Symbols, device flags, versions, choose-device#

Defined in code/simulator.h.

This page documents 8 symbols: 8 functions.

Functions#

Signature

Description

Availability

Location

Example

inline cudaError_t cudaChooseDevice(int* device, const cudaDeviceProp* )

sim only

simulator.h:2275

inline cudaError_t cudaDriverGetVersion(int* v)

sim only

simulator.h:2273

inline cudaError_t cudaGetDeviceFlags(unsigned* flags)

sim only

simulator.h:2272

inline cudaError_t cudaGetSymbolAddress(void** devPtr, void* symbol)

sim only

simulator.h:2269

View

inline cudaError_t cudaGetSymbolSize(size_t* size, const T& symbol)

sim only

simulator.h:2270

inline cudaError_t cudaMemGetInfo(size_t* freeMem, size_t* totalMem)

sim only

simulator.h:2276

inline cudaError_t cudaRuntimeGetVersion(int* v)

sim only

simulator.h:2274

inline cudaError_t cudaSetDeviceFlags(unsigned )

sim only

simulator.h:2271

View

Examples#

Code from the repository’s example corpus (code\*.cu); the highlighted line is the call site. These blocks are what the View links above open.

cudaGetSymbolAddress (test_web_enccs_reduce_static.cu:85)
75        y += g.shfl_down(y, i);
76
77    if (tid == 0)
78        d_y[bid] = y;
79}
80
81// host driver (the cudaEvent timing loop is dropped; runs the two passes once)
82static real reduce(const real *d_x)
83{
84    real *d_y;
85    CHECK(cudaMalloc(&d_y, sizeof(real) * GRID_SIZE));   // [SIM] was: cudaGetSymbolAddress(static_y)
86
87    const int smem = sizeof(real) * BLOCK_SIZE;
88    LAUNCH_DYN(reduce_cp, GRID_SIZE, BLOCK_SIZE, smem, d_x, d_y, NX);            // [SIM]
89    LAUNCH_DYN(reduce_cp, 1, 1024, sizeof(real) * 1024, d_y, d_y, GRID_SIZE);   // [SIM]
90
91    real h_y[1] = {0};
92    CHECK(cudaMemcpy(h_y, d_y, sizeof(real), cudaMemcpyDeviceToHost));
93    CHECK(cudaFree(d_y));
94    return h_y[0];
95}
cudaSetDeviceFlags (test_web_eeg_mapped.cu:111)
101    }
102    cudaDeviceProp prop{}; cudaGetDeviceProperties(&prop, 0);
103    std::printf("Device 0: %s (compute %d.%d)\n", prop.name, prop.major, prop.minor);
104
105    Vector<float> h_a, h_b;
106    h_a.length = FULL_DATA_SIZE;
107    h_b.length = FULL_DATA_SIZE;
108
109    // allocate pinned host memory accessible by the device
110#ifdef __CUDACC__
111    cudaSetDeviceFlags(cudaDeviceMapHost);
112    cudaHostAlloc((void**)&h_a.elements, FULL_ARRAY_BYTES,
113                  cudaHostAllocWriteCombined | cudaHostAllocMapped);
114    cudaHostAlloc((void**)&h_b.elements, FULL_ARRAY_BYTES,
115                  cudaHostAllocWriteCombined | cudaHostAllocMapped);
116#else
117    cudaHostAlloc((void**)&h_a.elements, FULL_ARRAY_BYTES);   // [SIM] plain pinned alloc
118    cudaHostAlloc((void**)&h_b.elements, FULL_ARRAY_BYTES);
119#endif
120
121    for (int i = 0; i < FULL_DATA_SIZE; i++) {
122        h_a.setElement(i, 1.0f);
On this page