Skip to content

[CANN]: Fix ggml_backend_cann_buffer_get_tensor #8871

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Aug 6, 2024
Merged

[CANN]: Fix ggml_backend_cann_buffer_get_tensor #8871

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
2340ff5
cann: fix ggml_backend_cann_buffer_get_tensor
MengqingCao Aug 5, 2024
f5ca40b
fix backend cann set_tensor
MengqingCao Aug 6, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
81 changes: 38 additions & 43 deletions ggml/src/ggml-cann.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -896,34 +896,32 @@ GGML_CALL static void ggml_backend_cann_buffer_init_tensor(
* @param size Size of the data to be copied, in bytes.
*/
GGML_CALL static void ggml_backend_cann_buffer_set_tensor(
ggml_backend_buffer_t buffer, ggml_tensor* tensor, const void* data,
ggml_backend_buffer_t buffer, ggml_tensor *tensor, const void *data,
size_t offset, size_t size) {
// GGML_ASSERT(size == ggml_nbytes(tensor));
ggml_backend_cann_buffer_context* ctx =
(ggml_backend_cann_buffer_context*)buffer->context;
ggml_backend_cann_buffer_context *ctx =
(ggml_backend_cann_buffer_context *)buffer->context;

ggml_cann_set_device(ctx->device);
// TODO: refer to cann(#6017), it use thread's default stream.
// For acl, synchronous functions use this default stream.
// Why aclrtSynchronizeDevice?

if (!need_transform(tensor->type)) {
ACL_CHECK(aclrtMemcpy(tensor->data, size, (const char*)data + offset,
size, ACL_MEMCPY_HOST_TO_DEVICE));
ACL_CHECK(aclrtMemcpy((char *)tensor->data + offset, size, data, size,
ACL_MEMCPY_HOST_TO_DEVICE));
} else {
void* transform_buffer = malloc(size);
ggml_backend_cann_transform(tensor, (const char*)data + offset,
transform_buffer);
void *transform_buffer = malloc(size);
ggml_backend_cann_transform(tensor, data, transform_buffer);

#ifndef NDEBUG
void* check_buffer = malloc(size);
void *check_buffer = malloc(size);
ggml_backend_cann_transform_back(tensor, transform_buffer,
check_buffer);
GGML_ASSERT(memcmp((const char*)data + offset, check_buffer, size) ==
0);
GGML_ASSERT(memcmp(data, check_buffer, size) == 0);
free(check_buffer);
#endif
ACL_CHECK(aclrtMemcpy(tensor->data, size, transform_buffer, size,
ACL_CHECK(aclrtMemcpy((char *)tensor->data + offset, size,
transform_buffer, size,
ACL_MEMCPY_HOST_TO_DEVICE));
free(transform_buffer);
}
Expand All @@ -945,21 +943,20 @@ GGML_CALL static void ggml_backend_cann_buffer_set_tensor(
GGML_CALL static void ggml_backend_cann_buffer_get_tensor(
ggml_backend_buffer_t buffer, const ggml_tensor* tensor, void* data,
size_t offset, size_t size) {
GGML_ASSERT(size == ggml_nbytes(tensor));
ggml_backend_cann_buffer_context* ctx =
(ggml_backend_cann_buffer_context*)buffer->context;

ggml_cann_set_device(ctx->device);

if (!need_transform(tensor->type)) {
ACL_CHECK(aclrtMemcpy((char*)data + offset, size, tensor->data, size,
ACL_CHECK(aclrtMemcpy(data, size, (char*)tensor->data + offset, size,
ACL_MEMCPY_DEVICE_TO_HOST));
} else {
void* transform_buffer = malloc(size);
ACL_CHECK(aclrtMemcpy(transform_buffer, size, tensor->data, size,
ACL_CHECK(aclrtMemcpy(transform_buffer, size,
(char*)tensor->data + offset, size,
ACL_MEMCPY_DEVICE_TO_HOST));
ggml_backend_cann_transform_back(tensor, transform_buffer,
(char*)data + offset);
ggml_backend_cann_transform_back(tensor, transform_buffer, data);
free(transform_buffer);
}
}
Expand Down Expand Up @@ -1448,60 +1445,58 @@ ggml_backend_cann_get_default_buffer_type(ggml_backend_t backend) {
* @param size Size of the data to copy in bytes.
*/
GGML_CALL static void ggml_backend_cann_set_tensor_async(ggml_backend_t backend,
ggml_tensor* tensor,
const void* data,
ggml_tensor *tensor,
const void *data,
size_t offset,
size_t size) {
ggml_backend_cann_context* cann_ctx =
(ggml_backend_cann_context*)backend->context;
ggml_backend_cann_context *cann_ctx =
(ggml_backend_cann_context *)backend->context;

if (!need_transform(tensor->type)) {
ACL_CHECK(aclrtMemcpyAsync(
tensor->data, size, (const char*)data + offset, size,
ACL_MEMCPY_HOST_TO_DEVICE, cann_ctx->stream()));
ACL_CHECK(aclrtMemcpyAsync((char *)tensor->data + offset, size, data,
size, ACL_MEMCPY_HOST_TO_DEVICE,
cann_ctx->stream()));
} else {
void* transform_buffer = malloc(size);
ggml_backend_cann_transform(tensor, (const char*)data + offset,
transform_buffer);
void *transform_buffer = malloc(size);
ggml_backend_cann_transform(tensor, data, transform_buffer);

#ifndef NDEBUG
void* check_buffer = malloc(size);
void *check_buffer = malloc(size);
ggml_backend_cann_transform_back(tensor, transform_buffer,
check_buffer);
GGML_ASSERT(memcmp((const char*)data + offset, check_buffer, size));
GGML_ASSERT(memcmp(data, check_buffer, size));
free(check_buffer);
#endif
ACL_CHECK(aclrtMemcpyAsync(tensor->data, size, transform_buffer, size,
ACL_MEMCPY_HOST_TO_DEVICE,
cann_ctx->stream()));
ACL_CHECK(aclrtMemcpyAsync(
(char *)tensor->data + offset, size, transform_buffer, size,
ACL_MEMCPY_HOST_TO_DEVICE, cann_ctx->stream()));
ACL_CHECK(aclrtSynchronizeStream(cann_ctx->stream()));
free(transform_buffer);
}
}

GGML_CALL static void ggml_backend_cann_get_tensor_async(
ggml_backend_t backend, const ggml_tensor* tensor, void* data,
ggml_backend_t backend, const ggml_tensor *tensor, void *data,
size_t offset, size_t size) {
ggml_backend_cann_context* cann_ctx =
(ggml_backend_cann_context*)backend->context;
ggml_backend_cann_context *cann_ctx =
(ggml_backend_cann_context *)backend->context;
ggml_backend_buffer_t buf =
tensor->view_src ? tensor->view_src->buffer : tensor->buffer;

GGML_ASSERT(buf->buft == ggml_backend_cann_buffer_type(cann_ctx->device) &&
"unsupported buffer type");

if (!need_transform(tensor->type)) {
ACL_CHECK(aclrtMemcpyAsync((char*)data + offset, size, tensor->data,
ACL_CHECK(aclrtMemcpyAsync(data, size, (char *)tensor->data + offset,
size, ACL_MEMCPY_DEVICE_TO_HOST,
cann_ctx->stream()));
} else {
void* transform_buffer = malloc(size);
ACL_CHECK(aclrtMemcpyAsync(transform_buffer, size, tensor->data, size,
ACL_MEMCPY_DEVICE_TO_HOST,
cann_ctx->stream()));
void *transform_buffer = malloc(size);
ACL_CHECK(aclrtMemcpyAsync(
transform_buffer, size, (char *)tensor->data + offset, size,
ACL_MEMCPY_DEVICE_TO_HOST, cann_ctx->stream()));
ACL_CHECK(aclrtSynchronizeStream(cann_ctx->stream()));
ggml_backend_cann_transform_back(tensor, transform_buffer,
(char*)data + offset);
ggml_backend_cann_transform_back(tensor, transform_buffer, data);
free(transform_buffer);
}
}
Expand Down
Loading