code object tracking and v3 code object support

Change-Id: I081ada13f6364ea4401a97a485dedfa9bf8a45fe


[ROCm/rocprofiler commit: 60043d198b]
Bu işleme şunda yer alıyor:
Evgeny committed 2019-11-19 20:18:09 -06:00
1 ebeveyn c6607f3f62
işleme 9b183d909e
12 files changed
+293 -12

No files matched your search

+18 -7
Dosyayı Görüntüle
@@ -148,9 +148,20 @@ class InterceptQueue {
}
// Prepareing dispatch callback data
const amd_kernel_code_t* kernel_code = GetKernelCode(dispatch_packet);
const uint64_t kernel_symbol = kernel_code->runtime_loader_kernel_symbol;
const char* kernel_name = GetKernelName(kernel_symbol);
uint64_t kernel_object = dispatch_packet->kernel_object;
const amd_kernel_code_t* kernel_code = GetKernelCode(kernel_object);
const uint16_t kernel_object_flag = *((uint64_t*)kernel_code + 1);
if (kernel_object_flag == 0) {
if (!util::HsaRsrcFactory::IsExecutableTracking()) {
fprintf(stderr, "Error: V3 code object detected - code objects tracking should be enabled\n");
abort();
}
}
const char* kernel_name = (util::HsaRsrcFactory::IsExecutableTracking()) ?
util::HsaRsrcFactory::GetKernelName(kernel_object) :
GetKernelName(kernel_code->runtime_loader_kernel_symbol);
rocprofiler_callback_data_t data = {obj->agent_info_->dev_id,
obj->agent_info_->dev_index,
obj->queue_,
@@ -159,7 +170,7 @@ class InterceptQueue {
completion_signal,
dispatch_packet,
kernel_name,
kernel_symbol,
kernel_object,
kernel_code,
syscall(__NR_gettid),
(tracker_entry) ? tracker_entry->record : NULL};
@@ -243,14 +254,14 @@ class InterceptQueue {
return static_cast<hsa_packet_type_t>((*header >> HSA_PACKET_HEADER_TYPE) & header_type_mask);
}
static const amd_kernel_code_t* GetKernelCode(const hsa_kernel_dispatch_packet_t* dispatch_packet) {
static const amd_kernel_code_t* GetKernelCode(uint64_t kernel_object) {
const amd_kernel_code_t* kernel_code = NULL;
hsa_status_t status =
util::HsaRsrcFactory::Instance().LoaderApi()->hsa_ven_amd_loader_query_host_address(
reinterpret_cast<const void*>(dispatch_packet->kernel_object),
reinterpret_cast<const void*>(kernel_object),
reinterpret_cast<const void**>(&kernel_code));
if (HSA_STATUS_SUCCESS != status) {
kernel_code = reinterpret_cast<amd_kernel_code_t*>(dispatch_packet->kernel_object);
kernel_code = reinterpret_cast<amd_kernel_code_t*>(kernel_object);
}
return kernel_code;
}
+10 -2
Dosyayı Görüntüle
@@ -144,7 +144,8 @@ void * tool_handle = NULL;
// Return true if intercepting mode is enabled
enum {
DISPATCH_INTERCEPT_MODE = 0x1,
MEMCOPY_INTERCEPT_MODE = 0x2
CODE_OBJ_TRACKING_MODE = 0x2,
MEMCOPY_INTERCEPT_MODE = 0x4,
};
uint32_t LoadTool() {
uint32_t intercept_mode = 0;
@@ -188,6 +189,7 @@ uint32_t LoadTool() {
util::HsaRsrcFactory::SetTimeoutNs(settings.timeout);
InterceptQueue::TrackerOn(settings.timestamp_on != 0);
if (settings.intercept_mode != 0) intercept_mode = DISPATCH_INTERCEPT_MODE;
if (settings.code_obj_tracking) intercept_mode |= CODE_OBJ_TRACKING_MODE;
if (settings.memcopy_tracking) intercept_mode |= MEMCOPY_INTERCEPT_MODE;
}
@@ -432,7 +434,13 @@ PUBLIC_API bool OnLoad(HsaApiTable* table, uint64_t runtime_version, uint64_t fa
// Loading a tool lib and setting of intercept mode
const uint32_t intercept_mode_mask = rocprofiler::LoadTool();
if (intercept_mode_mask & rocprofiler::DISPATCH_INTERCEPT_MODE) intercept_mode = true;
if (intercept_mode_mask & rocprofiler::DISPATCH_INTERCEPT_MODE) {
intercept_mode = true;
}
if (intercept_mode_mask & rocprofiler::CODE_OBJ_TRACKING_MODE) {
if (intercept_mode == false) EXC_RAISING(HSA_STATUS_ERROR, "code objects tracking without intercept mode enabled");
rocprofiler::util::HsaRsrcFactory::EnableExecutableTracking(table);
}
if (intercept_mode_mask & rocprofiler::MEMCOPY_INTERCEPT_MODE) {
hsa_status_t status = hsa_amd_profiling_async_copy_enable(true);
if (status != HSA_STATUS_SUCCESS) EXC_ABORT(status, "hsa_amd_profiling_async_copy_enable");