Skip to content

Commit 4ce13c8

Browse files
committed
Revert "Revert ggml : faster ssm scan (ggml-org#10558)"
This reverts commit 2d449c3.
1 parent 34906c2 commit 4ce13c8

File tree

1 file changed

+10
-0
lines changed

1 file changed

+10
-0
lines changed

ggml/src/ggml-cuda/ggml-cuda.cu

Lines changed: 10 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -34,6 +34,8 @@ bool g_mul_mat_q = true;
3434
#include "ggml-cuda/rope.cuh"
3535
#include "ggml-cuda/scale.cuh"
3636
#include "ggml-cuda/softmax.cuh"
37+
#include "ggml-cuda/ssm-conv.cuh"
38+
#include "ggml-cuda/ssm-scan.cuh"
3739
#include "ggml-cuda/sum.cuh"
3840
#include "ggml-cuda/sumrows.cuh"
3941
#include "ggml-cuda/tsembd.cuh"
@@ -2528,6 +2530,12 @@ static bool ggml_cuda_compute_forward(ggml_backend_cuda_context & ctx, struct gg
25282530
case GGML_OP_SUM_ROWS:
25292531
ggml_cuda_op_sum_rows(ctx, dst);
25302532
break;
2533+
case GGML_OP_SSM_CONV:
2534+
ggml_cuda_op_ssm_conv(ctx, dst);
2535+
break;
2536+
case GGML_OP_SSM_SCAN:
2537+
ggml_cuda_op_ssm_scan(ctx, dst);
2538+
break;
25312539
case GGML_OP_ARGSORT:
25322540
ggml_cuda_op_argsort(ctx, dst);
25332541
break;
@@ -3400,6 +3408,8 @@ static bool ggml_backend_cuda_device_supports_op(ggml_backend_dev_t dev, const g
34003408
case GGML_OP_SIN:
34013409
case GGML_OP_COS:
34023410
case GGML_OP_CLAMP:
3411+
case GGML_OP_SSM_SCAN:
3412+
case GGML_OP_SSM_CONV:
34033413
return true;
34043414
case GGML_OP_CONT:
34053415
return op->src[0]->type != GGML_TYPE_BF16;

0 commit comments

Comments
 (0)