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 let isCodeGenOnly = 1 in {
214 def rr_Int : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
216 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
217 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
218 [(set RC:$dst, (!cast<Intrinsic>(
219 !strconcat("int_x86_sse", SSEVer, "_", OpcodeStr, FPSizeStr))
220 RC:$src1, RC:$src2))], itins.rr>,
221 Sched<[itins.Sched]>;
222 def rm_Int : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, memopr:$src2),
224 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
225 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
226 [(set RC:$dst, (!cast<Intrinsic>(!strconcat("int_x86_sse",
227 SSEVer, "_", OpcodeStr, FPSizeStr))
228 RC:$src1, mem_cpat:$src2))], itins.rm>,
229 Sched<[itins.Sched.Folded, ReadAfterLd]>;
233 /// sse12_fp_packed - SSE 1 & 2 packed instructions class
234 multiclass sse12_fp_packed<bits<8> opc, string OpcodeStr, SDNode OpNode,
235 RegisterClass RC, ValueType vt,
236 X86MemOperand x86memop, PatFrag mem_frag,
237 Domain d, OpndItins itins, bit Is2Addr = 1> {
238 let isCommutable = 1 in
239 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
241 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
242 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
243 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], itins.rr, d>,
244 Sched<[itins.Sched]>;
246 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
248 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
249 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
250 [(set RC:$dst, (OpNode RC:$src1, (mem_frag addr:$src2)))],
252 Sched<[itins.Sched.Folded, ReadAfterLd]>;
255 /// sse12_fp_packed_logical_rm - SSE 1 & 2 packed instructions class
256 multiclass sse12_fp_packed_logical_rm<bits<8> opc, RegisterClass RC, Domain d,
257 string OpcodeStr, X86MemOperand x86memop,
258 list<dag> pat_rr, list<dag> pat_rm,
260 let isCommutable = 1, hasSideEffects = 0 in
261 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
263 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
264 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
265 pat_rr, NoItinerary, d>,
266 Sched<[WriteVecLogic]>;
267 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
269 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
270 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
271 pat_rm, NoItinerary, d>,
272 Sched<[WriteVecLogicLd, ReadAfterLd]>;
275 //===----------------------------------------------------------------------===//
276 // Non-instruction patterns
277 //===----------------------------------------------------------------------===//
279 // A vector extract of the first f32/f64 position is a subregister copy
280 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
281 (COPY_TO_REGCLASS (v4f32 VR128:$src), FR32)>;
282 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
283 (COPY_TO_REGCLASS (v2f64 VR128:$src), FR64)>;
285 // A 128-bit subvector extract from the first 256-bit vector position
286 // is a subregister copy that needs no instruction.
287 def : Pat<(v4i32 (extract_subvector (v8i32 VR256:$src), (iPTR 0))),
288 (v4i32 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm))>;
289 def : Pat<(v4f32 (extract_subvector (v8f32 VR256:$src), (iPTR 0))),
290 (v4f32 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_xmm))>;
292 def : Pat<(v2i64 (extract_subvector (v4i64 VR256:$src), (iPTR 0))),
293 (v2i64 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm))>;
294 def : Pat<(v2f64 (extract_subvector (v4f64 VR256:$src), (iPTR 0))),
295 (v2f64 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm))>;
297 def : Pat<(v8i16 (extract_subvector (v16i16 VR256:$src), (iPTR 0))),
298 (v8i16 (EXTRACT_SUBREG (v16i16 VR256:$src), sub_xmm))>;
299 def : Pat<(v16i8 (extract_subvector (v32i8 VR256:$src), (iPTR 0))),
300 (v16i8 (EXTRACT_SUBREG (v32i8 VR256:$src), sub_xmm))>;
302 // A 128-bit subvector insert to the first 256-bit vector position
303 // is a subregister copy that needs no instruction.
304 let AddedComplexity = 25 in { // to give priority over vinsertf128rm
305 def : Pat<(insert_subvector undef, (v2i64 VR128:$src), (iPTR 0)),
306 (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
307 def : Pat<(insert_subvector undef, (v2f64 VR128:$src), (iPTR 0)),
308 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
309 def : Pat<(insert_subvector undef, (v4i32 VR128:$src), (iPTR 0)),
310 (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
311 def : Pat<(insert_subvector undef, (v4f32 VR128:$src), (iPTR 0)),
312 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
313 def : Pat<(insert_subvector undef, (v8i16 VR128:$src), (iPTR 0)),
314 (INSERT_SUBREG (v16i16 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
315 def : Pat<(insert_subvector undef, (v16i8 VR128:$src), (iPTR 0)),
316 (INSERT_SUBREG (v32i8 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
319 // Implicitly promote a 32-bit scalar to a vector.
320 def : Pat<(v4f32 (scalar_to_vector FR32:$src)),
321 (COPY_TO_REGCLASS FR32:$src, VR128)>;
322 def : Pat<(v8f32 (scalar_to_vector FR32:$src)),
323 (COPY_TO_REGCLASS FR32:$src, VR128)>;
324 // Implicitly promote a 64-bit scalar to a vector.
325 def : Pat<(v2f64 (scalar_to_vector FR64:$src)),
326 (COPY_TO_REGCLASS FR64:$src, VR128)>;
327 def : Pat<(v4f64 (scalar_to_vector FR64:$src)),
328 (COPY_TO_REGCLASS FR64:$src, VR128)>;
330 // Bitcasts between 128-bit vector types. Return the original type since
331 // no instruction is needed for the conversion
332 let Predicates = [HasSSE2] in {
333 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
334 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
335 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
336 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
337 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
338 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
339 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
340 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
341 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
342 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
343 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
344 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
345 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
346 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
347 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
348 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
349 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
350 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
351 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
352 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
353 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
354 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
355 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
356 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
357 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
358 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
359 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
360 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
361 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
362 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
365 // Bitcasts between 256-bit vector types. Return the original type since
366 // no instruction is needed for the conversion
367 let Predicates = [HasAVX] in {
368 def : Pat<(v4f64 (bitconvert (v8f32 VR256:$src))), (v4f64 VR256:$src)>;
369 def : Pat<(v4f64 (bitconvert (v8i32 VR256:$src))), (v4f64 VR256:$src)>;
370 def : Pat<(v4f64 (bitconvert (v4i64 VR256:$src))), (v4f64 VR256:$src)>;
371 def : Pat<(v4f64 (bitconvert (v16i16 VR256:$src))), (v4f64 VR256:$src)>;
372 def : Pat<(v4f64 (bitconvert (v32i8 VR256:$src))), (v4f64 VR256:$src)>;
373 def : Pat<(v8f32 (bitconvert (v8i32 VR256:$src))), (v8f32 VR256:$src)>;
374 def : Pat<(v8f32 (bitconvert (v4i64 VR256:$src))), (v8f32 VR256:$src)>;
375 def : Pat<(v8f32 (bitconvert (v4f64 VR256:$src))), (v8f32 VR256:$src)>;
376 def : Pat<(v8f32 (bitconvert (v32i8 VR256:$src))), (v8f32 VR256:$src)>;
377 def : Pat<(v8f32 (bitconvert (v16i16 VR256:$src))), (v8f32 VR256:$src)>;
378 def : Pat<(v4i64 (bitconvert (v8f32 VR256:$src))), (v4i64 VR256:$src)>;
379 def : Pat<(v4i64 (bitconvert (v8i32 VR256:$src))), (v4i64 VR256:$src)>;
380 def : Pat<(v4i64 (bitconvert (v4f64 VR256:$src))), (v4i64 VR256:$src)>;
381 def : Pat<(v4i64 (bitconvert (v32i8 VR256:$src))), (v4i64 VR256:$src)>;
382 def : Pat<(v4i64 (bitconvert (v16i16 VR256:$src))), (v4i64 VR256:$src)>;
383 def : Pat<(v32i8 (bitconvert (v4f64 VR256:$src))), (v32i8 VR256:$src)>;
384 def : Pat<(v32i8 (bitconvert (v4i64 VR256:$src))), (v32i8 VR256:$src)>;
385 def : Pat<(v32i8 (bitconvert (v8f32 VR256:$src))), (v32i8 VR256:$src)>;
386 def : Pat<(v32i8 (bitconvert (v8i32 VR256:$src))), (v32i8 VR256:$src)>;
387 def : Pat<(v32i8 (bitconvert (v16i16 VR256:$src))), (v32i8 VR256:$src)>;
388 def : Pat<(v8i32 (bitconvert (v32i8 VR256:$src))), (v8i32 VR256:$src)>;
389 def : Pat<(v8i32 (bitconvert (v16i16 VR256:$src))), (v8i32 VR256:$src)>;
390 def : Pat<(v8i32 (bitconvert (v8f32 VR256:$src))), (v8i32 VR256:$src)>;
391 def : Pat<(v8i32 (bitconvert (v4i64 VR256:$src))), (v8i32 VR256:$src)>;
392 def : Pat<(v8i32 (bitconvert (v4f64 VR256:$src))), (v8i32 VR256:$src)>;
393 def : Pat<(v16i16 (bitconvert (v8f32 VR256:$src))), (v16i16 VR256:$src)>;
394 def : Pat<(v16i16 (bitconvert (v8i32 VR256:$src))), (v16i16 VR256:$src)>;
395 def : Pat<(v16i16 (bitconvert (v4i64 VR256:$src))), (v16i16 VR256:$src)>;
396 def : Pat<(v16i16 (bitconvert (v4f64 VR256:$src))), (v16i16 VR256:$src)>;
397 def : Pat<(v16i16 (bitconvert (v32i8 VR256:$src))), (v16i16 VR256:$src)>;
400 // Alias instructions that map fld0 to xorps for sse or vxorps for avx.
401 // This is expanded by ExpandPostRAPseudos.
402 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
403 isPseudo = 1, SchedRW = [WriteZero] in {
404 def FsFLD0SS : I<0, Pseudo, (outs FR32:$dst), (ins), "",
405 [(set FR32:$dst, fp32imm0)]>, Requires<[HasSSE1]>;
406 def FsFLD0SD : I<0, Pseudo, (outs FR64:$dst), (ins), "",
407 [(set FR64:$dst, fpimm0)]>, Requires<[HasSSE2]>;
410 //===----------------------------------------------------------------------===//
411 // AVX & SSE - Zero/One Vectors
412 //===----------------------------------------------------------------------===//
414 // Alias instruction that maps zero vector to pxor / xorp* for sse.
415 // This is expanded by ExpandPostRAPseudos to an xorps / vxorps, and then
416 // swizzled by ExecutionDepsFix to pxor.
417 // We set canFoldAsLoad because this can be converted to a constant-pool
418 // load of an all-zeros value if folding it would be beneficial.
419 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
420 isPseudo = 1, SchedRW = [WriteZero] in {
421 def V_SET0 : I<0, Pseudo, (outs VR128:$dst), (ins), "",
422 [(set VR128:$dst, (v4f32 immAllZerosV))]>;
425 def : Pat<(v2f64 immAllZerosV), (V_SET0)>;
426 def : Pat<(v4i32 immAllZerosV), (V_SET0)>;
427 def : Pat<(v2i64 immAllZerosV), (V_SET0)>;
428 def : Pat<(v8i16 immAllZerosV), (V_SET0)>;
429 def : Pat<(v16i8 immAllZerosV), (V_SET0)>;
432 // The same as done above but for AVX. The 256-bit AVX1 ISA doesn't support PI,
433 // and doesn't need it because on sandy bridge the register is set to zero
434 // at the rename stage without using any execution unit, so SET0PSY
435 // and SET0PDY can be used for vector int instructions without penalty
436 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
437 isPseudo = 1, Predicates = [HasAVX], SchedRW = [WriteZero] in {
438 def AVX_SET0 : I<0, Pseudo, (outs VR256:$dst), (ins), "",
439 [(set VR256:$dst, (v8f32 immAllZerosV))]>;
442 let Predicates = [HasAVX] in
443 def : Pat<(v4f64 immAllZerosV), (AVX_SET0)>;
445 let Predicates = [HasAVX2] in {
446 def : Pat<(v4i64 immAllZerosV), (AVX_SET0)>;
447 def : Pat<(v8i32 immAllZerosV), (AVX_SET0)>;
448 def : Pat<(v16i16 immAllZerosV), (AVX_SET0)>;
449 def : Pat<(v32i8 immAllZerosV), (AVX_SET0)>;
452 // AVX1 has no support for 256-bit integer instructions, but since the 128-bit
453 // VPXOR instruction writes zero to its upper part, it's safe build zeros.
454 let Predicates = [HasAVX1Only] in {
455 def : Pat<(v32i8 immAllZerosV), (SUBREG_TO_REG (i8 0), (V_SET0), sub_xmm)>;
456 def : Pat<(bc_v32i8 (v8f32 immAllZerosV)),
457 (SUBREG_TO_REG (i8 0), (V_SET0), sub_xmm)>;
459 def : Pat<(v16i16 immAllZerosV), (SUBREG_TO_REG (i16 0), (V_SET0), sub_xmm)>;
460 def : Pat<(bc_v16i16 (v8f32 immAllZerosV)),
461 (SUBREG_TO_REG (i16 0), (V_SET0), sub_xmm)>;
463 def : Pat<(v8i32 immAllZerosV), (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
464 def : Pat<(bc_v8i32 (v8f32 immAllZerosV)),
465 (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
467 def : Pat<(v4i64 immAllZerosV), (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
468 def : Pat<(bc_v4i64 (v8f32 immAllZerosV)),
469 (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
472 // We set canFoldAsLoad because this can be converted to a constant-pool
473 // load of an all-ones value if folding it would be beneficial.
474 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
475 isPseudo = 1, SchedRW = [WriteZero] in {
476 def V_SETALLONES : I<0, Pseudo, (outs VR128:$dst), (ins), "",
477 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
478 let Predicates = [HasAVX2] in
479 def AVX2_SETALLONES : I<0, Pseudo, (outs VR256:$dst), (ins), "",
480 [(set VR256:$dst, (v8i32 immAllOnesV))]>;
484 //===----------------------------------------------------------------------===//
485 // SSE 1 & 2 - Move FP Scalar Instructions
487 // Move Instructions. Register-to-register movss/movsd is not used for FR32/64
488 // register copies because it's a partial register update; Register-to-register
489 // movss/movsd is not modeled as an INSERT_SUBREG because INSERT_SUBREG requires
490 // that the insert be implementable in terms of a copy, and just mentioned, we
491 // don't use movss/movsd for copies.
492 //===----------------------------------------------------------------------===//
494 multiclass sse12_move_rr<RegisterClass RC, SDNode OpNode, ValueType vt,
495 X86MemOperand x86memop, string base_opc,
497 def rr : SI<0x10, MRMSrcReg, (outs VR128:$dst),
498 (ins VR128:$src1, RC:$src2),
499 !strconcat(base_opc, asm_opr),
500 [(set VR128:$dst, (vt (OpNode VR128:$src1,
501 (scalar_to_vector RC:$src2))))],
502 IIC_SSE_MOV_S_RR>, Sched<[WriteMove]>;
504 // For the disassembler
505 let isCodeGenOnly = 1, hasSideEffects = 0 in
506 def rr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
507 (ins VR128:$src1, RC:$src2),
508 !strconcat(base_opc, asm_opr),
509 [], IIC_SSE_MOV_S_RR>, Sched<[WriteMove]>;
512 multiclass sse12_move<RegisterClass RC, SDNode OpNode, ValueType vt,
513 X86MemOperand x86memop, string OpcodeStr> {
515 defm V#NAME : sse12_move_rr<RC, OpNode, vt, x86memop, OpcodeStr,
516 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">,
519 def V#NAME#mr : SI<0x11, MRMDestMem, (outs), (ins x86memop:$dst, RC:$src),
520 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
521 [(store RC:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
522 VEX, VEX_LIG, Sched<[WriteStore]>;
524 let Constraints = "$src1 = $dst" in {
525 defm NAME : sse12_move_rr<RC, OpNode, vt, x86memop, OpcodeStr,
526 "\t{$src2, $dst|$dst, $src2}">;
529 def NAME#mr : SI<0x11, MRMDestMem, (outs), (ins x86memop:$dst, RC:$src),
530 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
531 [(store RC:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
535 // Loading from memory automatically zeroing upper bits.
536 multiclass sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
537 PatFrag mem_pat, string OpcodeStr> {
538 def V#NAME#rm : SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
539 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
540 [(set RC:$dst, (mem_pat addr:$src))],
541 IIC_SSE_MOV_S_RM>, VEX, VEX_LIG, Sched<[WriteLoad]>;
542 def NAME#rm : SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
543 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
544 [(set RC:$dst, (mem_pat addr:$src))],
545 IIC_SSE_MOV_S_RM>, Sched<[WriteLoad]>;
548 defm MOVSS : sse12_move<FR32, X86Movss, v4f32, f32mem, "movss">, XS;
549 defm MOVSD : sse12_move<FR64, X86Movsd, v2f64, f64mem, "movsd">, XD;
551 let canFoldAsLoad = 1, isReMaterializable = 1 in {
552 defm MOVSS : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
554 let AddedComplexity = 20 in
555 defm MOVSD : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
559 let Predicates = [UseAVX] in {
560 let AddedComplexity = 15 in {
561 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
562 // MOVS{S,D} to the lower bits.
563 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
564 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
565 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
566 (VMOVSSrr (v4f32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
567 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
568 (VMOVSSrr (v4i32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
569 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
570 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
572 // Move low f32 and clear high bits.
573 def : Pat<(v8f32 (X86vzmovl (v8f32 VR256:$src))),
574 (SUBREG_TO_REG (i32 0),
575 (VMOVSSrr (v4f32 (V_SET0)),
576 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_xmm)), sub_xmm)>;
577 def : Pat<(v8i32 (X86vzmovl (v8i32 VR256:$src))),
578 (SUBREG_TO_REG (i32 0),
579 (VMOVSSrr (v4i32 (V_SET0)),
580 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm)), sub_xmm)>;
583 let AddedComplexity = 20 in {
584 // MOVSSrm zeros the high parts of the register; represent this
585 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
586 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
587 (COPY_TO_REGCLASS (VMOVSSrm addr:$src), VR128)>;
588 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
589 (COPY_TO_REGCLASS (VMOVSSrm addr:$src), VR128)>;
590 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
591 (COPY_TO_REGCLASS (VMOVSSrm addr:$src), VR128)>;
593 // MOVSDrm zeros the high parts of the register; represent this
594 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
595 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
596 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
597 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
598 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
599 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
600 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
601 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
602 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
603 def : Pat<(v2f64 (X86vzload addr:$src)),
604 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
606 // Represent the same patterns above but in the form they appear for
608 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
609 (v4i32 (scalar_to_vector (loadi32 addr:$src))), (iPTR 0)))),
610 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_xmm)>;
611 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
612 (v4f32 (scalar_to_vector (loadf32 addr:$src))), (iPTR 0)))),
613 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_xmm)>;
614 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
615 (v2f64 (scalar_to_vector (loadf64 addr:$src))), (iPTR 0)))),
616 (SUBREG_TO_REG (i32 0), (VMOVSDrm addr:$src), sub_xmm)>;
618 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
619 (v4f32 (scalar_to_vector FR32:$src)), (iPTR 0)))),
620 (SUBREG_TO_REG (i32 0),
621 (v4f32 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)),
623 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
624 (v2f64 (scalar_to_vector FR64:$src)), (iPTR 0)))),
625 (SUBREG_TO_REG (i64 0),
626 (v2f64 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)),
628 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
629 (v2i64 (scalar_to_vector (loadi64 addr:$src))), (iPTR 0)))),
630 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_xmm)>;
632 // Move low f64 and clear high bits.
633 def : Pat<(v4f64 (X86vzmovl (v4f64 VR256:$src))),
634 (SUBREG_TO_REG (i32 0),
635 (VMOVSDrr (v2f64 (V_SET0)),
636 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm)), sub_xmm)>;
638 def : Pat<(v4i64 (X86vzmovl (v4i64 VR256:$src))),
639 (SUBREG_TO_REG (i32 0),
640 (VMOVSDrr (v2i64 (V_SET0)),
641 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm)), sub_xmm)>;
643 // Extract and store.
644 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
646 (VMOVSSmr addr:$dst, (COPY_TO_REGCLASS (v4f32 VR128:$src), FR32))>;
647 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
649 (VMOVSDmr addr:$dst, (COPY_TO_REGCLASS (v2f64 VR128:$src), FR64))>;
651 // Shuffle with VMOVSS
652 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
653 (VMOVSSrr (v4i32 VR128:$src1),
654 (COPY_TO_REGCLASS (v4i32 VR128:$src2), FR32))>;
655 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
656 (VMOVSSrr (v4f32 VR128:$src1),
657 (COPY_TO_REGCLASS (v4f32 VR128:$src2), FR32))>;
660 def : Pat<(v8i32 (X86Movss VR256:$src1, VR256:$src2)),
661 (SUBREG_TO_REG (i32 0),
662 (VMOVSSrr (EXTRACT_SUBREG (v8i32 VR256:$src1), sub_xmm),
663 (EXTRACT_SUBREG (v8i32 VR256:$src2), sub_xmm)),
665 def : Pat<(v8f32 (X86Movss VR256:$src1, VR256:$src2)),
666 (SUBREG_TO_REG (i32 0),
667 (VMOVSSrr (EXTRACT_SUBREG (v8f32 VR256:$src1), sub_xmm),
668 (EXTRACT_SUBREG (v8f32 VR256:$src2), sub_xmm)),
671 // Shuffle with VMOVSD
672 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
673 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
674 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
675 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
676 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
677 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
678 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
679 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
682 def : Pat<(v4i64 (X86Movsd VR256:$src1, VR256:$src2)),
683 (SUBREG_TO_REG (i32 0),
684 (VMOVSDrr (EXTRACT_SUBREG (v4i64 VR256:$src1), sub_xmm),
685 (EXTRACT_SUBREG (v4i64 VR256:$src2), sub_xmm)),
687 def : Pat<(v4f64 (X86Movsd VR256:$src1, VR256:$src2)),
688 (SUBREG_TO_REG (i32 0),
689 (VMOVSDrr (EXTRACT_SUBREG (v4f64 VR256:$src1), sub_xmm),
690 (EXTRACT_SUBREG (v4f64 VR256:$src2), sub_xmm)),
694 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
695 // is during lowering, where it's not possible to recognize the fold cause
696 // it has two uses through a bitcast. One use disappears at isel time and the
697 // fold opportunity reappears.
698 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
699 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
700 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
701 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
702 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
703 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
704 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
705 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
708 let Predicates = [UseSSE1] in {
709 let AddedComplexity = 15 in {
710 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
711 // MOVSS to the lower bits.
712 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
713 (MOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
714 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
715 (MOVSSrr (v4f32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
716 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
717 (MOVSSrr (v4i32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
720 let AddedComplexity = 20 in {
721 // MOVSSrm already zeros the high parts of the register.
722 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
723 (COPY_TO_REGCLASS (MOVSSrm addr:$src), VR128)>;
724 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
725 (COPY_TO_REGCLASS (MOVSSrm addr:$src), VR128)>;
726 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
727 (COPY_TO_REGCLASS (MOVSSrm addr:$src), VR128)>;
730 // Extract and store.
731 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
733 (MOVSSmr addr:$dst, (COPY_TO_REGCLASS VR128:$src, FR32))>;
735 // Shuffle with MOVSS
736 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
737 (MOVSSrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR32))>;
738 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
739 (MOVSSrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR32))>;
742 let Predicates = [UseSSE2] in {
743 let AddedComplexity = 15 in {
744 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
745 // MOVSD to the lower bits.
746 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
747 (MOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
750 let AddedComplexity = 20 in {
751 // MOVSDrm already zeros the high parts of the register.
752 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
753 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
754 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
755 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
756 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
757 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
758 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
759 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
760 def : Pat<(v2f64 (X86vzload addr:$src)),
761 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
764 // Extract and store.
765 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
767 (MOVSDmr addr:$dst, (COPY_TO_REGCLASS VR128:$src, FR64))>;
769 // Shuffle with MOVSD
770 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
771 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
772 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
773 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
774 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
775 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
776 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
777 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
779 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
780 // is during lowering, where it's not possible to recognize the fold cause
781 // it has two uses through a bitcast. One use disappears at isel time and the
782 // fold opportunity reappears.
783 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
784 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
785 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
786 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
787 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
788 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
789 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
790 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
793 //===----------------------------------------------------------------------===//
794 // SSE 1 & 2 - Move Aligned/Unaligned FP Instructions
795 //===----------------------------------------------------------------------===//
797 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
798 X86MemOperand x86memop, PatFrag ld_frag,
799 string asm, Domain d,
801 bit IsReMaterializable = 1> {
802 let neverHasSideEffects = 1 in
803 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
804 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], itins.rr, d>,
806 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
807 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
808 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
809 [(set RC:$dst, (ld_frag addr:$src))], itins.rm, d>,
813 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
814 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
816 defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
817 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
819 defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
820 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
822 defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
823 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
826 defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
827 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
829 defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
830 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
831 TB, OpSize, VEX, VEX_L;
832 defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
833 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
835 defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
836 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
837 TB, OpSize, VEX, VEX_L;
838 defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
839 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
841 defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
842 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
844 defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
845 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
847 defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
848 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
851 let SchedRW = [WriteStore] in {
852 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
853 "movaps\t{$src, $dst|$dst, $src}",
854 [(alignedstore (v4f32 VR128:$src), addr:$dst)],
855 IIC_SSE_MOVA_P_MR>, VEX;
856 def VMOVAPDmr : VPDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
857 "movapd\t{$src, $dst|$dst, $src}",
858 [(alignedstore (v2f64 VR128:$src), addr:$dst)],
859 IIC_SSE_MOVA_P_MR>, VEX;
860 def VMOVUPSmr : VPSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
861 "movups\t{$src, $dst|$dst, $src}",
862 [(store (v4f32 VR128:$src), addr:$dst)],
863 IIC_SSE_MOVU_P_MR>, VEX;
864 def VMOVUPDmr : VPDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
865 "movupd\t{$src, $dst|$dst, $src}",
866 [(store (v2f64 VR128:$src), addr:$dst)],
867 IIC_SSE_MOVU_P_MR>, VEX;
868 def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
869 "movaps\t{$src, $dst|$dst, $src}",
870 [(alignedstore256 (v8f32 VR256:$src), addr:$dst)],
871 IIC_SSE_MOVA_P_MR>, VEX, VEX_L;
872 def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
873 "movapd\t{$src, $dst|$dst, $src}",
874 [(alignedstore256 (v4f64 VR256:$src), addr:$dst)],
875 IIC_SSE_MOVA_P_MR>, VEX, VEX_L;
876 def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
877 "movups\t{$src, $dst|$dst, $src}",
878 [(store (v8f32 VR256:$src), addr:$dst)],
879 IIC_SSE_MOVU_P_MR>, VEX, VEX_L;
880 def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
881 "movupd\t{$src, $dst|$dst, $src}",
882 [(store (v4f64 VR256:$src), addr:$dst)],
883 IIC_SSE_MOVU_P_MR>, VEX, VEX_L;
887 let isCodeGenOnly = 1, hasSideEffects = 0, SchedRW = [WriteMove] in {
888 def VMOVAPSrr_REV : VPSI<0x29, MRMDestReg, (outs VR128:$dst),
890 "movaps\t{$src, $dst|$dst, $src}", [],
891 IIC_SSE_MOVA_P_RR>, VEX;
892 def VMOVAPDrr_REV : VPDI<0x29, MRMDestReg, (outs VR128:$dst),
894 "movapd\t{$src, $dst|$dst, $src}", [],
895 IIC_SSE_MOVA_P_RR>, VEX;
896 def VMOVUPSrr_REV : VPSI<0x11, MRMDestReg, (outs VR128:$dst),
898 "movups\t{$src, $dst|$dst, $src}", [],
899 IIC_SSE_MOVU_P_RR>, VEX;
900 def VMOVUPDrr_REV : VPDI<0x11, MRMDestReg, (outs VR128:$dst),
902 "movupd\t{$src, $dst|$dst, $src}", [],
903 IIC_SSE_MOVU_P_RR>, VEX;
904 def VMOVAPSYrr_REV : VPSI<0x29, MRMDestReg, (outs VR256:$dst),
906 "movaps\t{$src, $dst|$dst, $src}", [],
907 IIC_SSE_MOVA_P_RR>, VEX, VEX_L;
908 def VMOVAPDYrr_REV : VPDI<0x29, MRMDestReg, (outs VR256:$dst),
910 "movapd\t{$src, $dst|$dst, $src}", [],
911 IIC_SSE_MOVA_P_RR>, VEX, VEX_L;
912 def VMOVUPSYrr_REV : VPSI<0x11, MRMDestReg, (outs VR256:$dst),
914 "movups\t{$src, $dst|$dst, $src}", [],
915 IIC_SSE_MOVU_P_RR>, VEX, VEX_L;
916 def VMOVUPDYrr_REV : VPDI<0x11, MRMDestReg, (outs VR256:$dst),
918 "movupd\t{$src, $dst|$dst, $src}", [],
919 IIC_SSE_MOVU_P_RR>, VEX, VEX_L;
922 let Predicates = [HasAVX] in {
923 def : Pat<(v8i32 (X86vzmovl
924 (insert_subvector undef, (v4i32 VR128:$src), (iPTR 0)))),
925 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
926 def : Pat<(v4i64 (X86vzmovl
927 (insert_subvector undef, (v2i64 VR128:$src), (iPTR 0)))),
928 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
929 def : Pat<(v8f32 (X86vzmovl
930 (insert_subvector undef, (v4f32 VR128:$src), (iPTR 0)))),
931 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
932 def : Pat<(v4f64 (X86vzmovl
933 (insert_subvector undef, (v2f64 VR128:$src), (iPTR 0)))),
934 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
938 def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
939 (VMOVUPSYmr addr:$dst, VR256:$src)>;
940 def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
941 (VMOVUPDYmr addr:$dst, VR256:$src)>;
943 let SchedRW = [WriteStore] in {
944 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
945 "movaps\t{$src, $dst|$dst, $src}",
946 [(alignedstore (v4f32 VR128:$src), addr:$dst)],
948 def MOVAPDmr : PDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
949 "movapd\t{$src, $dst|$dst, $src}",
950 [(alignedstore (v2f64 VR128:$src), addr:$dst)],
952 def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
953 "movups\t{$src, $dst|$dst, $src}",
954 [(store (v4f32 VR128:$src), addr:$dst)],
956 def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
957 "movupd\t{$src, $dst|$dst, $src}",
958 [(store (v2f64 VR128:$src), addr:$dst)],
963 let isCodeGenOnly = 1, hasSideEffects = 0, SchedRW = [WriteMove] in {
964 def MOVAPSrr_REV : PSI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
965 "movaps\t{$src, $dst|$dst, $src}", [],
967 def MOVAPDrr_REV : PDI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
968 "movapd\t{$src, $dst|$dst, $src}", [],
970 def MOVUPSrr_REV : PSI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
971 "movups\t{$src, $dst|$dst, $src}", [],
973 def MOVUPDrr_REV : PDI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
974 "movupd\t{$src, $dst|$dst, $src}", [],
978 let Predicates = [HasAVX] in {
979 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
980 (VMOVUPSmr addr:$dst, VR128:$src)>;
981 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
982 (VMOVUPDmr addr:$dst, VR128:$src)>;
985 let Predicates = [UseSSE1] in
986 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
987 (MOVUPSmr addr:$dst, VR128:$src)>;
988 let Predicates = [UseSSE2] in
989 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
990 (MOVUPDmr addr:$dst, VR128:$src)>;
992 // Use vmovaps/vmovups for AVX integer load/store.
993 let Predicates = [HasAVX] in {
994 // 128-bit load/store
995 def : Pat<(alignedloadv2i64 addr:$src),
996 (VMOVAPSrm addr:$src)>;
997 def : Pat<(loadv2i64 addr:$src),
998 (VMOVUPSrm addr:$src)>;
1000 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
1001 (VMOVAPSmr addr:$dst, VR128:$src)>;
1002 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
1003 (VMOVAPSmr addr:$dst, VR128:$src)>;
1004 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
1005 (VMOVAPSmr addr:$dst, VR128:$src)>;
1006 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
1007 (VMOVAPSmr addr:$dst, VR128:$src)>;
1008 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
1009 (VMOVUPSmr addr:$dst, VR128:$src)>;
1010 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
1011 (VMOVUPSmr addr:$dst, VR128:$src)>;
1012 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
1013 (VMOVUPSmr addr:$dst, VR128:$src)>;
1014 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
1015 (VMOVUPSmr addr:$dst, VR128:$src)>;
1017 // 256-bit load/store
1018 def : Pat<(alignedloadv4i64 addr:$src),
1019 (VMOVAPSYrm addr:$src)>;
1020 def : Pat<(loadv4i64 addr:$src),
1021 (VMOVUPSYrm addr:$src)>;
1022 def : Pat<(alignedstore256 (v4i64 VR256:$src), addr:$dst),
1023 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1024 def : Pat<(alignedstore256 (v8i32 VR256:$src), addr:$dst),
1025 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1026 def : Pat<(alignedstore256 (v16i16 VR256:$src), addr:$dst),
1027 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1028 def : Pat<(alignedstore256 (v32i8 VR256:$src), addr:$dst),
1029 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1030 def : Pat<(store (v4i64 VR256:$src), addr:$dst),
1031 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1032 def : Pat<(store (v8i32 VR256:$src), addr:$dst),
1033 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1034 def : Pat<(store (v16i16 VR256:$src), addr:$dst),
1035 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1036 def : Pat<(store (v32i8 VR256:$src), addr:$dst),
1037 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1039 // Special patterns for storing subvector extracts of lower 128-bits
1040 // Its cheaper to just use VMOVAPS/VMOVUPS instead of VEXTRACTF128mr
1041 def : Pat<(alignedstore (v2f64 (extract_subvector
1042 (v4f64 VR256:$src), (iPTR 0))), addr:$dst),
1043 (VMOVAPDmr addr:$dst, (v2f64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1044 def : Pat<(alignedstore (v4f32 (extract_subvector
1045 (v8f32 VR256:$src), (iPTR 0))), addr:$dst),
1046 (VMOVAPSmr addr:$dst, (v4f32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1047 def : Pat<(alignedstore (v2i64 (extract_subvector
1048 (v4i64 VR256:$src), (iPTR 0))), addr:$dst),
1049 (VMOVAPDmr addr:$dst, (v2i64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1050 def : Pat<(alignedstore (v4i32 (extract_subvector
1051 (v8i32 VR256:$src), (iPTR 0))), addr:$dst),
1052 (VMOVAPSmr addr:$dst, (v4i32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1053 def : Pat<(alignedstore (v8i16 (extract_subvector
1054 (v16i16 VR256:$src), (iPTR 0))), addr:$dst),
1055 (VMOVAPSmr addr:$dst, (v8i16 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1056 def : Pat<(alignedstore (v16i8 (extract_subvector
1057 (v32i8 VR256:$src), (iPTR 0))), addr:$dst),
1058 (VMOVAPSmr addr:$dst, (v16i8 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1060 def : Pat<(store (v2f64 (extract_subvector
1061 (v4f64 VR256:$src), (iPTR 0))), addr:$dst),
1062 (VMOVUPDmr addr:$dst, (v2f64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1063 def : Pat<(store (v4f32 (extract_subvector
1064 (v8f32 VR256:$src), (iPTR 0))), addr:$dst),
1065 (VMOVUPSmr addr:$dst, (v4f32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1066 def : Pat<(store (v2i64 (extract_subvector
1067 (v4i64 VR256:$src), (iPTR 0))), addr:$dst),
1068 (VMOVUPDmr addr:$dst, (v2i64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1069 def : Pat<(store (v4i32 (extract_subvector
1070 (v8i32 VR256:$src), (iPTR 0))), addr:$dst),
1071 (VMOVUPSmr addr:$dst, (v4i32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1072 def : Pat<(store (v8i16 (extract_subvector
1073 (v16i16 VR256:$src), (iPTR 0))), addr:$dst),
1074 (VMOVUPSmr addr:$dst, (v8i16 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1075 def : Pat<(store (v16i8 (extract_subvector
1076 (v32i8 VR256:$src), (iPTR 0))), addr:$dst),
1077 (VMOVUPSmr addr:$dst, (v16i8 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1080 // Use movaps / movups for SSE integer load / store (one byte shorter).
1081 // The instructions selected below are then converted to MOVDQA/MOVDQU
1082 // during the SSE domain pass.
1083 let Predicates = [UseSSE1] in {
1084 def : Pat<(alignedloadv2i64 addr:$src),
1085 (MOVAPSrm addr:$src)>;
1086 def : Pat<(loadv2i64 addr:$src),
1087 (MOVUPSrm addr:$src)>;
1089 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
1090 (MOVAPSmr addr:$dst, VR128:$src)>;
1091 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
1092 (MOVAPSmr addr:$dst, VR128:$src)>;
1093 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
1094 (MOVAPSmr addr:$dst, VR128:$src)>;
1095 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
1096 (MOVAPSmr addr:$dst, VR128:$src)>;
1097 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
1098 (MOVUPSmr addr:$dst, VR128:$src)>;
1099 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
1100 (MOVUPSmr addr:$dst, VR128:$src)>;
1101 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
1102 (MOVUPSmr addr:$dst, VR128:$src)>;
1103 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
1104 (MOVUPSmr addr:$dst, VR128:$src)>;
1107 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
1108 // bits are disregarded. FIXME: Set encoding to pseudo!
1109 let canFoldAsLoad = 1, isReMaterializable = 1, SchedRW = [WriteLoad] in {
1110 let isCodeGenOnly = 1 in {
1111 def FsVMOVAPSrm : VPSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1112 "movaps\t{$src, $dst|$dst, $src}",
1113 [(set FR32:$dst, (alignedloadfsf32 addr:$src))],
1114 IIC_SSE_MOVA_P_RM>, VEX;
1115 def FsVMOVAPDrm : VPDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1116 "movapd\t{$src, $dst|$dst, $src}",
1117 [(set FR64:$dst, (alignedloadfsf64 addr:$src))],
1118 IIC_SSE_MOVA_P_RM>, VEX;
1119 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1120 "movaps\t{$src, $dst|$dst, $src}",
1121 [(set FR32:$dst, (alignedloadfsf32 addr:$src))],
1123 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1124 "movapd\t{$src, $dst|$dst, $src}",
1125 [(set FR64:$dst, (alignedloadfsf64 addr:$src))],
1130 //===----------------------------------------------------------------------===//
1131 // SSE 1 & 2 - Move Low packed FP Instructions
1132 //===----------------------------------------------------------------------===//
1134 multiclass sse12_mov_hilo_packed_base<bits<8>opc, SDNode psnode, SDNode pdnode,
1135 string base_opc, string asm_opr,
1136 InstrItinClass itin> {
1137 def PSrm : PI<opc, MRMSrcMem,
1138 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
1139 !strconcat(base_opc, "s", asm_opr),
1141 (psnode VR128:$src1,
1142 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
1143 itin, SSEPackedSingle>, TB,
1144 Sched<[WriteShuffleLd, ReadAfterLd]>;
1146 def PDrm : PI<opc, MRMSrcMem,
1147 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
1148 !strconcat(base_opc, "d", asm_opr),
1149 [(set VR128:$dst, (v2f64 (pdnode VR128:$src1,
1150 (scalar_to_vector (loadf64 addr:$src2)))))],
1151 itin, SSEPackedDouble>, TB, OpSize,
1152 Sched<[WriteShuffleLd, ReadAfterLd]>;
1156 multiclass sse12_mov_hilo_packed<bits<8>opc, SDNode psnode, SDNode pdnode,
1157 string base_opc, InstrItinClass itin> {
1158 defm V#NAME : sse12_mov_hilo_packed_base<opc, psnode, pdnode, base_opc,
1159 "\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1162 let Constraints = "$src1 = $dst" in
1163 defm NAME : sse12_mov_hilo_packed_base<opc, psnode, pdnode, base_opc,
1164 "\t{$src2, $dst|$dst, $src2}",
1168 let AddedComplexity = 20 in {
1169 defm MOVL : sse12_mov_hilo_packed<0x12, X86Movlps, X86Movlpd, "movlp",
1173 let SchedRW = [WriteStore] in {
1174 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1175 "movlps\t{$src, $dst|$dst, $src}",
1176 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1177 (iPTR 0))), addr:$dst)],
1178 IIC_SSE_MOV_LH>, VEX;
1179 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1180 "movlpd\t{$src, $dst|$dst, $src}",
1181 [(store (f64 (vector_extract (v2f64 VR128:$src),
1182 (iPTR 0))), addr:$dst)],
1183 IIC_SSE_MOV_LH>, VEX;
1184 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1185 "movlps\t{$src, $dst|$dst, $src}",
1186 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1187 (iPTR 0))), addr:$dst)],
1189 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1190 "movlpd\t{$src, $dst|$dst, $src}",
1191 [(store (f64 (vector_extract (v2f64 VR128:$src),
1192 (iPTR 0))), addr:$dst)],
1196 let Predicates = [HasAVX] in {
1197 // Shuffle with VMOVLPS
1198 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1199 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1200 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1201 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1203 // Shuffle with VMOVLPD
1204 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1205 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1206 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1207 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1210 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1212 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1213 def : Pat<(store (v4i32 (X86Movlps
1214 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
1215 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1216 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1218 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1219 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1221 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1224 let Predicates = [UseSSE1] in {
1225 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
1226 def : Pat<(store (i64 (vector_extract (bc_v2i64 (v4f32 VR128:$src2)),
1227 (iPTR 0))), addr:$src1),
1228 (MOVLPSmr addr:$src1, VR128:$src2)>;
1230 // Shuffle with MOVLPS
1231 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1232 (MOVLPSrm VR128:$src1, addr:$src2)>;
1233 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1234 (MOVLPSrm VR128:$src1, addr:$src2)>;
1235 def : Pat<(X86Movlps VR128:$src1,
1236 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1237 (MOVLPSrm VR128:$src1, addr:$src2)>;
1240 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1242 (MOVLPSmr addr:$src1, VR128:$src2)>;
1243 def : Pat<(store (v4i32 (X86Movlps
1244 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
1246 (MOVLPSmr addr:$src1, VR128:$src2)>;
1249 let Predicates = [UseSSE2] in {
1250 // Shuffle with MOVLPD
1251 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1252 (MOVLPDrm VR128:$src1, addr:$src2)>;
1253 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1254 (MOVLPDrm VR128:$src1, addr:$src2)>;
1257 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1259 (MOVLPDmr addr:$src1, VR128:$src2)>;
1260 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1262 (MOVLPDmr addr:$src1, VR128:$src2)>;
1265 //===----------------------------------------------------------------------===//
1266 // SSE 1 & 2 - Move Hi packed FP Instructions
1267 //===----------------------------------------------------------------------===//
1269 let AddedComplexity = 20 in {
1270 defm MOVH : sse12_mov_hilo_packed<0x16, X86Movlhps, X86Movlhpd, "movhp",
1274 let SchedRW = [WriteStore] in {
1275 // v2f64 extract element 1 is always custom lowered to unpack high to low
1276 // and extract element 0 so the non-store version isn't too horrible.
1277 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1278 "movhps\t{$src, $dst|$dst, $src}",
1279 [(store (f64 (vector_extract
1280 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1281 (bc_v2f64 (v4f32 VR128:$src))),
1282 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>, VEX;
1283 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1284 "movhpd\t{$src, $dst|$dst, $src}",
1285 [(store (f64 (vector_extract
1286 (v2f64 (X86Unpckh VR128:$src, VR128:$src)),
1287 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>, VEX;
1288 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1289 "movhps\t{$src, $dst|$dst, $src}",
1290 [(store (f64 (vector_extract
1291 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1292 (bc_v2f64 (v4f32 VR128:$src))),
1293 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>;
1294 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1295 "movhpd\t{$src, $dst|$dst, $src}",
1296 [(store (f64 (vector_extract
1297 (v2f64 (X86Unpckh VR128:$src, VR128:$src)),
1298 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>;
1301 let Predicates = [HasAVX] in {
1303 def : Pat<(X86Movlhps VR128:$src1,
1304 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1305 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1306 def : Pat<(X86Movlhps VR128:$src1,
1307 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1308 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1310 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1311 // is during lowering, where it's not possible to recognize the load fold
1312 // cause it has two uses through a bitcast. One use disappears at isel time
1313 // and the fold opportunity reappears.
1314 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1315 (scalar_to_vector (loadf64 addr:$src2)))),
1316 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1319 let Predicates = [UseSSE1] in {
1321 def : Pat<(X86Movlhps VR128:$src1,
1322 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1323 (MOVHPSrm VR128:$src1, addr:$src2)>;
1324 def : Pat<(X86Movlhps VR128:$src1,
1325 (bc_v4f32 (v2i64 (X86vzload addr:$src2)))),
1326 (MOVHPSrm VR128:$src1, addr:$src2)>;
1329 let Predicates = [UseSSE2] in {
1330 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1331 // is during lowering, where it's not possible to recognize the load fold
1332 // cause it has two uses through a bitcast. One use disappears at isel time
1333 // and the fold opportunity reappears.
1334 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1335 (scalar_to_vector (loadf64 addr:$src2)))),
1336 (MOVHPDrm VR128:$src1, addr:$src2)>;
1339 //===----------------------------------------------------------------------===//
1340 // SSE 1 & 2 - Move Low to High and High to Low packed FP Instructions
1341 //===----------------------------------------------------------------------===//
1343 let AddedComplexity = 20, Predicates = [UseAVX] in {
1344 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
1345 (ins VR128:$src1, VR128:$src2),
1346 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1348 (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))],
1350 VEX_4V, Sched<[WriteShuffle]>;
1351 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
1352 (ins VR128:$src1, VR128:$src2),
1353 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1355 (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))],
1357 VEX_4V, Sched<[WriteShuffle]>;
1359 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1360 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
1361 (ins VR128:$src1, VR128:$src2),
1362 "movlhps\t{$src2, $dst|$dst, $src2}",
1364 (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))],
1365 IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>;
1366 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
1367 (ins VR128:$src1, VR128:$src2),
1368 "movhlps\t{$src2, $dst|$dst, $src2}",
1370 (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))],
1371 IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>;
1374 let Predicates = [UseAVX] in {
1376 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1377 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1378 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1379 (VMOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1382 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1383 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1386 let Predicates = [UseSSE1] in {
1388 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1389 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1390 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1391 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1394 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1395 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1398 //===----------------------------------------------------------------------===//
1399 // SSE 1 & 2 - Conversion Instructions
1400 //===----------------------------------------------------------------------===//
1402 def SSE_CVT_PD : OpndItins<
1403 IIC_SSE_CVT_PD_RR, IIC_SSE_CVT_PD_RM
1406 let Sched = WriteCvtI2F in
1407 def SSE_CVT_PS : OpndItins<
1408 IIC_SSE_CVT_PS_RR, IIC_SSE_CVT_PS_RM
1411 let Sched = WriteCvtI2F in
1412 def SSE_CVT_Scalar : OpndItins<
1413 IIC_SSE_CVT_Scalar_RR, IIC_SSE_CVT_Scalar_RM
1416 let Sched = WriteCvtF2I in
1417 def SSE_CVT_SS2SI_32 : OpndItins<
1418 IIC_SSE_CVT_SS2SI32_RR, IIC_SSE_CVT_SS2SI32_RM
1421 let Sched = WriteCvtF2I in
1422 def SSE_CVT_SS2SI_64 : OpndItins<
1423 IIC_SSE_CVT_SS2SI64_RR, IIC_SSE_CVT_SS2SI64_RM
1426 let Sched = WriteCvtF2I in
1427 def SSE_CVT_SD2SI : OpndItins<
1428 IIC_SSE_CVT_SD2SI_RR, IIC_SSE_CVT_SD2SI_RM
1431 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1432 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1433 string asm, OpndItins itins> {
1434 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1435 [(set DstRC:$dst, (OpNode SrcRC:$src))],
1436 itins.rr>, Sched<[itins.Sched]>;
1437 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1438 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))],
1439 itins.rm>, Sched<[itins.Sched.Folded]>;
1442 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1443 X86MemOperand x86memop, string asm, Domain d,
1445 let neverHasSideEffects = 1 in {
1446 def rr : I<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1447 [], itins.rr, d>, Sched<[itins.Sched]>;
1449 def rm : I<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1450 [], itins.rm, d>, Sched<[itins.Sched.Folded]>;
1454 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1455 X86MemOperand x86memop, string asm> {
1456 let neverHasSideEffects = 1, Predicates = [UseAVX] in {
1457 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
1458 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>,
1459 Sched<[WriteCvtI2F]>;
1461 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1462 (ins DstRC:$src1, x86memop:$src),
1463 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>,
1464 Sched<[WriteCvtI2FLd, ReadAfterLd]>;
1465 } // neverHasSideEffects = 1
1468 let Predicates = [UseAVX] in {
1469 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1470 "cvttss2si\t{$src, $dst|$dst, $src}",
1473 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1474 "cvttss2si\t{$src, $dst|$dst, $src}",
1476 XS, VEX, VEX_W, VEX_LIG;
1477 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1478 "cvttsd2si\t{$src, $dst|$dst, $src}",
1481 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1482 "cvttsd2si\t{$src, $dst|$dst, $src}",
1484 XD, VEX, VEX_W, VEX_LIG;
1486 def : InstAlias<"vcvttss2si{l}\t{$src, $dst|$dst, $src}",
1487 (VCVTTSS2SIrr GR32:$dst, FR32:$src), 0>;
1488 def : InstAlias<"vcvttss2si{l}\t{$src, $dst|$dst, $src}",
1489 (VCVTTSS2SIrm GR32:$dst, f32mem:$src), 0>;
1490 def : InstAlias<"vcvttsd2si{l}\t{$src, $dst|$dst, $src}",
1491 (VCVTTSD2SIrr GR32:$dst, FR64:$src), 0>;
1492 def : InstAlias<"vcvttsd2si{l}\t{$src, $dst|$dst, $src}",
1493 (VCVTTSD2SIrm GR32:$dst, f64mem:$src), 0>;
1494 def : InstAlias<"vcvttss2si{q}\t{$src, $dst|$dst, $src}",
1495 (VCVTTSS2SI64rr GR64:$dst, FR32:$src), 0>;
1496 def : InstAlias<"vcvttss2si{q}\t{$src, $dst|$dst, $src}",
1497 (VCVTTSS2SI64rm GR64:$dst, f32mem:$src), 0>;
1498 def : InstAlias<"vcvttsd2si{q}\t{$src, $dst|$dst, $src}",
1499 (VCVTTSD2SI64rr GR64:$dst, FR64:$src), 0>;
1500 def : InstAlias<"vcvttsd2si{q}\t{$src, $dst|$dst, $src}",
1501 (VCVTTSD2SI64rm GR64:$dst, f64mem:$src), 0>;
1503 // The assembler can recognize rr 64-bit instructions by seeing a rxx
1504 // register, but the same isn't true when only using memory operands,
1505 // provide other assembly "l" and "q" forms to address this explicitly
1506 // where appropriate to do so.
1507 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss{l}">,
1508 XS, VEX_4V, VEX_LIG;
1509 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">,
1510 XS, VEX_4V, VEX_W, VEX_LIG;
1511 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">,
1512 XD, VEX_4V, VEX_LIG;
1513 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">,
1514 XD, VEX_4V, VEX_W, VEX_LIG;
1516 let Predicates = [UseAVX] in {
1517 def : InstAlias<"vcvtsi2ss\t{$src, $src1, $dst|$dst, $src1, $src}",
1518 (VCVTSI2SSrm FR64:$dst, FR64:$src1, i32mem:$src)>;
1519 def : InstAlias<"vcvtsi2sd\t{$src, $src1, $dst|$dst, $src1, $src}",
1520 (VCVTSI2SDrm FR64:$dst, FR64:$src1, i32mem:$src)>;
1522 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
1523 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1524 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
1525 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
1526 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
1527 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
1528 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
1529 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
1531 def : Pat<(f32 (sint_to_fp GR32:$src)),
1532 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
1533 def : Pat<(f32 (sint_to_fp GR64:$src)),
1534 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
1535 def : Pat<(f64 (sint_to_fp GR32:$src)),
1536 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
1537 def : Pat<(f64 (sint_to_fp GR64:$src)),
1538 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
1541 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1542 "cvttss2si\t{$src, $dst|$dst, $src}",
1543 SSE_CVT_SS2SI_32>, XS;
1544 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1545 "cvttss2si\t{$src, $dst|$dst, $src}",
1546 SSE_CVT_SS2SI_64>, XS, REX_W;
1547 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1548 "cvttsd2si\t{$src, $dst|$dst, $src}",
1550 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1551 "cvttsd2si\t{$src, $dst|$dst, $src}",
1552 SSE_CVT_SD2SI>, XD, REX_W;
1553 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
1554 "cvtsi2ss{l}\t{$src, $dst|$dst, $src}",
1555 SSE_CVT_Scalar>, XS;
1556 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
1557 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}",
1558 SSE_CVT_Scalar>, XS, REX_W;
1559 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
1560 "cvtsi2sd{l}\t{$src, $dst|$dst, $src}",
1561 SSE_CVT_Scalar>, XD;
1562 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
1563 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}",
1564 SSE_CVT_Scalar>, XD, REX_W;
1566 def : InstAlias<"cvttss2si{l}\t{$src, $dst|$dst, $src}",
1567 (CVTTSS2SIrr GR32:$dst, FR32:$src), 0>;
1568 def : InstAlias<"cvttss2si{l}\t{$src, $dst|$dst, $src}",
1569 (CVTTSS2SIrm GR32:$dst, f32mem:$src), 0>;
1570 def : InstAlias<"cvttsd2si{l}\t{$src, $dst|$dst, $src}",
1571 (CVTTSD2SIrr GR32:$dst, FR64:$src), 0>;
1572 def : InstAlias<"cvttsd2si{l}\t{$src, $dst|$dst, $src}",
1573 (CVTTSD2SIrm GR32:$dst, f64mem:$src), 0>;
1574 def : InstAlias<"cvttss2si{q}\t{$src, $dst|$dst, $src}",
1575 (CVTTSS2SI64rr GR64:$dst, FR32:$src), 0>;
1576 def : InstAlias<"cvttss2si{q}\t{$src, $dst|$dst, $src}",
1577 (CVTTSS2SI64rm GR64:$dst, f32mem:$src), 0>;
1578 def : InstAlias<"cvttsd2si{q}\t{$src, $dst|$dst, $src}",
1579 (CVTTSD2SI64rr GR64:$dst, FR64:$src), 0>;
1580 def : InstAlias<"cvttsd2si{q}\t{$src, $dst|$dst, $src}",
1581 (CVTTSD2SI64rm GR64:$dst, f64mem:$src), 0>;
1583 def : InstAlias<"cvtsi2ss\t{$src, $dst|$dst, $src}",
1584 (CVTSI2SSrm FR64:$dst, i32mem:$src)>;
1585 def : InstAlias<"cvtsi2sd\t{$src, $dst|$dst, $src}",
1586 (CVTSI2SDrm FR64:$dst, i32mem:$src)>;
1588 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
1589 // and/or XMM operand(s).
1591 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1592 Intrinsic Int, Operand memop, ComplexPattern mem_cpat,
1593 string asm, OpndItins itins> {
1594 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
1595 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1596 [(set DstRC:$dst, (Int SrcRC:$src))], itins.rr>,
1597 Sched<[itins.Sched]>;
1598 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins memop:$src),
1599 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1600 [(set DstRC:$dst, (Int mem_cpat:$src))], itins.rm>,
1601 Sched<[itins.Sched.Folded]>;
1604 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
1605 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
1606 PatFrag ld_frag, string asm, OpndItins itins,
1608 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
1610 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1611 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1612 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))],
1613 itins.rr>, Sched<[itins.Sched]>;
1614 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1615 (ins DstRC:$src1, x86memop:$src2),
1617 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1618 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1619 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))],
1620 itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
1623 let Predicates = [UseAVX] in {
1624 defm VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32,
1625 int_x86_sse2_cvtsd2si, sdmem, sse_load_f64, "cvtsd2si",
1626 SSE_CVT_SD2SI>, XD, VEX, VEX_LIG;
1627 defm VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
1628 int_x86_sse2_cvtsd2si64, sdmem, sse_load_f64, "cvtsd2si",
1629 SSE_CVT_SD2SI>, XD, VEX, VEX_W, VEX_LIG;
1631 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1632 sdmem, sse_load_f64, "cvtsd2si", SSE_CVT_SD2SI>, XD;
1633 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
1634 sdmem, sse_load_f64, "cvtsd2si", SSE_CVT_SD2SI>, XD, REX_W;
1637 let isCodeGenOnly = 1 in {
1638 let Predicates = [UseAVX] in {
1639 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1640 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss{l}",
1641 SSE_CVT_Scalar, 0>, XS, VEX_4V;
1642 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1643 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss{q}",
1644 SSE_CVT_Scalar, 0>, XS, VEX_4V,
1646 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1647 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd{l}",
1648 SSE_CVT_Scalar, 0>, XD, VEX_4V;
1649 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1650 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd{q}",
1651 SSE_CVT_Scalar, 0>, XD,
1654 let Constraints = "$src1 = $dst" in {
1655 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1656 int_x86_sse_cvtsi2ss, i32mem, loadi32,
1657 "cvtsi2ss{l}", SSE_CVT_Scalar>, XS;
1658 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1659 int_x86_sse_cvtsi642ss, i64mem, loadi64,
1660 "cvtsi2ss{q}", SSE_CVT_Scalar>, XS, REX_W;
1661 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1662 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
1663 "cvtsi2sd{l}", SSE_CVT_Scalar>, XD;
1664 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1665 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
1666 "cvtsi2sd{q}", SSE_CVT_Scalar>, XD, REX_W;
1668 } // isCodeGenOnly = 1
1672 // Aliases for intrinsics
1673 let isCodeGenOnly = 1 in {
1674 let Predicates = [UseAVX] in {
1675 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1676 ssmem, sse_load_f32, "cvttss2si",
1677 SSE_CVT_SS2SI_32>, XS, VEX;
1678 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1679 int_x86_sse_cvttss2si64, ssmem, sse_load_f32,
1680 "cvttss2si", SSE_CVT_SS2SI_64>,
1682 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1683 sdmem, sse_load_f64, "cvttsd2si",
1684 SSE_CVT_SD2SI>, XD, VEX;
1685 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1686 int_x86_sse2_cvttsd2si64, sdmem, sse_load_f64,
1687 "cvttsd2si", SSE_CVT_SD2SI>,
1690 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1691 ssmem, sse_load_f32, "cvttss2si",
1692 SSE_CVT_SS2SI_32>, XS;
1693 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1694 int_x86_sse_cvttss2si64, ssmem, sse_load_f32,
1695 "cvttss2si", SSE_CVT_SS2SI_64>, XS, REX_W;
1696 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1697 sdmem, sse_load_f64, "cvttsd2si",
1699 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1700 int_x86_sse2_cvttsd2si64, sdmem, sse_load_f64,
1701 "cvttsd2si", SSE_CVT_SD2SI>, XD, REX_W;
1702 } // isCodeGenOnly = 1
1704 let Predicates = [UseAVX] in {
1705 defm VCVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
1706 ssmem, sse_load_f32, "cvtss2si",
1707 SSE_CVT_SS2SI_32>, XS, VEX, VEX_LIG;
1708 defm VCVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
1709 ssmem, sse_load_f32, "cvtss2si",
1710 SSE_CVT_SS2SI_64>, XS, VEX, VEX_W, VEX_LIG;
1712 defm CVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
1713 ssmem, sse_load_f32, "cvtss2si",
1714 SSE_CVT_SS2SI_32>, XS;
1715 defm CVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
1716 ssmem, sse_load_f32, "cvtss2si",
1717 SSE_CVT_SS2SI_64>, XS, REX_W;
1719 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, i128mem,
1720 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1721 SSEPackedSingle, SSE_CVT_PS>,
1722 TB, VEX, Requires<[HasAVX]>;
1723 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, i256mem,
1724 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1725 SSEPackedSingle, SSE_CVT_PS>,
1726 TB, VEX, VEX_L, Requires<[HasAVX]>;
1728 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, i128mem,
1729 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1730 SSEPackedSingle, SSE_CVT_PS>,
1731 TB, Requires<[UseSSE2]>;
1733 let Predicates = [UseAVX] in {
1734 def : InstAlias<"vcvtss2si{l}\t{$src, $dst|$dst, $src}",
1735 (VCVTSS2SIrr GR32:$dst, VR128:$src), 0>;
1736 def : InstAlias<"vcvtss2si{l}\t{$src, $dst|$dst, $src}",
1737 (VCVTSS2SIrm GR32:$dst, ssmem:$src), 0>;
1738 def : InstAlias<"vcvtsd2si{l}\t{$src, $dst|$dst, $src}",
1739 (VCVTSD2SIrr GR32:$dst, VR128:$src), 0>;
1740 def : InstAlias<"vcvtsd2si{l}\t{$src, $dst|$dst, $src}",
1741 (VCVTSD2SIrm GR32:$dst, sdmem:$src), 0>;
1742 def : InstAlias<"vcvtss2si{q}\t{$src, $dst|$dst, $src}",
1743 (VCVTSS2SI64rr GR64:$dst, VR128:$src), 0>;
1744 def : InstAlias<"vcvtss2si{q}\t{$src, $dst|$dst, $src}",
1745 (VCVTSS2SI64rm GR64:$dst, ssmem:$src), 0>;
1746 def : InstAlias<"vcvtsd2si{q}\t{$src, $dst|$dst, $src}",
1747 (VCVTSD2SI64rr GR64:$dst, VR128:$src), 0>;
1748 def : InstAlias<"vcvtsd2si{q}\t{$src, $dst|$dst, $src}",
1749 (VCVTSD2SI64rm GR64:$dst, sdmem:$src), 0>;
1752 def : InstAlias<"cvtss2si{l}\t{$src, $dst|$dst, $src}",
1753 (CVTSS2SIrr GR32:$dst, VR128:$src), 0>;
1754 def : InstAlias<"cvtss2si{l}\t{$src, $dst|$dst, $src}",
1755 (CVTSS2SIrm GR32:$dst, ssmem:$src), 0>;
1756 def : InstAlias<"cvtsd2si{l}\t{$src, $dst|$dst, $src}",
1757 (CVTSD2SIrr GR32:$dst, VR128:$src), 0>;
1758 def : InstAlias<"cvtsd2si{l}\t{$src, $dst|$dst, $src}",
1759 (CVTSD2SIrm GR32:$dst, sdmem:$src), 0>;
1760 def : InstAlias<"cvtss2si{q}\t{$src, $dst|$dst, $src}",
1761 (CVTSS2SI64rr GR64:$dst, VR128:$src), 0>;
1762 def : InstAlias<"cvtss2si{q}\t{$src, $dst|$dst, $src}",
1763 (CVTSS2SI64rm GR64:$dst, ssmem:$src), 0>;
1764 def : InstAlias<"cvtsd2si{q}\t{$src, $dst|$dst, $src}",
1765 (CVTSD2SI64rr GR64:$dst, VR128:$src), 0>;
1766 def : InstAlias<"cvtsd2si{q}\t{$src, $dst|$dst, $src}",
1767 (CVTSD2SI64rm GR64:$dst, sdmem:$src)>;
1771 // Convert scalar double to scalar single
1772 let neverHasSideEffects = 1, Predicates = [UseAVX] in {
1773 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
1774 (ins FR64:$src1, FR64:$src2),
1775 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
1776 IIC_SSE_CVT_Scalar_RR>, VEX_4V, VEX_LIG,
1777 Sched<[WriteCvtF2F]>;
1779 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
1780 (ins FR64:$src1, f64mem:$src2),
1781 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1782 [], IIC_SSE_CVT_Scalar_RM>,
1783 XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG,
1784 Sched<[WriteCvtF2FLd, ReadAfterLd]>;
1787 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
1790 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
1791 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1792 [(set FR32:$dst, (fround FR64:$src))],
1793 IIC_SSE_CVT_Scalar_RR>, Sched<[WriteCvtF2F]>;
1794 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
1795 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1796 [(set FR32:$dst, (fround (loadf64 addr:$src)))],
1797 IIC_SSE_CVT_Scalar_RM>,
1799 Requires<[UseSSE2, OptForSize]>, Sched<[WriteCvtF2FLd]>;
1801 let isCodeGenOnly = 1 in {
1802 def Int_VCVTSD2SSrr: I<0x5A, MRMSrcReg,
1803 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1804 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1806 (int_x86_sse2_cvtsd2ss VR128:$src1, VR128:$src2))],
1807 IIC_SSE_CVT_Scalar_RR>, XD, VEX_4V, Requires<[UseAVX]>,
1808 Sched<[WriteCvtF2F]>;
1809 def Int_VCVTSD2SSrm: I<0x5A, MRMSrcReg,
1810 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2),
1811 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1812 [(set VR128:$dst, (int_x86_sse2_cvtsd2ss
1813 VR128:$src1, sse_load_f64:$src2))],
1814 IIC_SSE_CVT_Scalar_RM>, XD, VEX_4V, Requires<[UseAVX]>,
1815 Sched<[WriteCvtF2FLd, ReadAfterLd]>;
1817 let Constraints = "$src1 = $dst" in {
1818 def Int_CVTSD2SSrr: I<0x5A, MRMSrcReg,
1819 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1820 "cvtsd2ss\t{$src2, $dst|$dst, $src2}",
1822 (int_x86_sse2_cvtsd2ss VR128:$src1, VR128:$src2))],
1823 IIC_SSE_CVT_Scalar_RR>, XD, Requires<[UseSSE2]>,
1824 Sched<[WriteCvtF2F]>;
1825 def Int_CVTSD2SSrm: I<0x5A, MRMSrcReg,
1826 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2),
1827 "cvtsd2ss\t{$src2, $dst|$dst, $src2}",
1828 [(set VR128:$dst, (int_x86_sse2_cvtsd2ss
1829 VR128:$src1, sse_load_f64:$src2))],
1830 IIC_SSE_CVT_Scalar_RM>, XD, Requires<[UseSSE2]>,
1831 Sched<[WriteCvtF2FLd, ReadAfterLd]>;
1833 } // isCodeGenOnly = 1
1835 // Convert scalar single to scalar double
1836 // SSE2 instructions with XS prefix
1837 let neverHasSideEffects = 1, Predicates = [UseAVX] in {
1838 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
1839 (ins FR32:$src1, FR32:$src2),
1840 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1841 [], IIC_SSE_CVT_Scalar_RR>,
1842 XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG,
1843 Sched<[WriteCvtF2F]>;
1845 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
1846 (ins FR32:$src1, f32mem:$src2),
1847 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1848 [], IIC_SSE_CVT_Scalar_RM>,
1849 XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>,
1850 Sched<[WriteCvtF2FLd, ReadAfterLd]>;
1853 def : Pat<(f64 (fextend FR32:$src)),
1854 (VCVTSS2SDrr FR32:$src, FR32:$src)>, Requires<[UseAVX]>;
1855 def : Pat<(fextend (loadf32 addr:$src)),
1856 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>, Requires<[UseAVX]>;
1858 def : Pat<(extloadf32 addr:$src),
1859 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>,
1860 Requires<[UseAVX, OptForSize]>;
1861 def : Pat<(extloadf32 addr:$src),
1862 (VCVTSS2SDrr (f32 (IMPLICIT_DEF)), (VMOVSSrm addr:$src))>,
1863 Requires<[UseAVX, OptForSpeed]>;
1865 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
1866 "cvtss2sd\t{$src, $dst|$dst, $src}",
1867 [(set FR64:$dst, (fextend FR32:$src))],
1868 IIC_SSE_CVT_Scalar_RR>, XS,
1869 Requires<[UseSSE2]>, Sched<[WriteCvtF2F]>;
1870 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
1871 "cvtss2sd\t{$src, $dst|$dst, $src}",
1872 [(set FR64:$dst, (extloadf32 addr:$src))],
1873 IIC_SSE_CVT_Scalar_RM>, XS,
1874 Requires<[UseSSE2, OptForSize]>, Sched<[WriteCvtF2FLd]>;
1876 // extload f32 -> f64. This matches load+fextend because we have a hack in
1877 // the isel (PreprocessForFPConvert) that can introduce loads after dag
1879 // Since these loads aren't folded into the fextend, we have to match it
1881 def : Pat<(fextend (loadf32 addr:$src)),
1882 (CVTSS2SDrm addr:$src)>, Requires<[UseSSE2]>;
1883 def : Pat<(extloadf32 addr:$src),
1884 (CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[UseSSE2, OptForSpeed]>;
1886 let isCodeGenOnly = 1 in {
1887 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
1888 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1889 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1891 (int_x86_sse2_cvtss2sd VR128:$src1, VR128:$src2))],
1892 IIC_SSE_CVT_Scalar_RR>, XS, VEX_4V, Requires<[UseAVX]>,
1893 Sched<[WriteCvtF2F]>;
1894 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
1895 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2),
1896 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1898 (int_x86_sse2_cvtss2sd VR128:$src1, sse_load_f32:$src2))],
1899 IIC_SSE_CVT_Scalar_RM>, XS, VEX_4V, Requires<[UseAVX]>,
1900 Sched<[WriteCvtF2FLd, ReadAfterLd]>;
1901 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
1902 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
1903 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1904 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1906 (int_x86_sse2_cvtss2sd VR128:$src1, VR128:$src2))],
1907 IIC_SSE_CVT_Scalar_RR>, XS, Requires<[UseSSE2]>,
1908 Sched<[WriteCvtF2F]>;
1909 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
1910 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2),
1911 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1913 (int_x86_sse2_cvtss2sd VR128:$src1, sse_load_f32:$src2))],
1914 IIC_SSE_CVT_Scalar_RM>, XS, Requires<[UseSSE2]>,
1915 Sched<[WriteCvtF2FLd, ReadAfterLd]>;
1917 } // isCodeGenOnly = 1
1919 // Convert packed single/double fp to doubleword
1920 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1921 "cvtps2dq\t{$src, $dst|$dst, $src}",
1922 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))],
1923 IIC_SSE_CVT_PS_RR>, VEX, Sched<[WriteCvtF2I]>;
1924 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1925 "cvtps2dq\t{$src, $dst|$dst, $src}",
1927 (int_x86_sse2_cvtps2dq (loadv4f32 addr:$src)))],
1928 IIC_SSE_CVT_PS_RM>, VEX, Sched<[WriteCvtF2ILd]>;
1929 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1930 "cvtps2dq\t{$src, $dst|$dst, $src}",
1932 (int_x86_avx_cvt_ps2dq_256 VR256:$src))],
1933 IIC_SSE_CVT_PS_RR>, VEX, VEX_L, Sched<[WriteCvtF2I]>;
1934 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1935 "cvtps2dq\t{$src, $dst|$dst, $src}",
1937 (int_x86_avx_cvt_ps2dq_256 (loadv8f32 addr:$src)))],
1938 IIC_SSE_CVT_PS_RM>, VEX, VEX_L, Sched<[WriteCvtF2ILd]>;
1939 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1940 "cvtps2dq\t{$src, $dst|$dst, $src}",
1941 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))],
1942 IIC_SSE_CVT_PS_RR>, Sched<[WriteCvtF2I]>;
1943 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1944 "cvtps2dq\t{$src, $dst|$dst, $src}",
1946 (int_x86_sse2_cvtps2dq (memopv4f32 addr:$src)))],
1947 IIC_SSE_CVT_PS_RM>, Sched<[WriteCvtF2ILd]>;
1950 // Convert Packed Double FP to Packed DW Integers
1951 let Predicates = [HasAVX] in {
1952 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1953 // register, but the same isn't true when using memory operands instead.
1954 // Provide other assembly rr and rm forms to address this explicitly.
1955 def VCVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1956 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1957 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1958 VEX, Sched<[WriteCvtF2I]>;
1961 def : InstAlias<"vcvtpd2dqx\t{$src, $dst|$dst, $src}",
1962 (VCVTPD2DQrr VR128:$dst, VR128:$src)>;
1963 def VCVTPD2DQXrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1964 "vcvtpd2dqx\t{$src, $dst|$dst, $src}",
1966 (int_x86_sse2_cvtpd2dq (loadv2f64 addr:$src)))]>, VEX,
1967 Sched<[WriteCvtF2ILd]>;
1970 def VCVTPD2DQYrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1971 "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}",
1973 (int_x86_avx_cvt_pd2dq_256 VR256:$src))]>, VEX, VEX_L,
1974 Sched<[WriteCvtF2I]>;
1975 def VCVTPD2DQYrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1976 "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}",
1978 (int_x86_avx_cvt_pd2dq_256 (loadv4f64 addr:$src)))]>,
1979 VEX, VEX_L, Sched<[WriteCvtF2ILd]>;
1980 def : InstAlias<"vcvtpd2dq\t{$src, $dst|$dst, $src}",
1981 (VCVTPD2DQYrr VR128:$dst, VR256:$src)>;
1984 def CVTPD2DQrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1985 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1987 (int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)))],
1988 IIC_SSE_CVT_PD_RM>, Sched<[WriteCvtF2ILd]>;
1989 def CVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1990 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1991 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))],
1992 IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtF2I]>;
1994 // Convert with truncation packed single/double fp to doubleword
1995 // SSE2 packed instructions with XS prefix
1996 def VCVTTPS2DQrr : VS2SI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1997 "cvttps2dq\t{$src, $dst|$dst, $src}",
1999 (int_x86_sse2_cvttps2dq VR128:$src))],
2000 IIC_SSE_CVT_PS_RR>, VEX, Sched<[WriteCvtF2I]>;
2001 def VCVTTPS2DQrm : VS2SI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2002 "cvttps2dq\t{$src, $dst|$dst, $src}",
2003 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
2004 (loadv4f32 addr:$src)))],
2005 IIC_SSE_CVT_PS_RM>, VEX, Sched<[WriteCvtF2ILd]>;
2006 def VCVTTPS2DQYrr : VS2SI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2007 "cvttps2dq\t{$src, $dst|$dst, $src}",
2009 (int_x86_avx_cvtt_ps2dq_256 VR256:$src))],
2010 IIC_SSE_CVT_PS_RR>, VEX, VEX_L, Sched<[WriteCvtF2I]>;
2011 def VCVTTPS2DQYrm : VS2SI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2012 "cvttps2dq\t{$src, $dst|$dst, $src}",
2013 [(set VR256:$dst, (int_x86_avx_cvtt_ps2dq_256
2014 (loadv8f32 addr:$src)))],
2015 IIC_SSE_CVT_PS_RM>, VEX, VEX_L,
2016 Sched<[WriteCvtF2ILd]>;
2018 def CVTTPS2DQrr : S2SI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2019 "cvttps2dq\t{$src, $dst|$dst, $src}",
2020 [(set VR128:$dst, (int_x86_sse2_cvttps2dq VR128:$src))],
2021 IIC_SSE_CVT_PS_RR>, Sched<[WriteCvtF2I]>;
2022 def CVTTPS2DQrm : S2SI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2023 "cvttps2dq\t{$src, $dst|$dst, $src}",
2025 (int_x86_sse2_cvttps2dq (memopv4f32 addr:$src)))],
2026 IIC_SSE_CVT_PS_RM>, Sched<[WriteCvtF2ILd]>;
2028 let Predicates = [HasAVX] in {
2029 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
2030 (VCVTDQ2PSrr VR128:$src)>;
2031 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (loadv2i64 addr:$src)))),
2032 (VCVTDQ2PSrm addr:$src)>;
2034 def : Pat<(int_x86_sse2_cvtdq2ps VR128:$src),
2035 (VCVTDQ2PSrr VR128:$src)>;
2036 def : Pat<(int_x86_sse2_cvtdq2ps (bc_v4i32 (loadv2i64 addr:$src))),
2037 (VCVTDQ2PSrm addr:$src)>;
2039 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
2040 (VCVTTPS2DQrr VR128:$src)>;
2041 def : Pat<(v4i32 (fp_to_sint (loadv4f32 addr:$src))),
2042 (VCVTTPS2DQrm addr:$src)>;
2044 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
2045 (VCVTDQ2PSYrr VR256:$src)>;
2046 def : Pat<(v8f32 (sint_to_fp (bc_v8i32 (loadv4i64 addr:$src)))),
2047 (VCVTDQ2PSYrm addr:$src)>;
2049 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
2050 (VCVTTPS2DQYrr VR256:$src)>;
2051 def : Pat<(v8i32 (fp_to_sint (loadv8f32 addr:$src))),
2052 (VCVTTPS2DQYrm addr:$src)>;
2055 let Predicates = [UseSSE2] in {
2056 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
2057 (CVTDQ2PSrr VR128:$src)>;
2058 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
2059 (CVTDQ2PSrm addr:$src)>;
2061 def : Pat<(int_x86_sse2_cvtdq2ps VR128:$src),
2062 (CVTDQ2PSrr VR128:$src)>;
2063 def : Pat<(int_x86_sse2_cvtdq2ps (bc_v4i32 (memopv2i64 addr:$src))),
2064 (CVTDQ2PSrm addr:$src)>;
2066 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
2067 (CVTTPS2DQrr VR128:$src)>;
2068 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
2069 (CVTTPS2DQrm addr:$src)>;
2072 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2073 "cvttpd2dq\t{$src, $dst|$dst, $src}",
2075 (int_x86_sse2_cvttpd2dq VR128:$src))],
2076 IIC_SSE_CVT_PD_RR>, VEX, Sched<[WriteCvtF2I]>;
2078 // The assembler can recognize rr 256-bit instructions by seeing a ymm
2079 // register, but the same isn't true when using memory operands instead.
2080 // Provide other assembly rr and rm forms to address this explicitly.
2083 def : InstAlias<"vcvttpd2dqx\t{$src, $dst|$dst, $src}",
2084 (VCVTTPD2DQrr VR128:$dst, VR128:$src)>;
2085 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2086 "cvttpd2dqx\t{$src, $dst|$dst, $src}",
2087 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
2088 (loadv2f64 addr:$src)))],
2089 IIC_SSE_CVT_PD_RM>, VEX, Sched<[WriteCvtF2ILd]>;
2092 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
2093 "cvttpd2dq{y}\t{$src, $dst|$dst, $src}",
2095 (int_x86_avx_cvtt_pd2dq_256 VR256:$src))],
2096 IIC_SSE_CVT_PD_RR>, VEX, VEX_L, Sched<[WriteCvtF2I]>;
2097 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
2098 "cvttpd2dq{y}\t{$src, $dst|$dst, $src}",
2100 (int_x86_avx_cvtt_pd2dq_256 (loadv4f64 addr:$src)))],
2101 IIC_SSE_CVT_PD_RM>, VEX, VEX_L, Sched<[WriteCvtF2ILd]>;
2102 def : InstAlias<"vcvttpd2dq\t{$src, $dst|$dst, $src}",
2103 (VCVTTPD2DQYrr VR128:$dst, VR256:$src)>;
2105 let Predicates = [HasAVX] in {
2106 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
2107 (VCVTTPD2DQYrr VR256:$src)>;
2108 def : Pat<(v4i32 (fp_to_sint (loadv4f64 addr:$src))),
2109 (VCVTTPD2DQYrm addr:$src)>;
2110 } // Predicates = [HasAVX]
2112 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2113 "cvttpd2dq\t{$src, $dst|$dst, $src}",
2114 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))],
2115 IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtF2I]>;
2116 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
2117 "cvttpd2dq\t{$src, $dst|$dst, $src}",
2118 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
2119 (memopv2f64 addr:$src)))],
2121 Sched<[WriteCvtF2ILd]>;
2123 // Convert packed single to packed double
2124 let Predicates = [HasAVX] in {
2125 // SSE2 instructions without OpSize prefix
2126 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2127 "vcvtps2pd\t{$src, $dst|$dst, $src}",
2128 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))],
2129 IIC_SSE_CVT_PD_RR>, TB, VEX, Sched<[WriteCvtF2F]>;
2130 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
2131 "vcvtps2pd\t{$src, $dst|$dst, $src}",
2132 [(set VR128:$dst, (v2f64 (extloadv2f32 addr:$src)))],
2133 IIC_SSE_CVT_PD_RM>, TB, VEX, Sched<[WriteCvtF2FLd]>;
2134 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
2135 "vcvtps2pd\t{$src, $dst|$dst, $src}",
2137 (int_x86_avx_cvt_ps2_pd_256 VR128:$src))],
2138 IIC_SSE_CVT_PD_RR>, TB, VEX, VEX_L, Sched<[WriteCvtF2F]>;
2139 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
2140 "vcvtps2pd\t{$src, $dst|$dst, $src}",
2142 (int_x86_avx_cvt_ps2_pd_256 (loadv4f32 addr:$src)))],
2143 IIC_SSE_CVT_PD_RM>, TB, VEX, VEX_L, Sched<[WriteCvtF2FLd]>;
2146 let Predicates = [UseSSE2] in {
2147 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2148 "cvtps2pd\t{$src, $dst|$dst, $src}",
2149 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))],
2150 IIC_SSE_CVT_PD_RR>, TB, Sched<[WriteCvtF2F]>;
2151 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
2152 "cvtps2pd\t{$src, $dst|$dst, $src}",
2153 [(set VR128:$dst, (v2f64 (extloadv2f32 addr:$src)))],
2154 IIC_SSE_CVT_PD_RM>, TB, Sched<[WriteCvtF2FLd]>;
2157 // Convert Packed DW Integers to Packed Double FP
2158 let Predicates = [HasAVX] in {
2159 let neverHasSideEffects = 1, mayLoad = 1 in
2160 def VCVTDQ2PDrm : S2SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2161 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2162 []>, VEX, Sched<[WriteCvtI2FLd]>;
2163 def VCVTDQ2PDrr : S2SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2164 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2166 (int_x86_sse2_cvtdq2pd VR128:$src))]>, VEX,
2167 Sched<[WriteCvtI2F]>;
2168 def VCVTDQ2PDYrm : S2SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
2169 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2171 (int_x86_avx_cvtdq2_pd_256
2172 (bitconvert (loadv2i64 addr:$src))))]>, VEX, VEX_L,
2173 Sched<[WriteCvtI2FLd]>;
2174 def VCVTDQ2PDYrr : S2SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
2175 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2177 (int_x86_avx_cvtdq2_pd_256 VR128:$src))]>, VEX, VEX_L,
2178 Sched<[WriteCvtI2F]>;
2181 let neverHasSideEffects = 1, mayLoad = 1 in
2182 def CVTDQ2PDrm : S2SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2183 "cvtdq2pd\t{$src, $dst|$dst, $src}", [],
2184 IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtI2FLd]>;
2185 def CVTDQ2PDrr : S2SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2186 "cvtdq2pd\t{$src, $dst|$dst, $src}",
2187 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))],
2188 IIC_SSE_CVT_PD_RM>, Sched<[WriteCvtI2F]>;
2190 // AVX 256-bit register conversion intrinsics
2191 let Predicates = [HasAVX] in {
2192 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
2193 (VCVTDQ2PDYrr VR128:$src)>;
2194 def : Pat<(v4f64 (sint_to_fp (bc_v4i32 (loadv2i64 addr:$src)))),
2195 (VCVTDQ2PDYrm addr:$src)>;
2196 } // Predicates = [HasAVX]
2198 // Convert packed double to packed single
2199 // The assembler can recognize rr 256-bit instructions by seeing a ymm
2200 // register, but the same isn't true when using memory operands instead.
2201 // Provide other assembly rr and rm forms to address this explicitly.
2202 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2203 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2204 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))],
2205 IIC_SSE_CVT_PD_RR>, VEX, Sched<[WriteCvtF2F]>;
2208 def : InstAlias<"vcvtpd2psx\t{$src, $dst|$dst, $src}",
2209 (VCVTPD2PSrr VR128:$dst, VR128:$src)>;
2210 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2211 "cvtpd2psx\t{$src, $dst|$dst, $src}",
2213 (int_x86_sse2_cvtpd2ps (loadv2f64 addr:$src)))],
2214 IIC_SSE_CVT_PD_RM>, VEX, Sched<[WriteCvtF2FLd]>;
2217 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
2218 "cvtpd2ps{y}\t{$src, $dst|$dst, $src}",
2220 (int_x86_avx_cvt_pd2_ps_256 VR256:$src))],
2221 IIC_SSE_CVT_PD_RR>, VEX, VEX_L, Sched<[WriteCvtF2F]>;
2222 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
2223 "cvtpd2ps{y}\t{$src, $dst|$dst, $src}",
2225 (int_x86_avx_cvt_pd2_ps_256 (loadv4f64 addr:$src)))],
2226 IIC_SSE_CVT_PD_RM>, VEX, VEX_L, Sched<[WriteCvtF2FLd]>;
2227 def : InstAlias<"vcvtpd2ps\t{$src, $dst|$dst, $src}",
2228 (VCVTPD2PSYrr VR128:$dst, VR256:$src)>;
2230 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2231 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2232 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))],
2233 IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtF2F]>;
2234 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2235 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2237 (int_x86_sse2_cvtpd2ps (memopv2f64 addr:$src)))],
2238 IIC_SSE_CVT_PD_RM>, Sched<[WriteCvtF2FLd]>;
2241 // AVX 256-bit register conversion intrinsics
2242 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
2243 // whenever possible to avoid declaring two versions of each one.
2244 let Predicates = [HasAVX] in {
2245 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
2246 (VCVTDQ2PSYrr VR256:$src)>;
2247 def : Pat<(int_x86_avx_cvtdq2_ps_256 (bitconvert (loadv4i64 addr:$src))),
2248 (VCVTDQ2PSYrm addr:$src)>;
2250 // Match fround and fextend for 128/256-bit conversions
2251 def : Pat<(v4f32 (X86vfpround (v2f64 VR128:$src))),
2252 (VCVTPD2PSrr VR128:$src)>;
2253 def : Pat<(v4f32 (X86vfpround (loadv2f64 addr:$src))),
2254 (VCVTPD2PSXrm addr:$src)>;
2255 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
2256 (VCVTPD2PSYrr VR256:$src)>;
2257 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
2258 (VCVTPD2PSYrm addr:$src)>;
2260 def : Pat<(v2f64 (X86vfpext (v4f32 VR128:$src))),
2261 (VCVTPS2PDrr VR128:$src)>;
2262 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
2263 (VCVTPS2PDYrr VR128:$src)>;
2264 def : Pat<(v4f64 (extloadv4f32 addr:$src)),
2265 (VCVTPS2PDYrm addr:$src)>;
2268 let Predicates = [UseSSE2] in {
2269 // Match fround and fextend for 128 conversions
2270 def : Pat<(v4f32 (X86vfpround (v2f64 VR128:$src))),
2271 (CVTPD2PSrr VR128:$src)>;
2272 def : Pat<(v4f32 (X86vfpround (memopv2f64 addr:$src))),
2273 (CVTPD2PSrm addr:$src)>;
2275 def : Pat<(v2f64 (X86vfpext (v4f32 VR128:$src))),
2276 (CVTPS2PDrr VR128:$src)>;
2279 //===----------------------------------------------------------------------===//
2280 // SSE 1 & 2 - Compare Instructions
2281 //===----------------------------------------------------------------------===//
2283 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
2284 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
2285 Operand CC, SDNode OpNode, ValueType VT,
2286 PatFrag ld_frag, string asm, string asm_alt,
2288 def rr : SIi8<0xC2, MRMSrcReg,
2289 (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
2290 [(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, imm:$cc))],
2291 itins.rr>, Sched<[itins.Sched]>;
2292 def rm : SIi8<0xC2, MRMSrcMem,
2293 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
2294 [(set RC:$dst, (OpNode (VT RC:$src1),
2295 (ld_frag addr:$src2), imm:$cc))],
2297 Sched<[itins.Sched.Folded, ReadAfterLd]>;
2299 // Accept explicit immediate argument form instead of comparison code.
2300 let neverHasSideEffects = 1 in {
2301 def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst),
2302 (ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, [],
2303 IIC_SSE_ALU_F32S_RR>, Sched<[itins.Sched]>;
2305 def rm_alt : SIi8<0xC2, MRMSrcMem, (outs RC:$dst),
2306 (ins RC:$src1, x86memop:$src2, i8imm:$cc), asm_alt, [],
2307 IIC_SSE_ALU_F32S_RM>,
2308 Sched<[itins.Sched.Folded, ReadAfterLd]>;
2312 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem, AVXCC, X86cmps, f32, loadf32,
2313 "cmp${cc}ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2314 "cmpss\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2316 XS, VEX_4V, VEX_LIG;
2317 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem, AVXCC, X86cmps, f64, loadf64,
2318 "cmp${cc}sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2319 "cmpsd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2320 SSE_ALU_F32S>, // same latency as 32 bit compare
2321 XD, VEX_4V, VEX_LIG;
2323 let Constraints = "$src1 = $dst" in {
2324 defm CMPSS : sse12_cmp_scalar<FR32, f32mem, SSECC, X86cmps, f32, loadf32,
2325 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
2326 "cmpss\t{$cc, $src2, $dst|$dst, $src2, $cc}", SSE_ALU_F32S>,
2328 defm CMPSD : sse12_cmp_scalar<FR64, f64mem, SSECC, X86cmps, f64, loadf64,
2329 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
2330 "cmpsd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2335 multiclass sse12_cmp_scalar_int<X86MemOperand x86memop, Operand CC,
2336 Intrinsic Int, string asm, OpndItins itins> {
2337 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
2338 (ins VR128:$src1, VR128:$src, CC:$cc), asm,
2339 [(set VR128:$dst, (Int VR128:$src1,
2340 VR128:$src, imm:$cc))],
2342 Sched<[itins.Sched]>;
2343 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
2344 (ins VR128:$src1, x86memop:$src, CC:$cc), asm,
2345 [(set VR128:$dst, (Int VR128:$src1,
2346 (load addr:$src), imm:$cc))],
2348 Sched<[itins.Sched.Folded, ReadAfterLd]>;
2351 let isCodeGenOnly = 1 in {
2352 // Aliases to match intrinsics which expect XMM operand(s).
2353 defm Int_VCMPSS : sse12_cmp_scalar_int<f32mem, AVXCC, int_x86_sse_cmp_ss,
2354 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
2357 defm Int_VCMPSD : sse12_cmp_scalar_int<f64mem, AVXCC, int_x86_sse2_cmp_sd,
2358 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
2359 SSE_ALU_F32S>, // same latency as f32
2361 let Constraints = "$src1 = $dst" in {
2362 defm Int_CMPSS : sse12_cmp_scalar_int<f32mem, SSECC, int_x86_sse_cmp_ss,
2363 "cmp${cc}ss\t{$src, $dst|$dst, $src}",
2365 defm Int_CMPSD : sse12_cmp_scalar_int<f64mem, SSECC, int_x86_sse2_cmp_sd,
2366 "cmp${cc}sd\t{$src, $dst|$dst, $src}",
2373 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
2374 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
2375 ValueType vt, X86MemOperand x86memop,
2376 PatFrag ld_frag, string OpcodeStr> {
2377 def rr: SI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
2378 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2379 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))],
2382 def rm: SI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
2383 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2384 [(set EFLAGS, (OpNode (vt RC:$src1),
2385 (ld_frag addr:$src2)))],
2387 Sched<[WriteFAddLd, ReadAfterLd]>;
2390 let Defs = [EFLAGS] in {
2391 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2392 "ucomiss">, TB, VEX, VEX_LIG;
2393 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2394 "ucomisd">, TB, OpSize, VEX, VEX_LIG;
2395 let Pattern = []<dag> in {
2396 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2397 "comiss">, TB, VEX, VEX_LIG;
2398 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2399 "comisd">, TB, OpSize, VEX, VEX_LIG;
2402 let isCodeGenOnly = 1 in {
2403 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2404 load, "ucomiss">, TB, VEX;
2405 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2406 load, "ucomisd">, TB, OpSize, VEX;
2408 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
2409 load, "comiss">, TB, VEX;
2410 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
2411 load, "comisd">, TB, OpSize, VEX;
2413 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2415 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2416 "ucomisd">, TB, OpSize;
2418 let Pattern = []<dag> in {
2419 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2421 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2422 "comisd">, TB, OpSize;
2425 let isCodeGenOnly = 1 in {
2426 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2427 load, "ucomiss">, TB;
2428 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2429 load, "ucomisd">, TB, OpSize;
2431 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
2433 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
2434 "comisd">, TB, OpSize;
2436 } // Defs = [EFLAGS]
2438 // sse12_cmp_packed - sse 1 & 2 compare packed instructions
2439 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
2440 Operand CC, Intrinsic Int, string asm,
2441 string asm_alt, Domain d,
2442 OpndItins itins = SSE_ALU_F32P> {
2443 def rri : PIi8<0xC2, MRMSrcReg,
2444 (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
2445 [(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))],
2448 def rmi : PIi8<0xC2, MRMSrcMem,
2449 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
2450 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))],
2452 Sched<[WriteFAddLd, ReadAfterLd]>;
2454 // Accept explicit immediate argument form instead of comparison code.
2455 let neverHasSideEffects = 1 in {
2456 def rri_alt : PIi8<0xC2, MRMSrcReg,
2457 (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
2458 asm_alt, [], itins.rr, d>, Sched<[WriteFAdd]>;
2459 def rmi_alt : PIi8<0xC2, MRMSrcMem,
2460 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc),
2461 asm_alt, [], itins.rm, d>,
2462 Sched<[WriteFAddLd, ReadAfterLd]>;
2466 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse_cmp_ps,
2467 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2468 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2469 SSEPackedSingle>, TB, VEX_4V;
2470 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse2_cmp_pd,
2471 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2472 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2473 SSEPackedDouble>, TB, OpSize, VEX_4V;
2474 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_ps_256,
2475 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2476 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2477 SSEPackedSingle>, TB, VEX_4V, VEX_L;
2478 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_pd_256,
2479 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2480 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2481 SSEPackedDouble>, TB, OpSize, VEX_4V, VEX_L;
2482 let Constraints = "$src1 = $dst" in {
2483 defm CMPPS : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse_cmp_ps,
2484 "cmp${cc}ps\t{$src2, $dst|$dst, $src2}",
2485 "cmpps\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2486 SSEPackedSingle, SSE_ALU_F32P>, TB;
2487 defm CMPPD : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse2_cmp_pd,
2488 "cmp${cc}pd\t{$src2, $dst|$dst, $src2}",
2489 "cmppd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2490 SSEPackedDouble, SSE_ALU_F64P>, TB, OpSize;
2493 let Predicates = [HasAVX] in {
2494 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2495 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2496 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2497 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2498 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2499 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2500 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2501 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2503 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
2504 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
2505 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
2506 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
2507 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
2508 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
2509 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
2510 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
2513 let Predicates = [UseSSE1] in {
2514 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2515 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2516 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2517 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2520 let Predicates = [UseSSE2] in {
2521 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2522 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2523 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2524 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2527 //===----------------------------------------------------------------------===//
2528 // SSE 1 & 2 - Shuffle Instructions
2529 //===----------------------------------------------------------------------===//
2531 /// sse12_shuffle - sse 1 & 2 shuffle instructions
2532 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
2533 ValueType vt, string asm, PatFrag mem_frag,
2534 Domain d, bit IsConvertibleToThreeAddress = 0> {
2535 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
2536 (ins RC:$src1, x86memop:$src2, i8imm:$src3), asm,
2537 [(set RC:$dst, (vt (X86Shufp RC:$src1, (mem_frag addr:$src2),
2538 (i8 imm:$src3))))], IIC_SSE_SHUFP, d>,
2539 Sched<[WriteShuffleLd, ReadAfterLd]>;
2540 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
2541 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
2542 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
2543 [(set RC:$dst, (vt (X86Shufp RC:$src1, RC:$src2,
2544 (i8 imm:$src3))))], IIC_SSE_SHUFP, d>,
2545 Sched<[WriteShuffle]>;
2548 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2549 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2550 loadv4f32, SSEPackedSingle>, TB, VEX_4V;
2551 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
2552 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2553 loadv8f32, SSEPackedSingle>, TB, VEX_4V, VEX_L;
2554 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2555 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2556 loadv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2557 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
2558 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2559 loadv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V, VEX_L;
2561 let Constraints = "$src1 = $dst" in {
2562 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2563 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2564 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
2566 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2567 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2568 memopv2f64, SSEPackedDouble, 1 /* cvt to pshufd */>,
2572 let Predicates = [HasAVX] in {
2573 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2574 (bc_v4i32 (loadv2i64 addr:$src2)), (i8 imm:$imm))),
2575 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2576 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2577 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2579 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2580 (loadv2i64 addr:$src2), (i8 imm:$imm))),
2581 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2582 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2583 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2586 def : Pat<(v8i32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2587 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2588 def : Pat<(v8i32 (X86Shufp VR256:$src1,
2589 (bc_v8i32 (loadv4i64 addr:$src2)), (i8 imm:$imm))),
2590 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2592 def : Pat<(v4i64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2593 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2594 def : Pat<(v4i64 (X86Shufp VR256:$src1,
2595 (loadv4i64 addr:$src2), (i8 imm:$imm))),
2596 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2599 let Predicates = [UseSSE1] in {
2600 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2601 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2602 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2603 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2604 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2607 let Predicates = [UseSSE2] in {
2608 // Generic SHUFPD patterns
2609 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2610 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2611 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2612 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2613 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2616 //===----------------------------------------------------------------------===//
2617 // SSE 1 & 2 - Unpack Instructions
2618 //===----------------------------------------------------------------------===//
2620 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
2621 multiclass sse12_unpack_interleave<bits<8> opc, SDNode OpNode, ValueType vt,
2622 PatFrag mem_frag, RegisterClass RC,
2623 X86MemOperand x86memop, string asm,
2625 def rr : PI<opc, MRMSrcReg,
2626 (outs RC:$dst), (ins RC:$src1, RC:$src2),
2628 (vt (OpNode RC:$src1, RC:$src2)))],
2629 IIC_SSE_UNPCK, d>, Sched<[WriteShuffle]>;
2630 def rm : PI<opc, MRMSrcMem,
2631 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
2633 (vt (OpNode RC:$src1,
2634 (mem_frag addr:$src2))))],
2636 Sched<[WriteShuffleLd, ReadAfterLd]>;
2639 defm VUNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, loadv4f32,
2640 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2641 SSEPackedSingle>, TB, VEX_4V;
2642 defm VUNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, loadv2f64,
2643 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2644 SSEPackedDouble>, TB, OpSize, VEX_4V;
2645 defm VUNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, loadv4f32,
2646 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2647 SSEPackedSingle>, TB, VEX_4V;
2648 defm VUNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, loadv2f64,
2649 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2650 SSEPackedDouble>, TB, OpSize, VEX_4V;
2652 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, X86Unpckh, v8f32, loadv8f32,
2653 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2654 SSEPackedSingle>, TB, VEX_4V, VEX_L;
2655 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, X86Unpckh, v4f64, loadv4f64,
2656 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2657 SSEPackedDouble>, TB, OpSize, VEX_4V, VEX_L;
2658 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, X86Unpckl, v8f32, loadv8f32,
2659 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2660 SSEPackedSingle>, TB, VEX_4V, VEX_L;
2661 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, X86Unpckl, v4f64, loadv4f64,
2662 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2663 SSEPackedDouble>, TB, OpSize, VEX_4V, VEX_L;
2665 let Constraints = "$src1 = $dst" in {
2666 defm UNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
2667 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
2668 SSEPackedSingle>, TB;
2669 defm UNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
2670 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
2671 SSEPackedDouble>, TB, OpSize;
2672 defm UNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
2673 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
2674 SSEPackedSingle>, TB;
2675 defm UNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
2676 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
2677 SSEPackedDouble>, TB, OpSize;
2678 } // Constraints = "$src1 = $dst"
2680 let Predicates = [HasAVX1Only] in {
2681 def : Pat<(v8i32 (X86Unpckl VR256:$src1, (bc_v8i32 (loadv4i64 addr:$src2)))),
2682 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
2683 def : Pat<(v8i32 (X86Unpckl VR256:$src1, VR256:$src2)),
2684 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
2685 def : Pat<(v8i32 (X86Unpckh VR256:$src1, (bc_v8i32 (loadv4i64 addr:$src2)))),
2686 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
2687 def : Pat<(v8i32 (X86Unpckh VR256:$src1, VR256:$src2)),
2688 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
2690 def : Pat<(v4i64 (X86Unpckl VR256:$src1, (loadv4i64 addr:$src2))),
2691 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
2692 def : Pat<(v4i64 (X86Unpckl VR256:$src1, VR256:$src2)),
2693 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
2694 def : Pat<(v4i64 (X86Unpckh VR256:$src1, (loadv4i64 addr:$src2))),
2695 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
2696 def : Pat<(v4i64 (X86Unpckh VR256:$src1, VR256:$src2)),
2697 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
2700 let Predicates = [HasAVX] in {
2701 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2702 // problem is during lowering, where it's not possible to recognize the load
2703 // fold cause it has two uses through a bitcast. One use disappears at isel
2704 // time and the fold opportunity reappears.
2705 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2706 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2709 let Predicates = [UseSSE2] in {
2710 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2711 // problem is during lowering, where it's not possible to recognize the load
2712 // fold cause it has two uses through a bitcast. One use disappears at isel
2713 // time and the fold opportunity reappears.
2714 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2715 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2718 //===----------------------------------------------------------------------===//
2719 // SSE 1 & 2 - Extract Floating-Point Sign mask
2720 //===----------------------------------------------------------------------===//
2722 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2723 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2725 def rr : PI<0x50, MRMSrcReg, (outs GR32orGR64:$dst), (ins RC:$src),
2726 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2727 [(set GR32orGR64:$dst, (Int RC:$src))], IIC_SSE_MOVMSK, d>,
2728 Sched<[WriteVecLogic]>;
2731 let Predicates = [HasAVX] in {
2732 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2733 "movmskps", SSEPackedSingle>, TB, VEX;
2734 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2735 "movmskpd", SSEPackedDouble>, TB,
2737 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2738 "movmskps", SSEPackedSingle>, TB,
2740 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2741 "movmskpd", SSEPackedDouble>, TB,
2744 def : Pat<(i32 (X86fgetsign FR32:$src)),
2745 (VMOVMSKPSrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
2746 def : Pat<(i64 (X86fgetsign FR32:$src)),
2747 (SUBREG_TO_REG (i64 0),
2748 (VMOVMSKPSrr (COPY_TO_REGCLASS FR32:$src, VR128)), sub_32bit)>;
2749 def : Pat<(i32 (X86fgetsign FR64:$src)),
2750 (VMOVMSKPDrr (COPY_TO_REGCLASS FR64:$src, VR128))>;
2751 def : Pat<(i64 (X86fgetsign FR64:$src)),
2752 (SUBREG_TO_REG (i64 0),
2753 (VMOVMSKPDrr (COPY_TO_REGCLASS FR64:$src, VR128)), sub_32bit)>;
2756 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2757 SSEPackedSingle>, TB;
2758 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2759 SSEPackedDouble>, TB, OpSize;
2761 def : Pat<(i32 (X86fgetsign FR32:$src)),
2762 (MOVMSKPSrr (COPY_TO_REGCLASS FR32:$src, VR128))>,
2763 Requires<[UseSSE1]>;
2764 def : Pat<(i64 (X86fgetsign FR32:$src)),
2765 (SUBREG_TO_REG (i64 0),
2766 (MOVMSKPSrr (COPY_TO_REGCLASS FR32:$src, VR128)), sub_32bit)>,
2767 Requires<[UseSSE1]>;
2768 def : Pat<(i32 (X86fgetsign FR64:$src)),
2769 (MOVMSKPDrr (COPY_TO_REGCLASS FR64:$src, VR128))>,
2770 Requires<[UseSSE2]>;
2771 def : Pat<(i64 (X86fgetsign FR64:$src)),
2772 (SUBREG_TO_REG (i64 0),
2773 (MOVMSKPDrr (COPY_TO_REGCLASS FR64:$src, VR128)), sub_32bit)>,
2774 Requires<[UseSSE2]>;
2776 //===---------------------------------------------------------------------===//
2777 // SSE2 - Packed Integer Logical Instructions
2778 //===---------------------------------------------------------------------===//
2780 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2782 /// PDI_binop_rm - Simple SSE2 binary operator.
2783 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2784 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
2785 X86MemOperand x86memop, OpndItins itins,
2786 bit IsCommutable, bit Is2Addr> {
2787 let isCommutable = IsCommutable in
2788 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
2789 (ins RC:$src1, RC:$src2),
2791 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2792 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2793 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>,
2794 Sched<[itins.Sched]>;
2795 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
2796 (ins RC:$src1, x86memop:$src2),
2798 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2799 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2800 [(set RC:$dst, (OpVT (OpNode RC:$src1,
2801 (bitconvert (memop_frag addr:$src2)))))],
2803 Sched<[itins.Sched.Folded, ReadAfterLd]>;
2805 } // ExeDomain = SSEPackedInt
2807 multiclass PDI_binop_all<bits<8> opc, string OpcodeStr, SDNode Opcode,
2808 ValueType OpVT128, ValueType OpVT256,
2809 OpndItins itins, bit IsCommutable = 0> {
2810 let Predicates = [HasAVX] in
2811 defm V#NAME : PDI_binop_rm<opc, !strconcat("v", OpcodeStr), Opcode, OpVT128,
2812 VR128, loadv2i64, i128mem, itins, IsCommutable, 0>, VEX_4V;
2814 let Constraints = "$src1 = $dst" in
2815 defm NAME : PDI_binop_rm<opc, OpcodeStr, Opcode, OpVT128, VR128,
2816 memopv2i64, i128mem, itins, IsCommutable, 1>;
2818 let Predicates = [HasAVX2] in
2819 defm V#NAME#Y : PDI_binop_rm<opc, !strconcat("v", OpcodeStr), Opcode,
2820 OpVT256, VR256, loadv4i64, i256mem, itins,
2821 IsCommutable, 0>, VEX_4V, VEX_L;
2824 // These are ordered here for pattern ordering requirements with the fp versions
2826 defm PAND : PDI_binop_all<0xDB, "pand", and, v2i64, v4i64, SSE_BIT_ITINS_P, 1>;
2827 defm POR : PDI_binop_all<0xEB, "por", or, v2i64, v4i64, SSE_BIT_ITINS_P, 1>;
2828 defm PXOR : PDI_binop_all<0xEF, "pxor", xor, v2i64, v4i64, SSE_BIT_ITINS_P, 1>;
2829 defm PANDN : PDI_binop_all<0xDF, "pandn", X86andnp, v2i64, v4i64,
2830 SSE_BIT_ITINS_P, 0>;
2832 //===----------------------------------------------------------------------===//
2833 // SSE 1 & 2 - Logical Instructions
2834 //===----------------------------------------------------------------------===//
2836 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2838 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2839 SDNode OpNode, OpndItins itins> {
2840 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2841 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, itins, 0>,
2844 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2845 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, itins, 0>,
2848 let Constraints = "$src1 = $dst" in {
2849 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2850 f32, f128mem, memopfsf32, SSEPackedSingle, itins>,
2853 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2854 f64, f128mem, memopfsf64, SSEPackedDouble, itins>,
2859 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2860 let isCodeGenOnly = 1 in {
2861 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand,
2863 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for,
2865 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor,
2868 let isCommutable = 0 in
2869 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", X86fandn,
2873 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2875 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2877 defm V#NAME#PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2878 !strconcat(OpcodeStr, "ps"), f256mem,
2879 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2880 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2881 (loadv4i64 addr:$src2)))], 0>, TB, VEX_4V, VEX_L;
2883 defm V#NAME#PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2884 !strconcat(OpcodeStr, "pd"), f256mem,
2885 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2886 (bc_v4i64 (v4f64 VR256:$src2))))],
2887 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2888 (loadv4i64 addr:$src2)))], 0>,
2889 TB, OpSize, VEX_4V, VEX_L;
2891 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2892 // are all promoted to v2i64, and the patterns are covered by the int
2893 // version. This is needed in SSE only, because v2i64 isn't supported on
2894 // SSE1, but only on SSE2.
2895 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2896 !strconcat(OpcodeStr, "ps"), f128mem, [],
2897 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2898 (loadv2i64 addr:$src2)))], 0>, TB, VEX_4V;
2900 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2901 !strconcat(OpcodeStr, "pd"), f128mem,
2902 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2903 (bc_v2i64 (v2f64 VR128:$src2))))],
2904 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2905 (loadv2i64 addr:$src2)))], 0>,
2908 let Constraints = "$src1 = $dst" in {
2909 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2910 !strconcat(OpcodeStr, "ps"), f128mem,
2911 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2912 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2913 (memopv2i64 addr:$src2)))]>, TB;
2915 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2916 !strconcat(OpcodeStr, "pd"), f128mem,
2917 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2918 (bc_v2i64 (v2f64 VR128:$src2))))],
2919 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2920 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2924 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2925 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2926 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2927 let isCommutable = 0 in
2928 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2930 //===----------------------------------------------------------------------===//
2931 // SSE 1 & 2 - Arithmetic Instructions
2932 //===----------------------------------------------------------------------===//
2934 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2937 /// In addition, we also have a special variant of the scalar form here to
2938 /// represent the associated intrinsic operation. This form is unlike the
2939 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2940 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2942 /// These three forms can each be reg+reg or reg+mem.
2945 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2947 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr,
2948 SDNode OpNode, SizeItins itins> {
2949 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2950 VR128, v4f32, f128mem, loadv4f32,
2951 SSEPackedSingle, itins.s, 0>, TB, VEX_4V;
2952 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2953 VR128, v2f64, f128mem, loadv2f64,
2954 SSEPackedDouble, itins.d, 0>, TB, OpSize, VEX_4V;
2956 defm V#NAME#PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"),
2957 OpNode, VR256, v8f32, f256mem, loadv8f32,
2958 SSEPackedSingle, itins.s, 0>, TB, VEX_4V, VEX_L;
2959 defm V#NAME#PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"),
2960 OpNode, VR256, v4f64, f256mem, loadv4f64,
2961 SSEPackedDouble, itins.d, 0>, TB, OpSize, VEX_4V, VEX_L;
2963 let Constraints = "$src1 = $dst" in {
2964 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2965 v4f32, f128mem, memopv4f32, SSEPackedSingle,
2967 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2968 v2f64, f128mem, memopv2f64, SSEPackedDouble,
2969 itins.d>, TB, OpSize;
2973 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2975 defm V#NAME#SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2976 OpNode, FR32, f32mem, itins.s, 0>, XS, VEX_4V, VEX_LIG;
2977 defm V#NAME#SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2978 OpNode, FR64, f64mem, itins.d, 0>, XD, VEX_4V, VEX_LIG;
2980 let Constraints = "$src1 = $dst" in {
2981 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2982 OpNode, FR32, f32mem, itins.s>, XS;
2983 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2984 OpNode, FR64, f64mem, itins.d>, XD;
2988 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2990 defm V#NAME#SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2991 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32,
2992 itins.s, 0>, XS, VEX_4V, VEX_LIG;
2993 defm V#NAME#SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2994 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64,
2995 itins.d, 0>, XD, VEX_4V, VEX_LIG;
2997 let Constraints = "$src1 = $dst" in {
2998 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2999 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32,
3001 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
3002 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64,
3007 // Binary Arithmetic instructions
3008 defm ADD : basic_sse12_fp_binop_p<0x58, "add", fadd, SSE_ALU_ITINS_P>,
3009 basic_sse12_fp_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S>,
3010 basic_sse12_fp_binop_s_int<0x58, "add", SSE_ALU_ITINS_S>;
3011 defm MUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, SSE_MUL_ITINS_P>,
3012 basic_sse12_fp_binop_s<0x59, "mul", fmul, SSE_MUL_ITINS_S>,
3013 basic_sse12_fp_binop_s_int<0x59, "mul", SSE_MUL_ITINS_S>;
3014 let isCommutable = 0 in {
3015 defm SUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, SSE_ALU_ITINS_P>,
3016 basic_sse12_fp_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S>,
3017 basic_sse12_fp_binop_s_int<0x5C, "sub", SSE_ALU_ITINS_S>;
3018 defm DIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, SSE_DIV_ITINS_P>,
3019 basic_sse12_fp_binop_s<0x5E, "div", fdiv, SSE_DIV_ITINS_S>,
3020 basic_sse12_fp_binop_s_int<0x5E, "div", SSE_DIV_ITINS_S>;
3021 defm MAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, SSE_ALU_ITINS_P>,
3022 basic_sse12_fp_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S>,
3023 basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S>;
3024 defm MIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, SSE_ALU_ITINS_P>,
3025 basic_sse12_fp_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S>,
3026 basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S>;
3029 let isCodeGenOnly = 1 in {
3030 defm MAXC: basic_sse12_fp_binop_p<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_P>,
3031 basic_sse12_fp_binop_s<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_S>;
3032 defm MINC: basic_sse12_fp_binop_p<0x5D, "min", X86fminc, SSE_ALU_ITINS_P>,
3033 basic_sse12_fp_binop_s<0x5D, "min", X86fminc, SSE_ALU_ITINS_S>;
3036 // Patterns used to select SSE scalar fp arithmetic instructions from
3037 // a scalar fp operation followed by a blend.
3039 // These patterns know, for example, how to select an ADDSS from a
3040 // float add plus vector insert.
3042 // The effect is that the backend no longer emits unnecessary vector
3043 // insert instructions immediately after SSE scalar fp instructions
3044 // like addss or mulss.
3046 // For example, given the following code:
3047 // __m128 foo(__m128 A, __m128 B) {
3052 // previously we generated:
3053 // addss %xmm0, %xmm1
3054 // movss %xmm1, %xmm0
3057 // addss %xmm1, %xmm0
3059 def : Pat<(v4f32 (X86Movss (v4f32 VR128:$dst), (v4f32 (scalar_to_vector (fadd
3060 (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
3062 (ADDSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
3063 def : Pat<(v4f32 (X86Movss (v4f32 VR128:$dst), (v4f32 (scalar_to_vector (fsub
3064 (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
3066 (SUBSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
3067 def : Pat<(v4f32 (X86Movss (v4f32 VR128:$dst), (v4f32 (scalar_to_vector (fmul
3068 (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
3070 (MULSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
3071 def : Pat<(v4f32 (X86Movss (v4f32 VR128:$dst), (v4f32 (scalar_to_vector (fdiv
3072 (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
3074 (DIVSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
3076 let Predicates = [HasSSE2] in {
3077 // SSE2 patterns to select scalar double-precision fp arithmetic instructions
3079 def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fadd
3080 (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
3082 (ADDSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
3083 def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fsub
3084 (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
3086 (SUBSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
3087 def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fmul
3088 (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
3090 (MULSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
3091 def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fdiv
3092 (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
3094 (DIVSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
3097 let Predicates = [UseSSE41] in {
3098 // If the subtarget has SSE4.1 but not AVX, the vector insert
3099 // instruction is lowered into a X86insrtps rather than a X86Movss.
3100 // When selecting SSE scalar single-precision fp arithmetic instructions,
3101 // make sure that we correctly match the X86insrtps.
3103 def : Pat<(v4f32 (X86insrtps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
3104 (fadd (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
3105 FR32:$src))), (iPTR 0))),
3106 (ADDSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
3107 def : Pat<(v4f32 (X86insrtps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
3108 (fsub (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
3109 FR32:$src))), (iPTR 0))),
3110 (SUBSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
3111 def : Pat<(v4f32 (X86insrtps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
3112 (fmul (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
3113 FR32:$src))), (iPTR 0))),
3114 (MULSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
3115 def : Pat<(v4f32 (X86insrtps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
3116 (fdiv (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
3117 FR32:$src))), (iPTR 0))),
3118 (DIVSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
3121 let AddedComplexity = 20, Predicates = [HasAVX] in {
3122 // The following patterns select AVX Scalar single/double precision fp
3123 // arithmetic instructions.
3124 // The 'AddedComplexity' is required to give them higher priority over
3125 // the equivalent SSE/SSE2 patterns.
3127 def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fadd
3128 (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
3130 (VADDSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
3131 def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fsub
3132 (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
3134 (VSUBSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
3135 def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fmul
3136 (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
3138 (VMULSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
3139 def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fdiv
3140 (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
3142 (VDIVSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
3143 def : Pat<(v4f32 (X86insrtps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
3144 (fadd (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
3145 FR32:$src))), (iPTR 0))),
3146 (VADDSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
3147 def : Pat<(v4f32 (X86insrtps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
3148 (fsub (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
3149 FR32:$src))), (iPTR 0))),
3150 (VSUBSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
3151 def : Pat<(v4f32 (X86insrtps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
3152 (fmul (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
3153 FR32:$src))), (iPTR 0))),
3154 (VMULSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
3155 def : Pat<(v4f32 (X86insrtps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
3156 (fdiv (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
3157 FR32:$src))), (iPTR 0))),
3158 (VDIVSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
3161 // Patterns used to select SSE scalar fp arithmetic instructions from
3162 // a vector packed single/double fp operation followed by a vector insert.
3164 // The effect is that the backend converts the packed fp instruction
3165 // followed by a vector insert into a single SSE scalar fp instruction.
3167 // For example, given the following code:
3168 // __m128 foo(__m128 A, __m128 B) {
3169 // __m128 C = A + B;
3170 // return (__m128) {c[0], a[1], a[2], a[3]};
3173 // previously we generated:
3174 // addps %xmm0, %xmm1
3175 // movss %xmm1, %xmm0
3178 // addss %xmm1, %xmm0
3180 def : Pat<(v4f32 (X86Movss (v4f32 VR128:$dst),
3181 (fadd (v4f32 VR128:$dst), (v4f32 VR128:$src)))),
3182 (ADDSSrr_Int v4f32:$dst, v4f32:$src)>;
3183 def : Pat<(v4f32 (X86Movss (v4f32 VR128:$dst),
3184 (fsub (v4f32 VR128:$dst), (v4f32 VR128:$src)))),
3185 (SUBSSrr_Int v4f32:$dst, v4f32:$src)>;
3186 def : Pat<(v4f32 (X86Movss (v4f32 VR128:$dst),
3187 (fmul (v4f32 VR128:$dst), (v4f32 VR128:$src)))),
3188 (MULSSrr_Int v4f32:$dst, v4f32:$src)>;
3189 def : Pat<(v4f32 (X86Movss (v4f32 VR128:$dst),
3190 (fdiv (v4f32 VR128:$dst), (v4f32 VR128:$src)))),
3191 (DIVSSrr_Int v4f32:$dst, v4f32:$src)>;
3193 let Predicates = [HasSSE2] in {
3194 // SSE2 patterns to select scalar double-precision fp arithmetic instructions
3195 // from a packed double-precision fp instruction plus movsd.
3197 def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst),
3198 (fadd (v2f64 VR128:$dst), (v2f64 VR128:$src)))),
3199 (ADDSDrr_Int v2f64:$dst, v2f64:$src)>;
3200 def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst),
3201 (fsub (v2f64 VR128:$dst), (v2f64 VR128:$src)))),
3202 (SUBSDrr_Int v2f64:$dst, v2f64:$src)>;
3203 def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst),
3204 (fmul (v2f64 VR128:$dst), (v2f64 VR128:$src)))),
3205 (MULSDrr_Int v2f64:$dst, v2f64:$src)>;
3206 def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst),
3207 (fdiv (v2f64 VR128:$dst), (v2f64 VR128:$src)))),
3208 (DIVSDrr_Int v2f64:$dst, v2f64:$src)>;
3211 let AddedComplexity = 20, Predicates = [HasAVX] in {
3212 // The following patterns select AVX Scalar single/double precision fp
3213 // arithmetic instructions from a packed single precision fp instruction
3214 // plus movss/movsd.
3215 // The 'AddedComplexity' is required to give them higher priority over
3216 // the equivalent SSE/SSE2 patterns.
3218 def : Pat<(v4f32 (X86Movss (v4f32 VR128:$dst),
3219 (fadd (v4f32 VR128:$dst), (v4f32 VR128:$src)))),
3220 (VADDSSrr_Int v4f32:$dst, v4f32:$src)>;
3221 def : Pat<(v4f32 (X86Movss (v4f32 VR128:$dst),
3222 (fsub (v4f32 VR128:$dst), (v4f32 VR128:$src)))),
3223 (VSUBSSrr_Int v4f32:$dst, v4f32:$src)>;
3224 def : Pat<(v4f32 (X86Movss (v4f32 VR128:$dst),
3225 (fmul (v4f32 VR128:$dst), (v4f32 VR128:$src)))),
3226 (VMULSSrr_Int v4f32:$dst, v4f32:$src)>;
3227 def : Pat<(v4f32 (X86Movss (v4f32 VR128:$dst),
3228 (fdiv (v4f32 VR128:$dst), (v4f32 VR128:$src)))),
3229 (VDIVSSrr_Int v4f32:$dst, v4f32:$src)>;
3230 def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst),
3231 (fadd (v2f64 VR128:$dst), (v2f64 VR128:$src)))),
3232 (VADDSDrr_Int v2f64:$dst, v2f64:$src)>;
3233 def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst),
3234 (fsub (v2f64 VR128:$dst), (v2f64 VR128:$src)))),
3235 (VSUBSDrr_Int v2f64:$dst, v2f64:$src)>;
3236 def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst),
3237 (fmul (v2f64 VR128:$dst), (v2f64 VR128:$src)))),
3238 (VMULSDrr_Int v2f64:$dst, v2f64:$src)>;
3239 def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst),
3240 (fdiv (v2f64 VR128:$dst), (v2f64 VR128:$src)))),
3241 (VDIVSDrr_Int v2f64:$dst, v2f64:$src)>;
3245 /// In addition, we also have a special variant of the scalar form here to
3246 /// represent the associated intrinsic operation. This form is unlike the
3247 /// plain scalar form, in that it takes an entire vector (instead of a
3248 /// scalar) and leaves the top elements undefined.
3250 /// And, we have a special variant form for a full-vector intrinsic form.
3252 let Sched = WriteFSqrt in {
3253 def SSE_SQRTPS : OpndItins<
3254 IIC_SSE_SQRTPS_RR, IIC_SSE_SQRTPS_RM
3257 def SSE_SQRTSS : OpndItins<
3258 IIC_SSE_SQRTSS_RR, IIC_SSE_SQRTSS_RM
3261 def SSE_SQRTPD : OpndItins<
3262 IIC_SSE_SQRTPD_RR, IIC_SSE_SQRTPD_RM
3265 def SSE_SQRTSD : OpndItins<
3266 IIC_SSE_SQRTSD_RR, IIC_SSE_SQRTSD_RM
3270 let Sched = WriteFRcp in {
3271 def SSE_RCPP : OpndItins<
3272 IIC_SSE_RCPP_RR, IIC_SSE_RCPP_RM
3275 def SSE_RCPS : OpndItins<
3276 IIC_SSE_RCPS_RR, IIC_SSE_RCPS_RM
3280 /// sse1_fp_unop_s - SSE1 unops in scalar form.
3281 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
3282 SDNode OpNode, Intrinsic F32Int, OpndItins itins> {
3283 let Predicates = [HasAVX], hasSideEffects = 0 in {
3284 def V#NAME#SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst),
3285 (ins FR32:$src1, FR32:$src2),
3286 !strconcat("v", OpcodeStr,
3287 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3288 []>, VEX_4V, VEX_LIG, Sched<[itins.Sched]>;
3289 let mayLoad = 1 in {
3290 def V#NAME#SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst),
3291 (ins FR32:$src1,f32mem:$src2),
3292 !strconcat("v", OpcodeStr,
3293 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3294 []>, VEX_4V, VEX_LIG,
3295 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3296 let isCodeGenOnly = 1 in
3297 def V#NAME#SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3298 (ins VR128:$src1, ssmem:$src2),
3299 !strconcat("v", OpcodeStr,
3300 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3301 []>, VEX_4V, VEX_LIG,
3302 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3306 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
3307 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3308 [(set FR32:$dst, (OpNode FR32:$src))]>, Sched<[itins.Sched]>;
3309 // For scalar unary operations, fold a load into the operation
3310 // only in OptForSize mode. It eliminates an instruction, but it also
3311 // eliminates a whole-register clobber (the load), so it introduces a
3312 // partial register update condition.
3313 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
3314 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3315 [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
3316 Requires<[UseSSE1, OptForSize]>, Sched<[itins.Sched.Folded]>;
3317 let isCodeGenOnly = 1 in {
3318 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3319 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3320 [(set VR128:$dst, (F32Int VR128:$src))], itins.rr>,
3321 Sched<[itins.Sched]>;
3322 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
3323 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3324 [(set VR128:$dst, (F32Int sse_load_f32:$src))], itins.rm>,
3325 Sched<[itins.Sched.Folded]>;
3329 /// sse1_fp_unop_s_rw - SSE1 unops where vector form has a read-write operand.
3330 multiclass sse1_fp_unop_rw<bits<8> opc, string OpcodeStr, SDNode OpNode,
3332 let Predicates = [HasAVX], hasSideEffects = 0 in {
3333 def V#NAME#SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst),
3334 (ins FR32:$src1, FR32:$src2),
3335 !strconcat("v", OpcodeStr,
3336 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3337 []>, VEX_4V, VEX_LIG, Sched<[itins.Sched]>;
3338 let mayLoad = 1 in {
3339 def V#NAME#SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst),
3340 (ins FR32:$src1,f32mem:$src2),
3341 !strconcat("v", OpcodeStr,
3342 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3343 []>, VEX_4V, VEX_LIG,
3344 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3345 let isCodeGenOnly = 1 in
3346 def V#NAME#SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3347 (ins VR128:$src1, ssmem:$src2),
3348 !strconcat("v", OpcodeStr,
3349 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3350 []>, VEX_4V, VEX_LIG,
3351 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3355 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
3356 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3357 [(set FR32:$dst, (OpNode FR32:$src))]>, Sched<[itins.Sched]>;
3358 // For scalar unary operations, fold a load into the operation
3359 // only in OptForSize mode. It eliminates an instruction, but it also
3360 // eliminates a whole-register clobber (the load), so it introduces a
3361 // partial register update condition.
3362 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
3363 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3364 [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
3365 Requires<[UseSSE1, OptForSize]>, Sched<[itins.Sched.Folded]>;
3366 let isCodeGenOnly = 1, Constraints = "$src1 = $dst" in {
3367 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst),
3368 (ins VR128:$src1, VR128:$src2),
3369 !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
3370 [], itins.rr>, Sched<[itins.Sched]>;
3371 let mayLoad = 1, hasSideEffects = 0 in
3372 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3373 (ins VR128:$src1, ssmem:$src2),
3374 !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
3375 [], itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
3379 /// sse1_fp_unop_p - SSE1 unops in packed form.
3380 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
3382 let Predicates = [HasAVX] in {
3383 def V#NAME#PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3384 !strconcat("v", OpcodeStr,
3385 "ps\t{$src, $dst|$dst, $src}"),
3386 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))],
3387 itins.rr>, VEX, Sched<[itins.Sched]>;
3388 def V#NAME#PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3389 !strconcat("v", OpcodeStr,
3390 "ps\t{$src, $dst|$dst, $src}"),
3391 [(set VR128:$dst, (OpNode (loadv4f32 addr:$src)))],
3392 itins.rm>, VEX, Sched<[itins.Sched.Folded]>;
3393 def V#NAME#PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3394 !strconcat("v", OpcodeStr,
3395 "ps\t{$src, $dst|$dst, $src}"),
3396 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))],
3397 itins.rr>, VEX, VEX_L, Sched<[itins.Sched]>;
3398 def V#NAME#PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3399 !strconcat("v", OpcodeStr,
3400 "ps\t{$src, $dst|$dst, $src}"),
3401 [(set VR256:$dst, (OpNode (loadv8f32 addr:$src)))],
3402 itins.rm>, VEX, VEX_L, Sched<[itins.Sched.Folded]>;
3405 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3406 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3407 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))], itins.rr>,
3408 Sched<[itins.Sched]>;
3409 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3410 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3411 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))], itins.rm>,
3412 Sched<[itins.Sched.Folded]>;
3415 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
3416 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3417 Intrinsic V4F32Int, Intrinsic V8F32Int,
3419 let isCodeGenOnly = 1 in {
3420 let Predicates = [HasAVX] in {
3421 def V#NAME#PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3422 !strconcat("v", OpcodeStr,
3423 "ps\t{$src, $dst|$dst, $src}"),
3424 [(set VR128:$dst, (V4F32Int VR128:$src))],
3425 itins.rr>, VEX, Sched<[itins.Sched]>;
3426 def V#NAME#PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3427 !strconcat("v", OpcodeStr,
3428 "ps\t{$src, $dst|$dst, $src}"),
3429 [(set VR128:$dst, (V4F32Int (loadv4f32 addr:$src)))],
3430 itins.rm>, VEX, Sched<[itins.Sched.Folded]>;
3431 def V#NAME#PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3432 !strconcat("v", OpcodeStr,
3433 "ps\t{$src, $dst|$dst, $src}"),
3434 [(set VR256:$dst, (V8F32Int VR256:$src))],
3435 itins.rr>, VEX, VEX_L, Sched<[itins.Sched]>;
3436 def V#NAME#PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst),
3438 !strconcat("v", OpcodeStr,
3439 "ps\t{$src, $dst|$dst, $src}"),
3440 [(set VR256:$dst, (V8F32Int (loadv8f32 addr:$src)))],
3441 itins.rm>, VEX, VEX_L, Sched<[itins.Sched.Folded]>;
3444 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3445 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3446 [(set VR128:$dst, (V4F32Int VR128:$src))],
3447 itins.rr>, Sched<[itins.Sched]>;
3448 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3449 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3450 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))],
3451 itins.rm>, Sched<[itins.Sched.Folded]>;
3452 } // isCodeGenOnly = 1
3455 /// sse2_fp_unop_s - SSE2 unops in scalar form.
3456 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
3457 SDNode OpNode, Intrinsic F64Int, OpndItins itins> {
3458 let Predicates = [HasAVX], hasSideEffects = 0 in {
3459 def V#NAME#SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst),
3460 (ins FR64:$src1, FR64:$src2),
3461 !strconcat("v", OpcodeStr,
3462 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3463 []>, VEX_4V, VEX_LIG, Sched<[itins.Sched]>;
3464 let mayLoad = 1 in {
3465 def V#NAME#SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst),
3466 (ins FR64:$src1,f64mem:$src2),
3467 !strconcat("v", OpcodeStr,
3468 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3469 []>, VEX_4V, VEX_LIG,
3470 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3471 let isCodeGenOnly = 1 in
3472 def V#NAME#SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
3473 (ins VR128:$src1, sdmem:$src2),
3474 !strconcat("v", OpcodeStr,
3475 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3476 []>, VEX_4V, VEX_LIG,
3477 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3481 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
3482 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3483 [(set FR64:$dst, (OpNode FR64:$src))], itins.rr>,
3484 Sched<[itins.Sched]>;
3485 // See the comments in sse1_fp_unop_s for why this is OptForSize.
3486 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
3487 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3488 [(set FR64:$dst, (OpNode (load addr:$src)))], itins.rm>, XD,
3489 Requires<[UseSSE2, OptForSize]>, Sched<[itins.Sched.Folded]>;
3490 let isCodeGenOnly = 1 in {
3491 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3492 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3493 [(set VR128:$dst, (F64Int VR128:$src))], itins.rr>,
3494 Sched<[itins.Sched]>;
3495 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
3496 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3497 [(set VR128:$dst, (F64Int sse_load_f64:$src))], itins.rm>,
3498 Sched<[itins.Sched.Folded]>;
3502 /// sse2_fp_unop_p - SSE2 unops in vector forms.
3503 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
3504 SDNode OpNode, OpndItins itins> {
3505 let Predicates = [HasAVX] in {
3506 def V#NAME#PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3507 !strconcat("v", OpcodeStr,
3508 "pd\t{$src, $dst|$dst, $src}"),
3509 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))],
3510 itins.rr>, VEX, Sched<[itins.Sched]>;
3511 def V#NAME#PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3512 !strconcat("v", OpcodeStr,
3513 "pd\t{$src, $dst|$dst, $src}"),
3514 [(set VR128:$dst, (OpNode (loadv2f64 addr:$src)))],
3515 itins.rm>, VEX, Sched<[itins.Sched.Folded]>;
3516 def V#NAME#PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3517 !strconcat("v", OpcodeStr,
3518 "pd\t{$src, $dst|$dst, $src}"),
3519 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))],
3520 itins.rr>, VEX, VEX_L, Sched<[itins.Sched]>;
3521 def V#NAME#PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3522 !strconcat("v", OpcodeStr,
3523 "pd\t{$src, $dst|$dst, $src}"),
3524 [(set VR256:$dst, (OpNode (loadv4f64 addr:$src)))],
3525 itins.rm>, VEX, VEX_L, Sched<[itins.Sched.Folded]>;
3528 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3529 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3530 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))], itins.rr>,
3531 Sched<[itins.Sched]>;
3532 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3533 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3534 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))], itins.rm>,
3535 Sched<[itins.Sched.Folded]>;
3539 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss,
3541 sse1_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTPS>,
3542 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd,
3544 sse2_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTPD>;
3546 // Reciprocal approximations. Note that these typically require refinement
3547 // in order to obtain suitable precision.
3548 defm RSQRT : sse1_fp_unop_rw<0x52, "rsqrt", X86frsqrt, SSE_SQRTSS>,
3549 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, SSE_SQRTPS>,
3550 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps,
3551 int_x86_avx_rsqrt_ps_256, SSE_SQRTPS>;
3552 defm RCP : sse1_fp_unop_rw<0x53, "rcp", X86frcp, SSE_RCPS>,
3553 sse1_fp_unop_p<0x53, "rcp", X86frcp, SSE_RCPP>,
3554 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps,
3555 int_x86_avx_rcp_ps_256, SSE_RCPP>;
3557 let Predicates = [UseAVX] in {
3558 def : Pat<(f32 (fsqrt FR32:$src)),
3559 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3560 def : Pat<(f32 (fsqrt (load addr:$src))),
3561 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3562 Requires<[HasAVX, OptForSize]>;
3563 def : Pat<(f64 (fsqrt FR64:$src)),
3564 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
3565 def : Pat<(f64 (fsqrt (load addr:$src))),
3566 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
3567 Requires<[HasAVX, OptForSize]>;
3569 def : Pat<(f32 (X86frsqrt FR32:$src)),
3570 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3571 def : Pat<(f32 (X86frsqrt (load addr:$src))),
3572 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3573 Requires<[HasAVX, OptForSize]>;
3575 def : Pat<(f32 (X86frcp FR32:$src)),
3576 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3577 def : Pat<(f32 (X86frcp (load addr:$src))),
3578 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3579 Requires<[HasAVX, OptForSize]>;
3581 let Predicates = [UseAVX] in {
3582 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
3583 (COPY_TO_REGCLASS (VSQRTSSr (f32 (IMPLICIT_DEF)),
3584 (COPY_TO_REGCLASS VR128:$src, FR32)),
3586 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
3587 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3589 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
3590 (COPY_TO_REGCLASS (VSQRTSDr (f64 (IMPLICIT_DEF)),
3591 (COPY_TO_REGCLASS VR128:$src, FR64)),
3593 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
3594 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
3597 let Predicates = [HasAVX] in {
3598 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3599 (COPY_TO_REGCLASS (VRSQRTSSr (f32 (IMPLICIT_DEF)),
3600 (COPY_TO_REGCLASS VR128:$src, FR32)),
3602 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
3603 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3605 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3606 (COPY_TO_REGCLASS (VRCPSSr (f32 (IMPLICIT_DEF)),
3607 (COPY_TO_REGCLASS VR128:$src, FR32)),
3609 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
3610 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3613 // Reciprocal approximations. Note that these typically require refinement
3614 // in order to obtain suitable precision.
3615 let Predicates = [UseSSE1] in {
3616 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3617 (RSQRTSSr_Int VR128:$src, VR128:$src)>;
3618 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3619 (RCPSSr_Int VR128:$src, VR128:$src)>;
3622 // There is no f64 version of the reciprocal approximation instructions.
3624 //===----------------------------------------------------------------------===//
3625 // SSE 1 & 2 - Non-temporal stores
3626 //===----------------------------------------------------------------------===//
3628 let AddedComplexity = 400 in { // Prefer non-temporal versions
3629 let SchedRW = [WriteStore] in {
3630 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3631 (ins f128mem:$dst, VR128:$src),
3632 "movntps\t{$src, $dst|$dst, $src}",
3633 [(alignednontemporalstore (v4f32 VR128:$src),
3635 IIC_SSE_MOVNT>, VEX;
3636 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3637 (ins f128mem:$dst, VR128:$src),
3638 "movntpd\t{$src, $dst|$dst, $src}",
3639 [(alignednontemporalstore (v2f64 VR128:$src),
3641 IIC_SSE_MOVNT>, VEX;
3643 let ExeDomain = SSEPackedInt in
3644 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3645 (ins f128mem:$dst, VR128:$src),
3646 "movntdq\t{$src, $dst|$dst, $src}",
3647 [(alignednontemporalstore (v2i64 VR128:$src),
3649 IIC_SSE_MOVNT>, VEX;
3651 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3652 (ins f256mem:$dst, VR256:$src),
3653 "movntps\t{$src, $dst|$dst, $src}",
3654 [(alignednontemporalstore (v8f32 VR256:$src),
3656 IIC_SSE_MOVNT>, VEX, VEX_L;
3657 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3658 (ins f256mem:$dst, VR256:$src),
3659 "movntpd\t{$src, $dst|$dst, $src}",
3660 [(alignednontemporalstore (v4f64 VR256:$src),
3662 IIC_SSE_MOVNT>, VEX, VEX_L;
3663 let ExeDomain = SSEPackedInt in
3664 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3665 (ins f256mem:$dst, VR256:$src),
3666 "movntdq\t{$src, $dst|$dst, $src}",
3667 [(alignednontemporalstore (v4i64 VR256:$src),
3669 IIC_SSE_MOVNT>, VEX, VEX_L;
3671 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3672 "movntps\t{$src, $dst|$dst, $src}",
3673 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)],
3675 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3676 "movntpd\t{$src, $dst|$dst, $src}",
3677 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)],
3680 let ExeDomain = SSEPackedInt in
3681 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3682 "movntdq\t{$src, $dst|$dst, $src}",
3683 [(alignednontemporalstore (v2i64 VR128:$src), addr:$dst)],
3686 // There is no AVX form for instructions below this point
3687 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3688 "movnti{l}\t{$src, $dst|$dst, $src}",
3689 [(nontemporalstore (i32 GR32:$src), addr:$dst)],
3691 TB, Requires<[HasSSE2]>;
3692 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3693 "movnti{q}\t{$src, $dst|$dst, $src}",
3694 [(nontemporalstore (i64 GR64:$src), addr:$dst)],
3696 TB, Requires<[HasSSE2]>;
3697 } // SchedRW = [WriteStore]
3699 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3700 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3702 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3703 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[UseSSE2]>;
3704 } // AddedComplexity
3706 //===----------------------------------------------------------------------===//
3707 // SSE 1 & 2 - Prefetch and memory fence
3708 //===----------------------------------------------------------------------===//
3710 // Prefetch intrinsic.
3711 let Predicates = [HasSSE1], SchedRW = [WriteLoad] in {
3712 def PREFETCHT0 : I<0x18, MRM1m, (outs), (ins i8mem:$src),
3713 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))],
3714 IIC_SSE_PREFETCH>, TB;
3715 def PREFETCHT1 : I<0x18, MRM2m, (outs), (ins i8mem:$src),
3716 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))],
3717 IIC_SSE_PREFETCH>, TB;
3718 def PREFETCHT2 : I<0x18, MRM3m, (outs), (ins i8mem:$src),
3719 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))],
3720 IIC_SSE_PREFETCH>, TB;
3721 def PREFETCHNTA : I<0x18, MRM0m, (outs), (ins i8mem:$src),
3722 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))],
3723 IIC_SSE_PREFETCH>, TB;
3726 // FIXME: How should these memory instructions be modeled?
3727 let SchedRW = [WriteLoad] in {
3729 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3730 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)],
3731 IIC_SSE_PREFETCH>, TB, Requires<[HasSSE2]>;
3733 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3734 // was introduced with SSE2, it's backward compatible.
3735 def PAUSE : I<0x90, RawFrm, (outs), (ins),
3736 "pause", [(int_x86_sse2_pause)], IIC_SSE_PAUSE>,
3737 REP, Requires<[HasSSE2]>;
3739 // Load, store, and memory fence
3740 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3741 "sfence", [(int_x86_sse_sfence)], IIC_SSE_SFENCE>,
3742 TB, Requires<[HasSSE1]>;
3743 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3744 "lfence", [(int_x86_sse2_lfence)], IIC_SSE_LFENCE>,
3745 TB, Requires<[HasSSE2]>;
3746 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3747 "mfence", [(int_x86_sse2_mfence)], IIC_SSE_MFENCE>,
3748 TB, Requires<[HasSSE2]>;
3751 def : Pat<(X86SFence), (SFENCE)>;
3752 def : Pat<(X86LFence), (LFENCE)>;
3753 def : Pat<(X86MFence), (MFENCE)>;
3755 //===----------------------------------------------------------------------===//
3756 // SSE 1 & 2 - Load/Store XCSR register
3757 //===----------------------------------------------------------------------===//
3759 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3760 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3761 IIC_SSE_LDMXCSR>, VEX, Sched<[WriteLoad]>;
3762 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3763 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3764 IIC_SSE_STMXCSR>, VEX, Sched<[WriteStore]>;
3766 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3767 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3768 IIC_SSE_LDMXCSR>, Sched<[WriteLoad]>;
3769 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3770 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3771 IIC_SSE_STMXCSR>, Sched<[WriteStore]>;
3773 //===---------------------------------------------------------------------===//
3774 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3775 //===---------------------------------------------------------------------===//
3777 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3779 let neverHasSideEffects = 1, SchedRW = [WriteMove] in {
3780 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3781 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3783 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3784 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3786 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3787 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3789 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3790 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3795 let isCodeGenOnly = 1, hasSideEffects = 0, SchedRW = [WriteMove] in {
3796 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3797 "movdqa\t{$src, $dst|$dst, $src}", [],
3800 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3801 "movdqa\t{$src, $dst|$dst, $src}", [],
3802 IIC_SSE_MOVA_P_RR>, VEX, VEX_L;
3803 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3804 "movdqu\t{$src, $dst|$dst, $src}", [],
3807 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3808 "movdqu\t{$src, $dst|$dst, $src}", [],
3809 IIC_SSE_MOVU_P_RR>, VEX, VEX_L;
3812 let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1,
3813 neverHasSideEffects = 1, SchedRW = [WriteLoad] in {
3814 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3815 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3817 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3818 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3820 let Predicates = [HasAVX] in {
3821 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3822 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3824 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3825 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3830 let mayStore = 1, neverHasSideEffects = 1, SchedRW = [WriteStore] in {
3831 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3832 (ins i128mem:$dst, VR128:$src),
3833 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3835 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3836 (ins i256mem:$dst, VR256:$src),
3837 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3839 let Predicates = [HasAVX] in {
3840 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3841 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3843 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3844 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3849 let SchedRW = [WriteMove] in {
3850 let neverHasSideEffects = 1 in
3851 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3852 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>;
3854 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3855 "movdqu\t{$src, $dst|$dst, $src}",
3856 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>;
3859 let isCodeGenOnly = 1, hasSideEffects = 0 in {
3860 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3861 "movdqa\t{$src, $dst|$dst, $src}", [],
3864 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3865 "movdqu\t{$src, $dst|$dst, $src}",
3866 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>;
3870 let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1,
3871 neverHasSideEffects = 1, SchedRW = [WriteLoad] in {
3872 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3873 "movdqa\t{$src, $dst|$dst, $src}",
3874 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/],
3876 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3877 "movdqu\t{$src, $dst|$dst, $src}",
3878 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/],
3880 XS, Requires<[UseSSE2]>;
3883 let mayStore = 1, neverHasSideEffects = 1, SchedRW = [WriteStore] in {
3884 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3885 "movdqa\t{$src, $dst|$dst, $src}",
3886 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/],
3888 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3889 "movdqu\t{$src, $dst|$dst, $src}",
3890 [/*(store (v2i64 VR128:$src), addr:$dst)*/],
3892 XS, Requires<[UseSSE2]>;
3895 } // ExeDomain = SSEPackedInt
3897 let Predicates = [HasAVX] in {
3898 def : Pat<(int_x86_sse2_storeu_dq addr:$dst, VR128:$src),
3899 (VMOVDQUmr addr:$dst, VR128:$src)>;
3900 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3901 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3903 let Predicates = [UseSSE2] in
3904 def : Pat<(int_x86_sse2_storeu_dq addr:$dst, VR128:$src),
3905 (MOVDQUmr addr:$dst, VR128:$src)>;
3907 //===---------------------------------------------------------------------===//
3908 // SSE2 - Packed Integer Arithmetic Instructions
3909 //===---------------------------------------------------------------------===//
3911 let Sched = WriteVecIMul in
3912 def SSE_PMADD : OpndItins<
3913 IIC_SSE_PMADD, IIC_SSE_PMADD
3916 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3918 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3919 RegisterClass RC, PatFrag memop_frag,
3920 X86MemOperand x86memop,
3922 bit IsCommutable = 0,
3924 let isCommutable = IsCommutable in
3925 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3926 (ins RC:$src1, RC:$src2),
3928 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3929 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3930 [(set RC:$dst, (IntId RC:$src1, RC:$src2))], itins.rr>,
3931 Sched<[itins.Sched]>;
3932 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3933 (ins RC:$src1, x86memop:$src2),
3935 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3936 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3937 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))],
3938 itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
3941 multiclass PDI_binop_all_int<bits<8> opc, string OpcodeStr, Intrinsic IntId128,
3942 Intrinsic IntId256, OpndItins itins,
3943 bit IsCommutable = 0> {
3944 let Predicates = [HasAVX] in
3945 defm V#NAME : PDI_binop_rm_int<opc, !strconcat("v", OpcodeStr), IntId128,
3946 VR128, loadv2i64, i128mem, itins,
3947 IsCommutable, 0>, VEX_4V;
3949 let Constraints = "$src1 = $dst" in
3950 defm NAME : PDI_binop_rm_int<opc, OpcodeStr, IntId128, VR128, memopv2i64,
3951 i128mem, itins, IsCommutable, 1>;
3953 let Predicates = [HasAVX2] in
3954 defm V#NAME#Y : PDI_binop_rm_int<opc, !strconcat("v", OpcodeStr), IntId256,
3955 VR256, loadv4i64, i256mem, itins,
3956 IsCommutable, 0>, VEX_4V, VEX_L;
3959 multiclass PDI_binop_rmi<bits<8> opc, bits<8> opc2, Format ImmForm,
3960 string OpcodeStr, SDNode OpNode,
3961 SDNode OpNode2, RegisterClass RC,
3962 ValueType DstVT, ValueType SrcVT, PatFrag bc_frag,
3963 ShiftOpndItins itins,
3965 // src2 is always 128-bit
3966 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3967 (ins RC:$src1, VR128:$src2),
3969 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3970 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3971 [(set RC:$dst, (DstVT (OpNode RC:$src1, (SrcVT VR128:$src2))))],
3972 itins.rr>, Sched<[WriteVecShift]>;
3973 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3974 (ins RC:$src1, i128mem:$src2),
3976 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3977 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3978 [(set RC:$dst, (DstVT (OpNode RC:$src1,
3979 (bc_frag (memopv2i64 addr:$src2)))))], itins.rm>,
3980 Sched<[WriteVecShiftLd, ReadAfterLd]>;
3981 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3982 (ins RC:$src1, i8imm:$src2),
3984 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3985 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3986 [(set RC:$dst, (DstVT (OpNode2 RC:$src1, (i8 imm:$src2))))], itins.ri>,
3987 Sched<[WriteVecShift]>;
3990 /// PDI_binop_rm2 - Simple SSE2 binary operator with different src and dst types
3991 multiclass PDI_binop_rm2<bits<8> opc, string OpcodeStr, SDNode OpNode,
3992 ValueType DstVT, ValueType SrcVT, RegisterClass RC,
3993 PatFrag memop_frag, X86MemOperand x86memop,
3995 bit IsCommutable = 0, bit Is2Addr = 1> {
3996 let isCommutable = IsCommutable in
3997 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3998 (ins RC:$src1, RC:$src2),
4000 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4001 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4002 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1), RC:$src2)))]>,
4003 Sched<[itins.Sched]>;
4004 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
4005 (ins RC:$src1, x86memop:$src2),
4007 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4008 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4009 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1),
4010 (bitconvert (memop_frag addr:$src2)))))]>,
4011 Sched<[itins.Sched.Folded, ReadAfterLd]>;
4013 } // ExeDomain = SSEPackedInt
4015 defm PADDB : PDI_binop_all<0xFC, "paddb", add, v16i8, v32i8,
4016 SSE_INTALU_ITINS_P, 1>;
4017 defm PADDW : PDI_binop_all<0xFD, "paddw", add, v8i16, v16i16,
4018 SSE_INTALU_ITINS_P, 1>;
4019 defm PADDD : PDI_binop_all<0xFE, "paddd", add, v4i32, v8i32,
4020 SSE_INTALU_ITINS_P, 1>;
4021 defm PADDQ : PDI_binop_all<0xD4, "paddq", add, v2i64, v4i64,
4022 SSE_INTALUQ_ITINS_P, 1>;
4023 defm PMULLW : PDI_binop_all<0xD5, "pmullw", mul, v8i16, v16i16,
4024 SSE_INTMUL_ITINS_P, 1>;
4025 defm PSUBB : PDI_binop_all<0xF8, "psubb", sub, v16i8, v32i8,
4026 SSE_INTALU_ITINS_P, 0>;
4027 defm PSUBW : PDI_binop_all<0xF9, "psubw", sub, v8i16, v16i16,
4028 SSE_INTALU_ITINS_P, 0>;
4029 defm PSUBD : PDI_binop_all<0xFA, "psubd", sub, v4i32, v8i32,
4030 SSE_INTALU_ITINS_P, 0>;
4031 defm PSUBQ : PDI_binop_all<0xFB, "psubq", sub, v2i64, v4i64,
4032 SSE_INTALUQ_ITINS_P, 0>;
4033 defm PSUBUSB : PDI_binop_all<0xD8, "psubusb", X86subus, v16i8, v32i8,
4034 SSE_INTALU_ITINS_P, 0>;
4035 defm PSUBUSW : PDI_binop_all<0xD9, "psubusw", X86subus, v8i16, v16i16,
4036 SSE_INTALU_ITINS_P, 0>;
4037 defm PMINUB : PDI_binop_all<0xDA, "pminub", X86umin, v16i8, v32i8,
4038 SSE_INTALU_ITINS_P, 1>;
4039 defm PMINSW : PDI_binop_all<0xEA, "pminsw", X86smin, v8i16, v16i16,
4040 SSE_INTALU_ITINS_P, 1>;
4041 defm PMAXUB : PDI_binop_all<0xDE, "pmaxub", X86umax, v16i8, v32i8,
4042 SSE_INTALU_ITINS_P, 1>;
4043 defm PMAXSW : PDI_binop_all<0xEE, "pmaxsw", X86smax, v8i16, v16i16,
4044 SSE_INTALU_ITINS_P, 1>;
4047 defm PSUBSB : PDI_binop_all_int<0xE8, "psubsb", int_x86_sse2_psubs_b,
4048 int_x86_avx2_psubs_b, SSE_INTALU_ITINS_P, 0>;
4049 defm PSUBSW : PDI_binop_all_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
4050 int_x86_avx2_psubs_w, SSE_INTALU_ITINS_P, 0>;
4051 defm PADDSB : PDI_binop_all_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
4052 int_x86_avx2_padds_b, SSE_INTALU_ITINS_P, 1>;
4053 defm PADDSW : PDI_binop_all_int<0xED, "paddsw" , int_x86_sse2_padds_w,
4054 int_x86_avx2_padds_w, SSE_INTALU_ITINS_P, 1>;
4055 defm PADDUSB : PDI_binop_all_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
4056 int_x86_avx2_paddus_b, SSE_INTALU_ITINS_P, 1>;
4057 defm PADDUSW : PDI_binop_all_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
4058 int_x86_avx2_paddus_w, SSE_INTALU_ITINS_P, 1>;
4059 defm PMULHUW : PDI_binop_all_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
4060 int_x86_avx2_pmulhu_w, SSE_INTMUL_ITINS_P, 1>;
4061 defm PMULHW : PDI_binop_all_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
4062 int_x86_avx2_pmulh_w, SSE_INTMUL_ITINS_P, 1>;
4063 defm PMADDWD : PDI_binop_all_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
4064 int_x86_avx2_pmadd_wd, SSE_PMADD, 1>;
4065 defm PAVGB : PDI_binop_all_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
4066 int_x86_avx2_pavg_b, SSE_INTALU_ITINS_P, 1>;
4067 defm PAVGW : PDI_binop_all_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
4068 int_x86_avx2_pavg_w, SSE_INTALU_ITINS_P, 1>;
4069 defm PSADBW : PDI_binop_all_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
4070 int_x86_avx2_psad_bw, SSE_PMADD, 1>;
4072 let Predicates = [HasAVX] in
4073 defm VPMULUDQ : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v2i64, v4i32, VR128,
4074 loadv2i64, i128mem, SSE_INTMUL_ITINS_P, 1, 0>,
4076 let Predicates = [HasAVX2] in
4077 defm VPMULUDQY : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v4i64, v8i32,
4078 VR256, loadv4i64, i256mem,
4079 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V, VEX_L;
4080 let Constraints = "$src1 = $dst" in
4081 defm PMULUDQ : PDI_binop_rm2<0xF4, "pmuludq", X86pmuludq, v2i64, v4i32, VR128,
4082 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1>;
4084 //===---------------------------------------------------------------------===//
4085 // SSE2 - Packed Integer Logical Instructions
4086 //===---------------------------------------------------------------------===//
4088 let Predicates = [HasAVX] in {
4089 defm VPSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
4090 VR128, v8i16, v8i16, bc_v8i16,
4091 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
4092 defm VPSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
4093 VR128, v4i32, v4i32, bc_v4i32,
4094 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
4095 defm VPSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
4096 VR128, v2i64, v2i64, bc_v2i64,
4097 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
4099 defm VPSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
4100 VR128, v8i16, v8i16, bc_v8i16,
4101 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
4102 defm VPSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
4103 VR128, v4i32, v4i32, bc_v4i32,
4104 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
4105 defm VPSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
4106 VR128, v2i64, v2i64, bc_v2i64,
4107 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
4109 defm VPSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
4110 VR128, v8i16, v8i16, bc_v8i16,
4111 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
4112 defm VPSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
4113 VR128, v4i32, v4i32, bc_v4i32,
4114 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
4116 let ExeDomain = SSEPackedInt, SchedRW = [WriteVecShift] in {
4117 // 128-bit logical shifts.
4118 def VPSLLDQri : PDIi8<0x73, MRM7r,
4119 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
4120 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4122 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>,
4124 def VPSRLDQri : PDIi8<0x73, MRM3r,
4125 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
4126 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4128 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>,
4130 // PSRADQri doesn't exist in SSE[1-3].
4132 } // Predicates = [HasAVX]
4134 let Predicates = [HasAVX2] in {
4135 defm VPSLLWY : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
4136 VR256, v16i16, v8i16, bc_v8i16,
4137 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
4138 defm VPSLLDY : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
4139 VR256, v8i32, v4i32, bc_v4i32,
4140 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
4141 defm VPSLLQY : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
4142 VR256, v4i64, v2i64, bc_v2i64,
4143 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
4145 defm VPSRLWY : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
4146 VR256, v16i16, v8i16, bc_v8i16,
4147 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
4148 defm VPSRLDY : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
4149 VR256, v8i32, v4i32, bc_v4i32,
4150 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
4151 defm VPSRLQY : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
4152 VR256, v4i64, v2i64, bc_v2i64,
4153 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
4155 defm VPSRAWY : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
4156 VR256, v16i16, v8i16, bc_v8i16,
4157 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
4158 defm VPSRADY : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
4159 VR256, v8i32, v4i32, bc_v4i32,
4160 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
4162 let ExeDomain = SSEPackedInt, SchedRW = [WriteVecShift] in {
4163 // 256-bit logical shifts.
4164 def VPSLLDQYri : PDIi8<0x73, MRM7r,
4165 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
4166 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4168 (int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2))]>,
4170 def VPSRLDQYri : PDIi8<0x73, MRM3r,
4171 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
4172 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4174 (int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2))]>,
4176 // PSRADQYri doesn't exist in SSE[1-3].
4178 } // Predicates = [HasAVX2]
4180 let Constraints = "$src1 = $dst" in {
4181 defm PSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "psllw", X86vshl, X86vshli,
4182 VR128, v8i16, v8i16, bc_v8i16,
4183 SSE_INTSHIFT_ITINS_P>;
4184 defm PSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "pslld", X86vshl, X86vshli,
4185 VR128, v4i32, v4i32, bc_v4i32,
4186 SSE_INTSHIFT_ITINS_P>;
4187 defm PSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "psllq", X86vshl, X86vshli,
4188 VR128, v2i64, v2i64, bc_v2i64,
4189 SSE_INTSHIFT_ITINS_P>;
4191 defm PSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "psrlw", X86vsrl, X86vsrli,
4192 VR128, v8i16, v8i16, bc_v8i16,
4193 SSE_INTSHIFT_ITINS_P>;
4194 defm PSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "psrld", X86vsrl, X86vsrli,
4195 VR128, v4i32, v4i32, bc_v4i32,
4196 SSE_INTSHIFT_ITINS_P>;
4197 defm PSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "psrlq", X86vsrl, X86vsrli,
4198 VR128, v2i64, v2i64, bc_v2i64,
4199 SSE_INTSHIFT_ITINS_P>;
4201 defm PSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "psraw", X86vsra, X86vsrai,
4202 VR128, v8i16, v8i16, bc_v8i16,
4203 SSE_INTSHIFT_ITINS_P>;
4204 defm PSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "psrad", X86vsra, X86vsrai,
4205 VR128, v4i32, v4i32, bc_v4i32,
4206 SSE_INTSHIFT_ITINS_P>;
4208 let ExeDomain = SSEPackedInt, SchedRW = [WriteVecShift] in {
4209 // 128-bit logical shifts.
4210 def PSLLDQri : PDIi8<0x73, MRM7r,
4211 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
4212 "pslldq\t{$src2, $dst|$dst, $src2}",
4214 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))],
4215 IIC_SSE_INTSHDQ_P_RI>;
4216 def PSRLDQri : PDIi8<0x73, MRM3r,
4217 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
4218 "psrldq\t{$src2, $dst|$dst, $src2}",
4220 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))],
4221 IIC_SSE_INTSHDQ_P_RI>;
4222 // PSRADQri doesn't exist in SSE[1-3].
4224 } // Constraints = "$src1 = $dst"
4226 let Predicates = [HasAVX] in {
4227 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
4228 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4229 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
4230 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4231 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
4232 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4234 // Shift up / down and insert zero's.
4235 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
4236 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4237 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
4238 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4241 let Predicates = [HasAVX2] in {
4242 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
4243 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
4244 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
4245 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
4248 let Predicates = [UseSSE2] in {
4249 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
4250 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4251 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
4252 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4253 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
4254 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4256 // Shift up / down and insert zero's.
4257 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
4258 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4259 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
4260 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4263 //===---------------------------------------------------------------------===//
4264 // SSE2 - Packed Integer Comparison Instructions
4265 //===---------------------------------------------------------------------===//
4267 defm PCMPEQB : PDI_binop_all<0x74, "pcmpeqb", X86pcmpeq, v16i8, v32i8,
4268 SSE_INTALU_ITINS_P, 1>;
4269 defm PCMPEQW : PDI_binop_all<0x75, "pcmpeqw", X86pcmpeq, v8i16, v16i16,
4270 SSE_INTALU_ITINS_P, 1>;
4271 defm PCMPEQD : PDI_binop_all<0x76, "pcmpeqd", X86pcmpeq, v4i32, v8i32,
4272 SSE_INTALU_ITINS_P, 1>;
4273 defm PCMPGTB : PDI_binop_all<0x64, "pcmpgtb", X86pcmpgt, v16i8, v32i8,
4274 SSE_INTALU_ITINS_P, 0>;
4275 defm PCMPGTW : PDI_binop_all<0x65, "pcmpgtw", X86pcmpgt, v8i16, v16i16,
4276 SSE_INTALU_ITINS_P, 0>;
4277 defm PCMPGTD : PDI_binop_all<0x66, "pcmpgtd", X86pcmpgt, v4i32, v8i32,
4278 SSE_INTALU_ITINS_P, 0>;
4280 //===---------------------------------------------------------------------===//
4281 // SSE2 - Packed Integer Pack Instructions
4282 //===---------------------------------------------------------------------===//
4284 defm PACKSSWB : PDI_binop_all_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
4285 int_x86_avx2_packsswb, SSE_INTALU_ITINS_P, 0>;
4286 defm PACKSSDW : PDI_binop_all_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
4287 int_x86_avx2_packssdw, SSE_INTALU_ITINS_P, 0>;
4288 defm PACKUSWB : PDI_binop_all_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
4289 int_x86_avx2_packuswb, SSE_INTALU_ITINS_P, 0>;
4291 //===---------------------------------------------------------------------===//
4292 // SSE2 - Packed Integer Shuffle Instructions
4293 //===---------------------------------------------------------------------===//
4295 let ExeDomain = SSEPackedInt in {
4296 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt128, ValueType vt256,
4298 let Predicates = [HasAVX] in {
4299 def V#NAME#ri : Ii8<0x70, MRMSrcReg, (outs VR128:$dst),
4300 (ins VR128:$src1, i8imm:$src2),
4301 !strconcat("v", OpcodeStr,
4302 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4304 (vt128 (OpNode VR128:$src1, (i8 imm:$src2))))],
4305 IIC_SSE_PSHUF_RI>, VEX, Sched<[WriteShuffle]>;
4306 def V#NAME#mi : Ii8<0x70, MRMSrcMem, (outs VR128:$dst),
4307 (ins i128mem:$src1, i8imm:$src2),
4308 !strconcat("v", OpcodeStr,
4309 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4311 (vt128 (OpNode (bitconvert (loadv2i64 addr:$src1)),
4312 (i8 imm:$src2))))], IIC_SSE_PSHUF_MI>, VEX,
4313 Sched<[WriteShuffleLd]>;
4316 let Predicates = [HasAVX2] in {
4317 def V#NAME#Yri : Ii8<0x70, MRMSrcReg, (outs VR256:$dst),
4318 (ins VR256:$src1, i8imm:$src2),
4319 !strconcat("v", OpcodeStr,
4320 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4322 (vt256 (OpNode VR256:$src1, (i8 imm:$src2))))],
4323 IIC_SSE_PSHUF_RI>, VEX, VEX_L, Sched<[WriteShuffle]>;
4324 def V#NAME#Ymi : Ii8<0x70, MRMSrcMem, (outs VR256:$dst),
4325 (ins i256mem:$src1, i8imm:$src2),
4326 !strconcat("v", OpcodeStr,
4327 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4329 (vt256 (OpNode (bitconvert (loadv4i64 addr:$src1)),
4330 (i8 imm:$src2))))], IIC_SSE_PSHUF_MI>, VEX, VEX_L,
4331 Sched<[WriteShuffleLd]>;
4334 let Predicates = [UseSSE2] in {
4335 def ri : Ii8<0x70, MRMSrcReg,
4336 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
4337 !strconcat(OpcodeStr,
4338 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4340 (vt128 (OpNode VR128:$src1, (i8 imm:$src2))))],
4341 IIC_SSE_PSHUF_RI>, Sched<[WriteShuffle]>;
4342 def mi : Ii8<0x70, MRMSrcMem,
4343 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
4344 !strconcat(OpcodeStr,
4345 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4347 (vt128 (OpNode (bitconvert (memopv2i64 addr:$src1)),
4348 (i8 imm:$src2))))], IIC_SSE_PSHUF_MI>,
4349 Sched<[WriteShuffleLd]>;
4352 } // ExeDomain = SSEPackedInt
4354 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, v8i32, X86PShufd>, TB, OpSize;
4355 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, v16i16, X86PShufhw>, XS;
4356 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, v16i16, X86PShuflw>, XD;
4358 let Predicates = [HasAVX] in {
4359 def : Pat<(v4f32 (X86PShufd (loadv4f32 addr:$src1), (i8 imm:$imm))),
4360 (VPSHUFDmi addr:$src1, imm:$imm)>;
4361 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4362 (VPSHUFDri VR128:$src1, imm:$imm)>;
4365 let Predicates = [UseSSE2] in {
4366 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
4367 (PSHUFDmi addr:$src1, imm:$imm)>;
4368 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4369 (PSHUFDri VR128:$src1, imm:$imm)>;
4372 //===---------------------------------------------------------------------===//
4373 // SSE2 - Packed Integer Unpack Instructions
4374 //===---------------------------------------------------------------------===//
4376 let ExeDomain = SSEPackedInt in {
4377 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
4378 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
4379 def rr : PDI<opc, MRMSrcReg,
4380 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4382 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4383 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4384 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))],
4385 IIC_SSE_UNPCK>, Sched<[WriteShuffle]>;
4386 def rm : PDI<opc, MRMSrcMem,
4387 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4389 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4390 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4391 [(set VR128:$dst, (OpNode VR128:$src1,
4392 (bc_frag (memopv2i64
4395 Sched<[WriteShuffleLd, ReadAfterLd]>;
4398 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
4399 SDNode OpNode, PatFrag bc_frag> {
4400 def Yrr : PDI<opc, MRMSrcReg,
4401 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4402 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4403 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>,
4404 Sched<[WriteShuffle]>;
4405 def Yrm : PDI<opc, MRMSrcMem,
4406 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4407 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4408 [(set VR256:$dst, (OpNode VR256:$src1,
4409 (bc_frag (memopv4i64 addr:$src2))))]>,
4410 Sched<[WriteShuffleLd, ReadAfterLd]>;
4413 let Predicates = [HasAVX] in {
4414 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
4415 bc_v16i8, 0>, VEX_4V;
4416 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
4417 bc_v8i16, 0>, VEX_4V;
4418 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
4419 bc_v4i32, 0>, VEX_4V;
4420 defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
4421 bc_v2i64, 0>, VEX_4V;
4423 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
4424 bc_v16i8, 0>, VEX_4V;
4425 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
4426 bc_v8i16, 0>, VEX_4V;
4427 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
4428 bc_v4i32, 0>, VEX_4V;
4429 defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
4430 bc_v2i64, 0>, VEX_4V;
4433 let Predicates = [HasAVX2] in {
4434 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
4435 bc_v32i8>, VEX_4V, VEX_L;
4436 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
4437 bc_v16i16>, VEX_4V, VEX_L;
4438 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
4439 bc_v8i32>, VEX_4V, VEX_L;
4440 defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
4441 bc_v4i64>, VEX_4V, VEX_L;
4443 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
4444 bc_v32i8>, VEX_4V, VEX_L;
4445 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
4446 bc_v16i16>, VEX_4V, VEX_L;
4447 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
4448 bc_v8i32>, VEX_4V, VEX_L;
4449 defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
4450 bc_v4i64>, VEX_4V, VEX_L;
4453 let Constraints = "$src1 = $dst" in {
4454 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
4456 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
4458 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
4460 defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
4463 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
4465 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
4467 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
4469 defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
4472 } // ExeDomain = SSEPackedInt
4474 //===---------------------------------------------------------------------===//
4475 // SSE2 - Packed Integer Extract and Insert
4476 //===---------------------------------------------------------------------===//
4478 let ExeDomain = SSEPackedInt in {
4479 multiclass sse2_pinsrw<bit Is2Addr = 1> {
4480 def rri : Ii8<0xC4, MRMSrcReg,
4481 (outs VR128:$dst), (ins VR128:$src1,
4482 GR32orGR64:$src2, i32i8imm:$src3),
4484 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4485 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4487 (X86pinsrw VR128:$src1, GR32orGR64:$src2, imm:$src3))],
4488 IIC_SSE_PINSRW>, Sched<[WriteShuffle]>;
4489 def rmi : Ii8<0xC4, MRMSrcMem,
4490 (outs VR128:$dst), (ins VR128:$src1,
4491 i16mem:$src2, i32i8imm:$src3),
4493 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4494 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4496 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
4497 imm:$src3))], IIC_SSE_PINSRW>,
4498 Sched<[WriteShuffleLd, ReadAfterLd]>;
4502 let Predicates = [HasAVX] in
4503 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
4504 (outs GR32orGR64:$dst), (ins VR128:$src1, i32i8imm:$src2),
4505 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4506 [(set GR32orGR64:$dst, (X86pextrw (v8i16 VR128:$src1),
4507 imm:$src2))]>, TB, OpSize, VEX,
4508 Sched<[WriteShuffle]>;
4509 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
4510 (outs GR32orGR64:$dst), (ins VR128:$src1, i32i8imm:$src2),
4511 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4512 [(set GR32orGR64:$dst, (X86pextrw (v8i16 VR128:$src1),
4513 imm:$src2))], IIC_SSE_PEXTRW>,
4514 Sched<[WriteShuffleLd, ReadAfterLd]>;
4517 let Predicates = [HasAVX] in
4518 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
4520 let Predicates = [UseSSE2], Constraints = "$src1 = $dst" in
4521 defm PINSRW : sse2_pinsrw, TB, OpSize;
4523 } // ExeDomain = SSEPackedInt
4525 //===---------------------------------------------------------------------===//
4526 // SSE2 - Packed Mask Creation
4527 //===---------------------------------------------------------------------===//
4529 let ExeDomain = SSEPackedInt, SchedRW = [WriteVecLogic] in {
4531 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32orGR64:$dst),
4533 "pmovmskb\t{$src, $dst|$dst, $src}",
4534 [(set GR32orGR64:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4535 IIC_SSE_MOVMSK>, VEX;
4537 let Predicates = [HasAVX2] in {
4538 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32orGR64:$dst),
4540 "pmovmskb\t{$src, $dst|$dst, $src}",
4541 [(set GR32orGR64:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>,
4545 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32orGR64:$dst), (ins VR128:$src),
4546 "pmovmskb\t{$src, $dst|$dst, $src}",
4547 [(set GR32orGR64:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4550 } // ExeDomain = SSEPackedInt
4552 //===---------------------------------------------------------------------===//
4553 // SSE2 - Conditional Store
4554 //===---------------------------------------------------------------------===//
4556 let ExeDomain = SSEPackedInt, SchedRW = [WriteStore] in {
4558 let Uses = [EDI], Predicates = [HasAVX,Not64BitMode] in
4559 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4560 (ins VR128:$src, VR128:$mask),
4561 "maskmovdqu\t{$mask, $src|$src, $mask}",
4562 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4563 IIC_SSE_MASKMOV>, VEX;
4564 let Uses = [RDI], Predicates = [HasAVX,In64BitMode] in
4565 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4566 (ins VR128:$src, VR128:$mask),
4567 "maskmovdqu\t{$mask, $src|$src, $mask}",
4568 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4569 IIC_SSE_MASKMOV>, VEX;
4571 let Uses = [EDI], Predicates = [UseSSE2,Not64BitMode] in
4572 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4573 "maskmovdqu\t{$mask, $src|$src, $mask}",
4574 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4576 let Uses = [RDI], Predicates = [UseSSE2,In64BitMode] in
4577 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4578 "maskmovdqu\t{$mask, $src|$src, $mask}",
4579 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4582 } // ExeDomain = SSEPackedInt
4584 //===---------------------------------------------------------------------===//
4585 // SSE2 - Move Doubleword
4586 //===---------------------------------------------------------------------===//
4588 //===---------------------------------------------------------------------===//
4589 // Move Int Doubleword to Packed Double Int
4591 def VMOVDI2PDIrr : VS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4592 "movd\t{$src, $dst|$dst, $src}",
4594 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>,
4595 VEX, Sched<[WriteMove]>;
4596 def VMOVDI2PDIrm : VS2I<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4597 "movd\t{$src, $dst|$dst, $src}",
4599 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4601 VEX, Sched<[WriteLoad]>;
4602 def VMOV64toPQIrr : VRS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4603 "movq\t{$src, $dst|$dst, $src}",
4605 (v2i64 (scalar_to_vector GR64:$src)))],
4606 IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
4607 let isCodeGenOnly = 1 in
4608 def VMOV64toSDrr : VRS2I<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4609 "movq\t{$src, $dst|$dst, $src}",
4610 [(set FR64:$dst, (bitconvert GR64:$src))],
4611 IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
4613 def MOVDI2PDIrr : S2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4614 "movd\t{$src, $dst|$dst, $src}",
4616 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>,
4618 def MOVDI2PDIrm : S2I<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4619 "movd\t{$src, $dst|$dst, $src}",
4621 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4622 IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
4623 def MOV64toPQIrr : RS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4624 "mov{d|q}\t{$src, $dst|$dst, $src}",
4626 (v2i64 (scalar_to_vector GR64:$src)))],
4627 IIC_SSE_MOVDQ>, Sched<[WriteMove]>;
4628 let isCodeGenOnly = 1 in
4629 def MOV64toSDrr : RS2I<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4630 "mov{d|q}\t{$src, $dst|$dst, $src}",
4631 [(set FR64:$dst, (bitconvert GR64:$src))],
4632 IIC_SSE_MOVDQ>, Sched<[WriteMove]>;
4634 //===---------------------------------------------------------------------===//
4635 // Move Int Doubleword to Single Scalar
4637 let isCodeGenOnly = 1 in {
4638 def VMOVDI2SSrr : VS2I<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4639 "movd\t{$src, $dst|$dst, $src}",
4640 [(set FR32:$dst, (bitconvert GR32:$src))],
4641 IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
4643 def VMOVDI2SSrm : VS2I<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4644 "movd\t{$src, $dst|$dst, $src}",
4645 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4647 VEX, Sched<[WriteLoad]>;
4648 def MOVDI2SSrr : S2I<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4649 "movd\t{$src, $dst|$dst, $src}",
4650 [(set FR32:$dst, (bitconvert GR32:$src))],
4651 IIC_SSE_MOVDQ>, Sched<[WriteMove]>;
4653 def MOVDI2SSrm : S2I<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4654 "movd\t{$src, $dst|$dst, $src}",
4655 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4656 IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
4659 //===---------------------------------------------------------------------===//
4660 // Move Packed Doubleword Int to Packed Double Int
4662 def VMOVPDI2DIrr : VS2I<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4663 "movd\t{$src, $dst|$dst, $src}",
4664 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4665 (iPTR 0)))], IIC_SSE_MOVD_ToGP>, VEX,
4667 def VMOVPDI2DImr : VS2I<0x7E, MRMDestMem, (outs),
4668 (ins i32mem:$dst, VR128:$src),
4669 "movd\t{$src, $dst|$dst, $src}",
4670 [(store (i32 (vector_extract (v4i32 VR128:$src),
4671 (iPTR 0))), addr:$dst)], IIC_SSE_MOVDQ>,
4672 VEX, Sched<[WriteLoad]>;
4673 def MOVPDI2DIrr : S2I<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4674 "movd\t{$src, $dst|$dst, $src}",
4675 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4676 (iPTR 0)))], IIC_SSE_MOVD_ToGP>,
4678 def MOVPDI2DImr : S2I<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4679 "movd\t{$src, $dst|$dst, $src}",
4680 [(store (i32 (vector_extract (v4i32 VR128:$src),
4681 (iPTR 0))), addr:$dst)],
4682 IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
4684 def : Pat<(v8i32 (X86Vinsert (v8i32 immAllZerosV), GR32:$src2, (iPTR 0))),
4685 (SUBREG_TO_REG (i32 0), (VMOVDI2PDIrr GR32:$src2), sub_xmm)>;
4687 def : Pat<(v4i64 (X86Vinsert (bc_v4i64 (v8i32 immAllZerosV)), GR64:$src2, (iPTR 0))),
4688 (SUBREG_TO_REG (i32 0), (VMOV64toPQIrr GR64:$src2), sub_xmm)>;
4690 def : Pat<(v8i32 (X86Vinsert undef, GR32:$src2, (iPTR 0))),
4691 (SUBREG_TO_REG (i32 0), (VMOVDI2PDIrr GR32:$src2), sub_xmm)>;
4693 def : Pat<(v4i64 (X86Vinsert undef, GR64:$src2, (iPTR 0))),
4694 (SUBREG_TO_REG (i32 0), (VMOV64toPQIrr GR64:$src2), sub_xmm)>;
4696 //===---------------------------------------------------------------------===//
4697 // Move Packed Doubleword Int first element to Doubleword Int
4699 let SchedRW = [WriteMove] in {
4700 def VMOVPQIto64rr : VRS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4701 "movq\t{$src, $dst|$dst, $src}",
4702 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4707 def MOVPQIto64rr : RS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4708 "mov{d|q}\t{$src, $dst|$dst, $src}",
4709 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4714 //===---------------------------------------------------------------------===//
4715 // Bitcast FR64 <-> GR64
4717 let isCodeGenOnly = 1 in {
4718 let Predicates = [UseAVX] in
4719 def VMOV64toSDrm : VS2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4720 "movq\t{$src, $dst|$dst, $src}",
4721 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4722 VEX, Sched<[WriteLoad]>;
4723 def VMOVSDto64rr : VRS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4724 "movq\t{$src, $dst|$dst, $src}",
4725 [(set GR64:$dst, (bitconvert FR64:$src))],
4726 IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
4727 def VMOVSDto64mr : VRS2I<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4728 "movq\t{$src, $dst|$dst, $src}",
4729 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4730 IIC_SSE_MOVDQ>, VEX, Sched<[WriteStore]>;
4732 def MOV64toSDrm : S2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4733 "movq\t{$src, $dst|$dst, $src}",
4734 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))],
4735 IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
4736 def MOVSDto64rr : RS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4737 "mov{d|q}\t{$src, $dst|$dst, $src}",
4738 [(set GR64:$dst, (bitconvert FR64:$src))],
4739 IIC_SSE_MOVD_ToGP>, Sched<[WriteMove]>;
4740 def MOVSDto64mr : RS2I<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4741 "movq\t{$src, $dst|$dst, $src}",
4742 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4743 IIC_SSE_MOVDQ>, Sched<[WriteStore]>;
4746 //===---------------------------------------------------------------------===//
4747 // Move Scalar Single to Double Int
4749 let isCodeGenOnly = 1 in {
4750 def VMOVSS2DIrr : VS2I<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4751 "movd\t{$src, $dst|$dst, $src}",
4752 [(set GR32:$dst, (bitconvert FR32:$src))],
4753 IIC_SSE_MOVD_ToGP>, VEX, Sched<[WriteMove]>;
4754 def VMOVSS2DImr : VS2I<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4755 "movd\t{$src, $dst|$dst, $src}",
4756 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4757 IIC_SSE_MOVDQ>, VEX, Sched<[WriteStore]>;
4758 def MOVSS2DIrr : S2I<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4759 "movd\t{$src, $dst|$dst, $src}",
4760 [(set GR32:$dst, (bitconvert FR32:$src))],
4761 IIC_SSE_MOVD_ToGP>, Sched<[WriteMove]>;
4762 def MOVSS2DImr : S2I<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4763 "movd\t{$src, $dst|$dst, $src}",
4764 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4765 IIC_SSE_MOVDQ>, Sched<[WriteStore]>;
4768 //===---------------------------------------------------------------------===//
4769 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4771 let isCodeGenOnly = 1, SchedRW = [WriteMove] in {
4772 let AddedComplexity = 15 in {
4773 def VMOVZQI2PQIrr : VS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4774 "movq\t{$src, $dst|$dst, $src}", // X86-64 only
4775 [(set VR128:$dst, (v2i64 (X86vzmovl
4776 (v2i64 (scalar_to_vector GR64:$src)))))],
4779 def MOVZQI2PQIrr : RS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4780 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4781 [(set VR128:$dst, (v2i64 (X86vzmovl
4782 (v2i64 (scalar_to_vector GR64:$src)))))],
4785 } // isCodeGenOnly, SchedRW
4787 let Predicates = [UseAVX] in {
4788 let AddedComplexity = 15 in
4789 def : Pat<(v4i32 (X86vzmovl (v4i32 (scalar_to_vector GR32:$src)))),
4790 (VMOVDI2PDIrr GR32:$src)>;
4792 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4793 let AddedComplexity = 20 in {
4794 def : Pat<(v4i32 (X86vzmovl (v4i32 (scalar_to_vector (loadi32 addr:$src))))),
4795 (VMOVDI2PDIrm addr:$src)>;
4796 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4797 (VMOVDI2PDIrm addr:$src)>;
4798 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4799 (VMOVDI2PDIrm addr:$src)>;
4801 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4802 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4803 (v4i32 (scalar_to_vector GR32:$src)),(iPTR 0)))),
4804 (SUBREG_TO_REG (i32 0), (VMOVDI2PDIrr GR32:$src), sub_xmm)>;
4805 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4806 (v2i64 (scalar_to_vector GR64:$src)),(iPTR 0)))),
4807 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4810 let Predicates = [UseSSE2] in {
4811 let AddedComplexity = 15 in
4812 def : Pat<(v4i32 (X86vzmovl (v4i32 (scalar_to_vector GR32:$src)))),
4813 (MOVDI2PDIrr GR32:$src)>;
4815 let AddedComplexity = 20 in {
4816 def : Pat<(v4i32 (X86vzmovl (v4i32 (scalar_to_vector (loadi32 addr:$src))))),
4817 (MOVDI2PDIrm addr:$src)>;
4818 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4819 (MOVDI2PDIrm addr:$src)>;
4820 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4821 (MOVDI2PDIrm addr:$src)>;
4825 // These are the correct encodings of the instructions so that we know how to
4826 // read correct assembly, even though we continue to emit the wrong ones for
4827 // compatibility with Darwin's buggy assembler.
4828 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4829 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4830 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4831 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4832 // Allow "vmovd" but print "vmovq" since we don't need compatibility for AVX.
4833 def : InstAlias<"vmovd\t{$src, $dst|$dst, $src}",
4834 (VMOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4835 def : InstAlias<"vmovd\t{$src, $dst|$dst, $src}",
4836 (VMOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4838 //===---------------------------------------------------------------------===//
4839 // SSE2 - Move Quadword
4840 //===---------------------------------------------------------------------===//
4842 //===---------------------------------------------------------------------===//
4843 // Move Quadword Int to Packed Quadword Int
4846 let SchedRW = [WriteLoad] in {
4847 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4848 "vmovq\t{$src, $dst|$dst, $src}",
4850 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4851 VEX, Requires<[UseAVX]>;
4852 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4853 "movq\t{$src, $dst|$dst, $src}",
4855 (v2i64 (scalar_to_vector (loadi64 addr:$src))))],
4857 Requires<[UseSSE2]>; // SSE2 instruction with XS Prefix
4860 //===---------------------------------------------------------------------===//
4861 // Move Packed Quadword Int to Quadword Int
4863 let SchedRW = [WriteStore] in {
4864 def VMOVPQI2QImr : VS2I<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4865 "movq\t{$src, $dst|$dst, $src}",
4866 [(store (i64 (vector_extract (v2i64 VR128:$src),
4867 (iPTR 0))), addr:$dst)],
4868 IIC_SSE_MOVDQ>, VEX;
4869 def MOVPQI2QImr : S2I<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4870 "movq\t{$src, $dst|$dst, $src}",
4871 [(store (i64 (vector_extract (v2i64 VR128:$src),
4872 (iPTR 0))), addr:$dst)],
4876 //===---------------------------------------------------------------------===//
4877 // Store / copy lower 64-bits of a XMM register.
4879 def VMOVLQ128mr : VS2I<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4880 "movq\t{$src, $dst|$dst, $src}",
4881 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX,
4882 Sched<[WriteStore]>;
4883 def MOVLQ128mr : S2I<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4884 "movq\t{$src, $dst|$dst, $src}",
4885 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)],
4886 IIC_SSE_MOVDQ>, Sched<[WriteStore]>;
4888 let isCodeGenOnly = 1, AddedComplexity = 20 in {
4889 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4890 "vmovq\t{$src, $dst|$dst, $src}",
4892 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4893 (loadi64 addr:$src))))))],
4895 XS, VEX, Requires<[UseAVX]>, Sched<[WriteLoad]>;
4897 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4898 "movq\t{$src, $dst|$dst, $src}",
4900 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4901 (loadi64 addr:$src))))))],
4903 XS, Requires<[UseSSE2]>, Sched<[WriteLoad]>;
4906 let Predicates = [UseAVX], AddedComplexity = 20 in {
4907 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4908 (VMOVZQI2PQIrm addr:$src)>;
4909 def : Pat<(v2i64 (X86vzload addr:$src)),
4910 (VMOVZQI2PQIrm addr:$src)>;
4913 let Predicates = [UseSSE2], AddedComplexity = 20 in {
4914 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4915 (MOVZQI2PQIrm addr:$src)>;
4916 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4919 let Predicates = [HasAVX] in {
4920 def : Pat<(v4i64 (alignedX86vzload addr:$src)),
4921 (SUBREG_TO_REG (i32 0), (VMOVAPSrm addr:$src), sub_xmm)>;
4922 def : Pat<(v4i64 (X86vzload addr:$src)),
4923 (SUBREG_TO_REG (i32 0), (VMOVUPSrm addr:$src), sub_xmm)>;
4926 //===---------------------------------------------------------------------===//
4927 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4928 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4930 let SchedRW = [WriteVecLogic] in {
4931 let AddedComplexity = 15 in
4932 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4933 "vmovq\t{$src, $dst|$dst, $src}",
4934 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4936 XS, VEX, Requires<[UseAVX]>;
4937 let AddedComplexity = 15 in
4938 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4939 "movq\t{$src, $dst|$dst, $src}",
4940 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4942 XS, Requires<[UseSSE2]>;
4945 let isCodeGenOnly = 1, SchedRW = [WriteVecLogicLd] in {
4946 let AddedComplexity = 20 in
4947 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4948 "vmovq\t{$src, $dst|$dst, $src}",
4949 [(set VR128:$dst, (v2i64 (X86vzmovl
4950 (loadv2i64 addr:$src))))],
4952 XS, VEX, Requires<[UseAVX]>;
4953 let AddedComplexity = 20 in {
4954 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4955 "movq\t{$src, $dst|$dst, $src}",
4956 [(set VR128:$dst, (v2i64 (X86vzmovl
4957 (loadv2i64 addr:$src))))],
4959 XS, Requires<[UseSSE2]>;
4961 } // isCodeGenOnly, SchedRW
4963 let AddedComplexity = 20 in {
4964 let Predicates = [UseAVX] in {
4965 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4966 (VMOVZPQILo2PQIrr VR128:$src)>;
4968 let Predicates = [UseSSE2] in {
4969 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4970 (MOVZPQILo2PQIrr VR128:$src)>;
4974 //===---------------------------------------------------------------------===//
4975 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4976 //===---------------------------------------------------------------------===//
4977 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4978 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4979 X86MemOperand x86memop> {
4980 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4981 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4982 [(set RC:$dst, (vt (OpNode RC:$src)))],
4983 IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>;
4984 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4985 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4986 [(set RC:$dst, (OpNode (mem_frag addr:$src)))],
4987 IIC_SSE_MOV_LH>, Sched<[WriteShuffleLd]>;
4990 let Predicates = [HasAVX] in {
4991 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4992 v4f32, VR128, loadv4f32, f128mem>, VEX;
4993 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4994 v4f32, VR128, loadv4f32, f128mem>, VEX;
4995 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4996 v8f32, VR256, loadv8f32, f256mem>, VEX, VEX_L;
4997 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4998 v8f32, VR256, loadv8f32, f256mem>, VEX, VEX_L;
5000 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
5001 memopv4f32, f128mem>;
5002 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
5003 memopv4f32, f128mem>;
5005 let Predicates = [HasAVX] in {
5006 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
5007 (VMOVSHDUPrr VR128:$src)>;
5008 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (loadv2i64 addr:$src)))),
5009 (VMOVSHDUPrm addr:$src)>;
5010 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
5011 (VMOVSLDUPrr VR128:$src)>;
5012 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (loadv2i64 addr:$src)))),
5013 (VMOVSLDUPrm addr:$src)>;
5014 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
5015 (VMOVSHDUPYrr VR256:$src)>;
5016 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (loadv4i64 addr:$src)))),
5017 (VMOVSHDUPYrm addr:$src)>;
5018 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
5019 (VMOVSLDUPYrr VR256:$src)>;
5020 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (loadv4i64 addr:$src)))),
5021 (VMOVSLDUPYrm addr:$src)>;
5024 let Predicates = [UseSSE3] in {
5025 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
5026 (MOVSHDUPrr VR128:$src)>;
5027 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
5028 (MOVSHDUPrm addr:$src)>;
5029 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
5030 (MOVSLDUPrr VR128:$src)>;
5031 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
5032 (MOVSLDUPrm addr:$src)>;
5035 //===---------------------------------------------------------------------===//
5036 // SSE3 - Replicate Double FP - MOVDDUP
5037 //===---------------------------------------------------------------------===//
5039 multiclass sse3_replicate_dfp<string OpcodeStr> {
5040 let neverHasSideEffects = 1 in
5041 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5042 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5043 [], IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>;
5044 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
5045 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5048 (scalar_to_vector (loadf64 addr:$src)))))],
5049 IIC_SSE_MOV_LH>, Sched<[WriteShuffleLd]>;
5052 // FIXME: Merge with above classe when there're patterns for the ymm version
5053 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
5054 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
5055 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5056 [(set VR256:$dst, (v4f64 (X86Movddup VR256:$src)))]>,
5057 Sched<[WriteShuffle]>;
5058 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
5059 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5062 (scalar_to_vector (loadf64 addr:$src)))))]>,
5063 Sched<[WriteShuffleLd]>;
5066 let Predicates = [HasAVX] in {
5067 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
5068 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX, VEX_L;
5071 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
5073 let Predicates = [HasAVX] in {
5074 def : Pat<(X86Movddup (loadv2f64 addr:$src)),
5075 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5076 def : Pat<(X86Movddup (bc_v2f64 (loadv4f32 addr:$src))),
5077 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5078 def : Pat<(X86Movddup (bc_v2f64 (loadv2i64 addr:$src))),
5079 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5080 def : Pat<(X86Movddup (bc_v2f64
5081 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5082 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5085 def : Pat<(X86Movddup (loadv4f64 addr:$src)),
5086 (VMOVDDUPYrm addr:$src)>;
5087 def : Pat<(X86Movddup (loadv4i64 addr:$src)),
5088 (VMOVDDUPYrm addr:$src)>;
5089 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
5090 (VMOVDDUPYrm addr:$src)>;
5091 def : Pat<(X86Movddup (v4i64 VR256:$src)),
5092 (VMOVDDUPYrr VR256:$src)>;
5095 let Predicates = [UseSSE3] in {
5096 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5097 (MOVDDUPrm addr:$src)>;
5098 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
5099 (MOVDDUPrm addr:$src)>;
5100 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
5101 (MOVDDUPrm addr:$src)>;
5102 def : Pat<(X86Movddup (bc_v2f64
5103 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5104 (MOVDDUPrm addr:$src)>;
5107 //===---------------------------------------------------------------------===//
5108 // SSE3 - Move Unaligned Integer
5109 //===---------------------------------------------------------------------===//
5111 let SchedRW = [WriteLoad] in {
5112 let Predicates = [HasAVX] in {
5113 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5114 "vlddqu\t{$src, $dst|$dst, $src}",
5115 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
5116 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
5117 "vlddqu\t{$src, $dst|$dst, $src}",
5118 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>,
5121 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5122 "lddqu\t{$src, $dst|$dst, $src}",
5123 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))],
5127 //===---------------------------------------------------------------------===//
5128 // SSE3 - Arithmetic
5129 //===---------------------------------------------------------------------===//
5131 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
5132 X86MemOperand x86memop, OpndItins itins,
5134 def rr : I<0xD0, MRMSrcReg,
5135 (outs RC:$dst), (ins RC:$src1, RC:$src2),
5137 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5138 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5139 [(set RC:$dst, (Int RC:$src1, RC:$src2))], itins.rr>,
5140 Sched<[itins.Sched]>;
5141 def rm : I<0xD0, MRMSrcMem,
5142 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5144 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5145 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5146 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))], itins.rr>,
5147 Sched<[itins.Sched.Folded, ReadAfterLd]>;
5150 let Predicates = [HasAVX] in {
5151 let ExeDomain = SSEPackedSingle in {
5152 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
5153 f128mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V;
5154 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
5155 f256mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V, VEX_L;
5157 let ExeDomain = SSEPackedDouble in {
5158 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
5159 f128mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V;
5160 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
5161 f256mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V, VEX_L;
5164 let Constraints = "$src1 = $dst", Predicates = [UseSSE3] in {
5165 let ExeDomain = SSEPackedSingle in
5166 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
5167 f128mem, SSE_ALU_F32P>, TB, XD;
5168 let ExeDomain = SSEPackedDouble in
5169 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
5170 f128mem, SSE_ALU_F64P>, TB, OpSize;
5173 //===---------------------------------------------------------------------===//
5174 // SSE3 Instructions
5175 //===---------------------------------------------------------------------===//
5178 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5179 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5180 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5182 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5183 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5184 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>,
5187 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5189 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5190 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5191 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
5192 IIC_SSE_HADDSUB_RM>, Sched<[WriteFAddLd, ReadAfterLd]>;
5194 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5195 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5196 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5198 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5199 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5200 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>,
5203 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5205 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5206 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5207 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
5208 IIC_SSE_HADDSUB_RM>, Sched<[WriteFAddLd, ReadAfterLd]>;
5211 let Predicates = [HasAVX] in {
5212 let ExeDomain = SSEPackedSingle in {
5213 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
5214 X86fhadd, 0>, VEX_4V;
5215 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
5216 X86fhsub, 0>, VEX_4V;
5217 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
5218 X86fhadd, 0>, VEX_4V, VEX_L;
5219 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
5220 X86fhsub, 0>, VEX_4V, VEX_L;
5222 let ExeDomain = SSEPackedDouble in {
5223 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
5224 X86fhadd, 0>, VEX_4V;
5225 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
5226 X86fhsub, 0>, VEX_4V;
5227 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
5228 X86fhadd, 0>, VEX_4V, VEX_L;
5229 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
5230 X86fhsub, 0>, VEX_4V, VEX_L;
5234 let Constraints = "$src1 = $dst" in {
5235 let ExeDomain = SSEPackedSingle in {
5236 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
5237 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
5239 let ExeDomain = SSEPackedDouble in {
5240 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
5241 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
5245 //===---------------------------------------------------------------------===//
5246 // SSSE3 - Packed Absolute Instructions
5247 //===---------------------------------------------------------------------===//
5250 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5251 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
5252 Intrinsic IntId128> {
5253 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5255 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5256 [(set VR128:$dst, (IntId128 VR128:$src))], IIC_SSE_PABS_RR>,
5257 OpSize, Sched<[WriteVecALU]>;
5259 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5261 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5264 (bitconvert (memopv2i64 addr:$src))))], IIC_SSE_PABS_RM>,
5265 OpSize, Sched<[WriteVecALULd]>;
5268 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5269 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
5270 Intrinsic IntId256> {
5271 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5273 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5274 [(set VR256:$dst, (IntId256 VR256:$src))]>,
5275 OpSize, Sched<[WriteVecALU]>;
5277 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5279 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5282 (bitconvert (memopv4i64 addr:$src))))]>, OpSize,
5283 Sched<[WriteVecALULd]>;
5286 // Helper fragments to match sext vXi1 to vXiY.
5287 def v16i1sextv16i8 : PatLeaf<(v16i8 (X86pcmpgt (bc_v16i8 (v4i32 immAllZerosV)),
5289 def v8i1sextv8i16 : PatLeaf<(v8i16 (X86vsrai VR128:$src, (i8 15)))>;
5290 def v4i1sextv4i32 : PatLeaf<(v4i32 (X86vsrai VR128:$src, (i8 31)))>;
5291 def v32i1sextv32i8 : PatLeaf<(v32i8 (X86pcmpgt (bc_v32i8 (v8i32 immAllZerosV)),
5293 def v16i1sextv16i16: PatLeaf<(v16i16 (X86vsrai VR256:$src, (i8 15)))>;
5294 def v8i1sextv8i32 : PatLeaf<(v8i32 (X86vsrai VR256:$src, (i8 31)))>;
5296 let Predicates = [HasAVX] in {
5297 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb",
5298 int_x86_ssse3_pabs_b_128>, VEX;
5299 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw",
5300 int_x86_ssse3_pabs_w_128>, VEX;
5301 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd",
5302 int_x86_ssse3_pabs_d_128>, VEX;
5305 (bc_v2i64 (v16i1sextv16i8)),
5306 (bc_v2i64 (add (v16i8 VR128:$src), (v16i1sextv16i8)))),
5307 (VPABSBrr128 VR128:$src)>;
5309 (bc_v2i64 (v8i1sextv8i16)),
5310 (bc_v2i64 (add (v8i16 VR128:$src), (v8i1sextv8i16)))),
5311 (VPABSWrr128 VR128:$src)>;
5313 (bc_v2i64 (v4i1sextv4i32)),
5314 (bc_v2i64 (add (v4i32 VR128:$src), (v4i1sextv4i32)))),
5315 (VPABSDrr128 VR128:$src)>;
5318 let Predicates = [HasAVX2] in {
5319 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb",
5320 int_x86_avx2_pabs_b>, VEX, VEX_L;
5321 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw",
5322 int_x86_avx2_pabs_w>, VEX, VEX_L;
5323 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd",
5324 int_x86_avx2_pabs_d>, VEX, VEX_L;
5327 (bc_v4i64 (v32i1sextv32i8)),
5328 (bc_v4i64 (add (v32i8 VR256:$src), (v32i1sextv32i8)))),
5329 (VPABSBrr256 VR256:$src)>;
5331 (bc_v4i64 (v16i1sextv16i16)),
5332 (bc_v4i64 (add (v16i16 VR256:$src), (v16i1sextv16i16)))),
5333 (VPABSWrr256 VR256:$src)>;
5335 (bc_v4i64 (v8i1sextv8i32)),
5336 (bc_v4i64 (add (v8i32 VR256:$src), (v8i1sextv8i32)))),
5337 (VPABSDrr256 VR256:$src)>;
5340 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb",
5341 int_x86_ssse3_pabs_b_128>;
5342 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw",
5343 int_x86_ssse3_pabs_w_128>;
5344 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
5345 int_x86_ssse3_pabs_d_128>;
5347 let Predicates = [HasSSSE3] in {
5349 (bc_v2i64 (v16i1sextv16i8)),
5350 (bc_v2i64 (add (v16i8 VR128:$src), (v16i1sextv16i8)))),
5351 (PABSBrr128 VR128:$src)>;
5353 (bc_v2i64 (v8i1sextv8i16)),
5354 (bc_v2i64 (add (v8i16 VR128:$src), (v8i1sextv8i16)))),
5355 (PABSWrr128 VR128:$src)>;
5357 (bc_v2i64 (v4i1sextv4i32)),
5358 (bc_v2i64 (add (v4i32 VR128:$src), (v4i1sextv4i32)))),
5359 (PABSDrr128 VR128:$src)>;
5362 //===---------------------------------------------------------------------===//
5363 // SSSE3 - Packed Binary Operator Instructions
5364 //===---------------------------------------------------------------------===//
5366 let Sched = WriteVecALU in {
5367 def SSE_PHADDSUBD : OpndItins<
5368 IIC_SSE_PHADDSUBD_RR, IIC_SSE_PHADDSUBD_RM
5370 def SSE_PHADDSUBSW : OpndItins<
5371 IIC_SSE_PHADDSUBSW_RR, IIC_SSE_PHADDSUBSW_RM
5373 def SSE_PHADDSUBW : OpndItins<
5374 IIC_SSE_PHADDSUBW_RR, IIC_SSE_PHADDSUBW_RM
5377 let Sched = WriteShuffle in
5378 def SSE_PSHUFB : OpndItins<
5379 IIC_SSE_PSHUFB_RR, IIC_SSE_PSHUFB_RM
5381 let Sched = WriteVecALU in
5382 def SSE_PSIGN : OpndItins<
5383 IIC_SSE_PSIGN_RR, IIC_SSE_PSIGN_RM
5385 let Sched = WriteVecIMul in
5386 def SSE_PMULHRSW : OpndItins<
5387 IIC_SSE_PMULHRSW, IIC_SSE_PMULHRSW
5390 /// SS3I_binop_rm - Simple SSSE3 bin op
5391 multiclass SS3I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
5392 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
5393 X86MemOperand x86memop, OpndItins itins,
5395 let isCommutable = 1 in
5396 def rr : SS38I<opc, MRMSrcReg, (outs RC:$dst),
5397 (ins RC:$src1, RC:$src2),
5399 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5400 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5401 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>,
5402 OpSize, Sched<[itins.Sched]>;
5403 def rm : SS38I<opc, MRMSrcMem, (outs RC:$dst),
5404 (ins RC:$src1, x86memop:$src2),
5406 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5407 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5409 (OpVT (OpNode RC:$src1,
5410 (bitconvert (memop_frag addr:$src2)))))], itins.rm>, OpSize,
5411 Sched<[itins.Sched.Folded, ReadAfterLd]>;
5414 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
5415 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
5416 Intrinsic IntId128, OpndItins itins,
5418 let isCommutable = 1 in
5419 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5420 (ins VR128:$src1, VR128:$src2),
5422 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5423 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5424 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5425 OpSize, Sched<[itins.Sched]>;
5426 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5427 (ins VR128:$src1, i128mem:$src2),
5429 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5430 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5432 (IntId128 VR128:$src1,
5433 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize,
5434 Sched<[itins.Sched.Folded, ReadAfterLd]>;
5437 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
5438 Intrinsic IntId256> {
5439 let isCommutable = 1 in
5440 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5441 (ins VR256:$src1, VR256:$src2),
5442 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5443 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
5445 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5446 (ins VR256:$src1, i256mem:$src2),
5447 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5449 (IntId256 VR256:$src1,
5450 (bitconvert (loadv4i64 addr:$src2))))]>, OpSize;
5453 let ImmT = NoImm, Predicates = [HasAVX] in {
5454 let isCommutable = 0 in {
5455 defm VPHADDW : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v8i16, VR128,
5457 SSE_PHADDSUBW, 0>, VEX_4V;
5458 defm VPHADDD : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v4i32, VR128,
5460 SSE_PHADDSUBD, 0>, VEX_4V;
5461 defm VPHSUBW : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v8i16, VR128,
5463 SSE_PHADDSUBW, 0>, VEX_4V;
5464 defm VPHSUBD : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v4i32, VR128,
5466 SSE_PHADDSUBD, 0>, VEX_4V;
5467 defm VPSIGNB : SS3I_binop_rm<0x08, "vpsignb", X86psign, v16i8, VR128,
5469 SSE_PSIGN, 0>, VEX_4V;
5470 defm VPSIGNW : SS3I_binop_rm<0x09, "vpsignw", X86psign, v8i16, VR128,
5472 SSE_PSIGN, 0>, VEX_4V;
5473 defm VPSIGND : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v4i32, VR128,
5475 SSE_PSIGN, 0>, VEX_4V;
5476 defm VPSHUFB : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v16i8, VR128,
5478 SSE_PSHUFB, 0>, VEX_4V;
5479 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw",
5480 int_x86_ssse3_phadd_sw_128,
5481 SSE_PHADDSUBSW, 0>, VEX_4V;
5482 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw",
5483 int_x86_ssse3_phsub_sw_128,
5484 SSE_PHADDSUBSW, 0>, VEX_4V;
5485 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw",
5486 int_x86_ssse3_pmadd_ub_sw_128,
5487 SSE_PMADD, 0>, VEX_4V;
5489 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw",
5490 int_x86_ssse3_pmul_hr_sw_128,
5491 SSE_PMULHRSW, 0>, VEX_4V;
5494 let ImmT = NoImm, Predicates = [HasAVX2] in {
5495 let isCommutable = 0 in {
5496 defm VPHADDWY : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v16i16, VR256,
5498 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5499 defm VPHADDDY : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v8i32, VR256,
5501 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5502 defm VPHSUBWY : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v16i16, VR256,
5504 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5505 defm VPHSUBDY : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v8i32, VR256,
5507 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5508 defm VPSIGNBY : SS3I_binop_rm<0x08, "vpsignb", X86psign, v32i8, VR256,
5510 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5511 defm VPSIGNWY : SS3I_binop_rm<0x09, "vpsignw", X86psign, v16i16, VR256,
5513 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5514 defm VPSIGNDY : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v8i32, VR256,
5516 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5517 defm VPSHUFBY : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v32i8, VR256,
5519 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5520 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
5521 int_x86_avx2_phadd_sw>, VEX_4V, VEX_L;
5522 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
5523 int_x86_avx2_phsub_sw>, VEX_4V, VEX_L;
5524 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
5525 int_x86_avx2_pmadd_ub_sw>, VEX_4V, VEX_L;
5527 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
5528 int_x86_avx2_pmul_hr_sw>, VEX_4V, VEX_L;
5531 // None of these have i8 immediate fields.
5532 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
5533 let isCommutable = 0 in {
5534 defm PHADDW : SS3I_binop_rm<0x01, "phaddw", X86hadd, v8i16, VR128,
5535 memopv2i64, i128mem, SSE_PHADDSUBW>;
5536 defm PHADDD : SS3I_binop_rm<0x02, "phaddd", X86hadd, v4i32, VR128,
5537 memopv2i64, i128mem, SSE_PHADDSUBD>;
5538 defm PHSUBW : SS3I_binop_rm<0x05, "phsubw", X86hsub, v8i16, VR128,
5539 memopv2i64, i128mem, SSE_PHADDSUBW>;
5540 defm PHSUBD : SS3I_binop_rm<0x06, "phsubd", X86hsub, v4i32, VR128,
5541 memopv2i64, i128mem, SSE_PHADDSUBD>;
5542 defm PSIGNB : SS3I_binop_rm<0x08, "psignb", X86psign, v16i8, VR128,
5543 memopv2i64, i128mem, SSE_PSIGN>;
5544 defm PSIGNW : SS3I_binop_rm<0x09, "psignw", X86psign, v8i16, VR128,
5545 memopv2i64, i128mem, SSE_PSIGN>;
5546 defm PSIGND : SS3I_binop_rm<0x0A, "psignd", X86psign, v4i32, VR128,
5547 memopv2i64, i128mem, SSE_PSIGN>;
5548 defm PSHUFB : SS3I_binop_rm<0x00, "pshufb", X86pshufb, v16i8, VR128,
5549 memopv2i64, i128mem, SSE_PSHUFB>;
5550 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw",
5551 int_x86_ssse3_phadd_sw_128,
5553 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw",
5554 int_x86_ssse3_phsub_sw_128,
5556 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw",
5557 int_x86_ssse3_pmadd_ub_sw_128, SSE_PMADD>;
5559 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw",
5560 int_x86_ssse3_pmul_hr_sw_128,
5564 //===---------------------------------------------------------------------===//
5565 // SSSE3 - Packed Align Instruction Patterns
5566 //===---------------------------------------------------------------------===//
5568 multiclass ssse3_palignr<string asm, bit Is2Addr = 1> {
5569 let neverHasSideEffects = 1 in {
5570 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
5571 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5573 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5575 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5576 [], IIC_SSE_PALIGNRR>, OpSize, Sched<[WriteShuffle]>;
5578 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
5579 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5581 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5583 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5584 [], IIC_SSE_PALIGNRM>, OpSize, Sched<[WriteShuffleLd, ReadAfterLd]>;
5588 multiclass ssse3_palignr_y<string asm, bit Is2Addr = 1> {
5589 let neverHasSideEffects = 1 in {
5590 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
5591 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5593 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5594 []>, OpSize, Sched<[WriteShuffle]>;
5596 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
5597 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
5599 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5600 []>, OpSize, Sched<[WriteShuffleLd, ReadAfterLd]>;
5604 let Predicates = [HasAVX] in
5605 defm VPALIGN : ssse3_palignr<"vpalignr", 0>, VEX_4V;
5606 let Predicates = [HasAVX2] in
5607 defm VPALIGN : ssse3_palignr_y<"vpalignr", 0>, VEX_4V, VEX_L;
5608 let Constraints = "$src1 = $dst", Predicates = [UseSSSE3] in
5609 defm PALIGN : ssse3_palignr<"palignr">;
5611 let Predicates = [HasAVX2] in {
5612 def : Pat<(v8i32 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5613 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5614 def : Pat<(v8f32 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5615 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5616 def : Pat<(v16i16 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5617 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5618 def : Pat<(v32i8 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5619 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5622 let Predicates = [HasAVX] in {
5623 def : Pat<(v4i32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5624 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5625 def : Pat<(v4f32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5626 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5627 def : Pat<(v8i16 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5628 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5629 def : Pat<(v16i8 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5630 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5633 let Predicates = [UseSSSE3] in {
5634 def : Pat<(v4i32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5635 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5636 def : Pat<(v4f32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5637 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5638 def : Pat<(v8i16 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5639 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5640 def : Pat<(v16i8 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5641 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5644 //===---------------------------------------------------------------------===//
5645 // SSSE3 - Thread synchronization
5646 //===---------------------------------------------------------------------===//
5648 let SchedRW = [WriteSystem] in {
5649 let usesCustomInserter = 1 in {
5650 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5651 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>,
5652 Requires<[HasSSE3]>;
5655 let Uses = [EAX, ECX, EDX] in
5656 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", [], IIC_SSE_MONITOR>,
5657 TB, Requires<[HasSSE3]>;
5658 let Uses = [ECX, EAX] in
5659 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait",
5660 [(int_x86_sse3_mwait ECX, EAX)], IIC_SSE_MWAIT>,
5661 TB, Requires<[HasSSE3]>;
5664 def : InstAlias<"mwait\t{%eax, %ecx|ecx, eax}", (MWAITrr)>, Requires<[Not64BitMode]>;
5665 def : InstAlias<"mwait\t{%rax, %rcx|rcx, rax}", (MWAITrr)>, Requires<[In64BitMode]>;
5667 def : InstAlias<"monitor\t{%eax, %ecx, %edx|edx, ecx, eax}", (MONITORrrr)>,
5668 Requires<[Not64BitMode]>;
5669 def : InstAlias<"monitor\t{%rax, %rcx, %rdx|rdx, rcx, rax}", (MONITORrrr)>,
5670 Requires<[In64BitMode]>;
5672 //===----------------------------------------------------------------------===//
5673 // SSE4.1 - Packed Move with Sign/Zero Extend
5674 //===----------------------------------------------------------------------===//
5676 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId,
5677 OpndItins itins = DEFAULT_ITINS> {
5678 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5679 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5680 [(set VR128:$dst, (IntId VR128:$src))], itins.rr>, OpSize;
5682 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5683 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5685 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))],
5689 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5691 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5692 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5693 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5695 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5696 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5697 [(set VR256:$dst, (IntId (load addr:$src)))]>,
5701 let Predicates = [HasAVX] in {
5702 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw",
5703 int_x86_sse41_pmovsxbw>, VEX;
5704 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd",
5705 int_x86_sse41_pmovsxwd>, VEX;
5706 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq",
5707 int_x86_sse41_pmovsxdq>, VEX;
5708 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw",
5709 int_x86_sse41_pmovzxbw>, VEX;
5710 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd",
5711 int_x86_sse41_pmovzxwd>, VEX;
5712 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq",
5713 int_x86_sse41_pmovzxdq>, VEX;
5716 let Predicates = [HasAVX2] in {
5717 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5718 int_x86_avx2_pmovsxbw>, VEX, VEX_L;
5719 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5720 int_x86_avx2_pmovsxwd>, VEX, VEX_L;
5721 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5722 int_x86_avx2_pmovsxdq>, VEX, VEX_L;
5723 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5724 int_x86_avx2_pmovzxbw>, VEX, VEX_L;
5725 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5726 int_x86_avx2_pmovzxwd>, VEX, VEX_L;
5727 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5728 int_x86_avx2_pmovzxdq>, VEX, VEX_L;
5731 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw, SSE_INTALU_ITINS_P>;
5732 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd, SSE_INTALU_ITINS_P>;
5733 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq, SSE_INTALU_ITINS_P>;
5734 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw, SSE_INTALU_ITINS_P>;
5735 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd, SSE_INTALU_ITINS_P>;
5736 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq, SSE_INTALU_ITINS_P>;
5738 let Predicates = [HasAVX] in {
5739 // Common patterns involving scalar load.
5740 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5741 (VPMOVSXBWrm addr:$src)>;
5742 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5743 (VPMOVSXBWrm addr:$src)>;
5744 def : Pat<(int_x86_sse41_pmovsxbw (bc_v16i8 (loadv2i64 addr:$src))),
5745 (VPMOVSXBWrm addr:$src)>;
5747 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5748 (VPMOVSXWDrm addr:$src)>;
5749 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5750 (VPMOVSXWDrm addr:$src)>;
5751 def : Pat<(int_x86_sse41_pmovsxwd (bc_v8i16 (loadv2i64 addr:$src))),
5752 (VPMOVSXWDrm addr:$src)>;
5754 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5755 (VPMOVSXDQrm addr:$src)>;
5756 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5757 (VPMOVSXDQrm addr:$src)>;
5758 def : Pat<(int_x86_sse41_pmovsxdq (bc_v4i32 (loadv2i64 addr:$src))),
5759 (VPMOVSXDQrm addr:$src)>;
5761 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5762 (VPMOVZXBWrm addr:$src)>;
5763 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5764 (VPMOVZXBWrm addr:$src)>;
5765 def : Pat<(int_x86_sse41_pmovzxbw (bc_v16i8 (loadv2i64 addr:$src))),
5766 (VPMOVZXBWrm addr:$src)>;
5768 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5769 (VPMOVZXWDrm addr:$src)>;
5770 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5771 (VPMOVZXWDrm addr:$src)>;
5772 def : Pat<(int_x86_sse41_pmovzxwd (bc_v8i16 (loadv2i64 addr:$src))),
5773 (VPMOVZXWDrm addr:$src)>;
5775 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5776 (VPMOVZXDQrm addr:$src)>;
5777 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5778 (VPMOVZXDQrm addr:$src)>;
5779 def : Pat<(int_x86_sse41_pmovzxdq (bc_v4i32 (loadv2i64 addr:$src))),
5780 (VPMOVZXDQrm addr:$src)>;
5783 let Predicates = [UseSSE41] in {
5784 // Common patterns involving scalar load.
5785 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5786 (PMOVSXBWrm addr:$src)>;
5787 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5788 (PMOVSXBWrm addr:$src)>;
5789 def : Pat<(int_x86_sse41_pmovsxbw (bc_v16i8 (loadv2i64 addr:$src))),
5790 (PMOVSXBWrm addr:$src)>;
5792 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5793 (PMOVSXWDrm addr:$src)>;
5794 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5795 (PMOVSXWDrm addr:$src)>;
5796 def : Pat<(int_x86_sse41_pmovsxwd (bc_v8i16 (loadv2i64 addr:$src))),
5797 (PMOVSXWDrm addr:$src)>;
5799 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5800 (PMOVSXDQrm addr:$src)>;
5801 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5802 (PMOVSXDQrm addr:$src)>;
5803 def : Pat<(int_x86_sse41_pmovsxdq (bc_v4i32 (loadv2i64 addr:$src))),
5804 (PMOVSXDQrm addr:$src)>;
5806 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5807 (PMOVZXBWrm addr:$src)>;
5808 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5809 (PMOVZXBWrm addr:$src)>;
5810 def : Pat<(int_x86_sse41_pmovzxbw (bc_v16i8 (loadv2i64 addr:$src))),
5811 (PMOVZXBWrm addr:$src)>;
5813 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5814 (PMOVZXWDrm addr:$src)>;
5815 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5816 (PMOVZXWDrm addr:$src)>;
5817 def : Pat<(int_x86_sse41_pmovzxwd (bc_v8i16 (loadv2i64 addr:$src))),
5818 (PMOVZXWDrm addr:$src)>;
5820 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5821 (PMOVZXDQrm addr:$src)>;
5822 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5823 (PMOVZXDQrm addr:$src)>;
5824 def : Pat<(int_x86_sse41_pmovzxdq (bc_v4i32 (loadv2i64 addr:$src))),
5825 (PMOVZXDQrm addr:$src)>;
5828 let Predicates = [HasAVX2] in {
5829 let AddedComplexity = 15 in {
5830 def : Pat<(v4i64 (X86vzmovly (v4i32 VR128:$src))),
5831 (VPMOVZXDQYrr VR128:$src)>;
5832 def : Pat<(v8i32 (X86vzmovly (v8i16 VR128:$src))),
5833 (VPMOVZXWDYrr VR128:$src)>;
5834 def : Pat<(v16i16 (X86vzmovly (v16i8 VR128:$src))),
5835 (VPMOVZXBWYrr VR128:$src)>;
5838 def : Pat<(v4i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>;
5839 def : Pat<(v8i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>;
5840 def : Pat<(v16i16 (X86vsmovl (v16i8 VR128:$src))), (VPMOVSXBWYrr VR128:$src)>;
5843 let Predicates = [HasAVX] in {
5844 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
5845 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
5846 def : Pat<(v8i16 (X86vsmovl (v16i8 VR128:$src))), (VPMOVSXBWrr VR128:$src)>;
5849 let Predicates = [UseSSE41] in {
5850 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
5851 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
5852 def : Pat<(v8i16 (X86vsmovl (v16i8 VR128:$src))), (PMOVSXBWrr VR128:$src)>;
5856 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId,
5857 OpndItins itins = DEFAULT_ITINS> {
5858 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5859 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5860 [(set VR128:$dst, (IntId VR128:$src))], itins.rr>, OpSize;
5862 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5863 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5865 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))],
5870 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5872 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5873 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5874 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5876 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5877 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5879 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5883 let Predicates = [HasAVX] in {
5884 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5886 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5888 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5890 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5894 let Predicates = [HasAVX2] in {
5895 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5896 int_x86_avx2_pmovsxbd>, VEX, VEX_L;
5897 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5898 int_x86_avx2_pmovsxwq>, VEX, VEX_L;
5899 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5900 int_x86_avx2_pmovzxbd>, VEX, VEX_L;
5901 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5902 int_x86_avx2_pmovzxwq>, VEX, VEX_L;
5905 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd,
5906 SSE_INTALU_ITINS_P>;
5907 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq,
5908 SSE_INTALU_ITINS_P>;
5909 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd,
5910 SSE_INTALU_ITINS_P>;
5911 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq,
5912 SSE_INTALU_ITINS_P>;
5914 let Predicates = [HasAVX] in {
5915 // Common patterns involving scalar load
5916 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5917 (VPMOVSXBDrm addr:$src)>;
5918 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5919 (VPMOVSXWQrm addr:$src)>;
5921 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5922 (VPMOVZXBDrm addr:$src)>;
5923 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5924 (VPMOVZXWQrm addr:$src)>;
5927 let Predicates = [UseSSE41] in {
5928 // Common patterns involving scalar load
5929 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5930 (PMOVSXBDrm addr:$src)>;
5931 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5932 (PMOVSXWQrm addr:$src)>;
5934 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5935 (PMOVZXBDrm addr:$src)>;
5936 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5937 (PMOVZXWQrm addr:$src)>;
5940 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId,
5941 OpndItins itins = DEFAULT_ITINS> {
5942 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5943 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5944 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5946 // Expecting a i16 load any extended to i32 value.
5947 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5948 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5949 [(set VR128:$dst, (IntId (bitconvert
5950 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5954 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5956 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5957 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5958 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5960 // Expecting a i16 load any extended to i32 value.
5961 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5962 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5963 [(set VR256:$dst, (IntId (bitconvert
5964 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5968 let Predicates = [HasAVX] in {
5969 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5971 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5974 let Predicates = [HasAVX2] in {
5975 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5976 int_x86_avx2_pmovsxbq>, VEX, VEX_L;
5977 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5978 int_x86_avx2_pmovzxbq>, VEX, VEX_L;
5980 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq,
5981 SSE_INTALU_ITINS_P>;
5982 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq,
5983 SSE_INTALU_ITINS_P>;
5985 let Predicates = [HasAVX2] in {
5986 def : Pat<(v16i16 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBWYrr VR128:$src)>;
5987 def : Pat<(v8i32 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBDYrr VR128:$src)>;
5988 def : Pat<(v4i64 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBQYrr VR128:$src)>;
5990 def : Pat<(v8i32 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>;
5991 def : Pat<(v4i64 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWQYrr VR128:$src)>;
5993 def : Pat<(v4i64 (X86vsext (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>;
5995 def : Pat<(v16i16 (X86vsext (v32i8 VR256:$src))),
5996 (VPMOVSXBWYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5997 def : Pat<(v8i32 (X86vsext (v32i8 VR256:$src))),
5998 (VPMOVSXBDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5999 def : Pat<(v4i64 (X86vsext (v32i8 VR256:$src))),
6000 (VPMOVSXBQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
6002 def : Pat<(v8i32 (X86vsext (v16i16 VR256:$src))),
6003 (VPMOVSXWDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
6004 def : Pat<(v4i64 (X86vsext (v16i16 VR256:$src))),
6005 (VPMOVSXWQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
6007 def : Pat<(v4i64 (X86vsext (v8i32 VR256:$src))),
6008 (VPMOVSXDQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
6010 def : Pat<(v8i32 (X86vsmovl (v8i16 (bitconvert (v2i64 (load addr:$src)))))),
6011 (VPMOVSXWDYrm addr:$src)>;
6012 def : Pat<(v4i64 (X86vsmovl (v4i32 (bitconvert (v2i64 (load addr:$src)))))),
6013 (VPMOVSXDQYrm addr:$src)>;
6015 def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2i64
6016 (scalar_to_vector (loadi64 addr:$src))))))),
6017 (VPMOVSXBDYrm addr:$src)>;
6018 def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2f64
6019 (scalar_to_vector (loadf64 addr:$src))))))),
6020 (VPMOVSXBDYrm addr:$src)>;
6022 def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2i64
6023 (scalar_to_vector (loadi64 addr:$src))))))),
6024 (VPMOVSXWQYrm addr:$src)>;
6025 def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2f64
6026 (scalar_to_vector (loadf64 addr:$src))))))),
6027 (VPMOVSXWQYrm addr:$src)>;
6029 def : Pat<(v4i64 (X86vsext (v16i8 (bitconvert (v4i32
6030 (scalar_to_vector (loadi32 addr:$src))))))),
6031 (VPMOVSXBQYrm addr:$src)>;
6034 let Predicates = [HasAVX] in {
6035 // Common patterns involving scalar load
6036 def : Pat<(int_x86_sse41_pmovsxbq
6037 (bitconvert (v4i32 (X86vzmovl
6038 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
6039 (VPMOVSXBQrm addr:$src)>;
6041 def : Pat<(int_x86_sse41_pmovzxbq
6042 (bitconvert (v4i32 (X86vzmovl
6043 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
6044 (VPMOVZXBQrm addr:$src)>;
6047 let Predicates = [UseSSE41] in {
6048 def : Pat<(v8i16 (X86vsext (v16i8 VR128:$src))), (PMOVSXBWrr VR128:$src)>;
6049 def : Pat<(v4i32 (X86vsext (v16i8 VR128:$src))), (PMOVSXBDrr VR128:$src)>;
6050 def : Pat<(v2i64 (X86vsext (v16i8 VR128:$src))), (PMOVSXBQrr VR128:$src)>;
6052 def : Pat<(v4i32 (X86vsext (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
6053 def : Pat<(v2i64 (X86vsext (v8i16 VR128:$src))), (PMOVSXWQrr VR128:$src)>;
6055 def : Pat<(v2i64 (X86vsext (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
6057 // Common patterns involving scalar load
6058 def : Pat<(int_x86_sse41_pmovsxbq
6059 (bitconvert (v4i32 (X86vzmovl
6060 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
6061 (PMOVSXBQrm addr:$src)>;
6063 def : Pat<(int_x86_sse41_pmovzxbq
6064 (bitconvert (v4i32 (X86vzmovl
6065 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
6066 (PMOVZXBQrm addr:$src)>;
6068 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64
6069 (scalar_to_vector (loadi64 addr:$src))))))),
6070 (PMOVSXWDrm addr:$src)>;
6071 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64
6072 (scalar_to_vector (loadf64 addr:$src))))))),
6073 (PMOVSXWDrm addr:$src)>;
6074 def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32
6075 (scalar_to_vector (loadi32 addr:$src))))))),
6076 (PMOVSXBDrm addr:$src)>;
6077 def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32
6078 (scalar_to_vector (loadi32 addr:$src))))))),
6079 (PMOVSXWQrm addr:$src)>;
6080 def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32
6081 (scalar_to_vector (extloadi32i16 addr:$src))))))),
6082 (PMOVSXBQrm addr:$src)>;
6083 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64
6084 (scalar_to_vector (loadi64 addr:$src))))))),
6085 (PMOVSXDQrm addr:$src)>;
6086 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64
6087 (scalar_to_vector (loadf64 addr:$src))))))),
6088 (PMOVSXDQrm addr:$src)>;
6089 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64
6090 (scalar_to_vector (loadi64 addr:$src))))))),
6091 (PMOVSXBWrm addr:$src)>;
6092 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64
6093 (scalar_to_vector (loadf64 addr:$src))))))),
6094 (PMOVSXBWrm addr:$src)>;
6097 let Predicates = [HasAVX2] in {
6098 def : Pat<(v16i16 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBWYrr VR128:$src)>;
6099 def : Pat<(v8i32 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBDYrr VR128:$src)>;
6100 def : Pat<(v4i64 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBQYrr VR128:$src)>;
6102 def : Pat<(v8i32 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWDYrr VR128:$src)>;
6103 def : Pat<(v4i64 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWQYrr VR128:$src)>;
6105 def : Pat<(v4i64 (X86vzext (v4i32 VR128:$src))), (VPMOVZXDQYrr VR128:$src)>;
6107 def : Pat<(v16i16 (X86vzext (v32i8 VR256:$src))),
6108 (VPMOVZXBWYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
6109 def : Pat<(v8i32 (X86vzext (v32i8 VR256:$src))),
6110 (VPMOVZXBDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
6111 def : Pat<(v4i64 (X86vzext (v32i8 VR256:$src))),
6112 (VPMOVZXBQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
6114 def : Pat<(v8i32 (X86vzext (v16i16 VR256:$src))),
6115 (VPMOVZXWDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
6116 def : Pat<(v4i64 (X86vzext (v16i16 VR256:$src))),
6117 (VPMOVZXWQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
6119 def : Pat<(v4i64 (X86vzext (v8i32 VR256:$src))),
6120 (VPMOVZXDQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
6123 let Predicates = [HasAVX] in {
6124 def : Pat<(v8i16 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBWrr VR128:$src)>;
6125 def : Pat<(v4i32 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBDrr VR128:$src)>;
6126 def : Pat<(v2i64 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBQrr VR128:$src)>;
6128 def : Pat<(v4i32 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWDrr VR128:$src)>;
6129 def : Pat<(v2i64 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWQrr VR128:$src)>;
6131 def : Pat<(v2i64 (X86vzext (v4i32 VR128:$src))), (VPMOVZXDQrr VR128:$src)>;
6133 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
6134 (VPMOVZXBWrm addr:$src)>;
6135 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
6136 (VPMOVZXBWrm addr:$src)>;
6137 def : Pat<(v4i32 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
6138 (VPMOVZXBDrm addr:$src)>;
6139 def : Pat<(v2i64 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))),
6140 (VPMOVZXBQrm addr:$src)>;
6142 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
6143 (VPMOVZXWDrm addr:$src)>;
6144 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
6145 (VPMOVZXWDrm addr:$src)>;
6146 def : Pat<(v2i64 (X86vzext (v8i16 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
6147 (VPMOVZXWQrm addr:$src)>;
6149 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
6150 (VPMOVZXDQrm addr:$src)>;
6151 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
6152 (VPMOVZXDQrm addr:$src)>;
6153 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (X86vzload addr:$src)))))),
6154 (VPMOVZXDQrm addr:$src)>;
6156 def : Pat<(v8i16 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBWrr VR128:$src)>;
6157 def : Pat<(v4i32 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBDrr VR128:$src)>;
6158 def : Pat<(v2i64 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBQrr VR128:$src)>;
6160 def : Pat<(v4i32 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
6161 def : Pat<(v2i64 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWQrr VR128:$src)>;
6163 def : Pat<(v2i64 (X86vsext (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
6165 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64
6166 (scalar_to_vector (loadi64 addr:$src))))))),
6167 (VPMOVSXWDrm addr:$src)>;
6168 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64
6169 (scalar_to_vector (loadi64 addr:$src))))))),
6170 (VPMOVSXDQrm addr:$src)>;
6171 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64
6172 (scalar_to_vector (loadf64 addr:$src))))))),
6173 (VPMOVSXWDrm addr:$src)>;
6174 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64
6175 (scalar_to_vector (loadf64 addr:$src))))))),
6176 (VPMOVSXDQrm addr:$src)>;
6177 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64
6178 (scalar_to_vector (loadi64 addr:$src))))))),
6179 (VPMOVSXBWrm addr:$src)>;
6180 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64
6181 (scalar_to_vector (loadf64 addr:$src))))))),
6182 (VPMOVSXBWrm addr:$src)>;
6184 def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32
6185 (scalar_to_vector (loadi32 addr:$src))))))),
6186 (VPMOVSXBDrm addr:$src)>;
6187 def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32
6188 (scalar_to_vector (loadi32 addr:$src))))))),
6189 (VPMOVSXWQrm addr:$src)>;
6190 def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32
6191 (scalar_to_vector (extloadi32i16 addr:$src))))))),
6192 (VPMOVSXBQrm addr:$src)>;
6195 let Predicates = [UseSSE41] in {
6196 def : Pat<(v8i16 (X86vzext (v16i8 VR128:$src))), (PMOVZXBWrr VR128:$src)>;
6197 def : Pat<(v4i32 (X86vzext (v16i8 VR128:$src))), (PMOVZXBDrr VR128:$src)>;
6198 def : Pat<(v2i64 (X86vzext (v16i8 VR128:$src))), (PMOVZXBQrr VR128:$src)>;
6200 def : Pat<(v4i32 (X86vzext (v8i16 VR128:$src))), (PMOVZXWDrr VR128:$src)>;
6201 def : Pat<(v2i64 (X86vzext (v8i16 VR128:$src))), (PMOVZXWQrr VR128:$src)>;
6203 def : Pat<(v2i64 (X86vzext (v4i32 VR128:$src))), (PMOVZXDQrr VR128:$src)>;
6205 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
6206 (PMOVZXBWrm addr:$src)>;
6207 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
6208 (PMOVZXBWrm addr:$src)>;
6209 def : Pat<(v4i32 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
6210 (PMOVZXBDrm addr:$src)>;
6211 def : Pat<(v2i64 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))),
6212 (PMOVZXBQrm addr:$src)>;
6214 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
6215 (PMOVZXWDrm addr:$src)>;
6216 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
6217 (PMOVZXWDrm addr:$src)>;
6218 def : Pat<(v2i64 (X86vzext (v8i16 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
6219 (PMOVZXWQrm addr:$src)>;
6221 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
6222 (PMOVZXDQrm addr:$src)>;
6223 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
6224 (PMOVZXDQrm addr:$src)>;
6225 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (X86vzload addr:$src)))))),
6226 (PMOVZXDQrm addr:$src)>;
6229 //===----------------------------------------------------------------------===//
6230 // SSE4.1 - Extract Instructions
6231 //===----------------------------------------------------------------------===//
6233 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
6234 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
6235 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32orGR64:$dst),
6236 (ins VR128:$src1, i32i8imm:$src2),
6237 !strconcat(OpcodeStr,
6238 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6239 [(set GR32orGR64:$dst, (X86pextrb (v16i8 VR128:$src1),
6242 let neverHasSideEffects = 1, mayStore = 1 in
6243 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6244 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
6245 !strconcat(OpcodeStr,
6246 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6249 // There's an AssertZext in the way of writing the store pattern
6250 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
6253 let Predicates = [HasAVX] in
6254 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
6256 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
6259 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
6260 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
6261 let isCodeGenOnly = 1, hasSideEffects = 0 in
6262 def rr_REV : SS4AIi8<opc, MRMDestReg, (outs GR32orGR64:$dst),
6263 (ins VR128:$src1, i32i8imm:$src2),
6264 !strconcat(OpcodeStr,
6265 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6268 let neverHasSideEffects = 1, mayStore = 1 in
6269 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6270 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
6271 !strconcat(OpcodeStr,
6272 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6275 // There's an AssertZext in the way of writing the store pattern
6276 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
6279 let Predicates = [HasAVX] in
6280 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
6282 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
6285 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
6286 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
6287 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
6288 (ins VR128:$src1, i32i8imm:$src2),
6289 !strconcat(OpcodeStr,
6290 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6292 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
6293 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6294 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
6295 !strconcat(OpcodeStr,
6296 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6297 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
6298 addr:$dst)]>, OpSize;
6301 let Predicates = [HasAVX] in
6302 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
6304 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
6306 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
6307 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
6308 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
6309 (ins VR128:$src1, i32i8imm:$src2),
6310 !strconcat(OpcodeStr,
6311 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6313 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
6314 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6315 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
6316 !strconcat(OpcodeStr,
6317 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6318 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
6319 addr:$dst)]>, OpSize, REX_W;
6322 let Predicates = [HasAVX] in
6323 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
6325 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
6327 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
6329 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr,
6330 OpndItins itins = DEFAULT_ITINS> {
6331 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32orGR64:$dst),
6332 (ins VR128:$src1, i32i8imm:$src2),
6333 !strconcat(OpcodeStr,
6334 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6335 [(set GR32orGR64:$dst,
6336 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))],
6339 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6340 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
6341 !strconcat(OpcodeStr,
6342 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6343 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
6344 addr:$dst)], itins.rm>, OpSize;
6347 let ExeDomain = SSEPackedSingle in {
6348 let Predicates = [UseAVX] in
6349 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
6350 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps", SSE_EXTRACT_ITINS>;
6353 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
6354 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
6357 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
6359 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
6362 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
6363 Requires<[UseSSE41]>;
6365 //===----------------------------------------------------------------------===//
6366 // SSE4.1 - Insert Instructions
6367 //===----------------------------------------------------------------------===//
6369 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
6370 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6371 (ins VR128:$src1, GR32orGR64:$src2, i32i8imm:$src3),
6373 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6375 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6377 (X86pinsrb VR128:$src1, GR32orGR64:$src2, imm:$src3))]>, OpSize;
6378 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6379 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
6381 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6383 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6385 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
6386 imm:$src3))]>, OpSize;
6389 let Predicates = [HasAVX] in
6390 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
6391 let Constraints = "$src1 = $dst" in
6392 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
6394 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
6395 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6396 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
6398 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6400 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6402 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
6404 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6405 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
6407 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6409 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6411 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
6412 imm:$src3)))]>, OpSize;
6415 let Predicates = [HasAVX] in
6416 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
6417 let Constraints = "$src1 = $dst" in
6418 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
6420 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
6421 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6422 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
6424 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6426 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6428 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
6430 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6431 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
6433 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6435 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6437 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
6438 imm:$src3)))]>, OpSize;
6441 let Predicates = [HasAVX] in
6442 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
6443 let Constraints = "$src1 = $dst" in
6444 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
6446 // insertps has a few different modes, there's the first two here below which
6447 // are optimized inserts that won't zero arbitrary elements in the destination
6448 // vector. The next one matches the intrinsic and could zero arbitrary elements
6449 // in the target vector.
6450 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1,
6451 OpndItins itins = DEFAULT_ITINS> {
6452 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6453 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
6455 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6457 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6459 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))], itins.rr>,
6461 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6462 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
6464 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6466 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6468 (X86insrtps VR128:$src1,
6469 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
6470 imm:$src3))], itins.rm>, OpSize;
6473 let ExeDomain = SSEPackedSingle in {
6474 let Predicates = [UseAVX] in
6475 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
6476 let Constraints = "$src1 = $dst" in
6477 defm INSERTPS : SS41I_insertf32<0x21, "insertps", 1, SSE_INSERT_ITINS>;
6480 //===----------------------------------------------------------------------===//
6481 // SSE4.1 - Round Instructions
6482 //===----------------------------------------------------------------------===//
6484 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
6485 X86MemOperand x86memop, RegisterClass RC,
6486 PatFrag mem_frag32, PatFrag mem_frag64,
6487 Intrinsic V4F32Int, Intrinsic V2F64Int> {
6488 let ExeDomain = SSEPackedSingle in {
6489 // Intrinsic operation, reg.
6490 // Vector intrinsic operation, reg
6491 def PSr : SS4AIi8<opcps, MRMSrcReg,
6492 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6493 !strconcat(OpcodeStr,
6494 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6495 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))],
6496 IIC_SSE_ROUNDPS_REG>,
6499 // Vector intrinsic operation, mem
6500 def PSm : SS4AIi8<opcps, MRMSrcMem,
6501 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6502 !strconcat(OpcodeStr,
6503 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6505 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))],
6506 IIC_SSE_ROUNDPS_MEM>,
6508 } // ExeDomain = SSEPackedSingle
6510 let ExeDomain = SSEPackedDouble in {
6511 // Vector intrinsic operation, reg
6512 def PDr : SS4AIi8<opcpd, MRMSrcReg,
6513 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6514 !strconcat(OpcodeStr,
6515 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6516 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))],
6517 IIC_SSE_ROUNDPS_REG>,
6520 // Vector intrinsic operation, mem
6521 def PDm : SS4AIi8<opcpd, MRMSrcMem,
6522 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6523 !strconcat(OpcodeStr,
6524 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6526 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))],
6527 IIC_SSE_ROUNDPS_REG>,
6529 } // ExeDomain = SSEPackedDouble
6532 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
6535 Intrinsic F64Int, bit Is2Addr = 1> {
6536 let ExeDomain = GenericDomain in {
6538 let hasSideEffects = 0 in
6539 def SSr : SS4AIi8<opcss, MRMSrcReg,
6540 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3),
6542 !strconcat(OpcodeStr,
6543 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6544 !strconcat(OpcodeStr,
6545 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6548 // Intrinsic operation, reg.
6549 let isCodeGenOnly = 1 in
6550 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
6551 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6553 !strconcat(OpcodeStr,
6554 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6555 !strconcat(OpcodeStr,
6556 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6557 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6560 // Intrinsic operation, mem.
6561 def SSm : SS4AIi8<opcss, MRMSrcMem,
6562 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
6564 !strconcat(OpcodeStr,
6565 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6566 !strconcat(OpcodeStr,
6567 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6569 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
6573 let hasSideEffects = 0 in
6574 def SDr : SS4AIi8<opcsd, MRMSrcReg,
6575 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3),
6577 !strconcat(OpcodeStr,
6578 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6579 !strconcat(OpcodeStr,
6580 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6583 // Intrinsic operation, reg.
6584 let isCodeGenOnly = 1 in
6585 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
6586 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6588 !strconcat(OpcodeStr,
6589 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6590 !strconcat(OpcodeStr,
6591 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6592 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6595 // Intrinsic operation, mem.
6596 def SDm : SS4AIi8<opcsd, MRMSrcMem,
6597 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
6599 !strconcat(OpcodeStr,
6600 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6601 !strconcat(OpcodeStr,
6602 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6604 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
6606 } // ExeDomain = GenericDomain
6609 // FP round - roundss, roundps, roundsd, roundpd
6610 let Predicates = [HasAVX] in {
6612 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
6613 loadv4f32, loadv2f64,
6614 int_x86_sse41_round_ps,
6615 int_x86_sse41_round_pd>, VEX;
6616 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
6617 loadv8f32, loadv4f64,
6618 int_x86_avx_round_ps_256,
6619 int_x86_avx_round_pd_256>, VEX, VEX_L;
6620 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
6621 int_x86_sse41_round_ss,
6622 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
6624 def : Pat<(ffloor FR32:$src),
6625 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6626 def : Pat<(f64 (ffloor FR64:$src)),
6627 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6628 def : Pat<(f32 (fnearbyint FR32:$src)),
6629 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6630 def : Pat<(f64 (fnearbyint FR64:$src)),
6631 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6632 def : Pat<(f32 (fceil FR32:$src)),
6633 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6634 def : Pat<(f64 (fceil FR64:$src)),
6635 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6636 def : Pat<(f32 (frint FR32:$src)),
6637 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6638 def : Pat<(f64 (frint FR64:$src)),
6639 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6640 def : Pat<(f32 (ftrunc FR32:$src)),
6641 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6642 def : Pat<(f64 (ftrunc FR64:$src)),
6643 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6645 def : Pat<(v4f32 (ffloor VR128:$src)),
6646 (VROUNDPSr VR128:$src, (i32 0x1))>;
6647 def : Pat<(v4f32 (fnearbyint VR128:$src)),
6648 (VROUNDPSr VR128:$src, (i32 0xC))>;
6649 def : Pat<(v4f32 (fceil VR128:$src)),
6650 (VROUNDPSr VR128:$src, (i32 0x2))>;
6651 def : Pat<(v4f32 (frint VR128:$src)),
6652 (VROUNDPSr VR128:$src, (i32 0x4))>;
6653 def : Pat<(v4f32 (ftrunc VR128:$src)),
6654 (VROUNDPSr VR128:$src, (i32 0x3))>;
6656 def : Pat<(v2f64 (ffloor VR128:$src)),
6657 (VROUNDPDr VR128:$src, (i32 0x1))>;
6658 def : Pat<(v2f64 (fnearbyint VR128:$src)),
6659 (VROUNDPDr VR128:$src, (i32 0xC))>;
6660 def : Pat<(v2f64 (fceil VR128:$src)),
6661 (VROUNDPDr VR128:$src, (i32 0x2))>;
6662 def : Pat<(v2f64 (frint VR128:$src)),
6663 (VROUNDPDr VR128:$src, (i32 0x4))>;
6664 def : Pat<(v2f64 (ftrunc VR128:$src)),
6665 (VROUNDPDr VR128:$src, (i32 0x3))>;
6667 def : Pat<(v8f32 (ffloor VR256:$src)),
6668 (VROUNDYPSr VR256:$src, (i32 0x1))>;
6669 def : Pat<(v8f32 (fnearbyint VR256:$src)),
6670 (VROUNDYPSr VR256:$src, (i32 0xC))>;
6671 def : Pat<(v8f32 (fceil VR256:$src)),
6672 (VROUNDYPSr VR256:$src, (i32 0x2))>;
6673 def : Pat<(v8f32 (frint VR256:$src)),
6674 (VROUNDYPSr VR256:$src, (i32 0x4))>;
6675 def : Pat<(v8f32 (ftrunc VR256:$src)),
6676 (VROUNDYPSr VR256:$src, (i32 0x3))>;
6678 def : Pat<(v4f64 (ffloor VR256:$src)),
6679 (VROUNDYPDr VR256:$src, (i32 0x1))>;
6680 def : Pat<(v4f64 (fnearbyint VR256:$src)),
6681 (VROUNDYPDr VR256:$src, (i32 0xC))>;
6682 def : Pat<(v4f64 (fceil VR256:$src)),
6683 (VROUNDYPDr VR256:$src, (i32 0x2))>;
6684 def : Pat<(v4f64 (frint VR256:$src)),
6685 (VROUNDYPDr VR256:$src, (i32 0x4))>;
6686 def : Pat<(v4f64 (ftrunc VR256:$src)),
6687 (VROUNDYPDr VR256:$src, (i32 0x3))>;
6690 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
6691 memopv4f32, memopv2f64,
6692 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
6693 let Constraints = "$src1 = $dst" in
6694 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
6695 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
6697 let Predicates = [UseSSE41] in {
6698 def : Pat<(ffloor FR32:$src),
6699 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6700 def : Pat<(f64 (ffloor FR64:$src)),
6701 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6702 def : Pat<(f32 (fnearbyint FR32:$src)),
6703 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6704 def : Pat<(f64 (fnearbyint FR64:$src)),
6705 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6706 def : Pat<(f32 (fceil FR32:$src)),
6707 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6708 def : Pat<(f64 (fceil FR64:$src)),
6709 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6710 def : Pat<(f32 (frint FR32:$src)),
6711 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6712 def : Pat<(f64 (frint FR64:$src)),
6713 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6714 def : Pat<(f32 (ftrunc FR32:$src)),
6715 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6716 def : Pat<(f64 (ftrunc FR64:$src)),
6717 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6719 def : Pat<(v4f32 (ffloor VR128:$src)),
6720 (ROUNDPSr VR128:$src, (i32 0x1))>;
6721 def : Pat<(v4f32 (fnearbyint VR128:$src)),
6722 (ROUNDPSr VR128:$src, (i32 0xC))>;
6723 def : Pat<(v4f32 (fceil VR128:$src)),
6724 (ROUNDPSr VR128:$src, (i32 0x2))>;
6725 def : Pat<(v4f32 (frint VR128:$src)),
6726 (ROUNDPSr VR128:$src, (i32 0x4))>;
6727 def : Pat<(v4f32 (ftrunc VR128:$src)),
6728 (ROUNDPSr VR128:$src, (i32 0x3))>;
6730 def : Pat<(v2f64 (ffloor VR128:$src)),
6731 (ROUNDPDr VR128:$src, (i32 0x1))>;
6732 def : Pat<(v2f64 (fnearbyint VR128:$src)),
6733 (ROUNDPDr VR128:$src, (i32 0xC))>;
6734 def : Pat<(v2f64 (fceil VR128:$src)),
6735 (ROUNDPDr VR128:$src, (i32 0x2))>;
6736 def : Pat<(v2f64 (frint VR128:$src)),
6737 (ROUNDPDr VR128:$src, (i32 0x4))>;
6738 def : Pat<(v2f64 (ftrunc VR128:$src)),
6739 (ROUNDPDr VR128:$src, (i32 0x3))>;
6742 //===----------------------------------------------------------------------===//
6743 // SSE4.1 - Packed Bit Test
6744 //===----------------------------------------------------------------------===//
6746 // ptest instruction we'll lower to this in X86ISelLowering primarily from
6747 // the intel intrinsic that corresponds to this.
6748 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6749 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6750 "vptest\t{$src2, $src1|$src1, $src2}",
6751 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6753 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6754 "vptest\t{$src2, $src1|$src1, $src2}",
6755 [(set EFLAGS,(X86ptest VR128:$src1, (loadv2i64 addr:$src2)))]>,
6758 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
6759 "vptest\t{$src2, $src1|$src1, $src2}",
6760 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
6762 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
6763 "vptest\t{$src2, $src1|$src1, $src2}",
6764 [(set EFLAGS,(X86ptest VR256:$src1, (loadv4i64 addr:$src2)))]>,
6768 let Defs = [EFLAGS] in {
6769 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6770 "ptest\t{$src2, $src1|$src1, $src2}",
6771 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6773 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6774 "ptest\t{$src2, $src1|$src1, $src2}",
6775 [(set EFLAGS, (X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6779 // The bit test instructions below are AVX only
6780 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
6781 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
6782 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
6783 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6784 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
6785 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
6786 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6787 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
6791 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6792 let ExeDomain = SSEPackedSingle in {
6793 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, loadv4f32, v4f32>;
6794 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, loadv8f32, v8f32>,
6797 let ExeDomain = SSEPackedDouble in {
6798 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, loadv2f64, v2f64>;
6799 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, loadv4f64, v4f64>,
6804 //===----------------------------------------------------------------------===//
6805 // SSE4.1 - Misc Instructions
6806 //===----------------------------------------------------------------------===//
6808 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
6809 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
6810 "popcnt{w}\t{$src, $dst|$dst, $src}",
6811 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)],
6814 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
6815 "popcnt{w}\t{$src, $dst|$dst, $src}",
6816 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
6817 (implicit EFLAGS)], IIC_SSE_POPCNT_RM>, OpSize, XS;
6819 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
6820 "popcnt{l}\t{$src, $dst|$dst, $src}",
6821 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)],
6824 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
6825 "popcnt{l}\t{$src, $dst|$dst, $src}",
6826 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
6827 (implicit EFLAGS)], IIC_SSE_POPCNT_RM>, XS;
6829 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
6830 "popcnt{q}\t{$src, $dst|$dst, $src}",
6831 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)],
6834 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
6835 "popcnt{q}\t{$src, $dst|$dst, $src}",
6836 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
6837 (implicit EFLAGS)], IIC_SSE_POPCNT_RM>, XS;
6842 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
6843 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
6844 Intrinsic IntId128> {
6845 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6847 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6848 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
6849 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6851 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6854 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
6857 let Predicates = [HasAVX] in
6858 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
6859 int_x86_sse41_phminposuw>, VEX;
6860 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
6861 int_x86_sse41_phminposuw>;
6863 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6864 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
6865 Intrinsic IntId128, bit Is2Addr = 1,
6866 OpndItins itins = DEFAULT_ITINS> {
6867 let isCommutable = 1 in
6868 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6869 (ins VR128:$src1, VR128:$src2),
6871 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6872 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6873 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))],
6875 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6876 (ins VR128:$src1, i128mem:$src2),
6878 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6879 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6881 (IntId128 VR128:$src1,
6882 (bitconvert (memopv2i64 addr:$src2))))],
6886 /// SS41I_binop_rm_int_y - Simple SSE 4.1 binary operator
6887 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6888 Intrinsic IntId256> {
6889 let isCommutable = 1 in
6890 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6891 (ins VR256:$src1, VR256:$src2),
6892 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6893 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
6894 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6895 (ins VR256:$src1, i256mem:$src2),
6896 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6898 (IntId256 VR256:$src1,
6899 (bitconvert (loadv4i64 addr:$src2))))]>, OpSize;
6903 /// SS48I_binop_rm - Simple SSE41 binary operator.
6904 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6905 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6906 X86MemOperand x86memop, bit Is2Addr = 1,
6907 OpndItins itins = DEFAULT_ITINS> {
6908 let isCommutable = 1 in
6909 def rr : SS48I<opc, MRMSrcReg, (outs RC:$dst),
6910 (ins RC:$src1, RC:$src2),
6912 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6913 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6914 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>, OpSize;
6915 def rm : SS48I<opc, MRMSrcMem, (outs RC:$dst),
6916 (ins RC:$src1, x86memop:$src2),
6918 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6919 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6921 (OpVT (OpNode RC:$src1,
6922 (bitconvert (memop_frag addr:$src2)))))]>, OpSize;
6925 let Predicates = [HasAVX] in {
6926 let isCommutable = 0 in
6927 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
6929 defm VPMINSB : SS48I_binop_rm<0x38, "vpminsb", X86smin, v16i8, VR128,
6930 loadv2i64, i128mem, 0>, VEX_4V;
6931 defm VPMINSD : SS48I_binop_rm<0x39, "vpminsd", X86smin, v4i32, VR128,
6932 loadv2i64, i128mem, 0>, VEX_4V;
6933 defm VPMINUD : SS48I_binop_rm<0x3B, "vpminud", X86umin, v4i32, VR128,
6934 loadv2i64, i128mem, 0>, VEX_4V;
6935 defm VPMINUW : SS48I_binop_rm<0x3A, "vpminuw", X86umin, v8i16, VR128,
6936 loadv2i64, i128mem, 0>, VEX_4V;
6937 defm VPMAXSB : SS48I_binop_rm<0x3C, "vpmaxsb", X86smax, v16i8, VR128,
6938 loadv2i64, i128mem, 0>, VEX_4V;
6939 defm VPMAXSD : SS48I_binop_rm<0x3D, "vpmaxsd", X86smax, v4i32, VR128,
6940 loadv2i64, i128mem, 0>, VEX_4V;
6941 defm VPMAXUD : SS48I_binop_rm<0x3F, "vpmaxud", X86umax, v4i32, VR128,
6942 loadv2i64, i128mem, 0>, VEX_4V;
6943 defm VPMAXUW : SS48I_binop_rm<0x3E, "vpmaxuw", X86umax, v8i16, VR128,
6944 loadv2i64, i128mem, 0>, VEX_4V;
6945 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
6949 let Predicates = [HasAVX2] in {
6950 let isCommutable = 0 in
6951 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
6952 int_x86_avx2_packusdw>, VEX_4V, VEX_L;
6953 defm VPMINSBY : SS48I_binop_rm<0x38, "vpminsb", X86smin, v32i8, VR256,
6954 loadv4i64, i256mem, 0>, VEX_4V, VEX_L;
6955 defm VPMINSDY : SS48I_binop_rm<0x39, "vpminsd", X86smin, v8i32, VR256,
6956 loadv4i64, i256mem, 0>, VEX_4V, VEX_L;
6957 defm VPMINUDY : SS48I_binop_rm<0x3B, "vpminud", X86umin, v8i32, VR256,
6958 loadv4i64, i256mem, 0>, VEX_4V, VEX_L;
6959 defm VPMINUWY : SS48I_binop_rm<0x3A, "vpminuw", X86umin, v16i16, VR256,
6960 loadv4i64, i256mem, 0>, VEX_4V, VEX_L;
6961 defm VPMAXSBY : SS48I_binop_rm<0x3C, "vpmaxsb", X86smax, v32i8, VR256,
6962 loadv4i64, i256mem, 0>, VEX_4V, VEX_L;
6963 defm VPMAXSDY : SS48I_binop_rm<0x3D, "vpmaxsd", X86smax, v8i32, VR256,
6964 loadv4i64, i256mem, 0>, VEX_4V, VEX_L;
6965 defm VPMAXUDY : SS48I_binop_rm<0x3F, "vpmaxud", X86umax, v8i32, VR256,
6966 loadv4i64, i256mem, 0>, VEX_4V, VEX_L;
6967 defm VPMAXUWY : SS48I_binop_rm<0x3E, "vpmaxuw", X86umax, v16i16, VR256,
6968 loadv4i64, i256mem, 0>, VEX_4V, VEX_L;
6969 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
6970 int_x86_avx2_pmul_dq>, VEX_4V, VEX_L;
6973 let Constraints = "$src1 = $dst" in {
6974 let isCommutable = 0 in
6975 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
6976 defm PMINSB : SS48I_binop_rm<0x38, "pminsb", X86smin, v16i8, VR128,
6977 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6978 defm PMINSD : SS48I_binop_rm<0x39, "pminsd", X86smin, v4i32, VR128,
6979 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6980 defm PMINUD : SS48I_binop_rm<0x3B, "pminud", X86umin, v4i32, VR128,
6981 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6982 defm PMINUW : SS48I_binop_rm<0x3A, "pminuw", X86umin, v8i16, VR128,
6983 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6984 defm PMAXSB : SS48I_binop_rm<0x3C, "pmaxsb", X86smax, v16i8, VR128,
6985 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6986 defm PMAXSD : SS48I_binop_rm<0x3D, "pmaxsd", X86smax, v4i32, VR128,
6987 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6988 defm PMAXUD : SS48I_binop_rm<0x3F, "pmaxud", X86umax, v4i32, VR128,
6989 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6990 defm PMAXUW : SS48I_binop_rm<0x3E, "pmaxuw", X86umax, v8i16, VR128,
6991 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6992 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq,
6993 1, SSE_INTMUL_ITINS_P>;
6996 let Predicates = [HasAVX] in {
6997 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, VR128,
6998 memopv2i64, i128mem, 0>, VEX_4V;
6999 defm VPCMPEQQ : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v2i64, VR128,
7000 memopv2i64, i128mem, 0>, VEX_4V;
7002 let Predicates = [HasAVX2] in {
7003 defm VPMULLDY : SS48I_binop_rm<0x40, "vpmulld", mul, v8i32, VR256,
7004 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
7005 defm VPCMPEQQY : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v4i64, VR256,
7006 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
7009 let Constraints = "$src1 = $dst" in {
7010 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32, VR128,
7011 memopv2i64, i128mem, 1, SSE_PMULLD_ITINS>;
7012 defm PCMPEQQ : SS48I_binop_rm<0x29, "pcmpeqq", X86pcmpeq, v2i64, VR128,
7013 memopv2i64, i128mem, 1, SSE_INTALUQ_ITINS_P>;
7016 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
7017 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
7018 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
7019 X86MemOperand x86memop, bit Is2Addr = 1,
7020 OpndItins itins = DEFAULT_ITINS> {
7021 let isCommutable = 1 in
7022 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
7023 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
7025 !strconcat(OpcodeStr,
7026 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
7027 !strconcat(OpcodeStr,
7028 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
7029 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))], itins.rr>,
7031 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
7032 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
7034 !strconcat(OpcodeStr,
7035 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
7036 !strconcat(OpcodeStr,
7037 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
7040 (bitconvert (memop_frag addr:$src2)), imm:$src3))], itins.rm>,
7044 let Predicates = [HasAVX] in {
7045 let isCommutable = 0 in {
7046 let ExeDomain = SSEPackedSingle in {
7047 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
7048 VR128, loadv4f32, f128mem, 0>, VEX_4V;
7049 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
7050 int_x86_avx_blend_ps_256, VR256, loadv8f32,
7051 f256mem, 0>, VEX_4V, VEX_L;
7053 let ExeDomain = SSEPackedDouble in {
7054 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
7055 VR128, loadv2f64, f128mem, 0>, VEX_4V;
7056 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
7057 int_x86_avx_blend_pd_256,VR256, loadv4f64,
7058 f256mem, 0>, VEX_4V, VEX_L;
7060 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
7061 VR128, loadv2i64, i128mem, 0>, VEX_4V;
7062 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
7063 VR128, loadv2i64, i128mem, 0>, VEX_4V;
7065 let ExeDomain = SSEPackedSingle in
7066 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
7067 VR128, loadv4f32, f128mem, 0>, VEX_4V;
7068 let ExeDomain = SSEPackedDouble in
7069 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
7070 VR128, loadv2f64, f128mem, 0>, VEX_4V;
7071 let ExeDomain = SSEPackedSingle in
7072 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
7073 VR256, loadv8f32, i256mem, 0>, VEX_4V, VEX_L;
7076 let Predicates = [HasAVX2] in {
7077 let isCommutable = 0 in {
7078 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
7079 VR256, loadv4i64, i256mem, 0>, VEX_4V, VEX_L;
7080 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
7081 VR256, loadv4i64, i256mem, 0>, VEX_4V, VEX_L;
7085 let Constraints = "$src1 = $dst" in {
7086 let isCommutable = 0 in {
7087 let ExeDomain = SSEPackedSingle in
7088 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
7089 VR128, memopv4f32, f128mem,
7090 1, SSE_INTALU_ITINS_P>;
7091 let ExeDomain = SSEPackedDouble in
7092 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
7093 VR128, memopv2f64, f128mem,
7094 1, SSE_INTALU_ITINS_P>;
7095 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
7096 VR128, memopv2i64, i128mem,
7097 1, SSE_INTALU_ITINS_P>;
7098 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
7099 VR128, memopv2i64, i128mem,
7100 1, SSE_INTMUL_ITINS_P>;
7102 let ExeDomain = SSEPackedSingle in
7103 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
7104 VR128, memopv4f32, f128mem, 1,
7106 let ExeDomain = SSEPackedDouble in
7107 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
7108 VR128, memopv2f64, f128mem, 1,
7112 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
7113 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
7114 RegisterClass RC, X86MemOperand x86memop,
7115 PatFrag mem_frag, Intrinsic IntId> {
7116 def rr : Ii8<opc, MRMSrcReg, (outs RC:$dst),
7117 (ins RC:$src1, RC:$src2, RC:$src3),
7118 !strconcat(OpcodeStr,
7119 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7120 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
7121 NoItinerary, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
7123 def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
7124 (ins RC:$src1, x86memop:$src2, RC:$src3),
7125 !strconcat(OpcodeStr,
7126 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7128 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
7130 NoItinerary, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
7133 let Predicates = [HasAVX] in {
7134 let ExeDomain = SSEPackedDouble in {
7135 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, f128mem,
7136 loadv2f64, int_x86_sse41_blendvpd>;
7137 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, f256mem,
7138 loadv4f64, int_x86_avx_blendv_pd_256>, VEX_L;
7139 } // ExeDomain = SSEPackedDouble
7140 let ExeDomain = SSEPackedSingle in {
7141 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, f128mem,
7142 loadv4f32, int_x86_sse41_blendvps>;
7143 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, f256mem,
7144 loadv8f32, int_x86_avx_blendv_ps_256>, VEX_L;
7145 } // ExeDomain = SSEPackedSingle
7146 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
7147 loadv2i64, int_x86_sse41_pblendvb>;
7150 let Predicates = [HasAVX2] in {
7151 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
7152 loadv4i64, int_x86_avx2_pblendvb>, VEX_L;
7155 let Predicates = [HasAVX] in {
7156 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
7157 (v16i8 VR128:$src2))),
7158 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
7159 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
7160 (v4i32 VR128:$src2))),
7161 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
7162 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
7163 (v4f32 VR128:$src2))),
7164 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
7165 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
7166 (v2i64 VR128:$src2))),
7167 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
7168 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
7169 (v2f64 VR128:$src2))),
7170 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
7171 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
7172 (v8i32 VR256:$src2))),
7173 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
7174 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
7175 (v8f32 VR256:$src2))),
7176 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
7177 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
7178 (v4i64 VR256:$src2))),
7179 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
7180 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
7181 (v4f64 VR256:$src2))),
7182 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
7184 def : Pat<(v8f32 (X86Blendi (v8f32 VR256:$src1), (v8f32 VR256:$src2),
7186 (VBLENDPSYrri VR256:$src1, VR256:$src2, imm:$mask)>;
7187 def : Pat<(v4f64 (X86Blendi (v4f64 VR256:$src1), (v4f64 VR256:$src2),
7189 (VBLENDPDYrri VR256:$src1, VR256:$src2, imm:$mask)>;
7191 def : Pat<(v8i16 (X86Blendi (v8i16 VR128:$src1), (v8i16 VR128:$src2),
7193 (VPBLENDWrri VR128:$src1, VR128:$src2, imm:$mask)>;
7194 def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$src1), (v4f32 VR128:$src2),
7196 (VBLENDPSrri VR128:$src1, VR128:$src2, imm:$mask)>;
7197 def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$src1), (v2f64 VR128:$src2),
7199 (VBLENDPDrri VR128:$src1, VR128:$src2, imm:$mask)>;
7202 let Predicates = [HasAVX2] in {
7203 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
7204 (v32i8 VR256:$src2))),
7205 (VPBLENDVBYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
7206 def : Pat<(v16i16 (X86Blendi (v16i16 VR256:$src1), (v16i16 VR256:$src2),
7208 (VPBLENDWYrri VR256:$src1, VR256:$src2, imm:$mask)>;
7211 /// SS41I_ternary_int - SSE 4.1 ternary operator
7212 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
7213 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7214 X86MemOperand x86memop, Intrinsic IntId,
7215 OpndItins itins = DEFAULT_ITINS> {
7216 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
7217 (ins VR128:$src1, VR128:$src2),
7218 !strconcat(OpcodeStr,
7219 "\t{$src2, $dst|$dst, $src2}"),
7220 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))],
7223 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
7224 (ins VR128:$src1, x86memop:$src2),
7225 !strconcat(OpcodeStr,
7226 "\t{$src2, $dst|$dst, $src2}"),
7229 (bitconvert (mem_frag addr:$src2)), XMM0))],
7234 let ExeDomain = SSEPackedDouble in
7235 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64, f128mem,
7236 int_x86_sse41_blendvpd>;
7237 let ExeDomain = SSEPackedSingle in
7238 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32, f128mem,
7239 int_x86_sse41_blendvps>;
7240 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64, i128mem,
7241 int_x86_sse41_pblendvb>;
7243 // Aliases with the implicit xmm0 argument
7244 def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7245 (BLENDVPDrr0 VR128:$dst, VR128:$src2)>;
7246 def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7247 (BLENDVPDrm0 VR128:$dst, f128mem:$src2)>;
7248 def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7249 (BLENDVPSrr0 VR128:$dst, VR128:$src2)>;
7250 def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7251 (BLENDVPSrm0 VR128:$dst, f128mem:$src2)>;
7252 def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7253 (PBLENDVBrr0 VR128:$dst, VR128:$src2)>;
7254 def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7255 (PBLENDVBrm0 VR128:$dst, i128mem:$src2)>;
7257 let Predicates = [UseSSE41] in {
7258 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
7259 (v16i8 VR128:$src2))),
7260 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
7261 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
7262 (v4i32 VR128:$src2))),
7263 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
7264 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
7265 (v4f32 VR128:$src2))),
7266 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
7267 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
7268 (v2i64 VR128:$src2))),
7269 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
7270 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
7271 (v2f64 VR128:$src2))),
7272 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
7274 def : Pat<(v8i16 (X86Blendi (v8i16 VR128:$src1), (v8i16 VR128:$src2),
7276 (PBLENDWrri VR128:$src1, VR128:$src2, imm:$mask)>;
7277 def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$src1), (v4f32 VR128:$src2),
7279 (BLENDPSrri VR128:$src1, VR128:$src2, imm:$mask)>;
7280 def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$src1), (v2f64 VR128:$src2),
7282 (BLENDPDrri VR128:$src1, VR128:$src2, imm:$mask)>;
7286 let Predicates = [HasAVX] in
7287 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
7288 "vmovntdqa\t{$src, $dst|$dst, $src}",
7289 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
7291 let Predicates = [HasAVX2] in
7292 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
7293 "vmovntdqa\t{$src, $dst|$dst, $src}",
7294 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
7296 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
7297 "movntdqa\t{$src, $dst|$dst, $src}",
7298 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
7301 //===----------------------------------------------------------------------===//
7302 // SSE4.2 - Compare Instructions
7303 //===----------------------------------------------------------------------===//
7305 /// SS42I_binop_rm - Simple SSE 4.2 binary operator
7306 multiclass SS42I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
7307 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
7308 X86MemOperand x86memop, bit Is2Addr = 1> {
7309 def rr : SS428I<opc, MRMSrcReg, (outs RC:$dst),
7310 (ins RC:$src1, RC:$src2),
7312 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7313 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7314 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>,
7316 def rm : SS428I<opc, MRMSrcMem, (outs RC:$dst),
7317 (ins RC:$src1, x86memop:$src2),
7319 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7320 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7322 (OpVT (OpNode RC:$src1, (memop_frag addr:$src2))))]>, OpSize;
7325 let Predicates = [HasAVX] in
7326 defm VPCMPGTQ : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v2i64, VR128,
7327 loadv2i64, i128mem, 0>, VEX_4V;
7329 let Predicates = [HasAVX2] in
7330 defm VPCMPGTQY : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v4i64, VR256,
7331 loadv4i64, i256mem, 0>, VEX_4V, VEX_L;
7333 let Constraints = "$src1 = $dst" in
7334 defm PCMPGTQ : SS42I_binop_rm<0x37, "pcmpgtq", X86pcmpgt, v2i64, VR128,
7335 memopv2i64, i128mem>;
7337 //===----------------------------------------------------------------------===//
7338 // SSE4.2 - String/text Processing Instructions
7339 //===----------------------------------------------------------------------===//
7341 // Packed Compare Implicit Length Strings, Return Mask
7342 multiclass pseudo_pcmpistrm<string asm> {
7343 def REG : PseudoI<(outs VR128:$dst),
7344 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7345 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
7347 def MEM : PseudoI<(outs VR128:$dst),
7348 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7349 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1,
7350 (bc_v16i8 (memopv2i64 addr:$src2)), imm:$src3))]>;
7353 let Defs = [EFLAGS], usesCustomInserter = 1 in {
7354 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
7355 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[UseSSE42]>;
7358 multiclass pcmpistrm_SS42AI<string asm> {
7359 def rr : SS42AI<0x62, MRMSrcReg, (outs),
7360 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7361 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
7364 def rm :SS42AI<0x62, MRMSrcMem, (outs),
7365 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7366 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
7370 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
7371 let Predicates = [HasAVX] in
7372 defm VPCMPISTRM128 : pcmpistrm_SS42AI<"vpcmpistrm">, VEX;
7373 defm PCMPISTRM128 : pcmpistrm_SS42AI<"pcmpistrm"> ;
7376 // Packed Compare Explicit Length Strings, Return Mask
7377 multiclass pseudo_pcmpestrm<string asm> {
7378 def REG : PseudoI<(outs VR128:$dst),
7379 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7380 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
7381 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
7382 def MEM : PseudoI<(outs VR128:$dst),
7383 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7384 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128 VR128:$src1, EAX,
7385 (bc_v16i8 (memopv2i64 addr:$src3)), EDX, imm:$src5))]>;
7388 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
7389 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
7390 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[UseSSE42]>;
7393 multiclass SS42AI_pcmpestrm<string asm> {
7394 def rr : SS42AI<0x60, MRMSrcReg, (outs),
7395 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7396 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7399 def rm : SS42AI<0x60, MRMSrcMem, (outs),
7400 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7401 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7405 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
7406 let Predicates = [HasAVX] in
7407 defm VPCMPESTRM128 : SS42AI_pcmpestrm<"vpcmpestrm">, VEX;
7408 defm PCMPESTRM128 : SS42AI_pcmpestrm<"pcmpestrm">;
7411 // Packed Compare Implicit Length Strings, Return Index
7412 multiclass pseudo_pcmpistri<string asm> {
7413 def REG : PseudoI<(outs GR32:$dst),
7414 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7415 [(set GR32:$dst, EFLAGS,
7416 (X86pcmpistri VR128:$src1, VR128:$src2, imm:$src3))]>;
7417 def MEM : PseudoI<(outs GR32:$dst),
7418 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7419 [(set GR32:$dst, EFLAGS, (X86pcmpistri VR128:$src1,
7420 (bc_v16i8 (memopv2i64 addr:$src2)), imm:$src3))]>;
7423 let Defs = [EFLAGS], usesCustomInserter = 1 in {
7424 defm VPCMPISTRI : pseudo_pcmpistri<"#VPCMPISTRI">, Requires<[HasAVX]>;
7425 defm PCMPISTRI : pseudo_pcmpistri<"#PCMPISTRI">, Requires<[UseSSE42]>;
7428 multiclass SS42AI_pcmpistri<string asm> {
7429 def rr : SS42AI<0x63, MRMSrcReg, (outs),
7430 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7431 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
7434 def rm : SS42AI<0x63, MRMSrcMem, (outs),
7435 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7436 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
7440 let Defs = [ECX, EFLAGS], neverHasSideEffects = 1 in {
7441 let Predicates = [HasAVX] in
7442 defm VPCMPISTRI : SS42AI_pcmpistri<"vpcmpistri">, VEX;
7443 defm PCMPISTRI : SS42AI_pcmpistri<"pcmpistri">;
7446 // Packed Compare Explicit Length Strings, Return Index
7447 multiclass pseudo_pcmpestri<string asm> {
7448 def REG : PseudoI<(outs GR32:$dst),
7449 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7450 [(set GR32:$dst, EFLAGS,
7451 (X86pcmpestri VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
7452 def MEM : PseudoI<(outs GR32:$dst),
7453 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7454 [(set GR32:$dst, EFLAGS,
7455 (X86pcmpestri VR128:$src1, EAX, (bc_v16i8 (memopv2i64 addr:$src3)), EDX,
7459 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
7460 defm VPCMPESTRI : pseudo_pcmpestri<"#VPCMPESTRI">, Requires<[HasAVX]>;
7461 defm PCMPESTRI : pseudo_pcmpestri<"#PCMPESTRI">, Requires<[UseSSE42]>;
7464 multiclass SS42AI_pcmpestri<string asm> {
7465 def rr : SS42AI<0x61, MRMSrcReg, (outs),
7466 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7467 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7470 def rm : SS42AI<0x61, MRMSrcMem, (outs),
7471 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7472 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7476 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
7477 let Predicates = [HasAVX] in
7478 defm VPCMPESTRI : SS42AI_pcmpestri<"vpcmpestri">, VEX;
7479 defm PCMPESTRI : SS42AI_pcmpestri<"pcmpestri">;
7482 //===----------------------------------------------------------------------===//
7483 // SSE4.2 - CRC Instructions
7484 //===----------------------------------------------------------------------===//
7486 // No CRC instructions have AVX equivalents
7488 // crc intrinsic instruction
7489 // This set of instructions are only rm, the only difference is the size
7491 class SS42I_crc32r<bits<8> opc, string asm, RegisterClass RCOut,
7492 RegisterClass RCIn, SDPatternOperator Int> :
7493 SS42FI<opc, MRMSrcReg, (outs RCOut:$dst), (ins RCOut:$src1, RCIn:$src2),
7494 !strconcat(asm, "\t{$src2, $src1|$src1, $src2}"),
7495 [(set RCOut:$dst, (Int RCOut:$src1, RCIn:$src2))], IIC_CRC32_REG>;
7497 class SS42I_crc32m<bits<8> opc, string asm, RegisterClass RCOut,
7498 X86MemOperand x86memop, SDPatternOperator Int> :
7499 SS42FI<opc, MRMSrcMem, (outs RCOut:$dst), (ins RCOut:$src1, x86memop:$src2),
7500 !strconcat(asm, "\t{$src2, $src1|$src1, $src2}"),
7501 [(set RCOut:$dst, (Int RCOut:$src1, (load addr:$src2)))],
7504 let Constraints = "$src1 = $dst" in {
7505 def CRC32r32m8 : SS42I_crc32m<0xF0, "crc32{b}", GR32, i8mem,
7506 int_x86_sse42_crc32_32_8>;
7507 def CRC32r32r8 : SS42I_crc32r<0xF0, "crc32{b}", GR32, GR8,
7508 int_x86_sse42_crc32_32_8>;
7509 def CRC32r32m16 : SS42I_crc32m<0xF1, "crc32{w}", GR32, i16mem,
7510 int_x86_sse42_crc32_32_16>, OpSize;
7511 def CRC32r32r16 : SS42I_crc32r<0xF1, "crc32{w}", GR32, GR16,
7512 int_x86_sse42_crc32_32_16>, OpSize;
7513 def CRC32r32m32 : SS42I_crc32m<0xF1, "crc32{l}", GR32, i32mem,
7514 int_x86_sse42_crc32_32_32>;
7515 def CRC32r32r32 : SS42I_crc32r<0xF1, "crc32{l}", GR32, GR32,
7516 int_x86_sse42_crc32_32_32>;
7517 def CRC32r64m64 : SS42I_crc32m<0xF1, "crc32{q}", GR64, i64mem,
7518 int_x86_sse42_crc32_64_64>, REX_W;
7519 def CRC32r64r64 : SS42I_crc32r<0xF1, "crc32{q}", GR64, GR64,
7520 int_x86_sse42_crc32_64_64>, REX_W;
7521 let hasSideEffects = 0 in {
7523 def CRC32r64m8 : SS42I_crc32m<0xF0, "crc32{b}", GR64, i8mem,
7525 def CRC32r64r8 : SS42I_crc32r<0xF0, "crc32{b}", GR64, GR8,
7530 //===----------------------------------------------------------------------===//
7531 // SHA-NI Instructions
7532 //===----------------------------------------------------------------------===//
7534 multiclass SHAI_binop<bits<8> Opc, string OpcodeStr, Intrinsic IntId,
7536 def rr : I<Opc, MRMSrcReg, (outs VR128:$dst),
7537 (ins VR128:$src1, VR128:$src2),
7538 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7540 (set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0)),
7541 (set VR128:$dst, (IntId VR128:$src1, VR128:$src2)))]>, T8;
7543 def rm : I<Opc, MRMSrcMem, (outs VR128:$dst),
7544 (ins VR128:$src1, i128mem:$src2),
7545 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7547 (set VR128:$dst, (IntId VR128:$src1,
7548 (bc_v4i32 (memopv2i64 addr:$src2)), XMM0)),
7549 (set VR128:$dst, (IntId VR128:$src1,
7550 (bc_v4i32 (memopv2i64 addr:$src2)))))]>, T8;
7553 let Constraints = "$src1 = $dst", Predicates = [HasSHA] in {
7554 def SHA1RNDS4rri : Ii8<0xCC, MRMSrcReg, (outs VR128:$dst),
7555 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7556 "sha1rnds4\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7558 (int_x86_sha1rnds4 VR128:$src1, VR128:$src2,
7559 (i8 imm:$src3)))]>, TA;
7560 def SHA1RNDS4rmi : Ii8<0xCC, MRMSrcMem, (outs VR128:$dst),
7561 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7562 "sha1rnds4\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7564 (int_x86_sha1rnds4 VR128:$src1,
7565 (bc_v4i32 (memopv2i64 addr:$src2)),
7566 (i8 imm:$src3)))]>, TA;
7568 defm SHA1NEXTE : SHAI_binop<0xC8, "sha1nexte", int_x86_sha1nexte>;
7569 defm SHA1MSG1 : SHAI_binop<0xC9, "sha1msg1", int_x86_sha1msg1>;
7570 defm SHA1MSG2 : SHAI_binop<0xCA, "sha1msg2", int_x86_sha1msg2>;
7573 defm SHA256RNDS2 : SHAI_binop<0xCB, "sha256rnds2", int_x86_sha256rnds2, 1>;
7575 defm SHA256MSG1 : SHAI_binop<0xCC, "sha256msg1", int_x86_sha256msg1>;
7576 defm SHA256MSG2 : SHAI_binop<0xCD, "sha256msg2", int_x86_sha256msg2>;
7579 // Aliases with explicit %xmm0
7580 def : InstAlias<"sha256rnds2\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7581 (SHA256RNDS2rr VR128:$dst, VR128:$src2)>;
7582 def : InstAlias<"sha256rnds2\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7583 (SHA256RNDS2rm VR128:$dst, i128mem:$src2)>;
7585 //===----------------------------------------------------------------------===//
7586 // AES-NI Instructions
7587 //===----------------------------------------------------------------------===//
7589 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
7590 Intrinsic IntId128, bit Is2Addr = 1> {
7591 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
7592 (ins VR128:$src1, VR128:$src2),
7594 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7595 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7596 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
7598 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
7599 (ins VR128:$src1, i128mem:$src2),
7601 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7602 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7604 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
7607 // Perform One Round of an AES Encryption/Decryption Flow
7608 let Predicates = [HasAVX, HasAES] in {
7609 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
7610 int_x86_aesni_aesenc, 0>, VEX_4V;
7611 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
7612 int_x86_aesni_aesenclast, 0>, VEX_4V;
7613 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
7614 int_x86_aesni_aesdec, 0>, VEX_4V;
7615 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
7616 int_x86_aesni_aesdeclast, 0>, VEX_4V;
7619 let Constraints = "$src1 = $dst" in {
7620 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
7621 int_x86_aesni_aesenc>;
7622 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
7623 int_x86_aesni_aesenclast>;
7624 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
7625 int_x86_aesni_aesdec>;
7626 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
7627 int_x86_aesni_aesdeclast>;
7630 // Perform the AES InvMixColumn Transformation
7631 let Predicates = [HasAVX, HasAES] in {
7632 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7634 "vaesimc\t{$src1, $dst|$dst, $src1}",
7636 (int_x86_aesni_aesimc VR128:$src1))]>,
7638 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7639 (ins i128mem:$src1),
7640 "vaesimc\t{$src1, $dst|$dst, $src1}",
7641 [(set VR128:$dst, (int_x86_aesni_aesimc (loadv2i64 addr:$src1)))]>,
7644 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7646 "aesimc\t{$src1, $dst|$dst, $src1}",
7648 (int_x86_aesni_aesimc VR128:$src1))]>,
7650 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7651 (ins i128mem:$src1),
7652 "aesimc\t{$src1, $dst|$dst, $src1}",
7653 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7656 // AES Round Key Generation Assist
7657 let Predicates = [HasAVX, HasAES] in {
7658 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7659 (ins VR128:$src1, i8imm:$src2),
7660 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7662 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7664 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7665 (ins i128mem:$src1, i8imm:$src2),
7666 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7668 (int_x86_aesni_aeskeygenassist (loadv2i64 addr:$src1), imm:$src2))]>,
7671 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7672 (ins VR128:$src1, i8imm:$src2),
7673 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7675 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7677 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7678 (ins i128mem:$src1, i8imm:$src2),
7679 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7681 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7684 //===----------------------------------------------------------------------===//
7685 // PCLMUL Instructions
7686 //===----------------------------------------------------------------------===//
7688 // AVX carry-less Multiplication instructions
7689 def VPCLMULQDQrr : AVXPCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7690 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7691 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7693 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
7695 def VPCLMULQDQrm : AVXPCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7696 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7697 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7698 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7699 (loadv2i64 addr:$src2), imm:$src3))]>;
7701 // Carry-less Multiplication instructions
7702 let Constraints = "$src1 = $dst" in {
7703 def PCLMULQDQrr : PCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7704 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7705 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7707 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))],
7708 IIC_SSE_PCLMULQDQ_RR>;
7710 def PCLMULQDQrm : PCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7711 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7712 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7713 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7714 (memopv2i64 addr:$src2), imm:$src3))],
7715 IIC_SSE_PCLMULQDQ_RM>;
7716 } // Constraints = "$src1 = $dst"
7719 multiclass pclmul_alias<string asm, int immop> {
7720 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7721 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
7723 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7724 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
7726 def : InstAlias<!strconcat("vpclmul", asm,
7727 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7728 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
7730 def : InstAlias<!strconcat("vpclmul", asm,
7731 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7732 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
7734 defm : pclmul_alias<"hqhq", 0x11>;
7735 defm : pclmul_alias<"hqlq", 0x01>;
7736 defm : pclmul_alias<"lqhq", 0x10>;
7737 defm : pclmul_alias<"lqlq", 0x00>;
7739 //===----------------------------------------------------------------------===//
7740 // SSE4A Instructions
7741 //===----------------------------------------------------------------------===//
7743 let Predicates = [HasSSE4A] in {
7745 let Constraints = "$src = $dst" in {
7746 def EXTRQI : Ii8<0x78, MRM0r, (outs VR128:$dst),
7747 (ins VR128:$src, i8imm:$len, i8imm:$idx),
7748 "extrq\t{$idx, $len, $src|$src, $len, $idx}",
7749 [(set VR128:$dst, (int_x86_sse4a_extrqi VR128:$src, imm:$len,
7750 imm:$idx))]>, TB, OpSize;
7751 def EXTRQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7752 (ins VR128:$src, VR128:$mask),
7753 "extrq\t{$mask, $src|$src, $mask}",
7754 [(set VR128:$dst, (int_x86_sse4a_extrq VR128:$src,
7755 VR128:$mask))]>, TB, OpSize;
7757 def INSERTQI : Ii8<0x78, MRMSrcReg, (outs VR128:$dst),
7758 (ins VR128:$src, VR128:$src2, i8imm:$len, i8imm:$idx),
7759 "insertq\t{$idx, $len, $src2, $src|$src, $src2, $len, $idx}",
7760 [(set VR128:$dst, (int_x86_sse4a_insertqi VR128:$src,
7761 VR128:$src2, imm:$len, imm:$idx))]>, XD;
7762 def INSERTQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7763 (ins VR128:$src, VR128:$mask),
7764 "insertq\t{$mask, $src|$src, $mask}",
7765 [(set VR128:$dst, (int_x86_sse4a_insertq VR128:$src,
7766 VR128:$mask))]>, XD;
7769 def MOVNTSS : I<0x2B, MRMDestMem, (outs), (ins f32mem:$dst, VR128:$src),
7770 "movntss\t{$src, $dst|$dst, $src}",
7771 [(int_x86_sse4a_movnt_ss addr:$dst, VR128:$src)]>, XS;
7773 def MOVNTSD : I<0x2B, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
7774 "movntsd\t{$src, $dst|$dst, $src}",
7775 [(int_x86_sse4a_movnt_sd addr:$dst, VR128:$src)]>, XD;
7778 //===----------------------------------------------------------------------===//
7780 //===----------------------------------------------------------------------===//
7782 //===----------------------------------------------------------------------===//
7783 // VBROADCAST - Load from memory and broadcast to all elements of the
7784 // destination operand
7786 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
7787 X86MemOperand x86memop, Intrinsic Int> :
7788 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7789 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7790 [(set RC:$dst, (Int addr:$src))]>, VEX;
7792 // AVX2 adds register forms
7793 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
7795 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7796 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7797 [(set RC:$dst, (Int VR128:$src))]>, VEX;
7799 let ExeDomain = SSEPackedSingle in {
7800 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
7801 int_x86_avx_vbroadcast_ss>;
7802 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
7803 int_x86_avx_vbroadcast_ss_256>, VEX_L;
7805 let ExeDomain = SSEPackedDouble in
7806 def VBROADCASTSDYrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
7807 int_x86_avx_vbroadcast_sd_256>, VEX_L;
7808 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
7809 int_x86_avx_vbroadcastf128_pd_256>, VEX_L;
7811 let ExeDomain = SSEPackedSingle in {
7812 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
7813 int_x86_avx2_vbroadcast_ss_ps>;
7814 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
7815 int_x86_avx2_vbroadcast_ss_ps_256>, VEX_L;
7817 let ExeDomain = SSEPackedDouble in
7818 def VBROADCASTSDYrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
7819 int_x86_avx2_vbroadcast_sd_pd_256>, VEX_L;
7821 let Predicates = [HasAVX2] in
7822 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
7823 int_x86_avx2_vbroadcasti128>, VEX_L;
7825 let Predicates = [HasAVX] in
7826 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
7827 (VBROADCASTF128 addr:$src)>;
7830 //===----------------------------------------------------------------------===//
7831 // VINSERTF128 - Insert packed floating-point values
7833 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7834 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
7835 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7836 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7839 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
7840 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
7841 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7845 let Predicates = [HasAVX] in {
7846 def : Pat<(vinsert128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7848 (VINSERTF128rr VR256:$src1, VR128:$src2,
7849 (INSERT_get_vinsert128_imm VR256:$ins))>;
7850 def : Pat<(vinsert128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7852 (VINSERTF128rr VR256:$src1, VR128:$src2,
7853 (INSERT_get_vinsert128_imm VR256:$ins))>;
7855 def : Pat<(vinsert128_insert:$ins (v8f32 VR256:$src1), (loadv4f32 addr:$src2),
7857 (VINSERTF128rm VR256:$src1, addr:$src2,
7858 (INSERT_get_vinsert128_imm VR256:$ins))>;
7859 def : Pat<(vinsert128_insert:$ins (v4f64 VR256:$src1), (loadv2f64 addr:$src2),
7861 (VINSERTF128rm VR256:$src1, addr:$src2,
7862 (INSERT_get_vinsert128_imm VR256:$ins))>;
7865 let Predicates = [HasAVX1Only] in {
7866 def : Pat<(vinsert128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7868 (VINSERTF128rr VR256:$src1, VR128:$src2,
7869 (INSERT_get_vinsert128_imm VR256:$ins))>;
7870 def : Pat<(vinsert128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7872 (VINSERTF128rr VR256:$src1, VR128:$src2,
7873 (INSERT_get_vinsert128_imm VR256:$ins))>;
7874 def : Pat<(vinsert128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7876 (VINSERTF128rr VR256:$src1, VR128:$src2,
7877 (INSERT_get_vinsert128_imm VR256:$ins))>;
7878 def : Pat<(vinsert128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7880 (VINSERTF128rr VR256:$src1, VR128:$src2,
7881 (INSERT_get_vinsert128_imm VR256:$ins))>;
7883 def : Pat<(vinsert128_insert:$ins (v4i64 VR256:$src1), (loadv2i64 addr:$src2),
7885 (VINSERTF128rm VR256:$src1, addr:$src2,
7886 (INSERT_get_vinsert128_imm VR256:$ins))>;
7887 def : Pat<(vinsert128_insert:$ins (v8i32 VR256:$src1),
7888 (bc_v4i32 (loadv2i64 addr:$src2)),
7890 (VINSERTF128rm VR256:$src1, addr:$src2,
7891 (INSERT_get_vinsert128_imm VR256:$ins))>;
7892 def : Pat<(vinsert128_insert:$ins (v32i8 VR256:$src1),
7893 (bc_v16i8 (loadv2i64 addr:$src2)),
7895 (VINSERTF128rm VR256:$src1, addr:$src2,
7896 (INSERT_get_vinsert128_imm VR256:$ins))>;
7897 def : Pat<(vinsert128_insert:$ins (v16i16 VR256:$src1),
7898 (bc_v8i16 (loadv2i64 addr:$src2)),
7900 (VINSERTF128rm VR256:$src1, addr:$src2,
7901 (INSERT_get_vinsert128_imm VR256:$ins))>;
7904 //===----------------------------------------------------------------------===//
7905 // VEXTRACTF128 - Extract packed floating-point values
7907 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7908 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
7909 (ins VR256:$src1, i8imm:$src2),
7910 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7913 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
7914 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
7915 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7920 let Predicates = [HasAVX] in {
7921 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7922 (v4f32 (VEXTRACTF128rr
7923 (v8f32 VR256:$src1),
7924 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7925 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7926 (v2f64 (VEXTRACTF128rr
7927 (v4f64 VR256:$src1),
7928 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7930 def : Pat<(store (v4f32 (vextract128_extract:$ext (v8f32 VR256:$src1),
7931 (iPTR imm))), addr:$dst),
7932 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7933 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7934 def : Pat<(store (v2f64 (vextract128_extract:$ext (v4f64 VR256:$src1),
7935 (iPTR imm))), addr:$dst),
7936 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7937 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7940 let Predicates = [HasAVX1Only] in {
7941 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7942 (v2i64 (VEXTRACTF128rr
7943 (v4i64 VR256:$src1),
7944 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7945 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7946 (v4i32 (VEXTRACTF128rr
7947 (v8i32 VR256:$src1),
7948 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7949 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7950 (v8i16 (VEXTRACTF128rr
7951 (v16i16 VR256:$src1),
7952 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7953 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7954 (v16i8 (VEXTRACTF128rr
7955 (v32i8 VR256:$src1),
7956 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7958 def : Pat<(alignedstore (v2i64 (vextract128_extract:$ext (v4i64 VR256:$src1),
7959 (iPTR imm))), addr:$dst),
7960 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7961 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7962 def : Pat<(alignedstore (v4i32 (vextract128_extract:$ext (v8i32 VR256:$src1),
7963 (iPTR imm))), addr:$dst),
7964 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7965 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7966 def : Pat<(alignedstore (v8i16 (vextract128_extract:$ext (v16i16 VR256:$src1),
7967 (iPTR imm))), addr:$dst),
7968 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7969 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7970 def : Pat<(alignedstore (v16i8 (vextract128_extract:$ext (v32i8 VR256:$src1),
7971 (iPTR imm))), addr:$dst),
7972 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7973 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7976 //===----------------------------------------------------------------------===//
7977 // VMASKMOV - Conditional SIMD Packed Loads and Stores
7979 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
7980 Intrinsic IntLd, Intrinsic IntLd256,
7981 Intrinsic IntSt, Intrinsic IntSt256> {
7982 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
7983 (ins VR128:$src1, f128mem:$src2),
7984 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7985 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
7987 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
7988 (ins VR256:$src1, f256mem:$src2),
7989 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7990 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7992 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
7993 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
7994 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7995 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7996 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
7997 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
7998 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7999 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V, VEX_L;
8002 let ExeDomain = SSEPackedSingle in
8003 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
8004 int_x86_avx_maskload_ps,
8005 int_x86_avx_maskload_ps_256,
8006 int_x86_avx_maskstore_ps,
8007 int_x86_avx_maskstore_ps_256>;
8008 let ExeDomain = SSEPackedDouble in
8009 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
8010 int_x86_avx_maskload_pd,
8011 int_x86_avx_maskload_pd_256,
8012 int_x86_avx_maskstore_pd,
8013 int_x86_avx_maskstore_pd_256>;
8015 //===----------------------------------------------------------------------===//
8016 // VPERMIL - Permute Single and Double Floating-Point Values
8018 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
8019 RegisterClass RC, X86MemOperand x86memop_f,
8020 X86MemOperand x86memop_i, PatFrag i_frag,
8021 Intrinsic IntVar, ValueType vt> {
8022 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
8023 (ins RC:$src1, RC:$src2),
8024 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8025 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
8026 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
8027 (ins RC:$src1, x86memop_i:$src2),
8028 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8029 [(set RC:$dst, (IntVar RC:$src1,
8030 (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
8032 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
8033 (ins RC:$src1, i8imm:$src2),
8034 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8035 [(set RC:$dst, (vt (X86VPermilp RC:$src1, (i8 imm:$src2))))]>, VEX;
8036 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
8037 (ins x86memop_f:$src1, i8imm:$src2),
8038 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8040 (vt (X86VPermilp (memop addr:$src1), (i8 imm:$src2))))]>, VEX;
8043 let ExeDomain = SSEPackedSingle in {
8044 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
8045 loadv2i64, int_x86_avx_vpermilvar_ps, v4f32>;
8046 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
8047 loadv4i64, int_x86_avx_vpermilvar_ps_256, v8f32>, VEX_L;
8049 let ExeDomain = SSEPackedDouble in {
8050 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
8051 loadv2i64, int_x86_avx_vpermilvar_pd, v2f64>;
8052 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
8053 loadv4i64, int_x86_avx_vpermilvar_pd_256, v4f64>, VEX_L;
8056 let Predicates = [HasAVX] in {
8057 def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
8058 (VPERMILPSYri VR256:$src1, imm:$imm)>;
8059 def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
8060 (VPERMILPDYri VR256:$src1, imm:$imm)>;
8061 def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (loadv4i64 addr:$src1)),
8063 (VPERMILPSYmi addr:$src1, imm:$imm)>;
8064 def : Pat<(v4i64 (X86VPermilp (loadv4i64 addr:$src1), (i8 imm:$imm))),
8065 (VPERMILPDYmi addr:$src1, imm:$imm)>;
8067 def : Pat<(v2i64 (X86VPermilp VR128:$src1, (i8 imm:$imm))),
8068 (VPERMILPDri VR128:$src1, imm:$imm)>;
8069 def : Pat<(v2i64 (X86VPermilp (loadv2i64 addr:$src1), (i8 imm:$imm))),
8070 (VPERMILPDmi addr:$src1, imm:$imm)>;
8073 //===----------------------------------------------------------------------===//
8074 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
8076 let ExeDomain = SSEPackedSingle in {
8077 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
8078 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
8079 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
8080 [(set VR256:$dst, (v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2,
8081 (i8 imm:$src3))))]>, VEX_4V, VEX_L;
8082 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
8083 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
8084 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
8085 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (loadv8f32 addr:$src2),
8086 (i8 imm:$src3)))]>, VEX_4V, VEX_L;
8089 let Predicates = [HasAVX] in {
8090 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
8091 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
8092 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
8093 (loadv4f64 addr:$src2), (i8 imm:$imm))),
8094 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
8097 let Predicates = [HasAVX1Only] in {
8098 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
8099 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
8100 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
8101 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
8102 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
8103 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
8104 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
8105 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
8107 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
8108 (bc_v8i32 (loadv4i64 addr:$src2)), (i8 imm:$imm))),
8109 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
8110 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
8111 (loadv4i64 addr:$src2), (i8 imm:$imm))),
8112 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
8113 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
8114 (bc_v32i8 (loadv4i64 addr:$src2)), (i8 imm:$imm))),
8115 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
8116 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
8117 (bc_v16i16 (loadv4i64 addr:$src2)), (i8 imm:$imm))),
8118 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
8121 //===----------------------------------------------------------------------===//
8122 // VZERO - Zero YMM registers
8124 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
8125 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
8126 // Zero All YMM registers
8127 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
8128 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
8130 // Zero Upper bits of YMM registers
8131 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
8132 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
8135 //===----------------------------------------------------------------------===//
8136 // Half precision conversion instructions
8137 //===----------------------------------------------------------------------===//
8138 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
8139 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
8140 "vcvtph2ps\t{$src, $dst|$dst, $src}",
8141 [(set RC:$dst, (Int VR128:$src))]>,
8143 let neverHasSideEffects = 1, mayLoad = 1 in
8144 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
8145 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
8148 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
8149 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
8150 (ins RC:$src1, i32i8imm:$src2),
8151 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
8152 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
8154 let neverHasSideEffects = 1, mayStore = 1 in
8155 def mr : Ii8<0x1D, MRMDestMem, (outs),
8156 (ins x86memop:$dst, RC:$src1, i32i8imm:$src2),
8157 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
8161 let Predicates = [HasF16C] in {
8162 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
8163 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>, VEX_L;
8164 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
8165 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>, VEX_L;
8168 //===----------------------------------------------------------------------===//
8169 // AVX2 Instructions
8170 //===----------------------------------------------------------------------===//
8172 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
8173 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
8174 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
8175 X86MemOperand x86memop> {
8176 let isCommutable = 1 in
8177 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
8178 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
8179 !strconcat(OpcodeStr,
8180 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
8181 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
8183 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
8184 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
8185 !strconcat(OpcodeStr,
8186 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
8189 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
8193 let isCommutable = 0 in {
8194 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
8195 VR128, loadv2i64, i128mem>;
8196 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
8197 VR256, loadv4i64, i256mem>, VEX_L;
8200 def : Pat<(v4i32 (X86Blendi (v4i32 VR128:$src1), (v4i32 VR128:$src2),
8202 (VPBLENDDrri VR128:$src1, VR128:$src2, imm:$mask)>;
8203 def : Pat<(v8i32 (X86Blendi (v8i32 VR256:$src1), (v8i32 VR256:$src2),
8205 (VPBLENDDYrri VR256:$src1, VR256:$src2, imm:$mask)>;
8207 //===----------------------------------------------------------------------===//
8208 // VPBROADCAST - Load from memory and broadcast to all elements of the
8209 // destination operand
8211 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
8212 X86MemOperand x86memop, PatFrag ld_frag,
8213 Intrinsic Int128, Intrinsic Int256> {
8214 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
8215 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
8216 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
8217 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
8218 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
8220 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
8221 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
8222 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
8223 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX, VEX_L;
8224 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
8225 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
8227 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>,
8231 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
8232 int_x86_avx2_pbroadcastb_128,
8233 int_x86_avx2_pbroadcastb_256>;
8234 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
8235 int_x86_avx2_pbroadcastw_128,
8236 int_x86_avx2_pbroadcastw_256>;
8237 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
8238 int_x86_avx2_pbroadcastd_128,
8239 int_x86_avx2_pbroadcastd_256>;
8240 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
8241 int_x86_avx2_pbroadcastq_128,
8242 int_x86_avx2_pbroadcastq_256>;
8244 let Predicates = [HasAVX2] in {
8245 def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
8246 (VPBROADCASTBrm addr:$src)>;
8247 def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
8248 (VPBROADCASTBYrm addr:$src)>;
8249 def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
8250 (VPBROADCASTWrm addr:$src)>;
8251 def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
8252 (VPBROADCASTWYrm addr:$src)>;
8253 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
8254 (VPBROADCASTDrm addr:$src)>;
8255 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
8256 (VPBROADCASTDYrm addr:$src)>;
8257 def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
8258 (VPBROADCASTQrm addr:$src)>;
8259 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
8260 (VPBROADCASTQYrm addr:$src)>;
8262 def : Pat<(v16i8 (X86VBroadcast (v16i8 VR128:$src))),
8263 (VPBROADCASTBrr VR128:$src)>;
8264 def : Pat<(v32i8 (X86VBroadcast (v16i8 VR128:$src))),
8265 (VPBROADCASTBYrr VR128:$src)>;
8266 def : Pat<(v8i16 (X86VBroadcast (v8i16 VR128:$src))),
8267 (VPBROADCASTWrr VR128:$src)>;
8268 def : Pat<(v16i16 (X86VBroadcast (v8i16 VR128:$src))),
8269 (VPBROADCASTWYrr VR128:$src)>;
8270 def : Pat<(v4i32 (X86VBroadcast (v4i32 VR128:$src))),
8271 (VPBROADCASTDrr VR128:$src)>;
8272 def : Pat<(v8i32 (X86VBroadcast (v4i32 VR128:$src))),
8273 (VPBROADCASTDYrr VR128:$src)>;
8274 def : Pat<(v2i64 (X86VBroadcast (v2i64 VR128:$src))),
8275 (VPBROADCASTQrr VR128:$src)>;
8276 def : Pat<(v4i64 (X86VBroadcast (v2i64 VR128:$src))),
8277 (VPBROADCASTQYrr VR128:$src)>;
8278 def : Pat<(v4f32 (X86VBroadcast (v4f32 VR128:$src))),
8279 (VBROADCASTSSrr VR128:$src)>;
8280 def : Pat<(v8f32 (X86VBroadcast (v4f32 VR128:$src))),
8281 (VBROADCASTSSYrr VR128:$src)>;
8282 def : Pat<(v2f64 (X86VBroadcast (v2f64 VR128:$src))),
8283 (VPBROADCASTQrr VR128:$src)>;
8284 def : Pat<(v4f64 (X86VBroadcast (v2f64 VR128:$src))),
8285 (VBROADCASTSDYrr VR128:$src)>;
8287 // Provide fallback in case the load node that is used in the patterns above
8288 // is used by additional users, which prevents the pattern selection.
8289 let AddedComplexity = 20 in {
8290 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
8291 (VBROADCASTSSrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
8292 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
8293 (VBROADCASTSSYrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
8294 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
8295 (VBROADCASTSDYrr (COPY_TO_REGCLASS FR64:$src, VR128))>;
8297 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
8298 (VBROADCASTSSrr (COPY_TO_REGCLASS GR32:$src, VR128))>;
8299 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
8300 (VBROADCASTSSYrr (COPY_TO_REGCLASS GR32:$src, VR128))>;
8301 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
8302 (VBROADCASTSDYrr (COPY_TO_REGCLASS GR64:$src, VR128))>;
8306 // AVX1 broadcast patterns
8307 let Predicates = [HasAVX1Only] in {
8308 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
8309 (VBROADCASTSSYrm addr:$src)>;
8310 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
8311 (VBROADCASTSDYrm addr:$src)>;
8312 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
8313 (VBROADCASTSSrm addr:$src)>;
8316 let Predicates = [HasAVX] in {
8317 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
8318 (VBROADCASTSSYrm addr:$src)>;
8319 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
8320 (VBROADCASTSDYrm addr:$src)>;
8321 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
8322 (VBROADCASTSSrm addr:$src)>;
8324 // Provide fallback in case the load node that is used in the patterns above
8325 // is used by additional users, which prevents the pattern selection.
8326 let AddedComplexity = 20 in {
8327 // 128bit broadcasts:
8328 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
8329 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0)>;
8330 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
8331 (VINSERTF128rr (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)),
8332 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0), sub_xmm),
8333 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0), 1)>;
8334 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
8335 (VINSERTF128rr (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)),
8336 (VPSHUFDri (COPY_TO_REGCLASS FR64:$src, VR128), 0x44), sub_xmm),
8337 (VPSHUFDri (COPY_TO_REGCLASS FR64:$src, VR128), 0x44), 1)>;
8339 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
8340 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0)>;
8341 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
8342 (VINSERTF128rr (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)),
8343 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0), sub_xmm),
8344 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0), 1)>;
8345 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
8346 (VINSERTF128rr (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)),
8347 (VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), sub_xmm),
8348 (VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), 1)>;
8352 //===----------------------------------------------------------------------===//
8353 // VPERM - Permute instructions
8356 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
8358 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
8359 (ins VR256:$src1, VR256:$src2),
8360 !strconcat(OpcodeStr,
8361 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8363 (OpVT (X86VPermv VR256:$src1, VR256:$src2)))]>,
8365 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
8366 (ins VR256:$src1, i256mem:$src2),
8367 !strconcat(OpcodeStr,
8368 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8370 (OpVT (X86VPermv VR256:$src1,
8371 (bitconvert (mem_frag addr:$src2)))))]>,
8375 defm VPERMD : avx2_perm<0x36, "vpermd", loadv4i64, v8i32>;
8376 let ExeDomain = SSEPackedSingle in
8377 defm VPERMPS : avx2_perm<0x16, "vpermps", loadv8f32, v8f32>;
8379 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
8381 def Yri : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
8382 (ins VR256:$src1, i8imm:$src2),
8383 !strconcat(OpcodeStr,
8384 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8386 (OpVT (X86VPermi VR256:$src1, (i8 imm:$src2))))]>,
8388 def Ymi : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
8389 (ins i256mem:$src1, i8imm:$src2),
8390 !strconcat(OpcodeStr,
8391 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8393 (OpVT (X86VPermi (mem_frag addr:$src1),
8394 (i8 imm:$src2))))]>, VEX, VEX_L;
8397 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", loadv4i64, v4i64>, VEX_W;
8398 let ExeDomain = SSEPackedDouble in
8399 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", loadv4f64, v4f64>, VEX_W;
8401 //===----------------------------------------------------------------------===//
8402 // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
8404 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
8405 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
8406 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
8407 [(set VR256:$dst, (v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2,
8408 (i8 imm:$src3))))]>, VEX_4V, VEX_L;
8409 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
8410 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
8411 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
8412 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (loadv4i64 addr:$src2),
8413 (i8 imm:$src3)))]>, VEX_4V, VEX_L;
8415 let Predicates = [HasAVX2] in {
8416 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
8417 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
8418 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
8419 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
8420 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
8421 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
8423 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (loadv4i64 addr:$src2)),
8425 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
8426 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
8427 (bc_v16i16 (loadv4i64 addr:$src2)), (i8 imm:$imm))),
8428 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
8429 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (loadv4i64 addr:$src2)),
8431 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
8435 //===----------------------------------------------------------------------===//
8436 // VINSERTI128 - Insert packed integer values
8438 let neverHasSideEffects = 1 in {
8439 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
8440 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
8441 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
8444 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
8445 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
8446 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
8450 let Predicates = [HasAVX2] in {
8451 def : Pat<(vinsert128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
8453 (VINSERTI128rr VR256:$src1, VR128:$src2,
8454 (INSERT_get_vinsert128_imm VR256:$ins))>;
8455 def : Pat<(vinsert128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
8457 (VINSERTI128rr VR256:$src1, VR128:$src2,
8458 (INSERT_get_vinsert128_imm VR256:$ins))>;
8459 def : Pat<(vinsert128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
8461 (VINSERTI128rr VR256:$src1, VR128:$src2,
8462 (INSERT_get_vinsert128_imm VR256:$ins))>;
8463 def : Pat<(vinsert128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
8465 (VINSERTI128rr VR256:$src1, VR128:$src2,
8466 (INSERT_get_vinsert128_imm VR256:$ins))>;
8468 def : Pat<(vinsert128_insert:$ins (v4i64 VR256:$src1), (loadv2i64 addr:$src2),
8470 (VINSERTI128rm VR256:$src1, addr:$src2,
8471 (INSERT_get_vinsert128_imm VR256:$ins))>;
8472 def : Pat<(vinsert128_insert:$ins (v8i32 VR256:$src1),
8473 (bc_v4i32 (loadv2i64 addr:$src2)),
8475 (VINSERTI128rm VR256:$src1, addr:$src2,
8476 (INSERT_get_vinsert128_imm VR256:$ins))>;
8477 def : Pat<(vinsert128_insert:$ins (v32i8 VR256:$src1),
8478 (bc_v16i8 (loadv2i64 addr:$src2)),
8480 (VINSERTI128rm VR256:$src1, addr:$src2,
8481 (INSERT_get_vinsert128_imm VR256:$ins))>;
8482 def : Pat<(vinsert128_insert:$ins (v16i16 VR256:$src1),
8483 (bc_v8i16 (loadv2i64 addr:$src2)),
8485 (VINSERTI128rm VR256:$src1, addr:$src2,
8486 (INSERT_get_vinsert128_imm VR256:$ins))>;
8489 //===----------------------------------------------------------------------===//
8490 // VEXTRACTI128 - Extract packed integer values
8492 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
8493 (ins VR256:$src1, i8imm:$src2),
8494 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
8496 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
8498 let neverHasSideEffects = 1, mayStore = 1 in
8499 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
8500 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
8501 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
8504 let Predicates = [HasAVX2] in {
8505 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
8506 (v2i64 (VEXTRACTI128rr
8507 (v4i64 VR256:$src1),
8508 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
8509 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
8510 (v4i32 (VEXTRACTI128rr
8511 (v8i32 VR256:$src1),
8512 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
8513 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
8514 (v8i16 (VEXTRACTI128rr
8515 (v16i16 VR256:$src1),
8516 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
8517 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
8518 (v16i8 (VEXTRACTI128rr
8519 (v32i8 VR256:$src1),
8520 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
8522 def : Pat<(store (v2i64 (vextract128_extract:$ext (v4i64 VR256:$src1),
8523 (iPTR imm))), addr:$dst),
8524 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8525 (EXTRACT_get_vextract128_imm VR128:$ext))>;
8526 def : Pat<(store (v4i32 (vextract128_extract:$ext (v8i32 VR256:$src1),
8527 (iPTR imm))), addr:$dst),
8528 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8529 (EXTRACT_get_vextract128_imm VR128:$ext))>;
8530 def : Pat<(store (v8i16 (vextract128_extract:$ext (v16i16 VR256:$src1),
8531 (iPTR imm))), addr:$dst),
8532 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8533 (EXTRACT_get_vextract128_imm VR128:$ext))>;
8534 def : Pat<(store (v16i8 (vextract128_extract:$ext (v32i8 VR256:$src1),
8535 (iPTR imm))), addr:$dst),
8536 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8537 (EXTRACT_get_vextract128_imm VR128:$ext))>;
8540 //===----------------------------------------------------------------------===//
8541 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
8543 multiclass avx2_pmovmask<string OpcodeStr,
8544 Intrinsic IntLd128, Intrinsic IntLd256,
8545 Intrinsic IntSt128, Intrinsic IntSt256> {
8546 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
8547 (ins VR128:$src1, i128mem:$src2),
8548 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8549 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
8550 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
8551 (ins VR256:$src1, i256mem:$src2),
8552 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8553 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
8555 def mr : AVX28I<0x8e, MRMDestMem, (outs),
8556 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
8557 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8558 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
8559 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
8560 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
8561 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8562 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V, VEX_L;
8565 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
8566 int_x86_avx2_maskload_d,
8567 int_x86_avx2_maskload_d_256,
8568 int_x86_avx2_maskstore_d,
8569 int_x86_avx2_maskstore_d_256>;
8570 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
8571 int_x86_avx2_maskload_q,
8572 int_x86_avx2_maskload_q_256,
8573 int_x86_avx2_maskstore_q,
8574 int_x86_avx2_maskstore_q_256>, VEX_W;
8577 //===----------------------------------------------------------------------===//
8578 // Variable Bit Shifts
8580 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
8581 ValueType vt128, ValueType vt256> {
8582 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
8583 (ins VR128:$src1, VR128:$src2),
8584 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8586 (vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
8588 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
8589 (ins VR128:$src1, i128mem:$src2),
8590 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8592 (vt128 (OpNode VR128:$src1,
8593 (vt128 (bitconvert (loadv2i64 addr:$src2))))))]>,
8595 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
8596 (ins VR256:$src1, VR256:$src2),
8597 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8599 (vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
8601 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
8602 (ins VR256:$src1, i256mem:$src2),
8603 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8605 (vt256 (OpNode VR256:$src1,
8606 (vt256 (bitconvert (loadv4i64 addr:$src2))))))]>,
8610 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
8611 defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", shl, v2i64, v4i64>, VEX_W;
8612 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", srl, v4i32, v8i32>;
8613 defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", srl, v2i64, v4i64>, VEX_W;
8614 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;
8616 //===----------------------------------------------------------------------===//
8617 // VGATHER - GATHER Operations
8618 multiclass avx2_gather<bits<8> opc, string OpcodeStr, RegisterClass RC256,
8619 X86MemOperand memop128, X86MemOperand memop256> {
8620 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst, VR128:$mask_wb),
8621 (ins VR128:$src1, memop128:$src2, VR128:$mask),
8622 !strconcat(OpcodeStr,
8623 "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
8625 def Yrm : AVX28I<opc, MRMSrcMem, (outs RC256:$dst, RC256:$mask_wb),
8626 (ins RC256:$src1, memop256:$src2, RC256:$mask),
8627 !strconcat(OpcodeStr,
8628 "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
8629 []>, VEX_4VOp3, VEX_L;
8632 let mayLoad = 1, Constraints
8633 = "@earlyclobber $dst,@earlyclobber $mask_wb, $src1 = $dst, $mask = $mask_wb"
8635 defm VGATHERDPD : avx2_gather<0x92, "vgatherdpd", VR256, vx64mem, vx64mem>, VEX_W;
8636 defm VGATHERQPD : avx2_gather<0x93, "vgatherqpd", VR256, vx64mem, vy64mem>, VEX_W;
8637 defm VGATHERDPS : avx2_gather<0x92, "vgatherdps", VR256, vx32mem, vy32mem>;
8638 defm VGATHERQPS : avx2_gather<0x93, "vgatherqps", VR128, vx32mem, vy32mem>;
8639 defm VPGATHERDQ : avx2_gather<0x90, "vpgatherdq", VR256, vx64mem, vx64mem>, VEX_W;
8640 defm VPGATHERQQ : avx2_gather<0x91, "vpgatherqq", VR256, vx64mem, vy64mem>, VEX_W;
8641 defm VPGATHERDD : avx2_gather<0x90, "vpgatherdd", VR256, vx32mem, vy32mem>;
8642 defm VPGATHERQD : avx2_gather<0x91, "vpgatherqd", VR128, vx32mem, vy32mem>;