diff options
| author | Andrea Di Biagio <Andrea_DiBiagio@sn.scee.net> | 2015-05-08 15:11:07 +0000 |
|---|---|---|
| committer | Andrea Di Biagio <Andrea_DiBiagio@sn.scee.net> | 2015-05-08 15:11:07 +0000 |
| commit | 84e22b90967d8f1b36d4254cc500493c7d0d7e41 (patch) | |
| tree | 038efc6e9bafc46e2ca7b518c1ca162d732ffcfe | |
| parent | 9e36fb4c1875dfc02485f9e9657f028f00d78bfb (diff) | |
| download | bcm5719-llvm-84e22b90967d8f1b36d4254cc500493c7d0d7e41.tar.gz bcm5719-llvm-84e22b90967d8f1b36d4254cc500493c7d0d7e41.zip | |
[X86] Teach 'getTargetShuffleMask' how to look through ISD::WrapperRIP when decoding a PSHUFB mask.
The function 'getTargetShuffleMask' already knows how to deal with PSHUFB nodes
where the mask node is a load from constant pool, and the constant pool node
is wrapped by a X86ISD::Wrapper node. This patch extends that logic by teaching
it how to also look through X86ISD::WrapperRIP.
This helps function combineX86ShufflesRecusively to combine more shuffle
sequences containing PSHUFB nodes if we are in RIPRel PIC mode.
Before this change, llc (with -relocation-model=pic -march=x86-64) was unable
to decode a pshufb where the mask was loaded from a constant pool. For example,
the no-op shuffle from test 'x86-fold-pshufb.ll' was not folded into its
operand, so instead of generating a single 'movaps' the backend always
generated a sub-optimal 'movdqa + pshufb' sequence.
Added test x86-fold-pshufb.ll.
llvm-svn: 236863
| -rw-r--r-- | llvm/lib/Target/X86/X86ISelLowering.cpp | 3 | ||||
| -rw-r--r-- | llvm/test/CodeGen/X86/x86-fold-pshufb.ll | 17 |
2 files changed, 19 insertions, 1 deletions
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp index cac5b54187b..35517212acc 100644 --- a/llvm/lib/Target/X86/X86ISelLowering.cpp +++ b/llvm/lib/Target/X86/X86ISelLowering.cpp @@ -4352,7 +4352,8 @@ static bool getTargetShuffleMask(SDNode *N, MVT VT, return false; SDValue Ptr = MaskLoad->getBasePtr(); - if (Ptr->getOpcode() == X86ISD::Wrapper) + if (Ptr->getOpcode() == X86ISD::Wrapper || + Ptr->getOpcode() == X86ISD::WrapperRIP) Ptr = Ptr->getOperand(0); auto *MaskCP = dyn_cast<ConstantPoolSDNode>(Ptr); diff --git a/llvm/test/CodeGen/X86/x86-fold-pshufb.ll b/llvm/test/CodeGen/X86/x86-fold-pshufb.ll new file mode 100644 index 00000000000..c29e592bfe8 --- /dev/null +++ b/llvm/test/CodeGen/X86/x86-fold-pshufb.ll @@ -0,0 +1,17 @@ +; RUN: llc -relocation-model=pic -march=x86-64 -mtriple=x86_64-unknown-unknown -mattr=+ssse3 < %s | FileCheck %s + +; Verify that the backend correctly folds the shuffle in function 'fold_pshufb' +; into a simple load from constant pool. + +define <2 x i64> @fold_pshufb() { +; CHECK-LABEL: fold_pshufb: +; CHECK: # BB#0: +; CHECK-NEXT: movaps {{.*#+}} xmm0 = [0,0,0,0,1,0,0,0,2,0,0,0,3,0,0,0] +; CHECK-NEXT: retq +entry: + %0 = tail call <16 x i8> @llvm.x86.ssse3.pshuf.b.128(<16 x i8> <i8 0, i8 0, i8 0, i8 0, i8 1, i8 0, i8 0, i8 0, i8 2, i8 0, i8 0, i8 0, i8 3, i8 0, i8 0, i8 0>, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15>) + %1 = bitcast <16 x i8> %0 to <2 x i64> + ret <2 x i64> %1 +} + +declare <16 x i8> @llvm.x86.ssse3.pshuf.b.128(<16 x i8>, <16 x i8>) |
