Files
totk-pc-recomp/runtime/vk_device.c
T
Valiantsin DzerakhandClaude Opus 5.5 57037ccc29 Deswizzle block-linear textures on the GPU when their storage is an imported pool
The replay thread deswizzled every block-linear texture on the CPU, and entering a new area stalled it for 50-236 ms. Now a compute pass reads the texture straight from the pool's imported buffer and writes a linear copy into a device-local scratch pool; the upload batch copies it into the image, or the ASTC decoder takes it. Large texture loads submit the batch early so the staging and scratch pools do not grow on the replay thread, and both pools are made at device creation.

On fight2 the render thread waits about 270 ms on textures per run instead of 900, and the load frame drops from 145 to about 80 ms. Mean fps is unchanged. TOTK_DESWIZZLE=cpu turns it off and check compares every upload with the CPU result (0 of 196k regions differ). Shadow mode (Linux/Wine) and textures outside the device's limits stay on the CPU.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-07 16:38:26 +03:00

520 lines
29 KiB
C

// Vulkan instance, physical device choice, logical device, VMA, timeline semaphore and the validation messenger.
// Ported from LAS Recomp, runtime/nvn/vulkan/device.c:33-330
// (https://github.com/peiikai/LASRecomp @ da35b83a9611d046a9a681aae668d7085954d37f, MIT, Copyright (c) 2026 Peiikai),
// without SDL. The VMA implementation is runtime/vk_vma.cpp (VMA is C++ only).
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include "cfg.h"
#include "host.h"
#include "vk.h"
#include "platform/platform.h"
#include "vk_dlaa.h"
static VkCtx g_vk;
static int g_ready;
static PlatMutex g_lock = PLAT_MUTEX_INIT;
static VKAPI_ATTR VkBool32 VKAPI_CALL debug_cb(VkDebugUtilsMessageSeverityFlagBitsEXT sev, VkDebugUtilsMessageTypeFlagsEXT type,
const VkDebugUtilsMessengerCallbackDataEXT *d, void *user) {
(void)type, (void)user;
int err = (sev & VK_DEBUG_UTILS_MESSAGE_SEVERITY_ERROR_BIT_EXT) != 0;
if (err) plat_inc32(&g_vk.validation_errors);
static volatile int32_t logged;
if (plat_inc32(&logged) <= 500) rt_log("vk validation %s: %s", err ? "error" : "warning", d->pMessage);
return VK_FALSE;
}
static int has_ext(const VkExtensionProperties *e, uint32_t n, const char *name) {
for (uint32_t i = 0; i < n; i++) if (!strcmp(e[i].extensionName, name)) return 1;
return 0;
}
static int layer_present(const char *name) {
uint32_t n = 0;
vkEnumerateInstanceLayerProperties(&n, NULL);
VkLayerProperties *l = calloc(n ? n : 1, sizeof *l);
vkEnumerateInstanceLayerProperties(&n, l);
int found = 0;
for (uint32_t i = 0; i < n; i++) if (!strcmp(l[i].layerName, name)) found = 1;
free(l);
return found;
}
static int instance_ext_present(const char *name) {
uint32_t n = 0;
vkEnumerateInstanceExtensionProperties(NULL, &n, NULL);
VkExtensionProperties *e = calloc(n ? n : 1, sizeof *e);
vkEnumerateInstanceExtensionProperties(NULL, &n, e);
int found = 0;
for (uint32_t i = 0; i < n; i++) if (!strcmp(e[i].extensionName, name)) found = 1;
free(e);
return found;
}
// Mesa's drivers (RADV on Linux and the Steam Deck) keep a shader cache of their own beside our pipeline cache, by
// default in ~/.cache and up to 1 GB: about 35 MB in a minute of the title screen. A Deck's /home is small and the game
// usually sits on the SD card, so unless MESA_SHADER_CACHE_DIR names a folder, it goes to mesa-cache beside the exe.
static void mesa_cache_beside_exe(void) {
char dir[PLAT_PATH_MAX * 2];
if (plat_getenv("MESA_SHADER_CACHE_DIR", dir, sizeof dir)) return;
int len = plat_exe_path(dir, sizeof dir - 16);
char *cut = len ? plat_last_sep(dir) : NULL;
if (!cut) return;
strcpy(cut + 1, "mesa-cache");
plat_setenv("MESA_SHADER_CACHE_DIR", dir);
}
static int create_instance(void) {
mesa_cache_beside_exe();
const char *ext[12];
uint32_t n_ext = (uint32_t)gfx_window_vk_extensions(ext, 7);
const char *layers[] = {"VK_LAYER_KHRONOS_validation"};
int validate = cfg_bool("TOTK_VK_VALIDATION");
if (validate && !layer_present(layers[0])) {
rt_log("vk: TOTK_VK_VALIDATION set but %s is not installed; continuing without", layers[0]);
validate = 0;
}
if (validate) ext[n_ext++] = VK_EXT_DEBUG_UTILS_EXTENSION_NAME;
if (vk_hdr_wanted() && instance_ext_present(VK_EXT_SWAPCHAIN_COLOR_SPACE_EXTENSION_NAME)) // HDR10 surface formats
ext[n_ext++] = VK_EXT_SWAPCHAIN_COLOR_SPACE_EXTENSION_NAME;
vk_dlaa_instance_exts(ext, &n_ext, 12);
VkApplicationInfo app = {.sType = VK_STRUCTURE_TYPE_APPLICATION_INFO, .pApplicationName = "TotK Recomp",
.pEngineName = "TotK Recomp", .apiVersion = VK_API_VERSION_1_3};
VkInstanceCreateInfo ci = {.sType = VK_STRUCTURE_TYPE_INSTANCE_CREATE_INFO, .pApplicationInfo = &app,
.enabledExtensionCount = n_ext, .ppEnabledExtensionNames = ext,
.enabledLayerCount = validate ? 1 : 0, .ppEnabledLayerNames = layers};
rdoc_init();
VkResult r = vkCreateInstance(&ci, NULL, &g_vk.instance);
if (r != VK_SUCCESS) { rt_log("vk: vkCreateInstance failed (%d); is a Vulkan 1.3 driver installed?", (int)r); return 0; }
g_vk.validation = validate;
if (validate) {
PFN_vkCreateDebugUtilsMessengerEXT f =
(PFN_vkCreateDebugUtilsMessengerEXT)vkGetInstanceProcAddr(g_vk.instance, "vkCreateDebugUtilsMessengerEXT");
VkDebugUtilsMessengerCreateInfoEXT mi = {
.sType = VK_STRUCTURE_TYPE_DEBUG_UTILS_MESSENGER_CREATE_INFO_EXT,
.messageSeverity = VK_DEBUG_UTILS_MESSAGE_SEVERITY_WARNING_BIT_EXT | VK_DEBUG_UTILS_MESSAGE_SEVERITY_ERROR_BIT_EXT,
.messageType = VK_DEBUG_UTILS_MESSAGE_TYPE_GENERAL_BIT_EXT | VK_DEBUG_UTILS_MESSAGE_TYPE_VALIDATION_BIT_EXT |
VK_DEBUG_UTILS_MESSAGE_TYPE_PERFORMANCE_BIT_EXT,
.pfnUserCallback = debug_cb};
if (f) f(g_vk.instance, &mi, NULL, &g_vk.messenger);
}
return 1;
}
static const char *const k_required_ext[] = {
VK_KHR_SWAPCHAIN_EXTENSION_NAME, VK_KHR_PUSH_DESCRIPTOR_EXTENSION_NAME, VK_EXT_DEPTH_CLIP_CONTROL_EXTENSION_NAME,
VK_EXT_EXTERNAL_MEMORY_HOST_EXTENSION_NAME};
// Name of the first missing required extension, or NULL.
static const char *missing_required(VkPhysicalDevice pd) {
uint32_t n = 0;
vkEnumerateDeviceExtensionProperties(pd, NULL, &n, NULL);
VkExtensionProperties *e = calloc(n ? n : 1, sizeof *e);
vkEnumerateDeviceExtensionProperties(pd, NULL, &n, e);
const char *miss = NULL;
for (size_t i = 0; i < sizeof k_required_ext / sizeof *k_required_ext && !miss; i++)
if (!has_ext(e, n, k_required_ext[i])) miss = k_required_ext[i];
free(e);
return miss;
}
// Graphics+compute family that can present to the window.
static uint32_t find_gfx_family(VkPhysicalDevice pd, const VkQueueFamilyProperties *qp, uint32_t nq) {
for (uint32_t i = 0; i < nq; i++)
if ((qp[i].queueFlags & VK_QUEUE_GRAPHICS_BIT) && (qp[i].queueFlags & VK_QUEUE_COMPUTE_BIT) &&
gfx_window_vk_presentable(g_vk.instance, pd, i))
return i;
return UINT32_MAX;
}
static int name_has(const char *s, const char *sub) {
for (size_t n = strlen(sub); *s; s++)
if (!strncasecmp(s, sub, n)) return 1;
return 0;
}
// First discrete GPU with Vulkan 1.3, the required extensions and a graphics+present queue; TOTK_GPU_INDEX (index) or
// TOTK_GPU_NAME (case-insensitive part of the name) overrides.
static int pick_device(void) {
uint32_t n = 0;
vkEnumeratePhysicalDevices(g_vk.instance, &n, NULL);
if (!n) { rt_log("vk: no Vulkan device"); return 0; }
VkPhysicalDevice *pd = calloc(n, sizeof *pd);
vkEnumeratePhysicalDevices(g_vk.instance, &n, pd);
const char *e = cfg_get("TOTK_GPU_INDEX"); // no fixed default: the best device when unset
int want = e && *e ? atoi(e) : -1;
const char *name = cfg_get("TOTK_GPU_NAME");
int best = -1, best_score = -1;
for (uint32_t i = 0; i < n; i++) {
VkPhysicalDeviceProperties p;
vkGetPhysicalDeviceProperties(pd[i], &p);
rt_log("vk: gpu %u: %s (type %d, api %u.%u.%u)", i, p.deviceName, (int)p.deviceType, VK_API_VERSION_MAJOR(p.apiVersion),
VK_API_VERSION_MINOR(p.apiVersion), VK_API_VERSION_PATCH(p.apiVersion));
if (want >= 0 && (int)i != want) continue;
if (name && *name && !name_has(p.deviceName, name)) continue;
if (p.apiVersion < VK_API_VERSION_1_3) { rt_log("vk: gpu %u skipped: Vulkan below 1.3", i); continue; }
const char *miss = missing_required(pd[i]);
if (miss) { rt_log("vk: gpu %u skipped: missing extension %s", i, miss); continue; }
uint32_t nq = 0;
vkGetPhysicalDeviceQueueFamilyProperties(pd[i], &nq, NULL);
VkQueueFamilyProperties *qp = calloc(nq ? nq : 1, sizeof *qp);
vkGetPhysicalDeviceQueueFamilyProperties(pd[i], &nq, qp);
uint32_t fam = find_gfx_family(pd[i], qp, nq);
free(qp);
if (fam == UINT32_MAX) { rt_log("vk: gpu %u skipped: no graphics+compute+present queue family", i); continue; }
int score = (int)i == want ? 100 : p.deviceType == VK_PHYSICAL_DEVICE_TYPE_DISCRETE_GPU ? 3
: p.deviceType == VK_PHYSICAL_DEVICE_TYPE_INTEGRATED_GPU ? 2 : 1;
if (score > best_score) { best_score = score; best = (int)i; }
}
if (best < 0) {
if (want >= 0) rt_log("vk: TOTK_GPU_INDEX=%d is not usable (see above)", want);
else if (name && *name) rt_log("vk: no usable gpu matches TOTK_GPU_NAME=%s (see above)", name);
else rt_log("vk: no device with Vulkan 1.3 and the required extensions (see above)");
free(pd);
return 0;
}
g_vk.phys = pd[best];
g_vk.device_index = best;
free(pd);
vkGetPhysicalDeviceProperties(g_vk.phys, &g_vk.props);
vkGetPhysicalDeviceMemoryProperties(g_vk.phys, &g_vk.mem_props);
uint32_t nq = 0;
vkGetPhysicalDeviceQueueFamilyProperties(g_vk.phys, &nq, NULL);
VkQueueFamilyProperties *qp = calloc(nq ? nq : 1, sizeof *qp);
vkGetPhysicalDeviceQueueFamilyProperties(g_vk.phys, &nq, qp);
g_vk.gfx_family = find_gfx_family(g_vk.phys, qp, nq);
g_vk.xfer_family = g_vk.gfx_family;
for (uint32_t i = 0; i < nq; i++) // a family that does transfer only (async DMA engine)
if ((qp[i].queueFlags & VK_QUEUE_TRANSFER_BIT) && !(qp[i].queueFlags & (VK_QUEUE_GRAPHICS_BIT | VK_QUEUE_COMPUTE_BIT))) {
g_vk.xfer_family = i;
break;
}
free(qp);
rt_log("vk: using gpu %d: %s", best, g_vk.props.deviceName);
vk_mem_class_log(&g_vk.props, &g_vk.mem_props);
return 1;
}
typedef struct {
VkPhysicalDeviceFeatures2 f2;
VkPhysicalDeviceVulkan11Features f11;
VkPhysicalDeviceVulkan12Features f12;
VkPhysicalDeviceVulkan13Features f13;
VkPhysicalDeviceDepthClipControlFeaturesEXT clip;
VkPhysicalDeviceCustomBorderColorFeaturesEXT cbc;
VkPhysicalDeviceExtendedDynamicStateFeaturesEXT eds;
VkPhysicalDeviceExtendedDynamicState2FeaturesEXT eds2;
VkPhysicalDeviceExtendedDynamicState3FeaturesEXT eds3;
VkPhysicalDeviceVertexInputDynamicStateFeaturesEXT vids;
VkPhysicalDeviceConditionalRenderingFeaturesEXT cond;
VkPhysicalDeviceTransformFeedbackFeaturesEXT xfb;
VkPhysicalDeviceProvokingVertexFeaturesEXT pv;
VkPhysicalDeviceRobustness2FeaturesEXT rob2;
VkPhysicalDeviceFaultFeaturesEXT fault;
VkPhysicalDeviceFragmentShaderInterlockFeaturesEXT fsi;
VkPhysicalDeviceGraphicsPipelineLibraryFeaturesEXT gpl;
VkPhysicalDeviceAttachmentFeedbackLoopLayoutFeaturesEXT fbl;
VkPhysicalDeviceDynamicRenderingLocalReadFeaturesKHR lr;
} FeatureChain;
static int create_device(void) {
uint32_t ne = 0;
vkEnumerateDeviceExtensionProperties(g_vk.phys, NULL, &ne, NULL);
VkExtensionProperties *ep = calloc(ne ? ne : 1, sizeof *ep);
vkEnumerateDeviceExtensionProperties(g_vk.phys, NULL, &ne, ep);
const char *ext[48];
uint32_t n_ext = 0;
for (size_t i = 0; i < sizeof k_required_ext / sizeof *k_required_ext; i++) ext[n_ext++] = k_required_ext[i];
FeatureChain *c = calloc(1, sizeof *c);
void **tail = &c->f2.pNext;
#define CHAIN(member, stype) do { c->member.sType = stype; *tail = &c->member; tail = &c->member.pNext; } while (0)
c->f2.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2;
CHAIN(f11, VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_FEATURES);
CHAIN(f12, VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_2_FEATURES);
CHAIN(f13, VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_3_FEATURES);
CHAIN(clip, VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DEPTH_CLIP_CONTROL_FEATURES_EXT);
#define OPTIONAL_EXT(flag, name, member, stype) do { \
if (has_ext(ep, ne, name)) { g_vk.flag = 1; ext[n_ext++] = name; CHAIN(member, stype); } } while (0)
OPTIONAL_EXT(has_custom_border_color, VK_EXT_CUSTOM_BORDER_COLOR_EXTENSION_NAME, cbc, VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CUSTOM_BORDER_COLOR_FEATURES_EXT);
OPTIONAL_EXT(has_eds, VK_EXT_EXTENDED_DYNAMIC_STATE_EXTENSION_NAME, eds, VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_EXTENDED_DYNAMIC_STATE_FEATURES_EXT);
OPTIONAL_EXT(has_eds2, VK_EXT_EXTENDED_DYNAMIC_STATE_2_EXTENSION_NAME, eds2, VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_EXTENDED_DYNAMIC_STATE_2_FEATURES_EXT);
OPTIONAL_EXT(has_eds3, VK_EXT_EXTENDED_DYNAMIC_STATE_3_EXTENSION_NAME, eds3, VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_EXTENDED_DYNAMIC_STATE_3_FEATURES_EXT);
OPTIONAL_EXT(has_vertex_input_dynamic_state, VK_EXT_VERTEX_INPUT_DYNAMIC_STATE_EXTENSION_NAME, vids, VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VERTEX_INPUT_DYNAMIC_STATE_FEATURES_EXT);
OPTIONAL_EXT(has_conditional_rendering, VK_EXT_CONDITIONAL_RENDERING_EXTENSION_NAME, cond, VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CONDITIONAL_RENDERING_FEATURES_EXT);
OPTIONAL_EXT(has_transform_feedback, VK_EXT_TRANSFORM_FEEDBACK_EXTENSION_NAME, xfb, VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_FEATURES_EXT);
OPTIONAL_EXT(has_provoking_vertex, VK_EXT_PROVOKING_VERTEX_EXTENSION_NAME, pv, VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROVOKING_VERTEX_FEATURES_EXT);
OPTIONAL_EXT(has_robustness2, VK_EXT_ROBUSTNESS_2_EXTENSION_NAME, rob2, VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ROBUSTNESS_2_FEATURES_EXT);
OPTIONAL_EXT(has_device_fault, VK_EXT_DEVICE_FAULT_EXTENSION_NAME, fault, VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FAULT_FEATURES_EXT);
// Ryujinx's SPIR-V declares fragment shader interlock whenever the host reports it (vk_pipeline.c spirv_fixup).
OPTIONAL_EXT(has_fragment_shader_interlock, VK_EXT_FRAGMENT_SHADER_INTERLOCK_EXTENSION_NAME, fsi,
VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FRAGMENT_SHADER_INTERLOCK_FEATURES_EXT);
// Feedback loops (vk.h VkCtx.feedback): the layout, and barriers inside dynamic rendering
int has_fbl = 0, has_lr = 0;
if (has_ext(ep, ne, VK_EXT_ATTACHMENT_FEEDBACK_LOOP_LAYOUT_EXTENSION_NAME) &&
has_ext(ep, ne, VK_KHR_DYNAMIC_RENDERING_LOCAL_READ_EXTENSION_NAME)) {
has_fbl = has_lr = 1;
ext[n_ext++] = VK_EXT_ATTACHMENT_FEEDBACK_LOOP_LAYOUT_EXTENSION_NAME;
ext[n_ext++] = VK_KHR_DYNAMIC_RENDERING_LOCAL_READ_EXTENSION_NAME;
CHAIN(fbl, VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ATTACHMENT_FEEDBACK_LOOP_LAYOUT_FEATURES_EXT);
CHAIN(lr, VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DYNAMIC_RENDERING_LOCAL_READ_FEATURES_KHR);
}
vk_dlaa_device_exts(ep, ne, ext, &n_ext, 40); // NGX's; the eight slots after it stay for the seven below
if (has_ext(ep, ne, VK_EXT_MEMORY_BUDGET_EXTENSION_NAME)) { g_vk.has_memory_budget = 1; ext[n_ext++] = VK_EXT_MEMORY_BUDGET_EXTENSION_NAME; }
// GPU timestamps on the CPU clock (vk_sync.c): the perf line splits GPU idle time into waiting for a submission or not.
if (has_ext(ep, ne, VK_KHR_CALIBRATED_TIMESTAMPS_EXTENSION_NAME)) { g_vk.has_calibrated = 2; ext[n_ext++] = VK_KHR_CALIBRATED_TIMESTAMPS_EXTENSION_NAME; }
else if (has_ext(ep, ne, VK_EXT_CALIBRATED_TIMESTAMPS_EXTENSION_NAME)) { g_vk.has_calibrated = 1; ext[n_ext++] = VK_EXT_CALIBRATED_TIMESTAMPS_EXTENSION_NAME; }
if (vk_hdr_wanted() && has_ext(ep, ne, VK_EXT_HDR_METADATA_EXTENSION_NAME)) { // the display's peak (vk_hdr_metadata)
g_vk.has_hdr_metadata = 1;
ext[n_ext++] = VK_EXT_HDR_METADATA_EXTENSION_NAME;
}
// GPU breadcrumbs for a lost device (vk_crumbs.c): AMD's drivers and RADV report it.
if (has_ext(ep, ne, VK_AMD_BUFFER_MARKER_EXTENSION_NAME)) { g_vk.has_buffer_marker = 1; ext[n_ext++] = VK_AMD_BUFFER_MARKER_EXTENSION_NAME; }
// Pass marks on NVIDIA (vk_crumbs.c): checkpoints only with TOTK_PASS_CRUMBS=1, since NVIDIA documents a cost for
// the extension. NVIDIA's drivers report it, AMD's do not.
if (vk_pass_marks_want_nv() && has_ext(ep, ne, VK_NV_DEVICE_DIAGNOSTIC_CHECKPOINTS_EXTENSION_NAME)) {
g_vk.has_nv_checkpoints = 1;
ext[n_ext++] = VK_NV_DEVICE_DIAGNOSTIC_CHECKPOINTS_EXTENSION_NAME;
}
// Graphics pipeline library: chained last so it can be dropped again when vk_caps_gpl declines it.
void **gpl_link = NULL;
if (has_ext(ep, ne, VK_KHR_PIPELINE_LIBRARY_EXTENSION_NAME) && has_ext(ep, ne, VK_EXT_GRAPHICS_PIPELINE_LIBRARY_EXTENSION_NAME)) {
ext[n_ext++] = VK_KHR_PIPELINE_LIBRARY_EXTENSION_NAME;
ext[n_ext++] = VK_EXT_GRAPHICS_PIPELINE_LIBRARY_EXTENSION_NAME;
gpl_link = tail;
CHAIN(gpl, VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_GRAPHICS_PIPELINE_LIBRARY_FEATURES_EXT);
}
free(ep);
// The driver fills the chain with what it supports; the same chain then enables all of it (robustBufferAccess
// stays on: guest shaders index constant buffers freely and out-of-range reads must stay harmless).
vkGetPhysicalDeviceFeatures2(g_vk.phys, &c->f2);
vk_caps_driver(g_vk.phys);
vk_caps_features(&c->f2.features);
if (gpl_link) {
VkPhysicalDeviceGraphicsPipelineLibraryPropertiesEXT gp = {
.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_GRAPHICS_PIPELINE_LIBRARY_PROPERTIES_EXT};
VkPhysicalDeviceProperties2 p2 = {.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2, .pNext = &gp};
vkGetPhysicalDeviceProperties2(g_vk.phys, &p2);
if (!vk_caps_gpl(&c->gpl, &gp)) { *gpl_link = NULL; n_ext -= 2; }
}
const VkPhysicalDeviceFeatures *h = &c->f2.features;
const char *miss = NULL;
if (!c->f12.timelineSemaphore) miss = "timelineSemaphore";
else if (!c->f13.dynamicRendering) miss = "dynamicRendering";
else if (!c->f13.synchronization2) miss = "synchronization2";
else if (!h->shaderStorageImageWriteWithoutFormat) miss = "shaderStorageImageWriteWithoutFormat";
else if (!c->clip.depthClipControl) miss = "depthClipControl";
else if (!h->shaderInt64) miss = "shaderInt64";
else if (!h->samplerAnisotropy) miss = "samplerAnisotropy";
else if (!h->independentBlend) miss = "independentBlend";
else if (!h->geometryShader) miss = "geometryShader";
else if (!h->textureCompressionBC) miss = "textureCompressionBC";
if (miss) { rt_log("vk: required feature missing on %s: %s", g_vk.props.deviceName, miss); free(c); return 0; }
if (!c->fault.deviceFault) g_vk.has_device_fault = 0; // vkGetDeviceFaultInfoEXT needs the feature, not just the extension
g_vk.has_buffer_device_address = c->f12.bufferDeviceAddress;
g_vk.has_astc_ldr = h->textureCompressionASTC_LDR;
{
const char *e = cfg_get("TOTK_FEEDBACK");
int layout_ok = has_fbl && has_lr && c->fbl.attachmentFeedbackLoopLayout && c->lr.dynamicRenderingLocalRead;
g_vk.feedback = e && !strcmp(e, "0") ? VK_FEEDBACK_OFF
: layout_ok && !(e && !strcmp(e, "restart")) ? VK_FEEDBACK_LAYOUT : VK_FEEDBACK_RESTART;
rt_log("vk: feedback loops: %s (attachmentFeedbackLoopLayout %d, dynamicRenderingLocalRead %d)",
g_vk.feedback == VK_FEEDBACK_LAYOUT ? "feedback layout + by-region barrier"
: g_vk.feedback == VK_FEEDBACK_RESTART ? "pass restart + full barrier" : "off (TOTK_FEEDBACK=0)",
has_fbl && c->fbl.attachmentFeedbackLoopLayout, has_lr && c->lr.dynamicRenderingLocalRead);
}
float prio = 1.0f;
VkDeviceQueueCreateInfo qi[2] = {
{.sType = VK_STRUCTURE_TYPE_DEVICE_QUEUE_CREATE_INFO, .queueFamilyIndex = g_vk.gfx_family, .queueCount = 1,
.pQueuePriorities = &prio}};
uint32_t nqi = 1;
if (g_vk.xfer_family != g_vk.gfx_family) {
qi[1] = (VkDeviceQueueCreateInfo){.sType = VK_STRUCTURE_TYPE_DEVICE_QUEUE_CREATE_INFO,
.queueFamilyIndex = g_vk.xfer_family, .queueCount = 1, .pQueuePriorities = &prio};
nqi = 2;
}
VkDeviceCreateInfo di = {.sType = VK_STRUCTURE_TYPE_DEVICE_CREATE_INFO, .pNext = &c->f2, .queueCreateInfoCount = nqi,
.pQueueCreateInfos = qi, .enabledExtensionCount = n_ext, .ppEnabledExtensionNames = ext};
VkResult r = vkCreateDevice(g_vk.phys, &di, NULL, &g_vk.device);
free(c);
if (r != VK_SUCCESS) { rt_log("vk: vkCreateDevice failed (%d)", (int)r); return 0; }
vkGetDeviceQueue(g_vk.device, g_vk.gfx_family, 0, &g_vk.gfx_queue);
if (g_vk.xfer_family != g_vk.gfx_family) vkGetDeviceQueue(g_vk.device, g_vk.xfer_family, 0, &g_vk.xfer_queue);
else { g_vk.xfer_queue = g_vk.gfx_queue; g_vk.xfer_shared = 1; }
VkPhysicalDeviceExternalMemoryHostPropertiesEXT emh = {.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_EXTERNAL_MEMORY_HOST_PROPERTIES_EXT};
VkPhysicalDeviceProperties2 p2 = {.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2, .pNext = &emh};
vkGetPhysicalDeviceProperties2(g_vk.phys, &p2);
g_vk.min_imported_host_alignment = (uint32_t)emh.minImportedHostPointerAlignment;
vk_caps_init(&g_vk);
VmaAllocatorCreateInfo ai = {.physicalDevice = g_vk.phys, .device = g_vk.device, .instance = g_vk.instance,
.vulkanApiVersion = VK_API_VERSION_1_3,
.flags = (g_vk.has_buffer_device_address ? VMA_ALLOCATOR_CREATE_BUFFER_DEVICE_ADDRESS_BIT : 0) |
(g_vk.has_memory_budget ? VMA_ALLOCATOR_CREATE_EXT_MEMORY_BUDGET_BIT : 0),
.pDeviceMemoryCallbacks = vk_headroom_callbacks()};
if ((r = vmaCreateAllocator(&ai, &g_vk.vma)) != VK_SUCCESS) { rt_log("vk: vmaCreateAllocator failed (%d)", (int)r); return 0; }
VkSemaphoreTypeCreateInfo st = {.sType = VK_STRUCTURE_TYPE_SEMAPHORE_TYPE_CREATE_INFO,
.semaphoreType = VK_SEMAPHORE_TYPE_TIMELINE, .initialValue = 0};
VkSemaphoreCreateInfo si = {.sType = VK_STRUCTURE_TYPE_SEMAPHORE_CREATE_INFO, .pNext = &st};
if ((r = vkCreateSemaphore(g_vk.device, &si, NULL, &g_vk.timeline)) != VK_SUCCESS) {
rt_log("vk: timeline semaphore creation failed (%d)", (int)r);
return 0;
}
vk_crumbs_init(&g_vk);
return 1;
}
static long g_errors_total; // outlives the device so the probe can print it after vk_shutdown
static void destroy_all(void) {
gfx_window_release_vk();
if (g_vk.device) vkDeviceWaitIdle(g_vk.device);
if (g_vk.device) vk_astc_release();
if (g_vk.device) vk_deswz_release();
if (g_vk.device) vk_crumbs_release();
if (g_vk.timeline) vkDestroySemaphore(g_vk.device, g_vk.timeline, NULL);
vk_headroom_stop();
if (g_vk.vma) vmaDestroyAllocator(g_vk.vma);
if (g_vk.device) vkDestroyDevice(g_vk.device, NULL);
if (g_vk.messenger) {
PFN_vkDestroyDebugUtilsMessengerEXT f =
(PFN_vkDestroyDebugUtilsMessengerEXT)vkGetInstanceProcAddr(g_vk.instance, "vkDestroyDebugUtilsMessengerEXT");
if (f) f(g_vk.instance, g_vk.messenger, NULL);
}
if (g_vk.instance) vkDestroyInstance(g_vk.instance, NULL);
g_errors_total += g_vk.validation_errors;
memset(&g_vk, 0, sizeof g_vk);
}
long vk_validation_errors_total(void) { return g_errors_total + (g_ready ? g_vk.validation_errors : 0); }
int vk_init(void) {
plat_mutex_lock(&g_lock);
if (!g_ready) {
memset(&g_vk, 0, sizeof g_vk);
if (create_instance() && pick_device() && create_device()) g_ready = 1, vk_upload_pools_init(), vk_headroom_start(), vk_dlaa_init();
else destroy_all();
}
int ok = g_ready;
plat_mutex_unlock(&g_lock);
return ok;
}
void vk_shutdown(void) {
plat_mutex_lock(&g_lock);
if (g_ready) { destroy_all(); g_ready = 0; }
plat_mutex_unlock(&g_lock);
}
VkCtx *vk_ctx(void) { return g_ready ? &g_vk : NULL; }
static const char *fault_address_type(VkDeviceFaultAddressTypeEXT t) {
switch (t) {
case VK_DEVICE_FAULT_ADDRESS_TYPE_NONE_EXT: return "none";
case VK_DEVICE_FAULT_ADDRESS_TYPE_READ_INVALID_EXT: return "read invalid";
case VK_DEVICE_FAULT_ADDRESS_TYPE_WRITE_INVALID_EXT: return "write invalid";
case VK_DEVICE_FAULT_ADDRESS_TYPE_EXECUTE_INVALID_EXT: return "execute invalid";
case VK_DEVICE_FAULT_ADDRESS_TYPE_INSTRUCTION_POINTER_UNKNOWN_EXT: return "instruction pointer (unknown)";
case VK_DEVICE_FAULT_ADDRESS_TYPE_INSTRUCTION_POINTER_INVALID_EXT: return "instruction pointer (invalid)";
case VK_DEVICE_FAULT_ADDRESS_TYPE_INSTRUCTION_POINTER_FAULT_EXT: return "instruction pointer (fault)";
default: return "unknown type";
}
}
// The query runs once per process. Threads that hit the loss at the same time wait on the lock until it is logged, so
// none of them reaches its rt_fatal and ends the process before the report is written.
void vk_device_fault_report(VkResult r) {
static PlatMutex lock = PLAT_MUTEX_INIT;
static int done;
if (r != VK_ERROR_DEVICE_LOST) return;
plat_mutex_lock(&lock);
if (done) { plat_mutex_unlock(&lock); return; }
done = 1;
vk_crumbs_report();
vk_pass_marks_report();
VkCtx *c = vk_ctx();
PFN_vkGetDeviceFaultInfoEXT get =
c && c->has_device_fault ? (PFN_vkGetDeviceFaultInfoEXT)vkGetDeviceProcAddr(c->device, "vkGetDeviceFaultInfoEXT")
: NULL;
if (!get) {
rt_log("vk: device lost; no fault report (VK_EXT_device_fault %s)", c && c->has_device_fault ? "has no entry point"
: "not enabled");
plat_mutex_unlock(&lock);
return;
}
VkDeviceFaultCountsEXT n = {.sType = VK_STRUCTURE_TYPE_DEVICE_FAULT_COUNTS_EXT};
VkResult q = get(c->device, &n, NULL);
VkDeviceFaultAddressInfoEXT *ai = calloc(n.addressInfoCount + 1, sizeof *ai);
VkDeviceFaultVendorInfoEXT *vi = calloc(n.vendorInfoCount + 1, sizeof *vi);
VkDeviceFaultInfoEXT info = {.sType = VK_STRUCTURE_TYPE_DEVICE_FAULT_INFO_EXT, .pAddressInfos = ai, .pVendorInfos = vi};
uint64_t binary = n.vendorBinarySize;
n.vendorBinarySize = 0; // the vendor binary is not fetched; its size alone is logged
if (q == VK_SUCCESS && ai && vi) q = get(c->device, &n, &info);
if (q != VK_SUCCESS && q != VK_INCOMPLETE) {
rt_log("vk: device lost; vkGetDeviceFaultInfoEXT failed (%d)", (int)q);
} else {
rt_log("vk: device fault: \"%s\" (%u address infos, %u vendor infos, vendor binary %llu bytes)", info.description,
n.addressInfoCount, n.vendorInfoCount, (unsigned long long)binary);
// The faulting access lies in the precision-aligned range around the reported address.
for (uint32_t i = 0; i < n.addressInfoCount; i++)
rt_log("vk: device fault address %u: %s at %#llx, precision %#llx", i, fault_address_type(ai[i].addressType),
(unsigned long long)ai[i].reportedAddress, (unsigned long long)ai[i].addressPrecision);
for (uint32_t i = 0; i < n.vendorInfoCount; i++)
rt_log("vk: device fault vendor info %u: \"%s\" code %#llx data %#llx", i, vi[i].description,
(unsigned long long)vi[i].vendorFaultCode, (unsigned long long)vi[i].vendorFaultData);
}
free(ai);
free(vi);
plat_mutex_unlock(&lock);
}
int vk_selftest(void) {
if (!vk_init()) return 1;
VkCtx *c = &g_vk;
VkBufferCreateInfo bi = {.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO, .size = 64 << 10,
.usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_BUFFER_USAGE_STORAGE_BUFFER_BIT};
VmaAllocationCreateInfo aci = {.usage = VMA_MEMORY_USAGE_AUTO, .flags = VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT};
VkBuffer buf;
VmaAllocation alloc;
VkResult r = vmaCreateBuffer(c->vma, &bi, &aci, &buf, &alloc, NULL);
if (r != VK_SUCCESS) { rt_log("vk selftest: vmaCreateBuffer failed (%d)", (int)r); return 2; }
vmaDestroyBuffer(c->vma, buf, alloc);
VkSemaphoreSignalInfo sig = {.sType = VK_STRUCTURE_TYPE_SEMAPHORE_SIGNAL_INFO, .semaphore = c->timeline, .value = 1};
if ((r = vkSignalSemaphore(c->device, &sig)) != VK_SUCCESS) { rt_log("vk selftest: vkSignalSemaphore failed (%d)", (int)r); return 3; }
uint64_t value = 1;
VkSemaphoreWaitInfo wi = {.sType = VK_STRUCTURE_TYPE_SEMAPHORE_WAIT_INFO, .semaphoreCount = 1, .pSemaphores = &c->timeline,
.pValues = &value};
if ((r = vkWaitSemaphores(c->device, &wi, 1000000000ull)) != VK_SUCCESS) { rt_log("vk selftest: vkWaitSemaphores failed (%d)", (int)r); return 4; }
uint64_t now = 0;
vkGetSemaphoreCounterValue(c->device, c->timeline, &now);
if (now != 1) { rt_log("vk selftest: timeline value %llu, expected 1", (unsigned long long)now); return 5; }
return 0;
}
// --probe-vkdevice (host.h)
int probe_vkdevice(void) {
int rc = vk_selftest();
VkCtx *c = vk_ctx();
if (rc || !c) { printf("probe-vkdevice: FAILED (selftest %d)\n", rc ? rc : 1); vk_shutdown(); return 1; }
uint32_t v = c->props.apiVersion;
printf("probe-vkdevice: device %d \"%s\" api %u.%u.%u, gfx queue family %u, transfer family %u%s, host import alignment %#x\n",
c->device_index, c->props.deviceName, VK_API_VERSION_MAJOR(v), VK_API_VERSION_MINOR(v), VK_API_VERSION_PATCH(v),
c->gfx_family, c->xfer_family, c->xfer_shared ? " (shared queue)" : "", c->min_imported_host_alignment);
printf("probe-vkdevice: optional: custom_border=%d eds=%d eds2=%d eds3=%d vertex_input=%d cond_render=%d xfb=%d provoking=%d "
"robustness2=%d fault=%d budget=%d bda=%d interlock=%d buffer_marker=%d nv_checkpoints=%d validation=%d\n",
c->has_custom_border_color, c->has_eds, c->has_eds2, c->has_eds3, c->has_vertex_input_dynamic_state,
c->has_conditional_rendering, c->has_transform_feedback, c->has_provoking_vertex, c->has_robustness2,
c->has_device_fault, c->has_memory_budget, c->has_buffer_device_address, c->has_fragment_shader_interlock,
c->has_buffer_marker, c->has_nv_checkpoints, c->validation);
vk_shutdown();
printf("probe-vkdevice: %s (validation errors %ld)\n", g_errors_total ? "FAILED" : "passed", g_errors_total);
return g_errors_total ? 1 : 0;
}