AMDGPU: Implement get_global_offset builtin

Also fix get_global_id to consider offset
No idea how to add this for ptx, so they are stuck with the old get_global_id
implementation.

v2: split to a separate patch

v3: Switch R600 to use implictarg.ptr

Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>

git-svn-id: https://llvm.org/svn/llvm-project/libclc/trunk@276443 91177308-0d34-0410-b5e6-96231b3b80d8

9 files changed, 33 insertions, 1 deletions

diff --git a/amdgcn/lib/SOURCES b/amdgcn/lib/SOURCES
index 49d9b53..ca88b93 100644
--- a/amdgcn/lib/SOURCES
+++ b/amdgcn/lib/SOURCES
@@ -1,5 +1,6 @@
 math/ldexp.cl
 synchronization/barrier_impl.ll
+workitem/get_global_offset.cl
 workitem/get_group_id.cl
 workitem/get_local_id.cl
 workitem/get_work_dim.cl
diff --git a/amdgcn/lib/workitem/get_global_offset.cl b/amdgcn/lib/workitem/get_global_offset.cl
new file mode 100644
index 0000000..32aaa4c
--- /dev/null
+++ b/amdgcn/lib/workitem/get_global_offset.cl
@@ -0,0 +1,11 @@
+#include <clc/clc.h>
+
+_CLC_DEF uint get_global_offset(uint dim)
+{
+	__attribute__((address_space(2))) uint * ptr =
+		(__attribute__((address_space(2))) uint *)
+		__builtin_amdgcn_implicitarg_ptr();
+	if (dim < 3)
+		return ptr[dim + 1];
+	return 0;
+}
diff --git a/generic/include/clc/clc.h b/generic/include/clc/clc.h
index 61f9b35..d7931f1 100644
--- a/generic/include/clc/clc.h
+++ b/generic/include/clc/clc.h
@@ -30,6 +30,7 @@
 #include <clc/workitem/get_local_id.h>
 #include <clc/workitem/get_num_groups.h>
 #include <clc/workitem/get_group_id.h>
+#include <clc/workitem/get_global_offset.h>
 
 /* 6.11.2 Math Functions */
 #include <clc/math/acos.h>
diff --git a/generic/include/clc/workitem/get_global_offset.h b/generic/include/clc/workitem/get_global_offset.h
new file mode 100644
index 0000000..7f4f603
--- /dev/null
+++ b/generic/include/clc/workitem/get_global_offset.h
@@ -0,0 +1 @@
+_CLC_DECL size_t get_global_offset(uint dim);
diff --git a/generic/lib/workitem/get_global_id.cl b/generic/lib/workitem/get_global_id.cl
index fdd83d2..b6c2ea1 100644
--- a/generic/lib/workitem/get_global_id.cl
+++ b/generic/lib/workitem/get_global_id.cl
@@ -1,5 +1,5 @@
 #include <clc/clc.h>
 
 _CLC_DEF size_t get_global_id(uint dim) {
-  return get_group_id(dim)*get_local_size(dim) + get_local_id(dim);
+  return get_group_id(dim) * get_local_size(dim) + get_local_id(dim) + get_global_offset(dim);
 }
diff --git a/ptx-nvidiacl/lib/SOURCES b/ptx-nvidiacl/lib/SOURCES
index 7cdbd85..ce26bcb 100644
--- a/ptx-nvidiacl/lib/SOURCES
+++ b/ptx-nvidiacl/lib/SOURCES
@@ -1,4 +1,5 @@
 synchronization/barrier.cl
+workitem/get_global_id.cl
 workitem/get_group_id.cl
 workitem/get_local_id.cl
 workitem/get_local_size.cl
diff --git a/ptx-nvidiacl/lib/workitem/get_global_id.cl b/ptx-nvidiacl/lib/workitem/get_global_id.cl
new file mode 100644
index 0000000..19bc195
--- /dev/null
+++ b/ptx-nvidiacl/lib/workitem/get_global_id.cl
@@ -0,0 +1,5 @@
+#include <clc/clc.h>
+
+_CLC_DEF size_t get_global_id(uint dim) {
+  return get_group_id(dim) * get_local_size(dim) + get_local_id(dim);
+}
diff --git a/r600/lib/SOURCES b/r600/lib/SOURCES
index 4178d70..33038f2 100644
--- a/r600/lib/SOURCES
+++ b/r600/lib/SOURCES
@@ -1,4 +1,5 @@
 synchronization/barrier_impl.ll
+workitem/get_global_offset.cl
 workitem/get_group_id.cl
 workitem/get_local_id.cl
 workitem/get_work_dim.cl
diff --git a/r600/lib/workitem/get_global_offset.cl b/r600/lib/workitem/get_global_offset.cl
new file mode 100644
index 0000000..b38ae33
--- /dev/null
+++ b/r600/lib/workitem/get_global_offset.cl
@@ -0,0 +1,11 @@
+#include <clc/clc.h>
+
+_CLC_DEF uint get_global_offset(uint dim)
+{
+	__attribute__((address_space(7))) uint * ptr =
+		(__attribute__((address_space(7))) uint *)
+		__builtin_r600_implicitarg_ptr();
+	if (dim < 3)
+		return ptr[dim + 1];
+	return 0;
+}