Skip to content

Commit

Permalink
w
Browse files Browse the repository at this point in the history
  • Loading branch information
@nihui
nihui committed Jan 25, 2025
1 parent 1cf015d commit 07d6fd0
Showing 1 changed file with 18 additions and 18 deletions.
36 changes: 18 additions & 18 deletions src/layer/loongarch/requantize_loongarch.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -96,8 +96,8 @@ static void requantize_relu(const int* intptr, signed char* ptr, const Mat& scal
for (; i + 7 < size; i += 8)
{
__builtin_prefetch(intptr + 32);
__m128 _v0 = (__m128)__msa_ffint_s_w(__lsx_vld(intptr, 0));
__m128 _v1 = (__m128)__msa_ffint_s_w(__lsx_vld(intptr + 4, 0));
__m128 _v0 = __lsx_vffint_s_w(__lsx_vld(intptr, 0));
__m128 _v1 = __lsx_vffint_s_w(__lsx_vld(intptr + 4, 0));
_v0 = __lsx_vfmul_s(_v0, _scale0);
_v1 = __lsx_vfmul_s(_v1, _scale1);
*((int64_t*)ptr) = float2int8relu(_v0, _v1);
Expand All @@ -106,7 +106,7 @@ static void requantize_relu(const int* intptr, signed char* ptr, const Mat& scal
}
for (; i + 3 < size; i += 4)
{
__m128 _v = (__m128)__msa_ffint_s_w(__lsx_vld(intptr, 0));
__m128 _v = __lsx_vffint_s_w(__lsx_vld(intptr, 0));
_v = __lsx_vfmul_s(_v, _scale0);
v16i8 v = (v16i8)float2int8relu(_v, _v);
ptr[0] = v[0];
Expand Down Expand Up @@ -158,8 +158,8 @@ static void requantize_relu(const int* intptr, signed char* ptr, const Mat& scal
for (; i + 7 < size; i += 8)
{
__builtin_prefetch(intptr + 32);
__m128 _v0 = (__m128)__msa_ffint_s_w(__lsx_vld(intptr, 0));
__m128 _v1 = (__m128)__msa_ffint_s_w(__lsx_vld(intptr + 4, 0));
__m128 _v0 = __lsx_vffint_s_w(__lsx_vld(intptr, 0));
__m128 _v1 = __lsx_vffint_s_w(__lsx_vld(intptr + 4, 0));
_v0 = __lsx_vfmadd_s(_v0, _scale0, _bias0);
_v1 = __lsx_vfmadd_s(_v1, _scale1, _bias1);
*((int64_t*)ptr) = float2int8relu(_v0, _v1);
Expand All @@ -168,7 +168,7 @@ static void requantize_relu(const int* intptr, signed char* ptr, const Mat& scal
}
for (; i + 3 < size; i += 4)
{
__m128 _v = (__m128)__msa_ffint_s_w(__lsx_vld(intptr, 0));
__m128 _v = __lsx_vffint_s_w(__lsx_vld(intptr, 0));
_v = __lsx_vfmadd_s(_v, _scale0, _bias0);
v16i8 v = (v16i8)float2int8relu(_v, _v);
ptr[0] = v[0];
Expand Down Expand Up @@ -257,8 +257,8 @@ static void requantize_leakyrelu(const int* intptr, signed char* ptr, const Mat&
for (; i + 7 < size; i += 8)
{
__builtin_prefetch(intptr + 32);
__m128 _v0 = (__m128)__msa_ffint_s_w(__lsx_vld(intptr, 0));
__m128 _v1 = (__m128)__msa_ffint_s_w(__lsx_vld(intptr + 4, 0));
__m128 _v0 = __lsx_vffint_s_w(__lsx_vld(intptr, 0));
__m128 _v1 = __lsx_vffint_s_w(__lsx_vld(intptr + 4, 0));
_v0 = __lsx_vfmul_s(_v0, _scale0);
_v1 = __lsx_vfmul_s(_v1, _scale1);
*((int64_t*)ptr) = float2int8leakyrelu(_v0, _v1, _slope);
Expand All @@ -267,7 +267,7 @@ static void requantize_leakyrelu(const int* intptr, signed char* ptr, const Mat&
}
for (; i + 3 < size; i += 4)
{
__m128 _v = (__m128)__msa_ffint_s_w(__lsx_vld(intptr, 0));
__m128 _v = __lsx_vffint_s_w(__lsx_vld(intptr, 0));
_v = __lsx_vfmul_s(_v, _scale0);
v16i8 v = (v16i8)float2int8leakyrelu(_v, _v, _slope);
ptr[0] = v[0];
Expand Down Expand Up @@ -319,8 +319,8 @@ static void requantize_leakyrelu(const int* intptr, signed char* ptr, const Mat&
for (; i + 7 < size; i += 8)
{
__builtin_prefetch(intptr + 32);
__m128 _v0 = (__m128)__msa_ffint_s_w(__lsx_vld(intptr, 0));
__m128 _v1 = (__m128)__msa_ffint_s_w(__lsx_vld(intptr + 4, 0));
__m128 _v0 = __lsx_vffint_s_w(__lsx_vld(intptr, 0));
__m128 _v1 = __lsx_vffint_s_w(__lsx_vld(intptr + 4, 0));
_v0 = __lsx_vfmadd_s(_v0, _scale0, _bias0);
_v1 = __lsx_vfmadd_s(_v1, _scale1, _bias1);
*((int64_t*)ptr) = float2int8leakyrelu(_v0, _v1, _slope);
Expand All @@ -329,7 +329,7 @@ static void requantize_leakyrelu(const int* intptr, signed char* ptr, const Mat&
}
for (; i + 3 < size; i += 4)
{
__m128 _v = (__m128)__msa_ffint_s_w(__lsx_vld(intptr, 0));
__m128 _v = __lsx_vffint_s_w(__lsx_vld(intptr, 0));
_v = __lsx_vfmadd_s(_v, _scale0, _bias0);
v16i8 v = (v16i8)float2int8leakyrelu(_v, _v, _slope);
ptr[0] = v[0];
Expand Down Expand Up @@ -418,8 +418,8 @@ static void requantize(const int* intptr, signed char* ptr, const Mat& scale_in_
for (; i + 7 < size; i += 8)
{
__builtin_prefetch(intptr + 32);
__m128 _v0 = (__m128)__msa_ffint_s_w(__lsx_vld(intptr, 0));
__m128 _v1 = (__m128)__msa_ffint_s_w(__lsx_vld(intptr + 4, 0));
__m128 _v0 = __lsx_vffint_s_w(__lsx_vld(intptr, 0));
__m128 _v1 = __lsx_vffint_s_w(__lsx_vld(intptr + 4, 0));
_v0 = __lsx_vfmul_s(_v0, _scale_in0);
_v1 = __lsx_vfmul_s(_v1, _scale_in1);
_v0 = activation_ps(_v0, activation_type, activation_params);
Expand All @@ -432,7 +432,7 @@ static void requantize(const int* intptr, signed char* ptr, const Mat& scale_in_
}
for (; i + 3 < size; i += 4)
{
__m128 _v = (__m128)__msa_ffint_s_w(__lsx_vld(intptr, 0));
__m128 _v = __lsx_vffint_s_w(__lsx_vld(intptr, 0));
_v = __lsx_vfmul_s(_v, _scale_in0);
_v = activation_ps(_v, activation_type, activation_params);
_v = __lsx_vfmul_s(_v, _scale_out0);
Expand Down Expand Up @@ -480,8 +480,8 @@ static void requantize(const int* intptr, signed char* ptr, const Mat& scale_in_
for (; i + 7 < size; i += 8)
{
__builtin_prefetch(intptr + 32);
__m128 _v0 = (__m128)__msa_ffint_s_w(__lsx_vld(intptr, 0));
__m128 _v1 = (__m128)__msa_ffint_s_w(__lsx_vld(intptr + 4, 0));
__m128 _v0 = __lsx_vffint_s_w(__lsx_vld(intptr, 0));
__m128 _v1 = __lsx_vffint_s_w(__lsx_vld(intptr + 4, 0));
_v0 = __lsx_vfmadd_s(_v0, _scale_in0, _bias0);
_v1 = __lsx_vfmadd_s(_v1, _scale_in1, _bias1);
_v0 = activation_ps(_v0, activation_type, activation_params);
Expand All @@ -494,7 +494,7 @@ static void requantize(const int* intptr, signed char* ptr, const Mat& scale_in_
}
for (; i + 3 < size; i += 4)
{
__m128 _v = (__m128)__msa_ffint_s_w(__lsx_vld(intptr, 0));
__m128 _v = __lsx_vffint_s_w(__lsx_vld(intptr, 0));
_v = __lsx_vfmadd_s(_v, _scale_in0, _bias0);
_v = activation_ps(_v, activation_type, activation_params);
_v = __lsx_vfmul_s(_v, _scale_out0);
Expand Down

0 comments on commit 07d6fd0

Please sign in to comment.