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

Fix bug in __gather_base_offsets_32.

Browse Source 
In short, we weren't correctly zeroing the compile-time constant portion
of the offsets for lanes that aren't executing. (!)

Fixes issue #235.
This commit is contained in:
Matt Pharr
2012-04-12 10:28:15 -07:00
parent ca7cc4744e
commit fd846fbe77
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
builtins/util.m4
View File

@@ -2880,11 +2880,11 @@ define <$1 x $2> @__gather_base_offsets32_$2(i8 * %ptr, <$1 x i32> %offsets, i32
%newDelta = load <$1 x i32> * %deltaPtr
%ret0 = call <$1 x $2> @__gather_elt32_$2(i8 * %ptr, <$1 x i32> %newOffsets,
i32 %offset_scale, <$1 x i32> %offset_delta,
i32 %offset_scale, <$1 x i32> %newDelta,
<$1 x $2> undef, i32 0)
forloop(lane, 1, eval($1-1),
`patsubst(patsubst(`%retLANE = call <$1 x $2> @__gather_elt32_$2(i8 * %ptr,
<$1 x i32> %newOffsets, i32 %offset_scale, <$1 x i32> %offset_delta,
<$1 x i32> %newOffsets, i32 %offset_scale, <$1 x i32> %newDelta,
<$1 x $2> %retPREV, i32 LANE)
', `LANE', lane), `PREV', eval(lane-1))')
ret <$1 x $2> %ret`'eval($1-1)