ec054dad7f
Emit synthetic functions before function constants. Support use of spvQuantizeToF16() in function constants for numerical behavior consistency with the op code. Ensure subnormal results from OpQuantizeToF16 are flushed to zero per SPIR-V spec. Adjust SPIRV-Cross unit test reference shaders to accommodate these changes. Any MSL reference shader that inclues a synthetic function is affected, since the location it is emitted has changed.
136 lines
2.8 KiB
Plaintext
136 lines
2.8 KiB
Plaintext
#pragma clang diagnostic ignored "-Wmissing-prototypes"
|
|
|
|
#include <metal_stdlib>
|
|
#include <simd/simd.h>
|
|
|
|
using namespace metal;
|
|
|
|
template<typename T, uint A>
|
|
inline void spvArrayCopyFromConstantToStack1(thread T (&dst)[A], constant T (&src)[A])
|
|
{
|
|
for (uint i = 0; i < A; i++)
|
|
{
|
|
dst[i] = src[i];
|
|
}
|
|
}
|
|
|
|
template<typename T, uint A>
|
|
inline void spvArrayCopyFromConstantToThreadGroup1(threadgroup T (&dst)[A], constant T (&src)[A])
|
|
{
|
|
for (uint i = 0; i < A; i++)
|
|
{
|
|
dst[i] = src[i];
|
|
}
|
|
}
|
|
|
|
template<typename T, uint A>
|
|
inline void spvArrayCopyFromStackToStack1(thread T (&dst)[A], thread const T (&src)[A])
|
|
{
|
|
for (uint i = 0; i < A; i++)
|
|
{
|
|
dst[i] = src[i];
|
|
}
|
|
}
|
|
|
|
template<typename T, uint A>
|
|
inline void spvArrayCopyFromStackToThreadGroup1(threadgroup T (&dst)[A], thread const T (&src)[A])
|
|
{
|
|
for (uint i = 0; i < A; i++)
|
|
{
|
|
dst[i] = src[i];
|
|
}
|
|
}
|
|
|
|
template<typename T, uint A>
|
|
inline void spvArrayCopyFromThreadGroupToStack1(thread T (&dst)[A], threadgroup const T (&src)[A])
|
|
{
|
|
for (uint i = 0; i < A; i++)
|
|
{
|
|
dst[i] = src[i];
|
|
}
|
|
}
|
|
|
|
template<typename T, uint A>
|
|
inline void spvArrayCopyFromThreadGroupToThreadGroup1(threadgroup T (&dst)[A], threadgroup const T (&src)[A])
|
|
{
|
|
for (uint i = 0; i < A; i++)
|
|
{
|
|
dst[i] = src[i];
|
|
}
|
|
}
|
|
|
|
template<typename T, uint A>
|
|
inline void spvArrayCopyFromDeviceToDevice1(device T (&dst)[A], device const T (&src)[A])
|
|
{
|
|
for (uint i = 0; i < A; i++)
|
|
{
|
|
dst[i] = src[i];
|
|
}
|
|
}
|
|
|
|
template<typename T, uint A>
|
|
inline void spvArrayCopyFromConstantToDevice1(device T (&dst)[A], constant T (&src)[A])
|
|
{
|
|
for (uint i = 0; i < A; i++)
|
|
{
|
|
dst[i] = src[i];
|
|
}
|
|
}
|
|
|
|
template<typename T, uint A>
|
|
inline void spvArrayCopyFromStackToDevice1(device T (&dst)[A], thread const T (&src)[A])
|
|
{
|
|
for (uint i = 0; i < A; i++)
|
|
{
|
|
dst[i] = src[i];
|
|
}
|
|
}
|
|
|
|
template<typename T, uint A>
|
|
inline void spvArrayCopyFromThreadGroupToDevice1(device T (&dst)[A], threadgroup const T (&src)[A])
|
|
{
|
|
for (uint i = 0; i < A; i++)
|
|
{
|
|
dst[i] = src[i];
|
|
}
|
|
}
|
|
|
|
template<typename T, uint A>
|
|
inline void spvArrayCopyFromDeviceToStack1(thread T (&dst)[A], device const T (&src)[A])
|
|
{
|
|
for (uint i = 0; i < A; i++)
|
|
{
|
|
dst[i] = src[i];
|
|
}
|
|
}
|
|
|
|
template<typename T, uint A>
|
|
inline void spvArrayCopyFromDeviceToThreadGroup1(threadgroup T (&dst)[A], device const T (&src)[A])
|
|
{
|
|
for (uint i = 0; i < A; i++)
|
|
{
|
|
dst[i] = src[i];
|
|
}
|
|
}
|
|
|
|
struct Block
|
|
{
|
|
uint2 _m0[2];
|
|
uint2 _m1[2];
|
|
};
|
|
|
|
struct SSBO
|
|
{
|
|
Block _m0[3];
|
|
};
|
|
|
|
kernel void main0(device SSBO& ssbo [[buffer(0)]])
|
|
{
|
|
threadgroup uint2 _18[2];
|
|
uint2 _27[2];
|
|
spvArrayCopyFromDeviceToStack1(_27, ssbo._m0[0u]._m1);
|
|
spvArrayCopyFromStackToDevice1(ssbo._m0[0u]._m0, _27);
|
|
spvArrayCopyFromStackToDevice1(ssbo._m0[0u]._m0, _27);
|
|
}
|
|
|