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

Saturation arithmetic (sub and add) was added for int32/int64.

Browse Source
This commit is contained in:
Vsevolod Livinskij
2014-02-17 18:55:40 +04:00
parent 04fda2fcbe
commit f5508db24f
17 changed files with 512 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

128
stdlib.ispc
View File

@@ -4411,6 +4411,42 @@ 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;
@@ -4435,6 +4471,34 @@ 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;
@@ -4461,6 +4525,42 @@ 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;
@@ -4484,6 +4584,34 @@ 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;
}
///////////////////////////////////////////////////////////////////////////
// rdrand