| /external/XNNPACK/src/qu8-requantization/ |
| D | q31-neon.c | 47 const int16x8_t vzero_point = vdupq_n_s16((int16_t)(uint16_t) zero_point); in xnn_qu8_requantize_q31__neon() local
83 …nst int16x8_t xy_packed = vqaddq_s16(vqmovn_high_s32(vqmovn_s32(x_scaled), y_scaled), vzero_point); in xnn_qu8_requantize_q31__neon()
84 …nst int16x8_t zw_packed = vqaddq_s16(vqmovn_high_s32(vqmovn_s32(z_scaled), w_scaled), vzero_point); in xnn_qu8_requantize_q31__neon()
87 …x8_t xy_packed = vqaddq_s16(vcombine_s16(vqmovn_s32(x_scaled), vqmovn_s32(y_scaled)), vzero_point); in xnn_qu8_requantize_q31__neon()
88 …x8_t zw_packed = vqaddq_s16(vcombine_s16(vqmovn_s32(z_scaled), vqmovn_s32(w_scaled)), vzero_point); in xnn_qu8_requantize_q31__neon()
|
| D | fp32-sse2.c | 32 const __m128i vzero_point = _mm_set1_epi16((short) (uint16_t) zero_point); in xnn_qu8_requantize_fp32__sse2() local
74 const __m128i xy_packed = _mm_adds_epi16(_mm_packs_epi32(x_rounded, y_rounded), vzero_point); in xnn_qu8_requantize_fp32__sse2()
75 const __m128i zw_packed = _mm_adds_epi16(_mm_packs_epi32(z_rounded, w_rounded), vzero_point); in xnn_qu8_requantize_fp32__sse2()
|
| D | precise-neon.c | 44 const int16x8_t vzero_point = vdupq_n_s16((int16_t)(uint16_t) zero_point); in xnn_qu8_requantize_precise__neon() local
115 …nst int16x8_t xy_packed = vqaddq_s16(vqmovn_high_s32(vqmovn_s32(x_scaled), y_scaled), vzero_point); in xnn_qu8_requantize_precise__neon()
116 …nst int16x8_t zw_packed = vqaddq_s16(vqmovn_high_s32(vqmovn_s32(z_scaled), w_scaled), vzero_point); in xnn_qu8_requantize_precise__neon()
124 …x8_t xy_packed = vqaddq_s16(vcombine_s16(vqmovn_s32(x_scaled), vqmovn_s32(y_scaled)), vzero_point); in xnn_qu8_requantize_precise__neon()
125 …x8_t zw_packed = vqaddq_s16(vcombine_s16(vqmovn_s32(z_scaled), vqmovn_s32(w_scaled)), vzero_point); in xnn_qu8_requantize_precise__neon()
|
| D | fp32-neon.c | 33 const int16x8_t vzero_point = vdupq_n_s16((int16_t)(uint16_t) zero_point); in xnn_qu8_requantize_fp32__neon() local
74 …t int16x8_t xy_packed = vqaddq_s16(vqmovn_high_s32(vqmovn_s32(x_rounded), y_rounded), vzero_point); in xnn_qu8_requantize_fp32__neon()
75 …t int16x8_t zw_packed = vqaddq_s16(vqmovn_high_s32(vqmovn_s32(z_rounded), w_rounded), vzero_point); in xnn_qu8_requantize_fp32__neon()
|
| D | precise-sse4.c | 41 const __m128i vzero_point = _mm_set1_epi16((short) (uint16_t) zero_point); in xnn_qu8_requantize_precise__sse4() local
93 const __m128i xy_packed = _mm_adds_epi16(_mm_packs_epi32(x_scaled, y_scaled), vzero_point); in xnn_qu8_requantize_precise__sse4()
94 const __m128i zw_packed = _mm_adds_epi16(_mm_packs_epi32(z_scaled, w_scaled), vzero_point); in xnn_qu8_requantize_precise__sse4()
|
| D | precise-ssse3.c | 41 const __m128i vzero_point = _mm_set1_epi16((short) (uint16_t) zero_point); in xnn_qu8_requantize_precise__ssse3() local
101 const __m128i xy_packed = _mm_adds_epi16(_mm_packs_epi32(x_scaled, y_scaled), vzero_point); in xnn_qu8_requantize_precise__ssse3()
102 const __m128i zw_packed = _mm_adds_epi16(_mm_packs_epi32(z_scaled, w_scaled), vzero_point); in xnn_qu8_requantize_precise__ssse3()
|
| D | q31-sse4.c | 47 const __m128i vzero_point = _mm_set1_epi16((short) (uint16_t) zero_point); in xnn_qu8_requantize_q31__sse4() local
109 const __m128i xy_packed = _mm_adds_epi16(_mm_packs_epi32(x_scaled, y_scaled), vzero_point); in xnn_qu8_requantize_q31__sse4()
110 const __m128i zw_packed = _mm_adds_epi16(_mm_packs_epi32(z_scaled, w_scaled), vzero_point); in xnn_qu8_requantize_q31__sse4()
|
| D | precise-sse2.c | 41 const __m128i vzero_point = _mm_set1_epi16((short) (uint16_t) zero_point); in xnn_qu8_requantize_precise__sse2() local
106 const __m128i xy_packed = _mm_adds_epi16(_mm_packs_epi32(x_scaled, y_scaled), vzero_point); in xnn_qu8_requantize_precise__sse2()
107 const __m128i zw_packed = _mm_adds_epi16(_mm_packs_epi32(z_scaled, w_scaled), vzero_point); in xnn_qu8_requantize_precise__sse2()
|
| D | q31-wasmsimd.c | 49 const v128_t vzero_point = wasm_i16x8_splat((int16_t) (uint16_t) zero_point); in xnn_qu8_requantize_q31__wasmsimd() local
112 …28_t xy_packed = wasm_i16x8_add_saturate(wasm_i16x8_narrow_i32x4(x_scaled, y_scaled), vzero_point); in xnn_qu8_requantize_q31__wasmsimd()
113 …28_t zw_packed = wasm_i16x8_add_saturate(wasm_i16x8_narrow_i32x4(z_scaled, w_scaled), vzero_point); in xnn_qu8_requantize_q31__wasmsimd()
|
| D | q31-ssse3.c | 47 const __m128i vzero_point = _mm_set1_epi16((short) (uint16_t) zero_point); in xnn_qu8_requantize_q31__ssse3() local
158 const __m128i xy_packed = _mm_adds_epi16(_mm_packs_epi32(x_scaled, y_scaled), vzero_point); in xnn_qu8_requantize_q31__ssse3()
159 const __m128i zw_packed = _mm_adds_epi16(_mm_packs_epi32(z_scaled, w_scaled), vzero_point); in xnn_qu8_requantize_q31__ssse3()
|
| D | q31-sse2.c | 47 const __m128i vzero_point = _mm_set1_epi16((short) (uint16_t) zero_point); in xnn_qu8_requantize_q31__sse2() local
158 const __m128i xy_packed = _mm_adds_epi16(_mm_packs_epi32(x_scaled, y_scaled), vzero_point); in xnn_qu8_requantize_q31__sse2()
159 const __m128i zw_packed = _mm_adds_epi16(_mm_packs_epi32(z_scaled, w_scaled), vzero_point); in xnn_qu8_requantize_q31__sse2()
|
| /external/XNNPACK/src/qs8-requantization/ |
| D | q31-neon.c | 47 const int16x8_t vzero_point = vdupq_n_s16((int16_t) zero_point); in xnn_qs8_requantize_q31__neon() local
83 …nst int16x8_t xy_packed = vqaddq_s16(vqmovn_high_s32(vqmovn_s32(x_scaled), y_scaled), vzero_point); in xnn_qs8_requantize_q31__neon()
84 …nst int16x8_t zw_packed = vqaddq_s16(vqmovn_high_s32(vqmovn_s32(z_scaled), w_scaled), vzero_point); in xnn_qs8_requantize_q31__neon()
87 …x8_t xy_packed = vqaddq_s16(vcombine_s16(vqmovn_s32(x_scaled), vqmovn_s32(y_scaled)), vzero_point); in xnn_qs8_requantize_q31__neon()
88 …x8_t zw_packed = vqaddq_s16(vcombine_s16(vqmovn_s32(z_scaled), vqmovn_s32(w_scaled)), vzero_point); in xnn_qs8_requantize_q31__neon()
|
| D | fp32-sse4.c | 32 const __m128i vzero_point = _mm_set1_epi16((short) zero_point); in xnn_qs8_requantize_fp32__sse4() local
74 const __m128i xy_packed = _mm_adds_epi16(_mm_packs_epi32(x_rounded, y_rounded), vzero_point); in xnn_qs8_requantize_fp32__sse4()
75 const __m128i zw_packed = _mm_adds_epi16(_mm_packs_epi32(z_rounded, w_rounded), vzero_point); in xnn_qs8_requantize_fp32__sse4()
|
| D | fp32-sse2.c | 32 const __m128i vzero_point = _mm_set1_epi16((short) (uint16_t) zero_point); in xnn_qs8_requantize_fp32__sse2() local
74 const __m128i xy_packed = _mm_adds_epi16(_mm_packs_epi32(x_rounded, y_rounded), vzero_point); in xnn_qs8_requantize_fp32__sse2()
75 const __m128i zw_packed = _mm_adds_epi16(_mm_packs_epi32(z_rounded, w_rounded), vzero_point); in xnn_qs8_requantize_fp32__sse2()
|
| D | precise-neon.c | 44 const int16x8_t vzero_point = vdupq_n_s16((int16_t) zero_point); in xnn_qs8_requantize_precise__neon() local
115 …nst int16x8_t xy_packed = vqaddq_s16(vqmovn_high_s32(vqmovn_s32(x_scaled), y_scaled), vzero_point); in xnn_qs8_requantize_precise__neon()
116 …nst int16x8_t zw_packed = vqaddq_s16(vqmovn_high_s32(vqmovn_s32(z_scaled), w_scaled), vzero_point); in xnn_qs8_requantize_precise__neon()
124 …x8_t xy_packed = vqaddq_s16(vcombine_s16(vqmovn_s32(x_scaled), vqmovn_s32(y_scaled)), vzero_point); in xnn_qs8_requantize_precise__neon()
125 …x8_t zw_packed = vqaddq_s16(vcombine_s16(vqmovn_s32(z_scaled), vqmovn_s32(w_scaled)), vzero_point); in xnn_qs8_requantize_precise__neon()
|
| D | fp32-neon.c | 33 const int16x8_t vzero_point = vdupq_n_s16((int16_t) zero_point); in xnn_qs8_requantize_fp32__neon() local
74 …t int16x8_t xy_packed = vqaddq_s16(vqmovn_high_s32(vqmovn_s32(x_rounded), y_rounded), vzero_point); in xnn_qs8_requantize_fp32__neon()
75 …t int16x8_t zw_packed = vqaddq_s16(vqmovn_high_s32(vqmovn_s32(z_rounded), w_rounded), vzero_point); in xnn_qs8_requantize_fp32__neon()
|
| D | precise-sse4.c | 41 const __m128i vzero_point = _mm_set1_epi16((short) zero_point); in xnn_qs8_requantize_precise__sse4() local
93 const __m128i xy_packed = _mm_adds_epi16(_mm_packs_epi32(x_scaled, y_scaled), vzero_point); in xnn_qs8_requantize_precise__sse4()
94 const __m128i zw_packed = _mm_adds_epi16(_mm_packs_epi32(z_scaled, w_scaled), vzero_point); in xnn_qs8_requantize_precise__sse4()
|
| D | precise-ssse3.c | 41 const __m128i vzero_point = _mm_set1_epi16((short) zero_point); in xnn_qs8_requantize_precise__ssse3() local
101 const __m128i xy_packed = _mm_adds_epi16(_mm_packs_epi32(x_scaled, y_scaled), vzero_point); in xnn_qs8_requantize_precise__ssse3()
102 const __m128i zw_packed = _mm_adds_epi16(_mm_packs_epi32(z_scaled, w_scaled), vzero_point); in xnn_qs8_requantize_precise__ssse3()
|
| D | q31-sse4.c | 47 const __m128i vzero_point = _mm_set1_epi16((short) zero_point); in xnn_qs8_requantize_q31__sse4() local
109 const __m128i xy_packed = _mm_adds_epi16(_mm_packs_epi32(x_scaled, y_scaled), vzero_point); in xnn_qs8_requantize_q31__sse4()
110 const __m128i zw_packed = _mm_adds_epi16(_mm_packs_epi32(z_scaled, w_scaled), vzero_point); in xnn_qs8_requantize_q31__sse4()
|
| D | q31-wasmsimd.c | 49 const v128_t vzero_point = wasm_i16x8_splat((int16_t) zero_point); in xnn_qs8_requantize_q31__wasmsimd() local
112 …28_t xy_packed = wasm_i16x8_add_saturate(wasm_i16x8_narrow_i32x4(x_scaled, y_scaled), vzero_point); in xnn_qs8_requantize_q31__wasmsimd()
113 …28_t zw_packed = wasm_i16x8_add_saturate(wasm_i16x8_narrow_i32x4(z_scaled, w_scaled), vzero_point); in xnn_qs8_requantize_q31__wasmsimd()
|
| D | precise-sse2.c | 41 const __m128i vzero_point = _mm_set1_epi16((short) zero_point); in xnn_qs8_requantize_precise__sse2() local
106 const __m128i xy_packed = _mm_adds_epi16(_mm_packs_epi32(x_scaled, y_scaled), vzero_point); in xnn_qs8_requantize_precise__sse2()
107 const __m128i zw_packed = _mm_adds_epi16(_mm_packs_epi32(z_scaled, w_scaled), vzero_point); in xnn_qs8_requantize_precise__sse2()
|
| D | q31-ssse3.c | 47 const __m128i vzero_point = _mm_set1_epi16((short) zero_point); in xnn_qs8_requantize_q31__ssse3() local
158 const __m128i xy_packed = _mm_adds_epi16(_mm_packs_epi32(x_scaled, y_scaled), vzero_point); in xnn_qs8_requantize_q31__ssse3()
159 const __m128i zw_packed = _mm_adds_epi16(_mm_packs_epi32(z_scaled, w_scaled), vzero_point); in xnn_qs8_requantize_q31__ssse3()
|
| D | q31-sse2.c | 47 const __m128i vzero_point = _mm_set1_epi16((short) zero_point); in xnn_qs8_requantize_q31__sse2() local
158 const __m128i xy_packed = _mm_adds_epi16(_mm_packs_epi32(x_scaled, y_scaled), vzero_point); in xnn_qs8_requantize_q31__sse2()
159 const __m128i zw_packed = _mm_adds_epi16(_mm_packs_epi32(z_scaled, w_scaled), vzero_point); in xnn_qs8_requantize_q31__sse2()
|