summaryrefslogtreecommitdiff
path: root/test
diff options
context:
space:
mode:
authorJakob Stoklund Olesen <stoklund@2pi.dk>2011-11-09 01:06:56 +0000
committerJakob Stoklund Olesen <stoklund@2pi.dk>2011-11-09 01:06:56 +0000
commitf4c4768fb2277cb940a90cb2f0e9a747ebc671c3 (patch)
treeb98651084e6e09a4538358964fbbb7f1b87dfeb3 /test
parentdbc372f47e3a77343e6ef1ab4a88bc46f532f774 (diff)
downloadllvm-f4c4768fb2277cb940a90cb2f0e9a747ebc671c3.tar.gz
llvm-f4c4768fb2277cb940a90cb2f0e9a747ebc671c3.tar.bz2
llvm-f4c4768fb2277cb940a90cb2f0e9a747ebc671c3.tar.xz
Collapse DomainValues across loop back-edges.
During the initial RPO traversal of the basic blocks, remember the ones that are incomplete because of back-edges from predecessors that haven't been visited yet. After the initial RPO, revisit all those loop headers so the incoming DomainValues on the back-edges can be properly collapsed. This will properly fix execution domains on software pipelined code, like the included test case. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@144151 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'test')
-rw-r--r--test/CodeGen/X86/sse-domains.ll45
1 files changed, 45 insertions, 0 deletions
diff --git a/test/CodeGen/X86/sse-domains.ll b/test/CodeGen/X86/sse-domains.ll
new file mode 100644
index 0000000000..d26d32287e
--- /dev/null
+++ b/test/CodeGen/X86/sse-domains.ll
@@ -0,0 +1,45 @@
+; RUN: llc < %s | FileCheck %s
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
+target triple = "x86_64-apple-macosx10.7"
+
+; CHECK: f
+;
+; This function contains load / store / and operations that all can execute in
+; any domain. The only domain-specific operation is the %add = shl... operation
+; which is <4 x i32>.
+;
+; The paddd instruction can only influence the other operations through the loop
+; back-edge. Check that everything is still moved into the integer domain.
+
+define void @f(<4 x i32>* nocapture %p, i32 %n) nounwind uwtable ssp {
+entry:
+ br label %while.body
+
+; Materialize a zeroinitializer and a constant-pool load in the integer domain.
+; The order is not important.
+; CHECK: pxor
+; CHECK: movdqa
+
+; The instructions in the loop must all be integer domain as well.
+; CHECK: while.body
+; CHECK: pand
+; CHECK: movdqa
+; CHECK: movdqa
+; Finally, the controlling integer-only instruction.
+; CHECK: paddd
+while.body:
+ %p.addr.04 = phi <4 x i32>* [ %incdec.ptr, %while.body ], [ %p, %entry ]
+ %n.addr.03 = phi i32 [ %dec, %while.body ], [ %n, %entry ]
+ %x.02 = phi <4 x i32> [ %add, %while.body ], [ zeroinitializer, %entry ]
+ %dec = add nsw i32 %n.addr.03, -1
+ %and = and <4 x i32> %x.02, <i32 127, i32 127, i32 127, i32 127>
+ %incdec.ptr = getelementptr inbounds <4 x i32>* %p.addr.04, i64 1
+ store <4 x i32> %and, <4 x i32>* %p.addr.04, align 16
+ %0 = load <4 x i32>* %incdec.ptr, align 16
+ %add = shl <4 x i32> %0, <i32 1, i32 1, i32 1, i32 1>
+ %tobool = icmp eq i32 %dec, 0
+ br i1 %tobool, label %while.end, label %while.body
+
+while.end:
+ ret void
+}