Skip to content

Commit

Permalink
drop requant pack4
Browse files Browse the repository at this point in the history
  • Loading branch information
@nihui
nihui committed Feb 7, 2025
1 parent 80177fe commit 3be03dd
Show file tree
Hide file tree
Showing 4 changed files with 0 additions and 171 deletions.
45 changes: 0 additions & 45 deletions src/layer/arm/requantize_arm.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -57,11 +57,6 @@ static void requantize_relu(const int* intptr, signed char* ptr, const Mat& scal
_scale_in0 = vld1q_f32((const float*)scale_in_data);
_scale_in1 = vld1q_f32((const float*)scale_in_data + 4);
}
if (elempack == 4)
{
_scale_in0 = vld1q_f32((const float*)scale_in_data);
_scale_in1 = _scale_in0;
}
}
#endif // __ARM_NEON

Expand All @@ -76,11 +71,6 @@ static void requantize_relu(const int* intptr, signed char* ptr, const Mat& scal
_scale_out0 = vld1q_f32((const float*)scale_out_data);
_scale_out1 = vld1q_f32((const float*)scale_out_data + 4);
}
if (elempack == 4)
{
_scale_out0 = vld1q_f32((const float*)scale_out_data);
_scale_out1 = _scale_out0;
}
}
#endif // __ARM_NEON

Expand Down Expand Up @@ -139,11 +129,6 @@ static void requantize_relu(const int* intptr, signed char* ptr, const Mat& scal
_bias0 = vld1q_f32((const float*)bias_data);
_bias1 = vld1q_f32((const float*)bias_data + 4);
}
if (elempack == 4)
{
_bias0 = vld1q_f32((const float*)bias_data);
_bias1 = _bias0;
}
}
#endif // __ARM_NEON

Expand Down Expand Up @@ -224,11 +209,6 @@ static void requantize_leakyrelu(const int* intptr, signed char* ptr, const Mat&
_scale_in0 = vld1q_f32((const float*)scale_in_data);
_scale_in1 = vld1q_f32((const float*)scale_in_data + 4);
}
if (elempack == 4)
{
_scale_in0 = vld1q_f32((const float*)scale_in_data);
_scale_in1 = _scale_in0;
}
}
#endif // __ARM_NEON

Expand All @@ -243,11 +223,6 @@ static void requantize_leakyrelu(const int* intptr, signed char* ptr, const Mat&
_scale_out0 = vld1q_f32((const float*)scale_out_data);
_scale_out1 = vld1q_f32((const float*)scale_out_data + 4);
}
if (elempack == 4)
{
_scale_out0 = vld1q_f32((const float*)scale_out_data);
_scale_out1 = _scale_out0;
}
}
#endif // __ARM_NEON

Expand Down Expand Up @@ -307,11 +282,6 @@ static void requantize_leakyrelu(const int* intptr, signed char* ptr, const Mat&
_bias0 = vld1q_f32((const float*)bias_data);
_bias1 = vld1q_f32((const float*)bias_data + 4);
}
if (elempack == 4)
{
_bias0 = vld1q_f32((const float*)bias_data);
_bias1 = _bias0;
}
}
#endif // __ARM_NEON

Expand Down Expand Up @@ -399,11 +369,6 @@ static void requantize(const int* intptr, signed char* ptr, const Mat& scale_in_
_scale_in0 = vld1q_f32((const float*)scale_in_data);
_scale_in1 = vld1q_f32((const float*)scale_in_data + 4);
}
if (elempack == 4)
{
_scale_in0 = vld1q_f32((const float*)scale_in_data);
_scale_in1 = _scale_in0;
}
}
#endif // __ARM_NEON

Expand All @@ -418,11 +383,6 @@ static void requantize(const int* intptr, signed char* ptr, const Mat& scale_in_
_scale_out0 = vld1q_f32((const float*)scale_out_data);
_scale_out1 = vld1q_f32((const float*)scale_out_data + 4);
}
if (elempack == 4)
{
_scale_out0 = vld1q_f32((const float*)scale_out_data);
_scale_out1 = _scale_out0;
}
}
#endif // __ARM_NEON

Expand Down Expand Up @@ -481,11 +441,6 @@ static void requantize(const int* intptr, signed char* ptr, const Mat& scale_in_
_bias0 = vld1q_f32((const float*)bias_data);
_bias1 = vld1q_f32((const float*)bias_data + 4);
}
if (elempack == 4)
{
_bias0 = vld1q_f32((const float*)bias_data);
_bias1 = _bias0;
}
}
#endif // __ARM_NEON

Expand Down
45 changes: 0 additions & 45 deletions src/layer/loongarch/requantize_loongarch.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -56,11 +56,6 @@ static void requantize_relu(const int* intptr, signed char* ptr, const Mat& scal
_scale_in0 = (__m128)__lsx_vld((const float*)scale_in_data, 0);
_scale_in1 = (__m128)__lsx_vld((const float*)scale_in_data + 4, 0);
}
if (elempack == 4)
{
_scale_in0 = (__m128)__lsx_vld((const float*)scale_in_data, 0);
_scale_in1 = _scale_in0;
}
}
#endif // __loongarch_sx

Expand All @@ -75,11 +70,6 @@ static void requantize_relu(const int* intptr, signed char* ptr, const Mat& scal
_scale_out0 = (__m128)__lsx_vld((const float*)scale_out_data, 0);
_scale_out1 = (__m128)__lsx_vld((const float*)scale_out_data + 4, 0);
}
if (elempack == 4)
{
_scale_out0 = (__m128)__lsx_vld((const float*)scale_out_data, 0);
_scale_out1 = _scale_out0;
}
}
#endif // __loongarch_sx

Expand Down Expand Up @@ -139,11 +129,6 @@ static void requantize_relu(const int* intptr, signed char* ptr, const Mat& scal
_bias0 = (__m128)__lsx_vld((const float*)bias_data, 0);
_bias1 = (__m128)__lsx_vld((const float*)bias_data + 4, 0);
}
if (elempack == 4)
{
_bias0 = (__m128)__lsx_vld((const float*)bias_data, 0);
_bias1 = _bias0;
}
}
#endif // __loongarch_sx

Expand Down Expand Up @@ -216,11 +201,6 @@ static void requantize_leakyrelu(const int* intptr, signed char* ptr, const Mat&
_scale_in0 = (__m128)__lsx_vld((const float*)scale_in_data, 0);
_scale_in1 = (__m128)__lsx_vld((const float*)scale_in_data + 4, 0);
}
if (elempack == 4)
{
_scale_in0 = (__m128)__lsx_vld((const float*)scale_in_data, 0);
_scale_in1 = _scale_in0;
}
}
#endif // __loongarch_sx

Expand All @@ -235,11 +215,6 @@ static void requantize_leakyrelu(const int* intptr, signed char* ptr, const Mat&
_scale_out0 = (__m128)__lsx_vld((const float*)scale_out_data, 0);
_scale_out1 = (__m128)__lsx_vld((const float*)scale_out_data + 4, 0);
}
if (elempack == 4)
{
_scale_out0 = (__m128)__lsx_vld((const float*)scale_out_data, 0);
_scale_out1 = _scale_out0;
}
}
#endif // __loongarch_sx

Expand Down Expand Up @@ -300,11 +275,6 @@ static void requantize_leakyrelu(const int* intptr, signed char* ptr, const Mat&
_bias0 = (__m128)__lsx_vld((const float*)bias_data, 0);
_bias1 = (__m128)__lsx_vld((const float*)bias_data + 4, 0);
}
if (elempack == 4)
{
_bias0 = (__m128)__lsx_vld((const float*)bias_data, 0);
_bias1 = _bias0;
}
}
#endif // __loongarch_sx

Expand Down Expand Up @@ -384,11 +354,6 @@ static void requantize(const int* intptr, signed char* ptr, const Mat& scale_in_
_scale_in0 = (__m128)__lsx_vld((const float*)scale_in_data, 0);
_scale_in1 = (__m128)__lsx_vld((const float*)scale_in_data + 4, 0);
}
if (elempack == 4)
{
_scale_in0 = (__m128)__lsx_vld((const float*)scale_in_data, 0);
_scale_in1 = _scale_in0;
}
}
#endif // __loongarch_sx

Expand All @@ -403,11 +368,6 @@ static void requantize(const int* intptr, signed char* ptr, const Mat& scale_in_
_scale_out0 = (__m128)__lsx_vld((const float*)scale_out_data, 0);
_scale_out1 = (__m128)__lsx_vld((const float*)scale_out_data + 4, 0);
}
if (elempack == 4)
{
_scale_out0 = (__m128)__lsx_vld((const float*)scale_out_data, 0);
_scale_out1 = _scale_out0;
}
}
#endif // __loongarch_sx

Expand Down Expand Up @@ -467,11 +427,6 @@ static void requantize(const int* intptr, signed char* ptr, const Mat& scale_in_
_bias0 = (__m128)__lsx_vld((const float*)bias_data, 0);
_bias1 = (__m128)__lsx_vld((const float*)bias_data + 4, 0);
}
if (elempack == 4)
{
_bias0 = (__m128)__lsx_vld((const float*)bias_data, 0);
_bias1 = _bias0;
}
}
#endif // __loongarch_sx

Expand Down
45 changes: 0 additions & 45 deletions src/layer/mips/requantize_mips.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -56,11 +56,6 @@ static void requantize_relu(const int* intptr, signed char* ptr, const Mat& scal
_scale_in0 = (v4f32)__msa_ld_w((const float*)scale_in_data, 0);
_scale_in1 = (v4f32)__msa_ld_w((const float*)scale_in_data + 4, 0);
}
if (elempack == 4)
{
_scale_in0 = (v4f32)__msa_ld_w((const float*)scale_in_data, 0);
_scale_in1 = _scale_in0;
}
}
#endif // __mips_msa

Expand All @@ -75,11 +70,6 @@ static void requantize_relu(const int* intptr, signed char* ptr, const Mat& scal
_scale_out0 = (v4f32)__msa_ld_w((const float*)scale_out_data, 0);
_scale_out1 = (v4f32)__msa_ld_w((const float*)scale_out_data + 4, 0);
}
if (elempack == 4)
{
_scale_out0 = (v4f32)__msa_ld_w((const float*)scale_out_data, 0);
_scale_out1 = _scale_out0;
}
}
#endif // __mips_msa

Expand Down Expand Up @@ -139,11 +129,6 @@ static void requantize_relu(const int* intptr, signed char* ptr, const Mat& scal
_bias0 = (v4f32)__msa_ld_w((const float*)bias_data, 0);
_bias1 = (v4f32)__msa_ld_w((const float*)bias_data + 4, 0);
}
if (elempack == 4)
{
_bias0 = (v4f32)__msa_ld_w((const float*)bias_data, 0);
_bias1 = _bias0;
}
}
#endif // __mips_msa

Expand Down Expand Up @@ -216,11 +201,6 @@ static void requantize_leakyrelu(const int* intptr, signed char* ptr, const Mat&
_scale_in0 = (v4f32)__msa_ld_w((const float*)scale_in_data, 0);
_scale_in1 = (v4f32)__msa_ld_w((const float*)scale_in_data + 4, 0);
}
if (elempack == 4)
{
_scale_in0 = (v4f32)__msa_ld_w((const float*)scale_in_data, 0);
_scale_in1 = _scale_in0;
}
}
#endif // __mips_msa

Expand All @@ -235,11 +215,6 @@ static void requantize_leakyrelu(const int* intptr, signed char* ptr, const Mat&
_scale_out0 = (v4f32)__msa_ld_w((const float*)scale_out_data, 0);
_scale_out1 = (v4f32)__msa_ld_w((const float*)scale_out_data + 4, 0);
}
if (elempack == 4)
{
_scale_out0 = (v4f32)__msa_ld_w((const float*)scale_out_data, 0);
_scale_out1 = _scale_out0;
}
}
#endif // __mips_msa

Expand Down Expand Up @@ -300,11 +275,6 @@ static void requantize_leakyrelu(const int* intptr, signed char* ptr, const Mat&
_bias0 = (v4f32)__msa_ld_w((const float*)bias_data, 0);
_bias1 = (v4f32)__msa_ld_w((const float*)bias_data + 4, 0);
}
if (elempack == 4)
{
_bias0 = (v4f32)__msa_ld_w((const float*)bias_data, 0);
_bias1 = _bias0;
}
}
#endif // __mips_msa

Expand Down Expand Up @@ -384,11 +354,6 @@ static void requantize(const int* intptr, signed char* ptr, const Mat& scale_in_
_scale_in0 = (v4f32)__msa_ld_w((const float*)scale_in_data, 0);
_scale_in1 = (v4f32)__msa_ld_w((const float*)scale_in_data + 4, 0);
}
if (elempack == 4)
{
_scale_in0 = (v4f32)__msa_ld_w((const float*)scale_in_data, 0);
_scale_in1 = _scale_in0;
}
}
#endif // __mips_msa

Expand All @@ -403,11 +368,6 @@ static void requantize(const int* intptr, signed char* ptr, const Mat& scale_in_
_scale_out0 = (v4f32)__msa_ld_w((const float*)scale_out_data, 0);
_scale_out1 = (v4f32)__msa_ld_w((const float*)scale_out_data + 4, 0);
}
if (elempack == 4)
{
_scale_out0 = (v4f32)__msa_ld_w((const float*)scale_out_data, 0);
_scale_out1 = _scale_out0;
}
}
#endif // __mips_msa

Expand Down Expand Up @@ -467,11 +427,6 @@ static void requantize(const int* intptr, signed char* ptr, const Mat& scale_in_
_bias0 = (v4f32)__msa_ld_w((const float*)bias_data, 0);
_bias1 = (v4f32)__msa_ld_w((const float*)bias_data + 4, 0);
}
if (elempack == 4)
{
_bias0 = (v4f32)__msa_ld_w((const float*)bias_data, 0);
_bias1 = _bias0;
}
}
#endif // __mips_msa

Expand Down
36 changes: 0 additions & 36 deletions src/layer/x86/requantize_x86.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -71,18 +71,6 @@ static void requantize(const int* intptr, signed char* ptr, const Mat& scale_in_
#else // __AVX__
_scale_in0 = _mm_loadu_ps((const float*)scale_in_data);
_scale_in1 = _mm_loadu_ps((const float*)scale_in_data + 4);
#endif // __AVX__
}
if (elempack == 4)
{
_scale_in0 = _mm_loadu_ps((const float*)scale_in_data);
#if __AVX__
_scale_in_avx = combine4x2_ps(_scale_in0, _scale_in0);
#if __AVX512F__
_scale_in_avx512 = combine8x2_ps(_scale_in_avx, _scale_in_avx);
#endif // __AVX512F__
#else // __AVX__
_scale_in1 = _scale_in0;
#endif // __AVX__
}
}
Expand Down Expand Up @@ -117,18 +105,6 @@ static void requantize(const int* intptr, signed char* ptr, const Mat& scale_in_
#else // __AVX__
_scale_out0 = _mm_loadu_ps((const float*)scale_out_data);
_scale_out1 = _mm_loadu_ps((const float*)scale_out_data + 4);
#endif // __AVX__
}
if (elempack == 4)
{
_scale_out0 = _mm_loadu_ps((const float*)scale_out_data);
#if __AVX__
_scale_out_avx = combine4x2_ps(_scale_out0, _scale_out0);
#if __AVX512F__
_scale_out_avx512 = combine8x2_ps(_scale_out_avx, _scale_out_avx);
#endif // __AVX512F__
#else // __AVX__
_scale_out1 = _scale_out0;
#endif // __AVX__
}
}
Expand Down Expand Up @@ -239,18 +215,6 @@ static void requantize(const int* intptr, signed char* ptr, const Mat& scale_in_
#else // __AVX__
_bias0 = _mm_loadu_ps((const float*)bias_data);
_bias1 = _mm_loadu_ps((const float*)bias_data + 4);
#endif // __AVX__
}
if (elempack == 4)
{
_bias0 = _mm_loadu_ps((const float*)bias_data);
#if __AVX__
_bias_avx = combine4x2_ps(_bias0, _bias0);
#if __AVX512F__
_bias_avx512 = combine8x2_ps(_bias_avx, _bias_avx);
#endif // __AVX512F__
#else // __AVX__
_bias1 = _bias0;
#endif // __AVX__
}
}
Expand Down

0 comments on commit 3be03dd

Please sign in to comment.