spacegame/src/gpu_mem.c

#include "vulkan/vulkan_core.h"
#include <string.h>
#include <stdlib.h>
#include <stdio.h>
#include <stdbool.h>
#include <gpu_mem.h>

GPUMemoryType pick_memory(VkPhysicalDeviceMemoryProperties memories, uint32_t filter, VkMemoryPropertyFlags include, VkMemoryPropertyFlags exclude) {
  for(uint32_t i = 0; i < memories.memoryTypeCount; i++){
    if((filter & (1 << i))
        && ((include & memories.memoryTypes[i].propertyFlags) == include)
        && ((exclude & memories.memoryTypes[i].propertyFlags) == 0)) {
      GPUMemoryType ret = {
        .flags = memories.memoryTypes[i].propertyFlags,
        .index = i,
      };
      return ret;
    }
  }

  GPUMemoryType err = {
    .flags = 0,
    .index = 0xFFFFFFFF,
  };
  return err;
}

VkResult gpu_page_allocate(VkDevice device, VkPhysicalDeviceMemoryProperties memories, VkDeviceSize size, uint32_t filter, VkMemoryPropertyFlags include, VkMemoryPropertyFlags exclude, VkMemoryAllocateFlags allocate_flags, GPUPage** handle) {
  if(handle == NULL) {
    return VK_ERROR_VALIDATION_FAILED_EXT;
  }

  GPUPage* output = malloc(sizeof(GPUPage));
  if(output == NULL) {
    return VK_ERROR_OUT_OF_HOST_MEMORY;
  }

  GPUMemoryChunk* initial_chunk = malloc(sizeof(GPUMemoryChunk));
  if(initial_chunk == 0) {
    free(output);
    return VK_ERROR_OUT_OF_HOST_MEMORY;
  }

  initial_chunk->size = size;
  initial_chunk->offset = 0;
  initial_chunk->next = NULL;

  GPUMemoryType memory_type = pick_memory(memories, filter, include, exclude);
  if(memory_type.index == 0xFFFFFFFF) {
    free(initial_chunk);
    free(output);
    return VK_ERROR_UNKNOWN;
  }

  VkMemoryAllocateFlagsInfo allocate_flags_info = {
    .sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_FLAGS_INFO,
    .flags = allocate_flags,
  };

  VkMemoryAllocateInfo allocate_info = {
    .sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO,
    .allocationSize = size,
    .memoryTypeIndex = memory_type.index,
    .pNext = &allocate_flags_info,
  };

  VkDeviceMemory memory = VK_NULL_HANDLE;
  VkResult result = vkAllocateMemory(device, &allocate_info, 0, &memory);
  if(result != VK_SUCCESS) {
    free(initial_chunk);
    free(output);
    return result;
  }

  if(include & VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT) {
    result = vkMapMemory(device, memory, 0, size, 0, &output->ptr);
    if(result != VK_SUCCESS) {
      free(initial_chunk);
      free(output);
      vkFreeMemory(device, memory, 0);
      return result;
    }
  } else {
    output->ptr = NULL;
  }

  output->memory = memory;
  output->size = size;
  output->type = memory_type;
  output->free = initial_chunk;
  output->allocated = NULL;

  *handle = output;

  return VK_SUCCESS;
}

void gpu_page_free(VkDevice device, GPUPage* page) {
  if(page == NULL) {
    return;
  }

  GPUMemoryChunk* cur = page->free;
  while(cur != NULL) {
    GPUMemoryChunk* last = cur;
    cur = cur->next;
    free(last);
  }

  vkFreeMemory(device, page->memory, 0);
  free(page);
}

void gpu_add_allocation(GPUPage* page, GPUMemoryChunk* allocation, VkDeviceSize size, GPUMemoryChunk* prev, GPUMemoryChunk* cur) {
  if(page->allocated == NULL) {
    page->allocated = allocation;
  } else {
    GPUMemoryChunk* alloc_cur = page->allocated;
    while(alloc_cur->next != NULL) {
      alloc_cur = alloc_cur->next;
    }
    alloc_cur->next = allocation;
  }

  if(cur->size == size && prev == NULL) {
    free(cur);
    page->free = NULL;
  } else if(cur->size == size && prev != NULL) {
    prev->next = cur->next;
    free(cur);
  } else if(cur->size > size) {
    cur->offset += size;
    cur->size -= size;
  }

}

VkResult gpu_new_allocation(GPUPage* page, GPUMemoryChunk** prev, GPUMemoryChunk** cur, GPUMemoryChunk** allocation, VkDeviceSize size) {
  if(prev == NULL || cur == NULL || allocation == NULL) {
    return VK_ERROR_VALIDATION_FAILED_EXT;
  }

  *cur = page->free;
  *prev = NULL;


  // Find a chunk
  while(*cur != NULL) {
    if((*cur)->size >= size) {
      break;
    }
    *prev = *cur;
    *cur = (*cur)->next;
  }

  if(*cur == NULL) {
    return VK_ERROR_OUT_OF_DEVICE_MEMORY;
  }

  *allocation = malloc(sizeof(GPUMemoryChunk));
  if(*allocation == NULL) {
    return VK_ERROR_OUT_OF_HOST_MEMORY;
  }


  (*allocation)->next = NULL;
  (*allocation)->size = size;
  (*allocation)->offset = (*cur)->offset;

  return VK_SUCCESS;
}

VkResult gpu_image_malloc(VkDevice device, GPUPage* page, VkImageCreateInfo* info, GPUImage* image) {
  if(image == NULL || info == NULL || page == NULL) {
    return VK_ERROR_VALIDATION_FAILED_EXT;
  }

  VkResult result = vkCreateImage(device, info, 0, &image->handle);
  if(result != VK_SUCCESS) {
    return result;
  }

  VkMemoryRequirements requirements;
  vkGetImageMemoryRequirements(device, image->handle, &requirements);

  GPUMemoryChunk* cur;
  GPUMemoryChunk* prev;
  GPUMemoryChunk* allocation;

  result = gpu_new_allocation(page, &prev, &cur, &allocation, requirements.size);
  if(result != VK_SUCCESS) {
    return result;
  }

  result = vkBindImageMemory(device, image->handle, page->memory, cur->offset);
  if(result != VK_SUCCESS) {
    return result;
  }

  image->page = page;
  image->memory = allocation;

  gpu_add_allocation(page, allocation, requirements.size, prev, cur);

  return VK_SUCCESS;
}

VkResult gpu_buffer_malloc(VkDevice device, GPUPage* page, VkDeviceSize size, VkBufferUsageFlags usage, GPUBuffer* buffer) {
  if(buffer == NULL || page == NULL) {
    return VK_ERROR_VALIDATION_FAILED_EXT;
  }

  //TODO: use real alignment size instead of hard-coded to 16
  size += size % 16;

  VkBufferCreateInfo buffer_info = {
    .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
    .size = size,
    .usage = usage,
    .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
  };

  VkResult result = vkCreateBuffer(device, &buffer_info, 0, &buffer->handle);
  if(result != VK_SUCCESS) {
    return result;
  }

  GPUMemoryChunk* cur;
  GPUMemoryChunk* prev;
  GPUMemoryChunk* allocation;

  result = gpu_new_allocation(page, &prev, &cur, &allocation, size);
  if(result != VK_SUCCESS) {
    return result;
  }

  result = vkBindBufferMemory(device, buffer->handle, page->memory, cur->offset);
  if(result != VK_SUCCESS) {
    return result;
  }

  buffer->page = page;
  buffer->memory = allocation;

  gpu_add_allocation(page, allocation, size, prev, cur);

  return VK_SUCCESS;
}

void gpu_image_free(VkDevice device, GPUImage image) {
  vkDestroyImage(device, image.handle, 0);
  gpu_free(image.page, image.memory);
}

void gpu_buffer_free(VkDevice device, GPUBuffer buffer) {
  vkDestroyBuffer(device, buffer.handle, 0);
  gpu_free(buffer.page, buffer.memory);
}

void gpu_free(GPUPage* page, GPUMemoryChunk* memory) {

  if(memory == page->allocated) {
    page->allocated = memory->next;
  } else {
    GPUMemoryChunk* cur = page->allocated;
    while(cur->next != NULL) {
      if(cur->next == memory) {
        cur->next = memory->next;
        break;
      }
      cur = cur->next;
    }
    if(cur == NULL) {
      return;
    }
  }

  memory->next = NULL;

  GPUMemoryChunk* free_cur = page->free;
  GPUMemoryChunk* free_prev = NULL;
  while(free_cur != NULL) {
    if(free_cur->offset > memory->offset) {
      break;
    }
    free_prev = free_cur;
    free_cur = free_cur->next;
  }
  if(free_cur == NULL && free_prev == NULL) {
    page->free = memory;
  } else {
    bool left_cont = false;
    if (free_prev != NULL) {
      left_cont = ((free_prev->offset + free_prev->size) == memory->offset);
    }
    bool right_cont = false;
    if (free_cur != NULL) {
      right_cont = ((memory->offset + memory->size) == free_cur->offset);
    }

    fprintf(stderr, "l: %d, r: %d\n", left_cont, right_cont);

    if(left_cont && right_cont) {
      free_prev->next = free_cur->next;
      free_prev->size += free_cur->size;
      free_prev->size += memory->size;

      free(free_cur);
      free(memory);
    } else if(!left_cont && right_cont) {
      free_cur->offset -= memory->size;
      free_cur->size += memory->size;
      free(memory);
    } else if(left_cont && !right_cont) {
      free_prev->size += memory->size;
      free(memory);
    } else if(!left_cont && !right_cont) {
      if(free_cur == NULL) {
        memory->next = NULL;
      } else {
        memory->next = free_cur->next;
      }

      if(free_prev == NULL) {
        memory->next = page->free;
        page->free = memory;
      } else {
        free_prev->next = memory;
      }
    }
  }
}

void fprintchunks(FILE* out, GPUMemoryChunk* start) {
  if(start == NULL) {
    fprintf(out, "Chunks: {}\n");
    return;
  }

  fprintf(out, "Chunks: {");
  GPUMemoryChunk* cur = start;
  while(cur != NULL) {
    if(cur->next == NULL) {
      fprintf(out, "%llu@%llu}", cur->size, cur->offset);
    } else {
      fprintf(out, "%llu@%llu, ", cur->size, cur->offset);
    }
    cur = cur->next;
  }
  fprintf(out, "\n");
}