1 //===- ARMInstrVFP.td - VFP support for ARM ----------------*- 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 ARM VFP instruction set.
12 //===----------------------------------------------------------------------===//
14 def SDT_FTOI : SDTypeProfile<1, 1, [SDTCisVT<0, f32>, SDTCisFP<1>]>;
15 def SDT_ITOF : SDTypeProfile<1, 1, [SDTCisFP<0>, SDTCisVT<1, f32>]>;
16 def SDT_CMPFP0 : SDTypeProfile<0, 1, [SDTCisFP<0>]>;
17 def SDT_VMOVDRR : SDTypeProfile<1, 2, [SDTCisVT<0, f64>, SDTCisVT<1, i32>,
20 def arm_ftoui : SDNode<"ARMISD::FTOUI", SDT_FTOI>;
21 def arm_ftosi : SDNode<"ARMISD::FTOSI", SDT_FTOI>;
22 def arm_sitof : SDNode<"ARMISD::SITOF", SDT_ITOF>;
23 def arm_uitof : SDNode<"ARMISD::UITOF", SDT_ITOF>;
24 def arm_fmstat : SDNode<"ARMISD::FMSTAT", SDTNone, [SDNPInFlag, SDNPOutFlag]>;
25 def arm_cmpfp : SDNode<"ARMISD::CMPFP", SDT_ARMCmp, [SDNPOutFlag]>;
26 def arm_cmpfp0 : SDNode<"ARMISD::CMPFPw0", SDT_CMPFP0, [SDNPOutFlag]>;
27 def arm_fmdrr : SDNode<"ARMISD::VMOVDRR", SDT_VMOVDRR>;
30 //===----------------------------------------------------------------------===//
31 // Operand Definitions.
34 def vfp_f32imm : Operand<f32>,
35 PatLeaf<(f32 fpimm), [{
36 return ARM::getVFPf32Imm(N->getValueAPF()) != -1;
38 let PrintMethod = "printVFPf32ImmOperand";
41 def vfp_f64imm : Operand<f64>,
42 PatLeaf<(f64 fpimm), [{
43 return ARM::getVFPf64Imm(N->getValueAPF()) != -1;
45 let PrintMethod = "printVFPf64ImmOperand";
49 //===----------------------------------------------------------------------===//
50 // Load / store Instructions.
53 let canFoldAsLoad = 1, isReMaterializable = 1 in {
54 def VLDRD : ADI5<0b1101, 0b01, (outs DPR:$Dd), (ins addrmode5:$addr),
55 IIC_fpLoad64, "vldr", ".64\t$Dd, $addr",
56 [(set DPR:$Dd, (f64 (load addrmode5:$addr)))]> {
57 // Instruction operands.
61 // Encode instruction operands.
62 let Inst{23} = addr{16}; // U (add = (U == '1'))
64 let Inst{19-16} = addr{20-17}; // Rn
65 let Inst{15-12} = Dd{3-0};
66 let Inst{7-0} = addr{7-0}; // imm8
69 def VLDRS : ASI5<0b1101, 0b01, (outs SPR:$dst), (ins addrmode5:$addr),
70 IIC_fpLoad32, "vldr", ".32\t$dst, $addr",
71 [(set SPR:$dst, (load addrmode5:$addr))]>;
74 def VSTRD : ADI5<0b1101, 0b00, (outs), (ins DPR:$src, addrmode5:$addr),
75 IIC_fpStore64, "vstr", ".64\t$src, $addr",
76 [(store (f64 DPR:$src), addrmode5:$addr)]>;
78 def VSTRS : ASI5<0b1101, 0b00, (outs), (ins SPR:$src, addrmode5:$addr),
79 IIC_fpStore32, "vstr", ".32\t$src, $addr",
80 [(store SPR:$src, addrmode5:$addr)]>;
82 //===----------------------------------------------------------------------===//
83 // Load / store multiple Instructions.
86 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1,
87 isCodeGenOnly = 1 in {
88 def VLDMD : AXDI4<(outs), (ins addrmode4:$addr, pred:$p, reglist:$dsts,
89 variable_ops), IndexModeNone, IIC_fpLoad_m,
90 "vldm${addr:submode}${p}\t$addr, $dsts", "", []> {
94 def VLDMS : AXSI4<(outs), (ins addrmode4:$addr, pred:$p, reglist:$dsts,
95 variable_ops), IndexModeNone, IIC_fpLoad_m,
96 "vldm${addr:submode}${p}\t$addr, $dsts", "", []> {
100 def VLDMD_UPD : AXDI4<(outs GPR:$wb), (ins addrmode4:$addr, pred:$p,
101 reglist:$dsts, variable_ops),
102 IndexModeUpd, IIC_fpLoad_mu,
103 "vldm${addr:submode}${p}\t$addr!, $dsts",
104 "$addr.addr = $wb", []> {
108 def VLDMS_UPD : AXSI4<(outs GPR:$wb), (ins addrmode4:$addr, pred:$p,
109 reglist:$dsts, variable_ops),
110 IndexModeUpd, IIC_fpLoad_mu,
111 "vldm${addr:submode}${p}\t$addr!, $dsts",
112 "$addr.addr = $wb", []> {
115 } // mayLoad, neverHasSideEffects, hasExtraDefRegAllocReq
117 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1,
118 isCodeGenOnly = 1 in {
119 def VSTMD : AXDI4<(outs), (ins addrmode4:$addr, pred:$p, reglist:$srcs,
120 variable_ops), IndexModeNone, IIC_fpStore_m,
121 "vstm${addr:submode}${p}\t$addr, $srcs", "", []> {
125 def VSTMS : AXSI4<(outs), (ins addrmode4:$addr, pred:$p, reglist:$srcs,
126 variable_ops), IndexModeNone, IIC_fpStore_m,
127 "vstm${addr:submode}${p}\t$addr, $srcs", "", []> {
131 def VSTMD_UPD : AXDI4<(outs GPR:$wb), (ins addrmode4:$addr, pred:$p,
132 reglist:$srcs, variable_ops),
133 IndexModeUpd, IIC_fpStore_mu,
134 "vstm${addr:submode}${p}\t$addr!, $srcs",
135 "$addr.addr = $wb", []> {
139 def VSTMS_UPD : AXSI4<(outs GPR:$wb), (ins addrmode4:$addr, pred:$p,
140 reglist:$srcs, variable_ops),
141 IndexModeUpd, IIC_fpStore_mu,
142 "vstm${addr:submode}${p}\t$addr!, $srcs",
143 "$addr.addr = $wb", []> {
146 } // mayStore, neverHasSideEffects, hasExtraSrcRegAllocReq
148 // FLDMX, FSTMX - mixing S/D registers for pre-armv6 cores
150 //===----------------------------------------------------------------------===//
151 // FP Binary Operations.
154 def VADDD : ADbI<0b11100, 0b11, 0, 0,
155 (outs DPR:$Dd), (ins DPR:$Dn, DPR:$Dm),
156 IIC_fpALU64, "vadd", ".f64\t$Dd, $Dn, $Dm",
157 [(set DPR:$Dd, (fadd DPR:$Dn, (f64 DPR:$Dm)))]>;
159 def VADDS : ASbIn<0b11100, 0b11, 0, 0,
160 (outs SPR:$Sd), (ins SPR:$Sn, SPR:$Sm),
161 IIC_fpALU32, "vadd", ".f32\t$Sd, $Sn, $Sm",
162 [(set SPR:$Sd, (fadd SPR:$Sn, SPR:$Sm))]>;
164 def VSUBD : ADbI<0b11100, 0b11, 1, 0,
165 (outs DPR:$Dd), (ins DPR:$Dn, DPR:$Dm),
166 IIC_fpALU64, "vsub", ".f64\t$Dd, $Dn, $Dm",
167 [(set DPR:$Dd, (fsub DPR:$Dn, (f64 DPR:$Dm)))]>;
169 def VSUBS : ASbIn<0b11100, 0b11, 1, 0,
170 (outs SPR:$Sd), (ins SPR:$Sn, SPR:$Sm),
171 IIC_fpALU32, "vsub", ".f32\t$Sd, $Sn, $Sm",
172 [(set SPR:$Sd, (fsub SPR:$Sn, SPR:$Sm))]>;
174 def VDIVD : ADbI<0b11101, 0b00, 0, 0,
175 (outs DPR:$Dd), (ins DPR:$Dn, DPR:$Dm),
176 IIC_fpDIV64, "vdiv", ".f64\t$Dd, $Dn, $Dm",
177 [(set DPR:$Dd, (fdiv DPR:$Dn, (f64 DPR:$Dm)))]>;
179 def VDIVS : ASbI<0b11101, 0b00, 0, 0,
180 (outs SPR:$Sd), (ins SPR:$Sn, SPR:$Sm),
181 IIC_fpDIV32, "vdiv", ".f32\t$Sd, $Sn, $Sm",
182 [(set SPR:$Sd, (fdiv SPR:$Sn, SPR:$Sm))]>;
184 def VMULD : ADbI<0b11100, 0b10, 0, 0,
185 (outs DPR:$Dd), (ins DPR:$Dn, DPR:$Dm),
186 IIC_fpMUL64, "vmul", ".f64\t$Dd, $Dn, $Dm",
187 [(set DPR:$Dd, (fmul DPR:$Dn, (f64 DPR:$Dm)))]>;
189 def VMULS : ASbIn<0b11100, 0b10, 0, 0,
190 (outs SPR:$Sd), (ins SPR:$Sn, SPR:$Sm),
191 IIC_fpMUL32, "vmul", ".f32\t$Sd, $Sn, $Sm",
192 [(set SPR:$Sd, (fmul SPR:$Sn, SPR:$Sm))]>;
194 def VNMULD : ADbI<0b11100, 0b10, 1, 0,
195 (outs DPR:$Dd), (ins DPR:$Dn, DPR:$Dm),
196 IIC_fpMUL64, "vnmul", ".f64\t$Dd, $Dn, $Dm",
197 [(set DPR:$Dd, (fneg (fmul DPR:$Dn, (f64 DPR:$Dm))))]>;
199 def VNMULS : ASbI<0b11100, 0b10, 1, 0,
200 (outs SPR:$Sd), (ins SPR:$Sn, SPR:$Sm),
201 IIC_fpMUL32, "vnmul", ".f32\t$Sd, $Sn, $Sm",
202 [(set SPR:$Sd, (fneg (fmul SPR:$Sn, SPR:$Sm)))]>;
204 // Match reassociated forms only if not sign dependent rounding.
205 def : Pat<(fmul (fneg DPR:$a), (f64 DPR:$b)),
206 (VNMULD DPR:$a, DPR:$b)>, Requires<[NoHonorSignDependentRounding]>;
207 def : Pat<(fmul (fneg SPR:$a), SPR:$b),
208 (VNMULS SPR:$a, SPR:$b)>, Requires<[NoHonorSignDependentRounding]>;
210 // These are encoded as unary instructions.
211 let Defs = [FPSCR] in {
212 def VCMPED : ADuI<0b11101, 0b11, 0b0100, 0b11, 0,
213 (outs), (ins DPR:$Dd, DPR:$Dm),
214 IIC_fpCMP64, "vcmpe", ".f64\t$Dd, $Dm",
215 [(arm_cmpfp DPR:$Dd, (f64 DPR:$Dm))]>;
217 def VCMPES : ASuI<0b11101, 0b11, 0b0100, 0b11, 0,
218 (outs), (ins SPR:$Sd, SPR:$Sm),
219 IIC_fpCMP32, "vcmpe", ".f32\t$Sd, $Sm",
220 [(arm_cmpfp SPR:$Sd, SPR:$Sm)]>;
222 // FIXME: Verify encoding after integrated assembler is working.
223 def VCMPD : ADuI<0b11101, 0b11, 0b0100, 0b01, 0,
224 (outs), (ins DPR:$Dd, DPR:$Dm),
225 IIC_fpCMP64, "vcmp", ".f64\t$Dd, $Dm",
226 [/* For disassembly only; pattern left blank */]>;
228 def VCMPS : ASuI<0b11101, 0b11, 0b0100, 0b01, 0,
229 (outs), (ins SPR:$Sd, SPR:$Sm),
230 IIC_fpCMP32, "vcmp", ".f32\t$Sd, $Sm",
231 [/* For disassembly only; pattern left blank */]>;
234 //===----------------------------------------------------------------------===//
235 // FP Unary Operations.
238 def VABSD : ADuI<0b11101, 0b11, 0b0000, 0b11, 0,
239 (outs DPR:$Dd), (ins DPR:$Dm),
240 IIC_fpUNA64, "vabs", ".f64\t$Dd, $Dm",
241 [(set DPR:$Dd, (fabs (f64 DPR:$Dm)))]>;
243 def VABSS : ASuIn<0b11101, 0b11, 0b0000, 0b11, 0,
244 (outs SPR:$Sd), (ins SPR:$Sm),
245 IIC_fpUNA32, "vabs", ".f32\t$Sd, $Sm",
246 [(set SPR:$Sd, (fabs SPR:$Sm))]>;
248 let Defs = [FPSCR] in {
249 def VCMPEZD : ADuI<0b11101, 0b11, 0b0101, 0b11, 0,
250 (outs), (ins DPR:$Dd),
251 IIC_fpCMP64, "vcmpe", ".f64\t$Dd, #0",
252 [(arm_cmpfp0 (f64 DPR:$Dd))]> {
253 let Inst{3-0} = 0b0000;
257 def VCMPEZS : ASuI<0b11101, 0b11, 0b0101, 0b11, 0,
258 (outs), (ins SPR:$Sd),
259 IIC_fpCMP32, "vcmpe", ".f32\t$Sd, #0",
260 [(arm_cmpfp0 SPR:$Sd)]> {
261 let Inst{3-0} = 0b0000;
265 // FIXME: Verify encoding after integrated assembler is working.
266 def VCMPZD : ADuI<0b11101, 0b11, 0b0101, 0b01, 0,
267 (outs), (ins DPR:$Dd),
268 IIC_fpCMP64, "vcmp", ".f64\t$Dd, #0",
269 [/* For disassembly only; pattern left blank */]> {
270 let Inst{3-0} = 0b0000;
274 def VCMPZS : ASuI<0b11101, 0b11, 0b0101, 0b01, 0,
275 (outs), (ins SPR:$Sd),
276 IIC_fpCMP32, "vcmp", ".f32\t$Sd, #0",
277 [/* For disassembly only; pattern left blank */]> {
278 let Inst{3-0} = 0b0000;
283 def VCVTDS : ASuI<0b11101, 0b11, 0b0111, 0b11, 0,
284 (outs DPR:$Dd), (ins SPR:$Sm),
285 IIC_fpCVTDS, "vcvt", ".f64.f32\t$Dd, $Sm",
286 [(set DPR:$Dd, (fextend SPR:$Sm))]> {
287 // Instruction operands.
291 // Encode instruction operands.
292 let Inst{3-0} = Sm{4-1};
294 let Inst{15-12} = Dd{3-0};
295 let Inst{22} = Dd{4};
298 // Special case encoding: bits 11-8 is 0b1011.
299 def VCVTSD : VFPAI<(outs SPR:$Sd), (ins DPR:$Dm), VFPUnaryFrm,
300 IIC_fpCVTSD, "vcvt", ".f32.f64\t$Sd, $Dm",
301 [(set SPR:$Sd, (fround DPR:$Dm))]> {
302 // Instruction operands.
306 // Encode instruction operands.
307 let Inst{3-0} = Dm{3-0};
309 let Inst{15-12} = Sd{4-1};
310 let Inst{22} = Sd{0};
312 let Inst{27-23} = 0b11101;
313 let Inst{21-16} = 0b110111;
314 let Inst{11-8} = 0b1011;
315 let Inst{7-6} = 0b11;
319 // Between half-precision and single-precision. For disassembly only.
321 // FIXME: Verify encoding after integrated assembler is working.
322 def VCVTBSH: ASuI<0b11101, 0b11, 0b0010, 0b01, 0, (outs SPR:$dst), (ins SPR:$a),
323 /* FIXME */ IIC_fpCVTSH, "vcvtb", ".f32.f16\t$dst, $a",
324 [/* For disassembly only; pattern left blank */]>;
326 def : ARMPat<(f32_to_f16 SPR:$a),
327 (i32 (COPY_TO_REGCLASS (VCVTBSH SPR:$a), GPR))>;
329 def VCVTBHS: ASuI<0b11101, 0b11, 0b0011, 0b01, 0, (outs SPR:$dst), (ins SPR:$a),
330 /* FIXME */ IIC_fpCVTHS, "vcvtb", ".f16.f32\t$dst, $a",
331 [/* For disassembly only; pattern left blank */]>;
333 def : ARMPat<(f16_to_f32 GPR:$a),
334 (VCVTBHS (COPY_TO_REGCLASS GPR:$a, SPR))>;
336 def VCVTTSH: ASuI<0b11101, 0b11, 0b0010, 0b11, 0, (outs SPR:$dst), (ins SPR:$a),
337 /* FIXME */ IIC_fpCVTSH, "vcvtt", ".f32.f16\t$dst, $a",
338 [/* For disassembly only; pattern left blank */]>;
340 def VCVTTHS: ASuI<0b11101, 0b11, 0b0011, 0b11, 0, (outs SPR:$dst), (ins SPR:$a),
341 /* FIXME */ IIC_fpCVTHS, "vcvtt", ".f16.f32\t$dst, $a",
342 [/* For disassembly only; pattern left blank */]>;
344 def VNEGD : ADuI<0b11101, 0b11, 0b0001, 0b01, 0,
345 (outs DPR:$Dd), (ins DPR:$Dm),
346 IIC_fpUNA64, "vneg", ".f64\t$Dd, $Dm",
347 [(set DPR:$Dd, (fneg (f64 DPR:$Dm)))]>;
349 def VNEGS : ASuIn<0b11101, 0b11, 0b0001, 0b01, 0,
350 (outs SPR:$Sd), (ins SPR:$Sm),
351 IIC_fpUNA32, "vneg", ".f32\t$Sd, $Sm",
352 [(set SPR:$Sd, (fneg SPR:$Sm))]>;
354 def VSQRTD : ADuI<0b11101, 0b11, 0b0001, 0b11, 0,
355 (outs DPR:$Dd), (ins DPR:$Dm),
356 IIC_fpSQRT64, "vsqrt", ".f64\t$Dd, $Dm",
357 [(set DPR:$Dd, (fsqrt (f64 DPR:$Dm)))]>;
359 def VSQRTS : ASuI<0b11101, 0b11, 0b0001, 0b11, 0,
360 (outs SPR:$Sd), (ins SPR:$Sm),
361 IIC_fpSQRT32, "vsqrt", ".f32\t$Sd, $Sm",
362 [(set SPR:$Sd, (fsqrt SPR:$Sm))]>;
364 let neverHasSideEffects = 1 in {
365 def VMOVD : ADuI<0b11101, 0b11, 0b0000, 0b01, 0,
366 (outs DPR:$Dd), (ins DPR:$Dm),
367 IIC_fpUNA64, "vmov", ".f64\t$Dd, $Dm", []>;
369 def VMOVS : ASuI<0b11101, 0b11, 0b0000, 0b01, 0,
370 (outs SPR:$Sd), (ins SPR:$Sm),
371 IIC_fpUNA32, "vmov", ".f32\t$Sd, $Sm", []>;
372 } // neverHasSideEffects
374 //===----------------------------------------------------------------------===//
375 // FP <-> GPR Copies. Int <-> FP Conversions.
378 def VMOVRS : AVConv2I<0b11100001, 0b1010,
379 (outs GPR:$Rt), (ins SPR:$Sn),
380 IIC_fpMOVSI, "vmov", "\t$Rt, $Sn",
381 [(set GPR:$Rt, (bitconvert SPR:$Sn))]> {
382 // Instruction operands.
386 // Encode instruction operands.
387 let Inst{19-16} = Sn{4-1};
389 let Inst{15-12} = Rt;
391 let Inst{6-5} = 0b00;
392 let Inst{3-0} = 0b0000;
395 def VMOVSR : AVConv4I<0b11100000, 0b1010,
396 (outs SPR:$Sn), (ins GPR:$Rt),
397 IIC_fpMOVIS, "vmov", "\t$Sn, $Rt",
398 [(set SPR:$Sn, (bitconvert GPR:$Rt))]> {
399 // Instruction operands.
403 // Encode instruction operands.
404 let Inst{19-16} = Sn{4-1};
406 let Inst{15-12} = Rt;
408 let Inst{6-5} = 0b00;
409 let Inst{3-0} = 0b0000;
412 let neverHasSideEffects = 1 in {
413 def VMOVRRD : AVConv3I<0b11000101, 0b1011,
414 (outs GPR:$Rt, GPR:$Rt2), (ins DPR:$Dm),
415 IIC_fpMOVDI, "vmov", "\t$Rt, $Rt2, $Dm",
416 [/* FIXME: Can't write pattern for multiple result instr*/]> {
417 // Instruction operands.
422 // Encode instruction operands.
423 let Inst{3-0} = Dm{3-0};
425 let Inst{15-12} = Rt;
426 let Inst{19-16} = Rt2;
428 let Inst{7-6} = 0b00;
431 def VMOVRRS : AVConv3I<0b11000101, 0b1010,
432 (outs GPR:$wb, GPR:$dst2), (ins SPR:$src1, SPR:$src2),
433 IIC_fpMOVDI, "vmov", "\t$wb, $dst2, $src1, $src2",
434 [/* For disassembly only; pattern left blank */]> {
435 let Inst{7-6} = 0b00;
437 } // neverHasSideEffects
442 def VMOVDRR : AVConv5I<0b11000100, 0b1011,
443 (outs DPR:$Dm), (ins GPR:$Rt, GPR:$Rt2),
444 IIC_fpMOVID, "vmov", "\t$Dm, $Rt, $Rt2",
445 [(set DPR:$Dm, (arm_fmdrr GPR:$Rt, GPR:$Rt2))]> {
446 // Instruction operands.
451 // Encode instruction operands.
452 let Inst{3-0} = Dm{3-0};
454 let Inst{15-12} = Rt;
455 let Inst{19-16} = Rt2;
457 let Inst{7-6} = 0b00;
460 let neverHasSideEffects = 1 in
461 def VMOVSRR : AVConv5I<0b11000100, 0b1010,
462 (outs SPR:$dst1, SPR:$dst2), (ins GPR:$src1, GPR:$src2),
463 IIC_fpMOVID, "vmov", "\t$dst1, $dst2, $src1, $src2",
464 [/* For disassembly only; pattern left blank */]> {
465 let Inst{7-6} = 0b00;
471 // FMRX: SPR system reg -> GPR
473 // FMXR: GPR -> VFP system reg
478 class AVConv1IDs_Encode<bits<5> opcod1, bits<2> opcod2, bits<4> opcod3,
479 bits<4> opcod4, dag oops, dag iops,
480 InstrItinClass itin, string opc, string asm,
482 : AVConv1I<opcod1, opcod2, opcod3, opcod4, oops, iops, itin, opc, asm,
484 // Instruction operands.
488 // Encode instruction operands.
489 let Inst{3-0} = Sm{4-1};
491 let Inst{15-12} = Dd{3-0};
492 let Inst{22} = Dd{4};
495 class AVConv1InSs_Encode<bits<5> opcod1, bits<2> opcod2, bits<4> opcod3,
496 bits<4> opcod4, dag oops, dag iops,InstrItinClass itin,
497 string opc, string asm, list<dag> pattern>
498 : AVConv1In<opcod1, opcod2, opcod3, opcod4, oops, iops, itin, opc, asm,
500 // Instruction operands.
504 // Encode instruction operands.
505 let Inst{3-0} = Sm{4-1};
507 let Inst{15-12} = Sd{4-1};
508 let Inst{22} = Sd{0};
511 def VSITOD : AVConv1IDs_Encode<0b11101, 0b11, 0b1000, 0b1011,
512 (outs DPR:$Dd), (ins SPR:$Sm),
513 IIC_fpCVTID, "vcvt", ".f64.s32\t$Dd, $Sm",
514 [(set DPR:$Dd, (f64 (arm_sitof SPR:$Sm)))]> {
515 let Inst{7} = 1; // s32
518 def VSITOS : AVConv1InSs_Encode<0b11101, 0b11, 0b1000, 0b1010,
519 (outs SPR:$Sd),(ins SPR:$Sm),
520 IIC_fpCVTIS, "vcvt", ".f32.s32\t$Sd, $Sm",
521 [(set SPR:$Sd, (arm_sitof SPR:$Sm))]> {
522 let Inst{7} = 1; // s32
525 def VUITOD : AVConv1IDs_Encode<0b11101, 0b11, 0b1000, 0b1011,
526 (outs DPR:$Dd), (ins SPR:$Sm),
527 IIC_fpCVTID, "vcvt", ".f64.u32\t$Dd, $Sm",
528 [(set DPR:$Dd, (f64 (arm_uitof SPR:$Sm)))]> {
529 let Inst{7} = 0; // u32
532 def VUITOS : AVConv1InSs_Encode<0b11101, 0b11, 0b1000, 0b1010,
533 (outs SPR:$Sd), (ins SPR:$Sm),
534 IIC_fpCVTIS, "vcvt", ".f32.u32\t$Sd, $Sm",
535 [(set SPR:$Sd, (arm_uitof SPR:$Sm))]> {
536 let Inst{7} = 0; // u32
541 class AVConv1IsD_Encode<bits<5> opcod1, bits<2> opcod2, bits<4> opcod3,
542 bits<4> opcod4, dag oops, dag iops,
543 InstrItinClass itin, string opc, string asm,
545 : AVConv1I<opcod1, opcod2, opcod3, opcod4, oops, iops, itin, opc, asm,
547 // Instruction operands.
551 // Encode instruction operands.
552 let Inst{3-0} = Dm{3-0};
554 let Inst{15-12} = Sd{4-1};
555 let Inst{22} = Sd{0};
558 class AVConv1InsS_Encode<bits<5> opcod1, bits<2> opcod2, bits<4> opcod3,
559 bits<4> opcod4, dag oops, dag iops,
560 InstrItinClass itin, string opc, string asm,
562 : AVConv1In<opcod1, opcod2, opcod3, opcod4, oops, iops, itin, opc, asm,
564 // Instruction operands.
568 // Encode instruction operands.
569 let Inst{3-0} = Sm{4-1};
571 let Inst{15-12} = Sd{4-1};
572 let Inst{22} = Sd{0};
575 // Always set Z bit in the instruction, i.e. "round towards zero" variants.
576 def VTOSIZD : AVConv1IsD_Encode<0b11101, 0b11, 0b1101, 0b1011,
577 (outs SPR:$Sd), (ins DPR:$Dm),
578 IIC_fpCVTDI, "vcvt", ".s32.f64\t$Sd, $Dm",
579 [(set SPR:$Sd, (arm_ftosi (f64 DPR:$Dm)))]> {
580 let Inst{7} = 1; // Z bit
583 def VTOSIZS : AVConv1InsS_Encode<0b11101, 0b11, 0b1101, 0b1010,
584 (outs SPR:$Sd), (ins SPR:$Sm),
585 IIC_fpCVTSI, "vcvt", ".s32.f32\t$Sd, $Sm",
586 [(set SPR:$Sd, (arm_ftosi SPR:$Sm))]> {
587 let Inst{7} = 1; // Z bit
590 def VTOUIZD : AVConv1IsD_Encode<0b11101, 0b11, 0b1100, 0b1011,
591 (outs SPR:$Sd), (ins DPR:$Dm),
592 IIC_fpCVTDI, "vcvt", ".u32.f64\t$Sd, $Dm",
593 [(set SPR:$Sd, (arm_ftoui (f64 DPR:$Dm)))]> {
594 let Inst{7} = 1; // Z bit
597 def VTOUIZS : AVConv1InsS_Encode<0b11101, 0b11, 0b1100, 0b1010,
598 (outs SPR:$Sd), (ins SPR:$Sm),
599 IIC_fpCVTSI, "vcvt", ".u32.f32\t$Sd, $Sm",
600 [(set SPR:$Sd, (arm_ftoui SPR:$Sm))]> {
601 let Inst{7} = 1; // Z bit
604 // And the Z bit '0' variants, i.e. use the rounding mode specified by FPSCR.
605 // For disassembly only.
606 let Uses = [FPSCR] in {
607 // FIXME: Verify encoding after integrated assembler is working.
608 def VTOSIRD : AVConv1IsD_Encode<0b11101, 0b11, 0b1101, 0b1011,
609 (outs SPR:$Sd), (ins DPR:$Dm),
610 IIC_fpCVTDI, "vcvtr", ".s32.f64\t$Sd, $Dm",
611 [(set SPR:$Sd, (int_arm_vcvtr (f64 DPR:$Dm)))]>{
612 let Inst{7} = 0; // Z bit
615 def VTOSIRS : AVConv1InsS_Encode<0b11101, 0b11, 0b1101, 0b1010,
616 (outs SPR:$Sd), (ins SPR:$Sm),
617 IIC_fpCVTSI, "vcvtr", ".s32.f32\t$Sd, $Sm",
618 [(set SPR:$Sd, (int_arm_vcvtr SPR:$Sm))]> {
619 let Inst{7} = 0; // Z bit
622 def VTOUIRD : AVConv1IsD_Encode<0b11101, 0b11, 0b1100, 0b1011,
623 (outs SPR:$Sd), (ins DPR:$Dm),
624 IIC_fpCVTDI, "vcvtr", ".u32.f64\t$Sd, $Dm",
625 [(set SPR:$Sd, (int_arm_vcvtru(f64 DPR:$Dm)))]>{
626 let Inst{7} = 0; // Z bit
629 def VTOUIRS : AVConv1InsS_Encode<0b11101, 0b11, 0b1100, 0b1010,
630 (outs SPR:$Sd), (ins SPR:$Sm),
631 IIC_fpCVTSI, "vcvtr", ".u32.f32\t$Sd, $Sm",
632 [(set SPR:$Sd, (int_arm_vcvtru SPR:$Sm))]> {
633 let Inst{7} = 0; // Z bit
637 // Convert between floating-point and fixed-point
638 // Data type for fixed-point naming convention:
639 // S16 (U=0, sx=0) -> SH
640 // U16 (U=1, sx=0) -> UH
641 // S32 (U=0, sx=1) -> SL
642 // U32 (U=1, sx=1) -> UL
644 // FIXME: Marking these as codegen only seems wrong. They are real
646 let Constraints = "$a = $dst", isCodeGenOnly = 1 in {
648 // FP to Fixed-Point:
650 def VTOSHS : AVConv1XI<0b11101, 0b11, 0b1110, 0b1010, 0,
651 (outs SPR:$dst), (ins SPR:$a, i32imm:$fbits),
652 IIC_fpCVTSI, "vcvt", ".s16.f32\t$dst, $a, $fbits",
653 [/* For disassembly only; pattern left blank */]>;
655 def VTOUHS : AVConv1XI<0b11101, 0b11, 0b1111, 0b1010, 0,
656 (outs SPR:$dst), (ins SPR:$a, i32imm:$fbits),
657 IIC_fpCVTSI, "vcvt", ".u16.f32\t$dst, $a, $fbits",
658 [/* For disassembly only; pattern left blank */]>;
660 def VTOSLS : AVConv1XI<0b11101, 0b11, 0b1110, 0b1010, 1,
661 (outs SPR:$dst), (ins SPR:$a, i32imm:$fbits),
662 IIC_fpCVTSI, "vcvt", ".s32.f32\t$dst, $a, $fbits",
663 [/* For disassembly only; pattern left blank */]>;
665 def VTOULS : AVConv1XI<0b11101, 0b11, 0b1111, 0b1010, 1,
666 (outs SPR:$dst), (ins SPR:$a, i32imm:$fbits),
667 IIC_fpCVTSI, "vcvt", ".u32.f32\t$dst, $a, $fbits",
668 [/* For disassembly only; pattern left blank */]>;
670 def VTOSHD : AVConv1XI<0b11101, 0b11, 0b1110, 0b1011, 0,
671 (outs DPR:$dst), (ins DPR:$a, i32imm:$fbits),
672 IIC_fpCVTDI, "vcvt", ".s16.f64\t$dst, $a, $fbits",
673 [/* For disassembly only; pattern left blank */]>;
675 def VTOUHD : AVConv1XI<0b11101, 0b11, 0b1111, 0b1011, 0,
676 (outs DPR:$dst), (ins DPR:$a, i32imm:$fbits),
677 IIC_fpCVTDI, "vcvt", ".u16.f64\t$dst, $a, $fbits",
678 [/* For disassembly only; pattern left blank */]>;
680 def VTOSLD : AVConv1XI<0b11101, 0b11, 0b1110, 0b1011, 1,
681 (outs DPR:$dst), (ins DPR:$a, i32imm:$fbits),
682 IIC_fpCVTDI, "vcvt", ".s32.f64\t$dst, $a, $fbits",
683 [/* For disassembly only; pattern left blank */]>;
685 def VTOULD : AVConv1XI<0b11101, 0b11, 0b1111, 0b1011, 1,
686 (outs DPR:$dst), (ins DPR:$a, i32imm:$fbits),
687 IIC_fpCVTDI, "vcvt", ".u32.f64\t$dst, $a, $fbits",
688 [/* For disassembly only; pattern left blank */]>;
690 // Fixed-Point to FP:
692 def VSHTOS : AVConv1XI<0b11101, 0b11, 0b1010, 0b1010, 0,
693 (outs SPR:$dst), (ins SPR:$a, i32imm:$fbits),
694 IIC_fpCVTIS, "vcvt", ".f32.s16\t$dst, $a, $fbits",
695 [/* For disassembly only; pattern left blank */]>;
697 def VUHTOS : AVConv1XI<0b11101, 0b11, 0b1011, 0b1010, 0,
698 (outs SPR:$dst), (ins SPR:$a, i32imm:$fbits),
699 IIC_fpCVTIS, "vcvt", ".f32.u16\t$dst, $a, $fbits",
700 [/* For disassembly only; pattern left blank */]>;
702 def VSLTOS : AVConv1XI<0b11101, 0b11, 0b1010, 0b1010, 1,
703 (outs SPR:$dst), (ins SPR:$a, i32imm:$fbits),
704 IIC_fpCVTIS, "vcvt", ".f32.s32\t$dst, $a, $fbits",
705 [/* For disassembly only; pattern left blank */]>;
707 def VULTOS : AVConv1XI<0b11101, 0b11, 0b1011, 0b1010, 1,
708 (outs SPR:$dst), (ins SPR:$a, i32imm:$fbits),
709 IIC_fpCVTIS, "vcvt", ".f32.u32\t$dst, $a, $fbits",
710 [/* For disassembly only; pattern left blank */]>;
712 def VSHTOD : AVConv1XI<0b11101, 0b11, 0b1010, 0b1011, 0,
713 (outs DPR:$dst), (ins DPR:$a, i32imm:$fbits),
714 IIC_fpCVTID, "vcvt", ".f64.s16\t$dst, $a, $fbits",
715 [/* For disassembly only; pattern left blank */]>;
717 def VUHTOD : AVConv1XI<0b11101, 0b11, 0b1011, 0b1011, 0,
718 (outs DPR:$dst), (ins DPR:$a, i32imm:$fbits),
719 IIC_fpCVTID, "vcvt", ".f64.u16\t$dst, $a, $fbits",
720 [/* For disassembly only; pattern left blank */]>;
722 def VSLTOD : AVConv1XI<0b11101, 0b11, 0b1010, 0b1011, 1,
723 (outs DPR:$dst), (ins DPR:$a, i32imm:$fbits),
724 IIC_fpCVTID, "vcvt", ".f64.s32\t$dst, $a, $fbits",
725 [/* For disassembly only; pattern left blank */]>;
727 def VULTOD : AVConv1XI<0b11101, 0b11, 0b1011, 0b1011, 1,
728 (outs DPR:$dst), (ins DPR:$a, i32imm:$fbits),
729 IIC_fpCVTID, "vcvt", ".f64.u32\t$dst, $a, $fbits",
730 [/* For disassembly only; pattern left blank */]>;
732 } // End of 'let Constraints = "$a = $dst", isCodeGenOnly = 1 in'
734 //===----------------------------------------------------------------------===//
735 // FP FMA Operations.
738 def VMLAD : ADbI_vmlX<0b11100, 0b00, 0, 0,
739 (outs DPR:$Dd), (ins DPR:$Ddin, DPR:$Dn, DPR:$Dm),
740 IIC_fpMAC64, "vmla", ".f64\t$Dd, $Dn, $Dm",
741 [(set DPR:$Dd, (fadd (fmul DPR:$Dn, DPR:$Dm),
743 RegConstraint<"$Ddin = $Dd">;
745 def VMLAS : ASbIn<0b11100, 0b00, 0, 0,
746 (outs SPR:$Sd), (ins SPR:$Sdin, SPR:$Sn, SPR:$Sm),
747 IIC_fpMAC32, "vmla", ".f32\t$Sd, $Sn, $Sm",
748 [(set SPR:$Sd, (fadd (fmul SPR:$Sn, SPR:$Sm),
750 RegConstraint<"$Sdin = $Sd">;
752 def : Pat<(fadd DPR:$dstin, (fmul DPR:$a, (f64 DPR:$b))),
753 (VMLAD DPR:$dstin, DPR:$a, DPR:$b)>, Requires<[DontUseNEONForFP]>;
754 def : Pat<(fadd SPR:$dstin, (fmul SPR:$a, SPR:$b)),
755 (VMLAS SPR:$dstin, SPR:$a, SPR:$b)>, Requires<[DontUseNEONForFP]>;
757 def VMLSD : ADbI_vmlX<0b11100, 0b00, 1, 0,
758 (outs DPR:$Dd), (ins DPR:$Ddin, DPR:$Dn, DPR:$Dm),
759 IIC_fpMAC64, "vmls", ".f64\t$Dd, $Dn, $Dm",
760 [(set DPR:$Dd, (fadd (fneg (fmul DPR:$Dn,DPR:$Dm)),
762 RegConstraint<"$Ddin = $Dd">;
764 def VMLSS : ASbIn<0b11100, 0b00, 1, 0,
765 (outs SPR:$Sd), (ins SPR:$Sdin, SPR:$Sn, SPR:$Sm),
766 IIC_fpMAC32, "vmls", ".f32\t$Sd, $Sn, $Sm",
767 [(set SPR:$Sd, (fadd (fneg (fmul SPR:$Sn, SPR:$Sm)),
769 RegConstraint<"$Sdin = $Sd">;
771 def : Pat<(fsub DPR:$dstin, (fmul DPR:$a, (f64 DPR:$b))),
772 (VMLSD DPR:$dstin, DPR:$a, DPR:$b)>, Requires<[DontUseNEONForFP]>;
773 def : Pat<(fsub SPR:$dstin, (fmul SPR:$a, SPR:$b)),
774 (VMLSS SPR:$dstin, SPR:$a, SPR:$b)>, Requires<[DontUseNEONForFP]>;
776 def VNMLAD : ADbI_vmlX<0b11100, 0b01, 1, 0,
777 (outs DPR:$Dd), (ins DPR:$Ddin, DPR:$Dn, DPR:$Dm),
778 IIC_fpMAC64, "vnmla", ".f64\t$Dd, $Dn, $Dm",
779 [(set DPR:$Dd,(fsub (fneg (fmul DPR:$Dn,DPR:$Dm)),
781 RegConstraint<"$Ddin = $Dd">;
783 def VNMLAS : ASbI<0b11100, 0b01, 1, 0,
784 (outs SPR:$Sd), (ins SPR:$Sdin, SPR:$Sn, SPR:$Sm),
785 IIC_fpMAC32, "vnmla", ".f32\t$Sd, $Sn, $Sm",
786 [(set SPR:$Sd, (fsub (fneg (fmul SPR:$Sn, SPR:$Sm)),
788 RegConstraint<"$Sdin = $Sd">;
790 def : Pat<(fsub (fneg (fmul DPR:$a, (f64 DPR:$b))), DPR:$dstin),
791 (VNMLAD DPR:$dstin, DPR:$a, DPR:$b)>, Requires<[DontUseNEONForFP]>;
792 def : Pat<(fsub (fneg (fmul SPR:$a, SPR:$b)), SPR:$dstin),
793 (VNMLAS SPR:$dstin, SPR:$a, SPR:$b)>, Requires<[DontUseNEONForFP]>;
795 def VNMLSD : ADbI_vmlX<0b11100, 0b01, 0, 0,
796 (outs DPR:$Dd), (ins DPR:$Ddin, DPR:$Dn, DPR:$Dm),
797 IIC_fpMAC64, "vnmls", ".f64\t$Dd, $Dn, $Dm",
798 [(set DPR:$Dd, (fsub (fmul DPR:$Dn, DPR:$Dm),
800 RegConstraint<"$Ddin = $Dd">;
802 def VNMLSS : ASbI<0b11100, 0b01, 0, 0,
803 (outs SPR:$Sd), (ins SPR:$Sdin, SPR:$Sn, SPR:$Sm),
804 IIC_fpMAC32, "vnmls", ".f32\t$Sd, $Sn, $Sm",
805 [(set SPR:$Sd, (fsub (fmul SPR:$Sn, SPR:$Sm), SPR:$Sdin))]>,
806 RegConstraint<"$Sdin = $Sd">;
808 def : Pat<(fsub (fmul DPR:$a, (f64 DPR:$b)), DPR:$dstin),
809 (VNMLSD DPR:$dstin, DPR:$a, DPR:$b)>, Requires<[DontUseNEONForFP]>;
810 def : Pat<(fsub (fmul SPR:$a, SPR:$b), SPR:$dstin),
811 (VNMLSS SPR:$dstin, SPR:$a, SPR:$b)>, Requires<[DontUseNEONForFP]>;
814 //===----------------------------------------------------------------------===//
815 // FP Conditional moves.
818 let neverHasSideEffects = 1 in {
819 def VMOVDcc : ADuI<0b11101, 0b11, 0b0000, 0b01, 0,
820 (outs DPR:$Dd), (ins DPR:$Dn, DPR:$Dm),
821 IIC_fpUNA64, "vmov", ".f64\t$Dd, $Dm",
822 [/*(set DPR:$Dd, (ARMcmov DPR:$Dn, DPR:$Dm, imm:$cc))*/]>,
823 RegConstraint<"$Dn = $Dd">;
825 def VMOVScc : ASuI<0b11101, 0b11, 0b0000, 0b01, 0,
826 (outs SPR:$Sd), (ins SPR:$Sn, SPR:$Sm),
827 IIC_fpUNA32, "vmov", ".f32\t$Sd, $Sm",
828 [/*(set SPR:$Sd, (ARMcmov SPR:$Sn, SPR:$Sm, imm:$cc))*/]>,
829 RegConstraint<"$Sn = $Sd">;
831 def VNEGDcc : ADuI<0b11101, 0b11, 0b0001, 0b01, 0,
832 (outs DPR:$Dd), (ins DPR:$Dn, DPR:$Dm),
833 IIC_fpUNA64, "vneg", ".f64\t$Dd, $Dm",
834 [/*(set DPR:$Dd, (ARMcneg DPR:$Dn, DPR:$Dm, imm:$cc))*/]>,
835 RegConstraint<"$Dn = $Dd">;
837 def VNEGScc : ASuI<0b11101, 0b11, 0b0001, 0b01, 0,
838 (outs SPR:$Sd), (ins SPR:$Sn, SPR:$Sm),
839 IIC_fpUNA32, "vneg", ".f32\t$Sd, $Sm",
840 [/*(set SPR:$Sd, (ARMcneg SPR:$Sn, SPR:$Sm, imm:$cc))*/]>,
841 RegConstraint<"$Sn = $Sd">;
842 } // neverHasSideEffects
844 //===----------------------------------------------------------------------===//
848 // APSR is the application level alias of CPSR. This FPSCR N, Z, C, V flags
850 let Defs = [CPSR], Uses = [FPSCR] in
851 def FMSTAT : VFPAI<(outs), (ins), VFPMiscFrm, IIC_fpSTAT,
852 "vmrs", "\tapsr_nzcv, fpscr",
854 let Inst{27-20} = 0b11101111;
855 let Inst{19-16} = 0b0001;
856 let Inst{15-12} = 0b1111;
857 let Inst{11-8} = 0b1010;
859 let Inst{6-5} = 0b00;
861 let Inst{3-0} = 0b0000;
865 let hasSideEffects = 1, Uses = [FPSCR] in
866 def VMRS : VFPAI<(outs GPR:$Rt), (ins), VFPMiscFrm, IIC_fpSTAT,
867 "vmrs", "\t$Rt, fpscr",
868 [(set GPR:$Rt, (int_arm_get_fpscr))]> {
869 // Instruction operand.
872 // Encode instruction operand.
873 let Inst{15-12} = Rt;
875 let Inst{27-20} = 0b11101111;
876 let Inst{19-16} = 0b0001;
877 let Inst{11-8} = 0b1010;
879 let Inst{6-5} = 0b00;
881 let Inst{3-0} = 0b0000;
884 let Defs = [FPSCR] in
885 def VMSR : VFPAI<(outs), (ins GPR:$src), VFPMiscFrm, IIC_fpSTAT,
886 "vmsr", "\tfpscr, $src",
887 [(int_arm_set_fpscr GPR:$src)]> {
888 // Instruction operand.
891 // Encode instruction operand.
892 let Inst{15-12} = src;
894 let Inst{27-20} = 0b11101110;
895 let Inst{19-16} = 0b0001;
896 let Inst{11-8} = 0b1010;
901 // Materialize FP immediates. VFP3 only.
902 let isReMaterializable = 1 in {
903 def FCONSTD : VFPAI<(outs DPR:$Dd), (ins vfp_f64imm:$imm),
904 VFPMiscFrm, IIC_fpUNA64,
905 "vmov", ".f64\t$Dd, $imm",
906 [(set DPR:$Dd, vfp_f64imm:$imm)]>, Requires<[HasVFP3]> {
907 // Instruction operands.
911 // Encode instruction operands.
912 let Inst{15-12} = Dd{3-0};
913 let Inst{22} = Dd{4};
914 let Inst{19} = imm{31};
915 let Inst{18-16} = imm{22-20};
916 let Inst{3-0} = imm{19-16};
918 // Encode remaining instruction bits.
919 let Inst{27-23} = 0b11101;
920 let Inst{21-20} = 0b11;
921 let Inst{11-9} = 0b101;
922 let Inst{8} = 1; // Double precision.
923 let Inst{7-4} = 0b0000;
926 def FCONSTS : VFPAI<(outs SPR:$Sd), (ins vfp_f32imm:$imm),
927 VFPMiscFrm, IIC_fpUNA32,
928 "vmov", ".f32\t$Sd, $imm",
929 [(set SPR:$Sd, vfp_f32imm:$imm)]>, Requires<[HasVFP3]> {
930 // Instruction operands.
934 // Encode instruction operands.
935 let Inst{15-12} = Sd{4-1};
936 let Inst{22} = Sd{0};
937 let Inst{19} = imm{31}; // The immediate is handled as a double.
938 let Inst{18-16} = imm{22-20};
939 let Inst{3-0} = imm{19-16};
941 // Encode remaining instruction bits.
942 let Inst{27-23} = 0b11101;
943 let Inst{21-20} = 0b11;
944 let Inst{11-9} = 0b101;
945 let Inst{8} = 0; // Single precision.
946 let Inst{7-4} = 0b0000;