Final touches

Author: qnixsynapse

ggml/src/ggml-sycl/binbcast.cpp

@@ -1,5 +1,4 @@
 #include "binbcast.hpp"
-#include "common.hpp"
 
 static __dpct_inline__ float op_repeat(const float a, const float b) {
     return b;

ggml/src/ggml-sycl/ggml-sycl.cpp

@@ -309,6 +309,7 @@ static void ggml_backend_sycl_buffer_set_tensor(ggml_backend_buffer_t buffer,
                                                 ggml_tensor *tensor,
                                                 const void *data, size_t offset,
                                                 size_t size) try {
+    GGML_SYCL_DEBUG("[SYCL] call %s", __func__);
 
     ggml_backend_sycl_buffer_context * ctx = ( ggml_backend_sycl_buffer_context *)buffer->context;
 
@@ -333,6 +334,7 @@ static void ggml_backend_sycl_buffer_get_tensor(ggml_backend_buffer_t buffer,
                                                 const ggml_tensor *tensor,
                                                 void *data, size_t offset,
                                                 size_t size) try {
+    GGML_SYCL_DEBUG("[SYCL] call %s", __func__);
 
     ggml_backend_sycl_buffer_context * ctx = ( ggml_backend_sycl_buffer_context *)buffer->context;
 
@@ -361,6 +363,7 @@ static bool
 ggml_backend_sycl_buffer_cpy_tensor(ggml_backend_buffer_t buffer,
                                     const ggml_tensor *src,
                                     ggml_tensor *dst) try {
+    GGML_SYCL_DEBUG("[SYCL] call %s", __func__);
     if (ggml_backend_buffer_is_sycl(src->buffer)) {
         ggml_backend_sycl_buffer_context * src_ctx = (ggml_backend_sycl_buffer_context *)src->buffer->context;
         ggml_backend_sycl_buffer_context * dst_ctx = (ggml_backend_sycl_buffer_context *)dst->buffer->context;
@@ -418,7 +421,8 @@ ggml_backend_sycl_buffer_cpy_tensor(ggml_backend_buffer_t buffer,
 
 static void ggml_backend_sycl_buffer_clear(ggml_backend_buffer_t buffer,
                                            uint8_t value) try {
-     ggml_backend_sycl_buffer_context * ctx = ( ggml_backend_sycl_buffer_context *)buffer->context;
+    GGML_SYCL_DEBUG("[SYCL] call %s", __func__);
+    ggml_backend_sycl_buffer_context * ctx = (ggml_backend_sycl_buffer_context *) buffer->context;
 
     ggml_sycl_set_device(ctx->device);
     queue_ptr stream = ctx->stream;
@@ -465,6 +469,7 @@ static const char * ggml_backend_sycl_buffer_type_get_name(ggml_backend_buffer_t
 static ggml_backend_buffer_t
 ggml_backend_sycl_buffer_type_alloc_buffer(ggml_backend_buffer_type_t buft,
                                            size_t size) try {
+    GGML_SYCL_DEBUG("[SYCL] call %s", __func__);
     ggml_backend_sycl_buffer_type_context * buft_ctx = (ggml_backend_sycl_buffer_type_context *)buft->context;
     ggml_sycl_set_device(buft_ctx->device);
     const queue_ptr stream = buft_ctx->stream;
@@ -708,6 +713,7 @@ static void * ggml_backend_sycl_split_buffer_get_base(ggml_backend_buffer_t buff
 static void
 ggml_backend_sycl_split_buffer_init_tensor(ggml_backend_buffer_t buffer,
                                            ggml_tensor *tensor) try {
+    GGML_SYCL_DEBUG("[SYCL] call %s", __func__);
     GGML_ASSERT(tensor->view_src == nullptr); // views of split tensors are not supported
 
     ggml_backend_sycl_split_buffer_context * ctx = (ggml_backend_sycl_split_buffer_context *)buffer->context;
@@ -791,6 +797,7 @@ static void
 ggml_backend_sycl_split_buffer_set_tensor(ggml_backend_buffer_t buffer,
                                           ggml_tensor *tensor, const void *data,
                                           size_t offset, size_t size) try {
+    GGML_SYCL_DEBUG("[SYCL] call %s", __func__);
     // split tensors must always be set in their entirety at once
     GGML_ASSERT(offset == 0);
     GGML_ASSERT(size == ggml_nbytes(tensor));
@@ -844,6 +851,7 @@ static void
 ggml_backend_sycl_split_buffer_get_tensor(ggml_backend_buffer_t buffer,
                                           const ggml_tensor *tensor, void *data,
                                           size_t offset, size_t size) try {
+    GGML_SYCL_DEBUG("[SYCL] call %s", __func__);
     // split tensors must always be set in their entirety at once
     GGML_ASSERT(offset == 0);
     GGML_ASSERT(size == ggml_nbytes(tensor));
@@ -894,6 +902,7 @@ catch (sycl::exception const &exc) {
 }
 
 static void ggml_backend_sycl_split_buffer_clear(ggml_backend_buffer_t buffer, uint8_t value) {
+    GGML_SYCL_DEBUG("[SYCL] call %s", __func__);
     GGML_UNUSED(buffer);
     GGML_UNUSED(value);
 }
@@ -1017,10 +1026,12 @@ static const char * ggml_backend_sycl_host_buffer_type_name(ggml_backend_buffer_
 }
 
 static void ggml_backend_sycl_host_buffer_free_buffer(ggml_backend_buffer_t buffer) {
+    GGML_SYCL_DEBUG("[SYCL] call %s", __func__);
     ggml_sycl_host_free(buffer->context);
 }
 
 static ggml_backend_buffer_t ggml_backend_sycl_host_buffer_type_alloc_buffer(ggml_backend_buffer_type_t buft, size_t size) {
+    GGML_SYCL_DEBUG("[SYCL] call %s", __func__);
     void * ptr = ggml_sycl_host_malloc(size);
 
     if (ptr == nullptr) {