Added the following mask tests: __any(), __all(), __none() for all supported targets.
This allows for more efficient code generation of KNC.
This commit is contained in:
@@ -55,6 +55,7 @@ typedef int64_t __vec1_i64;
|
||||
|
||||
struct __vec32_i1 {
|
||||
__vec32_i1() { }
|
||||
__vec32_i1(const uint32_t &vv) : v(vv) { }
|
||||
__vec32_i1(uint32_t v0, uint32_t v1, uint32_t v2, uint32_t v3,
|
||||
uint32_t v4, uint32_t v5, uint32_t v6, uint32_t v7,
|
||||
uint32_t v8, uint32_t v9, uint32_t v10, uint32_t v11,
|
||||
@@ -407,6 +408,18 @@ static FORCEINLINE uint64_t __movmsk(__vec32_i1 mask) {
|
||||
return (uint64_t)mask.v;
|
||||
}
|
||||
|
||||
static FORCEINLINE __vec32_i1 __any(__vec32_i1 mask) {
|
||||
return (mask.v!=0);
|
||||
}
|
||||
|
||||
static FORCEINLINE __vec32_i1 __all(__vec32_i1 mask) {
|
||||
return (mask.v==0xFFFFFFFF);
|
||||
}
|
||||
|
||||
static FORCEINLINE __vec32_i1 __none(__vec32_i1 mask) {
|
||||
return (mask.v==0);
|
||||
}
|
||||
|
||||
static FORCEINLINE __vec32_i1 __equal_i1(__vec32_i1 a, __vec32_i1 b) {
|
||||
__vec32_i1 r;
|
||||
r.v = (a.v & b.v) | (~a.v & ~b.v);
|
||||
|
||||
Reference in New Issue
Block a user