ferrum-kernels 0.8.3

Unified compute kernels (CUDA/Metal/CPU) and model runner for Ferrum inference
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

#include <metal_stdlib>
using namespace metal;

constant uint QK8_0 = 32;

struct LinearParams {
    uint rows;
    uint in_features;
    uint out_features;
    uint output_stride;
    uint output_column_offset;
};

struct SwiGluParams {
    uint rows;
    uint intermediate_size;
    uint gate_up_stride;
};

struct block_q8_0 {
    half d;
    char qs[QK8_0];
};

static inline float q8_0_value(device const block_q8_0 & block, uint index) {
    return float(block.d) * float(block.qs[index]);
}

kernel void vnext_linear_dense_f16(
    device const half * input [[buffer(0)]],
    device const half * weight [[buffer(1)]],
    device half * output [[buffer(2)]],
    constant LinearParams & params [[buffer(3)]],
    uint3 group [[threadgroup_position_in_grid]],
    uint simd_lane [[thread_index_in_simdgroup]],
    uint simd_group [[simdgroup_index_in_threadgroup]]) {
    const uint row = group.y;
    const uint first_output = group.x * 4 + simd_group * 2;
    float sums[2] = {0.0f, 0.0f};
    const ulong input_base = ulong(row) * params.in_features;
    for (uint column = simd_lane; column < params.in_features; column += 32) {
        const float activation = float(input[input_base + column]);
        for (uint local_output = 0; local_output < 2; ++local_output) {
            const uint output_column = first_output + local_output;
            if (output_column < params.out_features) {
                sums[local_output] += activation
                    * float(weight[ulong(output_column) * params.in_features + column]);
            }
        }
    }
    for (uint local_output = 0; local_output < 2; ++local_output) {
        const uint output_column = first_output + local_output;
        const float total = simd_sum(sums[local_output]);
        if (simd_lane == 0 && output_column < params.out_features) {
            output[ulong(row) * params.output_stride
                + params.output_column_offset + output_column] = half(total);
        }
    }
}

kernel void vnext_linear_q8_0_f16(
    device const half * input [[buffer(0)]],
    device const block_q8_0 * weight [[buffer(1)]],
    device half * output [[buffer(2)]],
    constant LinearParams & params [[buffer(3)]],
    uint3 group [[threadgroup_position_in_grid]],
    uint simd_lane [[thread_index_in_simdgroup]],
    uint simd_group [[simdgroup_index_in_threadgroup]]) {
    const uint row = group.y;
    const uint first_output = group.x * 4 + simd_group * 2;
    const uint blocks_per_row = params.in_features / QK8_0;
    float sums[2] = {0.0f, 0.0f};
    const ulong input_base = ulong(row) * params.in_features;
    for (uint column = simd_lane; column < params.in_features; column += 32) {
        const float activation = float(input[input_base + column]);
        for (uint local_output = 0; local_output < 2; ++local_output) {
            const uint output_column = first_output + local_output;
            if (output_column < params.out_features) {
                const ulong weight_base = ulong(output_column) * blocks_per_row;
                sums[local_output] += activation * q8_0_value(
                    weight[weight_base + column / QK8_0], column % QK8_0
                );
            }
        }
    }
    for (uint local_output = 0; local_output < 2; ++local_output) {
        const uint output_column = first_output + local_output;
        const float total = simd_sum(sums[local_output]);
        if (simd_lane == 0 && output_column < params.out_features) {
            output[ulong(row) * params.output_stride
                + params.output_column_offset + output_column] = half(total);
        }
    }
}

kernel void vnext_swiglu_f16(
    device const half * gate_up [[buffer(0)]],
    device half * activation [[buffer(1)]],
    constant SwiGluParams & params [[buffer(2)]],
    uint index [[thread_position_in_grid]]) {
    const uint elements = params.rows * params.intermediate_size;
    if (index >= elements) {
        return;
    }
    const uint row = index / params.intermediate_size;
    const uint column = index % params.intermediate_size;
    const ulong base = ulong(row) * params.gate_up_stride;
    const float gate = float(gate_up[base + column]);
    const float up = float(gate_up[base + params.intermediate_size + column]);
    activation[index] = half((gate / (1.0f + exp(-gate))) * up);
}

kernel void vnext_linear_dense_f32(
    device const float * input [[buffer(0)]],
    device const half * weight [[buffer(1)]],
    device float * output [[buffer(2)]],
    constant LinearParams & params [[buffer(3)]],
    uint3 group [[threadgroup_position_in_grid]],
    uint simd_lane [[thread_index_in_simdgroup]],
    uint simd_group [[simdgroup_index_in_threadgroup]]) {
    const uint row = group.y;
    const uint first_output = group.x * 4 + simd_group * 2;
    float sums[2] = {0.0f, 0.0f};
    const ulong input_base = ulong(row) * params.in_features;
    for (uint column = simd_lane; column < params.in_features; column += 32) {
        const float activation = input[input_base + column];
        for (uint local_output = 0; local_output < 2; ++local_output) {
            const uint output_column = first_output + local_output;
            if (output_column < params.out_features) {
                sums[local_output] += activation
                    * float(weight[ulong(output_column) * params.in_features + column]);
            }
        }
    }
    for (uint local_output = 0; local_output < 2; ++local_output) {
        const uint output_column = first_output + local_output;
        const float total = simd_sum(sums[local_output]);
        if (simd_lane == 0 && output_column < params.out_features) {
            output[ulong(row) * params.output_stride
                + params.output_column_offset + output_column] = total;
        }
    }
}

kernel void vnext_linear_q8_0_f32(
    device const float * input [[buffer(0)]],
    device const block_q8_0 * weight [[buffer(1)]],
    device float * output [[buffer(2)]],
    constant LinearParams & params [[buffer(3)]],
    uint3 group [[threadgroup_position_in_grid]],
    uint simd_lane [[thread_index_in_simdgroup]],
    uint simd_group [[simdgroup_index_in_threadgroup]]) {
    const uint row = group.y;
    const uint first_output = group.x * 4 + simd_group * 2;
    const uint blocks_per_row = params.in_features / QK8_0;
    float sums[2] = {0.0f, 0.0f};
    const ulong input_base = ulong(row) * params.in_features;
    for (uint column = simd_lane; column < params.in_features; column += 32) {
        const float activation = input[input_base + column];
        for (uint local_output = 0; local_output < 2; ++local_output) {
            const uint output_column = first_output + local_output;
            if (output_column < params.out_features) {
                const ulong weight_base = ulong(output_column) * blocks_per_row;
                sums[local_output] += activation * q8_0_value(
                    weight[weight_base + column / QK8_0], column % QK8_0
                );
            }
        }
    }
    for (uint local_output = 0; local_output < 2; ++local_output) {
        const uint output_column = first_output + local_output;
        const float total = simd_sum(sums[local_output]);
        if (simd_lane == 0 && output_column < params.out_features) {
            output[ulong(row) * params.output_stride
                + params.output_column_offset + output_column] = total;
        }
    }
}