blob: d868773d2d690cc952f80635845ac7f69ee2c6cf [file] [log] [blame]
Bruno Cardoso Lopes6b7e9162010-07-23 00:54:35 +00001//====- X86InstrFMA.td - Describe the X86 Instruction Set --*- tablegen -*-===//
2//
3// The LLVM Compiler Infrastructure
4//
5// This file is distributed under the University of Illinois Open Source
6// License. See LICENSE.TXT for details.
7//
8//===----------------------------------------------------------------------===//
9//
10// This file describes FMA (Fused Multiply-Add) instructions.
11//
12//===----------------------------------------------------------------------===//
13
14//===----------------------------------------------------------------------===//
15// FMA3 - Intel 3 operand Fused Multiply-Add instructions
16//===----------------------------------------------------------------------===//
17
18multiclass fma_rm<bits<8> opc, string OpcodeStr> {
19 def r : FMA3<opc, MRMSrcReg, (outs VR128:$dst),
20 (ins VR128:$src1, VR128:$src2),
21 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
22 []>;
23 def m : FMA3<opc, MRMSrcMem, (outs VR128:$dst),
24 (ins VR128:$src1, f128mem:$src2),
25 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
26 []>;
27 def rY : FMA3<opc, MRMSrcReg, (outs VR256:$dst),
28 (ins VR256:$src1, VR256:$src2),
29 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
30 []>;
31 def mY : FMA3<opc, MRMSrcMem, (outs VR256:$dst),
32 (ins VR256:$src1, f256mem:$src2),
33 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
34 []>;
35}
36
37multiclass fma_forms<bits<8> opc132, bits<8> opc213, bits<8> opc231,
38 string OpcodeStr, string PackTy> {
39 defm r132 : fma_rm<opc132, !strconcat(OpcodeStr, !strconcat("132", PackTy))>;
40 defm r213 : fma_rm<opc213, !strconcat(OpcodeStr, !strconcat("213", PackTy))>;
41 defm r231 : fma_rm<opc231, !strconcat(OpcodeStr, !strconcat("231", PackTy))>;
42}
43
44let isAsmParserOnly = 1 in {
45 // Fused Multiply-Add
46 defm VFMADDPS : fma_forms<0x98, 0xA8, 0xB8, "vfmadd", "ps">;
47 defm VFMADDPD : fma_forms<0x98, 0xA8, 0xB8, "vfmadd", "pd">, VEX_W;
48 defm VFMADDSUBPS : fma_forms<0x96, 0xA6, 0xB6, "vfmaddsub", "ps">;
49 defm VFMADDSUBPD : fma_forms<0x96, 0xA6, 0xB6, "vfmaddsub", "pd">, VEX_W;
50 defm VFMSUBADDPS : fma_forms<0x97, 0xA7, 0xB7, "vfmsubadd", "ps">;
51 defm VFMSUBADDPD : fma_forms<0x97, 0xA7, 0xB7, "vfmsubadd", "pd">, VEX_W;
52 defm VFMSUBPS : fma_forms<0x9A, 0xAA, 0xBA, "vfmsub", "ps">;
53 defm VFMSUBPD : fma_forms<0x9A, 0xAA, 0xBA, "vfmsub", "pd">, VEX_W;
54
55 // Fused Negative Multiply-Add
56 defm VFNMADDPS : fma_forms<0x9C, 0xAC, 0xBC, "vfnmadd", "ps">;
57 defm VFNMADDPD : fma_forms<0x9C, 0xAC, 0xBC, "vfnmadd", "pd">, VEX_W;
58 defm VFNMSUBPS : fma_forms<0x9E, 0xAE, 0xBE, "vfnmsub", "ps">;
59 defm VFNMSUBPD : fma_forms<0x9E, 0xAE, 0xBE, "vfnmsub", "pd">, VEX_W;
60}