Split the memory bandwidth reporting into its own function so that users trying to get memory usage information won't incur additional calls

PiperOrigin-RevId: 357122076
Change-Id: Ia23cbb2e06c36cf23a8b0f3f31ffb9d62d780b56
This commit is contained in:
Frank Chen 2021-02-11 21:41:51 -08:00 committed by TensorFlower Gardener
parent 0f270839dc
commit 2ad0a33259
5 changed files with 33 additions and 3 deletions

View File

@ -463,6 +463,7 @@ tf_kernel_library(
"//tensorflow/core:session_options",
"//tensorflow/core/kernels:ops_util",
"//tensorflow/core/profiler/lib:traceme",
"//tensorflow/core/profiler/lib:traceme_encode",
"@com_google_absl//absl/memory",
],
)

View File

@ -47,10 +47,12 @@ limitations under the License.
#include "tensorflow/core/lib/strings/stringprintf.h"
#include "tensorflow/core/platform/casts.h"
#include "tensorflow/core/platform/env.h"
#include "tensorflow/core/platform/mem.h"
#include "tensorflow/core/platform/mutex.h"
#include "tensorflow/core/platform/refcount.h"
#include "tensorflow/core/platform/resource.h"
#include "tensorflow/core/profiler/lib/traceme.h"
#include "tensorflow/core/profiler/lib/traceme_encode.h"
#include "tensorflow/core/public/session_options.h"
namespace tensorflow {
@ -965,9 +967,18 @@ void RecordElementSize(const std::vector<Tensor> element,
Status IteratorGetNextOp::DoCompute(OpKernelContext* ctx) {
profiler::TraceMe traceme(
[&] {
return strings::StrCat(
"IteratorGetNextOp::DoCompute#id=", ctx->step_id(),
",iter_num=", ctx->frame_iter().iter_id, "#");
int64 mem_bw = port::GetMemoryBandwidthInfo().bw_used;
if (mem_bw != INT64_MAX) {
return profiler::TraceMeEncode(
"IteratorGetNextOp::DoCompute",
{{"id", ctx->step_id()},
{"iter_num", ctx->frame_iter().iter_id},
{"mem_bw_used_megabytes_per_sec", mem_bw}});
}
return profiler::TraceMeEncode(
"IteratorGetNextOp::DoCompute",
{{"id", ctx->step_id()}, {"iter_num", ctx->frame_iter().iter_id}});
},
profiler::kInfo);
tensorflow::ResourceTagger tag(kTFDataResourceTag,

View File

@ -369,5 +369,10 @@ MemoryInfo GetMemoryInfo() {
return mem_info;
}
MemoryBandwidthInfo GetMemoryBandwidthInfo() {
MemoryBandwidthInfo membw_info = {INT64_MAX};
return membw_info;
}
} // namespace port
} // namespace tensorflow

View File

@ -64,11 +64,19 @@ struct MemoryInfo {
int64 free = 0;
};
struct MemoryBandwidthInfo {
int64 bw_used = 0; // memory bandwidth used across all CPU (in MBs/second)
};
// Retrieves the host memory information. If any of the fields in the returned
// MemoryInfo structure is INT64_MAX, it means such information is not
// available.
MemoryInfo GetMemoryInfo();
// Retrieves the host memory bandwidth information. If any field in the returned
// structure is INT64_MAX, it means such information is not available.
MemoryBandwidthInfo GetMemoryBandwidthInfo();
// Returns the amount of RAM available in bytes, or INT64_MAX if unknown.
static inline int64 AvailableRam() { return GetMemoryInfo().free; }

View File

@ -202,6 +202,11 @@ MemoryInfo GetMemoryInfo() {
return mem_info;
}
MemoryBandwidthInfo GetMemoryBandwidthInfo() {
MemoryBandwidthInfo membw_info = {INT64_MAX};
return membw_info;
}
int NumHyperthreadsPerCore() {
static const int ht_per_core = tensorflow::port::CPUIDNumSMT();
return (ht_per_core > 0) ? ht_per_core : 1;