summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorGrigore Lupescu <grigore.lupescu at intel.com>2016-04-11 17:36:16 +0300
committerYang Rong <rong.r.yang@intel.com>2016-04-14 14:13:36 +0800
commit5536699c9ddd0759560c3f55dfadab186e43c6be (patch)
tree5bf6dffa7996a409b9db9e9b82905a37f71764d7
parentd6ffeedec5366f531114bb25bc988de0eaa09ff4 (diff)
Backend: Workgroup scan cmp qword, workaround for execution width 16
Signed-off-by: Grigore Lupescu <grigore.lupescu at intel.com> Reviewed-by: Pan Xiuli <xiuli.pan@intel.com>
-rw-r--r--backend/src/backend/gen_context.cpp28
1 files changed, 24 insertions, 4 deletions
diff --git a/backend/src/backend/gen_context.cpp b/backend/src/backend/gen_context.cpp
index 056fe3a9..cac78a96 100644
--- a/backend/src/backend/gen_context.cpp
+++ b/backend/src/backend/gen_context.cpp
@@ -2768,12 +2768,32 @@ namespace gbe
if(wg_op == ir::WORKGROUP_OP_INCLUSIVE_ADD
|| wg_op == ir::WORKGROUP_OP_EXCLUSIVE_ADD)
p->ADD(dst, dst, partialData);
- else if(wg_op == ir::WORKGROUP_OP_INCLUSIVE_MIN
- || wg_op == ir::WORKGROUP_OP_EXCLUSIVE_MIN)
+ else if(wg_op == ir::WORKGROUP_OP_INCLUSIVE_MIN || wg_op == ir::WORKGROUP_OP_EXCLUSIVE_MIN)
+ {
p->SEL_CMP(GEN_CONDITIONAL_LE, dst, dst, partialData);
- else if(wg_op == ir::WORKGROUP_OP_INCLUSIVE_MAX
- || wg_op == ir::WORKGROUP_OP_EXCLUSIVE_MAX)
+ /* workaround QW datatype on CMP */
+ if(dst.type == GEN_TYPE_UL || dst.type == GEN_TYPE_L){
+ p->SEL_CMP(GEN_CONDITIONAL_LE, dst.offset(dst, 1, 0),
+ dst.offset(dst, 1, 0), partialData);
+ p->SEL_CMP(GEN_CONDITIONAL_LE, dst.offset(dst, 2, 0),
+ dst.offset(dst, 2, 0), partialData);
+ p->SEL_CMP(GEN_CONDITIONAL_LE, dst.offset(dst, 3, 0),
+ dst.offset(dst, 3, 0), partialData);
+ }
+ }
+ else if(wg_op == ir::WORKGROUP_OP_INCLUSIVE_MAX || wg_op == ir::WORKGROUP_OP_EXCLUSIVE_MAX)
+ {
p->SEL_CMP(GEN_CONDITIONAL_GE, dst, dst, partialData);
+ /* workaround QW datatype on CMP */
+ if(dst.type == GEN_TYPE_UL || dst.type == GEN_TYPE_L){
+ p->SEL_CMP(GEN_CONDITIONAL_GE, dst.offset(dst, 1, 0),
+ dst.offset(dst, 1, 0), partialData);
+ p->SEL_CMP(GEN_CONDITIONAL_GE, dst.offset(dst, 2, 0),
+ dst.offset(dst, 2, 0), partialData);
+ p->SEL_CMP(GEN_CONDITIONAL_GE, dst.offset(dst, 3, 0),
+ dst.offset(dst, 3, 0), partialData);
+ }
+ }
}
/* corner cases for threads 0 */