whisper.rn 0.4.0-rc.8 → 0.4.0

package/cpp/ggml-backend.cpp

@@ -0,0 +1,2016 @@
+// Note: porting this file to C++ is a work in progress
+
+#ifdef _WIN32
+#define WIN32_LEAN_AND_MEAN
+#ifndef NOMINMAX
+#   define NOMINMAX
+#endif
+#include <windows.h>
+#endif
+
+#include "ggml-backend.h"
+#include "ggml-backend-impl.h"
+#include "ggml-alloc.h"
+#include "ggml-impl.h"
+
+#include <assert.h>
+#include <limits.h>
+#include <stdarg.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <string>
+#include <vector>
+#include <algorithm>
+
+#ifdef __APPLE__
+#include <sys/types.h>
+#include <sys/sysctl.h>
+#endif
+
+
+// backend buffer type
+
+const char * wsp_ggml_backend_buft_name(wsp_ggml_backend_buffer_type_t buft) {
+    return buft->iface.get_name(buft);
+}
+
+wsp_ggml_backend_buffer_t wsp_ggml_backend_buft_alloc_buffer(wsp_ggml_backend_buffer_type_t buft, size_t size) {
+    if (size == 0) {
+        // return a dummy buffer for zero-sized allocations
+        return wsp_ggml_backend_buffer_init(buft, {}, NULL, 0);
+    }
+
+    return buft->iface.alloc_buffer(buft, size);
+}
+
+size_t wsp_ggml_backend_buft_get_alignment(wsp_ggml_backend_buffer_type_t buft) {
+    return buft->iface.get_alignment(buft);
+}
+
+size_t wsp_ggml_backend_buft_get_max_size(wsp_ggml_backend_buffer_type_t buft) {
+    // get_max_size is optional, defaults to SIZE_MAX
+    if (buft->iface.get_max_size) {
+        return buft->iface.get_max_size(buft);
+    }
+    return SIZE_MAX;
+}
+
+size_t wsp_ggml_backend_buft_get_alloc_size(wsp_ggml_backend_buffer_type_t buft, const struct wsp_ggml_tensor * tensor) {
+    // get_alloc_size is optional, defaults to wsp_ggml_nbytes
+    if (buft->iface.get_alloc_size) {
+        size_t size = buft->iface.get_alloc_size(buft, tensor);
+        assert(size >= wsp_ggml_nbytes(tensor));
+        return size;
+    }
+    return wsp_ggml_nbytes(tensor);
+}
+
+bool wsp_ggml_backend_buft_is_host(wsp_ggml_backend_buffer_type_t buft) {
+    if (buft->iface.is_host) {
+        return buft->iface.is_host(buft);
+    }
+    return false;
+}
+
+wsp_ggml_backend_dev_t wsp_ggml_backend_buft_get_device(wsp_ggml_backend_buffer_type_t buft) {
+    return buft->device;
+}
+
+// backend buffer
+
+wsp_ggml_backend_buffer_t wsp_ggml_backend_buffer_init(
+               wsp_ggml_backend_buffer_type_t buft,
+        struct wsp_ggml_backend_buffer_i      iface,
+               void *                     context,
+               size_t                     size) {
+    wsp_ggml_backend_buffer_t buffer = new wsp_ggml_backend_buffer {
+        /* .interface = */ iface,
+        /* .buft      = */ buft,
+        /* .context   = */ context,
+        /* .size      = */ size,
+        /* .usage     = */ WSP_GGML_BACKEND_BUFFER_USAGE_ANY
+    };
+
+    return buffer;
+}
+
+const char * wsp_ggml_backend_buffer_name(wsp_ggml_backend_buffer_t buffer) {
+    return wsp_ggml_backend_buft_name(wsp_ggml_backend_buffer_get_type(buffer));
+}
+
+void wsp_ggml_backend_buffer_free(wsp_ggml_backend_buffer_t buffer) {
+    if (buffer == NULL) {
+        return;
+    }
+
+    if (buffer->iface.free_buffer != NULL) {
+        buffer->iface.free_buffer(buffer);
+    }
+    delete buffer;
+}
+
+size_t wsp_ggml_backend_buffer_get_size(wsp_ggml_backend_buffer_t buffer) {
+    return buffer->size;
+}
+
+void * wsp_ggml_backend_buffer_get_base(wsp_ggml_backend_buffer_t buffer) {
+    // get_base is optional if the buffer is zero-sized
+    if (buffer->size == 0) {
+        return NULL;
+    }
+
+    void * base = buffer->iface.get_base(buffer);
+
+    WSP_GGML_ASSERT(base != NULL && "backend buffer base cannot be NULL");
+
+    return base;
+}
+
+enum wsp_ggml_status wsp_ggml_backend_buffer_init_tensor(wsp_ggml_backend_buffer_t buffer, struct wsp_ggml_tensor * tensor) {
+    // init_tensor is optional
+    if (buffer->iface.init_tensor) {
+        return buffer->iface.init_tensor(buffer, tensor);
+    }
+    return WSP_GGML_STATUS_SUCCESS;
+}
+
+void wsp_ggml_backend_buffer_clear(wsp_ggml_backend_buffer_t buffer, uint8_t value) {
+    // clear is optional if the buffer is zero-sized
+    if (buffer->size == 0) {
+        return;
+    }
+
+    buffer->iface.clear(buffer, value);
+}
+
+size_t wsp_ggml_backend_buffer_get_alignment(wsp_ggml_backend_buffer_t buffer) {
+    return wsp_ggml_backend_buft_get_alignment(wsp_ggml_backend_buffer_get_type(buffer));
+}
+
+size_t wsp_ggml_backend_buffer_get_max_size(wsp_ggml_backend_buffer_t buffer) {
+    return wsp_ggml_backend_buft_get_max_size(wsp_ggml_backend_buffer_get_type(buffer));
+}
+
+size_t wsp_ggml_backend_buffer_get_alloc_size(wsp_ggml_backend_buffer_t buffer, const struct wsp_ggml_tensor * tensor) {
+    return wsp_ggml_backend_buft_get_alloc_size(wsp_ggml_backend_buffer_get_type(buffer), tensor);
+}
+
+bool wsp_ggml_backend_buffer_is_host(wsp_ggml_backend_buffer_t buffer) {
+    return wsp_ggml_backend_buft_is_host(wsp_ggml_backend_buffer_get_type(buffer));
+}
+
+void wsp_ggml_backend_buffer_set_usage(wsp_ggml_backend_buffer_t buffer, enum wsp_ggml_backend_buffer_usage usage) {
+    buffer->usage = usage;
+
+    // FIXME: add a generic callback to the buffer interface
+    if (wsp_ggml_backend_buffer_is_multi_buffer(buffer)) {
+        wsp_ggml_backend_multi_buffer_set_usage(buffer, usage);
+    }
+}
+
+enum wsp_ggml_backend_buffer_usage wsp_ggml_backend_buffer_get_usage(wsp_ggml_backend_buffer_t buffer) {
+    return buffer->usage;
+}
+
+wsp_ggml_backend_buffer_type_t wsp_ggml_backend_buffer_get_type(wsp_ggml_backend_buffer_t buffer) {
+    return buffer->buft;
+}
+
+void wsp_ggml_backend_buffer_reset(wsp_ggml_backend_buffer_t buffer) {
+    if (buffer->iface.reset) {
+        buffer->iface.reset(buffer);
+    }
+}
+
+bool wsp_ggml_backend_buffer_copy_tensor(const struct wsp_ggml_tensor * src, struct wsp_ggml_tensor * dst) {
+    wsp_ggml_backend_buffer_t dst_buf = dst->view_src ? dst->view_src->buffer : dst->buffer;
+    if (dst_buf->iface.cpy_tensor) {
+        return dst_buf->iface.cpy_tensor(dst_buf, src, dst);
+    }
+    return false;
+}
+
+// backend
+
+wsp_ggml_guid_t wsp_ggml_backend_guid(wsp_ggml_backend_t backend) {
+    if (backend == NULL) {
+        return NULL;
+    }
+    return backend->guid;
+}
+
+const char * wsp_ggml_backend_name(wsp_ggml_backend_t backend) {
+    if (backend == NULL) {
+        return "NULL";
+    }
+    return backend->iface.get_name(backend);
+}
+
+void wsp_ggml_backend_free(wsp_ggml_backend_t backend) {
+    if (backend == NULL) {
+        return;
+    }
+
+    backend->iface.free(backend);
+}
+
+wsp_ggml_backend_buffer_type_t wsp_ggml_backend_get_default_buffer_type(wsp_ggml_backend_t backend) {
+    return wsp_ggml_backend_dev_buffer_type(backend->device);
+}
+
+wsp_ggml_backend_buffer_t wsp_ggml_backend_alloc_buffer(wsp_ggml_backend_t backend, size_t size) {
+    return wsp_ggml_backend_buft_alloc_buffer(wsp_ggml_backend_get_default_buffer_type(backend), size);
+}
+
+size_t wsp_ggml_backend_get_alignment(wsp_ggml_backend_t backend) {
+    return wsp_ggml_backend_buft_get_alignment(wsp_ggml_backend_get_default_buffer_type(backend));
+}
+
+size_t wsp_ggml_backend_get_max_size(wsp_ggml_backend_t backend) {
+    return wsp_ggml_backend_buft_get_max_size(wsp_ggml_backend_get_default_buffer_type(backend));
+}
+
+void wsp_ggml_backend_tensor_set_async(wsp_ggml_backend_t backend, struct wsp_ggml_tensor * tensor, const void * data, size_t offset, size_t size) {
+    WSP_GGML_ASSERT(tensor->data != NULL && "tensor not allocated");
+    WSP_GGML_ASSERT(offset + size <= wsp_ggml_nbytes(tensor) && "tensor write out of bounds");
+
+    if (backend->iface.set_tensor_async == NULL) {
+        wsp_ggml_backend_tensor_set(tensor, data, offset, size);
+    } else {
+        backend->iface.set_tensor_async(backend, tensor, data, offset, size);
+    }
+}
+
+void wsp_ggml_backend_tensor_get_async(wsp_ggml_backend_t backend, const struct wsp_ggml_tensor * tensor, void * data, size_t offset, size_t size) {
+    WSP_GGML_ASSERT(tensor->data != NULL && "tensor not allocated");
+    WSP_GGML_ASSERT(offset + size <= wsp_ggml_nbytes(tensor) && "tensor read out of bounds");
+
+    if (backend->iface.get_tensor_async == NULL) {
+        wsp_ggml_backend_tensor_get(tensor, data, offset, size);
+    } else {
+        backend->iface.get_tensor_async(backend, tensor, data, offset, size);
+    }
+}
+
+void wsp_ggml_backend_tensor_set(struct wsp_ggml_tensor * tensor, const void * data, size_t offset, size_t size) {
+    WSP_GGML_ASSERT(tensor);
+    wsp_ggml_backend_buffer_t buf = tensor->view_src ? tensor->view_src->buffer : tensor->buffer;
+
+    if (size == 0) {
+        return;
+    }
+
+    WSP_GGML_ASSERT(buf != NULL && "tensor buffer not set");
+    WSP_GGML_ASSERT(tensor->data != NULL && "tensor not allocated");
+    WSP_GGML_ASSERT(offset + size <= wsp_ggml_nbytes(tensor) && "tensor write out of bounds");
+
+    buf->iface.set_tensor(buf, tensor, data, offset, size);
+}
+
+void wsp_ggml_backend_tensor_get(const struct wsp_ggml_tensor * tensor, void * data, size_t offset, size_t size) {
+    WSP_GGML_ASSERT(tensor);
+    wsp_ggml_backend_buffer_t buf = tensor->view_src ? tensor->view_src->buffer : tensor->buffer;
+
+    if (size == 0) {
+        return;
+    }
+
+    WSP_GGML_ASSERT(buf != NULL && "tensor buffer not set");
+    WSP_GGML_ASSERT(tensor->data != NULL && "tensor not allocated");
+    WSP_GGML_ASSERT(offset + size <= wsp_ggml_nbytes(tensor) && "tensor read out of bounds");
+
+    buf->iface.get_tensor(buf, tensor, data, offset, size);
+}
+
+void wsp_ggml_backend_tensor_memset(struct wsp_ggml_tensor * tensor, uint8_t value, size_t offset, size_t size) {
+    wsp_ggml_backend_buffer_t buf = tensor->view_src ? tensor->view_src->buffer : tensor->buffer;
+
+    if (size == 0) {
+        return;
+    }
+
+    WSP_GGML_ASSERT(buf != NULL && "tensor buffer not set");
+    WSP_GGML_ASSERT(tensor->data != NULL && "tensor not allocated");
+    WSP_GGML_ASSERT(offset + size <= wsp_ggml_nbytes(tensor) && "tensor write out of bounds");
+    WSP_GGML_ASSERT(buf->iface.memset_tensor != NULL && "memset not implemented by backend buffer");
+
+    buf->iface.memset_tensor(buf, tensor, value, offset, size);
+}
+
+void wsp_ggml_backend_synchronize(wsp_ggml_backend_t backend) {
+    if (backend->iface.synchronize == NULL) {
+        return;
+    }
+
+    backend->iface.synchronize(backend);
+}
+
+wsp_ggml_backend_graph_plan_t wsp_ggml_backend_graph_plan_create(wsp_ggml_backend_t backend, struct wsp_ggml_cgraph * cgraph) {
+    WSP_GGML_ASSERT(backend->iface.graph_plan_create != NULL);
+
+    return backend->iface.graph_plan_create(backend, cgraph);
+}
+
+void wsp_ggml_backend_graph_plan_free(wsp_ggml_backend_t backend, wsp_ggml_backend_graph_plan_t plan) {
+    WSP_GGML_ASSERT(backend->iface.graph_plan_free != NULL);
+
+    backend->iface.graph_plan_free(backend, plan);
+}
+
+enum wsp_ggml_status wsp_ggml_backend_graph_plan_compute(wsp_ggml_backend_t backend, wsp_ggml_backend_graph_plan_t plan) {
+    WSP_GGML_ASSERT(backend->iface.graph_plan_compute != NULL);
+
+    return backend->iface.graph_plan_compute(backend, plan);
+}
+
+enum wsp_ggml_status wsp_ggml_backend_graph_compute(wsp_ggml_backend_t backend, struct wsp_ggml_cgraph * cgraph) {
+    enum wsp_ggml_status err = wsp_ggml_backend_graph_compute_async(backend, cgraph);
+    wsp_ggml_backend_synchronize(backend);
+    return err;
+}
+
+enum wsp_ggml_status wsp_ggml_backend_graph_compute_async(wsp_ggml_backend_t backend, struct wsp_ggml_cgraph * cgraph) {
+    return backend->iface.graph_compute(backend, cgraph);
+}
+
+bool wsp_ggml_backend_supports_op(wsp_ggml_backend_t backend, const struct wsp_ggml_tensor * op) {
+    return wsp_ggml_backend_dev_supports_op(backend->device, op);
+}
+
+bool wsp_ggml_backend_supports_buft(wsp_ggml_backend_t backend, wsp_ggml_backend_buffer_type_t buft) {
+    return wsp_ggml_backend_dev_supports_buft(backend->device, buft);
+}
+
+bool wsp_ggml_backend_offload_op(wsp_ggml_backend_t backend, const struct wsp_ggml_tensor * op) {
+    return wsp_ggml_backend_dev_offload_op(backend->device, op);
+}
+
+wsp_ggml_backend_dev_t wsp_ggml_backend_get_device(wsp_ggml_backend_t backend) {
+    return backend->device;
+}
+
+// backend copy
+
+static bool wsp_ggml_are_same_layout(const struct wsp_ggml_tensor * a, const struct wsp_ggml_tensor * b) {
+    if (a->type != b->type) {
+        return false;
+    }
+    for (int i = 0; i < WSP_GGML_MAX_DIMS; i++) {
+        if (a->ne[i] != b->ne[i]) {
+            return false;
+        }
+        if (a->nb[i] != b->nb[i]) {
+            return false;
+        }
+    }
+    return true;
+}
+
+void wsp_ggml_backend_tensor_copy(struct wsp_ggml_tensor * src, struct wsp_ggml_tensor * dst) {
+    WSP_GGML_ASSERT(wsp_ggml_are_same_layout(src, dst) && "cannot copy tensors with different layouts");
+
+    if (src == dst) {
+        return;
+    }
+
+    if (wsp_ggml_backend_buffer_is_host(src->buffer)) {
+        wsp_ggml_backend_tensor_set(dst, src->data, 0, wsp_ggml_nbytes(src));
+    } else if (wsp_ggml_backend_buffer_is_host(dst->buffer)) {
+        wsp_ggml_backend_tensor_get(src, dst->data, 0, wsp_ggml_nbytes(src));
+    } else if (!wsp_ggml_backend_buffer_copy_tensor(src, dst)) {
+#ifndef NDEBUG
+        WSP_GGML_LOG_DEBUG("%s: warning: slow copy from %s to %s\n", __func__, wsp_ggml_backend_buffer_name(src->buffer), wsp_ggml_backend_buffer_name(dst->buffer));
+#endif
+        size_t nbytes = wsp_ggml_nbytes(src);
+        void * data = malloc(nbytes);
+        wsp_ggml_backend_tensor_get(src, data, 0, nbytes);
+        wsp_ggml_backend_tensor_set(dst, data, 0, nbytes);
+        free(data);
+    }
+}
+
+void wsp_ggml_backend_tensor_copy_async(wsp_ggml_backend_t backend_src, wsp_ggml_backend_t backend_dst, struct wsp_ggml_tensor * src, struct wsp_ggml_tensor * dst) {
+    WSP_GGML_ASSERT(wsp_ggml_are_same_layout(src, dst) && "cannot copy tensors with different layouts");
+
+    if (src == dst) {
+        return;
+    }
+
+    if (backend_dst->iface.cpy_tensor_async != NULL) {
+        if (backend_dst->iface.cpy_tensor_async(backend_src, backend_dst, src, dst)) {
+            return;
+        }
+    }
+
+    // an async copy would normally happen after all the queued operations on both backends are completed
+    // to simulate the same behavior, we need to synchronize both backends first, and do a blocking copy
+    wsp_ggml_backend_synchronize(backend_src);
+    wsp_ggml_backend_synchronize(backend_dst);
+    wsp_ggml_backend_tensor_copy(src, dst);
+}
+
+// events
+
+wsp_ggml_backend_event_t wsp_ggml_backend_event_new(wsp_ggml_backend_dev_t device) {
+    // null device is allowed for the transition period to the device interface
+    if (device == NULL || device->iface.event_new == NULL) {
+        return NULL;
+    }
+    return device->iface.event_new(device);
+}
+
+void wsp_ggml_backend_event_free(wsp_ggml_backend_event_t event) {
+    if (event == NULL) {
+        return;
+    }
+    event->device->iface.event_free(event->device, event);
+}
+
+void wsp_ggml_backend_event_record(wsp_ggml_backend_event_t event, wsp_ggml_backend_t backend) {
+    WSP_GGML_ASSERT(backend->iface.event_record != NULL);
+
+    backend->iface.event_record(backend, event);
+}
+
+void wsp_ggml_backend_event_synchronize(wsp_ggml_backend_event_t event) {
+    WSP_GGML_ASSERT(event->device->iface.event_synchronize);
+
+    event->device->iface.event_synchronize(event->device, event);
+}
+
+void wsp_ggml_backend_event_wait(wsp_ggml_backend_t backend, wsp_ggml_backend_event_t event) {
+    WSP_GGML_ASSERT(backend->iface.event_wait != NULL);
+
+    backend->iface.event_wait(backend, event);
+}
+
+// Backend device
+
+const char * wsp_ggml_backend_dev_name(wsp_ggml_backend_dev_t device) {
+    return device->iface.get_name(device);
+}
+
+const char * wsp_ggml_backend_dev_description(wsp_ggml_backend_dev_t device) {
+    return device->iface.get_description(device);
+}
+
+void wsp_ggml_backend_dev_memory(wsp_ggml_backend_dev_t device, size_t * free, size_t * total) {
+    device->iface.get_memory(device, free, total);
+}
+
+enum wsp_ggml_backend_dev_type wsp_ggml_backend_dev_type(wsp_ggml_backend_dev_t device) {
+    return device->iface.get_type(device);
+}
+
+void wsp_ggml_backend_dev_get_props(wsp_ggml_backend_dev_t device, struct wsp_ggml_backend_dev_props * props) {
+    memset(props, 0, sizeof(*props));
+    device->iface.get_props(device, props);
+}
+
+wsp_ggml_backend_reg_t wsp_ggml_backend_dev_backend_reg(wsp_ggml_backend_dev_t device) {
+    return device->reg;
+}
+
+wsp_ggml_backend_t wsp_ggml_backend_dev_init(wsp_ggml_backend_dev_t device, const char * params) {
+    return device->iface.init_backend(device, params);
+}
+
+wsp_ggml_backend_buffer_type_t wsp_ggml_backend_dev_buffer_type(wsp_ggml_backend_dev_t device) {
+    return device->iface.get_buffer_type(device);
+}
+
+wsp_ggml_backend_buffer_type_t wsp_ggml_backend_dev_host_buffer_type(wsp_ggml_backend_dev_t device) {
+    if (device->iface.get_host_buffer_type == NULL) {
+        return NULL;
+    }
+
+    return device->iface.get_host_buffer_type(device);
+}
+
+wsp_ggml_backend_buffer_t wsp_ggml_backend_dev_buffer_from_host_ptr(wsp_ggml_backend_dev_t device, void * ptr, size_t size, size_t max_tensor_size) {
+    return device->iface.buffer_from_host_ptr(device, ptr, size, max_tensor_size);
+}
+
+bool wsp_ggml_backend_dev_supports_op(wsp_ggml_backend_dev_t device, const struct wsp_ggml_tensor * op) {
+    return device->iface.supports_op(device, op);
+}
+
+bool wsp_ggml_backend_dev_supports_buft(wsp_ggml_backend_dev_t device, wsp_ggml_backend_buffer_type_t buft) {
+    return device->iface.supports_buft(device, buft);
+}
+
+bool wsp_ggml_backend_dev_offload_op(wsp_ggml_backend_dev_t device, const struct wsp_ggml_tensor * op) {
+    if (device->iface.offload_op != NULL) {
+        return device->iface.offload_op(device, op);
+    }
+
+    return false;
+}
+
+// Backend (reg)
+
+const char * wsp_ggml_backend_reg_name(wsp_ggml_backend_reg_t reg) {
+    return reg->iface.get_name(reg);
+}
+
+size_t wsp_ggml_backend_reg_dev_count(wsp_ggml_backend_reg_t reg) {
+    return reg->iface.get_device_count(reg);
+}
+
+wsp_ggml_backend_dev_t wsp_ggml_backend_reg_dev_get(wsp_ggml_backend_reg_t reg, size_t index) {
+    return reg->iface.get_device(reg, index);
+}
+
+void * wsp_ggml_backend_reg_get_proc_address(wsp_ggml_backend_reg_t reg, const char * name) {
+    if (!reg->iface.get_proc_address) {
+        return NULL;
+    }
+    return reg->iface.get_proc_address(reg, name);
+}
+
+// multi-buffer buffer
+
+struct wsp_ggml_backend_multi_buffer_context {
+    wsp_ggml_backend_buffer_t * buffers;
+    size_t n_buffers;
+};
+
+static void wsp_ggml_backend_multi_buffer_free_buffer(wsp_ggml_backend_buffer_t buffer) {
+    wsp_ggml_backend_multi_buffer_context * ctx = (wsp_ggml_backend_multi_buffer_context *) buffer->context;
+    for (size_t i = 0; i < ctx->n_buffers; i++) {
+        wsp_ggml_backend_buffer_free(ctx->buffers[i]);
+    }
+
+    free(ctx->buffers);
+    free(ctx);
+}
+
+static void wsp_ggml_backend_multi_buffer_clear(wsp_ggml_backend_buffer_t buffer, uint8_t value) {
+    wsp_ggml_backend_multi_buffer_context * ctx = (wsp_ggml_backend_multi_buffer_context *) buffer->context;
+    for (size_t i = 0; i < ctx->n_buffers; i++) {
+        wsp_ggml_backend_buffer_clear(ctx->buffers[i], value);
+    }
+}
+
+static const struct wsp_ggml_backend_buffer_i wsp_ggml_backend_multi_buffer_i = {
+    /* .free_buffer     = */ wsp_ggml_backend_multi_buffer_free_buffer,
+    /* .get_base        = */ NULL,
+    /* .init_tensor     = */ NULL,
+    /* .memset_tensor   = */ NULL,
+    /* .set_tensor      = */ NULL,
+    /* .get_tensor      = */ NULL,
+    /* .cpy_tensor      = */ NULL,
+    /* .clear           = */ wsp_ggml_backend_multi_buffer_clear,
+    /* .reset           = */ NULL,
+};
+
+wsp_ggml_backend_buffer_t wsp_ggml_backend_multi_buffer_alloc_buffer(wsp_ggml_backend_buffer_t * buffers, size_t n_buffers) {
+    wsp_ggml_backend_multi_buffer_context * ctx = (wsp_ggml_backend_multi_buffer_context *) malloc(sizeof(struct wsp_ggml_backend_multi_buffer_context));
+    ctx->n_buffers = n_buffers;
+    ctx->buffers = (wsp_ggml_backend_buffer_t *) malloc(n_buffers * sizeof(wsp_ggml_backend_buffer_t));
+
+    WSP_GGML_ASSERT(ctx->buffers != NULL);
+
+    size_t total_size = 0;
+    for (size_t i = 0; i < n_buffers; i++) {
+        ctx->buffers[i] = buffers[i];
+        total_size += wsp_ggml_backend_buffer_get_size(buffers[i]);
+    }
+
+    return wsp_ggml_backend_buffer_init(buffers[0]->buft, wsp_ggml_backend_multi_buffer_i, ctx, total_size);
+}
+
+bool wsp_ggml_backend_buffer_is_multi_buffer(wsp_ggml_backend_buffer_t buffer) {
+    return buffer->iface.free_buffer == wsp_ggml_backend_multi_buffer_free_buffer;
+}
+
+void wsp_ggml_backend_multi_buffer_set_usage(wsp_ggml_backend_buffer_t buffer, enum wsp_ggml_backend_buffer_usage usage) {
+    WSP_GGML_ASSERT(wsp_ggml_backend_buffer_is_multi_buffer(buffer));
+    wsp_ggml_backend_multi_buffer_context * ctx = (wsp_ggml_backend_multi_buffer_context *) buffer->context;
+    for (size_t i = 0; i < ctx->n_buffers; i++) {
+        wsp_ggml_backend_buffer_set_usage(ctx->buffers[i], usage);
+    }
+}
+
+// creates a copy of the tensor with the same memory layout
+static struct wsp_ggml_tensor * wsp_ggml_dup_tensor_layout(struct wsp_ggml_context * ctx, const struct wsp_ggml_tensor * tensor) {
+    struct wsp_ggml_tensor * dup = wsp_ggml_dup_tensor(ctx, tensor);
+    for (int i = 0; i < WSP_GGML_MAX_DIMS; i++) {
+        dup->nb[i] = tensor->nb[i];
+    }
+    return dup;
+}
+
+static bool wsp_ggml_is_view_op(enum wsp_ggml_op op) {
+    return op == WSP_GGML_OP_VIEW || op == WSP_GGML_OP_RESHAPE || op == WSP_GGML_OP_PERMUTE || op == WSP_GGML_OP_TRANSPOSE;
+}
+
+// scheduler
+
+#ifndef WSP_GGML_SCHED_MAX_BACKENDS
+#define WSP_GGML_SCHED_MAX_BACKENDS 16
+#endif
+
+#ifndef WSP_GGML_SCHED_MAX_SPLIT_INPUTS
+#define WSP_GGML_SCHED_MAX_SPLIT_INPUTS WSP_GGML_MAX_SRC
+#endif
+
+#ifndef WSP_GGML_SCHED_MAX_COPIES
+#define WSP_GGML_SCHED_MAX_COPIES 4
+#endif
+
+struct wsp_ggml_backend_sched_split {
+    int backend_id;
+    int i_start;
+    int i_end;
+    struct wsp_ggml_tensor * inputs[WSP_GGML_SCHED_MAX_SPLIT_INPUTS];
+    int n_inputs;
+    // graph view of this split
+    struct wsp_ggml_cgraph graph;
+};
+
+struct wsp_ggml_backend_sched {
+    bool is_reset; // true if the scheduler has been reset since the last graph split
+    bool is_alloc;
+
+    int n_backends;
+
+    wsp_ggml_backend_t backends[WSP_GGML_SCHED_MAX_BACKENDS];
+    wsp_ggml_backend_buffer_type_t bufts[WSP_GGML_SCHED_MAX_BACKENDS];
+    wsp_ggml_gallocr_t galloc;
+
+    // hash map of the nodes in the graph
+    struct wsp_ggml_hash_set  hash_set;
+    int                 * hv_tensor_backend_ids; // [hash_set.size]
+    struct wsp_ggml_tensor ** hv_tensor_copies;      // [hash_set.size][n_backends][n_copies]
+
+    int * node_backend_ids; // [graph_size]
+    int * leaf_backend_ids; // [graph_size]
+
+    int * prev_node_backend_ids; // [graph_size]
+    int * prev_leaf_backend_ids; // [graph_size]
+
+    // copy of the graph with modified inputs
+    struct wsp_ggml_cgraph graph;
+
+    // graph splits
+    struct wsp_ggml_backend_sched_split * splits;
+    int n_splits;
+    int splits_capacity;
+
+    // pipeline parallelism support
+    int n_copies;
+    int cur_copy;
+    wsp_ggml_backend_event_t events[WSP_GGML_SCHED_MAX_BACKENDS][WSP_GGML_SCHED_MAX_COPIES];
+    struct wsp_ggml_tensor * graph_inputs[WSP_GGML_SCHED_MAX_SPLIT_INPUTS];
+    int n_graph_inputs;
+
+    struct wsp_ggml_context * ctx;
+
+    wsp_ggml_backend_sched_eval_callback callback_eval;
+    void * callback_eval_user_data;
+
+    char * context_buffer;
+    size_t context_buffer_size;
+
+    bool op_offload;
+
+    int debug;
+};
+
+#define hash_id(tensor) wsp_ggml_hash_find_or_insert(&sched->hash_set, tensor)
+#define tensor_backend_id(tensor) sched->hv_tensor_backend_ids[hash_id(tensor)]
+#define tensor_id_copy(id, backend_id, copy_id) sched->hv_tensor_copies[(id) * sched->n_backends * sched->n_copies + (backend_id) * sched->n_copies + (copy_id)]
+#define tensor_copy(tensor, backend_id, copy_id) tensor_id_copy(hash_id(tensor), backend_id, copy_id)
+
+// returns the priority of the backend, lower id is higher priority
+static int wsp_ggml_backend_sched_backend_id(wsp_ggml_backend_sched_t sched, wsp_ggml_backend_t backend) {
+    for (int i = 0; i < sched->n_backends; i++) {
+        if (sched->backends[i] == backend) {
+            return i;
+        }
+    }
+    return -1;
+}
+
+static int wsp_ggml_backend_sched_backend_from_buffer(wsp_ggml_backend_sched_t sched, const struct wsp_ggml_tensor * tensor, const struct wsp_ggml_tensor * op) {
+    wsp_ggml_backend_buffer_t buffer = tensor->view_src ? tensor->view_src->buffer : tensor->buffer;
+    if (buffer == NULL) {
+        return -1;
+    }
+
+    // find highest prio backend that supports the buffer type and the op
+    for (int i = 0; i < sched->n_backends; i++) {
+        if (wsp_ggml_backend_supports_buft(sched->backends[i], buffer->buft) &&
+            wsp_ggml_backend_supports_op(sched->backends[i], op)) {
+            return i;
+        }
+    }
+
+#ifndef NDEBUG
+    WSP_GGML_LOG_DEBUG("%s: warning: no backend supports op %s with a weight with buffer type %s used in tensor %s, the weight will need to be copied\n",
+        __func__, wsp_ggml_op_desc(tensor), wsp_ggml_backend_buffer_name(buffer), tensor->name);
+#endif
+
+    return -1;
+}
+
+#if 0
+#define WSP_GGML_SCHED_MAX_SPLITS_DEBUG 4096
+static char causes[WSP_GGML_DEFAULT_GRAPH_SIZE*16 + WSP_GGML_SCHED_MAX_SPLITS_DEBUG*WSP_GGML_SCHED_MAX_SPLIT_INPUTS][128]; // debug only
+#define SET_CAUSE(node, ...) sprintf(causes[hash_id(node)], __VA_ARGS__)
+#define GET_CAUSE(node) causes[hash_id(node)]
+#else
+#define SET_CAUSE(node, ...)
+#define GET_CAUSE(node) ""
+#endif
+
+// returns the backend that should be used for the node based on the current locations
+static int wsp_ggml_backend_sched_backend_id_from_cur(wsp_ggml_backend_sched_t sched, struct wsp_ggml_tensor * tensor) {
+    // assign pre-allocated nodes to their backend
+    int cur_backend_id = wsp_ggml_backend_sched_backend_from_buffer(sched, tensor, tensor);
+    if (cur_backend_id != -1) {
+        SET_CAUSE(tensor, "1.dst");
+        return cur_backend_id;
+    }
+
+    // view_src
+    if (tensor->view_src != NULL) {
+        cur_backend_id = wsp_ggml_backend_sched_backend_from_buffer(sched, tensor->view_src, tensor);
+        if (cur_backend_id != -1) {
+            SET_CAUSE(tensor, "1.vsrc");
+            return cur_backend_id;
+        }
+    }
+
+    if (tensor->buffer || (tensor->view_src && tensor->view_src->buffer)) {
+        // since the tensor is pre-allocated, it cannot be moved to another backend
+        wsp_ggml_backend_buffer_t buffer = tensor->view_src ? tensor->view_src->buffer : tensor->buffer;
+        WSP_GGML_ABORT("pre-allocated tensor (%s) in a buffer (%s) that cannot run the operation (%s)", tensor->name, wsp_ggml_backend_buffer_name(buffer), wsp_ggml_op_name(tensor->op));
+    }
+
+    // graph input
+    if (tensor->flags & WSP_GGML_TENSOR_FLAG_INPUT) {
+        cur_backend_id = sched->n_backends - 1; // last backend (assumed CPU)
+        SET_CAUSE(tensor, "1.inp");
+        return cur_backend_id;
+    }
+
+    // operations with weights are preferably run on the same backend as the weights
+    for (int i = 0; i < WSP_GGML_MAX_SRC; i++) {
+        const struct wsp_ggml_tensor * src = tensor->src[i];
+        if (src == NULL) {
+            continue;
+        }
+        // skip ROPE since the rope freqs tensor is too small to choose a backend based on it
+        // not an ideal solution
+        if (tensor->op != WSP_GGML_OP_ROPE && src->buffer != NULL && src->buffer->usage == WSP_GGML_BACKEND_BUFFER_USAGE_WEIGHTS) {
+            int src_backend_id = wsp_ggml_backend_sched_backend_from_buffer(sched, src, tensor);
+            // check if a backend with higher prio wants to offload the op
+            if (sched->op_offload && src_backend_id == sched->n_backends - 1 && wsp_ggml_backend_buffer_is_host(src->buffer)) {
+                for (int b = 0; b < src_backend_id; b++) {
+                    if (wsp_ggml_backend_supports_op(sched->backends[b], tensor) && wsp_ggml_backend_offload_op(sched->backends[b], tensor)) {
+                        SET_CAUSE(tensor, "1.off");
+                        return b;
+                    }
+                }
+            }
+            SET_CAUSE(tensor, "1.wgt%d", i);
+            return src_backend_id;
+        }
+    }
+
+    return -1;
+}
+
+static char * fmt_size(size_t size) {
+    static char buffer[128];
+    if (size >= 1024*1024) {
+        snprintf(buffer, sizeof(buffer), "%zuM", size/1024/1024);
+    } else {
+        snprintf(buffer, sizeof(buffer), "%zuK", size/1024);
+    }
+    return buffer;
+}
+
+static void wsp_ggml_backend_sched_print_assignments(wsp_ggml_backend_sched_t sched, struct wsp_ggml_cgraph * graph) {
+    int cur_split = 0;
+    for (int i = 0; i < graph->n_nodes; i++) {
+        if (cur_split < sched->n_splits && i == sched->splits[cur_split].i_start) {
+            wsp_ggml_backend_t split_backend = sched->backends[sched->splits[cur_split].backend_id];
+            WSP_GGML_LOG_DEBUG("\n## SPLIT #%d: %s # %d inputs", cur_split, wsp_ggml_backend_name(split_backend),
+                sched->splits[cur_split].n_inputs);
+            for (int j = 0; j < sched->splits[cur_split].n_inputs; j++) {
+                if (j == 0) {
+                    WSP_GGML_LOG_DEBUG(": ");
+                }
+                WSP_GGML_LOG_DEBUG("[%s (%5.5s)] ", sched->splits[cur_split].inputs[j]->name,
+                    fmt_size(wsp_ggml_nbytes(sched->splits[cur_split].inputs[j])));
+            }
+            WSP_GGML_LOG_DEBUG("\n");
+            cur_split++;
+        }
+        struct wsp_ggml_tensor * node = graph->nodes[i];
+        if (wsp_ggml_is_view_op(node->op)) {
+            continue;
+        }
+        if (sched->debug > 1) {
+            wsp_ggml_backend_t tensor_backend = wsp_ggml_backend_sched_get_tensor_backend(sched, node);
+            WSP_GGML_LOG_DEBUG("node #%3d (%10.10s): %20.20s (%5.5s) [%5.5s %8.8s]:", i, wsp_ggml_op_name(node->op), node->name,
+                fmt_size(wsp_ggml_nbytes(node)), tensor_backend ? wsp_ggml_backend_name(tensor_backend) : "NULL", GET_CAUSE(node));
+            for (int j = 0; j < WSP_GGML_MAX_SRC; j++) {
+                struct wsp_ggml_tensor * src = node->src[j];
+                if (src == NULL) {
+                    continue;
+                }
+                wsp_ggml_backend_t src_backend = wsp_ggml_backend_sched_get_tensor_backend(sched, src);
+                WSP_GGML_LOG_DEBUG(" %20.20s (%5.5s) [%5.5s %8.8s]", src->name,
+                    fmt_size(wsp_ggml_nbytes(src)), src_backend ? wsp_ggml_backend_name(src_backend) : "NULL", GET_CAUSE(src));
+            }
+            WSP_GGML_LOG_DEBUG("\n");
+        }
+    }
+}
+
+static bool wsp_ggml_backend_sched_buffer_supported(wsp_ggml_backend_sched_t sched, struct wsp_ggml_tensor * t, int backend_id) {
+    wsp_ggml_backend_buffer_t buf = t->view_src ? t->view_src->buffer : t->buffer;
+    wsp_ggml_backend_buffer_type_t buft = NULL;
+
+    if (buf) {
+        // the tensor is already allocated
+        buft = buf->buft;
+    } else {
+        // see if the tensor already has a backend assigned, and use the buffer type of that backend
+        int tensor_backend_id = tensor_backend_id(t);
+        if (tensor_backend_id == -1 && t->view_src) {
+            tensor_backend_id = tensor_backend_id(t->view_src);
+        }
+        if (tensor_backend_id != -1) {
+            buft = sched->bufts[tensor_backend_id];
+        }
+    }
+
+    return buft != NULL && wsp_ggml_backend_supports_buft(sched->backends[backend_id], buft);
+}
+
+static void wsp_ggml_backend_sched_set_if_supported(wsp_ggml_backend_sched_t sched, struct wsp_ggml_tensor * node, int cur_backend_id, int * node_backend_id) {
+    if (wsp_ggml_backend_supports_op(sched->backends[cur_backend_id], node)) {
+        *node_backend_id = cur_backend_id;
+        SET_CAUSE(node, "2.sup");
+    }
+}
+
+// assigns backends to ops and splits the graph into subgraphs that can be computed on the same backend
+static void wsp_ggml_backend_sched_split_graph(wsp_ggml_backend_sched_t sched, struct wsp_ggml_cgraph * graph) {
+    // reset splits
+    sched->n_splits = 0;
+    sched->n_graph_inputs = 0;
+    sched->is_reset = false;
+
+    struct wsp_ggml_init_params params = {
+        /* .mem_size =   */ sched->context_buffer_size,
+        /* .mem_buffer = */ sched->context_buffer,
+        /* .no_alloc =   */ true
+    };
+
+    wsp_ggml_free(sched->ctx);
+
+    sched->ctx = wsp_ggml_init(params);
+    if (sched->ctx == NULL) {
+        WSP_GGML_ABORT("%s: failed to initialize context\n", __func__);
+    }
+
+    // pass 1: assign backends to ops with pre-allocated inputs
+    for (int i = 0; i < graph->n_leafs; i++) {
+        struct wsp_ggml_tensor * leaf = graph->leafs[i];
+        int * leaf_backend_id = &tensor_backend_id(leaf);
+        // do not overwrite user assignments
+        if (*leaf_backend_id == -1) {
+            *leaf_backend_id = wsp_ggml_backend_sched_backend_id_from_cur(sched, leaf);
+        }
+    }
+
+    for (int i = 0; i < graph->n_nodes; i++) {
+        struct wsp_ggml_tensor * node = graph->nodes[i];
+        int * node_backend_id = &tensor_backend_id(node);
+        // do not overwrite user assignments
+        if (*node_backend_id == -1) {
+            *node_backend_id = wsp_ggml_backend_sched_backend_id_from_cur(sched, node);
+
+#if 0
+            // src
+            if (node->op == WSP_GGML_OP_NONE) {
+                continue;
+            }
+
+            for (int j = 0; j < WSP_GGML_MAX_SRC; j++) {
+                struct wsp_ggml_tensor * src = node->src[j];
+                if (src == NULL) {
+                    continue;
+                }
+                int * src_backend_id = &tensor_backend_id(src);
+                if (*src_backend_id == -1) {
+                    *src_backend_id = wsp_ggml_backend_sched_backend_id_from_cur(sched, src);
+                }
+            }
+#endif
+        }
+    }
+
+    // pass 2: expand current backend assignments
+    // assign the same backend to adjacent nodes
+    // expand gpu backends (i.e. non last prio) up and down, ignoring cpu (the lowest priority backend)
+    // thus, cpu will never be used unless weights are on cpu, or there are no gpu ops between cpu ops
+    // ops unsupported by the backend being expanded will be left unassigned so that they can be assigned later when the locations of its inputs are known
+    // expand gpu down
+    {
+        int cur_backend_id = -1;
+        for (int i = 0; i < graph->n_nodes; i++) {
+            struct wsp_ggml_tensor * node = graph->nodes[i];
+            if (wsp_ggml_is_view_op(node->op)) {
+                continue;
+            }
+            int * node_backend_id = &tensor_backend_id(node);
+            if (*node_backend_id != -1) {
+                if (*node_backend_id == sched->n_backends - 1) {
+                    // skip cpu (lowest prio backend)
+                    cur_backend_id = -1;
+                } else {
+                    cur_backend_id = *node_backend_id;
+                }
+            } else if (cur_backend_id != -1) {
+                wsp_ggml_backend_sched_set_if_supported(sched, node, cur_backend_id, node_backend_id);
+            }
+        }
+    }
+    // expand gpu up
+    {
+        int cur_backend_id = -1;
+        for (int i = graph->n_nodes - 1; i >= 0; i--) {
+            struct wsp_ggml_tensor * node = graph->nodes[i];
+            if (wsp_ggml_is_view_op(node->op)) {
+                continue;
+            }
+            int * node_backend_id = &tensor_backend_id(node);
+            if (*node_backend_id != -1) {
+                if (*node_backend_id == sched->n_backends - 1) {
+                    // skip cpu (lowest prio backend)
+                    cur_backend_id = -1;
+                } else {
+                    cur_backend_id = *node_backend_id;
+                }
+            } else if (cur_backend_id != -1) {
+                wsp_ggml_backend_sched_set_if_supported(sched, node, cur_backend_id, node_backend_id);
+            }
+        }
+    }
+    // expand rest down
+    {
+        int cur_backend_id = -1;
+        for (int i = 0; i < graph->n_nodes; i++) {
+            struct wsp_ggml_tensor * node = graph->nodes[i];
+            if (wsp_ggml_is_view_op(node->op)) {
+                continue;
+            }
+            int * node_backend_id = &tensor_backend_id(node);
+            if (*node_backend_id != -1) {
+                cur_backend_id = *node_backend_id;
+            } else if (cur_backend_id != -1) {
+                wsp_ggml_backend_sched_set_if_supported(sched, node, cur_backend_id, node_backend_id);
+            }
+        }
+    }
+    // expand rest up
+    {
+        int cur_backend_id = -1;
+        for (int i = graph->n_nodes - 1; i >= 0; i--) {
+            struct wsp_ggml_tensor * node = graph->nodes[i];
+            if (wsp_ggml_is_view_op(node->op)) {
+                continue;
+            }
+            int * node_backend_id = &tensor_backend_id(node);
+            if (*node_backend_id != -1) {
+                cur_backend_id = *node_backend_id;
+            } else if (cur_backend_id != -1) {
+                wsp_ggml_backend_sched_set_if_supported(sched, node, cur_backend_id, node_backend_id);
+            }
+        }
+    }
+
+    // pass 3: upgrade nodes to higher prio backends with compatible buffer types
+    // if the tensor is already in the same buffer type (*) as another higher priority backend, we should move it there
+    // however, we also need to verify that the sources are in compatible buffer types
+    // (*) the actual requirement is more relaxed, the buffer type of the backend should be supported by all the users of this tensor further down the graph
+    // however, this is slow to verify, so we have a more strict requirement that the buffer type is the same
+    // this is not uncommon since multiple backends can use host memory, with the same buffer type (eg. BLAS and CPU)
+    // additionally, set remaining unassigned nodes to the backend with the most supported inputs
+    // only nodes that could not be assigned during expansion due to the backend not supporting the op should be unassigned at this point
+    for (int i = 0; i < graph->n_nodes; i++) {
+        struct wsp_ggml_tensor * node = graph->nodes[i];
+        if (wsp_ggml_is_view_op(node->op)) {
+            continue;
+        }
+        int * node_backend_id = &tensor_backend_id(node);
+        if (*node_backend_id == -1) {
+            // unassigned node: find the backend with the most supported inputs
+            int n_supported_best = -1;
+            for (int b = 0; b < sched->n_backends; b++) {
+                if (wsp_ggml_backend_supports_op(sched->backends[b], node)) {
+                    int n_supported = 0;
+                    for (int j = 0; j < WSP_GGML_MAX_SRC; j++) {
+                        struct wsp_ggml_tensor * src = node->src[j];
+                        if (src == NULL) {
+                            continue;
+                        }
+                        if ((tensor_backend_id(src) != -1 || tensor_backend_id(src->view_src) != -1) && wsp_ggml_backend_sched_buffer_supported(sched, src, b)) {
+                            n_supported++;
+                        }
+                    }
+                    if (n_supported > n_supported_best) {
+                        n_supported_best = n_supported;
+                        *node_backend_id = b;
+                        SET_CAUSE(node, "3.best");
+                    }
+                }
+            }
+        } else {
+            // assigned node: upgrade to higher prio backend if possible
+            for (int b = 0; b < *node_backend_id; b++) {
+                if (sched->bufts[b] == sched->bufts[*node_backend_id] && wsp_ggml_backend_supports_op(sched->backends[b], node)) {
+                    bool supported = true;
+                    for (int j = 0; j < WSP_GGML_MAX_SRC; j++) {
+                        struct wsp_ggml_tensor * src = node->src[j];
+                        if (src == NULL) {
+                            continue;
+                        }
+                        if (!wsp_ggml_backend_sched_buffer_supported(sched, src, b)) {
+                            supported = false;
+                            break;
+                        }
+                    }
+                    if (supported) {
+                        *node_backend_id = b;
+                        SET_CAUSE(node, "3.upg");
+                        break;
+                    }
+                }
+            }
+        }
+    }
+
+    // pass 4: assign backends to remaining src from dst and view_src
+    for (int i = 0; i < graph->n_nodes; i++) {
+        struct wsp_ggml_tensor * node = graph->nodes[i];
+        int * cur_backend_id = &tensor_backend_id(node);
+        if (node->view_src != NULL && *cur_backend_id == -1) {
+            *cur_backend_id = tensor_backend_id(node->view_src);
+            SET_CAUSE(node, "4.vsrc");
+        }
+        for (int j = 0; j < WSP_GGML_MAX_SRC; j++) {
+            struct wsp_ggml_tensor * src = node->src[j];
+            if (src == NULL) {
+                continue;
+            }
+            int * src_backend_id = &tensor_backend_id(src);
+            if (*src_backend_id == -1) {
+                if (src->view_src != NULL) {
+                    // views are always on the same backend as the source
+                    *src_backend_id = tensor_backend_id(src->view_src);
+                    SET_CAUSE(src, "4.vsrc");
+                } else {
+                    *src_backend_id = *cur_backend_id;
+                    SET_CAUSE(src, "4.cur");
+                }
+            }
+        }
+    }
+
+    // pass 5: split graph, find tensors that need to be copied
+    {
+        int i_split = 0;
+        struct wsp_ggml_backend_sched_split * split = &sched->splits[0];
+        // find the backend of the first split, skipping view ops
+        int i = 0;
+        for (; i < graph->n_nodes; i++) {
+            struct wsp_ggml_tensor * node = graph->nodes[i];
+            if (!wsp_ggml_is_view_op(node->op)) {
+                split->backend_id = tensor_backend_id(node);
+                break;
+            }
+        }
+        split->i_start = 0;
+        split->n_inputs = 0;
+        int cur_backend_id = split->backend_id;
+        for (; i < graph->n_nodes; i++) {
+            struct wsp_ggml_tensor * node = graph->nodes[i];
+
+            if (wsp_ggml_is_view_op(node->op)) {
+                continue;
+            }
+
+            const int node_backend_id = tensor_backend_id(node);
+
+            assert(node_backend_id != -1); // all nodes should be assigned by now, this can happen if there is no CPU fallback
+
+            // check if we should start a new split based on the sources of the current node
+            bool need_new_split = false;
+            if (node_backend_id == cur_backend_id && split->n_inputs > 0) {
+                for (int j = 0; j < WSP_GGML_MAX_SRC; j++) {
+                    struct wsp_ggml_tensor * src = node->src[j];
+                    if (src == NULL) {
+                        continue;
+                    }
+                    // check if a weight is on a different and incompatible backend
+                    // by starting a new split, the memory of the previously offloaded weights can be reused
+                    if (src->buffer != NULL && src->buffer->usage == WSP_GGML_BACKEND_BUFFER_USAGE_WEIGHTS) {
+                        int src_backend_id = tensor_backend_id(src);
+                        if (src_backend_id != cur_backend_id && !wsp_ggml_backend_sched_buffer_supported(sched, src, cur_backend_id)) {
+                            need_new_split = true;
+                            break;
+                        }
+                    }
+                    // check if the split has too many inputs
+                    // FIXME: count the number of inputs instead of only checking when full
+                    if (split->n_inputs == WSP_GGML_SCHED_MAX_SPLIT_INPUTS) {
+                        const size_t id = hash_id(src);
+                        int src_backend_id = sched->hv_tensor_backend_ids[id];
+                        bool supported = wsp_ggml_backend_sched_buffer_supported(sched, src, cur_backend_id);
+                        if (src_backend_id != cur_backend_id && tensor_id_copy(id, cur_backend_id, 0) == NULL && !supported) {
+                            need_new_split = true;
+                            break;
+                        }
+                    }
+                }
+            }
+
+            if (node_backend_id != cur_backend_id || need_new_split) {
+                split->i_end = i;
+                i_split++;
+                if (i_split >= sched->splits_capacity) {
+                    sched->splits_capacity *= 2;
+                    sched->splits = (wsp_ggml_backend_sched_split *)
+                        realloc(sched->splits, sched->splits_capacity * sizeof(struct wsp_ggml_backend_sched_split));
+                    WSP_GGML_ASSERT(sched->splits != NULL);
+                }
+                split = &sched->splits[i_split];
+                split->backend_id = node_backend_id;
+                split->i_start = i;
+                split->n_inputs = 0;
+                cur_backend_id = node_backend_id;
+            }
+
+            // find inputs that are not on the same backend
+            for (int j = 0; j < WSP_GGML_MAX_SRC; j++) {
+                struct wsp_ggml_tensor * src = node->src[j];
+                if (src == NULL) {
+                    continue;
+                }
+
+                size_t src_id = hash_id(src);
+                const int src_backend_id = sched->hv_tensor_backend_ids[src_id];
+                assert(src_backend_id != -1); // all inputs should be assigned by now
+
+                if (src->flags & WSP_GGML_TENSOR_FLAG_INPUT && sched->n_copies > 1) {
+                    if (tensor_id_copy(src_id, src_backend_id, 0) == NULL) {
+                        wsp_ggml_backend_t backend = sched->backends[src_backend_id];
+                        for (int c = 0; c < sched->n_copies; c++) {
+                            struct wsp_ggml_tensor * tensor_copy;
+                            if (c == sched->cur_copy) {
+                                tensor_copy = src; // use the original tensor as the current copy
+                            } else {
+                                tensor_copy = wsp_ggml_dup_tensor_layout(sched->ctx, src);
+                                wsp_ggml_format_name(tensor_copy, "%s#%s#%d", wsp_ggml_backend_name(backend), src->name, c);
+                            }
+                            if (sched->n_copies > 1) {
+                                wsp_ggml_set_input(tensor_copy);
+                                wsp_ggml_set_output(tensor_copy); // prevent ggml-alloc from overwriting the tensor
+                            }
+                            tensor_id_copy(src_id, src_backend_id, c) = tensor_copy;
+                            SET_CAUSE(tensor_copy, "4.cpy");
+                        }
+                        int n_graph_inputs = sched->n_graph_inputs++;
+                        WSP_GGML_ASSERT(n_graph_inputs < WSP_GGML_SCHED_MAX_SPLIT_INPUTS);
+                        sched->graph_inputs[n_graph_inputs] = src;
+                    }
+                }
+
+                if (src_backend_id != cur_backend_id && !wsp_ggml_backend_sched_buffer_supported(sched, src, cur_backend_id)) {
+                    // create a copy of the input in the split's backend
+                    if (tensor_id_copy(src_id, cur_backend_id, 0) == NULL) {
+                        wsp_ggml_backend_t backend = sched->backends[cur_backend_id];
+                        for (int c = 0; c < sched->n_copies; c++) {
+                            struct wsp_ggml_tensor * tensor_copy = wsp_ggml_dup_tensor_layout(sched->ctx, src);
+                            wsp_ggml_format_name(tensor_copy, "%s#%s#%d", wsp_ggml_backend_name(backend), src->name, c);
+                            if (sched->n_copies > 1) {
+                                wsp_ggml_set_input(tensor_copy);
+                                wsp_ggml_set_output(tensor_copy); // prevent ggml-alloc from overwriting the tensor
+                            }
+                            tensor_id_copy(src_id, cur_backend_id, c) = tensor_copy;
+                            SET_CAUSE(tensor_copy, "4.cpy");
+                        }
+                        int n_inputs = split->n_inputs++;
+                        WSP_GGML_ASSERT(n_inputs < WSP_GGML_SCHED_MAX_SPLIT_INPUTS);
+                        split->inputs[n_inputs] = src;
+                    }
+                    node->src[j] = tensor_id_copy(src_id, cur_backend_id, sched->cur_copy);
+                }
+            }
+        }
+        split->i_end = graph->n_nodes;
+        sched->n_splits = i_split + 1;
+    }
+
+    if (sched->debug) {
+        wsp_ggml_backend_sched_print_assignments(sched, graph);
+    }
+
+    // swap node_backend_ids and leaf _backend_ids with prevs
+    {
+        int * tmp = sched->node_backend_ids;
+        sched->node_backend_ids = sched->prev_node_backend_ids;
+        sched->prev_node_backend_ids = tmp;
+
+        tmp = sched->leaf_backend_ids;
+        sched->leaf_backend_ids = sched->prev_leaf_backend_ids;
+        sched->prev_leaf_backend_ids = tmp;
+    }
+
+    int graph_size = std::max(graph->n_nodes, graph->n_leafs) + sched->n_splits*WSP_GGML_SCHED_MAX_SPLIT_INPUTS*2*sched->n_copies;
+    if (sched->graph.size < graph_size) {
+        sched->graph.size = graph_size;
+        sched->graph.nodes = (wsp_ggml_tensor **) realloc(sched->graph.nodes, graph_size * sizeof(struct wsp_ggml_tensor *));
+        sched->graph.leafs = (wsp_ggml_tensor **) realloc(sched->graph.leafs, graph_size * sizeof(struct wsp_ggml_tensor *));
+        WSP_GGML_ASSERT(sched->graph.nodes != NULL);
+        WSP_GGML_ASSERT(sched->graph.leafs != NULL);
+    }
+    sched->graph.n_nodes = 0;
+    sched->graph.n_leafs = 0;
+
+    struct wsp_ggml_cgraph * graph_copy = &sched->graph;
+
+    for (int i = 0; i < sched->n_splits; i++) {
+        struct wsp_ggml_backend_sched_split * split = &sched->splits[i];
+        split->graph = wsp_ggml_graph_view(graph, split->i_start, split->i_end);
+
+        // add inputs to the graph copy so that they are allocated by ggml-alloc at the start of the split
+        for (int j = 0; j < split->n_inputs; j++) {
+            assert(graph_copy->size > (graph_copy->n_nodes + 1));
+
+            struct wsp_ggml_tensor * input = split->inputs[j];
+            const size_t input_id = hash_id(input);
+            struct wsp_ggml_tensor * input_cpy = tensor_id_copy(input_id, split->backend_id, sched->cur_copy);
+
+            // add a dependency to the input source so that it is not freed before the copy is done
+            struct wsp_ggml_tensor * input_dep = wsp_ggml_view_tensor(sched->ctx, input);
+            input_dep->src[0] = input;
+            sched->node_backend_ids[graph_copy->n_nodes] = sched->hv_tensor_backend_ids[input_id];
+            graph_copy->nodes[graph_copy->n_nodes++] = input_dep;
+
+            // add a dependency to the input copy so that it is allocated at the start of the split
+            sched->node_backend_ids[graph_copy->n_nodes] = split->backend_id;
+            graph_copy->nodes[graph_copy->n_nodes++] = input_cpy;
+        }
+
+        for (int j = split->i_start; j < split->i_end; j++) {
+            assert(graph_copy->size > graph_copy->n_nodes);
+            sched->node_backend_ids[graph_copy->n_nodes] = tensor_backend_id(graph->nodes[j]);
+            graph_copy->nodes[graph_copy->n_nodes++] = graph->nodes[j];
+        }
+    }
+
+    if (sched->n_copies > 1) {
+        // add input copies as leafs so that they are allocated first
+        for (int i = 0; i < sched->n_graph_inputs; i++) {
+            struct wsp_ggml_tensor * input = sched->graph_inputs[i];
+            size_t id = hash_id(input);
+            int backend_id = tensor_backend_id(input);
+            for (int c = 0; c < sched->n_copies; c++) {
+                struct wsp_ggml_tensor * input_cpy = tensor_id_copy(id, backend_id, c);
+                sched->leaf_backend_ids[graph_copy->n_leafs] = backend_id;
+                assert(graph_copy->size > graph_copy->n_leafs);
+                graph_copy->leafs[graph_copy->n_leafs++] = input_cpy;
+            }
+        }
+
+        for (int i = 0; i < sched->n_splits; i++) {
+            struct wsp_ggml_backend_sched_split * split = &sched->splits[i];
+            int backend_id = split->backend_id;
+            for (int j = 0; j < split->n_inputs; j++) {
+                struct wsp_ggml_tensor * input = split->inputs[j];
+                size_t id = hash_id(input);
+                for (int c = 0; c < sched->n_copies; c++) {
+                    struct wsp_ggml_tensor * input_cpy = tensor_id_copy(id, backend_id, c);
+                    sched->leaf_backend_ids[graph_copy->n_leafs] = backend_id;
+                    assert(graph_copy->size > graph_copy->n_leafs);
+                    graph_copy->leafs[graph_copy->n_leafs++] = input_cpy;
+                }
+            }
+        }
+    }
+
+    // add leafs from the original graph
+    for (int i = 0; i < graph->n_leafs; i++) {
+        struct wsp_ggml_tensor * leaf = graph->leafs[i];
+        sched->leaf_backend_ids[graph_copy->n_leafs] = tensor_backend_id(leaf);
+        assert(graph_copy->size > graph_copy->n_leafs);
+        graph_copy->leafs[graph_copy->n_leafs++] = leaf;
+    }
+}
+
+static bool wsp_ggml_backend_sched_alloc_splits(wsp_ggml_backend_sched_t sched) {
+    bool backend_ids_changed = false;
+    for (int i = 0; i < sched->graph.n_nodes; i++) {
+        if (sched->node_backend_ids[i] != sched->prev_node_backend_ids[i] &&
+            sched->bufts[sched->node_backend_ids[i]] != sched->bufts[sched->prev_node_backend_ids[i]]) {
+            backend_ids_changed = true;
+            break;
+        }
+    }
+    if (!backend_ids_changed) {
+        for (int i = 0; i < sched->graph.n_leafs; i++) {
+            if (sched->leaf_backend_ids[i] != sched->prev_leaf_backend_ids[i] &&
+                sched->bufts[sched->leaf_backend_ids[i]] != sched->bufts[sched->prev_leaf_backend_ids[i]]) {
+                backend_ids_changed = true;
+                break;
+            }
+        }
+    }
+
+    // allocate graph
+    if (backend_ids_changed || !wsp_ggml_gallocr_alloc_graph(sched->galloc, &sched->graph)) {
+        // the re-allocation may cause the split inputs to be moved to a different address
+        // synchronize without wsp_ggml_backend_sched_synchronize to avoid changing cur_copy
+        for (int i = 0; i < sched->n_backends; i++) {
+            wsp_ggml_backend_synchronize(sched->backends[i]);
+        }
+#ifndef NDEBUG
+        WSP_GGML_LOG_DEBUG("%s: failed to allocate graph, reserving (backend_ids_changed = %d)\n", __func__, backend_ids_changed);
+#endif
+        wsp_ggml_gallocr_reserve_n(sched->galloc, &sched->graph, sched->node_backend_ids, sched->leaf_backend_ids);
+        if (!wsp_ggml_gallocr_alloc_graph(sched->galloc, &sched->graph)) {
+            WSP_GGML_LOG_ERROR("%s: failed to allocate graph\n", __func__);
+            return false;
+        }
+    }
+
+    return true;
+}
+
+static enum wsp_ggml_status wsp_ggml_backend_sched_compute_splits(wsp_ggml_backend_sched_t sched) {
+    struct wsp_ggml_backend_sched_split * splits = sched->splits;
+
+    for (int i = 0; i < sched->n_splits; i++) {
+        struct wsp_ggml_backend_sched_split * split = &splits[i];
+        int split_backend_id = split->backend_id;
+        wsp_ggml_backend_t split_backend = sched->backends[split_backend_id];
+
+        // copy the input tensors to the split backend
+        for (int j = 0; j < split->n_inputs; j++) {
+            wsp_ggml_backend_t input_backend = wsp_ggml_backend_sched_get_tensor_backend(sched, split->inputs[j]);
+            struct wsp_ggml_tensor * input = split->inputs[j];
+            struct wsp_ggml_tensor * input_cpy = tensor_copy(input, split_backend_id, sched->cur_copy);
+
+            if (input->flags & WSP_GGML_TENSOR_FLAG_INPUT) {
+                // inputs from the user must be copied immediately to prevent the user overwriting the data before the copy is done
+                if (sched->events[split_backend_id][sched->cur_copy] != NULL) {
+                    wsp_ggml_backend_event_synchronize(sched->events[split_backend_id][sched->cur_copy]);
+                } else {
+                    wsp_ggml_backend_synchronize(split_backend);
+                }
+                wsp_ggml_backend_tensor_copy(input, input_cpy);
+            } else {
+                // wait for the split backend to finish using the input before overwriting it
+                if (sched->events[split_backend_id][sched->cur_copy] != NULL) {
+                    wsp_ggml_backend_event_wait(split_backend, sched->events[split_backend_id][sched->cur_copy]);
+                } else {
+                    wsp_ggml_backend_synchronize(split_backend);
+                }
+                // try async copy, but if not possible, we can still use a sync copy without synchronizing the dst backend, since we handle the synchronization here with multiple copies and events
+                // TODO: add public function to facilitate this, since applications do not have direct access to the backend interface
+                if (!split_backend->iface.cpy_tensor_async || !split_backend->iface.cpy_tensor_async(input_backend, split_backend, input, input_cpy)) {
+                    wsp_ggml_backend_synchronize(input_backend);
+                    if (sched->events[split_backend_id][sched->cur_copy] != NULL) {
+                        wsp_ggml_backend_event_synchronize(sched->events[split_backend_id][sched->cur_copy]);
+                    } else {
+                        wsp_ggml_backend_synchronize(split_backend);
+                    }
+                    wsp_ggml_backend_tensor_copy(input, input_cpy);
+                }
+            }
+        }
+
+        if (!sched->callback_eval) {
+            enum wsp_ggml_status ec = wsp_ggml_backend_graph_compute_async(split_backend, &split->graph);
+            if (ec != WSP_GGML_STATUS_SUCCESS) {
+                return ec;
+            }
+        } else {
+            // similar to wsp_ggml_backend_compare_graph_backend
+            for (int j0 = 0; j0 < split->graph.n_nodes; j0++) {
+                struct wsp_ggml_tensor * t = split->graph.nodes[j0];
+
+                // check if the user needs data from this node
+                bool need = sched->callback_eval(t, true, sched->callback_eval_user_data);
+
+                int j1 = j0;
+
+                // determine the range [j0, j1] of nodes that can be computed together
+                while (!need && j1 < split->graph.n_nodes - 1) {
+                    t = split->graph.nodes[++j1];
+                    need = sched->callback_eval(t, true, sched->callback_eval_user_data);
+                }
+
+                struct wsp_ggml_cgraph gv = wsp_ggml_graph_view(&split->graph, j0, j1 + 1);
+
+                enum wsp_ggml_status ec = wsp_ggml_backend_graph_compute_async(split_backend, &gv);
+                if (ec != WSP_GGML_STATUS_SUCCESS) {
+                    return ec;
+                }
+
+                // TODO: pass backend to the callback, then the user can decide if they want to synchronize
+                wsp_ggml_backend_synchronize(split_backend);
+
+                if (need && !sched->callback_eval(t, false, sched->callback_eval_user_data)) {
+                    break;
+                }
+
+                j0 = j1;
+            }
+        }
+
+        // record the event of this copy
+        if (split->n_inputs > 0) {
+            if (sched->events[split_backend_id][sched->cur_copy] != NULL) {
+                wsp_ggml_backend_event_record(sched->events[split_backend_id][sched->cur_copy], split_backend);
+            }
+        }
+    }
+
+    sched->cur_copy = (sched->cur_copy + 1) % sched->n_copies;
+
+    return WSP_GGML_STATUS_SUCCESS;
+}
+
+wsp_ggml_backend_sched_t wsp_ggml_backend_sched_new(
+        wsp_ggml_backend_t * backends,
+        wsp_ggml_backend_buffer_type_t * bufts,
+        int n_backends,
+        size_t graph_size,
+        bool parallel,
+        bool op_offload) {
+    WSP_GGML_ASSERT(n_backends > 0);
+    WSP_GGML_ASSERT(n_backends <= WSP_GGML_SCHED_MAX_BACKENDS);
+    WSP_GGML_ASSERT(wsp_ggml_backend_dev_type(wsp_ggml_backend_get_device(backends[n_backends - 1])) == WSP_GGML_BACKEND_DEVICE_TYPE_CPU);
+
+    struct wsp_ggml_backend_sched * sched = (wsp_ggml_backend_sched *) calloc(1, sizeof(struct wsp_ggml_backend_sched));
+
+    const char * WSP_GGML_SCHED_DEBUG = getenv("WSP_GGML_SCHED_DEBUG");
+    sched->debug = WSP_GGML_SCHED_DEBUG ? atoi(WSP_GGML_SCHED_DEBUG) : 0;
+    sched->n_backends = n_backends;
+    sched->n_copies = parallel ? WSP_GGML_SCHED_MAX_COPIES : 1;
+
+    // initialize hash table
+    // FIXME: needs to be size*2 to account for leafs (do it in graph_split instead)
+    sched->hash_set    = wsp_ggml_hash_set_new(graph_size);
+    sched->hv_tensor_backend_ids = (int *) malloc(sched->hash_set.size * sizeof(sched->hv_tensor_backend_ids[0]));
+    sched->hv_tensor_copies      = (wsp_ggml_tensor **) malloc(sched->hash_set.size * sched->n_backends * sched->n_copies * sizeof(struct wsp_ggml_tensor *));
+
+    const size_t wsp_ggml_sched_max_splits = graph_size; // at most there is one split for each node in the graph
+    const size_t nodes_size = graph_size + wsp_ggml_sched_max_splits*WSP_GGML_SCHED_MAX_SPLIT_INPUTS*2;
+    sched->node_backend_ids = (int *) calloc(nodes_size, sizeof(sched->node_backend_ids[0]));
+    sched->leaf_backend_ids = (int *) calloc(nodes_size, sizeof(sched->leaf_backend_ids[0]));
+    sched->prev_node_backend_ids = (int *) calloc(nodes_size, sizeof(sched->prev_node_backend_ids[0]));
+    sched->prev_leaf_backend_ids = (int *) calloc(nodes_size, sizeof(sched->prev_leaf_backend_ids[0]));
+
+    sched->context_buffer_size = wsp_ggml_sched_max_splits*WSP_GGML_SCHED_MAX_SPLIT_INPUTS*2*sizeof(struct wsp_ggml_tensor) + wsp_ggml_graph_overhead_custom(graph_size, false);
+    sched->context_buffer = (char *) malloc(sched->context_buffer_size);
+
+    const int initial_splits_capacity = 16;
+    sched->splits = (wsp_ggml_backend_sched_split *) calloc(initial_splits_capacity, sizeof(sched->splits[0]));
+    sched->splits_capacity = initial_splits_capacity;
+
+    for (int b = 0; b < n_backends; b++) {
+        sched->backends[b] = backends[b];
+        sched->bufts[b] = bufts ? bufts[b] : wsp_ggml_backend_get_default_buffer_type(backends[b]);
+        WSP_GGML_ASSERT(wsp_ggml_backend_supports_buft(backends[b], sched->bufts[b]));
+
+        if (sched->n_copies > 1) {
+            for (int c = 0; c < sched->n_copies; c++) {
+                sched->events[b][c] = wsp_ggml_backend_event_new(backends[b]->device);
+            }
+        }
+    }
+
+    sched->galloc = wsp_ggml_gallocr_new_n(sched->bufts, n_backends);
+    sched->op_offload = op_offload;
+
+    wsp_ggml_backend_sched_reset(sched);
+
+    return sched;
+}
+
+void wsp_ggml_backend_sched_free(wsp_ggml_backend_sched_t sched) {
+    if (sched == NULL) {
+        return;
+    }
+    for (int b = 0; b < sched->n_backends; b++) {
+        for (int c = 0; c < sched->n_copies; c++) {
+            wsp_ggml_backend_event_free(sched->events[b][c]);
+        }
+    }
+    wsp_ggml_gallocr_free(sched->galloc);
+    wsp_ggml_free(sched->ctx);
+    wsp_ggml_hash_set_free(&sched->hash_set);
+    free(sched->splits);
+    free(sched->hv_tensor_backend_ids);
+    free(sched->hv_tensor_copies);
+    free(sched->node_backend_ids);
+    free(sched->leaf_backend_ids);
+    free(sched->prev_node_backend_ids);
+    free(sched->prev_leaf_backend_ids);
+    free(sched->context_buffer);
+    free(sched->graph.nodes);
+    free(sched->graph.leafs);
+    free(sched);
+}
+
+void wsp_ggml_backend_sched_reset(wsp_ggml_backend_sched_t sched) {
+    // reset state for the next run
+    if (!sched->is_reset) {
+        wsp_ggml_hash_set_reset(&sched->hash_set);
+        memset(sched->hv_tensor_backend_ids, -1, sched->hash_set.size * sizeof(sched->hv_tensor_backend_ids[0]));
+        memset(sched->hv_tensor_copies,       0, sched->hash_set.size * sched->n_backends * sched->n_copies * sizeof(struct wsp_ggml_tensor *));
+        sched->is_reset = true;
+    }
+    sched->is_alloc = false;
+}
+
+bool wsp_ggml_backend_sched_reserve(wsp_ggml_backend_sched_t sched, struct wsp_ggml_cgraph * measure_graph) {
+    WSP_GGML_ASSERT((int)sched->hash_set.size >= measure_graph->n_nodes + measure_graph->n_leafs);
+
+    wsp_ggml_backend_sched_split_graph(sched, measure_graph);
+
+    wsp_ggml_backend_sched_synchronize(sched);
+
+    if (!wsp_ggml_gallocr_reserve_n(sched->galloc, &sched->graph, sched->node_backend_ids, sched->leaf_backend_ids)) {
+        return false;
+    }
+
+    wsp_ggml_backend_sched_reset(sched);
+
+    return true;
+}
+
+bool wsp_ggml_backend_sched_alloc_graph(wsp_ggml_backend_sched_t sched, struct wsp_ggml_cgraph * graph) {
+    WSP_GGML_ASSERT((int)sched->hash_set.size >= graph->n_nodes + graph->n_leafs);
+
+    wsp_ggml_backend_sched_split_graph(sched, graph);
+
+    if (!wsp_ggml_backend_sched_alloc_splits(sched)) {
+        return false;
+    }
+
+    sched->is_alloc = true;
+
+    return true;
+}
+
+enum wsp_ggml_status wsp_ggml_backend_sched_graph_compute(wsp_ggml_backend_sched_t sched, struct wsp_ggml_cgraph * graph) {
+    enum wsp_ggml_status err = wsp_ggml_backend_sched_graph_compute_async(sched, graph);
+    wsp_ggml_backend_sched_synchronize(sched);
+    return err;
+}
+
+enum wsp_ggml_status wsp_ggml_backend_sched_graph_compute_async(wsp_ggml_backend_sched_t sched, struct wsp_ggml_cgraph * graph) {
+    if (!sched->is_reset && !sched->is_alloc) {
+        wsp_ggml_backend_sched_reset(sched);
+    }
+
+    if (!sched->is_alloc) {
+        if (!wsp_ggml_backend_sched_alloc_graph(sched, graph)) {
+            return WSP_GGML_STATUS_ALLOC_FAILED;
+        }
+    }
+
+    return wsp_ggml_backend_sched_compute_splits(sched);
+}
+
+void wsp_ggml_backend_sched_synchronize(wsp_ggml_backend_sched_t sched) {
+    for (int i = 0; i < sched->n_backends; i++) {
+        wsp_ggml_backend_synchronize(sched->backends[i]);
+    }
+    if (!sched->is_alloc) {
+        // if the graph is not already allocated, always use copy 0 after a synchronization
+        // this ensures that during generation the same copy is used every time,
+        // which avoids changes in the graph that could cause CUDA or other graphs to be disabled
+        sched->cur_copy = 0;
+    }
+}
+
+void wsp_ggml_backend_sched_set_eval_callback(wsp_ggml_backend_sched_t sched, wsp_ggml_backend_sched_eval_callback callback, void * user_data) {
+    sched->callback_eval = callback;
+    sched->callback_eval_user_data = user_data;
+}
+
+int wsp_ggml_backend_sched_get_n_splits(wsp_ggml_backend_sched_t sched) {
+    return sched->n_splits;
+}
+
+int wsp_ggml_backend_sched_get_n_copies(wsp_ggml_backend_sched_t sched) {
+    return sched->n_copies;
+}
+
+int wsp_ggml_backend_sched_get_n_backends(wsp_ggml_backend_sched_t sched) {
+    return sched->n_backends;
+}
+
+wsp_ggml_backend_t wsp_ggml_backend_sched_get_backend(wsp_ggml_backend_sched_t sched, int i) {
+    WSP_GGML_ASSERT(i >= 0 && i < sched->n_backends);
+    return sched->backends[i];
+}
+
+size_t wsp_ggml_backend_sched_get_buffer_size(wsp_ggml_backend_sched_t sched, wsp_ggml_backend_t backend) {
+    int backend_index = wsp_ggml_backend_sched_backend_id(sched, backend);
+    WSP_GGML_ASSERT(backend_index >= 0 && backend_index < sched->n_backends);
+
+    return wsp_ggml_gallocr_get_buffer_size(sched->galloc, backend_index);
+}
+
+void wsp_ggml_backend_sched_set_tensor_backend(wsp_ggml_backend_sched_t sched, struct wsp_ggml_tensor * node, wsp_ggml_backend_t backend) {
+    int backend_index = wsp_ggml_backend_sched_backend_id(sched, backend);
+    WSP_GGML_ASSERT(backend_index >= 0 && backend_index < sched->n_backends);
+    tensor_backend_id(node) = backend_index;
+    SET_CAUSE(node, "usr");
+    sched->is_reset = false;
+}
+
+wsp_ggml_backend_t wsp_ggml_backend_sched_get_tensor_backend(wsp_ggml_backend_sched_t sched, struct wsp_ggml_tensor * node) {
+    int backend_index = tensor_backend_id(node);
+    if (backend_index == -1) {
+        return NULL;
+    }
+    return sched->backends[backend_index];
+}
+
+// utils
+
+enum wsp_ggml_status wsp_ggml_backend_view_init(struct wsp_ggml_tensor * tensor) {
+    WSP_GGML_ASSERT(tensor->buffer == NULL);
+    WSP_GGML_ASSERT(tensor->view_src != NULL);
+    WSP_GGML_ASSERT(tensor->view_src->buffer != NULL);
+    WSP_GGML_ASSERT(tensor->view_src->data != NULL);
+
+    tensor->buffer = tensor->view_src->buffer;
+    tensor->data = (char *)tensor->view_src->data + tensor->view_offs;
+    return wsp_ggml_backend_buffer_init_tensor(tensor->buffer, tensor);
+}
+
+enum wsp_ggml_status wsp_ggml_backend_tensor_alloc(wsp_ggml_backend_buffer_t buffer, struct wsp_ggml_tensor * tensor, void * addr) {
+    WSP_GGML_ASSERT(tensor->buffer == NULL);
+    WSP_GGML_ASSERT(tensor->data == NULL);
+    WSP_GGML_ASSERT(tensor->view_src == NULL);
+    WSP_GGML_ASSERT(addr >= wsp_ggml_backend_buffer_get_base(buffer));
+    WSP_GGML_ASSERT((char *)addr + wsp_ggml_backend_buffer_get_alloc_size(buffer, tensor) <=
+                (char *)wsp_ggml_backend_buffer_get_base(buffer) + wsp_ggml_backend_buffer_get_size(buffer));
+
+    tensor->buffer = buffer;
+    tensor->data = addr;
+    return wsp_ggml_backend_buffer_init_tensor(buffer, tensor);
+}
+
+static struct wsp_ggml_tensor * graph_copy_dup_tensor(struct wsp_ggml_hash_set hash_set, struct wsp_ggml_tensor ** node_copies,
+    struct wsp_ggml_context * ctx_allocated, struct wsp_ggml_context * ctx_unallocated, struct wsp_ggml_tensor * src) {
+
+    WSP_GGML_ASSERT(src != NULL);
+    WSP_GGML_ASSERT(src->data && "graph must be allocated");
+
+    size_t id = wsp_ggml_hash_insert(&hash_set, src);
+    if (id == WSP_GGML_HASHSET_ALREADY_EXISTS) {
+        return node_copies[wsp_ggml_hash_find(&hash_set, src)];
+    }
+
+    struct wsp_ggml_tensor * dst = wsp_ggml_dup_tensor_layout(src->data && !src->view_src ? ctx_allocated : ctx_unallocated, src);
+    if (src->view_src != NULL) {
+        dst->view_src = graph_copy_dup_tensor(hash_set, node_copies, ctx_allocated, ctx_unallocated, src->view_src);
+        dst->view_offs = src->view_offs;
+    }
+    dst->op = src->op;
+    memcpy(dst->op_params, src->op_params, sizeof(dst->op_params));
+    wsp_ggml_set_name(dst, src->name);
+
+    // copy src
+    for (int i = 0; i < WSP_GGML_MAX_SRC; i++) {
+        struct wsp_ggml_tensor * s = src->src[i];
+        if (s == NULL) {
+            continue;
+        }
+        dst->src[i] = graph_copy_dup_tensor(hash_set, node_copies, ctx_allocated, ctx_unallocated, s);
+    }
+
+    node_copies[id] = dst;
+    return dst;
+}
+
+static void graph_copy_init_tensor(struct wsp_ggml_hash_set * hash_set, struct wsp_ggml_tensor ** node_copies, bool * node_init, struct wsp_ggml_tensor * src) {
+    size_t id = wsp_ggml_hash_find(hash_set, src);
+    if (node_init[id]) {
+        return;
+    }
+    node_init[id] = true;
+
+    struct wsp_ggml_tensor * dst = node_copies[id];
+    if (dst->view_src != NULL) {
+        graph_copy_init_tensor(hash_set, node_copies, node_init, src->view_src);
+        enum wsp_ggml_status status = wsp_ggml_backend_view_init(dst);
+        WSP_GGML_ASSERT(status == WSP_GGML_STATUS_SUCCESS);
+    }
+    else {
+        wsp_ggml_backend_tensor_copy(src, dst);
+    }
+
+    // init src
+    for (int i = 0; i < WSP_GGML_MAX_SRC; i++) {
+        struct wsp_ggml_tensor * s = src->src[i];
+        if (s == NULL) {
+            continue;
+        }
+        graph_copy_init_tensor(hash_set, node_copies, node_init, s);
+    }
+}
+
+struct wsp_ggml_backend_graph_copy wsp_ggml_backend_graph_copy(wsp_ggml_backend_t backend, struct wsp_ggml_cgraph * graph) {
+    struct wsp_ggml_hash_set hash_set = wsp_ggml_hash_set_new(graph->visited_hash_set.size);
+    struct wsp_ggml_tensor ** node_copies = (wsp_ggml_tensor **) calloc(hash_set.size, sizeof(node_copies[0])); // NOLINT
+    bool * node_init = (bool *) calloc(hash_set.size, sizeof(node_init[0]));
+
+    struct wsp_ggml_init_params params = {
+        /* .mem_size   = */ wsp_ggml_tensor_overhead()*hash_set.size + wsp_ggml_graph_overhead_custom(graph->size, false),
+        /* .mem_buffer = */ NULL,
+        /* .no_alloc   = */ true
+    };
+
+    struct wsp_ggml_context * ctx_allocated = wsp_ggml_init(params);
+    struct wsp_ggml_context * ctx_unallocated = wsp_ggml_init(params);
+
+    if (ctx_allocated == NULL || ctx_unallocated == NULL) {
+        WSP_GGML_LOG_ERROR("%s: failed to allocate context for graph copy\n", __func__);
+        wsp_ggml_hash_set_free(&hash_set);
+        free(node_copies);
+        free(node_init);
+        wsp_ggml_free(ctx_allocated);
+        wsp_ggml_free(ctx_unallocated);
+        return {
+            /* .buffer           = */ NULL,
+            /* .ctx_allocated    = */ NULL,
+            /* .ctx_unallocated  = */ NULL,
+            /* .graph            = */ NULL,
+        };
+    }
+
+    // dup nodes
+    for (int i = 0; i < graph->n_nodes; i++) {
+        struct wsp_ggml_tensor * node = graph->nodes[i];
+        graph_copy_dup_tensor(hash_set, node_copies, ctx_allocated, ctx_unallocated, node);
+    }
+
+    // allocate nodes
+    wsp_ggml_backend_buffer_t buffer = wsp_ggml_backend_alloc_ctx_tensors(ctx_allocated, backend);
+    if (buffer == NULL) {
+        WSP_GGML_LOG_ERROR("%s: failed to allocate buffer for graph copy\n", __func__);
+        wsp_ggml_hash_set_free(&hash_set);
+        free(node_copies);
+        free(node_init);
+        wsp_ggml_free(ctx_allocated);
+        wsp_ggml_free(ctx_unallocated);
+        return {
+            /* .buffer           = */ NULL,
+            /* .ctx_allocated    = */ NULL,
+            /* .ctx_unallocated  = */ NULL,
+            /* .graph            = */ NULL,
+        };
+    }
+
+    //printf("copy buffer size: %zu MB\n", wsp_ggml_backend_buffer_get_size(buffer) / 1024 / 1024);
+
+    // copy data and init views
+    for (int i = 0; i < graph->n_nodes; i++) {
+        struct wsp_ggml_tensor * node = graph->nodes[i];
+        graph_copy_init_tensor(&hash_set, node_copies, node_init, node);
+    }
+
+    // build graph copy
+    struct wsp_ggml_cgraph * graph_copy = wsp_ggml_new_graph_custom(ctx_allocated, graph->size, false);
+    for (int i = 0; i < graph->n_nodes; i++) {
+        struct wsp_ggml_tensor * node = graph->nodes[i];
+        struct wsp_ggml_tensor * node_copy = node_copies[wsp_ggml_hash_find(&hash_set, node)];
+        graph_copy->nodes[i] = node_copy;
+    }
+    graph_copy->n_nodes = graph->n_nodes;
+
+    wsp_ggml_hash_set_free(&hash_set);
+    free(node_copies);
+    free(node_init);
+
+    return {
+        /* .buffer           = */ buffer,
+        /* .ctx_allocated    = */ ctx_allocated,
+        /* .ctx_unallocated  = */ ctx_unallocated,
+        /* .graph            = */ graph_copy,
+    };
+}
+
+void wsp_ggml_backend_graph_copy_free(struct wsp_ggml_backend_graph_copy copy) {
+    wsp_ggml_backend_buffer_free(copy.buffer);
+    wsp_ggml_free(copy.ctx_allocated);
+    wsp_ggml_free(copy.ctx_unallocated);
+}
+
+bool wsp_ggml_backend_compare_graph_backend(wsp_ggml_backend_t backend1, wsp_ggml_backend_t backend2, struct wsp_ggml_cgraph * graph, wsp_ggml_backend_eval_callback callback, void * user_data) {
+    struct wsp_ggml_backend_graph_copy copy = wsp_ggml_backend_graph_copy(backend2, graph);
+    if (copy.buffer == NULL) {
+        return false;
+    }
+
+    struct wsp_ggml_cgraph * g1 = graph;
+    struct wsp_ggml_cgraph * g2 = copy.graph;
+
+    assert(g1->n_nodes == g2->n_nodes);
+
+    for (int i = 0; i < g1->n_nodes; i++) {
+        struct wsp_ggml_tensor * t1 = g1->nodes[i];
+        struct wsp_ggml_tensor * t2 = g2->nodes[i];
+
+        assert(t1->op == t2->op && wsp_ggml_are_same_layout(t1, t2));
+
+        struct wsp_ggml_cgraph g1v = wsp_ggml_graph_view(g1, i, i + 1);
+        struct wsp_ggml_cgraph g2v = wsp_ggml_graph_view(g2, i, i + 1);
+
+        wsp_ggml_backend_graph_compute(backend1, &g1v);
+        wsp_ggml_backend_graph_compute(backend2, &g2v);
+
+        if (wsp_ggml_is_view_op(t1->op)) {
+            continue;
+        }
+
+        // compare results, calculate rms etc
+        if (!callback(i, t1, t2, user_data)) {
+            break;
+        }
+    }
+
+    wsp_ggml_backend_graph_copy_free(copy);
+
+    return true;
+}
+
+// CPU backend - buffer
+
+static void * wsp_ggml_backend_cpu_buffer_get_base(wsp_ggml_backend_buffer_t buffer) {
+    uintptr_t data = (uintptr_t)buffer->context;
+
+    // align the buffer
+    if (data % TENSOR_ALIGNMENT != 0) {
+        data = WSP_GGML_PAD(data, TENSOR_ALIGNMENT);
+    }
+
+    return (void *)data;
+}
+
+static void wsp_ggml_backend_cpu_buffer_free_buffer(wsp_ggml_backend_buffer_t buffer) {
+    wsp_ggml_aligned_free(buffer->context, buffer->size);
+}
+
+static void wsp_ggml_backend_cpu_buffer_memset_tensor(wsp_ggml_backend_buffer_t buffer, struct wsp_ggml_tensor * tensor, uint8_t value, size_t offset, size_t size) {
+    memset((char *)tensor->data + offset, value, size);
+
+    WSP_GGML_UNUSED(buffer);
+}
+
+static void wsp_ggml_backend_cpu_buffer_set_tensor(wsp_ggml_backend_buffer_t buffer, struct wsp_ggml_tensor * tensor, const void * data, size_t offset, size_t size) {
+    memcpy((char *)tensor->data + offset, data, size);
+
+    WSP_GGML_UNUSED(buffer);
+}
+
+static void wsp_ggml_backend_cpu_buffer_get_tensor(wsp_ggml_backend_buffer_t buffer, const struct wsp_ggml_tensor * tensor, void * data, size_t offset, size_t size) {
+    memcpy(data, (const char *)tensor->data + offset, size);
+
+    WSP_GGML_UNUSED(buffer);
+}
+
+static bool wsp_ggml_backend_cpu_buffer_cpy_tensor(wsp_ggml_backend_buffer_t buffer, const struct wsp_ggml_tensor * src, struct wsp_ggml_tensor * dst) {
+    if (wsp_ggml_backend_buffer_is_host(src->buffer)) {
+        memcpy(dst->data, src->data, wsp_ggml_nbytes(src));
+        return true;
+    }
+    return false;
+
+    WSP_GGML_UNUSED(buffer);
+}
+
+static void wsp_ggml_backend_cpu_buffer_clear(wsp_ggml_backend_buffer_t buffer, uint8_t value) {
+    memset(buffer->context, value, buffer->size);
+}
+
+static const struct wsp_ggml_backend_buffer_i wsp_ggml_backend_cpu_buffer_i = {
+    /* .free_buffer     = */ wsp_ggml_backend_cpu_buffer_free_buffer,
+    /* .get_base        = */ wsp_ggml_backend_cpu_buffer_get_base,
+    /* .init_tensor     = */ NULL, // no initialization required
+    /* .memset_tensor   = */ wsp_ggml_backend_cpu_buffer_memset_tensor,
+    /* .set_tensor      = */ wsp_ggml_backend_cpu_buffer_set_tensor,
+    /* .get_tensor      = */ wsp_ggml_backend_cpu_buffer_get_tensor,
+    /* .cpy_tensor      = */ wsp_ggml_backend_cpu_buffer_cpy_tensor,
+    /* .clear           = */ wsp_ggml_backend_cpu_buffer_clear,
+    /* .reset           = */ NULL,
+};
+
+static const struct wsp_ggml_backend_buffer_i wsp_ggml_backend_cpu_buffer_from_ptr_i = {
+    /* .free_buffer     = */ NULL, // ptr is not owned by the buffer, so it does not need to be freed
+    /* .get_base        = */ wsp_ggml_backend_cpu_buffer_get_base,
+    /* .init_tensor     = */ NULL, // no initialization required
+    /* .memset_tensor   = */ wsp_ggml_backend_cpu_buffer_memset_tensor,
+    /* .set_tensor      = */ wsp_ggml_backend_cpu_buffer_set_tensor,
+    /* .get_tensor      = */ wsp_ggml_backend_cpu_buffer_get_tensor,
+    /* .cpy_tensor      = */ wsp_ggml_backend_cpu_buffer_cpy_tensor,
+    /* .clear           = */ wsp_ggml_backend_cpu_buffer_clear,
+    /* .reset           = */ NULL,
+};
+
+// CPU backend buffer type
+
+// this buffer type is defined here to make it available to all backends
+
+static const char * wsp_ggml_backend_cpu_buffer_type_get_name(wsp_ggml_backend_buffer_type_t buft) {
+    return "CPU";
+
+    WSP_GGML_UNUSED(buft);
+}
+
+static wsp_ggml_backend_buffer_t wsp_ggml_backend_cpu_buffer_type_alloc_buffer(wsp_ggml_backend_buffer_type_t buft, size_t size) {
+    void * data = wsp_ggml_aligned_malloc(size);
+
+    if (data == NULL) {
+        WSP_GGML_LOG_ERROR("%s: failed to allocate buffer of size %zu\n", __func__, size);
+        return NULL;
+    }
+
+    return wsp_ggml_backend_buffer_init(buft, wsp_ggml_backend_cpu_buffer_i, data, size);
+}
+
+static size_t wsp_ggml_backend_cpu_buffer_type_get_alignment(wsp_ggml_backend_buffer_type_t buft) {
+    return TENSOR_ALIGNMENT;
+
+    WSP_GGML_UNUSED(buft);
+}
+
+static bool wsp_ggml_backend_cpu_buffer_type_is_host(wsp_ggml_backend_buffer_type_t buft) {
+    return true;
+
+    WSP_GGML_UNUSED(buft);
+}
+
+wsp_ggml_backend_buffer_type_t wsp_ggml_backend_cpu_buffer_type(void) {
+    static struct wsp_ggml_backend_buffer_type wsp_ggml_backend_cpu_buffer_type = {
+        /* .iface   = */ {
+            /* .get_name         = */ wsp_ggml_backend_cpu_buffer_type_get_name,
+            /* .alloc_buffer     = */ wsp_ggml_backend_cpu_buffer_type_alloc_buffer,
+            /* .get_alignment    = */ wsp_ggml_backend_cpu_buffer_type_get_alignment,
+            /* .get_max_size     = */ NULL, // defaults to SIZE_MAX
+            /* .get_alloc_size   = */ NULL, // defaults to wsp_ggml_nbytes
+            /* .is_host          = */ wsp_ggml_backend_cpu_buffer_type_is_host,
+        },
+        /* .device  = */ NULL, // FIXME wsp_ggml_backend_reg_dev_get(wsp_ggml_backend_cpu_reg(), 0),
+        /* .context = */ NULL,
+    };
+
+    return &wsp_ggml_backend_cpu_buffer_type;
+}
+
+static const char * wsp_ggml_backend_cpu_buffer_from_ptr_type_get_name(wsp_ggml_backend_buffer_type_t buft) {
+    return "CPU_Mapped";
+
+    WSP_GGML_UNUSED(buft);
+}
+
+static wsp_ggml_backend_buffer_type_t wsp_ggml_backend_cpu_buffer_from_ptr_type(void) {
+    static struct wsp_ggml_backend_buffer_type wsp_ggml_backend_cpu_buffer_type = {
+        /* .iface   = */ {
+            /* .get_name         = */ wsp_ggml_backend_cpu_buffer_from_ptr_type_get_name,
+            /* .alloc_buffer     = */ wsp_ggml_backend_cpu_buffer_type_alloc_buffer,
+            /* .get_alignment    = */ wsp_ggml_backend_cpu_buffer_type_get_alignment,
+            /* .get_max_size     = */ NULL, // defaults to SIZE_MAX
+            /* .get_alloc_size   = */ NULL, // defaults to wsp_ggml_nbytes
+            /* .is_host          = */ wsp_ggml_backend_cpu_buffer_type_is_host,
+        },
+        /* .device  = */ NULL, // FIXME wsp_ggml_backend_reg_dev_get(wsp_ggml_backend_cpu_reg(), 0),
+        /* .context = */ NULL,
+    };
+
+    return &wsp_ggml_backend_cpu_buffer_type;
+}
+
+wsp_ggml_backend_buffer_t wsp_ggml_backend_cpu_buffer_from_ptr(void * ptr, size_t size) {
+    WSP_GGML_ASSERT((uintptr_t)ptr % TENSOR_ALIGNMENT == 0 && "buffer pointer must be aligned");
+    return wsp_ggml_backend_buffer_init(wsp_ggml_backend_cpu_buffer_from_ptr_type(), wsp_ggml_backend_cpu_buffer_from_ptr_i, ptr, size);
+}