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

Fix issue #2: use zero extend to convert bool->int, not sign extend.

Browse Source 
This way, we match C/C++ in that casting a bool to an int gives either the value
zero or the value one.  There is a new stdlib function int sign_extend(bool)
that does sign extension for cases where that's desired.
This commit is contained in:
Matt Pharr
2011-07-12 13:30:05 +01:00
parent 6e8af5038b
commit a535aa586b
8 changed files with 94 additions and 28 deletions
Download Patch File Download Diff File Expand all files Collapse all files

23
stdlib.ispc
View File

@@ -197,18 +197,25 @@ static inline unsigned int64 insert(unsigned int64 x, uniform int i,
return __insert_int64(x, (unsigned int)i, v);
}
static inline uniform int32 sign_extend(uniform bool v) {
return __sext_uniform_bool(v);
}
static inline int32 sign_extend(bool v) {
return __sext_varying_bool(v);
}
static inline uniform bool any(bool v) {
// We only care about whether "any" is true for the active program instances,
// so we have to make v with the current program mask.
return __movmsk(v & __mask) != 0;
return __movmsk(__sext_varying_bool(v) & __mask) != 0;
}
static inline uniform bool all(bool v) {
// As with any(), we need to explicitly mask v with the current program mask
// so we're only looking at the current lanes
bool match = ((v & __mask) == __mask);
return __movmsk((int)match) == (1 << programCount) - 1;
int32 match = __sext_varying_bool((__sext_varying_bool(v) & __mask) == __mask);
return __movmsk(match) == (1 << programCount) - 1;
}
static inline uniform int popcnt(uniform int v) {
@@ -235,7 +242,7 @@ static inline int popcnt(int64 v) {
static inline uniform int popcnt(bool v) {
// As with any() and all(), only count across the active lanes
return __popcnt_int32(__movmsk(v & __mask));
return __popcnt_int32(__movmsk(__sext_varying_bool(v) & __mask));
}
static inline uniform int lanemask() {
@@ -271,7 +278,7 @@ static inline uniform float reduce_max(float v) {
static inline uniform int reduce_add(int x) {
// Zero out the values for lanes that aren't running
return __reduce_add_int32(x & (int)__mask);
return __reduce_add_int32(x & __mask);
}
static inline uniform int reduce_min(int v) {
@@ -291,7 +298,7 @@ static inline uniform int reduce_max(int v) {
static inline uniform unsigned int reduce_add(unsigned int x) {
// Set values for non-running lanes to zero so they don't affect the
// result.
return __reduce_add_uint32(x & (int)__mask);
return __reduce_add_uint32(x & __mask);
}
static inline uniform unsigned int reduce_min(unsigned int v) {
@@ -329,7 +336,7 @@ static inline uniform double reduce_max(double v) {
static inline uniform int64 reduce_add(int64 x) {
// Zero out the values for lanes that aren't running
return __reduce_add_int64(x & (int64)__mask);
return __reduce_add_int64(x & (int64)(__mask));
}
static inline uniform int64 reduce_min(int64 v) {
@@ -349,7 +356,7 @@ static inline uniform int64 reduce_max(int64 v) {
static inline uniform unsigned int64 reduce_add(unsigned int64 x) {
// Set values for non-running lanes to zero so they don't affect the
// result.
return __reduce_add_int64(x & (int64)__mask);
return __reduce_add_int64(x & (int64)(__mask));
}
static inline uniform unsigned int64 reduce_min(unsigned int64 v) {