Merge pull request #10522 from sandipmgiri/master

np.float64("np.inf").astype(np.int32) is negative on x86 but positive on ppc64le
ShiSanChuan · Jun 30, 2017 · d0d2308 · d0d2308
2 parents 6fff4cb + c8d78a1
commit d0d2308
Show file tree

Hide file tree

Showing 4 changed files with 35 additions and 10 deletions.
diff --git a/tensorflow/core/platform/profile_utils/cpu_utils.cc b/tensorflow/core/platform/profile_utils/cpu_utils.cc
@@ -28,10 +28,17 @@ namespace profile_utils {
 
 static ICpuUtilsHelper* cpu_utils_helper_instance_ = nullptr;
 
-/* static */ int64 CpuUtils::GetCycleCounterFrequency() {
-  static const int64 cpu_frequency = GetCycleCounterFrequencyImpl();
-  return cpu_frequency;
+#if defined(__powerpc__) || defined(__ppc__) && ( __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)
+   /* static */ uint64 CpuUtils::GetCycleCounterFrequency() {
+     static const uint64 cpu_frequency = GetCycleCounterFrequencyImpl();
+     return cpu_frequency;
 }
+#else
+   /* static */ int64 CpuUtils::GetCycleCounterFrequency() {
+     static const int64 cpu_frequency = GetCycleCounterFrequencyImpl();
+     return cpu_frequency;
+}
+#endif
 
 /* static */ double CpuUtils::GetMicroSecPerClock() {
   static const double micro_sec_per_clock =

diff --git a/tensorflow/core/platform/profile_utils/cpu_utils.h b/tensorflow/core/platform/profile_utils/cpu_utils.h
@@ -97,7 +97,11 @@ class CpuUtils {
   // Return cycle counter frequency.
   // As this method caches the cpu frequency internally,
   // the first call will incur overhead, but not subsequent calls.
-  static int64 GetCycleCounterFrequency();
+  #if defined(__powerpc__) || defined(__ppc__) && ( __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)
+     static uint64 GetCycleCounterFrequency();
+  #else
+     static int64 GetCycleCounterFrequency();
+  #endif
 
   // Return micro secound per each clock
   // As this method caches the cpu frequency internally,

diff --git a/tensorflow/core/platform/profile_utils/cpu_utils_test.cc b/tensorflow/core/platform/profile_utils/cpu_utils_test.cc
@@ -53,9 +53,15 @@ TEST_F(CpuUtilsTest, CheckGetCurrentClockCycle) {
 }
 
 TEST_F(CpuUtilsTest, CheckCycleCounterFrequency) {
-  const int64 cpu_frequency = CpuUtils::GetCycleCounterFrequency();
-  CHECK_GT(cpu_frequency, 0);
-  CHECK_NE(cpu_frequency, CpuUtils::INVALID_FREQUENCY);
+  #if defined(__powerpc__) || defined(__ppc__) && ( __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)
+     const uint64 cpu_frequency = CpuUtils::GetCycleCounterFrequency();
+     CHECK_GT(cpu_frequency, 0);
+     CHECK_NE(cpu_frequency, unsigned(CpuUtils::INVALID_FREQUENCY));
+  #else
+     const int64 cpu_frequency = CpuUtils::GetCycleCounterFrequency();
+     CHECK_GT(cpu_frequency, 0);
+     CHECK_NE(cpu_frequency, CpuUtils::INVALID_FREQUENCY);
+  #endif
   if (DBG) {
     LOG(INFO) << "Cpu frequency = " << cpu_frequency;
   }

diff --git a/tensorflow/python/kernel_tests/cast_op_test.py b/tensorflow/python/kernel_tests/cast_op_test.py
@@ -20,6 +20,7 @@
 
 import numpy as np
 import sys
+import platform
 
 from tensorflow.python.framework import constant_op
 from tensorflow.python.framework import dtypes
@@ -146,9 +147,16 @@ def testInfNan(self):
     if sys.byteorder == "big":  
       self._compare(np.inf, np.int32, i4.max, False)  
       self._compare(np.inf, np.int64, i8.max, False)  
-    else:  
-      self._compare(np.inf, np.int32, i4.min, False)  
-      self._compare(np.inf, np.int64, i8.min, False)  
+    else:
+      # np.float64("np.inf").astype(np.int32) is negative on x86 but positive on ppc64le
+      # Numpy link to relevant discussion - https://github.com/numpy/numpy/issues/9040
+      # Tensorflow link to relevant discussion - https://github.com/tensorflow/tensorflow/issues/9360
+      if platform.machine() == "ppc64le":
+        self._compare(-np.inf, np.int32, i4.min, False)
+        self._compare(-np.inf, np.int64, i8.min, False)
+      else:
+        self._compare(np.inf, np.int32, i4.min, False)
+        self._compare(np.inf, np.int64, i8.min, False)  
     self._compare(-np.inf, np.float32, -np.inf, False)
     self._compare(-np.inf, np.float64, -np.inf, False)
     self._compare(-np.inf, np.int32, i4.min, False)