diff --git a/ggml.c b/ggml.c index d2f5e7275..33a6ffdc6 100644 --- a/ggml.c +++ b/ggml.c @@ -4229,6 +4229,15 @@ bool ggml_is_contiguous(const struct ggml_tensor * tensor) { tensor->nb[3] == tensor->nb[2]*tensor->ne[2]; } +static inline bool ggml_is_contiguous_except_dim_1(const struct ggml_tensor * tensor) { + static_assert(GGML_MAX_DIMS == 4, "GGML_MAX_DIMS is not 4 - update this function"); + + return + tensor->nb[0] == GGML_TYPE_SIZE[tensor->type] && + tensor->nb[2] == tensor->nb[1]*tensor->ne[1] && + tensor->nb[3] == tensor->nb[2]*tensor->ne[2]; +} + bool ggml_is_permuted(const struct ggml_tensor * tensor) { static_assert(GGML_MAX_DIMS == 4, "GGML_MAX_DIMS is not 4 - update this function"); @@ -9814,8 +9823,8 @@ static void ggml_compute_forward_gelu_f32( const struct ggml_compute_params * params, const struct ggml_tensor * src0, struct ggml_tensor * dst) { - GGML_ASSERT(ggml_is_contiguous(src0)); - GGML_ASSERT(ggml_is_contiguous(dst)); + GGML_ASSERT(ggml_is_contiguous_except_dim_1(src0)); + GGML_ASSERT(ggml_is_contiguous_except_dim_1(dst)); GGML_ASSERT(ggml_are_same_shape(src0, dst)); if (params->type == GGML_TASK_INIT || params->type == GGML_TASK_FINALIZE) { @@ -9873,8 +9882,8 @@ static void ggml_compute_forward_gelu_quick_f32( const struct ggml_compute_params * params, const struct ggml_tensor * src0, struct ggml_tensor * dst) { - GGML_ASSERT(ggml_is_contiguous(src0)); - GGML_ASSERT(ggml_is_contiguous(dst)); + GGML_ASSERT(ggml_is_contiguous_except_dim_1(src0)); + GGML_ASSERT(ggml_is_contiguous_except_dim_1(dst)); GGML_ASSERT(ggml_are_same_shape(src0, dst)); if (params->type == GGML_TASK_INIT || params->type == GGML_TASK_FINALIZE) { @@ -9932,8 +9941,8 @@ static void ggml_compute_forward_silu_f32( const struct ggml_compute_params * params, const struct ggml_tensor * src0, struct ggml_tensor * dst) { - GGML_ASSERT(ggml_is_contiguous(src0)); - GGML_ASSERT(ggml_is_contiguous(dst)); + GGML_ASSERT(ggml_is_contiguous_except_dim_1(src0)); + GGML_ASSERT(ggml_is_contiguous_except_dim_1(dst)); GGML_ASSERT(ggml_are_same_shape(src0, dst)); if (params->type == GGML_TASK_INIT || params->type == GGML_TASK_FINALIZE) { @@ -9992,9 +10001,9 @@ static void ggml_compute_forward_silu_back_f32( const struct ggml_tensor * src0, const struct ggml_tensor * grad, struct ggml_tensor * dst) { - GGML_ASSERT(ggml_is_contiguous(grad)); - GGML_ASSERT(ggml_is_contiguous(src0)); - GGML_ASSERT(ggml_is_contiguous(dst)); + GGML_ASSERT(ggml_is_contiguous_except_dim_1(grad)); + GGML_ASSERT(ggml_is_contiguous_except_dim_1(src0)); + GGML_ASSERT(ggml_is_contiguous_except_dim_1(dst)); GGML_ASSERT(ggml_are_same_shape(src0, dst)); GGML_ASSERT(ggml_are_same_shape(src0, grad));