diff options
| author | Hal Finkel <hfinkel@anl.gov> | 2013-04-09 17:27:09 +0000 |
|---|---|---|
| committer | Hal Finkel <hfinkel@anl.gov> | 2013-04-09 17:27:09 +0000 |
| commit | f6f8198d85f278ff03aaf32c9db6ae0b3826395c (patch) | |
| tree | b5b4e76c738f25c5402e812080d5af875f3edc55 /test/CodeGen/PowerPC | |
| parent | 307b8535ee4cb37ade40fdc7dd36baa7f27fa34c (diff) | |
| download | llvm-f6f8198d85f278ff03aaf32c9db6ae0b3826395c.tar.gz llvm-f6f8198d85f278ff03aaf32c9db6ae0b3826395c.tar.bz2 llvm-f6f8198d85f278ff03aaf32c9db6ae0b3826395c.tar.xz | |
Use virtual base registers on PPC
On PowerPC, non-vector loads and stores have r+i forms; however, in functions
with large stack frames these were not being used to access slots far from the
stack pointer because such slots were out of range for the signed 16-bit
immediate offset field. This increases register pressure because we need a
separate register for each offset (when the r+r form is used). By enabling
virtual base registers, we can deal with large stack frames without unduly
increasing register pressure.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@179105 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'test/CodeGen/PowerPC')
| -rw-r--r-- | test/CodeGen/PowerPC/2007-09-07-LoadStoreIdxForms.ll | 4 | ||||
| -rw-r--r-- | test/CodeGen/PowerPC/lsa.ll | 48 |
2 files changed, 52 insertions, 0 deletions
diff --git a/test/CodeGen/PowerPC/2007-09-07-LoadStoreIdxForms.ll b/test/CodeGen/PowerPC/2007-09-07-LoadStoreIdxForms.ll index 40f46fda46..ba5c8c172f 100644 --- a/test/CodeGen/PowerPC/2007-09-07-LoadStoreIdxForms.ll +++ b/test/CodeGen/PowerPC/2007-09-07-LoadStoreIdxForms.ll @@ -1,5 +1,9 @@ ; RUN: llc < %s -march=ppc64 | FileCheck %s +; Temporarily XFAIL this test until LSA stops creating single-use +; virtual base registers. +; XFAIL: * + %struct.__db_region = type { %struct.__mutex_t, [4 x i8], %struct.anon, i32, [1 x i32] } %struct.__mutex_t = type { i32 } %struct.anon = type { i64, i64 } diff --git a/test/CodeGen/PowerPC/lsa.ll b/test/CodeGen/PowerPC/lsa.ll new file mode 100644 index 0000000000..3daeccbf40 --- /dev/null +++ b/test/CodeGen/PowerPC/lsa.ll @@ -0,0 +1,48 @@ +; RUN: llc < %s -mtriple=powerpc64-unknown-linux-gnu -mcpu=pwr7 | 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-f128:128:128-v128:128:128-n32:64" +target triple = "powerpc64-unknown-linux-gnu" + +define signext i32 @foo() #0 { +entry: + %v = alloca [8200 x i32], align 4 + %w = alloca [8200 x i32], align 4 + %q = alloca [8200 x i32], align 4 + %0 = bitcast [8200 x i32]* %v to i8* + call void @llvm.lifetime.start(i64 32800, i8* %0) #0 + %1 = bitcast [8200 x i32]* %w to i8* + call void @llvm.lifetime.start(i64 32800, i8* %1) #0 + %2 = bitcast [8200 x i32]* %q to i8* + call void @llvm.lifetime.start(i64 32800, i8* %2) #0 + %arraydecay = getelementptr inbounds [8200 x i32]* %q, i64 0, i64 0 + %arraydecay1 = getelementptr inbounds [8200 x i32]* %v, i64 0, i64 0 + %arraydecay2 = getelementptr inbounds [8200 x i32]* %w, i64 0, i64 0 + call void @bar(i32* %arraydecay, i32* %arraydecay1, i32* %arraydecay2) #0 + %3 = load i32* %arraydecay2, align 4, !tbaa !0 + %arrayidx3 = getelementptr inbounds [8200 x i32]* %w, i64 0, i64 1 + %4 = load i32* %arrayidx3, align 4, !tbaa !0 + +; CHECK: @foo +; CHECK-NOT: lwzx +; CHECK: lwz {{[0-9]+}}, 4([[REG:[0-9]+]]) +; CHECK: lwz {{[0-9]+}}, 0([[REG]]) +; CHECK: blr + + %add = add nsw i32 %4, %3 + call void @llvm.lifetime.end(i64 32800, i8* %2) #0 + call void @llvm.lifetime.end(i64 32800, i8* %1) #0 + call void @llvm.lifetime.end(i64 32800, i8* %0) #0 + ret i32 %add +} + +declare void @llvm.lifetime.start(i64, i8* nocapture) #0 + +declare void @bar(i32*, i32*, i32*) + +declare void @llvm.lifetime.end(i64, i8* nocapture) #0 + +attributes #0 = { nounwind } + +!0 = metadata !{metadata !"int", metadata !1} +!1 = metadata !{metadata !"omnipotent char", metadata !2} +!2 = metadata !{metadata !"Simple C/C++ TBAA"} + |