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

Fix various small things that were broken with single-bit-per-lane masks.

Browse Source 
Also small cleanups to declarations, "no captures" added, etc.
This commit is contained in:
Matt Pharr
2012-01-04 12:26:22 -08:00
parent dea13979e0
commit 848a432640
4 changed files with 55 additions and 53 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
stdlib.ispc
View File

@@ -312,14 +312,14 @@ static inline int popcnt(int v) {
int r;
for (uniform int i = 0; i < programCount; ++i)
r = insert(r, i, popcnt(extract(v, i)));
return (r & __mask);
return __mask ? r : 0;
}
static inline int popcnt(int64 v) {
int r;
for (uniform int i = 0; i < programCount; ++i)
r = insert(r, i, popcnt(extract(v, i)));
return (r & __mask);
return __mask ? r : 0;
}
static inline uniform int popcnt(bool v) {
@@ -589,7 +589,7 @@ 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 & __mask);
return __reduce_add_int32(__mask ? x : 0);
}
static inline uniform int reduce_min(int v) {
@@ -609,7 +609,7 @@ 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 & __mask);
return __reduce_add_uint32(__mask ? x : 0);
}
static inline uniform unsigned int reduce_min(unsigned int v) {
@@ -647,7 +647,7 @@ 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(__mask ? x : 0);
}
static inline uniform int64 reduce_min(int64 v) {
@@ -667,7 +667,7 @@ 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(__mask ? x : 0);
}
static inline uniform unsigned int64 reduce_min(unsigned int64 v) {