Skip to content

Fix nan issue for certain shape #10

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 1 commit into
base: main
Choose a base branch
from
Open

Fix nan issue for certain shape #10

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
10 changes: 7 additions & 3 deletions batch_invariant_ops/batch_invariant_ops.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -70,7 +70,7 @@ def matmul_kernel_persistent(
k_tiles = tl.cdiv(K, BLOCK_SIZE_K)
num_tiles = num_pid_m * num_pid_n

tile_id_c = start_pid - NUM_SMS
# tile_id_c = start_pid - NUM_SMS

offs_k_for_mask = tl.arange(0, BLOCK_SIZE_K)
num_pid_in_group = GROUP_SIZE_M * num_pid_n
Expand Down Expand Up @@ -103,8 +103,8 @@ def matmul_kernel_persistent(
b = tl.load(b_ptrs, mask=offs_k_for_mask[:, None] < K - ki * BLOCK_SIZE_K, other=0.0)
accumulator = tl.dot(a, b, accumulator)

tile_id_c += NUM_SMS
pid_m, pid_n = _compute_pid(tile_id_c, num_pid_in_group, num_pid_m, GROUP_SIZE_M, NUM_SMS)
# tile_id_c += NUM_SMS
# pid_m, pid_n = _compute_pid(tile_id_c, num_pid_in_group, num_pid_m, GROUP_SIZE_M, NUM_SMS)
offs_cm = pid_m * BLOCK_SIZE_M + tl.arange(0, BLOCK_SIZE_M)
offs_cn = pid_n * BLOCK_SIZE_N + tl.arange(0, BLOCK_SIZE_N)
if C_LARGE:
Expand All @@ -118,6 +118,10 @@ def matmul_kernel_persistent(
accumulator += bias
if c_ptr.dtype.element_ty == tl.float8e4nv:
c = accumulator.to(tl.float8e4nv)
elif c_ptr.dtype.element_ty == tl.bfloat16:
c = accumulator.to(tl.bfloat16)
elif c_ptr.dtype.element_ty == tl.float32:
c = accumulator.to(tl.float32)
else:
c = accumulator.to(tl.float16)
tl.store(c_ptrs, c, mask=c_mask)
Expand Down
15 changes: 12 additions & 3 deletions test_batch_invariance.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,9 +8,18 @@
pass

def test_batch_invariance(dtype=torch.float32):
B, D = 2048, 4096
a = torch.linspace(-100, 100, B*D, dtype=dtype).reshape(B, D)
b = torch.linspace(-100, 100, D*D, dtype=dtype).reshape(D, D)
M = 32
K = 128
N = 1024
a = torch.linspace(-100, 100, M*K, dtype=dtype).reshape(M, K)

# Create non-contiguous tensor to mimic the nn.Linear case while weight is always transposed
# See ref: https://github.com/pytorch/pytorch/blob/v2.8.0/torch/nn/modules/linear.py#L50
b = torch.linspace(-100, 100, K*N, dtype=dtype).reshape(N, K)
b = b.transpose(0, 1)

print(f"a is contiguous: {a.is_contiguous()}")
print(f"b is contiguous: {b.is_contiguous()}")

# Method 1: Matrix-vector multiplication (batch size 1)
out1 = torch.mm(a[:1], b)
Expand Down