deftdawg/bitsandbytes-rocm
1
0
Fork 0
forked from mrq/bitsandbytes-rocm
Code Pull Requests Activity

Fixed issue where Pascal was not displaying proper error.

Browse Source
This commit is contained in:
Tim Dettmers 2022-08-23 16:00:26 -07:00
parent 7e0fb655e1
commit ee5b947e63
2 changed files with 7 additions and 22 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

23
bitsandbytes/functional.py
View File

@ -185,14 +185,9 @@ def create_dynamic_map(signed=True, n=7):
def get_special_format_str(): def get_special_format_str():
if not torch.cuda.is_available(): return 'col_turing'
major, minor = torch.cuda.get_device_capability() major, minor = torch.cuda.get_device_capability()
if major < 7: if major <= 7:
print(
f"Device with CUDA capability of {major} not supported for 8-bit matmul. Device has no tensor cores!"
)
assert major >= 7
if major == 7:
return "col_turing" return "col_turing"
elif major == 8: elif major == 8:
return "col_ampere" return "col_ampere"
@ -1685,20 +1680,6 @@ def double_quant(
return out_row, out_col, row_stats, col_stats, coo_tensor return out_row, out_col, row_stats, col_stats, coo_tensor
def get_special_format_str():
if not torch.cuda.is_available(): return 'col_turning'
major, minor = torch.cuda.get_device_capability()
if major < 7:
print(f"Device with CUDA capability of {major} not supported for 8-bit matmul. Device has no tensor cores!")
assert major >= 7
if major == 7: return 'col_turing'
elif major == 8: return 'col_ampere'
else: return 'col_turing'
def transform(A, to_order, from_order='row', out=None, transpose=False, state=None, ld=None): def transform(A, to_order, from_order='row', out=None, transpose=False, state=None, ld=None):
prev_device = pre_call(A.device) prev_device = pre_call(A.device)
if state is None: state = (A.shape, from_order) if state is None: state = (A.shape, from_order)

6
csrc/ops.cu
View File

@ -371,7 +371,11 @@ template void transform<int32_t, COL32, ROW, false, 32>(cublasLtHandle_t ltHandl
template <int FORMATB, int DTYPE_OUT, int SCALE_ROWS> int igemmlt(cublasLtHandle_t ltHandle, int m, int n, int k, const int8_t *A, const int8_t *B, void *C, float *row_scale, int lda, int ldb, int ldc) template <int FORMATB, int DTYPE_OUT, int SCALE_ROWS> int igemmlt(cublasLtHandle_t ltHandle, int m, int n, int k, const int8_t *A, const int8_t *B, void *C, float *row_scale, int lda, int ldb, int ldc)
{ {
#ifdef NO_CUBLASLT #ifdef NO_CUBLASLT
printf("ERROR: Your GPU does not support Int8 Matmul!"); cout << "" << endl;
cout << "=============================================" << endl;
cout << "ERROR: Your GPU does not support Int8 Matmul!" << endl;
cout << "=============================================" << endl;
cout << "" << endl;
assert(false); assert(false);
return 0; return 0;