1 //===-- X86InstrSSE.td - SSE Instruction Set ---------------*- tablegen -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the X86 SSE instruction set, defining the instructions,
11 // and properties of the instructions which are needed for code generation,
12 // machine code emission, and analysis.
14 //===----------------------------------------------------------------------===//
16 class OpndItins<InstrItinClass arg_rr, InstrItinClass arg_rm> {
17 InstrItinClass rr = arg_rr;
18 InstrItinClass rm = arg_rm;
19 // InstrSchedModel info.
20 X86FoldableSchedWrite Sched = WriteFAdd;
23 class SizeItins<OpndItins arg_s, OpndItins arg_d> {
29 class ShiftOpndItins<InstrItinClass arg_rr, InstrItinClass arg_rm,
30 InstrItinClass arg_ri> {
31 InstrItinClass rr = arg_rr;
32 InstrItinClass rm = arg_rm;
33 InstrItinClass ri = arg_ri;
38 let Sched = WriteFAdd in {
39 def SSE_ALU_F32S : OpndItins<
40 IIC_SSE_ALU_F32S_RR, IIC_SSE_ALU_F32S_RM
43 def SSE_ALU_F64S : OpndItins<
44 IIC_SSE_ALU_F64S_RR, IIC_SSE_ALU_F64S_RM
48 def SSE_ALU_ITINS_S : SizeItins<
49 SSE_ALU_F32S, SSE_ALU_F64S
52 let Sched = WriteFMul in {
53 def SSE_MUL_F32S : OpndItins<
54 IIC_SSE_MUL_F32S_RR, IIC_SSE_MUL_F64S_RM
57 def SSE_MUL_F64S : OpndItins<
58 IIC_SSE_MUL_F64S_RR, IIC_SSE_MUL_F64S_RM
62 def SSE_MUL_ITINS_S : SizeItins<
63 SSE_MUL_F32S, SSE_MUL_F64S
66 let Sched = WriteFDiv in {
67 def SSE_DIV_F32S : OpndItins<
68 IIC_SSE_DIV_F32S_RR, IIC_SSE_DIV_F64S_RM
71 def SSE_DIV_F64S : OpndItins<
72 IIC_SSE_DIV_F64S_RR, IIC_SSE_DIV_F64S_RM
76 def SSE_DIV_ITINS_S : SizeItins<
77 SSE_DIV_F32S, SSE_DIV_F64S
81 let Sched = WriteFAdd in {
82 def SSE_ALU_F32P : OpndItins<
83 IIC_SSE_ALU_F32P_RR, IIC_SSE_ALU_F32P_RM
86 def SSE_ALU_F64P : OpndItins<
87 IIC_SSE_ALU_F64P_RR, IIC_SSE_ALU_F64P_RM
91 def SSE_ALU_ITINS_P : SizeItins<
92 SSE_ALU_F32P, SSE_ALU_F64P
95 let Sched = WriteFMul in {
96 def SSE_MUL_F32P : OpndItins<
97 IIC_SSE_MUL_F32P_RR, IIC_SSE_MUL_F64P_RM
100 def SSE_MUL_F64P : OpndItins<
101 IIC_SSE_MUL_F64P_RR, IIC_SSE_MUL_F64P_RM
105 def SSE_MUL_ITINS_P : SizeItins<
106 SSE_MUL_F32P, SSE_MUL_F64P
109 let Sched = WriteFDiv in {
110 def SSE_DIV_F32P : OpndItins<
111 IIC_SSE_DIV_F32P_RR, IIC_SSE_DIV_F64P_RM
114 def SSE_DIV_F64P : OpndItins<
115 IIC_SSE_DIV_F64P_RR, IIC_SSE_DIV_F64P_RM
119 def SSE_DIV_ITINS_P : SizeItins<
120 SSE_DIV_F32P, SSE_DIV_F64P
123 def SSE_BIT_ITINS_P : OpndItins<
124 IIC_SSE_BIT_P_RR, IIC_SSE_BIT_P_RM
127 let Sched = WriteVecALU in {
128 def SSE_INTALU_ITINS_P : OpndItins<
129 IIC_SSE_INTALU_P_RR, IIC_SSE_INTALU_P_RM
132 def SSE_INTALUQ_ITINS_P : OpndItins<
133 IIC_SSE_INTALUQ_P_RR, IIC_SSE_INTALUQ_P_RM
137 let Sched = WriteVecIMul in
138 def SSE_INTMUL_ITINS_P : OpndItins<
139 IIC_SSE_INTMUL_P_RR, IIC_SSE_INTMUL_P_RM
142 def SSE_INTSHIFT_ITINS_P : ShiftOpndItins<
143 IIC_SSE_INTSH_P_RR, IIC_SSE_INTSH_P_RM, IIC_SSE_INTSH_P_RI
146 def SSE_MOVA_ITINS : OpndItins<
147 IIC_SSE_MOVA_P_RR, IIC_SSE_MOVA_P_RM
150 def SSE_MOVU_ITINS : OpndItins<
151 IIC_SSE_MOVU_P_RR, IIC_SSE_MOVU_P_RM
154 def SSE_DPPD_ITINS : OpndItins<
155 IIC_SSE_DPPD_RR, IIC_SSE_DPPD_RM
158 def SSE_DPPS_ITINS : OpndItins<
159 IIC_SSE_DPPS_RR, IIC_SSE_DPPD_RM
162 def DEFAULT_ITINS : OpndItins<
163 IIC_ALU_NONMEM, IIC_ALU_MEM
166 def SSE_EXTRACT_ITINS : OpndItins<
167 IIC_SSE_EXTRACTPS_RR, IIC_SSE_EXTRACTPS_RM
170 def SSE_INSERT_ITINS : OpndItins<
171 IIC_SSE_INSERTPS_RR, IIC_SSE_INSERTPS_RM
174 def SSE_MPSADBW_ITINS : OpndItins<
175 IIC_SSE_MPSADBW_RR, IIC_SSE_MPSADBW_RM
178 def SSE_PMULLD_ITINS : OpndItins<
179 IIC_SSE_PMULLD_RR, IIC_SSE_PMULLD_RM
182 //===----------------------------------------------------------------------===//
183 // SSE 1 & 2 Instructions Classes
184 //===----------------------------------------------------------------------===//
186 /// sse12_fp_scalar - SSE 1 & 2 scalar instructions class
187 multiclass sse12_fp_scalar<bits<8> opc, string OpcodeStr, SDNode OpNode,
188 RegisterClass RC, X86MemOperand x86memop,
191 let isCommutable = 1 in {
192 def rr : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
194 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
195 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
196 [(set RC:$dst, (OpNode RC:$src1, RC:$src2))], itins.rr>,
197 Sched<[itins.Sched]>;
199 def rm : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
201 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
202 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
203 [(set RC:$dst, (OpNode RC:$src1, (load addr:$src2)))], itins.rm>,
204 Sched<[itins.Sched.Folded, ReadAfterLd]>;
207 /// sse12_fp_scalar_int - SSE 1 & 2 scalar instructions intrinsics class
208 multiclass sse12_fp_scalar_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
209 string asm, string SSEVer, string FPSizeStr,
210 Operand memopr, ComplexPattern mem_cpat,
213 def rr_Int : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
215 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
216 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
217 [(set RC:$dst, (!cast<Intrinsic>(
218 !strconcat("int_x86_sse", SSEVer, "_", OpcodeStr, FPSizeStr))
219 RC:$src1, RC:$src2))], itins.rr>,
220 Sched<[itins.Sched]>;
221 def rm_Int : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, memopr:$src2),
223 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
224 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
225 [(set RC:$dst, (!cast<Intrinsic>(!strconcat("int_x86_sse",
226 SSEVer, "_", OpcodeStr, FPSizeStr))
227 RC:$src1, mem_cpat:$src2))], itins.rm>,
228 Sched<[itins.Sched.Folded, ReadAfterLd]>;
231 /// sse12_fp_packed - SSE 1 & 2 packed instructions class
232 multiclass sse12_fp_packed<bits<8> opc, string OpcodeStr, SDNode OpNode,
233 RegisterClass RC, ValueType vt,
234 X86MemOperand x86memop, PatFrag mem_frag,
235 Domain d, OpndItins itins, bit Is2Addr = 1> {
236 let isCommutable = 1 in
237 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
239 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
240 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
241 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], itins.rr, d>,
242 Sched<[itins.Sched]>;
244 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
246 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
247 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
248 [(set RC:$dst, (OpNode RC:$src1, (mem_frag addr:$src2)))],
250 Sched<[itins.Sched.Folded, ReadAfterLd]>;
253 /// sse12_fp_packed_logical_rm - SSE 1 & 2 packed instructions class
254 multiclass sse12_fp_packed_logical_rm<bits<8> opc, RegisterClass RC, Domain d,
255 string OpcodeStr, X86MemOperand x86memop,
256 list<dag> pat_rr, list<dag> pat_rm,
258 let isCommutable = 1, hasSideEffects = 0 in
259 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
261 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
262 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
263 pat_rr, NoItinerary, d>,
264 Sched<[WriteVecLogic]>;
265 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
267 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
268 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
269 pat_rm, NoItinerary, d>,
270 Sched<[WriteVecLogicLd, ReadAfterLd]>;
273 //===----------------------------------------------------------------------===//
274 // Non-instruction patterns
275 //===----------------------------------------------------------------------===//
277 // A vector extract of the first f32/f64 position is a subregister copy
278 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
279 (COPY_TO_REGCLASS (v4f32 VR128:$src), FR32)>;
280 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
281 (COPY_TO_REGCLASS (v2f64 VR128:$src), FR64)>;
283 // A 128-bit subvector extract from the first 256-bit vector position
284 // is a subregister copy that needs no instruction.
285 def : Pat<(v4i32 (extract_subvector (v8i32 VR256:$src), (iPTR 0))),
286 (v4i32 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm))>;
287 def : Pat<(v4f32 (extract_subvector (v8f32 VR256:$src), (iPTR 0))),
288 (v4f32 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_xmm))>;
290 def : Pat<(v2i64 (extract_subvector (v4i64 VR256:$src), (iPTR 0))),
291 (v2i64 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm))>;
292 def : Pat<(v2f64 (extract_subvector (v4f64 VR256:$src), (iPTR 0))),
293 (v2f64 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm))>;
295 def : Pat<(v8i16 (extract_subvector (v16i16 VR256:$src), (iPTR 0))),
296 (v8i16 (EXTRACT_SUBREG (v16i16 VR256:$src), sub_xmm))>;
297 def : Pat<(v16i8 (extract_subvector (v32i8 VR256:$src), (iPTR 0))),
298 (v16i8 (EXTRACT_SUBREG (v32i8 VR256:$src), sub_xmm))>;
300 // A 128-bit subvector insert to the first 256-bit vector position
301 // is a subregister copy that needs no instruction.
302 let AddedComplexity = 25 in { // to give priority over vinsertf128rm
303 def : Pat<(insert_subvector undef, (v2i64 VR128:$src), (iPTR 0)),
304 (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
305 def : Pat<(insert_subvector undef, (v2f64 VR128:$src), (iPTR 0)),
306 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
307 def : Pat<(insert_subvector undef, (v4i32 VR128:$src), (iPTR 0)),
308 (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
309 def : Pat<(insert_subvector undef, (v4f32 VR128:$src), (iPTR 0)),
310 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
311 def : Pat<(insert_subvector undef, (v8i16 VR128:$src), (iPTR 0)),
312 (INSERT_SUBREG (v16i16 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
313 def : Pat<(insert_subvector undef, (v16i8 VR128:$src), (iPTR 0)),
314 (INSERT_SUBREG (v32i8 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
317 // Implicitly promote a 32-bit scalar to a vector.
318 def : Pat<(v4f32 (scalar_to_vector FR32:$src)),
319 (COPY_TO_REGCLASS FR32:$src, VR128)>;
320 def : Pat<(v8f32 (scalar_to_vector FR32:$src)),
321 (COPY_TO_REGCLASS FR32:$src, VR128)>;
322 // Implicitly promote a 64-bit scalar to a vector.
323 def : Pat<(v2f64 (scalar_to_vector FR64:$src)),
324 (COPY_TO_REGCLASS FR64:$src, VR128)>;
325 def : Pat<(v4f64 (scalar_to_vector FR64:$src)),
326 (COPY_TO_REGCLASS FR64:$src, VR128)>;
328 // Bitcasts between 128-bit vector types. Return the original type since
329 // no instruction is needed for the conversion
330 let Predicates = [HasSSE2] in {
331 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
332 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
333 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
334 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
335 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
336 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
337 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
338 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
339 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
340 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
341 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
342 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
343 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
344 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
345 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
346 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
347 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
348 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
349 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
350 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
351 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
352 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
353 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
354 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
355 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
356 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
357 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
358 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
359 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
360 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
363 // Bitcasts between 256-bit vector types. Return the original type since
364 // no instruction is needed for the conversion
365 let Predicates = [HasAVX] in {
366 def : Pat<(v4f64 (bitconvert (v8f32 VR256:$src))), (v4f64 VR256:$src)>;
367 def : Pat<(v4f64 (bitconvert (v8i32 VR256:$src))), (v4f64 VR256:$src)>;
368 def : Pat<(v4f64 (bitconvert (v4i64 VR256:$src))), (v4f64 VR256:$src)>;
369 def : Pat<(v4f64 (bitconvert (v16i16 VR256:$src))), (v4f64 VR256:$src)>;
370 def : Pat<(v4f64 (bitconvert (v32i8 VR256:$src))), (v4f64 VR256:$src)>;
371 def : Pat<(v8f32 (bitconvert (v8i32 VR256:$src))), (v8f32 VR256:$src)>;
372 def : Pat<(v8f32 (bitconvert (v4i64 VR256:$src))), (v8f32 VR256:$src)>;
373 def : Pat<(v8f32 (bitconvert (v4f64 VR256:$src))), (v8f32 VR256:$src)>;
374 def : Pat<(v8f32 (bitconvert (v32i8 VR256:$src))), (v8f32 VR256:$src)>;
375 def : Pat<(v8f32 (bitconvert (v16i16 VR256:$src))), (v8f32 VR256:$src)>;
376 def : Pat<(v4i64 (bitconvert (v8f32 VR256:$src))), (v4i64 VR256:$src)>;
377 def : Pat<(v4i64 (bitconvert (v8i32 VR256:$src))), (v4i64 VR256:$src)>;
378 def : Pat<(v4i64 (bitconvert (v4f64 VR256:$src))), (v4i64 VR256:$src)>;
379 def : Pat<(v4i64 (bitconvert (v32i8 VR256:$src))), (v4i64 VR256:$src)>;
380 def : Pat<(v4i64 (bitconvert (v16i16 VR256:$src))), (v4i64 VR256:$src)>;
381 def : Pat<(v32i8 (bitconvert (v4f64 VR256:$src))), (v32i8 VR256:$src)>;
382 def : Pat<(v32i8 (bitconvert (v4i64 VR256:$src))), (v32i8 VR256:$src)>;
383 def : Pat<(v32i8 (bitconvert (v8f32 VR256:$src))), (v32i8 VR256:$src)>;
384 def : Pat<(v32i8 (bitconvert (v8i32 VR256:$src))), (v32i8 VR256:$src)>;
385 def : Pat<(v32i8 (bitconvert (v16i16 VR256:$src))), (v32i8 VR256:$src)>;
386 def : Pat<(v8i32 (bitconvert (v32i8 VR256:$src))), (v8i32 VR256:$src)>;
387 def : Pat<(v8i32 (bitconvert (v16i16 VR256:$src))), (v8i32 VR256:$src)>;
388 def : Pat<(v8i32 (bitconvert (v8f32 VR256:$src))), (v8i32 VR256:$src)>;
389 def : Pat<(v8i32 (bitconvert (v4i64 VR256:$src))), (v8i32 VR256:$src)>;
390 def : Pat<(v8i32 (bitconvert (v4f64 VR256:$src))), (v8i32 VR256:$src)>;
391 def : Pat<(v16i16 (bitconvert (v8f32 VR256:$src))), (v16i16 VR256:$src)>;
392 def : Pat<(v16i16 (bitconvert (v8i32 VR256:$src))), (v16i16 VR256:$src)>;
393 def : Pat<(v16i16 (bitconvert (v4i64 VR256:$src))), (v16i16 VR256:$src)>;
394 def : Pat<(v16i16 (bitconvert (v4f64 VR256:$src))), (v16i16 VR256:$src)>;
395 def : Pat<(v16i16 (bitconvert (v32i8 VR256:$src))), (v16i16 VR256:$src)>;
398 // Alias instructions that map fld0 to xorps for sse or vxorps for avx.
399 // This is expanded by ExpandPostRAPseudos.
400 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
401 isPseudo = 1, SchedRW = [WriteZero] in {
402 def FsFLD0SS : I<0, Pseudo, (outs FR32:$dst), (ins), "",
403 [(set FR32:$dst, fp32imm0)]>, Requires<[HasSSE1]>;
404 def FsFLD0SD : I<0, Pseudo, (outs FR64:$dst), (ins), "",
405 [(set FR64:$dst, fpimm0)]>, Requires<[HasSSE2]>;
408 //===----------------------------------------------------------------------===//
409 // AVX & SSE - Zero/One Vectors
410 //===----------------------------------------------------------------------===//
412 // Alias instruction that maps zero vector to pxor / xorp* for sse.
413 // This is expanded by ExpandPostRAPseudos to an xorps / vxorps, and then
414 // swizzled by ExecutionDepsFix to pxor.
415 // We set canFoldAsLoad because this can be converted to a constant-pool
416 // load of an all-zeros value if folding it would be beneficial.
417 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
418 isPseudo = 1, SchedRW = [WriteZero] in {
419 def V_SET0 : I<0, Pseudo, (outs VR128:$dst), (ins), "",
420 [(set VR128:$dst, (v4f32 immAllZerosV))]>;
423 def : Pat<(v2f64 immAllZerosV), (V_SET0)>;
424 def : Pat<(v4i32 immAllZerosV), (V_SET0)>;
425 def : Pat<(v2i64 immAllZerosV), (V_SET0)>;
426 def : Pat<(v8i16 immAllZerosV), (V_SET0)>;
427 def : Pat<(v16i8 immAllZerosV), (V_SET0)>;
430 // The same as done above but for AVX. The 256-bit AVX1 ISA doesn't support PI,
431 // and doesn't need it because on sandy bridge the register is set to zero
432 // at the rename stage without using any execution unit, so SET0PSY
433 // and SET0PDY can be used for vector int instructions without penalty
434 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
435 isPseudo = 1, Predicates = [HasAVX], SchedRW = [WriteZero] in {
436 def AVX_SET0 : I<0, Pseudo, (outs VR256:$dst), (ins), "",
437 [(set VR256:$dst, (v8f32 immAllZerosV))]>;
440 let Predicates = [HasAVX] in
441 def : Pat<(v4f64 immAllZerosV), (AVX_SET0)>;
443 let Predicates = [HasAVX2] in {
444 def : Pat<(v4i64 immAllZerosV), (AVX_SET0)>;
445 def : Pat<(v8i32 immAllZerosV), (AVX_SET0)>;
446 def : Pat<(v16i16 immAllZerosV), (AVX_SET0)>;
447 def : Pat<(v32i8 immAllZerosV), (AVX_SET0)>;
450 // AVX1 has no support for 256-bit integer instructions, but since the 128-bit
451 // VPXOR instruction writes zero to its upper part, it's safe build zeros.
452 let Predicates = [HasAVX1Only] in {
453 def : Pat<(v32i8 immAllZerosV), (SUBREG_TO_REG (i8 0), (V_SET0), sub_xmm)>;
454 def : Pat<(bc_v32i8 (v8f32 immAllZerosV)),
455 (SUBREG_TO_REG (i8 0), (V_SET0), sub_xmm)>;
457 def : Pat<(v16i16 immAllZerosV), (SUBREG_TO_REG (i16 0), (V_SET0), sub_xmm)>;
458 def : Pat<(bc_v16i16 (v8f32 immAllZerosV)),
459 (SUBREG_TO_REG (i16 0), (V_SET0), sub_xmm)>;
461 def : Pat<(v8i32 immAllZerosV), (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
462 def : Pat<(bc_v8i32 (v8f32 immAllZerosV)),
463 (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
465 def : Pat<(v4i64 immAllZerosV), (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
466 def : Pat<(bc_v4i64 (v8f32 immAllZerosV)),
467 (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
470 // We set canFoldAsLoad because this can be converted to a constant-pool
471 // load of an all-ones value if folding it would be beneficial.
472 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
473 isPseudo = 1, SchedRW = [WriteZero] in {
474 def V_SETALLONES : I<0, Pseudo, (outs VR128:$dst), (ins), "",
475 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
476 let Predicates = [HasAVX2] in
477 def AVX2_SETALLONES : I<0, Pseudo, (outs VR256:$dst), (ins), "",
478 [(set VR256:$dst, (v8i32 immAllOnesV))]>;
482 //===----------------------------------------------------------------------===//
483 // SSE 1 & 2 - Move FP Scalar Instructions
485 // Move Instructions. Register-to-register movss/movsd is not used for FR32/64
486 // register copies because it's a partial register update; Register-to-register
487 // movss/movsd is not modeled as an INSERT_SUBREG because INSERT_SUBREG requires
488 // that the insert be implementable in terms of a copy, and just mentioned, we
489 // don't use movss/movsd for copies.
490 //===----------------------------------------------------------------------===//
492 multiclass sse12_move_rr<RegisterClass RC, SDNode OpNode, ValueType vt,
493 X86MemOperand x86memop, string base_opc,
495 def rr : SI<0x10, MRMSrcReg, (outs VR128:$dst),
496 (ins VR128:$src1, RC:$src2),
497 !strconcat(base_opc, asm_opr),
498 [(set VR128:$dst, (vt (OpNode VR128:$src1,
499 (scalar_to_vector RC:$src2))))],
500 IIC_SSE_MOV_S_RR>, Sched<[WriteMove]>;
502 // For the disassembler
503 let isCodeGenOnly = 1, hasSideEffects = 0 in
504 def rr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
505 (ins VR128:$src1, RC:$src2),
506 !strconcat(base_opc, asm_opr),
507 [], IIC_SSE_MOV_S_RR>, Sched<[WriteMove]>;
510 multiclass sse12_move<RegisterClass RC, SDNode OpNode, ValueType vt,
511 X86MemOperand x86memop, string OpcodeStr> {
513 defm V#NAME : sse12_move_rr<RC, OpNode, vt, x86memop, OpcodeStr,
514 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">,
517 def V#NAME#mr : SI<0x11, MRMDestMem, (outs), (ins x86memop:$dst, RC:$src),
518 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
519 [(store RC:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
520 VEX, VEX_LIG, Sched<[WriteStore]>;
522 let Constraints = "$src1 = $dst" in {
523 defm NAME : sse12_move_rr<RC, OpNode, vt, x86memop, OpcodeStr,
524 "\t{$src2, $dst|$dst, $src2}">;
527 def NAME#mr : SI<0x11, MRMDestMem, (outs), (ins x86memop:$dst, RC:$src),
528 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
529 [(store RC:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
533 // Loading from memory automatically zeroing upper bits.
534 multiclass sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
535 PatFrag mem_pat, string OpcodeStr> {
536 def V#NAME#rm : SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
537 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
538 [(set RC:$dst, (mem_pat addr:$src))],
539 IIC_SSE_MOV_S_RM>, VEX, VEX_LIG, Sched<[WriteLoad]>;
540 def NAME#rm : SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
541 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
542 [(set RC:$dst, (mem_pat addr:$src))],
543 IIC_SSE_MOV_S_RM>, Sched<[WriteLoad]>;
546 defm MOVSS : sse12_move<FR32, X86Movss, v4f32, f32mem, "movss">, XS;
547 defm MOVSD : sse12_move<FR64, X86Movsd, v2f64, f64mem, "movsd">, XD;
549 let canFoldAsLoad = 1, isReMaterializable = 1 in {
550 defm MOVSS : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
552 let AddedComplexity = 20 in
553 defm MOVSD : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
557 let Predicates = [UseAVX] in {
558 let AddedComplexity = 15 in {
559 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
560 // MOVS{S,D} to the lower bits.
561 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
562 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
563 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
564 (VMOVSSrr (v4f32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
565 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
566 (VMOVSSrr (v4i32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
567 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
568 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
570 // Move low f32 and clear high bits.
571 def : Pat<(v8f32 (X86vzmovl (v8f32 VR256:$src))),
572 (SUBREG_TO_REG (i32 0),
573 (VMOVSSrr (v4f32 (V_SET0)),
574 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_xmm)), sub_xmm)>;
575 def : Pat<(v8i32 (X86vzmovl (v8i32 VR256:$src))),
576 (SUBREG_TO_REG (i32 0),
577 (VMOVSSrr (v4i32 (V_SET0)),
578 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm)), sub_xmm)>;
581 let AddedComplexity = 20 in {
582 // MOVSSrm zeros the high parts of the register; represent this
583 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
584 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
585 (COPY_TO_REGCLASS (VMOVSSrm addr:$src), VR128)>;
586 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
587 (COPY_TO_REGCLASS (VMOVSSrm addr:$src), VR128)>;
588 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
589 (COPY_TO_REGCLASS (VMOVSSrm addr:$src), VR128)>;
591 // MOVSDrm zeros the high parts of the register; represent this
592 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
593 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
594 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
595 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
596 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
597 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
598 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
599 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
600 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
601 def : Pat<(v2f64 (X86vzload addr:$src)),
602 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
604 // Represent the same patterns above but in the form they appear for
606 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
607 (v4i32 (scalar_to_vector (loadi32 addr:$src))), (iPTR 0)))),
608 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_xmm)>;
609 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
610 (v4f32 (scalar_to_vector (loadf32 addr:$src))), (iPTR 0)))),
611 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_xmm)>;
612 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
613 (v2f64 (scalar_to_vector (loadf64 addr:$src))), (iPTR 0)))),
614 (SUBREG_TO_REG (i32 0), (VMOVSDrm addr:$src), sub_xmm)>;
616 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
617 (v4f32 (scalar_to_vector FR32:$src)), (iPTR 0)))),
618 (SUBREG_TO_REG (i32 0),
619 (v4f32 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)),
621 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
622 (v2f64 (scalar_to_vector FR64:$src)), (iPTR 0)))),
623 (SUBREG_TO_REG (i64 0),
624 (v2f64 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)),
626 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
627 (v2i64 (scalar_to_vector (loadi64 addr:$src))), (iPTR 0)))),
628 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_xmm)>;
630 // Move low f64 and clear high bits.
631 def : Pat<(v4f64 (X86vzmovl (v4f64 VR256:$src))),
632 (SUBREG_TO_REG (i32 0),
633 (VMOVSDrr (v2f64 (V_SET0)),
634 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm)), sub_xmm)>;
636 def : Pat<(v4i64 (X86vzmovl (v4i64 VR256:$src))),
637 (SUBREG_TO_REG (i32 0),
638 (VMOVSDrr (v2i64 (V_SET0)),
639 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm)), sub_xmm)>;
641 // Extract and store.
642 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
644 (VMOVSSmr addr:$dst, (COPY_TO_REGCLASS (v4f32 VR128:$src), FR32))>;
645 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
647 (VMOVSDmr addr:$dst, (COPY_TO_REGCLASS (v2f64 VR128:$src), FR64))>;
649 // Shuffle with VMOVSS
650 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
651 (VMOVSSrr (v4i32 VR128:$src1),
652 (COPY_TO_REGCLASS (v4i32 VR128:$src2), FR32))>;
653 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
654 (VMOVSSrr (v4f32 VR128:$src1),
655 (COPY_TO_REGCLASS (v4f32 VR128:$src2), FR32))>;
658 def : Pat<(v8i32 (X86Movss VR256:$src1, VR256:$src2)),
659 (SUBREG_TO_REG (i32 0),
660 (VMOVSSrr (EXTRACT_SUBREG (v8i32 VR256:$src1), sub_xmm),
661 (EXTRACT_SUBREG (v8i32 VR256:$src2), sub_xmm)),
663 def : Pat<(v8f32 (X86Movss VR256:$src1, VR256:$src2)),
664 (SUBREG_TO_REG (i32 0),
665 (VMOVSSrr (EXTRACT_SUBREG (v8f32 VR256:$src1), sub_xmm),
666 (EXTRACT_SUBREG (v8f32 VR256:$src2), sub_xmm)),
669 // Shuffle with VMOVSD
670 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
671 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
672 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
673 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
674 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
675 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
676 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
677 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
680 def : Pat<(v4i64 (X86Movsd VR256:$src1, VR256:$src2)),
681 (SUBREG_TO_REG (i32 0),
682 (VMOVSDrr (EXTRACT_SUBREG (v4i64 VR256:$src1), sub_xmm),
683 (EXTRACT_SUBREG (v4i64 VR256:$src2), sub_xmm)),
685 def : Pat<(v4f64 (X86Movsd VR256:$src1, VR256:$src2)),
686 (SUBREG_TO_REG (i32 0),
687 (VMOVSDrr (EXTRACT_SUBREG (v4f64 VR256:$src1), sub_xmm),
688 (EXTRACT_SUBREG (v4f64 VR256:$src2), sub_xmm)),
692 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
693 // is during lowering, where it's not possible to recognize the fold cause
694 // it has two uses through a bitcast. One use disappears at isel time and the
695 // fold opportunity reappears.
696 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
697 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
698 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
699 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
700 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
701 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
702 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
703 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
706 let Predicates = [UseSSE1] in {
707 let AddedComplexity = 15 in {
708 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
709 // MOVSS to the lower bits.
710 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
711 (MOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
712 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
713 (MOVSSrr (v4f32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
714 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
715 (MOVSSrr (v4i32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
718 let AddedComplexity = 20 in {
719 // MOVSSrm already zeros the high parts of the register.
720 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
721 (COPY_TO_REGCLASS (MOVSSrm addr:$src), VR128)>;
722 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
723 (COPY_TO_REGCLASS (MOVSSrm addr:$src), VR128)>;
724 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
725 (COPY_TO_REGCLASS (MOVSSrm addr:$src), VR128)>;
728 // Extract and store.
729 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
731 (MOVSSmr addr:$dst, (COPY_TO_REGCLASS VR128:$src, FR32))>;
733 // Shuffle with MOVSS
734 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
735 (MOVSSrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR32))>;
736 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
737 (MOVSSrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR32))>;
740 let Predicates = [UseSSE2] in {
741 let AddedComplexity = 15 in {
742 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
743 // MOVSD to the lower bits.
744 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
745 (MOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
748 let AddedComplexity = 20 in {
749 // MOVSDrm already zeros the high parts of the register.
750 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
751 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
752 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
753 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
754 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
755 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
756 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
757 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
758 def : Pat<(v2f64 (X86vzload addr:$src)),
759 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
762 // Extract and store.
763 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
765 (MOVSDmr addr:$dst, (COPY_TO_REGCLASS VR128:$src, FR64))>;
767 // Shuffle with MOVSD
768 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
769 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
770 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
771 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
772 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
773 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
774 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
775 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
777 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
778 // is during lowering, where it's not possible to recognize the fold cause
779 // it has two uses through a bitcast. One use disappears at isel time and the
780 // fold opportunity reappears.
781 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
782 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
783 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
784 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
785 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
786 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
787 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
788 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
791 //===----------------------------------------------------------------------===//
792 // SSE 1 & 2 - Move Aligned/Unaligned FP Instructions
793 //===----------------------------------------------------------------------===//
795 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
796 X86MemOperand x86memop, PatFrag ld_frag,
797 string asm, Domain d,
799 bit IsReMaterializable = 1> {
800 let neverHasSideEffects = 1 in
801 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
802 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], itins.rr, d>,
804 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
805 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
806 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
807 [(set RC:$dst, (ld_frag addr:$src))], itins.rm, d>,
811 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
812 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
814 defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
815 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
817 defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
818 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
820 defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
821 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
824 defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
825 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
827 defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
828 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
829 TB, OpSize, VEX, VEX_L;
830 defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
831 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
833 defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
834 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
835 TB, OpSize, VEX, VEX_L;
836 defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
837 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
839 defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
840 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
842 defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
843 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
845 defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
846 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
849 let SchedRW = [WriteStore] in {
850 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
851 "movaps\t{$src, $dst|$dst, $src}",
852 [(alignedstore (v4f32 VR128:$src), addr:$dst)],
853 IIC_SSE_MOVA_P_MR>, VEX;
854 def VMOVAPDmr : VPDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
855 "movapd\t{$src, $dst|$dst, $src}",
856 [(alignedstore (v2f64 VR128:$src), addr:$dst)],
857 IIC_SSE_MOVA_P_MR>, VEX;
858 def VMOVUPSmr : VPSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
859 "movups\t{$src, $dst|$dst, $src}",
860 [(store (v4f32 VR128:$src), addr:$dst)],
861 IIC_SSE_MOVU_P_MR>, VEX;
862 def VMOVUPDmr : VPDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
863 "movupd\t{$src, $dst|$dst, $src}",
864 [(store (v2f64 VR128:$src), addr:$dst)],
865 IIC_SSE_MOVU_P_MR>, VEX;
866 def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
867 "movaps\t{$src, $dst|$dst, $src}",
868 [(alignedstore256 (v8f32 VR256:$src), addr:$dst)],
869 IIC_SSE_MOVA_P_MR>, VEX, VEX_L;
870 def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
871 "movapd\t{$src, $dst|$dst, $src}",
872 [(alignedstore256 (v4f64 VR256:$src), addr:$dst)],
873 IIC_SSE_MOVA_P_MR>, VEX, VEX_L;
874 def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
875 "movups\t{$src, $dst|$dst, $src}",
876 [(store (v8f32 VR256:$src), addr:$dst)],
877 IIC_SSE_MOVU_P_MR>, VEX, VEX_L;
878 def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
879 "movupd\t{$src, $dst|$dst, $src}",
880 [(store (v4f64 VR256:$src), addr:$dst)],
881 IIC_SSE_MOVU_P_MR>, VEX, VEX_L;
885 let isCodeGenOnly = 1, hasSideEffects = 0, SchedRW = [WriteMove] in {
886 def VMOVAPSrr_REV : VPSI<0x29, MRMDestReg, (outs VR128:$dst),
888 "movaps\t{$src, $dst|$dst, $src}", [],
889 IIC_SSE_MOVA_P_RR>, VEX;
890 def VMOVAPDrr_REV : VPDI<0x29, MRMDestReg, (outs VR128:$dst),
892 "movapd\t{$src, $dst|$dst, $src}", [],
893 IIC_SSE_MOVA_P_RR>, VEX;
894 def VMOVUPSrr_REV : VPSI<0x11, MRMDestReg, (outs VR128:$dst),
896 "movups\t{$src, $dst|$dst, $src}", [],
897 IIC_SSE_MOVU_P_RR>, VEX;
898 def VMOVUPDrr_REV : VPDI<0x11, MRMDestReg, (outs VR128:$dst),
900 "movupd\t{$src, $dst|$dst, $src}", [],
901 IIC_SSE_MOVU_P_RR>, VEX;
902 def VMOVAPSYrr_REV : VPSI<0x29, MRMDestReg, (outs VR256:$dst),
904 "movaps\t{$src, $dst|$dst, $src}", [],
905 IIC_SSE_MOVA_P_RR>, VEX, VEX_L;
906 def VMOVAPDYrr_REV : VPDI<0x29, MRMDestReg, (outs VR256:$dst),
908 "movapd\t{$src, $dst|$dst, $src}", [],
909 IIC_SSE_MOVA_P_RR>, VEX, VEX_L;
910 def VMOVUPSYrr_REV : VPSI<0x11, MRMDestReg, (outs VR256:$dst),
912 "movups\t{$src, $dst|$dst, $src}", [],
913 IIC_SSE_MOVU_P_RR>, VEX, VEX_L;
914 def VMOVUPDYrr_REV : VPDI<0x11, MRMDestReg, (outs VR256:$dst),
916 "movupd\t{$src, $dst|$dst, $src}", [],
917 IIC_SSE_MOVU_P_RR>, VEX, VEX_L;
920 let Predicates = [HasAVX] in {
921 def : Pat<(v8i32 (X86vzmovl
922 (insert_subvector undef, (v4i32 VR128:$src), (iPTR 0)))),
923 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
924 def : Pat<(v4i64 (X86vzmovl
925 (insert_subvector undef, (v2i64 VR128:$src), (iPTR 0)))),
926 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
927 def : Pat<(v8f32 (X86vzmovl
928 (insert_subvector undef, (v4f32 VR128:$src), (iPTR 0)))),
929 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
930 def : Pat<(v4f64 (X86vzmovl
931 (insert_subvector undef, (v2f64 VR128:$src), (iPTR 0)))),
932 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
936 def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
937 (VMOVUPSYmr addr:$dst, VR256:$src)>;
938 def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
939 (VMOVUPDYmr addr:$dst, VR256:$src)>;
941 let SchedRW = [WriteStore] in {
942 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
943 "movaps\t{$src, $dst|$dst, $src}",
944 [(alignedstore (v4f32 VR128:$src), addr:$dst)],
946 def MOVAPDmr : PDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
947 "movapd\t{$src, $dst|$dst, $src}",
948 [(alignedstore (v2f64 VR128:$src), addr:$dst)],
950 def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
951 "movups\t{$src, $dst|$dst, $src}",
952 [(store (v4f32 VR128:$src), addr:$dst)],
954 def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
955 "movupd\t{$src, $dst|$dst, $src}",
956 [(store (v2f64 VR128:$src), addr:$dst)],
961 let isCodeGenOnly = 1, hasSideEffects = 0, SchedRW = [WriteMove] in {
962 def MOVAPSrr_REV : PSI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
963 "movaps\t{$src, $dst|$dst, $src}", [],
965 def MOVAPDrr_REV : PDI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
966 "movapd\t{$src, $dst|$dst, $src}", [],
968 def MOVUPSrr_REV : PSI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
969 "movups\t{$src, $dst|$dst, $src}", [],
971 def MOVUPDrr_REV : PDI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
972 "movupd\t{$src, $dst|$dst, $src}", [],
976 let Predicates = [HasAVX] in {
977 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
978 (VMOVUPSmr addr:$dst, VR128:$src)>;
979 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
980 (VMOVUPDmr addr:$dst, VR128:$src)>;
983 let Predicates = [UseSSE1] in
984 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
985 (MOVUPSmr addr:$dst, VR128:$src)>;
986 let Predicates = [UseSSE2] in
987 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
988 (MOVUPDmr addr:$dst, VR128:$src)>;
990 // Use vmovaps/vmovups for AVX integer load/store.
991 let Predicates = [HasAVX] in {
992 // 128-bit load/store
993 def : Pat<(alignedloadv2i64 addr:$src),
994 (VMOVAPSrm addr:$src)>;
995 def : Pat<(loadv2i64 addr:$src),
996 (VMOVUPSrm addr:$src)>;
998 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
999 (VMOVAPSmr addr:$dst, VR128:$src)>;
1000 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
1001 (VMOVAPSmr addr:$dst, VR128:$src)>;
1002 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
1003 (VMOVAPSmr addr:$dst, VR128:$src)>;
1004 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
1005 (VMOVAPSmr addr:$dst, VR128:$src)>;
1006 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
1007 (VMOVUPSmr addr:$dst, VR128:$src)>;
1008 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
1009 (VMOVUPSmr addr:$dst, VR128:$src)>;
1010 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
1011 (VMOVUPSmr addr:$dst, VR128:$src)>;
1012 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
1013 (VMOVUPSmr addr:$dst, VR128:$src)>;
1015 // 256-bit load/store
1016 def : Pat<(alignedloadv4i64 addr:$src),
1017 (VMOVAPSYrm addr:$src)>;
1018 def : Pat<(loadv4i64 addr:$src),
1019 (VMOVUPSYrm addr:$src)>;
1020 def : Pat<(alignedstore256 (v4i64 VR256:$src), addr:$dst),
1021 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1022 def : Pat<(alignedstore256 (v8i32 VR256:$src), addr:$dst),
1023 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1024 def : Pat<(alignedstore256 (v16i16 VR256:$src), addr:$dst),
1025 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1026 def : Pat<(alignedstore256 (v32i8 VR256:$src), addr:$dst),
1027 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1028 def : Pat<(store (v4i64 VR256:$src), addr:$dst),
1029 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1030 def : Pat<(store (v8i32 VR256:$src), addr:$dst),
1031 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1032 def : Pat<(store (v16i16 VR256:$src), addr:$dst),
1033 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1034 def : Pat<(store (v32i8 VR256:$src), addr:$dst),
1035 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1037 // Special patterns for storing subvector extracts of lower 128-bits
1038 // Its cheaper to just use VMOVAPS/VMOVUPS instead of VEXTRACTF128mr
1039 def : Pat<(alignedstore (v2f64 (extract_subvector
1040 (v4f64 VR256:$src), (iPTR 0))), addr:$dst),
1041 (VMOVAPDmr addr:$dst, (v2f64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1042 def : Pat<(alignedstore (v4f32 (extract_subvector
1043 (v8f32 VR256:$src), (iPTR 0))), addr:$dst),
1044 (VMOVAPSmr addr:$dst, (v4f32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1045 def : Pat<(alignedstore (v2i64 (extract_subvector
1046 (v4i64 VR256:$src), (iPTR 0))), addr:$dst),
1047 (VMOVAPDmr addr:$dst, (v2i64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1048 def : Pat<(alignedstore (v4i32 (extract_subvector
1049 (v8i32 VR256:$src), (iPTR 0))), addr:$dst),
1050 (VMOVAPSmr addr:$dst, (v4i32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1051 def : Pat<(alignedstore (v8i16 (extract_subvector
1052 (v16i16 VR256:$src), (iPTR 0))), addr:$dst),
1053 (VMOVAPSmr addr:$dst, (v8i16 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1054 def : Pat<(alignedstore (v16i8 (extract_subvector
1055 (v32i8 VR256:$src), (iPTR 0))), addr:$dst),
1056 (VMOVAPSmr addr:$dst, (v16i8 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1058 def : Pat<(store (v2f64 (extract_subvector
1059 (v4f64 VR256:$src), (iPTR 0))), addr:$dst),
1060 (VMOVUPDmr addr:$dst, (v2f64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1061 def : Pat<(store (v4f32 (extract_subvector
1062 (v8f32 VR256:$src), (iPTR 0))), addr:$dst),
1063 (VMOVUPSmr addr:$dst, (v4f32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1064 def : Pat<(store (v2i64 (extract_subvector
1065 (v4i64 VR256:$src), (iPTR 0))), addr:$dst),
1066 (VMOVUPDmr addr:$dst, (v2i64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1067 def : Pat<(store (v4i32 (extract_subvector
1068 (v8i32 VR256:$src), (iPTR 0))), addr:$dst),
1069 (VMOVUPSmr addr:$dst, (v4i32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1070 def : Pat<(store (v8i16 (extract_subvector
1071 (v16i16 VR256:$src), (iPTR 0))), addr:$dst),
1072 (VMOVUPSmr addr:$dst, (v8i16 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1073 def : Pat<(store (v16i8 (extract_subvector
1074 (v32i8 VR256:$src), (iPTR 0))), addr:$dst),
1075 (VMOVUPSmr addr:$dst, (v16i8 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1078 // Use movaps / movups for SSE integer load / store (one byte shorter).
1079 // The instructions selected below are then converted to MOVDQA/MOVDQU
1080 // during the SSE domain pass.
1081 let Predicates = [UseSSE1] in {
1082 def : Pat<(alignedloadv2i64 addr:$src),
1083 (MOVAPSrm addr:$src)>;
1084 def : Pat<(loadv2i64 addr:$src),
1085 (MOVUPSrm addr:$src)>;
1087 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
1088 (MOVAPSmr addr:$dst, VR128:$src)>;
1089 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
1090 (MOVAPSmr addr:$dst, VR128:$src)>;
1091 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
1092 (MOVAPSmr addr:$dst, VR128:$src)>;
1093 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
1094 (MOVAPSmr addr:$dst, VR128:$src)>;
1095 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
1096 (MOVUPSmr addr:$dst, VR128:$src)>;
1097 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
1098 (MOVUPSmr addr:$dst, VR128:$src)>;
1099 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
1100 (MOVUPSmr addr:$dst, VR128:$src)>;
1101 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
1102 (MOVUPSmr addr:$dst, VR128:$src)>;
1105 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
1106 // bits are disregarded. FIXME: Set encoding to pseudo!
1107 let canFoldAsLoad = 1, isReMaterializable = 1, SchedRW = [WriteLoad] in {
1108 let isCodeGenOnly = 1 in {
1109 def FsVMOVAPSrm : VPSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1110 "movaps\t{$src, $dst|$dst, $src}",
1111 [(set FR32:$dst, (alignedloadfsf32 addr:$src))],
1112 IIC_SSE_MOVA_P_RM>, VEX;
1113 def FsVMOVAPDrm : VPDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1114 "movapd\t{$src, $dst|$dst, $src}",
1115 [(set FR64:$dst, (alignedloadfsf64 addr:$src))],
1116 IIC_SSE_MOVA_P_RM>, VEX;
1117 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1118 "movaps\t{$src, $dst|$dst, $src}",
1119 [(set FR32:$dst, (alignedloadfsf32 addr:$src))],
1121 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1122 "movapd\t{$src, $dst|$dst, $src}",
1123 [(set FR64:$dst, (alignedloadfsf64 addr:$src))],
1128 //===----------------------------------------------------------------------===//
1129 // SSE 1 & 2 - Move Low packed FP Instructions
1130 //===----------------------------------------------------------------------===//
1132 multiclass sse12_mov_hilo_packed_base<bits<8>opc, SDNode psnode, SDNode pdnode,
1133 string base_opc, string asm_opr,
1134 InstrItinClass itin> {
1135 def PSrm : PI<opc, MRMSrcMem,
1136 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
1137 !strconcat(base_opc, "s", asm_opr),
1139 (psnode VR128:$src1,
1140 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
1141 itin, SSEPackedSingle>, TB,
1142 Sched<[WriteShuffleLd, ReadAfterLd]>;
1144 def PDrm : PI<opc, MRMSrcMem,
1145 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
1146 !strconcat(base_opc, "d", asm_opr),
1147 [(set VR128:$dst, (v2f64 (pdnode VR128:$src1,
1148 (scalar_to_vector (loadf64 addr:$src2)))))],
1149 itin, SSEPackedDouble>, TB, OpSize,
1150 Sched<[WriteShuffleLd, ReadAfterLd]>;
1154 multiclass sse12_mov_hilo_packed<bits<8>opc, SDNode psnode, SDNode pdnode,
1155 string base_opc, InstrItinClass itin> {
1156 defm V#NAME : sse12_mov_hilo_packed_base<opc, psnode, pdnode, base_opc,
1157 "\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1160 let Constraints = "$src1 = $dst" in
1161 defm NAME : sse12_mov_hilo_packed_base<opc, psnode, pdnode, base_opc,
1162 "\t{$src2, $dst|$dst, $src2}",
1166 let AddedComplexity = 20 in {
1167 defm MOVL : sse12_mov_hilo_packed<0x12, X86Movlps, X86Movlpd, "movlp",
1171 let SchedRW = [WriteStore] in {
1172 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1173 "movlps\t{$src, $dst|$dst, $src}",
1174 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1175 (iPTR 0))), addr:$dst)],
1176 IIC_SSE_MOV_LH>, VEX;
1177 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1178 "movlpd\t{$src, $dst|$dst, $src}",
1179 [(store (f64 (vector_extract (v2f64 VR128:$src),
1180 (iPTR 0))), addr:$dst)],
1181 IIC_SSE_MOV_LH>, VEX;
1182 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1183 "movlps\t{$src, $dst|$dst, $src}",
1184 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1185 (iPTR 0))), addr:$dst)],
1187 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1188 "movlpd\t{$src, $dst|$dst, $src}",
1189 [(store (f64 (vector_extract (v2f64 VR128:$src),
1190 (iPTR 0))), addr:$dst)],
1194 let Predicates = [HasAVX] in {
1195 // Shuffle with VMOVLPS
1196 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1197 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1198 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1199 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1201 // Shuffle with VMOVLPD
1202 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1203 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1204 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1205 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1208 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1210 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1211 def : Pat<(store (v4i32 (X86Movlps
1212 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
1213 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1214 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1216 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1217 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1219 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1222 let Predicates = [UseSSE1] in {
1223 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
1224 def : Pat<(store (i64 (vector_extract (bc_v2i64 (v4f32 VR128:$src2)),
1225 (iPTR 0))), addr:$src1),
1226 (MOVLPSmr addr:$src1, VR128:$src2)>;
1228 // Shuffle with MOVLPS
1229 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1230 (MOVLPSrm VR128:$src1, addr:$src2)>;
1231 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1232 (MOVLPSrm VR128:$src1, addr:$src2)>;
1233 def : Pat<(X86Movlps VR128:$src1,
1234 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1235 (MOVLPSrm VR128:$src1, addr:$src2)>;
1238 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1240 (MOVLPSmr addr:$src1, VR128:$src2)>;
1241 def : Pat<(store (v4i32 (X86Movlps
1242 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
1244 (MOVLPSmr addr:$src1, VR128:$src2)>;
1247 let Predicates = [UseSSE2] in {
1248 // Shuffle with MOVLPD
1249 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1250 (MOVLPDrm VR128:$src1, addr:$src2)>;
1251 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1252 (MOVLPDrm VR128:$src1, addr:$src2)>;
1255 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1257 (MOVLPDmr addr:$src1, VR128:$src2)>;
1258 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1260 (MOVLPDmr addr:$src1, VR128:$src2)>;
1263 //===----------------------------------------------------------------------===//
1264 // SSE 1 & 2 - Move Hi packed FP Instructions
1265 //===----------------------------------------------------------------------===//
1267 let AddedComplexity = 20 in {
1268 defm MOVH : sse12_mov_hilo_packed<0x16, X86Movlhps, X86Movlhpd, "movhp",
1272 let SchedRW = [WriteStore] in {
1273 // v2f64 extract element 1 is always custom lowered to unpack high to low
1274 // and extract element 0 so the non-store version isn't too horrible.
1275 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1276 "movhps\t{$src, $dst|$dst, $src}",
1277 [(store (f64 (vector_extract
1278 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1279 (bc_v2f64 (v4f32 VR128:$src))),
1280 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>, VEX;
1281 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1282 "movhpd\t{$src, $dst|$dst, $src}",
1283 [(store (f64 (vector_extract
1284 (v2f64 (X86Unpckh VR128:$src, VR128:$src)),
1285 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>, VEX;
1286 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1287 "movhps\t{$src, $dst|$dst, $src}",
1288 [(store (f64 (vector_extract
1289 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1290 (bc_v2f64 (v4f32 VR128:$src))),
1291 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>;
1292 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1293 "movhpd\t{$src, $dst|$dst, $src}",
1294 [(store (f64 (vector_extract
1295 (v2f64 (X86Unpckh VR128:$src, VR128:$src)),
1296 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>;
1299 let Predicates = [HasAVX] in {
1301 def : Pat<(X86Movlhps VR128:$src1,
1302 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1303 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1304 def : Pat<(X86Movlhps VR128:$src1,
1305 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1306 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1308 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1309 // is during lowering, where it's not possible to recognize the load fold
1310 // cause it has two uses through a bitcast. One use disappears at isel time
1311 // and the fold opportunity reappears.
1312 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1313 (scalar_to_vector (loadf64 addr:$src2)))),
1314 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1317 let Predicates = [UseSSE1] in {
1319 def : Pat<(X86Movlhps VR128:$src1,
1320 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1321 (MOVHPSrm VR128:$src1, addr:$src2)>;
1322 def : Pat<(X86Movlhps VR128:$src1,
1323 (bc_v4f32 (v2i64 (X86vzload addr:$src2)))),
1324 (MOVHPSrm VR128:$src1, addr:$src2)>;
1327 let Predicates = [UseSSE2] in {
1328 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1329 // is during lowering, where it's not possible to recognize the load fold
1330 // cause it has two uses through a bitcast. One use disappears at isel time
1331 // and the fold opportunity reappears.
1332 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1333 (scalar_to_vector (loadf64 addr:$src2)))),
1334 (MOVHPDrm VR128:$src1, addr:$src2)>;
1337 //===----------------------------------------------------------------------===//
1338 // SSE 1 & 2 - Move Low to High and High to Low packed FP Instructions
1339 //===----------------------------------------------------------------------===//
1341 let AddedComplexity = 20, Predicates = [UseAVX] in {
1342 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
1343 (ins VR128:$src1, VR128:$src2),
1344 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1346 (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))],
1348 VEX_4V, Sched<[WriteShuffle]>;
1349 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
1350 (ins VR128:$src1, VR128:$src2),
1351 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1353 (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))],
1355 VEX_4V, Sched<[WriteShuffle]>;
1357 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1358 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
1359 (ins VR128:$src1, VR128:$src2),
1360 "movlhps\t{$src2, $dst|$dst, $src2}",
1362 (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))],
1363 IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>;
1364 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
1365 (ins VR128:$src1, VR128:$src2),
1366 "movhlps\t{$src2, $dst|$dst, $src2}",
1368 (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))],
1369 IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>;
1372 let Predicates = [UseAVX] in {
1374 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1375 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1376 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1377 (VMOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1380 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1381 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1384 let Predicates = [UseSSE1] in {
1386 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1387 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1388 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1389 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1392 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1393 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1396 //===----------------------------------------------------------------------===//
1397 // SSE 1 & 2 - Conversion Instructions
1398 //===----------------------------------------------------------------------===//
1400 def SSE_CVT_PD : OpndItins<
1401 IIC_SSE_CVT_PD_RR, IIC_SSE_CVT_PD_RM
1404 let Sched = WriteCvtI2F in
1405 def SSE_CVT_PS : OpndItins<
1406 IIC_SSE_CVT_PS_RR, IIC_SSE_CVT_PS_RM
1409 let Sched = WriteCvtI2F in
1410 def SSE_CVT_Scalar : OpndItins<
1411 IIC_SSE_CVT_Scalar_RR, IIC_SSE_CVT_Scalar_RM
1414 let Sched = WriteCvtF2I in
1415 def SSE_CVT_SS2SI_32 : OpndItins<
1416 IIC_SSE_CVT_SS2SI32_RR, IIC_SSE_CVT_SS2SI32_RM
1419 let Sched = WriteCvtF2I in
1420 def SSE_CVT_SS2SI_64 : OpndItins<
1421 IIC_SSE_CVT_SS2SI64_RR, IIC_SSE_CVT_SS2SI64_RM
1424 let Sched = WriteCvtF2I in
1425 def SSE_CVT_SD2SI : OpndItins<
1426 IIC_SSE_CVT_SD2SI_RR, IIC_SSE_CVT_SD2SI_RM
1429 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1430 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1431 string asm, OpndItins itins> {
1432 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1433 [(set DstRC:$dst, (OpNode SrcRC:$src))],
1434 itins.rr>, Sched<[itins.Sched]>;
1435 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1436 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))],
1437 itins.rm>, Sched<[itins.Sched.Folded]>;
1440 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1441 X86MemOperand x86memop, string asm, Domain d,
1443 let neverHasSideEffects = 1 in {
1444 def rr : I<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1445 [], itins.rr, d>, Sched<[itins.Sched]>;
1447 def rm : I<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1448 [], itins.rm, d>, Sched<[itins.Sched.Folded]>;
1452 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1453 X86MemOperand x86memop, string asm> {
1454 let neverHasSideEffects = 1, Predicates = [UseAVX] in {
1455 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
1456 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>,
1457 Sched<[WriteCvtI2F]>;
1459 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1460 (ins DstRC:$src1, x86memop:$src),
1461 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>,
1462 Sched<[WriteCvtI2FLd, ReadAfterLd]>;
1463 } // neverHasSideEffects = 1
1466 let Predicates = [UseAVX] in {
1467 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1468 "cvttss2si\t{$src, $dst|$dst, $src}",
1471 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1472 "cvttss2si\t{$src, $dst|$dst, $src}",
1474 XS, VEX, VEX_W, VEX_LIG;
1475 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1476 "cvttsd2si\t{$src, $dst|$dst, $src}",
1479 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1480 "cvttsd2si\t{$src, $dst|$dst, $src}",
1482 XD, VEX, VEX_W, VEX_LIG;
1484 def : InstAlias<"vcvttss2si{l}\t{$src, $dst|$dst, $src}",
1485 (VCVTTSS2SIrr GR32:$dst, FR32:$src), 0>;
1486 def : InstAlias<"vcvttss2si{l}\t{$src, $dst|$dst, $src}",
1487 (VCVTTSS2SIrm GR32:$dst, f32mem:$src), 0>;
1488 def : InstAlias<"vcvttsd2si{l}\t{$src, $dst|$dst, $src}",
1489 (VCVTTSD2SIrr GR32:$dst, FR64:$src), 0>;
1490 def : InstAlias<"vcvttsd2si{l}\t{$src, $dst|$dst, $src}",
1491 (VCVTTSD2SIrm GR32:$dst, f64mem:$src), 0>;
1492 def : InstAlias<"vcvttss2si{q}\t{$src, $dst|$dst, $src}",
1493 (VCVTTSS2SI64rr GR64:$dst, FR32:$src), 0>;
1494 def : InstAlias<"vcvttss2si{q}\t{$src, $dst|$dst, $src}",
1495 (VCVTTSS2SI64rm GR64:$dst, f32mem:$src), 0>;
1496 def : InstAlias<"vcvttsd2si{q}\t{$src, $dst|$dst, $src}",
1497 (VCVTTSD2SI64rr GR64:$dst, FR64:$src), 0>;
1498 def : InstAlias<"vcvttsd2si{q}\t{$src, $dst|$dst, $src}",
1499 (VCVTTSD2SI64rm GR64:$dst, f64mem:$src), 0>;
1501 // The assembler can recognize rr 64-bit instructions by seeing a rxx
1502 // register, but the same isn't true when only using memory operands,
1503 // provide other assembly "l" and "q" forms to address this explicitly
1504 // where appropriate to do so.
1505 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss{l}">,
1506 XS, VEX_4V, VEX_LIG;
1507 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">,
1508 XS, VEX_4V, VEX_W, VEX_LIG;
1509 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">,
1510 XD, VEX_4V, VEX_LIG;
1511 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">,
1512 XD, VEX_4V, VEX_W, VEX_LIG;
1514 let Predicates = [UseAVX] in {
1515 def : InstAlias<"vcvtsi2ss\t{$src, $src1, $dst|$dst, $src1, $src}",
1516 (VCVTSI2SSrm FR64:$dst, FR64:$src1, i32mem:$src)>;
1517 def : InstAlias<"vcvtsi2sd\t{$src, $src1, $dst|$dst, $src1, $src}",
1518 (VCVTSI2SDrm FR64:$dst, FR64:$src1, i32mem:$src)>;
1520 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
1521 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1522 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
1523 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
1524 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
1525 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
1526 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
1527 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
1529 def : Pat<(f32 (sint_to_fp GR32:$src)),
1530 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
1531 def : Pat<(f32 (sint_to_fp GR64:$src)),
1532 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
1533 def : Pat<(f64 (sint_to_fp GR32:$src)),
1534 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
1535 def : Pat<(f64 (sint_to_fp GR64:$src)),
1536 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
1539 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1540 "cvttss2si\t{$src, $dst|$dst, $src}",
1541 SSE_CVT_SS2SI_32>, XS;
1542 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1543 "cvttss2si\t{$src, $dst|$dst, $src}",
1544 SSE_CVT_SS2SI_64>, XS, REX_W;
1545 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1546 "cvttsd2si\t{$src, $dst|$dst, $src}",
1548 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1549 "cvttsd2si\t{$src, $dst|$dst, $src}",
1550 SSE_CVT_SD2SI>, XD, REX_W;
1551 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
1552 "cvtsi2ss{l}\t{$src, $dst|$dst, $src}",
1553 SSE_CVT_Scalar>, XS;
1554 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
1555 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}",
1556 SSE_CVT_Scalar>, XS, REX_W;
1557 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
1558 "cvtsi2sd{l}\t{$src, $dst|$dst, $src}",
1559 SSE_CVT_Scalar>, XD;
1560 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
1561 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}",
1562 SSE_CVT_Scalar>, XD, REX_W;
1564 def : InstAlias<"cvttss2si{l}\t{$src, $dst|$dst, $src}",
1565 (CVTTSS2SIrr GR32:$dst, FR32:$src), 0>;
1566 def : InstAlias<"cvttss2si{l}\t{$src, $dst|$dst, $src}",
1567 (CVTTSS2SIrm GR32:$dst, f32mem:$src), 0>;
1568 def : InstAlias<"cvttsd2si{l}\t{$src, $dst|$dst, $src}",
1569 (CVTTSD2SIrr GR32:$dst, FR64:$src), 0>;
1570 def : InstAlias<"cvttsd2si{l}\t{$src, $dst|$dst, $src}",
1571 (CVTTSD2SIrm GR32:$dst, f64mem:$src), 0>;
1572 def : InstAlias<"cvttss2si{q}\t{$src, $dst|$dst, $src}",
1573 (CVTTSS2SI64rr GR64:$dst, FR32:$src), 0>;
1574 def : InstAlias<"cvttss2si{q}\t{$src, $dst|$dst, $src}",
1575 (CVTTSS2SI64rm GR64:$dst, f32mem:$src), 0>;
1576 def : InstAlias<"cvttsd2si{q}\t{$src, $dst|$dst, $src}",
1577 (CVTTSD2SI64rr GR64:$dst, FR64:$src), 0>;
1578 def : InstAlias<"cvttsd2si{q}\t{$src, $dst|$dst, $src}",
1579 (CVTTSD2SI64rm GR64:$dst, f64mem:$src), 0>;
1581 def : InstAlias<"cvtsi2ss\t{$src, $dst|$dst, $src}",
1582 (CVTSI2SSrm FR64:$dst, i32mem:$src)>;
1583 def : InstAlias<"cvtsi2sd\t{$src, $dst|$dst, $src}",
1584 (CVTSI2SDrm FR64:$dst, i32mem:$src)>;
1586 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
1587 // and/or XMM operand(s).
1589 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1590 Intrinsic Int, Operand memop, ComplexPattern mem_cpat,
1591 string asm, OpndItins itins> {
1592 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
1593 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1594 [(set DstRC:$dst, (Int SrcRC:$src))], itins.rr>,
1595 Sched<[itins.Sched]>;
1596 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins memop:$src),
1597 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1598 [(set DstRC:$dst, (Int mem_cpat:$src))], itins.rm>,
1599 Sched<[itins.Sched.Folded]>;
1602 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
1603 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
1604 PatFrag ld_frag, string asm, OpndItins itins,
1606 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
1608 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1609 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1610 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))],
1611 itins.rr>, Sched<[itins.Sched]>;
1612 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1613 (ins DstRC:$src1, x86memop:$src2),
1615 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1616 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1617 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))],
1618 itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
1621 let Predicates = [UseAVX] in {
1622 defm VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32,
1623 int_x86_sse2_cvtsd2si, sdmem, sse_load_f64, "cvtsd2si",
1624 SSE_CVT_SD2SI>, XD, VEX, VEX_LIG;
1625 defm VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
1626 int_x86_sse2_cvtsd2si64, sdmem, sse_load_f64, "cvtsd2si",
1627 SSE_CVT_SD2SI>, XD, VEX, VEX_W, VEX_LIG;
1629 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1630 sdmem, sse_load_f64, "cvtsd2si", SSE_CVT_SD2SI>, XD;
1631 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
1632 sdmem, sse_load_f64, "cvtsd2si", SSE_CVT_SD2SI>, XD, REX_W;
1635 let Predicates = [UseAVX] in {
1636 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1637 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss{l}",
1638 SSE_CVT_Scalar, 0>, XS, VEX_4V;
1639 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1640 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss{q}",
1641 SSE_CVT_Scalar, 0>, XS, VEX_4V,
1643 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1644 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd{l}",
1645 SSE_CVT_Scalar, 0>, XD, VEX_4V;
1646 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1647 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd{q}",
1648 SSE_CVT_Scalar, 0>, XD,
1651 let Constraints = "$src1 = $dst" in {
1652 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1653 int_x86_sse_cvtsi2ss, i32mem, loadi32,
1654 "cvtsi2ss{l}", SSE_CVT_Scalar>, XS;
1655 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1656 int_x86_sse_cvtsi642ss, i64mem, loadi64,
1657 "cvtsi2ss{q}", SSE_CVT_Scalar>, XS, REX_W;
1658 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1659 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
1660 "cvtsi2sd{l}", SSE_CVT_Scalar>, XD;
1661 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1662 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
1663 "cvtsi2sd{q}", SSE_CVT_Scalar>, XD, REX_W;
1668 // Aliases for intrinsics
1669 let Predicates = [UseAVX] in {
1670 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1671 ssmem, sse_load_f32, "cvttss2si",
1672 SSE_CVT_SS2SI_32>, XS, VEX;
1673 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1674 int_x86_sse_cvttss2si64, ssmem, sse_load_f32,
1675 "cvttss2si", SSE_CVT_SS2SI_64>,
1677 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1678 sdmem, sse_load_f64, "cvttsd2si",
1679 SSE_CVT_SD2SI>, XD, VEX;
1680 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1681 int_x86_sse2_cvttsd2si64, sdmem, sse_load_f64,
1682 "cvttsd2si", SSE_CVT_SD2SI>,
1685 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1686 ssmem, sse_load_f32, "cvttss2si",
1687 SSE_CVT_SS2SI_32>, XS;
1688 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1689 int_x86_sse_cvttss2si64, ssmem, sse_load_f32,
1690 "cvttss2si", SSE_CVT_SS2SI_64>, XS, REX_W;
1691 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1692 sdmem, sse_load_f64, "cvttsd2si",
1694 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1695 int_x86_sse2_cvttsd2si64, sdmem, sse_load_f64,
1696 "cvttsd2si", SSE_CVT_SD2SI>, XD, REX_W;
1698 let Predicates = [UseAVX] in {
1699 defm VCVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
1700 ssmem, sse_load_f32, "cvtss2si",
1701 SSE_CVT_SS2SI_32>, XS, VEX, VEX_LIG;
1702 defm VCVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
1703 ssmem, sse_load_f32, "cvtss2si",
1704 SSE_CVT_SS2SI_64>, XS, VEX, VEX_W, VEX_LIG;
1706 defm CVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
1707 ssmem, sse_load_f32, "cvtss2si",
1708 SSE_CVT_SS2SI_32>, XS;
1709 defm CVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
1710 ssmem, sse_load_f32, "cvtss2si",
1711 SSE_CVT_SS2SI_64>, XS, REX_W;
1713 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, i128mem,
1714 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1715 SSEPackedSingle, SSE_CVT_PS>,
1716 TB, VEX, Requires<[HasAVX]>;
1717 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, i256mem,
1718 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1719 SSEPackedSingle, SSE_CVT_PS>,
1720 TB, VEX, VEX_L, Requires<[HasAVX]>;
1722 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, i128mem,
1723 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1724 SSEPackedSingle, SSE_CVT_PS>,
1725 TB, Requires<[UseSSE2]>;
1727 let Predicates = [UseAVX] in {
1728 def : InstAlias<"vcvtss2si{l}\t{$src, $dst|$dst, $src}",
1729 (VCVTSS2SIrr GR32:$dst, VR128:$src), 0>;
1730 def : InstAlias<"vcvtss2si{l}\t{$src, $dst|$dst, $src}",
1731 (VCVTSS2SIrm GR32:$dst, ssmem:$src), 0>;
1732 def : InstAlias<"vcvtsd2si{l}\t{$src, $dst|$dst, $src}",
1733 (VCVTSD2SIrr GR32:$dst, VR128:$src), 0>;
1734 def : InstAlias<"vcvtsd2si{l}\t{$src, $dst|$dst, $src}",
1735 (VCVTSD2SIrm GR32:$dst, sdmem:$src), 0>;
1736 def : InstAlias<"vcvtss2si{q}\t{$src, $dst|$dst, $src}",
1737 (VCVTSS2SI64rr GR64:$dst, VR128:$src), 0>;
1738 def : InstAlias<"vcvtss2si{q}\t{$src, $dst|$dst, $src}",
1739 (VCVTSS2SI64rm GR64:$dst, ssmem:$src), 0>;
1740 def : InstAlias<"vcvtsd2si{q}\t{$src, $dst|$dst, $src}",
1741 (VCVTSD2SI64rr GR64:$dst, VR128:$src), 0>;
1742 def : InstAlias<"vcvtsd2si{q}\t{$src, $dst|$dst, $src}",
1743 (VCVTSD2SI64rm GR64:$dst, sdmem:$src), 0>;
1746 def : InstAlias<"cvtss2si{l}\t{$src, $dst|$dst, $src}",
1747 (CVTSS2SIrr GR32:$dst, VR128:$src), 0>;
1748 def : InstAlias<"cvtss2si{l}\t{$src, $dst|$dst, $src}",
1749 (CVTSS2SIrm GR32:$dst, ssmem:$src), 0>;
1750 def : InstAlias<"cvtsd2si{l}\t{$src, $dst|$dst, $src}",
1751 (CVTSD2SIrr GR32:$dst, VR128:$src), 0>;
1752 def : InstAlias<"cvtsd2si{l}\t{$src, $dst|$dst, $src}",
1753 (CVTSD2SIrm GR32:$dst, sdmem:$src), 0>;
1754 def : InstAlias<"cvtss2si{q}\t{$src, $dst|$dst, $src}",
1755 (CVTSS2SI64rr GR64:$dst, VR128:$src), 0>;
1756 def : InstAlias<"cvtss2si{q}\t{$src, $dst|$dst, $src}",
1757 (CVTSS2SI64rm GR64:$dst, ssmem:$src), 0>;
1758 def : InstAlias<"cvtsd2si{q}\t{$src, $dst|$dst, $src}",
1759 (CVTSD2SI64rr GR64:$dst, VR128:$src), 0>;
1760 def : InstAlias<"cvtsd2si{q}\t{$src, $dst|$dst, $src}",
1761 (CVTSD2SI64rm GR64:$dst, sdmem:$src)>;
1765 // Convert scalar double to scalar single
1766 let neverHasSideEffects = 1, Predicates = [UseAVX] in {
1767 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
1768 (ins FR64:$src1, FR64:$src2),
1769 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
1770 IIC_SSE_CVT_Scalar_RR>, VEX_4V, VEX_LIG,
1771 Sched<[WriteCvtF2F]>;
1773 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
1774 (ins FR64:$src1, f64mem:$src2),
1775 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1776 [], IIC_SSE_CVT_Scalar_RM>,
1777 XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG,
1778 Sched<[WriteCvtF2FLd, ReadAfterLd]>;
1781 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
1784 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
1785 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1786 [(set FR32:$dst, (fround FR64:$src))],
1787 IIC_SSE_CVT_Scalar_RR>, Sched<[WriteCvtF2F]>;
1788 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
1789 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1790 [(set FR32:$dst, (fround (loadf64 addr:$src)))],
1791 IIC_SSE_CVT_Scalar_RM>,
1793 Requires<[UseSSE2, OptForSize]>, Sched<[WriteCvtF2FLd]>;
1795 def Int_VCVTSD2SSrr: I<0x5A, MRMSrcReg,
1796 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1797 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1799 (int_x86_sse2_cvtsd2ss VR128:$src1, VR128:$src2))],
1800 IIC_SSE_CVT_Scalar_RR>, XD, VEX_4V, Requires<[UseAVX]>,
1801 Sched<[WriteCvtF2F]>;
1802 def Int_VCVTSD2SSrm: I<0x5A, MRMSrcReg,
1803 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2),
1804 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1805 [(set VR128:$dst, (int_x86_sse2_cvtsd2ss
1806 VR128:$src1, sse_load_f64:$src2))],
1807 IIC_SSE_CVT_Scalar_RM>, XD, VEX_4V, Requires<[UseAVX]>,
1808 Sched<[WriteCvtF2FLd, ReadAfterLd]>;
1810 let Constraints = "$src1 = $dst" in {
1811 def Int_CVTSD2SSrr: I<0x5A, MRMSrcReg,
1812 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1813 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1815 (int_x86_sse2_cvtsd2ss VR128:$src1, VR128:$src2))],
1816 IIC_SSE_CVT_Scalar_RR>, XD, Requires<[UseSSE2]>,
1817 Sched<[WriteCvtF2F]>;
1818 def Int_CVTSD2SSrm: I<0x5A, MRMSrcReg,
1819 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2),
1820 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1821 [(set VR128:$dst, (int_x86_sse2_cvtsd2ss
1822 VR128:$src1, sse_load_f64:$src2))],
1823 IIC_SSE_CVT_Scalar_RM>, XD, Requires<[UseSSE2]>,
1824 Sched<[WriteCvtF2FLd, ReadAfterLd]>;
1827 // Convert scalar single to scalar double
1828 // SSE2 instructions with XS prefix
1829 let neverHasSideEffects = 1, Predicates = [UseAVX] in {
1830 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
1831 (ins FR32:$src1, FR32:$src2),
1832 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1833 [], IIC_SSE_CVT_Scalar_RR>,
1834 XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG,
1835 Sched<[WriteCvtF2F]>;
1837 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
1838 (ins FR32:$src1, f32mem:$src2),
1839 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1840 [], IIC_SSE_CVT_Scalar_RM>,
1841 XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>,
1842 Sched<[WriteCvtF2FLd, ReadAfterLd]>;
1845 def : Pat<(f64 (fextend FR32:$src)),
1846 (VCVTSS2SDrr FR32:$src, FR32:$src)>, Requires<[UseAVX]>;
1847 def : Pat<(fextend (loadf32 addr:$src)),
1848 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>, Requires<[UseAVX]>;
1850 def : Pat<(extloadf32 addr:$src),
1851 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>,
1852 Requires<[UseAVX, OptForSize]>;
1853 def : Pat<(extloadf32 addr:$src),
1854 (VCVTSS2SDrr (f32 (IMPLICIT_DEF)), (VMOVSSrm addr:$src))>,
1855 Requires<[UseAVX, OptForSpeed]>;
1857 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
1858 "cvtss2sd\t{$src, $dst|$dst, $src}",
1859 [(set FR64:$dst, (fextend FR32:$src))],
1860 IIC_SSE_CVT_Scalar_RR>, XS,
1861 Requires<[UseSSE2]>, Sched<[WriteCvtF2F]>;
1862 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
1863 "cvtss2sd\t{$src, $dst|$dst, $src}",
1864 [(set FR64:$dst, (extloadf32 addr:$src))],
1865 IIC_SSE_CVT_Scalar_RM>, XS,
1866 Requires<[UseSSE2, OptForSize]>, Sched<[WriteCvtF2FLd]>;
1868 // extload f32 -> f64. This matches load+fextend because we have a hack in
1869 // the isel (PreprocessForFPConvert) that can introduce loads after dag
1871 // Since these loads aren't folded into the fextend, we have to match it
1873 def : Pat<(fextend (loadf32 addr:$src)),
1874 (CVTSS2SDrm addr:$src)>, Requires<[UseSSE2]>;
1875 def : Pat<(extloadf32 addr:$src),
1876 (CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[UseSSE2, OptForSpeed]>;
1878 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
1879 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1880 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1882 (int_x86_sse2_cvtss2sd VR128:$src1, VR128:$src2))],
1883 IIC_SSE_CVT_Scalar_RR>, XS, VEX_4V, Requires<[UseAVX]>,
1884 Sched<[WriteCvtF2F]>;
1885 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
1886 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2),
1887 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1889 (int_x86_sse2_cvtss2sd VR128:$src1, sse_load_f32:$src2))],
1890 IIC_SSE_CVT_Scalar_RM>, XS, VEX_4V, Requires<[UseAVX]>,
1891 Sched<[WriteCvtF2FLd, ReadAfterLd]>;
1892 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
1893 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
1894 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1895 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1897 (int_x86_sse2_cvtss2sd VR128:$src1, VR128:$src2))],
1898 IIC_SSE_CVT_Scalar_RR>, XS, Requires<[UseSSE2]>,
1899 Sched<[WriteCvtF2F]>;
1900 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
1901 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2),
1902 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1904 (int_x86_sse2_cvtss2sd VR128:$src1, sse_load_f32:$src2))],
1905 IIC_SSE_CVT_Scalar_RM>, XS, Requires<[UseSSE2]>,
1906 Sched<[WriteCvtF2FLd, ReadAfterLd]>;
1909 // Convert packed single/double fp to doubleword
1910 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1911 "cvtps2dq\t{$src, $dst|$dst, $src}",
1912 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))],
1913 IIC_SSE_CVT_PS_RR>, VEX, Sched<[WriteCvtF2I]>;
1914 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1915 "cvtps2dq\t{$src, $dst|$dst, $src}",
1917 (int_x86_sse2_cvtps2dq (memopv4f32 addr:$src)))],
1918 IIC_SSE_CVT_PS_RM>, VEX, Sched<[WriteCvtF2ILd]>;
1919 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1920 "cvtps2dq\t{$src, $dst|$dst, $src}",
1922 (int_x86_avx_cvt_ps2dq_256 VR256:$src))],
1923 IIC_SSE_CVT_PS_RR>, VEX, VEX_L, Sched<[WriteCvtF2I]>;
1924 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1925 "cvtps2dq\t{$src, $dst|$dst, $src}",
1927 (int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)))],
1928 IIC_SSE_CVT_PS_RM>, VEX, VEX_L, Sched<[WriteCvtF2ILd]>;
1929 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1930 "cvtps2dq\t{$src, $dst|$dst, $src}",
1931 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))],
1932 IIC_SSE_CVT_PS_RR>, Sched<[WriteCvtF2I]>;
1933 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1934 "cvtps2dq\t{$src, $dst|$dst, $src}",
1936 (int_x86_sse2_cvtps2dq (memopv4f32 addr:$src)))],
1937 IIC_SSE_CVT_PS_RM>, Sched<[WriteCvtF2ILd]>;
1940 // Convert Packed Double FP to Packed DW Integers
1941 let Predicates = [HasAVX] in {
1942 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1943 // register, but the same isn't true when using memory operands instead.
1944 // Provide other assembly rr and rm forms to address this explicitly.
1945 def VCVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1946 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1947 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1948 VEX, Sched<[WriteCvtF2I]>;
1951 def : InstAlias<"vcvtpd2dqx\t{$src, $dst|$dst, $src}",
1952 (VCVTPD2DQrr VR128:$dst, VR128:$src)>;
1953 def VCVTPD2DQXrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1954 "vcvtpd2dqx\t{$src, $dst|$dst, $src}",
1956 (int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)))]>, VEX,
1957 Sched<[WriteCvtF2ILd]>;
1960 def VCVTPD2DQYrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1961 "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}",
1963 (int_x86_avx_cvt_pd2dq_256 VR256:$src))]>, VEX, VEX_L,
1964 Sched<[WriteCvtF2I]>;
1965 def VCVTPD2DQYrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1966 "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}",
1968 (int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)))]>,
1969 VEX, VEX_L, Sched<[WriteCvtF2ILd]>;
1970 def : InstAlias<"vcvtpd2dq\t{$src, $dst|$dst, $src}",
1971 (VCVTPD2DQYrr VR128:$dst, VR256:$src)>;
1974 def CVTPD2DQrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1975 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1977 (int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)))],
1978 IIC_SSE_CVT_PD_RM>, Sched<[WriteCvtF2ILd]>;
1979 def CVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1980 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1981 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))],
1982 IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtF2I]>;
1984 // Convert with truncation packed single/double fp to doubleword
1985 // SSE2 packed instructions with XS prefix
1986 def VCVTTPS2DQrr : VS2SI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1987 "cvttps2dq\t{$src, $dst|$dst, $src}",
1989 (int_x86_sse2_cvttps2dq VR128:$src))],
1990 IIC_SSE_CVT_PS_RR>, VEX, Sched<[WriteCvtF2I]>;
1991 def VCVTTPS2DQrm : VS2SI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1992 "cvttps2dq\t{$src, $dst|$dst, $src}",
1993 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
1994 (memopv4f32 addr:$src)))],
1995 IIC_SSE_CVT_PS_RM>, VEX, Sched<[WriteCvtF2ILd]>;
1996 def VCVTTPS2DQYrr : VS2SI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1997 "cvttps2dq\t{$src, $dst|$dst, $src}",
1999 (int_x86_avx_cvtt_ps2dq_256 VR256:$src))],
2000 IIC_SSE_CVT_PS_RR>, VEX, VEX_L, Sched<[WriteCvtF2I]>;
2001 def VCVTTPS2DQYrm : VS2SI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2002 "cvttps2dq\t{$src, $dst|$dst, $src}",
2003 [(set VR256:$dst, (int_x86_avx_cvtt_ps2dq_256
2004 (memopv8f32 addr:$src)))],
2005 IIC_SSE_CVT_PS_RM>, VEX, VEX_L,
2006 Sched<[WriteCvtF2ILd]>;
2008 def CVTTPS2DQrr : S2SI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2009 "cvttps2dq\t{$src, $dst|$dst, $src}",
2010 [(set VR128:$dst, (int_x86_sse2_cvttps2dq VR128:$src))],
2011 IIC_SSE_CVT_PS_RR>, Sched<[WriteCvtF2I]>;
2012 def CVTTPS2DQrm : S2SI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2013 "cvttps2dq\t{$src, $dst|$dst, $src}",
2015 (int_x86_sse2_cvttps2dq (memopv4f32 addr:$src)))],
2016 IIC_SSE_CVT_PS_RM>, Sched<[WriteCvtF2ILd]>;
2018 let Predicates = [HasAVX] in {
2019 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
2020 (VCVTDQ2PSrr VR128:$src)>;
2021 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
2022 (VCVTDQ2PSrm addr:$src)>;
2024 def : Pat<(int_x86_sse2_cvtdq2ps VR128:$src),
2025 (VCVTDQ2PSrr VR128:$src)>;
2026 def : Pat<(int_x86_sse2_cvtdq2ps (bc_v4i32 (memopv2i64 addr:$src))),
2027 (VCVTDQ2PSrm addr:$src)>;
2029 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
2030 (VCVTTPS2DQrr VR128:$src)>;
2031 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
2032 (VCVTTPS2DQrm addr:$src)>;
2034 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
2035 (VCVTDQ2PSYrr VR256:$src)>;
2036 def : Pat<(v8f32 (sint_to_fp (bc_v8i32 (memopv4i64 addr:$src)))),
2037 (VCVTDQ2PSYrm addr:$src)>;
2039 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
2040 (VCVTTPS2DQYrr VR256:$src)>;
2041 def : Pat<(v8i32 (fp_to_sint (memopv8f32 addr:$src))),
2042 (VCVTTPS2DQYrm addr:$src)>;
2045 let Predicates = [UseSSE2] in {
2046 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
2047 (CVTDQ2PSrr VR128:$src)>;
2048 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
2049 (CVTDQ2PSrm addr:$src)>;
2051 def : Pat<(int_x86_sse2_cvtdq2ps VR128:$src),
2052 (CVTDQ2PSrr VR128:$src)>;
2053 def : Pat<(int_x86_sse2_cvtdq2ps (bc_v4i32 (memopv2i64 addr:$src))),
2054 (CVTDQ2PSrm addr:$src)>;
2056 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
2057 (CVTTPS2DQrr VR128:$src)>;
2058 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
2059 (CVTTPS2DQrm addr:$src)>;
2062 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2063 "cvttpd2dq\t{$src, $dst|$dst, $src}",
2065 (int_x86_sse2_cvttpd2dq VR128:$src))],
2066 IIC_SSE_CVT_PD_RR>, VEX, Sched<[WriteCvtF2I]>;
2068 // The assembler can recognize rr 256-bit instructions by seeing a ymm
2069 // register, but the same isn't true when using memory operands instead.
2070 // Provide other assembly rr and rm forms to address this explicitly.
2073 def : InstAlias<"vcvttpd2dqx\t{$src, $dst|$dst, $src}",
2074 (VCVTTPD2DQrr VR128:$dst, VR128:$src)>;
2075 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2076 "cvttpd2dqx\t{$src, $dst|$dst, $src}",
2077 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
2078 (memopv2f64 addr:$src)))],
2079 IIC_SSE_CVT_PD_RM>, VEX, Sched<[WriteCvtF2ILd]>;
2082 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
2083 "cvttpd2dq{y}\t{$src, $dst|$dst, $src}",
2085 (int_x86_avx_cvtt_pd2dq_256 VR256:$src))],
2086 IIC_SSE_CVT_PD_RR>, VEX, VEX_L, Sched<[WriteCvtF2I]>;
2087 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
2088 "cvttpd2dq{y}\t{$src, $dst|$dst, $src}",
2090 (int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)))],
2091 IIC_SSE_CVT_PD_RM>, VEX, VEX_L, Sched<[WriteCvtF2ILd]>;
2092 def : InstAlias<"vcvttpd2dq\t{$src, $dst|$dst, $src}",
2093 (VCVTTPD2DQYrr VR128:$dst, VR256:$src)>;
2095 let Predicates = [HasAVX] in {
2096 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
2097 (VCVTTPD2DQYrr VR256:$src)>;
2098 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
2099 (VCVTTPD2DQYrm addr:$src)>;
2100 } // Predicates = [HasAVX]
2102 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2103 "cvttpd2dq\t{$src, $dst|$dst, $src}",
2104 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))],
2105 IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtF2I]>;
2106 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
2107 "cvttpd2dq\t{$src, $dst|$dst, $src}",
2108 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
2109 (memopv2f64 addr:$src)))],
2111 Sched<[WriteCvtF2ILd]>;
2113 // Convert packed single to packed double
2114 let Predicates = [HasAVX] in {
2115 // SSE2 instructions without OpSize prefix
2116 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2117 "vcvtps2pd\t{$src, $dst|$dst, $src}",
2118 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))],
2119 IIC_SSE_CVT_PD_RR>, TB, VEX, Sched<[WriteCvtF2F]>;
2120 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
2121 "vcvtps2pd\t{$src, $dst|$dst, $src}",
2122 [(set VR128:$dst, (v2f64 (extloadv2f32 addr:$src)))],
2123 IIC_SSE_CVT_PD_RM>, TB, VEX, Sched<[WriteCvtF2FLd]>;
2124 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
2125 "vcvtps2pd\t{$src, $dst|$dst, $src}",
2127 (int_x86_avx_cvt_ps2_pd_256 VR128:$src))],
2128 IIC_SSE_CVT_PD_RR>, TB, VEX, VEX_L, Sched<[WriteCvtF2F]>;
2129 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
2130 "vcvtps2pd\t{$src, $dst|$dst, $src}",
2132 (int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)))],
2133 IIC_SSE_CVT_PD_RM>, TB, VEX, VEX_L, Sched<[WriteCvtF2FLd]>;
2136 let Predicates = [UseSSE2] in {
2137 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2138 "cvtps2pd\t{$src, $dst|$dst, $src}",
2139 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))],
2140 IIC_SSE_CVT_PD_RR>, TB, Sched<[WriteCvtF2F]>;
2141 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
2142 "cvtps2pd\t{$src, $dst|$dst, $src}",
2143 [(set VR128:$dst, (v2f64 (extloadv2f32 addr:$src)))],
2144 IIC_SSE_CVT_PD_RM>, TB, Sched<[WriteCvtF2FLd]>;
2147 // Convert Packed DW Integers to Packed Double FP
2148 let Predicates = [HasAVX] in {
2149 let neverHasSideEffects = 1, mayLoad = 1 in
2150 def VCVTDQ2PDrm : S2SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2151 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2152 []>, VEX, Sched<[WriteCvtI2FLd]>;
2153 def VCVTDQ2PDrr : S2SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2154 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2156 (int_x86_sse2_cvtdq2pd VR128:$src))]>, VEX,
2157 Sched<[WriteCvtI2F]>;
2158 def VCVTDQ2PDYrm : S2SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
2159 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2161 (int_x86_avx_cvtdq2_pd_256
2162 (bitconvert (memopv2i64 addr:$src))))]>, VEX, VEX_L,
2163 Sched<[WriteCvtI2FLd]>;
2164 def VCVTDQ2PDYrr : S2SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
2165 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2167 (int_x86_avx_cvtdq2_pd_256 VR128:$src))]>, VEX, VEX_L,
2168 Sched<[WriteCvtI2F]>;
2171 let neverHasSideEffects = 1, mayLoad = 1 in
2172 def CVTDQ2PDrm : S2SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2173 "cvtdq2pd\t{$src, $dst|$dst, $src}", [],
2174 IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtI2FLd]>;
2175 def CVTDQ2PDrr : S2SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2176 "cvtdq2pd\t{$src, $dst|$dst, $src}",
2177 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))],
2178 IIC_SSE_CVT_PD_RM>, Sched<[WriteCvtI2F]>;
2180 // AVX 256-bit register conversion intrinsics
2181 let Predicates = [HasAVX] in {
2182 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
2183 (VCVTDQ2PDYrr VR128:$src)>;
2184 def : Pat<(v4f64 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
2185 (VCVTDQ2PDYrm addr:$src)>;
2186 } // Predicates = [HasAVX]
2188 // Convert packed double to packed single
2189 // The assembler can recognize rr 256-bit instructions by seeing a ymm
2190 // register, but the same isn't true when using memory operands instead.
2191 // Provide other assembly rr and rm forms to address this explicitly.
2192 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2193 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2194 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))],
2195 IIC_SSE_CVT_PD_RR>, VEX, Sched<[WriteCvtF2F]>;
2198 def : InstAlias<"vcvtpd2psx\t{$src, $dst|$dst, $src}",
2199 (VCVTPD2PSrr VR128:$dst, VR128:$src)>;
2200 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2201 "cvtpd2psx\t{$src, $dst|$dst, $src}",
2203 (int_x86_sse2_cvtpd2ps (memopv2f64 addr:$src)))],
2204 IIC_SSE_CVT_PD_RM>, VEX, Sched<[WriteCvtF2FLd]>;
2207 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
2208 "cvtpd2ps{y}\t{$src, $dst|$dst, $src}",
2210 (int_x86_avx_cvt_pd2_ps_256 VR256:$src))],
2211 IIC_SSE_CVT_PD_RR>, VEX, VEX_L, Sched<[WriteCvtF2F]>;
2212 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
2213 "cvtpd2ps{y}\t{$src, $dst|$dst, $src}",
2215 (int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)))],
2216 IIC_SSE_CVT_PD_RM>, VEX, VEX_L, Sched<[WriteCvtF2FLd]>;
2217 def : InstAlias<"vcvtpd2ps\t{$src, $dst|$dst, $src}",
2218 (VCVTPD2PSYrr VR128:$dst, VR256:$src)>;
2220 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2221 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2222 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))],
2223 IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtF2F]>;
2224 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2225 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2227 (int_x86_sse2_cvtpd2ps (memopv2f64 addr:$src)))],
2228 IIC_SSE_CVT_PD_RM>, Sched<[WriteCvtF2FLd]>;
2231 // AVX 256-bit register conversion intrinsics
2232 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
2233 // whenever possible to avoid declaring two versions of each one.
2234 let Predicates = [HasAVX] in {
2235 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
2236 (VCVTDQ2PSYrr VR256:$src)>;
2237 def : Pat<(int_x86_avx_cvtdq2_ps_256 (bitconvert (memopv4i64 addr:$src))),
2238 (VCVTDQ2PSYrm addr:$src)>;
2240 // Match fround and fextend for 128/256-bit conversions
2241 def : Pat<(v4f32 (X86vfpround (v2f64 VR128:$src))),
2242 (VCVTPD2PSrr VR128:$src)>;
2243 def : Pat<(v4f32 (X86vfpround (memopv2f64 addr:$src))),
2244 (VCVTPD2PSXrm addr:$src)>;
2245 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
2246 (VCVTPD2PSYrr VR256:$src)>;
2247 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
2248 (VCVTPD2PSYrm addr:$src)>;
2250 def : Pat<(v2f64 (X86vfpext (v4f32 VR128:$src))),
2251 (VCVTPS2PDrr VR128:$src)>;
2252 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
2253 (VCVTPS2PDYrr VR128:$src)>;
2254 def : Pat<(v4f64 (extloadv4f32 addr:$src)),
2255 (VCVTPS2PDYrm addr:$src)>;
2258 let Predicates = [UseSSE2] in {
2259 // Match fround and fextend for 128 conversions
2260 def : Pat<(v4f32 (X86vfpround (v2f64 VR128:$src))),
2261 (CVTPD2PSrr VR128:$src)>;
2262 def : Pat<(v4f32 (X86vfpround (memopv2f64 addr:$src))),
2263 (CVTPD2PSrm addr:$src)>;
2265 def : Pat<(v2f64 (X86vfpext (v4f32 VR128:$src))),
2266 (CVTPS2PDrr VR128:$src)>;
2269 //===----------------------------------------------------------------------===//
2270 // SSE 1 & 2 - Compare Instructions
2271 //===----------------------------------------------------------------------===//
2273 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
2274 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
2275 Operand CC, SDNode OpNode, ValueType VT,
2276 PatFrag ld_frag, string asm, string asm_alt,
2278 def rr : SIi8<0xC2, MRMSrcReg,
2279 (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
2280 [(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, imm:$cc))],
2281 itins.rr>, Sched<[itins.Sched]>;
2282 def rm : SIi8<0xC2, MRMSrcMem,
2283 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
2284 [(set RC:$dst, (OpNode (VT RC:$src1),
2285 (ld_frag addr:$src2), imm:$cc))],
2287 Sched<[itins.Sched.Folded, ReadAfterLd]>;
2289 // Accept explicit immediate argument form instead of comparison code.
2290 let neverHasSideEffects = 1 in {
2291 def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst),
2292 (ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, [],
2293 IIC_SSE_ALU_F32S_RR>, Sched<[itins.Sched]>;
2295 def rm_alt : SIi8<0xC2, MRMSrcMem, (outs RC:$dst),
2296 (ins RC:$src1, x86memop:$src2, i8imm:$cc), asm_alt, [],
2297 IIC_SSE_ALU_F32S_RM>,
2298 Sched<[itins.Sched.Folded, ReadAfterLd]>;
2302 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem, AVXCC, X86cmpss, f32, loadf32,
2303 "cmp${cc}ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2304 "cmpss\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2306 XS, VEX_4V, VEX_LIG;
2307 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem, AVXCC, X86cmpsd, f64, loadf64,
2308 "cmp${cc}sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2309 "cmpsd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2310 SSE_ALU_F32S>, // same latency as 32 bit compare
2311 XD, VEX_4V, VEX_LIG;
2313 let Constraints = "$src1 = $dst" in {
2314 defm CMPSS : sse12_cmp_scalar<FR32, f32mem, SSECC, X86cmpss, f32, loadf32,
2315 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
2316 "cmpss\t{$cc, $src2, $dst|$dst, $src2, $cc}", SSE_ALU_F32S>,
2318 defm CMPSD : sse12_cmp_scalar<FR64, f64mem, SSECC, X86cmpsd, f64, loadf64,
2319 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
2320 "cmpsd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2325 multiclass sse12_cmp_scalar_int<X86MemOperand x86memop, Operand CC,
2326 Intrinsic Int, string asm, OpndItins itins> {
2327 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
2328 (ins VR128:$src1, VR128:$src, CC:$cc), asm,
2329 [(set VR128:$dst, (Int VR128:$src1,
2330 VR128:$src, imm:$cc))],
2332 Sched<[itins.Sched]>;
2333 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
2334 (ins VR128:$src1, x86memop:$src, CC:$cc), asm,
2335 [(set VR128:$dst, (Int VR128:$src1,
2336 (load addr:$src), imm:$cc))],
2338 Sched<[itins.Sched.Folded, ReadAfterLd]>;
2341 // Aliases to match intrinsics which expect XMM operand(s).
2342 defm Int_VCMPSS : sse12_cmp_scalar_int<f32mem, AVXCC, int_x86_sse_cmp_ss,
2343 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
2346 defm Int_VCMPSD : sse12_cmp_scalar_int<f64mem, AVXCC, int_x86_sse2_cmp_sd,
2347 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
2348 SSE_ALU_F32S>, // same latency as f32
2350 let Constraints = "$src1 = $dst" in {
2351 defm Int_CMPSS : sse12_cmp_scalar_int<f32mem, SSECC, int_x86_sse_cmp_ss,
2352 "cmp${cc}ss\t{$src, $dst|$dst, $src}",
2354 defm Int_CMPSD : sse12_cmp_scalar_int<f64mem, SSECC, int_x86_sse2_cmp_sd,
2355 "cmp${cc}sd\t{$src, $dst|$dst, $src}",
2361 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
2362 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
2363 ValueType vt, X86MemOperand x86memop,
2364 PatFrag ld_frag, string OpcodeStr> {
2365 def rr: SI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
2366 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2367 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))],
2370 def rm: SI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
2371 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2372 [(set EFLAGS, (OpNode (vt RC:$src1),
2373 (ld_frag addr:$src2)))],
2375 Sched<[WriteFAddLd, ReadAfterLd]>;
2378 let Defs = [EFLAGS] in {
2379 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2380 "ucomiss">, TB, VEX, VEX_LIG;
2381 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2382 "ucomisd">, TB, OpSize, VEX, VEX_LIG;
2383 let Pattern = []<dag> in {
2384 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2385 "comiss">, TB, VEX, VEX_LIG;
2386 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2387 "comisd">, TB, OpSize, VEX, VEX_LIG;
2390 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2391 load, "ucomiss">, TB, VEX;
2392 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2393 load, "ucomisd">, TB, OpSize, VEX;
2395 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
2396 load, "comiss">, TB, VEX;
2397 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
2398 load, "comisd">, TB, OpSize, VEX;
2399 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2401 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2402 "ucomisd">, TB, OpSize;
2404 let Pattern = []<dag> in {
2405 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2407 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2408 "comisd">, TB, OpSize;
2411 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2412 load, "ucomiss">, TB;
2413 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2414 load, "ucomisd">, TB, OpSize;
2416 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
2418 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
2419 "comisd">, TB, OpSize;
2420 } // Defs = [EFLAGS]
2422 // sse12_cmp_packed - sse 1 & 2 compare packed instructions
2423 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
2424 Operand CC, Intrinsic Int, string asm,
2425 string asm_alt, Domain d,
2426 OpndItins itins = SSE_ALU_F32P> {
2427 def rri : PIi8<0xC2, MRMSrcReg,
2428 (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
2429 [(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))],
2432 def rmi : PIi8<0xC2, MRMSrcMem,
2433 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
2434 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))],
2436 Sched<[WriteFAddLd, ReadAfterLd]>;
2438 // Accept explicit immediate argument form instead of comparison code.
2439 let neverHasSideEffects = 1 in {
2440 def rri_alt : PIi8<0xC2, MRMSrcReg,
2441 (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
2442 asm_alt, [], itins.rr, d>, Sched<[WriteFAdd]>;
2443 def rmi_alt : PIi8<0xC2, MRMSrcMem,
2444 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc),
2445 asm_alt, [], itins.rm, d>,
2446 Sched<[WriteFAddLd, ReadAfterLd]>;
2450 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse_cmp_ps,
2451 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2452 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2453 SSEPackedSingle>, TB, VEX_4V;
2454 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse2_cmp_pd,
2455 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2456 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2457 SSEPackedDouble>, TB, OpSize, VEX_4V;
2458 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_ps_256,
2459 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2460 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2461 SSEPackedSingle>, TB, VEX_4V, VEX_L;
2462 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_pd_256,
2463 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2464 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2465 SSEPackedDouble>, TB, OpSize, VEX_4V, VEX_L;
2466 let Constraints = "$src1 = $dst" in {
2467 defm CMPPS : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse_cmp_ps,
2468 "cmp${cc}ps\t{$src2, $dst|$dst, $src2}",
2469 "cmpps\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2470 SSEPackedSingle, SSE_ALU_F32P>, TB;
2471 defm CMPPD : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse2_cmp_pd,
2472 "cmp${cc}pd\t{$src2, $dst|$dst, $src2}",
2473 "cmppd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2474 SSEPackedDouble, SSE_ALU_F64P>, TB, OpSize;
2477 let Predicates = [HasAVX] in {
2478 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2479 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2480 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2481 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2482 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2483 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2484 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2485 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2487 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
2488 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
2489 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
2490 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
2491 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
2492 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
2493 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
2494 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
2497 let Predicates = [UseSSE1] in {
2498 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2499 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2500 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2501 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2504 let Predicates = [UseSSE2] in {
2505 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2506 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2507 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2508 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2511 //===----------------------------------------------------------------------===//
2512 // SSE 1 & 2 - Shuffle Instructions
2513 //===----------------------------------------------------------------------===//
2515 /// sse12_shuffle - sse 1 & 2 shuffle instructions
2516 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
2517 ValueType vt, string asm, PatFrag mem_frag,
2518 Domain d, bit IsConvertibleToThreeAddress = 0> {
2519 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
2520 (ins RC:$src1, x86memop:$src2, i8imm:$src3), asm,
2521 [(set RC:$dst, (vt (X86Shufp RC:$src1, (mem_frag addr:$src2),
2522 (i8 imm:$src3))))], IIC_SSE_SHUFP, d>,
2523 Sched<[WriteShuffleLd, ReadAfterLd]>;
2524 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
2525 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
2526 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
2527 [(set RC:$dst, (vt (X86Shufp RC:$src1, RC:$src2,
2528 (i8 imm:$src3))))], IIC_SSE_SHUFP, d>,
2529 Sched<[WriteShuffle]>;
2532 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2533 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2534 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
2535 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
2536 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2537 memopv8f32, SSEPackedSingle>, TB, VEX_4V, VEX_L;
2538 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2539 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2540 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2541 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
2542 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2543 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V, VEX_L;
2545 let Constraints = "$src1 = $dst" in {
2546 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2547 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2548 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
2550 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2551 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2552 memopv2f64, SSEPackedDouble, 1 /* cvt to pshufd */>,
2556 let Predicates = [HasAVX] in {
2557 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2558 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2559 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2560 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2561 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2563 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2564 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2565 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2566 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2567 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2570 def : Pat<(v8i32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2571 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2572 def : Pat<(v8i32 (X86Shufp VR256:$src1,
2573 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
2574 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2576 def : Pat<(v4i64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2577 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2578 def : Pat<(v4i64 (X86Shufp VR256:$src1,
2579 (memopv4i64 addr:$src2), (i8 imm:$imm))),
2580 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2583 let Predicates = [UseSSE1] in {
2584 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2585 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2586 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2587 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2588 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2591 let Predicates = [UseSSE2] in {
2592 // Generic SHUFPD patterns
2593 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2594 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2595 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2596 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2597 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2600 //===----------------------------------------------------------------------===//
2601 // SSE 1 & 2 - Unpack Instructions
2602 //===----------------------------------------------------------------------===//
2604 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
2605 multiclass sse12_unpack_interleave<bits<8> opc, SDNode OpNode, ValueType vt,
2606 PatFrag mem_frag, RegisterClass RC,
2607 X86MemOperand x86memop, string asm,
2609 def rr : PI<opc, MRMSrcReg,
2610 (outs RC:$dst), (ins RC:$src1, RC:$src2),
2612 (vt (OpNode RC:$src1, RC:$src2)))],
2613 IIC_SSE_UNPCK, d>, Sched<[WriteShuffle]>;
2614 def rm : PI<opc, MRMSrcMem,
2615 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
2617 (vt (OpNode RC:$src1,
2618 (mem_frag addr:$src2))))],
2620 Sched<[WriteShuffleLd, ReadAfterLd]>;
2623 defm VUNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
2624 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2625 SSEPackedSingle>, TB, VEX_4V;
2626 defm VUNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
2627 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2628 SSEPackedDouble>, TB, OpSize, VEX_4V;
2629 defm VUNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
2630 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2631 SSEPackedSingle>, TB, VEX_4V;
2632 defm VUNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
2633 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2634 SSEPackedDouble>, TB, OpSize, VEX_4V;
2636 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, X86Unpckh, v8f32, memopv8f32,
2637 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2638 SSEPackedSingle>, TB, VEX_4V, VEX_L;
2639 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, X86Unpckh, v4f64, memopv4f64,
2640 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2641 SSEPackedDouble>, TB, OpSize, VEX_4V, VEX_L;
2642 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, X86Unpckl, v8f32, memopv8f32,
2643 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2644 SSEPackedSingle>, TB, VEX_4V, VEX_L;
2645 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, X86Unpckl, v4f64, memopv4f64,
2646 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2647 SSEPackedDouble>, TB, OpSize, VEX_4V, VEX_L;
2649 let Constraints = "$src1 = $dst" in {
2650 defm UNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
2651 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
2652 SSEPackedSingle>, TB;
2653 defm UNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
2654 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
2655 SSEPackedDouble>, TB, OpSize;
2656 defm UNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
2657 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
2658 SSEPackedSingle>, TB;
2659 defm UNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
2660 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
2661 SSEPackedDouble>, TB, OpSize;
2662 } // Constraints = "$src1 = $dst"
2664 let Predicates = [HasAVX1Only] in {
2665 def : Pat<(v8i32 (X86Unpckl VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
2666 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
2667 def : Pat<(v8i32 (X86Unpckl VR256:$src1, VR256:$src2)),
2668 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
2669 def : Pat<(v8i32 (X86Unpckh VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
2670 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
2671 def : Pat<(v8i32 (X86Unpckh VR256:$src1, VR256:$src2)),
2672 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
2674 def : Pat<(v4i64 (X86Unpckl VR256:$src1, (memopv4i64 addr:$src2))),
2675 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
2676 def : Pat<(v4i64 (X86Unpckl VR256:$src1, VR256:$src2)),
2677 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
2678 def : Pat<(v4i64 (X86Unpckh VR256:$src1, (memopv4i64 addr:$src2))),
2679 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
2680 def : Pat<(v4i64 (X86Unpckh VR256:$src1, VR256:$src2)),
2681 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
2684 let Predicates = [HasAVX] in {
2685 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2686 // problem is during lowering, where it's not possible to recognize the load
2687 // fold cause it has two uses through a bitcast. One use disappears at isel
2688 // time and the fold opportunity reappears.
2689 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2690 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2693 let Predicates = [UseSSE2] in {
2694 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2695 // problem is during lowering, where it's not possible to recognize the load
2696 // fold cause it has two uses through a bitcast. One use disappears at isel
2697 // time and the fold opportunity reappears.
2698 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2699 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2702 //===----------------------------------------------------------------------===//
2703 // SSE 1 & 2 - Extract Floating-Point Sign mask
2704 //===----------------------------------------------------------------------===//
2706 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2707 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2709 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
2710 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2711 [(set GR32:$dst, (Int RC:$src))], IIC_SSE_MOVMSK, d>,
2712 Sched<[WriteVecLogic]>;
2713 let isAsmParserOnly = 1, hasSideEffects = 0 in
2714 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
2715 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [],
2716 IIC_SSE_MOVMSK, d>, REX_W, Sched<[WriteVecLogic]>;
2719 let Predicates = [HasAVX] in {
2720 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2721 "movmskps", SSEPackedSingle>, TB, VEX;
2722 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2723 "movmskpd", SSEPackedDouble>, TB,
2725 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2726 "movmskps", SSEPackedSingle>, TB,
2728 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2729 "movmskpd", SSEPackedDouble>, TB,
2732 def : Pat<(i32 (X86fgetsign FR32:$src)),
2733 (VMOVMSKPSrr32 (COPY_TO_REGCLASS FR32:$src, VR128))>;
2734 def : Pat<(i64 (X86fgetsign FR32:$src)),
2735 (SUBREG_TO_REG (i64 0),
2736 (VMOVMSKPSrr32 (COPY_TO_REGCLASS FR32:$src, VR128)), sub_32bit)>;
2737 def : Pat<(i32 (X86fgetsign FR64:$src)),
2738 (VMOVMSKPDrr32 (COPY_TO_REGCLASS FR64:$src, VR128))>;
2739 def : Pat<(i64 (X86fgetsign FR64:$src)),
2740 (SUBREG_TO_REG (i64 0),
2741 (VMOVMSKPDrr32 (COPY_TO_REGCLASS FR64:$src, VR128)), sub_32bit)>;
2744 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2745 SSEPackedSingle>, TB;
2746 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2747 SSEPackedDouble>, TB, OpSize;
2749 def : Pat<(i32 (X86fgetsign FR32:$src)),
2750 (MOVMSKPSrr32 (COPY_TO_REGCLASS FR32:$src, VR128))>,
2751 Requires<[UseSSE1]>;
2752 def : Pat<(i64 (X86fgetsign FR32:$src)),
2753 (SUBREG_TO_REG (i64 0),
2754 (MOVMSKPSrr32 (COPY_TO_REGCLASS FR32:$src, VR128)), sub_32bit)>,
2755 Requires<[UseSSE1]>;
2756 def : Pat<(i32 (X86fgetsign FR64:$src)),
2757 (MOVMSKPDrr32 (COPY_TO_REGCLASS FR64:$src, VR128))>,
2758 Requires<[UseSSE2]>;
2759 def : Pat<(i64 (X86fgetsign FR64:$src)),
2760 (SUBREG_TO_REG (i64 0),
2761 (MOVMSKPDrr32 (COPY_TO_REGCLASS FR64:$src, VR128)), sub_32bit)>,
2762 Requires<[UseSSE2]>;
2764 //===---------------------------------------------------------------------===//
2765 // SSE2 - Packed Integer Logical Instructions
2766 //===---------------------------------------------------------------------===//
2768 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2770 /// PDI_binop_rm - Simple SSE2 binary operator.
2771 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2772 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
2773 X86MemOperand x86memop, OpndItins itins,
2774 bit IsCommutable, bit Is2Addr> {
2775 let isCommutable = IsCommutable in
2776 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
2777 (ins RC:$src1, RC:$src2),
2779 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2780 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2781 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>,
2782 Sched<[itins.Sched]>;
2783 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
2784 (ins RC:$src1, x86memop:$src2),
2786 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2787 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2788 [(set RC:$dst, (OpVT (OpNode RC:$src1,
2789 (bitconvert (memop_frag addr:$src2)))))],
2791 Sched<[itins.Sched.Folded, ReadAfterLd]>;
2793 } // ExeDomain = SSEPackedInt
2795 multiclass PDI_binop_all<bits<8> opc, string OpcodeStr, SDNode Opcode,
2796 ValueType OpVT128, ValueType OpVT256,
2797 OpndItins itins, bit IsCommutable = 0> {
2798 let Predicates = [HasAVX] in
2799 defm V#NAME : PDI_binop_rm<opc, !strconcat("v", OpcodeStr), Opcode, OpVT128,
2800 VR128, memopv2i64, i128mem, itins, IsCommutable, 0>, VEX_4V;
2802 let Constraints = "$src1 = $dst" in
2803 defm NAME : PDI_binop_rm<opc, OpcodeStr, Opcode, OpVT128, VR128,
2804 memopv2i64, i128mem, itins, IsCommutable, 1>;
2806 let Predicates = [HasAVX2] in
2807 defm V#NAME#Y : PDI_binop_rm<opc, !strconcat("v", OpcodeStr), Opcode,
2808 OpVT256, VR256, memopv4i64, i256mem, itins,
2809 IsCommutable, 0>, VEX_4V, VEX_L;
2812 // These are ordered here for pattern ordering requirements with the fp versions
2814 defm PAND : PDI_binop_all<0xDB, "pand", and, v2i64, v4i64, SSE_BIT_ITINS_P, 1>;
2815 defm POR : PDI_binop_all<0xEB, "por", or, v2i64, v4i64, SSE_BIT_ITINS_P, 1>;
2816 defm PXOR : PDI_binop_all<0xEF, "pxor", xor, v2i64, v4i64, SSE_BIT_ITINS_P, 1>;
2817 defm PANDN : PDI_binop_all<0xDF, "pandn", X86andnp, v2i64, v4i64,
2818 SSE_BIT_ITINS_P, 0>;
2820 //===----------------------------------------------------------------------===//
2821 // SSE 1 & 2 - Logical Instructions
2822 //===----------------------------------------------------------------------===//
2824 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2826 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2827 SDNode OpNode, OpndItins itins> {
2828 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2829 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, itins, 0>,
2832 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2833 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, itins, 0>,
2836 let Constraints = "$src1 = $dst" in {
2837 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2838 f32, f128mem, memopfsf32, SSEPackedSingle, itins>,
2841 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2842 f64, f128mem, memopfsf64, SSEPackedDouble, itins>,
2847 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2848 let isCodeGenOnly = 1 in {
2849 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand,
2851 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for,
2853 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor,
2856 let isCommutable = 0 in
2857 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", X86fandn,
2861 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2863 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2865 defm V#NAME#PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2866 !strconcat(OpcodeStr, "ps"), f256mem,
2867 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2868 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2869 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V, VEX_L;
2871 defm V#NAME#PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2872 !strconcat(OpcodeStr, "pd"), f256mem,
2873 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2874 (bc_v4i64 (v4f64 VR256:$src2))))],
2875 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2876 (memopv4i64 addr:$src2)))], 0>,
2877 TB, OpSize, VEX_4V, VEX_L;
2879 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2880 // are all promoted to v2i64, and the patterns are covered by the int
2881 // version. This is needed in SSE only, because v2i64 isn't supported on
2882 // SSE1, but only on SSE2.
2883 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2884 !strconcat(OpcodeStr, "ps"), f128mem, [],
2885 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2886 (memopv2i64 addr:$src2)))], 0>, TB, VEX_4V;
2888 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2889 !strconcat(OpcodeStr, "pd"), f128mem,
2890 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2891 (bc_v2i64 (v2f64 VR128:$src2))))],
2892 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2893 (memopv2i64 addr:$src2)))], 0>,
2896 let Constraints = "$src1 = $dst" in {
2897 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2898 !strconcat(OpcodeStr, "ps"), f128mem,
2899 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2900 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2901 (memopv2i64 addr:$src2)))]>, TB;
2903 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2904 !strconcat(OpcodeStr, "pd"), f128mem,
2905 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2906 (bc_v2i64 (v2f64 VR128:$src2))))],
2907 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2908 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2912 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2913 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2914 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2915 let isCommutable = 0 in
2916 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2918 //===----------------------------------------------------------------------===//
2919 // SSE 1 & 2 - Arithmetic Instructions
2920 //===----------------------------------------------------------------------===//
2922 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2925 /// In addition, we also have a special variant of the scalar form here to
2926 /// represent the associated intrinsic operation. This form is unlike the
2927 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2928 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2930 /// These three forms can each be reg+reg or reg+mem.
2933 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2935 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr,
2936 SDNode OpNode, SizeItins itins> {
2937 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2938 VR128, v4f32, f128mem, memopv4f32,
2939 SSEPackedSingle, itins.s, 0>, TB, VEX_4V;
2940 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2941 VR128, v2f64, f128mem, memopv2f64,
2942 SSEPackedDouble, itins.d, 0>, TB, OpSize, VEX_4V;
2944 defm V#NAME#PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"),
2945 OpNode, VR256, v8f32, f256mem, memopv8f32,
2946 SSEPackedSingle, itins.s, 0>, TB, VEX_4V, VEX_L;
2947 defm V#NAME#PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"),
2948 OpNode, VR256, v4f64, f256mem, memopv4f64,
2949 SSEPackedDouble, itins.d, 0>, TB, OpSize, VEX_4V, VEX_L;
2951 let Constraints = "$src1 = $dst" in {
2952 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2953 v4f32, f128mem, memopv4f32, SSEPackedSingle,
2955 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2956 v2f64, f128mem, memopv2f64, SSEPackedDouble,
2957 itins.d>, TB, OpSize;
2961 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2963 defm V#NAME#SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2964 OpNode, FR32, f32mem, itins.s, 0>, XS, VEX_4V, VEX_LIG;
2965 defm V#NAME#SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2966 OpNode, FR64, f64mem, itins.d, 0>, XD, VEX_4V, VEX_LIG;
2968 let Constraints = "$src1 = $dst" in {
2969 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2970 OpNode, FR32, f32mem, itins.s>, XS;
2971 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2972 OpNode, FR64, f64mem, itins.d>, XD;
2976 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2978 defm V#NAME#SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2979 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32,
2980 itins.s, 0>, XS, VEX_4V, VEX_LIG;
2981 defm V#NAME#SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2982 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64,
2983 itins.d, 0>, XD, VEX_4V, VEX_LIG;
2985 let Constraints = "$src1 = $dst" in {
2986 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2987 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32,
2989 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2990 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64,
2995 // Binary Arithmetic instructions
2996 defm ADD : basic_sse12_fp_binop_p<0x58, "add", fadd, SSE_ALU_ITINS_P>,
2997 basic_sse12_fp_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S>,
2998 basic_sse12_fp_binop_s_int<0x58, "add", SSE_ALU_ITINS_S>;
2999 defm MUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, SSE_MUL_ITINS_P>,
3000 basic_sse12_fp_binop_s<0x59, "mul", fmul, SSE_MUL_ITINS_S>,
3001 basic_sse12_fp_binop_s_int<0x59, "mul", SSE_MUL_ITINS_S>;
3002 let isCommutable = 0 in {
3003 defm SUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, SSE_ALU_ITINS_P>,
3004 basic_sse12_fp_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S>,
3005 basic_sse12_fp_binop_s_int<0x5C, "sub", SSE_ALU_ITINS_S>;
3006 defm DIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, SSE_DIV_ITINS_P>,
3007 basic_sse12_fp_binop_s<0x5E, "div", fdiv, SSE_DIV_ITINS_S>,
3008 basic_sse12_fp_binop_s_int<0x5E, "div", SSE_DIV_ITINS_S>;
3009 defm MAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, SSE_ALU_ITINS_P>,
3010 basic_sse12_fp_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S>,
3011 basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S>;
3012 defm MIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, SSE_ALU_ITINS_P>,
3013 basic_sse12_fp_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S>,
3014 basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S>;
3017 let isCodeGenOnly = 1 in {
3018 defm MAXC: basic_sse12_fp_binop_p<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_P>,
3019 basic_sse12_fp_binop_s<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_S>;
3020 defm MINC: basic_sse12_fp_binop_p<0x5D, "min", X86fminc, SSE_ALU_ITINS_P>,
3021 basic_sse12_fp_binop_s<0x5D, "min", X86fminc, SSE_ALU_ITINS_S>;
3025 /// In addition, we also have a special variant of the scalar form here to
3026 /// represent the associated intrinsic operation. This form is unlike the
3027 /// plain scalar form, in that it takes an entire vector (instead of a
3028 /// scalar) and leaves the top elements undefined.
3030 /// And, we have a special variant form for a full-vector intrinsic form.
3032 let Sched = WriteFSqrt in {
3033 def SSE_SQRTPS : OpndItins<
3034 IIC_SSE_SQRTPS_RR, IIC_SSE_SQRTPS_RM
3037 def SSE_SQRTSS : OpndItins<
3038 IIC_SSE_SQRTSS_RR, IIC_SSE_SQRTSS_RM
3041 def SSE_SQRTPD : OpndItins<
3042 IIC_SSE_SQRTPD_RR, IIC_SSE_SQRTPD_RM
3045 def SSE_SQRTSD : OpndItins<
3046 IIC_SSE_SQRTSD_RR, IIC_SSE_SQRTSD_RM
3050 let Sched = WriteFRcp in {
3051 def SSE_RCPP : OpndItins<
3052 IIC_SSE_RCPP_RR, IIC_SSE_RCPP_RM
3055 def SSE_RCPS : OpndItins<
3056 IIC_SSE_RCPS_RR, IIC_SSE_RCPS_RM
3060 /// sse1_fp_unop_s - SSE1 unops in scalar form.
3061 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
3062 SDNode OpNode, Intrinsic F32Int, OpndItins itins> {
3063 let Predicates = [HasAVX], hasSideEffects = 0 in {
3064 def V#NAME#SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst),
3065 (ins FR32:$src1, FR32:$src2),
3066 !strconcat("v", OpcodeStr,
3067 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3068 []>, VEX_4V, VEX_LIG, Sched<[itins.Sched]>;
3069 let mayLoad = 1 in {
3070 def V#NAME#SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst),
3071 (ins FR32:$src1,f32mem:$src2),
3072 !strconcat("v", OpcodeStr,
3073 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3074 []>, VEX_4V, VEX_LIG,
3075 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3076 def V#NAME#SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3077 (ins VR128:$src1, ssmem:$src2),
3078 !strconcat("v", OpcodeStr,
3079 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3080 []>, VEX_4V, VEX_LIG,
3081 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3085 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
3086 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3087 [(set FR32:$dst, (OpNode FR32:$src))]>, Sched<[itins.Sched]>;
3088 // For scalar unary operations, fold a load into the operation
3089 // only in OptForSize mode. It eliminates an instruction, but it also
3090 // eliminates a whole-register clobber (the load), so it introduces a
3091 // partial register update condition.
3092 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
3093 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3094 [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
3095 Requires<[UseSSE1, OptForSize]>, Sched<[itins.Sched.Folded]>;
3096 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3097 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3098 [(set VR128:$dst, (F32Int VR128:$src))], itins.rr>,
3099 Sched<[itins.Sched]>;
3100 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
3101 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3102 [(set VR128:$dst, (F32Int sse_load_f32:$src))], itins.rm>,
3103 Sched<[itins.Sched.Folded]>;
3106 /// sse1_fp_unop_s_rw - SSE1 unops where vector form has a read-write operand.
3107 multiclass sse1_fp_unop_rw<bits<8> opc, string OpcodeStr, SDNode OpNode,
3109 let Predicates = [HasAVX], hasSideEffects = 0 in {
3110 def V#NAME#SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst),
3111 (ins FR32:$src1, FR32:$src2),
3112 !strconcat("v", OpcodeStr,
3113 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3114 []>, VEX_4V, VEX_LIG, Sched<[itins.Sched]>;
3115 let mayLoad = 1 in {
3116 def V#NAME#SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst),
3117 (ins FR32:$src1,f32mem:$src2),
3118 !strconcat("v", OpcodeStr,
3119 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3120 []>, VEX_4V, VEX_LIG,
3121 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3122 def V#NAME#SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3123 (ins VR128:$src1, ssmem:$src2),
3124 !strconcat("v", OpcodeStr,
3125 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3126 []>, VEX_4V, VEX_LIG,
3127 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3131 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
3132 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3133 [(set FR32:$dst, (OpNode FR32:$src))]>, Sched<[itins.Sched]>;
3134 // For scalar unary operations, fold a load into the operation
3135 // only in OptForSize mode. It eliminates an instruction, but it also
3136 // eliminates a whole-register clobber (the load), so it introduces a
3137 // partial register update condition.
3138 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
3139 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3140 [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
3141 Requires<[UseSSE1, OptForSize]>, Sched<[itins.Sched.Folded]>;
3142 let Constraints = "$src1 = $dst" in {
3143 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst),
3144 (ins VR128:$src1, VR128:$src2),
3145 !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
3146 [], itins.rr>, Sched<[itins.Sched]>;
3147 let mayLoad = 1, hasSideEffects = 0 in
3148 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3149 (ins VR128:$src1, ssmem:$src2),
3150 !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
3151 [], itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
3155 /// sse1_fp_unop_p - SSE1 unops in packed form.
3156 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
3158 let Predicates = [HasAVX] in {
3159 def V#NAME#PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3160 !strconcat("v", OpcodeStr,
3161 "ps\t{$src, $dst|$dst, $src}"),
3162 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))],
3163 itins.rr>, VEX, Sched<[itins.Sched]>;
3164 def V#NAME#PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3165 !strconcat("v", OpcodeStr,
3166 "ps\t{$src, $dst|$dst, $src}"),
3167 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))],
3168 itins.rm>, VEX, Sched<[itins.Sched.Folded]>;
3169 def V#NAME#PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3170 !strconcat("v", OpcodeStr,
3171 "ps\t{$src, $dst|$dst, $src}"),
3172 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))],
3173 itins.rr>, VEX, VEX_L, Sched<[itins.Sched]>;
3174 def V#NAME#PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3175 !strconcat("v", OpcodeStr,
3176 "ps\t{$src, $dst|$dst, $src}"),
3177 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))],
3178 itins.rm>, VEX, VEX_L, Sched<[itins.Sched.Folded]>;
3181 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3182 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3183 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))], itins.rr>,
3184 Sched<[itins.Sched]>;
3185 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3186 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3187 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))], itins.rm>,
3188 Sched<[itins.Sched.Folded]>;
3191 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
3192 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3193 Intrinsic V4F32Int, Intrinsic V8F32Int,
3195 let Predicates = [HasAVX] in {
3196 def V#NAME#PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3197 !strconcat("v", OpcodeStr,
3198 "ps\t{$src, $dst|$dst, $src}"),
3199 [(set VR128:$dst, (V4F32Int VR128:$src))],
3200 itins.rr>, VEX, Sched<[itins.Sched]>;
3201 def V#NAME#PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3202 !strconcat("v", OpcodeStr,
3203 "ps\t{$src, $dst|$dst, $src}"),
3204 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))],
3205 itins.rm>, VEX, Sched<[itins.Sched.Folded]>;
3206 def V#NAME#PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3207 !strconcat("v", OpcodeStr,
3208 "ps\t{$src, $dst|$dst, $src}"),
3209 [(set VR256:$dst, (V8F32Int VR256:$src))],
3210 itins.rr>, VEX, VEX_L, Sched<[itins.Sched]>;
3211 def V#NAME#PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst),
3213 !strconcat("v", OpcodeStr,
3214 "ps\t{$src, $dst|$dst, $src}"),
3215 [(set VR256:$dst, (V8F32Int (memopv8f32 addr:$src)))],
3216 itins.rm>, VEX, VEX_L, Sched<[itins.Sched.Folded]>;
3219 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3220 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3221 [(set VR128:$dst, (V4F32Int VR128:$src))],
3222 itins.rr>, Sched<[itins.Sched]>;
3223 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3224 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3225 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))],
3226 itins.rm>, Sched<[itins.Sched.Folded]>;
3229 /// sse2_fp_unop_s - SSE2 unops in scalar form.
3230 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
3231 SDNode OpNode, Intrinsic F64Int, OpndItins itins> {
3232 let Predicates = [HasAVX], hasSideEffects = 0 in {
3233 def V#NAME#SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst),
3234 (ins FR64:$src1, FR64:$src2),
3235 !strconcat("v", OpcodeStr,
3236 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3237 []>, VEX_4V, VEX_LIG, Sched<[itins.Sched]>;
3238 let mayLoad = 1 in {
3239 def V#NAME#SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst),
3240 (ins FR64:$src1,f64mem:$src2),
3241 !strconcat("v", OpcodeStr,
3242 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3243 []>, VEX_4V, VEX_LIG,
3244 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3245 def V#NAME#SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
3246 (ins VR128:$src1, sdmem:$src2),
3247 !strconcat("v", OpcodeStr,
3248 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3249 []>, VEX_4V, VEX_LIG,
3250 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3254 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
3255 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3256 [(set FR64:$dst, (OpNode FR64:$src))], itins.rr>,
3257 Sched<[itins.Sched]>;
3258 // See the comments in sse1_fp_unop_s for why this is OptForSize.
3259 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
3260 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3261 [(set FR64:$dst, (OpNode (load addr:$src)))], itins.rm>, XD,
3262 Requires<[UseSSE2, OptForSize]>, Sched<[itins.Sched.Folded]>;
3263 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3264 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3265 [(set VR128:$dst, (F64Int VR128:$src))], itins.rr>,
3266 Sched<[itins.Sched]>;
3267 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
3268 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3269 [(set VR128:$dst, (F64Int sse_load_f64:$src))], itins.rm>,
3270 Sched<[itins.Sched.Folded]>;
3273 /// sse2_fp_unop_p - SSE2 unops in vector forms.
3274 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
3275 SDNode OpNode, OpndItins itins> {
3276 let Predicates = [HasAVX] in {
3277 def V#NAME#PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3278 !strconcat("v", OpcodeStr,
3279 "pd\t{$src, $dst|$dst, $src}"),
3280 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))],
3281 itins.rr>, VEX, Sched<[itins.Sched]>;
3282 def V#NAME#PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3283 !strconcat("v", OpcodeStr,
3284 "pd\t{$src, $dst|$dst, $src}"),
3285 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))],
3286 itins.rm>, VEX, Sched<[itins.Sched.Folded]>;
3287 def V#NAME#PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3288 !strconcat("v", OpcodeStr,
3289 "pd\t{$src, $dst|$dst, $src}"),
3290 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))],
3291 itins.rr>, VEX, VEX_L, Sched<[itins.Sched]>;
3292 def V#NAME#PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3293 !strconcat("v", OpcodeStr,
3294 "pd\t{$src, $dst|$dst, $src}"),
3295 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))],
3296 itins.rm>, VEX, VEX_L, Sched<[itins.Sched.Folded]>;
3299 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3300 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3301 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))], itins.rr>,
3302 Sched<[itins.Sched]>;
3303 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3304 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3305 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))], itins.rm>,
3306 Sched<[itins.Sched.Folded]>;
3310 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss,
3312 sse1_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTPS>,
3313 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd,
3315 sse2_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTPD>;
3317 // Reciprocal approximations. Note that these typically require refinement
3318 // in order to obtain suitable precision.
3319 defm RSQRT : sse1_fp_unop_rw<0x52, "rsqrt", X86frsqrt, SSE_SQRTSS>,
3320 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, SSE_SQRTPS>,
3321 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps,
3322 int_x86_avx_rsqrt_ps_256, SSE_SQRTPS>;
3323 defm RCP : sse1_fp_unop_rw<0x53, "rcp", X86frcp, SSE_RCPS>,
3324 sse1_fp_unop_p<0x53, "rcp", X86frcp, SSE_RCPP>,
3325 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps,
3326 int_x86_avx_rcp_ps_256, SSE_RCPP>;
3328 let Predicates = [UseAVX] in {
3329 def : Pat<(f32 (fsqrt FR32:$src)),
3330 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3331 def : Pat<(f32 (fsqrt (load addr:$src))),
3332 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3333 Requires<[HasAVX, OptForSize]>;
3334 def : Pat<(f64 (fsqrt FR64:$src)),
3335 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
3336 def : Pat<(f64 (fsqrt (load addr:$src))),
3337 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
3338 Requires<[HasAVX, OptForSize]>;
3340 def : Pat<(f32 (X86frsqrt FR32:$src)),
3341 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3342 def : Pat<(f32 (X86frsqrt (load addr:$src))),
3343 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3344 Requires<[HasAVX, OptForSize]>;
3346 def : Pat<(f32 (X86frcp FR32:$src)),
3347 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3348 def : Pat<(f32 (X86frcp (load addr:$src))),
3349 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3350 Requires<[HasAVX, OptForSize]>;
3352 let Predicates = [UseAVX] in {
3353 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
3354 (COPY_TO_REGCLASS (VSQRTSSr (f32 (IMPLICIT_DEF)),
3355 (COPY_TO_REGCLASS VR128:$src, FR32)),
3357 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
3358 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3360 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
3361 (COPY_TO_REGCLASS (VSQRTSDr (f64 (IMPLICIT_DEF)),
3362 (COPY_TO_REGCLASS VR128:$src, FR64)),
3364 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
3365 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
3368 let Predicates = [HasAVX] in {
3369 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3370 (COPY_TO_REGCLASS (VRSQRTSSr (f32 (IMPLICIT_DEF)),
3371 (COPY_TO_REGCLASS VR128:$src, FR32)),
3373 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
3374 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3376 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3377 (COPY_TO_REGCLASS (VRCPSSr (f32 (IMPLICIT_DEF)),
3378 (COPY_TO_REGCLASS VR128:$src, FR32)),
3380 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
3381 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3384 // Reciprocal approximations. Note that these typically require refinement
3385 // in order to obtain suitable precision.
3386 let Predicates = [UseSSE1] in {
3387 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3388 (RSQRTSSr_Int VR128:$src, VR128:$src)>;
3389 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3390 (RCPSSr_Int VR128:$src, VR128:$src)>;
3393 // There is no f64 version of the reciprocal approximation instructions.
3395 //===----------------------------------------------------------------------===//
3396 // SSE 1 & 2 - Non-temporal stores
3397 //===----------------------------------------------------------------------===//
3399 let AddedComplexity = 400 in { // Prefer non-temporal versions
3400 let SchedRW = [WriteStore] in {
3401 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3402 (ins f128mem:$dst, VR128:$src),
3403 "movntps\t{$src, $dst|$dst, $src}",
3404 [(alignednontemporalstore (v4f32 VR128:$src),
3406 IIC_SSE_MOVNT>, VEX;
3407 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3408 (ins f128mem:$dst, VR128:$src),
3409 "movntpd\t{$src, $dst|$dst, $src}",
3410 [(alignednontemporalstore (v2f64 VR128:$src),
3412 IIC_SSE_MOVNT>, VEX;
3414 let ExeDomain = SSEPackedInt in
3415 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3416 (ins f128mem:$dst, VR128:$src),
3417 "movntdq\t{$src, $dst|$dst, $src}",
3418 [(alignednontemporalstore (v2i64 VR128:$src),
3420 IIC_SSE_MOVNT>, VEX;
3422 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3423 (ins f256mem:$dst, VR256:$src),
3424 "movntps\t{$src, $dst|$dst, $src}",
3425 [(alignednontemporalstore (v8f32 VR256:$src),
3427 IIC_SSE_MOVNT>, VEX, VEX_L;
3428 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3429 (ins f256mem:$dst, VR256:$src),
3430 "movntpd\t{$src, $dst|$dst, $src}",
3431 [(alignednontemporalstore (v4f64 VR256:$src),
3433 IIC_SSE_MOVNT>, VEX, VEX_L;
3434 let ExeDomain = SSEPackedInt in
3435 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3436 (ins f256mem:$dst, VR256:$src),
3437 "movntdq\t{$src, $dst|$dst, $src}",
3438 [(alignednontemporalstore (v4i64 VR256:$src),
3440 IIC_SSE_MOVNT>, VEX, VEX_L;
3442 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3443 "movntps\t{$src, $dst|$dst, $src}",
3444 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)],
3446 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3447 "movntpd\t{$src, $dst|$dst, $src}",
3448 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)],
3451 let ExeDomain = SSEPackedInt in
3452 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3453 "movntdq\t{$src, $dst|$dst, $src}",
3454 [(alignednontemporalstore (v2i64 VR128:$src), addr:$dst)],
3457 // There is no AVX form for instructions below this point
3458 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3459 "movnti{l}\t{$src, $dst|$dst, $src}",
3460 [(nontemporalstore (i32 GR32:$src), addr:$dst)],
3462 TB, Requires<[HasSSE2]>;
3463 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3464 "movnti{q}\t{$src, $dst|$dst, $src}",
3465 [(nontemporalstore (i64 GR64:$src), addr:$dst)],
3467 TB, Requires<[HasSSE2]>;
3468 } // SchedRW = [WriteStore]
3470 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3471 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3473 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3474 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[UseSSE2]>;
3475 } // AddedComplexity
3477 //===----------------------------------------------------------------------===//
3478 // SSE 1 & 2 - Prefetch and memory fence
3479 //===----------------------------------------------------------------------===//
3481 // Prefetch intrinsic.
3482 let Predicates = [HasSSE1], SchedRW = [WriteLoad] in {
3483 def PREFETCHT0 : I<0x18, MRM1m, (outs), (ins i8mem:$src),
3484 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))],
3485 IIC_SSE_PREFETCH>, TB;
3486 def PREFETCHT1 : I<0x18, MRM2m, (outs), (ins i8mem:$src),
3487 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))],
3488 IIC_SSE_PREFETCH>, TB;
3489 def PREFETCHT2 : I<0x18, MRM3m, (outs), (ins i8mem:$src),
3490 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))],
3491 IIC_SSE_PREFETCH>, TB;
3492 def PREFETCHNTA : I<0x18, MRM0m, (outs), (ins i8mem:$src),
3493 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))],
3494 IIC_SSE_PREFETCH>, TB;
3497 // FIXME: How should these memory instructions be modeled?
3498 let SchedRW = [WriteLoad] in {
3500 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3501 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)],
3502 IIC_SSE_PREFETCH>, TB, Requires<[HasSSE2]>;
3504 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3505 // was introduced with SSE2, it's backward compatible.
3506 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", [], IIC_SSE_PAUSE>, REP;
3508 // Load, store, and memory fence
3509 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3510 "sfence", [(int_x86_sse_sfence)], IIC_SSE_SFENCE>,
3511 TB, Requires<[HasSSE1]>;
3512 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3513 "lfence", [(int_x86_sse2_lfence)], IIC_SSE_LFENCE>,
3514 TB, Requires<[HasSSE2]>;
3515 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3516 "mfence", [(int_x86_sse2_mfence)], IIC_SSE_MFENCE>,
3517 TB, Requires<[HasSSE2]>;
3520 def : Pat<(X86SFence), (SFENCE)>;
3521 def : Pat<(X86LFence), (LFENCE)>;
3522 def : Pat<(X86MFence), (MFENCE)>;
3524 //===----------------------------------------------------------------------===//
3525 // SSE 1 & 2 - Load/Store XCSR register
3526 //===----------------------------------------------------------------------===//
3528 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3529 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3530 IIC_SSE_LDMXCSR>, VEX, Sched<[WriteLoad]>;
3531 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3532 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3533 IIC_SSE_STMXCSR>, VEX, Sched<[WriteStore]>;
3535 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3536 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3537 IIC_SSE_LDMXCSR>, Sched<[WriteLoad]>;
3538 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3539 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3540 IIC_SSE_STMXCSR>, Sched<[WriteStore]>;
3542 //===---------------------------------------------------------------------===//
3543 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3544 //===---------------------------------------------------------------------===//
3546 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3548 let neverHasSideEffects = 1, SchedRW = [WriteMove] in {
3549 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3550 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3552 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3553 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3555 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3556 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3558 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3559 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3564 let isCodeGenOnly = 1, hasSideEffects = 0, SchedRW = [WriteMove] in {
3565 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3566 "movdqa\t{$src, $dst|$dst, $src}", [],
3569 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3570 "movdqa\t{$src, $dst|$dst, $src}", [],
3571 IIC_SSE_MOVA_P_RR>, VEX, VEX_L;
3572 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3573 "movdqu\t{$src, $dst|$dst, $src}", [],
3576 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3577 "movdqu\t{$src, $dst|$dst, $src}", [],
3578 IIC_SSE_MOVU_P_RR>, VEX, VEX_L;
3581 let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1,
3582 neverHasSideEffects = 1, SchedRW = [WriteLoad] in {
3583 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3584 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3586 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3587 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3589 let Predicates = [HasAVX] in {
3590 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3591 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3593 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3594 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3599 let mayStore = 1, neverHasSideEffects = 1, SchedRW = [WriteStore] in {
3600 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3601 (ins i128mem:$dst, VR128:$src),
3602 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3604 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3605 (ins i256mem:$dst, VR256:$src),
3606 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3608 let Predicates = [HasAVX] in {
3609 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3610 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3612 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3613 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3618 let SchedRW = [WriteMove] in {
3619 let neverHasSideEffects = 1 in
3620 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3621 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>;
3623 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3624 "movdqu\t{$src, $dst|$dst, $src}",
3625 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>;
3628 let isCodeGenOnly = 1, hasSideEffects = 0 in {
3629 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3630 "movdqa\t{$src, $dst|$dst, $src}", [],
3633 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3634 "movdqu\t{$src, $dst|$dst, $src}",
3635 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>;
3639 let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1,
3640 neverHasSideEffects = 1, SchedRW = [WriteLoad] in {
3641 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3642 "movdqa\t{$src, $dst|$dst, $src}",
3643 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/],
3645 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3646 "movdqu\t{$src, $dst|$dst, $src}",
3647 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/],
3649 XS, Requires<[UseSSE2]>;
3652 let mayStore = 1, neverHasSideEffects = 1, SchedRW = [WriteStore] in {
3653 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3654 "movdqa\t{$src, $dst|$dst, $src}",
3655 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/],
3657 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3658 "movdqu\t{$src, $dst|$dst, $src}",
3659 [/*(store (v2i64 VR128:$src), addr:$dst)*/],
3661 XS, Requires<[UseSSE2]>;
3664 } // ExeDomain = SSEPackedInt
3666 let Predicates = [HasAVX] in {
3667 def : Pat<(int_x86_sse2_storeu_dq addr:$dst, VR128:$src),
3668 (VMOVDQUmr addr:$dst, VR128:$src)>;
3669 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3670 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3672 let Predicates = [UseSSE2] in
3673 def : Pat<(int_x86_sse2_storeu_dq addr:$dst, VR128:$src),
3674 (MOVDQUmr addr:$dst, VR128:$src)>;
3676 //===---------------------------------------------------------------------===//
3677 // SSE2 - Packed Integer Arithmetic Instructions
3678 //===---------------------------------------------------------------------===//
3680 let Sched = WriteVecIMul in
3681 def SSE_PMADD : OpndItins<
3682 IIC_SSE_PMADD, IIC_SSE_PMADD
3685 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3687 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3688 RegisterClass RC, PatFrag memop_frag,
3689 X86MemOperand x86memop,
3691 bit IsCommutable = 0,
3693 let isCommutable = IsCommutable in
3694 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3695 (ins RC:$src1, RC:$src2),
3697 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3698 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3699 [(set RC:$dst, (IntId RC:$src1, RC:$src2))], itins.rr>,
3700 Sched<[itins.Sched]>;
3701 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3702 (ins RC:$src1, x86memop:$src2),
3704 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3705 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3706 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))],
3707 itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
3710 multiclass PDI_binop_all_int<bits<8> opc, string OpcodeStr, Intrinsic IntId128,
3711 Intrinsic IntId256, OpndItins itins,
3712 bit IsCommutable = 0> {
3713 let Predicates = [HasAVX] in
3714 defm V#NAME : PDI_binop_rm_int<opc, !strconcat("v", OpcodeStr), IntId128,
3715 VR128, memopv2i64, i128mem, itins,
3716 IsCommutable, 0>, VEX_4V;
3718 let Constraints = "$src1 = $dst" in
3719 defm NAME : PDI_binop_rm_int<opc, OpcodeStr, IntId128, VR128, memopv2i64,
3720 i128mem, itins, IsCommutable, 1>;
3722 let Predicates = [HasAVX2] in
3723 defm V#NAME#Y : PDI_binop_rm_int<opc, !strconcat("v", OpcodeStr), IntId256,
3724 VR256, memopv4i64, i256mem, itins,
3725 IsCommutable, 0>, VEX_4V, VEX_L;
3728 multiclass PDI_binop_rmi<bits<8> opc, bits<8> opc2, Format ImmForm,
3729 string OpcodeStr, SDNode OpNode,
3730 SDNode OpNode2, RegisterClass RC,
3731 ValueType DstVT, ValueType SrcVT, PatFrag bc_frag,
3732 ShiftOpndItins itins,
3734 // src2 is always 128-bit
3735 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3736 (ins RC:$src1, VR128:$src2),
3738 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3739 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3740 [(set RC:$dst, (DstVT (OpNode RC:$src1, (SrcVT VR128:$src2))))],
3741 itins.rr>, Sched<[WriteVecShift]>;
3742 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3743 (ins RC:$src1, i128mem:$src2),
3745 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3746 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3747 [(set RC:$dst, (DstVT (OpNode RC:$src1,
3748 (bc_frag (memopv2i64 addr:$src2)))))], itins.rm>,
3749 Sched<[WriteVecShiftLd, ReadAfterLd]>;
3750 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3751 (ins RC:$src1, i32i8imm:$src2),
3753 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3754 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3755 [(set RC:$dst, (DstVT (OpNode2 RC:$src1, (i32 imm:$src2))))], itins.ri>,
3756 Sched<[WriteVecShift]>;
3759 /// PDI_binop_rm2 - Simple SSE2 binary operator with different src and dst types
3760 multiclass PDI_binop_rm2<bits<8> opc, string OpcodeStr, SDNode OpNode,
3761 ValueType DstVT, ValueType SrcVT, RegisterClass RC,
3762 PatFrag memop_frag, X86MemOperand x86memop,
3764 bit IsCommutable = 0, bit Is2Addr = 1> {
3765 let isCommutable = IsCommutable in
3766 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3767 (ins RC:$src1, RC:$src2),
3769 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3770 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3771 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1), RC:$src2)))]>,
3772 Sched<[itins.Sched]>;
3773 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3774 (ins RC:$src1, x86memop:$src2),
3776 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3777 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3778 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1),
3779 (bitconvert (memop_frag addr:$src2)))))]>,
3780 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3782 } // ExeDomain = SSEPackedInt
3784 defm PADDB : PDI_binop_all<0xFC, "paddb", add, v16i8, v32i8,
3785 SSE_INTALU_ITINS_P, 1>;
3786 defm PADDW : PDI_binop_all<0xFD, "paddw", add, v8i16, v16i16,
3787 SSE_INTALU_ITINS_P, 1>;
3788 defm PADDD : PDI_binop_all<0xFE, "paddd", add, v4i32, v8i32,
3789 SSE_INTALU_ITINS_P, 1>;
3790 defm PADDQ : PDI_binop_all<0xD4, "paddq", add, v2i64, v4i64,
3791 SSE_INTALUQ_ITINS_P, 1>;
3792 defm PMULLW : PDI_binop_all<0xD5, "pmullw", mul, v8i16, v16i16,
3793 SSE_INTMUL_ITINS_P, 1>;
3794 defm PSUBB : PDI_binop_all<0xF8, "psubb", sub, v16i8, v32i8,
3795 SSE_INTALU_ITINS_P, 0>;
3796 defm PSUBW : PDI_binop_all<0xF9, "psubw", sub, v8i16, v16i16,
3797 SSE_INTALU_ITINS_P, 0>;
3798 defm PSUBD : PDI_binop_all<0xFA, "psubd", sub, v4i32, v8i32,
3799 SSE_INTALU_ITINS_P, 0>;
3800 defm PSUBQ : PDI_binop_all<0xFB, "psubq", sub, v2i64, v4i64,
3801 SSE_INTALUQ_ITINS_P, 0>;
3802 defm PSUBUSB : PDI_binop_all<0xD8, "psubusb", X86subus, v16i8, v32i8,
3803 SSE_INTALU_ITINS_P, 0>;
3804 defm PSUBUSW : PDI_binop_all<0xD9, "psubusw", X86subus, v8i16, v16i16,
3805 SSE_INTALU_ITINS_P, 0>;
3806 defm PMINUB : PDI_binop_all<0xDA, "pminub", X86umin, v16i8, v32i8,
3807 SSE_INTALU_ITINS_P, 1>;
3808 defm PMINSW : PDI_binop_all<0xEA, "pminsw", X86smin, v8i16, v16i16,
3809 SSE_INTALU_ITINS_P, 1>;
3810 defm PMAXUB : PDI_binop_all<0xDE, "pmaxub", X86umax, v16i8, v32i8,
3811 SSE_INTALU_ITINS_P, 1>;
3812 defm PMAXSW : PDI_binop_all<0xEE, "pmaxsw", X86smax, v8i16, v16i16,
3813 SSE_INTALU_ITINS_P, 1>;
3816 defm PSUBSB : PDI_binop_all_int<0xE8, "psubsb", int_x86_sse2_psubs_b,
3817 int_x86_avx2_psubs_b, SSE_INTALU_ITINS_P, 0>;
3818 defm PSUBSW : PDI_binop_all_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
3819 int_x86_avx2_psubs_w, SSE_INTALU_ITINS_P, 0>;
3820 defm PADDSB : PDI_binop_all_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
3821 int_x86_avx2_padds_b, SSE_INTALU_ITINS_P, 1>;
3822 defm PADDSW : PDI_binop_all_int<0xED, "paddsw" , int_x86_sse2_padds_w,
3823 int_x86_avx2_padds_w, SSE_INTALU_ITINS_P, 1>;
3824 defm PADDUSB : PDI_binop_all_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
3825 int_x86_avx2_paddus_b, SSE_INTALU_ITINS_P, 1>;
3826 defm PADDUSW : PDI_binop_all_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
3827 int_x86_avx2_paddus_w, SSE_INTALU_ITINS_P, 1>;
3828 defm PMULHUW : PDI_binop_all_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
3829 int_x86_avx2_pmulhu_w, SSE_INTMUL_ITINS_P, 1>;
3830 defm PMULHW : PDI_binop_all_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
3831 int_x86_avx2_pmulh_w, SSE_INTMUL_ITINS_P, 1>;
3832 defm PMADDWD : PDI_binop_all_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
3833 int_x86_avx2_pmadd_wd, SSE_PMADD, 1>;
3834 defm PAVGB : PDI_binop_all_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
3835 int_x86_avx2_pavg_b, SSE_INTALU_ITINS_P, 1>;
3836 defm PAVGW : PDI_binop_all_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
3837 int_x86_avx2_pavg_w, SSE_INTALU_ITINS_P, 1>;
3838 defm PSADBW : PDI_binop_all_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
3839 int_x86_avx2_psad_bw, SSE_PMADD, 1>;
3841 let Predicates = [HasAVX] in
3842 defm VPMULUDQ : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v2i64, v4i32, VR128,
3843 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1, 0>,
3845 let Predicates = [HasAVX2] in
3846 defm VPMULUDQY : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v4i64, v8i32,
3847 VR256, memopv4i64, i256mem,
3848 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V, VEX_L;
3849 let Constraints = "$src1 = $dst" in
3850 defm PMULUDQ : PDI_binop_rm2<0xF4, "pmuludq", X86pmuludq, v2i64, v4i32, VR128,
3851 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1>;
3853 //===---------------------------------------------------------------------===//
3854 // SSE2 - Packed Integer Logical Instructions
3855 //===---------------------------------------------------------------------===//
3857 let Predicates = [HasAVX] in {
3858 defm VPSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3859 VR128, v8i16, v8i16, bc_v8i16,
3860 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3861 defm VPSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3862 VR128, v4i32, v4i32, bc_v4i32,
3863 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3864 defm VPSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3865 VR128, v2i64, v2i64, bc_v2i64,
3866 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3868 defm VPSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3869 VR128, v8i16, v8i16, bc_v8i16,
3870 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3871 defm VPSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3872 VR128, v4i32, v4i32, bc_v4i32,
3873 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3874 defm VPSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3875 VR128, v2i64, v2i64, bc_v2i64,
3876 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3878 defm VPSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3879 VR128, v8i16, v8i16, bc_v8i16,
3880 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3881 defm VPSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3882 VR128, v4i32, v4i32, bc_v4i32,
3883 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3885 let ExeDomain = SSEPackedInt, SchedRW = [WriteVecShift] in {
3886 // 128-bit logical shifts.
3887 def VPSLLDQri : PDIi8<0x73, MRM7r,
3888 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3889 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3891 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>,
3893 def VPSRLDQri : PDIi8<0x73, MRM3r,
3894 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3895 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3897 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>,
3899 // PSRADQri doesn't exist in SSE[1-3].
3901 } // Predicates = [HasAVX]
3903 let Predicates = [HasAVX2] in {
3904 defm VPSLLWY : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3905 VR256, v16i16, v8i16, bc_v8i16,
3906 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3907 defm VPSLLDY : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3908 VR256, v8i32, v4i32, bc_v4i32,
3909 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3910 defm VPSLLQY : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3911 VR256, v4i64, v2i64, bc_v2i64,
3912 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3914 defm VPSRLWY : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3915 VR256, v16i16, v8i16, bc_v8i16,
3916 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3917 defm VPSRLDY : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3918 VR256, v8i32, v4i32, bc_v4i32,
3919 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3920 defm VPSRLQY : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3921 VR256, v4i64, v2i64, bc_v2i64,
3922 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3924 defm VPSRAWY : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3925 VR256, v16i16, v8i16, bc_v8i16,
3926 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3927 defm VPSRADY : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3928 VR256, v8i32, v4i32, bc_v4i32,
3929 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3931 let ExeDomain = SSEPackedInt, SchedRW = [WriteVecShift] in {
3932 // 256-bit logical shifts.
3933 def VPSLLDQYri : PDIi8<0x73, MRM7r,
3934 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3935 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3937 (int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2))]>,
3939 def VPSRLDQYri : PDIi8<0x73, MRM3r,
3940 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3941 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3943 (int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2))]>,
3945 // PSRADQYri doesn't exist in SSE[1-3].
3947 } // Predicates = [HasAVX2]
3949 let Constraints = "$src1 = $dst" in {
3950 defm PSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "psllw", X86vshl, X86vshli,
3951 VR128, v8i16, v8i16, bc_v8i16,
3952 SSE_INTSHIFT_ITINS_P>;
3953 defm PSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "pslld", X86vshl, X86vshli,
3954 VR128, v4i32, v4i32, bc_v4i32,
3955 SSE_INTSHIFT_ITINS_P>;
3956 defm PSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "psllq", X86vshl, X86vshli,
3957 VR128, v2i64, v2i64, bc_v2i64,
3958 SSE_INTSHIFT_ITINS_P>;
3960 defm PSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "psrlw", X86vsrl, X86vsrli,
3961 VR128, v8i16, v8i16, bc_v8i16,
3962 SSE_INTSHIFT_ITINS_P>;
3963 defm PSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "psrld", X86vsrl, X86vsrli,
3964 VR128, v4i32, v4i32, bc_v4i32,
3965 SSE_INTSHIFT_ITINS_P>;
3966 defm PSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "psrlq", X86vsrl, X86vsrli,
3967 VR128, v2i64, v2i64, bc_v2i64,
3968 SSE_INTSHIFT_ITINS_P>;
3970 defm PSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "psraw", X86vsra, X86vsrai,
3971 VR128, v8i16, v8i16, bc_v8i16,
3972 SSE_INTSHIFT_ITINS_P>;
3973 defm PSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "psrad", X86vsra, X86vsrai,
3974 VR128, v4i32, v4i32, bc_v4i32,
3975 SSE_INTSHIFT_ITINS_P>;
3977 let ExeDomain = SSEPackedInt, SchedRW = [WriteVecShift] in {
3978 // 128-bit logical shifts.
3979 def PSLLDQri : PDIi8<0x73, MRM7r,
3980 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3981 "pslldq\t{$src2, $dst|$dst, $src2}",
3983 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))],
3984 IIC_SSE_INTSHDQ_P_RI>;
3985 def PSRLDQri : PDIi8<0x73, MRM3r,
3986 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3987 "psrldq\t{$src2, $dst|$dst, $src2}",
3989 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))],
3990 IIC_SSE_INTSHDQ_P_RI>;
3991 // PSRADQri doesn't exist in SSE[1-3].
3993 } // Constraints = "$src1 = $dst"
3995 let Predicates = [HasAVX] in {
3996 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3997 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3998 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3999 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4000 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
4001 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4003 // Shift up / down and insert zero's.
4004 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
4005 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4006 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
4007 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4010 let Predicates = [HasAVX2] in {
4011 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
4012 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
4013 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
4014 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
4017 let Predicates = [UseSSE2] in {
4018 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
4019 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4020 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
4021 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4022 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
4023 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4025 // Shift up / down and insert zero's.
4026 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
4027 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4028 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
4029 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4032 //===---------------------------------------------------------------------===//
4033 // SSE2 - Packed Integer Comparison Instructions
4034 //===---------------------------------------------------------------------===//
4036 defm PCMPEQB : PDI_binop_all<0x74, "pcmpeqb", X86pcmpeq, v16i8, v32i8,
4037 SSE_INTALU_ITINS_P, 1>;
4038 defm PCMPEQW : PDI_binop_all<0x75, "pcmpeqw", X86pcmpeq, v8i16, v16i16,
4039 SSE_INTALU_ITINS_P, 1>;
4040 defm PCMPEQD : PDI_binop_all<0x76, "pcmpeqd", X86pcmpeq, v4i32, v8i32,
4041 SSE_INTALU_ITINS_P, 1>;
4042 defm PCMPGTB : PDI_binop_all<0x64, "pcmpgtb", X86pcmpgt, v16i8, v32i8,
4043 SSE_INTALU_ITINS_P, 0>;
4044 defm PCMPGTW : PDI_binop_all<0x65, "pcmpgtw", X86pcmpgt, v8i16, v16i16,
4045 SSE_INTALU_ITINS_P, 0>;
4046 defm PCMPGTD : PDI_binop_all<0x66, "pcmpgtd", X86pcmpgt, v4i32, v8i32,
4047 SSE_INTALU_ITINS_P, 0>;
4049 //===---------------------------------------------------------------------===//
4050 // SSE2 - Packed Integer Pack Instructions
4051 //===---------------------------------------------------------------------===//
4053 defm PACKSSWB : PDI_binop_all_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
4054 int_x86_avx2_packsswb, SSE_INTALU_ITINS_P, 0>;
4055 defm PACKSSDW : PDI_binop_all_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
4056 int_x86_avx2_packssdw, SSE_INTALU_ITINS_P, 0>;
4057 defm PACKUSWB : PDI_binop_all_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
4058 int_x86_avx2_packuswb, SSE_INTALU_ITINS_P, 0>;
4060 //===---------------------------------------------------------------------===//
4061 // SSE2 - Packed Integer Shuffle Instructions
4062 //===---------------------------------------------------------------------===//
4064 let ExeDomain = SSEPackedInt in {
4065 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt128, ValueType vt256,
4067 let Predicates = [HasAVX] in {
4068 def V#NAME#ri : Ii8<0x70, MRMSrcReg, (outs VR128:$dst),
4069 (ins VR128:$src1, i8imm:$src2),
4070 !strconcat("v", OpcodeStr,
4071 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4073 (vt128 (OpNode VR128:$src1, (i8 imm:$src2))))],
4074 IIC_SSE_PSHUF_RI>, VEX, Sched<[WriteShuffle]>;
4075 def V#NAME#mi : Ii8<0x70, MRMSrcMem, (outs VR128:$dst),
4076 (ins i128mem:$src1, i8imm:$src2),
4077 !strconcat("v", OpcodeStr,
4078 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4080 (vt128 (OpNode (bitconvert (memopv2i64 addr:$src1)),
4081 (i8 imm:$src2))))], IIC_SSE_PSHUF_MI>, VEX,
4082 Sched<[WriteShuffleLd]>;
4085 let Predicates = [HasAVX2] in {
4086 def V#NAME#Yri : Ii8<0x70, MRMSrcReg, (outs VR256:$dst),
4087 (ins VR256:$src1, i8imm:$src2),
4088 !strconcat("v", OpcodeStr,
4089 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4091 (vt256 (OpNode VR256:$src1, (i8 imm:$src2))))],
4092 IIC_SSE_PSHUF_RI>, VEX, VEX_L, Sched<[WriteShuffle]>;
4093 def V#NAME#Ymi : Ii8<0x70, MRMSrcMem, (outs VR256:$dst),
4094 (ins i256mem:$src1, i8imm:$src2),
4095 !strconcat("v", OpcodeStr,
4096 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4098 (vt256 (OpNode (bitconvert (memopv4i64 addr:$src1)),
4099 (i8 imm:$src2))))], IIC_SSE_PSHUF_MI>, VEX, VEX_L,
4100 Sched<[WriteShuffleLd]>;
4103 let Predicates = [UseSSE2] in {
4104 def ri : Ii8<0x70, MRMSrcReg,
4105 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
4106 !strconcat(OpcodeStr,
4107 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4109 (vt128 (OpNode VR128:$src1, (i8 imm:$src2))))],
4110 IIC_SSE_PSHUF_RI>, Sched<[WriteShuffle]>;
4111 def mi : Ii8<0x70, MRMSrcMem,
4112 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
4113 !strconcat(OpcodeStr,
4114 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4116 (vt128 (OpNode (bitconvert (memopv2i64 addr:$src1)),
4117 (i8 imm:$src2))))], IIC_SSE_PSHUF_MI>,
4118 Sched<[WriteShuffleLd]>;
4121 } // ExeDomain = SSEPackedInt
4123 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, v8i32, X86PShufd>, TB, OpSize;
4124 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, v16i16, X86PShufhw>, XS;
4125 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, v16i16, X86PShuflw>, XD;
4127 let Predicates = [HasAVX] in {
4128 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
4129 (VPSHUFDmi addr:$src1, imm:$imm)>;
4130 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4131 (VPSHUFDri VR128:$src1, imm:$imm)>;
4134 let Predicates = [UseSSE2] in {
4135 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
4136 (PSHUFDmi addr:$src1, imm:$imm)>;
4137 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4138 (PSHUFDri VR128:$src1, imm:$imm)>;
4141 //===---------------------------------------------------------------------===//
4142 // SSE2 - Packed Integer Unpack Instructions
4143 //===---------------------------------------------------------------------===//
4145 let ExeDomain = SSEPackedInt in {
4146 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
4147 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
4148 def rr : PDI<opc, MRMSrcReg,
4149 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4151 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4152 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4153 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))],
4154 IIC_SSE_UNPCK>, Sched<[WriteShuffle]>;
4155 def rm : PDI<opc, MRMSrcMem,
4156 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4158 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4159 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4160 [(set VR128:$dst, (OpNode VR128:$src1,
4161 (bc_frag (memopv2i64
4164 Sched<[WriteShuffleLd, ReadAfterLd]>;
4167 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
4168 SDNode OpNode, PatFrag bc_frag> {
4169 def Yrr : PDI<opc, MRMSrcReg,
4170 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4171 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4172 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>,
4173 Sched<[WriteShuffle]>;
4174 def Yrm : PDI<opc, MRMSrcMem,
4175 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4176 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4177 [(set VR256:$dst, (OpNode VR256:$src1,
4178 (bc_frag (memopv4i64 addr:$src2))))]>,
4179 Sched<[WriteShuffleLd, ReadAfterLd]>;
4182 let Predicates = [HasAVX] in {
4183 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
4184 bc_v16i8, 0>, VEX_4V;
4185 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
4186 bc_v8i16, 0>, VEX_4V;
4187 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
4188 bc_v4i32, 0>, VEX_4V;
4189 defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
4190 bc_v2i64, 0>, VEX_4V;
4192 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
4193 bc_v16i8, 0>, VEX_4V;
4194 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
4195 bc_v8i16, 0>, VEX_4V;
4196 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
4197 bc_v4i32, 0>, VEX_4V;
4198 defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
4199 bc_v2i64, 0>, VEX_4V;
4202 let Predicates = [HasAVX2] in {
4203 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
4204 bc_v32i8>, VEX_4V, VEX_L;
4205 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
4206 bc_v16i16>, VEX_4V, VEX_L;
4207 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
4208 bc_v8i32>, VEX_4V, VEX_L;
4209 defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
4210 bc_v4i64>, VEX_4V, VEX_L;
4212 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
4213 bc_v32i8>, VEX_4V, VEX_L;
4214 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
4215 bc_v16i16>, VEX_4V, VEX_L;
4216 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
4217 bc_v8i32>, VEX_4V, VEX_L;
4218 defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
4219 bc_v4i64>, VEX_4V, VEX_L;
4222 let Constraints = "$src1 = $dst" in {
4223 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
4225 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
4227 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
4229 defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
4232 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
4234 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
4236 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
4238 defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
4241 } // ExeDomain = SSEPackedInt
4243 //===---------------------------------------------------------------------===//
4244 // SSE2 - Packed Integer Extract and Insert
4245 //===---------------------------------------------------------------------===//
4247 let ExeDomain = SSEPackedInt in {
4248 multiclass sse2_pinsrw<bit Is2Addr = 1> {
4249 def rri : Ii8<0xC4, MRMSrcReg,
4250 (outs VR128:$dst), (ins VR128:$src1,
4251 GR32:$src2, i32i8imm:$src3),
4253 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4254 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4256 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))], IIC_SSE_PINSRW>,
4257 Sched<[WriteShuffle]>;
4258 def rmi : Ii8<0xC4, MRMSrcMem,
4259 (outs VR128:$dst), (ins VR128:$src1,
4260 i16mem:$src2, i32i8imm:$src3),
4262 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4263 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4265 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
4266 imm:$src3))], IIC_SSE_PINSRW>,
4267 Sched<[WriteShuffleLd, ReadAfterLd]>;
4271 let Predicates = [HasAVX] in
4272 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
4273 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4274 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4275 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4276 imm:$src2))]>, TB, OpSize, VEX,
4277 Sched<[WriteShuffle]>;
4278 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
4279 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4280 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4281 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4282 imm:$src2))], IIC_SSE_PEXTRW>,
4283 Sched<[WriteShuffleLd, ReadAfterLd]>;
4286 let Predicates = [HasAVX] in {
4287 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
4288 let isAsmParserOnly = 1, hasSideEffects = 0 in
4289 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
4290 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4291 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
4292 []>, TB, OpSize, VEX_4V, Sched<[WriteShuffle]>;
4295 let Predicates = [UseSSE2], Constraints = "$src1 = $dst" in {
4296 defm PINSRW : sse2_pinsrw, TB, OpSize;
4297 let isAsmParserOnly = 1, hasSideEffects = 0 in
4298 def PINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
4299 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4300 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4301 []>, TB, OpSize, Sched<[WriteShuffle]>;
4302 } // Predicates = [UseSSE2], Constraints = "$src1 = $dst"
4304 } // ExeDomain = SSEPackedInt
4306 //===---------------------------------------------------------------------===//
4307 // SSE2 - Packed Mask Creation
4308 //===---------------------------------------------------------------------===//
4310 let ExeDomain = SSEPackedInt, SchedRW = [WriteVecLogic] in {
4312 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4313 "pmovmskb\t{$src, $dst|$dst, $src}",
4314 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4315 IIC_SSE_MOVMSK>, VEX;
4316 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
4317 "pmovmskb\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK>, VEX;
4319 let Predicates = [HasAVX2] in {
4320 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
4321 "pmovmskb\t{$src, $dst|$dst, $src}",
4322 [(set GR32:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>, VEX, VEX_L;
4323 def VPMOVMSKBYr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
4324 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
4327 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4328 "pmovmskb\t{$src, $dst|$dst, $src}",
4329 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4331 def PMOVMSKBr64r : PDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
4332 "pmovmskb\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK>;
4334 } // ExeDomain = SSEPackedInt
4336 //===---------------------------------------------------------------------===//
4337 // SSE2 - Conditional Store
4338 //===---------------------------------------------------------------------===//
4340 let ExeDomain = SSEPackedInt, SchedRW = [WriteStore] in {
4342 let Uses = [EDI], Predicates = [HasAVX,In32BitMode] in
4343 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4344 (ins VR128:$src, VR128:$mask),
4345 "maskmovdqu\t{$mask, $src|$src, $mask}",
4346 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4347 IIC_SSE_MASKMOV>, VEX;
4348 let Uses = [RDI], Predicates = [HasAVX,In64BitMode] in
4349 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4350 (ins VR128:$src, VR128:$mask),
4351 "maskmovdqu\t{$mask, $src|$src, $mask}",
4352 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4353 IIC_SSE_MASKMOV>, VEX;
4355 let Uses = [EDI], Predicates = [UseSSE2,In32BitMode] in
4356 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4357 "maskmovdqu\t{$mask, $src|$src, $mask}",
4358 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4360 let Uses = [RDI], Predicates = [UseSSE2,In64BitMode] in
4361 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4362 "maskmovdqu\t{$mask, $src|$src, $mask}",
4363 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4366 } // ExeDomain = SSEPackedInt
4368 //===---------------------------------------------------------------------===//
4369 // SSE2 - Move Doubleword
4370 //===---------------------------------------------------------------------===//
4372 //===---------------------------------------------------------------------===//
4373 // Move Int Doubleword to Packed Double Int
4375 def VMOVDI2PDIrr : VS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4376 "movd\t{$src, $dst|$dst, $src}",
4378 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>,
4379 VEX, Sched<[WriteMove]>;
4380 def VMOVDI2PDIrm : VS2I<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4381 "movd\t{$src, $dst|$dst, $src}",
4383 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4385 VEX, Sched<[WriteLoad]>;
4386 def VMOV64toPQIrr : VRS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4387 "movq\t{$src, $dst|$dst, $src}",
4389 (v2i64 (scalar_to_vector GR64:$src)))],
4390 IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
4391 let isCodeGenOnly = 1 in
4392 def VMOV64toSDrr : VRS2I<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4393 "movq\t{$src, $dst|$dst, $src}",
4394 [(set FR64:$dst, (bitconvert GR64:$src))],
4395 IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
4397 def MOVDI2PDIrr : S2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4398 "movd\t{$src, $dst|$dst, $src}",
4400 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>,
4402 def MOVDI2PDIrm : S2I<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4403 "movd\t{$src, $dst|$dst, $src}",
4405 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4406 IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
4407 def MOV64toPQIrr : RS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4408 "mov{d|q}\t{$src, $dst|$dst, $src}",
4410 (v2i64 (scalar_to_vector GR64:$src)))],
4411 IIC_SSE_MOVDQ>, Sched<[WriteMove]>;
4412 let isCodeGenOnly = 1 in
4413 def MOV64toSDrr : RS2I<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4414 "mov{d|q}\t{$src, $dst|$dst, $src}",
4415 [(set FR64:$dst, (bitconvert GR64:$src))],
4416 IIC_SSE_MOVDQ>, Sched<[WriteMove]>;
4418 //===---------------------------------------------------------------------===//
4419 // Move Int Doubleword to Single Scalar
4421 let isCodeGenOnly = 1 in {
4422 def VMOVDI2SSrr : VS2I<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4423 "movd\t{$src, $dst|$dst, $src}",
4424 [(set FR32:$dst, (bitconvert GR32:$src))],
4425 IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
4427 def VMOVDI2SSrm : VS2I<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4428 "movd\t{$src, $dst|$dst, $src}",
4429 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4431 VEX, Sched<[WriteLoad]>;
4432 def MOVDI2SSrr : S2I<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4433 "movd\t{$src, $dst|$dst, $src}",
4434 [(set FR32:$dst, (bitconvert GR32:$src))],
4435 IIC_SSE_MOVDQ>, Sched<[WriteMove]>;
4437 def MOVDI2SSrm : S2I<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4438 "movd\t{$src, $dst|$dst, $src}",
4439 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4440 IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
4443 //===---------------------------------------------------------------------===//
4444 // Move Packed Doubleword Int to Packed Double Int
4446 def VMOVPDI2DIrr : VS2I<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4447 "movd\t{$src, $dst|$dst, $src}",
4448 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4449 (iPTR 0)))], IIC_SSE_MOVD_ToGP>, VEX,
4451 def VMOVPDI2DImr : VS2I<0x7E, MRMDestMem, (outs),
4452 (ins i32mem:$dst, VR128:$src),
4453 "movd\t{$src, $dst|$dst, $src}",
4454 [(store (i32 (vector_extract (v4i32 VR128:$src),
4455 (iPTR 0))), addr:$dst)], IIC_SSE_MOVDQ>,
4456 VEX, Sched<[WriteLoad]>;
4457 def MOVPDI2DIrr : S2I<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4458 "movd\t{$src, $dst|$dst, $src}",
4459 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4460 (iPTR 0)))], IIC_SSE_MOVD_ToGP>,
4462 def MOVPDI2DImr : S2I<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4463 "movd\t{$src, $dst|$dst, $src}",
4464 [(store (i32 (vector_extract (v4i32 VR128:$src),
4465 (iPTR 0))), addr:$dst)],
4466 IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
4468 def : Pat<(v8i32 (X86Vinsert (v8i32 immAllZerosV), GR32:$src2, (iPTR 0))),
4469 (SUBREG_TO_REG (i32 0), (VMOVDI2PDIrr GR32:$src2), sub_xmm)>;
4471 def : Pat<(v4i64 (X86Vinsert (bc_v4i64 (v8i32 immAllZerosV)), GR64:$src2, (iPTR 0))),
4472 (SUBREG_TO_REG (i32 0), (VMOV64toPQIrr GR64:$src2), sub_xmm)>;
4474 def : Pat<(v8i32 (X86Vinsert undef, GR32:$src2, (iPTR 0))),
4475 (SUBREG_TO_REG (i32 0), (VMOVDI2PDIrr GR32:$src2), sub_xmm)>;
4477 def : Pat<(v4i64 (X86Vinsert undef, GR64:$src2, (iPTR 0))),
4478 (SUBREG_TO_REG (i32 0), (VMOV64toPQIrr GR64:$src2), sub_xmm)>;
4480 //===---------------------------------------------------------------------===//
4481 // Move Packed Doubleword Int first element to Doubleword Int
4483 let SchedRW = [WriteMove] in {
4484 def VMOVPQIto64rr : VRS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4485 "movq\t{$src, $dst|$dst, $src}",
4486 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4491 def MOVPQIto64rr : RS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4492 "mov{d|q}\t{$src, $dst|$dst, $src}",
4493 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4498 //===---------------------------------------------------------------------===//
4499 // Bitcast FR64 <-> GR64
4501 let isCodeGenOnly = 1 in {
4502 let Predicates = [UseAVX] in
4503 def VMOV64toSDrm : VS2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4504 "movq\t{$src, $dst|$dst, $src}",
4505 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4506 VEX, Sched<[WriteLoad]>;
4507 def VMOVSDto64rr : VRS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4508 "movq\t{$src, $dst|$dst, $src}",
4509 [(set GR64:$dst, (bitconvert FR64:$src))],
4510 IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
4511 def VMOVSDto64mr : VRS2I<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4512 "movq\t{$src, $dst|$dst, $src}",
4513 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4514 IIC_SSE_MOVDQ>, VEX, Sched<[WriteStore]>;
4516 def MOV64toSDrm : S2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4517 "movq\t{$src, $dst|$dst, $src}",
4518 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))],
4519 IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
4520 def MOVSDto64rr : RS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4521 "mov{d|q}\t{$src, $dst|$dst, $src}",
4522 [(set GR64:$dst, (bitconvert FR64:$src))],
4523 IIC_SSE_MOVD_ToGP>, Sched<[WriteMove]>;
4524 def MOVSDto64mr : RS2I<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4525 "movq\t{$src, $dst|$dst, $src}",
4526 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4527 IIC_SSE_MOVDQ>, Sched<[WriteStore]>;
4530 //===---------------------------------------------------------------------===//
4531 // Move Scalar Single to Double Int
4533 let isCodeGenOnly = 1 in {
4534 def VMOVSS2DIrr : VS2I<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4535 "movd\t{$src, $dst|$dst, $src}",
4536 [(set GR32:$dst, (bitconvert FR32:$src))],
4537 IIC_SSE_MOVD_ToGP>, VEX, Sched<[WriteMove]>;
4538 def VMOVSS2DImr : VS2I<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4539 "movd\t{$src, $dst|$dst, $src}",
4540 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4541 IIC_SSE_MOVDQ>, VEX, Sched<[WriteStore]>;
4542 def MOVSS2DIrr : S2I<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4543 "movd\t{$src, $dst|$dst, $src}",
4544 [(set GR32:$dst, (bitconvert FR32:$src))],
4545 IIC_SSE_MOVD_ToGP>, Sched<[WriteMove]>;
4546 def MOVSS2DImr : S2I<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4547 "movd\t{$src, $dst|$dst, $src}",
4548 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4549 IIC_SSE_MOVDQ>, Sched<[WriteStore]>;
4552 //===---------------------------------------------------------------------===//
4553 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4555 let isCodeGenOnly = 1, SchedRW = [WriteMove] in {
4556 let AddedComplexity = 15 in {
4557 def VMOVZDI2PDIrr : VS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4558 "movd\t{$src, $dst|$dst, $src}",
4559 [(set VR128:$dst, (v4i32 (X86vzmovl
4560 (v4i32 (scalar_to_vector GR32:$src)))))],
4561 IIC_SSE_MOVDQ>, VEX;
4562 def VMOVZQI2PQIrr : VS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4563 "movq\t{$src, $dst|$dst, $src}", // X86-64 only
4564 [(set VR128:$dst, (v2i64 (X86vzmovl
4565 (v2i64 (scalar_to_vector GR64:$src)))))],
4569 let AddedComplexity = 15 in {
4570 def MOVZDI2PDIrr : S2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4571 "movd\t{$src, $dst|$dst, $src}",
4572 [(set VR128:$dst, (v4i32 (X86vzmovl
4573 (v4i32 (scalar_to_vector GR32:$src)))))],
4575 def MOVZQI2PQIrr : RS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4576 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4577 [(set VR128:$dst, (v2i64 (X86vzmovl
4578 (v2i64 (scalar_to_vector GR64:$src)))))],
4581 } // isCodeGenOnly, SchedRW
4583 let isCodeGenOnly = 1, AddedComplexity = 20, SchedRW = [WriteLoad] in {
4584 def VMOVZDI2PDIrm : VS2I<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4585 "movd\t{$src, $dst|$dst, $src}",
4587 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4588 (loadi32 addr:$src))))))],
4589 IIC_SSE_MOVDQ>, VEX;
4590 def MOVZDI2PDIrm : S2I<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4591 "movd\t{$src, $dst|$dst, $src}",
4593 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4594 (loadi32 addr:$src))))))],
4596 } // isCodeGenOnly, AddedComplexity, SchedRW
4598 let Predicates = [UseAVX] in {
4599 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4600 let AddedComplexity = 20 in {
4601 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4602 (VMOVZDI2PDIrm addr:$src)>;
4603 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4604 (VMOVZDI2PDIrm addr:$src)>;
4606 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4607 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4608 (v4i32 (scalar_to_vector GR32:$src)),(iPTR 0)))),
4609 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
4610 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4611 (v2i64 (scalar_to_vector GR64:$src)),(iPTR 0)))),
4612 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4615 let Predicates = [UseSSE2], AddedComplexity = 20 in {
4616 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4617 (MOVZDI2PDIrm addr:$src)>;
4618 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4619 (MOVZDI2PDIrm addr:$src)>;
4622 // These are the correct encodings of the instructions so that we know how to
4623 // read correct assembly, even though we continue to emit the wrong ones for
4624 // compatibility with Darwin's buggy assembler.
4625 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4626 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4627 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4628 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4629 // Allow "vmovd" but print "vmovq" since we don't need compatibility for AVX.
4630 def : InstAlias<"vmovd\t{$src, $dst|$dst, $src}",
4631 (VMOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4632 def : InstAlias<"vmovd\t{$src, $dst|$dst, $src}",
4633 (VMOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4635 //===---------------------------------------------------------------------===//
4636 // SSE2 - Move Quadword
4637 //===---------------------------------------------------------------------===//
4639 //===---------------------------------------------------------------------===//
4640 // Move Quadword Int to Packed Quadword Int
4643 let SchedRW = [WriteLoad] in {
4644 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4645 "vmovq\t{$src, $dst|$dst, $src}",
4647 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4648 VEX, Requires<[UseAVX]>;
4649 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4650 "movq\t{$src, $dst|$dst, $src}",
4652 (v2i64 (scalar_to_vector (loadi64 addr:$src))))],
4654 Requires<[UseSSE2]>; // SSE2 instruction with XS Prefix
4657 //===---------------------------------------------------------------------===//
4658 // Move Packed Quadword Int to Quadword Int
4660 let SchedRW = [WriteStore] in {
4661 def VMOVPQI2QImr : VS2I<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4662 "movq\t{$src, $dst|$dst, $src}",
4663 [(store (i64 (vector_extract (v2i64 VR128:$src),
4664 (iPTR 0))), addr:$dst)],
4665 IIC_SSE_MOVDQ>, VEX;
4666 def MOVPQI2QImr : S2I<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4667 "movq\t{$src, $dst|$dst, $src}",
4668 [(store (i64 (vector_extract (v2i64 VR128:$src),
4669 (iPTR 0))), addr:$dst)],
4673 //===---------------------------------------------------------------------===//
4674 // Store / copy lower 64-bits of a XMM register.
4676 def VMOVLQ128mr : VS2I<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4677 "movq\t{$src, $dst|$dst, $src}",
4678 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX,
4679 Sched<[WriteStore]>;
4680 def MOVLQ128mr : S2I<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4681 "movq\t{$src, $dst|$dst, $src}",
4682 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)],
4683 IIC_SSE_MOVDQ>, Sched<[WriteStore]>;
4685 let isCodeGenOnly = 1, AddedComplexity = 20 in {
4686 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4687 "vmovq\t{$src, $dst|$dst, $src}",
4689 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4690 (loadi64 addr:$src))))))],
4692 XS, VEX, Requires<[UseAVX]>, Sched<[WriteLoad]>;
4694 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4695 "movq\t{$src, $dst|$dst, $src}",
4697 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4698 (loadi64 addr:$src))))))],
4700 XS, Requires<[UseSSE2]>, Sched<[WriteLoad]>;
4703 let Predicates = [UseAVX], AddedComplexity = 20 in {
4704 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4705 (VMOVZQI2PQIrm addr:$src)>;
4706 def : Pat<(v2i64 (X86vzload addr:$src)),
4707 (VMOVZQI2PQIrm addr:$src)>;
4710 let Predicates = [UseSSE2], AddedComplexity = 20 in {
4711 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4712 (MOVZQI2PQIrm addr:$src)>;
4713 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4716 let Predicates = [HasAVX] in {
4717 def : Pat<(v4i64 (alignedX86vzload addr:$src)),
4718 (SUBREG_TO_REG (i32 0), (VMOVAPSrm addr:$src), sub_xmm)>;
4719 def : Pat<(v4i64 (X86vzload addr:$src)),
4720 (SUBREG_TO_REG (i32 0), (VMOVUPSrm addr:$src), sub_xmm)>;
4723 //===---------------------------------------------------------------------===//
4724 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4725 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4727 let SchedRW = [WriteVecLogic] in {
4728 let AddedComplexity = 15 in
4729 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4730 "vmovq\t{$src, $dst|$dst, $src}",
4731 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4733 XS, VEX, Requires<[UseAVX]>;
4734 let AddedComplexity = 15 in
4735 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4736 "movq\t{$src, $dst|$dst, $src}",
4737 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4739 XS, Requires<[UseSSE2]>;
4742 let isCodeGenOnly = 1, SchedRW = [WriteVecLogicLd] in {
4743 let AddedComplexity = 20 in
4744 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4745 "vmovq\t{$src, $dst|$dst, $src}",
4746 [(set VR128:$dst, (v2i64 (X86vzmovl
4747 (loadv2i64 addr:$src))))],
4749 XS, VEX, Requires<[UseAVX]>;
4750 let AddedComplexity = 20 in {
4751 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4752 "movq\t{$src, $dst|$dst, $src}",
4753 [(set VR128:$dst, (v2i64 (X86vzmovl
4754 (loadv2i64 addr:$src))))],
4756 XS, Requires<[UseSSE2]>;
4758 } // isCodeGenOnly, SchedRW
4760 let AddedComplexity = 20 in {
4761 let Predicates = [UseAVX] in {
4762 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4763 (VMOVZPQILo2PQIrr VR128:$src)>;
4765 let Predicates = [UseSSE2] in {
4766 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4767 (MOVZPQILo2PQIrr VR128:$src)>;
4771 //===---------------------------------------------------------------------===//
4772 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4773 //===---------------------------------------------------------------------===//
4774 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4775 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4776 X86MemOperand x86memop> {
4777 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4778 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4779 [(set RC:$dst, (vt (OpNode RC:$src)))],
4780 IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>;
4781 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4782 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4783 [(set RC:$dst, (OpNode (mem_frag addr:$src)))],
4784 IIC_SSE_MOV_LH>, Sched<[WriteShuffleLd]>;
4787 let Predicates = [HasAVX] in {
4788 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4789 v4f32, VR128, memopv4f32, f128mem>, VEX;
4790 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4791 v4f32, VR128, memopv4f32, f128mem>, VEX;
4792 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4793 v8f32, VR256, memopv8f32, f256mem>, VEX, VEX_L;
4794 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4795 v8f32, VR256, memopv8f32, f256mem>, VEX, VEX_L;
4797 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4798 memopv4f32, f128mem>;
4799 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4800 memopv4f32, f128mem>;
4802 let Predicates = [HasAVX] in {
4803 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4804 (VMOVSHDUPrr VR128:$src)>;
4805 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4806 (VMOVSHDUPrm addr:$src)>;
4807 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4808 (VMOVSLDUPrr VR128:$src)>;
4809 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4810 (VMOVSLDUPrm addr:$src)>;
4811 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4812 (VMOVSHDUPYrr VR256:$src)>;
4813 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4814 (VMOVSHDUPYrm addr:$src)>;
4815 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4816 (VMOVSLDUPYrr VR256:$src)>;
4817 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
4818 (VMOVSLDUPYrm addr:$src)>;
4821 let Predicates = [UseSSE3] in {
4822 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4823 (MOVSHDUPrr VR128:$src)>;
4824 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4825 (MOVSHDUPrm addr:$src)>;
4826 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4827 (MOVSLDUPrr VR128:$src)>;
4828 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4829 (MOVSLDUPrm addr:$src)>;
4832 //===---------------------------------------------------------------------===//
4833 // SSE3 - Replicate Double FP - MOVDDUP
4834 //===---------------------------------------------------------------------===//
4836 multiclass sse3_replicate_dfp<string OpcodeStr> {
4837 let neverHasSideEffects = 1 in
4838 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4839 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4840 [], IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>;
4841 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
4842 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4845 (scalar_to_vector (loadf64 addr:$src)))))],
4846 IIC_SSE_MOV_LH>, Sched<[WriteShuffleLd]>;
4849 // FIXME: Merge with above classe when there're patterns for the ymm version
4850 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
4851 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
4852 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4853 [(set VR256:$dst, (v4f64 (X86Movddup VR256:$src)))]>,
4854 Sched<[WriteShuffle]>;
4855 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
4856 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4859 (scalar_to_vector (loadf64 addr:$src)))))]>,
4860 Sched<[WriteShuffleLd]>;
4863 let Predicates = [HasAVX] in {
4864 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
4865 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX, VEX_L;
4868 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
4870 let Predicates = [HasAVX] in {
4871 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4872 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4873 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4874 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4875 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4876 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4877 def : Pat<(X86Movddup (bc_v2f64
4878 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4879 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4882 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
4883 (VMOVDDUPYrm addr:$src)>;
4884 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
4885 (VMOVDDUPYrm addr:$src)>;
4886 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
4887 (VMOVDDUPYrm addr:$src)>;
4888 def : Pat<(X86Movddup (v4i64 VR256:$src)),
4889 (VMOVDDUPYrr VR256:$src)>;
4892 let Predicates = [UseSSE3] in {
4893 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4894 (MOVDDUPrm addr:$src)>;
4895 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4896 (MOVDDUPrm addr:$src)>;
4897 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4898 (MOVDDUPrm addr:$src)>;
4899 def : Pat<(X86Movddup (bc_v2f64
4900 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4901 (MOVDDUPrm addr:$src)>;
4904 //===---------------------------------------------------------------------===//
4905 // SSE3 - Move Unaligned Integer
4906 //===---------------------------------------------------------------------===//
4908 let SchedRW = [WriteLoad] in {
4909 let Predicates = [HasAVX] in {
4910 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4911 "vlddqu\t{$src, $dst|$dst, $src}",
4912 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
4913 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
4914 "vlddqu\t{$src, $dst|$dst, $src}",
4915 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>,
4918 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4919 "lddqu\t{$src, $dst|$dst, $src}",
4920 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))],
4924 //===---------------------------------------------------------------------===//
4925 // SSE3 - Arithmetic
4926 //===---------------------------------------------------------------------===//
4928 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
4929 X86MemOperand x86memop, OpndItins itins,
4931 def rr : I<0xD0, MRMSrcReg,
4932 (outs RC:$dst), (ins RC:$src1, RC:$src2),
4934 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4935 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4936 [(set RC:$dst, (Int RC:$src1, RC:$src2))], itins.rr>,
4937 Sched<[itins.Sched]>;
4938 def rm : I<0xD0, MRMSrcMem,
4939 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4941 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4942 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4943 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))], itins.rr>,
4944 Sched<[itins.Sched.Folded, ReadAfterLd]>;
4947 let Predicates = [HasAVX] in {
4948 let ExeDomain = SSEPackedSingle in {
4949 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
4950 f128mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V;
4951 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
4952 f256mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V, VEX_L;
4954 let ExeDomain = SSEPackedDouble in {
4955 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
4956 f128mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V;
4957 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
4958 f256mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V, VEX_L;
4961 let Constraints = "$src1 = $dst", Predicates = [UseSSE3] in {
4962 let ExeDomain = SSEPackedSingle in
4963 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
4964 f128mem, SSE_ALU_F32P>, TB, XD;
4965 let ExeDomain = SSEPackedDouble in
4966 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
4967 f128mem, SSE_ALU_F64P>, TB, OpSize;
4970 //===---------------------------------------------------------------------===//
4971 // SSE3 Instructions
4972 //===---------------------------------------------------------------------===//
4975 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4976 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
4977 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4979 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4980 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4981 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>,
4984 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4986 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4987 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4988 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
4989 IIC_SSE_HADDSUB_RM>, Sched<[WriteFAddLd, ReadAfterLd]>;
4991 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4992 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
4993 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4995 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4996 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4997 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>,
5000 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5002 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5003 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5004 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
5005 IIC_SSE_HADDSUB_RM>, Sched<[WriteFAddLd, ReadAfterLd]>;
5008 let Predicates = [HasAVX] in {
5009 let ExeDomain = SSEPackedSingle in {
5010 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
5011 X86fhadd, 0>, VEX_4V;
5012 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
5013 X86fhsub, 0>, VEX_4V;
5014 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
5015 X86fhadd, 0>, VEX_4V, VEX_L;
5016 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
5017 X86fhsub, 0>, VEX_4V, VEX_L;
5019 let ExeDomain = SSEPackedDouble in {
5020 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
5021 X86fhadd, 0>, VEX_4V;
5022 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
5023 X86fhsub, 0>, VEX_4V;
5024 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
5025 X86fhadd, 0>, VEX_4V, VEX_L;
5026 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
5027 X86fhsub, 0>, VEX_4V, VEX_L;
5031 let Constraints = "$src1 = $dst" in {
5032 let ExeDomain = SSEPackedSingle in {
5033 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
5034 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
5036 let ExeDomain = SSEPackedDouble in {
5037 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
5038 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
5042 //===---------------------------------------------------------------------===//
5043 // SSSE3 - Packed Absolute Instructions
5044 //===---------------------------------------------------------------------===//
5047 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5048 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
5049 Intrinsic IntId128> {
5050 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5052 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5053 [(set VR128:$dst, (IntId128 VR128:$src))], IIC_SSE_PABS_RR>,
5054 OpSize, Sched<[WriteVecALU]>;
5056 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5058 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5061 (bitconvert (memopv2i64 addr:$src))))], IIC_SSE_PABS_RM>,
5062 OpSize, Sched<[WriteVecALULd]>;
5065 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5066 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
5067 Intrinsic IntId256> {
5068 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5070 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5071 [(set VR256:$dst, (IntId256 VR256:$src))]>,
5072 OpSize, Sched<[WriteVecALU]>;
5074 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5076 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5079 (bitconvert (memopv4i64 addr:$src))))]>, OpSize,
5080 Sched<[WriteVecALULd]>;
5083 // Helper fragments to match sext vXi1 to vXiY.
5084 def v16i1sextv16i8 : PatLeaf<(v16i8 (X86pcmpgt (bc_v16i8 (v4i32 immAllZerosV)),
5086 def v8i1sextv8i16 : PatLeaf<(v8i16 (X86vsrai VR128:$src, (i32 15)))>;
5087 def v4i1sextv4i32 : PatLeaf<(v4i32 (X86vsrai VR128:$src, (i32 31)))>;
5088 def v32i1sextv32i8 : PatLeaf<(v32i8 (X86pcmpgt (bc_v32i8 (v8i32 immAllZerosV)),
5090 def v16i1sextv16i16: PatLeaf<(v16i16 (X86vsrai VR256:$src, (i32 15)))>;
5091 def v8i1sextv8i32 : PatLeaf<(v8i32 (X86vsrai VR256:$src, (i32 31)))>;
5093 let Predicates = [HasAVX] in {
5094 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb",
5095 int_x86_ssse3_pabs_b_128>, VEX;
5096 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw",
5097 int_x86_ssse3_pabs_w_128>, VEX;
5098 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd",
5099 int_x86_ssse3_pabs_d_128>, VEX;
5102 (bc_v2i64 (v16i1sextv16i8)),
5103 (bc_v2i64 (add (v16i8 VR128:$src), (v16i1sextv16i8)))),
5104 (VPABSBrr128 VR128:$src)>;
5106 (bc_v2i64 (v8i1sextv8i16)),
5107 (bc_v2i64 (add (v8i16 VR128:$src), (v8i1sextv8i16)))),
5108 (VPABSWrr128 VR128:$src)>;
5110 (bc_v2i64 (v4i1sextv4i32)),
5111 (bc_v2i64 (add (v4i32 VR128:$src), (v4i1sextv4i32)))),
5112 (VPABSDrr128 VR128:$src)>;
5115 let Predicates = [HasAVX2] in {
5116 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb",
5117 int_x86_avx2_pabs_b>, VEX, VEX_L;
5118 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw",
5119 int_x86_avx2_pabs_w>, VEX, VEX_L;
5120 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd",
5121 int_x86_avx2_pabs_d>, VEX, VEX_L;
5124 (bc_v4i64 (v32i1sextv32i8)),
5125 (bc_v4i64 (add (v32i8 VR256:$src), (v32i1sextv32i8)))),
5126 (VPABSBrr256 VR256:$src)>;
5128 (bc_v4i64 (v16i1sextv16i16)),
5129 (bc_v4i64 (add (v16i16 VR256:$src), (v16i1sextv16i16)))),
5130 (VPABSWrr256 VR256:$src)>;
5132 (bc_v4i64 (v8i1sextv8i32)),
5133 (bc_v4i64 (add (v8i32 VR256:$src), (v8i1sextv8i32)))),
5134 (VPABSDrr256 VR256:$src)>;
5137 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb",
5138 int_x86_ssse3_pabs_b_128>;
5139 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw",
5140 int_x86_ssse3_pabs_w_128>;
5141 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
5142 int_x86_ssse3_pabs_d_128>;
5144 let Predicates = [HasSSSE3] in {
5146 (bc_v2i64 (v16i1sextv16i8)),
5147 (bc_v2i64 (add (v16i8 VR128:$src), (v16i1sextv16i8)))),
5148 (PABSBrr128 VR128:$src)>;
5150 (bc_v2i64 (v8i1sextv8i16)),
5151 (bc_v2i64 (add (v8i16 VR128:$src), (v8i1sextv8i16)))),
5152 (PABSWrr128 VR128:$src)>;
5154 (bc_v2i64 (v4i1sextv4i32)),
5155 (bc_v2i64 (add (v4i32 VR128:$src), (v4i1sextv4i32)))),
5156 (PABSDrr128 VR128:$src)>;
5159 //===---------------------------------------------------------------------===//
5160 // SSSE3 - Packed Binary Operator Instructions
5161 //===---------------------------------------------------------------------===//
5163 let Sched = WriteVecALU in {
5164 def SSE_PHADDSUBD : OpndItins<
5165 IIC_SSE_PHADDSUBD_RR, IIC_SSE_PHADDSUBD_RM
5167 def SSE_PHADDSUBSW : OpndItins<
5168 IIC_SSE_PHADDSUBSW_RR, IIC_SSE_PHADDSUBSW_RM
5170 def SSE_PHADDSUBW : OpndItins<
5171 IIC_SSE_PHADDSUBW_RR, IIC_SSE_PHADDSUBW_RM
5174 let Sched = WriteShuffle in
5175 def SSE_PSHUFB : OpndItins<
5176 IIC_SSE_PSHUFB_RR, IIC_SSE_PSHUFB_RM
5178 let Sched = WriteVecALU in
5179 def SSE_PSIGN : OpndItins<
5180 IIC_SSE_PSIGN_RR, IIC_SSE_PSIGN_RM
5182 let Sched = WriteVecIMul in
5183 def SSE_PMULHRSW : OpndItins<
5184 IIC_SSE_PMULHRSW, IIC_SSE_PMULHRSW
5187 /// SS3I_binop_rm - Simple SSSE3 bin op
5188 multiclass SS3I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
5189 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
5190 X86MemOperand x86memop, OpndItins itins,
5192 let isCommutable = 1 in
5193 def rr : SS38I<opc, MRMSrcReg, (outs RC:$dst),
5194 (ins RC:$src1, RC:$src2),
5196 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5197 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5198 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>,
5199 OpSize, Sched<[itins.Sched]>;
5200 def rm : SS38I<opc, MRMSrcMem, (outs RC:$dst),
5201 (ins RC:$src1, x86memop:$src2),
5203 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5204 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5206 (OpVT (OpNode RC:$src1,
5207 (bitconvert (memop_frag addr:$src2)))))], itins.rm>, OpSize,
5208 Sched<[itins.Sched.Folded, ReadAfterLd]>;
5211 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
5212 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
5213 Intrinsic IntId128, OpndItins itins,
5215 let isCommutable = 1 in
5216 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5217 (ins VR128:$src1, VR128:$src2),
5219 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5220 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5221 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5222 OpSize, Sched<[itins.Sched]>;
5223 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5224 (ins VR128:$src1, i128mem:$src2),
5226 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5227 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5229 (IntId128 VR128:$src1,
5230 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize,
5231 Sched<[itins.Sched.Folded, ReadAfterLd]>;
5234 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
5235 Intrinsic IntId256> {
5236 let isCommutable = 1 in
5237 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5238 (ins VR256:$src1, VR256:$src2),
5239 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5240 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
5242 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5243 (ins VR256:$src1, i256mem:$src2),
5244 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5246 (IntId256 VR256:$src1,
5247 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
5250 let ImmT = NoImm, Predicates = [HasAVX] in {
5251 let isCommutable = 0 in {
5252 defm VPHADDW : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v8i16, VR128,
5253 memopv2i64, i128mem,
5254 SSE_PHADDSUBW, 0>, VEX_4V;
5255 defm VPHADDD : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v4i32, VR128,
5256 memopv2i64, i128mem,
5257 SSE_PHADDSUBD, 0>, VEX_4V;
5258 defm VPHSUBW : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v8i16, VR128,
5259 memopv2i64, i128mem,
5260 SSE_PHADDSUBW, 0>, VEX_4V;
5261 defm VPHSUBD : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v4i32, VR128,
5262 memopv2i64, i128mem,
5263 SSE_PHADDSUBD, 0>, VEX_4V;
5264 defm VPSIGNB : SS3I_binop_rm<0x08, "vpsignb", X86psign, v16i8, VR128,
5265 memopv2i64, i128mem,
5266 SSE_PSIGN, 0>, VEX_4V;
5267 defm VPSIGNW : SS3I_binop_rm<0x09, "vpsignw", X86psign, v8i16, VR128,
5268 memopv2i64, i128mem,
5269 SSE_PSIGN, 0>, VEX_4V;
5270 defm VPSIGND : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v4i32, VR128,
5271 memopv2i64, i128mem,
5272 SSE_PSIGN, 0>, VEX_4V;
5273 defm VPSHUFB : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v16i8, VR128,
5274 memopv2i64, i128mem,
5275 SSE_PSHUFB, 0>, VEX_4V;
5276 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw",
5277 int_x86_ssse3_phadd_sw_128,
5278 SSE_PHADDSUBSW, 0>, VEX_4V;
5279 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw",
5280 int_x86_ssse3_phsub_sw_128,
5281 SSE_PHADDSUBSW, 0>, VEX_4V;
5282 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw",
5283 int_x86_ssse3_pmadd_ub_sw_128,
5284 SSE_PMADD, 0>, VEX_4V;
5286 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw",
5287 int_x86_ssse3_pmul_hr_sw_128,
5288 SSE_PMULHRSW, 0>, VEX_4V;
5291 let ImmT = NoImm, Predicates = [HasAVX2] in {
5292 let isCommutable = 0 in {
5293 defm VPHADDWY : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v16i16, VR256,
5294 memopv4i64, i256mem,
5295 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5296 defm VPHADDDY : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v8i32, VR256,
5297 memopv4i64, i256mem,
5298 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5299 defm VPHSUBWY : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v16i16, VR256,
5300 memopv4i64, i256mem,
5301 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5302 defm VPHSUBDY : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v8i32, VR256,
5303 memopv4i64, i256mem,
5304 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5305 defm VPSIGNBY : SS3I_binop_rm<0x08, "vpsignb", X86psign, v32i8, VR256,
5306 memopv4i64, i256mem,
5307 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5308 defm VPSIGNWY : SS3I_binop_rm<0x09, "vpsignw", X86psign, v16i16, VR256,
5309 memopv4i64, i256mem,
5310 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5311 defm VPSIGNDY : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v8i32, VR256,
5312 memopv4i64, i256mem,
5313 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5314 defm VPSHUFBY : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v32i8, VR256,
5315 memopv4i64, i256mem,
5316 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5317 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
5318 int_x86_avx2_phadd_sw>, VEX_4V, VEX_L;
5319 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
5320 int_x86_avx2_phsub_sw>, VEX_4V, VEX_L;
5321 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
5322 int_x86_avx2_pmadd_ub_sw>, VEX_4V, VEX_L;
5324 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
5325 int_x86_avx2_pmul_hr_sw>, VEX_4V, VEX_L;
5328 // None of these have i8 immediate fields.
5329 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
5330 let isCommutable = 0 in {
5331 defm PHADDW : SS3I_binop_rm<0x01, "phaddw", X86hadd, v8i16, VR128,
5332 memopv2i64, i128mem, SSE_PHADDSUBW>;
5333 defm PHADDD : SS3I_binop_rm<0x02, "phaddd", X86hadd, v4i32, VR128,
5334 memopv2i64, i128mem, SSE_PHADDSUBD>;
5335 defm PHSUBW : SS3I_binop_rm<0x05, "phsubw", X86hsub, v8i16, VR128,
5336 memopv2i64, i128mem, SSE_PHADDSUBW>;
5337 defm PHSUBD : SS3I_binop_rm<0x06, "phsubd", X86hsub, v4i32, VR128,
5338 memopv2i64, i128mem, SSE_PHADDSUBD>;
5339 defm PSIGNB : SS3I_binop_rm<0x08, "psignb", X86psign, v16i8, VR128,
5340 memopv2i64, i128mem, SSE_PSIGN>;
5341 defm PSIGNW : SS3I_binop_rm<0x09, "psignw", X86psign, v8i16, VR128,
5342 memopv2i64, i128mem, SSE_PSIGN>;
5343 defm PSIGND : SS3I_binop_rm<0x0A, "psignd", X86psign, v4i32, VR128,
5344 memopv2i64, i128mem, SSE_PSIGN>;
5345 defm PSHUFB : SS3I_binop_rm<0x00, "pshufb", X86pshufb, v16i8, VR128,
5346 memopv2i64, i128mem, SSE_PSHUFB>;
5347 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw",
5348 int_x86_ssse3_phadd_sw_128,
5350 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw",
5351 int_x86_ssse3_phsub_sw_128,
5353 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw",
5354 int_x86_ssse3_pmadd_ub_sw_128, SSE_PMADD>;
5356 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw",
5357 int_x86_ssse3_pmul_hr_sw_128,
5361 //===---------------------------------------------------------------------===//
5362 // SSSE3 - Packed Align Instruction Patterns
5363 //===---------------------------------------------------------------------===//
5365 multiclass ssse3_palignr<string asm, bit Is2Addr = 1> {
5366 let neverHasSideEffects = 1 in {
5367 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
5368 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5370 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5372 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5373 [], IIC_SSE_PALIGNRR>, OpSize, Sched<[WriteShuffle]>;
5375 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
5376 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5378 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5380 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5381 [], IIC_SSE_PALIGNRM>, OpSize, Sched<[WriteShuffleLd, ReadAfterLd]>;
5385 multiclass ssse3_palignr_y<string asm, bit Is2Addr = 1> {
5386 let neverHasSideEffects = 1 in {
5387 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
5388 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5390 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5391 []>, OpSize, Sched<[WriteShuffle]>;
5393 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
5394 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
5396 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5397 []>, OpSize, Sched<[WriteShuffleLd, ReadAfterLd]>;
5401 let Predicates = [HasAVX] in
5402 defm VPALIGN : ssse3_palignr<"vpalignr", 0>, VEX_4V;
5403 let Predicates = [HasAVX2] in
5404 defm VPALIGN : ssse3_palignr_y<"vpalignr", 0>, VEX_4V, VEX_L;
5405 let Constraints = "$src1 = $dst", Predicates = [UseSSSE3] in
5406 defm PALIGN : ssse3_palignr<"palignr">;
5408 let Predicates = [HasAVX2] in {
5409 def : Pat<(v8i32 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5410 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5411 def : Pat<(v8f32 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5412 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5413 def : Pat<(v16i16 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5414 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5415 def : Pat<(v32i8 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5416 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5419 let Predicates = [HasAVX] in {
5420 def : Pat<(v4i32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5421 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5422 def : Pat<(v4f32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5423 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5424 def : Pat<(v8i16 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5425 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5426 def : Pat<(v16i8 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5427 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5430 let Predicates = [UseSSSE3] in {
5431 def : Pat<(v4i32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5432 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5433 def : Pat<(v4f32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5434 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5435 def : Pat<(v8i16 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5436 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5437 def : Pat<(v16i8 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5438 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5441 //===---------------------------------------------------------------------===//
5442 // SSSE3 - Thread synchronization
5443 //===---------------------------------------------------------------------===//
5445 let SchedRW = [WriteSystem] in {
5446 let usesCustomInserter = 1 in {
5447 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5448 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>,
5449 Requires<[HasSSE3]>;
5452 let Uses = [EAX, ECX, EDX] in
5453 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", [], IIC_SSE_MONITOR>,
5454 TB, Requires<[HasSSE3]>;
5455 let Uses = [ECX, EAX] in
5456 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait",
5457 [(int_x86_sse3_mwait ECX, EAX)], IIC_SSE_MWAIT>,
5458 TB, Requires<[HasSSE3]>;
5461 def : InstAlias<"mwait\t{%eax, %ecx|ecx, eax}", (MWAITrr)>, Requires<[In32BitMode]>;
5462 def : InstAlias<"mwait\t{%rax, %rcx|rcx, rax}", (MWAITrr)>, Requires<[In64BitMode]>;
5464 def : InstAlias<"monitor\t{%eax, %ecx, %edx|edx, ecx, eax}", (MONITORrrr)>,
5465 Requires<[In32BitMode]>;
5466 def : InstAlias<"monitor\t{%rax, %rcx, %rdx|rdx, rcx, rax}", (MONITORrrr)>,
5467 Requires<[In64BitMode]>;
5469 //===----------------------------------------------------------------------===//
5470 // SSE4.1 - Packed Move with Sign/Zero Extend
5471 //===----------------------------------------------------------------------===//
5473 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId,
5474 OpndItins itins = DEFAULT_ITINS> {
5475 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5476 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5477 [(set VR128:$dst, (IntId VR128:$src))], itins.rr>, OpSize;
5479 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5480 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5482 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))],
5486 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5488 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5489 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5490 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5492 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5493 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5494 [(set VR256:$dst, (IntId (load addr:$src)))]>,
5498 let Predicates = [HasAVX] in {
5499 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw",
5500 int_x86_sse41_pmovsxbw>, VEX;
5501 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd",
5502 int_x86_sse41_pmovsxwd>, VEX;
5503 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq",
5504 int_x86_sse41_pmovsxdq>, VEX;
5505 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw",
5506 int_x86_sse41_pmovzxbw>, VEX;
5507 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd",
5508 int_x86_sse41_pmovzxwd>, VEX;
5509 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq",
5510 int_x86_sse41_pmovzxdq>, VEX;
5513 let Predicates = [HasAVX2] in {
5514 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5515 int_x86_avx2_pmovsxbw>, VEX, VEX_L;
5516 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5517 int_x86_avx2_pmovsxwd>, VEX, VEX_L;
5518 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5519 int_x86_avx2_pmovsxdq>, VEX, VEX_L;
5520 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5521 int_x86_avx2_pmovzxbw>, VEX, VEX_L;
5522 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5523 int_x86_avx2_pmovzxwd>, VEX, VEX_L;
5524 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5525 int_x86_avx2_pmovzxdq>, VEX, VEX_L;
5528 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw, SSE_INTALU_ITINS_P>;
5529 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd, SSE_INTALU_ITINS_P>;
5530 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq, SSE_INTALU_ITINS_P>;
5531 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw, SSE_INTALU_ITINS_P>;
5532 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd, SSE_INTALU_ITINS_P>;
5533 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq, SSE_INTALU_ITINS_P>;
5535 let Predicates = [HasAVX] in {
5536 // Common patterns involving scalar load.
5537 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5538 (VPMOVSXBWrm addr:$src)>;
5539 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5540 (VPMOVSXBWrm addr:$src)>;
5541 def : Pat<(int_x86_sse41_pmovsxbw (bc_v16i8 (loadv2i64 addr:$src))),
5542 (VPMOVSXBWrm addr:$src)>;
5544 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5545 (VPMOVSXWDrm addr:$src)>;
5546 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5547 (VPMOVSXWDrm addr:$src)>;
5548 def : Pat<(int_x86_sse41_pmovsxwd (bc_v8i16 (loadv2i64 addr:$src))),
5549 (VPMOVSXWDrm addr:$src)>;
5551 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5552 (VPMOVSXDQrm addr:$src)>;
5553 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5554 (VPMOVSXDQrm addr:$src)>;
5555 def : Pat<(int_x86_sse41_pmovsxdq (bc_v4i32 (loadv2i64 addr:$src))),
5556 (VPMOVSXDQrm addr:$src)>;
5558 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5559 (VPMOVZXBWrm addr:$src)>;
5560 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5561 (VPMOVZXBWrm addr:$src)>;
5562 def : Pat<(int_x86_sse41_pmovzxbw (bc_v16i8 (loadv2i64 addr:$src))),
5563 (VPMOVZXBWrm addr:$src)>;
5565 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5566 (VPMOVZXWDrm addr:$src)>;
5567 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5568 (VPMOVZXWDrm addr:$src)>;
5569 def : Pat<(int_x86_sse41_pmovzxwd (bc_v8i16 (loadv2i64 addr:$src))),
5570 (VPMOVZXWDrm addr:$src)>;
5572 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5573 (VPMOVZXDQrm addr:$src)>;
5574 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5575 (VPMOVZXDQrm addr:$src)>;
5576 def : Pat<(int_x86_sse41_pmovzxdq (bc_v4i32 (loadv2i64 addr:$src))),
5577 (VPMOVZXDQrm addr:$src)>;
5580 let Predicates = [UseSSE41] in {
5581 // Common patterns involving scalar load.
5582 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5583 (PMOVSXBWrm addr:$src)>;
5584 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5585 (PMOVSXBWrm addr:$src)>;
5586 def : Pat<(int_x86_sse41_pmovsxbw (bc_v16i8 (loadv2i64 addr:$src))),
5587 (PMOVSXBWrm addr:$src)>;
5589 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5590 (PMOVSXWDrm addr:$src)>;
5591 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5592 (PMOVSXWDrm addr:$src)>;
5593 def : Pat<(int_x86_sse41_pmovsxwd (bc_v8i16 (loadv2i64 addr:$src))),
5594 (PMOVSXWDrm addr:$src)>;
5596 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5597 (PMOVSXDQrm addr:$src)>;
5598 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5599 (PMOVSXDQrm addr:$src)>;
5600 def : Pat<(int_x86_sse41_pmovsxdq (bc_v4i32 (loadv2i64 addr:$src))),
5601 (PMOVSXDQrm addr:$src)>;
5603 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5604 (PMOVZXBWrm addr:$src)>;
5605 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5606 (PMOVZXBWrm addr:$src)>;
5607 def : Pat<(int_x86_sse41_pmovzxbw (bc_v16i8 (loadv2i64 addr:$src))),
5608 (PMOVZXBWrm addr:$src)>;
5610 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5611 (PMOVZXWDrm addr:$src)>;
5612 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5613 (PMOVZXWDrm addr:$src)>;
5614 def : Pat<(int_x86_sse41_pmovzxwd (bc_v8i16 (loadv2i64 addr:$src))),
5615 (PMOVZXWDrm addr:$src)>;
5617 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5618 (PMOVZXDQrm addr:$src)>;
5619 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5620 (PMOVZXDQrm addr:$src)>;
5621 def : Pat<(int_x86_sse41_pmovzxdq (bc_v4i32 (loadv2i64 addr:$src))),
5622 (PMOVZXDQrm addr:$src)>;
5625 let Predicates = [HasAVX2] in {
5626 let AddedComplexity = 15 in {
5627 def : Pat<(v4i64 (X86vzmovly (v4i32 VR128:$src))),
5628 (VPMOVZXDQYrr VR128:$src)>;
5629 def : Pat<(v8i32 (X86vzmovly (v8i16 VR128:$src))),
5630 (VPMOVZXWDYrr VR128:$src)>;
5633 def : Pat<(v4i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>;
5634 def : Pat<(v8i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>;
5637 let Predicates = [HasAVX] in {
5638 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
5639 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
5642 let Predicates = [UseSSE41] in {
5643 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
5644 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
5648 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId,
5649 OpndItins itins = DEFAULT_ITINS> {
5650 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5651 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5652 [(set VR128:$dst, (IntId VR128:$src))], itins.rr>, OpSize;
5654 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5655 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5657 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))],
5662 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5664 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5665 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5666 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5668 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5669 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5671 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5675 let Predicates = [HasAVX] in {
5676 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5678 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5680 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5682 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5686 let Predicates = [HasAVX2] in {
5687 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5688 int_x86_avx2_pmovsxbd>, VEX, VEX_L;
5689 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5690 int_x86_avx2_pmovsxwq>, VEX, VEX_L;
5691 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5692 int_x86_avx2_pmovzxbd>, VEX, VEX_L;
5693 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5694 int_x86_avx2_pmovzxwq>, VEX, VEX_L;
5697 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd,
5698 SSE_INTALU_ITINS_P>;
5699 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq,
5700 SSE_INTALU_ITINS_P>;
5701 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd,
5702 SSE_INTALU_ITINS_P>;
5703 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq,
5704 SSE_INTALU_ITINS_P>;
5706 let Predicates = [HasAVX] in {
5707 // Common patterns involving scalar load
5708 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5709 (VPMOVSXBDrm addr:$src)>;
5710 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5711 (VPMOVSXWQrm addr:$src)>;
5713 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5714 (VPMOVZXBDrm addr:$src)>;
5715 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5716 (VPMOVZXWQrm addr:$src)>;
5719 let Predicates = [UseSSE41] in {
5720 // Common patterns involving scalar load
5721 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5722 (PMOVSXBDrm addr:$src)>;
5723 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5724 (PMOVSXWQrm addr:$src)>;
5726 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5727 (PMOVZXBDrm addr:$src)>;
5728 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5729 (PMOVZXWQrm addr:$src)>;
5732 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId,
5733 OpndItins itins = DEFAULT_ITINS> {
5734 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5735 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5736 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5738 // Expecting a i16 load any extended to i32 value.
5739 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5740 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5741 [(set VR128:$dst, (IntId (bitconvert
5742 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5746 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5748 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5749 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5750 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5752 // Expecting a i16 load any extended to i32 value.
5753 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5754 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5755 [(set VR256:$dst, (IntId (bitconvert
5756 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5760 let Predicates = [HasAVX] in {
5761 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5763 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5766 let Predicates = [HasAVX2] in {
5767 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5768 int_x86_avx2_pmovsxbq>, VEX, VEX_L;
5769 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5770 int_x86_avx2_pmovzxbq>, VEX, VEX_L;
5772 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq,
5773 SSE_INTALU_ITINS_P>;
5774 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq,
5775 SSE_INTALU_ITINS_P>;
5777 let Predicates = [HasAVX2] in {
5778 def : Pat<(v16i16 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBWYrr VR128:$src)>;
5779 def : Pat<(v8i32 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBDYrr VR128:$src)>;
5780 def : Pat<(v4i64 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBQYrr VR128:$src)>;
5782 def : Pat<(v8i32 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>;
5783 def : Pat<(v4i64 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWQYrr VR128:$src)>;
5785 def : Pat<(v4i64 (X86vsext (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>;
5787 def : Pat<(v16i16 (X86vsext (v32i8 VR256:$src))),
5788 (VPMOVSXBWYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5789 def : Pat<(v8i32 (X86vsext (v32i8 VR256:$src))),
5790 (VPMOVSXBDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5791 def : Pat<(v4i64 (X86vsext (v32i8 VR256:$src))),
5792 (VPMOVSXBQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5794 def : Pat<(v8i32 (X86vsext (v16i16 VR256:$src))),
5795 (VPMOVSXWDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5796 def : Pat<(v4i64 (X86vsext (v16i16 VR256:$src))),
5797 (VPMOVSXWQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5799 def : Pat<(v4i64 (X86vsext (v8i32 VR256:$src))),
5800 (VPMOVSXDQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5802 def : Pat<(v8i32 (X86vsmovl (v8i16 (bitconvert (v2i64 (load addr:$src)))))),
5803 (VPMOVSXWDYrm addr:$src)>;
5804 def : Pat<(v4i64 (X86vsmovl (v4i32 (bitconvert (v2i64 (load addr:$src)))))),
5805 (VPMOVSXDQYrm addr:$src)>;
5807 def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2i64
5808 (scalar_to_vector (loadi64 addr:$src))))))),
5809 (VPMOVSXBDYrm addr:$src)>;
5810 def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2f64
5811 (scalar_to_vector (loadf64 addr:$src))))))),
5812 (VPMOVSXBDYrm addr:$src)>;
5814 def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2i64
5815 (scalar_to_vector (loadi64 addr:$src))))))),
5816 (VPMOVSXWQYrm addr:$src)>;
5817 def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2f64
5818 (scalar_to_vector (loadf64 addr:$src))))))),
5819 (VPMOVSXWQYrm addr:$src)>;
5821 def : Pat<(v4i64 (X86vsext (v16i8 (bitconvert (v4i32
5822 (scalar_to_vector (loadi32 addr:$src))))))),
5823 (VPMOVSXBQYrm addr:$src)>;
5826 let Predicates = [HasAVX] in {
5827 // Common patterns involving scalar load
5828 def : Pat<(int_x86_sse41_pmovsxbq
5829 (bitconvert (v4i32 (X86vzmovl
5830 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5831 (VPMOVSXBQrm addr:$src)>;
5833 def : Pat<(int_x86_sse41_pmovzxbq
5834 (bitconvert (v4i32 (X86vzmovl
5835 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5836 (VPMOVZXBQrm addr:$src)>;
5839 let Predicates = [UseSSE41] in {
5840 def : Pat<(v8i16 (X86vsext (v16i8 VR128:$src))), (PMOVSXBWrr VR128:$src)>;
5841 def : Pat<(v4i32 (X86vsext (v16i8 VR128:$src))), (PMOVSXBDrr VR128:$src)>;
5842 def : Pat<(v2i64 (X86vsext (v16i8 VR128:$src))), (PMOVSXBQrr VR128:$src)>;
5844 def : Pat<(v4i32 (X86vsext (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
5845 def : Pat<(v2i64 (X86vsext (v8i16 VR128:$src))), (PMOVSXWQrr VR128:$src)>;
5847 def : Pat<(v2i64 (X86vsext (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
5849 // Common patterns involving scalar load
5850 def : Pat<(int_x86_sse41_pmovsxbq
5851 (bitconvert (v4i32 (X86vzmovl
5852 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5853 (PMOVSXBQrm addr:$src)>;
5855 def : Pat<(int_x86_sse41_pmovzxbq
5856 (bitconvert (v4i32 (X86vzmovl
5857 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5858 (PMOVZXBQrm addr:$src)>;
5860 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64
5861 (scalar_to_vector (loadi64 addr:$src))))))),
5862 (PMOVSXWDrm addr:$src)>;
5863 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64
5864 (scalar_to_vector (loadf64 addr:$src))))))),
5865 (PMOVSXWDrm addr:$src)>;
5866 def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32
5867 (scalar_to_vector (loadi32 addr:$src))))))),
5868 (PMOVSXBDrm addr:$src)>;
5869 def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32
5870 (scalar_to_vector (loadi32 addr:$src))))))),
5871 (PMOVSXWQrm addr:$src)>;
5872 def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32
5873 (scalar_to_vector (extloadi32i16 addr:$src))))))),
5874 (PMOVSXBQrm addr:$src)>;
5875 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64
5876 (scalar_to_vector (loadi64 addr:$src))))))),
5877 (PMOVSXDQrm addr:$src)>;
5878 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64
5879 (scalar_to_vector (loadf64 addr:$src))))))),
5880 (PMOVSXDQrm addr:$src)>;
5881 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64
5882 (scalar_to_vector (loadi64 addr:$src))))))),
5883 (PMOVSXBWrm addr:$src)>;
5884 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64
5885 (scalar_to_vector (loadf64 addr:$src))))))),
5886 (PMOVSXBWrm addr:$src)>;
5889 let Predicates = [HasAVX2] in {
5890 def : Pat<(v16i16 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBWYrr VR128:$src)>;
5891 def : Pat<(v8i32 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBDYrr VR128:$src)>;
5892 def : Pat<(v4i64 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBQYrr VR128:$src)>;
5894 def : Pat<(v8i32 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWDYrr VR128:$src)>;
5895 def : Pat<(v4i64 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWQYrr VR128:$src)>;
5897 def : Pat<(v4i64 (X86vzext (v4i32 VR128:$src))), (VPMOVZXDQYrr VR128:$src)>;
5899 def : Pat<(v16i16 (X86vzext (v32i8 VR256:$src))),
5900 (VPMOVZXBWYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5901 def : Pat<(v8i32 (X86vzext (v32i8 VR256:$src))),
5902 (VPMOVZXBDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5903 def : Pat<(v4i64 (X86vzext (v32i8 VR256:$src))),
5904 (VPMOVZXBQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5906 def : Pat<(v8i32 (X86vzext (v16i16 VR256:$src))),
5907 (VPMOVZXWDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5908 def : Pat<(v4i64 (X86vzext (v16i16 VR256:$src))),
5909 (VPMOVZXWQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5911 def : Pat<(v4i64 (X86vzext (v8i32 VR256:$src))),
5912 (VPMOVZXDQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5915 let Predicates = [HasAVX] in {
5916 def : Pat<(v8i16 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBWrr VR128:$src)>;
5917 def : Pat<(v4i32 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBDrr VR128:$src)>;
5918 def : Pat<(v2i64 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBQrr VR128:$src)>;
5920 def : Pat<(v4i32 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWDrr VR128:$src)>;
5921 def : Pat<(v2i64 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWQrr VR128:$src)>;
5923 def : Pat<(v2i64 (X86vzext (v4i32 VR128:$src))), (VPMOVZXDQrr VR128:$src)>;
5925 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5926 (VPMOVZXBWrm addr:$src)>;
5927 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5928 (VPMOVZXBWrm addr:$src)>;
5929 def : Pat<(v4i32 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5930 (VPMOVZXBDrm addr:$src)>;
5931 def : Pat<(v2i64 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))),
5932 (VPMOVZXBQrm addr:$src)>;
5934 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5935 (VPMOVZXWDrm addr:$src)>;
5936 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5937 (VPMOVZXWDrm addr:$src)>;
5938 def : Pat<(v2i64 (X86vzext (v8i16 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5939 (VPMOVZXWQrm addr:$src)>;
5941 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5942 (VPMOVZXDQrm addr:$src)>;
5943 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5944 (VPMOVZXDQrm addr:$src)>;
5945 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (X86vzload addr:$src)))))),
5946 (VPMOVZXDQrm addr:$src)>;
5948 def : Pat<(v8i16 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBWrr VR128:$src)>;
5949 def : Pat<(v4i32 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBDrr VR128:$src)>;
5950 def : Pat<(v2i64 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBQrr VR128:$src)>;
5952 def : Pat<(v4i32 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
5953 def : Pat<(v2i64 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWQrr VR128:$src)>;
5955 def : Pat<(v2i64 (X86vsext (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
5957 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64
5958 (scalar_to_vector (loadi64 addr:$src))))))),
5959 (VPMOVSXWDrm addr:$src)>;
5960 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64
5961 (scalar_to_vector (loadi64 addr:$src))))))),
5962 (VPMOVSXDQrm addr:$src)>;
5963 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64
5964 (scalar_to_vector (loadf64 addr:$src))))))),
5965 (VPMOVSXWDrm addr:$src)>;
5966 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64
5967 (scalar_to_vector (loadf64 addr:$src))))))),
5968 (VPMOVSXDQrm addr:$src)>;
5969 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64
5970 (scalar_to_vector (loadi64 addr:$src))))))),
5971 (VPMOVSXBWrm addr:$src)>;
5972 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64
5973 (scalar_to_vector (loadf64 addr:$src))))))),
5974 (VPMOVSXBWrm addr:$src)>;
5976 def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32
5977 (scalar_to_vector (loadi32 addr:$src))))))),
5978 (VPMOVSXBDrm addr:$src)>;
5979 def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32
5980 (scalar_to_vector (loadi32 addr:$src))))))),
5981 (VPMOVSXWQrm addr:$src)>;
5982 def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32
5983 (scalar_to_vector (extloadi32i16 addr:$src))))))),
5984 (VPMOVSXBQrm addr:$src)>;
5987 let Predicates = [UseSSE41] in {
5988 def : Pat<(v8i16 (X86vzext (v16i8 VR128:$src))), (PMOVZXBWrr VR128:$src)>;
5989 def : Pat<(v4i32 (X86vzext (v16i8 VR128:$src))), (PMOVZXBDrr VR128:$src)>;
5990 def : Pat<(v2i64 (X86vzext (v16i8 VR128:$src))), (PMOVZXBQrr VR128:$src)>;
5992 def : Pat<(v4i32 (X86vzext (v8i16 VR128:$src))), (PMOVZXWDrr VR128:$src)>;
5993 def : Pat<(v2i64 (X86vzext (v8i16 VR128:$src))), (PMOVZXWQrr VR128:$src)>;
5995 def : Pat<(v2i64 (X86vzext (v4i32 VR128:$src))), (PMOVZXDQrr VR128:$src)>;
5997 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5998 (PMOVZXBWrm addr:$src)>;
5999 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
6000 (PMOVZXBWrm addr:$src)>;
6001 def : Pat<(v4i32 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
6002 (PMOVZXBDrm addr:$src)>;
6003 def : Pat<(v2i64 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))),
6004 (PMOVZXBQrm addr:$src)>;
6006 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
6007 (PMOVZXWDrm addr:$src)>;
6008 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
6009 (PMOVZXWDrm addr:$src)>;
6010 def : Pat<(v2i64 (X86vzext (v8i16 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
6011 (PMOVZXWQrm addr:$src)>;
6013 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
6014 (PMOVZXDQrm addr:$src)>;
6015 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
6016 (PMOVZXDQrm addr:$src)>;
6017 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (X86vzload addr:$src)))))),
6018 (PMOVZXDQrm addr:$src)>;
6021 //===----------------------------------------------------------------------===//
6022 // SSE4.1 - Extract Instructions
6023 //===----------------------------------------------------------------------===//
6025 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
6026 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
6027 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
6028 (ins VR128:$src1, i32i8imm:$src2),
6029 !strconcat(OpcodeStr,
6030 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6031 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
6033 let neverHasSideEffects = 1, mayStore = 1 in
6034 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6035 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
6036 !strconcat(OpcodeStr,
6037 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6040 // There's an AssertZext in the way of writing the store pattern
6041 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
6044 let Predicates = [HasAVX] in {
6045 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
6046 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
6047 (ins VR128:$src1, i32i8imm:$src2),
6048 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
6051 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
6054 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
6055 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
6056 let isCodeGenOnly = 1, hasSideEffects = 0 in
6057 def rr_REV : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
6058 (ins VR128:$src1, i32i8imm:$src2),
6059 !strconcat(OpcodeStr,
6060 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6063 let neverHasSideEffects = 1, mayStore = 1 in
6064 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6065 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
6066 !strconcat(OpcodeStr,
6067 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6070 // There's an AssertZext in the way of writing the store pattern
6071 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
6074 let Predicates = [HasAVX] in
6075 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
6077 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
6080 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
6081 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
6082 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
6083 (ins VR128:$src1, i32i8imm:$src2),
6084 !strconcat(OpcodeStr,
6085 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6087 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
6088 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6089 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
6090 !strconcat(OpcodeStr,
6091 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6092 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
6093 addr:$dst)]>, OpSize;
6096 let Predicates = [HasAVX] in
6097 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
6099 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
6101 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
6102 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
6103 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
6104 (ins VR128:$src1, i32i8imm:$src2),
6105 !strconcat(OpcodeStr,
6106 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6108 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
6109 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6110 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
6111 !strconcat(OpcodeStr,
6112 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6113 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
6114 addr:$dst)]>, OpSize, REX_W;
6117 let Predicates = [HasAVX] in
6118 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
6120 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
6122 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
6124 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr,
6125 OpndItins itins = DEFAULT_ITINS> {
6126 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
6127 (ins VR128:$src1, i32i8imm:$src2),
6128 !strconcat(OpcodeStr,
6129 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6131 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))],
6134 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6135 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
6136 !strconcat(OpcodeStr,
6137 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6138 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
6139 addr:$dst)], itins.rm>, OpSize;
6142 let ExeDomain = SSEPackedSingle in {
6143 let Predicates = [UseAVX] in {
6144 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
6145 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
6146 (ins VR128:$src1, i32i8imm:$src2),
6147 "vextractps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6150 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps", SSE_EXTRACT_ITINS>;
6153 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
6154 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
6157 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
6159 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
6162 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
6163 Requires<[UseSSE41]>;
6165 //===----------------------------------------------------------------------===//
6166 // SSE4.1 - Insert Instructions
6167 //===----------------------------------------------------------------------===//
6169 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
6170 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6171 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
6173 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6175 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6177 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
6178 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6179 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
6181 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6183 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6185 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
6186 imm:$src3))]>, OpSize;
6189 let Predicates = [HasAVX] in
6190 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
6191 let Constraints = "$src1 = $dst" in
6192 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
6194 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
6195 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6196 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
6198 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6200 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6202 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
6204 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6205 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
6207 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6209 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6211 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
6212 imm:$src3)))]>, OpSize;
6215 let Predicates = [HasAVX] in
6216 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
6217 let Constraints = "$src1 = $dst" in
6218 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
6220 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
6221 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6222 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
6224 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6226 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6228 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
6230 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6231 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
6233 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6235 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6237 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
6238 imm:$src3)))]>, OpSize;
6241 let Predicates = [HasAVX] in
6242 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
6243 let Constraints = "$src1 = $dst" in
6244 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
6246 // insertps has a few different modes, there's the first two here below which
6247 // are optimized inserts that won't zero arbitrary elements in the destination
6248 // vector. The next one matches the intrinsic and could zero arbitrary elements
6249 // in the target vector.
6250 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1,
6251 OpndItins itins = DEFAULT_ITINS> {
6252 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6253 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
6255 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6257 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6259 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))], itins.rr>,
6261 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6262 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
6264 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6266 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6268 (X86insrtps VR128:$src1,
6269 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
6270 imm:$src3))], itins.rm>, OpSize;
6273 let ExeDomain = SSEPackedSingle in {
6274 let Predicates = [UseAVX] in
6275 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
6276 let Constraints = "$src1 = $dst" in
6277 defm INSERTPS : SS41I_insertf32<0x21, "insertps", 1, SSE_INSERT_ITINS>;
6280 //===----------------------------------------------------------------------===//
6281 // SSE4.1 - Round Instructions
6282 //===----------------------------------------------------------------------===//
6284 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
6285 X86MemOperand x86memop, RegisterClass RC,
6286 PatFrag mem_frag32, PatFrag mem_frag64,
6287 Intrinsic V4F32Int, Intrinsic V2F64Int> {
6288 let ExeDomain = SSEPackedSingle in {
6289 // Intrinsic operation, reg.
6290 // Vector intrinsic operation, reg
6291 def PSr : SS4AIi8<opcps, MRMSrcReg,
6292 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6293 !strconcat(OpcodeStr,
6294 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6295 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))],
6296 IIC_SSE_ROUNDPS_REG>,
6299 // Vector intrinsic operation, mem
6300 def PSm : SS4AIi8<opcps, MRMSrcMem,
6301 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6302 !strconcat(OpcodeStr,
6303 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6305 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))],
6306 IIC_SSE_ROUNDPS_MEM>,
6308 } // ExeDomain = SSEPackedSingle
6310 let ExeDomain = SSEPackedDouble in {
6311 // Vector intrinsic operation, reg
6312 def PDr : SS4AIi8<opcpd, MRMSrcReg,
6313 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6314 !strconcat(OpcodeStr,
6315 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6316 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))],
6317 IIC_SSE_ROUNDPS_REG>,
6320 // Vector intrinsic operation, mem
6321 def PDm : SS4AIi8<opcpd, MRMSrcMem,
6322 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6323 !strconcat(OpcodeStr,
6324 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6326 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))],
6327 IIC_SSE_ROUNDPS_REG>,
6329 } // ExeDomain = SSEPackedDouble
6332 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
6335 Intrinsic F64Int, bit Is2Addr = 1> {
6336 let ExeDomain = GenericDomain in {
6338 let hasSideEffects = 0 in
6339 def SSr : SS4AIi8<opcss, MRMSrcReg,
6340 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3),
6342 !strconcat(OpcodeStr,
6343 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6344 !strconcat(OpcodeStr,
6345 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6348 // Intrinsic operation, reg.
6349 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
6350 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6352 !strconcat(OpcodeStr,
6353 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6354 !strconcat(OpcodeStr,
6355 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6356 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6359 // Intrinsic operation, mem.
6360 def SSm : SS4AIi8<opcss, MRMSrcMem,
6361 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
6363 !strconcat(OpcodeStr,
6364 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6365 !strconcat(OpcodeStr,
6366 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6368 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
6372 let hasSideEffects = 0 in
6373 def SDr : SS4AIi8<opcsd, MRMSrcReg,
6374 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3),
6376 !strconcat(OpcodeStr,
6377 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6378 !strconcat(OpcodeStr,
6379 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6382 // Intrinsic operation, reg.
6383 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
6384 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6386 !strconcat(OpcodeStr,
6387 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6388 !strconcat(OpcodeStr,
6389 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6390 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6393 // Intrinsic operation, mem.
6394 def SDm : SS4AIi8<opcsd, MRMSrcMem,
6395 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
6397 !strconcat(OpcodeStr,
6398 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6399 !strconcat(OpcodeStr,
6400 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6402 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
6404 } // ExeDomain = GenericDomain
6407 // FP round - roundss, roundps, roundsd, roundpd
6408 let Predicates = [HasAVX] in {
6410 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
6411 memopv4f32, memopv2f64,
6412 int_x86_sse41_round_ps,
6413 int_x86_sse41_round_pd>, VEX;
6414 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
6415 memopv8f32, memopv4f64,
6416 int_x86_avx_round_ps_256,
6417 int_x86_avx_round_pd_256>, VEX, VEX_L;
6418 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
6419 int_x86_sse41_round_ss,
6420 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
6422 def : Pat<(ffloor FR32:$src),
6423 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6424 def : Pat<(f64 (ffloor FR64:$src)),
6425 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6426 def : Pat<(f32 (fnearbyint FR32:$src)),
6427 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6428 def : Pat<(f64 (fnearbyint FR64:$src)),
6429 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6430 def : Pat<(f32 (fceil FR32:$src)),
6431 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6432 def : Pat<(f64 (fceil FR64:$src)),
6433 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6434 def : Pat<(f32 (frint FR32:$src)),
6435 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6436 def : Pat<(f64 (frint FR64:$src)),
6437 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6438 def : Pat<(f32 (ftrunc FR32:$src)),
6439 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6440 def : Pat<(f64 (ftrunc FR64:$src)),
6441 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6443 def : Pat<(v4f32 (ffloor VR128:$src)),
6444 (VROUNDPSr VR128:$src, (i32 0x1))>;
6445 def : Pat<(v4f32 (fnearbyint VR128:$src)),
6446 (VROUNDPSr VR128:$src, (i32 0xC))>;
6447 def : Pat<(v4f32 (fceil VR128:$src)),
6448 (VROUNDPSr VR128:$src, (i32 0x2))>;
6449 def : Pat<(v4f32 (frint VR128:$src)),
6450 (VROUNDPSr VR128:$src, (i32 0x4))>;
6451 def : Pat<(v4f32 (ftrunc VR128:$src)),
6452 (VROUNDPSr VR128:$src, (i32 0x3))>;
6454 def : Pat<(v2f64 (ffloor VR128:$src)),
6455 (VROUNDPDr VR128:$src, (i32 0x1))>;
6456 def : Pat<(v2f64 (fnearbyint VR128:$src)),
6457 (VROUNDPDr VR128:$src, (i32 0xC))>;
6458 def : Pat<(v2f64 (fceil VR128:$src)),
6459 (VROUNDPDr VR128:$src, (i32 0x2))>;
6460 def : Pat<(v2f64 (frint VR128:$src)),
6461 (VROUNDPDr VR128:$src, (i32 0x4))>;
6462 def : Pat<(v2f64 (ftrunc VR128:$src)),
6463 (VROUNDPDr VR128:$src, (i32 0x3))>;
6465 def : Pat<(v8f32 (ffloor VR256:$src)),
6466 (VROUNDYPSr VR256:$src, (i32 0x1))>;
6467 def : Pat<(v8f32 (fnearbyint VR256:$src)),
6468 (VROUNDYPSr VR256:$src, (i32 0xC))>;
6469 def : Pat<(v8f32 (fceil VR256:$src)),
6470 (VROUNDYPSr VR256:$src, (i32 0x2))>;
6471 def : Pat<(v8f32 (frint VR256:$src)),
6472 (VROUNDYPSr VR256:$src, (i32 0x4))>;
6473 def : Pat<(v8f32 (ftrunc VR256:$src)),
6474 (VROUNDYPSr VR256:$src, (i32 0x3))>;
6476 def : Pat<(v4f64 (ffloor VR256:$src)),
6477 (VROUNDYPDr VR256:$src, (i32 0x1))>;
6478 def : Pat<(v4f64 (fnearbyint VR256:$src)),
6479 (VROUNDYPDr VR256:$src, (i32 0xC))>;
6480 def : Pat<(v4f64 (fceil VR256:$src)),
6481 (VROUNDYPDr VR256:$src, (i32 0x2))>;
6482 def : Pat<(v4f64 (frint VR256:$src)),
6483 (VROUNDYPDr VR256:$src, (i32 0x4))>;
6484 def : Pat<(v4f64 (ftrunc VR256:$src)),
6485 (VROUNDYPDr VR256:$src, (i32 0x3))>;
6488 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
6489 memopv4f32, memopv2f64,
6490 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
6491 let Constraints = "$src1 = $dst" in
6492 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
6493 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
6495 let Predicates = [UseSSE41] in {
6496 def : Pat<(ffloor FR32:$src),
6497 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6498 def : Pat<(f64 (ffloor FR64:$src)),
6499 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6500 def : Pat<(f32 (fnearbyint FR32:$src)),
6501 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6502 def : Pat<(f64 (fnearbyint FR64:$src)),
6503 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6504 def : Pat<(f32 (fceil FR32:$src)),
6505 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6506 def : Pat<(f64 (fceil FR64:$src)),
6507 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6508 def : Pat<(f32 (frint FR32:$src)),
6509 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6510 def : Pat<(f64 (frint FR64:$src)),
6511 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6512 def : Pat<(f32 (ftrunc FR32:$src)),
6513 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6514 def : Pat<(f64 (ftrunc FR64:$src)),
6515 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6517 def : Pat<(v4f32 (ffloor VR128:$src)),
6518 (ROUNDPSr VR128:$src, (i32 0x1))>;
6519 def : Pat<(v4f32 (fnearbyint VR128:$src)),
6520 (ROUNDPSr VR128:$src, (i32 0xC))>;
6521 def : Pat<(v4f32 (fceil VR128:$src)),
6522 (ROUNDPSr VR128:$src, (i32 0x2))>;
6523 def : Pat<(v4f32 (frint VR128:$src)),
6524 (ROUNDPSr VR128:$src, (i32 0x4))>;
6525 def : Pat<(v4f32 (ftrunc VR128:$src)),
6526 (ROUNDPSr VR128:$src, (i32 0x3))>;
6528 def : Pat<(v2f64 (ffloor VR128:$src)),
6529 (ROUNDPDr VR128:$src, (i32 0x1))>;
6530 def : Pat<(v2f64 (fnearbyint VR128:$src)),
6531 (ROUNDPDr VR128:$src, (i32 0xC))>;
6532 def : Pat<(v2f64 (fceil VR128:$src)),
6533 (ROUNDPDr VR128:$src, (i32 0x2))>;
6534 def : Pat<(v2f64 (frint VR128:$src)),
6535 (ROUNDPDr VR128:$src, (i32 0x4))>;
6536 def : Pat<(v2f64 (ftrunc VR128:$src)),
6537 (ROUNDPDr VR128:$src, (i32 0x3))>;
6540 //===----------------------------------------------------------------------===//
6541 // SSE4.1 - Packed Bit Test
6542 //===----------------------------------------------------------------------===//
6544 // ptest instruction we'll lower to this in X86ISelLowering primarily from
6545 // the intel intrinsic that corresponds to this.
6546 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6547 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6548 "vptest\t{$src2, $src1|$src1, $src2}",
6549 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6551 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6552 "vptest\t{$src2, $src1|$src1, $src2}",
6553 [(set EFLAGS,(X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6556 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
6557 "vptest\t{$src2, $src1|$src1, $src2}",
6558 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
6560 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
6561 "vptest\t{$src2, $src1|$src1, $src2}",
6562 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
6566 let Defs = [EFLAGS] in {
6567 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6568 "ptest\t{$src2, $src1|$src1, $src2}",
6569 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6571 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6572 "ptest\t{$src2, $src1|$src1, $src2}",
6573 [(set EFLAGS, (X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6577 // The bit test instructions below are AVX only
6578 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
6579 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
6580 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
6581 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6582 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
6583 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
6584 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6585 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
6589 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6590 let ExeDomain = SSEPackedSingle in {
6591 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
6592 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>,
6595 let ExeDomain = SSEPackedDouble in {
6596 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
6597 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>,
6602 //===----------------------------------------------------------------------===//
6603 // SSE4.1 - Misc Instructions
6604 //===----------------------------------------------------------------------===//
6606 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
6607 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
6608 "popcnt{w}\t{$src, $dst|$dst, $src}",
6609 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)],
6612 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
6613 "popcnt{w}\t{$src, $dst|$dst, $src}",
6614 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
6615 (implicit EFLAGS)], IIC_SSE_POPCNT_RM>, OpSize, XS;
6617 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
6618 "popcnt{l}\t{$src, $dst|$dst, $src}",
6619 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)],
6622 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
6623 "popcnt{l}\t{$src, $dst|$dst, $src}",
6624 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
6625 (implicit EFLAGS)], IIC_SSE_POPCNT_RM>, XS;
6627 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
6628 "popcnt{q}\t{$src, $dst|$dst, $src}",
6629 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)],
6632 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
6633 "popcnt{q}\t{$src, $dst|$dst, $src}",
6634 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
6635 (implicit EFLAGS)], IIC_SSE_POPCNT_RM>, XS;
6640 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
6641 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
6642 Intrinsic IntId128> {
6643 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6645 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6646 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
6647 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6649 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6652 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
6655 let Predicates = [HasAVX] in
6656 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
6657 int_x86_sse41_phminposuw>, VEX;
6658 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
6659 int_x86_sse41_phminposuw>;
6661 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6662 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
6663 Intrinsic IntId128, bit Is2Addr = 1,
6664 OpndItins itins = DEFAULT_ITINS> {
6665 let isCommutable = 1 in
6666 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6667 (ins VR128:$src1, VR128:$src2),
6669 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6670 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6671 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))],
6673 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6674 (ins VR128:$src1, i128mem:$src2),
6676 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6677 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6679 (IntId128 VR128:$src1,
6680 (bitconvert (memopv2i64 addr:$src2))))],
6684 /// SS41I_binop_rm_int_y - Simple SSE 4.1 binary operator
6685 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6686 Intrinsic IntId256> {
6687 let isCommutable = 1 in
6688 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6689 (ins VR256:$src1, VR256:$src2),
6690 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6691 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
6692 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6693 (ins VR256:$src1, i256mem:$src2),
6694 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6696 (IntId256 VR256:$src1,
6697 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
6701 /// SS48I_binop_rm - Simple SSE41 binary operator.
6702 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6703 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6704 X86MemOperand x86memop, bit Is2Addr = 1,
6705 OpndItins itins = DEFAULT_ITINS> {
6706 let isCommutable = 1 in
6707 def rr : SS48I<opc, MRMSrcReg, (outs RC:$dst),
6708 (ins RC:$src1, RC:$src2),
6710 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6711 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6712 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>, OpSize;
6713 def rm : SS48I<opc, MRMSrcMem, (outs RC:$dst),
6714 (ins RC:$src1, x86memop:$src2),
6716 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6717 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6719 (OpVT (OpNode RC:$src1,
6720 (bitconvert (memop_frag addr:$src2)))))]>, OpSize;
6723 let Predicates = [HasAVX] in {
6724 let isCommutable = 0 in
6725 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
6727 defm VPMINSB : SS48I_binop_rm<0x38, "vpminsb", X86smin, v16i8, VR128,
6728 memopv2i64, i128mem, 0>, VEX_4V;
6729 defm VPMINSD : SS48I_binop_rm<0x39, "vpminsd", X86smin, v4i32, VR128,
6730 memopv2i64, i128mem, 0>, VEX_4V;
6731 defm VPMINUD : SS48I_binop_rm<0x3B, "vpminud", X86umin, v4i32, VR128,
6732 memopv2i64, i128mem, 0>, VEX_4V;
6733 defm VPMINUW : SS48I_binop_rm<0x3A, "vpminuw", X86umin, v8i16, VR128,
6734 memopv2i64, i128mem, 0>, VEX_4V;
6735 defm VPMAXSB : SS48I_binop_rm<0x3C, "vpmaxsb", X86smax, v16i8, VR128,
6736 memopv2i64, i128mem, 0>, VEX_4V;
6737 defm VPMAXSD : SS48I_binop_rm<0x3D, "vpmaxsd", X86smax, v4i32, VR128,
6738 memopv2i64, i128mem, 0>, VEX_4V;
6739 defm VPMAXUD : SS48I_binop_rm<0x3F, "vpmaxud", X86umax, v4i32, VR128,
6740 memopv2i64, i128mem, 0>, VEX_4V;
6741 defm VPMAXUW : SS48I_binop_rm<0x3E, "vpmaxuw", X86umax, v8i16, VR128,
6742 memopv2i64, i128mem, 0>, VEX_4V;
6743 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
6747 let Predicates = [HasAVX2] in {
6748 let isCommutable = 0 in
6749 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
6750 int_x86_avx2_packusdw>, VEX_4V, VEX_L;
6751 defm VPMINSBY : SS48I_binop_rm<0x38, "vpminsb", X86smin, v32i8, VR256,
6752 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6753 defm VPMINSDY : SS48I_binop_rm<0x39, "vpminsd", X86smin, v8i32, VR256,
6754 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6755 defm VPMINUDY : SS48I_binop_rm<0x3B, "vpminud", X86umin, v8i32, VR256,
6756 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6757 defm VPMINUWY : SS48I_binop_rm<0x3A, "vpminuw", X86umin, v16i16, VR256,
6758 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6759 defm VPMAXSBY : SS48I_binop_rm<0x3C, "vpmaxsb", X86smax, v32i8, VR256,
6760 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6761 defm VPMAXSDY : SS48I_binop_rm<0x3D, "vpmaxsd", X86smax, v8i32, VR256,
6762 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6763 defm VPMAXUDY : SS48I_binop_rm<0x3F, "vpmaxud", X86umax, v8i32, VR256,
6764 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6765 defm VPMAXUWY : SS48I_binop_rm<0x3E, "vpmaxuw", X86umax, v16i16, VR256,
6766 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6767 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
6768 int_x86_avx2_pmul_dq>, VEX_4V, VEX_L;
6771 let Constraints = "$src1 = $dst" in {
6772 let isCommutable = 0 in
6773 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
6774 defm PMINSB : SS48I_binop_rm<0x38, "pminsb", X86smin, v16i8, VR128,
6775 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6776 defm PMINSD : SS48I_binop_rm<0x39, "pminsd", X86smin, v4i32, VR128,
6777 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6778 defm PMINUD : SS48I_binop_rm<0x3B, "pminud", X86umin, v4i32, VR128,
6779 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6780 defm PMINUW : SS48I_binop_rm<0x3A, "pminuw", X86umin, v8i16, VR128,
6781 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6782 defm PMAXSB : SS48I_binop_rm<0x3C, "pmaxsb", X86smax, v16i8, VR128,
6783 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6784 defm PMAXSD : SS48I_binop_rm<0x3D, "pmaxsd", X86smax, v4i32, VR128,
6785 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6786 defm PMAXUD : SS48I_binop_rm<0x3F, "pmaxud", X86umax, v4i32, VR128,
6787 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6788 defm PMAXUW : SS48I_binop_rm<0x3E, "pmaxuw", X86umax, v8i16, VR128,
6789 memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
6790 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq,
6791 1, SSE_INTMUL_ITINS_P>;
6794 let Predicates = [HasAVX] in {
6795 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, VR128,
6796 memopv2i64, i128mem, 0>, VEX_4V;
6797 defm VPCMPEQQ : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v2i64, VR128,
6798 memopv2i64, i128mem, 0>, VEX_4V;
6800 let Predicates = [HasAVX2] in {
6801 defm VPMULLDY : SS48I_binop_rm<0x40, "vpmulld", mul, v8i32, VR256,
6802 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6803 defm VPCMPEQQY : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v4i64, VR256,
6804 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6807 let Constraints = "$src1 = $dst" in {
6808 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32, VR128,
6809 memopv2i64, i128mem, 1, SSE_PMULLD_ITINS>;
6810 defm PCMPEQQ : SS48I_binop_rm<0x29, "pcmpeqq", X86pcmpeq, v2i64, VR128,
6811 memopv2i64, i128mem, 1, SSE_INTALUQ_ITINS_P>;
6814 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
6815 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
6816 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6817 X86MemOperand x86memop, bit Is2Addr = 1,
6818 OpndItins itins = DEFAULT_ITINS> {
6819 let isCommutable = 1 in
6820 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
6821 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6823 !strconcat(OpcodeStr,
6824 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6825 !strconcat(OpcodeStr,
6826 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6827 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))], itins.rr>,
6829 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
6830 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6832 !strconcat(OpcodeStr,
6833 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6834 !strconcat(OpcodeStr,
6835 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6838 (bitconvert (memop_frag addr:$src2)), imm:$src3))], itins.rm>,
6842 let Predicates = [HasAVX] in {
6843 let isCommutable = 0 in {
6844 let ExeDomain = SSEPackedSingle in {
6845 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
6846 VR128, memopv4f32, f128mem, 0>, VEX_4V;
6847 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
6848 int_x86_avx_blend_ps_256, VR256, memopv8f32,
6849 f256mem, 0>, VEX_4V, VEX_L;
6851 let ExeDomain = SSEPackedDouble in {
6852 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
6853 VR128, memopv2f64, f128mem, 0>, VEX_4V;
6854 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
6855 int_x86_avx_blend_pd_256,VR256, memopv4f64,
6856 f256mem, 0>, VEX_4V, VEX_L;
6858 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
6859 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6860 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
6861 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6863 let ExeDomain = SSEPackedSingle in
6864 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
6865 VR128, memopv4f32, f128mem, 0>, VEX_4V;
6866 let ExeDomain = SSEPackedDouble in
6867 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
6868 VR128, memopv2f64, f128mem, 0>, VEX_4V;
6869 let ExeDomain = SSEPackedSingle in
6870 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
6871 VR256, memopv8f32, i256mem, 0>, VEX_4V, VEX_L;
6874 let Predicates = [HasAVX2] in {
6875 let isCommutable = 0 in {
6876 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
6877 VR256, memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6878 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
6879 VR256, memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6883 let Constraints = "$src1 = $dst" in {
6884 let isCommutable = 0 in {
6885 let ExeDomain = SSEPackedSingle in
6886 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
6887 VR128, memopv4f32, f128mem,
6888 1, SSE_INTALU_ITINS_P>;
6889 let ExeDomain = SSEPackedDouble in
6890 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
6891 VR128, memopv2f64, f128mem,
6892 1, SSE_INTALU_ITINS_P>;
6893 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
6894 VR128, memopv2i64, i128mem,
6895 1, SSE_INTALU_ITINS_P>;
6896 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
6897 VR128, memopv2i64, i128mem,
6898 1, SSE_INTMUL_ITINS_P>;
6900 let ExeDomain = SSEPackedSingle in
6901 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
6902 VR128, memopv4f32, f128mem, 1,
6904 let ExeDomain = SSEPackedDouble in
6905 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
6906 VR128, memopv2f64, f128mem, 1,
6910 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
6911 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
6912 RegisterClass RC, X86MemOperand x86memop,
6913 PatFrag mem_frag, Intrinsic IntId> {
6914 def rr : Ii8<opc, MRMSrcReg, (outs RC:$dst),
6915 (ins RC:$src1, RC:$src2, RC:$src3),
6916 !strconcat(OpcodeStr,
6917 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6918 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
6919 NoItinerary, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6921 def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
6922 (ins RC:$src1, x86memop:$src2, RC:$src3),
6923 !strconcat(OpcodeStr,
6924 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6926 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
6928 NoItinerary, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6931 let Predicates = [HasAVX] in {
6932 let ExeDomain = SSEPackedDouble in {
6933 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, f128mem,
6934 memopv2f64, int_x86_sse41_blendvpd>;
6935 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, f256mem,
6936 memopv4f64, int_x86_avx_blendv_pd_256>, VEX_L;
6937 } // ExeDomain = SSEPackedDouble
6938 let ExeDomain = SSEPackedSingle in {
6939 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, f128mem,
6940 memopv4f32, int_x86_sse41_blendvps>;
6941 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, f256mem,
6942 memopv8f32, int_x86_avx_blendv_ps_256>, VEX_L;
6943 } // ExeDomain = SSEPackedSingle
6944 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
6945 memopv2i64, int_x86_sse41_pblendvb>;
6948 let Predicates = [HasAVX2] in {
6949 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
6950 memopv4i64, int_x86_avx2_pblendvb>, VEX_L;
6953 let Predicates = [HasAVX] in {
6954 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
6955 (v16i8 VR128:$src2))),
6956 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6957 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
6958 (v4i32 VR128:$src2))),
6959 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6960 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
6961 (v4f32 VR128:$src2))),
6962 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6963 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
6964 (v2i64 VR128:$src2))),
6965 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6966 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
6967 (v2f64 VR128:$src2))),
6968 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6969 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
6970 (v8i32 VR256:$src2))),
6971 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6972 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
6973 (v8f32 VR256:$src2))),
6974 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6975 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
6976 (v4i64 VR256:$src2))),
6977 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6978 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
6979 (v4f64 VR256:$src2))),
6980 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6982 def : Pat<(v8f32 (X86Blendi (v8f32 VR256:$src1), (v8f32 VR256:$src2),
6984 (VBLENDPSYrri VR256:$src1, VR256:$src2, imm:$mask)>;
6985 def : Pat<(v4f64 (X86Blendi (v4f64 VR256:$src1), (v4f64 VR256:$src2),
6987 (VBLENDPDYrri VR256:$src1, VR256:$src2, imm:$mask)>;
6989 def : Pat<(v8i16 (X86Blendi (v8i16 VR128:$src1), (v8i16 VR128:$src2),
6991 (VPBLENDWrri VR128:$src1, VR128:$src2, imm:$mask)>;
6992 def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$src1), (v4f32 VR128:$src2),
6994 (VBLENDPSrri VR128:$src1, VR128:$src2, imm:$mask)>;
6995 def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$src1), (v2f64 VR128:$src2),
6997 (VBLENDPDrri VR128:$src1, VR128:$src2, imm:$mask)>;
7000 let Predicates = [HasAVX2] in {
7001 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
7002 (v32i8 VR256:$src2))),
7003 (VPBLENDVBYrr VR256:$src1, VR256:$src2, VR256:$mask)>;
7004 def : Pat<(v16i16 (X86Blendi (v16i16 VR256:$src1), (v16i16 VR256:$src2),
7006 (VPBLENDWYrri VR256:$src1, VR256:$src2, imm:$mask)>;
7009 /// SS41I_ternary_int - SSE 4.1 ternary operator
7010 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
7011 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7012 X86MemOperand x86memop, Intrinsic IntId,
7013 OpndItins itins = DEFAULT_ITINS> {
7014 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
7015 (ins VR128:$src1, VR128:$src2),
7016 !strconcat(OpcodeStr,
7017 "\t{$src2, $dst|$dst, $src2}"),
7018 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))],
7021 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
7022 (ins VR128:$src1, x86memop:$src2),
7023 !strconcat(OpcodeStr,
7024 "\t{$src2, $dst|$dst, $src2}"),
7027 (bitconvert (mem_frag addr:$src2)), XMM0))],
7032 let ExeDomain = SSEPackedDouble in
7033 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64, f128mem,
7034 int_x86_sse41_blendvpd>;
7035 let ExeDomain = SSEPackedSingle in
7036 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32, f128mem,
7037 int_x86_sse41_blendvps>;
7038 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64, i128mem,
7039 int_x86_sse41_pblendvb>;
7041 // Aliases with the implicit xmm0 argument
7042 def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7043 (BLENDVPDrr0 VR128:$dst, VR128:$src2)>;
7044 def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7045 (BLENDVPDrm0 VR128:$dst, f128mem:$src2)>;
7046 def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7047 (BLENDVPSrr0 VR128:$dst, VR128:$src2)>;
7048 def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7049 (BLENDVPSrm0 VR128:$dst, f128mem:$src2)>;
7050 def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7051 (PBLENDVBrr0 VR128:$dst, VR128:$src2)>;
7052 def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7053 (PBLENDVBrm0 VR128:$dst, i128mem:$src2)>;
7055 let Predicates = [UseSSE41] in {
7056 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
7057 (v16i8 VR128:$src2))),
7058 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
7059 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
7060 (v4i32 VR128:$src2))),
7061 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
7062 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
7063 (v4f32 VR128:$src2))),
7064 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
7065 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
7066 (v2i64 VR128:$src2))),
7067 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
7068 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
7069 (v2f64 VR128:$src2))),
7070 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
7072 def : Pat<(v8i16 (X86Blendi (v8i16 VR128:$src1), (v8i16 VR128:$src2),
7074 (PBLENDWrri VR128:$src1, VR128:$src2, imm:$mask)>;
7075 def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$src1), (v4f32 VR128:$src2),
7077 (BLENDPSrri VR128:$src1, VR128:$src2, imm:$mask)>;
7078 def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$src1), (v2f64 VR128:$src2),
7080 (BLENDPDrri VR128:$src1, VR128:$src2, imm:$mask)>;
7084 let Predicates = [HasAVX] in
7085 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
7086 "vmovntdqa\t{$src, $dst|$dst, $src}",
7087 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
7089 let Predicates = [HasAVX2] in
7090 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
7091 "vmovntdqa\t{$src, $dst|$dst, $src}",
7092 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
7094 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
7095 "movntdqa\t{$src, $dst|$dst, $src}",
7096 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
7099 //===----------------------------------------------------------------------===//
7100 // SSE4.2 - Compare Instructions
7101 //===----------------------------------------------------------------------===//
7103 /// SS42I_binop_rm - Simple SSE 4.2 binary operator
7104 multiclass SS42I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
7105 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
7106 X86MemOperand x86memop, bit Is2Addr = 1> {
7107 def rr : SS428I<opc, MRMSrcReg, (outs RC:$dst),
7108 (ins RC:$src1, RC:$src2),
7110 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7111 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7112 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>,
7114 def rm : SS428I<opc, MRMSrcMem, (outs RC:$dst),
7115 (ins RC:$src1, x86memop:$src2),
7117 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7118 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7120 (OpVT (OpNode RC:$src1, (memop_frag addr:$src2))))]>, OpSize;
7123 let Predicates = [HasAVX] in
7124 defm VPCMPGTQ : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v2i64, VR128,
7125 memopv2i64, i128mem, 0>, VEX_4V;
7127 let Predicates = [HasAVX2] in
7128 defm VPCMPGTQY : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v4i64, VR256,
7129 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
7131 let Constraints = "$src1 = $dst" in
7132 defm PCMPGTQ : SS42I_binop_rm<0x37, "pcmpgtq", X86pcmpgt, v2i64, VR128,
7133 memopv2i64, i128mem>;
7135 //===----------------------------------------------------------------------===//
7136 // SSE4.2 - String/text Processing Instructions
7137 //===----------------------------------------------------------------------===//
7139 // Packed Compare Implicit Length Strings, Return Mask
7140 multiclass pseudo_pcmpistrm<string asm> {
7141 def REG : PseudoI<(outs VR128:$dst),
7142 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7143 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
7145 def MEM : PseudoI<(outs VR128:$dst),
7146 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7147 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1,
7148 (bc_v16i8 (memopv2i64 addr:$src2)), imm:$src3))]>;
7151 let Defs = [EFLAGS], usesCustomInserter = 1 in {
7152 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
7153 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[UseSSE42]>;
7156 multiclass pcmpistrm_SS42AI<string asm> {
7157 def rr : SS42AI<0x62, MRMSrcReg, (outs),
7158 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7159 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
7162 def rm :SS42AI<0x62, MRMSrcMem, (outs),
7163 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7164 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
7168 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
7169 let Predicates = [HasAVX] in
7170 defm VPCMPISTRM128 : pcmpistrm_SS42AI<"vpcmpistrm">, VEX;
7171 defm PCMPISTRM128 : pcmpistrm_SS42AI<"pcmpistrm"> ;
7174 // Packed Compare Explicit Length Strings, Return Mask
7175 multiclass pseudo_pcmpestrm<string asm> {
7176 def REG : PseudoI<(outs VR128:$dst),
7177 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7178 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
7179 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
7180 def MEM : PseudoI<(outs VR128:$dst),
7181 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7182 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128 VR128:$src1, EAX,
7183 (bc_v16i8 (memopv2i64 addr:$src3)), EDX, imm:$src5))]>;
7186 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
7187 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
7188 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[UseSSE42]>;
7191 multiclass SS42AI_pcmpestrm<string asm> {
7192 def rr : SS42AI<0x60, MRMSrcReg, (outs),
7193 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7194 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7197 def rm : SS42AI<0x60, MRMSrcMem, (outs),
7198 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7199 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7203 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
7204 let Predicates = [HasAVX] in
7205 defm VPCMPESTRM128 : SS42AI_pcmpestrm<"vpcmpestrm">, VEX;
7206 defm PCMPESTRM128 : SS42AI_pcmpestrm<"pcmpestrm">;
7209 // Packed Compare Implicit Length Strings, Return Index
7210 multiclass pseudo_pcmpistri<string asm> {
7211 def REG : PseudoI<(outs GR32:$dst),
7212 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7213 [(set GR32:$dst, EFLAGS,
7214 (X86pcmpistri VR128:$src1, VR128:$src2, imm:$src3))]>;
7215 def MEM : PseudoI<(outs GR32:$dst),
7216 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7217 [(set GR32:$dst, EFLAGS, (X86pcmpistri VR128:$src1,
7218 (bc_v16i8 (memopv2i64 addr:$src2)), imm:$src3))]>;
7221 let Defs = [EFLAGS], usesCustomInserter = 1 in {
7222 defm VPCMPISTRI : pseudo_pcmpistri<"#VPCMPISTRI">, Requires<[HasAVX]>;
7223 defm PCMPISTRI : pseudo_pcmpistri<"#PCMPISTRI">, Requires<[UseSSE42]>;
7226 multiclass SS42AI_pcmpistri<string asm> {
7227 def rr : SS42AI<0x63, MRMSrcReg, (outs),
7228 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7229 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
7232 def rm : SS42AI<0x63, MRMSrcMem, (outs),
7233 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7234 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
7238 let Defs = [ECX, EFLAGS], neverHasSideEffects = 1 in {
7239 let Predicates = [HasAVX] in
7240 defm VPCMPISTRI : SS42AI_pcmpistri<"vpcmpistri">, VEX;
7241 defm PCMPISTRI : SS42AI_pcmpistri<"pcmpistri">;
7244 // Packed Compare Explicit Length Strings, Return Index
7245 multiclass pseudo_pcmpestri<string asm> {
7246 def REG : PseudoI<(outs GR32:$dst),
7247 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7248 [(set GR32:$dst, EFLAGS,
7249 (X86pcmpestri VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
7250 def MEM : PseudoI<(outs GR32:$dst),
7251 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7252 [(set GR32:$dst, EFLAGS,
7253 (X86pcmpestri VR128:$src1, EAX, (bc_v16i8 (memopv2i64 addr:$src3)), EDX,
7257 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
7258 defm VPCMPESTRI : pseudo_pcmpestri<"#VPCMPESTRI">, Requires<[HasAVX]>;
7259 defm PCMPESTRI : pseudo_pcmpestri<"#PCMPESTRI">, Requires<[UseSSE42]>;
7262 multiclass SS42AI_pcmpestri<string asm> {
7263 def rr : SS42AI<0x61, MRMSrcReg, (outs),
7264 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7265 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7268 def rm : SS42AI<0x61, MRMSrcMem, (outs),
7269 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7270 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7274 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
7275 let Predicates = [HasAVX] in
7276 defm VPCMPESTRI : SS42AI_pcmpestri<"vpcmpestri">, VEX;
7277 defm PCMPESTRI : SS42AI_pcmpestri<"pcmpestri">;
7280 //===----------------------------------------------------------------------===//
7281 // SSE4.2 - CRC Instructions
7282 //===----------------------------------------------------------------------===//
7284 // No CRC instructions have AVX equivalents
7286 // crc intrinsic instruction
7287 // This set of instructions are only rm, the only difference is the size
7289 let Constraints = "$src1 = $dst" in {
7290 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
7291 (ins GR32:$src1, i8mem:$src2),
7292 "crc32{b}\t{$src2, $src1|$src1, $src2}",
7294 (int_x86_sse42_crc32_32_8 GR32:$src1,
7295 (load addr:$src2)))], IIC_CRC32_MEM>;
7296 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
7297 (ins GR32:$src1, GR8:$src2),
7298 "crc32{b}\t{$src2, $src1|$src1, $src2}",
7300 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))],
7302 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
7303 (ins GR32:$src1, i16mem:$src2),
7304 "crc32{w}\t{$src2, $src1|$src1, $src2}",
7306 (int_x86_sse42_crc32_32_16 GR32:$src1,
7307 (load addr:$src2)))], IIC_CRC32_MEM>,
7309 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
7310 (ins GR32:$src1, GR16:$src2),
7311 "crc32{w}\t{$src2, $src1|$src1, $src2}",
7313 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))],
7316 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
7317 (ins GR32:$src1, i32mem:$src2),
7318 "crc32{l}\t{$src2, $src1|$src1, $src2}",
7320 (int_x86_sse42_crc32_32_32 GR32:$src1,
7321 (load addr:$src2)))], IIC_CRC32_MEM>;
7322 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
7323 (ins GR32:$src1, GR32:$src2),
7324 "crc32{l}\t{$src2, $src1|$src1, $src2}",
7326 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))],
7328 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
7329 (ins GR64:$src1, i8mem:$src2),
7330 "crc32{b}\t{$src2, $src1|$src1, $src2}",
7332 (int_x86_sse42_crc32_64_8 GR64:$src1,
7333 (load addr:$src2)))], IIC_CRC32_MEM>,
7335 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
7336 (ins GR64:$src1, GR8:$src2),
7337 "crc32{b}\t{$src2, $src1|$src1, $src2}",
7339 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))],
7342 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
7343 (ins GR64:$src1, i64mem:$src2),
7344 "crc32{q}\t{$src2, $src1|$src1, $src2}",
7346 (int_x86_sse42_crc32_64_64 GR64:$src1,
7347 (load addr:$src2)))], IIC_CRC32_MEM>,
7349 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
7350 (ins GR64:$src1, GR64:$src2),
7351 "crc32{q}\t{$src2, $src1|$src1, $src2}",
7353 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))],
7358 //===----------------------------------------------------------------------===//
7359 // SHA-NI Instructions
7360 //===----------------------------------------------------------------------===//
7362 multiclass SHAI_binop<bits<8> Opc, string OpcodeStr, Intrinsic IntId,
7364 def rr : I<Opc, MRMSrcReg, (outs VR128:$dst),
7365 (ins VR128:$src1, VR128:$src2),
7366 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7368 (set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0)),
7369 (set VR128:$dst, (IntId VR128:$src1, VR128:$src2)))]>, T8;
7371 def rm : I<Opc, MRMSrcMem, (outs VR128:$dst),
7372 (ins VR128:$src1, i128mem:$src2),
7373 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7375 (set VR128:$dst, (IntId VR128:$src1,
7376 (bc_v4i32 (memopv2i64 addr:$src2)), XMM0)),
7377 (set VR128:$dst, (IntId VR128:$src1,
7378 (bc_v4i32 (memopv2i64 addr:$src2)))))]>, T8;
7381 let Constraints = "$src1 = $dst", Predicates = [HasSHA] in {
7382 def SHA1RNDS4rri : Ii8<0xCC, MRMSrcReg, (outs VR128:$dst),
7383 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7384 "sha1rnds4\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7386 (int_x86_sha1rnds4 VR128:$src1, VR128:$src2,
7387 (i8 imm:$src3)))]>, TA;
7388 def SHA1RNDS4rmi : Ii8<0xCC, MRMSrcMem, (outs VR128:$dst),
7389 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7390 "sha1rnds4\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7392 (int_x86_sha1rnds4 VR128:$src1,
7393 (bc_v4i32 (memopv2i64 addr:$src2)),
7394 (i8 imm:$src3)))]>, TA;
7396 defm SHA1NEXTE : SHAI_binop<0xC8, "sha1nexte", int_x86_sha1nexte>;
7397 defm SHA1MSG1 : SHAI_binop<0xC9, "sha1msg1", int_x86_sha1msg1>;
7398 defm SHA1MSG2 : SHAI_binop<0xCA, "sha1msg2", int_x86_sha1msg2>;
7401 defm SHA256RNDS2 : SHAI_binop<0xCB, "sha256rnds2", int_x86_sha256rnds2, 1>;
7403 defm SHA256MSG1 : SHAI_binop<0xCC, "sha256msg1", int_x86_sha256msg1>;
7404 defm SHA256MSG2 : SHAI_binop<0xCD, "sha256msg2", int_x86_sha256msg2>;
7407 // Aliases with explicit %xmm0
7408 def : InstAlias<"sha256rnds2\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7409 (SHA256RNDS2rr VR128:$dst, VR128:$src2)>;
7410 def : InstAlias<"sha256rnds2\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7411 (SHA256RNDS2rm VR128:$dst, i128mem:$src2)>;
7413 //===----------------------------------------------------------------------===//
7414 // AES-NI Instructions
7415 //===----------------------------------------------------------------------===//
7417 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
7418 Intrinsic IntId128, bit Is2Addr = 1> {
7419 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
7420 (ins VR128:$src1, VR128:$src2),
7422 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7423 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7424 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
7426 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
7427 (ins VR128:$src1, i128mem:$src2),
7429 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7430 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7432 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
7435 // Perform One Round of an AES Encryption/Decryption Flow
7436 let Predicates = [HasAVX, HasAES] in {
7437 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
7438 int_x86_aesni_aesenc, 0>, VEX_4V;
7439 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
7440 int_x86_aesni_aesenclast, 0>, VEX_4V;
7441 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
7442 int_x86_aesni_aesdec, 0>, VEX_4V;
7443 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
7444 int_x86_aesni_aesdeclast, 0>, VEX_4V;
7447 let Constraints = "$src1 = $dst" in {
7448 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
7449 int_x86_aesni_aesenc>;
7450 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
7451 int_x86_aesni_aesenclast>;
7452 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
7453 int_x86_aesni_aesdec>;
7454 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
7455 int_x86_aesni_aesdeclast>;
7458 // Perform the AES InvMixColumn Transformation
7459 let Predicates = [HasAVX, HasAES] in {
7460 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7462 "vaesimc\t{$src1, $dst|$dst, $src1}",
7464 (int_x86_aesni_aesimc VR128:$src1))]>,
7466 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7467 (ins i128mem:$src1),
7468 "vaesimc\t{$src1, $dst|$dst, $src1}",
7469 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7472 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7474 "aesimc\t{$src1, $dst|$dst, $src1}",
7476 (int_x86_aesni_aesimc VR128:$src1))]>,
7478 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7479 (ins i128mem:$src1),
7480 "aesimc\t{$src1, $dst|$dst, $src1}",
7481 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7484 // AES Round Key Generation Assist
7485 let Predicates = [HasAVX, HasAES] in {
7486 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7487 (ins VR128:$src1, i8imm:$src2),
7488 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7490 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7492 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7493 (ins i128mem:$src1, i8imm:$src2),
7494 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7496 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7499 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7500 (ins VR128:$src1, i8imm:$src2),
7501 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7503 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7505 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7506 (ins i128mem:$src1, i8imm:$src2),
7507 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7509 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7512 //===----------------------------------------------------------------------===//
7513 // PCLMUL Instructions
7514 //===----------------------------------------------------------------------===//
7516 // AVX carry-less Multiplication instructions
7517 def VPCLMULQDQrr : AVXPCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7518 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7519 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7521 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
7523 def VPCLMULQDQrm : AVXPCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7524 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7525 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7526 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7527 (memopv2i64 addr:$src2), imm:$src3))]>;
7529 // Carry-less Multiplication instructions
7530 let Constraints = "$src1 = $dst" in {
7531 def PCLMULQDQrr : PCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7532 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7533 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7535 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))],
7536 IIC_SSE_PCLMULQDQ_RR>;
7538 def PCLMULQDQrm : PCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7539 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7540 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7541 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7542 (memopv2i64 addr:$src2), imm:$src3))],
7543 IIC_SSE_PCLMULQDQ_RM>;
7544 } // Constraints = "$src1 = $dst"
7547 multiclass pclmul_alias<string asm, int immop> {
7548 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7549 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
7551 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7552 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
7554 def : InstAlias<!strconcat("vpclmul", asm,
7555 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7556 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
7558 def : InstAlias<!strconcat("vpclmul", asm,
7559 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7560 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
7562 defm : pclmul_alias<"hqhq", 0x11>;
7563 defm : pclmul_alias<"hqlq", 0x01>;
7564 defm : pclmul_alias<"lqhq", 0x10>;
7565 defm : pclmul_alias<"lqlq", 0x00>;
7567 //===----------------------------------------------------------------------===//
7568 // SSE4A Instructions
7569 //===----------------------------------------------------------------------===//
7571 let Predicates = [HasSSE4A] in {
7573 let Constraints = "$src = $dst" in {
7574 def EXTRQI : Ii8<0x78, MRM0r, (outs VR128:$dst),
7575 (ins VR128:$src, i8imm:$len, i8imm:$idx),
7576 "extrq\t{$idx, $len, $src|$src, $len, $idx}",
7577 [(set VR128:$dst, (int_x86_sse4a_extrqi VR128:$src, imm:$len,
7578 imm:$idx))]>, TB, OpSize;
7579 def EXTRQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7580 (ins VR128:$src, VR128:$mask),
7581 "extrq\t{$mask, $src|$src, $mask}",
7582 [(set VR128:$dst, (int_x86_sse4a_extrq VR128:$src,
7583 VR128:$mask))]>, TB, OpSize;
7585 def INSERTQI : Ii8<0x78, MRMSrcReg, (outs VR128:$dst),
7586 (ins VR128:$src, VR128:$src2, i8imm:$len, i8imm:$idx),
7587 "insertq\t{$idx, $len, $src2, $src|$src, $src2, $len, $idx}",
7588 [(set VR128:$dst, (int_x86_sse4a_insertqi VR128:$src,
7589 VR128:$src2, imm:$len, imm:$idx))]>, XD;
7590 def INSERTQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7591 (ins VR128:$src, VR128:$mask),
7592 "insertq\t{$mask, $src|$src, $mask}",
7593 [(set VR128:$dst, (int_x86_sse4a_insertq VR128:$src,
7594 VR128:$mask))]>, XD;
7597 def MOVNTSS : I<0x2B, MRMDestMem, (outs), (ins f32mem:$dst, VR128:$src),
7598 "movntss\t{$src, $dst|$dst, $src}",
7599 [(int_x86_sse4a_movnt_ss addr:$dst, VR128:$src)]>, XS;
7601 def MOVNTSD : I<0x2B, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
7602 "movntsd\t{$src, $dst|$dst, $src}",
7603 [(int_x86_sse4a_movnt_sd addr:$dst, VR128:$src)]>, XD;
7606 //===----------------------------------------------------------------------===//
7608 //===----------------------------------------------------------------------===//
7610 //===----------------------------------------------------------------------===//
7611 // VBROADCAST - Load from memory and broadcast to all elements of the
7612 // destination operand
7614 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
7615 X86MemOperand x86memop, Intrinsic Int> :
7616 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7617 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7618 [(set RC:$dst, (Int addr:$src))]>, VEX;
7620 // AVX2 adds register forms
7621 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
7623 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7624 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7625 [(set RC:$dst, (Int VR128:$src))]>, VEX;
7627 let ExeDomain = SSEPackedSingle in {
7628 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
7629 int_x86_avx_vbroadcast_ss>;
7630 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
7631 int_x86_avx_vbroadcast_ss_256>, VEX_L;
7633 let ExeDomain = SSEPackedDouble in
7634 def VBROADCASTSDYrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
7635 int_x86_avx_vbroadcast_sd_256>, VEX_L;
7636 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
7637 int_x86_avx_vbroadcastf128_pd_256>, VEX_L;
7639 let ExeDomain = SSEPackedSingle in {
7640 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
7641 int_x86_avx2_vbroadcast_ss_ps>;
7642 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
7643 int_x86_avx2_vbroadcast_ss_ps_256>, VEX_L;
7645 let ExeDomain = SSEPackedDouble in
7646 def VBROADCASTSDYrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
7647 int_x86_avx2_vbroadcast_sd_pd_256>, VEX_L;
7649 let Predicates = [HasAVX2] in
7650 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
7651 int_x86_avx2_vbroadcasti128>, VEX_L;
7653 let Predicates = [HasAVX] in
7654 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
7655 (VBROADCASTF128 addr:$src)>;
7658 //===----------------------------------------------------------------------===//
7659 // VINSERTF128 - Insert packed floating-point values
7661 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7662 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
7663 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7664 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7667 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
7668 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
7669 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7673 let Predicates = [HasAVX] in {
7674 def : Pat<(vinsert128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7676 (VINSERTF128rr VR256:$src1, VR128:$src2,
7677 (INSERT_get_vinsert128_imm VR256:$ins))>;
7678 def : Pat<(vinsert128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7680 (VINSERTF128rr VR256:$src1, VR128:$src2,
7681 (INSERT_get_vinsert128_imm VR256:$ins))>;
7683 def : Pat<(vinsert128_insert:$ins (v8f32 VR256:$src1), (loadv4f32 addr:$src2),
7685 (VINSERTF128rm VR256:$src1, addr:$src2,
7686 (INSERT_get_vinsert128_imm VR256:$ins))>;
7687 def : Pat<(vinsert128_insert:$ins (v4f64 VR256:$src1), (loadv2f64 addr:$src2),
7689 (VINSERTF128rm VR256:$src1, addr:$src2,
7690 (INSERT_get_vinsert128_imm VR256:$ins))>;
7693 let Predicates = [HasAVX1Only] in {
7694 def : Pat<(vinsert128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7696 (VINSERTF128rr VR256:$src1, VR128:$src2,
7697 (INSERT_get_vinsert128_imm VR256:$ins))>;
7698 def : Pat<(vinsert128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7700 (VINSERTF128rr VR256:$src1, VR128:$src2,
7701 (INSERT_get_vinsert128_imm VR256:$ins))>;
7702 def : Pat<(vinsert128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7704 (VINSERTF128rr VR256:$src1, VR128:$src2,
7705 (INSERT_get_vinsert128_imm VR256:$ins))>;
7706 def : Pat<(vinsert128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7708 (VINSERTF128rr VR256:$src1, VR128:$src2,
7709 (INSERT_get_vinsert128_imm VR256:$ins))>;
7711 def : Pat<(vinsert128_insert:$ins (v4i64 VR256:$src1), (loadv2i64 addr:$src2),
7713 (VINSERTF128rm VR256:$src1, addr:$src2,
7714 (INSERT_get_vinsert128_imm VR256:$ins))>;
7715 def : Pat<(vinsert128_insert:$ins (v8i32 VR256:$src1),
7716 (bc_v4i32 (loadv2i64 addr:$src2)),
7718 (VINSERTF128rm VR256:$src1, addr:$src2,
7719 (INSERT_get_vinsert128_imm VR256:$ins))>;
7720 def : Pat<(vinsert128_insert:$ins (v32i8 VR256:$src1),
7721 (bc_v16i8 (loadv2i64 addr:$src2)),
7723 (VINSERTF128rm VR256:$src1, addr:$src2,
7724 (INSERT_get_vinsert128_imm VR256:$ins))>;
7725 def : Pat<(vinsert128_insert:$ins (v16i16 VR256:$src1),
7726 (bc_v8i16 (loadv2i64 addr:$src2)),
7728 (VINSERTF128rm VR256:$src1, addr:$src2,
7729 (INSERT_get_vinsert128_imm VR256:$ins))>;
7732 //===----------------------------------------------------------------------===//
7733 // VEXTRACTF128 - Extract packed floating-point values
7735 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7736 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
7737 (ins VR256:$src1, i8imm:$src2),
7738 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7741 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
7742 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
7743 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7748 let Predicates = [HasAVX] in {
7749 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7750 (v4f32 (VEXTRACTF128rr
7751 (v8f32 VR256:$src1),
7752 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7753 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7754 (v2f64 (VEXTRACTF128rr
7755 (v4f64 VR256:$src1),
7756 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7758 def : Pat<(store (v4f32 (vextract128_extract:$ext (v8f32 VR256:$src1),
7759 (iPTR imm))), addr:$dst),
7760 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7761 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7762 def : Pat<(store (v2f64 (vextract128_extract:$ext (v4f64 VR256:$src1),
7763 (iPTR imm))), addr:$dst),
7764 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7765 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7768 let Predicates = [HasAVX1Only] in {
7769 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7770 (v2i64 (VEXTRACTF128rr
7771 (v4i64 VR256:$src1),
7772 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7773 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7774 (v4i32 (VEXTRACTF128rr
7775 (v8i32 VR256:$src1),
7776 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7777 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7778 (v8i16 (VEXTRACTF128rr
7779 (v16i16 VR256:$src1),
7780 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7781 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7782 (v16i8 (VEXTRACTF128rr
7783 (v32i8 VR256:$src1),
7784 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7786 def : Pat<(alignedstore (v2i64 (vextract128_extract:$ext (v4i64 VR256:$src1),
7787 (iPTR imm))), addr:$dst),
7788 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7789 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7790 def : Pat<(alignedstore (v4i32 (vextract128_extract:$ext (v8i32 VR256:$src1),
7791 (iPTR imm))), addr:$dst),
7792 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7793 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7794 def : Pat<(alignedstore (v8i16 (vextract128_extract:$ext (v16i16 VR256:$src1),
7795 (iPTR imm))), addr:$dst),
7796 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7797 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7798 def : Pat<(alignedstore (v16i8 (vextract128_extract:$ext (v32i8 VR256:$src1),
7799 (iPTR imm))), addr:$dst),
7800 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7801 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7804 //===----------------------------------------------------------------------===//
7805 // VMASKMOV - Conditional SIMD Packed Loads and Stores
7807 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
7808 Intrinsic IntLd, Intrinsic IntLd256,
7809 Intrinsic IntSt, Intrinsic IntSt256> {
7810 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
7811 (ins VR128:$src1, f128mem:$src2),
7812 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7813 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
7815 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
7816 (ins VR256:$src1, f256mem:$src2),
7817 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7818 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7820 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
7821 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
7822 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7823 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7824 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
7825 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
7826 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7827 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V, VEX_L;
7830 let ExeDomain = SSEPackedSingle in
7831 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
7832 int_x86_avx_maskload_ps,
7833 int_x86_avx_maskload_ps_256,
7834 int_x86_avx_maskstore_ps,
7835 int_x86_avx_maskstore_ps_256>;
7836 let ExeDomain = SSEPackedDouble in
7837 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
7838 int_x86_avx_maskload_pd,
7839 int_x86_avx_maskload_pd_256,
7840 int_x86_avx_maskstore_pd,
7841 int_x86_avx_maskstore_pd_256>;
7843 //===----------------------------------------------------------------------===//
7844 // VPERMIL - Permute Single and Double Floating-Point Values
7846 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
7847 RegisterClass RC, X86MemOperand x86memop_f,
7848 X86MemOperand x86memop_i, PatFrag i_frag,
7849 Intrinsic IntVar, ValueType vt> {
7850 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
7851 (ins RC:$src1, RC:$src2),
7852 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7853 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
7854 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
7855 (ins RC:$src1, x86memop_i:$src2),
7856 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7857 [(set RC:$dst, (IntVar RC:$src1,
7858 (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
7860 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
7861 (ins RC:$src1, i8imm:$src2),
7862 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7863 [(set RC:$dst, (vt (X86VPermilp RC:$src1, (i8 imm:$src2))))]>, VEX;
7864 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
7865 (ins x86memop_f:$src1, i8imm:$src2),
7866 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7868 (vt (X86VPermilp (memop addr:$src1), (i8 imm:$src2))))]>, VEX;
7871 let ExeDomain = SSEPackedSingle in {
7872 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
7873 memopv2i64, int_x86_avx_vpermilvar_ps, v4f32>;
7874 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
7875 memopv4i64, int_x86_avx_vpermilvar_ps_256, v8f32>, VEX_L;
7877 let ExeDomain = SSEPackedDouble in {
7878 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
7879 memopv2i64, int_x86_avx_vpermilvar_pd, v2f64>;
7880 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
7881 memopv4i64, int_x86_avx_vpermilvar_pd_256, v4f64>, VEX_L;
7884 let Predicates = [HasAVX] in {
7885 def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7886 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7887 def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7888 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7889 def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (memopv4i64 addr:$src1)),
7891 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7892 def : Pat<(v4i64 (X86VPermilp (memopv4i64 addr:$src1), (i8 imm:$imm))),
7893 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7895 def : Pat<(v2i64 (X86VPermilp VR128:$src1, (i8 imm:$imm))),
7896 (VPERMILPDri VR128:$src1, imm:$imm)>;
7897 def : Pat<(v2i64 (X86VPermilp (memopv2i64 addr:$src1), (i8 imm:$imm))),
7898 (VPERMILPDmi addr:$src1, imm:$imm)>;
7901 //===----------------------------------------------------------------------===//
7902 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
7904 let ExeDomain = SSEPackedSingle in {
7905 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
7906 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7907 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7908 [(set VR256:$dst, (v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7909 (i8 imm:$src3))))]>, VEX_4V, VEX_L;
7910 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
7911 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7912 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7913 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv8f32 addr:$src2),
7914 (i8 imm:$src3)))]>, VEX_4V, VEX_L;
7917 let Predicates = [HasAVX] in {
7918 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7919 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7920 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
7921 (memopv4f64 addr:$src2), (i8 imm:$imm))),
7922 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7925 let Predicates = [HasAVX1Only] in {
7926 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7927 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7928 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7929 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7930 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7931 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7932 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7933 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7935 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
7936 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7937 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7938 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
7939 (memopv4i64 addr:$src2), (i8 imm:$imm))),
7940 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7941 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
7942 (bc_v32i8 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7943 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7944 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7945 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7946 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7949 //===----------------------------------------------------------------------===//
7950 // VZERO - Zero YMM registers
7952 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
7953 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
7954 // Zero All YMM registers
7955 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
7956 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
7958 // Zero Upper bits of YMM registers
7959 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
7960 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
7963 //===----------------------------------------------------------------------===//
7964 // Half precision conversion instructions
7965 //===----------------------------------------------------------------------===//
7966 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7967 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7968 "vcvtph2ps\t{$src, $dst|$dst, $src}",
7969 [(set RC:$dst, (Int VR128:$src))]>,
7971 let neverHasSideEffects = 1, mayLoad = 1 in
7972 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7973 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
7976 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7977 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
7978 (ins RC:$src1, i32i8imm:$src2),
7979 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7980 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
7982 let neverHasSideEffects = 1, mayStore = 1 in
7983 def mr : Ii8<0x1D, MRMDestMem, (outs),
7984 (ins x86memop:$dst, RC:$src1, i32i8imm:$src2),
7985 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
7989 let Predicates = [HasF16C] in {
7990 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
7991 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>, VEX_L;
7992 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
7993 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>, VEX_L;
7996 //===----------------------------------------------------------------------===//
7997 // AVX2 Instructions
7998 //===----------------------------------------------------------------------===//
8000 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
8001 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
8002 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
8003 X86MemOperand x86memop> {
8004 let isCommutable = 1 in
8005 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
8006 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
8007 !strconcat(OpcodeStr,
8008 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
8009 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
8011 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
8012 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
8013 !strconcat(OpcodeStr,
8014 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
8017 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
8021 let isCommutable = 0 in {
8022 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
8023 VR128, memopv2i64, i128mem>;
8024 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
8025 VR256, memopv4i64, i256mem>, VEX_L;
8028 def : Pat<(v4i32 (X86Blendi (v4i32 VR128:$src1), (v4i32 VR128:$src2),
8030 (VPBLENDDrri VR128:$src1, VR128:$src2, imm:$mask)>;
8031 def : Pat<(v8i32 (X86Blendi (v8i32 VR256:$src1), (v8i32 VR256:$src2),
8033 (VPBLENDDYrri VR256:$src1, VR256:$src2, imm:$mask)>;
8035 //===----------------------------------------------------------------------===//
8036 // VPBROADCAST - Load from memory and broadcast to all elements of the
8037 // destination operand
8039 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
8040 X86MemOperand x86memop, PatFrag ld_frag,
8041 Intrinsic Int128, Intrinsic Int256> {
8042 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
8043 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
8044 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
8045 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
8046 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
8048 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
8049 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
8050 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
8051 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX, VEX_L;
8052 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
8053 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
8055 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>,
8059 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
8060 int_x86_avx2_pbroadcastb_128,
8061 int_x86_avx2_pbroadcastb_256>;
8062 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
8063 int_x86_avx2_pbroadcastw_128,
8064 int_x86_avx2_pbroadcastw_256>;
8065 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
8066 int_x86_avx2_pbroadcastd_128,
8067 int_x86_avx2_pbroadcastd_256>;
8068 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
8069 int_x86_avx2_pbroadcastq_128,
8070 int_x86_avx2_pbroadcastq_256>;
8072 let Predicates = [HasAVX2] in {
8073 def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
8074 (VPBROADCASTBrm addr:$src)>;
8075 def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
8076 (VPBROADCASTBYrm addr:$src)>;
8077 def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
8078 (VPBROADCASTWrm addr:$src)>;
8079 def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
8080 (VPBROADCASTWYrm addr:$src)>;
8081 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
8082 (VPBROADCASTDrm addr:$src)>;
8083 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
8084 (VPBROADCASTDYrm addr:$src)>;
8085 def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
8086 (VPBROADCASTQrm addr:$src)>;
8087 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
8088 (VPBROADCASTQYrm addr:$src)>;
8090 def : Pat<(v16i8 (X86VBroadcast (v16i8 VR128:$src))),
8091 (VPBROADCASTBrr VR128:$src)>;
8092 def : Pat<(v32i8 (X86VBroadcast (v16i8 VR128:$src))),
8093 (VPBROADCASTBYrr VR128:$src)>;
8094 def : Pat<(v8i16 (X86VBroadcast (v8i16 VR128:$src))),
8095 (VPBROADCASTWrr VR128:$src)>;
8096 def : Pat<(v16i16 (X86VBroadcast (v8i16 VR128:$src))),
8097 (VPBROADCASTWYrr VR128:$src)>;
8098 def : Pat<(v4i32 (X86VBroadcast (v4i32 VR128:$src))),
8099 (VPBROADCASTDrr VR128:$src)>;
8100 def : Pat<(v8i32 (X86VBroadcast (v4i32 VR128:$src))),
8101 (VPBROADCASTDYrr VR128:$src)>;
8102 def : Pat<(v2i64 (X86VBroadcast (v2i64 VR128:$src))),
8103 (VPBROADCASTQrr VR128:$src)>;
8104 def : Pat<(v4i64 (X86VBroadcast (v2i64 VR128:$src))),
8105 (VPBROADCASTQYrr VR128:$src)>;
8106 def : Pat<(v4f32 (X86VBroadcast (v4f32 VR128:$src))),
8107 (VBROADCASTSSrr VR128:$src)>;
8108 def : Pat<(v8f32 (X86VBroadcast (v4f32 VR128:$src))),
8109 (VBROADCASTSSYrr VR128:$src)>;
8110 def : Pat<(v2f64 (X86VBroadcast (v2f64 VR128:$src))),
8111 (VPBROADCASTQrr VR128:$src)>;
8112 def : Pat<(v4f64 (X86VBroadcast (v2f64 VR128:$src))),
8113 (VBROADCASTSDYrr VR128:$src)>;
8115 // Provide fallback in case the load node that is used in the patterns above
8116 // is used by additional users, which prevents the pattern selection.
8117 let AddedComplexity = 20 in {
8118 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
8119 (VBROADCASTSSrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
8120 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
8121 (VBROADCASTSSYrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
8122 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
8123 (VBROADCASTSDYrr (COPY_TO_REGCLASS FR64:$src, VR128))>;
8125 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
8126 (VBROADCASTSSrr (COPY_TO_REGCLASS GR32:$src, VR128))>;
8127 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
8128 (VBROADCASTSSYrr (COPY_TO_REGCLASS GR32:$src, VR128))>;
8129 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
8130 (VBROADCASTSDYrr (COPY_TO_REGCLASS GR64:$src, VR128))>;
8134 // AVX1 broadcast patterns
8135 let Predicates = [HasAVX1Only] in {
8136 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
8137 (VBROADCASTSSYrm addr:$src)>;
8138 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
8139 (VBROADCASTSDYrm addr:$src)>;
8140 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
8141 (VBROADCASTSSrm addr:$src)>;
8144 let Predicates = [HasAVX] in {
8145 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
8146 (VBROADCASTSSYrm addr:$src)>;
8147 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
8148 (VBROADCASTSDYrm addr:$src)>;
8149 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
8150 (VBROADCASTSSrm addr:$src)>;
8152 // Provide fallback in case the load node that is used in the patterns above
8153 // is used by additional users, which prevents the pattern selection.
8154 let AddedComplexity = 20 in {
8155 // 128bit broadcasts:
8156 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
8157 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0)>;
8158 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
8159 (VINSERTF128rr (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)),
8160 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0), sub_xmm),
8161 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0), 1)>;
8162 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
8163 (VINSERTF128rr (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)),
8164 (VPSHUFDri (COPY_TO_REGCLASS FR64:$src, VR128), 0x44), sub_xmm),
8165 (VPSHUFDri (COPY_TO_REGCLASS FR64:$src, VR128), 0x44), 1)>;
8167 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
8168 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0)>;
8169 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
8170 (VINSERTF128rr (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)),
8171 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0), sub_xmm),
8172 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0), 1)>;
8173 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
8174 (VINSERTF128rr (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)),
8175 (VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), sub_xmm),
8176 (VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), 1)>;
8180 //===----------------------------------------------------------------------===//
8181 // VPERM - Permute instructions
8184 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
8186 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
8187 (ins VR256:$src1, VR256:$src2),
8188 !strconcat(OpcodeStr,
8189 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8191 (OpVT (X86VPermv VR256:$src1, VR256:$src2)))]>,
8193 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
8194 (ins VR256:$src1, i256mem:$src2),
8195 !strconcat(OpcodeStr,
8196 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8198 (OpVT (X86VPermv VR256:$src1,
8199 (bitconvert (mem_frag addr:$src2)))))]>,
8203 defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, v8i32>;
8204 let ExeDomain = SSEPackedSingle in
8205 defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, v8f32>;
8207 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
8209 def Yri : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
8210 (ins VR256:$src1, i8imm:$src2),
8211 !strconcat(OpcodeStr,
8212 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8214 (OpVT (X86VPermi VR256:$src1, (i8 imm:$src2))))]>,
8216 def Ymi : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
8217 (ins i256mem:$src1, i8imm:$src2),
8218 !strconcat(OpcodeStr,
8219 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8221 (OpVT (X86VPermi (mem_frag addr:$src1),
8222 (i8 imm:$src2))))]>, VEX, VEX_L;
8225 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, v4i64>, VEX_W;
8226 let ExeDomain = SSEPackedDouble in
8227 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, v4f64>, VEX_W;
8229 //===----------------------------------------------------------------------===//
8230 // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
8232 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
8233 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
8234 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
8235 [(set VR256:$dst, (v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2,
8236 (i8 imm:$src3))))]>, VEX_4V, VEX_L;
8237 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
8238 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
8239 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
8240 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
8241 (i8 imm:$src3)))]>, VEX_4V, VEX_L;
8243 let Predicates = [HasAVX2] in {
8244 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
8245 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
8246 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
8247 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
8248 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
8249 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
8251 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (memopv4i64 addr:$src2)),
8253 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
8254 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
8255 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
8256 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
8257 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)),
8259 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
8263 //===----------------------------------------------------------------------===//
8264 // VINSERTI128 - Insert packed integer values
8266 let neverHasSideEffects = 1 in {
8267 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
8268 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
8269 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
8272 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
8273 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
8274 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
8278 let Predicates = [HasAVX2] in {
8279 def : Pat<(vinsert128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
8281 (VINSERTI128rr VR256:$src1, VR128:$src2,
8282 (INSERT_get_vinsert128_imm VR256:$ins))>;
8283 def : Pat<(vinsert128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
8285 (VINSERTI128rr VR256:$src1, VR128:$src2,
8286 (INSERT_get_vinsert128_imm VR256:$ins))>;
8287 def : Pat<(vinsert128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
8289 (VINSERTI128rr VR256:$src1, VR128:$src2,
8290 (INSERT_get_vinsert128_imm VR256:$ins))>;
8291 def : Pat<(vinsert128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
8293 (VINSERTI128rr VR256:$src1, VR128:$src2,
8294 (INSERT_get_vinsert128_imm VR256:$ins))>;
8296 def : Pat<(vinsert128_insert:$ins (v4i64 VR256:$src1), (loadv2i64 addr:$src2),
8298 (VINSERTI128rm VR256:$src1, addr:$src2,
8299 (INSERT_get_vinsert128_imm VR256:$ins))>;
8300 def : Pat<(vinsert128_insert:$ins (v8i32 VR256:$src1),
8301 (bc_v4i32 (loadv2i64 addr:$src2)),
8303 (VINSERTI128rm VR256:$src1, addr:$src2,
8304 (INSERT_get_vinsert128_imm VR256:$ins))>;
8305 def : Pat<(vinsert128_insert:$ins (v32i8 VR256:$src1),
8306 (bc_v16i8 (loadv2i64 addr:$src2)),
8308 (VINSERTI128rm VR256:$src1, addr:$src2,
8309 (INSERT_get_vinsert128_imm VR256:$ins))>;
8310 def : Pat<(vinsert128_insert:$ins (v16i16 VR256:$src1),
8311 (bc_v8i16 (loadv2i64 addr:$src2)),
8313 (VINSERTI128rm VR256:$src1, addr:$src2,
8314 (INSERT_get_vinsert128_imm VR256:$ins))>;
8317 //===----------------------------------------------------------------------===//
8318 // VEXTRACTI128 - Extract packed integer values
8320 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
8321 (ins VR256:$src1, i8imm:$src2),
8322 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
8324 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
8326 let neverHasSideEffects = 1, mayStore = 1 in
8327 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
8328 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
8329 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
8332 let Predicates = [HasAVX2] in {
8333 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
8334 (v2i64 (VEXTRACTI128rr
8335 (v4i64 VR256:$src1),
8336 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
8337 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
8338 (v4i32 (VEXTRACTI128rr
8339 (v8i32 VR256:$src1),
8340 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
8341 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
8342 (v8i16 (VEXTRACTI128rr
8343 (v16i16 VR256:$src1),
8344 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
8345 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
8346 (v16i8 (VEXTRACTI128rr
8347 (v32i8 VR256:$src1),
8348 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
8350 def : Pat<(store (v2i64 (vextract128_extract:$ext (v4i64 VR256:$src1),
8351 (iPTR imm))), addr:$dst),
8352 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8353 (EXTRACT_get_vextract128_imm VR128:$ext))>;
8354 def : Pat<(store (v4i32 (vextract128_extract:$ext (v8i32 VR256:$src1),
8355 (iPTR imm))), addr:$dst),
8356 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8357 (EXTRACT_get_vextract128_imm VR128:$ext))>;
8358 def : Pat<(store (v8i16 (vextract128_extract:$ext (v16i16 VR256:$src1),
8359 (iPTR imm))), addr:$dst),
8360 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8361 (EXTRACT_get_vextract128_imm VR128:$ext))>;
8362 def : Pat<(store (v16i8 (vextract128_extract:$ext (v32i8 VR256:$src1),
8363 (iPTR imm))), addr:$dst),
8364 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8365 (EXTRACT_get_vextract128_imm VR128:$ext))>;
8368 //===----------------------------------------------------------------------===//
8369 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
8371 multiclass avx2_pmovmask<string OpcodeStr,
8372 Intrinsic IntLd128, Intrinsic IntLd256,
8373 Intrinsic IntSt128, Intrinsic IntSt256> {
8374 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
8375 (ins VR128:$src1, i128mem:$src2),
8376 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8377 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
8378 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
8379 (ins VR256:$src1, i256mem:$src2),
8380 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8381 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
8383 def mr : AVX28I<0x8e, MRMDestMem, (outs),
8384 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
8385 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8386 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
8387 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
8388 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
8389 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8390 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V, VEX_L;
8393 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
8394 int_x86_avx2_maskload_d,
8395 int_x86_avx2_maskload_d_256,
8396 int_x86_avx2_maskstore_d,
8397 int_x86_avx2_maskstore_d_256>;
8398 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
8399 int_x86_avx2_maskload_q,
8400 int_x86_avx2_maskload_q_256,
8401 int_x86_avx2_maskstore_q,
8402 int_x86_avx2_maskstore_q_256>, VEX_W;
8405 //===----------------------------------------------------------------------===//
8406 // Variable Bit Shifts
8408 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
8409 ValueType vt128, ValueType vt256> {
8410 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
8411 (ins VR128:$src1, VR128:$src2),
8412 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8414 (vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
8416 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
8417 (ins VR128:$src1, i128mem:$src2),
8418 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8420 (vt128 (OpNode VR128:$src1,
8421 (vt128 (bitconvert (memopv2i64 addr:$src2))))))]>,
8423 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
8424 (ins VR256:$src1, VR256:$src2),
8425 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8427 (vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
8429 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
8430 (ins VR256:$src1, i256mem:$src2),
8431 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8433 (vt256 (OpNode VR256:$src1,
8434 (vt256 (bitconvert (memopv4i64 addr:$src2))))))]>,
8438 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
8439 defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", shl, v2i64, v4i64>, VEX_W;
8440 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", srl, v4i32, v8i32>;
8441 defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", srl, v2i64, v4i64>, VEX_W;
8442 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;
8444 //===----------------------------------------------------------------------===//
8445 // VGATHER - GATHER Operations
8446 multiclass avx2_gather<bits<8> opc, string OpcodeStr, RegisterClass RC256,
8447 X86MemOperand memop128, X86MemOperand memop256> {
8448 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst, VR128:$mask_wb),
8449 (ins VR128:$src1, memop128:$src2, VR128:$mask),
8450 !strconcat(OpcodeStr,
8451 "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
8453 def Yrm : AVX28I<opc, MRMSrcMem, (outs RC256:$dst, RC256:$mask_wb),
8454 (ins RC256:$src1, memop256:$src2, RC256:$mask),
8455 !strconcat(OpcodeStr,
8456 "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
8457 []>, VEX_4VOp3, VEX_L;
8460 let mayLoad = 1, Constraints
8461 = "@earlyclobber $dst,@earlyclobber $mask_wb, $src1 = $dst, $mask = $mask_wb"
8463 defm VGATHERDPD : avx2_gather<0x92, "vgatherdpd", VR256, vx64mem, vx64mem>, VEX_W;
8464 defm VGATHERQPD : avx2_gather<0x93, "vgatherqpd", VR256, vx64mem, vy64mem>, VEX_W;
8465 defm VGATHERDPS : avx2_gather<0x92, "vgatherdps", VR256, vx32mem, vy32mem>;
8466 defm VGATHERQPS : avx2_gather<0x93, "vgatherqps", VR128, vx32mem, vy32mem>;
8467 defm VPGATHERDQ : avx2_gather<0x90, "vpgatherdq", VR256, vx64mem, vx64mem>, VEX_W;
8468 defm VPGATHERQQ : avx2_gather<0x91, "vpgatherqq", VR256, vx64mem, vy64mem>, VEX_W;
8469 defm VPGATHERDD : avx2_gather<0x90, "vpgatherdd", VR256, vx32mem, vy32mem>;
8470 defm VPGATHERQD : avx2_gather<0x91, "vpgatherqd", VR128, vx32mem, vy32mem>;