aaron/ispc
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Add reduce_add() for int8 and int16 types.

Browse Source 
This maps to specialized instructions (e.g. PSADBW) when available.
This commit is contained in:
Matt Pharr
2013-07-25 09:11:39 -07:00
parent 2d063925a1
commit b6df447b55
24 changed files with 464 additions and 44 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
examples/intrinsics/sse4.h
View File

@@ -2528,6 +2528,22 @@ static FORCEINLINE int64_t __count_leading_zeros_i64(uint64_t v) {
///////////////////////////////////////////////////////////////////////////
// reductions
static FORCEINLINE int16_t __reduce_add_int8(__vec4_i8 v) {
// TODO: improve
int16_t ret = 0;
for (int i = 0; i < 4; ++i)
ret += v.v[i];
return ret;
}
static FORCEINLINE int32_t __reduce_add_int16(__vec4_i16 v) {
// TODO: improve
int32_t ret = 0;
for (int i = 0; i < 4; ++i)
ret += v.v[i];
return ret;
}
static FORCEINLINE float __reduce_add_float(__vec4_f v) {
float r = bits_as_float(_mm_extract_ps(v.v, 0));
r += bits_as_float(_mm_extract_ps(v.v, 1));