#include <libsync.h>
#include <stdbool.h>
#include <string.h>
-#include <sys/mman.h>
#include <sys/sysinfo.h>
#include <unistd.h>
-#include <xf86drm.h>
+#include "compiler/glsl_types.h"
#include "util/debug.h"
#include "util/disk_cache.h"
-#include "util/strtod.h"
+#include "util/u_atomic.h"
#include "vk_format.h"
#include "vk_util.h"
-#include "drm/msm_drm.h"
+/* for fd_get_driver/device_uuid() */
+#include "freedreno/common/freedreno_uuid.h"
static int
tu_device_get_cache_uuid(uint16_t family, void *uuid)
return 0;
}
-static void
-tu_get_driver_uuid(void *uuid)
-{
- memset(uuid, 0, VK_UUID_SIZE);
- snprintf(uuid, VK_UUID_SIZE, "freedreno");
-}
-
-static void
-tu_get_device_uuid(void *uuid)
-{
- memset(uuid, 0, VK_UUID_SIZE);
-}
-
-VkResult
-tu_bo_init_new(struct tu_device *dev, struct tu_bo *bo, uint64_t size)
-{
- /* TODO: Choose better flags. As of 2018-11-12, freedreno/drm/msm_bo.c
- * always sets `flags = MSM_BO_WC`, and we copy that behavior here.
- */
- uint32_t gem_handle = tu_gem_new(dev, size, MSM_BO_WC);
- if (!gem_handle)
- goto fail_new;
-
- uint64_t iova = tu_gem_info_iova(dev, gem_handle);
- if (!iova)
- goto fail_info;
-
- *bo = (struct tu_bo) {
- .gem_handle = gem_handle,
- .size = size,
- .iova = iova,
- };
-
- return VK_SUCCESS;
-
-fail_info:
- tu_gem_close(dev, bo->gem_handle);
-fail_new:
- return vk_error(dev->instance, VK_ERROR_OUT_OF_DEVICE_MEMORY);
-}
-
VkResult
-tu_bo_map(struct tu_device *dev, struct tu_bo *bo)
-{
- if (bo->map)
- return VK_SUCCESS;
-
- uint64_t offset = tu_gem_info_offset(dev, bo->gem_handle);
- if (!offset)
- return vk_error(dev->instance, VK_ERROR_OUT_OF_DEVICE_MEMORY);
-
- /* TODO: Should we use the wrapper os_mmap() like Freedreno does? */
- void *map = mmap(0, bo->size, PROT_READ | PROT_WRITE, MAP_SHARED,
- dev->physical_device->local_fd, offset);
- if (map == MAP_FAILED)
- return vk_error(dev->instance, VK_ERROR_MEMORY_MAP_FAILED);
-
- bo->map = map;
- return VK_SUCCESS;
-}
-
-void
-tu_bo_finish(struct tu_device *dev, struct tu_bo *bo)
-{
- assert(bo->gem_handle);
-
- if (bo->map)
- munmap(bo->map, bo->size);
-
- tu_gem_close(dev, bo->gem_handle);
-}
-
-static VkResult
tu_physical_device_init(struct tu_physical_device *device,
- struct tu_instance *instance,
- drmDevicePtr drm_device)
+ struct tu_instance *instance)
{
- const char *path = drm_device->nodes[DRM_NODE_RENDER];
VkResult result = VK_SUCCESS;
- drmVersionPtr version;
- int fd;
- int master_fd = -1;
-
- fd = open(path, O_RDWR | O_CLOEXEC);
- if (fd < 0) {
- return vk_errorf(instance, VK_ERROR_INCOMPATIBLE_DRIVER,
- "failed to open device %s", path);
- }
-
- /* Version 1.3 added MSM_INFO_IOVA. */
- const int min_version_major = 1;
- const int min_version_minor = 3;
-
- version = drmGetVersion(fd);
- if (!version) {
- close(fd);
- return vk_errorf(instance, VK_ERROR_INCOMPATIBLE_DRIVER,
- "failed to query kernel driver version for device %s",
- path);
- }
-
- if (strcmp(version->name, "msm")) {
- drmFreeVersion(version);
- if (master_fd != -1)
- close(master_fd);
- close(fd);
- return vk_errorf(instance, VK_ERROR_INCOMPATIBLE_DRIVER,
- "device %s does not use the msm kernel driver", path);
- }
-
- if (version->version_major != min_version_major ||
- version->version_minor < min_version_minor) {
- result = vk_errorf(instance, VK_ERROR_INCOMPATIBLE_DRIVER,
- "kernel driver for device %s has version %d.%d, "
- "but Vulkan requires version >= %d.%d",
- path, version->version_major, version->version_minor,
- min_version_major, min_version_minor);
- drmFreeVersion(version);
- close(fd);
- return result;
- }
-
- drmFreeVersion(version);
-
- if (instance->debug_flags & TU_DEBUG_STARTUP)
- tu_logi("Found compatible device '%s'.", path);
-
- device->_loader_data.loaderMagic = ICD_LOADER_MAGIC;
- device->instance = instance;
- assert(strlen(path) < ARRAY_SIZE(device->path));
- strncpy(device->path, path, ARRAY_SIZE(device->path));
-
- if (instance->enabled_extensions.KHR_display) {
- master_fd =
- open(drm_device->nodes[DRM_NODE_PRIMARY], O_RDWR | O_CLOEXEC);
- if (master_fd >= 0) {
- /* TODO: free master_fd is accel is not working? */
- }
- }
-
- device->master_fd = master_fd;
- device->local_fd = fd;
-
- if (tu_drm_get_gpu_id(device, &device->gpu_id)) {
- if (instance->debug_flags & TU_DEBUG_STARTUP)
- tu_logi("Could not query the GPU ID");
- result = vk_errorf(instance, VK_ERROR_INITIALIZATION_FAILED,
- "could not get GPU ID");
- goto fail;
- }
-
- if (tu_drm_get_gmem_size(device, &device->gmem_size)) {
- if (instance->debug_flags & TU_DEBUG_STARTUP)
- tu_logi("Could not query the GMEM size");
- result = vk_errorf(instance, VK_ERROR_INITIALIZATION_FAILED,
- "could not get GMEM size");
- goto fail;
- }
memset(device->name, 0, sizeof(device->name));
sprintf(device->name, "FD%d", device->gpu_id);
+ device->limited_z24s8 = (device->gpu_id == 630);
+
switch (device->gpu_id) {
- case 530:
+ case 618:
+ device->ccu_offset_gmem = 0x7c000; /* 0x7e000 in some cases? */
+ device->ccu_offset_bypass = 0x10000;
+ device->tile_align_w = 32;
+ device->magic.PC_UNKNOWN_9805 = 0x0;
+ device->magic.SP_UNKNOWN_A0F8 = 0x0;
+ device->supports_multiview_mask = false; /* TODO */
+ break;
case 630:
+ case 640:
+ device->ccu_offset_gmem = 0xf8000;
+ device->ccu_offset_bypass = 0x20000;
+ device->tile_align_w = 32;
+ device->magic.PC_UNKNOWN_9805 = 0x1;
+ device->magic.SP_UNKNOWN_A0F8 = 0x1;
+ device->supports_multiview_mask = device->gpu_id != 630;
+ break;
+ case 650:
+ device->ccu_offset_gmem = 0x114000;
+ device->ccu_offset_bypass = 0x30000;
+ device->tile_align_w = 96;
+ device->magic.PC_UNKNOWN_9805 = 0x2;
+ device->magic.SP_UNKNOWN_A0F8 = 0x2;
+ device->supports_multiview_mask = true;
break;
default:
result = vk_errorf(instance, VK_ERROR_INITIALIZATION_FAILED,
fprintf(stderr, "WARNING: tu is not a conformant vulkan implementation, "
"testing use only.\n");
- tu_get_driver_uuid(&device->device_uuid);
- tu_get_device_uuid(&device->device_uuid);
+ fd_get_driver_uuid(device->driver_uuid);
+ fd_get_device_uuid(device->device_uuid, device->gpu_id);
- tu_fill_device_extension_table(device, &device->supported_extensions);
+ tu_physical_device_get_supported_extensions(device, &device->supported_extensions);
if (result != VK_SUCCESS) {
vk_error(instance, result);
goto fail;
}
+ result = tu_wsi_init(device);
+ if (result != VK_SUCCESS) {
+ vk_error(instance, result);
+ goto fail;
+ }
+
return VK_SUCCESS;
fail:
- close(fd);
- if (master_fd != -1)
- close(master_fd);
+ close(device->local_fd);
+ if (device->master_fd != -1)
+ close(device->master_fd);
return result;
}
static void
tu_physical_device_finish(struct tu_physical_device *device)
{
+ tu_wsi_finish(device);
+
disk_cache_destroy(device->disk_cache);
close(device->local_fd);
if (device->master_fd != -1)
close(device->master_fd);
+
+ vk_object_base_finish(&device->base);
}
-static void *
+static VKAPI_ATTR void *
default_alloc_func(void *pUserData,
size_t size,
size_t align,
return malloc(size);
}
-static void *
+static VKAPI_ATTR void *
default_realloc_func(void *pUserData,
void *pOriginal,
size_t size,
return realloc(pOriginal, size);
}
-static void
+static VKAPI_ATTR void
default_free_func(void *pUserData, void *pMemory)
{
free(pMemory);
};
static const struct debug_control tu_debug_options[] = {
- { "startup", TU_DEBUG_STARTUP }, { NULL, 0 }
+ { "startup", TU_DEBUG_STARTUP },
+ { "nir", TU_DEBUG_NIR },
+ { "ir3", TU_DEBUG_IR3 },
+ { "nobin", TU_DEBUG_NOBIN },
+ { "sysmem", TU_DEBUG_SYSMEM },
+ { "forcebin", TU_DEBUG_FORCEBIN },
+ { "noubwc", TU_DEBUG_NOUBWC },
+ { NULL, 0 }
};
const char *
instance = vk_zalloc2(&default_alloc, pAllocator, sizeof(*instance), 8,
VK_SYSTEM_ALLOCATION_SCOPE_INSTANCE);
+
if (!instance)
return vk_error(NULL, VK_ERROR_OUT_OF_HOST_MEMORY);
- instance->_loader_data.loaderMagic = ICD_LOADER_MAGIC;
+ vk_object_base_init(NULL, &instance->base, VK_OBJECT_TYPE_INSTANCE);
if (pAllocator)
instance->alloc = *pAllocator;
const char *ext_name = pCreateInfo->ppEnabledExtensionNames[i];
int index = tu_get_instance_extension_index(ext_name);
- if (index < 0 || !tu_supported_instance_extensions.extensions[index]) {
+ if (index < 0 || !tu_instance_extensions_supported.extensions[index]) {
+ vk_object_base_finish(&instance->base);
vk_free2(&default_alloc, pAllocator, instance);
return vk_error(instance, VK_ERROR_EXTENSION_NOT_PRESENT);
}
result = vk_debug_report_instance_init(&instance->debug_report_callbacks);
if (result != VK_SUCCESS) {
+ vk_object_base_finish(&instance->base);
vk_free2(&default_alloc, pAllocator, instance);
return vk_error(instance, result);
}
- _mesa_locale_init();
+ glsl_type_singleton_init_or_ref();
VG(VALGRIND_CREATE_MEMPOOL(instance, 0, false));
VG(VALGRIND_DESTROY_MEMPOOL(instance));
- _mesa_locale_fini();
+ glsl_type_singleton_decref();
vk_debug_report_instance_destroy(&instance->debug_report_callbacks);
+ vk_object_base_finish(&instance->base);
vk_free(&instance->alloc, instance);
}
-static VkResult
-tu_enumerate_devices(struct tu_instance *instance)
-{
- /* TODO: Check for more devices ? */
- drmDevicePtr devices[8];
- VkResult result = VK_ERROR_INCOMPATIBLE_DRIVER;
- int max_devices;
-
- instance->physical_device_count = 0;
-
- max_devices = drmGetDevices2(0, devices, ARRAY_SIZE(devices));
-
- if (instance->debug_flags & TU_DEBUG_STARTUP)
- tu_logi("Found %d drm nodes", max_devices);
-
- if (max_devices < 1)
- return vk_error(instance, VK_ERROR_INCOMPATIBLE_DRIVER);
-
- for (unsigned i = 0; i < (unsigned) max_devices; i++) {
- if (devices[i]->available_nodes & 1 << DRM_NODE_RENDER &&
- devices[i]->bustype == DRM_BUS_PLATFORM) {
-
- result = tu_physical_device_init(
- instance->physical_devices + instance->physical_device_count,
- instance, devices[i]);
- if (result == VK_SUCCESS)
- ++instance->physical_device_count;
- else if (result != VK_ERROR_INCOMPATIBLE_DRIVER)
- break;
- }
- }
- drmFreeDevices(devices, max_devices);
-
- return result;
-}
-
VkResult
tu_EnumeratePhysicalDevices(VkInstance _instance,
uint32_t *pPhysicalDeviceCount,
memset(pFeatures, 0, sizeof(*pFeatures));
*pFeatures = (VkPhysicalDeviceFeatures) {
- .robustBufferAccess = false,
- .fullDrawIndexUint32 = false,
- .imageCubeArray = false,
- .independentBlend = false,
- .geometryShader = false,
- .tessellationShader = false,
- .sampleRateShading = false,
- .dualSrcBlend = false,
- .logicOp = false,
- .multiDrawIndirect = false,
- .drawIndirectFirstInstance = false,
- .depthClamp = false,
- .depthBiasClamp = false,
- .fillModeNonSolid = false,
- .depthBounds = false,
+ .robustBufferAccess = true,
+ .fullDrawIndexUint32 = true,
+ .imageCubeArray = true,
+ .independentBlend = true,
+ .geometryShader = true,
+ .tessellationShader = true,
+ .sampleRateShading = true,
+ .dualSrcBlend = true,
+ .logicOp = true,
+ .multiDrawIndirect = true,
+ .drawIndirectFirstInstance = true,
+ .depthClamp = true,
+ .depthBiasClamp = true,
+ .fillModeNonSolid = true,
+ .depthBounds = true,
.wideLines = false,
- .largePoints = false,
- .alphaToOne = false,
+ .largePoints = true,
+ .alphaToOne = true,
.multiViewport = false,
- .samplerAnisotropy = false,
- .textureCompressionETC2 = false,
- .textureCompressionASTC_LDR = false,
- .textureCompressionBC = false,
- .occlusionQueryPrecise = false,
+ .samplerAnisotropy = true,
+ .textureCompressionETC2 = true,
+ .textureCompressionASTC_LDR = true,
+ .textureCompressionBC = true,
+ .occlusionQueryPrecise = true,
.pipelineStatisticsQuery = false,
- .vertexPipelineStoresAndAtomics = false,
- .fragmentStoresAndAtomics = false,
+ .vertexPipelineStoresAndAtomics = true,
+ .fragmentStoresAndAtomics = true,
.shaderTessellationAndGeometryPointSize = false,
.shaderImageGatherExtended = false,
.shaderStorageImageExtendedFormats = false,
.shaderStorageImageMultisample = false,
- .shaderUniformBufferArrayDynamicIndexing = false,
- .shaderSampledImageArrayDynamicIndexing = false,
- .shaderStorageBufferArrayDynamicIndexing = false,
- .shaderStorageImageArrayDynamicIndexing = false,
+ .shaderUniformBufferArrayDynamicIndexing = true,
+ .shaderSampledImageArrayDynamicIndexing = true,
+ .shaderStorageBufferArrayDynamicIndexing = true,
+ .shaderStorageImageArrayDynamicIndexing = true,
.shaderStorageImageReadWithoutFormat = false,
.shaderStorageImageWriteWithoutFormat = false,
.shaderClipDistance = false,
void
tu_GetPhysicalDeviceFeatures2(VkPhysicalDevice physicalDevice,
- VkPhysicalDeviceFeatures2KHR *pFeatures)
+ VkPhysicalDeviceFeatures2 *pFeatures)
{
vk_foreach_struct(ext, pFeatures->pNext)
{
switch (ext->sType) {
- case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VARIABLE_POINTER_FEATURES_KHR: {
- VkPhysicalDeviceVariablePointerFeaturesKHR *features = (void *) ext;
- features->variablePointersStorageBuffer = false;
- features->variablePointers = false;
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_FEATURES: {
+ VkPhysicalDeviceVulkan11Features *features = (void *) ext;
+ features->storageBuffer16BitAccess = false;
+ features->uniformAndStorageBuffer16BitAccess = false;
+ features->storagePushConstant16 = false;
+ features->storageInputOutput16 = false;
+ features->multiview = false;
+ features->multiviewGeometryShader = false;
+ features->multiviewTessellationShader = false;
+ features->variablePointersStorageBuffer = true;
+ features->variablePointers = true;
+ features->protectedMemory = false;
+ features->samplerYcbcrConversion = true;
+ features->shaderDrawParameters = true;
+ break;
+ }
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_2_FEATURES: {
+ VkPhysicalDeviceVulkan12Features *features = (void *) ext;
+ features->samplerMirrorClampToEdge = true;
+ features->drawIndirectCount = true;
+ features->storageBuffer8BitAccess = false;
+ features->uniformAndStorageBuffer8BitAccess = false;
+ features->storagePushConstant8 = false;
+ features->shaderBufferInt64Atomics = false;
+ features->shaderSharedInt64Atomics = false;
+ features->shaderFloat16 = false;
+ features->shaderInt8 = false;
+
+ features->descriptorIndexing = false;
+ features->shaderInputAttachmentArrayDynamicIndexing = false;
+ features->shaderUniformTexelBufferArrayDynamicIndexing = false;
+ features->shaderStorageTexelBufferArrayDynamicIndexing = false;
+ features->shaderUniformBufferArrayNonUniformIndexing = false;
+ features->shaderSampledImageArrayNonUniformIndexing = false;
+ features->shaderStorageBufferArrayNonUniformIndexing = false;
+ features->shaderStorageImageArrayNonUniformIndexing = false;
+ features->shaderInputAttachmentArrayNonUniformIndexing = false;
+ features->shaderUniformTexelBufferArrayNonUniformIndexing = false;
+ features->shaderStorageTexelBufferArrayNonUniformIndexing = false;
+ features->descriptorBindingUniformBufferUpdateAfterBind = false;
+ features->descriptorBindingSampledImageUpdateAfterBind = false;
+ features->descriptorBindingStorageImageUpdateAfterBind = false;
+ features->descriptorBindingStorageBufferUpdateAfterBind = false;
+ features->descriptorBindingUniformTexelBufferUpdateAfterBind = false;
+ features->descriptorBindingStorageTexelBufferUpdateAfterBind = false;
+ features->descriptorBindingUpdateUnusedWhilePending = false;
+ features->descriptorBindingPartiallyBound = false;
+ features->descriptorBindingVariableDescriptorCount = false;
+ features->runtimeDescriptorArray = false;
+
+ features->samplerFilterMinmax = true;
+ features->scalarBlockLayout = false;
+ features->imagelessFramebuffer = false;
+ features->uniformBufferStandardLayout = false;
+ features->shaderSubgroupExtendedTypes = false;
+ features->separateDepthStencilLayouts = false;
+ features->hostQueryReset = false;
+ features->timelineSemaphore = false;
+ features->bufferDeviceAddress = false;
+ features->bufferDeviceAddressCaptureReplay = false;
+ features->bufferDeviceAddressMultiDevice = false;
+ features->vulkanMemoryModel = false;
+ features->vulkanMemoryModelDeviceScope = false;
+ features->vulkanMemoryModelAvailabilityVisibilityChains = false;
+ features->shaderOutputViewportIndex = false;
+ features->shaderOutputLayer = false;
+ features->subgroupBroadcastDynamicId = false;
+ break;
+ }
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VARIABLE_POINTERS_FEATURES: {
+ VkPhysicalDeviceVariablePointersFeatures *features = (void *) ext;
+ features->variablePointersStorageBuffer = true;
+ features->variablePointers = true;
break;
}
- case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MULTIVIEW_FEATURES_KHR: {
- VkPhysicalDeviceMultiviewFeaturesKHR *features =
- (VkPhysicalDeviceMultiviewFeaturesKHR *) ext;
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MULTIVIEW_FEATURES: {
+ VkPhysicalDeviceMultiviewFeatures *features =
+ (VkPhysicalDeviceMultiviewFeatures *) ext;
features->multiview = false;
features->multiviewGeometryShader = false;
features->multiviewTessellationShader = false;
break;
}
- case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_DRAW_PARAMETER_FEATURES: {
- VkPhysicalDeviceShaderDrawParameterFeatures *features =
- (VkPhysicalDeviceShaderDrawParameterFeatures *) ext;
- features->shaderDrawParameters = false;
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_DRAW_PARAMETERS_FEATURES: {
+ VkPhysicalDeviceShaderDrawParametersFeatures *features =
+ (VkPhysicalDeviceShaderDrawParametersFeatures *) ext;
+ features->shaderDrawParameters = true;
break;
}
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROTECTED_MEMORY_FEATURES: {
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLER_YCBCR_CONVERSION_FEATURES: {
VkPhysicalDeviceSamplerYcbcrConversionFeatures *features =
(VkPhysicalDeviceSamplerYcbcrConversionFeatures *) ext;
- features->samplerYcbcrConversion = false;
+ features->samplerYcbcrConversion = true;
break;
}
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DESCRIPTOR_INDEXING_FEATURES_EXT: {
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CONDITIONAL_RENDERING_FEATURES_EXT: {
VkPhysicalDeviceConditionalRenderingFeaturesEXT *features =
(VkPhysicalDeviceConditionalRenderingFeaturesEXT *) ext;
- features->conditionalRendering = false;
- features->inheritedConditionalRendering = false;
+ features->conditionalRendering = true;
+ features->inheritedConditionalRendering = true;
+ break;
+ }
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_FEATURES_EXT: {
+ VkPhysicalDeviceTransformFeedbackFeaturesEXT *features =
+ (VkPhysicalDeviceTransformFeedbackFeaturesEXT *) ext;
+ features->transformFeedback = true;
+ features->geometryStreams = false;
+ break;
+ }
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_INDEX_TYPE_UINT8_FEATURES_EXT: {
+ VkPhysicalDeviceIndexTypeUint8FeaturesEXT *features =
+ (VkPhysicalDeviceIndexTypeUint8FeaturesEXT *)ext;
+ features->indexTypeUint8 = true;
+ break;
+ }
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VERTEX_ATTRIBUTE_DIVISOR_FEATURES_EXT: {
+ VkPhysicalDeviceVertexAttributeDivisorFeaturesEXT *features =
+ (VkPhysicalDeviceVertexAttributeDivisorFeaturesEXT *)ext;
+ features->vertexAttributeInstanceRateDivisor = true;
+ features->vertexAttributeInstanceRateZeroDivisor = true;
+ break;
+ }
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PRIVATE_DATA_FEATURES_EXT: {
+ VkPhysicalDevicePrivateDataFeaturesEXT *features =
+ (VkPhysicalDevicePrivateDataFeaturesEXT *)ext;
+ features->privateData = true;
+ break;
+ }
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DEPTH_CLIP_ENABLE_FEATURES_EXT: {
+ VkPhysicalDeviceDepthClipEnableFeaturesEXT *features =
+ (VkPhysicalDeviceDepthClipEnableFeaturesEXT *)ext;
+ features->depthClipEnable = true;
+ break;
+ }
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_4444_FORMATS_FEATURES_EXT: {
+ VkPhysicalDevice4444FormatsFeaturesEXT *features = (void *)ext;
+ features->formatA4R4G4B4 = true;
+ features->formatA4B4G4R4 = true;
+ break;
+ }
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CUSTOM_BORDER_COLOR_FEATURES_EXT: {
+ VkPhysicalDeviceCustomBorderColorFeaturesEXT *features = (void *) ext;
+ features->customBorderColors = true;
+ features->customBorderColorWithoutFormat = true;
break;
}
default:
VkPhysicalDeviceProperties *pProperties)
{
TU_FROM_HANDLE(tu_physical_device, pdevice, physicalDevice);
- VkSampleCountFlags sample_counts = 0xf;
-
- /* make sure that the entire descriptor set is addressable with a signed
- * 32-bit int. So the sum of all limits scaled by descriptor size has to
- * be at most 2 GiB. the combined image & samples object count as one of
- * both. This limit is for the pipeline layout, not for the set layout, but
- * there is no set limit, so we just set a pipeline limit. I don't think
- * any app is going to hit this soon. */
- size_t max_descriptor_set_size =
- ((1ull << 31) - 16 * MAX_DYNAMIC_BUFFERS) /
- (32 /* uniform buffer, 32 due to potential space wasted on alignment */ +
- 32 /* storage buffer, 32 due to potential space wasted on alignment */ +
- 32 /* sampler, largest when combined with image */ +
- 64 /* sampled image */ + 64 /* storage image */);
+ VkSampleCountFlags sample_counts =
+ VK_SAMPLE_COUNT_1_BIT | VK_SAMPLE_COUNT_2_BIT | VK_SAMPLE_COUNT_4_BIT;
+
+ /* I have no idea what the maximum size is, but the hardware supports very
+ * large numbers of descriptors (at least 2^16). This limit is based on
+ * CP_LOAD_STATE6, which has a 28-bit field for the DWORD offset, so that
+ * we don't have to think about what to do if that overflows, but really
+ * nothing is likely to get close to this.
+ */
+ const size_t max_descriptor_set_size = (1 << 28) / A6XX_TEX_CONST_DWORDS;
VkPhysicalDeviceLimits limits = {
.maxImageDimension1D = (1 << 14),
.maxImageDimensionCube = (1 << 14),
.maxImageArrayLayers = (1 << 11),
.maxTexelBufferElements = 128 * 1024 * 1024,
- .maxUniformBufferRange = UINT32_MAX,
- .maxStorageBufferRange = UINT32_MAX,
+ .maxUniformBufferRange = MAX_UNIFORM_BUFFER_RANGE,
+ .maxStorageBufferRange = MAX_STORAGE_BUFFER_RANGE,
.maxPushConstantsSize = MAX_PUSH_CONSTANTS_SIZE,
.maxMemoryAllocationCount = UINT32_MAX,
.maxSamplerAllocationCount = 64 * 1024,
.maxPerStageDescriptorStorageBuffers = max_descriptor_set_size,
.maxPerStageDescriptorSampledImages = max_descriptor_set_size,
.maxPerStageDescriptorStorageImages = max_descriptor_set_size,
- .maxPerStageDescriptorInputAttachments = max_descriptor_set_size,
+ .maxPerStageDescriptorInputAttachments = MAX_RTS,
.maxPerStageResources = max_descriptor_set_size,
.maxDescriptorSetSamplers = max_descriptor_set_size,
.maxDescriptorSetUniformBuffers = max_descriptor_set_size,
.maxDescriptorSetStorageBuffersDynamic = MAX_DYNAMIC_STORAGE_BUFFERS,
.maxDescriptorSetSampledImages = max_descriptor_set_size,
.maxDescriptorSetStorageImages = max_descriptor_set_size,
- .maxDescriptorSetInputAttachments = max_descriptor_set_size,
+ .maxDescriptorSetInputAttachments = MAX_RTS,
.maxVertexInputAttributes = 32,
.maxVertexInputBindings = 32,
- .maxVertexInputAttributeOffset = 2047,
+ .maxVertexInputAttributeOffset = 4095,
.maxVertexInputBindingStride = 2048,
.maxVertexOutputComponents = 128,
.maxTessellationGenerationLevel = 64,
.maxTessellationControlTotalOutputComponents = 4096,
.maxTessellationEvaluationInputComponents = 128,
.maxTessellationEvaluationOutputComponents = 128,
- .maxGeometryShaderInvocations = 127,
+ .maxGeometryShaderInvocations = 32,
.maxGeometryInputComponents = 64,
.maxGeometryOutputComponents = 128,
.maxGeometryOutputVertices = 256,
.maxGeometryTotalOutputComponents = 1024,
- .maxFragmentInputComponents = 128,
+ .maxFragmentInputComponents = 124,
.maxFragmentOutputAttachments = 8,
.maxFragmentDualSrcAttachments = 1,
.maxFragmentCombinedOutputResources = 8,
.maxComputeWorkGroupCount = { 65535, 65535, 65535 },
.maxComputeWorkGroupInvocations = 2048,
.maxComputeWorkGroupSize = { 2048, 2048, 2048 },
- .subPixelPrecisionBits = 4 /* FIXME */,
- .subTexelPrecisionBits = 4 /* FIXME */,
- .mipmapPrecisionBits = 4 /* FIXME */,
+ .subPixelPrecisionBits = 8,
+ .subTexelPrecisionBits = 8,
+ .mipmapPrecisionBits = 8,
.maxDrawIndexedIndexValue = UINT32_MAX,
.maxDrawIndirectCount = UINT32_MAX,
- .maxSamplerLodBias = 16,
+ .maxSamplerLodBias = 4095.0 / 256.0, /* [-16, 15.99609375] */
.maxSamplerAnisotropy = 16,
.maxViewports = MAX_VIEWPORTS,
.maxViewportDimensions = { (1 << 14), (1 << 14) },
.viewportBoundsRange = { INT16_MIN, INT16_MAX },
.viewportSubPixelBits = 8,
.minMemoryMapAlignment = 4096, /* A page */
- .minTexelBufferOffsetAlignment = 1,
- .minUniformBufferOffsetAlignment = 4,
- .minStorageBufferOffsetAlignment = 4,
- .minTexelOffset = -32,
- .maxTexelOffset = 31,
+ .minTexelBufferOffsetAlignment = 64,
+ .minUniformBufferOffsetAlignment = 64,
+ .minStorageBufferOffsetAlignment = 64,
+ .minTexelOffset = -16,
+ .maxTexelOffset = 15,
.minTexelGatherOffset = -32,
.maxTexelGatherOffset = 31,
- .minInterpolationOffset = -2,
- .maxInterpolationOffset = 2,
- .subPixelInterpolationOffsetBits = 8,
+ .minInterpolationOffset = -0.5,
+ .maxInterpolationOffset = 0.4375,
+ .subPixelInterpolationOffsetBits = 4,
.maxFramebufferWidth = (1 << 14),
.maxFramebufferHeight = (1 << 14),
.maxFramebufferLayers = (1 << 10),
.storageImageSampleCounts = VK_SAMPLE_COUNT_1_BIT,
.maxSampleMaskWords = 1,
.timestampComputeAndGraphics = true,
- .timestampPeriod = 1,
+ .timestampPeriod = 1000000000.0 / 19200000.0, /* CP_ALWAYS_ON_COUNTER is fixed 19.2MHz */
.maxClipDistances = 8,
.maxCullDistances = 8,
.maxCombinedClipAndCullDistances = 8,
.discreteQueuePriorities = 1,
- .pointSizeRange = { 0.125, 255.875 },
+ .pointSizeRange = { 1, 4092 },
.lineWidthRange = { 0.0, 7.9921875 },
- .pointSizeGranularity = (1.0 / 8.0),
+ .pointSizeGranularity = 0.0625,
.lineWidthGranularity = (1.0 / 128.0),
.strictLines = false, /* FINISHME */
.standardSampleLocations = true,
void
tu_GetPhysicalDeviceProperties2(VkPhysicalDevice physicalDevice,
- VkPhysicalDeviceProperties2KHR *pProperties)
+ VkPhysicalDeviceProperties2 *pProperties)
{
TU_FROM_HANDLE(tu_physical_device, pdevice, physicalDevice);
tu_GetPhysicalDeviceProperties(physicalDevice, &pProperties->properties);
properties->maxPushDescriptors = MAX_PUSH_DESCRIPTORS;
break;
}
- case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ID_PROPERTIES_KHR: {
- VkPhysicalDeviceIDPropertiesKHR *properties =
- (VkPhysicalDeviceIDPropertiesKHR *) ext;
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ID_PROPERTIES: {
+ VkPhysicalDeviceIDProperties *properties =
+ (VkPhysicalDeviceIDProperties *) ext;
memcpy(properties->driverUUID, pdevice->driver_uuid, VK_UUID_SIZE);
memcpy(properties->deviceUUID, pdevice->device_uuid, VK_UUID_SIZE);
properties->deviceLUIDValid = false;
break;
}
- case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MULTIVIEW_PROPERTIES_KHR: {
- VkPhysicalDeviceMultiviewPropertiesKHR *properties =
- (VkPhysicalDeviceMultiviewPropertiesKHR *) ext;
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MULTIVIEW_PROPERTIES: {
+ VkPhysicalDeviceMultiviewProperties *properties =
+ (VkPhysicalDeviceMultiviewProperties *) ext;
properties->maxMultiviewViewCount = MAX_VIEWS;
properties->maxMultiviewInstanceIndex = INT_MAX;
break;
}
- case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_POINT_CLIPPING_PROPERTIES_KHR: {
- VkPhysicalDevicePointClippingPropertiesKHR *properties =
- (VkPhysicalDevicePointClippingPropertiesKHR *) ext;
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_POINT_CLIPPING_PROPERTIES: {
+ VkPhysicalDevicePointClippingProperties *properties =
+ (VkPhysicalDevicePointClippingProperties *) ext;
properties->pointClippingBehavior =
- VK_POINT_CLIPPING_BEHAVIOR_ALL_CLIP_PLANES_KHR;
+ VK_POINT_CLIPPING_BEHAVIOR_ALL_CLIP_PLANES;
break;
}
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MAINTENANCE_3_PROPERTIES: {
properties->maxMemoryAllocationSize = 0xFFFFFFFFull;
break;
}
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_PROPERTIES_EXT: {
+ VkPhysicalDeviceTransformFeedbackPropertiesEXT *properties =
+ (VkPhysicalDeviceTransformFeedbackPropertiesEXT *)ext;
+
+ properties->maxTransformFeedbackStreams = IR3_MAX_SO_STREAMS;
+ properties->maxTransformFeedbackBuffers = IR3_MAX_SO_BUFFERS;
+ properties->maxTransformFeedbackBufferSize = UINT32_MAX;
+ properties->maxTransformFeedbackStreamDataSize = 512;
+ properties->maxTransformFeedbackBufferDataSize = 512;
+ properties->maxTransformFeedbackBufferDataStride = 512;
+ properties->transformFeedbackQueries = true;
+ properties->transformFeedbackStreamsLinesTriangles = false;
+ properties->transformFeedbackRasterizationStreamSelect = false;
+ properties->transformFeedbackDraw = true;
+ break;
+ }
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLE_LOCATIONS_PROPERTIES_EXT: {
+ VkPhysicalDeviceSampleLocationsPropertiesEXT *properties =
+ (VkPhysicalDeviceSampleLocationsPropertiesEXT *)ext;
+ properties->sampleLocationSampleCounts = 0;
+ if (pdevice->supported_extensions.EXT_sample_locations) {
+ properties->sampleLocationSampleCounts =
+ VK_SAMPLE_COUNT_1_BIT | VK_SAMPLE_COUNT_2_BIT | VK_SAMPLE_COUNT_4_BIT;
+ }
+ properties->maxSampleLocationGridSize = (VkExtent2D) { 1 , 1 };
+ properties->sampleLocationCoordinateRange[0] = 0.0f;
+ properties->sampleLocationCoordinateRange[1] = 0.9375f;
+ properties->sampleLocationSubPixelBits = 4;
+ properties->variableSampleLocations = true;
+ break;
+ }
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLER_FILTER_MINMAX_PROPERTIES: {
+ VkPhysicalDeviceSamplerFilterMinmaxProperties *properties =
+ (VkPhysicalDeviceSamplerFilterMinmaxProperties *)ext;
+ properties->filterMinmaxImageComponentMapping = true;
+ properties->filterMinmaxSingleComponentFormats = true;
+ break;
+ }
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_PROPERTIES: {
+ VkPhysicalDeviceSubgroupProperties *properties =
+ (VkPhysicalDeviceSubgroupProperties *)ext;
+ properties->subgroupSize = 64;
+ properties->supportedStages = VK_SHADER_STAGE_COMPUTE_BIT;
+ properties->supportedOperations = VK_SUBGROUP_FEATURE_BASIC_BIT |
+ VK_SUBGROUP_FEATURE_VOTE_BIT;
+ properties->quadOperationsInAllStages = false;
+ break;
+ }
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VERTEX_ATTRIBUTE_DIVISOR_PROPERTIES_EXT: {
+ VkPhysicalDeviceVertexAttributeDivisorPropertiesEXT *props =
+ (VkPhysicalDeviceVertexAttributeDivisorPropertiesEXT *)ext;
+ props->maxVertexAttribDivisor = UINT32_MAX;
+ break;
+ }
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CUSTOM_BORDER_COLOR_PROPERTIES_EXT: {
+ VkPhysicalDeviceCustomBorderColorPropertiesEXT *props = (void *)ext;
+ props->maxCustomBorderColorSamplers = TU_BORDER_COLOR_COUNT;
+ break;
+ }
default:
break;
}
.queueFlags =
VK_QUEUE_GRAPHICS_BIT | VK_QUEUE_COMPUTE_BIT | VK_QUEUE_TRANSFER_BIT,
.queueCount = 1,
- .timestampValidBits = 64,
- .minImageTransferGranularity = (VkExtent3D) { 1, 1, 1 },
+ .timestampValidBits = 48,
+ .minImageTransferGranularity = { 1, 1, 1 },
};
void
tu_GetPhysicalDeviceQueueFamilyProperties2(
VkPhysicalDevice physicalDevice,
uint32_t *pQueueFamilyPropertyCount,
- VkQueueFamilyProperties2KHR *pQueueFamilyProperties)
+ VkQueueFamilyProperties2 *pQueueFamilyProperties)
{
VK_OUTARRAY_MAKE(out, pQueueFamilyProperties, pQueueFamilyPropertyCount);
void
tu_GetPhysicalDeviceMemoryProperties2(
VkPhysicalDevice physicalDevice,
- VkPhysicalDeviceMemoryProperties2KHR *pMemoryProperties)
+ VkPhysicalDeviceMemoryProperties2 *pMemoryProperties)
{
return tu_GetPhysicalDeviceMemoryProperties(
physicalDevice, &pMemoryProperties->memoryProperties);
int idx,
VkDeviceQueueCreateFlags flags)
{
- queue->_loader_data.loaderMagic = ICD_LOADER_MAGIC;
+ vk_object_base_init(&device->vk, &queue->base, VK_OBJECT_TYPE_QUEUE);
+
queue->device = device;
queue->queue_family_index = queue_family_index;
queue->queue_idx = idx;
if (ret)
return VK_ERROR_INITIALIZATION_FAILED;
- queue->submit_fence_fd = -1;
+ tu_fence_init(&queue->submit_fence, false);
return VK_SUCCESS;
}
static void
tu_queue_finish(struct tu_queue *queue)
{
- if (queue->submit_fence_fd >= 0) {
- close(queue->submit_fence_fd);
- }
+ tu_fence_finish(&queue->submit_fence);
tu_drm_submitqueue_close(queue->device, queue->msm_queue_id);
}
TU_FROM_HANDLE(tu_physical_device, physical_device, physicalDevice);
VkResult result;
struct tu_device *device;
+ bool custom_border_colors = false;
/* Check enabled features */
if (pCreateInfo->pEnabledFeatures) {
}
}
+ vk_foreach_struct_const(ext, pCreateInfo->pNext) {
+ switch (ext->sType) {
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CUSTOM_BORDER_COLOR_FEATURES_EXT: {
+ const VkPhysicalDeviceCustomBorderColorFeaturesEXT *border_color_features = (const void *)ext;
+ custom_border_colors = border_color_features->customBorderColors;
+ break;
+ }
+ default:
+ break;
+ }
+ }
+
device = vk_zalloc2(&physical_device->instance->alloc, pAllocator,
sizeof(*device), 8, VK_SYSTEM_ALLOCATION_SCOPE_DEVICE);
if (!device)
return vk_error(physical_device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
- device->_loader_data.loaderMagic = ICD_LOADER_MAGIC;
+ vk_device_init(&device->vk, pCreateInfo,
+ &physical_device->instance->alloc, pAllocator);
+
device->instance = physical_device->instance;
device->physical_device = physical_device;
-
- if (pAllocator)
- device->alloc = *pAllocator;
- else
- device->alloc = physical_device->instance->alloc;
+ device->_lost = false;
for (uint32_t i = 0; i < pCreateInfo->enabledExtensionCount; i++) {
const char *ext_name = pCreateInfo->ppEnabledExtensionNames[i];
int index = tu_get_device_extension_index(ext_name);
if (index < 0 ||
!physical_device->supported_extensions.extensions[index]) {
- vk_free(&device->alloc, device);
+ vk_free(&device->vk.alloc, device);
return vk_error(physical_device->instance,
VK_ERROR_EXTENSION_NOT_PRESENT);
}
&pCreateInfo->pQueueCreateInfos[i];
uint32_t qfi = queue_create->queueFamilyIndex;
device->queues[qfi] = vk_alloc(
- &device->alloc, queue_create->queueCount * sizeof(struct tu_queue),
+ &device->vk.alloc, queue_create->queueCount * sizeof(struct tu_queue),
8, VK_SYSTEM_ALLOCATION_SCOPE_DEVICE);
if (!device->queues[qfi]) {
result = VK_ERROR_OUT_OF_HOST_MEMORY;
- goto fail;
+ goto fail_queues;
}
memset(device->queues[qfi], 0,
result = tu_queue_init(device, &device->queues[qfi][q], qfi, q,
queue_create->flags);
if (result != VK_SUCCESS)
- goto fail;
+ goto fail_queues;
}
}
+ device->compiler = ir3_compiler_create(NULL, physical_device->gpu_id);
+ if (!device->compiler)
+ goto fail_queues;
+
+ /* initial sizes, these will increase if there is overflow */
+ device->vsc_draw_strm_pitch = 0x1000 + VSC_PAD;
+ device->vsc_prim_strm_pitch = 0x4000 + VSC_PAD;
+
+ uint32_t global_size = sizeof(struct tu6_global);
+ if (custom_border_colors)
+ global_size += TU_BORDER_COLOR_COUNT * sizeof(struct bcolor_entry);
+
+ result = tu_bo_init_new(device, &device->global_bo, global_size);
+ if (result != VK_SUCCESS)
+ goto fail_global_bo;
+
+ result = tu_bo_map(device, &device->global_bo);
+ if (result != VK_SUCCESS)
+ goto fail_global_bo_map;
+
+ struct tu6_global *global = device->global_bo.map;
+ tu_init_clear_blit_shaders(device->global_bo.map);
+ global->predicate = 0;
+ tu6_pack_border_color(&global->bcolor_builtin[VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK],
+ &(VkClearColorValue) {}, false);
+ tu6_pack_border_color(&global->bcolor_builtin[VK_BORDER_COLOR_INT_TRANSPARENT_BLACK],
+ &(VkClearColorValue) {}, true);
+ tu6_pack_border_color(&global->bcolor_builtin[VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK],
+ &(VkClearColorValue) { .float32[3] = 1.0f }, false);
+ tu6_pack_border_color(&global->bcolor_builtin[VK_BORDER_COLOR_INT_OPAQUE_BLACK],
+ &(VkClearColorValue) { .int32[3] = 1 }, true);
+ tu6_pack_border_color(&global->bcolor_builtin[VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE],
+ &(VkClearColorValue) { .float32[0 ... 3] = 1.0f }, false);
+ tu6_pack_border_color(&global->bcolor_builtin[VK_BORDER_COLOR_INT_OPAQUE_WHITE],
+ &(VkClearColorValue) { .int32[0 ... 3] = 1 }, true);
+
+ /* initialize to ones so ffs can be used to find unused slots */
+ BITSET_ONES(device->custom_border_color);
+
VkPipelineCacheCreateInfo ci;
ci.sType = VK_STRUCTURE_TYPE_PIPELINE_CACHE_CREATE_INFO;
ci.pNext = NULL;
result =
tu_CreatePipelineCache(tu_device_to_handle(device), &ci, NULL, &pc);
if (result != VK_SUCCESS)
- goto fail;
+ goto fail_pipeline_cache;
device->mem_cache = tu_pipeline_cache_from_handle(pc);
+ for (unsigned i = 0; i < ARRAY_SIZE(device->scratch_bos); i++)
+ mtx_init(&device->scratch_bos[i].construct_mtx, mtx_plain);
+
+ mtx_init(&device->mutex, mtx_plain);
+
*pDevice = tu_device_to_handle(device);
return VK_SUCCESS;
-fail:
+fail_pipeline_cache:
+fail_global_bo_map:
+ tu_bo_finish(device, &device->global_bo);
+
+fail_global_bo:
+ ralloc_free(device->compiler);
+
+fail_queues:
for (unsigned i = 0; i < TU_MAX_QUEUE_FAMILIES; i++) {
for (unsigned q = 0; q < device->queue_count[i]; q++)
tu_queue_finish(&device->queues[i][q]);
if (device->queue_count[i])
- vk_free(&device->alloc, device->queues[i]);
+ vk_object_free(&device->vk, NULL, device->queues[i]);
}
- vk_free(&device->alloc, device);
+ vk_free(&device->vk.alloc, device);
return result;
}
for (unsigned q = 0; q < device->queue_count[i]; q++)
tu_queue_finish(&device->queues[i][q]);
if (device->queue_count[i])
- vk_free(&device->alloc, device->queues[i]);
+ vk_object_free(&device->vk, NULL, device->queues[i]);
}
+ for (unsigned i = 0; i < ARRAY_SIZE(device->scratch_bos); i++) {
+ if (device->scratch_bos[i].initialized)
+ tu_bo_finish(device, &device->scratch_bos[i].bo);
+ }
+
+ ir3_compiler_destroy(device->compiler);
+
VkPipelineCache pc = tu_pipeline_cache_to_handle(device->mem_cache);
tu_DestroyPipelineCache(tu_device_to_handle(device), pc, NULL);
- vk_free(&device->alloc, device);
+ vk_free(&device->vk.alloc, device);
+}
+
+VkResult
+_tu_device_set_lost(struct tu_device *device,
+ const char *file, int line,
+ const char *msg, ...)
+{
+ /* Set the flag indicating that waits should return in finite time even
+ * after device loss.
+ */
+ p_atomic_inc(&device->_lost);
+
+ /* TODO: Report the log message through VkDebugReportCallbackEXT instead */
+ fprintf(stderr, "%s:%d: ", file, line);
+ va_list ap;
+ va_start(ap, msg);
+ vfprintf(stderr, msg, ap);
+ va_end(ap);
+
+ if (env_var_as_boolean("TU_ABORT_ON_DEVICE_LOSS", false))
+ abort();
+
+ return VK_ERROR_DEVICE_LOST;
+}
+
+VkResult
+tu_get_scratch_bo(struct tu_device *dev, uint64_t size, struct tu_bo **bo)
+{
+ unsigned size_log2 = MAX2(util_logbase2_ceil64(size), MIN_SCRATCH_BO_SIZE_LOG2);
+ unsigned index = size_log2 - MIN_SCRATCH_BO_SIZE_LOG2;
+ assert(index < ARRAY_SIZE(dev->scratch_bos));
+
+ for (unsigned i = index; i < ARRAY_SIZE(dev->scratch_bos); i++) {
+ if (p_atomic_read(&dev->scratch_bos[i].initialized)) {
+ /* Fast path: just return the already-allocated BO. */
+ *bo = &dev->scratch_bos[i].bo;
+ return VK_SUCCESS;
+ }
+ }
+
+ /* Slow path: actually allocate the BO. We take a lock because the process
+ * of allocating it is slow, and we don't want to block the CPU while it
+ * finishes.
+ */
+ mtx_lock(&dev->scratch_bos[index].construct_mtx);
+
+ /* Another thread may have allocated it already while we were waiting on
+ * the lock. We need to check this in order to avoid double-allocating.
+ */
+ if (dev->scratch_bos[index].initialized) {
+ mtx_unlock(&dev->scratch_bos[index].construct_mtx);
+ *bo = &dev->scratch_bos[index].bo;
+ return VK_SUCCESS;
+ }
+
+ unsigned bo_size = 1ull << size_log2;
+ VkResult result = tu_bo_init_new(dev, &dev->scratch_bos[index].bo, bo_size);
+ if (result != VK_SUCCESS) {
+ mtx_unlock(&dev->scratch_bos[index].construct_mtx);
+ return result;
+ }
+
+ p_atomic_set(&dev->scratch_bos[index].initialized, true);
+
+ mtx_unlock(&dev->scratch_bos[index].construct_mtx);
+
+ *bo = &dev->scratch_bos[index].bo;
+ return VK_SUCCESS;
}
VkResult
tu_GetDeviceQueue2(_device, &info, pQueue);
}
-VkResult
-tu_QueueSubmit(VkQueue _queue,
- uint32_t submitCount,
- const VkSubmitInfo *pSubmits,
- VkFence _fence)
-{
- TU_FROM_HANDLE(tu_queue, queue, _queue);
-
- for (uint32_t i = 0; i < submitCount; ++i) {
- const VkSubmitInfo *submit = pSubmits + i;
- const bool last_submit = (i == submitCount - 1);
- struct tu_bo_list bo_list;
- tu_bo_list_init(&bo_list);
-
- uint32_t entry_count = 0;
- for(uint32_t j = 0; j < submit->commandBufferCount; ++j) {
- TU_FROM_HANDLE(tu_cmd_buffer, cmdbuf, submit->pCommandBuffers[j]);
- entry_count += cmdbuf->cs.entry_count;
- }
-
- struct drm_msm_gem_submit_cmd cmds[entry_count];
- uint32_t entry_idx = 0;
- for(uint32_t j = 0; j < submit->commandBufferCount; ++j) {
- TU_FROM_HANDLE(tu_cmd_buffer, cmdbuf, submit->pCommandBuffers[j]);
- struct tu_cmd_stream *stream = &cmdbuf->cs;
- for (unsigned i = 0; i < stream->entry_count; ++i, ++entry_idx) {
- cmds[entry_idx].type = MSM_SUBMIT_CMD_BUF;
- cmds[entry_idx].submit_idx = tu_bo_list_add(&bo_list, stream->entries[i].bo);
- cmds[entry_idx].submit_offset = stream->entries[i].offset;
- cmds[entry_idx].size = stream->entries[i].size;
- cmds[entry_idx].pad = 0;
- cmds[entry_idx].nr_relocs = 0;
- cmds[entry_idx].relocs = 0;
-
- }
- }
-
- struct drm_msm_gem_submit_bo bos[bo_list.count];
- for (unsigned i = 0; i < bo_list.count; ++i) {
- bos[i].flags = MSM_SUBMIT_BO_READ | MSM_SUBMIT_BO_WRITE;
- bos[i].handle = bo_list.handles[i];
- bos[i].presumed = 0;
- }
-
- uint32_t flags = MSM_PIPE_3D0;
- if (last_submit) {
- flags |= MSM_SUBMIT_FENCE_FD_OUT;
- }
-
- struct drm_msm_gem_submit req = {
- .flags = flags,
- .queueid = queue->msm_queue_id,
- .bos = (uint64_t)(uintptr_t)bos,
- .nr_bos = bo_list.count,
- .cmds = (uint64_t)(uintptr_t)cmds,
- .nr_cmds = entry_count,
- };
-
- int ret = drmCommandWriteRead(queue->device->physical_device->local_fd,
- DRM_MSM_GEM_SUBMIT,
- &req, sizeof(req));
- if (ret) {
- fprintf(stderr, "submit failed: %s\n", strerror(errno));
- abort();
- }
-
- tu_bo_list_destroy(&bo_list);
-
- if (last_submit) {
- /* no need to merge fences as queue execution is serialized */
- if (queue->submit_fence_fd >= 0) {
- close(queue->submit_fence_fd);
- }
- queue->submit_fence_fd = req.fence_fd;
- }
- }
- return VK_SUCCESS;
-}
-
VkResult
tu_QueueWaitIdle(VkQueue _queue)
{
TU_FROM_HANDLE(tu_queue, queue, _queue);
- if (queue->submit_fence_fd >= 0) {
- int ret = sync_wait(queue->submit_fence_fd, -1);
- if (ret)
- tu_loge("sync_wait on fence fd %d failed", queue->submit_fence_fd);
+ if (tu_device_is_lost(queue->device))
+ return VK_ERROR_DEVICE_LOST;
- close(queue->submit_fence_fd);
- queue->submit_fence_fd = -1;
- }
+ tu_fence_wait_idle(&queue->submit_fence);
return VK_SUCCESS;
}
{
TU_FROM_HANDLE(tu_device, device, _device);
+ if (tu_device_is_lost(device))
+ return VK_ERROR_DEVICE_LOST;
+
for (unsigned i = 0; i < TU_MAX_QUEUE_FAMILIES; i++) {
for (unsigned q = 0; q < device->queue_count[i]; q++) {
tu_QueueWaitIdle(tu_queue_to_handle(&device->queues[i][q]));
return vk_error(NULL, VK_ERROR_LAYER_NOT_PRESENT);
for (int i = 0; i < TU_INSTANCE_EXTENSION_COUNT; i++) {
- if (tu_supported_instance_extensions.extensions[i]) {
+ if (tu_instance_extensions_supported.extensions[i]) {
vk_outarray_append(&out, prop) { *prop = tu_instance_extensions[i]; }
}
}
return VK_SUCCESS;
}
- mem = vk_alloc2(&device->alloc, pAllocator, sizeof(*mem), 8,
- VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
+ mem = vk_object_alloc(&device->vk, pAllocator, sizeof(*mem),
+ VK_OBJECT_TYPE_DEVICE_MEMORY);
if (mem == NULL)
return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
- result = tu_bo_init_new(device, &mem->bo, pAllocateInfo->allocationSize);
+ const VkImportMemoryFdInfoKHR *fd_info =
+ vk_find_struct_const(pAllocateInfo->pNext, IMPORT_MEMORY_FD_INFO_KHR);
+ if (fd_info && !fd_info->handleType)
+ fd_info = NULL;
+
+ if (fd_info) {
+ assert(fd_info->handleType ==
+ VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT ||
+ fd_info->handleType ==
+ VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT);
+
+ /*
+ * TODO Importing the same fd twice gives us the same handle without
+ * reference counting. We need to maintain a per-instance handle-to-bo
+ * table and add reference count to tu_bo.
+ */
+ result = tu_bo_init_dmabuf(device, &mem->bo,
+ pAllocateInfo->allocationSize, fd_info->fd);
+ if (result == VK_SUCCESS) {
+ /* take ownership and close the fd */
+ close(fd_info->fd);
+ }
+ } else {
+ result =
+ tu_bo_init_new(device, &mem->bo, pAllocateInfo->allocationSize);
+ }
+
if (result != VK_SUCCESS) {
- vk_free2(&device->alloc, pAllocator, mem);
+ vk_object_free(&device->vk, pAllocator, mem);
return result;
}
return;
tu_bo_finish(device, &mem->bo);
- vk_free2(&device->alloc, pAllocator, mem);
+ vk_object_free(&device->vk, pAllocator, mem);
}
VkResult
TU_FROM_HANDLE(tu_buffer, buffer, _buffer);
pMemoryRequirements->memoryTypeBits = 1;
- pMemoryRequirements->alignment = 16;
+ pMemoryRequirements->alignment = 64;
pMemoryRequirements->size =
align64(buffer->size, pMemoryRequirements->alignment);
}
void
tu_GetBufferMemoryRequirements2(
VkDevice device,
- const VkBufferMemoryRequirementsInfo2KHR *pInfo,
- VkMemoryRequirements2KHR *pMemoryRequirements)
+ const VkBufferMemoryRequirementsInfo2 *pInfo,
+ VkMemoryRequirements2 *pMemoryRequirements)
{
tu_GetBufferMemoryRequirements(device, pInfo->buffer,
&pMemoryRequirements->memoryRequirements);
TU_FROM_HANDLE(tu_image, image, _image);
pMemoryRequirements->memoryTypeBits = 1;
- pMemoryRequirements->size = image->size;
- pMemoryRequirements->alignment = image->alignment;
+ pMemoryRequirements->size = image->total_size;
+ pMemoryRequirements->alignment = image->layout[0].base_align;
}
void
tu_GetImageMemoryRequirements2(VkDevice device,
- const VkImageMemoryRequirementsInfo2KHR *pInfo,
- VkMemoryRequirements2KHR *pMemoryRequirements)
+ const VkImageMemoryRequirementsInfo2 *pInfo,
+ VkMemoryRequirements2 *pMemoryRequirements)
{
tu_GetImageMemoryRequirements(device, pInfo->image,
&pMemoryRequirements->memoryRequirements);
void
tu_GetImageSparseMemoryRequirements2(
VkDevice device,
- const VkImageSparseMemoryRequirementsInfo2KHR *pInfo,
+ const VkImageSparseMemoryRequirementsInfo2 *pInfo,
uint32_t *pSparseMemoryRequirementCount,
- VkSparseImageMemoryRequirements2KHR *pSparseMemoryRequirements)
+ VkSparseImageMemoryRequirements2 *pSparseMemoryRequirements)
{
tu_stub();
}
VkResult
tu_BindBufferMemory2(VkDevice device,
uint32_t bindInfoCount,
- const VkBindBufferMemoryInfoKHR *pBindInfos)
+ const VkBindBufferMemoryInfo *pBindInfos)
{
+ for (uint32_t i = 0; i < bindInfoCount; ++i) {
+ TU_FROM_HANDLE(tu_device_memory, mem, pBindInfos[i].memory);
+ TU_FROM_HANDLE(tu_buffer, buffer, pBindInfos[i].buffer);
+
+ if (mem) {
+ buffer->bo = &mem->bo;
+ buffer->bo_offset = pBindInfos[i].memoryOffset;
+ } else {
+ buffer->bo = NULL;
+ }
+ }
return VK_SUCCESS;
}
VkDeviceMemory memory,
VkDeviceSize memoryOffset)
{
- const VkBindBufferMemoryInfoKHR info = {
- .sType = VK_STRUCTURE_TYPE_BIND_BUFFER_MEMORY_INFO_KHR,
+ const VkBindBufferMemoryInfo info = {
+ .sType = VK_STRUCTURE_TYPE_BIND_BUFFER_MEMORY_INFO,
.buffer = buffer,
.memory = memory,
.memoryOffset = memoryOffset
VkResult
tu_BindImageMemory2(VkDevice device,
uint32_t bindInfoCount,
- const VkBindImageMemoryInfoKHR *pBindInfos)
-{
+ const VkBindImageMemoryInfo *pBindInfos)
+{
+ for (uint32_t i = 0; i < bindInfoCount; ++i) {
+ TU_FROM_HANDLE(tu_image, image, pBindInfos[i].image);
+ TU_FROM_HANDLE(tu_device_memory, mem, pBindInfos[i].memory);
+
+ if (mem) {
+ image->bo = &mem->bo;
+ image->bo_offset = pBindInfos[i].memoryOffset;
+ } else {
+ image->bo = NULL;
+ image->bo_offset = 0;
+ }
+ }
+
return VK_SUCCESS;
}
VkDeviceMemory memory,
VkDeviceSize memoryOffset)
{
- const VkBindImageMemoryInfoKHR info = {
- .sType = VK_STRUCTURE_TYPE_BIND_BUFFER_MEMORY_INFO_KHR,
+ const VkBindImageMemoryInfo info = {
+ .sType = VK_STRUCTURE_TYPE_BIND_BUFFER_MEMORY_INFO,
.image = image,
.memory = memory,
.memoryOffset = memoryOffset
return VK_SUCCESS;
}
-VkResult
-tu_CreateFence(VkDevice _device,
- const VkFenceCreateInfo *pCreateInfo,
- const VkAllocationCallbacks *pAllocator,
- VkFence *pFence)
-{
- TU_FROM_HANDLE(tu_device, device, _device);
-
- struct tu_fence *fence =
- vk_alloc2(&device->alloc, pAllocator, sizeof(*fence), 8,
- VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
-
- if (!fence)
- return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
-
- *pFence = tu_fence_to_handle(fence);
-
- return VK_SUCCESS;
-}
-
-void
-tu_DestroyFence(VkDevice _device,
- VkFence _fence,
- const VkAllocationCallbacks *pAllocator)
-{
- TU_FROM_HANDLE(tu_device, device, _device);
- TU_FROM_HANDLE(tu_fence, fence, _fence);
-
- if (!fence)
- return;
-
- vk_free2(&device->alloc, pAllocator, fence);
-}
-
-VkResult
-tu_WaitForFences(VkDevice _device,
- uint32_t fenceCount,
- const VkFence *pFences,
- VkBool32 waitAll,
- uint64_t timeout)
-{
- return VK_SUCCESS;
-}
-
-VkResult
-tu_ResetFences(VkDevice _device, uint32_t fenceCount, const VkFence *pFences)
-{
- return VK_SUCCESS;
-}
-
-VkResult
-tu_GetFenceStatus(VkDevice _device, VkFence _fence)
-{
- return VK_SUCCESS;
-}
-
-// Queue semaphore functions
-
-VkResult
-tu_CreateSemaphore(VkDevice _device,
- const VkSemaphoreCreateInfo *pCreateInfo,
- const VkAllocationCallbacks *pAllocator,
- VkSemaphore *pSemaphore)
-{
- TU_FROM_HANDLE(tu_device, device, _device);
-
- struct tu_semaphore *sem =
- vk_alloc2(&device->alloc, pAllocator, sizeof(*sem), 8,
- VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
- if (!sem)
- return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
-
- *pSemaphore = tu_semaphore_to_handle(sem);
- return VK_SUCCESS;
-}
-
-void
-tu_DestroySemaphore(VkDevice _device,
- VkSemaphore _semaphore,
- const VkAllocationCallbacks *pAllocator)
-{
- TU_FROM_HANDLE(tu_device, device, _device);
- TU_FROM_HANDLE(tu_semaphore, sem, _semaphore);
- if (!_semaphore)
- return;
-
- vk_free2(&device->alloc, pAllocator, sem);
-}
VkResult
tu_CreateEvent(VkDevice _device,
VkEvent *pEvent)
{
TU_FROM_HANDLE(tu_device, device, _device);
- struct tu_event *event =
- vk_alloc2(&device->alloc, pAllocator, sizeof(*event), 8,
- VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
+ struct tu_event *event =
+ vk_object_alloc(&device->vk, pAllocator, sizeof(*event),
+ VK_OBJECT_TYPE_EVENT);
if (!event)
return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
+ VkResult result = tu_bo_init_new(device, &event->bo, 0x1000);
+ if (result != VK_SUCCESS)
+ goto fail_alloc;
+
+ result = tu_bo_map(device, &event->bo);
+ if (result != VK_SUCCESS)
+ goto fail_map;
+
*pEvent = tu_event_to_handle(event);
return VK_SUCCESS;
+
+fail_map:
+ tu_bo_finish(device, &event->bo);
+fail_alloc:
+ vk_object_free(&device->vk, pAllocator, event);
+ return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
}
void
if (!event)
return;
- vk_free2(&device->alloc, pAllocator, event);
+
+ tu_bo_finish(device, &event->bo);
+ vk_object_free(&device->vk, pAllocator, event);
}
VkResult
{
TU_FROM_HANDLE(tu_event, event, _event);
- if (*event->map == 1)
+ if (*(uint64_t*) event->bo.map == 1)
return VK_EVENT_SET;
return VK_EVENT_RESET;
}
tu_SetEvent(VkDevice _device, VkEvent _event)
{
TU_FROM_HANDLE(tu_event, event, _event);
- *event->map = 1;
+ *(uint64_t*) event->bo.map = 1;
return VK_SUCCESS;
}
tu_ResetEvent(VkDevice _device, VkEvent _event)
{
TU_FROM_HANDLE(tu_event, event, _event);
- *event->map = 0;
+ *(uint64_t*) event->bo.map = 0;
return VK_SUCCESS;
}
assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO);
- buffer = vk_alloc2(&device->alloc, pAllocator, sizeof(*buffer), 8,
- VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
+ buffer = vk_object_alloc(&device->vk, pAllocator, sizeof(*buffer),
+ VK_OBJECT_TYPE_BUFFER);
if (buffer == NULL)
return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
if (!buffer)
return;
- vk_free2(&device->alloc, pAllocator, buffer);
-}
-
-static uint32_t
-tu_surface_max_layer_count(struct tu_image_view *iview)
-{
- return iview->type == VK_IMAGE_VIEW_TYPE_3D
- ? iview->extent.depth
- : (iview->base_layer + iview->layer_count);
+ vk_object_free(&device->vk, pAllocator, buffer);
}
VkResult
VkFramebuffer *pFramebuffer)
{
TU_FROM_HANDLE(tu_device, device, _device);
+ TU_FROM_HANDLE(tu_render_pass, pass, pCreateInfo->renderPass);
struct tu_framebuffer *framebuffer;
assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO);
size_t size = sizeof(*framebuffer) + sizeof(struct tu_attachment_info) *
pCreateInfo->attachmentCount;
- framebuffer = vk_alloc2(&device->alloc, pAllocator, size, 8,
- VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
+ framebuffer = vk_object_alloc(&device->vk, pAllocator, size,
+ VK_OBJECT_TYPE_FRAMEBUFFER);
if (framebuffer == NULL)
return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
VkImageView _iview = pCreateInfo->pAttachments[i];
struct tu_image_view *iview = tu_image_view_from_handle(_iview);
framebuffer->attachments[i].attachment = iview;
-
- framebuffer->width = MIN2(framebuffer->width, iview->extent.width);
- framebuffer->height = MIN2(framebuffer->height, iview->extent.height);
- framebuffer->layers =
- MIN2(framebuffer->layers, tu_surface_max_layer_count(iview));
}
+ tu_framebuffer_tiling_config(framebuffer, device, pass);
+
*pFramebuffer = tu_framebuffer_to_handle(framebuffer);
return VK_SUCCESS;
}
if (!fb)
return;
- vk_free2(&device->alloc, pAllocator, fb);
+
+ vk_object_free(&device->vk, pAllocator, fb);
}
static void
struct tu_sampler *sampler,
const VkSamplerCreateInfo *pCreateInfo)
{
+ const struct VkSamplerReductionModeCreateInfo *reduction =
+ vk_find_struct_const(pCreateInfo->pNext, SAMPLER_REDUCTION_MODE_CREATE_INFO);
+ const struct VkSamplerYcbcrConversionInfo *ycbcr_conversion =
+ vk_find_struct_const(pCreateInfo->pNext, SAMPLER_YCBCR_CONVERSION_INFO);
+ const VkSamplerCustomBorderColorCreateInfoEXT *custom_border_color =
+ vk_find_struct_const(pCreateInfo->pNext, SAMPLER_CUSTOM_BORDER_COLOR_CREATE_INFO_EXT);
+ /* for non-custom border colors, the VK enum is translated directly to an offset in
+ * the border color buffer. custom border colors are located immediately after the
+ * builtin colors, and thus an offset of TU_BORDER_COLOR_BUILTIN is added.
+ */
+ uint32_t border_color = (unsigned) pCreateInfo->borderColor;
+ if (pCreateInfo->borderColor == VK_BORDER_COLOR_FLOAT_CUSTOM_EXT ||
+ pCreateInfo->borderColor == VK_BORDER_COLOR_INT_CUSTOM_EXT) {
+ mtx_lock(&device->mutex);
+ border_color = BITSET_FFS(device->custom_border_color);
+ BITSET_CLEAR(device->custom_border_color, border_color);
+ mtx_unlock(&device->mutex);
+ tu6_pack_border_color(device->global_bo.map + gb_offset(bcolor[border_color]),
+ &custom_border_color->customBorderColor,
+ pCreateInfo->borderColor == VK_BORDER_COLOR_INT_CUSTOM_EXT);
+ border_color += TU_BORDER_COLOR_BUILTIN;
+ }
+
+ unsigned aniso = pCreateInfo->anisotropyEnable ?
+ util_last_bit(MIN2((uint32_t)pCreateInfo->maxAnisotropy >> 1, 8)) : 0;
+ bool miplinear = (pCreateInfo->mipmapMode == VK_SAMPLER_MIPMAP_MODE_LINEAR);
+ float min_lod = CLAMP(pCreateInfo->minLod, 0.0f, 4095.0f / 256.0f);
+ float max_lod = CLAMP(pCreateInfo->maxLod, 0.0f, 4095.0f / 256.0f);
+
+ sampler->descriptor[0] =
+ COND(miplinear, A6XX_TEX_SAMP_0_MIPFILTER_LINEAR_NEAR) |
+ A6XX_TEX_SAMP_0_XY_MAG(tu6_tex_filter(pCreateInfo->magFilter, aniso)) |
+ A6XX_TEX_SAMP_0_XY_MIN(tu6_tex_filter(pCreateInfo->minFilter, aniso)) |
+ A6XX_TEX_SAMP_0_ANISO(aniso) |
+ A6XX_TEX_SAMP_0_WRAP_S(tu6_tex_wrap(pCreateInfo->addressModeU)) |
+ A6XX_TEX_SAMP_0_WRAP_T(tu6_tex_wrap(pCreateInfo->addressModeV)) |
+ A6XX_TEX_SAMP_0_WRAP_R(tu6_tex_wrap(pCreateInfo->addressModeW)) |
+ A6XX_TEX_SAMP_0_LOD_BIAS(pCreateInfo->mipLodBias);
+ sampler->descriptor[1] =
+ /* COND(!cso->seamless_cube_map, A6XX_TEX_SAMP_1_CUBEMAPSEAMLESSFILTOFF) | */
+ COND(pCreateInfo->unnormalizedCoordinates, A6XX_TEX_SAMP_1_UNNORM_COORDS) |
+ A6XX_TEX_SAMP_1_MIN_LOD(min_lod) |
+ A6XX_TEX_SAMP_1_MAX_LOD(max_lod) |
+ COND(pCreateInfo->compareEnable,
+ A6XX_TEX_SAMP_1_COMPARE_FUNC(tu6_compare_func(pCreateInfo->compareOp)));
+ sampler->descriptor[2] = A6XX_TEX_SAMP_2_BCOLOR(border_color);
+ sampler->descriptor[3] = 0;
+
+ if (reduction) {
+ sampler->descriptor[2] |= A6XX_TEX_SAMP_2_REDUCTION_MODE(
+ tu6_reduction_mode(reduction->reductionMode));
+ }
+
+ sampler->ycbcr_sampler = ycbcr_conversion ?
+ tu_sampler_ycbcr_conversion_from_handle(ycbcr_conversion->conversion) : NULL;
+
+ if (sampler->ycbcr_sampler &&
+ sampler->ycbcr_sampler->chroma_filter == VK_FILTER_LINEAR) {
+ sampler->descriptor[2] |= A6XX_TEX_SAMP_2_CHROMA_LINEAR;
+ }
+
+ /* TODO:
+ * A6XX_TEX_SAMP_1_MIPFILTER_LINEAR_FAR disables mipmapping, but vk has no NONE mipfilter?
+ */
}
VkResult
assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO);
- sampler = vk_alloc2(&device->alloc, pAllocator, sizeof(*sampler), 8,
- VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
+ sampler = vk_object_alloc(&device->vk, pAllocator, sizeof(*sampler),
+ VK_OBJECT_TYPE_SAMPLER);
if (!sampler)
return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
{
TU_FROM_HANDLE(tu_device, device, _device);
TU_FROM_HANDLE(tu_sampler, sampler, _sampler);
+ uint32_t border_color;
if (!sampler)
return;
- vk_free2(&device->alloc, pAllocator, sampler);
+
+ border_color = (sampler->descriptor[2] & A6XX_TEX_SAMP_2_BCOLOR__MASK) >> A6XX_TEX_SAMP_2_BCOLOR__SHIFT;
+ if (border_color >= TU_BORDER_COLOR_BUILTIN) {
+ border_color -= TU_BORDER_COLOR_BUILTIN;
+ /* if the sampler had a custom border color, free it. TODO: no lock */
+ mtx_lock(&device->mutex);
+ assert(!BITSET_TEST(device->custom_border_color, border_color));
+ BITSET_SET(device->custom_border_color, border_color);
+ mtx_unlock(&device->mutex);
+ }
+
+ vk_object_free(&device->vk, pAllocator, sampler);
}
/* vk_icd.h does not declare this function, so we declare it here to
return VK_SUCCESS;
}
-void
-tu_GetPhysicalDeviceExternalSemaphoreProperties(
- VkPhysicalDevice physicalDevice,
- const VkPhysicalDeviceExternalSemaphoreInfoKHR *pExternalSemaphoreInfo,
- VkExternalSemaphorePropertiesKHR *pExternalSemaphoreProperties)
+VkResult
+tu_GetMemoryFdKHR(VkDevice _device,
+ const VkMemoryGetFdInfoKHR *pGetFdInfo,
+ int *pFd)
+{
+ TU_FROM_HANDLE(tu_device, device, _device);
+ TU_FROM_HANDLE(tu_device_memory, memory, pGetFdInfo->memory);
+
+ assert(pGetFdInfo->sType == VK_STRUCTURE_TYPE_MEMORY_GET_FD_INFO_KHR);
+
+ /* At the moment, we support only the below handle types. */
+ assert(pGetFdInfo->handleType ==
+ VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT ||
+ pGetFdInfo->handleType ==
+ VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT);
+
+ int prime_fd = tu_bo_export_dmabuf(device, &memory->bo);
+ if (prime_fd < 0)
+ return vk_error(device->instance, VK_ERROR_OUT_OF_DEVICE_MEMORY);
+
+ *pFd = prime_fd;
+ return VK_SUCCESS;
+}
+
+VkResult
+tu_GetMemoryFdPropertiesKHR(VkDevice _device,
+ VkExternalMemoryHandleTypeFlagBits handleType,
+ int fd,
+ VkMemoryFdPropertiesKHR *pMemoryFdProperties)
+{
+ assert(handleType == VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT);
+ pMemoryFdProperties->memoryTypeBits = 1;
+ return VK_SUCCESS;
+}
+
+VkResult
+tu_ImportFenceFdKHR(VkDevice _device,
+ const VkImportFenceFdInfoKHR *pImportFenceFdInfo)
+{
+ tu_stub();
+
+ return VK_SUCCESS;
+}
+
+VkResult
+tu_GetFenceFdKHR(VkDevice _device,
+ const VkFenceGetFdInfoKHR *pGetFdInfo,
+ int *pFd)
{
- pExternalSemaphoreProperties->exportFromImportedHandleTypes = 0;
- pExternalSemaphoreProperties->compatibleHandleTypes = 0;
- pExternalSemaphoreProperties->externalSemaphoreFeatures = 0;
+ tu_stub();
+
+ return VK_SUCCESS;
}
void
tu_GetPhysicalDeviceExternalFenceProperties(
VkPhysicalDevice physicalDevice,
- const VkPhysicalDeviceExternalFenceInfoKHR *pExternalFenceInfo,
- VkExternalFencePropertiesKHR *pExternalFenceProperties)
+ const VkPhysicalDeviceExternalFenceInfo *pExternalFenceInfo,
+ VkExternalFenceProperties *pExternalFenceProperties)
{
pExternalFenceProperties->exportFromImportedHandleTypes = 0;
pExternalFenceProperties->compatibleHandleTypes = 0;
VK_PEER_MEMORY_FEATURE_GENERIC_SRC_BIT |
VK_PEER_MEMORY_FEATURE_GENERIC_DST_BIT;
}
+
+void tu_GetPhysicalDeviceMultisamplePropertiesEXT(
+ VkPhysicalDevice physicalDevice,
+ VkSampleCountFlagBits samples,
+ VkMultisamplePropertiesEXT* pMultisampleProperties)
+{
+ TU_FROM_HANDLE(tu_physical_device, pdevice, physicalDevice);
+
+ if (samples <= VK_SAMPLE_COUNT_4_BIT && pdevice->supported_extensions.EXT_sample_locations)
+ pMultisampleProperties->maxSampleLocationGridSize = (VkExtent2D){ 1, 1 };
+ else
+ pMultisampleProperties->maxSampleLocationGridSize = (VkExtent2D){ 0, 0 };
+}
+
+
+VkResult
+tu_CreatePrivateDataSlotEXT(VkDevice _device,
+ const VkPrivateDataSlotCreateInfoEXT* pCreateInfo,
+ const VkAllocationCallbacks* pAllocator,
+ VkPrivateDataSlotEXT* pPrivateDataSlot)
+{
+ TU_FROM_HANDLE(tu_device, device, _device);
+ return vk_private_data_slot_create(&device->vk,
+ pCreateInfo,
+ pAllocator,
+ pPrivateDataSlot);
+}
+
+void
+tu_DestroyPrivateDataSlotEXT(VkDevice _device,
+ VkPrivateDataSlotEXT privateDataSlot,
+ const VkAllocationCallbacks* pAllocator)
+{
+ TU_FROM_HANDLE(tu_device, device, _device);
+ vk_private_data_slot_destroy(&device->vk, privateDataSlot, pAllocator);
+}
+
+VkResult
+tu_SetPrivateDataEXT(VkDevice _device,
+ VkObjectType objectType,
+ uint64_t objectHandle,
+ VkPrivateDataSlotEXT privateDataSlot,
+ uint64_t data)
+{
+ TU_FROM_HANDLE(tu_device, device, _device);
+ return vk_object_base_set_private_data(&device->vk,
+ objectType,
+ objectHandle,
+ privateDataSlot,
+ data);
+}
+
+void
+tu_GetPrivateDataEXT(VkDevice _device,
+ VkObjectType objectType,
+ uint64_t objectHandle,
+ VkPrivateDataSlotEXT privateDataSlot,
+ uint64_t* pData)
+{
+ TU_FROM_HANDLE(tu_device, device, _device);
+ vk_object_base_get_private_data(&device->vk,
+ objectType,
+ objectHandle,
+ privateDataSlot,
+ pData);
+}