Merge pull request #741 from Vsevolod-Livinskij/master

Saturation arithmetic.

stdlib.ispc

@@ -4527,6 +4527,42 @@ static inline varying int16 saturating_add(varying int16 a, varying int16 b) {
     return __padds_vi16(a, b);
 }
 
+static inline uniform int32 saturating_add(uniform int32 a, uniform int32 b) {
+    uniform unsigned int32 a_unsig = a, b_unsig = b;
+    uniform unsigned int32 result = a_unsig + b_unsig;
+    a_unsig = (a_unsig >> 31) + INT32_MAX;
+    if ((uniform int32) ((a_unsig ^ b_unsig) | ~(b_unsig ^ result)) >= 0)
+        result = a_unsig;
+    return result;
+}
+
+static inline varying int32 saturating_add(varying int32 a, varying int32 b) {
+    varying unsigned int32 a_unsig = a, b_unsig = b;
+    varying unsigned int32 result = a_unsig + b_unsig;
+    a_unsig = (a_unsig >> 31) + INT32_MAX;
+    if ((varying int32) ((a_unsig ^ b_unsig) | ~(b_unsig ^ result)) >= 0)
+        result = a_unsig;
+    return result;
+}
+
+static inline uniform int64 saturating_add(uniform int64 a, uniform int64 b) {
+    uniform unsigned int64 a_unsig = a, b_unsig = b;
+    uniform unsigned int64 result = a_unsig + b_unsig;
+    a_unsig = (a_unsig >> 63) + INT64_MAX;
+    if ((uniform int64) ((a_unsig ^ b_unsig) | ~(b_unsig ^ result)) >= 0)
+        result = a_unsig;
+    return result;
+}
+
+static inline varying int64 saturating_add(varying int64 a, varying int64 b) {
+    varying unsigned int64 a_unsig = a, b_unsig = b;
+    varying unsigned int64 result = a_unsig + b_unsig;
+    a_unsig = (a_unsig >> 63) + INT64_MAX;
+    if ((varying int64) ((a_unsig ^ b_unsig) | ~(b_unsig ^ result)) >= 0)
+        result = a_unsig;
+    return result;
+}
+
 static inline uniform unsigned int8 saturating_add(uniform unsigned int8 a, 
                                                    uniform unsigned int8 b) {
     uniform unsigned int8 result = a + b;
@@ -4551,6 +4587,34 @@ static inline varying unsigned int16 saturating_add(varying unsigned int16 a,
     return __paddus_vi16(a, b);
 }
 
+static inline uniform unsigned int32 saturating_add(uniform unsigned int32 a, 
+                                                    uniform unsigned int32 b) {
+    uniform unsigned int32 result = a + b;
+    result |= (-(uniform int32)(result < a));
+    return result;
+}
+
+static inline varying unsigned int32 saturating_add(varying unsigned int32 a, 
+                                                    varying unsigned int32 b) {
+    varying unsigned int32 result = a + b;
+    result |= (-(varying int32)(result < a));
+    return result;
+}
+
+static inline uniform unsigned int64 saturating_add(uniform unsigned int64 a, 
+                                                    uniform unsigned int64 b) {
+    uniform unsigned int64 result = a + b;
+    result |= (-(uniform int64)(result < a));
+    return result;
+}
+
+static inline varying unsigned int64 saturating_add(varying unsigned int64 a, 
+                                                    varying unsigned int64 b) {
+    varying unsigned int64 result = a + b;
+    result |= (-(varying int64)(result < a));
+    return result;
+}
+
 static inline uniform int8 saturating_sub(uniform int8 a, uniform int8 b) {
     uniform unsigned int8 a_unsig = a, b_unsig = b;
     uniform unsigned int8 result = a_unsig - b_unsig;
@@ -4577,6 +4641,42 @@ static inline varying int16 saturating_sub(varying int16 a, varying int16 b) {
     return __psubs_vi16(a, b);
 }
 
+static inline uniform int32 saturating_sub(uniform int32 a, uniform int32 b) {
+    uniform unsigned int32 a_unsig = a, b_unsig = b;
+    uniform unsigned int32 result = a_unsig - b_unsig;
+    a_unsig = (a_unsig >> 31) + INT32_MAX;
+    if ((uniform int32) ((a_unsig ^ b_unsig) & (a_unsig ^ result)) < 0)
+        result = a_unsig;
+    return result;
+}
+
+static inline varying int32 saturating_sub(varying int32 a, varying int32 b) {
+    varying unsigned int32 a_unsig = a, b_unsig = b;
+    varying unsigned int32 result = a_unsig - b_unsig;
+    a_unsig = (a_unsig >> 31) + INT32_MAX;
+    if ((varying int32) ((a_unsig ^ b_unsig) & (a_unsig ^ result)) < 0)
+        result = a_unsig;
+    return result;
+}
+
+static inline uniform int64 saturating_sub(uniform int64 a, uniform int64 b) {
+    uniform unsigned int64 a_unsig = a, b_unsig = b;
+    uniform unsigned int64 result = a_unsig - b_unsig;
+    a_unsig = (a_unsig >> 63) + INT64_MAX;
+    if ((uniform int64) ((a_unsig ^ b_unsig) & (a_unsig ^ result)) < 0)
+        result = a_unsig;
+    return result;
+}
+
+static inline varying int64 saturating_sub(varying int64 a, varying int64 b) {
+    varying unsigned int64 a_unsig = a, b_unsig = b;
+    varying unsigned int64 result = a_unsig - b_unsig;
+    a_unsig = (a_unsig >> 63) + INT64_MAX;
+    if ((varying int64) ((a_unsig ^ b_unsig) & (a_unsig ^ result)) < 0)
+        result = a_unsig;
+    return result;
+}
+
 static inline uniform unsigned int8 saturating_sub(uniform unsigned int8 a, 
                                                    uniform unsigned int8 b) {
     uniform unsigned int8 result = a - b;
@@ -4600,6 +4700,244 @@ static inline varying unsigned int16 saturating_sub(varying unsigned int16 a,
                                                     varying unsigned int16 b) {
     return __psubus_vi16(a, b);
 }
+
+static inline uniform unsigned int32 saturating_sub(uniform unsigned int32 a, 
+                                                    uniform unsigned int32 b) {
+    uniform unsigned int32 result = a - b;
+    result &= (-(uniform int32)(result <= a));
+    return result;
+}
+
+static inline varying unsigned int32 saturating_sub(varying unsigned int32 a, 
+                                                    varying unsigned int32 b) {
+    varying unsigned int32 result = a - b;
+    result &= (-(varying int32)(result <= a));
+    return result;
+}
+
+static inline uniform unsigned int64 saturating_sub(uniform unsigned int64 a, 
+                                                    uniform unsigned int64 b) {
+    uniform unsigned int64 result = a - b;
+    result &= (-(uniform int64)(result <= a));
+    return result;
+}
+
+static inline varying unsigned int64 saturating_sub(varying unsigned int64 a, 
+                                                    varying unsigned int64 b) {
+    varying unsigned int64 result = a - b;
+    result &= (-(varying int64)(result <= a));
+    return result;
+}
+
+static inline uniform int8 saturating_div(uniform int8 a, uniform int8 b) {
+    /* Only one way to overflow, so test for and prevent it. */
+    a += !((b + 1) | ((uniform unsigned int8) a + INT8_MIN));
+    return a / b;
+}
+
+static inline varying int8 saturating_div(varying int8 a, varying int8 b) {
+    /* Only one way to overflow, so test for and prevent it. */
+    a += !((b + 1) | ((varying unsigned int8) a + INT8_MIN));
+    return a / b;
+}
+
+static inline uniform int16 saturating_div(uniform int16 a, uniform int16 b) {
+    /* Only one way to overflow, so test for and prevent it. */
+    a += !((b + 1) | ((uniform unsigned int16) a + INT16_MIN));
+    return a / b;
+}
+
+static inline varying int16 saturating_div(varying int16 a, varying int16 b) {
+    /* Only one way to overflow, so test for and prevent it. */
+    a += !((b + 1) | ((varying unsigned int16) a + INT16_MIN));
+    return a / b;
+}
+
+static inline uniform int32 saturating_div(uniform int32 a, uniform int32 b) {
+    /* Only one way to overflow, so test for and prevent it. */
+    a += !((b + 1) | ((uniform unsigned int32) a + INT32_MIN));
+    return a / b;
+}
+
+static inline varying int32 saturating_div(varying int32 a, varying int32 b) {
+    /* Only one way to overflow, so test for and prevent it. */
+    a += !((b + 1) | ((varying unsigned int32) a + INT32_MIN));
+    return a / b;
+}
+
+static inline uniform int64 saturating_div(uniform int64 a, uniform int64 b) {
+    /* Only one way to overflow, so test for and prevent it. */
+    a += !((b + 1) | ((uniform unsigned int64) a + INT64_MIN));
+    return a / b;
+}
+
+static inline varying int64 saturating_div(varying int64 a, varying int64 b) {
+    /* Only one way to overflow, so test for and prevent it. */
+    a += !((b + 1) | ((varying unsigned int64) a + INT64_MIN));
+    return a / b;
+}
+
+static inline uniform unsigned int8 saturating_div(uniform unsigned int8 a,
+	                                               uniform unsigned int8 b) {
+    /* No overflow possible */
+    return a / b;
+}
+
+static inline varying unsigned int8 saturating_div(varying unsigned int8 a, 
+                                                   varying unsigned int8 b) {
+    /* No overflow possible */
+    return a / b;
+}
+
+static inline uniform unsigned int16 saturating_div(uniform unsigned int16 a,
+                                                    uniform unsigned int16 b) {
+    /* No overflow possible */
+    return a / b;
+}
+
+static inline varying unsigned int16 saturating_div(varying unsigned int16 a, 
+                                                    varying unsigned int16 b) {
+    /* No overflow possible */
+    return a / b;
+}
+
+static inline uniform unsigned int32 saturating_div(uniform unsigned int32 a, 
+                                                    uniform unsigned int32 b) {
+    /* No overflow possible */
+    return a / b;
+}
+
+static inline varying unsigned int32 saturating_div(varying unsigned int32 a,
+                                                    varying unsigned int32 b) {
+    /* No overflow possible */
+    return a / b;
+}
+
+static inline uniform unsigned int64 saturating_div(uniform unsigned int64 a, 
+                                                    uniform unsigned int64 b) {
+    /* No overflow possible */
+    return a / b;
+}
+
+static inline varying unsigned int64 saturating_div(varying unsigned int64 a, 
+                                                    varying unsigned int64 b) {
+    /* No overflow possible */
+    return a / b;
+}
+
+static inline uniform int8 saturating_mul(uniform int8 a, uniform int8 b) {
+    uniform int16 result = (uniform int16) a * (uniform int16) b;	
+    uniform unsigned int8 result2 = ((uniform unsigned int8) (a ^ b) >> 7) + INT8_MAX;
+    uniform int8 hi = result >> 8;
+    uniform int8 lo = result;
+    if (hi != (lo >> 7)) 
+        result = result2;
+    return result; 
+}
+
+static inline varying int8 saturating_mul(varying int8 a, varying int8 b) {
+    varying int16 result = (varying int16) a * (varying int16) b;	
+    varying unsigned int8 result2 = ((varying unsigned int8) (a ^ b) >> 7) + INT8_MAX;
+    varying int8 hi = result >> 8;
+    varying int8 lo = result;
+    if (hi != (lo >> 7)) 
+        result = result2;
+    return result; 
+}
+
+static inline uniform int16 saturating_mul(uniform int16 a, uniform int16 b) {
+    uniform int32 result = (uniform int32) a * (uniform int32) b;	
+    uniform unsigned int16 result2 = ((uniform unsigned int16) (a ^ b) >> 15) + INT16_MAX;
+    uniform int16 hi = result >> 16;
+    uniform int16 lo = result;
+    if (hi != (lo >> 15)) 
+        result = result2;
+    return result; 
+}
+
+static inline varying int16 saturating_mul(varying int16 a, varying int16 b) {
+    varying int32 result = (varying int32) a * (varying int32) b;	
+    varying unsigned int16 result2 = ((varying unsigned int16) (a ^ b) >> 15) + INT16_MAX;
+    varying int16 hi = result >> 16;
+    varying int16 lo = result;
+    if (hi != (lo >> 15)) 
+        result = result2;
+    return result; 
+}
+
+static inline uniform int32 saturating_mul(uniform int32 a, uniform int32 b) {
+    uniform int64 result = (uniform int64) a * (uniform int64) b;	
+    uniform unsigned int32 result2 = ((uniform unsigned int32) (a ^ b) >> 31) + INT32_MAX;
+    uniform int32 hi = result >> 32;
+    uniform int32 lo = result;
+    if (hi != (lo >> 31)) 
+        result = result2;
+    return result; 
+}
+
+static inline varying int32 saturating_mul(varying int32 a, varying int32 b) {
+    varying int64 result = (varying int64) a * (varying int64) b;	
+    varying unsigned int32 result2 = ((varying unsigned int32) (a ^ b) >> 31) + INT32_MAX;
+    varying int32 hi = result >> 32;
+    varying int32 lo = result;
+    if (hi != (lo >> 31)) 
+        result = result2;
+    return result; 
+}
+
+static inline uniform unsigned int8 saturating_mul(uniform unsigned int8 a,
+                                                   uniform unsigned int8 b) {
+    uniform unsigned int16 result = (uniform unsigned int16) a * 
+                                    (uniform unsigned int16) b;
+    uniform unsigned int8 hi = result >> 8;
+    uniform unsigned int8 lo = result;
+    return lo | - (uniform int8) !! hi;
+}
+
+static inline varying unsigned int8 saturating_mul(varying unsigned int8 a,
+                                                   varying unsigned int8 b) {
+    varying unsigned int16 result = (varying unsigned int16) a * 
+                                    (varying unsigned int16) b;
+    varying unsigned int8 hi = result >> 8;
+    varying unsigned int8 lo = result;
+    return lo | - (varying int8) !! hi;
+}
+
+static inline uniform unsigned int16 saturating_mul(uniform unsigned int16 a,
+                                                    uniform unsigned int16 b) {
+    uniform unsigned int32 result = (uniform unsigned int32) a * 
+                                    (uniform unsigned int32) b;
+    uniform unsigned int16 hi = result >> 16;
+    uniform unsigned int16 lo = result;
+    return lo | - (uniform int16) !! hi;
+}
+
+static inline varying unsigned int16 saturating_mul(varying unsigned int16 a,
+                                                    varying unsigned int16 b) {
+    varying unsigned int32 result = (varying unsigned int32) a * 
+                                    (varying unsigned int32) b;
+    varying unsigned int16 hi = result >> 16;
+    varying unsigned int16 lo = result;
+    return lo | - (varying int16) !! hi;
+}
+
+static inline uniform unsigned int32 saturating_mul(uniform unsigned int32 a,
+                                                    uniform unsigned int32 b) {
+    uniform unsigned int64 result = (uniform unsigned int64) a * 
+                                    (uniform unsigned int64) b;
+    uniform unsigned int32 hi = result >> 32;
+    uniform unsigned int32 lo = result;
+    return lo | - (uniform int32) !! hi;
+}
+
+static inline varying unsigned int32 saturating_mul(varying unsigned int32 a,
+                                                    varying unsigned int32 b) {
+    varying unsigned int64 result = (varying unsigned int64) a * 
+                                    (varying unsigned int64) b;
+    varying unsigned int32 hi = result >> 32;
+    varying unsigned int32 lo = result;
+    return lo | - (varying int32) !! hi;
+}
 ///////////////////////////////////////////////////////////////////////////
 // rdrand
 

tests/padds_i32.ispc

@@ -0,0 +1,27 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    uniform int32 a_max = 0x7FFFFFFF, a_min = -0x80000000; // max and min signed int32
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = saturating_add(a_max, b);
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = saturating_add(a_min, -b);
+    }
+    else {
+        RET[programIndex] = saturating_add(a_min, b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = (uniform int32) 0x7FFFFFFF; // max signed int32
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = (uniform int32) -0x80000000; // min signed int32
+    }
+    else {
+        RET[programIndex] = (uniform int32) -0x7FFFFFFB; // min + 5
+    } 
+}

tests/padds_i64.ispc

@@ -0,0 +1,27 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    uniform int64 a_max = 0x7FFFFFFFFFFFFFFF, a_min = -0x8000000000000000; // max and min signed int64
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = saturating_add(a_max, b);
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = saturating_add(a_min, -b);
+    }
+    else {
+        RET[programIndex] = saturating_add(a_min, b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = (uniform int64) 0x7FFFFFFFFFFFFFFF; // max signed int64
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = (uniform int64) -0x8000000000000000; // min signed int64
+    }
+    else {
+        RET[programIndex] = (uniform int64) -0x7FFFFFFFFFFFFFFB; // min + 5
+    } 
+}

tests/padds_vi32.ispc

@@ -0,0 +1,27 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    varying int32 a_max = 0x7FFFFFFF, a_min = -0x80000000; // max and min signed int32
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = saturating_add(a_max, b);
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = saturating_add(a_min, -b);
+    }
+    else {
+        RET[programIndex] = saturating_add(a_min, b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = (varying int32) 0x7FFFFFFF; // max signed int32
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = (varying int32) -0x80000000; // min signed int32
+    }
+    else {
+        RET[programIndex] = (varying int32) -0x7FFFFFFB; // min + 5
+    } 
+}

tests/padds_vi64.ispc

@@ -0,0 +1,27 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    varying int64 a_max = 0x7FFFFFFFFFFFFFFF, a_min = -0x8000000000000000; // max and min signed int64
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = saturating_add(a_max, b);
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = saturating_add(a_min, -b);
+    }
+    else {
+        RET[programIndex] = saturating_add(a_min, b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = (varying int64) 0x7FFFFFFFFFFFFFFF; // max signed int64
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = (varying int64) -0x8000000000000000; // min signed int64
+    }
+    else {
+        RET[programIndex] = (varying int64) -0x7FFFFFFFFFFFFFFB; // min + 5
+    } 
+}

tests/paddus_i32.ispc

@@ -0,0 +1,21 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    uniform unsigned int32 a_max = 0xFFFFFFFF, a_min = 0; // max and min unsigned int32
+    if (programIndex % 2 == 0) {
+        RET[programIndex] = saturating_add(a_max, b);
+    }
+    else {
+        RET[programIndex] = saturating_add(a_min, b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 2 == 0) {
+        RET[programIndex] = (uniform unsigned int32) 0xFFFFFFFF; // max unsigned int32
+    }
+    else {
+        RET[programIndex] = (uniform unsigned int32) 5;
+    } 
+}

tests/paddus_i64.ispc

@@ -0,0 +1,21 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    uniform unsigned int64 a_max = 0xFFFFFFFFFFFFFFFF, a_min = 0; // max and min unsigned int64
+    if (programIndex % 2 == 0) {
+        RET[programIndex] = saturating_add(a_max, b);
+    }
+    else {
+        RET[programIndex] = saturating_add(a_min, b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 2 == 0) {
+        RET[programIndex] = (uniform unsigned int64) 0xFFFFFFFFFFFFFFFF; // max unsigned int64
+    }
+    else {
+        RET[programIndex] = (uniform unsigned int64) 5;
+    } 
+}

tests/paddus_vi32.ispc

@@ -0,0 +1,21 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    varying unsigned int32 a_max = 0xFFFFFFFF, a_min = 0; // max and min unsigned int32
+    if (programIndex % 2 == 0) {
+        RET[programIndex] = saturating_add(a_max, b);
+    }
+    else {
+        RET[programIndex] = saturating_add(a_min, b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 2 == 0) {
+        RET[programIndex] = (varying unsigned int32) 0xFFFFFFFF; // max unsigned int32
+    }
+    else {
+        RET[programIndex] = (varying unsigned int32) 5;
+    } 
+}

tests/paddus_vi64.ispc

@@ -0,0 +1,21 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    varying unsigned int64 a_max = 0xFFFFFFFFFFFFFFFF, a_min = 0; // max and min unsigned int64
+    if (programIndex % 2 == 0) {
+        RET[programIndex] = saturating_add(a_max, b);
+    }
+    else {
+        RET[programIndex] = saturating_add(a_min, b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 2 == 0) {
+        RET[programIndex] = (varying unsigned int64) 0xFFFFFFFFFFFFFFFF; // max unsigned int64
+    }
+    else {
+        RET[programIndex] = (varying unsigned int64) 5;
+    } 
+}

tests/pdivs_i16.ispc

@@ -0,0 +1,27 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    uniform int16 a_max = 0x7FFF, a_min = -0x8000; // max and min signed int16
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = saturating_div(a_max, (uniform int16) b);
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = saturating_div(a_min, (uniform int16) b);
+    }
+    else {
+        RET[programIndex] = saturating_div(a_min, (uniform int16) -1);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = (uniform int16) 6553.4; // max / 5
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = (uniform int16) -6553.6; // min / 5 
+    }
+    else {
+        RET[programIndex] = (uniform int16) 0x7FFF; // max signed int16
+    } 
+}

tests/pdivs_i32.ispc

@@ -0,0 +1,27 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    uniform int32 a_max = 0x7FFFFFFF, a_min = -0x80000000; // max and min signed int32
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = saturating_div(a_max, (uniform int32) b);
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = saturating_div(a_min, (uniform int32) b);
+    }
+    else {
+        RET[programIndex] = saturating_div(a_min, (uniform int32) -1);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = (uniform int32) 429496729.4; // max / 5
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = (uniform int32) -429496729.6; // min / 5 
+    }
+    else {
+        RET[programIndex] = (uniform int32) 0x7FFFFFFF; // max signed int32
+    } 
+}

tests/pdivs_i64.ispc

@@ -0,0 +1,27 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    uniform int64 a_max = 0x7FFFFFFFFFFFFFFF, a_min = -0x8000000000000000; // max and min signed int64
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = saturating_div(a_max, (uniform int64) b);
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = saturating_div(a_min, (uniform int64) b);
+    }
+    else {
+        RET[programIndex] = saturating_div(a_min, (uniform int64) -1);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = (uniform int64) 1844674407370955161.4; // max / 5
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = (uniform int64) -1844674407370955161.6; // min / 5 
+    }
+    else {
+        RET[programIndex] = (uniform int64) 0x7FFFFFFFFFFFFFFF; // max signed int64
+    } 
+}

tests/pdivs_i8.ispc

@@ -0,0 +1,27 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    uniform int8 a_max = 0x7F, a_min = -0x80; // max and min signed int8
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = saturating_div(a_max, (uniform int8) b);
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = saturating_div(a_min, (uniform int8) b);
+    }
+    else {
+        RET[programIndex] = saturating_div(a_min, (uniform int8) -1);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = (uniform int8) 25.4; // max / 5
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = (uniform int8) -25.6; // min / 5 
+    }
+    else {
+        RET[programIndex] = (uniform int8) 0x7F; // max signed int8
+    } 
+}

tests/pdivs_vi16.ispc

@@ -0,0 +1,27 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    varying int16 a_max = 0x7FFF, a_min = -0x8000; // max and min signed int16
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = saturating_div(a_max, (varying int16) b);
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = saturating_div(a_min, (varying int16) b);
+    }
+    else {
+        RET[programIndex] = saturating_div(a_min, (varying int16) -1);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = (varying int16) 6553.4; // max / 5
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = (varying int16) -6553.6; // min / 5 
+    }
+    else {
+        RET[programIndex] = (varying int16) 0x7FFF; // max signed int16
+    } 
+}

tests/pdivs_vi32.ispc

@@ -0,0 +1,27 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    varying int32 a_max = 0x7FFFFFFF, a_min = -0x80000000; // max and min signed int32
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = saturating_div(a_max, (varying int32) b);
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = saturating_div(a_min, (varying int32) b);
+    }
+    else {
+        RET[programIndex] = saturating_div(a_min, (varying int32) -1);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = (varying int32) 429496729.4; // max / 5
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = (varying int32) -429496729.6; // min / 5 
+    }
+    else {
+        RET[programIndex] = (varying int32) 0x7FFFFFFF; // max signed int32
+    } 
+}

tests/pdivs_vi64.ispc

@@ -0,0 +1,27 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    varying int64 a_max = 0x7FFFFFFFFFFFFFFF, a_min = -0x8000000000000000; // max and min signed int64
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = saturating_div(a_max, (varying int64) b);
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = saturating_div(a_min, (varying int64) b);
+    }
+    else {
+        RET[programIndex] = saturating_div(a_min, (varying int64) -1);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = (varying int64) 1844674407370955161.4; // max / 5
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = (varying int64) -1844674407370955161.6; // min / 5 
+    }
+    else {
+        RET[programIndex] = (varying int64) 0x7FFFFFFFFFFFFFFF; // max signed int64
+    } 
+}

tests/pdivs_vi8.ispc

@@ -0,0 +1,27 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    varying int8 a_max = 0x7F, a_min = -0x80; // max and min signed int8
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = saturating_div(a_max, (varying int8) b);
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = saturating_div(a_min, (varying int8) b);
+    }
+    else {
+        RET[programIndex] = saturating_div(a_min, (varying int8) -1);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = (varying int8) 25.4; // max / 5
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = (varying int8) -25.6; // min / 5 
+    }
+    else {
+        RET[programIndex] = (varying int8) 0x7F; // max signed int8
+    } 
+}

tests/pdivus_i16.ispc

@@ -0,0 +1,21 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    uniform unsigned int16 a_max = 0xFFFF, a_min = 0; // max and min unsigned int16
+    if (programIndex % 2 == 0) {
+        RET[programIndex] = saturating_div(a_min, (uniform unsigned int16) b);
+    }
+    else {
+        RET[programIndex] = saturating_div(a_max, (uniform unsigned int16) b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 2 == 0) {
+        RET[programIndex] = (uniform unsigned int16) 0; // min unsigned int16
+    }
+    else {
+        RET[programIndex] = (uniform unsigned int16) 0x3333; // max unsigned int16 / 5
+    } 
+}

tests/pdivus_i32.ispc

@@ -0,0 +1,21 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    uniform unsigned int32 a_max = 0xFFFFFFFF, a_min = 0; // max and min unsigned int32
+    if (programIndex % 2 == 0) {
+        RET[programIndex] = saturating_div(a_min, (uniform unsigned int32) b);
+    }
+    else {
+        RET[programIndex] = saturating_div(a_max, (uniform unsigned int32) b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 2 == 0) {
+        RET[programIndex] = (uniform unsigned int32) 0; // min unsigned int32
+    }
+    else {
+        RET[programIndex] = (uniform unsigned int32) 0x33333333; // max unsigned int32 / 5
+    } 
+}

tests/pdivus_i64.ispc

@@ -0,0 +1,21 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    uniform unsigned int64 a_max = 0xFFFFFFFFFFFFFFFF, a_min = 0; // max and min unsigned int64
+    if (programIndex % 2 == 0) {
+        RET[programIndex] = saturating_div(a_min, (uniform unsigned int64) b);
+    }
+    else {
+        RET[programIndex] = saturating_div(a_max, (uniform unsigned int64) b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 2 == 0) {
+        RET[programIndex] = (uniform unsigned int64) 0; // min unsigned int64
+    }
+    else {
+        RET[programIndex] = (uniform unsigned int64) 0x3333333333333333; // max unsigned int64 / 5
+    } 
+}

tests/pdivus_i8.ispc

@@ -0,0 +1,21 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    uniform unsigned int8 a_max = 0xFF, a_min = 0; // max and min unsigned int8
+    if (programIndex % 2 == 0) {
+        RET[programIndex] = saturating_div(a_min, (uniform unsigned int8) b);
+    }
+    else {
+        RET[programIndex] = saturating_div(a_max, (uniform unsigned int8) b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 2 == 0) {
+        RET[programIndex] = (uniform unsigned int8) 0; // min unsigned int8
+    }
+    else {
+        RET[programIndex] = (uniform unsigned int8) 0x33; // max unsigned int8 / 5
+    } 
+}

tests/pdivus_vi16.ispc

@@ -0,0 +1,21 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    varying unsigned int16 a_max = 0xFFFF, a_min = 0; // max and min unsigned int16
+    if (programIndex % 2 == 0) {
+        RET[programIndex] = saturating_div(a_min, (varying unsigned int16) b);
+    }
+    else {
+        RET[programIndex] = saturating_div(a_max, (varying unsigned int16) b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 2 == 0) {
+        RET[programIndex] = (varying unsigned int16) 0; // min unsigned int16
+    }
+    else {
+        RET[programIndex] = (varying unsigned int16) 0x3333; // max unsigned int16 / 5
+    } 
+}

tests/pdivus_vi32.ispc

@@ -0,0 +1,21 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    varying unsigned int32 a_max = 0xFFFFFFFF, a_min = 0; // max and min unsigned int32
+    if (programIndex % 2 == 0) {
+        RET[programIndex] = saturating_div(a_min, (varying unsigned int32) b);
+    }
+    else {
+        RET[programIndex] = saturating_div(a_max, (varying unsigned int32) b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 2 == 0) {
+        RET[programIndex] = (varying unsigned int32) 0; // min unsigned int32
+    }
+    else {
+        RET[programIndex] = (varying unsigned int32) 0x33333333; // max unsigned int32 / 5
+    } 
+}

tests/pdivus_vi64.ispc

@@ -0,0 +1,21 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    varying unsigned int64 a_max = 0xFFFFFFFFFFFFFFFF, a_min = 0; // max and min unsigned int64
+    if (programIndex % 2 == 0) {
+        RET[programIndex] = saturating_div(a_min, (varying unsigned int64) b);
+    }
+    else {
+        RET[programIndex] = saturating_div(a_max, (varying unsigned int64) b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 2 == 0) {
+        RET[programIndex] = (varying unsigned int64) 0; // min unsigned int64
+    }
+    else {
+        RET[programIndex] = (varying unsigned int64) 0x3333333333333333; // max unsigned int64 / 5
+    } 
+}

tests/pdivus_vi8.ispc

@@ -0,0 +1,21 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    varying unsigned int8 a_max = 0xFF, a_min = 0; // max and min unsigned int8
+    if (programIndex % 2 == 0) {
+        RET[programIndex] = saturating_div(a_min, (varying unsigned int8) b);
+    }
+    else {
+        RET[programIndex] = saturating_div(a_max, (varying unsigned int8) b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 2 == 0) {
+        RET[programIndex] = (varying unsigned int8) 0; // min unsigned int8
+    }
+    else {
+        RET[programIndex] = (varying unsigned int8) 0x33; // max unsigned int8 / 5
+    } 
+}

tests/pmuls_i16.ispc

@@ -0,0 +1,28 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    uniform int16 a_max = 0x7FFF, a_min = -0x8000; // max and min signed int16
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = saturating_mul(a_max, (uniform int16) b);
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = saturating_mul(a_min, (uniform int16) b);
+    }
+    else {
+        RET[programIndex] = saturating_mul((uniform int16) b,
+                                           (uniform int16) b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = (uniform int16) 0x7FFF; // max signed int16
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = (uniform int16) -0x8000; // min signed int16
+    }
+    else {
+        RET[programIndex] = (uniform int16) 25;
+    } 
+}

tests/pmuls_i32.ispc

@@ -0,0 +1,28 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    uniform int32 a_max = 0x7FFFFFFF, a_min = -0x80000000; // max and min signed int32
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = saturating_mul(a_max, (uniform int32) b);
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = saturating_mul(a_min, (uniform int32) b);
+    }
+    else {
+        RET[programIndex] = saturating_mul((uniform int32) b,
+                                           (uniform int32) b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = (uniform int32) 0x7FFFFFFF; // max signed int32
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = (uniform int32) -0x80000000; // min signed int32
+    }
+    else {
+        RET[programIndex] = (uniform int32) 25;
+    } 
+}

tests/pmuls_i8.ispc

@@ -0,0 +1,28 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    uniform int8 a_max = 0x7F, a_min = -0x80; // max and min signed int8
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = saturating_mul(a_max, (uniform int8) b);
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = saturating_mul(a_min, (uniform int8) b);
+    }
+    else {
+        RET[programIndex] = saturating_mul((uniform int8) b,
+                                           (uniform int8) b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = (uniform int8) 0x7F; // max signed int8
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = (uniform int8) -0x80; // min signed int8
+    }
+    else {
+        RET[programIndex] = (uniform int8) 25;
+    } 
+}

tests/pmuls_vi16.ispc

@@ -0,0 +1,28 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    varying int16 a_max = 0x7FFF, a_min = -0x8000; // max and min signed int16
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = saturating_mul(a_max, (varying int16) b);
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = saturating_mul(a_min, (varying int16) b);
+    }
+    else {
+        RET[programIndex] = saturating_mul((varying int16) b,
+                                           (varying int16) b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = (varying int16) 0x7FFF; // max signed int16
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = (varying int16) -0x8000; // min signed int16
+    }
+    else {
+        RET[programIndex] = (varying int16) 25;
+    } 
+}

tests/pmuls_vi32.ispc

@@ -0,0 +1,28 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    varying int32 a_max = 0x7FFFFFFF, a_min = -0x80000000; // max and min signed int32
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = saturating_mul(a_max, (varying int32) b);
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = saturating_mul(a_min, (varying int32) b);
+    }
+    else {
+        RET[programIndex] = saturating_mul((varying int32) b,
+                                           (varying int32) b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = (varying int32) 0x7FFFFFFF; // max signed int32
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = (varying int32) -0x80000000; // min signed int32
+    }
+    else {
+        RET[programIndex] = (varying int32) 25;
+    } 
+}

tests/pmuls_vi8.ispc

@@ -0,0 +1,28 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    varying int8 a_max = 0x7F, a_min = -0x80; // max and min signed int8
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = saturating_mul(a_max, (varying int8) b);
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = saturating_mul(a_min, (varying int8) b);
+    }
+    else {
+        RET[programIndex] = saturating_mul((varying int8) b,
+                                           (varying int8) b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = (varying int8) 0x7F; // max signed int8
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = (varying int8) -0x80; // min signed int8
+    }
+    else {
+        RET[programIndex] = (varying int8) 25;
+    } 
+}

tests/pmulus_i16.ispc

@@ -0,0 +1,28 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    uniform unsigned int16 a_max = 0xFFFF, a_min = 0; // max and min unsigned int16
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = saturating_mul(a_max, (uniform unsigned int16) b);
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = saturating_mul(a_min, (uniform unsigned int16) -b);
+    }
+    else {
+        RET[programIndex] = saturating_mul((uniform unsigned int16) b,
+                                           (uniform unsigned int16) b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = (uniform unsigned int16) 0xFFFF; // max unsigned int16
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = (uniform unsigned int16) 0; // min unsigned int16
+    }
+    else {
+        RET[programIndex] = (uniform unsigned int16) 25;
+    } 
+}

tests/pmulus_i32.ispc

@@ -0,0 +1,28 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    uniform unsigned int32 a_max = 0xFFFFFFFF, a_min = 0; // max and min unsigned int32
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = saturating_mul(a_max, (uniform unsigned int32) b);
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = saturating_mul(a_min, (uniform unsigned int32) -b);
+    }
+    else {
+        RET[programIndex] = saturating_mul((uniform unsigned int32) b,
+                                           (uniform unsigned int32) b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = (uniform unsigned int32) 0xFFFFFFFF; // max unsigned int32
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = (uniform unsigned int32) 0; // min unsigned int32
+    }
+    else {
+        RET[programIndex] = (uniform unsigned int32) 25;
+    } 
+}

tests/pmulus_i8.ispc

@@ -0,0 +1,28 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    uniform unsigned int8 a_max = 0xFF, a_min = 0; // max and min unsigned int8
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = saturating_mul(a_max, (uniform unsigned int8) b);
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = saturating_mul(a_min, (uniform unsigned int8) -b);
+    }
+    else {
+        RET[programIndex] = saturating_mul((uniform unsigned int8) b,
+                                           (uniform unsigned int8) b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = (uniform unsigned int8) 0xFF; // max unsigned int8
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = (uniform unsigned int8) 0; // min unsigned int8
+    }
+    else {
+        RET[programIndex] = (uniform unsigned int8) 25;
+    } 
+}

tests/pmulus_vi16.ispc

@@ -0,0 +1,28 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    varying unsigned int16 a_max = 0xFFFF, a_min = 0; // max and min unsigned int16
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = saturating_mul(a_max, (varying unsigned int16) b);
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = saturating_mul(a_min, (varying unsigned int16) -b);
+    }
+    else {
+        RET[programIndex] = saturating_mul((varying unsigned int16) b,
+                                           (varying unsigned int16) b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = (varying unsigned int16) 0xFFFF; // max unsigned int16
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = (varying unsigned int16) 0; // min unsigned int16
+    }
+    else {
+        RET[programIndex] = (varying unsigned int16) 25;
+    } 
+}

tests/pmulus_vi32.ispc

@@ -0,0 +1,28 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    varying unsigned int32 a_max = 0xFFFFFFFF, a_min = 0; // max and min unsigned int32
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = saturating_mul(a_max, (varying unsigned int32) b);
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = saturating_mul(a_min, (varying unsigned int32) -b);
+    }
+    else {
+        RET[programIndex] = saturating_mul((varying unsigned int32) b,
+                                           (varying unsigned int32) b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = (varying unsigned int32) 0xFFFFFFFF; // max unsigned int32
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = (varying unsigned int32) 0; // min unsigned int32
+    }
+    else {
+        RET[programIndex] = (varying unsigned int32) 25;
+    } 
+}

tests/pmulus_vi8.ispc

@@ -0,0 +1,28 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    varying unsigned int8 a_max = 0xFF, a_min = 0; // max and min unsigned int8
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = saturating_mul(a_max, (varying unsigned int8) b);
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = saturating_mul(a_min, (varying unsigned int8) -b);
+    }
+    else {
+        RET[programIndex] = saturating_mul((varying unsigned int8) b,
+                                           (varying unsigned int8) b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = (varying unsigned int8) 0xFF; // max unsigned int8
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = (varying unsigned int8) 0; // min unsigned int8
+    }
+    else {
+        RET[programIndex] = (varying unsigned int8) 25;
+    } 
+}

tests/psubs_i32.ispc

@@ -0,0 +1,27 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    uniform int32 a_max = 0x7FFFFFFF, a_min = -0x80000000; // max and min int32
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = saturating_sub(a_min, b);
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = saturating_sub(a_max, -b);
+    }
+    else {
+        RET[programIndex] = saturating_sub(a_max, b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = (uniform int32) -0x80000000; // min signed int32
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = (uniform int32) 0x7FFFFFFF; // max signed int32
+    }
+    else {
+        RET[programIndex] = (uniform int32) 0x7FFFFFFA;
+    } 
+}

tests/psubs_i64.ispc

@@ -0,0 +1,27 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    uniform int64 a_max = 0x7FFFFFFFFFFFFFFF, a_min = -0x8000000000000000; // max and min int64
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = saturating_sub(a_min, b);
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = saturating_sub(a_max, -b);
+    }
+    else {
+        RET[programIndex] = saturating_sub(a_max, b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = (uniform int64) -0x8000000000000000; // min signed int64
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = (uniform int64) 0x7FFFFFFFFFFFFFFF; // max signed int64
+    }
+    else {
+        RET[programIndex] = (uniform int64) 0x7FFFFFFFFFFFFFFA;
+    } 
+}

tests/psubs_vi32.ispc

@@ -0,0 +1,27 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    varying int32 a_max = 0x7FFFFFFF, a_min = -0x80000000; // max and min int32
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = saturating_sub(a_min, b);
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = saturating_sub(a_max, -b);
+    }
+    else {
+        RET[programIndex] = saturating_sub(a_max, b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = (varying int32) -0x80000000; // min signed int32
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = (varying int32) 0x7FFFFFFF; // max signed int32
+    }
+    else {
+        RET[programIndex] = (varying int32) 0x7FFFFFFA;
+    } 
+}

tests/psubs_vi64.ispc

@@ -0,0 +1,27 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    varying int64 a_max = 0x7FFFFFFFFFFFFFFF, a_min = -0x8000000000000000; // max and min int64
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = saturating_sub(a_min, b);
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = saturating_sub(a_max, -b);
+    }
+    else {
+        RET[programIndex] = saturating_sub(a_max, b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 3 == 0) {
+        RET[programIndex] = (varying int64) -0x8000000000000000; // min signed int64
+    }
+    else if (programIndex % 3 == 1) {
+        RET[programIndex] = (varying int64) 0x7FFFFFFFFFFFFFFF; // max signed int64
+    }
+    else {
+        RET[programIndex] = (varying int64) 0x7FFFFFFFFFFFFFFA;
+    } 
+}

tests/psubus_i32.ispc

@@ -0,0 +1,21 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    uniform unsigned int32 a_max = 0xFFFFFFFF, a_min = 0; // max and min unsigned int32
+    if (programIndex % 2 == 0) {
+        RET[programIndex] = saturating_sub(a_min, b);
+    }
+    else {
+        RET[programIndex] = saturating_sub(a_max, b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 2 == 0) {
+        RET[programIndex] = (uniform unsigned int32) 0; // min unsigned int32
+    }
+    else {
+        RET[programIndex] = (uniform unsigned int32) 0xFFFFFFFB; // max unsigned int32 - 5
+    } 
+}

tests/psubus_i64.ispc

@@ -0,0 +1,21 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    uniform unsigned int64 a_max = 0xFFFFFFFFFFFFFFFF, a_min = 0; // max and min unsigned int64
+    if (programIndex % 2 == 0) {
+        RET[programIndex] = saturating_sub(a_min, b);
+    }
+    else {
+        RET[programIndex] = saturating_sub(a_max, b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 2 == 0) {
+        RET[programIndex] = (uniform unsigned int64) 0; // min unsigned int64
+    }
+    else {
+        RET[programIndex] = (uniform unsigned int64) 0xFFFFFFFFFFFFFFFB; // max unsigned int64 - 5
+    } 
+}

tests/psubus_vi32.ispc

@@ -0,0 +1,21 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    varying unsigned int32 a_max = 0xFFFFFFFF, a_min = 0; // max and min unsigned int32
+    if (programIndex % 2 == 0) {
+        RET[programIndex] = saturating_sub(a_min, b);
+    }
+    else {
+        RET[programIndex] = saturating_sub(a_max, b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 2 == 0) {
+        RET[programIndex] = (varying unsigned int32) 0; // min unsigned int32
+    }
+    else {
+        RET[programIndex] = (varying unsigned int32) 0xFFFFFFFB; // max unsigned int32 - 5
+    } 
+}

tests/psubus_vi64.ispc

@@ -0,0 +1,21 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    varying unsigned int64 a_max = 0xFFFFFFFFFFFFFFFF, a_min = 0; // max and min unsigned int64
+    if (programIndex % 2 == 0) {
+        RET[programIndex] = saturating_sub(a_min, b);
+    }
+    else {
+        RET[programIndex] = saturating_sub(a_max, b);
+    } 
+}
+
+export void result(uniform float RET[]) {
+    if (programIndex % 2 == 0) {
+        RET[programIndex] = (varying unsigned int64) 0; // min unsigned int64
+    }
+    else {
+        RET[programIndex] = (varying unsigned int64) 0xFFFFFFFFFFFFFFFB; // max unsigned int64 - 5
+    } 
+}