PTX: move implementations of work-item and synchronisation functions

to lib, and add header files in generic.  Incorporates a patch by
Tom Stellard!

git-svn-id: https://llvm.org/svn/llvm-project/libclc/trunk@161313 91177308-0d34-0410-b5e6-96231b3b80d8

8 files changed, 19 insertions, 21 deletions

diff --git a/ptx-nvidiacl/include/clc/workitem/get_global_id.h b/ptx-nvidiacl/include/clc/workitem/get_global_id.h
deleted file mode 100644
index 026d2fe..0000000
--- a/ptx-nvidiacl/include/clc/workitem/get_global_id.h
+++ /dev/null
@@ -1,8 +0,0 @@
-_CLC_INLINE size_t get_global_id(uint dim) {
-  switch (dim) {
-  case 0:  return __builtin_ptx_read_ctaid_x()*__builtin_ptx_read_ntid_x()+__builtin_ptx_read_tid_x();
-  case 1:  return __builtin_ptx_read_ctaid_y()*__builtin_ptx_read_ntid_y()+__builtin_ptx_read_tid_y();
-  case 2:  return __builtin_ptx_read_ctaid_z()*__builtin_ptx_read_ntid_z()+__builtin_ptx_read_tid_z();
-  default: return 0;
-  }
-}
diff --git a/ptx-nvidiacl/include/clc/workitem/get_global_size.h b/ptx-nvidiacl/include/clc/workitem/get_global_size.h
deleted file mode 100644
index 5cd4222..0000000
--- a/ptx-nvidiacl/include/clc/workitem/get_global_size.h
+++ /dev/null
@@ -1,8 +0,0 @@
-_CLC_INLINE size_t get_global_size(uint dim) {
-  switch (dim) {
-  case 0:  return __builtin_ptx_read_nctaid_x()*__builtin_ptx_read_ntid_x();
-  case 1:  return __builtin_ptx_read_nctaid_y()*__builtin_ptx_read_ntid_y();
-  case 2:  return __builtin_ptx_read_nctaid_z()*__builtin_ptx_read_ntid_z();
-  default: return 0;
-  }
-}
diff --git a/ptx-nvidiacl/lib/SOURCES b/ptx-nvidiacl/lib/SOURCES
index e69de29..1a96a1a 100644
--- a/ptx-nvidiacl/lib/SOURCES
+++ b/ptx-nvidiacl/lib/SOURCES
@@ -0,0 +1,4 @@
+workitem/get_group_id.cl
+workitem/get_local_id.cl
+workitem/get_local_size.cl
+workitem/get_num_groups.cl
diff --git a/ptx-nvidiacl/include/clc/synchronization/barrier.h b/ptx-nvidiacl/lib/synchronization/barrier.cl
index cd9f327..fb36c26 100644
--- a/ptx-nvidiacl/include/clc/synchronization/barrier.h
+++ b/ptx-nvidiacl/lib/synchronization/barrier.cl
@@ -1,4 +1,6 @@
-_CLC_INLINE void barrier(cl_mem_fence_flags flags) {
+#include <clc/clc.h>
+
+_CLC_DEF void barrier(cl_mem_fence_flags flags) {
   if (flags & CLK_LOCAL_MEM_FENCE) {
     __builtin_ptx_bar_sync(0);
   }
diff --git a/ptx-nvidiacl/include/clc/workitem/get_group_id.h b/ptx-nvidiacl/lib/workitem/get_group_id.cl
index 18b1bd4..2b35b4e 100644
--- a/ptx-nvidiacl/include/clc/workitem/get_group_id.h
+++ b/ptx-nvidiacl/lib/workitem/get_group_id.cl
@@ -1,4 +1,6 @@
-_CLC_INLINE size_t get_group_id(uint dim) {
+#include <clc/clc.h>
+
+_CLC_DEF size_t get_group_id(uint dim) {
   switch (dim) {
   case 0:  return __builtin_ptx_read_ctaid_x();
   case 1:  return __builtin_ptx_read_ctaid_y();
diff --git a/ptx-nvidiacl/include/clc/workitem/get_local_id.h b/ptx-nvidiacl/lib/workitem/get_local_id.cl
index 1b8c776..f0cfdc0 100644
--- a/ptx-nvidiacl/include/clc/workitem/get_local_id.h
+++ b/ptx-nvidiacl/lib/workitem/get_local_id.cl
@@ -1,4 +1,6 @@
-_CLC_INLINE size_t get_local_id(uint dim) {
+#include <clc/clc.h>
+
+_CLC_DEF size_t get_local_id(uint dim) {
   switch (dim) {
   case 0:  return __builtin_ptx_read_tid_x();
   case 1:  return __builtin_ptx_read_tid_y();
diff --git a/ptx-nvidiacl/include/clc/workitem/get_local_size.h b/ptx-nvidiacl/lib/workitem/get_local_size.cl
index cbc1f6e..c3f5425 100644
--- a/ptx-nvidiacl/include/clc/workitem/get_local_size.h
+++ b/ptx-nvidiacl/lib/workitem/get_local_size.cl
@@ -1,4 +1,6 @@
-_CLC_INLINE size_t get_local_size(uint dim) {
+#include <clc/clc.h>
+
+_CLC_DEF size_t get_local_size(uint dim) {
   switch (dim) {
   case 0:  return __builtin_ptx_read_ntid_x();
   case 1:  return __builtin_ptx_read_ntid_y();
diff --git a/ptx-nvidiacl/include/clc/workitem/get_num_groups.h b/ptx-nvidiacl/lib/workitem/get_num_groups.cl
index 36ee849..90bdc2e 100644
--- a/ptx-nvidiacl/include/clc/workitem/get_num_groups.h
+++ b/ptx-nvidiacl/lib/workitem/get_num_groups.cl
@@ -1,4 +1,6 @@
-_CLC_INLINE size_t get_num_groups(uint dim) {
+#include <clc/clc.h>
+
+_CLC_DEF size_t get_num_groups(uint dim) {
   switch (dim) {
   case 0:  return __builtin_ptx_read_nctaid_x();
   case 1:  return __builtin_ptx_read_nctaid_y();