Skip to content

Commit

Permalink
w--
Browse files Browse the repository at this point in the history
  • Loading branch information
@nihui
nihui committed Dec 11, 2024
1 parent b38c5bf commit 383cb5f
Showing 1 changed file with 49 additions and 49 deletions.
98 changes: 49 additions & 49 deletions src/layer/x86/gemm_int8.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -4464,9 +4464,9 @@ static void transpose_pack_A_tile_fp32_to_int8(const Mat& A, Mat& AT, int i, int
{
const float* p0 = (const float*)A + k * A_hstep + (i + ii) * elempack;

#if __AVX512VNNI__ || __AVXVNNI__
#if __AVX512VNNI__
__m128i _v127 = _mm_set1_epi8(127);
#endif // __AVX512VNNI__ || __AVXVNNI__
#endif // __AVX512VNNI__

const float scale = scales[i + ii];

Expand Down Expand Up @@ -6952,9 +6952,9 @@ static void transpose_pack_B_tile_fp32_to_int8(const Mat& B, Mat& BT, int j, int
{
const float* p0 = (const float*)B + k * B_hstep + (j + jj) * elempack;

#if __AVX512VNNI__ || __AVXVNNI__
#if __AVX512VNNI__
__m128i _v127 = _mm_set1_epi8(127);
#endif // __AVX512VNNI__ || __AVXVNNI__
#endif // __AVX512VNNI__

#if __SSE2__
#if __AVX__
Expand Down Expand Up @@ -7094,7 +7094,7 @@ static void unpack_output_tile_int32_to_fp32(const Mat& topT, const Mat& C, Mat&

__m512 _descale = _mm512_loadu_ps((const float*)descales + i + ii);

__m512 _c0;
__m512 _c0 = _mm512_set1_ps(0.f);
if (pC)
{
if (broadcast_type_C == 0)
Expand Down Expand Up @@ -7357,7 +7357,7 @@ static void unpack_output_tile_int32_to_fp32(const Mat& topT, const Mat& C, Mat&
_cf = _mm512_loadu_ps(pC + 128 + 112);
pC += 256;
}
if (c_elempack == 8)
else if (c_elempack == 8)
{
__m512 _tmp0 = _mm512_loadu_ps(pC);
__m512 _tmp1 = _mm512_loadu_ps(pC + 16);
Expand Down Expand Up @@ -7395,7 +7395,7 @@ static void unpack_output_tile_int32_to_fp32(const Mat& topT, const Mat& C, Mat&

pC += 128;
}
if (c_elempack == 4)
else if (c_elempack == 4)
{
_c0 = _mm512_loadu_ps(pC);
_c1 = _mm512_loadu_ps(pC + 16);
Expand Down Expand Up @@ -7450,7 +7450,7 @@ static void unpack_output_tile_int32_to_fp32(const Mat& topT, const Mat& C, Mat&

pC += 64;
}
if (c_elempack == 1)
else // if (c_elempack == 1)
{
_c0 = _mm512_loadu_ps(pC);
_c1 = _mm512_loadu_ps(pC + c_hstep);
Expand Down Expand Up @@ -7938,7 +7938,7 @@ static void unpack_output_tile_int32_to_fp32(const Mat& topT, const Mat& C, Mat&
_c7 = _mm512_loadu_ps(pC + 112);
pC += 128;
}
if (c_elempack == 8)
else if (c_elempack == 8)
{
__m512 _tmp0 = _mm512_loadu_ps(pC);
__m512 _tmp1 = _mm512_loadu_ps(pC + 16);
Expand All @@ -7960,7 +7960,7 @@ static void unpack_output_tile_int32_to_fp32(const Mat& topT, const Mat& C, Mat&

pC += 64;
}
if (c_elempack == 4)
else if (c_elempack == 4)
{
_c0 = _mm512_loadu_ps(pC);
_c1 = _mm512_loadu_ps(pC + 16);
Expand Down Expand Up @@ -7991,7 +7991,7 @@ static void unpack_output_tile_int32_to_fp32(const Mat& topT, const Mat& C, Mat&

pC += 32;
}
if (c_elempack == 1)
else // if (c_elempack == 1)
{
__m256 _cc0 = _mm256_loadu_ps(pC);
__m256 _cc1 = _mm256_loadu_ps(pC + c_hstep);
Expand Down Expand Up @@ -8278,7 +8278,7 @@ static void unpack_output_tile_int32_to_fp32(const Mat& topT, const Mat& C, Mat&
_c3 = _mm512_loadu_ps(pC + 48);
pC += 64;
}
if (c_elempack == 8)
else if (c_elempack == 8)
{
__m512 _cc0 = _mm512_loadu_ps(pC);
__m512 _cc1 = _mm512_loadu_ps(pC + 16);
Expand All @@ -8290,7 +8290,7 @@ static void unpack_output_tile_int32_to_fp32(const Mat& topT, const Mat& C, Mat&
_c3 = _mm512_shuffle_f32x4(_cc1, _cc3, _MM_SHUFFLE(3, 2, 3, 2));
pC += 32;
}
if (c_elempack == 4)
else if (c_elempack == 4)
{
_c0 = _mm512_loadu_ps(pC);
_c1 = _mm512_loadu_ps(pC + c_hstep * 4);
Expand All @@ -8306,7 +8306,7 @@ static void unpack_output_tile_int32_to_fp32(const Mat& topT, const Mat& C, Mat&
_c3 = _mm512_shuffle_f32x4(_tmp2, _tmp3, _MM_SHUFFLE(3, 1, 3, 1));
pC += 16;
}
if (c_elempack == 1)
else // if (c_elempack == 1)
{
__m128 _cc0 = _mm_loadu_ps(pC);
__m128 _cc1 = _mm_loadu_ps(pC + c_hstep);
Expand Down Expand Up @@ -8562,15 +8562,15 @@ static void unpack_output_tile_int32_to_fp32(const Mat& topT, const Mat& C, Mat&
_c1 = _mm512_loadu_ps(pC + 16);
pC += 32;
}
if (c_elempack == 8)
else if (c_elempack == 8)
{
__m512 _cc0 = _mm512_loadu_ps(pC);
__m512 _cc1 = _mm512_loadu_ps(pC + c_hstep * 8);
_c0 = _mm512_shuffle_f32x4(_cc0, _cc1, _MM_SHUFFLE(1, 0, 1, 0));
_c1 = _mm512_shuffle_f32x4(_cc0, _cc1, _MM_SHUFFLE(3, 2, 3, 2));
pC += 16;
}
if (c_elempack == 4)
else if (c_elempack == 4)
{
__m128 _cc0 = _mm_loadu_ps(pC);
__m128 _cc1 = _mm_loadu_ps(pC + 4);
Expand All @@ -8588,7 +8588,7 @@ static void unpack_output_tile_int32_to_fp32(const Mat& topT, const Mat& C, Mat&
_c1 = _mm512_insertf32x8(_mm512_castps256_ps512(_cc13), _cc57, 1);
pC += 8;
}
if (c_elempack == 1)
else // if (c_elempack == 1)
{
__m512i _vindex = _mm512_mullo_epi32(_mm512_setr_epi32(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15), _mm512_set1_epi32(c_hstep));
_c0 = _mm512_i32gather_ps(_vindex, pC, sizeof(float));
Expand Down Expand Up @@ -8691,14 +8691,14 @@ static void unpack_output_tile_int32_to_fp32(const Mat& topT, const Mat& C, Mat&
_c0 = _mm512_loadu_ps(pC);
pC += 16;
}
if (c_elempack == 8)
else if (c_elempack == 8)
{
__m256 _cc0 = _mm256_loadu_ps(pC);
__m256 _cc1 = _mm256_loadu_ps(pC + c_hstep * 8);
_c0 = _mm512_insertf32x8(_mm512_castps256_ps512(_cc0), _cc1, 1);
pC += 8;
}
if (c_elempack == 4)
else if (c_elempack == 4)
{
__m128 _cc0 = _mm_loadu_ps(pC);
__m128 _cc1 = _mm_loadu_ps(pC + c_hstep * 4);
Expand All @@ -8709,7 +8709,7 @@ static void unpack_output_tile_int32_to_fp32(const Mat& topT, const Mat& C, Mat&
_c0 = _mm512_insertf32x8(_mm512_castps256_ps512(_cc01), _cc23, 1);
pC += 4;
}
if (c_elempack == 1)
else // if (c_elempack == 1)
{
__m512i _vindex = _mm512_mullo_epi32(_mm512_setr_epi32(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15), _mm512_set1_epi32(c_hstep));
_c0 = _mm512_i32gather_ps(_vindex, pC, sizeof(float));
Expand Down Expand Up @@ -8783,9 +8783,9 @@ static void unpack_output_tile_int32_to_fp32(const Mat& topT, const Mat& C, Mat&
__m512 _descale_avx512 = _mm512_broadcast_f32x8(_descale);
#endif

__m256 _c0;
__m256 _c0 = _mm256_set1_ps(0.f);
#if __AVX512F__
__m512 _c0_avx512;
__m512 _c0_avx512 = _mm512_set1_ps(0.f);
#endif
if (pC)
{
Expand Down Expand Up @@ -8970,7 +8970,7 @@ static void unpack_output_tile_int32_to_fp32(const Mat& topT, const Mat& C, Mat&

pC += 128;
}
if (c_elempack == 4)
else if (c_elempack == 4)
{
_c0_avx512 = _mm512_loadu_ps(pC);
_c1_avx512 = _mm512_loadu_ps(pC + 16);
Expand Down Expand Up @@ -9010,7 +9010,7 @@ static void unpack_output_tile_int32_to_fp32(const Mat& topT, const Mat& C, Mat&

pC += 64;
}
if (c_elempack == 1)
else // if (c_elempack == 1)
{
_c0_avx512 = _mm512_loadu_ps(pC);
_c1_avx512 = _mm512_loadu_ps(pC + c_hstep);
Expand Down Expand Up @@ -9506,7 +9506,7 @@ static void unpack_output_tile_int32_to_fp32(const Mat& topT, const Mat& C, Mat&
_c7 = _mm256_loadu_ps(pC + 56);
pC += 64;
}
if (c_elempack == 4)
else if (c_elempack == 4)
{
__m256 _tmp0 = _mm256_loadu_ps(pC);
__m256 _tmp1 = _mm256_loadu_ps(pC + 8);
Expand All @@ -9526,7 +9526,7 @@ static void unpack_output_tile_int32_to_fp32(const Mat& topT, const Mat& C, Mat&
_c7 = _mm256_permute2f128_ps(_tmp3, _tmp7, _MM_SHUFFLE(0, 3, 0, 1));
pC += 32;
}
if (c_elempack == 1)
else // if (c_elempack == 1)
{
_c0 = _mm256_loadu_ps(pC);
_c1 = _mm256_loadu_ps(pC + c_hstep);
Expand Down Expand Up @@ -9772,7 +9772,7 @@ static void unpack_output_tile_int32_to_fp32(const Mat& topT, const Mat& C, Mat&
_c3 = _mm256_loadu_ps(pC + 24);
pC += 32;
}
if (c_elempack == 4)
else if (c_elempack == 4)
{
__m256 _cc0 = _mm256_loadu_ps(pC);
__m256 _cc1 = _mm256_loadu_ps(pC + 8);
Expand All @@ -9784,7 +9784,7 @@ static void unpack_output_tile_int32_to_fp32(const Mat& topT, const Mat& C, Mat&
_c3 = _mm256_permute2f128_ps(_cc1, _cc3, _MM_SHUFFLE(0, 3, 0, 1));
pC += 16;
}
if (c_elempack == 1)
else // if (c_elempack == 1)
{
// __m256i _vindex = _mm256_setr_epi32(0, 1, 2, 3, 4, 5, 6, 7);
// _c0 = _mm256_i32gather_ps(pC, _vindex, c_hstep * sizeof(float));
Expand Down Expand Up @@ -9994,15 +9994,15 @@ static void unpack_output_tile_int32_to_fp32(const Mat& topT, const Mat& C, Mat&
_c1 = _mm256_loadu_ps(pC + 8);
pC += 16;
}
if (c_elempack == 4)
else if (c_elempack == 4)
{
__m256 _cc0 = _mm256_loadu_ps(pC);
__m256 _cc1 = _mm256_loadu_ps(pC + c_hstep * 4);
_c0 = _mm256_permute2f128_ps(_cc0, _cc1, _MM_SHUFFLE(0, 2, 0, 0));
_c1 = _mm256_permute2f128_ps(_cc0, _cc1, _MM_SHUFFLE(0, 3, 0, 1));
pC += 8;
}
if (c_elempack == 1)
else // if (c_elempack == 1)
{
#if __AVX2__
__m256i _vindex = _mm256_mullo_epi32(_mm256_setr_epi32(0, 1, 2, 3, 4, 5, 6, 7), _mm256_set1_epi32(c_hstep));
Expand Down Expand Up @@ -10126,14 +10126,14 @@ static void unpack_output_tile_int32_to_fp32(const Mat& topT, const Mat& C, Mat&
_c0 = _mm256_loadu_ps(pC);
pC += 8;
}
if (c_elempack == 4)
else if (c_elempack == 4)
{
__m128 _cc0 = _mm_loadu_ps(pC);
__m128 _cc1 = _mm_loadu_ps(pC + c_hstep * 4);
_c0 = _mm256_insertf128_ps(_mm256_castps128_ps256(_cc0), _cc1, 1);
pC += 4;
}
if (c_elempack == 1)
else // if (c_elempack == 1)
{
#if __AVX2__
__m256i _vindex = _mm256_mullo_epi32(_mm256_setr_epi32(0, 1, 2, 3, 4, 5, 6, 7), _mm256_set1_epi32(c_hstep));
Expand Down Expand Up @@ -10213,9 +10213,9 @@ static void unpack_output_tile_int32_to_fp32(const Mat& topT, const Mat& C, Mat&
__m512 _descale_avx512 = _mm512_broadcast_f32x4(_descale);
#endif

__m128 _c0;
__m128 _c0 = _mm_set1_ps(0.f);
#if __AVX512F__
__m512 _c0_avx512;
__m512 _c0_avx512 = _mm512_set1_ps(0.f);
#endif
if (pC)
{
Expand Down Expand Up @@ -10323,7 +10323,7 @@ static void unpack_output_tile_int32_to_fp32(const Mat& topT, const Mat& C, Mat&
_c2_avx512 = _mm512_shuffle_f32x4(_tmp2, _tmp3, _MM_SHUFFLE(2, 0, 2, 0));
_c3_avx512 = _mm512_shuffle_f32x4(_tmp2, _tmp3, _MM_SHUFFLE(3, 1, 3, 1));
}
if (c_elempack == 1)
else // if (c_elempack == 1)
{
_c0_avx512 = _mm512_loadu_ps(pC);
_c1_avx512 = _mm512_loadu_ps(pC + c_hstep);
Expand Down Expand Up @@ -10596,7 +10596,7 @@ static void unpack_output_tile_int32_to_fp32(const Mat& topT, const Mat& C, Mat&
_c2 = _mm_loadu_ps(pC + 8);
_c3 = _mm_loadu_ps(pC + 12);
}
if (c_elempack == 1)
else // if (c_elempack == 1)
{
_c0 = _mm_loadu_ps(pC);
_c1 = _mm_loadu_ps(pC + c_hstep);
Expand Down Expand Up @@ -10627,7 +10627,7 @@ static void unpack_output_tile_int32_to_fp32(const Mat& topT, const Mat& C, Mat&
_c3 = _mm_loadu_ps(pC + 28);
pC += 32;
}
if (c_elempack == 1)
else // if (c_elempack == 1)
{
_c0 = _mm_loadu_ps(pC + 4);
_c1 = _mm_loadu_ps(pC + c_hstep + 4);
Expand Down Expand Up @@ -10832,7 +10832,7 @@ static void unpack_output_tile_int32_to_fp32(const Mat& topT, const Mat& C, Mat&
_c3 = _mm_loadu_ps(pC + 12);
pC += 16;
}
if (c_elempack == 1)
else // if (c_elempack == 1)
{
_c0 = _mm_loadu_ps(pC);
_c1 = _mm_loadu_ps(pC + c_hstep);
Expand Down Expand Up @@ -10992,7 +10992,7 @@ static void unpack_output_tile_int32_to_fp32(const Mat& topT, const Mat& C, Mat&
_c1 = _mm_loadu_ps(pC + 4);
pC += 8;
}
if (c_elempack == 1)
else // if (c_elempack == 1)
{
_c0 = _mm_setr_ps(pC[0], pC[c_hstep], pC[c_hstep * 2], pC[c_hstep * 3]);
_c1 = _mm_setr_ps(pC[1], pC[c_hstep + 1], pC[c_hstep * 2 + 1], pC[c_hstep * 3 + 1]);
Expand Down Expand Up @@ -11084,7 +11084,7 @@ static void unpack_output_tile_int32_to_fp32(const Mat& topT, const Mat& C, Mat&
_c0 = _mm_loadu_ps(pC);
pC += 4;
}
if (c_elempack == 1)
else // if (c_elempack == 1)
{
_c0 = _mm_setr_ps(pC[0], pC[c_hstep], pC[c_hstep * 2], pC[c_hstep * 3]);
pC += 1;
Expand Down Expand Up @@ -11153,14 +11153,14 @@ static void unpack_output_tile_int32_to_fp32(const Mat& topT, const Mat& C, Mat&
#endif // __AVX512F__
#endif

float c0;
float c1;
float c0 = 0.f;
float c1 = 0.f;
#if __SSE2__
__m128 _c0;
__m128 _c1;
__m128 _c0 = _mm_set1_ps(0.f);
__m128 _c1 = _mm_set1_ps(0.f);
#if __AVX512F__
__m512 _c0_avx512;
__m512 _c1_avx512;
__m512 _c0_avx512 = _mm512_set1_ps(0.f);
__m512 _c1_avx512 = _mm512_set1_ps(0.f);
#endif // __AVX512F__
#endif
if (pC)
Expand Down Expand Up @@ -11726,11 +11726,11 @@ static void unpack_output_tile_int32_to_fp32(const Mat& topT, const Mat& C, Mat&
#endif // __AVX512F__
#endif

float c0;
float c0 = 0.f;
#if __SSE2__
__m128 _c0;
__m128 _c0 = _mm_set1_ps(0.f);
#if __AVX512F__
__m512 _c0_avx512;
__m512 _c0_avx512 = _mm512_set1_ps(0.f);
#endif // __AVX512F__
#endif
if (pC)
Expand Down

0 comments on commit 383cb5f

Please sign in to comment.