LLM inference in C/C++
C++ ggml
ggml-zdnn: impl buffer reset, fix memory leaks (#29637)
Aaron Teoeae11d2217fe
diff --git a/ggml/src/ggml-zdnn/ggml-zdnn.cpp b/ggml/src/ggml-zdnn/ggml-zdnn.cppindex 5ae83bbd13..46b37c05ee 100644--- a/ggml/src/ggml-zdnn/ggml-zdnn.cpp+++ b/ggml/src/ggml-zdnn/ggml-zdnn.cpp@@ -211,6 +211,10 @@ static void ggml_backend_zdnn_buffer_free_buffer(ggml_backend_buffer_t buffer) {if (buf->ztensor.buffer_size > 0) ZDNN_CHECK(zdnn_free_ztensor_buffer(&buf->ztensor));}+ if (ctx->owned) {+ ggml_aligned_free(ctx->all_data, ctx->all_size);+ }+delete ctx;}@@ -311,6 +315,28 @@ static void ggml_backend_zdnn_buffer_clear(ggml_backend_buffer_t buffer, uint8_tmemset(ctx->all_data, value, ctx->all_size);}+static void ggml_backend_zdnn_buffer_reset(ggml_backend_buffer_t buffer) {+ ggml_backend_zdnn_buffer_context * ctx = (ggml_backend_zdnn_buffer_context *)buffer->context;++ for (const auto & buf_ptr : ctx->buffers) {+ ggml_backend_zdnn_buffer * buf = buf_ptr.get();++ if (buf->extra != nullptr) {+ free(buf->extra->data);+ }++ if (buf->ztensor.buffer_size > 0) {+ ZDNN_CHECK(zdnn_free_ztensor_buffer(&buf->ztensor));+ }+ }++ if (ctx->buffers.size() > 1) {+ ctx->buffers.resize(1);+ }++ ctx->n_buffers = 1;+}+static ggml_backend_buffer_i ggml_backend_zdnn_buffer_i = {/* .free_buffer = */ ggml_backend_zdnn_buffer_free_buffer,/* .get_base = */ ggml_backend_zdnn_buffer_get_base,@@ -322,7 +348,7 @@ static ggml_backend_buffer_i ggml_backend_zdnn_buffer_i = {/* .get_tensor_2d = */ NULL,/* .cpy_tensor = */ NULL,/* .clear = */ ggml_backend_zdnn_buffer_clear,- /* .reset = */ NULL,+ /* .reset = */ ggml_backend_zdnn_buffer_reset,};//