[flang-commits] [flang] 166192e - [flang] Add POSIX implementation for CPU_TIME
Diana Picus via flang-commits
flang-commits at lists.llvm.org
Mon Jun 14 02:42:20 PDT 2021
Author: Diana Picus
Date: 2021-06-14T09:41:11Z
New Revision: 166192eb78cdeb41ee25a37e920259fb13f82f6a
URL: https://github.com/llvm/llvm-project/commit/166192eb78cdeb41ee25a37e920259fb13f82f6a
DIFF: https://github.com/llvm/llvm-project/commit/166192eb78cdeb41ee25a37e920259fb13f82f6a.diff
LOG: [flang] Add POSIX implementation for CPU_TIME
Add an implementation for CPU_TIME using the POSIX function
clock_gettime. I think on most POSIX systems this will be included for
free via <ctime>, which corresponds to "time.h" (YMMV, we can fix the
code if the need arises).
Detecting that clock_gettime is available is tricky. For instance, commit
827407a86aa07 used the following incantation in f18-parse-demo.cpp:
#if _POSIX_C_SOURCE >= 199309L && _POSIX_TIMERS > 0 && _POSIX_CPUTIME && \
defined CLOCK_PROCESS_CPUTIME_ID
This doesn't work on my AArch64 Ubuntu system, which provides
clock_gettime but doesn't define _POSIX_TIMERS. Since finding the right
combination of macros requires infinite time, patience and access to
sundry POSIX systems, we should probably try a different approach.
This patch attempts to use SFINAE instead of the preprocessor to choose
an implementation for CPU_TIME. We define a helper function template
which helps us check if clock_gettime is available (and has the
interface we expect). I hope the comments explain it well enough.
This approach has the advantage that it keeps the detection of
clock_gettime close to the code that uses it. An alternative would be to
use CMake to check for the symbol (I personally haven't used this before
so I don't know if there are any quirks).
Differential Revision: https://reviews.llvm.org/D104020
Added:
Modified:
flang/runtime/time-intrinsic.cpp
Removed:
################################################################################
diff --git a/flang/runtime/time-intrinsic.cpp b/flang/runtime/time-intrinsic.cpp
index 00d51a5bea86..bd84ef9bdb42 100644
--- a/flang/runtime/time-intrinsic.cpp
+++ b/flang/runtime/time-intrinsic.cpp
@@ -12,14 +12,29 @@
#include <ctime>
-namespace Fortran::runtime {
-extern "C" {
-
// CPU_TIME (Fortran 2018 16.9.57)
-double RTNAME(CpuTime)() {
- // This is part of the c++ standard, so it should at least exist everywhere.
- // It probably does not have the best resolution, so we prefer other
- // platform-specific alternatives if they exist.
+// We can use std::clock() from the <ctime> header as a fallback implementation
+// that should be available everywhere. This may not provide the best resolution
+// and is particularly troublesome on (some?) POSIX systems where CLOCKS_PER_SEC
+// is defined as 10^6 regardless of the actual precision of std::clock().
+// Therefore, we will usually prefer platform-specific alternatives when they
+// are available.
+//
+// We can use SFINAE to choose a platform-specific alternative. To do so, we
+// introduce a helper function template, whose overload set will contain only
+// implementations relying on interfaces which are actually available. Each
+// overload will have a dummy parameter whose type indicates whether or not it
+// should be preferred. Any other parameters required for SFINAE should have
+// default values provided.
+namespace {
+// Types for the dummy parameter indicating the priority of a given overload.
+// We will invoke our helper with an integer literal argument, so the overload
+// with the highest priority should have the type int.
+using fallback_implementation = double;
+using preferred_implementation = int;
+
+// This is the fallback implementation, which should work everywhere.
+template <typename Unused = void> double getCpuTime(fallback_implementation) {
std::clock_t timestamp{std::clock()};
if (timestamp != std::clock_t{-1}) {
return static_cast<double>(timestamp) / CLOCKS_PER_SEC;
@@ -28,5 +43,36 @@ double RTNAME(CpuTime)() {
// Return some negative value to represent failure.
return -1.0;
}
+
+// POSIX implementation using clock_gettime. This is only enabled if
+// clock_gettime is available.
+template <typename T = int, typename U = struct timespec>
+double getCpuTime(preferred_implementation,
+ // We need some dummy parameters to pass to decltype(clock_gettime).
+ T ClockId = 0, U *Timespec = nullptr,
+ decltype(clock_gettime(ClockId, Timespec)) *Enabled = nullptr) {
+#if defined CLOCK_THREAD_CPUTIME_ID
+#define CLOCKID CLOCK_THREAD_CPUTIME_ID
+#elif defined CLOCK_PROCESS_CPUTIME_ID
+#define CLOCKID CLOCK_PROCESS_CPUTIME_ID
+#elif defined CLOCK_MONOTONIC
+#define CLOCKID CLOCK_MONOTONIC
+#else
+#define CLOCKID CLOCK_REALTIME
+#endif
+ struct timespec tspec;
+ if (clock_gettime(CLOCKID, &tspec) == 0) {
+ return tspec.tv_nsec * 1.0e-9 + tspec.tv_sec;
+ }
+
+ // Return some negative value to represent failure.
+ return -1.0;
+}
+} // anonymous namespace
+
+namespace Fortran::runtime {
+extern "C" {
+
+double RTNAME(CpuTime)() { return getCpuTime(0); }
} // extern "C"
} // namespace Fortran::runtime
More information about the flang-commits
mailing list