Franscobec/gpufetch
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[v0.11] Fix compilation error and ambiguity with CUDA and Intel backend when enabled at the same time due to functions with the same name

Browse Source
This commit is contained in:
Dr-Noob
2021-12-18 10:01:42 +01:00
parent 93889b2b18
commit 6d4d8b621b
10 changed files with 21 additions and 23 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
src/cuda/cuda.cpp
View File

@@ -76,12 +76,12 @@ struct memory* get_memory_info(struct gpu_info* gpu, cudaDeviceProp prop) {
}
// Compute peak performance when using CUDA cores
int64_t get_peak_performance(struct gpu_info* gpu) {
int64_t get_peak_performance_cuda(struct gpu_info* gpu) {
return gpu->freq * 1000000 * gpu->topo->cuda_cores * 2;
}
// Compute peak performance when using tensor cores
int64_t get_peak_performance_t(struct gpu_info* gpu) {
int64_t get_peak_performance_tcu(struct gpu_info* gpu) {
return gpu->freq * 1000000 * 4 * 4 * 8 * gpu->topo->tensor_cores;
}
@@ -138,8 +138,8 @@ struct gpu_info* get_gpu_info_cuda(int gpu_idx) {
gpu->cach = get_cache_info(deviceProp);
gpu->mem = get_memory_info(gpu, deviceProp);
gpu->topo = get_topology_info(deviceProp);
gpu->peak_performance = get_peak_performance(gpu);
gpu->peak_performance_t = get_peak_performance_t(gpu);
gpu->peak_performance = get_peak_performance_cuda(gpu);
gpu->peak_performance_tcu = get_peak_performance_tcu(gpu);
return gpu;
}

2
src/cuda/pci.cpp
View File

@@ -19,7 +19,7 @@
* or in pci.ids itself)
*/
GPUCHIP get_chip_from_pci(struct pci* pci) {
GPUCHIP get_chip_from_pci_cuda(struct pci* pci) {
CHECK_PCI_START
CHECK_PCI(pci, 0x25e5, CHIP_GA107BM)
CHECK_PCI(pci, 0x25e2, CHIP_GA107BM)

3
src/cuda/pci.hpp
View File

@@ -14,7 +14,6 @@
struct pci;
struct pci* get_pci_from_pciutils(struct pci_dev *devices);
GPUCHIP get_chip_from_pci(struct pci* pci);
GPUCHIP get_chip_from_pci_cuda(struct pci* pci);
#endif

9
src/cuda/uarch.cpp
View File

@@ -39,7 +39,7 @@ static const char *uarch_str[] = {
#define CHECK_UARCH_START if (false) {}
#define CHECK_UARCH(arch, chip_, str, uarch, process) \
else if (arch->chip == chip_) fill_uarch(arch, str, uarch, process);
#define CHECK_UARCH_END else { printBug("map_chip_to_uarch: Unknown chip id: %d", arch->chip); fill_uarch(arch, STRING_UNKNOWN, UARCH_UNKNOWN, 0); }
#define CHECK_UARCH_END else { printBug("map_chip_to_uarch_cuda: Unknown chip id: %d", arch->chip); fill_uarch(arch, STRING_UNKNOWN, UARCH_UNKNOWN, 0); }
void fill_uarch(struct uarch* arch, char const *str, MICROARCH u, uint32_t process) {
arch->chip_str = (char *) emalloc(sizeof(char) * (strlen(str)+1));
@@ -54,7 +54,7 @@ void fill_uarch(struct uarch* arch, char const *str, MICROARCH u, uint32_t proce
* o CHIP_XXXGL: indicates a professional-class (Quadro/Tesla) chip
* o CHIP_XXXM: indicates a mobile chip
*/
void map_chip_to_uarch(struct uarch* arch) {
void map_chip_to_uarch_cuda(struct uarch* arch) {
CHECK_UARCH_START
// TESLA (1.0, 1.1, 1.2, 1.3) //
CHECK_UARCH(arch, CHIP_G80, "G80", UARCH_TESLA, 90)
@@ -243,9 +243,8 @@ struct uarch* get_uarch_from_cuda(struct gpu_info* gpu) {
arch->cc_major = deviceProp.major;
arch->cc_minor = deviceProp.minor;
arch->compute_capability = deviceProp.major * 10 + deviceProp.minor;
arch->chip = get_chip_from_pci(gpu->pci);
map_chip_to_uarch(arch);
arch->chip = get_chip_from_pci_cuda(gpu->pci);
map_chip_to_uarch_cuda(arch);
return arch;
}