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, $dst|$dst, $src2}",
1815 (int_x86_sse2_cvtsd2ss VR128:$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, $dst|$dst, $src2}",
1821 [(set VR128:$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, $src}",
1917 (int_x86_sse2_cvtps2dq (loadv4f32 addr:$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, $src}",
1927 (int_x86_avx_cvt_ps2dq_256 (loadv8f32 addr:$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, $src}",
1956 (int_x86_sse2_cvtpd2dq (loadv2f64 addr:$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, $src}",
1968 (int_x86_avx_cvt_pd2dq_256 (loadv4f64 addr:$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 (loadv4f32 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 (loadv8f32 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 (loadv2i64 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 (loadv2i64 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 (loadv4f32 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 (loadv4i64 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 (loadv8f32 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 (loadv2f64 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, $src}",
2090 (int_x86_avx_cvtt_pd2dq_256 (loadv4f64 addr:$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:$src)>;
2098 def : Pat<(v4i32 (fp_to_sint (loadv4f64 addr:$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, $src}",
2132 (int_x86_avx_cvt_ps2_pd_256 (loadv4f32 addr:$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, $src}",
2161 (int_x86_avx_cvtdq2_pd_256
2162 (bitconvert (loadv2i64 addr:$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:$src)>;
2184 def : Pat<(v4f64 (sint_to_fp (bc_v4i32 (loadv2i64 addr:$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, $src}",
2203 (int_x86_sse2_cvtpd2ps (loadv2f64 addr:$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, $src}",
2215 (int_x86_avx_cvt_pd2_ps_256 (loadv4f64 addr:$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 (loadv4i64 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 (loadv2f64 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, $src3}",
2534 loadv4f32, SSEPackedSingle>, TB, VEX_4V;
2535 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
2536 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2537 loadv8f32, SSEPackedSingle>, TB, VEX_4V, VEX_L;
2538 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2539 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2540 loadv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2541 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
2542 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2543 loadv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V, VEX_L;
2545 let Constraints = "$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 (loadv2i64 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 (loadv2i64 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 (loadv4i64 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 (loadv4i64 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:$src2))))],
2620 Sched<[WriteShuffleLd, ReadAfterLd]>;
2623 defm VUNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, loadv4f32,
2624 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2625 SSEPackedSingle>, TB, VEX_4V;
2626 defm VUNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, loadv2f64,
2627 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2628 SSEPackedDouble>, TB, OpSize, VEX_4V;
2629 defm VUNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, loadv4f32,
2630 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2631 SSEPackedSingle>, TB, VEX_4V;
2632 defm VUNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, loadv2f64,
2633 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2634 SSEPackedDouble>, TB, OpSize, VEX_4V;
2636 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, X86Unpckh, v8f32, loadv8f32,
2637 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2638 SSEPackedSingle>, TB, VEX_4V, VEX_L;
2639 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, X86Unpckh, v4f64, loadv4f64,
2640 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2641 SSEPackedDouble>, TB, OpSize, VEX_4V, VEX_L;
2642 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, X86Unpckl, v8f32, loadv8f32,
2643 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2644 SSEPackedSingle>, TB, VEX_4V, VEX_L;
2645 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, X86Unpckl, v4f64, loadv4f64,
2646 VR256, f256mem, "unpcklpd\t{$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 (loadv4i64 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 (loadv4i64 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, (loadv4i64 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, (loadv4i64 addr:$src2))),
2679 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
2680 def : Pat<(v4i64 (X86Unpckh VR256:$src1, VR256:$src2)),
2681 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
2684 let Predicates = [HasAVX] in {
2685 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2686 // problem is during lowering, where it's not possible to recognize the load
2687 // fold cause it has two uses through a bitcast. One use disappears at isel
2688 // time and the fold opportunity reappears.
2689 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2690 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2693 let Predicates = [UseSSE2] in {
2694 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2695 // problem is during lowering, where it's not possible to recognize the load
2696 // fold cause it has two uses through a bitcast. One use disappears at isel
2697 // time and the fold opportunity reappears.
2698 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2699 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2702 //===----------------------------------------------------------------------===//
2703 // SSE 1 & 2 - Extract Floating-Point Sign mask
2704 //===----------------------------------------------------------------------===//
2706 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2707 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2709 def rr : PI<0x50, MRMSrcReg, (outs GR32orGR64:$dst), (ins RC:$src),
2710 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2711 [(set GR32orGR64:$dst, (Int RC:$src))], IIC_SSE_MOVMSK, d>,
2712 Sched<[WriteVecLogic]>;
2715 let Predicates = [HasAVX] in {
2716 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2717 "movmskps", SSEPackedSingle>, TB, VEX;
2718 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2719 "movmskpd", SSEPackedDouble>, TB,
2721 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2722 "movmskps", SSEPackedSingle>, TB,
2724 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2725 "movmskpd", SSEPackedDouble>, TB,
2728 def : Pat<(i32 (X86fgetsign FR32:$src)),
2729 (VMOVMSKPSrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
2730 def : Pat<(i64 (X86fgetsign FR32:$src)),
2731 (SUBREG_TO_REG (i64 0),
2732 (VMOVMSKPSrr (COPY_TO_REGCLASS FR32:$src, VR128)), sub_32bit)>;
2733 def : Pat<(i32 (X86fgetsign FR64:$src)),
2734 (VMOVMSKPDrr (COPY_TO_REGCLASS FR64:$src, VR128))>;
2735 def : Pat<(i64 (X86fgetsign FR64:$src)),
2736 (SUBREG_TO_REG (i64 0),
2737 (VMOVMSKPDrr (COPY_TO_REGCLASS FR64:$src, VR128)), sub_32bit)>;
2740 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2741 SSEPackedSingle>, TB;
2742 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2743 SSEPackedDouble>, TB, OpSize;
2745 def : Pat<(i32 (X86fgetsign FR32:$src)),
2746 (MOVMSKPSrr (COPY_TO_REGCLASS FR32:$src, VR128))>,
2747 Requires<[UseSSE1]>;
2748 def : Pat<(i64 (X86fgetsign FR32:$src)),
2749 (SUBREG_TO_REG (i64 0),
2750 (MOVMSKPSrr (COPY_TO_REGCLASS FR32:$src, VR128)), sub_32bit)>,
2751 Requires<[UseSSE1]>;
2752 def : Pat<(i32 (X86fgetsign FR64:$src)),
2753 (MOVMSKPDrr (COPY_TO_REGCLASS FR64:$src, VR128))>,
2754 Requires<[UseSSE2]>;
2755 def : Pat<(i64 (X86fgetsign FR64:$src)),
2756 (SUBREG_TO_REG (i64 0),
2757 (MOVMSKPDrr (COPY_TO_REGCLASS FR64:$src, VR128)), sub_32bit)>,
2758 Requires<[UseSSE2]>;
2760 //===---------------------------------------------------------------------===//
2761 // SSE2 - Packed Integer Logical Instructions
2762 //===---------------------------------------------------------------------===//
2764 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2766 /// PDI_binop_rm - Simple SSE2 binary operator.
2767 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2768 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
2769 X86MemOperand x86memop, OpndItins itins,
2770 bit IsCommutable, bit Is2Addr> {
2771 let isCommutable = IsCommutable in
2772 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
2773 (ins RC:$src1, RC:$src2),
2775 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2776 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2777 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>,
2778 Sched<[itins.Sched]>;
2779 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
2780 (ins RC:$src1, x86memop:$src2),
2782 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2783 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2784 [(set RC:$dst, (OpVT (OpNode RC:$src1,
2785 (bitconvert (memop_frag addr:$src2)))))],
2787 Sched<[itins.Sched.Folded, ReadAfterLd]>;
2789 } // ExeDomain = SSEPackedInt
2791 multiclass PDI_binop_all<bits<8> opc, string OpcodeStr, SDNode Opcode,
2792 ValueType OpVT128, ValueType OpVT256,
2793 OpndItins itins, bit IsCommutable = 0> {
2794 let Predicates = [HasAVX] in
2795 defm V#NAME : PDI_binop_rm<opc, !strconcat("v", OpcodeStr), Opcode, OpVT128,
2796 VR128, loadv2i64, i128mem, itins, IsCommutable, 0>, VEX_4V;
2798 let Constraints = "$src1 = $dst" in
2799 defm NAME : PDI_binop_rm<opc, OpcodeStr, Opcode, OpVT128, VR128,
2800 memopv2i64, i128mem, itins, IsCommutable, 1>;
2802 let Predicates = [HasAVX2] in
2803 defm V#NAME#Y : PDI_binop_rm<opc, !strconcat("v", OpcodeStr), Opcode,
2804 OpVT256, VR256, loadv4i64, i256mem, itins,
2805 IsCommutable, 0>, VEX_4V, VEX_L;
2808 // These are ordered here for pattern ordering requirements with the fp versions
2810 defm PAND : PDI_binop_all<0xDB, "pand", and, v2i64, v4i64, SSE_BIT_ITINS_P, 1>;
2811 defm POR : PDI_binop_all<0xEB, "por", or, v2i64, v4i64, SSE_BIT_ITINS_P, 1>;
2812 defm PXOR : PDI_binop_all<0xEF, "pxor", xor, v2i64, v4i64, SSE_BIT_ITINS_P, 1>;
2813 defm PANDN : PDI_binop_all<0xDF, "pandn", X86andnp, v2i64, v4i64,
2814 SSE_BIT_ITINS_P, 0>;
2816 //===----------------------------------------------------------------------===//
2817 // SSE 1 & 2 - Logical Instructions
2818 //===----------------------------------------------------------------------===//
2820 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2822 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2823 SDNode OpNode, OpndItins itins> {
2824 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2825 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, itins, 0>,
2828 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2829 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, itins, 0>,
2832 let Constraints = "$src1 = $dst" in {
2833 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2834 f32, f128mem, memopfsf32, SSEPackedSingle, itins>,
2837 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2838 f64, f128mem, memopfsf64, SSEPackedDouble, itins>,
2843 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2844 let isCodeGenOnly = 1 in {
2845 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand,
2847 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for,
2849 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor,
2852 let isCommutable = 0 in
2853 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", X86fandn,
2857 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2859 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2861 defm V#NAME#PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2862 !strconcat(OpcodeStr, "ps"), f256mem,
2863 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2864 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2865 (loadv4i64 addr:$src2)))], 0>, TB, VEX_4V, VEX_L;
2867 defm V#NAME#PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2868 !strconcat(OpcodeStr, "pd"), f256mem,
2869 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2870 (bc_v4i64 (v4f64 VR256:$src2))))],
2871 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2872 (loadv4i64 addr:$src2)))], 0>,
2873 TB, OpSize, VEX_4V, VEX_L;
2875 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2876 // are all promoted to v2i64, and the patterns are covered by the int
2877 // version. This is needed in SSE only, because v2i64 isn't supported on
2878 // SSE1, but only on SSE2.
2879 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2880 !strconcat(OpcodeStr, "ps"), f128mem, [],
2881 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2882 (loadv2i64 addr:$src2)))], 0>, TB, VEX_4V;
2884 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2885 !strconcat(OpcodeStr, "pd"), f128mem,
2886 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2887 (bc_v2i64 (v2f64 VR128:$src2))))],
2888 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2889 (loadv2i64 addr:$src2)))], 0>,
2892 let Constraints = "$src1 = $dst" in {
2893 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2894 !strconcat(OpcodeStr, "ps"), f128mem,
2895 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2896 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2897 (memopv2i64 addr:$src2)))]>, TB;
2899 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2900 !strconcat(OpcodeStr, "pd"), f128mem,
2901 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2902 (bc_v2i64 (v2f64 VR128:$src2))))],
2903 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2904 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2908 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2909 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2910 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2911 let isCommutable = 0 in
2912 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2914 //===----------------------------------------------------------------------===//
2915 // SSE 1 & 2 - Arithmetic Instructions
2916 //===----------------------------------------------------------------------===//
2918 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2921 /// In addition, we also have a special variant of the scalar form here to
2922 /// represent the associated intrinsic operation. This form is unlike the
2923 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2924 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2926 /// These three forms can each be reg+reg or reg+mem.
2929 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2931 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr,
2932 SDNode OpNode, SizeItins itins> {
2933 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2934 VR128, v4f32, f128mem, loadv4f32,
2935 SSEPackedSingle, itins.s, 0>, TB, VEX_4V;
2936 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2937 VR128, v2f64, f128mem, loadv2f64,
2938 SSEPackedDouble, itins.d, 0>, TB, OpSize, VEX_4V;
2940 defm V#NAME#PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"),
2941 OpNode, VR256, v8f32, f256mem, loadv8f32,
2942 SSEPackedSingle, itins.s, 0>, TB, VEX_4V, VEX_L;
2943 defm V#NAME#PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"),
2944 OpNode, VR256, v4f64, f256mem, loadv4f64,
2945 SSEPackedDouble, itins.d, 0>, TB, OpSize, VEX_4V, VEX_L;
2947 let Constraints = "$src1 = $dst" in {
2948 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2949 v4f32, f128mem, memopv4f32, SSEPackedSingle,
2951 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2952 v2f64, f128mem, memopv2f64, SSEPackedDouble,
2953 itins.d>, TB, OpSize;
2957 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2959 defm V#NAME#SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2960 OpNode, FR32, f32mem, itins.s, 0>, XS, VEX_4V, VEX_LIG;
2961 defm V#NAME#SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2962 OpNode, FR64, f64mem, itins.d, 0>, XD, VEX_4V, VEX_LIG;
2964 let Constraints = "$src1 = $dst" in {
2965 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2966 OpNode, FR32, f32mem, itins.s>, XS;
2967 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2968 OpNode, FR64, f64mem, itins.d>, XD;
2972 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2974 defm V#NAME#SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2975 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32,
2976 itins.s, 0>, XS, VEX_4V, VEX_LIG;
2977 defm V#NAME#SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2978 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64,
2979 itins.d, 0>, XD, VEX_4V, VEX_LIG;
2981 let Constraints = "$src1 = $dst" in {
2982 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2983 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32,
2985 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2986 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64,
2991 // Binary Arithmetic instructions
2992 defm ADD : basic_sse12_fp_binop_p<0x58, "add", fadd, SSE_ALU_ITINS_P>,
2993 basic_sse12_fp_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S>,
2994 basic_sse12_fp_binop_s_int<0x58, "add", SSE_ALU_ITINS_S>;
2995 defm MUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, SSE_MUL_ITINS_P>,
2996 basic_sse12_fp_binop_s<0x59, "mul", fmul, SSE_MUL_ITINS_S>,
2997 basic_sse12_fp_binop_s_int<0x59, "mul", SSE_MUL_ITINS_S>;
2998 let isCommutable = 0 in {
2999 defm SUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, SSE_ALU_ITINS_P>,
3000 basic_sse12_fp_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S>,
3001 basic_sse12_fp_binop_s_int<0x5C, "sub", SSE_ALU_ITINS_S>;
3002 defm DIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, SSE_DIV_ITINS_P>,
3003 basic_sse12_fp_binop_s<0x5E, "div", fdiv, SSE_DIV_ITINS_S>,
3004 basic_sse12_fp_binop_s_int<0x5E, "div", SSE_DIV_ITINS_S>;
3005 defm MAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, SSE_ALU_ITINS_P>,
3006 basic_sse12_fp_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S>,
3007 basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S>;
3008 defm MIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, SSE_ALU_ITINS_P>,
3009 basic_sse12_fp_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S>,
3010 basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S>;
3013 let isCodeGenOnly = 1 in {
3014 defm MAXC: basic_sse12_fp_binop_p<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_P>,
3015 basic_sse12_fp_binop_s<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_S>;
3016 defm MINC: basic_sse12_fp_binop_p<0x5D, "min", X86fminc, SSE_ALU_ITINS_P>,
3017 basic_sse12_fp_binop_s<0x5D, "min", X86fminc, SSE_ALU_ITINS_S>;
3020 // Patterns used to select SSE scalar fp arithmetic instructions from
3021 // a scalar fp operation followed by a blend.
3023 // These patterns know, for example, how to select an ADDSS from a
3024 // float add plus vector insert.
3026 // The effect is that the backend no longer emits unnecessary vector
3027 // insert instructions immediately after SSE scalar fp instructions
3028 // like addss or mulss.
3030 // For example, given the following code:
3031 // __m128 foo(__m128 A, __m128 B) {
3036 // previously we generated:
3037 // addss %xmm0, %xmm1
3038 // movss %xmm1, %xmm0
3041 // addss %xmm1, %xmm0
3043 def : Pat<(v4f32 (X86Movss (v4f32 VR128:$dst), (v4f32 (scalar_to_vector (fadd
3044 (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
3046 (ADDSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
3047 def : Pat<(v4f32 (X86Movss (v4f32 VR128:$dst), (v4f32 (scalar_to_vector (fsub
3048 (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
3050 (SUBSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
3051 def : Pat<(v4f32 (X86Movss (v4f32 VR128:$dst), (v4f32 (scalar_to_vector (fmul
3052 (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
3054 (MULSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
3055 def : Pat<(v4f32 (X86Movss (v4f32 VR128:$dst), (v4f32 (scalar_to_vector (fdiv
3056 (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
3058 (DIVSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
3060 let Predicates = [HasSSE2] in {
3061 // SSE2 patterns to select scalar double-precision fp arithmetic instructions
3063 def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fadd
3064 (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
3066 (ADDSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
3067 def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fsub
3068 (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
3070 (SUBSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
3071 def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fmul
3072 (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
3074 (MULSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
3075 def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fdiv
3076 (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
3078 (DIVSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
3081 let Predicates = [UseSSE41] in {
3082 // If the subtarget has SSE4.1 but not AVX, the vector insert
3083 // instruction is lowered into a X86insrtps rather than a X86Movss.
3084 // When selecting SSE scalar single-precision fp arithmetic instructions,
3085 // make sure that we correctly match the X86insrtps.
3087 def : Pat<(v4f32 (X86insrtps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
3088 (fadd (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
3089 FR32:$src))), (iPTR 0))),
3090 (ADDSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
3091 def : Pat<(v4f32 (X86insrtps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
3092 (fsub (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
3093 FR32:$src))), (iPTR 0))),
3094 (SUBSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
3095 def : Pat<(v4f32 (X86insrtps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
3096 (fmul (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
3097 FR32:$src))), (iPTR 0))),
3098 (MULSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
3099 def : Pat<(v4f32 (X86insrtps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
3100 (fdiv (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
3101 FR32:$src))), (iPTR 0))),
3102 (DIVSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
3105 let AddedComplexity = 20, Predicates = [HasAVX] in {
3106 // The following patterns select AVX Scalar single/double precision fp
3107 // arithmetic instructions.
3108 // The 'AddedComplexity' is required to give them higher priority over
3109 // the equivalent SSE/SSE2 patterns.
3111 def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fadd
3112 (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
3114 (VADDSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
3115 def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fsub
3116 (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
3118 (VSUBSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
3119 def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fmul
3120 (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
3122 (VMULSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
3123 def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fdiv
3124 (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
3126 (VDIVSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
3127 def : Pat<(v4f32 (X86insrtps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
3128 (fadd (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
3129 FR32:$src))), (iPTR 0))),
3130 (VADDSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
3131 def : Pat<(v4f32 (X86insrtps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
3132 (fsub (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
3133 FR32:$src))), (iPTR 0))),
3134 (VSUBSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
3135 def : Pat<(v4f32 (X86insrtps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
3136 (fmul (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
3137 FR32:$src))), (iPTR 0))),
3138 (VMULSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
3139 def : Pat<(v4f32 (X86insrtps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
3140 (fdiv (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
3141 FR32:$src))), (iPTR 0))),
3142 (VDIVSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
3146 /// In addition, we also have a special variant of the scalar form here to
3147 /// represent the associated intrinsic operation. This form is unlike the
3148 /// plain scalar form, in that it takes an entire vector (instead of a
3149 /// scalar) and leaves the top elements undefined.
3151 /// And, we have a special variant form for a full-vector intrinsic form.
3153 let Sched = WriteFSqrt in {
3154 def SSE_SQRTPS : OpndItins<
3155 IIC_SSE_SQRTPS_RR, IIC_SSE_SQRTPS_RM
3158 def SSE_SQRTSS : OpndItins<
3159 IIC_SSE_SQRTSS_RR, IIC_SSE_SQRTSS_RM
3162 def SSE_SQRTPD : OpndItins<
3163 IIC_SSE_SQRTPD_RR, IIC_SSE_SQRTPD_RM
3166 def SSE_SQRTSD : OpndItins<
3167 IIC_SSE_SQRTSD_RR, IIC_SSE_SQRTSD_RM
3171 let Sched = WriteFRcp in {
3172 def SSE_RCPP : OpndItins<
3173 IIC_SSE_RCPP_RR, IIC_SSE_RCPP_RM
3176 def SSE_RCPS : OpndItins<
3177 IIC_SSE_RCPS_RR, IIC_SSE_RCPS_RM
3181 /// sse1_fp_unop_s - SSE1 unops in scalar form.
3182 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
3183 SDNode OpNode, Intrinsic F32Int, OpndItins itins> {
3184 let Predicates = [HasAVX], hasSideEffects = 0 in {
3185 def V#NAME#SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst),
3186 (ins FR32:$src1, FR32:$src2),
3187 !strconcat("v", OpcodeStr,
3188 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3189 []>, VEX_4V, VEX_LIG, Sched<[itins.Sched]>;
3190 let mayLoad = 1 in {
3191 def V#NAME#SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst),
3192 (ins FR32:$src1,f32mem:$src2),
3193 !strconcat("v", OpcodeStr,
3194 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3195 []>, VEX_4V, VEX_LIG,
3196 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3197 def V#NAME#SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3198 (ins VR128:$src1, ssmem:$src2),
3199 !strconcat("v", OpcodeStr,
3200 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3201 []>, VEX_4V, VEX_LIG,
3202 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3206 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
3207 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3208 [(set FR32:$dst, (OpNode FR32:$src))]>, Sched<[itins.Sched]>;
3209 // For scalar unary operations, fold a load into the operation
3210 // only in OptForSize mode. It eliminates an instruction, but it also
3211 // eliminates a whole-register clobber (the load), so it introduces a
3212 // partial register update condition.
3213 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
3214 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3215 [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
3216 Requires<[UseSSE1, OptForSize]>, Sched<[itins.Sched.Folded]>;
3217 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3218 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3219 [(set VR128:$dst, (F32Int VR128:$src))], itins.rr>,
3220 Sched<[itins.Sched]>;
3221 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
3222 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3223 [(set VR128:$dst, (F32Int sse_load_f32:$src))], itins.rm>,
3224 Sched<[itins.Sched.Folded]>;
3227 /// sse1_fp_unop_s_rw - SSE1 unops where vector form has a read-write operand.
3228 multiclass sse1_fp_unop_rw<bits<8> opc, string OpcodeStr, SDNode OpNode,
3230 let Predicates = [HasAVX], hasSideEffects = 0 in {
3231 def V#NAME#SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst),
3232 (ins FR32:$src1, FR32:$src2),
3233 !strconcat("v", OpcodeStr,
3234 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3235 []>, VEX_4V, VEX_LIG, Sched<[itins.Sched]>;
3236 let mayLoad = 1 in {
3237 def V#NAME#SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst),
3238 (ins FR32:$src1,f32mem:$src2),
3239 !strconcat("v", OpcodeStr,
3240 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3241 []>, VEX_4V, VEX_LIG,
3242 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3243 def V#NAME#SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3244 (ins VR128:$src1, ssmem:$src2),
3245 !strconcat("v", OpcodeStr,
3246 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3247 []>, VEX_4V, VEX_LIG,
3248 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3252 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
3253 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3254 [(set FR32:$dst, (OpNode FR32:$src))]>, Sched<[itins.Sched]>;
3255 // For scalar unary operations, fold a load into the operation
3256 // only in OptForSize mode. It eliminates an instruction, but it also
3257 // eliminates a whole-register clobber (the load), so it introduces a
3258 // partial register update condition.
3259 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
3260 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3261 [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
3262 Requires<[UseSSE1, OptForSize]>, Sched<[itins.Sched.Folded]>;
3263 let Constraints = "$src1 = $dst" in {
3264 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst),
3265 (ins VR128:$src1, VR128:$src2),
3266 !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
3267 [], itins.rr>, Sched<[itins.Sched]>;
3268 let mayLoad = 1, hasSideEffects = 0 in
3269 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3270 (ins VR128:$src1, ssmem:$src2),
3271 !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
3272 [], itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
3276 /// sse1_fp_unop_p - SSE1 unops in packed form.
3277 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
3279 let Predicates = [HasAVX] in {
3280 def V#NAME#PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3281 !strconcat("v", OpcodeStr,
3282 "ps\t{$src, $dst|$dst, $src}"),
3283 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))],
3284 itins.rr>, VEX, Sched<[itins.Sched]>;
3285 def V#NAME#PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3286 !strconcat("v", OpcodeStr,
3287 "ps\t{$src, $dst|$dst, $src}"),
3288 [(set VR128:$dst, (OpNode (loadv4f32 addr:$src)))],
3289 itins.rm>, VEX, Sched<[itins.Sched.Folded]>;
3290 def V#NAME#PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3291 !strconcat("v", OpcodeStr,
3292 "ps\t{$src, $dst|$dst, $src}"),
3293 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))],
3294 itins.rr>, VEX, VEX_L, Sched<[itins.Sched]>;
3295 def V#NAME#PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3296 !strconcat("v", OpcodeStr,
3297 "ps\t{$src, $dst|$dst, $src}"),
3298 [(set VR256:$dst, (OpNode (loadv8f32 addr:$src)))],
3299 itins.rm>, VEX, VEX_L, Sched<[itins.Sched.Folded]>;
3302 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3303 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3304 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))], itins.rr>,
3305 Sched<[itins.Sched]>;
3306 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3307 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3308 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))], itins.rm>,
3309 Sched<[itins.Sched.Folded]>;
3312 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
3313 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3314 Intrinsic V4F32Int, Intrinsic V8F32Int,
3316 let Predicates = [HasAVX] in {
3317 def V#NAME#PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3318 !strconcat("v", OpcodeStr,
3319 "ps\t{$src, $dst|$dst, $src}"),
3320 [(set VR128:$dst, (V4F32Int VR128:$src))],
3321 itins.rr>, VEX, Sched<[itins.Sched]>;
3322 def V#NAME#PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3323 !strconcat("v", OpcodeStr,
3324 "ps\t{$src, $dst|$dst, $src}"),
3325 [(set VR128:$dst, (V4F32Int (loadv4f32 addr:$src)))],
3326 itins.rm>, VEX, Sched<[itins.Sched.Folded]>;
3327 def V#NAME#PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3328 !strconcat("v", OpcodeStr,
3329 "ps\t{$src, $dst|$dst, $src}"),
3330 [(set VR256:$dst, (V8F32Int VR256:$src))],
3331 itins.rr>, VEX, VEX_L, Sched<[itins.Sched]>;
3332 def V#NAME#PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst),
3334 !strconcat("v", OpcodeStr,
3335 "ps\t{$src, $dst|$dst, $src}"),
3336 [(set VR256:$dst, (V8F32Int (loadv8f32 addr:$src)))],
3337 itins.rm>, VEX, VEX_L, Sched<[itins.Sched.Folded]>;
3340 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3341 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3342 [(set VR128:$dst, (V4F32Int VR128:$src))],
3343 itins.rr>, Sched<[itins.Sched]>;
3344 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3345 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3346 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))],
3347 itins.rm>, Sched<[itins.Sched.Folded]>;
3350 /// sse2_fp_unop_s - SSE2 unops in scalar form.
3351 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
3352 SDNode OpNode, Intrinsic F64Int, OpndItins itins> {
3353 let Predicates = [HasAVX], hasSideEffects = 0 in {
3354 def V#NAME#SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst),
3355 (ins FR64:$src1, FR64:$src2),
3356 !strconcat("v", OpcodeStr,
3357 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3358 []>, VEX_4V, VEX_LIG, Sched<[itins.Sched]>;
3359 let mayLoad = 1 in {
3360 def V#NAME#SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst),
3361 (ins FR64:$src1,f64mem:$src2),
3362 !strconcat("v", OpcodeStr,
3363 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3364 []>, VEX_4V, VEX_LIG,
3365 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3366 def V#NAME#SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
3367 (ins VR128:$src1, sdmem:$src2),
3368 !strconcat("v", OpcodeStr,
3369 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3370 []>, VEX_4V, VEX_LIG,
3371 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3375 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
3376 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3377 [(set FR64:$dst, (OpNode FR64:$src))], itins.rr>,
3378 Sched<[itins.Sched]>;
3379 // See the comments in sse1_fp_unop_s for why this is OptForSize.
3380 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
3381 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3382 [(set FR64:$dst, (OpNode (load addr:$src)))], itins.rm>, XD,
3383 Requires<[UseSSE2, OptForSize]>, Sched<[itins.Sched.Folded]>;
3384 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3385 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3386 [(set VR128:$dst, (F64Int VR128:$src))], itins.rr>,
3387 Sched<[itins.Sched]>;
3388 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
3389 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3390 [(set VR128:$dst, (F64Int sse_load_f64:$src))], itins.rm>,
3391 Sched<[itins.Sched.Folded]>;
3394 /// sse2_fp_unop_p - SSE2 unops in vector forms.
3395 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
3396 SDNode OpNode, OpndItins itins> {
3397 let Predicates = [HasAVX] in {
3398 def V#NAME#PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3399 !strconcat("v", OpcodeStr,
3400 "pd\t{$src, $dst|$dst, $src}"),
3401 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))],
3402 itins.rr>, VEX, Sched<[itins.Sched]>;
3403 def V#NAME#PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3404 !strconcat("v", OpcodeStr,
3405 "pd\t{$src, $dst|$dst, $src}"),
3406 [(set VR128:$dst, (OpNode (loadv2f64 addr:$src)))],
3407 itins.rm>, VEX, Sched<[itins.Sched.Folded]>;
3408 def V#NAME#PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3409 !strconcat("v", OpcodeStr,
3410 "pd\t{$src, $dst|$dst, $src}"),
3411 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))],
3412 itins.rr>, VEX, VEX_L, Sched<[itins.Sched]>;
3413 def V#NAME#PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3414 !strconcat("v", OpcodeStr,
3415 "pd\t{$src, $dst|$dst, $src}"),
3416 [(set VR256:$dst, (OpNode (loadv4f64 addr:$src)))],
3417 itins.rm>, VEX, VEX_L, Sched<[itins.Sched.Folded]>;
3420 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3421 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3422 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))], itins.rr>,
3423 Sched<[itins.Sched]>;
3424 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3425 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3426 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))], itins.rm>,
3427 Sched<[itins.Sched.Folded]>;
3431 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss,
3433 sse1_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTPS>,
3434 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd,
3436 sse2_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTPD>;
3438 // Reciprocal approximations. Note that these typically require refinement
3439 // in order to obtain suitable precision.
3440 defm RSQRT : sse1_fp_unop_rw<0x52, "rsqrt", X86frsqrt, SSE_SQRTSS>,
3441 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, SSE_SQRTPS>,
3442 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps,
3443 int_x86_avx_rsqrt_ps_256, SSE_SQRTPS>;
3444 defm RCP : sse1_fp_unop_rw<0x53, "rcp", X86frcp, SSE_RCPS>,
3445 sse1_fp_unop_p<0x53, "rcp", X86frcp, SSE_RCPP>,
3446 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps,
3447 int_x86_avx_rcp_ps_256, SSE_RCPP>;
3449 let Predicates = [UseAVX] in {
3450 def : Pat<(f32 (fsqrt FR32:$src)),
3451 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3452 def : Pat<(f32 (fsqrt (load addr:$src))),
3453 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3454 Requires<[HasAVX, OptForSize]>;
3455 def : Pat<(f64 (fsqrt FR64:$src)),
3456 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
3457 def : Pat<(f64 (fsqrt (load addr:$src))),
3458 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
3459 Requires<[HasAVX, OptForSize]>;
3461 def : Pat<(f32 (X86frsqrt FR32:$src)),
3462 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3463 def : Pat<(f32 (X86frsqrt (load addr:$src))),
3464 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3465 Requires<[HasAVX, OptForSize]>;
3467 def : Pat<(f32 (X86frcp FR32:$src)),
3468 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3469 def : Pat<(f32 (X86frcp (load addr:$src))),
3470 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3471 Requires<[HasAVX, OptForSize]>;
3473 let Predicates = [UseAVX] in {
3474 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
3475 (COPY_TO_REGCLASS (VSQRTSSr (f32 (IMPLICIT_DEF)),
3476 (COPY_TO_REGCLASS VR128:$src, FR32)),
3478 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
3479 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3481 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
3482 (COPY_TO_REGCLASS (VSQRTSDr (f64 (IMPLICIT_DEF)),
3483 (COPY_TO_REGCLASS VR128:$src, FR64)),
3485 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
3486 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
3489 let Predicates = [HasAVX] in {
3490 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3491 (COPY_TO_REGCLASS (VRSQRTSSr (f32 (IMPLICIT_DEF)),
3492 (COPY_TO_REGCLASS VR128:$src, FR32)),
3494 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
3495 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3497 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3498 (COPY_TO_REGCLASS (VRCPSSr (f32 (IMPLICIT_DEF)),
3499 (COPY_TO_REGCLASS VR128:$src, FR32)),
3501 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
3502 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3505 // Reciprocal approximations. Note that these typically require refinement
3506 // in order to obtain suitable precision.
3507 let Predicates = [UseSSE1] in {
3508 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3509 (RSQRTSSr_Int VR128:$src, VR128:$src)>;
3510 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3511 (RCPSSr_Int VR128:$src, VR128:$src)>;
3514 // There is no f64 version of the reciprocal approximation instructions.
3516 //===----------------------------------------------------------------------===//
3517 // SSE 1 & 2 - Non-temporal stores
3518 //===----------------------------------------------------------------------===//
3520 let AddedComplexity = 400 in { // Prefer non-temporal versions
3521 let SchedRW = [WriteStore] in {
3522 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3523 (ins f128mem:$dst, VR128:$src),
3524 "movntps\t{$src, $dst|$dst, $src}",
3525 [(alignednontemporalstore (v4f32 VR128:$src),
3527 IIC_SSE_MOVNT>, VEX;
3528 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3529 (ins f128mem:$dst, VR128:$src),
3530 "movntpd\t{$src, $dst|$dst, $src}",
3531 [(alignednontemporalstore (v2f64 VR128:$src),
3533 IIC_SSE_MOVNT>, VEX;
3535 let ExeDomain = SSEPackedInt in
3536 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3537 (ins f128mem:$dst, VR128:$src),
3538 "movntdq\t{$src, $dst|$dst, $src}",
3539 [(alignednontemporalstore (v2i64 VR128:$src),
3541 IIC_SSE_MOVNT>, VEX;
3543 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3544 (ins f256mem:$dst, VR256:$src),
3545 "movntps\t{$src, $dst|$dst, $src}",
3546 [(alignednontemporalstore (v8f32 VR256:$src),
3548 IIC_SSE_MOVNT>, VEX, VEX_L;
3549 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3550 (ins f256mem:$dst, VR256:$src),
3551 "movntpd\t{$src, $dst|$dst, $src}",
3552 [(alignednontemporalstore (v4f64 VR256:$src),
3554 IIC_SSE_MOVNT>, VEX, VEX_L;
3555 let ExeDomain = SSEPackedInt in
3556 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3557 (ins f256mem:$dst, VR256:$src),
3558 "movntdq\t{$src, $dst|$dst, $src}",
3559 [(alignednontemporalstore (v4i64 VR256:$src),
3561 IIC_SSE_MOVNT>, VEX, VEX_L;
3563 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3564 "movntps\t{$src, $dst|$dst, $src}",
3565 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)],
3567 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3568 "movntpd\t{$src, $dst|$dst, $src}",
3569 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)],
3572 let ExeDomain = SSEPackedInt in
3573 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3574 "movntdq\t{$src, $dst|$dst, $src}",
3575 [(alignednontemporalstore (v2i64 VR128:$src), addr:$dst)],
3578 // There is no AVX form for instructions below this point
3579 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3580 "movnti{l}\t{$src, $dst|$dst, $src}",
3581 [(nontemporalstore (i32 GR32:$src), addr:$dst)],
3583 TB, Requires<[HasSSE2]>;
3584 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3585 "movnti{q}\t{$src, $dst|$dst, $src}",
3586 [(nontemporalstore (i64 GR64:$src), addr:$dst)],
3588 TB, Requires<[HasSSE2]>;
3589 } // SchedRW = [WriteStore]
3591 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3592 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3594 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3595 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[UseSSE2]>;
3596 } // AddedComplexity
3598 //===----------------------------------------------------------------------===//
3599 // SSE 1 & 2 - Prefetch and memory fence
3600 //===----------------------------------------------------------------------===//
3602 // Prefetch intrinsic.
3603 let Predicates = [HasSSE1], SchedRW = [WriteLoad] in {
3604 def PREFETCHT0 : I<0x18, MRM1m, (outs), (ins i8mem:$src),
3605 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))],
3606 IIC_SSE_PREFETCH>, TB;
3607 def PREFETCHT1 : I<0x18, MRM2m, (outs), (ins i8mem:$src),
3608 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))],
3609 IIC_SSE_PREFETCH>, TB;
3610 def PREFETCHT2 : I<0x18, MRM3m, (outs), (ins i8mem:$src),
3611 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))],
3612 IIC_SSE_PREFETCH>, TB;
3613 def PREFETCHNTA : I<0x18, MRM0m, (outs), (ins i8mem:$src),
3614 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))],
3615 IIC_SSE_PREFETCH>, TB;
3618 // FIXME: How should these memory instructions be modeled?
3619 let SchedRW = [WriteLoad] in {
3621 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3622 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)],
3623 IIC_SSE_PREFETCH>, TB, Requires<[HasSSE2]>;
3625 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3626 // was introduced with SSE2, it's backward compatible.
3627 def PAUSE : I<0x90, RawFrm, (outs), (ins),
3628 "pause", [(int_x86_sse2_pause)], IIC_SSE_PAUSE>,
3629 REP, Requires<[HasSSE2]>;
3631 // Load, store, and memory fence
3632 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3633 "sfence", [(int_x86_sse_sfence)], IIC_SSE_SFENCE>,
3634 TB, Requires<[HasSSE1]>;
3635 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3636 "lfence", [(int_x86_sse2_lfence)], IIC_SSE_LFENCE>,
3637 TB, Requires<[HasSSE2]>;
3638 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3639 "mfence", [(int_x86_sse2_mfence)], IIC_SSE_MFENCE>,
3640 TB, Requires<[HasSSE2]>;
3643 def : Pat<(X86SFence), (SFENCE)>;
3644 def : Pat<(X86LFence), (LFENCE)>;
3645 def : Pat<(X86MFence), (MFENCE)>;
3647 //===----------------------------------------------------------------------===//
3648 // SSE 1 & 2 - Load/Store XCSR register
3649 //===----------------------------------------------------------------------===//
3651 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3652 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3653 IIC_SSE_LDMXCSR>, VEX, Sched<[WriteLoad]>;
3654 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3655 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3656 IIC_SSE_STMXCSR>, VEX, Sched<[WriteStore]>;
3658 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3659 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3660 IIC_SSE_LDMXCSR>, Sched<[WriteLoad]>;
3661 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3662 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3663 IIC_SSE_STMXCSR>, Sched<[WriteStore]>;
3665 //===---------------------------------------------------------------------===//
3666 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3667 //===---------------------------------------------------------------------===//
3669 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3671 let neverHasSideEffects = 1, SchedRW = [WriteMove] in {
3672 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3673 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3675 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3676 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3678 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3679 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3681 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3682 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3687 let isCodeGenOnly = 1, hasSideEffects = 0, SchedRW = [WriteMove] in {
3688 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3689 "movdqa\t{$src, $dst|$dst, $src}", [],
3692 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3693 "movdqa\t{$src, $dst|$dst, $src}", [],
3694 IIC_SSE_MOVA_P_RR>, VEX, VEX_L;
3695 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3696 "movdqu\t{$src, $dst|$dst, $src}", [],
3699 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3700 "movdqu\t{$src, $dst|$dst, $src}", [],
3701 IIC_SSE_MOVU_P_RR>, VEX, VEX_L;
3704 let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1,
3705 neverHasSideEffects = 1, SchedRW = [WriteLoad] in {
3706 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3707 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3709 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3710 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3712 let Predicates = [HasAVX] in {
3713 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3714 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3716 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3717 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3722 let mayStore = 1, neverHasSideEffects = 1, SchedRW = [WriteStore] in {
3723 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3724 (ins i128mem:$dst, VR128:$src),
3725 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3727 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3728 (ins i256mem:$dst, VR256:$src),
3729 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3731 let Predicates = [HasAVX] in {
3732 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3733 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3735 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3736 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3741 let SchedRW = [WriteMove] in {
3742 let neverHasSideEffects = 1 in
3743 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3744 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>;
3746 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3747 "movdqu\t{$src, $dst|$dst, $src}",
3748 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>;
3751 let isCodeGenOnly = 1, hasSideEffects = 0 in {
3752 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3753 "movdqa\t{$src, $dst|$dst, $src}", [],
3756 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3757 "movdqu\t{$src, $dst|$dst, $src}",
3758 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>;
3762 let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1,
3763 neverHasSideEffects = 1, SchedRW = [WriteLoad] in {
3764 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3765 "movdqa\t{$src, $dst|$dst, $src}",
3766 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/],
3768 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3769 "movdqu\t{$src, $dst|$dst, $src}",
3770 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/],
3772 XS, Requires<[UseSSE2]>;
3775 let mayStore = 1, neverHasSideEffects = 1, SchedRW = [WriteStore] in {
3776 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3777 "movdqa\t{$src, $dst|$dst, $src}",
3778 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/],
3780 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3781 "movdqu\t{$src, $dst|$dst, $src}",
3782 [/*(store (v2i64 VR128:$src), addr:$dst)*/],
3784 XS, Requires<[UseSSE2]>;
3787 } // ExeDomain = SSEPackedInt
3789 let Predicates = [HasAVX] in {
3790 def : Pat<(int_x86_sse2_storeu_dq addr:$dst, VR128:$src),
3791 (VMOVDQUmr addr:$dst, VR128:$src)>;
3792 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3793 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3795 let Predicates = [UseSSE2] in
3796 def : Pat<(int_x86_sse2_storeu_dq addr:$dst, VR128:$src),
3797 (MOVDQUmr addr:$dst, VR128:$src)>;
3799 //===---------------------------------------------------------------------===//
3800 // SSE2 - Packed Integer Arithmetic Instructions
3801 //===---------------------------------------------------------------------===//
3803 let Sched = WriteVecIMul in
3804 def SSE_PMADD : OpndItins<
3805 IIC_SSE_PMADD, IIC_SSE_PMADD
3808 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3810 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3811 RegisterClass RC, PatFrag memop_frag,
3812 X86MemOperand x86memop,
3814 bit IsCommutable = 0,
3816 let isCommutable = IsCommutable in
3817 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3818 (ins RC:$src1, RC:$src2),
3820 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3821 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3822 [(set RC:$dst, (IntId RC:$src1, RC:$src2))], itins.rr>,
3823 Sched<[itins.Sched]>;
3824 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3825 (ins RC:$src1, x86memop:$src2),
3827 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3828 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3829 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))],
3830 itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
3833 multiclass PDI_binop_all_int<bits<8> opc, string OpcodeStr, Intrinsic IntId128,
3834 Intrinsic IntId256, OpndItins itins,
3835 bit IsCommutable = 0> {
3836 let Predicates = [HasAVX] in
3837 defm V#NAME : PDI_binop_rm_int<opc, !strconcat("v", OpcodeStr), IntId128,
3838 VR128, loadv2i64, i128mem, itins,
3839 IsCommutable, 0>, VEX_4V;
3841 let Constraints = "$src1 = $dst" in
3842 defm NAME : PDI_binop_rm_int<opc, OpcodeStr, IntId128, VR128, memopv2i64,
3843 i128mem, itins, IsCommutable, 1>;
3845 let Predicates = [HasAVX2] in
3846 defm V#NAME#Y : PDI_binop_rm_int<opc, !strconcat("v", OpcodeStr), IntId256,
3847 VR256, loadv4i64, i256mem, itins,
3848 IsCommutable, 0>, VEX_4V, VEX_L;
3851 multiclass PDI_binop_rmi<bits<8> opc, bits<8> opc2, Format ImmForm,
3852 string OpcodeStr, SDNode OpNode,
3853 SDNode OpNode2, RegisterClass RC,
3854 ValueType DstVT, ValueType SrcVT, PatFrag bc_frag,
3855 ShiftOpndItins itins,
3857 // src2 is always 128-bit
3858 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3859 (ins RC:$src1, VR128:$src2),
3861 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3862 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3863 [(set RC:$dst, (DstVT (OpNode RC:$src1, (SrcVT VR128:$src2))))],
3864 itins.rr>, Sched<[WriteVecShift]>;
3865 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3866 (ins RC:$src1, i128mem:$src2),
3868 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3869 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3870 [(set RC:$dst, (DstVT (OpNode RC:$src1,
3871 (bc_frag (memopv2i64 addr:$src2)))))], itins.rm>,
3872 Sched<[WriteVecShiftLd, ReadAfterLd]>;
3873 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3874 (ins RC:$src1, i8imm:$src2),
3876 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3877 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3878 [(set RC:$dst, (DstVT (OpNode2 RC:$src1, (i8 imm:$src2))))], itins.ri>,
3879 Sched<[WriteVecShift]>;
3882 /// PDI_binop_rm2 - Simple SSE2 binary operator with different src and dst types
3883 multiclass PDI_binop_rm2<bits<8> opc, string OpcodeStr, SDNode OpNode,
3884 ValueType DstVT, ValueType SrcVT, RegisterClass RC,
3885 PatFrag memop_frag, X86MemOperand x86memop,
3887 bit IsCommutable = 0, bit Is2Addr = 1> {
3888 let isCommutable = IsCommutable in
3889 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3890 (ins RC:$src1, RC:$src2),
3892 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3893 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3894 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1), RC:$src2)))]>,
3895 Sched<[itins.Sched]>;
3896 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3897 (ins RC:$src1, x86memop:$src2),
3899 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3900 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3901 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1),
3902 (bitconvert (memop_frag addr:$src2)))))]>,
3903 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3905 } // ExeDomain = SSEPackedInt
3907 defm PADDB : PDI_binop_all<0xFC, "paddb", add, v16i8, v32i8,
3908 SSE_INTALU_ITINS_P, 1>;
3909 defm PADDW : PDI_binop_all<0xFD, "paddw", add, v8i16, v16i16,
3910 SSE_INTALU_ITINS_P, 1>;
3911 defm PADDD : PDI_binop_all<0xFE, "paddd", add, v4i32, v8i32,
3912 SSE_INTALU_ITINS_P, 1>;
3913 defm PADDQ : PDI_binop_all<0xD4, "paddq", add, v2i64, v4i64,
3914 SSE_INTALUQ_ITINS_P, 1>;
3915 defm PMULLW : PDI_binop_all<0xD5, "pmullw", mul, v8i16, v16i16,
3916 SSE_INTMUL_ITINS_P, 1>;
3917 defm PSUBB : PDI_binop_all<0xF8, "psubb", sub, v16i8, v32i8,
3918 SSE_INTALU_ITINS_P, 0>;
3919 defm PSUBW : PDI_binop_all<0xF9, "psubw", sub, v8i16, v16i16,
3920 SSE_INTALU_ITINS_P, 0>;
3921 defm PSUBD : PDI_binop_all<0xFA, "psubd", sub, v4i32, v8i32,
3922 SSE_INTALU_ITINS_P, 0>;
3923 defm PSUBQ : PDI_binop_all<0xFB, "psubq", sub, v2i64, v4i64,
3924 SSE_INTALUQ_ITINS_P, 0>;
3925 defm PSUBUSB : PDI_binop_all<0xD8, "psubusb", X86subus, v16i8, v32i8,
3926 SSE_INTALU_ITINS_P, 0>;
3927 defm PSUBUSW : PDI_binop_all<0xD9, "psubusw", X86subus, v8i16, v16i16,
3928 SSE_INTALU_ITINS_P, 0>;
3929 defm PMINUB : PDI_binop_all<0xDA, "pminub", X86umin, v16i8, v32i8,
3930 SSE_INTALU_ITINS_P, 1>;
3931 defm PMINSW : PDI_binop_all<0xEA, "pminsw", X86smin, v8i16, v16i16,
3932 SSE_INTALU_ITINS_P, 1>;
3933 defm PMAXUB : PDI_binop_all<0xDE, "pmaxub", X86umax, v16i8, v32i8,
3934 SSE_INTALU_ITINS_P, 1>;
3935 defm PMAXSW : PDI_binop_all<0xEE, "pmaxsw", X86smax, v8i16, v16i16,
3936 SSE_INTALU_ITINS_P, 1>;
3939 defm PSUBSB : PDI_binop_all_int<0xE8, "psubsb", int_x86_sse2_psubs_b,
3940 int_x86_avx2_psubs_b, SSE_INTALU_ITINS_P, 0>;
3941 defm PSUBSW : PDI_binop_all_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
3942 int_x86_avx2_psubs_w, SSE_INTALU_ITINS_P, 0>;
3943 defm PADDSB : PDI_binop_all_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
3944 int_x86_avx2_padds_b, SSE_INTALU_ITINS_P, 1>;
3945 defm PADDSW : PDI_binop_all_int<0xED, "paddsw" , int_x86_sse2_padds_w,
3946 int_x86_avx2_padds_w, SSE_INTALU_ITINS_P, 1>;
3947 defm PADDUSB : PDI_binop_all_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
3948 int_x86_avx2_paddus_b, SSE_INTALU_ITINS_P, 1>;
3949 defm PADDUSW : PDI_binop_all_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
3950 int_x86_avx2_paddus_w, SSE_INTALU_ITINS_P, 1>;
3951 defm PMULHUW : PDI_binop_all_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
3952 int_x86_avx2_pmulhu_w, SSE_INTMUL_ITINS_P, 1>;
3953 defm PMULHW : PDI_binop_all_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
3954 int_x86_avx2_pmulh_w, SSE_INTMUL_ITINS_P, 1>;
3955 defm PMADDWD : PDI_binop_all_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
3956 int_x86_avx2_pmadd_wd, SSE_PMADD, 1>;
3957 defm PAVGB : PDI_binop_all_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
3958 int_x86_avx2_pavg_b, SSE_INTALU_ITINS_P, 1>;
3959 defm PAVGW : PDI_binop_all_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
3960 int_x86_avx2_pavg_w, SSE_INTALU_ITINS_P, 1>;
3961 defm PSADBW : PDI_binop_all_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
3962 int_x86_avx2_psad_bw, SSE_PMADD, 1>;
3964 let Predicates = [HasAVX] in
3965 defm VPMULUDQ : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v2i64, v4i32, VR128,
3966 loadv2i64, i128mem, SSE_INTMUL_ITINS_P, 1, 0>,
3968 let Predicates = [HasAVX2] in
3969 defm VPMULUDQY : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v4i64, v8i32,
3970 VR256, loadv4i64, i256mem,
3971 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V, VEX_L;
3972 let Constraints = "$src1 = $dst" in
3973 defm PMULUDQ : PDI_binop_rm2<0xF4, "pmuludq", X86pmuludq, v2i64, v4i32, VR128,
3974 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1>;
3976 //===---------------------------------------------------------------------===//
3977 // SSE2 - Packed Integer Logical Instructions
3978 //===---------------------------------------------------------------------===//
3980 let Predicates = [HasAVX] in {
3981 defm VPSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3982 VR128, v8i16, v8i16, bc_v8i16,
3983 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3984 defm VPSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3985 VR128, v4i32, v4i32, bc_v4i32,
3986 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3987 defm VPSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3988 VR128, v2i64, v2i64, bc_v2i64,
3989 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3991 defm VPSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3992 VR128, v8i16, v8i16, bc_v8i16,
3993 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3994 defm VPSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3995 VR128, v4i32, v4i32, bc_v4i32,
3996 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3997 defm VPSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3998 VR128, v2i64, v2i64, bc_v2i64,
3999 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
4001 defm VPSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
4002 VR128, v8i16, v8i16, bc_v8i16,
4003 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
4004 defm VPSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
4005 VR128, v4i32, v4i32, bc_v4i32,
4006 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
4008 let ExeDomain = SSEPackedInt, SchedRW = [WriteVecShift] in {
4009 // 128-bit logical shifts.
4010 def VPSLLDQri : PDIi8<0x73, MRM7r,
4011 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
4012 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4014 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>,
4016 def VPSRLDQri : PDIi8<0x73, MRM3r,
4017 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
4018 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4020 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>,
4022 // PSRADQri doesn't exist in SSE[1-3].
4024 } // Predicates = [HasAVX]
4026 let Predicates = [HasAVX2] in {
4027 defm VPSLLWY : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
4028 VR256, v16i16, v8i16, bc_v8i16,
4029 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
4030 defm VPSLLDY : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
4031 VR256, v8i32, v4i32, bc_v4i32,
4032 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
4033 defm VPSLLQY : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
4034 VR256, v4i64, v2i64, bc_v2i64,
4035 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
4037 defm VPSRLWY : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
4038 VR256, v16i16, v8i16, bc_v8i16,
4039 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
4040 defm VPSRLDY : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
4041 VR256, v8i32, v4i32, bc_v4i32,
4042 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
4043 defm VPSRLQY : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
4044 VR256, v4i64, v2i64, bc_v2i64,
4045 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
4047 defm VPSRAWY : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
4048 VR256, v16i16, v8i16, bc_v8i16,
4049 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
4050 defm VPSRADY : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
4051 VR256, v8i32, v4i32, bc_v4i32,
4052 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
4054 let ExeDomain = SSEPackedInt, SchedRW = [WriteVecShift] in {
4055 // 256-bit logical shifts.
4056 def VPSLLDQYri : PDIi8<0x73, MRM7r,
4057 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
4058 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4060 (int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2))]>,
4062 def VPSRLDQYri : PDIi8<0x73, MRM3r,
4063 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
4064 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4066 (int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2))]>,
4068 // PSRADQYri doesn't exist in SSE[1-3].
4070 } // Predicates = [HasAVX2]
4072 let Constraints = "$src1 = $dst" in {
4073 defm PSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "psllw", X86vshl, X86vshli,
4074 VR128, v8i16, v8i16, bc_v8i16,
4075 SSE_INTSHIFT_ITINS_P>;
4076 defm PSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "pslld", X86vshl, X86vshli,
4077 VR128, v4i32, v4i32, bc_v4i32,
4078 SSE_INTSHIFT_ITINS_P>;
4079 defm PSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "psllq", X86vshl, X86vshli,
4080 VR128, v2i64, v2i64, bc_v2i64,
4081 SSE_INTSHIFT_ITINS_P>;
4083 defm PSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "psrlw", X86vsrl, X86vsrli,
4084 VR128, v8i16, v8i16, bc_v8i16,
4085 SSE_INTSHIFT_ITINS_P>;
4086 defm PSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "psrld", X86vsrl, X86vsrli,
4087 VR128, v4i32, v4i32, bc_v4i32,
4088 SSE_INTSHIFT_ITINS_P>;
4089 defm PSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "psrlq", X86vsrl, X86vsrli,
4090 VR128, v2i64, v2i64, bc_v2i64,
4091 SSE_INTSHIFT_ITINS_P>;
4093 defm PSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "psraw", X86vsra, X86vsrai,
4094 VR128, v8i16, v8i16, bc_v8i16,
4095 SSE_INTSHIFT_ITINS_P>;
4096 defm PSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "psrad", X86vsra, X86vsrai,
4097 VR128, v4i32, v4i32, bc_v4i32,
4098 SSE_INTSHIFT_ITINS_P>;
4100 let ExeDomain = SSEPackedInt, SchedRW = [WriteVecShift] in {
4101 // 128-bit logical shifts.
4102 def PSLLDQri : PDIi8<0x73, MRM7r,
4103 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
4104 "pslldq\t{$src2, $dst|$dst, $src2}",
4106 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))],
4107 IIC_SSE_INTSHDQ_P_RI>;
4108 def PSRLDQri : PDIi8<0x73, MRM3r,
4109 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
4110 "psrldq\t{$src2, $dst|$dst, $src2}",
4112 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))],
4113 IIC_SSE_INTSHDQ_P_RI>;
4114 // PSRADQri doesn't exist in SSE[1-3].
4116 } // Constraints = "$src1 = $dst"
4118 let Predicates = [HasAVX] in {
4119 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
4120 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4121 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
4122 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4123 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
4124 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4126 // Shift up / down and insert zero's.
4127 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
4128 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4129 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
4130 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4133 let Predicates = [HasAVX2] in {
4134 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
4135 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
4136 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
4137 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
4140 let Predicates = [UseSSE2] in {
4141 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
4142 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4143 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
4144 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4145 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
4146 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4148 // Shift up / down and insert zero's.
4149 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
4150 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4151 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
4152 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4155 //===---------------------------------------------------------------------===//
4156 // SSE2 - Packed Integer Comparison Instructions
4157 //===---------------------------------------------------------------------===//
4159 defm PCMPEQB : PDI_binop_all<0x74, "pcmpeqb", X86pcmpeq, v16i8, v32i8,
4160 SSE_INTALU_ITINS_P, 1>;
4161 defm PCMPEQW : PDI_binop_all<0x75, "pcmpeqw", X86pcmpeq, v8i16, v16i16,
4162 SSE_INTALU_ITINS_P, 1>;
4163 defm PCMPEQD : PDI_binop_all<0x76, "pcmpeqd", X86pcmpeq, v4i32, v8i32,
4164 SSE_INTALU_ITINS_P, 1>;
4165 defm PCMPGTB : PDI_binop_all<0x64, "pcmpgtb", X86pcmpgt, v16i8, v32i8,
4166 SSE_INTALU_ITINS_P, 0>;
4167 defm PCMPGTW : PDI_binop_all<0x65, "pcmpgtw", X86pcmpgt, v8i16, v16i16,
4168 SSE_INTALU_ITINS_P, 0>;
4169 defm PCMPGTD : PDI_binop_all<0x66, "pcmpgtd", X86pcmpgt, v4i32, v8i32,
4170 SSE_INTALU_ITINS_P, 0>;
4172 //===---------------------------------------------------------------------===//
4173 // SSE2 - Packed Integer Pack Instructions
4174 //===---------------------------------------------------------------------===//
4176 defm PACKSSWB : PDI_binop_all_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
4177 int_x86_avx2_packsswb, SSE_INTALU_ITINS_P, 0>;
4178 defm PACKSSDW : PDI_binop_all_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
4179 int_x86_avx2_packssdw, SSE_INTALU_ITINS_P, 0>;
4180 defm PACKUSWB : PDI_binop_all_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
4181 int_x86_avx2_packuswb, SSE_INTALU_ITINS_P, 0>;
4183 //===---------------------------------------------------------------------===//
4184 // SSE2 - Packed Integer Shuffle Instructions
4185 //===---------------------------------------------------------------------===//
4187 let ExeDomain = SSEPackedInt in {
4188 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt128, ValueType vt256,
4190 let Predicates = [HasAVX] in {
4191 def V#NAME#ri : Ii8<0x70, MRMSrcReg, (outs VR128:$dst),
4192 (ins VR128:$src1, i8imm:$src2),
4193 !strconcat("v", OpcodeStr,
4194 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4196 (vt128 (OpNode VR128:$src1, (i8 imm:$src2))))],
4197 IIC_SSE_PSHUF_RI>, VEX, Sched<[WriteShuffle]>;
4198 def V#NAME#mi : Ii8<0x70, MRMSrcMem, (outs VR128:$dst),
4199 (ins i128mem:$src1, i8imm:$src2),
4200 !strconcat("v", OpcodeStr,
4201 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4203 (vt128 (OpNode (bitconvert (loadv2i64 addr:$src1)),
4204 (i8 imm:$src2))))], IIC_SSE_PSHUF_MI>, VEX,
4205 Sched<[WriteShuffleLd]>;
4208 let Predicates = [HasAVX2] in {
4209 def V#NAME#Yri : Ii8<0x70, MRMSrcReg, (outs VR256:$dst),
4210 (ins VR256:$src1, i8imm:$src2),
4211 !strconcat("v", OpcodeStr,
4212 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4214 (vt256 (OpNode VR256:$src1, (i8 imm:$src2))))],
4215 IIC_SSE_PSHUF_RI>, VEX, VEX_L, Sched<[WriteShuffle]>;
4216 def V#NAME#Ymi : Ii8<0x70, MRMSrcMem, (outs VR256:$dst),
4217 (ins i256mem:$src1, i8imm:$src2),
4218 !strconcat("v", OpcodeStr,
4219 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4221 (vt256 (OpNode (bitconvert (loadv4i64 addr:$src1)),
4222 (i8 imm:$src2))))], IIC_SSE_PSHUF_MI>, VEX, VEX_L,
4223 Sched<[WriteShuffleLd]>;
4226 let Predicates = [UseSSE2] in {
4227 def ri : Ii8<0x70, MRMSrcReg,
4228 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
4229 !strconcat(OpcodeStr,
4230 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4232 (vt128 (OpNode VR128:$src1, (i8 imm:$src2))))],
4233 IIC_SSE_PSHUF_RI>, Sched<[WriteShuffle]>;
4234 def mi : Ii8<0x70, MRMSrcMem,
4235 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
4236 !strconcat(OpcodeStr,
4237 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4239 (vt128 (OpNode (bitconvert (memopv2i64 addr:$src1)),
4240 (i8 imm:$src2))))], IIC_SSE_PSHUF_MI>,
4241 Sched<[WriteShuffleLd]>;
4244 } // ExeDomain = SSEPackedInt
4246 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, v8i32, X86PShufd>, TB, OpSize;
4247 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, v16i16, X86PShufhw>, XS;
4248 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, v16i16, X86PShuflw>, XD;
4250 let Predicates = [HasAVX] in {
4251 def : Pat<(v4f32 (X86PShufd (loadv4f32 addr:$src1), (i8 imm:$imm))),
4252 (VPSHUFDmi addr:$src1, imm:$imm)>;
4253 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4254 (VPSHUFDri VR128:$src1, imm:$imm)>;
4257 let Predicates = [UseSSE2] in {
4258 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
4259 (PSHUFDmi addr:$src1, imm:$imm)>;
4260 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4261 (PSHUFDri VR128:$src1, imm:$imm)>;
4264 //===---------------------------------------------------------------------===//
4265 // SSE2 - Packed Integer Unpack Instructions
4266 //===---------------------------------------------------------------------===//
4268 let ExeDomain = SSEPackedInt in {
4269 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
4270 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
4271 def rr : PDI<opc, MRMSrcReg,
4272 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4274 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4275 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4276 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))],
4277 IIC_SSE_UNPCK>, Sched<[WriteShuffle]>;
4278 def rm : PDI<opc, MRMSrcMem,
4279 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4281 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4282 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4283 [(set VR128:$dst, (OpNode VR128:$src1,
4284 (bc_frag (memopv2i64
4287 Sched<[WriteShuffleLd, ReadAfterLd]>;
4290 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
4291 SDNode OpNode, PatFrag bc_frag> {
4292 def Yrr : PDI<opc, MRMSrcReg,
4293 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4294 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4295 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>,
4296 Sched<[WriteShuffle]>;
4297 def Yrm : PDI<opc, MRMSrcMem,
4298 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4299 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4300 [(set VR256:$dst, (OpNode VR256:$src1,
4301 (bc_frag (memopv4i64 addr:$src2))))]>,
4302 Sched<[WriteShuffleLd, ReadAfterLd]>;
4305 let Predicates = [HasAVX] in {
4306 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
4307 bc_v16i8, 0>, VEX_4V;
4308 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
4309 bc_v8i16, 0>, VEX_4V;
4310 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
4311 bc_v4i32, 0>, VEX_4V;
4312 defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
4313 bc_v2i64, 0>, VEX_4V;
4315 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
4316 bc_v16i8, 0>, VEX_4V;
4317 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
4318 bc_v8i16, 0>, VEX_4V;
4319 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
4320 bc_v4i32, 0>, VEX_4V;
4321 defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
4322 bc_v2i64, 0>, VEX_4V;
4325 let Predicates = [HasAVX2] in {
4326 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
4327 bc_v32i8>, VEX_4V, VEX_L;
4328 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
4329 bc_v16i16>, VEX_4V, VEX_L;
4330 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
4331 bc_v8i32>, VEX_4V, VEX_L;
4332 defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
4333 bc_v4i64>, VEX_4V, VEX_L;
4335 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
4336 bc_v32i8>, VEX_4V, VEX_L;
4337 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
4338 bc_v16i16>, VEX_4V, VEX_L;
4339 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
4340 bc_v8i32>, VEX_4V, VEX_L;
4341 defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
4342 bc_v4i64>, VEX_4V, VEX_L;
4345 let Constraints = "$src1 = $dst" in {
4346 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
4348 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
4350 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
4352 defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
4355 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
4357 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
4359 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
4361 defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
4364 } // ExeDomain = SSEPackedInt
4366 //===---------------------------------------------------------------------===//
4367 // SSE2 - Packed Integer Extract and Insert
4368 //===---------------------------------------------------------------------===//
4370 let ExeDomain = SSEPackedInt in {
4371 multiclass sse2_pinsrw<bit Is2Addr = 1> {
4372 def rri : Ii8<0xC4, MRMSrcReg,
4373 (outs VR128:$dst), (ins VR128:$src1,
4374 GR32orGR64:$src2, i32i8imm:$src3),
4376 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4377 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4379 (X86pinsrw VR128:$src1, GR32orGR64:$src2, imm:$src3))],
4380 IIC_SSE_PINSRW>, Sched<[WriteShuffle]>;
4381 def rmi : Ii8<0xC4, MRMSrcMem,
4382 (outs VR128:$dst), (ins VR128:$src1,
4383 i16mem:$src2, i32i8imm:$src3),
4385 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4386 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4388 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
4389 imm:$src3))], IIC_SSE_PINSRW>,
4390 Sched<[WriteShuffleLd, ReadAfterLd]>;
4394 let Predicates = [HasAVX] in
4395 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
4396 (outs GR32orGR64:$dst), (ins VR128:$src1, i32i8imm:$src2),
4397 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4398 [(set GR32orGR64:$dst, (X86pextrw (v8i16 VR128:$src1),
4399 imm:$src2))]>, TB, OpSize, VEX,
4400 Sched<[WriteShuffle]>;
4401 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
4402 (outs GR32orGR64:$dst), (ins VR128:$src1, i32i8imm:$src2),
4403 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4404 [(set GR32orGR64:$dst, (X86pextrw (v8i16 VR128:$src1),
4405 imm:$src2))], IIC_SSE_PEXTRW>,
4406 Sched<[WriteShuffleLd, ReadAfterLd]>;
4409 let Predicates = [HasAVX] in
4410 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
4412 let Predicates = [UseSSE2], Constraints = "$src1 = $dst" in
4413 defm PINSRW : sse2_pinsrw, TB, OpSize;
4415 } // ExeDomain = SSEPackedInt
4417 //===---------------------------------------------------------------------===//
4418 // SSE2 - Packed Mask Creation
4419 //===---------------------------------------------------------------------===//
4421 let ExeDomain = SSEPackedInt, SchedRW = [WriteVecLogic] in {
4423 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32orGR64:$dst),
4425 "pmovmskb\t{$src, $dst|$dst, $src}",
4426 [(set GR32orGR64:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4427 IIC_SSE_MOVMSK>, VEX;
4429 let Predicates = [HasAVX2] in {
4430 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32orGR64:$dst),
4432 "pmovmskb\t{$src, $dst|$dst, $src}",
4433 [(set GR32orGR64:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>,
4437 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32orGR64:$dst), (ins VR128:$src),
4438 "pmovmskb\t{$src, $dst|$dst, $src}",
4439 [(set GR32orGR64:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4442 } // ExeDomain = SSEPackedInt
4444 //===---------------------------------------------------------------------===//
4445 // SSE2 - Conditional Store
4446 //===---------------------------------------------------------------------===//
4448 let ExeDomain = SSEPackedInt, SchedRW = [WriteStore] in {
4450 let Uses = [EDI], Predicates = [HasAVX,In32BitMode] in
4451 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4452 (ins VR128:$src, VR128:$mask),
4453 "maskmovdqu\t{$mask, $src|$src, $mask}",
4454 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4455 IIC_SSE_MASKMOV>, VEX;
4456 let Uses = [RDI], Predicates = [HasAVX,In64BitMode] in
4457 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4458 (ins VR128:$src, VR128:$mask),
4459 "maskmovdqu\t{$mask, $src|$src, $mask}",
4460 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4461 IIC_SSE_MASKMOV>, VEX;
4463 let Uses = [EDI], Predicates = [UseSSE2,In32BitMode] in
4464 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4465 "maskmovdqu\t{$mask, $src|$src, $mask}",
4466 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4468 let Uses = [RDI], Predicates = [UseSSE2,In64BitMode] in
4469 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4470 "maskmovdqu\t{$mask, $src|$src, $mask}",
4471 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4474 } // ExeDomain = SSEPackedInt
4476 //===---------------------------------------------------------------------===//
4477 // SSE2 - Move Doubleword
4478 //===---------------------------------------------------------------------===//
4480 //===---------------------------------------------------------------------===//
4481 // Move Int Doubleword to Packed Double Int
4483 def VMOVDI2PDIrr : VS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4484 "movd\t{$src, $dst|$dst, $src}",
4486 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>,
4487 VEX, Sched<[WriteMove]>;
4488 def VMOVDI2PDIrm : VS2I<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4489 "movd\t{$src, $dst|$dst, $src}",
4491 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4493 VEX, Sched<[WriteLoad]>;
4494 def VMOV64toPQIrr : VRS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4495 "movq\t{$src, $dst|$dst, $src}",
4497 (v2i64 (scalar_to_vector GR64:$src)))],
4498 IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
4499 let isCodeGenOnly = 1 in
4500 def VMOV64toSDrr : VRS2I<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4501 "movq\t{$src, $dst|$dst, $src}",
4502 [(set FR64:$dst, (bitconvert GR64:$src))],
4503 IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
4505 def MOVDI2PDIrr : S2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4506 "movd\t{$src, $dst|$dst, $src}",
4508 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>,
4510 def MOVDI2PDIrm : S2I<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4511 "movd\t{$src, $dst|$dst, $src}",
4513 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4514 IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
4515 def MOV64toPQIrr : RS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4516 "mov{d|q}\t{$src, $dst|$dst, $src}",
4518 (v2i64 (scalar_to_vector GR64:$src)))],
4519 IIC_SSE_MOVDQ>, Sched<[WriteMove]>;
4520 let isCodeGenOnly = 1 in
4521 def MOV64toSDrr : RS2I<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4522 "mov{d|q}\t{$src, $dst|$dst, $src}",
4523 [(set FR64:$dst, (bitconvert GR64:$src))],
4524 IIC_SSE_MOVDQ>, Sched<[WriteMove]>;
4526 //===---------------------------------------------------------------------===//
4527 // Move Int Doubleword to Single Scalar
4529 let isCodeGenOnly = 1 in {
4530 def VMOVDI2SSrr : VS2I<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4531 "movd\t{$src, $dst|$dst, $src}",
4532 [(set FR32:$dst, (bitconvert GR32:$src))],
4533 IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
4535 def VMOVDI2SSrm : VS2I<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4536 "movd\t{$src, $dst|$dst, $src}",
4537 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4539 VEX, Sched<[WriteLoad]>;
4540 def MOVDI2SSrr : S2I<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4541 "movd\t{$src, $dst|$dst, $src}",
4542 [(set FR32:$dst, (bitconvert GR32:$src))],
4543 IIC_SSE_MOVDQ>, Sched<[WriteMove]>;
4545 def MOVDI2SSrm : S2I<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4546 "movd\t{$src, $dst|$dst, $src}",
4547 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4548 IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
4551 //===---------------------------------------------------------------------===//
4552 // Move Packed Doubleword Int to Packed Double Int
4554 def VMOVPDI2DIrr : VS2I<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4555 "movd\t{$src, $dst|$dst, $src}",
4556 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4557 (iPTR 0)))], IIC_SSE_MOVD_ToGP>, VEX,
4559 def VMOVPDI2DImr : VS2I<0x7E, MRMDestMem, (outs),
4560 (ins i32mem:$dst, VR128:$src),
4561 "movd\t{$src, $dst|$dst, $src}",
4562 [(store (i32 (vector_extract (v4i32 VR128:$src),
4563 (iPTR 0))), addr:$dst)], IIC_SSE_MOVDQ>,
4564 VEX, Sched<[WriteLoad]>;
4565 def MOVPDI2DIrr : S2I<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4566 "movd\t{$src, $dst|$dst, $src}",
4567 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4568 (iPTR 0)))], IIC_SSE_MOVD_ToGP>,
4570 def MOVPDI2DImr : S2I<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4571 "movd\t{$src, $dst|$dst, $src}",
4572 [(store (i32 (vector_extract (v4i32 VR128:$src),
4573 (iPTR 0))), addr:$dst)],
4574 IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
4576 def : Pat<(v8i32 (X86Vinsert (v8i32 immAllZerosV), GR32:$src2, (iPTR 0))),
4577 (SUBREG_TO_REG (i32 0), (VMOVDI2PDIrr GR32:$src2), sub_xmm)>;
4579 def : Pat<(v4i64 (X86Vinsert (bc_v4i64 (v8i32 immAllZerosV)), GR64:$src2, (iPTR 0))),
4580 (SUBREG_TO_REG (i32 0), (VMOV64toPQIrr GR64:$src2), sub_xmm)>;
4582 def : Pat<(v8i32 (X86Vinsert undef, GR32:$src2, (iPTR 0))),
4583 (SUBREG_TO_REG (i32 0), (VMOVDI2PDIrr GR32:$src2), sub_xmm)>;
4585 def : Pat<(v4i64 (X86Vinsert undef, GR64:$src2, (iPTR 0))),
4586 (SUBREG_TO_REG (i32 0), (VMOV64toPQIrr GR64:$src2), sub_xmm)>;
4588 //===---------------------------------------------------------------------===//
4589 // Move Packed Doubleword Int first element to Doubleword Int
4591 let SchedRW = [WriteMove] in {
4592 def VMOVPQIto64rr : VRS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4593 "movq\t{$src, $dst|$dst, $src}",
4594 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4599 def MOVPQIto64rr : RS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4600 "mov{d|q}\t{$src, $dst|$dst, $src}",
4601 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4606 //===---------------------------------------------------------------------===//
4607 // Bitcast FR64 <-> GR64
4609 let isCodeGenOnly = 1 in {
4610 let Predicates = [UseAVX] in
4611 def VMOV64toSDrm : VS2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4612 "movq\t{$src, $dst|$dst, $src}",
4613 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4614 VEX, Sched<[WriteLoad]>;
4615 def VMOVSDto64rr : VRS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4616 "movq\t{$src, $dst|$dst, $src}",
4617 [(set GR64:$dst, (bitconvert FR64:$src))],
4618 IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
4619 def VMOVSDto64mr : VRS2I<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4620 "movq\t{$src, $dst|$dst, $src}",
4621 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4622 IIC_SSE_MOVDQ>, VEX, Sched<[WriteStore]>;
4624 def MOV64toSDrm : S2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4625 "movq\t{$src, $dst|$dst, $src}",
4626 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))],
4627 IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
4628 def MOVSDto64rr : RS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4629 "mov{d|q}\t{$src, $dst|$dst, $src}",
4630 [(set GR64:$dst, (bitconvert FR64:$src))],
4631 IIC_SSE_MOVD_ToGP>, Sched<[WriteMove]>;
4632 def MOVSDto64mr : RS2I<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4633 "movq\t{$src, $dst|$dst, $src}",
4634 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4635 IIC_SSE_MOVDQ>, Sched<[WriteStore]>;
4638 //===---------------------------------------------------------------------===//
4639 // Move Scalar Single to Double Int
4641 let isCodeGenOnly = 1 in {
4642 def VMOVSS2DIrr : VS2I<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4643 "movd\t{$src, $dst|$dst, $src}",
4644 [(set GR32:$dst, (bitconvert FR32:$src))],
4645 IIC_SSE_MOVD_ToGP>, VEX, Sched<[WriteMove]>;
4646 def VMOVSS2DImr : VS2I<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4647 "movd\t{$src, $dst|$dst, $src}",
4648 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4649 IIC_SSE_MOVDQ>, VEX, Sched<[WriteStore]>;
4650 def MOVSS2DIrr : S2I<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4651 "movd\t{$src, $dst|$dst, $src}",
4652 [(set GR32:$dst, (bitconvert FR32:$src))],
4653 IIC_SSE_MOVD_ToGP>, Sched<[WriteMove]>;
4654 def MOVSS2DImr : S2I<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4655 "movd\t{$src, $dst|$dst, $src}",
4656 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4657 IIC_SSE_MOVDQ>, Sched<[WriteStore]>;
4660 //===---------------------------------------------------------------------===//
4661 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4663 let isCodeGenOnly = 1, SchedRW = [WriteMove] in {
4664 let AddedComplexity = 15 in {
4665 def VMOVZQI2PQIrr : VS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4666 "movq\t{$src, $dst|$dst, $src}", // X86-64 only
4667 [(set VR128:$dst, (v2i64 (X86vzmovl
4668 (v2i64 (scalar_to_vector GR64:$src)))))],
4671 def MOVZQI2PQIrr : RS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4672 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4673 [(set VR128:$dst, (v2i64 (X86vzmovl
4674 (v2i64 (scalar_to_vector GR64:$src)))))],
4677 } // isCodeGenOnly, SchedRW
4679 let Predicates = [UseAVX] in {
4680 let AddedComplexity = 15 in
4681 def : Pat<(v4i32 (X86vzmovl (v4i32 (scalar_to_vector GR32:$src)))),
4682 (VMOVDI2PDIrr GR32:$src)>;
4684 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4685 let AddedComplexity = 20 in {
4686 def : Pat<(v4i32 (X86vzmovl (v4i32 (scalar_to_vector (loadi32 addr:$src))))),
4687 (VMOVDI2PDIrm addr:$src)>;
4688 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4689 (VMOVDI2PDIrm addr:$src)>;
4690 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4691 (VMOVDI2PDIrm addr:$src)>;
4693 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4694 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4695 (v4i32 (scalar_to_vector GR32:$src)),(iPTR 0)))),
4696 (SUBREG_TO_REG (i32 0), (VMOVDI2PDIrr GR32:$src), sub_xmm)>;
4697 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4698 (v2i64 (scalar_to_vector GR64:$src)),(iPTR 0)))),
4699 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4702 let Predicates = [UseSSE2] in {
4703 let AddedComplexity = 15 in
4704 def : Pat<(v4i32 (X86vzmovl (v4i32 (scalar_to_vector GR32:$src)))),
4705 (MOVDI2PDIrr GR32:$src)>;
4707 let AddedComplexity = 20 in {
4708 def : Pat<(v4i32 (X86vzmovl (v4i32 (scalar_to_vector (loadi32 addr:$src))))),
4709 (MOVDI2PDIrm addr:$src)>;
4710 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4711 (MOVDI2PDIrm addr:$src)>;
4712 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4713 (MOVDI2PDIrm addr:$src)>;
4717 // These are the correct encodings of the instructions so that we know how to
4718 // read correct assembly, even though we continue to emit the wrong ones for
4719 // compatibility with Darwin's buggy assembler.
4720 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4721 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4722 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4723 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4724 // Allow "vmovd" but print "vmovq" since we don't need compatibility for AVX.
4725 def : InstAlias<"vmovd\t{$src, $dst|$dst, $src}",
4726 (VMOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4727 def : InstAlias<"vmovd\t{$src, $dst|$dst, $src}",
4728 (VMOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4730 //===---------------------------------------------------------------------===//
4731 // SSE2 - Move Quadword
4732 //===---------------------------------------------------------------------===//
4734 //===---------------------------------------------------------------------===//
4735 // Move Quadword Int to Packed Quadword Int
4738 let SchedRW = [WriteLoad] in {
4739 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4740 "vmovq\t{$src, $dst|$dst, $src}",
4742 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4743 VEX, Requires<[UseAVX]>;
4744 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4745 "movq\t{$src, $dst|$dst, $src}",
4747 (v2i64 (scalar_to_vector (loadi64 addr:$src))))],
4749 Requires<[UseSSE2]>; // SSE2 instruction with XS Prefix
4752 //===---------------------------------------------------------------------===//
4753 // Move Packed Quadword Int to Quadword Int
4755 let SchedRW = [WriteStore] in {
4756 def VMOVPQI2QImr : VS2I<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4757 "movq\t{$src, $dst|$dst, $src}",
4758 [(store (i64 (vector_extract (v2i64 VR128:$src),
4759 (iPTR 0))), addr:$dst)],
4760 IIC_SSE_MOVDQ>, VEX;
4761 def MOVPQI2QImr : S2I<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4762 "movq\t{$src, $dst|$dst, $src}",
4763 [(store (i64 (vector_extract (v2i64 VR128:$src),
4764 (iPTR 0))), addr:$dst)],
4768 //===---------------------------------------------------------------------===//
4769 // Store / copy lower 64-bits of a XMM register.
4771 def VMOVLQ128mr : VS2I<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4772 "movq\t{$src, $dst|$dst, $src}",
4773 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX,
4774 Sched<[WriteStore]>;
4775 def MOVLQ128mr : S2I<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4776 "movq\t{$src, $dst|$dst, $src}",
4777 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)],
4778 IIC_SSE_MOVDQ>, Sched<[WriteStore]>;
4780 let isCodeGenOnly = 1, AddedComplexity = 20 in {
4781 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4782 "vmovq\t{$src, $dst|$dst, $src}",
4784 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4785 (loadi64 addr:$src))))))],
4787 XS, VEX, Requires<[UseAVX]>, Sched<[WriteLoad]>;
4789 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4790 "movq\t{$src, $dst|$dst, $src}",
4792 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4793 (loadi64 addr:$src))))))],
4795 XS, Requires<[UseSSE2]>, Sched<[WriteLoad]>;
4798 let Predicates = [UseAVX], AddedComplexity = 20 in {
4799 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4800 (VMOVZQI2PQIrm addr:$src)>;
4801 def : Pat<(v2i64 (X86vzload addr:$src)),
4802 (VMOVZQI2PQIrm addr:$src)>;
4805 let Predicates = [UseSSE2], AddedComplexity = 20 in {
4806 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4807 (MOVZQI2PQIrm addr:$src)>;
4808 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4811 let Predicates = [HasAVX] in {
4812 def : Pat<(v4i64 (alignedX86vzload addr:$src)),
4813 (SUBREG_TO_REG (i32 0), (VMOVAPSrm addr:$src), sub_xmm)>;
4814 def : Pat<(v4i64 (X86vzload addr:$src)),
4815 (SUBREG_TO_REG (i32 0), (VMOVUPSrm addr:$src), sub_xmm)>;
4818 //===---------------------------------------------------------------------===//
4819 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4820 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4822 let SchedRW = [WriteVecLogic] in {
4823 let AddedComplexity = 15 in
4824 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4825 "vmovq\t{$src, $dst|$dst, $src}",
4826 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4828 XS, VEX, Requires<[UseAVX]>;
4829 let AddedComplexity = 15 in
4830 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4831 "movq\t{$src, $dst|$dst, $src}",
4832 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4834 XS, Requires<[UseSSE2]>;
4837 let isCodeGenOnly = 1, SchedRW = [WriteVecLogicLd] in {
4838 let AddedComplexity = 20 in
4839 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4840 "vmovq\t{$src, $dst|$dst, $src}",
4841 [(set VR128:$dst, (v2i64 (X86vzmovl
4842 (loadv2i64 addr:$src))))],
4844 XS, VEX, Requires<[UseAVX]>;
4845 let AddedComplexity = 20 in {
4846 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4847 "movq\t{$src, $dst|$dst, $src}",
4848 [(set VR128:$dst, (v2i64 (X86vzmovl
4849 (loadv2i64 addr:$src))))],
4851 XS, Requires<[UseSSE2]>;
4853 } // isCodeGenOnly, SchedRW
4855 let AddedComplexity = 20 in {
4856 let Predicates = [UseAVX] in {
4857 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4858 (VMOVZPQILo2PQIrr VR128:$src)>;
4860 let Predicates = [UseSSE2] in {
4861 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4862 (MOVZPQILo2PQIrr VR128:$src)>;
4866 //===---------------------------------------------------------------------===//
4867 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4868 //===---------------------------------------------------------------------===//
4869 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4870 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4871 X86MemOperand x86memop> {
4872 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4873 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4874 [(set RC:$dst, (vt (OpNode RC:$src)))],
4875 IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>;
4876 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4877 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4878 [(set RC:$dst, (OpNode (mem_frag addr:$src)))],
4879 IIC_SSE_MOV_LH>, Sched<[WriteShuffleLd]>;
4882 let Predicates = [HasAVX] in {
4883 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4884 v4f32, VR128, loadv4f32, f128mem>, VEX;
4885 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4886 v4f32, VR128, loadv4f32, f128mem>, VEX;
4887 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4888 v8f32, VR256, loadv8f32, f256mem>, VEX, VEX_L;
4889 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4890 v8f32, VR256, loadv8f32, f256mem>, VEX, VEX_L;
4892 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4893 memopv4f32, f128mem>;
4894 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4895 memopv4f32, f128mem>;
4897 let Predicates = [HasAVX] in {
4898 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4899 (VMOVSHDUPrr VR128:$src)>;
4900 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (loadv2i64 addr:$src)))),
4901 (VMOVSHDUPrm addr:$src)>;
4902 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4903 (VMOVSLDUPrr VR128:$src)>;
4904 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (loadv2i64 addr:$src)))),
4905 (VMOVSLDUPrm addr:$src)>;
4906 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4907 (VMOVSHDUPYrr VR256:$src)>;
4908 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (loadv4i64 addr:$src)))),
4909 (VMOVSHDUPYrm addr:$src)>;
4910 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4911 (VMOVSLDUPYrr VR256:$src)>;
4912 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (loadv4i64 addr:$src)))),
4913 (VMOVSLDUPYrm addr:$src)>;
4916 let Predicates = [UseSSE3] in {
4917 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4918 (MOVSHDUPrr VR128:$src)>;
4919 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4920 (MOVSHDUPrm addr:$src)>;
4921 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4922 (MOVSLDUPrr VR128:$src)>;
4923 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4924 (MOVSLDUPrm addr:$src)>;
4927 //===---------------------------------------------------------------------===//
4928 // SSE3 - Replicate Double FP - MOVDDUP
4929 //===---------------------------------------------------------------------===//
4931 multiclass sse3_replicate_dfp<string OpcodeStr> {
4932 let neverHasSideEffects = 1 in
4933 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4934 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4935 [], IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>;
4936 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
4937 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4940 (scalar_to_vector (loadf64 addr:$src)))))],
4941 IIC_SSE_MOV_LH>, Sched<[WriteShuffleLd]>;
4944 // FIXME: Merge with above classe when there're patterns for the ymm version
4945 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
4946 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
4947 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4948 [(set VR256:$dst, (v4f64 (X86Movddup VR256:$src)))]>,
4949 Sched<[WriteShuffle]>;
4950 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
4951 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4954 (scalar_to_vector (loadf64 addr:$src)))))]>,
4955 Sched<[WriteShuffleLd]>;
4958 let Predicates = [HasAVX] in {
4959 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
4960 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX, VEX_L;
4963 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
4965 let Predicates = [HasAVX] in {
4966 def : Pat<(X86Movddup (loadv2f64 addr:$src)),
4967 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4968 def : Pat<(X86Movddup (bc_v2f64 (loadv4f32 addr:$src))),
4969 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4970 def : Pat<(X86Movddup (bc_v2f64 (loadv2i64 addr:$src))),
4971 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4972 def : Pat<(X86Movddup (bc_v2f64
4973 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4974 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4977 def : Pat<(X86Movddup (loadv4f64 addr:$src)),
4978 (VMOVDDUPYrm addr:$src)>;
4979 def : Pat<(X86Movddup (loadv4i64 addr:$src)),
4980 (VMOVDDUPYrm addr:$src)>;
4981 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
4982 (VMOVDDUPYrm addr:$src)>;
4983 def : Pat<(X86Movddup (v4i64 VR256:$src)),
4984 (VMOVDDUPYrr VR256:$src)>;
4987 let Predicates = [UseSSE3] in {
4988 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4989 (MOVDDUPrm addr:$src)>;
4990 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4991 (MOVDDUPrm addr:$src)>;
4992 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4993 (MOVDDUPrm addr:$src)>;
4994 def : Pat<(X86Movddup (bc_v2f64
4995 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4996 (MOVDDUPrm addr:$src)>;
4999 //===---------------------------------------------------------------------===//
5000 // SSE3 - Move Unaligned Integer
5001 //===---------------------------------------------------------------------===//
5003 let SchedRW = [WriteLoad] in {
5004 let Predicates = [HasAVX] in {
5005 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5006 "vlddqu\t{$src, $dst|$dst, $src}",
5007 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
5008 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
5009 "vlddqu\t{$src, $dst|$dst, $src}",
5010 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>,
5013 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5014 "lddqu\t{$src, $dst|$dst, $src}",
5015 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))],
5019 //===---------------------------------------------------------------------===//
5020 // SSE3 - Arithmetic
5021 //===---------------------------------------------------------------------===//
5023 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
5024 X86MemOperand x86memop, OpndItins itins,
5026 def rr : I<0xD0, MRMSrcReg,
5027 (outs RC:$dst), (ins RC:$src1, RC:$src2),
5029 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5030 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5031 [(set RC:$dst, (Int RC:$src1, RC:$src2))], itins.rr>,
5032 Sched<[itins.Sched]>;
5033 def rm : I<0xD0, MRMSrcMem,
5034 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5036 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5037 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5038 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))], itins.rr>,
5039 Sched<[itins.Sched.Folded, ReadAfterLd]>;
5042 let Predicates = [HasAVX] in {
5043 let ExeDomain = SSEPackedSingle in {
5044 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
5045 f128mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V;
5046 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
5047 f256mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V, VEX_L;
5049 let ExeDomain = SSEPackedDouble in {
5050 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
5051 f128mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V;
5052 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
5053 f256mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V, VEX_L;
5056 let Constraints = "$src1 = $dst", Predicates = [UseSSE3] in {
5057 let ExeDomain = SSEPackedSingle in
5058 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
5059 f128mem, SSE_ALU_F32P>, TB, XD;
5060 let ExeDomain = SSEPackedDouble in
5061 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
5062 f128mem, SSE_ALU_F64P>, TB, OpSize;
5065 //===---------------------------------------------------------------------===//
5066 // SSE3 Instructions
5067 //===---------------------------------------------------------------------===//
5070 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5071 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5072 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5074 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5075 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5076 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>,
5079 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5081 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5082 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5083 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
5084 IIC_SSE_HADDSUB_RM>, Sched<[WriteFAddLd, ReadAfterLd]>;
5086 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5087 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5088 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5090 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5091 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5092 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>,
5095 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5097 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5098 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5099 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
5100 IIC_SSE_HADDSUB_RM>, Sched<[WriteFAddLd, ReadAfterLd]>;
5103 let Predicates = [HasAVX] in {
5104 let ExeDomain = SSEPackedSingle in {
5105 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
5106 X86fhadd, 0>, VEX_4V;
5107 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
5108 X86fhsub, 0>, VEX_4V;
5109 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
5110 X86fhadd, 0>, VEX_4V, VEX_L;
5111 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
5112 X86fhsub, 0>, VEX_4V, VEX_L;
5114 let ExeDomain = SSEPackedDouble in {
5115 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
5116 X86fhadd, 0>, VEX_4V;
5117 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
5118 X86fhsub, 0>, VEX_4V;
5119 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
5120 X86fhadd, 0>, VEX_4V, VEX_L;
5121 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
5122 X86fhsub, 0>, VEX_4V, VEX_L;
5126 let Constraints = "$src1 = $dst" in {
5127 let ExeDomain = SSEPackedSingle in {
5128 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
5129 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
5131 let ExeDomain = SSEPackedDouble in {
5132 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
5133 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
5137 //===---------------------------------------------------------------------===//
5138 // SSSE3 - Packed Absolute Instructions
5139 //===---------------------------------------------------------------------===//
5142 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5143 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
5144 Intrinsic IntId128> {
5145 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5147 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5148 [(set VR128:$dst, (IntId128 VR128:$src))], IIC_SSE_PABS_RR>,
5149 OpSize, Sched<[WriteVecALU]>;
5151 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5153 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5156 (bitconvert (memopv2i64 addr:$src))))], IIC_SSE_PABS_RM>,
5157 OpSize, Sched<[WriteVecALULd]>;
5160 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5161 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
5162 Intrinsic IntId256> {
5163 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5165 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5166 [(set VR256:$dst, (IntId256 VR256:$src))]>,
5167 OpSize, Sched<[WriteVecALU]>;
5169 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5171 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5174 (bitconvert (memopv4i64 addr:$src))))]>, OpSize,
5175 Sched<[WriteVecALULd]>;
5178 // Helper fragments to match sext vXi1 to vXiY.
5179 def v16i1sextv16i8 : PatLeaf<(v16i8 (X86pcmpgt (bc_v16i8 (v4i32 immAllZerosV)),
5181 def v8i1sextv8i16 : PatLeaf<(v8i16 (X86vsrai VR128:$src, (i8 15)))>;
5182 def v4i1sextv4i32 : PatLeaf<(v4i32 (X86vsrai VR128:$src, (i8 31)))>;
5183 def v32i1sextv32i8 : PatLeaf<(v32i8 (X86pcmpgt (bc_v32i8 (v8i32 immAllZerosV)),
5185 def v16i1sextv16i16: PatLeaf<(v16i16 (X86vsrai VR256:$src, (i8 15)))>;
5186 def v8i1sextv8i32 : PatLeaf<(v8i32 (X86vsrai VR256:$src, (i8 31)))>;
5188 let Predicates = [HasAVX] in {
5189 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb",
5190 int_x86_ssse3_pabs_b_128>, VEX;
5191 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw",
5192 int_x86_ssse3_pabs_w_128>, VEX;
5193 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd",
5194 int_x86_ssse3_pabs_d_128>, VEX;
5197 (bc_v2i64 (v16i1sextv16i8)),
5198 (bc_v2i64 (add (v16i8 VR128:$src), (v16i1sextv16i8)))),
5199 (VPABSBrr128 VR128:$src)>;
5201 (bc_v2i64 (v8i1sextv8i16)),
5202 (bc_v2i64 (add (v8i16 VR128:$src), (v8i1sextv8i16)))),
5203 (VPABSWrr128 VR128:$src)>;
5205 (bc_v2i64 (v4i1sextv4i32)),
5206 (bc_v2i64 (add (v4i32 VR128:$src), (v4i1sextv4i32)))),
5207 (VPABSDrr128 VR128:$src)>;
5210 let Predicates = [HasAVX2] in {
5211 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb",
5212 int_x86_avx2_pabs_b>, VEX, VEX_L;
5213 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw",
5214 int_x86_avx2_pabs_w>, VEX, VEX_L;
5215 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd",
5216 int_x86_avx2_pabs_d>, VEX, VEX_L;
5219 (bc_v4i64 (v32i1sextv32i8)),
5220 (bc_v4i64 (add (v32i8 VR256:$src), (v32i1sextv32i8)))),
5221 (VPABSBrr256 VR256:$src)>;
5223 (bc_v4i64 (v16i1sextv16i16)),
5224 (bc_v4i64 (add (v16i16 VR256:$src), (v16i1sextv16i16)))),
5225 (VPABSWrr256 VR256:$src)>;
5227 (bc_v4i64 (v8i1sextv8i32)),
5228 (bc_v4i64 (add (v8i32 VR256:$src), (v8i1sextv8i32)))),
5229 (VPABSDrr256 VR256:$src)>;
5232 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb",
5233 int_x86_ssse3_pabs_b_128>;
5234 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw",
5235 int_x86_ssse3_pabs_w_128>;
5236 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
5237 int_x86_ssse3_pabs_d_128>;
5239 let Predicates = [HasSSSE3] in {
5241 (bc_v2i64 (v16i1sextv16i8)),
5242 (bc_v2i64 (add (v16i8 VR128:$src), (v16i1sextv16i8)))),
5243 (PABSBrr128 VR128:$src)>;
5245 (bc_v2i64 (v8i1sextv8i16)),
5246 (bc_v2i64 (add (v8i16 VR128:$src), (v8i1sextv8i16)))),
5247 (PABSWrr128 VR128:$src)>;
5249 (bc_v2i64 (v4i1sextv4i32)),
5250 (bc_v2i64 (add (v4i32 VR128:$src), (v4i1sextv4i32)))),
5251 (PABSDrr128 VR128:$src)>;
5254 //===---------------------------------------------------------------------===//
5255 // SSSE3 - Packed Binary Operator Instructions
5256 //===---------------------------------------------------------------------===//
5258 let Sched = WriteVecALU in {
5259 def SSE_PHADDSUBD : OpndItins<
5260 IIC_SSE_PHADDSUBD_RR, IIC_SSE_PHADDSUBD_RM
5262 def SSE_PHADDSUBSW : OpndItins<
5263 IIC_SSE_PHADDSUBSW_RR, IIC_SSE_PHADDSUBSW_RM
5265 def SSE_PHADDSUBW : OpndItins<
5266 IIC_SSE_PHADDSUBW_RR, IIC_SSE_PHADDSUBW_RM
5269 let Sched = WriteShuffle in
5270 def SSE_PSHUFB : OpndItins<
5271 IIC_SSE_PSHUFB_RR, IIC_SSE_PSHUFB_RM
5273 let Sched = WriteVecALU in
5274 def SSE_PSIGN : OpndItins<
5275 IIC_SSE_PSIGN_RR, IIC_SSE_PSIGN_RM
5277 let Sched = WriteVecIMul in
5278 def SSE_PMULHRSW : OpndItins<
5279 IIC_SSE_PMULHRSW, IIC_SSE_PMULHRSW
5282 /// SS3I_binop_rm - Simple SSSE3 bin op
5283 multiclass SS3I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
5284 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
5285 X86MemOperand x86memop, OpndItins itins,
5287 let isCommutable = 1 in
5288 def rr : SS38I<opc, MRMSrcReg, (outs RC:$dst),
5289 (ins RC:$src1, RC:$src2),
5291 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5292 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5293 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>,
5294 OpSize, Sched<[itins.Sched]>;
5295 def rm : SS38I<opc, MRMSrcMem, (outs RC:$dst),
5296 (ins RC:$src1, x86memop:$src2),
5298 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5299 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5301 (OpVT (OpNode RC:$src1,
5302 (bitconvert (memop_frag addr:$src2)))))], itins.rm>, OpSize,
5303 Sched<[itins.Sched.Folded, ReadAfterLd]>;
5306 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
5307 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
5308 Intrinsic IntId128, OpndItins itins,
5310 let isCommutable = 1 in
5311 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5312 (ins VR128:$src1, VR128:$src2),
5314 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5315 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5316 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5317 OpSize, Sched<[itins.Sched]>;
5318 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5319 (ins VR128:$src1, i128mem:$src2),
5321 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5322 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5324 (IntId128 VR128:$src1,
5325 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize,
5326 Sched<[itins.Sched.Folded, ReadAfterLd]>;
5329 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
5330 Intrinsic IntId256> {
5331 let isCommutable = 1 in
5332 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5333 (ins VR256:$src1, VR256:$src2),
5334 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5335 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
5337 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5338 (ins VR256:$src1, i256mem:$src2),
5339 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5341 (IntId256 VR256:$src1,
5342 (bitconvert (loadv4i64 addr:$src2))))]>, OpSize;
5345 let ImmT = NoImm, Predicates = [HasAVX] in {
5346 let isCommutable = 0 in {
5347 defm VPHADDW : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v8i16, VR128,
5349 SSE_PHADDSUBW, 0>, VEX_4V;
5350 defm VPHADDD : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v4i32, VR128,
5352 SSE_PHADDSUBD, 0>, VEX_4V;
5353 defm VPHSUBW : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v8i16, VR128,
5355 SSE_PHADDSUBW, 0>, VEX_4V;
5356 defm VPHSUBD : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v4i32, VR128,
5358 SSE_PHADDSUBD, 0>, VEX_4V;
5359 defm VPSIGNB : SS3I_binop_rm<0x08, "vpsignb", X86psign, v16i8, VR128,
5361 SSE_PSIGN, 0>, VEX_4V;
5362 defm VPSIGNW : SS3I_binop_rm<0x09, "vpsignw", X86psign, v8i16, VR128,
5364 SSE_PSIGN, 0>, VEX_4V;
5365 defm VPSIGND : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v4i32, VR128,
5367 SSE_PSIGN, 0>, VEX_4V;
5368 defm VPSHUFB : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v16i8, VR128,
5370 SSE_PSHUFB, 0>, VEX_4V;
5371 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw",
5372 int_x86_ssse3_phadd_sw_128,
5373 SSE_PHADDSUBSW, 0>, VEX_4V;
5374 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw",
5375 int_x86_ssse3_phsub_sw_128,
5376 SSE_PHADDSUBSW, 0>, VEX_4V;
5377 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw",
5378 int_x86_ssse3_pmadd_ub_sw_128,
5379 SSE_PMADD, 0>, VEX_4V;
5381 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw",
5382 int_x86_ssse3_pmul_hr_sw_128,
5383 SSE_PMULHRSW, 0>, VEX_4V;
5386 let ImmT = NoImm, Predicates = [HasAVX2] in {
5387 let isCommutable = 0 in {
5388 defm VPHADDWY : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v16i16, VR256,
5390 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5391 defm VPHADDDY : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v8i32, VR256,
5393 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5394 defm VPHSUBWY : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v16i16, VR256,
5396 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5397 defm VPHSUBDY : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v8i32, VR256,
5399 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5400 defm VPSIGNBY : SS3I_binop_rm<0x08, "vpsignb", X86psign, v32i8, VR256,
5402 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5403 defm VPSIGNWY : SS3I_binop_rm<0x09, "vpsignw", X86psign, v16i16, VR256,
5405 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5406 defm VPSIGNDY : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v8i32, VR256,
5408 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5409 defm VPSHUFBY : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v32i8, VR256,
5411 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5412 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
5413 int_x86_avx2_phadd_sw>, VEX_4V, VEX_L;
5414 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
5415 int_x86_avx2_phsub_sw>, VEX_4V, VEX_L;
5416 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
5417 int_x86_avx2_pmadd_ub_sw>, VEX_4V, VEX_L;
5419 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
5420 int_x86_avx2_pmul_hr_sw>, VEX_4V, VEX_L;
5423 // None of these have i8 immediate fields.
5424 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
5425 let isCommutable = 0 in {
5426 defm PHADDW : SS3I_binop_rm<0x01, "phaddw", X86hadd, v8i16, VR128,
5427 memopv2i64, i128mem, SSE_PHADDSUBW>;
5428 defm PHADDD : SS3I_binop_rm<0x02, "phaddd", X86hadd, v4i32, VR128,
5429 memopv2i64, i128mem, SSE_PHADDSUBD>;
5430 defm PHSUBW : SS3I_binop_rm<0x05, "phsubw", X86hsub, v8i16, VR128,
5431 memopv2i64, i128mem, SSE_PHADDSUBW>;
5432 defm PHSUBD : SS3I_binop_rm<0x06, "phsubd", X86hsub, v4i32, VR128,
5433 memopv2i64, i128mem, SSE_PHADDSUBD>;
5434 defm PSIGNB : SS3I_binop_rm<0x08, "psignb", X86psign, v16i8, VR128,
5435 memopv2i64, i128mem, SSE_PSIGN>;
5436 defm PSIGNW : SS3I_binop_rm<0x09, "psignw", X86psign, v8i16, VR128,
5437 memopv2i64, i128mem, SSE_PSIGN>;
5438 defm PSIGND : SS3I_binop_rm<0x0A, "psignd", X86psign, v4i32, VR128,
5439 memopv2i64, i128mem, SSE_PSIGN>;
5440 defm PSHUFB : SS3I_binop_rm<0x00, "pshufb", X86pshufb, v16i8, VR128,
5441 memopv2i64, i128mem, SSE_PSHUFB>;
5442 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw",
5443 int_x86_ssse3_phadd_sw_128,
5445 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw",
5446 int_x86_ssse3_phsub_sw_128,
5448 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw",
5449 int_x86_ssse3_pmadd_ub_sw_128, SSE_PMADD>;
5451 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw",
5452 int_x86_ssse3_pmul_hr_sw_128,
5456 //===---------------------------------------------------------------------===//
5457 // SSSE3 - Packed Align Instruction Patterns
5458 //===---------------------------------------------------------------------===//
5460 multiclass ssse3_palignr<string asm, bit Is2Addr = 1> {
5461 let neverHasSideEffects = 1 in {
5462 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
5463 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5465 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5467 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5468 [], IIC_SSE_PALIGNRR>, OpSize, Sched<[WriteShuffle]>;
5470 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
5471 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5473 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5475 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5476 [], IIC_SSE_PALIGNRM>, OpSize, Sched<[WriteShuffleLd, ReadAfterLd]>;
5480 multiclass ssse3_palignr_y<string asm, bit Is2Addr = 1> {
5481 let neverHasSideEffects = 1 in {
5482 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
5483 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5485 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5486 []>, OpSize, Sched<[WriteShuffle]>;
5488 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
5489 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
5491 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5492 []>, OpSize, Sched<[WriteShuffleLd, ReadAfterLd]>;
5496 let Predicates = [HasAVX] in
5497 defm VPALIGN : ssse3_palignr<"vpalignr", 0>, VEX_4V;
5498 let Predicates = [HasAVX2] in
5499 defm VPALIGN : ssse3_palignr_y<"vpalignr", 0>, VEX_4V, VEX_L;
5500 let Constraints = "$src1 = $dst", Predicates = [UseSSSE3] in
5501 defm PALIGN : ssse3_palignr<"palignr">;
5503 let Predicates = [HasAVX2] in {
5504 def : Pat<(v8i32 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5505 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5506 def : Pat<(v8f32 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5507 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5508 def : Pat<(v16i16 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5509 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5510 def : Pat<(v32i8 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5511 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5514 let Predicates = [HasAVX] in {
5515 def : Pat<(v4i32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5516 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5517 def : Pat<(v4f32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5518 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5519 def : Pat<(v8i16 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5520 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5521 def : Pat<(v16i8 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5522 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5525 let Predicates = [UseSSSE3] in {
5526 def : Pat<(v4i32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5527 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5528 def : Pat<(v4f32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5529 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5530 def : Pat<(v8i16 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5531 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5532 def : Pat<(v16i8 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5533 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5536 //===---------------------------------------------------------------------===//
5537 // SSSE3 - Thread synchronization
5538 //===---------------------------------------------------------------------===//
5540 let SchedRW = [WriteSystem] in {
5541 let usesCustomInserter = 1 in {
5542 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5543 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>,
5544 Requires<[HasSSE3]>;
5547 let Uses = [EAX, ECX, EDX] in
5548 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", [], IIC_SSE_MONITOR>,
5549 TB, Requires<[HasSSE3]>;
5550 let Uses = [ECX, EAX] in
5551 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait",
5552 [(int_x86_sse3_mwait ECX, EAX)], IIC_SSE_MWAIT>,
5553 TB, Requires<[HasSSE3]>;
5556 def : InstAlias<"mwait\t{%eax, %ecx|ecx, eax}", (MWAITrr)>, Requires<[In32BitMode]>;
5557 def : InstAlias<"mwait\t{%rax, %rcx|rcx, rax}", (MWAITrr)>, Requires<[In64BitMode]>;
5559 def : InstAlias<"monitor\t{%eax, %ecx, %edx|edx, ecx, eax}", (MONITORrrr)>,
5560 Requires<[In32BitMode]>;
5561 def : InstAlias<"monitor\t{%rax, %rcx, %rdx|rdx, rcx, rax}", (MONITORrrr)>,
5562 Requires<[In64BitMode]>;
5564 //===----------------------------------------------------------------------===//
5565 // SSE4.1 - Packed Move with Sign/Zero Extend
5566 //===----------------------------------------------------------------------===//
5568 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId,
5569 OpndItins itins = DEFAULT_ITINS> {
5570 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5571 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5572 [(set VR128:$dst, (IntId VR128:$src))], itins.rr>, OpSize;
5574 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5575 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5577 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))],
5581 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5583 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5584 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5585 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5587 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5588 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5589 [(set VR256:$dst, (IntId (load addr:$src)))]>,
5593 let Predicates = [HasAVX] in {
5594 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw",
5595 int_x86_sse41_pmovsxbw>, VEX;
5596 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd",
5597 int_x86_sse41_pmovsxwd>, VEX;
5598 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq",
5599 int_x86_sse41_pmovsxdq>, VEX;
5600 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw",
5601 int_x86_sse41_pmovzxbw>, VEX;
5602 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd",
5603 int_x86_sse41_pmovzxwd>, VEX;
5604 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq",
5605 int_x86_sse41_pmovzxdq>, VEX;
5608 let Predicates = [HasAVX2] in {
5609 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5610 int_x86_avx2_pmovsxbw>, VEX, VEX_L;
5611 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5612 int_x86_avx2_pmovsxwd>, VEX, VEX_L;
5613 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5614 int_x86_avx2_pmovsxdq>, VEX, VEX_L;
5615 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5616 int_x86_avx2_pmovzxbw>, VEX, VEX_L;
5617 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5618 int_x86_avx2_pmovzxwd>, VEX, VEX_L;
5619 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5620 int_x86_avx2_pmovzxdq>, VEX, VEX_L;
5623 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw, SSE_INTALU_ITINS_P>;
5624 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd, SSE_INTALU_ITINS_P>;
5625 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq, SSE_INTALU_ITINS_P>;
5626 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw, SSE_INTALU_ITINS_P>;
5627 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd, SSE_INTALU_ITINS_P>;
5628 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq, SSE_INTALU_ITINS_P>;
5630 let Predicates = [HasAVX] in {
5631 // Common patterns involving scalar load.
5632 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5633 (VPMOVSXBWrm addr:$src)>;
5634 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5635 (VPMOVSXBWrm addr:$src)>;
5636 def : Pat<(int_x86_sse41_pmovsxbw (bc_v16i8 (loadv2i64 addr:$src))),
5637 (VPMOVSXBWrm addr:$src)>;
5639 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5640 (VPMOVSXWDrm addr:$src)>;
5641 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5642 (VPMOVSXWDrm addr:$src)>;
5643 def : Pat<(int_x86_sse41_pmovsxwd (bc_v8i16 (loadv2i64 addr:$src))),
5644 (VPMOVSXWDrm addr:$src)>;
5646 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5647 (VPMOVSXDQrm addr:$src)>;
5648 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5649 (VPMOVSXDQrm addr:$src)>;
5650 def : Pat<(int_x86_sse41_pmovsxdq (bc_v4i32 (loadv2i64 addr:$src))),
5651 (VPMOVSXDQrm addr:$src)>;
5653 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5654 (VPMOVZXBWrm addr:$src)>;
5655 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5656 (VPMOVZXBWrm addr:$src)>;
5657 def : Pat<(int_x86_sse41_pmovzxbw (bc_v16i8 (loadv2i64 addr:$src))),
5658 (VPMOVZXBWrm addr:$src)>;
5660 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5661 (VPMOVZXWDrm addr:$src)>;
5662 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5663 (VPMOVZXWDrm addr:$src)>;
5664 def : Pat<(int_x86_sse41_pmovzxwd (bc_v8i16 (loadv2i64 addr:$src))),
5665 (VPMOVZXWDrm addr:$src)>;
5667 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5668 (VPMOVZXDQrm addr:$src)>;
5669 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5670 (VPMOVZXDQrm addr:$src)>;
5671 def : Pat<(int_x86_sse41_pmovzxdq (bc_v4i32 (loadv2i64 addr:$src))),
5672 (VPMOVZXDQrm addr:$src)>;
5675 let Predicates = [UseSSE41] in {
5676 // Common patterns involving scalar load.
5677 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5678 (PMOVSXBWrm addr:$src)>;
5679 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5680 (PMOVSXBWrm addr:$src)>;
5681 def : Pat<(int_x86_sse41_pmovsxbw (bc_v16i8 (loadv2i64 addr:$src))),
5682 (PMOVSXBWrm addr:$src)>;
5684 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5685 (PMOVSXWDrm addr:$src)>;
5686 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5687 (PMOVSXWDrm addr:$src)>;
5688 def : Pat<(int_x86_sse41_pmovsxwd (bc_v8i16 (loadv2i64 addr:$src))),
5689 (PMOVSXWDrm addr:$src)>;
5691 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5692 (PMOVSXDQrm addr:$src)>;
5693 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5694 (PMOVSXDQrm addr:$src)>;
5695 def : Pat<(int_x86_sse41_pmovsxdq (bc_v4i32 (loadv2i64 addr:$src))),
5696 (PMOVSXDQrm addr:$src)>;
5698 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5699 (PMOVZXBWrm addr:$src)>;
5700 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5701 (PMOVZXBWrm addr:$src)>;
5702 def : Pat<(int_x86_sse41_pmovzxbw (bc_v16i8 (loadv2i64 addr:$src))),
5703 (PMOVZXBWrm addr:$src)>;
5705 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5706 (PMOVZXWDrm addr:$src)>;
5707 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5708 (PMOVZXWDrm addr:$src)>;
5709 def : Pat<(int_x86_sse41_pmovzxwd (bc_v8i16 (loadv2i64 addr:$src))),
5710 (PMOVZXWDrm addr:$src)>;
5712 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5713 (PMOVZXDQrm addr:$src)>;
5714 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5715 (PMOVZXDQrm addr:$src)>;
5716 def : Pat<(int_x86_sse41_pmovzxdq (bc_v4i32 (loadv2i64 addr:$src))),
5717 (PMOVZXDQrm addr:$src)>;
5720 let Predicates = [HasAVX2] in {
5721 let AddedComplexity = 15 in {
5722 def : Pat<(v4i64 (X86vzmovly (v4i32 VR128:$src))),
5723 (VPMOVZXDQYrr VR128:$src)>;
5724 def : Pat<(v8i32 (X86vzmovly (v8i16 VR128:$src))),
5725 (VPMOVZXWDYrr VR128:$src)>;
5726 def : Pat<(v16i16 (X86vzmovly (v16i8 VR128:$src))),
5727 (VPMOVZXBWYrr VR128:$src)>;
5730 def : Pat<(v4i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>;
5731 def : Pat<(v8i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>;
5732 def : Pat<(v16i16 (X86vsmovl (v16i8 VR128:$src))), (VPMOVSXBWYrr VR128:$src)>;
5735 let Predicates = [HasAVX] in {
5736 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
5737 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
5738 def : Pat<(v8i16 (X86vsmovl (v16i8 VR128:$src))), (VPMOVSXBWrr VR128:$src)>;
5741 let Predicates = [UseSSE41] in {
5742 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
5743 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
5744 def : Pat<(v8i16 (X86vsmovl (v16i8 VR128:$src))), (PMOVSXBWrr VR128:$src)>;
5748 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId,
5749 OpndItins itins = DEFAULT_ITINS> {
5750 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5751 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5752 [(set VR128:$dst, (IntId VR128:$src))], itins.rr>, OpSize;
5754 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5755 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5757 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))],
5762 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5764 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5765 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5766 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5768 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5769 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5771 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5775 let Predicates = [HasAVX] in {
5776 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5778 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5780 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5782 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5786 let Predicates = [HasAVX2] in {
5787 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5788 int_x86_avx2_pmovsxbd>, VEX, VEX_L;
5789 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5790 int_x86_avx2_pmovsxwq>, VEX, VEX_L;
5791 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5792 int_x86_avx2_pmovzxbd>, VEX, VEX_L;
5793 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5794 int_x86_avx2_pmovzxwq>, VEX, VEX_L;
5797 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd,
5798 SSE_INTALU_ITINS_P>;
5799 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq,
5800 SSE_INTALU_ITINS_P>;
5801 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd,
5802 SSE_INTALU_ITINS_P>;
5803 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq,
5804 SSE_INTALU_ITINS_P>;
5806 let Predicates = [HasAVX] in {
5807 // Common patterns involving scalar load
5808 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5809 (VPMOVSXBDrm addr:$src)>;
5810 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5811 (VPMOVSXWQrm addr:$src)>;
5813 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5814 (VPMOVZXBDrm addr:$src)>;
5815 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5816 (VPMOVZXWQrm addr:$src)>;
5819 let Predicates = [UseSSE41] in {
5820 // Common patterns involving scalar load
5821 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5822 (PMOVSXBDrm addr:$src)>;
5823 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5824 (PMOVSXWQrm addr:$src)>;
5826 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5827 (PMOVZXBDrm addr:$src)>;
5828 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5829 (PMOVZXWQrm addr:$src)>;
5832 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId,
5833 OpndItins itins = DEFAULT_ITINS> {
5834 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5835 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5836 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5838 // Expecting a i16 load any extended to i32 value.
5839 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5840 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5841 [(set VR128:$dst, (IntId (bitconvert
5842 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5846 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5848 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5849 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5850 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5852 // Expecting a i16 load any extended to i32 value.
5853 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5854 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5855 [(set VR256:$dst, (IntId (bitconvert
5856 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5860 let Predicates = [HasAVX] in {
5861 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5863 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5866 let Predicates = [HasAVX2] in {
5867 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5868 int_x86_avx2_pmovsxbq>, VEX, VEX_L;
5869 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5870 int_x86_avx2_pmovzxbq>, VEX, VEX_L;
5872 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq,
5873 SSE_INTALU_ITINS_P>;
5874 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq,
5875 SSE_INTALU_ITINS_P>;
5877 let Predicates = [HasAVX2] in {
5878 def : Pat<(v16i16 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBWYrr VR128:$src)>;
5879 def : Pat<(v8i32 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBDYrr VR128:$src)>;
5880 def : Pat<(v4i64 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBQYrr VR128:$src)>;
5882 def : Pat<(v8i32 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>;
5883 def : Pat<(v4i64 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWQYrr VR128:$src)>;
5885 def : Pat<(v4i64 (X86vsext (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>;
5887 def : Pat<(v16i16 (X86vsext (v32i8 VR256:$src))),
5888 (VPMOVSXBWYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5889 def : Pat<(v8i32 (X86vsext (v32i8 VR256:$src))),
5890 (VPMOVSXBDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5891 def : Pat<(v4i64 (X86vsext (v32i8 VR256:$src))),
5892 (VPMOVSXBQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5894 def : Pat<(v8i32 (X86vsext (v16i16 VR256:$src))),
5895 (VPMOVSXWDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5896 def : Pat<(v4i64 (X86vsext (v16i16 VR256:$src))),
5897 (VPMOVSXWQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5899 def : Pat<(v4i64 (X86vsext (v8i32 VR256:$src))),
5900 (VPMOVSXDQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5902 def : Pat<(v8i32 (X86vsmovl (v8i16 (bitconvert (v2i64 (load addr:$src)))))),
5903 (VPMOVSXWDYrm addr:$src)>;
5904 def : Pat<(v4i64 (X86vsmovl (v4i32 (bitconvert (v2i64 (load addr:$src)))))),
5905 (VPMOVSXDQYrm addr:$src)>;
5907 def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2i64
5908 (scalar_to_vector (loadi64 addr:$src))))))),
5909 (VPMOVSXBDYrm addr:$src)>;
5910 def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2f64
5911 (scalar_to_vector (loadf64 addr:$src))))))),
5912 (VPMOVSXBDYrm addr:$src)>;
5914 def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2i64
5915 (scalar_to_vector (loadi64 addr:$src))))))),
5916 (VPMOVSXWQYrm addr:$src)>;
5917 def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2f64
5918 (scalar_to_vector (loadf64 addr:$src))))))),
5919 (VPMOVSXWQYrm addr:$src)>;
5921 def : Pat<(v4i64 (X86vsext (v16i8 (bitconvert (v4i32
5922 (scalar_to_vector (loadi32 addr:$src))))))),
5923 (VPMOVSXBQYrm addr:$src)>;
5926 let Predicates = [HasAVX] in {
5927 // Common patterns involving scalar load
5928 def : Pat<(int_x86_sse41_pmovsxbq
5929 (bitconvert (v4i32 (X86vzmovl
5930 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5931 (VPMOVSXBQrm addr:$src)>;
5933 def : Pat<(int_x86_sse41_pmovzxbq
5934 (bitconvert (v4i32 (X86vzmovl
5935 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5936 (VPMOVZXBQrm addr:$src)>;
5939 let Predicates = [UseSSE41] in {
5940 def : Pat<(v8i16 (X86vsext (v16i8 VR128:$src))), (PMOVSXBWrr VR128:$src)>;
5941 def : Pat<(v4i32 (X86vsext (v16i8 VR128:$src))), (PMOVSXBDrr VR128:$src)>;
5942 def : Pat<(v2i64 (X86vsext (v16i8 VR128:$src))), (PMOVSXBQrr VR128:$src)>;
5944 def : Pat<(v4i32 (X86vsext (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
5945 def : Pat<(v2i64 (X86vsext (v8i16 VR128:$src))), (PMOVSXWQrr VR128:$src)>;
5947 def : Pat<(v2i64 (X86vsext (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
5949 // Common patterns involving scalar load
5950 def : Pat<(int_x86_sse41_pmovsxbq
5951 (bitconvert (v4i32 (X86vzmovl
5952 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5953 (PMOVSXBQrm addr:$src)>;
5955 def : Pat<(int_x86_sse41_pmovzxbq
5956 (bitconvert (v4i32 (X86vzmovl
5957 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5958 (PMOVZXBQrm addr:$src)>;
5960 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64
5961 (scalar_to_vector (loadi64 addr:$src))))))),
5962 (PMOVSXWDrm addr:$src)>;
5963 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64
5964 (scalar_to_vector (loadf64 addr:$src))))))),
5965 (PMOVSXWDrm addr:$src)>;
5966 def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32
5967 (scalar_to_vector (loadi32 addr:$src))))))),
5968 (PMOVSXBDrm addr:$src)>;
5969 def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32
5970 (scalar_to_vector (loadi32 addr:$src))))))),
5971 (PMOVSXWQrm addr:$src)>;
5972 def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32
5973 (scalar_to_vector (extloadi32i16 addr:$src))))))),
5974 (PMOVSXBQrm addr:$src)>;
5975 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64
5976 (scalar_to_vector (loadi64 addr:$src))))))),
5977 (PMOVSXDQrm addr:$src)>;
5978 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64
5979 (scalar_to_vector (loadf64 addr:$src))))))),
5980 (PMOVSXDQrm addr:$src)>;
5981 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64
5982 (scalar_to_vector (loadi64 addr:$src))))))),
5983 (PMOVSXBWrm addr:$src)>;
5984 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64
5985 (scalar_to_vector (loadf64 addr:$src))))))),
5986 (PMOVSXBWrm addr:$src)>;
5989 let Predicates = [HasAVX2] in {
5990 def : Pat<(v16i16 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBWYrr VR128:$src)>;
5991 def : Pat<(v8i32 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBDYrr VR128:$src)>;
5992 def : Pat<(v4i64 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBQYrr VR128:$src)>;
5994 def : Pat<(v8i32 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWDYrr VR128:$src)>;
5995 def : Pat<(v4i64 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWQYrr VR128:$src)>;
5997 def : Pat<(v4i64 (X86vzext (v4i32 VR128:$src))), (VPMOVZXDQYrr VR128:$src)>;
5999 def : Pat<(v16i16 (X86vzext (v32i8 VR256:$src))),
6000 (VPMOVZXBWYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
6001 def : Pat<(v8i32 (X86vzext (v32i8 VR256:$src))),
6002 (VPMOVZXBDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
6003 def : Pat<(v4i64 (X86vzext (v32i8 VR256:$src))),
6004 (VPMOVZXBQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
6006 def : Pat<(v8i32 (X86vzext (v16i16 VR256:$src))),
6007 (VPMOVZXWDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
6008 def : Pat<(v4i64 (X86vzext (v16i16 VR256:$src))),
6009 (VPMOVZXWQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
6011 def : Pat<(v4i64 (X86vzext (v8i32 VR256:$src))),
6012 (VPMOVZXDQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
6015 let Predicates = [HasAVX] in {
6016 def : Pat<(v8i16 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBWrr VR128:$src)>;
6017 def : Pat<(v4i32 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBDrr VR128:$src)>;
6018 def : Pat<(v2i64 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBQrr VR128:$src)>;
6020 def : Pat<(v4i32 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWDrr VR128:$src)>;
6021 def : Pat<(v2i64 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWQrr VR128:$src)>;
6023 def : Pat<(v2i64 (X86vzext (v4i32 VR128:$src))), (VPMOVZXDQrr VR128:$src)>;
6025 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
6026 (VPMOVZXBWrm addr:$src)>;
6027 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
6028 (VPMOVZXBWrm addr:$src)>;
6029 def : Pat<(v4i32 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
6030 (VPMOVZXBDrm addr:$src)>;
6031 def : Pat<(v2i64 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))),
6032 (VPMOVZXBQrm addr:$src)>;
6034 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
6035 (VPMOVZXWDrm addr:$src)>;
6036 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
6037 (VPMOVZXWDrm addr:$src)>;
6038 def : Pat<(v2i64 (X86vzext (v8i16 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
6039 (VPMOVZXWQrm addr:$src)>;
6041 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
6042 (VPMOVZXDQrm addr:$src)>;
6043 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
6044 (VPMOVZXDQrm addr:$src)>;
6045 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (X86vzload addr:$src)))))),
6046 (VPMOVZXDQrm addr:$src)>;
6048 def : Pat<(v8i16 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBWrr VR128:$src)>;
6049 def : Pat<(v4i32 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBDrr VR128:$src)>;
6050 def : Pat<(v2i64 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBQrr VR128:$src)>;
6052 def : Pat<(v4i32 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
6053 def : Pat<(v2i64 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWQrr VR128:$src)>;
6055 def : Pat<(v2i64 (X86vsext (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
6057 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64
6058 (scalar_to_vector (loadi64 addr:$src))))))),
6059 (VPMOVSXWDrm addr:$src)>;
6060 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64
6061 (scalar_to_vector (loadi64 addr:$src))))))),
6062 (VPMOVSXDQrm addr:$src)>;
6063 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64
6064 (scalar_to_vector (loadf64 addr:$src))))))),
6065 (VPMOVSXWDrm addr:$src)>;
6066 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64
6067 (scalar_to_vector (loadf64 addr:$src))))))),
6068 (VPMOVSXDQrm addr:$src)>;
6069 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64
6070 (scalar_to_vector (loadi64 addr:$src))))))),
6071 (VPMOVSXBWrm addr:$src)>;
6072 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64
6073 (scalar_to_vector (loadf64 addr:$src))))))),
6074 (VPMOVSXBWrm addr:$src)>;
6076 def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32
6077 (scalar_to_vector (loadi32 addr:$src))))))),
6078 (VPMOVSXBDrm addr:$src)>;
6079 def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32
6080 (scalar_to_vector (loadi32 addr:$src))))))),
6081 (VPMOVSXWQrm addr:$src)>;
6082 def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32
6083 (scalar_to_vector (extloadi32i16 addr:$src))))))),
6084 (VPMOVSXBQrm addr:$src)>;
6087 let Predicates = [UseSSE41] in {
6088 def : Pat<(v8i16 (X86vzext (v16i8 VR128:$src))), (PMOVZXBWrr VR128:$src)>;
6089 def : Pat<(v4i32 (X86vzext (v16i8 VR128:$src))), (PMOVZXBDrr VR128:$src)>;
6090 def : Pat<(v2i64 (X86vzext (v16i8 VR128:$src))), (PMOVZXBQrr VR128:$src)>;
6092 def : Pat<(v4i32 (X86vzext (v8i16 VR128:$src))), (PMOVZXWDrr VR128:$src)>;
6093 def : Pat<(v2i64 (X86vzext (v8i16 VR128:$src))), (PMOVZXWQrr VR128:$src)>;
6095 def : Pat<(v2i64 (X86vzext (v4i32 VR128:$src))), (PMOVZXDQrr VR128:$src)>;
6097 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
6098 (PMOVZXBWrm addr:$src)>;
6099 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
6100 (PMOVZXBWrm addr:$src)>;
6101 def : Pat<(v4i32 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
6102 (PMOVZXBDrm addr:$src)>;
6103 def : Pat<(v2i64 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))),
6104 (PMOVZXBQrm addr:$src)>;
6106 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
6107 (PMOVZXWDrm addr:$src)>;
6108 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
6109 (PMOVZXWDrm addr:$src)>;
6110 def : Pat<(v2i64 (X86vzext (v8i16 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
6111 (PMOVZXWQrm addr:$src)>;
6113 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
6114 (PMOVZXDQrm addr:$src)>;
6115 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
6116 (PMOVZXDQrm addr:$src)>;
6117 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (X86vzload addr:$src)))))),
6118 (PMOVZXDQrm addr:$src)>;
6121 //===----------------------------------------------------------------------===//
6122 // SSE4.1 - Extract Instructions
6123 //===----------------------------------------------------------------------===//
6125 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
6126 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
6127 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32orGR64:$dst),
6128 (ins VR128:$src1, i32i8imm:$src2),
6129 !strconcat(OpcodeStr,
6130 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6131 [(set GR32orGR64:$dst, (X86pextrb (v16i8 VR128:$src1),
6134 let neverHasSideEffects = 1, mayStore = 1 in
6135 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6136 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
6137 !strconcat(OpcodeStr,
6138 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6141 // There's an AssertZext in the way of writing the store pattern
6142 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
6145 let Predicates = [HasAVX] in
6146 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
6148 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
6151 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
6152 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
6153 let isCodeGenOnly = 1, hasSideEffects = 0 in
6154 def rr_REV : SS4AIi8<opc, MRMDestReg, (outs GR32orGR64:$dst),
6155 (ins VR128:$src1, i32i8imm:$src2),
6156 !strconcat(OpcodeStr,
6157 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6160 let neverHasSideEffects = 1, mayStore = 1 in
6161 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6162 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
6163 !strconcat(OpcodeStr,
6164 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6167 // There's an AssertZext in the way of writing the store pattern
6168 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
6171 let Predicates = [HasAVX] in
6172 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
6174 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
6177 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
6178 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
6179 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
6180 (ins VR128:$src1, i32i8imm:$src2),
6181 !strconcat(OpcodeStr,
6182 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6184 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
6185 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6186 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
6187 !strconcat(OpcodeStr,
6188 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6189 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
6190 addr:$dst)]>, OpSize;
6193 let Predicates = [HasAVX] in
6194 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
6196 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
6198 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
6199 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
6200 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
6201 (ins VR128:$src1, i32i8imm:$src2),
6202 !strconcat(OpcodeStr,
6203 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6205 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
6206 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6207 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
6208 !strconcat(OpcodeStr,
6209 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6210 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
6211 addr:$dst)]>, OpSize, REX_W;
6214 let Predicates = [HasAVX] in
6215 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
6217 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
6219 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
6221 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr,
6222 OpndItins itins = DEFAULT_ITINS> {
6223 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32orGR64:$dst),
6224 (ins VR128:$src1, i32i8imm:$src2),
6225 !strconcat(OpcodeStr,
6226 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6227 [(set GR32orGR64:$dst,
6228 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))],
6231 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6232 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
6233 !strconcat(OpcodeStr,
6234 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6235 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
6236 addr:$dst)], itins.rm>, OpSize;
6239 let ExeDomain = SSEPackedSingle in {
6240 let Predicates = [UseAVX] in
6241 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
6242 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps", SSE_EXTRACT_ITINS>;
6245 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
6246 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
6249 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
6251 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
6254 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
6255 Requires<[UseSSE41]>;
6257 //===----------------------------------------------------------------------===//
6258 // SSE4.1 - Insert Instructions
6259 //===----------------------------------------------------------------------===//
6261 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
6262 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6263 (ins VR128:$src1, GR32orGR64:$src2, i32i8imm:$src3),
6265 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6267 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6269 (X86pinsrb VR128:$src1, GR32orGR64:$src2, imm:$src3))]>, OpSize;
6270 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6271 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
6273 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6275 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6277 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
6278 imm:$src3))]>, OpSize;
6281 let Predicates = [HasAVX] in
6282 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
6283 let Constraints = "$src1 = $dst" in
6284 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
6286 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
6287 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6288 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
6290 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6292 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6294 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
6296 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6297 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
6299 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6301 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6303 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
6304 imm:$src3)))]>, OpSize;
6307 let Predicates = [HasAVX] in
6308 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
6309 let Constraints = "$src1 = $dst" in
6310 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
6312 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
6313 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6314 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
6316 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6318 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6320 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
6322 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6323 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
6325 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6327 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6329 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
6330 imm:$src3)))]>, OpSize;
6333 let Predicates = [HasAVX] in
6334 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
6335 let Constraints = "$src1 = $dst" in
6336 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
6338 // insertps has a few different modes, there's the first two here below which
6339 // are optimized inserts that won't zero arbitrary elements in the destination
6340 // vector. The next one matches the intrinsic and could zero arbitrary elements
6341 // in the target vector.
6342 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1,
6343 OpndItins itins = DEFAULT_ITINS> {
6344 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6345 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
6347 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6349 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6351 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))], itins.rr>,
6353 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6354 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
6356 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6358 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6360 (X86insrtps VR128:$src1,
6361 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
6362 imm:$src3))], itins.rm>, OpSize;
6365 let ExeDomain = SSEPackedSingle in {
6366 let Predicates = [UseAVX] in
6367 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
6368 let Constraints = "$src1 = $dst" in
6369 defm INSERTPS : SS41I_insertf32<0x21, "insertps", 1, SSE_INSERT_ITINS>;
6372 //===----------------------------------------------------------------------===//
6373 // SSE4.1 - Round Instructions
6374 //===----------------------------------------------------------------------===//
6376 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
6377 X86MemOperand x86memop, RegisterClass RC,
6378 PatFrag mem_frag32, PatFrag mem_frag64,
6379 Intrinsic V4F32Int, Intrinsic V2F64Int> {
6380 let ExeDomain = SSEPackedSingle in {
6381 // Intrinsic operation, reg.
6382 // Vector intrinsic operation, reg
6383 def PSr : SS4AIi8<opcps, MRMSrcReg,
6384 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6385 !strconcat(OpcodeStr,
6386 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6387 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))],
6388 IIC_SSE_ROUNDPS_REG>,
6391 // Vector intrinsic operation, mem
6392 def PSm : SS4AIi8<opcps, MRMSrcMem,
6393 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6394 !strconcat(OpcodeStr,
6395 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6397 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))],
6398 IIC_SSE_ROUNDPS_MEM>,
6400 } // ExeDomain = SSEPackedSingle
6402 let ExeDomain = SSEPackedDouble in {
6403 // Vector intrinsic operation, reg
6404 def PDr : SS4AIi8<opcpd, MRMSrcReg,
6405 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6406 !strconcat(OpcodeStr,
6407 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6408 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))],
6409 IIC_SSE_ROUNDPS_REG>,
6412 // Vector intrinsic operation, mem
6413 def PDm : SS4AIi8<opcpd, MRMSrcMem,
6414 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6415 !strconcat(OpcodeStr,
6416 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6418 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))],
6419 IIC_SSE_ROUNDPS_REG>,
6421 } // ExeDomain = SSEPackedDouble
6424 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
6427 Intrinsic F64Int, bit Is2Addr = 1> {
6428 let ExeDomain = GenericDomain in {
6430 let hasSideEffects = 0 in
6431 def SSr : SS4AIi8<opcss, MRMSrcReg,
6432 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3),
6434 !strconcat(OpcodeStr,
6435 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6436 !strconcat(OpcodeStr,
6437 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6440 // Intrinsic operation, reg.
6441 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
6442 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6444 !strconcat(OpcodeStr,
6445 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6446 !strconcat(OpcodeStr,
6447 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6448 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6451 // Intrinsic operation, mem.
6452 def SSm : SS4AIi8<opcss, MRMSrcMem,
6453 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
6455 !strconcat(OpcodeStr,
6456 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6457 !strconcat(OpcodeStr,
6458 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6460 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
6464 let hasSideEffects = 0 in
6465 def SDr : SS4AIi8<opcsd, MRMSrcReg,
6466 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3),
6468 !strconcat(OpcodeStr,
6469 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6470 !strconcat(OpcodeStr,
6471 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6474 // Intrinsic operation, reg.
6475 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
6476 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6478 !strconcat(OpcodeStr,
6479 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6480 !strconcat(OpcodeStr,
6481 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6482 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6485 // Intrinsic operation, mem.
6486 def SDm : SS4AIi8<opcsd, MRMSrcMem,
6487 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
6489 !strconcat(OpcodeStr,
6490 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6491 !strconcat(OpcodeStr,
6492 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6494 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
6496 } // ExeDomain = GenericDomain
6499 // FP round - roundss, roundps, roundsd, roundpd
6500 let Predicates = [HasAVX] in {
6502 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
6503 loadv4f32, loadv2f64,
6504 int_x86_sse41_round_ps,
6505 int_x86_sse41_round_pd>, VEX;
6506 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
6507 loadv8f32, loadv4f64,
6508 int_x86_avx_round_ps_256,
6509 int_x86_avx_round_pd_256>, VEX, VEX_L;
6510 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
6511 int_x86_sse41_round_ss,
6512 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
6514 def : Pat<(ffloor FR32:$src),
6515 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6516 def : Pat<(f64 (ffloor FR64:$src)),
6517 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6518 def : Pat<(f32 (fnearbyint FR32:$src)),
6519 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6520 def : Pat<(f64 (fnearbyint FR64:$src)),
6521 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6522 def : Pat<(f32 (fceil FR32:$src)),
6523 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6524 def : Pat<(f64 (fceil FR64:$src)),
6525 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6526 def : Pat<(f32 (frint FR32:$src)),
6527 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6528 def : Pat<(f64 (frint FR64:$src)),
6529 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6530 def : Pat<(f32 (ftrunc FR32:$src)),
6531 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6532 def : Pat<(f64 (ftrunc FR64:$src)),
6533 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6535 def : Pat<(v4f32 (ffloor VR128:$src)),
6536 (VROUNDPSr VR128:$src, (i32 0x1))>;
6537 def : Pat<(v4f32 (fnearbyint VR128:$src)),
6538 (VROUNDPSr VR128:$src, (i32 0xC))>;
6539 def : Pat<(v4f32 (fceil VR128:$src)),
6540 (VROUNDPSr VR128:$src, (i32 0x2))>;
6541 def : Pat<(v4f32 (frint VR128:$src)),
6542 (VROUNDPSr VR128:$src, (i32 0x4))>;
6543 def : Pat<(v4f32 (ftrunc VR128:$src)),
6544 (VROUNDPSr VR128:$src, (i32 0x3))>;
6546 def : Pat<(v2f64 (ffloor VR128:$src)),
6547 (VROUNDPDr VR128:$src, (i32 0x1))>;
6548 def : Pat<(v2f64 (fnearbyint VR128:$src)),
6549 (VROUNDPDr VR128:$src, (i32 0xC))>;
6550 def : Pat<(v2f64 (fceil VR128:$src)),
6551 (VROUNDPDr VR128:$src, (i32 0x2))>;
6552 def : Pat<(v2f64 (frint VR128:$src)),
6553 (VROUNDPDr VR128:$src, (i32 0x4))>;
6554 def : Pat<(v2f64 (ftrunc VR128:$src)),
6555 (VROUNDPDr VR128:$src, (i32 0x3))>;
6557 def : Pat<(v8f32 (ffloor VR256:$src)),
6558 (VROUNDYPSr VR256:$src, (i32 0x1))>;
6559 def : Pat<(v8f32 (fnearbyint VR256:$src)),
6560 (VROUNDYPSr VR256:$src, (i32 0xC))>;
6561 def : Pat<(v8f32 (fceil VR256:$src)),
6562 (VROUNDYPSr VR256:$src, (i32 0x2))>;
6563 def : Pat<(v8f32 (frint VR256:$src)),
6564 (VROUNDYPSr VR256:$src, (i32 0x4))>;
6565 def : Pat<(v8f32 (ftrunc VR256:$src)),
6566 (VROUNDYPSr VR256:$src, (i32 0x3))>;
6568 def : Pat<(v4f64 (ffloor VR256:$src)),
6569 (VROUNDYPDr VR256:$src, (i32 0x1))>;
6570 def : Pat<(v4f64 (fnearbyint VR256:$src)),
6571 (VROUNDYPDr VR256:$src, (i32 0xC))>;
6572 def : Pat<(v4f64 (fceil VR256:$src)),
6573 (VROUNDYPDr VR256:$src, (i32 0x2))>;
6574 def : Pat<(v4f64 (frint VR256:$src)),
6575 (VROUNDYPDr VR256:$src, (i32 0x4))>;
6576 def : Pat<(v4f64 (ftrunc VR256:$src)),
6577 (VROUNDYPDr VR256:$src, (i32 0x3))>;
6580 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
6581 memopv4f32, memopv2f64,
6582 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
6583 let Constraints = "$src1 = $dst" in
6584 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
6585 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
6587 let Predicates = [UseSSE41] in {
6588 def : Pat<(ffloor FR32:$src),
6589 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6590 def : Pat<(f64 (ffloor FR64:$src)),
6591 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6592 def : Pat<(f32 (fnearbyint FR32:$src)),
6593 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6594 def : Pat<(f64 (fnearbyint FR64:$src)),
6595 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6596 def : Pat<(f32 (fceil FR32:$src)),
6597 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6598 def : Pat<(f64 (fceil FR64:$src)),
6599 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6600 def : Pat<(f32 (frint FR32:$src)),
6601 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6602 def : Pat<(f64 (frint FR64:$src)),
6603 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6604 def : Pat<(f32 (ftrunc FR32:$src)),
6605 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6606 def : Pat<(f64 (ftrunc FR64:$src)),
6607 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6609 def : Pat<(v4f32 (ffloor VR128:$src)),
6610 (ROUNDPSr VR128:$src, (i32 0x1))>;
6611 def : Pat<(v4f32 (fnearbyint VR128:$src)),
6612 (ROUNDPSr VR128:$src, (i32 0xC))>;
6613 def : Pat<(v4f32 (fceil VR128:$src)),
6614 (ROUNDPSr VR128:$src, (i32 0x2))>;
6615 def : Pat<(v4f32 (frint VR128:$src)),
6616 (ROUNDPSr VR128:$src, (i32 0x4))>;
6617 def : Pat<(v4f32 (ftrunc VR128:$src)),
6618 (ROUNDPSr VR128:$src, (i32 0x3))>;
6620 def : Pat<(v2f64 (ffloor VR128:$src)),
6621 (ROUNDPDr VR128:$src, (i32 0x1))>;
6622 def : Pat<(v2f64 (fnearbyint VR128:$src)),
6623 (ROUNDPDr VR128:$src, (i32 0xC))>;
6624 def : Pat<(v2f64 (fceil VR128:$src)),
6625 (ROUNDPDr VR128:$src, (i32 0x2))>;
6626 def : Pat<(v2f64 (frint VR128:$src)),
6627 (ROUNDPDr VR128:$src, (i32 0x4))>;
6628 def : Pat<(v2f64 (ftrunc VR128:$src)),
6629 (ROUNDPDr VR128:$src, (i32 0x3))>;
6632 //===----------------------------------------------------------------------===//
6633 // SSE4.1 - Packed Bit Test
6634 //===----------------------------------------------------------------------===//
6636 // ptest instruction we'll lower to this in X86ISelLowering primarily from
6637 // the intel intrinsic that corresponds to this.
6638 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6639 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6640 "vptest\t{$src2, $src1|$src1, $src2}",
6641 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6643 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6644 "vptest\t{$src2, $src1|$src1, $src2}",
6645 [(set EFLAGS,(X86ptest VR128:$src1, (loadv2i64 addr:$src2)))]>,
6648 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
6649 "vptest\t{$src2, $src1|$src1, $src2}",
6650 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
6652 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
6653 "vptest\t{$src2, $src1|$src1, $src2}",
6654 [(set EFLAGS,(X86ptest VR256:$src1, (loadv4i64 addr:$src2)))]>,
6658 let Defs = [EFLAGS] in {
6659 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6660 "ptest\t{$src2, $src1|$src1, $src2}",
6661 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6663 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6664 "ptest\t{$src2, $src1|$src1, $src2}",
6665 [(set EFLAGS, (X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6669 // The bit test instructions below are AVX only
6670 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
6671 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
6672 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
6673 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6674 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
6675 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
6676 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6677 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
6681 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6682 let ExeDomain = SSEPackedSingle in {
6683 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, loadv4f32, v4f32>;
6684 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, loadv8f32, v8f32>,
6687 let ExeDomain = SSEPackedDouble in {
6688 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, loadv2f64, v2f64>;
6689 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, loadv4f64, v4f64>,
6694 //===----------------------------------------------------------------------===//
6695 // SSE4.1 - Misc Instructions
6696 //===----------------------------------------------------------------------===//
6698 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
6699 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
6700 "popcnt{w}\t{$src, $dst|$dst, $src}",
6701 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)],
6704 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
6705 "popcnt{w}\t{$src, $dst|$dst, $src}",
6706 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
6707 (implicit EFLAGS)], IIC_SSE_POPCNT_RM>, OpSize, XS;
6709 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
6710 "popcnt{l}\t{$src, $dst|$dst, $src}",
6711 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)],
6714 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
6715 "popcnt{l}\t{$src, $dst|$dst, $src}",
6716 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
6717 (implicit EFLAGS)], IIC_SSE_POPCNT_RM>, XS;
6719 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
6720 "popcnt{q}\t{$src, $dst|$dst, $src}",
6721 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)],
6724 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
6725 "popcnt{q}\t{$src, $dst|$dst, $src}",
6726 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
6727 (implicit EFLAGS)], IIC_SSE_POPCNT_RM>, XS;
6732 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
6733 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
6734 Intrinsic IntId128> {
6735 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6737 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6738 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
6739 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6741 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6744 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
6747 let Predicates = [HasAVX] in
6748 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
6749 int_x86_sse41_phminposuw>, VEX;
6750 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
6751 int_x86_sse41_phminposuw>;
6753 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6754 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
6755 Intrinsic IntId128, bit Is2Addr = 1,
6756 OpndItins itins = DEFAULT_ITINS> {
6757 let isCommutable = 1 in
6758 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6759 (ins VR128:$src1, VR128:$src2),
6761 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6762 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6763 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))],
6765 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6766 (ins VR128:$src1, i128mem:$src2),
6768 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6769 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6771 (IntId128 VR128:$src1,
6772 (bitconvert (memopv2i64 addr:$src2))))],
6776 /// SS41I_binop_rm_int_y - Simple SSE 4.1 binary operator
6777 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6778 Intrinsic IntId256> {
6779 let isCommutable = 1 in
6780 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6781 (ins VR256:$src1, VR256:$src2),
6782 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6783 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
6784 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6785 (ins VR256:$src1, i256mem:$src2),
6786 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6788 (IntId256 VR256:$src1,
6789 (bitconvert (loadv4i64 addr:$src2))))]>, OpSize;
6793 /// SS48I_binop_rm - Simple SSE41 binary operator.
6794 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6795 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6796 X86MemOperand x86memop, bit Is2Addr = 1,
6797 OpndItins itins = DEFAULT_ITINS> {
6798 let isCommutable = 1 in
6799 def rr : SS48I<opc, MRMSrcReg, (outs RC:$dst),
6800 (ins RC:$src1, RC:$src2),
6802 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6803 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6804 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>, OpSize;
6805 def rm : SS48I<opc, MRMSrcMem, (outs RC:$dst),
6806 (ins RC:$src1, x86memop:$src2),
6808 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6809 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6811 (OpVT (OpNode RC:$src1,
6812 (bitconvert (memop_frag addr:$src2)))))]>, OpSize;
6815 let Predicates = [HasAVX] in {
6816 let isCommutable = 0 in
6817 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
6819 defm VPMINSB : SS48I_binop_rm<0x38, "vpminsb", X86smin, v16i8, VR128,
6820 loadv2i64, i128mem, 0>, VEX_4V;
6821 defm VPMINSD : SS48I_binop_rm<0x39, "vpminsd", X86smin, v4i32, VR128,
6822 loadv2i64, i128mem, 0>, VEX_4V;
6823 defm VPMINUD : SS48I_binop_rm<0x3B, "vpminud", X86umin, v4i32, VR128,
6824 loadv2i64, i128mem, 0>, VEX_4V;
6825 defm VPMINUW : SS48I_binop_rm<0x3A, "vpminuw", X86umin, v8i16, VR128,
6826 loadv2i64, i128mem, 0>, VEX_4V;
6827 defm VPMAXSB : SS48I_binop_rm<0x3C, "vpmaxsb", X86smax, v16i8, VR128,
6828 loadv2i64, i128mem, 0>, VEX_4V;
6829 defm VPMAXSD : SS48I_binop_rm<0x3D, "vpmaxsd", X86smax, v4i32, VR128,
6830 loadv2i64, i128mem, 0>, VEX_4V;
6831 defm VPMAXUD : SS48I_binop_rm<0x3F, "vpmaxud", X86umax, v4i32, VR128,
6832 loadv2i64, i128mem, 0>, VEX_4V;
6833 defm VPMAXUW : SS48I_binop_rm<0x3E, "vpmaxuw", X86umax, v8i16, VR128,
6834 loadv2i64, i128mem, 0>, VEX_4V;
6835 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
6839 let Predicates = [HasAVX2] in {
6840 let isCommutable = 0 in
6841 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
6842 int_x86_avx2_packusdw>, VEX_4V, VEX_L;
6843 defm VPMINSBY : SS48I_binop_rm<0x38, "vpminsb", X86smin, v32i8, VR256,
6844 loadv4i64, i256mem, 0>, VEX_4V, VEX_L;
6845 defm VPMINSDY : SS48I_binop_rm<0x39, "vpminsd", X86smin, v8i32, VR256,
6846 loadv4i64, i256mem, 0>, VEX_4V, VEX_L;
6847 defm VPMINUDY : SS48I_binop_rm<0x3B, "vpminud", X86umin, v8i32, VR256,
6848 loadv4i64, i256mem, 0>, VEX_4V, VEX_L;
6849 defm VPMINUWY : SS48I_binop_rm<0x3A, "vpminuw", X86umin, v16i16, VR256,
6850 loadv4i64, i256mem, 0>, VEX_4V, VEX_L;
6851 defm VPMAXSBY : SS48I_binop_rm<0x3C, "vpmaxsb", X86smax, v32i8, VR256,
6852 loadv4i64, i256mem, 0>, VEX_4V, VEX_L;
6853 defm VPMAXSDY : SS48I_binop_rm<0x3D, "vpmaxsd", X86smax, v8i32, VR256,
6854 loadv4i64, i256mem, 0>, VEX_4V, VEX_L;
6855 defm VPMAXUDY : SS48I_binop_rm<0x3F, "vpmaxud", X86umax, v8i32, VR256,
6856 loadv4i64, i256mem, 0>, VEX_4V, VEX_L;
6857 defm VPMAXUWY : SS48I_binop_rm<0x3E, "vpmaxuw", X86umax, v16i16, VR256,
6858 loadv4i64, i256mem, 0>, VEX_4V, VEX_L;
6859 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
6860 int_x86_avx2_pmul_dq>, VEX_4V, VEX_L;
6863 let Constraints = "$src1 = $dst" in {
6864 let isCommutable = 0 in
6865 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
6866 defm PMINSB : SS48I_binop_rm<0x38, "pminsb", X86smin, v16i8, VR128,
6867 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6868 defm PMINSD : SS48I_binop_rm<0x39, "pminsd", X86smin, v4i32, VR128,
6869 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6870 defm PMINUD : SS48I_binop_rm<0x3B, "pminud", X86umin, v4i32, VR128,
6871 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6872 defm PMINUW : SS48I_binop_rm<0x3A, "pminuw", X86umin, v8i16, VR128,
6873 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6874 defm PMAXSB : SS48I_binop_rm<0x3C, "pmaxsb", X86smax, v16i8, VR128,
6875 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6876 defm PMAXSD : SS48I_binop_rm<0x3D, "pmaxsd", X86smax, v4i32, VR128,
6877 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6878 defm PMAXUD : SS48I_binop_rm<0x3F, "pmaxud", X86umax, v4i32, VR128,
6879 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6880 defm PMAXUW : SS48I_binop_rm<0x3E, "pmaxuw", X86umax, v8i16, VR128,
6881 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6882 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq,
6883 1, SSE_INTMUL_ITINS_P>;
6886 let Predicates = [HasAVX] in {
6887 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, VR128,
6888 memopv2i64, i128mem, 0>, VEX_4V;
6889 defm VPCMPEQQ : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v2i64, VR128,
6890 memopv2i64, i128mem, 0>, VEX_4V;
6892 let Predicates = [HasAVX2] in {
6893 defm VPMULLDY : SS48I_binop_rm<0x40, "vpmulld", mul, v8i32, VR256,
6894 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6895 defm VPCMPEQQY : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v4i64, VR256,
6896 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6899 let Constraints = "$src1 = $dst" in {
6900 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32, VR128,
6901 memopv2i64, i128mem, 1, SSE_PMULLD_ITINS>;
6902 defm PCMPEQQ : SS48I_binop_rm<0x29, "pcmpeqq", X86pcmpeq, v2i64, VR128,
6903 memopv2i64, i128mem, 1, SSE_INTALUQ_ITINS_P>;
6906 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
6907 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
6908 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6909 X86MemOperand x86memop, bit Is2Addr = 1,
6910 OpndItins itins = DEFAULT_ITINS> {
6911 let isCommutable = 1 in
6912 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
6913 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6915 !strconcat(OpcodeStr,
6916 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6917 !strconcat(OpcodeStr,
6918 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6919 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))], itins.rr>,
6921 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
6922 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6924 !strconcat(OpcodeStr,
6925 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6926 !strconcat(OpcodeStr,
6927 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6930 (bitconvert (memop_frag addr:$src2)), imm:$src3))], itins.rm>,
6934 let Predicates = [HasAVX] in {
6935 let isCommutable = 0 in {
6936 let ExeDomain = SSEPackedSingle in {
6937 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
6938 VR128, loadv4f32, f128mem, 0>, VEX_4V;
6939 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
6940 int_x86_avx_blend_ps_256, VR256, loadv8f32,
6941 f256mem, 0>, VEX_4V, VEX_L;
6943 let ExeDomain = SSEPackedDouble in {
6944 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
6945 VR128, loadv2f64, f128mem, 0>, VEX_4V;
6946 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
6947 int_x86_avx_blend_pd_256,VR256, loadv4f64,
6948 f256mem, 0>, VEX_4V, VEX_L;
6950 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
6951 VR128, loadv2i64, i128mem, 0>, VEX_4V;
6952 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
6953 VR128, loadv2i64, i128mem, 0>, VEX_4V;
6955 let ExeDomain = SSEPackedSingle in
6956 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
6957 VR128, loadv4f32, f128mem, 0>, VEX_4V;
6958 let ExeDomain = SSEPackedDouble in
6959 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
6960 VR128, loadv2f64, f128mem, 0>, VEX_4V;
6961 let ExeDomain = SSEPackedSingle in
6962 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
6963 VR256, loadv8f32, i256mem, 0>, VEX_4V, VEX_L;
6966 let Predicates = [HasAVX2] in {
6967 let isCommutable = 0 in {
6968 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
6969 VR256, loadv4i64, i256mem, 0>, VEX_4V, VEX_L;
6970 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
6971 VR256, loadv4i64, i256mem, 0>, VEX_4V, VEX_L;
6975 let Constraints = "$src1 = $dst" in {
6976 let isCommutable = 0 in {
6977 let ExeDomain = SSEPackedSingle in
6978 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
6979 VR128, memopv4f32, f128mem,
6980 1, SSE_INTALU_ITINS_P>;
6981 let ExeDomain = SSEPackedDouble in
6982 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
6983 VR128, memopv2f64, f128mem,
6984 1, SSE_INTALU_ITINS_P>;
6985 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
6986 VR128, memopv2i64, i128mem,
6987 1, SSE_INTALU_ITINS_P>;
6988 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
6989 VR128, memopv2i64, i128mem,
6990 1, SSE_INTMUL_ITINS_P>;
6992 let ExeDomain = SSEPackedSingle in
6993 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
6994 VR128, memopv4f32, f128mem, 1,
6996 let ExeDomain = SSEPackedDouble in
6997 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
6998 VR128, memopv2f64, f128mem, 1,
7002 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
7003 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
7004 RegisterClass RC, X86MemOperand x86memop,
7005 PatFrag mem_frag, Intrinsic IntId> {
7006 def rr : Ii8<opc, MRMSrcReg, (outs RC:$dst),
7007 (ins RC:$src1, RC:$src2, RC:$src3),
7008 !strconcat(OpcodeStr,
7009 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7010 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
7011 NoItinerary, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
7013 def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
7014 (ins RC:$src1, x86memop:$src2, RC:$src3),
7015 !strconcat(OpcodeStr,
7016 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7018 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
7020 NoItinerary, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
7023 let Predicates = [HasAVX] in {
7024 let ExeDomain = SSEPackedDouble in {
7025 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, f128mem,
7026 loadv2f64, int_x86_sse41_blendvpd>;
7027 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, f256mem,
7028 loadv4f64, int_x86_avx_blendv_pd_256>, VEX_L;
7029 } // ExeDomain = SSEPackedDouble
7030 let ExeDomain = SSEPackedSingle in {
7031 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, f128mem,
7032 loadv4f32, int_x86_sse41_blendvps>;
7033 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, f256mem,
7034 loadv8f32, int_x86_avx_blendv_ps_256>, VEX_L;
7035 } // ExeDomain = SSEPackedSingle
7036 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
7037 loadv2i64, int_x86_sse41_pblendvb>;
7040 let Predicates = [HasAVX2] in {
7041 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
7042 loadv4i64, int_x86_avx2_pblendvb>, VEX_L;
7045 let Predicates = [HasAVX] in {
7046 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
7047 (v16i8 VR128:$src2))),
7048 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
7049 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
7050 (v4i32 VR128:$src2))),
7051 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
7052 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
7053 (v4f32 VR128:$src2))),
7054 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
7055 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
7056 (v2i64 VR128:$src2))),
7057 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
7058 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
7059 (v2f64 VR128:$src2))),
7060 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
7061 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
7062 (v8i32 VR256:$src2))),
7063 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
7064 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
7065 (v8f32 VR256:$src2))),
7066 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
7067 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
7068 (v4i64 VR256:$src2))),
7069 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
7070 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
7071 (v4f64 VR256:$src2))),
7072 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
7074 def : Pat<(v8f32 (X86Blendi (v8f32 VR256:$src1), (v8f32 VR256:$src2),
7076 (VBLENDPSYrri VR256:$src1, VR256:$src2, imm:$mask)>;
7077 def : Pat<(v4f64 (X86Blendi (v4f64 VR256:$src1), (v4f64 VR256:$src2),
7079 (VBLENDPDYrri VR256:$src1, VR256:$src2, imm:$mask)>;
7081 def : Pat<(v8i16 (X86Blendi (v8i16 VR128:$src1), (v8i16 VR128:$src2),
7083 (VPBLENDWrri VR128:$src1, VR128:$src2, imm:$mask)>;
7084 def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$src1), (v4f32 VR128:$src2),
7086 (VBLENDPSrri VR128:$src1, VR128:$src2, imm:$mask)>;
7087 def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$src1), (v2f64 VR128:$src2),
7089 (VBLENDPDrri VR128:$src1, VR128:$src2, imm:$mask)>;
7092 let Predicates = [HasAVX2] in {
7093 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
7094 (v32i8 VR256:$src2))),
7095 (VPBLENDVBYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
7096 def : Pat<(v16i16 (X86Blendi (v16i16 VR256:$src1), (v16i16 VR256:$src2),
7098 (VPBLENDWYrri VR256:$src1, VR256:$src2, imm:$mask)>;
7101 /// SS41I_ternary_int - SSE 4.1 ternary operator
7102 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
7103 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7104 X86MemOperand x86memop, Intrinsic IntId,
7105 OpndItins itins = DEFAULT_ITINS> {
7106 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
7107 (ins VR128:$src1, VR128:$src2),
7108 !strconcat(OpcodeStr,
7109 "\t{$src2, $dst|$dst, $src2}"),
7110 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))],
7113 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
7114 (ins VR128:$src1, x86memop:$src2),
7115 !strconcat(OpcodeStr,
7116 "\t{$src2, $dst|$dst, $src2}"),
7119 (bitconvert (mem_frag addr:$src2)), XMM0))],
7124 let ExeDomain = SSEPackedDouble in
7125 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64, f128mem,
7126 int_x86_sse41_blendvpd>;
7127 let ExeDomain = SSEPackedSingle in
7128 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32, f128mem,
7129 int_x86_sse41_blendvps>;
7130 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64, i128mem,
7131 int_x86_sse41_pblendvb>;
7133 // Aliases with the implicit xmm0 argument
7134 def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7135 (BLENDVPDrr0 VR128:$dst, VR128:$src2)>;
7136 def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7137 (BLENDVPDrm0 VR128:$dst, f128mem:$src2)>;
7138 def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7139 (BLENDVPSrr0 VR128:$dst, VR128:$src2)>;
7140 def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7141 (BLENDVPSrm0 VR128:$dst, f128mem:$src2)>;
7142 def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7143 (PBLENDVBrr0 VR128:$dst, VR128:$src2)>;
7144 def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7145 (PBLENDVBrm0 VR128:$dst, i128mem:$src2)>;
7147 let Predicates = [UseSSE41] in {
7148 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
7149 (v16i8 VR128:$src2))),
7150 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
7151 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
7152 (v4i32 VR128:$src2))),
7153 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
7154 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
7155 (v4f32 VR128:$src2))),
7156 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
7157 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
7158 (v2i64 VR128:$src2))),
7159 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
7160 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
7161 (v2f64 VR128:$src2))),
7162 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
7164 def : Pat<(v8i16 (X86Blendi (v8i16 VR128:$src1), (v8i16 VR128:$src2),
7166 (PBLENDWrri VR128:$src1, VR128:$src2, imm:$mask)>;
7167 def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$src1), (v4f32 VR128:$src2),
7169 (BLENDPSrri VR128:$src1, VR128:$src2, imm:$mask)>;
7170 def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$src1), (v2f64 VR128:$src2),
7172 (BLENDPDrri VR128:$src1, VR128:$src2, imm:$mask)>;
7176 let Predicates = [HasAVX] in
7177 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
7178 "vmovntdqa\t{$src, $dst|$dst, $src}",
7179 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
7181 let Predicates = [HasAVX2] in
7182 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
7183 "vmovntdqa\t{$src, $dst|$dst, $src}",
7184 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
7186 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
7187 "movntdqa\t{$src, $dst|$dst, $src}",
7188 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
7191 //===----------------------------------------------------------------------===//
7192 // SSE4.2 - Compare Instructions
7193 //===----------------------------------------------------------------------===//
7195 /// SS42I_binop_rm - Simple SSE 4.2 binary operator
7196 multiclass SS42I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
7197 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
7198 X86MemOperand x86memop, bit Is2Addr = 1> {
7199 def rr : SS428I<opc, MRMSrcReg, (outs RC:$dst),
7200 (ins RC:$src1, RC:$src2),
7202 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7203 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7204 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>,
7206 def rm : SS428I<opc, MRMSrcMem, (outs RC:$dst),
7207 (ins RC:$src1, x86memop:$src2),
7209 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7210 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7212 (OpVT (OpNode RC:$src1, (memop_frag addr:$src2))))]>, OpSize;
7215 let Predicates = [HasAVX] in
7216 defm VPCMPGTQ : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v2i64, VR128,
7217 loadv2i64, i128mem, 0>, VEX_4V;
7219 let Predicates = [HasAVX2] in
7220 defm VPCMPGTQY : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v4i64, VR256,
7221 loadv4i64, i256mem, 0>, VEX_4V, VEX_L;
7223 let Constraints = "$src1 = $dst" in
7224 defm PCMPGTQ : SS42I_binop_rm<0x37, "pcmpgtq", X86pcmpgt, v2i64, VR128,
7225 memopv2i64, i128mem>;
7227 //===----------------------------------------------------------------------===//
7228 // SSE4.2 - String/text Processing Instructions
7229 //===----------------------------------------------------------------------===//
7231 // Packed Compare Implicit Length Strings, Return Mask
7232 multiclass pseudo_pcmpistrm<string asm> {
7233 def REG : PseudoI<(outs VR128:$dst),
7234 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7235 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
7237 def MEM : PseudoI<(outs VR128:$dst),
7238 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7239 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1,
7240 (bc_v16i8 (memopv2i64 addr:$src2)), imm:$src3))]>;
7243 let Defs = [EFLAGS], usesCustomInserter = 1 in {
7244 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
7245 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[UseSSE42]>;
7248 multiclass pcmpistrm_SS42AI<string asm> {
7249 def rr : SS42AI<0x62, MRMSrcReg, (outs),
7250 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7251 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
7254 def rm :SS42AI<0x62, MRMSrcMem, (outs),
7255 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7256 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
7260 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
7261 let Predicates = [HasAVX] in
7262 defm VPCMPISTRM128 : pcmpistrm_SS42AI<"vpcmpistrm">, VEX;
7263 defm PCMPISTRM128 : pcmpistrm_SS42AI<"pcmpistrm"> ;
7266 // Packed Compare Explicit Length Strings, Return Mask
7267 multiclass pseudo_pcmpestrm<string asm> {
7268 def REG : PseudoI<(outs VR128:$dst),
7269 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7270 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
7271 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
7272 def MEM : PseudoI<(outs VR128:$dst),
7273 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7274 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128 VR128:$src1, EAX,
7275 (bc_v16i8 (memopv2i64 addr:$src3)), EDX, imm:$src5))]>;
7278 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
7279 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
7280 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[UseSSE42]>;
7283 multiclass SS42AI_pcmpestrm<string asm> {
7284 def rr : SS42AI<0x60, MRMSrcReg, (outs),
7285 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7286 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7289 def rm : SS42AI<0x60, MRMSrcMem, (outs),
7290 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7291 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7295 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
7296 let Predicates = [HasAVX] in
7297 defm VPCMPESTRM128 : SS42AI_pcmpestrm<"vpcmpestrm">, VEX;
7298 defm PCMPESTRM128 : SS42AI_pcmpestrm<"pcmpestrm">;
7301 // Packed Compare Implicit Length Strings, Return Index
7302 multiclass pseudo_pcmpistri<string asm> {
7303 def REG : PseudoI<(outs GR32:$dst),
7304 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7305 [(set GR32:$dst, EFLAGS,
7306 (X86pcmpistri VR128:$src1, VR128:$src2, imm:$src3))]>;
7307 def MEM : PseudoI<(outs GR32:$dst),
7308 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7309 [(set GR32:$dst, EFLAGS, (X86pcmpistri VR128:$src1,
7310 (bc_v16i8 (memopv2i64 addr:$src2)), imm:$src3))]>;
7313 let Defs = [EFLAGS], usesCustomInserter = 1 in {
7314 defm VPCMPISTRI : pseudo_pcmpistri<"#VPCMPISTRI">, Requires<[HasAVX]>;
7315 defm PCMPISTRI : pseudo_pcmpistri<"#PCMPISTRI">, Requires<[UseSSE42]>;
7318 multiclass SS42AI_pcmpistri<string asm> {
7319 def rr : SS42AI<0x63, MRMSrcReg, (outs),
7320 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7321 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
7324 def rm : SS42AI<0x63, MRMSrcMem, (outs),
7325 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7326 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
7330 let Defs = [ECX, EFLAGS], neverHasSideEffects = 1 in {
7331 let Predicates = [HasAVX] in
7332 defm VPCMPISTRI : SS42AI_pcmpistri<"vpcmpistri">, VEX;
7333 defm PCMPISTRI : SS42AI_pcmpistri<"pcmpistri">;
7336 // Packed Compare Explicit Length Strings, Return Index
7337 multiclass pseudo_pcmpestri<string asm> {
7338 def REG : PseudoI<(outs GR32:$dst),
7339 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7340 [(set GR32:$dst, EFLAGS,
7341 (X86pcmpestri VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
7342 def MEM : PseudoI<(outs GR32:$dst),
7343 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7344 [(set GR32:$dst, EFLAGS,
7345 (X86pcmpestri VR128:$src1, EAX, (bc_v16i8 (memopv2i64 addr:$src3)), EDX,
7349 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
7350 defm VPCMPESTRI : pseudo_pcmpestri<"#VPCMPESTRI">, Requires<[HasAVX]>;
7351 defm PCMPESTRI : pseudo_pcmpestri<"#PCMPESTRI">, Requires<[UseSSE42]>;
7354 multiclass SS42AI_pcmpestri<string asm> {
7355 def rr : SS42AI<0x61, MRMSrcReg, (outs),
7356 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7357 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7360 def rm : SS42AI<0x61, MRMSrcMem, (outs),
7361 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7362 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7366 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
7367 let Predicates = [HasAVX] in
7368 defm VPCMPESTRI : SS42AI_pcmpestri<"vpcmpestri">, VEX;
7369 defm PCMPESTRI : SS42AI_pcmpestri<"pcmpestri">;
7372 //===----------------------------------------------------------------------===//
7373 // SSE4.2 - CRC Instructions
7374 //===----------------------------------------------------------------------===//
7376 // No CRC instructions have AVX equivalents
7378 // crc intrinsic instruction
7379 // This set of instructions are only rm, the only difference is the size
7381 class SS42I_crc32r<bits<8> opc, string asm, RegisterClass RCOut,
7382 RegisterClass RCIn, SDPatternOperator Int> :
7383 SS42FI<opc, MRMSrcReg, (outs RCOut:$dst), (ins RCOut:$src1, RCIn:$src2),
7384 !strconcat(asm, "\t{$src2, $src1|$src1, $src2}"),
7385 [(set RCOut:$dst, (Int RCOut:$src1, RCIn:$src2))], IIC_CRC32_REG>;
7387 class SS42I_crc32m<bits<8> opc, string asm, RegisterClass RCOut,
7388 X86MemOperand x86memop, SDPatternOperator Int> :
7389 SS42FI<opc, MRMSrcMem, (outs RCOut:$dst), (ins RCOut:$src1, x86memop:$src2),
7390 !strconcat(asm, "\t{$src2, $src1|$src1, $src2}"),
7391 [(set RCOut:$dst, (Int RCOut:$src1, (load addr:$src2)))],
7394 let Constraints = "$src1 = $dst" in {
7395 def CRC32r32m8 : SS42I_crc32m<0xF0, "crc32{b}", GR32, i8mem,
7396 int_x86_sse42_crc32_32_8>;
7397 def CRC32r32r8 : SS42I_crc32r<0xF0, "crc32{b}", GR32, GR8,
7398 int_x86_sse42_crc32_32_8>;
7399 def CRC32r32m16 : SS42I_crc32m<0xF1, "crc32{w}", GR32, i16mem,
7400 int_x86_sse42_crc32_32_16>, OpSize;
7401 def CRC32r32r16 : SS42I_crc32r<0xF1, "crc32{w}", GR32, GR16,
7402 int_x86_sse42_crc32_32_16>, OpSize;
7403 def CRC32r32m32 : SS42I_crc32m<0xF1, "crc32{l}", GR32, i32mem,
7404 int_x86_sse42_crc32_32_32>;
7405 def CRC32r32r32 : SS42I_crc32r<0xF1, "crc32{l}", GR32, GR32,
7406 int_x86_sse42_crc32_32_32>;
7407 def CRC32r64m64 : SS42I_crc32m<0xF1, "crc32{q}", GR64, i64mem,
7408 int_x86_sse42_crc32_64_64>, REX_W;
7409 def CRC32r64r64 : SS42I_crc32r<0xF1, "crc32{q}", GR64, GR64,
7410 int_x86_sse42_crc32_64_64>, REX_W;
7411 let hasSideEffects = 0 in {
7413 def CRC32r64m8 : SS42I_crc32m<0xF0, "crc32{b}", GR64, i8mem,
7415 def CRC32r64r8 : SS42I_crc32r<0xF0, "crc32{b}", GR64, GR8,
7420 //===----------------------------------------------------------------------===//
7421 // SHA-NI Instructions
7422 //===----------------------------------------------------------------------===//
7424 multiclass SHAI_binop<bits<8> Opc, string OpcodeStr, Intrinsic IntId,
7426 def rr : I<Opc, MRMSrcReg, (outs VR128:$dst),
7427 (ins VR128:$src1, VR128:$src2),
7428 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7430 (set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0)),
7431 (set VR128:$dst, (IntId VR128:$src1, VR128:$src2)))]>, T8;
7433 def rm : I<Opc, MRMSrcMem, (outs VR128:$dst),
7434 (ins VR128:$src1, i128mem:$src2),
7435 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7437 (set VR128:$dst, (IntId VR128:$src1,
7438 (bc_v4i32 (memopv2i64 addr:$src2)), XMM0)),
7439 (set VR128:$dst, (IntId VR128:$src1,
7440 (bc_v4i32 (memopv2i64 addr:$src2)))))]>, T8;
7443 let Constraints = "$src1 = $dst", Predicates = [HasSHA] in {
7444 def SHA1RNDS4rri : Ii8<0xCC, MRMSrcReg, (outs VR128:$dst),
7445 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7446 "sha1rnds4\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7448 (int_x86_sha1rnds4 VR128:$src1, VR128:$src2,
7449 (i8 imm:$src3)))]>, TA;
7450 def SHA1RNDS4rmi : Ii8<0xCC, MRMSrcMem, (outs VR128:$dst),
7451 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7452 "sha1rnds4\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7454 (int_x86_sha1rnds4 VR128:$src1,
7455 (bc_v4i32 (memopv2i64 addr:$src2)),
7456 (i8 imm:$src3)))]>, TA;
7458 defm SHA1NEXTE : SHAI_binop<0xC8, "sha1nexte", int_x86_sha1nexte>;
7459 defm SHA1MSG1 : SHAI_binop<0xC9, "sha1msg1", int_x86_sha1msg1>;
7460 defm SHA1MSG2 : SHAI_binop<0xCA, "sha1msg2", int_x86_sha1msg2>;
7463 defm SHA256RNDS2 : SHAI_binop<0xCB, "sha256rnds2", int_x86_sha256rnds2, 1>;
7465 defm SHA256MSG1 : SHAI_binop<0xCC, "sha256msg1", int_x86_sha256msg1>;
7466 defm SHA256MSG2 : SHAI_binop<0xCD, "sha256msg2", int_x86_sha256msg2>;
7469 // Aliases with explicit %xmm0
7470 def : InstAlias<"sha256rnds2\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7471 (SHA256RNDS2rr VR128:$dst, VR128:$src2)>;
7472 def : InstAlias<"sha256rnds2\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7473 (SHA256RNDS2rm VR128:$dst, i128mem:$src2)>;
7475 //===----------------------------------------------------------------------===//
7476 // AES-NI Instructions
7477 //===----------------------------------------------------------------------===//
7479 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
7480 Intrinsic IntId128, bit Is2Addr = 1> {
7481 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
7482 (ins VR128:$src1, VR128:$src2),
7484 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7485 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7486 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
7488 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
7489 (ins VR128:$src1, i128mem:$src2),
7491 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7492 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7494 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
7497 // Perform One Round of an AES Encryption/Decryption Flow
7498 let Predicates = [HasAVX, HasAES] in {
7499 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
7500 int_x86_aesni_aesenc, 0>, VEX_4V;
7501 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
7502 int_x86_aesni_aesenclast, 0>, VEX_4V;
7503 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
7504 int_x86_aesni_aesdec, 0>, VEX_4V;
7505 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
7506 int_x86_aesni_aesdeclast, 0>, VEX_4V;
7509 let Constraints = "$src1 = $dst" in {
7510 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
7511 int_x86_aesni_aesenc>;
7512 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
7513 int_x86_aesni_aesenclast>;
7514 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
7515 int_x86_aesni_aesdec>;
7516 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
7517 int_x86_aesni_aesdeclast>;
7520 // Perform the AES InvMixColumn Transformation
7521 let Predicates = [HasAVX, HasAES] in {
7522 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7524 "vaesimc\t{$src1, $dst|$dst, $src1}",
7526 (int_x86_aesni_aesimc VR128:$src1))]>,
7528 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7529 (ins i128mem:$src1),
7530 "vaesimc\t{$src1, $dst|$dst, $src1}",
7531 [(set VR128:$dst, (int_x86_aesni_aesimc (loadv2i64 addr:$src1)))]>,
7534 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7536 "aesimc\t{$src1, $dst|$dst, $src1}",
7538 (int_x86_aesni_aesimc VR128:$src1))]>,
7540 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7541 (ins i128mem:$src1),
7542 "aesimc\t{$src1, $dst|$dst, $src1}",
7543 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7546 // AES Round Key Generation Assist
7547 let Predicates = [HasAVX, HasAES] in {
7548 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7549 (ins VR128:$src1, i8imm:$src2),
7550 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7552 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7554 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7555 (ins i128mem:$src1, i8imm:$src2),
7556 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7558 (int_x86_aesni_aeskeygenassist (loadv2i64 addr:$src1), imm:$src2))]>,
7561 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7562 (ins VR128:$src1, i8imm:$src2),
7563 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7565 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7567 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7568 (ins i128mem:$src1, i8imm:$src2),
7569 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7571 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7574 //===----------------------------------------------------------------------===//
7575 // PCLMUL Instructions
7576 //===----------------------------------------------------------------------===//
7578 // AVX carry-less Multiplication instructions
7579 def VPCLMULQDQrr : AVXPCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7580 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7581 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7583 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
7585 def VPCLMULQDQrm : AVXPCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7586 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7587 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7588 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7589 (loadv2i64 addr:$src2), imm:$src3))]>;
7591 // Carry-less Multiplication instructions
7592 let Constraints = "$src1 = $dst" in {
7593 def PCLMULQDQrr : PCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7594 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7595 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7597 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))],
7598 IIC_SSE_PCLMULQDQ_RR>;
7600 def PCLMULQDQrm : PCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7601 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7602 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7603 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7604 (memopv2i64 addr:$src2), imm:$src3))],
7605 IIC_SSE_PCLMULQDQ_RM>;
7606 } // Constraints = "$src1 = $dst"
7609 multiclass pclmul_alias<string asm, int immop> {
7610 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7611 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
7613 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7614 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
7616 def : InstAlias<!strconcat("vpclmul", asm,
7617 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7618 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
7620 def : InstAlias<!strconcat("vpclmul", asm,
7621 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7622 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
7624 defm : pclmul_alias<"hqhq", 0x11>;
7625 defm : pclmul_alias<"hqlq", 0x01>;
7626 defm : pclmul_alias<"lqhq", 0x10>;
7627 defm : pclmul_alias<"lqlq", 0x00>;
7629 //===----------------------------------------------------------------------===//
7630 // SSE4A Instructions
7631 //===----------------------------------------------------------------------===//
7633 let Predicates = [HasSSE4A] in {
7635 let Constraints = "$src = $dst" in {
7636 def EXTRQI : Ii8<0x78, MRM0r, (outs VR128:$dst),
7637 (ins VR128:$src, i8imm:$len, i8imm:$idx),
7638 "extrq\t{$idx, $len, $src|$src, $len, $idx}",
7639 [(set VR128:$dst, (int_x86_sse4a_extrqi VR128:$src, imm:$len,
7640 imm:$idx))]>, TB, OpSize;
7641 def EXTRQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7642 (ins VR128:$src, VR128:$mask),
7643 "extrq\t{$mask, $src|$src, $mask}",
7644 [(set VR128:$dst, (int_x86_sse4a_extrq VR128:$src,
7645 VR128:$mask))]>, TB, OpSize;
7647 def INSERTQI : Ii8<0x78, MRMSrcReg, (outs VR128:$dst),
7648 (ins VR128:$src, VR128:$src2, i8imm:$len, i8imm:$idx),
7649 "insertq\t{$idx, $len, $src2, $src|$src, $src2, $len, $idx}",
7650 [(set VR128:$dst, (int_x86_sse4a_insertqi VR128:$src,
7651 VR128:$src2, imm:$len, imm:$idx))]>, XD;
7652 def INSERTQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7653 (ins VR128:$src, VR128:$mask),
7654 "insertq\t{$mask, $src|$src, $mask}",
7655 [(set VR128:$dst, (int_x86_sse4a_insertq VR128:$src,
7656 VR128:$mask))]>, XD;
7659 def MOVNTSS : I<0x2B, MRMDestMem, (outs), (ins f32mem:$dst, VR128:$src),
7660 "movntss\t{$src, $dst|$dst, $src}",
7661 [(int_x86_sse4a_movnt_ss addr:$dst, VR128:$src)]>, XS;
7663 def MOVNTSD : I<0x2B, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
7664 "movntsd\t{$src, $dst|$dst, $src}",
7665 [(int_x86_sse4a_movnt_sd addr:$dst, VR128:$src)]>, XD;
7668 //===----------------------------------------------------------------------===//
7670 //===----------------------------------------------------------------------===//
7672 //===----------------------------------------------------------------------===//
7673 // VBROADCAST - Load from memory and broadcast to all elements of the
7674 // destination operand
7676 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
7677 X86MemOperand x86memop, Intrinsic Int> :
7678 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7679 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7680 [(set RC:$dst, (Int addr:$src))]>, VEX;
7682 // AVX2 adds register forms
7683 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
7685 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7686 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7687 [(set RC:$dst, (Int VR128:$src))]>, VEX;
7689 let ExeDomain = SSEPackedSingle in {
7690 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
7691 int_x86_avx_vbroadcast_ss>;
7692 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
7693 int_x86_avx_vbroadcast_ss_256>, VEX_L;
7695 let ExeDomain = SSEPackedDouble in
7696 def VBROADCASTSDYrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
7697 int_x86_avx_vbroadcast_sd_256>, VEX_L;
7698 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
7699 int_x86_avx_vbroadcastf128_pd_256>, VEX_L;
7701 let ExeDomain = SSEPackedSingle in {
7702 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
7703 int_x86_avx2_vbroadcast_ss_ps>;
7704 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
7705 int_x86_avx2_vbroadcast_ss_ps_256>, VEX_L;
7707 let ExeDomain = SSEPackedDouble in
7708 def VBROADCASTSDYrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
7709 int_x86_avx2_vbroadcast_sd_pd_256>, VEX_L;
7711 let Predicates = [HasAVX2] in
7712 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
7713 int_x86_avx2_vbroadcasti128>, VEX_L;
7715 let Predicates = [HasAVX] in
7716 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
7717 (VBROADCASTF128 addr:$src)>;
7720 //===----------------------------------------------------------------------===//
7721 // VINSERTF128 - Insert packed floating-point values
7723 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7724 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
7725 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7726 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7729 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
7730 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
7731 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7735 let Predicates = [HasAVX] in {
7736 def : Pat<(vinsert128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7738 (VINSERTF128rr VR256:$src1, VR128:$src2,
7739 (INSERT_get_vinsert128_imm VR256:$ins))>;
7740 def : Pat<(vinsert128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7742 (VINSERTF128rr VR256:$src1, VR128:$src2,
7743 (INSERT_get_vinsert128_imm VR256:$ins))>;
7745 def : Pat<(vinsert128_insert:$ins (v8f32 VR256:$src1), (loadv4f32 addr:$src2),
7747 (VINSERTF128rm VR256:$src1, addr:$src2,
7748 (INSERT_get_vinsert128_imm VR256:$ins))>;
7749 def : Pat<(vinsert128_insert:$ins (v4f64 VR256:$src1), (loadv2f64 addr:$src2),
7751 (VINSERTF128rm VR256:$src1, addr:$src2,
7752 (INSERT_get_vinsert128_imm VR256:$ins))>;
7755 let Predicates = [HasAVX1Only] in {
7756 def : Pat<(vinsert128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7758 (VINSERTF128rr VR256:$src1, VR128:$src2,
7759 (INSERT_get_vinsert128_imm VR256:$ins))>;
7760 def : Pat<(vinsert128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7762 (VINSERTF128rr VR256:$src1, VR128:$src2,
7763 (INSERT_get_vinsert128_imm VR256:$ins))>;
7764 def : Pat<(vinsert128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7766 (VINSERTF128rr VR256:$src1, VR128:$src2,
7767 (INSERT_get_vinsert128_imm VR256:$ins))>;
7768 def : Pat<(vinsert128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7770 (VINSERTF128rr VR256:$src1, VR128:$src2,
7771 (INSERT_get_vinsert128_imm VR256:$ins))>;
7773 def : Pat<(vinsert128_insert:$ins (v4i64 VR256:$src1), (loadv2i64 addr:$src2),
7775 (VINSERTF128rm VR256:$src1, addr:$src2,
7776 (INSERT_get_vinsert128_imm VR256:$ins))>;
7777 def : Pat<(vinsert128_insert:$ins (v8i32 VR256:$src1),
7778 (bc_v4i32 (loadv2i64 addr:$src2)),
7780 (VINSERTF128rm VR256:$src1, addr:$src2,
7781 (INSERT_get_vinsert128_imm VR256:$ins))>;
7782 def : Pat<(vinsert128_insert:$ins (v32i8 VR256:$src1),
7783 (bc_v16i8 (loadv2i64 addr:$src2)),
7785 (VINSERTF128rm VR256:$src1, addr:$src2,
7786 (INSERT_get_vinsert128_imm VR256:$ins))>;
7787 def : Pat<(vinsert128_insert:$ins (v16i16 VR256:$src1),
7788 (bc_v8i16 (loadv2i64 addr:$src2)),
7790 (VINSERTF128rm VR256:$src1, addr:$src2,
7791 (INSERT_get_vinsert128_imm VR256:$ins))>;
7794 //===----------------------------------------------------------------------===//
7795 // VEXTRACTF128 - Extract packed floating-point values
7797 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7798 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
7799 (ins VR256:$src1, i8imm:$src2),
7800 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7803 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
7804 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
7805 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7810 let Predicates = [HasAVX] in {
7811 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7812 (v4f32 (VEXTRACTF128rr
7813 (v8f32 VR256:$src1),
7814 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7815 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7816 (v2f64 (VEXTRACTF128rr
7817 (v4f64 VR256:$src1),
7818 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7820 def : Pat<(store (v4f32 (vextract128_extract:$ext (v8f32 VR256:$src1),
7821 (iPTR imm))), addr:$dst),
7822 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7823 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7824 def : Pat<(store (v2f64 (vextract128_extract:$ext (v4f64 VR256:$src1),
7825 (iPTR imm))), addr:$dst),
7826 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7827 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7830 let Predicates = [HasAVX1Only] in {
7831 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7832 (v2i64 (VEXTRACTF128rr
7833 (v4i64 VR256:$src1),
7834 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7835 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7836 (v4i32 (VEXTRACTF128rr
7837 (v8i32 VR256:$src1),
7838 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7839 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7840 (v8i16 (VEXTRACTF128rr
7841 (v16i16 VR256:$src1),
7842 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7843 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7844 (v16i8 (VEXTRACTF128rr
7845 (v32i8 VR256:$src1),
7846 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7848 def : Pat<(alignedstore (v2i64 (vextract128_extract:$ext (v4i64 VR256:$src1),
7849 (iPTR imm))), addr:$dst),
7850 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7851 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7852 def : Pat<(alignedstore (v4i32 (vextract128_extract:$ext (v8i32 VR256:$src1),
7853 (iPTR imm))), addr:$dst),
7854 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7855 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7856 def : Pat<(alignedstore (v8i16 (vextract128_extract:$ext (v16i16 VR256:$src1),
7857 (iPTR imm))), addr:$dst),
7858 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7859 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7860 def : Pat<(alignedstore (v16i8 (vextract128_extract:$ext (v32i8 VR256:$src1),
7861 (iPTR imm))), addr:$dst),
7862 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7863 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7866 //===----------------------------------------------------------------------===//
7867 // VMASKMOV - Conditional SIMD Packed Loads and Stores
7869 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
7870 Intrinsic IntLd, Intrinsic IntLd256,
7871 Intrinsic IntSt, Intrinsic IntSt256> {
7872 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
7873 (ins VR128:$src1, f128mem:$src2),
7874 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7875 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
7877 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
7878 (ins VR256:$src1, f256mem:$src2),
7879 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7880 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7882 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
7883 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
7884 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7885 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7886 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
7887 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
7888 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7889 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V, VEX_L;
7892 let ExeDomain = SSEPackedSingle in
7893 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
7894 int_x86_avx_maskload_ps,
7895 int_x86_avx_maskload_ps_256,
7896 int_x86_avx_maskstore_ps,
7897 int_x86_avx_maskstore_ps_256>;
7898 let ExeDomain = SSEPackedDouble in
7899 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
7900 int_x86_avx_maskload_pd,
7901 int_x86_avx_maskload_pd_256,
7902 int_x86_avx_maskstore_pd,
7903 int_x86_avx_maskstore_pd_256>;
7905 //===----------------------------------------------------------------------===//
7906 // VPERMIL - Permute Single and Double Floating-Point Values
7908 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
7909 RegisterClass RC, X86MemOperand x86memop_f,
7910 X86MemOperand x86memop_i, PatFrag i_frag,
7911 Intrinsic IntVar, ValueType vt> {
7912 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
7913 (ins RC:$src1, RC:$src2),
7914 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7915 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
7916 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
7917 (ins RC:$src1, x86memop_i:$src2),
7918 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7919 [(set RC:$dst, (IntVar RC:$src1,
7920 (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
7922 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
7923 (ins RC:$src1, i8imm:$src2),
7924 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7925 [(set RC:$dst, (vt (X86VPermilp RC:$src1, (i8 imm:$src2))))]>, VEX;
7926 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
7927 (ins x86memop_f:$src1, i8imm:$src2),
7928 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7930 (vt (X86VPermilp (memop addr:$src1), (i8 imm:$src2))))]>, VEX;
7933 let ExeDomain = SSEPackedSingle in {
7934 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
7935 loadv2i64, int_x86_avx_vpermilvar_ps, v4f32>;
7936 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
7937 loadv4i64, int_x86_avx_vpermilvar_ps_256, v8f32>, VEX_L;
7939 let ExeDomain = SSEPackedDouble in {
7940 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
7941 loadv2i64, int_x86_avx_vpermilvar_pd, v2f64>;
7942 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
7943 loadv4i64, int_x86_avx_vpermilvar_pd_256, v4f64>, VEX_L;
7946 let Predicates = [HasAVX] in {
7947 def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7948 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7949 def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7950 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7951 def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (loadv4i64 addr:$src1)),
7953 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7954 def : Pat<(v4i64 (X86VPermilp (loadv4i64 addr:$src1), (i8 imm:$imm))),
7955 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7957 def : Pat<(v2i64 (X86VPermilp VR128:$src1, (i8 imm:$imm))),
7958 (VPERMILPDri VR128:$src1, imm:$imm)>;
7959 def : Pat<(v2i64 (X86VPermilp (loadv2i64 addr:$src1), (i8 imm:$imm))),
7960 (VPERMILPDmi addr:$src1, imm:$imm)>;
7963 //===----------------------------------------------------------------------===//
7964 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
7966 let ExeDomain = SSEPackedSingle in {
7967 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
7968 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7969 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7970 [(set VR256:$dst, (v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7971 (i8 imm:$src3))))]>, VEX_4V, VEX_L;
7972 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
7973 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7974 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7975 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (loadv8f32 addr:$src2),
7976 (i8 imm:$src3)))]>, VEX_4V, VEX_L;
7979 let Predicates = [HasAVX] in {
7980 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7981 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7982 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
7983 (loadv4f64 addr:$src2), (i8 imm:$imm))),
7984 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7987 let Predicates = [HasAVX1Only] in {
7988 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7989 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7990 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7991 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7992 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7993 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7994 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7995 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7997 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
7998 (bc_v8i32 (loadv4i64 addr:$src2)), (i8 imm:$imm))),
7999 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
8000 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
8001 (loadv4i64 addr:$src2), (i8 imm:$imm))),
8002 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
8003 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
8004 (bc_v32i8 (loadv4i64 addr:$src2)), (i8 imm:$imm))),
8005 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
8006 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
8007 (bc_v16i16 (loadv4i64 addr:$src2)), (i8 imm:$imm))),
8008 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
8011 //===----------------------------------------------------------------------===//
8012 // VZERO - Zero YMM registers
8014 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
8015 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
8016 // Zero All YMM registers
8017 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
8018 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
8020 // Zero Upper bits of YMM registers
8021 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
8022 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
8025 //===----------------------------------------------------------------------===//
8026 // Half precision conversion instructions
8027 //===----------------------------------------------------------------------===//
8028 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
8029 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
8030 "vcvtph2ps\t{$src, $dst|$dst, $src}",
8031 [(set RC:$dst, (Int VR128:$src))]>,
8033 let neverHasSideEffects = 1, mayLoad = 1 in
8034 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
8035 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
8038 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
8039 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
8040 (ins RC:$src1, i32i8imm:$src2),
8041 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
8042 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
8044 let neverHasSideEffects = 1, mayStore = 1 in
8045 def mr : Ii8<0x1D, MRMDestMem, (outs),
8046 (ins x86memop:$dst, RC:$src1, i32i8imm:$src2),
8047 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
8051 let Predicates = [HasF16C] in {
8052 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
8053 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>, VEX_L;
8054 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
8055 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>, VEX_L;
8058 //===----------------------------------------------------------------------===//
8059 // AVX2 Instructions
8060 //===----------------------------------------------------------------------===//
8062 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
8063 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
8064 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
8065 X86MemOperand x86memop> {
8066 let isCommutable = 1 in
8067 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
8068 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
8069 !strconcat(OpcodeStr,
8070 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
8071 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
8073 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
8074 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
8075 !strconcat(OpcodeStr,
8076 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
8079 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
8083 let isCommutable = 0 in {
8084 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
8085 VR128, loadv2i64, i128mem>;
8086 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
8087 VR256, loadv4i64, i256mem>, VEX_L;
8090 def : Pat<(v4i32 (X86Blendi (v4i32 VR128:$src1), (v4i32 VR128:$src2),
8092 (VPBLENDDrri VR128:$src1, VR128:$src2, imm:$mask)>;
8093 def : Pat<(v8i32 (X86Blendi (v8i32 VR256:$src1), (v8i32 VR256:$src2),
8095 (VPBLENDDYrri VR256:$src1, VR256:$src2, imm:$mask)>;
8097 //===----------------------------------------------------------------------===//
8098 // VPBROADCAST - Load from memory and broadcast to all elements of the
8099 // destination operand
8101 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
8102 X86MemOperand x86memop, PatFrag ld_frag,
8103 Intrinsic Int128, Intrinsic Int256> {
8104 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
8105 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
8106 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
8107 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
8108 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
8110 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
8111 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
8112 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
8113 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX, VEX_L;
8114 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
8115 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
8117 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>,
8121 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
8122 int_x86_avx2_pbroadcastb_128,
8123 int_x86_avx2_pbroadcastb_256>;
8124 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
8125 int_x86_avx2_pbroadcastw_128,
8126 int_x86_avx2_pbroadcastw_256>;
8127 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
8128 int_x86_avx2_pbroadcastd_128,
8129 int_x86_avx2_pbroadcastd_256>;
8130 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
8131 int_x86_avx2_pbroadcastq_128,
8132 int_x86_avx2_pbroadcastq_256>;
8134 let Predicates = [HasAVX2] in {
8135 def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
8136 (VPBROADCASTBrm addr:$src)>;
8137 def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
8138 (VPBROADCASTBYrm addr:$src)>;
8139 def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
8140 (VPBROADCASTWrm addr:$src)>;
8141 def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
8142 (VPBROADCASTWYrm addr:$src)>;
8143 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
8144 (VPBROADCASTDrm addr:$src)>;
8145 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
8146 (VPBROADCASTDYrm addr:$src)>;
8147 def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
8148 (VPBROADCASTQrm addr:$src)>;
8149 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
8150 (VPBROADCASTQYrm addr:$src)>;
8152 def : Pat<(v16i8 (X86VBroadcast (v16i8 VR128:$src))),
8153 (VPBROADCASTBrr VR128:$src)>;
8154 def : Pat<(v32i8 (X86VBroadcast (v16i8 VR128:$src))),
8155 (VPBROADCASTBYrr VR128:$src)>;
8156 def : Pat<(v8i16 (X86VBroadcast (v8i16 VR128:$src))),
8157 (VPBROADCASTWrr VR128:$src)>;
8158 def : Pat<(v16i16 (X86VBroadcast (v8i16 VR128:$src))),
8159 (VPBROADCASTWYrr VR128:$src)>;
8160 def : Pat<(v4i32 (X86VBroadcast (v4i32 VR128:$src))),
8161 (VPBROADCASTDrr VR128:$src)>;
8162 def : Pat<(v8i32 (X86VBroadcast (v4i32 VR128:$src))),
8163 (VPBROADCASTDYrr VR128:$src)>;
8164 def : Pat<(v2i64 (X86VBroadcast (v2i64 VR128:$src))),
8165 (VPBROADCASTQrr VR128:$src)>;
8166 def : Pat<(v4i64 (X86VBroadcast (v2i64 VR128:$src))),
8167 (VPBROADCASTQYrr VR128:$src)>;
8168 def : Pat<(v4f32 (X86VBroadcast (v4f32 VR128:$src))),
8169 (VBROADCASTSSrr VR128:$src)>;
8170 def : Pat<(v8f32 (X86VBroadcast (v4f32 VR128:$src))),
8171 (VBROADCASTSSYrr VR128:$src)>;
8172 def : Pat<(v2f64 (X86VBroadcast (v2f64 VR128:$src))),
8173 (VPBROADCASTQrr VR128:$src)>;
8174 def : Pat<(v4f64 (X86VBroadcast (v2f64 VR128:$src))),
8175 (VBROADCASTSDYrr VR128:$src)>;
8177 // Provide fallback in case the load node that is used in the patterns above
8178 // is used by additional users, which prevents the pattern selection.
8179 let AddedComplexity = 20 in {
8180 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
8181 (VBROADCASTSSrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
8182 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
8183 (VBROADCASTSSYrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
8184 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
8185 (VBROADCASTSDYrr (COPY_TO_REGCLASS FR64:$src, VR128))>;
8187 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
8188 (VBROADCASTSSrr (COPY_TO_REGCLASS GR32:$src, VR128))>;
8189 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
8190 (VBROADCASTSSYrr (COPY_TO_REGCLASS GR32:$src, VR128))>;
8191 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
8192 (VBROADCASTSDYrr (COPY_TO_REGCLASS GR64:$src, VR128))>;
8196 // AVX1 broadcast patterns
8197 let Predicates = [HasAVX1Only] in {
8198 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
8199 (VBROADCASTSSYrm addr:$src)>;
8200 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
8201 (VBROADCASTSDYrm addr:$src)>;
8202 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
8203 (VBROADCASTSSrm addr:$src)>;
8206 let Predicates = [HasAVX] in {
8207 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
8208 (VBROADCASTSSYrm addr:$src)>;
8209 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
8210 (VBROADCASTSDYrm addr:$src)>;
8211 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
8212 (VBROADCASTSSrm addr:$src)>;
8214 // Provide fallback in case the load node that is used in the patterns above
8215 // is used by additional users, which prevents the pattern selection.
8216 let AddedComplexity = 20 in {
8217 // 128bit broadcasts:
8218 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
8219 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0)>;
8220 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
8221 (VINSERTF128rr (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)),
8222 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0), sub_xmm),
8223 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0), 1)>;
8224 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
8225 (VINSERTF128rr (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)),
8226 (VPSHUFDri (COPY_TO_REGCLASS FR64:$src, VR128), 0x44), sub_xmm),
8227 (VPSHUFDri (COPY_TO_REGCLASS FR64:$src, VR128), 0x44), 1)>;
8229 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
8230 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0)>;
8231 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
8232 (VINSERTF128rr (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)),
8233 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0), sub_xmm),
8234 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0), 1)>;
8235 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
8236 (VINSERTF128rr (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)),
8237 (VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), sub_xmm),
8238 (VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), 1)>;
8242 //===----------------------------------------------------------------------===//
8243 // VPERM - Permute instructions
8246 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
8248 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
8249 (ins VR256:$src1, VR256:$src2),
8250 !strconcat(OpcodeStr,
8251 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8253 (OpVT (X86VPermv VR256:$src1, VR256:$src2)))]>,
8255 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
8256 (ins VR256:$src1, i256mem:$src2),
8257 !strconcat(OpcodeStr,
8258 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8260 (OpVT (X86VPermv VR256:$src1,
8261 (bitconvert (mem_frag addr:$src2)))))]>,
8265 defm VPERMD : avx2_perm<0x36, "vpermd", loadv4i64, v8i32>;
8266 let ExeDomain = SSEPackedSingle in
8267 defm VPERMPS : avx2_perm<0x16, "vpermps", loadv8f32, v8f32>;
8269 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
8271 def Yri : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
8272 (ins VR256:$src1, i8imm:$src2),
8273 !strconcat(OpcodeStr,
8274 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8276 (OpVT (X86VPermi VR256:$src1, (i8 imm:$src2))))]>,
8278 def Ymi : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
8279 (ins i256mem:$src1, i8imm:$src2),
8280 !strconcat(OpcodeStr,
8281 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8283 (OpVT (X86VPermi (mem_frag addr:$src1),
8284 (i8 imm:$src2))))]>, VEX, VEX_L;
8287 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", loadv4i64, v4i64>, VEX_W;
8288 let ExeDomain = SSEPackedDouble in
8289 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", loadv4f64, v4f64>, VEX_W;
8291 //===----------------------------------------------------------------------===//
8292 // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
8294 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
8295 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
8296 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
8297 [(set VR256:$dst, (v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2,
8298 (i8 imm:$src3))))]>, VEX_4V, VEX_L;
8299 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
8300 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
8301 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
8302 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (loadv4i64 addr:$src2),
8303 (i8 imm:$src3)))]>, VEX_4V, VEX_L;
8305 let Predicates = [HasAVX2] in {
8306 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
8307 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
8308 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
8309 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
8310 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
8311 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
8313 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (loadv4i64 addr:$src2)),
8315 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
8316 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
8317 (bc_v16i16 (loadv4i64 addr:$src2)), (i8 imm:$imm))),
8318 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
8319 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (loadv4i64 addr:$src2)),
8321 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
8325 //===----------------------------------------------------------------------===//
8326 // VINSERTI128 - Insert packed integer values
8328 let neverHasSideEffects = 1 in {
8329 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
8330 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
8331 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
8334 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
8335 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
8336 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
8340 let Predicates = [HasAVX2] in {
8341 def : Pat<(vinsert128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
8343 (VINSERTI128rr VR256:$src1, VR128:$src2,
8344 (INSERT_get_vinsert128_imm VR256:$ins))>;
8345 def : Pat<(vinsert128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
8347 (VINSERTI128rr VR256:$src1, VR128:$src2,
8348 (INSERT_get_vinsert128_imm VR256:$ins))>;
8349 def : Pat<(vinsert128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
8351 (VINSERTI128rr VR256:$src1, VR128:$src2,
8352 (INSERT_get_vinsert128_imm VR256:$ins))>;
8353 def : Pat<(vinsert128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
8355 (VINSERTI128rr VR256:$src1, VR128:$src2,
8356 (INSERT_get_vinsert128_imm VR256:$ins))>;
8358 def : Pat<(vinsert128_insert:$ins (v4i64 VR256:$src1), (loadv2i64 addr:$src2),
8360 (VINSERTI128rm VR256:$src1, addr:$src2,
8361 (INSERT_get_vinsert128_imm VR256:$ins))>;
8362 def : Pat<(vinsert128_insert:$ins (v8i32 VR256:$src1),
8363 (bc_v4i32 (loadv2i64 addr:$src2)),
8365 (VINSERTI128rm VR256:$src1, addr:$src2,
8366 (INSERT_get_vinsert128_imm VR256:$ins))>;
8367 def : Pat<(vinsert128_insert:$ins (v32i8 VR256:$src1),
8368 (bc_v16i8 (loadv2i64 addr:$src2)),
8370 (VINSERTI128rm VR256:$src1, addr:$src2,
8371 (INSERT_get_vinsert128_imm VR256:$ins))>;
8372 def : Pat<(vinsert128_insert:$ins (v16i16 VR256:$src1),
8373 (bc_v8i16 (loadv2i64 addr:$src2)),
8375 (VINSERTI128rm VR256:$src1, addr:$src2,
8376 (INSERT_get_vinsert128_imm VR256:$ins))>;
8379 //===----------------------------------------------------------------------===//
8380 // VEXTRACTI128 - Extract packed integer values
8382 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
8383 (ins VR256:$src1, i8imm:$src2),
8384 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
8386 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
8388 let neverHasSideEffects = 1, mayStore = 1 in
8389 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
8390 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
8391 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
8394 let Predicates = [HasAVX2] in {
8395 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
8396 (v2i64 (VEXTRACTI128rr
8397 (v4i64 VR256:$src1),
8398 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
8399 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
8400 (v4i32 (VEXTRACTI128rr
8401 (v8i32 VR256:$src1),
8402 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
8403 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
8404 (v8i16 (VEXTRACTI128rr
8405 (v16i16 VR256:$src1),
8406 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
8407 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
8408 (v16i8 (VEXTRACTI128rr
8409 (v32i8 VR256:$src1),
8410 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
8412 def : Pat<(store (v2i64 (vextract128_extract:$ext (v4i64 VR256:$src1),
8413 (iPTR imm))), addr:$dst),
8414 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8415 (EXTRACT_get_vextract128_imm VR128:$ext))>;
8416 def : Pat<(store (v4i32 (vextract128_extract:$ext (v8i32 VR256:$src1),
8417 (iPTR imm))), addr:$dst),
8418 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8419 (EXTRACT_get_vextract128_imm VR128:$ext))>;
8420 def : Pat<(store (v8i16 (vextract128_extract:$ext (v16i16 VR256:$src1),
8421 (iPTR imm))), addr:$dst),
8422 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8423 (EXTRACT_get_vextract128_imm VR128:$ext))>;
8424 def : Pat<(store (v16i8 (vextract128_extract:$ext (v32i8 VR256:$src1),
8425 (iPTR imm))), addr:$dst),
8426 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8427 (EXTRACT_get_vextract128_imm VR128:$ext))>;
8430 //===----------------------------------------------------------------------===//
8431 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
8433 multiclass avx2_pmovmask<string OpcodeStr,
8434 Intrinsic IntLd128, Intrinsic IntLd256,
8435 Intrinsic IntSt128, Intrinsic IntSt256> {
8436 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
8437 (ins VR128:$src1, i128mem:$src2),
8438 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8439 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
8440 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
8441 (ins VR256:$src1, i256mem:$src2),
8442 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8443 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
8445 def mr : AVX28I<0x8e, MRMDestMem, (outs),
8446 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
8447 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8448 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
8449 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
8450 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
8451 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8452 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V, VEX_L;
8455 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
8456 int_x86_avx2_maskload_d,
8457 int_x86_avx2_maskload_d_256,
8458 int_x86_avx2_maskstore_d,
8459 int_x86_avx2_maskstore_d_256>;
8460 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
8461 int_x86_avx2_maskload_q,
8462 int_x86_avx2_maskload_q_256,
8463 int_x86_avx2_maskstore_q,
8464 int_x86_avx2_maskstore_q_256>, VEX_W;
8467 //===----------------------------------------------------------------------===//
8468 // Variable Bit Shifts
8470 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
8471 ValueType vt128, ValueType vt256> {
8472 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
8473 (ins VR128:$src1, VR128:$src2),
8474 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8476 (vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
8478 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
8479 (ins VR128:$src1, i128mem:$src2),
8480 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8482 (vt128 (OpNode VR128:$src1,
8483 (vt128 (bitconvert (loadv2i64 addr:$src2))))))]>,
8485 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
8486 (ins VR256:$src1, VR256:$src2),
8487 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8489 (vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
8491 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
8492 (ins VR256:$src1, i256mem:$src2),
8493 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8495 (vt256 (OpNode VR256:$src1,
8496 (vt256 (bitconvert (loadv4i64 addr:$src2))))))]>,
8500 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
8501 defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", shl, v2i64, v4i64>, VEX_W;
8502 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", srl, v4i32, v8i32>;
8503 defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", srl, v2i64, v4i64>, VEX_W;
8504 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;
8506 //===----------------------------------------------------------------------===//
8507 // VGATHER - GATHER Operations
8508 multiclass avx2_gather<bits<8> opc, string OpcodeStr, RegisterClass RC256,
8509 X86MemOperand memop128, X86MemOperand memop256> {
8510 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst, VR128:$mask_wb),
8511 (ins VR128:$src1, memop128:$src2, VR128:$mask),
8512 !strconcat(OpcodeStr,
8513 "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
8515 def Yrm : AVX28I<opc, MRMSrcMem, (outs RC256:$dst, RC256:$mask_wb),
8516 (ins RC256:$src1, memop256:$src2, RC256:$mask),
8517 !strconcat(OpcodeStr,
8518 "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
8519 []>, VEX_4VOp3, VEX_L;
8522 let mayLoad = 1, Constraints
8523 = "@earlyclobber $dst,@earlyclobber $mask_wb, $src1 = $dst, $mask = $mask_wb"
8525 defm VGATHERDPD : avx2_gather<0x92, "vgatherdpd", VR256, vx64mem, vx64mem>, VEX_W;
8526 defm VGATHERQPD : avx2_gather<0x93, "vgatherqpd", VR256, vx64mem, vy64mem>, VEX_W;
8527 defm VGATHERDPS : avx2_gather<0x92, "vgatherdps", VR256, vx32mem, vy32mem>;
8528 defm VGATHERQPS : avx2_gather<0x93, "vgatherqps", VR128, vx32mem, vy32mem>;
8529 defm VPGATHERDQ : avx2_gather<0x90, "vpgatherdq", VR256, vx64mem, vx64mem>, VEX_W;
8530 defm VPGATHERQQ : avx2_gather<0x91, "vpgatherqq", VR256, vx64mem, vy64mem>, VEX_W;
8531 defm VPGATHERDD : avx2_gather<0x90, "vpgatherdd", VR256, vx32mem, vy32mem>;
8532 defm VPGATHERQD : avx2_gather<0x91, "vpgatherqd", VR128, vx32mem, vy32mem>;