mirror/tracy
1
0
Fork 0
mirror of https://github.com/wolfpld/tracy.git synced 2024-11-22 14:44:34 +00:00
Code Issues Packages Projects Releases Wiki Activity

Support use of dynamic vulkan symbols (#570)

Browse Source
This commit is contained in:
robertblaketaylor 2023-06-09 03:48:30 -07:00 committed by GitHub
parent 2b191e157e
commit 1b65a87c32
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 156 additions and 38 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
manual/tracy.tex
View File

@ -1531,6 +1531,12 @@ Vulkan 1.2 and \texttt{VK\_EXT\_host\_query\_reset} provide mechanics to reset t
However, using this feature requires the physical device to have calibrated device and host time domains. In addition to \texttt{VK\_TIME\_DOMAIN\_DEVICE\_EXT}, \texttt{vkGetPhysicalDeviceCalibrateableTimeDomainsEXT} will have to additionally return either \texttt{VK\_TIME\_DOMAIN\_CLOCK\_MONOTONIC\_RAW\_EXT} or \texttt{VK\_TIME\_DOMAIN\_QUERY\_PERFORMANCE\_COUNTER\_EXT} for Unix and Windows, respectively. If this is not the case, you will need to use \texttt{TracyVkContextCalibrated} or \texttt{TracyVkContext} macro instead. However, using this feature requires the physical device to have calibrated device and host time domains. In addition to \texttt{VK\_TIME\_DOMAIN\_DEVICE\_EXT}, \texttt{vkGetPhysicalDeviceCalibrateableTimeDomainsEXT} will have to additionally return either \texttt{VK\_TIME\_DOMAIN\_CLOCK\_MONOTONIC\_RAW\_EXT} or \texttt{VK\_TIME\_DOMAIN\_QUERY\_PERFORMANCE\_COUNTER\_EXT} for Unix and Windows, respectively. If this is not the case, you will need to use \texttt{TracyVkContextCalibrated} or \texttt{TracyVkContext} macro instead.
\subparagraph{Dynamically loading the Vulkan symbols}
Some applications dynamically link the Vulkan loader, and manage a local symbol table, to remove the trampoline overhead of calling through the Vulkan loader itself.
When \texttt{TRACY\_VK\_USE\_SYMBOL\_TABLE} is defined the signature of \texttt{TracyVkContext}, \texttt{TracyVkContextCalibrated}, and \texttt{TracyVkContextHostCalibrated} are adjusted to take in the \texttt{VkInstance}, \texttt{PFN\_vkGetInstanceProcAddr}, and \texttt{PFN\_vkGetDeviceProcAddr} to enable constructing a local symbol table to be used to call through the Vulkan API when tracing.
\subsubsection{Direct3D 11} \subsubsection{Direct3D 11}
To enable Direct3D 11 support, include the \texttt{public/tracy/TracyD3D11.hpp} header file, and create a \texttt{TracyD3D11Ctx} object with the \texttt{TracyD3D11Context(device, devicecontext)} macro. The object should later be cleaned up with the \texttt{TracyD3D11Destroy} macro. Tracy does not support D3D11 command lists. To set a custom name for the context, use the \texttt{TracyGpuContextName(name, size)} macro. To enable Direct3D 11 support, include the \texttt{public/tracy/TracyD3D11.hpp} header file, and create a \texttt{TracyD3D11Ctx} object with the \texttt{TracyD3D11Context(device, devicecontext)} macro. The object should later be cleaned up with the \texttt{TracyD3D11Destroy} macro. Tracy does not support D3D11 command lists. To set a custom name for the context, use the \texttt{TracyGpuContextName(name, size)} macro.

188
public/tracy/TracyVulkan.hpp
View File

@ -45,6 +45,41 @@ using TracyVkCtx = void*;
namespace tracy namespace tracy
{ {
#if defined TRACY_VK_USE_SYMBOL_TABLE
#define LoadVkDeviceCoreSymbols(Operation) \
Operation(vkBeginCommandBuffer) \
Operation(vkCmdResetQueryPool) \
Operation(vkCmdWriteTimestamp) \
Operation(vkCreateQueryPool) \
Operation(vkDestroyQueryPool) \
Operation(vkEndCommandBuffer) \
Operation(vkGetQueryPoolResults) \
Operation(vkQueueSubmit) \
Operation(vkQueueWaitIdle)
#define LoadVkDeviceExtensionSymbols(Operation) \
Operation(vkGetCalibratedTimestampsEXT) \
Operation(vkGetPhysicalDeviceCalibrateableTimeDomainsEXT)
#define LoadVkInstanceCoreSymbols(Operation) \
Operation(vkGetPhysicalDeviceProperties)
struct VkSymbolTable
{
#define MAKE_PFN(name) PFN_##name name;
LoadVkDeviceCoreSymbols(MAKE_PFN)
LoadVkDeviceExtensionSymbols(MAKE_PFN)
LoadVkInstanceCoreSymbols(MAKE_PFN)
#undef MAKE_PFN
};
#define VK_FUNCTION_WRAPPER(callSignature) m_symbols.callSignature
#define CONTEXT_VK_FUNCTION_WRAPPER(callSignature) m_ctx->m_symbols.callSignature
#else
#define VK_FUNCTION_WRAPPER(callSignature) callSignature
#define CONTEXT_VK_FUNCTION_WRAPPER(callSignature) callSignature
#endif
class VkCtx class VkCtx
{ {
friend class VkCtxScope; friend class VkCtxScope;
@ -52,7 +87,11 @@ class VkCtx
enum { QueryCount = 64 * 1024 }; enum { QueryCount = 64 * 1024 };
public: public:
VkCtx( VkPhysicalDevice physdev, VkDevice device, VkQueue queue, VkCommandBuffer cmdbuf, PFN_vkGetPhysicalDeviceCalibrateableTimeDomainsEXT _vkGetPhysicalDeviceCalibrateableTimeDomainsEXT, PFN_vkGetCalibratedTimestampsEXT _vkGetCalibratedTimestampsEXT ) #if defined TRACY_VK_USE_SYMBOL_TABLE
VkCtx( VkInstance instance, VkPhysicalDevice physdev, VkDevice device, VkQueue queue, VkCommandBuffer cmdbuf, PFN_vkGetInstanceProcAddr instanceProcAddr, PFN_vkGetDeviceProcAddr deviceProcAddr, bool calibrated )
#else
VkCtx( VkPhysicalDevice physdev, VkDevice device, VkQueue queue, VkCommandBuffer cmdbuf, PFN_vkGetPhysicalDeviceCalibrateableTimeDomainsEXT vkGetPhysicalDeviceCalibrateableTimeDomainsEXT, PFN_vkGetCalibratedTimestampsEXT vkGetCalibratedTimestampsEXT)
#endif
: m_device( device ) : m_device( device )
, m_timeDomain( VK_TIME_DOMAIN_DEVICE_EXT ) , m_timeDomain( VK_TIME_DOMAIN_DEVICE_EXT )
, m_context( GetGpuCtxCounter().fetch_add( 1, std::memory_order_relaxed ) ) , m_context( GetGpuCtxCounter().fetch_add( 1, std::memory_order_relaxed ) )
@ -60,13 +99,24 @@ public:
, m_tail( 0 ) , m_tail( 0 )
, m_oldCnt( 0 ) , m_oldCnt( 0 )
, m_queryCount( QueryCount ) , m_queryCount( QueryCount )
, m_vkGetCalibratedTimestampsEXT( _vkGetCalibratedTimestampsEXT ) #if !defined TRACY_VK_USE_SYMBOL_TABLE
, m_vkGetCalibratedTimestampsEXT( vkGetCalibratedTimestampsEXT )
#endif
{ {
assert( m_context != 255 ); assert( m_context != 255 );
if( _vkGetPhysicalDeviceCalibrateableTimeDomainsEXT && _vkGetCalibratedTimestampsEXT ) #if defined TRACY_VK_USE_SYMBOL_TABLE
PopulateSymbolTable(instance, instanceProcAddr, deviceProcAddr);
if ( calibrated )
{ {
FindAvailableTimeDomains( physdev, _vkGetPhysicalDeviceCalibrateableTimeDomainsEXT ); m_vkGetCalibratedTimestampsEXT = m_symbols.vkGetCalibratedTimestampsEXT;
}
#endif
if( VK_FUNCTION_WRAPPER( vkGetPhysicalDeviceCalibrateableTimeDomainsEXT ) && m_vkGetCalibratedTimestampsEXT )
{
FindAvailableTimeDomains( physdev, VK_FUNCTION_WRAPPER( vkGetPhysicalDeviceCalibrateableTimeDomainsEXT ) );
} }
CreateQueryPool(); CreateQueryPool();
@ -80,29 +130,29 @@ public:
submitInfo.commandBufferCount = 1; submitInfo.commandBufferCount = 1;
submitInfo.pCommandBuffers = &cmdbuf; submitInfo.pCommandBuffers = &cmdbuf;
vkBeginCommandBuffer( cmdbuf, &beginInfo ); VK_FUNCTION_WRAPPER( vkBeginCommandBuffer( cmdbuf, &beginInfo ) );
vkCmdResetQueryPool( cmdbuf, m_query, 0, m_queryCount ); VK_FUNCTION_WRAPPER( vkCmdResetQueryPool( cmdbuf, m_query, 0, m_queryCount ) );
vkEndCommandBuffer( cmdbuf ); VK_FUNCTION_WRAPPER( vkEndCommandBuffer( cmdbuf ) );
vkQueueSubmit( queue, 1, &submitInfo, VK_NULL_HANDLE ); VK_FUNCTION_WRAPPER( vkQueueSubmit( queue, 1, &submitInfo, VK_NULL_HANDLE ) );
vkQueueWaitIdle( queue ); VK_FUNCTION_WRAPPER( vkQueueWaitIdle( queue ) );
int64_t tcpu, tgpu; int64_t tcpu, tgpu;
if( m_timeDomain == VK_TIME_DOMAIN_DEVICE_EXT ) if( m_timeDomain == VK_TIME_DOMAIN_DEVICE_EXT )
{ {
vkBeginCommandBuffer( cmdbuf, &beginInfo ); VK_FUNCTION_WRAPPER( vkBeginCommandBuffer( cmdbuf, &beginInfo ) );
vkCmdWriteTimestamp( cmdbuf, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, m_query, 0 ); VK_FUNCTION_WRAPPER( vkCmdWriteTimestamp( cmdbuf, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, m_query, 0 ) );
vkEndCommandBuffer( cmdbuf ); VK_FUNCTION_WRAPPER( vkEndCommandBuffer( cmdbuf ) );
vkQueueSubmit( queue, 1, &submitInfo, VK_NULL_HANDLE ); VK_FUNCTION_WRAPPER( vkQueueSubmit( queue, 1, &submitInfo, VK_NULL_HANDLE ) );
vkQueueWaitIdle( queue ); VK_FUNCTION_WRAPPER( vkQueueWaitIdle( queue ) );
tcpu = Profiler::GetTime(); tcpu = Profiler::GetTime();
vkGetQueryPoolResults( device, m_query, 0, 1, sizeof( tgpu ), &tgpu, sizeof( tgpu ), VK_QUERY_RESULT_64_BIT | VK_QUERY_RESULT_WAIT_BIT ); VK_FUNCTION_WRAPPER( vkGetQueryPoolResults( device, m_query, 0, 1, sizeof( tgpu ), &tgpu, sizeof( tgpu ), VK_QUERY_RESULT_64_BIT | VK_QUERY_RESULT_WAIT_BIT ) );
vkBeginCommandBuffer( cmdbuf, &beginInfo ); VK_FUNCTION_WRAPPER( vkBeginCommandBuffer( cmdbuf, &beginInfo ) );
vkCmdResetQueryPool( cmdbuf, m_query, 0, 1 ); VK_FUNCTION_WRAPPER( vkCmdResetQueryPool( cmdbuf, m_query, 0, 1 ) );
vkEndCommandBuffer( cmdbuf ); VK_FUNCTION_WRAPPER( vkEndCommandBuffer( cmdbuf ) );
vkQueueSubmit( queue, 1, &submitInfo, VK_NULL_HANDLE ); VK_FUNCTION_WRAPPER( vkQueueSubmit( queue, 1, &submitInfo, VK_NULL_HANDLE ) );
vkQueueWaitIdle( queue ); VK_FUNCTION_WRAPPER( vkQueueWaitIdle( queue ) );
} }
else else
{ {
@ -122,7 +172,11 @@ public:
* VK_EXT_host_query_reset (core with 1.2 and non-optional) and VK_EXT_calibrated_timestamps. This requires * VK_EXT_host_query_reset (core with 1.2 and non-optional) and VK_EXT_calibrated_timestamps. This requires
* the physical device to have another time domain apart from DEVICE to be calibrateable. * the physical device to have another time domain apart from DEVICE to be calibrateable.
*/ */
VkCtx( VkPhysicalDevice physdev, VkDevice device, PFN_vkResetQueryPoolEXT _vkResetQueryPool, PFN_vkGetPhysicalDeviceCalibrateableTimeDomainsEXT _vkGetPhysicalDeviceCalibrateableTimeDomainsEXT, PFN_vkGetCalibratedTimestampsEXT _vkGetCalibratedTimestampsEXT ) #if defined TRACY_VK_USE_SYMBOL_TABLE
VkCtx( VkInstance instance, VkPhysicalDevice physdev, VkDevice device, PFN_vkGetInstanceProcAddr instanceProcAddr, PFN_vkGetDeviceProcAddr deviceProcAddr )
#else
VkCtx( VkPhysicalDevice physdev, VkDevice device, PFN_vkResetQueryPoolEXT vkResetQueryPool, PFN_vkGetPhysicalDeviceCalibrateableTimeDomainsEXT vkGetPhysicalDeviceCalibrateableTimeDomainsEXT, PFN_vkGetCalibratedTimestampsEXT vkGetCalibratedTimestampsEXT )
#endif
: m_device( device ) : m_device( device )
, m_timeDomain( VK_TIME_DOMAIN_DEVICE_EXT ) , m_timeDomain( VK_TIME_DOMAIN_DEVICE_EXT )
, m_context( GetGpuCtxCounter().fetch_add(1, std::memory_order_relaxed) ) , m_context( GetGpuCtxCounter().fetch_add(1, std::memory_order_relaxed) )
@ -130,14 +184,22 @@ public:
, m_tail( 0 ) , m_tail( 0 )
, m_oldCnt( 0 ) , m_oldCnt( 0 )
, m_queryCount( QueryCount ) , m_queryCount( QueryCount )
, m_vkGetCalibratedTimestampsEXT( _vkGetCalibratedTimestampsEXT ) #if !defined TRACY_VK_USE_SYMBOL_TABLE
, m_vkGetCalibratedTimestampsEXT( vkGetCalibratedTimestampsEXT )
#endif
{ {
assert( m_context != 255); assert( m_context != 255);
assert( _vkResetQueryPool != nullptr );
assert( _vkGetPhysicalDeviceCalibrateableTimeDomainsEXT != nullptr );
assert( _vkGetCalibratedTimestampsEXT != nullptr );
FindAvailableTimeDomains( physdev, _vkGetPhysicalDeviceCalibrateableTimeDomainsEXT ); #if defined TRACY_VK_USE_SYMBOL_TABLE
PopulateSymbolTable(instance, instanceProcAddr, deviceProcAddr);
m_vkGetCalibratedTimestampsEXT = m_symbols.vkGetCalibratedTimestampsEXT;
#endif
assert( VK_FUNCTION_WRAPPER( vkResetQueryPool ) != nullptr );
assert( VK_FUNCTION_WRAPPER( vkGetPhysicalDeviceCalibrateableTimeDomainsEXT ) != nullptr );
assert( VK_FUNCTION_WRAPPER( vkGetCalibratedTimestampsEXT ) != nullptr );
FindAvailableTimeDomains( physdev, VK_FUNCTION_WRAPPER( vkGetPhysicalDeviceCalibrateableTimeDomainsEXT ) );
// We require a host time domain to be available to properly calibrate. // We require a host time domain to be available to properly calibrate.
FindCalibratedTimestampDeviation(); FindCalibratedTimestampDeviation();
@ -146,7 +208,7 @@ public:
int64_t tcpu = Profiler::GetTime(); int64_t tcpu = Profiler::GetTime();
CreateQueryPool(); CreateQueryPool();
_vkResetQueryPool( device, m_query, 0, m_queryCount ); VK_FUNCTION_WRAPPER( vkResetQueryPool( device, m_query, 0, m_queryCount ) );
WriteInitialItem( physdev, tcpu, tgpu ); WriteInitialItem( physdev, tcpu, tgpu );
@ -157,7 +219,7 @@ public:
~VkCtx() ~VkCtx()
{ {
tracy_free( m_res ); tracy_free( m_res );
vkDestroyQueryPool( m_device, m_query, nullptr ); VK_FUNCTION_WRAPPER( vkDestroyQueryPool( m_device, m_query, nullptr ) );
} }
void Name( const char* name, uint16_t len ) void Name( const char* name, uint16_t len )
@ -185,7 +247,7 @@ public:
#ifdef TRACY_ON_DEMAND #ifdef TRACY_ON_DEMAND
if( !GetProfiler().IsConnected() ) if( !GetProfiler().IsConnected() )
{ {
vkCmdResetQueryPool( cmdbuf, m_query, 0, m_queryCount ); VK_FUNCTION_WRAPPER( vkCmdResetQueryPool( cmdbuf, m_query, 0, m_queryCount ) );
m_head = m_tail = m_oldCnt = 0; m_head = m_tail = m_oldCnt = 0;
int64_t tgpu; int64_t tgpu;
if( m_timeDomain != VK_TIME_DOMAIN_DEVICE_EXT ) Calibrate( m_device, m_prevCalibration, tgpu ); if( m_timeDomain != VK_TIME_DOMAIN_DEVICE_EXT ) Calibrate( m_device, m_prevCalibration, tgpu );
@ -204,7 +266,7 @@ public:
cnt = m_head < m_tail ? m_queryCount - m_tail : m_head - m_tail; cnt = m_head < m_tail ? m_queryCount - m_tail : m_head - m_tail;
} }
if( vkGetQueryPoolResults( m_device, m_query, m_tail, cnt, sizeof( int64_t ) * m_queryCount, m_res, sizeof( int64_t ), VK_QUERY_RESULT_64_BIT ) == VK_NOT_READY ) if( VK_FUNCTION_WRAPPER( vkGetQueryPoolResults( m_device, m_query, m_tail, cnt, sizeof( int64_t ) * m_queryCount, m_res, sizeof( int64_t ), VK_QUERY_RESULT_64_BIT ) == VK_NOT_READY ) )
{ {
m_oldCnt = cnt; m_oldCnt = cnt;
return; return;
@ -239,7 +301,7 @@ public:
} }
} }
vkCmdResetQueryPool( cmdbuf, m_query, m_tail, cnt ); VK_FUNCTION_WRAPPER( vkCmdResetQueryPool( cmdbuf, m_query, m_tail, cnt ) );
m_tail += cnt; m_tail += cnt;
if( m_tail == m_queryCount ) m_tail = 0; if( m_tail == m_queryCount ) m_tail = 0;
@ -291,7 +353,7 @@ private:
poolInfo.sType = VK_STRUCTURE_TYPE_QUERY_POOL_CREATE_INFO; poolInfo.sType = VK_STRUCTURE_TYPE_QUERY_POOL_CREATE_INFO;
poolInfo.queryCount = m_queryCount; poolInfo.queryCount = m_queryCount;
poolInfo.queryType = VK_QUERY_TYPE_TIMESTAMP; poolInfo.queryType = VK_QUERY_TYPE_TIMESTAMP;
while ( vkCreateQueryPool( m_device, &poolInfo, nullptr, &m_query ) != VK_SUCCESS ) while ( VK_FUNCTION_WRAPPER( vkCreateQueryPool( m_device, &poolInfo, nullptr, &m_query ) != VK_SUCCESS ) )
{ {
m_queryCount /= 2; m_queryCount /= 2;
poolInfo.queryCount = m_queryCount; poolInfo.queryCount = m_queryCount;
@ -351,7 +413,7 @@ private:
if( m_timeDomain != VK_TIME_DOMAIN_DEVICE_EXT ) flags |= GpuContextCalibration; if( m_timeDomain != VK_TIME_DOMAIN_DEVICE_EXT ) flags |= GpuContextCalibration;
VkPhysicalDeviceProperties prop; VkPhysicalDeviceProperties prop;
vkGetPhysicalDeviceProperties( physdev, &prop ); VK_FUNCTION_WRAPPER( vkGetPhysicalDeviceProperties( physdev, &prop ) );
const float period = prop.limits.timestampPeriod; const float period = prop.limits.timestampPeriod;
auto item = Profiler::QueueSerial(); auto item = Profiler::QueueSerial();
@ -370,9 +432,31 @@ private:
Profiler::QueueSerialFinish(); Profiler::QueueSerialFinish();
} }
#if defined TRACY_VK_USE_SYMBOL_TABLE
void PopulateSymbolTable( VkInstance instance, PFN_vkGetInstanceProcAddr instanceProcAddr, PFN_vkGetDeviceProcAddr deviceProcAddr )
{
#define VK_GET_DEVICE_SYMBOL( name ) \
(PFN_##name)deviceProcAddr( m_device, #name );
#define VK_LOAD_DEVICE_SYMBOL( name ) \
m_symbols.name = VK_GET_DEVICE_SYMBOL( name );
#define VK_GET_INSTANCE_SYMBOL( name ) \
(PFN_##name)instanceProcAddr( instance, #name );
#define VK_LOAD_INSTANCE_SYMBOL( name ) \
m_symbols.name = VK_GET_INSTANCE_SYMBOL( name );
LoadVkDeviceCoreSymbols( VK_LOAD_DEVICE_SYMBOL )
LoadVkDeviceExtensionSymbols( VK_LOAD_DEVICE_SYMBOL )
LoadVkInstanceCoreSymbols( VK_LOAD_INSTANCE_SYMBOL )
#undef VK_LOAD_DEVICE_SYMBOL
#undef VK_LOAD_INSTANCE_SYMBOL
}
VkDevice m_device; VkDevice m_device;
VkQueryPool m_query; VkQueryPool m_query;
VkTimeDomainEXT m_timeDomain; VkTimeDomainEXT m_timeDomain;
#if defined TRACY_VK_USE_SYMBOL_TABLE
VkSymbolTable m_symbols;
#endif
uint64_t m_deviation; uint64_t m_deviation;
int64_t m_qpcToNs; int64_t m_qpcToNs;
int64_t m_prevCalibration; int64_t m_prevCalibration;
@ -403,7 +487,7 @@ public:
m_ctx = ctx; m_ctx = ctx;
const auto queryId = ctx->NextQueryId(); const auto queryId = ctx->NextQueryId();
vkCmdWriteTimestamp( cmdbuf, VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, ctx->m_query, queryId ); CONTEXT_VK_FUNCTION_WRAPPER( vkCmdWriteTimestamp( cmdbuf, VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, ctx->m_query, queryId ) );
auto item = Profiler::QueueSerial(); auto item = Profiler::QueueSerial();
MemWrite( &item->hdr.type, QueueType::GpuZoneBeginSerial ); MemWrite( &item->hdr.type, QueueType::GpuZoneBeginSerial );
@ -427,7 +511,7 @@ public:
m_ctx = ctx; m_ctx = ctx;
const auto queryId = ctx->NextQueryId(); const auto queryId = ctx->NextQueryId();
vkCmdWriteTimestamp( cmdbuf, VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, ctx->m_query, queryId ); CONTEXT_VK_FUNCTION_WRAPPER( vkCmdWriteTimestamp( cmdbuf, VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, ctx->m_query, queryId ) );
auto item = Profiler::QueueSerialCallstack( Callstack( depth ) ); auto item = Profiler::QueueSerialCallstack( Callstack( depth ) );
MemWrite( &item->hdr.type, QueueType::GpuZoneBeginCallstackSerial ); MemWrite( &item->hdr.type, QueueType::GpuZoneBeginCallstackSerial );
@ -451,7 +535,7 @@ public:
m_ctx = ctx; m_ctx = ctx;
const auto queryId = ctx->NextQueryId(); const auto queryId = ctx->NextQueryId();
vkCmdWriteTimestamp( cmdbuf, VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, ctx->m_query, queryId ); CONTEXT_VK_FUNCTION_WRAPPER( vkCmdWriteTimestamp( cmdbuf, VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, ctx->m_query, queryId ) );
const auto srcloc = Profiler::AllocSourceLocation( line, source, sourceSz, function, functionSz, name, nameSz ); const auto srcloc = Profiler::AllocSourceLocation( line, source, sourceSz, function, functionSz, name, nameSz );
auto item = Profiler::QueueSerial(); auto item = Profiler::QueueSerial();
@ -476,7 +560,7 @@ public:
m_ctx = ctx; m_ctx = ctx;
const auto queryId = ctx->NextQueryId(); const auto queryId = ctx->NextQueryId();
vkCmdWriteTimestamp( cmdbuf, VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, ctx->m_query, queryId ); CONTEXT_VK_FUNCTION_WRAPPER( vkCmdWriteTimestamp( cmdbuf, VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, ctx->m_query, queryId ) );
const auto srcloc = Profiler::AllocSourceLocation( line, source, sourceSz, function, functionSz, name, nameSz ); const auto srcloc = Profiler::AllocSourceLocation( line, source, sourceSz, function, functionSz, name, nameSz );
auto item = Profiler::QueueSerialCallstack( Callstack( depth ) ); auto item = Profiler::QueueSerialCallstack( Callstack( depth ) );
@ -494,7 +578,7 @@ public:
if( !m_active ) return; if( !m_active ) return;
const auto queryId = m_ctx->NextQueryId(); const auto queryId = m_ctx->NextQueryId();
vkCmdWriteTimestamp( m_cmdbuf, VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, m_ctx->m_query, queryId ); CONTEXT_VK_FUNCTION_WRAPPER( vkCmdWriteTimestamp( m_cmdbuf, VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, m_ctx->m_query, queryId ) );
auto item = Profiler::QueueSerial(); auto item = Profiler::QueueSerial();
MemWrite( &item->hdr.type, QueueType::GpuZoneEndSerial ); MemWrite( &item->hdr.type, QueueType::GpuZoneEndSerial );
@ -512,18 +596,34 @@ private:
VkCtx* m_ctx; VkCtx* m_ctx;
}; };
#if defined TRACY_VK_USE_SYMBOL_TABLE
static inline VkCtx* CreateVkContext( VkInstance instance, VkPhysicalDevice physdev, VkDevice device, VkQueue queue, VkCommandBuffer cmdbuf, PFN_vkGetInstanceProcAddr instanceProcAddr, PFN_vkGetDeviceProcAddr getDeviceProcAddr, bool calibrated = false )
#else
static inline VkCtx* CreateVkContext( VkPhysicalDevice physdev, VkDevice device, VkQueue queue, VkCommandBuffer cmdbuf, PFN_vkGetPhysicalDeviceCalibrateableTimeDomainsEXT gpdctd, PFN_vkGetCalibratedTimestampsEXT gct ) static inline VkCtx* CreateVkContext( VkPhysicalDevice physdev, VkDevice device, VkQueue queue, VkCommandBuffer cmdbuf, PFN_vkGetPhysicalDeviceCalibrateableTimeDomainsEXT gpdctd, PFN_vkGetCalibratedTimestampsEXT gct )
#endif
{ {
auto ctx = (VkCtx*)tracy_malloc( sizeof( VkCtx ) ); auto ctx = (VkCtx*)tracy_malloc( sizeof( VkCtx ) );
#if defined TRACY_VK_USE_SYMBOL_TABLE
new(ctx) VkCtx( instance, physdev, device, queue, cmdbuf, instanceProcAddr, getDeviceProcAddr, calibrated );
#else
new(ctx) VkCtx( physdev, device, queue, cmdbuf, gpdctd, gct ); new(ctx) VkCtx( physdev, device, queue, cmdbuf, gpdctd, gct );
#endif
return ctx; return ctx;
} }
#if defined VK_EXT_host_query_reset #if defined VK_EXT_host_query_reset
#if defined TRACY_VK_USE_SYMBOL_TABLE
static inline VkCtx* CreateVkContext( VkInstance instance, VkPhysicalDevice physdev, VkDevice device, PFN_vkGetInstanceProcAddr instanceProcAddr, PFN_vkGetDeviceProcAddr getDeviceProcAddr )
#else
static inline VkCtx* CreateVkContext( VkPhysicalDevice physdev, VkDevice device, PFN_vkResetQueryPoolEXT qpreset, PFN_vkGetPhysicalDeviceCalibrateableTimeDomainsEXT gpdctd, PFN_vkGetCalibratedTimestampsEXT gct ) static inline VkCtx* CreateVkContext( VkPhysicalDevice physdev, VkDevice device, PFN_vkResetQueryPoolEXT qpreset, PFN_vkGetPhysicalDeviceCalibrateableTimeDomainsEXT gpdctd, PFN_vkGetCalibratedTimestampsEXT gct )
#endif
{ {
auto ctx = (VkCtx*)tracy_malloc( sizeof( VkCtx ) ); auto ctx = (VkCtx*)tracy_malloc( sizeof( VkCtx ) );
#if defined TRACY_VK_USE_SYMBOL_TABLE
new(ctx) VkCtx( instance, physdev, device, instanceProcAddr, getDeviceProcAddr );
#else
new(ctx) VkCtx( physdev, device, qpreset, gpdctd, gct ); new(ctx) VkCtx( physdev, device, qpreset, gpdctd, gct );
#endif
return ctx; return ctx;
} }
#endif #endif
@ -538,11 +638,23 @@ static inline void DestroyVkContext( VkCtx* ctx )
using TracyVkCtx = tracy::VkCtx*; using TracyVkCtx = tracy::VkCtx*;
#if defined TRACY_VK_USE_SYMBOL_TABLE
#define TracyVkContext( instance, physdev, device, queue, cmdbuf, instanceProcAddr, deviceProcAddr ) tracy::CreateVkContext( instance, physdev, device, queue, cmdbuf, instanceProcAddr, deviceProcAddr );
#else
#define TracyVkContext( physdev, device, queue, cmdbuf ) tracy::CreateVkContext( physdev, device, queue, cmdbuf, nullptr, nullptr ); #define TracyVkContext( physdev, device, queue, cmdbuf ) tracy::CreateVkContext( physdev, device, queue, cmdbuf, nullptr, nullptr );
#endif
#if defined TRACY_VK_USE_SYMBOL_TABLE
#define TracyVkContextCalibrated( instance, physdev, device, queue, cmdbuf, instanceProcAddr, deviceProcAddr ) tracy::CreateVkContext( instance, physdev, device, queue, cmdbuf, instanceProcAddr, deviceProcAddr, true );
#else
#define TracyVkContextCalibrated( physdev, device, queue, cmdbuf, gpdctd, gct ) tracy::CreateVkContext( physdev, device, queue, cmdbuf, gpdctd, gct ); #define TracyVkContextCalibrated( physdev, device, queue, cmdbuf, gpdctd, gct ) tracy::CreateVkContext( physdev, device, queue, cmdbuf, gpdctd, gct );
#endif
#if defined VK_EXT_host_query_reset #if defined VK_EXT_host_query_reset
#if defined TRACY_VK_USE_SYMBOL_TABLE
#define TracyVkContextHostCalibrated( instance, physdev, device, instanceProcAddr, deviceProcAddr ) tracy::CreateVkContext( instance, physdev, device, instanceProcAddr, deviceProcAddr );
#else
#define TracyVkContextHostCalibrated( physdev, device, qpreset, gpdctd, gct ) tracy::CreateVkContext( physdev, device, qpreset, gpdctd, gct ); #define TracyVkContextHostCalibrated( physdev, device, qpreset, gpdctd, gct ) tracy::CreateVkContext( physdev, device, qpreset, gpdctd, gct );
#endif #endif
#endif
#define TracyVkDestroy( ctx ) tracy::DestroyVkContext( ctx ); #define TracyVkDestroy( ctx ) tracy::DestroyVkContext( ctx );
#define TracyVkContextName( ctx, name, size ) ctx->Name( name, size ); #define TracyVkContextName( ctx, name, size ) ctx->Name( name, size );
#if defined TRACY_HAS_CALLSTACK && defined TRACY_CALLSTACK #if defined TRACY_HAS_CALLSTACK && defined TRACY_CALLSTACK