Some fixes for building on windows + cuda (#1614)

Summary: Upstreaming some work towards #1586 from conda-forge/faiss-split-feedstock#19; complements #1610 Pull Request resolved: #1614 Reviewed By: mdouze Differential Revision: D25864994 Pulled By: wickedfoo fbshipit-source-id: 4abaaebf1c4102c24c3a9dca544744318e780581
facebookresearch · Jan 11, 2021 · 3d3d539 · 3d3d539
1 parent 1c996d0
commit 3d3d539
Show file tree

Hide file tree

Showing 6 changed files with 12 additions and 12 deletions.
diff --git a/faiss/gpu/impl/PQCodeLoad.cuh b/faiss/gpu/impl/PQCodeLoad.cuh
@@ -49,7 +49,7 @@ inline __device__ unsigned int getByte(unsigned int v,
   return getBitfield(v, pos, width);
 }
 
-inline __device__ unsigned int getByte(unsigned long v,
+inline __device__ unsigned int getByte(uint64_t v,
                                        int pos,
                                        int width) {
   return getBitfield(v, pos, width);

diff --git a/faiss/gpu/utils/PtxUtils.cuh b/faiss/gpu/utils/PtxUtils.cuh
@@ -27,8 +27,8 @@ unsigned int getBitfield(unsigned int val, int pos, int len) {
 }
 
 __device__ __forceinline__
-unsigned long getBitfield(unsigned long val, int pos, int len) {
-  unsigned long ret;
+uint64_t getBitfield(uint64_t val, int pos, int len) {
+  uint64_t ret;
   asm("bfe.u64 %0, %1, %2, %3;" : "=l"(ret) : "l"(val), "r"(pos), "r"(len));
   return ret;
 }

diff --git a/faiss/gpu/utils/StackDeviceMemory.cpp b/faiss/gpu/utils/StackDeviceMemory.cpp
@@ -10,6 +10,7 @@
 #include <faiss/gpu/utils/DeviceUtils.h>
 #include <faiss/gpu/utils/StaticUtils.h>
 #include <faiss/impl/FaissAssert.h>
+#include <algorithm>
 #include <sstream>
 
 namespace faiss { namespace gpu {

diff --git a/faiss/gpu/utils/Tensor.cuh b/faiss/gpu/utils/Tensor.cuh
@@ -374,7 +374,7 @@ bool canUseIndexType() {
 
 template <typename IndexType, typename T, typename... U>
 bool canUseIndexType(const T& arg, const U&... args) {
-  return arg.template canUseIndexType() &&
+  return arg.template canUseIndexType<IndexType>() &&
     canUseIndexType(args...);
 }
 

diff --git a/faiss/gpu/utils/Timer.cpp b/faiss/gpu/utils/Timer.cpp
@@ -9,6 +9,7 @@
 #include <faiss/gpu/utils/Timer.h>
 #include <faiss/gpu/utils/DeviceUtils.h>
 #include <faiss/impl/FaissAssert.h>
+#include <chrono>
 
 namespace faiss { namespace gpu {
 
@@ -43,18 +44,16 @@ KernelTimer::elapsedMilliseconds() {
 }
 
 CpuTimer::CpuTimer() {
-  clock_gettime(CLOCK_REALTIME, &start_);
+  start_ = std::chrono::steady_clock::now();
 }
 
 float
 CpuTimer::elapsedMilliseconds() {
-  struct timespec end;
-  clock_gettime(CLOCK_REALTIME, &end);
+  auto end = std::chrono::steady_clock::now();
 
-  auto diffS = end.tv_sec - start_.tv_sec;
-  auto diffNs = end.tv_nsec - start_.tv_nsec;
+  std::chrono::duration<float, std::milli> duration = end - start_;
 
-  return 1000.0f * (float) diffS + ((float) diffNs) / 1000000.0f;
+  return duration.count();
 }
 
 } } // namespace
diff --git a/faiss/gpu/utils/Timer.h b/faiss/gpu/utils/Timer.h
@@ -9,7 +9,7 @@
 #pragma once
 
 #include <cuda_runtime.h>
-#include <time.h>
+#include <chrono>
 
 namespace faiss { namespace gpu {
 
@@ -46,7 +46,7 @@ class CpuTimer {
   float elapsedMilliseconds();
 
  private:
-  struct timespec start_;
+  std::chrono::time_point<std::chrono::steady_clock> start_;
 };
 
 } } // namespace