sycl : fix for untransposed GDA recurrent state (#20583)
This commit is contained in:
@@ -55,7 +55,7 @@ void gated_delta_net_sycl(const float * q,
|
|||||||
#pragma unroll
|
#pragma unroll
|
||||||
for (int r = 0; r < rows_per_lane; r++) {
|
for (int r = 0; r < rows_per_lane; r++) {
|
||||||
const int i = r * warp_size + lane;
|
const int i = r * warp_size + lane;
|
||||||
s_shard[r] = curr_state[i * S_v + col];
|
s_shard[r] = curr_state[col * S_v + i];
|
||||||
}
|
}
|
||||||
|
|
||||||
for (int t = 0; t < n_tokens; t++) {
|
for (int t = 0; t < n_tokens; t++) {
|
||||||
@@ -137,7 +137,7 @@ void gated_delta_net_sycl(const float * q,
|
|||||||
#pragma unroll
|
#pragma unroll
|
||||||
for (int r = 0; r < rows_per_lane; r++) {
|
for (int r = 0; r < rows_per_lane; r++) {
|
||||||
const int i = r * warp_size + lane;
|
const int i = r * warp_size + lane;
|
||||||
state[i * S_v + col] = s_shard[r];
|
state[col * S_v + i] = s_shard[r];
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|||||||
Reference in New Issue
Block a user