cera 0.5.5

Rust-native LLM inference engine
Documentation 
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
222
223

#include <metal_stdlib>
#include <metal_math>
#include <metal_texture>
using namespace metal;

#line 48 "cera/src/backend/shaders/slang/layernorm_batch.slang"
struct KernelContext_0
{
    packed_uint4 device* par_buf_0;
    float device* src_buf_0;
    float device* dst_buf_0;
    float device* weight_buf_0;
    float device* bias_buf_0;
    array<float, int(256)> threadgroup* scratch_0;
};


#line 42
float block_sum_0(uint tid_0, float v_0, KernelContext_0 thread* kernelContext_0)
{



    float sg_0 = simd_sum(v_0);
    if((tid_0 & 31U) == 0U)
    {

#line 48
        (*kernelContext_0->scratch_0)[tid_0 >> 5U] = sg_0;

#line 48
    }
    threadgroup_barrier(mem_flags::mem_threadgroup);

#line 49
    float lane_0;
    if(tid_0 < 8U)
    {

#line 50
        lane_0 = (*kernelContext_0->scratch_0)[tid_0];

#line 50
    }
    else
    {

#line 50
        lane_0 = 0.0f;

#line 50
    }
    float total_0 = simd_sum(lane_0);
    if(tid_0 == 0U)
    {

#line 52
        (*kernelContext_0->scratch_0)[int(0)] = total_0;

#line 52
    }
    threadgroup_barrier(mem_flags::mem_threadgroup);
    float _S1 = (*kernelContext_0->scratch_0)[int(0)];

#line 69
    return _S1;
}



[[kernel]] void layernorm_batch(uint3 lid_0 [[thread_position_in_threadgroup]], uint3 wid_0 [[threadgroup_position_in_grid]], packed_uint4 device* par_buf_1 [[buffer(4)]], float device* src_buf_1 [[buffer(0)]], float device* dst_buf_1 [[buffer(1)]], float device* weight_buf_1 [[buffer(2)]], float device* bias_buf_1 [[buffer(3)]])
{

#line 74
    thread KernelContext_0 kernelContext_1;

#line 74
    (&kernelContext_1)->par_buf_0 = par_buf_1;

#line 74
    (&kernelContext_1)->src_buf_0 = src_buf_1;

#line 74
    (&kernelContext_1)->dst_buf_0 = dst_buf_1;

#line 74
    (&kernelContext_1)->weight_buf_0 = weight_buf_1;

#line 74
    (&kernelContext_1)->bias_buf_0 = bias_buf_1;

#line 74
    threadgroup array<float, int(256)> scratch_1;

#line 74
    (&kernelContext_1)->scratch_0 = &scratch_1;
    uint tid_1 = lid_0.x;
    uint row_0 = wid_0.x;
    uint n_0 = (uint4(*(par_buf_1+int(0))) ).x;
    float eps_0 = (as_type<float>(((uint4(*(par_buf_1+int(0))) ).y)));
    uint _S2 = row_0 * (uint4(*(par_buf_1+int(0))) ).z;
    uint _S3 = row_0 * (uint4(*(par_buf_1+int(0))) ).w;

#line 80
    uint i_0 = tid_1;

#line 80
    float partial_0 = 0.0f;



    for(;;)
    {

#line 84
        if(i_0 < n_0)
        {
        }
        else
        {

#line 84
            break;
        }

#line 85
        float partial_1 = partial_0 + (&kernelContext_1)->src_buf_0[_S2 + i_0];

#line 84
        i_0 = i_0 + 256U;

#line 84
        partial_0 = partial_1;

#line 84
    }

#line 84
    float _S4 = block_sum_0(tid_1, partial_0, &kernelContext_1);


    float _S5 = float(n_0);

#line 87
    float _S6 = _S4 / _S5;


    threadgroup_barrier(mem_flags::mem_threadgroup);

#line 90
    i_0 = tid_1;

#line 90
    partial_0 = 0.0f;



    for(;;)
    {

#line 94
        if(i_0 < n_0)
        {
        }
        else
        {

#line 94
            break;
        }

#line 95
        float d_0 = (&kernelContext_1)->src_buf_0[_S2 + i_0] - _S6;
        float partial_2 = partial_0 + d_0 * d_0;

#line 94
        i_0 = i_0 + 256U;

#line 94
        partial_0 = partial_2;

#line 94
    }

#line 94
    float _S7 = block_sum_0(tid_1, partial_0, &kernelContext_1);



    float _S8 = 1.0f / sqrt(_S7 / _S5 + eps_0);

#line 98
    i_0 = tid_1;


    for(;;)
    {

#line 101
        if(i_0 < n_0)
        {
        }
        else
        {

#line 101
            break;
        }

#line 102
        *((&kernelContext_1)->dst_buf_0+(_S3 + i_0)) = ((&kernelContext_1)->src_buf_0[_S2 + i_0] - _S6) * _S8 * (&kernelContext_1)->weight_buf_0[i_0] + (&kernelContext_1)->bias_buf_0[i_0];

#line 101
        i_0 = i_0 + 256U;

#line 101
    }



    return;
}