Will Usher
078c589aaa
Use an enum + switch statements to avoid issues/limitations with function ptrs. Renable volumes with OpenVKL when running on CPU w/ enable volumes Put both CPU and GPU modules behind CMake flags workaround odd bug with AO renderer and intervals Add spheres geometry and port to use Embree sphere point Push math constants for RNG to USM instead of in global constant WA for `Material_dispatch_getTransparency` hang, working path tracer Re-enable subdivision surfaces for CPU-only Port Texture2D to use switch statement dispatch Migrate all materials to switch dispatch Add option for large grf, does improve performance Workaround for TextureParam get3f bug Update headers and APIs for L0 interop header updates/deprecations
105 lines
3.5 KiB
Plaintext
105 lines
3.5 KiB
Plaintext
// Copyright 2009 Intel Corporation
|
|
// SPDX-License-Identifier: Apache-2.0
|
|
|
|
#pragma once
|
|
|
|
#ifdef OSPRAY_TARGET_DPCPP
|
|
#include "common/OSPCommon.ih"
|
|
#include "rkcommon/math/LinearSpace.h"
|
|
#include "rkcommon/math/box.h"
|
|
using namespace rkcommon;
|
|
using namespace rkcommon::math;
|
|
#else
|
|
#include "rkcommon/math/LinearSpace.ih"
|
|
#include "rkcommon/math/box.ih"
|
|
#endif
|
|
|
|
OSPRAY_BEGIN_ISPC_NAMESPACE
|
|
|
|
struct Data1D
|
|
{
|
|
uint8 *addr;
|
|
int64 byteStride;
|
|
uint32 numItems;
|
|
bool huge; // 64bit address calculation necessary
|
|
};
|
|
|
|
inline void Data1D_Constructor(Data1D *uniform self)
|
|
{
|
|
self->addr = NULL;
|
|
self->byteStride = 0;
|
|
self->numItems = 0;
|
|
self->huge = false;
|
|
}
|
|
|
|
// ok to pass-by-value, will be inlined and optimized
|
|
inline uniform bool valid(const uniform Data1D data)
|
|
{
|
|
return data.addr != NULL;
|
|
}
|
|
|
|
// special 64-bit safe code:
|
|
#define BITS 20
|
|
|
|
// ok to pass-by-value, will be inlined and optimized
|
|
#ifdef OSPRAY_TARGET_DPCPP
|
|
// DPC++ execution in a thread is scalar
|
|
#define __define_get(T) \
|
|
inline uniform T get_##T(const uniform Data1D data, const uniform int index) \
|
|
{ \
|
|
return *((const T *)(data.addr + data.byteStride * index)); \
|
|
}
|
|
#else
|
|
#define __define_get(T) \
|
|
inline T get_##T(const uniform Data1D data, const varying int index) \
|
|
{ \
|
|
if (data.huge) { \
|
|
T v; \
|
|
const int index_lo = index & ((1 << BITS) - 1); \
|
|
const int index_hi = index - index_lo; \
|
|
const int scaledIndexLo = index_lo * data.byteStride; \
|
|
foreach_unique (hi in index_hi) { \
|
|
/* uniform offset for upper bits */ \
|
|
const uniform uint64 scaledIndexHi = (uint64)(hi)*data.byteStride; \
|
|
/* properly shifted base address (shifted by 64-bits) */ \
|
|
const uint8 *uniform base_hi = data.addr + scaledIndexHi; \
|
|
v = *((const T *)(base_hi + scaledIndexLo)); \
|
|
} \
|
|
return v; \
|
|
} else \
|
|
return *((const T *)(data.addr + data.byteStride * index)); \
|
|
} \
|
|
inline uniform T get_##T(const uniform Data1D data, const uniform int index) \
|
|
{ \
|
|
return *((const T *)(data.addr + data.byteStride * index)); \
|
|
}
|
|
#endif
|
|
|
|
__define_get(int32);
|
|
__define_get(vec2i);
|
|
__define_get(vec3i);
|
|
__define_get(vec4i);
|
|
|
|
__define_get(uint8);
|
|
__define_get(uint32);
|
|
__define_get(vec2ui);
|
|
__define_get(vec3ui);
|
|
__define_get(vec4ui);
|
|
|
|
__define_get(float);
|
|
__define_get(vec2f);
|
|
__define_get(vec3f);
|
|
__define_get(vec4f);
|
|
|
|
__define_get(box1f);
|
|
__define_get(box2f);
|
|
__define_get(box3f);
|
|
__define_get(box4f);
|
|
|
|
__define_get(linear3f);
|
|
|
|
#undef __define_get
|
|
#undef BITS
|
|
|
|
OSPRAY_END_ISPC_NAMESPACE
|