Skip to content

Commit bf78f54

Browse files
0cc4m0cc4m
authored
vulkan: add ops docs (ggml-org#14900)
1 parent bbfc849 commit bf78f54

File tree

2 files changed

+8222
-89
lines changed

2 files changed

+8222
-89
lines changed

‎docs/ops.md‎

Lines changed: 89 additions & 89 deletions
Original file line numberDiff line numberDiff line change
@@ -12,92 +12,92 @@ Legend:
1212
- 🟡 Partially supported by this backend
1313
- ❌ Not supported by this backend
1414

15-
| Operation | BLAS | CPU | CUDA | Metal | SYCL |
16-
|-----------|------|------|------|------|------|
17-
| ABS | ❌ | ✅ | 🟡 | ❌ | 🟡 |
18-
| ACC | ❌ | ✅ | ✅ | ✅ | ✅ |
19-
| ADD | ❌ | ✅ | ✅ | 🟡 | ✅ |
20-
| ADD1 | ❌ | ✅ | ✅ | ❌ | ✅ |
21-
| ARANGE | ❌ | ✅ | ✅ | ✅ | ❌ |
22-
| ARGMAX | ❌ | ✅ | ✅ | ✅ | ✅ |
23-
| ARGSORT | ❌ | ✅ | ✅ | ✅ | ✅ |
24-
| CLAMP | ❌ | ✅ | ✅ | 🟡 | ✅ |
25-
| CONCAT | ❌ | ✅ | 🟡 | ✅ | 🟡 |
26-
| CONT | ❌ | ✅ | ✅ | ✅ | 🟡 |
27-
| CONV_2D | ❌ | ✅ | ❌ | ❌ | ❌ |
28-
| CONV_2D_DW | ❌ | ✅ | ✅ | ❌ | ❌ |
29-
| CONV_TRANSPOSE_1D | ❌ | ✅ | ✅ | ✅ | ✅ |
30-
| CONV_TRANSPOSE_2D | ❌ | ✅ | ✅ | ❌ | ❌ |
31-
| COS | ❌ | ✅ | ✅ | 🟡 | ✅ |
32-
| COUNT_EQUAL | ❌ | ✅ | ✅ | ❌ | ❌ |
33-
| CPY | ❌ | 🟡 | 🟡 | 🟡 | 🟡 |
34-
| CROSS_ENTROPY_LOSS | ❌ | ✅ | ✅ | ❌ | ❌ |
35-
| CROSS_ENTROPY_LOSS_BACK | ❌ | ✅ | ✅ | ❌ | ❌ |
36-
| DIAG_MASK_INF | ❌ | ✅ | ✅ | 🟡 | ✅ |
37-
| DIV | ❌ | ✅ | ✅ | 🟡 | ✅ |
38-
| DUP | ❌ | ✅ | 🟡 | 🟡 | ✅ |
39-
| ELU | ❌ | ✅ | 🟡 | 🟡 | 🟡 |
40-
| EXP | ❌ | ✅ | 🟡 | ❌ | 🟡 |
41-
| FLASH_ATTN_EXT | ❌ | ✅ | 🟡 | 🟡 | ❌ |
42-
| GATED_LINEAR_ATTN | ❌ | ✅ | ✅ | ❌ | ✅ |
43-
| GEGLU | ❌ | ✅ | ✅ | 🟡 | ✅ |
44-
| GEGLU_ERF | ❌ | ✅ | ✅ | 🟡 | ✅ |
45-
| GEGLU_QUICK | ❌ | ✅ | ✅ | 🟡 | ✅ |
46-
| GELU | ❌ | ✅ | 🟡 | 🟡 | 🟡 |
47-
| GELU_ERF | ❌ | ✅ | 🟡 | 🟡 | 🟡 |
48-
| GELU_QUICK | ❌ | ✅ | 🟡 | 🟡 | 🟡 |
49-
| GET_ROWS | ❌ | ✅ | 🟡 | ✅ | 🟡 |
50-
| GET_ROWS_BACK | ❌ | 🟡 | 🟡 | ❌ | ❌ |
51-
| GROUP_NORM | ❌ | ✅ | ✅ | ✅ | ✅ |
52-
| HARDSIGMOID | ❌ | ✅ | 🟡 | ❌ | 🟡 |
53-
| HARDSWISH | ❌ | ✅ | 🟡 | ❌ | 🟡 |
54-
| IM2COL | ❌ | ✅ | ✅ | 🟡 | ✅ |
55-
| L2_NORM | ❌ | ✅ | ✅ | ✅ | ✅ |
56-
| LEAKY_RELU | ❌ | ✅ | ✅ | ✅ | ✅ |
57-
| LOG | ❌ | ✅ | ✅ | ❌ | ✅ |
58-
| MEAN | ❌ | ✅ | ✅ | ✅ | ❌ |
59-
| MUL | ❌ | ✅ | ✅ | 🟡 | ✅ |
60-
| MUL_MAT | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 |
61-
| MUL_MAT_ID | ❌ | ✅ | ✅ | ✅ | 🟡 |
62-
| NEG | ❌ | ✅ | 🟡 | 🟡 | 🟡 |
63-
| NORM | ❌ | ✅ | ✅ | 🟡 | ✅ |
64-
| OPT_STEP_ADAMW | ❌ | ✅ | ✅ | ❌ | ❌ |
65-
| OUT_PROD | 🟡 | 🟡 | 🟡 | ❌ | 🟡 |
66-
| PAD | ❌ | ✅ | ✅ | ✅ | ✅ |
67-
| PAD_REFLECT_1D | ❌ | ✅ | ❌ | ✅ | ❌ |
68-
| POOL_2D | ❌ | ✅ | ✅ | ✅ | ✅ |
69-
| REGLU | ❌ | ✅ | ✅ | 🟡 | ✅ |
70-
| RELU | ❌ | ✅ | 🟡 | 🟡 | 🟡 |
71-
| REPEAT | ❌ | ✅ | 🟡 | ✅ | ✅ |
72-
| REPEAT_BACK | ❌ | ✅ | ✅ | ❌ | ❌ |
73-
| RMS_NORM | ❌ | ✅ | ✅ | 🟡 | ✅ |
74-
| RMS_NORM_BACK | ❌ | ✅ | ✅ | ❌ | ❌ |
75-
| RMS_NORM_MUL | ❌ | ❌ | ❌ | ✅ | ❌ |
76-
| RMS_NORM_MUL_ADD | ❌ | ✅ | ✅ | ❌ | ✅ |
77-
| ROLL | ❌ | ✅ | ❌ | ❌ | ❌ |
78-
| ROPE | ❌ | ✅ | ✅ | ✅ | ✅ |
79-
| ROPE_BACK | ❌ | ✅ | ✅ | ❌ | ❌ |
80-
| RWKV_WKV6 | ❌ | ✅ | ✅ | ✅ | ✅ |
81-
| RWKV_WKV7 | ❌ | ✅ | ✅ | ✅ | ✅ |
82-
| SCALE | ❌ | ✅ | ✅ | ✅ | ✅ |
83-
| SET | ❌ | ✅ | ❌ | ✅ | ❌ |
84-
| SET_ROWS | ❌ | 🟡 | 🟡 | 🟡 | 🟡 |
85-
| SGN | ❌ | ✅ | 🟡 | ❌ | 🟡 |
86-
| SIGMOID | ❌ | ✅ | 🟡 | 🟡 | 🟡 |
87-
| SILU | ❌ | ✅ | 🟡 | 🟡 | 🟡 |
88-
| SILU_BACK | ❌ | ✅ | ✅ | ❌ | ❌ |
89-
| SIN | ❌ | ✅ | ✅ | 🟡 | ✅ |
90-
| SOFT_MAX | ❌ | ✅ | ✅ | ✅ | 🟡 |
91-
| SOFT_MAX_BACK | ❌ | 🟡 | 🟡 | ❌ | ❌ |
92-
| SQR | ❌ | ✅ | ✅ | 🟡 | ✅ |
93-
| SQRT | ❌ | ✅ | ✅ | 🟡 | ✅ |
94-
| SSM_CONV | ❌ | ✅ | ✅ | ✅ | ❌ |
95-
| SSM_SCAN | ❌ | ✅ | ✅ | ✅ | ❌ |
96-
| STEP | ❌ | ✅ | 🟡 | ❌ | 🟡 |
97-
| SUB | ❌ | ✅ | ✅ | 🟡 | ✅ |
98-
| SUM | ❌ | ✅ | ✅ | ❌ | ✅ |
99-
| SUM_ROWS | ❌ | ✅ | ✅ | ✅ | ✅ |
100-
| SWIGLU | ❌ | ✅ | ✅ | 🟡 | ✅ |
101-
| TANH | ❌ | ✅ | 🟡 | 🟡 | 🟡 |
102-
| TIMESTEP_EMBEDDING | ❌ | ✅ | ✅ | ✅ | ✅ |
103-
| UPSCALE | ❌ | ✅ | ✅ | 🟡 | 🟡 |
15+
| Operation | BLAS | CPU | CUDA | Metal | SYCL | Vulkan |
16+
|-----------|------|------|------|------|------|------|
17+
| ABS | ❌ | ✅ | 🟡 | ❌ | 🟡 | ❌ |
18+
| ACC | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ |
19+
| ADD | ❌ | ✅ | ✅ | 🟡 | ✅ | ✅ |
20+
| ADD1 | ❌ | ✅ | ✅ | ❌ | ✅ | ❌ |
21+
| ARANGE | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ |
22+
| ARGMAX | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ |
23+
| ARGSORT | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ |
24+
| CLAMP | ❌ | ✅ | ✅ | 🟡 | ✅ | 🟡 |
25+
| CONCAT | ❌ | ✅ | 🟡 | ✅ | 🟡 | ✅ |
26+
| CONT | ❌ | ✅ | ✅ | ✅ | 🟡 | 🟡 |
27+
| CONV_2D | ❌ | ✅ | ❌ | ❌ | ❌ | ✅ |
28+
| CONV_2D_DW | ❌ | ✅ | ✅ | ❌ | ❌ | ✅ |
29+
| CONV_TRANSPOSE_1D | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ |
30+
| CONV_TRANSPOSE_2D | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
31+
| COS | ❌ | ✅ | ✅ | 🟡 | ✅ | 🟡 |
32+
| COUNT_EQUAL | ❌ | ✅ | ✅ | ❌ | ❌ | ✅ |
33+
| CPY | ❌ | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 |
34+
| CROSS_ENTROPY_LOSS | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
35+
| CROSS_ENTROPY_LOSS_BACK | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
36+
| DIAG_MASK_INF | ❌ | ✅ | ✅ | 🟡 | ✅ | ✅ |
37+
| DIV | ❌ | ✅ | ✅ | 🟡 | ✅ | ✅ |
38+
| DUP | ❌ | ✅ | 🟡 | 🟡 | ✅ | 🟡 |
39+
| ELU | ❌ | ✅ | 🟡 | 🟡 | 🟡 | ❌ |
40+
| EXP | ❌ | ✅ | 🟡 | ❌ | 🟡 | ❌ |
41+
| FLASH_ATTN_EXT | ❌ | ✅ | 🟡 | 🟡 | ❌ | 🟡 |
42+
| GATED_LINEAR_ATTN | ❌ | ✅ | ✅ | ❌ | ✅ | ❌ |
43+
| GEGLU | ❌ | ✅ | ✅ | 🟡 | ✅ | 🟡 |
44+
| GEGLU_ERF | ❌ | ✅ | ✅ | 🟡 | ✅ | 🟡 |
45+
| GEGLU_QUICK | ❌ | ✅ | ✅ | 🟡 | ✅ | 🟡 |
46+
| GELU | ❌ | ✅ | 🟡 | 🟡 | 🟡 | 🟡 |
47+
| GELU_ERF | ❌ | ✅ | 🟡 | 🟡 | 🟡 | 🟡 |
48+
| GELU_QUICK | ❌ | ✅ | 🟡 | 🟡 | 🟡 | 🟡 |
49+
| GET_ROWS | ❌ | ✅ | 🟡 | ✅ | 🟡 | 🟡 |
50+
| GET_ROWS_BACK | ❌ | 🟡 | 🟡 | ❌ | ❌ | ❌ |
51+
| GROUP_NORM | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ |
52+
| HARDSIGMOID | ❌ | ✅ | 🟡 | ❌ | 🟡 | ❌ |
53+
| HARDSWISH | ❌ | ✅ | 🟡 | ❌ | 🟡 | ❌ |
54+
| IM2COL | ❌ | ✅ | ✅ | 🟡 | ✅ | ✅ |
55+
| L2_NORM | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ |
56+
| LEAKY_RELU | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ |
57+
| LOG | ❌ | ✅ | ✅ | ❌ | ✅ | ❌ |
58+
| MEAN | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ |
59+
| MUL | ❌ | ✅ | ✅ | 🟡 | ✅ | ✅ |
60+
| MUL_MAT | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 |
61+
| MUL_MAT_ID | ❌ | ✅ | ✅ | ✅ | 🟡 | ✅ |
62+
| NEG | ❌ | ✅ | 🟡 | 🟡 | 🟡 | ❌ |
63+
| NORM | ❌ | ✅ | ✅ | 🟡 | ✅ | 🟡 |
64+
| OPT_STEP_ADAMW | ❌ | ✅ | ✅ | ❌ | ❌ | ✅ |
65+
| OUT_PROD | 🟡 | 🟡 | 🟡 | ❌ | 🟡 | ❌ |
66+
| PAD | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ |
67+
| PAD_REFLECT_1D | ❌ | ✅ | ❌ | ✅ | ❌ | ❌ |
68+
| POOL_2D | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ |
69+
| REGLU | ❌ | ✅ | ✅ | 🟡 | ✅ | 🟡 |
70+
| RELU | ❌ | ✅ | 🟡 | 🟡 | 🟡 | 🟡 |
71+
| REPEAT | ❌ | ✅ | 🟡 | ✅ | ✅ | 🟡 |
72+
| REPEAT_BACK | ❌ | ✅ | ✅ | ❌ | ❌ | ✅ |
73+
| RMS_NORM | ❌ | ✅ | ✅ | 🟡 | ✅ | ✅ |
74+
| RMS_NORM_BACK | ❌ | ✅ | ✅ | ❌ | ❌ | ✅ |
75+
| RMS_NORM_MUL | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ |
76+
| RMS_NORM_MUL_ADD | ❌ | ✅ | ✅ | ❌ | ✅ | ✅ |
77+
| ROLL | ❌ | ✅ | ❌ | ❌ | ❌ | ✅ |
78+
| ROPE | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ |
79+
| ROPE_BACK | ❌ | ✅ | ✅ | ❌ | ❌ | ✅ |
80+
| RWKV_WKV6 | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ |
81+
| RWKV_WKV7 | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ |
82+
| SCALE | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ |
83+
| SET | ❌ | ✅ | ❌ | ✅ | ❌ | ❌ |
84+
| SET_ROWS | ❌ | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 |
85+
| SGN | ❌ | ✅ | 🟡 | ❌ | 🟡 | ❌ |
86+
| SIGMOID | ❌ | ✅ | 🟡 | 🟡 | 🟡 | 🟡 |
87+
| SILU | ❌ | ✅ | 🟡 | 🟡 | 🟡 | 🟡 |
88+
| SILU_BACK | ❌ | ✅ | ✅ | ❌ | ❌ | ✅ |
89+
| SIN | ❌ | ✅ | ✅ | 🟡 | ✅ | 🟡 |
90+
| SOFT_MAX | ❌ | ✅ | ✅ | ✅ | 🟡 | ✅ |
91+
| SOFT_MAX_BACK | ❌ | 🟡 | 🟡 | ❌ | ❌ | ✅ |
92+
| SQR | ❌ | ✅ | ✅ | 🟡 | ✅ | 🟡 |
93+
| SQRT | ❌ | ✅ | ✅ | 🟡 | ✅ | ❌ |
94+
| SSM_CONV | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ |
95+
| SSM_SCAN | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ |
96+
| STEP | ❌ | ✅ | 🟡 | ❌ | 🟡 | ❌ |
97+
| SUB | ❌ | ✅ | ✅ | 🟡 | ✅ | ✅ |
98+
| SUM | ❌ | ✅ | ✅ | ❌ | ✅ | ✅ |
99+
| SUM_ROWS | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ |
100+
| SWIGLU | ❌ | ✅ | ✅ | 🟡 | ✅ | 🟡 |
101+
| TANH | ❌ | ✅ | 🟡 | 🟡 | 🟡 | 🟡 |
102+
| TIMESTEP_EMBEDDING | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ |
103+
| UPSCALE | ❌ | ✅ | ✅ | 🟡 | 🟡 | ✅ |

0 commit comments

Comments
 (0)