diff options
| author | Nadav Rotem <nadav.rotem@intel.com> | 2011-02-20 12:37:50 +0000 |
|---|---|---|
| committer | Nadav Rotem <nadav.rotem@intel.com> | 2011-02-20 12:37:50 +0000 |
| commit | ed9b934f65d82324506f03e2db2834682c7a8914 (patch) | |
| tree | f9eb5310a278d45abf817090ac930efe6f9c42f2 /test/CodeGen/X86/vec_zext.ll | |
| parent | 1a4021a2be4a59e9f9010776cb6f72107241aeb5 (diff) | |
| download | llvm-ed9b934f65d82324506f03e2db2834682c7a8914.tar.gz llvm-ed9b934f65d82324506f03e2db2834682c7a8914.tar.bz2 llvm-ed9b934f65d82324506f03e2db2834682c7a8914.tar.xz | |
Fix 9267; Add vector zext support.
The DAGCombiner folds the zext into complex load instructions. This patch
prevents this optimization on vectors since none of the supported targets
knows how to perform load+vector_zext in one instruction.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@126080 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'test/CodeGen/X86/vec_zext.ll')
| -rw-r--r-- | test/CodeGen/X86/vec_zext.ll | 69 |
1 files changed, 69 insertions, 0 deletions
diff --git a/test/CodeGen/X86/vec_zext.ll b/test/CodeGen/X86/vec_zext.ll new file mode 100644 index 0000000000..615a50b7af --- /dev/null +++ b/test/CodeGen/X86/vec_zext.ll @@ -0,0 +1,69 @@ +; RUN: llc < %s -march=x86-64 +; PR 9267 + +define<4 x i32> @func_16_32() { + %F = load <4 x i16>* undef + %G = zext <4 x i16> %F to <4 x i32> + %H = load <4 x i16>* undef + %Y = zext <4 x i16> %H to <4 x i32> + %T = add <4 x i32> %Y, %G + store <4 x i32>%T , <4 x i32>* undef + ret <4 x i32> %T +} + +define<4 x i64> @func_16_64() { + %F = load <4 x i16>* undef + %G = zext <4 x i16> %F to <4 x i64> + %H = load <4 x i16>* undef + %Y = zext <4 x i16> %H to <4 x i64> + %T = xor <4 x i64> %Y, %G + store <4 x i64>%T , <4 x i64>* undef + ret <4 x i64> %T +} + +define<4 x i64> @func_32_64() { + %F = load <4 x i32>* undef + %G = zext <4 x i32> %F to <4 x i64> + %H = load <4 x i32>* undef + %Y = zext <4 x i32> %H to <4 x i64> + %T = or <4 x i64> %Y, %G + ret <4 x i64> %T +} + +define<4 x i16> @func_8_16() { + %F = load <4 x i8>* undef + %G = zext <4 x i8> %F to <4 x i16> + %H = load <4 x i8>* undef + %Y = zext <4 x i8> %H to <4 x i16> + %T = add <4 x i16> %Y, %G + ret <4 x i16> %T +} + +define<4 x i32> @func_8_32() { + %F = load <4 x i8>* undef + %G = zext <4 x i8> %F to <4 x i32> + %H = load <4 x i8>* undef + %Y = zext <4 x i8> %H to <4 x i32> + %T = sub <4 x i32> %Y, %G + ret <4 x i32> %T +} + +define<4 x i64> @func_8_64() { + %F = load <4 x i8>* undef + %G = zext <4 x i8> %F to <4 x i64> + %H = load <4 x i8>* undef + %Y = zext <4 x i8> %H to <4 x i64> + %T = add <4 x i64> %Y, %G + ret <4 x i64> %T +} + +define<4 x i32> @const_16_32() { + %G = zext <4 x i16> <i16 0, i16 3, i16 8, i16 7> to <4 x i32> + ret <4 x i32> %G +} + +define<4 x i64> @const_16_64() { + %G = zext <4 x i16> <i16 0, i16 3, i16 8, i16 7> to <4 x i64> + ret <4 x i64> %G +} + |