Skip to content

Fix/fix convert float to fp16 warning #6525

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
nihui merged 7 commits into from
Feb 3, 2026
+160 −160
Merged

Fix/fix convert float to fp16 warning #6525

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
cc37862
fix:small part of convolution_winograd_transform_packn_fp16s.h
bluemiao3 Jan 23, 2026
725dcc8
fix: no converting warning in self-test
bluemiao3 Jan 23, 2026
fc73c6f
fix: Fix the remaining warnings
bluemiao3 Jan 26, 2026
5732487
Fix: solve compile problem
bluemiao3 Jan 28, 2026
24a8db3
Fix: solve compile problem
bluemiao3 Jan 28, 2026
e53d0db
fix remaining issues
bluemiao3 Feb 2, 2026
2857988
fix remaining issues
bluemiao3 Feb 3, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion src/layer/riscv/bias_riscv_zfh.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,7 +25,7 @@ int Bias_riscv::forward_inplace_fp16s(Mat& bottom_top_blob, const Option& opt) c
for (int q = 0; q < channels; q++)
{
__fp16* ptr = bottom_top_blob.channel(q);
__fp16 bias = bias_data[q];
__fp16 bias = (__fp16)bias_data[q];

#if __riscv_zvfh
int n = size;
Expand Down
8 changes: 4 additions & 4 deletions src/layer/riscv/celu_riscv_zfh.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,12 +32,12 @@ int CELU_riscv::forward_inplace_fp16s(Mat& bottom_top_blob, const Option& opt) c
size_t vl = __riscv_vsetvl_e16m8(n);

vfloat16m8_t _p = __riscv_vle16_v_f16m8(ptr, vl);
vbool2_t _mask = __riscv_vmfgt_vf_f16m8_b2(_p, 0.f, vl);
vbool2_t _mask = __riscv_vmfgt_vf_f16m8_b2(_p, (__fp16)0.f, vl);

vfloat16m8_t _q = __riscv_vfdiv_vf_f16m8(_p, alpha, vl);
vfloat16m8_t _q = __riscv_vfdiv_vf_f16m8(_p, (__fp16)alpha, vl);
_q = exp_ps(_q, vl);
_q = __riscv_vfsub_vf_f16m8(_q, 1.f, vl);
_q = __riscv_vfmul_vf_f16m8(_q, alpha, vl);
_q = __riscv_vfsub_vf_f16m8(_q, (__fp16)1.f, vl);
_q = __riscv_vfmul_vf_f16m8(_q, (__fp16)alpha, vl);

vfloat16m8_t _res = __riscv_vmerge_vvm_f16m8(_q, _p, _mask, vl);
__riscv_vse16_v_f16m8(ptr, _res, vl);
Expand Down
14 changes: 7 additions & 7 deletions src/layer/riscv/convolution1d_riscv_zfh.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -143,7 +143,7 @@ int Convolution1D_riscv::forward_fp16s(const Mat& bottom_blob, Mat& top_blob, co
{
float val = (float)*slptr++;
vfloat16m1_t _w0 = __riscv_vle16_v_f16m1(kptr, vl);
_sum = __riscv_vfwmacc_vf_f32m2(_sum, val, _w0, vl);
_sum = __riscv_vfwmacc_vf_f32m2(_sum, (__fp16)val, _w0, vl);

kptr += packn;
}
Expand Down Expand Up @@ -186,7 +186,7 @@ int Convolution1D_riscv::forward_fp16s(const Mat& bottom_blob, Mat& top_blob, co
{
float val = (float)sptr[0];
vfloat16m1_t _w = __riscv_vle16_v_f16m1(kptr, vl);
_sum = __riscv_vfwmacc_vf_f32m2(_sum, val, _w, vl);
_sum = __riscv_vfwmacc_vf_f32m2(_sum, (__fp16)val, _w, vl);

sptr += dilation_w;
kptr += packn;
Expand Down Expand Up @@ -353,7 +353,7 @@ int Convolution1D_riscv::forward_fp16sa(const Mat& bottom_blob, Mat& top_blob, c

for (int j = 0; j < outw; j++)
{
vfloat16m1_t _sum = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);

if (bias_term)
{
Expand Down Expand Up @@ -400,7 +400,7 @@ int Convolution1D_riscv::forward_fp16sa(const Mat& bottom_blob, Mat& top_blob, c

for (int j = 0; j < outw; j++)
{
vfloat16m1_t _sum = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);

if (bias_term)
{
Expand Down Expand Up @@ -443,14 +443,14 @@ int Convolution1D_riscv::forward_fp16sa(const Mat& bottom_blob, Mat& top_blob, c

for (int j = 0; j < outw; j++)
{
__fp16 sum = 0.f;
__fp16 sum = (__fp16)0.f;

if (bias_term)
{
sum = ((const __fp16*)bias_data_fp16)[p];
}

vfloat16m1_t _sum = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);

const __fp16* kptr = weight_data_fp16.channel(p);

Expand All @@ -471,7 +471,7 @@ int Convolution1D_riscv::forward_fp16sa(const Mat& bottom_blob, Mat& top_blob, c

sum = __riscv_vfmv_f_s_f16m1_f16(__riscv_vfredusum_vs_f16m1_f16m1(_sum, __riscv_vfmv_s_f_f16m1(sum, vl), vl));

sum = activation_ss(sum, activation_type, activation_params);
sum = (__fp16)activation_ss(sum, activation_type, activation_params);

outptr[j] = sum;
}
Expand Down
4 changes: 2 additions & 2 deletions src/layer/riscv/convolution_3x3_pack1ton_fp16s.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,7 +18,7 @@ static void conv3x3s1_pack1ton_fp16sa_rvv(const Mat& bottom_blob, Mat& top_blob,
{
Mat out0 = top_blob.channel(p);

vfloat16m1_t _bias0 = bias ? __riscv_vle16_v_f16m1(bias + p * packn, vl) : __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _bias0 = bias ? __riscv_vle16_v_f16m1(bias + p * packn, vl) : __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);
out0.fill(_bias0);

const __fp16* k0 = kernel.channel(p);
Expand Down Expand Up @@ -296,7 +296,7 @@ static void conv3x3s2_pack1ton_fp16sa_rvv(const Mat& bottom_blob, Mat& top_blob,
{
Mat out0 = top_blob.channel(p);

vfloat16m1_t _bias0 = bias ? __riscv_vle16_v_f16m1(bias + p * packn, vl) : __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _bias0 = bias ? __riscv_vle16_v_f16m1(bias + p * packn, vl) : __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);
out0.fill(_bias0);

const __fp16* k0 = kernel.channel(p);
Expand Down
2 changes: 1 addition & 1 deletion src/layer/riscv/convolution_7x7_pack1ton_fp16s.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,7 +22,7 @@ static void conv7x7s2_pack1ton_fp16sa_rvv(const Mat& bottom_blob, Mat& top_blob,
{
Mat out0 = top_blob.channel(p);

vfloat16m1_t _bias0 = bias ? __riscv_vle16_v_f16m1(bias + p * packn, vl) : __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _bias0 = bias ? __riscv_vle16_v_f16m1(bias + p * packn, vl) : __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);
out0.fill(_bias0);

for (int q = 0; q < inch; q++)
Expand Down
4 changes: 2 additions & 2 deletions src/layer/riscv/convolution_pack1ton_fp16s.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -65,7 +65,7 @@ static void convolution_pack1ton_fp16s_rvv(const Mat& bottom_blob, Mat& top_blob
{
float val = (float)sptr[space_ofs[k]];
vfloat16m1_t _w = __riscv_vle16_v_f16m1(kptr, vl);
_sum = __riscv_vfwmacc_vf_f32m2(_sum, val, _w, vl);
_sum = __riscv_vfwmacc_vf_f32m2(_sum, (__fp16)val, _w, vl);

kptr += packn;
}
Expand Down Expand Up @@ -126,7 +126,7 @@ static void convolution_pack1ton_fp16sa_rvv(const Mat& bottom_blob, Mat& top_blo
{
for (int j = 0; j < outw; j++)
{
vfloat16m1_t _sum = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);

if (bias_data_ptr)
{
Expand Down
4 changes: 2 additions & 2 deletions src/layer/riscv/convolution_packn_fp16s.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -71,7 +71,7 @@ static void convolution_packn_fp16s_rvv(const Mat& bottom_blob, Mat& top_blob, c
vfloat16m1_t _w0 = __riscv_vle16_v_f16m1(kptr, vl);
// _sum = __riscv_vfwmacc_vf_f32m2(_sum, val, _w0, vl);

vfloat32m2_t _qwq = __riscv_vfwmul_vf_f32m2(_w0, val, vl);
vfloat32m2_t _qwq = __riscv_vfwmul_vf_f32m2(_w0, (__fp16)val, vl);
_sum = __riscv_vfadd_vv_f32m2(_sum, _qwq, vl);

kptr += packn;
Expand Down Expand Up @@ -134,7 +134,7 @@ static void convolution_packn_fp16sa_rvv(const Mat& bottom_blob, Mat& top_blob,
{
for (int j = 0; j < outw; j++)
{
vfloat16m1_t _sum = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);

if (bias_data_ptr)
{
Expand Down
8 changes: 4 additions & 4 deletions src/layer/riscv/convolution_packnto1_fp16s.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -85,7 +85,7 @@ static void convolution_packnto1_fp16s_rvv(const Mat& bottom_blob, Mat& top_blob
sum = __riscv_vfmv_f_s_f32m1_f32(__riscv_vfredusum_vs_f32m2_f32m1(_sum, __riscv_vfmv_s_f_f32m1(sum, vl), vl));
#endif

sum = activation_ss(sum, activation_type, activation_params);
sum = (__fp16)activation_ss(sum, activation_type, activation_params);

outptr[j] = (__fp16)sum;
}
Expand Down Expand Up @@ -140,14 +140,14 @@ static void convolution_packnto1_fp16sa_rvv(const Mat& bottom_blob, Mat& top_blo
{
for (int j = 0; j < outw; j++)
{
__fp16 sum = 0.f;
__fp16 sum = (__fp16)0.f;

if (bias_data_ptr)
{
sum = bias_data_ptr[p];
}

vfloat16m1_t _sum = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);

const __fp16* kptr = weight_data_fp16.channel(p);

Expand All @@ -169,7 +169,7 @@ static void convolution_packnto1_fp16sa_rvv(const Mat& bottom_blob, Mat& top_blo

sum = __riscv_vfmv_f_s_f16m1_f16(__riscv_vfredusum_vs_f16m1_f16m1(_sum, __riscv_vfmv_s_f_f16m1(sum, vl), vl));

sum = activation_ss(sum, activation_type, activation_params);
sum = (__fp16)activation_ss(sum, activation_type, activation_params);

outptr[j] = sum;
}
Expand Down
8 changes: 4 additions & 4 deletions src/layer/riscv/convolution_sgemm_fp16s.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -109,7 +109,7 @@ static void im2col_sgemm_fp16sa_rvv(const Mat& bottom_im2col, Mat& top_blob, con
__fp16* outptr6 = top_blob.channel(p + 6);
__fp16* outptr7 = top_blob.channel(p + 7);

const __fp16 zeros[8] = {0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f};
const __fp16 zeros[8] = {(__fp16)0.f, (__fp16)0.f, (__fp16)0.f, (__fp16)0.f, (__fp16)0.f, (__fp16)0.f, (__fp16)0.f, (__fp16)0.f};
const __fp16* biasptr = bias ? bias + p : zeros;

int i = 0;
Expand Down Expand Up @@ -224,7 +224,7 @@ static void im2col_sgemm_fp16sa_rvv(const Mat& bottom_im2col, Mat& top_blob, con
__fp16* outptr2 = top_blob.channel(p + 2);
__fp16* outptr3 = top_blob.channel(p + 3);

const __fp16 zeros[4] = {0.f, 0.f, 0.f, 0.f};
const __fp16 zeros[4] = {(__fp16)0.f, (__fp16)0.f, (__fp16)0.f, (__fp16)0.f};
const __fp16* biasptr = bias ? bias + p : zeros;

int i = 0;
Expand Down Expand Up @@ -302,7 +302,7 @@ static void im2col_sgemm_fp16sa_rvv(const Mat& bottom_im2col, Mat& top_blob, con
{
__fp16* outptr0 = top_blob.channel(p);

const __fp16 bias0 = bias ? bias[p] : 0.f;
const __fp16 bias0 = bias ? (__fp16)bias[p] : (__fp16)0.f;

int i = 0;
for (; i + (packn - 1) < size; i += packn)
Expand Down Expand Up @@ -352,7 +352,7 @@ static void im2col_sgemm_fp16sa_rvv(const Mat& bottom_im2col, Mat& top_blob, con
{
__fp16* outptr0 = top_blob.channel(p);

const __fp16 bias0 = bias ? bias[p] : 0.f;
const __fp16 bias0 = bias ? (__fp16)bias[p] : (__fp16)0.f;

for (int i = 0; i < size; i++)
{
Expand Down
2 changes: 1 addition & 1 deletion src/layer/riscv/convolution_sgemm_pack1ton_fp16s.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -52,7 +52,7 @@ static void im2col_sgemm_pack1ton_fp16sa_rvv(const Mat& bottom_im2col, Mat& top_

int nn = inch * maxk; // inch always > 0

vfloat16m1_t _sum = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);

if (bias)
{
Expand Down
30 changes: 15 additions & 15 deletions src/layer/riscv/convolution_sgemm_packn_fp16s.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -262,14 +262,14 @@ static void im2col_sgemm_packn_fp16sa_rvv(const Mat& bottom_im2col, Mat& top_blo

int nn = inch * maxk * packn; // inch always > 0

vfloat16m1_t _sum0 = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum1 = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum2 = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum3 = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum4 = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum5 = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum6 = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum7 = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum0 = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);
vfloat16m1_t _sum1 = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);
vfloat16m1_t _sum2 = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);
vfloat16m1_t _sum3 = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);
vfloat16m1_t _sum4 = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);
vfloat16m1_t _sum5 = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);
vfloat16m1_t _sum6 = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);
vfloat16m1_t _sum7 = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);

if (bias)
{
Expand Down Expand Up @@ -324,10 +324,10 @@ static void im2col_sgemm_packn_fp16sa_rvv(const Mat& bottom_im2col, Mat& top_blo

int nn = inch * maxk * packn; // inch always > 0

vfloat16m1_t _sum0 = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum1 = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum2 = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum3 = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum0 = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);
vfloat16m1_t _sum1 = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);
vfloat16m1_t _sum2 = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);
vfloat16m1_t _sum3 = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);

if (bias)
{
Expand Down Expand Up @@ -366,8 +366,8 @@ static void im2col_sgemm_packn_fp16sa_rvv(const Mat& bottom_im2col, Mat& top_blo

int nn = inch * maxk * packn; // inch always > 0

vfloat16m1_t _sum0 = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum1 = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum0 = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);
vfloat16m1_t _sum1 = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);

if (bias)
{
Expand Down Expand Up @@ -398,7 +398,7 @@ static void im2col_sgemm_packn_fp16sa_rvv(const Mat& bottom_im2col, Mat& top_blo

int nn = inch * maxk * packn; // inch always > 0

vfloat16m1_t _sum = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);

if (bias)
{
Expand Down
34 changes: 17 additions & 17 deletions src/layer/riscv/convolution_sgemm_packnto1_fp16s.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -197,7 +197,7 @@ static void im2col_sgemm_packnto1_fp16sa_rvv(const Mat& bottom_im2col, Mat& top_
#ifdef __clang__
const __fp16* zeros = _zero_tmp;
#else
const __fp16 zeros[packn] = {0.f};
const __fp16 zeros[packn] = {(__fp16)0.f};
#endif // __clang__
const __fp16* biasptr = bias ? bias + p : zeros;

Expand Down Expand Up @@ -353,7 +353,7 @@ static void im2col_sgemm_packnto1_fp16sa_rvv(const Mat& bottom_im2col, Mat& top_
{
__fp16* outptr0 = top_blob.channel(p);

const __fp16 bias0 = bias ? bias[p] : 0.f;
const __fp16 bias0 = bias ? (__fp16)bias[p] : (__fp16)0.f;

int i = 0;
for (; i + 7 < size; i += 8)
Expand All @@ -372,14 +372,14 @@ static void im2col_sgemm_packnto1_fp16sa_rvv(const Mat& bottom_im2col, Mat& top_
__fp16 sum6 = bias0;
__fp16 sum7 = bias0;

vfloat16m1_t _sum0 = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum1 = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum2 = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum3 = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum4 = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum5 = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum6 = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum7 = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum0 = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);
vfloat16m1_t _sum1 = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);
vfloat16m1_t _sum2 = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);
vfloat16m1_t _sum3 = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);
vfloat16m1_t _sum4 = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);
vfloat16m1_t _sum5 = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);
vfloat16m1_t _sum6 = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);
vfloat16m1_t _sum7 = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);

for (int j = 0; j < nn; j++)
{
Expand Down Expand Up @@ -460,10 +460,10 @@ static void im2col_sgemm_packnto1_fp16sa_rvv(const Mat& bottom_im2col, Mat& top_
__fp16 sum2 = bias0;
__fp16 sum3 = bias0;

vfloat16m1_t _sum0 = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum1 = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum2 = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum3 = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum0 = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);
vfloat16m1_t _sum1 = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);
vfloat16m1_t _sum2 = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);
vfloat16m1_t _sum3 = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);

for (int j = 0; j < nn; j++)
{
Expand Down Expand Up @@ -518,8 +518,8 @@ static void im2col_sgemm_packnto1_fp16sa_rvv(const Mat& bottom_im2col, Mat& top_
__fp16 sum0 = bias0;
__fp16 sum1 = bias0;

vfloat16m1_t _sum0 = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum1 = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum0 = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);
vfloat16m1_t _sum1 = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);

for (int j = 0; j < nn; j++)
{
Expand Down Expand Up @@ -561,7 +561,7 @@ static void im2col_sgemm_packnto1_fp16sa_rvv(const Mat& bottom_im2col, Mat& top_

__fp16 sum0 = bias0;

vfloat16m1_t _sum0 = __riscv_vfmv_v_f_f16m1(0.f, vl);
vfloat16m1_t _sum0 = __riscv_vfmv_v_f_f16m1((__fp16)0.f, vl);

for (int j = 0; j < nn; j++)
{
Expand Down
Loading
Loading