#include <string.h>
#include <unistd.h>
#include <fcntl.h>
+#include "radv_debug.h"
#include "radv_private.h"
#include "radv_cs.h"
#include "util/disk_cache.h"
#include "util/strtod.h"
-#include "util/vk_util.h"
+#include "vk_util.h"
#include <xf86drm.h>
#include <amdgpu.h>
#include <amdgpu_drm.h>
#include "ac_llvm_util.h"
#include "vk_format.h"
#include "sid.h"
+#include "gfx9d.h"
#include "util/debug.h"
static int
return 0;
}
+static void
+radv_get_driver_uuid(void *uuid)
+{
+ ac_compute_driver_uuid(uuid, VK_UUID_SIZE);
+}
+
+static void
+radv_get_device_uuid(struct radeon_info *info, void *uuid)
+{
+ ac_compute_device_uuid(info, uuid, VK_UUID_SIZE);
+}
+
static const VkExtensionProperties instance_extensions[] = {
{
.extensionName = VK_KHR_SURFACE_EXTENSION_NAME,
#ifdef VK_USE_PLATFORM_WAYLAND_KHR
{
.extensionName = VK_KHR_WAYLAND_SURFACE_EXTENSION_NAME,
- .specVersion = 5,
+ .specVersion = 6,
},
#endif
{
.extensionName = VK_KHR_GET_PHYSICAL_DEVICE_PROPERTIES_2_EXTENSION_NAME,
.specVersion = 1,
},
+ {
+ .extensionName = VK_KHR_EXTERNAL_MEMORY_CAPABILITIES_EXTENSION_NAME,
+ .specVersion = 1,
+ },
+ {
+ .extensionName = VK_KHR_EXTERNAL_SEMAPHORE_CAPABILITIES_EXTENSION_NAME,
+ .specVersion = 1,
+ },
};
static const VkExtensionProperties common_device_extensions[] = {
+ {
+ .extensionName = VK_KHR_DESCRIPTOR_UPDATE_TEMPLATE_EXTENSION_NAME,
+ .specVersion = 1,
+ },
+ {
+ .extensionName = VK_KHR_INCREMENTAL_PRESENT_EXTENSION_NAME,
+ .specVersion = 1,
+ },
{
.extensionName = VK_KHR_MAINTENANCE1_EXTENSION_NAME,
.specVersion = 1,
},
+ {
+ .extensionName = VK_KHR_PUSH_DESCRIPTOR_EXTENSION_NAME,
+ .specVersion = 1,
+ },
{
.extensionName = VK_KHR_SAMPLER_MIRROR_CLAMP_TO_EDGE_EXTENSION_NAME,
.specVersion = 1,
.specVersion = 1,
},
{
- .extensionName = VK_NV_DEDICATED_ALLOCATION_EXTENSION_NAME,
+ .extensionName = VK_KHR_GET_MEMORY_REQUIREMENTS_2_EXTENSION_NAME,
+ .specVersion = 1,
+ },
+ {
+ .extensionName = VK_KHR_DEDICATED_ALLOCATION_EXTENSION_NAME,
+ .specVersion = 1,
+ },
+ {
+ .extensionName = VK_KHR_EXTERNAL_MEMORY_EXTENSION_NAME,
+ .specVersion = 1,
+ },
+ {
+ .extensionName = VK_KHR_EXTERNAL_MEMORY_FD_EXTENSION_NAME,
+ .specVersion = 1,
+ },
+ {
+ .extensionName = VK_KHR_STORAGE_BUFFER_STORAGE_CLASS_EXTENSION_NAME,
+ .specVersion = 1,
+ },
+ {
+ .extensionName = VK_KHR_VARIABLE_POINTERS_EXTENSION_NAME,
+ .specVersion = 1,
+ },
+};
+static const VkExtensionProperties ext_sema_device_extensions[] = {
+ {
+ .extensionName = VK_KHR_EXTERNAL_SEMAPHORE_EXTENSION_NAME,
+ .specVersion = 1,
+ },
+ {
+ .extensionName = VK_KHR_EXTERNAL_SEMAPHORE_FD_EXTENSION_NAME,
+ .specVersion = 1,
+ },
+ {
+ .extensionName = VK_KHX_MULTIVIEW_EXTENSION_NAME,
.specVersion = 1,
},
};
return false;
}
+static const char *
+get_chip_name(enum radeon_family family)
+{
+ switch (family) {
+ case CHIP_TAHITI: return "AMD RADV TAHITI";
+ case CHIP_PITCAIRN: return "AMD RADV PITCAIRN";
+ case CHIP_VERDE: return "AMD RADV CAPE VERDE";
+ case CHIP_OLAND: return "AMD RADV OLAND";
+ case CHIP_HAINAN: return "AMD RADV HAINAN";
+ case CHIP_BONAIRE: return "AMD RADV BONAIRE";
+ case CHIP_KAVERI: return "AMD RADV KAVERI";
+ case CHIP_KABINI: return "AMD RADV KABINI";
+ case CHIP_HAWAII: return "AMD RADV HAWAII";
+ case CHIP_MULLINS: return "AMD RADV MULLINS";
+ case CHIP_TONGA: return "AMD RADV TONGA";
+ case CHIP_ICELAND: return "AMD RADV ICELAND";
+ case CHIP_CARRIZO: return "AMD RADV CARRIZO";
+ case CHIP_FIJI: return "AMD RADV FIJI";
+ case CHIP_POLARIS10: return "AMD RADV POLARIS10";
+ case CHIP_POLARIS11: return "AMD RADV POLARIS11";
+ case CHIP_POLARIS12: return "AMD RADV POLARIS12";
+ case CHIP_STONEY: return "AMD RADV STONEY";
+ case CHIP_VEGA10: return "AMD RADV VEGA";
+ case CHIP_RAVEN: return "AMD RADV RAVEN";
+ default: return "AMD RADV unknown";
+ }
+}
+
static VkResult
radv_physical_device_init(struct radv_physical_device *device,
struct radv_instance *instance,
- const char *path)
+ drmDevicePtr drm_device)
{
+ const char *path = drm_device->nodes[DRM_NODE_RENDER];
VkResult result;
drmVersionPtr version;
int fd;
assert(strlen(path) < ARRAY_SIZE(device->path));
strncpy(device->path, path, ARRAY_SIZE(device->path));
- device->ws = radv_amdgpu_winsys_create(fd, instance->debug_flags);
+ device->ws = radv_amdgpu_winsys_create(fd, instance->debug_flags,
+ instance->perftest_flags);
if (!device->ws) {
result = VK_ERROR_INCOMPATIBLE_DRIVER;
goto fail;
goto fail;
}
- if (radv_device_get_cache_uuid(device->rad_info.family, device->uuid)) {
+ if (radv_device_get_cache_uuid(device->rad_info.family, device->cache_uuid)) {
radv_finish_wsi(device);
device->ws->destroy(device->ws);
result = vk_errorf(VK_ERROR_INITIALIZATION_FAILED,
if (result != VK_SUCCESS)
goto fail;
+ if (device->rad_info.has_syncobj) {
+ result = radv_extensions_register(instance,
+ &device->extensions,
+ ext_sema_device_extensions,
+ ARRAY_SIZE(ext_sema_device_extensions));
+ if (result != VK_SUCCESS)
+ goto fail;
+ }
+
fprintf(stderr, "WARNING: radv is not a conformant vulkan implementation, testing use only.\n");
- device->name = device->rad_info.name;
+ device->name = get_chip_name(device->rad_info.family);
+
+ radv_get_driver_uuid(&device->device_uuid);
+ radv_get_device_uuid(&device->rad_info, &device->device_uuid);
+
+ if (device->rad_info.family == CHIP_STONEY ||
+ device->rad_info.chip_class >= GFX9) {
+ device->has_rbplus = true;
+ device->rbplus_allowed = device->rad_info.family == CHIP_STONEY;
+ }
return VK_SUCCESS;
close(device->local_fd);
}
-
static void *
default_alloc_func(void *pUserData, size_t size, size_t align,
VkSystemAllocationScope allocationScope)
{NULL, 0}
};
+static const struct debug_control radv_perftest_options[] = {
+ {"batchchain", RADV_PERFTEST_BATCHCHAIN},
+ {"sisched", RADV_PERFTEST_SISCHED},
+ {NULL, 0}
+};
+
VkResult radv_CreateInstance(
const VkInstanceCreateInfo* pCreateInfo,
const VkAllocationCallbacks* pAllocator,
instance->debug_flags = parse_debug_string(getenv("RADV_DEBUG"),
radv_debug_options);
+ instance->perftest_flags = parse_debug_string(getenv("RADV_PERFTEST"),
+ radv_perftest_options);
+
*pInstance = radv_instance_to_handle(instance);
return VK_SUCCESS;
instance->physicalDeviceCount = 0;
- max_devices = drmGetDevices2(0, devices, sizeof(devices));
+ max_devices = drmGetDevices2(0, devices, ARRAY_SIZE(devices));
if (max_devices < 1)
return VK_ERROR_INCOMPATIBLE_DRIVER;
result = radv_physical_device_init(instance->physicalDevices +
instance->physicalDeviceCount,
instance,
- devices[i]->nodes[DRM_NODE_RENDER]);
+ devices[i]);
if (result == VK_SUCCESS)
++instance->physicalDeviceCount;
else if (result != VK_ERROR_INCOMPATIBLE_DRIVER)
- return result;
+ break;
}
}
+ drmFreeDevices(devices, max_devices);
+
return result;
}
VkPhysicalDevice physicalDevice,
VkPhysicalDeviceFeatures* pFeatures)
{
- // RADV_FROM_HANDLE(radv_physical_device, pdevice, physicalDevice);
-
+ RADV_FROM_HANDLE(radv_physical_device, pdevice, physicalDevice);
+ bool is_gfx9 = pdevice->rad_info.chip_class >= GFX9;
memset(pFeatures, 0, sizeof(*pFeatures));
*pFeatures = (VkPhysicalDeviceFeatures) {
.fullDrawIndexUint32 = true,
.imageCubeArray = true,
.independentBlend = true,
- .geometryShader = true,
- .tessellationShader = false,
- .sampleRateShading = false,
+ .geometryShader = !is_gfx9,
+ .tessellationShader = !is_gfx9,
+ .sampleRateShading = true,
.dualSrcBlend = true,
.logicOp = true,
.multiDrawIndirect = true,
.textureCompressionASTC_LDR = false,
.textureCompressionBC = true,
.occlusionQueryPrecise = true,
- .pipelineStatisticsQuery = false,
+ .pipelineStatisticsQuery = true,
.vertexPipelineStoresAndAtomics = true,
.fragmentStoresAndAtomics = true,
.shaderTessellationAndGeometryPointSize = true,
.shaderClipDistance = true,
.shaderCullDistance = true,
.shaderFloat64 = true,
- .shaderInt64 = false,
+ .shaderInt64 = true,
.shaderInt16 = false,
.sparseBinding = true,
- .variableMultisampleRate = false,
- .inheritedQueries = false,
+ .variableMultisampleRate = true,
+ .inheritedQueries = true,
};
}
VkPhysicalDevice physicalDevice,
VkPhysicalDeviceFeatures2KHR *pFeatures)
{
- return radv_GetPhysicalDeviceFeatures(physicalDevice, &pFeatures->features);
-}
-
-static uint32_t radv_get_driver_version()
-{
- const char *minor_string = strchr(VERSION, '.');
- const char *patch_string = minor_string ? strchr(minor_string + 1, ','): NULL;
- int major = atoi(VERSION);
- int minor = minor_string ? atoi(minor_string + 1) : 0;
- int patch = patch_string ? atoi(patch_string + 1) : 0;
- if (strstr(VERSION, "devel")) {
- if (patch == 0) {
- patch = 99;
- if (minor == 0) {
- minor = 99;
- --major;
- } else
- --minor;
- } else
- --patch;
+ 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 = true;
+ features->variablePointers = false;
+ break;
+ }
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MULTIVIEW_FEATURES_KHX: {
+ VkPhysicalDeviceMultiviewFeaturesKHX *features = (VkPhysicalDeviceMultiviewFeaturesKHX*)ext;
+ features->multiview = true;
+ features->multiviewGeometryShader = true;
+ features->multiviewTessellationShader = true;
+ break;
+ }
+ default:
+ break;
+ }
}
- uint32_t version = VK_MAKE_VERSION(major, minor, patch);
- return version;
+ return radv_GetPhysicalDeviceFeatures(physicalDevice, &pFeatures->features);
}
void radv_GetPhysicalDeviceProperties(
{
RADV_FROM_HANDLE(radv_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 alignement */ +
+ 32 /* storage buffer, 32 due to potential space wasted on alignement */ +
+ 32 /* sampler, largest when combined with image */ +
+ 64 /* sampled image */ +
+ 64 /* storage image */);
+
VkPhysicalDeviceLimits limits = {
.maxImageDimension1D = (1 << 14),
.maxImageDimension2D = (1 << 14),
.bufferImageGranularity = 64, /* A cache line */
.sparseAddressSpaceSize = 0xffffffffu, /* buffer max size */
.maxBoundDescriptorSets = MAX_SETS,
- .maxPerStageDescriptorSamplers = 64,
- .maxPerStageDescriptorUniformBuffers = 64,
- .maxPerStageDescriptorStorageBuffers = 64,
- .maxPerStageDescriptorSampledImages = 64,
- .maxPerStageDescriptorStorageImages = 64,
- .maxPerStageDescriptorInputAttachments = 64,
- .maxPerStageResources = 128,
- .maxDescriptorSetSamplers = 256,
- .maxDescriptorSetUniformBuffers = 256,
- .maxDescriptorSetUniformBuffersDynamic = 256,
- .maxDescriptorSetStorageBuffers = 256,
- .maxDescriptorSetStorageBuffersDynamic = 256,
- .maxDescriptorSetSampledImages = 256,
- .maxDescriptorSetStorageImages = 256,
- .maxDescriptorSetInputAttachments = 256,
+ .maxPerStageDescriptorSamplers = max_descriptor_set_size,
+ .maxPerStageDescriptorUniformBuffers = max_descriptor_set_size,
+ .maxPerStageDescriptorStorageBuffers = max_descriptor_set_size,
+ .maxPerStageDescriptorSampledImages = max_descriptor_set_size,
+ .maxPerStageDescriptorStorageImages = max_descriptor_set_size,
+ .maxPerStageDescriptorInputAttachments = max_descriptor_set_size,
+ .maxPerStageResources = max_descriptor_set_size,
+ .maxDescriptorSetSamplers = max_descriptor_set_size,
+ .maxDescriptorSetUniformBuffers = max_descriptor_set_size,
+ .maxDescriptorSetUniformBuffersDynamic = MAX_DYNAMIC_BUFFERS / 2,
+ .maxDescriptorSetStorageBuffers = max_descriptor_set_size,
+ .maxDescriptorSetStorageBuffersDynamic = MAX_DYNAMIC_BUFFERS / 2,
+ .maxDescriptorSetSampledImages = max_descriptor_set_size,
+ .maxDescriptorSetStorageImages = max_descriptor_set_size,
+ .maxDescriptorSetInputAttachments = max_descriptor_set_size,
.maxVertexInputAttributes = 32,
.maxVertexInputBindings = 32,
.maxVertexInputAttributeOffset = 2047,
.maxVertexInputBindingStride = 2048,
.maxVertexOutputComponents = 128,
- .maxTessellationGenerationLevel = 0,
- .maxTessellationPatchSize = 0,
- .maxTessellationControlPerVertexInputComponents = 0,
- .maxTessellationControlPerVertexOutputComponents = 0,
- .maxTessellationControlPerPatchOutputComponents = 0,
- .maxTessellationControlTotalOutputComponents = 0,
- .maxTessellationEvaluationInputComponents = 0,
- .maxTessellationEvaluationOutputComponents = 0,
- .maxGeometryShaderInvocations = 32,
+ .maxTessellationGenerationLevel = 64,
+ .maxTessellationPatchSize = 32,
+ .maxTessellationControlPerVertexInputComponents = 128,
+ .maxTessellationControlPerVertexOutputComponents = 128,
+ .maxTessellationControlPerPatchOutputComponents = 120,
+ .maxTessellationControlTotalOutputComponents = 4096,
+ .maxTessellationEvaluationInputComponents = 128,
+ .maxTessellationEvaluationOutputComponents = 128,
+ .maxGeometryShaderInvocations = 127,
.maxGeometryInputComponents = 64,
.maxGeometryOutputComponents = 128,
.maxGeometryOutputVertices = 256,
.sampledImageStencilSampleCounts = sample_counts,
.storageImageSampleCounts = VK_SAMPLE_COUNT_1_BIT,
.maxSampleMaskWords = 1,
- .timestampComputeAndGraphics = false,
- .timestampPeriod = 100000.0 / pdevice->rad_info.clock_crystal_freq,
+ .timestampComputeAndGraphics = true,
+ .timestampPeriod = 1000000.0 / pdevice->rad_info.clock_crystal_freq,
.maxClipDistances = 8,
.maxCullDistances = 8,
.maxCombinedClipAndCullDistances = 8,
*pProperties = (VkPhysicalDeviceProperties) {
.apiVersion = VK_MAKE_VERSION(1, 0, 42),
- .driverVersion = radv_get_driver_version(),
+ .driverVersion = vk_get_driver_version(),
.vendorID = 0x1002,
.deviceID = pdevice->rad_info.pci_id,
- .deviceType = VK_PHYSICAL_DEVICE_TYPE_DISCRETE_GPU,
+ .deviceType = pdevice->rad_info.has_dedicated_vram ? VK_PHYSICAL_DEVICE_TYPE_DISCRETE_GPU : VK_PHYSICAL_DEVICE_TYPE_INTEGRATED_GPU,
.limits = limits,
- .sparseProperties = {0}, /* Broadwell doesn't do sparse. */
+ .sparseProperties = {0},
};
strcpy(pProperties->deviceName, pdevice->name);
- memcpy(pProperties->pipelineCacheUUID, pdevice->uuid, VK_UUID_SIZE);
+ memcpy(pProperties->pipelineCacheUUID, pdevice->cache_uuid, VK_UUID_SIZE);
}
void radv_GetPhysicalDeviceProperties2KHR(
VkPhysicalDevice physicalDevice,
VkPhysicalDeviceProperties2KHR *pProperties)
{
- return radv_GetPhysicalDeviceProperties(physicalDevice, &pProperties->properties);
+ RADV_FROM_HANDLE(radv_physical_device, pdevice, physicalDevice);
+ radv_GetPhysicalDeviceProperties(physicalDevice, &pProperties->properties);
+
+ vk_foreach_struct(ext, pProperties->pNext) {
+ switch (ext->sType) {
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PUSH_DESCRIPTOR_PROPERTIES_KHR: {
+ VkPhysicalDevicePushDescriptorPropertiesKHR *properties =
+ (VkPhysicalDevicePushDescriptorPropertiesKHR *) ext;
+ properties->maxPushDescriptors = MAX_PUSH_DESCRIPTORS;
+ break;
+ }
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ID_PROPERTIES_KHR: {
+ VkPhysicalDeviceIDPropertiesKHR *properties = (VkPhysicalDeviceIDPropertiesKHR*)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_KHX: {
+ VkPhysicalDeviceMultiviewPropertiesKHX *properties = (VkPhysicalDeviceMultiviewPropertiesKHX*)ext;
+ properties->maxMultiviewViewCount = MAX_VIEWS;
+ properties->maxMultiviewInstanceIndex = INT_MAX;
+ break;
+ }
+ default:
+ break;
+ }
+ }
}
static void radv_get_physical_device_queue_family_properties(
{
int num_queue_families = 1;
int idx;
- if (pdevice->rad_info.compute_rings > 0 &&
+ if (pdevice->rad_info.num_compute_rings > 0 &&
pdevice->rad_info.chip_class >= CIK &&
!(pdevice->instance->debug_flags & RADV_DEBUG_NO_COMPUTE_QUEUE))
num_queue_families++;
idx++;
}
- if (pdevice->rad_info.compute_rings > 0 &&
+ if (pdevice->rad_info.num_compute_rings > 0 &&
pdevice->rad_info.chip_class >= CIK &&
!(pdevice->instance->debug_flags & RADV_DEBUG_NO_COMPUTE_QUEUE)) {
if (*pCount > idx) {
.queueFlags = VK_QUEUE_COMPUTE_BIT |
VK_QUEUE_TRANSFER_BIT |
VK_QUEUE_SPARSE_BINDING_BIT,
- .queueCount = pdevice->rad_info.compute_rings,
+ .queueCount = pdevice->rad_info.num_compute_rings,
.timestampValidBits = 64,
.minImageTransferGranularity = (VkExtent3D) { 1, 1, 1 },
};
};
STATIC_ASSERT(RADV_MEM_HEAP_COUNT <= VK_MAX_MEMORY_HEAPS);
+ uint64_t visible_vram_size = MIN2(physical_device->rad_info.vram_size,
+ physical_device->rad_info.vram_vis_size);
pMemoryProperties->memoryHeapCount = RADV_MEM_HEAP_COUNT;
pMemoryProperties->memoryHeaps[RADV_MEM_HEAP_VRAM] = (VkMemoryHeap) {
.size = physical_device->rad_info.vram_size -
- physical_device->rad_info.visible_vram_size,
+ visible_vram_size,
.flags = VK_MEMORY_HEAP_DEVICE_LOCAL_BIT,
};
pMemoryProperties->memoryHeaps[RADV_MEM_HEAP_VRAM_CPU_ACCESS] = (VkMemoryHeap) {
- .size = physical_device->rad_info.visible_vram_size,
+ .size = visible_vram_size,
.flags = VK_MEMORY_HEAP_DEVICE_LOCAL_BIT,
};
pMemoryProperties->memoryHeaps[RADV_MEM_HEAP_GTT] = (VkMemoryHeap) {
queue->device->ws->buffer_destroy(queue->esgs_ring_bo);
if (queue->gsvs_ring_bo)
queue->device->ws->buffer_destroy(queue->gsvs_ring_bo);
+ if (queue->tess_factor_ring_bo)
+ queue->device->ws->buffer_destroy(queue->tess_factor_ring_bo);
+ if (queue->tess_offchip_ring_bo)
+ queue->device->ws->buffer_destroy(queue->tess_offchip_ring_bo);
if (queue->compute_scratch_bo)
queue->device->ws->buffer_destroy(queue->compute_scratch_bo);
}
case CHIP_FIJI:
case CHIP_POLARIS10:
case CHIP_POLARIS11:
+ case CHIP_POLARIS12:
+ case CHIP_VEGA10:
+ case CHIP_RAVEN:
device->gs_table_depth = 32;
return;
default:
return vk_error(VK_ERROR_EXTENSION_NOT_PRESENT);
}
+ /* Check enabled features */
+ if (pCreateInfo->pEnabledFeatures) {
+ VkPhysicalDeviceFeatures supported_features;
+ radv_GetPhysicalDeviceFeatures(physicalDevice, &supported_features);
+ VkBool32 *supported_feature = (VkBool32 *)&supported_features;
+ VkBool32 *enabled_feature = (VkBool32 *)pCreateInfo->pEnabledFeatures;
+ unsigned num_features = sizeof(VkPhysicalDeviceFeatures) / sizeof(VkBool32);
+ for (uint32_t i = 0; i < num_features; i++) {
+ if (enabled_feature[i] && !supported_feature[i])
+ return vk_error(VK_ERROR_FEATURE_NOT_PRESENT);
+ }
+ }
+
device = vk_alloc2(&physical_device->instance->alloc, pAllocator,
sizeof(*device), 8,
VK_SYSTEM_ALLOCATION_SCOPE_DEVICE);
else
device->alloc = physical_device->instance->alloc;
+ mtx_init(&device->shader_slab_mutex, mtx_plain);
+ list_inithead(&device->shader_slabs);
+
for (unsigned i = 0; i < pCreateInfo->queueCreateInfoCount; i++) {
const VkDeviceQueueCreateInfo *queue_create = &pCreateInfo->pQueueCreateInfos[i];
uint32_t qfi = queue_create->queueFamilyIndex;
radv_device_init_gs_info(device);
+ device->tess_offchip_block_dw_size =
+ device->physical_device->rad_info.family == CHIP_HAWAII ? 4096 : 8192;
+ device->has_distributed_tess =
+ device->physical_device->rad_info.chip_class >= VI &&
+ device->physical_device->rad_info.max_se >= 2;
+
result = radv_device_init_meta(device);
if (result != VK_SUCCESS)
goto fail;
case RADV_QUEUE_GENERAL:
case RADV_QUEUE_COMPUTE:
si_cs_emit_cache_flush(device->flush_cs[family],
+ false,
device->physical_device->rad_info.chip_class,
+ NULL, 0,
family == RADV_QUEUE_COMPUTE && device->physical_device->rad_info.chip_class >= CIK,
RADV_CMD_FLAG_INV_ICACHE |
RADV_CMD_FLAG_INV_SMEM_L1 |
break;
}
device->ws->cs_finalize(device->flush_cs[family]);
+
+ device->flush_shader_cs[family] = device->ws->cs_create(device->ws, family);
+ switch (family) {
+ case RADV_QUEUE_GENERAL:
+ case RADV_QUEUE_COMPUTE:
+ si_cs_emit_cache_flush(device->flush_shader_cs[family],
+ false,
+ device->physical_device->rad_info.chip_class,
+ NULL, 0,
+ family == RADV_QUEUE_COMPUTE && device->physical_device->rad_info.chip_class >= CIK,
+ family == RADV_QUEUE_COMPUTE ? RADV_CMD_FLAG_CS_PARTIAL_FLUSH : (RADV_CMD_FLAG_CS_PARTIAL_FLUSH | RADV_CMD_FLAG_PS_PARTIAL_FLUSH) |
+ RADV_CMD_FLAG_INV_ICACHE |
+ RADV_CMD_FLAG_INV_SMEM_L1 |
+ RADV_CMD_FLAG_INV_VMEM_L1 |
+ RADV_CMD_FLAG_INV_GLOBAL_L2);
+ break;
+ }
+ device->ws->cs_finalize(device->flush_shader_cs[family]);
}
if (getenv("RADV_TRACE_FILE")) {
- device->trace_bo = device->ws->buffer_create(device->ws, 4096, 8,
- RADEON_DOMAIN_VRAM, RADEON_FLAG_CPU_ACCESS);
- if (!device->trace_bo)
- goto fail;
-
- device->trace_id_ptr = device->ws->buffer_map(device->trace_bo);
- if (!device->trace_id_ptr)
+ if (!radv_init_trace(device))
goto fail;
}
device->ws->cs_destroy(device->empty_cs[i]);
if (device->flush_cs[i])
device->ws->cs_destroy(device->flush_cs[i]);
+ if (device->flush_shader_cs[i])
+ device->ws->cs_destroy(device->flush_shader_cs[i]);
}
radv_device_finish_meta(device);
VkPipelineCache pc = radv_pipeline_cache_to_handle(device->mem_cache);
radv_DestroyPipelineCache(radv_device_to_handle(device), pc, NULL);
+ radv_destroy_shader_slabs(device);
+
vk_free(&device->alloc, device);
}
*pQueue = radv_queue_to_handle(&device->queues[queueFamilyIndex][queueIndex]);
}
-static void radv_dump_trace(struct radv_device *device,
- struct radeon_winsys_cs *cs)
-{
- const char *filename = getenv("RADV_TRACE_FILE");
- FILE *f = fopen(filename, "w");
- if (!f) {
- fprintf(stderr, "Failed to write trace dump to %s\n", filename);
- return;
- }
-
- fprintf(f, "Trace ID: %x\n", *device->trace_id_ptr);
- device->ws->cs_dump(cs, f, *device->trace_id_ptr);
- fclose(f);
-}
-
static void
-fill_geom_rings(struct radv_queue *queue,
- uint32_t *map,
- uint32_t esgs_ring_size,
- struct radeon_winsys_bo *esgs_ring_bo,
- uint32_t gsvs_ring_size,
- struct radeon_winsys_bo *gsvs_ring_bo)
+fill_geom_tess_rings(struct radv_queue *queue,
+ uint32_t *map,
+ bool add_sample_positions,
+ uint32_t esgs_ring_size,
+ struct radeon_winsys_bo *esgs_ring_bo,
+ uint32_t gsvs_ring_size,
+ struct radeon_winsys_bo *gsvs_ring_bo,
+ uint32_t tess_factor_ring_size,
+ struct radeon_winsys_bo *tess_factor_ring_bo,
+ uint32_t tess_offchip_ring_size,
+ struct radeon_winsys_bo *tess_offchip_ring_bo)
{
uint64_t esgs_va = 0, gsvs_va = 0;
+ uint64_t tess_factor_va = 0, tess_offchip_va = 0;
uint32_t *desc = &map[4];
if (esgs_ring_bo)
esgs_va = queue->device->ws->buffer_get_va(esgs_ring_bo);
if (gsvs_ring_bo)
gsvs_va = queue->device->ws->buffer_get_va(gsvs_ring_bo);
+ if (tess_factor_ring_bo)
+ tess_factor_va = queue->device->ws->buffer_get_va(tess_factor_ring_bo);
+ if (tess_offchip_ring_bo)
+ tess_offchip_va = queue->device->ws->buffer_get_va(tess_offchip_ring_bo);
/* stride 0, num records - size, add tid, swizzle, elsize4,
index stride 64 */
S_008F0C_ELEMENT_SIZE(1) |
S_008F0C_INDEX_STRIDE(1) |
S_008F0C_ADD_TID_ENABLE(true);
+ desc += 4;
+
+ desc[0] = tess_factor_va;
+ desc[1] = S_008F04_BASE_ADDRESS_HI(tess_factor_va >> 32) |
+ S_008F04_STRIDE(0) |
+ S_008F04_SWIZZLE_ENABLE(false);
+ desc[2] = tess_factor_ring_size;
+ desc[3] = S_008F0C_DST_SEL_X(V_008F0C_SQ_SEL_X) |
+ S_008F0C_DST_SEL_Y(V_008F0C_SQ_SEL_Y) |
+ S_008F0C_DST_SEL_Z(V_008F0C_SQ_SEL_Z) |
+ S_008F0C_DST_SEL_W(V_008F0C_SQ_SEL_W) |
+ S_008F0C_NUM_FORMAT(V_008F0C_BUF_NUM_FORMAT_FLOAT) |
+ S_008F0C_DATA_FORMAT(V_008F0C_BUF_DATA_FORMAT_32) |
+ S_008F0C_ELEMENT_SIZE(0) |
+ S_008F0C_INDEX_STRIDE(0) |
+ S_008F0C_ADD_TID_ENABLE(false);
+ desc += 4;
+
+ desc[0] = tess_offchip_va;
+ desc[1] = S_008F04_BASE_ADDRESS_HI(tess_offchip_va >> 32) |
+ S_008F04_STRIDE(0) |
+ S_008F04_SWIZZLE_ENABLE(false);
+ desc[2] = tess_offchip_ring_size;
+ desc[3] = S_008F0C_DST_SEL_X(V_008F0C_SQ_SEL_X) |
+ S_008F0C_DST_SEL_Y(V_008F0C_SQ_SEL_Y) |
+ S_008F0C_DST_SEL_Z(V_008F0C_SQ_SEL_Z) |
+ S_008F0C_DST_SEL_W(V_008F0C_SQ_SEL_W) |
+ S_008F0C_NUM_FORMAT(V_008F0C_BUF_NUM_FORMAT_FLOAT) |
+ S_008F0C_DATA_FORMAT(V_008F0C_BUF_DATA_FORMAT_32) |
+ S_008F0C_ELEMENT_SIZE(0) |
+ S_008F0C_INDEX_STRIDE(0) |
+ S_008F0C_ADD_TID_ENABLE(false);
+ desc += 4;
+
+ /* add sample positions after all rings */
+ memcpy(desc, queue->device->sample_locations_1x, 8);
+ desc += 2;
+ memcpy(desc, queue->device->sample_locations_2x, 16);
+ desc += 4;
+ memcpy(desc, queue->device->sample_locations_4x, 32);
+ desc += 8;
+ memcpy(desc, queue->device->sample_locations_8x, 64);
+ desc += 16;
+ memcpy(desc, queue->device->sample_locations_16x, 128);
+}
+
+static unsigned
+radv_get_hs_offchip_param(struct radv_device *device, uint32_t *max_offchip_buffers_p)
+{
+ bool double_offchip_buffers = device->physical_device->rad_info.chip_class >= CIK &&
+ device->physical_device->rad_info.family != CHIP_CARRIZO &&
+ device->physical_device->rad_info.family != CHIP_STONEY;
+ unsigned max_offchip_buffers_per_se = double_offchip_buffers ? 128 : 64;
+ unsigned max_offchip_buffers = max_offchip_buffers_per_se *
+ device->physical_device->rad_info.max_se;
+ unsigned offchip_granularity;
+ unsigned hs_offchip_param;
+ switch (device->tess_offchip_block_dw_size) {
+ default:
+ assert(0);
+ /* fall through */
+ case 8192:
+ offchip_granularity = V_03093C_X_8K_DWORDS;
+ break;
+ case 4096:
+ offchip_granularity = V_03093C_X_4K_DWORDS;
+ break;
+ }
+
+ switch (device->physical_device->rad_info.chip_class) {
+ case SI:
+ max_offchip_buffers = MIN2(max_offchip_buffers, 126);
+ break;
+ case CIK:
+ case VI:
+ case GFX9:
+ default:
+ max_offchip_buffers = MIN2(max_offchip_buffers, 508);
+ break;
+ }
+
+ *max_offchip_buffers_p = max_offchip_buffers;
+ if (device->physical_device->rad_info.chip_class >= CIK) {
+ if (device->physical_device->rad_info.chip_class >= VI)
+ --max_offchip_buffers;
+ hs_offchip_param =
+ S_03093C_OFFCHIP_BUFFERING(max_offchip_buffers) |
+ S_03093C_OFFCHIP_GRANULARITY(offchip_granularity);
+ } else {
+ hs_offchip_param =
+ S_0089B0_OFFCHIP_BUFFERING(max_offchip_buffers);
+ }
+ return hs_offchip_param;
}
static VkResult
uint32_t compute_scratch_size,
uint32_t esgs_ring_size,
uint32_t gsvs_ring_size,
+ bool needs_tess_rings,
+ bool needs_sample_positions,
struct radeon_winsys_cs **initial_preamble_cs,
struct radeon_winsys_cs **continue_preamble_cs)
{
struct radeon_winsys_bo *compute_scratch_bo = NULL;
struct radeon_winsys_bo *esgs_ring_bo = NULL;
struct radeon_winsys_bo *gsvs_ring_bo = NULL;
+ struct radeon_winsys_bo *tess_factor_ring_bo = NULL;
+ struct radeon_winsys_bo *tess_offchip_ring_bo = NULL;
struct radeon_winsys_cs *dest_cs[2] = {0};
+ bool add_tess_rings = false, add_sample_positions = false;
+ unsigned tess_factor_ring_size = 0, tess_offchip_ring_size = 0;
+ unsigned max_offchip_buffers;
+ unsigned hs_offchip_param = 0;
+ if (!queue->has_tess_rings) {
+ if (needs_tess_rings)
+ add_tess_rings = true;
+ }
+ if (!queue->has_sample_positions) {
+ if (needs_sample_positions)
+ add_sample_positions = true;
+ }
+ tess_factor_ring_size = 32768 * queue->device->physical_device->rad_info.max_se;
+ hs_offchip_param = radv_get_hs_offchip_param(queue->device,
+ &max_offchip_buffers);
+ tess_offchip_ring_size = max_offchip_buffers *
+ queue->device->tess_offchip_block_dw_size * 4;
if (scratch_size <= queue->scratch_size &&
compute_scratch_size <= queue->compute_scratch_size &&
esgs_ring_size <= queue->esgs_ring_size &&
gsvs_ring_size <= queue->gsvs_ring_size &&
+ !add_tess_rings && !add_sample_positions &&
queue->initial_preamble_cs) {
*initial_preamble_cs = queue->initial_preamble_cs;
*continue_preamble_cs = queue->continue_preamble_cs;
gsvs_ring_size = queue->gsvs_ring_size;
}
+ if (add_tess_rings) {
+ tess_factor_ring_bo = queue->device->ws->buffer_create(queue->device->ws,
+ tess_factor_ring_size,
+ 256,
+ RADEON_DOMAIN_VRAM,
+ RADEON_FLAG_NO_CPU_ACCESS);
+ if (!tess_factor_ring_bo)
+ goto fail;
+ tess_offchip_ring_bo = queue->device->ws->buffer_create(queue->device->ws,
+ tess_offchip_ring_size,
+ 256,
+ RADEON_DOMAIN_VRAM,
+ RADEON_FLAG_NO_CPU_ACCESS);
+ if (!tess_offchip_ring_bo)
+ goto fail;
+ } else {
+ tess_factor_ring_bo = queue->tess_factor_ring_bo;
+ tess_offchip_ring_bo = queue->tess_offchip_ring_bo;
+ }
+
if (scratch_bo != queue->scratch_bo ||
esgs_ring_bo != queue->esgs_ring_bo ||
- gsvs_ring_bo != queue->gsvs_ring_bo) {
+ gsvs_ring_bo != queue->gsvs_ring_bo ||
+ tess_factor_ring_bo != queue->tess_factor_ring_bo ||
+ tess_offchip_ring_bo != queue->tess_offchip_ring_bo || add_sample_positions) {
uint32_t size = 0;
- if (gsvs_ring_bo || esgs_ring_bo)
- size = 80; /* 2 dword + 2 padding + 4 dword * 4 */
+ if (gsvs_ring_bo || esgs_ring_bo ||
+ tess_factor_ring_bo || tess_offchip_ring_bo || add_sample_positions) {
+ size = 112; /* 2 dword + 2 padding + 4 dword * 6 */
+ if (add_sample_positions)
+ size += 256; /* 32+16+8+4+2+1 samples * 4 * 2 = 248 bytes. */
+ }
else if (scratch_bo)
size = 8; /* 2 dword */
if (gsvs_ring_bo)
queue->device->ws->cs_add_buffer(cs, gsvs_ring_bo, 8);
+ if (tess_factor_ring_bo)
+ queue->device->ws->cs_add_buffer(cs, tess_factor_ring_bo, 8);
+
+ if (tess_offchip_ring_bo)
+ queue->device->ws->cs_add_buffer(cs, tess_offchip_ring_bo, 8);
+
if (descriptor_bo)
queue->device->ws->cs_add_buffer(cs, descriptor_bo, 8);
map[1] = rsrc1;
}
- if (esgs_ring_bo || gsvs_ring_bo)
- fill_geom_rings(queue, map, esgs_ring_size, esgs_ring_bo, gsvs_ring_size, gsvs_ring_bo);
+ if (esgs_ring_bo || gsvs_ring_bo || tess_factor_ring_bo || tess_offchip_ring_bo ||
+ add_sample_positions)
+ fill_geom_tess_rings(queue, map, add_sample_positions,
+ esgs_ring_size, esgs_ring_bo,
+ gsvs_ring_size, gsvs_ring_bo,
+ tess_factor_ring_size, tess_factor_ring_bo,
+ tess_offchip_ring_size, tess_offchip_ring_bo);
queue->device->ws->buffer_unmap(descriptor_bo);
}
- if (esgs_ring_bo || gsvs_ring_bo) {
+ if (esgs_ring_bo || gsvs_ring_bo || tess_factor_ring_bo || tess_offchip_ring_bo) {
radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
radeon_emit(cs, EVENT_TYPE(V_028A90_VS_PARTIAL_FLUSH) | EVENT_INDEX(4));
radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
radeon_emit(cs, EVENT_TYPE(V_028A90_VGT_FLUSH) | EVENT_INDEX(0));
+ }
+ if (esgs_ring_bo || gsvs_ring_bo) {
if (queue->device->physical_device->rad_info.chip_class >= CIK) {
radeon_set_uconfig_reg_seq(cs, R_030900_VGT_ESGS_RING_SIZE, 2);
radeon_emit(cs, esgs_ring_size >> 8);
}
}
+ if (tess_factor_ring_bo) {
+ uint64_t tf_va = queue->device->ws->buffer_get_va(tess_factor_ring_bo);
+ if (queue->device->physical_device->rad_info.chip_class >= CIK) {
+ radeon_set_uconfig_reg(cs, R_030938_VGT_TF_RING_SIZE,
+ S_030938_SIZE(tess_factor_ring_size / 4));
+ radeon_set_uconfig_reg(cs, R_030940_VGT_TF_MEMORY_BASE,
+ tf_va >> 8);
+ if (queue->device->physical_device->rad_info.chip_class >= GFX9) {
+ radeon_set_uconfig_reg(cs, R_030944_VGT_TF_MEMORY_BASE_HI,
+ tf_va >> 40);
+ }
+ radeon_set_uconfig_reg(cs, R_03093C_VGT_HS_OFFCHIP_PARAM, hs_offchip_param);
+ } else {
+ radeon_set_config_reg(cs, R_008988_VGT_TF_RING_SIZE,
+ S_008988_SIZE(tess_factor_ring_size / 4));
+ radeon_set_config_reg(cs, R_0089B8_VGT_TF_MEMORY_BASE,
+ tf_va >> 8);
+ radeon_set_config_reg(cs, R_0089B0_VGT_HS_OFFCHIP_PARAM,
+ hs_offchip_param);
+ }
+ }
+
if (descriptor_bo) {
uint32_t regs[] = {R_00B030_SPI_SHADER_USER_DATA_PS_0,
R_00B130_SPI_SHADER_USER_DATA_VS_0,
if (!i) {
si_cs_emit_cache_flush(cs,
+ false,
queue->device->physical_device->rad_info.chip_class,
+ NULL, 0,
queue->queue_family_index == RING_COMPUTE &&
queue->device->physical_device->rad_info.chip_class >= CIK,
RADV_CMD_FLAG_INV_ICACHE |
queue->gsvs_ring_size = gsvs_ring_size;
}
+ if (tess_factor_ring_bo != queue->tess_factor_ring_bo) {
+ queue->tess_factor_ring_bo = tess_factor_ring_bo;
+ }
+
+ if (tess_offchip_ring_bo != queue->tess_offchip_ring_bo) {
+ queue->tess_offchip_ring_bo = tess_offchip_ring_bo;
+ queue->has_tess_rings = true;
+ }
+
if (descriptor_bo != queue->descriptor_bo) {
if (queue->descriptor_bo)
queue->device->ws->buffer_destroy(queue->descriptor_bo);
queue->descriptor_bo = descriptor_bo;
}
+ if (add_sample_positions)
+ queue->has_sample_positions = true;
+
*initial_preamble_cs = queue->initial_preamble_cs;
*continue_preamble_cs = queue->continue_preamble_cs;
if (!scratch_size && !compute_scratch_size && !esgs_ring_size && !gsvs_ring_size)
queue->device->ws->buffer_destroy(esgs_ring_bo);
if (gsvs_ring_bo && gsvs_ring_bo != queue->gsvs_ring_bo)
queue->device->ws->buffer_destroy(gsvs_ring_bo);
+ if (tess_factor_ring_bo && tess_factor_ring_bo != queue->tess_factor_ring_bo)
+ queue->device->ws->buffer_destroy(tess_factor_ring_bo);
+ if (tess_offchip_ring_bo && tess_offchip_ring_bo != queue->tess_offchip_ring_bo)
+ queue->device->ws->buffer_destroy(tess_offchip_ring_bo);
return VK_ERROR_OUT_OF_DEVICE_MEMORY;
}
+static VkResult radv_alloc_sem_counts(struct radv_winsys_sem_counts *counts,
+ int num_sems,
+ const VkSemaphore *sems,
+ bool reset_temp)
+{
+ int syncobj_idx = 0, sem_idx = 0;
+
+ if (num_sems == 0)
+ return VK_SUCCESS;
+ for (uint32_t i = 0; i < num_sems; i++) {
+ RADV_FROM_HANDLE(radv_semaphore, sem, sems[i]);
+
+ if (sem->temp_syncobj || sem->syncobj)
+ counts->syncobj_count++;
+ else
+ counts->sem_count++;
+ }
+
+ if (counts->syncobj_count) {
+ counts->syncobj = (uint32_t *)malloc(sizeof(uint32_t) * counts->syncobj_count);
+ if (!counts->syncobj)
+ return VK_ERROR_OUT_OF_HOST_MEMORY;
+ }
+
+ if (counts->sem_count) {
+ counts->sem = (struct radeon_winsys_sem **)malloc(sizeof(struct radeon_winsys_sem *) * counts->sem_count);
+ if (!counts->sem) {
+ free(counts->syncobj);
+ return VK_ERROR_OUT_OF_HOST_MEMORY;
+ }
+ }
+
+ for (uint32_t i = 0; i < num_sems; i++) {
+ RADV_FROM_HANDLE(radv_semaphore, sem, sems[i]);
+
+ if (sem->temp_syncobj) {
+ counts->syncobj[syncobj_idx++] = sem->temp_syncobj;
+ if (reset_temp) {
+ /* after we wait on a temp import - drop it */
+ sem->temp_syncobj = 0;
+ }
+ }
+ else if (sem->syncobj)
+ counts->syncobj[syncobj_idx++] = sem->syncobj;
+ else {
+ assert(sem->sem);
+ counts->sem[sem_idx++] = sem->sem;
+ }
+ }
+
+ return VK_SUCCESS;
+}
+
+void radv_free_sem_info(struct radv_winsys_sem_info *sem_info)
+{
+ free(sem_info->wait.syncobj);
+ free(sem_info->wait.sem);
+ free(sem_info->signal.syncobj);
+ free(sem_info->signal.sem);
+}
+
+VkResult radv_alloc_sem_info(struct radv_winsys_sem_info *sem_info,
+ int num_wait_sems,
+ const VkSemaphore *wait_sems,
+ int num_signal_sems,
+ const VkSemaphore *signal_sems)
+{
+ VkResult ret;
+ memset(sem_info, 0, sizeof(*sem_info));
+
+ ret = radv_alloc_sem_counts(&sem_info->wait, num_wait_sems, wait_sems, true);
+ if (ret)
+ return ret;
+ ret = radv_alloc_sem_counts(&sem_info->signal, num_signal_sems, signal_sems, false);
+ if (ret)
+ radv_free_sem_info(sem_info);
+
+ /* caller can override these */
+ sem_info->cs_emit_wait = true;
+ sem_info->cs_emit_signal = true;
+ return ret;
+}
+
VkResult radv_QueueSubmit(
VkQueue _queue,
uint32_t submitCount,
struct radeon_winsys_cs *initial_preamble_cs = NULL, *continue_preamble_cs = NULL;
VkResult result;
bool fence_emitted = false;
+ bool tess_rings_needed = false;
+ bool sample_positions_needed = false;
/* Do this first so failing to allocate scratch buffers can't result in
* partially executed submissions. */
cmd_buffer->compute_scratch_size_needed);
esgs_ring_size = MAX2(esgs_ring_size, cmd_buffer->esgs_ring_size_needed);
gsvs_ring_size = MAX2(gsvs_ring_size, cmd_buffer->gsvs_ring_size_needed);
+ tess_rings_needed |= cmd_buffer->tess_rings_needed;
+ sample_positions_needed |= cmd_buffer->sample_positions_needed;
}
}
result = radv_get_preamble_cs(queue, scratch_size, compute_scratch_size,
- esgs_ring_size, gsvs_ring_size,
+ esgs_ring_size, gsvs_ring_size, tess_rings_needed,
+ sample_positions_needed,
&initial_preamble_cs, &continue_preamble_cs);
if (result != VK_SUCCESS)
return result;
for (uint32_t i = 0; i < submitCount; i++) {
struct radeon_winsys_cs **cs_array;
- bool do_flush = !i;
+ bool do_flush = !i || pSubmits[i].pWaitDstStageMask;
bool can_patch = !do_flush;
uint32_t advance;
+ struct radv_winsys_sem_info sem_info;
+
+ result = radv_alloc_sem_info(&sem_info,
+ pSubmits[i].waitSemaphoreCount,
+ pSubmits[i].pWaitSemaphores,
+ pSubmits[i].signalSemaphoreCount,
+ pSubmits[i].pSignalSemaphores);
+ if (result != VK_SUCCESS)
+ return result;
if (!pSubmits[i].commandBufferCount) {
if (pSubmits[i].waitSemaphoreCount || pSubmits[i].signalSemaphoreCount) {
ret = queue->device->ws->cs_submit(ctx, queue->queue_idx,
&queue->device->empty_cs[queue->queue_family_index],
1, NULL, NULL,
- (struct radeon_winsys_sem **)pSubmits[i].pWaitSemaphores,
- pSubmits[i].waitSemaphoreCount,
- (struct radeon_winsys_sem **)pSubmits[i].pSignalSemaphores,
- pSubmits[i].signalSemaphoreCount,
+ &sem_info,
false, base_fence);
if (ret) {
radv_loge("failed to submit CS %d\n", i);
}
fence_emitted = true;
}
+ radv_free_sem_info(&sem_info);
continue;
}
(pSubmits[i].commandBufferCount + do_flush));
if(do_flush)
- cs_array[0] = queue->device->flush_cs[queue->queue_family_index];
+ cs_array[0] = pSubmits[i].waitSemaphoreCount ?
+ queue->device->flush_shader_cs[queue->queue_family_index] :
+ queue->device->flush_cs[queue->queue_family_index];
for (uint32_t j = 0; j < pSubmits[i].commandBufferCount; j++) {
RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer,
for (uint32_t j = 0; j < pSubmits[i].commandBufferCount + do_flush; j += advance) {
advance = MIN2(max_cs_submission,
pSubmits[i].commandBufferCount + do_flush - j);
- bool b = j == 0;
- bool e = j + advance == pSubmits[i].commandBufferCount + do_flush;
if (queue->device->trace_bo)
*queue->device->trace_id_ptr = 0;
+ sem_info.cs_emit_wait = j == 0;
+ sem_info.cs_emit_signal = j + advance == pSubmits[i].commandBufferCount + do_flush;
+
ret = queue->device->ws->cs_submit(ctx, queue->queue_idx, cs_array + j,
advance, initial_preamble_cs, continue_preamble_cs,
- (struct radeon_winsys_sem **)pSubmits[i].pWaitSemaphores,
- b ? pSubmits[i].waitSemaphoreCount : 0,
- (struct radeon_winsys_sem **)pSubmits[i].pSignalSemaphores,
- e ? pSubmits[i].signalSemaphoreCount : 0,
+ &sem_info,
can_patch, base_fence);
if (ret) {
}
}
}
+
+ radv_free_sem_info(&sem_info);
free(cs_array);
}
if (fence) {
- if (!fence_emitted)
+ if (!fence_emitted) {
+ struct radv_winsys_sem_info sem_info = {0};
ret = queue->device->ws->cs_submit(ctx, queue->queue_idx,
&queue->device->empty_cs[queue->queue_family_index],
- 1, NULL, NULL, NULL, 0, NULL, 0,
+ 1, NULL, NULL, &sem_info,
false, base_fence);
-
+ }
fence->submitted = true;
}
VkResult result;
enum radeon_bo_domain domain;
uint32_t flags = 0;
- const VkDedicatedAllocationMemoryAllocateInfoNV *dedicate_info = NULL;
+
assert(pAllocateInfo->sType == VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO);
if (pAllocateInfo->allocationSize == 0) {
return VK_SUCCESS;
}
- vk_foreach_struct(ext, pAllocateInfo->pNext) {
- switch (ext->sType) {
- case VK_STRUCTURE_TYPE_DEDICATED_ALLOCATION_MEMORY_ALLOCATE_INFO_NV:
- dedicate_info = (const VkDedicatedAllocationMemoryAllocateInfoNV *)ext;
- break;
- default:
- break;
- }
- }
+ const VkImportMemoryFdInfoKHR *import_info =
+ vk_find_struct_const(pAllocateInfo->pNext, IMPORT_MEMORY_FD_INFO_KHR);
+ const VkMemoryDedicatedAllocateInfoKHR *dedicate_info =
+ vk_find_struct_const(pAllocateInfo->pNext, MEMORY_DEDICATED_ALLOCATE_INFO_KHR);
mem = vk_alloc2(&device->alloc, pAllocator, sizeof(*mem), 8,
VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
mem->buffer = NULL;
}
+ if (import_info) {
+ assert(import_info->handleType ==
+ VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT_KHR);
+ mem->bo = device->ws->buffer_from_fd(device->ws, import_info->fd,
+ NULL, NULL);
+ if (!mem->bo) {
+ result = VK_ERROR_INVALID_EXTERNAL_HANDLE_KHR;
+ goto fail;
+ } else {
+ close(import_info->fd);
+ goto out_success;
+ }
+ }
+
uint64_t alloc_size = align_u64(pAllocateInfo->allocationSize, 4096);
if (pAllocateInfo->memoryTypeIndex == RADV_MEM_TYPE_GTT_WRITE_COMBINE ||
pAllocateInfo->memoryTypeIndex == RADV_MEM_TYPE_GTT_CACHED)
if (pAllocateInfo->memoryTypeIndex == RADV_MEM_TYPE_GTT_WRITE_COMBINE)
flags |= RADEON_FLAG_GTT_WC;
- mem->bo = device->ws->buffer_create(device->ws, alloc_size, 65536,
+ mem->bo = device->ws->buffer_create(device->ws, alloc_size, device->physical_device->rad_info.max_alignment,
domain, flags);
if (!mem->bo) {
goto fail;
}
mem->type_index = pAllocateInfo->memoryTypeIndex;
-
+out_success:
*pMem = radv_device_memory_to_handle(mem);
return VK_SUCCESS;
pMemoryRequirements->size = align64(buffer->size, pMemoryRequirements->alignment);
}
+void radv_GetBufferMemoryRequirements2KHR(
+ VkDevice device,
+ const VkBufferMemoryRequirementsInfo2KHR* pInfo,
+ VkMemoryRequirements2KHR* pMemoryRequirements)
+{
+ radv_GetBufferMemoryRequirements(device, pInfo->buffer,
+ &pMemoryRequirements->memoryRequirements);
+
+ vk_foreach_struct(ext, pMemoryRequirements->pNext) {
+ switch (ext->sType) {
+ case VK_STRUCTURE_TYPE_MEMORY_DEDICATED_REQUIREMENTS_KHR: {
+ VkMemoryDedicatedRequirementsKHR *req =
+ (VkMemoryDedicatedRequirementsKHR *) ext;
+ req->requiresDedicatedAllocation = false;
+ req->prefersDedicatedAllocation = req->requiresDedicatedAllocation;
+ break;
+ }
+ default:
+ break;
+ }
+ }
+}
+
void radv_GetImageMemoryRequirements(
VkDevice device,
VkImage _image,
pMemoryRequirements->alignment = image->alignment;
}
+void radv_GetImageMemoryRequirements2KHR(
+ VkDevice device,
+ const VkImageMemoryRequirementsInfo2KHR* pInfo,
+ VkMemoryRequirements2KHR* pMemoryRequirements)
+{
+ radv_GetImageMemoryRequirements(device, pInfo->image,
+ &pMemoryRequirements->memoryRequirements);
+
+ RADV_FROM_HANDLE(radv_image, image, pInfo->image);
+
+ vk_foreach_struct(ext, pMemoryRequirements->pNext) {
+ switch (ext->sType) {
+ case VK_STRUCTURE_TYPE_MEMORY_DEDICATED_REQUIREMENTS_KHR: {
+ VkMemoryDedicatedRequirementsKHR *req =
+ (VkMemoryDedicatedRequirementsKHR *) ext;
+ req->requiresDedicatedAllocation = image->shareable;
+ req->prefersDedicatedAllocation = req->requiresDedicatedAllocation;
+ break;
+ }
+ default:
+ break;
+ }
+ }
+}
+
void radv_GetImageSparseMemoryRequirements(
VkDevice device,
VkImage image,
stub();
}
+void radv_GetImageSparseMemoryRequirements2KHR(
+ VkDevice device,
+ const VkImageSparseMemoryRequirementsInfo2KHR* pInfo,
+ uint32_t* pSparseMemoryRequirementCount,
+ VkSparseImageMemoryRequirements2KHR* pSparseMemoryRequirements)
+{
+ stub();
+}
+
void radv_GetDeviceMemoryCommitment(
VkDevice device,
VkDeviceMemory memory,
bool fence_emitted = false;
for (uint32_t i = 0; i < bindInfoCount; ++i) {
+ struct radv_winsys_sem_info sem_info;
for (uint32_t j = 0; j < pBindInfo[i].bufferBindCount; ++j) {
radv_sparse_buffer_bind_memory(queue->device,
pBindInfo[i].pBufferBinds + j);
pBindInfo[i].pImageOpaqueBinds + j);
}
+ VkResult result;
+ result = radv_alloc_sem_info(&sem_info,
+ pBindInfo[i].waitSemaphoreCount,
+ pBindInfo[i].pWaitSemaphores,
+ pBindInfo[i].signalSemaphoreCount,
+ pBindInfo[i].pSignalSemaphores);
+ if (result != VK_SUCCESS)
+ return result;
+
if (pBindInfo[i].waitSemaphoreCount || pBindInfo[i].signalSemaphoreCount) {
queue->device->ws->cs_submit(queue->hw_ctx, queue->queue_idx,
&queue->device->empty_cs[queue->queue_family_index],
1, NULL, NULL,
- (struct radeon_winsys_sem **)pBindInfo[i].pWaitSemaphores,
- pBindInfo[i].waitSemaphoreCount,
- (struct radeon_winsys_sem **)pBindInfo[i].pSignalSemaphores,
- pBindInfo[i].signalSemaphoreCount,
+ &sem_info,
false, base_fence);
fence_emitted = true;
if (fence)
fence->submitted = true;
}
+
+ radv_free_sem_info(&sem_info);
+
}
if (fence && !fence_emitted) {
VkSemaphore* pSemaphore)
{
RADV_FROM_HANDLE(radv_device, device, _device);
- struct radeon_winsys_sem *sem;
+ const VkExportSemaphoreCreateInfoKHR *export =
+ vk_find_struct_const(pCreateInfo->pNext, EXPORT_SEMAPHORE_CREATE_INFO_KHR);
+ VkExternalSemaphoreHandleTypeFlagsKHR handleTypes =
+ export ? export->handleTypes : 0;
- sem = device->ws->create_sem(device->ws);
+ struct radv_semaphore *sem = vk_alloc2(&device->alloc, pAllocator,
+ sizeof(*sem), 8,
+ VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
if (!sem)
return VK_ERROR_OUT_OF_HOST_MEMORY;
- *pSemaphore = radeon_winsys_sem_to_handle(sem);
+ sem->temp_syncobj = 0;
+ /* create a syncobject if we are going to export this semaphore */
+ if (handleTypes) {
+ assert (device->physical_device->rad_info.has_syncobj);
+ assert (handleTypes == VK_EXTERNAL_FENCE_HANDLE_TYPE_OPAQUE_FD_BIT_KHR);
+ int ret = device->ws->create_syncobj(device->ws, &sem->syncobj);
+ if (ret) {
+ vk_free2(&device->alloc, pAllocator, sem);
+ return VK_ERROR_OUT_OF_HOST_MEMORY;
+ }
+ sem->sem = NULL;
+ } else {
+ sem->sem = device->ws->create_sem(device->ws);
+ if (!sem->sem) {
+ vk_free2(&device->alloc, pAllocator, sem);
+ return VK_ERROR_OUT_OF_HOST_MEMORY;
+ }
+ sem->syncobj = 0;
+ }
+
+ *pSemaphore = radv_semaphore_to_handle(sem);
return VK_SUCCESS;
}
const VkAllocationCallbacks* pAllocator)
{
RADV_FROM_HANDLE(radv_device, device, _device);
- RADV_FROM_HANDLE(radeon_winsys_sem, sem, _semaphore);
+ RADV_FROM_HANDLE(radv_semaphore, sem, _semaphore);
if (!_semaphore)
return;
- device->ws->destroy_sem(sem);
+ if (sem->syncobj)
+ device->ws->destroy_syncobj(device->ws, sem->syncobj);
+ else
+ device->ws->destroy_sem(sem->sem);
+ vk_free2(&device->alloc, pAllocator, sem);
}
VkResult radv_CreateEvent(
si_tile_mode_index(const struct radv_image *image, unsigned level, bool stencil)
{
if (stencil)
- return image->surface.stencil_tiling_index[level];
+ return image->surface.u.legacy.stencil_tiling_index[level];
else
- return image->surface.tiling_index[level];
+ return image->surface.u.legacy.tiling_index[level];
}
static uint32_t radv_surface_layer_count(struct radv_image_view *iview)
const struct vk_format_description *desc;
unsigned ntype, format, swap, endian;
unsigned blend_clamp = 0, blend_bypass = 0;
- unsigned pitch_tile_max, slice_tile_max, tile_mode_index;
uint64_t va;
const struct radeon_surf *surf = &iview->image->surface;
- const struct radeon_surf_level *level_info = &surf->level[iview->base_mip];
desc = vk_format_description(iview->vk_format);
memset(cb, 0, sizeof(*cb));
+ /* Intensity is implemented as Red, so treat it that way. */
+ cb->cb_color_attrib = S_028C74_FORCE_DST_ALPHA_1(desc->swizzle[3] == VK_SWIZZLE_1);
+
va = device->ws->buffer_get_va(iview->bo) + iview->image->offset;
- va += level_info->offset;
+
cb->cb_color_base = va >> 8;
+ if (device->physical_device->rad_info.chip_class >= GFX9) {
+ struct gfx9_surf_meta_flags meta;
+ if (iview->image->dcc_offset)
+ meta = iview->image->surface.u.gfx9.dcc;
+ else
+ meta = iview->image->surface.u.gfx9.cmask;
+
+ cb->cb_color_attrib |= S_028C74_COLOR_SW_MODE(iview->image->surface.u.gfx9.surf.swizzle_mode) |
+ S_028C74_FMASK_SW_MODE(iview->image->surface.u.gfx9.fmask.swizzle_mode) |
+ S_028C74_RB_ALIGNED(meta.rb_aligned) |
+ S_028C74_PIPE_ALIGNED(meta.pipe_aligned);
+
+ cb->cb_color_base += iview->image->surface.u.gfx9.surf_offset >> 8;
+ cb->cb_color_base |= iview->image->surface.tile_swizzle;
+ } else {
+ const struct legacy_surf_level *level_info = &surf->u.legacy.level[iview->base_mip];
+ unsigned pitch_tile_max, slice_tile_max, tile_mode_index;
+
+ cb->cb_color_base += level_info->offset >> 8;
+ if (level_info->mode == RADEON_SURF_MODE_2D)
+ cb->cb_color_base |= iview->image->surface.tile_swizzle;
+
+ pitch_tile_max = level_info->nblk_x / 8 - 1;
+ slice_tile_max = (level_info->nblk_x * level_info->nblk_y) / 64 - 1;
+ tile_mode_index = si_tile_mode_index(iview->image, iview->base_mip, false);
+
+ cb->cb_color_pitch = S_028C64_TILE_MAX(pitch_tile_max);
+ cb->cb_color_slice = S_028C68_TILE_MAX(slice_tile_max);
+ cb->cb_color_cmask_slice = iview->image->cmask.slice_tile_max;
+
+ cb->cb_color_attrib |= S_028C74_TILE_MODE_INDEX(tile_mode_index);
+ cb->micro_tile_mode = iview->image->surface.micro_tile_mode;
+
+ if (iview->image->fmask.size) {
+ if (device->physical_device->rad_info.chip_class >= CIK)
+ cb->cb_color_pitch |= S_028C64_FMASK_TILE_MAX(iview->image->fmask.pitch_in_pixels / 8 - 1);
+ cb->cb_color_attrib |= S_028C74_FMASK_TILE_MODE_INDEX(iview->image->fmask.tile_mode_index);
+ cb->cb_color_fmask_slice = S_028C88_TILE_MAX(iview->image->fmask.slice_tile_max);
+ } else {
+ /* This must be set for fast clear to work without FMASK. */
+ if (device->physical_device->rad_info.chip_class >= CIK)
+ cb->cb_color_pitch |= S_028C64_FMASK_TILE_MAX(pitch_tile_max);
+ cb->cb_color_attrib |= S_028C74_FMASK_TILE_MODE_INDEX(tile_mode_index);
+ cb->cb_color_fmask_slice = S_028C88_TILE_MAX(slice_tile_max);
+ }
+ }
+
/* CMASK variables */
va = device->ws->buffer_get_va(iview->bo) + iview->image->offset;
va += iview->image->cmask.offset;
cb->cb_color_cmask = va >> 8;
- cb->cb_color_cmask_slice = iview->image->cmask.slice_tile_max;
va = device->ws->buffer_get_va(iview->bo) + iview->image->offset;
va += iview->image->dcc_offset;
cb->cb_dcc_base = va >> 8;
+ cb->cb_dcc_base |= iview->image->surface.tile_swizzle;
uint32_t max_slice = radv_surface_layer_count(iview);
cb->cb_color_view = S_028C6C_SLICE_START(iview->base_layer) |
S_028C6C_SLICE_MAX(iview->base_layer + max_slice - 1);
- cb->micro_tile_mode = iview->image->surface.micro_tile_mode;
- pitch_tile_max = level_info->nblk_x / 8 - 1;
- slice_tile_max = (level_info->nblk_x * level_info->nblk_y) / 64 - 1;
- tile_mode_index = si_tile_mode_index(iview->image, iview->base_mip, false);
-
- cb->cb_color_pitch = S_028C64_TILE_MAX(pitch_tile_max);
- cb->cb_color_slice = S_028C68_TILE_MAX(slice_tile_max);
-
- /* Intensity is implemented as Red, so treat it that way. */
- cb->cb_color_attrib = S_028C74_FORCE_DST_ALPHA_1(desc->swizzle[3] == VK_SWIZZLE_1) |
- S_028C74_TILE_MODE_INDEX(tile_mode_index);
-
- if (iview->image->samples > 1) {
- unsigned log_samples = util_logbase2(iview->image->samples);
+ if (iview->image->info.samples > 1) {
+ unsigned log_samples = util_logbase2(iview->image->info.samples);
cb->cb_color_attrib |= S_028C74_NUM_SAMPLES(log_samples) |
S_028C74_NUM_FRAGMENTS(log_samples);
if (iview->image->fmask.size) {
va = device->ws->buffer_get_va(iview->bo) + iview->image->offset + iview->image->fmask.offset;
- if (device->physical_device->rad_info.chip_class >= CIK)
- cb->cb_color_pitch |= S_028C64_FMASK_TILE_MAX(iview->image->fmask.pitch_in_pixels / 8 - 1);
- cb->cb_color_attrib |= S_028C74_FMASK_TILE_MODE_INDEX(iview->image->fmask.tile_mode_index);
cb->cb_color_fmask = va >> 8;
- cb->cb_color_fmask_slice = S_028C88_TILE_MAX(iview->image->fmask.slice_tile_max);
+ cb->cb_color_fmask |= iview->image->fmask.tile_swizzle;
} else {
- /* This must be set for fast clear to work without FMASK. */
- if (device->physical_device->rad_info.chip_class >= CIK)
- cb->cb_color_pitch |= S_028C64_FMASK_TILE_MAX(pitch_tile_max);
- cb->cb_color_attrib |= S_028C74_FMASK_TILE_MODE_INDEX(tile_mode_index);
cb->cb_color_fmask = cb->cb_color_base;
- cb->cb_color_fmask_slice = S_028C88_TILE_MAX(slice_tile_max);
}
ntype = radv_translate_color_numformat(iview->vk_format,
format != V_028C70_COLOR_24_8) |
S_028C70_NUMBER_TYPE(ntype) |
S_028C70_ENDIAN(endian);
- if (iview->image->samples > 1)
- if (iview->image->fmask.size)
- cb->cb_color_info |= S_028C70_COMPRESSION(1);
+ if ((iview->image->info.samples > 1) && iview->image->fmask.size) {
+ cb->cb_color_info |= S_028C70_COMPRESSION(1);
+ if (device->physical_device->rad_info.chip_class == SI) {
+ unsigned fmask_bankh = util_logbase2(iview->image->fmask.bank_height);
+ cb->cb_color_attrib |= S_028C74_FMASK_BANK_HEIGHT(fmask_bankh);
+ }
+ }
if (iview->image->cmask.size &&
!(device->debug_flags & RADV_DEBUG_NO_FAST_CLEARS))
cb->cb_color_info |= S_028C70_FAST_CLEAR(1);
- if (iview->image->surface.dcc_size && level_info->dcc_enabled)
+ if (iview->image->surface.dcc_size && iview->base_mip < surf->num_dcc_levels)
cb->cb_color_info |= S_028C70_DCC_ENABLE(1);
if (device->physical_device->rad_info.chip_class >= VI) {
unsigned max_uncompressed_block_size = 2;
- if (iview->image->samples > 1) {
+ if (iview->image->info.samples > 1) {
if (iview->image->surface.bpe == 1)
max_uncompressed_block_size = 0;
else if (iview->image->surface.bpe == 2)
/* This must be set for fast clear to work without FMASK. */
if (!iview->image->fmask.size &&
device->physical_device->rad_info.chip_class == SI) {
- unsigned bankh = util_logbase2(iview->image->surface.bankh);
+ unsigned bankh = util_logbase2(iview->image->surface.u.legacy.bankh);
cb->cb_color_attrib |= S_028C74_FMASK_BANK_HEIGHT(bankh);
}
+
+ if (device->physical_device->rad_info.chip_class >= GFX9) {
+ uint32_t max_slice = radv_surface_layer_count(iview);
+ unsigned mip0_depth = iview->base_layer + max_slice - 1;
+
+ cb->cb_color_view |= S_028C6C_MIP_LEVEL(iview->base_mip);
+ cb->cb_color_attrib |= S_028C74_MIP0_DEPTH(mip0_depth) |
+ S_028C74_RESOURCE_TYPE(iview->image->surface.u.gfx9.resource_type);
+ cb->cb_color_attrib2 = S_028C68_MIP0_WIDTH(iview->extent.width - 1) |
+ S_028C68_MIP0_HEIGHT(iview->extent.height - 1) |
+ S_028C68_MAX_MIP(iview->image->info.levels - 1);
+
+ cb->gfx9_epitch = S_0287A0_EPITCH(iview->image->surface.u.gfx9.surf.epitch);
+
+ }
}
static void
struct radv_image_view *iview)
{
unsigned level = iview->base_mip;
- unsigned format;
+ unsigned format, stencil_format;
uint64_t va, s_offs, z_offs;
- const struct radeon_surf_level *level_info = &iview->image->surface.level[level];
+ bool stencil_only = false;
memset(ds, 0, sizeof(*ds));
- switch (iview->vk_format) {
+ switch (iview->image->vk_format) {
case VK_FORMAT_D24_UNORM_S8_UINT:
case VK_FORMAT_X8_D24_UNORM_PACK32:
ds->pa_su_poly_offset_db_fmt_cntl = S_028B78_POLY_OFFSET_NEG_NUM_DB_BITS(-24);
S_028B78_POLY_OFFSET_DB_IS_FLOAT_FMT(1);
ds->offset_scale = 1.0f;
break;
+ case VK_FORMAT_S8_UINT:
+ stencil_only = true;
+ break;
default:
break;
}
- format = radv_translate_dbformat(iview->vk_format);
-
- va = device->ws->buffer_get_va(iview->bo) + iview->image->offset;
- s_offs = z_offs = va;
- z_offs += iview->image->surface.level[level].offset;
- s_offs += iview->image->surface.stencil_level[level].offset;
+ format = radv_translate_dbformat(iview->image->vk_format);
+ stencil_format = iview->image->surface.flags & RADEON_SURF_SBUFFER ?
+ V_028044_STENCIL_8 : V_028044_STENCIL_INVALID;
uint32_t max_slice = radv_surface_layer_count(iview);
ds->db_depth_view = S_028008_SLICE_START(iview->base_layer) |
S_028008_SLICE_MAX(iview->base_layer + max_slice - 1);
- ds->db_depth_info = S_02803C_ADDR5_SWIZZLE_MASK(1);
- ds->db_z_info = S_028040_FORMAT(format) | S_028040_ZRANGE_PRECISION(1);
- if (iview->image->samples > 1)
- ds->db_z_info |= S_028040_NUM_SAMPLES(util_logbase2(iview->image->samples));
+ ds->db_htile_data_base = 0;
+ ds->db_htile_surface = 0;
- if (iview->image->surface.flags & RADEON_SURF_SBUFFER)
- ds->db_stencil_info = S_028044_FORMAT(V_028044_STENCIL_8);
- else
- ds->db_stencil_info = S_028044_FORMAT(V_028044_STENCIL_INVALID);
+ va = device->ws->buffer_get_va(iview->bo) + iview->image->offset;
+ s_offs = z_offs = va;
- if (device->physical_device->rad_info.chip_class >= CIK) {
- struct radeon_info *info = &device->physical_device->rad_info;
- unsigned tiling_index = iview->image->surface.tiling_index[level];
- unsigned stencil_index = iview->image->surface.stencil_tiling_index[level];
- unsigned macro_index = iview->image->surface.macro_tile_index;
- unsigned tile_mode = info->si_tile_mode_array[tiling_index];
- unsigned stencil_tile_mode = info->si_tile_mode_array[stencil_index];
- unsigned macro_mode = info->cik_macrotile_mode_array[macro_index];
-
- ds->db_depth_info |=
- S_02803C_ARRAY_MODE(G_009910_ARRAY_MODE(tile_mode)) |
- S_02803C_PIPE_CONFIG(G_009910_PIPE_CONFIG(tile_mode)) |
- S_02803C_BANK_WIDTH(G_009990_BANK_WIDTH(macro_mode)) |
- S_02803C_BANK_HEIGHT(G_009990_BANK_HEIGHT(macro_mode)) |
- S_02803C_MACRO_TILE_ASPECT(G_009990_MACRO_TILE_ASPECT(macro_mode)) |
- S_02803C_NUM_BANKS(G_009990_NUM_BANKS(macro_mode));
- ds->db_z_info |= S_028040_TILE_SPLIT(G_009910_TILE_SPLIT(tile_mode));
- ds->db_stencil_info |= S_028044_TILE_SPLIT(G_009910_TILE_SPLIT(stencil_tile_mode));
- } else {
- unsigned tile_mode_index = si_tile_mode_index(iview->image, level, false);
- ds->db_z_info |= S_028040_TILE_MODE_INDEX(tile_mode_index);
- tile_mode_index = si_tile_mode_index(iview->image, level, true);
- ds->db_stencil_info |= S_028044_TILE_MODE_INDEX(tile_mode_index);
- }
-
- if (iview->image->surface.htile_size && !level) {
- ds->db_z_info |= S_028040_TILE_SURFACE_ENABLE(1) |
- S_028040_ALLOW_EXPCLEAR(1);
-
- if (iview->image->surface.flags & RADEON_SURF_SBUFFER) {
- /* Workaround: For a not yet understood reason, the
- * combination of MSAA, fast stencil clear and stencil
- * decompress messes with subsequent stencil buffer
- * uses. Problem was reproduced on Verde, Bonaire,
- * Tonga, and Carrizo.
- *
- * Disabling EXPCLEAR works around the problem.
- *
- * Check piglit's arb_texture_multisample-stencil-clear
- * test if you want to try changing this.
- */
- if (iview->image->samples <= 1)
- ds->db_stencil_info |= S_028044_ALLOW_EXPCLEAR(1);
- } else
- /* Use all of the htile_buffer for depth if there's no stencil. */
- ds->db_stencil_info |= S_028044_TILE_STENCIL_DISABLE(1);
-
- va = device->ws->buffer_get_va(iview->bo) + iview->image->offset +
- iview->image->htile_offset;
- ds->db_htile_data_base = va >> 8;
- ds->db_htile_surface = S_028ABC_FULL_CACHE(1);
+ if (device->physical_device->rad_info.chip_class >= GFX9) {
+ assert(iview->image->surface.u.gfx9.surf_offset == 0);
+ s_offs += iview->image->surface.u.gfx9.stencil_offset;
+
+ ds->db_z_info = S_028038_FORMAT(format) |
+ S_028038_NUM_SAMPLES(util_logbase2(iview->image->info.samples)) |
+ S_028038_SW_MODE(iview->image->surface.u.gfx9.surf.swizzle_mode) |
+ S_028038_MAXMIP(iview->image->info.levels - 1);
+ ds->db_stencil_info = S_02803C_FORMAT(stencil_format) |
+ S_02803C_SW_MODE(iview->image->surface.u.gfx9.stencil.swizzle_mode);
+
+ ds->db_z_info2 = S_028068_EPITCH(iview->image->surface.u.gfx9.surf.epitch);
+ ds->db_stencil_info2 = S_02806C_EPITCH(iview->image->surface.u.gfx9.stencil.epitch);
+ ds->db_depth_view |= S_028008_MIPID(level);
+
+ ds->db_depth_size = S_02801C_X_MAX(iview->image->info.width - 1) |
+ S_02801C_Y_MAX(iview->image->info.height - 1);
+
+ /* Only use HTILE for the first level. */
+ if (iview->image->surface.htile_size && !level) {
+ ds->db_z_info |= S_028038_TILE_SURFACE_ENABLE(1);
+
+ if (!(iview->image->surface.flags & RADEON_SURF_SBUFFER))
+ /* Use all of the htile_buffer for depth if there's no stencil. */
+ ds->db_stencil_info |= S_02803C_TILE_STENCIL_DISABLE(1);
+ va = device->ws->buffer_get_va(iview->bo) + iview->image->offset +
+ iview->image->htile_offset;
+ ds->db_htile_data_base = va >> 8;
+ ds->db_htile_surface = S_028ABC_FULL_CACHE(1) |
+ S_028ABC_PIPE_ALIGNED(iview->image->surface.u.gfx9.htile.pipe_aligned) |
+ S_028ABC_RB_ALIGNED(iview->image->surface.u.gfx9.htile.rb_aligned);
+ }
} else {
- ds->db_htile_data_base = 0;
- ds->db_htile_surface = 0;
+ const struct legacy_surf_level *level_info = &iview->image->surface.u.legacy.level[level];
+
+ if (stencil_only)
+ level_info = &iview->image->surface.u.legacy.stencil_level[level];
+
+ z_offs += iview->image->surface.u.legacy.level[level].offset;
+ s_offs += iview->image->surface.u.legacy.stencil_level[level].offset;
+
+ ds->db_depth_info = S_02803C_ADDR5_SWIZZLE_MASK(1);
+ ds->db_z_info = S_028040_FORMAT(format) | S_028040_ZRANGE_PRECISION(1);
+ ds->db_stencil_info = S_028044_FORMAT(stencil_format);
+
+ if (iview->image->info.samples > 1)
+ ds->db_z_info |= S_028040_NUM_SAMPLES(util_logbase2(iview->image->info.samples));
+
+ if (device->physical_device->rad_info.chip_class >= CIK) {
+ struct radeon_info *info = &device->physical_device->rad_info;
+ unsigned tiling_index = iview->image->surface.u.legacy.tiling_index[level];
+ unsigned stencil_index = iview->image->surface.u.legacy.stencil_tiling_index[level];
+ unsigned macro_index = iview->image->surface.u.legacy.macro_tile_index;
+ unsigned tile_mode = info->si_tile_mode_array[tiling_index];
+ unsigned stencil_tile_mode = info->si_tile_mode_array[stencil_index];
+ unsigned macro_mode = info->cik_macrotile_mode_array[macro_index];
+
+ if (stencil_only)
+ tile_mode = stencil_tile_mode;
+
+ ds->db_depth_info |=
+ S_02803C_ARRAY_MODE(G_009910_ARRAY_MODE(tile_mode)) |
+ S_02803C_PIPE_CONFIG(G_009910_PIPE_CONFIG(tile_mode)) |
+ S_02803C_BANK_WIDTH(G_009990_BANK_WIDTH(macro_mode)) |
+ S_02803C_BANK_HEIGHT(G_009990_BANK_HEIGHT(macro_mode)) |
+ S_02803C_MACRO_TILE_ASPECT(G_009990_MACRO_TILE_ASPECT(macro_mode)) |
+ S_02803C_NUM_BANKS(G_009990_NUM_BANKS(macro_mode));
+ ds->db_z_info |= S_028040_TILE_SPLIT(G_009910_TILE_SPLIT(tile_mode));
+ ds->db_stencil_info |= S_028044_TILE_SPLIT(G_009910_TILE_SPLIT(stencil_tile_mode));
+ } else {
+ unsigned tile_mode_index = si_tile_mode_index(iview->image, level, false);
+ ds->db_z_info |= S_028040_TILE_MODE_INDEX(tile_mode_index);
+ tile_mode_index = si_tile_mode_index(iview->image, level, true);
+ ds->db_stencil_info |= S_028044_TILE_MODE_INDEX(tile_mode_index);
+ if (stencil_only)
+ ds->db_z_info |= S_028040_TILE_MODE_INDEX(tile_mode_index);
+ }
+
+ ds->db_depth_size = S_028058_PITCH_TILE_MAX((level_info->nblk_x / 8) - 1) |
+ S_028058_HEIGHT_TILE_MAX((level_info->nblk_y / 8) - 1);
+ ds->db_depth_slice = S_02805C_SLICE_TILE_MAX((level_info->nblk_x * level_info->nblk_y) / 64 - 1);
+
+ if (iview->image->surface.htile_size && !level) {
+ ds->db_z_info |= S_028040_TILE_SURFACE_ENABLE(1);
+
+ if (!(iview->image->surface.flags & RADEON_SURF_SBUFFER))
+ /* Use all of the htile_buffer for depth if there's no stencil. */
+ ds->db_stencil_info |= S_028044_TILE_STENCIL_DISABLE(1);
+
+ va = device->ws->buffer_get_va(iview->bo) + iview->image->offset +
+ iview->image->htile_offset;
+ ds->db_htile_data_base = va >> 8;
+ ds->db_htile_surface = S_028ABC_FULL_CACHE(1);
+ }
}
ds->db_z_read_base = ds->db_z_write_base = z_offs >> 8;
ds->db_stencil_read_base = ds->db_stencil_write_base = s_offs >> 8;
-
- ds->db_depth_size = S_028058_PITCH_TILE_MAX((level_info->nblk_x / 8) - 1) |
- S_028058_HEIGHT_TILE_MAX((level_info->nblk_y / 8) - 1);
- ds->db_depth_slice = S_02805C_SLICE_TILE_MAX((level_info->nblk_x * level_info->nblk_y) / 64 - 1);
}
VkResult radv_CreateFramebuffer(
vk_free2(&device->alloc, pAllocator, sampler);
}
-
/* vk_icd.h does not declare this function, so we declare it here to
* suppress Wmissing-prototypes.
*/
*pSupportedVersion = MIN2(*pSupportedVersion, 3u);
return VK_SUCCESS;
}
+
+VkResult radv_GetMemoryFdKHR(VkDevice _device,
+ const VkMemoryGetFdInfoKHR *pGetFdInfo,
+ int *pFD)
+{
+ RADV_FROM_HANDLE(radv_device, device, _device);
+ RADV_FROM_HANDLE(radv_device_memory, memory, pGetFdInfo->memory);
+
+ assert(pGetFdInfo->sType == VK_STRUCTURE_TYPE_MEMORY_GET_FD_INFO_KHR);
+
+ /* We support only one handle type. */
+ assert(pGetFdInfo->handleType ==
+ VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT_KHR);
+
+ bool ret = radv_get_memory_fd(device, memory, pFD);
+ if (ret == false)
+ return VK_ERROR_OUT_OF_DEVICE_MEMORY;
+ return VK_SUCCESS;
+}
+
+VkResult radv_GetMemoryFdPropertiesKHR(VkDevice _device,
+ VkExternalMemoryHandleTypeFlagBitsKHR handleType,
+ int fd,
+ VkMemoryFdPropertiesKHR *pMemoryFdProperties)
+{
+ /* The valid usage section for this function says:
+ *
+ * "handleType must not be one of the handle types defined as opaque."
+ *
+ * Since we only handle opaque handles for now, there are no FD properties.
+ */
+ return VK_ERROR_INVALID_EXTERNAL_HANDLE_KHR;
+}
+
+VkResult radv_ImportSemaphoreFdKHR(VkDevice _device,
+ const VkImportSemaphoreFdInfoKHR *pImportSemaphoreFdInfo)
+{
+ RADV_FROM_HANDLE(radv_device, device, _device);
+ RADV_FROM_HANDLE(radv_semaphore, sem, pImportSemaphoreFdInfo->semaphore);
+ uint32_t syncobj_handle = 0;
+ assert(pImportSemaphoreFdInfo->handleType == VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT_KHR);
+
+ int ret = device->ws->import_syncobj(device->ws, pImportSemaphoreFdInfo->fd, &syncobj_handle);
+ if (ret != 0)
+ return VK_ERROR_INVALID_EXTERNAL_HANDLE_KHR;
+
+ if (pImportSemaphoreFdInfo->flags & VK_SEMAPHORE_IMPORT_TEMPORARY_BIT_KHR) {
+ sem->temp_syncobj = syncobj_handle;
+ } else {
+ sem->syncobj = syncobj_handle;
+ }
+ close(pImportSemaphoreFdInfo->fd);
+ return VK_SUCCESS;
+}
+
+VkResult radv_GetSemaphoreFdKHR(VkDevice _device,
+ const VkSemaphoreGetFdInfoKHR *pGetFdInfo,
+ int *pFd)
+{
+ RADV_FROM_HANDLE(radv_device, device, _device);
+ RADV_FROM_HANDLE(radv_semaphore, sem, pGetFdInfo->semaphore);
+ int ret;
+ uint32_t syncobj_handle;
+
+ assert(pGetFdInfo->handleType == VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT_KHR);
+ if (sem->temp_syncobj)
+ syncobj_handle = sem->temp_syncobj;
+ else
+ syncobj_handle = sem->syncobj;
+ ret = device->ws->export_syncobj(device->ws, syncobj_handle, pFd);
+ if (ret)
+ return vk_error(VK_ERROR_INVALID_EXTERNAL_HANDLE_KHR);
+ return VK_SUCCESS;
+}
+
+void radv_GetPhysicalDeviceExternalSemaphorePropertiesKHR(
+ VkPhysicalDevice physicalDevice,
+ const VkPhysicalDeviceExternalSemaphoreInfoKHR* pExternalSemaphoreInfo,
+ VkExternalSemaphorePropertiesKHR* pExternalSemaphoreProperties)
+{
+ if (pExternalSemaphoreInfo->handleType == VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT_KHR) {
+ pExternalSemaphoreProperties->exportFromImportedHandleTypes = VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT_KHR;
+ pExternalSemaphoreProperties->compatibleHandleTypes = VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT_KHR;
+ pExternalSemaphoreProperties->externalSemaphoreFeatures = VK_EXTERNAL_SEMAPHORE_FEATURE_EXPORTABLE_BIT_KHR |
+ VK_EXTERNAL_SEMAPHORE_FEATURE_IMPORTABLE_BIT_KHR;
+ } else {
+ pExternalSemaphoreProperties->exportFromImportedHandleTypes = 0;
+ pExternalSemaphoreProperties->compatibleHandleTypes = 0;
+ pExternalSemaphoreProperties->externalSemaphoreFeatures = 0;
+ }
+}