From 8e2fb2e7109b68277041a0e0fbebe3e7614b91fb Mon Sep 17 00:00:00 2001
From: nihuini <nihuini@tencent.com>
Date: Wed, 22 May 2019 11:25:06 +0800
Subject: [PATCH] expose timestamp_period and timestamp_valid_bits

---
 src/gpu.cpp | 21 +++++++++++++--------
 src/gpu.h   |  4 ++++
 2 files changed, 17 insertions(+), 8 deletions(-)

diff --git a/src/gpu.cpp b/src/gpu.cpp
index 58dcecb59..4394e089f 100644
--- a/src/gpu.cpp
+++ b/src/gpu.cpp
@@ -530,6 +530,8 @@ int create_gpu_instance()
         gpu_info.memory_map_alignment = physicalDeviceProperties.limits.minMemoryMapAlignment;
         gpu_info.buffer_offset_alignment = physicalDeviceProperties.limits.minStorageBufferOffsetAlignment;
 
+        gpu_info.timestamp_period = physicalDeviceProperties.limits.timestampPeriod;
+
 //         fprintf(stderr, "[%u] max_shared_memory_size = %u\n", i, gpu_info.max_shared_memory_size);
 //         fprintf(stderr, "[%u] max_workgroup_count = %u %u %u\n", i, gpu_info.max_workgroup_count[0], gpu_info.max_workgroup_count[1], gpu_info.max_workgroup_count[2]);
 //         fprintf(stderr, "[%u] max_workgroup_invocations = %u\n", i, gpu_info.max_workgroup_invocations);
@@ -550,6 +552,9 @@ int create_gpu_instance()
         gpu_info.compute_queue_count = queueFamilyProperties[gpu_info.compute_queue_family_index].queueCount;
         gpu_info.transfer_queue_count = queueFamilyProperties[gpu_info.transfer_queue_family_index].queueCount;
 
+        // compute queue
+        gpu_info.timestamp_valid_bits = queueFamilyProperties[gpu_info.compute_queue_family_index].timestampValidBits;
+
         // find memory type index
         VkPhysicalDeviceMemoryProperties physicalDeviceMemoryProperties;
         vkGetPhysicalDeviceMemoryProperties(physicalDevice, &physicalDeviceMemoryProperties);
@@ -683,14 +688,14 @@ int create_gpu_instance()
 
             vkGetPhysicalDeviceFeatures2KHR(physicalDevice, &queryFeatures);
 
-//             if (gpu_info.support_VK_KHR_8bit_storage)
-//             {
-//                 gpu_info.support_int8_storage = query8BitStorageFeatures.storageBuffer8BitAccess && query8BitStorageFeatures.uniformAndStorageBuffer8BitAccess;
-//             }
-//             if (gpu_info.support_VK_KHR_16bit_storage)
-//             {
-//                 gpu_info.support_fp16_storage = query16BitStorageFeatures.storageBuffer16BitAccess && query16BitStorageFeatures.uniformAndStorageBuffer16BitAccess;
-//             }
+            if (gpu_info.support_VK_KHR_8bit_storage)
+            {
+                gpu_info.support_int8_storage = query8BitStorageFeatures.storageBuffer8BitAccess && query8BitStorageFeatures.uniformAndStorageBuffer8BitAccess;
+            }
+            if (gpu_info.support_VK_KHR_16bit_storage)
+            {
+                gpu_info.support_fp16_storage = query16BitStorageFeatures.storageBuffer16BitAccess && query16BitStorageFeatures.uniformAndStorageBuffer16BitAccess;
+            }
 //             if (gpu_info.support_VK_KHR_shader_float16_int8)
 //             {
 //                 gpu_info.support_fp16_arithmetic = queryFloat16Int8Features.shaderFloat16;
diff --git a/src/gpu.h b/src/gpu.h
index 3969d5507..4d9bb6f7a 100644
--- a/src/gpu.h
+++ b/src/gpu.h
@@ -71,6 +71,7 @@ public:
     uint32_t max_workgroup_size[3];
     size_t memory_map_alignment;
     size_t buffer_offset_alignment;
+    float timestamp_period;
 
     // runtime
     uint32_t compute_queue_family_index;
@@ -79,6 +80,9 @@ public:
     uint32_t compute_queue_count;
     uint32_t transfer_queue_count;
 
+    // compute queue only
+    uint32_t timestamp_valid_bits;
+
     uint32_t unified_memory_index;
     uint32_t device_local_memory_index;
     uint32_t host_visible_memory_index;