diff options
author | Martin Braun <martin.braun@ettus.com> | 2019-07-18 15:36:11 -0700 |
---|---|---|
committer | Martin Braun <martin.braun@ettus.com> | 2019-11-26 11:49:10 -0800 |
commit | 9df26a9d89ef8fb50a667428066f3ef1732245c9 (patch) | |
tree | aa8aa5adf1c40e0aecb3e45a527511af96e05ca1 /host/lib/convert/sse2_sc8_to_fc64.cpp | |
parent | fed32af0806a730e0f4202003dc49cb736c832fb (diff) | |
download | uhd-9df26a9d89ef8fb50a667428066f3ef1732245c9.tar.gz uhd-9df26a9d89ef8fb50a667428066f3ef1732245c9.tar.bz2 uhd-9df26a9d89ef8fb50a667428066f3ef1732245c9.zip |
convert: sse2: Apply clang-format
Diffstat (limited to 'host/lib/convert/sse2_sc8_to_fc64.cpp')
-rw-r--r-- | host/lib/convert/sse2_sc8_to_fc64.cpp | 168 |
1 files changed, 86 insertions, 82 deletions
diff --git a/host/lib/convert/sse2_sc8_to_fc64.cpp b/host/lib/convert/sse2_sc8_to_fc64.cpp index f5b406152..3cc2fefd0 100644 --- a/host/lib/convert/sse2_sc8_to_fc64.cpp +++ b/host/lib/convert/sse2_sc8_to_fc64.cpp @@ -13,129 +13,133 @@ using namespace uhd::convert; static const __m128i zeroi = _mm_setzero_si128(); -UHD_INLINE void unpack_sc32_8x( - const __m128i &in, - __m128d &out0, __m128d &out1, - __m128d &out2, __m128d &out3, - __m128d &out4, __m128d &out5, - __m128d &out6, __m128d &out7, - const __m128d &scalar -){ +UHD_INLINE void unpack_sc32_8x(const __m128i& in, + __m128d& out0, + __m128d& out1, + __m128d& out2, + __m128d& out3, + __m128d& out4, + __m128d& out5, + __m128d& out6, + __m128d& out7, + const __m128d& scalar) +{ const int shuf = _MM_SHUFFLE(1, 0, 3, 2); __m128i tmp; const __m128i tmplo = _mm_unpacklo_epi8(zeroi, in); /* value in upper 8 bits */ - tmp = _mm_unpacklo_epi16(zeroi, tmplo); /* value in upper 16 bits */ - out0 = _mm_mul_pd(_mm_cvtepi32_pd(tmp), scalar); - tmp = _mm_shuffle_epi32(tmp, shuf); - out1 = _mm_mul_pd(_mm_cvtepi32_pd(tmp), scalar); - tmp = _mm_unpackhi_epi16(zeroi, tmplo); - out2 = _mm_mul_pd(_mm_cvtepi32_pd(tmp), scalar); - tmp = _mm_shuffle_epi32(tmp, shuf); - out3 = _mm_mul_pd(_mm_cvtepi32_pd(tmp), scalar); + tmp = _mm_unpacklo_epi16(zeroi, tmplo); /* value in upper 16 bits */ + out0 = _mm_mul_pd(_mm_cvtepi32_pd(tmp), scalar); + tmp = _mm_shuffle_epi32(tmp, shuf); + out1 = _mm_mul_pd(_mm_cvtepi32_pd(tmp), scalar); + tmp = _mm_unpackhi_epi16(zeroi, tmplo); + out2 = _mm_mul_pd(_mm_cvtepi32_pd(tmp), scalar); + tmp = _mm_shuffle_epi32(tmp, shuf); + out3 = _mm_mul_pd(_mm_cvtepi32_pd(tmp), scalar); const __m128i tmphi = _mm_unpackhi_epi8(zeroi, in); - tmp = _mm_unpacklo_epi16(zeroi, tmphi); - out4 = _mm_mul_pd(_mm_cvtepi32_pd(tmp), scalar); - tmp = _mm_shuffle_epi32(tmp, shuf); - out5 = _mm_mul_pd(_mm_cvtepi32_pd(tmp), scalar); - tmp = _mm_unpackhi_epi16(zeroi, tmphi); - out6 = _mm_mul_pd(_mm_cvtepi32_pd(tmp), scalar); - tmp = _mm_shuffle_epi32(tmp, shuf); - out7 = _mm_mul_pd(_mm_cvtepi32_pd(tmp), scalar); + tmp = _mm_unpacklo_epi16(zeroi, tmphi); + out4 = _mm_mul_pd(_mm_cvtepi32_pd(tmp), scalar); + tmp = _mm_shuffle_epi32(tmp, shuf); + out5 = _mm_mul_pd(_mm_cvtepi32_pd(tmp), scalar); + tmp = _mm_unpackhi_epi16(zeroi, tmphi); + out6 = _mm_mul_pd(_mm_cvtepi32_pd(tmp), scalar); + tmp = _mm_shuffle_epi32(tmp, shuf); + out7 = _mm_mul_pd(_mm_cvtepi32_pd(tmp), scalar); } -DECLARE_CONVERTER(sc8_item32_be, 1, fc64, 1, PRIORITY_SIMD){ - const item32_t *input = reinterpret_cast<const item32_t *>(size_t(inputs[0]) & ~0x3); - fc64_t *output = reinterpret_cast<fc64_t *>(outputs[0]); +DECLARE_CONVERTER(sc8_item32_be, 1, fc64, 1, PRIORITY_SIMD) +{ + const item32_t* input = reinterpret_cast<const item32_t*>(size_t(inputs[0]) & ~0x3); + fc64_t* output = reinterpret_cast<fc64_t*>(outputs[0]); - const __m128d scalar = _mm_set1_pd(scale_factor/(1 << 24)); + const __m128d scalar = _mm_set1_pd(scale_factor / (1 << 24)); size_t i = 0, j = 0; fc32_t dummy; size_t num_samps = nsamps; - if ((size_t(inputs[0]) & 0x3) != 0){ + if ((size_t(inputs[0]) & 0x3) != 0) { item32_sc8_to_xx<uhd::ntohx>(input++, output++, 1, scale_factor); num_samps--; } - #define convert_sc8_item32_1_to_fc64_1_bswap_guts(_al_) \ - for (; j+7 < num_samps; j+=8, i+=4){ \ - /* load from input */ \ - __m128i tmpi = _mm_loadu_si128(reinterpret_cast<const __m128i *>(input+i)); \ - \ - /* unpack */ \ - __m128d tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; \ +#define convert_sc8_item32_1_to_fc64_1_bswap_guts(_al_) \ + for (; j + 7 < num_samps; j += 8, i += 4) { \ + /* load from input */ \ + __m128i tmpi = _mm_loadu_si128(reinterpret_cast<const __m128i*>(input + i)); \ + \ + /* unpack */ \ + __m128d tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; \ unpack_sc32_8x(tmpi, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, scalar); \ - \ - /* store to output */ \ - _mm_store ## _al_ ## pd(reinterpret_cast<double *>(output+j+0), tmp0); \ - _mm_store ## _al_ ## pd(reinterpret_cast<double *>(output+j+1), tmp1); \ - _mm_store ## _al_ ## pd(reinterpret_cast<double *>(output+j+2), tmp2); \ - _mm_store ## _al_ ## pd(reinterpret_cast<double *>(output+j+3), tmp3); \ - _mm_store ## _al_ ## pd(reinterpret_cast<double *>(output+j+4), tmp4); \ - _mm_store ## _al_ ## pd(reinterpret_cast<double *>(output+j+5), tmp5); \ - _mm_store ## _al_ ## pd(reinterpret_cast<double *>(output+j+6), tmp6); \ - _mm_store ## _al_ ## pd(reinterpret_cast<double *>(output+j+7), tmp7); \ + \ + /* store to output */ \ + _mm_store##_al_##pd(reinterpret_cast<double*>(output + j + 0), tmp0); \ + _mm_store##_al_##pd(reinterpret_cast<double*>(output + j + 1), tmp1); \ + _mm_store##_al_##pd(reinterpret_cast<double*>(output + j + 2), tmp2); \ + _mm_store##_al_##pd(reinterpret_cast<double*>(output + j + 3), tmp3); \ + _mm_store##_al_##pd(reinterpret_cast<double*>(output + j + 4), tmp4); \ + _mm_store##_al_##pd(reinterpret_cast<double*>(output + j + 5), tmp5); \ + _mm_store##_al_##pd(reinterpret_cast<double*>(output + j + 6), tmp6); \ + _mm_store##_al_##pd(reinterpret_cast<double*>(output + j + 7), tmp7); \ } - //dispatch according to alignment - if ((size_t(output) & 0xf) == 0){ + // dispatch according to alignment + if ((size_t(output) & 0xf) == 0) { convert_sc8_item32_1_to_fc64_1_bswap_guts(_) - } - else{ + } else { convert_sc8_item32_1_to_fc64_1_bswap_guts(u_) } - //convert remainder - item32_sc8_to_xx<uhd::ntohx>(input+i, output+j, num_samps-j, scale_factor); + // convert remainder + item32_sc8_to_xx<uhd::ntohx>(input + i, output + j, num_samps - j, scale_factor); } -DECLARE_CONVERTER(sc8_item32_le, 1, fc64, 1, PRIORITY_SIMD){ - const item32_t *input = reinterpret_cast<const item32_t *>(size_t(inputs[0]) & ~0x3); - fc64_t *output = reinterpret_cast<fc64_t *>(outputs[0]); +DECLARE_CONVERTER(sc8_item32_le, 1, fc64, 1, PRIORITY_SIMD) +{ + const item32_t* input = reinterpret_cast<const item32_t*>(size_t(inputs[0]) & ~0x3); + fc64_t* output = reinterpret_cast<fc64_t*>(outputs[0]); - const __m128d scalar = _mm_set1_pd(scale_factor/(1 << 24)); + const __m128d scalar = _mm_set1_pd(scale_factor / (1 << 24)); size_t i = 0, j = 0; fc32_t dummy; size_t num_samps = nsamps; - if ((size_t(inputs[0]) & 0x3) != 0){ + if ((size_t(inputs[0]) & 0x3) != 0) { item32_sc8_to_xx<uhd::wtohx>(input++, output++, 1, scale_factor); num_samps--; } - #define convert_sc8_item32_1_to_fc64_1_nswap_guts(_al_) \ - for (; j+7 < num_samps; j+=8, i+=4){ \ - /* load from input */ \ - __m128i tmpi = _mm_loadu_si128(reinterpret_cast<const __m128i *>(input+i)); \ - \ - /* unpack */ \ - __m128d tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; \ - tmpi = _mm_or_si128(_mm_srli_epi16(tmpi, 8), _mm_slli_epi16(tmpi, 8)); /*byteswap*/\ - unpack_sc32_8x(tmpi, tmp1, tmp0, tmp3, tmp2, tmp5, tmp4, tmp7, tmp6, scalar); \ - \ - /* store to output */ \ - _mm_store ## _al_ ## pd(reinterpret_cast<double *>(output+j+0), tmp0); \ - _mm_store ## _al_ ## pd(reinterpret_cast<double *>(output+j+1), tmp1); \ - _mm_store ## _al_ ## pd(reinterpret_cast<double *>(output+j+2), tmp2); \ - _mm_store ## _al_ ## pd(reinterpret_cast<double *>(output+j+3), tmp3); \ - _mm_store ## _al_ ## pd(reinterpret_cast<double *>(output+j+4), tmp4); \ - _mm_store ## _al_ ## pd(reinterpret_cast<double *>(output+j+5), tmp5); \ - _mm_store ## _al_ ## pd(reinterpret_cast<double *>(output+j+6), tmp6); \ - _mm_store ## _al_ ## pd(reinterpret_cast<double *>(output+j+7), tmp7); \ +#define convert_sc8_item32_1_to_fc64_1_nswap_guts(_al_) \ + for (; j + 7 < num_samps; j += 8, i += 4) { \ + /* load from input */ \ + __m128i tmpi = _mm_loadu_si128(reinterpret_cast<const __m128i*>(input + i)); \ + \ + /* unpack */ \ + __m128d tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; \ + tmpi = \ + _mm_or_si128(_mm_srli_epi16(tmpi, 8), _mm_slli_epi16(tmpi, 8)); /*byteswap*/ \ + unpack_sc32_8x(tmpi, tmp1, tmp0, tmp3, tmp2, tmp5, tmp4, tmp7, tmp6, scalar); \ + \ + /* store to output */ \ + _mm_store##_al_##pd(reinterpret_cast<double*>(output + j + 0), tmp0); \ + _mm_store##_al_##pd(reinterpret_cast<double*>(output + j + 1), tmp1); \ + _mm_store##_al_##pd(reinterpret_cast<double*>(output + j + 2), tmp2); \ + _mm_store##_al_##pd(reinterpret_cast<double*>(output + j + 3), tmp3); \ + _mm_store##_al_##pd(reinterpret_cast<double*>(output + j + 4), tmp4); \ + _mm_store##_al_##pd(reinterpret_cast<double*>(output + j + 5), tmp5); \ + _mm_store##_al_##pd(reinterpret_cast<double*>(output + j + 6), tmp6); \ + _mm_store##_al_##pd(reinterpret_cast<double*>(output + j + 7), tmp7); \ } - //dispatch according to alignment - if ((size_t(output) & 0xf) == 0){ + // dispatch according to alignment + if ((size_t(output) & 0xf) == 0) { convert_sc8_item32_1_to_fc64_1_nswap_guts(_) - } - else{ + } else { convert_sc8_item32_1_to_fc64_1_nswap_guts(u_) } - //convert remainder - item32_sc8_to_xx<uhd::wtohx>(input+i, output+j, num_samps-j, scale_factor); + // convert remainder + item32_sc8_to_xx<uhd::wtohx>(input + i, output + j, num_samps - j, scale_factor); } |