#include <vulkan/vulkan.h>
#include <vulkan/vk_layer.h>
+#include "git_sha1.h"
+
#include "imgui.h"
#include "overlay_params.h"
#include "util/list.h"
#include "util/ralloc.h"
#include "util/os_time.h"
+#include "util/os_socket.h"
#include "util/simple_mtx.h"
#include "vk_enum_to_str.h"
bool pipeline_statistics_enabled;
bool first_line_printed;
+
+ int control_client;
+
+ /* Dumping of frame stats to a file has been enabled. */
+ bool capture_enabled;
+
+ /* Dumping of frame stats to a file has been enabled and started. */
+ bool capture_started;
};
struct frame_stat {
VkCommandBuffer command_buffer;
+ VkSemaphore cross_engine_semaphore;
+
VkSemaphore semaphore;
VkFence fence;
}
#define HKEY(obj) ((uint64_t)(obj))
-#define FIND_SWAPCHAIN_DATA(obj) ((struct swapchain_data *)find_object_data(HKEY(obj)))
-#define FIND_CMD_BUFFER_DATA(obj) ((struct command_buffer_data *)find_object_data(HKEY(obj)))
-#define FIND_DEVICE_DATA(obj) ((struct device_data *)find_object_data(HKEY(obj)))
-#define FIND_QUEUE_DATA(obj) ((struct queue_data *)find_object_data(HKEY(obj)))
-#define FIND_PHYSICAL_DEVICE_DATA(obj) ((struct instance_data *)find_object_data(HKEY(obj)))
-#define FIND_INSTANCE_DATA(obj) ((struct instance_data *)find_object_data(HKEY(obj)))
+#define FIND(type, obj) ((type *)find_object_data(HKEY(obj)))
+
static void *find_object_data(uint64_t obj)
{
simple_mtx_lock(&vk_object_to_data_mutex);
/**/
-static void check_vk_result(VkResult err)
-{
- if (err != VK_SUCCESS)
- printf("ERROR!\n");
-}
-
static struct instance_data *new_instance_data(VkInstance instance)
{
struct instance_data *data = rzalloc(NULL, struct instance_data);
data->instance = instance;
+ data->control_client = -1;
map_object(HKEY(data->instance), data);
return data;
}
{
if (data->params.output_file)
fclose(data->params.output_file);
+ if (data->params.control >= 0)
+ os_socket_close(data->params.control);
unmap_object(HKEY(data->instance));
ralloc_free(data);
}
data->flags = family_props->queueFlags;
data->timestamp_mask = (1ull << family_props->timestampValidBits) - 1;
data->family_index = family_index;
- LIST_INITHEAD(&data->running_command_buffer);
+ list_inithead(&data->running_command_buffer);
map_object(HKEY(data->queue), data);
/* Fence synchronizing access to queries on that queue. */
VkFenceCreateInfo fence_info = {};
fence_info.sType = VK_STRUCTURE_TYPE_FENCE_CREATE_INFO;
fence_info.flags = VK_FENCE_CREATE_SIGNALED_BIT;
- VkResult err = device_data->vtable.CreateFence(device_data->device,
- &fence_info,
- NULL,
- &data->queries_fence);
- check_vk_result(err);
+ VK_CHECK(device_data->vtable.CreateFence(device_data->device,
+ &fence_info,
+ NULL,
+ &data->queries_fence));
if (data->flags & VK_QUEUE_GRAPHICS_BIT)
device_data->graphic_queue = data;
struct overlay_draw *get_overlay_draw(struct swapchain_data *data)
{
struct device_data *device_data = data->device;
- struct overlay_draw *draw = list_empty(&data->draws) ?
+ struct overlay_draw *draw = list_is_empty(&data->draws) ?
NULL : list_first_entry(&data->draws, struct overlay_draw, link);
VkSemaphoreCreateInfo sem_info = {};
VK_CHECK(device_data->vtable.CreateSemaphore(device_data->device, &sem_info,
NULL, &draw->semaphore));
+ VK_CHECK(device_data->vtable.CreateSemaphore(device_data->device, &sem_info,
+ NULL, &draw->cross_engine_semaphore));
list_addtail(&draw->link, &data->draws);
switch (param) {
case OVERLAY_PARAM_ENABLED_frame_timing:
case OVERLAY_PARAM_ENABLED_acquire_timing:
+ case OVERLAY_PARAM_ENABLED_present_timing:
return "(us)";
case OVERLAY_PARAM_ENABLED_gpu_timing:
return "(ns)";
}
}
+static void parse_command(struct instance_data *instance_data,
+ const char *cmd, unsigned cmdlen,
+ const char *param, unsigned paramlen)
+{
+ if (!strncmp(cmd, "capture", cmdlen)) {
+ int value = atoi(param);
+ bool enabled = value > 0;
+
+ if (enabled) {
+ instance_data->capture_enabled = true;
+ } else {
+ instance_data->capture_enabled = false;
+ instance_data->capture_started = false;
+ }
+ }
+}
+
+#define BUFSIZE 4096
+
+/**
+ * This function will process commands through the control file.
+ *
+ * A command starts with a colon, followed by the command, and followed by an
+ * option '=' and a parameter. It has to end with a semi-colon. A full command
+ * + parameter looks like:
+ *
+ * :cmd=param;
+ */
+static void process_char(struct instance_data *instance_data, char c)
+{
+ static char cmd[BUFSIZE];
+ static char param[BUFSIZE];
+
+ static unsigned cmdpos = 0;
+ static unsigned parampos = 0;
+ static bool reading_cmd = false;
+ static bool reading_param = false;
+
+ switch (c) {
+ case ':':
+ cmdpos = 0;
+ parampos = 0;
+ reading_cmd = true;
+ reading_param = false;
+ break;
+ case ';':
+ if (!reading_cmd)
+ break;
+ cmd[cmdpos++] = '\0';
+ param[parampos++] = '\0';
+ parse_command(instance_data, cmd, cmdpos, param, parampos);
+ reading_cmd = false;
+ reading_param = false;
+ break;
+ case '=':
+ if (!reading_cmd)
+ break;
+ reading_param = true;
+ break;
+ default:
+ if (!reading_cmd)
+ break;
+
+ if (reading_param) {
+ /* overflow means an invalid parameter */
+ if (parampos >= BUFSIZE - 1) {
+ reading_cmd = false;
+ reading_param = false;
+ break;
+ }
+
+ param[parampos++] = c;
+ } else {
+ /* overflow means an invalid command */
+ if (cmdpos >= BUFSIZE - 1) {
+ reading_cmd = false;
+ break;
+ }
+
+ cmd[cmdpos++] = c;
+ }
+ }
+}
+
+static void control_send(struct instance_data *instance_data,
+ const char *cmd, unsigned cmdlen,
+ const char *param, unsigned paramlen)
+{
+ unsigned msglen = 0;
+ char buffer[BUFSIZE];
+
+ assert(cmdlen + paramlen + 3 < BUFSIZE);
+
+ buffer[msglen++] = ':';
+
+ memcpy(&buffer[msglen], cmd, cmdlen);
+ msglen += cmdlen;
+
+ if (paramlen > 0) {
+ buffer[msglen++] = '=';
+ memcpy(&buffer[msglen], param, paramlen);
+ msglen += paramlen;
+ buffer[msglen++] = ';';
+ }
+
+ os_socket_send(instance_data->control_client, buffer, msglen, 0);
+}
+
+static void control_send_connection_string(struct device_data *device_data)
+{
+ struct instance_data *instance_data = device_data->instance;
+
+ const char *controlVersionCmd = "MesaOverlayControlVersion";
+ const char *controlVersionString = "1";
+
+ control_send(instance_data, controlVersionCmd, strlen(controlVersionCmd),
+ controlVersionString, strlen(controlVersionString));
+
+ const char *deviceCmd = "DeviceName";
+ const char *deviceName = device_data->properties.deviceName;
+
+ control_send(instance_data, deviceCmd, strlen(deviceCmd),
+ deviceName, strlen(deviceName));
+
+ const char *mesaVersionCmd = "MesaVersion";
+ const char *mesaVersionString = "Mesa " PACKAGE_VERSION MESA_GIT_SHA1;
+
+ control_send(instance_data, mesaVersionCmd, strlen(mesaVersionCmd),
+ mesaVersionString, strlen(mesaVersionString));
+}
+
+static void control_client_check(struct device_data *device_data)
+{
+ struct instance_data *instance_data = device_data->instance;
+
+ /* Already connected, just return. */
+ if (instance_data->control_client >= 0)
+ return;
+
+ int socket = os_socket_accept(instance_data->params.control);
+ if (socket == -1) {
+ if (errno != EAGAIN && errno != EWOULDBLOCK && errno != ECONNABORTED)
+ fprintf(stderr, "ERROR on socket: %s\n", strerror(errno));
+ return;
+ }
+
+ if (socket >= 0) {
+ os_socket_block(socket, false);
+ instance_data->control_client = socket;
+ control_send_connection_string(device_data);
+ }
+}
+
+static void control_client_disconnected(struct instance_data *instance_data)
+{
+ os_socket_close(instance_data->control_client);
+ instance_data->control_client = -1;
+}
+
+static void process_control_socket(struct instance_data *instance_data)
+{
+ const int client = instance_data->control_client;
+ if (client >= 0) {
+ char buf[BUFSIZE];
+
+ while (true) {
+ ssize_t n = os_socket_recv(client, buf, BUFSIZE, 0);
+
+ if (n == -1) {
+ if (errno == EAGAIN || errno == EWOULDBLOCK) {
+ /* nothing to read, try again later */
+ break;
+ }
+
+ if (errno != ECONNRESET)
+ fprintf(stderr, "ERROR on connection: %s\n", strerror(errno));
+
+ control_client_disconnected(instance_data);
+ } else if (n == 0) {
+ /* recv() returns 0 when the client disconnects */
+ control_client_disconnected(instance_data);
+ }
+
+ for (ssize_t i = 0; i < n; i++) {
+ process_char(instance_data, buf[i]);
+ }
+
+ /* If we try to read BUFSIZE and receive BUFSIZE bytes from the
+ * socket, there's a good chance that there's still more data to be
+ * read, so we will try again. Otherwise, simply be done for this
+ * iteration and try again on the next frame.
+ */
+ if (n < BUFSIZE)
+ break;
+ }
+ }
+}
+
static void snapshot_swapchain_frame(struct swapchain_data *data)
{
struct device_data *device_data = data->device;
uint32_t f_idx = data->n_frames % ARRAY_SIZE(data->frames_stats);
uint64_t now = os_time_get(); /* us */
+ if (instance_data->params.control >= 0) {
+ control_client_check(device_data);
+ process_control_socket(instance_data);
+ }
+
if (data->last_present_time) {
data->frame_stats.stats[OVERLAY_PARAM_ENABLED_frame_timing] =
now - data->last_present_time;
data->accumulated_stats.stats[s] += device_data->frame_stats.stats[s] + data->frame_stats.stats[s];
}
+ /* If capture has been enabled but it hasn't started yet, it means we are on
+ * the first snapshot after it has been enabled. At this point we want to
+ * use the stats captured so far to update the display, but we don't want
+ * this data to cause noise to the stats that we want to capture from now
+ * on.
+ *
+ * capture_begin == true will trigger an update of the fps on display, and a
+ * flush of the data, but no stats will be written to the output file. This
+ * way, we will have only stats from after the capture has been enabled
+ * written to the output_file.
+ */
+ const bool capture_begin =
+ instance_data->capture_enabled && !instance_data->capture_started;
+
if (data->last_fps_update) {
double elapsed = (double)(now - data->last_fps_update); /* us */
- if (elapsed >= instance_data->params.fps_sampling_period) {
+ if (capture_begin ||
+ elapsed >= instance_data->params.fps_sampling_period) {
data->fps = 1000000.0f * data->n_frames_since_update / elapsed;
- if (instance_data->params.output_file) {
+ if (instance_data->capture_started) {
if (!instance_data->first_line_printed) {
bool first_column = true;
memset(&data->accumulated_stats, 0, sizeof(data->accumulated_stats));
data->n_frames_since_update = 0;
data->last_fps_update = now;
+
+ if (capture_begin)
+ instance_data->capture_started = true;
}
} else {
data->last_fps_update = now;
if (s == OVERLAY_PARAM_ENABLED_frame_timing ||
s == OVERLAY_PARAM_ENABLED_acquire_timing ||
+ s == OVERLAY_PARAM_ENABLED_present_timing ||
s == OVERLAY_PARAM_ENABLED_gpu_timing) {
double min_time = data->stats_min.stats[s] / data->time_dividor;
double max_time = data->stats_max.stats[s] / data->time_dividor;
}
static struct overlay_draw *render_swapchain_display(struct swapchain_data *data,
+ struct queue_data *present_queue,
const VkSemaphore *wait_semaphores,
unsigned n_wait_semaphores,
unsigned image_index)
imb.subresourceRange.levelCount = 1;
imb.subresourceRange.baseArrayLayer = 0;
imb.subresourceRange.layerCount = 1;
- imb.srcQueueFamilyIndex = device_data->graphic_queue->family_index;
+ imb.srcQueueFamilyIndex = present_queue->family_index;
imb.dstQueueFamilyIndex = device_data->graphic_queue->family_index;
device_data->vtable.CmdPipelineBarrier(draw->command_buffer,
VK_PIPELINE_STAGE_ALL_GRAPHICS_BIT,
}
device_data->vtable.CmdEndRenderPass(draw->command_buffer);
+
+ if (device_data->graphic_queue->family_index != present_queue->family_index)
+ {
+ /* Transfer the image back to the present queue family
+ * image layout was already changed to present by the render pass
+ */
+ imb.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER;
+ imb.pNext = nullptr;
+ imb.srcAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT;
+ imb.dstAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT;
+ imb.oldLayout = VK_IMAGE_LAYOUT_PRESENT_SRC_KHR;
+ imb.newLayout = VK_IMAGE_LAYOUT_PRESENT_SRC_KHR;
+ imb.image = data->images[image_index];
+ imb.subresourceRange.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
+ imb.subresourceRange.baseMipLevel = 0;
+ imb.subresourceRange.levelCount = 1;
+ imb.subresourceRange.baseArrayLayer = 0;
+ imb.subresourceRange.layerCount = 1;
+ imb.srcQueueFamilyIndex = device_data->graphic_queue->family_index;
+ imb.dstQueueFamilyIndex = present_queue->family_index;
+ device_data->vtable.CmdPipelineBarrier(draw->command_buffer,
+ VK_PIPELINE_STAGE_ALL_GRAPHICS_BIT,
+ VK_PIPELINE_STAGE_ALL_GRAPHICS_BIT,
+ 0, /* dependency flags */
+ 0, nullptr, /* memory barriers */
+ 0, nullptr, /* buffer memory barriers */
+ 1, &imb); /* image memory barriers */
+ }
+
device_data->vtable.EndCommandBuffer(draw->command_buffer);
- VkSubmitInfo submit_info = {};
- VkPipelineStageFlags stage_wait = VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT;
- submit_info.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO;
- submit_info.commandBufferCount = 1;
- submit_info.pCommandBuffers = &draw->command_buffer;
- submit_info.pWaitDstStageMask = &stage_wait;
- submit_info.waitSemaphoreCount = n_wait_semaphores;
- submit_info.pWaitSemaphores = wait_semaphores;
- submit_info.signalSemaphoreCount = 1;
- submit_info.pSignalSemaphores = &draw->semaphore;
+ /* When presenting on a different queue than where we're drawing the
+ * overlay *AND* when the application does not provide a semaphore to
+ * vkQueuePresent, insert our own cross engine synchronization
+ * semaphore.
+ */
+ if (n_wait_semaphores == 0 && device_data->graphic_queue->queue != present_queue->queue) {
+ VkPipelineStageFlags stages_wait = VK_PIPELINE_STAGE_ALL_COMMANDS_BIT;
+ VkSubmitInfo submit_info = {};
+ submit_info.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO;
+ submit_info.commandBufferCount = 0;
+ submit_info.pWaitDstStageMask = &stages_wait;
+ submit_info.waitSemaphoreCount = 0;
+ submit_info.signalSemaphoreCount = 1;
+ submit_info.pSignalSemaphores = &draw->cross_engine_semaphore;
+
+ device_data->vtable.QueueSubmit(present_queue->queue, 1, &submit_info, VK_NULL_HANDLE);
+
+ submit_info.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO;
+ submit_info.commandBufferCount = 1;
+ submit_info.pWaitDstStageMask = &stages_wait;
+ submit_info.pCommandBuffers = &draw->command_buffer;
+ submit_info.waitSemaphoreCount = 1;
+ submit_info.pWaitSemaphores = &draw->cross_engine_semaphore;
+ submit_info.signalSemaphoreCount = 1;
+ submit_info.pSignalSemaphores = &draw->semaphore;
+
+ device_data->vtable.QueueSubmit(device_data->graphic_queue->queue, 1, &submit_info, draw->fence);
+ } else {
+ VkPipelineStageFlags *stages_wait = (VkPipelineStageFlags*) malloc(sizeof(VkPipelineStageFlags) * n_wait_semaphores);
+ for (unsigned i = 0; i < n_wait_semaphores; i++)
+ {
+ // wait in the fragment stage until the swapchain image is ready
+ stages_wait[i] = VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT;
+ }
+
+ VkSubmitInfo submit_info = {};
+ submit_info.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO;
+ submit_info.commandBufferCount = 1;
+ submit_info.pCommandBuffers = &draw->command_buffer;
+ submit_info.pWaitDstStageMask = stages_wait;
+ submit_info.waitSemaphoreCount = n_wait_semaphores;
+ submit_info.pWaitSemaphores = wait_semaphores;
+ submit_info.signalSemaphoreCount = 1;
+ submit_info.pSignalSemaphores = &draw->semaphore;
- device_data->vtable.QueueSubmit(device_data->graphic_queue->queue, 1, &submit_info, draw->fence);
+ device_data->vtable.QueueSubmit(device_data->graphic_queue->queue, 1, &submit_info, draw->fence);
+
+ free(stages_wait);
+ }
return draw;
}
struct device_data *device_data = data->device;
list_for_each_entry_safe(struct overlay_draw, draw, &data->draws, link) {
+ device_data->vtable.DestroySemaphore(device_data->device, draw->cross_engine_semaphore, NULL);
device_data->vtable.DestroySemaphore(device_data->device, draw->semaphore, NULL);
device_data->vtable.DestroyFence(device_data->device, draw->fence, NULL);
device_data->vtable.DestroyBuffer(device_data->device, draw->vertex_buffer, NULL);
}
static struct overlay_draw *before_present(struct swapchain_data *swapchain_data,
+ struct queue_data *present_queue,
const VkSemaphore *wait_semaphores,
unsigned n_wait_semaphores,
unsigned imageIndex)
if (!instance_data->params.no_display && swapchain_data->n_frames > 0) {
compute_swapchain_display(swapchain_data);
- draw = render_swapchain_display(swapchain_data,
+ draw = render_swapchain_display(swapchain_data, present_queue,
wait_semaphores, n_wait_semaphores,
imageIndex);
}
const VkAllocationCallbacks* pAllocator,
VkSwapchainKHR* pSwapchain)
{
- struct device_data *device_data = FIND_DEVICE_DATA(device);
+ struct device_data *device_data = FIND(struct device_data, device);
VkResult result = device_data->vtable.CreateSwapchainKHR(device, pCreateInfo, pAllocator, pSwapchain);
if (result != VK_SUCCESS) return result;
VkSwapchainKHR swapchain,
const VkAllocationCallbacks* pAllocator)
{
- struct swapchain_data *swapchain_data = FIND_SWAPCHAIN_DATA(swapchain);
+ struct swapchain_data *swapchain_data =
+ FIND(struct swapchain_data, swapchain);
shutdown_swapchain_data(swapchain_data);
swapchain_data->device->vtable.DestroySwapchainKHR(device, swapchain, pAllocator);
VkQueue queue,
const VkPresentInfoKHR* pPresentInfo)
{
- struct queue_data *queue_data = FIND_QUEUE_DATA(queue);
+ struct queue_data *queue_data = FIND(struct queue_data, queue);
struct device_data *device_data = queue_data->device;
struct instance_data *instance_data = device_data->instance;
uint32_t query_results[OVERLAY_QUERY_COUNT];
/* Before getting the query results, make sure the operations have
* completed.
*/
- VkResult err = device_data->vtable.ResetFences(device_data->device,
- 1, &queue_data->queries_fence);
- check_vk_result(err);
- err = device_data->vtable.QueueSubmit(queue, 0, NULL, queue_data->queries_fence);
- check_vk_result(err);
- err = device_data->vtable.WaitForFences(device_data->device,
- 1, &queue_data->queries_fence,
- VK_FALSE, UINT64_MAX);
- check_vk_result(err);
+ VK_CHECK(device_data->vtable.ResetFences(device_data->device,
+ 1, &queue_data->queries_fence));
+ VK_CHECK(device_data->vtable.QueueSubmit(queue, 0, NULL, queue_data->queries_fence));
+ VK_CHECK(device_data->vtable.WaitForFences(device_data->device,
+ 1, &queue_data->queries_fence,
+ VK_FALSE, UINT64_MAX));
/* Now get the results. */
list_for_each_entry_safe(struct command_buffer_data, cmd_buffer_data,
if (cmd_buffer_data->pipeline_query_pool) {
memset(query_results, 0, sizeof(query_results));
- err =
- device_data->vtable.GetQueryPoolResults(device_data->device,
- cmd_buffer_data->pipeline_query_pool,
- cmd_buffer_data->query_index, 1,
- sizeof(uint32_t) * OVERLAY_QUERY_COUNT,
- query_results, 0, VK_QUERY_RESULT_WAIT_BIT);
- check_vk_result(err);
+ VK_CHECK(device_data->vtable.GetQueryPoolResults(device_data->device,
+ cmd_buffer_data->pipeline_query_pool,
+ cmd_buffer_data->query_index, 1,
+ sizeof(uint32_t) * OVERLAY_QUERY_COUNT,
+ query_results, 0, VK_QUERY_RESULT_WAIT_BIT));
for (uint32_t i = OVERLAY_PARAM_ENABLED_vertices;
i <= OVERLAY_PARAM_ENABLED_compute_invocations; i++) {
}
if (cmd_buffer_data->timestamp_query_pool) {
uint64_t gpu_timestamps[2] = { 0 };
- err =
- device_data->vtable.GetQueryPoolResults(device_data->device,
- cmd_buffer_data->timestamp_query_pool,
- cmd_buffer_data->query_index * 2, 2,
- 2 * sizeof(uint64_t), gpu_timestamps, sizeof(uint64_t),
- VK_QUERY_RESULT_WAIT_BIT | VK_QUERY_RESULT_64_BIT);
- check_vk_result(err);
+ VK_CHECK(device_data->vtable.GetQueryPoolResults(device_data->device,
+ cmd_buffer_data->timestamp_query_pool,
+ cmd_buffer_data->query_index * 2, 2,
+ 2 * sizeof(uint64_t), gpu_timestamps, sizeof(uint64_t),
+ VK_QUERY_RESULT_WAIT_BIT | VK_QUERY_RESULT_64_BIT));
gpu_timestamps[0] &= queue_data->timestamp_mask;
gpu_timestamps[1] &= queue_data->timestamp_mask;
if (instance_data->params.no_display) {
for (uint32_t i = 0; i < pPresentInfo->swapchainCount; i++) {
VkSwapchainKHR swapchain = pPresentInfo->pSwapchains[i];
- struct swapchain_data *swapchain_data = FIND_SWAPCHAIN_DATA(swapchain);
+ struct swapchain_data *swapchain_data =
+ FIND(struct swapchain_data, swapchain);
+
+ uint32_t image_index = pPresentInfo->pImageIndices[i];
before_present(swapchain_data,
+ queue_data,
pPresentInfo->pWaitSemaphores,
pPresentInfo->waitSemaphoreCount,
- pPresentInfo->pImageIndices[i]);
+ image_index);
+
+ VkPresentInfoKHR present_info = *pPresentInfo;
+ present_info.swapchainCount = 1;
+ present_info.pSwapchains = &swapchain;
+ present_info.pImageIndices = &image_index;
+
+ uint64_t ts0 = os_time_get();
+ result = queue_data->device->vtable.QueuePresentKHR(queue, &present_info);
+ uint64_t ts1 = os_time_get();
+ swapchain_data->frame_stats.stats[OVERLAY_PARAM_ENABLED_present_timing] += ts1 - ts0;
}
- result = queue_data->device->vtable.QueuePresentKHR(queue, pPresentInfo);
} else {
for (uint32_t i = 0; i < pPresentInfo->swapchainCount; i++) {
VkSwapchainKHR swapchain = pPresentInfo->pSwapchains[i];
- struct swapchain_data *swapchain_data = FIND_SWAPCHAIN_DATA(swapchain);
+ struct swapchain_data *swapchain_data =
+ FIND(struct swapchain_data, swapchain);
+
+ uint32_t image_index = pPresentInfo->pImageIndices[i];
+
VkPresentInfoKHR present_info = *pPresentInfo;
present_info.swapchainCount = 1;
present_info.pSwapchains = &swapchain;
-
- uint32_t image_index = pPresentInfo->pImageIndices[i];
+ present_info.pImageIndices = &image_index;
struct overlay_draw *draw = before_present(swapchain_data,
+ queue_data,
pPresentInfo->pWaitSemaphores,
pPresentInfo->waitSemaphoreCount,
image_index);
present_info.pWaitSemaphores = &draw->semaphore;
present_info.waitSemaphoreCount = 1;
+ uint64_t ts0 = os_time_get();
VkResult chain_result = queue_data->device->vtable.QueuePresentKHR(queue, &present_info);
+ uint64_t ts1 = os_time_get();
+ swapchain_data->frame_stats.stats[OVERLAY_PARAM_ENABLED_present_timing] += ts1 - ts0;
if (pPresentInfo->pResults)
pPresentInfo->pResults[i] = chain_result;
if (chain_result != VK_SUCCESS && result == VK_SUCCESS)
VkFence fence,
uint32_t* pImageIndex)
{
- struct swapchain_data *swapchain_data = FIND_SWAPCHAIN_DATA(swapchain);
+ struct swapchain_data *swapchain_data =
+ FIND(struct swapchain_data, swapchain);
struct device_data *device_data = swapchain_data->device;
uint64_t ts0 = os_time_get();
const VkAcquireNextImageInfoKHR* pAcquireInfo,
uint32_t* pImageIndex)
{
- struct swapchain_data *swapchain_data = FIND_SWAPCHAIN_DATA(pAcquireInfo->swapchain);
+ struct swapchain_data *swapchain_data =
+ FIND(struct swapchain_data, pAcquireInfo->swapchain);
struct device_data *device_data = swapchain_data->device;
uint64_t ts0 = os_time_get();
uint32_t firstVertex,
uint32_t firstInstance)
{
- struct command_buffer_data *cmd_buffer_data = FIND_CMD_BUFFER_DATA(commandBuffer);
+ struct command_buffer_data *cmd_buffer_data =
+ FIND(struct command_buffer_data, commandBuffer);
cmd_buffer_data->stats.stats[OVERLAY_PARAM_ENABLED_draw]++;
struct device_data *device_data = cmd_buffer_data->device;
device_data->vtable.CmdDraw(commandBuffer, vertexCount, instanceCount,
int32_t vertexOffset,
uint32_t firstInstance)
{
- struct command_buffer_data *cmd_buffer_data = FIND_CMD_BUFFER_DATA(commandBuffer);
+ struct command_buffer_data *cmd_buffer_data =
+ FIND(struct command_buffer_data, commandBuffer);
cmd_buffer_data->stats.stats[OVERLAY_PARAM_ENABLED_draw_indexed]++;
struct device_data *device_data = cmd_buffer_data->device;
device_data->vtable.CmdDrawIndexed(commandBuffer, indexCount, instanceCount,
uint32_t drawCount,
uint32_t stride)
{
- struct command_buffer_data *cmd_buffer_data = FIND_CMD_BUFFER_DATA(commandBuffer);
+ struct command_buffer_data *cmd_buffer_data =
+ FIND(struct command_buffer_data, commandBuffer);
cmd_buffer_data->stats.stats[OVERLAY_PARAM_ENABLED_draw_indirect]++;
struct device_data *device_data = cmd_buffer_data->device;
device_data->vtable.CmdDrawIndirect(commandBuffer, buffer, offset, drawCount, stride);
uint32_t drawCount,
uint32_t stride)
{
- struct command_buffer_data *cmd_buffer_data = FIND_CMD_BUFFER_DATA(commandBuffer);
+ struct command_buffer_data *cmd_buffer_data =
+ FIND(struct command_buffer_data, commandBuffer);
cmd_buffer_data->stats.stats[OVERLAY_PARAM_ENABLED_draw_indexed_indirect]++;
struct device_data *device_data = cmd_buffer_data->device;
device_data->vtable.CmdDrawIndexedIndirect(commandBuffer, buffer, offset, drawCount, stride);
}
-static void overlay_CmdDrawIndirectCountKHR(
+static void overlay_CmdDrawIndirectCount(
VkCommandBuffer commandBuffer,
VkBuffer buffer,
VkDeviceSize offset,
uint32_t maxDrawCount,
uint32_t stride)
{
- struct command_buffer_data *cmd_buffer_data = FIND_CMD_BUFFER_DATA(commandBuffer);
+ struct command_buffer_data *cmd_buffer_data =
+ FIND(struct command_buffer_data, commandBuffer);
cmd_buffer_data->stats.stats[OVERLAY_PARAM_ENABLED_draw_indirect_count]++;
struct device_data *device_data = cmd_buffer_data->device;
- device_data->vtable.CmdDrawIndirectCountKHR(commandBuffer, buffer, offset,
- countBuffer, countBufferOffset,
- maxDrawCount, stride);
+ device_data->vtable.CmdDrawIndirectCount(commandBuffer, buffer, offset,
+ countBuffer, countBufferOffset,
+ maxDrawCount, stride);
}
-static void overlay_CmdDrawIndexedIndirectCountKHR(
+static void overlay_CmdDrawIndexedIndirectCount(
VkCommandBuffer commandBuffer,
VkBuffer buffer,
VkDeviceSize offset,
uint32_t maxDrawCount,
uint32_t stride)
{
- struct command_buffer_data *cmd_buffer_data = FIND_CMD_BUFFER_DATA(commandBuffer);
+ struct command_buffer_data *cmd_buffer_data =
+ FIND(struct command_buffer_data, commandBuffer);
cmd_buffer_data->stats.stats[OVERLAY_PARAM_ENABLED_draw_indexed_indirect_count]++;
struct device_data *device_data = cmd_buffer_data->device;
- device_data->vtable.CmdDrawIndexedIndirectCountKHR(commandBuffer, buffer, offset,
- countBuffer, countBufferOffset,
- maxDrawCount, stride);
+ device_data->vtable.CmdDrawIndexedIndirectCount(commandBuffer, buffer, offset,
+ countBuffer, countBufferOffset,
+ maxDrawCount, stride);
}
static void overlay_CmdDispatch(
uint32_t groupCountY,
uint32_t groupCountZ)
{
- struct command_buffer_data *cmd_buffer_data = FIND_CMD_BUFFER_DATA(commandBuffer);
+ struct command_buffer_data *cmd_buffer_data =
+ FIND(struct command_buffer_data, commandBuffer);
cmd_buffer_data->stats.stats[OVERLAY_PARAM_ENABLED_dispatch]++;
struct device_data *device_data = cmd_buffer_data->device;
device_data->vtable.CmdDispatch(commandBuffer, groupCountX, groupCountY, groupCountZ);
VkBuffer buffer,
VkDeviceSize offset)
{
- struct command_buffer_data *cmd_buffer_data = FIND_CMD_BUFFER_DATA(commandBuffer);
+ struct command_buffer_data *cmd_buffer_data =
+ FIND(struct command_buffer_data, commandBuffer);
cmd_buffer_data->stats.stats[OVERLAY_PARAM_ENABLED_dispatch_indirect]++;
struct device_data *device_data = cmd_buffer_data->device;
device_data->vtable.CmdDispatchIndirect(commandBuffer, buffer, offset);
VkPipelineBindPoint pipelineBindPoint,
VkPipeline pipeline)
{
- struct command_buffer_data *cmd_buffer_data = FIND_CMD_BUFFER_DATA(commandBuffer);
+ struct command_buffer_data *cmd_buffer_data =
+ FIND(struct command_buffer_data, commandBuffer);
switch (pipelineBindPoint) {
case VK_PIPELINE_BIND_POINT_GRAPHICS: cmd_buffer_data->stats.stats[OVERLAY_PARAM_ENABLED_pipeline_graphics]++; break;
case VK_PIPELINE_BIND_POINT_COMPUTE: cmd_buffer_data->stats.stats[OVERLAY_PARAM_ENABLED_pipeline_compute]++; break;
VkCommandBuffer commandBuffer,
const VkCommandBufferBeginInfo* pBeginInfo)
{
- struct command_buffer_data *cmd_buffer_data = FIND_CMD_BUFFER_DATA(commandBuffer);
+ struct command_buffer_data *cmd_buffer_data =
+ FIND(struct command_buffer_data, commandBuffer);
struct device_data *device_data = cmd_buffer_data->device;
+ memset(&cmd_buffer_data->stats, 0, sizeof(cmd_buffer_data->stats));
+
/* We don't record any query in secondary command buffers, just make sure
* we have the right inheritance.
*/
return result;
}
- /* Primary command buffers with no queries. */
- if (!cmd_buffer_data->pipeline_query_pool && cmd_buffer_data->timestamp_query_pool)
- return device_data->vtable.BeginCommandBuffer(commandBuffer, pBeginInfo);
-
/* Otherwise record a begin query as first command. */
VkResult result = device_data->vtable.BeginCommandBuffer(commandBuffer, pBeginInfo);
static VkResult overlay_EndCommandBuffer(
VkCommandBuffer commandBuffer)
{
- struct command_buffer_data *cmd_buffer_data = FIND_CMD_BUFFER_DATA(commandBuffer);
+ struct command_buffer_data *cmd_buffer_data =
+ FIND(struct command_buffer_data, commandBuffer);
struct device_data *device_data = cmd_buffer_data->device;
if (cmd_buffer_data->timestamp_query_pool) {
VkCommandBuffer commandBuffer,
VkCommandBufferResetFlags flags)
{
- struct command_buffer_data *cmd_buffer_data = FIND_CMD_BUFFER_DATA(commandBuffer);
+ struct command_buffer_data *cmd_buffer_data =
+ FIND(struct command_buffer_data, commandBuffer);
struct device_data *device_data = cmd_buffer_data->device;
memset(&cmd_buffer_data->stats, 0, sizeof(cmd_buffer_data->stats));
uint32_t commandBufferCount,
const VkCommandBuffer* pCommandBuffers)
{
- struct command_buffer_data *cmd_buffer_data = FIND_CMD_BUFFER_DATA(commandBuffer);
+ struct command_buffer_data *cmd_buffer_data =
+ FIND(struct command_buffer_data, commandBuffer);
struct device_data *device_data = cmd_buffer_data->device;
/* Add the stats of the executed command buffers to the primary one. */
for (uint32_t c = 0; c < commandBufferCount; c++) {
- struct command_buffer_data *sec_cmd_buffer_data = FIND_CMD_BUFFER_DATA(pCommandBuffers[c]);
+ struct command_buffer_data *sec_cmd_buffer_data =
+ FIND(struct command_buffer_data, pCommandBuffers[c]);
for (uint32_t s = 0; s < OVERLAY_PARAM_ENABLED_MAX; s++)
cmd_buffer_data->stats.stats[s] += sec_cmd_buffer_data->stats.stats[s];
const VkCommandBufferAllocateInfo* pAllocateInfo,
VkCommandBuffer* pCommandBuffers)
{
- struct device_data *device_data = FIND_DEVICE_DATA(device);
+ struct device_data *device_data = FIND(struct device_data, device);
VkResult result =
device_data->vtable.AllocateCommandBuffers(device, pAllocateInfo, pCommandBuffers);
if (result != VK_SUCCESS)
pAllocateInfo->commandBufferCount,
overlay_query_flags,
};
- VkResult err =
- device_data->vtable.CreateQueryPool(device_data->device, &pool_info,
- NULL, &pipeline_query_pool);
- check_vk_result(err);
+ VK_CHECK(device_data->vtable.CreateQueryPool(device_data->device, &pool_info,
+ NULL, &pipeline_query_pool));
}
if (device_data->instance->params.enabled[OVERLAY_PARAM_ENABLED_gpu_timing]) {
VkQueryPoolCreateInfo pool_info = {
pAllocateInfo->commandBufferCount * 2,
0,
};
- VkResult err =
- device_data->vtable.CreateQueryPool(device_data->device, &pool_info,
- NULL, ×tamp_query_pool);
- check_vk_result(err);
+ VK_CHECK(device_data->vtable.CreateQueryPool(device_data->device, &pool_info,
+ NULL, ×tamp_query_pool));
}
for (uint32_t i = 0; i < pAllocateInfo->commandBufferCount; i++) {
uint32_t commandBufferCount,
const VkCommandBuffer* pCommandBuffers)
{
- struct device_data *device_data = FIND_DEVICE_DATA(device);
+ struct device_data *device_data = FIND(struct device_data, device);
for (uint32_t i = 0; i < commandBufferCount; i++) {
struct command_buffer_data *cmd_buffer_data =
- FIND_CMD_BUFFER_DATA(pCommandBuffers[i]);
+ FIND(struct command_buffer_data, pCommandBuffers[i]);
+
+ /* It is legal to free a NULL command buffer*/
+ if (!cmd_buffer_data)
+ continue;
+
uint64_t count = (uintptr_t)find_object_data(HKEY(cmd_buffer_data->pipeline_query_pool));
if (count == 1) {
unmap_object(HKEY(cmd_buffer_data->pipeline_query_pool));
const VkSubmitInfo* pSubmits,
VkFence fence)
{
- struct queue_data *queue_data = FIND_QUEUE_DATA(queue);
+ struct queue_data *queue_data = FIND(struct queue_data, queue);
struct device_data *device_data = queue_data->device;
device_data->frame_stats.stats[OVERLAY_PARAM_ENABLED_submit]++;
for (uint32_t s = 0; s < submitCount; s++) {
for (uint32_t c = 0; c < pSubmits[s].commandBufferCount; c++) {
struct command_buffer_data *cmd_buffer_data =
- FIND_CMD_BUFFER_DATA(pSubmits[s].pCommandBuffers[c]);
+ FIND(struct command_buffer_data, pSubmits[s].pCommandBuffers[c]);
/* Merge the submitted command buffer stats into the device. */
for (uint32_t st = 0; st < OVERLAY_PARAM_ENABLED_MAX; st++)
!cmd_buffer_data->timestamp_query_pool)
continue;
- if (list_empty(&cmd_buffer_data->link)) {
+ if (list_is_empty(&cmd_buffer_data->link)) {
list_addtail(&cmd_buffer_data->link,
&queue_data->running_command_buffer);
} else {
const VkAllocationCallbacks* pAllocator,
VkDevice* pDevice)
{
- struct instance_data *instance_data = FIND_PHYSICAL_DEVICE_DATA(physicalDevice);
+ struct instance_data *instance_data =
+ FIND(struct instance_data, physicalDevice);
VkLayerDeviceCreateInfo *chain_info =
get_device_chain_info(pCreateInfo, VK_LAYER_LINK_INFO);
VkDevice device,
const VkAllocationCallbacks* pAllocator)
{
- struct device_data *device_data = FIND_DEVICE_DATA(device);
+ struct device_data *device_data = FIND(struct device_data, device);
device_unmap_queues(device_data);
device_data->vtable.DestroyDevice(device, pAllocator);
destroy_device_data(device_data);
parse_overlay_env(&instance_data->params, getenv("VK_LAYER_MESA_OVERLAY_CONFIG"));
+ /* If there's no control file, and an output_file was specified, start
+ * capturing fps data right away.
+ */
+ instance_data->capture_enabled =
+ instance_data->params.output_file && instance_data->params.control < 0;
+ instance_data->capture_started = instance_data->capture_enabled;
+
for (int i = OVERLAY_PARAM_ENABLED_vertices;
i <= OVERLAY_PARAM_ENABLED_compute_invocations; i++) {
if (instance_data->params.enabled[i]) {
VkInstance instance,
const VkAllocationCallbacks* pAllocator)
{
- struct instance_data *instance_data = FIND_INSTANCE_DATA(instance);
+ struct instance_data *instance_data = FIND(struct instance_data, instance);
instance_data_map_physical_devices(instance_data, false);
instance_data->vtable.DestroyInstance(instance, pAllocator);
destroy_instance_data(instance_data);
} name_to_funcptr_map[] = {
{ "vkGetDeviceProcAddr", (void *) vkGetDeviceProcAddr },
#define ADD_HOOK(fn) { "vk" # fn, (void *) overlay_ ## fn }
+#define ADD_ALIAS_HOOK(alias, fn) { "vk" # alias, (void *) overlay_ ## fn }
ADD_HOOK(AllocateCommandBuffers),
ADD_HOOK(FreeCommandBuffers),
ADD_HOOK(ResetCommandBuffer),
ADD_HOOK(CmdDrawIndexedIndirect),
ADD_HOOK(CmdDispatch),
ADD_HOOK(CmdDispatchIndirect),
- ADD_HOOK(CmdDrawIndirectCountKHR),
- ADD_HOOK(CmdDrawIndexedIndirectCountKHR),
+ ADD_HOOK(CmdDrawIndirectCount),
+ ADD_ALIAS_HOOK(CmdDrawIndirectCountKHR, CmdDrawIndirectCount),
+ ADD_HOOK(CmdDrawIndexedIndirectCount),
+ ADD_ALIAS_HOOK(CmdDrawIndexedIndirectCountKHR, CmdDrawIndexedIndirectCount),
ADD_HOOK(CmdBindPipeline),
if (dev == NULL) return NULL;
- struct device_data *device_data = FIND_DEVICE_DATA(dev);
+ struct device_data *device_data = FIND(struct device_data, dev);
if (device_data->vtable.GetDeviceProcAddr == NULL) return NULL;
return device_data->vtable.GetDeviceProcAddr(dev, funcName);
}
if (instance == NULL) return NULL;
- struct instance_data *instance_data = FIND_INSTANCE_DATA(instance);
+ struct instance_data *instance_data = FIND(struct instance_data, instance);
if (instance_data->vtable.GetInstanceProcAddr == NULL) return NULL;
return instance_data->vtable.GetInstanceProcAddr(instance, funcName);
}
projects / mesa.git / blobdiff