Skip to content

Modified block and warp sizes for improved performance on XPU for both layernnorm and rmsnorm #661

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 5 commits into
base: main
Choose a base branch
from
Open

Modified block and warp sizes for improved performance on XPU for both layernnorm and rmsnorm #661

Show file tree
Hide file tree
Changes from 3 commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
e8f2c79
rms_norm.py
Tarakarevu1 Apr 10, 2025
333e5c5
layer_norm.py
Tarakarevu1 Apr 10, 2025
1af9b0e
layer_norm.py
Tarakarevu1 Apr 10, 2025
adad6cd
Merge branch 'main' into Tarakarevu1-patch-3
lancerts May 18, 2025
7bdb1ab
Merge branch 'main' into Tarakarevu1-patch-3
lancerts May 19, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
22 changes: 16 additions & 6 deletions src/liger_kernel/ops/layer_norm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -197,11 +197,21 @@ def layer_norm_backward(dY, X, W, B, Mean, RSTD):
_DW = torch.empty((sm_count, n_cols), dtype=W.dtype, device=W.device)
_DB = torch.empty((sm_count, n_cols), dtype=W.dtype, device=W.device)

BLOCK_SIZE, num_warps = calculate_settings(n_cols)
if n_cols > BLOCK_SIZE:
raise RuntimeError(
f"Feature dimension {n_cols} exceeds maximum supported size of {BLOCK_SIZE}. Consider using a smaller feature dimension."
)
if X.device.type == "xpu": # XPU-specific optimization
BLOCK_SIZE = torch.xpu.get_device_properties(X.device).max_work_group_size
Copy link
Collaborator

@shivam15s shivam15s Apr 11, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

afaik we want the block size here to be triton.next_power_of_2(n). torch.xpu.get_device_properties(X.device).max_work_group_size is supposedly giving the max possible block size on the xpu device. Better way to handle this is to change the calculate_settings function directly where you set the MAX_FUSED_SIZE according to the device.

Copy link
Contributor Author

@Tarakarevu1 Tarakarevu1 Apr 14, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@shivam15s These changes are specific for the layernorm and rmsnorm.That's why i made the changes locally and this will impact other kernels.

Copy link
Collaborator

@lancerts lancerts May 18, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@Tarakarevu1 how much is the perf improvment of layernorm and rmsnorm after this change?

else:
BLOCK_SIZE, num_warps = calculate_settings(n_cols)

if X.device.type == "xpu": # XPU-specific optimization
Copy link
Contributor

@abhilash1910 abhilash1910 Jul 31, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@Tarakarevu1 I think this is not an optimization but rather a guardrail to avoid spilling beyond 64 KB ?
LGTM on the code changes although the runtime behaviour error can be tweaked to show more xpu usage ?

if n_cols > 65536:
raise RuntimeError(
f"Feature dimension {n_cols} exceeds maximum supported size of {BLOCK_SIZE}. Consider using a smaller feature dimension."
)
else:
if n_cols > BLOCK_SIZE:
raise RuntimeError(
f"Feature dimension {n_cols} exceeds maximum supported size of {BLOCK_SIZE}. Consider using a smaller feature dimension."
)

rows_per_program = math.ceil(n_rows / sm_count)
grid = (sm_count,)
Expand All @@ -218,7 +228,7 @@ def layer_norm_backward(dY, X, W, B, Mean, RSTD):
# XPU-specific optimization
kernel_args = {}
if X.device.type == "xpu":
kernel_args.update({"grf_mode": "large", "num_warps": 32, "num_stages": 4})
kernel_args.update({"grf_mode": "large", "num_warps": 4, "num_stages": 4})

_layer_norm_backward_kernel[grid](
X,
Expand Down
16 changes: 13 additions & 3 deletions src/liger_kernel/ops/rms_norm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -212,7 +212,11 @@ def rms_norm_forward(X, W, eps, offset, casting_mode):
dim = shape[-1]
X = X.view(-1, dim)
n_rows, n_cols = X.shape
BLOCK_SIZE, num_warps = calculate_settings(n_cols)
if X.device.type == "xpu": # XPU-specific optimization
BLOCK_SIZE = torch.xpu.get_device_properties(X.device).max_work_group_size
num_warps = 4
else:
BLOCK_SIZE, num_warps = calculate_settings(n_cols)

Y = torch.empty((n_rows, n_cols), dtype=X.dtype, device=X.device)
# RSTD is to cache rstd for each row
Expand Down Expand Up @@ -262,8 +266,14 @@ def rms_norm_backward(dY, X, W, RSTD, offset, casting_mode, BLOCK_SIZE, num_warp
# fp32 for numerical stability especially.
_dW = torch.empty((sm_count, n_cols), dtype=torch.float32, device=W.device)

if n_cols > BLOCK_SIZE:
raise RuntimeError("This layer norm doesn't support feature dim >= 64KB.")
if X.device.type == "xpu": # XPU-specific optimization
if n_cols > 65536:
raise RuntimeError(
"This layer norm doesn't support feature dim >= 64KB."
) # TODO RuntimeError might need little more investigation in the future
else:
if n_cols > BLOCK_SIZE:
raise RuntimeError("This layer norm doesn't support feature dim >= 64KB.")
rows_per_program = math.ceil(n_rows / sm_count)
grid = (sm_count,)

Expand Down
Loading