mirror of https://github.com/bjornbytes/lovr.git
658 lines
23 KiB
C
658 lines
23 KiB
C
#include "graphics/graphics.h"
|
|
#include "data/image.h"
|
|
#include "core/gpu.h"
|
|
#include "core/maf.h"
|
|
#include "core/os.h"
|
|
#include "util.h"
|
|
#include <math.h>
|
|
#include <stdlib.h>
|
|
#include <string.h>
|
|
|
|
#define MAX_FRAME_MEMORY (1 << 30)
|
|
|
|
struct Buffer {
|
|
uint32_t ref;
|
|
uint32_t size;
|
|
gpu_buffer* gpu;
|
|
BufferInfo info;
|
|
char* pointer;
|
|
};
|
|
|
|
struct Texture {
|
|
uint32_t ref;
|
|
gpu_texture* gpu;
|
|
gpu_texture* renderView;
|
|
TextureInfo info;
|
|
};
|
|
|
|
struct Sampler {
|
|
uint32_t ref;
|
|
gpu_sampler* gpu;
|
|
SamplerInfo info;
|
|
};
|
|
|
|
struct Pass {
|
|
uint32_t ref;
|
|
PassInfo info;
|
|
gpu_stream* stream;
|
|
float* transform;
|
|
uint32_t transformIndex;
|
|
float transforms[16][16];
|
|
};
|
|
|
|
typedef struct {
|
|
char* memory;
|
|
uint32_t cursor;
|
|
uint32_t length;
|
|
} Allocator;
|
|
|
|
static struct {
|
|
bool initialized;
|
|
bool active;
|
|
uint32_t tick;
|
|
Pass* transfers;
|
|
gpu_device_info device;
|
|
gpu_features features;
|
|
gpu_limits limits;
|
|
Allocator allocator;
|
|
} state;
|
|
|
|
// Helpers
|
|
|
|
static void* tempAlloc(size_t size);
|
|
static void beginFrame(void);
|
|
static gpu_stream* getTransfers(void);
|
|
static size_t measureTexture(TextureFormat format, uint16_t w, uint16_t h, uint16_t d);
|
|
static void onMessage(void* context, const char* message, bool severe);
|
|
|
|
// Entry
|
|
|
|
bool lovrGraphicsInit(bool debug) {
|
|
if (state.initialized) return false;
|
|
|
|
float16Init();
|
|
|
|
gpu_config config = {
|
|
.debug = debug,
|
|
.callback = onMessage,
|
|
.engineName = "LOVR",
|
|
.engineVersion = { LOVR_VERSION_MAJOR, LOVR_VERSION_MINOR, LOVR_VERSION_PATCH },
|
|
.device = &state.device,
|
|
.features = &state.features,
|
|
.limits = &state.limits
|
|
};
|
|
|
|
if (!gpu_init(&config)) {
|
|
lovrThrow("Failed to initialize GPU");
|
|
}
|
|
|
|
// Temporary frame memory uses a large 1GB virtual memory allocation, committing pages as needed
|
|
state.allocator.length = 1 << 14;
|
|
state.allocator.memory = os_vm_init(MAX_FRAME_MEMORY);
|
|
os_vm_commit(state.allocator.memory, state.allocator.length);
|
|
|
|
state.initialized = true;
|
|
return true;
|
|
}
|
|
|
|
void lovrGraphicsDestroy() {
|
|
if (!state.initialized) return;
|
|
gpu_destroy();
|
|
os_vm_free(state.allocator.memory, MAX_FRAME_MEMORY);
|
|
memset(&state, 0, sizeof(state));
|
|
}
|
|
|
|
void lovrGraphicsGetDevice(GraphicsDevice* device) {
|
|
device->deviceId = state.device.deviceId;
|
|
device->vendorId = state.device.vendorId;
|
|
device->name = state.device.deviceName;
|
|
device->renderer = state.device.renderer;
|
|
device->subgroupSize = state.device.subgroupSize;
|
|
device->discrete = state.device.discrete;
|
|
}
|
|
|
|
void lovrGraphicsGetFeatures(GraphicsFeatures* features) {
|
|
features->textureBC = state.features.textureBC;
|
|
features->textureASTC = state.features.textureASTC;
|
|
features->wireframe = state.features.wireframe;
|
|
features->depthClamp = state.features.depthClamp;
|
|
features->indirectDrawFirstInstance = state.features.indirectDrawFirstInstance;
|
|
features->float64 = state.features.float64;
|
|
features->int64 = state.features.int64;
|
|
features->int16 = state.features.int16;
|
|
}
|
|
|
|
void lovrGraphicsGetLimits(GraphicsLimits* limits) {
|
|
limits->textureSize2D = state.limits.textureSize2D;
|
|
limits->textureSize3D = state.limits.textureSize3D;
|
|
limits->textureSizeCube = state.limits.textureSizeCube;
|
|
limits->textureLayers = state.limits.textureLayers;
|
|
limits->renderSize[0] = state.limits.renderSize[0];
|
|
limits->renderSize[1] = state.limits.renderSize[1];
|
|
limits->renderSize[2] = state.limits.renderSize[2];
|
|
limits->uniformBufferRange = state.limits.uniformBufferRange;
|
|
limits->storageBufferRange = state.limits.storageBufferRange;
|
|
limits->uniformBufferAlign = state.limits.uniformBufferAlign;
|
|
limits->storageBufferAlign = state.limits.storageBufferAlign;
|
|
limits->vertexAttributes = state.limits.vertexAttributes;
|
|
limits->vertexBufferStride = state.limits.vertexBufferStride;
|
|
limits->vertexShaderOutputs = state.limits.vertexShaderOutputs;
|
|
limits->clipDistances = state.limits.clipDistances;
|
|
limits->cullDistances = state.limits.cullDistances;
|
|
limits->clipAndCullDistances = state.limits.clipAndCullDistances;
|
|
memcpy(limits->computeDispatchCount, state.limits.computeDispatchCount, 3 * sizeof(uint32_t));
|
|
memcpy(limits->computeWorkgroupSize, state.limits.computeWorkgroupSize, 3 * sizeof(uint32_t));
|
|
limits->computeWorkgroupVolume = state.limits.computeWorkgroupVolume;
|
|
limits->computeSharedMemory = state.limits.computeSharedMemory;
|
|
limits->shaderConstantSize = state.limits.pushConstantSize;
|
|
limits->indirectDrawCount = state.limits.indirectDrawCount;
|
|
limits->instances = state.limits.instances;
|
|
limits->anisotropy = state.limits.anisotropy;
|
|
limits->pointSize = state.limits.pointSize;
|
|
}
|
|
|
|
bool lovrGraphicsIsFormatSupported(uint32_t format, uint32_t features) {
|
|
uint8_t supports = state.features.formats[format];
|
|
if (!features) return supports;
|
|
if ((features & TEXTURE_FEATURE_SAMPLE) && !(supports & GPU_FEATURE_SAMPLE)) return false;
|
|
if ((features & TEXTURE_FEATURE_FILTER) && !(supports & GPU_FEATURE_FILTER)) return false;
|
|
if ((features & TEXTURE_FEATURE_RENDER) && !(supports & GPU_FEATURE_RENDER)) return false;
|
|
if ((features & TEXTURE_FEATURE_BLEND) && !(supports & GPU_FEATURE_BLEND)) return false;
|
|
if ((features & TEXTURE_FEATURE_STORAGE) && !(supports & GPU_FEATURE_STORAGE)) return false;
|
|
if ((features & TEXTURE_FEATURE_ATOMIC) && !(supports & GPU_FEATURE_ATOMIC)) return false;
|
|
if ((features & TEXTURE_FEATURE_BLIT_SRC) && !(supports & GPU_FEATURE_BLIT_SRC)) return false;
|
|
if ((features & TEXTURE_FEATURE_BLIT_DST) && !(supports & GPU_FEATURE_BLIT_DST)) return false;
|
|
return true;
|
|
}
|
|
|
|
void lovrGraphicsSubmit(Pass** passes, uint32_t count) {
|
|
if (!state.active) {
|
|
return;
|
|
}
|
|
|
|
// Allocate a few extra stream handles for any internal passes we sneak in
|
|
gpu_stream** streams = tempAlloc((count + 3) * sizeof(gpu_stream*));
|
|
|
|
uint32_t extraPassCount = 0;
|
|
|
|
if (state.transfers) {
|
|
streams[extraPassCount++] = state.transfers->stream;
|
|
}
|
|
|
|
for (uint32_t i = 0; i < count; i++) {
|
|
streams[extraPassCount + i] = passes[i]->stream;
|
|
}
|
|
|
|
for (uint32_t i = 0; i < extraPassCount + count; i++) {
|
|
gpu_stream_end(streams[i]);
|
|
}
|
|
|
|
gpu_submit(streams, extraPassCount + count);
|
|
|
|
state.transfers = NULL;
|
|
state.active = false;
|
|
}
|
|
|
|
void lovrGraphicsWait() {
|
|
gpu_wait();
|
|
}
|
|
|
|
// Buffer
|
|
|
|
Buffer* lovrGraphicsGetBuffer(BufferInfo* info, void** data) {
|
|
uint32_t size = info->length * info->stride;
|
|
lovrCheck(size > 0, "Buffer size can not be zero");
|
|
lovrCheck(size <= 1 << 30, "Max buffer size is 1GB");
|
|
|
|
Buffer* buffer = tempAlloc(sizeof(Buffer) + gpu_sizeof_buffer());
|
|
buffer->ref = 1;
|
|
buffer->size = size;
|
|
buffer->gpu = (gpu_buffer*) (buffer + 1);
|
|
buffer->info = *info;
|
|
|
|
buffer->pointer = gpu_map(buffer->gpu, size, state.limits.uniformBufferAlign, GPU_MAP_WRITE);
|
|
|
|
if (data) {
|
|
*data = buffer->pointer;
|
|
}
|
|
|
|
return buffer;
|
|
}
|
|
|
|
Buffer* lovrBufferCreate(BufferInfo* info, void** data) {
|
|
uint32_t size = info->length * info->stride;
|
|
lovrCheck(size > 0, "Buffer size can not be zero");
|
|
lovrCheck(size <= 1 << 30, "Max buffer size is 1GB");
|
|
|
|
Buffer* buffer = calloc(1, sizeof(Buffer) + gpu_sizeof_buffer());
|
|
lovrAssert(buffer, "Out of memory");
|
|
buffer->ref = 1;
|
|
buffer->size = size;
|
|
buffer->gpu = (gpu_buffer*) (buffer + 1);
|
|
buffer->info = *info;
|
|
|
|
gpu_buffer_init(buffer->gpu, &(gpu_buffer_info) {
|
|
.size = buffer->size,
|
|
.label = info->label,
|
|
.pointer = data
|
|
});
|
|
|
|
if (data && *data == NULL) {
|
|
gpu_buffer* scratchpad = tempAlloc(gpu_sizeof_buffer());
|
|
*data = gpu_map(scratchpad, size, 4, GPU_MAP_WRITE);
|
|
// TODO copy scratchpad to buffer
|
|
}
|
|
|
|
return buffer;
|
|
}
|
|
|
|
void lovrBufferDestroy(void* ref) {
|
|
Buffer* buffer = ref;
|
|
if (buffer->pointer) return;
|
|
gpu_buffer_destroy(buffer->gpu);
|
|
free(buffer);
|
|
}
|
|
|
|
const BufferInfo* lovrBufferGetInfo(Buffer* buffer) {
|
|
return &buffer->info;
|
|
}
|
|
|
|
bool lovrBufferIsTemporary(Buffer* buffer) {
|
|
return !!buffer->pointer;
|
|
}
|
|
|
|
void* lovrBufferMap(Buffer* buffer, uint32_t offset, uint32_t size) {
|
|
if (size == ~0u) {
|
|
size = buffer->size - offset;
|
|
}
|
|
|
|
lovrCheck(offset + size <= buffer->size, "Buffer write range [%d,%d] exceeds buffer size", offset, offset + size);
|
|
|
|
if (buffer->pointer) {
|
|
return buffer->pointer + offset;
|
|
}
|
|
|
|
gpu_stream* transfers = getTransfers();
|
|
gpu_buffer* scratchpad = tempAlloc(gpu_sizeof_buffer());
|
|
void* data = gpu_map(scratchpad, size, 4, GPU_MAP_WRITE);
|
|
gpu_copy_buffers(transfers, scratchpad, buffer->gpu, 0, offset, size);
|
|
return data;
|
|
}
|
|
|
|
void lovrBufferClear(Buffer* buffer, uint32_t offset, uint32_t size) {
|
|
lovrCheck(size % 4 == 0, "Buffer clear size must be a multiple of 4");
|
|
lovrCheck(offset % 4 == 0, "Buffer clear offset must be a multiple of 4");
|
|
lovrCheck(offset + size <= buffer->size, "Tried to clear past the end of the Buffer");
|
|
if (buffer->pointer) {
|
|
memset(buffer->pointer + offset, 0, size);
|
|
} else {
|
|
gpu_stream* transfers = getTransfers();
|
|
gpu_clear_buffer(transfers, buffer->gpu, offset, size);
|
|
}
|
|
}
|
|
|
|
// Texture
|
|
|
|
Texture* lovrTextureCreate(TextureInfo* info) {
|
|
uint32_t limits[] = {
|
|
[TEXTURE_2D] = state.limits.textureSize2D,
|
|
[TEXTURE_3D] = state.limits.textureSize3D,
|
|
[TEXTURE_CUBE] = state.limits.textureSizeCube,
|
|
[TEXTURE_ARRAY] = state.limits.textureSize2D
|
|
};
|
|
|
|
uint32_t limit = limits[info->type];
|
|
uint32_t mipmapCap = log2(MAX(MAX(info->width, info->height), (info->type == TEXTURE_3D ? info->depth : 1))) + 1;
|
|
uint32_t mipmaps = CLAMP(info->mipmaps, 1, mipmapCap);
|
|
uint8_t supports = state.features.formats[info->format];
|
|
|
|
lovrCheck(info->width > 0, "Texture width must be greater than zero");
|
|
lovrCheck(info->height > 0, "Texture height must be greater than zero");
|
|
lovrCheck(info->depth > 0, "Texture depth must be greater than zero");
|
|
lovrCheck(info->width <= limit, "Texture %s exceeds the limit for this texture type (%d)", "width", limit);
|
|
lovrCheck(info->height <= limit, "Texture %s exceeds the limit for this texture type (%d)", "height", limit);
|
|
lovrCheck(info->depth <= limit || info->type != TEXTURE_3D, "Texture %s exceeds the limit for this texture type (%d)", "depth", limit);
|
|
lovrCheck(info->depth <= state.limits.textureLayers || info->type != TEXTURE_ARRAY, "Texture %s exceeds the limit for this texture type (%d)", "depth", limit);
|
|
lovrCheck(info->depth == 1 || info->type != TEXTURE_2D, "2D textures must have a depth of 1");
|
|
lovrCheck(info->depth == 6 || info->type != TEXTURE_CUBE, "Cubemaps must have a depth of 6");
|
|
lovrCheck(info->width == info->height || info->type != TEXTURE_CUBE, "Cubemaps must be square");
|
|
lovrCheck(measureTexture(info->format, info->width, info->height, info->depth) < 1 << 30, "Memory for a Texture can not exceed 1GB"); // TODO mip?
|
|
lovrCheck(info->samples == 1 || info->samples == 4, "Currently, Texture multisample count must be 1 or 4");
|
|
lovrCheck(info->samples == 1 || info->type != TEXTURE_CUBE, "Cubemaps can not be multisampled");
|
|
lovrCheck(info->samples == 1 || info->type != TEXTURE_3D, "Volume textures can not be multisampled");
|
|
lovrCheck(info->samples == 1 || ~info->usage & TEXTURE_STORAGE, "Currently, Textures with the 'storage' flag can not be multisampled");
|
|
lovrCheck(info->samples == 1 || mipmaps == 1, "Multisampled textures can only have 1 mipmap");
|
|
lovrCheck(~info->usage & TEXTURE_SAMPLE || (supports & GPU_FEATURE_SAMPLE), "GPU does not support the 'sample' flag for this format");
|
|
lovrCheck(~info->usage & TEXTURE_RENDER || (supports & GPU_FEATURE_RENDER), "GPU does not support the 'render' flag for this format");
|
|
lovrCheck(~info->usage & TEXTURE_STORAGE || (supports & GPU_FEATURE_STORAGE), "GPU does not support the 'storage' flag for this format");
|
|
lovrCheck(~info->usage & TEXTURE_RENDER || info->width <= state.limits.renderSize[0], "Texture has 'render' flag but its size exceeds the renderSize limit");
|
|
lovrCheck(~info->usage & TEXTURE_RENDER || info->height <= state.limits.renderSize[1], "Texture has 'render' flag but its size exceeds the renderSize limit");
|
|
lovrCheck(mipmaps <= mipmapCap, "Texture has more than the max number of mipmap levels for its size (%d)", mipmapCap);
|
|
lovrCheck((info->format < FORMAT_BC1 || info->format > FORMAT_BC7) || state.features.textureBC, "%s textures are not supported on this GPU", "BC");
|
|
lovrCheck(info->format < FORMAT_ASTC_4x4 || state.features.textureASTC, "%s textures are not supported on this GPU", "ASTC");
|
|
|
|
Texture* texture = calloc(1, sizeof(Texture) + gpu_sizeof_texture());
|
|
lovrAssert(texture, "Out of memory");
|
|
texture->ref = 1;
|
|
texture->gpu = (gpu_texture*) (texture + 1);
|
|
texture->info = *info;
|
|
texture->info.mipmaps = mipmaps;
|
|
|
|
uint32_t levelCount = 0;
|
|
uint32_t levelOffsets[16];
|
|
uint32_t levelSizes[16];
|
|
gpu_buffer* scratchpad = NULL;
|
|
|
|
if (info->imageCount > 0) {
|
|
levelCount = lovrImageGetLevelCount(info->images[0]);
|
|
lovrCheck(info->type != TEXTURE_3D || levelCount == 1, "Images used to initialize 3D textures can not have mipmaps");
|
|
|
|
uint32_t total = 0;
|
|
for (uint32_t level = 0; level < levelCount; level++) {
|
|
levelOffsets[level] = total;
|
|
uint32_t width = MAX(info->width >> level, 1);
|
|
uint32_t height = MAX(info->height >> level, 1);
|
|
levelSizes[level] = measureTexture(info->format, width, height, info->depth);
|
|
total += levelSizes[level];
|
|
}
|
|
|
|
scratchpad = tempAlloc(gpu_sizeof_buffer());
|
|
char* data = gpu_map(scratchpad, total, 64, GPU_MAP_WRITE);
|
|
|
|
for (uint32_t level = 0; level < levelCount; level++) {
|
|
for (uint32_t layer = 0; layer < info->depth; layer++) {
|
|
Image* image = info->imageCount == 1 ? info->images[0] : info->images[layer];
|
|
uint32_t slice = info->imageCount == 1 ? layer : 0;
|
|
uint32_t size = lovrImageGetLayerSize(image, level);
|
|
lovrCheck(size == levelSizes[level], "Texture/Image size mismatch!");
|
|
void* pixels = lovrImageGetLayerData(image, level, slice);
|
|
memcpy(data, pixels, size);
|
|
data += size;
|
|
}
|
|
}
|
|
}
|
|
|
|
gpu_texture_init(texture->gpu, &(gpu_texture_info) {
|
|
.type = (gpu_texture_type) info->type,
|
|
.format = (gpu_texture_format) info->format,
|
|
.size = { info->width, info->height, info->depth },
|
|
.mipmaps = texture->info.mipmaps,
|
|
.samples = MAX(info->samples, 1),
|
|
.usage =
|
|
((info->usage & TEXTURE_SAMPLE) ? GPU_TEXTURE_SAMPLE : 0) |
|
|
((info->usage & TEXTURE_RENDER) ? GPU_TEXTURE_RENDER : 0) |
|
|
((info->usage & TEXTURE_STORAGE) ? GPU_TEXTURE_STORAGE : 0) |
|
|
((info->usage & TEXTURE_COPY) ? GPU_TEXTURE_COPY_SRC | GPU_TEXTURE_COPY_DST : 0),
|
|
.srgb = info->srgb,
|
|
.handle = info->handle,
|
|
.label = info->label,
|
|
.upload = {
|
|
.stream = getTransfers(),
|
|
.buffer = scratchpad,
|
|
.levelCount = levelCount,
|
|
.levelOffsets = levelOffsets,
|
|
.generateMipmaps = levelCount < mipmaps
|
|
}
|
|
});
|
|
|
|
// Automatically create a renderable view for renderable non-volume textures
|
|
if ((info->usage & TEXTURE_RENDER) && info->type != TEXTURE_3D && info->depth <= state.limits.renderSize[2]) {
|
|
if (info->mipmaps == 1) {
|
|
texture->renderView = texture->gpu;
|
|
} else {
|
|
gpu_texture_view_info view = {
|
|
.source = texture->gpu,
|
|
.type = GPU_TEXTURE_ARRAY,
|
|
.layerCount = info->depth,
|
|
.levelCount = 1
|
|
};
|
|
|
|
texture->renderView = malloc(gpu_sizeof_texture());
|
|
lovrAssert(texture->renderView, "Out of memory");
|
|
lovrAssert(gpu_texture_init_view(texture->renderView, &view), "Failed to create texture view");
|
|
}
|
|
}
|
|
|
|
return texture;
|
|
}
|
|
|
|
Texture* lovrTextureCreateView(TextureViewInfo* view) {
|
|
const TextureInfo* info = &view->parent->info;
|
|
uint32_t maxDepth = info->type == TEXTURE_3D ? MAX(info->depth >> view->levelIndex, 1) : info->depth;
|
|
lovrCheck(!info->parent, "Can't nest texture views");
|
|
lovrCheck(view->type != TEXTURE_3D, "Texture views may not be volume textures");
|
|
lovrCheck(view->layerCount > 0, "Texture view must have at least one layer");
|
|
lovrCheck(view->levelCount > 0, "Texture view must have at least one mipmap");
|
|
lovrCheck(view->layerIndex + view->layerCount <= maxDepth, "Texture view layer range exceeds depth of parent texture");
|
|
lovrCheck(view->levelIndex + view->levelCount <= info->mipmaps, "Texture view mipmap range exceeds mipmap count of parent texture");
|
|
lovrCheck(view->layerCount == 1 || view->type != TEXTURE_2D, "2D texture can only have a single layer");
|
|
lovrCheck(view->levelCount == 1 || info->type != TEXTURE_3D, "Views of volume textures may only have a single mipmap level");
|
|
lovrCheck(view->layerCount == 6 || view->type != TEXTURE_CUBE, "Cubemaps can only have a six layers");
|
|
|
|
Texture* texture = calloc(1, sizeof(Texture) + gpu_sizeof_texture());
|
|
lovrAssert(texture, "Out of memory");
|
|
texture->ref = 1;
|
|
texture->gpu = (gpu_texture*) (texture + 1);
|
|
texture->info = *info;
|
|
|
|
texture->info.parent = view->parent;
|
|
texture->info.mipmaps = view->levelCount;
|
|
texture->info.width = MAX(info->width >> view->levelIndex, 1);
|
|
texture->info.height = MAX(info->height >> view->levelIndex, 1);
|
|
texture->info.depth = view->layerCount;
|
|
|
|
gpu_texture_init_view(texture->gpu, &(gpu_texture_view_info) {
|
|
.source = view->parent->gpu,
|
|
.type = (gpu_texture_type) view->type,
|
|
.layerIndex = view->layerIndex,
|
|
.layerCount = view->layerCount,
|
|
.levelIndex = view->levelIndex,
|
|
.levelCount = view->levelCount
|
|
});
|
|
|
|
if (view->levelCount == 1 && view->type != TEXTURE_3D && view->layerCount <= 6) {
|
|
texture->renderView = texture->gpu;
|
|
}
|
|
|
|
lovrRetain(view->parent);
|
|
return texture;
|
|
}
|
|
|
|
void lovrTextureDestroy(void* ref) {
|
|
Texture* texture = ref;
|
|
lovrRelease(texture->info.parent, lovrTextureDestroy);
|
|
if (texture->renderView && texture->renderView != texture->gpu) gpu_texture_destroy(texture->renderView);
|
|
if (texture->gpu) gpu_texture_destroy(texture->gpu);
|
|
free(texture);
|
|
}
|
|
|
|
const TextureInfo* lovrTextureGetInfo(Texture* texture) {
|
|
return &texture->info;
|
|
}
|
|
|
|
// Sampler
|
|
|
|
Sampler* lovrSamplerCreate(SamplerInfo* info) {
|
|
lovrCheck(info->range[1] < 0.f || info->range[1] >= info->range[0], "Invalid Sampler mipmap range");
|
|
lovrCheck(info->anisotropy <= state.limits.anisotropy, "Sampler anisotropy (%f) exceeds anisotropy limit (%f)", info->anisotropy, state.limits.anisotropy);
|
|
|
|
Sampler* sampler = calloc(1, sizeof(Sampler) + gpu_sizeof_sampler());
|
|
lovrAssert(sampler, "Out of memory");
|
|
sampler->gpu = (gpu_sampler*) (sampler + 1);
|
|
sampler->info = *info;
|
|
sampler->ref = 1;
|
|
|
|
gpu_sampler_info gpu = {
|
|
.min = (gpu_filter) info->min,
|
|
.mag = (gpu_filter) info->mag,
|
|
.mip = (gpu_filter) info->mip,
|
|
.wrap[0] = (gpu_wrap) info->wrap[0],
|
|
.wrap[1] = (gpu_wrap) info->wrap[1],
|
|
.wrap[2] = (gpu_wrap) info->wrap[2],
|
|
.compare = (gpu_compare_mode) info->compare,
|
|
.anisotropy = MIN(info->anisotropy, state.limits.anisotropy),
|
|
.lodClamp = { info->range[0], info->range[1] }
|
|
};
|
|
|
|
lovrAssert(gpu_sampler_init(sampler->gpu, &gpu), "Failed to initialize sampler");
|
|
return sampler;
|
|
}
|
|
|
|
void lovrSamplerDestroy(void* ref) {
|
|
Sampler* sampler = ref;
|
|
gpu_sampler_destroy(sampler->gpu);
|
|
free(sampler);
|
|
}
|
|
|
|
const SamplerInfo* lovrSamplerGetInfo(Sampler* sampler) {
|
|
return &sampler->info;
|
|
}
|
|
|
|
// Pass
|
|
|
|
Pass* lovrGraphicsGetPass(PassInfo* info) {
|
|
beginFrame();
|
|
Pass* pass = tempAlloc(sizeof(Pass));
|
|
pass->ref = 1;
|
|
pass->info = *info;
|
|
pass->stream = gpu_stream_begin(info->label);
|
|
return pass;
|
|
}
|
|
|
|
void lovrPassDestroy(void* ref) {
|
|
//
|
|
}
|
|
|
|
const PassInfo* lovrPassGetInfo(Pass* pass) {
|
|
return &pass->info;
|
|
}
|
|
|
|
void lovrPassPush(Pass* pass, StackType stack) {
|
|
lovrCheck(pass->info.type == PASS_RENDER, "This function can only be called on a render pass");
|
|
if (stack == STACK_TRANSFORM) {
|
|
pass->transform = pass->transforms[++pass->transformIndex];
|
|
lovrCheck(pass->transformIndex < COUNTOF(pass->transforms), "Transform stack overflow (more pushes than pops?)");
|
|
mat4_init(pass->transforms[pass->transformIndex], pass->transforms[pass->transformIndex - 1]);
|
|
}
|
|
}
|
|
|
|
void lovrPassPop(Pass* pass, StackType stack) {
|
|
lovrCheck(pass->info.type == PASS_RENDER, "This function can only be called on a render pass");
|
|
if (stack == STACK_TRANSFORM) {
|
|
pass->transform = pass->transforms[--pass->transformIndex];
|
|
lovrCheck(pass->transformIndex < COUNTOF(pass->transforms), "Transform stack underflow (more pops than pushes?)");
|
|
}
|
|
}
|
|
|
|
void lovrPassOrigin(Pass* pass) {
|
|
mat4_identity(pass->transform);
|
|
}
|
|
|
|
void lovrPassTranslate(Pass* pass, vec3 translation) {
|
|
mat4_translate(pass->transform, translation[0], translation[1], translation[2]);
|
|
}
|
|
|
|
void lovrPassRotate(Pass* pass, quat rotation) {
|
|
mat4_rotateQuat(pass->transform, rotation);
|
|
}
|
|
|
|
void lovrPassScale(Pass* pass, vec3 scale) {
|
|
mat4_scale(pass->transform, scale[0], scale[1], scale[2]);
|
|
}
|
|
|
|
void lovrPassTransform(Pass* pass, mat4 transform) {
|
|
mat4_mul(pass->transform, transform);
|
|
}
|
|
|
|
// Helpers
|
|
|
|
static void* tempAlloc(size_t size) {
|
|
while (state.allocator.cursor + size > state.allocator.length) {
|
|
lovrAssert(state.allocator.length << 1 <= MAX_FRAME_MEMORY, "Out of memory");
|
|
os_vm_commit(state.allocator.memory + state.allocator.length, state.allocator.length);
|
|
state.allocator.length <<= 1;
|
|
}
|
|
|
|
uint32_t cursor = ALIGN(state.allocator.cursor, 8);
|
|
state.allocator.cursor = cursor + size;
|
|
return state.allocator.memory + cursor;
|
|
}
|
|
|
|
static void beginFrame(void) {
|
|
if (state.active) {
|
|
return;
|
|
}
|
|
|
|
state.active = true;
|
|
state.tick = gpu_begin();
|
|
}
|
|
|
|
static gpu_stream* getTransfers(void) {
|
|
if (!state.transfers) {
|
|
state.transfers = lovrGraphicsGetPass(&(PassInfo) {
|
|
.type = PASS_TRANSFER,
|
|
.label = "Internal Transfers"
|
|
});
|
|
}
|
|
|
|
return state.transfers->stream;
|
|
}
|
|
|
|
// Returns number of bytes of a 3D texture region of a given format
|
|
static size_t measureTexture(TextureFormat format, uint16_t w, uint16_t h, uint16_t d) {
|
|
switch (format) {
|
|
case FORMAT_R8: return w * h * d;
|
|
case FORMAT_RG8:
|
|
case FORMAT_R16:
|
|
case FORMAT_R16F:
|
|
case FORMAT_RGB565:
|
|
case FORMAT_RGB5A1:
|
|
case FORMAT_D16: return w * h * d * 2;
|
|
case FORMAT_RGBA8:
|
|
case FORMAT_RG16:
|
|
case FORMAT_RG16F:
|
|
case FORMAT_R32F:
|
|
case FORMAT_RG11B10F:
|
|
case FORMAT_RGB10A2:
|
|
case FORMAT_D24S8:
|
|
case FORMAT_D32F: return w * h * d * 4;
|
|
case FORMAT_RGBA16:
|
|
case FORMAT_RGBA16F:
|
|
case FORMAT_RG32F: return w * h * d * 8;
|
|
case FORMAT_RGBA32F: return w * h * d * 16;
|
|
case FORMAT_BC1:
|
|
case FORMAT_BC2:
|
|
case FORMAT_BC3:
|
|
case FORMAT_BC4U:
|
|
case FORMAT_BC4S:
|
|
case FORMAT_BC5U:
|
|
case FORMAT_BC5S:
|
|
case FORMAT_BC6UF:
|
|
case FORMAT_BC6SF:
|
|
case FORMAT_BC7:
|
|
case FORMAT_ASTC_4x4: return ((w + 3) / 4) * ((h + 3) / 4) * d * 16;
|
|
case FORMAT_ASTC_5x4: return ((w + 4) / 5) * ((h + 3) / 4) * d * 16;
|
|
case FORMAT_ASTC_5x5: return ((w + 4) / 5) * ((h + 4) / 5) * d * 16;
|
|
case FORMAT_ASTC_6x5: return ((w + 5) / 6) * ((h + 4) / 5) * d * 16;
|
|
case FORMAT_ASTC_6x6: return ((w + 5) / 6) * ((h + 5) / 6) * d * 16;
|
|
case FORMAT_ASTC_8x5: return ((w + 7) / 8) * ((h + 4) / 5) * d * 16;
|
|
case FORMAT_ASTC_8x6: return ((w + 7) / 8) * ((h + 5) / 6) * d * 16;
|
|
case FORMAT_ASTC_8x8: return ((w + 7) / 8) * ((h + 7) / 8) * d * 16;
|
|
case FORMAT_ASTC_10x5: return ((w + 9) / 10) * ((h + 4) / 5) * d * 16;
|
|
case FORMAT_ASTC_10x6: return ((w + 9) / 10) * ((h + 5) / 6) * d * 16;
|
|
case FORMAT_ASTC_10x8: return ((w + 9) / 10) * ((h + 7) / 8) * d * 16;
|
|
case FORMAT_ASTC_10x10: return ((w + 9) / 10) * ((h + 9) / 10) * d * 16;
|
|
case FORMAT_ASTC_12x10: return ((w + 11) / 12) * ((h + 9) / 10) * d * 16;
|
|
case FORMAT_ASTC_12x12: return ((w + 11) / 12) * ((h + 11) / 12) * d * 16;
|
|
default: lovrUnreachable();
|
|
}
|
|
}
|
|
|
|
static void onMessage(void* context, const char* message, bool severe) {
|
|
if (severe) {
|
|
lovrLog(LOG_ERROR, "GPU", message);
|
|
} else {
|
|
lovrLog(LOG_DEBUG, "GPU", message);
|
|
}
|
|
}
|