NASP/cuda-samples
85
0
Fork 0
mirror of https://github.com/NVIDIA/cuda-samples.git synced 2026-10-25 22:28:17 +08:00
Code Issues Packages Projects Releases Wiki Activity

Apply consistent code formatting across the repo. Add clang-format and pre-commit hooks.

Browse Source
This commit is contained in:
Rob Armstrong
2025-03-27 10:30:07 -07:00
parent 2cd58fbc9a
commit ceab6e8bcc
782 changed files with 107230 additions and 106548 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1040
Samples/4_CUDA_Libraries/batchCUBLAS/batchCUBLAS.cpp
View File

File diff suppressed because it is too large Load Diff

152
Samples/4_CUDA_Libraries/batchCUBLAS/batchCUBLAS.h
View File

@@ -37,8 +37,8 @@
#define SWITCH_CHAR '-'
#define REFFUNC(funcname) ref_##funcname
#define TESTGEN(funcname) get_##funcname##_params
#define REFFUNC(funcname) ref_##funcname
#define TESTGEN(funcname) get_##funcname##_params
#define TESTPARAMS(funcname) funcname##TestParams
#define DEV_VER_DBL_SUPPORT (130)
@@ -54,10 +54,11 @@
//==============================================================================
#if defined(__cplusplus)
extern "C" {
extern "C"
{
#endif /* __cplusplus */
int getDeviceVersion(void);
size_t getDeviceMemory(void);
int getDeviceVersion(void);
size_t getDeviceMemory(void);
#if defined(__cplusplus)
}
#endif /* __cplusplus */
@@ -68,71 +69,58 @@ size_t getDeviceMemory(void);
static __inline__ int imax(int x, int y) { return (x > y) ? x : y; }
static __inline__ unsigned floatAsUInt(float x) {
volatile union {
float f;
unsigned i;
} xx;
xx.f = x;
return xx.i;
static __inline__ unsigned floatAsUInt(float x)
{
volatile union
{
float f;
unsigned i;
} xx;
xx.f = x;
return xx.i;
}
static __inline__ unsigned long long doubleAsULL(double x) {
volatile union {
double f;
unsigned long long i;
} xx;
xx.f = x;
return xx.i;
static __inline__ unsigned long long doubleAsULL(double x)
{
volatile union
{
double f;
unsigned long long i;
} xx;
xx.f = x;
return xx.i;
}
static __inline__ unsigned cuRand(void) {
/* George Marsaglia's fast inline random number generator */
static __inline__ unsigned cuRand(void)
{
/* George Marsaglia's fast inline random number generator */
#define CUDA_ZNEW (cuda_z = 36969 * (cuda_z & 65535) + (cuda_z >> 16))
#define CUDA_WNEW (cuda_w = 18000 * (cuda_w & 65535) + (cuda_w >> 16))
#define CUDA_MWC ((CUDA_ZNEW << 16) + CUDA_WNEW)
#define CUDA_SHR3 \
(cuda_jsr = cuda_jsr ^ (cuda_jsr << 17), \
cuda_jsr = cuda_jsr ^ (cuda_jsr >> 13), \
cuda_jsr = cuda_jsr ^ (cuda_jsr << 5))
#define CUDA_MWC ((CUDA_ZNEW << 16) + CUDA_WNEW)
#define CUDA_SHR3 \
(cuda_jsr = cuda_jsr ^ (cuda_jsr << 17), \
cuda_jsr = cuda_jsr ^ (cuda_jsr >> 13), \
cuda_jsr = cuda_jsr ^ (cuda_jsr << 5))
#define CUDA_CONG (cuda_jcong = 69069 * cuda_jcong + 1234567)
#define KISS ((CUDA_MWC ^ CUDA_CONG) + CUDA_SHR3)
static unsigned int cuda_z = 362436069, cuda_w = 521288629;
static unsigned int cuda_jsr = 123456789, cuda_jcong = 380116160;
return KISS;
#define KISS ((CUDA_MWC ^ CUDA_CONG) + CUDA_SHR3)
static unsigned int cuda_z = 362436069, cuda_w = 521288629;
static unsigned int cuda_jsr = 123456789, cuda_jcong = 380116160;
return KISS;
}
//==============================================================================
// cuGet and cuEqual versions
//==============================================================================
template <typename T_ELEM>
__inline__ __device__ __host__ T_ELEM cuGet(double);
template <>
__inline__ __device__ __host__ float cuGet<float>(double x) {
return float(x);
}
template <>
__inline__ __device__ __host__ double cuGet<double>(double x) {
return double(x);
}
template <typename T_ELEM> __inline__ __device__ __host__ T_ELEM cuGet(double);
template <> __inline__ __device__ __host__ float cuGet<float>(double x) { return float(x); }
template <> __inline__ __device__ __host__ double cuGet<double>(double x) { return double(x); }
template <typename T_ELEM>
__inline__ __device__ __host__ T_ELEM cuGet(double, double);
template <>
__inline__ __device__ __host__ float cuGet<float>(double x, double y) {
return float(x);
}
template <>
__inline__ __device__ __host__ double cuGet<double>(double x, double y) {
return double(x);
}
static __inline__ __device__ __host__ bool cuEqual(float x, float y) {
return (x == y);
}
static __inline__ __device__ __host__ bool cuEqual(double x, double y) {
return (x == y);
}
template <typename T_ELEM> __inline__ __device__ __host__ T_ELEM cuGet(double, double);
template <> __inline__ __device__ __host__ float cuGet<float>(double x, double y) { return float(x); }
template <> __inline__ __device__ __host__ double cuGet<double>(double x, double y) { return double(x); }
static __inline__ __device__ __host__ bool cuEqual(float x, float y) { return (x == y); }
static __inline__ __device__ __host__ bool cuEqual(double x, double y) { return (x == y); }
//==============================================================================
// Platform dependent timing utility
@@ -143,33 +131,36 @@ static __inline__ __device__ __host__ bool cuEqual(double x, double y) {
#define WIN32_LEAN_AND_MEAN
#endif
#include <windows.h>
static __inline__ double second(void) {
LARGE_INTEGER t;
static double oofreq;
static int checkedForHighResTimer;
static BOOL hasHighResTimer;
static __inline__ double second(void)
{
LARGE_INTEGER t;
static double oofreq;
static int checkedForHighResTimer;
static BOOL hasHighResTimer;
if (!checkedForHighResTimer) {
hasHighResTimer = QueryPerformanceFrequency(&t);
oofreq = 1.0 / (double)t.QuadPart;
checkedForHighResTimer = 1;
}
if (!checkedForHighResTimer) {
hasHighResTimer = QueryPerformanceFrequency(&t);
oofreq = 1.0 / (double)t.QuadPart;
checkedForHighResTimer = 1;
}
if (hasHighResTimer) {
QueryPerformanceCounter(&t);
return (double)t.QuadPart * oofreq;
} else {
return (double)GetTickCount() / 1000.0;
}
if (hasHighResTimer) {
QueryPerformanceCounter(&t);
return (double)t.QuadPart * oofreq;
}
else {
return (double)GetTickCount() / 1000.0;
}
}
#elif defined(__linux__) || defined(__QNX__)
#include <stddef.h>
#include <sys/resource.h>
#include <sys/time.h>
static double second(void) {
struct timeval tv;
gettimeofday(&tv, NULL);
return (double)tv.tv_sec + (double)tv.tv_usec / 1000000.0;
static double second(void)
{
struct timeval tv;
gettimeofday(&tv, NULL);
return (double)tv.tv_sec + (double)tv.tv_usec / 1000000.0;
}
#elif defined(__APPLE__)
#include <stddef.h>
@@ -177,10 +168,11 @@ static double second(void) {
#include <sys/sysctl.h>
#include <sys/time.h>
#include <sys/types.h>
static double second(void) {
struct timeval tv;
gettimeofday(&tv, NULL);
return (double)tv.tv_sec + (double)tv.tv_usec / 1000000.0;
static double second(void)
{
struct timeval tv;
gettimeofday(&tv, NULL);
return (double)tv.tv_sec + (double)tv.tv_usec / 1000000.0;
}
#else
#error unsupported platform