Skip to content

Commit c4e8479

Browse files
leo-ponyleo-pony
committed
Delete the commented code
1 parent cd9a324 commit c4e8479

File tree

2 files changed

+0
-71
lines changed

2 files changed

+0
-71
lines changed

‎ggml/src/ggml-cann/kernels/get_row_f16.cpp‎

Lines changed: 0 additions & 26 deletions
Original file line numberDiff line numberDiff line change
@@ -65,18 +65,7 @@ class GET_ROW_F16 {
6565
size_t tail = len % elem_per_block;
6666
len = len & ~(elem_per_block - 1);
6767
if(tail != 0) {
68-
//printf("f16 get_row: copy_in: offset:%d, len:%d, origin_len:%d, tail:%d, elem_per_block:%d.\n", offset, len, origin_len, tail, elem_per_block);
69-
//DumpTensor(input_local, 5, elem_per_block);
7068
len += elem_per_block;
71-
// DataCopy(input_local[len], input_gm[offset + len], elem_per_block);
72-
#if 0
73-
DataCopyExtParams dataCopyParams;
74-
dataCopyParams.blockCount = 1;
75-
dataCopyParams.blockLen = tail * sizeof(half);
76-
DataCopyPadExtParams<half> padParams;
77-
DataCopyPad(input_local[len], input_gm[offset + len],
78-
dataCopyParams, padParams);
79-
#endif
8069
}
8170
DataCopy(input_local, input_gm[offset], len);
8271
input_queue.EnQue(input_local);
@@ -90,30 +79,15 @@ class GET_ROW_F16 {
9079
if (len > 0) {
9180
DataCopy(output_gm[offset], output_local, len);
9281
}
93-
#if 1
9482
if(tail != 0) {
95-
/* printf("\nf16 BLOCK_IDX:%d get_row: Copy_Out AtomicAdd: offset:%d, len:%d, tail:%d, elem_per_block:%d.\n", op_block_idx, offset, len, tail, elem_per_block);
96-
DumpTensor(output_gm, 5, elem_per_block); */
9783
for (size_t i = tail; i < elem_per_block; i++) {
9884
output_local[len + i].SetValue(0, 0);
9985
}
100-
// DumpTensor(output_local[len], 5, elem_per_block);
10186
SetAtomicAdd<float>();
10287
DataCopy(output_gm[offset + len], output_local[len], elem_per_block);
10388
SetAtomicNone();
104-
// DumpTensor(output_gm, 5, elem_per_block);
10589
}
106-
#endif
10790

108-
#if 0
109-
if(tail != 0) {
110-
DataCopyExtParams dataCopyParams;
111-
dataCopyParams.blockCount = 1;
112-
dataCopyParams.blockLen = tail * sizeof(float);
113-
DataCopyPad(output_gm[offset + len], output_local[len],
114-
dataCopyParams);
115-
}
116-
#endif
11791
output_queue.FreeTensor(output_local);
11892
}
11993

‎ggml/src/ggml-cann/kernels/get_row_f32.cpp‎

Lines changed: 0 additions & 45 deletions
Original file line numberDiff line numberDiff line change
@@ -51,47 +51,18 @@ class GET_ROW_F32 {
5151
// All data should asign to 32. It's ok because all data is align to 32.
5252
pipe.InitBuffer(input_queue, BUFFER_NUM, local_buffer_size);
5353
pipe.InitBuffer(output_queue, BUFFER_NUM, local_buffer_size);
54-
// printf("f32 BLOCK_IDX:%d get_row: init: ir:%d, dr:%d, n_elements:%d.\n", op_block_idx, ir, dr, n_elements);
5554
}
5655

5756
__aicore__ inline void copy_in(uint32_t offset, size_t len) {
58-
size_t origin_len = len;
5957
LocalTensor<float> input_local = input_queue.AllocTensor<float>();
6058
const size_t elem_per_block = 32 / sizeof(float);
6159
size_t tail = len % elem_per_block;
6260
len = len & ~(elem_per_block - 1);
6361

64-
//printf("f32 BLOCK_IDX:%d get_row: Copy_in: offset:%d, len:%d, origin_len:%d, tail:%d, elem_per_block:%d.\n", op_block_idx, offset, len, origin_len, tail, elem_per_block);
6562
if (len > 0)
6663
DataCopy(input_local, input_gm[offset], len);
67-
//printf("f32 BLOCK_IDX:%d get_row: Copy_in executed: offset:%d, len:%d, origin_len:%d, tail:%d, elem_per_block:%d.\n", op_block_idx, offset, len, origin_len, tail, elem_per_block);
6864
if(tail != 0) {
69-
#if 1
70-
/* //printf("f32 BLOCK_IDX:%d get_row: Copy_in ENTER tail != 0: offset:%d, len:%d, origin_len:%d, tail:%d, elem_per_block:%d.\n", op_block_idx, offset, len, origin_len, tail, elem_per_block);
71-
for (int i = 0; i < elem_per_block; i++) {
72-
printf("f32 BLOCK_IDX:%d get_row: Copy_in: get value idx:%d, origin input local val:%f.\n", op_block_idx, i, input_local[len + i].GetValue(0));
73-
}
74-
//DumpTensor(input_gm[offset + len], 5, elem_per_block);
75-
for (int i = 0; i < tail; i++) {
76-
printf("f32 BLOCK_IDX:%d get_row: Copy_in: get value idx:%d, input local val:%f, input_gm:%f.\n", op_block_idx, len + i, input_local[len + i].GetValue(0), input_gm[offset + len + i]);
77-
} */
7865
DataCopy(input_local[len], input_gm[offset + len], elem_per_block);
79-
// clean
80-
/* for (int i = tail; i < elem_per_block; i++) {
81-
input_local[len + i].SetValue(0, 0);
82-
}
83-
for (int i = 0; i < elem_per_block; i++) {
84-
printf("f32 BLOCK_IDX:%d get_row: Copy_in: get value idx:%d, after clean and copy, input local val:%f.\n", op_block_idx, i, input_local[len + i].GetValue(0));
85-
} */
86-
#endif
87-
#if 0
88-
DataCopyExtParams dataCopyParams;
89-
dataCopyParams.blockCount = 1;
90-
dataCopyParams.blockLen = tail * sizeof(float);
91-
DataCopyPadExtParams<float> padParams;
92-
DataCopyPad(input_local[len], input_gm[offset + len],
93-
dataCopyParams, padParams);
94-
#endif
9566
}
9667
input_queue.EnQue(input_local);
9768
}
@@ -104,31 +75,15 @@ class GET_ROW_F32 {
10475
if (len > 0) {
10576
DataCopy(output_gm[offset], output_local, len);
10677
}
107-
108-
#if 1
10978
if(tail != 0) {
11079
for (size_t i = tail; i < elem_per_block; i++) {
11180
output_local[len + i].SetValue(0, 0);
11281
}
113-
//printf("\nf32 BLOCK_IDX:%d get_row: Copy_Out AtomicAdd: offset:%d, len:%d, tail:%d, elem_per_block:%d.\n", op_block_idx, offset, len, tail, elem_per_block);
114-
/* DumpTensor(output_gm[offset + len], 5, elem_per_block);
115-
DumpTensor(output_local[len], 5, elem_per_block); */
11682
SetAtomicAdd<float>();
11783
DataCopy(output_gm[offset + len], output_local[len], elem_per_block);
11884
SetAtomicNone();
119-
/* DumpTensor(output_gm[offset + len], 5, elem_per_block); */
12085
}
121-
#endif
122-
#if 0
123-
if(tail != 0) {
12486

125-
DataCopyExtParams dataCopyParams;
126-
dataCopyParams.blockCount = 1;
127-
dataCopyParams.blockLen = tail * sizeof(float);
128-
DataCopyPad(output_gm[offset + len], output_local[len],
129-
dataCopyParams);
130-
}
131-
#endif
13287
output_queue.FreeTensor(output_local);
13388
}
13489

0 commit comments

Comments
 (0)