diff options
author | Bruno Cardoso Lopes <bruno.cardoso@gmail.com> | 2011-07-28 01:26:39 +0000 |
---|---|---|
committer | Bruno Cardoso Lopes <bruno.cardoso@gmail.com> | 2011-07-28 01:26:39 +0000 |
commit | 9e2a301216df4f56f9fc904f1ff3e7b113efdfd5 (patch) | |
tree | 45193c79c43c62083c42f4df0e4c89ba541656e9 /llvm/lib | |
parent | e71b9c446dd5d984bd8a7f603ccce96febfa678a (diff) | |
download | bcm5719-llvm-9e2a301216df4f56f9fc904f1ff3e7b113efdfd5.tar.gz bcm5719-llvm-9e2a301216df4f56f9fc904f1ff3e7b113efdfd5.zip |
Add SINT_TO_FP and FP_TO_SINT support for v8i32 types. Also move
a convert pattern close to the instruction definition.
llvm-svn: 136320
Diffstat (limited to 'llvm/lib')
-rw-r--r-- | llvm/lib/Target/X86/X86ISelLowering.cpp | 3 | ||||
-rw-r--r-- | llvm/lib/Target/X86/X86InstrSSE.td | 17 |
2 files changed, 13 insertions, 7 deletions
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp index 081ac006930..e1c64ea04ef 100644 --- a/llvm/lib/Target/X86/X86ISelLowering.cpp +++ b/llvm/lib/Target/X86/X86ISelLowering.cpp @@ -964,6 +964,9 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM) setOperationAction(ISD::FSQRT, MVT::v4f64, Legal); setOperationAction(ISD::FNEG, MVT::v4f64, Custom); + setOperationAction(ISD::FP_TO_SINT, MVT::v8i32, Legal); + setOperationAction(ISD::SINT_TO_FP, MVT::v8i32, Legal); + // Custom lower several nodes for 256-bit types. for (unsigned i = (unsigned)MVT::FIRST_VECTOR_VALUETYPE; i <= (unsigned)MVT::LAST_VECTOR_VALUETYPE; ++i) { diff --git a/llvm/lib/Target/X86/X86InstrSSE.td b/llvm/lib/Target/X86/X86InstrSSE.td index 84f53298caf..e15d7bd76b8 100644 --- a/llvm/lib/Target/X86/X86InstrSSE.td +++ b/llvm/lib/Target/X86/X86InstrSSE.td @@ -884,7 +884,6 @@ def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src), [(set VR128:$dst, (int_x86_sse2_cvttps2dq (memop addr:$src)))]>; - def Int_VCVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "vcvttps2dq\t{$src, $dst|$dst, $src}", [(set VR128:$dst, @@ -896,6 +895,16 @@ def Int_VCVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src), (memop addr:$src)))]>, XS, VEX, Requires<[HasAVX]>; +def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))), + (Int_CVTDQ2PSrr VR128:$src)>, Requires<[HasSSE2]>; +def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))), + (CVTTPS2DQrr VR128:$src)>, Requires<[HasSSE2]>; + +def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))), + (VCVTDQ2PSYrr VR256:$src)>, Requires<[HasAVX]>; +def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))), + (VCVTTPS2DQYrr VR256:$src)>, Requires<[HasAVX]>; + def Int_VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "cvttpd2dq\t{$src, $dst|$dst, $src}", @@ -3857,12 +3866,6 @@ def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))), def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))), (MOVZPQILo2PQIrr VR128:$src)>, Requires<[HasSSE2]>; -// vector -> vector casts -def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))), - (Int_CVTDQ2PSrr VR128:$src)>, Requires<[HasSSE2]>; -def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))), - (CVTTPS2DQrr VR128:$src)>, Requires<[HasSSE2]>; - // Use movaps / movups for SSE integer load / store (one byte shorter). // The instructions selected below are then converted to MOVDQA/MOVDQU // during the SSE domain pass. |