File size: 9,696 Bytes
c1af2fa |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 |
#pragma once
namespace at::mps {
static const char* SCATTER_OPS_TEMPLATE = R"METAL_SCATTER(
template<typename Y, typename X>
Y cast(const X x);
template<>
{1} cast<{1}, {0}>(const {0} x) {{
return {2};
}}
kernel void scatter_kernel_n(uint linear_index [[thread_position_in_grid]],
constant void * src_ [[buffer(0)]],
device void * dst_ [[buffer(1)]],
constant uint32_t * size [[buffer(2)]],
constant uint32_t * stride [[buffer(3)]],
constant uint32_t & numel [[buffer(4)]],
constant int32_t & ndim [[buffer(5)]]) {{
if (linear_index >= numel) return;
constant {0} * src = (constant {0} *)src_;
device {1} * dst = (device {1} *)dst_;
uint64_t dst_offs = 0;
auto dst_idx = linear_index;
for(int dim = ndim - 1; dim >= 0; --dim) {{
dst_offs += stride[dim] * (dst_idx % size[dim]);
dst_idx /= size[dim];
}}
dst[dst_offs] = cast<{1}>(src[linear_index]);
}}
kernel void scatter_kernel_4(uint linear_index [[thread_position_in_grid]],
constant void * src_ [[buffer(0)]],
device void * dst_ [[buffer(1)]],
constant packed_uint4 & size [[buffer(2)]],
constant packed_uint4 & stride [[buffer(3)]],
constant uint32_t & numel [[buffer(4)]]) {{
if (linear_index >= numel) return;
constant {0} * src = (constant {0} *)src_;
device {1} * dst = (device {1} *)dst_;
packed_uint4 local_index;
local_index.x = linear_index / (size[3] * size[2] * size[1]) % size[0];
local_index.y = linear_index / (size[3] * size[2]) % size[1];
local_index.z = linear_index / size[3] % size[2];
local_index.w = linear_index % size[3];
const packed_uint4 strided_index = local_index * stride;
dst[strided_index.x + strided_index.y + strided_index.z + strided_index.w] = cast<{1}>(src[linear_index]);
}}
kernel void scatter_kernel_3(uint linear_index [[thread_position_in_grid]],
constant void * src_ [[buffer(0)]],
device void * dst_ [[buffer(1)]],
constant packed_uint3 & size [[buffer(2)]],
constant packed_uint3 & stride [[buffer(3)]],
constant uint32_t & numel [[buffer(4)]]) {{
if (linear_index >= numel) return;
constant {0} * src = (constant {0} *)src_;
device {1} * dst = (device {1} *)dst_;
packed_uint3 local_index;
local_index.x = linear_index / (size[2] * size[1]) % size[0];
local_index.y = linear_index / size[2] % size[1];
local_index.z = linear_index % size[2];
const packed_uint3 strided_index = local_index * stride;
dst[strided_index.x + strided_index.y + strided_index.z] = cast<{1}>(src[linear_index]);
}}
kernel void scatter_kernel_2(uint linear_index [[thread_position_in_grid]],
constant void * src_ [[buffer(0)]],
device void * dst_ [[buffer(1)]],
constant packed_uint2 & size [[buffer(2)]],
constant packed_uint2 & stride [[buffer(3)]],
constant uint32_t & numel [[buffer(4)]]) {{
if (linear_index >= numel) return;
constant {0} * src = (constant {0} *)src_;
device {1} * dst = (device {1} *)dst_;
packed_uint2 local_index;
local_index.x = linear_index / size[1] % size[0];
local_index.y = linear_index % size[1];
const packed_uint2 strided_index = local_index * stride;
dst[strided_index.x + strided_index.y] = cast<{1}>(src[linear_index]);
}}
kernel void scatter_kernel_1(uint linear_index [[thread_position_in_grid]],
constant void * src_ [[buffer(0)]],
device void * dst_ [[buffer(1)]],
constant int & size [[buffer(2)]],
constant int & stride [[buffer(3)]],
constant uint32_t & numel [[buffer(4)]]) {{
if (linear_index >= numel) return;
constant {0} * src = (constant {0} *)src_;
device {1} * dst = (device {1} *)dst_;
const int local_index = linear_index % size;
const int strided_index = local_index * stride;
dst[strided_index] = cast<{1}>(src[linear_index]);
}}
)METAL_SCATTER";
static const char* GATHER_OPS_TEMPLATE = R"METAL_GATHER(
template<typename Y, typename X>
Y cast(const X x);
template<>
{1} cast<{1}, {0}>(const {0} x) {{
return {2};
}}
kernel void gather_kernel_n(uint linear_index [[thread_position_in_grid]],
constant void * src_ [[buffer(0)]],
device void * dst_ [[buffer(1)]],
constant uint32_t * size [[buffer(2)]],
constant uint32_t * stride [[buffer(3)]],
constant uint32_t & numel [[buffer(4)]],
constant int32_t & ndim [[buffer(5)]]) {{
if (linear_index >= numel) return;
constant {0} * src = (constant {0} *)src_;
device {1} * dst = (device {1} *)dst_;
uint64_t src_offs = 0;
auto src_idx = linear_index;
for(int dim = ndim - 1; dim >= 0; --dim) {{
src_offs += stride[dim] * (src_idx % size[dim]);
src_idx /= size[dim];
}}
dst[linear_index] = cast<{1}>(src[src_offs]);
}}
kernel void gather_kernel_4(uint linear_index [[thread_position_in_grid]],
constant void * src_ [[buffer(0)]],
device void * dst_ [[buffer(1)]],
constant packed_uint4 & size [[buffer(2)]],
constant packed_uint4 & stride [[buffer(3)]],
constant uint32_t & numel [[buffer(4)]]) {{
if (linear_index >= numel) return;
constant {0} * src = (constant {0} *)src_;
device {1} * dst = (device {1} *)dst_;
packed_uint4 local_index;
local_index.x = linear_index / (size[3] * size[2] * size[1]) % size[0];
local_index.y = linear_index / (size[3] * size[2]) % size[1];
local_index.z = linear_index / size[3] % size[2];
local_index.w = linear_index % size[3];
const packed_uint4 strided_index = local_index * stride;
dst[linear_index] = cast<{1}>(src[strided_index.x + strided_index.y + strided_index.z + strided_index.w]);
}}
kernel void gather_kernel_3(uint linear_index [[thread_position_in_grid]],
constant void * src_ [[buffer(0)]],
device void * dst_ [[buffer(1)]],
constant packed_uint3 & size [[buffer(2)]],
constant packed_uint3 & stride [[buffer(3)]],
constant uint32_t & numel [[buffer(4)]]) {{
if (linear_index >= numel) return;
constant {0} * src = (constant {0} *)src_;
device {1} * dst = (device {1} *)dst_;
packed_uint3 local_index;
local_index.x = linear_index / (size[2] * size[1]) % size[0];
local_index.y = linear_index / size[2] % size[1];
local_index.z = linear_index % size[2];
const packed_uint3 strided_index = local_index * stride;
dst[linear_index] = cast<{1}>(src[strided_index.x + strided_index.y + strided_index.z]);
}}
kernel void gather_kernel_2(uint linear_index [[thread_position_in_grid]],
constant void * src_ [[buffer(0)]],
device void * dst_ [[buffer(1)]],
constant packed_uint2 & size [[buffer(2)]],
constant packed_uint2 & stride [[buffer(3)]],
constant uint32_t & numel [[buffer(4)]]) {{
if (linear_index >= numel) return;
constant {0} * src = (constant {0} *)src_;
device {1} * dst = (device {1} *)dst_;
packed_uint2 local_index;
local_index.x = linear_index / size[1] % size[0];
local_index.y = linear_index % size[1];
const packed_uint2 strided_index = local_index * stride;
dst[linear_index] = cast<{1}>(src[strided_index.x + strided_index.y]);
}}
kernel void gather_kernel_1(uint linear_index [[thread_position_in_grid]],
constant void * src_ [[buffer(0)]],
device void * dst_ [[buffer(1)]],
constant int & size [[buffer(2)]],
constant int & stride [[buffer(3)]],
constant uint32_t & numel [[buffer(4)]]) {{
if (linear_index >= numel) return;
constant {0} * src = (constant {0} *)src_;
device {1} * dst = (device {1} *)dst_;
const int local_index = linear_index % size;
const int strided_index = local_index * stride;
dst[linear_index] = cast<{1}>(src[strided_index]);
}}
)METAL_GATHER";
} // namespace at::mps
|