blob: fcea166c3acd20eeb4370a4f1ed3d819e542d74d (
plain)
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
|
#version 450
#include "generic_binary_head.comp"
#include "types.comp"
#extension GL_EXT_control_flow_attributes : enable
#define BLOCK_SIZE 512
layout(local_size_x = BLOCK_SIZE, local_size_y = 1, local_size_z = 1) in;
shared FLOAT_TYPE sum[BLOCK_SIZE];
void main() {
const uint ncols = p.ne00;
const uint nrows = gl_NumWorkGroups.x;
const uint nchannels = gl_NumWorkGroups.y;
const uint row = gl_WorkGroupID.x;
const uint channel = gl_WorkGroupID.y;
const uint samp = gl_WorkGroupID.z;
const uint tid = gl_LocalInvocationID.x;
const uint stride_row_a = p.nb01;
const uint stride_channel_a = p.nb02;
const uint stride_sample_a = p.nb03;
uint32_t a_offset = samp*stride_sample_a + channel*stride_channel_a + row*stride_row_a;
uint32_t d_offset = ((samp*nchannels + channel)*nrows + row)*ncols + get_doffset();
FLOAT_TYPE sumf = FLOAT_TYPE(0.0f);
[[unroll]] for (uint col = tid; col < ncols; col += BLOCK_SIZE) {
const FLOAT_TYPE xi = FLOAT_TYPE(data_a[a_offset + col]);
sumf += xi * xi;
}
sum[tid] = sumf;
// sum up partial sums and write back result
barrier();
[[unroll]] for (int s = BLOCK_SIZE / 2; s > 0; s >>= 1) {
if (tid < s) {
sum[tid] += sum[tid + s];
}
barrier();
}
const FLOAT_TYPE mean = sum[0] / FLOAT_TYPE(ncols);
const FLOAT_TYPE scale = inversesqrt(mean + FLOAT_TYPE(p.param1));
[[unroll]] for (uint col = tid; col < ncols; col += BLOCK_SIZE) {
data_d[d_offset + col] = D_TYPE(scale * FLOAT_TYPE(data_a[a_offset + col]) * FLOAT_TYPE(data_b[col]));
}
}
|
