[libclc] r276443 - AMDGPU: Implement get_global_offset builtin

[Jan Vesely via cfe-commits]

Author: jvesely
Date: Fri Jul 22 12:24:24 2016
New Revision: 276443

URL: http://llvm.org/viewvc/llvm-project?rev=276443&view=rev
Log:
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 at rutgers.edu>

Added:
    libclc/trunk/amdgcn/lib/workitem/get_global_offset.cl
    libclc/trunk/generic/include/clc/workitem/get_global_offset.h
    libclc/trunk/ptx-nvidiacl/lib/workitem/get_global_id.cl
    libclc/trunk/r600/lib/workitem/get_global_offset.cl
Modified:
    libclc/trunk/amdgcn/lib/SOURCES
    libclc/trunk/generic/include/clc/clc.h
    libclc/trunk/generic/lib/workitem/get_global_id.cl
    libclc/trunk/ptx-nvidiacl/lib/SOURCES
    libclc/trunk/r600/lib/SOURCES

Modified: libclc/trunk/amdgcn/lib/SOURCES
URL: http://llvm.org/viewvc/llvm-project/libclc/trunk/amdgcn/lib/SOURCES?rev=276443&r1=276442&r2=276443&view=diff
==============================================================================
--- libclc/trunk/amdgcn/lib/SOURCES (original)
+++ libclc/trunk/amdgcn/lib/SOURCES Fri Jul 22 12:24:24 2016
@@ -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

Added: libclc/trunk/amdgcn/lib/workitem/get_global_offset.cl
URL: http://llvm.org/viewvc/llvm-project/libclc/trunk/amdgcn/lib/workitem/get_global_offset.cl?rev=276443&view=auto
==============================================================================
--- libclc/trunk/amdgcn/lib/workitem/get_global_offset.cl (added)
+++ libclc/trunk/amdgcn/lib/workitem/get_global_offset.cl Fri Jul 22 12:24:24 2016
@@ -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;
+}

Modified: libclc/trunk/generic/include/clc/clc.h
URL: http://llvm.org/viewvc/llvm-project/libclc/trunk/generic/include/clc/clc.h?rev=276443&r1=276442&r2=276443&view=diff
==============================================================================
--- libclc/trunk/generic/include/clc/clc.h (original)
+++ libclc/trunk/generic/include/clc/clc.h Fri Jul 22 12:24:24 2016
@@ -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>

Added: libclc/trunk/generic/include/clc/workitem/get_global_offset.h
URL: http://llvm.org/viewvc/llvm-project/libclc/trunk/generic/include/clc/workitem/get_global_offset.h?rev=276443&view=auto
==============================================================================
--- libclc/trunk/generic/include/clc/workitem/get_global_offset.h (added)
+++ libclc/trunk/generic/include/clc/workitem/get_global_offset.h Fri Jul 22 12:24:24 2016
@@ -0,0 +1 @@
+_CLC_DECL size_t get_global_offset(uint dim);

Modified: libclc/trunk/generic/lib/workitem/get_global_id.cl
URL: http://llvm.org/viewvc/llvm-project/libclc/trunk/generic/lib/workitem/get_global_id.cl?rev=276443&r1=276442&r2=276443&view=diff
==============================================================================
--- libclc/trunk/generic/lib/workitem/get_global_id.cl (original)
+++ libclc/trunk/generic/lib/workitem/get_global_id.cl Fri Jul 22 12:24:24 2016
@@ -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);
 }

Modified: libclc/trunk/ptx-nvidiacl/lib/SOURCES
URL: http://llvm.org/viewvc/llvm-project/libclc/trunk/ptx-nvidiacl/lib/SOURCES?rev=276443&r1=276442&r2=276443&view=diff
==============================================================================
--- libclc/trunk/ptx-nvidiacl/lib/SOURCES (original)
+++ libclc/trunk/ptx-nvidiacl/lib/SOURCES Fri Jul 22 12:24:24 2016
@@ -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

Added: libclc/trunk/ptx-nvidiacl/lib/workitem/get_global_id.cl
URL: http://llvm.org/viewvc/llvm-project/libclc/trunk/ptx-nvidiacl/lib/workitem/get_global_id.cl?rev=276443&view=auto
==============================================================================
--- libclc/trunk/ptx-nvidiacl/lib/workitem/get_global_id.cl (added)
+++ libclc/trunk/ptx-nvidiacl/lib/workitem/get_global_id.cl Fri Jul 22 12:24:24 2016
@@ -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);
+}

Modified: libclc/trunk/r600/lib/SOURCES
URL: http://llvm.org/viewvc/llvm-project/libclc/trunk/r600/lib/SOURCES?rev=276443&r1=276442&r2=276443&view=diff
==============================================================================
--- libclc/trunk/r600/lib/SOURCES (original)
+++ libclc/trunk/r600/lib/SOURCES Fri Jul 22 12:24:24 2016
@@ -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

Added: libclc/trunk/r600/lib/workitem/get_global_offset.cl
URL: http://llvm.org/viewvc/llvm-project/libclc/trunk/r600/lib/workitem/get_global_offset.cl?rev=276443&view=auto
==============================================================================
--- libclc/trunk/r600/lib/workitem/get_global_offset.cl (added)
+++ libclc/trunk/r600/lib/workitem/get_global_offset.cl Fri Jul 22 12:24:24 2016
@@ -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;
+}