mirror of
https://github.com/blender/blender
synced 2026-10-25 03:07:01 +03:00
5fc85d1cb7
The new intersect_mnee kernel runs before shade_surface, and shade_surface_mnee is eliminated. That large kernel was causing problems for some GPU compilers. MNEE state is packed into a shadow path state to avoid significantly increasing the path state size. This shadow state is then either turned into an actual shadow ray state or discarded in shade_surface. MNEE was re-enabled on HIP RDNA2 as it works again now. Texture cache misses now also work correctly with MNEE. This adds some extra code to the regular shade_surface kernel even when MNEE is not used, to use the MNEE sampled point instead of sampling a light. But there seems to be no significant performance impact. Co-authored-by: Sergey Sharybin <sergey@blender.org> Pull Request: https://projects.blender.org/blender/blender/pulls/158698
81 lines
2.4 KiB
C++
81 lines
2.4 KiB
C++
/* SPDX-FileCopyrightText: 2011-2022 Blender Foundation
|
|
*
|
|
* SPDX-License-Identifier: Apache-2.0 */
|
|
|
|
#ifdef WITH_HIP
|
|
# include "kernel/types.h"
|
|
|
|
# include "device/hip/device_impl.h"
|
|
# include "device/hip/kernel.h"
|
|
|
|
CCL_NAMESPACE_BEGIN
|
|
|
|
bool HIPDeviceKernels::load_kernel(HIPDevice *device,
|
|
hipModule_t hip_module,
|
|
const DeviceKernel kernel)
|
|
{
|
|
if (available(kernel)) {
|
|
return true;
|
|
}
|
|
|
|
if (!device_kernel_has_gpu_function(kernel)) {
|
|
return false;
|
|
}
|
|
|
|
HIPDeviceKernel &hip_kernel = kernels_[int(kernel)];
|
|
|
|
const std::string function_name = std::string("kernel_gpu_") + device_kernel_as_string(kernel);
|
|
hip_device_assert(device,
|
|
hipModuleGetFunction(&hip_kernel.function, hip_module, function_name.c_str()));
|
|
if (!hip_kernel.function) {
|
|
LOG_ERROR << "Unable to load kernel " << function_name;
|
|
return false;
|
|
}
|
|
|
|
hip_device_assert(device, hipFuncSetCacheConfig(hip_kernel.function, hipFuncCachePreferL1));
|
|
hip_device_assert(
|
|
device,
|
|
hipModuleOccupancyMaxPotentialBlockSize(
|
|
&hip_kernel.min_blocks, &hip_kernel.num_threads_per_block, hip_kernel.function, 0, 0));
|
|
|
|
LOG_DEBUG << "Loaded kernel: " << function_name;
|
|
|
|
return true;
|
|
}
|
|
|
|
void HIPDeviceKernels::load_all(HIPDevice *device, hipModule_t hip_module)
|
|
{
|
|
LOG_DEBUG << "Loading all HIP kernels";
|
|
|
|
for (int i = 0; i < (int)DEVICE_KERNEL_NUM; i++) {
|
|
load_kernel(device, hip_module, DeviceKernel(i));
|
|
}
|
|
}
|
|
|
|
void HIPDeviceKernels::load_raytrace(HIPDevice *device, hipModule_t hip_module)
|
|
{
|
|
LOG_DEBUG << "Loading ray-tracing HIP-RT kernels";
|
|
|
|
load_kernel(device, hip_module, DEVICE_KERNEL_INTEGRATOR_INTERSECT_CLOSEST);
|
|
load_kernel(device, hip_module, DEVICE_KERNEL_INTEGRATOR_INTERSECT_SHADOW);
|
|
load_kernel(device, hip_module, DEVICE_KERNEL_INTEGRATOR_INTERSECT_SUBSURFACE);
|
|
load_kernel(device, hip_module, DEVICE_KERNEL_INTEGRATOR_INTERSECT_VOLUME_STACK);
|
|
load_kernel(device, hip_module, DEVICE_KERNEL_INTEGRATOR_INTERSECT_DEDICATED_LIGHT);
|
|
load_kernel(device, hip_module, DEVICE_KERNEL_INTEGRATOR_INTERSECT_MNEE);
|
|
|
|
load_kernel(device, hip_module, DEVICE_KERNEL_INTEGRATOR_SHADE_SURFACE_RAYTRACE);
|
|
}
|
|
|
|
const HIPDeviceKernel &HIPDeviceKernels::get(DeviceKernel kernel) const
|
|
{
|
|
return kernels_[(int)kernel];
|
|
}
|
|
|
|
bool HIPDeviceKernels::available(DeviceKernel kernel) const
|
|
{
|
|
return kernels_[(int)kernel].function != nullptr;
|
|
}
|
|
|
|
CCL_NAMESPACE_END
|
|
|
|
#endif /* WITH_HIP */
|