| 1 |
#pragma clang diagnostic ignored "-Wmissing-prototypes" |
| 2 |
|
| 3 |
#include <metal_stdlib> |
| 4 |
#include <simd/simd.h> |
| 5 |
|
| 6 |
using namespace metal; |
| 7 |
|
| 8 |
struct SSBO |
| 9 |
{ |
| 10 |
float FragColor; |
| 11 |
}; |
| 12 |
|
| 13 |
constant uint3 gl_WorkGroupSize [[maybe_unused]] = uint3(1u); |
| 14 |
|
| 15 |
template<typename T> |
| 16 |
inline T spvSubgroupShuffle(T value, ushort lane) |
| 17 |
{ |
| 18 |
return quad_shuffle(value, lane); |
| 19 |
} |
| 20 |
|
| 21 |
template<> |
| 22 |
inline bool spvSubgroupShuffle(bool value, ushort lane) |
| 23 |
{ |
| 24 |
return !!quad_shuffle((ushort)value, lane); |
| 25 |
} |
| 26 |
|
| 27 |
template<uint N> |
| 28 |
inline vec<bool, N> spvSubgroupShuffle(vec<bool, N> value, ushort lane) |
| 29 |
{ |
| 30 |
return (vec<bool, N>)quad_shuffle((vec<ushort, N>)value, lane); |
| 31 |
} |
| 32 |
|
| 33 |
template<typename T> |
| 34 |
inline T spvSubgroupShuffleXor(T value, ushort mask) |
| 35 |
{ |
| 36 |
return quad_shuffle_xor(value, mask); |
| 37 |
} |
| 38 |
|
| 39 |
template<> |
| 40 |
inline bool spvSubgroupShuffleXor(bool value, ushort mask) |
| 41 |
{ |
| 42 |
return !!quad_shuffle_xor((ushort)value, mask); |
| 43 |
} |
| 44 |
|
| 45 |
template<uint N> |
| 46 |
inline vec<bool, N> spvSubgroupShuffleXor(vec<bool, N> value, ushort mask) |
| 47 |
{ |
| 48 |
return (vec<bool, N>)quad_shuffle_xor((vec<ushort, N>)value, mask); |
| 49 |
} |
| 50 |
|
| 51 |
template<typename T> |
| 52 |
inline T spvSubgroupShuffleUp(T value, ushort delta) |
| 53 |
{ |
| 54 |
return quad_shuffle_up(value, delta); |
| 55 |
} |
| 56 |
|
| 57 |
template<> |
| 58 |
inline bool spvSubgroupShuffleUp(bool value, ushort delta) |
| 59 |
{ |
| 60 |
return !!quad_shuffle_up((ushort)value, delta); |
| 61 |
} |
| 62 |
|
| 63 |
template<uint N> |
| 64 |
inline vec<bool, N> spvSubgroupShuffleUp(vec<bool, N> value, ushort delta) |
| 65 |
{ |
| 66 |
return (vec<bool, N>)quad_shuffle_up((vec<ushort, N>)value, delta); |
| 67 |
} |
| 68 |
|
| 69 |
template<typename T> |
| 70 |
inline T spvSubgroupShuffleDown(T value, ushort delta) |
| 71 |
{ |
| 72 |
return quad_shuffle_down(value, delta); |
| 73 |
} |
| 74 |
|
| 75 |
template<> |
| 76 |
inline bool spvSubgroupShuffleDown(bool value, ushort delta) |
| 77 |
{ |
| 78 |
return !!quad_shuffle_down((ushort)value, delta); |
| 79 |
} |
| 80 |
|
| 81 |
template<uint N> |
| 82 |
inline vec<bool, N> spvSubgroupShuffleDown(vec<bool, N> value, ushort delta) |
| 83 |
{ |
| 84 |
return (vec<bool, N>)quad_shuffle_down((vec<ushort, N>)value, delta); |
| 85 |
} |
| 86 |
|
| 87 |
template<typename T> |
| 88 |
inline T spvQuadBroadcast(T value, uint lane) |
| 89 |
{ |
| 90 |
return quad_broadcast(value, lane); |
| 91 |
} |
| 92 |
|
| 93 |
template<> |
| 94 |
inline bool spvQuadBroadcast(bool value, uint lane) |
| 95 |
{ |
| 96 |
return !!quad_broadcast((ushort)value, lane); |
| 97 |
} |
| 98 |
|
| 99 |
template<uint N> |
| 100 |
inline vec<bool, N> spvQuadBroadcast(vec<bool, N> value, uint lane) |
| 101 |
{ |
| 102 |
return (vec<bool, N>)quad_broadcast((vec<ushort, N>)value, lane); |
| 103 |
} |
| 104 |
|
| 105 |
template<typename T> |
| 106 |
inline T spvQuadSwap(T value, uint dir) |
| 107 |
{ |
| 108 |
return quad_shuffle_xor(value, dir + 1); |
| 109 |
} |
| 110 |
|
| 111 |
template<> |
| 112 |
inline bool spvQuadSwap(bool value, uint dir) |
| 113 |
{ |
| 114 |
return !!quad_shuffle_xor((ushort)value, dir + 1); |
| 115 |
} |
| 116 |
|
| 117 |
template<uint N> |
| 118 |
inline vec<bool, N> spvQuadSwap(vec<bool, N> value, uint dir) |
| 119 |
{ |
| 120 |
return (vec<bool, N>)quad_shuffle_xor((vec<ushort, N>)value, dir + 1); |
| 121 |
} |
| 122 |
|
| 123 |
kernel void main0(device SSBO& _9 [[buffer(0)]], uint gl_NumSubgroups [[quadgroups_per_threadgroup]], uint gl_SubgroupID [[quadgroup_index_in_threadgroup]], uint gl_SubgroupSize [[thread_execution_width]], uint gl_SubgroupInvocationID [[thread_index_in_quadgroup]]) |
| 124 |
{ |
| 125 |
_9.FragColor = float(gl_NumSubgroups); |
| 126 |
_9.FragColor = float(gl_SubgroupID); |
| 127 |
_9.FragColor = float(gl_SubgroupSize); |
| 128 |
_9.FragColor = float(gl_SubgroupInvocationID); |
| 129 |
simdgroup_barrier(mem_flags::mem_device | mem_flags::mem_threadgroup | mem_flags::mem_texture); |
| 130 |
simdgroup_barrier(mem_flags::mem_device | mem_flags::mem_threadgroup | mem_flags::mem_texture); |
| 131 |
simdgroup_barrier(mem_flags::mem_device); |
| 132 |
simdgroup_barrier(mem_flags::mem_threadgroup); |
| 133 |
simdgroup_barrier(mem_flags::mem_texture); |
| 134 |
uint shuffled = spvSubgroupShuffle(10u, 8u); |
| 135 |
bool shuffled_bool = spvSubgroupShuffle(true, 9u); |
| 136 |
uint shuffled_xor = spvSubgroupShuffleXor(30u, 8u); |
| 137 |
bool shuffled_xor_bool = spvSubgroupShuffleXor(false, 9u); |
| 138 |
uint shuffled_up = spvSubgroupShuffleUp(20u, 4u); |
| 139 |
bool shuffled_up_bool = spvSubgroupShuffleUp(true, 4u); |
| 140 |
uint shuffled_down = spvSubgroupShuffleDown(20u, 4u); |
| 141 |
bool shuffled_down_bool = spvSubgroupShuffleDown(false, 4u); |
| 142 |
float4 swap_horiz = spvQuadSwap(float4(20.0), 0u); |
| 143 |
bool4 swap_horiz_bool = spvQuadSwap(bool4(true), 0u); |
| 144 |
float4 swap_vertical = spvQuadSwap(float4(20.0), 1u); |
| 145 |
bool4 swap_vertical_bool = spvQuadSwap(bool4(true), 1u); |
| 146 |
float4 swap_diagonal = spvQuadSwap(float4(20.0), 2u); |
| 147 |
bool4 swap_diagonal_bool = spvQuadSwap(bool4(true), 2u); |
| 148 |
float4 quad_broadcast0 = spvQuadBroadcast(float4(20.0), 3u); |
| 149 |
bool4 quad_broadcast_bool = spvQuadBroadcast(bool4(true), 3u); |
| 150 |
} |
| 151 |
|