From 53e3eee9eeac184125cb9003ed00ba58a12748d9 Mon Sep 17 00:00:00 2001
From: Bartosz Taudul <wolf.pld@gmail.com>
Date: Sun, 17 Jun 2018 20:56:46 +0200
Subject: [PATCH] Delay query until results are available.

This will query for the new values with a bit of lag, but it ignores
issues with broken drivers...
---
 TracyVulkan.hpp | 39 ++++++++++++++++-----------------------
 1 file changed, 16 insertions(+), 23 deletions(-)

diff --git a/TracyVulkan.hpp b/TracyVulkan.hpp
index 1e4fdace..1dcb9b83 100644
--- a/TracyVulkan.hpp
+++ b/TracyVulkan.hpp
@@ -40,6 +40,7 @@ public:
         , m_context( s_vkCtxCounter.fetch_add( 1, std::memory_order_relaxed ) )
         , m_head( 0 )
         , m_tail( 0 )
+        , m_oldCnt( 0 )
     {
         VkPhysicalDeviceProperties prop;
         vkGetPhysicalDeviceProperties( physdev, &prop );
@@ -107,31 +108,22 @@ public:
 
         if( m_tail == m_head ) return;
 
-        auto cnt = m_head < m_tail ? QueryCount - m_tail : m_head - m_tail;
-        int64_t res[QueryCount];
-
-        // This memset is required, because Nvidia drivers seem to break Vulkan spec: "If VK_QUERY_RESULT_WITH_AVAILABILITY_BIT
-        // is set, the final integer value written for each query is non-zero if the query's status was available or zero if the
-        // status was unavailable."
-        // Nvidia drivers in some cases do not write the non-zero value, even if the following zero values are written.
-        memset( res, 0xFF, sizeof( *res ) * cnt );
-
-        if( vkGetQueryPoolResults( m_device, m_query, m_tail, cnt, sizeof( res ), res, sizeof( *res ), VK_QUERY_RESULT_64_BIT | VK_QUERY_RESULT_WITH_AVAILABILITY_BIT ) == VK_NOT_READY )
+        unsigned int cnt;
+        if( m_oldCnt != 0 )
         {
-            unsigned int idx;
-            for( idx=0; idx<cnt; idx++ )
-            {
-                if( res[idx] == 0 ) break;
-            }
-            if( idx == 0 ) return;
-            cnt = idx;
+            cnt = m_oldCnt;
+            m_oldCnt = 0;
+        }
+        else
+        {
+            cnt = m_head < m_tail ? QueryCount - m_tail : m_head - m_tail;
+        }
 
-            // The spec states that the query values MUST be available: "When VK_QUERY_RESULT_WITH_AVAILABILITY_BIT is used,
-            // implementations must guarantee that if they return a non-zero availability value then the numerical results must
-            // be valid".
-            // In some cases Nvidia drivers can still return VK_NOT_READY here, so we have to use VK_QUERY_RESULT_WAIT_BIT.
-            auto status = vkGetQueryPoolResults( m_device, m_query, m_tail, cnt, sizeof( res ), res, sizeof( *res ), VK_QUERY_RESULT_64_BIT | VK_QUERY_RESULT_WAIT_BIT );
-            assert( status == VK_SUCCESS );
+        int64_t res[QueryCount];
+        if( vkGetQueryPoolResults( m_device, m_query, m_tail, cnt, sizeof( res ), res, sizeof( *res ), VK_QUERY_RESULT_64_BIT ) == VK_NOT_READY )
+        {
+            m_oldCnt = cnt;
+            return;
         }
 
         Magic magic;
@@ -176,6 +168,7 @@ private:
 
     unsigned int m_head;
     unsigned int m_tail;
+    unsigned int m_oldCnt;
 };
 
 extern VkCtxWrapper s_vkCtx;