diff options
| -rw-r--r-- | vec_avx_double4.h | 6 | ||||
| -rw-r--r-- | vec_avx_float8.h | 12 | ||||
| -rw-r--r-- | vec_qpx_double4.h | 9 | ||||
| -rw-r--r-- | vec_sse_float4.h | 12 |
4 files changed, 31 insertions, 8 deletions
diff --git a/vec_avx_double4.h b/vec_avx_double4.h index 9d60b3d..f45855c 100644 --- a/vec_avx_double4.h +++ b/vec_avx_double4.h @@ -533,7 +533,11 @@ namespace vecmathlib { } real_t prod() const { - return (*this)[0] * (*this)[1] * (*this)[2] * (*this)[3]; + // return (*this)[0] * (*this)[1] * (*this)[2] * (*this)[3]; + realvec_t x0123 = *this; + realvec_t x1032 = _mm256_shuffle_pd(x0123, x0123, 0b0101); + realvec_t y0022 = x0123 * x1032; + return y0022[0] * y0022[2]; } real_t sum() const { diff --git a/vec_avx_float8.h b/vec_avx_float8.h index 3bb13ec..02a1ec8 100644 --- a/vec_avx_float8.h +++ b/vec_avx_float8.h @@ -526,9 +526,15 @@ namespace vecmathlib { } real_t prod() const { - return - (*this)[0] * (*this)[1] * (*this)[2] * (*this)[3] * - (*this)[4] * (*this)[5] * (*this)[6] * (*this)[7]; + // return + // (*this)[0] * (*this)[1] * (*this)[2] * (*this)[3] * + // (*this)[4] * (*this)[5] * (*this)[6] * (*this)[7]; + realvec_t x01234567 = *this; + realvec_t x10325476 = _mm256_shuffle_ps(x01234567, x01234567, 0b10110001); + realvec_t y00224466 = x01234567 * x10325476; + realvec_t y22006644 = _mm256_shuffle_ps(y00224466, y00224466, 0b01001110); + realvec_t z00004444 = y00224466 * y22006644; + return z00004444[0] * z00004444[4]; } real_t sum() const { diff --git a/vec_qpx_double4.h b/vec_qpx_double4.h index 5bf830f..684c161 100644 --- a/vec_qpx_double4.h +++ b/vec_qpx_double4.h @@ -531,11 +531,16 @@ namespace vecmathlib { } real_t prod() const { - return (*this)[0] * (*this)[1] * (*this)[2] * (*this)[3]; + // return (*this)[0] * (*this)[1] * (*this)[2] * (*this)[3]; + realvec_t x = vec_xmul(v, v); + return x[1] * x[3]; } real_t sum() const { - return (*this)[0] + (*this)[1] + (*this)[2] + (*this)[3]; + // return (*this)[0] + (*this)[1] + (*this)[2] + (*this)[3]; + realvec_t c1 = vec_logical(v, v, 0b1111); + realvec_t x = vec_xxmadd(v, c1, v); + return x[0] + x[2]; } diff --git a/vec_sse_float4.h b/vec_sse_float4.h index 1466013..dc3ecb6 100644 --- a/vec_sse_float4.h +++ b/vec_sse_float4.h @@ -499,7 +499,11 @@ namespace vecmathlib { } real_t prod() const { - return (*this)[0] * (*this)[1] * (*this)[2] * (*this)[3]; + // return (*this)[0] * (*this)[1] * (*this)[2] * (*this)[3]; + realvec_t x0123 = *this; + realvec_t x1032 = _mm_shuffle_ps(x0123, x0123, 0b10110001); + realvec_t y0022 = x0123 * x1032; + return y0022[0] * y0022[2]; } real_t sum() const { @@ -509,7 +513,11 @@ namespace vecmathlib { x = _mm_hadd_ps(x.v, x.v); return x[0]; #else - return (*this)[0] + (*this)[1] + (*this)[2] + (*this)[3]; + // return (*this)[0] + (*this)[1] + (*this)[2] + (*this)[3]; + realvec_t x0123 = *this; + realvec_t x1032 = _mm_shuffle_ps(x0123, x0123, 0b10110001); + realvec_t y0022 = x0123 + x1032; + return y0022[0] + y0022[2]; #endif } |
