diff options
| author | Dan Gohman <gohman@apple.com> | 2009-08-20 18:23:44 +0000 |
|---|---|---|
| committer | Dan Gohman <gohman@apple.com> | 2009-08-20 18:23:44 +0000 |
| commit | ef74e9bf2af0f7443e85111adb497afd8802975a (patch) | |
| tree | c8ca804caec3947a289e169586b19cd7270d54f8 /test/CodeGen/X86/rip-rel-address.ll | |
| parent | 073f5b669d3f2ea60981d0c044bd43099e7a3dd1 (diff) | |
| download | llvm-ef74e9bf2af0f7443e85111adb497afd8802975a.tar.gz llvm-ef74e9bf2af0f7443e85111adb497afd8802975a.tar.bz2 llvm-ef74e9bf2af0f7443e85111adb497afd8802975a.tar.xz | |
Fix an x86 code size regression: prefer RIP-relative addressing
over absolute addressing even in non-PIC mode (unless the address
has an index or something else incompatible), because it has a
smaller encoding.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79553 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'test/CodeGen/X86/rip-rel-address.ll')
| -rw-r--r-- | test/CodeGen/X86/rip-rel-address.ll | 5 |
1 files changed, 4 insertions, 1 deletions
diff --git a/test/CodeGen/X86/rip-rel-address.ll b/test/CodeGen/X86/rip-rel-address.ll index 863c28785f..a41b8a960f 100644 --- a/test/CodeGen/X86/rip-rel-address.ll +++ b/test/CodeGen/X86/rip-rel-address.ll @@ -1,11 +1,14 @@ ; RUN: llvm-as < %s | llc -march=x86-64 -relocation-model=pic -mtriple=x86_64-apple-darwin10 | FileCheck %s -check-prefix=PIC64 ; RUN: llvm-as < %s | llc -mtriple=x86_64-unknown-linux-gnu -relocation-model=static | FileCheck %s -check-prefix=STATIC64 +; Use %rip-relative addressing even in static mode on x86-64, because +; it has a smaller encoding. + @a = internal global double 3.4 define double @foo() nounwind { %a = load double* @a ret double %a ; PIC64: movsd _a(%rip), %xmm0 -; STATIC64: movsd a, %xmm0 +; STATIC64: movsd a(%rip), %xmm0 } |