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 : FPImmLeaf<fAny, [{
61 return Imm.isExactlyValue(+0.0);
64 def fpimmneg0 : FPImmLeaf<fAny, [{
65 return Imm.isExactlyValue(-0.0);
68 def fpimm1 : FPImmLeaf<fAny, [{
69 return Imm.isExactlyValue(+1.0);
72 def fpimmneg1 : FPImmLeaf<fAny, [{
73 return Imm.isExactlyValue(-1.0);
76 // Some 'special' instructions - expanded after instruction selection.
77 let usesCustomInserter = 1, hasNoSchedulingInfo = 1 in {
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 let mayLoad = 1, hasSideEffects = 1 in {
143 // ST(0) = ST(0) + [mem]
144 def _Fp32m : FpIf32<(outs RFP32:$dst),
145 (ins RFP32:$src1, f32mem:$src2), OneArgFPRW,
148 (OpNode RFP32:$src1, (loadf32 addr:$src2))),
150 (OpNode (loadf32 addr:$src2), RFP32:$src1)))]>;
151 def _Fp64m : FpIf64<(outs RFP64:$dst),
152 (ins RFP64:$src1, f64mem:$src2), OneArgFPRW,
155 (OpNode RFP64:$src1, (loadf64 addr:$src2))),
157 (OpNode (loadf64 addr:$src2), RFP64:$src1)))]>;
158 def _Fp64m32: FpIf64<(outs RFP64:$dst),
159 (ins RFP64:$src1, f32mem:$src2), OneArgFPRW,
162 (OpNode RFP64:$src1, (f64 (extloadf32 addr:$src2)))),
164 (OpNode (f64 (extloadf32 addr:$src2)), RFP64:$src1)))]>;
165 def _Fp80m32: FpI_<(outs RFP80:$dst),
166 (ins RFP80:$src1, f32mem:$src2), OneArgFPRW,
169 (OpNode RFP80:$src1, (f80 (extloadf32 addr:$src2)))),
171 (OpNode (f80 (extloadf32 addr:$src2)), RFP80:$src1)))]>;
172 def _Fp80m64: FpI_<(outs RFP80:$dst),
173 (ins RFP80:$src1, f64mem:$src2), OneArgFPRW,
176 (OpNode RFP80:$src1, (f80 (extloadf64 addr:$src2)))),
178 (OpNode (f80 (extloadf64 addr:$src2)), RFP80:$src1)))]>;
179 def _F32m : FPI<0xD8, fp, (outs), (ins f32mem:$src),
180 !strconcat("f", asmstring, "{s}\t$src")>;
181 def _F64m : FPI<0xDC, fp, (outs), (ins f64mem:$src),
182 !strconcat("f", asmstring, "{l}\t$src")>;
183 // ST(0) = ST(0) + [memint]
184 def _FpI16m32 : FpIf32<(outs RFP32:$dst), (ins RFP32:$src1, i16mem:$src2),
188 (OpNode RFP32:$src1, (X86fild addr:$src2, i16))),
190 (OpNode (X86fild addr:$src2, i16), RFP32:$src1)))]>;
191 def _FpI32m32 : FpIf32<(outs RFP32:$dst), (ins RFP32:$src1, i32mem:$src2),
195 (OpNode RFP32:$src1, (X86fild addr:$src2, i32))),
197 (OpNode (X86fild addr:$src2, i32), RFP32:$src1)))]>;
198 def _FpI16m64 : FpIf64<(outs RFP64:$dst), (ins RFP64:$src1, i16mem:$src2),
202 (OpNode RFP64:$src1, (X86fild addr:$src2, i16))),
204 (OpNode (X86fild addr:$src2, i16), RFP64:$src1)))]>;
205 def _FpI32m64 : FpIf64<(outs RFP64:$dst), (ins RFP64:$src1, i32mem:$src2),
209 (OpNode RFP64:$src1, (X86fild addr:$src2, i32))),
211 (OpNode (X86fild addr:$src2, i32), RFP64:$src1)))]>;
212 def _FpI16m80 : FpI_<(outs RFP80:$dst), (ins RFP80:$src1, i16mem:$src2),
216 (OpNode RFP80:$src1, (X86fild addr:$src2, i16))),
218 (OpNode (X86fild addr:$src2, i16), RFP80:$src1)))]>;
219 def _FpI32m80 : FpI_<(outs RFP80:$dst), (ins RFP80:$src1, i32mem:$src2),
223 (OpNode RFP80:$src1, (X86fild addr:$src2, i32))),
225 (OpNode (X86fild addr:$src2, i32), RFP80:$src1)))]>;
226 def _FI16m : FPI<0xDE, fp, (outs), (ins i16mem:$src),
227 !strconcat("fi", asmstring, "{s}\t$src")>;
228 def _FI32m : FPI<0xDA, fp, (outs), (ins i32mem:$src),
229 !strconcat("fi", asmstring, "{l}\t$src")>;
230 } // mayLoad = 1, hasSideEffects = 1
233 let Defs = [FPSW] in {
234 // FPBinary_rr just defines pseudo-instructions, no need to set a scheduling
236 let hasNoSchedulingInfo = 1 in {
237 defm ADD : FPBinary_rr<fadd>;
238 defm SUB : FPBinary_rr<fsub>;
239 defm MUL : FPBinary_rr<fmul>;
240 defm DIV : FPBinary_rr<fdiv>;
243 // Sets the scheduling resources for the actual NAME#_F<size>m defintions.
244 let SchedRW = [WriteFAddLd] in {
245 defm ADD : FPBinary<fadd, MRM0m, "add">;
246 defm SUB : FPBinary<fsub, MRM4m, "sub">;
247 defm SUBR: FPBinary<fsub ,MRM5m, "subr", 0>;
250 let SchedRW = [WriteFMulLd] in {
251 defm MUL : FPBinary<fmul, MRM1m, "mul">;
254 let SchedRW = [WriteFDivLd] in {
255 defm DIV : FPBinary<fdiv, MRM6m, "div">;
256 defm DIVR: FPBinary<fdiv, MRM7m, "divr", 0>;
260 class FPST0rInst<Format fp, string asm>
261 : FPI<0xD8, fp, (outs), (ins RST:$op), asm>;
262 class FPrST0Inst<Format fp, string asm>
263 : FPI<0xDC, fp, (outs), (ins RST:$op), asm>;
264 class FPrST0PInst<Format fp, string asm>
265 : FPI<0xDE, fp, (outs), (ins RST:$op), asm>;
267 // NOTE: GAS and apparently all other AT&T style assemblers have a broken notion
268 // of some of the 'reverse' forms of the fsub and fdiv instructions. As such,
269 // we have to put some 'r's in and take them out of weird places.
270 let SchedRW = [WriteFAdd] in {
271 def ADD_FST0r : FPST0rInst <MRM0r, "fadd\t$op">;
272 def ADD_FrST0 : FPrST0Inst <MRM0r, "fadd\t{%st(0), $op|$op, st(0)}">;
273 def ADD_FPrST0 : FPrST0PInst<MRM0r, "faddp\t$op">;
274 def SUBR_FST0r : FPST0rInst <MRM5r, "fsubr\t$op">;
275 def SUB_FrST0 : FPrST0Inst <MRM5r, "fsub{r}\t{%st(0), $op|$op, st(0)}">;
276 def SUB_FPrST0 : FPrST0PInst<MRM5r, "fsub{r}p\t$op">;
277 def SUB_FST0r : FPST0rInst <MRM4r, "fsub\t$op">;
278 def SUBR_FrST0 : FPrST0Inst <MRM4r, "fsub{|r}\t{%st(0), $op|$op, st(0)}">;
279 def SUBR_FPrST0 : FPrST0PInst<MRM4r, "fsub{|r}p\t$op">;
281 let SchedRW = [WriteFCom] in {
282 def COM_FST0r : FPST0rInst <MRM2r, "fcom\t$op">;
283 def COMP_FST0r : FPST0rInst <MRM3r, "fcomp\t$op">;
285 let SchedRW = [WriteFMul] in {
286 def MUL_FST0r : FPST0rInst <MRM1r, "fmul\t$op">;
287 def MUL_FrST0 : FPrST0Inst <MRM1r, "fmul\t{%st(0), $op|$op, st(0)}">;
288 def MUL_FPrST0 : FPrST0PInst<MRM1r, "fmulp\t$op">;
290 let SchedRW = [WriteFDiv] in {
291 def DIVR_FST0r : FPST0rInst <MRM7r, "fdivr\t$op">;
292 def DIV_FrST0 : FPrST0Inst <MRM7r, "fdiv{r}\t{%st(0), $op|$op, st(0)}">;
293 def DIV_FPrST0 : FPrST0PInst<MRM7r, "fdiv{r}p\t$op">;
294 def DIV_FST0r : FPST0rInst <MRM6r, "fdiv\t$op">;
295 def DIVR_FrST0 : FPrST0Inst <MRM6r, "fdiv{|r}\t{%st(0), $op|$op, st(0)}">;
296 def DIVR_FPrST0 : FPrST0PInst<MRM6r, "fdiv{|r}p\t$op">;
300 multiclass FPUnary<SDNode OpNode, Format fp, string asmstring> {
301 def _Fp32 : FpIf32<(outs RFP32:$dst), (ins RFP32:$src), OneArgFPRW,
302 [(set RFP32:$dst, (OpNode RFP32:$src))]>;
303 def _Fp64 : FpIf64<(outs RFP64:$dst), (ins RFP64:$src), OneArgFPRW,
304 [(set RFP64:$dst, (OpNode RFP64:$src))]>;
305 def _Fp80 : FpI_<(outs RFP80:$dst), (ins RFP80:$src), OneArgFPRW,
306 [(set RFP80:$dst, (OpNode RFP80:$src))]>;
307 def _F : FPI<0xD9, fp, (outs), (ins), asmstring>;
310 let Defs = [FPSW] in {
312 let SchedRW = [WriteFSign] in {
313 defm CHS : FPUnary<fneg, MRM_E0, "fchs">;
314 defm ABS : FPUnary<fabs, MRM_E1, "fabs">;
317 let SchedRW = [WriteFSqrt80] in
318 defm SQRT: FPUnary<fsqrt,MRM_FA, "fsqrt">;
320 let SchedRW = [WriteMicrocoded] in {
321 defm SIN : FPUnary<fsin, MRM_FE, "fsin">;
322 defm COS : FPUnary<fcos, MRM_FF, "fcos">;
325 let SchedRW = [WriteFCom] in {
326 let hasSideEffects = 0 in {
327 def TST_Fp32 : FpIf32<(outs), (ins RFP32:$src), OneArgFP, []>;
328 def TST_Fp64 : FpIf64<(outs), (ins RFP64:$src), OneArgFP, []>;
329 def TST_Fp80 : FpI_<(outs), (ins RFP80:$src), OneArgFP, []>;
332 def TST_F : FPI<0xD9, MRM_E4, (outs), (ins), "ftst">;
336 // Versions of FP instructions that take a single memory operand. Added for the
337 // disassembler; remove as they are included with patterns elsewhere.
338 let SchedRW = [WriteFComLd] in {
339 def FCOM32m : FPI<0xD8, MRM2m, (outs), (ins f32mem:$src), "fcom{s}\t$src">;
340 def FCOMP32m : FPI<0xD8, MRM3m, (outs), (ins f32mem:$src), "fcomp{s}\t$src">;
342 def FCOM64m : FPI<0xDC, MRM2m, (outs), (ins f64mem:$src), "fcom{l}\t$src">;
343 def FCOMP64m : FPI<0xDC, MRM3m, (outs), (ins f64mem:$src), "fcomp{l}\t$src">;
345 def FICOM16m : FPI<0xDE, MRM2m, (outs), (ins i16mem:$src), "ficom{s}\t$src">;
346 def FICOMP16m: FPI<0xDE, MRM3m, (outs), (ins i16mem:$src), "ficomp{s}\t$src">;
348 def FICOM32m : FPI<0xDA, MRM2m, (outs), (ins i32mem:$src), "ficom{l}\t$src">;
349 def FICOMP32m: FPI<0xDA, MRM3m, (outs), (ins i32mem:$src), "ficomp{l}\t$src">;
352 let SchedRW = [WriteMicrocoded] in {
353 def FLDENVm : FPI<0xD9, MRM4m, (outs), (ins f32mem:$src), "fldenv\t$src">;
354 def FSTENVm : FPI<0xD9, MRM6m, (outs), (ins f32mem:$dst), "fnstenv\t$dst">;
356 def FRSTORm : FPI<0xDD, MRM4m, (outs), (ins f32mem:$dst), "frstor\t$dst">;
357 def FSAVEm : FPI<0xDD, MRM6m, (outs), (ins f32mem:$dst), "fnsave\t$dst">;
358 def FNSTSWm : FPI<0xDD, MRM7m, (outs), (ins i16mem:$dst), "fnstsw\t$dst">;
360 def FBLDm : FPI<0xDF, MRM4m, (outs), (ins f80mem:$src), "fbld\t$src">;
361 def FBSTPm : FPI<0xDF, MRM6m, (outs), (ins f80mem:$dst), "fbstp\t$dst">;
364 // Floating point cmovs.
365 class FpIf32CMov<dag outs, dag ins, FPFormat fp, list<dag> pattern> :
366 FpI_<outs, ins, fp, pattern>, Requires<[FPStackf32, HasCMov]>;
367 class FpIf64CMov<dag outs, dag ins, FPFormat fp, list<dag> pattern> :
368 FpI_<outs, ins, fp, pattern>, Requires<[FPStackf64, HasCMov]>;
370 multiclass FPCMov<PatLeaf cc> {
371 def _Fp32 : FpIf32CMov<(outs RFP32:$dst), (ins RFP32:$src1, RFP32:$src2),
373 [(set RFP32:$dst, (X86cmov RFP32:$src1, RFP32:$src2,
375 def _Fp64 : FpIf64CMov<(outs RFP64:$dst), (ins RFP64:$src1, RFP64:$src2),
377 [(set RFP64:$dst, (X86cmov RFP64:$src1, RFP64:$src2,
379 def _Fp80 : FpI_<(outs RFP80:$dst), (ins RFP80:$src1, RFP80:$src2),
381 [(set RFP80:$dst, (X86cmov RFP80:$src1, RFP80:$src2,
386 let Defs = [FPSW] in {
387 let SchedRW = [WriteFCMOV] in {
388 let Uses = [EFLAGS], Constraints = "$src1 = $dst" in {
389 defm CMOVB : FPCMov<X86_COND_B>;
390 defm CMOVBE : FPCMov<X86_COND_BE>;
391 defm CMOVE : FPCMov<X86_COND_E>;
392 defm CMOVP : FPCMov<X86_COND_P>;
393 defm CMOVNB : FPCMov<X86_COND_AE>;
394 defm CMOVNBE: FPCMov<X86_COND_A>;
395 defm CMOVNE : FPCMov<X86_COND_NE>;
396 defm CMOVNP : FPCMov<X86_COND_NP>;
397 } // Uses = [EFLAGS], Constraints = "$src1 = $dst"
399 let Predicates = [HasCMov] in {
400 // These are not factored because there's no clean way to pass DA/DB.
401 def CMOVB_F : FPI<0xDA, MRM0r, (outs), (ins RST:$op),
402 "fcmovb\t{$op, %st(0)|st(0), $op}">;
403 def CMOVBE_F : FPI<0xDA, MRM2r, (outs), (ins RST:$op),
404 "fcmovbe\t{$op, %st(0)|st(0), $op}">;
405 def CMOVE_F : FPI<0xDA, MRM1r, (outs), (ins RST:$op),
406 "fcmove\t{$op, %st(0)|st(0), $op}">;
407 def CMOVP_F : FPI<0xDA, MRM3r, (outs), (ins RST:$op),
408 "fcmovu\t{$op, %st(0)|st(0), $op}">;
409 def CMOVNB_F : FPI<0xDB, MRM0r, (outs), (ins RST:$op),
410 "fcmovnb\t{$op, %st(0)|st(0), $op}">;
411 def CMOVNBE_F: FPI<0xDB, MRM2r, (outs), (ins RST:$op),
412 "fcmovnbe\t{$op, %st(0)|st(0), $op}">;
413 def CMOVNE_F : FPI<0xDB, MRM1r, (outs), (ins RST:$op),
414 "fcmovne\t{$op, %st(0)|st(0), $op}">;
415 def CMOVNP_F : FPI<0xDB, MRM3r, (outs), (ins RST:$op),
416 "fcmovnu\t{$op, %st(0)|st(0), $op}">;
417 } // Predicates = [HasCMov]
420 // Floating point loads & stores.
421 let SchedRW = [WriteLoad] in {
422 let canFoldAsLoad = 1 in {
423 def LD_Fp32m : FpIf32<(outs RFP32:$dst), (ins f32mem:$src), ZeroArgFP,
424 [(set RFP32:$dst, (loadf32 addr:$src))]>;
425 let isReMaterializable = 1 in
426 def LD_Fp64m : FpIf64<(outs RFP64:$dst), (ins f64mem:$src), ZeroArgFP,
427 [(set RFP64:$dst, (loadf64 addr:$src))]>;
428 def LD_Fp80m : FpI_<(outs RFP80:$dst), (ins f80mem:$src), ZeroArgFP,
429 [(set RFP80:$dst, (loadf80 addr:$src))]>;
431 def LD_Fp32m64 : FpIf64<(outs RFP64:$dst), (ins f32mem:$src), ZeroArgFP,
432 [(set RFP64:$dst, (f64 (extloadf32 addr:$src)))]>;
433 def LD_Fp64m80 : FpI_<(outs RFP80:$dst), (ins f64mem:$src), ZeroArgFP,
434 [(set RFP80:$dst, (f80 (extloadf64 addr:$src)))]>;
435 def LD_Fp32m80 : FpI_<(outs RFP80:$dst), (ins f32mem:$src), ZeroArgFP,
436 [(set RFP80:$dst, (f80 (extloadf32 addr:$src)))]>;
437 def ILD_Fp16m32: FpIf32<(outs RFP32:$dst), (ins i16mem:$src), ZeroArgFP,
438 [(set RFP32:$dst, (X86fild addr:$src, i16))]>;
439 def ILD_Fp32m32: FpIf32<(outs RFP32:$dst), (ins i32mem:$src), ZeroArgFP,
440 [(set RFP32:$dst, (X86fild addr:$src, i32))]>;
441 def ILD_Fp64m32: FpIf32<(outs RFP32:$dst), (ins i64mem:$src), ZeroArgFP,
442 [(set RFP32:$dst, (X86fild addr:$src, i64))]>;
443 def ILD_Fp16m64: FpIf64<(outs RFP64:$dst), (ins i16mem:$src), ZeroArgFP,
444 [(set RFP64:$dst, (X86fild addr:$src, i16))]>;
445 def ILD_Fp32m64: FpIf64<(outs RFP64:$dst), (ins i32mem:$src), ZeroArgFP,
446 [(set RFP64:$dst, (X86fild addr:$src, i32))]>;
447 def ILD_Fp64m64: FpIf64<(outs RFP64:$dst), (ins i64mem:$src), ZeroArgFP,
448 [(set RFP64:$dst, (X86fild addr:$src, i64))]>;
449 def ILD_Fp16m80: FpI_<(outs RFP80:$dst), (ins i16mem:$src), ZeroArgFP,
450 [(set RFP80:$dst, (X86fild addr:$src, i16))]>;
451 def ILD_Fp32m80: FpI_<(outs RFP80:$dst), (ins i32mem:$src), ZeroArgFP,
452 [(set RFP80:$dst, (X86fild addr:$src, i32))]>;
453 def ILD_Fp64m80: FpI_<(outs RFP80:$dst), (ins i64mem:$src), ZeroArgFP,
454 [(set RFP80:$dst, (X86fild addr:$src, i64))]>;
457 let SchedRW = [WriteStore] in {
458 def ST_Fp32m : FpIf32<(outs), (ins f32mem:$op, RFP32:$src), OneArgFP,
459 [(store RFP32:$src, addr:$op)]>;
460 def ST_Fp64m32 : FpIf64<(outs), (ins f32mem:$op, RFP64:$src), OneArgFP,
461 [(truncstoref32 RFP64:$src, addr:$op)]>;
462 def ST_Fp64m : FpIf64<(outs), (ins f64mem:$op, RFP64:$src), OneArgFP,
463 [(store RFP64:$src, addr:$op)]>;
464 def ST_Fp80m32 : FpI_<(outs), (ins f32mem:$op, RFP80:$src), OneArgFP,
465 [(truncstoref32 RFP80:$src, addr:$op)]>;
466 def ST_Fp80m64 : FpI_<(outs), (ins f64mem:$op, RFP80:$src), OneArgFP,
467 [(truncstoref64 RFP80:$src, addr:$op)]>;
468 // FST does not support 80-bit memory target; FSTP must be used.
470 let mayStore = 1, hasSideEffects = 0 in {
471 def ST_FpP32m : FpIf32<(outs), (ins f32mem:$op, RFP32:$src), OneArgFP, []>;
472 def ST_FpP64m32 : FpIf64<(outs), (ins f32mem:$op, RFP64:$src), OneArgFP, []>;
473 def ST_FpP64m : FpIf64<(outs), (ins f64mem:$op, RFP64:$src), OneArgFP, []>;
474 def ST_FpP80m32 : FpI_<(outs), (ins f32mem:$op, RFP80:$src), OneArgFP, []>;
475 def ST_FpP80m64 : FpI_<(outs), (ins f64mem:$op, RFP80:$src), OneArgFP, []>;
478 def ST_FpP80m : FpI_<(outs), (ins f80mem:$op, RFP80:$src), OneArgFP,
479 [(store RFP80:$src, addr:$op)]>;
481 let mayStore = 1, hasSideEffects = 0 in {
482 def IST_Fp16m32 : FpIf32<(outs), (ins i16mem:$op, RFP32:$src), OneArgFP, []>;
483 def IST_Fp32m32 : FpIf32<(outs), (ins i32mem:$op, RFP32:$src), OneArgFP, []>;
484 def IST_Fp64m32 : FpIf32<(outs), (ins i64mem:$op, RFP32:$src), OneArgFP, []>;
485 def IST_Fp16m64 : FpIf64<(outs), (ins i16mem:$op, RFP64:$src), OneArgFP, []>;
486 def IST_Fp32m64 : FpIf64<(outs), (ins i32mem:$op, RFP64:$src), OneArgFP, []>;
487 def IST_Fp64m64 : FpIf64<(outs), (ins i64mem:$op, RFP64:$src), OneArgFP, []>;
488 def IST_Fp16m80 : FpI_<(outs), (ins i16mem:$op, RFP80:$src), OneArgFP, []>;
489 def IST_Fp32m80 : FpI_<(outs), (ins i32mem:$op, RFP80:$src), OneArgFP, []>;
490 def IST_Fp64m80 : FpI_<(outs), (ins i64mem:$op, RFP80:$src), OneArgFP, []>;
494 let mayLoad = 1, SchedRW = [WriteLoad] in {
495 def LD_F32m : FPI<0xD9, MRM0m, (outs), (ins f32mem:$src), "fld{s}\t$src">;
496 def LD_F64m : FPI<0xDD, MRM0m, (outs), (ins f64mem:$src), "fld{l}\t$src">;
497 def LD_F80m : FPI<0xDB, MRM5m, (outs), (ins f80mem:$src), "fld{t}\t$src">;
498 def ILD_F16m : FPI<0xDF, MRM0m, (outs), (ins i16mem:$src), "fild{s}\t$src">;
499 def ILD_F32m : FPI<0xDB, MRM0m, (outs), (ins i32mem:$src), "fild{l}\t$src">;
500 def ILD_F64m : FPI<0xDF, MRM5m, (outs), (ins i64mem:$src), "fild{ll}\t$src">;
502 let mayStore = 1, SchedRW = [WriteStore] in {
503 def ST_F32m : FPI<0xD9, MRM2m, (outs), (ins f32mem:$dst), "fst{s}\t$dst">;
504 def ST_F64m : FPI<0xDD, MRM2m, (outs), (ins f64mem:$dst), "fst{l}\t$dst">;
505 def ST_FP32m : FPI<0xD9, MRM3m, (outs), (ins f32mem:$dst), "fstp{s}\t$dst">;
506 def ST_FP64m : FPI<0xDD, MRM3m, (outs), (ins f64mem:$dst), "fstp{l}\t$dst">;
507 def ST_FP80m : FPI<0xDB, MRM7m, (outs), (ins f80mem:$dst), "fstp{t}\t$dst">;
508 def IST_F16m : FPI<0xDF, MRM2m, (outs), (ins i16mem:$dst), "fist{s}\t$dst">;
509 def IST_F32m : FPI<0xDB, MRM2m, (outs), (ins i32mem:$dst), "fist{l}\t$dst">;
510 def IST_FP16m : FPI<0xDF, MRM3m, (outs), (ins i16mem:$dst), "fistp{s}\t$dst">;
511 def IST_FP32m : FPI<0xDB, MRM3m, (outs), (ins i32mem:$dst), "fistp{l}\t$dst">;
512 def IST_FP64m : FPI<0xDF, MRM7m, (outs), (ins i64mem:$dst), "fistp{ll}\t$dst">;
515 // FISTTP requires SSE3 even though it's a FPStack op.
516 let Predicates = [HasSSE3], SchedRW = [WriteStore] in {
517 def ISTT_Fp16m32 : FpI_<(outs), (ins i16mem:$op, RFP32:$src), OneArgFP,
518 [(X86fp_to_i16mem RFP32:$src, addr:$op)]>;
519 def ISTT_Fp32m32 : FpI_<(outs), (ins i32mem:$op, RFP32:$src), OneArgFP,
520 [(X86fp_to_i32mem RFP32:$src, addr:$op)]>;
521 def ISTT_Fp64m32 : FpI_<(outs), (ins i64mem:$op, RFP32:$src), OneArgFP,
522 [(X86fp_to_i64mem RFP32:$src, addr:$op)]>;
523 def ISTT_Fp16m64 : FpI_<(outs), (ins i16mem:$op, RFP64:$src), OneArgFP,
524 [(X86fp_to_i16mem RFP64:$src, addr:$op)]>;
525 def ISTT_Fp32m64 : FpI_<(outs), (ins i32mem:$op, RFP64:$src), OneArgFP,
526 [(X86fp_to_i32mem RFP64:$src, addr:$op)]>;
527 def ISTT_Fp64m64 : FpI_<(outs), (ins i64mem:$op, RFP64:$src), OneArgFP,
528 [(X86fp_to_i64mem RFP64:$src, addr:$op)]>;
529 def ISTT_Fp16m80 : FpI_<(outs), (ins i16mem:$op, RFP80:$src), OneArgFP,
530 [(X86fp_to_i16mem RFP80:$src, addr:$op)]>;
531 def ISTT_Fp32m80 : FpI_<(outs), (ins i32mem:$op, RFP80:$src), OneArgFP,
532 [(X86fp_to_i32mem RFP80:$src, addr:$op)]>;
533 def ISTT_Fp64m80 : FpI_<(outs), (ins i64mem:$op, RFP80:$src), OneArgFP,
534 [(X86fp_to_i64mem RFP80:$src, addr:$op)]>;
535 } // Predicates = [HasSSE3]
537 let mayStore = 1, SchedRW = [WriteStore] in {
538 def ISTT_FP16m : FPI<0xDF, MRM1m, (outs), (ins i16mem:$dst), "fisttp{s}\t$dst">;
539 def ISTT_FP32m : FPI<0xDB, MRM1m, (outs), (ins i32mem:$dst), "fisttp{l}\t$dst">;
540 def ISTT_FP64m : FPI<0xDD, MRM1m, (outs), (ins i64mem:$dst), "fisttp{ll}\t$dst">;
543 // FP Stack manipulation instructions.
544 let SchedRW = [WriteMove] in {
545 def LD_Frr : FPI<0xD9, MRM0r, (outs), (ins RST:$op), "fld\t$op">;
546 def ST_Frr : FPI<0xDD, MRM2r, (outs), (ins RST:$op), "fst\t$op">;
547 def ST_FPrr : FPI<0xDD, MRM3r, (outs), (ins RST:$op), "fstp\t$op">;
548 def XCH_F : FPI<0xD9, MRM1r, (outs), (ins RST:$op), "fxch\t$op">;
551 // Floating point constant loads.
552 let isReMaterializable = 1, SchedRW = [WriteZero] in {
553 def LD_Fp032 : FpIf32<(outs RFP32:$dst), (ins), ZeroArgFP,
554 [(set RFP32:$dst, fpimm0)]>;
555 def LD_Fp132 : FpIf32<(outs RFP32:$dst), (ins), ZeroArgFP,
556 [(set RFP32:$dst, fpimm1)]>;
557 def LD_Fp064 : FpIf64<(outs RFP64:$dst), (ins), ZeroArgFP,
558 [(set RFP64:$dst, fpimm0)]>;
559 def LD_Fp164 : FpIf64<(outs RFP64:$dst), (ins), ZeroArgFP,
560 [(set RFP64:$dst, fpimm1)]>;
561 def LD_Fp080 : FpI_<(outs RFP80:$dst), (ins), ZeroArgFP,
562 [(set RFP80:$dst, fpimm0)]>;
563 def LD_Fp180 : FpI_<(outs RFP80:$dst), (ins), ZeroArgFP,
564 [(set RFP80:$dst, fpimm1)]>;
567 let SchedRW = [WriteFLD0] in
568 def LD_F0 : FPI<0xD9, MRM_EE, (outs), (ins), "fldz">;
570 let SchedRW = [WriteFLD1] in
571 def LD_F1 : FPI<0xD9, MRM_E8, (outs), (ins), "fld1">;
573 let SchedRW = [WriteFLDC], Defs = [FPSW] in {
574 def FLDL2T : I<0xD9, MRM_E9, (outs), (ins), "fldl2t", []>;
575 def FLDL2E : I<0xD9, MRM_EA, (outs), (ins), "fldl2e", []>;
576 def FLDPI : I<0xD9, MRM_EB, (outs), (ins), "fldpi", []>;
577 def FLDLG2 : I<0xD9, MRM_EC, (outs), (ins), "fldlg2", []>;
578 def FLDLN2 : I<0xD9, MRM_ED, (outs), (ins), "fldln2", []>;
581 // Floating point compares.
582 let SchedRW = [WriteFCom] in {
583 def UCOM_Fpr32 : FpIf32<(outs), (ins RFP32:$lhs, RFP32:$rhs), CompareFP,
584 [(set FPSW, (trunc (X86cmp RFP32:$lhs, RFP32:$rhs)))]>;
585 def UCOM_Fpr64 : FpIf64<(outs), (ins RFP64:$lhs, RFP64:$rhs), CompareFP,
586 [(set FPSW, (trunc (X86cmp RFP64:$lhs, RFP64:$rhs)))]>;
587 def UCOM_Fpr80 : FpI_ <(outs), (ins RFP80:$lhs, RFP80:$rhs), CompareFP,
588 [(set FPSW, (trunc (X86cmp RFP80:$lhs, RFP80:$rhs)))]>;
592 let SchedRW = [WriteFCom] in {
593 // CC = ST(0) cmp ST(i)
594 let Defs = [EFLAGS, FPSW] in {
595 let Predicates = [FPStackf32, HasCMov] in
596 def UCOM_FpIr32: FpIf32<(outs), (ins RFP32:$lhs, RFP32:$rhs), CompareFP,
597 [(set EFLAGS, (X86cmp RFP32:$lhs, RFP32:$rhs))]>;
598 let Predicates = [FPStackf64, HasCMov] in
599 def UCOM_FpIr64: FpIf64<(outs), (ins RFP64:$lhs, RFP64:$rhs), CompareFP,
600 [(set EFLAGS, (X86cmp RFP64:$lhs, RFP64:$rhs))]>;
601 let Predicates = [HasCMov] in
602 def UCOM_FpIr80: FpI_<(outs), (ins RFP80:$lhs, RFP80:$rhs), CompareFP,
603 [(set EFLAGS, (X86cmp RFP80:$lhs, RFP80:$rhs))]>;
606 let Defs = [FPSW], Uses = [ST0] in {
607 def UCOM_Fr : FPI<0xDD, MRM4r, // FPSW = cmp ST(0) with ST(i)
608 (outs), (ins RST:$reg), "fucom\t$reg">;
609 def UCOM_FPr : FPI<0xDD, MRM5r, // FPSW = cmp ST(0) with ST(i), pop
610 (outs), (ins RST:$reg), "fucomp\t$reg">;
611 def UCOM_FPPr : FPI<0xDA, MRM_E9, // cmp ST(0) with ST(1), pop, pop
612 (outs), (ins), "fucompp">;
615 let Defs = [EFLAGS, FPSW], Uses = [ST0] in {
616 def UCOM_FIr : FPI<0xDB, MRM5r, // CC = cmp ST(0) with ST(i)
617 (outs), (ins RST:$reg), "fucomi\t$reg">;
618 def UCOM_FIPr : FPI<0xDF, MRM5r, // CC = cmp ST(0) with ST(i), pop
619 (outs), (ins RST:$reg), "fucompi\t$reg">;
622 let Defs = [EFLAGS, FPSW] in {
623 def COM_FIr : FPI<0xDB, MRM6r, (outs), (ins RST:$reg), "fcomi\t$reg">;
624 def COM_FIPr : FPI<0xDF, MRM6r, (outs), (ins RST:$reg), "fcompi\t$reg">;
628 // Floating point flag ops.
629 let SchedRW = [WriteALU] in {
630 let Defs = [AX], Uses = [FPSW] in
631 def FNSTSW16r : I<0xDF, MRM_E0, // AX = fp flags
632 (outs), (ins), "fnstsw\t{%ax|ax}",
633 [(set AX, (X86fp_stsw FPSW))]>;
635 def FNSTCW16m : I<0xD9, MRM7m, // [mem16] = X87 control world
636 (outs), (ins i16mem:$dst), "fnstcw\t$dst",
637 [(X86fp_cwd_get16 addr:$dst)]>;
639 let Defs = [FPSW], mayLoad = 1 in
640 def FLDCW16m : I<0xD9, MRM5m, // X87 control world = [mem16]
641 (outs), (ins i16mem:$dst), "fldcw\t$dst", []>,
644 // FPU control instructions
645 let SchedRW = [WriteMicrocoded] in {
646 let Defs = [FPSW] in {
647 def FNINIT : I<0xDB, MRM_E3, (outs), (ins), "fninit", []>;
648 def FFREE : FPI<0xDD, MRM0r, (outs), (ins RST:$reg), "ffree\t$reg">;
649 def FFREEP : FPI<0xDF, MRM0r, (outs), (ins RST:$reg), "ffreep\t$reg">;
652 def FNCLEX : I<0xDB, MRM_E2, (outs), (ins), "fnclex", []>;
656 // Operand-less floating-point instructions for the disassembler.
657 def FNOP : I<0xD9, MRM_D0, (outs), (ins), "fnop", []>, Sched<[WriteNop]>;
659 let SchedRW = [WriteMicrocoded] in {
660 let Defs = [FPSW] in {
661 def WAIT : I<0x9B, RawFrm, (outs), (ins), "wait", []>;
662 def FXAM : I<0xD9, MRM_E5, (outs), (ins), "fxam", []>;
663 def F2XM1 : I<0xD9, MRM_F0, (outs), (ins), "f2xm1", []>;
664 def FYL2X : I<0xD9, MRM_F1, (outs), (ins), "fyl2x", []>;
665 def FPTAN : I<0xD9, MRM_F2, (outs), (ins), "fptan", []>;
666 def FPATAN : I<0xD9, MRM_F3, (outs), (ins), "fpatan", []>;
667 def FXTRACT : I<0xD9, MRM_F4, (outs), (ins), "fxtract", []>;
668 def FPREM1 : I<0xD9, MRM_F5, (outs), (ins), "fprem1", []>;
669 def FDECSTP : I<0xD9, MRM_F6, (outs), (ins), "fdecstp", []>;
670 def FINCSTP : I<0xD9, MRM_F7, (outs), (ins), "fincstp", []>;
671 def FPREM : I<0xD9, MRM_F8, (outs), (ins), "fprem", []>;
672 def FYL2XP1 : I<0xD9, MRM_F9, (outs), (ins), "fyl2xp1", []>;
673 def FSINCOS : I<0xD9, MRM_FB, (outs), (ins), "fsincos", []>;
674 def FRNDINT : I<0xD9, MRM_FC, (outs), (ins), "frndint", []>;
675 def FSCALE : I<0xD9, MRM_FD, (outs), (ins), "fscale", []>;
676 def FCOMPP : I<0xDE, MRM_D9, (outs), (ins), "fcompp", []>;
679 def FXSAVE : I<0xAE, MRM0m, (outs), (ins opaquemem:$dst),
680 "fxsave\t$dst", [(int_x86_fxsave addr:$dst)]>, TB,
682 def FXSAVE64 : RI<0xAE, MRM0m, (outs), (ins opaquemem:$dst),
683 "fxsave64\t$dst", [(int_x86_fxsave64 addr:$dst)]>,
684 TB, Requires<[HasFXSR, In64BitMode]>;
685 def FXRSTOR : I<0xAE, MRM1m, (outs), (ins opaquemem:$src),
686 "fxrstor\t$src", [(int_x86_fxrstor addr:$src)]>,
687 TB, Requires<[HasFXSR]>;
688 def FXRSTOR64 : RI<0xAE, MRM1m, (outs), (ins opaquemem:$src),
689 "fxrstor64\t$src", [(int_x86_fxrstor64 addr:$src)]>,
690 TB, Requires<[HasFXSR, In64BitMode]>;
693 //===----------------------------------------------------------------------===//
694 // Non-Instruction Patterns
695 //===----------------------------------------------------------------------===//
697 // Required for RET of f32 / f64 / f80 values.
698 def : Pat<(X86fld addr:$src, f32), (LD_Fp32m addr:$src)>;
699 def : Pat<(X86fld addr:$src, f64), (LD_Fp64m addr:$src)>;
700 def : Pat<(X86fld addr:$src, f80), (LD_Fp80m addr:$src)>;
702 // Required for CALL which return f32 / f64 / f80 values.
703 def : Pat<(X86fst RFP32:$src, addr:$op, f32), (ST_Fp32m addr:$op, RFP32:$src)>;
704 def : Pat<(X86fst RFP64:$src, addr:$op, f32), (ST_Fp64m32 addr:$op,
706 def : Pat<(X86fst RFP64:$src, addr:$op, f64), (ST_Fp64m addr:$op, RFP64:$src)>;
707 def : Pat<(X86fst RFP80:$src, addr:$op, f32), (ST_Fp80m32 addr:$op,
709 def : Pat<(X86fst RFP80:$src, addr:$op, f64), (ST_Fp80m64 addr:$op,
711 def : Pat<(X86fst RFP80:$src, addr:$op, f80), (ST_FpP80m addr:$op,
714 // Floating point constant -0.0 and -1.0
715 def : Pat<(f32 fpimmneg0), (CHS_Fp32 (LD_Fp032))>, Requires<[FPStackf32]>;
716 def : Pat<(f32 fpimmneg1), (CHS_Fp32 (LD_Fp132))>, Requires<[FPStackf32]>;
717 def : Pat<(f64 fpimmneg0), (CHS_Fp64 (LD_Fp064))>, Requires<[FPStackf64]>;
718 def : Pat<(f64 fpimmneg1), (CHS_Fp64 (LD_Fp164))>, Requires<[FPStackf64]>;
719 def : Pat<(f80 fpimmneg0), (CHS_Fp80 (LD_Fp080))>;
720 def : Pat<(f80 fpimmneg1), (CHS_Fp80 (LD_Fp180))>;
722 // Used to conv. i64 to f64 since there isn't a SSE version.
723 def : Pat<(X86fildflag addr:$src, i64), (ILD_Fp64m64 addr:$src)>;
725 // FP extensions map onto simple pseudo-value conversions if they are to/from
727 def : Pat<(f64 (fpextend RFP32:$src)), (COPY_TO_REGCLASS RFP32:$src, RFP64)>,
728 Requires<[FPStackf32]>;
729 def : Pat<(f80 (fpextend RFP32:$src)), (COPY_TO_REGCLASS RFP32:$src, RFP80)>,
730 Requires<[FPStackf32]>;
731 def : Pat<(f80 (fpextend RFP64:$src)), (COPY_TO_REGCLASS RFP64:$src, RFP80)>,
732 Requires<[FPStackf64]>;
734 // FP truncations map onto simple pseudo-value conversions if they are to/from
735 // the FP stack. We have validated that only value-preserving truncations make
737 def : Pat<(f32 (fpround RFP64:$src)), (COPY_TO_REGCLASS RFP64:$src, RFP32)>,
738 Requires<[FPStackf32]>;
739 def : Pat<(f32 (fpround RFP80:$src)), (COPY_TO_REGCLASS RFP80:$src, RFP32)>,
740 Requires<[FPStackf32]>;
741 def : Pat<(f64 (fpround RFP80:$src)), (COPY_TO_REGCLASS RFP80:$src, RFP64)>,
742 Requires<[FPStackf64]>;