happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Move to no timeout for WaitAny in graph submission to avoid deadlocks in some cases on llvm-pipe backends (#20618)

This commit is contained in:
Reese Levine 2026-03-18 10:23:47 -07:00 committed by GitHub
parent 78d550b541
commit 8ced5f41f9
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 24 additions and 36 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

60
ggml/src/ggml-webgpu/ggml-webgpu.cpp
View File

@ -509,50 +509,39 @@ static void ggml_backend_webgpu_wait_profile_futures(webgpu_global_context &
static void ggml_backend_webgpu_wait(webgpu_global_context & ctx, static void ggml_backend_webgpu_wait(webgpu_global_context & ctx,
std::vector<webgpu_submission> & subs, std::vector<webgpu_submission> & subs,
bool block = true) { bool block = true) {
// If we have too many in-flight submissions, wait on the oldest one first.
if (subs.empty()) { if (subs.empty()) {
return; return;
} }
while (subs.size() >= WEBGPU_MAX_INFLIGHT_SUBS_PER_THREAD) {
auto waitStatus = ctx->instance.WaitAny(1, &subs[0].submit_done, UINT64_MAX); bool blocking_wait = block || subs.size() >= WEBGPU_MAX_INFLIGHT_SUBS_PER_THREAD;
if (ggml_backend_webgpu_handle_wait_status(waitStatus)) { while (blocking_wait) {
auto waitStatus = ctx->instance.WaitAny(1, &subs[0].submit_done, 0);
if (ggml_backend_webgpu_handle_wait_status(waitStatus, true)) {
#ifdef GGML_WEBGPU_GPU_PROFILE #ifdef GGML_WEBGPU_GPU_PROFILE
ggml_backend_webgpu_wait_profile_futures(ctx, subs[0].profile_futures, true); ggml_backend_webgpu_wait_profile_futures(ctx, subs[0].profile_futures, true);
#endif #endif
subs.erase(subs.begin()); subs.erase(subs.begin());
} }
blocking_wait = (block && !subs.empty()) || subs.size() >= WEBGPU_MAX_INFLIGHT_SUBS_PER_THREAD;
} }
if (subs.empty()) { if (subs.empty()) {
return; return;
} }
if (block) { // Poll each submit future once and remove completed submissions.
for (auto & sub : subs) { for (auto sub = subs.begin(); sub != subs.end();) {
while (!sub.submit_done.completed) { auto waitStatus = ctx->instance.WaitAny(1, &sub->submit_done, 0);
auto waitStatus = ctx->instance.WaitAny(1, &sub.submit_done, UINT64_MAX); bool success = ggml_backend_webgpu_handle_wait_status(waitStatus, true);
ggml_backend_webgpu_handle_wait_status(waitStatus);
}
#ifdef GGML_WEBGPU_GPU_PROFILE #ifdef GGML_WEBGPU_GPU_PROFILE
ggml_backend_webgpu_wait_profile_futures(ctx, sub.profile_futures, true); ggml_backend_webgpu_wait_profile_futures(ctx, sub->profile_futures, false);
#endif if (success && sub->profile_futures.empty()) {
}
subs.clear();
} else {
// Poll each submit future once and remove completed submissions.
for (auto sub = subs.begin(); sub != subs.end();) {
auto waitStatus = ctx->instance.WaitAny(1, &sub->submit_done, 0);
ggml_backend_webgpu_handle_wait_status(waitStatus, true);
#ifdef GGML_WEBGPU_GPU_PROFILE
ggml_backend_webgpu_wait_profile_futures(ctx, sub->profile_futures, false);
if (sub->submit_done.completed && sub->profile_futures.empty()) {
#else #else
if (sub->submit_done.completed) { if (success) {
#endif #endif
sub = subs.erase(sub); sub = subs.erase(sub);
} else { } else {
++sub; ++sub;
}
} }
} }
} }
@ -2961,17 +2950,16 @@ static ggml_backend_buffer_type_t ggml_backend_webgpu_device_get_buffer_type(ggm
static struct ggml_backend_buffer_type ggml_backend_webgpu_buffer_type = { static struct ggml_backend_buffer_type ggml_backend_webgpu_buffer_type = {
/* .iface = */ { /* .iface = */ {
/* .get_name = */ ggml_backend_webgpu_buffer_type_get_name, /* .get_name = */ ggml_backend_webgpu_buffer_type_get_name,
/* .alloc_buffer = */ /* .alloc_buffer = */ ggml_backend_webgpu_buffer_type_alloc_buffer,
ggml_backend_webgpu_buffer_type_alloc_buffer, /* .get_alignment = */ /* .get_alignment = */ ggml_backend_webgpu_buffer_type_get_alignment,
ggml_backend_webgpu_buffer_type_get_alignment, /* .get_max_size = */ /* .get_max_size = */ ggml_backend_webgpu_buffer_type_get_max_size,
ggml_backend_webgpu_buffer_type_get_max_size, /* .get_alloc_size = */ /* .get_alloc_size = */ ggml_backend_webgpu_buffer_type_get_alloc_size,
ggml_backend_webgpu_buffer_type_get_alloc_size, /* .is_host = */ NULL, // defaults to false /* .is_host = */ NULL, // defaults to false
}, },
/* .device = */ /* .device = */
dev, dev,
/* .context = */ /* .context = */ NULL
NULL
}; };
return &ggml_backend_webgpu_buffer_type; return &ggml_backend_webgpu_buffer_type;