1 //===- X86InstrFPStack.td - FPU Instruction Set ------------*- tablegen -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the X86 x87 FPU instruction set, defining the
11 // instructions, and properties of the instructions which are needed for code
12 // generation, machine code emission, and analysis.
14 //===----------------------------------------------------------------------===//
16 //===----------------------------------------------------------------------===//
17 // FPStack specific DAG Nodes.
18 //===----------------------------------------------------------------------===//
20 def SDTX86FpGet2 : SDTypeProfile<2, 0, [SDTCisVT<0, f80>,
22 def SDTX86Fld : SDTypeProfile<1, 2, [SDTCisFP<0>,
24 SDTCisVT<2, OtherVT>]>;
25 def SDTX86Fst : SDTypeProfile<0, 3, [SDTCisFP<0>,
27 SDTCisVT<2, OtherVT>]>;
28 def SDTX86Fild : SDTypeProfile<1, 2, [SDTCisFP<0>, SDTCisPtrTy<1>,
29 SDTCisVT<2, OtherVT>]>;
30 def SDTX86Fnstsw : SDTypeProfile<1, 1, [SDTCisVT<0, i16>, SDTCisVT<1, i16>]>;
31 def SDTX86FpToIMem : SDTypeProfile<0, 2, [SDTCisFP<0>, SDTCisPtrTy<1>]>;
33 def SDTX86CwdStore : SDTypeProfile<0, 1, [SDTCisPtrTy<0>]>;
35 def X86fld : SDNode<"X86ISD::FLD", SDTX86Fld,
36 [SDNPHasChain, SDNPMayLoad, SDNPMemOperand]>;
37 def X86fst : SDNode<"X86ISD::FST", SDTX86Fst,
38 [SDNPHasChain, SDNPInGlue, SDNPMayStore,
40 def X86fild : SDNode<"X86ISD::FILD", SDTX86Fild,
41 [SDNPHasChain, SDNPMayLoad, SDNPMemOperand]>;
42 def X86fildflag : SDNode<"X86ISD::FILD_FLAG", SDTX86Fild,
43 [SDNPHasChain, SDNPOutGlue, SDNPMayLoad,
45 def X86fp_stsw : SDNode<"X86ISD::FNSTSW16r", SDTX86Fnstsw>;
46 def X86fp_to_i16mem : SDNode<"X86ISD::FP_TO_INT16_IN_MEM", SDTX86FpToIMem,
47 [SDNPHasChain, SDNPMayStore, SDNPMemOperand]>;
48 def X86fp_to_i32mem : SDNode<"X86ISD::FP_TO_INT32_IN_MEM", SDTX86FpToIMem,
49 [SDNPHasChain, SDNPMayStore, SDNPMemOperand]>;
50 def X86fp_to_i64mem : SDNode<"X86ISD::FP_TO_INT64_IN_MEM", SDTX86FpToIMem,
51 [SDNPHasChain, SDNPMayStore, SDNPMemOperand]>;
52 def X86fp_cwd_get16 : SDNode<"X86ISD::FNSTCW16m", SDTX86CwdStore,
53 [SDNPHasChain, SDNPMayStore, SDNPSideEffect,
56 //===----------------------------------------------------------------------===//
57 // FPStack pattern fragments
58 //===----------------------------------------------------------------------===//
60 def fpimm0 : PatLeaf<(fpimm), [{
61 return N->isExactlyValue(+0.0);
64 def fpimmneg0 : PatLeaf<(fpimm), [{
65 return N->isExactlyValue(-0.0);
68 def fpimm1 : PatLeaf<(fpimm), [{
69 return N->isExactlyValue(+1.0);
72 def fpimmneg1 : PatLeaf<(fpimm), [{
73 return N->isExactlyValue(-1.0);
76 // Some 'special' instructions
77 let usesCustomInserter = 1 in { // Expanded after instruction selection.
78 def FP32_TO_INT16_IN_MEM : PseudoI<(outs), (ins i16mem:$dst, RFP32:$src),
79 [(X86fp_to_i16mem RFP32:$src, addr:$dst)]>;
80 def FP32_TO_INT32_IN_MEM : PseudoI<(outs), (ins i32mem:$dst, RFP32:$src),
81 [(X86fp_to_i32mem RFP32:$src, addr:$dst)]>;
82 def FP32_TO_INT64_IN_MEM : PseudoI<(outs), (ins i64mem:$dst, RFP32:$src),
83 [(X86fp_to_i64mem RFP32:$src, addr:$dst)]>;
84 def FP64_TO_INT16_IN_MEM : PseudoI<(outs), (ins i16mem:$dst, RFP64:$src),
85 [(X86fp_to_i16mem RFP64:$src, addr:$dst)]>;
86 def FP64_TO_INT32_IN_MEM : PseudoI<(outs), (ins i32mem:$dst, RFP64:$src),
87 [(X86fp_to_i32mem RFP64:$src, addr:$dst)]>;
88 def FP64_TO_INT64_IN_MEM : PseudoI<(outs), (ins i64mem:$dst, RFP64:$src),
89 [(X86fp_to_i64mem RFP64:$src, addr:$dst)]>;
90 def FP80_TO_INT16_IN_MEM : PseudoI<(outs), (ins i16mem:$dst, RFP80:$src),
91 [(X86fp_to_i16mem RFP80:$src, addr:$dst)]>;
92 def FP80_TO_INT32_IN_MEM : PseudoI<(outs), (ins i32mem:$dst, RFP80:$src),
93 [(X86fp_to_i32mem RFP80:$src, addr:$dst)]>;
94 def FP80_TO_INT64_IN_MEM : PseudoI<(outs), (ins i64mem:$dst, RFP80:$src),
95 [(X86fp_to_i64mem RFP80:$src, addr:$dst)]>;
98 // All FP Stack operations are represented with four instructions here. The
99 // first three instructions, generated by the instruction selector, use "RFP32"
100 // "RFP64" or "RFP80" registers: traditional register files to reference 32-bit,
101 // 64-bit or 80-bit floating point values. These sizes apply to the values,
102 // not the registers, which are always 80 bits; RFP32, RFP64 and RFP80 can be
103 // copied to each other without losing information. These instructions are all
104 // pseudo instructions and use the "_Fp" suffix.
105 // In some cases there are additional variants with a mixture of different
107 // The second instruction is defined with FPI, which is the actual instruction
108 // emitted by the assembler. These use "RST" registers, although frequently
109 // the actual register(s) used are implicit. These are always 80 bits.
110 // The FP stackifier pass converts one to the other after register allocation
113 // Note that the FpI instruction should have instruction selection info (e.g.
114 // a pattern) and the FPI instruction should have emission info (e.g. opcode
115 // encoding and asm printing info).
117 // FpIf32, FpIf64 - Floating Point Pseudo Instruction template.
118 // f32 instructions can use SSE1 and are predicated on FPStackf32 == !SSE1.
119 // f64 instructions can use SSE2 and are predicated on FPStackf64 == !SSE2.
120 // f80 instructions cannot use SSE and use neither of these.
121 class FpIf32<dag outs, dag ins, FPFormat fp, list<dag> pattern> :
122 FpI_<outs, ins, fp, pattern>, Requires<[FPStackf32]>;
123 class FpIf64<dag outs, dag ins, FPFormat fp, list<dag> pattern> :
124 FpI_<outs, ins, fp, pattern>, Requires<[FPStackf64]>;
126 // Factoring for arithmetic.
127 multiclass FPBinary_rr<SDNode OpNode> {
128 // Register op register -> register
129 // These are separated out because they have no reversed form.
130 def _Fp32 : FpIf32<(outs RFP32:$dst), (ins RFP32:$src1, RFP32:$src2), TwoArgFP,
131 [(set RFP32:$dst, (OpNode RFP32:$src1, RFP32:$src2))]>;
132 def _Fp64 : FpIf64<(outs RFP64:$dst), (ins RFP64:$src1, RFP64:$src2), TwoArgFP,
133 [(set RFP64:$dst, (OpNode RFP64:$src1, RFP64:$src2))]>;
134 def _Fp80 : FpI_<(outs RFP80:$dst), (ins RFP80:$src1, RFP80:$src2), TwoArgFP,
135 [(set RFP80:$dst, (OpNode RFP80:$src1, RFP80:$src2))]>;
137 // The FopST0 series are not included here because of the irregularities
138 // in where the 'r' goes in assembly output.
139 // These instructions cannot address 80-bit memory.
140 multiclass FPBinary<SDNode OpNode, Format fp, string asmstring,
142 // ST(0) = ST(0) + [mem]
143 def _Fp32m : FpIf32<(outs RFP32:$dst),
144 (ins RFP32:$src1, f32mem:$src2), OneArgFPRW,
147 (OpNode RFP32:$src1, (loadf32 addr:$src2))),
149 (OpNode (loadf32 addr:$src2), RFP32:$src1)))]>;
150 def _Fp64m : FpIf64<(outs RFP64:$dst),
151 (ins RFP64:$src1, f64mem:$src2), OneArgFPRW,
154 (OpNode RFP64:$src1, (loadf64 addr:$src2))),
156 (OpNode (loadf64 addr:$src2), RFP64:$src1)))]>;
157 def _Fp64m32: FpIf64<(outs RFP64:$dst),
158 (ins RFP64:$src1, f32mem:$src2), OneArgFPRW,
161 (OpNode RFP64:$src1, (f64 (extloadf32 addr:$src2)))),
163 (OpNode (f64 (extloadf32 addr:$src2)), RFP64:$src1)))]>;
164 def _Fp80m32: FpI_<(outs RFP80:$dst),
165 (ins RFP80:$src1, f32mem:$src2), OneArgFPRW,
168 (OpNode RFP80:$src1, (f80 (extloadf32 addr:$src2)))),
170 (OpNode (f80 (extloadf32 addr:$src2)), RFP80:$src1)))]>;
171 def _Fp80m64: FpI_<(outs RFP80:$dst),
172 (ins RFP80:$src1, f64mem:$src2), OneArgFPRW,
175 (OpNode RFP80:$src1, (f80 (extloadf64 addr:$src2)))),
177 (OpNode (f80 (extloadf64 addr:$src2)), RFP80:$src1)))]>;
179 def _F32m : FPI<0xD8, fp, (outs), (ins f32mem:$src),
180 !strconcat("f", asmstring, "{s}\t$src")>;
182 def _F64m : FPI<0xDC, fp, (outs), (ins f64mem:$src),
183 !strconcat("f", asmstring, "{l}\t$src")>;
184 // ST(0) = ST(0) + [memint]
185 def _FpI16m32 : FpIf32<(outs RFP32:$dst), (ins RFP32:$src1, i16mem:$src2),
189 (OpNode RFP32:$src1, (X86fild addr:$src2, i16))),
191 (OpNode (X86fild addr:$src2, i16), RFP32:$src1)))]>;
192 def _FpI32m32 : FpIf32<(outs RFP32:$dst), (ins RFP32:$src1, i32mem:$src2),
196 (OpNode RFP32:$src1, (X86fild addr:$src2, i32))),
198 (OpNode (X86fild addr:$src2, i32), RFP32:$src1)))]>;
199 def _FpI16m64 : FpIf64<(outs RFP64:$dst), (ins RFP64:$src1, i16mem:$src2),
203 (OpNode RFP64:$src1, (X86fild addr:$src2, i16))),
205 (OpNode (X86fild addr:$src2, i16), RFP64:$src1)))]>;
206 def _FpI32m64 : FpIf64<(outs RFP64:$dst), (ins RFP64:$src1, i32mem:$src2),
210 (OpNode RFP64:$src1, (X86fild addr:$src2, i32))),
212 (OpNode (X86fild addr:$src2, i32), RFP64:$src1)))]>;
213 def _FpI16m80 : FpI_<(outs RFP80:$dst), (ins RFP80:$src1, i16mem:$src2),
217 (OpNode RFP80:$src1, (X86fild addr:$src2, i16))),
219 (OpNode (X86fild addr:$src2, i16), RFP80:$src1)))]>;
220 def _FpI32m80 : FpI_<(outs RFP80:$dst), (ins RFP80:$src1, i32mem:$src2),
224 (OpNode RFP80:$src1, (X86fild addr:$src2, i32))),
226 (OpNode (X86fild addr:$src2, i32), RFP80:$src1)))]>;
228 def _FI16m : FPI<0xDE, fp, (outs), (ins i16mem:$src),
229 !strconcat("fi", asmstring, "{s}\t$src")>;
231 def _FI32m : FPI<0xDA, fp, (outs), (ins i32mem:$src),
232 !strconcat("fi", asmstring, "{l}\t$src")>;
235 let Defs = [FPSW] in {
236 // FPBinary_rr just defines pseudo-instructions, no need to set a scheduling
238 defm ADD : FPBinary_rr<fadd>;
239 defm SUB : FPBinary_rr<fsub>;
240 defm MUL : FPBinary_rr<fmul>;
241 defm DIV : FPBinary_rr<fdiv>;
242 // Sets the scheduling resources for the actual NAME#_F<size>m defintions.
243 let SchedRW = [WriteFAddLd] in {
244 defm ADD : FPBinary<fadd, MRM0m, "add">;
245 defm SUB : FPBinary<fsub, MRM4m, "sub">;
246 defm SUBR: FPBinary<fsub ,MRM5m, "subr", 0>;
248 let SchedRW = [WriteFMulLd] in {
249 defm MUL : FPBinary<fmul, MRM1m, "mul">;
251 let SchedRW = [WriteFDivLd] in {
252 defm DIV : FPBinary<fdiv, MRM6m, "div">;
253 defm DIVR: FPBinary<fdiv, MRM7m, "divr", 0>;
257 class FPST0rInst<Format fp, string asm>
258 : FPI<0xD8, fp, (outs), (ins RST:$op), asm>;
259 class FPrST0Inst<Format fp, string asm>
260 : FPI<0xDC, fp, (outs), (ins RST:$op), asm>;
261 class FPrST0PInst<Format fp, string asm>
262 : FPI<0xDE, fp, (outs), (ins RST:$op), asm>;
264 // NOTE: GAS and apparently all other AT&T style assemblers have a broken notion
265 // of some of the 'reverse' forms of the fsub and fdiv instructions. As such,
266 // we have to put some 'r's in and take them out of weird places.
267 let SchedRW = [WriteFAdd] in {
268 def ADD_FST0r : FPST0rInst <MRM0r, "fadd\t$op">;
269 def ADD_FrST0 : FPrST0Inst <MRM0r, "fadd\t{%st(0), $op|$op, st(0)}">;
270 def ADD_FPrST0 : FPrST0PInst<MRM0r, "faddp\t$op">;
271 def SUBR_FST0r : FPST0rInst <MRM5r, "fsubr\t$op">;
272 def SUB_FrST0 : FPrST0Inst <MRM5r, "fsub{r}\t{%st(0), $op|$op, st(0)}">;
273 def SUB_FPrST0 : FPrST0PInst<MRM5r, "fsub{r}p\t$op">;
274 def SUB_FST0r : FPST0rInst <MRM4r, "fsub\t$op">;
275 def SUBR_FrST0 : FPrST0Inst <MRM4r, "fsub{|r}\t{%st(0), $op|$op, st(0)}">;
276 def SUBR_FPrST0 : FPrST0PInst<MRM4r, "fsub{|r}p\t$op">;
278 let SchedRW = [WriteFMul] in {
279 def MUL_FST0r : FPST0rInst <MRM1r, "fmul\t$op">;
280 def MUL_FrST0 : FPrST0Inst <MRM1r, "fmul\t{%st(0), $op|$op, st(0)}">;
281 def MUL_FPrST0 : FPrST0PInst<MRM1r, "fmulp\t$op">;
283 let SchedRW = [WriteFDiv] in {
284 def DIVR_FST0r : FPST0rInst <MRM7r, "fdivr\t$op">;
285 def DIV_FrST0 : FPrST0Inst <MRM7r, "fdiv{r}\t{%st(0), $op|$op, st(0)}">;
286 def DIV_FPrST0 : FPrST0PInst<MRM7r, "fdiv{r}p\t$op">;
287 def DIV_FST0r : FPST0rInst <MRM6r, "fdiv\t$op">;
288 def DIVR_FrST0 : FPrST0Inst <MRM6r, "fdiv{|r}\t{%st(0), $op|$op, st(0)}">;
289 def DIVR_FPrST0 : FPrST0PInst<MRM6r, "fdiv{|r}p\t$op">;
292 def COM_FST0r : FPST0rInst <MRM2r, "fcom\t$op">;
293 def COMP_FST0r : FPST0rInst <MRM3r, "fcomp\t$op">;
296 multiclass FPUnary<SDNode OpNode, Format fp, string asmstring> {
297 def _Fp32 : FpIf32<(outs RFP32:$dst), (ins RFP32:$src), OneArgFPRW,
298 [(set RFP32:$dst, (OpNode RFP32:$src))]>;
299 def _Fp64 : FpIf64<(outs RFP64:$dst), (ins RFP64:$src), OneArgFPRW,
300 [(set RFP64:$dst, (OpNode RFP64:$src))]>;
301 def _Fp80 : FpI_<(outs RFP80:$dst), (ins RFP80:$src), OneArgFPRW,
302 [(set RFP80:$dst, (OpNode RFP80:$src))]>;
303 def _F : FPI<0xD9, fp, (outs), (ins), asmstring>;
306 let Defs = [FPSW] in {
307 defm CHS : FPUnary<fneg, MRM_E0, "fchs">;
308 defm ABS : FPUnary<fabs, MRM_E1, "fabs">;
309 let SchedRW = [WriteFSqrt] in {
310 defm SQRT: FPUnary<fsqrt,MRM_FA, "fsqrt">;
312 defm SIN : FPUnary<fsin, MRM_FE, "fsin">;
313 defm COS : FPUnary<fcos, MRM_FF, "fcos">;
315 let hasSideEffects = 0 in {
316 def TST_Fp32 : FpIf32<(outs), (ins RFP32:$src), OneArgFP, []>;
317 def TST_Fp64 : FpIf64<(outs), (ins RFP64:$src), OneArgFP, []>;
318 def TST_Fp80 : FpI_<(outs), (ins RFP80:$src), OneArgFP, []>;
320 def TST_F : FPI<0xD9, MRM_E4, (outs), (ins), "ftst">;
323 // Versions of FP instructions that take a single memory operand. Added for the
324 // disassembler; remove as they are included with patterns elsewhere.
325 def FCOM32m : FPI<0xD8, MRM2m, (outs), (ins f32mem:$src), "fcom{s}\t$src">;
326 def FCOMP32m : FPI<0xD8, MRM3m, (outs), (ins f32mem:$src), "fcomp{s}\t$src">;
328 def FLDENVm : FPI<0xD9, MRM4m, (outs), (ins f32mem:$src), "fldenv\t$src">;
329 def FSTENVm : FPI<0xD9, MRM6m, (outs f32mem:$dst), (ins), "fnstenv\t$dst">;
331 def FICOM32m : FPI<0xDA, MRM2m, (outs), (ins i32mem:$src), "ficom{l}\t$src">;
332 def FICOMP32m: FPI<0xDA, MRM3m, (outs), (ins i32mem:$src), "ficomp{l}\t$src">;
334 def FCOM64m : FPI<0xDC, MRM2m, (outs), (ins f64mem:$src), "fcom{l}\t$src">;
335 def FCOMP64m : FPI<0xDC, MRM3m, (outs), (ins f64mem:$src), "fcomp{l}\t$src">;
337 def FRSTORm : FPI<0xDD, MRM4m, (outs f32mem:$dst), (ins), "frstor\t$dst">;
338 def FSAVEm : FPI<0xDD, MRM6m, (outs f32mem:$dst), (ins), "fnsave\t$dst">;
339 def FNSTSWm : FPI<0xDD, MRM7m, (outs i16mem:$dst), (ins), "fnstsw\t$dst">;
341 def FICOM16m : FPI<0xDE, MRM2m, (outs), (ins i16mem:$src), "ficom{s}\t$src">;
342 def FICOMP16m: FPI<0xDE, MRM3m, (outs), (ins i16mem:$src), "ficomp{s}\t$src">;
344 def FBLDm : FPI<0xDF, MRM4m, (outs), (ins f80mem:$src), "fbld\t$src">;
345 def FBSTPm : FPI<0xDF, MRM6m, (outs f80mem:$dst), (ins), "fbstp\t$dst">;
347 // Floating point cmovs.
348 class FpIf32CMov<dag outs, dag ins, FPFormat fp, list<dag> pattern> :
349 FpI_<outs, ins, fp, pattern>, Requires<[FPStackf32, HasCMov]>;
350 class FpIf64CMov<dag outs, dag ins, FPFormat fp, list<dag> pattern> :
351 FpI_<outs, ins, fp, pattern>, Requires<[FPStackf64, HasCMov]>;
353 multiclass FPCMov<PatLeaf cc> {
354 def _Fp32 : FpIf32CMov<(outs RFP32:$dst), (ins RFP32:$src1, RFP32:$src2),
356 [(set RFP32:$dst, (X86cmov RFP32:$src1, RFP32:$src2,
358 def _Fp64 : FpIf64CMov<(outs RFP64:$dst), (ins RFP64:$src1, RFP64:$src2),
360 [(set RFP64:$dst, (X86cmov RFP64:$src1, RFP64:$src2,
362 def _Fp80 : FpI_<(outs RFP80:$dst), (ins RFP80:$src1, RFP80:$src2),
364 [(set RFP80:$dst, (X86cmov RFP80:$src1, RFP80:$src2,
369 let Defs = [FPSW] in {
370 let Uses = [EFLAGS], Constraints = "$src1 = $dst" in {
371 defm CMOVB : FPCMov<X86_COND_B>;
372 defm CMOVBE : FPCMov<X86_COND_BE>;
373 defm CMOVE : FPCMov<X86_COND_E>;
374 defm CMOVP : FPCMov<X86_COND_P>;
375 defm CMOVNB : FPCMov<X86_COND_AE>;
376 defm CMOVNBE: FPCMov<X86_COND_A>;
377 defm CMOVNE : FPCMov<X86_COND_NE>;
378 defm CMOVNP : FPCMov<X86_COND_NP>;
379 } // Uses = [EFLAGS], Constraints = "$src1 = $dst"
381 let Predicates = [HasCMov] in {
382 // These are not factored because there's no clean way to pass DA/DB.
383 def CMOVB_F : FPI<0xDA, MRM0r, (outs), (ins RST:$op),
384 "fcmovb\t{$op, %st(0)|st(0), $op}">;
385 def CMOVBE_F : FPI<0xDA, MRM2r, (outs), (ins RST:$op),
386 "fcmovbe\t{$op, %st(0)|st(0), $op}">;
387 def CMOVE_F : FPI<0xDA, MRM1r, (outs), (ins RST:$op),
388 "fcmove\t{$op, %st(0)|st(0), $op}">;
389 def CMOVP_F : FPI<0xDA, MRM3r, (outs), (ins RST:$op),
390 "fcmovu\t{$op, %st(0)|st(0), $op}">;
391 def CMOVNB_F : FPI<0xDB, MRM0r, (outs), (ins RST:$op),
392 "fcmovnb\t{$op, %st(0)|st(0), $op}">;
393 def CMOVNBE_F: FPI<0xDB, MRM2r, (outs), (ins RST:$op),
394 "fcmovnbe\t{$op, %st(0)|st(0), $op}">;
395 def CMOVNE_F : FPI<0xDB, MRM1r, (outs), (ins RST:$op),
396 "fcmovne\t{$op, %st(0)|st(0), $op}">;
397 def CMOVNP_F : FPI<0xDB, MRM3r, (outs), (ins RST:$op),
398 "fcmovnu\t{$op, %st(0)|st(0), $op}">;
399 } // Predicates = [HasCMov]
401 // Floating point loads & stores.
402 let canFoldAsLoad = 1 in {
403 def LD_Fp32m : FpIf32<(outs RFP32:$dst), (ins f32mem:$src), ZeroArgFP,
404 [(set RFP32:$dst, (loadf32 addr:$src))]>;
405 let isReMaterializable = 1 in
406 def LD_Fp64m : FpIf64<(outs RFP64:$dst), (ins f64mem:$src), ZeroArgFP,
407 [(set RFP64:$dst, (loadf64 addr:$src))]>;
408 def LD_Fp80m : FpI_<(outs RFP80:$dst), (ins f80mem:$src), ZeroArgFP,
409 [(set RFP80:$dst, (loadf80 addr:$src))]>;
411 def LD_Fp32m64 : FpIf64<(outs RFP64:$dst), (ins f32mem:$src), ZeroArgFP,
412 [(set RFP64:$dst, (f64 (extloadf32 addr:$src)))]>;
413 def LD_Fp64m80 : FpI_<(outs RFP80:$dst), (ins f64mem:$src), ZeroArgFP,
414 [(set RFP80:$dst, (f80 (extloadf64 addr:$src)))]>;
415 def LD_Fp32m80 : FpI_<(outs RFP80:$dst), (ins f32mem:$src), ZeroArgFP,
416 [(set RFP80:$dst, (f80 (extloadf32 addr:$src)))]>;
417 def ILD_Fp16m32: FpIf32<(outs RFP32:$dst), (ins i16mem:$src), ZeroArgFP,
418 [(set RFP32:$dst, (X86fild addr:$src, i16))]>;
419 def ILD_Fp32m32: FpIf32<(outs RFP32:$dst), (ins i32mem:$src), ZeroArgFP,
420 [(set RFP32:$dst, (X86fild addr:$src, i32))]>;
421 def ILD_Fp64m32: FpIf32<(outs RFP32:$dst), (ins i64mem:$src), ZeroArgFP,
422 [(set RFP32:$dst, (X86fild addr:$src, i64))]>;
423 def ILD_Fp16m64: FpIf64<(outs RFP64:$dst), (ins i16mem:$src), ZeroArgFP,
424 [(set RFP64:$dst, (X86fild addr:$src, i16))]>;
425 def ILD_Fp32m64: FpIf64<(outs RFP64:$dst), (ins i32mem:$src), ZeroArgFP,
426 [(set RFP64:$dst, (X86fild addr:$src, i32))]>;
427 def ILD_Fp64m64: FpIf64<(outs RFP64:$dst), (ins i64mem:$src), ZeroArgFP,
428 [(set RFP64:$dst, (X86fild addr:$src, i64))]>;
429 def ILD_Fp16m80: FpI_<(outs RFP80:$dst), (ins i16mem:$src), ZeroArgFP,
430 [(set RFP80:$dst, (X86fild addr:$src, i16))]>;
431 def ILD_Fp32m80: FpI_<(outs RFP80:$dst), (ins i32mem:$src), ZeroArgFP,
432 [(set RFP80:$dst, (X86fild addr:$src, i32))]>;
433 def ILD_Fp64m80: FpI_<(outs RFP80:$dst), (ins i64mem:$src), ZeroArgFP,
434 [(set RFP80:$dst, (X86fild addr:$src, i64))]>;
436 def ST_Fp32m : FpIf32<(outs), (ins f32mem:$op, RFP32:$src), OneArgFP,
437 [(store RFP32:$src, addr:$op)]>;
438 def ST_Fp64m32 : FpIf64<(outs), (ins f32mem:$op, RFP64:$src), OneArgFP,
439 [(truncstoref32 RFP64:$src, addr:$op)]>;
440 def ST_Fp64m : FpIf64<(outs), (ins f64mem:$op, RFP64:$src), OneArgFP,
441 [(store RFP64:$src, addr:$op)]>;
442 def ST_Fp80m32 : FpI_<(outs), (ins f32mem:$op, RFP80:$src), OneArgFP,
443 [(truncstoref32 RFP80:$src, addr:$op)]>;
444 def ST_Fp80m64 : FpI_<(outs), (ins f64mem:$op, RFP80:$src), OneArgFP,
445 [(truncstoref64 RFP80:$src, addr:$op)]>;
446 // FST does not support 80-bit memory target; FSTP must be used.
448 let mayStore = 1, hasSideEffects = 0 in {
449 def ST_FpP32m : FpIf32<(outs), (ins f32mem:$op, RFP32:$src), OneArgFP, []>;
450 def ST_FpP64m32 : FpIf64<(outs), (ins f32mem:$op, RFP64:$src), OneArgFP, []>;
451 def ST_FpP64m : FpIf64<(outs), (ins f64mem:$op, RFP64:$src), OneArgFP, []>;
452 def ST_FpP80m32 : FpI_<(outs), (ins f32mem:$op, RFP80:$src), OneArgFP, []>;
453 def ST_FpP80m64 : FpI_<(outs), (ins f64mem:$op, RFP80:$src), OneArgFP, []>;
455 def ST_FpP80m : FpI_<(outs), (ins f80mem:$op, RFP80:$src), OneArgFP,
456 [(store RFP80:$src, addr:$op)]>;
457 let mayStore = 1, hasSideEffects = 0 in {
458 def IST_Fp16m32 : FpIf32<(outs), (ins i16mem:$op, RFP32:$src), OneArgFP, []>;
459 def IST_Fp32m32 : FpIf32<(outs), (ins i32mem:$op, RFP32:$src), OneArgFP, []>;
460 def IST_Fp64m32 : FpIf32<(outs), (ins i64mem:$op, RFP32:$src), OneArgFP, []>;
461 def IST_Fp16m64 : FpIf64<(outs), (ins i16mem:$op, RFP64:$src), OneArgFP, []>;
462 def IST_Fp32m64 : FpIf64<(outs), (ins i32mem:$op, RFP64:$src), OneArgFP, []>;
463 def IST_Fp64m64 : FpIf64<(outs), (ins i64mem:$op, RFP64:$src), OneArgFP, []>;
464 def IST_Fp16m80 : FpI_<(outs), (ins i16mem:$op, RFP80:$src), OneArgFP, []>;
465 def IST_Fp32m80 : FpI_<(outs), (ins i32mem:$op, RFP80:$src), OneArgFP, []>;
466 def IST_Fp64m80 : FpI_<(outs), (ins i64mem:$op, RFP80:$src), OneArgFP, []>;
469 let mayLoad = 1, SchedRW = [WriteLoad] in {
470 def LD_F32m : FPI<0xD9, MRM0m, (outs), (ins f32mem:$src), "fld{s}\t$src",
472 def LD_F64m : FPI<0xDD, MRM0m, (outs), (ins f64mem:$src), "fld{l}\t$src",
474 def LD_F80m : FPI<0xDB, MRM5m, (outs), (ins f80mem:$src), "fld{t}\t$src",
476 def ILD_F16m : FPI<0xDF, MRM0m, (outs), (ins i16mem:$src), "fild{s}\t$src",
478 def ILD_F32m : FPI<0xDB, MRM0m, (outs), (ins i32mem:$src), "fild{l}\t$src",
480 def ILD_F64m : FPI<0xDF, MRM5m, (outs), (ins i64mem:$src), "fild{ll}\t$src",
483 let mayStore = 1, SchedRW = [WriteStore] in {
484 def ST_F32m : FPI<0xD9, MRM2m, (outs), (ins f32mem:$dst), "fst{s}\t$dst",
486 def ST_F64m : FPI<0xDD, MRM2m, (outs), (ins f64mem:$dst), "fst{l}\t$dst",
488 def ST_FP32m : FPI<0xD9, MRM3m, (outs), (ins f32mem:$dst), "fstp{s}\t$dst",
490 def ST_FP64m : FPI<0xDD, MRM3m, (outs), (ins f64mem:$dst), "fstp{l}\t$dst",
492 def ST_FP80m : FPI<0xDB, MRM7m, (outs), (ins f80mem:$dst), "fstp{t}\t$dst",
494 def IST_F16m : FPI<0xDF, MRM2m, (outs), (ins i16mem:$dst), "fist{s}\t$dst",
496 def IST_F32m : FPI<0xDB, MRM2m, (outs), (ins i32mem:$dst), "fist{l}\t$dst",
498 def IST_FP16m : FPI<0xDF, MRM3m, (outs), (ins i16mem:$dst), "fistp{s}\t$dst",
500 def IST_FP32m : FPI<0xDB, MRM3m, (outs), (ins i32mem:$dst), "fistp{l}\t$dst",
502 def IST_FP64m : FPI<0xDF, MRM7m, (outs), (ins i64mem:$dst), "fistp{ll}\t$dst",
506 // FISTTP requires SSE3 even though it's a FPStack op.
507 let Predicates = [HasSSE3] in {
508 def ISTT_Fp16m32 : FpI_<(outs), (ins i16mem:$op, RFP32:$src), OneArgFP,
509 [(X86fp_to_i16mem RFP32:$src, addr:$op)]>;
510 def ISTT_Fp32m32 : FpI_<(outs), (ins i32mem:$op, RFP32:$src), OneArgFP,
511 [(X86fp_to_i32mem RFP32:$src, addr:$op)]>;
512 def ISTT_Fp64m32 : FpI_<(outs), (ins i64mem:$op, RFP32:$src), OneArgFP,
513 [(X86fp_to_i64mem RFP32:$src, addr:$op)]>;
514 def ISTT_Fp16m64 : FpI_<(outs), (ins i16mem:$op, RFP64:$src), OneArgFP,
515 [(X86fp_to_i16mem RFP64:$src, addr:$op)]>;
516 def ISTT_Fp32m64 : FpI_<(outs), (ins i32mem:$op, RFP64:$src), OneArgFP,
517 [(X86fp_to_i32mem RFP64:$src, addr:$op)]>;
518 def ISTT_Fp64m64 : FpI_<(outs), (ins i64mem:$op, RFP64:$src), OneArgFP,
519 [(X86fp_to_i64mem RFP64:$src, addr:$op)]>;
520 def ISTT_Fp16m80 : FpI_<(outs), (ins i16mem:$op, RFP80:$src), OneArgFP,
521 [(X86fp_to_i16mem RFP80:$src, addr:$op)]>;
522 def ISTT_Fp32m80 : FpI_<(outs), (ins i32mem:$op, RFP80:$src), OneArgFP,
523 [(X86fp_to_i32mem RFP80:$src, addr:$op)]>;
524 def ISTT_Fp64m80 : FpI_<(outs), (ins i64mem:$op, RFP80:$src), OneArgFP,
525 [(X86fp_to_i64mem RFP80:$src, addr:$op)]>;
526 } // Predicates = [HasSSE3]
528 let mayStore = 1, SchedRW = [WriteStore] in {
529 def ISTT_FP16m : FPI<0xDF, MRM1m, (outs), (ins i16mem:$dst), "fisttp{s}\t$dst",
531 def ISTT_FP32m : FPI<0xDB, MRM1m, (outs), (ins i32mem:$dst), "fisttp{l}\t$dst",
533 def ISTT_FP64m : FPI<0xDD, MRM1m, (outs), (ins i64mem:$dst),
534 "fisttp{ll}\t$dst", IIC_FST>;
537 // FP Stack manipulation instructions.
538 let SchedRW = [WriteMove] in {
539 def LD_Frr : FPI<0xD9, MRM0r, (outs), (ins RST:$op), "fld\t$op", IIC_FLD>;
540 def ST_Frr : FPI<0xDD, MRM2r, (outs), (ins RST:$op), "fst\t$op", IIC_FST>;
541 def ST_FPrr : FPI<0xDD, MRM3r, (outs), (ins RST:$op), "fstp\t$op", IIC_FST>;
542 def XCH_F : FPI<0xD9, MRM1r, (outs), (ins RST:$op), "fxch\t$op", IIC_FXCH>;
545 // Floating point constant loads.
546 let isReMaterializable = 1 in {
547 def LD_Fp032 : FpIf32<(outs RFP32:$dst), (ins), ZeroArgFP,
548 [(set RFP32:$dst, fpimm0)]>;
549 def LD_Fp132 : FpIf32<(outs RFP32:$dst), (ins), ZeroArgFP,
550 [(set RFP32:$dst, fpimm1)]>;
551 def LD_Fp064 : FpIf64<(outs RFP64:$dst), (ins), ZeroArgFP,
552 [(set RFP64:$dst, fpimm0)]>;
553 def LD_Fp164 : FpIf64<(outs RFP64:$dst), (ins), ZeroArgFP,
554 [(set RFP64:$dst, fpimm1)]>;
555 def LD_Fp080 : FpI_<(outs RFP80:$dst), (ins), ZeroArgFP,
556 [(set RFP80:$dst, fpimm0)]>;
557 def LD_Fp180 : FpI_<(outs RFP80:$dst), (ins), ZeroArgFP,
558 [(set RFP80:$dst, fpimm1)]>;
561 let SchedRW = [WriteZero] in {
562 def LD_F0 : FPI<0xD9, MRM_EE, (outs), (ins), "fldz", IIC_FLDZ>;
563 def LD_F1 : FPI<0xD9, MRM_E8, (outs), (ins), "fld1", IIC_FIST>;
566 // Floating point compares.
567 let SchedRW = [WriteFAdd] in {
568 def UCOM_Fpr32 : FpIf32<(outs), (ins RFP32:$lhs, RFP32:$rhs), CompareFP,
569 [(set FPSW, (trunc (X86cmp RFP32:$lhs, RFP32:$rhs)))]>;
570 def UCOM_Fpr64 : FpIf64<(outs), (ins RFP64:$lhs, RFP64:$rhs), CompareFP,
571 [(set FPSW, (trunc (X86cmp RFP64:$lhs, RFP64:$rhs)))]>;
572 def UCOM_Fpr80 : FpI_ <(outs), (ins RFP80:$lhs, RFP80:$rhs), CompareFP,
573 [(set FPSW, (trunc (X86cmp RFP80:$lhs, RFP80:$rhs)))]>;
577 let SchedRW = [WriteFAdd] in {
578 // CC = ST(0) cmp ST(i)
579 let Defs = [EFLAGS, FPSW] in {
580 def UCOM_FpIr32: FpIf32<(outs), (ins RFP32:$lhs, RFP32:$rhs), CompareFP,
581 [(set EFLAGS, (X86cmp RFP32:$lhs, RFP32:$rhs))]>;
582 def UCOM_FpIr64: FpIf64<(outs), (ins RFP64:$lhs, RFP64:$rhs), CompareFP,
583 [(set EFLAGS, (X86cmp RFP64:$lhs, RFP64:$rhs))]>;
584 def UCOM_FpIr80: FpI_<(outs), (ins RFP80:$lhs, RFP80:$rhs), CompareFP,
585 [(set EFLAGS, (X86cmp RFP80:$lhs, RFP80:$rhs))]>;
588 let Defs = [FPSW], Uses = [ST0] in {
589 def UCOM_Fr : FPI<0xDD, MRM4r, // FPSW = cmp ST(0) with ST(i)
590 (outs), (ins RST:$reg), "fucom\t$reg", IIC_FUCOM>;
591 def UCOM_FPr : FPI<0xDD, MRM5r, // FPSW = cmp ST(0) with ST(i), pop
592 (outs), (ins RST:$reg), "fucomp\t$reg", IIC_FUCOM>;
593 def UCOM_FPPr : FPI<0xDA, MRM_E9, // cmp ST(0) with ST(1), pop, pop
594 (outs), (ins), "fucompp", IIC_FUCOM>;
597 let Defs = [EFLAGS, FPSW], Uses = [ST0] in {
598 def UCOM_FIr : FPI<0xDB, MRM5r, // CC = cmp ST(0) with ST(i)
599 (outs), (ins RST:$reg), "fucomi\t$reg", IIC_FUCOMI>;
600 def UCOM_FIPr : FPI<0xDF, MRM5r, // CC = cmp ST(0) with ST(i), pop
601 (outs), (ins RST:$reg), "fucompi\t$reg", IIC_FUCOMI>;
604 let Defs = [EFLAGS, FPSW] in {
605 def COM_FIr : FPI<0xDB, MRM6r, (outs), (ins RST:$reg),
606 "fcomi\t$reg", IIC_FCOMI>;
607 def COM_FIPr : FPI<0xDF, MRM6r, (outs), (ins RST:$reg),
608 "fcompi\t$reg", IIC_FCOMI>;
612 // Floating point flag ops.
613 let SchedRW = [WriteALU] in {
614 let Defs = [AX], Uses = [FPSW] in
615 def FNSTSW16r : I<0xDF, MRM_E0, // AX = fp flags
616 (outs), (ins), "fnstsw\t{%ax|ax}",
617 [(set AX, (X86fp_stsw FPSW))], IIC_FNSTSW>;
619 def FNSTCW16m : I<0xD9, MRM7m, // [mem16] = X87 control world
620 (outs), (ins i16mem:$dst), "fnstcw\t$dst",
621 [(X86fp_cwd_get16 addr:$dst)], IIC_FNSTCW>;
624 def FLDCW16m : I<0xD9, MRM5m, // X87 control world = [mem16]
625 (outs), (ins i16mem:$dst), "fldcw\t$dst", [], IIC_FLDCW>,
628 // FPU control instructions
629 let SchedRW = [WriteMicrocoded] in {
631 def FNINIT : I<0xDB, MRM_E3, (outs), (ins), "fninit", [], IIC_FNINIT>;
632 def FFREE : FPI<0xDD, MRM0r, (outs), (ins RST:$reg),
633 "ffree\t$reg", IIC_FFREE>;
637 def FNCLEX : I<0xDB, MRM_E2, (outs), (ins), "fnclex", [], IIC_FNCLEX>;
640 // Operandless floating-point instructions for the disassembler.
641 let SchedRW = [WriteMicrocoded] in {
642 def WAIT : I<0x9B, RawFrm, (outs), (ins), "wait", [], IIC_WAIT>;
644 def FNOP : I<0xD9, MRM_D0, (outs), (ins), "fnop", [], IIC_FNOP>;
645 def FXAM : I<0xD9, MRM_E5, (outs), (ins), "fxam", [], IIC_FXAM>;
646 def FLDL2T : I<0xD9, MRM_E9, (outs), (ins), "fldl2t", [], IIC_FLDL>;
647 def FLDL2E : I<0xD9, MRM_EA, (outs), (ins), "fldl2e", [], IIC_FLDL>;
648 def FLDPI : I<0xD9, MRM_EB, (outs), (ins), "fldpi", [], IIC_FLDL>;
649 def FLDLG2 : I<0xD9, MRM_EC, (outs), (ins), "fldlg2", [], IIC_FLDL>;
650 def FLDLN2 : I<0xD9, MRM_ED, (outs), (ins), "fldln2", [], IIC_FLDL>;
651 def F2XM1 : I<0xD9, MRM_F0, (outs), (ins), "f2xm1", [], IIC_F2XM1>;
652 def FYL2X : I<0xD9, MRM_F1, (outs), (ins), "fyl2x", [], IIC_FYL2X>;
653 def FPTAN : I<0xD9, MRM_F2, (outs), (ins), "fptan", [], IIC_FPTAN>;
654 def FPATAN : I<0xD9, MRM_F3, (outs), (ins), "fpatan", [], IIC_FPATAN>;
655 def FXTRACT : I<0xD9, MRM_F4, (outs), (ins), "fxtract", [], IIC_FXTRACT>;
656 def FPREM1 : I<0xD9, MRM_F5, (outs), (ins), "fprem1", [], IIC_FPREM1>;
657 def FDECSTP : I<0xD9, MRM_F6, (outs), (ins), "fdecstp", [], IIC_FPSTP>;
658 def FINCSTP : I<0xD9, MRM_F7, (outs), (ins), "fincstp", [], IIC_FPSTP>;
659 def FPREM : I<0xD9, MRM_F8, (outs), (ins), "fprem", [], IIC_FPREM>;
660 def FYL2XP1 : I<0xD9, MRM_F9, (outs), (ins), "fyl2xp1", [], IIC_FYL2XP1>;
661 def FSINCOS : I<0xD9, MRM_FB, (outs), (ins), "fsincos", [], IIC_FSINCOS>;
662 def FRNDINT : I<0xD9, MRM_FC, (outs), (ins), "frndint", [], IIC_FRNDINT>;
663 def FSCALE : I<0xD9, MRM_FD, (outs), (ins), "fscale", [], IIC_FSCALE>;
664 def FCOMPP : I<0xDE, MRM_D9, (outs), (ins), "fcompp", [], IIC_FCOMPP>;
666 let Predicates = [HasFXSR] in {
667 def FXSAVE : I<0xAE, MRM0m, (outs), (ins opaque512mem:$dst),
668 "fxsave\t$dst", [(int_x86_fxsave addr:$dst)], IIC_FXSAVE>, TB;
669 def FXSAVE64 : RI<0xAE, MRM0m, (outs), (ins opaque512mem:$dst),
670 "fxsave64\t$dst", [(int_x86_fxsave64 addr:$dst)],
671 IIC_FXSAVE>, TB, Requires<[In64BitMode]>;
672 def FXRSTOR : I<0xAE, MRM1m, (outs), (ins opaque512mem:$src),
673 "fxrstor\t$src", [(int_x86_fxrstor addr:$src)], IIC_FXRSTOR>, TB;
674 def FXRSTOR64 : RI<0xAE, MRM1m, (outs), (ins opaque512mem:$src),
675 "fxrstor64\t$src", [(int_x86_fxrstor64 addr:$src)],
676 IIC_FXRSTOR>, TB, Requires<[In64BitMode]>;
677 } // Predicates = [FeatureFXSR]
680 //===----------------------------------------------------------------------===//
681 // Non-Instruction Patterns
682 //===----------------------------------------------------------------------===//
684 // Required for RET of f32 / f64 / f80 values.
685 def : Pat<(X86fld addr:$src, f32), (LD_Fp32m addr:$src)>;
686 def : Pat<(X86fld addr:$src, f64), (LD_Fp64m addr:$src)>;
687 def : Pat<(X86fld addr:$src, f80), (LD_Fp80m addr:$src)>;
689 // Required for CALL which return f32 / f64 / f80 values.
690 def : Pat<(X86fst RFP32:$src, addr:$op, f32), (ST_Fp32m addr:$op, RFP32:$src)>;
691 def : Pat<(X86fst RFP64:$src, addr:$op, f32), (ST_Fp64m32 addr:$op,
693 def : Pat<(X86fst RFP64:$src, addr:$op, f64), (ST_Fp64m addr:$op, RFP64:$src)>;
694 def : Pat<(X86fst RFP80:$src, addr:$op, f32), (ST_Fp80m32 addr:$op,
696 def : Pat<(X86fst RFP80:$src, addr:$op, f64), (ST_Fp80m64 addr:$op,
698 def : Pat<(X86fst RFP80:$src, addr:$op, f80), (ST_FpP80m addr:$op,
701 // Floating point constant -0.0 and -1.0
702 def : Pat<(f32 fpimmneg0), (CHS_Fp32 (LD_Fp032))>, Requires<[FPStackf32]>;
703 def : Pat<(f32 fpimmneg1), (CHS_Fp32 (LD_Fp132))>, Requires<[FPStackf32]>;
704 def : Pat<(f64 fpimmneg0), (CHS_Fp64 (LD_Fp064))>, Requires<[FPStackf64]>;
705 def : Pat<(f64 fpimmneg1), (CHS_Fp64 (LD_Fp164))>, Requires<[FPStackf64]>;
706 def : Pat<(f80 fpimmneg0), (CHS_Fp80 (LD_Fp080))>;
707 def : Pat<(f80 fpimmneg1), (CHS_Fp80 (LD_Fp180))>;
709 // Used to conv. i64 to f64 since there isn't a SSE version.
710 def : Pat<(X86fildflag addr:$src, i64), (ILD_Fp64m64 addr:$src)>;
712 // FP extensions map onto simple pseudo-value conversions if they are to/from
714 def : Pat<(f64 (fextend RFP32:$src)), (COPY_TO_REGCLASS RFP32:$src, RFP64)>,
715 Requires<[FPStackf32]>;
716 def : Pat<(f80 (fextend RFP32:$src)), (COPY_TO_REGCLASS RFP32:$src, RFP80)>,
717 Requires<[FPStackf32]>;
718 def : Pat<(f80 (fextend RFP64:$src)), (COPY_TO_REGCLASS RFP64:$src, RFP80)>,
719 Requires<[FPStackf64]>;
721 // FP truncations map onto simple pseudo-value conversions if they are to/from
722 // the FP stack. We have validated that only value-preserving truncations make
724 def : Pat<(f32 (fround RFP64:$src)), (COPY_TO_REGCLASS RFP64:$src, RFP32)>,
725 Requires<[FPStackf32]>;
726 def : Pat<(f32 (fround RFP80:$src)), (COPY_TO_REGCLASS RFP80:$src, RFP32)>,
727 Requires<[FPStackf32]>;
728 def : Pat<(f64 (fround RFP80:$src)), (COPY_TO_REGCLASS RFP80:$src, RFP64)>,
729 Requires<[FPStackf64]>;