AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

P4 to Git Change 1123622 by wchau@wchau_WINDOWS7_OCL on 2015/02/19 17:52:17

Browse Source 
ECR #399840 - OpenCL Runtime HW Debug support development

	- use device to control debugger registration and exception notification so that debug event will not be tied to any particular queue.
	- use aqlCodeInfo parameter for clHwDbgMapKernelCodeAMD() to be consistent with clHwDbgGetAqlPacketInfoAMD()

Affected files ...

... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_debugger_amd.cpp#6 edit
... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_debugger_amd.h#6 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudebugmanager.cpp#7 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudebugmanager.hpp#4 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuvirtual.cpp#352 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuvirtual.hpp#127 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gslbe/src/rt/GSLContext.cpp#70 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gslbe/src/rt/GSLContext.h#43 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/hwdebug.cpp#5 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/hwdebug.hpp#6 edit


[ROCm/clr commit: 3e53caa02e]
This commit is contained in:
foreman
2015-02-19 19:40:35 -05:00
parent d53616dbf2
commit 5ac60a3c29
8 changed files with 28 additions and 89 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/clr/rocclr/runtime/device/gpu/gpudebugmanager.cpp
+16 -31
View File
@@ -111,6 +111,15 @@ GpuDebugManager::executePostDispatchCallBack()
}
}
//! Map the kernel code for host access
void
GpuDebugManager::mapKernelCode(void* aqlCodeInfo) const
{
AqlCodeInfo* codeInfo = reinterpret_cast<AqlCodeInfo*>(aqlCodeInfo);
codeInfo->aqlCode_ = reinterpret_cast<amd_kernel_code_t*>(aqlCodeAddr_);
codeInfo->aqlCodeSize_ = aqlCodeSize_;
}
cl_int
GpuDebugManager::registerDebugger(amd::Context* context, uintptr_t messageStorage)
@@ -125,8 +134,11 @@ GpuDebugManager::registerDebugger(amd::Context* context, uintptr_t messageStorag
// first time register - set the message storage, flush queue and enable hw debug
if (!isRegistered()) {
debugMessages_ = messageStorage;
dbgMsgBufferReady_ = true;
isRegistered_ = false;
if (!device()->gslCtx()->registerHwDebugger(debugMessages_)) {
return CL_OUT_OF_RESOURCES;
}
isRegistered_ = true;
}
context_ = context;
@@ -142,35 +154,10 @@ GpuDebugManager::unregisterDebugger()
// reset the debugger registration flag
isRegistered_ = false;
dbgMsgBufferReady_ = false;
context_ = NULL;
}
}
cl_int
GpuDebugManager::registerDebuggerOnQueue(device::VirtualDevice* vDevice)
{
if (!isMsgBufferReady()) {
return CL_DEBUGGER_REGISTER_FAILURE_AMD;
}
if (isRegistered()) { // The debugger has already been registered,
return CL_SUCCESS; // nothing to be done
}
VirtualGPU* vGpu = reinterpret_cast<gpu::VirtualGPU*>(vDevice);
// populate the fields in the debugMessages structure used by the GPU exception notification
if (vGpu->RegisterHwDebugger(debugMessages_)) {
vGpu_ = vGpu;
isRegistered_ = true;
return CL_SUCCESS;
}
return CL_DEBUGGER_REGISTER_FAILURE_AMD;
}
void
GpuDebugManager::flushCache(uint32_t mask)
{
@@ -205,7 +192,6 @@ GpuDebugManager::setupTrapInformation(DebugToolInfo* toolInfo)
toolInfo->trapBuffer_ = rtTrapInfo_[kDebugTrapBufferLocation];
}
void
GpuDebugManager::getPacketAmdInfo(
const void* aqlCodeInfo,
@@ -259,8 +245,7 @@ GpuDebugManager::createDebugEvent(
osEventReset(shaderEvent); // initial state is non-signaled
if (vGpu_->ExceptionNotification(shaderEvent)) {
isRegistered_ = true;
if (device()->gslCtx()->exceptionNotification(shaderEvent)) {
return shaderEvent;
}
}
@@ -287,7 +272,7 @@ GpuDebugManager::destroyDebugEvent(DebugEvent* pEvent)
osEventDestroy(*pEvent);
*pEvent = 0;
vGpu_->ExceptionNotification(0);
device()->gslCtx()->exceptionNotification(0);
}
projects/clr/rocclr/runtime/device/gpu/gpudebugmanager.hpp
+3 -3
View File
@@ -65,9 +65,6 @@ public:
//! Register the debugger
cl_int registerDebugger(amd::Context*context, uintptr_t messageStorage);
//! Register the debugger with KMD after command queue has been created
cl_int registerDebuggerOnQueue(device::VirtualDevice* vDevice);
//! Unregister the debugger
void unregisterDebugger();
@@ -84,6 +81,9 @@ public:
uint64_t* watchMode,
DebugEvent* pEvent);
//! Map the kernel code for host access
void mapKernelCode(void* aqlCodeInfo) const;
//! Get the packet information for dispatch
void getPacketAmdInfo(const void* aqlCodeInfo, void* packetInfo) const;
projects/clr/rocclr/runtime/device/gpu/gpuvirtual.cpp
+6 -17
View File
@@ -403,7 +403,6 @@ VirtualGPU::VirtualGPU(
, schedParamIdx_(0)
, deviceQueueSize_(0)
, hsaQueueMem_(NULL)
, useHwDebug_(false)
{
memset(&cal_, 0, sizeof(CalVirtualDesc));
for (uint i = 0; i < AllEngines; ++i) {
@@ -587,15 +586,6 @@ VirtualGPU::create(
return false;
}
// Check if HW Debug is used and register the debugger if not done yet
amd::HwDebugManager * dbgManager = dev().hwDebugMgr();
if ( dbgManager && dbgManager->isMsgBufferReady() ) {
if ( dbgManager->registerDebuggerOnQueue(this) == CL_SUCCESS ) {
useHwDebug_ = true;
}
}
return true;
}
@@ -1648,6 +1638,7 @@ VirtualGPU::submitKernelInternalHSA(
uint64_t vmDefQueue = 0;
amd::DeviceQueue* defQueue = kernel.program().context().defDeviceQueue(dev());
VirtualGPU* gpuDefQueue = NULL;
amd::HwDebugManager * dbgManager = dev().hwDebugMgr();
// Get the HSA kernel object
const HSAILKernel& hsaKernel =
@@ -1735,8 +1726,8 @@ VirtualGPU::submitKernelInternalHSA(
// setup the storage for the memory pointers of the kernel parameters
uint numParams = kernel.signature().numParameters();
if (useHwDebug_) {
dev().hwDebugMgr()->allocParamMemList(numParams);
if (dbgManager) {
dbgManager->allocParamMemList(numParams);
}
// Program the kernel arguments for the GPU execution
@@ -1768,7 +1759,7 @@ VirtualGPU::submitKernelInternalHSA(
HwDbgKernelInfo kernelInfo;
HwDbgKernelInfo *pKernelInfo = NULL;
if (useHwDebug_) {
if (dbgManager) {
buildKernelInfo(hsaKernel, aqlPkt, kernelInfo, enqueueEvent);
pKernelInfo = &kernelInfo;
}
@@ -1778,10 +1769,8 @@ VirtualGPU::submitKernelInternalHSA(
runAqlDispatch(gpuEvent, aqlPkt, vmMems(), cal_.memCount_,
scratch, scratchOffset, hsaKernel.cpuAqlCode(), hsaQueueMem_->vmAddress(), pKernelInfo);
if (useHwDebug_) {
if (NULL != dev().hwDebugMgr()->postDispatchCallBackFunc()) {
dev().hwDebugMgr()->executePostDispatchCallBack();
}
if (dbgManager && (NULL != dbgManager->postDispatchCallBackFunc())) {
dbgManager->executePostDispatchCallBack();
}
if (hsaKernel.dynamicParallelism()) {
projects/clr/rocclr/runtime/device/gpu/gpuvirtual.hpp
-2
View File
@@ -554,8 +554,6 @@ private:
uint deviceQueueSize_; //!< Device queue size
Memory* hsaQueueMem_; //!< Memory for the amd_queue_t object
bool useHwDebug_; //!< Flag of using HW debug
};
/*@}*/} // namespace gpu
projects/clr/rocclr/runtime/device/gpu/gslbe/src/rt/GSLContext.cpp
-12
View File
@@ -1300,18 +1300,6 @@ CALGSLContext::virtualQueueHandshake(GpuEvent& event, const gslMemObject mem, mc
eventEnd(MainEngine, event);
}
bool
CALGSLContext::RegisterHwDebugger(uint64 debugMessages)
{
return m_cs->registerHwDebugger(debugMessages);
}
bool
CALGSLContext::ExceptionNotification(osEventHandle debugEvent)
{
return m_cs->exceptionNotification(debugEvent);
}
void
CALGSLContext::InvalidateSqCaches(bool instInvalidate, bool dataInvalidate, bool tcL1, bool tcL2)
{
projects/clr/rocclr/runtime/device/gpu/gslbe/src/rt/GSLContext.h
-2
View File
@@ -142,8 +142,6 @@ public:
void writeSurfRaw(GpuEvent& event, gslMemObject mem, size_t size, const void* data);
/// HW Debug support functions
bool RegisterHwDebugger(uint64 debugMessages);
bool ExceptionNotification(osEventHandle debugEvent);
void InvalidateSqCaches(bool instInvalidate = true, bool dataInvalidate = true, bool tcL1 = true, bool tcL2 = true);
protected:
projects/clr/rocclr/runtime/device/hwdebug.cpp
-10
View File
@@ -41,7 +41,6 @@ HwDebugManager::HwDebugManager(amd::Device* device)
, scratchRingAddr_(NULL)
, scratchRingSize_(0)
, isRegistered_(false)
, dbgMsgBufferReady_(false)
{
memset(&debugInfo_, 0, sizeof(debugInfo_));
@@ -81,7 +80,6 @@ HwDebugManager::getDispatchDebugInfo(void* debugInfo) const
memcpy(debugInfo, (void*) &debugInfo_, sizeof(DispatchDebugInfo));
}
//! Set the kernel code address and its size
void
HwDebugManager::setKernelCodeInfo(address aqlCodeAddr, uint32_t aqlCodeSize)
@@ -98,14 +96,6 @@ HwDebugManager::setScratchRing(address scratchRingAddr, uint32_t scratchRingSize
scratchRingSize_ = scratchRingSize;
}
//! Map the shader (AQL code) for host access
void
HwDebugManager::mapKernelCode(uint64_t* aqlCodeAddr, uint32_t* aqlCodeSize) const
{
*aqlCodeAddr = reinterpret_cast<uint64_t>(aqlCodeAddr_);
*aqlCodeSize = aqlCodeSize_;
}
//! Map the scratch ring for host access
void
HwDebugManager::mapScratchRing(uint64_t* scratchRingAddr, uint32_t* scratchRingSize) const
projects/clr/rocclr/runtime/device/hwdebug.hpp
+3 -12
View File
@@ -68,8 +68,6 @@ struct RuntimeTrapInfo {
//!< Buffer to backup the VGPR used by the runtime trap handler
};
/**
* Opaque pointer to trap event
*/
@@ -113,9 +111,6 @@ public:
//! Get the scratch ring
void setScratchRing(address scratchRingAddr, uint32_t scratchRingSize);
//! Map the shader (AQL code) for host access
void mapKernelCode(uint64_t* aqlCodeAddr, uint32_t* aqlCodeSize) const;
//! Map the scratch ring for host access
void mapScratchRing(uint64_t* scratchRingAddr, uint32_t* scratchRingSize) const;
@@ -177,9 +172,6 @@ public:
//! Register the debugger
virtual cl_int registerDebugger(amd::Context* context, uintptr_t pMessageStorage) = 0;
//! Call KMD to register the debugger
virtual cl_int registerDebuggerOnQueue(device::VirtualDevice* vDevice) = 0;
//! Unregister the debugger
virtual void unregisterDebugger() = 0;
@@ -197,6 +189,9 @@ public:
uint64_t* watchMode,
DebugEvent* event) = 0;
//! Map the shader (AQL code) for host access
virtual void mapKernelCode(void* aqlCodeInfo) const = 0;
//! Get the packet information for dispatch
virtual void getPacketAmdInfo(const void* aqlCodeInfo,
void* packetInfo) const = 0;
@@ -214,9 +209,6 @@ public:
virtual void executePreDispatchCallBack(void* aqlPacket,
void* toolInfo) = 0;
//! Return the use of HW DEBUG flag
bool isMsgBufferReady() const { return dbgMsgBufferReady_; }
protected:
//! Return the context
const amd::Context* context() const { return context_; }
@@ -252,7 +244,6 @@ protected:
uint32_t scratchRingSize_; //!< The size of the scratch ring
bool isRegistered_; //! flag to indicate the debugger has been registered
bool dbgMsgBufferReady_; //! flag to indicate the HW DEBUG is using
cl_dbg_exception_policy_amd excpPolicy_; //!< exception policy
cl_dbg_kernel_exec_mode_amd execMode_; //!< kernel execution mode