Skip to content

fix(kernels): rename kernel_add in fused_encoder_bwd #87

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

Merged
dndungu merged 1 commit into from
Apr 10, 2026
Merged

fix(kernels): rename kernel_add in fused_encoder_bwd #87

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: 5 additions & 5 deletions internal/cuda/kernels/fused_encoder_bwd.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -9,7 +9,7 @@
* kernel_gelu_bwd GELU derivative * upstream gradient
* kernel_softmax_bwd Softmax backward (Jacobian-vector product)
* kernel_bias_grad_reduce Sum rows to compute bias gradients
* kernel_add_elementwise Element-wise addition for residual gradients
* kernel_enc_bwd_add_elementwise Element-wise addition for residual gradients
* kernel_matmul_grad_accum Accumulate weight gradient: dW += A^T @ B
*
* cuBLAS calls (~14 total per layer):
Expand Down Expand Up @@ -313,7 +313,7 @@ __global__ void kernel_bias_grad_reduce(
/* out[i] = a[i] + b[i] */
/* ------------------------------------------------------------------ */

__global__ void kernel_add(
__global__ void kernel_enc_bwd_add(
const float* __restrict__ a,
const float* __restrict__ b,
float* __restrict__ out,
Expand All @@ -330,7 +330,7 @@ __global__ void kernel_add(
/* out[i] = a[i] + b[i] + c[i] */
/* ------------------------------------------------------------------ */

__global__ void kernel_add3(
__global__ void kernel_enc_bwd_add3(
const float* __restrict__ a,
const float* __restrict__ b,
const float* __restrict__ c,
Expand Down Expand Up @@ -558,7 +558,7 @@ cudaError_t fused_encoder_bwd_f32(
dXRes1, dg_norm2W, dg_norm2B, dModel);

/* Add residual skip: dXRes1 += dOutput */
kernel_add<<<elemGridTrDm, block256, 0, stream>>>(
kernel_enc_bwd_add<<<elemGridTrDm, block256, 0, stream>>>(
dXRes1, dOutput, dXRes1, trDm);

/* ------------------------------------------------------------ */
Expand Down Expand Up @@ -722,7 +722,7 @@ cudaError_t fused_encoder_bwd_f32(
temp, dg_norm1W, dg_norm1B, dModel);

/* dInput = temp + dXRes1 */
kernel_add<<<elemGridTrDm, block256, 0, stream>>>(
kernel_enc_bwd_add<<<elemGridTrDm, block256, 0, stream>>>(
temp, dXRes1, dInput, trDm);

return cudaGetLastError();
Expand Down
Loading