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

changed codestyle to 2 spaces in 'knc.h'

Browse Source
This commit is contained in:
Anton Mitrokhin
2014-09-25 12:55:21 +04:00
parent 832aff5d76
commit 4560df284b
1 changed files with 1081 additions and 1082 deletions
Download Patch File Download Diff File Expand all files Collapse all files

37
examples/intrinsics/knc.h
View File

@@ -4,7 +4,7 @@
Redistribution and use in source and binary forms, with or without Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are modification, are permitted provided that the following conditions are
met: met:
* Redistributions of source code must retain the above copyright * Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer. notice, this list of conditions and the following disclaimer.
@@ -29,7 +29,7 @@
LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/ */
#include <limits.h> // INT_MIN #include <limits.h> // INT_MIN
#include <stdint.h> #include <stdint.h>
@@ -86,10 +86,10 @@ struct __vec16_i32;
#if 0 #if 0
/* (iw) actually, this *SHOULD* be the right implementation for a /* (iw) actually, this *SHOULD* be the right implementation for a
vec16_i1: this one is a class that can have a constructor (which vec16_i1: this one is a class that can have a constructor (which
ISPC sometimes emits for these vectors...) This version might ISPC sometimes emits for these vectors...) This version might
not be working with embree's ISPC bindings, probably because not be working with embree's ISPC bindings, probably because
embree still uses the 'wrong' implementation */ embree still uses the 'wrong' implementation */
typedef struct PRE_ALIGN(2) __vec16_i1 typedef struct PRE_ALIGN(2) __vec16_i1
{ {
FORCEINLINE operator __mmask16() const { return v; } FORCEINLINE operator __mmask16() const { return v; }
@@ -277,9 +277,9 @@ inline std::ostream &operator<<(std::ostream &out, const __vec16_i64 &v)
out << "["; out << "[";
uint32_t *ptr = (uint32_t*)&v; uint32_t *ptr = (uint32_t*)&v;
for (int i=0;i<16;i++) { for (int i=0;i<16;i++) {
uint64_t val = (uint64_t(ptr[i])<<32)+ptr[i+16]; uint64_t val = (uint64_t(ptr[i])<<32)+ptr[i+16];
out << (i?",":"") << ((int*)val); out << (i?",":"") << ((int*)val);
} }
out << "]" << std::flush; out << "]" << std::flush;
return out; return out;
} }
@@ -382,14 +382,14 @@ static FORCEINLINE int64_t __extract_element(const __vec16_i64 &v, uint32_t inde
/* /*
static FORCEINLINE void __insert_element(__vec16_i1 *vec, int index, static FORCEINLINE void __insert_element(__vec16_i1 *vec, int index,
bool val) { bool val) {
if (val == false) if (val == false)
vec->v &= ~(1 << index); vec->v &= ~(1 << index);
else else
vec->v |= (1 << index); vec->v |= (1 << index);
} }
*/ */
template <int ALIGN> static FORCEINLINE __vec16_i1 __load(const __vec16_i1 *p) { template <int ALIGN> static FORCEINLINE __vec16_i1 __load(const __vec16_i1 *p) {
const uint16_t *ptr = (const uint16_t *)p; const uint16_t *ptr = (const uint16_t *)p;
@@ -424,7 +424,7 @@ template <> FORCEINLINE __vec16_i1 __undef_i1<__vec16_i1>() {
/* /*
TODO TODO
*/ */
@@ -435,7 +435,7 @@ TODO
/* /*
TODO TODO
*/ */
@@ -684,7 +684,7 @@ template <> FORCEINLINE void __store<64>(__vec16_i32 *p, __vec16_i32 v) {
/////////////////////////////////////////////////////////////////////////// ///////////////////////////////////////////////////////////////////////////
// int64 // int64
/////////////////////////////////////////////////////////////////////////// ///////////////////////////////////////////////////////////////////////////
static FORCEINLINE static FORCEINLINE
void __masked_store_i64(void *p, const __vec16_i64 &v, __vec16_i1 mask) void __masked_store_i64(void *p, const __vec16_i64 &v, __vec16_i1 mask)
{ {
__m512i v1; __m512i v1;
@@ -998,7 +998,7 @@ __gather_base_offsets32_i64(uint8_t *base, uint32_t scale, __vec16_i32 offsets,
poitners, so in here there's no way to tell - only thing we can do poitners, so in here there's no way to tell - only thing we can do
is pick one... is pick one...
*/ */
static FORCEINLINE __vec16_i64 static FORCEINLINE __vec16_i64
__gather64_i64(__vec16_i64 addr, __vec16_i1 mask) __gather64_i64(__vec16_i64 addr, __vec16_i1 mask)
{ {
__vec16_i64 ret; __vec16_i64 ret;
@@ -1862,6 +1862,7 @@ template <int ALIGN> static FORCEINLINE __vec16_i8 __load(const __vec16_i8 *p) {
template <int ALIGN> static FORCEINLINE void __store(__vec16_i8 *p, __vec16_i8 v) { template <int ALIGN> static FORCEINLINE void __store(__vec16_i8 *p, __vec16_i8 v) {
*p = v; *p = v;
} }
static FORCEINLINE void static FORCEINLINE void
__scatter_base_offsets32_i8(uint8_t *b, uint32_t scale, __vec16_i32 offsets, __scatter_base_offsets32_i8(uint8_t *b, uint32_t scale, __vec16_i32 offsets,
__vec16_i8 val, __vec16_i1 mask) __vec16_i8 val, __vec16_i1 mask)
@@ -2184,8 +2185,7 @@ __scatter_base_offsets32_float(void *base, uint32_t scale, __vec16_i32 offsets,
// packed load/store // packed load/store
/////////////////////////////////////////////////////////////////////////// ///////////////////////////////////////////////////////////////////////////
static FORCEINLINE int32_t __packed_load_active(uint32_t *p, __vec16_i32 *val, static FORCEINLINE int32_t __packed_load_active(uint32_t *p, __vec16_i32 *val, __vec16_i1 mask) {
__vec16_i1 mask) {
__vec16_i32 v = __load<64>(val); __vec16_i32 v = __load<64>(val);
v = _mm512_mask_extloadunpacklo_epi32(v, mask, p, _MM_UPCONV_EPI32_NONE, _MM_HINT_NONE); v = _mm512_mask_extloadunpacklo_epi32(v, mask, p, _MM_UPCONV_EPI32_NONE, _MM_HINT_NONE);
v = _mm512_mask_extloadunpackhi_epi32(v, mask, (uint8_t*)p+64, _MM_UPCONV_EPI32_NONE, _MM_HINT_NONE); v = _mm512_mask_extloadunpackhi_epi32(v, mask, (uint8_t*)p+64, _MM_UPCONV_EPI32_NONE, _MM_HINT_NONE);
@@ -2193,8 +2193,7 @@ static FORCEINLINE int32_t __packed_load_active(uint32_t *p, __vec16_i32 *val,
return _mm_countbits_32(uint32_t(mask)); return _mm_countbits_32(uint32_t(mask));
} }
static FORCEINLINE int32_t __packed_store_active(uint32_t *p, __vec16_i32 val, static FORCEINLINE int32_t __packed_store_active(uint32_t *p, __vec16_i32 val, __vec16_i1 mask) {
__vec16_i1 mask) {
_mm512_mask_extpackstorelo_epi32(p, mask, val, _MM_DOWNCONV_EPI32_NONE, _MM_HINT_NONE); _mm512_mask_extpackstorelo_epi32(p, mask, val, _MM_DOWNCONV_EPI32_NONE, _MM_HINT_NONE);
_mm512_mask_extpackstorehi_epi32((uint8_t*)p+64, mask, val, _MM_DOWNCONV_EPI32_NONE, _MM_HINT_NONE); _mm512_mask_extpackstorehi_epi32((uint8_t*)p+64, mask, val, _MM_DOWNCONV_EPI32_NONE, _MM_HINT_NONE);
return _mm_countbits_32(uint32_t(mask)); return _mm_countbits_32(uint32_t(mask));