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

reduce_add_int8/16 implementation fixed

Browse Source
This commit is contained in:
Anton Mitrokhin
2014-11-27 17:41:21 +04:00
parent 7333e6ab7a
commit 0a987ad06f
1 changed files with 10 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files

25
examples/intrinsics/knc.h
View File

@@ -2089,6 +2089,14 @@ static FORCEINLINE TYPE __select(bool cond, TYPE a, TYPE b) { \
return cond ? a : b; \ return cond ? a : b; \
} }
#define REDUCE_ADD(TYPE, VTYPE, NAME) \
static FORCEINLINE TYPE NAME(VTYPE v) { \
TYPE ret = v[0]; \
for (int i = 1; i < 16; ++i) \
ret = ret + v[i]; \
return ret; \
}
/////////////////////////////////////////////////////////////////////////// ///////////////////////////////////////////////////////////////////////////
// int8 // int8
/////////////////////////////////////////////////////////////////////////// ///////////////////////////////////////////////////////////////////////////
@@ -2463,21 +2471,8 @@ static FORCEINLINE int64_t __count_trailing_zeros_i64(const __vec1_i64 mask) {
// reductions // reductions
/////////////////////////////////////////////////////////////////////////// ///////////////////////////////////////////////////////////////////////////
static FORCEINLINE int8_t __reduce_add_int8(__vec16_i8 v) { REDUCE_ADD ( int16_t, __vec16_i8, __reduce_add_int8)
// TODO: improve this! REDUCE_ADD ( int32_t, __vec16_i16, __reduce_add_int16)
int16_t ret = 0;
for (int i = 0; i < 16; ++i)
ret += v.v[i];
return ret;
}
static FORCEINLINE int16_t __reduce_add_int16(__vec16_i16 v) {
// TODO: improve this!
int32_t ret = 0;
for (int i = 0; i < 16; ++i)
ret += v.v[i];
return ret;
}
static FORCEINLINE int32_t __reduce_add_int32(__vec16_i32 v) { static FORCEINLINE int32_t __reduce_add_int32(__vec16_i32 v) {
return _mm512_reduce_add_epi32(v); return _mm512_reduce_add_epi32(v);