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 //===----------------------------------------------------------------------===//
155 // SSE 1 & 2 Instructions Classes
156 //===----------------------------------------------------------------------===//
158 /// sse12_fp_scalar - SSE 1 & 2 scalar instructions class
159 multiclass sse12_fp_scalar<bits<8> opc, string OpcodeStr, SDNode OpNode,
160 RegisterClass RC, X86MemOperand x86memop,
163 let isCommutable = 1 in {
164 def rr : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
166 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
167 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
168 [(set RC:$dst, (OpNode RC:$src1, RC:$src2))], itins.rr>,
169 Sched<[itins.Sched]>;
171 def rm : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
173 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
174 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
175 [(set RC:$dst, (OpNode RC:$src1, (load addr:$src2)))], itins.rm>,
176 Sched<[itins.Sched.Folded, ReadAfterLd]>;
179 /// sse12_fp_scalar_int - SSE 1 & 2 scalar instructions intrinsics class
180 multiclass sse12_fp_scalar_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
181 string asm, string SSEVer, string FPSizeStr,
182 Operand memopr, ComplexPattern mem_cpat,
185 def rr_Int : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
187 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
188 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
189 [(set RC:$dst, (!cast<Intrinsic>(
190 !strconcat("int_x86_sse", SSEVer, "_", OpcodeStr, FPSizeStr))
191 RC:$src1, RC:$src2))], itins.rr>,
192 Sched<[itins.Sched]>;
193 def rm_Int : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, memopr:$src2),
195 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
196 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
197 [(set RC:$dst, (!cast<Intrinsic>(!strconcat("int_x86_sse",
198 SSEVer, "_", OpcodeStr, FPSizeStr))
199 RC:$src1, mem_cpat:$src2))], itins.rm>,
200 Sched<[itins.Sched.Folded, ReadAfterLd]>;
203 /// sse12_fp_packed - SSE 1 & 2 packed instructions class
204 multiclass sse12_fp_packed<bits<8> opc, string OpcodeStr, SDNode OpNode,
205 RegisterClass RC, ValueType vt,
206 X86MemOperand x86memop, PatFrag mem_frag,
207 Domain d, OpndItins itins, bit Is2Addr = 1> {
208 let isCommutable = 1 in
209 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
211 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
212 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
213 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], itins.rr, d>,
214 Sched<[itins.Sched]>;
216 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
218 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
219 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
220 [(set RC:$dst, (OpNode RC:$src1, (mem_frag addr:$src2)))],
222 Sched<[itins.Sched.Folded, ReadAfterLd]>;
225 /// sse12_fp_packed_logical_rm - SSE 1 & 2 packed instructions class
226 multiclass sse12_fp_packed_logical_rm<bits<8> opc, RegisterClass RC, Domain d,
227 string OpcodeStr, X86MemOperand x86memop,
228 list<dag> pat_rr, list<dag> pat_rm,
230 let isCommutable = 1, hasSideEffects = 0 in
231 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
233 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
234 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
235 pat_rr, NoItinerary, d>,
236 Sched<[WriteVecLogic]>;
237 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
239 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
240 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
241 pat_rm, NoItinerary, d>,
242 Sched<[WriteVecLogicLd, ReadAfterLd]>;
245 //===----------------------------------------------------------------------===//
246 // Non-instruction patterns
247 //===----------------------------------------------------------------------===//
249 // A vector extract of the first f32/f64 position is a subregister copy
250 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
251 (COPY_TO_REGCLASS (v4f32 VR128:$src), FR32)>;
252 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
253 (COPY_TO_REGCLASS (v2f64 VR128:$src), FR64)>;
255 // A 128-bit subvector extract from the first 256-bit vector position
256 // is a subregister copy that needs no instruction.
257 def : Pat<(v4i32 (extract_subvector (v8i32 VR256:$src), (iPTR 0))),
258 (v4i32 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm))>;
259 def : Pat<(v4f32 (extract_subvector (v8f32 VR256:$src), (iPTR 0))),
260 (v4f32 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_xmm))>;
262 def : Pat<(v2i64 (extract_subvector (v4i64 VR256:$src), (iPTR 0))),
263 (v2i64 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm))>;
264 def : Pat<(v2f64 (extract_subvector (v4f64 VR256:$src), (iPTR 0))),
265 (v2f64 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm))>;
267 def : Pat<(v8i16 (extract_subvector (v16i16 VR256:$src), (iPTR 0))),
268 (v8i16 (EXTRACT_SUBREG (v16i16 VR256:$src), sub_xmm))>;
269 def : Pat<(v16i8 (extract_subvector (v32i8 VR256:$src), (iPTR 0))),
270 (v16i8 (EXTRACT_SUBREG (v32i8 VR256:$src), sub_xmm))>;
272 // A 128-bit subvector insert to the first 256-bit vector position
273 // is a subregister copy that needs no instruction.
274 let AddedComplexity = 25 in { // to give priority over vinsertf128rm
275 def : Pat<(insert_subvector undef, (v2i64 VR128:$src), (iPTR 0)),
276 (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
277 def : Pat<(insert_subvector undef, (v2f64 VR128:$src), (iPTR 0)),
278 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
279 def : Pat<(insert_subvector undef, (v4i32 VR128:$src), (iPTR 0)),
280 (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
281 def : Pat<(insert_subvector undef, (v4f32 VR128:$src), (iPTR 0)),
282 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
283 def : Pat<(insert_subvector undef, (v8i16 VR128:$src), (iPTR 0)),
284 (INSERT_SUBREG (v16i16 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
285 def : Pat<(insert_subvector undef, (v16i8 VR128:$src), (iPTR 0)),
286 (INSERT_SUBREG (v32i8 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
289 // Implicitly promote a 32-bit scalar to a vector.
290 def : Pat<(v4f32 (scalar_to_vector FR32:$src)),
291 (COPY_TO_REGCLASS FR32:$src, VR128)>;
292 def : Pat<(v8f32 (scalar_to_vector FR32:$src)),
293 (COPY_TO_REGCLASS FR32:$src, VR128)>;
294 // Implicitly promote a 64-bit scalar to a vector.
295 def : Pat<(v2f64 (scalar_to_vector FR64:$src)),
296 (COPY_TO_REGCLASS FR64:$src, VR128)>;
297 def : Pat<(v4f64 (scalar_to_vector FR64:$src)),
298 (COPY_TO_REGCLASS FR64:$src, VR128)>;
300 // Bitcasts between 128-bit vector types. Return the original type since
301 // no instruction is needed for the conversion
302 let Predicates = [HasSSE2] in {
303 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
304 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
305 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
306 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
307 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
308 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
309 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
310 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
311 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
312 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
313 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
314 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
315 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
316 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
317 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
318 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
319 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
320 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
321 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
322 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
323 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
324 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
325 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
326 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
327 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
328 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
329 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
330 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
331 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
332 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
335 // Bitcasts between 256-bit vector types. Return the original type since
336 // no instruction is needed for the conversion
337 let Predicates = [HasAVX] in {
338 def : Pat<(v4f64 (bitconvert (v8f32 VR256:$src))), (v4f64 VR256:$src)>;
339 def : Pat<(v4f64 (bitconvert (v8i32 VR256:$src))), (v4f64 VR256:$src)>;
340 def : Pat<(v4f64 (bitconvert (v4i64 VR256:$src))), (v4f64 VR256:$src)>;
341 def : Pat<(v4f64 (bitconvert (v16i16 VR256:$src))), (v4f64 VR256:$src)>;
342 def : Pat<(v4f64 (bitconvert (v32i8 VR256:$src))), (v4f64 VR256:$src)>;
343 def : Pat<(v8f32 (bitconvert (v8i32 VR256:$src))), (v8f32 VR256:$src)>;
344 def : Pat<(v8f32 (bitconvert (v4i64 VR256:$src))), (v8f32 VR256:$src)>;
345 def : Pat<(v8f32 (bitconvert (v4f64 VR256:$src))), (v8f32 VR256:$src)>;
346 def : Pat<(v8f32 (bitconvert (v32i8 VR256:$src))), (v8f32 VR256:$src)>;
347 def : Pat<(v8f32 (bitconvert (v16i16 VR256:$src))), (v8f32 VR256:$src)>;
348 def : Pat<(v4i64 (bitconvert (v8f32 VR256:$src))), (v4i64 VR256:$src)>;
349 def : Pat<(v4i64 (bitconvert (v8i32 VR256:$src))), (v4i64 VR256:$src)>;
350 def : Pat<(v4i64 (bitconvert (v4f64 VR256:$src))), (v4i64 VR256:$src)>;
351 def : Pat<(v4i64 (bitconvert (v32i8 VR256:$src))), (v4i64 VR256:$src)>;
352 def : Pat<(v4i64 (bitconvert (v16i16 VR256:$src))), (v4i64 VR256:$src)>;
353 def : Pat<(v32i8 (bitconvert (v4f64 VR256:$src))), (v32i8 VR256:$src)>;
354 def : Pat<(v32i8 (bitconvert (v4i64 VR256:$src))), (v32i8 VR256:$src)>;
355 def : Pat<(v32i8 (bitconvert (v8f32 VR256:$src))), (v32i8 VR256:$src)>;
356 def : Pat<(v32i8 (bitconvert (v8i32 VR256:$src))), (v32i8 VR256:$src)>;
357 def : Pat<(v32i8 (bitconvert (v16i16 VR256:$src))), (v32i8 VR256:$src)>;
358 def : Pat<(v8i32 (bitconvert (v32i8 VR256:$src))), (v8i32 VR256:$src)>;
359 def : Pat<(v8i32 (bitconvert (v16i16 VR256:$src))), (v8i32 VR256:$src)>;
360 def : Pat<(v8i32 (bitconvert (v8f32 VR256:$src))), (v8i32 VR256:$src)>;
361 def : Pat<(v8i32 (bitconvert (v4i64 VR256:$src))), (v8i32 VR256:$src)>;
362 def : Pat<(v8i32 (bitconvert (v4f64 VR256:$src))), (v8i32 VR256:$src)>;
363 def : Pat<(v16i16 (bitconvert (v8f32 VR256:$src))), (v16i16 VR256:$src)>;
364 def : Pat<(v16i16 (bitconvert (v8i32 VR256:$src))), (v16i16 VR256:$src)>;
365 def : Pat<(v16i16 (bitconvert (v4i64 VR256:$src))), (v16i16 VR256:$src)>;
366 def : Pat<(v16i16 (bitconvert (v4f64 VR256:$src))), (v16i16 VR256:$src)>;
367 def : Pat<(v16i16 (bitconvert (v32i8 VR256:$src))), (v16i16 VR256:$src)>;
370 // Alias instructions that map fld0 to xorps for sse or vxorps for avx.
371 // This is expanded by ExpandPostRAPseudos.
372 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
373 isPseudo = 1, SchedRW = [WriteZero] in {
374 def FsFLD0SS : I<0, Pseudo, (outs FR32:$dst), (ins), "",
375 [(set FR32:$dst, fp32imm0)]>, Requires<[HasSSE1]>;
376 def FsFLD0SD : I<0, Pseudo, (outs FR64:$dst), (ins), "",
377 [(set FR64:$dst, fpimm0)]>, Requires<[HasSSE2]>;
380 //===----------------------------------------------------------------------===//
381 // AVX & SSE - Zero/One Vectors
382 //===----------------------------------------------------------------------===//
384 // Alias instruction that maps zero vector to pxor / xorp* for sse.
385 // This is expanded by ExpandPostRAPseudos to an xorps / vxorps, and then
386 // swizzled by ExecutionDepsFix to pxor.
387 // We set canFoldAsLoad because this can be converted to a constant-pool
388 // load of an all-zeros value if folding it would be beneficial.
389 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
390 isPseudo = 1, SchedRW = [WriteZero] in {
391 def V_SET0 : I<0, Pseudo, (outs VR128:$dst), (ins), "",
392 [(set VR128:$dst, (v4f32 immAllZerosV))]>;
395 def : Pat<(v2f64 immAllZerosV), (V_SET0)>;
396 def : Pat<(v4i32 immAllZerosV), (V_SET0)>;
397 def : Pat<(v2i64 immAllZerosV), (V_SET0)>;
398 def : Pat<(v8i16 immAllZerosV), (V_SET0)>;
399 def : Pat<(v16i8 immAllZerosV), (V_SET0)>;
402 // The same as done above but for AVX. The 256-bit AVX1 ISA doesn't support PI,
403 // and doesn't need it because on sandy bridge the register is set to zero
404 // at the rename stage without using any execution unit, so SET0PSY
405 // and SET0PDY can be used for vector int instructions without penalty
406 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
407 isPseudo = 1, Predicates = [HasAVX], SchedRW = [WriteZero] in {
408 def AVX_SET0 : I<0, Pseudo, (outs VR256:$dst), (ins), "",
409 [(set VR256:$dst, (v8f32 immAllZerosV))]>;
412 let Predicates = [HasAVX] in
413 def : Pat<(v4f64 immAllZerosV), (AVX_SET0)>;
415 let Predicates = [HasAVX2] in {
416 def : Pat<(v4i64 immAllZerosV), (AVX_SET0)>;
417 def : Pat<(v8i32 immAllZerosV), (AVX_SET0)>;
418 def : Pat<(v16i16 immAllZerosV), (AVX_SET0)>;
419 def : Pat<(v32i8 immAllZerosV), (AVX_SET0)>;
422 // AVX1 has no support for 256-bit integer instructions, but since the 128-bit
423 // VPXOR instruction writes zero to its upper part, it's safe build zeros.
424 let Predicates = [HasAVX1Only] in {
425 def : Pat<(v32i8 immAllZerosV), (SUBREG_TO_REG (i8 0), (V_SET0), sub_xmm)>;
426 def : Pat<(bc_v32i8 (v8f32 immAllZerosV)),
427 (SUBREG_TO_REG (i8 0), (V_SET0), sub_xmm)>;
429 def : Pat<(v16i16 immAllZerosV), (SUBREG_TO_REG (i16 0), (V_SET0), sub_xmm)>;
430 def : Pat<(bc_v16i16 (v8f32 immAllZerosV)),
431 (SUBREG_TO_REG (i16 0), (V_SET0), sub_xmm)>;
433 def : Pat<(v8i32 immAllZerosV), (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
434 def : Pat<(bc_v8i32 (v8f32 immAllZerosV)),
435 (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
437 def : Pat<(v4i64 immAllZerosV), (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
438 def : Pat<(bc_v4i64 (v8f32 immAllZerosV)),
439 (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
442 // We set canFoldAsLoad because this can be converted to a constant-pool
443 // load of an all-ones value if folding it would be beneficial.
444 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
445 isPseudo = 1, SchedRW = [WriteZero] in {
446 def V_SETALLONES : I<0, Pseudo, (outs VR128:$dst), (ins), "",
447 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
448 let Predicates = [HasAVX2] in
449 def AVX2_SETALLONES : I<0, Pseudo, (outs VR256:$dst), (ins), "",
450 [(set VR256:$dst, (v8i32 immAllOnesV))]>;
454 //===----------------------------------------------------------------------===//
455 // SSE 1 & 2 - Move FP Scalar Instructions
457 // Move Instructions. Register-to-register movss/movsd is not used for FR32/64
458 // register copies because it's a partial register update; FsMOVAPSrr/FsMOVAPDrr
459 // is used instead. Register-to-register movss/movsd is not modeled as an
460 // INSERT_SUBREG because INSERT_SUBREG requires that the insert be implementable
461 // in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
462 //===----------------------------------------------------------------------===//
464 multiclass sse12_move_rr<RegisterClass RC, SDNode OpNode, ValueType vt,
465 X86MemOperand x86memop, string base_opc,
467 def rr : SI<0x10, MRMSrcReg, (outs VR128:$dst),
468 (ins VR128:$src1, RC:$src2),
469 !strconcat(base_opc, asm_opr),
470 [(set VR128:$dst, (vt (OpNode VR128:$src1,
471 (scalar_to_vector RC:$src2))))],
472 IIC_SSE_MOV_S_RR>, Sched<[WriteMove]>;
474 // For the disassembler
475 let isCodeGenOnly = 1, hasSideEffects = 0 in
476 def rr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
477 (ins VR128:$src1, RC:$src2),
478 !strconcat(base_opc, asm_opr),
479 [], IIC_SSE_MOV_S_RR>, Sched<[WriteMove]>;
482 multiclass sse12_move<RegisterClass RC, SDNode OpNode, ValueType vt,
483 X86MemOperand x86memop, string OpcodeStr> {
485 defm V#NAME : sse12_move_rr<RC, OpNode, vt, x86memop, OpcodeStr,
486 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">,
489 def V#NAME#mr : SI<0x11, MRMDestMem, (outs), (ins x86memop:$dst, RC:$src),
490 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
491 [(store RC:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
492 VEX, VEX_LIG, Sched<[WriteStore]>;
494 let Constraints = "$src1 = $dst" in {
495 defm NAME : sse12_move_rr<RC, OpNode, vt, x86memop, OpcodeStr,
496 "\t{$src2, $dst|$dst, $src2}">;
499 def NAME#mr : SI<0x11, MRMDestMem, (outs), (ins x86memop:$dst, RC:$src),
500 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
501 [(store RC:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
505 // Loading from memory automatically zeroing upper bits.
506 multiclass sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
507 PatFrag mem_pat, string OpcodeStr> {
508 def V#NAME#rm : SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
509 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
510 [(set RC:$dst, (mem_pat addr:$src))],
511 IIC_SSE_MOV_S_RM>, VEX, VEX_LIG, Sched<[WriteLoad]>;
512 def NAME#rm : SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
513 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
514 [(set RC:$dst, (mem_pat addr:$src))],
515 IIC_SSE_MOV_S_RM>, Sched<[WriteLoad]>;
518 defm MOVSS : sse12_move<FR32, X86Movss, v4f32, f32mem, "movss">, XS;
519 defm MOVSD : sse12_move<FR64, X86Movsd, v2f64, f64mem, "movsd">, XD;
521 let canFoldAsLoad = 1, isReMaterializable = 1 in {
522 defm MOVSS : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
524 let AddedComplexity = 20 in
525 defm MOVSD : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
529 let Predicates = [HasAVX] in {
530 let AddedComplexity = 15 in {
531 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
532 // MOVS{S,D} to the lower bits.
533 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
534 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
535 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
536 (VMOVSSrr (v4f32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
537 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
538 (VMOVSSrr (v4i32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
539 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
540 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
542 // Move low f32 and clear high bits.
543 def : Pat<(v8f32 (X86vzmovl (v8f32 VR256:$src))),
544 (SUBREG_TO_REG (i32 0),
545 (VMOVSSrr (v4f32 (V_SET0)),
546 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_xmm)), sub_xmm)>;
547 def : Pat<(v8i32 (X86vzmovl (v8i32 VR256:$src))),
548 (SUBREG_TO_REG (i32 0),
549 (VMOVSSrr (v4i32 (V_SET0)),
550 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm)), sub_xmm)>;
553 let AddedComplexity = 20 in {
554 // MOVSSrm zeros the high parts of the register; represent this
555 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
556 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
557 (COPY_TO_REGCLASS (VMOVSSrm addr:$src), VR128)>;
558 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
559 (COPY_TO_REGCLASS (VMOVSSrm addr:$src), VR128)>;
560 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
561 (COPY_TO_REGCLASS (VMOVSSrm addr:$src), VR128)>;
563 // MOVSDrm zeros the high parts of the register; represent this
564 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
565 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
566 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
567 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
568 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
569 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
570 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
571 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
572 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
573 def : Pat<(v2f64 (X86vzload addr:$src)),
574 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
576 // Represent the same patterns above but in the form they appear for
578 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
579 (v4i32 (scalar_to_vector (loadi32 addr:$src))), (iPTR 0)))),
580 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_xmm)>;
581 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
582 (v4f32 (scalar_to_vector (loadf32 addr:$src))), (iPTR 0)))),
583 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_xmm)>;
584 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
585 (v2f64 (scalar_to_vector (loadf64 addr:$src))), (iPTR 0)))),
586 (SUBREG_TO_REG (i32 0), (VMOVSDrm addr:$src), sub_xmm)>;
588 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
589 (v4f32 (scalar_to_vector FR32:$src)), (iPTR 0)))),
590 (SUBREG_TO_REG (i32 0),
591 (v4f32 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)),
593 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
594 (v2f64 (scalar_to_vector FR64:$src)), (iPTR 0)))),
595 (SUBREG_TO_REG (i64 0),
596 (v2f64 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)),
598 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
599 (v2i64 (scalar_to_vector (loadi64 addr:$src))), (iPTR 0)))),
600 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_xmm)>;
602 // Move low f64 and clear high bits.
603 def : Pat<(v4f64 (X86vzmovl (v4f64 VR256:$src))),
604 (SUBREG_TO_REG (i32 0),
605 (VMOVSDrr (v2f64 (V_SET0)),
606 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm)), sub_xmm)>;
608 def : Pat<(v4i64 (X86vzmovl (v4i64 VR256:$src))),
609 (SUBREG_TO_REG (i32 0),
610 (VMOVSDrr (v2i64 (V_SET0)),
611 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm)), sub_xmm)>;
613 // Extract and store.
614 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
616 (VMOVSSmr addr:$dst, (COPY_TO_REGCLASS (v4f32 VR128:$src), FR32))>;
617 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
619 (VMOVSDmr addr:$dst, (COPY_TO_REGCLASS (v2f64 VR128:$src), FR64))>;
621 // Shuffle with VMOVSS
622 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
623 (VMOVSSrr (v4i32 VR128:$src1),
624 (COPY_TO_REGCLASS (v4i32 VR128:$src2), FR32))>;
625 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
626 (VMOVSSrr (v4f32 VR128:$src1),
627 (COPY_TO_REGCLASS (v4f32 VR128:$src2), FR32))>;
630 def : Pat<(v8i32 (X86Movss VR256:$src1, VR256:$src2)),
631 (SUBREG_TO_REG (i32 0),
632 (VMOVSSrr (EXTRACT_SUBREG (v8i32 VR256:$src1), sub_xmm),
633 (EXTRACT_SUBREG (v8i32 VR256:$src2), sub_xmm)),
635 def : Pat<(v8f32 (X86Movss VR256:$src1, VR256:$src2)),
636 (SUBREG_TO_REG (i32 0),
637 (VMOVSSrr (EXTRACT_SUBREG (v8f32 VR256:$src1), sub_xmm),
638 (EXTRACT_SUBREG (v8f32 VR256:$src2), sub_xmm)),
641 // Shuffle with VMOVSD
642 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
643 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
644 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
645 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
646 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
647 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
648 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
649 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
652 def : Pat<(v4i64 (X86Movsd VR256:$src1, VR256:$src2)),
653 (SUBREG_TO_REG (i32 0),
654 (VMOVSDrr (EXTRACT_SUBREG (v4i64 VR256:$src1), sub_xmm),
655 (EXTRACT_SUBREG (v4i64 VR256:$src2), sub_xmm)),
657 def : Pat<(v4f64 (X86Movsd VR256:$src1, VR256:$src2)),
658 (SUBREG_TO_REG (i32 0),
659 (VMOVSDrr (EXTRACT_SUBREG (v4f64 VR256:$src1), sub_xmm),
660 (EXTRACT_SUBREG (v4f64 VR256:$src2), sub_xmm)),
664 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
665 // is during lowering, where it's not possible to recognize the fold cause
666 // it has two uses through a bitcast. One use disappears at isel time and the
667 // fold opportunity reappears.
668 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
669 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
670 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
671 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
672 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
673 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
674 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
675 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
678 let Predicates = [UseSSE1] in {
679 let AddedComplexity = 15 in {
680 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
681 // MOVSS to the lower bits.
682 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
683 (MOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
684 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
685 (MOVSSrr (v4f32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
686 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
687 (MOVSSrr (v4i32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
690 let AddedComplexity = 20 in {
691 // MOVSSrm already zeros the high parts of the register.
692 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
693 (COPY_TO_REGCLASS (MOVSSrm addr:$src), VR128)>;
694 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
695 (COPY_TO_REGCLASS (MOVSSrm addr:$src), VR128)>;
696 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
697 (COPY_TO_REGCLASS (MOVSSrm addr:$src), VR128)>;
700 // Extract and store.
701 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
703 (MOVSSmr addr:$dst, (COPY_TO_REGCLASS VR128:$src, FR32))>;
705 // Shuffle with MOVSS
706 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
707 (MOVSSrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR32))>;
708 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
709 (MOVSSrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR32))>;
712 let Predicates = [UseSSE2] in {
713 let AddedComplexity = 15 in {
714 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
715 // MOVSD to the lower bits.
716 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
717 (MOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
720 let AddedComplexity = 20 in {
721 // MOVSDrm already zeros the high parts of the register.
722 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
723 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
724 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
725 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
726 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
727 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
728 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
729 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
730 def : Pat<(v2f64 (X86vzload addr:$src)),
731 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
734 // Extract and store.
735 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
737 (MOVSDmr addr:$dst, (COPY_TO_REGCLASS VR128:$src, FR64))>;
739 // Shuffle with MOVSD
740 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
741 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
742 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
743 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
744 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
745 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
746 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
747 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
749 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
750 // is during lowering, where it's not possible to recognize the fold cause
751 // it has two uses through a bitcast. One use disappears at isel time and the
752 // fold opportunity reappears.
753 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
754 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
755 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
756 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
757 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
758 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
759 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
760 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
763 //===----------------------------------------------------------------------===//
764 // SSE 1 & 2 - Move Aligned/Unaligned FP Instructions
765 //===----------------------------------------------------------------------===//
767 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
768 X86MemOperand x86memop, PatFrag ld_frag,
769 string asm, Domain d,
771 bit IsReMaterializable = 1> {
772 let neverHasSideEffects = 1 in
773 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
774 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], itins.rr, d>,
776 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
777 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
778 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
779 [(set RC:$dst, (ld_frag addr:$src))], itins.rm, d>,
783 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
784 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
786 defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
787 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
789 defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
790 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
792 defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
793 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
796 defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
797 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
799 defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
800 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
801 TB, OpSize, VEX, VEX_L;
802 defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
803 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
805 defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
806 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
807 TB, OpSize, VEX, VEX_L;
808 defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
809 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
811 defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
812 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
814 defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
815 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
817 defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
818 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
821 let SchedRW = [WriteStore] in {
822 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
823 "movaps\t{$src, $dst|$dst, $src}",
824 [(alignedstore (v4f32 VR128:$src), addr:$dst)],
825 IIC_SSE_MOVA_P_MR>, VEX;
826 def VMOVAPDmr : VPDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
827 "movapd\t{$src, $dst|$dst, $src}",
828 [(alignedstore (v2f64 VR128:$src), addr:$dst)],
829 IIC_SSE_MOVA_P_MR>, VEX;
830 def VMOVUPSmr : VPSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
831 "movups\t{$src, $dst|$dst, $src}",
832 [(store (v4f32 VR128:$src), addr:$dst)],
833 IIC_SSE_MOVU_P_MR>, VEX;
834 def VMOVUPDmr : VPDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
835 "movupd\t{$src, $dst|$dst, $src}",
836 [(store (v2f64 VR128:$src), addr:$dst)],
837 IIC_SSE_MOVU_P_MR>, VEX;
838 def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
839 "movaps\t{$src, $dst|$dst, $src}",
840 [(alignedstore256 (v8f32 VR256:$src), addr:$dst)],
841 IIC_SSE_MOVA_P_MR>, VEX, VEX_L;
842 def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
843 "movapd\t{$src, $dst|$dst, $src}",
844 [(alignedstore256 (v4f64 VR256:$src), addr:$dst)],
845 IIC_SSE_MOVA_P_MR>, VEX, VEX_L;
846 def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
847 "movups\t{$src, $dst|$dst, $src}",
848 [(store (v8f32 VR256:$src), addr:$dst)],
849 IIC_SSE_MOVU_P_MR>, VEX, VEX_L;
850 def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
851 "movupd\t{$src, $dst|$dst, $src}",
852 [(store (v4f64 VR256:$src), addr:$dst)],
853 IIC_SSE_MOVU_P_MR>, VEX, VEX_L;
857 let isCodeGenOnly = 1, hasSideEffects = 0, SchedRW = [WriteMove] in {
858 def VMOVAPSrr_REV : VPSI<0x29, MRMDestReg, (outs VR128:$dst),
860 "movaps\t{$src, $dst|$dst, $src}", [],
861 IIC_SSE_MOVA_P_RR>, VEX;
862 def VMOVAPDrr_REV : VPDI<0x29, MRMDestReg, (outs VR128:$dst),
864 "movapd\t{$src, $dst|$dst, $src}", [],
865 IIC_SSE_MOVA_P_RR>, VEX;
866 def VMOVUPSrr_REV : VPSI<0x11, MRMDestReg, (outs VR128:$dst),
868 "movups\t{$src, $dst|$dst, $src}", [],
869 IIC_SSE_MOVU_P_RR>, VEX;
870 def VMOVUPDrr_REV : VPDI<0x11, MRMDestReg, (outs VR128:$dst),
872 "movupd\t{$src, $dst|$dst, $src}", [],
873 IIC_SSE_MOVU_P_RR>, VEX;
874 def VMOVAPSYrr_REV : VPSI<0x29, MRMDestReg, (outs VR256:$dst),
876 "movaps\t{$src, $dst|$dst, $src}", [],
877 IIC_SSE_MOVA_P_RR>, VEX, VEX_L;
878 def VMOVAPDYrr_REV : VPDI<0x29, MRMDestReg, (outs VR256:$dst),
880 "movapd\t{$src, $dst|$dst, $src}", [],
881 IIC_SSE_MOVA_P_RR>, VEX, VEX_L;
882 def VMOVUPSYrr_REV : VPSI<0x11, MRMDestReg, (outs VR256:$dst),
884 "movups\t{$src, $dst|$dst, $src}", [],
885 IIC_SSE_MOVU_P_RR>, VEX, VEX_L;
886 def VMOVUPDYrr_REV : VPDI<0x11, MRMDestReg, (outs VR256:$dst),
888 "movupd\t{$src, $dst|$dst, $src}", [],
889 IIC_SSE_MOVU_P_RR>, VEX, VEX_L;
892 let Predicates = [HasAVX] in {
893 def : Pat<(v8i32 (X86vzmovl
894 (insert_subvector undef, (v4i32 VR128:$src), (iPTR 0)))),
895 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
896 def : Pat<(v4i64 (X86vzmovl
897 (insert_subvector undef, (v2i64 VR128:$src), (iPTR 0)))),
898 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
899 def : Pat<(v8f32 (X86vzmovl
900 (insert_subvector undef, (v4f32 VR128:$src), (iPTR 0)))),
901 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
902 def : Pat<(v4f64 (X86vzmovl
903 (insert_subvector undef, (v2f64 VR128:$src), (iPTR 0)))),
904 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
908 def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
909 (VMOVUPSYmr addr:$dst, VR256:$src)>;
910 def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
911 (VMOVUPDYmr addr:$dst, VR256:$src)>;
913 let SchedRW = [WriteStore] in {
914 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
915 "movaps\t{$src, $dst|$dst, $src}",
916 [(alignedstore (v4f32 VR128:$src), addr:$dst)],
918 def MOVAPDmr : PDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
919 "movapd\t{$src, $dst|$dst, $src}",
920 [(alignedstore (v2f64 VR128:$src), addr:$dst)],
922 def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
923 "movups\t{$src, $dst|$dst, $src}",
924 [(store (v4f32 VR128:$src), addr:$dst)],
926 def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
927 "movupd\t{$src, $dst|$dst, $src}",
928 [(store (v2f64 VR128:$src), addr:$dst)],
933 let isCodeGenOnly = 1, hasSideEffects = 0, SchedRW = [WriteMove] in {
934 def MOVAPSrr_REV : PSI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
935 "movaps\t{$src, $dst|$dst, $src}", [],
937 def MOVAPDrr_REV : PDI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
938 "movapd\t{$src, $dst|$dst, $src}", [],
940 def MOVUPSrr_REV : PSI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
941 "movups\t{$src, $dst|$dst, $src}", [],
943 def MOVUPDrr_REV : PDI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
944 "movupd\t{$src, $dst|$dst, $src}", [],
948 let Predicates = [HasAVX] in {
949 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
950 (VMOVUPSmr addr:$dst, VR128:$src)>;
951 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
952 (VMOVUPDmr addr:$dst, VR128:$src)>;
955 let Predicates = [UseSSE1] in
956 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
957 (MOVUPSmr addr:$dst, VR128:$src)>;
958 let Predicates = [UseSSE2] in
959 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
960 (MOVUPDmr addr:$dst, VR128:$src)>;
962 // Use vmovaps/vmovups for AVX integer load/store.
963 let Predicates = [HasAVX] in {
964 // 128-bit load/store
965 def : Pat<(alignedloadv2i64 addr:$src),
966 (VMOVAPSrm addr:$src)>;
967 def : Pat<(loadv2i64 addr:$src),
968 (VMOVUPSrm addr:$src)>;
970 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
971 (VMOVAPSmr addr:$dst, VR128:$src)>;
972 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
973 (VMOVAPSmr addr:$dst, VR128:$src)>;
974 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
975 (VMOVAPSmr addr:$dst, VR128:$src)>;
976 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
977 (VMOVAPSmr addr:$dst, VR128:$src)>;
978 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
979 (VMOVUPSmr addr:$dst, VR128:$src)>;
980 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
981 (VMOVUPSmr addr:$dst, VR128:$src)>;
982 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
983 (VMOVUPSmr addr:$dst, VR128:$src)>;
984 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
985 (VMOVUPSmr addr:$dst, VR128:$src)>;
987 // 256-bit load/store
988 def : Pat<(alignedloadv4i64 addr:$src),
989 (VMOVAPSYrm addr:$src)>;
990 def : Pat<(loadv4i64 addr:$src),
991 (VMOVUPSYrm addr:$src)>;
992 def : Pat<(alignedstore256 (v4i64 VR256:$src), addr:$dst),
993 (VMOVAPSYmr addr:$dst, VR256:$src)>;
994 def : Pat<(alignedstore256 (v8i32 VR256:$src), addr:$dst),
995 (VMOVAPSYmr addr:$dst, VR256:$src)>;
996 def : Pat<(alignedstore256 (v16i16 VR256:$src), addr:$dst),
997 (VMOVAPSYmr addr:$dst, VR256:$src)>;
998 def : Pat<(alignedstore256 (v32i8 VR256:$src), addr:$dst),
999 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1000 def : Pat<(store (v4i64 VR256:$src), addr:$dst),
1001 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1002 def : Pat<(store (v8i32 VR256:$src), addr:$dst),
1003 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1004 def : Pat<(store (v16i16 VR256:$src), addr:$dst),
1005 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1006 def : Pat<(store (v32i8 VR256:$src), addr:$dst),
1007 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1009 // Special patterns for storing subvector extracts of lower 128-bits
1010 // Its cheaper to just use VMOVAPS/VMOVUPS instead of VEXTRACTF128mr
1011 def : Pat<(alignedstore (v2f64 (extract_subvector
1012 (v4f64 VR256:$src), (iPTR 0))), addr:$dst),
1013 (VMOVAPDmr addr:$dst, (v2f64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1014 def : Pat<(alignedstore (v4f32 (extract_subvector
1015 (v8f32 VR256:$src), (iPTR 0))), addr:$dst),
1016 (VMOVAPSmr addr:$dst, (v4f32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1017 def : Pat<(alignedstore (v2i64 (extract_subvector
1018 (v4i64 VR256:$src), (iPTR 0))), addr:$dst),
1019 (VMOVAPDmr addr:$dst, (v2i64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1020 def : Pat<(alignedstore (v4i32 (extract_subvector
1021 (v8i32 VR256:$src), (iPTR 0))), addr:$dst),
1022 (VMOVAPSmr addr:$dst, (v4i32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1023 def : Pat<(alignedstore (v8i16 (extract_subvector
1024 (v16i16 VR256:$src), (iPTR 0))), addr:$dst),
1025 (VMOVAPSmr addr:$dst, (v8i16 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1026 def : Pat<(alignedstore (v16i8 (extract_subvector
1027 (v32i8 VR256:$src), (iPTR 0))), addr:$dst),
1028 (VMOVAPSmr addr:$dst, (v16i8 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1030 def : Pat<(store (v2f64 (extract_subvector
1031 (v4f64 VR256:$src), (iPTR 0))), addr:$dst),
1032 (VMOVUPDmr addr:$dst, (v2f64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1033 def : Pat<(store (v4f32 (extract_subvector
1034 (v8f32 VR256:$src), (iPTR 0))), addr:$dst),
1035 (VMOVUPSmr addr:$dst, (v4f32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1036 def : Pat<(store (v2i64 (extract_subvector
1037 (v4i64 VR256:$src), (iPTR 0))), addr:$dst),
1038 (VMOVUPDmr addr:$dst, (v2i64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1039 def : Pat<(store (v4i32 (extract_subvector
1040 (v8i32 VR256:$src), (iPTR 0))), addr:$dst),
1041 (VMOVUPSmr addr:$dst, (v4i32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1042 def : Pat<(store (v8i16 (extract_subvector
1043 (v16i16 VR256:$src), (iPTR 0))), addr:$dst),
1044 (VMOVUPSmr addr:$dst, (v8i16 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1045 def : Pat<(store (v16i8 (extract_subvector
1046 (v32i8 VR256:$src), (iPTR 0))), addr:$dst),
1047 (VMOVUPSmr addr:$dst, (v16i8 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1050 // Use movaps / movups for SSE integer load / store (one byte shorter).
1051 // The instructions selected below are then converted to MOVDQA/MOVDQU
1052 // during the SSE domain pass.
1053 let Predicates = [UseSSE1] in {
1054 def : Pat<(alignedloadv2i64 addr:$src),
1055 (MOVAPSrm addr:$src)>;
1056 def : Pat<(loadv2i64 addr:$src),
1057 (MOVUPSrm addr:$src)>;
1059 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
1060 (MOVAPSmr addr:$dst, VR128:$src)>;
1061 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
1062 (MOVAPSmr addr:$dst, VR128:$src)>;
1063 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
1064 (MOVAPSmr addr:$dst, VR128:$src)>;
1065 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
1066 (MOVAPSmr addr:$dst, VR128:$src)>;
1067 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
1068 (MOVUPSmr addr:$dst, VR128:$src)>;
1069 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
1070 (MOVUPSmr addr:$dst, VR128:$src)>;
1071 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
1072 (MOVUPSmr addr:$dst, VR128:$src)>;
1073 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
1074 (MOVUPSmr addr:$dst, VR128:$src)>;
1077 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
1078 // bits are disregarded. FIXME: Set encoding to pseudo!
1079 let neverHasSideEffects = 1, SchedRW = [WriteMove] in {
1080 def FsVMOVAPSrr : VPSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1081 "movaps\t{$src, $dst|$dst, $src}", [],
1082 IIC_SSE_MOVA_P_RR>, VEX;
1083 def FsVMOVAPDrr : VPDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1084 "movapd\t{$src, $dst|$dst, $src}", [],
1085 IIC_SSE_MOVA_P_RR>, VEX;
1086 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1087 "movaps\t{$src, $dst|$dst, $src}", [],
1089 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1090 "movapd\t{$src, $dst|$dst, $src}", [],
1094 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
1095 // bits are disregarded. FIXME: Set encoding to pseudo!
1096 let canFoldAsLoad = 1, isReMaterializable = 1, SchedRW = [WriteLoad] in {
1097 let isCodeGenOnly = 1 in {
1098 def FsVMOVAPSrm : VPSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1099 "movaps\t{$src, $dst|$dst, $src}",
1100 [(set FR32:$dst, (alignedloadfsf32 addr:$src))],
1101 IIC_SSE_MOVA_P_RM>, VEX;
1102 def FsVMOVAPDrm : VPDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1103 "movapd\t{$src, $dst|$dst, $src}",
1104 [(set FR64:$dst, (alignedloadfsf64 addr:$src))],
1105 IIC_SSE_MOVA_P_RM>, VEX;
1107 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1108 "movaps\t{$src, $dst|$dst, $src}",
1109 [(set FR32:$dst, (alignedloadfsf32 addr:$src))],
1111 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1112 "movapd\t{$src, $dst|$dst, $src}",
1113 [(set FR64:$dst, (alignedloadfsf64 addr:$src))],
1117 //===----------------------------------------------------------------------===//
1118 // SSE 1 & 2 - Move Low packed FP Instructions
1119 //===----------------------------------------------------------------------===//
1121 multiclass sse12_mov_hilo_packed_base<bits<8>opc, SDNode psnode, SDNode pdnode,
1122 string base_opc, string asm_opr,
1123 InstrItinClass itin> {
1124 def PSrm : PI<opc, MRMSrcMem,
1125 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
1126 !strconcat(base_opc, "s", asm_opr),
1128 (psnode VR128:$src1,
1129 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
1130 itin, SSEPackedSingle>, TB,
1131 Sched<[WriteShuffleLd, ReadAfterLd]>;
1133 def PDrm : PI<opc, MRMSrcMem,
1134 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
1135 !strconcat(base_opc, "d", asm_opr),
1136 [(set VR128:$dst, (v2f64 (pdnode VR128:$src1,
1137 (scalar_to_vector (loadf64 addr:$src2)))))],
1138 itin, SSEPackedDouble>, TB, OpSize,
1139 Sched<[WriteShuffleLd, ReadAfterLd]>;
1143 multiclass sse12_mov_hilo_packed<bits<8>opc, SDNode psnode, SDNode pdnode,
1144 string base_opc, InstrItinClass itin> {
1145 defm V#NAME : sse12_mov_hilo_packed_base<opc, psnode, pdnode, base_opc,
1146 "\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1149 let Constraints = "$src1 = $dst" in
1150 defm NAME : sse12_mov_hilo_packed_base<opc, psnode, pdnode, base_opc,
1151 "\t{$src2, $dst|$dst, $src2}",
1155 let AddedComplexity = 20 in {
1156 defm MOVL : sse12_mov_hilo_packed<0x12, X86Movlps, X86Movlpd, "movlp",
1160 let SchedRW = [WriteStore] in {
1161 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1162 "movlps\t{$src, $dst|$dst, $src}",
1163 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1164 (iPTR 0))), addr:$dst)],
1165 IIC_SSE_MOV_LH>, VEX;
1166 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1167 "movlpd\t{$src, $dst|$dst, $src}",
1168 [(store (f64 (vector_extract (v2f64 VR128:$src),
1169 (iPTR 0))), addr:$dst)],
1170 IIC_SSE_MOV_LH>, VEX;
1171 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1172 "movlps\t{$src, $dst|$dst, $src}",
1173 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1174 (iPTR 0))), addr:$dst)],
1176 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1177 "movlpd\t{$src, $dst|$dst, $src}",
1178 [(store (f64 (vector_extract (v2f64 VR128:$src),
1179 (iPTR 0))), addr:$dst)],
1183 let Predicates = [HasAVX] in {
1184 // Shuffle with VMOVLPS
1185 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1186 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1187 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1188 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1190 // Shuffle with VMOVLPD
1191 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1192 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1193 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1194 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1197 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1199 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1200 def : Pat<(store (v4i32 (X86Movlps
1201 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
1202 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1203 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1205 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1206 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1208 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1211 let Predicates = [UseSSE1] in {
1212 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
1213 def : Pat<(store (i64 (vector_extract (bc_v2i64 (v4f32 VR128:$src2)),
1214 (iPTR 0))), addr:$src1),
1215 (MOVLPSmr addr:$src1, VR128:$src2)>;
1217 // Shuffle with MOVLPS
1218 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1219 (MOVLPSrm VR128:$src1, addr:$src2)>;
1220 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1221 (MOVLPSrm VR128:$src1, addr:$src2)>;
1222 def : Pat<(X86Movlps VR128:$src1,
1223 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1224 (MOVLPSrm VR128:$src1, addr:$src2)>;
1227 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1229 (MOVLPSmr addr:$src1, VR128:$src2)>;
1230 def : Pat<(store (v4i32 (X86Movlps
1231 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
1233 (MOVLPSmr addr:$src1, VR128:$src2)>;
1236 let Predicates = [UseSSE2] in {
1237 // Shuffle with MOVLPD
1238 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1239 (MOVLPDrm VR128:$src1, addr:$src2)>;
1240 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1241 (MOVLPDrm VR128:$src1, addr:$src2)>;
1244 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1246 (MOVLPDmr addr:$src1, VR128:$src2)>;
1247 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1249 (MOVLPDmr addr:$src1, VR128:$src2)>;
1252 //===----------------------------------------------------------------------===//
1253 // SSE 1 & 2 - Move Hi packed FP Instructions
1254 //===----------------------------------------------------------------------===//
1256 let AddedComplexity = 20 in {
1257 defm MOVH : sse12_mov_hilo_packed<0x16, X86Movlhps, X86Movlhpd, "movhp",
1261 let SchedRW = [WriteStore] in {
1262 // v2f64 extract element 1 is always custom lowered to unpack high to low
1263 // and extract element 0 so the non-store version isn't too horrible.
1264 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1265 "movhps\t{$src, $dst|$dst, $src}",
1266 [(store (f64 (vector_extract
1267 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1268 (bc_v2f64 (v4f32 VR128:$src))),
1269 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>, VEX;
1270 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1271 "movhpd\t{$src, $dst|$dst, $src}",
1272 [(store (f64 (vector_extract
1273 (v2f64 (X86Unpckh VR128:$src, VR128:$src)),
1274 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>, VEX;
1275 def MOVHPSmr : PSI<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>;
1281 def MOVHPDmr : PDI<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>;
1288 let Predicates = [HasAVX] in {
1290 def : Pat<(X86Movlhps VR128:$src1,
1291 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1292 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1293 def : Pat<(X86Movlhps VR128:$src1,
1294 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1295 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1297 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1298 // is during lowering, where it's not possible to recognize the load fold
1299 // cause it has two uses through a bitcast. One use disappears at isel time
1300 // and the fold opportunity reappears.
1301 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1302 (scalar_to_vector (loadf64 addr:$src2)))),
1303 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1306 let Predicates = [UseSSE1] in {
1308 def : Pat<(X86Movlhps VR128:$src1,
1309 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1310 (MOVHPSrm VR128:$src1, addr:$src2)>;
1311 def : Pat<(X86Movlhps VR128:$src1,
1312 (bc_v4f32 (v2i64 (X86vzload addr:$src2)))),
1313 (MOVHPSrm VR128:$src1, addr:$src2)>;
1316 let Predicates = [UseSSE2] in {
1317 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1318 // is during lowering, where it's not possible to recognize the load fold
1319 // cause it has two uses through a bitcast. One use disappears at isel time
1320 // and the fold opportunity reappears.
1321 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1322 (scalar_to_vector (loadf64 addr:$src2)))),
1323 (MOVHPDrm VR128:$src1, addr:$src2)>;
1326 //===----------------------------------------------------------------------===//
1327 // SSE 1 & 2 - Move Low to High and High to Low packed FP Instructions
1328 //===----------------------------------------------------------------------===//
1330 let AddedComplexity = 20 in {
1331 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
1332 (ins VR128:$src1, VR128:$src2),
1333 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1335 (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))],
1337 VEX_4V, Sched<[WriteShuffle]>;
1338 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
1339 (ins VR128:$src1, VR128:$src2),
1340 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1342 (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))],
1344 VEX_4V, Sched<[WriteShuffle]>;
1346 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1347 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
1348 (ins VR128:$src1, VR128:$src2),
1349 "movlhps\t{$src2, $dst|$dst, $src2}",
1351 (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))],
1352 IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>;
1353 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
1354 (ins VR128:$src1, VR128:$src2),
1355 "movhlps\t{$src2, $dst|$dst, $src2}",
1357 (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))],
1358 IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>;
1361 let Predicates = [HasAVX] in {
1363 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1364 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1365 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1366 (VMOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1369 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1370 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1373 let Predicates = [UseSSE1] in {
1375 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1376 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1377 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1378 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1381 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1382 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1385 //===----------------------------------------------------------------------===//
1386 // SSE 1 & 2 - Conversion Instructions
1387 //===----------------------------------------------------------------------===//
1389 def SSE_CVT_PD : OpndItins<
1390 IIC_SSE_CVT_PD_RR, IIC_SSE_CVT_PD_RM
1393 let Sched = WriteCvtI2F in
1394 def SSE_CVT_PS : OpndItins<
1395 IIC_SSE_CVT_PS_RR, IIC_SSE_CVT_PS_RM
1398 let Sched = WriteCvtI2F in
1399 def SSE_CVT_Scalar : OpndItins<
1400 IIC_SSE_CVT_Scalar_RR, IIC_SSE_CVT_Scalar_RM
1403 let Sched = WriteCvtF2I in
1404 def SSE_CVT_SS2SI_32 : OpndItins<
1405 IIC_SSE_CVT_SS2SI32_RR, IIC_SSE_CVT_SS2SI32_RM
1408 let Sched = WriteCvtF2I in
1409 def SSE_CVT_SS2SI_64 : OpndItins<
1410 IIC_SSE_CVT_SS2SI64_RR, IIC_SSE_CVT_SS2SI64_RM
1413 let Sched = WriteCvtF2I in
1414 def SSE_CVT_SD2SI : OpndItins<
1415 IIC_SSE_CVT_SD2SI_RR, IIC_SSE_CVT_SD2SI_RM
1418 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1419 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1420 string asm, OpndItins itins> {
1421 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1422 [(set DstRC:$dst, (OpNode SrcRC:$src))],
1423 itins.rr>, Sched<[itins.Sched]>;
1424 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1425 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))],
1426 itins.rm>, Sched<[itins.Sched.Folded]>;
1429 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1430 X86MemOperand x86memop, string asm, Domain d,
1432 let neverHasSideEffects = 1 in {
1433 def rr : I<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1434 [], itins.rr, d>, Sched<[itins.Sched]>;
1436 def rm : I<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1437 [], itins.rm, d>, Sched<[itins.Sched.Folded]>;
1441 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1442 X86MemOperand x86memop, string asm> {
1443 let neverHasSideEffects = 1 in {
1444 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
1445 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>,
1446 Sched<[WriteCvtI2F]>;
1448 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1449 (ins DstRC:$src1, x86memop:$src),
1450 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>,
1451 Sched<[WriteCvtI2FLd, ReadAfterLd]>;
1452 } // neverHasSideEffects = 1
1455 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1456 "cvttss2si\t{$src, $dst|$dst, $src}",
1459 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1460 "cvttss2si\t{$src, $dst|$dst, $src}",
1462 XS, VEX, VEX_W, VEX_LIG;
1463 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1464 "cvttsd2si\t{$src, $dst|$dst, $src}",
1467 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1468 "cvttsd2si\t{$src, $dst|$dst, $src}",
1470 XD, VEX, VEX_W, VEX_LIG;
1472 def : InstAlias<"vcvttss2si{l}\t{$src, $dst|$dst, $src}",
1473 (VCVTTSS2SIrr GR32:$dst, FR32:$src), 0>;
1474 def : InstAlias<"vcvttss2si{l}\t{$src, $dst|$dst, $src}",
1475 (VCVTTSS2SIrm GR32:$dst, f32mem:$src), 0>;
1476 def : InstAlias<"vcvttsd2si{l}\t{$src, $dst|$dst, $src}",
1477 (VCVTTSD2SIrr GR32:$dst, FR64:$src), 0>;
1478 def : InstAlias<"vcvttsd2si{l}\t{$src, $dst|$dst, $src}",
1479 (VCVTTSD2SIrm GR32:$dst, f64mem:$src), 0>;
1480 def : InstAlias<"vcvttss2si{q}\t{$src, $dst|$dst, $src}",
1481 (VCVTTSS2SI64rr GR64:$dst, FR32:$src), 0>;
1482 def : InstAlias<"vcvttss2si{q}\t{$src, $dst|$dst, $src}",
1483 (VCVTTSS2SI64rm GR64:$dst, f32mem:$src), 0>;
1484 def : InstAlias<"vcvttsd2si{q}\t{$src, $dst|$dst, $src}",
1485 (VCVTTSD2SI64rr GR64:$dst, FR64:$src), 0>;
1486 def : InstAlias<"vcvttsd2si{q}\t{$src, $dst|$dst, $src}",
1487 (VCVTTSD2SI64rm GR64:$dst, f64mem:$src), 0>;
1489 // The assembler can recognize rr 64-bit instructions by seeing a rxx
1490 // register, but the same isn't true when only using memory operands,
1491 // provide other assembly "l" and "q" forms to address this explicitly
1492 // where appropriate to do so.
1493 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss{l}">,
1494 XS, VEX_4V, VEX_LIG;
1495 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">,
1496 XS, VEX_4V, VEX_W, VEX_LIG;
1497 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">,
1498 XD, VEX_4V, VEX_LIG;
1499 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">,
1500 XD, VEX_4V, VEX_W, VEX_LIG;
1502 def : InstAlias<"vcvtsi2ss\t{$src, $src1, $dst|$dst, $src1, $src}",
1503 (VCVTSI2SSrm FR64:$dst, FR64:$src1, i32mem:$src)>;
1504 def : InstAlias<"vcvtsi2sd\t{$src, $src1, $dst|$dst, $src1, $src}",
1505 (VCVTSI2SDrm FR64:$dst, FR64:$src1, i32mem:$src)>;
1507 let Predicates = [HasAVX] in {
1508 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
1509 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1510 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
1511 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
1512 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
1513 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
1514 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
1515 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
1517 def : Pat<(f32 (sint_to_fp GR32:$src)),
1518 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
1519 def : Pat<(f32 (sint_to_fp GR64:$src)),
1520 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
1521 def : Pat<(f64 (sint_to_fp GR32:$src)),
1522 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
1523 def : Pat<(f64 (sint_to_fp GR64:$src)),
1524 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
1527 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1528 "cvttss2si\t{$src, $dst|$dst, $src}",
1529 SSE_CVT_SS2SI_32>, XS;
1530 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1531 "cvttss2si\t{$src, $dst|$dst, $src}",
1532 SSE_CVT_SS2SI_64>, XS, REX_W;
1533 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1534 "cvttsd2si\t{$src, $dst|$dst, $src}",
1536 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1537 "cvttsd2si\t{$src, $dst|$dst, $src}",
1538 SSE_CVT_SD2SI>, XD, REX_W;
1539 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
1540 "cvtsi2ss{l}\t{$src, $dst|$dst, $src}",
1541 SSE_CVT_Scalar>, XS;
1542 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
1543 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}",
1544 SSE_CVT_Scalar>, XS, REX_W;
1545 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
1546 "cvtsi2sd{l}\t{$src, $dst|$dst, $src}",
1547 SSE_CVT_Scalar>, XD;
1548 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
1549 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}",
1550 SSE_CVT_Scalar>, XD, REX_W;
1552 def : InstAlias<"cvttss2si{l}\t{$src, $dst|$dst, $src}",
1553 (CVTTSS2SIrr GR32:$dst, FR32:$src), 0>;
1554 def : InstAlias<"cvttss2si{l}\t{$src, $dst|$dst, $src}",
1555 (CVTTSS2SIrm GR32:$dst, f32mem:$src), 0>;
1556 def : InstAlias<"cvttsd2si{l}\t{$src, $dst|$dst, $src}",
1557 (CVTTSD2SIrr GR32:$dst, FR64:$src), 0>;
1558 def : InstAlias<"cvttsd2si{l}\t{$src, $dst|$dst, $src}",
1559 (CVTTSD2SIrm GR32:$dst, f64mem:$src), 0>;
1560 def : InstAlias<"cvttss2si{q}\t{$src, $dst|$dst, $src}",
1561 (CVTTSS2SI64rr GR64:$dst, FR32:$src), 0>;
1562 def : InstAlias<"cvttss2si{q}\t{$src, $dst|$dst, $src}",
1563 (CVTTSS2SI64rm GR64:$dst, f32mem:$src), 0>;
1564 def : InstAlias<"cvttsd2si{q}\t{$src, $dst|$dst, $src}",
1565 (CVTTSD2SI64rr GR64:$dst, FR64:$src), 0>;
1566 def : InstAlias<"cvttsd2si{q}\t{$src, $dst|$dst, $src}",
1567 (CVTTSD2SI64rm GR64:$dst, f64mem:$src), 0>;
1569 def : InstAlias<"cvtsi2ss\t{$src, $dst|$dst, $src}",
1570 (CVTSI2SSrm FR64:$dst, i32mem:$src)>;
1571 def : InstAlias<"cvtsi2sd\t{$src, $dst|$dst, $src}",
1572 (CVTSI2SDrm FR64:$dst, i32mem:$src)>;
1574 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
1575 // and/or XMM operand(s).
1577 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1578 Intrinsic Int, Operand memop, ComplexPattern mem_cpat,
1579 string asm, OpndItins itins> {
1580 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
1581 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1582 [(set DstRC:$dst, (Int SrcRC:$src))], itins.rr>,
1583 Sched<[itins.Sched]>;
1584 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins memop:$src),
1585 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1586 [(set DstRC:$dst, (Int mem_cpat:$src))], itins.rm>,
1587 Sched<[itins.Sched.Folded]>;
1590 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
1591 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
1592 PatFrag ld_frag, string asm, OpndItins itins,
1594 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
1596 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1597 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1598 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))],
1599 itins.rr>, Sched<[itins.Sched]>;
1600 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1601 (ins DstRC:$src1, x86memop:$src2),
1603 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1604 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1605 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))],
1606 itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
1609 defm VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32,
1610 int_x86_sse2_cvtsd2si, sdmem, sse_load_f64, "cvtsd2si",
1611 SSE_CVT_SD2SI>, XD, VEX, VEX_LIG;
1612 defm VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
1613 int_x86_sse2_cvtsd2si64, sdmem, sse_load_f64, "cvtsd2si",
1614 SSE_CVT_SD2SI>, XD, VEX, VEX_W, VEX_LIG;
1616 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1617 sdmem, sse_load_f64, "cvtsd2si", SSE_CVT_SD2SI>, XD;
1618 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
1619 sdmem, sse_load_f64, "cvtsd2si", SSE_CVT_SD2SI>, XD, REX_W;
1622 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1623 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss{l}",
1624 SSE_CVT_Scalar, 0>, XS, VEX_4V;
1625 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1626 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss{q}",
1627 SSE_CVT_Scalar, 0>, XS, VEX_4V,
1629 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1630 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd{l}",
1631 SSE_CVT_Scalar, 0>, XD, VEX_4V;
1632 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1633 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd{q}",
1634 SSE_CVT_Scalar, 0>, XD,
1637 let Constraints = "$src1 = $dst" in {
1638 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1639 int_x86_sse_cvtsi2ss, i32mem, loadi32,
1640 "cvtsi2ss{l}", SSE_CVT_Scalar>, XS;
1641 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1642 int_x86_sse_cvtsi642ss, i64mem, loadi64,
1643 "cvtsi2ss{q}", SSE_CVT_Scalar>, XS, REX_W;
1644 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1645 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
1646 "cvtsi2sd{l}", SSE_CVT_Scalar>, XD;
1647 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1648 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
1649 "cvtsi2sd{q}", SSE_CVT_Scalar>, XD, REX_W;
1654 // Aliases for intrinsics
1655 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1656 ssmem, sse_load_f32, "cvttss2si",
1657 SSE_CVT_SS2SI_32>, XS, VEX;
1658 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1659 int_x86_sse_cvttss2si64, ssmem, sse_load_f32,
1660 "cvttss2si", SSE_CVT_SS2SI_64>,
1662 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1663 sdmem, sse_load_f64, "cvttsd2si",
1664 SSE_CVT_SD2SI>, XD, VEX;
1665 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1666 int_x86_sse2_cvttsd2si64, sdmem, sse_load_f64,
1667 "cvttsd2si", SSE_CVT_SD2SI>,
1669 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1670 ssmem, sse_load_f32, "cvttss2si",
1671 SSE_CVT_SS2SI_32>, XS;
1672 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1673 int_x86_sse_cvttss2si64, ssmem, sse_load_f32,
1674 "cvttss2si", SSE_CVT_SS2SI_64>, XS, REX_W;
1675 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1676 sdmem, sse_load_f64, "cvttsd2si",
1678 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1679 int_x86_sse2_cvttsd2si64, sdmem, sse_load_f64,
1680 "cvttsd2si", SSE_CVT_SD2SI>, XD, REX_W;
1682 defm VCVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
1683 ssmem, sse_load_f32, "cvtss2si",
1684 SSE_CVT_SS2SI_32>, XS, VEX, VEX_LIG;
1685 defm VCVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
1686 ssmem, sse_load_f32, "cvtss2si",
1687 SSE_CVT_SS2SI_64>, XS, VEX, VEX_W, VEX_LIG;
1689 defm CVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
1690 ssmem, sse_load_f32, "cvtss2si",
1691 SSE_CVT_SS2SI_32>, XS;
1692 defm CVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
1693 ssmem, sse_load_f32, "cvtss2si",
1694 SSE_CVT_SS2SI_64>, XS, REX_W;
1696 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, i128mem,
1697 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1698 SSEPackedSingle, SSE_CVT_PS>,
1699 TB, VEX, Requires<[HasAVX]>;
1700 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, i256mem,
1701 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1702 SSEPackedSingle, SSE_CVT_PS>,
1703 TB, VEX, VEX_L, Requires<[HasAVX]>;
1705 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, i128mem,
1706 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1707 SSEPackedSingle, SSE_CVT_PS>,
1708 TB, Requires<[UseSSE2]>;
1710 def : InstAlias<"vcvtss2si{l}\t{$src, $dst|$dst, $src}",
1711 (VCVTSS2SIrr GR32:$dst, VR128:$src), 0>;
1712 def : InstAlias<"vcvtss2si{l}\t{$src, $dst|$dst, $src}",
1713 (VCVTSS2SIrm GR32:$dst, ssmem:$src), 0>;
1714 def : InstAlias<"vcvtsd2si{l}\t{$src, $dst|$dst, $src}",
1715 (VCVTSD2SIrr GR32:$dst, VR128:$src), 0>;
1716 def : InstAlias<"vcvtsd2si{l}\t{$src, $dst|$dst, $src}",
1717 (VCVTSD2SIrm GR32:$dst, sdmem:$src), 0>;
1718 def : InstAlias<"vcvtss2si{q}\t{$src, $dst|$dst, $src}",
1719 (VCVTSS2SI64rr GR64:$dst, VR128:$src), 0>;
1720 def : InstAlias<"vcvtss2si{q}\t{$src, $dst|$dst, $src}",
1721 (VCVTSS2SI64rm GR64:$dst, ssmem:$src), 0>;
1722 def : InstAlias<"vcvtsd2si{q}\t{$src, $dst|$dst, $src}",
1723 (VCVTSD2SI64rr GR64:$dst, VR128:$src), 0>;
1724 def : InstAlias<"vcvtsd2si{q}\t{$src, $dst|$dst, $src}",
1725 (VCVTSD2SI64rm GR64:$dst, sdmem:$src), 0>;
1727 def : InstAlias<"cvtss2si{l}\t{$src, $dst|$dst, $src}",
1728 (CVTSS2SIrr GR32:$dst, VR128:$src), 0>;
1729 def : InstAlias<"cvtss2si{l}\t{$src, $dst|$dst, $src}",
1730 (CVTSS2SIrm GR32:$dst, ssmem:$src), 0>;
1731 def : InstAlias<"cvtsd2si{l}\t{$src, $dst|$dst, $src}",
1732 (CVTSD2SIrr GR32:$dst, VR128:$src), 0>;
1733 def : InstAlias<"cvtsd2si{l}\t{$src, $dst|$dst, $src}",
1734 (CVTSD2SIrm GR32:$dst, sdmem:$src), 0>;
1735 def : InstAlias<"cvtss2si{q}\t{$src, $dst|$dst, $src}",
1736 (CVTSS2SI64rr GR64:$dst, VR128:$src), 0>;
1737 def : InstAlias<"cvtss2si{q}\t{$src, $dst|$dst, $src}",
1738 (CVTSS2SI64rm GR64:$dst, ssmem:$src), 0>;
1739 def : InstAlias<"cvtsd2si{q}\t{$src, $dst|$dst, $src}",
1740 (CVTSD2SI64rr GR64:$dst, VR128:$src), 0>;
1741 def : InstAlias<"cvtsd2si{q}\t{$src, $dst|$dst, $src}",
1742 (CVTSD2SI64rm GR64:$dst, sdmem:$src)>;
1746 // Convert scalar double to scalar single
1747 let neverHasSideEffects = 1 in {
1748 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
1749 (ins FR64:$src1, FR64:$src2),
1750 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
1751 IIC_SSE_CVT_Scalar_RR>, VEX_4V, VEX_LIG,
1752 Sched<[WriteCvtF2F]>;
1754 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
1755 (ins FR64:$src1, f64mem:$src2),
1756 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1757 [], IIC_SSE_CVT_Scalar_RM>,
1758 XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG,
1759 Sched<[WriteCvtF2FLd, ReadAfterLd]>;
1762 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
1765 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
1766 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1767 [(set FR32:$dst, (fround FR64:$src))],
1768 IIC_SSE_CVT_Scalar_RR>, Sched<[WriteCvtF2F]>;
1769 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
1770 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1771 [(set FR32:$dst, (fround (loadf64 addr:$src)))],
1772 IIC_SSE_CVT_Scalar_RM>,
1774 Requires<[UseSSE2, OptForSize]>, Sched<[WriteCvtF2FLd]>;
1776 def Int_VCVTSD2SSrr: I<0x5A, MRMSrcReg,
1777 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1778 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1780 (int_x86_sse2_cvtsd2ss VR128:$src1, VR128:$src2))],
1781 IIC_SSE_CVT_Scalar_RR>, XD, VEX_4V, Requires<[HasAVX]>,
1782 Sched<[WriteCvtF2F]>;
1783 def Int_VCVTSD2SSrm: I<0x5A, MRMSrcReg,
1784 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2),
1785 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1786 [(set VR128:$dst, (int_x86_sse2_cvtsd2ss
1787 VR128:$src1, sse_load_f64:$src2))],
1788 IIC_SSE_CVT_Scalar_RM>, XD, VEX_4V, Requires<[HasAVX]>,
1789 Sched<[WriteCvtF2FLd, ReadAfterLd]>;
1791 let Constraints = "$src1 = $dst" in {
1792 def Int_CVTSD2SSrr: I<0x5A, MRMSrcReg,
1793 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1794 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1796 (int_x86_sse2_cvtsd2ss VR128:$src1, VR128:$src2))],
1797 IIC_SSE_CVT_Scalar_RR>, XD, Requires<[UseSSE2]>,
1798 Sched<[WriteCvtF2F]>;
1799 def Int_CVTSD2SSrm: I<0x5A, MRMSrcReg,
1800 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2),
1801 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1802 [(set VR128:$dst, (int_x86_sse2_cvtsd2ss
1803 VR128:$src1, sse_load_f64:$src2))],
1804 IIC_SSE_CVT_Scalar_RM>, XD, Requires<[UseSSE2]>,
1805 Sched<[WriteCvtF2FLd, ReadAfterLd]>;
1808 // Convert scalar single to scalar double
1809 // SSE2 instructions with XS prefix
1810 let neverHasSideEffects = 1 in {
1811 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
1812 (ins FR32:$src1, FR32:$src2),
1813 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1814 [], IIC_SSE_CVT_Scalar_RR>,
1815 XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG,
1816 Sched<[WriteCvtF2F]>;
1818 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
1819 (ins FR32:$src1, f32mem:$src2),
1820 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1821 [], IIC_SSE_CVT_Scalar_RM>,
1822 XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>,
1823 Sched<[WriteCvtF2FLd, ReadAfterLd]>;
1826 def : Pat<(f64 (fextend FR32:$src)),
1827 (VCVTSS2SDrr FR32:$src, FR32:$src)>, Requires<[HasAVX]>;
1828 def : Pat<(fextend (loadf32 addr:$src)),
1829 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>, Requires<[HasAVX]>;
1831 def : Pat<(extloadf32 addr:$src),
1832 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>,
1833 Requires<[HasAVX, OptForSize]>;
1834 def : Pat<(extloadf32 addr:$src),
1835 (VCVTSS2SDrr (f32 (IMPLICIT_DEF)), (VMOVSSrm addr:$src))>,
1836 Requires<[HasAVX, OptForSpeed]>;
1838 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
1839 "cvtss2sd\t{$src, $dst|$dst, $src}",
1840 [(set FR64:$dst, (fextend FR32:$src))],
1841 IIC_SSE_CVT_Scalar_RR>, XS,
1842 Requires<[UseSSE2]>, Sched<[WriteCvtF2F]>;
1843 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
1844 "cvtss2sd\t{$src, $dst|$dst, $src}",
1845 [(set FR64:$dst, (extloadf32 addr:$src))],
1846 IIC_SSE_CVT_Scalar_RM>, XS,
1847 Requires<[UseSSE2, OptForSize]>, Sched<[WriteCvtF2FLd]>;
1849 // extload f32 -> f64. This matches load+fextend because we have a hack in
1850 // the isel (PreprocessForFPConvert) that can introduce loads after dag
1852 // Since these loads aren't folded into the fextend, we have to match it
1854 def : Pat<(fextend (loadf32 addr:$src)),
1855 (CVTSS2SDrm addr:$src)>, Requires<[UseSSE2]>;
1856 def : Pat<(extloadf32 addr:$src),
1857 (CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[UseSSE2, OptForSpeed]>;
1859 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
1860 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1861 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1863 (int_x86_sse2_cvtss2sd VR128:$src1, VR128:$src2))],
1864 IIC_SSE_CVT_Scalar_RR>, XS, VEX_4V, Requires<[HasAVX]>,
1865 Sched<[WriteCvtF2F]>;
1866 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
1867 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2),
1868 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1870 (int_x86_sse2_cvtss2sd VR128:$src1, sse_load_f32:$src2))],
1871 IIC_SSE_CVT_Scalar_RM>, XS, VEX_4V, Requires<[HasAVX]>,
1872 Sched<[WriteCvtF2FLd, ReadAfterLd]>;
1873 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
1874 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
1875 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1876 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1878 (int_x86_sse2_cvtss2sd VR128:$src1, VR128:$src2))],
1879 IIC_SSE_CVT_Scalar_RR>, XS, Requires<[UseSSE2]>,
1880 Sched<[WriteCvtF2F]>;
1881 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
1882 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2),
1883 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1885 (int_x86_sse2_cvtss2sd VR128:$src1, sse_load_f32:$src2))],
1886 IIC_SSE_CVT_Scalar_RM>, XS, Requires<[UseSSE2]>,
1887 Sched<[WriteCvtF2FLd, ReadAfterLd]>;
1890 // Convert packed single/double fp to doubleword
1891 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1892 "cvtps2dq\t{$src, $dst|$dst, $src}",
1893 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))],
1894 IIC_SSE_CVT_PS_RR>, VEX, Sched<[WriteCvtF2I]>;
1895 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1896 "cvtps2dq\t{$src, $dst|$dst, $src}",
1898 (int_x86_sse2_cvtps2dq (memopv4f32 addr:$src)))],
1899 IIC_SSE_CVT_PS_RM>, VEX, Sched<[WriteCvtF2ILd]>;
1900 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1901 "cvtps2dq\t{$src, $dst|$dst, $src}",
1903 (int_x86_avx_cvt_ps2dq_256 VR256:$src))],
1904 IIC_SSE_CVT_PS_RR>, VEX, VEX_L, Sched<[WriteCvtF2I]>;
1905 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1906 "cvtps2dq\t{$src, $dst|$dst, $src}",
1908 (int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)))],
1909 IIC_SSE_CVT_PS_RM>, VEX, VEX_L, Sched<[WriteCvtF2ILd]>;
1910 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1911 "cvtps2dq\t{$src, $dst|$dst, $src}",
1912 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))],
1913 IIC_SSE_CVT_PS_RR>, Sched<[WriteCvtF2I]>;
1914 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$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>, Sched<[WriteCvtF2ILd]>;
1921 // Convert Packed Double FP to Packed DW Integers
1922 let Predicates = [HasAVX] in {
1923 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1924 // register, but the same isn't true when using memory operands instead.
1925 // Provide other assembly rr and rm forms to address this explicitly.
1926 def VCVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1927 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1928 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1929 VEX, Sched<[WriteCvtF2I]>;
1932 def : InstAlias<"vcvtpd2dqx\t{$src, $dst|$dst, $src}",
1933 (VCVTPD2DQrr VR128:$dst, VR128:$src)>;
1934 def VCVTPD2DQXrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1935 "vcvtpd2dqx\t{$src, $dst|$dst, $src}",
1937 (int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)))]>, VEX,
1938 Sched<[WriteCvtF2ILd]>;
1941 def VCVTPD2DQYrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1942 "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}",
1944 (int_x86_avx_cvt_pd2dq_256 VR256:$src))]>, VEX, VEX_L,
1945 Sched<[WriteCvtF2I]>;
1946 def VCVTPD2DQYrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1947 "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}",
1949 (int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)))]>,
1950 VEX, VEX_L, Sched<[WriteCvtF2ILd]>;
1951 def : InstAlias<"vcvtpd2dq\t{$src, $dst|$dst, $src}",
1952 (VCVTPD2DQYrr VR128:$dst, VR256:$src)>;
1955 def CVTPD2DQrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1956 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1958 (int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)))],
1959 IIC_SSE_CVT_PD_RM>, Sched<[WriteCvtF2ILd]>;
1960 def CVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1961 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1962 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))],
1963 IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtF2I]>;
1965 // Convert with truncation packed single/double fp to doubleword
1966 // SSE2 packed instructions with XS prefix
1967 def VCVTTPS2DQrr : VS2SI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1968 "cvttps2dq\t{$src, $dst|$dst, $src}",
1970 (int_x86_sse2_cvttps2dq VR128:$src))],
1971 IIC_SSE_CVT_PS_RR>, VEX, Sched<[WriteCvtF2I]>;
1972 def VCVTTPS2DQrm : VS2SI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1973 "cvttps2dq\t{$src, $dst|$dst, $src}",
1974 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
1975 (memopv4f32 addr:$src)))],
1976 IIC_SSE_CVT_PS_RM>, VEX, Sched<[WriteCvtF2ILd]>;
1977 def VCVTTPS2DQYrr : VS2SI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1978 "cvttps2dq\t{$src, $dst|$dst, $src}",
1980 (int_x86_avx_cvtt_ps2dq_256 VR256:$src))],
1981 IIC_SSE_CVT_PS_RR>, VEX, VEX_L, Sched<[WriteCvtF2I]>;
1982 def VCVTTPS2DQYrm : VS2SI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1983 "cvttps2dq\t{$src, $dst|$dst, $src}",
1984 [(set VR256:$dst, (int_x86_avx_cvtt_ps2dq_256
1985 (memopv8f32 addr:$src)))],
1986 IIC_SSE_CVT_PS_RM>, VEX, VEX_L,
1987 Sched<[WriteCvtF2ILd]>;
1989 def CVTTPS2DQrr : S2SI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1990 "cvttps2dq\t{$src, $dst|$dst, $src}",
1991 [(set VR128:$dst, (int_x86_sse2_cvttps2dq VR128:$src))],
1992 IIC_SSE_CVT_PS_RR>, Sched<[WriteCvtF2I]>;
1993 def CVTTPS2DQrm : S2SI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1994 "cvttps2dq\t{$src, $dst|$dst, $src}",
1996 (int_x86_sse2_cvttps2dq (memopv4f32 addr:$src)))],
1997 IIC_SSE_CVT_PS_RM>, Sched<[WriteCvtF2ILd]>;
1999 let Predicates = [HasAVX] in {
2000 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
2001 (VCVTDQ2PSrr VR128:$src)>;
2002 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
2003 (VCVTDQ2PSrm addr:$src)>;
2005 def : Pat<(int_x86_sse2_cvtdq2ps VR128:$src),
2006 (VCVTDQ2PSrr VR128:$src)>;
2007 def : Pat<(int_x86_sse2_cvtdq2ps (bc_v4i32 (memopv2i64 addr:$src))),
2008 (VCVTDQ2PSrm addr:$src)>;
2010 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
2011 (VCVTTPS2DQrr VR128:$src)>;
2012 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
2013 (VCVTTPS2DQrm addr:$src)>;
2015 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
2016 (VCVTDQ2PSYrr VR256:$src)>;
2017 def : Pat<(v8f32 (sint_to_fp (bc_v8i32 (memopv4i64 addr:$src)))),
2018 (VCVTDQ2PSYrm addr:$src)>;
2020 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
2021 (VCVTTPS2DQYrr VR256:$src)>;
2022 def : Pat<(v8i32 (fp_to_sint (memopv8f32 addr:$src))),
2023 (VCVTTPS2DQYrm addr:$src)>;
2026 let Predicates = [UseSSE2] in {
2027 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
2028 (CVTDQ2PSrr VR128:$src)>;
2029 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
2030 (CVTDQ2PSrm addr:$src)>;
2032 def : Pat<(int_x86_sse2_cvtdq2ps VR128:$src),
2033 (CVTDQ2PSrr VR128:$src)>;
2034 def : Pat<(int_x86_sse2_cvtdq2ps (bc_v4i32 (memopv2i64 addr:$src))),
2035 (CVTDQ2PSrm addr:$src)>;
2037 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
2038 (CVTTPS2DQrr VR128:$src)>;
2039 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
2040 (CVTTPS2DQrm addr:$src)>;
2043 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2044 "cvttpd2dq\t{$src, $dst|$dst, $src}",
2046 (int_x86_sse2_cvttpd2dq VR128:$src))],
2047 IIC_SSE_CVT_PD_RR>, VEX, Sched<[WriteCvtF2I]>;
2049 // The assembler can recognize rr 256-bit instructions by seeing a ymm
2050 // register, but the same isn't true when using memory operands instead.
2051 // Provide other assembly rr and rm forms to address this explicitly.
2054 def : InstAlias<"vcvttpd2dqx\t{$src, $dst|$dst, $src}",
2055 (VCVTTPD2DQrr VR128:$dst, VR128:$src)>;
2056 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2057 "cvttpd2dqx\t{$src, $dst|$dst, $src}",
2058 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
2059 (memopv2f64 addr:$src)))],
2060 IIC_SSE_CVT_PD_RM>, VEX, Sched<[WriteCvtF2ILd]>;
2063 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
2064 "cvttpd2dq{y}\t{$src, $dst|$dst, $src}",
2066 (int_x86_avx_cvtt_pd2dq_256 VR256:$src))],
2067 IIC_SSE_CVT_PD_RR>, VEX, VEX_L, Sched<[WriteCvtF2I]>;
2068 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
2069 "cvttpd2dq{y}\t{$src, $dst|$dst, $src}",
2071 (int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)))],
2072 IIC_SSE_CVT_PD_RM>, VEX, VEX_L, Sched<[WriteCvtF2ILd]>;
2073 def : InstAlias<"vcvttpd2dq\t{$src, $dst|$dst, $src}",
2074 (VCVTTPD2DQYrr VR128:$dst, VR256:$src)>;
2076 let Predicates = [HasAVX] in {
2077 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
2078 (VCVTTPD2DQYrr VR256:$src)>;
2079 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
2080 (VCVTTPD2DQYrm addr:$src)>;
2081 } // Predicates = [HasAVX]
2083 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2084 "cvttpd2dq\t{$src, $dst|$dst, $src}",
2085 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))],
2086 IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtF2I]>;
2087 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
2088 "cvttpd2dq\t{$src, $dst|$dst, $src}",
2089 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
2090 (memopv2f64 addr:$src)))],
2092 Sched<[WriteCvtF2ILd]>;
2094 // Convert packed single to packed double
2095 let Predicates = [HasAVX] in {
2096 // SSE2 instructions without OpSize prefix
2097 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2098 "vcvtps2pd\t{$src, $dst|$dst, $src}",
2099 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))],
2100 IIC_SSE_CVT_PD_RR>, TB, VEX, Sched<[WriteCvtF2F]>;
2101 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
2102 "vcvtps2pd\t{$src, $dst|$dst, $src}",
2103 [(set VR128:$dst, (v2f64 (extloadv2f32 addr:$src)))],
2104 IIC_SSE_CVT_PD_RM>, TB, VEX, Sched<[WriteCvtF2FLd]>;
2105 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
2106 "vcvtps2pd\t{$src, $dst|$dst, $src}",
2108 (int_x86_avx_cvt_ps2_pd_256 VR128:$src))],
2109 IIC_SSE_CVT_PD_RR>, TB, VEX, VEX_L, Sched<[WriteCvtF2F]>;
2110 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
2111 "vcvtps2pd\t{$src, $dst|$dst, $src}",
2113 (int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)))],
2114 IIC_SSE_CVT_PD_RM>, TB, VEX, VEX_L, Sched<[WriteCvtF2FLd]>;
2117 let Predicates = [UseSSE2] in {
2118 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2119 "cvtps2pd\t{$src, $dst|$dst, $src}",
2120 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))],
2121 IIC_SSE_CVT_PD_RR>, TB, Sched<[WriteCvtF2F]>;
2122 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
2123 "cvtps2pd\t{$src, $dst|$dst, $src}",
2124 [(set VR128:$dst, (v2f64 (extloadv2f32 addr:$src)))],
2125 IIC_SSE_CVT_PD_RM>, TB, Sched<[WriteCvtF2FLd]>;
2128 // Convert Packed DW Integers to Packed Double FP
2129 let Predicates = [HasAVX] in {
2130 let neverHasSideEffects = 1, mayLoad = 1 in
2131 def VCVTDQ2PDrm : S2SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2132 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2133 []>, VEX, Sched<[WriteCvtI2FLd]>;
2134 def VCVTDQ2PDrr : S2SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2135 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2137 (int_x86_sse2_cvtdq2pd VR128:$src))]>, VEX,
2138 Sched<[WriteCvtI2F]>;
2139 def VCVTDQ2PDYrm : S2SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
2140 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2142 (int_x86_avx_cvtdq2_pd_256
2143 (bitconvert (memopv2i64 addr:$src))))]>, VEX, VEX_L,
2144 Sched<[WriteCvtI2FLd]>;
2145 def VCVTDQ2PDYrr : S2SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
2146 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2148 (int_x86_avx_cvtdq2_pd_256 VR128:$src))]>, VEX, VEX_L,
2149 Sched<[WriteCvtI2F]>;
2152 let neverHasSideEffects = 1, mayLoad = 1 in
2153 def CVTDQ2PDrm : S2SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2154 "cvtdq2pd\t{$src, $dst|$dst, $src}", [],
2155 IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtI2FLd]>;
2156 def CVTDQ2PDrr : S2SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2157 "cvtdq2pd\t{$src, $dst|$dst, $src}",
2158 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))],
2159 IIC_SSE_CVT_PD_RM>, Sched<[WriteCvtI2F]>;
2161 // AVX 256-bit register conversion intrinsics
2162 let Predicates = [HasAVX] in {
2163 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
2164 (VCVTDQ2PDYrr VR128:$src)>;
2165 def : Pat<(v4f64 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
2166 (VCVTDQ2PDYrm addr:$src)>;
2167 } // Predicates = [HasAVX]
2169 // Convert packed double to packed single
2170 // The assembler can recognize rr 256-bit instructions by seeing a ymm
2171 // register, but the same isn't true when using memory operands instead.
2172 // Provide other assembly rr and rm forms to address this explicitly.
2173 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2174 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2175 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))],
2176 IIC_SSE_CVT_PD_RR>, VEX, Sched<[WriteCvtF2F]>;
2179 def : InstAlias<"vcvtpd2psx\t{$src, $dst|$dst, $src}",
2180 (VCVTPD2PSrr VR128:$dst, VR128:$src)>;
2181 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2182 "cvtpd2psx\t{$src, $dst|$dst, $src}",
2184 (int_x86_sse2_cvtpd2ps (memopv2f64 addr:$src)))],
2185 IIC_SSE_CVT_PD_RM>, VEX, Sched<[WriteCvtF2FLd]>;
2188 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
2189 "cvtpd2ps{y}\t{$src, $dst|$dst, $src}",
2191 (int_x86_avx_cvt_pd2_ps_256 VR256:$src))],
2192 IIC_SSE_CVT_PD_RR>, VEX, VEX_L, Sched<[WriteCvtF2F]>;
2193 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
2194 "cvtpd2ps{y}\t{$src, $dst|$dst, $src}",
2196 (int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)))],
2197 IIC_SSE_CVT_PD_RM>, VEX, VEX_L, Sched<[WriteCvtF2FLd]>;
2198 def : InstAlias<"vcvtpd2ps\t{$src, $dst|$dst, $src}",
2199 (VCVTPD2PSYrr VR128:$dst, VR256:$src)>;
2201 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2202 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2203 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))],
2204 IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtF2F]>;
2205 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2206 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2208 (int_x86_sse2_cvtpd2ps (memopv2f64 addr:$src)))],
2209 IIC_SSE_CVT_PD_RM>, Sched<[WriteCvtF2FLd]>;
2212 // AVX 256-bit register conversion intrinsics
2213 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
2214 // whenever possible to avoid declaring two versions of each one.
2215 let Predicates = [HasAVX] in {
2216 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
2217 (VCVTDQ2PSYrr VR256:$src)>;
2218 def : Pat<(int_x86_avx_cvtdq2_ps_256 (bitconvert (memopv4i64 addr:$src))),
2219 (VCVTDQ2PSYrm addr:$src)>;
2221 // Match fround and fextend for 128/256-bit conversions
2222 def : Pat<(v4f32 (X86vfpround (v2f64 VR128:$src))),
2223 (VCVTPD2PSrr VR128:$src)>;
2224 def : Pat<(v4f32 (X86vfpround (memopv2f64 addr:$src))),
2225 (VCVTPD2PSXrm addr:$src)>;
2226 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
2227 (VCVTPD2PSYrr VR256:$src)>;
2228 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
2229 (VCVTPD2PSYrm addr:$src)>;
2231 def : Pat<(v2f64 (X86vfpext (v4f32 VR128:$src))),
2232 (VCVTPS2PDrr VR128:$src)>;
2233 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
2234 (VCVTPS2PDYrr VR128:$src)>;
2235 def : Pat<(v4f64 (extloadv4f32 addr:$src)),
2236 (VCVTPS2PDYrm addr:$src)>;
2239 let Predicates = [UseSSE2] in {
2240 // Match fround and fextend for 128 conversions
2241 def : Pat<(v4f32 (X86vfpround (v2f64 VR128:$src))),
2242 (CVTPD2PSrr VR128:$src)>;
2243 def : Pat<(v4f32 (X86vfpround (memopv2f64 addr:$src))),
2244 (CVTPD2PSrm addr:$src)>;
2246 def : Pat<(v2f64 (X86vfpext (v4f32 VR128:$src))),
2247 (CVTPS2PDrr VR128:$src)>;
2250 //===----------------------------------------------------------------------===//
2251 // SSE 1 & 2 - Compare Instructions
2252 //===----------------------------------------------------------------------===//
2254 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
2255 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
2256 Operand CC, SDNode OpNode, ValueType VT,
2257 PatFrag ld_frag, string asm, string asm_alt,
2259 def rr : SIi8<0xC2, MRMSrcReg,
2260 (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
2261 [(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, imm:$cc))],
2262 itins.rr>, Sched<[itins.Sched]>;
2263 def rm : SIi8<0xC2, MRMSrcMem,
2264 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
2265 [(set RC:$dst, (OpNode (VT RC:$src1),
2266 (ld_frag addr:$src2), imm:$cc))],
2268 Sched<[itins.Sched.Folded, ReadAfterLd]>;
2270 // Accept explicit immediate argument form instead of comparison code.
2271 let neverHasSideEffects = 1 in {
2272 def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst),
2273 (ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, [],
2274 IIC_SSE_ALU_F32S_RR>, Sched<[itins.Sched]>;
2276 def rm_alt : SIi8<0xC2, MRMSrcMem, (outs RC:$dst),
2277 (ins RC:$src1, x86memop:$src2, i8imm:$cc), asm_alt, [],
2278 IIC_SSE_ALU_F32S_RM>,
2279 Sched<[itins.Sched.Folded, ReadAfterLd]>;
2283 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem, AVXCC, X86cmpss, f32, loadf32,
2284 "cmp${cc}ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2285 "cmpss\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2287 XS, VEX_4V, VEX_LIG;
2288 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem, AVXCC, X86cmpsd, f64, loadf64,
2289 "cmp${cc}sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2290 "cmpsd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2291 SSE_ALU_F32S>, // same latency as 32 bit compare
2292 XD, VEX_4V, VEX_LIG;
2294 let Constraints = "$src1 = $dst" in {
2295 defm CMPSS : sse12_cmp_scalar<FR32, f32mem, SSECC, X86cmpss, f32, loadf32,
2296 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
2297 "cmpss\t{$cc, $src2, $dst|$dst, $src2, $cc}", SSE_ALU_F32S>,
2299 defm CMPSD : sse12_cmp_scalar<FR64, f64mem, SSECC, X86cmpsd, f64, loadf64,
2300 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
2301 "cmpsd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2302 SSE_ALU_F32S>, // same latency as 32 bit compare
2306 multiclass sse12_cmp_scalar_int<X86MemOperand x86memop, Operand CC,
2307 Intrinsic Int, string asm, OpndItins itins> {
2308 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
2309 (ins VR128:$src1, VR128:$src, CC:$cc), asm,
2310 [(set VR128:$dst, (Int VR128:$src1,
2311 VR128:$src, imm:$cc))],
2313 Sched<[itins.Sched]>;
2314 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
2315 (ins VR128:$src1, x86memop:$src, CC:$cc), asm,
2316 [(set VR128:$dst, (Int VR128:$src1,
2317 (load addr:$src), imm:$cc))],
2319 Sched<[itins.Sched.Folded, ReadAfterLd]>;
2322 // Aliases to match intrinsics which expect XMM operand(s).
2323 defm Int_VCMPSS : sse12_cmp_scalar_int<f32mem, AVXCC, int_x86_sse_cmp_ss,
2324 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
2327 defm Int_VCMPSD : sse12_cmp_scalar_int<f64mem, AVXCC, int_x86_sse2_cmp_sd,
2328 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
2329 SSE_ALU_F32S>, // same latency as f32
2331 let Constraints = "$src1 = $dst" in {
2332 defm Int_CMPSS : sse12_cmp_scalar_int<f32mem, SSECC, int_x86_sse_cmp_ss,
2333 "cmp${cc}ss\t{$src, $dst|$dst, $src}",
2335 defm Int_CMPSD : sse12_cmp_scalar_int<f64mem, SSECC, int_x86_sse2_cmp_sd,
2336 "cmp${cc}sd\t{$src, $dst|$dst, $src}",
2337 SSE_ALU_F32S>, // same latency as f32
2342 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
2343 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
2344 ValueType vt, X86MemOperand x86memop,
2345 PatFrag ld_frag, string OpcodeStr> {
2346 def rr: SI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
2347 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2348 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))],
2351 def rm: SI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
2352 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2353 [(set EFLAGS, (OpNode (vt RC:$src1),
2354 (ld_frag addr:$src2)))],
2356 Sched<[WriteFAddLd, ReadAfterLd]>;
2359 let Defs = [EFLAGS] in {
2360 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2361 "ucomiss">, TB, VEX, VEX_LIG;
2362 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2363 "ucomisd">, TB, OpSize, VEX, VEX_LIG;
2364 let Pattern = []<dag> in {
2365 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2366 "comiss">, TB, VEX, VEX_LIG;
2367 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2368 "comisd">, TB, OpSize, VEX, VEX_LIG;
2371 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2372 load, "ucomiss">, TB, VEX;
2373 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2374 load, "ucomisd">, TB, OpSize, VEX;
2376 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
2377 load, "comiss">, TB, VEX;
2378 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
2379 load, "comisd">, TB, OpSize, VEX;
2380 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2382 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2383 "ucomisd">, TB, OpSize;
2385 let Pattern = []<dag> in {
2386 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2388 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2389 "comisd">, TB, OpSize;
2392 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2393 load, "ucomiss">, TB;
2394 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2395 load, "ucomisd">, TB, OpSize;
2397 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
2399 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
2400 "comisd">, TB, OpSize;
2401 } // Defs = [EFLAGS]
2403 // sse12_cmp_packed - sse 1 & 2 compare packed instructions
2404 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
2405 Operand CC, Intrinsic Int, string asm,
2406 string asm_alt, Domain d> {
2407 def rri : PIi8<0xC2, MRMSrcReg,
2408 (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
2409 [(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))],
2410 IIC_SSE_CMPP_RR, d>,
2412 def rmi : PIi8<0xC2, MRMSrcMem,
2413 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
2414 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))],
2415 IIC_SSE_CMPP_RM, d>,
2416 Sched<[WriteFAddLd, ReadAfterLd]>;
2418 // Accept explicit immediate argument form instead of comparison code.
2419 let neverHasSideEffects = 1 in {
2420 def rri_alt : PIi8<0xC2, MRMSrcReg,
2421 (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
2422 asm_alt, [], IIC_SSE_CMPP_RR, d>, Sched<[WriteFAdd]>;
2423 def rmi_alt : PIi8<0xC2, MRMSrcMem,
2424 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc),
2425 asm_alt, [], IIC_SSE_CMPP_RM, d>,
2426 Sched<[WriteFAddLd, ReadAfterLd]>;
2430 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse_cmp_ps,
2431 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2432 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2433 SSEPackedSingle>, TB, VEX_4V;
2434 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse2_cmp_pd,
2435 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2436 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2437 SSEPackedDouble>, TB, OpSize, VEX_4V;
2438 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_ps_256,
2439 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2440 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2441 SSEPackedSingle>, TB, VEX_4V, VEX_L;
2442 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_pd_256,
2443 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2444 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2445 SSEPackedDouble>, TB, OpSize, VEX_4V, VEX_L;
2446 let Constraints = "$src1 = $dst" in {
2447 defm CMPPS : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse_cmp_ps,
2448 "cmp${cc}ps\t{$src2, $dst|$dst, $src2}",
2449 "cmpps\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2450 SSEPackedSingle>, TB;
2451 defm CMPPD : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse2_cmp_pd,
2452 "cmp${cc}pd\t{$src2, $dst|$dst, $src2}",
2453 "cmppd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2454 SSEPackedDouble>, TB, OpSize;
2457 let Predicates = [HasAVX] in {
2458 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2459 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2460 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2461 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2462 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2463 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2464 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2465 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2467 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
2468 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
2469 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
2470 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
2471 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
2472 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
2473 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
2474 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
2477 let Predicates = [UseSSE1] in {
2478 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2479 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2480 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2481 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2484 let Predicates = [UseSSE2] in {
2485 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2486 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2487 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2488 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2491 //===----------------------------------------------------------------------===//
2492 // SSE 1 & 2 - Shuffle Instructions
2493 //===----------------------------------------------------------------------===//
2495 /// sse12_shuffle - sse 1 & 2 shuffle instructions
2496 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
2497 ValueType vt, string asm, PatFrag mem_frag,
2498 Domain d, bit IsConvertibleToThreeAddress = 0> {
2499 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
2500 (ins RC:$src1, x86memop:$src2, i8imm:$src3), asm,
2501 [(set RC:$dst, (vt (X86Shufp RC:$src1, (mem_frag addr:$src2),
2502 (i8 imm:$src3))))], IIC_SSE_SHUFP, d>,
2503 Sched<[WriteShuffleLd, ReadAfterLd]>;
2504 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
2505 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
2506 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
2507 [(set RC:$dst, (vt (X86Shufp RC:$src1, RC:$src2,
2508 (i8 imm:$src3))))], IIC_SSE_SHUFP, d>,
2509 Sched<[WriteShuffle]>;
2512 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2513 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2514 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
2515 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
2516 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2517 memopv8f32, SSEPackedSingle>, TB, VEX_4V, VEX_L;
2518 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2519 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2520 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2521 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
2522 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2523 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V, VEX_L;
2525 let Constraints = "$src1 = $dst" in {
2526 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2527 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2528 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
2530 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2531 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2532 memopv2f64, SSEPackedDouble, 1 /* cvt to pshufd */>,
2536 let Predicates = [HasAVX] in {
2537 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2538 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2539 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2540 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2541 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2543 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2544 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2545 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2546 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2547 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2550 def : Pat<(v8i32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2551 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2552 def : Pat<(v8i32 (X86Shufp VR256:$src1,
2553 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
2554 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2556 def : Pat<(v4i64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2557 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2558 def : Pat<(v4i64 (X86Shufp VR256:$src1,
2559 (memopv4i64 addr:$src2), (i8 imm:$imm))),
2560 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2563 let Predicates = [UseSSE1] in {
2564 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2565 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2566 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2567 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2568 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2571 let Predicates = [UseSSE2] in {
2572 // Generic SHUFPD patterns
2573 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2574 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2575 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2576 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2577 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2580 //===----------------------------------------------------------------------===//
2581 // SSE 1 & 2 - Unpack Instructions
2582 //===----------------------------------------------------------------------===//
2584 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
2585 multiclass sse12_unpack_interleave<bits<8> opc, SDNode OpNode, ValueType vt,
2586 PatFrag mem_frag, RegisterClass RC,
2587 X86MemOperand x86memop, string asm,
2589 def rr : PI<opc, MRMSrcReg,
2590 (outs RC:$dst), (ins RC:$src1, RC:$src2),
2592 (vt (OpNode RC:$src1, RC:$src2)))],
2593 IIC_SSE_UNPCK, d>, Sched<[WriteShuffle]>;
2594 def rm : PI<opc, MRMSrcMem,
2595 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
2597 (vt (OpNode RC:$src1,
2598 (mem_frag addr:$src2))))],
2600 Sched<[WriteShuffleLd, ReadAfterLd]>;
2603 defm VUNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
2604 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2605 SSEPackedSingle>, TB, VEX_4V;
2606 defm VUNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
2607 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2608 SSEPackedDouble>, TB, OpSize, VEX_4V;
2609 defm VUNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
2610 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2611 SSEPackedSingle>, TB, VEX_4V;
2612 defm VUNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
2613 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2614 SSEPackedDouble>, TB, OpSize, VEX_4V;
2616 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, X86Unpckh, v8f32, memopv8f32,
2617 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2618 SSEPackedSingle>, TB, VEX_4V, VEX_L;
2619 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, X86Unpckh, v4f64, memopv4f64,
2620 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2621 SSEPackedDouble>, TB, OpSize, VEX_4V, VEX_L;
2622 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, X86Unpckl, v8f32, memopv8f32,
2623 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2624 SSEPackedSingle>, TB, VEX_4V, VEX_L;
2625 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, X86Unpckl, v4f64, memopv4f64,
2626 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2627 SSEPackedDouble>, TB, OpSize, VEX_4V, VEX_L;
2629 let Constraints = "$src1 = $dst" in {
2630 defm UNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
2631 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
2632 SSEPackedSingle>, TB;
2633 defm UNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
2634 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
2635 SSEPackedDouble>, TB, OpSize;
2636 defm UNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
2637 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
2638 SSEPackedSingle>, TB;
2639 defm UNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
2640 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
2641 SSEPackedDouble>, TB, OpSize;
2642 } // Constraints = "$src1 = $dst"
2644 let Predicates = [HasAVX1Only] in {
2645 def : Pat<(v8i32 (X86Unpckl VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
2646 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
2647 def : Pat<(v8i32 (X86Unpckl VR256:$src1, VR256:$src2)),
2648 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
2649 def : Pat<(v8i32 (X86Unpckh VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
2650 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
2651 def : Pat<(v8i32 (X86Unpckh VR256:$src1, VR256:$src2)),
2652 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
2654 def : Pat<(v4i64 (X86Unpckl VR256:$src1, (memopv4i64 addr:$src2))),
2655 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
2656 def : Pat<(v4i64 (X86Unpckl VR256:$src1, VR256:$src2)),
2657 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
2658 def : Pat<(v4i64 (X86Unpckh VR256:$src1, (memopv4i64 addr:$src2))),
2659 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
2660 def : Pat<(v4i64 (X86Unpckh VR256:$src1, VR256:$src2)),
2661 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
2664 let Predicates = [HasAVX] in {
2665 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2666 // problem is during lowering, where it's not possible to recognize the load
2667 // fold cause it has two uses through a bitcast. One use disappears at isel
2668 // time and the fold opportunity reappears.
2669 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2670 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2673 let Predicates = [UseSSE2] in {
2674 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2675 // problem is during lowering, where it's not possible to recognize the load
2676 // fold cause it has two uses through a bitcast. One use disappears at isel
2677 // time and the fold opportunity reappears.
2678 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2679 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2682 //===----------------------------------------------------------------------===//
2683 // SSE 1 & 2 - Extract Floating-Point Sign mask
2684 //===----------------------------------------------------------------------===//
2686 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2687 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2689 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
2690 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2691 [(set GR32:$dst, (Int RC:$src))], IIC_SSE_MOVMSK, d>,
2692 Sched<[WriteVecLogic]>;
2693 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
2694 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [],
2695 IIC_SSE_MOVMSK, d>, REX_W, Sched<[WriteVecLogic]>;
2698 let Predicates = [HasAVX] in {
2699 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2700 "movmskps", SSEPackedSingle>, TB, VEX;
2701 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2702 "movmskpd", SSEPackedDouble>, TB,
2704 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2705 "movmskps", SSEPackedSingle>, TB,
2707 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2708 "movmskpd", SSEPackedDouble>, TB,
2711 def : Pat<(i32 (X86fgetsign FR32:$src)),
2712 (VMOVMSKPSrr32 (COPY_TO_REGCLASS FR32:$src, VR128))>;
2713 def : Pat<(i64 (X86fgetsign FR32:$src)),
2714 (VMOVMSKPSrr64 (COPY_TO_REGCLASS FR32:$src, VR128))>;
2715 def : Pat<(i32 (X86fgetsign FR64:$src)),
2716 (VMOVMSKPDrr32 (COPY_TO_REGCLASS FR64:$src, VR128))>;
2717 def : Pat<(i64 (X86fgetsign FR64:$src)),
2718 (VMOVMSKPDrr64 (COPY_TO_REGCLASS FR64:$src, VR128))>;
2721 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2722 "movmskps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2723 SSEPackedSingle>, TB, VEX, Sched<[WriteVecLogic]>;
2724 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2725 "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2726 SSEPackedDouble>, TB,
2727 OpSize, VEX, Sched<[WriteVecLogic]>;
2728 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2729 "movmskps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2730 SSEPackedSingle>, TB, VEX, VEX_L, Sched<[WriteVecLogic]>;
2731 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2732 "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2733 SSEPackedDouble>, TB,
2734 OpSize, VEX, VEX_L, Sched<[WriteVecLogic]>;
2737 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2738 SSEPackedSingle>, TB;
2739 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2740 SSEPackedDouble>, TB, OpSize;
2742 def : Pat<(i32 (X86fgetsign FR32:$src)),
2743 (MOVMSKPSrr32 (COPY_TO_REGCLASS FR32:$src, VR128))>,
2744 Requires<[UseSSE1]>;
2745 def : Pat<(i64 (X86fgetsign FR32:$src)),
2746 (MOVMSKPSrr64 (COPY_TO_REGCLASS FR32:$src, VR128))>,
2747 Requires<[UseSSE1]>;
2748 def : Pat<(i32 (X86fgetsign FR64:$src)),
2749 (MOVMSKPDrr32 (COPY_TO_REGCLASS FR64:$src, VR128))>,
2750 Requires<[UseSSE2]>;
2751 def : Pat<(i64 (X86fgetsign FR64:$src)),
2752 (MOVMSKPDrr64 (COPY_TO_REGCLASS FR64:$src, VR128))>,
2753 Requires<[UseSSE2]>;
2755 //===---------------------------------------------------------------------===//
2756 // SSE2 - Packed Integer Logical Instructions
2757 //===---------------------------------------------------------------------===//
2759 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2761 /// PDI_binop_rm - Simple SSE2 binary operator.
2762 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2763 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
2764 X86MemOperand x86memop, OpndItins itins,
2765 bit IsCommutable, bit Is2Addr> {
2766 let isCommutable = IsCommutable in
2767 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
2768 (ins RC:$src1, RC:$src2),
2770 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2771 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2772 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>,
2773 Sched<[itins.Sched]>;
2774 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
2775 (ins RC:$src1, x86memop:$src2),
2777 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2778 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2779 [(set RC:$dst, (OpVT (OpNode RC:$src1,
2780 (bitconvert (memop_frag addr:$src2)))))],
2782 Sched<[itins.Sched.Folded, ReadAfterLd]>;
2784 } // ExeDomain = SSEPackedInt
2786 multiclass PDI_binop_all<bits<8> opc, string OpcodeStr, SDNode Opcode,
2787 ValueType OpVT128, ValueType OpVT256,
2788 OpndItins itins, bit IsCommutable = 0> {
2789 let Predicates = [HasAVX] in
2790 defm V#NAME : PDI_binop_rm<opc, !strconcat("v", OpcodeStr), Opcode, OpVT128,
2791 VR128, memopv2i64, i128mem, itins, IsCommutable, 0>, VEX_4V;
2793 let Constraints = "$src1 = $dst" in
2794 defm NAME : PDI_binop_rm<opc, OpcodeStr, Opcode, OpVT128, VR128,
2795 memopv2i64, i128mem, itins, IsCommutable, 1>;
2797 let Predicates = [HasAVX2] in
2798 defm V#NAME#Y : PDI_binop_rm<opc, !strconcat("v", OpcodeStr), Opcode,
2799 OpVT256, VR256, memopv4i64, i256mem, itins,
2800 IsCommutable, 0>, VEX_4V, VEX_L;
2803 // These are ordered here for pattern ordering requirements with the fp versions
2805 defm PAND : PDI_binop_all<0xDB, "pand", and, v2i64, v4i64, SSE_BIT_ITINS_P, 1>;
2806 defm POR : PDI_binop_all<0xEB, "por", or, v2i64, v4i64, SSE_BIT_ITINS_P, 1>;
2807 defm PXOR : PDI_binop_all<0xEF, "pxor", xor, v2i64, v4i64, SSE_BIT_ITINS_P, 1>;
2808 defm PANDN : PDI_binop_all<0xDF, "pandn", X86andnp, v2i64, v4i64,
2809 SSE_BIT_ITINS_P, 0>;
2811 //===----------------------------------------------------------------------===//
2812 // SSE 1 & 2 - Logical Instructions
2813 //===----------------------------------------------------------------------===//
2815 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2817 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2818 SDNode OpNode, OpndItins itins> {
2819 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2820 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, itins, 0>,
2823 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2824 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, itins, 0>,
2827 let Constraints = "$src1 = $dst" in {
2828 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2829 f32, f128mem, memopfsf32, SSEPackedSingle, itins>,
2832 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2833 f64, f128mem, memopfsf64, SSEPackedDouble, itins>,
2838 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2839 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand,
2841 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for,
2843 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor,
2846 let isCommutable = 0 in
2847 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", X86fandn,
2850 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2852 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2854 defm V#NAME#PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2855 !strconcat(OpcodeStr, "ps"), f256mem,
2856 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2857 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2858 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V, VEX_L;
2860 defm V#NAME#PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2861 !strconcat(OpcodeStr, "pd"), f256mem,
2862 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2863 (bc_v4i64 (v4f64 VR256:$src2))))],
2864 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2865 (memopv4i64 addr:$src2)))], 0>,
2866 TB, OpSize, VEX_4V, VEX_L;
2868 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2869 // are all promoted to v2i64, and the patterns are covered by the int
2870 // version. This is needed in SSE only, because v2i64 isn't supported on
2871 // SSE1, but only on SSE2.
2872 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2873 !strconcat(OpcodeStr, "ps"), f128mem, [],
2874 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2875 (memopv2i64 addr:$src2)))], 0>, TB, VEX_4V;
2877 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2878 !strconcat(OpcodeStr, "pd"), f128mem,
2879 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2880 (bc_v2i64 (v2f64 VR128:$src2))))],
2881 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2882 (memopv2i64 addr:$src2)))], 0>,
2885 let Constraints = "$src1 = $dst" in {
2886 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2887 !strconcat(OpcodeStr, "ps"), f128mem,
2888 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2889 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2890 (memopv2i64 addr:$src2)))]>, TB;
2892 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2893 !strconcat(OpcodeStr, "pd"), f128mem,
2894 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2895 (bc_v2i64 (v2f64 VR128:$src2))))],
2896 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2897 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2901 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2902 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2903 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2904 let isCommutable = 0 in
2905 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2907 //===----------------------------------------------------------------------===//
2908 // SSE 1 & 2 - Arithmetic Instructions
2909 //===----------------------------------------------------------------------===//
2911 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2914 /// In addition, we also have a special variant of the scalar form here to
2915 /// represent the associated intrinsic operation. This form is unlike the
2916 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2917 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2919 /// These three forms can each be reg+reg or reg+mem.
2922 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2924 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr,
2925 SDNode OpNode, SizeItins itins> {
2926 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2927 VR128, v4f32, f128mem, memopv4f32,
2928 SSEPackedSingle, itins.s, 0>, TB, VEX_4V;
2929 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2930 VR128, v2f64, f128mem, memopv2f64,
2931 SSEPackedDouble, itins.d, 0>, TB, OpSize, VEX_4V;
2933 defm V#NAME#PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"),
2934 OpNode, VR256, v8f32, f256mem, memopv8f32,
2935 SSEPackedSingle, itins.s, 0>, TB, VEX_4V, VEX_L;
2936 defm V#NAME#PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"),
2937 OpNode, VR256, v4f64, f256mem, memopv4f64,
2938 SSEPackedDouble, itins.d, 0>, TB, OpSize, VEX_4V, VEX_L;
2940 let Constraints = "$src1 = $dst" in {
2941 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2942 v4f32, f128mem, memopv4f32, SSEPackedSingle,
2944 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2945 v2f64, f128mem, memopv2f64, SSEPackedDouble,
2946 itins.d>, TB, OpSize;
2950 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2952 defm V#NAME#SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2953 OpNode, FR32, f32mem, itins.s, 0>, XS, VEX_4V, VEX_LIG;
2954 defm V#NAME#SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2955 OpNode, FR64, f64mem, itins.d, 0>, XD, VEX_4V, VEX_LIG;
2957 let Constraints = "$src1 = $dst" in {
2958 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2959 OpNode, FR32, f32mem, itins.s>, XS;
2960 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2961 OpNode, FR64, f64mem, itins.d>, XD;
2965 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2967 defm V#NAME#SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2968 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32,
2969 itins.s, 0>, XS, VEX_4V, VEX_LIG;
2970 defm V#NAME#SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2971 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64,
2972 itins.d, 0>, XD, VEX_4V, VEX_LIG;
2974 let Constraints = "$src1 = $dst" in {
2975 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2976 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32,
2978 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2979 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64,
2984 // Binary Arithmetic instructions
2985 defm ADD : basic_sse12_fp_binop_p<0x58, "add", fadd, SSE_ALU_ITINS_P>,
2986 basic_sse12_fp_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S>,
2987 basic_sse12_fp_binop_s_int<0x58, "add", SSE_ALU_ITINS_S>;
2988 defm MUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, SSE_MUL_ITINS_P>,
2989 basic_sse12_fp_binop_s<0x59, "mul", fmul, SSE_MUL_ITINS_S>,
2990 basic_sse12_fp_binop_s_int<0x59, "mul", SSE_MUL_ITINS_S>;
2991 let isCommutable = 0 in {
2992 defm SUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, SSE_ALU_ITINS_P>,
2993 basic_sse12_fp_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S>,
2994 basic_sse12_fp_binop_s_int<0x5C, "sub", SSE_ALU_ITINS_S>;
2995 defm DIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, SSE_DIV_ITINS_P>,
2996 basic_sse12_fp_binop_s<0x5E, "div", fdiv, SSE_DIV_ITINS_S>,
2997 basic_sse12_fp_binop_s_int<0x5E, "div", SSE_DIV_ITINS_S>;
2998 defm MAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, SSE_ALU_ITINS_P>,
2999 basic_sse12_fp_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S>,
3000 basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S>;
3001 defm MIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, SSE_ALU_ITINS_P>,
3002 basic_sse12_fp_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S>,
3003 basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S>;
3006 let isCodeGenOnly = 1 in {
3007 defm MAXC: basic_sse12_fp_binop_p<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_P>,
3008 basic_sse12_fp_binop_s<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_S>;
3009 defm MINC: basic_sse12_fp_binop_p<0x5D, "min", X86fminc, SSE_ALU_ITINS_P>,
3010 basic_sse12_fp_binop_s<0x5D, "min", X86fminc, SSE_ALU_ITINS_S>;
3014 /// In addition, we also have a special variant of the scalar form here to
3015 /// represent the associated intrinsic operation. This form is unlike the
3016 /// plain scalar form, in that it takes an entire vector (instead of a
3017 /// scalar) and leaves the top elements undefined.
3019 /// And, we have a special variant form for a full-vector intrinsic form.
3021 let Sched = WriteFSqrt in {
3022 def SSE_SQRTPS : OpndItins<
3023 IIC_SSE_SQRTPS_RR, IIC_SSE_SQRTPS_RM
3026 def SSE_SQRTSS : OpndItins<
3027 IIC_SSE_SQRTSS_RR, IIC_SSE_SQRTSS_RM
3030 def SSE_SQRTPD : OpndItins<
3031 IIC_SSE_SQRTPD_RR, IIC_SSE_SQRTPD_RM
3034 def SSE_SQRTSD : OpndItins<
3035 IIC_SSE_SQRTSD_RR, IIC_SSE_SQRTSD_RM
3039 let Sched = WriteFRcp in {
3040 def SSE_RCPP : OpndItins<
3041 IIC_SSE_RCPP_RR, IIC_SSE_RCPP_RM
3044 def SSE_RCPS : OpndItins<
3045 IIC_SSE_RCPS_RR, IIC_SSE_RCPS_RM
3049 /// sse1_fp_unop_s - SSE1 unops in scalar form.
3050 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
3051 SDNode OpNode, Intrinsic F32Int, OpndItins itins> {
3052 let Predicates = [HasAVX], hasSideEffects = 0 in {
3053 def V#NAME#SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst),
3054 (ins FR32:$src1, FR32:$src2),
3055 !strconcat("v", OpcodeStr,
3056 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3057 []>, VEX_4V, VEX_LIG, Sched<[itins.Sched]>;
3058 let mayLoad = 1 in {
3059 def V#NAME#SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst),
3060 (ins FR32:$src1,f32mem:$src2),
3061 !strconcat("v", OpcodeStr,
3062 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3063 []>, VEX_4V, VEX_LIG,
3064 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3065 def V#NAME#SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3066 (ins VR128:$src1, ssmem:$src2),
3067 !strconcat("v", OpcodeStr,
3068 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3069 []>, VEX_4V, VEX_LIG,
3070 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3074 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
3075 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3076 [(set FR32:$dst, (OpNode FR32:$src))]>, Sched<[itins.Sched]>;
3077 // For scalar unary operations, fold a load into the operation
3078 // only in OptForSize mode. It eliminates an instruction, but it also
3079 // eliminates a whole-register clobber (the load), so it introduces a
3080 // partial register update condition.
3081 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
3082 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3083 [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
3084 Requires<[UseSSE1, OptForSize]>, Sched<[itins.Sched.Folded]>;
3085 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3086 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3087 [(set VR128:$dst, (F32Int VR128:$src))], itins.rr>,
3088 Sched<[itins.Sched]>;
3089 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
3090 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3091 [(set VR128:$dst, (F32Int sse_load_f32:$src))], itins.rm>,
3092 Sched<[itins.Sched.Folded]>;
3095 /// sse1_fp_unop_s_rw - SSE1 unops where vector form has a read-write operand.
3096 multiclass sse1_fp_unop_rw<bits<8> opc, string OpcodeStr, SDNode OpNode,
3098 let Predicates = [HasAVX], hasSideEffects = 0 in {
3099 def V#NAME#SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst),
3100 (ins FR32:$src1, FR32:$src2),
3101 !strconcat("v", OpcodeStr,
3102 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3103 []>, VEX_4V, VEX_LIG, Sched<[itins.Sched]>;
3104 let mayLoad = 1 in {
3105 def V#NAME#SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst),
3106 (ins FR32:$src1,f32mem:$src2),
3107 !strconcat("v", OpcodeStr,
3108 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3109 []>, VEX_4V, VEX_LIG,
3110 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3111 def V#NAME#SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3112 (ins VR128:$src1, ssmem:$src2),
3113 !strconcat("v", OpcodeStr,
3114 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3115 []>, VEX_4V, VEX_LIG,
3116 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3120 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
3121 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3122 [(set FR32:$dst, (OpNode FR32:$src))]>, Sched<[itins.Sched]>;
3123 // For scalar unary operations, fold a load into the operation
3124 // only in OptForSize mode. It eliminates an instruction, but it also
3125 // eliminates a whole-register clobber (the load), so it introduces a
3126 // partial register update condition.
3127 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
3128 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3129 [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
3130 Requires<[UseSSE1, OptForSize]>, Sched<[itins.Sched.Folded]>;
3131 let Constraints = "$src1 = $dst" in {
3132 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst),
3133 (ins VR128:$src1, VR128:$src2),
3134 !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
3135 [], itins.rr>, Sched<[itins.Sched]>;
3136 let mayLoad = 1, hasSideEffects = 0 in
3137 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3138 (ins VR128:$src1, ssmem:$src2),
3139 !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
3140 [], itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
3144 /// sse1_fp_unop_p - SSE1 unops in packed form.
3145 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
3147 let Predicates = [HasAVX] in {
3148 def V#NAME#PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3149 !strconcat("v", OpcodeStr,
3150 "ps\t{$src, $dst|$dst, $src}"),
3151 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))],
3152 itins.rr>, VEX, Sched<[itins.Sched]>;
3153 def V#NAME#PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3154 !strconcat("v", OpcodeStr,
3155 "ps\t{$src, $dst|$dst, $src}"),
3156 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))],
3157 itins.rm>, VEX, Sched<[itins.Sched.Folded]>;
3158 def V#NAME#PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3159 !strconcat("v", OpcodeStr,
3160 "ps\t{$src, $dst|$dst, $src}"),
3161 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))],
3162 itins.rr>, VEX, VEX_L, Sched<[itins.Sched]>;
3163 def V#NAME#PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3164 !strconcat("v", OpcodeStr,
3165 "ps\t{$src, $dst|$dst, $src}"),
3166 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))],
3167 itins.rm>, VEX, VEX_L, Sched<[itins.Sched.Folded]>;
3170 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3171 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3172 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))], itins.rr>,
3173 Sched<[itins.Sched]>;
3174 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3175 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3176 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))], itins.rm>,
3177 Sched<[itins.Sched.Folded]>;
3180 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
3181 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3182 Intrinsic V4F32Int, Intrinsic V8F32Int,
3184 let Predicates = [HasAVX] in {
3185 def V#NAME#PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3186 !strconcat("v", OpcodeStr,
3187 "ps\t{$src, $dst|$dst, $src}"),
3188 [(set VR128:$dst, (V4F32Int VR128:$src))],
3189 itins.rr>, VEX, Sched<[itins.Sched]>;
3190 def V#NAME#PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3191 !strconcat("v", OpcodeStr,
3192 "ps\t{$src, $dst|$dst, $src}"),
3193 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))],
3194 itins.rm>, VEX, Sched<[itins.Sched.Folded]>;
3195 def V#NAME#PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3196 !strconcat("v", OpcodeStr,
3197 "ps\t{$src, $dst|$dst, $src}"),
3198 [(set VR256:$dst, (V8F32Int VR256:$src))],
3199 itins.rr>, VEX, VEX_L, Sched<[itins.Sched]>;
3200 def V#NAME#PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst),
3202 !strconcat("v", OpcodeStr,
3203 "ps\t{$src, $dst|$dst, $src}"),
3204 [(set VR256:$dst, (V8F32Int (memopv8f32 addr:$src)))],
3205 itins.rm>, VEX, VEX_L, Sched<[itins.Sched.Folded]>;
3208 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3209 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3210 [(set VR128:$dst, (V4F32Int VR128:$src))],
3211 itins.rr>, Sched<[itins.Sched]>;
3212 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3213 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3214 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))],
3215 itins.rm>, Sched<[itins.Sched.Folded]>;
3218 /// sse2_fp_unop_s - SSE2 unops in scalar form.
3219 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
3220 SDNode OpNode, Intrinsic F64Int, OpndItins itins> {
3221 let Predicates = [HasAVX], hasSideEffects = 0 in {
3222 def V#NAME#SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst),
3223 (ins FR64:$src1, FR64:$src2),
3224 !strconcat("v", OpcodeStr,
3225 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3226 []>, VEX_4V, VEX_LIG, Sched<[itins.Sched]>;
3227 let mayLoad = 1 in {
3228 def V#NAME#SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst),
3229 (ins FR64:$src1,f64mem:$src2),
3230 !strconcat("v", OpcodeStr,
3231 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3232 []>, VEX_4V, VEX_LIG,
3233 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3234 def V#NAME#SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
3235 (ins VR128:$src1, sdmem:$src2),
3236 !strconcat("v", OpcodeStr,
3237 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3238 []>, VEX_4V, VEX_LIG,
3239 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3243 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
3244 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3245 [(set FR64:$dst, (OpNode FR64:$src))], itins.rr>,
3246 Sched<[itins.Sched]>;
3247 // See the comments in sse1_fp_unop_s for why this is OptForSize.
3248 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
3249 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3250 [(set FR64:$dst, (OpNode (load addr:$src)))], itins.rm>, XD,
3251 Requires<[UseSSE2, OptForSize]>, Sched<[itins.Sched.Folded]>;
3252 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3253 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3254 [(set VR128:$dst, (F64Int VR128:$src))], itins.rr>,
3255 Sched<[itins.Sched]>;
3256 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
3257 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3258 [(set VR128:$dst, (F64Int sse_load_f64:$src))], itins.rm>,
3259 Sched<[itins.Sched.Folded]>;
3262 /// sse2_fp_unop_p - SSE2 unops in vector forms.
3263 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
3264 SDNode OpNode, OpndItins itins> {
3265 let Predicates = [HasAVX] in {
3266 def V#NAME#PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3267 !strconcat("v", OpcodeStr,
3268 "pd\t{$src, $dst|$dst, $src}"),
3269 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))],
3270 itins.rr>, VEX, Sched<[itins.Sched]>;
3271 def V#NAME#PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3272 !strconcat("v", OpcodeStr,
3273 "pd\t{$src, $dst|$dst, $src}"),
3274 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))],
3275 itins.rm>, VEX, Sched<[itins.Sched.Folded]>;
3276 def V#NAME#PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3277 !strconcat("v", OpcodeStr,
3278 "pd\t{$src, $dst|$dst, $src}"),
3279 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))],
3280 itins.rr>, VEX, VEX_L, Sched<[itins.Sched]>;
3281 def V#NAME#PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3282 !strconcat("v", OpcodeStr,
3283 "pd\t{$src, $dst|$dst, $src}"),
3284 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))],
3285 itins.rm>, VEX, VEX_L, Sched<[itins.Sched.Folded]>;
3288 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3289 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3290 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))], itins.rr>,
3291 Sched<[itins.Sched]>;
3292 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3293 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3294 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))], itins.rm>,
3295 Sched<[itins.Sched.Folded]>;
3299 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss,
3301 sse1_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTPS>,
3302 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd,
3304 sse2_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTPD>;
3306 // Reciprocal approximations. Note that these typically require refinement
3307 // in order to obtain suitable precision.
3308 defm RSQRT : sse1_fp_unop_rw<0x52, "rsqrt", X86frsqrt, SSE_SQRTSS>,
3309 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, SSE_SQRTPS>,
3310 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps,
3311 int_x86_avx_rsqrt_ps_256, SSE_SQRTPS>;
3312 defm RCP : sse1_fp_unop_rw<0x53, "rcp", X86frcp, SSE_RCPS>,
3313 sse1_fp_unop_p<0x53, "rcp", X86frcp, SSE_RCPP>,
3314 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps,
3315 int_x86_avx_rcp_ps_256, SSE_RCPP>;
3317 def : Pat<(f32 (fsqrt FR32:$src)),
3318 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3319 def : Pat<(f32 (fsqrt (load addr:$src))),
3320 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3321 Requires<[HasAVX, OptForSize]>;
3322 def : Pat<(f64 (fsqrt FR64:$src)),
3323 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
3324 def : Pat<(f64 (fsqrt (load addr:$src))),
3325 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
3326 Requires<[HasAVX, OptForSize]>;
3328 def : Pat<(f32 (X86frsqrt FR32:$src)),
3329 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3330 def : Pat<(f32 (X86frsqrt (load addr:$src))),
3331 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3332 Requires<[HasAVX, OptForSize]>;
3334 def : Pat<(f32 (X86frcp FR32:$src)),
3335 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3336 def : Pat<(f32 (X86frcp (load addr:$src))),
3337 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3338 Requires<[HasAVX, OptForSize]>;
3340 let Predicates = [HasAVX] in {
3341 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
3342 (COPY_TO_REGCLASS (VSQRTSSr (f32 (IMPLICIT_DEF)),
3343 (COPY_TO_REGCLASS VR128:$src, FR32)),
3345 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
3346 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3348 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
3349 (COPY_TO_REGCLASS (VSQRTSDr (f64 (IMPLICIT_DEF)),
3350 (COPY_TO_REGCLASS VR128:$src, FR64)),
3352 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
3353 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
3355 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3356 (COPY_TO_REGCLASS (VRSQRTSSr (f32 (IMPLICIT_DEF)),
3357 (COPY_TO_REGCLASS VR128:$src, FR32)),
3359 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
3360 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3362 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3363 (COPY_TO_REGCLASS (VRCPSSr (f32 (IMPLICIT_DEF)),
3364 (COPY_TO_REGCLASS VR128:$src, FR32)),
3366 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
3367 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3370 // Reciprocal approximations. Note that these typically require refinement
3371 // in order to obtain suitable precision.
3372 let Predicates = [UseSSE1] in {
3373 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3374 (RSQRTSSr_Int VR128:$src, VR128:$src)>;
3375 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3376 (RCPSSr_Int VR128:$src, VR128:$src)>;
3379 // There is no f64 version of the reciprocal approximation instructions.
3381 //===----------------------------------------------------------------------===//
3382 // SSE 1 & 2 - Non-temporal stores
3383 //===----------------------------------------------------------------------===//
3385 let AddedComplexity = 400 in { // Prefer non-temporal versions
3386 let SchedRW = [WriteStore] in {
3387 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3388 (ins f128mem:$dst, VR128:$src),
3389 "movntps\t{$src, $dst|$dst, $src}",
3390 [(alignednontemporalstore (v4f32 VR128:$src),
3392 IIC_SSE_MOVNT>, VEX;
3393 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3394 (ins f128mem:$dst, VR128:$src),
3395 "movntpd\t{$src, $dst|$dst, $src}",
3396 [(alignednontemporalstore (v2f64 VR128:$src),
3398 IIC_SSE_MOVNT>, VEX;
3400 let ExeDomain = SSEPackedInt in
3401 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3402 (ins f128mem:$dst, VR128:$src),
3403 "movntdq\t{$src, $dst|$dst, $src}",
3404 [(alignednontemporalstore (v2i64 VR128:$src),
3406 IIC_SSE_MOVNT>, VEX;
3408 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3409 (ins f256mem:$dst, VR256:$src),
3410 "movntps\t{$src, $dst|$dst, $src}",
3411 [(alignednontemporalstore (v8f32 VR256:$src),
3413 IIC_SSE_MOVNT>, VEX, VEX_L;
3414 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3415 (ins f256mem:$dst, VR256:$src),
3416 "movntpd\t{$src, $dst|$dst, $src}",
3417 [(alignednontemporalstore (v4f64 VR256:$src),
3419 IIC_SSE_MOVNT>, VEX, VEX_L;
3420 let ExeDomain = SSEPackedInt in
3421 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3422 (ins f256mem:$dst, VR256:$src),
3423 "movntdq\t{$src, $dst|$dst, $src}",
3424 [(alignednontemporalstore (v4i64 VR256:$src),
3426 IIC_SSE_MOVNT>, VEX, VEX_L;
3428 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3429 "movntps\t{$src, $dst|$dst, $src}",
3430 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)],
3432 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3433 "movntpd\t{$src, $dst|$dst, $src}",
3434 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)],
3437 let ExeDomain = SSEPackedInt in
3438 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3439 "movntdq\t{$src, $dst|$dst, $src}",
3440 [(alignednontemporalstore (v2i64 VR128:$src), addr:$dst)],
3443 // There is no AVX form for instructions below this point
3444 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3445 "movnti{l}\t{$src, $dst|$dst, $src}",
3446 [(nontemporalstore (i32 GR32:$src), addr:$dst)],
3448 TB, Requires<[HasSSE2]>;
3449 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3450 "movnti{q}\t{$src, $dst|$dst, $src}",
3451 [(nontemporalstore (i64 GR64:$src), addr:$dst)],
3453 TB, Requires<[HasSSE2]>;
3454 } // SchedRW = [WriteStore]
3456 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3457 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3459 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3460 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[UseSSE2]>;
3461 } // AddedComplexity
3463 //===----------------------------------------------------------------------===//
3464 // SSE 1 & 2 - Prefetch and memory fence
3465 //===----------------------------------------------------------------------===//
3467 // Prefetch intrinsic.
3468 let Predicates = [HasSSE1], SchedRW = [WriteLoad] in {
3469 def PREFETCHT0 : I<0x18, MRM1m, (outs), (ins i8mem:$src),
3470 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))],
3471 IIC_SSE_PREFETCH>, TB;
3472 def PREFETCHT1 : I<0x18, MRM2m, (outs), (ins i8mem:$src),
3473 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))],
3474 IIC_SSE_PREFETCH>, TB;
3475 def PREFETCHT2 : I<0x18, MRM3m, (outs), (ins i8mem:$src),
3476 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))],
3477 IIC_SSE_PREFETCH>, TB;
3478 def PREFETCHNTA : I<0x18, MRM0m, (outs), (ins i8mem:$src),
3479 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))],
3480 IIC_SSE_PREFETCH>, TB;
3483 // FIXME: How should these memory instructions be modeled?
3484 let SchedRW = [WriteLoad] in {
3486 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3487 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)],
3488 IIC_SSE_PREFETCH>, TB, Requires<[HasSSE2]>;
3490 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3491 // was introduced with SSE2, it's backward compatible.
3492 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", [], IIC_SSE_PAUSE>, REP;
3494 // Load, store, and memory fence
3495 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3496 "sfence", [(int_x86_sse_sfence)], IIC_SSE_SFENCE>,
3497 TB, Requires<[HasSSE1]>;
3498 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3499 "lfence", [(int_x86_sse2_lfence)], IIC_SSE_LFENCE>,
3500 TB, Requires<[HasSSE2]>;
3501 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3502 "mfence", [(int_x86_sse2_mfence)], IIC_SSE_MFENCE>,
3503 TB, Requires<[HasSSE2]>;
3506 def : Pat<(X86SFence), (SFENCE)>;
3507 def : Pat<(X86LFence), (LFENCE)>;
3508 def : Pat<(X86MFence), (MFENCE)>;
3510 //===----------------------------------------------------------------------===//
3511 // SSE 1 & 2 - Load/Store XCSR register
3512 //===----------------------------------------------------------------------===//
3514 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3515 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3516 IIC_SSE_LDMXCSR>, VEX, Sched<[WriteLoad]>;
3517 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3518 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3519 IIC_SSE_STMXCSR>, VEX, Sched<[WriteStore]>;
3521 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3522 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3523 IIC_SSE_LDMXCSR>, Sched<[WriteLoad]>;
3524 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3525 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3526 IIC_SSE_STMXCSR>, Sched<[WriteStore]>;
3528 //===---------------------------------------------------------------------===//
3529 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3530 //===---------------------------------------------------------------------===//
3532 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3534 let neverHasSideEffects = 1, SchedRW = [WriteMove] in {
3535 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3536 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3538 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3539 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3541 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3542 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3544 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3545 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3550 let isCodeGenOnly = 1, hasSideEffects = 0, SchedRW = [WriteMove] in {
3551 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3552 "movdqa\t{$src, $dst|$dst, $src}", [],
3555 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3556 "movdqa\t{$src, $dst|$dst, $src}", [],
3557 IIC_SSE_MOVA_P_RR>, VEX, VEX_L;
3558 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3559 "movdqu\t{$src, $dst|$dst, $src}", [],
3562 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3563 "movdqu\t{$src, $dst|$dst, $src}", [],
3564 IIC_SSE_MOVU_P_RR>, VEX, VEX_L;
3567 let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1,
3568 neverHasSideEffects = 1, SchedRW = [WriteLoad] in {
3569 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3570 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3572 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3573 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3575 let Predicates = [HasAVX] in {
3576 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3577 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3579 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3580 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3585 let mayStore = 1, neverHasSideEffects = 1, SchedRW = [WriteStore] in {
3586 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3587 (ins i128mem:$dst, VR128:$src),
3588 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3590 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3591 (ins i256mem:$dst, VR256:$src),
3592 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3594 let Predicates = [HasAVX] in {
3595 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3596 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3598 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3599 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3604 let SchedRW = [WriteMove] in {
3605 let neverHasSideEffects = 1 in
3606 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3607 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>;
3609 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3610 "movdqu\t{$src, $dst|$dst, $src}",
3611 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>;
3614 let isCodeGenOnly = 1, hasSideEffects = 0 in {
3615 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3616 "movdqa\t{$src, $dst|$dst, $src}", [],
3619 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3620 "movdqu\t{$src, $dst|$dst, $src}",
3621 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>;
3625 let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1,
3626 neverHasSideEffects = 1, SchedRW = [WriteLoad] in {
3627 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3628 "movdqa\t{$src, $dst|$dst, $src}",
3629 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/],
3631 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3632 "movdqu\t{$src, $dst|$dst, $src}",
3633 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/],
3635 XS, Requires<[UseSSE2]>;
3638 let mayStore = 1, SchedRW = [WriteStore] in {
3639 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3640 "movdqa\t{$src, $dst|$dst, $src}",
3641 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/],
3643 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3644 "movdqu\t{$src, $dst|$dst, $src}",
3645 [/*(store (v2i64 VR128:$src), addr:$dst)*/],
3647 XS, Requires<[UseSSE2]>;
3650 } // ExeDomain = SSEPackedInt
3652 let Predicates = [HasAVX] in {
3653 def : Pat<(int_x86_sse2_storeu_dq addr:$dst, VR128:$src),
3654 (VMOVDQUmr addr:$dst, VR128:$src)>;
3655 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3656 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3658 let Predicates = [UseSSE2] in
3659 def : Pat<(int_x86_sse2_storeu_dq addr:$dst, VR128:$src),
3660 (MOVDQUmr addr:$dst, VR128:$src)>;
3662 //===---------------------------------------------------------------------===//
3663 // SSE2 - Packed Integer Arithmetic Instructions
3664 //===---------------------------------------------------------------------===//
3666 let Sched = WriteVecIMul in
3667 def SSE_PMADD : OpndItins<
3668 IIC_SSE_PMADD, IIC_SSE_PMADD
3671 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3673 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3674 RegisterClass RC, PatFrag memop_frag,
3675 X86MemOperand x86memop,
3677 bit IsCommutable = 0,
3679 let isCommutable = IsCommutable in
3680 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3681 (ins RC:$src1, RC:$src2),
3683 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3684 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3685 [(set RC:$dst, (IntId RC:$src1, RC:$src2))], itins.rr>,
3686 Sched<[itins.Sched]>;
3687 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3688 (ins RC:$src1, x86memop:$src2),
3690 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3691 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3692 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))],
3693 itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
3696 multiclass PDI_binop_all_int<bits<8> opc, string OpcodeStr, Intrinsic IntId128,
3697 Intrinsic IntId256, OpndItins itins,
3698 bit IsCommutable = 0> {
3699 let Predicates = [HasAVX] in
3700 defm V#NAME : PDI_binop_rm_int<opc, !strconcat("v", OpcodeStr), IntId128,
3701 VR128, memopv2i64, i128mem, itins,
3702 IsCommutable, 0>, VEX_4V;
3704 let Constraints = "$src1 = $dst" in
3705 defm NAME : PDI_binop_rm_int<opc, OpcodeStr, IntId128, VR128, memopv2i64,
3706 i128mem, itins, IsCommutable, 1>;
3708 let Predicates = [HasAVX2] in
3709 defm V#NAME#Y : PDI_binop_rm_int<opc, !strconcat("v", OpcodeStr), IntId256,
3710 VR256, memopv4i64, i256mem, itins,
3711 IsCommutable, 0>, VEX_4V, VEX_L;
3714 multiclass PDI_binop_rmi<bits<8> opc, bits<8> opc2, Format ImmForm,
3715 string OpcodeStr, SDNode OpNode,
3716 SDNode OpNode2, RegisterClass RC,
3717 ValueType DstVT, ValueType SrcVT, PatFrag bc_frag,
3718 ShiftOpndItins itins,
3720 // src2 is always 128-bit
3721 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3722 (ins RC:$src1, VR128:$src2),
3724 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3725 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3726 [(set RC:$dst, (DstVT (OpNode RC:$src1, (SrcVT VR128:$src2))))],
3727 itins.rr>, Sched<[WriteVecShift]>;
3728 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3729 (ins RC:$src1, i128mem:$src2),
3731 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3732 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3733 [(set RC:$dst, (DstVT (OpNode RC:$src1,
3734 (bc_frag (memopv2i64 addr:$src2)))))], itins.rm>,
3735 Sched<[WriteVecShiftLd, ReadAfterLd]>;
3736 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3737 (ins RC:$src1, i32i8imm:$src2),
3739 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3740 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3741 [(set RC:$dst, (DstVT (OpNode2 RC:$src1, (i32 imm:$src2))))], itins.ri>,
3742 Sched<[WriteVecShift]>;
3745 /// PDI_binop_rm2 - Simple SSE2 binary operator with different src and dst types
3746 multiclass PDI_binop_rm2<bits<8> opc, string OpcodeStr, SDNode OpNode,
3747 ValueType DstVT, ValueType SrcVT, RegisterClass RC,
3748 PatFrag memop_frag, X86MemOperand x86memop,
3750 bit IsCommutable = 0, bit Is2Addr = 1> {
3751 let isCommutable = IsCommutable in
3752 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3753 (ins RC:$src1, RC:$src2),
3755 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3756 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3757 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1), RC:$src2)))]>,
3758 Sched<[itins.Sched]>;
3759 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3760 (ins RC:$src1, x86memop:$src2),
3762 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3763 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3764 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1),
3765 (bitconvert (memop_frag addr:$src2)))))]>,
3766 Sched<[itins.Sched.Folded, ReadAfterLd]>;
3768 } // ExeDomain = SSEPackedInt
3770 defm PADDB : PDI_binop_all<0xFC, "paddb", add, v16i8, v32i8,
3771 SSE_INTALU_ITINS_P, 1>;
3772 defm PADDW : PDI_binop_all<0xFD, "paddw", add, v8i16, v16i16,
3773 SSE_INTALU_ITINS_P, 1>;
3774 defm PADDD : PDI_binop_all<0xFE, "paddd", add, v4i32, v8i32,
3775 SSE_INTALU_ITINS_P, 1>;
3776 defm PADDQ : PDI_binop_all<0xD4, "paddq", add, v2i64, v4i64,
3777 SSE_INTALUQ_ITINS_P, 1>;
3778 defm PMULLW : PDI_binop_all<0xD5, "pmullw", mul, v8i16, v16i16,
3779 SSE_INTMUL_ITINS_P, 1>;
3780 defm PSUBB : PDI_binop_all<0xF8, "psubb", sub, v16i8, v32i8,
3781 SSE_INTALU_ITINS_P, 0>;
3782 defm PSUBW : PDI_binop_all<0xF9, "psubw", sub, v8i16, v16i16,
3783 SSE_INTALU_ITINS_P, 0>;
3784 defm PSUBD : PDI_binop_all<0xFA, "psubd", sub, v4i32, v8i32,
3785 SSE_INTALU_ITINS_P, 0>;
3786 defm PSUBQ : PDI_binop_all<0xFB, "psubq", sub, v2i64, v4i64,
3787 SSE_INTALUQ_ITINS_P, 0>;
3788 defm PSUBUSB : PDI_binop_all<0xD8, "psubusb", X86subus, v16i8, v32i8,
3789 SSE_INTALU_ITINS_P, 0>;
3790 defm PSUBUSW : PDI_binop_all<0xD9, "psubusw", X86subus, v8i16, v16i16,
3791 SSE_INTALU_ITINS_P, 0>;
3792 defm PMINUB : PDI_binop_all<0xDA, "pminub", X86umin, v16i8, v32i8,
3793 SSE_INTALU_ITINS_P, 1>;
3794 defm PMINSW : PDI_binop_all<0xEA, "pminsw", X86smin, v8i16, v16i16,
3795 SSE_INTALU_ITINS_P, 1>;
3796 defm PMAXUB : PDI_binop_all<0xDE, "pmaxub", X86umax, v16i8, v32i8,
3797 SSE_INTALU_ITINS_P, 1>;
3798 defm PMAXSW : PDI_binop_all<0xEE, "pmaxsw", X86smax, v8i16, v16i16,
3799 SSE_INTALU_ITINS_P, 1>;
3802 defm PSUBSB : PDI_binop_all_int<0xE8, "psubsb", int_x86_sse2_psubs_b,
3803 int_x86_avx2_psubs_b, SSE_INTALU_ITINS_P, 0>;
3804 defm PSUBSW : PDI_binop_all_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
3805 int_x86_avx2_psubs_w, SSE_INTALU_ITINS_P, 0>;
3806 defm PADDSB : PDI_binop_all_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
3807 int_x86_avx2_padds_b, SSE_INTALU_ITINS_P, 1>;
3808 defm PADDSW : PDI_binop_all_int<0xED, "paddsw" , int_x86_sse2_padds_w,
3809 int_x86_avx2_padds_w, SSE_INTALU_ITINS_P, 1>;
3810 defm PADDUSB : PDI_binop_all_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
3811 int_x86_avx2_paddus_b, SSE_INTALU_ITINS_P, 1>;
3812 defm PADDUSW : PDI_binop_all_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
3813 int_x86_avx2_paddus_w, SSE_INTALU_ITINS_P, 1>;
3814 defm PMULHUW : PDI_binop_all_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
3815 int_x86_avx2_pmulhu_w, SSE_INTMUL_ITINS_P, 1>;
3816 defm PMULHW : PDI_binop_all_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
3817 int_x86_avx2_pmulh_w, SSE_INTMUL_ITINS_P, 1>;
3818 defm PMADDWD : PDI_binop_all_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
3819 int_x86_avx2_pmadd_wd, SSE_PMADD, 1>;
3820 defm PAVGB : PDI_binop_all_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
3821 int_x86_avx2_pavg_b, SSE_INTALU_ITINS_P, 1>;
3822 defm PAVGW : PDI_binop_all_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
3823 int_x86_avx2_pavg_w, SSE_INTALU_ITINS_P, 1>;
3824 defm PSADBW : PDI_binop_all_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
3825 int_x86_avx2_psad_bw, SSE_INTALU_ITINS_P, 1>;
3827 let Predicates = [HasAVX] in
3828 defm VPMULUDQ : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v2i64, v4i32, VR128,
3829 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1, 0>,
3831 let Predicates = [HasAVX2] in
3832 defm VPMULUDQY : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v4i64, v8i32,
3833 VR256, memopv4i64, i256mem,
3834 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V, VEX_L;
3835 let Constraints = "$src1 = $dst" in
3836 defm PMULUDQ : PDI_binop_rm2<0xF4, "pmuludq", X86pmuludq, v2i64, v4i32, VR128,
3837 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1>;
3839 //===---------------------------------------------------------------------===//
3840 // SSE2 - Packed Integer Logical Instructions
3841 //===---------------------------------------------------------------------===//
3843 let Predicates = [HasAVX] in {
3844 defm VPSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3845 VR128, v8i16, v8i16, bc_v8i16,
3846 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3847 defm VPSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3848 VR128, v4i32, v4i32, bc_v4i32,
3849 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3850 defm VPSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3851 VR128, v2i64, v2i64, bc_v2i64,
3852 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3854 defm VPSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3855 VR128, v8i16, v8i16, bc_v8i16,
3856 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3857 defm VPSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3858 VR128, v4i32, v4i32, bc_v4i32,
3859 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3860 defm VPSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3861 VR128, v2i64, v2i64, bc_v2i64,
3862 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3864 defm VPSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3865 VR128, v8i16, v8i16, bc_v8i16,
3866 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3867 defm VPSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3868 VR128, v4i32, v4i32, bc_v4i32,
3869 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3871 let ExeDomain = SSEPackedInt, SchedRW = [WriteVecShift] in {
3872 // 128-bit logical shifts.
3873 def VPSLLDQri : PDIi8<0x73, MRM7r,
3874 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3875 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3877 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>,
3879 def VPSRLDQri : PDIi8<0x73, MRM3r,
3880 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3881 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3883 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>,
3885 // PSRADQri doesn't exist in SSE[1-3].
3887 } // Predicates = [HasAVX]
3889 let Predicates = [HasAVX2] in {
3890 defm VPSLLWY : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3891 VR256, v16i16, v8i16, bc_v8i16,
3892 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3893 defm VPSLLDY : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3894 VR256, v8i32, v4i32, bc_v4i32,
3895 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3896 defm VPSLLQY : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3897 VR256, v4i64, v2i64, bc_v2i64,
3898 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3900 defm VPSRLWY : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3901 VR256, v16i16, v8i16, bc_v8i16,
3902 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3903 defm VPSRLDY : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3904 VR256, v8i32, v4i32, bc_v4i32,
3905 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3906 defm VPSRLQY : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3907 VR256, v4i64, v2i64, bc_v2i64,
3908 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3910 defm VPSRAWY : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3911 VR256, v16i16, v8i16, bc_v8i16,
3912 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3913 defm VPSRADY : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3914 VR256, v8i32, v4i32, bc_v4i32,
3915 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3917 let ExeDomain = SSEPackedInt, SchedRW = [WriteVecShift] in {
3918 // 256-bit logical shifts.
3919 def VPSLLDQYri : PDIi8<0x73, MRM7r,
3920 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3921 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3923 (int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2))]>,
3925 def VPSRLDQYri : PDIi8<0x73, MRM3r,
3926 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3927 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3929 (int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2))]>,
3931 // PSRADQYri doesn't exist in SSE[1-3].
3933 } // Predicates = [HasAVX2]
3935 let Constraints = "$src1 = $dst" in {
3936 defm PSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "psllw", X86vshl, X86vshli,
3937 VR128, v8i16, v8i16, bc_v8i16,
3938 SSE_INTSHIFT_ITINS_P>;
3939 defm PSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "pslld", X86vshl, X86vshli,
3940 VR128, v4i32, v4i32, bc_v4i32,
3941 SSE_INTSHIFT_ITINS_P>;
3942 defm PSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "psllq", X86vshl, X86vshli,
3943 VR128, v2i64, v2i64, bc_v2i64,
3944 SSE_INTSHIFT_ITINS_P>;
3946 defm PSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "psrlw", X86vsrl, X86vsrli,
3947 VR128, v8i16, v8i16, bc_v8i16,
3948 SSE_INTSHIFT_ITINS_P>;
3949 defm PSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "psrld", X86vsrl, X86vsrli,
3950 VR128, v4i32, v4i32, bc_v4i32,
3951 SSE_INTSHIFT_ITINS_P>;
3952 defm PSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "psrlq", X86vsrl, X86vsrli,
3953 VR128, v2i64, v2i64, bc_v2i64,
3954 SSE_INTSHIFT_ITINS_P>;
3956 defm PSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "psraw", X86vsra, X86vsrai,
3957 VR128, v8i16, v8i16, bc_v8i16,
3958 SSE_INTSHIFT_ITINS_P>;
3959 defm PSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "psrad", X86vsra, X86vsrai,
3960 VR128, v4i32, v4i32, bc_v4i32,
3961 SSE_INTSHIFT_ITINS_P>;
3963 let ExeDomain = SSEPackedInt, SchedRW = [WriteVecShift] in {
3964 // 128-bit logical shifts.
3965 def PSLLDQri : PDIi8<0x73, MRM7r,
3966 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3967 "pslldq\t{$src2, $dst|$dst, $src2}",
3969 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>;
3970 def PSRLDQri : PDIi8<0x73, MRM3r,
3971 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3972 "psrldq\t{$src2, $dst|$dst, $src2}",
3974 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>;
3975 // PSRADQri doesn't exist in SSE[1-3].
3977 } // Constraints = "$src1 = $dst"
3979 let Predicates = [HasAVX] in {
3980 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3981 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3982 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3983 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3984 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3985 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3987 // Shift up / down and insert zero's.
3988 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
3989 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3990 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
3991 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3994 let Predicates = [HasAVX2] in {
3995 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
3996 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3997 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
3998 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
4001 let Predicates = [UseSSE2] in {
4002 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
4003 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4004 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
4005 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4006 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
4007 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4009 // Shift up / down and insert zero's.
4010 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
4011 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4012 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
4013 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4016 //===---------------------------------------------------------------------===//
4017 // SSE2 - Packed Integer Comparison Instructions
4018 //===---------------------------------------------------------------------===//
4020 defm PCMPEQB : PDI_binop_all<0x74, "pcmpeqb", X86pcmpeq, v16i8, v32i8,
4021 SSE_INTALU_ITINS_P, 1>;
4022 defm PCMPEQW : PDI_binop_all<0x75, "pcmpeqw", X86pcmpeq, v8i16, v16i16,
4023 SSE_INTALU_ITINS_P, 1>;
4024 defm PCMPEQD : PDI_binop_all<0x76, "pcmpeqd", X86pcmpeq, v4i32, v8i32,
4025 SSE_INTALU_ITINS_P, 1>;
4026 defm PCMPGTB : PDI_binop_all<0x64, "pcmpgtb", X86pcmpgt, v16i8, v32i8,
4027 SSE_INTALU_ITINS_P, 0>;
4028 defm PCMPGTW : PDI_binop_all<0x65, "pcmpgtw", X86pcmpgt, v8i16, v16i16,
4029 SSE_INTALU_ITINS_P, 0>;
4030 defm PCMPGTD : PDI_binop_all<0x66, "pcmpgtd", X86pcmpgt, v4i32, v8i32,
4031 SSE_INTALU_ITINS_P, 0>;
4033 //===---------------------------------------------------------------------===//
4034 // SSE2 - Packed Integer Pack Instructions
4035 //===---------------------------------------------------------------------===//
4037 defm PACKSSWB : PDI_binop_all_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
4038 int_x86_avx2_packsswb, SSE_INTALU_ITINS_P, 0>;
4039 defm PACKSSDW : PDI_binop_all_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
4040 int_x86_avx2_packssdw, SSE_INTALU_ITINS_P, 0>;
4041 defm PACKUSWB : PDI_binop_all_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
4042 int_x86_avx2_packuswb, SSE_INTALU_ITINS_P, 0>;
4044 //===---------------------------------------------------------------------===//
4045 // SSE2 - Packed Integer Shuffle Instructions
4046 //===---------------------------------------------------------------------===//
4048 let ExeDomain = SSEPackedInt in {
4049 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt128, ValueType vt256,
4051 let Predicates = [HasAVX] in {
4052 def V#NAME#ri : Ii8<0x70, MRMSrcReg, (outs VR128:$dst),
4053 (ins VR128:$src1, i8imm:$src2),
4054 !strconcat("v", OpcodeStr,
4055 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4057 (vt128 (OpNode VR128:$src1, (i8 imm:$src2))))],
4058 IIC_SSE_PSHUF>, VEX, Sched<[WriteShuffle]>;
4059 def V#NAME#mi : Ii8<0x70, MRMSrcMem, (outs VR128:$dst),
4060 (ins i128mem:$src1, i8imm:$src2),
4061 !strconcat("v", OpcodeStr,
4062 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4064 (vt128 (OpNode (bitconvert (memopv2i64 addr:$src1)),
4065 (i8 imm:$src2))))], IIC_SSE_PSHUF>, VEX,
4066 Sched<[WriteShuffleLd]>;
4069 let Predicates = [HasAVX2] in {
4070 def V#NAME#Yri : Ii8<0x70, MRMSrcReg, (outs VR256:$dst),
4071 (ins VR256:$src1, i8imm:$src2),
4072 !strconcat("v", OpcodeStr,
4073 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4075 (vt256 (OpNode VR256:$src1, (i8 imm:$src2))))],
4076 IIC_SSE_PSHUF>, VEX, VEX_L, Sched<[WriteShuffle]>;
4077 def V#NAME#Ymi : Ii8<0x70, MRMSrcMem, (outs VR256:$dst),
4078 (ins i256mem:$src1, i8imm:$src2),
4079 !strconcat("v", OpcodeStr,
4080 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4082 (vt256 (OpNode (bitconvert (memopv4i64 addr:$src1)),
4083 (i8 imm:$src2))))], IIC_SSE_PSHUF>, VEX, VEX_L,
4084 Sched<[WriteShuffleLd]>;
4087 let Predicates = [UseSSE2] in {
4088 def ri : Ii8<0x70, MRMSrcReg,
4089 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
4090 !strconcat(OpcodeStr,
4091 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4093 (vt128 (OpNode VR128:$src1, (i8 imm:$src2))))],
4094 IIC_SSE_PSHUF>, Sched<[WriteShuffle]>;
4095 def mi : Ii8<0x70, MRMSrcMem,
4096 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
4097 !strconcat(OpcodeStr,
4098 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4100 (vt128 (OpNode (bitconvert (memopv2i64 addr:$src1)),
4101 (i8 imm:$src2))))], IIC_SSE_PSHUF>,
4102 Sched<[WriteShuffleLd]>;
4105 } // ExeDomain = SSEPackedInt
4107 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, v8i32, X86PShufd>, TB, OpSize;
4108 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, v16i16, X86PShufhw>, XS;
4109 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, v16i16, X86PShuflw>, XD;
4111 let Predicates = [HasAVX] in {
4112 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
4113 (VPSHUFDmi addr:$src1, imm:$imm)>;
4114 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4115 (VPSHUFDri VR128:$src1, imm:$imm)>;
4118 let Predicates = [UseSSE2] in {
4119 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
4120 (PSHUFDmi addr:$src1, imm:$imm)>;
4121 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4122 (PSHUFDri VR128:$src1, imm:$imm)>;
4125 //===---------------------------------------------------------------------===//
4126 // SSE2 - Packed Integer Unpack Instructions
4127 //===---------------------------------------------------------------------===//
4129 let ExeDomain = SSEPackedInt in {
4130 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
4131 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
4132 def rr : PDI<opc, MRMSrcReg,
4133 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4135 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4136 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4137 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))],
4138 IIC_SSE_UNPCK>, Sched<[WriteShuffle]>;
4139 def rm : PDI<opc, MRMSrcMem,
4140 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4142 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4143 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4144 [(set VR128:$dst, (OpNode VR128:$src1,
4145 (bc_frag (memopv2i64
4148 Sched<[WriteShuffleLd, ReadAfterLd]>;
4151 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
4152 SDNode OpNode, PatFrag bc_frag> {
4153 def Yrr : PDI<opc, MRMSrcReg,
4154 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4155 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4156 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>,
4157 Sched<[WriteShuffle]>;
4158 def Yrm : PDI<opc, MRMSrcMem,
4159 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4160 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4161 [(set VR256:$dst, (OpNode VR256:$src1,
4162 (bc_frag (memopv4i64 addr:$src2))))]>,
4163 Sched<[WriteShuffleLd, ReadAfterLd]>;
4166 let Predicates = [HasAVX] in {
4167 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
4168 bc_v16i8, 0>, VEX_4V;
4169 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
4170 bc_v8i16, 0>, VEX_4V;
4171 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
4172 bc_v4i32, 0>, VEX_4V;
4173 defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
4174 bc_v2i64, 0>, VEX_4V;
4176 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
4177 bc_v16i8, 0>, VEX_4V;
4178 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
4179 bc_v8i16, 0>, VEX_4V;
4180 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
4181 bc_v4i32, 0>, VEX_4V;
4182 defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
4183 bc_v2i64, 0>, VEX_4V;
4186 let Predicates = [HasAVX2] in {
4187 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
4188 bc_v32i8>, VEX_4V, VEX_L;
4189 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
4190 bc_v16i16>, VEX_4V, VEX_L;
4191 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
4192 bc_v8i32>, VEX_4V, VEX_L;
4193 defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
4194 bc_v4i64>, VEX_4V, VEX_L;
4196 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
4197 bc_v32i8>, VEX_4V, VEX_L;
4198 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
4199 bc_v16i16>, VEX_4V, VEX_L;
4200 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
4201 bc_v8i32>, VEX_4V, VEX_L;
4202 defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
4203 bc_v4i64>, VEX_4V, VEX_L;
4206 let Constraints = "$src1 = $dst" in {
4207 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
4209 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
4211 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
4213 defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
4216 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
4218 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
4220 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
4222 defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
4225 } // ExeDomain = SSEPackedInt
4227 //===---------------------------------------------------------------------===//
4228 // SSE2 - Packed Integer Extract and Insert
4229 //===---------------------------------------------------------------------===//
4231 let ExeDomain = SSEPackedInt in {
4232 multiclass sse2_pinsrw<bit Is2Addr = 1> {
4233 def rri : Ii8<0xC4, MRMSrcReg,
4234 (outs VR128:$dst), (ins VR128:$src1,
4235 GR32:$src2, i32i8imm:$src3),
4237 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4238 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4240 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))], IIC_SSE_PINSRW>,
4241 Sched<[WriteShuffle]>;
4242 def rmi : Ii8<0xC4, MRMSrcMem,
4243 (outs VR128:$dst), (ins VR128:$src1,
4244 i16mem:$src2, i32i8imm:$src3),
4246 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4247 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4249 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
4250 imm:$src3))], IIC_SSE_PINSRW>,
4251 Sched<[WriteShuffleLd, ReadAfterLd]>;
4255 let Predicates = [HasAVX] in
4256 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
4257 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4258 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4259 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4260 imm:$src2))]>, TB, OpSize, VEX,
4261 Sched<[WriteShuffle]>;
4262 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
4263 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4264 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4265 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4266 imm:$src2))], IIC_SSE_PEXTRW>,
4267 Sched<[WriteShuffleLd, ReadAfterLd]>;
4270 let Predicates = [HasAVX] in {
4271 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
4272 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
4273 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4274 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
4275 []>, TB, OpSize, VEX_4V, Sched<[WriteShuffle]>;
4278 let Constraints = "$src1 = $dst" in
4279 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[UseSSE2]>;
4281 } // ExeDomain = SSEPackedInt
4283 //===---------------------------------------------------------------------===//
4284 // SSE2 - Packed Mask Creation
4285 //===---------------------------------------------------------------------===//
4287 let ExeDomain = SSEPackedInt, SchedRW = [WriteVecLogic] in {
4289 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4290 "pmovmskb\t{$src, $dst|$dst, $src}",
4291 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4292 IIC_SSE_MOVMSK>, VEX;
4293 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
4294 "pmovmskb\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK>, VEX;
4296 let Predicates = [HasAVX2] in {
4297 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
4298 "pmovmskb\t{$src, $dst|$dst, $src}",
4299 [(set GR32:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>, VEX, VEX_L;
4300 def VPMOVMSKBYr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
4301 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
4304 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4305 "pmovmskb\t{$src, $dst|$dst, $src}",
4306 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4309 } // ExeDomain = SSEPackedInt
4311 //===---------------------------------------------------------------------===//
4312 // SSE2 - Conditional Store
4313 //===---------------------------------------------------------------------===//
4315 let ExeDomain = SSEPackedInt, SchedRW = [WriteStore] in {
4318 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4319 (ins VR128:$src, VR128:$mask),
4320 "maskmovdqu\t{$mask, $src|$src, $mask}",
4321 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4322 IIC_SSE_MASKMOV>, VEX;
4324 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4325 (ins VR128:$src, VR128:$mask),
4326 "maskmovdqu\t{$mask, $src|$src, $mask}",
4327 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4328 IIC_SSE_MASKMOV>, VEX;
4331 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4332 "maskmovdqu\t{$mask, $src|$src, $mask}",
4333 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4336 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4337 "maskmovdqu\t{$mask, $src|$src, $mask}",
4338 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4341 } // ExeDomain = SSEPackedInt
4343 //===---------------------------------------------------------------------===//
4344 // SSE2 - Move Doubleword
4345 //===---------------------------------------------------------------------===//
4347 //===---------------------------------------------------------------------===//
4348 // Move Int Doubleword to Packed Double Int
4350 def VMOVDI2PDIrr : VS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4351 "movd\t{$src, $dst|$dst, $src}",
4353 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>,
4354 VEX, Sched<[WriteMove]>;
4355 def VMOVDI2PDIrm : VS2I<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4356 "movd\t{$src, $dst|$dst, $src}",
4358 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4360 VEX, Sched<[WriteLoad]>;
4361 def VMOV64toPQIrr : VRS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4362 "mov{d|q}\t{$src, $dst|$dst, $src}",
4364 (v2i64 (scalar_to_vector GR64:$src)))],
4365 IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
4366 def VMOV64toSDrr : VRS2I<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4367 "mov{d|q}\t{$src, $dst|$dst, $src}",
4368 [(set FR64:$dst, (bitconvert GR64:$src))],
4369 IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
4371 def MOVDI2PDIrr : S2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4372 "movd\t{$src, $dst|$dst, $src}",
4374 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>,
4376 def MOVDI2PDIrm : S2I<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4377 "movd\t{$src, $dst|$dst, $src}",
4379 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4380 IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
4381 def MOV64toPQIrr : RS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4382 "mov{d|q}\t{$src, $dst|$dst, $src}",
4384 (v2i64 (scalar_to_vector GR64:$src)))],
4385 IIC_SSE_MOVDQ>, Sched<[WriteMove]>;
4386 def MOV64toSDrr : RS2I<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4387 "mov{d|q}\t{$src, $dst|$dst, $src}",
4388 [(set FR64:$dst, (bitconvert GR64:$src))],
4389 IIC_SSE_MOVDQ>, Sched<[WriteMove]>;
4391 //===---------------------------------------------------------------------===//
4392 // Move Int Doubleword to Single Scalar
4394 def VMOVDI2SSrr : VS2I<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4395 "movd\t{$src, $dst|$dst, $src}",
4396 [(set FR32:$dst, (bitconvert GR32:$src))],
4397 IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
4399 def VMOVDI2SSrm : VS2I<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4400 "movd\t{$src, $dst|$dst, $src}",
4401 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4403 VEX, Sched<[WriteLoad]>;
4404 def MOVDI2SSrr : S2I<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4405 "movd\t{$src, $dst|$dst, $src}",
4406 [(set FR32:$dst, (bitconvert GR32:$src))],
4407 IIC_SSE_MOVDQ>, Sched<[WriteMove]>;
4409 def MOVDI2SSrm : S2I<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4410 "movd\t{$src, $dst|$dst, $src}",
4411 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4412 IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
4414 //===---------------------------------------------------------------------===//
4415 // Move Packed Doubleword Int to Packed Double Int
4417 def VMOVPDI2DIrr : VS2I<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4418 "movd\t{$src, $dst|$dst, $src}",
4419 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4420 (iPTR 0)))], IIC_SSE_MOVD_ToGP>, VEX,
4422 def VMOVPDI2DImr : VS2I<0x7E, MRMDestMem, (outs),
4423 (ins i32mem:$dst, VR128:$src),
4424 "movd\t{$src, $dst|$dst, $src}",
4425 [(store (i32 (vector_extract (v4i32 VR128:$src),
4426 (iPTR 0))), addr:$dst)], IIC_SSE_MOVDQ>,
4427 VEX, Sched<[WriteLoad]>;
4428 def MOVPDI2DIrr : S2I<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4429 "movd\t{$src, $dst|$dst, $src}",
4430 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4431 (iPTR 0)))], IIC_SSE_MOVD_ToGP>,
4433 def MOVPDI2DImr : S2I<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4434 "movd\t{$src, $dst|$dst, $src}",
4435 [(store (i32 (vector_extract (v4i32 VR128:$src),
4436 (iPTR 0))), addr:$dst)],
4437 IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
4439 //===---------------------------------------------------------------------===//
4440 // Move Packed Doubleword Int first element to Doubleword Int
4442 let SchedRW = [WriteMove] in {
4443 def VMOVPQIto64rr : VRS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4444 "mov{d|q}\t{$src, $dst|$dst, $src}",
4445 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4450 def MOVPQIto64rr : RS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4451 "mov{d|q}\t{$src, $dst|$dst, $src}",
4452 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4457 //===---------------------------------------------------------------------===//
4458 // Bitcast FR64 <-> GR64
4460 let Predicates = [UseAVX] in
4461 def VMOV64toSDrm : VS2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4462 "vmovq\t{$src, $dst|$dst, $src}",
4463 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4464 VEX, Sched<[WriteLoad]>;
4465 def VMOVSDto64rr : VRS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4466 "mov{d|q}\t{$src, $dst|$dst, $src}",
4467 [(set GR64:$dst, (bitconvert FR64:$src))],
4468 IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
4469 def VMOVSDto64mr : VRS2I<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4470 "movq\t{$src, $dst|$dst, $src}",
4471 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4472 IIC_SSE_MOVDQ>, VEX, Sched<[WriteStore]>;
4474 def MOV64toSDrm : S2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4475 "movq\t{$src, $dst|$dst, $src}",
4476 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))],
4477 IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
4478 def MOVSDto64rr : RS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4479 "mov{d|q}\t{$src, $dst|$dst, $src}",
4480 [(set GR64:$dst, (bitconvert FR64:$src))],
4481 IIC_SSE_MOVD_ToGP>, Sched<[WriteMove]>;
4482 def MOVSDto64mr : RS2I<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4483 "movq\t{$src, $dst|$dst, $src}",
4484 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4485 IIC_SSE_MOVDQ>, Sched<[WriteStore]>;
4487 //===---------------------------------------------------------------------===//
4488 // Move Scalar Single to Double Int
4490 def VMOVSS2DIrr : VS2I<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4491 "movd\t{$src, $dst|$dst, $src}",
4492 [(set GR32:$dst, (bitconvert FR32:$src))],
4493 IIC_SSE_MOVD_ToGP>, VEX, Sched<[WriteMove]>;
4494 def VMOVSS2DImr : VS2I<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4495 "movd\t{$src, $dst|$dst, $src}",
4496 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4497 IIC_SSE_MOVDQ>, VEX, Sched<[WriteStore]>;
4498 def MOVSS2DIrr : S2I<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4499 "movd\t{$src, $dst|$dst, $src}",
4500 [(set GR32:$dst, (bitconvert FR32:$src))],
4501 IIC_SSE_MOVD_ToGP>, Sched<[WriteMove]>;
4502 def MOVSS2DImr : S2I<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4503 "movd\t{$src, $dst|$dst, $src}",
4504 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4505 IIC_SSE_MOVDQ>, Sched<[WriteStore]>;
4507 //===---------------------------------------------------------------------===//
4508 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4510 let SchedRW = [WriteMove] in {
4511 let AddedComplexity = 15 in {
4512 def VMOVZDI2PDIrr : VS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4513 "movd\t{$src, $dst|$dst, $src}",
4514 [(set VR128:$dst, (v4i32 (X86vzmovl
4515 (v4i32 (scalar_to_vector GR32:$src)))))],
4516 IIC_SSE_MOVDQ>, VEX;
4517 def VMOVZQI2PQIrr : VS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4518 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4519 [(set VR128:$dst, (v2i64 (X86vzmovl
4520 (v2i64 (scalar_to_vector GR64:$src)))))],
4524 let AddedComplexity = 15 in {
4525 def MOVZDI2PDIrr : S2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4526 "movd\t{$src, $dst|$dst, $src}",
4527 [(set VR128:$dst, (v4i32 (X86vzmovl
4528 (v4i32 (scalar_to_vector GR32:$src)))))],
4530 def MOVZQI2PQIrr : RS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4531 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4532 [(set VR128:$dst, (v2i64 (X86vzmovl
4533 (v2i64 (scalar_to_vector GR64:$src)))))],
4538 let AddedComplexity = 20, SchedRW = [WriteLoad] in {
4539 def VMOVZDI2PDIrm : VS2I<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4540 "movd\t{$src, $dst|$dst, $src}",
4542 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4543 (loadi32 addr:$src))))))],
4544 IIC_SSE_MOVDQ>, VEX;
4545 def MOVZDI2PDIrm : S2I<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4546 "movd\t{$src, $dst|$dst, $src}",
4548 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4549 (loadi32 addr:$src))))))],
4551 } // AddedComplexity, SchedRW
4553 let Predicates = [UseAVX] in {
4554 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4555 let AddedComplexity = 20 in {
4556 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4557 (VMOVZDI2PDIrm addr:$src)>;
4558 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4559 (VMOVZDI2PDIrm addr:$src)>;
4561 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4562 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4563 (v4i32 (scalar_to_vector GR32:$src)),(iPTR 0)))),
4564 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
4565 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4566 (v2i64 (scalar_to_vector GR64:$src)),(iPTR 0)))),
4567 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4570 let Predicates = [UseSSE2], AddedComplexity = 20 in {
4571 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4572 (MOVZDI2PDIrm addr:$src)>;
4573 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4574 (MOVZDI2PDIrm addr:$src)>;
4577 // These are the correct encodings of the instructions so that we know how to
4578 // read correct assembly, even though we continue to emit the wrong ones for
4579 // compatibility with Darwin's buggy assembler.
4580 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4581 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4582 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4583 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
4584 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4585 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4586 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4587 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
4588 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4589 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4590 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4591 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4593 //===---------------------------------------------------------------------===//
4594 // SSE2 - Move Quadword
4595 //===---------------------------------------------------------------------===//
4597 //===---------------------------------------------------------------------===//
4598 // Move Quadword Int to Packed Quadword Int
4601 let SchedRW = [WriteLoad] in {
4602 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4603 "vmovq\t{$src, $dst|$dst, $src}",
4605 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4606 VEX, Requires<[UseAVX]>;
4607 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4608 "movq\t{$src, $dst|$dst, $src}",
4610 (v2i64 (scalar_to_vector (loadi64 addr:$src))))],
4612 Requires<[UseSSE2]>; // SSE2 instruction with XS Prefix
4615 //===---------------------------------------------------------------------===//
4616 // Move Packed Quadword Int to Quadword Int
4618 let SchedRW = [WriteStore] in {
4619 def VMOVPQI2QImr : VS2I<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4620 "movq\t{$src, $dst|$dst, $src}",
4621 [(store (i64 (vector_extract (v2i64 VR128:$src),
4622 (iPTR 0))), addr:$dst)],
4623 IIC_SSE_MOVDQ>, VEX;
4624 def MOVPQI2QImr : S2I<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4625 "movq\t{$src, $dst|$dst, $src}",
4626 [(store (i64 (vector_extract (v2i64 VR128:$src),
4627 (iPTR 0))), addr:$dst)],
4631 //===---------------------------------------------------------------------===//
4632 // Store / copy lower 64-bits of a XMM register.
4634 def VMOVLQ128mr : VS2I<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4635 "movq\t{$src, $dst|$dst, $src}",
4636 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX,
4637 Sched<[WriteStore]>;
4638 def MOVLQ128mr : S2I<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4639 "movq\t{$src, $dst|$dst, $src}",
4640 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)],
4641 IIC_SSE_MOVDQ>, Sched<[WriteStore]>;
4643 let AddedComplexity = 20 in
4644 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4645 "vmovq\t{$src, $dst|$dst, $src}",
4647 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4648 (loadi64 addr:$src))))))],
4650 XS, VEX, Requires<[UseAVX]>, Sched<[WriteLoad]>;
4652 let AddedComplexity = 20 in
4653 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4654 "movq\t{$src, $dst|$dst, $src}",
4656 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4657 (loadi64 addr:$src))))))],
4659 XS, Requires<[UseSSE2]>, Sched<[WriteLoad]>;
4661 let Predicates = [UseAVX], AddedComplexity = 20 in {
4662 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4663 (VMOVZQI2PQIrm addr:$src)>;
4664 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4665 (VMOVZQI2PQIrm addr:$src)>;
4666 def : Pat<(v2i64 (X86vzload addr:$src)),
4667 (VMOVZQI2PQIrm addr:$src)>;
4670 let Predicates = [UseSSE2], AddedComplexity = 20 in {
4671 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4672 (MOVZQI2PQIrm addr:$src)>;
4673 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4674 (MOVZQI2PQIrm addr:$src)>;
4675 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4678 let Predicates = [HasAVX] in {
4679 def : Pat<(v4i64 (alignedX86vzload addr:$src)),
4680 (SUBREG_TO_REG (i32 0), (VMOVAPSrm addr:$src), sub_xmm)>;
4681 def : Pat<(v4i64 (X86vzload addr:$src)),
4682 (SUBREG_TO_REG (i32 0), (VMOVUPSrm addr:$src), sub_xmm)>;
4685 //===---------------------------------------------------------------------===//
4686 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4687 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4689 let SchedRW = [WriteVecLogic] in {
4690 let AddedComplexity = 15 in
4691 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4692 "vmovq\t{$src, $dst|$dst, $src}",
4693 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4695 XS, VEX, Requires<[UseAVX]>;
4696 let AddedComplexity = 15 in
4697 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4698 "movq\t{$src, $dst|$dst, $src}",
4699 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4701 XS, Requires<[UseSSE2]>;
4704 let SchedRW = [WriteVecLogicLd] in {
4705 let AddedComplexity = 20 in
4706 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4707 "vmovq\t{$src, $dst|$dst, $src}",
4708 [(set VR128:$dst, (v2i64 (X86vzmovl
4709 (loadv2i64 addr:$src))))],
4711 XS, VEX, Requires<[UseAVX]>;
4712 let AddedComplexity = 20 in {
4713 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4714 "movq\t{$src, $dst|$dst, $src}",
4715 [(set VR128:$dst, (v2i64 (X86vzmovl
4716 (loadv2i64 addr:$src))))],
4718 XS, Requires<[UseSSE2]>;
4722 let AddedComplexity = 20 in {
4723 let Predicates = [UseAVX] in {
4724 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4725 (VMOVZPQILo2PQIrm addr:$src)>;
4726 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4727 (VMOVZPQILo2PQIrr VR128:$src)>;
4729 let Predicates = [UseSSE2] in {
4730 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4731 (MOVZPQILo2PQIrm addr:$src)>;
4732 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4733 (MOVZPQILo2PQIrr VR128:$src)>;
4737 // Instructions to match in the assembler
4738 let SchedRW = [WriteMove] in {
4739 def VMOVQs64rr : VS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4740 "movq\t{$src, $dst|$dst, $src}", [],
4741 IIC_SSE_MOVDQ>, VEX, VEX_W;
4742 def VMOVQd64rr : VS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4743 "movq\t{$src, $dst|$dst, $src}", [],
4744 IIC_SSE_MOVDQ>, VEX, VEX_W;
4745 // Recognize "movd" with GR64 destination, but encode as a "movq"
4746 def VMOVQd64rr_alt : VS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4747 "movq\t{$src, $dst|$dst, $src}", [],
4748 IIC_SSE_MOVDQ>, VEX, VEX_W;
4751 // Instructions for the disassembler
4752 // xr = XMM register
4755 let SchedRW = [WriteMove] in {
4756 let Predicates = [UseAVX] in
4757 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4758 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
4759 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4760 "movq\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVQ_RR>, XS;
4763 //===---------------------------------------------------------------------===//
4764 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4765 //===---------------------------------------------------------------------===//
4766 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4767 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4768 X86MemOperand x86memop> {
4769 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4770 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4771 [(set RC:$dst, (vt (OpNode RC:$src)))],
4772 IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>;
4773 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4774 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4775 [(set RC:$dst, (OpNode (mem_frag addr:$src)))],
4776 IIC_SSE_MOV_LH>, Sched<[WriteShuffleLd]>;
4779 let Predicates = [HasAVX] in {
4780 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4781 v4f32, VR128, memopv4f32, f128mem>, VEX;
4782 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4783 v4f32, VR128, memopv4f32, f128mem>, VEX;
4784 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4785 v8f32, VR256, memopv8f32, f256mem>, VEX, VEX_L;
4786 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4787 v8f32, VR256, memopv8f32, f256mem>, VEX, VEX_L;
4789 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4790 memopv4f32, f128mem>;
4791 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4792 memopv4f32, f128mem>;
4794 let Predicates = [HasAVX] in {
4795 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4796 (VMOVSHDUPrr VR128:$src)>;
4797 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4798 (VMOVSHDUPrm addr:$src)>;
4799 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4800 (VMOVSLDUPrr VR128:$src)>;
4801 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4802 (VMOVSLDUPrm addr:$src)>;
4803 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4804 (VMOVSHDUPYrr VR256:$src)>;
4805 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4806 (VMOVSHDUPYrm addr:$src)>;
4807 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4808 (VMOVSLDUPYrr VR256:$src)>;
4809 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
4810 (VMOVSLDUPYrm addr:$src)>;
4813 let Predicates = [UseSSE3] in {
4814 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4815 (MOVSHDUPrr VR128:$src)>;
4816 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4817 (MOVSHDUPrm addr:$src)>;
4818 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4819 (MOVSLDUPrr VR128:$src)>;
4820 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4821 (MOVSLDUPrm addr:$src)>;
4824 //===---------------------------------------------------------------------===//
4825 // SSE3 - Replicate Double FP - MOVDDUP
4826 //===---------------------------------------------------------------------===//
4828 multiclass sse3_replicate_dfp<string OpcodeStr> {
4829 let neverHasSideEffects = 1 in
4830 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4831 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4832 [], IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>;
4833 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
4834 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4837 (scalar_to_vector (loadf64 addr:$src)))))],
4838 IIC_SSE_MOV_LH>, Sched<[WriteShuffleLd]>;
4841 // FIXME: Merge with above classe when there're patterns for the ymm version
4842 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
4843 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
4844 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4845 [(set VR256:$dst, (v4f64 (X86Movddup VR256:$src)))]>,
4846 Sched<[WriteShuffle]>;
4847 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
4848 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4851 (scalar_to_vector (loadf64 addr:$src)))))]>,
4852 Sched<[WriteShuffleLd]>;
4855 let Predicates = [HasAVX] in {
4856 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
4857 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX, VEX_L;
4860 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
4862 let Predicates = [HasAVX] in {
4863 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4864 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4865 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4866 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4867 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4868 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4869 def : Pat<(X86Movddup (bc_v2f64
4870 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4871 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4874 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
4875 (VMOVDDUPYrm addr:$src)>;
4876 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
4877 (VMOVDDUPYrm addr:$src)>;
4878 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
4879 (VMOVDDUPYrm addr:$src)>;
4880 def : Pat<(X86Movddup (v4i64 VR256:$src)),
4881 (VMOVDDUPYrr VR256:$src)>;
4884 let Predicates = [UseSSE3] in {
4885 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4886 (MOVDDUPrm addr:$src)>;
4887 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4888 (MOVDDUPrm addr:$src)>;
4889 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4890 (MOVDDUPrm addr:$src)>;
4891 def : Pat<(X86Movddup (bc_v2f64
4892 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4893 (MOVDDUPrm addr:$src)>;
4896 //===---------------------------------------------------------------------===//
4897 // SSE3 - Move Unaligned Integer
4898 //===---------------------------------------------------------------------===//
4900 let SchedRW = [WriteLoad] in {
4901 let Predicates = [HasAVX] in {
4902 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4903 "vlddqu\t{$src, $dst|$dst, $src}",
4904 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
4905 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
4906 "vlddqu\t{$src, $dst|$dst, $src}",
4907 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>,
4910 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4911 "lddqu\t{$src, $dst|$dst, $src}",
4912 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))],
4916 //===---------------------------------------------------------------------===//
4917 // SSE3 - Arithmetic
4918 //===---------------------------------------------------------------------===//
4920 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
4921 X86MemOperand x86memop, OpndItins itins,
4923 def rr : I<0xD0, MRMSrcReg,
4924 (outs RC:$dst), (ins RC:$src1, RC:$src2),
4926 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4927 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4928 [(set RC:$dst, (Int RC:$src1, RC:$src2))], itins.rr>,
4929 Sched<[itins.Sched]>;
4930 def rm : I<0xD0, MRMSrcMem,
4931 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4933 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4934 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4935 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))], itins.rr>,
4936 Sched<[itins.Sched.Folded, ReadAfterLd]>;
4939 let Predicates = [HasAVX] in {
4940 let ExeDomain = SSEPackedSingle in {
4941 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
4942 f128mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V;
4943 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
4944 f256mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V, VEX_L;
4946 let ExeDomain = SSEPackedDouble in {
4947 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
4948 f128mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V;
4949 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
4950 f256mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V, VEX_L;
4953 let Constraints = "$src1 = $dst", Predicates = [UseSSE3] in {
4954 let ExeDomain = SSEPackedSingle in
4955 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
4956 f128mem, SSE_ALU_F32P>, TB, XD;
4957 let ExeDomain = SSEPackedDouble in
4958 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
4959 f128mem, SSE_ALU_F64P>, TB, OpSize;
4962 //===---------------------------------------------------------------------===//
4963 // SSE3 Instructions
4964 //===---------------------------------------------------------------------===//
4967 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4968 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
4969 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4971 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4972 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4973 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>,
4976 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4978 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4979 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4980 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
4981 IIC_SSE_HADDSUB_RM>, Sched<[WriteFAddLd, ReadAfterLd]>;
4983 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4984 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
4985 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4987 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4988 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4989 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>,
4992 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4994 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4995 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4996 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
4997 IIC_SSE_HADDSUB_RM>, Sched<[WriteFAddLd, ReadAfterLd]>;
5000 let Predicates = [HasAVX] in {
5001 let ExeDomain = SSEPackedSingle in {
5002 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
5003 X86fhadd, 0>, VEX_4V;
5004 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
5005 X86fhsub, 0>, VEX_4V;
5006 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
5007 X86fhadd, 0>, VEX_4V, VEX_L;
5008 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
5009 X86fhsub, 0>, VEX_4V, VEX_L;
5011 let ExeDomain = SSEPackedDouble in {
5012 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
5013 X86fhadd, 0>, VEX_4V;
5014 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
5015 X86fhsub, 0>, VEX_4V;
5016 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
5017 X86fhadd, 0>, VEX_4V, VEX_L;
5018 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
5019 X86fhsub, 0>, VEX_4V, VEX_L;
5023 let Constraints = "$src1 = $dst" in {
5024 let ExeDomain = SSEPackedSingle in {
5025 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
5026 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
5028 let ExeDomain = SSEPackedDouble in {
5029 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
5030 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
5034 //===---------------------------------------------------------------------===//
5035 // SSSE3 - Packed Absolute Instructions
5036 //===---------------------------------------------------------------------===//
5039 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5040 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
5041 Intrinsic IntId128> {
5042 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5044 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5045 [(set VR128:$dst, (IntId128 VR128:$src))], IIC_SSE_PABS_RR>,
5046 OpSize, Sched<[WriteVecALU]>;
5048 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5050 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5053 (bitconvert (memopv2i64 addr:$src))))], IIC_SSE_PABS_RM>,
5054 OpSize, Sched<[WriteVecALULd]>;
5057 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5058 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
5059 Intrinsic IntId256> {
5060 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5062 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5063 [(set VR256:$dst, (IntId256 VR256:$src))]>,
5064 OpSize, Sched<[WriteVecALU]>;
5066 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5068 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5071 (bitconvert (memopv4i64 addr:$src))))]>, OpSize,
5072 Sched<[WriteVecALULd]>;
5075 // Helper fragments to match sext vXi1 to vXiY.
5076 def v16i1sextv16i8 : PatLeaf<(v16i8 (X86pcmpgt (bc_v16i8 (v4i32 immAllZerosV)),
5078 def v8i1sextv8i16 : PatLeaf<(v8i16 (X86vsrai VR128:$src, (i32 15)))>;
5079 def v4i1sextv4i32 : PatLeaf<(v4i32 (X86vsrai VR128:$src, (i32 31)))>;
5080 def v32i1sextv32i8 : PatLeaf<(v32i8 (X86pcmpgt (bc_v32i8 (v8i32 immAllZerosV)),
5082 def v16i1sextv16i16: PatLeaf<(v16i16 (X86vsrai VR256:$src, (i32 15)))>;
5083 def v8i1sextv8i32 : PatLeaf<(v8i32 (X86vsrai VR256:$src, (i32 31)))>;
5085 let Predicates = [HasAVX] in {
5086 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb",
5087 int_x86_ssse3_pabs_b_128>, VEX;
5088 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw",
5089 int_x86_ssse3_pabs_w_128>, VEX;
5090 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd",
5091 int_x86_ssse3_pabs_d_128>, VEX;
5094 (bc_v2i64 (v16i1sextv16i8)),
5095 (bc_v2i64 (add (v16i8 VR128:$src), (v16i1sextv16i8)))),
5096 (VPABSBrr128 VR128:$src)>;
5098 (bc_v2i64 (v8i1sextv8i16)),
5099 (bc_v2i64 (add (v8i16 VR128:$src), (v8i1sextv8i16)))),
5100 (VPABSWrr128 VR128:$src)>;
5102 (bc_v2i64 (v4i1sextv4i32)),
5103 (bc_v2i64 (add (v4i32 VR128:$src), (v4i1sextv4i32)))),
5104 (VPABSDrr128 VR128:$src)>;
5107 let Predicates = [HasAVX2] in {
5108 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb",
5109 int_x86_avx2_pabs_b>, VEX, VEX_L;
5110 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw",
5111 int_x86_avx2_pabs_w>, VEX, VEX_L;
5112 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd",
5113 int_x86_avx2_pabs_d>, VEX, VEX_L;
5116 (bc_v4i64 (v32i1sextv32i8)),
5117 (bc_v4i64 (add (v32i8 VR256:$src), (v32i1sextv32i8)))),
5118 (VPABSBrr256 VR256:$src)>;
5120 (bc_v4i64 (v16i1sextv16i16)),
5121 (bc_v4i64 (add (v16i16 VR256:$src), (v16i1sextv16i16)))),
5122 (VPABSWrr256 VR256:$src)>;
5124 (bc_v4i64 (v8i1sextv8i32)),
5125 (bc_v4i64 (add (v8i32 VR256:$src), (v8i1sextv8i32)))),
5126 (VPABSDrr256 VR256:$src)>;
5129 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb",
5130 int_x86_ssse3_pabs_b_128>;
5131 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw",
5132 int_x86_ssse3_pabs_w_128>;
5133 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
5134 int_x86_ssse3_pabs_d_128>;
5136 let Predicates = [HasSSSE3] in {
5138 (bc_v2i64 (v16i1sextv16i8)),
5139 (bc_v2i64 (add (v16i8 VR128:$src), (v16i1sextv16i8)))),
5140 (PABSBrr128 VR128:$src)>;
5142 (bc_v2i64 (v8i1sextv8i16)),
5143 (bc_v2i64 (add (v8i16 VR128:$src), (v8i1sextv8i16)))),
5144 (PABSWrr128 VR128:$src)>;
5146 (bc_v2i64 (v4i1sextv4i32)),
5147 (bc_v2i64 (add (v4i32 VR128:$src), (v4i1sextv4i32)))),
5148 (PABSDrr128 VR128:$src)>;
5151 //===---------------------------------------------------------------------===//
5152 // SSSE3 - Packed Binary Operator Instructions
5153 //===---------------------------------------------------------------------===//
5155 let Sched = WriteVecALU in {
5156 def SSE_PHADDSUBD : OpndItins<
5157 IIC_SSE_PHADDSUBD_RR, IIC_SSE_PHADDSUBD_RM
5159 def SSE_PHADDSUBSW : OpndItins<
5160 IIC_SSE_PHADDSUBSW_RR, IIC_SSE_PHADDSUBSW_RM
5162 def SSE_PHADDSUBW : OpndItins<
5163 IIC_SSE_PHADDSUBW_RR, IIC_SSE_PHADDSUBW_RM
5166 let Sched = WriteShuffle in
5167 def SSE_PSHUFB : OpndItins<
5168 IIC_SSE_PSHUFB_RR, IIC_SSE_PSHUFB_RM
5170 let Sched = WriteVecALU in
5171 def SSE_PSIGN : OpndItins<
5172 IIC_SSE_PSIGN_RR, IIC_SSE_PSIGN_RM
5174 let Sched = WriteVecIMul in
5175 def SSE_PMULHRSW : OpndItins<
5176 IIC_SSE_PMULHRSW, IIC_SSE_PMULHRSW
5179 /// SS3I_binop_rm - Simple SSSE3 bin op
5180 multiclass SS3I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
5181 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
5182 X86MemOperand x86memop, OpndItins itins,
5184 let isCommutable = 1 in
5185 def rr : SS38I<opc, MRMSrcReg, (outs RC:$dst),
5186 (ins RC:$src1, RC:$src2),
5188 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5189 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5190 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>,
5191 OpSize, Sched<[itins.Sched]>;
5192 def rm : SS38I<opc, MRMSrcMem, (outs RC:$dst),
5193 (ins RC:$src1, x86memop:$src2),
5195 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5196 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5198 (OpVT (OpNode RC:$src1,
5199 (bitconvert (memop_frag addr:$src2)))))], itins.rm>, OpSize,
5200 Sched<[itins.Sched.Folded, ReadAfterLd]>;
5203 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
5204 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
5205 Intrinsic IntId128, OpndItins itins,
5207 let isCommutable = 1 in
5208 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5209 (ins VR128:$src1, VR128:$src2),
5211 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5212 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5213 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5214 OpSize, Sched<[itins.Sched]>;
5215 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5216 (ins VR128:$src1, i128mem:$src2),
5218 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5219 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5221 (IntId128 VR128:$src1,
5222 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize,
5223 Sched<[itins.Sched.Folded, ReadAfterLd]>;
5226 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
5227 Intrinsic IntId256> {
5228 let isCommutable = 1 in
5229 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5230 (ins VR256:$src1, VR256:$src2),
5231 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5232 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
5234 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5235 (ins VR256:$src1, i256mem:$src2),
5236 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5238 (IntId256 VR256:$src1,
5239 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
5242 let ImmT = NoImm, Predicates = [HasAVX] in {
5243 let isCommutable = 0 in {
5244 defm VPHADDW : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v8i16, VR128,
5245 memopv2i64, i128mem,
5246 SSE_PHADDSUBW, 0>, VEX_4V;
5247 defm VPHADDD : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v4i32, VR128,
5248 memopv2i64, i128mem,
5249 SSE_PHADDSUBD, 0>, VEX_4V;
5250 defm VPHSUBW : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v8i16, VR128,
5251 memopv2i64, i128mem,
5252 SSE_PHADDSUBW, 0>, VEX_4V;
5253 defm VPHSUBD : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v4i32, VR128,
5254 memopv2i64, i128mem,
5255 SSE_PHADDSUBD, 0>, VEX_4V;
5256 defm VPSIGNB : SS3I_binop_rm<0x08, "vpsignb", X86psign, v16i8, VR128,
5257 memopv2i64, i128mem,
5258 SSE_PSIGN, 0>, VEX_4V;
5259 defm VPSIGNW : SS3I_binop_rm<0x09, "vpsignw", X86psign, v8i16, VR128,
5260 memopv2i64, i128mem,
5261 SSE_PSIGN, 0>, VEX_4V;
5262 defm VPSIGND : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v4i32, VR128,
5263 memopv2i64, i128mem,
5264 SSE_PSIGN, 0>, VEX_4V;
5265 defm VPSHUFB : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v16i8, VR128,
5266 memopv2i64, i128mem,
5267 SSE_PSHUFB, 0>, VEX_4V;
5268 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw",
5269 int_x86_ssse3_phadd_sw_128,
5270 SSE_PHADDSUBSW, 0>, VEX_4V;
5271 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw",
5272 int_x86_ssse3_phsub_sw_128,
5273 SSE_PHADDSUBSW, 0>, VEX_4V;
5274 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw",
5275 int_x86_ssse3_pmadd_ub_sw_128,
5276 SSE_PMADD, 0>, VEX_4V;
5278 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw",
5279 int_x86_ssse3_pmul_hr_sw_128,
5280 SSE_PMULHRSW, 0>, VEX_4V;
5283 let ImmT = NoImm, Predicates = [HasAVX2] in {
5284 let isCommutable = 0 in {
5285 defm VPHADDWY : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v16i16, VR256,
5286 memopv4i64, i256mem,
5287 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5288 defm VPHADDDY : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v8i32, VR256,
5289 memopv4i64, i256mem,
5290 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5291 defm VPHSUBWY : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v16i16, VR256,
5292 memopv4i64, i256mem,
5293 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5294 defm VPHSUBDY : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v8i32, VR256,
5295 memopv4i64, i256mem,
5296 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5297 defm VPSIGNBY : SS3I_binop_rm<0x08, "vpsignb", X86psign, v32i8, VR256,
5298 memopv4i64, i256mem,
5299 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5300 defm VPSIGNWY : SS3I_binop_rm<0x09, "vpsignw", X86psign, v16i16, VR256,
5301 memopv4i64, i256mem,
5302 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5303 defm VPSIGNDY : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v8i32, VR256,
5304 memopv4i64, i256mem,
5305 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5306 defm VPSHUFBY : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v32i8, VR256,
5307 memopv4i64, i256mem,
5308 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5309 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
5310 int_x86_avx2_phadd_sw>, VEX_4V, VEX_L;
5311 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
5312 int_x86_avx2_phsub_sw>, VEX_4V, VEX_L;
5313 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
5314 int_x86_avx2_pmadd_ub_sw>, VEX_4V, VEX_L;
5316 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
5317 int_x86_avx2_pmul_hr_sw>, VEX_4V, VEX_L;
5320 // None of these have i8 immediate fields.
5321 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
5322 let isCommutable = 0 in {
5323 defm PHADDW : SS3I_binop_rm<0x01, "phaddw", X86hadd, v8i16, VR128,
5324 memopv2i64, i128mem, SSE_PHADDSUBW>;
5325 defm PHADDD : SS3I_binop_rm<0x02, "phaddd", X86hadd, v4i32, VR128,
5326 memopv2i64, i128mem, SSE_PHADDSUBD>;
5327 defm PHSUBW : SS3I_binop_rm<0x05, "phsubw", X86hsub, v8i16, VR128,
5328 memopv2i64, i128mem, SSE_PHADDSUBW>;
5329 defm PHSUBD : SS3I_binop_rm<0x06, "phsubd", X86hsub, v4i32, VR128,
5330 memopv2i64, i128mem, SSE_PHADDSUBD>;
5331 defm PSIGNB : SS3I_binop_rm<0x08, "psignb", X86psign, v16i8, VR128,
5332 memopv2i64, i128mem, SSE_PSIGN>;
5333 defm PSIGNW : SS3I_binop_rm<0x09, "psignw", X86psign, v8i16, VR128,
5334 memopv2i64, i128mem, SSE_PSIGN>;
5335 defm PSIGND : SS3I_binop_rm<0x0A, "psignd", X86psign, v4i32, VR128,
5336 memopv2i64, i128mem, SSE_PSIGN>;
5337 defm PSHUFB : SS3I_binop_rm<0x00, "pshufb", X86pshufb, v16i8, VR128,
5338 memopv2i64, i128mem, SSE_PSHUFB>;
5339 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw",
5340 int_x86_ssse3_phadd_sw_128,
5342 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw",
5343 int_x86_ssse3_phsub_sw_128,
5345 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw",
5346 int_x86_ssse3_pmadd_ub_sw_128, SSE_PMADD>;
5348 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw",
5349 int_x86_ssse3_pmul_hr_sw_128,
5353 //===---------------------------------------------------------------------===//
5354 // SSSE3 - Packed Align Instruction Patterns
5355 //===---------------------------------------------------------------------===//
5357 multiclass ssse3_palignr<string asm, bit Is2Addr = 1> {
5358 let neverHasSideEffects = 1 in {
5359 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
5360 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5362 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5364 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5365 [], IIC_SSE_PALIGNR>, OpSize, Sched<[WriteShuffle]>;
5367 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
5368 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5370 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5372 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5373 [], IIC_SSE_PALIGNR>, OpSize, Sched<[WriteShuffleLd, ReadAfterLd]>;
5377 multiclass ssse3_palignr_y<string asm, bit Is2Addr = 1> {
5378 let neverHasSideEffects = 1 in {
5379 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
5380 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5382 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5383 []>, OpSize, Sched<[WriteShuffle]>;
5385 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
5386 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
5388 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5389 []>, OpSize, Sched<[WriteShuffleLd, ReadAfterLd]>;
5393 let Predicates = [HasAVX] in
5394 defm VPALIGN : ssse3_palignr<"vpalignr", 0>, VEX_4V;
5395 let Predicates = [HasAVX2] in
5396 defm VPALIGN : ssse3_palignr_y<"vpalignr", 0>, VEX_4V, VEX_L;
5397 let Constraints = "$src1 = $dst", Predicates = [UseSSSE3] in
5398 defm PALIGN : ssse3_palignr<"palignr">;
5400 let Predicates = [HasAVX2] in {
5401 def : Pat<(v8i32 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5402 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5403 def : Pat<(v8f32 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5404 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5405 def : Pat<(v16i16 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5406 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5407 def : Pat<(v32i8 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5408 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5411 let Predicates = [HasAVX] in {
5412 def : Pat<(v4i32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5413 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5414 def : Pat<(v4f32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5415 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5416 def : Pat<(v8i16 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5417 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5418 def : Pat<(v16i8 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5419 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5422 let Predicates = [UseSSSE3] in {
5423 def : Pat<(v4i32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5424 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5425 def : Pat<(v4f32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5426 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5427 def : Pat<(v8i16 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5428 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5429 def : Pat<(v16i8 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5430 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5433 //===---------------------------------------------------------------------===//
5434 // SSSE3 - Thread synchronization
5435 //===---------------------------------------------------------------------===//
5437 let SchedRW = [WriteSystem] in {
5438 let usesCustomInserter = 1 in {
5439 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5440 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>,
5441 Requires<[HasSSE3]>;
5444 let Uses = [EAX, ECX, EDX] in
5445 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", [], IIC_SSE_MONITOR>,
5446 TB, Requires<[HasSSE3]>;
5447 let Uses = [ECX, EAX] in
5448 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait",
5449 [(int_x86_sse3_mwait ECX, EAX)], IIC_SSE_MWAIT>,
5450 TB, Requires<[HasSSE3]>;
5453 def : InstAlias<"mwait\t{%eax, %ecx|ecx, eax}", (MWAITrr)>, Requires<[In32BitMode]>;
5454 def : InstAlias<"mwait\t{%rax, %rcx|rcx, rax}", (MWAITrr)>, Requires<[In64BitMode]>;
5456 def : InstAlias<"monitor\t{%eax, %ecx, %edx|edx, ecx, eax}", (MONITORrrr)>,
5457 Requires<[In32BitMode]>;
5458 def : InstAlias<"monitor\t{%rax, %rcx, %rdx|rdx, rcx, rax}", (MONITORrrr)>,
5459 Requires<[In64BitMode]>;
5461 //===----------------------------------------------------------------------===//
5462 // SSE4.1 - Packed Move with Sign/Zero Extend
5463 //===----------------------------------------------------------------------===//
5465 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5466 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5467 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5468 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5470 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5471 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5473 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5477 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5479 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5480 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5481 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5483 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5484 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5485 [(set VR256:$dst, (IntId (load addr:$src)))]>, OpSize;
5488 let Predicates = [HasAVX] in {
5489 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
5491 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
5493 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
5495 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
5497 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
5499 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
5503 let Predicates = [HasAVX2] in {
5504 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5505 int_x86_avx2_pmovsxbw>, VEX, VEX_L;
5506 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5507 int_x86_avx2_pmovsxwd>, VEX, VEX_L;
5508 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5509 int_x86_avx2_pmovsxdq>, VEX, VEX_L;
5510 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5511 int_x86_avx2_pmovzxbw>, VEX, VEX_L;
5512 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5513 int_x86_avx2_pmovzxwd>, VEX, VEX_L;
5514 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5515 int_x86_avx2_pmovzxdq>, VEX, VEX_L;
5518 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
5519 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
5520 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
5521 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
5522 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
5523 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
5525 let Predicates = [HasAVX] in {
5526 // Common patterns involving scalar load.
5527 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5528 (VPMOVSXBWrm addr:$src)>;
5529 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5530 (VPMOVSXBWrm addr:$src)>;
5531 def : Pat<(int_x86_sse41_pmovsxbw (bc_v16i8 (loadv2i64 addr:$src))),
5532 (VPMOVSXBWrm addr:$src)>;
5534 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5535 (VPMOVSXWDrm addr:$src)>;
5536 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5537 (VPMOVSXWDrm addr:$src)>;
5538 def : Pat<(int_x86_sse41_pmovsxwd (bc_v8i16 (loadv2i64 addr:$src))),
5539 (VPMOVSXWDrm addr:$src)>;
5541 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5542 (VPMOVSXDQrm addr:$src)>;
5543 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5544 (VPMOVSXDQrm addr:$src)>;
5545 def : Pat<(int_x86_sse41_pmovsxdq (bc_v4i32 (loadv2i64 addr:$src))),
5546 (VPMOVSXDQrm addr:$src)>;
5548 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5549 (VPMOVZXBWrm addr:$src)>;
5550 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5551 (VPMOVZXBWrm addr:$src)>;
5552 def : Pat<(int_x86_sse41_pmovzxbw (bc_v16i8 (loadv2i64 addr:$src))),
5553 (VPMOVZXBWrm addr:$src)>;
5555 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5556 (VPMOVZXWDrm addr:$src)>;
5557 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5558 (VPMOVZXWDrm addr:$src)>;
5559 def : Pat<(int_x86_sse41_pmovzxwd (bc_v8i16 (loadv2i64 addr:$src))),
5560 (VPMOVZXWDrm addr:$src)>;
5562 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5563 (VPMOVZXDQrm addr:$src)>;
5564 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5565 (VPMOVZXDQrm addr:$src)>;
5566 def : Pat<(int_x86_sse41_pmovzxdq (bc_v4i32 (loadv2i64 addr:$src))),
5567 (VPMOVZXDQrm addr:$src)>;
5570 let Predicates = [UseSSE41] in {
5571 // Common patterns involving scalar load.
5572 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5573 (PMOVSXBWrm addr:$src)>;
5574 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5575 (PMOVSXBWrm addr:$src)>;
5576 def : Pat<(int_x86_sse41_pmovsxbw (bc_v16i8 (loadv2i64 addr:$src))),
5577 (PMOVSXBWrm addr:$src)>;
5579 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5580 (PMOVSXWDrm addr:$src)>;
5581 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5582 (PMOVSXWDrm addr:$src)>;
5583 def : Pat<(int_x86_sse41_pmovsxwd (bc_v8i16 (loadv2i64 addr:$src))),
5584 (PMOVSXWDrm addr:$src)>;
5586 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5587 (PMOVSXDQrm addr:$src)>;
5588 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5589 (PMOVSXDQrm addr:$src)>;
5590 def : Pat<(int_x86_sse41_pmovsxdq (bc_v4i32 (loadv2i64 addr:$src))),
5591 (PMOVSXDQrm addr:$src)>;
5593 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5594 (PMOVZXBWrm addr:$src)>;
5595 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5596 (PMOVZXBWrm addr:$src)>;
5597 def : Pat<(int_x86_sse41_pmovzxbw (bc_v16i8 (loadv2i64 addr:$src))),
5598 (PMOVZXBWrm addr:$src)>;
5600 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5601 (PMOVZXWDrm addr:$src)>;
5602 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5603 (PMOVZXWDrm addr:$src)>;
5604 def : Pat<(int_x86_sse41_pmovzxwd (bc_v8i16 (loadv2i64 addr:$src))),
5605 (PMOVZXWDrm addr:$src)>;
5607 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5608 (PMOVZXDQrm addr:$src)>;
5609 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5610 (PMOVZXDQrm addr:$src)>;
5611 def : Pat<(int_x86_sse41_pmovzxdq (bc_v4i32 (loadv2i64 addr:$src))),
5612 (PMOVZXDQrm addr:$src)>;
5615 let Predicates = [HasAVX2] in {
5616 let AddedComplexity = 15 in {
5617 def : Pat<(v4i64 (X86vzmovly (v4i32 VR128:$src))),
5618 (VPMOVZXDQYrr VR128:$src)>;
5619 def : Pat<(v8i32 (X86vzmovly (v8i16 VR128:$src))),
5620 (VPMOVZXWDYrr VR128:$src)>;
5623 def : Pat<(v4i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>;
5624 def : Pat<(v8i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>;
5627 let Predicates = [HasAVX] in {
5628 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
5629 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
5632 let Predicates = [UseSSE41] in {
5633 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
5634 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
5638 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5639 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5640 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5641 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5643 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5644 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5646 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5650 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5652 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5653 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5654 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5656 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5657 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5659 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5663 let Predicates = [HasAVX] in {
5664 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5666 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5668 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5670 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5674 let Predicates = [HasAVX2] in {
5675 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5676 int_x86_avx2_pmovsxbd>, VEX, VEX_L;
5677 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5678 int_x86_avx2_pmovsxwq>, VEX, VEX_L;
5679 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5680 int_x86_avx2_pmovzxbd>, VEX, VEX_L;
5681 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5682 int_x86_avx2_pmovzxwq>, VEX, VEX_L;
5685 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
5686 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
5687 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
5688 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
5690 let Predicates = [HasAVX] in {
5691 // Common patterns involving scalar load
5692 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5693 (VPMOVSXBDrm addr:$src)>;
5694 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5695 (VPMOVSXWQrm addr:$src)>;
5697 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5698 (VPMOVZXBDrm addr:$src)>;
5699 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5700 (VPMOVZXWQrm addr:$src)>;
5703 let Predicates = [UseSSE41] in {
5704 // Common patterns involving scalar load
5705 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5706 (PMOVSXBDrm addr:$src)>;
5707 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5708 (PMOVSXWQrm addr:$src)>;
5710 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5711 (PMOVZXBDrm addr:$src)>;
5712 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5713 (PMOVZXWQrm addr:$src)>;
5716 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5717 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5718 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5719 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5721 // Expecting a i16 load any extended to i32 value.
5722 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5723 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5724 [(set VR128:$dst, (IntId (bitconvert
5725 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5729 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5731 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5732 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5733 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5735 // Expecting a i16 load any extended to i32 value.
5736 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5737 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5738 [(set VR256:$dst, (IntId (bitconvert
5739 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5743 let Predicates = [HasAVX] in {
5744 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5746 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5749 let Predicates = [HasAVX2] in {
5750 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5751 int_x86_avx2_pmovsxbq>, VEX, VEX_L;
5752 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5753 int_x86_avx2_pmovzxbq>, VEX, VEX_L;
5755 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
5756 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
5758 let Predicates = [HasAVX2] in {
5759 def : Pat<(v16i16 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBWYrr VR128:$src)>;
5760 def : Pat<(v8i32 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBDYrr VR128:$src)>;
5761 def : Pat<(v4i64 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBQYrr VR128:$src)>;
5763 def : Pat<(v8i32 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>;
5764 def : Pat<(v4i64 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWQYrr VR128:$src)>;
5766 def : Pat<(v4i64 (X86vsext (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>;
5768 def : Pat<(v16i16 (X86vsext (v32i8 VR256:$src))),
5769 (VPMOVSXBWYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5770 def : Pat<(v8i32 (X86vsext (v32i8 VR256:$src))),
5771 (VPMOVSXBDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5772 def : Pat<(v4i64 (X86vsext (v32i8 VR256:$src))),
5773 (VPMOVSXBQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5775 def : Pat<(v8i32 (X86vsext (v16i16 VR256:$src))),
5776 (VPMOVSXWDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5777 def : Pat<(v4i64 (X86vsext (v16i16 VR256:$src))),
5778 (VPMOVSXWQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5780 def : Pat<(v4i64 (X86vsext (v8i32 VR256:$src))),
5781 (VPMOVSXDQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5783 def : Pat<(v8i32 (X86vsmovl (v8i16 (bitconvert (v2i64 (load addr:$src)))))),
5784 (VPMOVSXWDYrm addr:$src)>;
5785 def : Pat<(v4i64 (X86vsmovl (v4i32 (bitconvert (v2i64 (load addr:$src)))))),
5786 (VPMOVSXDQYrm addr:$src)>;
5788 def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2i64
5789 (scalar_to_vector (loadi64 addr:$src))))))),
5790 (VPMOVSXBDYrm addr:$src)>;
5791 def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2f64
5792 (scalar_to_vector (loadf64 addr:$src))))))),
5793 (VPMOVSXBDYrm addr:$src)>;
5795 def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2i64
5796 (scalar_to_vector (loadi64 addr:$src))))))),
5797 (VPMOVSXWQYrm addr:$src)>;
5798 def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2f64
5799 (scalar_to_vector (loadf64 addr:$src))))))),
5800 (VPMOVSXWQYrm addr:$src)>;
5802 def : Pat<(v4i64 (X86vsext (v16i8 (bitconvert (v4i32
5803 (scalar_to_vector (loadi32 addr:$src))))))),
5804 (VPMOVSXBQYrm addr:$src)>;
5807 let Predicates = [HasAVX] in {
5808 // Common patterns involving scalar load
5809 def : Pat<(int_x86_sse41_pmovsxbq
5810 (bitconvert (v4i32 (X86vzmovl
5811 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5812 (VPMOVSXBQrm addr:$src)>;
5814 def : Pat<(int_x86_sse41_pmovzxbq
5815 (bitconvert (v4i32 (X86vzmovl
5816 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5817 (VPMOVZXBQrm addr:$src)>;
5820 let Predicates = [UseSSE41] in {
5821 def : Pat<(v8i16 (X86vsext (v16i8 VR128:$src))), (PMOVSXBWrr VR128:$src)>;
5822 def : Pat<(v4i32 (X86vsext (v16i8 VR128:$src))), (PMOVSXBDrr VR128:$src)>;
5823 def : Pat<(v2i64 (X86vsext (v16i8 VR128:$src))), (PMOVSXBQrr VR128:$src)>;
5825 def : Pat<(v4i32 (X86vsext (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
5826 def : Pat<(v2i64 (X86vsext (v8i16 VR128:$src))), (PMOVSXWQrr VR128:$src)>;
5828 def : Pat<(v2i64 (X86vsext (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
5830 // Common patterns involving scalar load
5831 def : Pat<(int_x86_sse41_pmovsxbq
5832 (bitconvert (v4i32 (X86vzmovl
5833 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5834 (PMOVSXBQrm addr:$src)>;
5836 def : Pat<(int_x86_sse41_pmovzxbq
5837 (bitconvert (v4i32 (X86vzmovl
5838 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5839 (PMOVZXBQrm addr:$src)>;
5841 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64
5842 (scalar_to_vector (loadi64 addr:$src))))))),
5843 (PMOVSXWDrm addr:$src)>;
5844 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64
5845 (scalar_to_vector (loadf64 addr:$src))))))),
5846 (PMOVSXWDrm addr:$src)>;
5847 def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32
5848 (scalar_to_vector (loadi32 addr:$src))))))),
5849 (PMOVSXBDrm addr:$src)>;
5850 def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32
5851 (scalar_to_vector (loadi32 addr:$src))))))),
5852 (PMOVSXWQrm addr:$src)>;
5853 def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32
5854 (scalar_to_vector (extloadi32i16 addr:$src))))))),
5855 (PMOVSXBQrm addr:$src)>;
5856 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64
5857 (scalar_to_vector (loadi64 addr:$src))))))),
5858 (PMOVSXDQrm addr:$src)>;
5859 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64
5860 (scalar_to_vector (loadf64 addr:$src))))))),
5861 (PMOVSXDQrm addr:$src)>;
5862 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64
5863 (scalar_to_vector (loadi64 addr:$src))))))),
5864 (PMOVSXBWrm addr:$src)>;
5865 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64
5866 (scalar_to_vector (loadf64 addr:$src))))))),
5867 (PMOVSXBWrm addr:$src)>;
5870 let Predicates = [HasAVX2] in {
5871 def : Pat<(v16i16 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBWYrr VR128:$src)>;
5872 def : Pat<(v8i32 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBDYrr VR128:$src)>;
5873 def : Pat<(v4i64 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBQYrr VR128:$src)>;
5875 def : Pat<(v8i32 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWDYrr VR128:$src)>;
5876 def : Pat<(v4i64 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWQYrr VR128:$src)>;
5878 def : Pat<(v4i64 (X86vzext (v4i32 VR128:$src))), (VPMOVZXDQYrr VR128:$src)>;
5880 def : Pat<(v16i16 (X86vzext (v32i8 VR256:$src))),
5881 (VPMOVZXBWYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5882 def : Pat<(v8i32 (X86vzext (v32i8 VR256:$src))),
5883 (VPMOVZXBDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5884 def : Pat<(v4i64 (X86vzext (v32i8 VR256:$src))),
5885 (VPMOVZXBQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5887 def : Pat<(v8i32 (X86vzext (v16i16 VR256:$src))),
5888 (VPMOVZXWDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5889 def : Pat<(v4i64 (X86vzext (v16i16 VR256:$src))),
5890 (VPMOVZXWQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5892 def : Pat<(v4i64 (X86vzext (v8i32 VR256:$src))),
5893 (VPMOVZXDQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5896 let Predicates = [HasAVX] in {
5897 def : Pat<(v8i16 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBWrr VR128:$src)>;
5898 def : Pat<(v4i32 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBDrr VR128:$src)>;
5899 def : Pat<(v2i64 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBQrr VR128:$src)>;
5901 def : Pat<(v4i32 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWDrr VR128:$src)>;
5902 def : Pat<(v2i64 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWQrr VR128:$src)>;
5904 def : Pat<(v2i64 (X86vzext (v4i32 VR128:$src))), (VPMOVZXDQrr VR128:$src)>;
5906 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5907 (VPMOVZXBWrm addr:$src)>;
5908 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5909 (VPMOVZXBWrm addr:$src)>;
5910 def : Pat<(v4i32 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5911 (VPMOVZXBDrm addr:$src)>;
5912 def : Pat<(v2i64 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))),
5913 (VPMOVZXBQrm addr:$src)>;
5915 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5916 (VPMOVZXWDrm addr:$src)>;
5917 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5918 (VPMOVZXWDrm addr:$src)>;
5919 def : Pat<(v2i64 (X86vzext (v8i16 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5920 (VPMOVZXWQrm addr:$src)>;
5922 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5923 (VPMOVZXDQrm addr:$src)>;
5924 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5925 (VPMOVZXDQrm addr:$src)>;
5926 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (X86vzload addr:$src)))))),
5927 (VPMOVZXDQrm addr:$src)>;
5929 def : Pat<(v8i16 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBWrr VR128:$src)>;
5930 def : Pat<(v4i32 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBDrr VR128:$src)>;
5931 def : Pat<(v2i64 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBQrr VR128:$src)>;
5933 def : Pat<(v4i32 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
5934 def : Pat<(v2i64 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWQrr VR128:$src)>;
5936 def : Pat<(v2i64 (X86vsext (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
5938 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64
5939 (scalar_to_vector (loadi64 addr:$src))))))),
5940 (VPMOVSXWDrm addr:$src)>;
5941 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64
5942 (scalar_to_vector (loadi64 addr:$src))))))),
5943 (VPMOVSXDQrm addr:$src)>;
5944 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64
5945 (scalar_to_vector (loadf64 addr:$src))))))),
5946 (VPMOVSXWDrm addr:$src)>;
5947 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64
5948 (scalar_to_vector (loadf64 addr:$src))))))),
5949 (VPMOVSXDQrm addr:$src)>;
5950 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64
5951 (scalar_to_vector (loadi64 addr:$src))))))),
5952 (VPMOVSXBWrm addr:$src)>;
5953 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64
5954 (scalar_to_vector (loadf64 addr:$src))))))),
5955 (VPMOVSXBWrm addr:$src)>;
5957 def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32
5958 (scalar_to_vector (loadi32 addr:$src))))))),
5959 (VPMOVSXBDrm addr:$src)>;
5960 def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32
5961 (scalar_to_vector (loadi32 addr:$src))))))),
5962 (VPMOVSXWQrm addr:$src)>;
5963 def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32
5964 (scalar_to_vector (extloadi32i16 addr:$src))))))),
5965 (VPMOVSXBQrm addr:$src)>;
5968 let Predicates = [UseSSE41] in {
5969 def : Pat<(v8i16 (X86vzext (v16i8 VR128:$src))), (PMOVZXBWrr VR128:$src)>;
5970 def : Pat<(v4i32 (X86vzext (v16i8 VR128:$src))), (PMOVZXBDrr VR128:$src)>;
5971 def : Pat<(v2i64 (X86vzext (v16i8 VR128:$src))), (PMOVZXBQrr VR128:$src)>;
5973 def : Pat<(v4i32 (X86vzext (v8i16 VR128:$src))), (PMOVZXWDrr VR128:$src)>;
5974 def : Pat<(v2i64 (X86vzext (v8i16 VR128:$src))), (PMOVZXWQrr VR128:$src)>;
5976 def : Pat<(v2i64 (X86vzext (v4i32 VR128:$src))), (PMOVZXDQrr VR128:$src)>;
5978 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5979 (PMOVZXBWrm addr:$src)>;
5980 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5981 (PMOVZXBWrm addr:$src)>;
5982 def : Pat<(v4i32 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5983 (PMOVZXBDrm addr:$src)>;
5984 def : Pat<(v2i64 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))),
5985 (PMOVZXBQrm addr:$src)>;
5987 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5988 (PMOVZXWDrm addr:$src)>;
5989 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5990 (PMOVZXWDrm addr:$src)>;
5991 def : Pat<(v2i64 (X86vzext (v8i16 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5992 (PMOVZXWQrm addr:$src)>;
5994 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5995 (PMOVZXDQrm addr:$src)>;
5996 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5997 (PMOVZXDQrm addr:$src)>;
5998 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (X86vzload addr:$src)))))),
5999 (PMOVZXDQrm addr:$src)>;
6002 //===----------------------------------------------------------------------===//
6003 // SSE4.1 - Extract Instructions
6004 //===----------------------------------------------------------------------===//
6006 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
6007 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
6008 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
6009 (ins VR128:$src1, i32i8imm:$src2),
6010 !strconcat(OpcodeStr,
6011 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6012 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
6014 let neverHasSideEffects = 1, mayStore = 1 in
6015 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6016 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
6017 !strconcat(OpcodeStr,
6018 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6021 // There's an AssertZext in the way of writing the store pattern
6022 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
6025 let Predicates = [HasAVX] in {
6026 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
6027 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
6028 (ins VR128:$src1, i32i8imm:$src2),
6029 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
6032 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
6035 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
6036 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
6037 let neverHasSideEffects = 1, mayStore = 1 in
6038 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6039 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
6040 !strconcat(OpcodeStr,
6041 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6044 // There's an AssertZext in the way of writing the store pattern
6045 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
6048 let Predicates = [HasAVX] in
6049 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
6051 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
6054 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
6055 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
6056 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
6057 (ins VR128:$src1, i32i8imm:$src2),
6058 !strconcat(OpcodeStr,
6059 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6061 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
6062 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6063 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
6064 !strconcat(OpcodeStr,
6065 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6066 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
6067 addr:$dst)]>, OpSize;
6070 let Predicates = [HasAVX] in
6071 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
6073 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
6075 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
6076 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
6077 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
6078 (ins VR128:$src1, i32i8imm:$src2),
6079 !strconcat(OpcodeStr,
6080 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6082 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
6083 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6084 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
6085 !strconcat(OpcodeStr,
6086 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6087 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
6088 addr:$dst)]>, OpSize, REX_W;
6091 let Predicates = [HasAVX] in
6092 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
6094 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
6096 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
6098 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
6099 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
6100 (ins VR128:$src1, i32i8imm:$src2),
6101 !strconcat(OpcodeStr,
6102 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6104 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
6106 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6107 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
6108 !strconcat(OpcodeStr,
6109 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6110 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
6111 addr:$dst)]>, OpSize;
6114 let ExeDomain = SSEPackedSingle in {
6115 let Predicates = [UseAVX] in {
6116 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
6117 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
6118 (ins VR128:$src1, i32i8imm:$src2),
6119 "vextractps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6122 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
6125 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
6126 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
6129 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
6131 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
6134 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
6135 Requires<[UseSSE41]>;
6137 //===----------------------------------------------------------------------===//
6138 // SSE4.1 - Insert Instructions
6139 //===----------------------------------------------------------------------===//
6141 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
6142 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6143 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
6145 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6147 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6149 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
6150 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6151 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
6153 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6155 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6157 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
6158 imm:$src3))]>, OpSize;
6161 let Predicates = [HasAVX] in
6162 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
6163 let Constraints = "$src1 = $dst" in
6164 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
6166 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
6167 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6168 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
6170 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6172 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6174 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
6176 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6177 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
6179 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6181 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6183 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
6184 imm:$src3)))]>, OpSize;
6187 let Predicates = [HasAVX] in
6188 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
6189 let Constraints = "$src1 = $dst" in
6190 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
6192 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
6193 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6194 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
6196 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6198 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6200 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
6202 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6203 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
6205 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6207 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6209 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
6210 imm:$src3)))]>, OpSize;
6213 let Predicates = [HasAVX] in
6214 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
6215 let Constraints = "$src1 = $dst" in
6216 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
6218 // insertps has a few different modes, there's the first two here below which
6219 // are optimized inserts that won't zero arbitrary elements in the destination
6220 // vector. The next one matches the intrinsic and could zero arbitrary elements
6221 // in the target vector.
6222 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
6223 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6224 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
6226 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6228 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6230 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
6232 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6233 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
6235 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6237 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6239 (X86insrtps VR128:$src1,
6240 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
6241 imm:$src3))]>, OpSize;
6244 let ExeDomain = SSEPackedSingle in {
6245 let Predicates = [HasAVX] in
6246 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
6247 let Constraints = "$src1 = $dst" in
6248 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
6251 //===----------------------------------------------------------------------===//
6252 // SSE4.1 - Round Instructions
6253 //===----------------------------------------------------------------------===//
6255 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
6256 X86MemOperand x86memop, RegisterClass RC,
6257 PatFrag mem_frag32, PatFrag mem_frag64,
6258 Intrinsic V4F32Int, Intrinsic V2F64Int> {
6259 let ExeDomain = SSEPackedSingle in {
6260 // Intrinsic operation, reg.
6261 // Vector intrinsic operation, reg
6262 def PSr : SS4AIi8<opcps, MRMSrcReg,
6263 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6264 !strconcat(OpcodeStr,
6265 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6266 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
6269 // Vector intrinsic operation, mem
6270 def PSm : SS4AIi8<opcps, MRMSrcMem,
6271 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6272 !strconcat(OpcodeStr,
6273 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6275 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
6277 } // ExeDomain = SSEPackedSingle
6279 let ExeDomain = SSEPackedDouble in {
6280 // Vector intrinsic operation, reg
6281 def PDr : SS4AIi8<opcpd, MRMSrcReg,
6282 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6283 !strconcat(OpcodeStr,
6284 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6285 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
6288 // Vector intrinsic operation, mem
6289 def PDm : SS4AIi8<opcpd, MRMSrcMem,
6290 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6291 !strconcat(OpcodeStr,
6292 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6294 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
6296 } // ExeDomain = SSEPackedDouble
6299 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
6302 Intrinsic F64Int, bit Is2Addr = 1> {
6303 let ExeDomain = GenericDomain in {
6305 let hasSideEffects = 0 in
6306 def SSr : SS4AIi8<opcss, MRMSrcReg,
6307 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3),
6309 !strconcat(OpcodeStr,
6310 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6311 !strconcat(OpcodeStr,
6312 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6315 // Intrinsic operation, reg.
6316 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
6317 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6319 !strconcat(OpcodeStr,
6320 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6321 !strconcat(OpcodeStr,
6322 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6323 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6326 // Intrinsic operation, mem.
6327 def SSm : SS4AIi8<opcss, MRMSrcMem,
6328 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
6330 !strconcat(OpcodeStr,
6331 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6332 !strconcat(OpcodeStr,
6333 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6335 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
6339 let hasSideEffects = 0 in
6340 def SDr : SS4AIi8<opcsd, MRMSrcReg,
6341 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3),
6343 !strconcat(OpcodeStr,
6344 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6345 !strconcat(OpcodeStr,
6346 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6349 // Intrinsic operation, reg.
6350 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
6351 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6353 !strconcat(OpcodeStr,
6354 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6355 !strconcat(OpcodeStr,
6356 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6357 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6360 // Intrinsic operation, mem.
6361 def SDm : SS4AIi8<opcsd, MRMSrcMem,
6362 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
6364 !strconcat(OpcodeStr,
6365 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6366 !strconcat(OpcodeStr,
6367 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6369 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
6371 } // ExeDomain = GenericDomain
6374 // FP round - roundss, roundps, roundsd, roundpd
6375 let Predicates = [HasAVX] in {
6377 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
6378 memopv4f32, memopv2f64,
6379 int_x86_sse41_round_ps,
6380 int_x86_sse41_round_pd>, VEX;
6381 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
6382 memopv8f32, memopv4f64,
6383 int_x86_avx_round_ps_256,
6384 int_x86_avx_round_pd_256>, VEX, VEX_L;
6385 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
6386 int_x86_sse41_round_ss,
6387 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
6389 def : Pat<(ffloor FR32:$src),
6390 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6391 def : Pat<(f64 (ffloor FR64:$src)),
6392 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6393 def : Pat<(f32 (fnearbyint FR32:$src)),
6394 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6395 def : Pat<(f64 (fnearbyint FR64:$src)),
6396 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6397 def : Pat<(f32 (fceil FR32:$src)),
6398 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6399 def : Pat<(f64 (fceil FR64:$src)),
6400 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6401 def : Pat<(f32 (frint FR32:$src)),
6402 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6403 def : Pat<(f64 (frint FR64:$src)),
6404 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6405 def : Pat<(f32 (ftrunc FR32:$src)),
6406 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6407 def : Pat<(f64 (ftrunc FR64:$src)),
6408 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6410 def : Pat<(v4f32 (ffloor VR128:$src)),
6411 (VROUNDPSr VR128:$src, (i32 0x1))>;
6412 def : Pat<(v4f32 (fnearbyint VR128:$src)),
6413 (VROUNDPSr VR128:$src, (i32 0xC))>;
6414 def : Pat<(v4f32 (fceil VR128:$src)),
6415 (VROUNDPSr VR128:$src, (i32 0x2))>;
6416 def : Pat<(v4f32 (frint VR128:$src)),
6417 (VROUNDPSr VR128:$src, (i32 0x4))>;
6418 def : Pat<(v4f32 (ftrunc VR128:$src)),
6419 (VROUNDPSr VR128:$src, (i32 0x3))>;
6421 def : Pat<(v2f64 (ffloor VR128:$src)),
6422 (VROUNDPDr VR128:$src, (i32 0x1))>;
6423 def : Pat<(v2f64 (fnearbyint VR128:$src)),
6424 (VROUNDPDr VR128:$src, (i32 0xC))>;
6425 def : Pat<(v2f64 (fceil VR128:$src)),
6426 (VROUNDPDr VR128:$src, (i32 0x2))>;
6427 def : Pat<(v2f64 (frint VR128:$src)),
6428 (VROUNDPDr VR128:$src, (i32 0x4))>;
6429 def : Pat<(v2f64 (ftrunc VR128:$src)),
6430 (VROUNDPDr VR128:$src, (i32 0x3))>;
6432 def : Pat<(v8f32 (ffloor VR256:$src)),
6433 (VROUNDYPSr VR256:$src, (i32 0x1))>;
6434 def : Pat<(v8f32 (fnearbyint VR256:$src)),
6435 (VROUNDYPSr VR256:$src, (i32 0xC))>;
6436 def : Pat<(v8f32 (fceil VR256:$src)),
6437 (VROUNDYPSr VR256:$src, (i32 0x2))>;
6438 def : Pat<(v8f32 (frint VR256:$src)),
6439 (VROUNDYPSr VR256:$src, (i32 0x4))>;
6440 def : Pat<(v8f32 (ftrunc VR256:$src)),
6441 (VROUNDYPSr VR256:$src, (i32 0x3))>;
6443 def : Pat<(v4f64 (ffloor VR256:$src)),
6444 (VROUNDYPDr VR256:$src, (i32 0x1))>;
6445 def : Pat<(v4f64 (fnearbyint VR256:$src)),
6446 (VROUNDYPDr VR256:$src, (i32 0xC))>;
6447 def : Pat<(v4f64 (fceil VR256:$src)),
6448 (VROUNDYPDr VR256:$src, (i32 0x2))>;
6449 def : Pat<(v4f64 (frint VR256:$src)),
6450 (VROUNDYPDr VR256:$src, (i32 0x4))>;
6451 def : Pat<(v4f64 (ftrunc VR256:$src)),
6452 (VROUNDYPDr VR256:$src, (i32 0x3))>;
6455 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
6456 memopv4f32, memopv2f64,
6457 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
6458 let Constraints = "$src1 = $dst" in
6459 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
6460 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
6462 let Predicates = [UseSSE41] in {
6463 def : Pat<(ffloor FR32:$src),
6464 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6465 def : Pat<(f64 (ffloor FR64:$src)),
6466 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6467 def : Pat<(f32 (fnearbyint FR32:$src)),
6468 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6469 def : Pat<(f64 (fnearbyint FR64:$src)),
6470 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6471 def : Pat<(f32 (fceil FR32:$src)),
6472 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6473 def : Pat<(f64 (fceil FR64:$src)),
6474 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6475 def : Pat<(f32 (frint FR32:$src)),
6476 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6477 def : Pat<(f64 (frint FR64:$src)),
6478 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6479 def : Pat<(f32 (ftrunc FR32:$src)),
6480 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6481 def : Pat<(f64 (ftrunc FR64:$src)),
6482 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6484 def : Pat<(v4f32 (ffloor VR128:$src)),
6485 (ROUNDPSr VR128:$src, (i32 0x1))>;
6486 def : Pat<(v4f32 (fnearbyint VR128:$src)),
6487 (ROUNDPSr VR128:$src, (i32 0xC))>;
6488 def : Pat<(v4f32 (fceil VR128:$src)),
6489 (ROUNDPSr VR128:$src, (i32 0x2))>;
6490 def : Pat<(v4f32 (frint VR128:$src)),
6491 (ROUNDPSr VR128:$src, (i32 0x4))>;
6492 def : Pat<(v4f32 (ftrunc VR128:$src)),
6493 (ROUNDPSr VR128:$src, (i32 0x3))>;
6495 def : Pat<(v2f64 (ffloor VR128:$src)),
6496 (ROUNDPDr VR128:$src, (i32 0x1))>;
6497 def : Pat<(v2f64 (fnearbyint VR128:$src)),
6498 (ROUNDPDr VR128:$src, (i32 0xC))>;
6499 def : Pat<(v2f64 (fceil VR128:$src)),
6500 (ROUNDPDr VR128:$src, (i32 0x2))>;
6501 def : Pat<(v2f64 (frint VR128:$src)),
6502 (ROUNDPDr VR128:$src, (i32 0x4))>;
6503 def : Pat<(v2f64 (ftrunc VR128:$src)),
6504 (ROUNDPDr VR128:$src, (i32 0x3))>;
6507 //===----------------------------------------------------------------------===//
6508 // SSE4.1 - Packed Bit Test
6509 //===----------------------------------------------------------------------===//
6511 // ptest instruction we'll lower to this in X86ISelLowering primarily from
6512 // the intel intrinsic that corresponds to this.
6513 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6514 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6515 "vptest\t{$src2, $src1|$src1, $src2}",
6516 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6518 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6519 "vptest\t{$src2, $src1|$src1, $src2}",
6520 [(set EFLAGS,(X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6523 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
6524 "vptest\t{$src2, $src1|$src1, $src2}",
6525 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
6527 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
6528 "vptest\t{$src2, $src1|$src1, $src2}",
6529 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
6533 let Defs = [EFLAGS] in {
6534 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6535 "ptest\t{$src2, $src1|$src1, $src2}",
6536 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6538 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6539 "ptest\t{$src2, $src1|$src1, $src2}",
6540 [(set EFLAGS, (X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6544 // The bit test instructions below are AVX only
6545 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
6546 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
6547 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
6548 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6549 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
6550 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
6551 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6552 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
6556 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6557 let ExeDomain = SSEPackedSingle in {
6558 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
6559 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>,
6562 let ExeDomain = SSEPackedDouble in {
6563 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
6564 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>,
6569 //===----------------------------------------------------------------------===//
6570 // SSE4.1 - Misc Instructions
6571 //===----------------------------------------------------------------------===//
6573 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
6574 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
6575 "popcnt{w}\t{$src, $dst|$dst, $src}",
6576 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)]>,
6578 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
6579 "popcnt{w}\t{$src, $dst|$dst, $src}",
6580 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
6581 (implicit EFLAGS)]>, OpSize, XS;
6583 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
6584 "popcnt{l}\t{$src, $dst|$dst, $src}",
6585 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)]>,
6587 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
6588 "popcnt{l}\t{$src, $dst|$dst, $src}",
6589 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
6590 (implicit EFLAGS)]>, XS;
6592 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
6593 "popcnt{q}\t{$src, $dst|$dst, $src}",
6594 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)]>,
6596 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
6597 "popcnt{q}\t{$src, $dst|$dst, $src}",
6598 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
6599 (implicit EFLAGS)]>, XS;
6604 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
6605 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
6606 Intrinsic IntId128> {
6607 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6609 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6610 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
6611 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6613 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6616 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
6619 let Predicates = [HasAVX] in
6620 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
6621 int_x86_sse41_phminposuw>, VEX;
6622 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
6623 int_x86_sse41_phminposuw>;
6625 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6626 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
6627 Intrinsic IntId128, bit Is2Addr = 1> {
6628 let isCommutable = 1 in
6629 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6630 (ins VR128:$src1, VR128:$src2),
6632 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6633 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6634 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
6635 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6636 (ins VR128:$src1, i128mem:$src2),
6638 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6639 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6641 (IntId128 VR128:$src1,
6642 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
6645 /// SS41I_binop_rm_int_y - Simple SSE 4.1 binary operator
6646 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6647 Intrinsic IntId256> {
6648 let isCommutable = 1 in
6649 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6650 (ins VR256:$src1, VR256:$src2),
6651 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6652 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
6653 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6654 (ins VR256:$src1, i256mem:$src2),
6655 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6657 (IntId256 VR256:$src1,
6658 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
6662 /// SS48I_binop_rm - Simple SSE41 binary operator.
6663 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6664 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6665 X86MemOperand x86memop, bit Is2Addr = 1> {
6666 let isCommutable = 1 in
6667 def rr : SS48I<opc, MRMSrcReg, (outs RC:$dst),
6668 (ins RC:$src1, RC:$src2),
6670 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6671 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6672 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>, OpSize;
6673 def rm : SS48I<opc, MRMSrcMem, (outs RC:$dst),
6674 (ins RC:$src1, x86memop:$src2),
6676 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6677 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6679 (OpVT (OpNode RC:$src1,
6680 (bitconvert (memop_frag addr:$src2)))))]>, OpSize;
6683 let Predicates = [HasAVX] in {
6684 let isCommutable = 0 in
6685 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
6687 defm VPMINSB : SS48I_binop_rm<0x38, "vpminsb", X86smin, v16i8, VR128,
6688 memopv2i64, i128mem, 0>, VEX_4V;
6689 defm VPMINSD : SS48I_binop_rm<0x39, "vpminsd", X86smin, v4i32, VR128,
6690 memopv2i64, i128mem, 0>, VEX_4V;
6691 defm VPMINUD : SS48I_binop_rm<0x3B, "vpminud", X86umin, v4i32, VR128,
6692 memopv2i64, i128mem, 0>, VEX_4V;
6693 defm VPMINUW : SS48I_binop_rm<0x3A, "vpminuw", X86umin, v8i16, VR128,
6694 memopv2i64, i128mem, 0>, VEX_4V;
6695 defm VPMAXSB : SS48I_binop_rm<0x3C, "vpmaxsb", X86smax, v16i8, VR128,
6696 memopv2i64, i128mem, 0>, VEX_4V;
6697 defm VPMAXSD : SS48I_binop_rm<0x3D, "vpmaxsd", X86smax, v4i32, VR128,
6698 memopv2i64, i128mem, 0>, VEX_4V;
6699 defm VPMAXUD : SS48I_binop_rm<0x3F, "vpmaxud", X86umax, v4i32, VR128,
6700 memopv2i64, i128mem, 0>, VEX_4V;
6701 defm VPMAXUW : SS48I_binop_rm<0x3E, "vpmaxuw", X86umax, v8i16, VR128,
6702 memopv2i64, i128mem, 0>, VEX_4V;
6703 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
6707 let Predicates = [HasAVX2] in {
6708 let isCommutable = 0 in
6709 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
6710 int_x86_avx2_packusdw>, VEX_4V, VEX_L;
6711 defm VPMINSBY : SS48I_binop_rm<0x38, "vpminsb", X86smin, v32i8, VR256,
6712 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6713 defm VPMINSDY : SS48I_binop_rm<0x39, "vpminsd", X86smin, v8i32, VR256,
6714 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6715 defm VPMINUDY : SS48I_binop_rm<0x3B, "vpminud", X86umin, v8i32, VR256,
6716 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6717 defm VPMINUWY : SS48I_binop_rm<0x3A, "vpminuw", X86umin, v16i16, VR256,
6718 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6719 defm VPMAXSBY : SS48I_binop_rm<0x3C, "vpmaxsb", X86smax, v32i8, VR256,
6720 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6721 defm VPMAXSDY : SS48I_binop_rm<0x3D, "vpmaxsd", X86smax, v8i32, VR256,
6722 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6723 defm VPMAXUDY : SS48I_binop_rm<0x3F, "vpmaxud", X86umax, v8i32, VR256,
6724 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6725 defm VPMAXUWY : SS48I_binop_rm<0x3E, "vpmaxuw", X86umax, v16i16, VR256,
6726 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6727 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
6728 int_x86_avx2_pmul_dq>, VEX_4V, VEX_L;
6731 let Constraints = "$src1 = $dst" in {
6732 let isCommutable = 0 in
6733 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
6734 defm PMINSB : SS48I_binop_rm<0x38, "pminsb", X86smin, v16i8, VR128,
6735 memopv2i64, i128mem>;
6736 defm PMINSD : SS48I_binop_rm<0x39, "pminsd", X86smin, v4i32, VR128,
6737 memopv2i64, i128mem>;
6738 defm PMINUD : SS48I_binop_rm<0x3B, "pminud", X86umin, v4i32, VR128,
6739 memopv2i64, i128mem>;
6740 defm PMINUW : SS48I_binop_rm<0x3A, "pminuw", X86umin, v8i16, VR128,
6741 memopv2i64, i128mem>;
6742 defm PMAXSB : SS48I_binop_rm<0x3C, "pmaxsb", X86smax, v16i8, VR128,
6743 memopv2i64, i128mem>;
6744 defm PMAXSD : SS48I_binop_rm<0x3D, "pmaxsd", X86smax, v4i32, VR128,
6745 memopv2i64, i128mem>;
6746 defm PMAXUD : SS48I_binop_rm<0x3F, "pmaxud", X86umax, v4i32, VR128,
6747 memopv2i64, i128mem>;
6748 defm PMAXUW : SS48I_binop_rm<0x3E, "pmaxuw", X86umax, v8i16, VR128,
6749 memopv2i64, i128mem>;
6750 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
6753 let Predicates = [HasAVX] in {
6754 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, VR128,
6755 memopv2i64, i128mem, 0>, VEX_4V;
6756 defm VPCMPEQQ : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v2i64, VR128,
6757 memopv2i64, i128mem, 0>, VEX_4V;
6759 let Predicates = [HasAVX2] in {
6760 defm VPMULLDY : SS48I_binop_rm<0x40, "vpmulld", mul, v8i32, VR256,
6761 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6762 defm VPCMPEQQY : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v4i64, VR256,
6763 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6766 let Constraints = "$src1 = $dst" in {
6767 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32, VR128,
6768 memopv2i64, i128mem>;
6769 defm PCMPEQQ : SS48I_binop_rm<0x29, "pcmpeqq", X86pcmpeq, v2i64, VR128,
6770 memopv2i64, i128mem>;
6773 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
6774 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
6775 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6776 X86MemOperand x86memop, bit Is2Addr = 1> {
6777 let isCommutable = 1 in
6778 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
6779 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6781 !strconcat(OpcodeStr,
6782 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6783 !strconcat(OpcodeStr,
6784 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6785 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
6787 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
6788 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6790 !strconcat(OpcodeStr,
6791 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6792 !strconcat(OpcodeStr,
6793 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6796 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
6800 let Predicates = [HasAVX] in {
6801 let isCommutable = 0 in {
6802 let ExeDomain = SSEPackedSingle in {
6803 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
6804 VR128, memopv4f32, f128mem, 0>, VEX_4V;
6805 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
6806 int_x86_avx_blend_ps_256, VR256, memopv8f32,
6807 f256mem, 0>, VEX_4V, VEX_L;
6809 let ExeDomain = SSEPackedDouble in {
6810 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
6811 VR128, memopv2f64, f128mem, 0>, VEX_4V;
6812 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
6813 int_x86_avx_blend_pd_256,VR256, memopv4f64,
6814 f256mem, 0>, VEX_4V, VEX_L;
6816 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
6817 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6818 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
6819 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6821 let ExeDomain = SSEPackedSingle in
6822 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
6823 VR128, memopv4f32, f128mem, 0>, VEX_4V;
6824 let ExeDomain = SSEPackedDouble in
6825 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
6826 VR128, memopv2f64, f128mem, 0>, VEX_4V;
6827 let ExeDomain = SSEPackedSingle in
6828 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
6829 VR256, memopv8f32, i256mem, 0>, VEX_4V, VEX_L;
6832 let Predicates = [HasAVX2] in {
6833 let isCommutable = 0 in {
6834 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
6835 VR256, memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6836 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
6837 VR256, memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6841 let Constraints = "$src1 = $dst" in {
6842 let isCommutable = 0 in {
6843 let ExeDomain = SSEPackedSingle in
6844 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
6845 VR128, memopv4f32, f128mem>;
6846 let ExeDomain = SSEPackedDouble in
6847 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
6848 VR128, memopv2f64, f128mem>;
6849 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
6850 VR128, memopv2i64, i128mem>;
6851 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
6852 VR128, memopv2i64, i128mem>;
6854 let ExeDomain = SSEPackedSingle in
6855 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
6856 VR128, memopv4f32, f128mem>;
6857 let ExeDomain = SSEPackedDouble in
6858 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
6859 VR128, memopv2f64, f128mem>;
6862 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
6863 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
6864 RegisterClass RC, X86MemOperand x86memop,
6865 PatFrag mem_frag, Intrinsic IntId> {
6866 def rr : Ii8<opc, MRMSrcReg, (outs RC:$dst),
6867 (ins RC:$src1, RC:$src2, RC:$src3),
6868 !strconcat(OpcodeStr,
6869 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6870 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
6871 NoItinerary, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6873 def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
6874 (ins RC:$src1, x86memop:$src2, RC:$src3),
6875 !strconcat(OpcodeStr,
6876 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6878 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
6880 NoItinerary, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6883 let Predicates = [HasAVX] in {
6884 let ExeDomain = SSEPackedDouble in {
6885 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, f128mem,
6886 memopv2f64, int_x86_sse41_blendvpd>;
6887 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, f256mem,
6888 memopv4f64, int_x86_avx_blendv_pd_256>, VEX_L;
6889 } // ExeDomain = SSEPackedDouble
6890 let ExeDomain = SSEPackedSingle in {
6891 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, f128mem,
6892 memopv4f32, int_x86_sse41_blendvps>;
6893 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, f256mem,
6894 memopv8f32, int_x86_avx_blendv_ps_256>, VEX_L;
6895 } // ExeDomain = SSEPackedSingle
6896 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
6897 memopv2i64, int_x86_sse41_pblendvb>;
6900 let Predicates = [HasAVX2] in {
6901 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
6902 memopv4i64, int_x86_avx2_pblendvb>, VEX_L;
6905 let Predicates = [HasAVX] in {
6906 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
6907 (v16i8 VR128:$src2))),
6908 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6909 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
6910 (v4i32 VR128:$src2))),
6911 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6912 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
6913 (v4f32 VR128:$src2))),
6914 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6915 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
6916 (v2i64 VR128:$src2))),
6917 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6918 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
6919 (v2f64 VR128:$src2))),
6920 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6921 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
6922 (v8i32 VR256:$src2))),
6923 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6924 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
6925 (v8f32 VR256:$src2))),
6926 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6927 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
6928 (v4i64 VR256:$src2))),
6929 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6930 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
6931 (v4f64 VR256:$src2))),
6932 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6934 def : Pat<(v8f32 (X86Blendi (v8f32 VR256:$src1), (v8f32 VR256:$src2),
6936 (VBLENDPSYrri VR256:$src1, VR256:$src2, imm:$mask)>;
6937 def : Pat<(v4f64 (X86Blendi (v4f64 VR256:$src1), (v4f64 VR256:$src2),
6939 (VBLENDPDYrri VR256:$src1, VR256:$src2, imm:$mask)>;
6941 def : Pat<(v8i16 (X86Blendi (v8i16 VR128:$src1), (v8i16 VR128:$src2),
6943 (VPBLENDWrri VR128:$src1, VR128:$src2, imm:$mask)>;
6944 def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$src1), (v4f32 VR128:$src2),
6946 (VBLENDPSrri VR128:$src1, VR128:$src2, imm:$mask)>;
6947 def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$src1), (v2f64 VR128:$src2),
6949 (VBLENDPDrri VR128:$src1, VR128:$src2, imm:$mask)>;
6952 let Predicates = [HasAVX2] in {
6953 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
6954 (v32i8 VR256:$src2))),
6955 (VPBLENDVBYrr VR256:$src1, VR256:$src2, VR256:$mask)>;
6956 def : Pat<(v16i16 (X86Blendi (v16i16 VR256:$src1), (v16i16 VR256:$src2),
6958 (VPBLENDWYrri VR256:$src1, VR256:$src2, imm:$mask)>;
6961 /// SS41I_ternary_int - SSE 4.1 ternary operator
6962 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
6963 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
6964 X86MemOperand x86memop, Intrinsic IntId> {
6965 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6966 (ins VR128:$src1, VR128:$src2),
6967 !strconcat(OpcodeStr,
6968 "\t{$src2, $dst|$dst, $src2}"),
6969 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
6972 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6973 (ins VR128:$src1, x86memop:$src2),
6974 !strconcat(OpcodeStr,
6975 "\t{$src2, $dst|$dst, $src2}"),
6978 (bitconvert (mem_frag addr:$src2)), XMM0))]>, OpSize;
6982 let ExeDomain = SSEPackedDouble in
6983 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64, f128mem,
6984 int_x86_sse41_blendvpd>;
6985 let ExeDomain = SSEPackedSingle in
6986 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32, f128mem,
6987 int_x86_sse41_blendvps>;
6988 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64, i128mem,
6989 int_x86_sse41_pblendvb>;
6991 // Aliases with the implicit xmm0 argument
6992 def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
6993 (BLENDVPDrr0 VR128:$dst, VR128:$src2)>;
6994 def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
6995 (BLENDVPDrm0 VR128:$dst, f128mem:$src2)>;
6996 def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
6997 (BLENDVPSrr0 VR128:$dst, VR128:$src2)>;
6998 def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
6999 (BLENDVPSrm0 VR128:$dst, f128mem:$src2)>;
7000 def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7001 (PBLENDVBrr0 VR128:$dst, VR128:$src2)>;
7002 def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
7003 (PBLENDVBrm0 VR128:$dst, i128mem:$src2)>;
7005 let Predicates = [UseSSE41] in {
7006 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
7007 (v16i8 VR128:$src2))),
7008 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
7009 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
7010 (v4i32 VR128:$src2))),
7011 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
7012 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
7013 (v4f32 VR128:$src2))),
7014 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
7015 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
7016 (v2i64 VR128:$src2))),
7017 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
7018 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
7019 (v2f64 VR128:$src2))),
7020 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
7022 def : Pat<(v8i16 (X86Blendi (v8i16 VR128:$src1), (v8i16 VR128:$src2),
7024 (PBLENDWrri VR128:$src1, VR128:$src2, imm:$mask)>;
7025 def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$src1), (v4f32 VR128:$src2),
7027 (BLENDPSrri VR128:$src1, VR128:$src2, imm:$mask)>;
7028 def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$src1), (v2f64 VR128:$src2),
7030 (BLENDPDrri VR128:$src1, VR128:$src2, imm:$mask)>;
7034 let Predicates = [HasAVX] in
7035 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
7036 "vmovntdqa\t{$src, $dst|$dst, $src}",
7037 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
7039 let Predicates = [HasAVX2] in
7040 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
7041 "vmovntdqa\t{$src, $dst|$dst, $src}",
7042 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
7044 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
7045 "movntdqa\t{$src, $dst|$dst, $src}",
7046 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
7049 //===----------------------------------------------------------------------===//
7050 // SSE4.2 - Compare Instructions
7051 //===----------------------------------------------------------------------===//
7053 /// SS42I_binop_rm - Simple SSE 4.2 binary operator
7054 multiclass SS42I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
7055 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
7056 X86MemOperand x86memop, bit Is2Addr = 1> {
7057 def rr : SS428I<opc, MRMSrcReg, (outs RC:$dst),
7058 (ins RC:$src1, RC:$src2),
7060 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7061 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7062 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>,
7064 def rm : SS428I<opc, MRMSrcMem, (outs RC:$dst),
7065 (ins RC:$src1, x86memop:$src2),
7067 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7068 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7070 (OpVT (OpNode RC:$src1, (memop_frag addr:$src2))))]>, OpSize;
7073 let Predicates = [HasAVX] in
7074 defm VPCMPGTQ : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v2i64, VR128,
7075 memopv2i64, i128mem, 0>, VEX_4V;
7077 let Predicates = [HasAVX2] in
7078 defm VPCMPGTQY : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v4i64, VR256,
7079 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
7081 let Constraints = "$src1 = $dst" in
7082 defm PCMPGTQ : SS42I_binop_rm<0x37, "pcmpgtq", X86pcmpgt, v2i64, VR128,
7083 memopv2i64, i128mem>;
7085 //===----------------------------------------------------------------------===//
7086 // SSE4.2 - String/text Processing Instructions
7087 //===----------------------------------------------------------------------===//
7089 // Packed Compare Implicit Length Strings, Return Mask
7090 multiclass pseudo_pcmpistrm<string asm> {
7091 def REG : PseudoI<(outs VR128:$dst),
7092 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7093 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
7095 def MEM : PseudoI<(outs VR128:$dst),
7096 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7097 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1,
7098 (bc_v16i8 (memopv2i64 addr:$src2)), imm:$src3))]>;
7101 let Defs = [EFLAGS], usesCustomInserter = 1 in {
7102 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
7103 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[UseSSE42]>;
7106 multiclass pcmpistrm_SS42AI<string asm> {
7107 def rr : SS42AI<0x62, MRMSrcReg, (outs),
7108 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7109 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
7112 def rm :SS42AI<0x62, MRMSrcMem, (outs),
7113 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7114 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
7118 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
7119 let Predicates = [HasAVX] in
7120 defm VPCMPISTRM128 : pcmpistrm_SS42AI<"vpcmpistrm">, VEX;
7121 defm PCMPISTRM128 : pcmpistrm_SS42AI<"pcmpistrm"> ;
7124 // Packed Compare Explicit Length Strings, Return Mask
7125 multiclass pseudo_pcmpestrm<string asm> {
7126 def REG : PseudoI<(outs VR128:$dst),
7127 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7128 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
7129 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
7130 def MEM : PseudoI<(outs VR128:$dst),
7131 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7132 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128 VR128:$src1, EAX,
7133 (bc_v16i8 (memopv2i64 addr:$src3)), EDX, imm:$src5))]>;
7136 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
7137 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
7138 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[UseSSE42]>;
7141 multiclass SS42AI_pcmpestrm<string asm> {
7142 def rr : SS42AI<0x60, MRMSrcReg, (outs),
7143 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7144 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7147 def rm : SS42AI<0x60, MRMSrcMem, (outs),
7148 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7149 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7153 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
7154 let Predicates = [HasAVX] in
7155 defm VPCMPESTRM128 : SS42AI_pcmpestrm<"vpcmpestrm">, VEX;
7156 defm PCMPESTRM128 : SS42AI_pcmpestrm<"pcmpestrm">;
7159 // Packed Compare Implicit Length Strings, Return Index
7160 multiclass pseudo_pcmpistri<string asm> {
7161 def REG : PseudoI<(outs GR32:$dst),
7162 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7163 [(set GR32:$dst, EFLAGS,
7164 (X86pcmpistri VR128:$src1, VR128:$src2, imm:$src3))]>;
7165 def MEM : PseudoI<(outs GR32:$dst),
7166 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7167 [(set GR32:$dst, EFLAGS, (X86pcmpistri VR128:$src1,
7168 (bc_v16i8 (memopv2i64 addr:$src2)), imm:$src3))]>;
7171 let Defs = [EFLAGS], usesCustomInserter = 1 in {
7172 defm VPCMPISTRI : pseudo_pcmpistri<"#VPCMPISTRI">, Requires<[HasAVX]>;
7173 defm PCMPISTRI : pseudo_pcmpistri<"#PCMPISTRI">, Requires<[UseSSE42]>;
7176 multiclass SS42AI_pcmpistri<string asm> {
7177 def rr : SS42AI<0x63, MRMSrcReg, (outs),
7178 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7179 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
7182 def rm : SS42AI<0x63, MRMSrcMem, (outs),
7183 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7184 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
7188 let Defs = [ECX, EFLAGS], neverHasSideEffects = 1 in {
7189 let Predicates = [HasAVX] in
7190 defm VPCMPISTRI : SS42AI_pcmpistri<"vpcmpistri">, VEX;
7191 defm PCMPISTRI : SS42AI_pcmpistri<"pcmpistri">;
7194 // Packed Compare Explicit Length Strings, Return Index
7195 multiclass pseudo_pcmpestri<string asm> {
7196 def REG : PseudoI<(outs GR32:$dst),
7197 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7198 [(set GR32:$dst, EFLAGS,
7199 (X86pcmpestri VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
7200 def MEM : PseudoI<(outs GR32:$dst),
7201 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7202 [(set GR32:$dst, EFLAGS,
7203 (X86pcmpestri VR128:$src1, EAX, (bc_v16i8 (memopv2i64 addr:$src3)), EDX,
7207 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
7208 defm VPCMPESTRI : pseudo_pcmpestri<"#VPCMPESTRI">, Requires<[HasAVX]>;
7209 defm PCMPESTRI : pseudo_pcmpestri<"#PCMPESTRI">, Requires<[UseSSE42]>;
7212 multiclass SS42AI_pcmpestri<string asm> {
7213 def rr : SS42AI<0x61, MRMSrcReg, (outs),
7214 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7215 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7218 def rm : SS42AI<0x61, MRMSrcMem, (outs),
7219 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7220 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7224 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
7225 let Predicates = [HasAVX] in
7226 defm VPCMPESTRI : SS42AI_pcmpestri<"vpcmpestri">, VEX;
7227 defm PCMPESTRI : SS42AI_pcmpestri<"pcmpestri">;
7230 //===----------------------------------------------------------------------===//
7231 // SSE4.2 - CRC Instructions
7232 //===----------------------------------------------------------------------===//
7234 // No CRC instructions have AVX equivalents
7236 // crc intrinsic instruction
7237 // This set of instructions are only rm, the only difference is the size
7239 let Constraints = "$src1 = $dst" in {
7240 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
7241 (ins GR32:$src1, i8mem:$src2),
7242 "crc32{b}\t{$src2, $src1|$src1, $src2}",
7244 (int_x86_sse42_crc32_32_8 GR32:$src1,
7245 (load addr:$src2)))]>;
7246 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
7247 (ins GR32:$src1, GR8:$src2),
7248 "crc32{b}\t{$src2, $src1|$src1, $src2}",
7250 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
7251 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
7252 (ins GR32:$src1, i16mem:$src2),
7253 "crc32{w}\t{$src2, $src1|$src1, $src2}",
7255 (int_x86_sse42_crc32_32_16 GR32:$src1,
7256 (load addr:$src2)))]>,
7258 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
7259 (ins GR32:$src1, GR16:$src2),
7260 "crc32{w}\t{$src2, $src1|$src1, $src2}",
7262 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
7264 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
7265 (ins GR32:$src1, i32mem:$src2),
7266 "crc32{l}\t{$src2, $src1|$src1, $src2}",
7268 (int_x86_sse42_crc32_32_32 GR32:$src1,
7269 (load addr:$src2)))]>;
7270 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
7271 (ins GR32:$src1, GR32:$src2),
7272 "crc32{l}\t{$src2, $src1|$src1, $src2}",
7274 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
7275 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
7276 (ins GR64:$src1, i8mem:$src2),
7277 "crc32{b}\t{$src2, $src1|$src1, $src2}",
7279 (int_x86_sse42_crc32_64_8 GR64:$src1,
7280 (load addr:$src2)))]>,
7282 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
7283 (ins GR64:$src1, GR8:$src2),
7284 "crc32{b}\t{$src2, $src1|$src1, $src2}",
7286 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
7288 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
7289 (ins GR64:$src1, i64mem:$src2),
7290 "crc32{q}\t{$src2, $src1|$src1, $src2}",
7292 (int_x86_sse42_crc32_64_64 GR64:$src1,
7293 (load addr:$src2)))]>,
7295 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
7296 (ins GR64:$src1, GR64:$src2),
7297 "crc32{q}\t{$src2, $src1|$src1, $src2}",
7299 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
7303 //===----------------------------------------------------------------------===//
7304 // AES-NI Instructions
7305 //===----------------------------------------------------------------------===//
7307 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
7308 Intrinsic IntId128, bit Is2Addr = 1> {
7309 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
7310 (ins VR128:$src1, VR128:$src2),
7312 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7313 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7314 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
7316 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
7317 (ins VR128:$src1, i128mem:$src2),
7319 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7320 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7322 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
7325 // Perform One Round of an AES Encryption/Decryption Flow
7326 let Predicates = [HasAVX, HasAES] in {
7327 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
7328 int_x86_aesni_aesenc, 0>, VEX_4V;
7329 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
7330 int_x86_aesni_aesenclast, 0>, VEX_4V;
7331 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
7332 int_x86_aesni_aesdec, 0>, VEX_4V;
7333 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
7334 int_x86_aesni_aesdeclast, 0>, VEX_4V;
7337 let Constraints = "$src1 = $dst" in {
7338 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
7339 int_x86_aesni_aesenc>;
7340 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
7341 int_x86_aesni_aesenclast>;
7342 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
7343 int_x86_aesni_aesdec>;
7344 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
7345 int_x86_aesni_aesdeclast>;
7348 // Perform the AES InvMixColumn Transformation
7349 let Predicates = [HasAVX, HasAES] in {
7350 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7352 "vaesimc\t{$src1, $dst|$dst, $src1}",
7354 (int_x86_aesni_aesimc VR128:$src1))]>,
7356 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7357 (ins i128mem:$src1),
7358 "vaesimc\t{$src1, $dst|$dst, $src1}",
7359 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7362 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7364 "aesimc\t{$src1, $dst|$dst, $src1}",
7366 (int_x86_aesni_aesimc VR128:$src1))]>,
7368 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7369 (ins i128mem:$src1),
7370 "aesimc\t{$src1, $dst|$dst, $src1}",
7371 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7374 // AES Round Key Generation Assist
7375 let Predicates = [HasAVX, HasAES] in {
7376 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7377 (ins VR128:$src1, i8imm:$src2),
7378 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7380 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7382 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7383 (ins i128mem:$src1, i8imm:$src2),
7384 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7386 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7389 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7390 (ins VR128:$src1, i8imm:$src2),
7391 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7393 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7395 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7396 (ins i128mem:$src1, i8imm:$src2),
7397 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7399 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7402 //===----------------------------------------------------------------------===//
7403 // PCLMUL Instructions
7404 //===----------------------------------------------------------------------===//
7406 // AVX carry-less Multiplication instructions
7407 def VPCLMULQDQrr : AVXPCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7408 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7409 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7411 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
7413 def VPCLMULQDQrm : AVXPCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7414 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7415 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7416 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7417 (memopv2i64 addr:$src2), imm:$src3))]>;
7419 // Carry-less Multiplication instructions
7420 let Constraints = "$src1 = $dst" in {
7421 def PCLMULQDQrr : PCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7422 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7423 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7425 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
7427 def PCLMULQDQrm : PCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7428 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7429 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7430 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7431 (memopv2i64 addr:$src2), imm:$src3))]>;
7432 } // Constraints = "$src1 = $dst"
7435 multiclass pclmul_alias<string asm, int immop> {
7436 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7437 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
7439 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7440 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
7442 def : InstAlias<!strconcat("vpclmul", asm,
7443 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7444 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
7446 def : InstAlias<!strconcat("vpclmul", asm,
7447 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7448 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
7450 defm : pclmul_alias<"hqhq", 0x11>;
7451 defm : pclmul_alias<"hqlq", 0x01>;
7452 defm : pclmul_alias<"lqhq", 0x10>;
7453 defm : pclmul_alias<"lqlq", 0x00>;
7455 //===----------------------------------------------------------------------===//
7456 // SSE4A Instructions
7457 //===----------------------------------------------------------------------===//
7459 let Predicates = [HasSSE4A] in {
7461 let Constraints = "$src = $dst" in {
7462 def EXTRQI : Ii8<0x78, MRM0r, (outs VR128:$dst),
7463 (ins VR128:$src, i8imm:$len, i8imm:$idx),
7464 "extrq\t{$idx, $len, $src|$src, $len, $idx}",
7465 [(set VR128:$dst, (int_x86_sse4a_extrqi VR128:$src, imm:$len,
7466 imm:$idx))]>, TB, OpSize;
7467 def EXTRQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7468 (ins VR128:$src, VR128:$mask),
7469 "extrq\t{$mask, $src|$src, $mask}",
7470 [(set VR128:$dst, (int_x86_sse4a_extrq VR128:$src,
7471 VR128:$mask))]>, TB, OpSize;
7473 def INSERTQI : Ii8<0x78, MRMSrcReg, (outs VR128:$dst),
7474 (ins VR128:$src, VR128:$src2, i8imm:$len, i8imm:$idx),
7475 "insertq\t{$idx, $len, $src2, $src|$src, $src2, $len, $idx}",
7476 [(set VR128:$dst, (int_x86_sse4a_insertqi VR128:$src,
7477 VR128:$src2, imm:$len, imm:$idx))]>, XD;
7478 def INSERTQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7479 (ins VR128:$src, VR128:$mask),
7480 "insertq\t{$mask, $src|$src, $mask}",
7481 [(set VR128:$dst, (int_x86_sse4a_insertq VR128:$src,
7482 VR128:$mask))]>, XD;
7485 def MOVNTSS : I<0x2B, MRMDestMem, (outs), (ins f32mem:$dst, VR128:$src),
7486 "movntss\t{$src, $dst|$dst, $src}",
7487 [(int_x86_sse4a_movnt_ss addr:$dst, VR128:$src)]>, XS;
7489 def MOVNTSD : I<0x2B, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
7490 "movntsd\t{$src, $dst|$dst, $src}",
7491 [(int_x86_sse4a_movnt_sd addr:$dst, VR128:$src)]>, XD;
7494 //===----------------------------------------------------------------------===//
7496 //===----------------------------------------------------------------------===//
7498 //===----------------------------------------------------------------------===//
7499 // VBROADCAST - Load from memory and broadcast to all elements of the
7500 // destination operand
7502 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
7503 X86MemOperand x86memop, Intrinsic Int> :
7504 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7505 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7506 [(set RC:$dst, (Int addr:$src))]>, VEX;
7508 // AVX2 adds register forms
7509 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
7511 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7512 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7513 [(set RC:$dst, (Int VR128:$src))]>, VEX;
7515 let ExeDomain = SSEPackedSingle in {
7516 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
7517 int_x86_avx_vbroadcast_ss>;
7518 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
7519 int_x86_avx_vbroadcast_ss_256>, VEX_L;
7521 let ExeDomain = SSEPackedDouble in
7522 def VBROADCASTSDYrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
7523 int_x86_avx_vbroadcast_sd_256>, VEX_L;
7524 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
7525 int_x86_avx_vbroadcastf128_pd_256>, VEX_L;
7527 let ExeDomain = SSEPackedSingle in {
7528 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
7529 int_x86_avx2_vbroadcast_ss_ps>;
7530 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
7531 int_x86_avx2_vbroadcast_ss_ps_256>, VEX_L;
7533 let ExeDomain = SSEPackedDouble in
7534 def VBROADCASTSDYrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
7535 int_x86_avx2_vbroadcast_sd_pd_256>, VEX_L;
7537 let Predicates = [HasAVX2] in
7538 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
7539 int_x86_avx2_vbroadcasti128>, VEX_L;
7541 let Predicates = [HasAVX] in
7542 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
7543 (VBROADCASTF128 addr:$src)>;
7546 //===----------------------------------------------------------------------===//
7547 // VINSERTF128 - Insert packed floating-point values
7549 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7550 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
7551 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7552 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7555 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
7556 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
7557 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7561 let Predicates = [HasAVX] in {
7562 def : Pat<(vinsert128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7564 (VINSERTF128rr VR256:$src1, VR128:$src2,
7565 (INSERT_get_vinsert128_imm VR256:$ins))>;
7566 def : Pat<(vinsert128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7568 (VINSERTF128rr VR256:$src1, VR128:$src2,
7569 (INSERT_get_vinsert128_imm VR256:$ins))>;
7571 def : Pat<(vinsert128_insert:$ins (v8f32 VR256:$src1), (memopv4f32 addr:$src2),
7573 (VINSERTF128rm VR256:$src1, addr:$src2,
7574 (INSERT_get_vinsert128_imm VR256:$ins))>;
7575 def : Pat<(vinsert128_insert:$ins (v4f64 VR256:$src1), (memopv2f64 addr:$src2),
7577 (VINSERTF128rm VR256:$src1, addr:$src2,
7578 (INSERT_get_vinsert128_imm VR256:$ins))>;
7581 let Predicates = [HasAVX1Only] in {
7582 def : Pat<(vinsert128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7584 (VINSERTF128rr VR256:$src1, VR128:$src2,
7585 (INSERT_get_vinsert128_imm VR256:$ins))>;
7586 def : Pat<(vinsert128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7588 (VINSERTF128rr VR256:$src1, VR128:$src2,
7589 (INSERT_get_vinsert128_imm VR256:$ins))>;
7590 def : Pat<(vinsert128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7592 (VINSERTF128rr VR256:$src1, VR128:$src2,
7593 (INSERT_get_vinsert128_imm VR256:$ins))>;
7594 def : Pat<(vinsert128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7596 (VINSERTF128rr VR256:$src1, VR128:$src2,
7597 (INSERT_get_vinsert128_imm VR256:$ins))>;
7599 def : Pat<(vinsert128_insert:$ins (v4i64 VR256:$src1), (memopv2i64 addr:$src2),
7601 (VINSERTF128rm VR256:$src1, addr:$src2,
7602 (INSERT_get_vinsert128_imm VR256:$ins))>;
7603 def : Pat<(vinsert128_insert:$ins (v8i32 VR256:$src1),
7604 (bc_v4i32 (memopv2i64 addr:$src2)),
7606 (VINSERTF128rm VR256:$src1, addr:$src2,
7607 (INSERT_get_vinsert128_imm VR256:$ins))>;
7608 def : Pat<(vinsert128_insert:$ins (v32i8 VR256:$src1),
7609 (bc_v16i8 (memopv2i64 addr:$src2)),
7611 (VINSERTF128rm VR256:$src1, addr:$src2,
7612 (INSERT_get_vinsert128_imm VR256:$ins))>;
7613 def : Pat<(vinsert128_insert:$ins (v16i16 VR256:$src1),
7614 (bc_v8i16 (memopv2i64 addr:$src2)),
7616 (VINSERTF128rm VR256:$src1, addr:$src2,
7617 (INSERT_get_vinsert128_imm VR256:$ins))>;
7620 //===----------------------------------------------------------------------===//
7621 // VEXTRACTF128 - Extract packed floating-point values
7623 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7624 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
7625 (ins VR256:$src1, i8imm:$src2),
7626 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7629 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
7630 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
7631 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7636 let Predicates = [HasAVX] in {
7637 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7638 (v4f32 (VEXTRACTF128rr
7639 (v8f32 VR256:$src1),
7640 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7641 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7642 (v2f64 (VEXTRACTF128rr
7643 (v4f64 VR256:$src1),
7644 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7646 def : Pat<(alignedstore (v4f32 (vextract128_extract:$ext (v8f32 VR256:$src1),
7647 (iPTR imm))), addr:$dst),
7648 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7649 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7650 def : Pat<(alignedstore (v2f64 (vextract128_extract:$ext (v4f64 VR256:$src1),
7651 (iPTR imm))), addr:$dst),
7652 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7653 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7656 let Predicates = [HasAVX1Only] in {
7657 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7658 (v2i64 (VEXTRACTF128rr
7659 (v4i64 VR256:$src1),
7660 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7661 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7662 (v4i32 (VEXTRACTF128rr
7663 (v8i32 VR256:$src1),
7664 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7665 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7666 (v8i16 (VEXTRACTF128rr
7667 (v16i16 VR256:$src1),
7668 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7669 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
7670 (v16i8 (VEXTRACTF128rr
7671 (v32i8 VR256:$src1),
7672 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
7674 def : Pat<(alignedstore (v2i64 (vextract128_extract:$ext (v4i64 VR256:$src1),
7675 (iPTR imm))), addr:$dst),
7676 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7677 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7678 def : Pat<(alignedstore (v4i32 (vextract128_extract:$ext (v8i32 VR256:$src1),
7679 (iPTR imm))), addr:$dst),
7680 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7681 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7682 def : Pat<(alignedstore (v8i16 (vextract128_extract:$ext (v16i16 VR256:$src1),
7683 (iPTR imm))), addr:$dst),
7684 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7685 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7686 def : Pat<(alignedstore (v16i8 (vextract128_extract:$ext (v32i8 VR256:$src1),
7687 (iPTR imm))), addr:$dst),
7688 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7689 (EXTRACT_get_vextract128_imm VR128:$ext))>;
7692 //===----------------------------------------------------------------------===//
7693 // VMASKMOV - Conditional SIMD Packed Loads and Stores
7695 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
7696 Intrinsic IntLd, Intrinsic IntLd256,
7697 Intrinsic IntSt, Intrinsic IntSt256> {
7698 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
7699 (ins VR128:$src1, f128mem:$src2),
7700 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7701 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
7703 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
7704 (ins VR256:$src1, f256mem:$src2),
7705 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7706 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7708 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
7709 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
7710 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7711 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7712 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
7713 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
7714 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7715 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V, VEX_L;
7718 let ExeDomain = SSEPackedSingle in
7719 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
7720 int_x86_avx_maskload_ps,
7721 int_x86_avx_maskload_ps_256,
7722 int_x86_avx_maskstore_ps,
7723 int_x86_avx_maskstore_ps_256>;
7724 let ExeDomain = SSEPackedDouble in
7725 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
7726 int_x86_avx_maskload_pd,
7727 int_x86_avx_maskload_pd_256,
7728 int_x86_avx_maskstore_pd,
7729 int_x86_avx_maskstore_pd_256>;
7731 //===----------------------------------------------------------------------===//
7732 // VPERMIL - Permute Single and Double Floating-Point Values
7734 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
7735 RegisterClass RC, X86MemOperand x86memop_f,
7736 X86MemOperand x86memop_i, PatFrag i_frag,
7737 Intrinsic IntVar, ValueType vt> {
7738 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
7739 (ins RC:$src1, RC:$src2),
7740 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7741 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
7742 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
7743 (ins RC:$src1, x86memop_i:$src2),
7744 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7745 [(set RC:$dst, (IntVar RC:$src1,
7746 (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
7748 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
7749 (ins RC:$src1, i8imm:$src2),
7750 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7751 [(set RC:$dst, (vt (X86VPermilp RC:$src1, (i8 imm:$src2))))]>, VEX;
7752 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
7753 (ins x86memop_f:$src1, i8imm:$src2),
7754 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7756 (vt (X86VPermilp (memop addr:$src1), (i8 imm:$src2))))]>, VEX;
7759 let ExeDomain = SSEPackedSingle in {
7760 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
7761 memopv2i64, int_x86_avx_vpermilvar_ps, v4f32>;
7762 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
7763 memopv4i64, int_x86_avx_vpermilvar_ps_256, v8f32>, VEX_L;
7765 let ExeDomain = SSEPackedDouble in {
7766 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
7767 memopv2i64, int_x86_avx_vpermilvar_pd, v2f64>;
7768 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
7769 memopv4i64, int_x86_avx_vpermilvar_pd_256, v4f64>, VEX_L;
7772 let Predicates = [HasAVX] in {
7773 def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7774 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7775 def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7776 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7777 def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (memopv4i64 addr:$src1)),
7779 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7780 def : Pat<(v4i64 (X86VPermilp (memopv4i64 addr:$src1), (i8 imm:$imm))),
7781 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7783 def : Pat<(v2i64 (X86VPermilp VR128:$src1, (i8 imm:$imm))),
7784 (VPERMILPDri VR128:$src1, imm:$imm)>;
7785 def : Pat<(v2i64 (X86VPermilp (memopv2i64 addr:$src1), (i8 imm:$imm))),
7786 (VPERMILPDmi addr:$src1, imm:$imm)>;
7789 //===----------------------------------------------------------------------===//
7790 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
7792 let ExeDomain = SSEPackedSingle in {
7793 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
7794 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7795 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7796 [(set VR256:$dst, (v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7797 (i8 imm:$src3))))]>, VEX_4V, VEX_L;
7798 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
7799 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7800 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7801 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv8f32 addr:$src2),
7802 (i8 imm:$src3)))]>, VEX_4V, VEX_L;
7805 let Predicates = [HasAVX] in {
7806 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7807 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7808 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
7809 (memopv4f64 addr:$src2), (i8 imm:$imm))),
7810 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7813 let Predicates = [HasAVX1Only] in {
7814 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7815 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7816 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7817 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7818 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7819 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7820 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7821 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7823 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
7824 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7825 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7826 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
7827 (memopv4i64 addr:$src2), (i8 imm:$imm))),
7828 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7829 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
7830 (bc_v32i8 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7831 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7832 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7833 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7834 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7837 //===----------------------------------------------------------------------===//
7838 // VZERO - Zero YMM registers
7840 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
7841 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
7842 // Zero All YMM registers
7843 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
7844 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
7846 // Zero Upper bits of YMM registers
7847 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
7848 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
7851 //===----------------------------------------------------------------------===//
7852 // Half precision conversion instructions
7853 //===----------------------------------------------------------------------===//
7854 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7855 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7856 "vcvtph2ps\t{$src, $dst|$dst, $src}",
7857 [(set RC:$dst, (Int VR128:$src))]>,
7859 let neverHasSideEffects = 1, mayLoad = 1 in
7860 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7861 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
7864 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7865 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
7866 (ins RC:$src1, i32i8imm:$src2),
7867 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7868 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
7870 let neverHasSideEffects = 1, mayStore = 1 in
7871 def mr : Ii8<0x1D, MRMDestMem, (outs),
7872 (ins x86memop:$dst, RC:$src1, i32i8imm:$src2),
7873 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
7877 let Predicates = [HasAVX, HasF16C] in {
7878 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
7879 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>, VEX_L;
7880 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
7881 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>, VEX_L;
7884 //===----------------------------------------------------------------------===//
7885 // AVX2 Instructions
7886 //===----------------------------------------------------------------------===//
7888 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
7889 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
7890 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
7891 X86MemOperand x86memop> {
7892 let isCommutable = 1 in
7893 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
7894 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
7895 !strconcat(OpcodeStr,
7896 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7897 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
7899 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
7900 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
7901 !strconcat(OpcodeStr,
7902 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7905 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
7909 let isCommutable = 0 in {
7910 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
7911 VR128, memopv2i64, i128mem>;
7912 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
7913 VR256, memopv4i64, i256mem>, VEX_L;
7916 def : Pat<(v4i32 (X86Blendi (v4i32 VR128:$src1), (v4i32 VR128:$src2),
7918 (VPBLENDDrri VR128:$src1, VR128:$src2, imm:$mask)>;
7919 def : Pat<(v8i32 (X86Blendi (v8i32 VR256:$src1), (v8i32 VR256:$src2),
7921 (VPBLENDDYrri VR256:$src1, VR256:$src2, imm:$mask)>;
7923 //===----------------------------------------------------------------------===//
7924 // VPBROADCAST - Load from memory and broadcast to all elements of the
7925 // destination operand
7927 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
7928 X86MemOperand x86memop, PatFrag ld_frag,
7929 Intrinsic Int128, Intrinsic Int256> {
7930 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
7931 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7932 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
7933 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
7934 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7936 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7937 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
7938 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7939 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX, VEX_L;
7940 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
7941 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7943 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>,
7947 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
7948 int_x86_avx2_pbroadcastb_128,
7949 int_x86_avx2_pbroadcastb_256>;
7950 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
7951 int_x86_avx2_pbroadcastw_128,
7952 int_x86_avx2_pbroadcastw_256>;
7953 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
7954 int_x86_avx2_pbroadcastd_128,
7955 int_x86_avx2_pbroadcastd_256>;
7956 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
7957 int_x86_avx2_pbroadcastq_128,
7958 int_x86_avx2_pbroadcastq_256>;
7960 let Predicates = [HasAVX2] in {
7961 def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
7962 (VPBROADCASTBrm addr:$src)>;
7963 def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
7964 (VPBROADCASTBYrm addr:$src)>;
7965 def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
7966 (VPBROADCASTWrm addr:$src)>;
7967 def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
7968 (VPBROADCASTWYrm addr:$src)>;
7969 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7970 (VPBROADCASTDrm addr:$src)>;
7971 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7972 (VPBROADCASTDYrm addr:$src)>;
7973 def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
7974 (VPBROADCASTQrm addr:$src)>;
7975 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7976 (VPBROADCASTQYrm addr:$src)>;
7978 def : Pat<(v16i8 (X86VBroadcast (v16i8 VR128:$src))),
7979 (VPBROADCASTBrr VR128:$src)>;
7980 def : Pat<(v32i8 (X86VBroadcast (v16i8 VR128:$src))),
7981 (VPBROADCASTBYrr VR128:$src)>;
7982 def : Pat<(v8i16 (X86VBroadcast (v8i16 VR128:$src))),
7983 (VPBROADCASTWrr VR128:$src)>;
7984 def : Pat<(v16i16 (X86VBroadcast (v8i16 VR128:$src))),
7985 (VPBROADCASTWYrr VR128:$src)>;
7986 def : Pat<(v4i32 (X86VBroadcast (v4i32 VR128:$src))),
7987 (VPBROADCASTDrr VR128:$src)>;
7988 def : Pat<(v8i32 (X86VBroadcast (v4i32 VR128:$src))),
7989 (VPBROADCASTDYrr VR128:$src)>;
7990 def : Pat<(v2i64 (X86VBroadcast (v2i64 VR128:$src))),
7991 (VPBROADCASTQrr VR128:$src)>;
7992 def : Pat<(v4i64 (X86VBroadcast (v2i64 VR128:$src))),
7993 (VPBROADCASTQYrr VR128:$src)>;
7994 def : Pat<(v4f32 (X86VBroadcast (v4f32 VR128:$src))),
7995 (VBROADCASTSSrr VR128:$src)>;
7996 def : Pat<(v8f32 (X86VBroadcast (v4f32 VR128:$src))),
7997 (VBROADCASTSSYrr VR128:$src)>;
7998 def : Pat<(v2f64 (X86VBroadcast (v2f64 VR128:$src))),
7999 (VPBROADCASTQrr VR128:$src)>;
8000 def : Pat<(v4f64 (X86VBroadcast (v2f64 VR128:$src))),
8001 (VBROADCASTSDYrr VR128:$src)>;
8003 // Provide fallback in case the load node that is used in the patterns above
8004 // is used by additional users, which prevents the pattern selection.
8005 let AddedComplexity = 20 in {
8006 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
8007 (VBROADCASTSSrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
8008 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
8009 (VBROADCASTSSYrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
8010 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
8011 (VBROADCASTSDYrr (COPY_TO_REGCLASS FR64:$src, VR128))>;
8013 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
8014 (VBROADCASTSSrr (COPY_TO_REGCLASS GR32:$src, VR128))>;
8015 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
8016 (VBROADCASTSSYrr (COPY_TO_REGCLASS GR32:$src, VR128))>;
8017 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
8018 (VBROADCASTSDYrr (COPY_TO_REGCLASS GR64:$src, VR128))>;
8022 // AVX1 broadcast patterns
8023 let Predicates = [HasAVX1Only] in {
8024 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
8025 (VBROADCASTSSYrm addr:$src)>;
8026 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
8027 (VBROADCASTSDYrm addr:$src)>;
8028 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
8029 (VBROADCASTSSrm addr:$src)>;
8032 let Predicates = [HasAVX] in {
8033 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
8034 (VBROADCASTSSYrm addr:$src)>;
8035 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
8036 (VBROADCASTSDYrm addr:$src)>;
8037 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
8038 (VBROADCASTSSrm addr:$src)>;
8040 // Provide fallback in case the load node that is used in the patterns above
8041 // is used by additional users, which prevents the pattern selection.
8042 let AddedComplexity = 20 in {
8043 // 128bit broadcasts:
8044 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
8045 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0)>;
8046 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
8047 (VINSERTF128rr (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)),
8048 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0), sub_xmm),
8049 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0), 1)>;
8050 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
8051 (VINSERTF128rr (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)),
8052 (VPSHUFDri (COPY_TO_REGCLASS FR64:$src, VR128), 0x44), sub_xmm),
8053 (VPSHUFDri (COPY_TO_REGCLASS FR64:$src, VR128), 0x44), 1)>;
8055 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
8056 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0)>;
8057 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
8058 (VINSERTF128rr (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)),
8059 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0), sub_xmm),
8060 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0), 1)>;
8061 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
8062 (VINSERTF128rr (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)),
8063 (VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), sub_xmm),
8064 (VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), 1)>;
8068 //===----------------------------------------------------------------------===//
8069 // VPERM - Permute instructions
8072 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
8074 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
8075 (ins VR256:$src1, VR256:$src2),
8076 !strconcat(OpcodeStr,
8077 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8079 (OpVT (X86VPermv VR256:$src1, VR256:$src2)))]>,
8081 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
8082 (ins VR256:$src1, i256mem:$src2),
8083 !strconcat(OpcodeStr,
8084 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8086 (OpVT (X86VPermv VR256:$src1,
8087 (bitconvert (mem_frag addr:$src2)))))]>,
8091 defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, v8i32>;
8092 let ExeDomain = SSEPackedSingle in
8093 defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, v8f32>;
8095 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
8097 def Yri : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
8098 (ins VR256:$src1, i8imm:$src2),
8099 !strconcat(OpcodeStr,
8100 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8102 (OpVT (X86VPermi VR256:$src1, (i8 imm:$src2))))]>,
8104 def Ymi : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
8105 (ins i256mem:$src1, i8imm:$src2),
8106 !strconcat(OpcodeStr,
8107 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8109 (OpVT (X86VPermi (mem_frag addr:$src1),
8110 (i8 imm:$src2))))]>, VEX, VEX_L;
8113 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, v4i64>, VEX_W;
8114 let ExeDomain = SSEPackedDouble in
8115 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, v4f64>, VEX_W;
8117 //===----------------------------------------------------------------------===//
8118 // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
8120 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
8121 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
8122 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
8123 [(set VR256:$dst, (v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2,
8124 (i8 imm:$src3))))]>, VEX_4V, VEX_L;
8125 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
8126 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
8127 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
8128 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
8129 (i8 imm:$src3)))]>, VEX_4V, VEX_L;
8131 let Predicates = [HasAVX2] in {
8132 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
8133 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
8134 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
8135 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
8136 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
8137 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
8139 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (memopv4i64 addr:$src2)),
8141 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
8142 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
8143 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
8144 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
8145 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)),
8147 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
8151 //===----------------------------------------------------------------------===//
8152 // VINSERTI128 - Insert packed integer values
8154 let neverHasSideEffects = 1 in {
8155 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
8156 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
8157 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
8160 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
8161 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
8162 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
8166 let Predicates = [HasAVX2] in {
8167 def : Pat<(vinsert128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
8169 (VINSERTI128rr VR256:$src1, VR128:$src2,
8170 (INSERT_get_vinsert128_imm VR256:$ins))>;
8171 def : Pat<(vinsert128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
8173 (VINSERTI128rr VR256:$src1, VR128:$src2,
8174 (INSERT_get_vinsert128_imm VR256:$ins))>;
8175 def : Pat<(vinsert128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
8177 (VINSERTI128rr VR256:$src1, VR128:$src2,
8178 (INSERT_get_vinsert128_imm VR256:$ins))>;
8179 def : Pat<(vinsert128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
8181 (VINSERTI128rr VR256:$src1, VR128:$src2,
8182 (INSERT_get_vinsert128_imm VR256:$ins))>;
8184 def : Pat<(vinsert128_insert:$ins (v4i64 VR256:$src1), (memopv2i64 addr:$src2),
8186 (VINSERTI128rm VR256:$src1, addr:$src2,
8187 (INSERT_get_vinsert128_imm VR256:$ins))>;
8188 def : Pat<(vinsert128_insert:$ins (v8i32 VR256:$src1),
8189 (bc_v4i32 (memopv2i64 addr:$src2)),
8191 (VINSERTI128rm VR256:$src1, addr:$src2,
8192 (INSERT_get_vinsert128_imm VR256:$ins))>;
8193 def : Pat<(vinsert128_insert:$ins (v32i8 VR256:$src1),
8194 (bc_v16i8 (memopv2i64 addr:$src2)),
8196 (VINSERTI128rm VR256:$src1, addr:$src2,
8197 (INSERT_get_vinsert128_imm VR256:$ins))>;
8198 def : Pat<(vinsert128_insert:$ins (v16i16 VR256:$src1),
8199 (bc_v8i16 (memopv2i64 addr:$src2)),
8201 (VINSERTI128rm VR256:$src1, addr:$src2,
8202 (INSERT_get_vinsert128_imm VR256:$ins))>;
8205 //===----------------------------------------------------------------------===//
8206 // VEXTRACTI128 - Extract packed integer values
8208 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
8209 (ins VR256:$src1, i8imm:$src2),
8210 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
8212 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
8214 let neverHasSideEffects = 1, mayStore = 1 in
8215 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
8216 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
8217 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
8220 let Predicates = [HasAVX2] in {
8221 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
8222 (v2i64 (VEXTRACTI128rr
8223 (v4i64 VR256:$src1),
8224 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
8225 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
8226 (v4i32 (VEXTRACTI128rr
8227 (v8i32 VR256:$src1),
8228 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
8229 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
8230 (v8i16 (VEXTRACTI128rr
8231 (v16i16 VR256:$src1),
8232 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
8233 def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
8234 (v16i8 (VEXTRACTI128rr
8235 (v32i8 VR256:$src1),
8236 (EXTRACT_get_vextract128_imm VR128:$ext)))>;
8238 def : Pat<(alignedstore (v2i64 (vextract128_extract:$ext (v4i64 VR256:$src1),
8239 (iPTR imm))), addr:$dst),
8240 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8241 (EXTRACT_get_vextract128_imm VR128:$ext))>;
8242 def : Pat<(alignedstore (v4i32 (vextract128_extract:$ext (v8i32 VR256:$src1),
8243 (iPTR imm))), addr:$dst),
8244 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8245 (EXTRACT_get_vextract128_imm VR128:$ext))>;
8246 def : Pat<(alignedstore (v8i16 (vextract128_extract:$ext (v16i16 VR256:$src1),
8247 (iPTR imm))), addr:$dst),
8248 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8249 (EXTRACT_get_vextract128_imm VR128:$ext))>;
8250 def : Pat<(alignedstore (v16i8 (vextract128_extract:$ext (v32i8 VR256:$src1),
8251 (iPTR imm))), addr:$dst),
8252 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8253 (EXTRACT_get_vextract128_imm VR128:$ext))>;
8256 //===----------------------------------------------------------------------===//
8257 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
8259 multiclass avx2_pmovmask<string OpcodeStr,
8260 Intrinsic IntLd128, Intrinsic IntLd256,
8261 Intrinsic IntSt128, Intrinsic IntSt256> {
8262 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
8263 (ins VR128:$src1, i128mem:$src2),
8264 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8265 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
8266 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
8267 (ins VR256:$src1, i256mem:$src2),
8268 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8269 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
8271 def mr : AVX28I<0x8e, MRMDestMem, (outs),
8272 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
8273 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8274 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
8275 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
8276 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
8277 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8278 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V, VEX_L;
8281 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
8282 int_x86_avx2_maskload_d,
8283 int_x86_avx2_maskload_d_256,
8284 int_x86_avx2_maskstore_d,
8285 int_x86_avx2_maskstore_d_256>;
8286 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
8287 int_x86_avx2_maskload_q,
8288 int_x86_avx2_maskload_q_256,
8289 int_x86_avx2_maskstore_q,
8290 int_x86_avx2_maskstore_q_256>, VEX_W;
8293 //===----------------------------------------------------------------------===//
8294 // Variable Bit Shifts
8296 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
8297 ValueType vt128, ValueType vt256> {
8298 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
8299 (ins VR128:$src1, VR128:$src2),
8300 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8302 (vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
8304 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
8305 (ins VR128:$src1, i128mem:$src2),
8306 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8308 (vt128 (OpNode VR128:$src1,
8309 (vt128 (bitconvert (memopv2i64 addr:$src2))))))]>,
8311 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
8312 (ins VR256:$src1, VR256:$src2),
8313 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8315 (vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
8317 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
8318 (ins VR256:$src1, i256mem:$src2),
8319 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8321 (vt256 (OpNode VR256:$src1,
8322 (vt256 (bitconvert (memopv4i64 addr:$src2))))))]>,
8326 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
8327 defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", shl, v2i64, v4i64>, VEX_W;
8328 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", srl, v4i32, v8i32>;
8329 defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", srl, v2i64, v4i64>, VEX_W;
8330 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;
8332 //===----------------------------------------------------------------------===//
8333 // VGATHER - GATHER Operations
8334 multiclass avx2_gather<bits<8> opc, string OpcodeStr, RegisterClass RC256,
8335 X86MemOperand memop128, X86MemOperand memop256> {
8336 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst, VR128:$mask_wb),
8337 (ins VR128:$src1, memop128:$src2, VR128:$mask),
8338 !strconcat(OpcodeStr,
8339 "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
8341 def Yrm : AVX28I<opc, MRMSrcMem, (outs RC256:$dst, RC256:$mask_wb),
8342 (ins RC256:$src1, memop256:$src2, RC256:$mask),
8343 !strconcat(OpcodeStr,
8344 "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
8345 []>, VEX_4VOp3, VEX_L;
8348 let mayLoad = 1, Constraints
8349 = "@earlyclobber $dst,@earlyclobber $mask_wb, $src1 = $dst, $mask = $mask_wb"
8351 defm VGATHERDPD : avx2_gather<0x92, "vgatherdpd", VR256, vx64mem, vx64mem>, VEX_W;
8352 defm VGATHERQPD : avx2_gather<0x93, "vgatherqpd", VR256, vx64mem, vy64mem>, VEX_W;
8353 defm VGATHERDPS : avx2_gather<0x92, "vgatherdps", VR256, vx32mem, vy32mem>;
8354 defm VGATHERQPS : avx2_gather<0x93, "vgatherqps", VR128, vx32mem, vy32mem>;
8355 defm VPGATHERDQ : avx2_gather<0x90, "vpgatherdq", VR256, vx64mem, vx64mem>, VEX_W;
8356 defm VPGATHERQQ : avx2_gather<0x91, "vpgatherqq", VR256, vx64mem, vy64mem>, VEX_W;
8357 defm VPGATHERDD : avx2_gather<0x90, "vpgatherdd", VR256, vx32mem, vy32mem>;
8358 defm VPGATHERQD : avx2_gather<0x91, "vpgatherqd", VR128, vx32mem, vy32mem>;