SQTT memory checks

[ROCm/rocprofiler commit: ee5bc95019]
Tá an tiomantas seo le fáil i:
Evgeny
2018-04-04 12:14:29 -05:00
tuismitheoir 3c98242e09
tiomantas f1a5ba7386
D'athraigh 4 comhad le 60 breiseanna agus 44 scriosta
+24 -21
Féach ar an gComhad
@@ -322,6 +322,7 @@ class Context {
}
struct callback_data_t {
const profile_t* profile;
info_vector_t* info_vector;
size_t index;
char* ptr;
@@ -340,7 +341,7 @@ class Context {
if (!complete) WARN_LOGGING("timeout");
}
for (rocprofiler_feature_t* rinfo : *(tuple.info_vector)) rinfo->data.kind = ROCPROFILER_DATA_KIND_UNINIT;
callback_data_t callback_data{tuple.info_vector, tuple.info_vector->size(), NULL};
callback_data_t callback_data{tuple.profile, tuple.info_vector, tuple.info_vector->size(), NULL};
const hsa_status_t status =
api_->hsa_ven_amd_aqlprofile_iterate_data(tuple.profile, DataCallback, &callback_data);
if (status != HSA_STATUS_SUCCESS) AQL_EXC_RAISING(status, "context iterate data failed");
@@ -404,6 +405,7 @@ class Context {
hsa_ven_amd_aqlprofile_info_data_t* ainfo_data, void* data) {
hsa_status_t status = HSA_STATUS_SUCCESS;
callback_data_t* callback_data = reinterpret_cast<callback_data_t*>(data);
const profile_t* profile = callback_data->profile;
info_vector_t& info_vector = *(callback_data->info_vector);
uint32_t index = callback_data->index;
const uint32_t sample_id = ainfo_data->sample_id;
@@ -425,39 +427,40 @@ class Context {
} else if (ainfo_type == HSA_VEN_AMD_AQLPROFILE_INFO_SQTT_DATA) {
if (rinfo->data.result_bytes.copy) {
if (sample_id == 0) {
if (rinfo->data.result_bytes.size == 0) {
const uint32_t output_buffer_size = SqttProfile::GetSize();
const uint32_t output_buffer_size = profile->output_buffer.size;
const uint32_t output_buffer_size64 = output_buffer_size / sizeof(uint64_t);
rinfo->data.result_bytes.ptr = calloc(output_buffer_size64, sizeof(uint64_t));
void* ptr = calloc(output_buffer_size64, sizeof(uint64_t));
rinfo->data.result_bytes.size = output_buffer_size;
} else if (rinfo->data.result_bytes.size != SqttProfile::GetSize()) {
EXC_RAISING(HSA_STATUS_ERROR, "result bytes copy mode, data array size mismatch(" << rinfo->data.result_bytes.size << ")");
}
rinfo->data.result_bytes.ptr = ptr;
callback_data->ptr = reinterpret_cast<char*>(ptr);
}
char* result_bytes_ptr = reinterpret_cast<char*>(rinfo->data.result_bytes.ptr);
const char* end = result_bytes_ptr + rinfo->data.result_bytes.size;
const char* src = reinterpret_cast<char*>(ainfo_data->sqtt_data.ptr);
const uint32_t size = ainfo_data->sqtt_data.size;
char* ptr = (sample_id == 0) ? result_bytes_ptr : callback_data->ptr;
uint64_t* header = reinterpret_cast<uint64_t*>(ptr);
uint32_t size = ainfo_data->sqtt_data.size;
char* ptr = callback_data->ptr;
uint32_t* header = reinterpret_cast<uint32_t*>(ptr);
char* dest = ptr + sizeof(*header);
if ((dest + size) < end) {
hsa_status_t status = hsa_memory_copy(dest, src, size);
if (status == HSA_STATUS_SUCCESS) {
*header = size;
callback_data->ptr = dest + align_size(size, sizeof(uint64_t));
rinfo->data.result_bytes.instance_count = sample_id + 1;
rinfo->data.kind = ROCPROFILER_DATA_KIND_BYTES;
}
} else {
EXC_RAISING(HSA_STATUS_ERROR, "result bytes copy mode, out of boundary, size = " << rinfo->data.result_bytes.size);
if ((dest + size) >= end) {
if (dest < end) size = end - dest;
else EXC_RAISING(HSA_STATUS_ERROR, "SQTT data out of output buffer");
}
hsa_status_t status = hsa_memory_copy(dest, src, size);
if (status == HSA_STATUS_SUCCESS) {
*header = size;
callback_data->ptr = dest + align_size(size, sizeof(uint32_t));
rinfo->data.result_bytes.instance_count = sample_id + 1;
rinfo->data.kind = ROCPROFILER_DATA_KIND_BYTES;
}
} else {
if (sample_id == 0) {
rinfo->data.result_bytes.ptr = ainfo_data->sqtt_data.ptr;
rinfo->data.result_bytes.ptr = profile->output_buffer.ptr;
rinfo->data.result_bytes.size = profile->output_buffer.size;
rinfo->data.result_bytes.instance_count = UINT32_MAX;
}
rinfo->data.result_bytes.instance_count += 1;
rinfo->data.kind = ROCPROFILER_DATA_KIND_BYTES;
}
+1 -1
Féach ar an gComhad
@@ -14,5 +14,5 @@ InterceptQueue::obj_map_t* InterceptQueue::obj_map_ = NULL;
const char* InterceptQueue::kernel_none_ = "";
uint64_t InterceptQueue::timeout_ = UINT64_MAX;
Tracker* InterceptQueue::tracker_ = NULL;
bool InterceptQueue::tracker_on_ = true;
bool InterceptQueue::tracker_on_ = false;
} // namespace rocprofiler
+4 -3
Féach ar an gComhad
@@ -113,7 +113,8 @@ class Profile {
// Check the profile buffer sizes
status = api->hsa_ven_amd_aqlprofile_start(&profile_, NULL);
if (status != HSA_STATUS_SUCCESS) AQL_EXC_RAISING(status, "aqlprofile_start(NULL)");
Allocate(rsrc);
status = Allocate(rsrc);
if (status != HSA_STATUS_SUCCESS) AQL_EXC_RAISING(status, "Allocate()");
// Generate start/stop profiling packets
status = api->hsa_ven_amd_aqlprofile_start(&profile_, &start);
if (status != HSA_STATUS_SUCCESS) AQL_EXC_RAISING(status, "aqlprofile_start");
@@ -211,7 +212,7 @@ class PmcProfile : public Profile {
class SqttProfile : public Profile {
public:
static inline void SetSize(const uint32_t& size) { output_buffer_size_ = size; }
static inline uint32_t GetSize() { return output_buffer_size_; }
// static inline uint32_t GetSize() { return output_buffer_size_; }
SqttProfile(const util::AgentInfo* agent_info) : Profile(agent_info) {
profile_.type = HSA_VEN_AMD_AQLPROFILE_EVENT_TYPE_SQTT;
@@ -225,9 +226,9 @@ class SqttProfile : public Profile {
}
hsa_status_t Allocate(util::HsaRsrcFactory* rsrc) {
profile_.output_buffer.size = output_buffer_size_;
profile_.command_buffer.ptr =
rsrc->AllocateSysMemory(agent_info_, profile_.command_buffer.size);
profile_.output_buffer.size = output_buffer_size_;
profile_.output_buffer.ptr =
rsrc->AllocateLocalMemory(agent_info_, profile_.output_buffer.size);
return (profile_.command_buffer.ptr && profile_.output_buffer.ptr) ? HSA_STATUS_SUCCESS