mikhail/llama.cpp-mtp-turboquant
1
0
Fork 0
Code Issues Pull Requests Actions 30 Packages Projects Releases Wiki Activity

ggml-virtgpu: make the code thread safe (#19204)

Browse Source 
* ggml-virtgpu: regenerate_remoting.py: add the ability to deprecate a function

* ggml-virtgpu: deprecate buffer_type is_host remoting

not necessary

* ggml-virtgpu: stop using static vars as cache

The static init isn't thread safe.

* ggml-virtgpu: protect the use of the shared memory to transfer data

* ggml-virtgpu: make the remote calls thread-safe

* ggml-virtgpu: backend: don't continue if couldn't allocate the tensor memory

* ggml-virtgpu: add a cleanup function for consistency

* ggml-virtgpu: backend: don't crash if buft->iface.get_max_size is missing

* fix style and ordering

* Remove the static variable in apir_device_get_count

* ggml-virtgpu: improve the logging

* fix review minor formatting changes
This commit is contained in:
Kevin Pouget
2026-02-04 03:46:18 +01:00
committed by GitHub
parent 2ceda3f662
commit 015deb9048
27 changed files with 397 additions and 237 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-virtgpu/virtgpu-forward-device.cpp
+7 -15
View File
@@ -2,11 +2,6 @@
#include "virtgpu-shm.h"
int apir_device_get_count(virtgpu * gpu) {
static int32_t dev_count = -1;
if (dev_count != -1) {
return dev_count;
}
apir_encoder * encoder;
apir_decoder * decoder;
ApirForwardReturnCode ret;
@@ -14,6 +9,7 @@ int apir_device_get_count(virtgpu * gpu) {
REMOTE_CALL_PREPARE(gpu, encoder, APIR_COMMAND_TYPE_DEVICE_GET_COUNT);
REMOTE_CALL(gpu, encoder, decoder, ret);
int32_t dev_count = -1;
apir_decode_int32_t(decoder, &dev_count);
remote_call_finish(gpu, encoder, decoder);
@@ -21,11 +17,7 @@ int apir_device_get_count(virtgpu * gpu) {
return dev_count;
}
const char * apir_device_get_name(virtgpu * gpu) {
static char * string = nullptr;
if (string) {
return string;
}
char * apir_device_get_name(virtgpu * gpu) {
apir_encoder * encoder;
apir_decoder * decoder;
ApirForwardReturnCode ret;
@@ -34,9 +26,9 @@ const char * apir_device_get_name(virtgpu * gpu) {
REMOTE_CALL(gpu, encoder, decoder, ret);
const size_t string_size = apir_decode_array_size_unchecked(decoder);
string = (char *) apir_decoder_alloc_array(sizeof(char), string_size);
char * string = (char *) apir_decoder_alloc_array(sizeof(char), string_size);
if (!string) {
GGML_LOG_ERROR("%s: Could not allocate the device name buffer\n", __func__);
GGML_LOG_ERROR(GGML_VIRTGPU "%s: Could not allocate the device name buffer\n", __func__);
return NULL;
}
apir_decode_char_array(decoder, string, string_size);
@@ -46,7 +38,7 @@ const char * apir_device_get_name(virtgpu * gpu) {
return string;
}
const char * apir_device_get_description(virtgpu * gpu) {
char * apir_device_get_description(virtgpu * gpu) {
apir_encoder * encoder;
apir_decoder * decoder;
ApirForwardReturnCode ret;
@@ -58,7 +50,7 @@ const char * apir_device_get_description(virtgpu * gpu) {
const size_t string_size = apir_decode_array_size_unchecked(decoder);
char * string = (char *) apir_decoder_alloc_array(sizeof(char), string_size);
if (!string) {
GGML_LOG_ERROR("%s: Could not allocate the device description buffer\n", __func__);
GGML_LOG_ERROR(GGML_VIRTGPU "%s: Could not allocate the device description buffer\n", __func__);
return NULL;
}
@@ -181,7 +173,7 @@ apir_buffer_context_t apir_device_buffer_from_ptr(virtgpu * gpu, size_t size, si
REMOTE_CALL_PREPARE(gpu, encoder, APIR_COMMAND_TYPE_DEVICE_BUFFER_FROM_PTR);
if (virtgpu_shmem_create(gpu, size, &buffer_context.shmem)) {
GGML_ABORT("Couldn't allocate the guest-host shared buffer");
GGML_ABORT(GGML_VIRTGPU "Couldn't allocate the guest-host shared buffer");
}
apir_encode_virtgpu_shmem_res_id(encoder, buffer_context.shmem.res_id);