Split sources for amdgcn and r600

Most files remain in a common amdgpu directory.

Also switches barriers to to use convergent,
and use llvm.amdgcn.s.barrier.

This now requires 3.9/trunk to build amdgcn.

llvm-svn: 260777

libclc/amdgcn/lib/SOURCES

@@ -0,0 +1 @@
+synchronization/barrier_impl.ll

libclc/amdgcn/lib/synchronization/barrier_impl.ll

@@ -0,0 +1,32 @@
+declare i32 @__clc_clk_local_mem_fence() #1
+declare i32 @__clc_clk_global_mem_fence() #1
+declare void @llvm.amdgcn.s.barrier() #0
+
+define void @barrier(i32 %flags) #2 {
+barrier_local_test:
+  %CLK_LOCAL_MEM_FENCE = call i32 @__clc_clk_local_mem_fence()
+  %0 = and i32 %flags, %CLK_LOCAL_MEM_FENCE
+  %1 = icmp ne i32 %0, 0
+  br i1 %1, label %barrier_local, label %barrier_global_test
+
+barrier_local:
+  call void @llvm.amdgcn.s.barrier()
+  br label %barrier_global_test
+
+barrier_global_test:
+  %CLK_GLOBAL_MEM_FENCE = call i32 @__clc_clk_global_mem_fence()
+  %2 = and i32 %flags, %CLK_GLOBAL_MEM_FENCE
+  %3 = icmp ne i32 %2, 0
+  br i1 %3, label %barrier_global, label %done
+
+barrier_global:
+  call void @llvm.amdgcn.s.barrier()
+  br label %done
+
+done:
+  ret void
+}
+
+attributes #0 = { nounwind convergent }
+attributes #1 = { nounwind alwaysinline }
+attributes #2 = { nounwind convergent alwaysinline }

libclc/amdgpu/lib/OVERRIDES

@@ -0,0 +1,2 @@
+workitem/get_group_id.cl
+workitem/get_global_size.cl

libclc/amdgpu/lib/SOURCES

@@ -0,0 +1,25 @@
+atomic/atomic.cl
+math/ldexp.cl
+math/nextafter.cl
+math/sqrt.cl
+workitem/get_num_groups.ll
+workitem/get_group_id.ll
+workitem/get_local_size.ll
+workitem/get_local_id.ll
+workitem/get_global_size.ll
+workitem/get_work_dim.ll
+synchronization/barrier.cl
+image/get_image_width.cl
+image/get_image_height.cl
+image/get_image_depth.cl
+image/get_image_channel_data_type.cl
+image/get_image_channel_order.cl
+image/get_image_attributes_impl.ll
+image/read_imagef.cl
+image/read_imagei.cl
+image/read_imageui.cl
+image/read_image_impl.ll
+image/write_imagef.cl
+image/write_imagei.cl
+image/write_imageui.cl
+image/write_image_impl.ll

libclc/configure.py

@@ -69,8 +69,8 @@ llvm_version = string.split(string.replace(llvm_config(['--version']), 'svn', ''
 llvm_int_version = int(llvm_version[0]) * 100 + int(llvm_version[1]) * 10
 llvm_string_version = 'LLVM' + llvm_version[0] + '.' + llvm_version[1]
 
-if llvm_int_version < 370:
-    print "libclc requires LLVM >= 3.7"
+if llvm_int_version < 390:
+    print "libclc requires LLVM >= 3.9"
     sys.exit(1)
 
 llvm_system_libs = llvm_config(['--system-libs'])
@@ -175,8 +175,8 @@ for target in targets:
     subdirs.append("%s-%s-%s" % (arch, t_vendor, t_os))
     subdirs.append("%s-%s" % (arch, t_os))
     subdirs.append(arch)
-    if arch == 'amdgcn':
-        subdirs.append('r600')
+    if arch == 'amdgcn' or arch == 'r600':
+        subdirs.append('amdgpu')
 
   incdirs = filter(os.path.isdir,
                [os.path.join(srcdir, subdir, 'include') for subdir in subdirs])

libclc/r600/lib/OVERRIDES

@@ -1,2 +0,0 @@
-workitem/get_group_id.cl
-workitem/get_global_size.cl

libclc/r600/lib/SOURCES

@@ -1,26 +1 @@
-atomic/atomic.cl
-math/ldexp.cl
-math/nextafter.cl
-math/sqrt.cl
-workitem/get_num_groups.ll
-workitem/get_group_id.ll
-workitem/get_local_size.ll
-workitem/get_local_id.ll
-workitem/get_global_size.ll
-workitem/get_work_dim.ll
-synchronization/barrier.cl
 synchronization/barrier_impl.ll
-image/get_image_width.cl
-image/get_image_height.cl
-image/get_image_depth.cl
-image/get_image_channel_data_type.cl
-image/get_image_channel_order.cl
-image/get_image_attributes_impl.ll
-image/read_imagef.cl
-image/read_imagei.cl
-image/read_imageui.cl
-image/read_image_impl.ll
-image/write_imagef.cl
-image/write_imagei.cl
-image/write_imageui.cl
-image/write_image_impl.ll

libclc/r600/lib/synchronization/barrier_impl.ll

@@ -1,9 +1,9 @@
-declare i32 @__clc_clk_local_mem_fence() nounwind alwaysinline
-declare i32 @__clc_clk_global_mem_fence() nounwind alwaysinline
-declare void @llvm.AMDGPU.barrier.local() nounwind noduplicate
-declare void @llvm.AMDGPU.barrier.global() nounwind noduplicate
+declare i32 @__clc_clk_local_mem_fence() #1
+declare i32 @__clc_clk_global_mem_fence() #1
+declare void @llvm.AMDGPU.barrier.local() #0
+declare void @llvm.AMDGPU.barrier.global() #0
 
-define void @barrier(i32 %flags) nounwind noduplicate alwaysinline {
+define void @barrier(i32 %flags) #2 {
 barrier_local_test:
   %CLK_LOCAL_MEM_FENCE = call i32 @__clc_clk_local_mem_fence()
   %0 = and i32 %flags, %CLK_LOCAL_MEM_FENCE
@@ -11,7 +11,7 @@ barrier_local_test:
   br i1 %1, label %barrier_local, label %barrier_global_test
 
 barrier_local:
-  call void @llvm.AMDGPU.barrier.local() noduplicate
+  call void @llvm.AMDGPU.barrier.local()
   br label %barrier_global_test
 
 barrier_global_test:
@@ -21,9 +21,13 @@ barrier_global_test:
   br i1 %3, label %barrier_global, label %done
 
 barrier_global:
-  call void @llvm.AMDGPU.barrier.global() noduplicate
+  call void @llvm.AMDGPU.barrier.global()
   br label %done
 
 done:
   ret void
 }
+
+attributes #0 = { nounwind convergent }
+attributes #1 = { nounwind alwaysinline }
+attributes #2 = { nounwind convergent alwaysinline }