commit 9ea3c385973c395c4798145f71ff3ae34ccaabeb
author Alexey Bataev <a.bataev@hotmail.com> Tue Oct 09 14:49:00 2018 +0000
committer Alexey Bataev <a.bataev@hotmail.com> Tue Oct 09 14:49:00 2018 +0000
tree 0a2f511ca7fb094143065d6ee32b2976e7079a80
parent f5fac1826a867d5e65310456670cc1d5a495ed02

[OPENMP][NVPTX] Support memory coalescing for globalized variables.

Added support for memory coalescing for better performance for
globalized variables. From now on all the globalized variables are
represented as arrays of 32 elements and each thread accesses these
elements using `tid & 31` as index.

llvm-svn: 344049

clang/test/OpenMP/nvptx_teams_reduction_codegen.cpp

diff --git a/clang/test/OpenMP/nvptx_teams_reduction_codegen.cpp b/clang/test/OpenMP/nvptx_teams_reduction_codegen.cpp
index 6925b17..b73b5a8 100644
--- a/clang/test/OpenMP/nvptx_teams_reduction_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_teams_reduction_codegen.cpp
@@ -597,9 +597,9 @@
   // CHECK: [[OF:%.+]] = mul nuw i[[SZ]] [[NUM_TEAMS]], 1
   // CHECK: [[POS1:%.+]] = add nuw i[[SZ]] [[SCRATCHPAD]], [[OF]]
   // CHECK: [[POS2:%.+]] = sub nuw i[[SZ]] [[POS1]], 1
-  // CHECK: [[POS3:%.+]] = udiv i[[SZ]] [[POS2]], 256
+  // CHECK: [[POS3:%.+]] = udiv i[[SZ]] [[POS2]], 128
   // CHECK: [[POS4:%.+]] = add nuw i[[SZ]] [[POS3]], 1
-  // CHECK: [[SCRATCHPAD_NEXT:%.+]] = mul nuw i[[SZ]] [[POS4]], 256
+  // CHECK: [[SCRATCHPAD_NEXT:%.+]] = mul nuw i[[SZ]] [[POS4]], 128
   //
   // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST]], i[[SZ]] 0, i[[SZ]] 1
   // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]],
@@ -643,9 +643,9 @@
   // CHECK: [[OF:%.+]] = mul nuw i[[SZ]] [[NUM_TEAMS]], 1
   // CHECK: [[POS1:%.+]] = add nuw i[[SZ]] [[SCRATCHPAD]], [[OF]]
   // CHECK: [[POS2:%.+]] = sub nuw i[[SZ]] [[POS1]], 1
-  // CHECK: [[POS3:%.+]] = udiv i[[SZ]] [[POS2]], 256
+  // CHECK: [[POS3:%.+]] = udiv i[[SZ]] [[POS2]], 128
   // CHECK: [[POS4:%.+]] = add nuw i[[SZ]] [[POS3]], 1
-  // CHECK: [[SCRATCHPAD_NEXT:%.+]] = mul nuw i[[SZ]] [[POS4]], 256
+  // CHECK: [[SCRATCHPAD_NEXT:%.+]] = mul nuw i[[SZ]] [[POS4]], 128
   //
   // CHECK: [[P:%.+]] = mul nuw i[[SZ]] 4, [[TEAM]]
   // CHECK: [[SCRATCHPAD_ELT_PTR64:%.+]] = add nuw i[[SZ]] [[SCRATCHPAD_NEXT]], [[P]]
@@ -1024,9 +1024,9 @@
   // CHECK: [[OF:%.+]] = mul nuw i[[SZ]] [[NUM_TEAMS]], 4
   // CHECK: [[POS1:%.+]] = add nuw i[[SZ]] [[SCRATCHPAD]], [[OF]]
   // CHECK: [[POS2:%.+]] = sub nuw i[[SZ]] [[POS1]], 1
-  // CHECK: [[POS3:%.+]] = udiv i[[SZ]] [[POS2]], 256
+  // CHECK: [[POS3:%.+]] = udiv i[[SZ]] [[POS2]], 128
   // CHECK: [[POS4:%.+]] = add nuw i[[SZ]] [[POS3]], 1
-  // CHECK: [[SCRATCHPAD_NEXT:%.+]] = mul nuw i[[SZ]] [[POS4]], 256
+  // CHECK: [[SCRATCHPAD_NEXT:%.+]] = mul nuw i[[SZ]] [[POS4]], 128
   //
   // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST]], i[[SZ]] 0, i[[SZ]] 1
   // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]],
@@ -1072,9 +1072,9 @@
   // CHECK: [[OF:%.+]] = mul nuw i[[SZ]] [[NUM_TEAMS]], 4
   // CHECK: [[POS1:%.+]] = add nuw i[[SZ]] [[SCRATCHPAD]], [[OF]]
   // CHECK: [[POS2:%.+]] = sub nuw i[[SZ]] [[POS1]], 1
-  // CHECK: [[POS3:%.+]] = udiv i[[SZ]] [[POS2]], 256
+  // CHECK: [[POS3:%.+]] = udiv i[[SZ]] [[POS2]], 128
   // CHECK: [[POS4:%.+]] = add nuw i[[SZ]] [[POS3]], 1
-  // CHECK: [[SCRATCHPAD_NEXT:%.+]] = mul nuw i[[SZ]] [[POS4]], 256
+  // CHECK: [[SCRATCHPAD_NEXT:%.+]] = mul nuw i[[SZ]] [[POS4]], 128
   //
   // CHECK: [[P:%.+]] = mul nuw i[[SZ]] 2, [[TEAM]]
   // CHECK: [[SCRATCHPAD_ELT_PTR64:%.+]] = add nuw i[[SZ]] [[SCRATCHPAD_NEXT]], [[P]]