(A))\n",
    "KernelAbstractions.copyto!(backend, dA, A)\n",
    "dB = KernelAbstractions.allocate(backend, eltype(B), size(B))\n",
    "KernelAbstractions.copyto!(backend, dB, B)\n",
    "dC = KernelAbstractions.zeros(backend, eltype(A), 1)\n",
    "@show typeof(dC)\n",
    "\n",
    "k = ka_rmse_kernel(backend)\n",
    "k(dC, dA, dB; ndrange=size(A))\n",
    "\n",
    "sqrt(Array(dC)[] / length(A))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The programming interface is now much closer to CUDA.jl's, while retaining platform portability!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 80,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "typeof(dC) = CuArray{Float64, 1, CUDA.Mem.DeviceBuffer}\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "0.40829909701268174"
      ]
     },
     "execution_count": 80,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "backend = CUDABackend()\n",
    "dA = KernelAbstractions.allocate(backend, eltype(A), size(A))\n",
    "KernelAbstractions.copyto!(backend, dA, A)\n",
    "dB = KernelAbstractions.allocate(backend, eltype(B), size(B))\n",
    "KernelAbstractions.copyto!(backend, dB, B)\n",
    "dC = KernelAbstractions.zeros(backend, eltype(A), 1)\n",
    "@show typeof(dC)\n",
    "\n",
    "k = ka_rmse_kernel(backend)\n",
    "k(dC, dA, dB; ndrange=size(A))\n",
    "\n",
    "sqrt(Array(dC)[] / length(A))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The disadvantage of platform portability of course is that KernelAbstraction.jl's feature set is limited to the common denominator of all supported platforms. That means many CUDA features, like dynamic shared memory or warp-level programming, are not supported. The framework is also slightly opinionated wrt. indexing characteristics, so it's trickier to implement e.g. a grid-stride loop."
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Julia 1.10.0-rc1",
   "language": "julia",
   "name": "julia-1.10"
  },
  "language_info": {
   "file_extension": ".jl",
   "mimetype": "application/julia",
   "name": "julia",
   "version": "1.10.0"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}