| /external/XNNPACK/src/qs8-requantization/ |
| D | rndnu-neon-mull.c | 83 …const int32x4_t z_scaled = vuzp1q_s32(vreinterpretq_s32_s64(z01_scaled), vreinterpretq_s32_s64(z23… in xnn_qs8_requantize_rndnu__neon_mull() local
87 …const int16x8_t zw_packed = vqaddq_s16(vqmovn_high_s32(vqmovn_s32(z_scaled), w_scaled), vzero_poin… in xnn_qs8_requantize_rndnu__neon_mull()
92 const int32x4_t z_scaled = vcombine_s32(vmovn_s64(z01_scaled), vmovn_s64(z23_scaled)); in xnn_qs8_requantize_rndnu__neon_mull() local
96 …const int16x8_t zw_packed = vqaddq_s16(vcombine_s16(vqmovn_s32(z_scaled), vqmovn_s32(w_scaled)), v… in xnn_qs8_requantize_rndnu__neon_mull()
|
| D | rndnu-neon-qdmulh.c | 70 const int32x4_t z_scaled = vrshlq_s32(z_product, vpost_shift); in xnn_qs8_requantize_rndnu__neon_qdmulh() local
75 …const int16x8_t zw_packed = vqaddq_s16(vqmovn_high_s32(vqmovn_s32(z_scaled), w_scaled), vzero_poin… in xnn_qs8_requantize_rndnu__neon_qdmulh()
79 …const int16x8_t zw_packed = vqaddq_s16(vcombine_s16(vqmovn_s32(z_scaled), vqmovn_s32(w_scaled)), v… in xnn_qs8_requantize_rndnu__neon_qdmulh()
|
| D | gemmlowp-neon.c | 78 const int32x4_t z_scaled = vrshlq_s32(z_adjusted_product, vshift); in xnn_qs8_requantize_gemmlowp__neon() local
83 …const int16x8_t zw_packed = vqaddq_s16(vqmovn_high_s32(vqmovn_s32(z_scaled), w_scaled), vzero_poin… in xnn_qs8_requantize_gemmlowp__neon()
87 …const int16x8_t zw_packed = vqaddq_s16(vcombine_s16(vqmovn_s32(z_scaled), vqmovn_s32(w_scaled)), v… in xnn_qs8_requantize_gemmlowp__neon()
|
| D | fp32-neon.c | 58 const float32x4_t z_scaled = vmulq_f32(vcvtq_f32_s32(z), vscale); in xnn_qs8_requantize_fp32__neon() local
67 const int32x4_t z_rounded = vcvtnq_s32_f32(z_scaled); in xnn_qs8_requantize_fp32__neon()
116 const float32x4_t z_clamped = vminq_f32(vmaxq_f32(z_scaled, vfmin), vfmax); in xnn_qs8_requantize_fp32__neon()
|
| D | rndna-neon.c | 111 …const int32x4_t z_scaled = vuzp1q_s32(vreinterpretq_s32_s64(z01_scaled), vreinterpretq_s32_s64(z23… in xnn_qs8_requantize_rndna__neon() local
115 …const int16x8_t zw_packed = vqaddq_s16(vqmovn_high_s32(vqmovn_s32(z_scaled), w_scaled), vzero_poin… in xnn_qs8_requantize_rndna__neon()
120 const int32x4_t z_scaled = vcombine_s32(vmovn_s64(z01_scaled), vmovn_s64(z23_scaled)); in xnn_qs8_requantize_rndna__neon() local
124 …const int16x8_t zw_packed = vqaddq_s16(vcombine_s16(vqmovn_s32(z_scaled), vqmovn_s32(w_scaled)), v… in xnn_qs8_requantize_rndna__neon()
|
| D | fp32-scalar-fmagic.c | 44 const float z_scaled = (float) z * scale; in xnn_qs8_requantize_fp32__scalar_fmagic() local
49 const float z_clamped = math_min_f32(math_max_f32(z_scaled, fmin), fmax); in xnn_qs8_requantize_fp32__scalar_fmagic()
|
| D | fp32-sse4.c | 50 const __m128 z_scaled = _mm_mul_ps(_mm_cvtepi32_ps(z), vscale); in xnn_qs8_requantize_fp32__sse4() local
66 const __m128i z_rounded = _mm_cvtps_epi32(z_scaled); in xnn_qs8_requantize_fp32__sse4()
|
| D | fp32-scalar-lrintf.c | 42 const float z_scaled = (float) z * scale; in xnn_qs8_requantize_fp32__scalar_lrintf() local
47 const float z_clamped = math_min_f32(math_max_f32(z_scaled, fmin), fmax); in xnn_qs8_requantize_fp32__scalar_lrintf()
|
| D | fp32-sse2.c | 50 const __m128 z_scaled = _mm_mul_ps(_mm_cvtepi32_ps(z), vscale); in xnn_qs8_requantize_fp32__sse2() local
66 const __m128i z_rounded = _mm_cvtps_epi32(z_scaled); in xnn_qs8_requantize_fp32__sse2()
|
| D | fp32-wasmsimd.c | 48 const v128_t z_scaled = wasm_f32x4_mul(wasm_f32x4_convert_i32x4(z), vscale); in xnn_qs8_requantize_fp32__wasmsimd() local
60 const v128_t z_clamped = wasm_f32x4_min(wasm_f32x4_max(z_scaled, vfmin), vfmax); in xnn_qs8_requantize_fp32__wasmsimd()
|
| D | rndnu-scalar.c | 62 const int32_t z_scaled = (int32_t) math_asr_s64(z_product + rounding, shift); in xnn_qs8_requantize_rndnu__scalar() local
68 const int32_t z_clamped = math_min_s32(math_max_s32(z_scaled, smin), smax); in xnn_qs8_requantize_rndnu__scalar()
|
| D | rndna-scalar-signed64.c | 68 const int32_t z_scaled = (int32_t) math_asr_s64(z_adjusted_product + rounding, shift); in xnn_qs8_requantize_rndna__scalar_signed64() local
74 const int32_t z_clamped = math_min_s32(math_max_s32(z_scaled, smin), smax); in xnn_qs8_requantize_rndna__scalar_signed64()
|
| D | rndna-scalar-unsigned64.c | 75 const int32_t z_scaled = (int32_t) (z >= 0 ? z_abs_scaled : -z_abs_scaled); in xnn_qs8_requantize_rndna__scalar_unsigned64() local
81 const int32_t z_clamped = math_min_s32(math_max_s32(z_scaled, smin), smax); in xnn_qs8_requantize_rndna__scalar_unsigned64()
|
| D | gemmlowp-scalar.c | 103 … const int32_t z_scaled = math_asr_s32(z_q31product, shift) + (int32_t) (z_remainder > threshold); in xnn_qs8_requantize_gemmlowp__scalar() local
109 const int32_t z_clamped = math_min_s32(math_max_s32(z_scaled, smin), smax); in xnn_qs8_requantize_gemmlowp__scalar()
|
| D | rndnu-sse4-sra.c | 80 const __m128i z_scaled = _mm_sra_epi32(z_prescaled, vshift); in xnn_qs8_requantize_rndnu__sse4_sra() local
84 const __m128i zw_packed = _mm_adds_epi16(_mm_packs_epi32(z_scaled, w_scaled), vzero_point); in xnn_qs8_requantize_rndnu__sse4_sra()
|
| /external/XNNPACK/src/qu8-requantization/ |
| D | gemmlowp-neon.c | 78 const int32x4_t z_scaled = vrshlq_s32(z_adjusted_product, vshift); in xnn_qu8_requantize_gemmlowp__neon() local
83 …const int16x8_t zw_packed = vqaddq_s16(vqmovn_high_s32(vqmovn_s32(z_scaled), w_scaled), vzero_poin… in xnn_qu8_requantize_gemmlowp__neon()
87 …const int16x8_t zw_packed = vqaddq_s16(vcombine_s16(vqmovn_s32(z_scaled), vqmovn_s32(w_scaled)), v… in xnn_qu8_requantize_gemmlowp__neon()
|
| D | fp32-neon.c | 58 const float32x4_t z_scaled = vmulq_f32(vcvtq_f32_s32(z), vscale); in xnn_qu8_requantize_fp32__neon() local
67 const int32x4_t z_rounded = vcvtnq_s32_f32(z_scaled); in xnn_qu8_requantize_fp32__neon()
92 const float32x4_t z_clamped = vminq_f32(vmaxq_f32(z_scaled, vfmin), vfmax); in xnn_qu8_requantize_fp32__neon()
|
| D | rndna-neon.c | 111 …const int32x4_t z_scaled = vuzp1q_s32(vreinterpretq_s32_s64(z01_scaled), vreinterpretq_s32_s64(z23… in xnn_qu8_requantize_rndna__neon() local
115 …const int16x8_t zw_packed = vqaddq_s16(vqmovn_high_s32(vqmovn_s32(z_scaled), w_scaled), vzero_poin… in xnn_qu8_requantize_rndna__neon()
120 const int32x4_t z_scaled = vcombine_s32(vmovn_s64(z01_scaled), vmovn_s64(z23_scaled)); in xnn_qu8_requantize_rndna__neon() local
124 …const int16x8_t zw_packed = vqaddq_s16(vcombine_s16(vqmovn_s32(z_scaled), vqmovn_s32(w_scaled)), v… in xnn_qu8_requantize_rndna__neon()
|
| D | fp32-sse2.c | 50 const __m128 z_scaled = _mm_mul_ps(_mm_cvtepi32_ps(z), vscale); in xnn_qu8_requantize_fp32__sse2() local
66 const __m128i z_rounded = _mm_cvtps_epi32(z_scaled); in xnn_qu8_requantize_fp32__sse2()
|
| D | fp32-scalar-fmagic.c | 44 const float z_scaled = (float) z * scale; in xnn_qu8_requantize_fp32__scalar_fmagic() local
49 const float z_clamped = math_min_f32(math_max_f32(z_scaled, fmin), fmax); in xnn_qu8_requantize_fp32__scalar_fmagic()
|
| D | fp32-scalar-lrintf.c | 42 const float z_scaled = (float) z * scale; in xnn_qu8_requantize_fp32__scalar_lrintf() local
47 const float z_clamped = math_min_f32(math_max_f32(z_scaled, fmin), fmax); in xnn_qu8_requantize_fp32__scalar_lrintf()
|
| D | fp32-wasmsimd.c | 48 const v128_t z_scaled = wasm_f32x4_mul(wasm_f32x4_convert_i32x4(z), vscale); in xnn_qu8_requantize_fp32__wasmsimd() local
60 const v128_t z_clamped = wasm_f32x4_min(wasm_f32x4_max(z_scaled, vfmin), vfmax); in xnn_qu8_requantize_fp32__wasmsimd()
|
| D | rndna-scalar-signed64.c | 68 const int32_t z_scaled = (int32_t) math_asr_s64(z_adjusted_product + rounding, shift); in xnn_qu8_requantize_rndna__scalar_signed64() local
74 const int32_t z_clamped = math_min_s32(math_max_s32(z_scaled, smin), smax); in xnn_qu8_requantize_rndna__scalar_signed64()
|
| D | rndna-scalar-unsigned64.c | 75 const int32_t z_scaled = (int32_t) (z >= 0 ? z_abs_scaled : -z_abs_scaled); in xnn_qu8_requantize_rndna__scalar_unsigned64() local
81 const int32_t z_clamped = math_min_s32(math_max_s32(z_scaled, smin), smax); in xnn_qu8_requantize_rndna__scalar_unsigned64()
|
| D | gemmlowp-scalar.c | 103 … const int32_t z_scaled = math_asr_s32(z_q31product, shift) + (int32_t) (z_remainder > threshold); in xnn_qu8_requantize_gemmlowp__scalar() local
109 const int32_t z_clamped = math_min_s32(math_max_s32(z_scaled, smin), smax); in xnn_qu8_requantize_gemmlowp__scalar()
|