1 //===-- X86InstrSSE.td - SSE 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 SSE instruction set, defining the instructions,
11 // and properties of the instructions which are needed for code generation,
12 // machine code emission, and analysis.
14 //===----------------------------------------------------------------------===//
16 class OpndItins<InstrItinClass arg_rr, InstrItinClass arg_rm> {
17 InstrItinClass rr = arg_rr;
18 InstrItinClass rm = arg_rm;
19 // InstrSchedModel info.
20 X86FoldableSchedWrite Sched = WriteFAdd;
23 class SizeItins<OpndItins arg_s, OpndItins arg_d> {
29 class ShiftOpndItins<InstrItinClass arg_rr, InstrItinClass arg_rm,
30 InstrItinClass arg_ri> {
31 InstrItinClass rr = arg_rr;
32 InstrItinClass rm = arg_rm;
33 InstrItinClass ri = arg_ri;
38 let Sched = WriteFAdd in {
39 def SSE_ALU_F32S : OpndItins<
40 IIC_SSE_ALU_F32S_RR, IIC_SSE_ALU_F32S_RM
43 def SSE_ALU_F64S : OpndItins<
44 IIC_SSE_ALU_F64S_RR, IIC_SSE_ALU_F64S_RM
48 def SSE_ALU_ITINS_S : SizeItins<
49 SSE_ALU_F32S, SSE_ALU_F64S
52 let Sched = WriteFMul in {
53 def SSE_MUL_F32S : OpndItins<
54 IIC_SSE_MUL_F32S_RR, IIC_SSE_MUL_F64S_RM
57 def SSE_MUL_F64S : OpndItins<
58 IIC_SSE_MUL_F64S_RR, IIC_SSE_MUL_F64S_RM
62 def SSE_MUL_ITINS_S : SizeItins<
63 SSE_MUL_F32S, SSE_MUL_F64S
66 let Sched = WriteFDiv in {
67 def SSE_DIV_F32S : OpndItins<
68 IIC_SSE_DIV_F32S_RR, IIC_SSE_DIV_F64S_RM
71 def SSE_DIV_F64S : OpndItins<
72 IIC_SSE_DIV_F64S_RR, IIC_SSE_DIV_F64S_RM
76 def SSE_DIV_ITINS_S : SizeItins<
77 SSE_DIV_F32S, SSE_DIV_F64S
81 let Sched = WriteFAdd in {
82 def SSE_ALU_F32P : OpndItins<
83 IIC_SSE_ALU_F32P_RR, IIC_SSE_ALU_F32P_RM
86 def SSE_ALU_F64P : OpndItins<
87 IIC_SSE_ALU_F64P_RR, IIC_SSE_ALU_F64P_RM
91 def SSE_ALU_ITINS_P : SizeItins<
92 SSE_ALU_F32P, SSE_ALU_F64P
95 let Sched = WriteFMul in {
96 def SSE_MUL_F32P : OpndItins<
97 IIC_SSE_MUL_F32P_RR, IIC_SSE_MUL_F64P_RM
100 def SSE_MUL_F64P : OpndItins<
101 IIC_SSE_MUL_F64P_RR, IIC_SSE_MUL_F64P_RM
105 def SSE_MUL_ITINS_P : SizeItins<
106 SSE_MUL_F32P, SSE_MUL_F64P
109 let Sched = WriteFDiv in {
110 def SSE_DIV_F32P : OpndItins<
111 IIC_SSE_DIV_F32P_RR, IIC_SSE_DIV_F64P_RM
114 def SSE_DIV_F64P : OpndItins<
115 IIC_SSE_DIV_F64P_RR, IIC_SSE_DIV_F64P_RM
119 def SSE_DIV_ITINS_P : SizeItins<
120 SSE_DIV_F32P, SSE_DIV_F64P
123 def SSE_BIT_ITINS_P : OpndItins<
124 IIC_SSE_BIT_P_RR, IIC_SSE_BIT_P_RM
127 let Sched = WriteVecALU in {
128 def SSE_INTALU_ITINS_P : OpndItins<
129 IIC_SSE_INTALU_P_RR, IIC_SSE_INTALU_P_RM
132 def SSE_INTALUQ_ITINS_P : OpndItins<
133 IIC_SSE_INTALUQ_P_RR, IIC_SSE_INTALUQ_P_RM
137 let Sched = WriteVecIMul in
138 def SSE_INTMUL_ITINS_P : OpndItins<
139 IIC_SSE_INTMUL_P_RR, IIC_SSE_INTMUL_P_RM
142 def SSE_INTSHIFT_ITINS_P : ShiftOpndItins<
143 IIC_SSE_INTSH_P_RR, IIC_SSE_INTSH_P_RM, IIC_SSE_INTSH_P_RI
146 def SSE_MOVA_ITINS : OpndItins<
147 IIC_SSE_MOVA_P_RR, IIC_SSE_MOVA_P_RM
150 def SSE_MOVU_ITINS : OpndItins<
151 IIC_SSE_MOVU_P_RR, IIC_SSE_MOVU_P_RM
154 def SSE_DPPD_ITINS : OpndItins<
155 IIC_SSE_DPPD_RR, IIC_SSE_DPPD_RM
158 def SSE_DPPS_ITINS : OpndItins<
159 IIC_SSE_DPPS_RR, IIC_SSE_DPPD_RM
162 def DEFAULT_ITINS : OpndItins<
163 IIC_ALU_NONMEM, IIC_ALU_MEM
166 def SSE_EXTRACT_ITINS : OpndItins<
167 IIC_SSE_EXTRACTPS_RR, IIC_SSE_EXTRACTPS_RM
170 def SSE_INSERT_ITINS : OpndItins<
171 IIC_SSE_INSERTPS_RR, IIC_SSE_INSERTPS_RM
174 def SSE_MPSADBW_ITINS : OpndItins<
175 IIC_SSE_MPSADBW_RR, IIC_SSE_MPSADBW_RM
178 def SSE_PMULLD_ITINS : OpndItins<
179 IIC_SSE_PMULLD_RR, IIC_SSE_PMULLD_RM
182 //===----------------------------------------------------------------------===//
183 // SSE 1 & 2 Instructions Classes
184 //===----------------------------------------------------------------------===//
186 /// sse12_fp_scalar - SSE 1 & 2 scalar instructions class
187 multiclass sse12_fp_scalar<bits<8> opc, string OpcodeStr, SDNode OpNode,
188 RegisterClass RC, X86MemOperand x86memop,
191 let isCommutable = 1 in {
192 def rr : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
194 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
195 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
196 [(set RC:$dst, (OpNode RC:$src1, RC:$src2))], itins.rr>,
197 Sched<[itins.Sched]>;
199 def rm : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
201 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
202 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
203 [(set RC:$dst, (OpNode RC:$src1, (load addr:$src2)))], itins.rm>,
204 Sched<[itins.Sched.Folded, ReadAfterLd]>;
207 /// sse12_fp_scalar_int - SSE 1 & 2 scalar instructions intrinsics class
208 multiclass sse12_fp_scalar_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
209 string asm, string SSEVer, string FPSizeStr,
210 Operand memopr, ComplexPattern mem_cpat,
213 def rr_Int : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
215 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
216 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
217 [(set RC:$dst, (!cast<Intrinsic>(
218 !strconcat("int_x86_sse", SSEVer, "_", OpcodeStr, FPSizeStr))
219 RC:$src1, RC:$src2))], itins.rr>,
220 Sched<[itins.Sched]>;
221 def rm_Int : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, memopr:$src2),
223 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
224 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
225 [(set RC:$dst, (!cast<Intrinsic>(!strconcat("int_x86_sse",
226 SSEVer, "_", OpcodeStr, FPSizeStr))
227 RC:$src1, mem_cpat:$src2))], itins.rm>,
228 Sched<[itins.Sched.Folded, ReadAfterLd]>;
231 /// sse12_fp_packed - SSE 1 & 2 packed instructions class
232 multiclass sse12_fp_packed<bits<8> opc, string OpcodeStr, SDNode OpNode,
233 RegisterClass RC, ValueType vt,
234 X86MemOperand x86memop, PatFrag mem_frag,
235 Domain d, OpndItins itins, bit Is2Addr = 1> {
236 let isCommutable = 1 in
237 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
239 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
240 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
241 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], itins.rr, d>,
242 Sched<[itins.Sched]>;
244 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
246 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
247 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
248 [(set RC:$dst, (OpNode RC:$src1, (mem_frag addr:$src2)))],
250 Sched<[itins.Sched.Folded, ReadAfterLd]>;
253 /// sse12_fp_packed_logical_rm - SSE 1 & 2 packed instructions class
254 multiclass sse12_fp_packed_logical_rm<bits<8> opc, RegisterClass RC, Domain d,
255 string OpcodeStr, X86MemOperand x86memop,
256 list<dag> pat_rr, list<dag> pat_rm,
258 let isCommutable = 1, hasSideEffects = 0 in
259 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
261 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
262 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
263 pat_rr, NoItinerary, d>,
264 Sched<[WriteVecLogic]>;
265 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
267 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
268 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
269 pat_rm, NoItinerary, d>,
270 Sched<[WriteVecLogicLd, ReadAfterLd]>;
273 //===----------------------------------------------------------------------===//
274 // Non-instruction patterns
275 //===----------------------------------------------------------------------===//
277 // A vector extract of the first f32/f64 position is a subregister copy
278 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
279 (COPY_TO_REGCLASS (v4f32 VR128:$src), FR32)>;
280 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
281 (COPY_TO_REGCLASS (v2f64 VR128:$src), FR64)>;
283 // A 128-bit subvector extract from the first 256-bit vector position
284 // is a subregister copy that needs no instruction.
285 def : Pat<(v4i32 (extract_subvector (v8i32 VR256:$src), (iPTR 0))),
286 (v4i32 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm))>;
287 def : Pat<(v4f32 (extract_subvector (v8f32 VR256:$src), (iPTR 0))),
288 (v4f32 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_xmm))>;
290 def : Pat<(v2i64 (extract_subvector (v4i64 VR256:$src), (iPTR 0))),
291 (v2i64 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm))>;
292 def : Pat<(v2f64 (extract_subvector (v4f64 VR256:$src), (iPTR 0))),
293 (v2f64 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm))>;
295 def : Pat<(v8i16 (extract_subvector (v16i16 VR256:$src), (iPTR 0))),
296 (v8i16 (EXTRACT_SUBREG (v16i16 VR256:$src), sub_xmm))>;
297 def : Pat<(v16i8 (extract_subvector (v32i8 VR256:$src), (iPTR 0))),
298 (v16i8 (EXTRACT_SUBREG (v32i8 VR256:$src), sub_xmm))>;
300 // A 128-bit subvector insert to the first 256-bit vector position
301 // is a subregister copy that needs no instruction.
302 let AddedComplexity = 25 in { // to give priority over vinsertf128rm
303 def : Pat<(insert_subvector undef, (v2i64 VR128:$src), (iPTR 0)),
304 (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
305 def : Pat<(insert_subvector undef, (v2f64 VR128:$src), (iPTR 0)),
306 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
307 def : Pat<(insert_subvector undef, (v4i32 VR128:$src), (iPTR 0)),
308 (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
309 def : Pat<(insert_subvector undef, (v4f32 VR128:$src), (iPTR 0)),
310 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
311 def : Pat<(insert_subvector undef, (v8i16 VR128:$src), (iPTR 0)),
312 (INSERT_SUBREG (v16i16 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
313 def : Pat<(insert_subvector undef, (v16i8 VR128:$src), (iPTR 0)),
314 (INSERT_SUBREG (v32i8 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
317 // Implicitly promote a 32-bit scalar to a vector.
318 def : Pat<(v4f32 (scalar_to_vector FR32:$src)),
319 (COPY_TO_REGCLASS FR32:$src, VR128)>;
320 def : Pat<(v8f32 (scalar_to_vector FR32:$src)),
321 (COPY_TO_REGCLASS FR32:$src, VR128)>;
322 // Implicitly promote a 64-bit scalar to a vector.
323 def : Pat<(v2f64 (scalar_to_vector FR64:$src)),
324 (COPY_TO_REGCLASS FR64:$src, VR128)>;
325 def : Pat<(v4f64 (scalar_to_vector FR64:$src)),
326 (COPY_TO_REGCLASS FR64:$src, VR128)>;
328 // Bitcasts between 128-bit vector types. Return the original type since
329 // no instruction is needed for the conversion
330 let Predicates = [HasSSE2] in {
331 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
332 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
333 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
334 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
335 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
336 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
337 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
338 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
339 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
340 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
341 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
342 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
343 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
344 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
345 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
346 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
347 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
348 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
349 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
350 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
351 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
352 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
353 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
354 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
355 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
356 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
357 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
358 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
359 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
360 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
363 // Bitcasts between 256-bit vector types. Return the original type since
364 // no instruction is needed for the conversion
365 let Predicates = [HasAVX] in {
366 def : Pat<(v4f64 (bitconvert (v8f32 VR256:$src))), (v4f64 VR256:$src)>;
367 def : Pat<(v4f64 (bitconvert (v8i32 VR256:$src))), (v4f64 VR256:$src)>;
368 def : Pat<(v4f64 (bitconvert (v4i64 VR256:$src))), (v4f64 VR256:$src)>;
369 def : Pat<(v4f64 (bitconvert (v16i16 VR256:$src))), (v4f64 VR256:$src)>;
370 def : Pat<(v4f64 (bitconvert (v32i8 VR256:$src))), (v4f64 VR256:$src)>;
371 def : Pat<(v8f32 (bitconvert (v8i32 VR256:$src))), (v8f32 VR256:$src)>;
372 def : Pat<(v8f32 (bitconvert (v4i64 VR256:$src))), (v8f32 VR256:$src)>;
373 def : Pat<(v8f32 (bitconvert (v4f64 VR256:$src))), (v8f32 VR256:$src)>;
374 def : Pat<(v8f32 (bitconvert (v32i8 VR256:$src))), (v8f32 VR256:$src)>;
375 def : Pat<(v8f32 (bitconvert (v16i16 VR256:$src))), (v8f32 VR256:$src)>;
376 def : Pat<(v4i64 (bitconvert (v8f32 VR256:$src))), (v4i64 VR256:$src)>;
377 def : Pat<(v4i64 (bitconvert (v8i32 VR256:$src))), (v4i64 VR256:$src)>;
378 def : Pat<(v4i64 (bitconvert (v4f64 VR256:$src))), (v4i64 VR256:$src)>;
379 def : Pat<(v4i64 (bitconvert (v32i8 VR256:$src))), (v4i64 VR256:$src)>;
380 def : Pat<(v4i64 (bitconvert (v16i16 VR256:$src))), (v4i64 VR256:$src)>;
381 def : Pat<(v32i8 (bitconvert (v4f64 VR256:$src))), (v32i8 VR256:$src)>;
382 def : Pat<(v32i8 (bitconvert (v4i64 VR256:$src))), (v32i8 VR256:$src)>;
383 def : Pat<(v32i8 (bitconvert (v8f32 VR256:$src))), (v32i8 VR256:$src)>;
384 def : Pat<(v32i8 (bitconvert (v8i32 VR256:$src))), (v32i8 VR256:$src)>;
385 def : Pat<(v32i8 (bitconvert (v16i16 VR256:$src))), (v32i8 VR256:$src)>;
386 def : Pat<(v8i32 (bitconvert (v32i8 VR256:$src))), (v8i32 VR256:$src)>;
387 def : Pat<(v8i32 (bitconvert (v16i16 VR256:$src))), (v8i32 VR256:$src)>;
388 def : Pat<(v8i32 (bitconvert (v8f32 VR256:$src))), (v8i32 VR256:$src)>;
389 def : Pat<(v8i32 (bitconvert (v4i64 VR256:$src))), (v8i32 VR256:$src)>;
390 def : Pat<(v8i32 (bitconvert (v4f64 VR256:$src))), (v8i32 VR256:$src)>;
391 def : Pat<(v16i16 (bitconvert (v8f32 VR256:$src))), (v16i16 VR256:$src)>;
392 def : Pat<(v16i16 (bitconvert (v8i32 VR256:$src))), (v16i16 VR256:$src)>;
393 def : Pat<(v16i16 (bitconvert (v4i64 VR256:$src))), (v16i16 VR256:$src)>;
394 def : Pat<(v16i16 (bitconvert (v4f64 VR256:$src))), (v16i16 VR256:$src)>;
395 def : Pat<(v16i16 (bitconvert (v32i8 VR256:$src))), (v16i16 VR256:$src)>;
398 // Alias instructions that map fld0 to xorps for sse or vxorps for avx.
399 // This is expanded by ExpandPostRAPseudos.
400 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
401 isPseudo = 1, SchedRW = [WriteZero] in {
402 def FsFLD0SS : I<0, Pseudo, (outs FR32:$dst), (ins), "",
403 [(set FR32:$dst, fp32imm0)]>, Requires<[HasSSE1]>;
404 def FsFLD0SD : I<0, Pseudo, (outs FR64:$dst), (ins), "",
405 [(set FR64:$dst, fpimm0)]>, Requires<[HasSSE2]>;
408 //===----------------------------------------------------------------------===//
409 // AVX & SSE - Zero/One Vectors
410 //===----------------------------------------------------------------------===//
412 // Alias instruction that maps zero vector to pxor / xorp* for sse.
413 // This is expanded by ExpandPostRAPseudos to an xorps / vxorps, and then
414 // swizzled by ExecutionDepsFix to pxor.
415 // We set canFoldAsLoad because this can be converted to a constant-pool
416 // load of an all-zeros value if folding it would be beneficial.
417 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
418 isPseudo = 1, SchedRW = [WriteZero] in {
419 def V_SET0 : I<0, Pseudo, (outs VR128:$dst), (ins), "",
420 [(set VR128:$dst, (v4f32 immAllZerosV))]>;
423 def : Pat<(v2f64 immAllZerosV), (V_SET0)>;
424 def : Pat<(v4i32 immAllZerosV), (V_SET0)>;
425 def : Pat<(v2i64 immAllZerosV), (V_SET0)>;
426 def : Pat<(v8i16 immAllZerosV), (V_SET0)>;
427 def : Pat<(v16i8 immAllZerosV), (V_SET0)>;
430 // The same as done above but for AVX. The 256-bit AVX1 ISA doesn't support PI,
431 // and doesn't need it because on sandy bridge the register is set to zero
432 // at the rename stage without using any execution unit, so SET0PSY
433 // and SET0PDY can be used for vector int instructions without penalty
434 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
435 isPseudo = 1, Predicates = [HasAVX], SchedRW = [WriteZero] in {
436 def AVX_SET0 : I<0, Pseudo, (outs VR256:$dst), (ins), "",
437 [(set VR256:$dst, (v8f32 immAllZerosV))]>;
440 let Predicates = [HasAVX] in
441 def : Pat<(v4f64 immAllZerosV), (AVX_SET0)>;
443 let Predicates = [HasAVX2] in {
444 def : Pat<(v4i64 immAllZerosV), (AVX_SET0)>;
445 def : Pat<(v8i32 immAllZerosV), (AVX_SET0)>;
446 def : Pat<(v16i16 immAllZerosV), (AVX_SET0)>;
447 def : Pat<(v32i8 immAllZerosV), (AVX_SET0)>;
450 // AVX1 has no support for 256-bit integer instructions, but since the 128-bit
451 // VPXOR instruction writes zero to its upper part, it's safe build zeros.
452 let Predicates = [HasAVX1Only] in {
453 def : Pat<(v32i8 immAllZerosV), (SUBREG_TO_REG (i8 0), (V_SET0), sub_xmm)>;
454 def : Pat<(bc_v32i8 (v8f32 immAllZerosV)),
455 (SUBREG_TO_REG (i8 0), (V_SET0), sub_xmm)>;
457 def : Pat<(v16i16 immAllZerosV), (SUBREG_TO_REG (i16 0), (V_SET0), sub_xmm)>;
458 def : Pat<(bc_v16i16 (v8f32 immAllZerosV)),
459 (SUBREG_TO_REG (i16 0), (V_SET0), sub_xmm)>;
461 def : Pat<(v8i32 immAllZerosV), (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
462 def : Pat<(bc_v8i32 (v8f32 immAllZerosV)),
463 (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
465 def : Pat<(v4i64 immAllZerosV), (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
466 def : Pat<(bc_v4i64 (v8f32 immAllZerosV)),
467 (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
470 // We set canFoldAsLoad because this can be converted to a constant-pool
471 // load of an all-ones value if folding it would be beneficial.
472 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
473 isPseudo = 1, SchedRW = [WriteZero] in {
474 def V_SETALLONES : I<0, Pseudo, (outs VR128:$dst), (ins), "",
475 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
476 let Predicates = [HasAVX2] in
477 def AVX2_SETALLONES : I<0, Pseudo, (outs VR256:$dst), (ins), "",
478 [(set VR256:$dst, (v8i32 immAllOnesV))]>;
482 //===----------------------------------------------------------------------===//
483 // SSE 1 & 2 - Move FP Scalar Instructions
485 // Move Instructions. Register-to-register movss/movsd is not used for FR32/64
486 // register copies because it's a partial register update; Register-to-register
487 // movss/movsd is not modeled as an INSERT_SUBREG because INSERT_SUBREG requires
488 // that the insert be implementable in terms of a copy, and just mentioned, we
489 // don't use movss/movsd for copies.
490 //===----------------------------------------------------------------------===//
492 multiclass sse12_move_rr<RegisterClass RC, SDNode OpNode, ValueType vt,
493 X86MemOperand x86memop, string base_opc,
495 def rr : SI<0x10, MRMSrcReg, (outs VR128:$dst),
496 (ins VR128:$src1, RC:$src2),
497 !strconcat(base_opc, asm_opr),
498 [(set VR128:$dst, (vt (OpNode VR128:$src1,
499 (scalar_to_vector RC:$src2))))],
500 IIC_SSE_MOV_S_RR>, Sched<[WriteMove]>;
502 // For the disassembler
503 let isCodeGenOnly = 1, hasSideEffects = 0 in
504 def rr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
505 (ins VR128:$src1, RC:$src2),
506 !strconcat(base_opc, asm_opr),
507 [], IIC_SSE_MOV_S_RR>, Sched<[WriteMove]>;
510 multiclass sse12_move<RegisterClass RC, SDNode OpNode, ValueType vt,
511 X86MemOperand x86memop, string OpcodeStr> {
513 defm V#NAME : sse12_move_rr<RC, OpNode, vt, x86memop, OpcodeStr,
514 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">,
517 def V#NAME#mr : SI<0x11, MRMDestMem, (outs), (ins x86memop:$dst, RC:$src),
518 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
519 [(store RC:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
520 VEX, VEX_LIG, Sched<[WriteStore]>;
522 let Constraints = "$src1 = $dst" in {
523 defm NAME : sse12_move_rr<RC, OpNode, vt, x86memop, OpcodeStr,
524 "\t{$src2, $dst|$dst, $src2}">;
527 def NAME#mr : SI<0x11, MRMDestMem, (outs), (ins x86memop:$dst, RC:$src),
528 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
529 [(store RC:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
533 // Loading from memory automatically zeroing upper bits.
534 multiclass sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
535 PatFrag mem_pat, string OpcodeStr> {
536 def V#NAME#rm : SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
537 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
538 [(set RC:$dst, (mem_pat addr:$src))],
539 IIC_SSE_MOV_S_RM>, VEX, VEX_LIG, Sched<[WriteLoad]>;
540 def NAME#rm : SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
541 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
542 [(set RC:$dst, (mem_pat addr:$src))],
543 IIC_SSE_MOV_S_RM>, Sched<[WriteLoad]>;
546 defm MOVSS : sse12_move<FR32, X86Movss, v4f32, f32mem, "movss">, XS;
547 defm MOVSD : sse12_move<FR64, X86Movsd, v2f64, f64mem, "movsd">, XD;
549 let canFoldAsLoad = 1, isReMaterializable = 1 in {
550 defm MOVSS : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
552 let AddedComplexity = 20 in
553 defm MOVSD : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
557 let Predicates = [UseAVX] in {
558 let AddedComplexity = 15 in {
559 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
560 // MOVS{S,D} to the lower bits.
561 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
562 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
563 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
564 (VMOVSSrr (v4f32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
565 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
566 (VMOVSSrr (v4i32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
567 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
568 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
570 // Move low f32 and clear high bits.
571 def : Pat<(v8f32 (X86vzmovl (v8f32 VR256:$src))),
572 (SUBREG_TO_REG (i32 0),
573 (VMOVSSrr (v4f32 (V_SET0)),
574 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_xmm)), sub_xmm)>;
575 def : Pat<(v8i32 (X86vzmovl (v8i32 VR256:$src))),
576 (SUBREG_TO_REG (i32 0),
577 (VMOVSSrr (v4i32 (V_SET0)),
578 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm)), sub_xmm)>;
581 let AddedComplexity = 20 in {
582 // MOVSSrm zeros the high parts of the register; represent this
583 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
584 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
585 (COPY_TO_REGCLASS (VMOVSSrm addr:$src), VR128)>;
586 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
587 (COPY_TO_REGCLASS (VMOVSSrm addr:$src), VR128)>;
588 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
589 (COPY_TO_REGCLASS (VMOVSSrm addr:$src), VR128)>;
591 // MOVSDrm zeros the high parts of the register; represent this
592 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
593 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
594 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
595 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
596 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
597 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
598 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
599 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
600 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
601 def : Pat<(v2f64 (X86vzload addr:$src)),
602 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
604 // Represent the same patterns above but in the form they appear for
606 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
607 (v4i32 (scalar_to_vector (loadi32 addr:$src))), (iPTR 0)))),
608 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_xmm)>;
609 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
610 (v4f32 (scalar_to_vector (loadf32 addr:$src))), (iPTR 0)))),
611 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_xmm)>;
612 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
613 (v2f64 (scalar_to_vector (loadf64 addr:$src))), (iPTR 0)))),
614 (SUBREG_TO_REG (i32 0), (VMOVSDrm addr:$src), sub_xmm)>;
616 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
617 (v4f32 (scalar_to_vector FR32:$src)), (iPTR 0)))),
618 (SUBREG_TO_REG (i32 0),
619 (v4f32 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)),
621 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
622 (v2f64 (scalar_to_vector FR64:$src)), (iPTR 0)))),
623 (SUBREG_TO_REG (i64 0),
624 (v2f64 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)),
626 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
627 (v2i64 (scalar_to_vector (loadi64 addr:$src))), (iPTR 0)))),
628 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_xmm)>;
630 // Move low f64 and clear high bits.
631 def : Pat<(v4f64 (X86vzmovl (v4f64 VR256:$src))),
632 (SUBREG_TO_REG (i32 0),
633 (VMOVSDrr (v2f64 (V_SET0)),
634 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm)), sub_xmm)>;
636 def : Pat<(v4i64 (X86vzmovl (v4i64 VR256:$src))),
637 (SUBREG_TO_REG (i32 0),
638 (VMOVSDrr (v2i64 (V_SET0)),
639 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm)), sub_xmm)>;
641 // Extract and store.
642 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
644 (VMOVSSmr addr:$dst, (COPY_TO_REGCLASS (v4f32 VR128:$src), FR32))>;
645 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
647 (VMOVSDmr addr:$dst, (COPY_TO_REGCLASS (v2f64 VR128:$src), FR64))>;
649 // Shuffle with VMOVSS
650 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
651 (VMOVSSrr (v4i32 VR128:$src1),
652 (COPY_TO_REGCLASS (v4i32 VR128:$src2), FR32))>;
653 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
654 (VMOVSSrr (v4f32 VR128:$src1),
655 (COPY_TO_REGCLASS (v4f32 VR128:$src2), FR32))>;
658 def : Pat<(v8i32 (X86Movss VR256:$src1, VR256:$src2)),
659 (SUBREG_TO_REG (i32 0),
660 (VMOVSSrr (EXTRACT_SUBREG (v8i32 VR256:$src1), sub_xmm),
661 (EXTRACT_SUBREG (v8i32 VR256:$src2), sub_xmm)),
663 def : Pat<(v8f32 (X86Movss VR256:$src1, VR256:$src2)),
664 (SUBREG_TO_REG (i32 0),
665 (VMOVSSrr (EXTRACT_SUBREG (v8f32 VR256:$src1), sub_xmm),
666 (EXTRACT_SUBREG (v8f32 VR256:$src2), sub_xmm)),
669 // Shuffle with VMOVSD
670 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
671 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
672 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
673 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
674 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
675 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
676 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
677 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
680 def : Pat<(v4i64 (X86Movsd VR256:$src1, VR256:$src2)),
681 (SUBREG_TO_REG (i32 0),
682 (VMOVSDrr (EXTRACT_SUBREG (v4i64 VR256:$src1), sub_xmm),
683 (EXTRACT_SUBREG (v4i64 VR256:$src2), sub_xmm)),
685 def : Pat<(v4f64 (X86Movsd VR256:$src1, VR256:$src2)),
686 (SUBREG_TO_REG (i32 0),
687 (VMOVSDrr (EXTRACT_SUBREG (v4f64 VR256:$src1), sub_xmm),
688 (EXTRACT_SUBREG (v4f64 VR256:$src2), sub_xmm)),
692 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
693 // is during lowering, where it's not possible to recognize the fold cause
694 // it has two uses through a bitcast. One use disappears at isel time and the
695 // fold opportunity reappears.
696 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
697 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
698 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
699 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
700 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
701 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
702 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
703 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
706 let Predicates = [UseSSE1] in {
707 let AddedComplexity = 15 in {
708 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
709 // MOVSS to the lower bits.
710 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
711 (MOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
712 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
713 (MOVSSrr (v4f32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
714 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
715 (MOVSSrr (v4i32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
718 let AddedComplexity = 20 in {
719 // MOVSSrm already zeros the high parts of the register.
720 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
721 (COPY_TO_REGCLASS (MOVSSrm addr:$src), VR128)>;
722 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
723 (COPY_TO_REGCLASS (MOVSSrm addr:$src), VR128)>;
724 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
725 (COPY_TO_REGCLASS (MOVSSrm addr:$src), VR128)>;
728 // Extract and store.
729 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
731 (MOVSSmr addr:$dst, (COPY_TO_REGCLASS VR128:$src, FR32))>;
733 // Shuffle with MOVSS
734 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
735 (MOVSSrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR32))>;
736 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
737 (MOVSSrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR32))>;
740 let Predicates = [UseSSE2] in {
741 let AddedComplexity = 15 in {
742 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
743 // MOVSD to the lower bits.
744 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
745 (MOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
748 let AddedComplexity = 20 in {
749 // MOVSDrm already zeros the high parts of the register.
750 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
751 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
752 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
753 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
754 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
755 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
756 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
757 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
758 def : Pat<(v2f64 (X86vzload addr:$src)),
759 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
762 // Extract and store.
763 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
765 (MOVSDmr addr:$dst, (COPY_TO_REGCLASS VR128:$src, FR64))>;
767 // Shuffle with MOVSD
768 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
769 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
770 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
771 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
772 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
773 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
774 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
775 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
777 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
778 // is during lowering, where it's not possible to recognize the fold cause
779 // it has two uses through a bitcast. One use disappears at isel time and the
780 // fold opportunity reappears.
781 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
782 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
783 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
784 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
785 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
786 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
787 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
788 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
791 //===----------------------------------------------------------------------===//
792 // SSE 1 & 2 - Move Aligned/Unaligned FP Instructions
793 //===----------------------------------------------------------------------===//
795 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
796 X86MemOperand x86memop, PatFrag ld_frag,
797 string asm, Domain d,
799 bit IsReMaterializable = 1> {
800 let neverHasSideEffects = 1 in
801 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
802 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], itins.rr, d>,
804 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
805 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
806 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
807 [(set RC:$dst, (ld_frag addr:$src))], itins.rm, d>,
811 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
812 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
814 defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
815 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
817 defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
818 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
820 defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
821 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
824 defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
825 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
827 defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
828 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
829 TB, OpSize, VEX, VEX_L;
830 defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
831 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
833 defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
834 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
835 TB, OpSize, VEX, VEX_L;
836 defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
837 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
839 defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
840 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
842 defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
843 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
845 defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
846 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
849 let SchedRW = [WriteStore] in {
850 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
851 "movaps\t{$src, $dst|$dst, $src}",
852 [(alignedstore (v4f32 VR128:$src), addr:$dst)],
853 IIC_SSE_MOVA_P_MR>, VEX;
854 def VMOVAPDmr : VPDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
855 "movapd\t{$src, $dst|$dst, $src}",
856 [(alignedstore (v2f64 VR128:$src), addr:$dst)],
857 IIC_SSE_MOVA_P_MR>, VEX;
858 def VMOVUPSmr : VPSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
859 "movups\t{$src, $dst|$dst, $src}",
860 [(store (v4f32 VR128:$src), addr:$dst)],
861 IIC_SSE_MOVU_P_MR>, VEX;
862 def VMOVUPDmr : VPDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
863 "movupd\t{$src, $dst|$dst, $src}",
864 [(store (v2f64 VR128:$src), addr:$dst)],
865 IIC_SSE_MOVU_P_MR>, VEX;
866 def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
867 "movaps\t{$src, $dst|$dst, $src}",
868 [(alignedstore256 (v8f32 VR256:$src), addr:$dst)],
869 IIC_SSE_MOVA_P_MR>, VEX, VEX_L;
870 def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
871 "movapd\t{$src, $dst|$dst, $src}",
872 [(alignedstore256 (v4f64 VR256:$src), addr:$dst)],
873 IIC_SSE_MOVA_P_MR>, VEX, VEX_L;
874 def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
875 "movups\t{$src, $dst|$dst, $src}",
876 [(store (v8f32 VR256:$src), addr:$dst)],
877 IIC_SSE_MOVU_P_MR>, VEX, VEX_L;
878 def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
879 "movupd\t{$src, $dst|$dst, $src}",
880 [(store (v4f64 VR256:$src), addr:$dst)],
881 IIC_SSE_MOVU_P_MR>, VEX, VEX_L;
885 let isCodeGenOnly = 1, hasSideEffects = 0, SchedRW = [WriteMove] in {
886 def VMOVAPSrr_REV : VPSI<0x29, MRMDestReg, (outs VR128:$dst),
888 "movaps\t{$src, $dst|$dst, $src}", [],
889 IIC_SSE_MOVA_P_RR>, VEX;
890 def VMOVAPDrr_REV : VPDI<0x29, MRMDestReg, (outs VR128:$dst),
892 "movapd\t{$src, $dst|$dst, $src}", [],
893 IIC_SSE_MOVA_P_RR>, VEX;
894 def VMOVUPSrr_REV : VPSI<0x11, MRMDestReg, (outs VR128:$dst),
896 "movups\t{$src, $dst|$dst, $src}", [],
897 IIC_SSE_MOVU_P_RR>, VEX;
898 def VMOVUPDrr_REV : VPDI<0x11, MRMDestReg, (outs VR128:$dst),
900 "movupd\t{$src, $dst|$dst, $src}", [],
901 IIC_SSE_MOVU_P_RR>, VEX;
902 def VMOVAPSYrr_REV : VPSI<0x29, MRMDestReg, (outs VR256:$dst),
904 "movaps\t{$src, $dst|$dst, $src}", [],
905 IIC_SSE_MOVA_P_RR>, VEX, VEX_L;
906 def VMOVAPDYrr_REV : VPDI<0x29, MRMDestReg, (outs VR256:$dst),
908 "movapd\t{$src, $dst|$dst, $src}", [],
909 IIC_SSE_MOVA_P_RR>, VEX, VEX_L;
910 def VMOVUPSYrr_REV : VPSI<0x11, MRMDestReg, (outs VR256:$dst),
912 "movups\t{$src, $dst|$dst, $src}", [],
913 IIC_SSE_MOVU_P_RR>, VEX, VEX_L;
914 def VMOVUPDYrr_REV : VPDI<0x11, MRMDestReg, (outs VR256:$dst),
916 "movupd\t{$src, $dst|$dst, $src}", [],
917 IIC_SSE_MOVU_P_RR>, VEX, VEX_L;
920 let Predicates = [HasAVX] in {
921 def : Pat<(v8i32 (X86vzmovl
922 (insert_subvector undef, (v4i32 VR128:$src), (iPTR 0)))),
923 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
924 def : Pat<(v4i64 (X86vzmovl
925 (insert_subvector undef, (v2i64 VR128:$src), (iPTR 0)))),
926 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
927 def : Pat<(v8f32 (X86vzmovl
928 (insert_subvector undef, (v4f32 VR128:$src), (iPTR 0)))),
929 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
930 def : Pat<(v4f64 (X86vzmovl
931 (insert_subvector undef, (v2f64 VR128:$src), (iPTR 0)))),
932 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
936 def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
937 (VMOVUPSYmr addr:$dst, VR256:$src)>;
938 def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
939 (VMOVUPDYmr addr:$dst, VR256:$src)>;
941 let SchedRW = [WriteStore] in {
942 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
943 "movaps\t{$src, $dst|$dst, $src}",
944 [(alignedstore (v4f32 VR128:$src), addr:$dst)],
946 def MOVAPDmr : PDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
947 "movapd\t{$src, $dst|$dst, $src}",
948 [(alignedstore (v2f64 VR128:$src), addr:$dst)],
950 def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
951 "movups\t{$src, $dst|$dst, $src}",
952 [(store (v4f32 VR128:$src), addr:$dst)],
954 def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
955 "movupd\t{$src, $dst|$dst, $src}",
956 [(store (v2f64 VR128:$src), addr:$dst)],
961 let isCodeGenOnly = 1, hasSideEffects = 0, SchedRW = [WriteMove] in {
962 def MOVAPSrr_REV : PSI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
963 "movaps\t{$src, $dst|$dst, $src}", [],
965 def MOVAPDrr_REV : PDI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
966 "movapd\t{$src, $dst|$dst, $src}", [],
968 def MOVUPSrr_REV : PSI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
969 "movups\t{$src, $dst|$dst, $src}", [],
971 def MOVUPDrr_REV : PDI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
972 "movupd\t{$src, $dst|$dst, $src}", [],
976 let Predicates = [HasAVX] in {
977 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
978 (VMOVUPSmr addr:$dst, VR128:$src)>;
979 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
980 (VMOVUPDmr addr:$dst, VR128:$src)>;
983 let Predicates = [UseSSE1] in
984 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
985 (MOVUPSmr addr:$dst, VR128:$src)>;
986 let Predicates = [UseSSE2] in
987 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
988 (MOVUPDmr addr:$dst, VR128:$src)>;
990 // Use vmovaps/vmovups for AVX integer load/store.
991 let Predicates = [HasAVX] in {
992 // 128-bit load/store
993 def : Pat<(alignedloadv2i64 addr:$src),
994 (VMOVAPSrm addr:$src)>;
995 def : Pat<(loadv2i64 addr:$src),
996 (VMOVUPSrm addr:$src)>;
998 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
999 (VMOVAPSmr addr:$dst, VR128:$src)>;
1000 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
1001 (VMOVAPSmr addr:$dst, VR128:$src)>;
1002 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
1003 (VMOVAPSmr addr:$dst, VR128:$src)>;
1004 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
1005 (VMOVAPSmr addr:$dst, VR128:$src)>;
1006 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
1007 (VMOVUPSmr addr:$dst, VR128:$src)>;
1008 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
1009 (VMOVUPSmr addr:$dst, VR128:$src)>;
1010 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
1011 (VMOVUPSmr addr:$dst, VR128:$src)>;
1012 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
1013 (VMOVUPSmr addr:$dst, VR128:$src)>;
1015 // 256-bit load/store
1016 def : Pat<(alignedloadv4i64 addr:$src),
1017 (VMOVAPSYrm addr:$src)>;
1018 def : Pat<(loadv4i64 addr:$src),
1019 (VMOVUPSYrm addr:$src)>;
1020 def : Pat<(alignedstore256 (v4i64 VR256:$src), addr:$dst),
1021 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1022 def : Pat<(alignedstore256 (v8i32 VR256:$src), addr:$dst),
1023 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1024 def : Pat<(alignedstore256 (v16i16 VR256:$src), addr:$dst),
1025 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1026 def : Pat<(alignedstore256 (v32i8 VR256:$src), addr:$dst),
1027 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1028 def : Pat<(store (v4i64 VR256:$src), addr:$dst),
1029 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1030 def : Pat<(store (v8i32 VR256:$src), addr:$dst),
1031 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1032 def : Pat<(store (v16i16 VR256:$src), addr:$dst),
1033 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1034 def : Pat<(store (v32i8 VR256:$src), addr:$dst),
1035 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1037 // Special patterns for storing subvector extracts of lower 128-bits
1038 // Its cheaper to just use VMOVAPS/VMOVUPS instead of VEXTRACTF128mr
1039 def : Pat<(alignedstore (v2f64 (extract_subvector
1040 (v4f64 VR256:$src), (iPTR 0))), addr:$dst),
1041 (VMOVAPDmr addr:$dst, (v2f64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1042 def : Pat<(alignedstore (v4f32 (extract_subvector
1043 (v8f32 VR256:$src), (iPTR 0))), addr:$dst),
1044 (VMOVAPSmr addr:$dst, (v4f32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1045 def : Pat<(alignedstore (v2i64 (extract_subvector
1046 (v4i64 VR256:$src), (iPTR 0))), addr:$dst),
1047 (VMOVAPDmr addr:$dst, (v2i64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1048 def : Pat<(alignedstore (v4i32 (extract_subvector
1049 (v8i32 VR256:$src), (iPTR 0))), addr:$dst),
1050 (VMOVAPSmr addr:$dst, (v4i32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1051 def : Pat<(alignedstore (v8i16 (extract_subvector
1052 (v16i16 VR256:$src), (iPTR 0))), addr:$dst),
1053 (VMOVAPSmr addr:$dst, (v8i16 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1054 def : Pat<(alignedstore (v16i8 (extract_subvector
1055 (v32i8 VR256:$src), (iPTR 0))), addr:$dst),
1056 (VMOVAPSmr addr:$dst, (v16i8 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1058 def : Pat<(store (v2f64 (extract_subvector
1059 (v4f64 VR256:$src), (iPTR 0))), addr:$dst),
1060 (VMOVUPDmr addr:$dst, (v2f64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1061 def : Pat<(store (v4f32 (extract_subvector
1062 (v8f32 VR256:$src), (iPTR 0))), addr:$dst),
1063 (VMOVUPSmr addr:$dst, (v4f32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1064 def : Pat<(store (v2i64 (extract_subvector
1065 (v4i64 VR256:$src), (iPTR 0))), addr:$dst),
1066 (VMOVUPDmr addr:$dst, (v2i64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1067 def : Pat<(store (v4i32 (extract_subvector
1068 (v8i32 VR256:$src), (iPTR 0))), addr:$dst),
1069 (VMOVUPSmr addr:$dst, (v4i32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1070 def : Pat<(store (v8i16 (extract_subvector
1071 (v16i16 VR256:$src), (iPTR 0))), addr:$dst),
1072 (VMOVUPSmr addr:$dst, (v8i16 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1073 def : Pat<(store (v16i8 (extract_subvector
1074 (v32i8 VR256:$src), (iPTR 0))), addr:$dst),
1075 (VMOVUPSmr addr:$dst, (v16i8 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1078 // Use movaps / movups for SSE integer load / store (one byte shorter).
1079 // The instructions selected below are then converted to MOVDQA/MOVDQU
1080 // during the SSE domain pass.
1081 let Predicates = [UseSSE1] in {
1082 def : Pat<(alignedloadv2i64 addr:$src),
1083 (MOVAPSrm addr:$src)>;
1084 def : Pat<(loadv2i64 addr:$src),
1085 (MOVUPSrm addr:$src)>;
1087 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
1088 (MOVAPSmr addr:$dst, VR128:$src)>;
1089 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
1090 (MOVAPSmr addr:$dst, VR128:$src)>;
1091 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
1092 (MOVAPSmr addr:$dst, VR128:$src)>;
1093 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
1094 (MOVAPSmr addr:$dst, VR128:$src)>;
1095 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
1096 (MOVUPSmr addr:$dst, VR128:$src)>;
1097 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
1098 (MOVUPSmr addr:$dst, VR128:$src)>;
1099 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
1100 (MOVUPSmr addr:$dst, VR128:$src)>;
1101 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
1102 (MOVUPSmr addr:$dst, VR128:$src)>;
1105 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
1106 // bits are disregarded. FIXME: Set encoding to pseudo!
1107 let canFoldAsLoad = 1, isReMaterializable = 1, SchedRW = [WriteLoad] in {
1108 let isCodeGenOnly = 1 in {
1109 def FsVMOVAPSrm : VPSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1110 "movaps\t{$src, $dst|$dst, $src}",
1111 [(set FR32:$dst, (alignedloadfsf32 addr:$src))],
1112 IIC_SSE_MOVA_P_RM>, VEX;
1113 def FsVMOVAPDrm : VPDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1114 "movapd\t{$src, $dst|$dst, $src}",
1115 [(set FR64:$dst, (alignedloadfsf64 addr:$src))],
1116 IIC_SSE_MOVA_P_RM>, VEX;
1117 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1118 "movaps\t{$src, $dst|$dst, $src}",
1119 [(set FR32:$dst, (alignedloadfsf32 addr:$src))],
1121 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1122 "movapd\t{$src, $dst|$dst, $src}",
1123 [(set FR64:$dst, (alignedloadfsf64 addr:$src))],
1128 //===----------------------------------------------------------------------===//
1129 // SSE 1 & 2 - Move Low packed FP Instructions
1130 //===----------------------------------------------------------------------===//
1132 multiclass sse12_mov_hilo_packed_base<bits<8>opc, SDNode psnode, SDNode pdnode,
1133 string base_opc, string asm_opr,
1134 InstrItinClass itin> {
1135 def PSrm : PI<opc, MRMSrcMem,
1136 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
1137 !strconcat(base_opc, "s", asm_opr),
1139 (psnode VR128:$src1,
1140 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
1141 itin, SSEPackedSingle>, TB,
1142 Sched<[WriteShuffleLd, ReadAfterLd]>;
1144 def PDrm : PI<opc, MRMSrcMem,
1145 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
1146 !strconcat(base_opc, "d", asm_opr),
1147 [(set VR128:$dst, (v2f64 (pdnode VR128:$src1,
1148 (scalar_to_vector (loadf64 addr:$src2)))))],
1149 itin, SSEPackedDouble>, TB, OpSize,
1150 Sched<[WriteShuffleLd, ReadAfterLd]>;
1154 multiclass sse12_mov_hilo_packed<bits<8>opc, SDNode psnode, SDNode pdnode,
1155 string base_opc, InstrItinClass itin> {
1156 defm V#NAME : sse12_mov_hilo_packed_base<opc, psnode, pdnode, base_opc,
1157 "\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1160 let Constraints = "$src1 = $dst" in
1161 defm NAME : sse12_mov_hilo_packed_base<opc, psnode, pdnode, base_opc,
1162 "\t{$src2, $dst|$dst, $src2}",
1166 let AddedComplexity = 20 in {
1167 defm MOVL : sse12_mov_hilo_packed<0x12, X86Movlps, X86Movlpd, "movlp",
1171 let SchedRW = [WriteStore] in {
1172 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1173 "movlps\t{$src, $dst|$dst, $src}",
1174 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1175 (iPTR 0))), addr:$dst)],
1176 IIC_SSE_MOV_LH>, VEX;
1177 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1178 "movlpd\t{$src, $dst|$dst, $src}",
1179 [(store (f64 (vector_extract (v2f64 VR128:$src),
1180 (iPTR 0))), addr:$dst)],
1181 IIC_SSE_MOV_LH>, VEX;
1182 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1183 "movlps\t{$src, $dst|$dst, $src}",
1184 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1185 (iPTR 0))), addr:$dst)],
1187 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1188 "movlpd\t{$src, $dst|$dst, $src}",
1189 [(store (f64 (vector_extract (v2f64 VR128:$src),
1190 (iPTR 0))), addr:$dst)],
1194 let Predicates = [HasAVX] in {
1195 // Shuffle with VMOVLPS
1196 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1197 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1198 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1199 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1201 // Shuffle with VMOVLPD
1202 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1203 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1204 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1205 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1208 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1210 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1211 def : Pat<(store (v4i32 (X86Movlps
1212 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
1213 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1214 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1216 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1217 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1219 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1222 let Predicates = [UseSSE1] in {
1223 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
1224 def : Pat<(store (i64 (vector_extract (bc_v2i64 (v4f32 VR128:$src2)),
1225 (iPTR 0))), addr:$src1),
1226 (MOVLPSmr addr:$src1, VR128:$src2)>;
1228 // Shuffle with MOVLPS
1229 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1230 (MOVLPSrm VR128:$src1, addr:$src2)>;
1231 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1232 (MOVLPSrm VR128:$src1, addr:$src2)>;
1233 def : Pat<(X86Movlps VR128:$src1,
1234 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1235 (MOVLPSrm VR128:$src1, addr:$src2)>;
1238 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1240 (MOVLPSmr addr:$src1, VR128:$src2)>;
1241 def : Pat<(store (v4i32 (X86Movlps
1242 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
1244 (MOVLPSmr addr:$src1, VR128:$src2)>;
1247 let Predicates = [UseSSE2] in {
1248 // Shuffle with MOVLPD
1249 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1250 (MOVLPDrm VR128:$src1, addr:$src2)>;
1251 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1252 (MOVLPDrm VR128:$src1, addr:$src2)>;
1255 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1257 (MOVLPDmr addr:$src1, VR128:$src2)>;
1258 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1260 (MOVLPDmr addr:$src1, VR128:$src2)>;
1263 //===----------------------------------------------------------------------===//
1264 // SSE 1 & 2 - Move Hi packed FP Instructions
1265 //===----------------------------------------------------------------------===//
1267 let AddedComplexity = 20 in {
1268 defm MOVH : sse12_mov_hilo_packed<0x16, X86Movlhps, X86Movlhpd, "movhp",
1272 let SchedRW = [WriteStore] in {
1273 // v2f64 extract element 1 is always custom lowered to unpack high to low
1274 // and extract element 0 so the non-store version isn't too horrible.
1275 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1276 "movhps\t{$src, $dst|$dst, $src}",
1277 [(store (f64 (vector_extract
1278 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1279 (bc_v2f64 (v4f32 VR128:$src))),
1280 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>, VEX;
1281 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1282 "movhpd\t{$src, $dst|$dst, $src}",
1283 [(store (f64 (vector_extract
1284 (v2f64 (X86Unpckh VR128:$src, VR128:$src)),
1285 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>, VEX;
1286 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1287 "movhps\t{$src, $dst|$dst, $src}",
1288 [(store (f64 (vector_extract
1289 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1290 (bc_v2f64 (v4f32 VR128:$src))),
1291 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>;
1292 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1293 "movhpd\t{$src, $dst|$dst, $src}",
1294 [(store (f64 (vector_extract
1295 (v2f64 (X86Unpckh VR128:$src, VR128:$src)),
1296 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>;
1299 let Predicates = [HasAVX] in {
1301 def : Pat<(X86Movlhps VR128:$src1,
1302 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1303 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1304 def : Pat<(X86Movlhps VR128:$src1,
1305 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1306 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1308 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1309 // is during lowering, where it's not possible to recognize the load fold
1310 // cause it has two uses through a bitcast. One use disappears at isel time
1311 // and the fold opportunity reappears.
1312 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1313 (scalar_to_vector (loadf64 addr:$src2)))),
1314 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1317 let Predicates = [UseSSE1] in {
1319 def : Pat<(X86Movlhps VR128:$src1,
1320 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1321 (MOVHPSrm VR128:$src1, addr:$src2)>;
1322 def : Pat<(X86Movlhps VR128:$src1,
1323 (bc_v4f32 (v2i64 (X86vzload addr:$src2)))),
1324 (MOVHPSrm VR128:$src1, addr:$src2)>;
1327 let Predicates = [UseSSE2] in {
1328 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1329 // is during lowering, where it's not possible to recognize the load fold
1330 // cause it has two uses through a bitcast. One use disappears at isel time
1331 // and the fold opportunity reappears.
1332 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1333 (scalar_to_vector (loadf64 addr:$src2)))),
1334 (MOVHPDrm VR128:$src1, addr:$src2)>;
1337 //===----------------------------------------------------------------------===//
1338 // SSE 1 & 2 - Move Low to High and High to Low packed FP Instructions
1339 //===----------------------------------------------------------------------===//
1341 let AddedComplexity = 20, Predicates = [UseAVX] in {
1342 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
1343 (ins VR128:$src1, VR128:$src2),
1344 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1346 (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))],
1348 VEX_4V, Sched<[WriteShuffle]>;
1349 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
1350 (ins VR128:$src1, VR128:$src2),
1351 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1353 (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))],
1355 VEX_4V, Sched<[WriteShuffle]>;
1357 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1358 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
1359 (ins VR128:$src1, VR128:$src2),
1360 "movlhps\t{$src2, $dst|$dst, $src2}",
1362 (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))],
1363 IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>;
1364 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
1365 (ins VR128:$src1, VR128:$src2),
1366 "movhlps\t{$src2, $dst|$dst, $src2}",
1368 (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))],
1369 IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>;
1372 let Predicates = [UseAVX] in {
1374 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1375 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1376 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1377 (VMOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1380 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1381 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1384 let Predicates = [UseSSE1] in {
1386 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1387 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1388 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1389 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1392 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1393 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1396 //===----------------------------------------------------------------------===//
1397 // SSE 1 & 2 - Conversion Instructions
1398 //===----------------------------------------------------------------------===//
1400 def SSE_CVT_PD : OpndItins<
1401 IIC_SSE_CVT_PD_RR, IIC_SSE_CVT_PD_RM
1404 let Sched = WriteCvtI2F in
1405 def SSE_CVT_PS : OpndItins<
1406 IIC_SSE_CVT_PS_RR, IIC_SSE_CVT_PS_RM
1409 let Sched = WriteCvtI2F in
1410 def SSE_CVT_Scalar : OpndItins<
1411 IIC_SSE_CVT_Scalar_RR, IIC_SSE_CVT_Scalar_RM
1414 let Sched = WriteCvtF2I in
1415 def SSE_CVT_SS2SI_32 : OpndItins<
1416 IIC_SSE_CVT_SS2SI32_RR, IIC_SSE_CVT_SS2SI32_RM
1419 let Sched = WriteCvtF2I in
1420 def SSE_CVT_SS2SI_64 : OpndItins<
1421 IIC_SSE_CVT_SS2SI64_RR, IIC_SSE_CVT_SS2SI64_RM
1424 let Sched = WriteCvtF2I in
1425 def SSE_CVT_SD2SI : OpndItins<
1426 IIC_SSE_CVT_SD2SI_RR, IIC_SSE_CVT_SD2SI_RM
1429 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1430 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1431 string asm, OpndItins itins> {
1432 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1433 [(set DstRC:$dst, (OpNode SrcRC:$src))],
1434 itins.rr>, Sched<[itins.Sched]>;
1435 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1436 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))],
1437 itins.rm>, Sched<[itins.Sched.Folded]>;
1440 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1441 X86MemOperand x86memop, string asm, Domain d,
1443 let neverHasSideEffects = 1 in {
1444 def rr : I<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1445 [], itins.rr, d>, Sched<[itins.Sched]>;
1447 def rm : I<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1448 [], itins.rm, d>, Sched<[itins.Sched.Folded]>;
1452 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1453 X86MemOperand x86memop, string asm> {
1454 let neverHasSideEffects = 1, Predicates = [UseAVX] in {
1455 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
1456 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>,
1457 Sched<[WriteCvtI2F]>;
1459 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1460 (ins DstRC:$src1, x86memop:$src),
1461 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>,
1462 Sched<[WriteCvtI2FLd, ReadAfterLd]>;
1463 } // neverHasSideEffects = 1
1466 let Predicates = [UseAVX] in {
1467 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1468 "cvttss2si\t{$src, $dst|$dst, $src}",
1471 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1472 "cvttss2si\t{$src, $dst|$dst, $src}",
1474 XS, VEX, VEX_W, VEX_LIG;
1475 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1476 "cvttsd2si\t{$src, $dst|$dst, $src}",
1479 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1480 "cvttsd2si\t{$src, $dst|$dst, $src}",
1482 XD, VEX, VEX_W, VEX_LIG;
1484 def : InstAlias<"vcvttss2si{l}\t{$src, $dst|$dst, $src}",
1485 (VCVTTSS2SIrr GR32:$dst, FR32:$src), 0>;
1486 def : InstAlias<"vcvttss2si{l}\t{$src, $dst|$dst, $src}",
1487 (VCVTTSS2SIrm GR32:$dst, f32mem:$src), 0>;
1488 def : InstAlias<"vcvttsd2si{l}\t{$src, $dst|$dst, $src}",
1489 (VCVTTSD2SIrr GR32:$dst, FR64:$src), 0>;
1490 def : InstAlias<"vcvttsd2si{l}\t{$src, $dst|$dst, $src}",
1491 (VCVTTSD2SIrm GR32:$dst, f64mem:$src), 0>;
1492 def : InstAlias<"vcvttss2si{q}\t{$src, $dst|$dst, $src}",
1493 (VCVTTSS2SI64rr GR64:$dst, FR32:$src), 0>;
1494 def : InstAlias<"vcvttss2si{q}\t{$src, $dst|$dst, $src}",
1495 (VCVTTSS2SI64rm GR64:$dst, f32mem:$src), 0>;
1496 def : InstAlias<"vcvttsd2si{q}\t{$src, $dst|$dst, $src}",
1497 (VCVTTSD2SI64rr GR64:$dst, FR64:$src), 0>;
1498 def : InstAlias<"vcvttsd2si{q}\t{$src, $dst|$dst, $src}",
1499 (VCVTTSD2SI64rm GR64:$dst, f64mem:$src), 0>;
1501 // The assembler can recognize rr 64-bit instructions by seeing a rxx
1502 // register, but the same isn't true when only using memory operands,
1503 // provide other assembly "l" and "q" forms to address this explicitly
1504 // where appropriate to do so.
1505 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss{l}">,
1506 XS, VEX_4V, VEX_LIG;
1507 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">,
1508 XS, VEX_4V, VEX_W, VEX_LIG;
1509 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">,
1510 XD, VEX_4V, VEX_LIG;
1511 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">,
1512 XD, VEX_4V, VEX_W, VEX_LIG;
1514 let Predicates = [UseAVX] in {
1515 def : InstAlias<"vcvtsi2ss\t{$src, $src1, $dst|$dst, $src1, $src}",
1516 (VCVTSI2SSrm FR64:$dst, FR64:$src1, i32mem:$src)>;
1517 def : InstAlias<"vcvtsi2sd\t{$src, $src1, $dst|$dst, $src1, $src}",
1518 (VCVTSI2SDrm FR64:$dst, FR64:$src1, i32mem:$src)>;
1520 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
1521 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1522 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
1523 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
1524 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
1525 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
1526 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
1527 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
1529 def : Pat<(f32 (sint_to_fp GR32:$src)),
1530 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
1531 def : Pat<(f32 (sint_to_fp GR64:$src)),
1532 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
1533 def : Pat<(f64 (sint_to_fp GR32:$src)),
1534 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
1535 def : Pat<(f64 (sint_to_fp GR64:$src)),
1536 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
1539 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1540 "cvttss2si\t{$src, $dst|$dst, $src}",
1541 SSE_CVT_SS2SI_32>, XS;
1542 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1543 "cvttss2si\t{$src, $dst|$dst, $src}",
1544 SSE_CVT_SS2SI_64>, XS, REX_W;
1545 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1546 "cvttsd2si\t{$src, $dst|$dst, $src}",
1548 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1549 "cvttsd2si\t{$src, $dst|$dst, $src}",
1550 SSE_CVT_SD2SI>, XD, REX_W;
1551 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
1552 "cvtsi2ss{l}\t{$src, $dst|$dst, $src}",
1553 SSE_CVT_Scalar>, XS;
1554 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
1555 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}",
1556 SSE_CVT_Scalar>, XS, REX_W;
1557 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
1558 "cvtsi2sd{l}\t{$src, $dst|$dst, $src}",
1559 SSE_CVT_Scalar>, XD;
1560 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
1561 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}",
1562 SSE_CVT_Scalar>, XD, REX_W;
1564 def : InstAlias<"cvttss2si{l}\t{$src, $dst|$dst, $src}",
1565 (CVTTSS2SIrr GR32:$dst, FR32:$src), 0>;
1566 def : InstAlias<"cvttss2si{l}\t{$src, $dst|$dst, $src}",
1567 (CVTTSS2SIrm GR32:$dst, f32mem:$src), 0>;
1568 def : InstAlias<"cvttsd2si{l}\t{$src, $dst|$dst, $src}",
1569 (CVTTSD2SIrr GR32:$dst, FR64:$src), 0>;
1570 def : InstAlias<"cvttsd2si{l}\t{$src, $dst|$dst, $src}",
1571 (CVTTSD2SIrm GR32:$dst, f64mem:$src), 0>;
1572 def : InstAlias<"cvttss2si{q}\t{$src, $dst|$dst, $src}",
1573 (CVTTSS2SI64rr GR64:$dst, FR32:$src), 0>;
1574 def : InstAlias<"cvttss2si{q}\t{$src, $dst|$dst, $src}",
1575 (CVTTSS2SI64rm GR64:$dst, f32mem:$src), 0>;
1576 def : InstAlias<"cvttsd2si{q}\t{$src, $dst|$dst, $src}",
1577 (CVTTSD2SI64rr GR64:$dst, FR64:$src), 0>;
1578 def : InstAlias<"cvttsd2si{q}\t{$src, $dst|$dst, $src}",
1579 (CVTTSD2SI64rm GR64:$dst, f64mem:$src), 0>;
1581 def : InstAlias<"cvtsi2ss\t{$src, $dst|$dst, $src}",
1582 (CVTSI2SSrm FR64:$dst, i32mem:$src)>;
1583 def : InstAlias<"cvtsi2sd\t{$src, $dst|$dst, $src}",
1584 (CVTSI2SDrm FR64:$dst, i32mem:$src)>;
1586 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
1587 // and/or XMM operand(s).
1589 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1590 Intrinsic Int, Operand memop, ComplexPattern mem_cpat,
1591 string asm, OpndItins itins> {
1592 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
1593 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1594 [(set DstRC:$dst, (Int SrcRC:$src))], itins.rr>,
1595 Sched<[itins.Sched]>;
1596 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins memop:$src),
1597 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1598 [(set DstRC:$dst, (Int mem_cpat:$src))], itins.rm>,
1599 Sched<[itins.Sched.Folded]>;
1602 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
1603 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
1604 PatFrag ld_frag, string asm, OpndItins itins,
1606 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
1608 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1609 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1610 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))],
1611 itins.rr>, Sched<[itins.Sched]>;
1612 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1613 (ins DstRC:$src1, x86memop:$src2),
1615 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1616 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1617 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))],
1618 itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
1621 let Predicates = [UseAVX] in {
1622 defm VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32,
1623 int_x86_sse2_cvtsd2si, sdmem, sse_load_f64, "cvtsd2si",
1624 SSE_CVT_SD2SI>, XD, VEX, VEX_LIG;
1625 defm VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
1626 int_x86_sse2_cvtsd2si64, sdmem, sse_load_f64, "cvtsd2si",
1627 SSE_CVT_SD2SI>, XD, VEX, VEX_W, VEX_LIG;
1629 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1630 sdmem, sse_load_f64, "cvtsd2si", SSE_CVT_SD2SI>, XD;
1631 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
1632 sdmem, sse_load_f64, "cvtsd2si", SSE_CVT_SD2SI>, XD, REX_W;
1635 let Predicates = [UseAVX] in {
1636 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1637 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss{l}",
1638 SSE_CVT_Scalar, 0>, XS, VEX_4V;
1639 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1640 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss{q}",
1641 SSE_CVT_Scalar, 0>, XS, VEX_4V,
1643 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1644 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd{l}",
1645 SSE_CVT_Scalar, 0>, XD, VEX_4V;
1646 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1647 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd{q}",
1648 SSE_CVT_Scalar, 0>, XD,
1651 let Constraints = "$src1 = $dst" in {
1652 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1653 int_x86_sse_cvtsi2ss, i32mem, loadi32,
1654 "cvtsi2ss{l}", SSE_CVT_Scalar>, XS;
1655 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1656 int_x86_sse_cvtsi642ss, i64mem, loadi64,
1657 "cvtsi2ss{q}", SSE_CVT_Scalar>, XS, REX_W;
1658 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1659 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
1660 "cvtsi2sd{l}", SSE_CVT_Scalar>, XD;
1661 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1662 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
1663 "cvtsi2sd{q}", SSE_CVT_Scalar>, XD, REX_W;
1668 // Aliases for intrinsics
1669 let Predicates = [UseAVX] in {
1670 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1671 ssmem, sse_load_f32, "cvttss2si",
1672 SSE_CVT_SS2SI_32>, XS, VEX;
1673 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1674 int_x86_sse_cvttss2si64, ssmem, sse_load_f32,
1675 "cvttss2si", SSE_CVT_SS2SI_64>,
1677 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1678 sdmem, sse_load_f64, "cvttsd2si",
1679 SSE_CVT_SD2SI>, XD, VEX;
1680 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1681 int_x86_sse2_cvttsd2si64, sdmem, sse_load_f64,
1682 "cvttsd2si", SSE_CVT_SD2SI>,
1685 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1686 ssmem, sse_load_f32, "cvttss2si",
1687 SSE_CVT_SS2SI_32>, XS;
1688 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1689 int_x86_sse_cvttss2si64, ssmem, sse_load_f32,
1690 "cvttss2si", SSE_CVT_SS2SI_64>, XS, REX_W;
1691 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1692 sdmem, sse_load_f64, "cvttsd2si",
1694 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1695 int_x86_sse2_cvttsd2si64, sdmem, sse_load_f64,
1696 "cvttsd2si", SSE_CVT_SD2SI>, XD, REX_W;
1698 let Predicates = [UseAVX] in {
1699 defm VCVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
1700 ssmem, sse_load_f32, "cvtss2si",
1701 SSE_CVT_SS2SI_32>, XS, VEX, VEX_LIG;
1702 defm VCVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
1703 ssmem, sse_load_f32, "cvtss2si",
1704 SSE_CVT_SS2SI_64>, XS, VEX, VEX_W, VEX_LIG;
1706 defm CVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
1707 ssmem, sse_load_f32, "cvtss2si",
1708 SSE_CVT_SS2SI_32>, XS;
1709 defm CVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
1710 ssmem, sse_load_f32, "cvtss2si",
1711 SSE_CVT_SS2SI_64>, XS, REX_W;
1713 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, i128mem,
1714 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1715 SSEPackedSingle, SSE_CVT_PS>,
1716 TB, VEX, Requires<[HasAVX]>;
1717 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, i256mem,
1718 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1719 SSEPackedSingle, SSE_CVT_PS>,
1720 TB, VEX, VEX_L, Requires<[HasAVX]>;
1722 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, i128mem,
1723 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1724 SSEPackedSingle, SSE_CVT_PS>,
1725 TB, Requires<[UseSSE2]>;
1727 let Predicates = [UseAVX] in {
1728 def : InstAlias<"vcvtss2si{l}\t{$src, $dst|$dst, $src}",
1729 (VCVTSS2SIrr GR32:$dst, VR128:$src), 0>;
1730 def : InstAlias<"vcvtss2si{l}\t{$src, $dst|$dst, $src}",
1731 (VCVTSS2SIrm GR32:$dst, ssmem:$src), 0>;
1732 def : InstAlias<"vcvtsd2si{l}\t{$src, $dst|$dst, $src}",
1733 (VCVTSD2SIrr GR32:$dst, VR128:$src), 0>;
1734 def : InstAlias<"vcvtsd2si{l}\t{$src, $dst|$dst, $src}",
1735 (VCVTSD2SIrm GR32:$dst, sdmem:$src), 0>;
1736 def : InstAlias<"vcvtss2si{q}\t{$src, $dst|$dst, $src}",
1737 (VCVTSS2SI64rr GR64:$dst, VR128:$src), 0>;
1738 def : InstAlias<"vcvtss2si{q}\t{$src, $dst|$dst, $src}",
1739 (VCVTSS2SI64rm GR64:$dst, ssmem:$src), 0>;
1740 def : InstAlias<"vcvtsd2si{q}\t{$src, $dst|$dst, $src}",
1741 (VCVTSD2SI64rr GR64:$dst, VR128:$src), 0>;
1742 def : InstAlias<"vcvtsd2si{q}\t{$src, $dst|$dst, $src}",
1743 (VCVTSD2SI64rm GR64:$dst, sdmem:$src), 0>;
1746 def : InstAlias<"cvtss2si{l}\t{$src, $dst|$dst, $src}",
1747 (CVTSS2SIrr GR32:$dst, VR128:$src), 0>;
1748 def : InstAlias<"cvtss2si{l}\t{$src, $dst|$dst, $src}",
1749 (CVTSS2SIrm GR32:$dst, ssmem:$src), 0>;
1750 def : InstAlias<"cvtsd2si{l}\t{$src, $dst|$dst, $src}",
1751 (CVTSD2SIrr GR32:$dst, VR128:$src), 0>;
1752 def : InstAlias<"cvtsd2si{l}\t{$src, $dst|$dst, $src}",
1753 (CVTSD2SIrm GR32:$dst, sdmem:$src), 0>;
1754 def : InstAlias<"cvtss2si{q}\t{$src, $dst|$dst, $src}",
1755 (CVTSS2SI64rr GR64:$dst, VR128:$src), 0>;
1756 def : InstAlias<"cvtss2si{q}\t{$src, $dst|$dst, $src}",
1757 (CVTSS2SI64rm GR64:$dst, ssmem:$src), 0>;
1758 def : InstAlias<"cvtsd2si{q}\t{$src, $dst|$dst, $src}",
1759 (CVTSD2SI64rr GR64:$dst, VR128:$src), 0>;
1760 def : InstAlias<"cvtsd2si{q}\t{$src, $dst|$dst, $src}",
1761 (CVTSD2SI64rm GR64:$dst, sdmem:$src)>;
1765 // Convert scalar double to scalar single
1766 let neverHasSideEffects = 1, Predicates = [UseAVX] in {
1767 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
1768 (ins FR64:$src1, FR64:$src2),
1769 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
1770 IIC_SSE_CVT_Scalar_RR>, VEX_4V, VEX_LIG,
1771 Sched<[WriteCvtF2F]>;
1773 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
1774 (ins FR64:$src1, f64mem:$src2),
1775 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1776 [], IIC_SSE_CVT_Scalar_RM>,
1777 XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG,
1778 Sched<[WriteCvtF2FLd, ReadAfterLd]>;
1781 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
1784 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
1785 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1786 [(set FR32:$dst, (fround FR64:$src))],
1787 IIC_SSE_CVT_Scalar_RR>, Sched<[WriteCvtF2F]>;
1788 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
1789 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1790 [(set FR32:$dst, (fround (loadf64 addr:$src)))],
1791 IIC_SSE_CVT_Scalar_RM>,
1793 Requires<[UseSSE2, OptForSize]>, Sched<[WriteCvtF2FLd]>;
1795 def Int_VCVTSD2SSrr: I<0x5A, MRMSrcReg,
1796 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1797 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1799 (int_x86_sse2_cvtsd2ss VR128:$src1, VR128:$src2))],
1800 IIC_SSE_CVT_Scalar_RR>, XD, VEX_4V, Requires<[UseAVX]>,
1801 Sched<[WriteCvtF2F]>;
1802 def Int_VCVTSD2SSrm: I<0x5A, MRMSrcReg,
1803 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2),
1804 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1805 [(set VR128:$dst, (int_x86_sse2_cvtsd2ss
1806 VR128:$src1, sse_load_f64:$src2))],
1807 IIC_SSE_CVT_Scalar_RM>, XD, VEX_4V, Requires<[UseAVX]>,
1808 Sched<[WriteCvtF2FLd, ReadAfterLd]>;
1810 let Constraints = "$src1 = $dst" in {
1811 def Int_CVTSD2SSrr: I<0x5A, MRMSrcReg,
1812 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1813 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1815 (int_x86_sse2_cvtsd2ss VR128:$src1, VR128:$src2))],
1816 IIC_SSE_CVT_Scalar_RR>, XD, Requires<[UseSSE2]>,
1817 Sched<[WriteCvtF2F]>;
1818 def Int_CVTSD2SSrm: I<0x5A, MRMSrcReg,
1819 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2),
1820 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1821 [(set VR128:$dst, (int_x86_sse2_cvtsd2ss
1822 VR128:$src1, sse_load_f64:$src2))],
1823 IIC_SSE_CVT_Scalar_RM>, XD, Requires<[UseSSE2]>,
1824 Sched<[WriteCvtF2FLd, ReadAfterLd]>;
1827 // Convert scalar single to scalar double
1828 // SSE2 instructions with XS prefix
1829 let neverHasSideEffects = 1, Predicates = [UseAVX] in {
1830 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
1831 (ins FR32:$src1, FR32:$src2),
1832 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1833 [], IIC_SSE_CVT_Scalar_RR>,
1834 XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG,
1835 Sched<[WriteCvtF2F]>;
1837 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
1838 (ins FR32:$src1, f32mem:$src2),
1839 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1840 [], IIC_SSE_CVT_Scalar_RM>,
1841 XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>,
1842 Sched<[WriteCvtF2FLd, ReadAfterLd]>;
1845 def : Pat<(f64 (fextend FR32:$src)),
1846 (VCVTSS2SDrr FR32:$src, FR32:$src)>, Requires<[UseAVX]>;
1847 def : Pat<(fextend (loadf32 addr:$src)),
1848 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>, Requires<[UseAVX]>;
1850 def : Pat<(extloadf32 addr:$src),
1851 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>,
1852 Requires<[UseAVX, OptForSize]>;
1853 def : Pat<(extloadf32 addr:$src),
1854 (VCVTSS2SDrr (f32 (IMPLICIT_DEF)), (VMOVSSrm addr:$src))>,
1855 Requires<[UseAVX, OptForSpeed]>;
1857 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
1858 "cvtss2sd\t{$src, $dst|$dst, $src}",
1859 [(set FR64:$dst, (fextend FR32:$src))],
1860 IIC_SSE_CVT_Scalar_RR>, XS,
1861 Requires<[UseSSE2]>, Sched<[WriteCvtF2F]>;
1862 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
1863 "cvtss2sd\t{$src, $dst|$dst, $src}",
1864 [(set FR64:$dst, (extloadf32 addr:$src))],
1865 IIC_SSE_CVT_Scalar_RM>, XS,
1866 Requires<[UseSSE2, OptForSize]>, Sched<[WriteCvtF2FLd]>;
1868 // extload f32 -> f64. This matches load+fextend because we have a hack in
1869 // the isel (PreprocessForFPConvert) that can introduce loads after dag
1871 // Since these loads aren't folded into the fextend, we have to match it
1873 def : Pat<(fextend (loadf32 addr:$src)),
1874 (CVTSS2SDrm addr:$src)>, Requires<[UseSSE2]>;
1875 def : Pat<(extloadf32 addr:$src),
1876 (CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[UseSSE2, OptForSpeed]>;
1878 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
1879 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1880 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1882 (int_x86_sse2_cvtss2sd VR128:$src1, VR128:$src2))],
1883 IIC_SSE_CVT_Scalar_RR>, XS, VEX_4V, Requires<[UseAVX]>,
1884 Sched<[WriteCvtF2F]>;
1885 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
1886 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2),
1887 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1889 (int_x86_sse2_cvtss2sd VR128:$src1, sse_load_f32:$src2))],
1890 IIC_SSE_CVT_Scalar_RM>, XS, VEX_4V, Requires<[UseAVX]>,
1891 Sched<[WriteCvtF2FLd, ReadAfterLd]>;
1892 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
1893 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
1894 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1895 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1897 (int_x86_sse2_cvtss2sd VR128:$src1, VR128:$src2))],
1898 IIC_SSE_CVT_Scalar_RR>, XS, Requires<[UseSSE2]>,
1899 Sched<[WriteCvtF2F]>;
1900 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
1901 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2),
1902 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1904 (int_x86_sse2_cvtss2sd VR128:$src1, sse_load_f32:$src2))],
1905 IIC_SSE_CVT_Scalar_RM>, XS, Requires<[UseSSE2]>,
1906 Sched<[WriteCvtF2FLd, ReadAfterLd]>;
1909 // Convert packed single/double fp to doubleword
1910 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1911 "cvtps2dq\t{$src, $dst|$dst, $src}",
1912 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))],
1913 IIC_SSE_CVT_PS_RR>, VEX, Sched<[WriteCvtF2I]>;
1914 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1915 "cvtps2dq\t{$src, $dst|$dst, $src}",
1917 (int_x86_sse2_cvtps2dq (memopv4f32 addr:$src)))],
1918 IIC_SSE_CVT_PS_RM>, VEX, Sched<[WriteCvtF2ILd]>;
1919 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1920 "cvtps2dq\t{$src, $dst|$dst, $src}",
1922 (int_x86_avx_cvt_ps2dq_256 VR256:$src))],
1923 IIC_SSE_CVT_PS_RR>, VEX, VEX_L, Sched<[WriteCvtF2I]>;
1924 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1925 "cvtps2dq\t{$src, $dst|$dst, $src}",
1927 (int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)))],
1928 IIC_SSE_CVT_PS_RM>, VEX, VEX_L, Sched<[WriteCvtF2ILd]>;
1929 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1930 "cvtps2dq\t{$src, $dst|$dst, $src}",
1931 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))],
1932 IIC_SSE_CVT_PS_RR>, Sched<[WriteCvtF2I]>;
1933 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1934 "cvtps2dq\t{$src, $dst|$dst, $src}",
1936 (int_x86_sse2_cvtps2dq (memopv4f32 addr:$src)))],
1937 IIC_SSE_CVT_PS_RM>, Sched<[WriteCvtF2ILd]>;
1940 // Convert Packed Double FP to Packed DW Integers
1941 let Predicates = [HasAVX] in {
1942 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1943 // register, but the same isn't true when using memory operands instead.
1944 // Provide other assembly rr and rm forms to address this explicitly.
1945 def VCVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1946 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1947 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1948 VEX, Sched<[WriteCvtF2I]>;
1951 def : InstAlias<"vcvtpd2dqx\t{$src, $dst|$dst, $src}",
1952 (VCVTPD2DQrr VR128:$dst, VR128:$src)>;
1953 def VCVTPD2DQXrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1954 "vcvtpd2dqx\t{$src, $dst|$dst, $src}",
1956 (int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)))]>, VEX,
1957 Sched<[WriteCvtF2ILd]>;
1960 def VCVTPD2DQYrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1961 "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}",
1963 (int_x86_avx_cvt_pd2dq_256 VR256:$src))]>, VEX, VEX_L,
1964 Sched<[WriteCvtF2I]>;
1965 def VCVTPD2DQYrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1966 "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}",
1968 (int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)))]>,
1969 VEX, VEX_L, Sched<[WriteCvtF2ILd]>;
1970 def : InstAlias<"vcvtpd2dq\t{$src, $dst|$dst, $src}",
1971 (VCVTPD2DQYrr VR128:$dst, VR256:$src)>;
1974 def CVTPD2DQrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1975 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1977 (int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)))],
1978 IIC_SSE_CVT_PD_RM>, Sched<[WriteCvtF2ILd]>;
1979 def CVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1980 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1981 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))],
1982 IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtF2I]>;
1984 // Convert with truncation packed single/double fp to doubleword
1985 // SSE2 packed instructions with XS prefix
1986 def VCVTTPS2DQrr : VS2SI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1987 "cvttps2dq\t{$src, $dst|$dst, $src}",
1989 (int_x86_sse2_cvttps2dq VR128:$src))],
1990 IIC_SSE_CVT_PS_RR>, VEX, Sched<[WriteCvtF2I]>;
1991 def VCVTTPS2DQrm : VS2SI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1992 "cvttps2dq\t{$src, $dst|$dst, $src}",
1993 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
1994 (memopv4f32 addr:$src)))],
1995 IIC_SSE_CVT_PS_RM>, VEX, Sched<[WriteCvtF2ILd]>;
1996 def VCVTTPS2DQYrr : VS2SI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1997 "cvttps2dq\t{$src, $dst|$dst, $src}",
1999 (int_x86_avx_cvtt_ps2dq_256 VR256:$src))],
2000 IIC_SSE_CVT_PS_RR>, VEX, VEX_L, Sched<[WriteCvtF2I]>;
2001 def VCVTTPS2DQYrm : VS2SI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2002 "cvttps2dq\t{$src, $dst|$dst, $src}",
2003 [(set VR256:$dst, (int_x86_avx_cvtt_ps2dq_256
2004 (memopv8f32 addr:$src)))],
2005 IIC_SSE_CVT_PS_RM>, VEX, VEX_L,
2006 Sched<[WriteCvtF2ILd]>;
2008 def CVTTPS2DQrr : S2SI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2009 "cvttps2dq\t{$src, $dst|$dst, $src}",
2010 [(set VR128:$dst, (int_x86_sse2_cvttps2dq VR128:$src))],
2011 IIC_SSE_CVT_PS_RR>, Sched<[WriteCvtF2I]>;
2012 def CVTTPS2DQrm : S2SI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2013 "cvttps2dq\t{$src, $dst|$dst, $src}",
2015 (int_x86_sse2_cvttps2dq (memopv4f32 addr:$src)))],
2016 IIC_SSE_CVT_PS_RM>, Sched<[WriteCvtF2ILd]>;
2018 let Predicates = [HasAVX] in {
2019 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
2020 (VCVTDQ2PSrr VR128:$src)>;
2021 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
2022 (VCVTDQ2PSrm addr:$src)>;
2024 def : Pat<(int_x86_sse2_cvtdq2ps VR128:$src),
2025 (VCVTDQ2PSrr VR128:$src)>;
2026 def : Pat<(int_x86_sse2_cvtdq2ps (bc_v4i32 (memopv2i64 addr:$src))),
2027 (VCVTDQ2PSrm addr:$src)>;
2029 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
2030 (VCVTTPS2DQrr VR128:$src)>;
2031 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
2032 (VCVTTPS2DQrm addr:$src)>;
2034 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
2035 (VCVTDQ2PSYrr VR256:$src)>;
2036 def : Pat<(v8f32 (sint_to_fp (bc_v8i32 (memopv4i64 addr:$src)))),
2037 (VCVTDQ2PSYrm addr:$src)>;
2039 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
2040 (VCVTTPS2DQYrr VR256:$src)>;
2041 def : Pat<(v8i32 (fp_to_sint (memopv8f32 addr:$src))),
2042 (VCVTTPS2DQYrm addr:$src)>;
2045 let Predicates = [UseSSE2] in {
2046 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
2047 (CVTDQ2PSrr VR128:$src)>;
2048 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
2049 (CVTDQ2PSrm addr:$src)>;
2051 def : Pat<(int_x86_sse2_cvtdq2ps VR128:$src),
2052 (CVTDQ2PSrr VR128:$src)>;
2053 def : Pat<(int_x86_sse2_cvtdq2ps (bc_v4i32 (memopv2i64 addr:$src))),
2054 (CVTDQ2PSrm addr:$src)>;
2056 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
2057 (CVTTPS2DQrr VR128:$src)>;
2058 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
2059 (CVTTPS2DQrm addr:$src)>;
2062 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2063 "cvttpd2dq\t{$src, $dst|$dst, $src}",
2065 (int_x86_sse2_cvttpd2dq VR128:$src))],
2066 IIC_SSE_CVT_PD_RR>, VEX, Sched<[WriteCvtF2I]>;
2068 // The assembler can recognize rr 256-bit instructions by seeing a ymm
2069 // register, but the same isn't true when using memory operands instead.
2070 // Provide other assembly rr and rm forms to address this explicitly.
2073 def : InstAlias<"vcvttpd2dqx\t{$src, $dst|$dst, $src}",
2074 (VCVTTPD2DQrr VR128:$dst, VR128:$src)>;
2075 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2076 "cvttpd2dqx\t{$src, $dst|$dst, $src}",
2077 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
2078 (memopv2f64 addr:$src)))],
2079 IIC_SSE_CVT_PD_RM>, VEX, Sched<[WriteCvtF2ILd]>;
2082 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
2083 "cvttpd2dq{y}\t{$src, $dst|$dst, $src}",
2085 (int_x86_avx_cvtt_pd2dq_256 VR256:$src))],
2086 IIC_SSE_CVT_PD_RR>, VEX, VEX_L, Sched<[WriteCvtF2I]>;
2087 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
2088 "cvttpd2dq{y}\t{$src, $dst|$dst, $src}",
2090 (int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)))],
2091 IIC_SSE_CVT_PD_RM>, VEX, VEX_L, Sched<[WriteCvtF2ILd]>;
2092 def : InstAlias<"vcvttpd2dq\t{$src, $dst|$dst, $src}",
2093 (VCVTTPD2DQYrr VR128:$dst, VR256:$src)>;
2095 let Predicates = [HasAVX] in {
2096 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
2097 (VCVTTPD2DQYrr VR256:$src)>;
2098 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
2099 (VCVTTPD2DQYrm addr:$src)>;
2100 } // Predicates = [HasAVX]
2102 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2103 "cvttpd2dq\t{$src, $dst|$dst, $src}",
2104 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))],
2105 IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtF2I]>;
2106 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
2107 "cvttpd2dq\t{$src, $dst|$dst, $src}",
2108 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
2109 (memopv2f64 addr:$src)))],
2111 Sched<[WriteCvtF2ILd]>;
2113 // Convert packed single to packed double
2114 let Predicates = [HasAVX] in {
2115 // SSE2 instructions without OpSize prefix
2116 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2117 "vcvtps2pd\t{$src, $dst|$dst, $src}",
2118 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))],
2119 IIC_SSE_CVT_PD_RR>, TB, VEX, Sched<[WriteCvtF2F]>;
2120 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
2121 "vcvtps2pd\t{$src, $dst|$dst, $src}",
2122 [(set VR128:$dst, (v2f64 (extloadv2f32 addr:$src)))],
2123 IIC_SSE_CVT_PD_RM>, TB, VEX, Sched<[WriteCvtF2FLd]>;
2124 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
2125 "vcvtps2pd\t{$src, $dst|$dst, $src}",
2127 (int_x86_avx_cvt_ps2_pd_256 VR128:$src))],
2128 IIC_SSE_CVT_PD_RR>, TB, VEX, VEX_L, Sched<[WriteCvtF2F]>;
2129 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
2130 "vcvtps2pd\t{$src, $dst|$dst, $src}",
2132 (int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)))],
2133 IIC_SSE_CVT_PD_RM>, TB, VEX, VEX_L, Sched<[WriteCvtF2FLd]>;
2136 let Predicates = [UseSSE2] in {
2137 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2138 "cvtps2pd\t{$src, $dst|$dst, $src}",
2139 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))],
2140 IIC_SSE_CVT_PD_RR>, TB, Sched<[WriteCvtF2F]>;
2141 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
2142 "cvtps2pd\t{$src, $dst|$dst, $src}",
2143 [(set VR128:$dst, (v2f64 (extloadv2f32 addr:$src)))],
2144 IIC_SSE_CVT_PD_RM>, TB, Sched<[WriteCvtF2FLd]>;
2147 // Convert Packed DW Integers to Packed Double FP
2148 let Predicates = [HasAVX] in {
2149 let neverHasSideEffects = 1, mayLoad = 1 in
2150 def VCVTDQ2PDrm : S2SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2151 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2152 []>, VEX, Sched<[WriteCvtI2FLd]>;
2153 def VCVTDQ2PDrr : S2SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2154 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2156 (int_x86_sse2_cvtdq2pd VR128:$src))]>, VEX,
2157 Sched<[WriteCvtI2F]>;
2158 def VCVTDQ2PDYrm : S2SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
2159 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2161 (int_x86_avx_cvtdq2_pd_256
2162 (bitconvert (memopv2i64 addr:$src))))]>, VEX, VEX_L,
2163 Sched<[WriteCvtI2FLd]>;
2164 def VCVTDQ2PDYrr : S2SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
2165 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2167 (int_x86_avx_cvtdq2_pd_256 VR128:$src))]>, VEX, VEX_L,
2168 Sched<[WriteCvtI2F]>;
2171 let neverHasSideEffects = 1, mayLoad = 1 in
2172 def CVTDQ2PDrm : S2SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2173 "cvtdq2pd\t{$src, $dst|$dst, $src}", [],
2174 IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtI2FLd]>;
2175 def CVTDQ2PDrr : S2SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2176 "cvtdq2pd\t{$src, $dst|$dst, $src}",
2177 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))],
2178 IIC_SSE_CVT_PD_RM>, Sched<[WriteCvtI2F]>;
2180 // AVX 256-bit register conversion intrinsics
2181 let Predicates = [HasAVX] in {
2182 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
2183 (VCVTDQ2PDYrr VR128:$src)>;
2184 def : Pat<(v4f64 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
2185 (VCVTDQ2PDYrm addr:$src)>;
2186 } // Predicates = [HasAVX]
2188 // Convert packed double to packed single
2189 // The assembler can recognize rr 256-bit instructions by seeing a ymm
2190 // register, but the same isn't true when using memory operands instead.
2191 // Provide other assembly rr and rm forms to address this explicitly.
2192 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2193 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2194 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))],
2195 IIC_SSE_CVT_PD_RR>, VEX, Sched<[WriteCvtF2F]>;
2198 def : InstAlias<"vcvtpd2psx\t{$src, $dst|$dst, $src}",
2199 (VCVTPD2PSrr VR128:$dst, VR128:$src)>;
2200 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2201 "cvtpd2psx\t{$src, $dst|$dst, $src}",
2203 (int_x86_sse2_cvtpd2ps (memopv2f64 addr:$src)))],
2204 IIC_SSE_CVT_PD_RM>, VEX, Sched<[WriteCvtF2FLd]>;
2207 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
2208 "cvtpd2ps{y}\t{$src, $dst|$dst, $src}",
2210 (int_x86_avx_cvt_pd2_ps_256 VR256:$src))],
2211 IIC_SSE_CVT_PD_RR>, VEX, VEX_L, Sched<[WriteCvtF2F]>;
2212 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
2213 "cvtpd2ps{y}\t{$src, $dst|$dst, $src}",
2215 (int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)))],
2216 IIC_SSE_CVT_PD_RM>, VEX, VEX_L, Sched<[WriteCvtF2FLd]>;
2217 def : InstAlias<"vcvtpd2ps\t{$src, $dst|$dst, $src}",
2218 (VCVTPD2PSYrr VR128:$dst, VR256:$src)>;
2220 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2221 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2222 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))],
2223 IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtF2F]>;
2224 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2225 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2227 (int_x86_sse2_cvtpd2ps (memopv2f64 addr:$src)))],
2228 IIC_SSE_CVT_PD_RM>, Sched<[WriteCvtF2FLd]>;
2231 // AVX 256-bit register conversion intrinsics
2232 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
2233 // whenever possible to avoid declaring two versions of each one.
2234 let Predicates = [HasAVX] in {
2235 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
2236 (VCVTDQ2PSYrr VR256:$src)>;
2237 def : Pat<(int_x86_avx_cvtdq2_ps_256 (bitconvert (memopv4i64 addr:$src))),
2238 (VCVTDQ2PSYrm addr:$src)>;
2240 // Match fround and fextend for 128/256-bit conversions
2241 def : Pat<(v4f32 (X86vfpround (v2f64 VR128:$src))),
2242 (VCVTPD2PSrr VR128:$src)>;
2243 def : Pat<(v4f32 (X86vfpround (memopv2f64 addr:$src))),
2244 (VCVTPD2PSXrm addr:$src)>;
2245 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
2246 (VCVTPD2PSYrr VR256:$src)>;
2247 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
2248 (VCVTPD2PSYrm addr:$src)>;
2250 def : Pat<(v2f64 (X86vfpext (v4f32 VR128:$src))),
2251 (VCVTPS2PDrr VR128:$src)>;
2252 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
2253 (VCVTPS2PDYrr VR128:$src)>;
2254 def : Pat<(v4f64 (extloadv4f32 addr:$src)),
2255 (VCVTPS2PDYrm addr:$src)>;
2258 let Predicates = [UseSSE2] in {
2259 // Match fround and fextend for 128 conversions
2260 def : Pat<(v4f32 (X86vfpround (v2f64 VR128:$src))),
2261 (CVTPD2PSrr VR128:$src)>;
2262 def : Pat<(v4f32 (X86vfpround (memopv2f64 addr:$src))),
2263 (CVTPD2PSrm addr:$src)>;
2265 def : Pat<(v2f64 (X86vfpext (v4f32 VR128:$src))),
2266 (CVTPS2PDrr VR128:$src)>;
2269 //===----------------------------------------------------------------------===//
2270 // SSE 1 & 2 - Compare Instructions
2271 //===----------------------------------------------------------------------===//
2273 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
2274 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
2275 Operand CC, SDNode OpNode, ValueType VT,
2276 PatFrag ld_frag, string asm, string asm_alt,
2278 def rr : SIi8<0xC2, MRMSrcReg,
2279 (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
2280 [(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, imm:$cc))],
2281 itins.rr>, Sched<[itins.Sched]>;
2282 def rm : SIi8<0xC2, MRMSrcMem,
2283 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
2284 [(set RC:$dst, (OpNode (VT RC:$src1),
2285 (ld_frag addr:$src2), imm:$cc))],
2287 Sched<[itins.Sched.Folded, ReadAfterLd]>;
2289 // Accept explicit immediate argument form instead of comparison code.
2290 let neverHasSideEffects = 1 in {
2291 def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst),
2292 (ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, [],
2293 IIC_SSE_ALU_F32S_RR>, Sched<[itins.Sched]>;
2295 def rm_alt : SIi8<0xC2, MRMSrcMem, (outs RC:$dst),
2296 (ins RC:$src1, x86memop:$src2, i8imm:$cc), asm_alt, [],
2297 IIC_SSE_ALU_F32S_RM>,
2298 Sched<[itins.Sched.Folded, ReadAfterLd]>;
2302 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem, AVXCC, X86cmpss, f32, loadf32,
2303 "cmp${cc}ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2304 "cmpss\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2306 XS, VEX_4V, VEX_LIG;
2307 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem, AVXCC, X86cmpsd, f64, loadf64,
2308 "cmp${cc}sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2309 "cmpsd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2310 SSE_ALU_F32S>, // same latency as 32 bit compare
2311 XD, VEX_4V, VEX_LIG;
2313 let Constraints = "$src1 = $dst" in {
2314 defm CMPSS : sse12_cmp_scalar<FR32, f32mem, SSECC, X86cmpss, f32, loadf32,
2315 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
2316 "cmpss\t{$cc, $src2, $dst|$dst, $src2, $cc}", SSE_ALU_F32S>,
2318 defm CMPSD : sse12_cmp_scalar<FR64, f64mem, SSECC, X86cmpsd, f64, loadf64,
2319 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
2320 "cmpsd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2325 multiclass sse12_cmp_scalar_int<X86MemOperand x86memop, Operand CC,
2326 Intrinsic Int, string asm, OpndItins itins> {
2327 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
2328 (ins VR128:$src1, VR128:$src, CC:$cc), asm,
2329 [(set VR128:$dst, (Int VR128:$src1,
2330 VR128:$src, imm:$cc))],
2332 Sched<[itins.Sched]>;
2333 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
2334 (ins VR128:$src1, x86memop:$src, CC:$cc), asm,
2335 [(set VR128:$dst, (Int VR128:$src1,
2336 (load addr:$src), imm:$cc))],
2338 Sched<[itins.Sched.Folded, ReadAfterLd]>;
2341 // Aliases to match intrinsics which expect XMM operand(s).
2342 defm Int_VCMPSS : sse12_cmp_scalar_int<f32mem, AVXCC, int_x86_sse_cmp_ss,
2343 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
2346 defm Int_VCMPSD : sse12_cmp_scalar_int<f64mem, AVXCC, int_x86_sse2_cmp_sd,
2347 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
2348 SSE_ALU_F32S>, // same latency as f32
2350 let Constraints = "$src1 = $dst" in {
2351 defm Int_CMPSS : sse12_cmp_scalar_int<f32mem, SSECC, int_x86_sse_cmp_ss,
2352 "cmp${cc}ss\t{$src, $dst|$dst, $src}",
2354 defm Int_CMPSD : sse12_cmp_scalar_int<f64mem, SSECC, int_x86_sse2_cmp_sd,
2355 "cmp${cc}sd\t{$src, $dst|$dst, $src}",
2361 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
2362 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
2363 ValueType vt, X86MemOperand x86memop,
2364 PatFrag ld_frag, string OpcodeStr> {
2365 def rr: SI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
2366 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2367 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))],
2370 def rm: SI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
2371 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2372 [(set EFLAGS, (OpNode (vt RC:$src1),
2373 (ld_frag addr:$src2)))],
2375 Sched<[WriteFAddLd, ReadAfterLd]>;
2378 let Defs = [EFLAGS] in {
2379 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2380 "ucomiss">, TB, VEX, VEX_LIG;
2381 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2382 "ucomisd">, TB, OpSize, VEX, VEX_LIG;
2383 let Pattern = []<dag> in {
2384 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2385 "comiss">, TB, VEX, VEX_LIG;
2386 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2387 "comisd">, TB, OpSize, VEX, VEX_LIG;
2390 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2391 load, "ucomiss">, TB, VEX;
2392 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2393 load, "ucomisd">, TB, OpSize, VEX;
2395 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
2396 load, "comiss">, TB, VEX;
2397 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
2398 load, "comisd">, TB, OpSize, VEX;
2399 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2401 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2402 "ucomisd">, TB, OpSize;
2404 let Pattern = []<dag> in {
2405 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2407 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2408 "comisd">, TB, OpSize;
2411 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2412 load, "ucomiss">, TB;
2413 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2414 load, "ucomisd">, TB, OpSize;
2416 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
2418 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
2419 "comisd">, TB, OpSize;
2420 } // Defs = [EFLAGS]
2422 // sse12_cmp_packed - sse 1 & 2 compare packed instructions
2423 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
2424 Operand CC, Intrinsic Int, string asm,
2425 string asm_alt, Domain d,
2426 OpndItins itins = SSE_ALU_F32P> {
2427 def rri : PIi8<0xC2, MRMSrcReg,
2428 (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
2429 [(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))],
2432 def rmi : PIi8<0xC2, MRMSrcMem,
2433 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
2434 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))],
2436 Sched<[WriteFAddLd, ReadAfterLd]>;
2438 // Accept explicit immediate argument form instead of comparison code.
2439 let neverHasSideEffects = 1 in {
2440 def rri_alt : PIi8<0xC2, MRMSrcReg,
2441 (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
2442 asm_alt, [], itins.rr, d>, Sched<[WriteFAdd]>;
2443 def rmi_alt : PIi8<0xC2, MRMSrcMem,
2444 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc),
2445 asm_alt, [], itins.rm, d>,
2446 Sched<[WriteFAddLd, ReadAfterLd]>;
2450 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse_cmp_ps,
2451 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2452 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2453 SSEPackedSingle>, TB, VEX_4V;
2454 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse2_cmp_pd,
2455 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2456 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2457 SSEPackedDouble>, TB, OpSize, VEX_4V;
2458 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_ps_256,
2459 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2460 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2461 SSEPackedSingle>, TB, VEX_4V, VEX_L;
2462 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_pd_256,
2463 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2464 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2465 SSEPackedDouble>, TB, OpSize, VEX_4V, VEX_L;
2466 let Constraints = "$src1 = $dst" in {
2467 defm CMPPS : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse_cmp_ps,
2468 "cmp${cc}ps\t{$src2, $dst|$dst, $src2}",
2469 "cmpps\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2470 SSEPackedSingle, SSE_ALU_F32P>, TB;
2471 defm CMPPD : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse2_cmp_pd,
2472 "cmp${cc}pd\t{$src2, $dst|$dst, $src2}",
2473 "cmppd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2474 SSEPackedDouble, SSE_ALU_F64P>, TB, OpSize;
2477 let Predicates = [HasAVX] in {
2478 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2479 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2480 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2481 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2482 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2483 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2484 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2485 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2487 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
2488 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
2489 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
2490 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
2491 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
2492 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
2493 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
2494 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
2497 let Predicates = [UseSSE1] in {
2498 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2499 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2500 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2501 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2504 let Predicates = [UseSSE2] in {
2505 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2506 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2507 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2508 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2511 //===----------------------------------------------------------------------===//
2512 // SSE 1 & 2 - Shuffle Instructions
2513 //===----------------------------------------------------------------------===//
2515 /// sse12_shuffle - sse 1 & 2 shuffle instructions
2516 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
2517 ValueType vt, string asm, PatFrag mem_frag,
2518 Domain d, bit IsConvertibleToThreeAddress = 0> {
2519 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
2520 (ins RC:$src1, x86memop:$src2, i8imm:$src3), asm,
2521 [(set RC:$dst, (vt (X86Shufp RC:$src1, (mem_frag addr:$src2),
2522 (i8 imm:$src3))))], IIC_SSE_SHUFP, d>,
2523 Sched<[WriteShuffleLd, ReadAfterLd]>;
2524 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
2525 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
2526 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
2527 [(set RC:$dst, (vt (X86Shufp RC:$src1, RC:$src2,
2528 (i8 imm:$src3))))], IIC_SSE_SHUFP, d>,
2529 Sched<[WriteShuffle]>;
2532 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2533 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2534 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
2535 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
2536 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2537 memopv8f32, SSEPackedSingle>, TB, VEX_4V, VEX_L;
2538 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2539 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2540 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2541 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
2542 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2543 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V, VEX_L;
2545 let Constraints = "$src1 = $dst" in {
2546 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2547 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2548 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
2550 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2551 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2552 memopv2f64, SSEPackedDouble, 1 /* cvt to pshufd */>,
2556 let Predicates = [HasAVX] in {
2557 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2558 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2559 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2560 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2561 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2563 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2564 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2565 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2566 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2567 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2570 def : Pat<(v8i32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2571 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2572 def : Pat<(v8i32 (X86Shufp VR256:$src1,
2573 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
2574 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2576 def : Pat<(v4i64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2577 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2578 def : Pat<(v4i64 (X86Shufp VR256:$src1,
2579 (memopv4i64 addr:$src2), (i8 imm:$imm))),
2580 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2583 let Predicates = [UseSSE1] in {
2584 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2585 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2586 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2587 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2588 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2591 let Predicates = [UseSSE2] in {
2592 // Generic SHUFPD patterns
2593 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2594 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2595 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2596 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2597 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2600 //===----------------------------------------------------------------------===//
2601 // SSE 1 & 2 - Unpack Instructions
2602 //===----------------------------------------------------------------------===//
2604 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
2605 multiclass sse12_unpack_interleave<bits<8> opc, SDNode OpNode, ValueType vt,
2606 PatFrag mem_frag, RegisterClass RC,
2607 X86MemOperand x86memop, string asm,
2609 def rr : PI<opc, MRMSrcReg,
2610 (outs RC:$dst), (ins RC:$src1, RC:$src2),
2612 (vt (OpNode RC:$src1, RC:$src2)))],
2613 IIC_SSE_UNPCK, d>, Sched<[WriteShuffle]>;
2614 def rm : PI<opc, MRMSrcMem,
2615 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
2617 (vt (OpNode RC:$src1,
2618 (mem_frag addr:$src2))))],
2620 Sched<[WriteShuffleLd, ReadAfterLd]>;
2623 defm VUNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
2624 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2625 SSEPackedSingle>, TB, VEX_4V;
2626 defm VUNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
2627 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2628 SSEPackedDouble>, TB, OpSize, VEX_4V;
2629 defm VUNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
2630 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2631 SSEPackedSingle>, TB, VEX_4V;
2632 defm VUNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
2633 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2634 SSEPackedDouble>, TB, OpSize, VEX_4V;
2636 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, X86Unpckh, v8f32, memopv8f32,
2637 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2638 SSEPackedSingle>, TB, VEX_4V, VEX_L;
2639 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, X86Unpckh, v4f64, memopv4f64,
2640 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2641 SSEPackedDouble>, TB, OpSize, VEX_4V, VEX_L;
2642 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, X86Unpckl, v8f32, memopv8f32,
2643 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2644 SSEPackedSingle>, TB, VEX_4V, VEX_L;
2645 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, X86Unpckl, v4f64, memopv4f64,
2646 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2647 SSEPackedDouble>, TB, OpSize, VEX_4V, VEX_L;
2649 let Constraints = "$src1 = $dst" in {
2650 defm UNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
2651 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
2652 SSEPackedSingle>, TB;
2653 defm UNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
2654 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
2655 SSEPackedDouble>, TB, OpSize;
2656 defm UNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
2657 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
2658 SSEPackedSingle>, TB;
2659 defm UNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
2660 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
2661 SSEPackedDouble>, TB, OpSize;
2662 } // Constraints = "$src1 = $dst"
2664 let Predicates = [HasAVX1Only] in {
2665 def : Pat<(v8i32 (X86Unpckl VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
2666 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
2667 def : Pat<(v8i32 (X86Unpckl VR256:$src1, VR256:$src2)),
2668 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
2669 def : Pat<(v8i32 (X86Unpckh VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
2670 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
2671 def : Pat<(v8i32 (X86Unpckh VR256:$src1, VR256:$src2)),
2672 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
2674 def : Pat<(v4i64 (X86Unpckl VR256:$src1, (memopv4i64 addr:$src2))),
2675 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
2676 def : Pat<(v4i64 (X86Unpckl VR256:$src1, VR256:$src2)),
2677 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
2678 def : Pat<(v4i64 (X86Unpckh VR256:$src1, (memopv4i64 addr:$src2))),
2679 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
2680 def : Pat<(v4i64 (X86Unpckh VR256:$src1, VR256:$src2)),
2681 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
2684 let Predicates = [HasAVX] in {
2685 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2686 // problem is during lowering, where it's not possible to recognize the load
2687 // fold cause it has two uses through a bitcast. One use disappears at isel
2688 // time and the fold opportunity reappears.
2689 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2690 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2693 let Predicates = [UseSSE2] in {
2694 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2695 // problem is during lowering, where it's not possible to recognize the load
2696 // fold cause it has two uses through a bitcast. One use disappears at isel
2697 // time and the fold opportunity reappears.
2698 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2699 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2702 //===----------------------------------------------------------------------===//
2703 // SSE 1 & 2 - Extract Floating-Point Sign mask
2704 //===----------------------------------------------------------------------===//
2706 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2707 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2709 def rr : PI<0x50, MRMSrcReg, (outs GR32orGR64:$dst), (ins RC:$src),
2710 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2711 [(set GR32orGR64:$dst, (Int RC:$src))], IIC_SSE_MOVMSK, d>,
2712 Sched<[WriteVecLogic]>;
2715 let Predicates = [HasAVX] in {
2716 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2717 "movmskps", SSEPackedSingle>, TB, VEX;
2718 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2719 "movmskpd", SSEPackedDouble>, TB,
2721 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2722 "movmskps", SSEPackedSingle>, TB,
2724 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2725 "movmskpd", SSEPackedDouble>, TB,
2728 def : Pat<(i32 (X86fgetsign FR32:$src)),
2729 (VMOVMSKPSrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
2730 def : Pat<(i64 (X86fgetsign FR32:$src)),
2731 (SUBREG_TO_REG (i64 0),
2732 (VMOVMSKPSrr (COPY_TO_REGCLASS FR32:$src, VR128)), sub_32bit)>;
2733 def : Pat<(i32 (X86fgetsign FR64:$src)),
2734 (VMOVMSKPDrr (COPY_TO_REGCLASS FR64:$src, VR128))>;
2735 def : Pat<(i64 (X86fgetsign FR64:$src)),
2736 (SUBREG_TO_REG (i64 0),
2737 (VMOVMSKPDrr (COPY_TO_REGCLASS FR64:$src, VR128)), sub_32bit)>;
2740 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2741 SSEPackedSingle>, TB;
2742 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2743 SSEPackedDouble>, TB, OpSize;
2745 def : Pat<(i32 (X86fgetsign FR32:$src)),
2746 (MOVMSKPSrr (COPY_TO_REGCLASS FR32:$src, VR128))>,
2747 Requires<[UseSSE1]>;
2748 def : Pat<(i64 (X86fgetsign FR32:$src)),
2749 (SUBREG_TO_REG (i64 0),
2750 (MOVMSKPSrr (COPY_TO_REGCLASS FR32:$src, VR128)), sub_32bit)>,
2751 Requires<[UseSSE1]>;
2752 def : Pat<(i32 (X86fgetsign FR64:$src)),
2753 (MOVMSKPDrr (COPY_TO_REGCLASS FR64:$src, VR128))>,
2754 Requires<[UseSSE2]>;
2755 def : Pat<(i64 (X86fgetsign FR64:$src)),
2756 (SUBREG_TO_REG (i64 0),
2757 (MOVMSKPDrr (COPY_TO_REGCLASS FR64:$src, VR128)), sub_32bit)>,
2758 Requires<[UseSSE2]>;
2760 //===---------------------------------------------------------------------===//
2761 // SSE2 - Packed Integer Logical Instructions
2762 //===---------------------------------------------------------------------===//
2764 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2766 /// PDI_binop_rm - Simple SSE2 binary operator.
2767 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2768 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
2769 X86MemOperand x86memop, OpndItins itins,
2770 bit IsCommutable, bit Is2Addr> {
2771 let isCommutable = IsCommutable in
2772 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
2773 (ins RC:$src1, RC:$src2),
2775 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2776 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2777 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>,
2778 Sched<[itins.Sched]>;
2779 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
2780 (ins RC:$src1, x86memop:$src2),
2782 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2783 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2784 [(set RC:$dst, (OpVT (OpNode RC:$src1,
2785 (bitconvert (memop_frag addr:$src2)))))],
2787 Sched<[itins.Sched.Folded, ReadAfterLd]>;
2789 } // ExeDomain = SSEPackedInt
2791 multiclass PDI_binop_all<bits<8> opc, string OpcodeStr, SDNode Opcode,
2792 ValueType OpVT128, ValueType OpVT256,
2793 OpndItins itins, bit IsCommutable = 0> {
2794 let Predicates = [HasAVX] in
2795 defm V#NAME : PDI_binop_rm<opc, !strconcat("v", OpcodeStr), Opcode, OpVT128,
2796 VR128, memopv2i64, i128mem, itins, IsCommutable, 0>, VEX_4V;
2798 let Constraints = "$src1 = $dst" in
2799 defm NAME : PDI_binop_rm<opc, OpcodeStr, Opcode, OpVT128, VR128,
2800 memopv2i64, i128mem, itins, IsCommutable, 1>;
2802 let Predicates = [HasAVX2] in
2803 defm V#NAME#Y : PDI_binop_rm<opc, !strconcat("v", OpcodeStr), Opcode,
2804 OpVT256, VR256, memopv4i64, i256mem, itins,
2805 IsCommutable, 0>, VEX_4V, VEX_L;
2808 // These are ordered here for pattern ordering requirements with the fp versions
2810 defm PAND : PDI_binop_all<0xDB, "pand", and, v2i64, v4i64, SSE_BIT_ITINS_P, 1>;
2811 defm POR : PDI_binop_all<0xEB, "por", or, v2i64, v4i64, SSE_BIT_ITINS_P, 1>;
2812 defm PXOR : PDI_binop_all<0xEF, "pxor", xor, v2i64, v4i64, SSE_BIT_ITINS_P, 1>;
2813 defm PANDN : PDI_binop_all<0xDF, "pandn", X86andnp, v2i64, v4i64,
2814 SSE_BIT_ITINS_P, 0>;
2816 //===----------------------------------------------------------------------===//
2817 // SSE 1 & 2 - Logical Instructions
2818 //===----------------------------------------------------------------------===//
2820 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2822 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2823 SDNode OpNode, OpndItins itins> {
2824 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2825 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, itins, 0>,
2828 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2829 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, itins, 0>,
2832 let Constraints = "$src1 = $dst" in {
2833 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2834 f32, f128mem, memopfsf32, SSEPackedSingle, itins>,
2837 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2838 f64, f128mem, memopfsf64, SSEPackedDouble, itins>,
2843 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2844 let isCodeGenOnly = 1 in {
2845 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand,
2847 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for,
2849 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor,
2852 let isCommutable = 0 in
2853 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", X86fandn,
2857 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2859 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2861 defm V#NAME#PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2862 !strconcat(OpcodeStr, "ps"), f256mem,
2863 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2864 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2865 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V, VEX_L;
2867 defm V#NAME#PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2868 !strconcat(OpcodeStr, "pd"), f256mem,
2869 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2870 (bc_v4i64 (v4f64 VR256:$src2))))],
2871 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2872 (memopv4i64 addr:$src2)))], 0>,
2873 TB, OpSize, VEX_4V, VEX_L;
2875 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2876 // are all promoted to v2i64, and the patterns are covered by the int
2877 // version. This is needed in SSE only, because v2i64 isn't supported on
2878 // SSE1, but only on SSE2.
2879 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2880 !strconcat(OpcodeStr, "ps"), f128mem, [],
2881 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2882 (memopv2i64 addr:$src2)))], 0>, TB, VEX_4V;
2884 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2885 !strconcat(OpcodeStr, "pd"), f128mem,
2886 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2887 (bc_v2i64 (v2f64 VR128:$src2))))],
2888 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2889 (memopv2i64 addr:$src2)))], 0>,
2892 let Constraints = "$src1 = $dst" in {
2893 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2894 !strconcat(OpcodeStr, "ps"), f128mem,
2895 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2896 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2897 (memopv2i64 addr:$src2)))]>, TB;
2899 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2900 !strconcat(OpcodeStr, "pd"), f128mem,
2901 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2902 (bc_v2i64 (v2f64 VR128:$src2))))],
2903 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2904 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2908 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2909 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2910 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2911 let isCommutable = 0 in
2912 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2914 //===----------------------------------------------------------------------===//
2915 // SSE 1 & 2 - Arithmetic Instructions
2916 //===----------------------------------------------------------------------===//
2918 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2921 /// In addition, we also have a special variant of the scalar form here to
2922 /// represent the associated intrinsic operation. This form is unlike the
2923 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2924 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2926 /// These three forms can each be reg+reg or reg+mem.
2929 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2931 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr,
2932 SDNode OpNode, SizeItins itins> {
2933 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2934 VR128, v4f32, f128mem, memopv4f32,
2935 SSEPackedSingle, itins.s, 0>, TB, VEX_4V;
2936 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2937 VR128, v2f64, f128mem, memopv2f64,
2938 SSEPackedDouble, itins.d, 0>, TB, OpSize, VEX_4V;
2940 defm V#NAME#PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"),
2941 OpNode, VR256, v8f32, f256mem, memopv8f32,
2942 SSEPackedSingle, itins.s, 0>, TB, VEX_4V, VEX_L;
2943 defm V#NAME#PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"),
2944 OpNode, VR256, v4f64, f256mem, memopv4f64,
2945 SSEPackedDouble, itins.d, 0>, TB, OpSize, VEX_4V, VEX_L;
2947 let Constraints = "$src1 = $dst" in {
2948 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2949 v4f32, f128mem, memopv4f32, SSEPackedSingle,
2951 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2952 v2f64, f128mem, memopv2f64, SSEPackedDouble,
2953 itins.d>, TB, OpSize;
2957 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2959 defm V#NAME#SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2960 OpNode, FR32, f32mem, itins.s, 0>, XS, VEX_4V, VEX_LIG;
2961 defm V#NAME#SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2962 OpNode, FR64, f64mem, itins.d, 0>, XD, VEX_4V, VEX_LIG;
2964 let Constraints = "$src1 = $dst" in {
2965 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2966 OpNode, FR32, f32mem, itins.s>, XS;
2967 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2968 OpNode, FR64, f64mem, itins.d>, XD;
2972 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2974 defm V#NAME#SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2975 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32,
2976 itins.s, 0>, XS, VEX_4V, VEX_LIG;
2977 defm V#NAME#SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2978 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64,
2979 itins.d, 0>, XD, VEX_4V, VEX_LIG;
2981 let Constraints = "$src1 = $dst" in {
2982 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2983 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32,
2985 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2986 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64,
2991 // Binary Arithmetic instructions
2992 defm ADD : basic_sse12_fp_binop_p<0x58, "add", fadd, SSE_ALU_ITINS_P>,
2993 basic_sse12_fp_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S>,
2994 basic_sse12_fp_binop_s_int<0x58, "add", SSE_ALU_ITINS_S>;
2995 defm MUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, SSE_MUL_ITINS_P>,
2996 basic_sse12_fp_binop_s<0x59, "mul", fmul, SSE_MUL_ITINS_S>,
2997 basic_sse12_fp_binop_s_int<0x59, "mul", SSE_MUL_ITINS_S>;
2998 let isCommutable = 0 in {
2999 defm SUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, SSE_ALU_ITINS_P>,
3000 basic_sse12_fp_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S>,
3001 basic_sse12_fp_binop_s_int<0x5C, "sub", SSE_ALU_ITINS_S>;
3002 defm DIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, SSE_DIV_ITINS_P>,
3003 basic_sse12_fp_binop_s<0x5E, "div", fdiv, SSE_DIV_ITINS_S>,
3004 basic_sse12_fp_binop_s_int<0x5E, "div", SSE_DIV_ITINS_S>;
3005 defm MAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, SSE_ALU_ITINS_P>,
3006 basic_sse12_fp_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S>,
3007 basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S>;
3008 defm MIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, SSE_ALU_ITINS_P>,
3009 basic_sse12_fp_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S>,
3010 basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S>;
3013 let isCodeGenOnly = 1 in {
3014 defm MAXC: basic_sse12_fp_binop_p<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_P>,
3015 basic_sse12_fp_binop_s<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_S>;
3016 defm MINC: basic_sse12_fp_binop_p<0x5D, "min", X86fminc, SSE_ALU_ITINS_P>,
3017 basic_sse12_fp_binop_s<0x5D, "min", X86fminc, SSE_ALU_ITINS_S>;
3021 /// In addition, we also have a special variant of the scalar form here to
3022 /// represent the associated intrinsic operation. This form is unlike the
3023 /// plain scalar form, in that it takes an entire vector (instead of a
3024 /// scalar) and leaves the top elements undefined.
3026 /// And, we have a special variant form for a full-vector intrinsic form.
3028 let Sched = WriteFSqrt in {
3029 def SSE_SQRTPS : OpndItins<
3030 IIC_SSE_SQRTPS_RR, IIC_SSE_SQRTPS_RM
3033 def SSE_SQRTSS : OpndItins<
3034 IIC_SSE_SQRTSS_RR, IIC_SSE_SQRTSS_RM
3037 def SSE_SQRTPD : OpndItins<
3038 IIC_SSE_SQRTPD_RR, IIC_SSE_SQRTPD_RM
3041 def SSE_SQRTSD : OpndItins<
3042 IIC_SSE_SQRTSD_RR, IIC_SSE_SQRTSD_RM
3046 let Sched = WriteFRcp in {
3047 def SSE_RCPP : OpndItins<
3048 IIC_SSE_RCPP_RR, IIC_SSE_RCPP_RM
3051 def SSE_RCPS : OpndItins<
3052 IIC_SSE_RCPS_RR, IIC_SSE_RCPS_RM
3056 /// sse1_fp_unop_s - SSE1 unops in scalar form.
3057 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
3058 SDNode OpNode, Intrinsic F32Int, OpndItins itins> {
3059 let Predicates = [HasAVX], hasSideEffects = 0 in {
3060 def V#NAME#SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst),
3061 (ins FR32:$src1, FR32:$src2),
3062 !strconcat("v", OpcodeStr,
3063 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3064 []>, VEX_4V, VEX_LIG, Sched<[itins.Sched]>;
3065 let mayLoad = 1 in {
3066 def V#NAME#SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst),
3067 (ins FR32:$src1,f32mem:$src2),
3068 !strconcat("v", OpcodeStr,
3069 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3070 []>, VEX_4V, VEX_LIG,
3071 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3072 def V#NAME#SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3073 (ins VR128:$src1, ssmem:$src2),
3074 !strconcat("v", OpcodeStr,
3075 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3076 []>, VEX_4V, VEX_LIG,
3077 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3081 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
3082 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3083 [(set FR32:$dst, (OpNode FR32:$src))]>, Sched<[itins.Sched]>;
3084 // For scalar unary operations, fold a load into the operation
3085 // only in OptForSize mode. It eliminates an instruction, but it also
3086 // eliminates a whole-register clobber (the load), so it introduces a
3087 // partial register update condition.
3088 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
3089 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3090 [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
3091 Requires<[UseSSE1, OptForSize]>, Sched<[itins.Sched.Folded]>;
3092 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3093 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3094 [(set VR128:$dst, (F32Int VR128:$src))], itins.rr>,
3095 Sched<[itins.Sched]>;
3096 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
3097 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3098 [(set VR128:$dst, (F32Int sse_load_f32:$src))], itins.rm>,
3099 Sched<[itins.Sched.Folded]>;
3102 /// sse1_fp_unop_s_rw - SSE1 unops where vector form has a read-write operand.
3103 multiclass sse1_fp_unop_rw<bits<8> opc, string OpcodeStr, SDNode OpNode,
3105 let Predicates = [HasAVX], hasSideEffects = 0 in {
3106 def V#NAME#SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst),
3107 (ins FR32:$src1, FR32:$src2),
3108 !strconcat("v", OpcodeStr,
3109 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3110 []>, VEX_4V, VEX_LIG, Sched<[itins.Sched]>;
3111 let mayLoad = 1 in {
3112 def V#NAME#SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst),
3113 (ins FR32:$src1,f32mem:$src2),
3114 !strconcat("v", OpcodeStr,
3115 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3116 []>, VEX_4V, VEX_LIG,
3117 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3118 def V#NAME#SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3119 (ins VR128:$src1, ssmem:$src2),
3120 !strconcat("v", OpcodeStr,
3121 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3122 []>, VEX_4V, VEX_LIG,
3123 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3127 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
3128 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3129 [(set FR32:$dst, (OpNode FR32:$src))]>, Sched<[itins.Sched]>;
3130 // For scalar unary operations, fold a load into the operation
3131 // only in OptForSize mode. It eliminates an instruction, but it also
3132 // eliminates a whole-register clobber (the load), so it introduces a
3133 // partial register update condition.
3134 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
3135 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3136 [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
3137 Requires<[UseSSE1, OptForSize]>, Sched<[itins.Sched.Folded]>;
3138 let Constraints = "$src1 = $dst" in {
3139 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst),
3140 (ins VR128:$src1, VR128:$src2),
3141 !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
3142 [], itins.rr>, Sched<[itins.Sched]>;
3143 let mayLoad = 1, hasSideEffects = 0 in
3144 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3145 (ins VR128:$src1, ssmem:$src2),
3146 !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
3147 [], itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
3151 /// sse1_fp_unop_p - SSE1 unops in packed form.
3152 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
3154 let Predicates = [HasAVX] in {
3155 def V#NAME#PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3156 !strconcat("v", OpcodeStr,
3157 "ps\t{$src, $dst|$dst, $src}"),
3158 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))],
3159 itins.rr>, VEX, Sched<[itins.Sched]>;
3160 def V#NAME#PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3161 !strconcat("v", OpcodeStr,
3162 "ps\t{$src, $dst|$dst, $src}"),
3163 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))],
3164 itins.rm>, VEX, Sched<[itins.Sched.Folded]>;
3165 def V#NAME#PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3166 !strconcat("v", OpcodeStr,
3167 "ps\t{$src, $dst|$dst, $src}"),
3168 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))],
3169 itins.rr>, VEX, VEX_L, Sched<[itins.Sched]>;
3170 def V#NAME#PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3171 !strconcat("v", OpcodeStr,
3172 "ps\t{$src, $dst|$dst, $src}"),
3173 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))],
3174 itins.rm>, VEX, VEX_L, Sched<[itins.Sched.Folded]>;
3177 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3178 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3179 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))], itins.rr>,
3180 Sched<[itins.Sched]>;
3181 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3182 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3183 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))], itins.rm>,
3184 Sched<[itins.Sched.Folded]>;
3187 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
3188 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3189 Intrinsic V4F32Int, Intrinsic V8F32Int,
3191 let Predicates = [HasAVX] in {
3192 def V#NAME#PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3193 !strconcat("v", OpcodeStr,
3194 "ps\t{$src, $dst|$dst, $src}"),
3195 [(set VR128:$dst, (V4F32Int VR128:$src))],
3196 itins.rr>, VEX, Sched<[itins.Sched]>;
3197 def V#NAME#PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3198 !strconcat("v", OpcodeStr,
3199 "ps\t{$src, $dst|$dst, $src}"),
3200 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))],
3201 itins.rm>, VEX, Sched<[itins.Sched.Folded]>;
3202 def V#NAME#PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3203 !strconcat("v", OpcodeStr,
3204 "ps\t{$src, $dst|$dst, $src}"),
3205 [(set VR256:$dst, (V8F32Int VR256:$src))],
3206 itins.rr>, VEX, VEX_L, Sched<[itins.Sched]>;
3207 def V#NAME#PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst),
3209 !strconcat("v", OpcodeStr,
3210 "ps\t{$src, $dst|$dst, $src}"),
3211 [(set VR256:$dst, (V8F32Int (memopv8f32 addr:$src)))],
3212 itins.rm>, VEX, VEX_L, Sched<[itins.Sched.Folded]>;
3215 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3216 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3217 [(set VR128:$dst, (V4F32Int VR128:$src))],
3218 itins.rr>, Sched<[itins.Sched]>;
3219 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3220 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3221 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))],
3222 itins.rm>, Sched<[itins.Sched.Folded]>;
3225 /// sse2_fp_unop_s - SSE2 unops in scalar form.
3226 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
3227 SDNode OpNode, Intrinsic F64Int, OpndItins itins> {
3228 let Predicates = [HasAVX], hasSideEffects = 0 in {
3229 def V#NAME#SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst),
3230 (ins FR64:$src1, FR64:$src2),
3231 !strconcat("v", OpcodeStr,
3232 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3233 []>, VEX_4V, VEX_LIG, Sched<[itins.Sched]>;
3234 let mayLoad = 1 in {
3235 def V#NAME#SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst),
3236 (ins FR64:$src1,f64mem:$src2),
3237 !strconcat("v", OpcodeStr,
3238 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3239 []>, VEX_4V, VEX_LIG,
3240 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3241 def V#NAME#SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
3242 (ins VR128:$src1, sdmem:$src2),
3243 !strconcat("v", OpcodeStr,
3244 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3245 []>, VEX_4V, VEX_LIG,
3246 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3250 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
3251 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3252 [(set FR64:$dst, (OpNode FR64:$src))], itins.rr>,
3253 Sched<[itins.Sched]>;
3254 // See the comments in sse1_fp_unop_s for why this is OptForSize.
3255 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
3256 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3257 [(set FR64:$dst, (OpNode (load addr:$src)))], itins.rm>, XD,
3258 Requires<[UseSSE2, OptForSize]>, Sched<[itins.Sched.Folded]>;
3259 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3260 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3261 [(set VR128:$dst, (F64Int VR128:$src))], itins.rr>,
3262 Sched<[itins.Sched]>;
3263 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
3264 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3265 [(set VR128:$dst, (F64Int sse_load_f64:$src))], itins.rm>,
3266 Sched<[itins.Sched.Folded]>;
3269 /// sse2_fp_unop_p - SSE2 unops in vector forms.
3270 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
3271 SDNode OpNode, OpndItins itins> {
3272 let Predicates = [HasAVX] in {
3273 def V#NAME#PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3274 !strconcat("v", OpcodeStr,
3275 "pd\t{$src, $dst|$dst, $src}"),
3276 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))],
3277 itins.rr>, VEX, Sched<[itins.Sched]>;
3278 def V#NAME#PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3279 !strconcat("v", OpcodeStr,
3280 "pd\t{$src, $dst|$dst, $src}"),
3281 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))],
3282 itins.rm>, VEX, Sched<[itins.Sched.Folded]>;
3283 def V#NAME#PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3284 !strconcat("v", OpcodeStr,
3285 "pd\t{$src, $dst|$dst, $src}"),
3286 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))],
3287 itins.rr>, VEX, VEX_L, Sched<[itins.Sched]>;
3288 def V#NAME#PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3289 !strconcat("v", OpcodeStr,
3290 "pd\t{$src, $dst|$dst, $src}"),
3291 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))],
3292 itins.rm>, VEX, VEX_L, Sched<[itins.Sched.Folded]>;
3295 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3296 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3297 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))], itins.rr>,
3298 Sched<[itins.Sched]>;
3299 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3300 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3301 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))], itins.rm>,
3302 Sched<[itins.Sched.Folded]>;
3306 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss,
3308 sse1_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTPS>,
3309 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd,
3311 sse2_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTPD>;
3313 // Reciprocal approximations. Note that these typically require refinement
3314 // in order to obtain suitable precision.
3315 defm RSQRT : sse1_fp_unop_rw<0x52, "rsqrt", X86frsqrt, SSE_SQRTSS>,
3316 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, SSE_SQRTPS>,
3317 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps,
3318 int_x86_avx_rsqrt_ps_256, SSE_SQRTPS>;
3319 defm RCP : sse1_fp_unop_rw<0x53, "rcp", X86frcp, SSE_RCPS>,
3320 sse1_fp_unop_p<0x53, "rcp", X86frcp, SSE_RCPP>,
3321 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps,
3322 int_x86_avx_rcp_ps_256, SSE_RCPP>;
3324 let Predicates = [UseAVX] in {
3325 def : Pat<(f32 (fsqrt FR32:$src)),
3326 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3327 def : Pat<(f32 (fsqrt (load addr:$src))),
3328 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3329 Requires<[HasAVX, OptForSize]>;
3330 def : Pat<(f64 (fsqrt FR64:$src)),
3331 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
3332 def : Pat<(f64 (fsqrt (load addr:$src))),
3333 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
3334 Requires<[HasAVX, OptForSize]>;
3336 def : Pat<(f32 (X86frsqrt FR32:$src)),
3337 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3338 def : Pat<(f32 (X86frsqrt (load addr:$src))),
3339 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3340 Requires<[HasAVX, OptForSize]>;
3342 def : Pat<(f32 (X86frcp FR32:$src)),
3343 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3344 def : Pat<(f32 (X86frcp (load addr:$src))),
3345 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3346 Requires<[HasAVX, OptForSize]>;
3348 let Predicates = [UseAVX] in {
3349 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
3350 (COPY_TO_REGCLASS (VSQRTSSr (f32 (IMPLICIT_DEF)),
3351 (COPY_TO_REGCLASS VR128:$src, FR32)),
3353 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
3354 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3356 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
3357 (COPY_TO_REGCLASS (VSQRTSDr (f64 (IMPLICIT_DEF)),
3358 (COPY_TO_REGCLASS VR128:$src, FR64)),
3360 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
3361 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
3364 let Predicates = [HasAVX] in {
3365 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3366 (COPY_TO_REGCLASS (VRSQRTSSr (f32 (IMPLICIT_DEF)),
3367 (COPY_TO_REGCLASS VR128:$src, FR32)),
3369 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
3370 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3372 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3373 (COPY_TO_REGCLASS (VRCPSSr (f32 (IMPLICIT_DEF)),
3374 (COPY_TO_REGCLASS VR128:$src, FR32)),
3376 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
3377 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3380 // Reciprocal approximations. Note that these typically require refinement
3381 // in order to obtain suitable precision.
3382 let Predicates = [UseSSE1] in {
3383 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3384 (RSQRTSSr_Int VR128:$src, VR128:$src)>;
3385 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3386 (RCPSSr_Int VR128:$src, VR128:$src)>;
3389 // There is no f64 version of the reciprocal approximation instructions.
3391 //===----------------------------------------------------------------------===//
3392 // SSE 1 & 2 - Non-temporal stores
3393 //===----------------------------------------------------------------------===//
3395 let AddedComplexity = 400 in { // Prefer non-temporal versions
3396 let SchedRW = [WriteStore] in {
3397 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3398 (ins f128mem:$dst, VR128:$src),
3399 "movntps\t{$src, $dst|$dst, $src}",
3400 [(alignednontemporalstore (v4f32 VR128:$src),
3402 IIC_SSE_MOVNT>, VEX;
3403 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3404 (ins f128mem:$dst, VR128:$src),
3405 "movntpd\t{$src, $dst|$dst, $src}",
3406 [(alignednontemporalstore (v2f64 VR128:$src),
3408 IIC_SSE_MOVNT>, VEX;
3410 let ExeDomain = SSEPackedInt in
3411 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3412 (ins f128mem:$dst, VR128:$src),
3413 "movntdq\t{$src, $dst|$dst, $src}",
3414 [(alignednontemporalstore (v2i64 VR128:$src),
3416 IIC_SSE_MOVNT>, VEX;
3418 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3419 (ins f256mem:$dst, VR256:$src),
3420 "movntps\t{$src, $dst|$dst, $src}",
3421 [(alignednontemporalstore (v8f32 VR256:$src),
3423 IIC_SSE_MOVNT>, VEX, VEX_L;
3424 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3425 (ins f256mem:$dst, VR256:$src),
3426 "movntpd\t{$src, $dst|$dst, $src}",
3427 [(alignednontemporalstore (v4f64 VR256:$src),
3429 IIC_SSE_MOVNT>, VEX, VEX_L;
3430 let ExeDomain = SSEPackedInt in
3431 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3432 (ins f256mem:$dst, VR256:$src),
3433 "movntdq\t{$src, $dst|$dst, $src}",
3434 [(alignednontemporalstore (v4i64 VR256:$src),
3436 IIC_SSE_MOVNT>, VEX, VEX_L;
3438 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3439 "movntps\t{$src, $dst|$dst, $src}",
3440 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)],
3442 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3443 "movntpd\t{$src, $dst|$dst, $src}",
3444 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)],
3447 let ExeDomain = SSEPackedInt in
3448 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3449 "movntdq\t{$src, $dst|$dst, $src}",
3450 [(alignednontemporalstore (v2i64 VR128:$src), addr:$dst)],
3453 // There is no AVX form for instructions below this point
3454 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3455 "movnti{l}\t{$src, $dst|$dst, $src}",
3456 [(nontemporalstore (i32 GR32:$src), addr:$dst)],
3458 TB, Requires<[HasSSE2]>;
3459 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3460 "movnti{q}\t{$src, $dst|$dst, $src}",
3461 [(nontemporalstore (i64 GR64:$src), addr:$dst)],
3463 TB, Requires<[HasSSE2]>;
3464 } // SchedRW = [WriteStore]
3466 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3467 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3469 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3470 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[UseSSE2]>;
3471 } // AddedComplexity
3473 //===----------------------------------------------------------------------===//
3474 // SSE 1 & 2 - Prefetch and memory fence
3475 //===----------------------------------------------------------------------===//
3477 // Prefetch intrinsic.
3478 let Predicates = [HasSSE1], SchedRW = [WriteLoad] in {
3479 def PREFETCHT0 : I<0x18, MRM1m, (outs), (ins i8mem:$src),
3480 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))],
3481 IIC_SSE_PREFETCH>, TB;
3482 def PREFETCHT1 : I<0x18, MRM2m, (outs), (ins i8mem:$src),
3483 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))],
3484 IIC_SSE_PREFETCH>, TB;
3485 def PREFETCHT2 : I<0x18, MRM3m, (outs), (ins i8mem:$src),
3486 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))],
3487 IIC_SSE_PREFETCH>, TB;
3488 def PREFETCHNTA : I<0x18, MRM0m, (outs), (ins i8mem:$src),
3489 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))],
3490 IIC_SSE_PREFETCH>, TB;
3493 // FIXME: How should these memory instructions be modeled?
3494 let SchedRW = [WriteLoad] in {
3496 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3497 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)],
3498 IIC_SSE_PREFETCH>, TB, Requires<[HasSSE2]>;
3500 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3501 // was introduced with SSE2, it's backward compatible.
3502 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", [], IIC_SSE_PAUSE>, REP;
3504 // Load, store, and memory fence
3505 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3506 "sfence", [(int_x86_sse_sfence)], IIC_SSE_SFENCE>,
3507 TB, Requires<[HasSSE1]>;
3508 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3509 "lfence", [(int_x86_sse2_lfence)], IIC_SSE_LFENCE>,
3510 TB, Requires<[HasSSE2]>;
3511 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3512 "mfence", [(int_x86_sse2_mfence)], IIC_SSE_MFENCE>,
3513 TB, Requires<[HasSSE2]>;
3516 def : Pat<(X86SFence), (SFENCE)>;
3517 def : Pat<(X86LFence), (LFENCE)>;
3518 def : Pat<(X86MFence), (MFENCE)>;
3520 //===----------------------------------------------------------------------===//
3521 // SSE 1 & 2 - Load/Store XCSR register
3522 //===----------------------------------------------------------------------===//
3524 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3525 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3526 IIC_SSE_LDMXCSR>, VEX, Sched<[WriteLoad]>;
3527 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3528 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3529 IIC_SSE_STMXCSR>, VEX, Sched<[WriteStore]>;
3531 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3532 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3533 IIC_SSE_LDMXCSR>, Sched<[WriteLoad]>;
3534 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3535 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3536 IIC_SSE_STMXCSR>, Sched<[WriteStore]>;
3538 //===---------------------------------------------------------------------===//
3539 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3540 //===---------------------------------------------------------------------===//
3542 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3544 let neverHasSideEffects = 1, SchedRW = [WriteMove] in {
3545 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3546 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3548 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3549 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3551 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3552 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3554 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3555 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3560 let isCodeGenOnly = 1, hasSideEffects = 0, SchedRW = [WriteMove] in {
3561 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3562 "movdqa\t{$src, $dst|$dst, $src}", [],
3565 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3566 "movdqa\t{$src, $dst|$dst, $src}", [],
3567 IIC_SSE_MOVA_P_RR>, VEX, VEX_L;
3568 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3569 "movdqu\t{$src, $dst|$dst, $src}", [],
3572 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3573 "movdqu\t{$src, $dst|$dst, $src}", [],
3574 IIC_SSE_MOVU_P_RR>, VEX, VEX_L;
3577 let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1,
3578 neverHasSideEffects = 1, SchedRW = [WriteLoad] in {
3579 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3580 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3582 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3583 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3585 let Predicates = [HasAVX] in {
3586 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3587 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3589 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3590 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3595 let mayStore = 1, neverHasSideEffects = 1, SchedRW = [WriteStore] in {
3596 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3597 (ins i128mem:$dst, VR128:$src),
3598 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3600 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3601 (ins i256mem:$dst, VR256:$src),
3602 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3604 let Predicates = [HasAVX] in {
3605 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3606 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3608 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3609 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3614 let SchedRW = [WriteMove] in {
3615 let neverHasSideEffects = 1 in
3616 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3617 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>;
3619 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3620 "movdqu\t{$src, $dst|$dst, $src}",
3621 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>;
3624 let isCodeGenOnly = 1, hasSideEffects = 0 in {
3625 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3626 "movdqa\t{$src, $dst|$dst, $src}", [],
3629 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3630 "movdqu\t{$src, $dst|$dst, $src}",
3631 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>;
3635 let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1,
3636 neverHasSideEffects = 1, SchedRW = [WriteLoad] in {
3637 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3638 "movdqa\t{$src, $dst|$dst, $src}",
3639 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/],
3641 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3642 "movdqu\t{$src, $dst|$dst, $src}",
3643 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/],
3645 XS, Requires<[UseSSE2]>;
3648 let mayStore = 1, neverHasSideEffects = 1, SchedRW = [WriteStore] in {
3649 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3650 "movdqa\t{$src, $dst|$dst, $src}",
3651 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/],
3653 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3654 "movdqu\t{$src, $dst|$dst, $src}",
3655 [/*(store (v2i64 VR128:$src), addr:$dst)*/],
3657 XS, Requires<[UseSSE2]>;
3660 } // ExeDomain = SSEPackedInt
3662 let Predicates = [HasAVX] in {
3663 def : Pat<(int_x86_sse2_storeu_dq addr:$dst, VR128:$src),
3664 (VMOVDQUmr addr:$dst, VR128:$src)>;
3665 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3666 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3668 let Predicates = [UseSSE2] in
3669 def : Pat<(int_x86_sse2_storeu_dq addr:$dst, VR128:$src),
3670 (MOVDQUmr addr:$dst, VR128:$src)>;
3672 //===---------------------------------------------------------------------===//
3673 // SSE2 - Packed Integer Arithmetic Instructions
3674 //===---------------------------------------------------------------------===//
3676 let Sched = WriteVecIMul in
3677 def SSE_PMADD : OpndItins<
3678 IIC_SSE_PMADD, IIC_SSE_PMADD
3681 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3683 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3684 RegisterClass RC, PatFrag memop_frag,
3685 X86MemOperand x86memop,
3687 bit IsCommutable = 0,
3689 let isCommutable = IsCommutable in
3690 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3691 (ins RC:$src1, RC:$src2),
3693 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3694 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3695 [(set RC:$dst, (IntId RC:$src1, RC:$src2))], itins.rr>,
3696 Sched<[itins.Sched]>;
3697 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3698 (ins RC:$src1, x86memop:$src2),
3700 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3701 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3702 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))],
3703 itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
3706 multiclass PDI_binop_all_int<bits<8> opc, string OpcodeStr, Intrinsic IntId128,
3707 Intrinsic IntId256, OpndItins itins,
3708 bit IsCommutable = 0> {
3709 let Predicates = [HasAVX] in
3710 defm V#NAME : PDI_binop_rm_int<opc, !strconcat("v", OpcodeStr), IntId128,
3711 VR128, memopv2i64, i128mem, itins,
3712 IsCommutable, 0>, VEX_4V;
3714 let Constraints = "$src1 = $dst" in
3715 defm NAME : PDI_binop_rm_int<opc, OpcodeStr, IntId128, VR128, memopv2i64,
3716 i128mem, itins, IsCommutable, 1>;
3718 let Predicates = [HasAVX2] in
3719 defm V#NAME#Y : PDI_binop_rm_int<opc, !strconcat("v", OpcodeStr), IntId256,
3720 VR256, memopv4i64, i256mem, itins,
3721 IsCommutable, 0>, VEX_4V, VEX_L;
3724 multiclass PDI_binop_rmi<bits<8> opc, bits<8> opc2, Format ImmForm,
3725 string OpcodeStr, SDNode OpNode,
3726 SDNode OpNode2, RegisterClass RC,
3727 ValueType DstVT, ValueType SrcVT, PatFrag bc_frag,
3728 ShiftOpndItins itins,
3730 // src2 is always 128-bit
3731 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3732 (ins RC:$src1, VR128:$src2),
3734 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3735 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3736 [(set RC:$dst, (DstVT (OpNode RC:$src1, (SrcVT VR128:$src2))))],
3737 itins.rr>, Sched<[WriteVecShift]>;
3738 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3739 (ins RC:$src1, i128mem:$src2),
3741 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3742 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3743 [(set RC:$dst, (DstVT (OpNode RC:$src1,
3744 (bc_frag (memopv2i64 addr:$src2)))))], itins.rm>,
3745 Sched<[WriteVecShiftLd, ReadAfterLd]>;
3746 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3747 (ins RC:$src1, i8imm:$src2),
3749 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3750 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3751 [(set RC:$dst, (DstVT (OpNode2 RC:$src1, (i8 imm:$src2))))], itins.ri>,
3752 Sched<[WriteVecShift]>;
3755 /// PDI_binop_rm2 - Simple SSE2 binary operator with different src and dst types
3756 multiclass PDI_binop_rm2<bits<8> opc, string OpcodeStr, SDNode OpNode,
3757 ValueType DstVT, ValueType SrcVT, RegisterClass RC,
3758 PatFrag memop_frag, X86MemOperand x86memop,
3760 bit IsCommutable = 0, bit Is2Addr = 1> {
3761 let isCommutable = IsCommutable in
3762 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3763 (ins RC:$src1, RC:$src2),
3765 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3766 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3767 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1), RC:$src2)))]>,
3768 Sched<[itins.Sched]>;
3769 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3770 (ins RC:$src1, x86memop:$src2),
3772 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3773 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3774 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1),
3775 (bitconvert (memop_frag addr:$src2)))))]>,
3776 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3778 } // ExeDomain = SSEPackedInt
3780 defm PADDB : PDI_binop_all<0xFC, "paddb", add, v16i8, v32i8,
3781 SSE_INTALU_ITINS_P, 1>;
3782 defm PADDW : PDI_binop_all<0xFD, "paddw", add, v8i16, v16i16,
3783 SSE_INTALU_ITINS_P, 1>;
3784 defm PADDD : PDI_binop_all<0xFE, "paddd", add, v4i32, v8i32,
3785 SSE_INTALU_ITINS_P, 1>;
3786 defm PADDQ : PDI_binop_all<0xD4, "paddq", add, v2i64, v4i64,
3787 SSE_INTALUQ_ITINS_P, 1>;
3788 defm PMULLW : PDI_binop_all<0xD5, "pmullw", mul, v8i16, v16i16,
3789 SSE_INTMUL_ITINS_P, 1>;
3790 defm PSUBB : PDI_binop_all<0xF8, "psubb", sub, v16i8, v32i8,
3791 SSE_INTALU_ITINS_P, 0>;
3792 defm PSUBW : PDI_binop_all<0xF9, "psubw", sub, v8i16, v16i16,
3793 SSE_INTALU_ITINS_P, 0>;
3794 defm PSUBD : PDI_binop_all<0xFA, "psubd", sub, v4i32, v8i32,
3795 SSE_INTALU_ITINS_P, 0>;
3796 defm PSUBQ : PDI_binop_all<0xFB, "psubq", sub, v2i64, v4i64,
3797 SSE_INTALUQ_ITINS_P, 0>;
3798 defm PSUBUSB : PDI_binop_all<0xD8, "psubusb", X86subus, v16i8, v32i8,
3799 SSE_INTALU_ITINS_P, 0>;
3800 defm PSUBUSW : PDI_binop_all<0xD9, "psubusw", X86subus, v8i16, v16i16,
3801 SSE_INTALU_ITINS_P, 0>;
3802 defm PMINUB : PDI_binop_all<0xDA, "pminub", X86umin, v16i8, v32i8,
3803 SSE_INTALU_ITINS_P, 1>;
3804 defm PMINSW : PDI_binop_all<0xEA, "pminsw", X86smin, v8i16, v16i16,
3805 SSE_INTALU_ITINS_P, 1>;
3806 defm PMAXUB : PDI_binop_all<0xDE, "pmaxub", X86umax, v16i8, v32i8,
3807 SSE_INTALU_ITINS_P, 1>;
3808 defm PMAXSW : PDI_binop_all<0xEE, "pmaxsw", X86smax, v8i16, v16i16,
3809 SSE_INTALU_ITINS_P, 1>;
3812 defm PSUBSB : PDI_binop_all_int<0xE8, "psubsb", int_x86_sse2_psubs_b,
3813 int_x86_avx2_psubs_b, SSE_INTALU_ITINS_P, 0>;
3814 defm PSUBSW : PDI_binop_all_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
3815 int_x86_avx2_psubs_w, SSE_INTALU_ITINS_P, 0>;
3816 defm PADDSB : PDI_binop_all_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
3817 int_x86_avx2_padds_b, SSE_INTALU_ITINS_P, 1>;
3818 defm PADDSW : PDI_binop_all_int<0xED, "paddsw" , int_x86_sse2_padds_w,
3819 int_x86_avx2_padds_w, SSE_INTALU_ITINS_P, 1>;
3820 defm PADDUSB : PDI_binop_all_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
3821 int_x86_avx2_paddus_b, SSE_INTALU_ITINS_P, 1>;
3822 defm PADDUSW : PDI_binop_all_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
3823 int_x86_avx2_paddus_w, SSE_INTALU_ITINS_P, 1>;
3824 defm PMULHUW : PDI_binop_all_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
3825 int_x86_avx2_pmulhu_w, SSE_INTMUL_ITINS_P, 1>;
3826 defm PMULHW : PDI_binop_all_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
3827 int_x86_avx2_pmulh_w, SSE_INTMUL_ITINS_P, 1>;
3828 defm PMADDWD : PDI_binop_all_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
3829 int_x86_avx2_pmadd_wd, SSE_PMADD, 1>;
3830 defm PAVGB : PDI_binop_all_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
3831 int_x86_avx2_pavg_b, SSE_INTALU_ITINS_P, 1>;
3832 defm PAVGW : PDI_binop_all_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
3833 int_x86_avx2_pavg_w, SSE_INTALU_ITINS_P, 1>;
3834 defm PSADBW : PDI_binop_all_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
3835 int_x86_avx2_psad_bw, SSE_PMADD, 1>;
3837 let Predicates = [HasAVX] in
3838 defm VPMULUDQ : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v2i64, v4i32, VR128,
3839 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1, 0>,
3841 let Predicates = [HasAVX2] in
3842 defm VPMULUDQY : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v4i64, v8i32,
3843 VR256, memopv4i64, i256mem,
3844 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V, VEX_L;
3845 let Constraints = "$src1 = $dst" in
3846 defm PMULUDQ : PDI_binop_rm2<0xF4, "pmuludq", X86pmuludq, v2i64, v4i32, VR128,
3847 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1>;
3849 //===---------------------------------------------------------------------===//
3850 // SSE2 - Packed Integer Logical Instructions
3851 //===---------------------------------------------------------------------===//
3853 let Predicates = [HasAVX] in {
3854 defm VPSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3855 VR128, v8i16, v8i16, bc_v8i16,
3856 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3857 defm VPSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3858 VR128, v4i32, v4i32, bc_v4i32,
3859 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3860 defm VPSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3861 VR128, v2i64, v2i64, bc_v2i64,
3862 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3864 defm VPSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3865 VR128, v8i16, v8i16, bc_v8i16,
3866 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3867 defm VPSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3868 VR128, v4i32, v4i32, bc_v4i32,
3869 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3870 defm VPSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3871 VR128, v2i64, v2i64, bc_v2i64,
3872 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3874 defm VPSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3875 VR128, v8i16, v8i16, bc_v8i16,
3876 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3877 defm VPSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3878 VR128, v4i32, v4i32, bc_v4i32,
3879 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3881 let ExeDomain = SSEPackedInt, SchedRW = [WriteVecShift] in {
3882 // 128-bit logical shifts.
3883 def VPSLLDQri : PDIi8<0x73, MRM7r,
3884 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3885 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3887 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>,
3889 def VPSRLDQri : PDIi8<0x73, MRM3r,
3890 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3891 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3893 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>,
3895 // PSRADQri doesn't exist in SSE[1-3].
3897 } // Predicates = [HasAVX]
3899 let Predicates = [HasAVX2] in {
3900 defm VPSLLWY : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3901 VR256, v16i16, v8i16, bc_v8i16,
3902 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3903 defm VPSLLDY : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3904 VR256, v8i32, v4i32, bc_v4i32,
3905 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3906 defm VPSLLQY : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3907 VR256, v4i64, v2i64, bc_v2i64,
3908 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3910 defm VPSRLWY : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3911 VR256, v16i16, v8i16, bc_v8i16,
3912 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3913 defm VPSRLDY : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3914 VR256, v8i32, v4i32, bc_v4i32,
3915 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3916 defm VPSRLQY : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3917 VR256, v4i64, v2i64, bc_v2i64,
3918 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3920 defm VPSRAWY : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3921 VR256, v16i16, v8i16, bc_v8i16,
3922 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3923 defm VPSRADY : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3924 VR256, v8i32, v4i32, bc_v4i32,
3925 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3927 let ExeDomain = SSEPackedInt, SchedRW = [WriteVecShift] in {
3928 // 256-bit logical shifts.
3929 def VPSLLDQYri : PDIi8<0x73, MRM7r,
3930 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3931 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3933 (int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2))]>,
3935 def VPSRLDQYri : PDIi8<0x73, MRM3r,
3936 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3937 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3939 (int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2))]>,
3941 // PSRADQYri doesn't exist in SSE[1-3].
3943 } // Predicates = [HasAVX2]
3945 let Constraints = "$src1 = $dst" in {
3946 defm PSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "psllw", X86vshl, X86vshli,
3947 VR128, v8i16, v8i16, bc_v8i16,
3948 SSE_INTSHIFT_ITINS_P>;
3949 defm PSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "pslld", X86vshl, X86vshli,
3950 VR128, v4i32, v4i32, bc_v4i32,
3951 SSE_INTSHIFT_ITINS_P>;
3952 defm PSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "psllq", X86vshl, X86vshli,
3953 VR128, v2i64, v2i64, bc_v2i64,
3954 SSE_INTSHIFT_ITINS_P>;
3956 defm PSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "psrlw", X86vsrl, X86vsrli,
3957 VR128, v8i16, v8i16, bc_v8i16,
3958 SSE_INTSHIFT_ITINS_P>;
3959 defm PSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "psrld", X86vsrl, X86vsrli,
3960 VR128, v4i32, v4i32, bc_v4i32,
3961 SSE_INTSHIFT_ITINS_P>;
3962 defm PSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "psrlq", X86vsrl, X86vsrli,
3963 VR128, v2i64, v2i64, bc_v2i64,
3964 SSE_INTSHIFT_ITINS_P>;
3966 defm PSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "psraw", X86vsra, X86vsrai,
3967 VR128, v8i16, v8i16, bc_v8i16,
3968 SSE_INTSHIFT_ITINS_P>;
3969 defm PSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "psrad", X86vsra, X86vsrai,
3970 VR128, v4i32, v4i32, bc_v4i32,
3971 SSE_INTSHIFT_ITINS_P>;
3973 let ExeDomain = SSEPackedInt, SchedRW = [WriteVecShift] in {
3974 // 128-bit logical shifts.
3975 def PSLLDQri : PDIi8<0x73, MRM7r,
3976 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3977 "pslldq\t{$src2, $dst|$dst, $src2}",
3979 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))],
3980 IIC_SSE_INTSHDQ_P_RI>;
3981 def PSRLDQri : PDIi8<0x73, MRM3r,
3982 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3983 "psrldq\t{$src2, $dst|$dst, $src2}",
3985 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))],
3986 IIC_SSE_INTSHDQ_P_RI>;
3987 // PSRADQri doesn't exist in SSE[1-3].
3989 } // Constraints = "$src1 = $dst"
3991 let Predicates = [HasAVX] in {
3992 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3993 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3994 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3995 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3996 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3997 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3999 // Shift up / down and insert zero's.
4000 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
4001 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4002 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
4003 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4006 let Predicates = [HasAVX2] in {
4007 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
4008 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
4009 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
4010 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
4013 let Predicates = [UseSSE2] in {
4014 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
4015 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4016 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
4017 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4018 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
4019 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4021 // Shift up / down and insert zero's.
4022 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
4023 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4024 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
4025 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4028 //===---------------------------------------------------------------------===//
4029 // SSE2 - Packed Integer Comparison Instructions
4030 //===---------------------------------------------------------------------===//
4032 defm PCMPEQB : PDI_binop_all<0x74, "pcmpeqb", X86pcmpeq, v16i8, v32i8,
4033 SSE_INTALU_ITINS_P, 1>;
4034 defm PCMPEQW : PDI_binop_all<0x75, "pcmpeqw", X86pcmpeq, v8i16, v16i16,
4035 SSE_INTALU_ITINS_P, 1>;
4036 defm PCMPEQD : PDI_binop_all<0x76, "pcmpeqd", X86pcmpeq, v4i32, v8i32,
4037 SSE_INTALU_ITINS_P, 1>;
4038 defm PCMPGTB : PDI_binop_all<0x64, "pcmpgtb", X86pcmpgt, v16i8, v32i8,
4039 SSE_INTALU_ITINS_P, 0>;
4040 defm PCMPGTW : PDI_binop_all<0x65, "pcmpgtw", X86pcmpgt, v8i16, v16i16,
4041 SSE_INTALU_ITINS_P, 0>;
4042 defm PCMPGTD : PDI_binop_all<0x66, "pcmpgtd", X86pcmpgt, v4i32, v8i32,
4043 SSE_INTALU_ITINS_P, 0>;
4045 //===---------------------------------------------------------------------===//
4046 // SSE2 - Packed Integer Pack Instructions
4047 //===---------------------------------------------------------------------===//
4049 defm PACKSSWB : PDI_binop_all_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
4050 int_x86_avx2_packsswb, SSE_INTALU_ITINS_P, 0>;
4051 defm PACKSSDW : PDI_binop_all_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
4052 int_x86_avx2_packssdw, SSE_INTALU_ITINS_P, 0>;
4053 defm PACKUSWB : PDI_binop_all_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
4054 int_x86_avx2_packuswb, SSE_INTALU_ITINS_P, 0>;
4056 //===---------------------------------------------------------------------===//
4057 // SSE2 - Packed Integer Shuffle Instructions
4058 //===---------------------------------------------------------------------===//
4060 let ExeDomain = SSEPackedInt in {
4061 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt128, ValueType vt256,
4063 let Predicates = [HasAVX] in {
4064 def V#NAME#ri : Ii8<0x70, MRMSrcReg, (outs VR128:$dst),
4065 (ins VR128:$src1, i8imm:$src2),
4066 !strconcat("v", OpcodeStr,
4067 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4069 (vt128 (OpNode VR128:$src1, (i8 imm:$src2))))],
4070 IIC_SSE_PSHUF_RI>, VEX, Sched<[WriteShuffle]>;
4071 def V#NAME#mi : Ii8<0x70, MRMSrcMem, (outs VR128:$dst),
4072 (ins i128mem:$src1, i8imm:$src2),
4073 !strconcat("v", OpcodeStr,
4074 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4076 (vt128 (OpNode (bitconvert (memopv2i64 addr:$src1)),
4077 (i8 imm:$src2))))], IIC_SSE_PSHUF_MI>, VEX,
4078 Sched<[WriteShuffleLd]>;
4081 let Predicates = [HasAVX2] in {
4082 def V#NAME#Yri : Ii8<0x70, MRMSrcReg, (outs VR256:$dst),
4083 (ins VR256:$src1, i8imm:$src2),
4084 !strconcat("v", OpcodeStr,
4085 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4087 (vt256 (OpNode VR256:$src1, (i8 imm:$src2))))],
4088 IIC_SSE_PSHUF_RI>, VEX, VEX_L, Sched<[WriteShuffle]>;
4089 def V#NAME#Ymi : Ii8<0x70, MRMSrcMem, (outs VR256:$dst),
4090 (ins i256mem:$src1, i8imm:$src2),
4091 !strconcat("v", OpcodeStr,
4092 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4094 (vt256 (OpNode (bitconvert (memopv4i64 addr:$src1)),
4095 (i8 imm:$src2))))], IIC_SSE_PSHUF_MI>, VEX, VEX_L,
4096 Sched<[WriteShuffleLd]>;
4099 let Predicates = [UseSSE2] in {
4100 def ri : Ii8<0x70, MRMSrcReg,
4101 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
4102 !strconcat(OpcodeStr,
4103 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4105 (vt128 (OpNode VR128:$src1, (i8 imm:$src2))))],
4106 IIC_SSE_PSHUF_RI>, Sched<[WriteShuffle]>;
4107 def mi : Ii8<0x70, MRMSrcMem,
4108 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
4109 !strconcat(OpcodeStr,
4110 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4112 (vt128 (OpNode (bitconvert (memopv2i64 addr:$src1)),
4113 (i8 imm:$src2))))], IIC_SSE_PSHUF_MI>,
4114 Sched<[WriteShuffleLd]>;
4117 } // ExeDomain = SSEPackedInt
4119 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, v8i32, X86PShufd>, TB, OpSize;
4120 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, v16i16, X86PShufhw>, XS;
4121 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, v16i16, X86PShuflw>, XD;
4123 let Predicates = [HasAVX] in {
4124 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
4125 (VPSHUFDmi addr:$src1, imm:$imm)>;
4126 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4127 (VPSHUFDri VR128:$src1, imm:$imm)>;
4130 let Predicates = [UseSSE2] in {
4131 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
4132 (PSHUFDmi addr:$src1, imm:$imm)>;
4133 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4134 (PSHUFDri VR128:$src1, imm:$imm)>;
4137 //===---------------------------------------------------------------------===//
4138 // SSE2 - Packed Integer Unpack Instructions
4139 //===---------------------------------------------------------------------===//
4141 let ExeDomain = SSEPackedInt in {
4142 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
4143 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
4144 def rr : PDI<opc, MRMSrcReg,
4145 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4147 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4148 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4149 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))],
4150 IIC_SSE_UNPCK>, Sched<[WriteShuffle]>;
4151 def rm : PDI<opc, MRMSrcMem,
4152 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4154 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4155 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4156 [(set VR128:$dst, (OpNode VR128:$src1,
4157 (bc_frag (memopv2i64
4160 Sched<[WriteShuffleLd, ReadAfterLd]>;
4163 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
4164 SDNode OpNode, PatFrag bc_frag> {
4165 def Yrr : PDI<opc, MRMSrcReg,
4166 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4167 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4168 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>,
4169 Sched<[WriteShuffle]>;
4170 def Yrm : PDI<opc, MRMSrcMem,
4171 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4172 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4173 [(set VR256:$dst, (OpNode VR256:$src1,
4174 (bc_frag (memopv4i64 addr:$src2))))]>,
4175 Sched<[WriteShuffleLd, ReadAfterLd]>;
4178 let Predicates = [HasAVX] in {
4179 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
4180 bc_v16i8, 0>, VEX_4V;
4181 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
4182 bc_v8i16, 0>, VEX_4V;
4183 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
4184 bc_v4i32, 0>, VEX_4V;
4185 defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
4186 bc_v2i64, 0>, VEX_4V;
4188 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
4189 bc_v16i8, 0>, VEX_4V;
4190 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
4191 bc_v8i16, 0>, VEX_4V;
4192 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
4193 bc_v4i32, 0>, VEX_4V;
4194 defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
4195 bc_v2i64, 0>, VEX_4V;
4198 let Predicates = [HasAVX2] in {
4199 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
4200 bc_v32i8>, VEX_4V, VEX_L;
4201 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
4202 bc_v16i16>, VEX_4V, VEX_L;
4203 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
4204 bc_v8i32>, VEX_4V, VEX_L;
4205 defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
4206 bc_v4i64>, VEX_4V, VEX_L;
4208 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
4209 bc_v32i8>, VEX_4V, VEX_L;
4210 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
4211 bc_v16i16>, VEX_4V, VEX_L;
4212 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
4213 bc_v8i32>, VEX_4V, VEX_L;
4214 defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
4215 bc_v4i64>, VEX_4V, VEX_L;
4218 let Constraints = "$src1 = $dst" in {
4219 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
4221 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
4223 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
4225 defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
4228 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
4230 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
4232 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
4234 defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
4237 } // ExeDomain = SSEPackedInt
4239 //===---------------------------------------------------------------------===//
4240 // SSE2 - Packed Integer Extract and Insert
4241 //===---------------------------------------------------------------------===//
4243 let ExeDomain = SSEPackedInt in {
4244 multiclass sse2_pinsrw<bit Is2Addr = 1> {
4245 def rri : Ii8<0xC4, MRMSrcReg,
4246 (outs VR128:$dst), (ins VR128:$src1,
4247 GR32orGR64:$src2, i32i8imm:$src3),
4249 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4250 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4252 (X86pinsrw VR128:$src1, GR32orGR64:$src2, imm:$src3))],
4253 IIC_SSE_PINSRW>, Sched<[WriteShuffle]>;
4254 def rmi : Ii8<0xC4, MRMSrcMem,
4255 (outs VR128:$dst), (ins VR128:$src1,
4256 i16mem:$src2, i32i8imm:$src3),
4258 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4259 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4261 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
4262 imm:$src3))], IIC_SSE_PINSRW>,
4263 Sched<[WriteShuffleLd, ReadAfterLd]>;
4267 let Predicates = [HasAVX] in
4268 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
4269 (outs GR32orGR64:$dst), (ins VR128:$src1, i32i8imm:$src2),
4270 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4271 [(set GR32orGR64:$dst, (X86pextrw (v8i16 VR128:$src1),
4272 imm:$src2))]>, TB, OpSize, VEX,
4273 Sched<[WriteShuffle]>;
4274 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
4275 (outs GR32orGR64:$dst), (ins VR128:$src1, i32i8imm:$src2),
4276 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4277 [(set GR32orGR64:$dst, (X86pextrw (v8i16 VR128:$src1),
4278 imm:$src2))], IIC_SSE_PEXTRW>,
4279 Sched<[WriteShuffleLd, ReadAfterLd]>;
4282 let Predicates = [HasAVX] in
4283 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
4285 let Predicates = [UseSSE2], Constraints = "$src1 = $dst" in
4286 defm PINSRW : sse2_pinsrw, TB, OpSize;
4288 } // ExeDomain = SSEPackedInt
4290 //===---------------------------------------------------------------------===//
4291 // SSE2 - Packed Mask Creation
4292 //===---------------------------------------------------------------------===//
4294 let ExeDomain = SSEPackedInt, SchedRW = [WriteVecLogic] in {
4296 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32orGR64:$dst),
4298 "pmovmskb\t{$src, $dst|$dst, $src}",
4299 [(set GR32orGR64:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4300 IIC_SSE_MOVMSK>, VEX;
4302 let Predicates = [HasAVX2] in {
4303 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32orGR64:$dst),
4305 "pmovmskb\t{$src, $dst|$dst, $src}",
4306 [(set GR32orGR64:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>,
4310 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32orGR64:$dst), (ins VR128:$src),
4311 "pmovmskb\t{$src, $dst|$dst, $src}",
4312 [(set GR32orGR64:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4315 } // ExeDomain = SSEPackedInt
4317 //===---------------------------------------------------------------------===//
4318 // SSE2 - Conditional Store
4319 //===---------------------------------------------------------------------===//
4321 let ExeDomain = SSEPackedInt, SchedRW = [WriteStore] in {
4323 let Uses = [EDI], Predicates = [HasAVX,In32BitMode] in
4324 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4325 (ins VR128:$src, VR128:$mask),
4326 "maskmovdqu\t{$mask, $src|$src, $mask}",
4327 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4328 IIC_SSE_MASKMOV>, VEX;
4329 let Uses = [RDI], Predicates = [HasAVX,In64BitMode] in
4330 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4331 (ins VR128:$src, VR128:$mask),
4332 "maskmovdqu\t{$mask, $src|$src, $mask}",
4333 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4334 IIC_SSE_MASKMOV>, VEX;
4336 let Uses = [EDI], Predicates = [UseSSE2,In32BitMode] in
4337 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4338 "maskmovdqu\t{$mask, $src|$src, $mask}",
4339 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4341 let Uses = [RDI], Predicates = [UseSSE2,In64BitMode] in
4342 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4343 "maskmovdqu\t{$mask, $src|$src, $mask}",
4344 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4347 } // ExeDomain = SSEPackedInt
4349 //===---------------------------------------------------------------------===//
4350 // SSE2 - Move Doubleword
4351 //===---------------------------------------------------------------------===//
4353 //===---------------------------------------------------------------------===//
4354 // Move Int Doubleword to Packed Double Int
4356 def VMOVDI2PDIrr : VS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4357 "movd\t{$src, $dst|$dst, $src}",
4359 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>,
4360 VEX, Sched<[WriteMove]>;
4361 def VMOVDI2PDIrm : VS2I<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4362 "movd\t{$src, $dst|$dst, $src}",
4364 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4366 VEX, Sched<[WriteLoad]>;
4367 def VMOV64toPQIrr : VRS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4368 "movq\t{$src, $dst|$dst, $src}",
4370 (v2i64 (scalar_to_vector GR64:$src)))],
4371 IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
4372 let isCodeGenOnly = 1 in
4373 def VMOV64toSDrr : VRS2I<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4374 "movq\t{$src, $dst|$dst, $src}",
4375 [(set FR64:$dst, (bitconvert GR64:$src))],
4376 IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
4378 def MOVDI2PDIrr : S2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4379 "movd\t{$src, $dst|$dst, $src}",
4381 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>,
4383 def MOVDI2PDIrm : S2I<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4384 "movd\t{$src, $dst|$dst, $src}",
4386 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4387 IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
4388 def MOV64toPQIrr : RS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4389 "mov{d|q}\t{$src, $dst|$dst, $src}",
4391 (v2i64 (scalar_to_vector GR64:$src)))],
4392 IIC_SSE_MOVDQ>, Sched<[WriteMove]>;
4393 let isCodeGenOnly = 1 in
4394 def MOV64toSDrr : RS2I<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4395 "mov{d|q}\t{$src, $dst|$dst, $src}",
4396 [(set FR64:$dst, (bitconvert GR64:$src))],
4397 IIC_SSE_MOVDQ>, Sched<[WriteMove]>;
4399 //===---------------------------------------------------------------------===//
4400 // Move Int Doubleword to Single Scalar
4402 let isCodeGenOnly = 1 in {
4403 def VMOVDI2SSrr : VS2I<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4404 "movd\t{$src, $dst|$dst, $src}",
4405 [(set FR32:$dst, (bitconvert GR32:$src))],
4406 IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
4408 def VMOVDI2SSrm : VS2I<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4409 "movd\t{$src, $dst|$dst, $src}",
4410 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4412 VEX, Sched<[WriteLoad]>;
4413 def MOVDI2SSrr : S2I<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4414 "movd\t{$src, $dst|$dst, $src}",
4415 [(set FR32:$dst, (bitconvert GR32:$src))],
4416 IIC_SSE_MOVDQ>, Sched<[WriteMove]>;
4418 def MOVDI2SSrm : S2I<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4419 "movd\t{$src, $dst|$dst, $src}",
4420 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4421 IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
4424 //===---------------------------------------------------------------------===//
4425 // Move Packed Doubleword Int to Packed Double Int
4427 def VMOVPDI2DIrr : VS2I<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4428 "movd\t{$src, $dst|$dst, $src}",
4429 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4430 (iPTR 0)))], IIC_SSE_MOVD_ToGP>, VEX,
4432 def VMOVPDI2DImr : VS2I<0x7E, MRMDestMem, (outs),
4433 (ins i32mem:$dst, VR128:$src),
4434 "movd\t{$src, $dst|$dst, $src}",
4435 [(store (i32 (vector_extract (v4i32 VR128:$src),
4436 (iPTR 0))), addr:$dst)], IIC_SSE_MOVDQ>,
4437 VEX, Sched<[WriteLoad]>;
4438 def MOVPDI2DIrr : S2I<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4439 "movd\t{$src, $dst|$dst, $src}",
4440 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4441 (iPTR 0)))], IIC_SSE_MOVD_ToGP>,
4443 def MOVPDI2DImr : S2I<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4444 "movd\t{$src, $dst|$dst, $src}",
4445 [(store (i32 (vector_extract (v4i32 VR128:$src),
4446 (iPTR 0))), addr:$dst)],
4447 IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
4449 def : Pat<(v8i32 (X86Vinsert (v8i32 immAllZerosV), GR32:$src2, (iPTR 0))),
4450 (SUBREG_TO_REG (i32 0), (VMOVDI2PDIrr GR32:$src2), sub_xmm)>;
4452 def : Pat<(v4i64 (X86Vinsert (bc_v4i64 (v8i32 immAllZerosV)), GR64:$src2, (iPTR 0))),
4453 (SUBREG_TO_REG (i32 0), (VMOV64toPQIrr GR64:$src2), sub_xmm)>;
4455 def : Pat<(v8i32 (X86Vinsert undef, GR32:$src2, (iPTR 0))),
4456 (SUBREG_TO_REG (i32 0), (VMOVDI2PDIrr GR32:$src2), sub_xmm)>;
4458 def : Pat<(v4i64 (X86Vinsert undef, GR64:$src2, (iPTR 0))),
4459 (SUBREG_TO_REG (i32 0), (VMOV64toPQIrr GR64:$src2), sub_xmm)>;
4461 //===---------------------------------------------------------------------===//
4462 // Move Packed Doubleword Int first element to Doubleword Int
4464 let SchedRW = [WriteMove] in {
4465 def VMOVPQIto64rr : VRS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4466 "movq\t{$src, $dst|$dst, $src}",
4467 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4472 def MOVPQIto64rr : RS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4473 "mov{d|q}\t{$src, $dst|$dst, $src}",
4474 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4479 //===---------------------------------------------------------------------===//
4480 // Bitcast FR64 <-> GR64
4482 let isCodeGenOnly = 1 in {
4483 let Predicates = [UseAVX] in
4484 def VMOV64toSDrm : VS2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4485 "movq\t{$src, $dst|$dst, $src}",
4486 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4487 VEX, Sched<[WriteLoad]>;
4488 def VMOVSDto64rr : VRS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4489 "movq\t{$src, $dst|$dst, $src}",
4490 [(set GR64:$dst, (bitconvert FR64:$src))],
4491 IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
4492 def VMOVSDto64mr : VRS2I<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4493 "movq\t{$src, $dst|$dst, $src}",
4494 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4495 IIC_SSE_MOVDQ>, VEX, Sched<[WriteStore]>;
4497 def MOV64toSDrm : S2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4498 "movq\t{$src, $dst|$dst, $src}",
4499 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))],
4500 IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
4501 def MOVSDto64rr : RS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4502 "mov{d|q}\t{$src, $dst|$dst, $src}",
4503 [(set GR64:$dst, (bitconvert FR64:$src))],
4504 IIC_SSE_MOVD_ToGP>, Sched<[WriteMove]>;
4505 def MOVSDto64mr : RS2I<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4506 "movq\t{$src, $dst|$dst, $src}",
4507 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4508 IIC_SSE_MOVDQ>, Sched<[WriteStore]>;
4511 //===---------------------------------------------------------------------===//
4512 // Move Scalar Single to Double Int
4514 let isCodeGenOnly = 1 in {
4515 def VMOVSS2DIrr : VS2I<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4516 "movd\t{$src, $dst|$dst, $src}",
4517 [(set GR32:$dst, (bitconvert FR32:$src))],
4518 IIC_SSE_MOVD_ToGP>, VEX, Sched<[WriteMove]>;
4519 def VMOVSS2DImr : VS2I<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4520 "movd\t{$src, $dst|$dst, $src}",
4521 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4522 IIC_SSE_MOVDQ>, VEX, Sched<[WriteStore]>;
4523 def MOVSS2DIrr : S2I<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4524 "movd\t{$src, $dst|$dst, $src}",
4525 [(set GR32:$dst, (bitconvert FR32:$src))],
4526 IIC_SSE_MOVD_ToGP>, Sched<[WriteMove]>;
4527 def MOVSS2DImr : S2I<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4528 "movd\t{$src, $dst|$dst, $src}",
4529 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4530 IIC_SSE_MOVDQ>, Sched<[WriteStore]>;
4533 //===---------------------------------------------------------------------===//
4534 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4536 let isCodeGenOnly = 1, SchedRW = [WriteMove] in {
4537 let AddedComplexity = 15 in {
4538 def VMOVZQI2PQIrr : VS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4539 "movq\t{$src, $dst|$dst, $src}", // X86-64 only
4540 [(set VR128:$dst, (v2i64 (X86vzmovl
4541 (v2i64 (scalar_to_vector GR64:$src)))))],
4544 def MOVZQI2PQIrr : RS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4545 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4546 [(set VR128:$dst, (v2i64 (X86vzmovl
4547 (v2i64 (scalar_to_vector GR64:$src)))))],
4550 } // isCodeGenOnly, SchedRW
4552 let Predicates = [UseAVX] in {
4553 let AddedComplexity = 15 in
4554 def : Pat<(v4i32 (X86vzmovl (v4i32 (scalar_to_vector GR32:$src)))),
4555 (VMOVDI2PDIrr GR32:$src)>;
4557 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4558 let AddedComplexity = 20 in {
4559 def : Pat<(v4i32 (X86vzmovl (v4i32 (scalar_to_vector (loadi32 addr:$src))))),
4560 (VMOVDI2PDIrm addr:$src)>;
4561 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4562 (VMOVDI2PDIrm addr:$src)>;
4563 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4564 (VMOVDI2PDIrm addr:$src)>;
4566 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4567 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4568 (v4i32 (scalar_to_vector GR32:$src)),(iPTR 0)))),
4569 (SUBREG_TO_REG (i32 0), (VMOVDI2PDIrr GR32:$src), sub_xmm)>;
4570 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4571 (v2i64 (scalar_to_vector GR64:$src)),(iPTR 0)))),
4572 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4575 let Predicates = [UseSSE2] in {
4576 let AddedComplexity = 15 in
4577 def : Pat<(v4i32 (X86vzmovl (v4i32 (scalar_to_vector GR32:$src)))),
4578 (MOVDI2PDIrr GR32:$src)>;
4580 let AddedComplexity = 20 in {
4581 def : Pat<(v4i32 (X86vzmovl (v4i32 (scalar_to_vector (loadi32 addr:$src))))),
4582 (MOVDI2PDIrm addr:$src)>;
4583 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4584 (MOVDI2PDIrm addr:$src)>;
4585 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4586 (MOVDI2PDIrm addr:$src)>;
4590 // These are the correct encodings of the instructions so that we know how to
4591 // read correct assembly, even though we continue to emit the wrong ones for
4592 // compatibility with Darwin's buggy assembler.
4593 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4594 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4595 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4596 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4597 // Allow "vmovd" but print "vmovq" since we don't need compatibility for AVX.
4598 def : InstAlias<"vmovd\t{$src, $dst|$dst, $src}",
4599 (VMOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4600 def : InstAlias<"vmovd\t{$src, $dst|$dst, $src}",
4601 (VMOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4603 //===---------------------------------------------------------------------===//
4604 // SSE2 - Move Quadword
4605 //===---------------------------------------------------------------------===//
4607 //===---------------------------------------------------------------------===//
4608 // Move Quadword Int to Packed Quadword Int
4611 let SchedRW = [WriteLoad] in {
4612 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4613 "vmovq\t{$src, $dst|$dst, $src}",
4615 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4616 VEX, Requires<[UseAVX]>;
4617 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4618 "movq\t{$src, $dst|$dst, $src}",
4620 (v2i64 (scalar_to_vector (loadi64 addr:$src))))],
4622 Requires<[UseSSE2]>; // SSE2 instruction with XS Prefix
4625 //===---------------------------------------------------------------------===//
4626 // Move Packed Quadword Int to Quadword Int
4628 let SchedRW = [WriteStore] in {
4629 def VMOVPQI2QImr : VS2I<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4630 "movq\t{$src, $dst|$dst, $src}",
4631 [(store (i64 (vector_extract (v2i64 VR128:$src),
4632 (iPTR 0))), addr:$dst)],
4633 IIC_SSE_MOVDQ>, VEX;
4634 def MOVPQI2QImr : S2I<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4635 "movq\t{$src, $dst|$dst, $src}",
4636 [(store (i64 (vector_extract (v2i64 VR128:$src),
4637 (iPTR 0))), addr:$dst)],
4641 //===---------------------------------------------------------------------===//
4642 // Store / copy lower 64-bits of a XMM register.
4644 def VMOVLQ128mr : VS2I<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4645 "movq\t{$src, $dst|$dst, $src}",
4646 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX,
4647 Sched<[WriteStore]>;
4648 def MOVLQ128mr : S2I<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4649 "movq\t{$src, $dst|$dst, $src}",
4650 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)],
4651 IIC_SSE_MOVDQ>, Sched<[WriteStore]>;
4653 let isCodeGenOnly = 1, AddedComplexity = 20 in {
4654 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4655 "vmovq\t{$src, $dst|$dst, $src}",
4657 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4658 (loadi64 addr:$src))))))],
4660 XS, VEX, Requires<[UseAVX]>, Sched<[WriteLoad]>;
4662 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4663 "movq\t{$src, $dst|$dst, $src}",
4665 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4666 (loadi64 addr:$src))))))],
4668 XS, Requires<[UseSSE2]>, Sched<[WriteLoad]>;
4671 let Predicates = [UseAVX], AddedComplexity = 20 in {
4672 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4673 (VMOVZQI2PQIrm addr:$src)>;
4674 def : Pat<(v2i64 (X86vzload addr:$src)),
4675 (VMOVZQI2PQIrm addr:$src)>;
4678 let Predicates = [UseSSE2], AddedComplexity = 20 in {
4679 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4680 (MOVZQI2PQIrm addr:$src)>;
4681 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4684 let Predicates = [HasAVX] in {
4685 def : Pat<(v4i64 (alignedX86vzload addr:$src)),
4686 (SUBREG_TO_REG (i32 0), (VMOVAPSrm addr:$src), sub_xmm)>;
4687 def : Pat<(v4i64 (X86vzload addr:$src)),
4688 (SUBREG_TO_REG (i32 0), (VMOVUPSrm addr:$src), sub_xmm)>;
4691 //===---------------------------------------------------------------------===//
4692 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4693 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4695 let SchedRW = [WriteVecLogic] in {
4696 let AddedComplexity = 15 in
4697 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4698 "vmovq\t{$src, $dst|$dst, $src}",
4699 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4701 XS, VEX, Requires<[UseAVX]>;
4702 let AddedComplexity = 15 in
4703 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4704 "movq\t{$src, $dst|$dst, $src}",
4705 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4707 XS, Requires<[UseSSE2]>;
4710 let isCodeGenOnly = 1, SchedRW = [WriteVecLogicLd] in {
4711 let AddedComplexity = 20 in
4712 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4713 "vmovq\t{$src, $dst|$dst, $src}",
4714 [(set VR128:$dst, (v2i64 (X86vzmovl
4715 (loadv2i64 addr:$src))))],
4717 XS, VEX, Requires<[UseAVX]>;
4718 let AddedComplexity = 20 in {
4719 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4720 "movq\t{$src, $dst|$dst, $src}",
4721 [(set VR128:$dst, (v2i64 (X86vzmovl
4722 (loadv2i64 addr:$src))))],
4724 XS, Requires<[UseSSE2]>;
4726 } // isCodeGenOnly, SchedRW
4728 let AddedComplexity = 20 in {
4729 let Predicates = [UseAVX] in {
4730 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4731 (VMOVZPQILo2PQIrr VR128:$src)>;
4733 let Predicates = [UseSSE2] in {
4734 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4735 (MOVZPQILo2PQIrr VR128:$src)>;
4739 //===---------------------------------------------------------------------===//
4740 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4741 //===---------------------------------------------------------------------===//
4742 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4743 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4744 X86MemOperand x86memop> {
4745 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4746 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4747 [(set RC:$dst, (vt (OpNode RC:$src)))],
4748 IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>;
4749 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4750 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4751 [(set RC:$dst, (OpNode (mem_frag addr:$src)))],
4752 IIC_SSE_MOV_LH>, Sched<[WriteShuffleLd]>;
4755 let Predicates = [HasAVX] in {
4756 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4757 v4f32, VR128, memopv4f32, f128mem>, VEX;
4758 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4759 v4f32, VR128, memopv4f32, f128mem>, VEX;
4760 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4761 v8f32, VR256, memopv8f32, f256mem>, VEX, VEX_L;
4762 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4763 v8f32, VR256, memopv8f32, f256mem>, VEX, VEX_L;
4765 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4766 memopv4f32, f128mem>;
4767 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4768 memopv4f32, f128mem>;
4770 let Predicates = [HasAVX] in {
4771 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4772 (VMOVSHDUPrr VR128:$src)>;
4773 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4774 (VMOVSHDUPrm addr:$src)>;
4775 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4776 (VMOVSLDUPrr VR128:$src)>;
4777 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4778 (VMOVSLDUPrm addr:$src)>;
4779 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4780 (VMOVSHDUPYrr VR256:$src)>;
4781 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4782 (VMOVSHDUPYrm addr:$src)>;
4783 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4784 (VMOVSLDUPYrr VR256:$src)>;
4785 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
4786 (VMOVSLDUPYrm addr:$src)>;
4789 let Predicates = [UseSSE3] in {
4790 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4791 (MOVSHDUPrr VR128:$src)>;
4792 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4793 (MOVSHDUPrm addr:$src)>;
4794 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4795 (MOVSLDUPrr VR128:$src)>;
4796 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4797 (MOVSLDUPrm addr:$src)>;
4800 //===---------------------------------------------------------------------===//
4801 // SSE3 - Replicate Double FP - MOVDDUP
4802 //===---------------------------------------------------------------------===//
4804 multiclass sse3_replicate_dfp<string OpcodeStr> {
4805 let neverHasSideEffects = 1 in
4806 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4807 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4808 [], IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>;
4809 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
4810 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4813 (scalar_to_vector (loadf64 addr:$src)))))],
4814 IIC_SSE_MOV_LH>, Sched<[WriteShuffleLd]>;
4817 // FIXME: Merge with above classe when there're patterns for the ymm version
4818 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
4819 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
4820 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4821 [(set VR256:$dst, (v4f64 (X86Movddup VR256:$src)))]>,
4822 Sched<[WriteShuffle]>;
4823 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
4824 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4827 (scalar_to_vector (loadf64 addr:$src)))))]>,
4828 Sched<[WriteShuffleLd]>;
4831 let Predicates = [HasAVX] in {
4832 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
4833 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX, VEX_L;
4836 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
4838 let Predicates = [HasAVX] in {
4839 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4840 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4841 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4842 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4843 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4844 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4845 def : Pat<(X86Movddup (bc_v2f64
4846 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4847 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4850 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
4851 (VMOVDDUPYrm addr:$src)>;
4852 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
4853 (VMOVDDUPYrm addr:$src)>;
4854 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
4855 (VMOVDDUPYrm addr:$src)>;
4856 def : Pat<(X86Movddup (v4i64 VR256:$src)),
4857 (VMOVDDUPYrr VR256:$src)>;
4860 let Predicates = [UseSSE3] in {
4861 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4862 (MOVDDUPrm addr:$src)>;
4863 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4864 (MOVDDUPrm addr:$src)>;
4865 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4866 (MOVDDUPrm addr:$src)>;
4867 def : Pat<(X86Movddup (bc_v2f64
4868 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4869 (MOVDDUPrm addr:$src)>;
4872 //===---------------------------------------------------------------------===//
4873 // SSE3 - Move Unaligned Integer
4874 //===---------------------------------------------------------------------===//
4876 let SchedRW = [WriteLoad] in {
4877 let Predicates = [HasAVX] in {
4878 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4879 "vlddqu\t{$src, $dst|$dst, $src}",
4880 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
4881 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
4882 "vlddqu\t{$src, $dst|$dst, $src}",
4883 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>,
4886 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4887 "lddqu\t{$src, $dst|$dst, $src}",
4888 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))],
4892 //===---------------------------------------------------------------------===//
4893 // SSE3 - Arithmetic
4894 //===---------------------------------------------------------------------===//
4896 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
4897 X86MemOperand x86memop, OpndItins itins,
4899 def rr : I<0xD0, MRMSrcReg,
4900 (outs RC:$dst), (ins RC:$src1, RC:$src2),
4902 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4903 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4904 [(set RC:$dst, (Int RC:$src1, RC:$src2))], itins.rr>,
4905 Sched<[itins.Sched]>;
4906 def rm : I<0xD0, MRMSrcMem,
4907 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4909 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4910 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4911 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))], itins.rr>,
4912 Sched<[itins.Sched.Folded, ReadAfterLd]>;
4915 let Predicates = [HasAVX] in {
4916 let ExeDomain = SSEPackedSingle in {
4917 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
4918 f128mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V;
4919 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
4920 f256mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V, VEX_L;
4922 let ExeDomain = SSEPackedDouble in {
4923 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
4924 f128mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V;
4925 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
4926 f256mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V, VEX_L;
4929 let Constraints = "$src1 = $dst", Predicates = [UseSSE3] in {
4930 let ExeDomain = SSEPackedSingle in
4931 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
4932 f128mem, SSE_ALU_F32P>, TB, XD;
4933 let ExeDomain = SSEPackedDouble in
4934 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
4935 f128mem, SSE_ALU_F64P>, TB, OpSize;
4938 //===---------------------------------------------------------------------===//
4939 // SSE3 Instructions
4940 //===---------------------------------------------------------------------===//
4943 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4944 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
4945 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4947 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4948 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4949 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>,
4952 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4954 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4955 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4956 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
4957 IIC_SSE_HADDSUB_RM>, Sched<[WriteFAddLd, ReadAfterLd]>;
4959 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4960 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
4961 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4963 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4964 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4965 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>,
4968 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4970 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4971 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4972 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
4973 IIC_SSE_HADDSUB_RM>, Sched<[WriteFAddLd, ReadAfterLd]>;
4976 let Predicates = [HasAVX] in {
4977 let ExeDomain = SSEPackedSingle in {
4978 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
4979 X86fhadd, 0>, VEX_4V;
4980 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
4981 X86fhsub, 0>, VEX_4V;
4982 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
4983 X86fhadd, 0>, VEX_4V, VEX_L;
4984 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
4985 X86fhsub, 0>, VEX_4V, VEX_L;
4987 let ExeDomain = SSEPackedDouble in {
4988 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
4989 X86fhadd, 0>, VEX_4V;
4990 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
4991 X86fhsub, 0>, VEX_4V;
4992 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
4993 X86fhadd, 0>, VEX_4V, VEX_L;
4994 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
4995 X86fhsub, 0>, VEX_4V, VEX_L;
4999 let Constraints = "$src1 = $dst" in {
5000 let ExeDomain = SSEPackedSingle in {
5001 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
5002 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
5004 let ExeDomain = SSEPackedDouble in {
5005 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
5006 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
5010 //===---------------------------------------------------------------------===//
5011 // SSSE3 - Packed Absolute Instructions
5012 //===---------------------------------------------------------------------===//
5015 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5016 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
5017 Intrinsic IntId128> {
5018 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5020 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5021 [(set VR128:$dst, (IntId128 VR128:$src))], IIC_SSE_PABS_RR>,
5022 OpSize, Sched<[WriteVecALU]>;
5024 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5026 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5029 (bitconvert (memopv2i64 addr:$src))))], IIC_SSE_PABS_RM>,
5030 OpSize, Sched<[WriteVecALULd]>;
5033 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5034 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
5035 Intrinsic IntId256> {
5036 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5038 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5039 [(set VR256:$dst, (IntId256 VR256:$src))]>,
5040 OpSize, Sched<[WriteVecALU]>;
5042 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5044 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5047 (bitconvert (memopv4i64 addr:$src))))]>, OpSize,
5048 Sched<[WriteVecALULd]>;
5051 // Helper fragments to match sext vXi1 to vXiY.
5052 def v16i1sextv16i8 : PatLeaf<(v16i8 (X86pcmpgt (bc_v16i8 (v4i32 immAllZerosV)),
5054 def v8i1sextv8i16 : PatLeaf<(v8i16 (X86vsrai VR128:$src, (i8 15)))>;
5055 def v4i1sextv4i32 : PatLeaf<(v4i32 (X86vsrai VR128:$src, (i8 31)))>;
5056 def v32i1sextv32i8 : PatLeaf<(v32i8 (X86pcmpgt (bc_v32i8 (v8i32 immAllZerosV)),
5058 def v16i1sextv16i16: PatLeaf<(v16i16 (X86vsrai VR256:$src, (i8 15)))>;
5059 def v8i1sextv8i32 : PatLeaf<(v8i32 (X86vsrai VR256:$src, (i8 31)))>;
5061 let Predicates = [HasAVX] in {
5062 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb",
5063 int_x86_ssse3_pabs_b_128>, VEX;
5064 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw",
5065 int_x86_ssse3_pabs_w_128>, VEX;
5066 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd",
5067 int_x86_ssse3_pabs_d_128>, VEX;
5070 (bc_v2i64 (v16i1sextv16i8)),
5071 (bc_v2i64 (add (v16i8 VR128:$src), (v16i1sextv16i8)))),
5072 (VPABSBrr128 VR128:$src)>;
5074 (bc_v2i64 (v8i1sextv8i16)),
5075 (bc_v2i64 (add (v8i16 VR128:$src), (v8i1sextv8i16)))),
5076 (VPABSWrr128 VR128:$src)>;
5078 (bc_v2i64 (v4i1sextv4i32)),
5079 (bc_v2i64 (add (v4i32 VR128:$src), (v4i1sextv4i32)))),
5080 (VPABSDrr128 VR128:$src)>;
5083 let Predicates = [HasAVX2] in {
5084 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb",
5085 int_x86_avx2_pabs_b>, VEX, VEX_L;
5086 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw",
5087 int_x86_avx2_pabs_w>, VEX, VEX_L;
5088 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd",
5089 int_x86_avx2_pabs_d>, VEX, VEX_L;
5092 (bc_v4i64 (v32i1sextv32i8)),
5093 (bc_v4i64 (add (v32i8 VR256:$src), (v32i1sextv32i8)))),
5094 (VPABSBrr256 VR256:$src)>;
5096 (bc_v4i64 (v16i1sextv16i16)),
5097 (bc_v4i64 (add (v16i16 VR256:$src), (v16i1sextv16i16)))),
5098 (VPABSWrr256 VR256:$src)>;
5100 (bc_v4i64 (v8i1sextv8i32)),
5101 (bc_v4i64 (add (v8i32 VR256:$src), (v8i1sextv8i32)))),
5102 (VPABSDrr256 VR256:$src)>;
5105 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb",
5106 int_x86_ssse3_pabs_b_128>;
5107 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw",
5108 int_x86_ssse3_pabs_w_128>;
5109 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
5110 int_x86_ssse3_pabs_d_128>;
5112 let Predicates = [HasSSSE3] in {
5114 (bc_v2i64 (v16i1sextv16i8)),
5115 (bc_v2i64 (add (v16i8 VR128:$src), (v16i1sextv16i8)))),
5116 (PABSBrr128 VR128:$src)>;
5118 (bc_v2i64 (v8i1sextv8i16)),
5119 (bc_v2i64 (add (v8i16 VR128:$src), (v8i1sextv8i16)))),
5120 (PABSWrr128 VR128:$src)>;
5122 (bc_v2i64 (v4i1sextv4i32)),
5123 (bc_v2i64 (add (v4i32 VR128:$src), (v4i1sextv4i32)))),
5124 (PABSDrr128 VR128:$src)>;
5127 //===---------------------------------------------------------------------===//
5128 // SSSE3 - Packed Binary Operator Instructions
5129 //===---------------------------------------------------------------------===//
5131 let Sched = WriteVecALU in {
5132 def SSE_PHADDSUBD : OpndItins<
5133 IIC_SSE_PHADDSUBD_RR, IIC_SSE_PHADDSUBD_RM
5135 def SSE_PHADDSUBSW : OpndItins<
5136 IIC_SSE_PHADDSUBSW_RR, IIC_SSE_PHADDSUBSW_RM
5138 def SSE_PHADDSUBW : OpndItins<
5139 IIC_SSE_PHADDSUBW_RR, IIC_SSE_PHADDSUBW_RM
5142 let Sched = WriteShuffle in
5143 def SSE_PSHUFB : OpndItins<
5144 IIC_SSE_PSHUFB_RR, IIC_SSE_PSHUFB_RM
5146 let Sched = WriteVecALU in
5147 def SSE_PSIGN : OpndItins<
5148 IIC_SSE_PSIGN_RR, IIC_SSE_PSIGN_RM
5150 let Sched = WriteVecIMul in
5151 def SSE_PMULHRSW : OpndItins<
5152 IIC_SSE_PMULHRSW, IIC_SSE_PMULHRSW
5155 /// SS3I_binop_rm - Simple SSSE3 bin op
5156 multiclass SS3I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
5157 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
5158 X86MemOperand x86memop, OpndItins itins,
5160 let isCommutable = 1 in
5161 def rr : SS38I<opc, MRMSrcReg, (outs RC:$dst),
5162 (ins RC:$src1, RC:$src2),
5164 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5165 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5166 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>,
5167 OpSize, Sched<[itins.Sched]>;
5168 def rm : SS38I<opc, MRMSrcMem, (outs RC:$dst),
5169 (ins RC:$src1, x86memop:$src2),
5171 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5172 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5174 (OpVT (OpNode RC:$src1,
5175 (bitconvert (memop_frag addr:$src2)))))], itins.rm>, OpSize,
5176 Sched<[itins.Sched.Folded, ReadAfterLd]>;
5179 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
5180 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
5181 Intrinsic IntId128, OpndItins itins,
5183 let isCommutable = 1 in
5184 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5185 (ins VR128:$src1, VR128:$src2),
5187 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5188 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5189 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5190 OpSize, Sched<[itins.Sched]>;
5191 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5192 (ins VR128:$src1, i128mem:$src2),
5194 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5195 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5197 (IntId128 VR128:$src1,
5198 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize,
5199 Sched<[itins.Sched.Folded, ReadAfterLd]>;
5202 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
5203 Intrinsic IntId256> {
5204 let isCommutable = 1 in
5205 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5206 (ins VR256:$src1, VR256:$src2),
5207 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5208 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
5210 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5211 (ins VR256:$src1, i256mem:$src2),
5212 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5214 (IntId256 VR256:$src1,
5215 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
5218 let ImmT = NoImm, Predicates = [HasAVX] in {
5219 let isCommutable = 0 in {
5220 defm VPHADDW : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v8i16, VR128,
5221 memopv2i64, i128mem,
5222 SSE_PHADDSUBW, 0>, VEX_4V;
5223 defm VPHADDD : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v4i32, VR128,
5224 memopv2i64, i128mem,
5225 SSE_PHADDSUBD, 0>, VEX_4V;
5226 defm VPHSUBW : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v8i16, VR128,
5227 memopv2i64, i128mem,
5228 SSE_PHADDSUBW, 0>, VEX_4V;
5229 defm VPHSUBD : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v4i32, VR128,
5230 memopv2i64, i128mem,
5231 SSE_PHADDSUBD, 0>, VEX_4V;
5232 defm VPSIGNB : SS3I_binop_rm<0x08, "vpsignb", X86psign, v16i8, VR128,
5233 memopv2i64, i128mem,
5234 SSE_PSIGN, 0>, VEX_4V;
5235 defm VPSIGNW : SS3I_binop_rm<0x09, "vpsignw", X86psign, v8i16, VR128,
5236 memopv2i64, i128mem,
5237 SSE_PSIGN, 0>, VEX_4V;
5238 defm VPSIGND : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v4i32, VR128,
5239 memopv2i64, i128mem,
5240 SSE_PSIGN, 0>, VEX_4V;
5241 defm VPSHUFB : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v16i8, VR128,
5242 memopv2i64, i128mem,
5243 SSE_PSHUFB, 0>, VEX_4V;
5244 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw",
5245 int_x86_ssse3_phadd_sw_128,
5246 SSE_PHADDSUBSW, 0>, VEX_4V;
5247 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw",
5248 int_x86_ssse3_phsub_sw_128,
5249 SSE_PHADDSUBSW, 0>, VEX_4V;
5250 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw",
5251 int_x86_ssse3_pmadd_ub_sw_128,
5252 SSE_PMADD, 0>, VEX_4V;
5254 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw",
5255 int_x86_ssse3_pmul_hr_sw_128,
5256 SSE_PMULHRSW, 0>, VEX_4V;
5259 let ImmT = NoImm, Predicates = [HasAVX2] in {
5260 let isCommutable = 0 in {
5261 defm VPHADDWY : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v16i16, VR256,
5262 memopv4i64, i256mem,
5263 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5264 defm VPHADDDY : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v8i32, VR256,
5265 memopv4i64, i256mem,
5266 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5267 defm VPHSUBWY : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v16i16, VR256,
5268 memopv4i64, i256mem,
5269 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5270 defm VPHSUBDY : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v8i32, VR256,
5271 memopv4i64, i256mem,
5272 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5273 defm VPSIGNBY : SS3I_binop_rm<0x08, "vpsignb", X86psign, v32i8, VR256,
5274 memopv4i64, i256mem,
5275 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5276 defm VPSIGNWY : SS3I_binop_rm<0x09, "vpsignw", X86psign, v16i16, VR256,
5277 memopv4i64, i256mem,
5278 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5279 defm VPSIGNDY : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v8i32, VR256,
5280 memopv4i64, i256mem,
5281 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5282 defm VPSHUFBY : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v32i8, VR256,
5283 memopv4i64, i256mem,
5284 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5285 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
5286 int_x86_avx2_phadd_sw>, VEX_4V, VEX_L;
5287 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
5288 int_x86_avx2_phsub_sw>, VEX_4V, VEX_L;
5289 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
5290 int_x86_avx2_pmadd_ub_sw>, VEX_4V, VEX_L;
5292 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
5293 int_x86_avx2_pmul_hr_sw>, VEX_4V, VEX_L;
5296 // None of these have i8 immediate fields.
5297 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
5298 let isCommutable = 0 in {
5299 defm PHADDW : SS3I_binop_rm<0x01, "phaddw", X86hadd, v8i16, VR128,
5300 memopv2i64, i128mem, SSE_PHADDSUBW>;
5301 defm PHADDD : SS3I_binop_rm<0x02, "phaddd", X86hadd, v4i32, VR128,
5302 memopv2i64, i128mem, SSE_PHADDSUBD>;
5303 defm PHSUBW : SS3I_binop_rm<0x05, "phsubw", X86hsub, v8i16, VR128,
5304 memopv2i64, i128mem, SSE_PHADDSUBW>;
5305 defm PHSUBD : SS3I_binop_rm<0x06, "phsubd", X86hsub, v4i32, VR128,
5306 memopv2i64, i128mem, SSE_PHADDSUBD>;
5307 defm PSIGNB : SS3I_binop_rm<0x08, "psignb", X86psign, v16i8, VR128,
5308 memopv2i64, i128mem, SSE_PSIGN>;
5309 defm PSIGNW : SS3I_binop_rm<0x09, "psignw", X86psign, v8i16, VR128,
5310 memopv2i64, i128mem, SSE_PSIGN>;
5311 defm PSIGND : SS3I_binop_rm<0x0A, "psignd", X86psign, v4i32, VR128,
5312 memopv2i64, i128mem, SSE_PSIGN>;
5313 defm PSHUFB : SS3I_binop_rm<0x00, "pshufb", X86pshufb, v16i8, VR128,
5314 memopv2i64, i128mem, SSE_PSHUFB>;
5315 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw",
5316 int_x86_ssse3_phadd_sw_128,
5318 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw",
5319 int_x86_ssse3_phsub_sw_128,
5321 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw",
5322 int_x86_ssse3_pmadd_ub_sw_128, SSE_PMADD>;
5324 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw",
5325 int_x86_ssse3_pmul_hr_sw_128,
5329 //===---------------------------------------------------------------------===//
5330 // SSSE3 - Packed Align Instruction Patterns
5331 //===---------------------------------------------------------------------===//
5333 multiclass ssse3_palignr<string asm, bit Is2Addr = 1> {
5334 let neverHasSideEffects = 1 in {
5335 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
5336 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5338 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5340 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5341 [], IIC_SSE_PALIGNRR>, OpSize, Sched<[WriteShuffle]>;
5343 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
5344 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5346 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5348 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5349 [], IIC_SSE_PALIGNRM>, OpSize, Sched<[WriteShuffleLd, ReadAfterLd]>;
5353 multiclass ssse3_palignr_y<string asm, bit Is2Addr = 1> {
5354 let neverHasSideEffects = 1 in {
5355 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
5356 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5358 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5359 []>, OpSize, Sched<[WriteShuffle]>;
5361 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
5362 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
5364 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5365 []>, OpSize, Sched<[WriteShuffleLd, ReadAfterLd]>;
5369 let Predicates = [HasAVX] in
5370 defm VPALIGN : ssse3_palignr<"vpalignr", 0>, VEX_4V;
5371 let Predicates = [HasAVX2] in
5372 defm VPALIGN : ssse3_palignr_y<"vpalignr", 0>, VEX_4V, VEX_L;
5373 let Constraints = "$src1 = $dst", Predicates = [UseSSSE3] in
5374 defm PALIGN : ssse3_palignr<"palignr">;
5376 let Predicates = [HasAVX2] in {
5377 def : Pat<(v8i32 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5378 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5379 def : Pat<(v8f32 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5380 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5381 def : Pat<(v16i16 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5382 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5383 def : Pat<(v32i8 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5384 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5387 let Predicates = [HasAVX] in {
5388 def : Pat<(v4i32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5389 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5390 def : Pat<(v4f32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5391 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5392 def : Pat<(v8i16 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5393 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5394 def : Pat<(v16i8 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5395 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5398 let Predicates = [UseSSSE3] in {
5399 def : Pat<(v4i32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5400 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5401 def : Pat<(v4f32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5402 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5403 def : Pat<(v8i16 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5404 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5405 def : Pat<(v16i8 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5406 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5409 //===---------------------------------------------------------------------===//
5410 // SSSE3 - Thread synchronization
5411 //===---------------------------------------------------------------------===//
5413 let SchedRW = [WriteSystem] in {
5414 let usesCustomInserter = 1 in {
5415 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5416 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>,
5417 Requires<[HasSSE3]>;
5420 let Uses = [EAX, ECX, EDX] in
5421 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", [], IIC_SSE_MONITOR>,
5422 TB, Requires<[HasSSE3]>;
5423 let Uses = [ECX, EAX] in
5424 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait",
5425 [(int_x86_sse3_mwait ECX, EAX)], IIC_SSE_MWAIT>,
5426 TB, Requires<[HasSSE3]>;
5429 def : InstAlias<"mwait\t{%eax, %ecx|ecx, eax}", (MWAITrr)>, Requires<[In32BitMode]>;
5430 def : InstAlias<"mwait\t{%rax, %rcx|rcx, rax}", (MWAITrr)>, Requires<[In64BitMode]>;
5432 def : InstAlias<"monitor\t{%eax, %ecx, %edx|edx, ecx, eax}", (MONITORrrr)>,
5433 Requires<[In32BitMode]>;
5434 def : InstAlias<"monitor\t{%rax, %rcx, %rdx|rdx, rcx, rax}", (MONITORrrr)>,
5435 Requires<[In64BitMode]>;
5437 //===----------------------------------------------------------------------===//
5438 // SSE4.1 - Packed Move with Sign/Zero Extend
5439 //===----------------------------------------------------------------------===//
5441 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId,
5442 OpndItins itins = DEFAULT_ITINS> {
5443 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5444 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5445 [(set VR128:$dst, (IntId VR128:$src))], itins.rr>, OpSize;
5447 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5448 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5450 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))],
5454 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5456 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5457 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5458 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5460 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5461 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5462 [(set VR256:$dst, (IntId (load addr:$src)))]>,
5466 let Predicates = [HasAVX] in {
5467 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw",
5468 int_x86_sse41_pmovsxbw>, VEX;
5469 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd",
5470 int_x86_sse41_pmovsxwd>, VEX;
5471 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq",
5472 int_x86_sse41_pmovsxdq>, VEX;
5473 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw",
5474 int_x86_sse41_pmovzxbw>, VEX;
5475 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd",
5476 int_x86_sse41_pmovzxwd>, VEX;
5477 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq",
5478 int_x86_sse41_pmovzxdq>, VEX;
5481 let Predicates = [HasAVX2] in {
5482 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5483 int_x86_avx2_pmovsxbw>, VEX, VEX_L;
5484 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5485 int_x86_avx2_pmovsxwd>, VEX, VEX_L;
5486 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5487 int_x86_avx2_pmovsxdq>, VEX, VEX_L;
5488 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5489 int_x86_avx2_pmovzxbw>, VEX, VEX_L;
5490 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5491 int_x86_avx2_pmovzxwd>, VEX, VEX_L;
5492 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5493 int_x86_avx2_pmovzxdq>, VEX, VEX_L;
5496 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw, SSE_INTALU_ITINS_P>;
5497 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd, SSE_INTALU_ITINS_P>;
5498 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq, SSE_INTALU_ITINS_P>;
5499 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw, SSE_INTALU_ITINS_P>;
5500 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd, SSE_INTALU_ITINS_P>;
5501 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq, SSE_INTALU_ITINS_P>;
5503 let Predicates = [HasAVX] in {
5504 // Common patterns involving scalar load.
5505 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5506 (VPMOVSXBWrm addr:$src)>;
5507 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5508 (VPMOVSXBWrm addr:$src)>;
5509 def : Pat<(int_x86_sse41_pmovsxbw (bc_v16i8 (loadv2i64 addr:$src))),
5510 (VPMOVSXBWrm addr:$src)>;
5512 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5513 (VPMOVSXWDrm addr:$src)>;
5514 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5515 (VPMOVSXWDrm addr:$src)>;
5516 def : Pat<(int_x86_sse41_pmovsxwd (bc_v8i16 (loadv2i64 addr:$src))),
5517 (VPMOVSXWDrm addr:$src)>;
5519 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5520 (VPMOVSXDQrm addr:$src)>;
5521 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5522 (VPMOVSXDQrm addr:$src)>;
5523 def : Pat<(int_x86_sse41_pmovsxdq (bc_v4i32 (loadv2i64 addr:$src))),
5524 (VPMOVSXDQrm addr:$src)>;
5526 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5527 (VPMOVZXBWrm addr:$src)>;
5528 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5529 (VPMOVZXBWrm addr:$src)>;
5530 def : Pat<(int_x86_sse41_pmovzxbw (bc_v16i8 (loadv2i64 addr:$src))),
5531 (VPMOVZXBWrm addr:$src)>;
5533 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5534 (VPMOVZXWDrm addr:$src)>;
5535 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5536 (VPMOVZXWDrm addr:$src)>;
5537 def : Pat<(int_x86_sse41_pmovzxwd (bc_v8i16 (loadv2i64 addr:$src))),
5538 (VPMOVZXWDrm addr:$src)>;
5540 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5541 (VPMOVZXDQrm addr:$src)>;
5542 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5543 (VPMOVZXDQrm addr:$src)>;
5544 def : Pat<(int_x86_sse41_pmovzxdq (bc_v4i32 (loadv2i64 addr:$src))),
5545 (VPMOVZXDQrm addr:$src)>;
5548 let Predicates = [UseSSE41] in {
5549 // Common patterns involving scalar load.
5550 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5551 (PMOVSXBWrm addr:$src)>;
5552 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5553 (PMOVSXBWrm addr:$src)>;
5554 def : Pat<(int_x86_sse41_pmovsxbw (bc_v16i8 (loadv2i64 addr:$src))),
5555 (PMOVSXBWrm addr:$src)>;
5557 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5558 (PMOVSXWDrm addr:$src)>;
5559 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5560 (PMOVSXWDrm addr:$src)>;
5561 def : Pat<(int_x86_sse41_pmovsxwd (bc_v8i16 (loadv2i64 addr:$src))),
5562 (PMOVSXWDrm addr:$src)>;
5564 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5565 (PMOVSXDQrm addr:$src)>;
5566 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5567 (PMOVSXDQrm addr:$src)>;
5568 def : Pat<(int_x86_sse41_pmovsxdq (bc_v4i32 (loadv2i64 addr:$src))),
5569 (PMOVSXDQrm addr:$src)>;
5571 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5572 (PMOVZXBWrm addr:$src)>;
5573 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5574 (PMOVZXBWrm addr:$src)>;
5575 def : Pat<(int_x86_sse41_pmovzxbw (bc_v16i8 (loadv2i64 addr:$src))),
5576 (PMOVZXBWrm addr:$src)>;
5578 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5579 (PMOVZXWDrm addr:$src)>;
5580 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5581 (PMOVZXWDrm addr:$src)>;
5582 def : Pat<(int_x86_sse41_pmovzxwd (bc_v8i16 (loadv2i64 addr:$src))),
5583 (PMOVZXWDrm addr:$src)>;
5585 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5586 (PMOVZXDQrm addr:$src)>;
5587 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5588 (PMOVZXDQrm addr:$src)>;
5589 def : Pat<(int_x86_sse41_pmovzxdq (bc_v4i32 (loadv2i64 addr:$src))),
5590 (PMOVZXDQrm addr:$src)>;
5593 let Predicates = [HasAVX2] in {
5594 let AddedComplexity = 15 in {
5595 def : Pat<(v4i64 (X86vzmovly (v4i32 VR128:$src))),
5596 (VPMOVZXDQYrr VR128:$src)>;
5597 def : Pat<(v8i32 (X86vzmovly (v8i16 VR128:$src))),
5598 (VPMOVZXWDYrr VR128:$src)>;
5599 def : Pat<(v16i16 (X86vzmovly (v16i8 VR128:$src))),
5600 (VPMOVZXBWYrr VR128:$src)>;
5603 def : Pat<(v4i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>;
5604 def : Pat<(v8i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>;
5605 def : Pat<(v16i16 (X86vsmovl (v16i8 VR128:$src))), (VPMOVSXBWYrr VR128:$src)>;
5608 let Predicates = [HasAVX] in {
5609 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
5610 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
5611 def : Pat<(v8i16 (X86vsmovl (v16i8 VR128:$src))), (VPMOVSXBWrr VR128:$src)>;
5614 let Predicates = [UseSSE41] in {
5615 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
5616 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
5617 def : Pat<(v8i16 (X86vsmovl (v16i8 VR128:$src))), (PMOVSXBWrr VR128:$src)>;
5621 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId,
5622 OpndItins itins = DEFAULT_ITINS> {
5623 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5624 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5625 [(set VR128:$dst, (IntId VR128:$src))], itins.rr>, OpSize;
5627 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5628 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5630 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))],
5635 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5637 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5638 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5639 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5641 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5642 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5644 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5648 let Predicates = [HasAVX] in {
5649 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5651 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5653 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5655 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5659 let Predicates = [HasAVX2] in {
5660 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5661 int_x86_avx2_pmovsxbd>, VEX, VEX_L;
5662 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5663 int_x86_avx2_pmovsxwq>, VEX, VEX_L;
5664 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5665 int_x86_avx2_pmovzxbd>, VEX, VEX_L;
5666 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5667 int_x86_avx2_pmovzxwq>, VEX, VEX_L;
5670 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd,
5671 SSE_INTALU_ITINS_P>;
5672 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq,
5673 SSE_INTALU_ITINS_P>;
5674 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd,
5675 SSE_INTALU_ITINS_P>;
5676 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq,
5677 SSE_INTALU_ITINS_P>;
5679 let Predicates = [HasAVX] in {
5680 // Common patterns involving scalar load
5681 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5682 (VPMOVSXBDrm addr:$src)>;
5683 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5684 (VPMOVSXWQrm addr:$src)>;
5686 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5687 (VPMOVZXBDrm addr:$src)>;
5688 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5689 (VPMOVZXWQrm addr:$src)>;
5692 let Predicates = [UseSSE41] in {
5693 // Common patterns involving scalar load
5694 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5695 (PMOVSXBDrm addr:$src)>;
5696 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5697 (PMOVSXWQrm addr:$src)>;
5699 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5700 (PMOVZXBDrm addr:$src)>;
5701 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5702 (PMOVZXWQrm addr:$src)>;
5705 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId,
5706 OpndItins itins = DEFAULT_ITINS> {
5707 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5708 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5709 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5711 // Expecting a i16 load any extended to i32 value.
5712 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5713 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5714 [(set VR128:$dst, (IntId (bitconvert
5715 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5719 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5721 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5722 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5723 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5725 // Expecting a i16 load any extended to i32 value.
5726 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5727 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5728 [(set VR256:$dst, (IntId (bitconvert
5729 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5733 let Predicates = [HasAVX] in {
5734 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5736 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5739 let Predicates = [HasAVX2] in {
5740 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5741 int_x86_avx2_pmovsxbq>, VEX, VEX_L;
5742 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5743 int_x86_avx2_pmovzxbq>, VEX, VEX_L;
5745 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq,
5746 SSE_INTALU_ITINS_P>;
5747 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq,
5748 SSE_INTALU_ITINS_P>;
5750 let Predicates = [HasAVX2] in {
5751 def : Pat<(v16i16 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBWYrr VR128:$src)>;
5752 def : Pat<(v8i32 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBDYrr VR128:$src)>;
5753 def : Pat<(v4i64 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBQYrr VR128:$src)>;
5755 def : Pat<(v8i32 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>;
5756 def : Pat<(v4i64 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWQYrr VR128:$src)>;
5758 def : Pat<(v4i64 (X86vsext (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>;
5760 def : Pat<(v16i16 (X86vsext (v32i8 VR256:$src))),
5761 (VPMOVSXBWYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5762 def : Pat<(v8i32 (X86vsext (v32i8 VR256:$src))),
5763 (VPMOVSXBDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5764 def : Pat<(v4i64 (X86vsext (v32i8 VR256:$src))),
5765 (VPMOVSXBQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5767 def : Pat<(v8i32 (X86vsext (v16i16 VR256:$src))),
5768 (VPMOVSXWDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5769 def : Pat<(v4i64 (X86vsext (v16i16 VR256:$src))),
5770 (VPMOVSXWQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5772 def : Pat<(v4i64 (X86vsext (v8i32 VR256:$src))),
5773 (VPMOVSXDQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5775 def : Pat<(v8i32 (X86vsmovl (v8i16 (bitconvert (v2i64 (load addr:$src)))))),
5776 (VPMOVSXWDYrm addr:$src)>;
5777 def : Pat<(v4i64 (X86vsmovl (v4i32 (bitconvert (v2i64 (load addr:$src)))))),
5778 (VPMOVSXDQYrm addr:$src)>;
5780 def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2i64
5781 (scalar_to_vector (loadi64 addr:$src))))))),
5782 (VPMOVSXBDYrm addr:$src)>;
5783 def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2f64
5784 (scalar_to_vector (loadf64 addr:$src))))))),
5785 (VPMOVSXBDYrm addr:$src)>;
5787 def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2i64
5788 (scalar_to_vector (loadi64 addr:$src))))))),
5789 (VPMOVSXWQYrm addr:$src)>;
5790 def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2f64
5791 (scalar_to_vector (loadf64 addr:$src))))))),
5792 (VPMOVSXWQYrm addr:$src)>;
5794 def : Pat<(v4i64 (X86vsext (v16i8 (bitconvert (v4i32
5795 (scalar_to_vector (loadi32 addr:$src))))))),
5796 (VPMOVSXBQYrm addr:$src)>;
5799 let Predicates = [HasAVX] in {
5800 // Common patterns involving scalar load
5801 def : Pat<(int_x86_sse41_pmovsxbq
5802 (bitconvert (v4i32 (X86vzmovl
5803 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5804 (VPMOVSXBQrm addr:$src)>;
5806 def : Pat<(int_x86_sse41_pmovzxbq
5807 (bitconvert (v4i32 (X86vzmovl
5808 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5809 (VPMOVZXBQrm addr:$src)>;
5812 let Predicates = [UseSSE41] in {
5813 def : Pat<(v8i16 (X86vsext (v16i8 VR128:$src))), (PMOVSXBWrr VR128:$src)>;
5814 def : Pat<(v4i32 (X86vsext (v16i8 VR128:$src))), (PMOVSXBDrr VR128:$src)>;
5815 def : Pat<(v2i64 (X86vsext (v16i8 VR128:$src))), (PMOVSXBQrr VR128:$src)>;
5817 def : Pat<(v4i32 (X86vsext (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
5818 def : Pat<(v2i64 (X86vsext (v8i16 VR128:$src))), (PMOVSXWQrr VR128:$src)>;
5820 def : Pat<(v2i64 (X86vsext (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
5822 // Common patterns involving scalar load
5823 def : Pat<(int_x86_sse41_pmovsxbq
5824 (bitconvert (v4i32 (X86vzmovl
5825 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5826 (PMOVSXBQrm addr:$src)>;
5828 def : Pat<(int_x86_sse41_pmovzxbq
5829 (bitconvert (v4i32 (X86vzmovl
5830 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5831 (PMOVZXBQrm addr:$src)>;
5833 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64
5834 (scalar_to_vector (loadi64 addr:$src))))))),
5835 (PMOVSXWDrm addr:$src)>;
5836 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64
5837 (scalar_to_vector (loadf64 addr:$src))))))),
5838 (PMOVSXWDrm addr:$src)>;
5839 def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32
5840 (scalar_to_vector (loadi32 addr:$src))))))),
5841 (PMOVSXBDrm addr:$src)>;
5842 def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32
5843 (scalar_to_vector (loadi32 addr:$src))))))),
5844 (PMOVSXWQrm addr:$src)>;
5845 def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32
5846 (scalar_to_vector (extloadi32i16 addr:$src))))))),
5847 (PMOVSXBQrm addr:$src)>;
5848 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64
5849 (scalar_to_vector (loadi64 addr:$src))))))),
5850 (PMOVSXDQrm addr:$src)>;
5851 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64
5852 (scalar_to_vector (loadf64 addr:$src))))))),
5853 (PMOVSXDQrm addr:$src)>;
5854 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64
5855 (scalar_to_vector (loadi64 addr:$src))))))),
5856 (PMOVSXBWrm addr:$src)>;
5857 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64
5858 (scalar_to_vector (loadf64 addr:$src))))))),
5859 (PMOVSXBWrm addr:$src)>;
5862 let Predicates = [HasAVX2] in {
5863 def : Pat<(v16i16 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBWYrr VR128:$src)>;
5864 def : Pat<(v8i32 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBDYrr VR128:$src)>;
5865 def : Pat<(v4i64 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBQYrr VR128:$src)>;
5867 def : Pat<(v8i32 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWDYrr VR128:$src)>;
5868 def : Pat<(v4i64 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWQYrr VR128:$src)>;
5870 def : Pat<(v4i64 (X86vzext (v4i32 VR128:$src))), (VPMOVZXDQYrr VR128:$src)>;
5872 def : Pat<(v16i16 (X86vzext (v32i8 VR256:$src))),
5873 (VPMOVZXBWYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5874 def : Pat<(v8i32 (X86vzext (v32i8 VR256:$src))),
5875 (VPMOVZXBDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5876 def : Pat<(v4i64 (X86vzext (v32i8 VR256:$src))),
5877 (VPMOVZXBQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5879 def : Pat<(v8i32 (X86vzext (v16i16 VR256:$src))),
5880 (VPMOVZXWDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5881 def : Pat<(v4i64 (X86vzext (v16i16 VR256:$src))),
5882 (VPMOVZXWQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5884 def : Pat<(v4i64 (X86vzext (v8i32 VR256:$src))),
5885 (VPMOVZXDQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5888 let Predicates = [HasAVX] in {
5889 def : Pat<(v8i16 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBWrr VR128:$src)>;
5890 def : Pat<(v4i32 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBDrr VR128:$src)>;
5891 def : Pat<(v2i64 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBQrr VR128:$src)>;
5893 def : Pat<(v4i32 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWDrr VR128:$src)>;
5894 def : Pat<(v2i64 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWQrr VR128:$src)>;
5896 def : Pat<(v2i64 (X86vzext (v4i32 VR128:$src))), (VPMOVZXDQrr VR128:$src)>;
5898 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5899 (VPMOVZXBWrm addr:$src)>;
5900 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5901 (VPMOVZXBWrm addr:$src)>;
5902 def : Pat<(v4i32 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5903 (VPMOVZXBDrm addr:$src)>;
5904 def : Pat<(v2i64 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))),
5905 (VPMOVZXBQrm addr:$src)>;
5907 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5908 (VPMOVZXWDrm addr:$src)>;
5909 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5910 (VPMOVZXWDrm addr:$src)>;
5911 def : Pat<(v2i64 (X86vzext (v8i16 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5912 (VPMOVZXWQrm addr:$src)>;
5914 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5915 (VPMOVZXDQrm addr:$src)>;
5916 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5917 (VPMOVZXDQrm addr:$src)>;
5918 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (X86vzload addr:$src)))))),
5919 (VPMOVZXDQrm addr:$src)>;
5921 def : Pat<(v8i16 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBWrr VR128:$src)>;
5922 def : Pat<(v4i32 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBDrr VR128:$src)>;
5923 def : Pat<(v2i64 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBQrr VR128:$src)>;
5925 def : Pat<(v4i32 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
5926 def : Pat<(v2i64 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWQrr VR128:$src)>;
5928 def : Pat<(v2i64 (X86vsext (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
5930 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64
5931 (scalar_to_vector (loadi64 addr:$src))))))),
5932 (VPMOVSXWDrm addr:$src)>;
5933 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64
5934 (scalar_to_vector (loadi64 addr:$src))))))),
5935 (VPMOVSXDQrm addr:$src)>;
5936 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64
5937 (scalar_to_vector (loadf64 addr:$src))))))),
5938 (VPMOVSXWDrm addr:$src)>;
5939 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64
5940 (scalar_to_vector (loadf64 addr:$src))))))),
5941 (VPMOVSXDQrm addr:$src)>;
5942 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64
5943 (scalar_to_vector (loadi64 addr:$src))))))),
5944 (VPMOVSXBWrm addr:$src)>;
5945 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64
5946 (scalar_to_vector (loadf64 addr:$src))))))),
5947 (VPMOVSXBWrm addr:$src)>;
5949 def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32
5950 (scalar_to_vector (loadi32 addr:$src))))))),
5951 (VPMOVSXBDrm addr:$src)>;
5952 def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32
5953 (scalar_to_vector (loadi32 addr:$src))))))),
5954 (VPMOVSXWQrm addr:$src)>;
5955 def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32
5956 (scalar_to_vector (extloadi32i16 addr:$src))))))),
5957 (VPMOVSXBQrm addr:$src)>;
5960 let Predicates = [UseSSE41] in {
5961 def : Pat<(v8i16 (X86vzext (v16i8 VR128:$src))), (PMOVZXBWrr VR128:$src)>;
5962 def : Pat<(v4i32 (X86vzext (v16i8 VR128:$src))), (PMOVZXBDrr VR128:$src)>;
5963 def : Pat<(v2i64 (X86vzext (v16i8 VR128:$src))), (PMOVZXBQrr VR128:$src)>;
5965 def : Pat<(v4i32 (X86vzext (v8i16 VR128:$src))), (PMOVZXWDrr VR128:$src)>;
5966 def : Pat<(v2i64 (X86vzext (v8i16 VR128:$src))), (PMOVZXWQrr VR128:$src)>;
5968 def : Pat<(v2i64 (X86vzext (v4i32 VR128:$src))), (PMOVZXDQrr VR128:$src)>;
5970 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5971 (PMOVZXBWrm addr:$src)>;
5972 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5973 (PMOVZXBWrm addr:$src)>;
5974 def : Pat<(v4i32 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5975 (PMOVZXBDrm addr:$src)>;
5976 def : Pat<(v2i64 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))),
5977 (PMOVZXBQrm addr:$src)>;
5979 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5980 (PMOVZXWDrm addr:$src)>;
5981 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5982 (PMOVZXWDrm addr:$src)>;
5983 def : Pat<(v2i64 (X86vzext (v8i16 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5984 (PMOVZXWQrm addr:$src)>;
5986 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5987 (PMOVZXDQrm addr:$src)>;
5988 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5989 (PMOVZXDQrm addr:$src)>;
5990 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (X86vzload addr:$src)))))),
5991 (PMOVZXDQrm addr:$src)>;
5994 //===----------------------------------------------------------------------===//
5995 // SSE4.1 - Extract Instructions
5996 //===----------------------------------------------------------------------===//
5998 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
5999 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
6000 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32orGR64:$dst),
6001 (ins VR128:$src1, i32i8imm:$src2),
6002 !strconcat(OpcodeStr,
6003 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6004 [(set GR32orGR64:$dst, (X86pextrb (v16i8 VR128:$src1),
6007 let neverHasSideEffects = 1, mayStore = 1 in
6008 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6009 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
6010 !strconcat(OpcodeStr,
6011 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6014 // There's an AssertZext in the way of writing the store pattern
6015 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
6018 let Predicates = [HasAVX] in
6019 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
6021 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
6024 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
6025 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
6026 let isCodeGenOnly = 1, hasSideEffects = 0 in
6027 def rr_REV : SS4AIi8<opc, MRMDestReg, (outs GR32orGR64:$dst),
6028 (ins VR128:$src1, i32i8imm:$src2),
6029 !strconcat(OpcodeStr,
6030 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6033 let neverHasSideEffects = 1, mayStore = 1 in
6034 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6035 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
6036 !strconcat(OpcodeStr,
6037 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6040 // There's an AssertZext in the way of writing the store pattern
6041 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
6044 let Predicates = [HasAVX] in
6045 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
6047 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
6050 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
6051 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
6052 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
6053 (ins VR128:$src1, i32i8imm:$src2),
6054 !strconcat(OpcodeStr,
6055 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6057 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
6058 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6059 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
6060 !strconcat(OpcodeStr,
6061 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6062 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
6063 addr:$dst)]>, OpSize;
6066 let Predicates = [HasAVX] in
6067 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
6069 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
6071 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
6072 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
6073 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
6074 (ins VR128:$src1, i32i8imm:$src2),
6075 !strconcat(OpcodeStr,
6076 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6078 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
6079 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6080 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
6081 !strconcat(OpcodeStr,
6082 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6083 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
6084 addr:$dst)]>, OpSize, REX_W;
6087 let Predicates = [HasAVX] in
6088 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
6090 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
6092 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
6094 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr,
6095 OpndItins itins = DEFAULT_ITINS> {
6096 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32orGR64:$dst),
6097 (ins VR128:$src1, i32i8imm:$src2),
6098 !strconcat(OpcodeStr,
6099 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6100 [(set GR32orGR64:$dst,
6101 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))],
6104 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6105 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
6106 !strconcat(OpcodeStr,
6107 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6108 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
6109 addr:$dst)], itins.rm>, OpSize;
6112 let ExeDomain = SSEPackedSingle in {
6113 let Predicates = [UseAVX] in
6114 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
6115 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps", SSE_EXTRACT_ITINS>;
6118 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
6119 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
6122 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
6124 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
6127 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
6128 Requires<[UseSSE41]>;
6130 //===----------------------------------------------------------------------===//
6131 // SSE4.1 - Insert Instructions
6132 //===----------------------------------------------------------------------===//
6134 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
6135 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6136 (ins VR128:$src1, GR32orGR64:$src2, i32i8imm:$src3),
6138 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6140 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6142 (X86pinsrb VR128:$src1, GR32orGR64:$src2, imm:$src3))]>, OpSize;
6143 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6144 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
6146 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6148 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6150 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
6151 imm:$src3))]>, OpSize;
6154 let Predicates = [HasAVX] in
6155 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
6156 let Constraints = "$src1 = $dst" in
6157 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
6159 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
6160 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6161 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
6163 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6165 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6167 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
6169 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6170 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
6172 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6174 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6176 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
6177 imm:$src3)))]>, OpSize;
6180 let Predicates = [HasAVX] in
6181 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
6182 let Constraints = "$src1 = $dst" in
6183 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
6185 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
6186 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6187 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
6189 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6191 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6193 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
6195 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6196 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
6198 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6200 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6202 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
6203 imm:$src3)))]>, OpSize;
6206 let Predicates = [HasAVX] in
6207 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
6208 let Constraints = "$src1 = $dst" in
6209 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
6211 // insertps has a few different modes, there's the first two here below which
6212 // are optimized inserts that won't zero arbitrary elements in the destination
6213 // vector. The next one matches the intrinsic and could zero arbitrary elements
6214 // in the target vector.
6215 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1,
6216 OpndItins itins = DEFAULT_ITINS> {
6217 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6218 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
6220 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6222 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6224 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))], itins.rr>,
6226 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6227 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
6229 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6231 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6233 (X86insrtps VR128:$src1,
6234 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
6235 imm:$src3))], itins.rm>, OpSize;
6238 let ExeDomain = SSEPackedSingle in {
6239 let Predicates = [UseAVX] in
6240 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
6241 let Constraints = "$src1 = $dst" in
6242 defm INSERTPS : SS41I_insertf32<0x21, "insertps", 1, SSE_INSERT_ITINS>;
6245 //===----------------------------------------------------------------------===//
6246 // SSE4.1 - Round Instructions
6247 //===----------------------------------------------------------------------===//
6249 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
6250 X86MemOperand x86memop, RegisterClass RC,
6251 PatFrag mem_frag32, PatFrag mem_frag64,
6252 Intrinsic V4F32Int, Intrinsic V2F64Int> {
6253 let ExeDomain = SSEPackedSingle in {
6254 // Intrinsic operation, reg.
6255 // Vector intrinsic operation, reg
6256 def PSr : SS4AIi8<opcps, MRMSrcReg,
6257 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6258 !strconcat(OpcodeStr,
6259 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6260 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))],
6261 IIC_SSE_ROUNDPS_REG>,
6264 // Vector intrinsic operation, mem
6265 def PSm : SS4AIi8<opcps, MRMSrcMem,
6266 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6267 !strconcat(OpcodeStr,
6268 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6270 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))],
6271 IIC_SSE_ROUNDPS_MEM>,
6273 } // ExeDomain = SSEPackedSingle
6275 let ExeDomain = SSEPackedDouble in {
6276 // Vector intrinsic operation, reg
6277 def PDr : SS4AIi8<opcpd, MRMSrcReg,
6278 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6279 !strconcat(OpcodeStr,
6280 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6281 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))],
6282 IIC_SSE_ROUNDPS_REG>,
6285 // Vector intrinsic operation, mem
6286 def PDm : SS4AIi8<opcpd, MRMSrcMem,
6287 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6288 !strconcat(OpcodeStr,
6289 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6291 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))],
6292 IIC_SSE_ROUNDPS_REG>,
6294 } // ExeDomain = SSEPackedDouble
6297 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
6300 Intrinsic F64Int, bit Is2Addr = 1> {
6301 let ExeDomain = GenericDomain in {
6303 let hasSideEffects = 0 in
6304 def SSr : SS4AIi8<opcss, MRMSrcReg,
6305 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3),
6307 !strconcat(OpcodeStr,
6308 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6309 !strconcat(OpcodeStr,
6310 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6313 // Intrinsic operation, reg.
6314 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
6315 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6317 !strconcat(OpcodeStr,
6318 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6319 !strconcat(OpcodeStr,
6320 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6321 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6324 // Intrinsic operation, mem.
6325 def SSm : SS4AIi8<opcss, MRMSrcMem,
6326 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
6328 !strconcat(OpcodeStr,
6329 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6330 !strconcat(OpcodeStr,
6331 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6333 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
6337 let hasSideEffects = 0 in
6338 def SDr : SS4AIi8<opcsd, MRMSrcReg,
6339 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3),
6341 !strconcat(OpcodeStr,
6342 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6343 !strconcat(OpcodeStr,
6344 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6347 // Intrinsic operation, reg.
6348 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
6349 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6351 !strconcat(OpcodeStr,
6352 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6353 !strconcat(OpcodeStr,
6354 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6355 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6358 // Intrinsic operation, mem.
6359 def SDm : SS4AIi8<opcsd, MRMSrcMem,
6360 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
6362 !strconcat(OpcodeStr,
6363 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6364 !strconcat(OpcodeStr,
6365 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6367 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
6369 } // ExeDomain = GenericDomain
6372 // FP round - roundss, roundps, roundsd, roundpd
6373 let Predicates = [HasAVX] in {
6375 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
6376 memopv4f32, memopv2f64,
6377 int_x86_sse41_round_ps,
6378 int_x86_sse41_round_pd>, VEX;
6379 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
6380 memopv8f32, memopv4f64,
6381 int_x86_avx_round_ps_256,
6382 int_x86_avx_round_pd_256>, VEX, VEX_L;
6383 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
6384 int_x86_sse41_round_ss,
6385 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
6387 def : Pat<(ffloor FR32:$src),
6388 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6389 def : Pat<(f64 (ffloor FR64:$src)),
6390 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6391 def : Pat<(f32 (fnearbyint FR32:$src)),
6392 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6393 def : Pat<(f64 (fnearbyint FR64:$src)),
6394 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6395 def : Pat<(f32 (fceil FR32:$src)),
6396 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6397 def : Pat<(f64 (fceil FR64:$src)),
6398 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6399 def : Pat<(f32 (frint FR32:$src)),
6400 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6401 def : Pat<(f64 (frint FR64:$src)),
6402 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6403 def : Pat<(f32 (ftrunc FR32:$src)),
6404 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6405 def : Pat<(f64 (ftrunc FR64:$src)),
6406 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6408 def : Pat<(v4f32 (ffloor VR128:$src)),
6409 (VROUNDPSr VR128:$src, (i32 0x1))>;
6410 def : Pat<(v4f32 (fnearbyint VR128:$src)),
6411 (VROUNDPSr VR128:$src, (i32 0xC))>;
6412 def : Pat<(v4f32 (fceil VR128:$src)),
6413 (VROUNDPSr VR128:$src, (i32 0x2))>;
6414 def : Pat<(v4f32 (frint VR128:$src)),
6415 (VROUNDPSr VR128:$src, (i32 0x4))>;
6416 def : Pat<(v4f32 (ftrunc VR128:$src)),
6417 (VROUNDPSr VR128:$src, (i32 0x3))>;
6419 def : Pat<(v2f64 (ffloor VR128:$src)),
6420 (VROUNDPDr VR128:$src, (i32 0x1))>;
6421 def : Pat<(v2f64 (fnearbyint VR128:$src)),
6422 (VROUNDPDr VR128:$src, (i32 0xC))>;
6423 def : Pat<(v2f64 (fceil VR128:$src)),
6424 (VROUNDPDr VR128:$src, (i32 0x2))>;
6425 def : Pat<(v2f64 (frint VR128:$src)),
6426 (VROUNDPDr VR128:$src, (i32 0x4))>;
6427 def : Pat<(v2f64 (ftrunc VR128:$src)),
6428 (VROUNDPDr VR128:$src, (i32 0x3))>;
6430 def : Pat<(v8f32 (ffloor VR256:$src)),
6431 (VROUNDYPSr VR256:$src, (i32 0x1))>;
6432 def : Pat<(v8f32 (fnearbyint VR256:$src)),
6433 (VROUNDYPSr VR256:$src, (i32 0xC))>;
6434 def : Pat<(v8f32 (fceil VR256:$src)),
6435 (VROUNDYPSr VR256:$src, (i32 0x2))>;
6436 def : Pat<(v8f32 (frint VR256:$src)),
6437 (VROUNDYPSr VR256:$src, (i32 0x4))>;
6438 def : Pat<(v8f32 (ftrunc VR256:$src)),
6439 (VROUNDYPSr VR256:$src, (i32 0x3))>;
6441 def : Pat<(v4f64 (ffloor VR256:$src)),
6442 (VROUNDYPDr VR256:$src, (i32 0x1))>;
6443 def : Pat<(v4f64 (fnearbyint VR256:$src)),
6444 (VROUNDYPDr VR256:$src, (i32 0xC))>;
6445 def : Pat<(v4f64 (fceil VR256:$src)),
6446 (VROUNDYPDr VR256:$src, (i32 0x2))>;
6447 def : Pat<(v4f64 (frint VR256:$src)),
6448 (VROUNDYPDr VR256:$src, (i32 0x4))>;
6449 def : Pat<(v4f64 (ftrunc VR256:$src)),
6450 (VROUNDYPDr VR256:$src, (i32 0x3))>;
6453 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
6454 memopv4f32, memopv2f64,
6455 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
6456 let Constraints = "$src1 = $dst" in
6457 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
6458 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
6460 let Predicates = [UseSSE41] in {
6461 def : Pat<(ffloor FR32:$src),
6462 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6463 def : Pat<(f64 (ffloor FR64:$src)),
6464 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6465 def : Pat<(f32 (fnearbyint FR32:$src)),
6466 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6467 def : Pat<(f64 (fnearbyint FR64:$src)),
6468 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6469 def : Pat<(f32 (fceil FR32:$src)),
6470 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6471 def : Pat<(f64 (fceil FR64:$src)),
6472 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6473 def : Pat<(f32 (frint FR32:$src)),
6474 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6475 def : Pat<(f64 (frint FR64:$src)),
6476 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6477 def : Pat<(f32 (ftrunc FR32:$src)),
6478 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6479 def : Pat<(f64 (ftrunc FR64:$src)),
6480 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6482 def : Pat<(v4f32 (ffloor VR128:$src)),
6483 (ROUNDPSr VR128:$src, (i32 0x1))>;
6484 def : Pat<(v4f32 (fnearbyint VR128:$src)),
6485 (ROUNDPSr VR128:$src, (i32 0xC))>;
6486 def : Pat<(v4f32 (fceil VR128:$src)),
6487 (ROUNDPSr VR128:$src, (i32 0x2))>;
6488 def : Pat<(v4f32 (frint VR128:$src)),
6489 (ROUNDPSr VR128:$src, (i32 0x4))>;
6490 def : Pat<(v4f32 (ftrunc VR128:$src)),
6491 (ROUNDPSr VR128:$src, (i32 0x3))>;
6493 def : Pat<(v2f64 (ffloor VR128:$src)),
6494 (ROUNDPDr VR128:$src, (i32 0x1))>;
6495 def : Pat<(v2f64 (fnearbyint VR128:$src)),
6496 (ROUNDPDr VR128:$src, (i32 0xC))>;
6497 def : Pat<(v2f64 (fceil VR128:$src)),
6498 (ROUNDPDr VR128:$src, (i32 0x2))>;
6499 def : Pat<(v2f64 (frint VR128:$src)),
6500 (ROUNDPDr VR128:$src, (i32 0x4))>;
6501 def : Pat<(v2f64 (ftrunc VR128:$src)),
6502 (ROUNDPDr VR128:$src, (i32 0x3))>;
6505 //===----------------------------------------------------------------------===//
6506 // SSE4.1 - Packed Bit Test
6507 //===----------------------------------------------------------------------===//
6509 // ptest instruction we'll lower to this in X86ISelLowering primarily from
6510 // the intel intrinsic that corresponds to this.
6511 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6512 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6513 "vptest\t{$src2, $src1|$src1, $src2}",
6514 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6516 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6517 "vptest\t{$src2, $src1|$src1, $src2}",
6518 [(set EFLAGS,(X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6521 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
6522 "vptest\t{$src2, $src1|$src1, $src2}",
6523 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
6525 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
6526 "vptest\t{$src2, $src1|$src1, $src2}",
6527 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
6531 let Defs = [EFLAGS] in {
6532 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6533 "ptest\t{$src2, $src1|$src1, $src2}",
6534 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6536 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6537 "ptest\t{$src2, $src1|$src1, $src2}",
6538 [(set EFLAGS, (X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6542 // The bit test instructions below are AVX only
6543 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
6544 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
6545 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
6546 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6547 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
6548 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
6549 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6550 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
6554 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6555 let ExeDomain = SSEPackedSingle in {
6556 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
6557 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>,
6560 let ExeDomain = SSEPackedDouble in {
6561 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
6562 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>,
6567 //===----------------------------------------------------------------------===//
6568 // SSE4.1 - Misc Instructions
6569 //===----------------------------------------------------------------------===//
6571 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
6572 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
6573 "popcnt{w}\t{$src, $dst|$dst, $src}",
6574 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)],
6577 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
6578 "popcnt{w}\t{$src, $dst|$dst, $src}",
6579 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
6580 (implicit EFLAGS)], IIC_SSE_POPCNT_RM>, OpSize, XS;
6582 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
6583 "popcnt{l}\t{$src, $dst|$dst, $src}",
6584 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)],
6587 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
6588 "popcnt{l}\t{$src, $dst|$dst, $src}",
6589 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
6590 (implicit EFLAGS)], IIC_SSE_POPCNT_RM>, XS;
6592 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
6593 "popcnt{q}\t{$src, $dst|$dst, $src}",
6594 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)],
6597 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
6598 "popcnt{q}\t{$src, $dst|$dst, $src}",
6599 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
6600 (implicit EFLAGS)], IIC_SSE_POPCNT_RM>, XS;
6605 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
6606 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
6607 Intrinsic IntId128> {
6608 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6610 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6611 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
6612 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6614 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6617 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
6620 let Predicates = [HasAVX] in
6621 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
6622 int_x86_sse41_phminposuw>, VEX;
6623 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
6624 int_x86_sse41_phminposuw>;
6626 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6627 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
6628 Intrinsic IntId128, bit Is2Addr = 1,
6629 OpndItins itins = DEFAULT_ITINS> {
6630 let isCommutable = 1 in
6631 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6632 (ins VR128:$src1, VR128:$src2),
6634 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6635 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6636 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))],
6638 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6639 (ins VR128:$src1, i128mem:$src2),
6641 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6642 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6644 (IntId128 VR128:$src1,
6645 (bitconvert (memopv2i64 addr:$src2))))],
6649 /// SS41I_binop_rm_int_y - Simple SSE 4.1 binary operator
6650 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6651 Intrinsic IntId256> {
6652 let isCommutable = 1 in
6653 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6654 (ins VR256:$src1, VR256:$src2),
6655 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6656 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
6657 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6658 (ins VR256:$src1, i256mem:$src2),
6659 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6661 (IntId256 VR256:$src1,
6662 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
6666 /// SS48I_binop_rm - Simple SSE41 binary operator.
6667 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6668 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6669 X86MemOperand x86memop, bit Is2Addr = 1,
6670 OpndItins itins = DEFAULT_ITINS> {
6671 let isCommutable = 1 in
6672 def rr : SS48I<opc, MRMSrcReg, (outs RC:$dst),
6673 (ins RC:$src1, RC:$src2),
6675 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6676 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6677 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>, OpSize;
6678 def rm : SS48I<opc, MRMSrcMem, (outs RC:$dst),
6679 (ins RC:$src1, x86memop:$src2),
6681 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6682 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6684 (OpVT (OpNode RC:$src1,
6685 (bitconvert (memop_frag addr:$src2)))))]>, OpSize;
6688 let Predicates = [HasAVX] in {
6689 let isCommutable = 0 in
6690 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
6692 defm VPMINSB : SS48I_binop_rm<0x38, "vpminsb", X86smin, v16i8, VR128,
6693 memopv2i64, i128mem, 0>, VEX_4V;
6694 defm VPMINSD : SS48I_binop_rm<0x39, "vpminsd", X86smin, v4i32, VR128,
6695 memopv2i64, i128mem, 0>, VEX_4V;
6696 defm VPMINUD : SS48I_binop_rm<0x3B, "vpminud", X86umin, v4i32, VR128,
6697 memopv2i64, i128mem, 0>, VEX_4V;
6698 defm VPMINUW : SS48I_binop_rm<0x3A, "vpminuw", X86umin, v8i16, VR128,
6699 memopv2i64, i128mem, 0>, VEX_4V;
6700 defm VPMAXSB : SS48I_binop_rm<0x3C, "vpmaxsb", X86smax, v16i8, VR128,
6701 memopv2i64, i128mem, 0>, VEX_4V;
6702 defm VPMAXSD : SS48I_binop_rm<0x3D, "vpmaxsd", X86smax, v4i32, VR128,
6703 memopv2i64, i128mem, 0>, VEX_4V;
6704 defm VPMAXUD : SS48I_binop_rm<0x3F, "vpmaxud", X86umax, v4i32, VR128,
6705 memopv2i64, i128mem, 0>, VEX_4V;
6706 defm VPMAXUW : SS48I_binop_rm<0x3E, "vpmaxuw", X86umax, v8i16, VR128,
6707 memopv2i64, i128mem, 0>, VEX_4V;
6708 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
6712 let Predicates = [HasAVX2] in {
6713 let isCommutable = 0 in
6714 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
6715 int_x86_avx2_packusdw>, VEX_4V, VEX_L;
6716 defm VPMINSBY : SS48I_binop_rm<0x38, "vpminsb", X86smin, v32i8, VR256,
6717 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6718 defm VPMINSDY : SS48I_binop_rm<0x39, "vpminsd", X86smin, v8i32, VR256,
6719 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6720 defm VPMINUDY : SS48I_binop_rm<0x3B, "vpminud", X86umin, v8i32, VR256,
6721 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6722 defm VPMINUWY : SS48I_binop_rm<0x3A, "vpminuw", X86umin, v16i16, VR256,
6723 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6724 defm VPMAXSBY : SS48I_binop_rm<0x3C, "vpmaxsb", X86smax, v32i8, VR256,
6725 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6726 defm VPMAXSDY : SS48I_binop_rm<0x3D, "vpmaxsd", X86smax, v8i32, VR256,
6727 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6728 defm VPMAXUDY : SS48I_binop_rm<0x3F, "vpmaxud", X86umax, v8i32, VR256,
6729 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6730 defm VPMAXUWY : SS48I_binop_rm<0x3E, "vpmaxuw", X86umax, v16i16, VR256,
6731 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6732 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
6733 int_x86_avx2_pmul_dq>, VEX_4V, VEX_L;
6736 let Constraints = "$src1 = $dst" in {
6737 let isCommutable = 0 in
6738 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
6739 defm PMINSB : SS48I_binop_rm<0x38, "pminsb", X86smin, v16i8, VR128,
6740 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6741 defm PMINSD : SS48I_binop_rm<0x39, "pminsd", X86smin, v4i32, VR128,
6742 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6743 defm PMINUD : SS48I_binop_rm<0x3B, "pminud", X86umin, v4i32, VR128,
6744 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6745 defm PMINUW : SS48I_binop_rm<0x3A, "pminuw", X86umin, v8i16, VR128,
6746 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6747 defm PMAXSB : SS48I_binop_rm<0x3C, "pmaxsb", X86smax, v16i8, VR128,
6748 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6749 defm PMAXSD : SS48I_binop_rm<0x3D, "pmaxsd", X86smax, v4i32, VR128,
6750 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6751 defm PMAXUD : SS48I_binop_rm<0x3F, "pmaxud", X86umax, v4i32, VR128,
6752 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6753 defm PMAXUW : SS48I_binop_rm<0x3E, "pmaxuw", X86umax, v8i16, VR128,
6754 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6755 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq,
6756 1, SSE_INTMUL_ITINS_P>;
6759 let Predicates = [HasAVX] in {
6760 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, VR128,
6761 memopv2i64, i128mem, 0>, VEX_4V;
6762 defm VPCMPEQQ : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v2i64, VR128,
6763 memopv2i64, i128mem, 0>, VEX_4V;
6765 let Predicates = [HasAVX2] in {
6766 defm VPMULLDY : SS48I_binop_rm<0x40, "vpmulld", mul, v8i32, VR256,
6767 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6768 defm VPCMPEQQY : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v4i64, VR256,
6769 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6772 let Constraints = "$src1 = $dst" in {
6773 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32, VR128,
6774 memopv2i64, i128mem, 1, SSE_PMULLD_ITINS>;
6775 defm PCMPEQQ : SS48I_binop_rm<0x29, "pcmpeqq", X86pcmpeq, v2i64, VR128,
6776 memopv2i64, i128mem, 1, SSE_INTALUQ_ITINS_P>;
6779 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
6780 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
6781 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6782 X86MemOperand x86memop, bit Is2Addr = 1,
6783 OpndItins itins = DEFAULT_ITINS> {
6784 let isCommutable = 1 in
6785 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
6786 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6788 !strconcat(OpcodeStr,
6789 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6790 !strconcat(OpcodeStr,
6791 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6792 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))], itins.rr>,
6794 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
6795 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6797 !strconcat(OpcodeStr,
6798 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6799 !strconcat(OpcodeStr,
6800 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6803 (bitconvert (memop_frag addr:$src2)), imm:$src3))], itins.rm>,
6807 let Predicates = [HasAVX] in {
6808 let isCommutable = 0 in {
6809 let ExeDomain = SSEPackedSingle in {
6810 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
6811 VR128, memopv4f32, f128mem, 0>, VEX_4V;
6812 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
6813 int_x86_avx_blend_ps_256, VR256, memopv8f32,
6814 f256mem, 0>, VEX_4V, VEX_L;
6816 let ExeDomain = SSEPackedDouble in {
6817 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
6818 VR128, memopv2f64, f128mem, 0>, VEX_4V;
6819 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
6820 int_x86_avx_blend_pd_256,VR256, memopv4f64,
6821 f256mem, 0>, VEX_4V, VEX_L;
6823 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
6824 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6825 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
6826 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6828 let ExeDomain = SSEPackedSingle in
6829 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
6830 VR128, memopv4f32, f128mem, 0>, VEX_4V;
6831 let ExeDomain = SSEPackedDouble in
6832 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
6833 VR128, memopv2f64, f128mem, 0>, VEX_4V;
6834 let ExeDomain = SSEPackedSingle in
6835 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
6836 VR256, memopv8f32, i256mem, 0>, VEX_4V, VEX_L;
6839 let Predicates = [HasAVX2] in {
6840 let isCommutable = 0 in {
6841 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
6842 VR256, memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6843 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
6844 VR256, memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6848 let Constraints = "$src1 = $dst" in {
6849 let isCommutable = 0 in {
6850 let ExeDomain = SSEPackedSingle in
6851 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
6852 VR128, memopv4f32, f128mem,
6853 1, SSE_INTALU_ITINS_P>;
6854 let ExeDomain = SSEPackedDouble in
6855 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
6856 VR128, memopv2f64, f128mem,
6857 1, SSE_INTALU_ITINS_P>;
6858 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
6859 VR128, memopv2i64, i128mem,
6860 1, SSE_INTALU_ITINS_P>;
6861 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
6862 VR128, memopv2i64, i128mem,
6863 1, SSE_INTMUL_ITINS_P>;
6865 let ExeDomain = SSEPackedSingle in
6866 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
6867 VR128, memopv4f32, f128mem, 1,
6869 let ExeDomain = SSEPackedDouble in
6870 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
6871 VR128, memopv2f64, f128mem, 1,
6875 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
6876 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
6877 RegisterClass RC, X86MemOperand x86memop,
6878 PatFrag mem_frag, Intrinsic IntId> {
6879 def rr : Ii8<opc, MRMSrcReg, (outs RC:$dst),
6880 (ins RC:$src1, RC:$src2, RC:$src3),
6881 !strconcat(OpcodeStr,
6882 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6883 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
6884 NoItinerary, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6886 def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
6887 (ins RC:$src1, x86memop:$src2, RC:$src3),
6888 !strconcat(OpcodeStr,
6889 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6891 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
6893 NoItinerary, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6896 let Predicates = [HasAVX] in {
6897 let ExeDomain = SSEPackedDouble in {
6898 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, f128mem,
6899 memopv2f64, int_x86_sse41_blendvpd>;
6900 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, f256mem,
6901 memopv4f64, int_x86_avx_blendv_pd_256>, VEX_L;
6902 } // ExeDomain = SSEPackedDouble
6903 let ExeDomain = SSEPackedSingle in {
6904 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, f128mem,
6905 memopv4f32, int_x86_sse41_blendvps>;
6906 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, f256mem,
6907 memopv8f32, int_x86_avx_blendv_ps_256>, VEX_L;
6908 } // ExeDomain = SSEPackedSingle
6909 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
6910 memopv2i64, int_x86_sse41_pblendvb>;
6913 let Predicates = [HasAVX2] in {
6914 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
6915 memopv4i64, int_x86_avx2_pblendvb>, VEX_L;
6918 let Predicates = [HasAVX] in {
6919 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
6920 (v16i8 VR128:$src2))),
6921 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6922 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
6923 (v4i32 VR128:$src2))),
6924 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6925 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
6926 (v4f32 VR128:$src2))),
6927 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6928 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
6929 (v2i64 VR128:$src2))),
6930 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6931 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
6932 (v2f64 VR128:$src2))),
6933 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6934 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
6935 (v8i32 VR256:$src2))),
6936 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6937 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
6938 (v8f32 VR256:$src2))),
6939 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6940 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
6941 (v4i64 VR256:$src2))),
6942 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6943 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
6944 (v4f64 VR256:$src2))),
6945 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6947 def : Pat<(v8f32 (X86Blendi (v8f32 VR256:$src1), (v8f32 VR256:$src2),
6949 (VBLENDPSYrri VR256:$src1, VR256:$src2, imm:$mask)>;
6950 def : Pat<(v4f64 (X86Blendi (v4f64 VR256:$src1), (v4f64 VR256:$src2),
6952 (VBLENDPDYrri VR256:$src1, VR256:$src2, imm:$mask)>;
6954 def : Pat<(v8i16 (X86Blendi (v8i16 VR128:$src1), (v8i16 VR128:$src2),
6956 (VPBLENDWrri VR128:$src1, VR128:$src2, imm:$mask)>;
6957 def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$src1), (v4f32 VR128:$src2),
6959 (VBLENDPSrri VR128:$src1, VR128:$src2, imm:$mask)>;
6960 def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$src1), (v2f64 VR128:$src2),
6962 (VBLENDPDrri VR128:$src1, VR128:$src2, imm:$mask)>;
6965 let Predicates = [HasAVX2] in {
6966 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
6967 (v32i8 VR256:$src2))),
6968 (VPBLENDVBYrr VR256:$src1, VR256:$src2, VR256:$mask)>;
6969 def : Pat<(v16i16 (X86Blendi (v16i16 VR256:$src1), (v16i16 VR256:$src2),
6971 (VPBLENDWYrri VR256:$src1, VR256:$src2, imm:$mask)>;
6974 /// SS41I_ternary_int - SSE 4.1 ternary operator
6975 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
6976 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
6977 X86MemOperand x86memop, Intrinsic IntId,
6978 OpndItins itins = DEFAULT_ITINS> {
6979 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6980 (ins VR128:$src1, VR128:$src2),
6981 !strconcat(OpcodeStr,
6982 "\t{$src2, $dst|$dst, $src2}"),
6983 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))],
6986 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6987 (ins VR128:$src1, x86memop:$src2),
6988 !strconcat(OpcodeStr,
6989 "\t{$src2, $dst|$dst, $src2}"),
6992 (bitconvert (mem_frag addr:$src2)), XMM0))],
6997 let ExeDomain = SSEPackedDouble in
6998 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64, f128mem,
6999 int_x86_sse41_blendvpd>;
7000 let ExeDomain = SSEPackedSingle in
7001 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32, f128mem,
7002 int_x86_sse41_blendvps>;
7003 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64, i128mem,
7004 int_x86_sse41_pblendvb>;
7006 // Aliases with the implicit xmm0 argument
7007 def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7008 (BLENDVPDrr0 VR128:$dst, VR128:$src2)>;
7009 def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7010 (BLENDVPDrm0 VR128:$dst, f128mem:$src2)>;
7011 def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7012 (BLENDVPSrr0 VR128:$dst, VR128:$src2)>;
7013 def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7014 (BLENDVPSrm0 VR128:$dst, f128mem:$src2)>;
7015 def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7016 (PBLENDVBrr0 VR128:$dst, VR128:$src2)>;
7017 def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7018 (PBLENDVBrm0 VR128:$dst, i128mem:$src2)>;
7020 let Predicates = [UseSSE41] in {
7021 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
7022 (v16i8 VR128:$src2))),
7023 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
7024 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
7025 (v4i32 VR128:$src2))),
7026 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
7027 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
7028 (v4f32 VR128:$src2))),
7029 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
7030 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
7031 (v2i64 VR128:$src2))),
7032 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
7033 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
7034 (v2f64 VR128:$src2))),
7035 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
7037 def : Pat<(v8i16 (X86Blendi (v8i16 VR128:$src1), (v8i16 VR128:$src2),
7039 (PBLENDWrri VR128:$src1, VR128:$src2, imm:$mask)>;
7040 def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$src1), (v4f32 VR128:$src2),
7042 (BLENDPSrri VR128:$src1, VR128:$src2, imm:$mask)>;
7043 def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$src1), (v2f64 VR128:$src2),
7045 (BLENDPDrri VR128:$src1, VR128:$src2, imm:$mask)>;
7049 let Predicates = [HasAVX] in
7050 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
7051 "vmovntdqa\t{$src, $dst|$dst, $src}",
7052 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
7054 let Predicates = [HasAVX2] in
7055 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
7056 "vmovntdqa\t{$src, $dst|$dst, $src}",
7057 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
7059 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
7060 "movntdqa\t{$src, $dst|$dst, $src}",
7061 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
7064 //===----------------------------------------------------------------------===//
7065 // SSE4.2 - Compare Instructions
7066 //===----------------------------------------------------------------------===//
7068 /// SS42I_binop_rm - Simple SSE 4.2 binary operator
7069 multiclass SS42I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
7070 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
7071 X86MemOperand x86memop, bit Is2Addr = 1> {
7072 def rr : SS428I<opc, MRMSrcReg, (outs RC:$dst),
7073 (ins RC:$src1, RC:$src2),
7075 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7076 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7077 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>,
7079 def rm : SS428I<opc, MRMSrcMem, (outs RC:$dst),
7080 (ins RC:$src1, x86memop:$src2),
7082 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7083 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7085 (OpVT (OpNode RC:$src1, (memop_frag addr:$src2))))]>, OpSize;
7088 let Predicates = [HasAVX] in
7089 defm VPCMPGTQ : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v2i64, VR128,
7090 memopv2i64, i128mem, 0>, VEX_4V;
7092 let Predicates = [HasAVX2] in
7093 defm VPCMPGTQY : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v4i64, VR256,
7094 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
7096 let Constraints = "$src1 = $dst" in
7097 defm PCMPGTQ : SS42I_binop_rm<0x37, "pcmpgtq", X86pcmpgt, v2i64, VR128,
7098 memopv2i64, i128mem>;
7100 //===----------------------------------------------------------------------===//
7101 // SSE4.2 - String/text Processing Instructions
7102 //===----------------------------------------------------------------------===//
7104 // Packed Compare Implicit Length Strings, Return Mask
7105 multiclass pseudo_pcmpistrm<string asm> {
7106 def REG : PseudoI<(outs VR128:$dst),
7107 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7108 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
7110 def MEM : PseudoI<(outs VR128:$dst),
7111 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7112 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1,
7113 (bc_v16i8 (memopv2i64 addr:$src2)), imm:$src3))]>;
7116 let Defs = [EFLAGS], usesCustomInserter = 1 in {
7117 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
7118 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[UseSSE42]>;
7121 multiclass pcmpistrm_SS42AI<string asm> {
7122 def rr : SS42AI<0x62, MRMSrcReg, (outs),
7123 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7124 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
7127 def rm :SS42AI<0x62, MRMSrcMem, (outs),
7128 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7129 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
7133 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
7134 let Predicates = [HasAVX] in
7135 defm VPCMPISTRM128 : pcmpistrm_SS42AI<"vpcmpistrm">, VEX;
7136 defm PCMPISTRM128 : pcmpistrm_SS42AI<"pcmpistrm"> ;
7139 // Packed Compare Explicit Length Strings, Return Mask
7140 multiclass pseudo_pcmpestrm<string asm> {
7141 def REG : PseudoI<(outs VR128:$dst),
7142 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7143 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
7144 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
7145 def MEM : PseudoI<(outs VR128:$dst),
7146 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7147 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128 VR128:$src1, EAX,
7148 (bc_v16i8 (memopv2i64 addr:$src3)), EDX, imm:$src5))]>;
7151 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
7152 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
7153 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[UseSSE42]>;
7156 multiclass SS42AI_pcmpestrm<string asm> {
7157 def rr : SS42AI<0x60, MRMSrcReg, (outs),
7158 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7159 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7162 def rm : SS42AI<0x60, MRMSrcMem, (outs),
7163 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7164 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7168 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
7169 let Predicates = [HasAVX] in
7170 defm VPCMPESTRM128 : SS42AI_pcmpestrm<"vpcmpestrm">, VEX;
7171 defm PCMPESTRM128 : SS42AI_pcmpestrm<"pcmpestrm">;
7174 // Packed Compare Implicit Length Strings, Return Index
7175 multiclass pseudo_pcmpistri<string asm> {
7176 def REG : PseudoI<(outs GR32:$dst),
7177 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7178 [(set GR32:$dst, EFLAGS,
7179 (X86pcmpistri VR128:$src1, VR128:$src2, imm:$src3))]>;
7180 def MEM : PseudoI<(outs GR32:$dst),
7181 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7182 [(set GR32:$dst, EFLAGS, (X86pcmpistri VR128:$src1,
7183 (bc_v16i8 (memopv2i64 addr:$src2)), imm:$src3))]>;
7186 let Defs = [EFLAGS], usesCustomInserter = 1 in {
7187 defm VPCMPISTRI : pseudo_pcmpistri<"#VPCMPISTRI">, Requires<[HasAVX]>;
7188 defm PCMPISTRI : pseudo_pcmpistri<"#PCMPISTRI">, Requires<[UseSSE42]>;
7191 multiclass SS42AI_pcmpistri<string asm> {
7192 def rr : SS42AI<0x63, MRMSrcReg, (outs),
7193 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7194 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
7197 def rm : SS42AI<0x63, MRMSrcMem, (outs),
7198 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7199 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
7203 let Defs = [ECX, EFLAGS], neverHasSideEffects = 1 in {
7204 let Predicates = [HasAVX] in
7205 defm VPCMPISTRI : SS42AI_pcmpistri<"vpcmpistri">, VEX;
7206 defm PCMPISTRI : SS42AI_pcmpistri<"pcmpistri">;
7209 // Packed Compare Explicit Length Strings, Return Index
7210 multiclass pseudo_pcmpestri<string asm> {
7211 def REG : PseudoI<(outs GR32:$dst),
7212 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7213 [(set GR32:$dst, EFLAGS,
7214 (X86pcmpestri VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
7215 def MEM : PseudoI<(outs GR32:$dst),
7216 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7217 [(set GR32:$dst, EFLAGS,
7218 (X86pcmpestri VR128:$src1, EAX, (bc_v16i8 (memopv2i64 addr:$src3)), EDX,
7222 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
7223 defm VPCMPESTRI : pseudo_pcmpestri<"#VPCMPESTRI">, Requires<[HasAVX]>;
7224 defm PCMPESTRI : pseudo_pcmpestri<"#PCMPESTRI">, Requires<[UseSSE42]>;
7227 multiclass SS42AI_pcmpestri<string asm> {
7228 def rr : SS42AI<0x61, MRMSrcReg, (outs),
7229 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7230 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7233 def rm : SS42AI<0x61, MRMSrcMem, (outs),
7234 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7235 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7239 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
7240 let Predicates = [HasAVX] in
7241 defm VPCMPESTRI : SS42AI_pcmpestri<"vpcmpestri">, VEX;
7242 defm PCMPESTRI : SS42AI_pcmpestri<"pcmpestri">;
7245 //===----------------------------------------------------------------------===//
7246 // SSE4.2 - CRC Instructions
7247 //===----------------------------------------------------------------------===//
7249 // No CRC instructions have AVX equivalents
7251 // crc intrinsic instruction
7252 // This set of instructions are only rm, the only difference is the size
7254 class SS42I_crc32r<bits<8> opc, string asm, RegisterClass RCOut,
7255 RegisterClass RCIn, SDPatternOperator Int> :
7256 SS42FI<opc, MRMSrcReg, (outs RCOut:$dst), (ins RCOut:$src1, RCIn:$src2),
7257 !strconcat(asm, "\t{$src2, $src1|$src1, $src2}"),
7258 [(set RCOut:$dst, (Int RCOut:$src1, RCIn:$src2))], IIC_CRC32_REG>;
7260 class SS42I_crc32m<bits<8> opc, string asm, RegisterClass RCOut,
7261 X86MemOperand x86memop, SDPatternOperator Int> :
7262 SS42FI<opc, MRMSrcMem, (outs RCOut:$dst), (ins RCOut:$src1, x86memop:$src2),
7263 !strconcat(asm, "\t{$src2, $src1|$src1, $src2}"),
7264 [(set RCOut:$dst, (Int RCOut:$src1, (load addr:$src2)))],
7267 let Constraints = "$src1 = $dst" in {
7268 def CRC32r32m8 : SS42I_crc32m<0xF0, "crc32{b}", GR32, i8mem,
7269 int_x86_sse42_crc32_32_8>;
7270 def CRC32r32r8 : SS42I_crc32r<0xF0, "crc32{b}", GR32, GR8,
7271 int_x86_sse42_crc32_32_8>;
7272 def CRC32r32m16 : SS42I_crc32m<0xF1, "crc32{w}", GR32, i16mem,
7273 int_x86_sse42_crc32_32_16>, OpSize;
7274 def CRC32r32r16 : SS42I_crc32r<0xF1, "crc32{w}", GR32, GR16,
7275 int_x86_sse42_crc32_32_16>, OpSize;
7276 def CRC32r32m32 : SS42I_crc32m<0xF1, "crc32{l}", GR32, i32mem,
7277 int_x86_sse42_crc32_32_32>;
7278 def CRC32r32r32 : SS42I_crc32r<0xF1, "crc32{l}", GR32, GR32,
7279 int_x86_sse42_crc32_32_32>;
7280 def CRC32r64m64 : SS42I_crc32m<0xF1, "crc32{q}", GR64, i64mem,
7281 int_x86_sse42_crc32_64_64>, REX_W;
7282 def CRC32r64r64 : SS42I_crc32r<0xF1, "crc32{q}", GR64, GR64,
7283 int_x86_sse42_crc32_64_64>, REX_W;
7284 let hasSideEffects = 0 in {
7286 def CRC32r64m8 : SS42I_crc32m<0xF0, "crc32{b}", GR64, i8mem,
7288 def CRC32r64r8 : SS42I_crc32r<0xF0, "crc32{b}", GR64, GR8,
7293 //===----------------------------------------------------------------------===//
7294 // SHA-NI Instructions
7295 //===----------------------------------------------------------------------===//
7297 multiclass SHAI_binop<bits<8> Opc, string OpcodeStr, Intrinsic IntId,
7299 def rr : I<Opc, MRMSrcReg, (outs VR128:$dst),
7300 (ins VR128:$src1, VR128:$src2),
7301 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7303 (set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0)),
7304 (set VR128:$dst, (IntId VR128:$src1, VR128:$src2)))]>, T8;
7306 def rm : I<Opc, MRMSrcMem, (outs VR128:$dst),
7307 (ins VR128:$src1, i128mem:$src2),
7308 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7310 (set VR128:$dst, (IntId VR128:$src1,
7311 (bc_v4i32 (memopv2i64 addr:$src2)), XMM0)),
7312 (set VR128:$dst, (IntId VR128:$src1,
7313 (bc_v4i32 (memopv2i64 addr:$src2)))))]>, T8;
7316 let Constraints = "$src1 = $dst", Predicates = [HasSHA] in {
7317 def SHA1RNDS4rri : Ii8<0xCC, MRMSrcReg, (outs VR128:$dst),
7318 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7319 "sha1rnds4\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7321 (int_x86_sha1rnds4 VR128:$src1, VR128:$src2,
7322 (i8 imm:$src3)))]>, TA;
7323 def SHA1RNDS4rmi : Ii8<0xCC, MRMSrcMem, (outs VR128:$dst),
7324 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7325 "sha1rnds4\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7327 (int_x86_sha1rnds4 VR128:$src1,
7328 (bc_v4i32 (memopv2i64 addr:$src2)),
7329 (i8 imm:$src3)))]>, TA;
7331 defm SHA1NEXTE : SHAI_binop<0xC8, "sha1nexte", int_x86_sha1nexte>;
7332 defm SHA1MSG1 : SHAI_binop<0xC9, "sha1msg1", int_x86_sha1msg1>;
7333 defm SHA1MSG2 : SHAI_binop<0xCA, "sha1msg2", int_x86_sha1msg2>;
7336 defm SHA256RNDS2 : SHAI_binop<0xCB, "sha256rnds2", int_x86_sha256rnds2, 1>;
7338 defm SHA256MSG1 : SHAI_binop<0xCC, "sha256msg1", int_x86_sha256msg1>;
7339 defm SHA256MSG2 : SHAI_binop<0xCD, "sha256msg2", int_x86_sha256msg2>;
7342 // Aliases with explicit %xmm0
7343 def : InstAlias<"sha256rnds2\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7344 (SHA256RNDS2rr VR128:$dst, VR128:$src2)>;
7345 def : InstAlias<"sha256rnds2\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7346 (SHA256RNDS2rm VR128:$dst, i128mem:$src2)>;
7348 //===----------------------------------------------------------------------===//
7349 // AES-NI Instructions
7350 //===----------------------------------------------------------------------===//
7352 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
7353 Intrinsic IntId128, bit Is2Addr = 1> {
7354 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
7355 (ins VR128:$src1, VR128:$src2),
7357 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7358 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7359 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
7361 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
7362 (ins VR128:$src1, i128mem:$src2),
7364 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7365 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7367 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
7370 // Perform One Round of an AES Encryption/Decryption Flow
7371 let Predicates = [HasAVX, HasAES] in {
7372 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
7373 int_x86_aesni_aesenc, 0>, VEX_4V;
7374 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
7375 int_x86_aesni_aesenclast, 0>, VEX_4V;
7376 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
7377 int_x86_aesni_aesdec, 0>, VEX_4V;
7378 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
7379 int_x86_aesni_aesdeclast, 0>, VEX_4V;
7382 let Constraints = "$src1 = $dst" in {
7383 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
7384 int_x86_aesni_aesenc>;
7385 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
7386 int_x86_aesni_aesenclast>;
7387 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
7388 int_x86_aesni_aesdec>;
7389 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
7390 int_x86_aesni_aesdeclast>;
7393 // Perform the AES InvMixColumn Transformation
7394 let Predicates = [HasAVX, HasAES] in {
7395 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7397 "vaesimc\t{$src1, $dst|$dst, $src1}",
7399 (int_x86_aesni_aesimc VR128:$src1))]>,
7401 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7402 (ins i128mem:$src1),
7403 "vaesimc\t{$src1, $dst|$dst, $src1}",
7404 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7407 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7409 "aesimc\t{$src1, $dst|$dst, $src1}",
7411 (int_x86_aesni_aesimc VR128:$src1))]>,
7413 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7414 (ins i128mem:$src1),
7415 "aesimc\t{$src1, $dst|$dst, $src1}",
7416 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7419 // AES Round Key Generation Assist
7420 let Predicates = [HasAVX, HasAES] in {
7421 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7422 (ins VR128:$src1, i8imm:$src2),
7423 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7425 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7427 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7428 (ins i128mem:$src1, i8imm:$src2),
7429 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7431 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7434 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7435 (ins VR128:$src1, i8imm:$src2),
7436 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7438 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7440 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7441 (ins i128mem:$src1, i8imm:$src2),
7442 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7444 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7447 //===----------------------------------------------------------------------===//
7448 // PCLMUL Instructions
7449 //===----------------------------------------------------------------------===//
7451 // AVX carry-less Multiplication instructions
7452 def VPCLMULQDQrr : AVXPCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7453 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7454 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7456 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
7458 def VPCLMULQDQrm : AVXPCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7459 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7460 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7461 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7462 (memopv2i64 addr:$src2), imm:$src3))]>;
7464 // Carry-less Multiplication instructions
7465 let Constraints = "$src1 = $dst" in {
7466 def PCLMULQDQrr : PCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7467 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7468 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7470 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))],
7471 IIC_SSE_PCLMULQDQ_RR>;
7473 def PCLMULQDQrm : PCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7474 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7475 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7476 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7477 (memopv2i64 addr:$src2), imm:$src3))],
7478 IIC_SSE_PCLMULQDQ_RM>;
7479 } // Constraints = "$src1 = $dst"
7482 multiclass pclmul_alias<string asm, int immop> {
7483 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7484 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
7486 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7487 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
7489 def : InstAlias<!strconcat("vpclmul", asm,
7490 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7491 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
7493 def : InstAlias<!strconcat("vpclmul", asm,
7494 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7495 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
7497 defm : pclmul_alias<"hqhq", 0x11>;
7498 defm : pclmul_alias<"hqlq", 0x01>;
7499 defm : pclmul_alias<"lqhq", 0x10>;
7500 defm : pclmul_alias<"lqlq", 0x00>;
7502 //===----------------------------------------------------------------------===//
7503 // SSE4A Instructions
7504 //===----------------------------------------------------------------------===//
7506 let Predicates = [HasSSE4A] in {
7508 let Constraints = "$src = $dst" in {
7509 def EXTRQI : Ii8<0x78, MRM0r, (outs VR128:$dst),
7510 (ins VR128:$src, i8imm:$len, i8imm:$idx),
7511 "extrq\t{$idx, $len, $src|$src, $len, $idx}",
7512 [(set VR128:$dst, (int_x86_sse4a_extrqi VR128:$src, imm:$len,
7513 imm:$idx))]>, TB, OpSize;
7514 def EXTRQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7515 (ins VR128:$src, VR128:$mask),
7516 "extrq\t{$mask, $src|$src, $mask}",
7517 [(set VR128:$dst, (int_x86_sse4a_extrq VR128:$src,
7518 VR128:$mask))]>, TB, OpSize;
7520 def INSERTQI : Ii8<0x78, MRMSrcReg, (outs VR128:$dst),
7521 (ins VR128:$src, VR128:$src2, i8imm:$len, i8imm:$idx),
7522 "insertq\t{$idx, $len, $src2, $src|$src, $src2, $len, $idx}",
7523 [(set VR128:$dst, (int_x86_sse4a_insertqi VR128:$src,
7524 VR128:$src2, imm:$len, imm:$idx))]>, XD;
7525 def INSERTQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7526 (ins VR128:$src, VR128:$mask),
7527 "insertq\t{$mask, $src|$src, $mask}",
7528 [(set VR128:$dst, (int_x86_sse4a_insertq VR128:$src,
7529 VR128:$mask))]>, XD;
7532 def MOVNTSS : I<0x2B, MRMDestMem, (outs), (ins f32mem:$dst, VR128:$src),
7533 "movntss\t{$src, $dst|$dst, $src}",
7534 [(int_x86_sse4a_movnt_ss addr:$dst, VR128:$src)]>, XS;
7536 def MOVNTSD : I<0x2B, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
7537 "movntsd\t{$src, $dst|$dst, $src}",
7538 [(int_x86_sse4a_movnt_sd addr:$dst, VR128:$src)]>, XD;
7541 //===----------------------------------------------------------------------===//
7543 //===----------------------------------------------------------------------===//
7545 //===----------------------------------------------------------------------===//
7546 // VBROADCAST - Load from memory and broadcast to all elements of the
7547 // destination operand
7549 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
7550 X86MemOperand x86memop, Intrinsic Int> :
7551 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7552 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7553 [(set RC:$dst, (Int addr:$src))]>, VEX;
7555 // AVX2 adds register forms
7556 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
7558 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7559 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7560 [(set RC:$dst, (Int VR128:$src))]>, VEX;
7562 let ExeDomain = SSEPackedSingle in {
7563 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
7564 int_x86_avx_vbroadcast_ss>;
7565 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
7566 int_x86_avx_vbroadcast_ss_256>, VEX_L;
7568 let ExeDomain = SSEPackedDouble in
7569 def VBROADCASTSDYrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
7570 int_x86_avx_vbroadcast_sd_256>, VEX_L;
7571 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
7572 int_x86_avx_vbroadcastf128_pd_256>, VEX_L;
7574 let ExeDomain = SSEPackedSingle in {
7575 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
7576 int_x86_avx2_vbroadcast_ss_ps>;
7577 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
7578 int_x86_avx2_vbroadcast_ss_ps_256>, VEX_L;
7580 let ExeDomain = SSEPackedDouble in
7581 def VBROADCASTSDYrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
7582 int_x86_avx2_vbroadcast_sd_pd_256>, VEX_L;
7584 let Predicates = [HasAVX2] in
7585 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
7586 int_x86_avx2_vbroadcasti128>, VEX_L;
7588 let Predicates = [HasAVX] in
7589 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
7590 (VBROADCASTF128 addr:$src)>;
7593 //===----------------------------------------------------------------------===//
7594 // VINSERTF128 - Insert packed floating-point values
7596 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7597 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
7598 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7599 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7602 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
7603 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
7604 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7608 let Predicates = [HasAVX] in {
7609 def : Pat<(vinsert128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7611 (VINSERTF128rr VR256:$src1, VR128:$src2,
7612 (INSERT_get_vinsert128_imm VR256:$ins))>;
7613 def : Pat<(vinsert128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7615 (VINSERTF128rr VR256:$src1, VR128:$src2,
7616 (INSERT_get_vinsert128_imm VR256:$ins))>;
7618 def : Pat<(vinsert128_insert:$ins (v8f32 VR256:$src1), (loadv4f32 addr:$src2),
7620 (VINSERTF128rm VR256:$src1, addr:$src2,
7621 (INSERT_get_vinsert128_imm VR256:$ins))>;
7622 def : Pat<(vinsert128_insert:$ins (v4f64 VR256:$src1), (loadv2f64 addr:$src2),
7624 (VINSERTF128rm VR256:$src1, addr:$src2,
7625 (INSERT_get_vinsert128_imm VR256:$ins))>;
7628 let Predicates = [HasAVX1Only] in {
7629 def : Pat<(vinsert128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7631 (VINSERTF128rr VR256:$src1, VR128:$src2,
7632 (INSERT_get_vinsert128_imm VR256:$ins))>;
7633 def : Pat<(vinsert128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7635 (VINSERTF128rr VR256:$src1, VR128:$src2,
7636 (INSERT_get_vinsert128_imm VR256:$ins))>;
7637 def : Pat<(vinsert128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7639 (VINSERTF128rr VR256:$src1, VR128:$src2,
7640 (INSERT_get_vinsert128_imm VR256:$ins))>;
7641 def : Pat<(vinsert128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7643 (VINSERTF128rr VR256:$src1, VR128:$src2,
7644 (INSERT_get_vinsert128_imm VR256:$ins))>;
7646 def : Pat<(vinsert128_insert:$ins (v4i64 VR256:$src1), (loadv2i64 addr:$src2),
7648 (VINSERTF128rm VR256:$src1, addr:$src2,
7649 (INSERT_get_vinsert128_imm VR256:$ins))>;
7650 def : Pat<(vinsert128_insert:$ins (v8i32 VR256:$src1),
7651 (bc_v4i32 (loadv2i64 addr:$src2)),
7653 (VINSERTF128rm VR256:$src1, addr:$src2,
7654 (INSERT_get_vinsert128_imm VR256:$ins))>;
7655 def : Pat<(vinsert128_insert:$ins (v32i8 VR256:$src1),
7656 (bc_v16i8 (loadv2i64 addr:$src2)),
7658 (VINSERTF128rm VR256:$src1, addr:$src2,
7659 (INSERT_get_vinsert128_imm VR256:$ins))>;
7660 def : Pat<(vinsert128_insert:$ins (v16i16 VR256:$src1),
7661 (bc_v8i16 (loadv2i64 addr:$src2)),
7663 (VINSERTF128rm VR256:$src1, addr:$src2,
7664 (INSERT_get_vinsert128_imm VR256:$ins))>;
7667 //===----------------------------------------------------------------------===//
7668 // VEXTRACTF128 - Extract packed floating-point values
7670 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7671 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
7672 (ins VR256:$src1, i8imm:$src2),
7673 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7676 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
7677 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
7678 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7683 let Predicates = [HasAVX] in {
7684 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7685 (v4f32 (VEXTRACTF128rr
7686 (v8f32 VR256:$src1),
7687 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7688 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7689 (v2f64 (VEXTRACTF128rr
7690 (v4f64 VR256:$src1),
7691 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7693 def : Pat<(store (v4f32 (vextract128_extract:$ext (v8f32 VR256:$src1),
7694 (iPTR imm))), addr:$dst),
7695 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7696 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7697 def : Pat<(store (v2f64 (vextract128_extract:$ext (v4f64 VR256:$src1),
7698 (iPTR imm))), addr:$dst),
7699 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7700 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7703 let Predicates = [HasAVX1Only] in {
7704 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7705 (v2i64 (VEXTRACTF128rr
7706 (v4i64 VR256:$src1),
7707 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7708 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7709 (v4i32 (VEXTRACTF128rr
7710 (v8i32 VR256:$src1),
7711 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7712 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7713 (v8i16 (VEXTRACTF128rr
7714 (v16i16 VR256:$src1),
7715 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7716 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7717 (v16i8 (VEXTRACTF128rr
7718 (v32i8 VR256:$src1),
7719 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7721 def : Pat<(alignedstore (v2i64 (vextract128_extract:$ext (v4i64 VR256:$src1),
7722 (iPTR imm))), addr:$dst),
7723 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7724 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7725 def : Pat<(alignedstore (v4i32 (vextract128_extract:$ext (v8i32 VR256:$src1),
7726 (iPTR imm))), addr:$dst),
7727 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7728 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7729 def : Pat<(alignedstore (v8i16 (vextract128_extract:$ext (v16i16 VR256:$src1),
7730 (iPTR imm))), addr:$dst),
7731 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7732 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7733 def : Pat<(alignedstore (v16i8 (vextract128_extract:$ext (v32i8 VR256:$src1),
7734 (iPTR imm))), addr:$dst),
7735 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7736 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7739 //===----------------------------------------------------------------------===//
7740 // VMASKMOV - Conditional SIMD Packed Loads and Stores
7742 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
7743 Intrinsic IntLd, Intrinsic IntLd256,
7744 Intrinsic IntSt, Intrinsic IntSt256> {
7745 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
7746 (ins VR128:$src1, f128mem:$src2),
7747 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7748 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
7750 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
7751 (ins VR256:$src1, f256mem:$src2),
7752 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7753 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7755 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
7756 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
7757 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7758 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7759 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
7760 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
7761 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7762 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V, VEX_L;
7765 let ExeDomain = SSEPackedSingle in
7766 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
7767 int_x86_avx_maskload_ps,
7768 int_x86_avx_maskload_ps_256,
7769 int_x86_avx_maskstore_ps,
7770 int_x86_avx_maskstore_ps_256>;
7771 let ExeDomain = SSEPackedDouble in
7772 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
7773 int_x86_avx_maskload_pd,
7774 int_x86_avx_maskload_pd_256,
7775 int_x86_avx_maskstore_pd,
7776 int_x86_avx_maskstore_pd_256>;
7778 //===----------------------------------------------------------------------===//
7779 // VPERMIL - Permute Single and Double Floating-Point Values
7781 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
7782 RegisterClass RC, X86MemOperand x86memop_f,
7783 X86MemOperand x86memop_i, PatFrag i_frag,
7784 Intrinsic IntVar, ValueType vt> {
7785 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
7786 (ins RC:$src1, RC:$src2),
7787 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7788 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
7789 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
7790 (ins RC:$src1, x86memop_i:$src2),
7791 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7792 [(set RC:$dst, (IntVar RC:$src1,
7793 (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
7795 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
7796 (ins RC:$src1, i8imm:$src2),
7797 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7798 [(set RC:$dst, (vt (X86VPermilp RC:$src1, (i8 imm:$src2))))]>, VEX;
7799 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
7800 (ins x86memop_f:$src1, i8imm:$src2),
7801 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7803 (vt (X86VPermilp (memop addr:$src1), (i8 imm:$src2))))]>, VEX;
7806 let ExeDomain = SSEPackedSingle in {
7807 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
7808 memopv2i64, int_x86_avx_vpermilvar_ps, v4f32>;
7809 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
7810 memopv4i64, int_x86_avx_vpermilvar_ps_256, v8f32>, VEX_L;
7812 let ExeDomain = SSEPackedDouble in {
7813 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
7814 memopv2i64, int_x86_avx_vpermilvar_pd, v2f64>;
7815 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
7816 memopv4i64, int_x86_avx_vpermilvar_pd_256, v4f64>, VEX_L;
7819 let Predicates = [HasAVX] in {
7820 def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7821 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7822 def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7823 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7824 def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (memopv4i64 addr:$src1)),
7826 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7827 def : Pat<(v4i64 (X86VPermilp (memopv4i64 addr:$src1), (i8 imm:$imm))),
7828 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7830 def : Pat<(v2i64 (X86VPermilp VR128:$src1, (i8 imm:$imm))),
7831 (VPERMILPDri VR128:$src1, imm:$imm)>;
7832 def : Pat<(v2i64 (X86VPermilp (memopv2i64 addr:$src1), (i8 imm:$imm))),
7833 (VPERMILPDmi addr:$src1, imm:$imm)>;
7836 //===----------------------------------------------------------------------===//
7837 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
7839 let ExeDomain = SSEPackedSingle in {
7840 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
7841 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7842 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7843 [(set VR256:$dst, (v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7844 (i8 imm:$src3))))]>, VEX_4V, VEX_L;
7845 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
7846 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7847 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7848 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv8f32 addr:$src2),
7849 (i8 imm:$src3)))]>, VEX_4V, VEX_L;
7852 let Predicates = [HasAVX] in {
7853 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7854 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7855 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
7856 (memopv4f64 addr:$src2), (i8 imm:$imm))),
7857 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7860 let Predicates = [HasAVX1Only] in {
7861 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7862 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7863 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7864 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7865 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7866 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7867 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7868 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7870 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
7871 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7872 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7873 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
7874 (memopv4i64 addr:$src2), (i8 imm:$imm))),
7875 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7876 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
7877 (bc_v32i8 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7878 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7879 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7880 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7881 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7884 //===----------------------------------------------------------------------===//
7885 // VZERO - Zero YMM registers
7887 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
7888 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
7889 // Zero All YMM registers
7890 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
7891 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
7893 // Zero Upper bits of YMM registers
7894 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
7895 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
7898 //===----------------------------------------------------------------------===//
7899 // Half precision conversion instructions
7900 //===----------------------------------------------------------------------===//
7901 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7902 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7903 "vcvtph2ps\t{$src, $dst|$dst, $src}",
7904 [(set RC:$dst, (Int VR128:$src))]>,
7906 let neverHasSideEffects = 1, mayLoad = 1 in
7907 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7908 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
7911 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7912 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
7913 (ins RC:$src1, i32i8imm:$src2),
7914 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7915 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
7917 let neverHasSideEffects = 1, mayStore = 1 in
7918 def mr : Ii8<0x1D, MRMDestMem, (outs),
7919 (ins x86memop:$dst, RC:$src1, i32i8imm:$src2),
7920 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
7924 let Predicates = [HasF16C] in {
7925 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
7926 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>, VEX_L;
7927 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
7928 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>, VEX_L;
7931 //===----------------------------------------------------------------------===//
7932 // AVX2 Instructions
7933 //===----------------------------------------------------------------------===//
7935 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
7936 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
7937 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
7938 X86MemOperand x86memop> {
7939 let isCommutable = 1 in
7940 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
7941 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
7942 !strconcat(OpcodeStr,
7943 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7944 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
7946 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
7947 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
7948 !strconcat(OpcodeStr,
7949 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7952 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
7956 let isCommutable = 0 in {
7957 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
7958 VR128, memopv2i64, i128mem>;
7959 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
7960 VR256, memopv4i64, i256mem>, VEX_L;
7963 def : Pat<(v4i32 (X86Blendi (v4i32 VR128:$src1), (v4i32 VR128:$src2),
7965 (VPBLENDDrri VR128:$src1, VR128:$src2, imm:$mask)>;
7966 def : Pat<(v8i32 (X86Blendi (v8i32 VR256:$src1), (v8i32 VR256:$src2),
7968 (VPBLENDDYrri VR256:$src1, VR256:$src2, imm:$mask)>;
7970 //===----------------------------------------------------------------------===//
7971 // VPBROADCAST - Load from memory and broadcast to all elements of the
7972 // destination operand
7974 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
7975 X86MemOperand x86memop, PatFrag ld_frag,
7976 Intrinsic Int128, Intrinsic Int256> {
7977 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
7978 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7979 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
7980 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
7981 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7983 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7984 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
7985 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7986 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX, VEX_L;
7987 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
7988 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7990 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>,
7994 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
7995 int_x86_avx2_pbroadcastb_128,
7996 int_x86_avx2_pbroadcastb_256>;
7997 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
7998 int_x86_avx2_pbroadcastw_128,
7999 int_x86_avx2_pbroadcastw_256>;
8000 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
8001 int_x86_avx2_pbroadcastd_128,
8002 int_x86_avx2_pbroadcastd_256>;
8003 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
8004 int_x86_avx2_pbroadcastq_128,
8005 int_x86_avx2_pbroadcastq_256>;
8007 let Predicates = [HasAVX2] in {
8008 def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
8009 (VPBROADCASTBrm addr:$src)>;
8010 def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
8011 (VPBROADCASTBYrm addr:$src)>;
8012 def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
8013 (VPBROADCASTWrm addr:$src)>;
8014 def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
8015 (VPBROADCASTWYrm addr:$src)>;
8016 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
8017 (VPBROADCASTDrm addr:$src)>;
8018 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
8019 (VPBROADCASTDYrm addr:$src)>;
8020 def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
8021 (VPBROADCASTQrm addr:$src)>;
8022 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
8023 (VPBROADCASTQYrm addr:$src)>;
8025 def : Pat<(v16i8 (X86VBroadcast (v16i8 VR128:$src))),
8026 (VPBROADCASTBrr VR128:$src)>;
8027 def : Pat<(v32i8 (X86VBroadcast (v16i8 VR128:$src))),
8028 (VPBROADCASTBYrr VR128:$src)>;
8029 def : Pat<(v8i16 (X86VBroadcast (v8i16 VR128:$src))),
8030 (VPBROADCASTWrr VR128:$src)>;
8031 def : Pat<(v16i16 (X86VBroadcast (v8i16 VR128:$src))),
8032 (VPBROADCASTWYrr VR128:$src)>;
8033 def : Pat<(v4i32 (X86VBroadcast (v4i32 VR128:$src))),
8034 (VPBROADCASTDrr VR128:$src)>;
8035 def : Pat<(v8i32 (X86VBroadcast (v4i32 VR128:$src))),
8036 (VPBROADCASTDYrr VR128:$src)>;
8037 def : Pat<(v2i64 (X86VBroadcast (v2i64 VR128:$src))),
8038 (VPBROADCASTQrr VR128:$src)>;
8039 def : Pat<(v4i64 (X86VBroadcast (v2i64 VR128:$src))),
8040 (VPBROADCASTQYrr VR128:$src)>;
8041 def : Pat<(v4f32 (X86VBroadcast (v4f32 VR128:$src))),
8042 (VBROADCASTSSrr VR128:$src)>;
8043 def : Pat<(v8f32 (X86VBroadcast (v4f32 VR128:$src))),
8044 (VBROADCASTSSYrr VR128:$src)>;
8045 def : Pat<(v2f64 (X86VBroadcast (v2f64 VR128:$src))),
8046 (VPBROADCASTQrr VR128:$src)>;
8047 def : Pat<(v4f64 (X86VBroadcast (v2f64 VR128:$src))),
8048 (VBROADCASTSDYrr VR128:$src)>;
8050 // Provide fallback in case the load node that is used in the patterns above
8051 // is used by additional users, which prevents the pattern selection.
8052 let AddedComplexity = 20 in {
8053 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
8054 (VBROADCASTSSrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
8055 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
8056 (VBROADCASTSSYrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
8057 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
8058 (VBROADCASTSDYrr (COPY_TO_REGCLASS FR64:$src, VR128))>;
8060 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
8061 (VBROADCASTSSrr (COPY_TO_REGCLASS GR32:$src, VR128))>;
8062 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
8063 (VBROADCASTSSYrr (COPY_TO_REGCLASS GR32:$src, VR128))>;
8064 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
8065 (VBROADCASTSDYrr (COPY_TO_REGCLASS GR64:$src, VR128))>;
8069 // AVX1 broadcast patterns
8070 let Predicates = [HasAVX1Only] in {
8071 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
8072 (VBROADCASTSSYrm addr:$src)>;
8073 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
8074 (VBROADCASTSDYrm addr:$src)>;
8075 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
8076 (VBROADCASTSSrm addr:$src)>;
8079 let Predicates = [HasAVX] in {
8080 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
8081 (VBROADCASTSSYrm addr:$src)>;
8082 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
8083 (VBROADCASTSDYrm addr:$src)>;
8084 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
8085 (VBROADCASTSSrm addr:$src)>;
8087 // Provide fallback in case the load node that is used in the patterns above
8088 // is used by additional users, which prevents the pattern selection.
8089 let AddedComplexity = 20 in {
8090 // 128bit broadcasts:
8091 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
8092 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0)>;
8093 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
8094 (VINSERTF128rr (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)),
8095 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0), sub_xmm),
8096 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0), 1)>;
8097 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
8098 (VINSERTF128rr (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)),
8099 (VPSHUFDri (COPY_TO_REGCLASS FR64:$src, VR128), 0x44), sub_xmm),
8100 (VPSHUFDri (COPY_TO_REGCLASS FR64:$src, VR128), 0x44), 1)>;
8102 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
8103 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0)>;
8104 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
8105 (VINSERTF128rr (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)),
8106 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0), sub_xmm),
8107 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0), 1)>;
8108 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
8109 (VINSERTF128rr (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)),
8110 (VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), sub_xmm),
8111 (VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), 1)>;
8115 //===----------------------------------------------------------------------===//
8116 // VPERM - Permute instructions
8119 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
8121 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
8122 (ins VR256:$src1, VR256:$src2),
8123 !strconcat(OpcodeStr,
8124 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8126 (OpVT (X86VPermv VR256:$src1, VR256:$src2)))]>,
8128 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
8129 (ins VR256:$src1, i256mem:$src2),
8130 !strconcat(OpcodeStr,
8131 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8133 (OpVT (X86VPermv VR256:$src1,
8134 (bitconvert (mem_frag addr:$src2)))))]>,
8138 defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, v8i32>;
8139 let ExeDomain = SSEPackedSingle in
8140 defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, v8f32>;
8142 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
8144 def Yri : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
8145 (ins VR256:$src1, i8imm:$src2),
8146 !strconcat(OpcodeStr,
8147 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8149 (OpVT (X86VPermi VR256:$src1, (i8 imm:$src2))))]>,
8151 def Ymi : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
8152 (ins i256mem:$src1, i8imm:$src2),
8153 !strconcat(OpcodeStr,
8154 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8156 (OpVT (X86VPermi (mem_frag addr:$src1),
8157 (i8 imm:$src2))))]>, VEX, VEX_L;
8160 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, v4i64>, VEX_W;
8161 let ExeDomain = SSEPackedDouble in
8162 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, v4f64>, VEX_W;
8164 //===----------------------------------------------------------------------===//
8165 // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
8167 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
8168 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
8169 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
8170 [(set VR256:$dst, (v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2,
8171 (i8 imm:$src3))))]>, VEX_4V, VEX_L;
8172 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
8173 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
8174 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
8175 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
8176 (i8 imm:$src3)))]>, VEX_4V, VEX_L;
8178 let Predicates = [HasAVX2] in {
8179 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
8180 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
8181 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
8182 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
8183 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
8184 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
8186 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (memopv4i64 addr:$src2)),
8188 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
8189 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
8190 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
8191 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
8192 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)),
8194 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
8198 //===----------------------------------------------------------------------===//
8199 // VINSERTI128 - Insert packed integer values
8201 let neverHasSideEffects = 1 in {
8202 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
8203 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
8204 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
8207 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
8208 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
8209 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
8213 let Predicates = [HasAVX2] in {
8214 def : Pat<(vinsert128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
8216 (VINSERTI128rr VR256:$src1, VR128:$src2,
8217 (INSERT_get_vinsert128_imm VR256:$ins))>;
8218 def : Pat<(vinsert128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
8220 (VINSERTI128rr VR256:$src1, VR128:$src2,
8221 (INSERT_get_vinsert128_imm VR256:$ins))>;
8222 def : Pat<(vinsert128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
8224 (VINSERTI128rr VR256:$src1, VR128:$src2,
8225 (INSERT_get_vinsert128_imm VR256:$ins))>;
8226 def : Pat<(vinsert128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
8228 (VINSERTI128rr VR256:$src1, VR128:$src2,
8229 (INSERT_get_vinsert128_imm VR256:$ins))>;
8231 def : Pat<(vinsert128_insert:$ins (v4i64 VR256:$src1), (loadv2i64 addr:$src2),
8233 (VINSERTI128rm VR256:$src1, addr:$src2,
8234 (INSERT_get_vinsert128_imm VR256:$ins))>;
8235 def : Pat<(vinsert128_insert:$ins (v8i32 VR256:$src1),
8236 (bc_v4i32 (loadv2i64 addr:$src2)),
8238 (VINSERTI128rm VR256:$src1, addr:$src2,
8239 (INSERT_get_vinsert128_imm VR256:$ins))>;
8240 def : Pat<(vinsert128_insert:$ins (v32i8 VR256:$src1),
8241 (bc_v16i8 (loadv2i64 addr:$src2)),
8243 (VINSERTI128rm VR256:$src1, addr:$src2,
8244 (INSERT_get_vinsert128_imm VR256:$ins))>;
8245 def : Pat<(vinsert128_insert:$ins (v16i16 VR256:$src1),
8246 (bc_v8i16 (loadv2i64 addr:$src2)),
8248 (VINSERTI128rm VR256:$src1, addr:$src2,
8249 (INSERT_get_vinsert128_imm VR256:$ins))>;
8252 //===----------------------------------------------------------------------===//
8253 // VEXTRACTI128 - Extract packed integer values
8255 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
8256 (ins VR256:$src1, i8imm:$src2),
8257 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
8259 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
8261 let neverHasSideEffects = 1, mayStore = 1 in
8262 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
8263 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
8264 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
8267 let Predicates = [HasAVX2] in {
8268 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
8269 (v2i64 (VEXTRACTI128rr
8270 (v4i64 VR256:$src1),
8271 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
8272 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
8273 (v4i32 (VEXTRACTI128rr
8274 (v8i32 VR256:$src1),
8275 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
8276 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
8277 (v8i16 (VEXTRACTI128rr
8278 (v16i16 VR256:$src1),
8279 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
8280 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
8281 (v16i8 (VEXTRACTI128rr
8282 (v32i8 VR256:$src1),
8283 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
8285 def : Pat<(store (v2i64 (vextract128_extract:$ext (v4i64 VR256:$src1),
8286 (iPTR imm))), addr:$dst),
8287 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8288 (EXTRACT_get_vextract128_imm VR128:$ext))>;
8289 def : Pat<(store (v4i32 (vextract128_extract:$ext (v8i32 VR256:$src1),
8290 (iPTR imm))), addr:$dst),
8291 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8292 (EXTRACT_get_vextract128_imm VR128:$ext))>;
8293 def : Pat<(store (v8i16 (vextract128_extract:$ext (v16i16 VR256:$src1),
8294 (iPTR imm))), addr:$dst),
8295 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8296 (EXTRACT_get_vextract128_imm VR128:$ext))>;
8297 def : Pat<(store (v16i8 (vextract128_extract:$ext (v32i8 VR256:$src1),
8298 (iPTR imm))), addr:$dst),
8299 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8300 (EXTRACT_get_vextract128_imm VR128:$ext))>;
8303 //===----------------------------------------------------------------------===//
8304 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
8306 multiclass avx2_pmovmask<string OpcodeStr,
8307 Intrinsic IntLd128, Intrinsic IntLd256,
8308 Intrinsic IntSt128, Intrinsic IntSt256> {
8309 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
8310 (ins VR128:$src1, i128mem:$src2),
8311 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8312 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
8313 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
8314 (ins VR256:$src1, i256mem:$src2),
8315 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8316 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
8318 def mr : AVX28I<0x8e, MRMDestMem, (outs),
8319 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
8320 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8321 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
8322 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
8323 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
8324 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8325 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V, VEX_L;
8328 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
8329 int_x86_avx2_maskload_d,
8330 int_x86_avx2_maskload_d_256,
8331 int_x86_avx2_maskstore_d,
8332 int_x86_avx2_maskstore_d_256>;
8333 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
8334 int_x86_avx2_maskload_q,
8335 int_x86_avx2_maskload_q_256,
8336 int_x86_avx2_maskstore_q,
8337 int_x86_avx2_maskstore_q_256>, VEX_W;
8340 //===----------------------------------------------------------------------===//
8341 // Variable Bit Shifts
8343 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
8344 ValueType vt128, ValueType vt256> {
8345 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
8346 (ins VR128:$src1, VR128:$src2),
8347 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8349 (vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
8351 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
8352 (ins VR128:$src1, i128mem:$src2),
8353 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8355 (vt128 (OpNode VR128:$src1,
8356 (vt128 (bitconvert (memopv2i64 addr:$src2))))))]>,
8358 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
8359 (ins VR256:$src1, VR256:$src2),
8360 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8362 (vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
8364 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
8365 (ins VR256:$src1, i256mem:$src2),
8366 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8368 (vt256 (OpNode VR256:$src1,
8369 (vt256 (bitconvert (memopv4i64 addr:$src2))))))]>,
8373 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
8374 defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", shl, v2i64, v4i64>, VEX_W;
8375 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", srl, v4i32, v8i32>;
8376 defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", srl, v2i64, v4i64>, VEX_W;
8377 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;
8379 //===----------------------------------------------------------------------===//
8380 // VGATHER - GATHER Operations
8381 multiclass avx2_gather<bits<8> opc, string OpcodeStr, RegisterClass RC256,
8382 X86MemOperand memop128, X86MemOperand memop256> {
8383 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst, VR128:$mask_wb),
8384 (ins VR128:$src1, memop128:$src2, VR128:$mask),
8385 !strconcat(OpcodeStr,
8386 "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
8388 def Yrm : AVX28I<opc, MRMSrcMem, (outs RC256:$dst, RC256:$mask_wb),
8389 (ins RC256:$src1, memop256:$src2, RC256:$mask),
8390 !strconcat(OpcodeStr,
8391 "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
8392 []>, VEX_4VOp3, VEX_L;
8395 let mayLoad = 1, Constraints
8396 = "@earlyclobber $dst,@earlyclobber $mask_wb, $src1 = $dst, $mask = $mask_wb"
8398 defm VGATHERDPD : avx2_gather<0x92, "vgatherdpd", VR256, vx64mem, vx64mem>, VEX_W;
8399 defm VGATHERQPD : avx2_gather<0x93, "vgatherqpd", VR256, vx64mem, vy64mem>, VEX_W;
8400 defm VGATHERDPS : avx2_gather<0x92, "vgatherdps", VR256, vx32mem, vy32mem>;
8401 defm VGATHERQPS : avx2_gather<0x93, "vgatherqps", VR128, vx32mem, vy32mem>;
8402 defm VPGATHERDQ : avx2_gather<0x90, "vpgatherdq", VR256, vx64mem, vx64mem>, VEX_W;
8403 defm VPGATHERQQ : avx2_gather<0x91, "vpgatherqq", VR256, vx64mem, vy64mem>, VEX_W;
8404 defm VPGATHERDD : avx2_gather<0x90, "vpgatherdd", VR256, vx32mem, vy32mem>;
8405 defm VPGATHERQD : avx2_gather<0x91, "vpgatherqd", VR128, vx32mem, vy32mem>;