Skip to content

replace tl.libdevice.llrint with tl.extra.cuda.libdevice.rint #1372

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 2 commits into
base: main
Choose a base branch
from
Open

replace tl.libdevice.llrint with tl.extra.cuda.libdevice.rint #1372

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
84e188e
replace tl.libdevice.llrint with tl.extra.cuda.libdevice.rint
mdattack Sep 26, 2024
49ef0c2
Merge branch 'main' into develop
bjmsong Oct 9, 2024
File filter

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion bitsandbytes/triton/quantize_columnwise_and_transpose.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -54,7 +54,7 @@ def _quantize_columnwise_and_transpose(
x = tl.load(x_ptr + offsets, mask=p2_arange_mask)
abs_x = tl.abs(x)
max_val = tl.max(tl.where(p2_arange_mask, abs_x, 0), axis=0)
output = tl.libdevice.llrint(127.0 * (x / max_val))
output = tl.extra.cuda.libdevice.rint(127.0 * (x / max_val))

new_start = pid * M
new_offsets = new_start + p2_arange
Expand Down
4 changes: 2 additions & 2 deletions bitsandbytes/triton/quantize_global.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -35,7 +35,7 @@ def _quantize_global(
mask = offsets < n_elements
x = tl.load(x_ptr + offsets, mask=mask)
absmax_inv = tl.load(absmax_inv_ptr)
output = tl.libdevice.llrint(127.0 * (x * absmax_inv))
output = tl.extra.cuda.libdevice.rint(127.0 * (x * absmax_inv))
tl.store(output_ptr + offsets, output, mask=mask)

def quantize_global(x: torch.Tensor):
Expand Down Expand Up @@ -95,7 +95,7 @@ def _quantize_global_transpose(
B = B + (rm[:, None] * stride_bm + rn[None, :] * stride_bn)
mask = (rm < M)[:, None] & (rn < N)[None, :]

output = tl.libdevice.llrint(127.0 * (a * absmax_inv))
output = tl.extra.cuda.libdevice.rint(127.0 * (a * absmax_inv))

tl.store(B, output, mask=mask)

Expand Down
2 changes: 1 addition & 1 deletion bitsandbytes/triton/quantize_rowwise.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -50,7 +50,7 @@ def _quantize_rowwise(

abs_x = tl.abs(x)
max_val = tl.max(tl.where(row_mask, abs_x, 0), axis=0)
output = tl.libdevice.llrint(127.0 * (x / max_val))
output = tl.extra.cuda.libdevice.rint(127.0 * (x / max_val))
tl.store(output_ptr + offsets, output, mask=row_mask)
tl.store(output_maxs + pid, max_val)

Expand Down