[Openmp-commits] [openmp] r308423 - OpenMP RTL cleanup: two PAUSEs per spin loop iteration replaced with single one

[Andrey Churbanov via Openmp-commits]

Author: achurbanov
Date: Wed Jul 19 02:26:13 2017
New Revision: 308423

URL: http://llvm.org/viewvc/llvm-project?rev=308423&view=rev
Log:
OpenMP RTL cleanup: two PAUSEs per spin loop iteration replaced with single one

Differential Revision: https://reviews.llvm.org/D35490

Modified:
    openmp/trunk/runtime/src/kmp.h
    openmp/trunk/runtime/src/kmp_wait_release.h

Modified: openmp/trunk/runtime/src/kmp.h
URL: http://llvm.org/viewvc/llvm-project/openmp/trunk/runtime/src/kmp.h?rev=308423&r1=308422&r2=308423&view=diff
==============================================================================
--- openmp/trunk/runtime/src/kmp.h (original)
+++ openmp/trunk/runtime/src/kmp.h Wed Jul 19 02:26:13 2017
@@ -1040,7 +1040,11 @@ extern void __kmp_x86_cpuid(int mode, in
 #if KMP_ARCH_X86
 extern void __kmp_x86_pause(void);
 #elif KMP_MIC
-static void __kmp_x86_pause(void) { _mm_delay_32(100); }
+// Performance testing on KNC (C0QS-7120 P/A/X/D, 61-core, 16 GB Memory) showed
+// regression after removal of extra PAUSE from KMP_YIELD_SPIN(). Changing
+// the delay from 100 to 300 showed even better performance than double PAUSE
+// on Spec OMP2001 and LCPC tasking tests, no regressions on EPCC.
+static void __kmp_x86_pause(void) { _mm_delay_32(300); }
 #else
 static void __kmp_x86_pause(void) { _mm_pause(); }
 #endif
@@ -1076,7 +1080,7 @@ static void __kmp_x86_pause(void) { _mm_
     KMP_CPU_PAUSE();                                                           \
     (count) -= 2;                                                              \
     if (!(count)) {                                                            \
-      KMP_YIELD(cond);                                                         \
+      __kmp_yield(cond);                                                       \
       (count) = __kmp_yield_next;                                              \
     }                                                                          \
   }
@@ -1085,7 +1089,7 @@ static void __kmp_x86_pause(void) { _mm_
     KMP_CPU_PAUSE();                                                           \
     (count) -= 2;                                                              \
     if (!(count)) {                                                            \
-      KMP_YIELD(1);                                                            \
+      __kmp_yield(1);                                                          \
       (count) = __kmp_yield_next;                                              \
     }                                                                          \
   }

Modified: openmp/trunk/runtime/src/kmp_wait_release.h
URL: http://llvm.org/viewvc/llvm-project/openmp/trunk/runtime/src/kmp_wait_release.h?rev=308423&r1=308422&r2=308423&view=diff
==============================================================================
--- openmp/trunk/runtime/src/kmp_wait_release.h (original)
+++ openmp/trunk/runtime/src/kmp_wait_release.h Wed Jul 19 02:26:13 2017
@@ -47,7 +47,7 @@ enum flag_type {
  */
 template <typename P> class kmp_flag {
   volatile P
-    *loc; /**< Pointer to the flag storage that is modified by another thread
+      *loc; /**< Pointer to the flag storage that is modified by another thread
              */
   flag_type t; /**< "Type" of the flag in loc */
 public:
@@ -225,11 +225,14 @@ __kmp_wait_template(kmp_info_t *this_thr
 
     // If we are oversubscribed, or have waited a bit (and
     // KMP_LIBRARY=throughput), then yield
-    KMP_YIELD(oversubscribed);
     // TODO: Should it be number of cores instead of thread contexts? Like:
     // KMP_YIELD(TCR_4(__kmp_nth) > __kmp_ncores);
     // Need performance improvement data to make the change...
-    KMP_YIELD_SPIN(spins);
+    if (oversubscribed) {
+      KMP_YIELD(1);
+    } else {
+      KMP_YIELD_SPIN(spins);
+    }
     // Check if this thread was transferred from a team
     // to the thread pool (or vice-versa) while spinning.
     in_pool = !!TCR_4(this_thr->th.th_in_pool);