From 2aed6ca1d33468f7e6f8509213a1b9dcd8026ec1 Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault@amd.com>
Date: Sat, 19 Dec 2015 01:46:41 +0000
Subject: AMDGPU: Switch barrier intrinsics to using convergent

noduplicate prevents unrolling of small loops that happen to have
barriers in them. If a loop has a barrier in it, it is OK to duplicate
it for the unroll.

llvm-svn: 256075
---
 .../Transforms/LoopUnroll/AMDGPU/lit.local.cfg     |  3 ++
 .../Transforms/LoopUnroll/AMDGPU/unroll-barrier.ll | 33 ++++++++++++++++++++++
 2 files changed, 36 insertions(+)
 create mode 100644 llvm/test/Transforms/LoopUnroll/AMDGPU/lit.local.cfg
 create mode 100644 llvm/test/Transforms/LoopUnroll/AMDGPU/unroll-barrier.ll

(limited to 'llvm/test/Transforms/LoopUnroll')

diff --git a/llvm/test/Transforms/LoopUnroll/AMDGPU/lit.local.cfg b/llvm/test/Transforms/LoopUnroll/AMDGPU/lit.local.cfg
new file mode 100644
index 00000000000..6baccf05fff
--- /dev/null
+++ b/llvm/test/Transforms/LoopUnroll/AMDGPU/lit.local.cfg
@@ -0,0 +1,3 @@
+if not 'AMDGPU' in config.root.targets:
+    config.unsupported = True
+
diff --git a/llvm/test/Transforms/LoopUnroll/AMDGPU/unroll-barrier.ll b/llvm/test/Transforms/LoopUnroll/AMDGPU/unroll-barrier.ll
new file mode 100644
index 00000000000..3cbb70274da
--- /dev/null
+++ b/llvm/test/Transforms/LoopUnroll/AMDGPU/unroll-barrier.ll
@@ -0,0 +1,33 @@
+; RUN: opt -mtriple=amdgcn-unknown-amdhsa -mcpu=hawaii -loop-unroll -S < %s | FileCheck %s
+
+; CHECK-LABEL: @test_unroll_convergent_barrier(
+; CHECK: call void @llvm.AMDGPU.barrier.global()
+; CHECK: call void @llvm.AMDGPU.barrier.global()
+; CHECK: call void @llvm.AMDGPU.barrier.global()
+; CHECK: call void @llvm.AMDGPU.barrier.global()
+; CHECK-NOT: br
+define void @test_unroll_convergent_barrier(i32 addrspace(1)* noalias nocapture %out, i32 addrspace(1)* noalias nocapture %in) #0 {
+entry:
+  br label %for.body
+
+for.body:                                         ; preds = %for.body, %entry
+  %indvars.iv = phi i32 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
+  %sum.02 = phi i32 [ %add, %for.body ], [ 0, %entry ]
+  %arrayidx.in = getelementptr inbounds i32, i32 addrspace(1)* %in, i32 %indvars.iv
+  %arrayidx.out = getelementptr inbounds i32, i32 addrspace(1)* %out, i32 %indvars.iv
+  %load = load i32, i32 addrspace(1)* %arrayidx.in
+  call void @llvm.AMDGPU.barrier.global() #1
+  %add = add i32 %load, %sum.02
+  store i32 %add, i32 addrspace(1)* %arrayidx.out
+  %indvars.iv.next = add i32 %indvars.iv, 1
+  %exitcond = icmp eq i32 %indvars.iv.next, 4
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:                                          ; preds = %for.body, %entry
+  ret void
+}
+
+declare void @llvm.AMDGPU.barrier.global() #1
+
+attributes #0 = { nounwind }
+attributes #1 = { nounwind convergent }
-- 
cgit v1.2.3