Magic-Mirror/llvm-capstone
1
0
Fork 0
mirror of https://github.com/capstone-engine/llvm-capstone.git synced 2025-01-25 18:55:48 +00:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[ARM] f16 conversions

Browse Source 
This is a follow up of r324321, adding f16 <-> f32 and f16 <-> f64 conversion
match patterns.

Differential Revision: https://reviews.llvm.org/D42954

llvm-svn: 324360
This commit is contained in:
Sjoerd Meijer 2018-02-06 16:28:43 +00:00
parent 27721e8617
commit d2718ba95e
2 changed files with 68 additions and 16 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

39
llvm/lib/Target/ARM/ARMInstrVFP.td
View File

@ -675,19 +675,26 @@ def VCVTSD : VFPAI<(outs SPR:$Sd), (ins DPR:$Dm), VFPUnaryFrm,
// Between half, single and double-precision.
def VCVTBHS: ASuI<0b11101, 0b11, 0b0010, 0b01, 0, (outs SPR:$Sd), (ins SPR:$Sm),
/* FIXME */ IIC_fpCVTSH, "vcvtb", ".f32.f16\t$Sd, $Sm",
[ /* intentionally left blank, see rule below */ ]>,
[/* Intentionally left blank, see patterns below */]>,
Requires<[HasFP16]>,
Sched<[WriteFPCVT]>;
def : FullFP16Pat<(f32 (fpextend HPR:$Sm)),
(VCVTBHS (COPY_TO_REGCLASS HPR:$Sm, SPR))>;
def : FP16Pat<(f16_to_fp GPR:$a),
(VCVTBHS (COPY_TO_REGCLASS GPR:$a, SPR))>;
def VCVTBSH: ASuI<0b11101, 0b11, 0b0011, 0b01, 0, (outs SPR:$Sd), (ins SPR:$Sm),
/* FIXME */ IIC_fpCVTHS, "vcvtb", ".f16.f32\t$Sd, $Sm",
[]>,
[/* Intentionally left blank, see patterns below */]>,
Requires<[HasFP16]>,
Sched<[WriteFPCVT]>;
def : FullFP16Pat<(f16 (fpround SPR:$Sm)),
(COPY_TO_REGCLASS (VCVTBSH SPR:$Sm), HPR)>;
def : FP16Pat<(fp_to_f16 SPR:$a),
(i32 (COPY_TO_REGCLASS (VCVTBSH SPR:$a), GPR))>;
def VCVTTHS: ASuI<0b11101, 0b11, 0b0010, 0b11, 0, (outs SPR:$Sd), (ins SPR:$Sm),
/* FIXME */ IIC_fpCVTSH, "vcvtt", ".f32.f16\t$Sd, $Sm",
[/* For disassembly only; pattern left blank */]>,
@ -703,7 +710,8 @@ def VCVTTSH: ASuI<0b11101, 0b11, 0b0011, 0b11, 0, (outs SPR:$Sd), (ins SPR:$Sm),
def VCVTBHD : ADuI<0b11101, 0b11, 0b0010, 0b01, 0,
(outs DPR:$Dd), (ins SPR:$Sm),
NoItinerary, "vcvtb", ".f64.f16\t$Dd, $Sm",
[]>, Requires<[HasFPARMv8, HasDPVFP]>,
[/* Intentionally left blank, see patterns below */]>,
Requires<[HasFPARMv8, HasDPVFP]>,
Sched<[WriteFPCVT]> {
// Instruction operands.
bits<5> Sm;
@ -713,10 +721,16 @@ def VCVTBHD : ADuI<0b11101, 0b11, 0b0010, 0b01, 0,
let Inst{5} = Sm{0};
}
def : FullFP16Pat<(f64 (fpextend HPR:$Sm)),
(VCVTBHD (COPY_TO_REGCLASS HPR:$Sm, SPR))>;
def : FP16Pat<(f64 (f16_to_fp GPR:$a)),
(VCVTBHD (COPY_TO_REGCLASS GPR:$a, SPR))>;
def VCVTBDH : ADuI<0b11101, 0b11, 0b0011, 0b01, 0,
(outs SPR:$Sd), (ins DPR:$Dm),
NoItinerary, "vcvtb", ".f16.f64\t$Sd, $Dm",
[]>, Requires<[HasFPARMv8, HasDPVFP]> {
[/* Intentionally left blank, see patterns below */]>,
Requires<[HasFPARMv8, HasDPVFP]> {
// Instruction operands.
bits<5> Sd;
bits<5> Dm;
@ -728,6 +742,11 @@ def VCVTBDH : ADuI<0b11101, 0b11, 0b0011, 0b01, 0,
let Inst{22} = Sd{0};
}
def : FullFP16Pat<(f16 (fpround DPR:$Dm)),
(COPY_TO_REGCLASS (VCVTBDH DPR:$Dm), HPR)>;
def : FP16Pat<(fp_to_f16 (f64 DPR:$a)),
(i32 (COPY_TO_REGCLASS (VCVTBDH DPR:$a), GPR))>;
def VCVTTHD : ADuI<0b11101, 0b11, 0b0010, 0b11, 0,
(outs DPR:$Dd), (ins SPR:$Sm),
NoItinerary, "vcvtt", ".f64.f16\t$Dd, $Sm",
@ -755,18 +774,6 @@ def VCVTTDH : ADuI<0b11101, 0b11, 0b0011, 0b11, 0,
let Inst{5} = Dm{4};
}
def : FP16Pat<(fp_to_f16 SPR:$a),
(i32 (COPY_TO_REGCLASS (VCVTBSH SPR:$a), GPR))>;
def : FP16Pat<(fp_to_f16 (f64 DPR:$a)),
(i32 (COPY_TO_REGCLASS (VCVTBDH DPR:$a), GPR))>;
def : FP16Pat<(f16_to_fp GPR:$a),
(VCVTBHS (COPY_TO_REGCLASS GPR:$a, SPR))>;
def : FP16Pat<(f64 (f16_to_fp GPR:$a)),
(VCVTBHD (COPY_TO_REGCLASS GPR:$a, SPR))>;
multiclass vcvt_inst<string opc, bits<2> rm,
SDPatternOperator node = null_frag> {
let PostEncoderMethod = "", DecoderNamespace = "VFPV8" in {

45
llvm/test/CodeGen/ARM/fp16-instructions.ll
View File

@ -212,6 +212,51 @@ entry:
; CHECK-HARDFP-FULLFP16-NEXT: vcvt.f16.s32 s0, s0
}
define i32 @f2h(float %f) {
entry:
%conv = fptrunc float %f to half
%0 = bitcast half %conv to i16
%tmp.0.insert.ext = zext i16 %0 to i32
ret i32 %tmp.0.insert.ext
; CHECK-LABEL: f2h:
; CHECK-HARDFP-FULLFP16: vcvtb.f16.f32 s0, s0
}
define float @h2f(i32 %h.coerce) {
entry:
%tmp.0.extract.trunc = trunc i32 %h.coerce to i16
%0 = bitcast i16 %tmp.0.extract.trunc to half
%conv = fpext half %0 to float
ret float %conv
; CHECK-LABEL: h2f:
; CHECK-HARDFP-FULLFP16: vcvtb.f32.f16 s0, s0
}
define double @h2d(i32 %h.coerce) {
entry:
%tmp.0.extract.trunc = trunc i32 %h.coerce to i16
%0 = bitcast i16 %tmp.0.extract.trunc to half
%conv = fpext half %0 to double
ret double %conv
; CHECK-LABEL: h2d:
; CHECK-HARDFP-FULLFP16: vcvtb.f64.f16 d{{.*}}, s{{.}}
}
define i32 @d2h(double %d) {
entry:
%conv = fptrunc double %d to half
%0 = bitcast half %conv to i16
%tmp.0.insert.ext = zext i16 %0 to i32
ret i32 %tmp.0.insert.ext
; CHECK-LABEL: d2h:
; CHECK-HARDFP-FULLFP16: vcvtb.f16.f64 s0, d{{.*}}
}
; TODO:
; 7. VCVTA
; 8. VCVTM