Moved dispatch commands to a different command buffer, still observing issue with no validation errors

client/include/gpu.h

@@ -79,11 +79,13 @@ typedef struct FrameContextStruct {
   VkSemaphore image;
   VkSemaphore render;
   VkSemaphore transfer;
+  VkSemaphore compute;
   VkSemaphore frame;
 
   uint64_t frame_index;
-  uint64_t transfer_index;
+  uint64_t compute_index;
 
+  VkCommandBuffer compute_commands;
   VkCommandBuffer transfer_commands;
   VkBuffer        transfer_buffer;
   VmaAllocation   transfer_memory;

client/src/draw.c

@@ -71,38 +71,24 @@ VkResult draw_frame(
     VK_RESULT(vkBeginCommandBuffer(transfer_commands, &begin_info));
 
     VkDeviceSize src_offset = 0;
-    for(uint32_t transfer_index = 0; transfer_index < context->frame[context->current_frame].transfer_count; transfer_index++) {
+    for(uint32_t tid = 0; tid < context->frame[context->current_frame].transfer_count; tid++) {
       command_copy_buffer(
           transfer_commands,
           context->frame[context->current_frame].transfer_buffer,
-          context->frame[context->current_frame].transfer_infos[transfer_index].buffer,
+          context->frame[context->current_frame].transfer_infos[tid].buffer,
           src_offset,
-          context->frame[context->current_frame].transfer_infos[transfer_index].offset,
+          context->frame[context->current_frame].transfer_infos[tid].offset,
-          context->frame[context->current_frame].transfer_infos[transfer_index].size);
+          context->frame[context->current_frame].transfer_infos[tid].size);
-      src_offset += context->frame[context->current_frame].transfer_infos[transfer_index].size;
+      src_offset += context->frame[context->current_frame].transfer_infos[tid].size;
-      VkBufferMemoryBarrier barrier = {
-        .sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER,
-        .buffer = context->frame[context->current_frame].transfer_infos[transfer_index].buffer,
-        .offset = context->frame[context->current_frame].transfer_infos[transfer_index].offset,
-        .size = context->frame[context->current_frame].transfer_infos[transfer_index].size,
-        .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
-        .dstAccessMask = VK_ACCESS_INDIRECT_COMMAND_READ_BIT | VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT | VK_ACCESS_SHADER_READ_BIT,
-      };
-      vkCmdPipelineBarrier(transfer_commands, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT | VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT, 0, 0, NULL, 1, &barrier, 0, NULL);
     }
 
-    record_ui_compute(transfer_commands, ui, context->current_frame);
     VK_RESULT(vkEndCommandBuffer(transfer_commands));
-    VkPipelineStageFlags wait_stages[] = {VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT};
+    VkPipelineStageFlags wait_stages[] = {VK_PIPELINE_STAGE_ALL_COMMANDS_BIT};
-    context->frame[context->current_frame].transfer_index += 1;
     VkSemaphore transfer_signals[] = {context->frame[context->current_frame].transfer};
-    uint64_t transfer_signal_values[] = {context->frame[context->current_frame].transfer_index};
     VkSemaphore transfer_waits[] = {context->frame[context->current_frame].frame};
     uint64_t transfer_wait_values[] = {context->frame[context->current_frame].frame_index};
     VkTimelineSemaphoreSubmitInfo timeline_info = {
       .sType = VK_STRUCTURE_TYPE_TIMELINE_SEMAPHORE_SUBMIT_INFO,
-      .signalSemaphoreValueCount = sizeof(transfer_signal_values)/sizeof(uint64_t),
-      .pSignalSemaphoreValues = transfer_signal_values,
       .waitSemaphoreValueCount = sizeof(transfer_wait_values)/sizeof(uint64_t),
       .pWaitSemaphoreValues = transfer_wait_values,
     };
@@ -120,6 +106,35 @@ VkResult draw_frame(
     VK_RESULT(vkQueueSubmit(context->transfer_queue.handle, 1, &submit_info, VK_NULL_HANDLE));
     context->frame[context->current_frame].transfer_count = 0;
     context->frame[context->current_frame].transfer_written = 0;
+
+
+    VkCommandBuffer compute_commands = context->frame[context->current_frame].compute_commands;
+    VK_RESULT(vkResetCommandBuffer(compute_commands, 0));
+    VK_RESULT(vkBeginCommandBuffer(compute_commands, &begin_info));
+    record_ui_compute(compute_commands, ui, context->current_frame);
+    VK_RESULT(vkEndCommandBuffer(compute_commands));
+    VkPipelineStageFlags compute_wait_stages[] = {VK_PIPELINE_STAGE_ALL_COMMANDS_BIT};
+    context->frame[context->current_frame].compute_index += 1;
+    VkSemaphore compute_signals[] = {context->frame[context->current_frame].compute};
+    uint64_t compute_signal_values[] = {context->frame[context->current_frame].compute_index};
+    VkSemaphore compute_waits[] = {context->frame[context->current_frame].transfer};
+    VkTimelineSemaphoreSubmitInfo compute_timeline = {
+      .sType = VK_STRUCTURE_TYPE_TIMELINE_SEMAPHORE_SUBMIT_INFO,
+      .signalSemaphoreValueCount = sizeof(compute_signal_values)/sizeof(uint64_t),
+      .pSignalSemaphoreValues = compute_signal_values,
+    };
+    VkSubmitInfo compute_submit = {
+      .sType = VK_STRUCTURE_TYPE_SUBMIT_INFO,
+      .commandBufferCount = 1,
+      .pCommandBuffers = &compute_commands,
+      .pSignalSemaphores = compute_signals,
+      .signalSemaphoreCount = sizeof(compute_signals)/sizeof(VkSemaphore),
+      .pWaitSemaphores = compute_waits,
+      .pWaitDstStageMask = compute_wait_stages,
+      .waitSemaphoreCount = sizeof(compute_waits)/sizeof(VkSemaphore),
+      .pNext = &compute_timeline,
+    };
+    VK_RESULT(vkQueueSubmit(context->transfer_queue.handle, 1, &compute_submit, VK_NULL_HANDLE));
   }
 
   uint32_t image_index;
@@ -165,10 +180,10 @@ VkResult draw_frame(
   VK_RESULT(vkEndCommandBuffer(command_buffer));
 
   VkPipelineStageFlags wait_stages[] = {VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT, VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT};
-  VkSemaphore wait_semaphores[] = {context->frame[context->current_frame].image, context->frame[context->current_frame].transfer};
+  VkSemaphore wait_semaphores[] = {context->frame[context->current_frame].image, context->frame[context->current_frame].compute};
   VkSemaphore signal_semaphores[] = {context->frame[context->current_frame].render, context->frame[context->current_frame].frame};
   context->frame[context->current_frame].frame_index += 1;
-  uint64_t wait_values[] = {0, context->frame[context->current_frame].transfer_index};
+  uint64_t wait_values[] = {0, context->frame[context->current_frame].compute_index};
   uint64_t signal_values[] = {0, context->frame[context->current_frame].frame_index};
   VkTimelineSemaphoreSubmitInfo timeline_info = {
     .sType = VK_STRUCTURE_TYPE_TIMELINE_SEMAPHORE_SUBMIT_INFO,

client/src/gpu.c

@@ -781,11 +781,12 @@ VkResult create_frame_context(VkDevice device, VmaAllocator allocator, VkCommand
 
   frame->image = create_semaphore(device);
   frame->render = create_semaphore(device);
-  frame->transfer = create_timeline_semaphore(device);
+  frame->transfer = create_semaphore(device);
   frame->frame = create_timeline_semaphore(device);
+  frame->compute = create_timeline_semaphore(device);
 
   frame->frame_index = 0;
-  frame->transfer_index = 0;
+  frame->compute_index = 0;
 
   VkCommandBufferAllocateInfo command_info = {
     .sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_ALLOCATE_INFO,
@@ -794,6 +795,7 @@ VkResult create_frame_context(VkDevice device, VmaAllocator allocator, VkCommand
     .level = VK_COMMAND_BUFFER_LEVEL_PRIMARY,
   };
   VK_RESULT(vkAllocateCommandBuffers(device, &command_info, &frame->transfer_commands));
+  VK_RESULT(vkAllocateCommandBuffers(device, &command_info, &frame->compute_commands));
 
   // TODO: better defaults
   frame->transfer_max_size = 1;