diff options
author | Cameron Zwarich <zwarich@apple.com> | 2011-03-30 23:01:21 +0000 |
---|---|---|
committer | Cameron Zwarich <zwarich@apple.com> | 2011-03-30 23:01:21 +0000 |
commit | c0e6d780cd7a0935f545a0ec0a9ad4a6ae8db2a9 (patch) | |
tree | ec953e12dc862895c78118ce64ecc6240cb92a11 /lib/Target/ARM/ARMInstrNEON.td | |
parent | f5b23368d27ef7dd44b2c45b5c08812ac99c7661 (diff) | |
download | external_llvm-c0e6d780cd7a0935f545a0ec0a9ad4a6ae8db2a9.zip external_llvm-c0e6d780cd7a0935f545a0ec0a9ad4a6ae8db2a9.tar.gz external_llvm-c0e6d780cd7a0935f545a0ec0a9ad4a6ae8db2a9.tar.bz2 |
Add a ARM-specific SD node for VBSL so that forms with a constant first operand
can be recognized. This fixes <rdar://problem/9183078>.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@128584 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'lib/Target/ARM/ARMInstrNEON.td')
-rw-r--r-- | lib/Target/ARM/ARMInstrNEON.td | 23 |
1 files changed, 17 insertions, 6 deletions
diff --git a/lib/Target/ARM/ARMInstrNEON.td b/lib/Target/ARM/ARMInstrNEON.td index 374dac0..c4fb374 100644 --- a/lib/Target/ARM/ARMInstrNEON.td +++ b/lib/Target/ARM/ARMInstrNEON.td @@ -80,6 +80,12 @@ def SDTARMVORRIMM : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>, def NEONvorrImm : SDNode<"ARMISD::VORRIMM", SDTARMVORRIMM>; def NEONvbicImm : SDNode<"ARMISD::VBICIMM", SDTARMVORRIMM>; +def NEONvbsl : SDNode<"ARMISD::VBSL", + SDTypeProfile<1, 3, [SDTCisVec<0>, + SDTCisSameAs<0, 1>, + SDTCisSameAs<0, 2>, + SDTCisSameAs<0, 3>]>>; + def NEONvdup : SDNode<"ARMISD::VDUP", SDTypeProfile<1, 1, [SDTCisVec<0>]>>; // VDUPLANE can produce a quad-register result from a double-register source, @@ -3767,16 +3773,21 @@ def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VCNTiD, "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd", - [(set DPR:$Vd, - (v2i32 (or (and DPR:$Vn, DPR:$src1), - (and DPR:$Vm, (vnotd DPR:$src1)))))]>; + [(set DPR:$Vd, (v2i32 (NEONvbsl DPR:$src1, DPR:$Vn, DPR:$Vm)))]>; + +def : Pat<(v2i32 (or (and DPR:$Vn, DPR:$Vd), + (and DPR:$Vm, (vnotd DPR:$Vd)))), + (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm)>; + def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VCNTiQ, "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd", - [(set QPR:$Vd, - (v4i32 (or (and QPR:$Vn, QPR:$src1), - (and QPR:$Vm, (vnotq QPR:$src1)))))]>; + [(set QPR:$Vd, (v4i32 (NEONvbsl QPR:$src1, QPR:$Vn, QPR:$Vm)))]>; + +def : Pat<(v4i32 (or (and QPR:$Vn, QPR:$Vd), + (and QPR:$Vm, (vnotq QPR:$Vd)))), + (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm)>; // VBIF : Vector Bitwise Insert if False // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst", |