1 //====- X86InstrMMX.td - Describe the X86 Instruction Set --*- tablegen -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file was developed by the Evan Cheng and is distributed under
6 // the University of Illinois Open Source License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the X86 MMX instruction set, defining the instructions,
11 // and properties of the instructions which are needed for code generation,
12 // machine code emission, and analysis.
14 //===----------------------------------------------------------------------===//
16 //===----------------------------------------------------------------------===//
17 // Instruction templates
18 //===----------------------------------------------------------------------===//
20 // MMXI - MMX instructions with TB prefix.
21 // MMX2I - MMX / SSE2 instructions with TB and OpSize prefixes.
22 // MMXIi8 - MMX instructions with ImmT == Imm8 and TB prefix.
23 class MMXI<bits<8> o, Format F, dag ops, string asm, list<dag> pattern>
24 : I<o, F, ops, asm, pattern>, TB, Requires<[HasMMX]>;
25 class MMX2I<bits<8> o, Format F, dag ops, string asm, list<dag> pattern>
26 : I<o, F, ops, asm, pattern>, TB, OpSize, Requires<[HasSSE2]>;
27 class MMXIi8<bits<8> o, Format F, dag ops, string asm, list<dag> pattern>
28 : Ii8<o, F, ops, asm, pattern>, TB, Requires<[HasMMX]>;
30 // Some 'special' instructions
31 def IMPLICIT_DEF_VR64 : I<0, Pseudo, (ops VR64:$dst),
33 [(set VR64:$dst, (v8i8 (undef)))]>,
36 // 64-bit vector undef's.
37 def : Pat<(v8i8 (undef)), (IMPLICIT_DEF_VR64)>;
38 def : Pat<(v4i16 (undef)), (IMPLICIT_DEF_VR64)>;
39 def : Pat<(v2i32 (undef)), (IMPLICIT_DEF_VR64)>;
41 //===----------------------------------------------------------------------===//
42 // MMX Pattern Fragments
43 //===----------------------------------------------------------------------===//
45 def loadv2i32 : PatFrag<(ops node:$ptr), (v2i32 (load node:$ptr))>;
47 //===----------------------------------------------------------------------===//
49 //===----------------------------------------------------------------------===//
51 let isTwoAddress = 1 in {
52 // MMXI_binop_rm - Simple MMX binary operator.
53 multiclass MMXI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
54 ValueType OpVT, bit Commutable = 0> {
55 def rr : MMXI<opc, MRMSrcReg, (ops VR64:$dst, VR64:$src1, VR64:$src2),
56 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2}"),
57 [(set VR64:$dst, (OpVT (OpNode VR64:$src1, VR64:$src2)))]> {
58 let isCommutable = Commutable;
60 def rm : MMXI<opc, MRMSrcMem, (ops VR64:$dst, VR64:$src1, i64mem:$src2),
61 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2}"),
62 [(set VR64:$dst, (OpVT (OpNode VR64:$src1,
64 (loadv2i32 addr:$src2)))))]>;
67 multiclass MMXI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
69 def rr : MMXI<opc, MRMSrcReg, (ops VR64:$dst, VR64:$src1, VR64:$src2),
70 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2}"),
71 [(set VR64:$dst, (IntId VR64:$src1, VR64:$src2))]> {
72 let isCommutable = Commutable;
74 def rm : MMXI<opc, MRMSrcMem, (ops VR64:$dst, VR64:$src1, i64mem:$src2),
75 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2}"),
76 [(set VR64:$dst, (IntId VR64:$src1,
77 (bitconvert (loadv2i32 addr:$src2))))]>;
80 // MMXI_binop_rm_v2i32 - Simple MMX binary operator whose type is v2i32.
82 // FIXME: we could eliminate this and use MMXI_binop_rm instead if tblgen knew
83 // to collapse (bitconvert VT to VT) into its operand.
85 multiclass MMXI_binop_rm_v2i32<bits<8> opc, string OpcodeStr, SDNode OpNode,
87 def rr : MMXI<opc, MRMSrcReg, (ops VR64:$dst, VR64:$src1, VR64:$src2),
88 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2}"),
89 [(set VR64:$dst, (v2i32 (OpNode VR64:$src1, VR64:$src2)))]> {
90 let isCommutable = Commutable;
92 def rm : MMXI<opc, MRMSrcMem, (ops VR64:$dst, VR64:$src1, i64mem:$src2),
93 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2}"),
95 (OpNode VR64:$src1,(loadv2i32 addr:$src2)))]>;
99 //===----------------------------------------------------------------------===//
100 // MMX EMMS Instruction
101 //===----------------------------------------------------------------------===//
103 def EMMS : MMXI<0x77, RawFrm, (ops), "emms", [(int_x86_mmx_emms)]>;
105 //===----------------------------------------------------------------------===//
106 // MMX Scalar Instructions
107 //===----------------------------------------------------------------------===//
109 // Arithmetic Instructions
110 defm MMX_PADDB : MMXI_binop_rm<0xFC, "paddb", add, v8i8, 1>;
111 defm MMX_PADDW : MMXI_binop_rm<0xFD, "paddw", add, v4i16, 1>;
112 defm MMX_PADDD : MMXI_binop_rm<0xFE, "paddd", add, v2i32, 1>;
114 defm MMX_PADDSB : MMXI_binop_rm_int<0xEC, "paddsb" , int_x86_mmx_padds_b, 1>;
115 defm MMX_PADDSW : MMXI_binop_rm_int<0xED, "paddsw" , int_x86_mmx_padds_w, 1>;
117 defm MMX_PADDUSB : MMXI_binop_rm_int<0xDC, "paddusb", int_x86_mmx_paddus_b, 1>;
118 defm MMX_PADDUSW : MMXI_binop_rm_int<0xDD, "paddusw", int_x86_mmx_paddus_w, 1>;
120 defm MMX_PSUBB : MMXI_binop_rm<0xF8, "psubb", sub, v8i8>;
121 defm MMX_PSUBW : MMXI_binop_rm<0xF9, "psubw", sub, v4i16>;
122 defm MMX_PSUBD : MMXI_binop_rm<0xFA, "psubd", sub, v2i32>;
124 defm MMX_PSUBSB : MMXI_binop_rm_int<0xE8, "psubsb" , int_x86_mmx_psubs_b>;
125 defm MMX_PSUBSW : MMXI_binop_rm_int<0xE9, "psubsw" , int_x86_mmx_psubs_w>;
127 defm MMX_PSUBUSB : MMXI_binop_rm_int<0xD8, "psubusb", int_x86_mmx_psubus_b>;
128 defm MMX_PSUBUSW : MMXI_binop_rm_int<0xD9, "psubusw", int_x86_mmx_psubus_w>;
130 defm MMX_PMULLW : MMXI_binop_rm<0xD5, "pmullw", mul, v4i16, 1>;
132 defm MMX_PMULHW : MMXI_binop_rm_int<0xE5, "pmulhw" , int_x86_mmx_pmulh_w , 1>;
133 defm MMX_PMADDWD : MMXI_binop_rm_int<0xF5, "pmaddwd", int_x86_mmx_pmadd_wd, 1>;
135 // Logical Instructions
136 defm MMX_PAND : MMXI_binop_rm_v2i32<0xDB, "pand", and, 1>;
137 defm MMX_POR : MMXI_binop_rm_v2i32<0xEB, "por" , or, 1>;
138 defm MMX_PXOR : MMXI_binop_rm_v2i32<0xEF, "pxor", xor, 1>;
140 let isTwoAddress = 1 in {
141 def MMX_PANDNrr : MMXI<0xDF, MRMSrcReg,
142 (ops VR64:$dst, VR64:$src1, VR64:$src2),
143 "pandn {$src2, $dst|$dst, $src2}",
144 [(set VR64:$dst, (v2i32 (and (vnot VR64:$src1),
146 def MMX_PANDNrm : MMXI<0xDF, MRMSrcMem,
147 (ops VR64:$dst, VR64:$src1, i64mem:$src2),
148 "pandn {$src2, $dst|$dst, $src2}",
149 [(set VR64:$dst, (v2i32 (and (vnot VR64:$src1),
150 (load addr:$src2))))]>;
154 def MOVD64rr : MMXI<0x6E, MRMSrcReg, (ops VR64:$dst, GR32:$src),
155 "movd {$src, $dst|$dst, $src}", []>;
156 def MOVD64rm : MMXI<0x6E, MRMSrcMem, (ops VR64:$dst, i32mem:$src),
157 "movd {$src, $dst|$dst, $src}", []>;
158 def MOVD64mr : MMXI<0x7E, MRMDestMem, (ops i32mem:$dst, VR64:$src),
159 "movd {$src, $dst|$dst, $src}", []>;
161 def MOVQ64rr : MMXI<0x6F, MRMSrcReg, (ops VR64:$dst, VR64:$src),
162 "movq {$src, $dst|$dst, $src}", []>;
163 def MOVQ64rm : MMXI<0x6F, MRMSrcMem, (ops VR64:$dst, i64mem:$src),
164 "movq {$src, $dst|$dst, $src}",
165 [(set VR64:$dst, (loadv2i32 addr:$src))]>;
166 def MOVQ64mr : MMXI<0x7F, MRMDestMem, (ops i64mem:$dst, VR64:$src),
167 "movq {$src, $dst|$dst, $src}",
168 [(store (v2i32 VR64:$src), addr:$dst)]>;
170 // Conversion instructions
171 def CVTPI2PSrr : MMXI<0x2A, MRMSrcReg, (ops VR128:$dst, VR64:$src),
172 "cvtpi2ps {$src, $dst|$dst, $src}", []>;
173 def CVTPI2PSrm : MMXI<0x2A, MRMSrcMem, (ops VR128:$dst, i64mem:$src),
174 "cvtpi2ps {$src, $dst|$dst, $src}", []>;
175 def CVTPI2PDrr : MMX2I<0x2A, MRMSrcReg, (ops VR128:$dst, VR64:$src),
176 "cvtpi2pd {$src, $dst|$dst, $src}", []>;
177 def CVTPI2PDrm : MMX2I<0x2A, MRMSrcMem, (ops VR128:$dst, i64mem:$src),
178 "cvtpi2pd {$src, $dst|$dst, $src}", []>;
179 def CVTTPS2PIrr: I<0x2C, MRMSrcReg, (ops VR64:$dst, VR128:$src),
180 "cvttps2pi {$src, $dst|$dst, $src}", []>, TB,
182 def CVTTPS2PIrm: I<0x2C, MRMSrcMem, (ops VR64:$dst, f64mem:$src),
183 "cvttps2pi {$src, $dst|$dst, $src}", []>, TB,
185 def CVTPS2PIrr : MMXI<0x2D, MRMSrcReg, (ops VR64:$dst, VR128:$src),
186 "cvtps2pi {$src, $dst|$dst, $src}", []>;
187 def CVTPS2PIrm : MMXI<0x2D, MRMSrcMem, (ops VR64:$dst, f64mem:$src),
188 "cvtps2pi {$src, $dst|$dst, $src}", []>;
189 def CVTPD2PIrr : MMX2I<0x2D, MRMSrcReg, (ops VR64:$dst, VR128:$src),
190 "cvtpd2pi {$src, $dst|$dst, $src}", []>;
191 def CVTPD2PIrm : MMX2I<0x2D, MRMSrcMem, (ops VR64:$dst, f128mem:$src),
192 "cvtpd2pi {$src, $dst|$dst, $src}", []>;
194 // Shuffle and unpack instructions
195 def PSHUFWri : MMXIi8<0x70, MRMSrcReg,
196 (ops VR64:$dst, VR64:$src1, i8imm:$src2),
197 "pshufw {$src2, $src1, $dst|$dst, $src1, $src2}", []>;
198 def PSHUFWmi : MMXIi8<0x70, MRMSrcMem,
199 (ops VR64:$dst, i64mem:$src1, i8imm:$src2),
200 "pshufw {$src2, $src1, $dst|$dst, $src1, $src2}", []>;
203 def MOVNTQ : I<0xE7, MRMDestMem, (ops i64mem:$dst, VR64:$src),
204 "movntq {$src, $dst|$dst, $src}", []>, TB,
207 def MASKMOVQ : I<0xF7, MRMDestMem, (ops VR64:$src, VR64:$mask),
208 "maskmovq {$mask, $src|$src, $mask}", []>, TB,
211 //===----------------------------------------------------------------------===//
212 // Non-Instruction Patterns
213 //===----------------------------------------------------------------------===//
215 // Store 64-bit integer vector values.
216 def : Pat<(store (v8i8 VR64:$src), addr:$dst),
217 (MOVQ64mr addr:$dst, VR64:$src)>;
218 def : Pat<(store (v4i16 VR64:$src), addr:$dst),
219 (MOVQ64mr addr:$dst, VR64:$src)>;
222 def : Pat<(v8i8 (bitconvert (v2i32 VR64:$src))), (v8i8 VR64:$src)>;
223 def : Pat<(v8i8 (bitconvert (v4i16 VR64:$src))), (v8i8 VR64:$src)>;
224 def : Pat<(v4i16 (bitconvert (v2i32 VR64:$src))), (v4i16 VR64:$src)>;
225 def : Pat<(v4i16 (bitconvert (v8i8 VR64:$src))), (v4i16 VR64:$src)>;
226 def : Pat<(v2i32 (bitconvert (v4i16 VR64:$src))), (v2i32 VR64:$src)>;
227 def : Pat<(v2i32 (bitconvert (v8i8 VR64:$src))), (v2i32 VR64:$src)>;