frozenleaves/pytorch
1
0
Fork 0
mirror of https://github.com/pytorch/pytorch.git synced 2025-10-20 21:14:14 +08:00
Code Issues Packages Projects Releases Wiki Activity

build: set -DNDEBUG in Release (#32719)

Browse Source 
Summary:
This might lead to silent undefined behaviour (e.g. with out-of-bound indices). This affects `test_multinomial_invalid_probs_cuda` which is now removed.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/32719

Test Plan:
* Build with VERBOSE=1 and manually inspect `less ndebug.build.log | grep 'c++' | grep -v -- -DNDEBUG` (only with nina on Linux)
* CI

Fixes https://github.com/pytorch/pytorch/issues/22745

Differential Revision: D20104340

Pulled By: yf225

fbshipit-source-id: 2ebfd7ddae632258a36316999eeb5c968fb7642c
This commit is contained in:
Wojciech Baranowski
2020-02-26 12:48:06 -08:00
committed by Facebook Github Bot
parent 93e30c16cb
commit 8aa09de19e
3 changed files with 57 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
aten/src/ATen/native/cuda/MultinomialKernel.cu
View File

@ -35,13 +35,13 @@ __global__ void renormRowsL1(scalar_t* dist, long rows, long cols) {
scalar_t sum = static_cast<scalar_t>(0);
for (int64_t col = threadIdx.x; col < cols; col += blockDim.x) {
val = dist[row * cols + col];
assert(!THCNumerics<scalar_t>::lt(val, zero)); // ! < 0 for NaN handling
CUDA_ALWAYS_ASSERT(!THCNumerics<scalar_t>::lt(val, zero)); // ! < 0 for NaN handling
sum = sum + val;
}
sum = reduceBlock(smem, blockDim.x, sum, ReduceAdd<scalar_t>(), zero);
if (threadIdx.x == 0) {
assert(!THCNumerics<scalar_t>::lt(val, zero)); // ! < 0 for NaN handling
CUDA_ALWAYS_ASSERT(!THCNumerics<scalar_t>::lt(val, zero)); // ! < 0 for NaN handling
smem[0] = sum;
}
__syncthreads();
@ -61,7 +61,7 @@ void renormRows(Tensor& t) {
int64_t cols = t.size(1);
auto props = at::cuda::getCurrentDeviceProperties();
assert(props != NULL);
CUDA_ALWAYS_ASSERT(props != NULL);
int numSM = props->multiProcessorCount;
int maxThreads = props->maxThreadsPerBlock;
@ -84,7 +84,7 @@ __device__ int binarySearchForMultinomial(scalar_t* cumdist,
int start = 0;
int end = size;
// cumdist[size - 1] = 0 => all zero prob dist
assert(cumdist[size - 1] > static_cast<scalar_t>(0));
CUDA_ALWAYS_ASSERT(cumdist[size - 1] > static_cast<scalar_t>(0));
while (end - start > 0) {
int mid = start + (end - start) / 2;
@ -240,9 +240,9 @@ sampleMultinomialOnce(int64_t* dest,
scalar_t val;
for (int cat = threadIdx.x; cat < categories; cat += blockDim.x) {
val = dist[curDist * stride_dist + cat * stride_categories];
assert(val >= zero);
assert(!THCNumerics<scalar_t>::isinf(val));
assert(!THCNumerics<scalar_t>::isnan(val));
CUDA_ALWAYS_ASSERT(val >= zero);
CUDA_ALWAYS_ASSERT(!THCNumerics<scalar_t>::isinf(val));
CUDA_ALWAYS_ASSERT(!THCNumerics<scalar_t>::isnan(val));
sum = sum + static_cast<accscalar_t>(val);
}
@ -252,8 +252,8 @@ sampleMultinomialOnce(int64_t* dest,
// Broadcast sum and sample value
if (threadIdx.x == 0) {
// Make sure the sum of our distribution didn't overflow
assert(!THCNumerics<accscalar_t>::isinf(sum));
assert(sum > accZero);
CUDA_ALWAYS_ASSERT(!THCNumerics<accscalar_t>::isinf(sum));
CUDA_ALWAYS_ASSERT(sum > accZero);
asmem[0] = sum;
smem[0] = sampled[curDist];
@ -363,7 +363,7 @@ void multinomial_kernel_impl(Tensor& result, const Tensor& self, const int64_t n
AT_DISPATCH_FLOATING_TYPES_AND_HALF(self_v.scalar_type(), "multinomial_kernel_cuda", [&] {
using accscalar_t = at::acc_type<scalar_t, true>;
auto props = at::cuda::getCurrentDeviceProperties();
assert(props != NULL);
CUDA_ALWAYS_ASSERT(props != NULL);
int numSM = props->multiProcessorCount;
int maxThreads = props->maxThreadsPerBlock;
int maxShared = props->sharedMemPerBlock;

33
c10/macros/Macros.h
View File

@ -192,6 +192,10 @@ constexpr uint32_t CUDA_THREADS_PER_BLOCK_FALLBACK = 256;
#define C10_WARP_SIZE 32
#endif
#if defined(_MSC_VER) && _MSC_VER <= 1900
#define __func__ __FUNCTION__
#endif
// CUDA_KERNEL_ASSERT is a macro that wraps an assert() call inside cuda
// kernels. This is not supported by Apple platforms so we special case it.
// See http://docs.nvidia.com/cuda/cuda-c-programming-guide/#assertion
@ -201,6 +205,35 @@ constexpr uint32_t CUDA_THREADS_PER_BLOCK_FALLBACK = 256;
#define CUDA_KERNEL_ASSERT(...) assert(__VA_ARGS__)
#endif // __APPLE__
// CUDA_ALWAYS_ASSERT is similar to CUDA_KERNEL_ASSERT but checks the assertion
// even when NDEBUG is defined. This is useful for important assertions in CUDA
// code that when building Release.
#if defined(__APPLE__) || defined(__HIP_PLATFORM_HCC__)
// Those platforms do not support assert()
#define CUDA_ALWAYS_ASSERT(cond)
#elif defined(_MSC_VER)
// TODO: This should be defined but I don't have the environment to properly
// test it. See e.g., https://github.com/pytorch/pytorch/pull/32719#discussion_r379918384
#define CUDA_ALWAYS_ASSERT(cond)
#else // __APPLE__, _MSC_VER
#if defined(NDEBUG)
extern "C" {
[[noreturn]]
#if defined(__CUDA_ARCH__) || defined(__HIP_ARCH__) || defined(__HIP__)
__host__ __device__
#endif // __CUDA_ARCH__
void __assert_fail(const char *assertion, const char *file,
unsigned int line, const char *function)
throw();
}
#endif // NDEBUG
#define CUDA_ALWAYS_ASSERT(cond) \
if (C10_UNLIKELY(!(cond))) { \
__assert_fail(#cond, __FILE__, static_cast<unsigned int>(__LINE__), \
__func__); \
}
#endif // __APPLE__
#ifdef __APPLE__
#include <TargetConditionals.h>
#endif

17
cmake/Dependencies.cmake
View File

@ -1257,16 +1257,21 @@ if (NOT INTERN_BUILD_MOBILE)
MESSAGE(STATUS "Could not find CUDA with FP16 support, compiling without torch.CudaHalfTensor")
ENDIF()
OPTION(NDEBUG "disable asserts (WARNING: this may result in silent UB e.g. with out-of-bound indices)")
IF (NOT NDEBUG)
STRING(APPEND CMAKE_C_FLAGS_RELEASE " -DNDEBUG")
STRING(APPEND CMAKE_CXX_FLAGS_RELEASE " -DNDEBUG")
IF (NOT GENERATOR_IS_MULTI_CONFIG)
IF (${CMAKE_BUILD_TYPE} STREQUAL "Release")
MESSAGE(STATUS "Adding -DNDEBUG to compile flags")
STRING(APPEND CMAKE_C_FLAGS " -DNDEBUG")
STRING(APPEND CMAKE_CXX_FLAGS " -DNDEBUG")
ELSE()
MESSAGE(STATUS "Removing -DNDEBUG from compile flags")
STRING(REGEX REPLACE "[-/]DNDEBUG" "" CMAKE_C_FLAGS "" ${CMAKE_C_FLAGS})
STRING(REGEX REPLACE "[-/]DNDEBUG" "" CMAKE_C_FLAGS_DEBUG "" ${CMAKE_C_FLAGS_DEBUG})
STRING(REGEX REPLACE "[-/]DNDEBUG" "" CMAKE_C_FLAGS_RELEASE "" ${CMAKE_C_FLAGS_RELEASE})
STRING(REGEX REPLACE "[-/]DNDEBUG" "" CMAKE_CXX_FLAGS "" ${CMAKE_CXX_FLAGS})
STRING(REGEX REPLACE "[-/]DNDEBUG" "" CMAKE_CXX_FLAGS_DEBUG "" ${CMAKE_CXX_FLAGS_DEBUG})
STRING(REGEX REPLACE "[-/]DNDEBUG" "" CMAKE_CXX_FLAGS_RELEASE "" ${CMAKE_CXX_FLAGS_RELEASE})
ENDIF()
ENDIF()
STRING(REGEX REPLACE "[-/]DNDEBUG" "" CMAKE_C_FLAGS_DEBUG "" ${CMAKE_C_FLAGS_DEBUG})
STRING(REGEX REPLACE "[-/]DNDEBUG" "" CMAKE_CXX_FLAGS_DEBUG "" ${CMAKE_CXX_FLAGS_DEBUG})
SET(CUDA_ATTACH_VS_BUILD_RULE_TO_CUDA_FILE OFF)