Skip to content

Commit bf93446

Browse files
jerryz123jerryz123
committed
Merge remote-tracking branch 'origin/dev' into xcpt-handler
2 parents b67e6cf + 2802ca4 commit bf93446

File tree

1 file changed

+31
-31
lines changed

1 file changed

+31
-31
lines changed

‎include/gemmini.h‎

Lines changed: 31 additions & 31 deletions
Original file line numberDiff line numberDiff line change
@@ -919,7 +919,7 @@ void tiled_matmul(size_t dim_I, size_t dim_J, size_t dim_K,
919919
exit(1);
920920
}
921921

922-
char matmul_type_str[][3] = {"OS", "WS", "CPU"};
922+
char matmul_type_str[][4] = {"OS", "WS", "CPU"};
923923

924924
// Check if transpose options are correct
925925
if (((tiled_matmul_type == OS || tiled_matmul_type == CPU) && (transpose_A || transpose_B)) ||
@@ -948,7 +948,7 @@ void tiled_matmul(size_t dim_I, size_t dim_J, size_t dim_K,
948948
(int)tiled_matmul_type);
949949
} else /*if (tiled_matmul_type == CPU)*/ {
950950
matmul_cpu(dim_I, dim_J, dim_K,
951-
A, B, D, (elem_t*)C,
951+
A, B, (const acc_t*) D, (elem_t*)C,
952952
stride_A, stride_B, stride_D, stride_C,
953953
A_scale_factor, B_scale_factor, D_scale_factor,
954954
act, scale, relu6_shift, repeating_bias);
@@ -1321,11 +1321,11 @@ void sp_tiled_conv_ds(
13211321
int lpad, int rpad, int upad, int dpad,
13221322
int plpad, int prpad, int pupad, int pdpad,
13231323

1324-
elem_t * input,
1324+
const elem_t * input,
13251325
uint32_t B_sp_addr_outer,
1326-
elem_t * weights,
1326+
const elem_t * weights,
13271327
elem_t * output,
1328-
acc_t * bias,
1328+
const acc_t * bias,
13291329

13301330
int act, acc_scale_t scale, int relu6_shift,
13311331
bool no_bias, bool no_pool,
@@ -1403,7 +1403,7 @@ int bidims = batches*idims;
14031403

14041404
for (int b = 0; b < batches; b++) {
14051405
for (int irow = 0; irow < irows; irow++) {
1406-
elem_t * in = input + (b*in_dim*in_dim + irow*in_dim) * in_channels;// + ich;
1406+
const elem_t * in = input + (b*in_dim*in_dim + irow*in_dim) * in_channels;// + ich;
14071407
const uint32_t A_sp_addr = A_sp_addr_start + b * idims + irow * icols;
14081408
for (int ich = 0; ich < ichs; ich += DIM) {
14091409
// const int K = ichs - ich > DIM ? DIM : ichs - ich;
@@ -1870,10 +1870,10 @@ void sp_tiled_conv_ws_original(
18701870
int lpad, int rpad, int upad, int dpad,
18711871
int plpad, int prpad, int pupad, int pdpad,
18721872

1873-
elem_t * input,
1874-
elem_t * weights,
1873+
const elem_t * input,
1874+
const elem_t * weights,
18751875
elem_t * output,
1876-
acc_t * bias,
1876+
const acc_t * bias,
18771877

18781878
int act, int scale, int relu6_shift,
18791879
bool no_bias, bool no_pool,
@@ -1955,7 +1955,7 @@ void sp_tiled_conv_ws_original(
19551955

19561956
for (int icol = -lpad; icol < icols_unpadded + rpad;) {
19571957
int I = icols_unpadded - icol > DIM ? DIM : icols_unpadded - icol;
1958-
elem_t * in = input + (b*in_dim*in_dim + irow*in_dim + icol) * in_channels;// + ich;
1958+
const elem_t * in = input + (b*in_dim*in_dim + irow*in_dim + icol) * in_channels;// + ich;
19591959

19601960
if (icol < 0) {
19611961
I = -icol > DIM ? DIM : -icol;
@@ -2255,9 +2255,9 @@ void conv_cpu_without_pool(
22552255
int out_channels, int out_dim,
22562256
int stride, int padding, int kernel_dim,
22572257

2258-
elem_t * input,
2259-
elem_t * weights,
2260-
acc_t * bias,
2258+
const elem_t * input,
2259+
const elem_t * weights,
2260+
const acc_t * bias,
22612261
elem_t * output,
22622262

22632263
int act, acc_scale_t scale, size_t relu6_shift) {
@@ -2303,9 +2303,9 @@ void conv_cpu(
23032303
int out_channels, int out_dim,
23042304
int stride, int padding, int kernel_dim,
23052305

2306-
elem_t * input,
2307-
elem_t * weights,
2308-
acc_t * bias,
2306+
const elem_t * input,
2307+
const elem_t * weights,
2308+
const acc_t * bias,
23092309
elem_t * output,
23102310

23112311
int act, acc_scale_t scale, size_t relu6_shift,
@@ -2776,11 +2776,11 @@ void sp_tiled_conv_ws(
27762776
int lpad, int rpad, int upad, int dpad,
27772777
int plpad, int prpad, int pupad, int pdpad,
27782778

2779-
elem_t * input,
2779+
const elem_t * input,
27802780
uint32_t B_sp_addr_start,
27812781
//elem_t * weights,
27822782
elem_t * output,
2783-
acc_t * bias,
2783+
const acc_t * bias,
27842784

27852785
int act, acc_scale_t scale, int relu6_shift,
27862786
bool no_bias, bool no_pool,
@@ -2850,7 +2850,7 @@ void sp_tiled_conv_ws(
28502850

28512851
for (int icol = -lpad; icol < icols_unpadded + rpad;) {
28522852
int I = icols_unpadded - icol > DIM ? DIM : icols_unpadded - icol;
2853-
elem_t * in = input + (b*in_dim*in_dim + irow*in_dim + icol) * in_channels;// + ich;
2853+
const elem_t * in = input + (b*in_dim*in_dim + irow*in_dim + icol) * in_channels;// + ich;
28542854

28552855
if (icol < 0) {
28562856
I = -icol > DIM ? DIM : -icol;
@@ -3054,9 +3054,9 @@ void tiled_conv_original(
30543054
int porows, int pocols, int pochs,
30553055
int kcols, int kchs,
30563056

3057-
elem_t * input,
3058-
elem_t * weights,
3059-
acc_t * bias,
3057+
const elem_t * input,
3058+
const elem_t * weights,
3059+
const acc_t * bias,
30603060
elem_t * output,
30613061

30623062
int act, acc_scale_t scale, size_t relu6_shift,
@@ -3151,7 +3151,7 @@ void tiled_conv_original(
31513151
if (kch + kchs < in_channels) {
31523152
out = NULL;
31533153
}
3154-
acc_t * bias_ = bias + poch;
3154+
const acc_t * bias_ = bias + poch;
31553155
if (kch > 0) {
31563156
bias_ = NULL;
31573157
}
@@ -3245,9 +3245,9 @@ void tiled_conv(
32453245
int porows, int pocols, int pochs,
32463246
int krows, int kcols, int kchs,
32473247

3248-
elem_t * input,
3249-
elem_t * weights,
3250-
acc_t * bias,
3248+
const elem_t * input,
3249+
const elem_t * weights,
3250+
const acc_t * bias,
32513251
elem_t * output,
32523252

32533253
int act, acc_scale_t scale, size_t relu6_shift,
@@ -3321,7 +3321,7 @@ void tiled_conv(
33213321

33223322
for (int poch = 0; poch < out_channels; poch += pochs) {
33233323
const int pochs_ = out_channels - poch > pochs ? pochs : out_channels - poch;
3324-
acc_t * bias_ = bias + poch;
3324+
const acc_t * bias_ = bias + poch;
33253325

33263326
const int kchs_ = in_channels;
33273327
gemmini_config_ld(out_channels*sizeof(elem_t));
@@ -3332,7 +3332,7 @@ void tiled_conv(
33323332
for (int ich = 0; ich < kchs_; ich += DIM) {
33333333
const int K = kchs_ - ich > DIM ? DIM : kchs_ - ich;
33343334
for (int krow = 0; krow < kcols; krow++){
3335-
elem_t * weight = weights + poch + (krow*kcols*in_channels + ich) * out_channels + och;
3335+
const elem_t * weight = weights + poch + (krow*kcols*in_channels + ich) * out_channels + och;
33363336
const uint32_t B_sp_addr = B_sp_addr_start + (och / DIM) * kdims * kchs_ + ich*kdims + krow*kcols*K;// + kcol*K;
33373337
for (int kcol = 0; kcol < kcols; kcol++){
33383338
gemmini_extended_mvin(weight + kcol*in_channels*out_channels,
@@ -3890,9 +3890,9 @@ void tiled_conv_auto(
38903890
int out_channels, int out_dim,
38913891
int stride, int padding, int kernel_dim,
38923892

3893-
elem_t * input,
3894-
elem_t * weights,
3895-
acc_t * bias,
3893+
const elem_t * input,
3894+
const elem_t * weights,
3895+
const acc_t * bias,
38963896
elem_t * output,
38973897

38983898
int act, acc_scale_t scale, size_t relu6_shift,

0 commit comments

Comments
 (0)