forked from OSchip/llvm-project
Add new itin classes for FP16 <-> FP32 conversions and make uise of them for A9.
llvm-svn: 100647
This commit is contained in:
parent
baeb210be7
commit
4c1da0f82a
|
@ -256,25 +256,25 @@ def VCVTSD : VFPAI<(outs SPR:$dst), (ins DPR:$a), VFPUnaryFrm,
|
|||
// Between half-precision and single-precision. For disassembly only.
|
||||
|
||||
def VCVTBSH : ASuI<0b11101, 0b11, 0b0010, 0b01, 0, (outs SPR:$dst), (ins SPR:$a),
|
||||
/* FIXME */ IIC_fpCVTDS, "vcvtb", ".f32.f16\t$dst, $a",
|
||||
/* FIXME */ IIC_fpCVTSH, "vcvtb", ".f32.f16\t$dst, $a",
|
||||
[/* For disassembly only; pattern left blank */]>;
|
||||
|
||||
def : ARMPat<(f32_to_f16 SPR:$a),
|
||||
(i32 (COPY_TO_REGCLASS (VCVTBSH SPR:$a), GPR))>;
|
||||
|
||||
def VCVTBHS : ASuI<0b11101, 0b11, 0b0011, 0b01, 0, (outs SPR:$dst), (ins SPR:$a),
|
||||
/* FIXME */ IIC_fpCVTDS, "vcvtb", ".f16.f32\t$dst, $a",
|
||||
/* FIXME */ IIC_fpCVTHS, "vcvtb", ".f16.f32\t$dst, $a",
|
||||
[/* For disassembly only; pattern left blank */]>;
|
||||
|
||||
def : ARMPat<(f16_to_f32 GPR:$a),
|
||||
(VCVTBHS (COPY_TO_REGCLASS GPR:$a, SPR))>;
|
||||
|
||||
def VCVTTSH : ASuI<0b11101, 0b11, 0b0010, 0b11, 0, (outs SPR:$dst), (ins SPR:$a),
|
||||
/* FIXME */ IIC_fpCVTDS, "vcvtt", ".f32.f16\t$dst, $a",
|
||||
/* FIXME */ IIC_fpCVTSH, "vcvtt", ".f32.f16\t$dst, $a",
|
||||
[/* For disassembly only; pattern left blank */]>;
|
||||
|
||||
def VCVTTHS : ASuI<0b11101, 0b11, 0b0011, 0b11, 0, (outs SPR:$dst), (ins SPR:$a),
|
||||
/* FIXME */ IIC_fpCVTDS, "vcvtt", ".f16.f32\t$dst, $a",
|
||||
/* FIXME */ IIC_fpCVTHS, "vcvtt", ".f16.f32\t$dst, $a",
|
||||
[/* For disassembly only; pattern left blank */]>;
|
||||
|
||||
let neverHasSideEffects = 1 in {
|
||||
|
|
|
@ -71,6 +71,8 @@ def IIC_fpCMP32 : InstrItinClass;
|
|||
def IIC_fpCMP64 : InstrItinClass;
|
||||
def IIC_fpCVTSD : InstrItinClass;
|
||||
def IIC_fpCVTDS : InstrItinClass;
|
||||
def IIC_fpCVTSH : InstrItinClass;
|
||||
def IIC_fpCVTHS : InstrItinClass;
|
||||
def IIC_fpCVTIS : InstrItinClass;
|
||||
def IIC_fpCVTID : InstrItinClass;
|
||||
def IIC_fpCVTSI : InstrItinClass;
|
||||
|
|
|
@ -651,6 +651,20 @@ def CortexA9Itineraries : ProcessorItineraries<[
|
|||
InstrStage2<5, [FU_DRegsN], 0, Reserved>,
|
||||
InstrStage<1, [FU_Pipe0, FU_Pipe1]>,
|
||||
InstrStage<1, [FU_NPipe]>], [4, 1]>,
|
||||
|
||||
//
|
||||
// Single to Half FP Convert
|
||||
InstrItinData<IIC_fpCVTSH , [InstrStage2<1, [FU_DRegsVFP], 0, Required>,
|
||||
InstrStage2<5, [FU_DRegsN], 0, Reserved>,
|
||||
InstrStage<1, [FU_Pipe0, FU_Pipe1]>,
|
||||
InstrStage<1, [FU_NPipe]>], [4, 1]>,
|
||||
//
|
||||
// Half to Single FP Convert
|
||||
InstrItinData<IIC_fpCVTHS , [InstrStage2<1, [FU_DRegsVFP], 0, Required>,
|
||||
InstrStage2<3, [FU_DRegsN], 0, Reserved>,
|
||||
InstrStage<1, [FU_Pipe0, FU_Pipe1]>,
|
||||
InstrStage<1, [FU_NPipe]>], [2, 1]>,
|
||||
|
||||
//
|
||||
// Single-Precision FP to Integer Convert
|
||||
InstrItinData<IIC_fpCVTSI , [InstrStage2<1, [FU_DRegsVFP], 0, Required>,
|
||||
|
|
Loading…
Reference in New Issue