Cycles: oneAPI: Trigger compilation of used kernels only

JIT compilation of oneAPI kernels now happens during load stage
and proper message gets shown in the GUI during compilation.
Also, this implementation skips kernels that aren't needed for
the used scene, reducing overall (re)compilation time.
This commit is contained in:
Nikita Sirgienko 2022-10-10 16:37:40 +02:00
parent bb8dba8609
commit 82a5790d2a
3 changed files with 61 additions and 5 deletions

View File

@ -88,18 +88,26 @@ BVHLayoutMask OneapiDevice::get_bvh_layout_mask() const
bool OneapiDevice::load_kernels(const uint requested_features)
{
assert(device_queue_);
/* NOTE(@nsirgien): oneAPI can support compilation of kernel code with certain feature set
* with specialization constants, but it hasn't been implemented yet. */
(void)requested_features;
bool is_finished_ok = oneapi_run_test_kernel(device_queue_);
if (is_finished_ok == false) {
set_error("oneAPI kernel load: got runtime exception \"" + oneapi_error_string_ + "\"");
set_error("oneAPI test kernel execution: got a runtime exception \"" + oneapi_error_string_ +
"\"");
return false;
}
else {
VLOG_INFO << "Runtime compilation done for \"" << info.description << "\"";
VLOG_INFO << "Test kernel has been executed successfully for \"" << info.description << "\"";
assert(device_queue_);
}
is_finished_ok = oneapi_load_kernels(device_queue_, (const unsigned int)requested_features);
if (is_finished_ok == false) {
set_error("oneAPI kernels loading: got a runtime exception \"" + oneapi_error_string_ + "\"");
}
else {
VLOG_INFO << "Kernels loading (compilation) has been done for \"" << info.description << "\"";
}
return is_finished_ok;
}

View File

@ -123,6 +123,52 @@ size_t oneapi_kernel_preferred_local_size(SyclQueue *queue,
return std::min(limit_work_group_size, preferred_work_group_size);
}
bool oneapi_load_kernels(SyclQueue *queue_, const uint requested_features)
{
assert(queue_);
sycl::queue *queue = reinterpret_cast<sycl::queue *>(queue_);
try {
sycl::kernel_bundle<sycl::bundle_state::input> all_kernels_bundle =
sycl::get_kernel_bundle<sycl::bundle_state::input>(queue->get_context(),
{queue->get_device()});
for (const sycl::kernel_id &kernel_id : all_kernels_bundle.get_kernel_ids()) {
const std::string &kernel_name = kernel_id.get_name();
/* NOTE(@nsirgien): Names in this conditions below should match names from
* oneapi_call macro in oneapi_enqueue_kernel below */
if (((requested_features & KERNEL_FEATURE_VOLUME) == 0) &&
kernel_name.find("oneapi_kernel_integrator_shade_volume") != std::string::npos) {
continue;
}
if (((requested_features & KERNEL_FEATURE_MNEE) == 0) &&
kernel_name.find("oneapi_kernel_integrator_shade_surface_mnee") != std::string::npos) {
continue;
}
if (((requested_features & KERNEL_FEATURE_NODE_RAYTRACE) == 0) &&
kernel_name.find("oneapi_kernel_integrator_shade_surface_raytrace") !=
std::string::npos) {
continue;
}
sycl::kernel_bundle<sycl::bundle_state::input> one_kernel_bundle =
sycl::get_kernel_bundle<sycl::bundle_state::input>(queue->get_context(), {kernel_id});
sycl::build(one_kernel_bundle, {queue->get_device()}, sycl::property::queue::in_order());
}
}
catch (sycl::exception const &e) {
if (s_error_cb) {
s_error_cb(e.what(), s_error_user_ptr);
}
return false;
}
return true;
}
bool oneapi_enqueue_kernel(KernelContext *kernel_context,
int kernel,
size_t global_size,

View File

@ -48,6 +48,8 @@ CYCLES_KERNEL_ONEAPI_EXPORT bool oneapi_enqueue_kernel(KernelContext *context,
int kernel,
size_t global_size,
void **args);
CYCLES_KERNEL_ONEAPI_EXPORT bool oneapi_load_kernels(SyclQueue *queue,
const unsigned int requested_features);
# ifdef __cplusplus
}
# endif