Initial commit.

7 files changed, 268 insertions, 0 deletions

diff --git a/CMakeLists.txt b/CMakeLists.txt
new file mode 100644
index 0000000..c3ae680
--- /dev/null
+++ b/CMakeLists.txt
@@ -0,0 +1,6 @@
+cmake_minimum_required(VERSION 3.28)
+
+add_subdirectory(hello)
+add_subdirectory(julia)
+add_subdirectory(ptracer)
+add_subdirectory(vector_sum)
diff --git a/hello/CMakeLists.txt b/hello/CMakeLists.txt
new file mode 100644
index 0000000..e4b4acc
--- /dev/null
+++ b/hello/CMakeLists.txt
@@ -0,0 +1,11 @@
+cmake_minimum_required(VERSION 3.28)
+
+project(cuda_hello LANGUAGES CUDA CXX)
+
+add_executable(cuda_hello
+  hello.cu)
+
+# -Wpedantic causes warnings due to nvcc emitting non-standard (gcc-specific)
+# host code.
+# https://stackoverflow.com/questions/31000996/warning-when-compiling-cu-with-wpedantic-style-of-line-directive-is-a-gcc-ex
+target_compile_options(cuda_hello PRIVATE -Wall -Wextra -Wno-pedantic)
diff --git a/hello/hello.cu b/hello/hello.cu
new file mode 100644
index 0000000..691b18c
--- /dev/null
+++ b/hello/hello.cu
@@ -0,0 +1,59 @@
+#include <cstdio>
+
+void logDevices() {
+  int count;
+  if (cudaGetDeviceCount(&count) != cudaSuccess) {
+    printf("No CUDA devices found\n");
+    return;
+  }
+
+  printf("CUDA devices found: %d\n", count);
+  for (int i = 0; i < count; ++i) {
+    cudaDeviceProp properties;
+    if (cudaGetDeviceProperties(&properties, i) == cudaSuccess) {
+      printf("Device [%d]: %s\n", i, properties.name);
+    }
+  }
+}
+
+__global__ void kernel(int* array, int N) {
+  for (int i = 0; i < N; ++i) {
+    array[i] = i;
+  }
+}
+
+int main() {
+  logDevices();
+
+  constexpr int N = 100;
+
+  int* host_array   = new int[N];
+  int* device_array = nullptr;
+  bool success      = false;
+
+  if (cudaMalloc(&device_array, N * sizeof(int)) != cudaSuccess) {
+    goto cleanup;
+  }
+
+  kernel<<<1, 1>>>(device_array, N);
+
+  if (cudaMemcpy(
+          host_array, device_array, N * sizeof(int), cudaMemcpyDeviceToHost) !=
+      cudaSuccess) {
+    goto cleanup;
+  }
+
+  for (int i = 0; i < N; ++i) {
+    printf("%d ", host_array[i]);
+  }
+  printf("\n");
+
+  success = true;
+
+cleanup:
+  delete[] host_array;
+  if (device_array != nullptr) {
+    cudaFree(device_array);
+  }
+  return success ? 0 : 1;
+}
diff --git a/julia/CMakeLists.txt b/julia/CMakeLists.txt
new file mode 100644
index 0000000..e5428fb
--- /dev/null
+++ b/julia/CMakeLists.txt
@@ -0,0 +1,11 @@
+cmake_minimum_required(VERSION 3.28)
+
+project(cuda_julia LANGUAGES CUDA CXX)
+
+add_executable(cuda_julia
+  julia.cu)
+
+# -Wpedantic causes warnings due to nvcc emitting non-standard (gcc-specific)
+# host code.
+# https://stackoverflow.com/questions/31000996/warning-when-compiling-cu-with-wpedantic-style-of-line-directive-is-a-gcc-ex
+target_compile_options(cuda_julia PRIVATE -Wall -Wextra -Wno-pedantic)
diff --git a/julia/julia.cu b/julia/julia.cu
new file mode 100644
index 0000000..f3ecb80
--- /dev/null
+++ b/julia/julia.cu
@@ -0,0 +1,108 @@
+#include <cstdint>
+#include <cstdio>
+#include <cstdlib>
+
+struct Pixel {
+  uint8_t r, g, b;
+};
+
+struct Complex {
+  float r, i;
+
+  __device__ float norm2() const { return r * r + i * i; }
+};
+
+__device__ Complex operator*(Complex a, Complex b) {
+  return Complex{(a.r * b.r) - (a.i * b.i), (a.i * b.r) + (a.r * b.i)};
+}
+
+__device__ Complex operator+(Complex a, Complex b) {
+  return Complex{a.r + b.r, a.i + b.i};
+}
+
+__device__ int julia(int width, int height, int x, int y) {
+  constexpr float scale = 1.5;
+  constexpr int   N     = 200;
+
+  const float jx = scale * (width / 2 - x) / (width / 2);
+  const float jy = scale * (height / 2 - y) / (height / 2);
+
+  const Complex c{-0.8, 0.156};
+  Complex       a{jx, jy};
+
+  for (int i = 0; i < N; ++i) {
+    a = a * a + c;
+    if (a.norm2() > 1000) {
+      return 0;
+    }
+  }
+  return 1;
+}
+
+__global__ void juliaMain(int width, int height, Pixel* image) {
+  const int x = blockIdx.x;
+  const int y = blockIdx.y;
+
+  constexpr Pixel background{41, 95, 152};
+  constexpr Pixel juliaColour{228, 192, 135};
+
+  const Pixel pixel =
+      julia(width, height, x, y) == 1 ? juliaColour : background;
+
+  image[y * width + x] = pixel;
+}
+
+bool write_pbm(const Pixel* image, int width, int height, const char* path) {
+  const size_t num_pixels = width * height;
+
+  FILE* file = fopen(path, "wb");
+  if (!file) {
+    return false;
+  }
+
+  fprintf(file, "P6\n%d %d\n255\n", width, height);
+  if (fwrite(image, sizeof(Pixel), num_pixels, file) != num_pixels) {
+    fclose(file);
+    return false;
+  }
+
+  fclose(file);
+  return true;
+}
+
+int main(int argc, const char** argv) {
+  const int width  = argc > 1 ? atoi(argv[1]) : 1920;
+  const int height = argc > 2 ? atoi(argv[2]) : 1080;
+
+  bool success = false;
+
+  const dim3 dim(width, height);
+  const int  image_size_bytes = width * height * sizeof(Pixel);
+  auto       image_host       = new Pixel[width * height];
+  Pixel*     image_dev        = nullptr;
+
+  if (cudaMalloc(&image_dev, image_size_bytes) != cudaSuccess) {
+    goto cleanup;
+  }
+
+  juliaMain<<<dim, 1>>>(width, height, image_dev);
+
+  if (cudaMemcpy(
+          image_host, image_dev, image_size_bytes, cudaMemcpyDeviceToHost) !=
+      cudaSuccess) {
+    goto cleanup;
+  }
+
+  if (!write_pbm(image_host, width, height, "julia.pbm")) {
+    goto cleanup;
+  }
+
+  success = true;
+
+cleanup:
+  delete[] image_host;
+  if (image_dev) {
+    cudaFree(image_dev);
+  }
+  return success ? 0 : 1;
+}
diff --git a/vector_sum/CMakeLists.txt b/vector_sum/CMakeLists.txt
new file mode 100644
index 0000000..1eea51b
--- /dev/null
+++ b/vector_sum/CMakeLists.txt
@@ -0,0 +1,11 @@
+cmake_minimum_required(VERSION 3.28)
+
+project(vector_sum LANGUAGES CUDA CXX)
+
+add_executable(vector_sum
+  main.cu)
+
+# -Wpedantic causes warnings due to nvcc emitting non-standard (gcc-specific)
+# host code.
+# https://stackoverflow.com/questions/31000996/warning-when-compiling-cu-with-wpedantic-style-of-line-directive-is-a-gcc-ex
+target_compile_options(vector_sum PRIVATE -Wall -Wextra -Wno-pedantic)
diff --git a/vector_sum/main.cu b/vector_sum/main.cu
new file mode 100644
index 0000000..ba2e964
--- /dev/null
+++ b/vector_sum/main.cu
@@ -0,0 +1,62 @@
+#include <cstdio>
+
+__global__ void add(int N, int* a, int* b, int* out) {
+  const int id = blockIdx.x;
+  out[id]      = a[id] + b[id];
+}
+
+int main() {
+  constexpr int N = 100;
+
+  bool success       = false;
+  int  host_array[N] = {0};
+  int* dev_arrays[3] = {nullptr};
+
+  // Allocate device arrays.
+  for (int i = 0; i < 3; ++i) {
+    if (cudaMalloc(&dev_arrays[i], N * sizeof(int)) != cudaSuccess) {
+      goto cleanup;
+    }
+  }
+
+  // Fill the host array with values 0..N-1.
+  for (int i = 0; i < N; ++i) {
+    host_array[i] = i;
+  }
+
+  // Copy the host array to each of the first two device arrays.
+  for (int i = 0; i < 2; ++i) {
+    if (cudaMemcpy(
+            dev_arrays[i], host_array, N * sizeof(int),
+            cudaMemcpyHostToDevice) != cudaSuccess) {
+      goto cleanup;
+    }
+  }
+
+  // Add the first two arrays.
+  // N blocks, 1 thread per block.
+  add<<<N, 1>>>(N, dev_arrays[0], dev_arrays[1], dev_arrays[2]);
+
+  // Copy the result from the third array to the host.
+  if (cudaMemcpy(
+          host_array, dev_arrays[2], N * sizeof(int), cudaMemcpyDeviceToHost) !=
+      cudaSuccess) {
+    goto cleanup;
+  }
+
+  // Print the result.
+  for (int i = 0; i < N; ++i) {
+    printf("%d ", host_array[i]);
+  }
+  printf("\n");
+
+  success = true;
+
+cleanup:
+  for (int i = 0; i < 3; ++i) {
+    if (dev_arrays[i] != nullptr) {
+      cudaFree(dev_arrays[i]);
+    }
+  }
+  return success ? 0 : 1;
+}