LLM inference in C/C++
C++ ggml
hexagon: add missing FARF logs for cpy/get_rows/set_rows/gdn ops (#28217)
Trivikram Reddy69320fef12d3
diff --git a/ggml/src/ggml-hexagon/htp/cpy-ops.c b/ggml/src/ggml-hexagon/htp/cpy-ops.cindex c945425dab..b151b757f4 100644--- a/ggml/src/ggml-hexagon/htp/cpy-ops.c+++ b/ggml/src/ggml-hexagon/htp/cpy-ops.c@@ -323,6 +323,10 @@ int op_cpy(struct htp_ops_context * octx) {return HTP_STATUS_NO_SUPPORT;}+ FARF(HIGH, "cpy-%s-%s: (%ux%ux%ux%u) -> (%ux%ux%ux%u) : use_dma=%d n_threads %u\n",+ src0->type == HTP_TYPE_F32 ? "f32" : "f16", dst->type == HTP_TYPE_F32 ? "f32" : "f16",+ ne00, ne01, ne02, ne03, ne0, ne1, ne2, ne3, use_dma, n_threads);+if (use_dma) {cpy_dma_sametype_sameshape(octx, dst, src0, ct.src0_type_size, ne00, ne01, ne02, ne03, nb01, nb02, nb03, nb1, nb2, nb3);} else {diff --git a/ggml/src/ggml-hexagon/htp/gated-delta-net-ops.c b/ggml/src/ggml-hexagon/htp/gated-delta-net-ops.cindex 35518e6111..9665521529 100644--- a/ggml/src/ggml-hexagon/htp/gated-delta-net-ops.c+++ b/ggml/src/ggml-hexagon/htp/gated-delta-net-ops.c@@ -1138,6 +1138,15 @@ int op_gated_delta_net(struct htp_ops_context * octx) {gctx.vtcm_base = octx->ctx->vtcm_base;gctx.vtcm_per_thread = 2 * state_aligned;+ FARF(HIGH, "gated-delta-net-f32: q(%ux%ux%ux%u) k(%ux%ux%ux%u) v(%ux%ux%ux%u) state(%ux%ux%ux%u) -> (%ux%ux%ux%u) : "+ "vtcm-size %zu n_threads %u\n",+ q->ne[0], q->ne[1], q->ne[2], q->ne[3],+ k->ne[0], k->ne[1], k->ne[2], k->ne[3],+ v->ne[0], v->ne[1], v->ne[2], v->ne[3],+ state->ne[0], state->ne[1], state->ne[2], state->ne[3],+ dst->ne[0], dst->ne[1], dst->ne[2], dst->ne[3],+ gctx.vtcm_per_thread * octx->n_threads, octx->n_threads);+if (n_tokens == 1) {worker_pool_run_func(octx->ctx->worker_pool, gated_delta_net_f32_tg_thread, &gctx, octx->n_threads);} else {diff --git a/ggml/src/ggml-hexagon/htp/get-rows-ops.c b/ggml/src/ggml-hexagon/htp/get-rows-ops.cindex 05769d17f7..a87962d229 100644--- a/ggml/src/ggml-hexagon/htp/get-rows-ops.c+++ b/ggml/src/ggml-hexagon/htp/get-rows-ops.c@@ -247,6 +247,14 @@ int op_get_rows(struct htp_ops_context * octx) {}}+ FARF(HIGH, "get-rows: (%ux%ux%ux%u) x (%ux%ux%ux%u) -> (%ux%ux%ux%u) : src0-vtcm-size %zu dst-vtcm-size %zu use_dma=%d n_threads %d\n",+ octx->src[0]->ne[0], octx->src[0]->ne[1], octx->src[0]->ne[2], octx->src[0]->ne[3],+ octx->src[1]->ne[0], octx->src[1]->ne[1], octx->src[1]->ne[2], octx->src[1]->ne[3],+ octx->dst->ne[0], octx->dst->ne[1], octx->dst->ne[2], octx->dst->ne[3],+ grctx.vtcm_layout.src0_bytes_per_thread * kparams->n_threads,+ grctx.vtcm_layout.dst_bytes_per_thread * kparams->n_threads,+ kparams->use_dma, kparams->n_threads);+work_queue_run(octx->ctx->work_queue, q_func, &grctx, kparams->n_threads);return HTP_STATUS_OK;}diff --git a/ggml/src/ggml-hexagon/htp/main.c b/ggml/src/ggml-hexagon/htp/main.cindex 27d1dedcdf..72cf02a326 100644--- a/ggml/src/ggml-hexagon/htp/main.c+++ b/ggml/src/ggml-hexagon/htp/main.c@@ -981,7 +981,7 @@ static int proc_op_req(struct htp_ops_context * octx, struct htp_tensor *tens, uoctx->src_dma[i] = octx->ctx->dma; // FIXME: ? octx->ctx->dma_cached : octx->ctx->dma;FARF(HIGH, "prep-src #%u: data %p size %u : %u:%u:%u:%u", op->src[i], (void*) src->data, src->size,- src->ne[0], src->ne[1], src->ne[3], src->ne[3]);+ src->ne[0], src->ne[1], src->ne[2], src->ne[3]);}htp_tensor_flush_all(octx->ctx, octx->src, HTP_OP_MAX_INPUTS);diff --git a/ggml/src/ggml-hexagon/htp/set-rows-ops.c b/ggml/src/ggml-hexagon/htp/set-rows-ops.cindex fa14bf0ef6..340a497f7a 100644--- a/ggml/src/ggml-hexagon/htp/set-rows-ops.c+++ b/ggml/src/ggml-hexagon/htp/set-rows-ops.c@@ -216,6 +216,14 @@ int op_set_rows(struct htp_ops_context * octx) {default: return HTP_STATUS_NO_SUPPORT;}+ FARF(HIGH, "set-rows: (%ux%ux%ux%u) x (%ux%ux%ux%u) -> (%ux%ux%ux%u) : src0-vtcm-size %zu dst-vtcm-size %zu n_threads %d\n",+ octx->src[0]->ne[0], octx->src[0]->ne[1], octx->src[0]->ne[2], octx->src[0]->ne[3],+ octx->src[1]->ne[0], octx->src[1]->ne[1], octx->src[1]->ne[2], octx->src[1]->ne[3],+ octx->dst->ne[0], octx->dst->ne[1], octx->dst->ne[2], octx->dst->ne[3],+ srctx.vtcm_layout.src0_bytes_per_thread * kparams->n_threads,+ srctx.vtcm_layout.dst_bytes_per_thread * kparams->n_threads,+ kparams->n_threads);+work_queue_run(octx->ctx->work_queue, q_func, &srctx, kparams->n_threads);return HTP_STATUS_OK;