Added the following mask tests: __any(), __all(), __none() for all supported targets.

This allows for more efficient code generation of KNC.

examples/intrinsics/generic-16.h

@@ -55,6 +55,7 @@ typedef int64_t __vec1_i64;
 
 struct __vec16_i1 {
     __vec16_i1() { }
+    __vec16_i1(const uint16_t &vv) : v(vv) { }
     __vec16_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,
@@ -342,6 +343,18 @@ static FORCEINLINE uint64_t __movmsk(__vec16_i1 mask) {
     return (uint64_t)mask.v;
 }
 
+static FORCEINLINE __vec16_i1 __any(__vec16_i1 mask) {
+    return (mask.v!=0);
+}
+
+static FORCEINLINE __vec16_i1 __all(__vec16_i1 mask) {
+    return (mask.v==0xFFFF);
+}
+
+static FORCEINLINE __vec16_i1 __none(__vec16_i1 mask) {
+    return (mask.v==0);
+}
+
 static FORCEINLINE __vec16_i1 __equal_i1(__vec16_i1 a, __vec16_i1 b) {
     __vec16_i1 r;
     r.v = (a.v & b.v) | (~a.v & ~b.v);

examples/intrinsics/generic-32.h

@@ -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);

examples/intrinsics/generic-64.h

@@ -55,6 +55,7 @@ typedef int64_t __vec1_i64;
 
 struct __vec64_i1 {
     __vec64_i1() { }
+    __vec64_i1(const uint64_t &vv) : v(vv) { }
     __vec64_i1(uint64_t v0, uint64_t v1, uint64_t v2, uint64_t v3,
                uint64_t v4, uint64_t v5, uint64_t v6, uint64_t v7,
                uint64_t v8, uint64_t v9, uint64_t v10, uint64_t v11,
@@ -532,6 +533,18 @@ static FORCEINLINE uint64_t __movmsk(__vec64_i1 mask) {
     return (uint64_t)mask.v;
 }
 
+static FORCEINLINE __vec64_i1 __any(__vec64_i1 mask) {
+    return (mask.v!=0);
+}
+
+static FORCEINLINE __vec64_i1 __all(__vec64_i1 mask) {
+    return (mask.v==0xFFFFFFFFFFFFFFFF);
+}
+
+static FORCEINLINE __vec64_i1 __none(__vec64_i1 mask) {
+    return (mask.v==0);
+}
+
 static FORCEINLINE __vec64_i1 __equal_i1(__vec64_i1 a, __vec64_i1 b) {
     __vec64_i1 r;
     r.v = (a.v & b.v) | (~a.v & ~b.v);

examples/intrinsics/knc.h

@@ -428,6 +428,18 @@ static FORCEINLINE __vec16_i1 __movmsk(__vec16_i1 mask) {
     return _mm512_kmov(mask);
 }
 
+static FORCEINLINE __vec16_i1 __any(__vec16_i1 mask) {
+    return (mask!=0);
+}
+
+static FORCEINLINE __vec16_i1 __all(__vec16_i1 mask) {
+    return (mask=0xFFFF);
+}
+
+static FORCEINLINE __vec16_i1 __none(__vec16_i1 mask) {
+    return (mask==0);
+}
+
 static FORCEINLINE __vec16_i1 __equal_i1(__vec16_i1 a, __vec16_i1 b) {
     return _mm512_knot( _mm512_kandn(a, b));
 }

examples/intrinsics/knc2x.h

@@ -297,6 +297,18 @@ static FORCEINLINE uint32_t __movmsk(__vec32_i1 mask) {
     return ((m1<<16)|m2);
 }
 
+static FORCEINLINE uint32_t __any(__vec32_i1 mask) {
+    return (mask.m!=0);
+}
+
+static FORCEINLINE uint32_t __all(__vec32_i1 mask) {
+    return (mask.m==0xFFFFFFFF);
+}
+
+static FORCEINLINE uint32_t __none(__vec32_i1 mask) {
+    return (mask.m==0x0);
+}
+
 static FORCEINLINE __vec32_i1 __equal(__vec32_i1 a, __vec32_i1 b) {
     __vec32_i1 ret;
     ret.m16.m1 = _mm512_knot(_mm512_kandn(a.m16.m1, b.m16.m1));

examples/intrinsics/sse4.h

@@ -269,6 +269,18 @@ static FORCEINLINE uint64_t __movmsk(__vec4_i1 mask) {
     return (uint64_t)_mm_movemask_ps(mask.v);
 }
 
+static FORCEINLINE __vec4_i1 __any(__vec4_i1 mask) {
+    return (_mm_movemask_ps(mask.v)!=0);
+}
+
+static FORCEINLINE __vec4_i1 __all(__vec4_i1 mask) {
+    return (_mm_movemask_ps(mask.v)=0xF);
+}
+
+static FORCEINLINE __vec4_i1 __none(__vec4_i1 mask) {
+    return (_mm_movemask_ps(mask.v)==0);
+}
+
 static FORCEINLINE __vec4_i1 __equal_i1(__vec4_i1 a, __vec4_i1 b) {
     return _mm_cmpeq_epi32(_mm_castps_si128(a.v), _mm_castps_si128(b.v));
 }