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:$src)>;
2702 //===----------------------------------------------------------------------===//
2703 // SSE 1 & 2 - Extract Floating-Point Sign mask
2704 //===----------------------------------------------------------------------===//
2706 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2707 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2709 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
2710 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2711 [(set GR32:$dst, (Int RC:$src))], IIC_SSE_MOVMSK, d>,
2712 Sched<[WriteVecLogic]>;
2713 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
2714 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [],
2715 IIC_SSE_MOVMSK, d>, REX_W, Sched<[WriteVecLogic]>;
2718 let Predicates = [HasAVX] in {
2719 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2720 "movmskps", SSEPackedSingle>, TB, VEX;
2721 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2722 "movmskpd", SSEPackedDouble>, TB,
2724 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2725 "movmskps", SSEPackedSingle>, TB,
2727 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2728 "movmskpd", SSEPackedDouble>, TB,
2731 def : Pat<(i32 (X86fgetsign FR32:$src)),
2732 (VMOVMSKPSrr32 (COPY_TO_REGCLASS FR32:$src, VR128))>;
2733 def : Pat<(i64 (X86fgetsign FR32:$src)),
2734 (SUBREG_TO_REG (i64 0),
2735 (VMOVMSKPSrr32 (COPY_TO_REGCLASS FR32:$src, VR128)), sub_32bit)>;
2736 def : Pat<(i32 (X86fgetsign FR64:$src)),
2737 (VMOVMSKPDrr32 (COPY_TO_REGCLASS FR64:$src, VR128))>;
2738 def : Pat<(i64 (X86fgetsign FR64:$src)),
2739 (SUBREG_TO_REG (i64 0),
2740 (VMOVMSKPDrr32 (COPY_TO_REGCLASS FR64:$src, VR128)), sub_32bit)>;
2743 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2744 SSEPackedSingle>, TB;
2745 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2746 SSEPackedDouble>, TB, OpSize;
2748 def : Pat<(i32 (X86fgetsign FR32:$src)),
2749 (MOVMSKPSrr32 (COPY_TO_REGCLASS FR32:$src, VR128))>,
2750 Requires<[UseSSE1]>;
2751 def : Pat<(i64 (X86fgetsign FR32:$src)),
2752 (SUBREG_TO_REG (i64 0),
2753 (MOVMSKPSrr32 (COPY_TO_REGCLASS FR32:$src, VR128)), sub_32bit)>,
2754 Requires<[UseSSE1]>;
2755 def : Pat<(i32 (X86fgetsign FR64:$src)),
2756 (MOVMSKPDrr32 (COPY_TO_REGCLASS FR64:$src, VR128))>,
2757 Requires<[UseSSE2]>;
2758 def : Pat<(i64 (X86fgetsign FR64:$src)),
2759 (SUBREG_TO_REG (i64 0),
2760 (MOVMSKPDrr32 (COPY_TO_REGCLASS FR64:$src, VR128)), sub_32bit)>,
2761 Requires<[UseSSE2]>;
2763 //===---------------------------------------------------------------------===//
2764 // SSE2 - Packed Integer Logical Instructions
2765 //===---------------------------------------------------------------------===//
2767 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2769 /// PDI_binop_rm - Simple SSE2 binary operator.
2770 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2771 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
2772 X86MemOperand x86memop, OpndItins itins,
2773 bit IsCommutable, bit Is2Addr> {
2774 let isCommutable = IsCommutable in
2775 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
2776 (ins RC:$src1, RC:$src2),
2778 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2779 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2780 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>,
2781 Sched<[itins.Sched]>;
2782 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
2783 (ins RC:$src1, x86memop:$src2),
2785 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2786 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2787 [(set RC:$dst, (OpVT (OpNode RC:$src1,
2788 (bitconvert (memop_frag addr:$src2)))))],
2790 Sched<[itins.Sched.Folded, ReadAfterLd]>;
2792 } // ExeDomain = SSEPackedInt
2794 multiclass PDI_binop_all<bits<8> opc, string OpcodeStr, SDNode Opcode,
2795 ValueType OpVT128, ValueType OpVT256,
2796 OpndItins itins, bit IsCommutable = 0> {
2797 let Predicates = [HasAVX] in
2798 defm V#NAME : PDI_binop_rm<opc, !strconcat("v", OpcodeStr), Opcode, OpVT128,
2799 VR128, memopv2i64, i128mem, itins, IsCommutable, 0>, VEX_4V;
2801 let Constraints = "$src1 = $dst" in
2802 defm NAME : PDI_binop_rm<opc, OpcodeStr, Opcode, OpVT128, VR128,
2803 memopv2i64, i128mem, itins, IsCommutable, 1>;
2805 let Predicates = [HasAVX2] in
2806 defm V#NAME#Y : PDI_binop_rm<opc, !strconcat("v", OpcodeStr), Opcode,
2807 OpVT256, VR256, memopv4i64, i256mem, itins,
2808 IsCommutable, 0>, VEX_4V, VEX_L;
2811 // These are ordered here for pattern ordering requirements with the fp versions
2813 defm PAND : PDI_binop_all<0xDB, "pand", and, v2i64, v4i64, SSE_BIT_ITINS_P, 1>;
2814 defm POR : PDI_binop_all<0xEB, "por", or, v2i64, v4i64, SSE_BIT_ITINS_P, 1>;
2815 defm PXOR : PDI_binop_all<0xEF, "pxor", xor, v2i64, v4i64, SSE_BIT_ITINS_P, 1>;
2816 defm PANDN : PDI_binop_all<0xDF, "pandn", X86andnp, v2i64, v4i64,
2817 SSE_BIT_ITINS_P, 0>;
2819 //===----------------------------------------------------------------------===//
2820 // SSE 1 & 2 - Logical Instructions
2821 //===----------------------------------------------------------------------===//
2823 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2825 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2826 SDNode OpNode, OpndItins itins> {
2827 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2828 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, itins, 0>,
2831 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2832 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, itins, 0>,
2835 let Constraints = "$src1 = $dst" in {
2836 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2837 f32, f128mem, memopfsf32, SSEPackedSingle, itins>,
2840 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2841 f64, f128mem, memopfsf64, SSEPackedDouble, itins>,
2846 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2847 let isCodeGenOnly = 1 in {
2848 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand,
2850 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for,
2852 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor,
2855 let isCommutable = 0 in
2856 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", X86fandn,
2860 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2862 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2864 defm V#NAME#PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2865 !strconcat(OpcodeStr, "ps"), f256mem,
2866 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2867 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2868 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V, VEX_L;
2870 defm V#NAME#PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2871 !strconcat(OpcodeStr, "pd"), f256mem,
2872 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2873 (bc_v4i64 (v4f64 VR256:$src2))))],
2874 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2875 (memopv4i64 addr:$src2)))], 0>,
2876 TB, OpSize, VEX_4V, VEX_L;
2878 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2879 // are all promoted to v2i64, and the patterns are covered by the int
2880 // version. This is needed in SSE only, because v2i64 isn't supported on
2881 // SSE1, but only on SSE2.
2882 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2883 !strconcat(OpcodeStr, "ps"), f128mem, [],
2884 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2885 (memopv2i64 addr:$src2)))], 0>, TB, VEX_4V;
2887 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2888 !strconcat(OpcodeStr, "pd"), f128mem,
2889 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2890 (bc_v2i64 (v2f64 VR128:$src2))))],
2891 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2892 (memopv2i64 addr:$src2)))], 0>,
2895 let Constraints = "$src1 = $dst" in {
2896 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2897 !strconcat(OpcodeStr, "ps"), f128mem,
2898 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2899 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2900 (memopv2i64 addr:$src2)))]>, TB;
2902 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2903 !strconcat(OpcodeStr, "pd"), f128mem,
2904 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2905 (bc_v2i64 (v2f64 VR128:$src2))))],
2906 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2907 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2911 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2912 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2913 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2914 let isCommutable = 0 in
2915 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2917 //===----------------------------------------------------------------------===//
2918 // SSE 1 & 2 - Arithmetic Instructions
2919 //===----------------------------------------------------------------------===//
2921 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2924 /// In addition, we also have a special variant of the scalar form here to
2925 /// represent the associated intrinsic operation. This form is unlike the
2926 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2927 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2929 /// These three forms can each be reg+reg or reg+mem.
2932 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2934 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr,
2935 SDNode OpNode, SizeItins itins> {
2936 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2937 VR128, v4f32, f128mem, memopv4f32,
2938 SSEPackedSingle, itins.s, 0>, TB, VEX_4V;
2939 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2940 VR128, v2f64, f128mem, memopv2f64,
2941 SSEPackedDouble, itins.d, 0>, TB, OpSize, VEX_4V;
2943 defm V#NAME#PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"),
2944 OpNode, VR256, v8f32, f256mem, memopv8f32,
2945 SSEPackedSingle, itins.s, 0>, TB, VEX_4V, VEX_L;
2946 defm V#NAME#PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"),
2947 OpNode, VR256, v4f64, f256mem, memopv4f64,
2948 SSEPackedDouble, itins.d, 0>, TB, OpSize, VEX_4V, VEX_L;
2950 let Constraints = "$src1 = $dst" in {
2951 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2952 v4f32, f128mem, memopv4f32, SSEPackedSingle,
2954 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2955 v2f64, f128mem, memopv2f64, SSEPackedDouble,
2956 itins.d>, TB, OpSize;
2960 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2962 defm V#NAME#SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2963 OpNode, FR32, f32mem, itins.s, 0>, XS, VEX_4V, VEX_LIG;
2964 defm V#NAME#SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2965 OpNode, FR64, f64mem, itins.d, 0>, XD, VEX_4V, VEX_LIG;
2967 let Constraints = "$src1 = $dst" in {
2968 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2969 OpNode, FR32, f32mem, itins.s>, XS;
2970 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2971 OpNode, FR64, f64mem, itins.d>, XD;
2975 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2977 defm V#NAME#SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2978 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32,
2979 itins.s, 0>, XS, VEX_4V, VEX_LIG;
2980 defm V#NAME#SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2981 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64,
2982 itins.d, 0>, XD, VEX_4V, VEX_LIG;
2984 let Constraints = "$src1 = $dst" in {
2985 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2986 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32,
2988 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2989 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64,
2994 // Binary Arithmetic instructions
2995 defm ADD : basic_sse12_fp_binop_p<0x58, "add", fadd, SSE_ALU_ITINS_P>,
2996 basic_sse12_fp_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S>,
2997 basic_sse12_fp_binop_s_int<0x58, "add", SSE_ALU_ITINS_S>;
2998 defm MUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, SSE_MUL_ITINS_P>,
2999 basic_sse12_fp_binop_s<0x59, "mul", fmul, SSE_MUL_ITINS_S>,
3000 basic_sse12_fp_binop_s_int<0x59, "mul", SSE_MUL_ITINS_S>;
3001 let isCommutable = 0 in {
3002 defm SUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, SSE_ALU_ITINS_P>,
3003 basic_sse12_fp_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S>,
3004 basic_sse12_fp_binop_s_int<0x5C, "sub", SSE_ALU_ITINS_S>;
3005 defm DIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, SSE_DIV_ITINS_P>,
3006 basic_sse12_fp_binop_s<0x5E, "div", fdiv, SSE_DIV_ITINS_S>,
3007 basic_sse12_fp_binop_s_int<0x5E, "div", SSE_DIV_ITINS_S>;
3008 defm MAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, SSE_ALU_ITINS_P>,
3009 basic_sse12_fp_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S>,
3010 basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S>;
3011 defm MIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, SSE_ALU_ITINS_P>,
3012 basic_sse12_fp_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S>,
3013 basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S>;
3016 let isCodeGenOnly = 1 in {
3017 defm MAXC: basic_sse12_fp_binop_p<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_P>,
3018 basic_sse12_fp_binop_s<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_S>;
3019 defm MINC: basic_sse12_fp_binop_p<0x5D, "min", X86fminc, SSE_ALU_ITINS_P>,
3020 basic_sse12_fp_binop_s<0x5D, "min", X86fminc, SSE_ALU_ITINS_S>;
3024 /// In addition, we also have a special variant of the scalar form here to
3025 /// represent the associated intrinsic operation. This form is unlike the
3026 /// plain scalar form, in that it takes an entire vector (instead of a
3027 /// scalar) and leaves the top elements undefined.
3029 /// And, we have a special variant form for a full-vector intrinsic form.
3031 let Sched = WriteFSqrt in {
3032 def SSE_SQRTPS : OpndItins<
3033 IIC_SSE_SQRTPS_RR, IIC_SSE_SQRTPS_RM
3036 def SSE_SQRTSS : OpndItins<
3037 IIC_SSE_SQRTSS_RR, IIC_SSE_SQRTSS_RM
3040 def SSE_SQRTPD : OpndItins<
3041 IIC_SSE_SQRTPD_RR, IIC_SSE_SQRTPD_RM
3044 def SSE_SQRTSD : OpndItins<
3045 IIC_SSE_SQRTSD_RR, IIC_SSE_SQRTSD_RM
3049 let Sched = WriteFRcp in {
3050 def SSE_RCPP : OpndItins<
3051 IIC_SSE_RCPP_RR, IIC_SSE_RCPP_RM
3054 def SSE_RCPS : OpndItins<
3055 IIC_SSE_RCPS_RR, IIC_SSE_RCPS_RM
3059 /// sse1_fp_unop_s - SSE1 unops in scalar form.
3060 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
3061 SDNode OpNode, Intrinsic F32Int, OpndItins itins> {
3062 let Predicates = [HasAVX], hasSideEffects = 0 in {
3063 def V#NAME#SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst),
3064 (ins FR32:$src1, FR32:$src2),
3065 !strconcat("v", OpcodeStr,
3066 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3067 []>, VEX_4V, VEX_LIG, Sched<[itins.Sched]>;
3068 let mayLoad = 1 in {
3069 def V#NAME#SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst),
3070 (ins FR32:$src1,f32mem:$src2),
3071 !strconcat("v", OpcodeStr,
3072 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3073 []>, VEX_4V, VEX_LIG,
3074 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3075 def V#NAME#SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3076 (ins VR128:$src1, ssmem:$src2),
3077 !strconcat("v", OpcodeStr,
3078 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3079 []>, VEX_4V, VEX_LIG,
3080 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3084 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
3085 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3086 [(set FR32:$dst, (OpNode FR32:$src))]>, Sched<[itins.Sched]>;
3087 // For scalar unary operations, fold a load into the operation
3088 // only in OptForSize mode. It eliminates an instruction, but it also
3089 // eliminates a whole-register clobber (the load), so it introduces a
3090 // partial register update condition.
3091 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
3092 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3093 [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
3094 Requires<[UseSSE1, OptForSize]>, Sched<[itins.Sched.Folded]>;
3095 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3096 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3097 [(set VR128:$dst, (F32Int VR128:$src))], itins.rr>,
3098 Sched<[itins.Sched]>;
3099 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
3100 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3101 [(set VR128:$dst, (F32Int sse_load_f32:$src))], itins.rm>,
3102 Sched<[itins.Sched.Folded]>;
3105 /// sse1_fp_unop_s_rw - SSE1 unops where vector form has a read-write operand.
3106 multiclass sse1_fp_unop_rw<bits<8> opc, string OpcodeStr, SDNode OpNode,
3108 let Predicates = [HasAVX], hasSideEffects = 0 in {
3109 def V#NAME#SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst),
3110 (ins FR32:$src1, FR32:$src2),
3111 !strconcat("v", OpcodeStr,
3112 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3113 []>, VEX_4V, VEX_LIG, Sched<[itins.Sched]>;
3114 let mayLoad = 1 in {
3115 def V#NAME#SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst),
3116 (ins FR32:$src1,f32mem:$src2),
3117 !strconcat("v", OpcodeStr,
3118 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3119 []>, VEX_4V, VEX_LIG,
3120 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3121 def V#NAME#SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3122 (ins VR128:$src1, ssmem:$src2),
3123 !strconcat("v", OpcodeStr,
3124 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3125 []>, VEX_4V, VEX_LIG,
3126 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3130 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
3131 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3132 [(set FR32:$dst, (OpNode FR32:$src))]>, Sched<[itins.Sched]>;
3133 // For scalar unary operations, fold a load into the operation
3134 // only in OptForSize mode. It eliminates an instruction, but it also
3135 // eliminates a whole-register clobber (the load), so it introduces a
3136 // partial register update condition.
3137 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
3138 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3139 [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
3140 Requires<[UseSSE1, OptForSize]>, Sched<[itins.Sched.Folded]>;
3141 let Constraints = "$src1 = $dst" in {
3142 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst),
3143 (ins VR128:$src1, VR128:$src2),
3144 !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
3145 [], itins.rr>, Sched<[itins.Sched]>;
3146 let mayLoad = 1, hasSideEffects = 0 in
3147 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3148 (ins VR128:$src1, ssmem:$src2),
3149 !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
3150 [], itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
3154 /// sse1_fp_unop_p - SSE1 unops in packed form.
3155 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
3157 let Predicates = [HasAVX] in {
3158 def V#NAME#PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3159 !strconcat("v", OpcodeStr,
3160 "ps\t{$src, $dst|$dst, $src}"),
3161 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))],
3162 itins.rr>, VEX, Sched<[itins.Sched]>;
3163 def V#NAME#PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3164 !strconcat("v", OpcodeStr,
3165 "ps\t{$src, $dst|$dst, $src}"),
3166 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))],
3167 itins.rm>, VEX, Sched<[itins.Sched.Folded]>;
3168 def V#NAME#PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3169 !strconcat("v", OpcodeStr,
3170 "ps\t{$src, $dst|$dst, $src}"),
3171 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))],
3172 itins.rr>, VEX, VEX_L, Sched<[itins.Sched]>;
3173 def V#NAME#PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3174 !strconcat("v", OpcodeStr,
3175 "ps\t{$src, $dst|$dst, $src}"),
3176 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))],
3177 itins.rm>, VEX, VEX_L, Sched<[itins.Sched.Folded]>;
3180 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3181 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3182 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))], itins.rr>,
3183 Sched<[itins.Sched]>;
3184 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3185 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3186 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))], itins.rm>,
3187 Sched<[itins.Sched.Folded]>;
3190 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
3191 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3192 Intrinsic V4F32Int, Intrinsic V8F32Int,
3194 let Predicates = [HasAVX] in {
3195 def V#NAME#PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3196 !strconcat("v", OpcodeStr,
3197 "ps\t{$src, $dst|$dst, $src}"),
3198 [(set VR128:$dst, (V4F32Int VR128:$src))],
3199 itins.rr>, VEX, Sched<[itins.Sched]>;
3200 def V#NAME#PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3201 !strconcat("v", OpcodeStr,
3202 "ps\t{$src, $dst|$dst, $src}"),
3203 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))],
3204 itins.rm>, VEX, Sched<[itins.Sched.Folded]>;
3205 def V#NAME#PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3206 !strconcat("v", OpcodeStr,
3207 "ps\t{$src, $dst|$dst, $src}"),
3208 [(set VR256:$dst, (V8F32Int VR256:$src))],
3209 itins.rr>, VEX, VEX_L, Sched<[itins.Sched]>;
3210 def V#NAME#PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst),
3212 !strconcat("v", OpcodeStr,
3213 "ps\t{$src, $dst|$dst, $src}"),
3214 [(set VR256:$dst, (V8F32Int (memopv8f32 addr:$src)))],
3215 itins.rm>, VEX, VEX_L, Sched<[itins.Sched.Folded]>;
3218 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3219 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3220 [(set VR128:$dst, (V4F32Int VR128:$src))],
3221 itins.rr>, Sched<[itins.Sched]>;
3222 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3223 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3224 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))],
3225 itins.rm>, Sched<[itins.Sched.Folded]>;
3228 /// sse2_fp_unop_s - SSE2 unops in scalar form.
3229 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
3230 SDNode OpNode, Intrinsic F64Int, OpndItins itins> {
3231 let Predicates = [HasAVX], hasSideEffects = 0 in {
3232 def V#NAME#SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst),
3233 (ins FR64:$src1, FR64:$src2),
3234 !strconcat("v", OpcodeStr,
3235 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3236 []>, VEX_4V, VEX_LIG, Sched<[itins.Sched]>;
3237 let mayLoad = 1 in {
3238 def V#NAME#SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst),
3239 (ins FR64:$src1,f64mem:$src2),
3240 !strconcat("v", OpcodeStr,
3241 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3242 []>, VEX_4V, VEX_LIG,
3243 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3244 def V#NAME#SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
3245 (ins VR128:$src1, sdmem:$src2),
3246 !strconcat("v", OpcodeStr,
3247 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3248 []>, VEX_4V, VEX_LIG,
3249 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3253 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
3254 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3255 [(set FR64:$dst, (OpNode FR64:$src))], itins.rr>,
3256 Sched<[itins.Sched]>;
3257 // See the comments in sse1_fp_unop_s for why this is OptForSize.
3258 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
3259 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3260 [(set FR64:$dst, (OpNode (load addr:$src)))], itins.rm>, XD,
3261 Requires<[UseSSE2, OptForSize]>, Sched<[itins.Sched.Folded]>;
3262 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3263 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3264 [(set VR128:$dst, (F64Int VR128:$src))], itins.rr>,
3265 Sched<[itins.Sched]>;
3266 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
3267 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3268 [(set VR128:$dst, (F64Int sse_load_f64:$src))], itins.rm>,
3269 Sched<[itins.Sched.Folded]>;
3272 /// sse2_fp_unop_p - SSE2 unops in vector forms.
3273 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
3274 SDNode OpNode, OpndItins itins> {
3275 let Predicates = [HasAVX] in {
3276 def V#NAME#PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3277 !strconcat("v", OpcodeStr,
3278 "pd\t{$src, $dst|$dst, $src}"),
3279 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))],
3280 itins.rr>, VEX, Sched<[itins.Sched]>;
3281 def V#NAME#PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3282 !strconcat("v", OpcodeStr,
3283 "pd\t{$src, $dst|$dst, $src}"),
3284 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))],
3285 itins.rm>, VEX, Sched<[itins.Sched.Folded]>;
3286 def V#NAME#PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3287 !strconcat("v", OpcodeStr,
3288 "pd\t{$src, $dst|$dst, $src}"),
3289 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))],
3290 itins.rr>, VEX, VEX_L, Sched<[itins.Sched]>;
3291 def V#NAME#PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3292 !strconcat("v", OpcodeStr,
3293 "pd\t{$src, $dst|$dst, $src}"),
3294 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))],
3295 itins.rm>, VEX, VEX_L, Sched<[itins.Sched.Folded]>;
3298 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3299 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3300 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))], itins.rr>,
3301 Sched<[itins.Sched]>;
3302 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3303 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3304 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))], itins.rm>,
3305 Sched<[itins.Sched.Folded]>;
3309 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss,
3311 sse1_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTPS>,
3312 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd,
3314 sse2_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTPD>;
3316 // Reciprocal approximations. Note that these typically require refinement
3317 // in order to obtain suitable precision.
3318 defm RSQRT : sse1_fp_unop_rw<0x52, "rsqrt", X86frsqrt, SSE_SQRTSS>,
3319 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, SSE_SQRTPS>,
3320 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps,
3321 int_x86_avx_rsqrt_ps_256, SSE_SQRTPS>;
3322 defm RCP : sse1_fp_unop_rw<0x53, "rcp", X86frcp, SSE_RCPS>,
3323 sse1_fp_unop_p<0x53, "rcp", X86frcp, SSE_RCPP>,
3324 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps,
3325 int_x86_avx_rcp_ps_256, SSE_RCPP>;
3327 let Predicates = [UseAVX] in {
3328 def : Pat<(f32 (fsqrt FR32:$src)),
3329 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3330 def : Pat<(f32 (fsqrt (load addr:$src))),
3331 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3332 Requires<[HasAVX, OptForSize]>;
3333 def : Pat<(f64 (fsqrt FR64:$src)),
3334 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
3335 def : Pat<(f64 (fsqrt (load addr:$src))),
3336 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
3337 Requires<[HasAVX, OptForSize]>;
3339 def : Pat<(f32 (X86frsqrt FR32:$src)),
3340 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3341 def : Pat<(f32 (X86frsqrt (load addr:$src))),
3342 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3343 Requires<[HasAVX, OptForSize]>;
3345 def : Pat<(f32 (X86frcp FR32:$src)),
3346 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3347 def : Pat<(f32 (X86frcp (load addr:$src))),
3348 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3349 Requires<[HasAVX, OptForSize]>;
3351 let Predicates = [UseAVX] in {
3352 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
3353 (COPY_TO_REGCLASS (VSQRTSSr (f32 (IMPLICIT_DEF)),
3354 (COPY_TO_REGCLASS VR128:$src, FR32)),
3356 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
3357 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3359 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
3360 (COPY_TO_REGCLASS (VSQRTSDr (f64 (IMPLICIT_DEF)),
3361 (COPY_TO_REGCLASS VR128:$src, FR64)),
3363 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
3364 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
3367 let Predicates = [HasAVX] in {
3368 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3369 (COPY_TO_REGCLASS (VRSQRTSSr (f32 (IMPLICIT_DEF)),
3370 (COPY_TO_REGCLASS VR128:$src, FR32)),
3372 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
3373 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3375 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3376 (COPY_TO_REGCLASS (VRCPSSr (f32 (IMPLICIT_DEF)),
3377 (COPY_TO_REGCLASS VR128:$src, FR32)),
3379 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
3380 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3383 // Reciprocal approximations. Note that these typically require refinement
3384 // in order to obtain suitable precision.
3385 let Predicates = [UseSSE1] in {
3386 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3387 (RSQRTSSr_Int VR128:$src, VR128:$src)>;
3388 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3389 (RCPSSr_Int VR128:$src, VR128:$src)>;
3392 // There is no f64 version of the reciprocal approximation instructions.
3394 //===----------------------------------------------------------------------===//
3395 // SSE 1 & 2 - Non-temporal stores
3396 //===----------------------------------------------------------------------===//
3398 let AddedComplexity = 400 in { // Prefer non-temporal versions
3399 let SchedRW = [WriteStore] in {
3400 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3401 (ins f128mem:$dst, VR128:$src),
3402 "movntps\t{$src, $dst|$dst, $src}",
3403 [(alignednontemporalstore (v4f32 VR128:$src),
3405 IIC_SSE_MOVNT>, VEX;
3406 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3407 (ins f128mem:$dst, VR128:$src),
3408 "movntpd\t{$src, $dst|$dst, $src}",
3409 [(alignednontemporalstore (v2f64 VR128:$src),
3411 IIC_SSE_MOVNT>, VEX;
3413 let ExeDomain = SSEPackedInt in
3414 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3415 (ins f128mem:$dst, VR128:$src),
3416 "movntdq\t{$src, $dst|$dst, $src}",
3417 [(alignednontemporalstore (v2i64 VR128:$src),
3419 IIC_SSE_MOVNT>, VEX;
3421 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3422 (ins f256mem:$dst, VR256:$src),
3423 "movntps\t{$src, $dst|$dst, $src}",
3424 [(alignednontemporalstore (v8f32 VR256:$src),
3426 IIC_SSE_MOVNT>, VEX, VEX_L;
3427 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3428 (ins f256mem:$dst, VR256:$src),
3429 "movntpd\t{$src, $dst|$dst, $src}",
3430 [(alignednontemporalstore (v4f64 VR256:$src),
3432 IIC_SSE_MOVNT>, VEX, VEX_L;
3433 let ExeDomain = SSEPackedInt in
3434 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3435 (ins f256mem:$dst, VR256:$src),
3436 "movntdq\t{$src, $dst|$dst, $src}",
3437 [(alignednontemporalstore (v4i64 VR256:$src),
3439 IIC_SSE_MOVNT>, VEX, VEX_L;
3441 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3442 "movntps\t{$src, $dst|$dst, $src}",
3443 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)],
3445 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3446 "movntpd\t{$src, $dst|$dst, $src}",
3447 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)],
3450 let ExeDomain = SSEPackedInt in
3451 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3452 "movntdq\t{$src, $dst|$dst, $src}",
3453 [(alignednontemporalstore (v2i64 VR128:$src), addr:$dst)],
3456 // There is no AVX form for instructions below this point
3457 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3458 "movnti{l}\t{$src, $dst|$dst, $src}",
3459 [(nontemporalstore (i32 GR32:$src), addr:$dst)],
3461 TB, Requires<[HasSSE2]>;
3462 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3463 "movnti{q}\t{$src, $dst|$dst, $src}",
3464 [(nontemporalstore (i64 GR64:$src), addr:$dst)],
3466 TB, Requires<[HasSSE2]>;
3467 } // SchedRW = [WriteStore]
3469 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3470 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3472 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3473 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[UseSSE2]>;
3474 } // AddedComplexity
3476 //===----------------------------------------------------------------------===//
3477 // SSE 1 & 2 - Prefetch and memory fence
3478 //===----------------------------------------------------------------------===//
3480 // Prefetch intrinsic.
3481 let Predicates = [HasSSE1], SchedRW = [WriteLoad] in {
3482 def PREFETCHT0 : I<0x18, MRM1m, (outs), (ins i8mem:$src),
3483 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))],
3484 IIC_SSE_PREFETCH>, TB;
3485 def PREFETCHT1 : I<0x18, MRM2m, (outs), (ins i8mem:$src),
3486 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))],
3487 IIC_SSE_PREFETCH>, TB;
3488 def PREFETCHT2 : I<0x18, MRM3m, (outs), (ins i8mem:$src),
3489 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))],
3490 IIC_SSE_PREFETCH>, TB;
3491 def PREFETCHNTA : I<0x18, MRM0m, (outs), (ins i8mem:$src),
3492 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))],
3493 IIC_SSE_PREFETCH>, TB;
3496 // FIXME: How should these memory instructions be modeled?
3497 let SchedRW = [WriteLoad] in {
3499 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3500 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)],
3501 IIC_SSE_PREFETCH>, TB, Requires<[HasSSE2]>;
3503 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3504 // was introduced with SSE2, it's backward compatible.
3505 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", [], IIC_SSE_PAUSE>, REP;
3507 // Load, store, and memory fence
3508 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3509 "sfence", [(int_x86_sse_sfence)], IIC_SSE_SFENCE>,
3510 TB, Requires<[HasSSE1]>;
3511 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3512 "lfence", [(int_x86_sse2_lfence)], IIC_SSE_LFENCE>,
3513 TB, Requires<[HasSSE2]>;
3514 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3515 "mfence", [(int_x86_sse2_mfence)], IIC_SSE_MFENCE>,
3516 TB, Requires<[HasSSE2]>;
3519 def : Pat<(X86SFence), (SFENCE)>;
3520 def : Pat<(X86LFence), (LFENCE)>;
3521 def : Pat<(X86MFence), (MFENCE)>;
3523 //===----------------------------------------------------------------------===//
3524 // SSE 1 & 2 - Load/Store XCSR register
3525 //===----------------------------------------------------------------------===//
3527 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3528 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3529 IIC_SSE_LDMXCSR>, VEX, Sched<[WriteLoad]>;
3530 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3531 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3532 IIC_SSE_STMXCSR>, VEX, Sched<[WriteStore]>;
3534 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3535 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3536 IIC_SSE_LDMXCSR>, Sched<[WriteLoad]>;
3537 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3538 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3539 IIC_SSE_STMXCSR>, Sched<[WriteStore]>;
3541 //===---------------------------------------------------------------------===//
3542 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3543 //===---------------------------------------------------------------------===//
3545 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3547 let neverHasSideEffects = 1, SchedRW = [WriteMove] in {
3548 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3549 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3551 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3552 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3554 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3555 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3557 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3558 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3563 let isCodeGenOnly = 1, hasSideEffects = 0, SchedRW = [WriteMove] in {
3564 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3565 "movdqa\t{$src, $dst|$dst, $src}", [],
3568 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3569 "movdqa\t{$src, $dst|$dst, $src}", [],
3570 IIC_SSE_MOVA_P_RR>, VEX, VEX_L;
3571 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3572 "movdqu\t{$src, $dst|$dst, $src}", [],
3575 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3576 "movdqu\t{$src, $dst|$dst, $src}", [],
3577 IIC_SSE_MOVU_P_RR>, VEX, VEX_L;
3580 let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1,
3581 neverHasSideEffects = 1, SchedRW = [WriteLoad] in {
3582 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3583 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3585 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3586 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3588 let Predicates = [HasAVX] in {
3589 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3590 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3592 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3593 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3598 let mayStore = 1, neverHasSideEffects = 1, SchedRW = [WriteStore] in {
3599 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3600 (ins i128mem:$dst, VR128:$src),
3601 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3603 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3604 (ins i256mem:$dst, VR256:$src),
3605 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3607 let Predicates = [HasAVX] in {
3608 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3609 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3611 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3612 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3617 let SchedRW = [WriteMove] in {
3618 let neverHasSideEffects = 1 in
3619 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3620 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>;
3622 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3623 "movdqu\t{$src, $dst|$dst, $src}",
3624 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>;
3627 let isCodeGenOnly = 1, hasSideEffects = 0 in {
3628 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3629 "movdqa\t{$src, $dst|$dst, $src}", [],
3632 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3633 "movdqu\t{$src, $dst|$dst, $src}",
3634 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>;
3638 let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1,
3639 neverHasSideEffects = 1, SchedRW = [WriteLoad] in {
3640 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3641 "movdqa\t{$src, $dst|$dst, $src}",
3642 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/],
3644 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3645 "movdqu\t{$src, $dst|$dst, $src}",
3646 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/],
3648 XS, Requires<[UseSSE2]>;
3651 let mayStore = 1, neverHasSideEffects = 1, SchedRW = [WriteStore] in {
3652 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3653 "movdqa\t{$src, $dst|$dst, $src}",
3654 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/],
3656 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3657 "movdqu\t{$src, $dst|$dst, $src}",
3658 [/*(store (v2i64 VR128:$src), addr:$dst)*/],
3660 XS, Requires<[UseSSE2]>;
3663 } // ExeDomain = SSEPackedInt
3665 let Predicates = [HasAVX] in {
3666 def : Pat<(int_x86_sse2_storeu_dq addr:$dst, VR128:$src),
3667 (VMOVDQUmr addr:$dst, VR128:$src)>;
3668 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3669 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3671 let Predicates = [UseSSE2] in
3672 def : Pat<(int_x86_sse2_storeu_dq addr:$dst, VR128:$src),
3673 (MOVDQUmr addr:$dst, VR128:$src)>;
3675 //===---------------------------------------------------------------------===//
3676 // SSE2 - Packed Integer Arithmetic Instructions
3677 //===---------------------------------------------------------------------===//
3679 let Sched = WriteVecIMul in
3680 def SSE_PMADD : OpndItins<
3681 IIC_SSE_PMADD, IIC_SSE_PMADD
3684 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3686 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3687 RegisterClass RC, PatFrag memop_frag,
3688 X86MemOperand x86memop,
3690 bit IsCommutable = 0,
3692 let isCommutable = IsCommutable in
3693 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3694 (ins RC:$src1, RC:$src2),
3696 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3697 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3698 [(set RC:$dst, (IntId RC:$src1, RC:$src2))], itins.rr>,
3699 Sched<[itins.Sched]>;
3700 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3701 (ins RC:$src1, x86memop:$src2),
3703 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3704 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3705 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))],
3706 itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
3709 multiclass PDI_binop_all_int<bits<8> opc, string OpcodeStr, Intrinsic IntId128,
3710 Intrinsic IntId256, OpndItins itins,
3711 bit IsCommutable = 0> {
3712 let Predicates = [HasAVX] in
3713 defm V#NAME : PDI_binop_rm_int<opc, !strconcat("v", OpcodeStr), IntId128,
3714 VR128, memopv2i64, i128mem, itins,
3715 IsCommutable, 0>, VEX_4V;
3717 let Constraints = "$src1 = $dst" in
3718 defm NAME : PDI_binop_rm_int<opc, OpcodeStr, IntId128, VR128, memopv2i64,
3719 i128mem, itins, IsCommutable, 1>;
3721 let Predicates = [HasAVX2] in
3722 defm V#NAME#Y : PDI_binop_rm_int<opc, !strconcat("v", OpcodeStr), IntId256,
3723 VR256, memopv4i64, i256mem, itins,
3724 IsCommutable, 0>, VEX_4V, VEX_L;
3727 multiclass PDI_binop_rmi<bits<8> opc, bits<8> opc2, Format ImmForm,
3728 string OpcodeStr, SDNode OpNode,
3729 SDNode OpNode2, RegisterClass RC,
3730 ValueType DstVT, ValueType SrcVT, PatFrag bc_frag,
3731 ShiftOpndItins itins,
3733 // src2 is always 128-bit
3734 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3735 (ins RC:$src1, VR128:$src2),
3737 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3738 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3739 [(set RC:$dst, (DstVT (OpNode RC:$src1, (SrcVT VR128:$src2))))],
3740 itins.rr>, Sched<[WriteVecShift]>;
3741 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3742 (ins RC:$src1, i128mem:$src2),
3744 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3745 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3746 [(set RC:$dst, (DstVT (OpNode RC:$src1,
3747 (bc_frag (memopv2i64 addr:$src2)))))], itins.rm>,
3748 Sched<[WriteVecShiftLd, ReadAfterLd]>;
3749 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3750 (ins RC:$src1, i32i8imm:$src2),
3752 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3753 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3754 [(set RC:$dst, (DstVT (OpNode2 RC:$src1, (i32 imm:$src2))))], itins.ri>,
3755 Sched<[WriteVecShift]>;
3758 /// PDI_binop_rm2 - Simple SSE2 binary operator with different src and dst types
3759 multiclass PDI_binop_rm2<bits<8> opc, string OpcodeStr, SDNode OpNode,
3760 ValueType DstVT, ValueType SrcVT, RegisterClass RC,
3761 PatFrag memop_frag, X86MemOperand x86memop,
3763 bit IsCommutable = 0, bit Is2Addr = 1> {
3764 let isCommutable = IsCommutable in
3765 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3766 (ins RC:$src1, RC:$src2),
3768 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3769 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3770 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1), RC:$src2)))]>,
3771 Sched<[itins.Sched]>;
3772 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3773 (ins RC:$src1, x86memop:$src2),
3775 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3776 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3777 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1),
3778 (bitconvert (memop_frag addr:$src2)))))]>,
3779 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3781 } // ExeDomain = SSEPackedInt
3783 defm PADDB : PDI_binop_all<0xFC, "paddb", add, v16i8, v32i8,
3784 SSE_INTALU_ITINS_P, 1>;
3785 defm PADDW : PDI_binop_all<0xFD, "paddw", add, v8i16, v16i16,
3786 SSE_INTALU_ITINS_P, 1>;
3787 defm PADDD : PDI_binop_all<0xFE, "paddd", add, v4i32, v8i32,
3788 SSE_INTALU_ITINS_P, 1>;
3789 defm PADDQ : PDI_binop_all<0xD4, "paddq", add, v2i64, v4i64,
3790 SSE_INTALUQ_ITINS_P, 1>;
3791 defm PMULLW : PDI_binop_all<0xD5, "pmullw", mul, v8i16, v16i16,
3792 SSE_INTMUL_ITINS_P, 1>;
3793 defm PSUBB : PDI_binop_all<0xF8, "psubb", sub, v16i8, v32i8,
3794 SSE_INTALU_ITINS_P, 0>;
3795 defm PSUBW : PDI_binop_all<0xF9, "psubw", sub, v8i16, v16i16,
3796 SSE_INTALU_ITINS_P, 0>;
3797 defm PSUBD : PDI_binop_all<0xFA, "psubd", sub, v4i32, v8i32,
3798 SSE_INTALU_ITINS_P, 0>;
3799 defm PSUBQ : PDI_binop_all<0xFB, "psubq", sub, v2i64, v4i64,
3800 SSE_INTALUQ_ITINS_P, 0>;
3801 defm PSUBUSB : PDI_binop_all<0xD8, "psubusb", X86subus, v16i8, v32i8,
3802 SSE_INTALU_ITINS_P, 0>;
3803 defm PSUBUSW : PDI_binop_all<0xD9, "psubusw", X86subus, v8i16, v16i16,
3804 SSE_INTALU_ITINS_P, 0>;
3805 defm PMINUB : PDI_binop_all<0xDA, "pminub", X86umin, v16i8, v32i8,
3806 SSE_INTALU_ITINS_P, 1>;
3807 defm PMINSW : PDI_binop_all<0xEA, "pminsw", X86smin, v8i16, v16i16,
3808 SSE_INTALU_ITINS_P, 1>;
3809 defm PMAXUB : PDI_binop_all<0xDE, "pmaxub", X86umax, v16i8, v32i8,
3810 SSE_INTALU_ITINS_P, 1>;
3811 defm PMAXSW : PDI_binop_all<0xEE, "pmaxsw", X86smax, v8i16, v16i16,
3812 SSE_INTALU_ITINS_P, 1>;
3815 defm PSUBSB : PDI_binop_all_int<0xE8, "psubsb", int_x86_sse2_psubs_b,
3816 int_x86_avx2_psubs_b, SSE_INTALU_ITINS_P, 0>;
3817 defm PSUBSW : PDI_binop_all_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
3818 int_x86_avx2_psubs_w, SSE_INTALU_ITINS_P, 0>;
3819 defm PADDSB : PDI_binop_all_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
3820 int_x86_avx2_padds_b, SSE_INTALU_ITINS_P, 1>;
3821 defm PADDSW : PDI_binop_all_int<0xED, "paddsw" , int_x86_sse2_padds_w,
3822 int_x86_avx2_padds_w, SSE_INTALU_ITINS_P, 1>;
3823 defm PADDUSB : PDI_binop_all_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
3824 int_x86_avx2_paddus_b, SSE_INTALU_ITINS_P, 1>;
3825 defm PADDUSW : PDI_binop_all_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
3826 int_x86_avx2_paddus_w, SSE_INTALU_ITINS_P, 1>;
3827 defm PMULHUW : PDI_binop_all_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
3828 int_x86_avx2_pmulhu_w, SSE_INTMUL_ITINS_P, 1>;
3829 defm PMULHW : PDI_binop_all_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
3830 int_x86_avx2_pmulh_w, SSE_INTMUL_ITINS_P, 1>;
3831 defm PMADDWD : PDI_binop_all_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
3832 int_x86_avx2_pmadd_wd, SSE_PMADD, 1>;
3833 defm PAVGB : PDI_binop_all_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
3834 int_x86_avx2_pavg_b, SSE_INTALU_ITINS_P, 1>;
3835 defm PAVGW : PDI_binop_all_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
3836 int_x86_avx2_pavg_w, SSE_INTALU_ITINS_P, 1>;
3837 defm PSADBW : PDI_binop_all_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
3838 int_x86_avx2_psad_bw, SSE_PMADD, 1>;
3840 let Predicates = [HasAVX] in
3841 defm VPMULUDQ : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v2i64, v4i32, VR128,
3842 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1, 0>,
3844 let Predicates = [HasAVX2] in
3845 defm VPMULUDQY : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v4i64, v8i32,
3846 VR256, memopv4i64, i256mem,
3847 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V, VEX_L;
3848 let Constraints = "$src1 = $dst" in
3849 defm PMULUDQ : PDI_binop_rm2<0xF4, "pmuludq", X86pmuludq, v2i64, v4i32, VR128,
3850 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1>;
3852 //===---------------------------------------------------------------------===//
3853 // SSE2 - Packed Integer Logical Instructions
3854 //===---------------------------------------------------------------------===//
3856 let Predicates = [HasAVX] in {
3857 defm VPSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3858 VR128, v8i16, v8i16, bc_v8i16,
3859 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3860 defm VPSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3861 VR128, v4i32, v4i32, bc_v4i32,
3862 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3863 defm VPSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3864 VR128, v2i64, v2i64, bc_v2i64,
3865 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3867 defm VPSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3868 VR128, v8i16, v8i16, bc_v8i16,
3869 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3870 defm VPSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3871 VR128, v4i32, v4i32, bc_v4i32,
3872 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3873 defm VPSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3874 VR128, v2i64, v2i64, bc_v2i64,
3875 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3877 defm VPSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3878 VR128, v8i16, v8i16, bc_v8i16,
3879 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3880 defm VPSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3881 VR128, v4i32, v4i32, bc_v4i32,
3882 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3884 let ExeDomain = SSEPackedInt, SchedRW = [WriteVecShift] in {
3885 // 128-bit logical shifts.
3886 def VPSLLDQri : PDIi8<0x73, MRM7r,
3887 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3888 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3890 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>,
3892 def VPSRLDQri : PDIi8<0x73, MRM3r,
3893 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3894 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3896 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>,
3898 // PSRADQri doesn't exist in SSE[1-3].
3900 } // Predicates = [HasAVX]
3902 let Predicates = [HasAVX2] in {
3903 defm VPSLLWY : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3904 VR256, v16i16, v8i16, bc_v8i16,
3905 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3906 defm VPSLLDY : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3907 VR256, v8i32, v4i32, bc_v4i32,
3908 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3909 defm VPSLLQY : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3910 VR256, v4i64, v2i64, bc_v2i64,
3911 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3913 defm VPSRLWY : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3914 VR256, v16i16, v8i16, bc_v8i16,
3915 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3916 defm VPSRLDY : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3917 VR256, v8i32, v4i32, bc_v4i32,
3918 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3919 defm VPSRLQY : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3920 VR256, v4i64, v2i64, bc_v2i64,
3921 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3923 defm VPSRAWY : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3924 VR256, v16i16, v8i16, bc_v8i16,
3925 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3926 defm VPSRADY : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3927 VR256, v8i32, v4i32, bc_v4i32,
3928 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3930 let ExeDomain = SSEPackedInt, SchedRW = [WriteVecShift] in {
3931 // 256-bit logical shifts.
3932 def VPSLLDQYri : PDIi8<0x73, MRM7r,
3933 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3934 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3936 (int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2))]>,
3938 def VPSRLDQYri : PDIi8<0x73, MRM3r,
3939 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3940 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3942 (int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2))]>,
3944 // PSRADQYri doesn't exist in SSE[1-3].
3946 } // Predicates = [HasAVX2]
3948 let Constraints = "$src1 = $dst" in {
3949 defm PSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "psllw", X86vshl, X86vshli,
3950 VR128, v8i16, v8i16, bc_v8i16,
3951 SSE_INTSHIFT_ITINS_P>;
3952 defm PSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "pslld", X86vshl, X86vshli,
3953 VR128, v4i32, v4i32, bc_v4i32,
3954 SSE_INTSHIFT_ITINS_P>;
3955 defm PSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "psllq", X86vshl, X86vshli,
3956 VR128, v2i64, v2i64, bc_v2i64,
3957 SSE_INTSHIFT_ITINS_P>;
3959 defm PSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "psrlw", X86vsrl, X86vsrli,
3960 VR128, v8i16, v8i16, bc_v8i16,
3961 SSE_INTSHIFT_ITINS_P>;
3962 defm PSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "psrld", X86vsrl, X86vsrli,
3963 VR128, v4i32, v4i32, bc_v4i32,
3964 SSE_INTSHIFT_ITINS_P>;
3965 defm PSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "psrlq", X86vsrl, X86vsrli,
3966 VR128, v2i64, v2i64, bc_v2i64,
3967 SSE_INTSHIFT_ITINS_P>;
3969 defm PSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "psraw", X86vsra, X86vsrai,
3970 VR128, v8i16, v8i16, bc_v8i16,
3971 SSE_INTSHIFT_ITINS_P>;
3972 defm PSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "psrad", X86vsra, X86vsrai,
3973 VR128, v4i32, v4i32, bc_v4i32,
3974 SSE_INTSHIFT_ITINS_P>;
3976 let ExeDomain = SSEPackedInt, SchedRW = [WriteVecShift] in {
3977 // 128-bit logical shifts.
3978 def PSLLDQri : PDIi8<0x73, MRM7r,
3979 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3980 "pslldq\t{$src2, $dst|$dst, $src2}",
3982 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))],
3983 IIC_SSE_INTSHDQ_P_RI>;
3984 def PSRLDQri : PDIi8<0x73, MRM3r,
3985 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3986 "psrldq\t{$src2, $dst|$dst, $src2}",
3988 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))],
3989 IIC_SSE_INTSHDQ_P_RI>;
3990 // PSRADQri doesn't exist in SSE[1-3].
3992 } // Constraints = "$src1 = $dst"
3994 let Predicates = [HasAVX] in {
3995 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3996 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3997 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3998 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3999 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
4000 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4002 // Shift up / down and insert zero's.
4003 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
4004 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4005 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
4006 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4009 let Predicates = [HasAVX2] in {
4010 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
4011 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
4012 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
4013 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
4016 let Predicates = [UseSSE2] in {
4017 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
4018 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4019 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
4020 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4021 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
4022 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4024 // Shift up / down and insert zero's.
4025 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
4026 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4027 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
4028 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4031 //===---------------------------------------------------------------------===//
4032 // SSE2 - Packed Integer Comparison Instructions
4033 //===---------------------------------------------------------------------===//
4035 defm PCMPEQB : PDI_binop_all<0x74, "pcmpeqb", X86pcmpeq, v16i8, v32i8,
4036 SSE_INTALU_ITINS_P, 1>;
4037 defm PCMPEQW : PDI_binop_all<0x75, "pcmpeqw", X86pcmpeq, v8i16, v16i16,
4038 SSE_INTALU_ITINS_P, 1>;
4039 defm PCMPEQD : PDI_binop_all<0x76, "pcmpeqd", X86pcmpeq, v4i32, v8i32,
4040 SSE_INTALU_ITINS_P, 1>;
4041 defm PCMPGTB : PDI_binop_all<0x64, "pcmpgtb", X86pcmpgt, v16i8, v32i8,
4042 SSE_INTALU_ITINS_P, 0>;
4043 defm PCMPGTW : PDI_binop_all<0x65, "pcmpgtw", X86pcmpgt, v8i16, v16i16,
4044 SSE_INTALU_ITINS_P, 0>;
4045 defm PCMPGTD : PDI_binop_all<0x66, "pcmpgtd", X86pcmpgt, v4i32, v8i32,
4046 SSE_INTALU_ITINS_P, 0>;
4048 //===---------------------------------------------------------------------===//
4049 // SSE2 - Packed Integer Pack Instructions
4050 //===---------------------------------------------------------------------===//
4052 defm PACKSSWB : PDI_binop_all_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
4053 int_x86_avx2_packsswb, SSE_INTALU_ITINS_P, 0>;
4054 defm PACKSSDW : PDI_binop_all_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
4055 int_x86_avx2_packssdw, SSE_INTALU_ITINS_P, 0>;
4056 defm PACKUSWB : PDI_binop_all_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
4057 int_x86_avx2_packuswb, SSE_INTALU_ITINS_P, 0>;
4059 //===---------------------------------------------------------------------===//
4060 // SSE2 - Packed Integer Shuffle Instructions
4061 //===---------------------------------------------------------------------===//
4063 let ExeDomain = SSEPackedInt in {
4064 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt128, ValueType vt256,
4066 let Predicates = [HasAVX] in {
4067 def V#NAME#ri : Ii8<0x70, MRMSrcReg, (outs VR128:$dst),
4068 (ins VR128:$src1, i8imm:$src2),
4069 !strconcat("v", OpcodeStr,
4070 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4072 (vt128 (OpNode VR128:$src1, (i8 imm:$src2))))],
4073 IIC_SSE_PSHUF_RI>, VEX, Sched<[WriteShuffle]>;
4074 def V#NAME#mi : Ii8<0x70, MRMSrcMem, (outs VR128:$dst),
4075 (ins i128mem:$src1, i8imm:$src2),
4076 !strconcat("v", OpcodeStr,
4077 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4079 (vt128 (OpNode (bitconvert (memopv2i64 addr:$src1)),
4080 (i8 imm:$src2))))], IIC_SSE_PSHUF_MI>, VEX,
4081 Sched<[WriteShuffleLd]>;
4084 let Predicates = [HasAVX2] in {
4085 def V#NAME#Yri : Ii8<0x70, MRMSrcReg, (outs VR256:$dst),
4086 (ins VR256:$src1, i8imm:$src2),
4087 !strconcat("v", OpcodeStr,
4088 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4090 (vt256 (OpNode VR256:$src1, (i8 imm:$src2))))],
4091 IIC_SSE_PSHUF_RI>, VEX, VEX_L, Sched<[WriteShuffle]>;
4092 def V#NAME#Ymi : Ii8<0x70, MRMSrcMem, (outs VR256:$dst),
4093 (ins i256mem:$src1, i8imm:$src2),
4094 !strconcat("v", OpcodeStr,
4095 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4097 (vt256 (OpNode (bitconvert (memopv4i64 addr:$src1)),
4098 (i8 imm:$src2))))], IIC_SSE_PSHUF_MI>, VEX, VEX_L,
4099 Sched<[WriteShuffleLd]>;
4102 let Predicates = [UseSSE2] in {
4103 def ri : Ii8<0x70, MRMSrcReg,
4104 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
4105 !strconcat(OpcodeStr,
4106 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4108 (vt128 (OpNode VR128:$src1, (i8 imm:$src2))))],
4109 IIC_SSE_PSHUF_RI>, Sched<[WriteShuffle]>;
4110 def mi : Ii8<0x70, MRMSrcMem,
4111 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
4112 !strconcat(OpcodeStr,
4113 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4115 (vt128 (OpNode (bitconvert (memopv2i64 addr:$src1)),
4116 (i8 imm:$src2))))], IIC_SSE_PSHUF_MI>,
4117 Sched<[WriteShuffleLd]>;
4120 } // ExeDomain = SSEPackedInt
4122 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, v8i32, X86PShufd>, TB, OpSize;
4123 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, v16i16, X86PShufhw>, XS;
4124 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, v16i16, X86PShuflw>, XD;
4126 let Predicates = [HasAVX] in {
4127 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
4128 (VPSHUFDmi addr:$src1, imm:$imm)>;
4129 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4130 (VPSHUFDri VR128:$src1, imm:$imm)>;
4133 let Predicates = [UseSSE2] in {
4134 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
4135 (PSHUFDmi addr:$src1, imm:$imm)>;
4136 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4137 (PSHUFDri VR128:$src1, imm:$imm)>;
4140 //===---------------------------------------------------------------------===//
4141 // SSE2 - Packed Integer Unpack Instructions
4142 //===---------------------------------------------------------------------===//
4144 let ExeDomain = SSEPackedInt in {
4145 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
4146 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
4147 def rr : PDI<opc, MRMSrcReg,
4148 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4150 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4151 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4152 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))],
4153 IIC_SSE_UNPCK>, Sched<[WriteShuffle]>;
4154 def rm : PDI<opc, MRMSrcMem,
4155 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4157 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4158 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4159 [(set VR128:$dst, (OpNode VR128:$src1,
4160 (bc_frag (memopv2i64
4163 Sched<[WriteShuffleLd, ReadAfterLd]>;
4166 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
4167 SDNode OpNode, PatFrag bc_frag> {
4168 def Yrr : PDI<opc, MRMSrcReg,
4169 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4170 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4171 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>,
4172 Sched<[WriteShuffle]>;
4173 def Yrm : PDI<opc, MRMSrcMem,
4174 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4175 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4176 [(set VR256:$dst, (OpNode VR256:$src1,
4177 (bc_frag (memopv4i64 addr:$src2))))]>,
4178 Sched<[WriteShuffleLd, ReadAfterLd]>;
4181 let Predicates = [HasAVX] in {
4182 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
4183 bc_v16i8, 0>, VEX_4V;
4184 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
4185 bc_v8i16, 0>, VEX_4V;
4186 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
4187 bc_v4i32, 0>, VEX_4V;
4188 defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
4189 bc_v2i64, 0>, VEX_4V;
4191 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
4192 bc_v16i8, 0>, VEX_4V;
4193 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
4194 bc_v8i16, 0>, VEX_4V;
4195 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
4196 bc_v4i32, 0>, VEX_4V;
4197 defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
4198 bc_v2i64, 0>, VEX_4V;
4201 let Predicates = [HasAVX2] in {
4202 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
4203 bc_v32i8>, VEX_4V, VEX_L;
4204 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
4205 bc_v16i16>, VEX_4V, VEX_L;
4206 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
4207 bc_v8i32>, VEX_4V, VEX_L;
4208 defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
4209 bc_v4i64>, VEX_4V, VEX_L;
4211 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
4212 bc_v32i8>, VEX_4V, VEX_L;
4213 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
4214 bc_v16i16>, VEX_4V, VEX_L;
4215 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
4216 bc_v8i32>, VEX_4V, VEX_L;
4217 defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
4218 bc_v4i64>, VEX_4V, VEX_L;
4221 let Constraints = "$src1 = $dst" in {
4222 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
4224 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
4226 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
4228 defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
4231 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
4233 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
4235 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
4237 defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
4240 } // ExeDomain = SSEPackedInt
4242 //===---------------------------------------------------------------------===//
4243 // SSE2 - Packed Integer Extract and Insert
4244 //===---------------------------------------------------------------------===//
4246 let ExeDomain = SSEPackedInt in {
4247 multiclass sse2_pinsrw<bit Is2Addr = 1> {
4248 def rri : Ii8<0xC4, MRMSrcReg,
4249 (outs VR128:$dst), (ins VR128:$src1,
4250 GR32:$src2, i32i8imm:$src3),
4252 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4253 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4255 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))], IIC_SSE_PINSRW>,
4256 Sched<[WriteShuffle]>;
4257 def rmi : Ii8<0xC4, MRMSrcMem,
4258 (outs VR128:$dst), (ins VR128:$src1,
4259 i16mem:$src2, i32i8imm:$src3),
4261 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4262 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4264 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
4265 imm:$src3))], IIC_SSE_PINSRW>,
4266 Sched<[WriteShuffleLd, ReadAfterLd]>;
4270 let Predicates = [HasAVX] in
4271 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
4272 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4273 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4274 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4275 imm:$src2))]>, TB, OpSize, VEX,
4276 Sched<[WriteShuffle]>;
4277 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
4278 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4279 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4280 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4281 imm:$src2))], IIC_SSE_PEXTRW>,
4282 Sched<[WriteShuffleLd, ReadAfterLd]>;
4285 let Predicates = [HasAVX] in {
4286 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
4287 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
4288 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4289 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
4290 []>, TB, OpSize, VEX_4V, Sched<[WriteShuffle]>;
4293 let Constraints = "$src1 = $dst" in
4294 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[UseSSE2]>;
4296 } // ExeDomain = SSEPackedInt
4298 //===---------------------------------------------------------------------===//
4299 // SSE2 - Packed Mask Creation
4300 //===---------------------------------------------------------------------===//
4302 let ExeDomain = SSEPackedInt, SchedRW = [WriteVecLogic] in {
4304 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4305 "pmovmskb\t{$src, $dst|$dst, $src}",
4306 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4307 IIC_SSE_MOVMSK>, VEX;
4308 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
4309 "pmovmskb\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK>, VEX;
4311 let Predicates = [HasAVX2] in {
4312 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
4313 "pmovmskb\t{$src, $dst|$dst, $src}",
4314 [(set GR32:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>, VEX, VEX_L;
4315 def VPMOVMSKBYr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
4316 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
4319 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4320 "pmovmskb\t{$src, $dst|$dst, $src}",
4321 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4323 def PMOVMSKBr64r : PDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
4324 "pmovmskb\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK>;
4326 } // ExeDomain = SSEPackedInt
4328 //===---------------------------------------------------------------------===//
4329 // SSE2 - Conditional Store
4330 //===---------------------------------------------------------------------===//
4332 let ExeDomain = SSEPackedInt, SchedRW = [WriteStore] in {
4334 let Uses = [EDI], Predicates = [HasAVX,In32BitMode] in
4335 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4336 (ins VR128:$src, VR128:$mask),
4337 "maskmovdqu\t{$mask, $src|$src, $mask}",
4338 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4339 IIC_SSE_MASKMOV>, VEX;
4340 let Uses = [RDI], Predicates = [HasAVX,In64BitMode] in
4341 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4342 (ins VR128:$src, VR128:$mask),
4343 "maskmovdqu\t{$mask, $src|$src, $mask}",
4344 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4345 IIC_SSE_MASKMOV>, VEX;
4347 let Uses = [EDI], Predicates = [UseSSE2,In32BitMode] in
4348 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4349 "maskmovdqu\t{$mask, $src|$src, $mask}",
4350 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4352 let Uses = [RDI], Predicates = [UseSSE2,In64BitMode] in
4353 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4354 "maskmovdqu\t{$mask, $src|$src, $mask}",
4355 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4358 } // ExeDomain = SSEPackedInt
4360 //===---------------------------------------------------------------------===//
4361 // SSE2 - Move Doubleword
4362 //===---------------------------------------------------------------------===//
4364 //===---------------------------------------------------------------------===//
4365 // Move Int Doubleword to Packed Double Int
4367 def VMOVDI2PDIrr : VS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4368 "movd\t{$src, $dst|$dst, $src}",
4370 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>,
4371 VEX, Sched<[WriteMove]>;
4372 def VMOVDI2PDIrm : VS2I<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4373 "movd\t{$src, $dst|$dst, $src}",
4375 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4377 VEX, Sched<[WriteLoad]>;
4378 def VMOV64toPQIrr : VRS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4379 "movq\t{$src, $dst|$dst, $src}",
4381 (v2i64 (scalar_to_vector GR64:$src)))],
4382 IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
4383 let isCodeGenOnly = 1 in
4384 def VMOV64toSDrr : VRS2I<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4385 "movq\t{$src, $dst|$dst, $src}",
4386 [(set FR64:$dst, (bitconvert GR64:$src))],
4387 IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
4389 def MOVDI2PDIrr : S2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4390 "movd\t{$src, $dst|$dst, $src}",
4392 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>,
4394 def MOVDI2PDIrm : S2I<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4395 "movd\t{$src, $dst|$dst, $src}",
4397 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4398 IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
4399 def MOV64toPQIrr : RS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4400 "mov{d|q}\t{$src, $dst|$dst, $src}",
4402 (v2i64 (scalar_to_vector GR64:$src)))],
4403 IIC_SSE_MOVDQ>, Sched<[WriteMove]>;
4404 let isCodeGenOnly = 1 in
4405 def MOV64toSDrr : RS2I<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4406 "mov{d|q}\t{$src, $dst|$dst, $src}",
4407 [(set FR64:$dst, (bitconvert GR64:$src))],
4408 IIC_SSE_MOVDQ>, Sched<[WriteMove]>;
4410 //===---------------------------------------------------------------------===//
4411 // Move Int Doubleword to Single Scalar
4413 let isCodeGenOnly = 1 in {
4414 def VMOVDI2SSrr : VS2I<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4415 "movd\t{$src, $dst|$dst, $src}",
4416 [(set FR32:$dst, (bitconvert GR32:$src))],
4417 IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
4419 def VMOVDI2SSrm : VS2I<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4420 "movd\t{$src, $dst|$dst, $src}",
4421 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4423 VEX, Sched<[WriteLoad]>;
4424 def MOVDI2SSrr : S2I<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4425 "movd\t{$src, $dst|$dst, $src}",
4426 [(set FR32:$dst, (bitconvert GR32:$src))],
4427 IIC_SSE_MOVDQ>, Sched<[WriteMove]>;
4429 def MOVDI2SSrm : S2I<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4430 "movd\t{$src, $dst|$dst, $src}",
4431 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4432 IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
4435 //===---------------------------------------------------------------------===//
4436 // Move Packed Doubleword Int to Packed Double Int
4438 def VMOVPDI2DIrr : VS2I<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4439 "movd\t{$src, $dst|$dst, $src}",
4440 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4441 (iPTR 0)))], IIC_SSE_MOVD_ToGP>, VEX,
4443 def VMOVPDI2DImr : VS2I<0x7E, MRMDestMem, (outs),
4444 (ins i32mem:$dst, VR128:$src),
4445 "movd\t{$src, $dst|$dst, $src}",
4446 [(store (i32 (vector_extract (v4i32 VR128:$src),
4447 (iPTR 0))), addr:$dst)], IIC_SSE_MOVDQ>,
4448 VEX, Sched<[WriteLoad]>;
4449 def MOVPDI2DIrr : S2I<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4450 "movd\t{$src, $dst|$dst, $src}",
4451 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4452 (iPTR 0)))], IIC_SSE_MOVD_ToGP>,
4454 def MOVPDI2DImr : S2I<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4455 "movd\t{$src, $dst|$dst, $src}",
4456 [(store (i32 (vector_extract (v4i32 VR128:$src),
4457 (iPTR 0))), addr:$dst)],
4458 IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
4460 def : Pat<(v8i32 (X86Vinsert (v8i32 immAllZerosV), GR32:$src2, (iPTR 0))),
4461 (SUBREG_TO_REG (i32 0), (VMOVDI2PDIrr GR32:$src2), sub_xmm)>;
4463 def : Pat<(v4i64 (X86Vinsert (bc_v4i64 (v8i32 immAllZerosV)), GR64:$src2, (iPTR 0))),
4464 (SUBREG_TO_REG (i32 0), (VMOV64toPQIrr GR64:$src2), sub_xmm)>;
4466 def : Pat<(v8i32 (X86Vinsert undef, GR32:$src2, (iPTR 0))),
4467 (SUBREG_TO_REG (i32 0), (VMOVDI2PDIrr GR32:$src2), sub_xmm)>;
4469 def : Pat<(v4i64 (X86Vinsert undef, GR64:$src2, (iPTR 0))),
4470 (SUBREG_TO_REG (i32 0), (VMOV64toPQIrr GR64:$src2), sub_xmm)>;
4472 //===---------------------------------------------------------------------===//
4473 // Move Packed Doubleword Int first element to Doubleword Int
4475 let SchedRW = [WriteMove] in {
4476 def VMOVPQIto64rr : VRS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4477 "movq\t{$src, $dst|$dst, $src}",
4478 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4483 def MOVPQIto64rr : RS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4484 "mov{d|q}\t{$src, $dst|$dst, $src}",
4485 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4490 //===---------------------------------------------------------------------===//
4491 // Bitcast FR64 <-> GR64
4493 let isCodeGenOnly = 1 in {
4494 let Predicates = [UseAVX] in
4495 def VMOV64toSDrm : VS2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4496 "movq\t{$src, $dst|$dst, $src}",
4497 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4498 VEX, Sched<[WriteLoad]>;
4499 def VMOVSDto64rr : VRS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4500 "movq\t{$src, $dst|$dst, $src}",
4501 [(set GR64:$dst, (bitconvert FR64:$src))],
4502 IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
4503 def VMOVSDto64mr : VRS2I<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4504 "movq\t{$src, $dst|$dst, $src}",
4505 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4506 IIC_SSE_MOVDQ>, VEX, Sched<[WriteStore]>;
4508 def MOV64toSDrm : S2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4509 "movq\t{$src, $dst|$dst, $src}",
4510 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))],
4511 IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
4512 def MOVSDto64rr : RS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4513 "mov{d|q}\t{$src, $dst|$dst, $src}",
4514 [(set GR64:$dst, (bitconvert FR64:$src))],
4515 IIC_SSE_MOVD_ToGP>, Sched<[WriteMove]>;
4516 def MOVSDto64mr : RS2I<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4517 "movq\t{$src, $dst|$dst, $src}",
4518 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4519 IIC_SSE_MOVDQ>, Sched<[WriteStore]>;
4522 //===---------------------------------------------------------------------===//
4523 // Move Scalar Single to Double Int
4525 let isCodeGenOnly = 1 in {
4526 def VMOVSS2DIrr : VS2I<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4527 "movd\t{$src, $dst|$dst, $src}",
4528 [(set GR32:$dst, (bitconvert FR32:$src))],
4529 IIC_SSE_MOVD_ToGP>, VEX, Sched<[WriteMove]>;
4530 def VMOVSS2DImr : VS2I<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4531 "movd\t{$src, $dst|$dst, $src}",
4532 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4533 IIC_SSE_MOVDQ>, VEX, Sched<[WriteStore]>;
4534 def MOVSS2DIrr : S2I<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4535 "movd\t{$src, $dst|$dst, $src}",
4536 [(set GR32:$dst, (bitconvert FR32:$src))],
4537 IIC_SSE_MOVD_ToGP>, Sched<[WriteMove]>;
4538 def MOVSS2DImr : S2I<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4539 "movd\t{$src, $dst|$dst, $src}",
4540 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4541 IIC_SSE_MOVDQ>, Sched<[WriteStore]>;
4544 //===---------------------------------------------------------------------===//
4545 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4547 let isCodeGenOnly = 1, SchedRW = [WriteMove] in {
4548 let AddedComplexity = 15 in {
4549 def VMOVZDI2PDIrr : VS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4550 "movd\t{$src, $dst|$dst, $src}",
4551 [(set VR128:$dst, (v4i32 (X86vzmovl
4552 (v4i32 (scalar_to_vector GR32:$src)))))],
4553 IIC_SSE_MOVDQ>, VEX;
4554 def VMOVZQI2PQIrr : VS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4555 "movq\t{$src, $dst|$dst, $src}", // X86-64 only
4556 [(set VR128:$dst, (v2i64 (X86vzmovl
4557 (v2i64 (scalar_to_vector GR64:$src)))))],
4561 let AddedComplexity = 15 in {
4562 def MOVZDI2PDIrr : S2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4563 "movd\t{$src, $dst|$dst, $src}",
4564 [(set VR128:$dst, (v4i32 (X86vzmovl
4565 (v4i32 (scalar_to_vector GR32:$src)))))],
4567 def MOVZQI2PQIrr : RS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4568 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4569 [(set VR128:$dst, (v2i64 (X86vzmovl
4570 (v2i64 (scalar_to_vector GR64:$src)))))],
4573 } // isCodeGenOnly, SchedRW
4575 let isCodeGenOnly = 1, AddedComplexity = 20, SchedRW = [WriteLoad] in {
4576 def VMOVZDI2PDIrm : VS2I<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4577 "movd\t{$src, $dst|$dst, $src}",
4579 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4580 (loadi32 addr:$src))))))],
4581 IIC_SSE_MOVDQ>, VEX;
4582 def MOVZDI2PDIrm : S2I<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4583 "movd\t{$src, $dst|$dst, $src}",
4585 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4586 (loadi32 addr:$src))))))],
4588 } // isCodeGenOnly, AddedComplexity, SchedRW
4590 let Predicates = [UseAVX] in {
4591 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4592 let AddedComplexity = 20 in {
4593 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4594 (VMOVZDI2PDIrm addr:$src)>;
4595 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4596 (VMOVZDI2PDIrm addr:$src)>;
4598 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4599 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4600 (v4i32 (scalar_to_vector GR32:$src)),(iPTR 0)))),
4601 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
4602 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4603 (v2i64 (scalar_to_vector GR64:$src)),(iPTR 0)))),
4604 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4607 let Predicates = [UseSSE2], AddedComplexity = 20 in {
4608 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4609 (MOVZDI2PDIrm addr:$src)>;
4610 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4611 (MOVZDI2PDIrm addr:$src)>;
4614 // These are the correct encodings of the instructions so that we know how to
4615 // read correct assembly, even though we continue to emit the wrong ones for
4616 // compatibility with Darwin's buggy assembler.
4617 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4618 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4619 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4620 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4621 // Allow "vmovd" but print "vmovq" since we don't need compatibility for AVX.
4622 def : InstAlias<"vmovd\t{$src, $dst|$dst, $src}",
4623 (VMOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4624 def : InstAlias<"vmovd\t{$src, $dst|$dst, $src}",
4625 (VMOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4627 //===---------------------------------------------------------------------===//
4628 // SSE2 - Move Quadword
4629 //===---------------------------------------------------------------------===//
4631 //===---------------------------------------------------------------------===//
4632 // Move Quadword Int to Packed Quadword Int
4635 let SchedRW = [WriteLoad] in {
4636 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4637 "vmovq\t{$src, $dst|$dst, $src}",
4639 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4640 VEX, Requires<[UseAVX]>;
4641 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4642 "movq\t{$src, $dst|$dst, $src}",
4644 (v2i64 (scalar_to_vector (loadi64 addr:$src))))],
4646 Requires<[UseSSE2]>; // SSE2 instruction with XS Prefix
4649 //===---------------------------------------------------------------------===//
4650 // Move Packed Quadword Int to Quadword Int
4652 let SchedRW = [WriteStore] in {
4653 def VMOVPQI2QImr : VS2I<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4654 "movq\t{$src, $dst|$dst, $src}",
4655 [(store (i64 (vector_extract (v2i64 VR128:$src),
4656 (iPTR 0))), addr:$dst)],
4657 IIC_SSE_MOVDQ>, VEX;
4658 def MOVPQI2QImr : S2I<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4659 "movq\t{$src, $dst|$dst, $src}",
4660 [(store (i64 (vector_extract (v2i64 VR128:$src),
4661 (iPTR 0))), addr:$dst)],
4665 //===---------------------------------------------------------------------===//
4666 // Store / copy lower 64-bits of a XMM register.
4668 def VMOVLQ128mr : VS2I<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4669 "movq\t{$src, $dst|$dst, $src}",
4670 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX,
4671 Sched<[WriteStore]>;
4672 def MOVLQ128mr : S2I<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4673 "movq\t{$src, $dst|$dst, $src}",
4674 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)],
4675 IIC_SSE_MOVDQ>, Sched<[WriteStore]>;
4677 let isCodeGenOnly = 1, AddedComplexity = 20 in {
4678 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4679 "vmovq\t{$src, $dst|$dst, $src}",
4681 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4682 (loadi64 addr:$src))))))],
4684 XS, VEX, Requires<[UseAVX]>, Sched<[WriteLoad]>;
4686 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4687 "movq\t{$src, $dst|$dst, $src}",
4689 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4690 (loadi64 addr:$src))))))],
4692 XS, Requires<[UseSSE2]>, Sched<[WriteLoad]>;
4695 let Predicates = [UseAVX], AddedComplexity = 20 in {
4696 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4697 (VMOVZQI2PQIrm addr:$src)>;
4698 def : Pat<(v2i64 (X86vzload addr:$src)),
4699 (VMOVZQI2PQIrm addr:$src)>;
4702 let Predicates = [UseSSE2], AddedComplexity = 20 in {
4703 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4704 (MOVZQI2PQIrm addr:$src)>;
4705 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4708 let Predicates = [HasAVX] in {
4709 def : Pat<(v4i64 (alignedX86vzload addr:$src)),
4710 (SUBREG_TO_REG (i32 0), (VMOVAPSrm addr:$src), sub_xmm)>;
4711 def : Pat<(v4i64 (X86vzload addr:$src)),
4712 (SUBREG_TO_REG (i32 0), (VMOVUPSrm addr:$src), sub_xmm)>;
4715 //===---------------------------------------------------------------------===//
4716 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4717 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4719 let SchedRW = [WriteVecLogic] in {
4720 let AddedComplexity = 15 in
4721 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4722 "vmovq\t{$src, $dst|$dst, $src}",
4723 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4725 XS, VEX, Requires<[UseAVX]>;
4726 let AddedComplexity = 15 in
4727 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4728 "movq\t{$src, $dst|$dst, $src}",
4729 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4731 XS, Requires<[UseSSE2]>;
4734 let isCodeGenOnly = 1, SchedRW = [WriteVecLogicLd] in {
4735 let AddedComplexity = 20 in
4736 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4737 "vmovq\t{$src, $dst|$dst, $src}",
4738 [(set VR128:$dst, (v2i64 (X86vzmovl
4739 (loadv2i64 addr:$src))))],
4741 XS, VEX, Requires<[UseAVX]>;
4742 let AddedComplexity = 20 in {
4743 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4744 "movq\t{$src, $dst|$dst, $src}",
4745 [(set VR128:$dst, (v2i64 (X86vzmovl
4746 (loadv2i64 addr:$src))))],
4748 XS, Requires<[UseSSE2]>;
4750 } // isCodeGenOnly, SchedRW
4752 let AddedComplexity = 20 in {
4753 let Predicates = [UseAVX] in {
4754 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4755 (VMOVZPQILo2PQIrr VR128:$src)>;
4757 let Predicates = [UseSSE2] in {
4758 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4759 (MOVZPQILo2PQIrr VR128:$src)>;
4763 //===---------------------------------------------------------------------===//
4764 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4765 //===---------------------------------------------------------------------===//
4766 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4767 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4768 X86MemOperand x86memop> {
4769 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4770 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4771 [(set RC:$dst, (vt (OpNode RC:$src)))],
4772 IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>;
4773 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4774 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4775 [(set RC:$dst, (OpNode (mem_frag addr:$src)))],
4776 IIC_SSE_MOV_LH>, Sched<[WriteShuffleLd]>;
4779 let Predicates = [HasAVX] in {
4780 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4781 v4f32, VR128, memopv4f32, f128mem>, VEX;
4782 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4783 v4f32, VR128, memopv4f32, f128mem>, VEX;
4784 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4785 v8f32, VR256, memopv8f32, f256mem>, VEX, VEX_L;
4786 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4787 v8f32, VR256, memopv8f32, f256mem>, VEX, VEX_L;
4789 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4790 memopv4f32, f128mem>;
4791 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4792 memopv4f32, f128mem>;
4794 let Predicates = [HasAVX] in {
4795 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4796 (VMOVSHDUPrr VR128:$src)>;
4797 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4798 (VMOVSHDUPrm addr:$src)>;
4799 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4800 (VMOVSLDUPrr VR128:$src)>;
4801 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4802 (VMOVSLDUPrm addr:$src)>;
4803 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4804 (VMOVSHDUPYrr VR256:$src)>;
4805 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4806 (VMOVSHDUPYrm addr:$src)>;
4807 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4808 (VMOVSLDUPYrr VR256:$src)>;
4809 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
4810 (VMOVSLDUPYrm addr:$src)>;
4813 let Predicates = [UseSSE3] in {
4814 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4815 (MOVSHDUPrr VR128:$src)>;
4816 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4817 (MOVSHDUPrm addr:$src)>;
4818 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4819 (MOVSLDUPrr VR128:$src)>;
4820 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4821 (MOVSLDUPrm addr:$src)>;
4824 //===---------------------------------------------------------------------===//
4825 // SSE3 - Replicate Double FP - MOVDDUP
4826 //===---------------------------------------------------------------------===//
4828 multiclass sse3_replicate_dfp<string OpcodeStr> {
4829 let neverHasSideEffects = 1 in
4830 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4831 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4832 [], IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>;
4833 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
4834 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4837 (scalar_to_vector (loadf64 addr:$src)))))],
4838 IIC_SSE_MOV_LH>, Sched<[WriteShuffleLd]>;
4841 // FIXME: Merge with above classe when there're patterns for the ymm version
4842 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
4843 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
4844 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4845 [(set VR256:$dst, (v4f64 (X86Movddup VR256:$src)))]>,
4846 Sched<[WriteShuffle]>;
4847 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
4848 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4851 (scalar_to_vector (loadf64 addr:$src)))))]>,
4852 Sched<[WriteShuffleLd]>;
4855 let Predicates = [HasAVX] in {
4856 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
4857 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX, VEX_L;
4860 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
4862 let Predicates = [HasAVX] in {
4863 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4864 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4865 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4866 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4867 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4868 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4869 def : Pat<(X86Movddup (bc_v2f64
4870 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4871 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4874 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
4875 (VMOVDDUPYrm addr:$src)>;
4876 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
4877 (VMOVDDUPYrm addr:$src)>;
4878 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
4879 (VMOVDDUPYrm addr:$src)>;
4880 def : Pat<(X86Movddup (v4i64 VR256:$src)),
4881 (VMOVDDUPYrr VR256:$src)>;
4884 let Predicates = [UseSSE3] in {
4885 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4886 (MOVDDUPrm addr:$src)>;
4887 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4888 (MOVDDUPrm addr:$src)>;
4889 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4890 (MOVDDUPrm addr:$src)>;
4891 def : Pat<(X86Movddup (bc_v2f64
4892 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4893 (MOVDDUPrm addr:$src)>;
4896 //===---------------------------------------------------------------------===//
4897 // SSE3 - Move Unaligned Integer
4898 //===---------------------------------------------------------------------===//
4900 let SchedRW = [WriteLoad] in {
4901 let Predicates = [HasAVX] in {
4902 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4903 "vlddqu\t{$src, $dst|$dst, $src}",
4904 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
4905 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
4906 "vlddqu\t{$src, $dst|$dst, $src}",
4907 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>,
4910 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4911 "lddqu\t{$src, $dst|$dst, $src}",
4912 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))],
4916 //===---------------------------------------------------------------------===//
4917 // SSE3 - Arithmetic
4918 //===---------------------------------------------------------------------===//
4920 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
4921 X86MemOperand x86memop, OpndItins itins,
4923 def rr : I<0xD0, MRMSrcReg,
4924 (outs RC:$dst), (ins RC:$src1, RC:$src2),
4926 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4927 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4928 [(set RC:$dst, (Int RC:$src1, RC:$src2))], itins.rr>,
4929 Sched<[itins.Sched]>;
4930 def rm : I<0xD0, MRMSrcMem,
4931 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4933 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4934 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4935 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))], itins.rr>,
4936 Sched<[itins.Sched.Folded, ReadAfterLd]>;
4939 let Predicates = [HasAVX] in {
4940 let ExeDomain = SSEPackedSingle in {
4941 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
4942 f128mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V;
4943 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
4944 f256mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V, VEX_L;
4946 let ExeDomain = SSEPackedDouble in {
4947 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
4948 f128mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V;
4949 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
4950 f256mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V, VEX_L;
4953 let Constraints = "$src1 = $dst", Predicates = [UseSSE3] in {
4954 let ExeDomain = SSEPackedSingle in
4955 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
4956 f128mem, SSE_ALU_F32P>, TB, XD;
4957 let ExeDomain = SSEPackedDouble in
4958 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
4959 f128mem, SSE_ALU_F64P>, TB, OpSize;
4962 //===---------------------------------------------------------------------===//
4963 // SSE3 Instructions
4964 //===---------------------------------------------------------------------===//
4967 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4968 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
4969 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4971 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4972 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4973 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>,
4976 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4978 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4979 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4980 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
4981 IIC_SSE_HADDSUB_RM>, Sched<[WriteFAddLd, ReadAfterLd]>;
4983 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4984 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
4985 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4987 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4988 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4989 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>,
4992 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4994 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4995 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4996 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
4997 IIC_SSE_HADDSUB_RM>, Sched<[WriteFAddLd, ReadAfterLd]>;
5000 let Predicates = [HasAVX] in {
5001 let ExeDomain = SSEPackedSingle in {
5002 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
5003 X86fhadd, 0>, VEX_4V;
5004 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
5005 X86fhsub, 0>, VEX_4V;
5006 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
5007 X86fhadd, 0>, VEX_4V, VEX_L;
5008 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
5009 X86fhsub, 0>, VEX_4V, VEX_L;
5011 let ExeDomain = SSEPackedDouble in {
5012 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
5013 X86fhadd, 0>, VEX_4V;
5014 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
5015 X86fhsub, 0>, VEX_4V;
5016 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
5017 X86fhadd, 0>, VEX_4V, VEX_L;
5018 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
5019 X86fhsub, 0>, VEX_4V, VEX_L;
5023 let Constraints = "$src1 = $dst" in {
5024 let ExeDomain = SSEPackedSingle in {
5025 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
5026 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
5028 let ExeDomain = SSEPackedDouble in {
5029 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
5030 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
5034 //===---------------------------------------------------------------------===//
5035 // SSSE3 - Packed Absolute Instructions
5036 //===---------------------------------------------------------------------===//
5039 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5040 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
5041 Intrinsic IntId128> {
5042 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5044 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5045 [(set VR128:$dst, (IntId128 VR128:$src))], IIC_SSE_PABS_RR>,
5046 OpSize, Sched<[WriteVecALU]>;
5048 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5050 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5053 (bitconvert (memopv2i64 addr:$src))))], IIC_SSE_PABS_RM>,
5054 OpSize, Sched<[WriteVecALULd]>;
5057 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5058 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
5059 Intrinsic IntId256> {
5060 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5062 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5063 [(set VR256:$dst, (IntId256 VR256:$src))]>,
5064 OpSize, Sched<[WriteVecALU]>;
5066 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5068 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5071 (bitconvert (memopv4i64 addr:$src))))]>, OpSize,
5072 Sched<[WriteVecALULd]>;
5075 // Helper fragments to match sext vXi1 to vXiY.
5076 def v16i1sextv16i8 : PatLeaf<(v16i8 (X86pcmpgt (bc_v16i8 (v4i32 immAllZerosV)),
5078 def v8i1sextv8i16 : PatLeaf<(v8i16 (X86vsrai VR128:$src, (i32 15)))>;
5079 def v4i1sextv4i32 : PatLeaf<(v4i32 (X86vsrai VR128:$src, (i32 31)))>;
5080 def v32i1sextv32i8 : PatLeaf<(v32i8 (X86pcmpgt (bc_v32i8 (v8i32 immAllZerosV)),
5082 def v16i1sextv16i16: PatLeaf<(v16i16 (X86vsrai VR256:$src, (i32 15)))>;
5083 def v8i1sextv8i32 : PatLeaf<(v8i32 (X86vsrai VR256:$src, (i32 31)))>;
5085 let Predicates = [HasAVX] in {
5086 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb",
5087 int_x86_ssse3_pabs_b_128>, VEX;
5088 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw",
5089 int_x86_ssse3_pabs_w_128>, VEX;
5090 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd",
5091 int_x86_ssse3_pabs_d_128>, VEX;
5094 (bc_v2i64 (v16i1sextv16i8)),
5095 (bc_v2i64 (add (v16i8 VR128:$src), (v16i1sextv16i8)))),
5096 (VPABSBrr128 VR128:$src)>;
5098 (bc_v2i64 (v8i1sextv8i16)),
5099 (bc_v2i64 (add (v8i16 VR128:$src), (v8i1sextv8i16)))),
5100 (VPABSWrr128 VR128:$src)>;
5102 (bc_v2i64 (v4i1sextv4i32)),
5103 (bc_v2i64 (add (v4i32 VR128:$src), (v4i1sextv4i32)))),
5104 (VPABSDrr128 VR128:$src)>;
5107 let Predicates = [HasAVX2] in {
5108 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb",
5109 int_x86_avx2_pabs_b>, VEX, VEX_L;
5110 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw",
5111 int_x86_avx2_pabs_w>, VEX, VEX_L;
5112 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd",
5113 int_x86_avx2_pabs_d>, VEX, VEX_L;
5116 (bc_v4i64 (v32i1sextv32i8)),
5117 (bc_v4i64 (add (v32i8 VR256:$src), (v32i1sextv32i8)))),
5118 (VPABSBrr256 VR256:$src)>;
5120 (bc_v4i64 (v16i1sextv16i16)),
5121 (bc_v4i64 (add (v16i16 VR256:$src), (v16i1sextv16i16)))),
5122 (VPABSWrr256 VR256:$src)>;
5124 (bc_v4i64 (v8i1sextv8i32)),
5125 (bc_v4i64 (add (v8i32 VR256:$src), (v8i1sextv8i32)))),
5126 (VPABSDrr256 VR256:$src)>;
5129 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb",
5130 int_x86_ssse3_pabs_b_128>;
5131 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw",
5132 int_x86_ssse3_pabs_w_128>;
5133 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
5134 int_x86_ssse3_pabs_d_128>;
5136 let Predicates = [HasSSSE3] in {
5138 (bc_v2i64 (v16i1sextv16i8)),
5139 (bc_v2i64 (add (v16i8 VR128:$src), (v16i1sextv16i8)))),
5140 (PABSBrr128 VR128:$src)>;
5142 (bc_v2i64 (v8i1sextv8i16)),
5143 (bc_v2i64 (add (v8i16 VR128:$src), (v8i1sextv8i16)))),
5144 (PABSWrr128 VR128:$src)>;
5146 (bc_v2i64 (v4i1sextv4i32)),
5147 (bc_v2i64 (add (v4i32 VR128:$src), (v4i1sextv4i32)))),
5148 (PABSDrr128 VR128:$src)>;
5151 //===---------------------------------------------------------------------===//
5152 // SSSE3 - Packed Binary Operator Instructions
5153 //===---------------------------------------------------------------------===//
5155 let Sched = WriteVecALU in {
5156 def SSE_PHADDSUBD : OpndItins<
5157 IIC_SSE_PHADDSUBD_RR, IIC_SSE_PHADDSUBD_RM
5159 def SSE_PHADDSUBSW : OpndItins<
5160 IIC_SSE_PHADDSUBSW_RR, IIC_SSE_PHADDSUBSW_RM
5162 def SSE_PHADDSUBW : OpndItins<
5163 IIC_SSE_PHADDSUBW_RR, IIC_SSE_PHADDSUBW_RM
5166 let Sched = WriteShuffle in
5167 def SSE_PSHUFB : OpndItins<
5168 IIC_SSE_PSHUFB_RR, IIC_SSE_PSHUFB_RM
5170 let Sched = WriteVecALU in
5171 def SSE_PSIGN : OpndItins<
5172 IIC_SSE_PSIGN_RR, IIC_SSE_PSIGN_RM
5174 let Sched = WriteVecIMul in
5175 def SSE_PMULHRSW : OpndItins<
5176 IIC_SSE_PMULHRSW, IIC_SSE_PMULHRSW
5179 /// SS3I_binop_rm - Simple SSSE3 bin op
5180 multiclass SS3I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
5181 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
5182 X86MemOperand x86memop, OpndItins itins,
5184 let isCommutable = 1 in
5185 def rr : SS38I<opc, MRMSrcReg, (outs RC:$dst),
5186 (ins RC:$src1, RC:$src2),
5188 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5189 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5190 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>,
5191 OpSize, Sched<[itins.Sched]>;
5192 def rm : SS38I<opc, MRMSrcMem, (outs RC:$dst),
5193 (ins RC:$src1, x86memop:$src2),
5195 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5196 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5198 (OpVT (OpNode RC:$src1,
5199 (bitconvert (memop_frag addr:$src2)))))], itins.rm>, OpSize,
5200 Sched<[itins.Sched.Folded, ReadAfterLd]>;
5203 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
5204 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
5205 Intrinsic IntId128, OpndItins itins,
5207 let isCommutable = 1 in
5208 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5209 (ins VR128:$src1, VR128:$src2),
5211 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5212 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5213 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5214 OpSize, Sched<[itins.Sched]>;
5215 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5216 (ins VR128:$src1, i128mem:$src2),
5218 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5219 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5221 (IntId128 VR128:$src1,
5222 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize,
5223 Sched<[itins.Sched.Folded, ReadAfterLd]>;
5226 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
5227 Intrinsic IntId256> {
5228 let isCommutable = 1 in
5229 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5230 (ins VR256:$src1, VR256:$src2),
5231 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5232 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
5234 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5235 (ins VR256:$src1, i256mem:$src2),
5236 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5238 (IntId256 VR256:$src1,
5239 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
5242 let ImmT = NoImm, Predicates = [HasAVX] in {
5243 let isCommutable = 0 in {
5244 defm VPHADDW : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v8i16, VR128,
5245 memopv2i64, i128mem,
5246 SSE_PHADDSUBW, 0>, VEX_4V;
5247 defm VPHADDD : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v4i32, VR128,
5248 memopv2i64, i128mem,
5249 SSE_PHADDSUBD, 0>, VEX_4V;
5250 defm VPHSUBW : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v8i16, VR128,
5251 memopv2i64, i128mem,
5252 SSE_PHADDSUBW, 0>, VEX_4V;
5253 defm VPHSUBD : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v4i32, VR128,
5254 memopv2i64, i128mem,
5255 SSE_PHADDSUBD, 0>, VEX_4V;
5256 defm VPSIGNB : SS3I_binop_rm<0x08, "vpsignb", X86psign, v16i8, VR128,
5257 memopv2i64, i128mem,
5258 SSE_PSIGN, 0>, VEX_4V;
5259 defm VPSIGNW : SS3I_binop_rm<0x09, "vpsignw", X86psign, v8i16, VR128,
5260 memopv2i64, i128mem,
5261 SSE_PSIGN, 0>, VEX_4V;
5262 defm VPSIGND : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v4i32, VR128,
5263 memopv2i64, i128mem,
5264 SSE_PSIGN, 0>, VEX_4V;
5265 defm VPSHUFB : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v16i8, VR128,
5266 memopv2i64, i128mem,
5267 SSE_PSHUFB, 0>, VEX_4V;
5268 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw",
5269 int_x86_ssse3_phadd_sw_128,
5270 SSE_PHADDSUBSW, 0>, VEX_4V;
5271 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw",
5272 int_x86_ssse3_phsub_sw_128,
5273 SSE_PHADDSUBSW, 0>, VEX_4V;
5274 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw",
5275 int_x86_ssse3_pmadd_ub_sw_128,
5276 SSE_PMADD, 0>, VEX_4V;
5278 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw",
5279 int_x86_ssse3_pmul_hr_sw_128,
5280 SSE_PMULHRSW, 0>, VEX_4V;
5283 let ImmT = NoImm, Predicates = [HasAVX2] in {
5284 let isCommutable = 0 in {
5285 defm VPHADDWY : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v16i16, VR256,
5286 memopv4i64, i256mem,
5287 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5288 defm VPHADDDY : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v8i32, VR256,
5289 memopv4i64, i256mem,
5290 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5291 defm VPHSUBWY : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v16i16, VR256,
5292 memopv4i64, i256mem,
5293 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5294 defm VPHSUBDY : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v8i32, VR256,
5295 memopv4i64, i256mem,
5296 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5297 defm VPSIGNBY : SS3I_binop_rm<0x08, "vpsignb", X86psign, v32i8, VR256,
5298 memopv4i64, i256mem,
5299 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5300 defm VPSIGNWY : SS3I_binop_rm<0x09, "vpsignw", X86psign, v16i16, VR256,
5301 memopv4i64, i256mem,
5302 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5303 defm VPSIGNDY : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v8i32, VR256,
5304 memopv4i64, i256mem,
5305 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5306 defm VPSHUFBY : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v32i8, VR256,
5307 memopv4i64, i256mem,
5308 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5309 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
5310 int_x86_avx2_phadd_sw>, VEX_4V, VEX_L;
5311 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
5312 int_x86_avx2_phsub_sw>, VEX_4V, VEX_L;
5313 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
5314 int_x86_avx2_pmadd_ub_sw>, VEX_4V, VEX_L;
5316 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
5317 int_x86_avx2_pmul_hr_sw>, VEX_4V, VEX_L;
5320 // None of these have i8 immediate fields.
5321 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
5322 let isCommutable = 0 in {
5323 defm PHADDW : SS3I_binop_rm<0x01, "phaddw", X86hadd, v8i16, VR128,
5324 memopv2i64, i128mem, SSE_PHADDSUBW>;
5325 defm PHADDD : SS3I_binop_rm<0x02, "phaddd", X86hadd, v4i32, VR128,
5326 memopv2i64, i128mem, SSE_PHADDSUBD>;
5327 defm PHSUBW : SS3I_binop_rm<0x05, "phsubw", X86hsub, v8i16, VR128,
5328 memopv2i64, i128mem, SSE_PHADDSUBW>;
5329 defm PHSUBD : SS3I_binop_rm<0x06, "phsubd", X86hsub, v4i32, VR128,
5330 memopv2i64, i128mem, SSE_PHADDSUBD>;
5331 defm PSIGNB : SS3I_binop_rm<0x08, "psignb", X86psign, v16i8, VR128,
5332 memopv2i64, i128mem, SSE_PSIGN>;
5333 defm PSIGNW : SS3I_binop_rm<0x09, "psignw", X86psign, v8i16, VR128,
5334 memopv2i64, i128mem, SSE_PSIGN>;
5335 defm PSIGND : SS3I_binop_rm<0x0A, "psignd", X86psign, v4i32, VR128,
5336 memopv2i64, i128mem, SSE_PSIGN>;
5337 defm PSHUFB : SS3I_binop_rm<0x00, "pshufb", X86pshufb, v16i8, VR128,
5338 memopv2i64, i128mem, SSE_PSHUFB>;
5339 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw",
5340 int_x86_ssse3_phadd_sw_128,
5342 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw",
5343 int_x86_ssse3_phsub_sw_128,
5345 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw",
5346 int_x86_ssse3_pmadd_ub_sw_128, SSE_PMADD>;
5348 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw",
5349 int_x86_ssse3_pmul_hr_sw_128,
5353 //===---------------------------------------------------------------------===//
5354 // SSSE3 - Packed Align Instruction Patterns
5355 //===---------------------------------------------------------------------===//
5357 multiclass ssse3_palignr<string asm, bit Is2Addr = 1> {
5358 let neverHasSideEffects = 1 in {
5359 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
5360 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5362 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5364 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5365 [], IIC_SSE_PALIGNRR>, OpSize, Sched<[WriteShuffle]>;
5367 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
5368 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5370 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5372 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5373 [], IIC_SSE_PALIGNRM>, OpSize, Sched<[WriteShuffleLd, ReadAfterLd]>;
5377 multiclass ssse3_palignr_y<string asm, bit Is2Addr = 1> {
5378 let neverHasSideEffects = 1 in {
5379 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
5380 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5382 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5383 []>, OpSize, Sched<[WriteShuffle]>;
5385 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
5386 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
5388 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5389 []>, OpSize, Sched<[WriteShuffleLd, ReadAfterLd]>;
5393 let Predicates = [HasAVX] in
5394 defm VPALIGN : ssse3_palignr<"vpalignr", 0>, VEX_4V;
5395 let Predicates = [HasAVX2] in
5396 defm VPALIGN : ssse3_palignr_y<"vpalignr", 0>, VEX_4V, VEX_L;
5397 let Constraints = "$src1 = $dst", Predicates = [UseSSSE3] in
5398 defm PALIGN : ssse3_palignr<"palignr">;
5400 let Predicates = [HasAVX2] in {
5401 def : Pat<(v8i32 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5402 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5403 def : Pat<(v8f32 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5404 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5405 def : Pat<(v16i16 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5406 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5407 def : Pat<(v32i8 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5408 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5411 let Predicates = [HasAVX] in {
5412 def : Pat<(v4i32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5413 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5414 def : Pat<(v4f32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5415 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5416 def : Pat<(v8i16 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5417 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5418 def : Pat<(v16i8 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5419 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5422 let Predicates = [UseSSSE3] in {
5423 def : Pat<(v4i32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5424 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5425 def : Pat<(v4f32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5426 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5427 def : Pat<(v8i16 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5428 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5429 def : Pat<(v16i8 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5430 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5433 //===---------------------------------------------------------------------===//
5434 // SSSE3 - Thread synchronization
5435 //===---------------------------------------------------------------------===//
5437 let SchedRW = [WriteSystem] in {
5438 let usesCustomInserter = 1 in {
5439 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5440 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>,
5441 Requires<[HasSSE3]>;
5444 let Uses = [EAX, ECX, EDX] in
5445 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", [], IIC_SSE_MONITOR>,
5446 TB, Requires<[HasSSE3]>;
5447 let Uses = [ECX, EAX] in
5448 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait",
5449 [(int_x86_sse3_mwait ECX, EAX)], IIC_SSE_MWAIT>,
5450 TB, Requires<[HasSSE3]>;
5453 def : InstAlias<"mwait\t{%eax, %ecx|ecx, eax}", (MWAITrr)>, Requires<[In32BitMode]>;
5454 def : InstAlias<"mwait\t{%rax, %rcx|rcx, rax}", (MWAITrr)>, Requires<[In64BitMode]>;
5456 def : InstAlias<"monitor\t{%eax, %ecx, %edx|edx, ecx, eax}", (MONITORrrr)>,
5457 Requires<[In32BitMode]>;
5458 def : InstAlias<"monitor\t{%rax, %rcx, %rdx|rdx, rcx, rax}", (MONITORrrr)>,
5459 Requires<[In64BitMode]>;
5461 //===----------------------------------------------------------------------===//
5462 // SSE4.1 - Packed Move with Sign/Zero Extend
5463 //===----------------------------------------------------------------------===//
5465 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId,
5466 OpndItins itins = DEFAULT_ITINS> {
5467 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5468 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5469 [(set VR128:$dst, (IntId VR128:$src))], itins.rr>, OpSize;
5471 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5472 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5474 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))],
5478 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5480 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5481 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5482 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5484 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5485 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5486 [(set VR256:$dst, (IntId (load addr:$src)))]>,
5490 let Predicates = [HasAVX] in {
5491 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw",
5492 int_x86_sse41_pmovsxbw>, VEX;
5493 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd",
5494 int_x86_sse41_pmovsxwd>, VEX;
5495 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq",
5496 int_x86_sse41_pmovsxdq>, VEX;
5497 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw",
5498 int_x86_sse41_pmovzxbw>, VEX;
5499 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd",
5500 int_x86_sse41_pmovzxwd>, VEX;
5501 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq",
5502 int_x86_sse41_pmovzxdq>, VEX;
5505 let Predicates = [HasAVX2] in {
5506 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5507 int_x86_avx2_pmovsxbw>, VEX, VEX_L;
5508 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5509 int_x86_avx2_pmovsxwd>, VEX, VEX_L;
5510 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5511 int_x86_avx2_pmovsxdq>, VEX, VEX_L;
5512 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5513 int_x86_avx2_pmovzxbw>, VEX, VEX_L;
5514 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5515 int_x86_avx2_pmovzxwd>, VEX, VEX_L;
5516 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5517 int_x86_avx2_pmovzxdq>, VEX, VEX_L;
5520 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw, SSE_INTALU_ITINS_P>;
5521 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd, SSE_INTALU_ITINS_P>;
5522 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq, SSE_INTALU_ITINS_P>;
5523 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw, SSE_INTALU_ITINS_P>;
5524 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd, SSE_INTALU_ITINS_P>;
5525 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq, SSE_INTALU_ITINS_P>;
5527 let Predicates = [HasAVX] in {
5528 // Common patterns involving scalar load.
5529 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5530 (VPMOVSXBWrm addr:$src)>;
5531 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5532 (VPMOVSXBWrm addr:$src)>;
5533 def : Pat<(int_x86_sse41_pmovsxbw (bc_v16i8 (loadv2i64 addr:$src))),
5534 (VPMOVSXBWrm addr:$src)>;
5536 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5537 (VPMOVSXWDrm addr:$src)>;
5538 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5539 (VPMOVSXWDrm addr:$src)>;
5540 def : Pat<(int_x86_sse41_pmovsxwd (bc_v8i16 (loadv2i64 addr:$src))),
5541 (VPMOVSXWDrm addr:$src)>;
5543 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5544 (VPMOVSXDQrm addr:$src)>;
5545 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5546 (VPMOVSXDQrm addr:$src)>;
5547 def : Pat<(int_x86_sse41_pmovsxdq (bc_v4i32 (loadv2i64 addr:$src))),
5548 (VPMOVSXDQrm addr:$src)>;
5550 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5551 (VPMOVZXBWrm addr:$src)>;
5552 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5553 (VPMOVZXBWrm addr:$src)>;
5554 def : Pat<(int_x86_sse41_pmovzxbw (bc_v16i8 (loadv2i64 addr:$src))),
5555 (VPMOVZXBWrm addr:$src)>;
5557 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5558 (VPMOVZXWDrm addr:$src)>;
5559 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5560 (VPMOVZXWDrm addr:$src)>;
5561 def : Pat<(int_x86_sse41_pmovzxwd (bc_v8i16 (loadv2i64 addr:$src))),
5562 (VPMOVZXWDrm addr:$src)>;
5564 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5565 (VPMOVZXDQrm addr:$src)>;
5566 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5567 (VPMOVZXDQrm addr:$src)>;
5568 def : Pat<(int_x86_sse41_pmovzxdq (bc_v4i32 (loadv2i64 addr:$src))),
5569 (VPMOVZXDQrm addr:$src)>;
5572 let Predicates = [UseSSE41] in {
5573 // Common patterns involving scalar load.
5574 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5575 (PMOVSXBWrm addr:$src)>;
5576 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5577 (PMOVSXBWrm addr:$src)>;
5578 def : Pat<(int_x86_sse41_pmovsxbw (bc_v16i8 (loadv2i64 addr:$src))),
5579 (PMOVSXBWrm addr:$src)>;
5581 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5582 (PMOVSXWDrm addr:$src)>;
5583 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5584 (PMOVSXWDrm addr:$src)>;
5585 def : Pat<(int_x86_sse41_pmovsxwd (bc_v8i16 (loadv2i64 addr:$src))),
5586 (PMOVSXWDrm addr:$src)>;
5588 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5589 (PMOVSXDQrm addr:$src)>;
5590 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5591 (PMOVSXDQrm addr:$src)>;
5592 def : Pat<(int_x86_sse41_pmovsxdq (bc_v4i32 (loadv2i64 addr:$src))),
5593 (PMOVSXDQrm addr:$src)>;
5595 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5596 (PMOVZXBWrm addr:$src)>;
5597 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5598 (PMOVZXBWrm addr:$src)>;
5599 def : Pat<(int_x86_sse41_pmovzxbw (bc_v16i8 (loadv2i64 addr:$src))),
5600 (PMOVZXBWrm addr:$src)>;
5602 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5603 (PMOVZXWDrm addr:$src)>;
5604 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5605 (PMOVZXWDrm addr:$src)>;
5606 def : Pat<(int_x86_sse41_pmovzxwd (bc_v8i16 (loadv2i64 addr:$src))),
5607 (PMOVZXWDrm addr:$src)>;
5609 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5610 (PMOVZXDQrm addr:$src)>;
5611 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5612 (PMOVZXDQrm addr:$src)>;
5613 def : Pat<(int_x86_sse41_pmovzxdq (bc_v4i32 (loadv2i64 addr:$src))),
5614 (PMOVZXDQrm addr:$src)>;
5617 let Predicates = [HasAVX2] in {
5618 let AddedComplexity = 15 in {
5619 def : Pat<(v4i64 (X86vzmovly (v4i32 VR128:$src))),
5620 (VPMOVZXDQYrr VR128:$src)>;
5621 def : Pat<(v8i32 (X86vzmovly (v8i16 VR128:$src))),
5622 (VPMOVZXWDYrr VR128:$src)>;
5625 def : Pat<(v4i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>;
5626 def : Pat<(v8i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>;
5629 let Predicates = [HasAVX] in {
5630 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
5631 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
5634 let Predicates = [UseSSE41] in {
5635 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
5636 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
5640 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId,
5641 OpndItins itins = DEFAULT_ITINS> {
5642 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5643 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5644 [(set VR128:$dst, (IntId VR128:$src))], itins.rr>, OpSize;
5646 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5647 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5649 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))],
5654 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5656 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5657 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5658 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5660 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5661 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5663 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5667 let Predicates = [HasAVX] in {
5668 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5670 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5672 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5674 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5678 let Predicates = [HasAVX2] in {
5679 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5680 int_x86_avx2_pmovsxbd>, VEX, VEX_L;
5681 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5682 int_x86_avx2_pmovsxwq>, VEX, VEX_L;
5683 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5684 int_x86_avx2_pmovzxbd>, VEX, VEX_L;
5685 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5686 int_x86_avx2_pmovzxwq>, VEX, VEX_L;
5689 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd,
5690 SSE_INTALU_ITINS_P>;
5691 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq,
5692 SSE_INTALU_ITINS_P>;
5693 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd,
5694 SSE_INTALU_ITINS_P>;
5695 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq,
5696 SSE_INTALU_ITINS_P>;
5698 let Predicates = [HasAVX] in {
5699 // Common patterns involving scalar load
5700 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5701 (VPMOVSXBDrm addr:$src)>;
5702 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5703 (VPMOVSXWQrm addr:$src)>;
5705 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5706 (VPMOVZXBDrm addr:$src)>;
5707 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5708 (VPMOVZXWQrm addr:$src)>;
5711 let Predicates = [UseSSE41] in {
5712 // Common patterns involving scalar load
5713 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5714 (PMOVSXBDrm addr:$src)>;
5715 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5716 (PMOVSXWQrm addr:$src)>;
5718 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5719 (PMOVZXBDrm addr:$src)>;
5720 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5721 (PMOVZXWQrm addr:$src)>;
5724 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId,
5725 OpndItins itins = DEFAULT_ITINS> {
5726 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5727 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5728 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5730 // Expecting a i16 load any extended to i32 value.
5731 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5732 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5733 [(set VR128:$dst, (IntId (bitconvert
5734 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5738 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5740 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5741 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5742 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5744 // Expecting a i16 load any extended to i32 value.
5745 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5746 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5747 [(set VR256:$dst, (IntId (bitconvert
5748 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5752 let Predicates = [HasAVX] in {
5753 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5755 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5758 let Predicates = [HasAVX2] in {
5759 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5760 int_x86_avx2_pmovsxbq>, VEX, VEX_L;
5761 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5762 int_x86_avx2_pmovzxbq>, VEX, VEX_L;
5764 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq,
5765 SSE_INTALU_ITINS_P>;
5766 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq,
5767 SSE_INTALU_ITINS_P>;
5769 let Predicates = [HasAVX2] in {
5770 def : Pat<(v16i16 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBWYrr VR128:$src)>;
5771 def : Pat<(v8i32 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBDYrr VR128:$src)>;
5772 def : Pat<(v4i64 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBQYrr VR128:$src)>;
5774 def : Pat<(v8i32 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>;
5775 def : Pat<(v4i64 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWQYrr VR128:$src)>;
5777 def : Pat<(v4i64 (X86vsext (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>;
5779 def : Pat<(v16i16 (X86vsext (v32i8 VR256:$src))),
5780 (VPMOVSXBWYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5781 def : Pat<(v8i32 (X86vsext (v32i8 VR256:$src))),
5782 (VPMOVSXBDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5783 def : Pat<(v4i64 (X86vsext (v32i8 VR256:$src))),
5784 (VPMOVSXBQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5786 def : Pat<(v8i32 (X86vsext (v16i16 VR256:$src))),
5787 (VPMOVSXWDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5788 def : Pat<(v4i64 (X86vsext (v16i16 VR256:$src))),
5789 (VPMOVSXWQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5791 def : Pat<(v4i64 (X86vsext (v8i32 VR256:$src))),
5792 (VPMOVSXDQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5794 def : Pat<(v8i32 (X86vsmovl (v8i16 (bitconvert (v2i64 (load addr:$src)))))),
5795 (VPMOVSXWDYrm addr:$src)>;
5796 def : Pat<(v4i64 (X86vsmovl (v4i32 (bitconvert (v2i64 (load addr:$src)))))),
5797 (VPMOVSXDQYrm addr:$src)>;
5799 def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2i64
5800 (scalar_to_vector (loadi64 addr:$src))))))),
5801 (VPMOVSXBDYrm addr:$src)>;
5802 def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2f64
5803 (scalar_to_vector (loadf64 addr:$src))))))),
5804 (VPMOVSXBDYrm addr:$src)>;
5806 def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2i64
5807 (scalar_to_vector (loadi64 addr:$src))))))),
5808 (VPMOVSXWQYrm addr:$src)>;
5809 def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2f64
5810 (scalar_to_vector (loadf64 addr:$src))))))),
5811 (VPMOVSXWQYrm addr:$src)>;
5813 def : Pat<(v4i64 (X86vsext (v16i8 (bitconvert (v4i32
5814 (scalar_to_vector (loadi32 addr:$src))))))),
5815 (VPMOVSXBQYrm addr:$src)>;
5818 let Predicates = [HasAVX] in {
5819 // Common patterns involving scalar load
5820 def : Pat<(int_x86_sse41_pmovsxbq
5821 (bitconvert (v4i32 (X86vzmovl
5822 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5823 (VPMOVSXBQrm addr:$src)>;
5825 def : Pat<(int_x86_sse41_pmovzxbq
5826 (bitconvert (v4i32 (X86vzmovl
5827 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5828 (VPMOVZXBQrm addr:$src)>;
5831 let Predicates = [UseSSE41] in {
5832 def : Pat<(v8i16 (X86vsext (v16i8 VR128:$src))), (PMOVSXBWrr VR128:$src)>;
5833 def : Pat<(v4i32 (X86vsext (v16i8 VR128:$src))), (PMOVSXBDrr VR128:$src)>;
5834 def : Pat<(v2i64 (X86vsext (v16i8 VR128:$src))), (PMOVSXBQrr VR128:$src)>;
5836 def : Pat<(v4i32 (X86vsext (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
5837 def : Pat<(v2i64 (X86vsext (v8i16 VR128:$src))), (PMOVSXWQrr VR128:$src)>;
5839 def : Pat<(v2i64 (X86vsext (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
5841 // Common patterns involving scalar load
5842 def : Pat<(int_x86_sse41_pmovsxbq
5843 (bitconvert (v4i32 (X86vzmovl
5844 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5845 (PMOVSXBQrm addr:$src)>;
5847 def : Pat<(int_x86_sse41_pmovzxbq
5848 (bitconvert (v4i32 (X86vzmovl
5849 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5850 (PMOVZXBQrm addr:$src)>;
5852 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64
5853 (scalar_to_vector (loadi64 addr:$src))))))),
5854 (PMOVSXWDrm addr:$src)>;
5855 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64
5856 (scalar_to_vector (loadf64 addr:$src))))))),
5857 (PMOVSXWDrm addr:$src)>;
5858 def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32
5859 (scalar_to_vector (loadi32 addr:$src))))))),
5860 (PMOVSXBDrm addr:$src)>;
5861 def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32
5862 (scalar_to_vector (loadi32 addr:$src))))))),
5863 (PMOVSXWQrm addr:$src)>;
5864 def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32
5865 (scalar_to_vector (extloadi32i16 addr:$src))))))),
5866 (PMOVSXBQrm addr:$src)>;
5867 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64
5868 (scalar_to_vector (loadi64 addr:$src))))))),
5869 (PMOVSXDQrm addr:$src)>;
5870 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64
5871 (scalar_to_vector (loadf64 addr:$src))))))),
5872 (PMOVSXDQrm addr:$src)>;
5873 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64
5874 (scalar_to_vector (loadi64 addr:$src))))))),
5875 (PMOVSXBWrm addr:$src)>;
5876 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64
5877 (scalar_to_vector (loadf64 addr:$src))))))),
5878 (PMOVSXBWrm addr:$src)>;
5881 let Predicates = [HasAVX2] in {
5882 def : Pat<(v16i16 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBWYrr VR128:$src)>;
5883 def : Pat<(v8i32 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBDYrr VR128:$src)>;
5884 def : Pat<(v4i64 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBQYrr VR128:$src)>;
5886 def : Pat<(v8i32 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWDYrr VR128:$src)>;
5887 def : Pat<(v4i64 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWQYrr VR128:$src)>;
5889 def : Pat<(v4i64 (X86vzext (v4i32 VR128:$src))), (VPMOVZXDQYrr VR128:$src)>;
5891 def : Pat<(v16i16 (X86vzext (v32i8 VR256:$src))),
5892 (VPMOVZXBWYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5893 def : Pat<(v8i32 (X86vzext (v32i8 VR256:$src))),
5894 (VPMOVZXBDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5895 def : Pat<(v4i64 (X86vzext (v32i8 VR256:$src))),
5896 (VPMOVZXBQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5898 def : Pat<(v8i32 (X86vzext (v16i16 VR256:$src))),
5899 (VPMOVZXWDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5900 def : Pat<(v4i64 (X86vzext (v16i16 VR256:$src))),
5901 (VPMOVZXWQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5903 def : Pat<(v4i64 (X86vzext (v8i32 VR256:$src))),
5904 (VPMOVZXDQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5907 let Predicates = [HasAVX] in {
5908 def : Pat<(v8i16 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBWrr VR128:$src)>;
5909 def : Pat<(v4i32 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBDrr VR128:$src)>;
5910 def : Pat<(v2i64 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBQrr VR128:$src)>;
5912 def : Pat<(v4i32 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWDrr VR128:$src)>;
5913 def : Pat<(v2i64 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWQrr VR128:$src)>;
5915 def : Pat<(v2i64 (X86vzext (v4i32 VR128:$src))), (VPMOVZXDQrr VR128:$src)>;
5917 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5918 (VPMOVZXBWrm addr:$src)>;
5919 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5920 (VPMOVZXBWrm addr:$src)>;
5921 def : Pat<(v4i32 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5922 (VPMOVZXBDrm addr:$src)>;
5923 def : Pat<(v2i64 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))),
5924 (VPMOVZXBQrm addr:$src)>;
5926 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5927 (VPMOVZXWDrm addr:$src)>;
5928 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5929 (VPMOVZXWDrm addr:$src)>;
5930 def : Pat<(v2i64 (X86vzext (v8i16 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5931 (VPMOVZXWQrm addr:$src)>;
5933 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5934 (VPMOVZXDQrm addr:$src)>;
5935 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5936 (VPMOVZXDQrm addr:$src)>;
5937 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (X86vzload addr:$src)))))),
5938 (VPMOVZXDQrm addr:$src)>;
5940 def : Pat<(v8i16 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBWrr VR128:$src)>;
5941 def : Pat<(v4i32 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBDrr VR128:$src)>;
5942 def : Pat<(v2i64 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBQrr VR128:$src)>;
5944 def : Pat<(v4i32 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
5945 def : Pat<(v2i64 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWQrr VR128:$src)>;
5947 def : Pat<(v2i64 (X86vsext (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
5949 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64
5950 (scalar_to_vector (loadi64 addr:$src))))))),
5951 (VPMOVSXWDrm addr:$src)>;
5952 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64
5953 (scalar_to_vector (loadi64 addr:$src))))))),
5954 (VPMOVSXDQrm addr:$src)>;
5955 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64
5956 (scalar_to_vector (loadf64 addr:$src))))))),
5957 (VPMOVSXWDrm addr:$src)>;
5958 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64
5959 (scalar_to_vector (loadf64 addr:$src))))))),
5960 (VPMOVSXDQrm addr:$src)>;
5961 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64
5962 (scalar_to_vector (loadi64 addr:$src))))))),
5963 (VPMOVSXBWrm addr:$src)>;
5964 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64
5965 (scalar_to_vector (loadf64 addr:$src))))))),
5966 (VPMOVSXBWrm addr:$src)>;
5968 def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32
5969 (scalar_to_vector (loadi32 addr:$src))))))),
5970 (VPMOVSXBDrm addr:$src)>;
5971 def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32
5972 (scalar_to_vector (loadi32 addr:$src))))))),
5973 (VPMOVSXWQrm addr:$src)>;
5974 def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32
5975 (scalar_to_vector (extloadi32i16 addr:$src))))))),
5976 (VPMOVSXBQrm addr:$src)>;
5979 let Predicates = [UseSSE41] in {
5980 def : Pat<(v8i16 (X86vzext (v16i8 VR128:$src))), (PMOVZXBWrr VR128:$src)>;
5981 def : Pat<(v4i32 (X86vzext (v16i8 VR128:$src))), (PMOVZXBDrr VR128:$src)>;
5982 def : Pat<(v2i64 (X86vzext (v16i8 VR128:$src))), (PMOVZXBQrr VR128:$src)>;
5984 def : Pat<(v4i32 (X86vzext (v8i16 VR128:$src))), (PMOVZXWDrr VR128:$src)>;
5985 def : Pat<(v2i64 (X86vzext (v8i16 VR128:$src))), (PMOVZXWQrr VR128:$src)>;
5987 def : Pat<(v2i64 (X86vzext (v4i32 VR128:$src))), (PMOVZXDQrr VR128:$src)>;
5989 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5990 (PMOVZXBWrm addr:$src)>;
5991 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5992 (PMOVZXBWrm addr:$src)>;
5993 def : Pat<(v4i32 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5994 (PMOVZXBDrm addr:$src)>;
5995 def : Pat<(v2i64 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))),
5996 (PMOVZXBQrm addr:$src)>;
5998 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5999 (PMOVZXWDrm addr:$src)>;
6000 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
6001 (PMOVZXWDrm addr:$src)>;
6002 def : Pat<(v2i64 (X86vzext (v8i16 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
6003 (PMOVZXWQrm addr:$src)>;
6005 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
6006 (PMOVZXDQrm addr:$src)>;
6007 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
6008 (PMOVZXDQrm addr:$src)>;
6009 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (X86vzload addr:$src)))))),
6010 (PMOVZXDQrm addr:$src)>;
6013 //===----------------------------------------------------------------------===//
6014 // SSE4.1 - Extract Instructions
6015 //===----------------------------------------------------------------------===//
6017 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
6018 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
6019 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
6020 (ins VR128:$src1, i32i8imm:$src2),
6021 !strconcat(OpcodeStr,
6022 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6023 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
6025 let neverHasSideEffects = 1, mayStore = 1 in
6026 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6027 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
6028 !strconcat(OpcodeStr,
6029 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6032 // There's an AssertZext in the way of writing the store pattern
6033 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
6036 let Predicates = [HasAVX] in {
6037 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
6038 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
6039 (ins VR128:$src1, i32i8imm:$src2),
6040 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
6043 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
6046 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
6047 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
6048 let neverHasSideEffects = 1, mayStore = 1 in
6049 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6050 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
6051 !strconcat(OpcodeStr,
6052 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6055 // There's an AssertZext in the way of writing the store pattern
6056 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
6059 let Predicates = [HasAVX] in
6060 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
6062 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
6065 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
6066 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
6067 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
6068 (ins VR128:$src1, i32i8imm:$src2),
6069 !strconcat(OpcodeStr,
6070 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6072 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
6073 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6074 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
6075 !strconcat(OpcodeStr,
6076 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6077 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
6078 addr:$dst)]>, OpSize;
6081 let Predicates = [HasAVX] in
6082 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
6084 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
6086 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
6087 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
6088 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
6089 (ins VR128:$src1, i32i8imm:$src2),
6090 !strconcat(OpcodeStr,
6091 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6093 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
6094 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6095 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
6096 !strconcat(OpcodeStr,
6097 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6098 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
6099 addr:$dst)]>, OpSize, REX_W;
6102 let Predicates = [HasAVX] in
6103 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
6105 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
6107 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
6109 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr,
6110 OpndItins itins = DEFAULT_ITINS> {
6111 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
6112 (ins VR128:$src1, i32i8imm:$src2),
6113 !strconcat(OpcodeStr,
6114 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6116 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))],
6119 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6120 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
6121 !strconcat(OpcodeStr,
6122 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6123 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
6124 addr:$dst)], itins.rm>, OpSize;
6127 let ExeDomain = SSEPackedSingle in {
6128 let Predicates = [UseAVX] in {
6129 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
6130 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
6131 (ins VR128:$src1, i32i8imm:$src2),
6132 "vextractps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6135 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps", SSE_EXTRACT_ITINS>;
6138 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
6139 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
6142 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
6144 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
6147 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
6148 Requires<[UseSSE41]>;
6150 //===----------------------------------------------------------------------===//
6151 // SSE4.1 - Insert Instructions
6152 //===----------------------------------------------------------------------===//
6154 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
6155 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6156 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
6158 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6160 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6162 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
6163 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6164 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
6166 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6168 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6170 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
6171 imm:$src3))]>, OpSize;
6174 let Predicates = [HasAVX] in
6175 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
6176 let Constraints = "$src1 = $dst" in
6177 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
6179 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
6180 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6181 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
6183 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6185 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6187 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
6189 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6190 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
6192 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6194 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6196 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
6197 imm:$src3)))]>, OpSize;
6200 let Predicates = [HasAVX] in
6201 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
6202 let Constraints = "$src1 = $dst" in
6203 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
6205 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
6206 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6207 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
6209 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6211 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6213 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
6215 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6216 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
6218 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6220 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6222 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
6223 imm:$src3)))]>, OpSize;
6226 let Predicates = [HasAVX] in
6227 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
6228 let Constraints = "$src1 = $dst" in
6229 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
6231 // insertps has a few different modes, there's the first two here below which
6232 // are optimized inserts that won't zero arbitrary elements in the destination
6233 // vector. The next one matches the intrinsic and could zero arbitrary elements
6234 // in the target vector.
6235 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1,
6236 OpndItins itins = DEFAULT_ITINS> {
6237 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6238 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
6240 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6242 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6244 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))], itins.rr>,
6246 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6247 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
6249 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6251 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6253 (X86insrtps VR128:$src1,
6254 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
6255 imm:$src3))], itins.rm>, OpSize;
6258 let ExeDomain = SSEPackedSingle in {
6259 let Predicates = [UseAVX] in
6260 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
6261 let Constraints = "$src1 = $dst" in
6262 defm INSERTPS : SS41I_insertf32<0x21, "insertps", 1, SSE_INSERT_ITINS>;
6265 //===----------------------------------------------------------------------===//
6266 // SSE4.1 - Round Instructions
6267 //===----------------------------------------------------------------------===//
6269 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
6270 X86MemOperand x86memop, RegisterClass RC,
6271 PatFrag mem_frag32, PatFrag mem_frag64,
6272 Intrinsic V4F32Int, Intrinsic V2F64Int> {
6273 let ExeDomain = SSEPackedSingle in {
6274 // Intrinsic operation, reg.
6275 // Vector intrinsic operation, reg
6276 def PSr : SS4AIi8<opcps, MRMSrcReg,
6277 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6278 !strconcat(OpcodeStr,
6279 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6280 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))],
6281 IIC_SSE_ROUNDPS_REG>,
6284 // Vector intrinsic operation, mem
6285 def PSm : SS4AIi8<opcps, MRMSrcMem,
6286 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6287 !strconcat(OpcodeStr,
6288 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6290 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))],
6291 IIC_SSE_ROUNDPS_MEM>,
6293 } // ExeDomain = SSEPackedSingle
6295 let ExeDomain = SSEPackedDouble in {
6296 // Vector intrinsic operation, reg
6297 def PDr : SS4AIi8<opcpd, MRMSrcReg,
6298 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6299 !strconcat(OpcodeStr,
6300 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6301 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))],
6302 IIC_SSE_ROUNDPS_REG>,
6305 // Vector intrinsic operation, mem
6306 def PDm : SS4AIi8<opcpd, MRMSrcMem,
6307 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6308 !strconcat(OpcodeStr,
6309 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6311 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))],
6312 IIC_SSE_ROUNDPS_REG>,
6314 } // ExeDomain = SSEPackedDouble
6317 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
6320 Intrinsic F64Int, bit Is2Addr = 1> {
6321 let ExeDomain = GenericDomain in {
6323 let hasSideEffects = 0 in
6324 def SSr : SS4AIi8<opcss, MRMSrcReg,
6325 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3),
6327 !strconcat(OpcodeStr,
6328 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6329 !strconcat(OpcodeStr,
6330 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6333 // Intrinsic operation, reg.
6334 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
6335 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6337 !strconcat(OpcodeStr,
6338 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6339 !strconcat(OpcodeStr,
6340 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6341 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6344 // Intrinsic operation, mem.
6345 def SSm : SS4AIi8<opcss, MRMSrcMem,
6346 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
6348 !strconcat(OpcodeStr,
6349 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6350 !strconcat(OpcodeStr,
6351 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6353 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
6357 let hasSideEffects = 0 in
6358 def SDr : SS4AIi8<opcsd, MRMSrcReg,
6359 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3),
6361 !strconcat(OpcodeStr,
6362 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6363 !strconcat(OpcodeStr,
6364 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6367 // Intrinsic operation, reg.
6368 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
6369 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6371 !strconcat(OpcodeStr,
6372 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6373 !strconcat(OpcodeStr,
6374 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6375 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6378 // Intrinsic operation, mem.
6379 def SDm : SS4AIi8<opcsd, MRMSrcMem,
6380 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
6382 !strconcat(OpcodeStr,
6383 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6384 !strconcat(OpcodeStr,
6385 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6387 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
6389 } // ExeDomain = GenericDomain
6392 // FP round - roundss, roundps, roundsd, roundpd
6393 let Predicates = [HasAVX] in {
6395 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
6396 memopv4f32, memopv2f64,
6397 int_x86_sse41_round_ps,
6398 int_x86_sse41_round_pd>, VEX;
6399 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
6400 memopv8f32, memopv4f64,
6401 int_x86_avx_round_ps_256,
6402 int_x86_avx_round_pd_256>, VEX, VEX_L;
6403 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
6404 int_x86_sse41_round_ss,
6405 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
6407 def : Pat<(ffloor FR32:$src),
6408 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6409 def : Pat<(f64 (ffloor FR64:$src)),
6410 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6411 def : Pat<(f32 (fnearbyint FR32:$src)),
6412 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6413 def : Pat<(f64 (fnearbyint FR64:$src)),
6414 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6415 def : Pat<(f32 (fceil FR32:$src)),
6416 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6417 def : Pat<(f64 (fceil FR64:$src)),
6418 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6419 def : Pat<(f32 (frint FR32:$src)),
6420 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6421 def : Pat<(f64 (frint FR64:$src)),
6422 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6423 def : Pat<(f32 (ftrunc FR32:$src)),
6424 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6425 def : Pat<(f64 (ftrunc FR64:$src)),
6426 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6428 def : Pat<(v4f32 (ffloor VR128:$src)),
6429 (VROUNDPSr VR128:$src, (i32 0x1))>;
6430 def : Pat<(v4f32 (fnearbyint VR128:$src)),
6431 (VROUNDPSr VR128:$src, (i32 0xC))>;
6432 def : Pat<(v4f32 (fceil VR128:$src)),
6433 (VROUNDPSr VR128:$src, (i32 0x2))>;
6434 def : Pat<(v4f32 (frint VR128:$src)),
6435 (VROUNDPSr VR128:$src, (i32 0x4))>;
6436 def : Pat<(v4f32 (ftrunc VR128:$src)),
6437 (VROUNDPSr VR128:$src, (i32 0x3))>;
6439 def : Pat<(v2f64 (ffloor VR128:$src)),
6440 (VROUNDPDr VR128:$src, (i32 0x1))>;
6441 def : Pat<(v2f64 (fnearbyint VR128:$src)),
6442 (VROUNDPDr VR128:$src, (i32 0xC))>;
6443 def : Pat<(v2f64 (fceil VR128:$src)),
6444 (VROUNDPDr VR128:$src, (i32 0x2))>;
6445 def : Pat<(v2f64 (frint VR128:$src)),
6446 (VROUNDPDr VR128:$src, (i32 0x4))>;
6447 def : Pat<(v2f64 (ftrunc VR128:$src)),
6448 (VROUNDPDr VR128:$src, (i32 0x3))>;
6450 def : Pat<(v8f32 (ffloor VR256:$src)),
6451 (VROUNDYPSr VR256:$src, (i32 0x1))>;
6452 def : Pat<(v8f32 (fnearbyint VR256:$src)),
6453 (VROUNDYPSr VR256:$src, (i32 0xC))>;
6454 def : Pat<(v8f32 (fceil VR256:$src)),
6455 (VROUNDYPSr VR256:$src, (i32 0x2))>;
6456 def : Pat<(v8f32 (frint VR256:$src)),
6457 (VROUNDYPSr VR256:$src, (i32 0x4))>;
6458 def : Pat<(v8f32 (ftrunc VR256:$src)),
6459 (VROUNDYPSr VR256:$src, (i32 0x3))>;
6461 def : Pat<(v4f64 (ffloor VR256:$src)),
6462 (VROUNDYPDr VR256:$src, (i32 0x1))>;
6463 def : Pat<(v4f64 (fnearbyint VR256:$src)),
6464 (VROUNDYPDr VR256:$src, (i32 0xC))>;
6465 def : Pat<(v4f64 (fceil VR256:$src)),
6466 (VROUNDYPDr VR256:$src, (i32 0x2))>;
6467 def : Pat<(v4f64 (frint VR256:$src)),
6468 (VROUNDYPDr VR256:$src, (i32 0x4))>;
6469 def : Pat<(v4f64 (ftrunc VR256:$src)),
6470 (VROUNDYPDr VR256:$src, (i32 0x3))>;
6473 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
6474 memopv4f32, memopv2f64,
6475 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
6476 let Constraints = "$src1 = $dst" in
6477 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
6478 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
6480 let Predicates = [UseSSE41] in {
6481 def : Pat<(ffloor FR32:$src),
6482 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6483 def : Pat<(f64 (ffloor FR64:$src)),
6484 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6485 def : Pat<(f32 (fnearbyint FR32:$src)),
6486 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6487 def : Pat<(f64 (fnearbyint FR64:$src)),
6488 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6489 def : Pat<(f32 (fceil FR32:$src)),
6490 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6491 def : Pat<(f64 (fceil FR64:$src)),
6492 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6493 def : Pat<(f32 (frint FR32:$src)),
6494 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6495 def : Pat<(f64 (frint FR64:$src)),
6496 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6497 def : Pat<(f32 (ftrunc FR32:$src)),
6498 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6499 def : Pat<(f64 (ftrunc FR64:$src)),
6500 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6502 def : Pat<(v4f32 (ffloor VR128:$src)),
6503 (ROUNDPSr VR128:$src, (i32 0x1))>;
6504 def : Pat<(v4f32 (fnearbyint VR128:$src)),
6505 (ROUNDPSr VR128:$src, (i32 0xC))>;
6506 def : Pat<(v4f32 (fceil VR128:$src)),
6507 (ROUNDPSr VR128:$src, (i32 0x2))>;
6508 def : Pat<(v4f32 (frint VR128:$src)),
6509 (ROUNDPSr VR128:$src, (i32 0x4))>;
6510 def : Pat<(v4f32 (ftrunc VR128:$src)),
6511 (ROUNDPSr VR128:$src, (i32 0x3))>;
6513 def : Pat<(v2f64 (ffloor VR128:$src)),
6514 (ROUNDPDr VR128:$src, (i32 0x1))>;
6515 def : Pat<(v2f64 (fnearbyint VR128:$src)),
6516 (ROUNDPDr VR128:$src, (i32 0xC))>;
6517 def : Pat<(v2f64 (fceil VR128:$src)),
6518 (ROUNDPDr VR128:$src, (i32 0x2))>;
6519 def : Pat<(v2f64 (frint VR128:$src)),
6520 (ROUNDPDr VR128:$src, (i32 0x4))>;
6521 def : Pat<(v2f64 (ftrunc VR128:$src)),
6522 (ROUNDPDr VR128:$src, (i32 0x3))>;
6525 //===----------------------------------------------------------------------===//
6526 // SSE4.1 - Packed Bit Test
6527 //===----------------------------------------------------------------------===//
6529 // ptest instruction we'll lower to this in X86ISelLowering primarily from
6530 // the intel intrinsic that corresponds to this.
6531 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6532 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6533 "vptest\t{$src2, $src1|$src1, $src2}",
6534 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6536 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6537 "vptest\t{$src2, $src1|$src1, $src2}",
6538 [(set EFLAGS,(X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6541 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
6542 "vptest\t{$src2, $src1|$src1, $src2}",
6543 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
6545 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
6546 "vptest\t{$src2, $src1|$src1, $src2}",
6547 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
6551 let Defs = [EFLAGS] in {
6552 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6553 "ptest\t{$src2, $src1|$src1, $src2}",
6554 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6556 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6557 "ptest\t{$src2, $src1|$src1, $src2}",
6558 [(set EFLAGS, (X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6562 // The bit test instructions below are AVX only
6563 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
6564 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
6565 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
6566 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6567 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
6568 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
6569 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6570 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
6574 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6575 let ExeDomain = SSEPackedSingle in {
6576 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
6577 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>,
6580 let ExeDomain = SSEPackedDouble in {
6581 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
6582 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>,
6587 //===----------------------------------------------------------------------===//
6588 // SSE4.1 - Misc Instructions
6589 //===----------------------------------------------------------------------===//
6591 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
6592 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
6593 "popcnt{w}\t{$src, $dst|$dst, $src}",
6594 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)],
6597 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
6598 "popcnt{w}\t{$src, $dst|$dst, $src}",
6599 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
6600 (implicit EFLAGS)], IIC_SSE_POPCNT_RM>, OpSize, XS;
6602 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
6603 "popcnt{l}\t{$src, $dst|$dst, $src}",
6604 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)],
6607 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
6608 "popcnt{l}\t{$src, $dst|$dst, $src}",
6609 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
6610 (implicit EFLAGS)], IIC_SSE_POPCNT_RM>, XS;
6612 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
6613 "popcnt{q}\t{$src, $dst|$dst, $src}",
6614 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)],
6617 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
6618 "popcnt{q}\t{$src, $dst|$dst, $src}",
6619 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
6620 (implicit EFLAGS)], IIC_SSE_POPCNT_RM>, XS;
6625 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
6626 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
6627 Intrinsic IntId128> {
6628 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6630 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6631 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
6632 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6634 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6637 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
6640 let Predicates = [HasAVX] in
6641 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
6642 int_x86_sse41_phminposuw>, VEX;
6643 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
6644 int_x86_sse41_phminposuw>;
6646 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6647 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
6648 Intrinsic IntId128, bit Is2Addr = 1,
6649 OpndItins itins = DEFAULT_ITINS> {
6650 let isCommutable = 1 in
6651 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6652 (ins VR128:$src1, VR128:$src2),
6654 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6655 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6656 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))],
6658 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6659 (ins VR128:$src1, i128mem:$src2),
6661 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6662 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6664 (IntId128 VR128:$src1,
6665 (bitconvert (memopv2i64 addr:$src2))))],
6669 /// SS41I_binop_rm_int_y - Simple SSE 4.1 binary operator
6670 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6671 Intrinsic IntId256> {
6672 let isCommutable = 1 in
6673 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6674 (ins VR256:$src1, VR256:$src2),
6675 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6676 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
6677 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6678 (ins VR256:$src1, i256mem:$src2),
6679 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6681 (IntId256 VR256:$src1,
6682 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
6686 /// SS48I_binop_rm - Simple SSE41 binary operator.
6687 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6688 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6689 X86MemOperand x86memop, bit Is2Addr = 1,
6690 OpndItins itins = DEFAULT_ITINS> {
6691 let isCommutable = 1 in
6692 def rr : SS48I<opc, MRMSrcReg, (outs RC:$dst),
6693 (ins RC:$src1, RC:$src2),
6695 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6696 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6697 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>, OpSize;
6698 def rm : SS48I<opc, MRMSrcMem, (outs RC:$dst),
6699 (ins RC:$src1, x86memop:$src2),
6701 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6702 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6704 (OpVT (OpNode RC:$src1,
6705 (bitconvert (memop_frag addr:$src2)))))]>, OpSize;
6708 let Predicates = [HasAVX] in {
6709 let isCommutable = 0 in
6710 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
6712 defm VPMINSB : SS48I_binop_rm<0x38, "vpminsb", X86smin, v16i8, VR128,
6713 memopv2i64, i128mem, 0>, VEX_4V;
6714 defm VPMINSD : SS48I_binop_rm<0x39, "vpminsd", X86smin, v4i32, VR128,
6715 memopv2i64, i128mem, 0>, VEX_4V;
6716 defm VPMINUD : SS48I_binop_rm<0x3B, "vpminud", X86umin, v4i32, VR128,
6717 memopv2i64, i128mem, 0>, VEX_4V;
6718 defm VPMINUW : SS48I_binop_rm<0x3A, "vpminuw", X86umin, v8i16, VR128,
6719 memopv2i64, i128mem, 0>, VEX_4V;
6720 defm VPMAXSB : SS48I_binop_rm<0x3C, "vpmaxsb", X86smax, v16i8, VR128,
6721 memopv2i64, i128mem, 0>, VEX_4V;
6722 defm VPMAXSD : SS48I_binop_rm<0x3D, "vpmaxsd", X86smax, v4i32, VR128,
6723 memopv2i64, i128mem, 0>, VEX_4V;
6724 defm VPMAXUD : SS48I_binop_rm<0x3F, "vpmaxud", X86umax, v4i32, VR128,
6725 memopv2i64, i128mem, 0>, VEX_4V;
6726 defm VPMAXUW : SS48I_binop_rm<0x3E, "vpmaxuw", X86umax, v8i16, VR128,
6727 memopv2i64, i128mem, 0>, VEX_4V;
6728 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
6732 let Predicates = [HasAVX2] in {
6733 let isCommutable = 0 in
6734 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
6735 int_x86_avx2_packusdw>, VEX_4V, VEX_L;
6736 defm VPMINSBY : SS48I_binop_rm<0x38, "vpminsb", X86smin, v32i8, VR256,
6737 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6738 defm VPMINSDY : SS48I_binop_rm<0x39, "vpminsd", X86smin, v8i32, VR256,
6739 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6740 defm VPMINUDY : SS48I_binop_rm<0x3B, "vpminud", X86umin, v8i32, VR256,
6741 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6742 defm VPMINUWY : SS48I_binop_rm<0x3A, "vpminuw", X86umin, v16i16, VR256,
6743 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6744 defm VPMAXSBY : SS48I_binop_rm<0x3C, "vpmaxsb", X86smax, v32i8, VR256,
6745 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6746 defm VPMAXSDY : SS48I_binop_rm<0x3D, "vpmaxsd", X86smax, v8i32, VR256,
6747 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6748 defm VPMAXUDY : SS48I_binop_rm<0x3F, "vpmaxud", X86umax, v8i32, VR256,
6749 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6750 defm VPMAXUWY : SS48I_binop_rm<0x3E, "vpmaxuw", X86umax, v16i16, VR256,
6751 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6752 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
6753 int_x86_avx2_pmul_dq>, VEX_4V, VEX_L;
6756 let Constraints = "$src1 = $dst" in {
6757 let isCommutable = 0 in
6758 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
6759 defm PMINSB : SS48I_binop_rm<0x38, "pminsb", X86smin, v16i8, VR128,
6760 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6761 defm PMINSD : SS48I_binop_rm<0x39, "pminsd", X86smin, v4i32, VR128,
6762 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6763 defm PMINUD : SS48I_binop_rm<0x3B, "pminud", X86umin, v4i32, VR128,
6764 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6765 defm PMINUW : SS48I_binop_rm<0x3A, "pminuw", X86umin, v8i16, VR128,
6766 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6767 defm PMAXSB : SS48I_binop_rm<0x3C, "pmaxsb", X86smax, v16i8, VR128,
6768 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6769 defm PMAXSD : SS48I_binop_rm<0x3D, "pmaxsd", X86smax, v4i32, VR128,
6770 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6771 defm PMAXUD : SS48I_binop_rm<0x3F, "pmaxud", X86umax, v4i32, VR128,
6772 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6773 defm PMAXUW : SS48I_binop_rm<0x3E, "pmaxuw", X86umax, v8i16, VR128,
6774 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6775 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq,
6776 1, SSE_INTMUL_ITINS_P>;
6779 let Predicates = [HasAVX] in {
6780 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, VR128,
6781 memopv2i64, i128mem, 0>, VEX_4V;
6782 defm VPCMPEQQ : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v2i64, VR128,
6783 memopv2i64, i128mem, 0>, VEX_4V;
6785 let Predicates = [HasAVX2] in {
6786 defm VPMULLDY : SS48I_binop_rm<0x40, "vpmulld", mul, v8i32, VR256,
6787 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6788 defm VPCMPEQQY : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v4i64, VR256,
6789 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6792 let Constraints = "$src1 = $dst" in {
6793 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32, VR128,
6794 memopv2i64, i128mem, 1, SSE_PMULLD_ITINS>;
6795 defm PCMPEQQ : SS48I_binop_rm<0x29, "pcmpeqq", X86pcmpeq, v2i64, VR128,
6796 memopv2i64, i128mem, 1, SSE_INTALUQ_ITINS_P>;
6799 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
6800 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
6801 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6802 X86MemOperand x86memop, bit Is2Addr = 1,
6803 OpndItins itins = DEFAULT_ITINS> {
6804 let isCommutable = 1 in
6805 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
6806 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6808 !strconcat(OpcodeStr,
6809 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6810 !strconcat(OpcodeStr,
6811 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6812 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))], itins.rr>,
6814 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
6815 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6817 !strconcat(OpcodeStr,
6818 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6819 !strconcat(OpcodeStr,
6820 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6823 (bitconvert (memop_frag addr:$src2)), imm:$src3))], itins.rm>,
6827 let Predicates = [HasAVX] in {
6828 let isCommutable = 0 in {
6829 let ExeDomain = SSEPackedSingle in {
6830 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
6831 VR128, memopv4f32, f128mem, 0>, VEX_4V;
6832 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
6833 int_x86_avx_blend_ps_256, VR256, memopv8f32,
6834 f256mem, 0>, VEX_4V, VEX_L;
6836 let ExeDomain = SSEPackedDouble in {
6837 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
6838 VR128, memopv2f64, f128mem, 0>, VEX_4V;
6839 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
6840 int_x86_avx_blend_pd_256,VR256, memopv4f64,
6841 f256mem, 0>, VEX_4V, VEX_L;
6843 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
6844 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6845 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
6846 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6848 let ExeDomain = SSEPackedSingle in
6849 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
6850 VR128, memopv4f32, f128mem, 0>, VEX_4V;
6851 let ExeDomain = SSEPackedDouble in
6852 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
6853 VR128, memopv2f64, f128mem, 0>, VEX_4V;
6854 let ExeDomain = SSEPackedSingle in
6855 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
6856 VR256, memopv8f32, i256mem, 0>, VEX_4V, VEX_L;
6859 let Predicates = [HasAVX2] in {
6860 let isCommutable = 0 in {
6861 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
6862 VR256, memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6863 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
6864 VR256, memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6868 let Constraints = "$src1 = $dst" in {
6869 let isCommutable = 0 in {
6870 let ExeDomain = SSEPackedSingle in
6871 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
6872 VR128, memopv4f32, f128mem,
6873 1, SSE_INTALU_ITINS_P>;
6874 let ExeDomain = SSEPackedDouble in
6875 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
6876 VR128, memopv2f64, f128mem,
6877 1, SSE_INTALU_ITINS_P>;
6878 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
6879 VR128, memopv2i64, i128mem,
6880 1, SSE_INTALU_ITINS_P>;
6881 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
6882 VR128, memopv2i64, i128mem,
6883 1, SSE_INTMUL_ITINS_P>;
6885 let ExeDomain = SSEPackedSingle in
6886 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
6887 VR128, memopv4f32, f128mem, 1,
6889 let ExeDomain = SSEPackedDouble in
6890 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
6891 VR128, memopv2f64, f128mem, 1,
6895 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
6896 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
6897 RegisterClass RC, X86MemOperand x86memop,
6898 PatFrag mem_frag, Intrinsic IntId> {
6899 def rr : Ii8<opc, MRMSrcReg, (outs RC:$dst),
6900 (ins RC:$src1, RC:$src2, RC:$src3),
6901 !strconcat(OpcodeStr,
6902 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6903 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
6904 NoItinerary, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6906 def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
6907 (ins RC:$src1, x86memop:$src2, RC:$src3),
6908 !strconcat(OpcodeStr,
6909 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6911 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
6913 NoItinerary, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6916 let Predicates = [HasAVX] in {
6917 let ExeDomain = SSEPackedDouble in {
6918 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, f128mem,
6919 memopv2f64, int_x86_sse41_blendvpd>;
6920 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, f256mem,
6921 memopv4f64, int_x86_avx_blendv_pd_256>, VEX_L;
6922 } // ExeDomain = SSEPackedDouble
6923 let ExeDomain = SSEPackedSingle in {
6924 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, f128mem,
6925 memopv4f32, int_x86_sse41_blendvps>;
6926 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, f256mem,
6927 memopv8f32, int_x86_avx_blendv_ps_256>, VEX_L;
6928 } // ExeDomain = SSEPackedSingle
6929 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
6930 memopv2i64, int_x86_sse41_pblendvb>;
6933 let Predicates = [HasAVX2] in {
6934 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
6935 memopv4i64, int_x86_avx2_pblendvb>, VEX_L;
6938 let Predicates = [HasAVX] in {
6939 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
6940 (v16i8 VR128:$src2))),
6941 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6942 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
6943 (v4i32 VR128:$src2))),
6944 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6945 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
6946 (v4f32 VR128:$src2))),
6947 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6948 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
6949 (v2i64 VR128:$src2))),
6950 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6951 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
6952 (v2f64 VR128:$src2))),
6953 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6954 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
6955 (v8i32 VR256:$src2))),
6956 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6957 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
6958 (v8f32 VR256:$src2))),
6959 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6960 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
6961 (v4i64 VR256:$src2))),
6962 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6963 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
6964 (v4f64 VR256:$src2))),
6965 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6967 def : Pat<(v8f32 (X86Blendi (v8f32 VR256:$src1), (v8f32 VR256:$src2),
6969 (VBLENDPSYrri VR256:$src1, VR256:$src2, imm:$mask)>;
6970 def : Pat<(v4f64 (X86Blendi (v4f64 VR256:$src1), (v4f64 VR256:$src2),
6972 (VBLENDPDYrri VR256:$src1, VR256:$src2, imm:$mask)>;
6974 def : Pat<(v8i16 (X86Blendi (v8i16 VR128:$src1), (v8i16 VR128:$src2),
6976 (VPBLENDWrri VR128:$src1, VR128:$src2, imm:$mask)>;
6977 def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$src1), (v4f32 VR128:$src2),
6979 (VBLENDPSrri VR128:$src1, VR128:$src2, imm:$mask)>;
6980 def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$src1), (v2f64 VR128:$src2),
6982 (VBLENDPDrri VR128:$src1, VR128:$src2, imm:$mask)>;
6985 let Predicates = [HasAVX2] in {
6986 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
6987 (v32i8 VR256:$src2))),
6988 (VPBLENDVBYrr VR256:$src1, VR256:$src2, VR256:$mask)>;
6989 def : Pat<(v16i16 (X86Blendi (v16i16 VR256:$src1), (v16i16 VR256:$src2),
6991 (VPBLENDWYrri VR256:$src1, VR256:$src2, imm:$mask)>;
6994 /// SS41I_ternary_int - SSE 4.1 ternary operator
6995 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
6996 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
6997 X86MemOperand x86memop, Intrinsic IntId,
6998 OpndItins itins = DEFAULT_ITINS> {
6999 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
7000 (ins VR128:$src1, VR128:$src2),
7001 !strconcat(OpcodeStr,
7002 "\t{$src2, $dst|$dst, $src2}"),
7003 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))],
7006 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
7007 (ins VR128:$src1, x86memop:$src2),
7008 !strconcat(OpcodeStr,
7009 "\t{$src2, $dst|$dst, $src2}"),
7012 (bitconvert (mem_frag addr:$src2)), XMM0))],
7017 let ExeDomain = SSEPackedDouble in
7018 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64, f128mem,
7019 int_x86_sse41_blendvpd>;
7020 let ExeDomain = SSEPackedSingle in
7021 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32, f128mem,
7022 int_x86_sse41_blendvps>;
7023 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64, i128mem,
7024 int_x86_sse41_pblendvb>;
7026 // Aliases with the implicit xmm0 argument
7027 def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7028 (BLENDVPDrr0 VR128:$dst, VR128:$src2)>;
7029 def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7030 (BLENDVPDrm0 VR128:$dst, f128mem:$src2)>;
7031 def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7032 (BLENDVPSrr0 VR128:$dst, VR128:$src2)>;
7033 def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7034 (BLENDVPSrm0 VR128:$dst, f128mem:$src2)>;
7035 def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7036 (PBLENDVBrr0 VR128:$dst, VR128:$src2)>;
7037 def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7038 (PBLENDVBrm0 VR128:$dst, i128mem:$src2)>;
7040 let Predicates = [UseSSE41] in {
7041 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
7042 (v16i8 VR128:$src2))),
7043 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
7044 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
7045 (v4i32 VR128:$src2))),
7046 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
7047 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
7048 (v4f32 VR128:$src2))),
7049 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
7050 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
7051 (v2i64 VR128:$src2))),
7052 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
7053 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
7054 (v2f64 VR128:$src2))),
7055 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
7057 def : Pat<(v8i16 (X86Blendi (v8i16 VR128:$src1), (v8i16 VR128:$src2),
7059 (PBLENDWrri VR128:$src1, VR128:$src2, imm:$mask)>;
7060 def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$src1), (v4f32 VR128:$src2),
7062 (BLENDPSrri VR128:$src1, VR128:$src2, imm:$mask)>;
7063 def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$src1), (v2f64 VR128:$src2),
7065 (BLENDPDrri VR128:$src1, VR128:$src2, imm:$mask)>;
7069 let Predicates = [HasAVX] in
7070 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
7071 "vmovntdqa\t{$src, $dst|$dst, $src}",
7072 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
7074 let Predicates = [HasAVX2] in
7075 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
7076 "vmovntdqa\t{$src, $dst|$dst, $src}",
7077 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
7079 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
7080 "movntdqa\t{$src, $dst|$dst, $src}",
7081 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
7084 //===----------------------------------------------------------------------===//
7085 // SSE4.2 - Compare Instructions
7086 //===----------------------------------------------------------------------===//
7088 /// SS42I_binop_rm - Simple SSE 4.2 binary operator
7089 multiclass SS42I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
7090 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
7091 X86MemOperand x86memop, bit Is2Addr = 1> {
7092 def rr : SS428I<opc, MRMSrcReg, (outs RC:$dst),
7093 (ins RC:$src1, RC:$src2),
7095 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7096 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7097 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>,
7099 def rm : SS428I<opc, MRMSrcMem, (outs RC:$dst),
7100 (ins RC:$src1, x86memop:$src2),
7102 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7103 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7105 (OpVT (OpNode RC:$src1, (memop_frag addr:$src2))))]>, OpSize;
7108 let Predicates = [HasAVX] in
7109 defm VPCMPGTQ : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v2i64, VR128,
7110 memopv2i64, i128mem, 0>, VEX_4V;
7112 let Predicates = [HasAVX2] in
7113 defm VPCMPGTQY : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v4i64, VR256,
7114 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
7116 let Constraints = "$src1 = $dst" in
7117 defm PCMPGTQ : SS42I_binop_rm<0x37, "pcmpgtq", X86pcmpgt, v2i64, VR128,
7118 memopv2i64, i128mem>;
7120 //===----------------------------------------------------------------------===//
7121 // SSE4.2 - String/text Processing Instructions
7122 //===----------------------------------------------------------------------===//
7124 // Packed Compare Implicit Length Strings, Return Mask
7125 multiclass pseudo_pcmpistrm<string asm> {
7126 def REG : PseudoI<(outs VR128:$dst),
7127 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7128 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
7130 def MEM : PseudoI<(outs VR128:$dst),
7131 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7132 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1,
7133 (bc_v16i8 (memopv2i64 addr:$src2)), imm:$src3))]>;
7136 let Defs = [EFLAGS], usesCustomInserter = 1 in {
7137 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
7138 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[UseSSE42]>;
7141 multiclass pcmpistrm_SS42AI<string asm> {
7142 def rr : SS42AI<0x62, MRMSrcReg, (outs),
7143 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7144 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
7147 def rm :SS42AI<0x62, MRMSrcMem, (outs),
7148 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7149 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
7153 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
7154 let Predicates = [HasAVX] in
7155 defm VPCMPISTRM128 : pcmpistrm_SS42AI<"vpcmpistrm">, VEX;
7156 defm PCMPISTRM128 : pcmpistrm_SS42AI<"pcmpistrm"> ;
7159 // Packed Compare Explicit Length Strings, Return Mask
7160 multiclass pseudo_pcmpestrm<string asm> {
7161 def REG : PseudoI<(outs VR128:$dst),
7162 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7163 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
7164 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
7165 def MEM : PseudoI<(outs VR128:$dst),
7166 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7167 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128 VR128:$src1, EAX,
7168 (bc_v16i8 (memopv2i64 addr:$src3)), EDX, imm:$src5))]>;
7171 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
7172 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
7173 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[UseSSE42]>;
7176 multiclass SS42AI_pcmpestrm<string asm> {
7177 def rr : SS42AI<0x60, MRMSrcReg, (outs),
7178 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7179 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7182 def rm : SS42AI<0x60, MRMSrcMem, (outs),
7183 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7184 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7188 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
7189 let Predicates = [HasAVX] in
7190 defm VPCMPESTRM128 : SS42AI_pcmpestrm<"vpcmpestrm">, VEX;
7191 defm PCMPESTRM128 : SS42AI_pcmpestrm<"pcmpestrm">;
7194 // Packed Compare Implicit Length Strings, Return Index
7195 multiclass pseudo_pcmpistri<string asm> {
7196 def REG : PseudoI<(outs GR32:$dst),
7197 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7198 [(set GR32:$dst, EFLAGS,
7199 (X86pcmpistri VR128:$src1, VR128:$src2, imm:$src3))]>;
7200 def MEM : PseudoI<(outs GR32:$dst),
7201 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7202 [(set GR32:$dst, EFLAGS, (X86pcmpistri VR128:$src1,
7203 (bc_v16i8 (memopv2i64 addr:$src2)), imm:$src3))]>;
7206 let Defs = [EFLAGS], usesCustomInserter = 1 in {
7207 defm VPCMPISTRI : pseudo_pcmpistri<"#VPCMPISTRI">, Requires<[HasAVX]>;
7208 defm PCMPISTRI : pseudo_pcmpistri<"#PCMPISTRI">, Requires<[UseSSE42]>;
7211 multiclass SS42AI_pcmpistri<string asm> {
7212 def rr : SS42AI<0x63, MRMSrcReg, (outs),
7213 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7214 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
7217 def rm : SS42AI<0x63, MRMSrcMem, (outs),
7218 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7219 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
7223 let Defs = [ECX, EFLAGS], neverHasSideEffects = 1 in {
7224 let Predicates = [HasAVX] in
7225 defm VPCMPISTRI : SS42AI_pcmpistri<"vpcmpistri">, VEX;
7226 defm PCMPISTRI : SS42AI_pcmpistri<"pcmpistri">;
7229 // Packed Compare Explicit Length Strings, Return Index
7230 multiclass pseudo_pcmpestri<string asm> {
7231 def REG : PseudoI<(outs GR32:$dst),
7232 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7233 [(set GR32:$dst, EFLAGS,
7234 (X86pcmpestri VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
7235 def MEM : PseudoI<(outs GR32:$dst),
7236 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7237 [(set GR32:$dst, EFLAGS,
7238 (X86pcmpestri VR128:$src1, EAX, (bc_v16i8 (memopv2i64 addr:$src3)), EDX,
7242 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
7243 defm VPCMPESTRI : pseudo_pcmpestri<"#VPCMPESTRI">, Requires<[HasAVX]>;
7244 defm PCMPESTRI : pseudo_pcmpestri<"#PCMPESTRI">, Requires<[UseSSE42]>;
7247 multiclass SS42AI_pcmpestri<string asm> {
7248 def rr : SS42AI<0x61, MRMSrcReg, (outs),
7249 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7250 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7253 def rm : SS42AI<0x61, MRMSrcMem, (outs),
7254 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7255 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7259 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
7260 let Predicates = [HasAVX] in
7261 defm VPCMPESTRI : SS42AI_pcmpestri<"vpcmpestri">, VEX;
7262 defm PCMPESTRI : SS42AI_pcmpestri<"pcmpestri">;
7265 //===----------------------------------------------------------------------===//
7266 // SSE4.2 - CRC Instructions
7267 //===----------------------------------------------------------------------===//
7269 // No CRC instructions have AVX equivalents
7271 // crc intrinsic instruction
7272 // This set of instructions are only rm, the only difference is the size
7274 let Constraints = "$src1 = $dst" in {
7275 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
7276 (ins GR32:$src1, i8mem:$src2),
7277 "crc32{b}\t{$src2, $src1|$src1, $src2}",
7279 (int_x86_sse42_crc32_32_8 GR32:$src1,
7280 (load addr:$src2)))], IIC_CRC32_MEM>;
7281 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
7282 (ins GR32:$src1, GR8:$src2),
7283 "crc32{b}\t{$src2, $src1|$src1, $src2}",
7285 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))],
7287 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
7288 (ins GR32:$src1, i16mem:$src2),
7289 "crc32{w}\t{$src2, $src1|$src1, $src2}",
7291 (int_x86_sse42_crc32_32_16 GR32:$src1,
7292 (load addr:$src2)))], IIC_CRC32_MEM>,
7294 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
7295 (ins GR32:$src1, GR16:$src2),
7296 "crc32{w}\t{$src2, $src1|$src1, $src2}",
7298 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))],
7301 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
7302 (ins GR32:$src1, i32mem:$src2),
7303 "crc32{l}\t{$src2, $src1|$src1, $src2}",
7305 (int_x86_sse42_crc32_32_32 GR32:$src1,
7306 (load addr:$src2)))], IIC_CRC32_MEM>;
7307 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
7308 (ins GR32:$src1, GR32:$src2),
7309 "crc32{l}\t{$src2, $src1|$src1, $src2}",
7311 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))],
7313 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
7314 (ins GR64:$src1, i8mem:$src2),
7315 "crc32{b}\t{$src2, $src1|$src1, $src2}",
7317 (int_x86_sse42_crc32_64_8 GR64:$src1,
7318 (load addr:$src2)))], IIC_CRC32_MEM>,
7320 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
7321 (ins GR64:$src1, GR8:$src2),
7322 "crc32{b}\t{$src2, $src1|$src1, $src2}",
7324 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))],
7327 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
7328 (ins GR64:$src1, i64mem:$src2),
7329 "crc32{q}\t{$src2, $src1|$src1, $src2}",
7331 (int_x86_sse42_crc32_64_64 GR64:$src1,
7332 (load addr:$src2)))], IIC_CRC32_MEM>,
7334 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
7335 (ins GR64:$src1, GR64:$src2),
7336 "crc32{q}\t{$src2, $src1|$src1, $src2}",
7338 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))],
7343 //===----------------------------------------------------------------------===//
7344 // SHA-NI Instructions
7345 //===----------------------------------------------------------------------===//
7347 multiclass SHAI_binop<bits<8> Opc, string OpcodeStr, Intrinsic IntId,
7349 def rr : I<Opc, MRMSrcReg, (outs VR128:$dst),
7350 (ins VR128:$src1, VR128:$src2),
7351 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7353 (set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0)),
7354 (set VR128:$dst, (IntId VR128:$src1, VR128:$src2)))]>, T8;
7356 def rm : I<Opc, MRMSrcMem, (outs VR128:$dst),
7357 (ins VR128:$src1, i128mem:$src2),
7358 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7360 (set VR128:$dst, (IntId VR128:$src1,
7361 (bc_v4i32 (memopv2i64 addr:$src2)), XMM0)),
7362 (set VR128:$dst, (IntId VR128:$src1,
7363 (bc_v4i32 (memopv2i64 addr:$src2)))))]>, T8;
7366 let Constraints = "$src1 = $dst", Predicates = [HasSHA] in {
7367 def SHA1RNDS4rri : Ii8<0xCC, MRMSrcReg, (outs VR128:$dst),
7368 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7369 "sha1rnds4\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7371 (int_x86_sha1rnds4 VR128:$src1, VR128:$src2,
7372 (i8 imm:$src3)))]>, TA;
7373 def SHA1RNDS4rmi : Ii8<0xCC, MRMSrcMem, (outs VR128:$dst),
7374 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7375 "sha1rnds4\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7377 (int_x86_sha1rnds4 VR128:$src1,
7378 (bc_v4i32 (memopv2i64 addr:$src2)),
7379 (i8 imm:$src3)))]>, TA;
7381 defm SHA1NEXTE : SHAI_binop<0xC8, "sha1nexte", int_x86_sha1nexte>;
7382 defm SHA1MSG1 : SHAI_binop<0xC9, "sha1msg1", int_x86_sha1msg1>;
7383 defm SHA1MSG2 : SHAI_binop<0xCA, "sha1msg2", int_x86_sha1msg2>;
7386 defm SHA256RNDS2 : SHAI_binop<0xCB, "sha256rnds2", int_x86_sha256rnds2, 1>;
7388 defm SHA256MSG1 : SHAI_binop<0xCC, "sha256msg1", int_x86_sha256msg1>;
7389 defm SHA256MSG2 : SHAI_binop<0xCD, "sha256msg2", int_x86_sha256msg2>;
7392 // Aliases with explicit %xmm0
7393 def : InstAlias<"sha256rnds2\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7394 (SHA256RNDS2rr VR128:$dst, VR128:$src2)>;
7395 def : InstAlias<"sha256rnds2\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7396 (SHA256RNDS2rm VR128:$dst, i128mem:$src2)>;
7398 //===----------------------------------------------------------------------===//
7399 // AES-NI Instructions
7400 //===----------------------------------------------------------------------===//
7402 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
7403 Intrinsic IntId128, bit Is2Addr = 1> {
7404 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
7405 (ins VR128:$src1, VR128:$src2),
7407 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7408 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7409 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
7411 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
7412 (ins VR128:$src1, i128mem:$src2),
7414 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7415 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7417 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
7420 // Perform One Round of an AES Encryption/Decryption Flow
7421 let Predicates = [HasAVX, HasAES] in {
7422 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
7423 int_x86_aesni_aesenc, 0>, VEX_4V;
7424 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
7425 int_x86_aesni_aesenclast, 0>, VEX_4V;
7426 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
7427 int_x86_aesni_aesdec, 0>, VEX_4V;
7428 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
7429 int_x86_aesni_aesdeclast, 0>, VEX_4V;
7432 let Constraints = "$src1 = $dst" in {
7433 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
7434 int_x86_aesni_aesenc>;
7435 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
7436 int_x86_aesni_aesenclast>;
7437 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
7438 int_x86_aesni_aesdec>;
7439 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
7440 int_x86_aesni_aesdeclast>;
7443 // Perform the AES InvMixColumn Transformation
7444 let Predicates = [HasAVX, HasAES] in {
7445 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7447 "vaesimc\t{$src1, $dst|$dst, $src1}",
7449 (int_x86_aesni_aesimc VR128:$src1))]>,
7451 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7452 (ins i128mem:$src1),
7453 "vaesimc\t{$src1, $dst|$dst, $src1}",
7454 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7457 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7459 "aesimc\t{$src1, $dst|$dst, $src1}",
7461 (int_x86_aesni_aesimc VR128:$src1))]>,
7463 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7464 (ins i128mem:$src1),
7465 "aesimc\t{$src1, $dst|$dst, $src1}",
7466 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7469 // AES Round Key Generation Assist
7470 let Predicates = [HasAVX, HasAES] in {
7471 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7472 (ins VR128:$src1, i8imm:$src2),
7473 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7475 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7477 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7478 (ins i128mem:$src1, i8imm:$src2),
7479 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7481 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7484 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7485 (ins VR128:$src1, i8imm:$src2),
7486 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7488 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7490 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7491 (ins i128mem:$src1, i8imm:$src2),
7492 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7494 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7497 //===----------------------------------------------------------------------===//
7498 // PCLMUL Instructions
7499 //===----------------------------------------------------------------------===//
7501 // AVX carry-less Multiplication instructions
7502 def VPCLMULQDQrr : AVXPCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7503 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7504 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7506 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
7508 def VPCLMULQDQrm : AVXPCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7509 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7510 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7511 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7512 (memopv2i64 addr:$src2), imm:$src3))]>;
7514 // Carry-less Multiplication instructions
7515 let Constraints = "$src1 = $dst" in {
7516 def PCLMULQDQrr : PCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7517 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7518 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7520 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))],
7521 IIC_SSE_PCLMULQDQ_RR>;
7523 def PCLMULQDQrm : PCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7524 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7525 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7526 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7527 (memopv2i64 addr:$src2), imm:$src3))],
7528 IIC_SSE_PCLMULQDQ_RM>;
7529 } // Constraints = "$src1 = $dst"
7532 multiclass pclmul_alias<string asm, int immop> {
7533 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7534 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
7536 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7537 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
7539 def : InstAlias<!strconcat("vpclmul", asm,
7540 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7541 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
7543 def : InstAlias<!strconcat("vpclmul", asm,
7544 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7545 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
7547 defm : pclmul_alias<"hqhq", 0x11>;
7548 defm : pclmul_alias<"hqlq", 0x01>;
7549 defm : pclmul_alias<"lqhq", 0x10>;
7550 defm : pclmul_alias<"lqlq", 0x00>;
7552 //===----------------------------------------------------------------------===//
7553 // SSE4A Instructions
7554 //===----------------------------------------------------------------------===//
7556 let Predicates = [HasSSE4A] in {
7558 let Constraints = "$src = $dst" in {
7559 def EXTRQI : Ii8<0x78, MRM0r, (outs VR128:$dst),
7560 (ins VR128:$src, i8imm:$len, i8imm:$idx),
7561 "extrq\t{$idx, $len, $src|$src, $len, $idx}",
7562 [(set VR128:$dst, (int_x86_sse4a_extrqi VR128:$src, imm:$len,
7563 imm:$idx))]>, TB, OpSize;
7564 def EXTRQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7565 (ins VR128:$src, VR128:$mask),
7566 "extrq\t{$mask, $src|$src, $mask}",
7567 [(set VR128:$dst, (int_x86_sse4a_extrq VR128:$src,
7568 VR128:$mask))]>, TB, OpSize;
7570 def INSERTQI : Ii8<0x78, MRMSrcReg, (outs VR128:$dst),
7571 (ins VR128:$src, VR128:$src2, i8imm:$len, i8imm:$idx),
7572 "insertq\t{$idx, $len, $src2, $src|$src, $src2, $len, $idx}",
7573 [(set VR128:$dst, (int_x86_sse4a_insertqi VR128:$src,
7574 VR128:$src2, imm:$len, imm:$idx))]>, XD;
7575 def INSERTQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7576 (ins VR128:$src, VR128:$mask),
7577 "insertq\t{$mask, $src|$src, $mask}",
7578 [(set VR128:$dst, (int_x86_sse4a_insertq VR128:$src,
7579 VR128:$mask))]>, XD;
7582 def MOVNTSS : I<0x2B, MRMDestMem, (outs), (ins f32mem:$dst, VR128:$src),
7583 "movntss\t{$src, $dst|$dst, $src}",
7584 [(int_x86_sse4a_movnt_ss addr:$dst, VR128:$src)]>, XS;
7586 def MOVNTSD : I<0x2B, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
7587 "movntsd\t{$src, $dst|$dst, $src}",
7588 [(int_x86_sse4a_movnt_sd addr:$dst, VR128:$src)]>, XD;
7591 //===----------------------------------------------------------------------===//
7593 //===----------------------------------------------------------------------===//
7595 //===----------------------------------------------------------------------===//
7596 // VBROADCAST - Load from memory and broadcast to all elements of the
7597 // destination operand
7599 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
7600 X86MemOperand x86memop, Intrinsic Int> :
7601 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7602 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7603 [(set RC:$dst, (Int addr:$src))]>, VEX;
7605 // AVX2 adds register forms
7606 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
7608 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7609 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7610 [(set RC:$dst, (Int VR128:$src))]>, VEX;
7612 let ExeDomain = SSEPackedSingle in {
7613 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
7614 int_x86_avx_vbroadcast_ss>;
7615 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
7616 int_x86_avx_vbroadcast_ss_256>, VEX_L;
7618 let ExeDomain = SSEPackedDouble in
7619 def VBROADCASTSDYrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
7620 int_x86_avx_vbroadcast_sd_256>, VEX_L;
7621 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
7622 int_x86_avx_vbroadcastf128_pd_256>, VEX_L;
7624 let ExeDomain = SSEPackedSingle in {
7625 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
7626 int_x86_avx2_vbroadcast_ss_ps>;
7627 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
7628 int_x86_avx2_vbroadcast_ss_ps_256>, VEX_L;
7630 let ExeDomain = SSEPackedDouble in
7631 def VBROADCASTSDYrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
7632 int_x86_avx2_vbroadcast_sd_pd_256>, VEX_L;
7634 let Predicates = [HasAVX2] in
7635 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
7636 int_x86_avx2_vbroadcasti128>, VEX_L;
7638 let Predicates = [HasAVX] in
7639 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
7640 (VBROADCASTF128 addr:$src)>;
7643 //===----------------------------------------------------------------------===//
7644 // VINSERTF128 - Insert packed floating-point values
7646 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7647 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
7648 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7649 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7652 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
7653 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
7654 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7658 let Predicates = [HasAVX] in {
7659 def : Pat<(vinsert128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7661 (VINSERTF128rr VR256:$src1, VR128:$src2,
7662 (INSERT_get_vinsert128_imm VR256:$ins))>;
7663 def : Pat<(vinsert128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7665 (VINSERTF128rr VR256:$src1, VR128:$src2,
7666 (INSERT_get_vinsert128_imm VR256:$ins))>;
7668 def : Pat<(vinsert128_insert:$ins (v8f32 VR256:$src1), (loadv4f32 addr:$src2),
7670 (VINSERTF128rm VR256:$src1, addr:$src2,
7671 (INSERT_get_vinsert128_imm VR256:$ins))>;
7672 def : Pat<(vinsert128_insert:$ins (v4f64 VR256:$src1), (loadv2f64 addr:$src2),
7674 (VINSERTF128rm VR256:$src1, addr:$src2,
7675 (INSERT_get_vinsert128_imm VR256:$ins))>;
7678 let Predicates = [HasAVX1Only] in {
7679 def : Pat<(vinsert128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7681 (VINSERTF128rr VR256:$src1, VR128:$src2,
7682 (INSERT_get_vinsert128_imm VR256:$ins))>;
7683 def : Pat<(vinsert128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7685 (VINSERTF128rr VR256:$src1, VR128:$src2,
7686 (INSERT_get_vinsert128_imm VR256:$ins))>;
7687 def : Pat<(vinsert128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7689 (VINSERTF128rr VR256:$src1, VR128:$src2,
7690 (INSERT_get_vinsert128_imm VR256:$ins))>;
7691 def : Pat<(vinsert128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7693 (VINSERTF128rr VR256:$src1, VR128:$src2,
7694 (INSERT_get_vinsert128_imm VR256:$ins))>;
7696 def : Pat<(vinsert128_insert:$ins (v4i64 VR256:$src1), (loadv2i64 addr:$src2),
7698 (VINSERTF128rm VR256:$src1, addr:$src2,
7699 (INSERT_get_vinsert128_imm VR256:$ins))>;
7700 def : Pat<(vinsert128_insert:$ins (v8i32 VR256:$src1),
7701 (bc_v4i32 (loadv2i64 addr:$src2)),
7703 (VINSERTF128rm VR256:$src1, addr:$src2,
7704 (INSERT_get_vinsert128_imm VR256:$ins))>;
7705 def : Pat<(vinsert128_insert:$ins (v32i8 VR256:$src1),
7706 (bc_v16i8 (loadv2i64 addr:$src2)),
7708 (VINSERTF128rm VR256:$src1, addr:$src2,
7709 (INSERT_get_vinsert128_imm VR256:$ins))>;
7710 def : Pat<(vinsert128_insert:$ins (v16i16 VR256:$src1),
7711 (bc_v8i16 (loadv2i64 addr:$src2)),
7713 (VINSERTF128rm VR256:$src1, addr:$src2,
7714 (INSERT_get_vinsert128_imm VR256:$ins))>;
7717 //===----------------------------------------------------------------------===//
7718 // VEXTRACTF128 - Extract packed floating-point values
7720 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7721 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
7722 (ins VR256:$src1, i8imm:$src2),
7723 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7726 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
7727 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
7728 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7733 let Predicates = [HasAVX] in {
7734 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7735 (v4f32 (VEXTRACTF128rr
7736 (v8f32 VR256:$src1),
7737 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7738 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7739 (v2f64 (VEXTRACTF128rr
7740 (v4f64 VR256:$src1),
7741 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7743 def : Pat<(store (v4f32 (vextract128_extract:$ext (v8f32 VR256:$src1),
7744 (iPTR imm))), addr:$dst),
7745 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7746 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7747 def : Pat<(store (v2f64 (vextract128_extract:$ext (v4f64 VR256:$src1),
7748 (iPTR imm))), addr:$dst),
7749 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7750 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7753 let Predicates = [HasAVX1Only] in {
7754 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7755 (v2i64 (VEXTRACTF128rr
7756 (v4i64 VR256:$src1),
7757 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7758 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7759 (v4i32 (VEXTRACTF128rr
7760 (v8i32 VR256:$src1),
7761 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7762 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7763 (v8i16 (VEXTRACTF128rr
7764 (v16i16 VR256:$src1),
7765 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7766 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7767 (v16i8 (VEXTRACTF128rr
7768 (v32i8 VR256:$src1),
7769 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7771 def : Pat<(alignedstore (v2i64 (vextract128_extract:$ext (v4i64 VR256:$src1),
7772 (iPTR imm))), addr:$dst),
7773 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7774 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7775 def : Pat<(alignedstore (v4i32 (vextract128_extract:$ext (v8i32 VR256:$src1),
7776 (iPTR imm))), addr:$dst),
7777 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7778 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7779 def : Pat<(alignedstore (v8i16 (vextract128_extract:$ext (v16i16 VR256:$src1),
7780 (iPTR imm))), addr:$dst),
7781 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7782 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7783 def : Pat<(alignedstore (v16i8 (vextract128_extract:$ext (v32i8 VR256:$src1),
7784 (iPTR imm))), addr:$dst),
7785 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7786 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7789 //===----------------------------------------------------------------------===//
7790 // VMASKMOV - Conditional SIMD Packed Loads and Stores
7792 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
7793 Intrinsic IntLd, Intrinsic IntLd256,
7794 Intrinsic IntSt, Intrinsic IntSt256> {
7795 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
7796 (ins VR128:$src1, f128mem:$src2),
7797 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7798 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
7800 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
7801 (ins VR256:$src1, f256mem:$src2),
7802 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7803 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7805 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
7806 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
7807 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7808 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7809 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
7810 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
7811 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7812 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V, VEX_L;
7815 let ExeDomain = SSEPackedSingle in
7816 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
7817 int_x86_avx_maskload_ps,
7818 int_x86_avx_maskload_ps_256,
7819 int_x86_avx_maskstore_ps,
7820 int_x86_avx_maskstore_ps_256>;
7821 let ExeDomain = SSEPackedDouble in
7822 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
7823 int_x86_avx_maskload_pd,
7824 int_x86_avx_maskload_pd_256,
7825 int_x86_avx_maskstore_pd,
7826 int_x86_avx_maskstore_pd_256>;
7828 //===----------------------------------------------------------------------===//
7829 // VPERMIL - Permute Single and Double Floating-Point Values
7831 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
7832 RegisterClass RC, X86MemOperand x86memop_f,
7833 X86MemOperand x86memop_i, PatFrag i_frag,
7834 Intrinsic IntVar, ValueType vt> {
7835 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
7836 (ins RC:$src1, RC:$src2),
7837 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7838 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
7839 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
7840 (ins RC:$src1, x86memop_i:$src2),
7841 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7842 [(set RC:$dst, (IntVar RC:$src1,
7843 (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
7845 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
7846 (ins RC:$src1, i8imm:$src2),
7847 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7848 [(set RC:$dst, (vt (X86VPermilp RC:$src1, (i8 imm:$src2))))]>, VEX;
7849 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
7850 (ins x86memop_f:$src1, i8imm:$src2),
7851 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7853 (vt (X86VPermilp (memop addr:$src1), (i8 imm:$src2))))]>, VEX;
7856 let ExeDomain = SSEPackedSingle in {
7857 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
7858 memopv2i64, int_x86_avx_vpermilvar_ps, v4f32>;
7859 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
7860 memopv4i64, int_x86_avx_vpermilvar_ps_256, v8f32>, VEX_L;
7862 let ExeDomain = SSEPackedDouble in {
7863 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
7864 memopv2i64, int_x86_avx_vpermilvar_pd, v2f64>;
7865 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
7866 memopv4i64, int_x86_avx_vpermilvar_pd_256, v4f64>, VEX_L;
7869 let Predicates = [HasAVX] in {
7870 def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7871 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7872 def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7873 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7874 def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (memopv4i64 addr:$src1)),
7876 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7877 def : Pat<(v4i64 (X86VPermilp (memopv4i64 addr:$src1), (i8 imm:$imm))),
7878 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7880 def : Pat<(v2i64 (X86VPermilp VR128:$src1, (i8 imm:$imm))),
7881 (VPERMILPDri VR128:$src1, imm:$imm)>;
7882 def : Pat<(v2i64 (X86VPermilp (memopv2i64 addr:$src1), (i8 imm:$imm))),
7883 (VPERMILPDmi addr:$src1, imm:$imm)>;
7886 //===----------------------------------------------------------------------===//
7887 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
7889 let ExeDomain = SSEPackedSingle in {
7890 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
7891 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7892 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7893 [(set VR256:$dst, (v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7894 (i8 imm:$src3))))]>, VEX_4V, VEX_L;
7895 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
7896 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7897 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7898 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv8f32 addr:$src2),
7899 (i8 imm:$src3)))]>, VEX_4V, VEX_L;
7902 let Predicates = [HasAVX] in {
7903 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7904 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7905 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
7906 (memopv4f64 addr:$src2), (i8 imm:$imm))),
7907 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7910 let Predicates = [HasAVX1Only] in {
7911 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7912 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7913 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7914 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7915 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7916 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7917 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7918 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7920 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
7921 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7922 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7923 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
7924 (memopv4i64 addr:$src2), (i8 imm:$imm))),
7925 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7926 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
7927 (bc_v32i8 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7928 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7929 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7930 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7931 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7934 //===----------------------------------------------------------------------===//
7935 // VZERO - Zero YMM registers
7937 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
7938 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
7939 // Zero All YMM registers
7940 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
7941 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
7943 // Zero Upper bits of YMM registers
7944 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
7945 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
7948 //===----------------------------------------------------------------------===//
7949 // Half precision conversion instructions
7950 //===----------------------------------------------------------------------===//
7951 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7952 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7953 "vcvtph2ps\t{$src, $dst|$dst, $src}",
7954 [(set RC:$dst, (Int VR128:$src))]>,
7956 let neverHasSideEffects = 1, mayLoad = 1 in
7957 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7958 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
7961 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7962 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
7963 (ins RC:$src1, i32i8imm:$src2),
7964 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7965 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
7967 let neverHasSideEffects = 1, mayStore = 1 in
7968 def mr : Ii8<0x1D, MRMDestMem, (outs),
7969 (ins x86memop:$dst, RC:$src1, i32i8imm:$src2),
7970 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
7974 let Predicates = [HasF16C] in {
7975 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
7976 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>, VEX_L;
7977 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
7978 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>, VEX_L;
7981 //===----------------------------------------------------------------------===//
7982 // AVX2 Instructions
7983 //===----------------------------------------------------------------------===//
7985 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
7986 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
7987 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
7988 X86MemOperand x86memop> {
7989 let isCommutable = 1 in
7990 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
7991 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
7992 !strconcat(OpcodeStr,
7993 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7994 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
7996 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
7997 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
7998 !strconcat(OpcodeStr,
7999 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
8002 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
8006 let isCommutable = 0 in {
8007 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
8008 VR128, memopv2i64, i128mem>;
8009 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
8010 VR256, memopv4i64, i256mem>, VEX_L;
8013 def : Pat<(v4i32 (X86Blendi (v4i32 VR128:$src1), (v4i32 VR128:$src2),
8015 (VPBLENDDrri VR128:$src1, VR128:$src2, imm:$mask)>;
8016 def : Pat<(v8i32 (X86Blendi (v8i32 VR256:$src1), (v8i32 VR256:$src2),
8018 (VPBLENDDYrri VR256:$src1, VR256:$src2, imm:$mask)>;
8020 //===----------------------------------------------------------------------===//
8021 // VPBROADCAST - Load from memory and broadcast to all elements of the
8022 // destination operand
8024 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
8025 X86MemOperand x86memop, PatFrag ld_frag,
8026 Intrinsic Int128, Intrinsic Int256> {
8027 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
8028 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
8029 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
8030 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
8031 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
8033 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
8034 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
8035 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
8036 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX, VEX_L;
8037 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
8038 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
8040 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>,
8044 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
8045 int_x86_avx2_pbroadcastb_128,
8046 int_x86_avx2_pbroadcastb_256>;
8047 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
8048 int_x86_avx2_pbroadcastw_128,
8049 int_x86_avx2_pbroadcastw_256>;
8050 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
8051 int_x86_avx2_pbroadcastd_128,
8052 int_x86_avx2_pbroadcastd_256>;
8053 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
8054 int_x86_avx2_pbroadcastq_128,
8055 int_x86_avx2_pbroadcastq_256>;
8057 let Predicates = [HasAVX2] in {
8058 def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
8059 (VPBROADCASTBrm addr:$src)>;
8060 def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
8061 (VPBROADCASTBYrm addr:$src)>;
8062 def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
8063 (VPBROADCASTWrm addr:$src)>;
8064 def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
8065 (VPBROADCASTWYrm addr:$src)>;
8066 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
8067 (VPBROADCASTDrm addr:$src)>;
8068 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
8069 (VPBROADCASTDYrm addr:$src)>;
8070 def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
8071 (VPBROADCASTQrm addr:$src)>;
8072 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
8073 (VPBROADCASTQYrm addr:$src)>;
8075 def : Pat<(v16i8 (X86VBroadcast (v16i8 VR128:$src))),
8076 (VPBROADCASTBrr VR128:$src)>;
8077 def : Pat<(v32i8 (X86VBroadcast (v16i8 VR128:$src))),
8078 (VPBROADCASTBYrr VR128:$src)>;
8079 def : Pat<(v8i16 (X86VBroadcast (v8i16 VR128:$src))),
8080 (VPBROADCASTWrr VR128:$src)>;
8081 def : Pat<(v16i16 (X86VBroadcast (v8i16 VR128:$src))),
8082 (VPBROADCASTWYrr VR128:$src)>;
8083 def : Pat<(v4i32 (X86VBroadcast (v4i32 VR128:$src))),
8084 (VPBROADCASTDrr VR128:$src)>;
8085 def : Pat<(v8i32 (X86VBroadcast (v4i32 VR128:$src))),
8086 (VPBROADCASTDYrr VR128:$src)>;
8087 def : Pat<(v2i64 (X86VBroadcast (v2i64 VR128:$src))),
8088 (VPBROADCASTQrr VR128:$src)>;
8089 def : Pat<(v4i64 (X86VBroadcast (v2i64 VR128:$src))),
8090 (VPBROADCASTQYrr VR128:$src)>;
8091 def : Pat<(v4f32 (X86VBroadcast (v4f32 VR128:$src))),
8092 (VBROADCASTSSrr VR128:$src)>;
8093 def : Pat<(v8f32 (X86VBroadcast (v4f32 VR128:$src))),
8094 (VBROADCASTSSYrr VR128:$src)>;
8095 def : Pat<(v2f64 (X86VBroadcast (v2f64 VR128:$src))),
8096 (VPBROADCASTQrr VR128:$src)>;
8097 def : Pat<(v4f64 (X86VBroadcast (v2f64 VR128:$src))),
8098 (VBROADCASTSDYrr VR128:$src)>;
8100 // Provide fallback in case the load node that is used in the patterns above
8101 // is used by additional users, which prevents the pattern selection.
8102 let AddedComplexity = 20 in {
8103 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
8104 (VBROADCASTSSrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
8105 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
8106 (VBROADCASTSSYrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
8107 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
8108 (VBROADCASTSDYrr (COPY_TO_REGCLASS FR64:$src, VR128))>;
8110 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
8111 (VBROADCASTSSrr (COPY_TO_REGCLASS GR32:$src, VR128))>;
8112 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
8113 (VBROADCASTSSYrr (COPY_TO_REGCLASS GR32:$src, VR128))>;
8114 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
8115 (VBROADCASTSDYrr (COPY_TO_REGCLASS GR64:$src, VR128))>;
8119 // AVX1 broadcast patterns
8120 let Predicates = [HasAVX1Only] in {
8121 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
8122 (VBROADCASTSSYrm addr:$src)>;
8123 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
8124 (VBROADCASTSDYrm addr:$src)>;
8125 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
8126 (VBROADCASTSSrm addr:$src)>;
8129 let Predicates = [HasAVX] in {
8130 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
8131 (VBROADCASTSSYrm addr:$src)>;
8132 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
8133 (VBROADCASTSDYrm addr:$src)>;
8134 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
8135 (VBROADCASTSSrm addr:$src)>;
8137 // Provide fallback in case the load node that is used in the patterns above
8138 // is used by additional users, which prevents the pattern selection.
8139 let AddedComplexity = 20 in {
8140 // 128bit broadcasts:
8141 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
8142 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0)>;
8143 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
8144 (VINSERTF128rr (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)),
8145 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0), sub_xmm),
8146 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0), 1)>;
8147 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
8148 (VINSERTF128rr (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)),
8149 (VPSHUFDri (COPY_TO_REGCLASS FR64:$src, VR128), 0x44), sub_xmm),
8150 (VPSHUFDri (COPY_TO_REGCLASS FR64:$src, VR128), 0x44), 1)>;
8152 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
8153 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0)>;
8154 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
8155 (VINSERTF128rr (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)),
8156 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0), sub_xmm),
8157 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0), 1)>;
8158 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
8159 (VINSERTF128rr (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)),
8160 (VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), sub_xmm),
8161 (VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), 1)>;
8165 //===----------------------------------------------------------------------===//
8166 // VPERM - Permute instructions
8169 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
8171 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
8172 (ins VR256:$src1, VR256:$src2),
8173 !strconcat(OpcodeStr,
8174 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8176 (OpVT (X86VPermv VR256:$src1, VR256:$src2)))]>,
8178 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
8179 (ins VR256:$src1, i256mem:$src2),
8180 !strconcat(OpcodeStr,
8181 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8183 (OpVT (X86VPermv VR256:$src1,
8184 (bitconvert (mem_frag addr:$src2)))))]>,
8188 defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, v8i32>;
8189 let ExeDomain = SSEPackedSingle in
8190 defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, v8f32>;
8192 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
8194 def Yri : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
8195 (ins VR256:$src1, i8imm:$src2),
8196 !strconcat(OpcodeStr,
8197 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8199 (OpVT (X86VPermi VR256:$src1, (i8 imm:$src2))))]>,
8201 def Ymi : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
8202 (ins i256mem:$src1, i8imm:$src2),
8203 !strconcat(OpcodeStr,
8204 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8206 (OpVT (X86VPermi (mem_frag addr:$src1),
8207 (i8 imm:$src2))))]>, VEX, VEX_L;
8210 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, v4i64>, VEX_W;
8211 let ExeDomain = SSEPackedDouble in
8212 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, v4f64>, VEX_W;
8214 //===----------------------------------------------------------------------===//
8215 // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
8217 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
8218 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
8219 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
8220 [(set VR256:$dst, (v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2,
8221 (i8 imm:$src3))))]>, VEX_4V, VEX_L;
8222 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
8223 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
8224 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
8225 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
8226 (i8 imm:$src3)))]>, VEX_4V, VEX_L;
8228 let Predicates = [HasAVX2] in {
8229 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
8230 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
8231 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
8232 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
8233 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
8234 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
8236 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (memopv4i64 addr:$src2)),
8238 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
8239 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
8240 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
8241 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
8242 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)),
8244 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
8248 //===----------------------------------------------------------------------===//
8249 // VINSERTI128 - Insert packed integer values
8251 let neverHasSideEffects = 1 in {
8252 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
8253 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
8254 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
8257 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
8258 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
8259 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
8263 let Predicates = [HasAVX2] in {
8264 def : Pat<(vinsert128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
8266 (VINSERTI128rr VR256:$src1, VR128:$src2,
8267 (INSERT_get_vinsert128_imm VR256:$ins))>;
8268 def : Pat<(vinsert128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
8270 (VINSERTI128rr VR256:$src1, VR128:$src2,
8271 (INSERT_get_vinsert128_imm VR256:$ins))>;
8272 def : Pat<(vinsert128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
8274 (VINSERTI128rr VR256:$src1, VR128:$src2,
8275 (INSERT_get_vinsert128_imm VR256:$ins))>;
8276 def : Pat<(vinsert128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
8278 (VINSERTI128rr VR256:$src1, VR128:$src2,
8279 (INSERT_get_vinsert128_imm VR256:$ins))>;
8281 def : Pat<(vinsert128_insert:$ins (v4i64 VR256:$src1), (loadv2i64 addr:$src2),
8283 (VINSERTI128rm VR256:$src1, addr:$src2,
8284 (INSERT_get_vinsert128_imm VR256:$ins))>;
8285 def : Pat<(vinsert128_insert:$ins (v8i32 VR256:$src1),
8286 (bc_v4i32 (loadv2i64 addr:$src2)),
8288 (VINSERTI128rm VR256:$src1, addr:$src2,
8289 (INSERT_get_vinsert128_imm VR256:$ins))>;
8290 def : Pat<(vinsert128_insert:$ins (v32i8 VR256:$src1),
8291 (bc_v16i8 (loadv2i64 addr:$src2)),
8293 (VINSERTI128rm VR256:$src1, addr:$src2,
8294 (INSERT_get_vinsert128_imm VR256:$ins))>;
8295 def : Pat<(vinsert128_insert:$ins (v16i16 VR256:$src1),
8296 (bc_v8i16 (loadv2i64 addr:$src2)),
8298 (VINSERTI128rm VR256:$src1, addr:$src2,
8299 (INSERT_get_vinsert128_imm VR256:$ins))>;
8302 //===----------------------------------------------------------------------===//
8303 // VEXTRACTI128 - Extract packed integer values
8305 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
8306 (ins VR256:$src1, i8imm:$src2),
8307 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
8309 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
8311 let neverHasSideEffects = 1, mayStore = 1 in
8312 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
8313 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
8314 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
8317 let Predicates = [HasAVX2] in {
8318 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
8319 (v2i64 (VEXTRACTI128rr
8320 (v4i64 VR256:$src1),
8321 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
8322 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
8323 (v4i32 (VEXTRACTI128rr
8324 (v8i32 VR256:$src1),
8325 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
8326 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
8327 (v8i16 (VEXTRACTI128rr
8328 (v16i16 VR256:$src1),
8329 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
8330 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
8331 (v16i8 (VEXTRACTI128rr
8332 (v32i8 VR256:$src1),
8333 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
8335 def : Pat<(store (v2i64 (vextract128_extract:$ext (v4i64 VR256:$src1),
8336 (iPTR imm))), addr:$dst),
8337 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8338 (EXTRACT_get_vextract128_imm VR128:$ext))>;
8339 def : Pat<(store (v4i32 (vextract128_extract:$ext (v8i32 VR256:$src1),
8340 (iPTR imm))), addr:$dst),
8341 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8342 (EXTRACT_get_vextract128_imm VR128:$ext))>;
8343 def : Pat<(store (v8i16 (vextract128_extract:$ext (v16i16 VR256:$src1),
8344 (iPTR imm))), addr:$dst),
8345 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8346 (EXTRACT_get_vextract128_imm VR128:$ext))>;
8347 def : Pat<(store (v16i8 (vextract128_extract:$ext (v32i8 VR256:$src1),
8348 (iPTR imm))), addr:$dst),
8349 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8350 (EXTRACT_get_vextract128_imm VR128:$ext))>;
8353 //===----------------------------------------------------------------------===//
8354 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
8356 multiclass avx2_pmovmask<string OpcodeStr,
8357 Intrinsic IntLd128, Intrinsic IntLd256,
8358 Intrinsic IntSt128, Intrinsic IntSt256> {
8359 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
8360 (ins VR128:$src1, i128mem:$src2),
8361 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8362 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
8363 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
8364 (ins VR256:$src1, i256mem:$src2),
8365 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8366 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
8368 def mr : AVX28I<0x8e, MRMDestMem, (outs),
8369 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
8370 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8371 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
8372 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
8373 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
8374 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8375 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V, VEX_L;
8378 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
8379 int_x86_avx2_maskload_d,
8380 int_x86_avx2_maskload_d_256,
8381 int_x86_avx2_maskstore_d,
8382 int_x86_avx2_maskstore_d_256>;
8383 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
8384 int_x86_avx2_maskload_q,
8385 int_x86_avx2_maskload_q_256,
8386 int_x86_avx2_maskstore_q,
8387 int_x86_avx2_maskstore_q_256>, VEX_W;
8390 //===----------------------------------------------------------------------===//
8391 // Variable Bit Shifts
8393 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
8394 ValueType vt128, ValueType vt256> {
8395 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
8396 (ins VR128:$src1, VR128:$src2),
8397 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8399 (vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
8401 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
8402 (ins VR128:$src1, i128mem:$src2),
8403 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8405 (vt128 (OpNode VR128:$src1,
8406 (vt128 (bitconvert (memopv2i64 addr:$src2))))))]>,
8408 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
8409 (ins VR256:$src1, VR256:$src2),
8410 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8412 (vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
8414 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
8415 (ins VR256:$src1, i256mem:$src2),
8416 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8418 (vt256 (OpNode VR256:$src1,
8419 (vt256 (bitconvert (memopv4i64 addr:$src2))))))]>,
8423 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
8424 defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", shl, v2i64, v4i64>, VEX_W;
8425 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", srl, v4i32, v8i32>;
8426 defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", srl, v2i64, v4i64>, VEX_W;
8427 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;
8429 //===----------------------------------------------------------------------===//
8430 // VGATHER - GATHER Operations
8431 multiclass avx2_gather<bits<8> opc, string OpcodeStr, RegisterClass RC256,
8432 X86MemOperand memop128, X86MemOperand memop256> {
8433 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst, VR128:$mask_wb),
8434 (ins VR128:$src1, memop128:$src2, VR128:$mask),
8435 !strconcat(OpcodeStr,
8436 "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
8438 def Yrm : AVX28I<opc, MRMSrcMem, (outs RC256:$dst, RC256:$mask_wb),
8439 (ins RC256:$src1, memop256:$src2, RC256:$mask),
8440 !strconcat(OpcodeStr,
8441 "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
8442 []>, VEX_4VOp3, VEX_L;
8445 let mayLoad = 1, Constraints
8446 = "@earlyclobber $dst,@earlyclobber $mask_wb, $src1 = $dst, $mask = $mask_wb"
8448 defm VGATHERDPD : avx2_gather<0x92, "vgatherdpd", VR256, vx64mem, vx64mem>, VEX_W;
8449 defm VGATHERQPD : avx2_gather<0x93, "vgatherqpd", VR256, vx64mem, vy64mem>, VEX_W;
8450 defm VGATHERDPS : avx2_gather<0x92, "vgatherdps", VR256, vx32mem, vy32mem>;
8451 defm VGATHERQPS : avx2_gather<0x93, "vgatherqps", VR128, vx32mem, vy32mem>;
8452 defm VPGATHERDQ : avx2_gather<0x90, "vpgatherdq", VR256, vx64mem, vx64mem>, VEX_W;
8453 defm VPGATHERQQ : avx2_gather<0x91, "vpgatherqq", VR256, vx64mem, vy64mem>, VEX_W;
8454 defm VPGATHERDD : avx2_gather<0x90, "vpgatherdd", VR256, vx32mem, vy32mem>;
8455 defm VPGATHERQD : avx2_gather<0x91, "vpgatherqd", VR128, vx32mem, vy32mem>;