diff options
| author | Erik Schnetter <schnetter@gmail.com> | 2013-07-04 15:35:35 -0400 |
|---|---|---|
| committer | Erik Schnetter <schnetter@gmail.com> | 2013-07-04 15:35:35 -0400 |
| commit | 1a456c91a43494036111dea2db89ae656aa9c987 (patch) | |
| tree | e784c81254e8ab8139435c8ead79821060eb51bb /vec_sse_float4.h | |
| parent | 00578fbb97c8b839ea8675400eae41e761d2a8e2 (diff) | |
| download | vecmathlib-1a456c91a43494036111dea2db89ae656aa9c987.zip vecmathlib-1a456c91a43494036111dea2db89ae656aa9c987.tar.gz | |
Optimize prod and sum
Diffstat (limited to 'vec_sse_float4.h')
| -rw-r--r-- | vec_sse_float4.h | 12 |
1 files changed, 10 insertions, 2 deletions
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 } |
